1// SPDX-License-Identifier: GPL-2.0-or-later
 2/*
 3 * Copyright (C) 2022-2023 Oracle.  All Rights Reserved.
 4 * Author: Darrick J. Wong <djwong@kernel.org>
 5 */
 6#ifndef XFS_DRAIN_H_
 7#define XFS_DRAIN_H_
 8
 9struct xfs_group;
10struct xfs_perag;
11
12#ifdef CONFIG_XFS_DRAIN_INTENTS
13/*
14 * Passive drain mechanism.  This data structure tracks a count of some items
15 * and contains a waitqueue for callers who would like to wake up when the
16 * count hits zero.
17 */
18struct xfs_defer_drain {
19	/* Number of items pending in some part of the filesystem. */
20	atomic_t		dr_count;
21
22	/* Queue to wait for dri_count to go to zero */
23	struct wait_queue_head	dr_waiters;
24};
25
26void xfs_defer_drain_init(struct xfs_defer_drain *dr);
27void xfs_defer_drain_free(struct xfs_defer_drain *dr);
28
29void xfs_drain_wait_disable(void);
30void xfs_drain_wait_enable(void);
31
32/*
33 * Deferred Work Intent Drains
34 * ===========================
35 *
36 * When a writer thread executes a chain of log intent items, the AG header
37 * buffer locks will cycle during a transaction roll to get from one intent
38 * item to the next in a chain.  Although scrub takes all AG header buffer
39 * locks, this isn't sufficient to guard against scrub checking an AG while
40 * that writer thread is in the middle of finishing a chain because there's no
41 * higher level locking primitive guarding allocation groups.
42 *
43 * When there's a collision, cross-referencing between data structures (e.g.
44 * rmapbt and refcountbt) yields false corruption events; if repair is running,
45 * this results in incorrect repairs, which is catastrophic.
46 *
47 * The solution is to the perag structure the count of active intents and make
48 * scrub wait until it has both AG header buffer locks and the intent counter
49 * reaches zero.  It is therefore critical that deferred work threads hold the
50 * AGI or AGF buffers when decrementing the intent counter.
51 *
52 * Given a list of deferred work items, the deferred work manager will complete
53 * a work item and all the sub-items that the parent item creates before moving
54 * on to the next work item in the list.  This is also true for all levels of
55 * sub-items.  Writer threads are permitted to queue multiple work items
56 * targetting the same AG, so a deferred work item (such as a BUI) that creates
57 * sub-items (such as RUIs) must bump the intent counter and maintain it until
58 * the sub-items can themselves bump the intent counter.
59 *
60 * Therefore, the intent count tracks entire lifetimes of deferred work items.
61 * All functions that create work items must increment the intent counter as
62 * soon as the item is added to the transaction and cannot drop the counter
63 * until the item is finished or cancelled.
64 */
65struct xfs_group *xfs_group_intent_get(struct xfs_mount *mp,
66		xfs_fsblock_t fsbno, enum xfs_group_type type);
67void xfs_group_intent_put(struct xfs_group *rtg);
68
69int xfs_group_intent_drain(struct xfs_group *xg);
70bool xfs_group_intent_busy(struct xfs_group *xg);
71
72#else
73struct xfs_defer_drain { /* empty */ };
74
75#define xfs_defer_drain_free(dr)		((void)0)
76#define xfs_defer_drain_init(dr)		((void)0)
77
78#define xfs_group_intent_get(_mp, _fsbno, _type) \
79	xfs_group_get_by_fsb((_mp), (_fsbno), (_type))
80#define xfs_group_intent_put(xg)		xfs_group_put(xg)
81
82#endif /* CONFIG_XFS_DRAIN_INTENTS */
83
84#endif /* XFS_DRAIN_H_ */