Source: gVisor documentation (overview and platforms)
gVisor is a Go user-space application kernel (Sentry plus Gofer) that intercepts container syscalls via systrap or KVM platforms, trading syscall overhead and compatibility for isolation.
Contents
- What: official project documentation
- Publisher: gVisor project (created by Google, now a CNCF sandbox project)
- Date: living documents, read 2026-11-14
- Original: gVisor docs, Platforms
Summary
gVisor implements a Linux-like kernel interface in user space so that container syscalls never reach the host kernel directly. It ships runsc, an OCI runtime usable from Docker and Kubernetes.
Key claims
- The Sentry acts as the guest kernel (syscalls, signals, memory, threads) and itself makes only a limited set of host syscalls under seccomp. The Gofer, a per-container host process, mediates all filesystem access over 9P; the Sentry has no direct filesystem access.
- It is neither a seccomp filter nor a VM: no fixed hardware-virtualization overhead, faster startup, flexible footprint.
- Platforms: systrap (default since mid-2023, uses
SECCOMP_RET_TRAP/SIGSYS, best inside VMs), KVM (best on bare metal; worse under nested virtualization), ptrace (deprecated, high context-switch cost). - Trade-offs: higher per-syscall overhead (syscall-heavy workloads suffer) and incomplete syscall,
/procand/syscoverage, so some apps break, although most run unmodified. - Written in memory-safe Go.