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
9 struct xfs_perag;
10
11 #ifdef CONFIG_XFS_DRAIN_INTENTS
12 /*
13 * Passive drain mechanism. This data structure tracks a count of some items
14 * and contains a waitqueue for callers who would like to wake up when the
15 * count hits zero.
16 */
17 struct xfs_defer_drain {
18 /* Number of items pending in some part of the filesystem. */
19 atomic_t dr_count;
20
21 /* Queue to wait for dri_count to go to zero */
22 struct wait_queue_head dr_waiters;
23 };
24
25 void xfs_defer_drain_init(struct xfs_defer_drain *dr);
26 void xfs_defer_drain_free(struct xfs_defer_drain *dr);
27
28 void xfs_drain_wait_disable(void);
29 void xfs_drain_wait_enable(void);
30
31 /*
32 * Deferred Work Intent Drains
33 * ===========================
34 *
35 * When a writer thread executes a chain of log intent items, the AG header
36 * buffer locks will cycle during a transaction roll to get from one intent
37 * item to the next in a chain. Although scrub takes all AG header buffer
38 * locks, this isn't sufficient to guard against scrub checking an AG while
39 * that writer thread is in the middle of finishing a chain because there's no
40 * higher level locking primitive guarding allocation groups.
41 *
42 * When there's a collision, cross-referencing between data structures (e.g.
43 * rmapbt and refcountbt) yields false corruption events; if repair is running,
44 * this results in incorrect repairs, which is catastrophic.
45 *
46 * The solution is to the perag structure the count of active intents and make
47 * scrub wait until it has both AG header buffer locks and the intent counter
48 * reaches zero. It is therefore critical that deferred work threads hold the
49 * AGI or AGF buffers when decrementing the intent counter.
50 *
51 * Given a list of deferred work items, the deferred work manager will complete
52 * a work item and all the sub-items that the parent item creates before moving
53 * on to the next work item in the list. This is also true for all levels of
54 * sub-items. Writer threads are permitted to queue multiple work items
55 * targetting the same AG, so a deferred work item (such as a BUI) that creates
56 * sub-items (such as RUIs) must bump the intent counter and maintain it until
57 * the sub-items can themselves bump the intent counter.
58 *
59 * Therefore, the intent count tracks entire lifetimes of deferred work items.
60 * All functions that create work items must increment the intent counter as
61 * soon as the item is added to the transaction and cannot drop the counter
62 * until the item is finished or cancelled.
63 */
64 struct xfs_perag *xfs_perag_intent_get(struct xfs_mount *mp,
65 xfs_agnumber_t agno);
66 void xfs_perag_intent_put(struct xfs_perag *pag);
67
68 void xfs_perag_intent_hold(struct xfs_perag *pag);
69 void xfs_perag_intent_rele(struct xfs_perag *pag);
70
71 int xfs_perag_intent_drain(struct xfs_perag *pag);
72 bool xfs_perag_intent_busy(struct xfs_perag *pag);
73 #else
74 struct xfs_defer_drain { /* empty */ };
75
76 #define xfs_defer_drain_free(dr) ((void)0)
77 #define xfs_defer_drain_init(dr) ((void)0)
78
79 #define xfs_perag_intent_get(mp, agno) xfs_perag_get((mp), (agno))
80 #define xfs_perag_intent_put(pag) xfs_perag_put(pag)
81
xfs_perag_intent_hold(struct xfs_perag * pag)82 static inline void xfs_perag_intent_hold(struct xfs_perag *pag) { }
xfs_perag_intent_rele(struct xfs_perag * pag)83 static inline void xfs_perag_intent_rele(struct xfs_perag *pag) { }
84
85 #endif /* CONFIG_XFS_DRAIN_INTENTS */
86
87 #endif /* XFS_DRAIN_H_ */
88