O-SAFT / fetchfile
With the server protocol extension O-SAFT (Offer Simple Asynchronous File
Transfer) and the matching client fetchfile there is an easy method of
retrieving files from a SAFT server. This is a direct analogy to the SMTP and
POP or APOP protocol suite in the world of e-mail transfer.
- How does O-SAFT/fetchfile work?
- What to do on the client side?
- What to do on the server side?
- How about security issues?
O-SAFT is an extension to the existing SAFT protocol and allows athenticated
clients to retrieve files from a (remote) server. The implemention is the
server sendfiled and the client fetchfile.
O-SAFT uses a dedicated pgp key pair to authenticate the fetchfile session. The
private key will be kept on the client side, the public key must tbe present
at the server side. For security reasons this will NOT be your regular e-mail
pgp key pair, but a separate pair of pgp keys, uniquely assigned for fetchfile
transfers. You will have to create a pair of pgp keys for this purpose befor
using the fetchfile client for the first time (see below).
Fetchfile can provide a directory listing of available files from the server,
retrieve files or delete files. After retrieving a file, it will be placed in
the regular spool directory, not in the current directory! You will have to
use the receive command to transfer the files from the spool directory to your
current directory afterwards.
If there already exists a regular sendfile spool directory /var/spool/sendfile
on the client side it will be used, otherwise a $HOME/.sfspool will be
created. Fetchfile will be running without using root permissions on the
You must have pgp-2.6.x installed and the binaries must be available through
your $PATH environment variable.
First, and ONLY ONCE before using fetchfile the very first time, you have to
create a fetchfile pgp key pair (only pgp-2.6.x is supported!):
- fetchfile -I
Please only hit 'ENTER' when being asked for a pass phrase! This will create a
special non-passphrase protected key pair for O-SAFT.
After this initialization you will have a file
/var/spool/sendfile/$USER/config/public.pgp resp. $HOME/.sfspool/public.pgp
Please send this file to root@SAFT-server, who has to save this public key file
into the appropiate user configuration directory.
sendfile -c 'my O-SAFT puplic key'
(This prelimary action will enable you to use the SAFT server and will prevent
othes from abusing your name or SAFT-account on the server.)
After preparing the pgp keys an both sides, you can invoke fetchfile on a
- fetchfile -l
- list files on the server
- fetchfile -a
- retrieve all files from server
- fetchfile -daf *aol.com
- delete all files from the AOL domain
There is a detailed description of all capabilities in the fetchfile(1) man
For configuring the server SAFT account by the client user there are two
Using this the two local configuration files will be transfered from the local
current directory to the SAFT server. The details of the configuration can be
found in the sendfile(1) man page.
the files will be retrieved back and will be displayed to STDOUT.
pgp-2.6.x must be installed. The system adminsitrator needs to run sfdconf -e
add set the following option:
- fetchfile = on
The system administrator must create a user account (if it does not yet exist).
This account does not need an interactive login shell and does not need a
valid password; the login shell could be /bin/false. The only purpose is to
enable the sendfiled to check out the user and to create a local spool
directory (this method is well known for creating POP mail accounts).
The client user will create the initial pgp key pair and the public key
(public.pgp) will be sent to the system administrator of the server. This key
has to be placed into the config directory for the particular user. Assuming
the user name is bozo, the system administrator will have to type the
following (under root permissions):
receive -f bozo@* -b bozo public.pgp
(the first receive resends the file public.pgp from the sender bozo@* to the
local user bozo)
O-SAFT uses a tcp challenge/response authentication with a pgp signature. This
opens the possibility that the session can be attacked through tcp hijacking.
We are well aware of this, but tcp hijacking is not easy and only possible if
the attacker has direct access to the transport media (e.g. listening on the
same ethernet cable/segment) and has access to a set of pretty nice cracker
tools. With regular operating system supplied software it is not possible to
attack a session.
Ulli Horlacher - email@example.com
translated by firstname.lastname@example.org