summaryrefslogtreecommitdiff
path: root/libbcachefs/alloc_types.h
blob: 1bf48ef93a7390734c4bb9a8e0a1d042b39c402d (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
#ifndef _BCACHE_ALLOC_TYPES_H
#define _BCACHE_ALLOC_TYPES_H

#include <linux/mutex.h>

#include "clock_types.h"

/*
 * There's two of these clocks, one for reads and one for writes:
 *
 * All fields protected by bucket_lock
 */
struct prio_clock {
	/*
	 * "now" in (read/write) IO time - incremented whenever we do X amount
	 * of reads or writes.
	 *
	 * Goes with the bucket read/write prios: when we read or write to a
	 * bucket we reset the bucket's prio to the current hand; thus hand -
	 * prio = time since bucket was last read/written.
	 *
	 * The units are some amount (bytes/sectors) of data read/written, and
	 * the units can change on the fly if we need to rescale to fit
	 * everything in a u16 - your only guarantee is that the units are
	 * consistent.
	 */
	u16			hand;
	u16			min_prio;

	int			rw;

	struct io_timer		rescale;
};

/* There is one reserve for each type of btree, one for prios and gens
 * and one for moving GC */
enum alloc_reserve {
	RESERVE_PRIO,
	RESERVE_BTREE,
	RESERVE_METADATA_LAST = RESERVE_BTREE,
	RESERVE_MOVINGGC,

	RESERVE_NONE,
	RESERVE_NR,
};

static inline bool allocation_is_metadata(enum alloc_reserve id)
{
	return id <= RESERVE_METADATA_LAST;
}

struct dev_group {
	spinlock_t		lock;
	unsigned		nr;
	unsigned		cur_device;
	struct {
		u64		weight;
		struct bch_dev	*dev;
	}			d[BCH_SB_MEMBERS_MAX];
};

/* Enough for 16 cache devices, 2 tiers and some left over for pipelining */
#define OPEN_BUCKETS_COUNT	256

#define WRITE_POINT_COUNT	16

struct open_bucket {
	struct list_head	list;
	struct mutex		lock;
	atomic_t		pin;
	bool			has_full_ptrs;
	/*
	 * recalculated every time we allocate from this open_bucket based on
	 * how many pointers we're actually going to use:
	 */
	unsigned		sectors_free;
	unsigned		nr_ptrs;
	struct bch_extent_ptr	ptrs[BCH_REPLICAS_MAX];
	unsigned		ptr_offset[BCH_REPLICAS_MAX];
};

struct write_point {
	struct open_bucket	*b;

	/*
	 * Throttle writes to this write point if tier 0 is full?
	 */
	bool			throttle;

	/*
	 * If not NULL, cache group for tiering, promotion and moving GC -
	 * always allocates a single replica
	 */
	struct dev_group	*group;

	/*
	 * Otherwise do a normal replicated bucket allocation that could come
	 * from any device in tier 0 (foreground write)
	 */
};

#endif /* _BCACHE_ALLOC_TYPES_H */