This is a purely informative rendering of an RFC that includes verified errata. This rendering may not be used as a reference.
The following 'Verified' errata have been incorporated in this document:
EID 800,
EID 803
Network Working Group P. Ford-Hutchinson
Request for Comments: 4217 IBM UK Ltd
Category: Standards Track October 2005
Securing FTP with TLS
Status of This Memo
This document specifies an Internet standards track protocol for the
Internet community, and requests discussion and suggestions for
improvements. Please refer to the current edition of the "Internet
Official Protocol Standards" (STD 1) for the standardization state
and status of this protocol. Distribution of this memo is unlimited.
Copyright Notice
Copyright (C) The Internet Society (2005).
Abstract
This document describes a mechanism that can be used by FTP clients
and servers to implement security and authentication using the TLS
protocol defined by RFC 2246, "The TLS Protocol Version 1.0.", and
the extensions to the FTP protocol defined by RFC 2228, "FTP Security
Extensions". It describes the subset of the extensions that are
required and the parameters to be used, discusses some of the policy
issues that clients and servers will need to take, considers some of
the implications of those policies, and discusses some expected
behaviours of implementations to allow interoperation. This document
is intended to provide TLS support for FTP in a similar way to that
provided for SMTP in RFC 2487, "SMTP Service Extension for Secure
SMTP over Transport Layer Security", and HTTP in RFC 2817, "Upgrading
to TLS Within HTTP/1.1.".
This specification is in accordance with RFC 959, "File Transfer
Protocol". It relies on RFC 2246, "The TLS Protocol Version 1.0.",
and RFC 2228, "FTP Security Extensions".
Table of Contents
1. Introduction ....................................................3
2. Audience ........................................................5
3. Overview ........................................................5
4. Session Negotiation on the Control Port .........................5
4.1. Client Wants a Secured Session .............................5
4.2. Server Wants a Secured Session .............................6
5. Clearing the Control Port .......................................6
6. Response to the FEAT Command ....................................7
7. Data Connection Behaviour .......................................8
8. Mechanisms for the AUTH Command .................................9
9. Data Connection Security ........................................9
10. A Discussion of Negotiation Behaviour .........................11
10.1. The Server's View of the Control Connection ..............11
10.2. The Server's View of the Data Connection .................12
10.3. The Client's View of the Control Connection ..............14
10.4. The Client's View of the Data Connection .................15
11. Who Negotiates What, Where, and How ...........................15
11.1. Do we protect at all? ....................................15
11.2. What level of protection do we use on the Control
connection? ..............................................15
11.3. Do we protect data connections in general? ...............16
11.4. Is protection required for a particular data transfer? ...16
11.5. What level of protection is required for a
particular data ..........................................16
12. Timing Diagrams ...............................................16
12.1. Establishing a Protected Session .........................17
12.2. Establishing a Protected Session Without a
Password Request .........................................18
12.3. Establishing a Protected Session and then
Clearing with the CCC ....................................19
12.4. A Standard Data Transfer Without Protection ..............20
12.5. A Firewall-Friendly Data Transfer Without Protection .....20
12.6. A Standard Data Transfer with Protection .................21
12.7. A Firewall-Friendly Data Transfer with Protection ........21
13. Discussion of the REIN Command ................................22
14. Discussion of the STAT and ABOR Commands ......................22
15. Security Considerations .......................................23
15.1. Verification of Authentication Tokens ....................23
15.1.1. Server Certificates ...............................23
15.1.2. Client Certificates ...............................23
15.2. Addressing FTP Security Considerations [RFC-2577] ........24
15.2.1. Bounce Attack .....................................24
15.2.2. Restricting Access ................................24
15.2.3. Protecting Passwords ..............................24
15.2.4. Privacy ...........................................24
15.2.5. Protecting Usernames ..............................24
15.2.6. Port Stealing .....................................25
15.2.7. Software-Based Security Problems ..................25
15.3. Issues with the CCC Command ..............................25
16. IANA Considerations ...........................................25
17. Other Parameters ..............................................25
18. Scalability and Limits ........................................26
19. Applicability .................................................26
20. Acknowledgements ..............................................26
21. References ....................................................26
21.1. Normative References .....................................26
21.2. Informative References ...................................27
1. Introduction
This document describes how three other documents should be combined
to provide a useful, interoperable, and secure file transfer
protocol. Those documents are:
RFC 959 [RFC-959]
The description of the Internet File Transfer Protocol.
RFC 2246 [RFC-2246]
The description of the Transport Layer Security protocol
(developed from the Netscape Secure Sockets Layer (SSL)
protocol version 3.0).
RFC 2228 [RFC-2228]
Extensions to the FTP protocol to allow negotiation of security
mechanisms to allow authentication, confidentiality, and
message integrity.
This document is intended to provide TLS support for FTP in a similar
way to that provided for SMTP in RFC 3207 [RFC-3207] and HTTP in RFC
2817 [RFC-2817].
The security extensions to FTP in [RFC-2228] offer a comprehensive
set of commands and responses that can be used to add authentication,
integrity, and confidentiality to the FTP protocol. The TLS protocol
is a popular (due to its wholesale adoption in the HTTP environment)
mechanism for generally securing a socket connection.
Although TLS is not the only mechanism for securing file transfer, it
does offer some of the following positive attributes:
- Flexible security levels. TLS can support confidentiality,
integrity, authentication, or some combination of all of these.
During a session, this allows clients and servers to dynamically
decide on the level of security required for a particular data
transfer.
- Ability to provide strong authentication of the FTP server.
- It is possible to use TLS identities to authenticate client
users and client hosts.
- Formalised public key management. By use of well established
client identity mechanisms (supported by TLS) during the
authentication phase, certificate management may be built into a
central function. Whilst this may not be desirable for all uses
of secured file transfer, it offers advantages in certain
structured environments.
- Co-existence and interoperation with authentication mechanisms
that are already in place for the HTTPS protocol. This allows
web browsers to incorporate secure file transfer using the same
infrastructure that has been set up to allow secure web
browsing.
The TLS protocol is a development of the Netscape Communication
Corporation's SSL protocol and this document can be used to allow the
FTP protocol to be used with either SSL or TLS. The actual protocol
used will be decided by the negotiation of the protected session by
the TLS/SSL layer. This document will only refer to the TLS
protocol; however, it is understood that the Client and Server MAY
actually be using SSL if they are so configured.
There are many ways in which these three protocols can be combined.
This document selects one method by which FTP can operate securely,
while providing both flexibility and interoperation. This
necessitates a brief description of the actual negotiation mechanism,
a detailed description of the required policies and practices, and a
discussion of the expected behaviours of clients and servers to allow
either party to impose their security requirements on the FTP
session.
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
"SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY" and "OPTIONAL" that
appear in this document are to be interpreted as described in
[RFC-2119].
2. Audience
This document is aimed at developers who wish to implement TLS as a
security mechanism to secure FTP clients and/or servers.
Systems administrators and architects should be fully aware of the
security implications discussed in [RFC-2228], which need to be
considered when choosing an implementation of this protocol and
configuring it to provide their required security.
3. Overview
A full description of the FTP security protocol enhancements is
contained in [RFC-2228]. This document describes how the AUTH, PROT,
PBSZ, and CCC commands, defined therein, should be implemented with
the TLS protocol.
In summary, an FTP session is established on the normal control port.
A client requests TLS with the AUTH command and then decides if it
wishes to secure the data connections by use of the PBSZ and PROT
commands. Should a client wish to make the control connection revert
back into plaintext (for example, once the authentication phase is
completed), then the CCC command can be used.
Implementation of this protocol extension does not ensure that each
and every session and data transfer is secure, it merely provides the
tools that allow a client and/or server to negotiate an acceptable or
required level of security for that given session or data transfer.
However, it is possible to have a server implementation that is
capable of refusing to operate in an insecure fashion.
4. Session Negotiation on the Control Port
The server listens on the normal FTP control port {FTP-PORT} and the
session initiation is not secured at all. Once the client wishes to
secure the session, the AUTH command is sent and the server MAY then
allow TLS negotiation to take place.
4.1. Client Wants a Secured Session
If a client wishes to attempt to secure a session, then it SHOULD, in
accordance with [RFC-2228], send the AUTH command with the parameter
requesting TLS {TLS-PARM} ('TLS').
The client then needs to behave according to its policies depending
on the response received from the server and also the result of the
TLS negotiation. A client that receives an AUTH rejection MAY choose
to continue with the session unprotected if it so desires.
4.2. Server Wants a Secured Session
The FTP protocol does not allow a server to directly dictate client
behaviour; however, the same effect can be achieved by refusing to
accept certain FTP commands until the session is secured to a level
that is acceptable to the server.
In either case, '234' is the server response to an 'AUTH TLS' command
that it will honour.
The '334' response, as defined in [RFC-2228], implies that an ADAT
exchange will follow. This document does not use the ADAT command
and so the '334' reply is incorrect.
The FTP protocol insists that a USER command be used to identify the
entity attempting to use the ftp server. Although the TLS
negotiation may be providing authentication information, the USER
command MUST still be issued by the client. However, it will be a
server implementation issue to decide which credentials to accept and
what consistency checks to make between the client cert used and the
parameter on the USER command.
[RFC-2228] states that the user must reauthorize (that is, reissue
some or all of the USER, PASS, and ACCT commands) following an AUTH
command. Additionally, this document specifies that all other
transfer parameters (other than the AUTH parameter) must be reset,
almost as if a REIN command was issued.
Reset transfer parameters after the AUTH command, including (but
are not limited to): user identity, default data ports, TYPE,
STRU, MODE, and current working directory.
5. Clearing the Control Port
There are circumstances in which it may be desirable to protect the
control connection only during part of the session and then to revert
back to a plaintext connection. This is often due to the limitations
of boundary devices such as NAT and firewalls, which expect to be
able to examine the content of the control connection in order to
modify their behaviour.
Typically the AUTH, USER, PASS, PBSZ, and PROT commands would be
protected within the TLS protocol and then the CCC command would be
issued to return to a plaintext socket state. This has important
Security Issues (which are discussed in the Security Considerations
section), but this document describes how the command should be used,
if the client and server still wish to use it after having considered
the issues.
When a server receives the CCC command, it should behave as follows:
If the server does not accept CCC commands (or does not understand
them), then a 500 reply should be sent.
Otherwise, if the control connection is not protected with TLS,
then a 533 reply should be sent.
Otherwise, if the server does not wish to allow the control
connection to be cleared at this time, then a 534 reply should be
sent.
Otherwise, the server is accepting the CCC command and should do
the following:
o Send a 200 reply.
o Shutdown the TLS session on the socket and leave it open.
o Continue the control connection in plaintext, expecting the
next command from the client to be in plaintext.
o Not accept any more PBSZ or PROT commands. All subsequent
data transfers must be protected with the current PROT
settings.
6. Response to the FEAT Command
The FEAT command (introduced in [RFC-2389]) allows servers with
additional features to advertise these to a client by responding to
the FEAT command. If a server supports the FEAT command, then it
MUST advertise supported AUTH, PBSZ, and PROT commands in the reply,
as described in section 3.2 of [RFC-2389]. Additionally, the AUTH
command should have a reply that identifies 'TLS' as one of the
possible parameters to AUTH. It is not necessary to identify the
'TLS-C' synonym separately.
Example reply (in the same style as [RFC-2389])
C> FEAT
S> 211-Extensions supported
S> AUTH TLS
S> PBSZ
S> PROT
S> 211 END
7. Data Connection Behaviour
The Data Connection in the FTP model can be used in one of three
ways. (Note: These descriptions are not necessarily placed in exact
chronological order, but do describe the steps required. See
diagrams later for clarification.)
i) Classic FTP client/server data exchange
- The client obtains a port; sends the port number to
the server; the server connects to the client. The
client issues a send or receive request to the server
on the control connection and the data transfer
commences on the data connection.
ii) Firewall-Friendly client/server data exchange (as
discussed in [RFC-1579]) using the PASV command to reverse
the direction of the data connection.
- The client requests that the server open a port; the
server obtains a port and returns the address and
port number to the client; the client connects to the
server on this port. The client issues a send or
receive request on the control connection, and the
data transfer commences on the data connection.
iii) Client-initiated server/server data exchange (proxy or
PASV connections).
- The client requests that server A opens a port;
server A obtains a port and returns it to the client;
the client sends this port number to server B.
Server B connects to server A. The client sends a
send or receive request to server A and the
complement to server B and the data transfer
commences. In this model, server A is the proxy or
PASV host and is a client for the Data Connection to
server B.
For i) and ii), the FTP client MUST be the TLS client and the FTP
server MUST be the TLS server.
That is to say, it does not matter which side initiates the
connection with a connect() call or which side reacts to the
connection via the accept() call; the FTP client, as defined in
[RFC-959], is always the TLS client, as defined in [RFC-2246].
In scenario iii), there is a problem in that neither server A nor
server B is the TLS client, given the fact that an FTP server must
act as a TLS server for Firewall-Friendly FTP [RFC-1579]. Thus, this
is explicitly excluded in the security extensions document [RFC-2228]
and in this document.
8. Mechanisms for the AUTH Command
The AUTH command takes a single parameter to define the security
mechanism to be negotiated. As the SSL/TLS protocols self-negotiate
their levels, there is no need to distinguish between SSL and TLS in
the application layer. The mechanism name for negotiating TLS is the
character string identified in {TLS-PARM}. This allows the client
and server to negotiate TLS on the control connection without
altering the protection of the data channel. To protect the data
channel as well, the PBSZ command, followed by the PROT command
sequence, MUST be used.
Note: The data connection state MAY be modified by the client issuing
the PROT command with the new desired level of data channel
protection and the server replying in the affirmative. This data
channel protection negotiation can happen at any point in the session
(even straight after a PORT or PASV command) and as often as is
required.
See also Section 16, "IANA Considerations".
9. Data Connection Security
The Data Connection security level is determined by the PROT command.
The PROT command, as specified in [RFC-2228], allows client/server
negotiation of the security level of the data connection. Once a
PROT command has been issued by the client and accepted by the
server returning the '200' reply, the security of subsequent data
connections MUST be at that level until another PROT command is
issued and accepted; the session ends and a REIN command is
issued, or the security of the session (via an AUTH command) is
re-negotiated.
Data Connection Security Negotiation (the PROT command)
Note: In line with [RFC-2228], there is no facility for securing
the Data connection with an insecure Control connection.
Specifically, the PROT command MUST be preceded by a PBSZ command,
and a PBSZ command MUST be preceded by a successful security data
exchange (the TLS negotiation in this case).
The command defined in [RFC-2228] to negotiate data connection
security is the PROT command. As defined, there are four values
that the PROT command parameter can take.
'C' - Clear - neither Integrity nor Privacy
'S' - Safe - Integrity without Privacy
'E' - Confidential - Privacy without Integrity
'P' - Private - Integrity and Privacy
As TLS negotiation encompasses (and exceeds) the Safe /
Confidential / Private distinction, only Private (use TLS) and
Clear (don't use TLS) are used.
For TLS, the data connection can have one of two security levels.
1) Clear (requested by 'PROT C')
2) Private (requested by 'PROT P')
With 'Clear' protection level, the data connection is made without
TLS. Thus, the connection is unauthenticated and has no
confidentiality or integrity. This might be the desired behaviour
for servers sending file lists, pre-encrypted data, or non-
sensitive data (e.g., for anonymous FTP servers).
If the data connection security level is 'Private', then a TLS
negotiation must take place on the data connection to the
satisfaction of the Client and Server prior to any data being
transmitted over the connection. The TLS layers of the Client and
Server will be responsible for negotiating the exact TLS Cipher
Suites that will be used (and thus the eventual security of the
connection).
In addition, the PBSZ (protection buffer size) command, as
detailed in [RFC-2228], is compulsory prior to any PROT command.
This document also defines a data channel encapsulation mechanism
for protected data buffers. For FTP-TLS, which appears to the FTP
application as a streaming protection mechanism, this is not
required. Thus, the PBSZ command MUST still be issued, but must
have a parameter of '0' to indicate that no buffering is taking
place and the data connection should not be encapsulated.
Note that PBSZ 0 is not in the grammar of [RFC-2228], section 8.1,
where it is stated:
PBSZ <sp> <decimal-integer> <CRLF> <decimal-integer> ::= any
decimal integer from 1 to (2^32)-1
However, it should be noted that using a value of '0' to mean a
streaming protocol is a reasonable use of '0' for that parameter
and is not ambiguous.
Initial Data Connection Security
The initial state of the data connection MUST be 'Clear' (this is
the behaviour as indicated by [RFC-2228]).
10. A Discussion of Negotiation Behaviour
As [RFC-2228] allows security qualities to be negotiated, enabled,
and disabled dynamically, this can make implementations seem quite
complex. However, in any given instance the behaviour should be
quite straightforward. Either the server will be enforcing the
policy of the server host or it will be providing security
capabilities requested by the client. Either the client will be
conforming to the server's policy or will be endeavouring to provide
the capabilities that the user desires.
10.1. The Server's View of the Control Connection
A server MAY have a policy statement somewhere that might:
- Deny any command before TLS is negotiated (this might cause
problems if a SITE or some such command is required prior to
login).
- Deny certain commands before TLS is negotiated (e.g., USER,
PASS, or ACCT).
- Deny insecure USER commands for certain users (e.g., not
ftp/anonymous).
- Deny secure USER commands for certain users (e.g.,
ftp/anonymous).
- Define the level(s) of TLS to be allowed.
- Define the CipherSuites allowed to be used (perhaps on a per
host/domain/... basis).
- Allow TLS authentication as a substitute for local
authentication.
- Define data connection policies (see next section).
It is possible that the TLS negotiation may not be completed
satisfactorily for the server, in which case it can be one of
these states.
The TLS negotiation failed completely
In this case, the control connection should still be in an
unprotected mode and the server SHOULD issue an unprotected
'421' reply to end the session.
The TLS negotiation completed successfully, but the server
decides that the session parameters are not acceptable (e.g.,
Distinguished Name in the client certificate is not permitted
to use the server).
In this case, the control connection should still be in a
protected state, so the server MAY either continue to refuse
to service commands or issue a protected '421' reply and
close the connection.
The TLS negotiation failed during the TLS handshake
In this case, the control connection is in an unknown state
and the server SHOULD simply drop the control connection.
The server code will be responsible for implementing the required
policies and ensuring that the client is prevented from circumventing
the chosen security by refusing to service those commands that are
against policy.
10.2. The Server's View of the Data Connection
The server can take one of four basic views of the data connection.
1 - Don't allow encryption at all (in which case the PROT command
should not allow any value other than 'C' - if it is allowed
at all).
2 - Allow the client to choose protection or not.
3 - Insist on data protection (in which case the PROT command must
be issued prior to the first attempted data transfer).
4 - Decide on one of the above three for each and every data
connection.
The server SHOULD only check the status of the data protection level
(for options 3 and 4 above) on the actual command that will initiate
the data transfer (and not on the PORT or PASV). The following
commands, defined in [RFC-959], cause data connections to be opened
and thus may be rejected before any 1xx message due to an incorrect
PROT setting.
STOR
RETR
NLST
LIST
STOU
APPE
The reply to indicate that the PROT setting is incorrect is '521 data
connection cannot be opened with this PROT setting'
If the protection level indicates that TLS is required, then it
should be negotiated once the data connection is made. Thus, the
'150' reply only states that the command can be used given the
current PROT level. Should the server not like the TLS negotiation,
then it will close the data port immediately and follow the '150'
command with a '522' reply, which indicates that the TLS negotiation
failed or was unacceptable. (Note: This means that the application
can pass a standard list of CipherSuites to the TLS layer for
negotiation, and review the one negotiated for applicability in each
instance).
The Security Considerations section discusses the issue of cross-
checking any certificates used to authenticate the data connection
with the one(s) used to authenticate the control connection. This is
an important security step.
It is reasonable for the server to insist that the data connection
uses a TLS cached session. This might be a cache of a previous data
connection or of a cleared control connection. If this is the reason
for the refusal to allow the data transfer, then the '522' reply
should indicate this.
Note: This has an important impact on client design, but allows
servers to minimise the cycles used during TLS negotiation by
refusing to perform a full negotiation with a previously
authenticated client.
It should be noted that the TLS authentication of the server will be
authentication of the server host itself and not a user on the server
host.
10.3. The Client's View of the Control Connection
In most cases, it is likely that the client will be using TLS because
the server would refuse to interact insecurely. To allow for this,
clients SHOULD be flexible enough to manage the securing of a session
at the appropriate time and still allow the user/server policies to
dictate exactly when during the session the security is negotiated.
In the case where it is the client that is insisting on the securing
of the session, the client will need to ensure that the negotiations
are all completed satisfactorily and will need to be able to sensibly
inform the user should the server not support, or not be prepared to
use, the required security levels.
Clients SHOULD be coded in such a manner as to allow the timing of
the AUTH, PBSZ, and PROT commands to be flexible and dictated by the
server. It is quite reasonable for a server to refuse certain
commands prior to these commands. Similarly, it is quite possible
that a SITE or quoted command might be needed by a server prior to
the AUTH. A client MUST allow a user to override the timing of these
commands to suit a specific server.
For example, a client SHOULD NOT insist on sending the AUTH as the
first command in a session, nor should it insist on issuing a
PBSZ/PROT pair directly after the AUTH. This may well be the default
behaviour, but must be overridable by a user.
The TLS negotiation may not be completed satisfactorily for the
client, in which case it will be in one of these states:
The TLS negotiation failed completely
In this case, the control connection should still be in an
unprotected mode and the client should issue an unprotected
QUIT command to end the session.
The TLS negotiation completed successfully, but the client decides
that the session parameters are not acceptable (e.g.,
Distinguished Name in certificate is not the actual server
expected).
In this case, the control connection should still be up in a
protected state, so the client should issue a protected QUIT
command to end the session.
The TLS negotiation failed during the TLS handshake.
In this case, the control connection is in an unknown state and
the client should simply drop the control connection.
10.4. The Client's View of the Data Connection
Client security policies
Clients do not typically have 'policies' as such, instead they
rely on the user to define their actions and, to a certain extent,
are reactive to the server policy. Thus, a client will need to
have commands that will allow the user to switch the protection
level of the data connection dynamically; however, there may be a
general 'policy' that attempts all LIST and NLST commands on a
Clear connection first (and automatically switches to Private if
it fails). In this case, there would need to be a user command
available to ensure that a given data transfer was not attempted
on an insecure data connection.
Clients also need to understand that the level of the PROT setting
is only checked for a particular data transfer after that transfer
has been requested. Thus, a refusal by the server to accept a
particular data transfer should not be read by the client as a
refusal to accept that data protection level completely, as not
only may other data transfers be acceptable at that protection
level, but it is entirely possible that the same transfer may be
accepted at the same protection level at a later point in the
session.
It should be noted that the TLS authentication of the client
should be an authentication of a user on the client host and not
the client host itself.
11. Who Negotiates What, Where, and How
11.1. Do we protect at all?
Client issues 'AUTH TLS', server accepts or rejects. If the server
needs AUTH, then it refuses to accept certain commands until it gets
a successfully protected session.
11.2. What level of protection do we use on the Control connection?
Decided entirely by the TLS CipherSuite negotiation.
11.3. Do we protect data connections in general?
Client issues PROT command, server accepts or rejects.
11.4. Is protection required for a particular data transfer?
A client would have already issued a PROT command if it required the
connection to be protected.
If a server needs to have the connection protected, then it will
reply to the STOR/RETR/NLST/... command with a '522', indicating that
the current state of the data connection protection level is not
sufficient for that data transfer at that time.
11.5. What level of protection is required for a particular data
transfer?
Decided entirely by the TLS CipherSuite negotiation.
Thus, for flexibility, it can be seen that it is desirable for the
FTP application to be able to interact with the TLS layer upon which
it sits to define and discover the exact TLS CipherSuites that are to
be/have been negotiated and to make decisions accordingly.
12. Timing Diagrams
These timing diagrams aim to help explain exactly how the TLS
handshake and session protection fits into the existing logic of the
FTP protocol. Of course, the FTP protocol itself is not well
described with respect to the timing of commands and responses in
[RFC-959], so this is partly based on empirical observation of
existing widespread client and server implementations.
12.1. Establishing a Protected Session
Client Server
control data data control
====================================================================
socket()
bind()
socket()
connect() ----------------------------------------------> accept()
<---------------------------------------------- 220
AUTH TLS ---------------------------------------------->
<---------------------------------------------- 234
TLSneg() <----------------------------------------------> TLSneg()
PBSZ 0 ---------------------------------------------->
<---------------------------------------------- 200
PROT P ---------------------------------------------->
<---------------------------------------------- 200
USER fred ---------------------------------------------->
<---------------------------------------------- 331
PASS pass ---------------------------------------------->
<---------------------------------------------- 230
Note 1: The order of the PBSZ/PROT pair and the USER/PASS pair (with
respect to each other) is not important (i.e., the USER/PASS can
happen prior to the PBSZ/PROT, or the server can refuse to allow a
PBSZ/PROT pair until the USER/PASS pair has happened).
Note 2: The PASS command might not be required at all (if the USER
parameter and any client identity presented provide sufficient
authentication). The server would indicate this by issuing a '232'
reply to the USER command instead of the '331', which requests a PASS
from the client (see below).
Note 3: The AUTH command might not be the first command after the
receipt of the 220 welcome message.
12.2. Establishing a Protected Session Without a Password Request
(The TLS Authentication is Sufficient)
Client Server
control data data control
====================================================================
socket()
bind()
socket()
connect() ----------------------------------------------> accept()
<---------------------------------------------- 220
AUTH TLS ---------------------------------------------->
<---------------------------------------------- 234
TLSneg() <----------------------------------------------> TLSneg()
PBSZ 0 ---------------------------------------------->
<---------------------------------------------- 200
PROT P ---------------------------------------------->
<---------------------------------------------- 200
USER fred ---------------------------------------------->
<---------------------------------------------- 232
12.3. Establishing a Protected Session and then Clearing with the CCC
Command
Client Server
control data data control
====================================================================
socket()
bind()
socket()
connect() ----------------------------------------------> accept()
<---------------------------------------------- 220
AUTH TLS ---------------------------------------------->
<---------------------------------------------- 234
TLSneg() <----------------------------------------------> TLSneg()
PBSZ 0 ---------------------------------------------->
<---------------------------------------------- 200
PROT P ---------------------------------------------->
<---------------------------------------------- 200
USER fred ---------------------------------------------->
<---------------------------------------------- 232
CCC ---------------------------------------------->
<---------------------------------------------- 200
TLSshutdown() <-------------------------------------> TLSshutdown()
- The rest of the control session continues in plaintext with
protected data transfers (due to PROT P).
Note: This has serious security issues (see Security Considerations
section) but may be useful in a firewall/NAT scenario.
12.4. A Standard Data Transfer Without Protection
Client Server
control data data control
====================================================================
socket()
bind()
PORT w,x,y,z,a,b ----------------------------------------->
<----------------------------------------------------- 200
STOR file ------------------------------------------------>
socket()
bind()
<----------------------------------------------------- 150
accept() <----------- connect()
write() -----------> read()
close() -----------> close()
<----------------------------------------------------- 226
12.5. A Firewall-Friendly Data Transfer Without Protection
EID 800 (Verified) is as follows:
Section: 12.5
Original Text:
"... This contrasts the with situation when ..."
Corrected Text:
"... This contrasts with the situation when ..."
Notes:
from pending
EID 803 (Verified) is as follows:
Section: 12.5
Original Text:
"o The various people who have help author this document ..."
Corrected Text:
"o The various people who have helped [to] author this document ..."
or:
"o The various people who have helped the author of this document ..."
Notes:
from pending
Client Server
control data data control
====================================================================
PASV -------------------------------------------------------->
socket()
bind()
<------------------------------------------ 227 (w,x,y,z,a,b)
socket()
STOR file --------------------------------------------------->
connect() ----------> accept()
<-------------------------------------------------------- 150
write() ----------> read()
close() ----------> close()
<-------------------------------------------------------- 226
Note: Implementers should be aware that the connect()/accept()
function is performed prior to the receipt of the reply from the STOR
command. This contrasts the with situation when a non-firewall-
friendly PORT is used prior to the STOR, and the accept()/connect()
is performed after the reply from the aforementioned STOR has been
dealt with.
12.6. A Standard Data Transfer with Protection
Client Server
control data data control
====================================================================
socket()
bind()
PORT w,x,y,z,a,b -------------------------------------------->
<-------------------------------------------------------- 200
STOR file --------------------------------------------------->
socket()
bind()
<-------------------------------------------------------- 150
accept() <---------- connect()
TLSneg() <----------> TLSneg()
TLSwrite() ----------> TLSread()
TLSshutdown() -------> TLSshutdown()
close() ----------> close()
<-------------------------------------------------------- 226
12.7. A Firewall-Friendly Data Transfer with Protection
Client Server
control data data control
====================================================================
PASV -------------------------------------------------------->
socket()
bind()
<------------------------------------------ 227 (w,x,y,z,a,b)
socket()
STOR file --------------------------------------------------->
connect() ----------> accept()
<-------------------------------------------------------- 150
TLSneg() <---------> TLSneg()
TLSwrite() ---------> TLSread()
TLSshutdown() -------> TLSshutdown()
close() ---------> close()
<-------------------------------------------------------- 226
13. Discussion of the REIN Command
The REIN command, defined in [RFC-959], allows the user to reset the
state of the FTP session. From [RFC-959]:
REINITIALIZE (REIN)
This command terminates a USER, flushing all I/O and account
information, except to allow any transfer in progress to be
completed. All parameters are reset to the default settings
and the control connection is left open. This is identical to
the state in which a user finds himself immediately after the
control connection is opened. A USER command may be expected
to follow.
When this command is processed by the server, the TLS session(s) MUST
be cleared and the control and data connections revert to
unprotected, clear communications. It MAY be acceptable to use
cached TLS sessions for subsequent connections, however, a server
MUST NOT mandate this.
If the REIN command is being used to clear a TLS session, then the
reply to the REIN command MUST be sent in a protected session prior
to the session(s) being cleared.
14. Discussion of the STAT and ABOR Commands
The ABOR and STAT commands and the use of TCP Urgent Pointers
[RFC-959] describes the use of Telnet commands (IP and DM) and the
TCP Urgent pointer to indicate the transmission of commands on the
control channel during the execution of a data transfer. FTP uses
the Telnet Interrupt Process and Data Mark commands in conjunction
with Urgent data to preface two commands: ABOR (Abort Transfer)
and STAT (Status request).
The Urgent Pointer was used because, in a Unix implementation, the
receipt of a TCP packet marked as Urgent would result in the
execution of the SIGURG interrupt handler. This reliance on
interrupt handlers was necessary on systems that did not implement
select() or did not support multiple threads. TLS does not
support the notion of Urgent data.
When TLS is implemented as a security method in FTP, the server
SHOULD NOT rely on the use of SIGURG to process input on the
control channel during data transfers. The client MUST send all
data, including Telnet commands, across the TLS session.
15. Security Considerations
This document discusses how TLS may be used in conjunction with
[RFC-2228] to provide mechanisms for securing FTP sessions.
Discussions about security rationale and security properties are
contained within the [RFC-2228] document and are not repeated here.
15.1. Verification of Authentication Tokens
In this section, we assume that X.509 certificates will be used for
the TLS authentication. If some other identity token is used (e.g.,
kerberos tickets - see [RFC-2712]), then similar, mechanism-specific
considerations will need to be made.
15.1.1. Server Certificates
- Although it is entirely an implementation decision, it is
recommended that certificates used for server authentication of the
TLS session contain the server identification information in a
similar manner to those used for http servers (see [RFC-2818]).
- It is strongly recommended that the certificate used for server
authentication of Data connections be the same certificate as that
used for the corresponding Control connection. If different
certificates are to be used, there should be some other mechanism
that the client can use to cross-check the data and control
connection server identities.
- If Server Certificates are not used, then many of the security
benefits will not be realised. For Example, in an anonymous
Diffie-Hellman environment, there is no server identity
authentication, so there is little protection against man-in-the-
middle attacks.
15.1.2. Client Certificates
- Deciding which client certificates to allow and defining which
fields define what authentication information is entirely a server
implementation issue.
- However, it is strongly recommended that the certificate used for
client authentication of Data connections be the same certificate
as that used for the corresponding Control connection. If
different certificates are to be used, there should be some other
mechanism that the server can use to cross-check the data and
control connection client identities.
- If Client Certificates are not used, then many of the security
benefits will not be realised. For Example, it would still be
possible for a malicious client to hijack a data connection.
15.2. Addressing FTP Security Considerations [RFC-2577]
15.2.1. Bounce Attack
A bounce attack should be harder in a secured FTP environment
because:
- The FTP server that is being used to initiate a false connection
will always be a 'server' in the TLS context. Therefore, only
services that act as 'clients' in the TLS context could be
vulnerable. This would be a counter-intuitive way to implement
TLS on a service.
- The FTP server would detect that the authentication credentials
for the data connection are not the same as those for the
control connection, thus the server policies could be set to
drop the data connection.
- Genuine users are less likely to initiate such attacks when the
authentication is strong, and malicious users are less likely to
gain access to the FTP server if the authentication is not
easily subverted (password guessing, network tracing, etc...)
15.2.2. Restricting Access
This document presents a strong mechanism for solving the issue
raised in this section.
15.2.3. Protecting Passwords
The twin solutions of strong authentication and data confidentiality
ensure that this is not an issue when TLS is used to protect the
control session.
15.2.4. Privacy
The TLS protocol ensures data confidentiality by encryption. Privacy
(e.g., access to download logs, user profile information, etc...) is
outside the scope of this document (and [RFC-2577] presumably).
15.2.5. Protecting Usernames
This is not an issue when TLS is used as the primary authentication
mechanism.
15.2.6. Port Stealing
This specification will do little for the Denial of Service element
of this section; however, strong authentication on the data
connection will prevent unauthorised connections from retrieving or
submitting files. Of course, this is only the case where strong
client authentication is being used. If client certificates are not
used, then port stealing by a rogue client is still a problem. If no
strong authentication is in use at all (e.g., anonymous Diffie-
Hellman), then the port stealing problem will remain.
15.2.7. Software-Based Security Problems
Nothing in this specification will affect the discussion in this
section.
15.3. Issues with the CCC Command
Using the CCC command can create security issues. For a full
description, see the "CLEAR COMMAND CHANNEL (CCC)" section of
[RFC-2228]. Clients should not assume that a server will allow the
CCC command to be processed.
Server implementations may wish to refuse to process the CCC command
on a session that has not passed through some form of client
authentication (e.g., TLS client auth or FTP USER/PASS). This can
prevent anonymous clients from repeatedly requesting AUTH TLS
followed by CCC to tie up resources on the server.
16. IANA Considerations
{FTP-PORT} - The port assigned to the FTP control connection is 21.
17. Other Parameters
{TLS-PARM} - The parameter for the AUTH command to indicate that TLS
is required. To request the TLS protocol in accordance with this
document, the client MUST use 'TLS'
To maintain backward compatibility with older versions of this
document, the server SHOULD accept 'TLS-C' as a synonym for 'TLS'.
Note: [RFC-2228] states that these parameters are case-
insensitive.
18. Scalability and Limits
There are no issues other than those concerned with the ability of
the server to refuse to have a complete TLS negotiation for each and
every data connection, which will allow servers to retain throughput
whilst using cycles only when necessary.
19. Applicability
This mechanism is generally applicable as a mechanism for securing
the FTP protocol. It is unlikely that anonymous FTP clients or
servers will require such security (although some might like the
authentication features without the confidentiality).
20. Acknowledgements
o Netscape Communications Corporation for the original SSL protocol.
o Eric Young for the SSLeay libraries.
o University of California, Berkeley for the original
implementations of FTP and ftpd, on which the initial
implementation of these extensions were layered.
o IETF CAT working group.
o IETF TLS working group.
o IETF FTPEXT working group.
o Jeff Altman for the ABOR and STAT discussion.
o The various people who have help author this document throughout
its protracted draft stages, namely Martin Carpenter, Eric Murray,
Tim Hudson, and Volker Wiegand.
21. References
21.1. Normative References
[RFC-959] Postel, J. and J. Reynolds, "File Transfer Protocol", STD
9, RFC 959, October 1985.
[RFC-2119] Bradner, S., "Key words for use in RFCs to Indicate
Requirement Levels", BCP 14, RFC 2119, March 1997.
[RFC-2228] Horowitz, M. and S. Lunt, "FTP Security Extensions", RFC
2228, October 1997.
[RFC-2246] Dierks, T. and C. Allen, "The TLS Protocol Version 1.0",
RFC 2246, January 1999.
[RFC-2389] Hethmon, P. and R. Elz, "Feature negotiation mechanism for
the File Transfer Protocol", RFC 2389, August 1998.
21.2. Informative References
[RFC-1579] Bellovin, S., "Firewall-Friendly FTP", RFC 1579, February
1994.
[RFC-2222] Myers, J., "Simple Authentication and Security Layer
(SASL)", RFC 2222, October 1997.
[RFC-2577] Allman, M. and S. Ostermann, "FTP Security
Considerations", RFC 2577, May 1999.
[RFC-2712] Medvinsky, A. and M. Hur, "Addition of Kerberos Cipher
Suites to Transport Layer Security (TLS)", RFC 2712,
October 1999.
[RFC-2817] Khare, R. and S. Lawrence, "Upgrading to TLS Within
HTTP/1.1", RFC 2817, May 2000.
[RFC-2818] Rescorla, E., "HTTP Over TLS", RFC 2818, May 2000.
[RFC-3207] Hoffman, P., "SMTP Service Extension for Secure SMTP over
Transport Layer Security", RFC 3207, February 2002.
Contributors
Tim Hudson
RSA Data Security
Australia Pty Ltd
Phone: +61 7 3227 4444
EMail: tjh@rsasecurity.com.au
Volker Wiegand
SuSE Linux
EMail: wiegand@suse.de
Martin Carpenter
Verisign Ltd
EMail: mcarpenter@verisign.com
Eric Murray
Wave Systems Inc.
EMail: ericm@lne.com
Author's Address
Paul Ford-Hutchinson
IBM UK Ltd
PO Box 31
Birmingham Road
Warwick
United Kingdom
Phone: +44 1926 462005
EMail: rfc4217@ford-hutchinson.com
Full Copyright Statement
Copyright (C) The Internet Society (2005).
This document is subject to the rights, licenses and restrictions
contained in BCP 78, and except as set forth therein, the authors
retain all their rights.
This document and the information contained herein are provided on an
"AS IS" basis and THE CONTRIBUTOR, THE ORGANIZATION HE/SHE REPRESENTS
OR IS SPONSORED BY (IF ANY), THE INTERNET SOCIETY AND THE INTERNET
ENGINEERING TASK FORCE DISCLAIM ALL WARRANTIES, EXPRESS OR IMPLIED,
INCLUDING BUT NOT LIMITED TO ANY WARRANTY THAT THE USE OF THE
INFORMATION HEREIN WILL NOT INFRINGE ANY RIGHTS OR ANY IMPLIED
WARRANTIES OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
Intellectual Property
The IETF takes no position regarding the validity or scope of any
Intellectual Property Rights or other rights that might be claimed to
pertain to the implementation or use of the technology described in
this document or the extent to which any license under such rights
might or might not be available; nor does it represent that it has
made any independent effort to identify any such rights. Information
on the procedures with respect to rights in RFC documents can be
found in BCP 78 and BCP 79.
Copies of IPR disclosures made to the IETF Secretariat and any
assurances of licenses to be made available, or the result of an
attempt made to obtain a general license or permission for the use of
such proprietary rights by implementers or users of this
specification can be obtained from the IETF on-line IPR repository at
http://www.ietf.org/ipr.
The IETF invites any interested party to bring to its attention any
copyrights, patents or patent applications, or other proprietary
rights that may cover technology that may be required to implement
this standard. Please address the information to the IETF at ietf-
ipr@ietf.org.
Acknowledgement
Funding for the RFC Editor function is currently provided by the
Internet Society.