Download Latest Version v0.22.0_ Linux 7.1 compat, BPF tokens and vmlinux BTF caching changes source code.zip (2.6 MB)
Email in envelope

Get an email when there's a new version of eBPF

Home / v0.22.0
Name Modified Size InfoDownloads / Week
Parent folder
README.md 2026-06-26 10.1 kB
0
v0.22.0_ Linux 7.1 compat, BPF tokens and vmlinux BTF caching changes source code.tar.gz 2026-06-26 2.3 MB
1 1 weekly downloads
v0.22.0_ Linux 7.1 compat, BPF tokens and vmlinux BTF caching changes source code.zip 2026-06-26 2.6 MB
2 2 weekly downloads
Totals: 3 Items   5.0 MB 3

Hi everyone! Today we announce the v0.22.0 release of ebpf-go. We ship Linux 7.1 compatibility, older versions of the library will not work on 7.1 kernels and above, so please upgrade if you are targeting the latest releases of Linux. We also made a breaking change to BTF caching, please read those notes. And of course a few fixes, improvements and minor features.

Kernel 7.1 compatibility

In kernel 7.1 the BTF header was extended to introduce a new feature called BTF layout. This change caused our BTF parser to fail when parsing vmlinux for this kernel. This has been fixed in this latest release, we recommend upgrading to this latest release to avoid breakage on 7.1 and newer kernels.

See https://github.com/cilium/ebpf/pull/2042 for details. Special thanks to @Capricornus007 for making the bug report that allowed us to fix this in a timely manner.

BPF token support

Some BPF-related actions require the user to have root privileges (CAP_SYS_ADMIN). One example is loading and attaching programs that can inspect kernel memory. In some environments, you may want to permit a known good process to load such BPF programs, but you don't want to grant it CAP_SYS_ADMIN to avoid privilege escalation if that process gets compromised.

BPF tokens are a mechanism that allows a privileged process to delegate fine-grained BPF capabilities to an unprivileged process. The process of delegating is complex and typically handled by a container runtime such as LXC or a process manager like systemd. This part of the handshake is currently out of scope of ebpf-go, as the library is not in charge of process creation where this delegation takes place.

However, the consuming side is what's included in this release. ebpf-go will automatically detect when the current process is running in a namespace where a BPF token is provided, and will automatically try to obtain an use it for interacting with the BPF syscall. Unlike libbpf, this currently requires no extra configuration on behalf of the application.

See https://github.com/cilium/ebpf/pull/1953 for more details.

BTF cache changes

Package btf used to cache kernel (vmlinux) BTF specs globally. This is a significant time gain when loading multiple Collections in a row. However, doing so comes at a fairly significant memory cost (~20 MiB), so users could flush this cache with btf.FlushKernelSpec (pkg.go.dev). Unfortunately, the caching behaviour being opt-out means users would typically discover it while investigating memory usage, and would then have to find an appropriate time during execution to call the flush function. This was always a band-aid for something we didn't have a clear solution to.

With this update, we've removed the global cache and the btf.FlushKernelSpec (pkg.go.dev) function, which may slow down subsequent collection loading on busy systems. To opt back in, users can now maintain their own cache object, obtained from btf.NewCache (pkg.go.dev) and pass it to NewCollectionWithOptions via CollectionOptions.Cache (pkg.go.dev). Typically, you would put this in a global variable in a bpf-related package in your application, or keep it around in function scope if you load multiple collections in a row.

See https://github.com/cilium/ebpf/pull/1988 for more details. Thank you @matthyx for these changes.

New features

Bug fixes and improvements

Miscellaneous changes

New Contributors

Full Changelog: https://github.com/cilium/ebpf/compare/v0.21.0...v0.22.0

Source: README.md, updated 2026-06-26