tftp

ServiceConfig = Options

Options = [option()]

Most of the options are common for both the client and the server side, but some of them differs a little. The available option()s are as follows:

{debug, Level}

Level = none | error | warning | brief | normal | verbose | all

Controls the level of debug printouts. Default is none.

{host, Host}

Host = hostname(), see inet(3).

The name or IP address of the host where the TFTP daemon resides. This option is only used by the client.

{port, Port}

Port = int()

The TFTP port where the daemon listens. Defaults is the standardized number 69. On the server side, it can sometimes make sense to set it to 0, meaning that the daemon just picks a free port (which one is returned by function info/1).

If a socket is connected already, option {udp, [{fd, integer()}]} can be used to pass the open file descriptor to gen_udp. This can be automated by using a command-line argument stating the prebound file descriptor number. For example, if the port is 69 and file descriptor 22 is opened by setuid_socket_wrap, the command-line argument "-tftpd_69 22" triggers the prebound file descriptor 22 to be used instead of opening port 69. The UDP option {udp, [{fd, 22}]} is automatically added. See init:get_argument/ about command-line arguments and gen_udp:open/2 about UDP options.

{port_policy, Policy}

Policy = random | Port | {range, MinPort, MaxPort}

Port = MinPort = MaxPort = int()

Policy for the selection of the temporary port that is used by the server/client during the file transfer. Default is random, which is the standardized policy. With this policy a randomized free port is used. A single port or a range of ports can be useful if the protocol passes through a firewall.

{udp, Options}

Options = [Opt], see gen_udp:open/2.

{use_tsize, Bool}

Bool = bool()

Flag for automated use of option tsize. With this set to true, the write_file/3 client determines the filesize and sends it to the server as the standardized tsize option. A read_file/3 client acquires only a filesize from the server by sending a zero tsize.

{max_tsize, MaxTsize}

MaxTsize = int() | infinity

Threshold for the maximal filesize in bytes. The transfer is aborted if the limit is exceeded. Default is infinity.

{max_conn, MaxConn}

MaxConn = int() | infinity

Threshold for the maximal number of active connections. The daemon rejects the setup of new connections if the limit is exceeded. Default is infinity.

{TftpKey, TftpVal}

TftpKey = string()
TftpVal = string()

Name and value of a TFTP option.

{reject, Feature}

Feature = Mode | TftpKey
 Mode = read | write
 TftpKey = string()

Controls which features to reject. This is mostly useful for the server as it can restrict the use of certain TFTP options or read/write access.

{callback, {RegExp, Module, State}}

RegExp = string()
Module = atom()
State = term()

Registration of a callback module. When a file is to be transferred, its local filename is matched to the regular expressions of the registered callbacks. The first matching callback is used during the transfer. See read_file/3 and write_file/3.

The callback module must implement the tftp behavior, see CALLBACK FUNCTIONS.

{logger, Module}

Module = module()()

Callback module for customized logging of errors, warnings, and info messages. The callback module must implement the tftp_logger behavior, see LOGGER FUNCTIONS. The default module is tftp_logger.

{max_retries, MaxRetries}

MaxRetries = int()

Threshold for the maximal number of retries. By default the server/client tries to resend a message up to five times when the time-out expires.

A tftp callback module is to be implemented as a tftp behavior and export the functions listed in the following.

On the server side, the callback interaction starts with a call to open/5 with the registered initial callback state. open/5 is expected to open the (virtual) file. Then either function read/1 or write/2 is invoked repeatedly, once per transferred block. At each function call, the state returned from the previous call is obtained. When the last block is encountered, function read/1 or write/2 is expected to close the (virtual) file and return its last state. Function abort/3 is only used in error situations. Function prepare/5 is not used on the server side.

On the client side, the callback interaction is the same, but it starts and ends a bit differently. It starts with a call to prepare/5 with the same arguments as open/5 takes. prepare/5 is expected to validate the TFTP options suggested by the user and to return the subset of them that it accepts. Then the options are sent to the server, which performs the same TFTP option negotiation procedure. The options that are accepted by the server are forwarded to function open/5 on the client side. On the client side, function open/5 must accept all option as-is or reject the transfer. Then the callback interaction follows the same pattern as described for the server side. When the last block is encountered in read/1 or write/2, the returned state is forwarded to the user and returned from read_file/3 or write_file/3.

If a callback (performing the file access in the TFTP server) takes too long time (more than the double TFTP time-out), the server aborts the connection and sends an error reply to the client. This implies that the server releases resources attached to the connection faster than before. The server simply assumes that the client has given up.

If the TFTP server receives yet another request from the same client (same host and port) while it already has an active connection to the client, it ignores the new request if the request is equal to the first one (same filename and options). This implies that the (new) client will be served by the already ongoing connection on the server side. By not setting up yet another connection, in parallel with the ongoing one, the server consumes less resources.

A tftp_logger callback module is to be implemented as a tftp_logger behavior and export the following functions: