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 YZ |
1179 | if (ret == 0) { |
1180 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID); | |
61a18f1c NB |
1181 | update_inline_extent_backref(path, iref, refs_to_add, |
1182 | extent_op, NULL); | |
5d4f98a2 | 1183 | } else if (ret == -ENOENT) { |
a639cdeb | 1184 | setup_inline_extent_backref(trans->fs_info, path, iref, parent, |
143bede5 JM |
1185 | root_objectid, owner, offset, |
1186 | refs_to_add, extent_op); | |
1187 | ret = 0; | |
771ed689 | 1188 | } |
5d4f98a2 YZ |
1189 | return ret; |
1190 | } | |
31840ae1 | 1191 | |
5d4f98a2 | 1192 | static int remove_extent_backref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1193 | struct btrfs_path *path, |
1194 | struct btrfs_extent_inline_ref *iref, | |
fcebe456 | 1195 | int refs_to_drop, int is_data, int *last_ref) |
5d4f98a2 | 1196 | { |
143bede5 | 1197 | int ret = 0; |
b9473439 | 1198 | |
5d4f98a2 YZ |
1199 | BUG_ON(!is_data && refs_to_drop != 1); |
1200 | if (iref) { | |
61a18f1c NB |
1201 | update_inline_extent_backref(path, iref, -refs_to_drop, NULL, |
1202 | last_ref); | |
5d4f98a2 | 1203 | } else if (is_data) { |
e9f6290d | 1204 | ret = remove_extent_data_ref(trans, path, refs_to_drop, |
fcebe456 | 1205 | last_ref); |
5d4f98a2 | 1206 | } else { |
fcebe456 | 1207 | *last_ref = 1; |
87cc7a8a | 1208 | ret = btrfs_del_item(trans, trans->fs_info->extent_root, path); |
5d4f98a2 YZ |
1209 | } |
1210 | return ret; | |
1211 | } | |
1212 | ||
d04c6b88 JM |
1213 | static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len, |
1214 | u64 *discarded_bytes) | |
5d4f98a2 | 1215 | { |
86557861 JM |
1216 | int j, ret = 0; |
1217 | u64 bytes_left, end; | |
4d89d377 | 1218 | u64 aligned_start = ALIGN(start, 1 << 9); |
d04c6b88 | 1219 | |
4d89d377 JM |
1220 | if (WARN_ON(start != aligned_start)) { |
1221 | len -= aligned_start - start; | |
1222 | len = round_down(len, 1 << 9); | |
1223 | start = aligned_start; | |
1224 | } | |
d04c6b88 | 1225 | |
4d89d377 | 1226 | *discarded_bytes = 0; |
86557861 JM |
1227 | |
1228 | if (!len) | |
1229 | return 0; | |
1230 | ||
1231 | end = start + len; | |
1232 | bytes_left = len; | |
1233 | ||
1234 | /* Skip any superblocks on this device. */ | |
1235 | for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) { | |
1236 | u64 sb_start = btrfs_sb_offset(j); | |
1237 | u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE; | |
1238 | u64 size = sb_start - start; | |
1239 | ||
1240 | if (!in_range(sb_start, start, bytes_left) && | |
1241 | !in_range(sb_end, start, bytes_left) && | |
1242 | !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE)) | |
1243 | continue; | |
1244 | ||
1245 | /* | |
1246 | * Superblock spans beginning of range. Adjust start and | |
1247 | * try again. | |
1248 | */ | |
1249 | if (sb_start <= start) { | |
1250 | start += sb_end - start; | |
1251 | if (start > end) { | |
1252 | bytes_left = 0; | |
1253 | break; | |
1254 | } | |
1255 | bytes_left = end - start; | |
1256 | continue; | |
1257 | } | |
1258 | ||
1259 | if (size) { | |
1260 | ret = blkdev_issue_discard(bdev, start >> 9, size >> 9, | |
1261 | GFP_NOFS, 0); | |
1262 | if (!ret) | |
1263 | *discarded_bytes += size; | |
1264 | else if (ret != -EOPNOTSUPP) | |
1265 | return ret; | |
1266 | } | |
1267 | ||
1268 | start = sb_end; | |
1269 | if (start > end) { | |
1270 | bytes_left = 0; | |
1271 | break; | |
1272 | } | |
1273 | bytes_left = end - start; | |
1274 | } | |
1275 | ||
1276 | if (bytes_left) { | |
1277 | ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9, | |
4d89d377 JM |
1278 | GFP_NOFS, 0); |
1279 | if (!ret) | |
86557861 | 1280 | *discarded_bytes += bytes_left; |
4d89d377 | 1281 | } |
d04c6b88 | 1282 | return ret; |
5d4f98a2 | 1283 | } |
5d4f98a2 | 1284 | |
2ff7e61e | 1285 | int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr, |
1edb647b | 1286 | u64 num_bytes, u64 *actual_bytes) |
5d4f98a2 | 1287 | { |
6b7faadd | 1288 | int ret = 0; |
5378e607 | 1289 | u64 discarded_bytes = 0; |
6b7faadd QW |
1290 | u64 end = bytenr + num_bytes; |
1291 | u64 cur = bytenr; | |
a1d3c478 | 1292 | struct btrfs_bio *bbio = NULL; |
5d4f98a2 | 1293 | |
e244a0ae | 1294 | |
2999241d FM |
1295 | /* |
1296 | * Avoid races with device replace and make sure our bbio has devices | |
1297 | * associated to its stripes that don't go away while we are discarding. | |
1298 | */ | |
0b246afa | 1299 | btrfs_bio_counter_inc_blocked(fs_info); |
6b7faadd QW |
1300 | while (cur < end) { |
1301 | struct btrfs_bio_stripe *stripe; | |
5d4f98a2 YZ |
1302 | int i; |
1303 | ||
6b7faadd QW |
1304 | num_bytes = end - cur; |
1305 | /* Tell the block device(s) that the sectors can be discarded */ | |
1306 | ret = btrfs_map_block(fs_info, BTRFS_MAP_DISCARD, cur, | |
1307 | &num_bytes, &bbio, 0); | |
1308 | /* | |
1309 | * Error can be -ENOMEM, -ENOENT (no such chunk mapping) or | |
1310 | * -EOPNOTSUPP. For any such error, @num_bytes is not updated, | |
1311 | * thus we can't continue anyway. | |
1312 | */ | |
1313 | if (ret < 0) | |
1314 | goto out; | |
5d4f98a2 | 1315 | |
6b7faadd | 1316 | stripe = bbio->stripes; |
a1d3c478 | 1317 | for (i = 0; i < bbio->num_stripes; i++, stripe++) { |
d04c6b88 | 1318 | u64 bytes; |
38b5f68e AJ |
1319 | struct request_queue *req_q; |
1320 | ||
627e0873 FM |
1321 | if (!stripe->dev->bdev) { |
1322 | ASSERT(btrfs_test_opt(fs_info, DEGRADED)); | |
1323 | continue; | |
1324 | } | |
38b5f68e AJ |
1325 | req_q = bdev_get_queue(stripe->dev->bdev); |
1326 | if (!blk_queue_discard(req_q)) | |
d5e2003c JB |
1327 | continue; |
1328 | ||
5378e607 LD |
1329 | ret = btrfs_issue_discard(stripe->dev->bdev, |
1330 | stripe->physical, | |
d04c6b88 JM |
1331 | stripe->length, |
1332 | &bytes); | |
6b7faadd | 1333 | if (!ret) { |
d04c6b88 | 1334 | discarded_bytes += bytes; |
6b7faadd QW |
1335 | } else if (ret != -EOPNOTSUPP) { |
1336 | /* | |
1337 | * Logic errors or -ENOMEM, or -EIO, but | |
1338 | * unlikely to happen. | |
1339 | * | |
1340 | * And since there are two loops, explicitly | |
1341 | * go to out to avoid confusion. | |
1342 | */ | |
1343 | btrfs_put_bbio(bbio); | |
1344 | goto out; | |
1345 | } | |
d5e2003c JB |
1346 | |
1347 | /* | |
1348 | * Just in case we get back EOPNOTSUPP for some reason, | |
1349 | * just ignore the return value so we don't screw up | |
1350 | * people calling discard_extent. | |
1351 | */ | |
1352 | ret = 0; | |
5d4f98a2 | 1353 | } |
6e9606d2 | 1354 | btrfs_put_bbio(bbio); |
6b7faadd | 1355 | cur += num_bytes; |
5d4f98a2 | 1356 | } |
6b7faadd | 1357 | out: |
0b246afa | 1358 | btrfs_bio_counter_dec(fs_info); |
5378e607 LD |
1359 | |
1360 | if (actual_bytes) | |
1361 | *actual_bytes = discarded_bytes; | |
1362 | ||
5d4f98a2 | 1363 | |
53b381b3 DW |
1364 | if (ret == -EOPNOTSUPP) |
1365 | ret = 0; | |
5d4f98a2 | 1366 | return ret; |
5d4f98a2 YZ |
1367 | } |
1368 | ||
79787eaa | 1369 | /* Can return -ENOMEM */ |
5d4f98a2 | 1370 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
82fa113f | 1371 | struct btrfs_ref *generic_ref) |
5d4f98a2 | 1372 | { |
82fa113f | 1373 | struct btrfs_fs_info *fs_info = trans->fs_info; |
d7eae340 | 1374 | int old_ref_mod, new_ref_mod; |
5d4f98a2 | 1375 | int ret; |
66d7e7f0 | 1376 | |
82fa113f QW |
1377 | ASSERT(generic_ref->type != BTRFS_REF_NOT_SET && |
1378 | generic_ref->action); | |
1379 | BUG_ON(generic_ref->type == BTRFS_REF_METADATA && | |
1380 | generic_ref->tree_ref.root == BTRFS_TREE_LOG_OBJECTID); | |
5d4f98a2 | 1381 | |
82fa113f QW |
1382 | if (generic_ref->type == BTRFS_REF_METADATA) |
1383 | ret = btrfs_add_delayed_tree_ref(trans, generic_ref, | |
ed4f255b | 1384 | NULL, &old_ref_mod, &new_ref_mod); |
82fa113f QW |
1385 | else |
1386 | ret = btrfs_add_delayed_data_ref(trans, generic_ref, 0, | |
d7eae340 | 1387 | &old_ref_mod, &new_ref_mod); |
d7eae340 | 1388 | |
82fa113f | 1389 | btrfs_ref_tree_mod(fs_info, generic_ref); |
8a5040f7 | 1390 | |
ddf30cf0 | 1391 | if (ret == 0 && old_ref_mod < 0 && new_ref_mod >= 0) |
78192442 | 1392 | sub_pinned_bytes(fs_info, generic_ref); |
d7eae340 | 1393 | |
5d4f98a2 YZ |
1394 | return ret; |
1395 | } | |
1396 | ||
bd3c685e NB |
1397 | /* |
1398 | * __btrfs_inc_extent_ref - insert backreference for a given extent | |
1399 | * | |
1400 | * @trans: Handle of transaction | |
1401 | * | |
1402 | * @node: The delayed ref node used to get the bytenr/length for | |
1403 | * extent whose references are incremented. | |
1404 | * | |
1405 | * @parent: If this is a shared extent (BTRFS_SHARED_DATA_REF_KEY/ | |
1406 | * BTRFS_SHARED_BLOCK_REF_KEY) then it holds the logical | |
1407 | * bytenr of the parent block. Since new extents are always | |
1408 | * created with indirect references, this will only be the case | |
1409 | * when relocating a shared extent. In that case, root_objectid | |
1410 | * will be BTRFS_TREE_RELOC_OBJECTID. Otheriwse, parent must | |
1411 | * be 0 | |
1412 | * | |
1413 | * @root_objectid: The id of the root where this modification has originated, | |
1414 | * this can be either one of the well-known metadata trees or | |
1415 | * the subvolume id which references this extent. | |
1416 | * | |
1417 | * @owner: For data extents it is the inode number of the owning file. | |
1418 | * For metadata extents this parameter holds the level in the | |
1419 | * tree of the extent. | |
1420 | * | |
1421 | * @offset: For metadata extents the offset is ignored and is currently | |
1422 | * always passed as 0. For data extents it is the fileoffset | |
1423 | * this extent belongs to. | |
1424 | * | |
1425 | * @refs_to_add Number of references to add | |
1426 | * | |
1427 | * @extent_op Pointer to a structure, holding information necessary when | |
1428 | * updating a tree block's flags | |
1429 | * | |
1430 | */ | |
5d4f98a2 | 1431 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
c682f9b3 | 1432 | struct btrfs_delayed_ref_node *node, |
5d4f98a2 YZ |
1433 | u64 parent, u64 root_objectid, |
1434 | u64 owner, u64 offset, int refs_to_add, | |
1435 | struct btrfs_delayed_extent_op *extent_op) | |
1436 | { | |
1437 | struct btrfs_path *path; | |
1438 | struct extent_buffer *leaf; | |
1439 | struct btrfs_extent_item *item; | |
fcebe456 | 1440 | struct btrfs_key key; |
c682f9b3 QW |
1441 | u64 bytenr = node->bytenr; |
1442 | u64 num_bytes = node->num_bytes; | |
5d4f98a2 YZ |
1443 | u64 refs; |
1444 | int ret; | |
5d4f98a2 YZ |
1445 | |
1446 | path = btrfs_alloc_path(); | |
1447 | if (!path) | |
1448 | return -ENOMEM; | |
1449 | ||
5d4f98a2 YZ |
1450 | path->leave_spinning = 1; |
1451 | /* this will setup the path even if it fails to insert the back ref */ | |
a639cdeb NB |
1452 | ret = insert_inline_extent_backref(trans, path, bytenr, num_bytes, |
1453 | parent, root_objectid, owner, | |
1454 | offset, refs_to_add, extent_op); | |
0ed4792a | 1455 | if ((ret < 0 && ret != -EAGAIN) || !ret) |
5d4f98a2 | 1456 | goto out; |
fcebe456 JB |
1457 | |
1458 | /* | |
1459 | * Ok we had -EAGAIN which means we didn't have space to insert and | |
1460 | * inline extent ref, so just update the reference count and add a | |
1461 | * normal backref. | |
1462 | */ | |
5d4f98a2 | 1463 | leaf = path->nodes[0]; |
fcebe456 | 1464 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
5d4f98a2 YZ |
1465 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1466 | refs = btrfs_extent_refs(leaf, item); | |
1467 | btrfs_set_extent_refs(leaf, item, refs + refs_to_add); | |
1468 | if (extent_op) | |
1469 | __run_delayed_extent_op(extent_op, leaf, item); | |
56bec294 | 1470 | |
5d4f98a2 | 1471 | btrfs_mark_buffer_dirty(leaf); |
b3b4aa74 | 1472 | btrfs_release_path(path); |
56bec294 | 1473 | |
b9473439 | 1474 | path->leave_spinning = 1; |
56bec294 | 1475 | /* now insert the actual backref */ |
65cd6d9e NB |
1476 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
1477 | BUG_ON(refs_to_add != 1); | |
1478 | ret = insert_tree_block_ref(trans, path, bytenr, parent, | |
1479 | root_objectid); | |
1480 | } else { | |
1481 | ret = insert_extent_data_ref(trans, path, bytenr, parent, | |
1482 | root_objectid, owner, offset, | |
1483 | refs_to_add); | |
1484 | } | |
79787eaa | 1485 | if (ret) |
66642832 | 1486 | btrfs_abort_transaction(trans, ret); |
5d4f98a2 | 1487 | out: |
56bec294 | 1488 | btrfs_free_path(path); |
30d133fc | 1489 | return ret; |
56bec294 CM |
1490 | } |
1491 | ||
5d4f98a2 | 1492 | static int run_delayed_data_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1493 | struct btrfs_delayed_ref_node *node, |
1494 | struct btrfs_delayed_extent_op *extent_op, | |
1495 | int insert_reserved) | |
56bec294 | 1496 | { |
5d4f98a2 YZ |
1497 | int ret = 0; |
1498 | struct btrfs_delayed_data_ref *ref; | |
1499 | struct btrfs_key ins; | |
1500 | u64 parent = 0; | |
1501 | u64 ref_root = 0; | |
1502 | u64 flags = 0; | |
1503 | ||
1504 | ins.objectid = node->bytenr; | |
1505 | ins.offset = node->num_bytes; | |
1506 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
1507 | ||
1508 | ref = btrfs_delayed_node_to_data_ref(node); | |
2bf98ef3 | 1509 | trace_run_delayed_data_ref(trans->fs_info, node, ref, node->action); |
599c75ec | 1510 | |
5d4f98a2 YZ |
1511 | if (node->type == BTRFS_SHARED_DATA_REF_KEY) |
1512 | parent = ref->parent; | |
fcebe456 | 1513 | ref_root = ref->root; |
5d4f98a2 YZ |
1514 | |
1515 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
3173a18f | 1516 | if (extent_op) |
5d4f98a2 | 1517 | flags |= extent_op->flags_to_set; |
ef89b824 NB |
1518 | ret = alloc_reserved_file_extent(trans, parent, ref_root, |
1519 | flags, ref->objectid, | |
1520 | ref->offset, &ins, | |
1521 | node->ref_mod); | |
5d4f98a2 | 1522 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2590d0f1 NB |
1523 | ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root, |
1524 | ref->objectid, ref->offset, | |
1525 | node->ref_mod, extent_op); | |
5d4f98a2 | 1526 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
e72cb923 | 1527 | ret = __btrfs_free_extent(trans, node, parent, |
5d4f98a2 YZ |
1528 | ref_root, ref->objectid, |
1529 | ref->offset, node->ref_mod, | |
c682f9b3 | 1530 | extent_op); |
5d4f98a2 YZ |
1531 | } else { |
1532 | BUG(); | |
1533 | } | |
1534 | return ret; | |
1535 | } | |
1536 | ||
1537 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
1538 | struct extent_buffer *leaf, | |
1539 | struct btrfs_extent_item *ei) | |
1540 | { | |
1541 | u64 flags = btrfs_extent_flags(leaf, ei); | |
1542 | if (extent_op->update_flags) { | |
1543 | flags |= extent_op->flags_to_set; | |
1544 | btrfs_set_extent_flags(leaf, ei, flags); | |
1545 | } | |
1546 | ||
1547 | if (extent_op->update_key) { | |
1548 | struct btrfs_tree_block_info *bi; | |
1549 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)); | |
1550 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
1551 | btrfs_set_tree_block_key(leaf, bi, &extent_op->key); | |
1552 | } | |
1553 | } | |
1554 | ||
1555 | static int run_delayed_extent_op(struct btrfs_trans_handle *trans, | |
d278850e | 1556 | struct btrfs_delayed_ref_head *head, |
5d4f98a2 YZ |
1557 | struct btrfs_delayed_extent_op *extent_op) |
1558 | { | |
20b9a2d6 | 1559 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5d4f98a2 YZ |
1560 | struct btrfs_key key; |
1561 | struct btrfs_path *path; | |
1562 | struct btrfs_extent_item *ei; | |
1563 | struct extent_buffer *leaf; | |
1564 | u32 item_size; | |
56bec294 | 1565 | int ret; |
5d4f98a2 | 1566 | int err = 0; |
b1c79e09 | 1567 | int metadata = !extent_op->is_data; |
5d4f98a2 | 1568 | |
bf31f87f | 1569 | if (TRANS_ABORTED(trans)) |
79787eaa JM |
1570 | return 0; |
1571 | ||
0b246afa | 1572 | if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
3173a18f JB |
1573 | metadata = 0; |
1574 | ||
5d4f98a2 YZ |
1575 | path = btrfs_alloc_path(); |
1576 | if (!path) | |
1577 | return -ENOMEM; | |
1578 | ||
d278850e | 1579 | key.objectid = head->bytenr; |
5d4f98a2 | 1580 | |
3173a18f | 1581 | if (metadata) { |
3173a18f | 1582 | key.type = BTRFS_METADATA_ITEM_KEY; |
b1c79e09 | 1583 | key.offset = extent_op->level; |
3173a18f JB |
1584 | } else { |
1585 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
d278850e | 1586 | key.offset = head->num_bytes; |
3173a18f JB |
1587 | } |
1588 | ||
1589 | again: | |
5d4f98a2 | 1590 | path->leave_spinning = 1; |
0b246afa | 1591 | ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 1); |
5d4f98a2 YZ |
1592 | if (ret < 0) { |
1593 | err = ret; | |
1594 | goto out; | |
1595 | } | |
1596 | if (ret > 0) { | |
3173a18f | 1597 | if (metadata) { |
55994887 FDBM |
1598 | if (path->slots[0] > 0) { |
1599 | path->slots[0]--; | |
1600 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
1601 | path->slots[0]); | |
d278850e | 1602 | if (key.objectid == head->bytenr && |
55994887 | 1603 | key.type == BTRFS_EXTENT_ITEM_KEY && |
d278850e | 1604 | key.offset == head->num_bytes) |
55994887 FDBM |
1605 | ret = 0; |
1606 | } | |
1607 | if (ret > 0) { | |
1608 | btrfs_release_path(path); | |
1609 | metadata = 0; | |
3173a18f | 1610 | |
d278850e JB |
1611 | key.objectid = head->bytenr; |
1612 | key.offset = head->num_bytes; | |
55994887 FDBM |
1613 | key.type = BTRFS_EXTENT_ITEM_KEY; |
1614 | goto again; | |
1615 | } | |
1616 | } else { | |
1617 | err = -EIO; | |
1618 | goto out; | |
3173a18f | 1619 | } |
5d4f98a2 YZ |
1620 | } |
1621 | ||
1622 | leaf = path->nodes[0]; | |
1623 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
ba3c2b19 | 1624 | |
6d8ff4e4 | 1625 | if (unlikely(item_size < sizeof(*ei))) { |
ba3c2b19 NB |
1626 | err = -EINVAL; |
1627 | btrfs_print_v0_err(fs_info); | |
1628 | btrfs_abort_transaction(trans, err); | |
1629 | goto out; | |
1630 | } | |
1631 | ||
5d4f98a2 YZ |
1632 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1633 | __run_delayed_extent_op(extent_op, leaf, ei); | |
56bec294 | 1634 | |
5d4f98a2 YZ |
1635 | btrfs_mark_buffer_dirty(leaf); |
1636 | out: | |
1637 | btrfs_free_path(path); | |
1638 | return err; | |
56bec294 CM |
1639 | } |
1640 | ||
5d4f98a2 | 1641 | static int run_delayed_tree_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1642 | struct btrfs_delayed_ref_node *node, |
1643 | struct btrfs_delayed_extent_op *extent_op, | |
1644 | int insert_reserved) | |
56bec294 CM |
1645 | { |
1646 | int ret = 0; | |
5d4f98a2 | 1647 | struct btrfs_delayed_tree_ref *ref; |
5d4f98a2 YZ |
1648 | u64 parent = 0; |
1649 | u64 ref_root = 0; | |
56bec294 | 1650 | |
5d4f98a2 | 1651 | ref = btrfs_delayed_node_to_tree_ref(node); |
f97806f2 | 1652 | trace_run_delayed_tree_ref(trans->fs_info, node, ref, node->action); |
599c75ec | 1653 | |
5d4f98a2 YZ |
1654 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) |
1655 | parent = ref->parent; | |
fcebe456 | 1656 | ref_root = ref->root; |
5d4f98a2 | 1657 | |
02794222 | 1658 | if (node->ref_mod != 1) { |
f97806f2 | 1659 | btrfs_err(trans->fs_info, |
02794222 LB |
1660 | "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu", |
1661 | node->bytenr, node->ref_mod, node->action, ref_root, | |
1662 | parent); | |
1663 | return -EIO; | |
1664 | } | |
5d4f98a2 | 1665 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { |
3173a18f | 1666 | BUG_ON(!extent_op || !extent_op->update_flags); |
21ebfbe7 | 1667 | ret = alloc_reserved_tree_block(trans, node, extent_op); |
5d4f98a2 | 1668 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2590d0f1 NB |
1669 | ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root, |
1670 | ref->level, 0, 1, extent_op); | |
5d4f98a2 | 1671 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
e72cb923 | 1672 | ret = __btrfs_free_extent(trans, node, parent, ref_root, |
c682f9b3 | 1673 | ref->level, 0, 1, extent_op); |
5d4f98a2 YZ |
1674 | } else { |
1675 | BUG(); | |
1676 | } | |
56bec294 CM |
1677 | return ret; |
1678 | } | |
1679 | ||
1680 | /* helper function to actually process a single delayed ref entry */ | |
5d4f98a2 | 1681 | static int run_one_delayed_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1682 | struct btrfs_delayed_ref_node *node, |
1683 | struct btrfs_delayed_extent_op *extent_op, | |
1684 | int insert_reserved) | |
56bec294 | 1685 | { |
79787eaa JM |
1686 | int ret = 0; |
1687 | ||
bf31f87f | 1688 | if (TRANS_ABORTED(trans)) { |
857cc2fc | 1689 | if (insert_reserved) |
b25c36f8 | 1690 | btrfs_pin_extent(trans, node->bytenr, node->num_bytes, 1); |
79787eaa | 1691 | return 0; |
857cc2fc | 1692 | } |
79787eaa | 1693 | |
5d4f98a2 YZ |
1694 | if (node->type == BTRFS_TREE_BLOCK_REF_KEY || |
1695 | node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
f97806f2 | 1696 | ret = run_delayed_tree_ref(trans, node, extent_op, |
5d4f98a2 YZ |
1697 | insert_reserved); |
1698 | else if (node->type == BTRFS_EXTENT_DATA_REF_KEY || | |
1699 | node->type == BTRFS_SHARED_DATA_REF_KEY) | |
2bf98ef3 | 1700 | ret = run_delayed_data_ref(trans, node, extent_op, |
5d4f98a2 YZ |
1701 | insert_reserved); |
1702 | else | |
1703 | BUG(); | |
80ee54bf | 1704 | if (ret && insert_reserved) |
b25c36f8 | 1705 | btrfs_pin_extent(trans, node->bytenr, node->num_bytes, 1); |
5d4f98a2 | 1706 | return ret; |
56bec294 CM |
1707 | } |
1708 | ||
c6fc2454 | 1709 | static inline struct btrfs_delayed_ref_node * |
56bec294 CM |
1710 | select_delayed_ref(struct btrfs_delayed_ref_head *head) |
1711 | { | |
cffc3374 FM |
1712 | struct btrfs_delayed_ref_node *ref; |
1713 | ||
e3d03965 | 1714 | if (RB_EMPTY_ROOT(&head->ref_tree.rb_root)) |
c6fc2454 | 1715 | return NULL; |
d7df2c79 | 1716 | |
cffc3374 FM |
1717 | /* |
1718 | * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first. | |
1719 | * This is to prevent a ref count from going down to zero, which deletes | |
1720 | * the extent item from the extent tree, when there still are references | |
1721 | * to add, which would fail because they would not find the extent item. | |
1722 | */ | |
1d57ee94 WX |
1723 | if (!list_empty(&head->ref_add_list)) |
1724 | return list_first_entry(&head->ref_add_list, | |
1725 | struct btrfs_delayed_ref_node, add_list); | |
1726 | ||
e3d03965 | 1727 | ref = rb_entry(rb_first_cached(&head->ref_tree), |
0e0adbcf | 1728 | struct btrfs_delayed_ref_node, ref_node); |
1d57ee94 WX |
1729 | ASSERT(list_empty(&ref->add_list)); |
1730 | return ref; | |
56bec294 CM |
1731 | } |
1732 | ||
2eadaa22 JB |
1733 | static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs, |
1734 | struct btrfs_delayed_ref_head *head) | |
1735 | { | |
1736 | spin_lock(&delayed_refs->lock); | |
1737 | head->processing = 0; | |
1738 | delayed_refs->num_heads_ready++; | |
1739 | spin_unlock(&delayed_refs->lock); | |
1740 | btrfs_delayed_ref_unlock(head); | |
1741 | } | |
1742 | ||
bedc6617 JB |
1743 | static struct btrfs_delayed_extent_op *cleanup_extent_op( |
1744 | struct btrfs_delayed_ref_head *head) | |
b00e6250 JB |
1745 | { |
1746 | struct btrfs_delayed_extent_op *extent_op = head->extent_op; | |
b00e6250 JB |
1747 | |
1748 | if (!extent_op) | |
bedc6617 JB |
1749 | return NULL; |
1750 | ||
b00e6250 | 1751 | if (head->must_insert_reserved) { |
bedc6617 | 1752 | head->extent_op = NULL; |
b00e6250 | 1753 | btrfs_free_delayed_extent_op(extent_op); |
bedc6617 | 1754 | return NULL; |
b00e6250 | 1755 | } |
bedc6617 JB |
1756 | return extent_op; |
1757 | } | |
1758 | ||
1759 | static int run_and_cleanup_extent_op(struct btrfs_trans_handle *trans, | |
1760 | struct btrfs_delayed_ref_head *head) | |
1761 | { | |
1762 | struct btrfs_delayed_extent_op *extent_op; | |
1763 | int ret; | |
1764 | ||
1765 | extent_op = cleanup_extent_op(head); | |
1766 | if (!extent_op) | |
1767 | return 0; | |
1768 | head->extent_op = NULL; | |
b00e6250 | 1769 | spin_unlock(&head->lock); |
20b9a2d6 | 1770 | ret = run_delayed_extent_op(trans, head, extent_op); |
b00e6250 JB |
1771 | btrfs_free_delayed_extent_op(extent_op); |
1772 | return ret ? ret : 1; | |
1773 | } | |
1774 | ||
31890da0 JB |
1775 | void btrfs_cleanup_ref_head_accounting(struct btrfs_fs_info *fs_info, |
1776 | struct btrfs_delayed_ref_root *delayed_refs, | |
1777 | struct btrfs_delayed_ref_head *head) | |
07c47775 | 1778 | { |
ba2c4d4e | 1779 | int nr_items = 1; /* Dropping this ref head update. */ |
07c47775 JB |
1780 | |
1781 | if (head->total_ref_mod < 0) { | |
1782 | struct btrfs_space_info *space_info; | |
1783 | u64 flags; | |
1784 | ||
1785 | if (head->is_data) | |
1786 | flags = BTRFS_BLOCK_GROUP_DATA; | |
1787 | else if (head->is_system) | |
1788 | flags = BTRFS_BLOCK_GROUP_SYSTEM; | |
1789 | else | |
1790 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
280c2908 | 1791 | space_info = btrfs_find_space_info(fs_info, flags); |
07c47775 JB |
1792 | ASSERT(space_info); |
1793 | percpu_counter_add_batch(&space_info->total_bytes_pinned, | |
1794 | -head->num_bytes, | |
1795 | BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
1796 | ||
ba2c4d4e JB |
1797 | /* |
1798 | * We had csum deletions accounted for in our delayed refs rsv, | |
1799 | * we need to drop the csum leaves for this update from our | |
1800 | * delayed_refs_rsv. | |
1801 | */ | |
07c47775 JB |
1802 | if (head->is_data) { |
1803 | spin_lock(&delayed_refs->lock); | |
1804 | delayed_refs->pending_csums -= head->num_bytes; | |
1805 | spin_unlock(&delayed_refs->lock); | |
ba2c4d4e JB |
1806 | nr_items += btrfs_csum_bytes_to_leaves(fs_info, |
1807 | head->num_bytes); | |
07c47775 JB |
1808 | } |
1809 | } | |
1810 | ||
ba2c4d4e | 1811 | btrfs_delayed_refs_rsv_release(fs_info, nr_items); |
07c47775 JB |
1812 | } |
1813 | ||
194ab0bc | 1814 | static int cleanup_ref_head(struct btrfs_trans_handle *trans, |
194ab0bc JB |
1815 | struct btrfs_delayed_ref_head *head) |
1816 | { | |
f9871edd NB |
1817 | |
1818 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
194ab0bc JB |
1819 | struct btrfs_delayed_ref_root *delayed_refs; |
1820 | int ret; | |
1821 | ||
1822 | delayed_refs = &trans->transaction->delayed_refs; | |
1823 | ||
bedc6617 | 1824 | ret = run_and_cleanup_extent_op(trans, head); |
194ab0bc JB |
1825 | if (ret < 0) { |
1826 | unselect_delayed_ref_head(delayed_refs, head); | |
1827 | btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret); | |
1828 | return ret; | |
1829 | } else if (ret) { | |
1830 | return ret; | |
1831 | } | |
1832 | ||
1833 | /* | |
1834 | * Need to drop our head ref lock and re-acquire the delayed ref lock | |
1835 | * and then re-check to make sure nobody got added. | |
1836 | */ | |
1837 | spin_unlock(&head->lock); | |
1838 | spin_lock(&delayed_refs->lock); | |
1839 | spin_lock(&head->lock); | |
e3d03965 | 1840 | if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root) || head->extent_op) { |
194ab0bc JB |
1841 | spin_unlock(&head->lock); |
1842 | spin_unlock(&delayed_refs->lock); | |
1843 | return 1; | |
1844 | } | |
d7baffda | 1845 | btrfs_delete_ref_head(delayed_refs, head); |
c1103f7a | 1846 | spin_unlock(&head->lock); |
1e7a1421 | 1847 | spin_unlock(&delayed_refs->lock); |
c1103f7a | 1848 | |
c1103f7a | 1849 | if (head->must_insert_reserved) { |
b25c36f8 | 1850 | btrfs_pin_extent(trans, head->bytenr, head->num_bytes, 1); |
c1103f7a | 1851 | if (head->is_data) { |
40e046ac FM |
1852 | ret = btrfs_del_csums(trans, fs_info->csum_root, |
1853 | head->bytenr, head->num_bytes); | |
c1103f7a JB |
1854 | } |
1855 | } | |
1856 | ||
31890da0 | 1857 | btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head); |
07c47775 JB |
1858 | |
1859 | trace_run_delayed_ref_head(fs_info, head, 0); | |
c1103f7a | 1860 | btrfs_delayed_ref_unlock(head); |
d278850e | 1861 | btrfs_put_delayed_ref_head(head); |
194ab0bc JB |
1862 | return 0; |
1863 | } | |
1864 | ||
b1cdbcb5 NB |
1865 | static struct btrfs_delayed_ref_head *btrfs_obtain_ref_head( |
1866 | struct btrfs_trans_handle *trans) | |
1867 | { | |
1868 | struct btrfs_delayed_ref_root *delayed_refs = | |
1869 | &trans->transaction->delayed_refs; | |
1870 | struct btrfs_delayed_ref_head *head = NULL; | |
1871 | int ret; | |
1872 | ||
1873 | spin_lock(&delayed_refs->lock); | |
5637c74b | 1874 | head = btrfs_select_ref_head(delayed_refs); |
b1cdbcb5 NB |
1875 | if (!head) { |
1876 | spin_unlock(&delayed_refs->lock); | |
1877 | return head; | |
1878 | } | |
1879 | ||
1880 | /* | |
1881 | * Grab the lock that says we are going to process all the refs for | |
1882 | * this head | |
1883 | */ | |
9e920a6f | 1884 | ret = btrfs_delayed_ref_lock(delayed_refs, head); |
b1cdbcb5 NB |
1885 | spin_unlock(&delayed_refs->lock); |
1886 | ||
1887 | /* | |
1888 | * We may have dropped the spin lock to get the head mutex lock, and | |
1889 | * that might have given someone else time to free the head. If that's | |
1890 | * true, it has been removed from our list and we can move on. | |
1891 | */ | |
1892 | if (ret == -EAGAIN) | |
1893 | head = ERR_PTR(-EAGAIN); | |
1894 | ||
1895 | return head; | |
1896 | } | |
1897 | ||
e7261386 NB |
1898 | static int btrfs_run_delayed_refs_for_head(struct btrfs_trans_handle *trans, |
1899 | struct btrfs_delayed_ref_head *locked_ref, | |
1900 | unsigned long *run_refs) | |
1901 | { | |
1902 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
1903 | struct btrfs_delayed_ref_root *delayed_refs; | |
1904 | struct btrfs_delayed_extent_op *extent_op; | |
1905 | struct btrfs_delayed_ref_node *ref; | |
1906 | int must_insert_reserved = 0; | |
1907 | int ret; | |
1908 | ||
1909 | delayed_refs = &trans->transaction->delayed_refs; | |
1910 | ||
0110a4c4 NB |
1911 | lockdep_assert_held(&locked_ref->mutex); |
1912 | lockdep_assert_held(&locked_ref->lock); | |
1913 | ||
e7261386 NB |
1914 | while ((ref = select_delayed_ref(locked_ref))) { |
1915 | if (ref->seq && | |
1916 | btrfs_check_delayed_seq(fs_info, ref->seq)) { | |
1917 | spin_unlock(&locked_ref->lock); | |
1918 | unselect_delayed_ref_head(delayed_refs, locked_ref); | |
1919 | return -EAGAIN; | |
1920 | } | |
1921 | ||
1922 | (*run_refs)++; | |
1923 | ref->in_tree = 0; | |
1924 | rb_erase_cached(&ref->ref_node, &locked_ref->ref_tree); | |
1925 | RB_CLEAR_NODE(&ref->ref_node); | |
1926 | if (!list_empty(&ref->add_list)) | |
1927 | list_del(&ref->add_list); | |
1928 | /* | |
1929 | * When we play the delayed ref, also correct the ref_mod on | |
1930 | * head | |
1931 | */ | |
1932 | switch (ref->action) { | |
1933 | case BTRFS_ADD_DELAYED_REF: | |
1934 | case BTRFS_ADD_DELAYED_EXTENT: | |
1935 | locked_ref->ref_mod -= ref->ref_mod; | |
1936 | break; | |
1937 | case BTRFS_DROP_DELAYED_REF: | |
1938 | locked_ref->ref_mod += ref->ref_mod; | |
1939 | break; | |
1940 | default: | |
1941 | WARN_ON(1); | |
1942 | } | |
1943 | atomic_dec(&delayed_refs->num_entries); | |
1944 | ||
1945 | /* | |
1946 | * Record the must_insert_reserved flag before we drop the | |
1947 | * spin lock. | |
1948 | */ | |
1949 | must_insert_reserved = locked_ref->must_insert_reserved; | |
1950 | locked_ref->must_insert_reserved = 0; | |
1951 | ||
1952 | extent_op = locked_ref->extent_op; | |
1953 | locked_ref->extent_op = NULL; | |
1954 | spin_unlock(&locked_ref->lock); | |
1955 | ||
1956 | ret = run_one_delayed_ref(trans, ref, extent_op, | |
1957 | must_insert_reserved); | |
1958 | ||
1959 | btrfs_free_delayed_extent_op(extent_op); | |
1960 | if (ret) { | |
1961 | unselect_delayed_ref_head(delayed_refs, locked_ref); | |
1962 | btrfs_put_delayed_ref(ref); | |
1963 | btrfs_debug(fs_info, "run_one_delayed_ref returned %d", | |
1964 | ret); | |
1965 | return ret; | |
1966 | } | |
1967 | ||
1968 | btrfs_put_delayed_ref(ref); | |
1969 | cond_resched(); | |
1970 | ||
1971 | spin_lock(&locked_ref->lock); | |
1972 | btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref); | |
1973 | } | |
1974 | ||
1975 | return 0; | |
1976 | } | |
1977 | ||
79787eaa JM |
1978 | /* |
1979 | * Returns 0 on success or if called with an already aborted transaction. | |
1980 | * Returns -ENOMEM or -EIO on failure and will abort the transaction. | |
1981 | */ | |
d7df2c79 | 1982 | static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, |
d7df2c79 | 1983 | unsigned long nr) |
56bec294 | 1984 | { |
0a1e458a | 1985 | struct btrfs_fs_info *fs_info = trans->fs_info; |
56bec294 | 1986 | struct btrfs_delayed_ref_root *delayed_refs; |
56bec294 | 1987 | struct btrfs_delayed_ref_head *locked_ref = NULL; |
0a2b2a84 | 1988 | ktime_t start = ktime_get(); |
56bec294 | 1989 | int ret; |
d7df2c79 | 1990 | unsigned long count = 0; |
0a2b2a84 | 1991 | unsigned long actual_count = 0; |
56bec294 CM |
1992 | |
1993 | delayed_refs = &trans->transaction->delayed_refs; | |
0110a4c4 | 1994 | do { |
56bec294 | 1995 | if (!locked_ref) { |
b1cdbcb5 | 1996 | locked_ref = btrfs_obtain_ref_head(trans); |
0110a4c4 NB |
1997 | if (IS_ERR_OR_NULL(locked_ref)) { |
1998 | if (PTR_ERR(locked_ref) == -EAGAIN) { | |
1999 | continue; | |
2000 | } else { | |
2001 | break; | |
2002 | } | |
56bec294 | 2003 | } |
0110a4c4 | 2004 | count++; |
56bec294 | 2005 | } |
2c3cf7d5 FM |
2006 | /* |
2007 | * We need to try and merge add/drops of the same ref since we | |
2008 | * can run into issues with relocate dropping the implicit ref | |
2009 | * and then it being added back again before the drop can | |
2010 | * finish. If we merged anything we need to re-loop so we can | |
2011 | * get a good ref. | |
2012 | * Or we can get node references of the same type that weren't | |
2013 | * merged when created due to bumps in the tree mod seq, and | |
2014 | * we need to merge them to prevent adding an inline extent | |
2015 | * backref before dropping it (triggering a BUG_ON at | |
2016 | * insert_inline_extent_backref()). | |
2017 | */ | |
d7df2c79 | 2018 | spin_lock(&locked_ref->lock); |
be97f133 | 2019 | btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref); |
ae1e206b | 2020 | |
0110a4c4 NB |
2021 | ret = btrfs_run_delayed_refs_for_head(trans, locked_ref, |
2022 | &actual_count); | |
2023 | if (ret < 0 && ret != -EAGAIN) { | |
2024 | /* | |
2025 | * Error, btrfs_run_delayed_refs_for_head already | |
2026 | * unlocked everything so just bail out | |
2027 | */ | |
2028 | return ret; | |
2029 | } else if (!ret) { | |
2030 | /* | |
2031 | * Success, perform the usual cleanup of a processed | |
2032 | * head | |
2033 | */ | |
f9871edd | 2034 | ret = cleanup_ref_head(trans, locked_ref); |
194ab0bc | 2035 | if (ret > 0 ) { |
b00e6250 JB |
2036 | /* We dropped our lock, we need to loop. */ |
2037 | ret = 0; | |
d7df2c79 | 2038 | continue; |
194ab0bc JB |
2039 | } else if (ret) { |
2040 | return ret; | |
5d4f98a2 | 2041 | } |
22cd2e7d | 2042 | } |
1ce7a5ec | 2043 | |
b00e6250 | 2044 | /* |
0110a4c4 NB |
2045 | * Either success case or btrfs_run_delayed_refs_for_head |
2046 | * returned -EAGAIN, meaning we need to select another head | |
b00e6250 | 2047 | */ |
b00e6250 | 2048 | |
0110a4c4 | 2049 | locked_ref = NULL; |
c3e69d58 | 2050 | cond_resched(); |
0110a4c4 | 2051 | } while ((nr != -1 && count < nr) || locked_ref); |
0a2b2a84 JB |
2052 | |
2053 | /* | |
2054 | * We don't want to include ref heads since we can have empty ref heads | |
2055 | * and those will drastically skew our runtime down since we just do | |
2056 | * accounting, no actual extent tree updates. | |
2057 | */ | |
2058 | if (actual_count > 0) { | |
2059 | u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start)); | |
2060 | u64 avg; | |
2061 | ||
2062 | /* | |
2063 | * We weigh the current average higher than our current runtime | |
2064 | * to avoid large swings in the average. | |
2065 | */ | |
2066 | spin_lock(&delayed_refs->lock); | |
2067 | avg = fs_info->avg_delayed_ref_runtime * 3 + runtime; | |
f8c269d7 | 2068 | fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */ |
0a2b2a84 JB |
2069 | spin_unlock(&delayed_refs->lock); |
2070 | } | |
d7df2c79 | 2071 | return 0; |
c3e69d58 CM |
2072 | } |
2073 | ||
709c0486 AJ |
2074 | #ifdef SCRAMBLE_DELAYED_REFS |
2075 | /* | |
2076 | * Normally delayed refs get processed in ascending bytenr order. This | |
2077 | * correlates in most cases to the order added. To expose dependencies on this | |
2078 | * order, we start to process the tree in the middle instead of the beginning | |
2079 | */ | |
2080 | static u64 find_middle(struct rb_root *root) | |
2081 | { | |
2082 | struct rb_node *n = root->rb_node; | |
2083 | struct btrfs_delayed_ref_node *entry; | |
2084 | int alt = 1; | |
2085 | u64 middle; | |
2086 | u64 first = 0, last = 0; | |
2087 | ||
2088 | n = rb_first(root); | |
2089 | if (n) { | |
2090 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2091 | first = entry->bytenr; | |
2092 | } | |
2093 | n = rb_last(root); | |
2094 | if (n) { | |
2095 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2096 | last = entry->bytenr; | |
2097 | } | |
2098 | n = root->rb_node; | |
2099 | ||
2100 | while (n) { | |
2101 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2102 | WARN_ON(!entry->in_tree); | |
2103 | ||
2104 | middle = entry->bytenr; | |
2105 | ||
2106 | if (alt) | |
2107 | n = n->rb_left; | |
2108 | else | |
2109 | n = n->rb_right; | |
2110 | ||
2111 | alt = 1 - alt; | |
2112 | } | |
2113 | return middle; | |
2114 | } | |
2115 | #endif | |
2116 | ||
1262133b JB |
2117 | /* |
2118 | * Takes the number of bytes to be csumm'ed and figures out how many leaves it | |
2119 | * would require to store the csums for that many bytes. | |
2120 | */ | |
2ff7e61e | 2121 | u64 btrfs_csum_bytes_to_leaves(struct btrfs_fs_info *fs_info, u64 csum_bytes) |
1262133b JB |
2122 | { |
2123 | u64 csum_size; | |
2124 | u64 num_csums_per_leaf; | |
2125 | u64 num_csums; | |
2126 | ||
0b246afa | 2127 | csum_size = BTRFS_MAX_ITEM_SIZE(fs_info); |
1262133b | 2128 | num_csums_per_leaf = div64_u64(csum_size, |
0b246afa JM |
2129 | (u64)btrfs_super_csum_size(fs_info->super_copy)); |
2130 | num_csums = div64_u64(csum_bytes, fs_info->sectorsize); | |
1262133b JB |
2131 | num_csums += num_csums_per_leaf - 1; |
2132 | num_csums = div64_u64(num_csums, num_csums_per_leaf); | |
2133 | return num_csums; | |
2134 | } | |
2135 | ||
c3e69d58 CM |
2136 | /* |
2137 | * this starts processing the delayed reference count updates and | |
2138 | * extent insertions we have queued up so far. count can be | |
2139 | * 0, which means to process everything in the tree at the start | |
2140 | * of the run (but not newly added entries), or it can be some target | |
2141 | * number you'd like to process. | |
79787eaa JM |
2142 | * |
2143 | * Returns 0 on success or if called with an aborted transaction | |
2144 | * Returns <0 on error and aborts the transaction | |
c3e69d58 CM |
2145 | */ |
2146 | int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, | |
c79a70b1 | 2147 | unsigned long count) |
c3e69d58 | 2148 | { |
c79a70b1 | 2149 | struct btrfs_fs_info *fs_info = trans->fs_info; |
c3e69d58 CM |
2150 | struct rb_node *node; |
2151 | struct btrfs_delayed_ref_root *delayed_refs; | |
c46effa6 | 2152 | struct btrfs_delayed_ref_head *head; |
c3e69d58 CM |
2153 | int ret; |
2154 | int run_all = count == (unsigned long)-1; | |
c3e69d58 | 2155 | |
79787eaa | 2156 | /* We'll clean this up in btrfs_cleanup_transaction */ |
bf31f87f | 2157 | if (TRANS_ABORTED(trans)) |
79787eaa JM |
2158 | return 0; |
2159 | ||
0b246afa | 2160 | if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags)) |
511711af CM |
2161 | return 0; |
2162 | ||
c3e69d58 | 2163 | delayed_refs = &trans->transaction->delayed_refs; |
26455d33 | 2164 | if (count == 0) |
d7df2c79 | 2165 | count = atomic_read(&delayed_refs->num_entries) * 2; |
bb721703 | 2166 | |
c3e69d58 | 2167 | again: |
709c0486 AJ |
2168 | #ifdef SCRAMBLE_DELAYED_REFS |
2169 | delayed_refs->run_delayed_start = find_middle(&delayed_refs->root); | |
2170 | #endif | |
0a1e458a | 2171 | ret = __btrfs_run_delayed_refs(trans, count); |
d7df2c79 | 2172 | if (ret < 0) { |
66642832 | 2173 | btrfs_abort_transaction(trans, ret); |
d7df2c79 | 2174 | return ret; |
eb099670 | 2175 | } |
c3e69d58 | 2176 | |
56bec294 | 2177 | if (run_all) { |
119e80df | 2178 | btrfs_create_pending_block_groups(trans); |
ea658bad | 2179 | |
d7df2c79 | 2180 | spin_lock(&delayed_refs->lock); |
5c9d028b | 2181 | node = rb_first_cached(&delayed_refs->href_root); |
d7df2c79 JB |
2182 | if (!node) { |
2183 | spin_unlock(&delayed_refs->lock); | |
56bec294 | 2184 | goto out; |
d7df2c79 | 2185 | } |
d278850e JB |
2186 | head = rb_entry(node, struct btrfs_delayed_ref_head, |
2187 | href_node); | |
2188 | refcount_inc(&head->refs); | |
2189 | spin_unlock(&delayed_refs->lock); | |
e9d0b13b | 2190 | |
d278850e JB |
2191 | /* Mutex was contended, block until it's released and retry. */ |
2192 | mutex_lock(&head->mutex); | |
2193 | mutex_unlock(&head->mutex); | |
56bec294 | 2194 | |
d278850e | 2195 | btrfs_put_delayed_ref_head(head); |
d7df2c79 | 2196 | cond_resched(); |
56bec294 | 2197 | goto again; |
5f39d397 | 2198 | } |
54aa1f4d | 2199 | out: |
a28ec197 CM |
2200 | return 0; |
2201 | } | |
2202 | ||
5d4f98a2 | 2203 | int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans, |
42c9d0b5 | 2204 | struct extent_buffer *eb, u64 flags, |
b1c79e09 | 2205 | int level, int is_data) |
5d4f98a2 YZ |
2206 | { |
2207 | struct btrfs_delayed_extent_op *extent_op; | |
2208 | int ret; | |
2209 | ||
78a6184a | 2210 | extent_op = btrfs_alloc_delayed_extent_op(); |
5d4f98a2 YZ |
2211 | if (!extent_op) |
2212 | return -ENOMEM; | |
2213 | ||
2214 | extent_op->flags_to_set = flags; | |
35b3ad50 DS |
2215 | extent_op->update_flags = true; |
2216 | extent_op->update_key = false; | |
2217 | extent_op->is_data = is_data ? true : false; | |
b1c79e09 | 2218 | extent_op->level = level; |
5d4f98a2 | 2219 | |
42c9d0b5 | 2220 | ret = btrfs_add_delayed_extent_op(trans, eb->start, eb->len, extent_op); |
5d4f98a2 | 2221 | if (ret) |
78a6184a | 2222 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 YZ |
2223 | return ret; |
2224 | } | |
2225 | ||
e4c3b2dc | 2226 | static noinline int check_delayed_ref(struct btrfs_root *root, |
5d4f98a2 YZ |
2227 | struct btrfs_path *path, |
2228 | u64 objectid, u64 offset, u64 bytenr) | |
2229 | { | |
2230 | struct btrfs_delayed_ref_head *head; | |
2231 | struct btrfs_delayed_ref_node *ref; | |
2232 | struct btrfs_delayed_data_ref *data_ref; | |
2233 | struct btrfs_delayed_ref_root *delayed_refs; | |
e4c3b2dc | 2234 | struct btrfs_transaction *cur_trans; |
0e0adbcf | 2235 | struct rb_node *node; |
5d4f98a2 YZ |
2236 | int ret = 0; |
2237 | ||
998ac6d2 | 2238 | spin_lock(&root->fs_info->trans_lock); |
e4c3b2dc | 2239 | cur_trans = root->fs_info->running_transaction; |
998ac6d2 | 2240 | if (cur_trans) |
2241 | refcount_inc(&cur_trans->use_count); | |
2242 | spin_unlock(&root->fs_info->trans_lock); | |
e4c3b2dc LB |
2243 | if (!cur_trans) |
2244 | return 0; | |
2245 | ||
2246 | delayed_refs = &cur_trans->delayed_refs; | |
5d4f98a2 | 2247 | spin_lock(&delayed_refs->lock); |
f72ad18e | 2248 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
d7df2c79 JB |
2249 | if (!head) { |
2250 | spin_unlock(&delayed_refs->lock); | |
998ac6d2 | 2251 | btrfs_put_transaction(cur_trans); |
d7df2c79 JB |
2252 | return 0; |
2253 | } | |
5d4f98a2 YZ |
2254 | |
2255 | if (!mutex_trylock(&head->mutex)) { | |
d278850e | 2256 | refcount_inc(&head->refs); |
5d4f98a2 YZ |
2257 | spin_unlock(&delayed_refs->lock); |
2258 | ||
b3b4aa74 | 2259 | btrfs_release_path(path); |
5d4f98a2 | 2260 | |
8cc33e5c DS |
2261 | /* |
2262 | * Mutex was contended, block until it's released and let | |
2263 | * caller try again | |
2264 | */ | |
5d4f98a2 YZ |
2265 | mutex_lock(&head->mutex); |
2266 | mutex_unlock(&head->mutex); | |
d278850e | 2267 | btrfs_put_delayed_ref_head(head); |
998ac6d2 | 2268 | btrfs_put_transaction(cur_trans); |
5d4f98a2 YZ |
2269 | return -EAGAIN; |
2270 | } | |
d7df2c79 | 2271 | spin_unlock(&delayed_refs->lock); |
5d4f98a2 | 2272 | |
d7df2c79 | 2273 | spin_lock(&head->lock); |
0e0adbcf JB |
2274 | /* |
2275 | * XXX: We should replace this with a proper search function in the | |
2276 | * future. | |
2277 | */ | |
e3d03965 LB |
2278 | for (node = rb_first_cached(&head->ref_tree); node; |
2279 | node = rb_next(node)) { | |
0e0adbcf | 2280 | ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node); |
d7df2c79 JB |
2281 | /* If it's a shared ref we know a cross reference exists */ |
2282 | if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) { | |
2283 | ret = 1; | |
2284 | break; | |
2285 | } | |
5d4f98a2 | 2286 | |
d7df2c79 | 2287 | data_ref = btrfs_delayed_node_to_data_ref(ref); |
5d4f98a2 | 2288 | |
d7df2c79 JB |
2289 | /* |
2290 | * If our ref doesn't match the one we're currently looking at | |
2291 | * then we have a cross reference. | |
2292 | */ | |
2293 | if (data_ref->root != root->root_key.objectid || | |
2294 | data_ref->objectid != objectid || | |
2295 | data_ref->offset != offset) { | |
2296 | ret = 1; | |
2297 | break; | |
2298 | } | |
5d4f98a2 | 2299 | } |
d7df2c79 | 2300 | spin_unlock(&head->lock); |
5d4f98a2 | 2301 | mutex_unlock(&head->mutex); |
998ac6d2 | 2302 | btrfs_put_transaction(cur_trans); |
5d4f98a2 YZ |
2303 | return ret; |
2304 | } | |
2305 | ||
e4c3b2dc | 2306 | static noinline int check_committed_ref(struct btrfs_root *root, |
5d4f98a2 | 2307 | struct btrfs_path *path, |
a84d5d42 BB |
2308 | u64 objectid, u64 offset, u64 bytenr, |
2309 | bool strict) | |
be20aa9d | 2310 | { |
0b246afa JM |
2311 | struct btrfs_fs_info *fs_info = root->fs_info; |
2312 | struct btrfs_root *extent_root = fs_info->extent_root; | |
f321e491 | 2313 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
2314 | struct btrfs_extent_data_ref *ref; |
2315 | struct btrfs_extent_inline_ref *iref; | |
2316 | struct btrfs_extent_item *ei; | |
f321e491 | 2317 | struct btrfs_key key; |
5d4f98a2 | 2318 | u32 item_size; |
3de28d57 | 2319 | int type; |
be20aa9d | 2320 | int ret; |
925baedd | 2321 | |
be20aa9d | 2322 | key.objectid = bytenr; |
31840ae1 | 2323 | key.offset = (u64)-1; |
f321e491 | 2324 | key.type = BTRFS_EXTENT_ITEM_KEY; |
be20aa9d | 2325 | |
be20aa9d CM |
2326 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
2327 | if (ret < 0) | |
2328 | goto out; | |
79787eaa | 2329 | BUG_ON(ret == 0); /* Corruption */ |
80ff3856 YZ |
2330 | |
2331 | ret = -ENOENT; | |
2332 | if (path->slots[0] == 0) | |
31840ae1 | 2333 | goto out; |
be20aa9d | 2334 | |
31840ae1 | 2335 | path->slots[0]--; |
f321e491 | 2336 | leaf = path->nodes[0]; |
5d4f98a2 | 2337 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
be20aa9d | 2338 | |
5d4f98a2 | 2339 | if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY) |
be20aa9d | 2340 | goto out; |
f321e491 | 2341 | |
5d4f98a2 YZ |
2342 | ret = 1; |
2343 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
5d4f98a2 | 2344 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
bd09835d | 2345 | |
a6bd9cd1 | 2346 | /* If extent item has more than 1 inline ref then it's shared */ |
5d4f98a2 YZ |
2347 | if (item_size != sizeof(*ei) + |
2348 | btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY)) | |
2349 | goto out; | |
be20aa9d | 2350 | |
a84d5d42 BB |
2351 | /* |
2352 | * If extent created before last snapshot => it's shared unless the | |
2353 | * snapshot has been deleted. Use the heuristic if strict is false. | |
2354 | */ | |
2355 | if (!strict && | |
2356 | (btrfs_extent_generation(leaf, ei) <= | |
2357 | btrfs_root_last_snapshot(&root->root_item))) | |
5d4f98a2 YZ |
2358 | goto out; |
2359 | ||
2360 | iref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
3de28d57 | 2361 | |
a6bd9cd1 | 2362 | /* If this extent has SHARED_DATA_REF then it's shared */ |
3de28d57 LB |
2363 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA); |
2364 | if (type != BTRFS_EXTENT_DATA_REF_KEY) | |
5d4f98a2 YZ |
2365 | goto out; |
2366 | ||
2367 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
2368 | if (btrfs_extent_refs(leaf, ei) != | |
2369 | btrfs_extent_data_ref_count(leaf, ref) || | |
2370 | btrfs_extent_data_ref_root(leaf, ref) != | |
2371 | root->root_key.objectid || | |
2372 | btrfs_extent_data_ref_objectid(leaf, ref) != objectid || | |
2373 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
2374 | goto out; | |
2375 | ||
2376 | ret = 0; | |
2377 | out: | |
2378 | return ret; | |
2379 | } | |
2380 | ||
e4c3b2dc | 2381 | int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset, |
a84d5d42 | 2382 | u64 bytenr, bool strict) |
5d4f98a2 YZ |
2383 | { |
2384 | struct btrfs_path *path; | |
2385 | int ret; | |
5d4f98a2 YZ |
2386 | |
2387 | path = btrfs_alloc_path(); | |
2388 | if (!path) | |
9132c4ff | 2389 | return -ENOMEM; |
5d4f98a2 YZ |
2390 | |
2391 | do { | |
e4c3b2dc | 2392 | ret = check_committed_ref(root, path, objectid, |
a84d5d42 | 2393 | offset, bytenr, strict); |
5d4f98a2 | 2394 | if (ret && ret != -ENOENT) |
f321e491 | 2395 | goto out; |
80ff3856 | 2396 | |
380fd066 MT |
2397 | ret = check_delayed_ref(root, path, objectid, offset, bytenr); |
2398 | } while (ret == -EAGAIN); | |
5d4f98a2 | 2399 | |
be20aa9d | 2400 | out: |
80ff3856 | 2401 | btrfs_free_path(path); |
f0486c68 YZ |
2402 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) |
2403 | WARN_ON(ret > 0); | |
f321e491 | 2404 | return ret; |
be20aa9d | 2405 | } |
c5739bba | 2406 | |
5d4f98a2 | 2407 | static int __btrfs_mod_ref(struct btrfs_trans_handle *trans, |
b7a9f29f | 2408 | struct btrfs_root *root, |
5d4f98a2 | 2409 | struct extent_buffer *buf, |
e339a6b0 | 2410 | int full_backref, int inc) |
31840ae1 | 2411 | { |
0b246afa | 2412 | struct btrfs_fs_info *fs_info = root->fs_info; |
31840ae1 | 2413 | u64 bytenr; |
5d4f98a2 YZ |
2414 | u64 num_bytes; |
2415 | u64 parent; | |
31840ae1 | 2416 | u64 ref_root; |
31840ae1 | 2417 | u32 nritems; |
31840ae1 ZY |
2418 | struct btrfs_key key; |
2419 | struct btrfs_file_extent_item *fi; | |
82fa113f QW |
2420 | struct btrfs_ref generic_ref = { 0 }; |
2421 | bool for_reloc = btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC); | |
31840ae1 | 2422 | int i; |
82fa113f | 2423 | int action; |
31840ae1 ZY |
2424 | int level; |
2425 | int ret = 0; | |
fccb84c9 | 2426 | |
0b246afa | 2427 | if (btrfs_is_testing(fs_info)) |
faa2dbf0 | 2428 | return 0; |
fccb84c9 | 2429 | |
31840ae1 | 2430 | ref_root = btrfs_header_owner(buf); |
31840ae1 ZY |
2431 | nritems = btrfs_header_nritems(buf); |
2432 | level = btrfs_header_level(buf); | |
2433 | ||
92a7cc42 | 2434 | if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && level == 0) |
5d4f98a2 | 2435 | return 0; |
31840ae1 | 2436 | |
5d4f98a2 YZ |
2437 | if (full_backref) |
2438 | parent = buf->start; | |
2439 | else | |
2440 | parent = 0; | |
82fa113f QW |
2441 | if (inc) |
2442 | action = BTRFS_ADD_DELAYED_REF; | |
2443 | else | |
2444 | action = BTRFS_DROP_DELAYED_REF; | |
5d4f98a2 YZ |
2445 | |
2446 | for (i = 0; i < nritems; i++) { | |
31840ae1 | 2447 | if (level == 0) { |
5d4f98a2 | 2448 | btrfs_item_key_to_cpu(buf, &key, i); |
962a298f | 2449 | if (key.type != BTRFS_EXTENT_DATA_KEY) |
31840ae1 | 2450 | continue; |
5d4f98a2 | 2451 | fi = btrfs_item_ptr(buf, i, |
31840ae1 ZY |
2452 | struct btrfs_file_extent_item); |
2453 | if (btrfs_file_extent_type(buf, fi) == | |
2454 | BTRFS_FILE_EXTENT_INLINE) | |
2455 | continue; | |
2456 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
2457 | if (bytenr == 0) | |
2458 | continue; | |
5d4f98a2 YZ |
2459 | |
2460 | num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi); | |
2461 | key.offset -= btrfs_file_extent_offset(buf, fi); | |
82fa113f QW |
2462 | btrfs_init_generic_ref(&generic_ref, action, bytenr, |
2463 | num_bytes, parent); | |
2464 | generic_ref.real_root = root->root_key.objectid; | |
2465 | btrfs_init_data_ref(&generic_ref, ref_root, key.objectid, | |
2466 | key.offset); | |
2467 | generic_ref.skip_qgroup = for_reloc; | |
dd28b6a5 | 2468 | if (inc) |
82fa113f | 2469 | ret = btrfs_inc_extent_ref(trans, &generic_ref); |
dd28b6a5 | 2470 | else |
ffd4bb2a | 2471 | ret = btrfs_free_extent(trans, &generic_ref); |
31840ae1 ZY |
2472 | if (ret) |
2473 | goto fail; | |
2474 | } else { | |
5d4f98a2 | 2475 | bytenr = btrfs_node_blockptr(buf, i); |
0b246afa | 2476 | num_bytes = fs_info->nodesize; |
82fa113f QW |
2477 | btrfs_init_generic_ref(&generic_ref, action, bytenr, |
2478 | num_bytes, parent); | |
2479 | generic_ref.real_root = root->root_key.objectid; | |
2480 | btrfs_init_tree_ref(&generic_ref, level - 1, ref_root); | |
2481 | generic_ref.skip_qgroup = for_reloc; | |
dd28b6a5 | 2482 | if (inc) |
82fa113f | 2483 | ret = btrfs_inc_extent_ref(trans, &generic_ref); |
dd28b6a5 | 2484 | else |
ffd4bb2a | 2485 | ret = btrfs_free_extent(trans, &generic_ref); |
31840ae1 ZY |
2486 | if (ret) |
2487 | goto fail; | |
2488 | } | |
2489 | } | |
2490 | return 0; | |
2491 | fail: | |
5d4f98a2 YZ |
2492 | return ret; |
2493 | } | |
2494 | ||
2495 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 2496 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 2497 | { |
e339a6b0 | 2498 | return __btrfs_mod_ref(trans, root, buf, full_backref, 1); |
5d4f98a2 YZ |
2499 | } |
2500 | ||
2501 | int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 2502 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 2503 | { |
e339a6b0 | 2504 | return __btrfs_mod_ref(trans, root, buf, full_backref, 0); |
31840ae1 ZY |
2505 | } |
2506 | ||
2ff7e61e | 2507 | int btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr) |
d2fb3437 | 2508 | { |
32da5386 | 2509 | struct btrfs_block_group *block_group; |
d2fb3437 YZ |
2510 | int readonly = 0; |
2511 | ||
0b246afa | 2512 | block_group = btrfs_lookup_block_group(fs_info, bytenr); |
d2fb3437 YZ |
2513 | if (!block_group || block_group->ro) |
2514 | readonly = 1; | |
2515 | if (block_group) | |
fa9c0d79 | 2516 | btrfs_put_block_group(block_group); |
d2fb3437 YZ |
2517 | return readonly; |
2518 | } | |
2519 | ||
1b86826d | 2520 | static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data) |
9ed74f2d | 2521 | { |
0b246afa | 2522 | struct btrfs_fs_info *fs_info = root->fs_info; |
b742bb82 | 2523 | u64 flags; |
53b381b3 | 2524 | u64 ret; |
9ed74f2d | 2525 | |
b742bb82 YZ |
2526 | if (data) |
2527 | flags = BTRFS_BLOCK_GROUP_DATA; | |
0b246afa | 2528 | else if (root == fs_info->chunk_root) |
b742bb82 | 2529 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
9ed74f2d | 2530 | else |
b742bb82 | 2531 | flags = BTRFS_BLOCK_GROUP_METADATA; |
9ed74f2d | 2532 | |
878d7b67 | 2533 | ret = btrfs_get_alloc_profile(fs_info, flags); |
53b381b3 | 2534 | return ret; |
6a63209f | 2535 | } |
9ed74f2d | 2536 | |
2ff7e61e | 2537 | static u64 first_logical_byte(struct btrfs_fs_info *fs_info, u64 search_start) |
a061fc8d | 2538 | { |
32da5386 | 2539 | struct btrfs_block_group *cache; |
d2fb3437 | 2540 | u64 bytenr; |
0f9dd46c | 2541 | |
0b246afa JM |
2542 | spin_lock(&fs_info->block_group_cache_lock); |
2543 | bytenr = fs_info->first_logical_byte; | |
2544 | spin_unlock(&fs_info->block_group_cache_lock); | |
a1897fdd LB |
2545 | |
2546 | if (bytenr < (u64)-1) | |
2547 | return bytenr; | |
2548 | ||
0b246afa | 2549 | cache = btrfs_lookup_first_block_group(fs_info, search_start); |
0f9dd46c | 2550 | if (!cache) |
a061fc8d | 2551 | return 0; |
0f9dd46c | 2552 | |
b3470b5d | 2553 | bytenr = cache->start; |
fa9c0d79 | 2554 | btrfs_put_block_group(cache); |
d2fb3437 YZ |
2555 | |
2556 | return bytenr; | |
a061fc8d CM |
2557 | } |
2558 | ||
6690d071 NB |
2559 | static int pin_down_extent(struct btrfs_trans_handle *trans, |
2560 | struct btrfs_block_group *cache, | |
f0486c68 | 2561 | u64 bytenr, u64 num_bytes, int reserved) |
324ae4df | 2562 | { |
fdf08605 DS |
2563 | struct btrfs_fs_info *fs_info = cache->fs_info; |
2564 | ||
11833d66 YZ |
2565 | spin_lock(&cache->space_info->lock); |
2566 | spin_lock(&cache->lock); | |
2567 | cache->pinned += num_bytes; | |
bb96c4e5 JB |
2568 | btrfs_space_info_update_bytes_pinned(fs_info, cache->space_info, |
2569 | num_bytes); | |
11833d66 YZ |
2570 | if (reserved) { |
2571 | cache->reserved -= num_bytes; | |
2572 | cache->space_info->bytes_reserved -= num_bytes; | |
2573 | } | |
2574 | spin_unlock(&cache->lock); | |
2575 | spin_unlock(&cache->space_info->lock); | |
68b38550 | 2576 | |
dec59fa3 EL |
2577 | percpu_counter_add_batch(&cache->space_info->total_bytes_pinned, |
2578 | num_bytes, BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
fe119a6e | 2579 | set_extent_dirty(&trans->transaction->pinned_extents, bytenr, |
f0486c68 YZ |
2580 | bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL); |
2581 | return 0; | |
2582 | } | |
68b38550 | 2583 | |
b25c36f8 | 2584 | int btrfs_pin_extent(struct btrfs_trans_handle *trans, |
f0486c68 YZ |
2585 | u64 bytenr, u64 num_bytes, int reserved) |
2586 | { | |
32da5386 | 2587 | struct btrfs_block_group *cache; |
68b38550 | 2588 | |
b25c36f8 | 2589 | cache = btrfs_lookup_block_group(trans->fs_info, bytenr); |
79787eaa | 2590 | BUG_ON(!cache); /* Logic error */ |
f0486c68 | 2591 | |
6690d071 | 2592 | pin_down_extent(trans, cache, bytenr, num_bytes, reserved); |
f0486c68 YZ |
2593 | |
2594 | btrfs_put_block_group(cache); | |
11833d66 YZ |
2595 | return 0; |
2596 | } | |
2597 | ||
f0486c68 | 2598 | /* |
e688b725 CM |
2599 | * this function must be called within transaction |
2600 | */ | |
9fce5704 | 2601 | int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans, |
e688b725 CM |
2602 | u64 bytenr, u64 num_bytes) |
2603 | { | |
32da5386 | 2604 | struct btrfs_block_group *cache; |
b50c6e25 | 2605 | int ret; |
e688b725 | 2606 | |
f2fb7298 NB |
2607 | btrfs_add_excluded_extent(trans->fs_info, bytenr, num_bytes); |
2608 | ||
9fce5704 | 2609 | cache = btrfs_lookup_block_group(trans->fs_info, bytenr); |
b50c6e25 JB |
2610 | if (!cache) |
2611 | return -EINVAL; | |
e688b725 CM |
2612 | |
2613 | /* | |
2614 | * pull in the free space cache (if any) so that our pin | |
2615 | * removes the free space from the cache. We have load_only set | |
2616 | * to one because the slow code to read in the free extents does check | |
2617 | * the pinned extents. | |
2618 | */ | |
676f1f75 | 2619 | btrfs_cache_block_group(cache, 1); |
e688b725 | 2620 | |
6690d071 | 2621 | pin_down_extent(trans, cache, bytenr, num_bytes, 0); |
e688b725 CM |
2622 | |
2623 | /* remove us from the free space cache (if we're there at all) */ | |
b50c6e25 | 2624 | ret = btrfs_remove_free_space(cache, bytenr, num_bytes); |
e688b725 | 2625 | btrfs_put_block_group(cache); |
b50c6e25 | 2626 | return ret; |
e688b725 CM |
2627 | } |
2628 | ||
2ff7e61e JM |
2629 | static int __exclude_logged_extent(struct btrfs_fs_info *fs_info, |
2630 | u64 start, u64 num_bytes) | |
8c2a1a30 JB |
2631 | { |
2632 | int ret; | |
32da5386 | 2633 | struct btrfs_block_group *block_group; |
8c2a1a30 JB |
2634 | struct btrfs_caching_control *caching_ctl; |
2635 | ||
0b246afa | 2636 | block_group = btrfs_lookup_block_group(fs_info, start); |
8c2a1a30 JB |
2637 | if (!block_group) |
2638 | return -EINVAL; | |
2639 | ||
676f1f75 | 2640 | btrfs_cache_block_group(block_group, 0); |
e3cb339f | 2641 | caching_ctl = btrfs_get_caching_control(block_group); |
8c2a1a30 JB |
2642 | |
2643 | if (!caching_ctl) { | |
2644 | /* Logic error */ | |
32da5386 | 2645 | BUG_ON(!btrfs_block_group_done(block_group)); |
8c2a1a30 JB |
2646 | ret = btrfs_remove_free_space(block_group, start, num_bytes); |
2647 | } else { | |
2648 | mutex_lock(&caching_ctl->mutex); | |
2649 | ||
2650 | if (start >= caching_ctl->progress) { | |
6f410d1b JB |
2651 | ret = btrfs_add_excluded_extent(fs_info, start, |
2652 | num_bytes); | |
8c2a1a30 JB |
2653 | } else if (start + num_bytes <= caching_ctl->progress) { |
2654 | ret = btrfs_remove_free_space(block_group, | |
2655 | start, num_bytes); | |
2656 | } else { | |
2657 | num_bytes = caching_ctl->progress - start; | |
2658 | ret = btrfs_remove_free_space(block_group, | |
2659 | start, num_bytes); | |
2660 | if (ret) | |
2661 | goto out_lock; | |
2662 | ||
2663 | num_bytes = (start + num_bytes) - | |
2664 | caching_ctl->progress; | |
2665 | start = caching_ctl->progress; | |
6f410d1b JB |
2666 | ret = btrfs_add_excluded_extent(fs_info, start, |
2667 | num_bytes); | |
8c2a1a30 JB |
2668 | } |
2669 | out_lock: | |
2670 | mutex_unlock(&caching_ctl->mutex); | |
e3cb339f | 2671 | btrfs_put_caching_control(caching_ctl); |
8c2a1a30 JB |
2672 | } |
2673 | btrfs_put_block_group(block_group); | |
2674 | return ret; | |
2675 | } | |
2676 | ||
bcdc428c | 2677 | int btrfs_exclude_logged_extents(struct extent_buffer *eb) |
8c2a1a30 | 2678 | { |
bcdc428c | 2679 | struct btrfs_fs_info *fs_info = eb->fs_info; |
8c2a1a30 JB |
2680 | struct btrfs_file_extent_item *item; |
2681 | struct btrfs_key key; | |
2682 | int found_type; | |
2683 | int i; | |
b89311ef | 2684 | int ret = 0; |
8c2a1a30 | 2685 | |
2ff7e61e | 2686 | if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) |
8c2a1a30 JB |
2687 | return 0; |
2688 | ||
2689 | for (i = 0; i < btrfs_header_nritems(eb); i++) { | |
2690 | btrfs_item_key_to_cpu(eb, &key, i); | |
2691 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
2692 | continue; | |
2693 | item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item); | |
2694 | found_type = btrfs_file_extent_type(eb, item); | |
2695 | if (found_type == BTRFS_FILE_EXTENT_INLINE) | |
2696 | continue; | |
2697 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0) | |
2698 | continue; | |
2699 | key.objectid = btrfs_file_extent_disk_bytenr(eb, item); | |
2700 | key.offset = btrfs_file_extent_disk_num_bytes(eb, item); | |
b89311ef GJ |
2701 | ret = __exclude_logged_extent(fs_info, key.objectid, key.offset); |
2702 | if (ret) | |
2703 | break; | |
8c2a1a30 JB |
2704 | } |
2705 | ||
b89311ef | 2706 | return ret; |
8c2a1a30 JB |
2707 | } |
2708 | ||
9cfa3e34 | 2709 | static void |
32da5386 | 2710 | btrfs_inc_block_group_reservations(struct btrfs_block_group *bg) |
9cfa3e34 FM |
2711 | { |
2712 | atomic_inc(&bg->reservations); | |
2713 | } | |
2714 | ||
8b74c03e | 2715 | void btrfs_prepare_extent_commit(struct btrfs_fs_info *fs_info) |
e8569813 | 2716 | { |
11833d66 YZ |
2717 | struct btrfs_caching_control *next; |
2718 | struct btrfs_caching_control *caching_ctl; | |
32da5386 | 2719 | struct btrfs_block_group *cache; |
e8569813 | 2720 | |
9e351cc8 | 2721 | down_write(&fs_info->commit_root_sem); |
25179201 | 2722 | |
11833d66 YZ |
2723 | list_for_each_entry_safe(caching_ctl, next, |
2724 | &fs_info->caching_block_groups, list) { | |
2725 | cache = caching_ctl->block_group; | |
32da5386 | 2726 | if (btrfs_block_group_done(cache)) { |
11833d66 YZ |
2727 | cache->last_byte_to_unpin = (u64)-1; |
2728 | list_del_init(&caching_ctl->list); | |
e3cb339f | 2729 | btrfs_put_caching_control(caching_ctl); |
e8569813 | 2730 | } else { |
11833d66 | 2731 | cache->last_byte_to_unpin = caching_ctl->progress; |
e8569813 | 2732 | } |
e8569813 | 2733 | } |
11833d66 | 2734 | |
9e351cc8 | 2735 | up_write(&fs_info->commit_root_sem); |
8929ecfa | 2736 | |
67f9c220 | 2737 | btrfs_update_global_block_rsv(fs_info); |
e8569813 ZY |
2738 | } |
2739 | ||
c759c4e1 JB |
2740 | /* |
2741 | * Returns the free cluster for the given space info and sets empty_cluster to | |
2742 | * what it should be based on the mount options. | |
2743 | */ | |
2744 | static struct btrfs_free_cluster * | |
2ff7e61e JM |
2745 | fetch_cluster_info(struct btrfs_fs_info *fs_info, |
2746 | struct btrfs_space_info *space_info, u64 *empty_cluster) | |
c759c4e1 JB |
2747 | { |
2748 | struct btrfs_free_cluster *ret = NULL; | |
c759c4e1 JB |
2749 | |
2750 | *empty_cluster = 0; | |
2751 | if (btrfs_mixed_space_info(space_info)) | |
2752 | return ret; | |
2753 | ||
c759c4e1 | 2754 | if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) { |
0b246afa | 2755 | ret = &fs_info->meta_alloc_cluster; |
583b7231 HK |
2756 | if (btrfs_test_opt(fs_info, SSD)) |
2757 | *empty_cluster = SZ_2M; | |
2758 | else | |
ee22184b | 2759 | *empty_cluster = SZ_64K; |
583b7231 HK |
2760 | } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) && |
2761 | btrfs_test_opt(fs_info, SSD_SPREAD)) { | |
2762 | *empty_cluster = SZ_2M; | |
0b246afa | 2763 | ret = &fs_info->data_alloc_cluster; |
c759c4e1 JB |
2764 | } |
2765 | ||
2766 | return ret; | |
2767 | } | |
2768 | ||
2ff7e61e JM |
2769 | static int unpin_extent_range(struct btrfs_fs_info *fs_info, |
2770 | u64 start, u64 end, | |
678886bd | 2771 | const bool return_free_space) |
ccd467d6 | 2772 | { |
32da5386 | 2773 | struct btrfs_block_group *cache = NULL; |
7b398f8e JB |
2774 | struct btrfs_space_info *space_info; |
2775 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
c759c4e1 | 2776 | struct btrfs_free_cluster *cluster = NULL; |
11833d66 | 2777 | u64 len; |
c759c4e1 JB |
2778 | u64 total_unpinned = 0; |
2779 | u64 empty_cluster = 0; | |
7b398f8e | 2780 | bool readonly; |
ccd467d6 | 2781 | |
11833d66 | 2782 | while (start <= end) { |
7b398f8e | 2783 | readonly = false; |
11833d66 | 2784 | if (!cache || |
b3470b5d | 2785 | start >= cache->start + cache->length) { |
11833d66 YZ |
2786 | if (cache) |
2787 | btrfs_put_block_group(cache); | |
c759c4e1 | 2788 | total_unpinned = 0; |
11833d66 | 2789 | cache = btrfs_lookup_block_group(fs_info, start); |
79787eaa | 2790 | BUG_ON(!cache); /* Logic error */ |
c759c4e1 | 2791 | |
2ff7e61e | 2792 | cluster = fetch_cluster_info(fs_info, |
c759c4e1 JB |
2793 | cache->space_info, |
2794 | &empty_cluster); | |
2795 | empty_cluster <<= 1; | |
11833d66 YZ |
2796 | } |
2797 | ||
b3470b5d | 2798 | len = cache->start + cache->length - start; |
11833d66 YZ |
2799 | len = min(len, end + 1 - start); |
2800 | ||
2801 | if (start < cache->last_byte_to_unpin) { | |
2802 | len = min(len, cache->last_byte_to_unpin - start); | |
678886bd FM |
2803 | if (return_free_space) |
2804 | btrfs_add_free_space(cache, start, len); | |
11833d66 YZ |
2805 | } |
2806 | ||
f0486c68 | 2807 | start += len; |
c759c4e1 | 2808 | total_unpinned += len; |
7b398f8e | 2809 | space_info = cache->space_info; |
f0486c68 | 2810 | |
c759c4e1 JB |
2811 | /* |
2812 | * If this space cluster has been marked as fragmented and we've | |
2813 | * unpinned enough in this block group to potentially allow a | |
2814 | * cluster to be created inside of it go ahead and clear the | |
2815 | * fragmented check. | |
2816 | */ | |
2817 | if (cluster && cluster->fragmented && | |
2818 | total_unpinned > empty_cluster) { | |
2819 | spin_lock(&cluster->lock); | |
2820 | cluster->fragmented = 0; | |
2821 | spin_unlock(&cluster->lock); | |
2822 | } | |
2823 | ||
7b398f8e | 2824 | spin_lock(&space_info->lock); |
11833d66 YZ |
2825 | spin_lock(&cache->lock); |
2826 | cache->pinned -= len; | |
bb96c4e5 | 2827 | btrfs_space_info_update_bytes_pinned(fs_info, space_info, -len); |
4f4db217 | 2828 | space_info->max_extent_size = 0; |
dec59fa3 EL |
2829 | percpu_counter_add_batch(&space_info->total_bytes_pinned, |
2830 | -len, BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
7b398f8e JB |
2831 | if (cache->ro) { |
2832 | space_info->bytes_readonly += len; | |
2833 | readonly = true; | |
2834 | } | |
11833d66 | 2835 | spin_unlock(&cache->lock); |
957780eb JB |
2836 | if (!readonly && return_free_space && |
2837 | global_rsv->space_info == space_info) { | |
2838 | u64 to_add = len; | |
92ac58ec | 2839 | |
7b398f8e JB |
2840 | spin_lock(&global_rsv->lock); |
2841 | if (!global_rsv->full) { | |
957780eb JB |
2842 | to_add = min(len, global_rsv->size - |
2843 | global_rsv->reserved); | |
2844 | global_rsv->reserved += to_add; | |
bb96c4e5 JB |
2845 | btrfs_space_info_update_bytes_may_use(fs_info, |
2846 | space_info, to_add); | |
7b398f8e JB |
2847 | if (global_rsv->reserved >= global_rsv->size) |
2848 | global_rsv->full = 1; | |
957780eb | 2849 | len -= to_add; |
7b398f8e JB |
2850 | } |
2851 | spin_unlock(&global_rsv->lock); | |
957780eb JB |
2852 | /* Add to any tickets we may have */ |
2853 | if (len) | |
18fa2284 JB |
2854 | btrfs_try_granting_tickets(fs_info, |
2855 | space_info); | |
7b398f8e JB |
2856 | } |
2857 | spin_unlock(&space_info->lock); | |
ccd467d6 | 2858 | } |
11833d66 YZ |
2859 | |
2860 | if (cache) | |
2861 | btrfs_put_block_group(cache); | |
ccd467d6 CM |
2862 | return 0; |
2863 | } | |
2864 | ||
5ead2dd0 | 2865 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans) |
a28ec197 | 2866 | { |
5ead2dd0 | 2867 | struct btrfs_fs_info *fs_info = trans->fs_info; |
32da5386 | 2868 | struct btrfs_block_group *block_group, *tmp; |
e33e17ee | 2869 | struct list_head *deleted_bgs; |
11833d66 | 2870 | struct extent_io_tree *unpin; |
1a5bc167 CM |
2871 | u64 start; |
2872 | u64 end; | |
a28ec197 | 2873 | int ret; |
a28ec197 | 2874 | |
fe119a6e | 2875 | unpin = &trans->transaction->pinned_extents; |
11833d66 | 2876 | |
bf31f87f | 2877 | while (!TRANS_ABORTED(trans)) { |
0e6ec385 FM |
2878 | struct extent_state *cached_state = NULL; |
2879 | ||
d4b450cd | 2880 | mutex_lock(&fs_info->unused_bg_unpin_mutex); |
1a5bc167 | 2881 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
0e6ec385 | 2882 | EXTENT_DIRTY, &cached_state); |
d4b450cd FM |
2883 | if (ret) { |
2884 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); | |
a28ec197 | 2885 | break; |
d4b450cd | 2886 | } |
fe119a6e NB |
2887 | if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) |
2888 | clear_extent_bits(&fs_info->excluded_extents, start, | |
f2fb7298 | 2889 | end, EXTENT_UPTODATE); |
1f3c79a2 | 2890 | |
46b27f50 | 2891 | if (btrfs_test_opt(fs_info, DISCARD_SYNC)) |
2ff7e61e | 2892 | ret = btrfs_discard_extent(fs_info, start, |
5378e607 | 2893 | end + 1 - start, NULL); |
1f3c79a2 | 2894 | |
0e6ec385 | 2895 | clear_extent_dirty(unpin, start, end, &cached_state); |
2ff7e61e | 2896 | unpin_extent_range(fs_info, start, end, true); |
d4b450cd | 2897 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
0e6ec385 | 2898 | free_extent_state(cached_state); |
b9473439 | 2899 | cond_resched(); |
a28ec197 | 2900 | } |
817d52f8 | 2901 | |
a2309300 DZ |
2902 | if (btrfs_test_opt(fs_info, DISCARD_ASYNC)) { |
2903 | btrfs_discard_calc_delay(&fs_info->discard_ctl); | |
b0643e59 | 2904 | btrfs_discard_schedule_work(&fs_info->discard_ctl, true); |
a2309300 | 2905 | } |
b0643e59 | 2906 | |
e33e17ee JM |
2907 | /* |
2908 | * Transaction is finished. We don't need the lock anymore. We | |
2909 | * do need to clean up the block groups in case of a transaction | |
2910 | * abort. | |
2911 | */ | |
2912 | deleted_bgs = &trans->transaction->deleted_bgs; | |
2913 | list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) { | |
2914 | u64 trimmed = 0; | |
2915 | ||
2916 | ret = -EROFS; | |
bf31f87f | 2917 | if (!TRANS_ABORTED(trans)) |
2ff7e61e | 2918 | ret = btrfs_discard_extent(fs_info, |
b3470b5d DS |
2919 | block_group->start, |
2920 | block_group->length, | |
e33e17ee JM |
2921 | &trimmed); |
2922 | ||
2923 | list_del_init(&block_group->bg_list); | |
6b7304af | 2924 | btrfs_unfreeze_block_group(block_group); |
e33e17ee JM |
2925 | btrfs_put_block_group(block_group); |
2926 | ||
2927 | if (ret) { | |
2928 | const char *errstr = btrfs_decode_error(ret); | |
2929 | btrfs_warn(fs_info, | |
913e1535 | 2930 | "discard failed while removing blockgroup: errno=%d %s", |
e33e17ee JM |
2931 | ret, errstr); |
2932 | } | |
2933 | } | |
2934 | ||
e20d96d6 CM |
2935 | return 0; |
2936 | } | |
2937 | ||
5d4f98a2 | 2938 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
e72cb923 NB |
2939 | struct btrfs_delayed_ref_node *node, u64 parent, |
2940 | u64 root_objectid, u64 owner_objectid, | |
2941 | u64 owner_offset, int refs_to_drop, | |
2942 | struct btrfs_delayed_extent_op *extent_op) | |
a28ec197 | 2943 | { |
e72cb923 | 2944 | struct btrfs_fs_info *info = trans->fs_info; |
e2fa7227 | 2945 | struct btrfs_key key; |
5d4f98a2 | 2946 | struct btrfs_path *path; |
1261ec42 | 2947 | struct btrfs_root *extent_root = info->extent_root; |
5f39d397 | 2948 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
2949 | struct btrfs_extent_item *ei; |
2950 | struct btrfs_extent_inline_ref *iref; | |
a28ec197 | 2951 | int ret; |
5d4f98a2 | 2952 | int is_data; |
952fccac CM |
2953 | int extent_slot = 0; |
2954 | int found_extent = 0; | |
2955 | int num_to_del = 1; | |
5d4f98a2 YZ |
2956 | u32 item_size; |
2957 | u64 refs; | |
c682f9b3 QW |
2958 | u64 bytenr = node->bytenr; |
2959 | u64 num_bytes = node->num_bytes; | |
fcebe456 | 2960 | int last_ref = 0; |
0b246afa | 2961 | bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA); |
037e6390 | 2962 | |
5caf2a00 | 2963 | path = btrfs_alloc_path(); |
54aa1f4d CM |
2964 | if (!path) |
2965 | return -ENOMEM; | |
5f26f772 | 2966 | |
b9473439 | 2967 | path->leave_spinning = 1; |
5d4f98a2 YZ |
2968 | |
2969 | is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID; | |
2970 | BUG_ON(!is_data && refs_to_drop != 1); | |
2971 | ||
3173a18f | 2972 | if (is_data) |
897ca819 | 2973 | skinny_metadata = false; |
3173a18f | 2974 | |
fbe4801b NB |
2975 | ret = lookup_extent_backref(trans, path, &iref, bytenr, num_bytes, |
2976 | parent, root_objectid, owner_objectid, | |
5d4f98a2 | 2977 | owner_offset); |
7bb86316 | 2978 | if (ret == 0) { |
952fccac | 2979 | extent_slot = path->slots[0]; |
5d4f98a2 YZ |
2980 | while (extent_slot >= 0) { |
2981 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
952fccac | 2982 | extent_slot); |
5d4f98a2 | 2983 | if (key.objectid != bytenr) |
952fccac | 2984 | break; |
5d4f98a2 YZ |
2985 | if (key.type == BTRFS_EXTENT_ITEM_KEY && |
2986 | key.offset == num_bytes) { | |
952fccac CM |
2987 | found_extent = 1; |
2988 | break; | |
2989 | } | |
3173a18f JB |
2990 | if (key.type == BTRFS_METADATA_ITEM_KEY && |
2991 | key.offset == owner_objectid) { | |
2992 | found_extent = 1; | |
2993 | break; | |
2994 | } | |
952fccac CM |
2995 | if (path->slots[0] - extent_slot > 5) |
2996 | break; | |
5d4f98a2 | 2997 | extent_slot--; |
952fccac | 2998 | } |
a79865c6 | 2999 | |
31840ae1 | 3000 | if (!found_extent) { |
5d4f98a2 | 3001 | BUG_ON(iref); |
87cc7a8a | 3002 | ret = remove_extent_backref(trans, path, NULL, |
87bde3cd | 3003 | refs_to_drop, |
fcebe456 | 3004 | is_data, &last_ref); |
005d6427 | 3005 | if (ret) { |
66642832 | 3006 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
3007 | goto out; |
3008 | } | |
b3b4aa74 | 3009 | btrfs_release_path(path); |
b9473439 | 3010 | path->leave_spinning = 1; |
5d4f98a2 YZ |
3011 | |
3012 | key.objectid = bytenr; | |
3013 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
3014 | key.offset = num_bytes; | |
3015 | ||
3173a18f JB |
3016 | if (!is_data && skinny_metadata) { |
3017 | key.type = BTRFS_METADATA_ITEM_KEY; | |
3018 | key.offset = owner_objectid; | |
3019 | } | |
3020 | ||
31840ae1 ZY |
3021 | ret = btrfs_search_slot(trans, extent_root, |
3022 | &key, path, -1, 1); | |
3173a18f JB |
3023 | if (ret > 0 && skinny_metadata && path->slots[0]) { |
3024 | /* | |
3025 | * Couldn't find our skinny metadata item, | |
3026 | * see if we have ye olde extent item. | |
3027 | */ | |
3028 | path->slots[0]--; | |
3029 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
3030 | path->slots[0]); | |
3031 | if (key.objectid == bytenr && | |
3032 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
3033 | key.offset == num_bytes) | |
3034 | ret = 0; | |
3035 | } | |
3036 | ||
3037 | if (ret > 0 && skinny_metadata) { | |
3038 | skinny_metadata = false; | |
9ce49a0b | 3039 | key.objectid = bytenr; |
3173a18f JB |
3040 | key.type = BTRFS_EXTENT_ITEM_KEY; |
3041 | key.offset = num_bytes; | |
3042 | btrfs_release_path(path); | |
3043 | ret = btrfs_search_slot(trans, extent_root, | |
3044 | &key, path, -1, 1); | |
3045 | } | |
3046 | ||
f3465ca4 | 3047 | if (ret) { |
5d163e0e JM |
3048 | btrfs_err(info, |
3049 | "umm, got %d back from search, was looking for %llu", | |
3050 | ret, bytenr); | |
b783e62d | 3051 | if (ret > 0) |
a4f78750 | 3052 | btrfs_print_leaf(path->nodes[0]); |
f3465ca4 | 3053 | } |
005d6427 | 3054 | if (ret < 0) { |
66642832 | 3055 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
3056 | goto out; |
3057 | } | |
31840ae1 ZY |
3058 | extent_slot = path->slots[0]; |
3059 | } | |
fae7f21c | 3060 | } else if (WARN_ON(ret == -ENOENT)) { |
a4f78750 | 3061 | btrfs_print_leaf(path->nodes[0]); |
c2cf52eb SK |
3062 | btrfs_err(info, |
3063 | "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu", | |
c1c9ff7c GU |
3064 | bytenr, parent, root_objectid, owner_objectid, |
3065 | owner_offset); | |
66642832 | 3066 | btrfs_abort_transaction(trans, ret); |
c4a050bb | 3067 | goto out; |
79787eaa | 3068 | } else { |
66642832 | 3069 | btrfs_abort_transaction(trans, ret); |
005d6427 | 3070 | goto out; |
7bb86316 | 3071 | } |
5f39d397 CM |
3072 | |
3073 | leaf = path->nodes[0]; | |
5d4f98a2 | 3074 | item_size = btrfs_item_size_nr(leaf, extent_slot); |
6d8ff4e4 | 3075 | if (unlikely(item_size < sizeof(*ei))) { |
ba3c2b19 NB |
3076 | ret = -EINVAL; |
3077 | btrfs_print_v0_err(info); | |
3078 | btrfs_abort_transaction(trans, ret); | |
3079 | goto out; | |
3080 | } | |
952fccac | 3081 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 3082 | struct btrfs_extent_item); |
3173a18f JB |
3083 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID && |
3084 | key.type == BTRFS_EXTENT_ITEM_KEY) { | |
5d4f98a2 YZ |
3085 | struct btrfs_tree_block_info *bi; |
3086 | BUG_ON(item_size < sizeof(*ei) + sizeof(*bi)); | |
3087 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
3088 | WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi)); | |
3089 | } | |
56bec294 | 3090 | |
5d4f98a2 | 3091 | refs = btrfs_extent_refs(leaf, ei); |
32b02538 | 3092 | if (refs < refs_to_drop) { |
5d163e0e JM |
3093 | btrfs_err(info, |
3094 | "trying to drop %d refs but we only have %Lu for bytenr %Lu", | |
3095 | refs_to_drop, refs, bytenr); | |
32b02538 | 3096 | ret = -EINVAL; |
66642832 | 3097 | btrfs_abort_transaction(trans, ret); |
32b02538 JB |
3098 | goto out; |
3099 | } | |
56bec294 | 3100 | refs -= refs_to_drop; |
5f39d397 | 3101 | |
5d4f98a2 YZ |
3102 | if (refs > 0) { |
3103 | if (extent_op) | |
3104 | __run_delayed_extent_op(extent_op, leaf, ei); | |
3105 | /* | |
3106 | * In the case of inline back ref, reference count will | |
3107 | * be updated by remove_extent_backref | |
952fccac | 3108 | */ |
5d4f98a2 YZ |
3109 | if (iref) { |
3110 | BUG_ON(!found_extent); | |
3111 | } else { | |
3112 | btrfs_set_extent_refs(leaf, ei, refs); | |
3113 | btrfs_mark_buffer_dirty(leaf); | |
3114 | } | |
3115 | if (found_extent) { | |
87cc7a8a NB |
3116 | ret = remove_extent_backref(trans, path, iref, |
3117 | refs_to_drop, is_data, | |
3118 | &last_ref); | |
005d6427 | 3119 | if (ret) { |
66642832 | 3120 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
3121 | goto out; |
3122 | } | |
952fccac | 3123 | } |
5d4f98a2 | 3124 | } else { |
5d4f98a2 YZ |
3125 | if (found_extent) { |
3126 | BUG_ON(is_data && refs_to_drop != | |
9ed0dea0 | 3127 | extent_data_ref_count(path, iref)); |
5d4f98a2 YZ |
3128 | if (iref) { |
3129 | BUG_ON(path->slots[0] != extent_slot); | |
3130 | } else { | |
3131 | BUG_ON(path->slots[0] != extent_slot + 1); | |
3132 | path->slots[0] = extent_slot; | |
3133 | num_to_del = 2; | |
3134 | } | |
78fae27e | 3135 | } |
b9473439 | 3136 | |
fcebe456 | 3137 | last_ref = 1; |
952fccac CM |
3138 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
3139 | num_to_del); | |
005d6427 | 3140 | if (ret) { |
66642832 | 3141 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
3142 | goto out; |
3143 | } | |
b3b4aa74 | 3144 | btrfs_release_path(path); |
21af804c | 3145 | |
5d4f98a2 | 3146 | if (is_data) { |
40e046ac FM |
3147 | ret = btrfs_del_csums(trans, info->csum_root, bytenr, |
3148 | num_bytes); | |
005d6427 | 3149 | if (ret) { |
66642832 | 3150 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
3151 | goto out; |
3152 | } | |
459931ec CM |
3153 | } |
3154 | ||
e7355e50 | 3155 | ret = add_to_free_space_tree(trans, bytenr, num_bytes); |
1e144fb8 | 3156 | if (ret) { |
66642832 | 3157 | btrfs_abort_transaction(trans, ret); |
1e144fb8 OS |
3158 | goto out; |
3159 | } | |
3160 | ||
ade4b516 | 3161 | ret = btrfs_update_block_group(trans, bytenr, num_bytes, 0); |
005d6427 | 3162 | if (ret) { |
66642832 | 3163 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
3164 | goto out; |
3165 | } | |
a28ec197 | 3166 | } |
fcebe456 JB |
3167 | btrfs_release_path(path); |
3168 | ||
79787eaa | 3169 | out: |
5caf2a00 | 3170 | btrfs_free_path(path); |
a28ec197 CM |
3171 | return ret; |
3172 | } | |
3173 | ||
1887be66 | 3174 | /* |
f0486c68 | 3175 | * when we free an block, it is possible (and likely) that we free the last |
1887be66 CM |
3176 | * delayed ref for that extent as well. This searches the delayed ref tree for |
3177 | * a given extent, and if there are no other delayed refs to be processed, it | |
3178 | * removes it from the tree. | |
3179 | */ | |
3180 | static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans, | |
2ff7e61e | 3181 | u64 bytenr) |
1887be66 CM |
3182 | { |
3183 | struct btrfs_delayed_ref_head *head; | |
3184 | struct btrfs_delayed_ref_root *delayed_refs; | |
f0486c68 | 3185 | int ret = 0; |
1887be66 CM |
3186 | |
3187 | delayed_refs = &trans->transaction->delayed_refs; | |
3188 | spin_lock(&delayed_refs->lock); | |
f72ad18e | 3189 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
1887be66 | 3190 | if (!head) |
cf93da7b | 3191 | goto out_delayed_unlock; |
1887be66 | 3192 | |
d7df2c79 | 3193 | spin_lock(&head->lock); |
e3d03965 | 3194 | if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root)) |
1887be66 CM |
3195 | goto out; |
3196 | ||
bedc6617 JB |
3197 | if (cleanup_extent_op(head) != NULL) |
3198 | goto out; | |
5d4f98a2 | 3199 | |
1887be66 CM |
3200 | /* |
3201 | * waiting for the lock here would deadlock. If someone else has it | |
3202 | * locked they are already in the process of dropping it anyway | |
3203 | */ | |
3204 | if (!mutex_trylock(&head->mutex)) | |
3205 | goto out; | |
3206 | ||
d7baffda | 3207 | btrfs_delete_ref_head(delayed_refs, head); |
d7df2c79 | 3208 | head->processing = 0; |
d7baffda | 3209 | |
d7df2c79 | 3210 | spin_unlock(&head->lock); |
1887be66 CM |
3211 | spin_unlock(&delayed_refs->lock); |
3212 | ||
f0486c68 YZ |
3213 | BUG_ON(head->extent_op); |
3214 | if (head->must_insert_reserved) | |
3215 | ret = 1; | |
3216 | ||
31890da0 | 3217 | btrfs_cleanup_ref_head_accounting(trans->fs_info, delayed_refs, head); |
f0486c68 | 3218 | mutex_unlock(&head->mutex); |
d278850e | 3219 | btrfs_put_delayed_ref_head(head); |
f0486c68 | 3220 | return ret; |
1887be66 | 3221 | out: |
d7df2c79 | 3222 | spin_unlock(&head->lock); |
cf93da7b CM |
3223 | |
3224 | out_delayed_unlock: | |
1887be66 CM |
3225 | spin_unlock(&delayed_refs->lock); |
3226 | return 0; | |
3227 | } | |
3228 | ||
f0486c68 YZ |
3229 | void btrfs_free_tree_block(struct btrfs_trans_handle *trans, |
3230 | struct btrfs_root *root, | |
3231 | struct extent_buffer *buf, | |
5581a51a | 3232 | u64 parent, int last_ref) |
f0486c68 | 3233 | { |
0b246afa | 3234 | struct btrfs_fs_info *fs_info = root->fs_info; |
ed4f255b | 3235 | struct btrfs_ref generic_ref = { 0 }; |
b150a4f1 | 3236 | int pin = 1; |
f0486c68 YZ |
3237 | int ret; |
3238 | ||
ed4f255b QW |
3239 | btrfs_init_generic_ref(&generic_ref, BTRFS_DROP_DELAYED_REF, |
3240 | buf->start, buf->len, parent); | |
3241 | btrfs_init_tree_ref(&generic_ref, btrfs_header_level(buf), | |
3242 | root->root_key.objectid); | |
3243 | ||
f0486c68 | 3244 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
d7eae340 OS |
3245 | int old_ref_mod, new_ref_mod; |
3246 | ||
8a5040f7 | 3247 | btrfs_ref_tree_mod(fs_info, &generic_ref); |
ed4f255b | 3248 | ret = btrfs_add_delayed_tree_ref(trans, &generic_ref, NULL, |
d7eae340 | 3249 | &old_ref_mod, &new_ref_mod); |
79787eaa | 3250 | BUG_ON(ret); /* -ENOMEM */ |
d7eae340 | 3251 | pin = old_ref_mod >= 0 && new_ref_mod < 0; |
f0486c68 YZ |
3252 | } |
3253 | ||
0a16c7d7 | 3254 | if (last_ref && btrfs_header_generation(buf) == trans->transid) { |
32da5386 | 3255 | struct btrfs_block_group *cache; |
6219872d | 3256 | |
f0486c68 | 3257 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
2ff7e61e | 3258 | ret = check_ref_cleanup(trans, buf->start); |
f0486c68 | 3259 | if (!ret) |
37be25bc | 3260 | goto out; |
f0486c68 YZ |
3261 | } |
3262 | ||
4da8b76d | 3263 | pin = 0; |
0b246afa | 3264 | cache = btrfs_lookup_block_group(fs_info, buf->start); |
6219872d | 3265 | |
f0486c68 | 3266 | if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { |
6690d071 | 3267 | pin_down_extent(trans, cache, buf->start, buf->len, 1); |
6219872d | 3268 | btrfs_put_block_group(cache); |
37be25bc | 3269 | goto out; |
f0486c68 YZ |
3270 | } |
3271 | ||
3272 | WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)); | |
3273 | ||
3274 | btrfs_add_free_space(cache, buf->start, buf->len); | |
4824f1f4 | 3275 | btrfs_free_reserved_bytes(cache, buf->len, 0); |
6219872d | 3276 | btrfs_put_block_group(cache); |
71ff6437 | 3277 | trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len); |
f0486c68 YZ |
3278 | } |
3279 | out: | |
b150a4f1 | 3280 | if (pin) |
78192442 | 3281 | add_pinned_bytes(fs_info, &generic_ref); |
b150a4f1 | 3282 | |
0a16c7d7 OS |
3283 | if (last_ref) { |
3284 | /* | |
3285 | * Deleting the buffer, clear the corrupt flag since it doesn't | |
3286 | * matter anymore. | |
3287 | */ | |
3288 | clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags); | |
3289 | } | |
f0486c68 YZ |
3290 | } |
3291 | ||
79787eaa | 3292 | /* Can return -ENOMEM */ |
ffd4bb2a | 3293 | int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_ref *ref) |
925baedd | 3294 | { |
ffd4bb2a | 3295 | struct btrfs_fs_info *fs_info = trans->fs_info; |
d7eae340 | 3296 | int old_ref_mod, new_ref_mod; |
925baedd CM |
3297 | int ret; |
3298 | ||
f5ee5c9a | 3299 | if (btrfs_is_testing(fs_info)) |
faa2dbf0 | 3300 | return 0; |
fccb84c9 | 3301 | |
56bec294 CM |
3302 | /* |
3303 | * tree log blocks never actually go into the extent allocation | |
3304 | * tree, just update pinning info and exit early. | |
56bec294 | 3305 | */ |
ffd4bb2a QW |
3306 | if ((ref->type == BTRFS_REF_METADATA && |
3307 | ref->tree_ref.root == BTRFS_TREE_LOG_OBJECTID) || | |
3308 | (ref->type == BTRFS_REF_DATA && | |
3309 | ref->data_ref.ref_root == BTRFS_TREE_LOG_OBJECTID)) { | |
b9473439 | 3310 | /* unlocks the pinned mutex */ |
b25c36f8 | 3311 | btrfs_pin_extent(trans, ref->bytenr, ref->len, 1); |
d7eae340 | 3312 | old_ref_mod = new_ref_mod = 0; |
56bec294 | 3313 | ret = 0; |
ffd4bb2a QW |
3314 | } else if (ref->type == BTRFS_REF_METADATA) { |
3315 | ret = btrfs_add_delayed_tree_ref(trans, ref, NULL, | |
d7eae340 | 3316 | &old_ref_mod, &new_ref_mod); |
5d4f98a2 | 3317 | } else { |
ffd4bb2a | 3318 | ret = btrfs_add_delayed_data_ref(trans, ref, 0, |
d7eae340 | 3319 | &old_ref_mod, &new_ref_mod); |
56bec294 | 3320 | } |
d7eae340 | 3321 | |
ffd4bb2a QW |
3322 | if (!((ref->type == BTRFS_REF_METADATA && |
3323 | ref->tree_ref.root == BTRFS_TREE_LOG_OBJECTID) || | |
3324 | (ref->type == BTRFS_REF_DATA && | |
3325 | ref->data_ref.ref_root == BTRFS_TREE_LOG_OBJECTID))) | |
3326 | btrfs_ref_tree_mod(fs_info, ref); | |
8a5040f7 | 3327 | |
ddf30cf0 | 3328 | if (ret == 0 && old_ref_mod >= 0 && new_ref_mod < 0) |
78192442 | 3329 | add_pinned_bytes(fs_info, ref); |
d7eae340 | 3330 | |
925baedd CM |
3331 | return ret; |
3332 | } | |
3333 | ||
817d52f8 | 3334 | enum btrfs_loop_type { |
f262fa8d DS |
3335 | LOOP_CACHING_NOWAIT, |
3336 | LOOP_CACHING_WAIT, | |
3337 | LOOP_ALLOC_CHUNK, | |
3338 | LOOP_NO_EMPTY_SIZE, | |
817d52f8 JB |
3339 | }; |
3340 | ||
e570fd27 | 3341 | static inline void |
32da5386 | 3342 | btrfs_lock_block_group(struct btrfs_block_group *cache, |
e570fd27 MX |
3343 | int delalloc) |
3344 | { | |
3345 | if (delalloc) | |
3346 | down_read(&cache->data_rwsem); | |
3347 | } | |
3348 | ||
32da5386 | 3349 | static inline void btrfs_grab_block_group(struct btrfs_block_group *cache, |
e570fd27 MX |
3350 | int delalloc) |
3351 | { | |
3352 | btrfs_get_block_group(cache); | |
3353 | if (delalloc) | |
3354 | down_read(&cache->data_rwsem); | |
3355 | } | |
3356 | ||
32da5386 DS |
3357 | static struct btrfs_block_group *btrfs_lock_cluster( |
3358 | struct btrfs_block_group *block_group, | |
e570fd27 MX |
3359 | struct btrfs_free_cluster *cluster, |
3360 | int delalloc) | |
c142c6a4 | 3361 | __acquires(&cluster->refill_lock) |
e570fd27 | 3362 | { |
32da5386 | 3363 | struct btrfs_block_group *used_bg = NULL; |
6719afdc | 3364 | |
e570fd27 | 3365 | spin_lock(&cluster->refill_lock); |
6719afdc GU |
3366 | while (1) { |
3367 | used_bg = cluster->block_group; | |
3368 | if (!used_bg) | |
3369 | return NULL; | |
3370 | ||
3371 | if (used_bg == block_group) | |
e570fd27 MX |
3372 | return used_bg; |
3373 | ||
6719afdc | 3374 | btrfs_get_block_group(used_bg); |
e570fd27 | 3375 | |
6719afdc GU |
3376 | if (!delalloc) |
3377 | return used_bg; | |
e570fd27 | 3378 | |
6719afdc GU |
3379 | if (down_read_trylock(&used_bg->data_rwsem)) |
3380 | return used_bg; | |
e570fd27 | 3381 | |
6719afdc | 3382 | spin_unlock(&cluster->refill_lock); |
e570fd27 | 3383 | |
e321f8a8 LB |
3384 | /* We should only have one-level nested. */ |
3385 | down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING); | |
e570fd27 | 3386 | |
6719afdc GU |
3387 | spin_lock(&cluster->refill_lock); |
3388 | if (used_bg == cluster->block_group) | |
3389 | return used_bg; | |
e570fd27 | 3390 | |
6719afdc GU |
3391 | up_read(&used_bg->data_rwsem); |
3392 | btrfs_put_block_group(used_bg); | |
3393 | } | |
e570fd27 MX |
3394 | } |
3395 | ||
3396 | static inline void | |
32da5386 | 3397 | btrfs_release_block_group(struct btrfs_block_group *cache, |
e570fd27 MX |
3398 | int delalloc) |
3399 | { | |
3400 | if (delalloc) | |
3401 | up_read(&cache->data_rwsem); | |
3402 | btrfs_put_block_group(cache); | |
3403 | } | |
3404 | ||
cb2f96f8 NA |
3405 | enum btrfs_extent_allocation_policy { |
3406 | BTRFS_EXTENT_ALLOC_CLUSTERED, | |
3407 | }; | |
3408 | ||
b4bd745d QW |
3409 | /* |
3410 | * Structure used internally for find_free_extent() function. Wraps needed | |
3411 | * parameters. | |
3412 | */ | |
3413 | struct find_free_extent_ctl { | |
3414 | /* Basic allocation info */ | |
b4bd745d QW |
3415 | u64 num_bytes; |
3416 | u64 empty_size; | |
3417 | u64 flags; | |
3418 | int delalloc; | |
3419 | ||
3420 | /* Where to start the search inside the bg */ | |
3421 | u64 search_start; | |
3422 | ||
3423 | /* For clustered allocation */ | |
3424 | u64 empty_cluster; | |
c10859be NA |
3425 | struct btrfs_free_cluster *last_ptr; |
3426 | bool use_cluster; | |
b4bd745d QW |
3427 | |
3428 | bool have_caching_bg; | |
3429 | bool orig_have_caching_bg; | |
3430 | ||
3431 | /* RAID index, converted from flags */ | |
3432 | int index; | |
3433 | ||
e72d79d6 QW |
3434 | /* |
3435 | * Current loop number, check find_free_extent_update_loop() for details | |
3436 | */ | |
b4bd745d QW |
3437 | int loop; |
3438 | ||
3439 | /* | |
3440 | * Whether we're refilling a cluster, if true we need to re-search | |
3441 | * current block group but don't try to refill the cluster again. | |
3442 | */ | |
3443 | bool retry_clustered; | |
3444 | ||
3445 | /* | |
3446 | * Whether we're updating free space cache, if true we need to re-search | |
3447 | * current block group but don't try updating free space cache again. | |
3448 | */ | |
3449 | bool retry_unclustered; | |
3450 | ||
3451 | /* If current block group is cached */ | |
3452 | int cached; | |
3453 | ||
3454 | /* Max contiguous hole found */ | |
3455 | u64 max_extent_size; | |
3456 | ||
3457 | /* Total free space from free space cache, not always contiguous */ | |
3458 | u64 total_free_space; | |
3459 | ||
3460 | /* Found result */ | |
3461 | u64 found_offset; | |
cb2f96f8 | 3462 | |
ea544149 NA |
3463 | /* Hint where to start looking for an empty space */ |
3464 | u64 hint_byte; | |
3465 | ||
cb2f96f8 NA |
3466 | /* Allocation policy */ |
3467 | enum btrfs_extent_allocation_policy policy; | |
b4bd745d QW |
3468 | }; |
3469 | ||
d06e3bb6 QW |
3470 | |
3471 | /* | |
3472 | * Helper function for find_free_extent(). | |
3473 | * | |
3474 | * Return -ENOENT to inform caller that we need fallback to unclustered mode. | |
3475 | * Return -EAGAIN to inform caller that we need to re-search this block group | |
3476 | * Return >0 to inform caller that we find nothing | |
3477 | * Return 0 means we have found a location and set ffe_ctl->found_offset. | |
3478 | */ | |
32da5386 | 3479 | static int find_free_extent_clustered(struct btrfs_block_group *bg, |
897cae79 NA |
3480 | struct find_free_extent_ctl *ffe_ctl, |
3481 | struct btrfs_block_group **cluster_bg_ret) | |
d06e3bb6 | 3482 | { |
32da5386 | 3483 | struct btrfs_block_group *cluster_bg; |
897cae79 | 3484 | struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr; |
d06e3bb6 QW |
3485 | u64 aligned_cluster; |
3486 | u64 offset; | |
3487 | int ret; | |
3488 | ||
3489 | cluster_bg = btrfs_lock_cluster(bg, last_ptr, ffe_ctl->delalloc); | |
3490 | if (!cluster_bg) | |
3491 | goto refill_cluster; | |
3492 | if (cluster_bg != bg && (cluster_bg->ro || | |
3493 | !block_group_bits(cluster_bg, ffe_ctl->flags))) | |
3494 | goto release_cluster; | |
3495 | ||
3496 | offset = btrfs_alloc_from_cluster(cluster_bg, last_ptr, | |
b3470b5d | 3497 | ffe_ctl->num_bytes, cluster_bg->start, |
d06e3bb6 QW |
3498 | &ffe_ctl->max_extent_size); |
3499 | if (offset) { | |
3500 | /* We have a block, we're done */ | |
3501 | spin_unlock(&last_ptr->refill_lock); | |
3502 | trace_btrfs_reserve_extent_cluster(cluster_bg, | |
3503 | ffe_ctl->search_start, ffe_ctl->num_bytes); | |
3504 | *cluster_bg_ret = cluster_bg; | |
3505 | ffe_ctl->found_offset = offset; | |
3506 | return 0; | |
3507 | } | |
3508 | WARN_ON(last_ptr->block_group != cluster_bg); | |
3509 | ||
3510 | release_cluster: | |
3511 | /* | |
3512 | * If we are on LOOP_NO_EMPTY_SIZE, we can't set up a new clusters, so | |
3513 | * lets just skip it and let the allocator find whatever block it can | |
3514 | * find. If we reach this point, we will have tried the cluster | |
3515 | * allocator plenty of times and not have found anything, so we are | |
3516 | * likely way too fragmented for the clustering stuff to find anything. | |
3517 | * | |
3518 | * However, if the cluster is taken from the current block group, | |
3519 | * release the cluster first, so that we stand a better chance of | |
3520 | * succeeding in the unclustered allocation. | |
3521 | */ | |
3522 | if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE && cluster_bg != bg) { | |
3523 | spin_unlock(&last_ptr->refill_lock); | |
3524 | btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc); | |
3525 | return -ENOENT; | |
3526 | } | |
3527 | ||
3528 | /* This cluster didn't work out, free it and start over */ | |
3529 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
3530 | ||
3531 | if (cluster_bg != bg) | |
3532 | btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc); | |
3533 | ||
3534 | refill_cluster: | |
3535 | if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE) { | |
3536 | spin_unlock(&last_ptr->refill_lock); | |
3537 | return -ENOENT; | |
3538 | } | |
3539 | ||
3540 | aligned_cluster = max_t(u64, | |
3541 | ffe_ctl->empty_cluster + ffe_ctl->empty_size, | |
3542 | bg->full_stripe_len); | |
2ceeae2e DS |
3543 | ret = btrfs_find_space_cluster(bg, last_ptr, ffe_ctl->search_start, |
3544 | ffe_ctl->num_bytes, aligned_cluster); | |
d06e3bb6 QW |
3545 | if (ret == 0) { |
3546 | /* Now pull our allocation out of this cluster */ | |
3547 | offset = btrfs_alloc_from_cluster(bg, last_ptr, | |
3548 | ffe_ctl->num_bytes, ffe_ctl->search_start, | |
3549 | &ffe_ctl->max_extent_size); | |
3550 | if (offset) { | |
3551 | /* We found one, proceed */ | |
3552 | spin_unlock(&last_ptr->refill_lock); | |
3553 | trace_btrfs_reserve_extent_cluster(bg, | |
3554 | ffe_ctl->search_start, | |
3555 | ffe_ctl->num_bytes); | |
3556 | ffe_ctl->found_offset = offset; | |
3557 | return 0; | |
3558 | } | |
3559 | } else if (!ffe_ctl->cached && ffe_ctl->loop > LOOP_CACHING_NOWAIT && | |
3560 | !ffe_ctl->retry_clustered) { | |
3561 | spin_unlock(&last_ptr->refill_lock); | |
3562 | ||
3563 | ffe_ctl->retry_clustered = true; | |
676f1f75 | 3564 | btrfs_wait_block_group_cache_progress(bg, ffe_ctl->num_bytes + |
d06e3bb6 QW |
3565 | ffe_ctl->empty_cluster + ffe_ctl->empty_size); |
3566 | return -EAGAIN; | |
3567 | } | |
3568 | /* | |
3569 | * At this point we either didn't find a cluster or we weren't able to | |
3570 | * allocate a block from our cluster. Free the cluster we've been | |
3571 | * trying to use, and go to the next block group. | |
3572 | */ | |
3573 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
3574 | spin_unlock(&last_ptr->refill_lock); | |
3575 | return 1; | |
3576 | } | |
3577 | ||
e1a41848 QW |
3578 | /* |
3579 | * Return >0 to inform caller that we find nothing | |
3580 | * Return 0 when we found an free extent and set ffe_ctrl->found_offset | |
3581 | * Return -EAGAIN to inform caller that we need to re-search this block group | |
3582 | */ | |
32da5386 | 3583 | static int find_free_extent_unclustered(struct btrfs_block_group *bg, |
897cae79 | 3584 | struct find_free_extent_ctl *ffe_ctl) |
e1a41848 | 3585 | { |
897cae79 | 3586 | struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr; |
e1a41848 QW |
3587 | u64 offset; |
3588 | ||
3589 | /* | |
3590 | * We are doing an unclustered allocation, set the fragmented flag so | |
3591 | * we don't bother trying to setup a cluster again until we get more | |
3592 | * space. | |
3593 | */ | |
3594 | if (unlikely(last_ptr)) { | |
3595 | spin_lock(&last_ptr->lock); | |
3596 | last_ptr->fragmented = 1; | |
3597 | spin_unlock(&last_ptr->lock); | |
3598 | } | |
3599 | if (ffe_ctl->cached) { | |
3600 | struct btrfs_free_space_ctl *free_space_ctl; | |
3601 | ||
3602 | free_space_ctl = bg->free_space_ctl; | |
3603 | spin_lock(&free_space_ctl->tree_lock); | |
3604 | if (free_space_ctl->free_space < | |
3605 | ffe_ctl->num_bytes + ffe_ctl->empty_cluster + | |
3606 | ffe_ctl->empty_size) { | |
3607 | ffe_ctl->total_free_space = max_t(u64, | |
3608 | ffe_ctl->total_free_space, | |
3609 | free_space_ctl->free_space); | |
3610 | spin_unlock(&free_space_ctl->tree_lock); | |
3611 | return 1; | |
3612 | } | |
3613 | spin_unlock(&free_space_ctl->tree_lock); | |
3614 | } | |
3615 | ||
3616 | offset = btrfs_find_space_for_alloc(bg, ffe_ctl->search_start, | |
3617 | ffe_ctl->num_bytes, ffe_ctl->empty_size, | |
3618 | &ffe_ctl->max_extent_size); | |
3619 | ||
3620 | /* | |
3621 | * If we didn't find a chunk, and we haven't failed on this block group | |
3622 | * before, and this block group is in the middle of caching and we are | |
3623 | * ok with waiting, then go ahead and wait for progress to be made, and | |
3624 | * set @retry_unclustered to true. | |
3625 | * | |
3626 | * If @retry_unclustered is true then we've already waited on this | |
3627 | * block group once and should move on to the next block group. | |
3628 | */ | |
3629 | if (!offset && !ffe_ctl->retry_unclustered && !ffe_ctl->cached && | |
3630 | ffe_ctl->loop > LOOP_CACHING_NOWAIT) { | |
676f1f75 JB |
3631 | btrfs_wait_block_group_cache_progress(bg, ffe_ctl->num_bytes + |
3632 | ffe_ctl->empty_size); | |
e1a41848 QW |
3633 | ffe_ctl->retry_unclustered = true; |
3634 | return -EAGAIN; | |
3635 | } else if (!offset) { | |
3636 | return 1; | |
3637 | } | |
3638 | ffe_ctl->found_offset = offset; | |
3639 | return 0; | |
3640 | } | |
3641 | ||
c668690d NA |
3642 | static int do_allocation_clustered(struct btrfs_block_group *block_group, |
3643 | struct find_free_extent_ctl *ffe_ctl, | |
3644 | struct btrfs_block_group **bg_ret) | |
3645 | { | |
3646 | int ret; | |
3647 | ||
3648 | /* We want to try and use the cluster allocator, so lets look there */ | |
3649 | if (ffe_ctl->last_ptr && ffe_ctl->use_cluster) { | |
897cae79 | 3650 | ret = find_free_extent_clustered(block_group, ffe_ctl, bg_ret); |
c668690d NA |
3651 | if (ret >= 0 || ret == -EAGAIN) |
3652 | return ret; | |
3653 | /* ret == -ENOENT case falls through */ | |
3654 | } | |
3655 | ||
897cae79 | 3656 | return find_free_extent_unclustered(block_group, ffe_ctl); |
c668690d NA |
3657 | } |
3658 | ||
3659 | static int do_allocation(struct btrfs_block_group *block_group, | |
3660 | struct find_free_extent_ctl *ffe_ctl, | |
3661 | struct btrfs_block_group **bg_ret) | |
3662 | { | |
3663 | switch (ffe_ctl->policy) { | |
3664 | case BTRFS_EXTENT_ALLOC_CLUSTERED: | |
3665 | return do_allocation_clustered(block_group, ffe_ctl, bg_ret); | |
3666 | default: | |
3667 | BUG(); | |
3668 | } | |
3669 | } | |
3670 | ||
baba5062 NA |
3671 | static void release_block_group(struct btrfs_block_group *block_group, |
3672 | struct find_free_extent_ctl *ffe_ctl, | |
3673 | int delalloc) | |
3674 | { | |
3675 | switch (ffe_ctl->policy) { | |
3676 | case BTRFS_EXTENT_ALLOC_CLUSTERED: | |
3677 | ffe_ctl->retry_clustered = false; | |
3678 | ffe_ctl->retry_unclustered = false; | |
3679 | break; | |
3680 | default: | |
3681 | BUG(); | |
3682 | } | |
3683 | ||
3684 | BUG_ON(btrfs_bg_flags_to_raid_index(block_group->flags) != | |
3685 | ffe_ctl->index); | |
3686 | btrfs_release_block_group(block_group, delalloc); | |
3687 | } | |
3688 | ||
0ab9724b NA |
3689 | static void found_extent_clustered(struct find_free_extent_ctl *ffe_ctl, |
3690 | struct btrfs_key *ins) | |
3691 | { | |
3692 | struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr; | |
3693 | ||
3694 | if (!ffe_ctl->use_cluster && last_ptr) { | |
3695 | spin_lock(&last_ptr->lock); | |
3696 | last_ptr->window_start = ins->objectid; | |
3697 | spin_unlock(&last_ptr->lock); | |
3698 | } | |
3699 | } | |
3700 | ||
3701 | static void found_extent(struct find_free_extent_ctl *ffe_ctl, | |
3702 | struct btrfs_key *ins) | |
3703 | { | |
3704 | switch (ffe_ctl->policy) { | |
3705 | case BTRFS_EXTENT_ALLOC_CLUSTERED: | |
3706 | found_extent_clustered(ffe_ctl, ins); | |
3707 | break; | |
3708 | default: | |
3709 | BUG(); | |
3710 | } | |
3711 | } | |
3712 | ||
c70e2139 NA |
3713 | static int chunk_allocation_failed(struct find_free_extent_ctl *ffe_ctl) |
3714 | { | |
3715 | switch (ffe_ctl->policy) { | |
3716 | case BTRFS_EXTENT_ALLOC_CLUSTERED: | |
3717 | /* | |
3718 | * If we can't allocate a new chunk we've already looped through | |
3719 | * at least once, move on to the NO_EMPTY_SIZE case. | |
3720 | */ | |
3721 | ffe_ctl->loop = LOOP_NO_EMPTY_SIZE; | |
3722 | return 0; | |
3723 | default: | |
3724 | BUG(); | |
3725 | } | |
3726 | } | |
3727 | ||
e72d79d6 QW |
3728 | /* |
3729 | * Return >0 means caller needs to re-search for free extent | |
3730 | * Return 0 means we have the needed free extent. | |
3731 | * Return <0 means we failed to locate any free extent. | |
3732 | */ | |
3733 | static int find_free_extent_update_loop(struct btrfs_fs_info *fs_info, | |
e72d79d6 QW |
3734 | struct btrfs_key *ins, |
3735 | struct find_free_extent_ctl *ffe_ctl, | |
15b7ee65 | 3736 | bool full_search) |
e72d79d6 QW |
3737 | { |
3738 | struct btrfs_root *root = fs_info->extent_root; | |
3739 | int ret; | |
3740 | ||
3741 | if ((ffe_ctl->loop == LOOP_CACHING_NOWAIT) && | |
3742 | ffe_ctl->have_caching_bg && !ffe_ctl->orig_have_caching_bg) | |
3743 | ffe_ctl->orig_have_caching_bg = true; | |
3744 | ||
3745 | if (!ins->objectid && ffe_ctl->loop >= LOOP_CACHING_WAIT && | |
3746 | ffe_ctl->have_caching_bg) | |
3747 | return 1; | |
3748 | ||
3749 | if (!ins->objectid && ++(ffe_ctl->index) < BTRFS_NR_RAID_TYPES) | |
3750 | return 1; | |
3751 | ||
3752 | if (ins->objectid) { | |
0ab9724b | 3753 | found_extent(ffe_ctl, ins); |
e72d79d6 QW |
3754 | return 0; |
3755 | } | |
3756 | ||
3757 | /* | |
3758 | * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking | |
3759 | * caching kthreads as we move along | |
3760 | * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching | |
3761 | * LOOP_ALLOC_CHUNK, force a chunk allocation and try again | |
3762 | * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try | |
3763 | * again | |
3764 | */ | |
3765 | if (ffe_ctl->loop < LOOP_NO_EMPTY_SIZE) { | |
3766 | ffe_ctl->index = 0; | |
3767 | if (ffe_ctl->loop == LOOP_CACHING_NOWAIT) { | |
3768 | /* | |
3769 | * We want to skip the LOOP_CACHING_WAIT step if we | |
3770 | * don't have any uncached bgs and we've already done a | |
3771 | * full search through. | |
3772 | */ | |
3773 | if (ffe_ctl->orig_have_caching_bg || !full_search) | |
3774 | ffe_ctl->loop = LOOP_CACHING_WAIT; | |
3775 | else | |
3776 | ffe_ctl->loop = LOOP_ALLOC_CHUNK; | |
3777 | } else { | |
3778 | ffe_ctl->loop++; | |
3779 | } | |
3780 | ||
3781 | if (ffe_ctl->loop == LOOP_ALLOC_CHUNK) { | |
3782 | struct btrfs_trans_handle *trans; | |
3783 | int exist = 0; | |
3784 | ||
3785 | trans = current->journal_info; | |
3786 | if (trans) | |
3787 | exist = 1; | |
3788 | else | |
3789 | trans = btrfs_join_transaction(root); | |
3790 | ||
3791 | if (IS_ERR(trans)) { | |
3792 | ret = PTR_ERR(trans); | |
3793 | return ret; | |
3794 | } | |
3795 | ||
fc471cb0 JB |
3796 | ret = btrfs_chunk_alloc(trans, ffe_ctl->flags, |
3797 | CHUNK_ALLOC_FORCE); | |
e72d79d6 | 3798 | |
e72d79d6 | 3799 | /* Do not bail out on ENOSPC since we can do more. */ |
c70e2139 NA |
3800 | if (ret == -ENOSPC) |
3801 | ret = chunk_allocation_failed(ffe_ctl); | |
3802 | else if (ret < 0) | |
e72d79d6 QW |
3803 | btrfs_abort_transaction(trans, ret); |
3804 | else | |
3805 | ret = 0; | |
3806 | if (!exist) | |
3807 | btrfs_end_transaction(trans); | |
3808 | if (ret) | |
3809 | return ret; | |
3810 | } | |
3811 | ||
3812 | if (ffe_ctl->loop == LOOP_NO_EMPTY_SIZE) { | |
45d8e033 NA |
3813 | if (ffe_ctl->policy != BTRFS_EXTENT_ALLOC_CLUSTERED) |
3814 | return -ENOSPC; | |
3815 | ||
e72d79d6 QW |
3816 | /* |
3817 | * Don't loop again if we already have no empty_size and | |
3818 | * no empty_cluster. | |
3819 | */ | |
3820 | if (ffe_ctl->empty_size == 0 && | |
3821 | ffe_ctl->empty_cluster == 0) | |
3822 | return -ENOSPC; | |
3823 | ffe_ctl->empty_size = 0; | |
3824 | ffe_ctl->empty_cluster = 0; | |
3825 | } | |
3826 | return 1; | |
3827 | } | |
3828 | return -ENOSPC; | |
3829 | } | |
3830 | ||
7e895409 NA |
3831 | static int prepare_allocation_clustered(struct btrfs_fs_info *fs_info, |
3832 | struct find_free_extent_ctl *ffe_ctl, | |
3833 | struct btrfs_space_info *space_info, | |
3834 | struct btrfs_key *ins) | |
3835 | { | |
3836 | /* | |
3837 | * If our free space is heavily fragmented we may not be able to make | |
3838 | * big contiguous allocations, so instead of doing the expensive search | |
3839 | * for free space, simply return ENOSPC with our max_extent_size so we | |
3840 | * can go ahead and search for a more manageable chunk. | |
3841 | * | |
3842 | * If our max_extent_size is large enough for our allocation simply | |
3843 | * disable clustering since we will likely not be able to find enough | |
3844 | * space to create a cluster and induce latency trying. | |
3845 | */ | |
3846 | if (space_info->max_extent_size) { | |
3847 | spin_lock(&space_info->lock); | |
3848 | if (space_info->max_extent_size && | |
3849 | ffe_ctl->num_bytes > space_info->max_extent_size) { | |
3850 | ins->offset = space_info->max_extent_size; | |
3851 | spin_unlock(&space_info->lock); | |
3852 | return -ENOSPC; | |
3853 | } else if (space_info->max_extent_size) { | |
3854 | ffe_ctl->use_cluster = false; | |
3855 | } | |
3856 | spin_unlock(&space_info->lock); | |
3857 | } | |
3858 | ||
3859 | ffe_ctl->last_ptr = fetch_cluster_info(fs_info, space_info, | |
3860 | &ffe_ctl->empty_cluster); | |
3861 | if (ffe_ctl->last_ptr) { | |
3862 | struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr; | |
3863 | ||
3864 | spin_lock(&last_ptr->lock); | |
3865 | if (last_ptr->block_group) | |
3866 | ffe_ctl->hint_byte = last_ptr->window_start; | |
3867 | if (last_ptr->fragmented) { | |
3868 | /* | |
3869 | * We still set window_start so we can keep track of the | |
3870 | * last place we found an allocation to try and save | |
3871 | * some time. | |
3872 | */ | |
3873 | ffe_ctl->hint_byte = last_ptr->window_start; | |
3874 | ffe_ctl->use_cluster = false; | |
3875 | } | |
3876 | spin_unlock(&last_ptr->lock); | |
3877 | } | |
3878 | ||
3879 | return 0; | |
3880 | } | |
3881 | ||
3882 | static int prepare_allocation(struct btrfs_fs_info *fs_info, | |
3883 | struct find_free_extent_ctl *ffe_ctl, | |
3884 | struct btrfs_space_info *space_info, | |
3885 | struct btrfs_key *ins) | |
3886 | { | |
3887 | switch (ffe_ctl->policy) { | |
3888 | case BTRFS_EXTENT_ALLOC_CLUSTERED: | |
3889 | return prepare_allocation_clustered(fs_info, ffe_ctl, | |
3890 | space_info, ins); | |
3891 | default: | |
3892 | BUG(); | |
3893 | } | |
3894 | } | |
3895 | ||
fec577fb CM |
3896 | /* |
3897 | * walks the btree of allocated extents and find a hole of a given size. | |
3898 | * The key ins is changed to record the hole: | |
a4820398 | 3899 | * ins->objectid == start position |
62e2749e | 3900 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
a4820398 | 3901 | * ins->offset == the size of the hole. |
fec577fb | 3902 | * Any available blocks before search_start are skipped. |
a4820398 MX |
3903 | * |
3904 | * If there is no suitable free space, we will record the max size of | |
3905 | * the free space extent currently. | |
e72d79d6 QW |
3906 | * |
3907 | * The overall logic and call chain: | |
3908 | * | |
3909 | * find_free_extent() | |
3910 | * |- Iterate through all block groups | |
3911 | * | |- Get a valid block group | |
3912 | * | |- Try to do clustered allocation in that block group | |
3913 | * | |- Try to do unclustered allocation in that block group | |
3914 | * | |- Check if the result is valid | |
3915 | * | | |- If valid, then exit | |
3916 | * | |- Jump to next block group | |
3917 | * | | |
3918 | * |- Push harder to find free extents | |
3919 | * |- If not found, re-iterate all block groups | |
fec577fb | 3920 | */ |
437490fe | 3921 | static noinline int find_free_extent(struct btrfs_root *root, |
18513091 | 3922 | u64 ram_bytes, u64 num_bytes, u64 empty_size, |
ea544149 | 3923 | u64 hint_byte_orig, struct btrfs_key *ins, |
18513091 | 3924 | u64 flags, int delalloc) |
fec577fb | 3925 | { |
437490fe | 3926 | struct btrfs_fs_info *fs_info = root->fs_info; |
80eb234a | 3927 | int ret = 0; |
db8fe64f | 3928 | int cache_block_group_error = 0; |
32da5386 | 3929 | struct btrfs_block_group *block_group = NULL; |
b4bd745d | 3930 | struct find_free_extent_ctl ffe_ctl = {0}; |
80eb234a | 3931 | struct btrfs_space_info *space_info; |
a5e681d9 | 3932 | bool full_search = false; |
fec577fb | 3933 | |
0b246afa | 3934 | WARN_ON(num_bytes < fs_info->sectorsize); |
b4bd745d | 3935 | |
b4bd745d QW |
3936 | ffe_ctl.num_bytes = num_bytes; |
3937 | ffe_ctl.empty_size = empty_size; | |
3938 | ffe_ctl.flags = flags; | |
3939 | ffe_ctl.search_start = 0; | |
b4bd745d QW |
3940 | ffe_ctl.delalloc = delalloc; |
3941 | ffe_ctl.index = btrfs_bg_flags_to_raid_index(flags); | |
3942 | ffe_ctl.have_caching_bg = false; | |
3943 | ffe_ctl.orig_have_caching_bg = false; | |
3944 | ffe_ctl.found_offset = 0; | |
ea544149 | 3945 | ffe_ctl.hint_byte = hint_byte_orig; |
cb2f96f8 | 3946 | ffe_ctl.policy = BTRFS_EXTENT_ALLOC_CLUSTERED; |
b4bd745d | 3947 | |
c10859be NA |
3948 | /* For clustered allocation */ |
3949 | ffe_ctl.retry_clustered = false; | |
3950 | ffe_ctl.retry_unclustered = false; | |
3951 | ffe_ctl.last_ptr = NULL; | |
3952 | ffe_ctl.use_cluster = true; | |
3953 | ||
962a298f | 3954 | ins->type = BTRFS_EXTENT_ITEM_KEY; |
80eb234a JB |
3955 | ins->objectid = 0; |
3956 | ins->offset = 0; | |
b1a4d965 | 3957 | |
437490fe | 3958 | trace_find_free_extent(root, num_bytes, empty_size, flags); |
3f7de037 | 3959 | |
280c2908 | 3960 | space_info = btrfs_find_space_info(fs_info, flags); |
1b1d1f66 | 3961 | if (!space_info) { |
0b246afa | 3962 | btrfs_err(fs_info, "No space info for %llu", flags); |
1b1d1f66 JB |
3963 | return -ENOSPC; |
3964 | } | |
2552d17e | 3965 | |
7e895409 NA |
3966 | ret = prepare_allocation(fs_info, &ffe_ctl, space_info, ins); |
3967 | if (ret < 0) | |
3968 | return ret; | |
fa9c0d79 | 3969 | |
b4bd745d QW |
3970 | ffe_ctl.search_start = max(ffe_ctl.search_start, |
3971 | first_logical_byte(fs_info, 0)); | |
ea544149 NA |
3972 | ffe_ctl.search_start = max(ffe_ctl.search_start, ffe_ctl.hint_byte); |
3973 | if (ffe_ctl.search_start == ffe_ctl.hint_byte) { | |
b4bd745d QW |
3974 | block_group = btrfs_lookup_block_group(fs_info, |
3975 | ffe_ctl.search_start); | |
817d52f8 JB |
3976 | /* |
3977 | * we don't want to use the block group if it doesn't match our | |
3978 | * allocation bits, or if its not cached. | |
ccf0e725 JB |
3979 | * |
3980 | * However if we are re-searching with an ideal block group | |
3981 | * picked out then we don't care that the block group is cached. | |
817d52f8 | 3982 | */ |
b6919a58 | 3983 | if (block_group && block_group_bits(block_group, flags) && |
285ff5af | 3984 | block_group->cached != BTRFS_CACHE_NO) { |
2552d17e | 3985 | down_read(&space_info->groups_sem); |
44fb5511 CM |
3986 | if (list_empty(&block_group->list) || |
3987 | block_group->ro) { | |
3988 | /* | |
3989 | * someone is removing this block group, | |
3990 | * we can't jump into the have_block_group | |
3991 | * target because our list pointers are not | |
3992 | * valid | |
3993 | */ | |
3994 | btrfs_put_block_group(block_group); | |
3995 | up_read(&space_info->groups_sem); | |
ccf0e725 | 3996 | } else { |
b4bd745d | 3997 | ffe_ctl.index = btrfs_bg_flags_to_raid_index( |
3e72ee88 | 3998 | block_group->flags); |
e570fd27 | 3999 | btrfs_lock_block_group(block_group, delalloc); |
44fb5511 | 4000 | goto have_block_group; |
ccf0e725 | 4001 | } |
2552d17e | 4002 | } else if (block_group) { |
fa9c0d79 | 4003 | btrfs_put_block_group(block_group); |
2552d17e | 4004 | } |
42e70e7a | 4005 | } |
2552d17e | 4006 | search: |
b4bd745d QW |
4007 | ffe_ctl.have_caching_bg = false; |
4008 | if (ffe_ctl.index == btrfs_bg_flags_to_raid_index(flags) || | |
4009 | ffe_ctl.index == 0) | |
a5e681d9 | 4010 | full_search = true; |
80eb234a | 4011 | down_read(&space_info->groups_sem); |
b4bd745d QW |
4012 | list_for_each_entry(block_group, |
4013 | &space_info->block_groups[ffe_ctl.index], list) { | |
c668690d NA |
4014 | struct btrfs_block_group *bg_ret; |
4015 | ||
14443937 JM |
4016 | /* If the block group is read-only, we can skip it entirely. */ |
4017 | if (unlikely(block_group->ro)) | |
4018 | continue; | |
4019 | ||
e570fd27 | 4020 | btrfs_grab_block_group(block_group, delalloc); |
b3470b5d | 4021 | ffe_ctl.search_start = block_group->start; |
42e70e7a | 4022 | |
83a50de9 CM |
4023 | /* |
4024 | * this can happen if we end up cycling through all the | |
4025 | * raid types, but we want to make sure we only allocate | |
4026 | * for the proper type. | |
4027 | */ | |
b6919a58 | 4028 | if (!block_group_bits(block_group, flags)) { |
bece2e82 | 4029 | u64 extra = BTRFS_BLOCK_GROUP_DUP | |
c7369b3f | 4030 | BTRFS_BLOCK_GROUP_RAID1_MASK | |
a07e8a46 | 4031 | BTRFS_BLOCK_GROUP_RAID56_MASK | |
83a50de9 CM |
4032 | BTRFS_BLOCK_GROUP_RAID10; |
4033 | ||
4034 | /* | |
4035 | * if they asked for extra copies and this block group | |
4036 | * doesn't provide them, bail. This does allow us to | |
4037 | * fill raid0 from raid1. | |
4038 | */ | |
b6919a58 | 4039 | if ((flags & extra) && !(block_group->flags & extra)) |
83a50de9 | 4040 | goto loop; |
2a28468e QW |
4041 | |
4042 | /* | |
4043 | * This block group has different flags than we want. | |
4044 | * It's possible that we have MIXED_GROUP flag but no | |
4045 | * block group is mixed. Just skip such block group. | |
4046 | */ | |
4047 | btrfs_release_block_group(block_group, delalloc); | |
4048 | continue; | |
83a50de9 CM |
4049 | } |
4050 | ||
2552d17e | 4051 | have_block_group: |
32da5386 | 4052 | ffe_ctl.cached = btrfs_block_group_done(block_group); |
b4bd745d QW |
4053 | if (unlikely(!ffe_ctl.cached)) { |
4054 | ffe_ctl.have_caching_bg = true; | |
676f1f75 | 4055 | ret = btrfs_cache_block_group(block_group, 0); |
db8fe64f JB |
4056 | |
4057 | /* | |
4058 | * If we get ENOMEM here or something else we want to | |
4059 | * try other block groups, because it may not be fatal. | |
4060 | * However if we can't find anything else we need to | |
4061 | * save our return here so that we return the actual | |
4062 | * error that caused problems, not ENOSPC. | |
4063 | */ | |
4064 | if (ret < 0) { | |
4065 | if (!cache_block_group_error) | |
4066 | cache_block_group_error = ret; | |
4067 | ret = 0; | |
4068 | goto loop; | |
4069 | } | |
1d4284bd | 4070 | ret = 0; |
817d52f8 JB |
4071 | } |
4072 | ||
36cce922 JB |
4073 | if (unlikely(block_group->cached == BTRFS_CACHE_ERROR)) |
4074 | goto loop; | |
0f9dd46c | 4075 | |
c668690d NA |
4076 | bg_ret = NULL; |
4077 | ret = do_allocation(block_group, &ffe_ctl, &bg_ret); | |
4078 | if (ret == 0) { | |
4079 | if (bg_ret && bg_ret != block_group) { | |
4080 | btrfs_release_block_group(block_group, delalloc); | |
4081 | block_group = bg_ret; | |
fa9c0d79 | 4082 | } |
c668690d | 4083 | } else if (ret == -EAGAIN) { |
817d52f8 | 4084 | goto have_block_group; |
c668690d | 4085 | } else if (ret > 0) { |
1cdda9b8 | 4086 | goto loop; |
c668690d NA |
4087 | } |
4088 | ||
4089 | /* Checks */ | |
b4bd745d QW |
4090 | ffe_ctl.search_start = round_up(ffe_ctl.found_offset, |
4091 | fs_info->stripesize); | |
25179201 | 4092 | |
2552d17e | 4093 | /* move on to the next group */ |
b4bd745d | 4094 | if (ffe_ctl.search_start + num_bytes > |
b3470b5d | 4095 | block_group->start + block_group->length) { |
b4bd745d QW |
4096 | btrfs_add_free_space(block_group, ffe_ctl.found_offset, |
4097 | num_bytes); | |
2552d17e | 4098 | goto loop; |
6226cb0a | 4099 | } |
f5a31e16 | 4100 | |
b4bd745d QW |
4101 | if (ffe_ctl.found_offset < ffe_ctl.search_start) |
4102 | btrfs_add_free_space(block_group, ffe_ctl.found_offset, | |
4103 | ffe_ctl.search_start - ffe_ctl.found_offset); | |
2552d17e | 4104 | |
18513091 WX |
4105 | ret = btrfs_add_reserved_bytes(block_group, ram_bytes, |
4106 | num_bytes, delalloc); | |
f0486c68 | 4107 | if (ret == -EAGAIN) { |
b4bd745d QW |
4108 | btrfs_add_free_space(block_group, ffe_ctl.found_offset, |
4109 | num_bytes); | |
2552d17e | 4110 | goto loop; |
0f9dd46c | 4111 | } |
9cfa3e34 | 4112 | btrfs_inc_block_group_reservations(block_group); |
0b86a832 | 4113 | |
f0486c68 | 4114 | /* we are all good, lets return */ |
b4bd745d | 4115 | ins->objectid = ffe_ctl.search_start; |
2552d17e | 4116 | ins->offset = num_bytes; |
d2fb3437 | 4117 | |
b4bd745d QW |
4118 | trace_btrfs_reserve_extent(block_group, ffe_ctl.search_start, |
4119 | num_bytes); | |
e570fd27 | 4120 | btrfs_release_block_group(block_group, delalloc); |
2552d17e JB |
4121 | break; |
4122 | loop: | |
baba5062 | 4123 | release_block_group(block_group, &ffe_ctl, delalloc); |
14443937 | 4124 | cond_resched(); |
2552d17e JB |
4125 | } |
4126 | up_read(&space_info->groups_sem); | |
4127 | ||
15b7ee65 | 4128 | ret = find_free_extent_update_loop(fs_info, ins, &ffe_ctl, full_search); |
e72d79d6 | 4129 | if (ret > 0) |
b742bb82 YZ |
4130 | goto search; |
4131 | ||
db8fe64f | 4132 | if (ret == -ENOSPC && !cache_block_group_error) { |
b4bd745d QW |
4133 | /* |
4134 | * Use ffe_ctl->total_free_space as fallback if we can't find | |
4135 | * any contiguous hole. | |
4136 | */ | |
4137 | if (!ffe_ctl.max_extent_size) | |
4138 | ffe_ctl.max_extent_size = ffe_ctl.total_free_space; | |
4f4db217 | 4139 | spin_lock(&space_info->lock); |
b4bd745d | 4140 | space_info->max_extent_size = ffe_ctl.max_extent_size; |
4f4db217 | 4141 | spin_unlock(&space_info->lock); |
b4bd745d | 4142 | ins->offset = ffe_ctl.max_extent_size; |
db8fe64f JB |
4143 | } else if (ret == -ENOSPC) { |
4144 | ret = cache_block_group_error; | |
4f4db217 | 4145 | } |
0f70abe2 | 4146 | return ret; |
fec577fb | 4147 | } |
ec44a35c | 4148 | |
6f47c706 NB |
4149 | /* |
4150 | * btrfs_reserve_extent - entry point to the extent allocator. Tries to find a | |
4151 | * hole that is at least as big as @num_bytes. | |
4152 | * | |
4153 | * @root - The root that will contain this extent | |
4154 | * | |
4155 | * @ram_bytes - The amount of space in ram that @num_bytes take. This | |
4156 | * is used for accounting purposes. This value differs | |
4157 | * from @num_bytes only in the case of compressed extents. | |
4158 | * | |
4159 | * @num_bytes - Number of bytes to allocate on-disk. | |
4160 | * | |
4161 | * @min_alloc_size - Indicates the minimum amount of space that the | |
4162 | * allocator should try to satisfy. In some cases | |
4163 | * @num_bytes may be larger than what is required and if | |
4164 | * the filesystem is fragmented then allocation fails. | |
4165 | * However, the presence of @min_alloc_size gives a | |
4166 | * chance to try and satisfy the smaller allocation. | |
4167 | * | |
4168 | * @empty_size - A hint that you plan on doing more COW. This is the | |
4169 | * size in bytes the allocator should try to find free | |
4170 | * next to the block it returns. This is just a hint and | |
4171 | * may be ignored by the allocator. | |
4172 | * | |
4173 | * @hint_byte - Hint to the allocator to start searching above the byte | |
4174 | * address passed. It might be ignored. | |
4175 | * | |
4176 | * @ins - This key is modified to record the found hole. It will | |
4177 | * have the following values: | |
4178 | * ins->objectid == start position | |
4179 | * ins->flags = BTRFS_EXTENT_ITEM_KEY | |
4180 | * ins->offset == the size of the hole. | |
4181 | * | |
4182 | * @is_data - Boolean flag indicating whether an extent is | |
4183 | * allocated for data (true) or metadata (false) | |
4184 | * | |
4185 | * @delalloc - Boolean flag indicating whether this allocation is for | |
4186 | * delalloc or not. If 'true' data_rwsem of block groups | |
4187 | * is going to be acquired. | |
4188 | * | |
4189 | * | |
4190 | * Returns 0 when an allocation succeeded or < 0 when an error occurred. In | |
4191 | * case -ENOSPC is returned then @ins->offset will contain the size of the | |
4192 | * largest available hole the allocator managed to find. | |
4193 | */ | |
18513091 | 4194 | int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes, |
11833d66 YZ |
4195 | u64 num_bytes, u64 min_alloc_size, |
4196 | u64 empty_size, u64 hint_byte, | |
e570fd27 | 4197 | struct btrfs_key *ins, int is_data, int delalloc) |
fec577fb | 4198 | { |
ab8d0fc4 | 4199 | struct btrfs_fs_info *fs_info = root->fs_info; |
36af4e07 | 4200 | bool final_tried = num_bytes == min_alloc_size; |
b6919a58 | 4201 | u64 flags; |
fec577fb | 4202 | int ret; |
925baedd | 4203 | |
1b86826d | 4204 | flags = get_alloc_profile_by_root(root, is_data); |
98d20f67 | 4205 | again: |
0b246afa | 4206 | WARN_ON(num_bytes < fs_info->sectorsize); |
437490fe | 4207 | ret = find_free_extent(root, ram_bytes, num_bytes, empty_size, |
18513091 | 4208 | hint_byte, ins, flags, delalloc); |
9cfa3e34 | 4209 | if (!ret && !is_data) { |
ab8d0fc4 | 4210 | btrfs_dec_block_group_reservations(fs_info, ins->objectid); |
9cfa3e34 | 4211 | } else if (ret == -ENOSPC) { |
a4820398 MX |
4212 | if (!final_tried && ins->offset) { |
4213 | num_bytes = min(num_bytes >> 1, ins->offset); | |
da17066c | 4214 | num_bytes = round_down(num_bytes, |
0b246afa | 4215 | fs_info->sectorsize); |
9e622d6b | 4216 | num_bytes = max(num_bytes, min_alloc_size); |
18513091 | 4217 | ram_bytes = num_bytes; |
9e622d6b MX |
4218 | if (num_bytes == min_alloc_size) |
4219 | final_tried = true; | |
4220 | goto again; | |
ab8d0fc4 | 4221 | } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
9e622d6b MX |
4222 | struct btrfs_space_info *sinfo; |
4223 | ||
280c2908 | 4224 | sinfo = btrfs_find_space_info(fs_info, flags); |
0b246afa | 4225 | btrfs_err(fs_info, |
5d163e0e JM |
4226 | "allocation failed flags %llu, wanted %llu", |
4227 | flags, num_bytes); | |
53804280 | 4228 | if (sinfo) |
5da6afeb JB |
4229 | btrfs_dump_space_info(fs_info, sinfo, |
4230 | num_bytes, 1); | |
9e622d6b | 4231 | } |
925baedd | 4232 | } |
0f9dd46c JB |
4233 | |
4234 | return ret; | |
e6dcd2dc CM |
4235 | } |
4236 | ||
a0fbf736 NB |
4237 | int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info, |
4238 | u64 start, u64 len, int delalloc) | |
65b51a00 | 4239 | { |
32da5386 | 4240 | struct btrfs_block_group *cache; |
0f9dd46c | 4241 | |
0b246afa | 4242 | cache = btrfs_lookup_block_group(fs_info, start); |
0f9dd46c | 4243 | if (!cache) { |
0b246afa JM |
4244 | btrfs_err(fs_info, "Unable to find block group for %llu", |
4245 | start); | |
0f9dd46c JB |
4246 | return -ENOSPC; |
4247 | } | |
1f3c79a2 | 4248 | |
a0fbf736 NB |
4249 | btrfs_add_free_space(cache, start, len); |
4250 | btrfs_free_reserved_bytes(cache, len, delalloc); | |
4251 | trace_btrfs_reserved_extent_free(fs_info, start, len); | |
4252 | ||
fa9c0d79 | 4253 | btrfs_put_block_group(cache); |
a0fbf736 | 4254 | return 0; |
e6dcd2dc CM |
4255 | } |
4256 | ||
7bfc1007 NB |
4257 | int btrfs_pin_reserved_extent(struct btrfs_trans_handle *trans, u64 start, |
4258 | u64 len) | |
e688b725 | 4259 | { |
7ef54d54 | 4260 | struct btrfs_block_group *cache; |
a0fbf736 | 4261 | int ret = 0; |
7ef54d54 | 4262 | |
7bfc1007 | 4263 | cache = btrfs_lookup_block_group(trans->fs_info, start); |
7ef54d54 | 4264 | if (!cache) { |
7bfc1007 NB |
4265 | btrfs_err(trans->fs_info, "unable to find block group for %llu", |
4266 | start); | |
7ef54d54 NB |
4267 | return -ENOSPC; |
4268 | } | |
4269 | ||
6690d071 | 4270 | ret = pin_down_extent(trans, cache, start, len, 1); |
7ef54d54 | 4271 | btrfs_put_block_group(cache); |
a0fbf736 | 4272 | return ret; |
e688b725 CM |
4273 | } |
4274 | ||
5d4f98a2 | 4275 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
4276 | u64 parent, u64 root_objectid, |
4277 | u64 flags, u64 owner, u64 offset, | |
4278 | struct btrfs_key *ins, int ref_mod) | |
e6dcd2dc | 4279 | { |
ef89b824 | 4280 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e6dcd2dc | 4281 | int ret; |
e6dcd2dc | 4282 | struct btrfs_extent_item *extent_item; |
5d4f98a2 | 4283 | struct btrfs_extent_inline_ref *iref; |
e6dcd2dc | 4284 | struct btrfs_path *path; |
5d4f98a2 YZ |
4285 | struct extent_buffer *leaf; |
4286 | int type; | |
4287 | u32 size; | |
26b8003f | 4288 | |
5d4f98a2 YZ |
4289 | if (parent > 0) |
4290 | type = BTRFS_SHARED_DATA_REF_KEY; | |
4291 | else | |
4292 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
58176a96 | 4293 | |
5d4f98a2 | 4294 | size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type); |
7bb86316 CM |
4295 | |
4296 | path = btrfs_alloc_path(); | |
db5b493a TI |
4297 | if (!path) |
4298 | return -ENOMEM; | |
47e4bb98 | 4299 | |
b9473439 | 4300 | path->leave_spinning = 1; |
5d4f98a2 YZ |
4301 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
4302 | ins, size); | |
79787eaa JM |
4303 | if (ret) { |
4304 | btrfs_free_path(path); | |
4305 | return ret; | |
4306 | } | |
0f9dd46c | 4307 | |
5d4f98a2 YZ |
4308 | leaf = path->nodes[0]; |
4309 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
47e4bb98 | 4310 | struct btrfs_extent_item); |
5d4f98a2 YZ |
4311 | btrfs_set_extent_refs(leaf, extent_item, ref_mod); |
4312 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
4313 | btrfs_set_extent_flags(leaf, extent_item, | |
4314 | flags | BTRFS_EXTENT_FLAG_DATA); | |
4315 | ||
4316 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
4317 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
4318 | if (parent > 0) { | |
4319 | struct btrfs_shared_data_ref *ref; | |
4320 | ref = (struct btrfs_shared_data_ref *)(iref + 1); | |
4321 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
4322 | btrfs_set_shared_data_ref_count(leaf, ref, ref_mod); | |
4323 | } else { | |
4324 | struct btrfs_extent_data_ref *ref; | |
4325 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
4326 | btrfs_set_extent_data_ref_root(leaf, ref, root_objectid); | |
4327 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
4328 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
4329 | btrfs_set_extent_data_ref_count(leaf, ref, ref_mod); | |
4330 | } | |
47e4bb98 CM |
4331 | |
4332 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
7bb86316 | 4333 | btrfs_free_path(path); |
f510cfec | 4334 | |
25a356d3 | 4335 | ret = remove_from_free_space_tree(trans, ins->objectid, ins->offset); |
1e144fb8 OS |
4336 | if (ret) |
4337 | return ret; | |
4338 | ||
ade4b516 | 4339 | ret = btrfs_update_block_group(trans, ins->objectid, ins->offset, 1); |
79787eaa | 4340 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 4341 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
c1c9ff7c | 4342 | ins->objectid, ins->offset); |
f5947066 CM |
4343 | BUG(); |
4344 | } | |
71ff6437 | 4345 | trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, ins->offset); |
e6dcd2dc CM |
4346 | return ret; |
4347 | } | |
4348 | ||
5d4f98a2 | 4349 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, |
4e6bd4e0 | 4350 | struct btrfs_delayed_ref_node *node, |
21ebfbe7 | 4351 | struct btrfs_delayed_extent_op *extent_op) |
e6dcd2dc | 4352 | { |
9dcdbe01 | 4353 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e6dcd2dc | 4354 | int ret; |
5d4f98a2 | 4355 | struct btrfs_extent_item *extent_item; |
4e6bd4e0 | 4356 | struct btrfs_key extent_key; |
5d4f98a2 YZ |
4357 | struct btrfs_tree_block_info *block_info; |
4358 | struct btrfs_extent_inline_ref *iref; | |
4359 | struct btrfs_path *path; | |
4360 | struct extent_buffer *leaf; | |
4e6bd4e0 | 4361 | struct btrfs_delayed_tree_ref *ref; |
3173a18f | 4362 | u32 size = sizeof(*extent_item) + sizeof(*iref); |
4e6bd4e0 | 4363 | u64 num_bytes; |
21ebfbe7 | 4364 | u64 flags = extent_op->flags_to_set; |
0b246afa | 4365 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
3173a18f | 4366 | |
4e6bd4e0 NB |
4367 | ref = btrfs_delayed_node_to_tree_ref(node); |
4368 | ||
4e6bd4e0 NB |
4369 | extent_key.objectid = node->bytenr; |
4370 | if (skinny_metadata) { | |
4371 | extent_key.offset = ref->level; | |
4372 | extent_key.type = BTRFS_METADATA_ITEM_KEY; | |
4373 | num_bytes = fs_info->nodesize; | |
4374 | } else { | |
4375 | extent_key.offset = node->num_bytes; | |
4376 | extent_key.type = BTRFS_EXTENT_ITEM_KEY; | |
3173a18f | 4377 | size += sizeof(*block_info); |
4e6bd4e0 NB |
4378 | num_bytes = node->num_bytes; |
4379 | } | |
1c2308f8 | 4380 | |
5d4f98a2 | 4381 | path = btrfs_alloc_path(); |
80ee54bf | 4382 | if (!path) |
d8926bb3 | 4383 | return -ENOMEM; |
56bec294 | 4384 | |
5d4f98a2 YZ |
4385 | path->leave_spinning = 1; |
4386 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, | |
4e6bd4e0 | 4387 | &extent_key, size); |
79787eaa | 4388 | if (ret) { |
dd825259 | 4389 | btrfs_free_path(path); |
79787eaa JM |
4390 | return ret; |
4391 | } | |
5d4f98a2 YZ |
4392 | |
4393 | leaf = path->nodes[0]; | |
4394 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
4395 | struct btrfs_extent_item); | |
4396 | btrfs_set_extent_refs(leaf, extent_item, 1); | |
4397 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
4398 | btrfs_set_extent_flags(leaf, extent_item, | |
4399 | flags | BTRFS_EXTENT_FLAG_TREE_BLOCK); | |
5d4f98a2 | 4400 | |
3173a18f JB |
4401 | if (skinny_metadata) { |
4402 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
4403 | } else { | |
4404 | block_info = (struct btrfs_tree_block_info *)(extent_item + 1); | |
21ebfbe7 | 4405 | btrfs_set_tree_block_key(leaf, block_info, &extent_op->key); |
4e6bd4e0 | 4406 | btrfs_set_tree_block_level(leaf, block_info, ref->level); |
3173a18f JB |
4407 | iref = (struct btrfs_extent_inline_ref *)(block_info + 1); |
4408 | } | |
5d4f98a2 | 4409 | |
d4b20733 | 4410 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) { |
5d4f98a2 YZ |
4411 | BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); |
4412 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
4413 | BTRFS_SHARED_BLOCK_REF_KEY); | |
d4b20733 | 4414 | btrfs_set_extent_inline_ref_offset(leaf, iref, ref->parent); |
5d4f98a2 YZ |
4415 | } else { |
4416 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
4417 | BTRFS_TREE_BLOCK_REF_KEY); | |
4e6bd4e0 | 4418 | btrfs_set_extent_inline_ref_offset(leaf, iref, ref->root); |
5d4f98a2 YZ |
4419 | } |
4420 | ||
4421 | btrfs_mark_buffer_dirty(leaf); | |
4422 | btrfs_free_path(path); | |
4423 | ||
4e6bd4e0 NB |
4424 | ret = remove_from_free_space_tree(trans, extent_key.objectid, |
4425 | num_bytes); | |
1e144fb8 OS |
4426 | if (ret) |
4427 | return ret; | |
4428 | ||
ade4b516 JB |
4429 | ret = btrfs_update_block_group(trans, extent_key.objectid, |
4430 | fs_info->nodesize, 1); | |
79787eaa | 4431 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 4432 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
4e6bd4e0 | 4433 | extent_key.objectid, extent_key.offset); |
5d4f98a2 YZ |
4434 | BUG(); |
4435 | } | |
0be5dc67 | 4436 | |
4e6bd4e0 | 4437 | trace_btrfs_reserved_extent_alloc(fs_info, extent_key.objectid, |
0b246afa | 4438 | fs_info->nodesize); |
5d4f98a2 YZ |
4439 | return ret; |
4440 | } | |
4441 | ||
4442 | int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
84f7d8e6 | 4443 | struct btrfs_root *root, u64 owner, |
5846a3c2 QW |
4444 | u64 offset, u64 ram_bytes, |
4445 | struct btrfs_key *ins) | |
5d4f98a2 | 4446 | { |
76675593 | 4447 | struct btrfs_ref generic_ref = { 0 }; |
5d4f98a2 YZ |
4448 | int ret; |
4449 | ||
84f7d8e6 | 4450 | BUG_ON(root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID); |
5d4f98a2 | 4451 | |
76675593 QW |
4452 | btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT, |
4453 | ins->objectid, ins->offset, 0); | |
4454 | btrfs_init_data_ref(&generic_ref, root->root_key.objectid, owner, offset); | |
8a5040f7 | 4455 | btrfs_ref_tree_mod(root->fs_info, &generic_ref); |
76675593 QW |
4456 | ret = btrfs_add_delayed_data_ref(trans, &generic_ref, |
4457 | ram_bytes, NULL, NULL); | |
e6dcd2dc CM |
4458 | return ret; |
4459 | } | |
e02119d5 CM |
4460 | |
4461 | /* | |
4462 | * this is used by the tree logging recovery code. It records that | |
4463 | * an extent has been allocated and makes sure to clear the free | |
4464 | * space cache bits as well | |
4465 | */ | |
5d4f98a2 | 4466 | int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
4467 | u64 root_objectid, u64 owner, u64 offset, |
4468 | struct btrfs_key *ins) | |
e02119d5 | 4469 | { |
61da2abf | 4470 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e02119d5 | 4471 | int ret; |
32da5386 | 4472 | struct btrfs_block_group *block_group; |
ed7a6948 | 4473 | struct btrfs_space_info *space_info; |
11833d66 | 4474 | |
8c2a1a30 JB |
4475 | /* |
4476 | * Mixed block groups will exclude before processing the log so we only | |
01327610 | 4477 | * need to do the exclude dance if this fs isn't mixed. |
8c2a1a30 | 4478 | */ |
0b246afa | 4479 | if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) { |
2ff7e61e JM |
4480 | ret = __exclude_logged_extent(fs_info, ins->objectid, |
4481 | ins->offset); | |
b50c6e25 | 4482 | if (ret) |
8c2a1a30 | 4483 | return ret; |
11833d66 YZ |
4484 | } |
4485 | ||
0b246afa | 4486 | block_group = btrfs_lookup_block_group(fs_info, ins->objectid); |
8c2a1a30 JB |
4487 | if (!block_group) |
4488 | return -EINVAL; | |
4489 | ||
ed7a6948 WX |
4490 | space_info = block_group->space_info; |
4491 | spin_lock(&space_info->lock); | |
4492 | spin_lock(&block_group->lock); | |
4493 | space_info->bytes_reserved += ins->offset; | |
4494 | block_group->reserved += ins->offset; | |
4495 | spin_unlock(&block_group->lock); | |
4496 | spin_unlock(&space_info->lock); | |
4497 | ||
ef89b824 NB |
4498 | ret = alloc_reserved_file_extent(trans, 0, root_objectid, 0, owner, |
4499 | offset, ins, 1); | |
bd727173 | 4500 | if (ret) |
ab9b2c7b | 4501 | btrfs_pin_extent(trans, ins->objectid, ins->offset, 1); |
b50c6e25 | 4502 | btrfs_put_block_group(block_group); |
e02119d5 CM |
4503 | return ret; |
4504 | } | |
4505 | ||
48a3b636 ES |
4506 | static struct extent_buffer * |
4507 | btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
bc877d28 | 4508 | u64 bytenr, int level, u64 owner) |
65b51a00 | 4509 | { |
0b246afa | 4510 | struct btrfs_fs_info *fs_info = root->fs_info; |
65b51a00 CM |
4511 | struct extent_buffer *buf; |
4512 | ||
2ff7e61e | 4513 | buf = btrfs_find_create_tree_block(fs_info, bytenr); |
c871b0f2 LB |
4514 | if (IS_ERR(buf)) |
4515 | return buf; | |
4516 | ||
b72c3aba QW |
4517 | /* |
4518 | * Extra safety check in case the extent tree is corrupted and extent | |
4519 | * allocator chooses to use a tree block which is already used and | |
4520 | * locked. | |
4521 | */ | |
4522 | if (buf->lock_owner == current->pid) { | |
4523 | btrfs_err_rl(fs_info, | |
4524 | "tree block %llu owner %llu already locked by pid=%d, extent tree corruption detected", | |
4525 | buf->start, btrfs_header_owner(buf), current->pid); | |
4526 | free_extent_buffer(buf); | |
4527 | return ERR_PTR(-EUCLEAN); | |
4528 | } | |
4529 | ||
ad244665 | 4530 | btrfs_set_buffer_lockdep_class(owner, buf, level); |
65b51a00 | 4531 | btrfs_tree_lock(buf); |
6a884d7d | 4532 | btrfs_clean_tree_block(buf); |
3083ee2e | 4533 | clear_bit(EXTENT_BUFFER_STALE, &buf->bflags); |
b4ce94de | 4534 | |
8bead258 | 4535 | btrfs_set_lock_blocking_write(buf); |
4db8c528 | 4536 | set_extent_buffer_uptodate(buf); |
b4ce94de | 4537 | |
bc877d28 NB |
4538 | memzero_extent_buffer(buf, 0, sizeof(struct btrfs_header)); |
4539 | btrfs_set_header_level(buf, level); | |
4540 | btrfs_set_header_bytenr(buf, buf->start); | |
4541 | btrfs_set_header_generation(buf, trans->transid); | |
4542 | btrfs_set_header_backref_rev(buf, BTRFS_MIXED_BACKREF_REV); | |
4543 | btrfs_set_header_owner(buf, owner); | |
de37aa51 | 4544 | write_extent_buffer_fsid(buf, fs_info->fs_devices->metadata_uuid); |
bc877d28 | 4545 | write_extent_buffer_chunk_tree_uuid(buf, fs_info->chunk_tree_uuid); |
d0c803c4 | 4546 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { |
656f30db | 4547 | buf->log_index = root->log_transid % 2; |
8cef4e16 YZ |
4548 | /* |
4549 | * we allow two log transactions at a time, use different | |
52042d8e | 4550 | * EXTENT bit to differentiate dirty pages. |
8cef4e16 | 4551 | */ |
656f30db | 4552 | if (buf->log_index == 0) |
8cef4e16 YZ |
4553 | set_extent_dirty(&root->dirty_log_pages, buf->start, |
4554 | buf->start + buf->len - 1, GFP_NOFS); | |
4555 | else | |
4556 | set_extent_new(&root->dirty_log_pages, buf->start, | |
3744dbeb | 4557 | buf->start + buf->len - 1); |
d0c803c4 | 4558 | } else { |
656f30db | 4559 | buf->log_index = -1; |
d0c803c4 | 4560 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, |
65b51a00 | 4561 | buf->start + buf->len - 1, GFP_NOFS); |
d0c803c4 | 4562 | } |
64c12921 | 4563 | trans->dirty = true; |
b4ce94de | 4564 | /* this returns a buffer locked for blocking */ |
65b51a00 CM |
4565 | return buf; |
4566 | } | |
4567 | ||
fec577fb | 4568 | /* |
f0486c68 | 4569 | * finds a free extent and does all the dirty work required for allocation |
67b7859e | 4570 | * returns the tree buffer or an ERR_PTR on error. |
fec577fb | 4571 | */ |
4d75f8a9 | 4572 | struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans, |
310712b2 OS |
4573 | struct btrfs_root *root, |
4574 | u64 parent, u64 root_objectid, | |
4575 | const struct btrfs_disk_key *key, | |
4576 | int level, u64 hint, | |
4577 | u64 empty_size) | |
fec577fb | 4578 | { |
0b246afa | 4579 | struct btrfs_fs_info *fs_info = root->fs_info; |
e2fa7227 | 4580 | struct btrfs_key ins; |
f0486c68 | 4581 | struct btrfs_block_rsv *block_rsv; |
5f39d397 | 4582 | struct extent_buffer *buf; |
67b7859e | 4583 | struct btrfs_delayed_extent_op *extent_op; |
ed4f255b | 4584 | struct btrfs_ref generic_ref = { 0 }; |
f0486c68 YZ |
4585 | u64 flags = 0; |
4586 | int ret; | |
0b246afa JM |
4587 | u32 blocksize = fs_info->nodesize; |
4588 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); | |
fec577fb | 4589 | |
05653ef3 | 4590 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
0b246afa | 4591 | if (btrfs_is_testing(fs_info)) { |
faa2dbf0 | 4592 | buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr, |
bc877d28 | 4593 | level, root_objectid); |
faa2dbf0 JB |
4594 | if (!IS_ERR(buf)) |
4595 | root->alloc_bytenr += blocksize; | |
4596 | return buf; | |
4597 | } | |
05653ef3 | 4598 | #endif |
fccb84c9 | 4599 | |
67f9c220 | 4600 | block_rsv = btrfs_use_block_rsv(trans, root, blocksize); |
f0486c68 YZ |
4601 | if (IS_ERR(block_rsv)) |
4602 | return ERR_CAST(block_rsv); | |
4603 | ||
18513091 | 4604 | ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize, |
e570fd27 | 4605 | empty_size, hint, &ins, 0, 0); |
67b7859e OS |
4606 | if (ret) |
4607 | goto out_unuse; | |
55c69072 | 4608 | |
bc877d28 NB |
4609 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, level, |
4610 | root_objectid); | |
67b7859e OS |
4611 | if (IS_ERR(buf)) { |
4612 | ret = PTR_ERR(buf); | |
4613 | goto out_free_reserved; | |
4614 | } | |
f0486c68 YZ |
4615 | |
4616 | if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
4617 | if (parent == 0) | |
4618 | parent = ins.objectid; | |
4619 | flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
4620 | } else | |
4621 | BUG_ON(parent > 0); | |
4622 | ||
4623 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { | |
78a6184a | 4624 | extent_op = btrfs_alloc_delayed_extent_op(); |
67b7859e OS |
4625 | if (!extent_op) { |
4626 | ret = -ENOMEM; | |
4627 | goto out_free_buf; | |
4628 | } | |
f0486c68 YZ |
4629 | if (key) |
4630 | memcpy(&extent_op->key, key, sizeof(extent_op->key)); | |
4631 | else | |
4632 | memset(&extent_op->key, 0, sizeof(extent_op->key)); | |
4633 | extent_op->flags_to_set = flags; | |
35b3ad50 DS |
4634 | extent_op->update_key = skinny_metadata ? false : true; |
4635 | extent_op->update_flags = true; | |
4636 | extent_op->is_data = false; | |
b1c79e09 | 4637 | extent_op->level = level; |
f0486c68 | 4638 | |
ed4f255b QW |
4639 | btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT, |
4640 | ins.objectid, ins.offset, parent); | |
4641 | generic_ref.real_root = root->root_key.objectid; | |
4642 | btrfs_init_tree_ref(&generic_ref, level, root_objectid); | |
8a5040f7 | 4643 | btrfs_ref_tree_mod(fs_info, &generic_ref); |
ed4f255b | 4644 | ret = btrfs_add_delayed_tree_ref(trans, &generic_ref, |
7be07912 | 4645 | extent_op, NULL, NULL); |
67b7859e OS |
4646 | if (ret) |
4647 | goto out_free_delayed; | |
f0486c68 | 4648 | } |
fec577fb | 4649 | return buf; |
67b7859e OS |
4650 | |
4651 | out_free_delayed: | |
4652 | btrfs_free_delayed_extent_op(extent_op); | |
4653 | out_free_buf: | |
4654 | free_extent_buffer(buf); | |
4655 | out_free_reserved: | |
2ff7e61e | 4656 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0); |
67b7859e | 4657 | out_unuse: |
67f9c220 | 4658 | btrfs_unuse_block_rsv(fs_info, block_rsv, blocksize); |
67b7859e | 4659 | return ERR_PTR(ret); |
fec577fb | 4660 | } |
a28ec197 | 4661 | |
2c47e605 YZ |
4662 | struct walk_control { |
4663 | u64 refs[BTRFS_MAX_LEVEL]; | |
4664 | u64 flags[BTRFS_MAX_LEVEL]; | |
4665 | struct btrfs_key update_progress; | |
aea6f028 JB |
4666 | struct btrfs_key drop_progress; |
4667 | int drop_level; | |
2c47e605 YZ |
4668 | int stage; |
4669 | int level; | |
4670 | int shared_level; | |
4671 | int update_ref; | |
4672 | int keep_locks; | |
1c4850e2 YZ |
4673 | int reada_slot; |
4674 | int reada_count; | |
78c52d9e | 4675 | int restarted; |
2c47e605 YZ |
4676 | }; |
4677 | ||
4678 | #define DROP_REFERENCE 1 | |
4679 | #define UPDATE_BACKREF 2 | |
4680 | ||
1c4850e2 YZ |
4681 | static noinline void reada_walk_down(struct btrfs_trans_handle *trans, |
4682 | struct btrfs_root *root, | |
4683 | struct walk_control *wc, | |
4684 | struct btrfs_path *path) | |
6407bf6d | 4685 | { |
0b246afa | 4686 | struct btrfs_fs_info *fs_info = root->fs_info; |
1c4850e2 YZ |
4687 | u64 bytenr; |
4688 | u64 generation; | |
4689 | u64 refs; | |
94fcca9f | 4690 | u64 flags; |
5d4f98a2 | 4691 | u32 nritems; |
1c4850e2 YZ |
4692 | struct btrfs_key key; |
4693 | struct extent_buffer *eb; | |
6407bf6d | 4694 | int ret; |
1c4850e2 YZ |
4695 | int slot; |
4696 | int nread = 0; | |
6407bf6d | 4697 | |
1c4850e2 YZ |
4698 | if (path->slots[wc->level] < wc->reada_slot) { |
4699 | wc->reada_count = wc->reada_count * 2 / 3; | |
4700 | wc->reada_count = max(wc->reada_count, 2); | |
4701 | } else { | |
4702 | wc->reada_count = wc->reada_count * 3 / 2; | |
4703 | wc->reada_count = min_t(int, wc->reada_count, | |
0b246afa | 4704 | BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
1c4850e2 | 4705 | } |
7bb86316 | 4706 | |
1c4850e2 YZ |
4707 | eb = path->nodes[wc->level]; |
4708 | nritems = btrfs_header_nritems(eb); | |
bd56b302 | 4709 | |
1c4850e2 YZ |
4710 | for (slot = path->slots[wc->level]; slot < nritems; slot++) { |
4711 | if (nread >= wc->reada_count) | |
4712 | break; | |
bd56b302 | 4713 | |
2dd3e67b | 4714 | cond_resched(); |
1c4850e2 YZ |
4715 | bytenr = btrfs_node_blockptr(eb, slot); |
4716 | generation = btrfs_node_ptr_generation(eb, slot); | |
2dd3e67b | 4717 | |
1c4850e2 YZ |
4718 | if (slot == path->slots[wc->level]) |
4719 | goto reada; | |
5d4f98a2 | 4720 | |
1c4850e2 YZ |
4721 | if (wc->stage == UPDATE_BACKREF && |
4722 | generation <= root->root_key.offset) | |
bd56b302 CM |
4723 | continue; |
4724 | ||
94fcca9f | 4725 | /* We don't lock the tree block, it's OK to be racy here */ |
2ff7e61e | 4726 | ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, |
3173a18f JB |
4727 | wc->level - 1, 1, &refs, |
4728 | &flags); | |
79787eaa JM |
4729 | /* We don't care about errors in readahead. */ |
4730 | if (ret < 0) | |
4731 | continue; | |
94fcca9f YZ |
4732 | BUG_ON(refs == 0); |
4733 | ||
1c4850e2 | 4734 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 YZ |
4735 | if (refs == 1) |
4736 | goto reada; | |
bd56b302 | 4737 | |
94fcca9f YZ |
4738 | if (wc->level == 1 && |
4739 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
4740 | continue; | |
1c4850e2 YZ |
4741 | if (!wc->update_ref || |
4742 | generation <= root->root_key.offset) | |
4743 | continue; | |
4744 | btrfs_node_key_to_cpu(eb, &key, slot); | |
4745 | ret = btrfs_comp_cpu_keys(&key, | |
4746 | &wc->update_progress); | |
4747 | if (ret < 0) | |
4748 | continue; | |
94fcca9f YZ |
4749 | } else { |
4750 | if (wc->level == 1 && | |
4751 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
4752 | continue; | |
6407bf6d | 4753 | } |
1c4850e2 | 4754 | reada: |
2ff7e61e | 4755 | readahead_tree_block(fs_info, bytenr); |
1c4850e2 | 4756 | nread++; |
20524f02 | 4757 | } |
1c4850e2 | 4758 | wc->reada_slot = slot; |
20524f02 | 4759 | } |
2c47e605 | 4760 | |
f82d02d9 | 4761 | /* |
2c016dc2 | 4762 | * helper to process tree block while walking down the tree. |
2c47e605 | 4763 | * |
2c47e605 YZ |
4764 | * when wc->stage == UPDATE_BACKREF, this function updates |
4765 | * back refs for pointers in the block. | |
4766 | * | |
4767 | * NOTE: return value 1 means we should stop walking down. | |
f82d02d9 | 4768 | */ |
2c47e605 | 4769 | static noinline int walk_down_proc(struct btrfs_trans_handle *trans, |
5d4f98a2 | 4770 | struct btrfs_root *root, |
2c47e605 | 4771 | struct btrfs_path *path, |
94fcca9f | 4772 | struct walk_control *wc, int lookup_info) |
f82d02d9 | 4773 | { |
2ff7e61e | 4774 | struct btrfs_fs_info *fs_info = root->fs_info; |
2c47e605 YZ |
4775 | int level = wc->level; |
4776 | struct extent_buffer *eb = path->nodes[level]; | |
2c47e605 | 4777 | u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF; |
f82d02d9 YZ |
4778 | int ret; |
4779 | ||
2c47e605 YZ |
4780 | if (wc->stage == UPDATE_BACKREF && |
4781 | btrfs_header_owner(eb) != root->root_key.objectid) | |
4782 | return 1; | |
f82d02d9 | 4783 | |
2c47e605 YZ |
4784 | /* |
4785 | * when reference count of tree block is 1, it won't increase | |
4786 | * again. once full backref flag is set, we never clear it. | |
4787 | */ | |
94fcca9f YZ |
4788 | if (lookup_info && |
4789 | ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) || | |
4790 | (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) { | |
2c47e605 | 4791 | BUG_ON(!path->locks[level]); |
2ff7e61e | 4792 | ret = btrfs_lookup_extent_info(trans, fs_info, |
3173a18f | 4793 | eb->start, level, 1, |
2c47e605 YZ |
4794 | &wc->refs[level], |
4795 | &wc->flags[level]); | |
79787eaa JM |
4796 | BUG_ON(ret == -ENOMEM); |
4797 | if (ret) | |
4798 | return ret; | |
2c47e605 YZ |
4799 | BUG_ON(wc->refs[level] == 0); |
4800 | } | |
5d4f98a2 | 4801 | |
2c47e605 YZ |
4802 | if (wc->stage == DROP_REFERENCE) { |
4803 | if (wc->refs[level] > 1) | |
4804 | return 1; | |
f82d02d9 | 4805 | |
2c47e605 | 4806 | if (path->locks[level] && !wc->keep_locks) { |
bd681513 | 4807 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
4808 | path->locks[level] = 0; |
4809 | } | |
4810 | return 0; | |
4811 | } | |
f82d02d9 | 4812 | |
2c47e605 YZ |
4813 | /* wc->stage == UPDATE_BACKREF */ |
4814 | if (!(wc->flags[level] & flag)) { | |
4815 | BUG_ON(!path->locks[level]); | |
e339a6b0 | 4816 | ret = btrfs_inc_ref(trans, root, eb, 1); |
79787eaa | 4817 | BUG_ON(ret); /* -ENOMEM */ |
e339a6b0 | 4818 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 4819 | BUG_ON(ret); /* -ENOMEM */ |
42c9d0b5 | 4820 | ret = btrfs_set_disk_extent_flags(trans, eb, flag, |
b1c79e09 | 4821 | btrfs_header_level(eb), 0); |
79787eaa | 4822 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 YZ |
4823 | wc->flags[level] |= flag; |
4824 | } | |
4825 | ||
4826 | /* | |
4827 | * the block is shared by multiple trees, so it's not good to | |
4828 | * keep the tree lock | |
4829 | */ | |
4830 | if (path->locks[level] && level > 0) { | |
bd681513 | 4831 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
4832 | path->locks[level] = 0; |
4833 | } | |
4834 | return 0; | |
4835 | } | |
4836 | ||
78c52d9e JB |
4837 | /* |
4838 | * This is used to verify a ref exists for this root to deal with a bug where we | |
4839 | * would have a drop_progress key that hadn't been updated properly. | |
4840 | */ | |
4841 | static int check_ref_exists(struct btrfs_trans_handle *trans, | |
4842 | struct btrfs_root *root, u64 bytenr, u64 parent, | |
4843 | int level) | |
4844 | { | |
4845 | struct btrfs_path *path; | |
4846 | struct btrfs_extent_inline_ref *iref; | |
4847 | int ret; | |
4848 | ||
4849 | path = btrfs_alloc_path(); | |
4850 | if (!path) | |
4851 | return -ENOMEM; | |
4852 | ||
4853 | ret = lookup_extent_backref(trans, path, &iref, bytenr, | |
4854 | root->fs_info->nodesize, parent, | |
4855 | root->root_key.objectid, level, 0); | |
4856 | btrfs_free_path(path); | |
4857 | if (ret == -ENOENT) | |
4858 | return 0; | |
4859 | if (ret < 0) | |
4860 | return ret; | |
4861 | return 1; | |
4862 | } | |
4863 | ||
1c4850e2 | 4864 | /* |
2c016dc2 | 4865 | * helper to process tree block pointer. |
1c4850e2 YZ |
4866 | * |
4867 | * when wc->stage == DROP_REFERENCE, this function checks | |
4868 | * reference count of the block pointed to. if the block | |
4869 | * is shared and we need update back refs for the subtree | |
4870 | * rooted at the block, this function changes wc->stage to | |
4871 | * UPDATE_BACKREF. if the block is shared and there is no | |
4872 | * need to update back, this function drops the reference | |
4873 | * to the block. | |
4874 | * | |
4875 | * NOTE: return value 1 means we should stop walking down. | |
4876 | */ | |
4877 | static noinline int do_walk_down(struct btrfs_trans_handle *trans, | |
4878 | struct btrfs_root *root, | |
4879 | struct btrfs_path *path, | |
94fcca9f | 4880 | struct walk_control *wc, int *lookup_info) |
1c4850e2 | 4881 | { |
0b246afa | 4882 | struct btrfs_fs_info *fs_info = root->fs_info; |
1c4850e2 YZ |
4883 | u64 bytenr; |
4884 | u64 generation; | |
4885 | u64 parent; | |
1c4850e2 | 4886 | struct btrfs_key key; |
581c1760 | 4887 | struct btrfs_key first_key; |
ffd4bb2a | 4888 | struct btrfs_ref ref = { 0 }; |
1c4850e2 YZ |
4889 | struct extent_buffer *next; |
4890 | int level = wc->level; | |
4891 | int reada = 0; | |
4892 | int ret = 0; | |
1152651a | 4893 | bool need_account = false; |
1c4850e2 YZ |
4894 | |
4895 | generation = btrfs_node_ptr_generation(path->nodes[level], | |
4896 | path->slots[level]); | |
4897 | /* | |
4898 | * if the lower level block was created before the snapshot | |
4899 | * was created, we know there is no need to update back refs | |
4900 | * for the subtree | |
4901 | */ | |
4902 | if (wc->stage == UPDATE_BACKREF && | |
94fcca9f YZ |
4903 | generation <= root->root_key.offset) { |
4904 | *lookup_info = 1; | |
1c4850e2 | 4905 | return 1; |
94fcca9f | 4906 | } |
1c4850e2 YZ |
4907 | |
4908 | bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]); | |
581c1760 QW |
4909 | btrfs_node_key_to_cpu(path->nodes[level], &first_key, |
4910 | path->slots[level]); | |
1c4850e2 | 4911 | |
0b246afa | 4912 | next = find_extent_buffer(fs_info, bytenr); |
1c4850e2 | 4913 | if (!next) { |
2ff7e61e | 4914 | next = btrfs_find_create_tree_block(fs_info, bytenr); |
c871b0f2 LB |
4915 | if (IS_ERR(next)) |
4916 | return PTR_ERR(next); | |
4917 | ||
b2aaaa3b JB |
4918 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, next, |
4919 | level - 1); | |
1c4850e2 YZ |
4920 | reada = 1; |
4921 | } | |
4922 | btrfs_tree_lock(next); | |
8bead258 | 4923 | btrfs_set_lock_blocking_write(next); |
1c4850e2 | 4924 | |
2ff7e61e | 4925 | ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1, |
94fcca9f YZ |
4926 | &wc->refs[level - 1], |
4927 | &wc->flags[level - 1]); | |
4867268c JB |
4928 | if (ret < 0) |
4929 | goto out_unlock; | |
79787eaa | 4930 | |
c2cf52eb | 4931 | if (unlikely(wc->refs[level - 1] == 0)) { |
0b246afa | 4932 | btrfs_err(fs_info, "Missing references."); |
4867268c JB |
4933 | ret = -EIO; |
4934 | goto out_unlock; | |
c2cf52eb | 4935 | } |
94fcca9f | 4936 | *lookup_info = 0; |
1c4850e2 | 4937 | |
94fcca9f | 4938 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 | 4939 | if (wc->refs[level - 1] > 1) { |
1152651a | 4940 | need_account = true; |
94fcca9f YZ |
4941 | if (level == 1 && |
4942 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
4943 | goto skip; | |
4944 | ||
1c4850e2 YZ |
4945 | if (!wc->update_ref || |
4946 | generation <= root->root_key.offset) | |
4947 | goto skip; | |
4948 | ||
4949 | btrfs_node_key_to_cpu(path->nodes[level], &key, | |
4950 | path->slots[level]); | |
4951 | ret = btrfs_comp_cpu_keys(&key, &wc->update_progress); | |
4952 | if (ret < 0) | |
4953 | goto skip; | |
4954 | ||
4955 | wc->stage = UPDATE_BACKREF; | |
4956 | wc->shared_level = level - 1; | |
4957 | } | |
94fcca9f YZ |
4958 | } else { |
4959 | if (level == 1 && | |
4960 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
4961 | goto skip; | |
1c4850e2 YZ |
4962 | } |
4963 | ||
b9fab919 | 4964 | if (!btrfs_buffer_uptodate(next, generation, 0)) { |
1c4850e2 YZ |
4965 | btrfs_tree_unlock(next); |
4966 | free_extent_buffer(next); | |
4967 | next = NULL; | |
94fcca9f | 4968 | *lookup_info = 1; |
1c4850e2 YZ |
4969 | } |
4970 | ||
4971 | if (!next) { | |
4972 | if (reada && level == 1) | |
4973 | reada_walk_down(trans, root, wc, path); | |
581c1760 QW |
4974 | next = read_tree_block(fs_info, bytenr, generation, level - 1, |
4975 | &first_key); | |
64c043de LB |
4976 | if (IS_ERR(next)) { |
4977 | return PTR_ERR(next); | |
4978 | } else if (!extent_buffer_uptodate(next)) { | |
416bc658 | 4979 | free_extent_buffer(next); |
97d9a8a4 | 4980 | return -EIO; |
416bc658 | 4981 | } |
1c4850e2 | 4982 | btrfs_tree_lock(next); |
8bead258 | 4983 | btrfs_set_lock_blocking_write(next); |
1c4850e2 YZ |
4984 | } |
4985 | ||
4986 | level--; | |
4867268c JB |
4987 | ASSERT(level == btrfs_header_level(next)); |
4988 | if (level != btrfs_header_level(next)) { | |
4989 | btrfs_err(root->fs_info, "mismatched level"); | |
4990 | ret = -EIO; | |
4991 | goto out_unlock; | |
4992 | } | |
1c4850e2 YZ |
4993 | path->nodes[level] = next; |
4994 | path->slots[level] = 0; | |
bd681513 | 4995 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
1c4850e2 YZ |
4996 | wc->level = level; |
4997 | if (wc->level == 1) | |
4998 | wc->reada_slot = 0; | |
4999 | return 0; | |
5000 | skip: | |
5001 | wc->refs[level - 1] = 0; | |
5002 | wc->flags[level - 1] = 0; | |
94fcca9f YZ |
5003 | if (wc->stage == DROP_REFERENCE) { |
5004 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
5005 | parent = path->nodes[level]->start; | |
5006 | } else { | |
4867268c | 5007 | ASSERT(root->root_key.objectid == |
94fcca9f | 5008 | btrfs_header_owner(path->nodes[level])); |
4867268c JB |
5009 | if (root->root_key.objectid != |
5010 | btrfs_header_owner(path->nodes[level])) { | |
5011 | btrfs_err(root->fs_info, | |
5012 | "mismatched block owner"); | |
5013 | ret = -EIO; | |
5014 | goto out_unlock; | |
5015 | } | |
94fcca9f YZ |
5016 | parent = 0; |
5017 | } | |
1c4850e2 | 5018 | |
78c52d9e JB |
5019 | /* |
5020 | * If we had a drop_progress we need to verify the refs are set | |
5021 | * as expected. If we find our ref then we know that from here | |
5022 | * on out everything should be correct, and we can clear the | |
5023 | * ->restarted flag. | |
5024 | */ | |
5025 | if (wc->restarted) { | |
5026 | ret = check_ref_exists(trans, root, bytenr, parent, | |
5027 | level - 1); | |
5028 | if (ret < 0) | |
5029 | goto out_unlock; | |
5030 | if (ret == 0) | |
5031 | goto no_delete; | |
5032 | ret = 0; | |
5033 | wc->restarted = 0; | |
5034 | } | |
5035 | ||
2cd86d30 QW |
5036 | /* |
5037 | * Reloc tree doesn't contribute to qgroup numbers, and we have | |
5038 | * already accounted them at merge time (replace_path), | |
5039 | * thus we could skip expensive subtree trace here. | |
5040 | */ | |
5041 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID && | |
5042 | need_account) { | |
deb40627 | 5043 | ret = btrfs_qgroup_trace_subtree(trans, next, |
33d1f05c | 5044 | generation, level - 1); |
1152651a | 5045 | if (ret) { |
0b246afa | 5046 | btrfs_err_rl(fs_info, |
5d163e0e JM |
5047 | "Error %d accounting shared subtree. Quota is out of sync, rescan required.", |
5048 | ret); | |
1152651a MF |
5049 | } |
5050 | } | |
aea6f028 JB |
5051 | |
5052 | /* | |
5053 | * We need to update the next key in our walk control so we can | |
5054 | * update the drop_progress key accordingly. We don't care if | |
5055 | * find_next_key doesn't find a key because that means we're at | |
5056 | * the end and are going to clean up now. | |
5057 | */ | |
5058 | wc->drop_level = level; | |
5059 | find_next_key(path, level, &wc->drop_progress); | |
5060 | ||
ffd4bb2a QW |
5061 | btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr, |
5062 | fs_info->nodesize, parent); | |
5063 | btrfs_init_tree_ref(&ref, level - 1, root->root_key.objectid); | |
5064 | ret = btrfs_free_extent(trans, &ref); | |
4867268c JB |
5065 | if (ret) |
5066 | goto out_unlock; | |
1c4850e2 | 5067 | } |
78c52d9e | 5068 | no_delete: |
4867268c JB |
5069 | *lookup_info = 1; |
5070 | ret = 1; | |
5071 | ||
5072 | out_unlock: | |
1c4850e2 YZ |
5073 | btrfs_tree_unlock(next); |
5074 | free_extent_buffer(next); | |
4867268c JB |
5075 | |
5076 | return ret; | |
1c4850e2 YZ |
5077 | } |
5078 | ||
2c47e605 | 5079 | /* |
2c016dc2 | 5080 | * helper to process tree block while walking up the tree. |
2c47e605 YZ |
5081 | * |
5082 | * when wc->stage == DROP_REFERENCE, this function drops | |
5083 | * reference count on the block. | |
5084 | * | |
5085 | * when wc->stage == UPDATE_BACKREF, this function changes | |
5086 | * wc->stage back to DROP_REFERENCE if we changed wc->stage | |
5087 | * to UPDATE_BACKREF previously while processing the block. | |
5088 | * | |
5089 | * NOTE: return value 1 means we should stop walking up. | |
5090 | */ | |
5091 | static noinline int walk_up_proc(struct btrfs_trans_handle *trans, | |
5092 | struct btrfs_root *root, | |
5093 | struct btrfs_path *path, | |
5094 | struct walk_control *wc) | |
5095 | { | |
0b246afa | 5096 | struct btrfs_fs_info *fs_info = root->fs_info; |
f0486c68 | 5097 | int ret; |
2c47e605 YZ |
5098 | int level = wc->level; |
5099 | struct extent_buffer *eb = path->nodes[level]; | |
5100 | u64 parent = 0; | |
5101 | ||
5102 | if (wc->stage == UPDATE_BACKREF) { | |
5103 | BUG_ON(wc->shared_level < level); | |
5104 | if (level < wc->shared_level) | |
5105 | goto out; | |
5106 | ||
2c47e605 YZ |
5107 | ret = find_next_key(path, level + 1, &wc->update_progress); |
5108 | if (ret > 0) | |
5109 | wc->update_ref = 0; | |
5110 | ||
5111 | wc->stage = DROP_REFERENCE; | |
5112 | wc->shared_level = -1; | |
5113 | path->slots[level] = 0; | |
5114 | ||
5115 | /* | |
5116 | * check reference count again if the block isn't locked. | |
5117 | * we should start walking down the tree again if reference | |
5118 | * count is one. | |
5119 | */ | |
5120 | if (!path->locks[level]) { | |
5121 | BUG_ON(level == 0); | |
5122 | btrfs_tree_lock(eb); | |
8bead258 | 5123 | btrfs_set_lock_blocking_write(eb); |
bd681513 | 5124 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 5125 | |
2ff7e61e | 5126 | ret = btrfs_lookup_extent_info(trans, fs_info, |
3173a18f | 5127 | eb->start, level, 1, |
2c47e605 YZ |
5128 | &wc->refs[level], |
5129 | &wc->flags[level]); | |
79787eaa JM |
5130 | if (ret < 0) { |
5131 | btrfs_tree_unlock_rw(eb, path->locks[level]); | |
3268a246 | 5132 | path->locks[level] = 0; |
79787eaa JM |
5133 | return ret; |
5134 | } | |
2c47e605 YZ |
5135 | BUG_ON(wc->refs[level] == 0); |
5136 | if (wc->refs[level] == 1) { | |
bd681513 | 5137 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
3268a246 | 5138 | path->locks[level] = 0; |
2c47e605 YZ |
5139 | return 1; |
5140 | } | |
f82d02d9 | 5141 | } |
2c47e605 | 5142 | } |
f82d02d9 | 5143 | |
2c47e605 YZ |
5144 | /* wc->stage == DROP_REFERENCE */ |
5145 | BUG_ON(wc->refs[level] > 1 && !path->locks[level]); | |
5d4f98a2 | 5146 | |
2c47e605 YZ |
5147 | if (wc->refs[level] == 1) { |
5148 | if (level == 0) { | |
5149 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
e339a6b0 | 5150 | ret = btrfs_dec_ref(trans, root, eb, 1); |
2c47e605 | 5151 | else |
e339a6b0 | 5152 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 5153 | BUG_ON(ret); /* -ENOMEM */ |
c4140cbf QW |
5154 | if (is_fstree(root->root_key.objectid)) { |
5155 | ret = btrfs_qgroup_trace_leaf_items(trans, eb); | |
5156 | if (ret) { | |
5157 | btrfs_err_rl(fs_info, | |
5158 | "error %d accounting leaf items, quota is out of sync, rescan required", | |
5d163e0e | 5159 | ret); |
c4140cbf | 5160 | } |
1152651a | 5161 | } |
2c47e605 | 5162 | } |
6a884d7d | 5163 | /* make block locked assertion in btrfs_clean_tree_block happy */ |
2c47e605 YZ |
5164 | if (!path->locks[level] && |
5165 | btrfs_header_generation(eb) == trans->transid) { | |
5166 | btrfs_tree_lock(eb); | |
8bead258 | 5167 | btrfs_set_lock_blocking_write(eb); |
bd681513 | 5168 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 5169 | } |
6a884d7d | 5170 | btrfs_clean_tree_block(eb); |
2c47e605 YZ |
5171 | } |
5172 | ||
5173 | if (eb == root->node) { | |
5174 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
5175 | parent = eb->start; | |
65c6e82b QW |
5176 | else if (root->root_key.objectid != btrfs_header_owner(eb)) |
5177 | goto owner_mismatch; | |
2c47e605 YZ |
5178 | } else { |
5179 | if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
5180 | parent = path->nodes[level + 1]->start; | |
65c6e82b QW |
5181 | else if (root->root_key.objectid != |
5182 | btrfs_header_owner(path->nodes[level + 1])) | |
5183 | goto owner_mismatch; | |
f82d02d9 | 5184 | } |
f82d02d9 | 5185 | |
5581a51a | 5186 | btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1); |
2c47e605 YZ |
5187 | out: |
5188 | wc->refs[level] = 0; | |
5189 | wc->flags[level] = 0; | |
f0486c68 | 5190 | return 0; |
65c6e82b QW |
5191 | |
5192 | owner_mismatch: | |
5193 | btrfs_err_rl(fs_info, "unexpected tree owner, have %llu expect %llu", | |
5194 | btrfs_header_owner(eb), root->root_key.objectid); | |
5195 | return -EUCLEAN; | |
2c47e605 YZ |
5196 | } |
5197 | ||
5198 | static noinline int walk_down_tree(struct btrfs_trans_handle *trans, | |
5199 | struct btrfs_root *root, | |
5200 | struct btrfs_path *path, | |
5201 | struct walk_control *wc) | |
5202 | { | |
2c47e605 | 5203 | int level = wc->level; |
94fcca9f | 5204 | int lookup_info = 1; |
2c47e605 YZ |
5205 | int ret; |
5206 | ||
5207 | while (level >= 0) { | |
94fcca9f | 5208 | ret = walk_down_proc(trans, root, path, wc, lookup_info); |
2c47e605 YZ |
5209 | if (ret > 0) |
5210 | break; | |
5211 | ||
5212 | if (level == 0) | |
5213 | break; | |
5214 | ||
7a7965f8 YZ |
5215 | if (path->slots[level] >= |
5216 | btrfs_header_nritems(path->nodes[level])) | |
5217 | break; | |
5218 | ||
94fcca9f | 5219 | ret = do_walk_down(trans, root, path, wc, &lookup_info); |
1c4850e2 YZ |
5220 | if (ret > 0) { |
5221 | path->slots[level]++; | |
5222 | continue; | |
90d2c51d MX |
5223 | } else if (ret < 0) |
5224 | return ret; | |
1c4850e2 | 5225 | level = wc->level; |
f82d02d9 | 5226 | } |
f82d02d9 YZ |
5227 | return 0; |
5228 | } | |
5229 | ||
d397712b | 5230 | static noinline int walk_up_tree(struct btrfs_trans_handle *trans, |
98ed5174 | 5231 | struct btrfs_root *root, |
f82d02d9 | 5232 | struct btrfs_path *path, |
2c47e605 | 5233 | struct walk_control *wc, int max_level) |
20524f02 | 5234 | { |
2c47e605 | 5235 | int level = wc->level; |
20524f02 | 5236 | int ret; |
9f3a7427 | 5237 | |
2c47e605 YZ |
5238 | path->slots[level] = btrfs_header_nritems(path->nodes[level]); |
5239 | while (level < max_level && path->nodes[level]) { | |
5240 | wc->level = level; | |
5241 | if (path->slots[level] + 1 < | |
5242 | btrfs_header_nritems(path->nodes[level])) { | |
5243 | path->slots[level]++; | |
20524f02 CM |
5244 | return 0; |
5245 | } else { | |
2c47e605 YZ |
5246 | ret = walk_up_proc(trans, root, path, wc); |
5247 | if (ret > 0) | |
5248 | return 0; | |
65c6e82b QW |
5249 | if (ret < 0) |
5250 | return ret; | |
bd56b302 | 5251 | |
2c47e605 | 5252 | if (path->locks[level]) { |
bd681513 CM |
5253 | btrfs_tree_unlock_rw(path->nodes[level], |
5254 | path->locks[level]); | |
2c47e605 | 5255 | path->locks[level] = 0; |
f82d02d9 | 5256 | } |
2c47e605 YZ |
5257 | free_extent_buffer(path->nodes[level]); |
5258 | path->nodes[level] = NULL; | |
5259 | level++; | |
20524f02 CM |
5260 | } |
5261 | } | |
5262 | return 1; | |
5263 | } | |
5264 | ||
9aca1d51 | 5265 | /* |
2c47e605 YZ |
5266 | * drop a subvolume tree. |
5267 | * | |
5268 | * this function traverses the tree freeing any blocks that only | |
5269 | * referenced by the tree. | |
5270 | * | |
5271 | * when a shared tree block is found. this function decreases its | |
5272 | * reference count by one. if update_ref is true, this function | |
5273 | * also make sure backrefs for the shared block and all lower level | |
5274 | * blocks are properly updated. | |
9d1a2a3a DS |
5275 | * |
5276 | * If called with for_reloc == 0, may exit early with -EAGAIN | |
9aca1d51 | 5277 | */ |
0078a9f9 | 5278 | int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref, int for_reloc) |
20524f02 | 5279 | { |
ab8d0fc4 | 5280 | struct btrfs_fs_info *fs_info = root->fs_info; |
5caf2a00 | 5281 | struct btrfs_path *path; |
2c47e605 | 5282 | struct btrfs_trans_handle *trans; |
ab8d0fc4 | 5283 | struct btrfs_root *tree_root = fs_info->tree_root; |
9f3a7427 | 5284 | struct btrfs_root_item *root_item = &root->root_item; |
2c47e605 YZ |
5285 | struct walk_control *wc; |
5286 | struct btrfs_key key; | |
5287 | int err = 0; | |
5288 | int ret; | |
5289 | int level; | |
d29a9f62 | 5290 | bool root_dropped = false; |
20524f02 | 5291 | |
4fd786e6 | 5292 | btrfs_debug(fs_info, "Drop subvolume %llu", root->root_key.objectid); |
1152651a | 5293 | |
5caf2a00 | 5294 | path = btrfs_alloc_path(); |
cb1b69f4 TI |
5295 | if (!path) { |
5296 | err = -ENOMEM; | |
5297 | goto out; | |
5298 | } | |
20524f02 | 5299 | |
2c47e605 | 5300 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
38a1a919 MF |
5301 | if (!wc) { |
5302 | btrfs_free_path(path); | |
cb1b69f4 TI |
5303 | err = -ENOMEM; |
5304 | goto out; | |
38a1a919 | 5305 | } |
2c47e605 | 5306 | |
f3e3d9cc QW |
5307 | /* |
5308 | * Use join to avoid potential EINTR from transaction start. See | |
5309 | * wait_reserve_ticket and the whole reservation callchain. | |
5310 | */ | |
5311 | if (for_reloc) | |
5312 | trans = btrfs_join_transaction(tree_root); | |
5313 | else | |
5314 | trans = btrfs_start_transaction(tree_root, 0); | |
79787eaa JM |
5315 | if (IS_ERR(trans)) { |
5316 | err = PTR_ERR(trans); | |
5317 | goto out_free; | |
5318 | } | |
98d5dc13 | 5319 | |
0568e82d JB |
5320 | err = btrfs_run_delayed_items(trans); |
5321 | if (err) | |
5322 | goto out_end_trans; | |
5323 | ||
83354f07 JB |
5324 | /* |
5325 | * This will help us catch people modifying the fs tree while we're | |
5326 | * dropping it. It is unsafe to mess with the fs tree while it's being | |
5327 | * dropped as we unlock the root node and parent nodes as we walk down | |
5328 | * the tree, assuming nothing will change. If something does change | |
5329 | * then we'll have stale information and drop references to blocks we've | |
5330 | * already dropped. | |
5331 | */ | |
5332 | set_bit(BTRFS_ROOT_DELETING, &root->state); | |
9f3a7427 | 5333 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
2c47e605 | 5334 | level = btrfs_header_level(root->node); |
5d4f98a2 | 5335 | path->nodes[level] = btrfs_lock_root_node(root); |
8bead258 | 5336 | btrfs_set_lock_blocking_write(path->nodes[level]); |
9f3a7427 | 5337 | path->slots[level] = 0; |
bd681513 | 5338 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
5339 | memset(&wc->update_progress, 0, |
5340 | sizeof(wc->update_progress)); | |
9f3a7427 | 5341 | } else { |
9f3a7427 | 5342 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
2c47e605 YZ |
5343 | memcpy(&wc->update_progress, &key, |
5344 | sizeof(wc->update_progress)); | |
5345 | ||
6702ed49 | 5346 | level = root_item->drop_level; |
2c47e605 | 5347 | BUG_ON(level == 0); |
6702ed49 | 5348 | path->lowest_level = level; |
2c47e605 YZ |
5349 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
5350 | path->lowest_level = 0; | |
5351 | if (ret < 0) { | |
5352 | err = ret; | |
79787eaa | 5353 | goto out_end_trans; |
9f3a7427 | 5354 | } |
1c4850e2 | 5355 | WARN_ON(ret > 0); |
2c47e605 | 5356 | |
7d9eb12c CM |
5357 | /* |
5358 | * unlock our path, this is safe because only this | |
5359 | * function is allowed to delete this snapshot | |
5360 | */ | |
5d4f98a2 | 5361 | btrfs_unlock_up_safe(path, 0); |
2c47e605 YZ |
5362 | |
5363 | level = btrfs_header_level(root->node); | |
5364 | while (1) { | |
5365 | btrfs_tree_lock(path->nodes[level]); | |
8bead258 | 5366 | btrfs_set_lock_blocking_write(path->nodes[level]); |
fec386ac | 5367 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 5368 | |
2ff7e61e | 5369 | ret = btrfs_lookup_extent_info(trans, fs_info, |
2c47e605 | 5370 | path->nodes[level]->start, |
3173a18f | 5371 | level, 1, &wc->refs[level], |
2c47e605 | 5372 | &wc->flags[level]); |
79787eaa JM |
5373 | if (ret < 0) { |
5374 | err = ret; | |
5375 | goto out_end_trans; | |
5376 | } | |
2c47e605 YZ |
5377 | BUG_ON(wc->refs[level] == 0); |
5378 | ||
5379 | if (level == root_item->drop_level) | |
5380 | break; | |
5381 | ||
5382 | btrfs_tree_unlock(path->nodes[level]); | |
fec386ac | 5383 | path->locks[level] = 0; |
2c47e605 YZ |
5384 | WARN_ON(wc->refs[level] != 1); |
5385 | level--; | |
5386 | } | |
9f3a7427 | 5387 | } |
2c47e605 | 5388 | |
78c52d9e | 5389 | wc->restarted = test_bit(BTRFS_ROOT_DEAD_TREE, &root->state); |
2c47e605 YZ |
5390 | wc->level = level; |
5391 | wc->shared_level = -1; | |
5392 | wc->stage = DROP_REFERENCE; | |
5393 | wc->update_ref = update_ref; | |
5394 | wc->keep_locks = 0; | |
0b246afa | 5395 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info); |
2c47e605 | 5396 | |
d397712b | 5397 | while (1) { |
9d1a2a3a | 5398 | |
2c47e605 YZ |
5399 | ret = walk_down_tree(trans, root, path, wc); |
5400 | if (ret < 0) { | |
5401 | err = ret; | |
20524f02 | 5402 | break; |
2c47e605 | 5403 | } |
9aca1d51 | 5404 | |
2c47e605 YZ |
5405 | ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL); |
5406 | if (ret < 0) { | |
5407 | err = ret; | |
20524f02 | 5408 | break; |
2c47e605 YZ |
5409 | } |
5410 | ||
5411 | if (ret > 0) { | |
5412 | BUG_ON(wc->stage != DROP_REFERENCE); | |
e7a84565 CM |
5413 | break; |
5414 | } | |
2c47e605 YZ |
5415 | |
5416 | if (wc->stage == DROP_REFERENCE) { | |
aea6f028 JB |
5417 | wc->drop_level = wc->level; |
5418 | btrfs_node_key_to_cpu(path->nodes[wc->drop_level], | |
5419 | &wc->drop_progress, | |
5420 | path->slots[wc->drop_level]); | |
5421 | } | |
5422 | btrfs_cpu_key_to_disk(&root_item->drop_progress, | |
5423 | &wc->drop_progress); | |
5424 | root_item->drop_level = wc->drop_level; | |
2c47e605 YZ |
5425 | |
5426 | BUG_ON(wc->level == 0); | |
3a45bb20 | 5427 | if (btrfs_should_end_transaction(trans) || |
2ff7e61e | 5428 | (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) { |
2c47e605 YZ |
5429 | ret = btrfs_update_root(trans, tree_root, |
5430 | &root->root_key, | |
5431 | root_item); | |
79787eaa | 5432 | if (ret) { |
66642832 | 5433 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
5434 | err = ret; |
5435 | goto out_end_trans; | |
5436 | } | |
2c47e605 | 5437 | |
3a45bb20 | 5438 | btrfs_end_transaction_throttle(trans); |
2ff7e61e | 5439 | if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) { |
ab8d0fc4 JM |
5440 | btrfs_debug(fs_info, |
5441 | "drop snapshot early exit"); | |
3c8f2422 JB |
5442 | err = -EAGAIN; |
5443 | goto out_free; | |
5444 | } | |
5445 | ||
a22285a6 | 5446 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
5447 | if (IS_ERR(trans)) { |
5448 | err = PTR_ERR(trans); | |
5449 | goto out_free; | |
5450 | } | |
c3e69d58 | 5451 | } |
20524f02 | 5452 | } |
b3b4aa74 | 5453 | btrfs_release_path(path); |
79787eaa JM |
5454 | if (err) |
5455 | goto out_end_trans; | |
2c47e605 | 5456 | |
ab9ce7d4 | 5457 | ret = btrfs_del_root(trans, &root->root_key); |
79787eaa | 5458 | if (ret) { |
66642832 | 5459 | btrfs_abort_transaction(trans, ret); |
e19182c0 | 5460 | err = ret; |
79787eaa JM |
5461 | goto out_end_trans; |
5462 | } | |
2c47e605 | 5463 | |
76dda93c | 5464 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
cb517eab MX |
5465 | ret = btrfs_find_root(tree_root, &root->root_key, path, |
5466 | NULL, NULL); | |
79787eaa | 5467 | if (ret < 0) { |
66642832 | 5468 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
5469 | err = ret; |
5470 | goto out_end_trans; | |
5471 | } else if (ret > 0) { | |
84cd948c JB |
5472 | /* if we fail to delete the orphan item this time |
5473 | * around, it'll get picked up the next time. | |
5474 | * | |
5475 | * The most common failure here is just -ENOENT. | |
5476 | */ | |
5477 | btrfs_del_orphan_item(trans, tree_root, | |
5478 | root->root_key.objectid); | |
76dda93c YZ |
5479 | } |
5480 | } | |
5481 | ||
a3cf0e43 QW |
5482 | /* |
5483 | * This subvolume is going to be completely dropped, and won't be | |
5484 | * recorded as dirty roots, thus pertrans meta rsv will not be freed at | |
5485 | * commit transaction time. So free it here manually. | |
5486 | */ | |
5487 | btrfs_qgroup_convert_reserved_meta(root, INT_MAX); | |
5488 | btrfs_qgroup_free_meta_all_pertrans(root); | |
5489 | ||
8c38938c | 5490 | if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) |
2b9dbef2 | 5491 | btrfs_add_dropped_root(trans, root); |
8c38938c | 5492 | else |
00246528 | 5493 | btrfs_put_root(root); |
d29a9f62 | 5494 | root_dropped = true; |
79787eaa | 5495 | out_end_trans: |
3a45bb20 | 5496 | btrfs_end_transaction_throttle(trans); |
79787eaa | 5497 | out_free: |
2c47e605 | 5498 | kfree(wc); |
5caf2a00 | 5499 | btrfs_free_path(path); |
cb1b69f4 | 5500 | out: |
d29a9f62 JB |
5501 | /* |
5502 | * So if we need to stop dropping the snapshot for whatever reason we | |
5503 | * need to make sure to add it back to the dead root list so that we | |
5504 | * keep trying to do the work later. This also cleans up roots if we | |
5505 | * don't have it in the radix (like when we recover after a power fail | |
5506 | * or unmount) so we don't leak memory. | |
5507 | */ | |
897ca819 | 5508 | if (!for_reloc && !root_dropped) |
d29a9f62 | 5509 | btrfs_add_dead_root(root); |
2c536799 | 5510 | return err; |
20524f02 | 5511 | } |
9078a3e1 | 5512 | |
2c47e605 YZ |
5513 | /* |
5514 | * drop subtree rooted at tree block 'node'. | |
5515 | * | |
5516 | * NOTE: this function will unlock and release tree block 'node' | |
66d7e7f0 | 5517 | * only used by relocation code |
2c47e605 | 5518 | */ |
f82d02d9 YZ |
5519 | int btrfs_drop_subtree(struct btrfs_trans_handle *trans, |
5520 | struct btrfs_root *root, | |
5521 | struct extent_buffer *node, | |
5522 | struct extent_buffer *parent) | |
5523 | { | |
0b246afa | 5524 | struct btrfs_fs_info *fs_info = root->fs_info; |
f82d02d9 | 5525 | struct btrfs_path *path; |
2c47e605 | 5526 | struct walk_control *wc; |
f82d02d9 YZ |
5527 | int level; |
5528 | int parent_level; | |
5529 | int ret = 0; | |
5530 | int wret; | |
5531 | ||
2c47e605 YZ |
5532 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); |
5533 | ||
f82d02d9 | 5534 | path = btrfs_alloc_path(); |
db5b493a TI |
5535 | if (!path) |
5536 | return -ENOMEM; | |
f82d02d9 | 5537 | |
2c47e605 | 5538 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
db5b493a TI |
5539 | if (!wc) { |
5540 | btrfs_free_path(path); | |
5541 | return -ENOMEM; | |
5542 | } | |
2c47e605 | 5543 | |
b9447ef8 | 5544 | btrfs_assert_tree_locked(parent); |
f82d02d9 | 5545 | parent_level = btrfs_header_level(parent); |
67439dad | 5546 | atomic_inc(&parent->refs); |
f82d02d9 YZ |
5547 | path->nodes[parent_level] = parent; |
5548 | path->slots[parent_level] = btrfs_header_nritems(parent); | |
5549 | ||
b9447ef8 | 5550 | btrfs_assert_tree_locked(node); |
f82d02d9 | 5551 | level = btrfs_header_level(node); |
f82d02d9 YZ |
5552 | path->nodes[level] = node; |
5553 | path->slots[level] = 0; | |
bd681513 | 5554 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
5555 | |
5556 | wc->refs[parent_level] = 1; | |
5557 | wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
5558 | wc->level = level; | |
5559 | wc->shared_level = -1; | |
5560 | wc->stage = DROP_REFERENCE; | |
5561 | wc->update_ref = 0; | |
5562 | wc->keep_locks = 1; | |
0b246afa | 5563 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info); |
f82d02d9 YZ |
5564 | |
5565 | while (1) { | |
2c47e605 YZ |
5566 | wret = walk_down_tree(trans, root, path, wc); |
5567 | if (wret < 0) { | |
f82d02d9 | 5568 | ret = wret; |
f82d02d9 | 5569 | break; |
2c47e605 | 5570 | } |
f82d02d9 | 5571 | |
2c47e605 | 5572 | wret = walk_up_tree(trans, root, path, wc, parent_level); |
f82d02d9 YZ |
5573 | if (wret < 0) |
5574 | ret = wret; | |
5575 | if (wret != 0) | |
5576 | break; | |
5577 | } | |
5578 | ||
2c47e605 | 5579 | kfree(wc); |
f82d02d9 YZ |
5580 | btrfs_free_path(path); |
5581 | return ret; | |
5582 | } | |
5583 | ||
6d07bcec MX |
5584 | /* |
5585 | * helper to account the unused space of all the readonly block group in the | |
633c0aad | 5586 | * space_info. takes mirrors into account. |
6d07bcec | 5587 | */ |
633c0aad | 5588 | u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo) |
6d07bcec | 5589 | { |
32da5386 | 5590 | struct btrfs_block_group *block_group; |
6d07bcec MX |
5591 | u64 free_bytes = 0; |
5592 | int factor; | |
5593 | ||
01327610 | 5594 | /* It's df, we don't care if it's racy */ |
633c0aad JB |
5595 | if (list_empty(&sinfo->ro_bgs)) |
5596 | return 0; | |
5597 | ||
5598 | spin_lock(&sinfo->lock); | |
5599 | list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) { | |
6d07bcec MX |
5600 | spin_lock(&block_group->lock); |
5601 | ||
5602 | if (!block_group->ro) { | |
5603 | spin_unlock(&block_group->lock); | |
5604 | continue; | |
5605 | } | |
5606 | ||
46df06b8 | 5607 | factor = btrfs_bg_type_to_factor(block_group->flags); |
b3470b5d | 5608 | free_bytes += (block_group->length - |
bf38be65 | 5609 | block_group->used) * factor; |
6d07bcec MX |
5610 | |
5611 | spin_unlock(&block_group->lock); | |
5612 | } | |
6d07bcec MX |
5613 | spin_unlock(&sinfo->lock); |
5614 | ||
5615 | return free_bytes; | |
5616 | } | |
5617 | ||
2ff7e61e JM |
5618 | int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info, |
5619 | u64 start, u64 end) | |
acce952b | 5620 | { |
2ff7e61e | 5621 | return unpin_extent_range(fs_info, start, end, false); |
acce952b | 5622 | } |
5623 | ||
499f377f JM |
5624 | /* |
5625 | * It used to be that old block groups would be left around forever. | |
5626 | * Iterating over them would be enough to trim unused space. Since we | |
5627 | * now automatically remove them, we also need to iterate over unallocated | |
5628 | * space. | |
5629 | * | |
5630 | * We don't want a transaction for this since the discard may take a | |
5631 | * substantial amount of time. We don't require that a transaction be | |
5632 | * running, but we do need to take a running transaction into account | |
fee7acc3 JM |
5633 | * to ensure that we're not discarding chunks that were released or |
5634 | * allocated in the current transaction. | |
499f377f JM |
5635 | * |
5636 | * Holding the chunks lock will prevent other threads from allocating | |
5637 | * or releasing chunks, but it won't prevent a running transaction | |
5638 | * from committing and releasing the memory that the pending chunks | |
5639 | * list head uses. For that, we need to take a reference to the | |
fee7acc3 JM |
5640 | * transaction and hold the commit root sem. We only need to hold |
5641 | * it while performing the free space search since we have already | |
5642 | * held back allocations. | |
499f377f | 5643 | */ |
8103d10b | 5644 | static int btrfs_trim_free_extents(struct btrfs_device *device, u64 *trimmed) |
499f377f | 5645 | { |
8103d10b | 5646 | u64 start = SZ_1M, len = 0, end = 0; |
499f377f JM |
5647 | int ret; |
5648 | ||
5649 | *trimmed = 0; | |
5650 | ||
0be88e36 JM |
5651 | /* Discard not supported = nothing to do. */ |
5652 | if (!blk_queue_discard(bdev_get_queue(device->bdev))) | |
5653 | return 0; | |
5654 | ||
52042d8e | 5655 | /* Not writable = nothing to do. */ |
ebbede42 | 5656 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
499f377f JM |
5657 | return 0; |
5658 | ||
5659 | /* No free space = nothing to do. */ | |
5660 | if (device->total_bytes <= device->bytes_used) | |
5661 | return 0; | |
5662 | ||
5663 | ret = 0; | |
5664 | ||
5665 | while (1) { | |
fb456252 | 5666 | struct btrfs_fs_info *fs_info = device->fs_info; |
499f377f JM |
5667 | u64 bytes; |
5668 | ||
5669 | ret = mutex_lock_interruptible(&fs_info->chunk_mutex); | |
5670 | if (ret) | |
fee7acc3 | 5671 | break; |
499f377f | 5672 | |
929be17a NB |
5673 | find_first_clear_extent_bit(&device->alloc_state, start, |
5674 | &start, &end, | |
5675 | CHUNK_TRIMMED | CHUNK_ALLOCATED); | |
53460a45 | 5676 | |
c57dd1f2 QW |
5677 | /* Check if there are any CHUNK_* bits left */ |
5678 | if (start > device->total_bytes) { | |
5679 | WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); | |
5680 | btrfs_warn_in_rcu(fs_info, | |
5681 | "ignoring attempt to trim beyond device size: offset %llu length %llu device %s device size %llu", | |
5682 | start, end - start + 1, | |
5683 | rcu_str_deref(device->name), | |
5684 | device->total_bytes); | |
5685 | mutex_unlock(&fs_info->chunk_mutex); | |
5686 | ret = 0; | |
5687 | break; | |
5688 | } | |
5689 | ||
53460a45 NB |
5690 | /* Ensure we skip the reserved area in the first 1M */ |
5691 | start = max_t(u64, start, SZ_1M); | |
5692 | ||
929be17a NB |
5693 | /* |
5694 | * If find_first_clear_extent_bit find a range that spans the | |
5695 | * end of the device it will set end to -1, in this case it's up | |
5696 | * to the caller to trim the value to the size of the device. | |
5697 | */ | |
5698 | end = min(end, device->total_bytes - 1); | |
53460a45 | 5699 | |
929be17a | 5700 | len = end - start + 1; |
499f377f | 5701 | |
929be17a NB |
5702 | /* We didn't find any extents */ |
5703 | if (!len) { | |
499f377f | 5704 | mutex_unlock(&fs_info->chunk_mutex); |
929be17a | 5705 | ret = 0; |
499f377f JM |
5706 | break; |
5707 | } | |
5708 | ||
929be17a NB |
5709 | ret = btrfs_issue_discard(device->bdev, start, len, |
5710 | &bytes); | |
5711 | if (!ret) | |
5712 | set_extent_bits(&device->alloc_state, start, | |
5713 | start + bytes - 1, | |
5714 | CHUNK_TRIMMED); | |
499f377f JM |
5715 | mutex_unlock(&fs_info->chunk_mutex); |
5716 | ||
5717 | if (ret) | |
5718 | break; | |
5719 | ||
5720 | start += len; | |
5721 | *trimmed += bytes; | |
5722 | ||
5723 | if (fatal_signal_pending(current)) { | |
5724 | ret = -ERESTARTSYS; | |
5725 | break; | |
5726 | } | |
5727 | ||
5728 | cond_resched(); | |
5729 | } | |
5730 | ||
5731 | return ret; | |
5732 | } | |
5733 | ||
93bba24d QW |
5734 | /* |
5735 | * Trim the whole filesystem by: | |
5736 | * 1) trimming the free space in each block group | |
5737 | * 2) trimming the unallocated space on each device | |
5738 | * | |
5739 | * This will also continue trimming even if a block group or device encounters | |
5740 | * an error. The return value will be the last error, or 0 if nothing bad | |
5741 | * happens. | |
5742 | */ | |
2ff7e61e | 5743 | int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range) |
f7039b1d | 5744 | { |
32da5386 | 5745 | struct btrfs_block_group *cache = NULL; |
499f377f JM |
5746 | struct btrfs_device *device; |
5747 | struct list_head *devices; | |
f7039b1d | 5748 | u64 group_trimmed; |
07301df7 | 5749 | u64 range_end = U64_MAX; |
f7039b1d LD |
5750 | u64 start; |
5751 | u64 end; | |
5752 | u64 trimmed = 0; | |
93bba24d QW |
5753 | u64 bg_failed = 0; |
5754 | u64 dev_failed = 0; | |
5755 | int bg_ret = 0; | |
5756 | int dev_ret = 0; | |
f7039b1d LD |
5757 | int ret = 0; |
5758 | ||
07301df7 QW |
5759 | /* |
5760 | * Check range overflow if range->len is set. | |
5761 | * The default range->len is U64_MAX. | |
5762 | */ | |
5763 | if (range->len != U64_MAX && | |
5764 | check_add_overflow(range->start, range->len, &range_end)) | |
5765 | return -EINVAL; | |
5766 | ||
6ba9fc8e | 5767 | cache = btrfs_lookup_first_block_group(fs_info, range->start); |
2e405ad8 | 5768 | for (; cache; cache = btrfs_next_block_group(cache)) { |
b3470b5d | 5769 | if (cache->start >= range_end) { |
f7039b1d LD |
5770 | btrfs_put_block_group(cache); |
5771 | break; | |
5772 | } | |
5773 | ||
b3470b5d DS |
5774 | start = max(range->start, cache->start); |
5775 | end = min(range_end, cache->start + cache->length); | |
f7039b1d LD |
5776 | |
5777 | if (end - start >= range->minlen) { | |
32da5386 | 5778 | if (!btrfs_block_group_done(cache)) { |
676f1f75 | 5779 | ret = btrfs_cache_block_group(cache, 0); |
1be41b78 | 5780 | if (ret) { |
93bba24d QW |
5781 | bg_failed++; |
5782 | bg_ret = ret; | |
5783 | continue; | |
1be41b78 | 5784 | } |
676f1f75 | 5785 | ret = btrfs_wait_block_group_cache_done(cache); |
1be41b78 | 5786 | if (ret) { |
93bba24d QW |
5787 | bg_failed++; |
5788 | bg_ret = ret; | |
5789 | continue; | |
1be41b78 | 5790 | } |
f7039b1d LD |
5791 | } |
5792 | ret = btrfs_trim_block_group(cache, | |
5793 | &group_trimmed, | |
5794 | start, | |
5795 | end, | |
5796 | range->minlen); | |
5797 | ||
5798 | trimmed += group_trimmed; | |
5799 | if (ret) { | |
93bba24d QW |
5800 | bg_failed++; |
5801 | bg_ret = ret; | |
5802 | continue; | |
f7039b1d LD |
5803 | } |
5804 | } | |
f7039b1d LD |
5805 | } |
5806 | ||
93bba24d QW |
5807 | if (bg_failed) |
5808 | btrfs_warn(fs_info, | |
5809 | "failed to trim %llu block group(s), last error %d", | |
5810 | bg_failed, bg_ret); | |
0b246afa | 5811 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
d4e329de JM |
5812 | devices = &fs_info->fs_devices->devices; |
5813 | list_for_each_entry(device, devices, dev_list) { | |
8103d10b | 5814 | ret = btrfs_trim_free_extents(device, &group_trimmed); |
93bba24d QW |
5815 | if (ret) { |
5816 | dev_failed++; | |
5817 | dev_ret = ret; | |
499f377f | 5818 | break; |
93bba24d | 5819 | } |
499f377f JM |
5820 | |
5821 | trimmed += group_trimmed; | |
5822 | } | |
0b246afa | 5823 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
499f377f | 5824 | |
93bba24d QW |
5825 | if (dev_failed) |
5826 | btrfs_warn(fs_info, | |
5827 | "failed to trim %llu device(s), last error %d", | |
5828 | dev_failed, dev_ret); | |
f7039b1d | 5829 | range->len = trimmed; |
93bba24d QW |
5830 | if (bg_ret) |
5831 | return bg_ret; | |
5832 | return dev_ret; | |
f7039b1d | 5833 | } |