OpenTelemetry eBPF Instrumentation: Java TLS ioctl kprobe allows kernel memory disclosure
Description
Summary
The Java TLS ioctl probe reads user-controlled ioctl pointers with bpf_probe_read instead of bpf_probe_read_user. An instrumented local process can therefore point OBI at kernel memory and cause that memory to be copied into telemetry.
Details
The vulnerable path is in bpf/generictracer/java_tls.c. The kprobe hooks do_vfs_ioctl, filters on fd == 0 and the Java TLS magic command, and then treats the third ioctl argument as a structured buffer. It reads fields from that pointer using bpf_probe_read, including:
- the operation byte from
arg - connection metadata from
arg + 1 - the payload length from
arg + 1 + sizeof(connection_info_t)
If len > 0, it computes buf = arg + 1 + sizeof(connection_info_t) + sizeof(u32) and passes that pointer into `handle_buf_with_connection`.
The next stage, bpf/generictracer/k_tracer_defs.h, uses `bpf_probe_read(args->small_buf, MIN_HTTP2_SIZE, (void *)args->u_buf);` on the supplied pointer and tail-calls deeper protocol logic. The HTTP protocol path then reads from u_buf and emits the bytes through bpf_ringbuf_output in bpf/generictracer/protocol_http.h.
Because the ioctl pointer originates in user space, the probe should be using bpf_probe_read_user with strict length validation. Using bpf_probe_read instead makes it possible for an instrumented process to supply a kernel pointer and exfiltrate kernel-resident bytes into telemetry.
PoC
A complete lab reproduction requires:
- a vulnerable build of OBI with Java TLS instrumentation enabled
- a host capable of loading the BPF program
- a local process that issues the Java TLS magic ioctl with an attacker-controlled pointer
Suggested reproduction steps:
git checkout v0.0.0-rc.1+build
make build
sudo ./bin/obi
Then run a local helper that issues the matching ioctl command against fd=0 and supplies a crafted pointer.
// save as /tmp/ioctl_kernel_ptr.c
#include <stdio.h>
#include <stdint.h>
#include <sys/ioctl.h>
#include <unistd.h>
#define JAVA_TLS_MAGIC 0x0b10b1
int main(void) {
void *ptr = (void *)0xffff888000000000ULL;
long rc = ioctl(0, JAVA_TLS_MAGIC, ptr);
printf("ioctl rc=%ld\n", rc);
return 0;
}
Compile and run:
cc -O2 -o /tmp/ioctl_kernel_ptr /tmp/ioctl_kernel_ptr.c
/tmp/ioctl_kernel_ptr
On a vulnerable system, if the supplied pointer references readable kernel memory and the bytes satisfy the expected Java TLS structure enough to pass the early checks, OBI can read from that address and emit the resulting bytes into telemetry. The remaining local prerequisite is a host session with sufficient BPF capability to load and inspect the probe; the compile side of the reproduction is already satisfied here.
Impact
This is a local kernel memory disclosure primitive reachable from unprivileged instrumented processes. It affects deployments that enable Java TLS support. Successful exploitation can expose kernel memory contents to the privileged OBI agent and then to downstream telemetry systems.
AI Insight
LLM-synthesized narrative grounded in this CVE's description and references.
OpenTelemetry eBPF instrumentation (OBI) v0.0.0-rc.1+build allows a local instrumented process to leak kernel memory via an ioctl kprobe that uses bpf_probe_read instead of bpf_probe_read_user.
Vulnerability
The Java TLS ioctl probe in OpenTelemetry eBPF Instrumentation (OBI) versions up to v0.0.0-rc.1+build uses bpf_probe_read to read user-controlled ioctl pointers instead of the safer bpf_probe_read_user [1][2][3]. The vulnerable code path is in bpf/generictracer/java_tls.c and bpf/generictracer/k_tracer_defs.h. The kprobe hooks do_vfs_ioctl, filters on fd == 0 and the Java TLS magic command (0x0b10b1), then treats the third ioctl argument as a structured buffer [2][3]. It reads fields (operation byte, connection metadata, payload length) using bpf_probe_read, and if len > 0, computes a buffer pointer and passes it to handle_buf_with_connection. The next stage uses bpf_probe_read(args->small_buf, MIN_HTTP2_SIZE, (void *)args->u_buf) and tail-calls protocol logic that eventually outputs bytes via bpf_ringbuf_output [2][3].
Exploitation
An attacker must have a local process that is already being instrumented by a vulnerable build of OBI with Java TLS instrumentation enabled [2][3]. The attacker does not need special privileges beyond the ability to execute code as the instrumented process, and the host must be capable of loading the BPF program [2][3]. The attacker runs a helper that issues the Java TLS magic ioctl command against fd=0 with a crafted pointer pointing to kernel memory [3]. No user interaction or race window is required beyond issuing the ioctl call in the instrumented process.
Impact
Upon success, the attacker can cause kernel memory at the supplied kernel address to be copied into telemetry output (ring buffer), leading to information disclosure of kernel-resident bytes [2][3]. The attacker gains no code execution or privilege escalation directly, but can exfiltrate sensitive kernel data (e.g., pointers, cryptographic material, other confidential data) that may facilitate further attacks.
Mitigation
OBI is currently in development with breaking changes expected between minor releases [1]. As of this CVE's publication date (2026-05-18), no fixed version is available in the provided references; the advisory notes the fix should replace bpf_probe_read with bpf_probe_read_user and enforce strict length validation [2][3]. Users are urged to pin to a specific semver release and review security advisories before upgrading [1]. If the Java TLS ioctl probe is not essential, the best workaround is to disable it by not loading the BPF program or by removing the corresponding kernel hook.
References
- [1] OpenTelemetry eBPF Instrumentation repository
- [2] GitHub Advisory for CVE-2026-45683
- [3] OBI security advisory (GHSA-fjq3-ffvr-vm46)
AI Insight generated on May 21, 2026. Synthesized from this CVE's description and the cited reference URLs; citations are validated against the source bundle.
Affected products
1- Range: < 0.9.0
Patches
0No patches discovered yet.
Vulnerability mechanics
AI mechanics synthesis has not run for this CVE yet.
References
2News mentions
0No linked articles in our index yet.