SSLproxy - transparent SSL/TLS proxy for decrypting and diverting network traffic to other programs for deep SSL inspection

Copyright (C) 2017-2021, Soner Tari. https://github.com/sonertari/SSLproxy

Copyright (C) 2009-2019, Daniel Roethlisberger. https://www.roe.ch/SSLsplit

Overview

SSLproxy is a proxy for SSL/TLS encrypted network connections. It is intended to be used for decrypting and diverting network traffic to other programs, such as UTM services, for deep SSL inspection.

The UTMFW project uses SSLproxy to decyrpt and feed network traffic into its UTM services: Web Filter, POP3 Proxy, SMTP Proxy, and Inline IPS; and also indirectly into Virus Scanner and Spam Filter through those UTM software. Given that most of the Internet traffic is encrypted now, it wouldn't be possible without SSLproxy to deeply inspect most of the network traffic passing through UTMFW.

See this presentation for a summary of SSL interception and potential issues with middleboxes that support it.

Mode of operation

SSLproxy is designed to transparently terminate connections that are redirected to it using a network address translation engine. SSLproxy then terminates SSL/TLS and initiates a new SSL/TLS connection to the original destination address. Packets received on the client side are decrypted and sent to the program listening on a port given in the proxy specification. SSLproxy inserts in the first packet the address and port it is expecting to receive the packets back from the program. Upon receiving the packets back, SSLproxy re-encrypts and sends them to their original destination. The return traffic follows the same path back to the client in reverse order.

This is similar in principle to divert sockets, where the packet filter diverts the packets to a program listening on a divert socket, and after processing the packets the program reinjects them into the kernel. If there is no program listening on that divert socket or the program does not reinject the packets into the kernel, the connection is effectively blocked. In the case of SSLproxy, SSLproxy acts as both the packet filter and the kernel, and the communication occurs over networking sockets.

SSLproxy supports split mode of operation similar to SSLsplit as well. In split mode, packets are not diverted to listening programs, effectively making SSLproxy behave similar to SSLsplit, but not exactly like it, because SSLproxy has certain features non-existent in SSLsplit, such as user authentication and protocol validation. Also, note that the implementation of proxy core in SSLproxy is different from the one in SSLsplit, for example the proxy core in SSLproxy runs lockless, whereas SSLsplit implementation uses thread manager level locks (which does not necessarily make sslproxy run faster than sslsplit). In SSLproxy, split mode can be defined globally or per-proxyspec.

SSLproxy does not automagically redirect any network traffic. To actually implement a proxy, you also need to redirect the traffic to the system running sslproxy. Your options include running sslproxy on a legitimate router, ARP spoofing, ND spoofing, DNS poisoning, deploying a rogue access point (e.g. using hostap mode), physical recabling, malicious VLAN reconfiguration or route injection, /etc/hosts modification and so on.

Proxy specification

For example, given the following proxy specification:

https 127.0.0.1 8443 up:8080

• SSLproxy listens for HTTPS connections on 127.0.0.1:8443.
• Upon receiving a connection from the Client, it decrypts and diverts the packets to a Program listening on 127.0.0.1:8080. The default divert address is 127.0.0.1, which can be configured by the ua option.
• After processing the packets, the Program gives them back to SSLproxy listening on a dynamically assigned address, which the Program obtains from the SSLproxy line in the first packet in the connection.
• Then SSLproxy re-encrypts and sends the packets to the Server.

The response from the Server follows the same path back to the Client in reverse order.

See the SSLsplit documentation for split style and structured proxyspecs.

SSLproxy line

A sample line SSLproxy inserts into the first packet in the connection is the following:

SSLproxy: [127.0.0.1]:34649,[192.168.3.24]:47286,[192.168.111.130]:443,s

• The first IP:port pair is a dynamically assigned address that SSLproxy expects the program send the packets back to it.
• The second and third IP:port pairs are the actual source and destination addresses of the connection respectively. Since the program receives the packets from SSLproxy, it cannot determine the source and destination addresses of the packets by itself, e.g by asking the NAT engine, hence must rely on the information in the SSLproxy line.
• The last letter is either s or p, for SSL/TLS encrypted or plain traffic respectively. This information is also important for the program, because it cannot reliably determine if the actual network traffic it is processing was encrypted or not before being diverted to it.

Listening program

The program that packets are diverted to should support this mode of operation. Specifically, it should be able to recognize the SSLproxy address in the first packet, and give the first and subsequent packets back to SSLproxy listening on that address, instead of sending them to the original destination as it normally would.

You can use any software as a listening program as long as it supports this mode of operation. So existing or new software developed in any programming language can be modified to be used with SSLproxy to inspect and/or modify any or all parts of the packets diverted to it.

You can offload the system SSLproxy is running on by diverting packets to remote listening programs too. For example, given the following proxy specification:

https 127.0.0.1 8443 up:8080 ua:192.168.0.1 ra:192.168.1.1

• The ua option instructs SSLproxy to divert packets to 192.168.0.1:8080, instead of 127.0.0.1:8080 as in the previous proxyspec example.
• The ra option instructs SSLproxy to listen for returned packets from the program on 192.168.1.1, instead of 127.0.0.1 as in the previous SSLproxy line.

Accordingly, the SSLproxy line now becomes:

SSLproxy: [192.168.1.1]:34649,[192.168.3.24]:47286,[192.168.111.130]:443,s

So, the listening program can be running on a machine anywhere in the world. Since the packets between SSLproxy and the listening program are always unencrypted, you should be careful while using such a setup.

Protocols

Supported protocols

SSLproxy supports plain TCP, plain SSL, HTTP, HTTPS, POP3, POP3S, SMTP, and SMTPS connections over both IPv4 and IPv6. It also has the ability to dynamically upgrade plain TCP to SSL in order to generically support SMTP STARTTLS and similar upgrade mechanisms. Depending on the version of OpenSSL, SSLproxy supports SSL 3.0, TLS 1.0, TLS 1.1, TLS 1.2, and TLS 1.3, and optionally SSL 2.0 as well. SSLproxy supports Server Name Indication (SNI), but not Encrypted SNI in TLS 1.3. It is able to work with RSA, DSA and ECDSA keys and DHE and ECDHE cipher suites.

The following features of SSLproxy are IPv4 only:

• Divert addresses for listening programs in proxyspecs
• SSLproxy return addresses dynamically assigned to connections
• IP addresses in the ua and ra options
• IP and ethernet addresses of clients in user authentication
• Target IP and ethernet addresses in mirror logging

OCSP, HPKP, HSTS, Upgrade et al.

SSLproxy implements a number of defences against mechanisms which would normally prevent MitM attacks or make them more difficult. SSLproxy can deny OCSP requests in a generic way. For HTTP and HTTPS connections, SSLproxy mangles headers to prevent server-instructed public key pinning (HPKP), avoid strict transport security restrictions (HSTS), avoid Certificate Transparency enforcement (Expect-CT) and prevent switching to QUIC/SPDY, HTTP/2 or WebSockets (Upgrade, Alternate Protocols). HTTP compression, encodings and keep-alive are disabled to make the logs more readable.

Another reason to disable persistent connections is to reduce file descriptor usage. Accordingly, connections are closed if they remain idle for a certain period of time. The default timeout is 120 seconds, which can be configured by the ConnIdleTimeout option.

Protocol validation

Protocol validation makes sure the traffic handled by a proxyspec is using the protocol specified in that proxyspec. The ValidateProto option can be used to enable global and/or per-proxyspec protocol validation. This feature currently supports HTTP, POP3, and SMTP protocols. If a connection cannot pass protocol validation, then it is terminated.

SSLproxy uses only client requests for protocol validation. However, it also validates SMTP responses until it starts processing the packets from the client. If there is no excessive fragmentation, the first couple of packets in the connection should be enough for validating protocols.

Certificates

Certificate forging

For SSL and HTTPS connections, SSLproxy generates and signs forged X509v3 certificates on-the-fly, mimicking the original server certificate's subject DN, subjectAltName extension and other characteristics. SSLproxy has the ability to use existing certificates of which the private key is available, instead of generating forged ones. SSLproxy supports NULL-prefix CN certificates but otherwise does not implement exploits against specific certificate verification vulnerabilities in SSL/TLS stacks.

Certificate verification

SSLproxy verifies upstream certificates by default. If the verification fails, the connection is terminated immediately. This is in contrast to SSLsplit, because in order to maximize the chances that a connection can be successfully split, SSLsplit accepts all certificates by default, including self-signed ones. See The Risks of SSL Inspection for the reasons of this difference. You can disable this feature by the VerifyPeer option.

Client certificates

SSLproxy uses the certificate and key from the pemfiles configured by the ClientCert and ClientKey options when the destination requests client certificates. These options can be defined globally and/or per-proxyspec.

Alternatively, you can use Pass filtering rules to pass through certain destinations requesting client certificates.

User authentication

If the UserAuth option is enabled, SSLproxy requires network users to log in to the system to establish connections to the external network.

SSLproxy determines the user owner of a connection using a users table in an SQLite3 database configured by the UserDBPath option. The users table should be created using the following SQL statement:

CREATE TABLE USERS(
   IP             CHAR(45)     PRIMARY KEY     NOT NULL,
   USER           CHAR(31)     NOT NULL,
   ETHER          CHAR(17)     NOT NULL,
   ATIME          INT          NOT NULL,
   DESC           CHAR(50)
);

SSLproxy does not create this users table or the database file by itself, nor does it log users in or out. So the database file and the users table should already exist at the location pointed to by the UserDBPath option. An external program should log users in and out on the users table. The external program should fill out all the fields in user records, except perhaps for the DESC field, which can be left blank.

When SSLproxy accepts a connection,

• It searches the client IP address of the connection in the users table. If the client IP address is not in the users table, the connection is redirected to a login page configured by the UserAuthURL option.
• If SSLproxy finds a user record for the client IP address in the users table, it obtains the ethernet address of the client IP address from the arp cache of the system, and compares it with the value in the user record for that IP address. If the ethernet addresses do not match, the connection is redirected to the login page.
• If the ethernet addresses match, SSLproxy compares the atime value in the user record with the current system time. If the difference is greater than the value configured by the UserTimeout option, the connection is redirected to the login page.

If the connection passes all these checks, SSLproxy proceeds with establishing the connection.

The atime of the IP address in the users table is updated with the system time while the connection is being terminated. Since this atime update is executed using a privsep command, it is expensive. So, to reduce the frequency of such updates, it is deferred until after the user idle time is more than half of the timeout period.

If a description text is provided in the DESC field, it can be used with filtering rules to treat the user logged in from different locations, i.e. from different client IP addresses, separately.

If the UserAuth option is enabled, the user owner of the connection is appended at the end of the SSLproxy line, so that the listening program can parse and use this information in its logic and/or logging:

SSLproxy: [127.0.0.1]:34649,[192.168.3.24]:47286,[192.168.111.130]:443,s,soner

The user authentication feature is currently available on OpenBSD and Linux only.

Filtering rules

SSLproxy can divert, split, pass, block, or match connections based on filtering rules. Filtering rules can be defined globally or per-proxyspec.

• Divert action diverts packets to listening program, allowing SSL inspection by listening program and content logging of packets
• Split action splits the connection but does not divert packets to listening program, effectively disabling SSL inspection by listening program, but allowing content logging of packets
• Pass action passes the connection through by engaging passthrough mode, effectively disabling SSL inspection and content logging of packets
• Block action terminates the connection
• Match action specifies log actions for the connection without changing its filter action

The syntax of filtering rules is as follows:

(Divert|Split|Pass|Block|Match)
 ([from (
     user (username[*]|$macro|*) [desc (desc[*]|$macro|*)]|
     desc (desc[*]|$macro|*)|
     ip (clientip[*]|$macro|*)|
     *)]
  [to (
     sni (servername[*]|$macro|*)|
     cn (commonname[*]|$macro|*)|
     host (host[*]|$macro|*)|
     uri (uri[*]|$macro|*)|
     ip (serverip[*]|$macro|*) [port (serverport[*]|$macro|*)]|
     port (serverport[*]|$macro|*)|
     *)]
  [log ([[!]connect] [[!]master] [[!]cert]
        [[!]content] [[!]pcap] [[!]mirror] [$macro]|[!]*)]
  |*)

The definition of which connections the rule action will be applied to is achieved by the from and to parts of a filtering rule and by the proxyspec that the rule is defined for.

• The from part of a rule defines source filter based on client IP address, user and/or description, or * for all.
• The to part defines destination filter based on server IP address, SNI or Common Names of SSL connections, Host or URI fields in HTTP Request headers, or * for all.
  • Dst Host type of rules use ip and port site fields
  • SSL type of rules use sni and cn site fields
  • HTTP type of rules use host and uri site fields
• The proxyspec handling the connection defines the protocol filter for the connection.

If and how a connection should be logged is specified using the log part of filtering rules:

• connect enables logging connection information to connect log file
• master enables logging of master keys
• cert enables logging of generated certificates
• content enables logging packet contents to content log file
• pcap enables writing packets to pcap files
• mirror enables mirroring packets to mirror interfaces or targets

You can add a negation prefix ! to a log action to disable that logging.

For example, if the following rules are defined in a structured HTTPS proxyspec,

Split from user soner desc notebook to sni example.com log content
Pass from user soner desc android to cn .fbcdn.net*

The first filtering rule above splits but does not divert HTTPS connections from the user soner who has logged in with the description notebook to SSL sites with the SNI of example.com. Also, the rule specifies that the packet contents of the matching connection be written to content log file configured globally.

The second rule passes through HTTPS connections from the user soner who has logged in with the description android to SSL sites with the Common Names containing the substring .fbcdn.net anywhere in it (notice the asterisk at the end). Since connection contents cannot be written to log files in passthrough mode, the rule does not specify any content log action.

The default filter action is Divert. So, if those are the only filtering rules in that proxyspec, the other connections are diverted to the listening program specified in that proxyspec, without writing any logs.

If you want to enable, say, connect logging for the other connections handled by that proxyspec, without changing their default Divert filter action, you can add a third filtering rule to that proxyspec:

Match * log connect

Note that the second example above is a filtering rule you can use to resolve one of the certificate issues preventing the Facebook application on Android smartphones to connect to the Internet behind sslproxy.

Filtering rules are applied based on certain precedence orders:

• More specific rules have higher precedence.
• The precedence of filter types is as HTTP > SSL > Dst Host. Because, the application order of filter types is as Dst Host > SSL > HTTP, and a filter type can override the actions of a preceding filter type.
• The precedence of filter actions is as Divert > Split > Pass > Block. This is only for the same type of filtering rules.
• The precedence of site fields is as sni > cn for SSL filter and host > uri for HTTP filter.

For example, the pass action of a Dst Host filter rule is taken before the split action of an SSL filter rule with the same from definition, due to the precedence order of filter types. Or, the pass action of a rule with sni site field is taken before the split action of the same rule with cn site field, due to the precedence order of site fields.

In terms of possible filter actions,

• Dst Host filtering rules can take all of the filter and log actions.
• SSL filtering rules can take all of the filter and log actions.
• HTTP filtering rules can take match and block filter actions, can keep enabled divert and split modes, but cannot take pass action. Also, HTTP filtering rules can only disable logging.

Log actions do not configure any loggers. Global loggers for respective log actions should have been configured for those log actions to have any effect.

If no filtering rules are defined for a proxyspec, all log actions for that proxyspec are enabled. Otherwise, all log actions are disabled, and filtering rules should enable them specifically.

Macro expansion is supported. The Define option can be used for defining macros to be used in filtering rules. Macro names must start with a $ char. The macro name must be followed by words separated with spaces.

You can append an asterisk * to the fields in filtering rules for substring matching. Otherwise, the filter searches for an exact match with the field in the rule. The filter uses B-trees for exact string matching and Aho-Corasick machines for substring matching.

The ordering of filtering rules is important. The ordering of from, to, and log parts is not important. The ordering of log actions is not important.

If the UserAuth option is disabled, only client IP addresses can be used in the from part of filtering rules.

Excluding sites from SSL inspection

PassSite option is a special form of Pass filtering rule. PassSite rules can be written as Pass filtering rules. The PassSite option will be deprecated in favor of filtering rules in the future.

PassSite option allows certain SSL sites to be excluded from SSL inspection. If a PassSite matches the SNI or common names in the SSL certificate of a connection, that connection is passed through the proxy without being diverted to the listening program. SSLproxy engages the Passthrough mode for that purpose. For example, sites requiring client authentication can be added as PassSite.

Per-site filters can be defined using client IP addresses, users, and description. If the UserAuth option is disabled, only client IP addresses can be used in PassSite filters. Multiple sites can be defined, one on each line. PassSite rules can search for exact or substring matches. PassSite rules do not support macro expansion.

User control lists

User control lists can be implemented using filtering rules. The DivertUsers and PassUsers options will be deprecated in favor of filtering rules in the future.

DivertUsers and PassUsers options can be used to divert, pass through, or block users.

• If neither DivertUsers nor PassUsers is defined, all users are diverted to listening programs.
• Connections from users in DivertUsers, if defined, are diverted to listening programs.
• Connections from users in PassUsers, if defined, are simply passed through to their original destinations. SSLproxy engages the Passthrough mode for that purpose.
• If both DivertUsers and PassUsers are defined, users not listed in either of the lists are blocked. SSLproxy simply terminates their connections.
• If no DivertUsers list is defined, only users not listed in PassUsers are diverted to listening programs.

These user control lists can be defined globally or per-proxyspec. User control lists do not support macro expansion.

Logging

Logging options include traditional SSLproxy connect and content log files as well as PCAP files and mirroring decrypted traffic to a network interface. Additionally, certificates, master secrets and local process information can be logged. Filtering rules can selectively modify connection logging.

See the manual pages sslproxy(1) and sslproxy.conf(5) for details on using SSLproxy, setting up the various NAT engines, and for examples.

Requirements

SSLproxy depends on the OpenSSL, libevent 2.x, libpcap, libnet 1.1.x, and sqlite3 libraries by default. Libpcap and libnet are not needed if the mirroring feature is omitted. Sqlite3 is not needed if the user authentication feature is omitted. The build depends on GNU make and a POSIX.2 environment in PATH. If available, pkg-config is used to locate and configure the dependencies. The optional unit tests depend on the check library. The optional end-to-end tests depend on the TestProxy tool, and are supported only on Linux.

SSLproxy currently supports the following operating systems and NAT mechanisms:

• FreeBSD: pf rdr and divert-to, ipfw fwd, ipfilter rdr
• OpenBSD: pf rdr-to and divert-to
• Linux: netfilter REDIRECT and TPROXY
• Mac OS X: pf rdr and ipfw fwd

Support for local process information (-i) is currently available on Mac OS X and FreeBSD.

SSL/TLS features and compatibility greatly depend on the version of OpenSSL linked against. For optimal results, use a recent release of OpenSSL or LibreSSL.

Installation

With the requirements above available, run:

make
make test       # optional unit and e2e tests
make sudotest   # optional unit tests requiring privileges
make install    # optional install

Dependencies are autoconfigured using pkg-config. If dependencies are not picked up and fixing PKG_CONFIG_PATH does not help, you can specify their respective locations manually by setting OPENSSL_BASE, LIBEVENT_BASE, LIBPCAP_BASE, LIBNET_BASE, SQLITE_BASE and/or CHECK_BASE to the respective prefixes.

You can override the default install prefix (/usr/local) by setting PREFIX. For more build options and build-time defaults see GNUmakefile and defaults.h.

Documentation

See the manual pages sslproxy(1) and sslproxy.conf(5) for user documentation. See NEWS.md for release notes listing significant changes between releases and SECURITY.md for information on security vulnerability disclosure.

License

SSLproxy is provided under a 2-clause BSD license. SSLproxy contains components licensed under the MIT and APSL licenses. See LICENSE, LICENSE.contrib and LICENSE.third as well as the respective source file headers for details.

Credits

See AUTHORS.md for the list of contributors.

SSLproxy was inspired by and has been developed based on SSLsplit by Daniel Roethlisberger.