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-2026-91784 Overview
CVE-2026-91784 is a local argument injection vulnerability in cjbassi/gotop, a terminal-based graphical activity monitor written in Go. The flaw exists in the process termination feature, which passes the target process name directly to pkill without sanitization. A local attacker can create a process whose name starts with --, causing pkill to parse that name as a command-line option. When a gotop user invokes the kill action against the malicious process, pkill terminates processes matching the injected option rather than the intended target. The project is no longer actively maintained, and no fix is available. The vulnerability was confirmed in version 3.0.0; untested versions may also be affected.
Critical Impact
A local attacker with the ability to spawn processes can trick a gotop user into terminating arbitrary processes owned by a targeted user, including all processes of that user.
Affected Products
- cjbassi/gotop version 3.0.0 (confirmed)
- Other versions of cjbassi/gotop (untested, potentially affected)
- Project is unmaintained; no vendor patch is planned
Discovery Timeline
- 2026-10-02 - CVE-2026-91784 published to the National Vulnerability Database (NVD)
- 2026-10-02 - Last updated in NVD database
Technical Details for CVE-2026-91784
Vulnerability Analysis
The vulnerability is classified as argument injection under CWE-88: Improper Neutralization of Argument Delimiters in a Command. gotop exposes an interactive feature that lets the operator terminate a selected process. Internally, gotop builds a pkill invocation that includes the process name as a positional argument. Because the name is not separated from option parsing, strings beginning with -- are interpreted by pkill as flags rather than literal process names.
pkill supports selection options such as -U and --uid, which match processes by effective user ID. If a crafted process name resembles one of these flags, pkill will act on the matching selector. The attacker exploits the trust boundary between gotop's user interface and the underlying command executed with the UI user's privileges.
User interaction is required: the targeted gotop user must initiate the kill action on the crafted process entry. The attack is local only and does not require prior authentication to gotop itself.
Root Cause
The root cause is the absence of an argument-parsing terminator (--) and missing input sanitization before the process name is passed to pkill. Without the -- separator, pkill cannot distinguish user-supplied names from option flags. Any process name beginning with a dash is parsed as an option by the receiving binary.
Attack Vector
A local, unprivileged attacker creates a process whose argv[0] or process name begins with a pkill option such as --uid=<target_uid>. The attacker then waits for the gotop user, typically an administrator monitoring system activity, to select the entry and press the kill key. gotop invokes pkill <crafted_name>, which pkill interprets as an instruction to terminate every process owned by the targeted UID. This causes a denial-of-service condition against the victim user's workloads. Technical details are available in the CERT Polska advisory for CVE-2026-91784 and the cjbassi/gotop GitHub repository.
// No verified exploit code is published. The vulnerability is described in prose above.
// Refer to the CERT Polska advisory for additional technical context.
Detection Methods for CVE-2026-91784
Indicators of Compromise
- Processes with names or argv[0] values beginning with --, -U, --uid, --euid, or other pkill selectors
- Unexpected mass termination of processes owned by a single UID on hosts where gotop is in use
- pkill invocations in audit logs whose arguments contain user-controlled strings beginning with a dash
Detection Strategies
- Enable Linux auditd rules to log execve calls to /usr/bin/pkill and alert when any argument begins with -- or -
- Correlate gotop process activity with subsequent pkill executions and resulting process terminations
- Hunt for processes started by non-privileged users whose command names or titles begin with dash characters
Monitoring Recommendations
- Monitor execution of gotop on multi-user systems and track which account launched it
- Alert on unexpected termination of long-running services, especially when the terminating process is pkill invoked by an interactive shell session
- Collect endpoint telemetry on process name manipulation using prctl(PR_SET_NAME) or argv[0] rewrites by unprivileged users
How to Mitigate CVE-2026-91784
Immediate Actions Required
- Stop using cjbassi/gotop on multi-user systems or any host where untrusted local users can create processes
- Replace gotop with an actively maintained alternative such as btop, htop, or glances
- Restrict gotop usage to single-user workstations where no untrusted local accounts exist
Patch Information
No patch is available. The upstream cjbassi/gotop project is unmaintained, and the maintainer has not released a fix. Consult the cjbassi/gotop GitHub repository for the current project status and the CERT Polska advisory for coordinated disclosure details.
Workarounds
- Avoid using the in-app kill feature in gotop; terminate processes manually with kill <pid> using an explicit numeric PID
- Remove or restrict the pkill binary's availability to the account running gotop where feasible
- Apply a local fork or patch that inserts a -- argument-parsing terminator before the process name when invoking pkill
- Use mandatory access controls such as AppArmor or SELinux to constrain what gotop can execute
# Safer manual termination avoiding pkill name parsing
kill -TERM <numeric_pid>
# If forking gotop, ensure pkill invocations use the end-of-options terminator
# Example corrected invocation pattern:
pkill -- "$process_name"
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

