[go: up one dir, main page]

Skip to content

oskardrums/ebpf

Repository files navigation

ebpf

Erlang CI

Erlang eBPF library

Overview

ebpf is an Erlang library for creating and interacting with eBPF programs. The following modules are currently included:

  • ebpf_user: load eBPF programs and use loaded programs
  • ebpf_kern: generate eBPF instructions according to different parameters
  • ebpf_asm: eBPF assembly and disassembly routines
  • ebpf_maps: userspace API to eBPF maps, mimics the Erlang/OTP maps interface with eBPF maps

Documentation

The documentation for the latest release can be browsed on hexdocs. Documentation for the main branch is also available here. ebpf is documented with edoc, the docs can be built locally with

$ rebar3 edoc

Usage

Checkout the examples.

A minimal example is given below:

% Drop all packets
BinProg = ebpf_asm:assemble(ebpf_kern:return(0)),

{ok, FilterProg} = ebpf_user:load(socket_filter, BinProg),
{ok, Sock} = socket:open(inet, stream, {raw, 0}),
ok = ebpf_user:attach(Sock, FilterProg), % All new input to Sock is dropped
ok = ebpf_user:detach_socket_filter(Sock), % Sock is back to normal and FilterProg can be
ok = ebpf_user:close(FilterProg), % FilterProg is unloaded from the kernel

{ok, XdpProg} = ebpf_user:load(xdp, BinProg),
ok = ebpf_user:attach("lo", XdpProg), % Try pinging 127.0.0.1, go ahead
ok = ebpf_user:detach_xdp("lo"), % Now, that's better :)
ok = ebpf_user:close(XdpProg).

Add ebpf as a dependency in rebar.config:

% From hex
{deps, [ebpf]}.
% Or from github
{deps, [{ebpf, {git, "https://github.com/oskardrums/ebpf.git", "main"}}]}.

{error, eperm}

Most BPF operations require elevated permissions on most Linux systems. Lack of permissions usually manifests in ebpf in function calls failing with {error, eperm}.

To allow ebpf to run privileged operations, BEAM needs to be given permission to do so. The quickest way to do that for local testing is to run your program as super user, e.g.

$ sudo `which rebar3` shell

For production systems, Linux capabilities should be given to the user or to the BEAM executable. Most bpf(2) operations demand CAP_SYS_ADMIN capability, and some XDP operations demand CAP_NET_ADMIN.

Since Linux 4.4, socket_filter type eBPF programs can be loaded without elevated permissions under some conditions. For more information see the bpf(2) man page.

Build

$ rebar3 compile

ebpf uses NIFs to communicate with the Linux kernel eBPF system. You will need make, a C compiler and Linux headers for rebar3 to build the .so that contains those NIFs.

Test

$ rebar3 do ct, proper

Contributions

Are welcome :)

Feel free to open an issue or a PR if you encounter any problem or have an idea for an improvement.