X41 D-Sec GmbH Security Advisory: X41-2017-008 Multiple Vulnerabilities in Shadowsocks ======================================= Overview -------- Confirmed Affected Versions: Latest commit 2ab8c6b on Sep 6 Confirmed Patched Versions: N/A Vendor: Shadowsocks Vendor URL: https://github.com/shadowsocks/shadowsocks/tree/master Vector: Network Credit: X41 D-Sec GmbH, Niklas Abel Status: Public Advisory-URL: https://www.x41-dsec.de/lab/advisories/x41-2017-008-shadowsocks/ Summary and Impact ------------------ Several issues have been identified, which allow attackers to manipulate log files, execute commands and to brute force Shadowsocks with enabled autoban.py brute force detection. Brute force detection from autoban.py does not work with suggested tail command. The key of captured Shadowsocks traffic can be brute forced. Product Description ------------------- Shadowsocks is a fast tunnel proxy that helps you bypass firewalls. Log file manipulation ===================== Severity Rating: Medium Confirmed Affected Versions: Latest commit 2ab8c6b on Sep 6 Confirmed Patched Versions: N/A Vector: Network CVE: not yet issued CWE: 117 CVSS Score: 4.3 CVSS Vector: CVSS:3.0/AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:L/A:N Summary and Impact ------------------ Log file manipulation is possible with a manipulated hostname, sent to the server from a client, even if Shadowsocks is as quiet as possible with "-qq". Therefore a string like "\nI could be any log entry\n" could be sent as hostname to Shadowsocks. The server would log an additional line with "I could be any log entry". Workarounds ----------- There is no workaround available, do not trust the logfiles until a patch is released. Command Execution ================= Severity Rating: Critical Confirmed Affected Versions: Latest commit 2ab8c6b on Sep 6 Confirmed Patched Versions: N/A Vector: Network CVE: not yet issued CWE: 78 CVSS Score: 9.0 CVSS Vector: CVSS:3.0/AV:N/AC:H/PR:N/UI:N/S:C/C:H/I:H/A:H Summary and Impact ------------------ When the brute force detection with autoban.py is enabled, remote attackers are able to execute arbitrary commands. Command execution is possible because of because of line 53 "os.system(cmd)" in autoban.py, which executes "cmd = 'iptables -A INPUT -s %s -j DROP' % ip". The "ip" parameter gets parsed from the log file, whose contents can be controlled by a third party sending unauthenticated packets. Proof of Concept ---------------- When, a string like "can not parse header when ||ls&:\n" is sent as host name to Shadowsocks, it would end up in the logfile and lead to the execution of "ls". Autoban.py does not execute commands with spaces due to internal sanitization. A requested hostname like: " can not parse header when ||ls&:\ntouch /etc/evil.txt\nexit\ncan not parse header when ||/bin/bash log 2>log &" does not block IP's. The "for line in sys.stdin:" from autoban.py parses the input until there is an end of file (EOF). As "tail -F" will never pipe an EOF into the pyhon script, the sys.stdin will block the script forever. So the "tail -F /var/log/shodowsocks | autoban.py" will never block anything except itself. Workarounds ----------- Use python "autoban.py < /var/log/shadowsocks.log" in a cronjob. Do not use autoban.py until the command execution issue gets fixed. Bruteforcable Shadowsocks traffic because of MD5 ================================================ Confirmed Affected Versions: Latest commit 2ab8c6b on Sep 6 Confirmed Patched Versions: N/A Summary and Impact ------------------ Shadowsocks uses no brute force prevention for it's key derivation function. The key for Shadowsocks traffic encryption is static and derived from the password, using MD5. The password derivation is in encrypt.py in line 56 to 63: " while len(b''.join(m)) < (key_len + iv_len): md5 = hashlib.md5() data = password if i > 0: data = m[i - 1] + password md5.update(data) m.append(md5.digest()) i += 1 " MD5 should not be used to generate keys, since it is a hash function. A proper key derivation function increases the costs for this operation, which is a small burden for a user, but a big one for an attacker, which performs this operation many more times. As passwords usually have low-entropy, a good password derivation function has to be slow. Workarounds ----------- Use a secure password generated by a cryptographically secure random generator. Wait for a patch that uses a password based key derivation function like "Argon2" instead of a hash. About X41 D-Sec GmbH -------------------- X41 D-Sec is a provider of application security services. We focus on application code reviews, design review and security testing. X41 D-Sec GmbH was founded in 2015 by Markus Vervier. We support customers in various industries such as finance, software development and public institutions. Timeline -------- 2017-09-28 Issues found 2017-10-05 Vendor contacted 2017-10-09 Vendor contacted, replied to use GitHub for a full disclosure 2017-10-11 Vendor contacted, asked if the vendor is sure to want a full disclosure 2017-10-12 Vendor contacted, replied to create a public issue on GitHub 2017-10-13 Created public issues on GitHub 2017-10-13 Advisory release