The File Transfer Protocol.
A. Bhushan, B. Braden, W. Crowther, E. Harslem, J. Heafner, A. McKenzie, J. Melvin, B. Sundberg, D. Watson, J. White. June 1971.
Network Working Group 23 June 1971
Request for Comments #172 Abhay Bhushan, MIT
NIC 6794 Bob Braden, UCLA
Categories: D.4, D.5, and D.7 Will Crowther, BBN
Updates: 114 Eric Harslem, Rand
Obsolete: None John Heafner, Rand
Alex McKenzie, BBN
John Melvin, SRI
Bob Sundberg, Harvard
Dick Watson, SRI
Jim White, UCSB
THE FILE TRANSFER PROTOCOL
The file transfer protocol (FTP) is a user-level protocol for file
transfer between host computers (including terminal IMP's), on the ARPA
computer network. The primary function of FTP is to facilitate transfer
of files between hosts, and to allow convenient use of storage and file
handling capabilities of other hosts. FTP uses the data transfer
protocol described in RFC 171 to achieve transfer of data. This paper
assumes knowledge of RFC 171.
The objectives of FTP are to promote sharing of files (computer
programs and/or data), encourage indirect use (without login or
implicit) of computers, and shield the user from variations in file and
storage systems of different hosts, to the extent it is practical.
These objectives are achieved by specifying a standard file transfer
socket and initial connection protocol for indirect use, and using
standard conventions for file transfer and related operations.
A file is considered here to be an ordered set of arbitrary
length, consisting of computer (including instructions) data. Files are
uniquely identified in a system by their pathnames. A pathname is
(loosely) defined to be the data string which must be input to the file
system by a network user in order to identify a file. Pathname usually
contains device and/or directory names, and file names in case of named
files. FTP specifications provide standard file system commands, but do
not provide standard naming convention at this time. Each user must
follow the naming convention of the file system he wishes to use. FTP
may be extended later to include standard conventions for pathname
A file may or may not have access controls associated with it.
The access controls designate the users' access privilege. In the
absence of access controls, the files cannot be protected from
accidental or unauthorized usage. It is the prerogative of a resident
file system to provide protection, and selective access. FTP only
provides identifier and password mechanisms for exchange of access
control information. It should however be noted, that for file sharing,
it is necessary that a user be allowed (subject to access controls) to
access files not created by him.
FTP does not restrict the nature of information in the file. For
example, a file could contain ASCII text, binary data computer program,
or any other information. A provision for indicating data structure
(type and byte size) exists in FTP to aid in parsing, interpretation,
reconfiguration, and storage of data. To facilitate indirect usage, the
cooperating file transfer processes may be disowned "daemon" processes
which "listen" to agreed-upon sockets, and follow the standard initial
connection protocol for establishing a full-duplex connection. It should
be noted that FTP could also used directly by logging into a remote
host, and arranging for file transfer over specific sockets.
FTP is readily extensible, in that additional commands and data
types may be defined by those agreeing to implement them.
Implementation of a subset of commands is specifically permitted, and an
initial subset for implementation is recommended. The protocol may
also be extended to enable remote execution of programs, but no standard
procedure is suggested.
For transferring data, FTP uses the data transfer protocol
specified in RFC 171. As the data transfer protocol does not specify the
manner in which it is to be used by FTP, implementation may vary at
different host sites. Hosts not wishing to separate the data transfer
and file transfer functions, should take particular care in conforming
to the data transfer protocol specifications of RFC 171.
It should be noted, that FTP specifications do not require
knowledge of transfer modes used by data transfer protocol. However, as
file transfer protocol requires the transfer of more than a single
control transaction over the same connection, it is essential that hosts
be able to send control transactions in either 'transparent block' (type
B9) of 'descriptor and counts' (type BA) modes. (Type BB, the indefinite
bit stream mode is not suitable as it limits transfer to singles
The use of data transfer aborts (type B6) is neither required, nor
defined in FTP. FTP has its own error terminate which may be used to
abort a file transfer request. FTP also does not define the structure of
files, and there are no conventions on the use of group, record and unit
separators. A file separator is, however, used to indicate the end of
file. It is strongly recommended that default options be provided in
implementation to facilitate use of file transfer service. For example,
the main file directory on disk, a pool directory, user directory or
directory last accessed could serve as standard pathname defaults.
Default mechanisms are convenient, as the user doesn't have to specify
the complete pathname each time he wishes to use the file transfer
1. Data Transfer
FTP uses the data transfer protocol described in RFC 171, for
transferring data and/or control transaction. Both data and control
transactions are communicated over the same connection.
2. Data Transactions
Data transactions represent the data contained in a file. There is no
data type or byte size information contained in data transactions.
The structure of data is instead communicated via control
transactions. A file may be transferred as one or more data
transactions. The protocol neither specifies nor imposes any
limitations on the structure (record, group, etc) or length of file.
Such limitations may however be imposed by a serving host. The end of
a file may be indicated either by a file separator (as defined in
data transfer protocol), or by closing connection (in type B0). In
particular a serving or using host should not send the ETX, or other
end of file character, unless such a character is part of the data in
file (i.e., not provided by system).
3. Control Transactions
The control transactions may be typified as requests, identifiers,
and terminates. A request fulfillment sequence begins with a request
and ends with receipt of data (followed by End-of-File) or a
3A. Op Codes
Control transactions are distinguished by their first byte referred
as op code. A standard set of opcodes is defined below.
Implementation of a workable subset of opcodes is specifically
permitted. Additional standard opcodes may be assigned later. Opcodes
hex 5A (octal 100) through hex FF (octal 377) are for experimental
Op Code Operation
00 000 Change data type identifier
01 001 Retrieve Request
02 002 Store request (replaced if file already
03 003 Delete request
04 004 Rename_from request
05 005 Rename_to request
06 006 List_file_directory request
07 007 Username identifier
08 010 Password identifier
09 011 Error or unsuccessful terminate
04 012 Acknowledge or successful terminate
0B 013 Append request (add to existing file)
through through Reserved for standard assignment
through through Reserved for experimental use
3B. Syntax and Semantics
3B.1 Data Types
The 'change data type' control transactions identifies the structure
of data (data type and byte size) in succeeding data transactions.
This transaction shall contain two more bytes in addition to the
opcode byte. The first of these bytes shall convey a data type or
code information and the second byte may convey the data byte size,
where applicable. This information may be used to define the manner
in which data is to be parsed, interpreted, reconfigured or stored.
Change data type need be sent only when structure of data is changed
from the preceding.
Although, a number of data types are defined, specific
implementations may handle only limited data types or completely
ignore the data type and byte size descriptors. Even if a host
process does not "recognize" a data type, it must accept data (i.e.,
there is no such thing as a data type error.) These descriptors are
provided only for convenience, and it is not essential that they be
used. The standard default is to assume nothing about the information
and treat it as a bit stream (binary data, byte size 1) whose
interpretation is left to a higher level process, or the user.
* It is, however, possible that this bit stream is treated like
ASCII characters in specific instances such as transmitting a file
to a line printer.
The following data type codes are currently assigned. Where a
byte size is not implicit in data type, it may be provided by the
00 000 1 Bit stream (standard default)
01 001 none Binary data bytes
02 002 8 Network ASCII characters
03 003 8 EBCDIC characters
04 004 36 DEC-packed ASCII (five 7-bit
characters, 36th bit 1 or 0)
05 005 8 Decimal numbers, net. ASCII
06 006 8 Octal numbers, net. ASCII
07 007 8 Hexadecimal numbers, net. ASCII
through through Reserved for standard assignment
through through Reserved for experimental use
3B.2 Requests and Identifiers
Retrieve, delete, name_from, rename_t, and append requests contain a
pathname, following the op code, in the information field. A pathname
may also follow the opcode in list_file_directory request.
A pathname must uniquely identify a file in the serving host. The
syntax of pathnames and identifying information shall conform to
serving host conventions, except that standard network ASCII (as
defined in the TELNET protocol) shall be used.
The store request has a 4-byte (32 bits) 'allocate size' field
followed by pathname. 'Allocate size' indicates the number of bits of
storage to be allocated to the file. A size of zero indicates that
server should use his default.
Retrieve request achieves the transfer of a copy of file specified in
pathname, from serving to using host. The status and contents of file
in serving host should be unaffected.
Store request achieves the transfer of a copy of file specified in
pathname, from using to serving host. If file specified in pathname
exists on serving hosts, then its contents shall be replaced by the
contents of the file being transferred. A new file is created at the
serving host, if the file specified in pathname does not exist.
Append request achieves the transfer of data from using to serving
host. The transferred data is appended to file specified in pathname,
at serving host.
Rename-from and rename-to requests cause the name of the file
specified in pathname of rename_from to be changed to the name
specified in pathname of rename_to. A rename_from must always be
followed by a rename_to request.
Delete request causes file specified in pathname to be deleted from
the serving host. If an extra level of protection is desired such as
the query "Do you really wish to delete this file?", it is to be a
local implementation in the using system. Such queries should not be
transmitted over network connections.
Username and password identifiers contain the respective identifying
information. Normally, the information will be supplied by the user
of the file transfer service. These identifiers are normally sent at
the start of connection.
3B.3 Error and Acknowledge Terminates
The error transactions may have an error code indicated by the second
descriptor byte. Transmission of an error message in text is also
permitted. The following error codes are defined.
Error Code (2nd descriptor byte) Meaning
00 000 Error condition indicated by computer
system (external to protocol)
01 001 Name syntax error
02 003 Access control violation
03 003 Abort
04 004 Allocate size too big
05 005 Allocate size overflow
06 006 Improper order for transactions
07 007 Opcode not implemented
08 010 File search failed
09 011 Error described in text message
(ASCII characters follow code)
At present, no completion codes are defined for acknowledge. It is
assumed that acknowledge refers to the current request being
4. Order of Transactions
4A. A certain order of transactions must be maintained in fulfilling
file transfer requests. The exact sequence in which transactions
occur depends on the type of request, as described in section
4B. The fulfilling of a request may be aborted anytime by either
host, as explained in section 4C.
4B. Identifier transactions (change data type, username, and password)
may be sent by user at any time. The usual order would be a
username transaction followed by a password transaction at the
start of the connection. No acknowledge is required, or
permitted. The identifiers are to be used for default handling,
and access control.
Retrieve and list_file_directory requests cause the transfer of
file from server to user. After a complete file has been
transferred, the server should indicate end-of-file (by sending
CLS or file separator) to complete the request fulfillment
sequence, as shown below.
Read / List_file_directory request
User <File -- data> Server
End of file indication
Store and append requests cause the transfer of file from user to
server. After a complete file has been transferred, the user
should send an end-of-file indication. The receipt of the file
must be acknowledged by the server, as shown below.
User Store / Append request Server
<File -- data>
End of file indication
Rename_from request must be followed by a rename_to request. The
request must be acknowledged as shown below.
User Rename_from request Server
The delete request requires the server to acknowledge it, as shown
User Delete Server
Error transactions may be sent by either host at any time, and
these terminate the current request fulfillment sequence.
4C. Aborts. Either host may abort a request fulfillment sequence at
any time by sending an error terminate, or by closing the
connection (NCP to transmit a CLS for the connection). CLS is a
more drastic type of abort and shall be used when there is a
catastrophic failure, or when abort is desired in the middle of a
long transaction. The abort indicates to the receiving host that
sender of abort wishes to terminate request fulfillment and is now
ready to initiate or fulfill new requests. When CLS is used to
abort, the using host will be responsible for reopening
connection. The file transfer abort described here is different
from the data transfer abort which is sent only by the sender of
data. The use of the data transfer abort is not defined in this
6. Initial Connection, CLS, and Access Control
6A. There will be a preassigned permanent socket number for the
cooperating file transfer process at the serving host. The
connection establishment will be in accordance with the standard
initial connection protocol, establishing a full-duplex
6B. The connection will be broken by trading a CLS between the NCP's
for each of the two connections. Normally, the user will initiate
CLS may also be used by either user or server, to abort a
transaction in the middle. If CLS is received in the middle of
transaction, the current request fulfillment sequence will be
aborted. The using host will then reopen connection.
6C. It is recommended that identifier (user name and password)
transactions be sent by user to server , at the start, as this
would facilitate default handling and access control for the
entire duration of connection. The identifier transactions do not
require or permit and acknowledge, and the user can proceed
directly with requests. If the identifier information is incorrect
or not received, the server may send an error transaction
indicating access control, violation, upon subsequent requests.
 Alex McKenzie, BBN, is conducting a survey of network file systems
to determine the practicality of standard pathname conventions, and to
disseminate information to network users on host file systems.
 This initial subset represents control functions necessary for basic
file transfer operations, and some elementary file manipulation
operations. There is no attempt to provide a data management or complete
file management capability.
 It is possible that we may, at a later date, assign meaning to these
information separators within FTP.
 A workable subset is any request, plus terminates. Identifiers may
be required in addition when using protected file systems.
 It is, however, possible that this bit stream is treated like ASCII
characters in specific instances such as transmitting a file to a line
 It seems that socket 1 has been assigned to logger. Socket 3 seems a
reasonable choice for File Transfer.
 RFC 165, or any subsequent standard applicable in initial connection
[ This RFC was put into machine readable form for entry ]
[ into the online RFC archives by Glenn Forbes Fleming Larratt 5/97 ]