]>
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 CM |
43 | memset(transaction, 0, sizeof(*transaction)); |
44 | kmem_cache_free(btrfs_transaction_cachep, transaction); | |
78fae27e | 45 | } |
79154b1b CM |
46 | } |
47 | ||
817d52f8 JB |
48 | static noinline void switch_commit_root(struct btrfs_root *root) |
49 | { | |
817d52f8 JB |
50 | free_extent_buffer(root->commit_root); |
51 | root->commit_root = btrfs_root_node(root); | |
817d52f8 JB |
52 | } |
53 | ||
d352ac68 CM |
54 | /* |
55 | * either allocate a new transaction or hop into the existing one | |
56 | */ | |
354aa0fb | 57 | static noinline int join_transaction(struct btrfs_root *root, int type) |
79154b1b CM |
58 | { |
59 | struct btrfs_transaction *cur_trans; | |
19ae4e81 | 60 | struct btrfs_fs_info *fs_info = root->fs_info; |
a4abeea4 | 61 | |
19ae4e81 | 62 | spin_lock(&fs_info->trans_lock); |
d43317dc | 63 | loop: |
49b25e05 | 64 | /* The file system has been taken offline. No new transactions. */ |
19ae4e81 JS |
65 | if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) { |
66 | spin_unlock(&fs_info->trans_lock); | |
49b25e05 JM |
67 | return -EROFS; |
68 | } | |
69 | ||
19ae4e81 | 70 | if (fs_info->trans_no_join) { |
354aa0fb MX |
71 | /* |
72 | * If we are JOIN_NOLOCK we're already committing a current | |
73 | * transaction, we just need a handle to deal with something | |
74 | * when committing the transaction, such as inode cache and | |
75 | * space cache. It is a special case. | |
76 | */ | |
77 | if (type != TRANS_JOIN_NOLOCK) { | |
19ae4e81 | 78 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 JB |
79 | return -EBUSY; |
80 | } | |
81 | } | |
82 | ||
19ae4e81 | 83 | cur_trans = fs_info->running_transaction; |
a4abeea4 | 84 | if (cur_trans) { |
871383be | 85 | if (cur_trans->aborted) { |
19ae4e81 | 86 | spin_unlock(&fs_info->trans_lock); |
49b25e05 | 87 | return cur_trans->aborted; |
871383be | 88 | } |
a4abeea4 | 89 | atomic_inc(&cur_trans->use_count); |
13c5a93e | 90 | atomic_inc(&cur_trans->num_writers); |
15ee9bc7 | 91 | cur_trans->num_joined++; |
19ae4e81 | 92 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 93 | return 0; |
79154b1b | 94 | } |
19ae4e81 | 95 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 96 | |
354aa0fb MX |
97 | /* |
98 | * If we are ATTACH, we just want to catch the current transaction, | |
99 | * and commit it. If there is no transaction, just return ENOENT. | |
100 | */ | |
101 | if (type == TRANS_ATTACH) | |
102 | return -ENOENT; | |
103 | ||
a4abeea4 JB |
104 | cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, GFP_NOFS); |
105 | if (!cur_trans) | |
106 | return -ENOMEM; | |
d43317dc | 107 | |
19ae4e81 JS |
108 | spin_lock(&fs_info->trans_lock); |
109 | if (fs_info->running_transaction) { | |
d43317dc CM |
110 | /* |
111 | * someone started a transaction after we unlocked. Make sure | |
112 | * to redo the trans_no_join checks above | |
113 | */ | |
a4abeea4 | 114 | kmem_cache_free(btrfs_transaction_cachep, cur_trans); |
19ae4e81 | 115 | cur_trans = fs_info->running_transaction; |
d43317dc | 116 | goto loop; |
e4b50e14 DC |
117 | } else if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) { |
118 | spin_unlock(&fs_info->trans_lock); | |
7b8b92af JB |
119 | kmem_cache_free(btrfs_transaction_cachep, cur_trans); |
120 | return -EROFS; | |
79154b1b | 121 | } |
d43317dc | 122 | |
a4abeea4 JB |
123 | atomic_set(&cur_trans->num_writers, 1); |
124 | cur_trans->num_joined = 0; | |
125 | init_waitqueue_head(&cur_trans->writer_wait); | |
126 | init_waitqueue_head(&cur_trans->commit_wait); | |
127 | cur_trans->in_commit = 0; | |
128 | cur_trans->blocked = 0; | |
129 | /* | |
130 | * One for this trans handle, one so it will live on until we | |
131 | * commit the transaction. | |
132 | */ | |
133 | atomic_set(&cur_trans->use_count, 2); | |
134 | cur_trans->commit_done = 0; | |
135 | cur_trans->start_time = get_seconds(); | |
136 | ||
137 | cur_trans->delayed_refs.root = RB_ROOT; | |
138 | cur_trans->delayed_refs.num_entries = 0; | |
139 | cur_trans->delayed_refs.num_heads_ready = 0; | |
140 | cur_trans->delayed_refs.num_heads = 0; | |
141 | cur_trans->delayed_refs.flushing = 0; | |
142 | cur_trans->delayed_refs.run_delayed_start = 0; | |
20b297d6 JS |
143 | |
144 | /* | |
145 | * although the tree mod log is per file system and not per transaction, | |
146 | * the log must never go across transaction boundaries. | |
147 | */ | |
148 | smp_mb(); | |
31b1a2bd JL |
149 | if (!list_empty(&fs_info->tree_mod_seq_list)) |
150 | WARN(1, KERN_ERR "btrfs: tree_mod_seq_list not empty when " | |
20b297d6 | 151 | "creating a fresh transaction\n"); |
31b1a2bd JL |
152 | if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log)) |
153 | WARN(1, KERN_ERR "btrfs: tree_mod_log rb tree not empty when " | |
20b297d6 | 154 | "creating a fresh transaction\n"); |
20b297d6 JS |
155 | atomic_set(&fs_info->tree_mod_seq, 0); |
156 | ||
a4abeea4 JB |
157 | spin_lock_init(&cur_trans->commit_lock); |
158 | spin_lock_init(&cur_trans->delayed_refs.lock); | |
159 | ||
160 | INIT_LIST_HEAD(&cur_trans->pending_snapshots); | |
19ae4e81 | 161 | list_add_tail(&cur_trans->list, &fs_info->trans_list); |
a4abeea4 | 162 | extent_io_tree_init(&cur_trans->dirty_pages, |
19ae4e81 JS |
163 | fs_info->btree_inode->i_mapping); |
164 | fs_info->generation++; | |
165 | cur_trans->transid = fs_info->generation; | |
166 | fs_info->running_transaction = cur_trans; | |
49b25e05 | 167 | cur_trans->aborted = 0; |
19ae4e81 | 168 | spin_unlock(&fs_info->trans_lock); |
15ee9bc7 | 169 | |
79154b1b CM |
170 | return 0; |
171 | } | |
172 | ||
d352ac68 | 173 | /* |
d397712b CM |
174 | * this does all the record keeping required to make sure that a reference |
175 | * counted root is properly recorded in a given transaction. This is required | |
176 | * to make sure the old root from before we joined the transaction is deleted | |
177 | * when the transaction commits | |
d352ac68 | 178 | */ |
7585717f | 179 | static int record_root_in_trans(struct btrfs_trans_handle *trans, |
a4abeea4 | 180 | struct btrfs_root *root) |
6702ed49 | 181 | { |
5d4f98a2 | 182 | if (root->ref_cows && root->last_trans < trans->transid) { |
6702ed49 | 183 | WARN_ON(root == root->fs_info->extent_root); |
5d4f98a2 YZ |
184 | WARN_ON(root->commit_root != root->node); |
185 | ||
7585717f CM |
186 | /* |
187 | * see below for in_trans_setup usage rules | |
188 | * we have the reloc mutex held now, so there | |
189 | * is only one writer in this function | |
190 | */ | |
191 | root->in_trans_setup = 1; | |
192 | ||
193 | /* make sure readers find in_trans_setup before | |
194 | * they find our root->last_trans update | |
195 | */ | |
196 | smp_wmb(); | |
197 | ||
a4abeea4 JB |
198 | spin_lock(&root->fs_info->fs_roots_radix_lock); |
199 | if (root->last_trans == trans->transid) { | |
200 | spin_unlock(&root->fs_info->fs_roots_radix_lock); | |
201 | return 0; | |
202 | } | |
5d4f98a2 YZ |
203 | radix_tree_tag_set(&root->fs_info->fs_roots_radix, |
204 | (unsigned long)root->root_key.objectid, | |
205 | BTRFS_ROOT_TRANS_TAG); | |
a4abeea4 | 206 | spin_unlock(&root->fs_info->fs_roots_radix_lock); |
7585717f CM |
207 | root->last_trans = trans->transid; |
208 | ||
209 | /* this is pretty tricky. We don't want to | |
210 | * take the relocation lock in btrfs_record_root_in_trans | |
211 | * unless we're really doing the first setup for this root in | |
212 | * this transaction. | |
213 | * | |
214 | * Normally we'd use root->last_trans as a flag to decide | |
215 | * if we want to take the expensive mutex. | |
216 | * | |
217 | * But, we have to set root->last_trans before we | |
218 | * init the relocation root, otherwise, we trip over warnings | |
219 | * in ctree.c. The solution used here is to flag ourselves | |
220 | * with root->in_trans_setup. When this is 1, we're still | |
221 | * fixing up the reloc trees and everyone must wait. | |
222 | * | |
223 | * When this is zero, they can trust root->last_trans and fly | |
224 | * through btrfs_record_root_in_trans without having to take the | |
225 | * lock. smp_wmb() makes sure that all the writes above are | |
226 | * done before we pop in the zero below | |
227 | */ | |
5d4f98a2 | 228 | btrfs_init_reloc_root(trans, root); |
7585717f CM |
229 | smp_wmb(); |
230 | root->in_trans_setup = 0; | |
5d4f98a2 YZ |
231 | } |
232 | return 0; | |
233 | } | |
bcc63abb | 234 | |
7585717f CM |
235 | |
236 | int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, | |
237 | struct btrfs_root *root) | |
238 | { | |
239 | if (!root->ref_cows) | |
240 | return 0; | |
241 | ||
242 | /* | |
243 | * see record_root_in_trans for comments about in_trans_setup usage | |
244 | * and barriers | |
245 | */ | |
246 | smp_rmb(); | |
247 | if (root->last_trans == trans->transid && | |
248 | !root->in_trans_setup) | |
249 | return 0; | |
250 | ||
251 | mutex_lock(&root->fs_info->reloc_mutex); | |
252 | record_root_in_trans(trans, root); | |
253 | mutex_unlock(&root->fs_info->reloc_mutex); | |
254 | ||
255 | return 0; | |
256 | } | |
257 | ||
d352ac68 CM |
258 | /* wait for commit against the current transaction to become unblocked |
259 | * when this is done, it is safe to start a new transaction, but the current | |
260 | * transaction might not be fully on disk. | |
261 | */ | |
37d1aeee | 262 | static void wait_current_trans(struct btrfs_root *root) |
79154b1b | 263 | { |
f9295749 | 264 | struct btrfs_transaction *cur_trans; |
79154b1b | 265 | |
a4abeea4 | 266 | spin_lock(&root->fs_info->trans_lock); |
f9295749 | 267 | cur_trans = root->fs_info->running_transaction; |
37d1aeee | 268 | if (cur_trans && cur_trans->blocked) { |
13c5a93e | 269 | atomic_inc(&cur_trans->use_count); |
a4abeea4 | 270 | spin_unlock(&root->fs_info->trans_lock); |
72d63ed6 LZ |
271 | |
272 | wait_event(root->fs_info->transaction_wait, | |
273 | !cur_trans->blocked); | |
f9295749 | 274 | put_transaction(cur_trans); |
a4abeea4 JB |
275 | } else { |
276 | spin_unlock(&root->fs_info->trans_lock); | |
f9295749 | 277 | } |
37d1aeee CM |
278 | } |
279 | ||
a22285a6 YZ |
280 | static int may_wait_transaction(struct btrfs_root *root, int type) |
281 | { | |
a4abeea4 JB |
282 | if (root->fs_info->log_root_recovering) |
283 | return 0; | |
284 | ||
285 | if (type == TRANS_USERSPACE) | |
286 | return 1; | |
287 | ||
288 | if (type == TRANS_START && | |
289 | !atomic_read(&root->fs_info->open_ioctl_trans)) | |
a22285a6 | 290 | return 1; |
a4abeea4 | 291 | |
a22285a6 YZ |
292 | return 0; |
293 | } | |
294 | ||
08e007d2 MX |
295 | static struct btrfs_trans_handle * |
296 | start_transaction(struct btrfs_root *root, u64 num_items, int type, | |
297 | enum btrfs_reserve_flush_enum flush) | |
37d1aeee | 298 | { |
a22285a6 YZ |
299 | struct btrfs_trans_handle *h; |
300 | struct btrfs_transaction *cur_trans; | |
b5009945 | 301 | u64 num_bytes = 0; |
37d1aeee | 302 | int ret; |
c5567237 | 303 | u64 qgroup_reserved = 0; |
acce952b | 304 | |
305 | if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) | |
306 | return ERR_PTR(-EROFS); | |
2a1eb461 JB |
307 | |
308 | if (current->journal_info) { | |
309 | WARN_ON(type != TRANS_JOIN && type != TRANS_JOIN_NOLOCK); | |
310 | h = current->journal_info; | |
311 | h->use_count++; | |
b7d5b0a8 | 312 | WARN_ON(h->use_count > 2); |
2a1eb461 JB |
313 | h->orig_rsv = h->block_rsv; |
314 | h->block_rsv = NULL; | |
315 | goto got_it; | |
316 | } | |
b5009945 JB |
317 | |
318 | /* | |
319 | * Do the reservation before we join the transaction so we can do all | |
320 | * the appropriate flushing if need be. | |
321 | */ | |
322 | if (num_items > 0 && root != root->fs_info->chunk_root) { | |
c5567237 AJ |
323 | if (root->fs_info->quota_enabled && |
324 | is_fstree(root->root_key.objectid)) { | |
325 | qgroup_reserved = num_items * root->leafsize; | |
326 | ret = btrfs_qgroup_reserve(root, qgroup_reserved); | |
327 | if (ret) | |
328 | return ERR_PTR(ret); | |
329 | } | |
330 | ||
b5009945 | 331 | num_bytes = btrfs_calc_trans_metadata_size(root, num_items); |
08e007d2 MX |
332 | ret = btrfs_block_rsv_add(root, |
333 | &root->fs_info->trans_block_rsv, | |
334 | num_bytes, flush); | |
b5009945 JB |
335 | if (ret) |
336 | return ERR_PTR(ret); | |
337 | } | |
a22285a6 YZ |
338 | again: |
339 | h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS); | |
340 | if (!h) | |
341 | return ERR_PTR(-ENOMEM); | |
37d1aeee | 342 | |
98114659 JB |
343 | /* |
344 | * If we are JOIN_NOLOCK we're already committing a transaction and | |
345 | * waiting on this guy, so we don't need to do the sb_start_intwrite | |
346 | * because we're already holding a ref. We need this because we could | |
347 | * have raced in and did an fsync() on a file which can kick a commit | |
348 | * and then we deadlock with somebody doing a freeze. | |
354aa0fb MX |
349 | * |
350 | * If we are ATTACH, it means we just want to catch the current | |
351 | * transaction and commit it, so we needn't do sb_start_intwrite(). | |
98114659 | 352 | */ |
354aa0fb | 353 | if (type < TRANS_JOIN_NOLOCK) |
60376ce4 | 354 | sb_start_intwrite(root->fs_info->sb); |
b2b5ef5c | 355 | |
a22285a6 | 356 | if (may_wait_transaction(root, type)) |
37d1aeee | 357 | wait_current_trans(root); |
a22285a6 | 358 | |
a4abeea4 | 359 | do { |
354aa0fb | 360 | ret = join_transaction(root, type); |
a4abeea4 JB |
361 | if (ret == -EBUSY) |
362 | wait_current_trans(root); | |
363 | } while (ret == -EBUSY); | |
364 | ||
db5b493a | 365 | if (ret < 0) { |
354aa0fb MX |
366 | /* We must get the transaction if we are JOIN_NOLOCK. */ |
367 | BUG_ON(type == TRANS_JOIN_NOLOCK); | |
368 | ||
369 | if (type < TRANS_JOIN_NOLOCK) | |
370 | sb_end_intwrite(root->fs_info->sb); | |
6e8df2ae | 371 | kmem_cache_free(btrfs_trans_handle_cachep, h); |
db5b493a TI |
372 | return ERR_PTR(ret); |
373 | } | |
0f7d52f4 | 374 | |
a22285a6 | 375 | cur_trans = root->fs_info->running_transaction; |
a22285a6 YZ |
376 | |
377 | h->transid = cur_trans->transid; | |
378 | h->transaction = cur_trans; | |
79154b1b | 379 | h->blocks_used = 0; |
a22285a6 | 380 | h->bytes_reserved = 0; |
d13603ef | 381 | h->root = root; |
56bec294 | 382 | h->delayed_ref_updates = 0; |
2a1eb461 | 383 | h->use_count = 1; |
0e721106 | 384 | h->adding_csums = 0; |
f0486c68 | 385 | h->block_rsv = NULL; |
2a1eb461 | 386 | h->orig_rsv = NULL; |
49b25e05 | 387 | h->aborted = 0; |
c5567237 | 388 | h->qgroup_reserved = qgroup_reserved; |
bed92eae | 389 | h->delayed_ref_elem.seq = 0; |
a698d075 | 390 | h->type = type; |
bed92eae | 391 | INIT_LIST_HEAD(&h->qgroup_ref_list); |
ea658bad | 392 | INIT_LIST_HEAD(&h->new_bgs); |
b7ec40d7 | 393 | |
a22285a6 YZ |
394 | smp_mb(); |
395 | if (cur_trans->blocked && may_wait_transaction(root, type)) { | |
396 | btrfs_commit_transaction(h, root); | |
397 | goto again; | |
398 | } | |
399 | ||
b5009945 | 400 | if (num_bytes) { |
8c2a3ca2 | 401 | trace_btrfs_space_reservation(root->fs_info, "transaction", |
2bcc0328 | 402 | h->transid, num_bytes, 1); |
b5009945 JB |
403 | h->block_rsv = &root->fs_info->trans_block_rsv; |
404 | h->bytes_reserved = num_bytes; | |
a22285a6 | 405 | } |
9ed74f2d | 406 | |
2a1eb461 | 407 | got_it: |
a4abeea4 | 408 | btrfs_record_root_in_trans(h, root); |
a22285a6 YZ |
409 | |
410 | if (!current->journal_info && type != TRANS_USERSPACE) | |
411 | current->journal_info = h; | |
79154b1b CM |
412 | return h; |
413 | } | |
414 | ||
f9295749 | 415 | struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, |
a22285a6 | 416 | int num_items) |
f9295749 | 417 | { |
08e007d2 MX |
418 | return start_transaction(root, num_items, TRANS_START, |
419 | BTRFS_RESERVE_FLUSH_ALL); | |
f9295749 | 420 | } |
8407aa46 | 421 | |
08e007d2 | 422 | struct btrfs_trans_handle *btrfs_start_transaction_lflush( |
8407aa46 MX |
423 | struct btrfs_root *root, int num_items) |
424 | { | |
08e007d2 MX |
425 | return start_transaction(root, num_items, TRANS_START, |
426 | BTRFS_RESERVE_FLUSH_LIMIT); | |
8407aa46 MX |
427 | } |
428 | ||
7a7eaa40 | 429 | struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root) |
f9295749 | 430 | { |
8407aa46 | 431 | return start_transaction(root, 0, TRANS_JOIN, 0); |
f9295749 CM |
432 | } |
433 | ||
7a7eaa40 | 434 | struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root) |
0af3d00b | 435 | { |
8407aa46 | 436 | return start_transaction(root, 0, TRANS_JOIN_NOLOCK, 0); |
0af3d00b JB |
437 | } |
438 | ||
7a7eaa40 | 439 | struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root) |
9ca9ee09 | 440 | { |
8407aa46 | 441 | return start_transaction(root, 0, TRANS_USERSPACE, 0); |
9ca9ee09 SW |
442 | } |
443 | ||
354aa0fb | 444 | struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root) |
60376ce4 | 445 | { |
354aa0fb | 446 | return start_transaction(root, 0, TRANS_ATTACH, 0); |
60376ce4 JB |
447 | } |
448 | ||
d352ac68 | 449 | /* wait for a transaction commit to be fully complete */ |
b9c8300c | 450 | static noinline void wait_for_commit(struct btrfs_root *root, |
89ce8a63 CM |
451 | struct btrfs_transaction *commit) |
452 | { | |
72d63ed6 | 453 | wait_event(commit->commit_wait, commit->commit_done); |
89ce8a63 CM |
454 | } |
455 | ||
46204592 SW |
456 | int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid) |
457 | { | |
458 | struct btrfs_transaction *cur_trans = NULL, *t; | |
459 | int ret; | |
460 | ||
46204592 SW |
461 | ret = 0; |
462 | if (transid) { | |
463 | if (transid <= root->fs_info->last_trans_committed) | |
a4abeea4 | 464 | goto out; |
46204592 SW |
465 | |
466 | /* find specified transaction */ | |
a4abeea4 | 467 | spin_lock(&root->fs_info->trans_lock); |
46204592 SW |
468 | list_for_each_entry(t, &root->fs_info->trans_list, list) { |
469 | if (t->transid == transid) { | |
470 | cur_trans = t; | |
a4abeea4 | 471 | atomic_inc(&cur_trans->use_count); |
46204592 SW |
472 | break; |
473 | } | |
474 | if (t->transid > transid) | |
475 | break; | |
476 | } | |
a4abeea4 | 477 | spin_unlock(&root->fs_info->trans_lock); |
46204592 SW |
478 | ret = -EINVAL; |
479 | if (!cur_trans) | |
a4abeea4 | 480 | goto out; /* bad transid */ |
46204592 SW |
481 | } else { |
482 | /* find newest transaction that is committing | committed */ | |
a4abeea4 | 483 | spin_lock(&root->fs_info->trans_lock); |
46204592 SW |
484 | list_for_each_entry_reverse(t, &root->fs_info->trans_list, |
485 | list) { | |
486 | if (t->in_commit) { | |
487 | if (t->commit_done) | |
3473f3c0 | 488 | break; |
46204592 | 489 | cur_trans = t; |
a4abeea4 | 490 | atomic_inc(&cur_trans->use_count); |
46204592 SW |
491 | break; |
492 | } | |
493 | } | |
a4abeea4 | 494 | spin_unlock(&root->fs_info->trans_lock); |
46204592 | 495 | if (!cur_trans) |
a4abeea4 | 496 | goto out; /* nothing committing|committed */ |
46204592 SW |
497 | } |
498 | ||
46204592 SW |
499 | wait_for_commit(root, cur_trans); |
500 | ||
46204592 SW |
501 | put_transaction(cur_trans); |
502 | ret = 0; | |
a4abeea4 | 503 | out: |
46204592 SW |
504 | return ret; |
505 | } | |
506 | ||
37d1aeee CM |
507 | void btrfs_throttle(struct btrfs_root *root) |
508 | { | |
a4abeea4 | 509 | if (!atomic_read(&root->fs_info->open_ioctl_trans)) |
9ca9ee09 | 510 | wait_current_trans(root); |
37d1aeee CM |
511 | } |
512 | ||
8929ecfa YZ |
513 | static int should_end_transaction(struct btrfs_trans_handle *trans, |
514 | struct btrfs_root *root) | |
515 | { | |
516 | int ret; | |
36ba022a JB |
517 | |
518 | ret = btrfs_block_rsv_check(root, &root->fs_info->global_block_rsv, 5); | |
8929ecfa YZ |
519 | return ret ? 1 : 0; |
520 | } | |
521 | ||
522 | int btrfs_should_end_transaction(struct btrfs_trans_handle *trans, | |
523 | struct btrfs_root *root) | |
524 | { | |
525 | struct btrfs_transaction *cur_trans = trans->transaction; | |
526 | int updates; | |
49b25e05 | 527 | int err; |
8929ecfa | 528 | |
a4abeea4 | 529 | smp_mb(); |
8929ecfa YZ |
530 | if (cur_trans->blocked || cur_trans->delayed_refs.flushing) |
531 | return 1; | |
532 | ||
533 | updates = trans->delayed_ref_updates; | |
534 | trans->delayed_ref_updates = 0; | |
49b25e05 JM |
535 | if (updates) { |
536 | err = btrfs_run_delayed_refs(trans, root, updates); | |
537 | if (err) /* Error code will also eval true */ | |
538 | return err; | |
539 | } | |
8929ecfa YZ |
540 | |
541 | return should_end_transaction(trans, root); | |
542 | } | |
543 | ||
89ce8a63 | 544 | static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, |
a698d075 | 545 | struct btrfs_root *root, int throttle) |
79154b1b | 546 | { |
8929ecfa | 547 | struct btrfs_transaction *cur_trans = trans->transaction; |
ab78c84d | 548 | struct btrfs_fs_info *info = root->fs_info; |
c3e69d58 | 549 | int count = 0; |
a698d075 | 550 | int lock = (trans->type != TRANS_JOIN_NOLOCK); |
4edc2ca3 | 551 | int err = 0; |
c3e69d58 | 552 | |
2a1eb461 JB |
553 | if (--trans->use_count) { |
554 | trans->block_rsv = trans->orig_rsv; | |
555 | return 0; | |
556 | } | |
557 | ||
edf39272 JS |
558 | /* |
559 | * do the qgroup accounting as early as possible | |
560 | */ | |
561 | err = btrfs_delayed_refs_qgroup_accounting(trans, info); | |
562 | ||
b24e03db | 563 | btrfs_trans_release_metadata(trans, root); |
4c13d758 | 564 | trans->block_rsv = NULL; |
d13603ef AJ |
565 | /* |
566 | * the same root has to be passed to start_transaction and | |
567 | * end_transaction. Subvolume quota depends on this. | |
568 | */ | |
569 | WARN_ON(trans->root != root); | |
c5567237 AJ |
570 | |
571 | if (trans->qgroup_reserved) { | |
572 | btrfs_qgroup_free(root, trans->qgroup_reserved); | |
573 | trans->qgroup_reserved = 0; | |
574 | } | |
575 | ||
ea658bad JB |
576 | if (!list_empty(&trans->new_bgs)) |
577 | btrfs_create_pending_block_groups(trans, root); | |
578 | ||
203bf287 | 579 | while (count < 2) { |
c3e69d58 CM |
580 | unsigned long cur = trans->delayed_ref_updates; |
581 | trans->delayed_ref_updates = 0; | |
582 | if (cur && | |
583 | trans->transaction->delayed_refs.num_heads_ready > 64) { | |
584 | trans->delayed_ref_updates = 0; | |
585 | btrfs_run_delayed_refs(trans, root, cur); | |
586 | } else { | |
587 | break; | |
588 | } | |
589 | count++; | |
56bec294 | 590 | } |
0e721106 JB |
591 | btrfs_trans_release_metadata(trans, root); |
592 | trans->block_rsv = NULL; | |
56bec294 | 593 | |
ea658bad JB |
594 | if (!list_empty(&trans->new_bgs)) |
595 | btrfs_create_pending_block_groups(trans, root); | |
596 | ||
a4abeea4 JB |
597 | if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) && |
598 | should_end_transaction(trans, root)) { | |
8929ecfa | 599 | trans->transaction->blocked = 1; |
a4abeea4 JB |
600 | smp_wmb(); |
601 | } | |
8929ecfa | 602 | |
0af3d00b | 603 | if (lock && cur_trans->blocked && !cur_trans->in_commit) { |
81317fde JB |
604 | if (throttle) { |
605 | /* | |
606 | * We may race with somebody else here so end up having | |
607 | * to call end_transaction on ourselves again, so inc | |
608 | * our use_count. | |
609 | */ | |
610 | trans->use_count++; | |
8929ecfa | 611 | return btrfs_commit_transaction(trans, root); |
81317fde | 612 | } else { |
8929ecfa | 613 | wake_up_process(info->transaction_kthread); |
81317fde | 614 | } |
8929ecfa YZ |
615 | } |
616 | ||
354aa0fb | 617 | if (trans->type < TRANS_JOIN_NOLOCK) |
98114659 | 618 | sb_end_intwrite(root->fs_info->sb); |
6df7881a | 619 | |
8929ecfa | 620 | WARN_ON(cur_trans != info->running_transaction); |
13c5a93e JB |
621 | WARN_ON(atomic_read(&cur_trans->num_writers) < 1); |
622 | atomic_dec(&cur_trans->num_writers); | |
89ce8a63 | 623 | |
99d16cbc | 624 | smp_mb(); |
79154b1b CM |
625 | if (waitqueue_active(&cur_trans->writer_wait)) |
626 | wake_up(&cur_trans->writer_wait); | |
79154b1b | 627 | put_transaction(cur_trans); |
9ed74f2d JB |
628 | |
629 | if (current->journal_info == trans) | |
630 | current->journal_info = NULL; | |
ab78c84d | 631 | |
24bbcf04 YZ |
632 | if (throttle) |
633 | btrfs_run_delayed_iputs(root); | |
634 | ||
49b25e05 JM |
635 | if (trans->aborted || |
636 | root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) { | |
4edc2ca3 | 637 | err = -EIO; |
49b25e05 | 638 | } |
edf39272 | 639 | assert_qgroups_uptodate(trans); |
49b25e05 | 640 | |
4edc2ca3 DJ |
641 | memset(trans, 0, sizeof(*trans)); |
642 | kmem_cache_free(btrfs_trans_handle_cachep, trans); | |
643 | return err; | |
79154b1b CM |
644 | } |
645 | ||
89ce8a63 CM |
646 | int btrfs_end_transaction(struct btrfs_trans_handle *trans, |
647 | struct btrfs_root *root) | |
648 | { | |
16cdcec7 MX |
649 | int ret; |
650 | ||
a698d075 | 651 | ret = __btrfs_end_transaction(trans, root, 0); |
16cdcec7 MX |
652 | if (ret) |
653 | return ret; | |
654 | return 0; | |
89ce8a63 CM |
655 | } |
656 | ||
657 | int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans, | |
658 | struct btrfs_root *root) | |
659 | { | |
16cdcec7 MX |
660 | int ret; |
661 | ||
a698d075 | 662 | ret = __btrfs_end_transaction(trans, root, 1); |
16cdcec7 MX |
663 | if (ret) |
664 | return ret; | |
665 | return 0; | |
666 | } | |
667 | ||
668 | int btrfs_end_transaction_dmeta(struct btrfs_trans_handle *trans, | |
669 | struct btrfs_root *root) | |
670 | { | |
a698d075 | 671 | return __btrfs_end_transaction(trans, root, 1); |
89ce8a63 CM |
672 | } |
673 | ||
d352ac68 CM |
674 | /* |
675 | * when btree blocks are allocated, they have some corresponding bits set for | |
676 | * them in one of two extent_io trees. This is used to make sure all of | |
690587d1 | 677 | * those extents are sent to disk but does not wait on them |
d352ac68 | 678 | */ |
690587d1 | 679 | int btrfs_write_marked_extents(struct btrfs_root *root, |
8cef4e16 | 680 | struct extent_io_tree *dirty_pages, int mark) |
79154b1b | 681 | { |
777e6bd7 | 682 | int err = 0; |
7c4452b9 | 683 | int werr = 0; |
1728366e | 684 | struct address_space *mapping = root->fs_info->btree_inode->i_mapping; |
e6138876 | 685 | struct extent_state *cached_state = NULL; |
777e6bd7 | 686 | u64 start = 0; |
5f39d397 | 687 | u64 end; |
7c4452b9 | 688 | |
1728366e | 689 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
e6138876 JB |
690 | mark, &cached_state)) { |
691 | convert_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT, | |
692 | mark, &cached_state, GFP_NOFS); | |
693 | cached_state = NULL; | |
1728366e JB |
694 | err = filemap_fdatawrite_range(mapping, start, end); |
695 | if (err) | |
696 | werr = err; | |
697 | cond_resched(); | |
698 | start = end + 1; | |
7c4452b9 | 699 | } |
690587d1 CM |
700 | if (err) |
701 | werr = err; | |
702 | return werr; | |
703 | } | |
704 | ||
705 | /* | |
706 | * when btree blocks are allocated, they have some corresponding bits set for | |
707 | * them in one of two extent_io trees. This is used to make sure all of | |
708 | * those extents are on disk for transaction or log commit. We wait | |
709 | * on all the pages and clear them from the dirty pages state tree | |
710 | */ | |
711 | int btrfs_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 712 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 | 713 | { |
690587d1 CM |
714 | int err = 0; |
715 | int werr = 0; | |
1728366e | 716 | struct address_space *mapping = root->fs_info->btree_inode->i_mapping; |
e6138876 | 717 | struct extent_state *cached_state = NULL; |
690587d1 CM |
718 | u64 start = 0; |
719 | u64 end; | |
777e6bd7 | 720 | |
1728366e | 721 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
e6138876 JB |
722 | EXTENT_NEED_WAIT, &cached_state)) { |
723 | clear_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT, | |
724 | 0, 0, &cached_state, GFP_NOFS); | |
1728366e JB |
725 | err = filemap_fdatawait_range(mapping, start, end); |
726 | if (err) | |
727 | werr = err; | |
728 | cond_resched(); | |
729 | start = end + 1; | |
777e6bd7 | 730 | } |
7c4452b9 CM |
731 | if (err) |
732 | werr = err; | |
733 | return werr; | |
79154b1b CM |
734 | } |
735 | ||
690587d1 CM |
736 | /* |
737 | * when btree blocks are allocated, they have some corresponding bits set for | |
738 | * them in one of two extent_io trees. This is used to make sure all of | |
739 | * those extents are on disk for transaction or log commit | |
740 | */ | |
741 | int btrfs_write_and_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 742 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 CM |
743 | { |
744 | int ret; | |
745 | int ret2; | |
746 | ||
8cef4e16 YZ |
747 | ret = btrfs_write_marked_extents(root, dirty_pages, mark); |
748 | ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark); | |
bf0da8c1 CM |
749 | |
750 | if (ret) | |
751 | return ret; | |
752 | if (ret2) | |
753 | return ret2; | |
754 | return 0; | |
690587d1 CM |
755 | } |
756 | ||
d0c803c4 CM |
757 | int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans, |
758 | struct btrfs_root *root) | |
759 | { | |
760 | if (!trans || !trans->transaction) { | |
761 | struct inode *btree_inode; | |
762 | btree_inode = root->fs_info->btree_inode; | |
763 | return filemap_write_and_wait(btree_inode->i_mapping); | |
764 | } | |
765 | return btrfs_write_and_wait_marked_extents(root, | |
8cef4e16 YZ |
766 | &trans->transaction->dirty_pages, |
767 | EXTENT_DIRTY); | |
d0c803c4 CM |
768 | } |
769 | ||
d352ac68 CM |
770 | /* |
771 | * this is used to update the root pointer in the tree of tree roots. | |
772 | * | |
773 | * But, in the case of the extent allocation tree, updating the root | |
774 | * pointer may allocate blocks which may change the root of the extent | |
775 | * allocation tree. | |
776 | * | |
777 | * So, this loops and repeats and makes sure the cowonly root didn't | |
778 | * change while the root pointer was being updated in the metadata. | |
779 | */ | |
0b86a832 CM |
780 | static int update_cowonly_root(struct btrfs_trans_handle *trans, |
781 | struct btrfs_root *root) | |
79154b1b CM |
782 | { |
783 | int ret; | |
0b86a832 | 784 | u64 old_root_bytenr; |
86b9f2ec | 785 | u64 old_root_used; |
0b86a832 | 786 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
79154b1b | 787 | |
86b9f2ec | 788 | old_root_used = btrfs_root_used(&root->root_item); |
0b86a832 | 789 | btrfs_write_dirty_block_groups(trans, root); |
56bec294 | 790 | |
d397712b | 791 | while (1) { |
0b86a832 | 792 | old_root_bytenr = btrfs_root_bytenr(&root->root_item); |
86b9f2ec YZ |
793 | if (old_root_bytenr == root->node->start && |
794 | old_root_used == btrfs_root_used(&root->root_item)) | |
79154b1b | 795 | break; |
87ef2bb4 | 796 | |
5d4f98a2 | 797 | btrfs_set_root_node(&root->root_item, root->node); |
79154b1b | 798 | ret = btrfs_update_root(trans, tree_root, |
0b86a832 CM |
799 | &root->root_key, |
800 | &root->root_item); | |
49b25e05 JM |
801 | if (ret) |
802 | return ret; | |
56bec294 | 803 | |
86b9f2ec | 804 | old_root_used = btrfs_root_used(&root->root_item); |
4a8c9a62 | 805 | ret = btrfs_write_dirty_block_groups(trans, root); |
49b25e05 JM |
806 | if (ret) |
807 | return ret; | |
0b86a832 | 808 | } |
276e680d YZ |
809 | |
810 | if (root != root->fs_info->extent_root) | |
811 | switch_commit_root(root); | |
812 | ||
0b86a832 CM |
813 | return 0; |
814 | } | |
815 | ||
d352ac68 CM |
816 | /* |
817 | * update all the cowonly tree roots on disk | |
49b25e05 JM |
818 | * |
819 | * The error handling in this function may not be obvious. Any of the | |
820 | * failures will cause the file system to go offline. We still need | |
821 | * to clean up the delayed refs. | |
d352ac68 | 822 | */ |
5d4f98a2 YZ |
823 | static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans, |
824 | struct btrfs_root *root) | |
0b86a832 CM |
825 | { |
826 | struct btrfs_fs_info *fs_info = root->fs_info; | |
827 | struct list_head *next; | |
84234f3a | 828 | struct extent_buffer *eb; |
56bec294 | 829 | int ret; |
84234f3a | 830 | |
56bec294 | 831 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
49b25e05 JM |
832 | if (ret) |
833 | return ret; | |
87ef2bb4 | 834 | |
84234f3a | 835 | eb = btrfs_lock_root_node(fs_info->tree_root); |
49b25e05 JM |
836 | ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, |
837 | 0, &eb); | |
84234f3a YZ |
838 | btrfs_tree_unlock(eb); |
839 | free_extent_buffer(eb); | |
0b86a832 | 840 | |
49b25e05 JM |
841 | if (ret) |
842 | return ret; | |
843 | ||
56bec294 | 844 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
49b25e05 JM |
845 | if (ret) |
846 | return ret; | |
87ef2bb4 | 847 | |
733f4fbb | 848 | ret = btrfs_run_dev_stats(trans, root->fs_info); |
8dabb742 SB |
849 | WARN_ON(ret); |
850 | ret = btrfs_run_dev_replace(trans, root->fs_info); | |
851 | WARN_ON(ret); | |
733f4fbb | 852 | |
546adb0d JS |
853 | ret = btrfs_run_qgroups(trans, root->fs_info); |
854 | BUG_ON(ret); | |
855 | ||
856 | /* run_qgroups might have added some more refs */ | |
857 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); | |
858 | BUG_ON(ret); | |
859 | ||
d397712b | 860 | while (!list_empty(&fs_info->dirty_cowonly_roots)) { |
0b86a832 CM |
861 | next = fs_info->dirty_cowonly_roots.next; |
862 | list_del_init(next); | |
863 | root = list_entry(next, struct btrfs_root, dirty_list); | |
87ef2bb4 | 864 | |
49b25e05 JM |
865 | ret = update_cowonly_root(trans, root); |
866 | if (ret) | |
867 | return ret; | |
79154b1b | 868 | } |
276e680d YZ |
869 | |
870 | down_write(&fs_info->extent_commit_sem); | |
871 | switch_commit_root(fs_info->extent_root); | |
872 | up_write(&fs_info->extent_commit_sem); | |
873 | ||
8dabb742 SB |
874 | btrfs_after_dev_replace_commit(fs_info); |
875 | ||
79154b1b CM |
876 | return 0; |
877 | } | |
878 | ||
d352ac68 CM |
879 | /* |
880 | * dead roots are old snapshots that need to be deleted. This allocates | |
881 | * a dirty root struct and adds it into the list of dead roots that need to | |
882 | * be deleted | |
883 | */ | |
5d4f98a2 | 884 | int btrfs_add_dead_root(struct btrfs_root *root) |
5eda7b5e | 885 | { |
a4abeea4 | 886 | spin_lock(&root->fs_info->trans_lock); |
5d4f98a2 | 887 | list_add(&root->root_list, &root->fs_info->dead_roots); |
a4abeea4 | 888 | spin_unlock(&root->fs_info->trans_lock); |
5eda7b5e CM |
889 | return 0; |
890 | } | |
891 | ||
d352ac68 | 892 | /* |
5d4f98a2 | 893 | * update all the cowonly tree roots on disk |
d352ac68 | 894 | */ |
5d4f98a2 YZ |
895 | static noinline int commit_fs_roots(struct btrfs_trans_handle *trans, |
896 | struct btrfs_root *root) | |
0f7d52f4 | 897 | { |
0f7d52f4 | 898 | struct btrfs_root *gang[8]; |
5d4f98a2 | 899 | struct btrfs_fs_info *fs_info = root->fs_info; |
0f7d52f4 CM |
900 | int i; |
901 | int ret; | |
54aa1f4d CM |
902 | int err = 0; |
903 | ||
a4abeea4 | 904 | spin_lock(&fs_info->fs_roots_radix_lock); |
d397712b | 905 | while (1) { |
5d4f98a2 YZ |
906 | ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, |
907 | (void **)gang, 0, | |
0f7d52f4 CM |
908 | ARRAY_SIZE(gang), |
909 | BTRFS_ROOT_TRANS_TAG); | |
910 | if (ret == 0) | |
911 | break; | |
912 | for (i = 0; i < ret; i++) { | |
913 | root = gang[i]; | |
5d4f98a2 YZ |
914 | radix_tree_tag_clear(&fs_info->fs_roots_radix, |
915 | (unsigned long)root->root_key.objectid, | |
916 | BTRFS_ROOT_TRANS_TAG); | |
a4abeea4 | 917 | spin_unlock(&fs_info->fs_roots_radix_lock); |
31153d81 | 918 | |
e02119d5 | 919 | btrfs_free_log(trans, root); |
5d4f98a2 | 920 | btrfs_update_reloc_root(trans, root); |
d68fc57b | 921 | btrfs_orphan_commit_root(trans, root); |
bcc63abb | 922 | |
82d5902d LZ |
923 | btrfs_save_ino_cache(root, trans); |
924 | ||
f1ebcc74 LB |
925 | /* see comments in should_cow_block() */ |
926 | root->force_cow = 0; | |
927 | smp_wmb(); | |
928 | ||
978d910d | 929 | if (root->commit_root != root->node) { |
581bb050 | 930 | mutex_lock(&root->fs_commit_mutex); |
817d52f8 | 931 | switch_commit_root(root); |
581bb050 LZ |
932 | btrfs_unpin_free_ino(root); |
933 | mutex_unlock(&root->fs_commit_mutex); | |
934 | ||
978d910d YZ |
935 | btrfs_set_root_node(&root->root_item, |
936 | root->node); | |
937 | } | |
5d4f98a2 | 938 | |
5d4f98a2 | 939 | err = btrfs_update_root(trans, fs_info->tree_root, |
0f7d52f4 CM |
940 | &root->root_key, |
941 | &root->root_item); | |
a4abeea4 | 942 | spin_lock(&fs_info->fs_roots_radix_lock); |
54aa1f4d CM |
943 | if (err) |
944 | break; | |
0f7d52f4 CM |
945 | } |
946 | } | |
a4abeea4 | 947 | spin_unlock(&fs_info->fs_roots_radix_lock); |
54aa1f4d | 948 | return err; |
0f7d52f4 CM |
949 | } |
950 | ||
d352ac68 CM |
951 | /* |
952 | * defrag a given btree. If cacheonly == 1, this won't read from the disk, | |
953 | * otherwise every leaf in the btree is read and defragged. | |
954 | */ | |
e9d0b13b CM |
955 | int btrfs_defrag_root(struct btrfs_root *root, int cacheonly) |
956 | { | |
957 | struct btrfs_fs_info *info = root->fs_info; | |
e9d0b13b | 958 | struct btrfs_trans_handle *trans; |
8929ecfa | 959 | int ret; |
e9d0b13b | 960 | |
8929ecfa | 961 | if (xchg(&root->defrag_running, 1)) |
e9d0b13b | 962 | return 0; |
8929ecfa | 963 | |
6b80053d | 964 | while (1) { |
8929ecfa YZ |
965 | trans = btrfs_start_transaction(root, 0); |
966 | if (IS_ERR(trans)) | |
967 | return PTR_ERR(trans); | |
968 | ||
e9d0b13b | 969 | ret = btrfs_defrag_leaves(trans, root, cacheonly); |
8929ecfa | 970 | |
e9d0b13b | 971 | btrfs_end_transaction(trans, root); |
b53d3f5d | 972 | btrfs_btree_balance_dirty(info->tree_root); |
e9d0b13b CM |
973 | cond_resched(); |
974 | ||
7841cb28 | 975 | if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN) |
e9d0b13b CM |
976 | break; |
977 | } | |
978 | root->defrag_running = 0; | |
8929ecfa | 979 | return ret; |
e9d0b13b CM |
980 | } |
981 | ||
d352ac68 CM |
982 | /* |
983 | * new snapshots need to be created at a very specific time in the | |
984 | * transaction commit. This does the actual creation | |
985 | */ | |
80b6794d | 986 | static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, |
3063d29f CM |
987 | struct btrfs_fs_info *fs_info, |
988 | struct btrfs_pending_snapshot *pending) | |
989 | { | |
990 | struct btrfs_key key; | |
80b6794d | 991 | struct btrfs_root_item *new_root_item; |
3063d29f CM |
992 | struct btrfs_root *tree_root = fs_info->tree_root; |
993 | struct btrfs_root *root = pending->root; | |
6bdb72de | 994 | struct btrfs_root *parent_root; |
98c9942a | 995 | struct btrfs_block_rsv *rsv; |
6bdb72de | 996 | struct inode *parent_inode; |
42874b3d MX |
997 | struct btrfs_path *path; |
998 | struct btrfs_dir_item *dir_item; | |
6a912213 | 999 | struct dentry *parent; |
a22285a6 | 1000 | struct dentry *dentry; |
3063d29f | 1001 | struct extent_buffer *tmp; |
925baedd | 1002 | struct extent_buffer *old; |
8ea05e3a | 1003 | struct timespec cur_time = CURRENT_TIME; |
3063d29f | 1004 | int ret; |
d68fc57b | 1005 | u64 to_reserve = 0; |
6bdb72de | 1006 | u64 index = 0; |
a22285a6 | 1007 | u64 objectid; |
b83cc969 | 1008 | u64 root_flags; |
8ea05e3a | 1009 | uuid_le new_uuid; |
3063d29f | 1010 | |
42874b3d MX |
1011 | path = btrfs_alloc_path(); |
1012 | if (!path) { | |
1013 | ret = pending->error = -ENOMEM; | |
1014 | goto path_alloc_fail; | |
1015 | } | |
1016 | ||
80b6794d CM |
1017 | new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS); |
1018 | if (!new_root_item) { | |
49b25e05 | 1019 | ret = pending->error = -ENOMEM; |
6fa9700e | 1020 | goto root_item_alloc_fail; |
80b6794d | 1021 | } |
a22285a6 | 1022 | |
581bb050 | 1023 | ret = btrfs_find_free_objectid(tree_root, &objectid); |
a22285a6 YZ |
1024 | if (ret) { |
1025 | pending->error = ret; | |
6fa9700e | 1026 | goto no_free_objectid; |
a22285a6 | 1027 | } |
3063d29f | 1028 | |
3fd0a558 | 1029 | btrfs_reloc_pre_snapshot(trans, pending, &to_reserve); |
d68fc57b YZ |
1030 | |
1031 | if (to_reserve > 0) { | |
08e007d2 MX |
1032 | ret = btrfs_block_rsv_add(root, &pending->block_rsv, |
1033 | to_reserve, | |
1034 | BTRFS_RESERVE_NO_FLUSH); | |
d68fc57b YZ |
1035 | if (ret) { |
1036 | pending->error = ret; | |
6fa9700e | 1037 | goto no_free_objectid; |
d68fc57b YZ |
1038 | } |
1039 | } | |
1040 | ||
6f72c7e2 AJ |
1041 | ret = btrfs_qgroup_inherit(trans, fs_info, root->root_key.objectid, |
1042 | objectid, pending->inherit); | |
6f72c7e2 AJ |
1043 | if (ret) { |
1044 | pending->error = ret; | |
6fa9700e | 1045 | goto no_free_objectid; |
6f72c7e2 AJ |
1046 | } |
1047 | ||
3063d29f | 1048 | key.objectid = objectid; |
a22285a6 YZ |
1049 | key.offset = (u64)-1; |
1050 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3063d29f | 1051 | |
6fa9700e | 1052 | rsv = trans->block_rsv; |
a22285a6 | 1053 | trans->block_rsv = &pending->block_rsv; |
3de4586c | 1054 | |
a22285a6 | 1055 | dentry = pending->dentry; |
6a912213 JB |
1056 | parent = dget_parent(dentry); |
1057 | parent_inode = parent->d_inode; | |
a22285a6 | 1058 | parent_root = BTRFS_I(parent_inode)->root; |
7585717f | 1059 | record_root_in_trans(trans, parent_root); |
a22285a6 | 1060 | |
3063d29f CM |
1061 | /* |
1062 | * insert the directory item | |
1063 | */ | |
3de4586c | 1064 | ret = btrfs_set_inode_index(parent_inode, &index); |
49b25e05 | 1065 | BUG_ON(ret); /* -ENOMEM */ |
42874b3d MX |
1066 | |
1067 | /* check if there is a file/dir which has the same name. */ | |
1068 | dir_item = btrfs_lookup_dir_item(NULL, parent_root, path, | |
1069 | btrfs_ino(parent_inode), | |
1070 | dentry->d_name.name, | |
1071 | dentry->d_name.len, 0); | |
1072 | if (dir_item != NULL && !IS_ERR(dir_item)) { | |
fe66a05a | 1073 | pending->error = -EEXIST; |
fe66a05a | 1074 | goto fail; |
42874b3d MX |
1075 | } else if (IS_ERR(dir_item)) { |
1076 | ret = PTR_ERR(dir_item); | |
8732d44f MX |
1077 | btrfs_abort_transaction(trans, root, ret); |
1078 | goto fail; | |
79787eaa | 1079 | } |
42874b3d | 1080 | btrfs_release_path(path); |
52c26179 | 1081 | |
e999376f CM |
1082 | /* |
1083 | * pull in the delayed directory update | |
1084 | * and the delayed inode item | |
1085 | * otherwise we corrupt the FS during | |
1086 | * snapshot | |
1087 | */ | |
1088 | ret = btrfs_run_delayed_items(trans, root); | |
8732d44f MX |
1089 | if (ret) { /* Transaction aborted */ |
1090 | btrfs_abort_transaction(trans, root, ret); | |
1091 | goto fail; | |
1092 | } | |
e999376f | 1093 | |
7585717f | 1094 | record_root_in_trans(trans, root); |
6bdb72de SW |
1095 | btrfs_set_root_last_snapshot(&root->root_item, trans->transid); |
1096 | memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); | |
08fe4db1 | 1097 | btrfs_check_and_init_root_item(new_root_item); |
6bdb72de | 1098 | |
b83cc969 LZ |
1099 | root_flags = btrfs_root_flags(new_root_item); |
1100 | if (pending->readonly) | |
1101 | root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; | |
1102 | else | |
1103 | root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; | |
1104 | btrfs_set_root_flags(new_root_item, root_flags); | |
1105 | ||
8ea05e3a AB |
1106 | btrfs_set_root_generation_v2(new_root_item, |
1107 | trans->transid); | |
1108 | uuid_le_gen(&new_uuid); | |
1109 | memcpy(new_root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE); | |
1110 | memcpy(new_root_item->parent_uuid, root->root_item.uuid, | |
1111 | BTRFS_UUID_SIZE); | |
1112 | new_root_item->otime.sec = cpu_to_le64(cur_time.tv_sec); | |
dadd1105 | 1113 | new_root_item->otime.nsec = cpu_to_le32(cur_time.tv_nsec); |
8ea05e3a AB |
1114 | btrfs_set_root_otransid(new_root_item, trans->transid); |
1115 | memset(&new_root_item->stime, 0, sizeof(new_root_item->stime)); | |
1116 | memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime)); | |
1117 | btrfs_set_root_stransid(new_root_item, 0); | |
1118 | btrfs_set_root_rtransid(new_root_item, 0); | |
1119 | ||
6bdb72de | 1120 | old = btrfs_lock_root_node(root); |
49b25e05 | 1121 | ret = btrfs_cow_block(trans, root, old, NULL, 0, &old); |
79787eaa JM |
1122 | if (ret) { |
1123 | btrfs_tree_unlock(old); | |
1124 | free_extent_buffer(old); | |
8732d44f MX |
1125 | btrfs_abort_transaction(trans, root, ret); |
1126 | goto fail; | |
79787eaa | 1127 | } |
49b25e05 | 1128 | |
6bdb72de SW |
1129 | btrfs_set_lock_blocking(old); |
1130 | ||
49b25e05 | 1131 | ret = btrfs_copy_root(trans, root, old, &tmp, objectid); |
79787eaa | 1132 | /* clean up in any case */ |
6bdb72de SW |
1133 | btrfs_tree_unlock(old); |
1134 | free_extent_buffer(old); | |
8732d44f MX |
1135 | if (ret) { |
1136 | btrfs_abort_transaction(trans, root, ret); | |
1137 | goto fail; | |
1138 | } | |
6bdb72de | 1139 | |
f1ebcc74 LB |
1140 | /* see comments in should_cow_block() */ |
1141 | root->force_cow = 1; | |
1142 | smp_wmb(); | |
1143 | ||
6bdb72de | 1144 | btrfs_set_root_node(new_root_item, tmp); |
a22285a6 YZ |
1145 | /* record when the snapshot was created in key.offset */ |
1146 | key.offset = trans->transid; | |
1147 | ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); | |
6bdb72de SW |
1148 | btrfs_tree_unlock(tmp); |
1149 | free_extent_buffer(tmp); | |
8732d44f MX |
1150 | if (ret) { |
1151 | btrfs_abort_transaction(trans, root, ret); | |
1152 | goto fail; | |
1153 | } | |
6bdb72de | 1154 | |
a22285a6 YZ |
1155 | /* |
1156 | * insert root back/forward references | |
1157 | */ | |
1158 | ret = btrfs_add_root_ref(trans, tree_root, objectid, | |
0660b5af | 1159 | parent_root->root_key.objectid, |
33345d01 | 1160 | btrfs_ino(parent_inode), index, |
a22285a6 | 1161 | dentry->d_name.name, dentry->d_name.len); |
8732d44f MX |
1162 | if (ret) { |
1163 | btrfs_abort_transaction(trans, root, ret); | |
1164 | goto fail; | |
1165 | } | |
0660b5af | 1166 | |
a22285a6 YZ |
1167 | key.offset = (u64)-1; |
1168 | pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key); | |
79787eaa JM |
1169 | if (IS_ERR(pending->snap)) { |
1170 | ret = PTR_ERR(pending->snap); | |
8732d44f MX |
1171 | btrfs_abort_transaction(trans, root, ret); |
1172 | goto fail; | |
79787eaa | 1173 | } |
d68fc57b | 1174 | |
49b25e05 | 1175 | ret = btrfs_reloc_post_snapshot(trans, pending); |
8732d44f MX |
1176 | if (ret) { |
1177 | btrfs_abort_transaction(trans, root, ret); | |
1178 | goto fail; | |
1179 | } | |
361048f5 MX |
1180 | |
1181 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); | |
8732d44f MX |
1182 | if (ret) { |
1183 | btrfs_abort_transaction(trans, root, ret); | |
1184 | goto fail; | |
1185 | } | |
42874b3d MX |
1186 | |
1187 | ret = btrfs_insert_dir_item(trans, parent_root, | |
1188 | dentry->d_name.name, dentry->d_name.len, | |
1189 | parent_inode, &key, | |
1190 | BTRFS_FT_DIR, index); | |
1191 | /* We have check then name at the beginning, so it is impossible. */ | |
1192 | BUG_ON(ret == -EEXIST); | |
8732d44f MX |
1193 | if (ret) { |
1194 | btrfs_abort_transaction(trans, root, ret); | |
1195 | goto fail; | |
1196 | } | |
42874b3d MX |
1197 | |
1198 | btrfs_i_size_write(parent_inode, parent_inode->i_size + | |
1199 | dentry->d_name.len * 2); | |
1200 | parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME; | |
be6aef60 | 1201 | ret = btrfs_update_inode_fallback(trans, parent_root, parent_inode); |
42874b3d | 1202 | if (ret) |
8732d44f | 1203 | btrfs_abort_transaction(trans, root, ret); |
3063d29f | 1204 | fail: |
6fa9700e | 1205 | dput(parent); |
98c9942a | 1206 | trans->block_rsv = rsv; |
6fa9700e MX |
1207 | no_free_objectid: |
1208 | kfree(new_root_item); | |
1209 | root_item_alloc_fail: | |
42874b3d MX |
1210 | btrfs_free_path(path); |
1211 | path_alloc_fail: | |
a22285a6 | 1212 | btrfs_block_rsv_release(root, &pending->block_rsv, (u64)-1); |
49b25e05 | 1213 | return ret; |
3063d29f CM |
1214 | } |
1215 | ||
d352ac68 CM |
1216 | /* |
1217 | * create all the snapshots we've scheduled for creation | |
1218 | */ | |
80b6794d CM |
1219 | static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans, |
1220 | struct btrfs_fs_info *fs_info) | |
3de4586c CM |
1221 | { |
1222 | struct btrfs_pending_snapshot *pending; | |
1223 | struct list_head *head = &trans->transaction->pending_snapshots; | |
3de4586c | 1224 | |
fe66a05a CM |
1225 | list_for_each_entry(pending, head, list) |
1226 | create_pending_snapshot(trans, fs_info, pending); | |
3de4586c CM |
1227 | return 0; |
1228 | } | |
1229 | ||
5d4f98a2 YZ |
1230 | static void update_super_roots(struct btrfs_root *root) |
1231 | { | |
1232 | struct btrfs_root_item *root_item; | |
1233 | struct btrfs_super_block *super; | |
1234 | ||
6c41761f | 1235 | super = root->fs_info->super_copy; |
5d4f98a2 YZ |
1236 | |
1237 | root_item = &root->fs_info->chunk_root->root_item; | |
1238 | super->chunk_root = root_item->bytenr; | |
1239 | super->chunk_root_generation = root_item->generation; | |
1240 | super->chunk_root_level = root_item->level; | |
1241 | ||
1242 | root_item = &root->fs_info->tree_root->root_item; | |
1243 | super->root = root_item->bytenr; | |
1244 | super->generation = root_item->generation; | |
1245 | super->root_level = root_item->level; | |
73bc1876 | 1246 | if (btrfs_test_opt(root, SPACE_CACHE)) |
0af3d00b | 1247 | super->cache_generation = root_item->generation; |
5d4f98a2 YZ |
1248 | } |
1249 | ||
f36f3042 CM |
1250 | int btrfs_transaction_in_commit(struct btrfs_fs_info *info) |
1251 | { | |
1252 | int ret = 0; | |
a4abeea4 | 1253 | spin_lock(&info->trans_lock); |
f36f3042 CM |
1254 | if (info->running_transaction) |
1255 | ret = info->running_transaction->in_commit; | |
a4abeea4 | 1256 | spin_unlock(&info->trans_lock); |
f36f3042 CM |
1257 | return ret; |
1258 | } | |
1259 | ||
8929ecfa YZ |
1260 | int btrfs_transaction_blocked(struct btrfs_fs_info *info) |
1261 | { | |
1262 | int ret = 0; | |
a4abeea4 | 1263 | spin_lock(&info->trans_lock); |
8929ecfa YZ |
1264 | if (info->running_transaction) |
1265 | ret = info->running_transaction->blocked; | |
a4abeea4 | 1266 | spin_unlock(&info->trans_lock); |
8929ecfa YZ |
1267 | return ret; |
1268 | } | |
1269 | ||
bb9c12c9 SW |
1270 | /* |
1271 | * wait for the current transaction commit to start and block subsequent | |
1272 | * transaction joins | |
1273 | */ | |
1274 | static void wait_current_trans_commit_start(struct btrfs_root *root, | |
1275 | struct btrfs_transaction *trans) | |
1276 | { | |
72d63ed6 | 1277 | wait_event(root->fs_info->transaction_blocked_wait, trans->in_commit); |
bb9c12c9 SW |
1278 | } |
1279 | ||
1280 | /* | |
1281 | * wait for the current transaction to start and then become unblocked. | |
1282 | * caller holds ref. | |
1283 | */ | |
1284 | static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root, | |
1285 | struct btrfs_transaction *trans) | |
1286 | { | |
72d63ed6 LZ |
1287 | wait_event(root->fs_info->transaction_wait, |
1288 | trans->commit_done || (trans->in_commit && !trans->blocked)); | |
bb9c12c9 SW |
1289 | } |
1290 | ||
1291 | /* | |
1292 | * commit transactions asynchronously. once btrfs_commit_transaction_async | |
1293 | * returns, any subsequent transaction will not be allowed to join. | |
1294 | */ | |
1295 | struct btrfs_async_commit { | |
1296 | struct btrfs_trans_handle *newtrans; | |
1297 | struct btrfs_root *root; | |
1298 | struct delayed_work work; | |
1299 | }; | |
1300 | ||
1301 | static void do_async_commit(struct work_struct *work) | |
1302 | { | |
1303 | struct btrfs_async_commit *ac = | |
1304 | container_of(work, struct btrfs_async_commit, work.work); | |
1305 | ||
6fc4e354 SW |
1306 | /* |
1307 | * We've got freeze protection passed with the transaction. | |
1308 | * Tell lockdep about it. | |
1309 | */ | |
ff7c1d33 MX |
1310 | if (ac->newtrans->type < TRANS_JOIN_NOLOCK) |
1311 | rwsem_acquire_read( | |
1312 | &ac->root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1], | |
1313 | 0, 1, _THIS_IP_); | |
6fc4e354 | 1314 | |
e209db7a SW |
1315 | current->journal_info = ac->newtrans; |
1316 | ||
bb9c12c9 SW |
1317 | btrfs_commit_transaction(ac->newtrans, ac->root); |
1318 | kfree(ac); | |
1319 | } | |
1320 | ||
1321 | int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans, | |
1322 | struct btrfs_root *root, | |
1323 | int wait_for_unblock) | |
1324 | { | |
1325 | struct btrfs_async_commit *ac; | |
1326 | struct btrfs_transaction *cur_trans; | |
1327 | ||
1328 | ac = kmalloc(sizeof(*ac), GFP_NOFS); | |
db5b493a TI |
1329 | if (!ac) |
1330 | return -ENOMEM; | |
bb9c12c9 SW |
1331 | |
1332 | INIT_DELAYED_WORK(&ac->work, do_async_commit); | |
1333 | ac->root = root; | |
7a7eaa40 | 1334 | ac->newtrans = btrfs_join_transaction(root); |
3612b495 TI |
1335 | if (IS_ERR(ac->newtrans)) { |
1336 | int err = PTR_ERR(ac->newtrans); | |
1337 | kfree(ac); | |
1338 | return err; | |
1339 | } | |
bb9c12c9 SW |
1340 | |
1341 | /* take transaction reference */ | |
bb9c12c9 | 1342 | cur_trans = trans->transaction; |
13c5a93e | 1343 | atomic_inc(&cur_trans->use_count); |
bb9c12c9 SW |
1344 | |
1345 | btrfs_end_transaction(trans, root); | |
6fc4e354 SW |
1346 | |
1347 | /* | |
1348 | * Tell lockdep we've released the freeze rwsem, since the | |
1349 | * async commit thread will be the one to unlock it. | |
1350 | */ | |
ff7c1d33 MX |
1351 | if (trans->type < TRANS_JOIN_NOLOCK) |
1352 | rwsem_release( | |
1353 | &root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1], | |
1354 | 1, _THIS_IP_); | |
6fc4e354 | 1355 | |
bb9c12c9 SW |
1356 | schedule_delayed_work(&ac->work, 0); |
1357 | ||
1358 | /* wait for transaction to start and unblock */ | |
bb9c12c9 SW |
1359 | if (wait_for_unblock) |
1360 | wait_current_trans_commit_start_and_unblock(root, cur_trans); | |
1361 | else | |
1362 | wait_current_trans_commit_start(root, cur_trans); | |
bb9c12c9 | 1363 | |
38e88054 SW |
1364 | if (current->journal_info == trans) |
1365 | current->journal_info = NULL; | |
1366 | ||
1367 | put_transaction(cur_trans); | |
bb9c12c9 SW |
1368 | return 0; |
1369 | } | |
1370 | ||
49b25e05 JM |
1371 | |
1372 | static void cleanup_transaction(struct btrfs_trans_handle *trans, | |
7b8b92af | 1373 | struct btrfs_root *root, int err) |
49b25e05 JM |
1374 | { |
1375 | struct btrfs_transaction *cur_trans = trans->transaction; | |
1376 | ||
1377 | WARN_ON(trans->use_count > 1); | |
1378 | ||
7b8b92af JB |
1379 | btrfs_abort_transaction(trans, root, err); |
1380 | ||
49b25e05 JM |
1381 | spin_lock(&root->fs_info->trans_lock); |
1382 | list_del_init(&cur_trans->list); | |
d7096fc3 JB |
1383 | if (cur_trans == root->fs_info->running_transaction) { |
1384 | root->fs_info->running_transaction = NULL; | |
1385 | root->fs_info->trans_no_join = 0; | |
1386 | } | |
49b25e05 JM |
1387 | spin_unlock(&root->fs_info->trans_lock); |
1388 | ||
1389 | btrfs_cleanup_one_transaction(trans->transaction, root); | |
1390 | ||
1391 | put_transaction(cur_trans); | |
1392 | put_transaction(cur_trans); | |
1393 | ||
1394 | trace_btrfs_transaction_commit(root); | |
1395 | ||
1396 | btrfs_scrub_continue(root); | |
1397 | ||
1398 | if (current->journal_info == trans) | |
1399 | current->journal_info = NULL; | |
1400 | ||
1401 | kmem_cache_free(btrfs_trans_handle_cachep, trans); | |
1402 | } | |
1403 | ||
ca469637 MX |
1404 | static int btrfs_flush_all_pending_stuffs(struct btrfs_trans_handle *trans, |
1405 | struct btrfs_root *root) | |
1406 | { | |
1407 | int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT); | |
1408 | int snap_pending = 0; | |
1409 | int ret; | |
1410 | ||
1411 | if (!flush_on_commit) { | |
1412 | spin_lock(&root->fs_info->trans_lock); | |
1413 | if (!list_empty(&trans->transaction->pending_snapshots)) | |
1414 | snap_pending = 1; | |
1415 | spin_unlock(&root->fs_info->trans_lock); | |
1416 | } | |
1417 | ||
1418 | if (flush_on_commit || snap_pending) { | |
1419 | btrfs_start_delalloc_inodes(root, 1); | |
1420 | btrfs_wait_ordered_extents(root, 1); | |
1421 | } | |
1422 | ||
1423 | ret = btrfs_run_delayed_items(trans, root); | |
1424 | if (ret) | |
1425 | return ret; | |
1426 | ||
1427 | /* | |
1428 | * running the delayed items may have added new refs. account | |
1429 | * them now so that they hinder processing of more delayed refs | |
1430 | * as little as possible. | |
1431 | */ | |
1432 | btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info); | |
1433 | ||
1434 | /* | |
1435 | * rename don't use btrfs_join_transaction, so, once we | |
1436 | * set the transaction to blocked above, we aren't going | |
1437 | * to get any new ordered operations. We can safely run | |
1438 | * it here and no for sure that nothing new will be added | |
1439 | * to the list | |
1440 | */ | |
1441 | btrfs_run_ordered_operations(root, 1); | |
1442 | ||
1443 | return 0; | |
1444 | } | |
1445 | ||
bb9c12c9 SW |
1446 | /* |
1447 | * btrfs_transaction state sequence: | |
1448 | * in_commit = 0, blocked = 0 (initial) | |
1449 | * in_commit = 1, blocked = 1 | |
1450 | * blocked = 0 | |
1451 | * commit_done = 1 | |
1452 | */ | |
79154b1b CM |
1453 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans, |
1454 | struct btrfs_root *root) | |
1455 | { | |
15ee9bc7 | 1456 | unsigned long joined = 0; |
49b25e05 | 1457 | struct btrfs_transaction *cur_trans = trans->transaction; |
8fd17795 | 1458 | struct btrfs_transaction *prev_trans = NULL; |
79154b1b | 1459 | DEFINE_WAIT(wait); |
25287e0a | 1460 | int ret; |
89573b9c CM |
1461 | int should_grow = 0; |
1462 | unsigned long now = get_seconds(); | |
79154b1b | 1463 | |
25287e0a MX |
1464 | ret = btrfs_run_ordered_operations(root, 0); |
1465 | if (ret) { | |
1466 | btrfs_abort_transaction(trans, root, ret); | |
1467 | goto cleanup_transaction; | |
1468 | } | |
5a3f23d5 | 1469 | |
25287e0a MX |
1470 | if (cur_trans->aborted) { |
1471 | ret = cur_trans->aborted; | |
49b25e05 | 1472 | goto cleanup_transaction; |
25287e0a | 1473 | } |
49b25e05 | 1474 | |
56bec294 CM |
1475 | /* make a pass through all the delayed refs we have so far |
1476 | * any runnings procs may add more while we are here | |
1477 | */ | |
1478 | ret = btrfs_run_delayed_refs(trans, root, 0); | |
49b25e05 JM |
1479 | if (ret) |
1480 | goto cleanup_transaction; | |
56bec294 | 1481 | |
0e721106 JB |
1482 | btrfs_trans_release_metadata(trans, root); |
1483 | trans->block_rsv = NULL; | |
1484 | ||
b7ec40d7 | 1485 | cur_trans = trans->transaction; |
49b25e05 | 1486 | |
56bec294 CM |
1487 | /* |
1488 | * set the flushing flag so procs in this transaction have to | |
1489 | * start sending their work down. | |
1490 | */ | |
b7ec40d7 | 1491 | cur_trans->delayed_refs.flushing = 1; |
56bec294 | 1492 | |
ea658bad JB |
1493 | if (!list_empty(&trans->new_bgs)) |
1494 | btrfs_create_pending_block_groups(trans, root); | |
1495 | ||
c3e69d58 | 1496 | ret = btrfs_run_delayed_refs(trans, root, 0); |
49b25e05 JM |
1497 | if (ret) |
1498 | goto cleanup_transaction; | |
56bec294 | 1499 | |
a4abeea4 | 1500 | spin_lock(&cur_trans->commit_lock); |
b7ec40d7 | 1501 | if (cur_trans->in_commit) { |
a4abeea4 | 1502 | spin_unlock(&cur_trans->commit_lock); |
13c5a93e | 1503 | atomic_inc(&cur_trans->use_count); |
49b25e05 | 1504 | ret = btrfs_end_transaction(trans, root); |
ccd467d6 | 1505 | |
b9c8300c | 1506 | wait_for_commit(root, cur_trans); |
15ee9bc7 | 1507 | |
79154b1b | 1508 | put_transaction(cur_trans); |
15ee9bc7 | 1509 | |
49b25e05 | 1510 | return ret; |
79154b1b | 1511 | } |
4313b399 | 1512 | |
2c90e5d6 | 1513 | trans->transaction->in_commit = 1; |
f9295749 | 1514 | trans->transaction->blocked = 1; |
a4abeea4 | 1515 | spin_unlock(&cur_trans->commit_lock); |
bb9c12c9 SW |
1516 | wake_up(&root->fs_info->transaction_blocked_wait); |
1517 | ||
a4abeea4 | 1518 | spin_lock(&root->fs_info->trans_lock); |
ccd467d6 CM |
1519 | if (cur_trans->list.prev != &root->fs_info->trans_list) { |
1520 | prev_trans = list_entry(cur_trans->list.prev, | |
1521 | struct btrfs_transaction, list); | |
1522 | if (!prev_trans->commit_done) { | |
13c5a93e | 1523 | atomic_inc(&prev_trans->use_count); |
a4abeea4 | 1524 | spin_unlock(&root->fs_info->trans_lock); |
ccd467d6 CM |
1525 | |
1526 | wait_for_commit(root, prev_trans); | |
ccd467d6 | 1527 | |
15ee9bc7 | 1528 | put_transaction(prev_trans); |
a4abeea4 JB |
1529 | } else { |
1530 | spin_unlock(&root->fs_info->trans_lock); | |
ccd467d6 | 1531 | } |
a4abeea4 JB |
1532 | } else { |
1533 | spin_unlock(&root->fs_info->trans_lock); | |
ccd467d6 | 1534 | } |
15ee9bc7 | 1535 | |
e39e64ac CM |
1536 | if (!btrfs_test_opt(root, SSD) && |
1537 | (now < cur_trans->start_time || now - cur_trans->start_time < 1)) | |
89573b9c CM |
1538 | should_grow = 1; |
1539 | ||
15ee9bc7 JB |
1540 | do { |
1541 | joined = cur_trans->num_joined; | |
7ea394f1 | 1542 | |
2c90e5d6 | 1543 | WARN_ON(cur_trans != trans->transaction); |
15ee9bc7 | 1544 | |
ca469637 | 1545 | ret = btrfs_flush_all_pending_stuffs(trans, root); |
49b25e05 JM |
1546 | if (ret) |
1547 | goto cleanup_transaction; | |
16cdcec7 | 1548 | |
ed3b3d31 CM |
1549 | prepare_to_wait(&cur_trans->writer_wait, &wait, |
1550 | TASK_UNINTERRUPTIBLE); | |
1551 | ||
13c5a93e | 1552 | if (atomic_read(&cur_trans->num_writers) > 1) |
99d16cbc SW |
1553 | schedule_timeout(MAX_SCHEDULE_TIMEOUT); |
1554 | else if (should_grow) | |
1555 | schedule_timeout(1); | |
15ee9bc7 | 1556 | |
15ee9bc7 | 1557 | finish_wait(&cur_trans->writer_wait, &wait); |
13c5a93e | 1558 | } while (atomic_read(&cur_trans->num_writers) > 1 || |
89573b9c | 1559 | (should_grow && cur_trans->num_joined != joined)); |
15ee9bc7 | 1560 | |
ca469637 MX |
1561 | ret = btrfs_flush_all_pending_stuffs(trans, root); |
1562 | if (ret) | |
1563 | goto cleanup_transaction; | |
1564 | ||
ed0ca140 JB |
1565 | /* |
1566 | * Ok now we need to make sure to block out any other joins while we | |
1567 | * commit the transaction. We could have started a join before setting | |
1568 | * no_join so make sure to wait for num_writers to == 1 again. | |
1569 | */ | |
1570 | spin_lock(&root->fs_info->trans_lock); | |
1571 | root->fs_info->trans_no_join = 1; | |
1572 | spin_unlock(&root->fs_info->trans_lock); | |
1573 | wait_event(cur_trans->writer_wait, | |
1574 | atomic_read(&cur_trans->num_writers) == 1); | |
1575 | ||
7585717f CM |
1576 | /* |
1577 | * the reloc mutex makes sure that we stop | |
1578 | * the balancing code from coming in and moving | |
1579 | * extents around in the middle of the commit | |
1580 | */ | |
1581 | mutex_lock(&root->fs_info->reloc_mutex); | |
1582 | ||
42874b3d MX |
1583 | /* |
1584 | * We needn't worry about the delayed items because we will | |
1585 | * deal with them in create_pending_snapshot(), which is the | |
1586 | * core function of the snapshot creation. | |
1587 | */ | |
1588 | ret = create_pending_snapshots(trans, root->fs_info); | |
49b25e05 JM |
1589 | if (ret) { |
1590 | mutex_unlock(&root->fs_info->reloc_mutex); | |
1591 | goto cleanup_transaction; | |
1592 | } | |
3063d29f | 1593 | |
42874b3d MX |
1594 | /* |
1595 | * We insert the dir indexes of the snapshots and update the inode | |
1596 | * of the snapshots' parents after the snapshot creation, so there | |
1597 | * are some delayed items which are not dealt with. Now deal with | |
1598 | * them. | |
1599 | * | |
1600 | * We needn't worry that this operation will corrupt the snapshots, | |
1601 | * because all the tree which are snapshoted will be forced to COW | |
1602 | * the nodes and leaves. | |
1603 | */ | |
1604 | ret = btrfs_run_delayed_items(trans, root); | |
49b25e05 JM |
1605 | if (ret) { |
1606 | mutex_unlock(&root->fs_info->reloc_mutex); | |
1607 | goto cleanup_transaction; | |
1608 | } | |
16cdcec7 | 1609 | |
56bec294 | 1610 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
49b25e05 JM |
1611 | if (ret) { |
1612 | mutex_unlock(&root->fs_info->reloc_mutex); | |
1613 | goto cleanup_transaction; | |
1614 | } | |
56bec294 | 1615 | |
e999376f CM |
1616 | /* |
1617 | * make sure none of the code above managed to slip in a | |
1618 | * delayed item | |
1619 | */ | |
1620 | btrfs_assert_delayed_root_empty(root); | |
1621 | ||
2c90e5d6 | 1622 | WARN_ON(cur_trans != trans->transaction); |
dc17ff8f | 1623 | |
a2de733c | 1624 | btrfs_scrub_pause(root); |
e02119d5 CM |
1625 | /* btrfs_commit_tree_roots is responsible for getting the |
1626 | * various roots consistent with each other. Every pointer | |
1627 | * in the tree of tree roots has to point to the most up to date | |
1628 | * root for every subvolume and other tree. So, we have to keep | |
1629 | * the tree logging code from jumping in and changing any | |
1630 | * of the trees. | |
1631 | * | |
1632 | * At this point in the commit, there can't be any tree-log | |
1633 | * writers, but a little lower down we drop the trans mutex | |
1634 | * and let new people in. By holding the tree_log_mutex | |
1635 | * from now until after the super is written, we avoid races | |
1636 | * with the tree-log code. | |
1637 | */ | |
1638 | mutex_lock(&root->fs_info->tree_log_mutex); | |
1639 | ||
5d4f98a2 | 1640 | ret = commit_fs_roots(trans, root); |
49b25e05 JM |
1641 | if (ret) { |
1642 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
871383be | 1643 | mutex_unlock(&root->fs_info->reloc_mutex); |
49b25e05 JM |
1644 | goto cleanup_transaction; |
1645 | } | |
54aa1f4d | 1646 | |
5d4f98a2 | 1647 | /* commit_fs_roots gets rid of all the tree log roots, it is now |
e02119d5 CM |
1648 | * safe to free the root of tree log roots |
1649 | */ | |
1650 | btrfs_free_log_root_tree(trans, root->fs_info); | |
1651 | ||
5d4f98a2 | 1652 | ret = commit_cowonly_roots(trans, root); |
49b25e05 JM |
1653 | if (ret) { |
1654 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
871383be | 1655 | mutex_unlock(&root->fs_info->reloc_mutex); |
49b25e05 JM |
1656 | goto cleanup_transaction; |
1657 | } | |
54aa1f4d | 1658 | |
11833d66 YZ |
1659 | btrfs_prepare_extent_commit(trans, root); |
1660 | ||
78fae27e | 1661 | cur_trans = root->fs_info->running_transaction; |
5d4f98a2 YZ |
1662 | |
1663 | btrfs_set_root_node(&root->fs_info->tree_root->root_item, | |
1664 | root->fs_info->tree_root->node); | |
817d52f8 | 1665 | switch_commit_root(root->fs_info->tree_root); |
5d4f98a2 YZ |
1666 | |
1667 | btrfs_set_root_node(&root->fs_info->chunk_root->root_item, | |
1668 | root->fs_info->chunk_root->node); | |
817d52f8 | 1669 | switch_commit_root(root->fs_info->chunk_root); |
5d4f98a2 | 1670 | |
edf39272 | 1671 | assert_qgroups_uptodate(trans); |
5d4f98a2 | 1672 | update_super_roots(root); |
e02119d5 CM |
1673 | |
1674 | if (!root->fs_info->log_root_recovering) { | |
6c41761f DS |
1675 | btrfs_set_super_log_root(root->fs_info->super_copy, 0); |
1676 | btrfs_set_super_log_root_level(root->fs_info->super_copy, 0); | |
e02119d5 CM |
1677 | } |
1678 | ||
6c41761f DS |
1679 | memcpy(root->fs_info->super_for_commit, root->fs_info->super_copy, |
1680 | sizeof(*root->fs_info->super_copy)); | |
ccd467d6 | 1681 | |
f9295749 | 1682 | trans->transaction->blocked = 0; |
a4abeea4 JB |
1683 | spin_lock(&root->fs_info->trans_lock); |
1684 | root->fs_info->running_transaction = NULL; | |
1685 | root->fs_info->trans_no_join = 0; | |
1686 | spin_unlock(&root->fs_info->trans_lock); | |
7585717f | 1687 | mutex_unlock(&root->fs_info->reloc_mutex); |
b7ec40d7 | 1688 | |
f9295749 | 1689 | wake_up(&root->fs_info->transaction_wait); |
e6dcd2dc | 1690 | |
79154b1b | 1691 | ret = btrfs_write_and_wait_transaction(trans, root); |
49b25e05 JM |
1692 | if (ret) { |
1693 | btrfs_error(root->fs_info, ret, | |
1694 | "Error while writing out transaction."); | |
1695 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1696 | goto cleanup_transaction; | |
1697 | } | |
1698 | ||
1699 | ret = write_ctree_super(trans, root, 0); | |
1700 | if (ret) { | |
1701 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1702 | goto cleanup_transaction; | |
1703 | } | |
4313b399 | 1704 | |
e02119d5 CM |
1705 | /* |
1706 | * the super is written, we can safely allow the tree-loggers | |
1707 | * to go about their business | |
1708 | */ | |
1709 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1710 | ||
11833d66 | 1711 | btrfs_finish_extent_commit(trans, root); |
4313b399 | 1712 | |
2c90e5d6 | 1713 | cur_trans->commit_done = 1; |
b7ec40d7 | 1714 | |
15ee9bc7 | 1715 | root->fs_info->last_trans_committed = cur_trans->transid; |
817d52f8 | 1716 | |
2c90e5d6 | 1717 | wake_up(&cur_trans->commit_wait); |
3de4586c | 1718 | |
a4abeea4 | 1719 | spin_lock(&root->fs_info->trans_lock); |
13c5a93e | 1720 | list_del_init(&cur_trans->list); |
a4abeea4 JB |
1721 | spin_unlock(&root->fs_info->trans_lock); |
1722 | ||
78fae27e | 1723 | put_transaction(cur_trans); |
79154b1b | 1724 | put_transaction(cur_trans); |
58176a96 | 1725 | |
354aa0fb MX |
1726 | if (trans->type < TRANS_JOIN_NOLOCK) |
1727 | sb_end_intwrite(root->fs_info->sb); | |
b2b5ef5c | 1728 | |
1abe9b8a | 1729 | trace_btrfs_transaction_commit(root); |
1730 | ||
a2de733c AJ |
1731 | btrfs_scrub_continue(root); |
1732 | ||
9ed74f2d JB |
1733 | if (current->journal_info == trans) |
1734 | current->journal_info = NULL; | |
1735 | ||
2c90e5d6 | 1736 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
24bbcf04 YZ |
1737 | |
1738 | if (current != root->fs_info->transaction_kthread) | |
1739 | btrfs_run_delayed_iputs(root); | |
1740 | ||
79154b1b | 1741 | return ret; |
49b25e05 JM |
1742 | |
1743 | cleanup_transaction: | |
0e721106 JB |
1744 | btrfs_trans_release_metadata(trans, root); |
1745 | trans->block_rsv = NULL; | |
49b25e05 JM |
1746 | btrfs_printk(root->fs_info, "Skipping commit of aborted transaction.\n"); |
1747 | // WARN_ON(1); | |
1748 | if (current->journal_info == trans) | |
1749 | current->journal_info = NULL; | |
7b8b92af | 1750 | cleanup_transaction(trans, root, ret); |
49b25e05 JM |
1751 | |
1752 | return ret; | |
79154b1b CM |
1753 | } |
1754 | ||
d352ac68 CM |
1755 | /* |
1756 | * interface function to delete all the snapshots we have scheduled for deletion | |
1757 | */ | |
e9d0b13b CM |
1758 | int btrfs_clean_old_snapshots(struct btrfs_root *root) |
1759 | { | |
5d4f98a2 YZ |
1760 | LIST_HEAD(list); |
1761 | struct btrfs_fs_info *fs_info = root->fs_info; | |
1762 | ||
a4abeea4 | 1763 | spin_lock(&fs_info->trans_lock); |
5d4f98a2 | 1764 | list_splice_init(&fs_info->dead_roots, &list); |
a4abeea4 | 1765 | spin_unlock(&fs_info->trans_lock); |
e9d0b13b | 1766 | |
5d4f98a2 | 1767 | while (!list_empty(&list)) { |
2c536799 JM |
1768 | int ret; |
1769 | ||
5d4f98a2 | 1770 | root = list_entry(list.next, struct btrfs_root, root_list); |
76dda93c YZ |
1771 | list_del(&root->root_list); |
1772 | ||
16cdcec7 MX |
1773 | btrfs_kill_all_delayed_nodes(root); |
1774 | ||
76dda93c YZ |
1775 | if (btrfs_header_backref_rev(root->node) < |
1776 | BTRFS_MIXED_BACKREF_REV) | |
2c536799 | 1777 | ret = btrfs_drop_snapshot(root, NULL, 0, 0); |
76dda93c | 1778 | else |
2c536799 JM |
1779 | ret =btrfs_drop_snapshot(root, NULL, 1, 0); |
1780 | BUG_ON(ret < 0); | |
e9d0b13b CM |
1781 | } |
1782 | return 0; | |
1783 | } |