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
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
use core::{
cell::UnsafeCell,
mem,
mem::MaybeUninit,
ops::{Deref, DerefMut},
};
#[cfg(feature = "const_assert")]
use const_assert::{Assert, IsTrue};
use crate::{
bindings::{bpf_map_def, bpf_map_type::BPF_MAP_TYPE_RINGBUF},
helpers::{
bpf_ringbuf_discard, bpf_ringbuf_output, bpf_ringbuf_query, bpf_ringbuf_reserve,
bpf_ringbuf_submit,
},
maps::PinningType,
};
#[repr(transparent)]
pub struct RingBuf {
def: UnsafeCell<bpf_map_def>,
}
unsafe impl Sync for RingBuf {}
/// A ring buffer entry, returned from [`RingBuf::reserve`].
///
/// You must [`submit`] or [`discard`] this entry before it gets dropped.
///
/// [`submit`]: RingBufEntry::submit
/// [`discard`]: RingBufEntry::discard
#[must_use = "eBPF verifier requires ring buffer entries to be either submitted or discarded"]
pub struct RingBufEntry<T: 'static>(&'static mut MaybeUninit<T>);
impl<T> Deref for RingBufEntry<T> {
type Target = MaybeUninit<T>;
fn deref(&self) -> &Self::Target {
self.0
}
}
impl<T> DerefMut for RingBufEntry<T> {
fn deref_mut(&mut self) -> &mut Self::Target {
self.0
}
}
impl<T> RingBufEntry<T> {
/// Discard this ring buffer entry. The entry will be skipped by the userspace reader.
pub fn discard(self, flags: u64) {
unsafe { bpf_ringbuf_discard(self.0.as_mut_ptr() as *mut _, flags) };
}
/// Commit this ring buffer entry. The entry will be made visible to the userspace reader.
pub fn submit(self, flags: u64) {
unsafe { bpf_ringbuf_submit(self.0.as_mut_ptr() as *mut _, flags) };
}
}
impl RingBuf {
/// Declare an eBPF ring buffer.
///
/// The linux kernel requires that `byte_size` be a power-of-2 multiple of the page size. The
/// loading program may coerce the size when loading the map.
pub const fn with_byte_size(byte_size: u32, flags: u32) -> Self {
Self::new(byte_size, flags, PinningType::None)
}
/// Declare a pinned eBPF ring buffer.
///
/// The linux kernel requires that `byte_size` be a power-of-2 multiple of the page size. The
/// loading program may coerce the size when loading the map.
pub const fn pinned(byte_size: u32, flags: u32) -> Self {
Self::new(byte_size, flags, PinningType::ByName)
}
const fn new(byte_size: u32, flags: u32, pinning_type: PinningType) -> Self {
Self {
def: UnsafeCell::new(bpf_map_def {
type_: BPF_MAP_TYPE_RINGBUF,
key_size: 0,
value_size: 0,
max_entries: byte_size,
map_flags: flags,
id: 0,
pinning: pinning_type as u32,
}),
}
}
/// Reserve memory in the ring buffer that can fit `T`.
///
/// Returns `None` if the ring buffer is full.
#[cfg(feature = "const_assert")]
pub fn reserve<T: 'static>(&self, flags: u64) -> Option<RingBufEntry<T>>
where
Assert<{ 8 % mem::align_of::<T>() == 0 }>: IsTrue,
{
self.reserve_impl(flags)
}
/// Reserve memory in the ring buffer that can fit `T`.
///
/// Returns `None` if the ring buffer is full.
///
/// The kernel will reserve memory at an 8-bytes aligned boundary, so `mem::align_of<T>()` must
/// be equal or smaller than 8. If you use this with a `T` that isn't properly aligned, this
/// function will be compiled to a panic; depending on your panic_handler, this may make
/// the eBPF program fail to load, or it may make it have undefined behavior.
#[cfg(not(feature = "const_assert"))]
pub fn reserve<T: 'static>(&self, flags: u64) -> Option<RingBufEntry<T>> {
assert_eq!(8 % mem::align_of::<T>(), 0);
self.reserve_impl(flags)
}
fn reserve_impl<T: 'static>(&self, flags: u64) -> Option<RingBufEntry<T>> {
let ptr = unsafe {
bpf_ringbuf_reserve(self.def.get() as *mut _, mem::size_of::<T>() as _, flags)
} as *mut MaybeUninit<T>;
unsafe { ptr.as_mut() }.map(|ptr| RingBufEntry(ptr))
}
/// Copy `data` to the ring buffer output.
///
/// Consider using [`reserve`] and [`submit`] if `T` is statically sized and you want to save a
/// copy from either a map buffer or the stack.
///
/// Unlike [`reserve`], this function can handle dynamically sized types (which is hard to
/// create in eBPF but still possible, e.g. by slicing an array).
///
/// Note: `T` must be aligned to no more than 8 bytes; it's not possible to fulfill larger
/// alignment requests. If you use this with a `T` that isn't properly aligned, this function will
/// be compiled to a panic and silently make your eBPF program fail to load.
/// See [here](https://github.com/torvalds/linux/blob/3f01e9fed/kernel/bpf/ringbuf.c#L418).
///
/// [`reserve`]: RingBuf::reserve
/// [`submit`]: RingBufEntry::submit
pub fn output<T: ?Sized>(&self, data: &T, flags: u64) -> Result<(), i64> {
assert_eq!(8 % mem::align_of_val(data), 0);
let ret = unsafe {
bpf_ringbuf_output(
self.def.get() as *mut _,
data as *const _ as *mut _,
mem::size_of_val(data) as _,
flags,
)
};
if ret < 0 {
Err(ret)
} else {
Ok(())
}
}
/// Query various information about the ring buffer.
///
/// Consult `bpf_ringbuf_query` documentation for a list of allowed flags.
pub fn query(&self, flags: u64) -> u64 {
unsafe { bpf_ringbuf_query(self.def.get() as *mut _, flags) }
}
}