]>
Commit | Line | Data |
---|---|---|
6cbd5570 CM |
1 | /* |
2 | * Copyright (C) 2007 Oracle. All rights reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
18 | ||
79154b1b | 19 | #include <linux/fs.h> |
5a0e3ad6 | 20 | #include <linux/slab.h> |
34088780 | 21 | #include <linux/sched.h> |
d3c2fdcf | 22 | #include <linux/writeback.h> |
5f39d397 | 23 | #include <linux/pagemap.h> |
5f2cc086 | 24 | #include <linux/blkdev.h> |
8ea05e3a | 25 | #include <linux/uuid.h> |
79154b1b CM |
26 | #include "ctree.h" |
27 | #include "disk-io.h" | |
28 | #include "transaction.h" | |
925baedd | 29 | #include "locking.h" |
e02119d5 | 30 | #include "tree-log.h" |
581bb050 | 31 | #include "inode-map.h" |
733f4fbb | 32 | #include "volumes.h" |
8dabb742 | 33 | #include "dev-replace.h" |
79154b1b | 34 | |
0f7d52f4 CM |
35 | #define BTRFS_ROOT_TRANS_TAG 0 |
36 | ||
49b25e05 | 37 | void put_transaction(struct btrfs_transaction *transaction) |
79154b1b | 38 | { |
13c5a93e JB |
39 | WARN_ON(atomic_read(&transaction->use_count) == 0); |
40 | if (atomic_dec_and_test(&transaction->use_count)) { | |
a4abeea4 | 41 | BUG_ON(!list_empty(&transaction->list)); |
00f04b88 | 42 | WARN_ON(transaction->delayed_refs.root.rb_node); |
2c90e5d6 | 43 | kmem_cache_free(btrfs_transaction_cachep, transaction); |
78fae27e | 44 | } |
79154b1b CM |
45 | } |
46 | ||
817d52f8 JB |
47 | static noinline void switch_commit_root(struct btrfs_root *root) |
48 | { | |
817d52f8 JB |
49 | free_extent_buffer(root->commit_root); |
50 | root->commit_root = btrfs_root_node(root); | |
817d52f8 JB |
51 | } |
52 | ||
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. */ |
87533c47 | 64 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { |
19ae4e81 | 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; |
87533c47 | 116 | } else if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { |
e4b50e14 | 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); | |
569e0f35 | 160 | INIT_LIST_HEAD(&cur_trans->ordered_operations); |
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 | |
87533c47 | 305 | if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) |
acce952b | 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 | 335 | if (ret) |
843fcf35 | 336 | goto reserve_fail; |
b5009945 | 337 | } |
a22285a6 YZ |
338 | again: |
339 | h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS); | |
843fcf35 MX |
340 | if (!h) { |
341 | ret = -ENOMEM; | |
342 | goto alloc_fail; | |
343 | } | |
37d1aeee | 344 | |
98114659 JB |
345 | /* |
346 | * If we are JOIN_NOLOCK we're already committing a transaction and | |
347 | * waiting on this guy, so we don't need to do the sb_start_intwrite | |
348 | * because we're already holding a ref. We need this because we could | |
349 | * have raced in and did an fsync() on a file which can kick a commit | |
350 | * and then we deadlock with somebody doing a freeze. | |
354aa0fb MX |
351 | * |
352 | * If we are ATTACH, it means we just want to catch the current | |
353 | * transaction and commit it, so we needn't do sb_start_intwrite(). | |
98114659 | 354 | */ |
354aa0fb | 355 | if (type < TRANS_JOIN_NOLOCK) |
60376ce4 | 356 | sb_start_intwrite(root->fs_info->sb); |
b2b5ef5c | 357 | |
a22285a6 | 358 | if (may_wait_transaction(root, type)) |
37d1aeee | 359 | wait_current_trans(root); |
a22285a6 | 360 | |
a4abeea4 | 361 | do { |
354aa0fb | 362 | ret = join_transaction(root, type); |
a4abeea4 JB |
363 | if (ret == -EBUSY) |
364 | wait_current_trans(root); | |
365 | } while (ret == -EBUSY); | |
366 | ||
db5b493a | 367 | if (ret < 0) { |
354aa0fb MX |
368 | /* We must get the transaction if we are JOIN_NOLOCK. */ |
369 | BUG_ON(type == TRANS_JOIN_NOLOCK); | |
843fcf35 | 370 | goto join_fail; |
db5b493a | 371 | } |
0f7d52f4 | 372 | |
a22285a6 | 373 | cur_trans = root->fs_info->running_transaction; |
a22285a6 YZ |
374 | |
375 | h->transid = cur_trans->transid; | |
376 | h->transaction = cur_trans; | |
79154b1b | 377 | h->blocks_used = 0; |
a22285a6 | 378 | h->bytes_reserved = 0; |
d13603ef | 379 | h->root = root; |
56bec294 | 380 | h->delayed_ref_updates = 0; |
2a1eb461 | 381 | h->use_count = 1; |
0e721106 | 382 | h->adding_csums = 0; |
f0486c68 | 383 | h->block_rsv = NULL; |
2a1eb461 | 384 | h->orig_rsv = NULL; |
49b25e05 | 385 | h->aborted = 0; |
c5567237 | 386 | h->qgroup_reserved = qgroup_reserved; |
bed92eae | 387 | h->delayed_ref_elem.seq = 0; |
a698d075 | 388 | h->type = type; |
c6b305a8 | 389 | h->allocating_chunk = false; |
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 | 411 | return h; |
843fcf35 MX |
412 | |
413 | join_fail: | |
414 | if (type < TRANS_JOIN_NOLOCK) | |
415 | sb_end_intwrite(root->fs_info->sb); | |
416 | kmem_cache_free(btrfs_trans_handle_cachep, h); | |
417 | alloc_fail: | |
418 | if (num_bytes) | |
419 | btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv, | |
420 | num_bytes); | |
421 | reserve_fail: | |
422 | if (qgroup_reserved) | |
423 | btrfs_qgroup_free(root, qgroup_reserved); | |
424 | return ERR_PTR(ret); | |
79154b1b CM |
425 | } |
426 | ||
f9295749 | 427 | struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, |
a22285a6 | 428 | int num_items) |
f9295749 | 429 | { |
08e007d2 MX |
430 | return start_transaction(root, num_items, TRANS_START, |
431 | BTRFS_RESERVE_FLUSH_ALL); | |
f9295749 | 432 | } |
8407aa46 | 433 | |
08e007d2 | 434 | struct btrfs_trans_handle *btrfs_start_transaction_lflush( |
8407aa46 MX |
435 | struct btrfs_root *root, int num_items) |
436 | { | |
08e007d2 MX |
437 | return start_transaction(root, num_items, TRANS_START, |
438 | BTRFS_RESERVE_FLUSH_LIMIT); | |
8407aa46 MX |
439 | } |
440 | ||
7a7eaa40 | 441 | struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root) |
f9295749 | 442 | { |
8407aa46 | 443 | return start_transaction(root, 0, TRANS_JOIN, 0); |
f9295749 CM |
444 | } |
445 | ||
7a7eaa40 | 446 | struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root) |
0af3d00b | 447 | { |
8407aa46 | 448 | return start_transaction(root, 0, TRANS_JOIN_NOLOCK, 0); |
0af3d00b JB |
449 | } |
450 | ||
7a7eaa40 | 451 | struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root) |
9ca9ee09 | 452 | { |
8407aa46 | 453 | return start_transaction(root, 0, TRANS_USERSPACE, 0); |
9ca9ee09 SW |
454 | } |
455 | ||
354aa0fb | 456 | struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root) |
60376ce4 | 457 | { |
354aa0fb | 458 | return start_transaction(root, 0, TRANS_ATTACH, 0); |
60376ce4 JB |
459 | } |
460 | ||
d352ac68 | 461 | /* wait for a transaction commit to be fully complete */ |
b9c8300c | 462 | static noinline void wait_for_commit(struct btrfs_root *root, |
89ce8a63 CM |
463 | struct btrfs_transaction *commit) |
464 | { | |
72d63ed6 | 465 | wait_event(commit->commit_wait, commit->commit_done); |
89ce8a63 CM |
466 | } |
467 | ||
46204592 SW |
468 | int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid) |
469 | { | |
470 | struct btrfs_transaction *cur_trans = NULL, *t; | |
8cd2807f | 471 | int ret = 0; |
46204592 | 472 | |
46204592 SW |
473 | if (transid) { |
474 | if (transid <= root->fs_info->last_trans_committed) | |
a4abeea4 | 475 | goto out; |
46204592 | 476 | |
8cd2807f | 477 | ret = -EINVAL; |
46204592 | 478 | /* find specified transaction */ |
a4abeea4 | 479 | spin_lock(&root->fs_info->trans_lock); |
46204592 SW |
480 | list_for_each_entry(t, &root->fs_info->trans_list, list) { |
481 | if (t->transid == transid) { | |
482 | cur_trans = t; | |
a4abeea4 | 483 | atomic_inc(&cur_trans->use_count); |
8cd2807f | 484 | ret = 0; |
46204592 SW |
485 | break; |
486 | } | |
8cd2807f MX |
487 | if (t->transid > transid) { |
488 | ret = 0; | |
46204592 | 489 | break; |
8cd2807f | 490 | } |
46204592 | 491 | } |
a4abeea4 | 492 | spin_unlock(&root->fs_info->trans_lock); |
8cd2807f | 493 | /* The specified transaction doesn't exist */ |
46204592 | 494 | if (!cur_trans) |
8cd2807f | 495 | goto out; |
46204592 SW |
496 | } else { |
497 | /* find newest transaction that is committing | committed */ | |
a4abeea4 | 498 | spin_lock(&root->fs_info->trans_lock); |
46204592 SW |
499 | list_for_each_entry_reverse(t, &root->fs_info->trans_list, |
500 | list) { | |
501 | if (t->in_commit) { | |
502 | if (t->commit_done) | |
3473f3c0 | 503 | break; |
46204592 | 504 | cur_trans = t; |
a4abeea4 | 505 | atomic_inc(&cur_trans->use_count); |
46204592 SW |
506 | break; |
507 | } | |
508 | } | |
a4abeea4 | 509 | spin_unlock(&root->fs_info->trans_lock); |
46204592 | 510 | if (!cur_trans) |
a4abeea4 | 511 | goto out; /* nothing committing|committed */ |
46204592 SW |
512 | } |
513 | ||
46204592 | 514 | wait_for_commit(root, cur_trans); |
46204592 | 515 | put_transaction(cur_trans); |
a4abeea4 | 516 | out: |
46204592 SW |
517 | return ret; |
518 | } | |
519 | ||
37d1aeee CM |
520 | void btrfs_throttle(struct btrfs_root *root) |
521 | { | |
a4abeea4 | 522 | if (!atomic_read(&root->fs_info->open_ioctl_trans)) |
9ca9ee09 | 523 | wait_current_trans(root); |
37d1aeee CM |
524 | } |
525 | ||
8929ecfa YZ |
526 | static int should_end_transaction(struct btrfs_trans_handle *trans, |
527 | struct btrfs_root *root) | |
528 | { | |
529 | int ret; | |
36ba022a JB |
530 | |
531 | ret = btrfs_block_rsv_check(root, &root->fs_info->global_block_rsv, 5); | |
8929ecfa YZ |
532 | return ret ? 1 : 0; |
533 | } | |
534 | ||
535 | int btrfs_should_end_transaction(struct btrfs_trans_handle *trans, | |
536 | struct btrfs_root *root) | |
537 | { | |
538 | struct btrfs_transaction *cur_trans = trans->transaction; | |
539 | int updates; | |
49b25e05 | 540 | int err; |
8929ecfa | 541 | |
a4abeea4 | 542 | smp_mb(); |
8929ecfa YZ |
543 | if (cur_trans->blocked || cur_trans->delayed_refs.flushing) |
544 | return 1; | |
545 | ||
546 | updates = trans->delayed_ref_updates; | |
547 | trans->delayed_ref_updates = 0; | |
49b25e05 JM |
548 | if (updates) { |
549 | err = btrfs_run_delayed_refs(trans, root, updates); | |
550 | if (err) /* Error code will also eval true */ | |
551 | return err; | |
552 | } | |
8929ecfa YZ |
553 | |
554 | return should_end_transaction(trans, root); | |
555 | } | |
556 | ||
89ce8a63 | 557 | static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, |
a698d075 | 558 | struct btrfs_root *root, int throttle) |
79154b1b | 559 | { |
8929ecfa | 560 | struct btrfs_transaction *cur_trans = trans->transaction; |
ab78c84d | 561 | struct btrfs_fs_info *info = root->fs_info; |
c3e69d58 | 562 | int count = 0; |
a698d075 | 563 | int lock = (trans->type != TRANS_JOIN_NOLOCK); |
4edc2ca3 | 564 | int err = 0; |
c3e69d58 | 565 | |
2a1eb461 JB |
566 | if (--trans->use_count) { |
567 | trans->block_rsv = trans->orig_rsv; | |
568 | return 0; | |
569 | } | |
570 | ||
edf39272 JS |
571 | /* |
572 | * do the qgroup accounting as early as possible | |
573 | */ | |
574 | err = btrfs_delayed_refs_qgroup_accounting(trans, info); | |
575 | ||
b24e03db | 576 | btrfs_trans_release_metadata(trans, root); |
4c13d758 | 577 | trans->block_rsv = NULL; |
d13603ef AJ |
578 | /* |
579 | * the same root has to be passed to start_transaction and | |
580 | * end_transaction. Subvolume quota depends on this. | |
581 | */ | |
582 | WARN_ON(trans->root != root); | |
c5567237 AJ |
583 | |
584 | if (trans->qgroup_reserved) { | |
585 | btrfs_qgroup_free(root, trans->qgroup_reserved); | |
586 | trans->qgroup_reserved = 0; | |
587 | } | |
588 | ||
ea658bad JB |
589 | if (!list_empty(&trans->new_bgs)) |
590 | btrfs_create_pending_block_groups(trans, root); | |
591 | ||
203bf287 | 592 | while (count < 2) { |
c3e69d58 CM |
593 | unsigned long cur = trans->delayed_ref_updates; |
594 | trans->delayed_ref_updates = 0; | |
595 | if (cur && | |
596 | trans->transaction->delayed_refs.num_heads_ready > 64) { | |
597 | trans->delayed_ref_updates = 0; | |
598 | btrfs_run_delayed_refs(trans, root, cur); | |
599 | } else { | |
600 | break; | |
601 | } | |
602 | count++; | |
56bec294 | 603 | } |
0e721106 JB |
604 | btrfs_trans_release_metadata(trans, root); |
605 | trans->block_rsv = NULL; | |
56bec294 | 606 | |
ea658bad JB |
607 | if (!list_empty(&trans->new_bgs)) |
608 | btrfs_create_pending_block_groups(trans, root); | |
609 | ||
a4abeea4 JB |
610 | if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) && |
611 | should_end_transaction(trans, root)) { | |
8929ecfa | 612 | trans->transaction->blocked = 1; |
a4abeea4 JB |
613 | smp_wmb(); |
614 | } | |
8929ecfa | 615 | |
0af3d00b | 616 | if (lock && cur_trans->blocked && !cur_trans->in_commit) { |
81317fde JB |
617 | if (throttle) { |
618 | /* | |
619 | * We may race with somebody else here so end up having | |
620 | * to call end_transaction on ourselves again, so inc | |
621 | * our use_count. | |
622 | */ | |
623 | trans->use_count++; | |
8929ecfa | 624 | return btrfs_commit_transaction(trans, root); |
81317fde | 625 | } else { |
8929ecfa | 626 | wake_up_process(info->transaction_kthread); |
81317fde | 627 | } |
8929ecfa YZ |
628 | } |
629 | ||
354aa0fb | 630 | if (trans->type < TRANS_JOIN_NOLOCK) |
98114659 | 631 | sb_end_intwrite(root->fs_info->sb); |
6df7881a | 632 | |
8929ecfa | 633 | WARN_ON(cur_trans != info->running_transaction); |
13c5a93e JB |
634 | WARN_ON(atomic_read(&cur_trans->num_writers) < 1); |
635 | atomic_dec(&cur_trans->num_writers); | |
89ce8a63 | 636 | |
99d16cbc | 637 | smp_mb(); |
79154b1b CM |
638 | if (waitqueue_active(&cur_trans->writer_wait)) |
639 | wake_up(&cur_trans->writer_wait); | |
79154b1b | 640 | put_transaction(cur_trans); |
9ed74f2d JB |
641 | |
642 | if (current->journal_info == trans) | |
643 | current->journal_info = NULL; | |
ab78c84d | 644 | |
24bbcf04 YZ |
645 | if (throttle) |
646 | btrfs_run_delayed_iputs(root); | |
647 | ||
49b25e05 | 648 | if (trans->aborted || |
87533c47 | 649 | test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) |
4edc2ca3 | 650 | err = -EIO; |
edf39272 | 651 | assert_qgroups_uptodate(trans); |
49b25e05 | 652 | |
4edc2ca3 DJ |
653 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
654 | return err; | |
79154b1b CM |
655 | } |
656 | ||
89ce8a63 CM |
657 | int btrfs_end_transaction(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, 0); |
16cdcec7 MX |
663 | if (ret) |
664 | return ret; | |
665 | return 0; | |
89ce8a63 CM |
666 | } |
667 | ||
668 | int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans, | |
669 | struct btrfs_root *root) | |
670 | { | |
16cdcec7 MX |
671 | int ret; |
672 | ||
a698d075 | 673 | ret = __btrfs_end_transaction(trans, root, 1); |
16cdcec7 MX |
674 | if (ret) |
675 | return ret; | |
676 | return 0; | |
677 | } | |
678 | ||
679 | int btrfs_end_transaction_dmeta(struct btrfs_trans_handle *trans, | |
680 | struct btrfs_root *root) | |
681 | { | |
a698d075 | 682 | return __btrfs_end_transaction(trans, root, 1); |
89ce8a63 CM |
683 | } |
684 | ||
d352ac68 CM |
685 | /* |
686 | * when btree blocks are allocated, they have some corresponding bits set for | |
687 | * them in one of two extent_io trees. This is used to make sure all of | |
690587d1 | 688 | * those extents are sent to disk but does not wait on them |
d352ac68 | 689 | */ |
690587d1 | 690 | int btrfs_write_marked_extents(struct btrfs_root *root, |
8cef4e16 | 691 | struct extent_io_tree *dirty_pages, int mark) |
79154b1b | 692 | { |
777e6bd7 | 693 | int err = 0; |
7c4452b9 | 694 | int werr = 0; |
1728366e | 695 | struct address_space *mapping = root->fs_info->btree_inode->i_mapping; |
e6138876 | 696 | struct extent_state *cached_state = NULL; |
777e6bd7 | 697 | u64 start = 0; |
5f39d397 | 698 | u64 end; |
7c4452b9 | 699 | |
1728366e | 700 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
e6138876 JB |
701 | mark, &cached_state)) { |
702 | convert_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT, | |
703 | mark, &cached_state, GFP_NOFS); | |
704 | cached_state = NULL; | |
1728366e JB |
705 | err = filemap_fdatawrite_range(mapping, start, end); |
706 | if (err) | |
707 | werr = err; | |
708 | cond_resched(); | |
709 | start = end + 1; | |
7c4452b9 | 710 | } |
690587d1 CM |
711 | if (err) |
712 | werr = err; | |
713 | return werr; | |
714 | } | |
715 | ||
716 | /* | |
717 | * when btree blocks are allocated, they have some corresponding bits set for | |
718 | * them in one of two extent_io trees. This is used to make sure all of | |
719 | * those extents are on disk for transaction or log commit. We wait | |
720 | * on all the pages and clear them from the dirty pages state tree | |
721 | */ | |
722 | int btrfs_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 723 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 | 724 | { |
690587d1 CM |
725 | int err = 0; |
726 | int werr = 0; | |
1728366e | 727 | struct address_space *mapping = root->fs_info->btree_inode->i_mapping; |
e6138876 | 728 | struct extent_state *cached_state = NULL; |
690587d1 CM |
729 | u64 start = 0; |
730 | u64 end; | |
777e6bd7 | 731 | |
1728366e | 732 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
e6138876 JB |
733 | EXTENT_NEED_WAIT, &cached_state)) { |
734 | clear_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT, | |
735 | 0, 0, &cached_state, GFP_NOFS); | |
1728366e JB |
736 | err = filemap_fdatawait_range(mapping, start, end); |
737 | if (err) | |
738 | werr = err; | |
739 | cond_resched(); | |
740 | start = end + 1; | |
777e6bd7 | 741 | } |
7c4452b9 CM |
742 | if (err) |
743 | werr = err; | |
744 | return werr; | |
79154b1b CM |
745 | } |
746 | ||
690587d1 CM |
747 | /* |
748 | * when btree blocks are allocated, they have some corresponding bits set for | |
749 | * them in one of two extent_io trees. This is used to make sure all of | |
750 | * those extents are on disk for transaction or log commit | |
751 | */ | |
752 | int btrfs_write_and_wait_marked_extents(struct btrfs_root *root, | |
8cef4e16 | 753 | struct extent_io_tree *dirty_pages, int mark) |
690587d1 CM |
754 | { |
755 | int ret; | |
756 | int ret2; | |
757 | ||
8cef4e16 YZ |
758 | ret = btrfs_write_marked_extents(root, dirty_pages, mark); |
759 | ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark); | |
bf0da8c1 CM |
760 | |
761 | if (ret) | |
762 | return ret; | |
763 | if (ret2) | |
764 | return ret2; | |
765 | return 0; | |
690587d1 CM |
766 | } |
767 | ||
d0c803c4 CM |
768 | int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans, |
769 | struct btrfs_root *root) | |
770 | { | |
771 | if (!trans || !trans->transaction) { | |
772 | struct inode *btree_inode; | |
773 | btree_inode = root->fs_info->btree_inode; | |
774 | return filemap_write_and_wait(btree_inode->i_mapping); | |
775 | } | |
776 | return btrfs_write_and_wait_marked_extents(root, | |
8cef4e16 YZ |
777 | &trans->transaction->dirty_pages, |
778 | EXTENT_DIRTY); | |
d0c803c4 CM |
779 | } |
780 | ||
d352ac68 CM |
781 | /* |
782 | * this is used to update the root pointer in the tree of tree roots. | |
783 | * | |
784 | * But, in the case of the extent allocation tree, updating the root | |
785 | * pointer may allocate blocks which may change the root of the extent | |
786 | * allocation tree. | |
787 | * | |
788 | * So, this loops and repeats and makes sure the cowonly root didn't | |
789 | * change while the root pointer was being updated in the metadata. | |
790 | */ | |
0b86a832 CM |
791 | static int update_cowonly_root(struct btrfs_trans_handle *trans, |
792 | struct btrfs_root *root) | |
79154b1b CM |
793 | { |
794 | int ret; | |
0b86a832 | 795 | u64 old_root_bytenr; |
86b9f2ec | 796 | u64 old_root_used; |
0b86a832 | 797 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
79154b1b | 798 | |
86b9f2ec | 799 | old_root_used = btrfs_root_used(&root->root_item); |
0b86a832 | 800 | btrfs_write_dirty_block_groups(trans, root); |
56bec294 | 801 | |
d397712b | 802 | while (1) { |
0b86a832 | 803 | old_root_bytenr = btrfs_root_bytenr(&root->root_item); |
86b9f2ec YZ |
804 | if (old_root_bytenr == root->node->start && |
805 | old_root_used == btrfs_root_used(&root->root_item)) | |
79154b1b | 806 | break; |
87ef2bb4 | 807 | |
5d4f98a2 | 808 | btrfs_set_root_node(&root->root_item, root->node); |
79154b1b | 809 | ret = btrfs_update_root(trans, tree_root, |
0b86a832 CM |
810 | &root->root_key, |
811 | &root->root_item); | |
49b25e05 JM |
812 | if (ret) |
813 | return ret; | |
56bec294 | 814 | |
86b9f2ec | 815 | old_root_used = btrfs_root_used(&root->root_item); |
4a8c9a62 | 816 | ret = btrfs_write_dirty_block_groups(trans, root); |
49b25e05 JM |
817 | if (ret) |
818 | return ret; | |
0b86a832 | 819 | } |
276e680d YZ |
820 | |
821 | if (root != root->fs_info->extent_root) | |
822 | switch_commit_root(root); | |
823 | ||
0b86a832 CM |
824 | return 0; |
825 | } | |
826 | ||
d352ac68 CM |
827 | /* |
828 | * update all the cowonly tree roots on disk | |
49b25e05 JM |
829 | * |
830 | * The error handling in this function may not be obvious. Any of the | |
831 | * failures will cause the file system to go offline. We still need | |
832 | * to clean up the delayed refs. | |
d352ac68 | 833 | */ |
5d4f98a2 YZ |
834 | static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans, |
835 | struct btrfs_root *root) | |
0b86a832 CM |
836 | { |
837 | struct btrfs_fs_info *fs_info = root->fs_info; | |
838 | struct list_head *next; | |
84234f3a | 839 | struct extent_buffer *eb; |
56bec294 | 840 | int ret; |
84234f3a | 841 | |
56bec294 | 842 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
49b25e05 JM |
843 | if (ret) |
844 | return ret; | |
87ef2bb4 | 845 | |
84234f3a | 846 | eb = btrfs_lock_root_node(fs_info->tree_root); |
49b25e05 JM |
847 | ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, |
848 | 0, &eb); | |
84234f3a YZ |
849 | btrfs_tree_unlock(eb); |
850 | free_extent_buffer(eb); | |
0b86a832 | 851 | |
49b25e05 JM |
852 | if (ret) |
853 | return ret; | |
854 | ||
56bec294 | 855 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
49b25e05 JM |
856 | if (ret) |
857 | return ret; | |
87ef2bb4 | 858 | |
733f4fbb | 859 | ret = btrfs_run_dev_stats(trans, root->fs_info); |
8dabb742 SB |
860 | WARN_ON(ret); |
861 | ret = btrfs_run_dev_replace(trans, root->fs_info); | |
862 | WARN_ON(ret); | |
733f4fbb | 863 | |
546adb0d JS |
864 | ret = btrfs_run_qgroups(trans, root->fs_info); |
865 | BUG_ON(ret); | |
866 | ||
867 | /* run_qgroups might have added some more refs */ | |
868 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); | |
869 | BUG_ON(ret); | |
870 | ||
d397712b | 871 | while (!list_empty(&fs_info->dirty_cowonly_roots)) { |
0b86a832 CM |
872 | next = fs_info->dirty_cowonly_roots.next; |
873 | list_del_init(next); | |
874 | root = list_entry(next, struct btrfs_root, dirty_list); | |
87ef2bb4 | 875 | |
49b25e05 JM |
876 | ret = update_cowonly_root(trans, root); |
877 | if (ret) | |
878 | return ret; | |
79154b1b | 879 | } |
276e680d YZ |
880 | |
881 | down_write(&fs_info->extent_commit_sem); | |
882 | switch_commit_root(fs_info->extent_root); | |
883 | up_write(&fs_info->extent_commit_sem); | |
884 | ||
8dabb742 SB |
885 | btrfs_after_dev_replace_commit(fs_info); |
886 | ||
79154b1b CM |
887 | return 0; |
888 | } | |
889 | ||
d352ac68 CM |
890 | /* |
891 | * dead roots are old snapshots that need to be deleted. This allocates | |
892 | * a dirty root struct and adds it into the list of dead roots that need to | |
893 | * be deleted | |
894 | */ | |
5d4f98a2 | 895 | int btrfs_add_dead_root(struct btrfs_root *root) |
5eda7b5e | 896 | { |
a4abeea4 | 897 | spin_lock(&root->fs_info->trans_lock); |
5d4f98a2 | 898 | list_add(&root->root_list, &root->fs_info->dead_roots); |
a4abeea4 | 899 | spin_unlock(&root->fs_info->trans_lock); |
5eda7b5e CM |
900 | return 0; |
901 | } | |
902 | ||
d352ac68 | 903 | /* |
5d4f98a2 | 904 | * update all the cowonly tree roots on disk |
d352ac68 | 905 | */ |
5d4f98a2 YZ |
906 | static noinline int commit_fs_roots(struct btrfs_trans_handle *trans, |
907 | struct btrfs_root *root) | |
0f7d52f4 | 908 | { |
0f7d52f4 | 909 | struct btrfs_root *gang[8]; |
5d4f98a2 | 910 | struct btrfs_fs_info *fs_info = root->fs_info; |
0f7d52f4 CM |
911 | int i; |
912 | int ret; | |
54aa1f4d CM |
913 | int err = 0; |
914 | ||
a4abeea4 | 915 | spin_lock(&fs_info->fs_roots_radix_lock); |
d397712b | 916 | while (1) { |
5d4f98a2 YZ |
917 | ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, |
918 | (void **)gang, 0, | |
0f7d52f4 CM |
919 | ARRAY_SIZE(gang), |
920 | BTRFS_ROOT_TRANS_TAG); | |
921 | if (ret == 0) | |
922 | break; | |
923 | for (i = 0; i < ret; i++) { | |
924 | root = gang[i]; | |
5d4f98a2 YZ |
925 | radix_tree_tag_clear(&fs_info->fs_roots_radix, |
926 | (unsigned long)root->root_key.objectid, | |
927 | BTRFS_ROOT_TRANS_TAG); | |
a4abeea4 | 928 | spin_unlock(&fs_info->fs_roots_radix_lock); |
31153d81 | 929 | |
e02119d5 | 930 | btrfs_free_log(trans, root); |
5d4f98a2 | 931 | btrfs_update_reloc_root(trans, root); |
d68fc57b | 932 | btrfs_orphan_commit_root(trans, root); |
bcc63abb | 933 | |
82d5902d LZ |
934 | btrfs_save_ino_cache(root, trans); |
935 | ||
f1ebcc74 LB |
936 | /* see comments in should_cow_block() */ |
937 | root->force_cow = 0; | |
938 | smp_wmb(); | |
939 | ||
978d910d | 940 | if (root->commit_root != root->node) { |
581bb050 | 941 | mutex_lock(&root->fs_commit_mutex); |
817d52f8 | 942 | switch_commit_root(root); |
581bb050 LZ |
943 | btrfs_unpin_free_ino(root); |
944 | mutex_unlock(&root->fs_commit_mutex); | |
945 | ||
978d910d YZ |
946 | btrfs_set_root_node(&root->root_item, |
947 | root->node); | |
948 | } | |
5d4f98a2 | 949 | |
5d4f98a2 | 950 | err = btrfs_update_root(trans, fs_info->tree_root, |
0f7d52f4 CM |
951 | &root->root_key, |
952 | &root->root_item); | |
a4abeea4 | 953 | spin_lock(&fs_info->fs_roots_radix_lock); |
54aa1f4d CM |
954 | if (err) |
955 | break; | |
0f7d52f4 CM |
956 | } |
957 | } | |
a4abeea4 | 958 | spin_unlock(&fs_info->fs_roots_radix_lock); |
54aa1f4d | 959 | return err; |
0f7d52f4 CM |
960 | } |
961 | ||
d352ac68 | 962 | /* |
de78b51a ES |
963 | * defrag a given btree. |
964 | * Every leaf in the btree is read and defragged. | |
d352ac68 | 965 | */ |
de78b51a | 966 | int btrfs_defrag_root(struct btrfs_root *root) |
e9d0b13b CM |
967 | { |
968 | struct btrfs_fs_info *info = root->fs_info; | |
e9d0b13b | 969 | struct btrfs_trans_handle *trans; |
8929ecfa | 970 | int ret; |
e9d0b13b | 971 | |
8929ecfa | 972 | if (xchg(&root->defrag_running, 1)) |
e9d0b13b | 973 | return 0; |
8929ecfa | 974 | |
6b80053d | 975 | while (1) { |
8929ecfa YZ |
976 | trans = btrfs_start_transaction(root, 0); |
977 | if (IS_ERR(trans)) | |
978 | return PTR_ERR(trans); | |
979 | ||
de78b51a | 980 | ret = btrfs_defrag_leaves(trans, root); |
8929ecfa | 981 | |
e9d0b13b | 982 | btrfs_end_transaction(trans, root); |
b53d3f5d | 983 | btrfs_btree_balance_dirty(info->tree_root); |
e9d0b13b CM |
984 | cond_resched(); |
985 | ||
7841cb28 | 986 | if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN) |
e9d0b13b | 987 | break; |
210549eb DS |
988 | |
989 | if (btrfs_defrag_cancelled(root->fs_info)) { | |
990 | printk(KERN_DEBUG "btrfs: defrag_root cancelled\n"); | |
991 | ret = -EAGAIN; | |
992 | break; | |
993 | } | |
e9d0b13b CM |
994 | } |
995 | root->defrag_running = 0; | |
8929ecfa | 996 | return ret; |
e9d0b13b CM |
997 | } |
998 | ||
d352ac68 CM |
999 | /* |
1000 | * new snapshots need to be created at a very specific time in the | |
1001 | * transaction commit. This does the actual creation | |
1002 | */ | |
80b6794d | 1003 | static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, |
3063d29f CM |
1004 | struct btrfs_fs_info *fs_info, |
1005 | struct btrfs_pending_snapshot *pending) | |
1006 | { | |
1007 | struct btrfs_key key; | |
80b6794d | 1008 | struct btrfs_root_item *new_root_item; |
3063d29f CM |
1009 | struct btrfs_root *tree_root = fs_info->tree_root; |
1010 | struct btrfs_root *root = pending->root; | |
6bdb72de | 1011 | struct btrfs_root *parent_root; |
98c9942a | 1012 | struct btrfs_block_rsv *rsv; |
6bdb72de | 1013 | struct inode *parent_inode; |
42874b3d MX |
1014 | struct btrfs_path *path; |
1015 | struct btrfs_dir_item *dir_item; | |
6a912213 | 1016 | struct dentry *parent; |
a22285a6 | 1017 | struct dentry *dentry; |
3063d29f | 1018 | struct extent_buffer *tmp; |
925baedd | 1019 | struct extent_buffer *old; |
8ea05e3a | 1020 | struct timespec cur_time = CURRENT_TIME; |
3063d29f | 1021 | int ret; |
d68fc57b | 1022 | u64 to_reserve = 0; |
6bdb72de | 1023 | u64 index = 0; |
a22285a6 | 1024 | u64 objectid; |
b83cc969 | 1025 | u64 root_flags; |
8ea05e3a | 1026 | uuid_le new_uuid; |
3063d29f | 1027 | |
42874b3d MX |
1028 | path = btrfs_alloc_path(); |
1029 | if (!path) { | |
1030 | ret = pending->error = -ENOMEM; | |
1031 | goto path_alloc_fail; | |
1032 | } | |
1033 | ||
80b6794d CM |
1034 | new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS); |
1035 | if (!new_root_item) { | |
49b25e05 | 1036 | ret = pending->error = -ENOMEM; |
6fa9700e | 1037 | goto root_item_alloc_fail; |
80b6794d | 1038 | } |
a22285a6 | 1039 | |
581bb050 | 1040 | ret = btrfs_find_free_objectid(tree_root, &objectid); |
a22285a6 YZ |
1041 | if (ret) { |
1042 | pending->error = ret; | |
6fa9700e | 1043 | goto no_free_objectid; |
a22285a6 | 1044 | } |
3063d29f | 1045 | |
3fd0a558 | 1046 | btrfs_reloc_pre_snapshot(trans, pending, &to_reserve); |
d68fc57b YZ |
1047 | |
1048 | if (to_reserve > 0) { | |
08e007d2 MX |
1049 | ret = btrfs_block_rsv_add(root, &pending->block_rsv, |
1050 | to_reserve, | |
1051 | BTRFS_RESERVE_NO_FLUSH); | |
d68fc57b YZ |
1052 | if (ret) { |
1053 | pending->error = ret; | |
6fa9700e | 1054 | goto no_free_objectid; |
d68fc57b YZ |
1055 | } |
1056 | } | |
1057 | ||
6f72c7e2 AJ |
1058 | ret = btrfs_qgroup_inherit(trans, fs_info, root->root_key.objectid, |
1059 | objectid, pending->inherit); | |
6f72c7e2 AJ |
1060 | if (ret) { |
1061 | pending->error = ret; | |
6fa9700e | 1062 | goto no_free_objectid; |
6f72c7e2 AJ |
1063 | } |
1064 | ||
3063d29f | 1065 | key.objectid = objectid; |
a22285a6 YZ |
1066 | key.offset = (u64)-1; |
1067 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3063d29f | 1068 | |
6fa9700e | 1069 | rsv = trans->block_rsv; |
a22285a6 | 1070 | trans->block_rsv = &pending->block_rsv; |
3de4586c | 1071 | |
a22285a6 | 1072 | dentry = pending->dentry; |
6a912213 JB |
1073 | parent = dget_parent(dentry); |
1074 | parent_inode = parent->d_inode; | |
a22285a6 | 1075 | parent_root = BTRFS_I(parent_inode)->root; |
7585717f | 1076 | record_root_in_trans(trans, parent_root); |
a22285a6 | 1077 | |
3063d29f CM |
1078 | /* |
1079 | * insert the directory item | |
1080 | */ | |
3de4586c | 1081 | ret = btrfs_set_inode_index(parent_inode, &index); |
49b25e05 | 1082 | BUG_ON(ret); /* -ENOMEM */ |
42874b3d MX |
1083 | |
1084 | /* check if there is a file/dir which has the same name. */ | |
1085 | dir_item = btrfs_lookup_dir_item(NULL, parent_root, path, | |
1086 | btrfs_ino(parent_inode), | |
1087 | dentry->d_name.name, | |
1088 | dentry->d_name.len, 0); | |
1089 | if (dir_item != NULL && !IS_ERR(dir_item)) { | |
fe66a05a | 1090 | pending->error = -EEXIST; |
fe66a05a | 1091 | goto fail; |
42874b3d MX |
1092 | } else if (IS_ERR(dir_item)) { |
1093 | ret = PTR_ERR(dir_item); | |
8732d44f MX |
1094 | btrfs_abort_transaction(trans, root, ret); |
1095 | goto fail; | |
79787eaa | 1096 | } |
42874b3d | 1097 | btrfs_release_path(path); |
52c26179 | 1098 | |
e999376f CM |
1099 | /* |
1100 | * pull in the delayed directory update | |
1101 | * and the delayed inode item | |
1102 | * otherwise we corrupt the FS during | |
1103 | * snapshot | |
1104 | */ | |
1105 | ret = btrfs_run_delayed_items(trans, root); | |
8732d44f MX |
1106 | if (ret) { /* Transaction aborted */ |
1107 | btrfs_abort_transaction(trans, root, ret); | |
1108 | goto fail; | |
1109 | } | |
e999376f | 1110 | |
7585717f | 1111 | record_root_in_trans(trans, root); |
6bdb72de SW |
1112 | btrfs_set_root_last_snapshot(&root->root_item, trans->transid); |
1113 | memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); | |
08fe4db1 | 1114 | btrfs_check_and_init_root_item(new_root_item); |
6bdb72de | 1115 | |
b83cc969 LZ |
1116 | root_flags = btrfs_root_flags(new_root_item); |
1117 | if (pending->readonly) | |
1118 | root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; | |
1119 | else | |
1120 | root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; | |
1121 | btrfs_set_root_flags(new_root_item, root_flags); | |
1122 | ||
8ea05e3a AB |
1123 | btrfs_set_root_generation_v2(new_root_item, |
1124 | trans->transid); | |
1125 | uuid_le_gen(&new_uuid); | |
1126 | memcpy(new_root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE); | |
1127 | memcpy(new_root_item->parent_uuid, root->root_item.uuid, | |
1128 | BTRFS_UUID_SIZE); | |
1129 | new_root_item->otime.sec = cpu_to_le64(cur_time.tv_sec); | |
dadd1105 | 1130 | new_root_item->otime.nsec = cpu_to_le32(cur_time.tv_nsec); |
8ea05e3a AB |
1131 | btrfs_set_root_otransid(new_root_item, trans->transid); |
1132 | memset(&new_root_item->stime, 0, sizeof(new_root_item->stime)); | |
1133 | memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime)); | |
1134 | btrfs_set_root_stransid(new_root_item, 0); | |
1135 | btrfs_set_root_rtransid(new_root_item, 0); | |
1136 | ||
6bdb72de | 1137 | old = btrfs_lock_root_node(root); |
49b25e05 | 1138 | ret = btrfs_cow_block(trans, root, old, NULL, 0, &old); |
79787eaa JM |
1139 | if (ret) { |
1140 | btrfs_tree_unlock(old); | |
1141 | free_extent_buffer(old); | |
8732d44f MX |
1142 | btrfs_abort_transaction(trans, root, ret); |
1143 | goto fail; | |
79787eaa | 1144 | } |
49b25e05 | 1145 | |
6bdb72de SW |
1146 | btrfs_set_lock_blocking(old); |
1147 | ||
49b25e05 | 1148 | ret = btrfs_copy_root(trans, root, old, &tmp, objectid); |
79787eaa | 1149 | /* clean up in any case */ |
6bdb72de SW |
1150 | btrfs_tree_unlock(old); |
1151 | free_extent_buffer(old); | |
8732d44f MX |
1152 | if (ret) { |
1153 | btrfs_abort_transaction(trans, root, ret); | |
1154 | goto fail; | |
1155 | } | |
6bdb72de | 1156 | |
f1ebcc74 LB |
1157 | /* see comments in should_cow_block() */ |
1158 | root->force_cow = 1; | |
1159 | smp_wmb(); | |
1160 | ||
6bdb72de | 1161 | btrfs_set_root_node(new_root_item, tmp); |
a22285a6 YZ |
1162 | /* record when the snapshot was created in key.offset */ |
1163 | key.offset = trans->transid; | |
1164 | ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); | |
6bdb72de SW |
1165 | btrfs_tree_unlock(tmp); |
1166 | free_extent_buffer(tmp); | |
8732d44f MX |
1167 | if (ret) { |
1168 | btrfs_abort_transaction(trans, root, ret); | |
1169 | goto fail; | |
1170 | } | |
6bdb72de | 1171 | |
a22285a6 YZ |
1172 | /* |
1173 | * insert root back/forward references | |
1174 | */ | |
1175 | ret = btrfs_add_root_ref(trans, tree_root, objectid, | |
0660b5af | 1176 | parent_root->root_key.objectid, |
33345d01 | 1177 | btrfs_ino(parent_inode), index, |
a22285a6 | 1178 | dentry->d_name.name, dentry->d_name.len); |
8732d44f MX |
1179 | if (ret) { |
1180 | btrfs_abort_transaction(trans, root, ret); | |
1181 | goto fail; | |
1182 | } | |
0660b5af | 1183 | |
a22285a6 YZ |
1184 | key.offset = (u64)-1; |
1185 | pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key); | |
79787eaa JM |
1186 | if (IS_ERR(pending->snap)) { |
1187 | ret = PTR_ERR(pending->snap); | |
8732d44f MX |
1188 | btrfs_abort_transaction(trans, root, ret); |
1189 | goto fail; | |
79787eaa | 1190 | } |
d68fc57b | 1191 | |
49b25e05 | 1192 | ret = btrfs_reloc_post_snapshot(trans, pending); |
8732d44f MX |
1193 | if (ret) { |
1194 | btrfs_abort_transaction(trans, root, ret); | |
1195 | goto fail; | |
1196 | } | |
361048f5 MX |
1197 | |
1198 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); | |
8732d44f MX |
1199 | if (ret) { |
1200 | btrfs_abort_transaction(trans, root, ret); | |
1201 | goto fail; | |
1202 | } | |
42874b3d MX |
1203 | |
1204 | ret = btrfs_insert_dir_item(trans, parent_root, | |
1205 | dentry->d_name.name, dentry->d_name.len, | |
1206 | parent_inode, &key, | |
1207 | BTRFS_FT_DIR, index); | |
1208 | /* We have check then name at the beginning, so it is impossible. */ | |
9c52057c | 1209 | BUG_ON(ret == -EEXIST || ret == -EOVERFLOW); |
8732d44f MX |
1210 | if (ret) { |
1211 | btrfs_abort_transaction(trans, root, ret); | |
1212 | goto fail; | |
1213 | } | |
42874b3d MX |
1214 | |
1215 | btrfs_i_size_write(parent_inode, parent_inode->i_size + | |
1216 | dentry->d_name.len * 2); | |
1217 | parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME; | |
be6aef60 | 1218 | ret = btrfs_update_inode_fallback(trans, parent_root, parent_inode); |
42874b3d | 1219 | if (ret) |
8732d44f | 1220 | btrfs_abort_transaction(trans, root, ret); |
3063d29f | 1221 | fail: |
6fa9700e | 1222 | dput(parent); |
98c9942a | 1223 | trans->block_rsv = rsv; |
6fa9700e MX |
1224 | no_free_objectid: |
1225 | kfree(new_root_item); | |
1226 | root_item_alloc_fail: | |
42874b3d MX |
1227 | btrfs_free_path(path); |
1228 | path_alloc_fail: | |
a22285a6 | 1229 | btrfs_block_rsv_release(root, &pending->block_rsv, (u64)-1); |
49b25e05 | 1230 | return ret; |
3063d29f CM |
1231 | } |
1232 | ||
d352ac68 CM |
1233 | /* |
1234 | * create all the snapshots we've scheduled for creation | |
1235 | */ | |
80b6794d CM |
1236 | static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans, |
1237 | struct btrfs_fs_info *fs_info) | |
3de4586c CM |
1238 | { |
1239 | struct btrfs_pending_snapshot *pending; | |
1240 | struct list_head *head = &trans->transaction->pending_snapshots; | |
3de4586c | 1241 | |
fe66a05a CM |
1242 | list_for_each_entry(pending, head, list) |
1243 | create_pending_snapshot(trans, fs_info, pending); | |
3de4586c CM |
1244 | return 0; |
1245 | } | |
1246 | ||
5d4f98a2 YZ |
1247 | static void update_super_roots(struct btrfs_root *root) |
1248 | { | |
1249 | struct btrfs_root_item *root_item; | |
1250 | struct btrfs_super_block *super; | |
1251 | ||
6c41761f | 1252 | super = root->fs_info->super_copy; |
5d4f98a2 YZ |
1253 | |
1254 | root_item = &root->fs_info->chunk_root->root_item; | |
1255 | super->chunk_root = root_item->bytenr; | |
1256 | super->chunk_root_generation = root_item->generation; | |
1257 | super->chunk_root_level = root_item->level; | |
1258 | ||
1259 | root_item = &root->fs_info->tree_root->root_item; | |
1260 | super->root = root_item->bytenr; | |
1261 | super->generation = root_item->generation; | |
1262 | super->root_level = root_item->level; | |
73bc1876 | 1263 | if (btrfs_test_opt(root, SPACE_CACHE)) |
0af3d00b | 1264 | super->cache_generation = root_item->generation; |
5d4f98a2 YZ |
1265 | } |
1266 | ||
f36f3042 CM |
1267 | int btrfs_transaction_in_commit(struct btrfs_fs_info *info) |
1268 | { | |
1269 | int ret = 0; | |
a4abeea4 | 1270 | spin_lock(&info->trans_lock); |
f36f3042 CM |
1271 | if (info->running_transaction) |
1272 | ret = info->running_transaction->in_commit; | |
a4abeea4 | 1273 | spin_unlock(&info->trans_lock); |
f36f3042 CM |
1274 | return ret; |
1275 | } | |
1276 | ||
8929ecfa YZ |
1277 | int btrfs_transaction_blocked(struct btrfs_fs_info *info) |
1278 | { | |
1279 | int ret = 0; | |
a4abeea4 | 1280 | spin_lock(&info->trans_lock); |
8929ecfa YZ |
1281 | if (info->running_transaction) |
1282 | ret = info->running_transaction->blocked; | |
a4abeea4 | 1283 | spin_unlock(&info->trans_lock); |
8929ecfa YZ |
1284 | return ret; |
1285 | } | |
1286 | ||
bb9c12c9 SW |
1287 | /* |
1288 | * wait for the current transaction commit to start and block subsequent | |
1289 | * transaction joins | |
1290 | */ | |
1291 | static void wait_current_trans_commit_start(struct btrfs_root *root, | |
1292 | struct btrfs_transaction *trans) | |
1293 | { | |
72d63ed6 | 1294 | wait_event(root->fs_info->transaction_blocked_wait, trans->in_commit); |
bb9c12c9 SW |
1295 | } |
1296 | ||
1297 | /* | |
1298 | * wait for the current transaction to start and then become unblocked. | |
1299 | * caller holds ref. | |
1300 | */ | |
1301 | static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root, | |
1302 | struct btrfs_transaction *trans) | |
1303 | { | |
72d63ed6 LZ |
1304 | wait_event(root->fs_info->transaction_wait, |
1305 | trans->commit_done || (trans->in_commit && !trans->blocked)); | |
bb9c12c9 SW |
1306 | } |
1307 | ||
1308 | /* | |
1309 | * commit transactions asynchronously. once btrfs_commit_transaction_async | |
1310 | * returns, any subsequent transaction will not be allowed to join. | |
1311 | */ | |
1312 | struct btrfs_async_commit { | |
1313 | struct btrfs_trans_handle *newtrans; | |
1314 | struct btrfs_root *root; | |
7892b5af | 1315 | struct work_struct work; |
bb9c12c9 SW |
1316 | }; |
1317 | ||
1318 | static void do_async_commit(struct work_struct *work) | |
1319 | { | |
1320 | struct btrfs_async_commit *ac = | |
7892b5af | 1321 | container_of(work, struct btrfs_async_commit, work); |
bb9c12c9 | 1322 | |
6fc4e354 SW |
1323 | /* |
1324 | * We've got freeze protection passed with the transaction. | |
1325 | * Tell lockdep about it. | |
1326 | */ | |
ff7c1d33 MX |
1327 | if (ac->newtrans->type < TRANS_JOIN_NOLOCK) |
1328 | rwsem_acquire_read( | |
1329 | &ac->root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1], | |
1330 | 0, 1, _THIS_IP_); | |
6fc4e354 | 1331 | |
e209db7a SW |
1332 | current->journal_info = ac->newtrans; |
1333 | ||
bb9c12c9 SW |
1334 | btrfs_commit_transaction(ac->newtrans, ac->root); |
1335 | kfree(ac); | |
1336 | } | |
1337 | ||
1338 | int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans, | |
1339 | struct btrfs_root *root, | |
1340 | int wait_for_unblock) | |
1341 | { | |
1342 | struct btrfs_async_commit *ac; | |
1343 | struct btrfs_transaction *cur_trans; | |
1344 | ||
1345 | ac = kmalloc(sizeof(*ac), GFP_NOFS); | |
db5b493a TI |
1346 | if (!ac) |
1347 | return -ENOMEM; | |
bb9c12c9 | 1348 | |
7892b5af | 1349 | INIT_WORK(&ac->work, do_async_commit); |
bb9c12c9 | 1350 | ac->root = root; |
7a7eaa40 | 1351 | ac->newtrans = btrfs_join_transaction(root); |
3612b495 TI |
1352 | if (IS_ERR(ac->newtrans)) { |
1353 | int err = PTR_ERR(ac->newtrans); | |
1354 | kfree(ac); | |
1355 | return err; | |
1356 | } | |
bb9c12c9 SW |
1357 | |
1358 | /* take transaction reference */ | |
bb9c12c9 | 1359 | cur_trans = trans->transaction; |
13c5a93e | 1360 | atomic_inc(&cur_trans->use_count); |
bb9c12c9 SW |
1361 | |
1362 | btrfs_end_transaction(trans, root); | |
6fc4e354 SW |
1363 | |
1364 | /* | |
1365 | * Tell lockdep we've released the freeze rwsem, since the | |
1366 | * async commit thread will be the one to unlock it. | |
1367 | */ | |
ff7c1d33 MX |
1368 | if (trans->type < TRANS_JOIN_NOLOCK) |
1369 | rwsem_release( | |
1370 | &root->fs_info->sb->s_writers.lock_map[SB_FREEZE_FS-1], | |
1371 | 1, _THIS_IP_); | |
6fc4e354 | 1372 | |
7892b5af | 1373 | schedule_work(&ac->work); |
bb9c12c9 SW |
1374 | |
1375 | /* wait for transaction to start and unblock */ | |
bb9c12c9 SW |
1376 | if (wait_for_unblock) |
1377 | wait_current_trans_commit_start_and_unblock(root, cur_trans); | |
1378 | else | |
1379 | wait_current_trans_commit_start(root, cur_trans); | |
bb9c12c9 | 1380 | |
38e88054 SW |
1381 | if (current->journal_info == trans) |
1382 | current->journal_info = NULL; | |
1383 | ||
1384 | put_transaction(cur_trans); | |
bb9c12c9 SW |
1385 | return 0; |
1386 | } | |
1387 | ||
49b25e05 JM |
1388 | |
1389 | static void cleanup_transaction(struct btrfs_trans_handle *trans, | |
7b8b92af | 1390 | struct btrfs_root *root, int err) |
49b25e05 JM |
1391 | { |
1392 | struct btrfs_transaction *cur_trans = trans->transaction; | |
1393 | ||
1394 | WARN_ON(trans->use_count > 1); | |
1395 | ||
7b8b92af JB |
1396 | btrfs_abort_transaction(trans, root, err); |
1397 | ||
49b25e05 JM |
1398 | spin_lock(&root->fs_info->trans_lock); |
1399 | list_del_init(&cur_trans->list); | |
d7096fc3 JB |
1400 | if (cur_trans == root->fs_info->running_transaction) { |
1401 | root->fs_info->running_transaction = NULL; | |
1402 | root->fs_info->trans_no_join = 0; | |
1403 | } | |
49b25e05 JM |
1404 | spin_unlock(&root->fs_info->trans_lock); |
1405 | ||
1406 | btrfs_cleanup_one_transaction(trans->transaction, root); | |
1407 | ||
1408 | put_transaction(cur_trans); | |
1409 | put_transaction(cur_trans); | |
1410 | ||
1411 | trace_btrfs_transaction_commit(root); | |
1412 | ||
1413 | btrfs_scrub_continue(root); | |
1414 | ||
1415 | if (current->journal_info == trans) | |
1416 | current->journal_info = NULL; | |
1417 | ||
1418 | kmem_cache_free(btrfs_trans_handle_cachep, trans); | |
1419 | } | |
1420 | ||
ca469637 MX |
1421 | static int btrfs_flush_all_pending_stuffs(struct btrfs_trans_handle *trans, |
1422 | struct btrfs_root *root) | |
1423 | { | |
1424 | int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT); | |
1425 | int snap_pending = 0; | |
1426 | int ret; | |
1427 | ||
1428 | if (!flush_on_commit) { | |
1429 | spin_lock(&root->fs_info->trans_lock); | |
1430 | if (!list_empty(&trans->transaction->pending_snapshots)) | |
1431 | snap_pending = 1; | |
1432 | spin_unlock(&root->fs_info->trans_lock); | |
1433 | } | |
1434 | ||
1435 | if (flush_on_commit || snap_pending) { | |
3edb2a68 MX |
1436 | ret = btrfs_start_delalloc_inodes(root, 1); |
1437 | if (ret) | |
1438 | return ret; | |
ca469637 MX |
1439 | btrfs_wait_ordered_extents(root, 1); |
1440 | } | |
1441 | ||
1442 | ret = btrfs_run_delayed_items(trans, root); | |
1443 | if (ret) | |
1444 | return ret; | |
1445 | ||
1446 | /* | |
1447 | * running the delayed items may have added new refs. account | |
1448 | * them now so that they hinder processing of more delayed refs | |
1449 | * as little as possible. | |
1450 | */ | |
1451 | btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info); | |
1452 | ||
1453 | /* | |
1454 | * rename don't use btrfs_join_transaction, so, once we | |
1455 | * set the transaction to blocked above, we aren't going | |
1456 | * to get any new ordered operations. We can safely run | |
1457 | * it here and no for sure that nothing new will be added | |
1458 | * to the list | |
1459 | */ | |
569e0f35 | 1460 | ret = btrfs_run_ordered_operations(trans, root, 1); |
ca469637 | 1461 | |
eebc6084 | 1462 | return ret; |
ca469637 MX |
1463 | } |
1464 | ||
bb9c12c9 SW |
1465 | /* |
1466 | * btrfs_transaction state sequence: | |
1467 | * in_commit = 0, blocked = 0 (initial) | |
1468 | * in_commit = 1, blocked = 1 | |
1469 | * blocked = 0 | |
1470 | * commit_done = 1 | |
1471 | */ | |
79154b1b CM |
1472 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans, |
1473 | struct btrfs_root *root) | |
1474 | { | |
15ee9bc7 | 1475 | unsigned long joined = 0; |
49b25e05 | 1476 | struct btrfs_transaction *cur_trans = trans->transaction; |
8fd17795 | 1477 | struct btrfs_transaction *prev_trans = NULL; |
79154b1b | 1478 | DEFINE_WAIT(wait); |
25287e0a | 1479 | int ret; |
89573b9c CM |
1480 | int should_grow = 0; |
1481 | unsigned long now = get_seconds(); | |
79154b1b | 1482 | |
569e0f35 | 1483 | ret = btrfs_run_ordered_operations(trans, root, 0); |
25287e0a MX |
1484 | if (ret) { |
1485 | btrfs_abort_transaction(trans, root, ret); | |
e4a2bcac JB |
1486 | btrfs_end_transaction(trans, root); |
1487 | return ret; | |
25287e0a | 1488 | } |
5a3f23d5 | 1489 | |
8d25a086 MX |
1490 | /* Stop the commit early if ->aborted is set */ |
1491 | if (unlikely(ACCESS_ONCE(cur_trans->aborted))) { | |
25287e0a | 1492 | ret = cur_trans->aborted; |
e4a2bcac JB |
1493 | btrfs_end_transaction(trans, root); |
1494 | return ret; | |
25287e0a | 1495 | } |
49b25e05 | 1496 | |
56bec294 CM |
1497 | /* make a pass through all the delayed refs we have so far |
1498 | * any runnings procs may add more while we are here | |
1499 | */ | |
1500 | ret = btrfs_run_delayed_refs(trans, root, 0); | |
e4a2bcac JB |
1501 | if (ret) { |
1502 | btrfs_end_transaction(trans, root); | |
1503 | return ret; | |
1504 | } | |
56bec294 | 1505 | |
0e721106 JB |
1506 | btrfs_trans_release_metadata(trans, root); |
1507 | trans->block_rsv = NULL; | |
1508 | ||
b7ec40d7 | 1509 | cur_trans = trans->transaction; |
49b25e05 | 1510 | |
56bec294 CM |
1511 | /* |
1512 | * set the flushing flag so procs in this transaction have to | |
1513 | * start sending their work down. | |
1514 | */ | |
b7ec40d7 | 1515 | cur_trans->delayed_refs.flushing = 1; |
56bec294 | 1516 | |
ea658bad JB |
1517 | if (!list_empty(&trans->new_bgs)) |
1518 | btrfs_create_pending_block_groups(trans, root); | |
1519 | ||
c3e69d58 | 1520 | ret = btrfs_run_delayed_refs(trans, root, 0); |
e4a2bcac JB |
1521 | if (ret) { |
1522 | btrfs_end_transaction(trans, root); | |
1523 | return ret; | |
1524 | } | |
56bec294 | 1525 | |
a4abeea4 | 1526 | spin_lock(&cur_trans->commit_lock); |
b7ec40d7 | 1527 | if (cur_trans->in_commit) { |
a4abeea4 | 1528 | spin_unlock(&cur_trans->commit_lock); |
13c5a93e | 1529 | atomic_inc(&cur_trans->use_count); |
49b25e05 | 1530 | ret = btrfs_end_transaction(trans, root); |
ccd467d6 | 1531 | |
b9c8300c | 1532 | wait_for_commit(root, cur_trans); |
15ee9bc7 | 1533 | |
79154b1b | 1534 | put_transaction(cur_trans); |
15ee9bc7 | 1535 | |
49b25e05 | 1536 | return ret; |
79154b1b | 1537 | } |
4313b399 | 1538 | |
2c90e5d6 | 1539 | trans->transaction->in_commit = 1; |
f9295749 | 1540 | trans->transaction->blocked = 1; |
a4abeea4 | 1541 | spin_unlock(&cur_trans->commit_lock); |
bb9c12c9 SW |
1542 | wake_up(&root->fs_info->transaction_blocked_wait); |
1543 | ||
a4abeea4 | 1544 | spin_lock(&root->fs_info->trans_lock); |
ccd467d6 CM |
1545 | if (cur_trans->list.prev != &root->fs_info->trans_list) { |
1546 | prev_trans = list_entry(cur_trans->list.prev, | |
1547 | struct btrfs_transaction, list); | |
1548 | if (!prev_trans->commit_done) { | |
13c5a93e | 1549 | atomic_inc(&prev_trans->use_count); |
a4abeea4 | 1550 | spin_unlock(&root->fs_info->trans_lock); |
ccd467d6 CM |
1551 | |
1552 | wait_for_commit(root, prev_trans); | |
ccd467d6 | 1553 | |
15ee9bc7 | 1554 | put_transaction(prev_trans); |
a4abeea4 JB |
1555 | } else { |
1556 | spin_unlock(&root->fs_info->trans_lock); | |
ccd467d6 | 1557 | } |
a4abeea4 JB |
1558 | } else { |
1559 | spin_unlock(&root->fs_info->trans_lock); | |
ccd467d6 | 1560 | } |
15ee9bc7 | 1561 | |
e39e64ac CM |
1562 | if (!btrfs_test_opt(root, SSD) && |
1563 | (now < cur_trans->start_time || now - cur_trans->start_time < 1)) | |
89573b9c CM |
1564 | should_grow = 1; |
1565 | ||
15ee9bc7 JB |
1566 | do { |
1567 | joined = cur_trans->num_joined; | |
7ea394f1 | 1568 | |
2c90e5d6 | 1569 | WARN_ON(cur_trans != trans->transaction); |
15ee9bc7 | 1570 | |
ca469637 | 1571 | ret = btrfs_flush_all_pending_stuffs(trans, root); |
49b25e05 JM |
1572 | if (ret) |
1573 | goto cleanup_transaction; | |
16cdcec7 | 1574 | |
ed3b3d31 CM |
1575 | prepare_to_wait(&cur_trans->writer_wait, &wait, |
1576 | TASK_UNINTERRUPTIBLE); | |
1577 | ||
13c5a93e | 1578 | if (atomic_read(&cur_trans->num_writers) > 1) |
99d16cbc SW |
1579 | schedule_timeout(MAX_SCHEDULE_TIMEOUT); |
1580 | else if (should_grow) | |
1581 | schedule_timeout(1); | |
15ee9bc7 | 1582 | |
15ee9bc7 | 1583 | finish_wait(&cur_trans->writer_wait, &wait); |
13c5a93e | 1584 | } while (atomic_read(&cur_trans->num_writers) > 1 || |
89573b9c | 1585 | (should_grow && cur_trans->num_joined != joined)); |
15ee9bc7 | 1586 | |
ca469637 MX |
1587 | ret = btrfs_flush_all_pending_stuffs(trans, root); |
1588 | if (ret) | |
1589 | goto cleanup_transaction; | |
1590 | ||
ed0ca140 JB |
1591 | /* |
1592 | * Ok now we need to make sure to block out any other joins while we | |
1593 | * commit the transaction. We could have started a join before setting | |
1594 | * no_join so make sure to wait for num_writers to == 1 again. | |
1595 | */ | |
1596 | spin_lock(&root->fs_info->trans_lock); | |
1597 | root->fs_info->trans_no_join = 1; | |
1598 | spin_unlock(&root->fs_info->trans_lock); | |
1599 | wait_event(cur_trans->writer_wait, | |
1600 | atomic_read(&cur_trans->num_writers) == 1); | |
1601 | ||
2cba30f1 MX |
1602 | /* ->aborted might be set after the previous check, so check it */ |
1603 | if (unlikely(ACCESS_ONCE(cur_trans->aborted))) { | |
1604 | ret = cur_trans->aborted; | |
1605 | goto cleanup_transaction; | |
1606 | } | |
7585717f CM |
1607 | /* |
1608 | * the reloc mutex makes sure that we stop | |
1609 | * the balancing code from coming in and moving | |
1610 | * extents around in the middle of the commit | |
1611 | */ | |
1612 | mutex_lock(&root->fs_info->reloc_mutex); | |
1613 | ||
42874b3d MX |
1614 | /* |
1615 | * We needn't worry about the delayed items because we will | |
1616 | * deal with them in create_pending_snapshot(), which is the | |
1617 | * core function of the snapshot creation. | |
1618 | */ | |
1619 | ret = create_pending_snapshots(trans, root->fs_info); | |
49b25e05 JM |
1620 | if (ret) { |
1621 | mutex_unlock(&root->fs_info->reloc_mutex); | |
1622 | goto cleanup_transaction; | |
1623 | } | |
3063d29f | 1624 | |
42874b3d MX |
1625 | /* |
1626 | * We insert the dir indexes of the snapshots and update the inode | |
1627 | * of the snapshots' parents after the snapshot creation, so there | |
1628 | * are some delayed items which are not dealt with. Now deal with | |
1629 | * them. | |
1630 | * | |
1631 | * We needn't worry that this operation will corrupt the snapshots, | |
1632 | * because all the tree which are snapshoted will be forced to COW | |
1633 | * the nodes and leaves. | |
1634 | */ | |
1635 | ret = btrfs_run_delayed_items(trans, root); | |
49b25e05 JM |
1636 | if (ret) { |
1637 | mutex_unlock(&root->fs_info->reloc_mutex); | |
1638 | goto cleanup_transaction; | |
1639 | } | |
16cdcec7 | 1640 | |
56bec294 | 1641 | ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); |
49b25e05 JM |
1642 | if (ret) { |
1643 | mutex_unlock(&root->fs_info->reloc_mutex); | |
1644 | goto cleanup_transaction; | |
1645 | } | |
56bec294 | 1646 | |
e999376f CM |
1647 | /* |
1648 | * make sure none of the code above managed to slip in a | |
1649 | * delayed item | |
1650 | */ | |
1651 | btrfs_assert_delayed_root_empty(root); | |
1652 | ||
2c90e5d6 | 1653 | WARN_ON(cur_trans != trans->transaction); |
dc17ff8f | 1654 | |
a2de733c | 1655 | btrfs_scrub_pause(root); |
e02119d5 CM |
1656 | /* btrfs_commit_tree_roots is responsible for getting the |
1657 | * various roots consistent with each other. Every pointer | |
1658 | * in the tree of tree roots has to point to the most up to date | |
1659 | * root for every subvolume and other tree. So, we have to keep | |
1660 | * the tree logging code from jumping in and changing any | |
1661 | * of the trees. | |
1662 | * | |
1663 | * At this point in the commit, there can't be any tree-log | |
1664 | * writers, but a little lower down we drop the trans mutex | |
1665 | * and let new people in. By holding the tree_log_mutex | |
1666 | * from now until after the super is written, we avoid races | |
1667 | * with the tree-log code. | |
1668 | */ | |
1669 | mutex_lock(&root->fs_info->tree_log_mutex); | |
1670 | ||
5d4f98a2 | 1671 | ret = commit_fs_roots(trans, root); |
49b25e05 JM |
1672 | if (ret) { |
1673 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
871383be | 1674 | mutex_unlock(&root->fs_info->reloc_mutex); |
49b25e05 JM |
1675 | goto cleanup_transaction; |
1676 | } | |
54aa1f4d | 1677 | |
5d4f98a2 | 1678 | /* commit_fs_roots gets rid of all the tree log roots, it is now |
e02119d5 CM |
1679 | * safe to free the root of tree log roots |
1680 | */ | |
1681 | btrfs_free_log_root_tree(trans, root->fs_info); | |
1682 | ||
5d4f98a2 | 1683 | ret = commit_cowonly_roots(trans, root); |
49b25e05 JM |
1684 | if (ret) { |
1685 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
871383be | 1686 | mutex_unlock(&root->fs_info->reloc_mutex); |
49b25e05 JM |
1687 | goto cleanup_transaction; |
1688 | } | |
54aa1f4d | 1689 | |
2cba30f1 MX |
1690 | /* |
1691 | * The tasks which save the space cache and inode cache may also | |
1692 | * update ->aborted, check it. | |
1693 | */ | |
1694 | if (unlikely(ACCESS_ONCE(cur_trans->aborted))) { | |
1695 | ret = cur_trans->aborted; | |
1696 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1697 | mutex_unlock(&root->fs_info->reloc_mutex); | |
1698 | goto cleanup_transaction; | |
1699 | } | |
1700 | ||
11833d66 YZ |
1701 | btrfs_prepare_extent_commit(trans, root); |
1702 | ||
78fae27e | 1703 | cur_trans = root->fs_info->running_transaction; |
5d4f98a2 YZ |
1704 | |
1705 | btrfs_set_root_node(&root->fs_info->tree_root->root_item, | |
1706 | root->fs_info->tree_root->node); | |
817d52f8 | 1707 | switch_commit_root(root->fs_info->tree_root); |
5d4f98a2 YZ |
1708 | |
1709 | btrfs_set_root_node(&root->fs_info->chunk_root->root_item, | |
1710 | root->fs_info->chunk_root->node); | |
817d52f8 | 1711 | switch_commit_root(root->fs_info->chunk_root); |
5d4f98a2 | 1712 | |
edf39272 | 1713 | assert_qgroups_uptodate(trans); |
5d4f98a2 | 1714 | update_super_roots(root); |
e02119d5 CM |
1715 | |
1716 | if (!root->fs_info->log_root_recovering) { | |
6c41761f DS |
1717 | btrfs_set_super_log_root(root->fs_info->super_copy, 0); |
1718 | btrfs_set_super_log_root_level(root->fs_info->super_copy, 0); | |
e02119d5 CM |
1719 | } |
1720 | ||
6c41761f DS |
1721 | memcpy(root->fs_info->super_for_commit, root->fs_info->super_copy, |
1722 | sizeof(*root->fs_info->super_copy)); | |
ccd467d6 | 1723 | |
f9295749 | 1724 | trans->transaction->blocked = 0; |
a4abeea4 JB |
1725 | spin_lock(&root->fs_info->trans_lock); |
1726 | root->fs_info->running_transaction = NULL; | |
1727 | root->fs_info->trans_no_join = 0; | |
1728 | spin_unlock(&root->fs_info->trans_lock); | |
7585717f | 1729 | mutex_unlock(&root->fs_info->reloc_mutex); |
b7ec40d7 | 1730 | |
f9295749 | 1731 | wake_up(&root->fs_info->transaction_wait); |
e6dcd2dc | 1732 | |
79154b1b | 1733 | ret = btrfs_write_and_wait_transaction(trans, root); |
49b25e05 JM |
1734 | if (ret) { |
1735 | btrfs_error(root->fs_info, ret, | |
1736 | "Error while writing out transaction."); | |
1737 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1738 | goto cleanup_transaction; | |
1739 | } | |
1740 | ||
1741 | ret = write_ctree_super(trans, root, 0); | |
1742 | if (ret) { | |
1743 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1744 | goto cleanup_transaction; | |
1745 | } | |
4313b399 | 1746 | |
e02119d5 CM |
1747 | /* |
1748 | * the super is written, we can safely allow the tree-loggers | |
1749 | * to go about their business | |
1750 | */ | |
1751 | mutex_unlock(&root->fs_info->tree_log_mutex); | |
1752 | ||
11833d66 | 1753 | btrfs_finish_extent_commit(trans, root); |
4313b399 | 1754 | |
2c90e5d6 | 1755 | cur_trans->commit_done = 1; |
b7ec40d7 | 1756 | |
15ee9bc7 | 1757 | root->fs_info->last_trans_committed = cur_trans->transid; |
817d52f8 | 1758 | |
2c90e5d6 | 1759 | wake_up(&cur_trans->commit_wait); |
3de4586c | 1760 | |
a4abeea4 | 1761 | spin_lock(&root->fs_info->trans_lock); |
13c5a93e | 1762 | list_del_init(&cur_trans->list); |
a4abeea4 JB |
1763 | spin_unlock(&root->fs_info->trans_lock); |
1764 | ||
78fae27e | 1765 | put_transaction(cur_trans); |
79154b1b | 1766 | put_transaction(cur_trans); |
58176a96 | 1767 | |
354aa0fb MX |
1768 | if (trans->type < TRANS_JOIN_NOLOCK) |
1769 | sb_end_intwrite(root->fs_info->sb); | |
b2b5ef5c | 1770 | |
1abe9b8a | 1771 | trace_btrfs_transaction_commit(root); |
1772 | ||
a2de733c AJ |
1773 | btrfs_scrub_continue(root); |
1774 | ||
9ed74f2d JB |
1775 | if (current->journal_info == trans) |
1776 | current->journal_info = NULL; | |
1777 | ||
2c90e5d6 | 1778 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
24bbcf04 YZ |
1779 | |
1780 | if (current != root->fs_info->transaction_kthread) | |
1781 | btrfs_run_delayed_iputs(root); | |
1782 | ||
79154b1b | 1783 | return ret; |
49b25e05 JM |
1784 | |
1785 | cleanup_transaction: | |
0e721106 JB |
1786 | btrfs_trans_release_metadata(trans, root); |
1787 | trans->block_rsv = NULL; | |
49b25e05 JM |
1788 | btrfs_printk(root->fs_info, "Skipping commit of aborted transaction.\n"); |
1789 | // WARN_ON(1); | |
1790 | if (current->journal_info == trans) | |
1791 | current->journal_info = NULL; | |
7b8b92af | 1792 | cleanup_transaction(trans, root, ret); |
49b25e05 JM |
1793 | |
1794 | return ret; | |
79154b1b CM |
1795 | } |
1796 | ||
d352ac68 CM |
1797 | /* |
1798 | * interface function to delete all the snapshots we have scheduled for deletion | |
1799 | */ | |
e9d0b13b CM |
1800 | int btrfs_clean_old_snapshots(struct btrfs_root *root) |
1801 | { | |
5d4f98a2 YZ |
1802 | LIST_HEAD(list); |
1803 | struct btrfs_fs_info *fs_info = root->fs_info; | |
1804 | ||
a4abeea4 | 1805 | spin_lock(&fs_info->trans_lock); |
5d4f98a2 | 1806 | list_splice_init(&fs_info->dead_roots, &list); |
a4abeea4 | 1807 | spin_unlock(&fs_info->trans_lock); |
e9d0b13b | 1808 | |
5d4f98a2 | 1809 | while (!list_empty(&list)) { |
2c536799 JM |
1810 | int ret; |
1811 | ||
5d4f98a2 | 1812 | root = list_entry(list.next, struct btrfs_root, root_list); |
76dda93c YZ |
1813 | list_del(&root->root_list); |
1814 | ||
16cdcec7 MX |
1815 | btrfs_kill_all_delayed_nodes(root); |
1816 | ||
76dda93c YZ |
1817 | if (btrfs_header_backref_rev(root->node) < |
1818 | BTRFS_MIXED_BACKREF_REV) | |
2c536799 | 1819 | ret = btrfs_drop_snapshot(root, NULL, 0, 0); |
76dda93c | 1820 | else |
2c536799 JM |
1821 | ret =btrfs_drop_snapshot(root, NULL, 1, 0); |
1822 | BUG_ON(ret < 0); | |
e9d0b13b CM |
1823 | } |
1824 | return 0; | |
1825 | } |