Hi, Joe.
Thanks very much for the feedback. You make some good points, and I am working
on a new draft where I have incorporated a lot of the semantic changes that you
recommended as they will make some areas of the draft a little clearer. As far
as changing the title of the draft is concern, I am not opposed to changing the
title, but I tried to follow the examples from other RFCs. Looking at a couple
of recent RFCs like 5789 and 5689 perhaps you might prefer "HOST Command for
File Transfer Protocol."
I have a few key points that I would like to discuss with regard to your
analysis of the draft, although I would like to mention in advance that I
believe that some areas of the draft and your subsequent analysis might be
saying the same thing and we only differ semantically. If so, I might need to
clarify something in order to remove any ambiguity between what I wrote and
what you understood. <grin>
--------------------
1. Identifying the use of the HOST command with regard to which type of host
names to which the draft is referring.
--------------------
Just to make sure that we're on the same page, the use of the HOST command will
be the most advantageous to servers that want to support multiple FQDN-based
FTP sites on a single server. For example, the FTP sites ftp.foo.com and
ftp.bar.com are both physically located on a server at 192.168.1.1 and those
two names are mapped via DNS to that IP address, or the FTP sites
foo.example.com and bar.example.com are both physically located on a server at
192.168.2.1 and once again those two names are mapped via DNS to that IP
address. As you alluded to in your email, HTTP accommodates this through the
use of the Host header.
It is the intention for the HOST command to work in much the same way, in that
a typical scenario will involve a user-FTP process making a connection to an
FTP server using an IP address, and then send a HOST command with the name of
the host to connect to that the server-FTP process will process accordingly. In
actual practice this will most often resemble a person entering an FQDN into an
FTP client software application and that application will look up the IP
address to physically connect to the FTP server. Once the connection is
established using the IP address, the FTP client application will then send the
HOST command with the FQDN for the desired FTP site, which the FTP server will
use to route the connection according to its design.
So with that in mind, the parameter for the HOST command is supposed to be
indicating a hostname; this will typically be an FQDN, but that's not a
requirement. For example, Windows networks can use short host names for network
names, but everything in the above scenario would still hold true for a Windows
network where the virtual hosts "foobar1" and "foobar2" are both physically
located on the same FTP server.
But in any case, my expectation was that the parameter for the HOST command
should be mapped to the IP address. As your email indicated, these are multiple
virtual hosts on a single physical host. If I have not made that clear then
please let me know.
--------------------
2. Overloading the USER command to differentiate hosts.
--------------------
This was an early consideration of ours, and I have seen that there are some
limited FTP server implementations that resemble something similar to that
already. As just one example, some versions of the Microsoft FTP server allow
for USER syntax like "USER hostname|username" or "USER hostname\username" where
the hostname is used to internally route the FTP request to the correct FTP
site on the server. There are several reasons why this suggestion is
impractical outside of a limited-scope server implementation, but one of my
arguments against overloading the USER command stems from a series of
discussions that were held several years ago where it was agreed that syntax
like "username(_at_)hostname" is unfortunately already overloaded in a variety
of scenarios and therefore simply unworkable in a practical sense.
For example, in network environments where "username(_at_)hostname is already
in use as a network login where the hostname portion refers to the location of
the user's credentials within the network hierarchy, that would necessitate
having to determine how to implement differentiation between FTP hosts and
network credential stores, and you could easily wind up with syntax like "USER
username(_at_)credentialstore@ftphostname" or "USER
username(_at_)ftphostname@credentialstore", or you could wind up with a
scenario where an FTP service won't know the difference between "USER
username(_at_)credentialstore" and "USER username(_at_)ftphostname".
--------------------
3. Passing the parameter for the HOST command as an opaque string.
--------------------
I disagree with the suggestion that the parameter for the HOST command should
be an opaque string rather than constraining the parameter through the
protocol. The intention of this draft is to allow an FTP client to specify the
correct hostname (typically an FQDN) for an FTP server to use after an FTP
client has connected, so that limits the parameter for the HOST command to
hostnames that are already constrained by protocol, although I tried to make
allowances for an FTP client to send the IP address via the HOST command rather
than force the client to know the difference between an FQDN and an IP address.
(An actual human using an FTP client should be able to tell the difference, of
course, but their FTP software application might not. <grin>)
In any event, I am still working on an updated draft. Thanks again for the
feedback.
Robert
-----Original Message-----
From: Joe Touch [mailto:touch(_at_)ISI(_dot_)EDU]
Sent: Thursday, May 06, 2010 10:48 AM
To: IETF discussion list
Cc: TSV Dir; draft-hethmon-mcmurray-ftp-hosts(_at_)tools(_dot_)ietf(_dot_)org;
tsv-ads(_at_)tools(_dot_)ietf(_dot_)org
Subject: TSV-DIR review of draft-hethmon-mcmurray-ftp-hosts-11.txt
Hi, all,
I've reviewed this document as part of the transport area directorate's ongoing
effort to review key IETF documents. These comments were written primarily for
the transport area directors, but are copied to the document's authors for
their information and to allow them to address any issues raised. The authors
should consider this review together with any other last-call comments they
receive. Please always CC tsv-dir(_at_)ietf(_dot_)org if you reply to or
forward this review.
The document defines a new FTP command called "HOST" that allows an FTP server
to support virtual hosts, i.e., multiple hosts on the same IP address.
The document does raise an important concern from a transport viewpoint.
The document extends FTP to support the same kind of virtual hosting common in
the web. Although the Internet defines hosts by IP address, virtual hosts are
specified in various ways, including DNS names, that may or may not map to the
IP address of the connection to the server.
The extension is discussed in ways that imply that the user is connecting to a
HOST as indicated in the command, rather than the IP address of the FTP
connection. IMO, the HOST command ought to be presented more as a way to
provide context within a host, not to indicate a host. That might be acheived
by a name change (VHOST?) and/or by revising some of the discussion accordingly.
Thus, I would strongly suggest the document be revised as follows. Notes are
included below that indicate places where thes critical issues should be
addressed, flagged by "####".
--
This document should emphasize that the argument to the host command is, at the
protocol level, an opaque string passed to the application that indicates only
a session context (including authentication, file system, etc.). There should
be no attempt for the protocol to validate, translate, or interpret that string
- e.g., as matching the IP address of the connection, or providing a valid IP
address or fqdn. That validation can be performed by the server implementation,
but should not be a constraint of the protocol.
There are places where SHOULDs are used where I expected MUSTs. Any place a
SHOULD is used, the document ought to include a description of what happens if
not, i.e., if the SHOULD is not followed, and it'd be useful to also include
reasons why and conditions where this might be useful or valid.
Also, in terminology, the document would benefit from a bit of clarification
and revision, notably its title and description of the purpose of this command.
The HOST command performes much the same function as HTTP including the full
URL in the GET command, which similarly allows virtual hosting. The document
should avoid confusing this with multi-homing, which is not necessarily related.
--
I have provided some other feedback interspersed in the text below (flagged
with "++++"). The most notable is the issue of how to handle multiple HOST
commands, the potential for HOST commands after the USER command (without a
REIN command), and updating the state machine to include the effects of REIN
interactions with the HOST command.
I also found numerous places where "should" or "must" are used but not
capitalized; the document should be checked for these uses to confirm the
intent. If case is important, then Section 2 should include text to indicate
such (as noted there).
I'd like to also suggest that the alternatives discussed might consider an
opportunity to integrate this command with the USER command to accept
"name[(_at_)hoststring]" arguments, e.g.,
smith(_at_)host(_dot_)example(_dot_)com This syntax would suffice, and might
need only an additional error code (to indicate that the hoststring is not
valid, rather than the entire command argument), rather than defining a new
name and introducing an additional round trip into the login process. It seems
equally compatible with the wrapper approach discussed (which, IMO, should also
be noted as a goal of the design, FWIW).
I'd be glad to discuss this further on whatever list would be useful.
Joe
---------------------
Network Working Group P. Hethmon
Internet-Draft Hethmon Brothers
Updates: 959 R. McMurray
Intended status: Standards Track Microsoft
Expires: October 6, 2010 April 2010
File Transfer Protocol HOST Command
draft-hethmon-mcmurray-ftp-hosts-11
####
The title should more accurately indicate the nature of the command, e.g., "The
File Transfer Protocol HOST Command for Virtual Hosts"
####
Abstract
The File Transfer Protocol, as defined in RFC 959 and Section 4
of RFC 1123, is one of the oldest and most widely used protocols on
the Internet.
This document addresses the subject of creating multi-homed hostname-
based FTP servers on a single IP address. This is achieved by
extending the FTP specification to add a HOST command that is used
to specify individual FTP hosts.
####
Multihoming is when a single host has more than one IP address This extension
supports 'virtual hosts', where more than one fully-qualified DNS name (fqdn)
resolves to the same IP address.
The terminology needs to be updated throughout to address this issue.
####
1. Introduction
It is common on the Internet for many domain names to be allocated
++++
It is common in the Internet to allow multiple DNS host names to resolve to the
same IP address.
(they're not allocated, and they're not domain names)
++++
to a single IP address. This practice has introduced the concept of
a "virtual host", where a host appears to exist as an independent
entity, but in reality shares all of its resources with one or more
++++
physical resources
++++
similar hosts.
Such an arrangement presents some problems for FTP servers, as all
an FTP server can detect is an incoming FTP connection to a
particular IP address. That is, all domain names that share an IP
####
an FTP server distinguishes incoming FTP connections by their IP address, not
their DNS names, because hosts are uniquely identified by address rather than
name. That is, all DNS names that share an IP...
####
address also share the same FTP server, and more importantly, its
++++
are handled by the same FTP server
++++
Hethmon & McMurray Expires October 6, 2010 [Page 2]
INTERNET-DRAFT File Transfer Protocol HOST Command April 2010
Network Virtual File System (NVFS).
This means that different virtual hosts cannot offer different
virtual file systems to clients, nor can they offer different
authentication systems.
++++
[it's worth pointing out here that a similar situation existed for the web many
years ago, and how it was resolved, and that you want to do the same thing here
basically]
++++
No scheme can overcome this without modifications of some kind to the
user-PI and to the user-FTP process. That process is the only entity
####
I think you mean to say that "Any scheme to overcome this needs to indicate not
only the destination IP address, but also a name associated with the desired
virtual FTP server. This information is currently discarded in the FTP
application at the user interface of the client."
(note: I used the term "name" - not DNS name here; see preface text as to why)
####
that knows which virtual host is required. It has performed the
domain name-to-IP address translation, and thus has the original
domain name available.
++++
I don't know why the following section is in the intro. You do need to discuss
alternatives not pursued, but maybe later.
++++
One method that could be used to allow a style of virtual host would
be for the client to simply send a "CWD" command after connecting,
using the virtual host name as the argument to the CWD command.
This would allow the server-FTP process to implement the file stores
of the virtual hosts as sub-directories in its NVFS. This is simple,
and most server-FTP implementations support this without requiring
any code changes.
While that method is simple to describe, and to implement, it suffers
from several drawbacks. First, the "CWD" command is available only
after the user-PI has authenticated itself to the server-FTP process.
Thus, all virtual hosts would be required to share a common
authentication scheme if they used this method. Secondly, to make
the virtual host truly transparent, either the server-FTP process
needs to be modified to include information that shows the special
nature of this first CWD command (negating most of the advantage of
this scheme), or all users must see the same identical NVFS view upon
connecting (they must connect in the same initial directory), or the
NVFS must implement the full set of virtual host directories at each
possible initial directory for any possible user. Thirdly, and
again, unless the server is specially modified, a user connecting
this way to a virtual host would be able to easily move to any other
virtual host supported at the same server-FTP process, exposing the
nature of the virtual host.
++++
I think the above is too glib. The first point is worth noting - that it is
available only after the connection is established, and so cannot be used to
determine per-host parameters (including, but not limited to, authentication).
The second point can easily be overcome with an optional standard prefix to the
CWD command, e.g.:
CWD fqdn:{path}
The ability of the user to move between virtual FTP servers can be seen as a
benefit, not a liability. The question is whether the benefit is needed, and
whether it presents new problems (e.g., re-authenticating on different servers
may not be feasible).
++++
Another suggested method is to simply overload the "ACCT" for FTP
virtual hosts, but this proposal is unacceptable for several
reasons with regard to when the ACCT command is sent during the
request flow. Sections 5.4 and 6 of [RFC959] document the request
flow for a login sequence as USER -PASS -ACCT. This flow of
commands may be acceptable when you are considering a single user
having multiple accounts on an FTP server, but fails to
differentiate between virtual hosts when you consider the following
two issues:
++++
I think this is easier to state. The problem is that you need to do ACCT before
you do USER or PASS, since you don't know on what virtual server to validate
the first two.
I.e., same issue as before. Don't beat a dead horse, just note it and move on.
++++
a. The first problem with overloading the ACCT command is
certificate negotiation when using the FTP security extensions
that are documented in [RFC2228] and [RFC4217]. In order to
safeguard user credentials, security mechanism and certificate
Hethmon & McMurray Expires October 6, 2010 [Page 3]
INTERNET-DRAFT File Transfer Protocol HOST Command April 2010
negotiation must occur before login credentials are sent by the
client. The problem with using the ACCT command in this scenario
is that there is no way of ensuring that the certificate matches
the correct virtual host before the user credentials are sent.
b. The second problem with overloading the ACCT command is how
user credentials are implemented for FTP virtual hosts. FTP
server implementations may allow the use of custom user
credentials on a per-virtual-host basis. For example, in one
particular implementation the virtual host negotiation occurs,
and then the user credentials are looked up using the account
mechanism that is specific to that virtual host. So once again
the virtual host negotiation must take place before the user
credentials are sent.
The conclusion from the examination of the existing possibilities
seems to be that to obtain an adequate emulation of "real" FTP
servers, server modifications to support virtual hosts are
necessary. A new FTP command seems the most likely solution to
provide the required level of support.
++++
I think you may need to say:
a new command that *precedes* USER is required.
However, I think it isn't. You can easily do the colon trick:
vhostname:username
or even the at-sign trick:
username(_at_)vhostname
That seems like it's also potentially more easily understood to users (it's
already supported by many FTP command lines using that syntax), and reduces the
login by one round trip exchange as well.
++++
2. Document Conventions
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
"SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in
this document are to be interpreted as described in [RFC2119].
++++
The document uses these terms in lower case, presumably not as key words. If
that is desired, it would be useful to be explicit about the difference, e.g.,
by including the following text:
When used in lower case, these words have their conventional meaning
and do not convey the interpretations in [RFC2119].
++++
In examples, "C>" and "S>" indicate lines sent by the client and
server, respectively.
This document also uses notation defined in [RFC959] and [RFC1123].
In particular, the terms "reply", "user", "NVFS", "NVT", "file",
"pathname", "FTP commands", "DTP", "user-FTP process", "user-PI",
"user-DTP", "server-FTP process", "server-PI", "server-DTP", "mode",
"type", "control connection", "data connection", and "ASCII", are
all used here as defined there.
Syntax required is defined using the Augmented BNF defined in
[RFC5234]. Some general ABNF definitions are required throughout the
document, those will be defined later in this section. At first
reading, it may be wise to simply recall that these definitions
exist here, and skip to the next section.
2.1. Basic Tokens
This document imports the core definitions given in Appendix B of
[RFC5234]. There definitions will be found for basic ABNF elements
like ALPHA, DIGIT, SP, etc. To that, the following term is added
for use in this document.
TCHAR = VCHAR / SP / HTAB ; visible plus white space
Hethmon & McMurray Expires October 6, 2010 [Page 4]
INTERNET-DRAFT File Transfer Protocol HOST Command April 2010
The VCHAR (from [RFC5234]) and TCHAR rules give basic character
types from varying sub-sets of the ASCII character set for use in
various commands and responses.
Note that in ABNF, string literals are case insensitive. That
convention is preserved in this document, and implies that FTP
commands and parameters that are added by this specification have
values that can be represented in any case. That is, "HOST" is the
same as "host", "Host", "HoSt", etc., and "ftp.example.com" is the
same as "Ftp.Example.Com", "fTp.eXample.cOm", etc.
2.2. Server Replies
Section 4.2 of [RFC959] defines the format and meaning of replies
by the server-PI to FTP commands from the user-PI. Those reply
conventions are used here without change.
error-response = error-code SP *TCHAR CRLF
error-code = ("4" / "5") 2DIGIT
Implementers should note that the ABNF syntax (which was not used in
[RFC959]) used in this document, and other FTP related documents,
sometimes shows replies using the one line format. Unless otherwise
explicitly stated, that is not intended to imply that multi-line
responses are not permitted. Implementers should assume that, unless
stated to the contrary, any reply to any FTP command (including QUIT)
may be of the multi-line format described in [RFC959].
Throughout this document, replies will be identified by the three
digit code that is their first element. Thus the term "500 reply"
means a reply from the server-PI using the three digit code "500".
3. The HOST command
A new command "HOST" is added to the FTP command set to allow the
server-FTP process to determine to which of possibly many virtual
hosts the client wishes to connect. This command SHOULD be issued
before the user is authenticated, allowing the authentication scheme,
####
When can this command ever come AFTER a user is authenticated? Is that ever
intended to be valid? If not, then change the SHOULD to MUST. However, as noted
above, embedding this info in the USER command would have achieved the same
effect without requiring protocol machinery to determine the order of commands.
####
and set of legal users, to be dependent upon the virtual host chosen.
Server-FTP processes SHOULD treat a situation where the HOST command
is issued after the user has been authenticated using one of the
following two behaviors:
a. Treat the late HOST command as an erroneous sequence of
commands and return a 503 reply.
b. Treat the late HOST command as a though a REIN command was
sent before the HOST command and reset the user-PI to the state
that existed after the TCP connection was first established and
before the initial user authentication, and then return the
Hethmon & McMurray Expires October 6, 2010 [Page 5]
INTERNET-DRAFT File Transfer Protocol HOST Command April 2010
appropriate reply for the HOST command.
++++
This is further evidence that this info might be easier to embed in the USER
command. The REIN effectively does a logout, and then the next USER command
would be expected to have the needed vhostname information anyway.
++++
Servers should note that the response to the HOST command is a
sensible time to send their "welcome" message. This allows the
message to be personalized for any virtual hosts that are supported,
and also allows the client to determine the supported languages, or
representations, for the message, and other messages, via the FEAT
response, and select an appropriate one via the LANG command. See
[RFC2640] for more information.
3.1. Syntax of the HOST command
The HOST command is defined as follows.
host-command = "HOST" SP hostname CRLF
hostname = domain / IP-literal
####
IMO,
hostname = bytetring
I see no reason to specify it further. To the protocol, it can/should be
opaque. The server can decide whether it likes it or not. If I want to call a
vhost "mickey_mouse!!@@##" I should be able to, IMO.
The rest of the syntax can be omitted.
####
domain = sub-domain *("." sub-domain)
sub-domain = let-dig [ldh-str]
let-dig = ALPHA / DIGIT
ldh-str = *( ALPHA / DIGIT / "-" ) let-dig
IP-literal = ( "[" IPv6address "]" ) / IPv4address
IPv6address = 6( h16 ":" ) ls32
/ "::" 5( h16 ":" ) ls32
/ [ h16 ] "::" 4( h16 ":" ) ls32
/ [ *1( h16 ":" ) h16 ] "::" 3( h16 ":" ) ls32
/ [ *2( h16 ":" ) h16 ] "::" 2( h16 ":" ) ls32
/ [ *3( h16 ":" ) h16 ] "::" h16 ":" ls32
/ [ *4( h16 ":" ) h16 ] "::" ls32
/ [ *5( h16 ":" ) h16 ] "::" h16
/ [ *6( h16 ":" ) h16 ] "::"
ls32 = ( h16 ":" h16 ) / IPv4address
; least-significant 32 bits of address
h16 = 1*4HEXDIG
; 16 bits of address represented in hexadecimal
IPv4address = dec-octet "." dec-octet "." dec-octet "." dec-octet
dec-octet = DIGIT ; 0-9
/ %x31-39 DIGIT ; 10-99
/ "1" 2DIGIT ; 100-199
/ "2" %x30-34 DIGIT ; 200-249
/ "25" %x30-35 ; 250-255
host-response = host-ok / error-response
host-ok = "220" [ SP *TCHAR ] CRLF
Hethmon & McMurray Expires October 6, 2010 [Page 6]
INTERNET-DRAFT File Transfer Protocol HOST Command April 2010
As with all FTP commands, the "HOST" command word is case
independent, and may be specified in any character case desired.
The "hostname" (given as a parameter) specifies the virtual host to
which access is desired. It should normally be the same host name
that was used to obtain the IP address to which the FTP control
connection was made, after any client conversions have been completed
that convert an abbreviated or local alias to a complete (fully
qualified) domain name, but before resolving a DNS alias (owner of a
CNAME resource record) to its canonical name.
####
I disagree with the above paragraph. The contents of the hostname argument MUST
be opaque to the protocol.
The remainder of this section (and/or document) needs to be updated
accordingly, IMO.
That also means that the client can send what it wants - fully qualified or
not. That's up to the client implentation to decide. The semantics of that name
should be meaningless to the protocol.
####
Internationalization of domain names is only supported through the
use of Punycode as described in [RFC3492].
####
This might say that internationalization of the opaque string is handled via
Punycode. It should not refer to domain names per se; that is an interpretation
that can be made by the server, but must not be assumed in the protocol.
####
If the user was given an IPv4 or IPv6 literal address, and
consequently was not required to derive the literal address from a
hostname, the client should send the HOST command with the IPv4 or
IPv6 literal address as specified to it. While it may seem
counter-intuitive to specify a literal address by using the HOST
command after the client has already connected to the server using
a literal address, this should be expected behavior because a
user-FTP process should not be required to differentiate between a
fully qualified domain name and an IPv4 or IPv6 network literal
address. That being said, if the IPv4 or IPv6 literal address
specified by the client does not match the literal address for the
server, the server SHOULD respond with a 504 reply to indicate that
the IPv4 or IPv6 literal address is not valid.
####
First, these strings ought to be opaque to the protocol. The server can decide
to check them, but the protocol MUST NOT, IMO.
Note, though, that even if the server interprets them, let's discuss whether
the server should do the above or not (i.e., this would be an
implementation/configuration choice, not a protocol requirement, if so).
If you are intending to support only virtual hosting, then there is never a
need to send the IP address in a HOST command. However, if you do want to allow
it for completeness, then it MUST respond with a 504 if the address does not
match (this can't be a SHOULD unless there is a valid reason for either another
response, or a valid action, and this doc doesn't argue for either).
However, there could be a case where the FTP server listens on multiple
addresses on a multihomed host. In that case, this command would appear to
provide a valid way to connect to one address and access the data on the other.
Since this would succeed (even if not
intended!) if the user had provided a fqdn, I see no reason why it should fail
when they provide an IP address. Consider:
a.example.net = 10.0.0.1
b.example.com = 192.0.0.2
A user might issue:
ftp a.example.net joe(_at_)b(_dot_)example(_dot_)net
connecting to: 10.0.0.1
issuing: HOST b.example.net
if that works (and I can't see why it wouldn't, so far), then the following
should also work:
ftp 10.0.0.1 joe(_at_)192(_dot_)0(_dot_)0(_dot_)2
connecting to: 10.0.0.1
issuing: HOST 192.0.0.2
If you don't want either example to work, then you missed a step on the HOST
command:
HOST fqdn
resolve fqdn to a set of IP addresses "S"
the current connection IP destination address MUST
match an address in S, otherwise the server MUST
return an error
HOWEVER, I strongly discourage this sort of rule. Overall, IMO, the HOST
command takes an *arbitrary* argument that is interpreted *only* as the context
of access control and the NVFS. As a result, *all* such checks that attempt to
match any contents of this identifier to the current connection MUST NOT be
prohibited by the protocol (the server configuration MAY check these or not, as
supported and/or configured).
####
When the hostname parameter contains a literal address, square
brackets are expected to disambiguate port numbers syntax from IPv6
address syntax.
++++
This doesn't make sense; below you prohibit the use of port numbers in the
command, so why do you need to disambiguate them?
++++
Therefore, if the literal address is an IPv6
address, the IPv6 address should be enclosed in square brackets
(after eliminating any syntax that might also - but is not required
to - be enclosed in brackets, and from which the server deduced that
a literal address had been specified.) For example:
HOST 192.0.2.1
HOST [FE80::c000:0201]
HOST [::192.0.2.1]
should be sent if the client had been instructed to respectively
connect to "192.0.2.1", "FE80::c000:0201", or "192.0.2.1" and IPv6
syntax is preferred.
The client MUST NOT send the port number as part of the HOST
command, even when the client has been instructed to connect to a
non-standard port. For example, the server-PI should respond with
a 501 reply if the client sends a HOST command with syntax like
"HOST 192.0.2.1:2112" or "HOST [FE80::c000:0201]:2112".
####
If raw addresses are deemed useful, then ports should be as well.
Further ports MUST be allowed to be specified either by their number or their
service name (e.g., "ftp"), IMO.
I would assume that service names MUST NOT be translated or validated by the
protocol, just as the fqdn MUST NOT be. The server implementation MAY
check/translate these as needed to coordinate with their configuration, but
this is outside the scope of the protocol, IMO.
####
The hostname parameter is otherwise to be treated as a fully
Hethmon & McMurray Expires October 6, 2010 [Page 7]
INTERNET-DRAFT File Transfer Protocol HOST Command April 2010
qualified domain name or relative name as those terms are defined
in section 3.1 of [RFC1034]. This implies that the name is to be
treated as a case-independent string, meaning that uppercase ASCII
characters are to be treated as equivalent to their corresponding
lowercase ASCII characters, but otherwise preserved as given. It
also implies some limits on the length of the parameter and of the
components that create its internal structure. Those limits are not
altered in any way here.
####
None of this should be validated, translated, or checked in the protocol. The
server implementation can decide whether it likes the name, or it matches its
configuration tables.
####
Neither [RFC1034] nor [RFC1035] impose any other restrictions upon
what kinds of names can be stored in the DNS. This specification,
however, allows only the of names that can be inferred from the ABNF
grammar given for the "hostname".
3.2. HOST command semantics
Upon receiving the HOST command, before authenticating the user-PI, a
server-FTP process should validate that the hostname given represents
a valid virtual host for that server, and, if it is valid, establish
the appropriate environment for that virtual host. The resultant
actions needed to create that environment are not specified here, and
may range from doing nothing at all, to performing a simple change of
working directory, to making much more elaborate state changes, as
required.
++++
It might be useful to mention other changes, such as using different
username/password lists.
++++
If the hostname specified is unknown at the server, or if the server
is otherwise unwilling to treat the particular connection as a
connection to the hostname specified, the server will respond with a
504 reply.
Note: servers may require that the name specified is in some sense
equivalent to the particular network address that was used to reach
the server.
####
Agreed - that is a server configuration issue. It's important to note here that
this MUST NOT be checked in the protocol itself, though.
####
If the hostname specified would normally be acceptable, but for any
reason is temporarily unavailable, the server-FTP process SHOULD
reply to the HOST command with a 421 reply and close the connection.
++++
If you say SHOULD, you ought to indicate conditions under which you might not,
i.e., when would you not reply with 421? would you reply with a different code,
or just wait? It's not useful to include SHOULDs without giving a hint as to
why they are not MUSTs, IMO.
++++
The "220" reply code for the HOST command is the same as the code
that is used in the initial "welcome" message that is sent after
the connection is established. This reply code is used deliberately
in order to allow the implementation of the front-end FTP server as
a wrapper, which simply waits for the HOST command, and then invokes
a server that is compliant with [RFC959] in the appropriate
environment for the particular hostname received.
++++
A wrapper could as easily wait for a custom response command. I am not sure why
the response command is the same as for the initial connection. Many other
commands have their own response codes; IMO, this should as well:
260 HOST command accepted
++++
3.2.1. REIN command semantics
As specified in [RFC959], the REIN command returns the state of
the connection to that it was immediately after the transport
connection was opened. This specification makes no changes to that
Hethmon & McMurray Expires October 6, 2010 [Page 8]
INTERNET-DRAFT File Transfer Protocol HOST Command April 2010
behavior. The effect of a HOST command will be lost if a REIN
command is performed, and a new HOST command must be issued.
++++
The effect of a HOST command MUST be reset if a REIN command is performed.
(i.e., the FTP authentication and NVFS parameters are set to the initial
connection default after a REIN command).
++++
Implementers of user-FTP should be aware that server-FTP
implementations that implement the HOST command as a wrapper around
older implementations will be unable to correctly implement the REIN
command. If they do, REIN will typically return the server-FTP to
the state that it was in immediately after the HOST command was
issued, instead of to the state that it was in immediately after the
connection was opened.
++++
This seems like a problem; it seems like you need a few things here:
1) a way to find out what the current HOST environment is
CHOSTIS
SHOST a.example.com
2) a way to set the host to the default of a server, e.g.,
using the HOST command with no argument??
CHOST
SHOST set to server default
(note: you now need to allow successive HOST commands to be permitted after
either login on REIN)
++++
3.2.2. User-PI usage of HOST
A user-PI that conforms to this specification MUST send the HOST
command after opening the transport connection, or after any REIN
command, before attempting to authenticate the user with the USER
command. The following example illustrates what a typical login
sequence might look like when the HOST command is used:
CHOST foobar.com
S220 Host accepted
++++
I think you need to say:
S220 Service ready for new user
However, I prefer:
S260 HOST accepted
++++
CUSER foo
S331 Password required
CPASS bar
S230 User logged in
The HOST command can be used in combination with the ACCT command
to differentiate between a user's various accounts on a specific
virtual host. In this scenario, the user-PI sends a HOST command
which the server-PI uses to route activity to the correct virtual
host, the user-PI sends credentials using the USER and PASS commands
which the server-PI validates, then the user-PI sends an ACCT
command to specify any additional account information for the
server-PI implementation. The following example illustrates a
sequential series of client commands that specify both a HOST and
ACCT, with the server responses omitted for brevity:
CHOST foobar.com
CUSER foo
CPASS bar
CACCT project1
This is also true when the HOST command is used with the AUTH and
ADAT commands that are discussed in [RFC2228] and [RFC4217]. In
this scenario, the user-PI sends a HOST command which the server-PI
uses to route activity to the correct virtual host, then the user-PI
uses the AUTH and ADAT commands to negotiate the security mechanism
and certificate with the server-PI, then the user-PI sends user
credentials using the USER and PASS commands which the server-PI
validates. After which the user-PI may send an ACCT command to
specify any additional account information for the server-PI
Hethmon & McMurray Expires October 6, 2010 [Page 9]
INTERNET-DRAFT File Transfer Protocol HOST Command April 2010
implementation. The following example illustrates a sequential
series of client commands that specify both a HOST and ACCT when
used in conjunction with the security commands that are discussed
in [RFC2228] and [RFC4217], with the server responses omitted for
brevity:
CHOST foobar.com
CAUTH <mechanism-name>
CADAT <base64data>
CUSER foo
CPASS bar
CACCT project1
3.2.3. State Diagrams
The state diagrams in this section illustrate typical sequences for
command and reply interchange between the user-PI and server-PI.
These diagrams are modeled on the similar diagrams in section 6 of
[RFC959].
In both diagrams, the (B) "begin" state is assumed to occur after
the transport connection has opened, or after a REIN command has
succeeded. Other commands (such as FEAT [RFC2389]) that require no
authentication may have intervened.
In each diagram, a three-digit reply indicates a precise server reply
code. A single digit on a reply path indicates any server reply that
begins with that digit, except where a precise server reply code is
defined on another path. For example, a single digit "5" will apply
to "500", "501", "502", etc., when those reply codes are not
expressly defined in the diagram. For each command there are three
possible outcomes: success (S), failure (F), and error (E). In the
state diagrams below we use the symbol B for "begin", and the
symbol W for "wait for reply".
####
I think this should allow for multiple HOST commands.
It should also show how the REIN command sets you back to a state where you can
accept new HOST commands.
####
+---+ HOST +---+ 1,3,5
| B |---------->| W |-----------------
+---+ +---+ |
| | |
2,500,502 | | 4,501,503,504 |
-------------- ------------- |
| | |
V 1 | V
+---+ USER +---+-------------->+---+
| |---------->| W | 2 ----->| E |
+---+ +---+------ | --->+---+
| | | | | |
3 | | 4,5 | | | |
-------------- ----- | | | |
| | | | | |
Hethmon & McMurray Expires October 6, 2010 [Page 10]
INTERNET-DRAFT File Transfer Protocol HOST Command April 2010
| | | | | |
| ---------- | |
| 1| | | | |
V | | | | |
+---+ PASS +---+ 2 | ------->+---+
| |---------->| W |-------------->| S |
+---+ +---+ ----------->+---+
| | | | | |
3 | |4,5| | | |
-------------- -------- | ----
| | | | | |
| | | | | |
| ------------ |
| 1,3| | | | |
V | 2| | | V
+---+ ACCT +---+-- | ------>+---+
| |---------->| W | 4,5 --------->| F |
+---+ +---+-------------->+---+
When the HOST command is used in combination with the FTP security
extensions that were introduced in [RFC2228], it SHOULD precede
the security handshake. This allows both user-PI and server-FTP
processes to map an FTP HOST to security data appropriately.
The following state diagram shows a typical sequence of flow of
control when HOST is used with the AUTH and ADAT commands that are
discussed in [RFC2228]. The explanations that were given for the
previous diagram also apply to this diagram.
+---+ HOST +---+ 1,3,5
| B |---------->| W |-----------------
+---+ +---+ |
| | |
2,500,502 | | 4,501,503,504 |
+-------------- ------------- |
| | |
V | |
+---+ AUTH +---+ 4,5 | |
| |---------->| W |----------->| |
+---+ +---+ | |
234 | | | |
--------- | 334 | |
| | | |
---------------|------ | |
| | | | | |
V | V 335 | | |
+---+ | ADAT +---+---- | |
| |---------->| W | 4,5 | |
Hethmon & McMurray Expires October 6, 2010 [Page 11]
INTERNET-DRAFT File Transfer Protocol HOST Command April 2010
+---+ | +---+----------->| |
| | | |
---- 235| | |
| -------------- | |
| | | |
V V 1 | V
+---+ USER +---+-------------->+---+
| |---------->| W | 2 ----->| E |
+---+ +---+------ | --->+---+
| | | | | |
3 | | 4,5 | | | |
-------------- ----- | | | |
| | | | | |
| | | | | |
| ---------- | |
| 1| | | | |
V | | | | |
+---+ PASS +---+ 2 | ------->+---+
| |---------->| W |-------------->| S |
+---+ +---+ ----------->+---+
| | | | | |
3 | |4,5| | | ----
-------------- -------- | |
| | | | | |
| | | | | |
| ------------ |
| 1,3| | | | |
V | 2| | | V
+---+ ACCT +---+-- | ------>+---+
| |---------->| W | 4,5 --------->| F |
+---+ +---+-------------->+---+
3.3. HOST command errors
The server-PI should reply with a 500 or 502 reply if the HOST
command is unrecognized or unimplemented. A 503 reply may be sent
if the HOST command is given after a previous HOST command, or after
a user has been authenticated. Alternately, the server may define
++++
Please explain why successive HOST commands should be invalid. I can't see a
reason, and there's potential utility (e.g., to allow REIN wrappers, as
described earlier).
++++
behavior that allows the HOST command after a user has been
authenticated. A 501 reply should be sent if the hostname given is
syntactically invalid, and a 504 reply should be sent if a
syntactically valid hostname is not a valid virtual host name for the
server. In all such cases, the server-FTP process should act as if
no HOST command had been given.
A user-PI receiving a 500 or 502 reply should assume that the
server-PI does not implement virtual servers by using the HOST
command. The user-PI may then proceed to login as if the HOST
command had succeeded, and may attempt a CWD command to the hostname
after authenticating the user.
Hethmon & McMurray Expires October 6, 2010 [Page 12]
INTERNET-DRAFT File Transfer Protocol HOST Command April 2010
A user-PI receiving an error reply that is different from the errors
that have been described here should assume that the virtual HOST is
unavailable, and terminate communications.
A server-PI that receives a USER command, beginning the
authentication sequence, without having received a HOST command,
SHOULD NOT reject the USER command. Clients conforming to earlier
FTP specifications do not send HOST commands. In this case the
server may act as if some default virtual host had been explicitly
selected, or may enter an environment different from that of any
supported virtual hosts, perhaps one in which a union of all
available accounts exists, and which presents an NVFS that appears
to contain subdirectories that contain the NVFS for all supported
virtual hosts.
3.4. FEAT response for HOST command
A server-FTP process that supports the HOST command and virtual FTP
servers MUST include, in the response to the FEAT command [RFC2389],
a feature line indicating that the HOST command is supported. This
line should contain the single word "HOST". This command word is
case insensitive, but it SHOULD be transmitted in upper case only.
That is, the response SHOULD be:
CFEAT
S211- <any descriptive text>
S ...
S HOST
S ...
S211 End
The ellipses indicate place holders where other features may be
included, and are not required. The single space indentation of each
feature line is mandatory [RFC2389].
4. Security Considerations
With the introduction of virtual hosts to FTP, and the possible
accompanying multiple authentication environments, server
implementers will need to take some care to ensure that the integrity
of user credentials is maintained. A general discussion of issues
related to the security of FTP can be found in [RFC2577].
++++
I think you need to have more here, esp. on how the server MUST reset security
parameters when a HOST command is issued, and new commands MUST be interpreted
in the security context of the most recent HOST command.
++++
####
There's also the issue of whether a server should validate the HOST command
parameters as matching the incoming connection. IMO, that is something that the
server should be checking, not the protocol, but it might be important to note
whether there should be a default (MUST
validate?) or whether there should be no such assumptions because the HOST
command argument is treated by the protocol as an opaque string (IMO, that's
the better answer here).
####
5. IANA Considerations
IANA is requested to register the following FTP extension according
to the procedure established by [RFC5797]:
+-------+------+-------------------+------+------+------------------+
| cmd | FEAT | description | type | conf | RFC#s/References |
Hethmon & McMurray Expires October 6, 2010 [Page 13]
INTERNET-DRAFT File Transfer Protocol HOST Command April 2010
| | Code | | | | and Notes |
+-------+------+-------------------+------+------+------------------+
| HOST | HOST | Hostname | a | o | [#1] |
+-------+------+-------------------+------+------+------------------+
[[ RFC Editor: Please update note #1 in the above table with the
number of this document. ]]
6. References
6.1. Normative References
[RFC959] Postel, J., Reynolds, J., "File Transfer Protocol (FTP)",
STD 9, RFC 959, October 1985
[RFC1034] Mockapetris, P., "Domain Names - Concepts and Facilities",
STD 13, RFC 1034, November 1987
[RFC1035] Mockapetris, P., "Domain Names - Implementation and
Specification", STD 13, RFC 1035, November 1987
[RFC1123] Braden, R,. "Requirements for Internet Hosts --
Application and Support", STD 3, RFC 1123, October 1989
[RFC2119] Bradner, S., "Key words for use in RFCs to Indicate
Requirement Levels", BCP 14, RFC 2119, March 1997
[RFC2228] Horowitz, M., Lunt, S., "FTP Security Extensions",
RFC 2228, October 1997
[RFC2389] Hethmon, P., Elz, R., "Feature negotiation mechanism for
the File Transfer Protocol", RFC 2389, August 1998
[RFC2640] Curtin, W., "Internationalization of the File Transfer
Protocol", RFC 2640, July 1999
[RFC3492] Costello, A., "Punycode: A Bootstring encoding of Unicode
for Internationalized Domain Names in Applications (IDNA)",
RFC 3492, March 2003
[RFC4217] Ford-Hutchinson, P., "Securing FTP with TLS", RFC 4217,
October 2005
[RFC5234] Crocker, D., Overell, P., "Augmented BNF for Syntax
Specifications: ABNF", RFC 5234, January 2008
6.2. Informative References
[RFC2577] Allman, M., Ostermann, S., "FTP Security Considerations",
RFC 2577, May 1999
Hethmon & McMurray Expires October 6, 2010 [Page 14]
INTERNET-DRAFT File Transfer Protocol HOST Command April 2010
[RFC5797] Klensin, J., Hoenes, A., "FTP Command and Extension
Registry", March 2010
7. Acknowledgments
Robert Elz and Paul Hethmon provided a detailed discussion of the
HOST command in their Internet draft titled "Extensions to FTP"
as part of their work with the FTPEXT Working Group at the IETF.
Their work formed the basis for much of this document, and their
help has been greatly appreciated. They would also like to credit
Bernhard Rosenkraenzer for having first suggested and described the
HOST command.
Alexey Melnikov, Alfred Hoenes, and John Klensin have made several
suggestions about earlier versions of this document, most of which
have been incorporated and their contributions are gratefully
acknowledged.
8. Authors' Addresses
Paul Hethmon
Hethmon Brothers
2305 Chukar Road
Knoxville, TN 37923 USA
Email: phethmon(_at_)hethmon(_dot_)com
Robert McMurray
Microsoft Corporation
One Microsoft Way
Redmond, WA 98052-6399
Email: robmcm(_at_)microsoft(_dot_)com
Hethmon & McMurray Expires October 6, 2010 [Page 15]
_______________________________________________
Ietf mailing list
Ietf(_at_)ietf(_dot_)org
https://www.ietf.org/mailman/listinfo/ietf