Btf-rs

Rust library for the BPF Type Format (BTF). The BPF Type Format is a metadata format encoding debugging information such as types, function prototypes, structure layouts, etc. and is often used, but not limited, to deal with eBPF programs.

[dependencies]
btf-rs = "2.0"

The integration tests give good examples on how to use this library. We recommend reading the official BTF documentation as this library is offering a low-level API.

This library was initially developed for a Linux kernel packet tracking tool, Retis, but is exported on its own as there are no specific ties with the mentioned tool and can be (re)used in all kinds of Rust projects.

Overview

Parsing BTF

The main object this library offers is Btf, which represents a parsed BTF object. It offers helpers to resolve ids (u32), names (String) and types / chained types (Type).

Btf can be constructed using a BTF file or a split BTF one. BTF files hold self-contained information, while split BTF files are built upon a base BTF file and extend it. For example, in a standard Linux environment BTF files and split files can be found under /sys/kernel/btf, /sys/kernel/btf/vmlinux being the BTF file for the kernel and other files matching /sys/kernel/btf/<module-name> being BTF split files for its modules.

use btf_rs::Btf;

let base = Btf::from_file("/sys/kernel/btf/vmlinux").unwrap();

let ovs = Btf::from_split_file("/sys/kernel/btf/openvswitch", &base).unwrap();
let bbr = Btf::from_split_file("/sys/kernel/btf/tcp_bbr", &base).unwrap();

Btf-rs also supports constructing Btf using byte slices.

use std::fs;
use btf_rs::Btf;

let base = Btf::from_bytes(&fs::read("/sys/kernel/btf/vmlinux").unwrap()).unwrap();

let ovs = Btf::from_split_bytes(&fs::read("/sys/kernel/btf/openvswitch").unwrap(), &base)
          .unwrap();
let bbr = Btf::from_split_bytes(&fs::read("/sys/kernel/btf/bbr").unwrap(), &base).unwrap();

Resolving types

Types can be resolved using a Btf object. The following is an example of how a function can be inspected to retrieve information about its first parameter. Here the function kfree_skb_reason is taking a struct sk_buff * as its first argument.

use btf_rs::*;

let btf = Btf::from_file("/sys/kernel/btf/vmlinux").unwrap();

let func = match btf
    .resolve_types_by_name("kfree_skb_reason")
    .unwrap()
    .pop()
    .unwrap()
{
    Type::Func(func) => func,
    _ => panic!("Resolved type is not a function"),
};

let proto = match btf.resolve_chained_type(&func).unwrap() {
    Type::FuncProto(proto) => proto,
    _ => panic!("Resolved type is not a function proto"),
};

assert!(proto.parameters.len() > 1);

// The following prints "skb".
println!("{}", btf.resolve_name(&proto.parameters[0]).unwrap());

let ptr = match btf.resolve_chained_type(&proto.parameters[0]).unwrap() {
    Type::Ptr(ptr) => ptr,
    _ => panic!("Resolved type is not a pointer"),
};

let r#struct = match btf.resolve_chained_type(&ptr).unwrap() {
    Type::Struct(r#struct) => r#struct,
    _ => panic!("Resolved type is not a struct"),
};

// The following prints "sk_buff".
println!("{}", btf.resolve_name(&r#struct).unwrap());

Other information such as function scope and return value, structure size and members, etc. can be retrieved. For all those see the Type and its associated structures documentation.

Additional objects

Additional objects and helpers built on top of the ones described here can be found in the utils sub-module. Those are aimed at easing BTF consumption in common cases, such as parsing all BTF files of a running Linux kernel and its modules or dealing with .BTF ELF sections.

Feature flags

• elf: Enable helpers parsing the .BTF section of ELF files in utils::elf.
• elf-compression: Enable handling of compressed ELF files (e.g. compressed kernel modules) in utils::elf. The Bzip2, Gzip, Xz and Zstd compression algorithms are currently supported.
• regex: Enable name resolutions by regex (regex::Regex).

Btf backends

The Btf object supports different internal backends (Backend), optimized for different uses: Backend::Cache and Backend::Mmap. The former provides a faster API at the cost of slower initialization and increase in memory footprint. The latter provides a fast initialization time and lower memory footprint at the cast of slower API performances.

Backend::Cache is used by default. The backend can also be explicitly selected using dedicated constructors (see from_*_with_backend helpers). All BTF data read from bytes use Backend::Cache (this includes all elf feature helpers) and Backend::Mmap is only supported for base BTF files.

A benchmark utility is provided to gather actual numbers. It should be run on the target machine, to investigate how btf-rs would perform. E.g.

$ cargo build --release -F regex --example benchmark
$ ./target/release/examples/benchmark --help
[...]
$ ./target/release/examples/benchmark -i 100 \
    --id 77378 --name __kfree_skb --regex "^[[:alnum:]]+_drop_reason$" \
    --backend cache
[...]

A few parameters are required to configure the type id and name to use for querying the BTF data. Those can be selected by first inspecting the BTF data using bpftool (bpftool btf dump file /sys/kernel/btf/vmlinux).