1 | /* SPDX-License-Identifier: GPL-2.0 */ |
2 | /* |
3 | * Copyright (C) 2007 Oracle. All rights reserved. |
4 | */ |
5 | |
6 | #ifndef BTRFS_TRANSACTION_H |
7 | #define BTRFS_TRANSACTION_H |
8 | |
9 | #include <linux/refcount.h> |
10 | #include "btrfs_inode.h" |
11 | #include "delayed-ref.h" |
12 | #include "ctree.h" |
13 | #include "misc.h" |
14 | |
15 | enum btrfs_trans_state { |
16 | TRANS_STATE_RUNNING, |
17 | TRANS_STATE_COMMIT_PREP, |
18 | TRANS_STATE_COMMIT_START, |
19 | TRANS_STATE_COMMIT_DOING, |
20 | TRANS_STATE_UNBLOCKED, |
21 | TRANS_STATE_SUPER_COMMITTED, |
22 | TRANS_STATE_COMPLETED, |
23 | TRANS_STATE_MAX, |
24 | }; |
25 | |
26 | #define BTRFS_TRANS_HAVE_FREE_BGS 0 |
27 | #define BTRFS_TRANS_DIRTY_BG_RUN 1 |
28 | #define BTRFS_TRANS_CACHE_ENOSPC 2 |
29 | |
30 | struct btrfs_transaction { |
31 | u64 transid; |
32 | /* |
33 | * total external writers(USERSPACE/START/ATTACH) in this |
34 | * transaction, it must be zero before the transaction is |
35 | * being committed |
36 | */ |
37 | atomic_t num_extwriters; |
38 | /* |
39 | * total writers in this transaction, it must be zero before the |
40 | * transaction can end |
41 | */ |
42 | atomic_t num_writers; |
43 | refcount_t use_count; |
44 | |
45 | unsigned long flags; |
46 | |
47 | /* Be protected by fs_info->trans_lock when we want to change it. */ |
48 | enum btrfs_trans_state state; |
49 | int aborted; |
50 | struct list_head list; |
51 | struct extent_io_tree dirty_pages; |
52 | time64_t start_time; |
53 | wait_queue_head_t writer_wait; |
54 | wait_queue_head_t commit_wait; |
55 | struct list_head pending_snapshots; |
56 | struct list_head dev_update_list; |
57 | struct list_head switch_commits; |
58 | struct list_head dirty_bgs; |
59 | |
60 | /* |
61 | * There is no explicit lock which protects io_bgs, rather its |
62 | * consistency is implied by the fact that all the sites which modify |
63 | * it do so under some form of transaction critical section, namely: |
64 | * |
65 | * - btrfs_start_dirty_block_groups - This function can only ever be |
66 | * run by one of the transaction committers. Refer to |
67 | * BTRFS_TRANS_DIRTY_BG_RUN usage in btrfs_commit_transaction |
68 | * |
69 | * - btrfs_write_dirty_blockgroups - this is called by |
70 | * commit_cowonly_roots from transaction critical section |
71 | * (TRANS_STATE_COMMIT_DOING) |
72 | * |
73 | * - btrfs_cleanup_dirty_bgs - called on transaction abort |
74 | */ |
75 | struct list_head io_bgs; |
76 | struct list_head dropped_roots; |
77 | struct extent_io_tree pinned_extents; |
78 | |
79 | /* |
80 | * we need to make sure block group deletion doesn't race with |
81 | * free space cache writeout. This mutex keeps them from stomping |
82 | * on each other |
83 | */ |
84 | struct mutex cache_write_mutex; |
85 | spinlock_t dirty_bgs_lock; |
86 | /* Protected by spin lock fs_info->unused_bgs_lock. */ |
87 | struct list_head deleted_bgs; |
88 | spinlock_t dropped_roots_lock; |
89 | struct btrfs_delayed_ref_root delayed_refs; |
90 | struct btrfs_fs_info *fs_info; |
91 | |
92 | /* |
93 | * Number of ordered extents the transaction must wait for before |
94 | * committing. These are ordered extents started by a fast fsync. |
95 | */ |
96 | atomic_t pending_ordered; |
97 | wait_queue_head_t pending_wait; |
98 | }; |
99 | |
100 | enum { |
101 | ENUM_BIT(__TRANS_FREEZABLE), |
102 | ENUM_BIT(__TRANS_START), |
103 | ENUM_BIT(__TRANS_ATTACH), |
104 | ENUM_BIT(__TRANS_JOIN), |
105 | ENUM_BIT(__TRANS_JOIN_NOLOCK), |
106 | ENUM_BIT(__TRANS_DUMMY), |
107 | ENUM_BIT(__TRANS_JOIN_NOSTART), |
108 | }; |
109 | |
110 | #define TRANS_START (__TRANS_START | __TRANS_FREEZABLE) |
111 | #define TRANS_ATTACH (__TRANS_ATTACH) |
112 | #define TRANS_JOIN (__TRANS_JOIN | __TRANS_FREEZABLE) |
113 | #define TRANS_JOIN_NOLOCK (__TRANS_JOIN_NOLOCK) |
114 | #define TRANS_JOIN_NOSTART (__TRANS_JOIN_NOSTART) |
115 | |
116 | #define TRANS_EXTWRITERS (__TRANS_START | __TRANS_ATTACH) |
117 | |
118 | struct btrfs_trans_handle { |
119 | u64 transid; |
120 | u64 bytes_reserved; |
121 | u64 delayed_refs_bytes_reserved; |
122 | u64 chunk_bytes_reserved; |
123 | unsigned long delayed_ref_updates; |
124 | unsigned long delayed_ref_csum_deletions; |
125 | struct btrfs_transaction *transaction; |
126 | struct btrfs_block_rsv *block_rsv; |
127 | struct btrfs_block_rsv *orig_rsv; |
128 | /* Set by a task that wants to create a snapshot. */ |
129 | struct btrfs_pending_snapshot *pending_snapshot; |
130 | refcount_t use_count; |
131 | unsigned int type; |
132 | /* |
133 | * Error code of transaction abort, set outside of locks and must use |
134 | * the READ_ONCE/WRITE_ONCE access |
135 | */ |
136 | short aborted; |
137 | bool adding_csums; |
138 | bool allocating_chunk; |
139 | bool removing_chunk; |
140 | bool reloc_reserved; |
141 | bool in_fsync; |
142 | struct btrfs_fs_info *fs_info; |
143 | struct list_head new_bgs; |
144 | struct btrfs_block_rsv delayed_rsv; |
145 | }; |
146 | |
147 | /* |
148 | * The abort status can be changed between calls and is not protected by locks. |
149 | * This accepts btrfs_transaction and btrfs_trans_handle as types. Once it's |
150 | * set to a non-zero value it does not change, so the macro should be in checks |
151 | * but is not necessary for further reads of the value. |
152 | */ |
153 | #define TRANS_ABORTED(trans) (unlikely(READ_ONCE((trans)->aborted))) |
154 | |
155 | struct btrfs_pending_snapshot { |
156 | struct dentry *dentry; |
157 | struct inode *dir; |
158 | struct btrfs_root *root; |
159 | struct btrfs_root_item *root_item; |
160 | struct btrfs_root *snap; |
161 | struct btrfs_qgroup_inherit *inherit; |
162 | struct btrfs_path *path; |
163 | /* block reservation for the operation */ |
164 | struct btrfs_block_rsv block_rsv; |
165 | /* extra metadata reservation for relocation */ |
166 | int error; |
167 | /* Preallocated anonymous block device number */ |
168 | dev_t anon_dev; |
169 | bool readonly; |
170 | struct list_head list; |
171 | }; |
172 | |
173 | static inline void btrfs_set_inode_last_trans(struct btrfs_trans_handle *trans, |
174 | struct btrfs_inode *inode) |
175 | { |
176 | spin_lock(lock: &inode->lock); |
177 | inode->last_trans = trans->transaction->transid; |
178 | inode->last_sub_trans = btrfs_get_root_log_transid(root: inode->root); |
179 | inode->last_log_commit = inode->last_sub_trans - 1; |
180 | spin_unlock(lock: &inode->lock); |
181 | } |
182 | |
183 | /* |
184 | * Make qgroup codes to skip given qgroupid, means the old/new_roots for |
185 | * qgroup won't contain the qgroupid in it. |
186 | */ |
187 | static inline void btrfs_set_skip_qgroup(struct btrfs_trans_handle *trans, |
188 | u64 qgroupid) |
189 | { |
190 | struct btrfs_delayed_ref_root *delayed_refs; |
191 | |
192 | delayed_refs = &trans->transaction->delayed_refs; |
193 | WARN_ON(delayed_refs->qgroup_to_skip); |
194 | delayed_refs->qgroup_to_skip = qgroupid; |
195 | } |
196 | |
197 | static inline void btrfs_clear_skip_qgroup(struct btrfs_trans_handle *trans) |
198 | { |
199 | struct btrfs_delayed_ref_root *delayed_refs; |
200 | |
201 | delayed_refs = &trans->transaction->delayed_refs; |
202 | WARN_ON(!delayed_refs->qgroup_to_skip); |
203 | delayed_refs->qgroup_to_skip = 0; |
204 | } |
205 | |
206 | bool __cold abort_should_print_stack(int error); |
207 | |
208 | /* |
209 | * Call btrfs_abort_transaction as early as possible when an error condition is |
210 | * detected, that way the exact stack trace is reported for some errors. |
211 | */ |
212 | #define btrfs_abort_transaction(trans, error) \ |
213 | do { \ |
214 | bool first = false; \ |
215 | /* Report first abort since mount */ \ |
216 | if (!test_and_set_bit(BTRFS_FS_STATE_TRANS_ABORTED, \ |
217 | &((trans)->fs_info->fs_state))) { \ |
218 | first = true; \ |
219 | if (WARN(abort_should_print_stack(error), \ |
220 | KERN_ERR \ |
221 | "BTRFS: Transaction aborted (error %d)\n", \ |
222 | (error))) { \ |
223 | /* Stack trace printed. */ \ |
224 | } else { \ |
225 | btrfs_err((trans)->fs_info, \ |
226 | "Transaction aborted (error %d)", \ |
227 | (error)); \ |
228 | } \ |
229 | } \ |
230 | __btrfs_abort_transaction((trans), __func__, \ |
231 | __LINE__, (error), first); \ |
232 | } while (0) |
233 | |
234 | int btrfs_end_transaction(struct btrfs_trans_handle *trans); |
235 | struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, |
236 | unsigned int num_items); |
237 | struct btrfs_trans_handle *btrfs_start_transaction_fallback_global_rsv( |
238 | struct btrfs_root *root, |
239 | unsigned int num_items); |
240 | struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root); |
241 | struct btrfs_trans_handle *btrfs_join_transaction_spacecache(struct btrfs_root *root); |
242 | struct btrfs_trans_handle *btrfs_join_transaction_nostart(struct btrfs_root *root); |
243 | struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root); |
244 | struct btrfs_trans_handle *btrfs_attach_transaction_barrier( |
245 | struct btrfs_root *root); |
246 | int btrfs_wait_for_commit(struct btrfs_fs_info *fs_info, u64 transid); |
247 | |
248 | void btrfs_add_dead_root(struct btrfs_root *root); |
249 | void btrfs_maybe_wake_unfinished_drop(struct btrfs_fs_info *fs_info); |
250 | int btrfs_clean_one_deleted_snapshot(struct btrfs_fs_info *fs_info); |
251 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans); |
252 | void btrfs_commit_transaction_async(struct btrfs_trans_handle *trans); |
253 | int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans); |
254 | bool btrfs_should_end_transaction(struct btrfs_trans_handle *trans); |
255 | void btrfs_throttle(struct btrfs_fs_info *fs_info); |
256 | int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, |
257 | struct btrfs_root *root); |
258 | int btrfs_write_marked_extents(struct btrfs_fs_info *fs_info, |
259 | struct extent_io_tree *dirty_pages, int mark); |
260 | int btrfs_wait_tree_log_extents(struct btrfs_root *root, int mark); |
261 | int btrfs_transaction_blocked(struct btrfs_fs_info *info); |
262 | int btrfs_transaction_in_commit(struct btrfs_fs_info *info); |
263 | void btrfs_put_transaction(struct btrfs_transaction *transaction); |
264 | void btrfs_add_dropped_root(struct btrfs_trans_handle *trans, |
265 | struct btrfs_root *root); |
266 | void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans); |
267 | void __cold __btrfs_abort_transaction(struct btrfs_trans_handle *trans, |
268 | const char *function, |
269 | unsigned int line, int error, bool first_hit); |
270 | |
271 | int __init btrfs_transaction_init(void); |
272 | void __cold btrfs_transaction_exit(void); |
273 | |
274 | #endif |
275 | |