Disclaimer: This content was generated using AI. While we strive for accuracy, please verify critical information with official sources.
Sources: National Vulnerability Database, CISA KEV, Exploit-DB
CVE-2024-10603 Overview
CVE-2024-10603 affects Google gVisor, an application kernel that provides a sandboxed container runtime. The vulnerability stems from weaknesses in how gVisor generates TCP/UDP source ports and other network header values. An external attacker can predict these values under certain circumstances, undermining assumptions that protect network sessions from off-path attacks. The weakness maps to [CWE-340: Generation of Predictable Numbers or Identifiers]. Google addressed the issue through multiple patches that randomize source port selection, strengthen IPv6 fragment ID generation, and route randomness through a cryptographically secure RNG.
Critical Impact
Predictable transport-layer identifiers enable off-path network attackers to guess source ports and header fields, facilitating connection fingerprinting, traffic injection, and session interference against sandboxed workloads running on gVisor.
Affected Products
- Google gVisor (all versions prior to the fix commits)
- Google gVisor release 20231106.0
- Container workloads using gVisor's netstack networking backend
Discovery Timeline
- 2025-01-30 - CVE-2024-10603 published to NVD
- 2026-06-17 - Last updated in NVD database
Technical Details for CVE-2024-10603
Vulnerability Analysis
gVisor implements its own user-space TCP/IP stack called netstack. Three subsystems in that stack produced identifiers that were insufficiently random. TCP source ports were selected deterministically rather than randomized, IPv6 fragment identifiers were derived from a hash function (hash/fnv) rather than a secure source, and the general-purpose RNG used for non-Linux builds was not guaranteed to be cryptographically secure. An attacker who can observe or guess any of these values gains a foothold for off-path attacks that normally require on-path access, including TCP reset injection, data injection into unencrypted sessions, and fragment-based identification of a sandboxed host.
Root Cause
The root cause is predictable identifier generation [CWE-340]. The pkg/tcpip/ports/ports.go port allocator relied on atomicbitops counters without random selection. The pkg/tcpip/network/ipv6/ipv6.go fragment ID path used encoding/binary with FNV hashing. The pkg/rand/rand.go package on non-Linux platforms did not consistently use crypto/rand for security-sensitive consumers.
Attack Vector
Exploitation requires network access to a workload sandboxed by gVisor but no authentication or user interaction. An off-path adversary can predict upcoming TCP source ports or IPv6 fragment IDs, then forge packets that match in-flight sessions. The impact is bounded to confidentiality and integrity degradation of network flows rather than full code execution.
// Patch: Randomize TCP source port selection in pkg/tcpip/ports/ports.go
import (
"math"
- "gvisor.dev/gvisor/pkg/atomicbitops"
"gvisor.dev/gvisor/pkg/rand"
"gvisor.dev/gvisor/pkg/sync"
"gvisor.dev/gvisor/pkg/tcpip"
Source: gVisor commit cbdb2c6
// Patch: Replace FNV hash with secure source for IPv6 fragment IDs in pkg/tcpip/network/ipv6/ipv6.go
package ipv6
import (
- "encoding/binary"
"fmt"
- "hash/fnv"
"math"
"reflect"
"sort"
Source: gVisor commit 5d2bf25
// Patch: Use cryptographically secure RNG in pkg/rand/rand.go
//go:build !linux
// +build !linux
-// Package rand implements a cryptographically secure pseudorandom number
-// generator.
package rand
import "crypto/rand"
Source: gVisor commit 83f7508
Detection Methods for CVE-2024-10603
Indicators of Compromise
- Unexpected TCP RST floods targeting established connections from sandboxed workloads
- Repeated IPv6 fragmented packets with sequential or guessable identification fields
- Anomalous spikes in packets matching active connection 5-tuples from external sources
Detection Strategies
- Collect netstack logs from gVisor runsc instances and alert on repeated connection resets originating from untrusted networks
- Analyze packet captures for IPv6 fragment IDs that lack entropy across sampled flows
- Correlate workload-level connection termination events with concurrent off-path probe traffic
Monitoring Recommendations
- Ingest gVisor runtime and host network telemetry into a central analytics pipeline for cross-session correlation
- Monitor the installed gVisor commit or release version across container hosts to confirm patched builds are deployed
- Baseline source-port distributions from sandboxed workloads and alert when distributions collapse to narrow ranges
How to Mitigate CVE-2024-10603
Immediate Actions Required
- Upgrade gVisor to a release that includes commits cbdb2c6, 5d2bf25, and 83f7508
- Audit container platforms that embed gVisor (such as GKE Sandbox) and apply vendor-shipped updates
- Restrict direct network exposure of sandboxed workloads to untrusted sources where possible
Patch Information
Google merged three fixes into gVisor: TCP source port randomization in commit cbdb2c6, IPv6 fragment ID algorithm change in commit 5d2bf25, and consistent use of a cryptographically secure RNG in commit 83f7508. Technical background is published in the NDSS 2025 paper.
Workarounds
- Terminate sensitive sessions inside TLS or IPsec tunnels so predictable transport headers do not expose plaintext for injection
- Place sandboxed workloads behind stateful firewalls that validate connection ownership before forwarding packets
- Prefer IPv4 where feasible until IPv6 fragment ID randomization is confirmed deployed
# Verify installed gVisor runtime version on container hosts
runsc --version
# Pull a patched release and reinstall
curl -fsSL https://storage.googleapis.com/gvisor/releases/release/latest/x86_64/runsc -o runsc
chmod +x runsc && sudo mv runsc /usr/local/bin/
runsc --version
Disclaimer: This content was generated using AI. While we strive for accuracy, please verify critical information with official sources.
Sources: National Vulnerability Database, CISA KEV, Exploit-DB

