Crate rxdp

Expand description

Rust bindings for working with XDP programs & eBPF maps.

This library has bindings for some of the common, basic operations needed when working with XDP programs & eBPF maps from user-space. It is built on top of libbpf-sys. At the time of writing, it supports only a subset of all the possibe eBPF map types (see tests directory for a good indication of which maps are supported).

§Prerequisites

Linux OS
libbpf-sys dependencies

§Examples

§Create an object from an ELF file

use rxdp;

let obj_path = "/path/to/elf/file";
let obj = match rxdp::XDPObject::new(obj_path) {
    Ok(obj) => {
        println!("Successfully created object from {}", obj_path);
        obj
    },
    Err(err) => panic!("{:?}", err),
};

§Set the pinned maps.

Maps that have already been pinned will be loaded from the fs, provided the map name matches the name in the fs. Any new maps in the HashSet will set the pin path so that once the program is loaded, they will get automatically pinned.


let mut pinned_maps = HashSet::new();
let pin_path = None; // Will default to /sys/fs/bpf
pinned_maps.insert("my_map_name".to_string());
obj.pinned_maps(&pinned_maps, pin_path).unwrap();

§Load the object (programs + maps) into the kernel.

This will consume the XDPObject created above and return an XDPLoadedObject.

let obj = obj.load().unwrap();

§Get a reference to a specific XDP program and attach it to an interface

let dev = "eth0";
let flags = rxdp::AttachFlags::SKB_MODE;

let prog = obj.get_program("prog_name").unwrap();
match prog.attach_to_interface(dev, flags) {
    Ok(_) => println!("Successfully attached to {}", dev),
    Err(e) => panic!("{:?}", e),
}

§Get access to an underlying eBPF `Map`

use rxdp::MapLike;

let m: rxdp::Map<u32, u64> = rxdp::Map::new(&obj, "map_name").unwrap();

NOTE: the key/value sizes MUST match the key/value sizes defined in the eBPF code, otherwise creating the map will fail.

§Perform map operations

use rxdp::MapLike;

let key = 0u32;
let value = 1000u64;
m.update(&key, &value, rxdp::MapFlags::BpfAny).unwrap();
let got = m.lookup(&key).unwrap();
assert_eq!(value, got.into_single());

// iterate through all items
for kv in m.items().unwrap() {
    println!("key: {}, value: {}", kv.key, kv.value.into_single());
}

§For per-cpu maps, use `PerCpuMap`

let m: rxdp::PerCpuMap<u32, u64> = rxdp::PerCpuMap::new(&obj, "map_name").unwrap();

NOTE: the key size MUST match the key size defined in the eBPF code, otherwise creating the map will fail.

§Per CPU map operations

Per CPU maps return the MapValue::Multi(Vec<T>) variant during lookup, one for each possible CPU:

use rxdp::MapLike;

let key = 0u32;
let value = 1000u64;
m.update(&key, &value, rxdp::MapFlags::BpfAny).unwrap();
let got = m.lookup(&key).unwrap();
assert_eq!(got.into_vec(), vec![value; rxdp::num_cpus()]);
// iterate through all items
for kv in m.items().unwrap() {
    println!("key: {}", kv.key);
    for v in kv.value.into_vec() {
        println!("value: {}", v);
    }
}

§Perf event Map

Perf events sent from eBPF can be retrieved via PerfMap.

let mut perfmap = rxdp::PerfMap::<u32>::new(&obj, "map_name").unwrap();
let r: Receiver<rxdp::PerfEvent<u32>> = perfmap.start_polling(10000);

// Wait for events on the receiver side of the channel
loop {
    r.recv().map_or_else(
        |e| println!("error: {:?}", e),
        |event| println!("event: {:?}", event),
    );
}

§Batching support (kernel dependent)

If the kernel supports it, you can do batch operations for update/lookups:

use rxdp::MapLike;

if rxdp::is_batching_supported() {
    let mut next_key = None;
    let r = m.lookup_batch(10u32, next_key).unwrap();
    // do something with r.items...
    next_key = r.next_key;
}

Structs§

AttachFlags: Flags that control how the XDP program is attached to the interface.
BatchResult: The result of a batch operation.
KeyValue: Holds key/value pair when getting all items from a map.
Map: Used for working with normal eBPF maps.
PerCpuMap: Used for working with per-cpu eBPF maps.
PerfEvent: The event sent from eBPF.
PerfMap: Used for working with a perf eBPF map.
XDPError: Error information about the attempted BPF operation
XDPLoadedObject: Struct for an XDP object that has been loaded
XDPObject: Convenience wrapper around an XDP object
XDPProgram: Convenience wrapper around an XDP program

Enums§

EventType: Event type from eBPF perf event map.
MapFlags: Flags that control map update behaviour.
MapType: Valid eBPF map types
MapValue: Return value from eBPF maps.

Traits§

ByteAligned: Trait used to convert types to/from 8 byte aligned Vec<u8> (required by per-cpu eBPF maps).
MapLike: This trait exposes the functionality of update/lookup/delete of underlying eBPF maps.

Functions§

is_batching_supported: True if kernel supports eBPF batch syscalls
load_pinned_object: Load a pinned object from a path. Returns the object fd.
num_cpus: Number of possible CPUs (not online CPUs).

Type Aliases§

XDPResult

Crate rxdpCopy item path