]>
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" |
79154b1b | 33 | |
0f7d52f4 CM |
34 | #define BTRFS_ROOT_TRANS_TAG 0 |
35 | ||
49b25e05 | 36 | void put_transaction(struct btrfs_transaction *transaction) |
79154b1b | 37 | { |
13c5a93e JB |
38 | WARN_ON(atomic_read(&transaction->use_count) == 0); |
39 | if (atomic_dec_and_test(&transaction->use_count)) { | |
a4abeea4 | 40 | BUG_ON(!list_empty(&transaction->list)); |
00f04b88 | 41 | WARN_ON(transaction->delayed_refs.root.rb_node); |
2c90e5d6 CM |
42 | memset(transaction, 0, sizeof(*transaction)); |
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 | ||
d352ac68 CM |
53 | /* |
54 | * either allocate a new transaction or hop into the existing one | |
55 | */ | |
354aa0fb | 56 | static noinline int join_transaction(struct btrfs_root *root, int type) |
79154b1b CM |
57 | { |
58 | struct btrfs_transaction *cur_trans; | |
19ae4e81 | 59 | struct btrfs_fs_info *fs_info = root->fs_info; |
a4abeea4 | 60 | |
19ae4e81 | 61 | spin_lock(&fs_info->trans_lock); |
d43317dc | 62 | loop: |
49b25e05 | 63 | /* The file system has been taken offline. No new transactions. */ |
19ae4e81 JS |
64 | if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) { |
65 | spin_unlock(&fs_info->trans_lock); | |
49b25e05 JM |
66 | return -EROFS; |
67 | } | |
68 | ||
19ae4e81 | 69 | if (fs_info->trans_no_join) { |
354aa0fb MX |
70 | /* |
71 | * If we are JOIN_NOLOCK we're already committing a current | |
72 | * transaction, we just need a handle to deal with something | |
73 | * when committing the transaction, such as inode cache and | |
74 | * space cache. It is a special case. | |
75 | */ | |
76 | if (type != TRANS_JOIN_NOLOCK) { | |
19ae4e81 | 77 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 JB |
78 | return -EBUSY; |
79 | } | |
80 | } | |
81 | ||
19ae4e81 | 82 | cur_trans = fs_info->running_transaction; |
a4abeea4 | 83 | if (cur_trans) { |
871383be | 84 | if (cur_trans->aborted) { |
19ae4e81 | 85 | spin_unlock(&fs_info->trans_lock); |
49b25e05 | 86 | return cur_trans->aborted; |
871383be | 87 | } |
a4abeea4 | 88 | atomic_inc(&cur_trans->use_count); |
13c5a93e | 89 | atomic_inc(&cur_trans->num_writers); |
15ee9bc7 | 90 | cur_trans->num_joined++; |
19ae4e81 | 91 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 92 | return 0; |
79154b1b | 93 | } |
19ae4e81 | 94 | spin_unlock(&fs_info->trans_lock); |
a4abeea4 | 95 | |
354aa0fb MX |
96 | /* |
97 | * If we are ATTACH, we just want to catch the current transaction, | |
98 | * and commit it. If there is no transaction, just return ENOENT. | |
99 | */ | |
100 | if (type == TRANS_ATTACH) | |
101 | return -ENOENT; | |
102 | ||
a4abeea4 JB |
103 | cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, GFP_NOFS); |
104 | if (!cur_trans) | |
105 | return -ENOMEM; | |
d43317dc | 106 | |
19ae4e81 JS |
107 | spin_lock(&fs_info->trans_lock); |
108 | if (fs_info->running_transaction) { | |
d43317dc CM |
109 | /* |
110 | * someone started a transaction after we unlocked. Make sure | |
111 | * to redo the trans_no_join checks above | |
112 | */ | |
a4abeea4 | 113 | kmem_cache_free(btrfs_transaction_cachep, cur_trans); |
19ae4e81 | 114 | cur_trans = fs_info->running_transaction; |
d43317dc | 115 | goto loop; |
e4b50e14 DC |
116 | } else if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) { |
117 | spin_unlock(&fs_info->trans_lock); | |
7b8b92af JB |
118 | kmem_cache_free(btrfs_transaction_cachep, cur_trans); |
119 | return -EROFS; | |
79154b1b | 120 | } |
d43317dc | 121 | |
a4abeea4 JB |
122 | atomic_set(&cur_trans->num_writers, 1); |
123 | cur_trans->num_joined = 0; | |
124 | init_waitqueue_head(&cur_trans->writer_wait); | |
125 | init_waitqueue_head(&cur_trans->commit_wait); | |
126 | cur_trans->in_commit = 0; | |
127 | cur_trans->blocked = 0; | |
128 | /* | |
129 | * One for this trans handle, one so it will live on until we | |
130 | * commit the transaction. | |
131 | */ | |
132 | atomic_set(&cur_trans->use_count, 2); | |
133 | cur_trans->commit_done = 0; | |
134 | cur_trans->start_time = get_seconds(); | |
135 | ||
136 | cur_trans->delayed_refs.root = RB_ROOT; | |
137 | cur_trans->delayed_refs.num_entries = 0; | |
138 | cur_trans->delayed_refs.num_heads_ready = 0; | |
139 | cur_trans->delayed_refs.num_heads = 0; | |
140 | cur_trans->delayed_refs.flushing = 0; | |
141 | cur_trans->delayed_refs.run_delayed_start = 0; | |
20b297d6 JS |
142 | |
143 | /* | |
144 | * although the tree mod log is per file system and not per transaction, | |
145 | * the log must never go across transaction boundaries. | |
146 | */ | |
147 | smp_mb(); | |
31b1a2bd JL |
148 | if (!list_empty(&fs_info->tree_mod_seq_list)) |
149 | WARN(1, KERN_ERR "btrfs: tree_mod_seq_list not empty when " | |
20b297d6 | 150 | "creating a fresh transaction\n"); |
31b1a2bd JL |
151 | if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log)) |
152 | WARN(1, KERN_ERR "btrfs: tree_mod_log rb tree not empty when " | |
20b297d6 | 153 | "creating a fresh transaction\n"); |
20b297d6 JS |
154 | atomic_set(&fs_info->tree_mod_seq, 0); |
155 | ||
a4abeea4 JB |
156 | spin_lock_init(&cur_trans->commit_lock); |
157 | spin_lock_init(&cur_trans->delayed_refs.lock); | |
158 | ||
159 | INIT_LIST_HEAD(&cur_trans->pending_snapshots); | |
19ae4e81 | 160 | list_add_tail(&cur_trans->list, &fs_info->trans_list); |
a4abeea4 | 161 | extent_io_tree_init(&cur_trans->dirty_pages, |
19ae4e81 JS |
162 | fs_info->btree_inode->i_mapping); |
163 | fs_info->generation++; | |
164 | cur_trans->transid = fs_info->generation; | |
165 | fs_info->running_transaction = cur_trans; | |
49b25e05 | 166 | cur_trans->aborted = 0; |
19ae4e81 | 167 | spin_unlock(&fs_info->trans_lock); |
15ee9bc7 | 168 | |
79154b1b CM |
169 | return 0; |
170 | } | |
171 | ||
d352ac68 | 172 | /* |
d397712b CM |
173 | * this does all the record keeping required to make sure that a reference |
174 | * counted root is properly recorded in a given transaction. This is required | |
175 | * to make sure the old root from before we joined the transaction is deleted | |
176 | * when the transaction commits | |
d352ac68 | 177 | */ |
7585717f | 178 | static int record_root_in_trans(struct btrfs_trans_handle *trans, |
a4abeea4 | 179 | struct btrfs_root *root) |
6702ed49 | 180 | { |
5d4f98a2 | 181 | if (root->ref_cows && root->last_trans < trans->transid) { |
6702ed49 | 182 | WARN_ON(root == root->fs_info->extent_root); |
5d4f98a2 YZ |
183 | WARN_ON(root->commit_root != root->node); |
184 | ||
7585717f CM |
185 | /* |
186 | * see below for in_trans_setup usage rules | |
187 | * we have the reloc mutex held now, so there | |
188 | * is only one writer in this function | |
189 | */ | |
190 | root->in_trans_setup = 1; | |
191 | ||
192 | /* make sure readers find in_trans_setup before | |
193 | * they find our root->last_trans update | |
194 | */ | |
195 | smp_wmb(); | |
196 | ||
a4abeea4 JB |
197 | spin_lock(&root->fs_info->fs_roots_radix_lock); |
198 | if (root->last_trans == trans->transid) { | |
199 | spin_unlock(&root->fs_info->fs_roots_radix_lock); | |
200 | return 0; | |
201 | } | |
5d4f98a2 YZ |
202 | radix_tree_tag_set(&root->fs_info->fs_roots_radix, |
203 | (unsigned long)root->root_key.objectid, | |
204 | BTRFS_ROOT_TRANS_TAG); | |
a4abeea4 | 205 | spin_unlock(&root->fs_info->fs_roots_radix_lock); |
7585717f CM |
206 | root->last_trans = trans->transid; |
207 | ||
208 | /* this is pretty tricky. We don't want to | |
209 | * take the relocation lock in btrfs_record_root_in_trans | |
210 | * unless we're really doing the first setup for this root in | |
211 | * this transaction. | |
212 | * | |
213 | * Normally we'd use root->last_trans as a flag to decide | |
214 | * if we want to take the expensive mutex. | |
215 | * | |
216 | * But, we have to set root->last_trans before we | |
217 | * init the relocation root, otherwise, we trip over warnings | |
218 | * in ctree.c. The solution used here is to flag ourselves | |
219 | * with root->in_trans_setup. When this is 1, we're still | |
220 | * fixing up the reloc trees and everyone must wait. | |
221 | * | |
222 | * When this is zero, they can trust root->last_trans and fly | |
223 | * through btrfs_record_root_in_trans without having to take the | |
224 | * lock. smp_wmb() makes sure that all the writes above are | |
225 | * done before we pop in the zero below | |
226 | */ | |
5d4f98a2 | 227 | btrfs_init_reloc_root(trans, root); |
7585717f CM |
228 | smp_wmb(); |
229 | root->in_trans_setup = 0; | |
5d4f98a2 YZ |
230 | } |
231 | return 0; | |
232 | } | |
bcc63abb | 233 | |
7585717f CM |
234 | |
235 | int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, | |
236 | struct btrfs_root *root) | |
237 | { | |
238 | if (!root->ref_cows) | |
239 | return 0; | |
240 | ||
241 | /* | |
242 | * see record_root_in_trans for comments about in_trans_setup usage | |
243 | * and barriers | |
244 | */ | |
245 | smp_rmb(); | |
246 | if (root->last_trans == trans->transid && | |
247 | !root->in_trans_setup) | |
248 | return 0; | |
249 | ||
250 | mutex_lock(&root->fs_info->reloc_mutex); | |
251 | record_root_in_trans(trans, root); | |
252 | mutex_unlock(&root->fs_info->reloc_mutex); | |
253 | ||
254 | return 0; | |
255 | } | |
256 | ||
d352ac68 CM |
257 | /* wait for commit against the current transaction to become unblocked |
258 | * when this is done, it is safe to start a new transaction, but the current | |
259 | * transaction might not be fully on disk. | |
260 | */ | |
37d1aeee | 261 | static void wait_current_trans(struct btrfs_root *root) |
79154b1b | 262 | { |
f9295749 | 263 | struct btrfs_transaction *cur_trans; |
79154b1b | 264 | |
a4abeea4 | 265 | spin_lock(&root->fs_info->trans_lock); |
f9295749 | 266 | cur_trans = root->fs_info->running_transaction; |
37d1aeee | 267 | if (cur_trans && cur_trans->blocked) { |
13c5a93e | 268 | atomic_inc(&cur_trans->use_count); |
a4abeea4 | 269 | spin_unlock(&root->fs_info->trans_lock); |
72d63ed6 LZ |
270 | |
271 | wait_event(root->fs_info->transaction_wait, | |
272 | !cur_trans->blocked); | |
f9295749 | 273 | put_transaction(cur_trans); |
a4abeea4 JB |
274 | } else { |
275 | spin_unlock(&root->fs_info->trans_lock); | |
f9295749 | 276 | } |
37d1aeee CM |
277 | } |
278 | ||
a22285a6 YZ |
279 | static int may_wait_transaction(struct btrfs_root *root, int type) |
280 | { | |
a4abeea4 JB |
281 | if (root->fs_info->log_root_recovering) |
282 | return 0; | |
283 | ||
284 | if (type == TRANS_USERSPACE) | |
285 | return 1; | |
286 | ||
287 | if (type == TRANS_START && | |
288 | !atomic_read(&root->fs_info->open_ioctl_trans)) | |
a22285a6 | 289 | return 1; |
a4abeea4 | 290 | |
a22285a6 YZ |
291 | return 0; |
292 | } | |
293 | ||
08e007d2 MX |
294 | static struct btrfs_trans_handle * |
295 | start_transaction(struct btrfs_root *root, u64 num_items, int type, | |
296 | enum btrfs_reserve_flush_enum flush) | |
37d1aeee | 297 | { |
a22285a6 YZ |
298 | struct btrfs_trans_handle *h; |
299 | struct btrfs_transaction *cur_trans; | |
b5009945 | 300 | u64 num_bytes = 0; |
37d1aeee | 301 | int ret; |
c5567237 | 302 | u64 qgroup_reserved = 0; |
acce952b | 303 | |
304 | if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) | |
305 | return ERR_PTR(-EROFS); | |
2a1eb461 JB |
306 | |
307 | if (current->journal_info) { | |
308 | WARN_ON(type != TRANS_JOIN && type != TRANS_JOIN_NOLOCK); | |
309 | h = current->journal_info; | |
310 | h->use_count++; | |
b7d5b0a8 | 311 | WARN_ON(h->use_count > 2); |
2a1eb461 JB |
312 | h->orig_rsv = h->block_rsv; |
313 | h->block_rsv = NULL; | |
314 | goto got_it; | |
315 | } | |
b5009945 JB |
316 | |
317 | /* | |
318 | * Do the reservation before we join the transaction so we can do all | |
319 | * the appropriate flushing if need be. | |
320 | */ | |
321 | if (num_items > 0 && root != root->fs_info->chunk_root) { | |
c5567237 AJ |
322 | if (root->fs_info->quota_enabled && |
323 | is_fstree(root->root_key.objectid)) { | |
324 | qgroup_reserved = num_items * root->leafsize; | |
325 | ret = btrfs_qgroup_reserve(root, qgroup_reserved); | |
326 | if (ret) | |
327 | return ERR_PTR(ret); | |
328 | } | |
329 | ||
b5009945 | 330 | num_bytes = btrfs_calc_trans_metadata_size(root, num_items); |
08e007d2 MX |
331 | ret = btrfs_block_rsv_add(root, |
332 | &root->fs_info->trans_block_rsv, | |
333 | num_bytes, flush); | |
b5009945 JB |
334 | if (ret) |
335 | return ERR_PTR(ret); | |
336 | } | |
a22285a6 YZ |
337 | again: |
338 | h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS); | |
339 | if (!h) | |
340 | return ERR_PTR(-ENOMEM); | |
37d1aeee | 341 | |
98114659 JB |
342 | /* |
343 | * If we are JOIN_NOLOCK we're already committing a transaction and | |
344 | * waiting on this guy, so we don't need to do the sb_start_intwrite | |
345 | * because we're already holding a ref. We need this because we could | |
346 | * have raced in and did an fsync() on a file which can kick a commit | |
347 | * and then we deadlock with somebody doing a freeze. | |
354aa0fb MX |
348 | * |
349 | * If we are ATTACH, it means we just want to catch the current | |
350 | * transaction and commit it, so we needn't do sb_start_intwrite(). | |
98114659 | 351 | */ |
354aa0fb | 352 | if (type < TRANS_JOIN_NOLOCK) |
60376ce4 | 353 | sb_start_intwrite(root->fs_info->sb); |
b2b5ef5c | 354 | |
a22285a6 | 355 | if (may_wait_transaction(root, type)) |
37d1aeee | 356 | wait_current_trans(root); |
a22285a6 | 357 | |
a4abeea4 | 358 | do { |
354aa0fb | 359 | ret = join_transaction(root, type); |
a4abeea4 JB |
360 | if (ret == -EBUSY) |
361 | wait_current_trans(root); | |
362 | } while (ret == -EBUSY); | |
363 | ||
db5b493a | 364 | if (ret < 0) { |
354aa0fb MX |
365 | /* We must get the transaction if we are JOIN_NOLOCK. */ |
366 | BUG_ON(type == TRANS_JOIN_NOLOCK); | |
367 | ||
368 | if (type < TRANS_JOIN_NOLOCK) | |
369 | sb_end_intwrite(root->fs_info->sb); | |
6e8df2ae | 370 | kmem_cache_free(btrfs_trans_handle_cachep, h); |
db5b493a TI |
371 | return ERR_PTR(ret); |
372 | } | |
0f7d52f4 | 373 | |
a22285a6 | 374 | cur_trans = root->fs_info->running_transaction; |
a22285a6 YZ |
375 | |
376 | h->transid = cur_trans->transid; | |
377 | h->transaction = cur_trans; | |
79154b1b | 378 | h->blocks_used = 0; |
a22285a6 | 379 | h->bytes_reserved = 0; |
d13603ef | 380 | h->root = root; |
56bec294 | 381 | h->delayed_ref_updates = 0; |
2a1eb461 | 382 | h->use_count = 1; |
0e721106 | 383 | h->adding_csums = 0; |
f0486c68 | 384 | h->block_rsv = NULL; |
2a1eb461 | 385 | h->orig_rsv = NULL; |
49b25e05 | 386 | h->aborted = 0; |
c5567237 | 387 | h->qgroup_reserved = qgroup_reserved; |
bed92eae | 388 | h->delayed_ref_elem.seq = 0; |
a698d075 | 389 | h->type = type; |
bed92eae | 390 | INIT_LIST_HEAD(&h->qgroup_ref_list); |
ea658bad | 391 | INIT_LIST_HEAD(&h->new_bgs); |
b7ec40d7 | 392 | |
a22285a6 YZ |
393 | smp_mb(); |
394 | if (cur_trans->blocked && may_wait_transaction(root, type)) { | |
395 | btrfs_commit_transaction(h, root); | |
396 | goto again; | |
397 | } | |
398 | ||
b5009945 | 399 | if (num_bytes) { |
8c2a3ca2 | 400 | trace_btrfs_space_reservation(root->fs_info, "transaction", |
2bcc0328 | 401 | h->transid, num_bytes, 1); |
b5009945 JB |
402 | h->block_rsv = &root->fs_info->trans_block_rsv; |
403 | h->bytes_reserved = num_bytes; | |
a22285a6 | 404 | } |
9ed74f2d | 405 | |
2a1eb461 | 406 | got_it: |
a4abeea4 | 407 | btrfs_record_root_in_trans(h, root); |
a22285a6 YZ |
408 | |
409 | if (!current->journal_info && type != TRANS_USERSPACE) | |
410 | current->journal_info = h; | |
79154b1b CM |
411 | return h; |
412 | } | |
413 | ||
f9295749 | 414 | struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, |
a22285a6 | 415 | int num_items) |
f9295749 | 416 | { |
08e007d2 MX |
417 | return start_transaction(root, num_items, TRANS_START, |
418 | BTRFS_RESERVE_FLUSH_ALL); | |
f9295749 | 419 | } |
8407aa46 | 420 | |
08e007d2 | 421 | struct btrfs_trans_handle *btrfs_start_transaction_lflush( |
8407aa46 MX |
422 | struct btrfs_root *root, int num_items) |
423 | { | |
08e007d2 MX |
424 | return start_transaction(root, num_items, TRANS_START, |
425 | BTRFS_RESERVE_FLUSH_LIMIT); | |
8407aa46 MX |
426 | } |
427 | ||
7a7eaa40 | 428 | struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root) |
f9295749 | 429 | { |
8407aa46 | 430 | return start_transaction(root, 0, TRANS_JOIN, 0); |
f9295749 CM |
431 | } |
432 | ||
7a7eaa40 | 433 | struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root) |
0af3d00b | 434 | { |
8407aa46 | 435 | return start_transaction(root, 0, TRANS_JOIN_NOLOCK, 0); |
0af3d00b JB |
436 | } |
437 | ||
7a7eaa40 | 438 | struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root) |
9ca9ee09 | 439 | { |
8407aa46 | 440 | return start_transaction(root, 0, TRANS_USERSPACE, 0); |
9ca9ee09 SW |
441 | } |
442 | ||
354aa0fb | 443 | struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root) |
60376ce4 | 444 | { |
354aa0fb | 445 | return start_transaction(root, 0, TRANS_ATTACH, 0); |
60376ce4 JB |
446 | } |
447 | ||
d352ac68 | 448 | /* wait for a transaction commit to be fully complete */ |
b9c8300c | 449 | static noinline void wait_for_commit(struct btrfs_root *root, |
89ce8a63 CM |
450 | struct btrfs_transaction *commit) |
451 | { | |
72d63ed6 | 452 | wait_event(commit->commit_wait, commit->commit_done); |
89ce8a63 CM |
453 | } |
454 | ||
46204592 SW |
455 | int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid) |
456 | { | |
457 | struct btrfs_transaction *cur_trans = NULL, *t; | |
458 | int ret; | |
459 | ||
46204592 SW |
460 | ret = 0; |
461 | if (transid) { | |
462 | if (transid <= root->fs_info->last_trans_committed) | |
a4abeea4 | 463 | goto out; |
46204592 SW |
464 | |
465 | /* find specified transaction */ | |
a4abeea4 | 466 | spin_lock(&root->fs_info->trans_lock); |
46204592 SW |
467 | list_for_each_entry(t, &root->fs_info->trans_list, list) { |
468 | if (t->transid == transid) { | |
469 | cur_trans = t; | |
a4abeea4 | 470 | atomic_inc(&cur_trans->use_count); |
46204592 SW |
471 | break; |
472 | } | |
473 | if (t->transid > transid) | |
474 | break; | |
475 | } | |
a4abeea4 | 476 | spin_unlock(&root->fs_info->trans_lock); |
46204592 SW |
477 | ret = -EINVAL; |
478 | if (!cur_trans) | |
a4abeea4 | 479 | goto out; /* bad transid */ |
46204592 SW |
480 | } else { |
481 | /* find newest transaction that is committing | committed */ | |
a4abeea4 | 482 | spin_lock(&root->fs_info->trans_lock); |
46204592 SW |
483 | list_for_each_entry_reverse(t, &root->fs_info->trans_list, |
484 | list) { | |
485 | if (t->in_commit) { | |
486 | if (t->commit_done) | |
3473f3c0 | 487 | break; |
46204592 | 488 | cur_trans = t; |
a4abeea4 | 489 | atomic_inc(&cur_trans->use_count); |
46204592 SW |
490 | break; |
491 | } | |
492 | } | |
a4abeea4 | 493 | spin_unlock(&root->fs_info->trans_lock); |
46204592 | 494 | if (!cur_trans) |
a4abeea4 | 495 | goto out; /* nothing committing|committed */ |
46204592 SW |
496 | } |
497 | ||
46204592 SW |
498 | wait_for_commit(root, cur_trans); |
499 | ||
46204592 SW |
500 | put_transaction(cur_trans); |
501 | ret = 0; | |
a4abeea4 | 502 | out: |
46204592 SW |
503 | return ret; |
504 | } | |
505 | ||
37d1aeee CM |
506 | void btrfs_throttle(struct btrfs_root *root) |
507 | { | |
a4abeea4 | 508 | if (!atomic_read(&root->fs_info->open_ioctl_trans)) |
9ca9ee09 | 509 | wait_current_trans(root); |
37d1aeee CM |
510 | } |
511 | ||
8929ecfa YZ |
512 | static int should_end_transaction(struct btrfs_trans_handle *trans, |
513 | struct btrfs_root *root) | |
514 | { | |
515 | int ret; | |
36ba022a JB |
516 | |
517 | ret = btrfs_block_rsv_check(root, &root->fs_info->global_block_rsv, 5); | |
8929ecfa YZ |
518 | return ret ? 1 : 0; |
519 | } | |
520 | ||
521 | int btrfs_should_end_transaction(struct btrfs_trans_handle *trans, | |
522 | struct btrfs_root *root) | |
523 | { | |
524 | struct btrfs_transaction *cur_trans = trans->transaction; | |
525 | int updates; | |
49b25e05 | 526 | int err; |
8929ecfa | 527 | |
a4abeea4 | 528 | smp_mb(); |
8929ecfa YZ |
529 | if (cur_trans->blocked || cur_trans->delayed_refs.flushing) |
530 | return 1; | |
531 | ||
532 | updates = trans->delayed_ref_updates; | |
533 | trans->delayed_ref_updates = 0; | |
49b25e05 JM |
534 | if (updates) { |
535 | err = btrfs_run_delayed_refs(trans, root, updates); | |
536 | if (err) /* Error code will also eval true */ | |
537 | return err; | |
538 | } | |
8929ecfa YZ |
539 | |
540 | return should_end_transaction(trans, root); | |
541 | } | |
542 | ||
89ce8a63 | 543 | static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, |
a698d075 | 544 | struct btrfs_root *root, int throttle) |
79154b1b | 545 | { |
8929ecfa | 546 | struct btrfs_transaction *cur_trans = trans->transaction; |
ab78c84d | 547 | struct btrfs_fs_info *info = root->fs_info; |
c3e69d58 | 548 | int count = 0; |
a698d075 | 549 | int lock = (trans->type != TRANS_JOIN_NOLOCK); |
4edc2ca3 | 550 | int err = 0; |
c3e69d58 | 551 | |
2a1eb461 JB |
552 | if (--trans->use_count) { |
553 | trans->block_rsv = trans->orig_rsv; | |
554 | return 0; | |
555 | } | |
556 | ||
edf39272 JS |
557 | /* |
558 | * do the qgroup accounting as early as possible | |
559 | */ | |
560 | err = btrfs_delayed_refs_qgroup_accounting(trans, info); | |
561 | ||
b24e03db | 562 | btrfs_trans_release_metadata(trans, root); |
4c13d758 | 563 | trans->block_rsv = NULL; |
d13603ef AJ |
564 | /* |
565 | * the same root has to be passed to start_transaction and | |
566 | * end_transaction. Subvolume quota depends on this. | |
567 | */ | |
568 | WARN_ON(trans->root != root); | |
c5567237 AJ |
569 | |
570 | if (trans->qgroup_reserved) { | |
571 | btrfs_qgroup_free(root, trans->qgroup_reserved); | |
572 | trans->qgroup_reserved = 0; | |
573 | } | |
574 | ||
ea658bad JB |
575 | if (!list_empty(&trans->new_bgs)) |
576 | btrfs_create_pending_block_groups(trans, root); | |
577 | ||
203bf287 | 578 | while (count < 2) { |
c3e69d58 CM |
579 | unsigned long cur = trans->delayed_ref_updates; |
580 | trans->delayed_ref_updates = 0; | |
581 | if (cur && | |
582 | trans->transaction->delayed_refs.num_heads_ready > 64) { | |
583 | trans->delayed_ref_updates = 0; | |
584 | btrfs_run_delayed_refs(trans, root, cur); | |
585 | } else { | |
586 | break; | |
587 | } | |
588 | count++; | |
56bec294 | 589 | } |
0e721106 JB |
590 | btrfs_trans_release_metadata(trans, root); |
591 | trans->block_rsv = NULL; | |
56bec294 | 592 | |
ea658bad JB |
593 | if (!list_empty(&trans->new_bgs)) |
594 | btrfs_create_pending_block_groups(trans, root); | |
595 | ||
a4abeea4 JB |
596 | if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) && |
597 | should_end_transaction(trans, root)) { | |
8929ecfa | 598 | trans->transaction->blocked = 1; |
a4abeea4 JB |
599 | smp_wmb(); |
600 | } | |
8929ecfa | 601 | |
0af3d00b | 602 | if (lock && cur_trans->blocked && !cur_trans->in_commit) { |
81317fde JB |
603 | if (throttle) { |
604 | /* | |
605 | * We may race with somebody else here so end up having | |
606 | * to call end_transaction on ourselves again, so inc | |
607 | * our use_count. | |
608 | */ | |
609 | trans->use_count++; | |
8929ecfa | 610 | return btrfs_commit_transaction(trans, root); |
81317fde | 611 | } else { |
8929ecfa | 612 | wake_up_process(info->transaction_kthread); |
81317fde | 613 | } |
8929ecfa YZ |
614 | } |
615 | ||
354aa0fb | 616 | if (trans->type < TRANS_JOIN_NOLOCK) |
98114659 | 617 | sb_end_intwrite(root->fs_info->sb); |
6df7881a | 618 | |
8929ecfa | 619 | WARN_ON(cur_trans != info->running_transaction); |
13c5a93e JB |
620 | WARN_ON(atomic_read(&cur_trans->num_writers) < 1); |
621 | atomic_dec(&cur_trans->num_writers); | |
89ce8a63 | 622 | |
99d16cbc | 623 | smp_mb(); |
79154b1b CM |
624 | if (waitqueue_active(&cur_trans->writer_wait)) |
625 | wake_up(&cur_trans->writer_wait); | |
79154b1b | 626 | put_transaction(cur_trans); |
9ed74f2d JB |
627 | |
628 | if (current->journal_info == trans) | |
629 | current->journal_info = NULL; | |
ab78c84d | 630 | |
24bbcf04 YZ |
631 | if (throttle) |
632 | btrfs_run_delayed_iputs(root); | |
633 | ||
49b25e05 JM |
634 | if (trans->aborted || |
635 | root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) { | |
4edc2ca3 | 636 | err = -EIO; |
49b25e05 | 637 | } |
edf39272 | 638 | assert_qgroups_uptodate(trans); |
49b25e05 | 639 | |
4edc2ca3 DJ |
640 | memset(trans, 0, sizeof(*trans)); |
641 | kmem_cache_free(btrfs_trans_handle_cachep, trans); | |
642 | return err; | |
79154b1b CM |
643 | } |
644 | ||
89ce8a63 CM |
645 | int btrfs_end_transaction(struct btrfs_trans_handle *trans, |
646 | struct btrfs_root *root) | |
647 | { | |
16cdcec7 MX |
648 | int ret; |
649 | ||
a698d075 | 650 | ret = __btrfs_end_transaction(trans, root, 0); |
16cdcec7 MX |
651 | if (ret) |
652 | return ret; | |
653 | return 0; | |
89ce8a63 CM |
654 | } |
655 | ||
656 | int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans, | |
657 | struct btrfs_root *root) | |
658 | { | |
16cdcec7 MX |
659 | int ret; |
660 | ||
a698d075 | 661 | ret = __btrfs_end_transaction(trans, root, 1); |
16cdcec7 MX |
662 | if (ret) |
663 | return ret; | |
664 | return 0; | |
665 | } | |
666 | ||
667 | int btrfs_end_transaction_dmeta(struct btrfs_trans_handle *trans, | |
668 | struct btrfs_root *root) | |
669 | { | |
a698d075 | 670 | return __btrfs_end_transaction(trans, root, 1); |
89ce8a63 CM |
671 | } |
672 | ||
d352ac68 CM |
673 | /* |
674 | * when btree blocks are allocated, they have some corresponding bits set for | |
675 | * them in one of two extent_io trees. This is used to make sure all of | |
690587d1 | 676 | * those extents are sent to disk but does not wait on them |
d352ac68 | 677 | */ |
690587d1 | 678 | int btrfs_write_marked_extents(struct btrfs_root *root, |
8cef4e16 | 679 | struct extent_io_tree *dirty_pages, int mark) |
79154b1b | 680 | { |
777e6bd7 | 681 | int err = 0; |
7c4452b9 | 682 | int werr = 0; |
1728366e | 683 | struct address_space *mapping = root->fs_info->btree_inode->i_mapping; |
e6138876 | 684 | struct extent_state *cached_state = NULL; |
777e6bd7 | 685 | u64 start = 0; |
5f39d397 | 686 | u64 end; |
7c4452b9 | 687 | |
1728366e | 688 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
e6138876 JB |
689 | mark, &cached_state)) { |
690 | convert_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT, | |
691 | mark, &cached_state, GFP_NOFS); | |
692 | cached_state = NULL; | |
1728366e JB |
693 | err = filemap_fdatawrite_range(mapping, start, end); |
694 | if (err) | |
695 | werr = err; | |
696 | cond_resched(); | |
697 | start = end + 1; | |
7c4452b9 | 698 | } |
690587d1 CM |
699 | if (err) |
700 | werr = err; | |
701 | return werr; | |
702 | } | |
703 | ||
704 | /* | |
705 | * when btree blocks are allocated, they have some corresponding bits set for | |
706 | * them in one of two extent_io trees. This is used to make sure all of | |
707 | * those extents are on disk for transaction or log commit. We wait | |
708 | * on all the pages and clear them from the dirty pages state tree | |
709 | */ | |
710 | int btrfs_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 711 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 | 712 | { |
690587d1 CM |
713 | int err = 0; |
714 | int werr = 0; | |
1728366e | 715 | struct address_space *mapping = root->fs_info->btree_inode->i_mapping; |
e6138876 | 716 | struct extent_state *cached_state = NULL; |
690587d1 CM |
717 | u64 start = 0; |
718 | u64 end; | |
777e6bd7 | 719 | |
1728366e | 720 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
e6138876 JB |
721 | EXTENT_NEED_WAIT, &cached_state)) { |
722 | clear_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT, | |
723 | 0, 0, &cached_state, GFP_NOFS); | |
1728366e JB |
724 | err = filemap_fdatawait_range(mapping, start, end); |
725 | if (err) | |
726 | werr = err; | |
727 | cond_resched(); | |
728 | start = end + 1; | |
777e6bd7 | 729 | } |
7c4452b9 CM |
730 | if (err) |
731 | werr = err; | |
732 | return werr; | |
79154b1b CM |
733 | } |
734 | ||
690587d1 CM |
735 | /* |
736 | * when btree blocks are allocated, they have some corresponding bits set for | |
737 | * them in one of two extent_io trees. This is used to make sure all of | |
738 | * those extents are on disk for transaction or log commit | |
739 | */ | |
740 | int btrfs_write_and_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 741 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 CM |
742 | { |
743 | int ret; | |
744 | int ret2; | |
745 | ||
8cef4e16 YZ |
746 | ret = btrfs_write_marked_extents(root, dirty_pages, mark); |
747 | ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark); | |
bf0da8c1 CM |
748 | |
749 | if (ret) | |
750 | return ret; | |
751 | if (ret2) | |
752 | return ret2; | |
753 | return 0; | |
690587d1 CM |
754 | } |
755 | ||
d0c803c4 CM |
756 | int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans, |
757 | struct btrfs_root *root) | |
758 | { | |
759 | if (!trans || !trans->transaction) { | |
760 | struct inode *btree_inode; | |
761 | btree_inode = root->fs_info->btree_inode; | |
762 | return filemap_write_and_wait(btree_inode->i_mapping); | |
763 | } | |
764 | return btrfs_write_and_wait_marked_extents(root, | |
8cef4e16 YZ |
765 | &trans->transaction->dirty_pages, |
766 | EXTENT_DIRTY); | |
d0c803c4 CM |
767 | } |
768 | ||
d352ac68 CM |
769 | /* |
770 | * this is used to update the root pointer in the tree of tree roots. | |
771 | * | |
772 | * But, in the case of the extent allocation tree, updating the root | |
773 | * pointer may allocate blocks which may change the root of the extent | |
774 | * allocation tree. | |
775 | * | |
776 | * So, this loops and repeats and makes sure the cowonly root didn't | |
777 | * change while the root pointer was being updated in the metadata. | |
778 | */ | |
0b86a832 CM |
779 | static int update_cowonly_root(struct btrfs_trans_handle *trans, |
780 | struct btrfs_root *root) | |
79154b1b CM |
781 | { |
782 | int ret; | |
0b86a832 | 783 | u64 old_root_bytenr; |
86b9f2ec | 784 | u64 old_root_used; |
0b86a832 | 785 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
79154b1b | 786 | |
86b9f2ec | 787 | old_root_used = btrfs_root_used(&root->root_item); |
0b86a832 | 788 | btrfs_write_dirty_block_groups(trans, root); |
56bec294 | 789 | |
d397712b | 790 | while (1) { |
0b86a832 | 791 | old_root_bytenr = btrfs_root_bytenr(&root->root_item); |
86b9f2ec YZ |
792 | if (old_root_bytenr == root->node->start && |
793 | old_root_used == btrfs_root_used(&root->root_item)) | |
79154b1b | 794 | break; |
87ef2bb4 | 795 | |
5d4f98a2 | 796 | btrfs_set_root_node(&root->root_item, root->node); |
79154b1b | 797 | ret = btrfs_update_root(trans, tree_root, |
0b86a832 CM |
798 | &root->root_key, |
799 | &root->root_item); | |
49b25e05 JM |
800 | if (ret) |
801 | return ret; | |
56bec294 | 802 | |
86b9f2ec | 803 | old_root_used = btrfs_root_used(&root->root_item); |
4a8c9a62 | 804 | ret = btrfs_write_dirty_block_groups(trans, root); |
49b25e05 JM |
805 | if (ret) |
806 | return ret; | |
0b86a832 | 807 | } |
276e680d YZ |
808 | |
809 | if (root != root->fs_info->extent_root) | |
810 | switch_commit_root(root); | |
811 | ||
0b86a832 CM |
812 | return 0; |
813 | } | |
814 | ||
d352ac68 CM |
815 | /* |
816 | * update all the cowonly tree roots on disk | |
49b25e05 JM |
817 | * |
818 | * The error handling in this function may not be obvious. Any of the | |
819 | * failures will cause the file system to go offline. We still need | |
820 | * to clean up the delayed refs. | |
d352ac68 | 821 | */ |
5d4f98a2 YZ |
822 | static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans, |
823 | struct btrfs_root *root) | |
0b86a832 CM |
824 | { |
825 | struct btrfs_fs_info *fs_info = root->fs_info; | |
826 | struct list_head *next; | |
84234f3a | 827 | struct extent_buffer *eb; |
56bec294 | 828 | int ret; |
84234f3a | 829 | |
56bec294 | 830 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
49b25e05 JM |
831 | if (ret) |
832 | return ret; | |
87ef2bb4 | 833 | |
84234f3a | 834 | eb = btrfs_lock_root_node(fs_info->tree_root); |
49b25e05 JM |
835 | ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, |
836 | 0, &eb); | |
84234f3a YZ |
837 | btrfs_tree_unlock(eb); |
838 | free_extent_buffer(eb); | |
0b86a832 | 839 | |
49b25e05 JM |
840 | if (ret) |
841 | return ret; | |
842 | ||
56bec294 | 843 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
49b25e05 JM |
844 | if (ret) |
845 | return ret; | |
87ef2bb4 | 846 | |
733f4fbb SB |
847 | ret = btrfs_run_dev_stats(trans, root->fs_info); |
848 | BUG_ON(ret); | |
849 | ||
546adb0d JS |
850 | ret = btrfs_run_qgroups(trans, root->fs_info); |
851 | BUG_ON(ret); | |
852 | ||
853 | /* run_qgroups might have added some more refs */ | |
854 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); | |
855 | BUG_ON(ret); | |
856 | ||
d397712b | 857 | while (!list_empty(&fs_info->dirty_cowonly_roots)) { |
0b86a832 CM |
858 | next = fs_info->dirty_cowonly_roots.next; |
859 | list_del_init(next); | |
860 | root = list_entry(next, struct btrfs_root, dirty_list); | |
87ef2bb4 | 861 | |
49b25e05 JM |
862 | ret = update_cowonly_root(trans, root); |
863 | if (ret) | |
864 | return ret; | |
79154b1b | 865 | } |
276e680d YZ |
866 | |
867 | down_write(&fs_info->extent_commit_sem); | |
868 | switch_commit_root(fs_info->extent_root); | |
869 | up_write(&fs_info->extent_commit_sem); | |
870 | ||
79154b1b CM |
871 | return 0; |
872 | } | |
873 | ||
d352ac68 CM |
874 | /* |
875 | * dead roots are old snapshots that need to be deleted. This allocates | |
876 | * a dirty root struct and adds it into the list of dead roots that need to | |
877 | * be deleted | |
878 | */ | |
5d4f98a2 | 879 | int btrfs_add_dead_root(struct btrfs_root *root) |
5eda7b5e | 880 | { |
a4abeea4 | 881 | spin_lock(&root->fs_info->trans_lock); |
5d4f98a2 | 882 | list_add(&root->root_list, &root->fs_info->dead_roots); |
a4abeea4 | 883 | spin_unlock(&root->fs_info->trans_lock); |
5eda7b5e CM |
884 | return 0; |
885 | } | |
886 | ||
d352ac68 | 887 | /* |
5d4f98a2 | 888 | * update all the cowonly tree roots on disk |
d352ac68 | 889 | */ |
5d4f98a2 YZ |
890 | static noinline int commit_fs_roots(struct btrfs_trans_handle *trans, |
891 | struct btrfs_root *root) | |
0f7d52f4 | 892 | { |
0f7d52f4 | 893 | struct btrfs_root *gang[8]; |
5d4f98a2 | 894 | struct btrfs_fs_info *fs_info = root->fs_info; |
0f7d52f4 CM |
895 | int i; |
896 | int ret; | |
54aa1f4d CM |
897 | int err = 0; |
898 | ||
a4abeea4 | 899 | spin_lock(&fs_info->fs_roots_radix_lock); |
d397712b | 900 | while (1) { |
5d4f98a2 YZ |
901 | ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, |
902 | (void **)gang, 0, | |
0f7d52f4 CM |
903 | ARRAY_SIZE(gang), |
904 | BTRFS_ROOT_TRANS_TAG); | |
905 | if (ret == 0) | |
906 | break; | |
907 | for (i = 0; i < ret; i++) { | |
908 | root = gang[i]; | |
5d4f98a2 YZ |
909 | radix_tree_tag_clear(&fs_info->fs_roots_radix, |
910 | (unsigned long)root->root_key.objectid, | |
911 | BTRFS_ROOT_TRANS_TAG); | |
a4abeea4 | 912 | spin_unlock(&fs_info->fs_roots_radix_lock); |
31153d81 | 913 | |
e02119d5 | 914 | btrfs_free_log(trans, root); |
5d4f98a2 | 915 | btrfs_update_reloc_root(trans, root); |
d68fc57b | 916 | btrfs_orphan_commit_root(trans, root); |
bcc63abb | 917 | |
82d5902d LZ |
918 | btrfs_save_ino_cache(root, trans); |
919 | ||
f1ebcc74 LB |
920 | /* see comments in should_cow_block() */ |
921 | root->force_cow = 0; | |
922 | smp_wmb(); | |
923 | ||
978d910d | 924 | if (root->commit_root != root->node) { |
581bb050 | 925 | mutex_lock(&root->fs_commit_mutex); |
817d52f8 | 926 | switch_commit_root(root); |
581bb050 LZ |
927 | btrfs_unpin_free_ino(root); |
928 | mutex_unlock(&root->fs_commit_mutex); | |
929 | ||
978d910d YZ |
930 | btrfs_set_root_node(&root->root_item, |
931 | root->node); | |
932 | } | |
5d4f98a2 | 933 | |
5d4f98a2 | 934 | err = btrfs_update_root(trans, fs_info->tree_root, |
0f7d52f4 CM |
935 | &root->root_key, |
936 | &root->root_item); | |
a4abeea4 | 937 | spin_lock(&fs_info->fs_roots_radix_lock); |
54aa1f4d CM |
938 | if (err) |
939 | break; | |
0f7d52f4 CM |
940 | } |
941 | } | |
a4abeea4 | 942 | spin_unlock(&fs_info->fs_roots_radix_lock); |
54aa1f4d | 943 | return err; |
0f7d52f4 CM |
944 | } |
945 | ||
d352ac68 CM |
946 | /* |
947 | * defrag a given btree. If cacheonly == 1, this won't read from the disk, | |
948 | * otherwise every leaf in the btree is read and defragged. | |
949 | */ | |
e9d0b13b CM |
950 | int btrfs_defrag_root(struct btrfs_root *root, int cacheonly) |
951 | { | |
952 | struct btrfs_fs_info *info = root->fs_info; | |
e9d0b13b | 953 | struct btrfs_trans_handle *trans; |
8929ecfa | 954 | int ret; |
d3c2fdcf | 955 | unsigned long nr; |
e9d0b13b | 956 | |
8929ecfa | 957 | if (xchg(&root->defrag_running, 1)) |
e9d0b13b | 958 | return 0; |
8929ecfa | 959 | |
6b80053d | 960 | while (1) { |
8929ecfa YZ |
961 | trans = btrfs_start_transaction(root, 0); |
962 | if (IS_ERR(trans)) | |
963 | return PTR_ERR(trans); | |
964 | ||
e9d0b13b | 965 | ret = btrfs_defrag_leaves(trans, root, cacheonly); |
8929ecfa | 966 | |
d3c2fdcf | 967 | nr = trans->blocks_used; |
e9d0b13b | 968 | btrfs_end_transaction(trans, root); |
d3c2fdcf | 969 | btrfs_btree_balance_dirty(info->tree_root, nr); |
e9d0b13b CM |
970 | cond_resched(); |
971 | ||
7841cb28 | 972 | if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN) |
e9d0b13b CM |
973 | break; |
974 | } | |
975 | root->defrag_running = 0; | |
8929ecfa | 976 | return ret; |
e9d0b13b CM |
977 | } |
978 | ||
d352ac68 CM |
979 | /* |
980 | * new snapshots need to be created at a very specific time in the | |
981 | * transaction commit. This does the actual creation | |
982 | */ | |
80b6794d | 983 | static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, |
3063d29f CM |
984 | struct btrfs_fs_info *fs_info, |
985 | struct btrfs_pending_snapshot *pending) | |
986 | { | |
987 | struct btrfs_key key; | |
80b6794d | 988 | struct btrfs_root_item *new_root_item; |
3063d29f CM |
989 | struct btrfs_root *tree_root = fs_info->tree_root; |
990 | struct btrfs_root *root = pending->root; | |
6bdb72de | 991 | struct btrfs_root *parent_root; |
98c9942a | 992 | struct btrfs_block_rsv *rsv; |
6bdb72de | 993 | struct inode *parent_inode; |
42874b3d MX |
994 | struct btrfs_path *path; |
995 | struct btrfs_dir_item *dir_item; | |
6a912213 | 996 | struct dentry *parent; |
a22285a6 | 997 | struct dentry *dentry; |
3063d29f | 998 | struct extent_buffer *tmp; |
925baedd | 999 | struct extent_buffer *old; |
8ea05e3a | 1000 | struct timespec cur_time = CURRENT_TIME; |
3063d29f | 1001 | int ret; |
d68fc57b | 1002 | u64 to_reserve = 0; |
6bdb72de | 1003 | u64 index = 0; |
a22285a6 | 1004 | u64 objectid; |
b83cc969 | 1005 | u64 root_flags; |
8ea05e3a | 1006 | uuid_le new_uuid; |
3063d29f | 1007 | |
42874b3d MX |
1008 | path = btrfs_alloc_path(); |
1009 | if (!path) { | |
1010 | ret = pending->error = -ENOMEM; | |
1011 | goto path_alloc_fail; | |
1012 | } | |
1013 | ||
80b6794d CM |
1014 | new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS); |
1015 | if (!new_root_item) { | |
49b25e05 | 1016 | ret = pending->error = -ENOMEM; |
6fa9700e | 1017 | goto root_item_alloc_fail; |
80b6794d | 1018 | } |
a22285a6 | 1019 | |
581bb050 | 1020 | ret = btrfs_find_free_objectid(tree_root, &objectid); |
a22285a6 YZ |
1021 | if (ret) { |
1022 | pending->error = ret; | |
6fa9700e | 1023 | goto no_free_objectid; |
a22285a6 | 1024 | } |
3063d29f | 1025 | |
3fd0a558 | 1026 | btrfs_reloc_pre_snapshot(trans, pending, &to_reserve); |
d68fc57b YZ |
1027 | |
1028 | if (to_reserve > 0) { | |
08e007d2 MX |
1029 | ret = btrfs_block_rsv_add(root, &pending->block_rsv, |
1030 | to_reserve, | |
1031 | BTRFS_RESERVE_NO_FLUSH); | |
d68fc57b YZ |
1032 | if (ret) { |
1033 | pending->error = ret; | |
6fa9700e | 1034 | goto no_free_objectid; |
d68fc57b YZ |
1035 | } |
1036 | } | |
1037 | ||
6f72c7e2 AJ |
1038 | ret = btrfs_qgroup_inherit(trans, fs_info, root->root_key.objectid, |
1039 | objectid, pending->inherit); | |
6f72c7e2 AJ |
1040 | if (ret) { |
1041 | pending->error = ret; | |
6fa9700e | 1042 | goto no_free_objectid; |
6f72c7e2 AJ |
1043 | } |
1044 | ||
3063d29f | 1045 | key.objectid = objectid; |
a22285a6 YZ |
1046 | key.offset = (u64)-1; |
1047 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3063d29f | 1048 | |
6fa9700e | 1049 | rsv = trans->block_rsv; |
a22285a6 | 1050 | trans->block_rsv = &pending->block_rsv; |
3de4586c | 1051 | |
a22285a6 | 1052 | dentry = pending->dentry; |
6a912213 JB |
1053 | parent = dget_parent(dentry); |
1054 | parent_inode = parent->d_inode; | |
a22285a6 | 1055 | parent_root = BTRFS_I(parent_inode)->root; |
7585717f | 1056 | record_root_in_trans(trans, parent_root); |
a22285a6 | 1057 | |
3063d29f CM |
1058 | /* |
1059 | * insert the directory item | |
1060 | */ | |
3de4586c | 1061 | ret = btrfs_set_inode_index(parent_inode, &index); |
49b25e05 | 1062 | BUG_ON(ret); /* -ENOMEM */ |
42874b3d MX |
1063 | |
1064 | /* check if there is a file/dir which has the same name. */ | |
1065 | dir_item = btrfs_lookup_dir_item(NULL, parent_root, path, | |
1066 | btrfs_ino(parent_inode), | |
1067 | dentry->d_name.name, | |
1068 | dentry->d_name.len, 0); | |
1069 | if (dir_item != NULL && !IS_ERR(dir_item)) { | |
fe66a05a | 1070 | pending->error = -EEXIST; |
fe66a05a | 1071 | goto fail; |
42874b3d MX |
1072 | } else if (IS_ERR(dir_item)) { |
1073 | ret = PTR_ERR(dir_item); | |
8732d44f MX |
1074 | btrfs_abort_transaction(trans, root, ret); |
1075 | goto fail; | |
79787eaa | 1076 | } |
42874b3d | 1077 | btrfs_release_path(path); |
52c26179 | 1078 | |
e999376f CM |
1079 | /* |
1080 | * pull in the delayed directory update | |
1081 | * and the delayed inode item | |
1082 | * otherwise we corrupt the FS during | |
1083 | * snapshot | |
1084 | */ | |
1085 | ret = btrfs_run_delayed_items(trans, root); | |
8732d44f MX |
1086 | if (ret) { /* Transaction aborted */ |
1087 | btrfs_abort_transaction(trans, root, ret); | |
1088 | goto fail; | |
1089 | } | |
e999376f | 1090 | |
7585717f | 1091 | record_root_in_trans(trans, root); |
6bdb72de SW |
1092 | btrfs_set_root_last_snapshot(&root->root_item, trans->transid); |
1093 | memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); | |
08fe4db1 | 1094 | btrfs_check_and_init_root_item(new_root_item); |
6bdb72de | 1095 | |
b83cc969 LZ |
1096 | root_flags = btrfs_root_flags(new_root_item); |
1097 | if (pending->readonly) | |
1098 | root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; | |
1099 | else | |
1100 | root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; | |
1101 | btrfs_set_root_flags(new_root_item, root_flags); | |
1102 | ||
8ea05e3a AB |
1103 | btrfs_set_root_generation_v2(new_root_item, |
1104 | trans->transid); | |
1105 | uuid_le_gen(&new_uuid); | |
1106 | memcpy(new_root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE); | |
1107 | memcpy(new_root_item->parent_uuid, root->root_item.uuid, | |
1108 | BTRFS_UUID_SIZE); | |
1109 | new_root_item->otime.sec = cpu_to_le64(cur_time.tv_sec); | |
dadd1105 | 1110 | new_root_item->otime.nsec = cpu_to_le32(cur_time.tv_nsec); |
8ea05e3a AB |
1111 | btrfs_set_root_otransid(new_root_item, trans->transid); |
1112 | memset(&new_root_item->stime, 0, sizeof(new_root_item->stime)); | |
1113 | memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime)); | |
1114 | btrfs_set_root_stransid(new_root_item, 0); | |
1115 | btrfs_set_root_rtransid(new_root_item, 0); | |
1116 | ||
6bdb72de | 1117 | old = btrfs_lock_root_node(root); |
49b25e05 | 1118 | ret = btrfs_cow_block(trans, root, old, NULL, 0, &old); |
79787eaa JM |
1119 | if (ret) { |
1120 | btrfs_tree_unlock(old); | |
1121 | free_extent_buffer(old); | |
8732d44f MX |
1122 | btrfs_abort_transaction(trans, root, ret); |
1123 | goto fail; | |
79787eaa | 1124 | } |
49b25e05 | 1125 | |
6bdb72de SW |
1126 | btrfs_set_lock_blocking(old); |
1127 | ||
49b25e05 | 1128 | ret = btrfs_copy_root(trans, root, old, &tmp, objectid); |
79787eaa | 1129 | /* clean up in any case */ |
6bdb72de SW |
1130 | btrfs_tree_unlock(old); |
1131 | free_extent_buffer(old); | |
8732d44f MX |
1132 | if (ret) { |
1133 | btrfs_abort_transaction(trans, root, ret); | |
1134 | goto fail; | |
1135 | } | |
6bdb72de | 1136 | |
f1ebcc74 LB |
1137 | /* see comments in should_cow_block() */ |
1138 | root->force_cow = 1; | |
1139 | smp_wmb(); | |
1140 | ||
6bdb72de | 1141 | btrfs_set_root_node(new_root_item, tmp); |
a22285a6 YZ |
1142 | /* record when the snapshot was created in key.offset */ |
1143 | key.offset = trans->transid; | |
1144 | ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); | |
6bdb72de SW |
1145 | btrfs_tree_unlock(tmp); |
1146 | free_extent_buffer(tmp); | |
8732d44f MX |
1147 | if (ret) { |
1148 | btrfs_abort_transaction(trans, root, ret); | |
1149 | goto fail; | |
1150 | } | |
6bdb72de | 1151 | |
a22285a6 YZ |
1152 | /* |
1153 | * insert root back/forward references | |
1154 | */ | |
1155 | ret = btrfs_add_root_ref(trans, tree_root, objectid, | |
0660b5af | 1156 | parent_root->root_key.objectid, |
33345d01 | 1157 | btrfs_ino(parent_inode), index, |
a22285a6 | 1158 | dentry->d_name.name, dentry->d_name.len); |
8732d44f MX |
1159 | if (ret) { |
1160 | btrfs_abort_transaction(trans, root, ret); | |
1161 | goto fail; | |
1162 | } | |
0660b5af | 1163 | |
a22285a6 YZ |
1164 | key.offset = (u64)-1; |
1165 | pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key); | |
79787eaa JM |
1166 | if (IS_ERR(pending->snap)) { |
1167 | ret = PTR_ERR(pending->snap); | |
8732d44f MX |
1168 | btrfs_abort_transaction(trans, root, ret); |
1169 | goto fail; | |
79787eaa | 1170 | } |
d68fc57b | 1171 | |
49b25e05 | 1172 | ret = btrfs_reloc_post_snapshot(trans, pending); |
8732d44f MX |
1173 | if (ret) { |
1174 | btrfs_abort_transaction(trans, root, ret); | |
1175 | goto fail; | |
1176 | } | |
361048f5 MX |
1177 | |
1178 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); | |
8732d44f MX |
1179 | if (ret) { |
1180 | btrfs_abort_transaction(trans, root, ret); | |
1181 | goto fail; | |
1182 | } | |
42874b3d MX |
1183 | |
1184 | ret = btrfs_insert_dir_item(trans, parent_root, | |
1185 | dentry->d_name.name, dentry->d_name.len, | |
1186 | parent_inode, &key, | |
1187 | BTRFS_FT_DIR, index); | |
1188 | /* We have check then name at the beginning, so it is impossible. */ | |
1189 | BUG_ON(ret == -EEXIST); | |
8732d44f MX |
1190 | if (ret) { |
1191 | btrfs_abort_transaction(trans, root, ret); | |
1192 | goto fail; | |
1193 | } | |
42874b3d MX |
1194 | |
1195 | btrfs_i_size_write(parent_inode, parent_inode->i_size + | |
1196 | dentry->d_name.len * 2); | |
1197 | parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME; | |
be6aef60 | 1198 | ret = btrfs_update_inode_fallback(trans, parent_root, parent_inode); |
42874b3d | 1199 | if (ret) |
8732d44f | 1200 | btrfs_abort_transaction(trans, root, ret); |
3063d29f | 1201 | fail: |
6fa9700e | 1202 | dput(parent); |
98c9942a | 1203 | trans->block_rsv = rsv; |
6fa9700e MX |
1204 | no_free_objectid: |
1205 | kfree(new_root_item); | |
1206 | root_item_alloc_fail: | |
42874b3d MX |
1207 | btrfs_free_path(path); |
1208 | path_alloc_fail: | |
a22285a6 | 1209 | btrfs_block_rsv_release(root, &pending->block_rsv, (u64)-1); |
49b25e05 | 1210 | return ret; |
3063d29f CM |
1211 | } |
1212 | ||
d352ac68 CM |
1213 | /* |
1214 | * create all the snapshots we've scheduled for creation | |
1215 | */ | |
80b6794d CM |
1216 | static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans, |
1217 | struct btrfs_fs_info *fs_info) | |
3de4586c CM |
1218 | { |
1219 | struct btrfs_pending_snapshot *pending; | |
1220 | struct list_head *head = &trans->transaction->pending_snapshots; | |
3de4586c | 1221 | |
fe66a05a CM |
1222 | list_for_each_entry(pending, head, list) |
1223 | create_pending_snapshot(trans, fs_info, pending); | |
3de4586c CM |
1224 | return 0; |
1225 | } | |
1226 | ||
5d4f98a2 YZ |
1227 | static void update_super_roots(struct btrfs_root *root) |
1228 | { | |
1229 | struct btrfs_root_item *root_item; | |
1230 | struct btrfs_super_block *super; | |
1231 | ||
6c41761f | 1232 | super = root->fs_info->super_copy; |
5d4f98a2 YZ |
1233 | |
1234 | root_item = &root->fs_info->chunk_root->root_item; | |
1235 | super->chunk_root = root_item->bytenr; | |
1236 | super->chunk_root_generation = root_item->generation; | |
1237 | super->chunk_root_level = root_item->level; | |
1238 | ||
1239 | root_item = &root->fs_info->tree_root->root_item; | |
1240 | super->root = root_item->bytenr; | |
1241 | super->generation = root_item->generation; | |
1242 | super->root_level = root_item->level; | |
73bc1876 | 1243 | if (btrfs_test_opt(root, SPACE_CACHE)) |
0af3d00b | 1244 | super->cache_generation = root_item->generation; |
5d4f98a2 YZ |
1245 | } |
1246 | ||
f36f3042 CM |
1247 | int btrfs_transaction_in_commit(struct btrfs_fs_info *info) |
1248 | { | |
1249 | int ret = 0; | |
a4abeea4 | 1250 | spin_lock(&info->trans_lock); |
f36f3042 CM |
1251 | if (info->running_transaction) |
1252 | ret = info->running_transaction->in_commit; | |
a4abeea4 | 1253 | spin_unlock(&info->trans_lock); |
f36f3042 CM |
1254 | return ret; |
1255 | } | |
1256 | ||
8929ecfa YZ |
1257 | int btrfs_transaction_blocked(struct btrfs_fs_info *info) |
1258 | { | |
1259 | int ret = 0; | |
a4abeea4 | 1260 | spin_lock(&info->trans_lock); |
8929ecfa YZ |
1261 | if (info->running_transaction) |
1262 | ret = info->running_transaction->blocked; | |
a4abeea4 | 1263 | spin_unlock(&info->trans_lock); |
8929ecfa YZ |
1264 | return ret; |
1265 | } | |
1266 | ||
bb9c12c9 SW |
1267 | /* |
1268 | * wait for the current transaction commit to start and block subsequent | |
1269 | * transaction joins | |
1270 | */ | |
1271 | static void wait_current_trans_commit_start(struct btrfs_root *root, | |
1272 | struct btrfs_transaction *trans) | |
1273 | { | |
72d63ed6 | 1274 | wait_event(root->fs_info->transaction_blocked_wait, trans->in_commit); |
bb9c12c9 SW |
1275 | } |
1276 | ||
1277 | /* | |
1278 | * wait for the current transaction to start and then become unblocked. | |
1279 | * caller holds ref. | |
1280 | */ | |
1281 | static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root, | |
1282 | struct btrfs_transaction *trans) | |
1283 | { | |
72d63ed6 LZ |
1284 | wait_event(root->fs_info->transaction_wait, |
1285 | trans->commit_done || (trans->in_commit && !trans->blocked)); | |
bb9c12c9 SW |
1286 | } |
1287 | ||
1288 | /* | |
1289 | * commit transactions asynchronously. once btrfs_commit_transaction_async | |
1290 | * returns, any subsequent transaction will not be allowed to join. | |
1291 | */ | |
1292 | struct btrfs_async_commit { | |
1293 | struct btrfs_trans_handle *newtrans; | |
1294 | struct btrfs_root *root; | |
1295 | struct delayed_work work; | |
1296 | }; | |
1297 | ||
1298 | static void do_async_commit(struct work_struct *work) | |
1299 | { | |
1300 | struct btrfs_async_commit *ac = | |
1301 | container_of(work, struct btrfs_async_commit, work.work); | |
1302 | ||
6fc4e354 SW |
1303 | /* |
1304 | * We've got freeze protection passed with the transaction. | |
1305 | * Tell lockdep about it. | |
1306 | */ | |
1307 | rwsem_acquire_read( | |
1308 | &ac->root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1], | |
1309 | 0, 1, _THIS_IP_); | |
1310 | ||
e209db7a SW |
1311 | current->journal_info = ac->newtrans; |
1312 | ||
bb9c12c9 SW |
1313 | btrfs_commit_transaction(ac->newtrans, ac->root); |
1314 | kfree(ac); | |
1315 | } | |
1316 | ||
1317 | int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans, | |
1318 | struct btrfs_root *root, | |
1319 | int wait_for_unblock) | |
1320 | { | |
1321 | struct btrfs_async_commit *ac; | |
1322 | struct btrfs_transaction *cur_trans; | |
1323 | ||
1324 | ac = kmalloc(sizeof(*ac), GFP_NOFS); | |
db5b493a TI |
1325 | if (!ac) |
1326 | return -ENOMEM; | |
bb9c12c9 SW |
1327 | |
1328 | INIT_DELAYED_WORK(&ac->work, do_async_commit); | |
1329 | ac->root = root; | |
7a7eaa40 | 1330 | ac->newtrans = btrfs_join_transaction(root); |
3612b495 TI |
1331 | if (IS_ERR(ac->newtrans)) { |
1332 | int err = PTR_ERR(ac->newtrans); | |
1333 | kfree(ac); | |
1334 | return err; | |
1335 | } | |
bb9c12c9 SW |
1336 | |
1337 | /* take transaction reference */ | |
bb9c12c9 | 1338 | cur_trans = trans->transaction; |
13c5a93e | 1339 | atomic_inc(&cur_trans->use_count); |
bb9c12c9 SW |
1340 | |
1341 | btrfs_end_transaction(trans, root); | |
6fc4e354 SW |
1342 | |
1343 | /* | |
1344 | * Tell lockdep we've released the freeze rwsem, since the | |
1345 | * async commit thread will be the one to unlock it. | |
1346 | */ | |
1347 | rwsem_release(&root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1], | |
1348 | 1, _THIS_IP_); | |
1349 | ||
bb9c12c9 SW |
1350 | schedule_delayed_work(&ac->work, 0); |
1351 | ||
1352 | /* wait for transaction to start and unblock */ | |
bb9c12c9 SW |
1353 | if (wait_for_unblock) |
1354 | wait_current_trans_commit_start_and_unblock(root, cur_trans); | |
1355 | else | |
1356 | wait_current_trans_commit_start(root, cur_trans); | |
bb9c12c9 | 1357 | |
38e88054 SW |
1358 | if (current->journal_info == trans) |
1359 | current->journal_info = NULL; | |
1360 | ||
1361 | put_transaction(cur_trans); | |
bb9c12c9 SW |
1362 | return 0; |
1363 | } | |
1364 | ||
49b25e05 JM |
1365 | |
1366 | static void cleanup_transaction(struct btrfs_trans_handle *trans, | |
7b8b92af | 1367 | struct btrfs_root *root, int err) |
49b25e05 JM |
1368 | { |
1369 | struct btrfs_transaction *cur_trans = trans->transaction; | |
1370 | ||
1371 | WARN_ON(trans->use_count > 1); | |
1372 | ||
7b8b92af JB |
1373 | btrfs_abort_transaction(trans, root, err); |
1374 | ||
49b25e05 JM |
1375 | spin_lock(&root->fs_info->trans_lock); |
1376 | list_del_init(&cur_trans->list); | |
d7096fc3 JB |
1377 | if (cur_trans == root->fs_info->running_transaction) { |
1378 | root->fs_info->running_transaction = NULL; | |
1379 | root->fs_info->trans_no_join = 0; | |
1380 | } | |
49b25e05 JM |
1381 | spin_unlock(&root->fs_info->trans_lock); |
1382 | ||
1383 | btrfs_cleanup_one_transaction(trans->transaction, root); | |
1384 | ||
1385 | put_transaction(cur_trans); | |
1386 | put_transaction(cur_trans); | |
1387 | ||
1388 | trace_btrfs_transaction_commit(root); | |
1389 | ||
1390 | btrfs_scrub_continue(root); | |
1391 | ||
1392 | if (current->journal_info == trans) | |
1393 | current->journal_info = NULL; | |
1394 | ||
1395 | kmem_cache_free(btrfs_trans_handle_cachep, trans); | |
1396 | } | |
1397 | ||
ca469637 MX |
1398 | static int btrfs_flush_all_pending_stuffs(struct btrfs_trans_handle *trans, |
1399 | struct btrfs_root *root) | |
1400 | { | |
1401 | int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT); | |
1402 | int snap_pending = 0; | |
1403 | int ret; | |
1404 | ||
1405 | if (!flush_on_commit) { | |
1406 | spin_lock(&root->fs_info->trans_lock); | |
1407 | if (!list_empty(&trans->transaction->pending_snapshots)) | |
1408 | snap_pending = 1; | |
1409 | spin_unlock(&root->fs_info->trans_lock); | |
1410 | } | |
1411 | ||
1412 | if (flush_on_commit || snap_pending) { | |
1413 | btrfs_start_delalloc_inodes(root, 1); | |
1414 | btrfs_wait_ordered_extents(root, 1); | |
1415 | } | |
1416 | ||
1417 | ret = btrfs_run_delayed_items(trans, root); | |
1418 | if (ret) | |
1419 | return ret; | |
1420 | ||
1421 | /* | |
1422 | * running the delayed items may have added new refs. account | |
1423 | * them now so that they hinder processing of more delayed refs | |
1424 | * as little as possible. | |
1425 | */ | |
1426 | btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info); | |
1427 | ||
1428 | /* | |
1429 | * rename don't use btrfs_join_transaction, so, once we | |
1430 | * set the transaction to blocked above, we aren't going | |
1431 | * to get any new ordered operations. We can safely run | |
1432 | * it here and no for sure that nothing new will be added | |
1433 | * to the list | |
1434 | */ | |
1435 | btrfs_run_ordered_operations(root, 1); | |
1436 | ||
1437 | return 0; | |
1438 | } | |
1439 | ||
bb9c12c9 SW |
1440 | /* |
1441 | * btrfs_transaction state sequence: | |
1442 | * in_commit = 0, blocked = 0 (initial) | |
1443 | * in_commit = 1, blocked = 1 | |
1444 | * blocked = 0 | |
1445 | * commit_done = 1 | |
1446 | */ | |
79154b1b CM |
1447 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans, |
1448 | struct btrfs_root *root) | |
1449 | { | |
15ee9bc7 | 1450 | unsigned long joined = 0; |
49b25e05 | 1451 | struct btrfs_transaction *cur_trans = trans->transaction; |
8fd17795 | 1452 | struct btrfs_transaction *prev_trans = NULL; |
79154b1b | 1453 | DEFINE_WAIT(wait); |
25287e0a | 1454 | int ret; |
89573b9c CM |
1455 | int should_grow = 0; |
1456 | unsigned long now = get_seconds(); | |
79154b1b | 1457 | |
25287e0a MX |
1458 | ret = btrfs_run_ordered_operations(root, 0); |
1459 | if (ret) { | |
1460 | btrfs_abort_transaction(trans, root, ret); | |
1461 | goto cleanup_transaction; | |
1462 | } | |
5a3f23d5 | 1463 | |
25287e0a MX |
1464 | if (cur_trans->aborted) { |
1465 | ret = cur_trans->aborted; | |
49b25e05 | 1466 | goto cleanup_transaction; |
25287e0a | 1467 | } |
49b25e05 | 1468 | |
56bec294 CM |
1469 | /* make a pass through all the delayed refs we have so far |
1470 | * any runnings procs may add more while we are here | |
1471 | */ | |
1472 | ret = btrfs_run_delayed_refs(trans, root, 0); | |
49b25e05 JM |
1473 | if (ret) |
1474 | goto cleanup_transaction; | |
56bec294 | 1475 | |
0e721106 JB |
1476 | btrfs_trans_release_metadata(trans, root); |
1477 | trans->block_rsv = NULL; | |
1478 | ||
b7ec40d7 | 1479 | cur_trans = trans->transaction; |
49b25e05 | 1480 | |
56bec294 CM |
1481 | /* |
1482 | * set the flushing flag so procs in this transaction have to | |
1483 | * start sending their work down. | |
1484 | */ | |
b7ec40d7 | 1485 | cur_trans->delayed_refs.flushing = 1; |
56bec294 | 1486 | |
ea658bad JB |
1487 | if (!list_empty(&trans->new_bgs)) |
1488 | btrfs_create_pending_block_groups(trans, root); | |
1489 | ||
c3e69d58 | 1490 | ret = btrfs_run_delayed_refs(trans, root, 0); |
49b25e05 JM |
1491 | if (ret) |
1492 | goto cleanup_transaction; | |
56bec294 | 1493 | |
a4abeea4 | 1494 | spin_lock(&cur_trans->commit_lock); |
b7ec40d7 | 1495 | if (cur_trans->in_commit) { |
a4abeea4 | 1496 | spin_unlock(&cur_trans->commit_lock); |
13c5a93e | 1497 | atomic_inc(&cur_trans->use_count); |
49b25e05 | 1498 | ret = btrfs_end_transaction(trans, root); |
ccd467d6 | 1499 | |
b9c8300c | 1500 | wait_for_commit(root, cur_trans); |
15ee9bc7 | 1501 | |
79154b1b | 1502 | put_transaction(cur_trans); |
15ee9bc7 | 1503 | |
49b25e05 | 1504 | return ret; |
79154b1b | 1505 | } |
4313b399 | 1506 | |
2c90e5d6 | 1507 | trans->transaction->in_commit = 1; |
f9295749 | 1508 | trans->transaction->blocked = 1; |
a4abeea4 | 1509 | spin_unlock(&cur_trans->commit_lock); |
bb9c12c9 SW |
1510 | wake_up(&root->fs_info->transaction_blocked_wait); |
1511 | ||
a4abeea4 | 1512 | spin_lock(&root->fs_info->trans_lock); |
ccd467d6 CM |
1513 | if (cur_trans->list.prev != &root->fs_info->trans_list) { |
1514 | prev_trans = list_entry(cur_trans->list.prev, | |
1515 | struct btrfs_transaction, list); | |
1516 | if (!prev_trans->commit_done) { | |
13c5a93e | 1517 | atomic_inc(&prev_trans->use_count); |
a4abeea4 | 1518 | spin_unlock(&root->fs_info->trans_lock); |
ccd467d6 CM |
1519 | |
1520 | wait_for_commit(root, prev_trans); | |
ccd467d6 | 1521 | |
15ee9bc7 | 1522 | put_transaction(prev_trans); |
a4abeea4 JB |
1523 | } else { |
1524 | spin_unlock(&root->fs_info->trans_lock); | |
ccd467d6 | 1525 | } |
a4abeea4 JB |
1526 | } else { |
1527 | spin_unlock(&root->fs_info->trans_lock); | |
ccd467d6 | 1528 | } |
15ee9bc7 | 1529 | |
e39e64ac CM |
1530 | if (!btrfs_test_opt(root, SSD) && |
1531 | (now < cur_trans->start_time || now - cur_trans->start_time < 1)) | |
89573b9c CM |
1532 | should_grow = 1; |
1533 | ||
15ee9bc7 JB |
1534 | do { |
1535 | joined = cur_trans->num_joined; | |
7ea394f1 | 1536 | |
2c90e5d6 | 1537 | WARN_ON(cur_trans != trans->transaction); |
15ee9bc7 | 1538 | |
ca469637 | 1539 | ret = btrfs_flush_all_pending_stuffs(trans, root); |
49b25e05 JM |
1540 | if (ret) |
1541 | goto cleanup_transaction; | |
16cdcec7 | 1542 | |
ed3b3d31 CM |
1543 | prepare_to_wait(&cur_trans->writer_wait, &wait, |
1544 | TASK_UNINTERRUPTIBLE); | |
1545 | ||
13c5a93e | 1546 | if (atomic_read(&cur_trans->num_writers) > 1) |
99d16cbc SW |
1547 | schedule_timeout(MAX_SCHEDULE_TIMEOUT); |
1548 | else if (should_grow) | |
1549 | schedule_timeout(1); | |
15ee9bc7 | 1550 | |
15ee9bc7 | 1551 | finish_wait(&cur_trans->writer_wait, &wait); |
13c5a93e | 1552 | } while (atomic_read(&cur_trans->num_writers) > 1 || |
89573b9c | 1553 | (should_grow && cur_trans->num_joined != joined)); |
15ee9bc7 | 1554 | |
ca469637 MX |
1555 | ret = btrfs_flush_all_pending_stuffs(trans, root); |
1556 | if (ret) | |
1557 | goto cleanup_transaction; | |
1558 | ||
ed0ca140 JB |
1559 | /* |
1560 | * Ok now we need to make sure to block out any other joins while we | |
1561 | * commit the transaction. We could have started a join before setting | |
1562 | * no_join so make sure to wait for num_writers to == 1 again. | |
1563 | */ | |
1564 | spin_lock(&root->fs_info->trans_lock); | |
1565 | root->fs_info->trans_no_join = 1; | |
1566 | spin_unlock(&root->fs_info->trans_lock); | |
1567 | wait_event(cur_trans->writer_wait, | |
1568 | atomic_read(&cur_trans->num_writers) == 1); | |
1569 | ||
7585717f CM |
1570 | /* |
1571 | * the reloc mutex makes sure that we stop | |
1572 | * the balancing code from coming in and moving | |
1573 | * extents around in the middle of the commit | |
1574 | */ | |
1575 | mutex_lock(&root->fs_info->reloc_mutex); | |
1576 | ||
42874b3d MX |
1577 | /* |
1578 | * We needn't worry about the delayed items because we will | |
1579 | * deal with them in create_pending_snapshot(), which is the | |
1580 | * core function of the snapshot creation. | |
1581 | */ | |
1582 | ret = create_pending_snapshots(trans, root->fs_info); | |
49b25e05 JM |
1583 | if (ret) { |
1584 | mutex_unlock(&root->fs_info->reloc_mutex); | |
1585 | goto cleanup_transaction; | |
1586 | } | |
3063d29f | 1587 | |
42874b3d MX |
1588 | /* |
1589 | * We insert the dir indexes of the snapshots and update the inode | |
1590 | * of the snapshots' parents after the snapshot creation, so there | |
1591 | * are some delayed items which are not dealt with. Now deal with | |
1592 | * them. | |
1593 | * | |
1594 | * We needn't worry that this operation will corrupt the snapshots, | |
1595 | * because all the tree which are snapshoted will be forced to COW | |
1596 | * the nodes and leaves. | |
1597 | */ | |
1598 | ret = btrfs_run_delayed_items(trans, root); | |
49b25e05 JM |
1599 | if (ret) { |
1600 | mutex_unlock(&root->fs_info->reloc_mutex); | |
1601 | goto cleanup_transaction; | |
1602 | } | |
16cdcec7 | 1603 | |
56bec294 | 1604 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
49b25e05 JM |
1605 | if (ret) { |
1606 | mutex_unlock(&root->fs_info->reloc_mutex); | |
1607 | goto cleanup_transaction; | |
1608 | } | |
56bec294 | 1609 | |
e999376f CM |
1610 | /* |
1611 | * make sure none of the code above managed to slip in a | |
1612 | * delayed item | |
1613 | */ | |
1614 | btrfs_assert_delayed_root_empty(root); | |
1615 | ||
2c90e5d6 | 1616 | WARN_ON(cur_trans != trans->transaction); |
dc17ff8f | 1617 | |
a2de733c | 1618 | btrfs_scrub_pause(root); |
e02119d5 CM |
1619 | /* btrfs_commit_tree_roots is responsible for getting the |
1620 | * various roots consistent with each other. Every pointer | |
1621 | * in the tree of tree roots has to point to the most up to date | |
1622 | * root for every subvolume and other tree. So, we have to keep | |
1623 | * the tree logging code from jumping in and changing any | |
1624 | * of the trees. | |
1625 | * | |
1626 | * At this point in the commit, there can't be any tree-log | |
1627 | * writers, but a little lower down we drop the trans mutex | |
1628 | * and let new people in. By holding the tree_log_mutex | |
1629 | * from now until after the super is written, we avoid races | |
1630 | * with the tree-log code. | |
1631 | */ | |
1632 | mutex_lock(&root->fs_info->tree_log_mutex); | |
1633 | ||
5d4f98a2 | 1634 | ret = commit_fs_roots(trans, root); |
49b25e05 JM |
1635 | if (ret) { |
1636 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
871383be | 1637 | mutex_unlock(&root->fs_info->reloc_mutex); |
49b25e05 JM |
1638 | goto cleanup_transaction; |
1639 | } | |
54aa1f4d | 1640 | |
5d4f98a2 | 1641 | /* commit_fs_roots gets rid of all the tree log roots, it is now |
e02119d5 CM |
1642 | * safe to free the root of tree log roots |
1643 | */ | |
1644 | btrfs_free_log_root_tree(trans, root->fs_info); | |
1645 | ||
5d4f98a2 | 1646 | ret = commit_cowonly_roots(trans, root); |
49b25e05 JM |
1647 | if (ret) { |
1648 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
871383be | 1649 | mutex_unlock(&root->fs_info->reloc_mutex); |
49b25e05 JM |
1650 | goto cleanup_transaction; |
1651 | } | |
54aa1f4d | 1652 | |
11833d66 YZ |
1653 | btrfs_prepare_extent_commit(trans, root); |
1654 | ||
78fae27e | 1655 | cur_trans = root->fs_info->running_transaction; |
5d4f98a2 YZ |
1656 | |
1657 | btrfs_set_root_node(&root->fs_info->tree_root->root_item, | |
1658 | root->fs_info->tree_root->node); | |
817d52f8 | 1659 | switch_commit_root(root->fs_info->tree_root); |
5d4f98a2 YZ |
1660 | |
1661 | btrfs_set_root_node(&root->fs_info->chunk_root->root_item, | |
1662 | root->fs_info->chunk_root->node); | |
817d52f8 | 1663 | switch_commit_root(root->fs_info->chunk_root); |
5d4f98a2 | 1664 | |
edf39272 | 1665 | assert_qgroups_uptodate(trans); |
5d4f98a2 | 1666 | update_super_roots(root); |
e02119d5 CM |
1667 | |
1668 | if (!root->fs_info->log_root_recovering) { | |
6c41761f DS |
1669 | btrfs_set_super_log_root(root->fs_info->super_copy, 0); |
1670 | btrfs_set_super_log_root_level(root->fs_info->super_copy, 0); | |
e02119d5 CM |
1671 | } |
1672 | ||
6c41761f DS |
1673 | memcpy(root->fs_info->super_for_commit, root->fs_info->super_copy, |
1674 | sizeof(*root->fs_info->super_copy)); | |
ccd467d6 | 1675 | |
f9295749 | 1676 | trans->transaction->blocked = 0; |
a4abeea4 JB |
1677 | spin_lock(&root->fs_info->trans_lock); |
1678 | root->fs_info->running_transaction = NULL; | |
1679 | root->fs_info->trans_no_join = 0; | |
1680 | spin_unlock(&root->fs_info->trans_lock); | |
7585717f | 1681 | mutex_unlock(&root->fs_info->reloc_mutex); |
b7ec40d7 | 1682 | |
f9295749 | 1683 | wake_up(&root->fs_info->transaction_wait); |
e6dcd2dc | 1684 | |
79154b1b | 1685 | ret = btrfs_write_and_wait_transaction(trans, root); |
49b25e05 JM |
1686 | if (ret) { |
1687 | btrfs_error(root->fs_info, ret, | |
1688 | "Error while writing out transaction."); | |
1689 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1690 | goto cleanup_transaction; | |
1691 | } | |
1692 | ||
1693 | ret = write_ctree_super(trans, root, 0); | |
1694 | if (ret) { | |
1695 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1696 | goto cleanup_transaction; | |
1697 | } | |
4313b399 | 1698 | |
e02119d5 CM |
1699 | /* |
1700 | * the super is written, we can safely allow the tree-loggers | |
1701 | * to go about their business | |
1702 | */ | |
1703 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1704 | ||
11833d66 | 1705 | btrfs_finish_extent_commit(trans, root); |
4313b399 | 1706 | |
2c90e5d6 | 1707 | cur_trans->commit_done = 1; |
b7ec40d7 | 1708 | |
15ee9bc7 | 1709 | root->fs_info->last_trans_committed = cur_trans->transid; |
817d52f8 | 1710 | |
2c90e5d6 | 1711 | wake_up(&cur_trans->commit_wait); |
3de4586c | 1712 | |
a4abeea4 | 1713 | spin_lock(&root->fs_info->trans_lock); |
13c5a93e | 1714 | list_del_init(&cur_trans->list); |
a4abeea4 JB |
1715 | spin_unlock(&root->fs_info->trans_lock); |
1716 | ||
78fae27e | 1717 | put_transaction(cur_trans); |
79154b1b | 1718 | put_transaction(cur_trans); |
58176a96 | 1719 | |
354aa0fb MX |
1720 | if (trans->type < TRANS_JOIN_NOLOCK) |
1721 | sb_end_intwrite(root->fs_info->sb); | |
b2b5ef5c | 1722 | |
1abe9b8a | 1723 | trace_btrfs_transaction_commit(root); |
1724 | ||
a2de733c AJ |
1725 | btrfs_scrub_continue(root); |
1726 | ||
9ed74f2d JB |
1727 | if (current->journal_info == trans) |
1728 | current->journal_info = NULL; | |
1729 | ||
2c90e5d6 | 1730 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
24bbcf04 YZ |
1731 | |
1732 | if (current != root->fs_info->transaction_kthread) | |
1733 | btrfs_run_delayed_iputs(root); | |
1734 | ||
79154b1b | 1735 | return ret; |
49b25e05 JM |
1736 | |
1737 | cleanup_transaction: | |
0e721106 JB |
1738 | btrfs_trans_release_metadata(trans, root); |
1739 | trans->block_rsv = NULL; | |
49b25e05 JM |
1740 | btrfs_printk(root->fs_info, "Skipping commit of aborted transaction.\n"); |
1741 | // WARN_ON(1); | |
1742 | if (current->journal_info == trans) | |
1743 | current->journal_info = NULL; | |
7b8b92af | 1744 | cleanup_transaction(trans, root, ret); |
49b25e05 JM |
1745 | |
1746 | return ret; | |
79154b1b CM |
1747 | } |
1748 | ||
d352ac68 CM |
1749 | /* |
1750 | * interface function to delete all the snapshots we have scheduled for deletion | |
1751 | */ | |
e9d0b13b CM |
1752 | int btrfs_clean_old_snapshots(struct btrfs_root *root) |
1753 | { | |
5d4f98a2 YZ |
1754 | LIST_HEAD(list); |
1755 | struct btrfs_fs_info *fs_info = root->fs_info; | |
1756 | ||
a4abeea4 | 1757 | spin_lock(&fs_info->trans_lock); |
5d4f98a2 | 1758 | list_splice_init(&fs_info->dead_roots, &list); |
a4abeea4 | 1759 | spin_unlock(&fs_info->trans_lock); |
e9d0b13b | 1760 | |
5d4f98a2 | 1761 | while (!list_empty(&list)) { |
2c536799 JM |
1762 | int ret; |
1763 | ||
5d4f98a2 | 1764 | root = list_entry(list.next, struct btrfs_root, root_list); |
76dda93c YZ |
1765 | list_del(&root->root_list); |
1766 | ||
16cdcec7 MX |
1767 | btrfs_kill_all_delayed_nodes(root); |
1768 | ||
76dda93c YZ |
1769 | if (btrfs_header_backref_rev(root->node) < |
1770 | BTRFS_MIXED_BACKREF_REV) | |
2c536799 | 1771 | ret = btrfs_drop_snapshot(root, NULL, 0, 0); |
76dda93c | 1772 | else |
2c536799 JM |
1773 | ret =btrfs_drop_snapshot(root, NULL, 1, 0); |
1774 | BUG_ON(ret < 0); | |
e9d0b13b CM |
1775 | } |
1776 | return 0; | |
1777 | } |