transaction.h

  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
 18struct dentry;
 19struct inode;
 20struct btrfs_pending_snapshot;
 21struct btrfs_fs_info;
 22struct btrfs_root_item;
 23struct btrfs_root;
 24struct btrfs_path;
 25
 26/*
 27 * Signal that a direct IO write is in progress, to avoid deadlock for sync
 28 * direct IO writes when fsync is called during the direct IO write path.
 29 */
 30#define BTRFS_TRANS_DIO_WRITE_STUB	((void *) 1)
 31
 32/* Radix-tree tag for roots that are part of the transaction. */
 33#define BTRFS_ROOT_TRANS_TAG			0
 34
 35enum btrfs_trans_state {
 36	TRANS_STATE_RUNNING,
 37	TRANS_STATE_COMMIT_PREP,
 38	TRANS_STATE_COMMIT_START,
 39	TRANS_STATE_COMMIT_DOING,
 40	TRANS_STATE_UNBLOCKED,
 41	TRANS_STATE_SUPER_COMMITTED,
 42	TRANS_STATE_COMPLETED,
 43	TRANS_STATE_MAX,
 44};
 45
 46#define BTRFS_TRANS_HAVE_FREE_BGS	0
 47#define BTRFS_TRANS_DIRTY_BG_RUN	1
 48#define BTRFS_TRANS_CACHE_ENOSPC	2
 49
 50struct btrfs_transaction {
 51	u64 transid;
 52	/*
 53	 * total external writers(USERSPACE/START/ATTACH) in this
 54	 * transaction, it must be zero before the transaction is
 55	 * being committed
 56	 */
 57	atomic_t num_extwriters;
 58	/*
 59	 * total writers in this transaction, it must be zero before the
 60	 * transaction can end
 61	 */
 62	atomic_t num_writers;
 63	refcount_t use_count;
 64
 65	unsigned long flags;
 66
 67	/* Be protected by fs_info->trans_lock when we want to change it. */
 68	enum btrfs_trans_state state;
 69	int aborted;
 70	struct list_head list;
 71	struct extent_io_tree dirty_pages;
 72	time64_t start_time;
 73	wait_queue_head_t writer_wait;
 74	wait_queue_head_t commit_wait;
 75	struct list_head pending_snapshots;
 76	struct list_head dev_update_list;
 77	struct list_head switch_commits;
 78	struct list_head dirty_bgs;
 79
 80	/*
 81	 * There is no explicit lock which protects io_bgs, rather its
 82	 * consistency is implied by the fact that all the sites which modify
 83	 * it do so under some form of transaction critical section, namely:
 84	 *
 85	 * - btrfs_start_dirty_block_groups - This function can only ever be
 86	 *   run by one of the transaction committers. Refer to
 87	 *   BTRFS_TRANS_DIRTY_BG_RUN usage in btrfs_commit_transaction
 88	 *
 89	 * - btrfs_write_dirty_blockgroups - this is called by
 90	 *   commit_cowonly_roots from transaction critical section
 91	 *   (TRANS_STATE_COMMIT_DOING)
 92	 *
 93	 * - btrfs_cleanup_dirty_bgs - called on transaction abort
 94	 */
 95	struct list_head io_bgs;
 96	struct list_head dropped_roots;
 97	struct extent_io_tree pinned_extents;
 98
 99	/*
100	 * we need to make sure block group deletion doesn't race with
101	 * free space cache writeout.  This mutex keeps them from stomping
102	 * on each other
103	 */
104	struct mutex cache_write_mutex;
105	spinlock_t dirty_bgs_lock;
106	/* Protected by spin lock fs_info->unused_bgs_lock. */
107	struct list_head deleted_bgs;
108	spinlock_t dropped_roots_lock;
109	struct btrfs_delayed_ref_root delayed_refs;
110	struct btrfs_fs_info *fs_info;
111
112	/*
113	 * Number of ordered extents the transaction must wait for before
114	 * committing. These are ordered extents started by a fast fsync.
115	 */
116	atomic_t pending_ordered;
117	wait_queue_head_t pending_wait;
118};
119
120enum {
121	ENUM_BIT(__TRANS_FREEZABLE),
122	ENUM_BIT(__TRANS_START),
123	ENUM_BIT(__TRANS_ATTACH),
124	ENUM_BIT(__TRANS_JOIN),
125	ENUM_BIT(__TRANS_JOIN_NOLOCK),
126	ENUM_BIT(__TRANS_DUMMY),
127	ENUM_BIT(__TRANS_JOIN_NOSTART),
128};
129
130#define TRANS_START		(__TRANS_START | __TRANS_FREEZABLE)
131#define TRANS_ATTACH		(__TRANS_ATTACH)
132#define TRANS_JOIN		(__TRANS_JOIN | __TRANS_FREEZABLE)
133#define TRANS_JOIN_NOLOCK	(__TRANS_JOIN_NOLOCK)
134#define TRANS_JOIN_NOSTART	(__TRANS_JOIN_NOSTART)
135
136#define TRANS_EXTWRITERS	(__TRANS_START | __TRANS_ATTACH)
137
138struct btrfs_trans_handle {
139	u64 transid;
140	u64 bytes_reserved;
141	u64 delayed_refs_bytes_reserved;
142	u64 chunk_bytes_reserved;
143	unsigned long delayed_ref_updates;
144	unsigned long delayed_ref_csum_deletions;
145	struct btrfs_transaction *transaction;
146	struct btrfs_block_rsv *block_rsv;
147	struct btrfs_block_rsv *orig_rsv;
148	/* Set by a task that wants to create a snapshot. */
149	struct btrfs_pending_snapshot *pending_snapshot;
150	refcount_t use_count;
151	unsigned int type;
152	/*
153	 * Error code of transaction abort, set outside of locks and must use
154	 * the READ_ONCE/WRITE_ONCE access
155	 */
156	short aborted;
157	bool adding_csums;
158	bool allocating_chunk;
159	bool removing_chunk;
160	bool reloc_reserved;
161	bool in_fsync;
162	struct btrfs_fs_info *fs_info;
163	struct list_head new_bgs;
164	struct btrfs_block_rsv delayed_rsv;
165};
166
167/*
168 * The abort status can be changed between calls and is not protected by locks.
169 * This accepts btrfs_transaction and btrfs_trans_handle as types. Once it's
170 * set to a non-zero value it does not change, so the macro should be in checks
171 * but is not necessary for further reads of the value.
172 */
173#define TRANS_ABORTED(trans)		(unlikely(READ_ONCE((trans)->aborted)))
174
175struct btrfs_pending_snapshot {
176	struct dentry *dentry;
177	struct btrfs_inode *dir;
178	struct btrfs_root *root;
179	struct btrfs_root_item *root_item;
180	struct btrfs_root *snap;
181	struct btrfs_qgroup_inherit *inherit;
182	struct btrfs_path *path;
183	/* block reservation for the operation */
184	struct btrfs_block_rsv block_rsv;
185	/* extra metadata reservation for relocation */
186	int error;
187	/* Preallocated anonymous block device number */
188	dev_t anon_dev;
189	bool readonly;
190	struct list_head list;
191};
192
193static inline void btrfs_set_inode_last_trans(struct btrfs_trans_handle *trans,
194					      struct btrfs_inode *inode)
195{
196	spin_lock(&inode->lock);
197	inode->last_trans = trans->transaction->transid;
198	inode->last_sub_trans = btrfs_get_root_log_transid(inode->root);
199	inode->last_log_commit = inode->last_sub_trans - 1;
200	spin_unlock(&inode->lock);
201}
202
203/*
204 * Make qgroup codes to skip given qgroupid, means the old/new_roots for
205 * qgroup won't contain the qgroupid in it.
206 */
207static inline void btrfs_set_skip_qgroup(struct btrfs_trans_handle *trans,
208					 u64 qgroupid)
209{
210	struct btrfs_delayed_ref_root *delayed_refs;
211
212	delayed_refs = &trans->transaction->delayed_refs;
213	WARN_ON(delayed_refs->qgroup_to_skip);
214	delayed_refs->qgroup_to_skip = qgroupid;
215}
216
217static inline void btrfs_clear_skip_qgroup(struct btrfs_trans_handle *trans)
218{
219	struct btrfs_delayed_ref_root *delayed_refs;
220
221	delayed_refs = &trans->transaction->delayed_refs;
222	WARN_ON(!delayed_refs->qgroup_to_skip);
223	delayed_refs->qgroup_to_skip = 0;
224}
225
226/*
227 * We want the transaction abort to print stack trace only for errors where the
228 * cause could be a bug, eg. due to ENOSPC, and not for common errors that are
229 * caused by external factors.
230 */
231static inline bool btrfs_abort_should_print_stack(int error)
232{
233	switch (error) {
234	case -EIO:
235	case -EROFS:
236	case -ENOMEM:
237		return false;
238	}
239	return true;
240}
241
242/*
243 * Call btrfs_abort_transaction as early as possible when an error condition is
244 * detected, that way the exact stack trace is reported for some errors.
245 */
246#define btrfs_abort_transaction(trans, error)		\
247do {								\
248	bool __first = false;					\
249	/* Report first abort since mount */			\
250	if (!test_and_set_bit(BTRFS_FS_STATE_TRANS_ABORTED,	\
251			&((trans)->fs_info->fs_state))) {	\
252		__first = true;					\
253		if (WARN(btrfs_abort_should_print_stack(error),	\
254			KERN_ERR				\
255			"BTRFS: Transaction aborted (error %d)\n",	\
256			(error))) {					\
257			/* Stack trace printed. */			\
258		} else {						\
259			btrfs_err((trans)->fs_info,			\
260				  "Transaction aborted (error %d)",	\
261				  (error));			\
262		}						\
263	}							\
264	__btrfs_abort_transaction((trans), __func__,		\
265				  __LINE__, (error), __first);	\
266} while (0)
267
268int btrfs_end_transaction(struct btrfs_trans_handle *trans);
269struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
270						   unsigned int num_items);
271struct btrfs_trans_handle *btrfs_start_transaction_fallback_global_rsv(
272					struct btrfs_root *root,
273					unsigned int num_items);
274struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root);
275struct btrfs_trans_handle *btrfs_join_transaction_spacecache(struct btrfs_root *root);
276struct btrfs_trans_handle *btrfs_join_transaction_nostart(struct btrfs_root *root);
277struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root);
278struct btrfs_trans_handle *btrfs_attach_transaction_barrier(
279					struct btrfs_root *root);
280int btrfs_wait_for_commit(struct btrfs_fs_info *fs_info, u64 transid);
281
282void btrfs_add_dead_root(struct btrfs_root *root);
283void btrfs_maybe_wake_unfinished_drop(struct btrfs_fs_info *fs_info);
284int btrfs_clean_one_deleted_snapshot(struct btrfs_fs_info *fs_info);
285int btrfs_commit_transaction(struct btrfs_trans_handle *trans);
286void btrfs_commit_transaction_async(struct btrfs_trans_handle *trans);
287int btrfs_commit_current_transaction(struct btrfs_root *root);
288int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans);
289bool btrfs_should_end_transaction(struct btrfs_trans_handle *trans);
290void btrfs_throttle(struct btrfs_fs_info *fs_info);
291int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
292				struct btrfs_root *root);
293int btrfs_write_marked_extents(struct btrfs_fs_info *fs_info,
294				struct extent_io_tree *dirty_pages, int mark);
295int btrfs_wait_tree_log_extents(struct btrfs_root *root, int mark);
296int btrfs_transaction_blocked(struct btrfs_fs_info *info);
297void btrfs_put_transaction(struct btrfs_transaction *transaction);
298void btrfs_add_dropped_root(struct btrfs_trans_handle *trans,
299			    struct btrfs_root *root);
300void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans);
301void __cold __btrfs_abort_transaction(struct btrfs_trans_handle *trans,
302				      const char *function,
303				      unsigned int line, int error, bool first_hit);
304
305int __init btrfs_transaction_init(void);
306void __cold btrfs_transaction_exit(void);
307
308#endif