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 
14 enum btrfs_trans_state {
15 	TRANS_STATE_RUNNING,
16 	TRANS_STATE_COMMIT_START,
17 	TRANS_STATE_COMMIT_DOING,
18 	TRANS_STATE_UNBLOCKED,
19 	TRANS_STATE_SUPER_COMMITTED,
20 	TRANS_STATE_COMPLETED,
21 	TRANS_STATE_MAX,
22 };
23 
24 #define BTRFS_TRANS_HAVE_FREE_BGS	0
25 #define BTRFS_TRANS_DIRTY_BG_RUN	1
26 #define BTRFS_TRANS_CACHE_ENOSPC	2
27 
28 struct btrfs_transaction {
29 	u64 transid;
30 	/*
31 	 * total external writers(USERSPACE/START/ATTACH) in this
32 	 * transaction, it must be zero before the transaction is
33 	 * being committed
34 	 */
35 	atomic_t num_extwriters;
36 	/*
37 	 * total writers in this transaction, it must be zero before the
38 	 * transaction can end
39 	 */
40 	atomic_t num_writers;
41 	refcount_t use_count;
42 
43 	unsigned long flags;
44 
45 	/* Be protected by fs_info->trans_lock when we want to change it. */
46 	enum btrfs_trans_state state;
47 	int aborted;
48 	struct list_head list;
49 	struct extent_io_tree dirty_pages;
50 	time64_t start_time;
51 	wait_queue_head_t writer_wait;
52 	wait_queue_head_t commit_wait;
53 	struct list_head pending_snapshots;
54 	struct list_head dev_update_list;
55 	struct list_head switch_commits;
56 	struct list_head dirty_bgs;
57 
58 	/*
59 	 * There is no explicit lock which protects io_bgs, rather its
60 	 * consistency is implied by the fact that all the sites which modify
61 	 * it do so under some form of transaction critical section, namely:
62 	 *
63 	 * - btrfs_start_dirty_block_groups - This function can only ever be
64 	 *   run by one of the transaction committers. Refer to
65 	 *   BTRFS_TRANS_DIRTY_BG_RUN usage in btrfs_commit_transaction
66 	 *
67 	 * - btrfs_write_dirty_blockgroups - this is called by
68 	 *   commit_cowonly_roots from transaction critical section
69 	 *   (TRANS_STATE_COMMIT_DOING)
70 	 *
71 	 * - btrfs_cleanup_dirty_bgs - called on transaction abort
72 	 */
73 	struct list_head io_bgs;
74 	struct list_head dropped_roots;
75 	struct extent_io_tree pinned_extents;
76 
77 	/*
78 	 * we need to make sure block group deletion doesn't race with
79 	 * free space cache writeout.  This mutex keeps them from stomping
80 	 * on each other
81 	 */
82 	struct mutex cache_write_mutex;
83 	spinlock_t dirty_bgs_lock;
84 	/* Protected by spin lock fs_info->unused_bgs_lock. */
85 	struct list_head deleted_bgs;
86 	spinlock_t dropped_roots_lock;
87 	struct btrfs_delayed_ref_root delayed_refs;
88 	struct btrfs_fs_info *fs_info;
89 
90 	/*
91 	 * Number of ordered extents the transaction must wait for before
92 	 * committing. These are ordered extents started by a fast fsync.
93 	 */
94 	atomic_t pending_ordered;
95 	wait_queue_head_t pending_wait;
96 
97 	spinlock_t releasing_ebs_lock;
98 	struct list_head releasing_ebs;
99 
100 	/*
101 	 * The number of bytes currently reserved, by all transaction handles
102 	 * attached to this transaction, for metadata extents of the chunk tree.
103 	 */
104 	atomic64_t chunk_bytes_reserved;
105 	wait_queue_head_t chunk_reserve_wait;
106 };
107 
108 #define __TRANS_FREEZABLE	(1U << 0)
109 
110 #define __TRANS_START		(1U << 9)
111 #define __TRANS_ATTACH		(1U << 10)
112 #define __TRANS_JOIN		(1U << 11)
113 #define __TRANS_JOIN_NOLOCK	(1U << 12)
114 #define __TRANS_DUMMY		(1U << 13)
115 #define __TRANS_JOIN_NOSTART	(1U << 14)
116 
117 #define TRANS_START		(__TRANS_START | __TRANS_FREEZABLE)
118 #define TRANS_ATTACH		(__TRANS_ATTACH)
119 #define TRANS_JOIN		(__TRANS_JOIN | __TRANS_FREEZABLE)
120 #define TRANS_JOIN_NOLOCK	(__TRANS_JOIN_NOLOCK)
121 #define TRANS_JOIN_NOSTART	(__TRANS_JOIN_NOSTART)
122 
123 #define TRANS_EXTWRITERS	(__TRANS_START | __TRANS_ATTACH)
124 
125 #define BTRFS_SEND_TRANS_STUB	((void *)1)
126 
127 struct btrfs_trans_handle {
128 	u64 transid;
129 	u64 bytes_reserved;
130 	u64 chunk_bytes_reserved;
131 	unsigned long delayed_ref_updates;
132 	struct btrfs_transaction *transaction;
133 	struct btrfs_block_rsv *block_rsv;
134 	struct btrfs_block_rsv *orig_rsv;
135 	refcount_t use_count;
136 	unsigned int type;
137 	/*
138 	 * Error code of transaction abort, set outside of locks and must use
139 	 * the READ_ONCE/WRITE_ONCE access
140 	 */
141 	short aborted;
142 	bool adding_csums;
143 	bool allocating_chunk;
144 	bool can_flush_pending_bgs;
145 	bool reloc_reserved;
146 	bool dirty;
147 	bool in_fsync;
148 	struct btrfs_root *root;
149 	struct btrfs_fs_info *fs_info;
150 	struct list_head new_bgs;
151 };
152 
153 /*
154  * The abort status can be changed between calls and is not protected by locks.
155  * This accepts btrfs_transaction and btrfs_trans_handle as types. Once it's
156  * set to a non-zero value it does not change, so the macro should be in checks
157  * but is not necessary for further reads of the value.
158  */
159 #define TRANS_ABORTED(trans)		(unlikely(READ_ONCE((trans)->aborted)))
160 
161 struct btrfs_pending_snapshot {
162 	struct dentry *dentry;
163 	struct inode *dir;
164 	struct btrfs_root *root;
165 	struct btrfs_root_item *root_item;
166 	struct btrfs_root *snap;
167 	struct btrfs_qgroup_inherit *inherit;
168 	struct btrfs_path *path;
169 	/* block reservation for the operation */
170 	struct btrfs_block_rsv block_rsv;
171 	/* extra metadata reservation for relocation */
172 	int error;
173 	/* Preallocated anonymous block device number */
174 	dev_t anon_dev;
175 	bool readonly;
176 	struct list_head list;
177 };
178 
btrfs_set_inode_last_trans(struct btrfs_trans_handle * trans,struct btrfs_inode * inode)179 static inline void btrfs_set_inode_last_trans(struct btrfs_trans_handle *trans,
180 					      struct btrfs_inode *inode)
181 {
182 	spin_lock(&inode->lock);
183 	inode->last_trans = trans->transaction->transid;
184 	inode->last_sub_trans = inode->root->log_transid;
185 	inode->last_log_commit = inode->last_sub_trans - 1;
186 	spin_unlock(&inode->lock);
187 }
188 
189 /*
190  * Make qgroup codes to skip given qgroupid, means the old/new_roots for
191  * qgroup won't contain the qgroupid in it.
192  */
btrfs_set_skip_qgroup(struct btrfs_trans_handle * trans,u64 qgroupid)193 static inline void btrfs_set_skip_qgroup(struct btrfs_trans_handle *trans,
194 					 u64 qgroupid)
195 {
196 	struct btrfs_delayed_ref_root *delayed_refs;
197 
198 	delayed_refs = &trans->transaction->delayed_refs;
199 	WARN_ON(delayed_refs->qgroup_to_skip);
200 	delayed_refs->qgroup_to_skip = qgroupid;
201 }
202 
btrfs_clear_skip_qgroup(struct btrfs_trans_handle * trans)203 static inline void btrfs_clear_skip_qgroup(struct btrfs_trans_handle *trans)
204 {
205 	struct btrfs_delayed_ref_root *delayed_refs;
206 
207 	delayed_refs = &trans->transaction->delayed_refs;
208 	WARN_ON(!delayed_refs->qgroup_to_skip);
209 	delayed_refs->qgroup_to_skip = 0;
210 }
211 
212 int btrfs_end_transaction(struct btrfs_trans_handle *trans);
213 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
214 						   unsigned int num_items);
215 struct btrfs_trans_handle *btrfs_start_transaction_fallback_global_rsv(
216 					struct btrfs_root *root,
217 					unsigned int num_items);
218 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root);
219 struct btrfs_trans_handle *btrfs_join_transaction_spacecache(struct btrfs_root *root);
220 struct btrfs_trans_handle *btrfs_join_transaction_nostart(struct btrfs_root *root);
221 struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root);
222 struct btrfs_trans_handle *btrfs_attach_transaction_barrier(
223 					struct btrfs_root *root);
224 int btrfs_wait_for_commit(struct btrfs_fs_info *fs_info, u64 transid);
225 
226 void btrfs_add_dead_root(struct btrfs_root *root);
227 int btrfs_defrag_root(struct btrfs_root *root);
228 int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root);
229 int btrfs_commit_transaction(struct btrfs_trans_handle *trans);
230 int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans,
231 				   int wait_for_unblock);
232 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans);
233 bool btrfs_should_end_transaction(struct btrfs_trans_handle *trans);
234 void btrfs_throttle(struct btrfs_fs_info *fs_info);
235 int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
236 				struct btrfs_root *root);
237 int btrfs_write_marked_extents(struct btrfs_fs_info *fs_info,
238 				struct extent_io_tree *dirty_pages, int mark);
239 int btrfs_wait_tree_log_extents(struct btrfs_root *root, int mark);
240 int btrfs_transaction_blocked(struct btrfs_fs_info *info);
241 int btrfs_transaction_in_commit(struct btrfs_fs_info *info);
242 void btrfs_put_transaction(struct btrfs_transaction *transaction);
243 void btrfs_apply_pending_changes(struct btrfs_fs_info *fs_info);
244 void btrfs_add_dropped_root(struct btrfs_trans_handle *trans,
245 			    struct btrfs_root *root);
246 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans);
247 
248 #endif
249