Request for Comments: 607 Mark Krilanovich
NIC # 21255 George Gregg
references: RFC #542 UCSB Jan 1974
Comments on the File Transfer Protocol
There are several aspects of the File Transfer Protocol that constitute
serious drawbacks. Some of these are quite basic in nature, and imply
substantial design changes; these will be discussed in a later RFC.
Others could be remedied with very little effort, and this should be done
as soon as possible.
Following is a list of those problems that can be easily solved, together
with their proposed solutions:
of data connections, there is no way for the user to make him "active"
again. SOLUTION: define a new command, with a command verb of "ACTV", to
mean that the server is to issue a CONNECT rather than a LISTEN on the data
socket. If the server is already "active", the command is a no op. "ACTV"
is to have the same reply codes as "PASV".
same length, and design of an FTP user would be simpler if either all
command verbs were the same length, or if multiple blanks were allowed
following the verb. SOLUTION: replace the only three-letter verb, "BYE",
with a four-letter one, such as "QUIT", and constrain future command verbs
to be four letters long.
is left unspecified. After sending a STOR command, for example, a user
process has no way of knowing which to wait for first, the "250 FILE
TRANSFER STARTED" reply, or establishment of the data connection. SOLUTION:
specify that the server is to send a "250" reply before attempting to
establish the data connection. If it is desired to check if the user is
logged in, if the file exists, or if the user is to be allowed access to
the file, these checks must be made before any reply is sent. The text of
the "250" reply would perhaps be more appropriate as "250 OPENING DATA
CONNECTION", since it comes before actual data transfer begins. If the
server wishes to send an error reply in the event that the data connection
cannot be opened, it is to be sent in lieu of the "252 TRANSFER COMPLETE"
reply.
that is unimplemented because it is not needed (e.g., "ACCT" or "ALLO").
Even though the text of the reply indicates that the command has been
ignored, it is obviously impossible for a user process to know that there
is no real "error". SOLUTION: require that any server that does not support
a particular command because it is not needed in that system must return a
success reply.
password, account, or pathname. It is true that every system implementing
an FTP server likely has different maxima for its own parameters, but it is
nearly impossible for the writer of an FTP user (which must converse with
many FTP servers) to construct an indefinite length buffer. Typically some
-1-
arbitrary maximum must be chosen. SOLUTION: specify a maximum length for
TELNET lines, user names, passwords, account numbers, and pathnames. This
is to be done after conducting a poll of serving sites concerning their
individual maxima.
solves a minor problem for a few server FTP implementers at the expense of
creating a major problem for all user FTP implementers. The logic needed to
decode a multi-line reply is unnecessarily complex, and made an order of
magnitude more so by the fact that multi-line replies are allowed to be
nested. SOLUTION: assign a unique (numeric) reply code, such as "009", to
be used on all lines of a multi-line reply after the first.
the maximum allowed level of nesting is left unspecified creates a hardship
for implementers of user FTPs. This hardship is somewhat easily solved on a
machine that has hardware stacks, but not so for other machines. SOLUTION:
specify a maximum level of nesting of multi-line replies.
(EOR) are explicit, including the final one." This prohibits sending data
between the final end of record and the end of file, but does not specify
what the receiver of data is to do if this rule is broken. That is, should
the intervening data be discarded or treated as a new (final) record?
SOLUTION: specify that if an end-of-file marker is not immediately preceded
by an end-of-record marker, the intervening data is to be discarded.
A major complaint about the protocol concerns the fact that the writer of
an FTP user process must handle a considerable number of special cases
merely to determine whether or not the last command sent was successful. It
is admitted that the protocol is well-defined in all the following areas,
but it is important to realize that the characteristic "well-defined" is
necessary, but not sufficient; for many reasons, it is very desirable to
employ the simplest mechanism that satisfies all the needs. Following is a
list of those drawbacks that unduly complicate the flow chart of an FTP
user process:
"USER" command, for example, returns a "230" whereas a successful "BYTE"
command returns a "200". The concept that success replies should have an
even first digit and failure replies an odd first digit does not apply, as
"100, means success for "STAT", and "402" means failure for "BYTE".
SOLUTION: specify that any command must return a reply code beginning with
some unique digit, such as "2", if successful, and anything other than that
digit if not successful.
"REIN", and "BYE". It is undesirable for an FTP user to be required to keep
a list of reply codes for each command, all of which mean "command
accepted, continue". SOLUTION: same as for (9.) above. The desire to
communicate more specific information than simply "yes" or "no", such as
the difficulty in the login procedure that some sites do not need all the
parameters, may be solved by having, for example, "238" mean "PASSWORD
ACCEPTED, YOU ARE NOW LOGGED IN", and "237" mean "PASSWORD ACCEPTED,
ACCOUNT NOW NEEDED" The important point is that the idea of "command
accepted" is conveyed by the initial "2", and that finer gradations of
meaning can be deduced by the user process, if desired.
-2-
of view of a user FTP process, and their reply codes have no characteristic
that makes them easily distinguishable. For example, "010 message from
operator" and "050 FTP commentary" are superfluous to a user process, and
"000 announcing FTP" (in place of "300" greeting) is not. SOLUTION: specify
that any reply that has meaning only to a human user and not to a user
process must have a reply code beginning with a unique digit, such as "0".
The continuation line reply code proposed in (8.) above falls into this
category, and therefore must start with the same unique digit.
expected delay" immediately after completion of the ICP if input cannot be
accepted right away is another instance of multiple reply codes having the
same meaning to a user process. SOLUTION: require that the server send a
reply with a "020" reply code in the situation cited. If it is desired to
communicate more detailed information, the text of the reply may used for
this purpose.
In addition to the above mentioned weaknesses in the protocol, the
following is believed to be a typographical error:
commands "BYTE", "SOCK", "PASV", "TYPE", "STRU", "MODE", "ALLO", "REST",
"SITE", AND "STAT". This reply code means "ENTER ACCOUNT" (if required as
part of login sequence), and perhaps should be "332 LOGIN FIRST, PLEASE".
This is especially indicated by the fact that "332" is not listed anywhere
in the command-reply correspondence table.
-3-