Loading...
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/* Radix-tree tag for roots that are part of the trasaction. */
16#define BTRFS_ROOT_TRANS_TAG 0
17
18enum btrfs_trans_state {
19 TRANS_STATE_RUNNING,
20 TRANS_STATE_COMMIT_PREP,
21 TRANS_STATE_COMMIT_START,
22 TRANS_STATE_COMMIT_DOING,
23 TRANS_STATE_UNBLOCKED,
24 TRANS_STATE_SUPER_COMMITTED,
25 TRANS_STATE_COMPLETED,
26 TRANS_STATE_MAX,
27};
28
29#define BTRFS_TRANS_HAVE_FREE_BGS 0
30#define BTRFS_TRANS_DIRTY_BG_RUN 1
31#define BTRFS_TRANS_CACHE_ENOSPC 2
32
33struct btrfs_transaction {
34 u64 transid;
35 /*
36 * total external writers(USERSPACE/START/ATTACH) in this
37 * transaction, it must be zero before the transaction is
38 * being committed
39 */
40 atomic_t num_extwriters;
41 /*
42 * total writers in this transaction, it must be zero before the
43 * transaction can end
44 */
45 atomic_t num_writers;
46 refcount_t use_count;
47
48 unsigned long flags;
49
50 /* Be protected by fs_info->trans_lock when we want to change it. */
51 enum btrfs_trans_state state;
52 int aborted;
53 struct list_head list;
54 struct extent_io_tree dirty_pages;
55 time64_t start_time;
56 wait_queue_head_t writer_wait;
57 wait_queue_head_t commit_wait;
58 struct list_head pending_snapshots;
59 struct list_head dev_update_list;
60 struct list_head switch_commits;
61 struct list_head dirty_bgs;
62
63 /*
64 * There is no explicit lock which protects io_bgs, rather its
65 * consistency is implied by the fact that all the sites which modify
66 * it do so under some form of transaction critical section, namely:
67 *
68 * - btrfs_start_dirty_block_groups - This function can only ever be
69 * run by one of the transaction committers. Refer to
70 * BTRFS_TRANS_DIRTY_BG_RUN usage in btrfs_commit_transaction
71 *
72 * - btrfs_write_dirty_blockgroups - this is called by
73 * commit_cowonly_roots from transaction critical section
74 * (TRANS_STATE_COMMIT_DOING)
75 *
76 * - btrfs_cleanup_dirty_bgs - called on transaction abort
77 */
78 struct list_head io_bgs;
79 struct list_head dropped_roots;
80 struct extent_io_tree pinned_extents;
81
82 /*
83 * we need to make sure block group deletion doesn't race with
84 * free space cache writeout. This mutex keeps them from stomping
85 * on each other
86 */
87 struct mutex cache_write_mutex;
88 spinlock_t dirty_bgs_lock;
89 /* Protected by spin lock fs_info->unused_bgs_lock. */
90 struct list_head deleted_bgs;
91 spinlock_t dropped_roots_lock;
92 struct btrfs_delayed_ref_root delayed_refs;
93 struct btrfs_fs_info *fs_info;
94
95 /*
96 * Number of ordered extents the transaction must wait for before
97 * committing. These are ordered extents started by a fast fsync.
98 */
99 atomic_t pending_ordered;
100 wait_queue_head_t pending_wait;
101};
102
103enum {
104 ENUM_BIT(__TRANS_FREEZABLE),
105 ENUM_BIT(__TRANS_START),
106 ENUM_BIT(__TRANS_ATTACH),
107 ENUM_BIT(__TRANS_JOIN),
108 ENUM_BIT(__TRANS_JOIN_NOLOCK),
109 ENUM_BIT(__TRANS_DUMMY),
110 ENUM_BIT(__TRANS_JOIN_NOSTART),
111};
112
113#define TRANS_START (__TRANS_START | __TRANS_FREEZABLE)
114#define TRANS_ATTACH (__TRANS_ATTACH)
115#define TRANS_JOIN (__TRANS_JOIN | __TRANS_FREEZABLE)
116#define TRANS_JOIN_NOLOCK (__TRANS_JOIN_NOLOCK)
117#define TRANS_JOIN_NOSTART (__TRANS_JOIN_NOSTART)
118
119#define TRANS_EXTWRITERS (__TRANS_START | __TRANS_ATTACH)
120
121struct btrfs_trans_handle {
122 u64 transid;
123 u64 bytes_reserved;
124 u64 delayed_refs_bytes_reserved;
125 u64 chunk_bytes_reserved;
126 unsigned long delayed_ref_updates;
127 unsigned long delayed_ref_csum_deletions;
128 struct btrfs_transaction *transaction;
129 struct btrfs_block_rsv *block_rsv;
130 struct btrfs_block_rsv *orig_rsv;
131 /* Set by a task that wants to create a snapshot. */
132 struct btrfs_pending_snapshot *pending_snapshot;
133 refcount_t use_count;
134 unsigned int type;
135 /*
136 * Error code of transaction abort, set outside of locks and must use
137 * the READ_ONCE/WRITE_ONCE access
138 */
139 short aborted;
140 bool adding_csums;
141 bool allocating_chunk;
142 bool removing_chunk;
143 bool reloc_reserved;
144 bool in_fsync;
145 struct btrfs_fs_info *fs_info;
146 struct list_head new_bgs;
147 struct btrfs_block_rsv delayed_rsv;
148};
149
150/*
151 * The abort status can be changed between calls and is not protected by locks.
152 * This accepts btrfs_transaction and btrfs_trans_handle as types. Once it's
153 * set to a non-zero value it does not change, so the macro should be in checks
154 * but is not necessary for further reads of the value.
155 */
156#define TRANS_ABORTED(trans) (unlikely(READ_ONCE((trans)->aborted)))
157
158struct btrfs_pending_snapshot {
159 struct dentry *dentry;
160 struct inode *dir;
161 struct btrfs_root *root;
162 struct btrfs_root_item *root_item;
163 struct btrfs_root *snap;
164 struct btrfs_qgroup_inherit *inherit;
165 struct btrfs_path *path;
166 /* block reservation for the operation */
167 struct btrfs_block_rsv block_rsv;
168 /* extra metadata reservation for relocation */
169 int error;
170 /* Preallocated anonymous block device number */
171 dev_t anon_dev;
172 bool readonly;
173 struct list_head list;
174};
175
176static inline void btrfs_set_inode_last_trans(struct btrfs_trans_handle *trans,
177 struct btrfs_inode *inode)
178{
179 spin_lock(&inode->lock);
180 inode->last_trans = trans->transaction->transid;
181 inode->last_sub_trans = btrfs_get_root_log_transid(inode->root);
182 inode->last_log_commit = inode->last_sub_trans - 1;
183 spin_unlock(&inode->lock);
184}
185
186/*
187 * Make qgroup codes to skip given qgroupid, means the old/new_roots for
188 * qgroup won't contain the qgroupid in it.
189 */
190static inline void btrfs_set_skip_qgroup(struct btrfs_trans_handle *trans,
191 u64 qgroupid)
192{
193 struct btrfs_delayed_ref_root *delayed_refs;
194
195 delayed_refs = &trans->transaction->delayed_refs;
196 WARN_ON(delayed_refs->qgroup_to_skip);
197 delayed_refs->qgroup_to_skip = qgroupid;
198}
199
200static inline void btrfs_clear_skip_qgroup(struct btrfs_trans_handle *trans)
201{
202 struct btrfs_delayed_ref_root *delayed_refs;
203
204 delayed_refs = &trans->transaction->delayed_refs;
205 WARN_ON(!delayed_refs->qgroup_to_skip);
206 delayed_refs->qgroup_to_skip = 0;
207}
208
209bool __cold abort_should_print_stack(int error);
210
211/*
212 * Call btrfs_abort_transaction as early as possible when an error condition is
213 * detected, that way the exact stack trace is reported for some errors.
214 */
215#define btrfs_abort_transaction(trans, error) \
216do { \
217 bool first = false; \
218 /* Report first abort since mount */ \
219 if (!test_and_set_bit(BTRFS_FS_STATE_TRANS_ABORTED, \
220 &((trans)->fs_info->fs_state))) { \
221 first = true; \
222 if (WARN(abort_should_print_stack(error), \
223 KERN_ERR \
224 "BTRFS: Transaction aborted (error %d)\n", \
225 (error))) { \
226 /* Stack trace printed. */ \
227 } else { \
228 btrfs_err((trans)->fs_info, \
229 "Transaction aborted (error %d)", \
230 (error)); \
231 } \
232 } \
233 __btrfs_abort_transaction((trans), __func__, \
234 __LINE__, (error), first); \
235} while (0)
236
237int btrfs_end_transaction(struct btrfs_trans_handle *trans);
238struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
239 unsigned int num_items);
240struct btrfs_trans_handle *btrfs_start_transaction_fallback_global_rsv(
241 struct btrfs_root *root,
242 unsigned int num_items);
243struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root);
244struct btrfs_trans_handle *btrfs_join_transaction_spacecache(struct btrfs_root *root);
245struct btrfs_trans_handle *btrfs_join_transaction_nostart(struct btrfs_root *root);
246struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root);
247struct btrfs_trans_handle *btrfs_attach_transaction_barrier(
248 struct btrfs_root *root);
249int btrfs_wait_for_commit(struct btrfs_fs_info *fs_info, u64 transid);
250
251void btrfs_add_dead_root(struct btrfs_root *root);
252void btrfs_maybe_wake_unfinished_drop(struct btrfs_fs_info *fs_info);
253int btrfs_clean_one_deleted_snapshot(struct btrfs_fs_info *fs_info);
254int btrfs_commit_transaction(struct btrfs_trans_handle *trans);
255void btrfs_commit_transaction_async(struct btrfs_trans_handle *trans);
256int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans);
257bool btrfs_should_end_transaction(struct btrfs_trans_handle *trans);
258void btrfs_throttle(struct btrfs_fs_info *fs_info);
259int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
260 struct btrfs_root *root);
261int btrfs_write_marked_extents(struct btrfs_fs_info *fs_info,
262 struct extent_io_tree *dirty_pages, int mark);
263int btrfs_wait_tree_log_extents(struct btrfs_root *root, int mark);
264int btrfs_transaction_blocked(struct btrfs_fs_info *info);
265int btrfs_transaction_in_commit(struct btrfs_fs_info *info);
266void btrfs_put_transaction(struct btrfs_transaction *transaction);
267void btrfs_add_dropped_root(struct btrfs_trans_handle *trans,
268 struct btrfs_root *root);
269void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans);
270void __cold __btrfs_abort_transaction(struct btrfs_trans_handle *trans,
271 const char *function,
272 unsigned int line, int error, bool first_hit);
273
274int __init btrfs_transaction_init(void);
275void __cold btrfs_transaction_exit(void);
276
277#endif
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/atomic.h>
10#include <linux/refcount.h>
11#include <linux/list.h>
12#include <linux/time64.h>
13#include <linux/mutex.h>
14#include <linux/wait.h>
15#include "btrfs_inode.h"
16#include "delayed-ref.h"
17#include "extent-io-tree.h"
18#include "block-rsv.h"
19#include "messages.h"
20#include "misc.h"
21
22struct dentry;
23struct inode;
24struct btrfs_pending_snapshot;
25struct btrfs_fs_info;
26struct btrfs_root_item;
27struct btrfs_root;
28struct btrfs_path;
29
30/*
31 * Signal that a direct IO write is in progress, to avoid deadlock for sync
32 * direct IO writes when fsync is called during the direct IO write path.
33 */
34#define BTRFS_TRANS_DIO_WRITE_STUB ((void *) 1)
35
36/* Radix-tree tag for roots that are part of the transaction. */
37#define BTRFS_ROOT_TRANS_TAG 0
38
39enum btrfs_trans_state {
40 TRANS_STATE_RUNNING,
41 TRANS_STATE_COMMIT_PREP,
42 TRANS_STATE_COMMIT_START,
43 TRANS_STATE_COMMIT_DOING,
44 TRANS_STATE_UNBLOCKED,
45 TRANS_STATE_SUPER_COMMITTED,
46 TRANS_STATE_COMPLETED,
47 TRANS_STATE_MAX,
48};
49
50#define BTRFS_TRANS_HAVE_FREE_BGS 0
51#define BTRFS_TRANS_DIRTY_BG_RUN 1
52#define BTRFS_TRANS_CACHE_ENOSPC 2
53
54struct btrfs_transaction {
55 u64 transid;
56 /*
57 * total external writers(USERSPACE/START/ATTACH) in this
58 * transaction, it must be zero before the transaction is
59 * being committed
60 */
61 atomic_t num_extwriters;
62 /*
63 * total writers in this transaction, it must be zero before the
64 * transaction can end
65 */
66 atomic_t num_writers;
67 refcount_t use_count;
68
69 unsigned long flags;
70
71 /* Be protected by fs_info->trans_lock when we want to change it. */
72 enum btrfs_trans_state state;
73 int aborted;
74 struct list_head list;
75 struct extent_io_tree dirty_pages;
76 time64_t start_time;
77 wait_queue_head_t writer_wait;
78 wait_queue_head_t commit_wait;
79 struct list_head pending_snapshots;
80 struct list_head dev_update_list;
81 struct list_head switch_commits;
82 struct list_head dirty_bgs;
83
84 /*
85 * There is no explicit lock which protects io_bgs, rather its
86 * consistency is implied by the fact that all the sites which modify
87 * it do so under some form of transaction critical section, namely:
88 *
89 * - btrfs_start_dirty_block_groups - This function can only ever be
90 * run by one of the transaction committers. Refer to
91 * BTRFS_TRANS_DIRTY_BG_RUN usage in btrfs_commit_transaction
92 *
93 * - btrfs_write_dirty_blockgroups - this is called by
94 * commit_cowonly_roots from transaction critical section
95 * (TRANS_STATE_COMMIT_DOING)
96 *
97 * - btrfs_cleanup_dirty_bgs - called on transaction abort
98 */
99 struct list_head io_bgs;
100 struct list_head dropped_roots;
101 struct extent_io_tree pinned_extents;
102
103 /*
104 * we need to make sure block group deletion doesn't race with
105 * free space cache writeout. This mutex keeps them from stomping
106 * on each other
107 */
108 struct mutex cache_write_mutex;
109 spinlock_t dirty_bgs_lock;
110 /* Protected by spin lock fs_info->unused_bgs_lock. */
111 struct list_head deleted_bgs;
112 spinlock_t dropped_roots_lock;
113 struct btrfs_delayed_ref_root delayed_refs;
114 struct btrfs_fs_info *fs_info;
115
116 /*
117 * Number of ordered extents the transaction must wait for before
118 * committing. These are ordered extents started by a fast fsync.
119 */
120 atomic_t pending_ordered;
121 wait_queue_head_t pending_wait;
122};
123
124enum {
125 ENUM_BIT(__TRANS_FREEZABLE),
126 ENUM_BIT(__TRANS_START),
127 ENUM_BIT(__TRANS_ATTACH),
128 ENUM_BIT(__TRANS_JOIN),
129 ENUM_BIT(__TRANS_JOIN_NOLOCK),
130 ENUM_BIT(__TRANS_DUMMY),
131 ENUM_BIT(__TRANS_JOIN_NOSTART),
132};
133
134#define TRANS_START (__TRANS_START | __TRANS_FREEZABLE)
135#define TRANS_ATTACH (__TRANS_ATTACH)
136#define TRANS_JOIN (__TRANS_JOIN | __TRANS_FREEZABLE)
137#define TRANS_JOIN_NOLOCK (__TRANS_JOIN_NOLOCK)
138#define TRANS_JOIN_NOSTART (__TRANS_JOIN_NOSTART)
139
140#define TRANS_EXTWRITERS (__TRANS_START | __TRANS_ATTACH)
141
142struct btrfs_trans_handle {
143 u64 transid;
144 u64 bytes_reserved;
145 u64 delayed_refs_bytes_reserved;
146 u64 chunk_bytes_reserved;
147 unsigned long delayed_ref_updates;
148 unsigned long delayed_ref_csum_deletions;
149 struct btrfs_transaction *transaction;
150 struct btrfs_block_rsv *block_rsv;
151 struct btrfs_block_rsv *orig_rsv;
152 /* Set by a task that wants to create a snapshot. */
153 struct btrfs_pending_snapshot *pending_snapshot;
154 refcount_t use_count;
155 unsigned int type;
156 /*
157 * Error code of transaction abort, set outside of locks and must use
158 * the READ_ONCE/WRITE_ONCE access
159 */
160 short aborted;
161 bool adding_csums;
162 bool allocating_chunk;
163 bool removing_chunk;
164 bool reloc_reserved;
165 bool in_fsync;
166 struct btrfs_fs_info *fs_info;
167 struct list_head new_bgs;
168 struct btrfs_block_rsv delayed_rsv;
169};
170
171/*
172 * The abort status can be changed between calls and is not protected by locks.
173 * This accepts btrfs_transaction and btrfs_trans_handle as types. Once it's
174 * set to a non-zero value it does not change, so the macro should be in checks
175 * but is not necessary for further reads of the value.
176 */
177#define TRANS_ABORTED(trans) (unlikely(READ_ONCE((trans)->aborted)))
178
179struct btrfs_pending_snapshot {
180 struct dentry *dentry;
181 struct btrfs_inode *dir;
182 struct btrfs_root *root;
183 struct btrfs_root_item *root_item;
184 struct btrfs_root *snap;
185 struct btrfs_qgroup_inherit *inherit;
186 struct btrfs_path *path;
187 /* block reservation for the operation */
188 struct btrfs_block_rsv block_rsv;
189 /* extra metadata reservation for relocation */
190 int error;
191 /* Preallocated anonymous block device number */
192 dev_t anon_dev;
193 bool readonly;
194 struct list_head list;
195};
196
197static inline void btrfs_set_inode_last_trans(struct btrfs_trans_handle *trans,
198 struct btrfs_inode *inode)
199{
200 spin_lock(&inode->lock);
201 inode->last_trans = trans->transaction->transid;
202 inode->last_sub_trans = btrfs_get_root_log_transid(inode->root);
203 inode->last_log_commit = inode->last_sub_trans - 1;
204 spin_unlock(&inode->lock);
205}
206
207/*
208 * Make qgroup codes to skip given qgroupid, means the old/new_roots for
209 * qgroup won't contain the qgroupid in it.
210 */
211static inline void btrfs_set_skip_qgroup(struct btrfs_trans_handle *trans,
212 u64 qgroupid)
213{
214 struct btrfs_delayed_ref_root *delayed_refs;
215
216 delayed_refs = &trans->transaction->delayed_refs;
217 WARN_ON(delayed_refs->qgroup_to_skip);
218 delayed_refs->qgroup_to_skip = qgroupid;
219}
220
221static inline void btrfs_clear_skip_qgroup(struct btrfs_trans_handle *trans)
222{
223 struct btrfs_delayed_ref_root *delayed_refs;
224
225 delayed_refs = &trans->transaction->delayed_refs;
226 WARN_ON(!delayed_refs->qgroup_to_skip);
227 delayed_refs->qgroup_to_skip = 0;
228}
229
230bool __cold abort_should_print_stack(int error);
231
232/*
233 * Call btrfs_abort_transaction as early as possible when an error condition is
234 * detected, that way the exact stack trace is reported for some errors.
235 */
236#define btrfs_abort_transaction(trans, error) \
237do { \
238 bool __first = false; \
239 /* Report first abort since mount */ \
240 if (!test_and_set_bit(BTRFS_FS_STATE_TRANS_ABORTED, \
241 &((trans)->fs_info->fs_state))) { \
242 __first = true; \
243 if (WARN(abort_should_print_stack(error), \
244 KERN_ERR \
245 "BTRFS: Transaction aborted (error %d)\n", \
246 (error))) { \
247 /* Stack trace printed. */ \
248 } else { \
249 btrfs_err((trans)->fs_info, \
250 "Transaction aborted (error %d)", \
251 (error)); \
252 } \
253 } \
254 __btrfs_abort_transaction((trans), __func__, \
255 __LINE__, (error), __first); \
256} while (0)
257
258int btrfs_end_transaction(struct btrfs_trans_handle *trans);
259struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
260 unsigned int num_items);
261struct btrfs_trans_handle *btrfs_start_transaction_fallback_global_rsv(
262 struct btrfs_root *root,
263 unsigned int num_items);
264struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root);
265struct btrfs_trans_handle *btrfs_join_transaction_spacecache(struct btrfs_root *root);
266struct btrfs_trans_handle *btrfs_join_transaction_nostart(struct btrfs_root *root);
267struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root);
268struct btrfs_trans_handle *btrfs_attach_transaction_barrier(
269 struct btrfs_root *root);
270int btrfs_wait_for_commit(struct btrfs_fs_info *fs_info, u64 transid);
271
272void btrfs_add_dead_root(struct btrfs_root *root);
273void btrfs_maybe_wake_unfinished_drop(struct btrfs_fs_info *fs_info);
274int btrfs_clean_one_deleted_snapshot(struct btrfs_fs_info *fs_info);
275int btrfs_commit_transaction(struct btrfs_trans_handle *trans);
276void btrfs_commit_transaction_async(struct btrfs_trans_handle *trans);
277int btrfs_commit_current_transaction(struct btrfs_root *root);
278int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans);
279bool btrfs_should_end_transaction(struct btrfs_trans_handle *trans);
280void btrfs_throttle(struct btrfs_fs_info *fs_info);
281int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
282 struct btrfs_root *root);
283int btrfs_write_marked_extents(struct btrfs_fs_info *fs_info,
284 struct extent_io_tree *dirty_pages, int mark);
285int btrfs_wait_tree_log_extents(struct btrfs_root *root, int mark);
286int btrfs_transaction_blocked(struct btrfs_fs_info *info);
287void btrfs_put_transaction(struct btrfs_transaction *transaction);
288void btrfs_add_dropped_root(struct btrfs_trans_handle *trans,
289 struct btrfs_root *root);
290void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans);
291void __cold __btrfs_abort_transaction(struct btrfs_trans_handle *trans,
292 const char *function,
293 unsigned int line, int error, bool first_hit);
294
295int __init btrfs_transaction_init(void);
296void __cold btrfs_transaction_exit(void);
297
298#endif