|Title||Memo to FTP group: Proposal for File Access Protocol
Network Working Group J. Day
Request for Comments: 520 Center for Advanced Computation
NIC: 16819 25 June 1973
A Proposed File Access Protocol Specification
Attached is a proposal for the File Access Protocol. FAP is an
extension to FTP. I believe the specification is fairly general and
should provide a good jumping-off place. I hope the protocol is
specified in such a way as to fit with idiosyncrasies of most
systems. If the protocol would cause an inordinate amount of burden
on your system for one reason or another I would like to hear about
At some later date when the difficulties of implementation are better
known, I would like to see several levels of implementation specified
and implementation be done in terms of those levels.
From rumors I have heard I believe this will also allow creation and
transfer of what TENEX calls "holey" files. But, I am not sure of
all of the implications of that, or what would happen (or should
happen) when a "holey" file is moved to a site that doesn't really
have such a thing, per se. Comments from the TENEX crowd would be
I think some further work could be done to make FAP easier for record
oriented systems. This would probably require an extra command or
parameter to specify all operations are in terms of records.
Comments are invited.
In the long run though, I would like to see FAP thrown away. The
commands as they are described merely add a finer structure to the
present RETR, STOR, and APPE without much additional overhead. The
OPEN R FOO.BAR CRLF
READ ALL CRLF
is equivalent to RETR FOO.BAR CRLF. FAP could be merged with FTP to
give a much richer, coherent whole.
In writing this document, I ran into the deficiency of reply codes
for protocols. Three digits is no where near enough. I would like
to suggest that as another interim solution we go to a five digit
reply with two for specific categories (such as Primary access, FTP
results, etc.) and two for specific results. In the meantime, the
NWG should begin considering a general scheme for reply codes -- one
that doesn't need revising every two years.
Comments, complaints, etc. are welcomed. I may be reached through
network mail at ISI (DAY) or Multics (DAY Cnet) or by phone at the
University of Illinois (217) 333-6544.
File Access Protocol
The purpose of the File Access Protocol is to provide a method for
processes to access non-local files in either a sequential or non-
sequential manner. Unlike the proposed Mail Protocol, FAP is an
extension of FTP and not a subsystem. In general FAP is compatible
with the rest of FTP. Those modifications which are necessary are
The intent of this protocol is to allow processes to specify to the
remote file system where in the file they wish the next operation to
start and how much data to move. Thus only the part of a file
necessary for a process' computation need be transferred, rather than
the entire file. Thus transmission times and storage requirements
may be held down. In short, the rationale for a File Access Protocol
on the network is the same as the rationale for "random-accessed"
files in a standard operating system.
The file Access Protocol uses the connection model, data
representations, and transmission methods of the File transfer
Protocol. All data transmissions in FAP are handled according to the
description in FTP Section III.C with the following modifications.
In Stream mode, the minimum byte size is increased to 4 bits.
Another control code (value 4) is used to indicate "end of
transmission". An combination of EOT, EOR, or EOF may be indicated
by the proper control code. With this method it is not necessary to
close the connection after each access; a practice not highly
recommended. In Block mode, bit 5 of the descriptor field of the
header is set noting that this block is the end of transmission. In
addition to this, FAP uses a File Pointer (FP). The file pointer
points into the file and is the point at which the next FAP read or
write will commence. The file pointer is a general mechanism for
addressing a file and should be flexible enough to handle both stream
and record oriented systems.
II. PROBLEMS OF IMPLEMENTATION
As usual, not all systems will be able to implement this protocol in
its full generality. The approach that should be taken is that no
host should be required to provide for network users (in the name of
complete protocol implementation) service it does not provide its
Some systems allow "random" access to some kinds of files on its
system and not to others. In this case, this should be their
implementation, i.e., not all operations are valid for all kinds of
Some systems cannot move the byte pointer backwards without opening
and closing the file. They should not be required to do this
(although they may if they wish), but they should allow "spacing"
down a file some distance before starting a transfer.
Some systems may not allow read and write access to be available
without closing and reopening the file. Systems should not be
required to do both.
In general, the rules of implementation are:
1) If a system normally allows that particular kind of access to
that particular file then it should be allowed; if not, the system
should not be forced to implement it. (In many cases, the legality
cannot be known until the operation is attempted; i.e., it cannot
be told of the first two cases above if they are legal when the
file is opened but only on the read or write which violates the
2) A system should not try to simulate a facility if the
simulation has side effects. For example, if simulating the
capability of moving the byte pointer to the desired position has
some side effects, then the simulation should be left to the
process accessing the file.
3) All implementors should make known the capabilities of their
implementations via NIC documents.
III. FILE ACCESS PROTOCOL
The FAP extension to FTP includes 6 new commands and the file
pointer. Any implementation requires the file pointer and all six
commands. But, as described above, it is not necessary to implement
the commands in their full generality.
III.1 THE FILE POINTER
The file pointer represents an index or address within the file. The
units by which the index is measured, is "logical byte size" and does
not include any bytes related to transmission or structure. In
particular, for transmission mode Stream and structure Record, the
EOR and EOF markers are not counted. Local transformations on data
must be taken into account. For example, Multics stores CRLF as NL.
In this case, NL counts as two ASCII bytes since it was transmitted
to or will be sent from Multics as CRLF. If transmission Mode is
Image then the logical byte size is taken as the transmission byte
size. There are two commands which operate on the file pointer: 1)
SETP to move the pointer and 2) GETP to find out where it is at.
These are described below in more detail.
The file pointer may take on three classes of values. All may be
mapped to some decimal number. The value B represents the beginning
of the file (Byte 0). The value E represents the end of the file (or
Byte n for a file n bytes long). The byte pointer may also take on
any value between 0 and n.
A file of n bytes
^ 1 2 3 4 n-4 n-3 n-2 n-1 ^
If a file is stored under set of parameters (TYPE, etc.) and
operations are attempted on it under different parameters, the server
does not guarantee that the information will be valid.
III.2.1 OPEN <direction> <pathname>
This command instructs the server to "open" the file <pathname> for
access in the direction specified. The directions are read, R write,
W; or both, B. A read direction implies that the data connection is
from server to user; write, from user to server; and both implies
connections each ways. Functionally, this command corresponds to
RETR or STOR. Therefore, all the FTP parameter commands (TYPE, MODE,
etc.) must be sent before the file is opened. If the direction is
write (W) and the file specified by the pathname does not exist,
there is an implied create with the open. The success of this
create, is, of course, dependent on local access privileges and
possibly whether or not an ALL command was sent. If applicable, the
file created should be of the most general kind of file on which
"random" access is allowed. (This is to allow the largest degree of
compatibility with operations that may follow). This should be
ignored if some site specific command has already specified the kind
of file. This command identifies the file on which subsequent
operations are to be performed. After the file is opened, the file
pointer is at B and any of the other five FAP commands may be sent.
It is acknowledged that some systems cannot open a file for access in
both directions; an error reply 402 should be sent for this response.
258 451 500 504 550
402 454 501 505
434 455 502 506
4550 457 503 507
III.2.2 SETP <argument>
This command causes the file pointer to be set to the number
specified in the argument. This value will be the ordinal number of
the starting position of the next operation. (Byte 0 is the first
byte in the file). The argument may take on two other values besides
<decimal number> : B, for BEGIN, which sets the file pointer at the
beginning of a file (i.e. 0) and E, for END, which sets the file
pointer to the last byte in the file. Two error conditions are
possible. If the argument specifies an illegal change of file
pointer (such as moving it backwards on some systems), then the error
reply 402 should be sent. If the argument attempts to move the file
pointer off the end of the file, then the EOF: <byte number> reply
should be sent with the address of the end of the file (E), and the
file pointer left at E.
This command requests the server to return the value of the file
pointer as a decimal number.
III.2.4 READ <arg>
This command instructs the server to move as many bytes as specified
(of size logical byte size) from the server to the user. The values
the argument may take on are <decimal number> and ALL. ALL is
interpreted as all data from the present position of the file pointer
to the end-of-file. If a read requests more bytes than in the file,
the number of bytes from the present position to the end of file
should be transferred and an EOF: <byte number> response returned
noting the position of the end of file. If the file is Record
structured and a READ requests more bytes than in the record, then
the number of bytes in the record from the file pointer are moved and
the EOR: <byte number> reply is sent noting the end of record. The
action of a READ leaves the file pointer at the position before the
read plus the number of bytes moved, (i.e., updated). The EOF
condition leaves it at E.
III.2.5 WRITe <arg>
This command instructs the server to accept as many bytes as
specified from the user. The result updates the value of the file
pointer. The values the argument may take on are <decimal number> or
ALL. ALL is interpreted as all data from the present position of the
byte pointer to the end-of-file (or beyond). Associated with the
write is an implied "append", if necessary previous information has
been sent (such as allocation) and if the file's access privilege
allow the append. If a write specifies more bytes than there are
between the file pointer and the end-of-file, and expansion is not
allowed, no data is sent and the file pointer is not moved. An error
is returned specifying the byte position of the EOF. If the file is
Record structured and a WRIT attempts to move more bytes than there
are in the record, the file pointer is not moved and the EOR: <byte
number> reply is sent noting the end of record.
This command instructs the server to "close" the presently open file,
if any. The receipt of a CLOS without an open file is not an error.
The effect is to notify the server that further operations are not
directed at the file which is presently open. If an open is received
by the server and it has a file open, it should close the open file
and open the new one.
OPEN <direction> <pathname> CRLF
SETP <byte pointer arg> CRLF
READ <transfer argument> CRLF
WRIT <transfer argument> CRLF
<byte pointer argument>::= B|E|<decimal number>
<transfer argument>::=ALL|<decimal number>
<byte number>::= <decimal number>
IV.2 REPLIES USED BY FAP
258 Operation successful
402 Command not implemented for requested value or action
433 Cannot transfer files w/o valid account. Enter account &
450 FTP: file not found
451 FTP: file access denied
452 FTP: file transfer incomplete, data connection closed.
453 FTP: file transfer incomplete, insufficient storage space.
454 FTP: cannot connect to your data socket
455 FTP: file system error not covered by other reply codes.
457 FTP: transfer parameters in error.
480 EOR: <byte number>
481 EOF: <byte number>
482 File not open for operation
483 FP: <byte pointer>
500 Last command line completely unrecognized.
501 Syntax of last command is incorrect.
502 Last command invalid (ignored), illegal parameter combination.
504 Last command invalid, action not possible at this time.
505 Last command conflicts illegally with previous command(s).
506 Last command not implemented by the server.
507 Catchall error reply.
550 Bad pathname specification (e.g., syntax error).
[ This RFC was put into machine readable form for entry ]
[ into the online RFC archives by Via Genie ]