xref: /openbmc/linux/fs/btrfs/transaction.h (revision 7e2e638c)
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 /*
16  * Signal that a direct IO write is in progress, to avoid deadlock for sync
17  * direct IO writes when fsync is called during the direct IO write path.
18  */
19 #define BTRFS_TRANS_DIO_WRITE_STUB	((void *) 1)
20 
21 /* Radix-tree tag for roots that are part of the trasaction. */
22 #define BTRFS_ROOT_TRANS_TAG			0
23 
24 enum btrfs_trans_state {
25 	TRANS_STATE_RUNNING,
26 	TRANS_STATE_COMMIT_PREP,
27 	TRANS_STATE_COMMIT_START,
28 	TRANS_STATE_COMMIT_DOING,
29 	TRANS_STATE_UNBLOCKED,
30 	TRANS_STATE_SUPER_COMMITTED,
31 	TRANS_STATE_COMPLETED,
32 	TRANS_STATE_MAX,
33 };
34 
35 #define BTRFS_TRANS_HAVE_FREE_BGS	0
36 #define BTRFS_TRANS_DIRTY_BG_RUN	1
37 #define BTRFS_TRANS_CACHE_ENOSPC	2
38 
39 struct btrfs_transaction {
40 	u64 transid;
41 	/*
42 	 * total external writers(USERSPACE/START/ATTACH) in this
43 	 * transaction, it must be zero before the transaction is
44 	 * being committed
45 	 */
46 	atomic_t num_extwriters;
47 	/*
48 	 * total writers in this transaction, it must be zero before the
49 	 * transaction can end
50 	 */
51 	atomic_t num_writers;
52 	refcount_t use_count;
53 
54 	unsigned long flags;
55 
56 	/* Be protected by fs_info->trans_lock when we want to change it. */
57 	enum btrfs_trans_state state;
58 	int aborted;
59 	struct list_head list;
60 	struct extent_io_tree dirty_pages;
61 	time64_t start_time;
62 	wait_queue_head_t writer_wait;
63 	wait_queue_head_t commit_wait;
64 	struct list_head pending_snapshots;
65 	struct list_head dev_update_list;
66 	struct list_head switch_commits;
67 	struct list_head dirty_bgs;
68 
69 	/*
70 	 * There is no explicit lock which protects io_bgs, rather its
71 	 * consistency is implied by the fact that all the sites which modify
72 	 * it do so under some form of transaction critical section, namely:
73 	 *
74 	 * - btrfs_start_dirty_block_groups - This function can only ever be
75 	 *   run by one of the transaction committers. Refer to
76 	 *   BTRFS_TRANS_DIRTY_BG_RUN usage in btrfs_commit_transaction
77 	 *
78 	 * - btrfs_write_dirty_blockgroups - this is called by
79 	 *   commit_cowonly_roots from transaction critical section
80 	 *   (TRANS_STATE_COMMIT_DOING)
81 	 *
82 	 * - btrfs_cleanup_dirty_bgs - called on transaction abort
83 	 */
84 	struct list_head io_bgs;
85 	struct list_head dropped_roots;
86 	struct extent_io_tree pinned_extents;
87 
88 	/*
89 	 * we need to make sure block group deletion doesn't race with
90 	 * free space cache writeout.  This mutex keeps them from stomping
91 	 * on each other
92 	 */
93 	struct mutex cache_write_mutex;
94 	spinlock_t dirty_bgs_lock;
95 	/* Protected by spin lock fs_info->unused_bgs_lock. */
96 	struct list_head deleted_bgs;
97 	spinlock_t dropped_roots_lock;
98 	struct btrfs_delayed_ref_root delayed_refs;
99 	struct btrfs_fs_info *fs_info;
100 
101 	/*
102 	 * Number of ordered extents the transaction must wait for before
103 	 * committing. These are ordered extents started by a fast fsync.
104 	 */
105 	atomic_t pending_ordered;
106 	wait_queue_head_t pending_wait;
107 };
108 
109 enum {
110 	ENUM_BIT(__TRANS_FREEZABLE),
111 	ENUM_BIT(__TRANS_START),
112 	ENUM_BIT(__TRANS_ATTACH),
113 	ENUM_BIT(__TRANS_JOIN),
114 	ENUM_BIT(__TRANS_JOIN_NOLOCK),
115 	ENUM_BIT(__TRANS_DUMMY),
116 	ENUM_BIT(__TRANS_JOIN_NOSTART),
117 };
118 
119 #define TRANS_START		(__TRANS_START | __TRANS_FREEZABLE)
120 #define TRANS_ATTACH		(__TRANS_ATTACH)
121 #define TRANS_JOIN		(__TRANS_JOIN | __TRANS_FREEZABLE)
122 #define TRANS_JOIN_NOLOCK	(__TRANS_JOIN_NOLOCK)
123 #define TRANS_JOIN_NOSTART	(__TRANS_JOIN_NOSTART)
124 
125 #define TRANS_EXTWRITERS	(__TRANS_START | __TRANS_ATTACH)
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 	/* Set by a task that wants to create a snapshot. */
136 	struct btrfs_pending_snapshot *pending_snapshot;
137 	refcount_t use_count;
138 	unsigned int type;
139 	/*
140 	 * Error code of transaction abort, set outside of locks and must use
141 	 * the READ_ONCE/WRITE_ONCE access
142 	 */
143 	short aborted;
144 	bool adding_csums;
145 	bool allocating_chunk;
146 	bool removing_chunk;
147 	bool reloc_reserved;
148 	bool in_fsync;
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 
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  */
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 
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 bool __cold abort_should_print_stack(int errno);
213 
214 /*
215  * Call btrfs_abort_transaction as early as possible when an error condition is
216  * detected, that way the exact stack trace is reported for some errors.
217  */
218 #define btrfs_abort_transaction(trans, errno)		\
219 do {								\
220 	bool first = false;					\
221 	/* Report first abort since mount */			\
222 	if (!test_and_set_bit(BTRFS_FS_STATE_TRANS_ABORTED,	\
223 			&((trans)->fs_info->fs_state))) {	\
224 		first = true;					\
225 		if (WARN(abort_should_print_stack(errno),	\
226 			KERN_ERR				\
227 			"BTRFS: Transaction aborted (error %d)\n",	\
228 			(errno))) {					\
229 			/* Stack trace printed. */			\
230 		} else {						\
231 			btrfs_err((trans)->fs_info,			\
232 				  "Transaction aborted (error %d)",	\
233 				  (errno));			\
234 		}						\
235 	}							\
236 	__btrfs_abort_transaction((trans), __func__,		\
237 				  __LINE__, (errno), first);	\
238 } while (0)
239 
240 int btrfs_end_transaction(struct btrfs_trans_handle *trans);
241 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
242 						   unsigned int num_items);
243 struct btrfs_trans_handle *btrfs_start_transaction_fallback_global_rsv(
244 					struct btrfs_root *root,
245 					unsigned int num_items);
246 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root);
247 struct btrfs_trans_handle *btrfs_join_transaction_spacecache(struct btrfs_root *root);
248 struct btrfs_trans_handle *btrfs_join_transaction_nostart(struct btrfs_root *root);
249 struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root);
250 struct btrfs_trans_handle *btrfs_attach_transaction_barrier(
251 					struct btrfs_root *root);
252 int btrfs_wait_for_commit(struct btrfs_fs_info *fs_info, u64 transid);
253 
254 void btrfs_add_dead_root(struct btrfs_root *root);
255 int btrfs_defrag_root(struct btrfs_root *root);
256 void btrfs_maybe_wake_unfinished_drop(struct btrfs_fs_info *fs_info);
257 int btrfs_clean_one_deleted_snapshot(struct btrfs_fs_info *fs_info);
258 int btrfs_commit_transaction(struct btrfs_trans_handle *trans);
259 void btrfs_commit_transaction_async(struct btrfs_trans_handle *trans);
260 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans);
261 bool btrfs_should_end_transaction(struct btrfs_trans_handle *trans);
262 void btrfs_throttle(struct btrfs_fs_info *fs_info);
263 int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
264 				struct btrfs_root *root);
265 int btrfs_write_marked_extents(struct btrfs_fs_info *fs_info,
266 				struct extent_io_tree *dirty_pages, int mark);
267 int btrfs_wait_tree_log_extents(struct btrfs_root *root, int mark);
268 int btrfs_transaction_blocked(struct btrfs_fs_info *info);
269 int btrfs_transaction_in_commit(struct btrfs_fs_info *info);
270 void btrfs_put_transaction(struct btrfs_transaction *transaction);
271 void btrfs_add_dropped_root(struct btrfs_trans_handle *trans,
272 			    struct btrfs_root *root);
273 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans);
274 void __cold __btrfs_abort_transaction(struct btrfs_trans_handle *trans,
275 				      const char *function,
276 				      unsigned int line, int errno, bool first_hit);
277 
278 int __init btrfs_transaction_init(void);
279 void __cold btrfs_transaction_exit(void);
280 
281 #endif
282