bpftime: Userspace eBPF runtime for fast Uprobe & Syscall Hook & Extensions

bpftime, a full-featured, high-performance eBPF runtime designed to operate in userspace. It offers fast Uprobe and Syscall hook capabilities: Userspace uprobe can be 10x faster than kernel uprobe! and can programmatically hook all syscalls of a process safely and efficiently.

📦 Key Features
🔨 Quick Start
🔌 Examples & Use Cases
⌨️ Linux Plumbers 23 talk
📖 Slides
📚 Arxiv preprint

Checkout our documents in eunomia.dev!

Key Features

Uprobe and Syscall hooks based on binary rewriting: Run eBPF programs in userspace, attaching them to Uprobes and Syscall tracepoints: No manual instrumentation or restart required!. It can trace or change the execution of a function, hook or filter all syscalls of a process safely, and efficiently with an eBPF userspace runtime. Can inject eBPF runtime into any running process without the need for a restart or manual recompilation.
Performance: Experience up to a 10x speedup in Uprobe overhead compared to kernel uprobe and uretprobe. Read/Write userspace memory is also faster than kernel eBPF.
Interprocess eBPF Maps: Implement userspace eBPF maps in shared userspace memory for summary aggregation or control plane communication.
Compatibility: use existing eBPF toolchains like clang, libbpf and bpftrace to develop userspace eBPF application without any modifications. Supporting CO-RE via BTF, and offering userspace ufunc access.
Multi JIT Support: Support llvmbpf, a high-speed JIT/AOT compiler powered by LLVM, or using ubpf JIT and INTERPRETER. The vm can be built as a standalone library like ubpf.
Run with kernel eBPF: Can load userspace eBPF from kernel, and using kernel eBPF maps to cooperate with kernel eBPF programs like kprobes and network filters.

Components

vm: The eBPF VM and JIT compiler for bpftime, you can choose from bpftime LLVM JIT/AOT compiler and ubpf. The llvm-based vm in bpftime can also be built as a standalone library and integrated into other projects, similar to ubpf.
runtime: The userspace runtime for eBPF, including the bpf-syscall loader(syscall-server) and agent, support attaching eBPF programs to Uprobes, Syscall tracepoints and other events, as well as eBPF maps in shared memory.
verifier: Support using PREVAIL as userspace verifier, or using Linux kernel verifier as an option.
daemon: A daemon to make userspace eBPF working with kernel and compatible with kernel uprobe. Monitor and modify kernel eBPF events and syscalls, load eBPF in userspace from kernel.

Quick Start

With bpftime, you can build eBPF applications using familiar tools like clang and libbpf, and execute them in userspace. For instance, the malloc eBPF program traces malloc calls using uprobe and aggregates the counts using a hash map.

You can refer to eunomia.dev/bpftime/documents/build-and-test for how to build the project, or using the container images from GitHub packages.

To get started, you can build and run a libbpf based eBPF program starts with bpftime cli:

make -C example/malloc # Build the eBPF program example
export PATH=$PATH:~/.bpftime/
bpftime load ./example/malloc/malloc

In another shell, Run the target program with eBPF inside:

$ bpftime start ./example/malloc/victim
Hello malloc!
malloc called from pid 250215
continue malloc...
malloc called from pid 250215

You can also dynamically attach the eBPF program with a running process:

$ ./example/malloc/victim & echo $! # The pid is 101771
[1] 101771
101771
continue malloc...
continue malloc...

And attach to it:

$ sudo bpftime attach 101771 # You may need to run make install in root
Inject: "/root/.bpftime/libbpftime-agent.so"
Successfully injected. ID: 1

You can see the output from original program:

$ bpftime load ./example/malloc/malloc
...
12:44:35 
        pid=247299      malloc calls: 10
        pid=247322      malloc calls: 10

Alternatively, you can also run our sample eBPF program directly in the kernel eBPF, to see the similar output. This can be an example of how bpftime can work compatibly with kernel eBPF.

$ sudo example/malloc/malloc
15:38:05
        pid=30415       malloc calls: 1079
        pid=30393       malloc calls: 203
        pid=29882       malloc calls: 1076
        pid=34809       malloc calls: 8

See eunomia.dev/bpftime/documents/usage for more details.

Examples & Use Cases

⚠️ Note: bpftime is actively under development, and it's not yet recommended for production use. See our roadmap for details. We'd love to hear your feedback and suggestions! Please feel free to open an issue or Contact us.

For more examples and details, please refer to eunomia.dev/bpftime/documents/examples/ webpage.

Examples including:

Minimal examples of eBPF programs.
eBPF Uprobe/USDT tracing and syscall tracing:
- sslsniff for trace SSL/TLS unencrypted data.
- opensnoop for trace file open syscalls.
- More bcc/libbpf-tools.
- Run with bpftrace commands or scripts.
error injection: change function behavior with bpf_override_return.
Use the eBPF LLVM JIT/AOT vm as a standalone library.
Userspace XDP eBPF with DPDK

In-Depth

How it Works

bpftime supports two modes:

Running in userspace only

Left: original kernel eBPF | Right: bpftime

In this mode, bpftime can run eBPF programs in userspace without kernel, so it can be ported into low version of Linux or event other systems, and running without root permissions. It relies on a userspace verifier to ensure the safety of eBPF programs.

Run with kernel eBPF

In this mode, bpftime can run together with kernel eBPF. It can load eBPF programs from kernel, and using kernel eBPF maps to cooperate with kernel eBPF programs like kprobes and network filters.

Instrumentation implementation

Current hook implementation is based on binary rewriting and the underly technique is inspired by:

Userspace function hook: frida-gum
Syscall hooks: zpoline and pmem/syscall_intercept.

The hook can be easily replaced with other DBI methods or frameworks, or add more hook mechanisms in the future.

See our draft arxiv paper bpftime: userspace eBPF Runtime for Uprobe, Syscall and Kernel-User Interactions for details.

Performance Benchmarks

How is the performance of userspace uprobe compared to kernel uprobes?

Probe/Tracepoint Types	Kernel (ns)	Userspace (ns)
Uprobe	3224.172760	314.569110
Uretprobe	3996.799580	381.270270
Syscall Tracepoint	151.82801	232.57691
Manually Instrument	Not avaliable	110.008430

It can be attached to functions in running process just like the kernel uprobe does.

How is the performance of LLVM JIT/AOT compared to other eBPF userspace runtimes, native code or wasm runtimes?

Across all tests, the LLVM JIT for bpftime consistently showcased superior performance. Both demonstrated high efficiency in integer computations (as seen in log2_int), complex mathematical operations (as observed in prime), and memory operations (evident in memcpy and strcmp). While they lead in performance across the board, each runtime exhibits unique strengths and weaknesses. These insights can be invaluable for users when choosing the most appropriate runtime for their specific use-cases.

see github.com/eunomia-bpf/bpf-benchmark for how we evaluate and details.

Hash map or ring buffer compared to kernel(TODO)

See benchmark dir for detail performance benchmarks.

Comparing with Kernel eBPF Runtime

bpftime allows you to use clang and libbpf to build eBPF programs, and run them directly in this runtime. We have tested it with a libbpf version in third_party/libbpf. No specify libbpf or clang version needed.
Some kernel helpers and kfuncs may not be available in userspace.
It does not support direct access to kernel data structures or functions like task_struct.

Refer to eunomia.dev/bpftime/documents/available-features for more details.

Build and test

See eunomia.dev/bpftime/documents/build-and-test for details.

Roadmap

bpftime is continuously evolving with more features in the pipeline:

An AOT compiler for eBPF based on the LLVM.
More examples and usecases:
- Network on userspace eBPF
- Hotpatch userspace application
- Error injection and filter syscall
- Hotpatch and use iouring to batch syscall
- etc...
More map types and distribution maps support.
More program types support.

Stay tuned for more developments from this promising project! You can find bpftime on GitHub.

License

This project is licensed under the MIT License.

Contact and citations

Have any questions or suggestions on future development? Free free to open an issue or contact yunwei356@gmail.com !

Our arxiv preprint: https://arxiv.org/abs/2311.07923

@misc{zheng2023bpftime,
      title={bpftime: userspace eBPF Runtime for Uprobe, Syscall and Kernel-User Interactions}, 
      author={Yusheng Zheng and Tong Yu and Yiwei Yang and Yanpeng Hu and XiaoZheng Lai and Andrew Quinn},
      year={2023},
      eprint={2311.07923},
      archivePrefix={arXiv},
      primaryClass={cs.OS}
}

Acknowledgement

eunomia-bpf community is sponsored by PLCT Lab from ISCAS.

Thanks for other sponsors and discussions help building this project: Prof. Marios Kogias from Imperial College London, Prof. Xiaozheng lai from SCUT, Prof lijun chen from XUPT, Prof. Qi Li from THU NISL Lab, and Linux eBPF maintainers in the LPC 23 eBPF track.

zbs4017/bpftime