Commit | Line | Data |
---|---|---|
6cbd5570 CM |
1 | /* |
2 | * Copyright (C) 2007 Oracle. All rights reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
18 | ||
79154b1b | 19 | #include <linux/fs.h> |
5a0e3ad6 | 20 | #include <linux/slab.h> |
34088780 | 21 | #include <linux/sched.h> |
d3c2fdcf | 22 | #include <linux/writeback.h> |
5f39d397 | 23 | #include <linux/pagemap.h> |
5f2cc086 | 24 | #include <linux/blkdev.h> |
8ea05e3a | 25 | #include <linux/uuid.h> |
79154b1b CM |
26 | #include "ctree.h" |
27 | #include "disk-io.h" | |
28 | #include "transaction.h" | |
925baedd | 29 | #include "locking.h" |
e02119d5 | 30 | #include "tree-log.h" |
581bb050 | 31 | #include "inode-map.h" |
733f4fbb | 32 | #include "volumes.h" |
8dabb742 | 33 | #include "dev-replace.h" |
79154b1b | 34 | |
0f7d52f4 CM |
35 | #define BTRFS_ROOT_TRANS_TAG 0 |
36 | ||
49b25e05 | 37 | void put_transaction(struct btrfs_transaction *transaction) |
79154b1b | 38 | { |
13c5a93e JB |
39 | WARN_ON(atomic_read(&transaction->use_count) == 0); |
40 | if (atomic_dec_and_test(&transaction->use_count)) { | |
a4abeea4 | 41 | BUG_ON(!list_empty(&transaction->list)); |
00f04b88 | 42 | WARN_ON(transaction->delayed_refs.root.rb_node); |
2c90e5d6 | 43 | kmem_cache_free(btrfs_transaction_cachep, transaction); |
78fae27e | 44 | } |
79154b1b CM |
45 | } |
46 | ||
817d52f8 JB |
47 | static noinline void switch_commit_root(struct btrfs_root *root) |
48 | { | |
817d52f8 JB |
49 | free_extent_buffer(root->commit_root); |
50 | root->commit_root = btrfs_root_node(root); | |
817d52f8 JB |
51 | } |
52 | ||
178260b2 MX |
53 | static inline int can_join_transaction(struct btrfs_transaction *trans, |
54 | int type) | |
55 | { | |
56 | return !(trans->in_commit && | |
57 | type != TRANS_JOIN && | |
58 | type != TRANS_JOIN_NOLOCK); | |
59 | } | |
60 | ||
d352ac68 CM |
61 | /* |
62 | * either allocate a new transaction or hop into the existing one | |
63 | */ | |
354aa0fb | 64 | static noinline int join_transaction(struct btrfs_root *root, int type) |
79154b1b CM |
65 | { |
66 | struct btrfs_transaction *cur_trans; | |
19ae4e81 | 67 | struct btrfs_fs_info *fs_info = root->fs_info; |
a4abeea4 | 68 | |
19ae4e81 | 69 | spin_lock(&fs_info->trans_lock); |
d43317dc | 70 | loop: |
49b25e05 | 71 | /* The file system has been taken offline. No new transactions. */ |
87533c47 | 72 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { |
19ae4e81 | 73 | spin_unlock(&fs_info->trans_lock); |
49b25e05 JM |
74 | return -EROFS; |
75 | } | |
76 | ||
19ae4e81 | 77 | if (fs_info->trans_no_join) { |
354aa0fb MX |
78 | /* |
79 | * If we are JOIN_NOLOCK we're already committing a current | |
80 | * transaction, we just need a handle to deal with something | |
81 | * when committing the transaction, such as inode cache and | |
82 | * space cache. It is a special case. | |
83 | */ | |
84 | if (type != TRANS_JOIN_NOLOCK) { | |
19ae4e81 | 85 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 JB |
86 | return -EBUSY; |
87 | } | |
88 | } | |
89 | ||
19ae4e81 | 90 | cur_trans = fs_info->running_transaction; |
a4abeea4 | 91 | if (cur_trans) { |
871383be | 92 | if (cur_trans->aborted) { |
19ae4e81 | 93 | spin_unlock(&fs_info->trans_lock); |
49b25e05 | 94 | return cur_trans->aborted; |
871383be | 95 | } |
178260b2 MX |
96 | if (!can_join_transaction(cur_trans, type)) { |
97 | spin_unlock(&fs_info->trans_lock); | |
98 | return -EBUSY; | |
99 | } | |
a4abeea4 | 100 | atomic_inc(&cur_trans->use_count); |
13c5a93e | 101 | atomic_inc(&cur_trans->num_writers); |
15ee9bc7 | 102 | cur_trans->num_joined++; |
19ae4e81 | 103 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 104 | return 0; |
79154b1b | 105 | } |
19ae4e81 | 106 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 107 | |
354aa0fb MX |
108 | /* |
109 | * If we are ATTACH, we just want to catch the current transaction, | |
110 | * and commit it. If there is no transaction, just return ENOENT. | |
111 | */ | |
112 | if (type == TRANS_ATTACH) | |
113 | return -ENOENT; | |
114 | ||
a4abeea4 JB |
115 | cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, GFP_NOFS); |
116 | if (!cur_trans) | |
117 | return -ENOMEM; | |
d43317dc | 118 | |
19ae4e81 JS |
119 | spin_lock(&fs_info->trans_lock); |
120 | if (fs_info->running_transaction) { | |
d43317dc CM |
121 | /* |
122 | * someone started a transaction after we unlocked. Make sure | |
123 | * to redo the trans_no_join checks above | |
124 | */ | |
a4abeea4 | 125 | kmem_cache_free(btrfs_transaction_cachep, cur_trans); |
19ae4e81 | 126 | cur_trans = fs_info->running_transaction; |
d43317dc | 127 | goto loop; |
87533c47 | 128 | } else if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { |
e4b50e14 | 129 | spin_unlock(&fs_info->trans_lock); |
7b8b92af JB |
130 | kmem_cache_free(btrfs_transaction_cachep, cur_trans); |
131 | return -EROFS; | |
79154b1b | 132 | } |
d43317dc | 133 | |
a4abeea4 JB |
134 | atomic_set(&cur_trans->num_writers, 1); |
135 | cur_trans->num_joined = 0; | |
136 | init_waitqueue_head(&cur_trans->writer_wait); | |
137 | init_waitqueue_head(&cur_trans->commit_wait); | |
138 | cur_trans->in_commit = 0; | |
139 | cur_trans->blocked = 0; | |
140 | /* | |
141 | * One for this trans handle, one so it will live on until we | |
142 | * commit the transaction. | |
143 | */ | |
144 | atomic_set(&cur_trans->use_count, 2); | |
145 | cur_trans->commit_done = 0; | |
146 | cur_trans->start_time = get_seconds(); | |
147 | ||
148 | cur_trans->delayed_refs.root = RB_ROOT; | |
149 | cur_trans->delayed_refs.num_entries = 0; | |
150 | cur_trans->delayed_refs.num_heads_ready = 0; | |
151 | cur_trans->delayed_refs.num_heads = 0; | |
152 | cur_trans->delayed_refs.flushing = 0; | |
153 | cur_trans->delayed_refs.run_delayed_start = 0; | |
20b297d6 JS |
154 | |
155 | /* | |
156 | * although the tree mod log is per file system and not per transaction, | |
157 | * the log must never go across transaction boundaries. | |
158 | */ | |
159 | smp_mb(); | |
31b1a2bd JL |
160 | if (!list_empty(&fs_info->tree_mod_seq_list)) |
161 | WARN(1, KERN_ERR "btrfs: tree_mod_seq_list not empty when " | |
20b297d6 | 162 | "creating a fresh transaction\n"); |
31b1a2bd JL |
163 | if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log)) |
164 | WARN(1, KERN_ERR "btrfs: tree_mod_log rb tree not empty when " | |
20b297d6 | 165 | "creating a fresh transaction\n"); |
20b297d6 JS |
166 | atomic_set(&fs_info->tree_mod_seq, 0); |
167 | ||
a4abeea4 JB |
168 | spin_lock_init(&cur_trans->commit_lock); |
169 | spin_lock_init(&cur_trans->delayed_refs.lock); | |
bb721703 CM |
170 | atomic_set(&cur_trans->delayed_refs.procs_running_refs, 0); |
171 | atomic_set(&cur_trans->delayed_refs.ref_seq, 0); | |
172 | init_waitqueue_head(&cur_trans->delayed_refs.wait); | |
a4abeea4 JB |
173 | |
174 | INIT_LIST_HEAD(&cur_trans->pending_snapshots); | |
569e0f35 | 175 | INIT_LIST_HEAD(&cur_trans->ordered_operations); |
19ae4e81 | 176 | list_add_tail(&cur_trans->list, &fs_info->trans_list); |
a4abeea4 | 177 | extent_io_tree_init(&cur_trans->dirty_pages, |
19ae4e81 JS |
178 | fs_info->btree_inode->i_mapping); |
179 | fs_info->generation++; | |
180 | cur_trans->transid = fs_info->generation; | |
181 | fs_info->running_transaction = cur_trans; | |
49b25e05 | 182 | cur_trans->aborted = 0; |
19ae4e81 | 183 | spin_unlock(&fs_info->trans_lock); |
15ee9bc7 | 184 | |
79154b1b CM |
185 | return 0; |
186 | } | |
187 | ||
d352ac68 | 188 | /* |
d397712b CM |
189 | * this does all the record keeping required to make sure that a reference |
190 | * counted root is properly recorded in a given transaction. This is required | |
191 | * to make sure the old root from before we joined the transaction is deleted | |
192 | * when the transaction commits | |
d352ac68 | 193 | */ |
7585717f | 194 | static int record_root_in_trans(struct btrfs_trans_handle *trans, |
a4abeea4 | 195 | struct btrfs_root *root) |
6702ed49 | 196 | { |
5d4f98a2 | 197 | if (root->ref_cows && root->last_trans < trans->transid) { |
6702ed49 | 198 | WARN_ON(root == root->fs_info->extent_root); |
5d4f98a2 YZ |
199 | WARN_ON(root->commit_root != root->node); |
200 | ||
7585717f CM |
201 | /* |
202 | * see below for in_trans_setup usage rules | |
203 | * we have the reloc mutex held now, so there | |
204 | * is only one writer in this function | |
205 | */ | |
206 | root->in_trans_setup = 1; | |
207 | ||
208 | /* make sure readers find in_trans_setup before | |
209 | * they find our root->last_trans update | |
210 | */ | |
211 | smp_wmb(); | |
212 | ||
a4abeea4 JB |
213 | spin_lock(&root->fs_info->fs_roots_radix_lock); |
214 | if (root->last_trans == trans->transid) { | |
215 | spin_unlock(&root->fs_info->fs_roots_radix_lock); | |
216 | return 0; | |
217 | } | |
5d4f98a2 YZ |
218 | radix_tree_tag_set(&root->fs_info->fs_roots_radix, |
219 | (unsigned long)root->root_key.objectid, | |
220 | BTRFS_ROOT_TRANS_TAG); | |
a4abeea4 | 221 | spin_unlock(&root->fs_info->fs_roots_radix_lock); |
7585717f CM |
222 | root->last_trans = trans->transid; |
223 | ||
224 | /* this is pretty tricky. We don't want to | |
225 | * take the relocation lock in btrfs_record_root_in_trans | |
226 | * unless we're really doing the first setup for this root in | |
227 | * this transaction. | |
228 | * | |
229 | * Normally we'd use root->last_trans as a flag to decide | |
230 | * if we want to take the expensive mutex. | |
231 | * | |
232 | * But, we have to set root->last_trans before we | |
233 | * init the relocation root, otherwise, we trip over warnings | |
234 | * in ctree.c. The solution used here is to flag ourselves | |
235 | * with root->in_trans_setup. When this is 1, we're still | |
236 | * fixing up the reloc trees and everyone must wait. | |
237 | * | |
238 | * When this is zero, they can trust root->last_trans and fly | |
239 | * through btrfs_record_root_in_trans without having to take the | |
240 | * lock. smp_wmb() makes sure that all the writes above are | |
241 | * done before we pop in the zero below | |
242 | */ | |
5d4f98a2 | 243 | btrfs_init_reloc_root(trans, root); |
7585717f CM |
244 | smp_wmb(); |
245 | root->in_trans_setup = 0; | |
5d4f98a2 YZ |
246 | } |
247 | return 0; | |
248 | } | |
bcc63abb | 249 | |
7585717f CM |
250 | |
251 | int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, | |
252 | struct btrfs_root *root) | |
253 | { | |
254 | if (!root->ref_cows) | |
255 | return 0; | |
256 | ||
257 | /* | |
258 | * see record_root_in_trans for comments about in_trans_setup usage | |
259 | * and barriers | |
260 | */ | |
261 | smp_rmb(); | |
262 | if (root->last_trans == trans->transid && | |
263 | !root->in_trans_setup) | |
264 | return 0; | |
265 | ||
266 | mutex_lock(&root->fs_info->reloc_mutex); | |
267 | record_root_in_trans(trans, root); | |
268 | mutex_unlock(&root->fs_info->reloc_mutex); | |
269 | ||
270 | return 0; | |
271 | } | |
272 | ||
d352ac68 CM |
273 | /* wait for commit against the current transaction to become unblocked |
274 | * when this is done, it is safe to start a new transaction, but the current | |
275 | * transaction might not be fully on disk. | |
276 | */ | |
37d1aeee | 277 | static void wait_current_trans(struct btrfs_root *root) |
79154b1b | 278 | { |
f9295749 | 279 | struct btrfs_transaction *cur_trans; |
79154b1b | 280 | |
a4abeea4 | 281 | spin_lock(&root->fs_info->trans_lock); |
f9295749 | 282 | cur_trans = root->fs_info->running_transaction; |
37d1aeee | 283 | if (cur_trans && cur_trans->blocked) { |
13c5a93e | 284 | atomic_inc(&cur_trans->use_count); |
a4abeea4 | 285 | spin_unlock(&root->fs_info->trans_lock); |
72d63ed6 LZ |
286 | |
287 | wait_event(root->fs_info->transaction_wait, | |
288 | !cur_trans->blocked); | |
f9295749 | 289 | put_transaction(cur_trans); |
a4abeea4 JB |
290 | } else { |
291 | spin_unlock(&root->fs_info->trans_lock); | |
f9295749 | 292 | } |
37d1aeee CM |
293 | } |
294 | ||
a22285a6 YZ |
295 | static int may_wait_transaction(struct btrfs_root *root, int type) |
296 | { | |
a4abeea4 JB |
297 | if (root->fs_info->log_root_recovering) |
298 | return 0; | |
299 | ||
300 | if (type == TRANS_USERSPACE) | |
301 | return 1; | |
302 | ||
303 | if (type == TRANS_START && | |
304 | !atomic_read(&root->fs_info->open_ioctl_trans)) | |
a22285a6 | 305 | return 1; |
a4abeea4 | 306 | |
a22285a6 YZ |
307 | return 0; |
308 | } | |
309 | ||
08e007d2 MX |
310 | static struct btrfs_trans_handle * |
311 | start_transaction(struct btrfs_root *root, u64 num_items, int type, | |
312 | enum btrfs_reserve_flush_enum flush) | |
37d1aeee | 313 | { |
a22285a6 YZ |
314 | struct btrfs_trans_handle *h; |
315 | struct btrfs_transaction *cur_trans; | |
b5009945 | 316 | u64 num_bytes = 0; |
37d1aeee | 317 | int ret; |
c5567237 | 318 | u64 qgroup_reserved = 0; |
acce952b | 319 | |
87533c47 | 320 | if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) |
acce952b | 321 | return ERR_PTR(-EROFS); |
2a1eb461 JB |
322 | |
323 | if (current->journal_info) { | |
324 | WARN_ON(type != TRANS_JOIN && type != TRANS_JOIN_NOLOCK); | |
325 | h = current->journal_info; | |
326 | h->use_count++; | |
b7d5b0a8 | 327 | WARN_ON(h->use_count > 2); |
2a1eb461 JB |
328 | h->orig_rsv = h->block_rsv; |
329 | h->block_rsv = NULL; | |
330 | goto got_it; | |
331 | } | |
b5009945 JB |
332 | |
333 | /* | |
334 | * Do the reservation before we join the transaction so we can do all | |
335 | * the appropriate flushing if need be. | |
336 | */ | |
337 | if (num_items > 0 && root != root->fs_info->chunk_root) { | |
c5567237 AJ |
338 | if (root->fs_info->quota_enabled && |
339 | is_fstree(root->root_key.objectid)) { | |
340 | qgroup_reserved = num_items * root->leafsize; | |
341 | ret = btrfs_qgroup_reserve(root, qgroup_reserved); | |
342 | if (ret) | |
343 | return ERR_PTR(ret); | |
344 | } | |
345 | ||
b5009945 | 346 | num_bytes = btrfs_calc_trans_metadata_size(root, num_items); |
08e007d2 MX |
347 | ret = btrfs_block_rsv_add(root, |
348 | &root->fs_info->trans_block_rsv, | |
349 | num_bytes, flush); | |
b5009945 | 350 | if (ret) |
843fcf35 | 351 | goto reserve_fail; |
b5009945 | 352 | } |
a22285a6 YZ |
353 | again: |
354 | h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS); | |
843fcf35 MX |
355 | if (!h) { |
356 | ret = -ENOMEM; | |
357 | goto alloc_fail; | |
358 | } | |
37d1aeee | 359 | |
98114659 JB |
360 | /* |
361 | * If we are JOIN_NOLOCK we're already committing a transaction and | |
362 | * waiting on this guy, so we don't need to do the sb_start_intwrite | |
363 | * because we're already holding a ref. We need this because we could | |
364 | * have raced in and did an fsync() on a file which can kick a commit | |
365 | * and then we deadlock with somebody doing a freeze. | |
354aa0fb MX |
366 | * |
367 | * If we are ATTACH, it means we just want to catch the current | |
368 | * transaction and commit it, so we needn't do sb_start_intwrite(). | |
98114659 | 369 | */ |
354aa0fb | 370 | if (type < TRANS_JOIN_NOLOCK) |
60376ce4 | 371 | sb_start_intwrite(root->fs_info->sb); |
b2b5ef5c | 372 | |
a22285a6 | 373 | if (may_wait_transaction(root, type)) |
37d1aeee | 374 | wait_current_trans(root); |
a22285a6 | 375 | |
a4abeea4 | 376 | do { |
354aa0fb | 377 | ret = join_transaction(root, type); |
178260b2 | 378 | if (ret == -EBUSY) { |
a4abeea4 | 379 | wait_current_trans(root); |
178260b2 MX |
380 | if (unlikely(type == TRANS_ATTACH)) |
381 | ret = -ENOENT; | |
382 | } | |
a4abeea4 JB |
383 | } while (ret == -EBUSY); |
384 | ||
db5b493a | 385 | if (ret < 0) { |
354aa0fb MX |
386 | /* We must get the transaction if we are JOIN_NOLOCK. */ |
387 | BUG_ON(type == TRANS_JOIN_NOLOCK); | |
843fcf35 | 388 | goto join_fail; |
db5b493a | 389 | } |
0f7d52f4 | 390 | |
a22285a6 | 391 | cur_trans = root->fs_info->running_transaction; |
a22285a6 YZ |
392 | |
393 | h->transid = cur_trans->transid; | |
394 | h->transaction = cur_trans; | |
79154b1b | 395 | h->blocks_used = 0; |
a22285a6 | 396 | h->bytes_reserved = 0; |
d13603ef | 397 | h->root = root; |
56bec294 | 398 | h->delayed_ref_updates = 0; |
2a1eb461 | 399 | h->use_count = 1; |
0e721106 | 400 | h->adding_csums = 0; |
f0486c68 | 401 | h->block_rsv = NULL; |
2a1eb461 | 402 | h->orig_rsv = NULL; |
49b25e05 | 403 | h->aborted = 0; |
4b824906 | 404 | h->qgroup_reserved = 0; |
bed92eae | 405 | h->delayed_ref_elem.seq = 0; |
a698d075 | 406 | h->type = type; |
c6b305a8 | 407 | h->allocating_chunk = false; |
bed92eae | 408 | INIT_LIST_HEAD(&h->qgroup_ref_list); |
ea658bad | 409 | INIT_LIST_HEAD(&h->new_bgs); |
b7ec40d7 | 410 | |
a22285a6 YZ |
411 | smp_mb(); |
412 | if (cur_trans->blocked && may_wait_transaction(root, type)) { | |
413 | btrfs_commit_transaction(h, root); | |
414 | goto again; | |
415 | } | |
416 | ||
b5009945 | 417 | if (num_bytes) { |
8c2a3ca2 | 418 | trace_btrfs_space_reservation(root->fs_info, "transaction", |
2bcc0328 | 419 | h->transid, num_bytes, 1); |
b5009945 JB |
420 | h->block_rsv = &root->fs_info->trans_block_rsv; |
421 | h->bytes_reserved = num_bytes; | |
a22285a6 | 422 | } |
4b824906 | 423 | h->qgroup_reserved = qgroup_reserved; |
9ed74f2d | 424 | |
2a1eb461 | 425 | got_it: |
a4abeea4 | 426 | btrfs_record_root_in_trans(h, root); |
a22285a6 YZ |
427 | |
428 | if (!current->journal_info && type != TRANS_USERSPACE) | |
429 | current->journal_info = h; | |
79154b1b | 430 | return h; |
843fcf35 MX |
431 | |
432 | join_fail: | |
433 | if (type < TRANS_JOIN_NOLOCK) | |
434 | sb_end_intwrite(root->fs_info->sb); | |
435 | kmem_cache_free(btrfs_trans_handle_cachep, h); | |
436 | alloc_fail: | |
437 | if (num_bytes) | |
438 | btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv, | |
439 | num_bytes); | |
440 | reserve_fail: | |
441 | if (qgroup_reserved) | |
442 | btrfs_qgroup_free(root, qgroup_reserved); | |
443 | return ERR_PTR(ret); | |
79154b1b CM |
444 | } |
445 | ||
f9295749 | 446 | struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, |
a22285a6 | 447 | int num_items) |
f9295749 | 448 | { |
08e007d2 MX |
449 | return start_transaction(root, num_items, TRANS_START, |
450 | BTRFS_RESERVE_FLUSH_ALL); | |
f9295749 | 451 | } |
8407aa46 | 452 | |
08e007d2 | 453 | struct btrfs_trans_handle *btrfs_start_transaction_lflush( |
8407aa46 MX |
454 | struct btrfs_root *root, int num_items) |
455 | { | |
08e007d2 MX |
456 | return start_transaction(root, num_items, TRANS_START, |
457 | BTRFS_RESERVE_FLUSH_LIMIT); | |
8407aa46 MX |
458 | } |
459 | ||
7a7eaa40 | 460 | struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root) |
f9295749 | 461 | { |
8407aa46 | 462 | return start_transaction(root, 0, TRANS_JOIN, 0); |
f9295749 CM |
463 | } |
464 | ||
7a7eaa40 | 465 | struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root) |
0af3d00b | 466 | { |
8407aa46 | 467 | return start_transaction(root, 0, TRANS_JOIN_NOLOCK, 0); |
0af3d00b JB |
468 | } |
469 | ||
7a7eaa40 | 470 | struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root) |
9ca9ee09 | 471 | { |
8407aa46 | 472 | return start_transaction(root, 0, TRANS_USERSPACE, 0); |
9ca9ee09 SW |
473 | } |
474 | ||
d4edf39b MX |
475 | /* |
476 | * btrfs_attach_transaction() - catch the running transaction | |
477 | * | |
478 | * It is used when we want to commit the current the transaction, but | |
479 | * don't want to start a new one. | |
480 | * | |
481 | * Note: If this function return -ENOENT, it just means there is no | |
482 | * running transaction. But it is possible that the inactive transaction | |
483 | * is still in the memory, not fully on disk. If you hope there is no | |
484 | * inactive transaction in the fs when -ENOENT is returned, you should | |
485 | * invoke | |
486 | * btrfs_attach_transaction_barrier() | |
487 | */ | |
354aa0fb | 488 | struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root) |
60376ce4 | 489 | { |
354aa0fb | 490 | return start_transaction(root, 0, TRANS_ATTACH, 0); |
60376ce4 JB |
491 | } |
492 | ||
d4edf39b MX |
493 | /* |
494 | * btrfs_attach_transaction() - catch the running transaction | |
495 | * | |
496 | * It is similar to the above function, the differentia is this one | |
497 | * will wait for all the inactive transactions until they fully | |
498 | * complete. | |
499 | */ | |
500 | struct btrfs_trans_handle * | |
501 | btrfs_attach_transaction_barrier(struct btrfs_root *root) | |
502 | { | |
503 | struct btrfs_trans_handle *trans; | |
504 | ||
505 | trans = start_transaction(root, 0, TRANS_ATTACH, 0); | |
506 | if (IS_ERR(trans) && PTR_ERR(trans) == -ENOENT) | |
507 | btrfs_wait_for_commit(root, 0); | |
508 | ||
509 | return trans; | |
510 | } | |
511 | ||
d352ac68 | 512 | /* wait for a transaction commit to be fully complete */ |
b9c8300c | 513 | static noinline void wait_for_commit(struct btrfs_root *root, |
89ce8a63 CM |
514 | struct btrfs_transaction *commit) |
515 | { | |
72d63ed6 | 516 | wait_event(commit->commit_wait, commit->commit_done); |
89ce8a63 CM |
517 | } |
518 | ||
46204592 SW |
519 | int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid) |
520 | { | |
521 | struct btrfs_transaction *cur_trans = NULL, *t; | |
8cd2807f | 522 | int ret = 0; |
46204592 | 523 | |
46204592 SW |
524 | if (transid) { |
525 | if (transid <= root->fs_info->last_trans_committed) | |
a4abeea4 | 526 | goto out; |
46204592 | 527 | |
8cd2807f | 528 | ret = -EINVAL; |
46204592 | 529 | /* find specified transaction */ |
a4abeea4 | 530 | spin_lock(&root->fs_info->trans_lock); |
46204592 SW |
531 | list_for_each_entry(t, &root->fs_info->trans_list, list) { |
532 | if (t->transid == transid) { | |
533 | cur_trans = t; | |
a4abeea4 | 534 | atomic_inc(&cur_trans->use_count); |
8cd2807f | 535 | ret = 0; |
46204592 SW |
536 | break; |
537 | } | |
8cd2807f MX |
538 | if (t->transid > transid) { |
539 | ret = 0; | |
46204592 | 540 | break; |
8cd2807f | 541 | } |
46204592 | 542 | } |
a4abeea4 | 543 | spin_unlock(&root->fs_info->trans_lock); |
8cd2807f | 544 | /* The specified transaction doesn't exist */ |
46204592 | 545 | if (!cur_trans) |
8cd2807f | 546 | goto out; |
46204592 SW |
547 | } else { |
548 | /* find newest transaction that is committing | committed */ | |
a4abeea4 | 549 | spin_lock(&root->fs_info->trans_lock); |
46204592 SW |
550 | list_for_each_entry_reverse(t, &root->fs_info->trans_list, |
551 | list) { | |
552 | if (t->in_commit) { | |
553 | if (t->commit_done) | |
3473f3c0 | 554 | break; |
46204592 | 555 | cur_trans = t; |
a4abeea4 | 556 | atomic_inc(&cur_trans->use_count); |
46204592 SW |
557 | break; |
558 | } | |
559 | } | |
a4abeea4 | 560 | spin_unlock(&root->fs_info->trans_lock); |
46204592 | 561 | if (!cur_trans) |
a4abeea4 | 562 | goto out; /* nothing committing|committed */ |
46204592 SW |
563 | } |
564 | ||
46204592 | 565 | wait_for_commit(root, cur_trans); |
46204592 | 566 | put_transaction(cur_trans); |
a4abeea4 | 567 | out: |
46204592 SW |
568 | return ret; |
569 | } | |
570 | ||
37d1aeee CM |
571 | void btrfs_throttle(struct btrfs_root *root) |
572 | { | |
a4abeea4 | 573 | if (!atomic_read(&root->fs_info->open_ioctl_trans)) |
9ca9ee09 | 574 | wait_current_trans(root); |
37d1aeee CM |
575 | } |
576 | ||
8929ecfa YZ |
577 | static int should_end_transaction(struct btrfs_trans_handle *trans, |
578 | struct btrfs_root *root) | |
579 | { | |
580 | int ret; | |
36ba022a JB |
581 | |
582 | ret = btrfs_block_rsv_check(root, &root->fs_info->global_block_rsv, 5); | |
8929ecfa YZ |
583 | return ret ? 1 : 0; |
584 | } | |
585 | ||
586 | int btrfs_should_end_transaction(struct btrfs_trans_handle *trans, | |
587 | struct btrfs_root *root) | |
588 | { | |
589 | struct btrfs_transaction *cur_trans = trans->transaction; | |
590 | int updates; | |
49b25e05 | 591 | int err; |
8929ecfa | 592 | |
a4abeea4 | 593 | smp_mb(); |
8929ecfa YZ |
594 | if (cur_trans->blocked || cur_trans->delayed_refs.flushing) |
595 | return 1; | |
596 | ||
597 | updates = trans->delayed_ref_updates; | |
598 | trans->delayed_ref_updates = 0; | |
49b25e05 JM |
599 | if (updates) { |
600 | err = btrfs_run_delayed_refs(trans, root, updates); | |
601 | if (err) /* Error code will also eval true */ | |
602 | return err; | |
603 | } | |
8929ecfa YZ |
604 | |
605 | return should_end_transaction(trans, root); | |
606 | } | |
607 | ||
89ce8a63 | 608 | static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, |
a698d075 | 609 | struct btrfs_root *root, int throttle) |
79154b1b | 610 | { |
8929ecfa | 611 | struct btrfs_transaction *cur_trans = trans->transaction; |
ab78c84d | 612 | struct btrfs_fs_info *info = root->fs_info; |
c3e69d58 | 613 | int count = 0; |
a698d075 | 614 | int lock = (trans->type != TRANS_JOIN_NOLOCK); |
4edc2ca3 | 615 | int err = 0; |
c3e69d58 | 616 | |
2a1eb461 JB |
617 | if (--trans->use_count) { |
618 | trans->block_rsv = trans->orig_rsv; | |
619 | return 0; | |
620 | } | |
621 | ||
edf39272 JS |
622 | /* |
623 | * do the qgroup accounting as early as possible | |
624 | */ | |
625 | err = btrfs_delayed_refs_qgroup_accounting(trans, info); | |
626 | ||
b24e03db | 627 | btrfs_trans_release_metadata(trans, root); |
4c13d758 | 628 | trans->block_rsv = NULL; |
d13603ef AJ |
629 | /* |
630 | * the same root has to be passed to start_transaction and | |
631 | * end_transaction. Subvolume quota depends on this. | |
632 | */ | |
633 | WARN_ON(trans->root != root); | |
c5567237 AJ |
634 | |
635 | if (trans->qgroup_reserved) { | |
636 | btrfs_qgroup_free(root, trans->qgroup_reserved); | |
637 | trans->qgroup_reserved = 0; | |
638 | } | |
639 | ||
ea658bad JB |
640 | if (!list_empty(&trans->new_bgs)) |
641 | btrfs_create_pending_block_groups(trans, root); | |
642 | ||
bb721703 | 643 | while (count < 1) { |
c3e69d58 CM |
644 | unsigned long cur = trans->delayed_ref_updates; |
645 | trans->delayed_ref_updates = 0; | |
646 | if (cur && | |
647 | trans->transaction->delayed_refs.num_heads_ready > 64) { | |
648 | trans->delayed_ref_updates = 0; | |
649 | btrfs_run_delayed_refs(trans, root, cur); | |
650 | } else { | |
651 | break; | |
652 | } | |
653 | count++; | |
56bec294 | 654 | } |
bb721703 | 655 | |
0e721106 JB |
656 | btrfs_trans_release_metadata(trans, root); |
657 | trans->block_rsv = NULL; | |
56bec294 | 658 | |
ea658bad JB |
659 | if (!list_empty(&trans->new_bgs)) |
660 | btrfs_create_pending_block_groups(trans, root); | |
661 | ||
a4abeea4 JB |
662 | if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) && |
663 | should_end_transaction(trans, root)) { | |
8929ecfa | 664 | trans->transaction->blocked = 1; |
a4abeea4 JB |
665 | smp_wmb(); |
666 | } | |
8929ecfa | 667 | |
0af3d00b | 668 | if (lock && cur_trans->blocked && !cur_trans->in_commit) { |
81317fde JB |
669 | if (throttle) { |
670 | /* | |
671 | * We may race with somebody else here so end up having | |
672 | * to call end_transaction on ourselves again, so inc | |
673 | * our use_count. | |
674 | */ | |
675 | trans->use_count++; | |
8929ecfa | 676 | return btrfs_commit_transaction(trans, root); |
81317fde | 677 | } else { |
8929ecfa | 678 | wake_up_process(info->transaction_kthread); |
81317fde | 679 | } |
8929ecfa YZ |
680 | } |
681 | ||
354aa0fb | 682 | if (trans->type < TRANS_JOIN_NOLOCK) |
98114659 | 683 | sb_end_intwrite(root->fs_info->sb); |
6df7881a | 684 | |
8929ecfa | 685 | WARN_ON(cur_trans != info->running_transaction); |
13c5a93e JB |
686 | WARN_ON(atomic_read(&cur_trans->num_writers) < 1); |
687 | atomic_dec(&cur_trans->num_writers); | |
89ce8a63 | 688 | |
99d16cbc | 689 | smp_mb(); |
79154b1b CM |
690 | if (waitqueue_active(&cur_trans->writer_wait)) |
691 | wake_up(&cur_trans->writer_wait); | |
79154b1b | 692 | put_transaction(cur_trans); |
9ed74f2d JB |
693 | |
694 | if (current->journal_info == trans) | |
695 | current->journal_info = NULL; | |
ab78c84d | 696 | |
24bbcf04 YZ |
697 | if (throttle) |
698 | btrfs_run_delayed_iputs(root); | |
699 | ||
49b25e05 | 700 | if (trans->aborted || |
87533c47 | 701 | test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) |
4edc2ca3 | 702 | err = -EIO; |
edf39272 | 703 | assert_qgroups_uptodate(trans); |
49b25e05 | 704 | |
4edc2ca3 DJ |
705 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
706 | return err; | |
79154b1b CM |
707 | } |
708 | ||
89ce8a63 CM |
709 | int btrfs_end_transaction(struct btrfs_trans_handle *trans, |
710 | struct btrfs_root *root) | |
711 | { | |
16cdcec7 MX |
712 | int ret; |
713 | ||
a698d075 | 714 | ret = __btrfs_end_transaction(trans, root, 0); |
16cdcec7 MX |
715 | if (ret) |
716 | return ret; | |
717 | return 0; | |
89ce8a63 CM |
718 | } |
719 | ||
720 | int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans, | |
721 | struct btrfs_root *root) | |
722 | { | |
16cdcec7 MX |
723 | int ret; |
724 | ||
a698d075 | 725 | ret = __btrfs_end_transaction(trans, root, 1); |
16cdcec7 MX |
726 | if (ret) |
727 | return ret; | |
728 | return 0; | |
729 | } | |
730 | ||
731 | int btrfs_end_transaction_dmeta(struct btrfs_trans_handle *trans, | |
732 | struct btrfs_root *root) | |
733 | { | |
a698d075 | 734 | return __btrfs_end_transaction(trans, root, 1); |
89ce8a63 CM |
735 | } |
736 | ||
d352ac68 CM |
737 | /* |
738 | * when btree blocks are allocated, they have some corresponding bits set for | |
739 | * them in one of two extent_io trees. This is used to make sure all of | |
690587d1 | 740 | * those extents are sent to disk but does not wait on them |
d352ac68 | 741 | */ |
690587d1 | 742 | int btrfs_write_marked_extents(struct btrfs_root *root, |
8cef4e16 | 743 | struct extent_io_tree *dirty_pages, int mark) |
79154b1b | 744 | { |
777e6bd7 | 745 | int err = 0; |
7c4452b9 | 746 | int werr = 0; |
1728366e | 747 | struct address_space *mapping = root->fs_info->btree_inode->i_mapping; |
e6138876 | 748 | struct extent_state *cached_state = NULL; |
777e6bd7 | 749 | u64 start = 0; |
5f39d397 | 750 | u64 end; |
53b381b3 | 751 | struct blk_plug plug; |
7c4452b9 | 752 | |
53b381b3 | 753 | blk_start_plug(&plug); |
1728366e | 754 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
e6138876 JB |
755 | mark, &cached_state)) { |
756 | convert_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT, | |
757 | mark, &cached_state, GFP_NOFS); | |
758 | cached_state = NULL; | |
1728366e JB |
759 | err = filemap_fdatawrite_range(mapping, start, end); |
760 | if (err) | |
761 | werr = err; | |
762 | cond_resched(); | |
763 | start = end + 1; | |
7c4452b9 | 764 | } |
690587d1 CM |
765 | if (err) |
766 | werr = err; | |
53b381b3 | 767 | blk_finish_plug(&plug); |
690587d1 CM |
768 | return werr; |
769 | } | |
770 | ||
771 | /* | |
772 | * when btree blocks are allocated, they have some corresponding bits set for | |
773 | * them in one of two extent_io trees. This is used to make sure all of | |
774 | * those extents are on disk for transaction or log commit. We wait | |
775 | * on all the pages and clear them from the dirty pages state tree | |
776 | */ | |
777 | int btrfs_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 778 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 | 779 | { |
690587d1 CM |
780 | int err = 0; |
781 | int werr = 0; | |
1728366e | 782 | struct address_space *mapping = root->fs_info->btree_inode->i_mapping; |
e6138876 | 783 | struct extent_state *cached_state = NULL; |
690587d1 CM |
784 | u64 start = 0; |
785 | u64 end; | |
777e6bd7 | 786 | |
1728366e | 787 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
e6138876 JB |
788 | EXTENT_NEED_WAIT, &cached_state)) { |
789 | clear_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT, | |
790 | 0, 0, &cached_state, GFP_NOFS); | |
1728366e JB |
791 | err = filemap_fdatawait_range(mapping, start, end); |
792 | if (err) | |
793 | werr = err; | |
794 | cond_resched(); | |
795 | start = end + 1; | |
777e6bd7 | 796 | } |
7c4452b9 CM |
797 | if (err) |
798 | werr = err; | |
799 | return werr; | |
79154b1b CM |
800 | } |
801 | ||
690587d1 CM |
802 | /* |
803 | * when btree blocks are allocated, they have some corresponding bits set for | |
804 | * them in one of two extent_io trees. This is used to make sure all of | |
805 | * those extents are on disk for transaction or log commit | |
806 | */ | |
807 | int btrfs_write_and_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 808 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 CM |
809 | { |
810 | int ret; | |
811 | int ret2; | |
812 | ||
8cef4e16 YZ |
813 | ret = btrfs_write_marked_extents(root, dirty_pages, mark); |
814 | ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark); | |
bf0da8c1 CM |
815 | |
816 | if (ret) | |
817 | return ret; | |
818 | if (ret2) | |
819 | return ret2; | |
820 | return 0; | |
690587d1 CM |
821 | } |
822 | ||
d0c803c4 CM |
823 | int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans, |
824 | struct btrfs_root *root) | |
825 | { | |
826 | if (!trans || !trans->transaction) { | |
827 | struct inode *btree_inode; | |
828 | btree_inode = root->fs_info->btree_inode; | |
829 | return filemap_write_and_wait(btree_inode->i_mapping); | |
830 | } | |
831 | return btrfs_write_and_wait_marked_extents(root, | |
8cef4e16 YZ |
832 | &trans->transaction->dirty_pages, |
833 | EXTENT_DIRTY); | |
d0c803c4 CM |
834 | } |
835 | ||
d352ac68 CM |
836 | /* |
837 | * this is used to update the root pointer in the tree of tree roots. | |
838 | * | |
839 | * But, in the case of the extent allocation tree, updating the root | |
840 | * pointer may allocate blocks which may change the root of the extent | |
841 | * allocation tree. | |
842 | * | |
843 | * So, this loops and repeats and makes sure the cowonly root didn't | |
844 | * change while the root pointer was being updated in the metadata. | |
845 | */ | |
0b86a832 CM |
846 | static int update_cowonly_root(struct btrfs_trans_handle *trans, |
847 | struct btrfs_root *root) | |
79154b1b CM |
848 | { |
849 | int ret; | |
0b86a832 | 850 | u64 old_root_bytenr; |
86b9f2ec | 851 | u64 old_root_used; |
0b86a832 | 852 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
79154b1b | 853 | |
86b9f2ec | 854 | old_root_used = btrfs_root_used(&root->root_item); |
0b86a832 | 855 | btrfs_write_dirty_block_groups(trans, root); |
56bec294 | 856 | |
d397712b | 857 | while (1) { |
0b86a832 | 858 | old_root_bytenr = btrfs_root_bytenr(&root->root_item); |
86b9f2ec YZ |
859 | if (old_root_bytenr == root->node->start && |
860 | old_root_used == btrfs_root_used(&root->root_item)) | |
79154b1b | 861 | break; |
87ef2bb4 | 862 | |
5d4f98a2 | 863 | btrfs_set_root_node(&root->root_item, root->node); |
79154b1b | 864 | ret = btrfs_update_root(trans, tree_root, |
0b86a832 CM |
865 | &root->root_key, |
866 | &root->root_item); | |
49b25e05 JM |
867 | if (ret) |
868 | return ret; | |
56bec294 | 869 | |
86b9f2ec | 870 | old_root_used = btrfs_root_used(&root->root_item); |
4a8c9a62 | 871 | ret = btrfs_write_dirty_block_groups(trans, root); |
49b25e05 JM |
872 | if (ret) |
873 | return ret; | |
0b86a832 | 874 | } |
276e680d YZ |
875 | |
876 | if (root != root->fs_info->extent_root) | |
877 | switch_commit_root(root); | |
878 | ||
0b86a832 CM |
879 | return 0; |
880 | } | |
881 | ||
d352ac68 CM |
882 | /* |
883 | * update all the cowonly tree roots on disk | |
49b25e05 JM |
884 | * |
885 | * The error handling in this function may not be obvious. Any of the | |
886 | * failures will cause the file system to go offline. We still need | |
887 | * to clean up the delayed refs. | |
d352ac68 | 888 | */ |
5d4f98a2 YZ |
889 | static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans, |
890 | struct btrfs_root *root) | |
0b86a832 CM |
891 | { |
892 | struct btrfs_fs_info *fs_info = root->fs_info; | |
893 | struct list_head *next; | |
84234f3a | 894 | struct extent_buffer *eb; |
56bec294 | 895 | int ret; |
84234f3a | 896 | |
56bec294 | 897 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
49b25e05 JM |
898 | if (ret) |
899 | return ret; | |
87ef2bb4 | 900 | |
84234f3a | 901 | eb = btrfs_lock_root_node(fs_info->tree_root); |
49b25e05 JM |
902 | ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, |
903 | 0, &eb); | |
84234f3a YZ |
904 | btrfs_tree_unlock(eb); |
905 | free_extent_buffer(eb); | |
0b86a832 | 906 | |
49b25e05 JM |
907 | if (ret) |
908 | return ret; | |
909 | ||
56bec294 | 910 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
49b25e05 JM |
911 | if (ret) |
912 | return ret; | |
87ef2bb4 | 913 | |
733f4fbb | 914 | ret = btrfs_run_dev_stats(trans, root->fs_info); |
8dabb742 SB |
915 | WARN_ON(ret); |
916 | ret = btrfs_run_dev_replace(trans, root->fs_info); | |
917 | WARN_ON(ret); | |
733f4fbb | 918 | |
546adb0d JS |
919 | ret = btrfs_run_qgroups(trans, root->fs_info); |
920 | BUG_ON(ret); | |
921 | ||
922 | /* run_qgroups might have added some more refs */ | |
923 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); | |
924 | BUG_ON(ret); | |
925 | ||
d397712b | 926 | while (!list_empty(&fs_info->dirty_cowonly_roots)) { |
0b86a832 CM |
927 | next = fs_info->dirty_cowonly_roots.next; |
928 | list_del_init(next); | |
929 | root = list_entry(next, struct btrfs_root, dirty_list); | |
87ef2bb4 | 930 | |
49b25e05 JM |
931 | ret = update_cowonly_root(trans, root); |
932 | if (ret) | |
933 | return ret; | |
79154b1b | 934 | } |
276e680d YZ |
935 | |
936 | down_write(&fs_info->extent_commit_sem); | |
937 | switch_commit_root(fs_info->extent_root); | |
938 | up_write(&fs_info->extent_commit_sem); | |
939 | ||
8dabb742 SB |
940 | btrfs_after_dev_replace_commit(fs_info); |
941 | ||
79154b1b CM |
942 | return 0; |
943 | } | |
944 | ||
d352ac68 CM |
945 | /* |
946 | * dead roots are old snapshots that need to be deleted. This allocates | |
947 | * a dirty root struct and adds it into the list of dead roots that need to | |
948 | * be deleted | |
949 | */ | |
5d4f98a2 | 950 | int btrfs_add_dead_root(struct btrfs_root *root) |
5eda7b5e | 951 | { |
a4abeea4 | 952 | spin_lock(&root->fs_info->trans_lock); |
5d4f98a2 | 953 | list_add(&root->root_list, &root->fs_info->dead_roots); |
a4abeea4 | 954 | spin_unlock(&root->fs_info->trans_lock); |
5eda7b5e CM |
955 | return 0; |
956 | } | |
957 | ||
d352ac68 | 958 | /* |
5d4f98a2 | 959 | * update all the cowonly tree roots on disk |
d352ac68 | 960 | */ |
5d4f98a2 YZ |
961 | static noinline int commit_fs_roots(struct btrfs_trans_handle *trans, |
962 | struct btrfs_root *root) | |
0f7d52f4 | 963 | { |
0f7d52f4 | 964 | struct btrfs_root *gang[8]; |
5d4f98a2 | 965 | struct btrfs_fs_info *fs_info = root->fs_info; |
0f7d52f4 CM |
966 | int i; |
967 | int ret; | |
54aa1f4d CM |
968 | int err = 0; |
969 | ||
a4abeea4 | 970 | spin_lock(&fs_info->fs_roots_radix_lock); |
d397712b | 971 | while (1) { |
5d4f98a2 YZ |
972 | ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, |
973 | (void **)gang, 0, | |
0f7d52f4 CM |
974 | ARRAY_SIZE(gang), |
975 | BTRFS_ROOT_TRANS_TAG); | |
976 | if (ret == 0) | |
977 | break; | |
978 | for (i = 0; i < ret; i++) { | |
979 | root = gang[i]; | |
5d4f98a2 YZ |
980 | radix_tree_tag_clear(&fs_info->fs_roots_radix, |
981 | (unsigned long)root->root_key.objectid, | |
982 | BTRFS_ROOT_TRANS_TAG); | |
a4abeea4 | 983 | spin_unlock(&fs_info->fs_roots_radix_lock); |
31153d81 | 984 | |
e02119d5 | 985 | btrfs_free_log(trans, root); |
5d4f98a2 | 986 | btrfs_update_reloc_root(trans, root); |
d68fc57b | 987 | btrfs_orphan_commit_root(trans, root); |
bcc63abb | 988 | |
82d5902d LZ |
989 | btrfs_save_ino_cache(root, trans); |
990 | ||
f1ebcc74 LB |
991 | /* see comments in should_cow_block() */ |
992 | root->force_cow = 0; | |
993 | smp_wmb(); | |
994 | ||
978d910d | 995 | if (root->commit_root != root->node) { |
581bb050 | 996 | mutex_lock(&root->fs_commit_mutex); |
817d52f8 | 997 | switch_commit_root(root); |
581bb050 LZ |
998 | btrfs_unpin_free_ino(root); |
999 | mutex_unlock(&root->fs_commit_mutex); | |
1000 | ||
978d910d YZ |
1001 | btrfs_set_root_node(&root->root_item, |
1002 | root->node); | |
1003 | } | |
5d4f98a2 | 1004 | |
5d4f98a2 | 1005 | err = btrfs_update_root(trans, fs_info->tree_root, |
0f7d52f4 CM |
1006 | &root->root_key, |
1007 | &root->root_item); | |
a4abeea4 | 1008 | spin_lock(&fs_info->fs_roots_radix_lock); |
54aa1f4d CM |
1009 | if (err) |
1010 | break; | |
0f7d52f4 CM |
1011 | } |
1012 | } | |
a4abeea4 | 1013 | spin_unlock(&fs_info->fs_roots_radix_lock); |
54aa1f4d | 1014 | return err; |
0f7d52f4 CM |
1015 | } |
1016 | ||
d352ac68 | 1017 | /* |
de78b51a ES |
1018 | * defrag a given btree. |
1019 | * Every leaf in the btree is read and defragged. | |
d352ac68 | 1020 | */ |
de78b51a | 1021 | int btrfs_defrag_root(struct btrfs_root *root) |
e9d0b13b CM |
1022 | { |
1023 | struct btrfs_fs_info *info = root->fs_info; | |
e9d0b13b | 1024 | struct btrfs_trans_handle *trans; |
8929ecfa | 1025 | int ret; |
e9d0b13b | 1026 | |
8929ecfa | 1027 | if (xchg(&root->defrag_running, 1)) |
e9d0b13b | 1028 | return 0; |
8929ecfa | 1029 | |
6b80053d | 1030 | while (1) { |
8929ecfa YZ |
1031 | trans = btrfs_start_transaction(root, 0); |
1032 | if (IS_ERR(trans)) | |
1033 | return PTR_ERR(trans); | |
1034 | ||
de78b51a | 1035 | ret = btrfs_defrag_leaves(trans, root); |
8929ecfa | 1036 | |
e9d0b13b | 1037 | btrfs_end_transaction(trans, root); |
b53d3f5d | 1038 | btrfs_btree_balance_dirty(info->tree_root); |
e9d0b13b CM |
1039 | cond_resched(); |
1040 | ||
7841cb28 | 1041 | if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN) |
e9d0b13b | 1042 | break; |
210549eb DS |
1043 | |
1044 | if (btrfs_defrag_cancelled(root->fs_info)) { | |
1045 | printk(KERN_DEBUG "btrfs: defrag_root cancelled\n"); | |
1046 | ret = -EAGAIN; | |
1047 | break; | |
1048 | } | |
e9d0b13b CM |
1049 | } |
1050 | root->defrag_running = 0; | |
8929ecfa | 1051 | return ret; |
e9d0b13b CM |
1052 | } |
1053 | ||
d352ac68 CM |
1054 | /* |
1055 | * new snapshots need to be created at a very specific time in the | |
1056 | * transaction commit. This does the actual creation | |
1057 | */ | |
80b6794d | 1058 | static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, |
3063d29f CM |
1059 | struct btrfs_fs_info *fs_info, |
1060 | struct btrfs_pending_snapshot *pending) | |
1061 | { | |
1062 | struct btrfs_key key; | |
80b6794d | 1063 | struct btrfs_root_item *new_root_item; |
3063d29f CM |
1064 | struct btrfs_root *tree_root = fs_info->tree_root; |
1065 | struct btrfs_root *root = pending->root; | |
6bdb72de | 1066 | struct btrfs_root *parent_root; |
98c9942a | 1067 | struct btrfs_block_rsv *rsv; |
6bdb72de | 1068 | struct inode *parent_inode; |
42874b3d MX |
1069 | struct btrfs_path *path; |
1070 | struct btrfs_dir_item *dir_item; | |
6a912213 | 1071 | struct dentry *parent; |
a22285a6 | 1072 | struct dentry *dentry; |
3063d29f | 1073 | struct extent_buffer *tmp; |
925baedd | 1074 | struct extent_buffer *old; |
8ea05e3a | 1075 | struct timespec cur_time = CURRENT_TIME; |
3063d29f | 1076 | int ret; |
d68fc57b | 1077 | u64 to_reserve = 0; |
6bdb72de | 1078 | u64 index = 0; |
a22285a6 | 1079 | u64 objectid; |
b83cc969 | 1080 | u64 root_flags; |
8ea05e3a | 1081 | uuid_le new_uuid; |
3063d29f | 1082 | |
42874b3d MX |
1083 | path = btrfs_alloc_path(); |
1084 | if (!path) { | |
1085 | ret = pending->error = -ENOMEM; | |
1086 | goto path_alloc_fail; | |
1087 | } | |
1088 | ||
80b6794d CM |
1089 | new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS); |
1090 | if (!new_root_item) { | |
49b25e05 | 1091 | ret = pending->error = -ENOMEM; |
6fa9700e | 1092 | goto root_item_alloc_fail; |
80b6794d | 1093 | } |
a22285a6 | 1094 | |
581bb050 | 1095 | ret = btrfs_find_free_objectid(tree_root, &objectid); |
a22285a6 YZ |
1096 | if (ret) { |
1097 | pending->error = ret; | |
6fa9700e | 1098 | goto no_free_objectid; |
a22285a6 | 1099 | } |
3063d29f | 1100 | |
3fd0a558 | 1101 | btrfs_reloc_pre_snapshot(trans, pending, &to_reserve); |
d68fc57b YZ |
1102 | |
1103 | if (to_reserve > 0) { | |
08e007d2 MX |
1104 | ret = btrfs_block_rsv_add(root, &pending->block_rsv, |
1105 | to_reserve, | |
1106 | BTRFS_RESERVE_NO_FLUSH); | |
d68fc57b YZ |
1107 | if (ret) { |
1108 | pending->error = ret; | |
6fa9700e | 1109 | goto no_free_objectid; |
d68fc57b YZ |
1110 | } |
1111 | } | |
1112 | ||
6f72c7e2 AJ |
1113 | ret = btrfs_qgroup_inherit(trans, fs_info, root->root_key.objectid, |
1114 | objectid, pending->inherit); | |
6f72c7e2 AJ |
1115 | if (ret) { |
1116 | pending->error = ret; | |
6fa9700e | 1117 | goto no_free_objectid; |
6f72c7e2 AJ |
1118 | } |
1119 | ||
3063d29f | 1120 | key.objectid = objectid; |
a22285a6 YZ |
1121 | key.offset = (u64)-1; |
1122 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3063d29f | 1123 | |
6fa9700e | 1124 | rsv = trans->block_rsv; |
a22285a6 | 1125 | trans->block_rsv = &pending->block_rsv; |
3de4586c | 1126 | |
a22285a6 | 1127 | dentry = pending->dentry; |
6a912213 JB |
1128 | parent = dget_parent(dentry); |
1129 | parent_inode = parent->d_inode; | |
a22285a6 | 1130 | parent_root = BTRFS_I(parent_inode)->root; |
7585717f | 1131 | record_root_in_trans(trans, parent_root); |
a22285a6 | 1132 | |
3063d29f CM |
1133 | /* |
1134 | * insert the directory item | |
1135 | */ | |
3de4586c | 1136 | ret = btrfs_set_inode_index(parent_inode, &index); |
49b25e05 | 1137 | BUG_ON(ret); /* -ENOMEM */ |
42874b3d MX |
1138 | |
1139 | /* check if there is a file/dir which has the same name. */ | |
1140 | dir_item = btrfs_lookup_dir_item(NULL, parent_root, path, | |
1141 | btrfs_ino(parent_inode), | |
1142 | dentry->d_name.name, | |
1143 | dentry->d_name.len, 0); | |
1144 | if (dir_item != NULL && !IS_ERR(dir_item)) { | |
fe66a05a | 1145 | pending->error = -EEXIST; |
fe66a05a | 1146 | goto fail; |
42874b3d MX |
1147 | } else if (IS_ERR(dir_item)) { |
1148 | ret = PTR_ERR(dir_item); | |
8732d44f MX |
1149 | btrfs_abort_transaction(trans, root, ret); |
1150 | goto fail; | |
79787eaa | 1151 | } |
42874b3d | 1152 | btrfs_release_path(path); |
52c26179 | 1153 | |
e999376f CM |
1154 | /* |
1155 | * pull in the delayed directory update | |
1156 | * and the delayed inode item | |
1157 | * otherwise we corrupt the FS during | |
1158 | * snapshot | |
1159 | */ | |
1160 | ret = btrfs_run_delayed_items(trans, root); | |
8732d44f MX |
1161 | if (ret) { /* Transaction aborted */ |
1162 | btrfs_abort_transaction(trans, root, ret); | |
1163 | goto fail; | |
1164 | } | |
e999376f | 1165 | |
7585717f | 1166 | record_root_in_trans(trans, root); |
6bdb72de SW |
1167 | btrfs_set_root_last_snapshot(&root->root_item, trans->transid); |
1168 | memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); | |
08fe4db1 | 1169 | btrfs_check_and_init_root_item(new_root_item); |
6bdb72de | 1170 | |
b83cc969 LZ |
1171 | root_flags = btrfs_root_flags(new_root_item); |
1172 | if (pending->readonly) | |
1173 | root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; | |
1174 | else | |
1175 | root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; | |
1176 | btrfs_set_root_flags(new_root_item, root_flags); | |
1177 | ||
8ea05e3a AB |
1178 | btrfs_set_root_generation_v2(new_root_item, |
1179 | trans->transid); | |
1180 | uuid_le_gen(&new_uuid); | |
1181 | memcpy(new_root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE); | |
1182 | memcpy(new_root_item->parent_uuid, root->root_item.uuid, | |
1183 | BTRFS_UUID_SIZE); | |
1184 | new_root_item->otime.sec = cpu_to_le64(cur_time.tv_sec); | |
dadd1105 | 1185 | new_root_item->otime.nsec = cpu_to_le32(cur_time.tv_nsec); |
8ea05e3a AB |
1186 | btrfs_set_root_otransid(new_root_item, trans->transid); |
1187 | memset(&new_root_item->stime, 0, sizeof(new_root_item->stime)); | |
1188 | memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime)); | |
1189 | btrfs_set_root_stransid(new_root_item, 0); | |
1190 | btrfs_set_root_rtransid(new_root_item, 0); | |
1191 | ||
6bdb72de | 1192 | old = btrfs_lock_root_node(root); |
49b25e05 | 1193 | ret = btrfs_cow_block(trans, root, old, NULL, 0, &old); |
79787eaa JM |
1194 | if (ret) { |
1195 | btrfs_tree_unlock(old); | |
1196 | free_extent_buffer(old); | |
8732d44f MX |
1197 | btrfs_abort_transaction(trans, root, ret); |
1198 | goto fail; | |
79787eaa | 1199 | } |
49b25e05 | 1200 | |
6bdb72de SW |
1201 | btrfs_set_lock_blocking(old); |
1202 | ||
49b25e05 | 1203 | ret = btrfs_copy_root(trans, root, old, &tmp, objectid); |
79787eaa | 1204 | /* clean up in any case */ |
6bdb72de SW |
1205 | btrfs_tree_unlock(old); |
1206 | free_extent_buffer(old); | |
8732d44f MX |
1207 | if (ret) { |
1208 | btrfs_abort_transaction(trans, root, ret); | |
1209 | goto fail; | |
1210 | } | |
6bdb72de | 1211 | |
f1ebcc74 LB |
1212 | /* see comments in should_cow_block() */ |
1213 | root->force_cow = 1; | |
1214 | smp_wmb(); | |
1215 | ||
6bdb72de | 1216 | btrfs_set_root_node(new_root_item, tmp); |
a22285a6 YZ |
1217 | /* record when the snapshot was created in key.offset */ |
1218 | key.offset = trans->transid; | |
1219 | ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); | |
6bdb72de SW |
1220 | btrfs_tree_unlock(tmp); |
1221 | free_extent_buffer(tmp); | |
8732d44f MX |
1222 | if (ret) { |
1223 | btrfs_abort_transaction(trans, root, ret); | |
1224 | goto fail; | |
1225 | } | |
6bdb72de | 1226 | |
a22285a6 YZ |
1227 | /* |
1228 | * insert root back/forward references | |
1229 | */ | |
1230 | ret = btrfs_add_root_ref(trans, tree_root, objectid, | |
0660b5af | 1231 | parent_root->root_key.objectid, |
33345d01 | 1232 | btrfs_ino(parent_inode), index, |
a22285a6 | 1233 | dentry->d_name.name, dentry->d_name.len); |
8732d44f MX |
1234 | if (ret) { |
1235 | btrfs_abort_transaction(trans, root, ret); | |
1236 | goto fail; | |
1237 | } | |
0660b5af | 1238 | |
a22285a6 YZ |
1239 | key.offset = (u64)-1; |
1240 | pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key); | |
79787eaa JM |
1241 | if (IS_ERR(pending->snap)) { |
1242 | ret = PTR_ERR(pending->snap); | |
8732d44f MX |
1243 | btrfs_abort_transaction(trans, root, ret); |
1244 | goto fail; | |
79787eaa | 1245 | } |
d68fc57b | 1246 | |
49b25e05 | 1247 | ret = btrfs_reloc_post_snapshot(trans, pending); |
8732d44f MX |
1248 | if (ret) { |
1249 | btrfs_abort_transaction(trans, root, ret); | |
1250 | goto fail; | |
1251 | } | |
361048f5 MX |
1252 | |
1253 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); | |
8732d44f MX |
1254 | if (ret) { |
1255 | btrfs_abort_transaction(trans, root, ret); | |
1256 | goto fail; | |
1257 | } | |
42874b3d MX |
1258 | |
1259 | ret = btrfs_insert_dir_item(trans, parent_root, | |
1260 | dentry->d_name.name, dentry->d_name.len, | |
1261 | parent_inode, &key, | |
1262 | BTRFS_FT_DIR, index); | |
1263 | /* We have check then name at the beginning, so it is impossible. */ | |
9c52057c | 1264 | BUG_ON(ret == -EEXIST || ret == -EOVERFLOW); |
8732d44f MX |
1265 | if (ret) { |
1266 | btrfs_abort_transaction(trans, root, ret); | |
1267 | goto fail; | |
1268 | } | |
42874b3d MX |
1269 | |
1270 | btrfs_i_size_write(parent_inode, parent_inode->i_size + | |
1271 | dentry->d_name.len * 2); | |
1272 | parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME; | |
be6aef60 | 1273 | ret = btrfs_update_inode_fallback(trans, parent_root, parent_inode); |
42874b3d | 1274 | if (ret) |
8732d44f | 1275 | btrfs_abort_transaction(trans, root, ret); |
3063d29f | 1276 | fail: |
6fa9700e | 1277 | dput(parent); |
98c9942a | 1278 | trans->block_rsv = rsv; |
6fa9700e MX |
1279 | no_free_objectid: |
1280 | kfree(new_root_item); | |
1281 | root_item_alloc_fail: | |
42874b3d MX |
1282 | btrfs_free_path(path); |
1283 | path_alloc_fail: | |
a22285a6 | 1284 | btrfs_block_rsv_release(root, &pending->block_rsv, (u64)-1); |
49b25e05 | 1285 | return ret; |
3063d29f CM |
1286 | } |
1287 | ||
d352ac68 CM |
1288 | /* |
1289 | * create all the snapshots we've scheduled for creation | |
1290 | */ | |
80b6794d CM |
1291 | static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans, |
1292 | struct btrfs_fs_info *fs_info) | |
3de4586c CM |
1293 | { |
1294 | struct btrfs_pending_snapshot *pending; | |
1295 | struct list_head *head = &trans->transaction->pending_snapshots; | |
3de4586c | 1296 | |
fe66a05a CM |
1297 | list_for_each_entry(pending, head, list) |
1298 | create_pending_snapshot(trans, fs_info, pending); | |
3de4586c CM |
1299 | return 0; |
1300 | } | |
1301 | ||
5d4f98a2 YZ |
1302 | static void update_super_roots(struct btrfs_root *root) |
1303 | { | |
1304 | struct btrfs_root_item *root_item; | |
1305 | struct btrfs_super_block *super; | |
1306 | ||
6c41761f | 1307 | super = root->fs_info->super_copy; |
5d4f98a2 YZ |
1308 | |
1309 | root_item = &root->fs_info->chunk_root->root_item; | |
1310 | super->chunk_root = root_item->bytenr; | |
1311 | super->chunk_root_generation = root_item->generation; | |
1312 | super->chunk_root_level = root_item->level; | |
1313 | ||
1314 | root_item = &root->fs_info->tree_root->root_item; | |
1315 | super->root = root_item->bytenr; | |
1316 | super->generation = root_item->generation; | |
1317 | super->root_level = root_item->level; | |
73bc1876 | 1318 | if (btrfs_test_opt(root, SPACE_CACHE)) |
0af3d00b | 1319 | super->cache_generation = root_item->generation; |
5d4f98a2 YZ |
1320 | } |
1321 | ||
f36f3042 CM |
1322 | int btrfs_transaction_in_commit(struct btrfs_fs_info *info) |
1323 | { | |
1324 | int ret = 0; | |
a4abeea4 | 1325 | spin_lock(&info->trans_lock); |
f36f3042 CM |
1326 | if (info->running_transaction) |
1327 | ret = info->running_transaction->in_commit; | |
a4abeea4 | 1328 | spin_unlock(&info->trans_lock); |
f36f3042 CM |
1329 | return ret; |
1330 | } | |
1331 | ||
8929ecfa YZ |
1332 | int btrfs_transaction_blocked(struct btrfs_fs_info *info) |
1333 | { | |
1334 | int ret = 0; | |
a4abeea4 | 1335 | spin_lock(&info->trans_lock); |
8929ecfa YZ |
1336 | if (info->running_transaction) |
1337 | ret = info->running_transaction->blocked; | |
a4abeea4 | 1338 | spin_unlock(&info->trans_lock); |
8929ecfa YZ |
1339 | return ret; |
1340 | } | |
1341 | ||
bb9c12c9 SW |
1342 | /* |
1343 | * wait for the current transaction commit to start and block subsequent | |
1344 | * transaction joins | |
1345 | */ | |
1346 | static void wait_current_trans_commit_start(struct btrfs_root *root, | |
1347 | struct btrfs_transaction *trans) | |
1348 | { | |
72d63ed6 | 1349 | wait_event(root->fs_info->transaction_blocked_wait, trans->in_commit); |
bb9c12c9 SW |
1350 | } |
1351 | ||
1352 | /* | |
1353 | * wait for the current transaction to start and then become unblocked. | |
1354 | * caller holds ref. | |
1355 | */ | |
1356 | static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root, | |
1357 | struct btrfs_transaction *trans) | |
1358 | { | |
72d63ed6 LZ |
1359 | wait_event(root->fs_info->transaction_wait, |
1360 | trans->commit_done || (trans->in_commit && !trans->blocked)); | |
bb9c12c9 SW |
1361 | } |
1362 | ||
1363 | /* | |
1364 | * commit transactions asynchronously. once btrfs_commit_transaction_async | |
1365 | * returns, any subsequent transaction will not be allowed to join. | |
1366 | */ | |
1367 | struct btrfs_async_commit { | |
1368 | struct btrfs_trans_handle *newtrans; | |
1369 | struct btrfs_root *root; | |
7892b5af | 1370 | struct work_struct work; |
bb9c12c9 SW |
1371 | }; |
1372 | ||
1373 | static void do_async_commit(struct work_struct *work) | |
1374 | { | |
1375 | struct btrfs_async_commit *ac = | |
7892b5af | 1376 | container_of(work, struct btrfs_async_commit, work); |
bb9c12c9 | 1377 | |
6fc4e354 SW |
1378 | /* |
1379 | * We've got freeze protection passed with the transaction. | |
1380 | * Tell lockdep about it. | |
1381 | */ | |
ff7c1d33 MX |
1382 | if (ac->newtrans->type < TRANS_JOIN_NOLOCK) |
1383 | rwsem_acquire_read( | |
1384 | &ac->root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1], | |
1385 | 0, 1, _THIS_IP_); | |
6fc4e354 | 1386 | |
e209db7a SW |
1387 | current->journal_info = ac->newtrans; |
1388 | ||
bb9c12c9 SW |
1389 | btrfs_commit_transaction(ac->newtrans, ac->root); |
1390 | kfree(ac); | |
1391 | } | |
1392 | ||
1393 | int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans, | |
1394 | struct btrfs_root *root, | |
1395 | int wait_for_unblock) | |
1396 | { | |
1397 | struct btrfs_async_commit *ac; | |
1398 | struct btrfs_transaction *cur_trans; | |
1399 | ||
1400 | ac = kmalloc(sizeof(*ac), GFP_NOFS); | |
db5b493a TI |
1401 | if (!ac) |
1402 | return -ENOMEM; | |
bb9c12c9 | 1403 | |
7892b5af | 1404 | INIT_WORK(&ac->work, do_async_commit); |
bb9c12c9 | 1405 | ac->root = root; |
7a7eaa40 | 1406 | ac->newtrans = btrfs_join_transaction(root); |
3612b495 TI |
1407 | if (IS_ERR(ac->newtrans)) { |
1408 | int err = PTR_ERR(ac->newtrans); | |
1409 | kfree(ac); | |
1410 | return err; | |
1411 | } | |
bb9c12c9 SW |
1412 | |
1413 | /* take transaction reference */ | |
bb9c12c9 | 1414 | cur_trans = trans->transaction; |
13c5a93e | 1415 | atomic_inc(&cur_trans->use_count); |
bb9c12c9 SW |
1416 | |
1417 | btrfs_end_transaction(trans, root); | |
6fc4e354 SW |
1418 | |
1419 | /* | |
1420 | * Tell lockdep we've released the freeze rwsem, since the | |
1421 | * async commit thread will be the one to unlock it. | |
1422 | */ | |
ff7c1d33 MX |
1423 | if (trans->type < TRANS_JOIN_NOLOCK) |
1424 | rwsem_release( | |
1425 | &root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1], | |
1426 | 1, _THIS_IP_); | |
6fc4e354 | 1427 | |
7892b5af | 1428 | schedule_work(&ac->work); |
bb9c12c9 SW |
1429 | |
1430 | /* wait for transaction to start and unblock */ | |
bb9c12c9 SW |
1431 | if (wait_for_unblock) |
1432 | wait_current_trans_commit_start_and_unblock(root, cur_trans); | |
1433 | else | |
1434 | wait_current_trans_commit_start(root, cur_trans); | |
bb9c12c9 | 1435 | |
38e88054 SW |
1436 | if (current->journal_info == trans) |
1437 | current->journal_info = NULL; | |
1438 | ||
1439 | put_transaction(cur_trans); | |
bb9c12c9 SW |
1440 | return 0; |
1441 | } | |
1442 | ||
49b25e05 JM |
1443 | |
1444 | static void cleanup_transaction(struct btrfs_trans_handle *trans, | |
7b8b92af | 1445 | struct btrfs_root *root, int err) |
49b25e05 JM |
1446 | { |
1447 | struct btrfs_transaction *cur_trans = trans->transaction; | |
1448 | ||
1449 | WARN_ON(trans->use_count > 1); | |
1450 | ||
7b8b92af JB |
1451 | btrfs_abort_transaction(trans, root, err); |
1452 | ||
49b25e05 JM |
1453 | spin_lock(&root->fs_info->trans_lock); |
1454 | list_del_init(&cur_trans->list); | |
d7096fc3 JB |
1455 | if (cur_trans == root->fs_info->running_transaction) { |
1456 | root->fs_info->running_transaction = NULL; | |
1457 | root->fs_info->trans_no_join = 0; | |
1458 | } | |
49b25e05 JM |
1459 | spin_unlock(&root->fs_info->trans_lock); |
1460 | ||
1461 | btrfs_cleanup_one_transaction(trans->transaction, root); | |
1462 | ||
1463 | put_transaction(cur_trans); | |
1464 | put_transaction(cur_trans); | |
1465 | ||
1466 | trace_btrfs_transaction_commit(root); | |
1467 | ||
1468 | btrfs_scrub_continue(root); | |
1469 | ||
1470 | if (current->journal_info == trans) | |
1471 | current->journal_info = NULL; | |
1472 | ||
1473 | kmem_cache_free(btrfs_trans_handle_cachep, trans); | |
1474 | } | |
1475 | ||
ca469637 MX |
1476 | static int btrfs_flush_all_pending_stuffs(struct btrfs_trans_handle *trans, |
1477 | struct btrfs_root *root) | |
1478 | { | |
1479 | int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT); | |
1480 | int snap_pending = 0; | |
1481 | int ret; | |
1482 | ||
1483 | if (!flush_on_commit) { | |
1484 | spin_lock(&root->fs_info->trans_lock); | |
1485 | if (!list_empty(&trans->transaction->pending_snapshots)) | |
1486 | snap_pending = 1; | |
1487 | spin_unlock(&root->fs_info->trans_lock); | |
1488 | } | |
1489 | ||
1490 | if (flush_on_commit || snap_pending) { | |
3edb2a68 MX |
1491 | ret = btrfs_start_delalloc_inodes(root, 1); |
1492 | if (ret) | |
1493 | return ret; | |
ca469637 MX |
1494 | btrfs_wait_ordered_extents(root, 1); |
1495 | } | |
1496 | ||
1497 | ret = btrfs_run_delayed_items(trans, root); | |
1498 | if (ret) | |
1499 | return ret; | |
1500 | ||
1501 | /* | |
1502 | * running the delayed items may have added new refs. account | |
1503 | * them now so that they hinder processing of more delayed refs | |
1504 | * as little as possible. | |
1505 | */ | |
1506 | btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info); | |
1507 | ||
1508 | /* | |
1509 | * rename don't use btrfs_join_transaction, so, once we | |
1510 | * set the transaction to blocked above, we aren't going | |
1511 | * to get any new ordered operations. We can safely run | |
1512 | * it here and no for sure that nothing new will be added | |
1513 | * to the list | |
1514 | */ | |
569e0f35 | 1515 | ret = btrfs_run_ordered_operations(trans, root, 1); |
ca469637 | 1516 | |
eebc6084 | 1517 | return ret; |
ca469637 MX |
1518 | } |
1519 | ||
bb9c12c9 SW |
1520 | /* |
1521 | * btrfs_transaction state sequence: | |
1522 | * in_commit = 0, blocked = 0 (initial) | |
1523 | * in_commit = 1, blocked = 1 | |
1524 | * blocked = 0 | |
1525 | * commit_done = 1 | |
1526 | */ | |
79154b1b CM |
1527 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans, |
1528 | struct btrfs_root *root) | |
1529 | { | |
15ee9bc7 | 1530 | unsigned long joined = 0; |
49b25e05 | 1531 | struct btrfs_transaction *cur_trans = trans->transaction; |
8fd17795 | 1532 | struct btrfs_transaction *prev_trans = NULL; |
79154b1b | 1533 | DEFINE_WAIT(wait); |
25287e0a | 1534 | int ret; |
89573b9c CM |
1535 | int should_grow = 0; |
1536 | unsigned long now = get_seconds(); | |
79154b1b | 1537 | |
569e0f35 | 1538 | ret = btrfs_run_ordered_operations(trans, root, 0); |
25287e0a MX |
1539 | if (ret) { |
1540 | btrfs_abort_transaction(trans, root, ret); | |
e4a2bcac JB |
1541 | btrfs_end_transaction(trans, root); |
1542 | return ret; | |
25287e0a | 1543 | } |
5a3f23d5 | 1544 | |
8d25a086 MX |
1545 | /* Stop the commit early if ->aborted is set */ |
1546 | if (unlikely(ACCESS_ONCE(cur_trans->aborted))) { | |
25287e0a | 1547 | ret = cur_trans->aborted; |
e4a2bcac JB |
1548 | btrfs_end_transaction(trans, root); |
1549 | return ret; | |
25287e0a | 1550 | } |
49b25e05 | 1551 | |
56bec294 CM |
1552 | /* make a pass through all the delayed refs we have so far |
1553 | * any runnings procs may add more while we are here | |
1554 | */ | |
1555 | ret = btrfs_run_delayed_refs(trans, root, 0); | |
e4a2bcac JB |
1556 | if (ret) { |
1557 | btrfs_end_transaction(trans, root); | |
1558 | return ret; | |
1559 | } | |
56bec294 | 1560 | |
0e721106 JB |
1561 | btrfs_trans_release_metadata(trans, root); |
1562 | trans->block_rsv = NULL; | |
272d26d0 MX |
1563 | if (trans->qgroup_reserved) { |
1564 | btrfs_qgroup_free(root, trans->qgroup_reserved); | |
1565 | trans->qgroup_reserved = 0; | |
1566 | } | |
0e721106 | 1567 | |
b7ec40d7 | 1568 | cur_trans = trans->transaction; |
49b25e05 | 1569 | |
56bec294 CM |
1570 | /* |
1571 | * set the flushing flag so procs in this transaction have to | |
1572 | * start sending their work down. | |
1573 | */ | |
b7ec40d7 | 1574 | cur_trans->delayed_refs.flushing = 1; |
56bec294 | 1575 | |
ea658bad JB |
1576 | if (!list_empty(&trans->new_bgs)) |
1577 | btrfs_create_pending_block_groups(trans, root); | |
1578 | ||
c3e69d58 | 1579 | ret = btrfs_run_delayed_refs(trans, root, 0); |
e4a2bcac JB |
1580 | if (ret) { |
1581 | btrfs_end_transaction(trans, root); | |
1582 | return ret; | |
1583 | } | |
56bec294 | 1584 | |
a4abeea4 | 1585 | spin_lock(&cur_trans->commit_lock); |
b7ec40d7 | 1586 | if (cur_trans->in_commit) { |
a4abeea4 | 1587 | spin_unlock(&cur_trans->commit_lock); |
13c5a93e | 1588 | atomic_inc(&cur_trans->use_count); |
49b25e05 | 1589 | ret = btrfs_end_transaction(trans, root); |
ccd467d6 | 1590 | |
b9c8300c | 1591 | wait_for_commit(root, cur_trans); |
15ee9bc7 | 1592 | |
79154b1b | 1593 | put_transaction(cur_trans); |
15ee9bc7 | 1594 | |
49b25e05 | 1595 | return ret; |
79154b1b | 1596 | } |
4313b399 | 1597 | |
2c90e5d6 | 1598 | trans->transaction->in_commit = 1; |
f9295749 | 1599 | trans->transaction->blocked = 1; |
a4abeea4 | 1600 | spin_unlock(&cur_trans->commit_lock); |
bb9c12c9 SW |
1601 | wake_up(&root->fs_info->transaction_blocked_wait); |
1602 | ||
a4abeea4 | 1603 | spin_lock(&root->fs_info->trans_lock); |
ccd467d6 CM |
1604 | if (cur_trans->list.prev != &root->fs_info->trans_list) { |
1605 | prev_trans = list_entry(cur_trans->list.prev, | |
1606 | struct btrfs_transaction, list); | |
1607 | if (!prev_trans->commit_done) { | |
13c5a93e | 1608 | atomic_inc(&prev_trans->use_count); |
a4abeea4 | 1609 | spin_unlock(&root->fs_info->trans_lock); |
ccd467d6 CM |
1610 | |
1611 | wait_for_commit(root, prev_trans); | |
ccd467d6 | 1612 | |
15ee9bc7 | 1613 | put_transaction(prev_trans); |
a4abeea4 JB |
1614 | } else { |
1615 | spin_unlock(&root->fs_info->trans_lock); | |
ccd467d6 | 1616 | } |
a4abeea4 JB |
1617 | } else { |
1618 | spin_unlock(&root->fs_info->trans_lock); | |
ccd467d6 | 1619 | } |
15ee9bc7 | 1620 | |
e39e64ac CM |
1621 | if (!btrfs_test_opt(root, SSD) && |
1622 | (now < cur_trans->start_time || now - cur_trans->start_time < 1)) | |
89573b9c CM |
1623 | should_grow = 1; |
1624 | ||
15ee9bc7 JB |
1625 | do { |
1626 | joined = cur_trans->num_joined; | |
7ea394f1 | 1627 | |
2c90e5d6 | 1628 | WARN_ON(cur_trans != trans->transaction); |
15ee9bc7 | 1629 | |
ca469637 | 1630 | ret = btrfs_flush_all_pending_stuffs(trans, root); |
49b25e05 JM |
1631 | if (ret) |
1632 | goto cleanup_transaction; | |
16cdcec7 | 1633 | |
ed3b3d31 CM |
1634 | prepare_to_wait(&cur_trans->writer_wait, &wait, |
1635 | TASK_UNINTERRUPTIBLE); | |
1636 | ||
13c5a93e | 1637 | if (atomic_read(&cur_trans->num_writers) > 1) |
99d16cbc SW |
1638 | schedule_timeout(MAX_SCHEDULE_TIMEOUT); |
1639 | else if (should_grow) | |
1640 | schedule_timeout(1); | |
15ee9bc7 | 1641 | |
15ee9bc7 | 1642 | finish_wait(&cur_trans->writer_wait, &wait); |
13c5a93e | 1643 | } while (atomic_read(&cur_trans->num_writers) > 1 || |
89573b9c | 1644 | (should_grow && cur_trans->num_joined != joined)); |
15ee9bc7 | 1645 | |
ca469637 MX |
1646 | ret = btrfs_flush_all_pending_stuffs(trans, root); |
1647 | if (ret) | |
1648 | goto cleanup_transaction; | |
1649 | ||
ed0ca140 JB |
1650 | /* |
1651 | * Ok now we need to make sure to block out any other joins while we | |
1652 | * commit the transaction. We could have started a join before setting | |
1653 | * no_join so make sure to wait for num_writers to == 1 again. | |
1654 | */ | |
1655 | spin_lock(&root->fs_info->trans_lock); | |
1656 | root->fs_info->trans_no_join = 1; | |
1657 | spin_unlock(&root->fs_info->trans_lock); | |
1658 | wait_event(cur_trans->writer_wait, | |
1659 | atomic_read(&cur_trans->num_writers) == 1); | |
1660 | ||
2cba30f1 MX |
1661 | /* ->aborted might be set after the previous check, so check it */ |
1662 | if (unlikely(ACCESS_ONCE(cur_trans->aborted))) { | |
1663 | ret = cur_trans->aborted; | |
1664 | goto cleanup_transaction; | |
1665 | } | |
7585717f CM |
1666 | /* |
1667 | * the reloc mutex makes sure that we stop | |
1668 | * the balancing code from coming in and moving | |
1669 | * extents around in the middle of the commit | |
1670 | */ | |
1671 | mutex_lock(&root->fs_info->reloc_mutex); | |
1672 | ||
42874b3d MX |
1673 | /* |
1674 | * We needn't worry about the delayed items because we will | |
1675 | * deal with them in create_pending_snapshot(), which is the | |
1676 | * core function of the snapshot creation. | |
1677 | */ | |
1678 | ret = create_pending_snapshots(trans, root->fs_info); | |
49b25e05 JM |
1679 | if (ret) { |
1680 | mutex_unlock(&root->fs_info->reloc_mutex); | |
1681 | goto cleanup_transaction; | |
1682 | } | |
3063d29f | 1683 | |
42874b3d MX |
1684 | /* |
1685 | * We insert the dir indexes of the snapshots and update the inode | |
1686 | * of the snapshots' parents after the snapshot creation, so there | |
1687 | * are some delayed items which are not dealt with. Now deal with | |
1688 | * them. | |
1689 | * | |
1690 | * We needn't worry that this operation will corrupt the snapshots, | |
1691 | * because all the tree which are snapshoted will be forced to COW | |
1692 | * the nodes and leaves. | |
1693 | */ | |
1694 | ret = btrfs_run_delayed_items(trans, root); | |
49b25e05 JM |
1695 | if (ret) { |
1696 | mutex_unlock(&root->fs_info->reloc_mutex); | |
1697 | goto cleanup_transaction; | |
1698 | } | |
16cdcec7 | 1699 | |
56bec294 | 1700 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
49b25e05 JM |
1701 | if (ret) { |
1702 | mutex_unlock(&root->fs_info->reloc_mutex); | |
1703 | goto cleanup_transaction; | |
1704 | } | |
56bec294 | 1705 | |
e999376f CM |
1706 | /* |
1707 | * make sure none of the code above managed to slip in a | |
1708 | * delayed item | |
1709 | */ | |
1710 | btrfs_assert_delayed_root_empty(root); | |
1711 | ||
2c90e5d6 | 1712 | WARN_ON(cur_trans != trans->transaction); |
dc17ff8f | 1713 | |
a2de733c | 1714 | btrfs_scrub_pause(root); |
e02119d5 CM |
1715 | /* btrfs_commit_tree_roots is responsible for getting the |
1716 | * various roots consistent with each other. Every pointer | |
1717 | * in the tree of tree roots has to point to the most up to date | |
1718 | * root for every subvolume and other tree. So, we have to keep | |
1719 | * the tree logging code from jumping in and changing any | |
1720 | * of the trees. | |
1721 | * | |
1722 | * At this point in the commit, there can't be any tree-log | |
1723 | * writers, but a little lower down we drop the trans mutex | |
1724 | * and let new people in. By holding the tree_log_mutex | |
1725 | * from now until after the super is written, we avoid races | |
1726 | * with the tree-log code. | |
1727 | */ | |
1728 | mutex_lock(&root->fs_info->tree_log_mutex); | |
1729 | ||
5d4f98a2 | 1730 | ret = commit_fs_roots(trans, root); |
49b25e05 JM |
1731 | if (ret) { |
1732 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
871383be | 1733 | mutex_unlock(&root->fs_info->reloc_mutex); |
49b25e05 JM |
1734 | goto cleanup_transaction; |
1735 | } | |
54aa1f4d | 1736 | |
5d4f98a2 | 1737 | /* commit_fs_roots gets rid of all the tree log roots, it is now |
e02119d5 CM |
1738 | * safe to free the root of tree log roots |
1739 | */ | |
1740 | btrfs_free_log_root_tree(trans, root->fs_info); | |
1741 | ||
5d4f98a2 | 1742 | ret = commit_cowonly_roots(trans, root); |
49b25e05 JM |
1743 | if (ret) { |
1744 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
871383be | 1745 | mutex_unlock(&root->fs_info->reloc_mutex); |
49b25e05 JM |
1746 | goto cleanup_transaction; |
1747 | } | |
54aa1f4d | 1748 | |
2cba30f1 MX |
1749 | /* |
1750 | * The tasks which save the space cache and inode cache may also | |
1751 | * update ->aborted, check it. | |
1752 | */ | |
1753 | if (unlikely(ACCESS_ONCE(cur_trans->aborted))) { | |
1754 | ret = cur_trans->aborted; | |
1755 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1756 | mutex_unlock(&root->fs_info->reloc_mutex); | |
1757 | goto cleanup_transaction; | |
1758 | } | |
1759 | ||
11833d66 YZ |
1760 | btrfs_prepare_extent_commit(trans, root); |
1761 | ||
78fae27e | 1762 | cur_trans = root->fs_info->running_transaction; |
5d4f98a2 YZ |
1763 | |
1764 | btrfs_set_root_node(&root->fs_info->tree_root->root_item, | |
1765 | root->fs_info->tree_root->node); | |
817d52f8 | 1766 | switch_commit_root(root->fs_info->tree_root); |
5d4f98a2 YZ |
1767 | |
1768 | btrfs_set_root_node(&root->fs_info->chunk_root->root_item, | |
1769 | root->fs_info->chunk_root->node); | |
817d52f8 | 1770 | switch_commit_root(root->fs_info->chunk_root); |
5d4f98a2 | 1771 | |
edf39272 | 1772 | assert_qgroups_uptodate(trans); |
5d4f98a2 | 1773 | update_super_roots(root); |
e02119d5 CM |
1774 | |
1775 | if (!root->fs_info->log_root_recovering) { | |
6c41761f DS |
1776 | btrfs_set_super_log_root(root->fs_info->super_copy, 0); |
1777 | btrfs_set_super_log_root_level(root->fs_info->super_copy, 0); | |
e02119d5 CM |
1778 | } |
1779 | ||
6c41761f DS |
1780 | memcpy(root->fs_info->super_for_commit, root->fs_info->super_copy, |
1781 | sizeof(*root->fs_info->super_copy)); | |
ccd467d6 | 1782 | |
f9295749 | 1783 | trans->transaction->blocked = 0; |
a4abeea4 JB |
1784 | spin_lock(&root->fs_info->trans_lock); |
1785 | root->fs_info->running_transaction = NULL; | |
1786 | root->fs_info->trans_no_join = 0; | |
1787 | spin_unlock(&root->fs_info->trans_lock); | |
7585717f | 1788 | mutex_unlock(&root->fs_info->reloc_mutex); |
b7ec40d7 | 1789 | |
f9295749 | 1790 | wake_up(&root->fs_info->transaction_wait); |
e6dcd2dc | 1791 | |
79154b1b | 1792 | ret = btrfs_write_and_wait_transaction(trans, root); |
49b25e05 JM |
1793 | if (ret) { |
1794 | btrfs_error(root->fs_info, ret, | |
1795 | "Error while writing out transaction."); | |
1796 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1797 | goto cleanup_transaction; | |
1798 | } | |
1799 | ||
1800 | ret = write_ctree_super(trans, root, 0); | |
1801 | if (ret) { | |
1802 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1803 | goto cleanup_transaction; | |
1804 | } | |
4313b399 | 1805 | |
e02119d5 CM |
1806 | /* |
1807 | * the super is written, we can safely allow the tree-loggers | |
1808 | * to go about their business | |
1809 | */ | |
1810 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1811 | ||
11833d66 | 1812 | btrfs_finish_extent_commit(trans, root); |
4313b399 | 1813 | |
2c90e5d6 | 1814 | cur_trans->commit_done = 1; |
b7ec40d7 | 1815 | |
15ee9bc7 | 1816 | root->fs_info->last_trans_committed = cur_trans->transid; |
817d52f8 | 1817 | |
2c90e5d6 | 1818 | wake_up(&cur_trans->commit_wait); |
3de4586c | 1819 | |
a4abeea4 | 1820 | spin_lock(&root->fs_info->trans_lock); |
13c5a93e | 1821 | list_del_init(&cur_trans->list); |
a4abeea4 JB |
1822 | spin_unlock(&root->fs_info->trans_lock); |
1823 | ||
78fae27e | 1824 | put_transaction(cur_trans); |
79154b1b | 1825 | put_transaction(cur_trans); |
58176a96 | 1826 | |
354aa0fb MX |
1827 | if (trans->type < TRANS_JOIN_NOLOCK) |
1828 | sb_end_intwrite(root->fs_info->sb); | |
b2b5ef5c | 1829 | |
1abe9b8a | 1830 | trace_btrfs_transaction_commit(root); |
1831 | ||
a2de733c AJ |
1832 | btrfs_scrub_continue(root); |
1833 | ||
9ed74f2d JB |
1834 | if (current->journal_info == trans) |
1835 | current->journal_info = NULL; | |
1836 | ||
2c90e5d6 | 1837 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
24bbcf04 YZ |
1838 | |
1839 | if (current != root->fs_info->transaction_kthread) | |
1840 | btrfs_run_delayed_iputs(root); | |
1841 | ||
79154b1b | 1842 | return ret; |
49b25e05 JM |
1843 | |
1844 | cleanup_transaction: | |
0e721106 JB |
1845 | btrfs_trans_release_metadata(trans, root); |
1846 | trans->block_rsv = NULL; | |
272d26d0 MX |
1847 | if (trans->qgroup_reserved) { |
1848 | btrfs_qgroup_free(root, trans->qgroup_reserved); | |
1849 | trans->qgroup_reserved = 0; | |
1850 | } | |
49b25e05 JM |
1851 | btrfs_printk(root->fs_info, "Skipping commit of aborted transaction.\n"); |
1852 | // WARN_ON(1); | |
1853 | if (current->journal_info == trans) | |
1854 | current->journal_info = NULL; | |
7b8b92af | 1855 | cleanup_transaction(trans, root, ret); |
49b25e05 JM |
1856 | |
1857 | return ret; | |
79154b1b CM |
1858 | } |
1859 | ||
d352ac68 CM |
1860 | /* |
1861 | * interface function to delete all the snapshots we have scheduled for deletion | |
1862 | */ | |
e9d0b13b CM |
1863 | int btrfs_clean_old_snapshots(struct btrfs_root *root) |
1864 | { | |
5d4f98a2 YZ |
1865 | LIST_HEAD(list); |
1866 | struct btrfs_fs_info *fs_info = root->fs_info; | |
1867 | ||
a4abeea4 | 1868 | spin_lock(&fs_info->trans_lock); |
5d4f98a2 | 1869 | list_splice_init(&fs_info->dead_roots, &list); |
a4abeea4 | 1870 | spin_unlock(&fs_info->trans_lock); |
e9d0b13b | 1871 | |
5d4f98a2 | 1872 | while (!list_empty(&list)) { |
2c536799 JM |
1873 | int ret; |
1874 | ||
5d4f98a2 | 1875 | root = list_entry(list.next, struct btrfs_root, root_list); |
76dda93c YZ |
1876 | list_del(&root->root_list); |
1877 | ||
16cdcec7 MX |
1878 | btrfs_kill_all_delayed_nodes(root); |
1879 | ||
76dda93c YZ |
1880 | if (btrfs_header_backref_rev(root->node) < |
1881 | BTRFS_MIXED_BACKREF_REV) | |
2c536799 | 1882 | ret = btrfs_drop_snapshot(root, NULL, 0, 0); |
76dda93c | 1883 | else |
2c536799 JM |
1884 | ret =btrfs_drop_snapshot(root, NULL, 1, 0); |
1885 | BUG_ON(ret < 0); | |
e9d0b13b CM |
1886 | } |
1887 | return 0; | |
1888 | } |