register_tcp_functions (9)

Leading comments

Copyright (c) 2016 Jonathan Looney <jtl@FreeBSD.org>
All rights reserved.

Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions
are met:
1. Redistributions of source code must retain the above copyright
   notice, this list of conditions and the following disclaimer.
2. Redistributions in binary form must reproduce the above copyright
   notice, this list of conditions and the following disclaimer in the
   docum...

(The comments found at the beginning of the groff file "man9/register_tcp_functions.9freebsd".)

NAME

tcp_functions - Alternate TCP Stack Framework

SYNOPSIS

In netinet/tcp.h In netinet/tcp_var.h Ft int Fn register_tcp_functions struct tcp_function_block *blk int wait Ft int Fn deregister_tcp_functions struct tcp_function_block *blk

DESCRIPTION

The framework allows a kernel developer to implement alternate TCP stacks. The alternate stacks can be compiled in the kernel or can be implemented in loadable kernel modules. This functionality is intended to encourage experimentation with the TCP stack and to allow alternate behaviors to be deployed for different TCP connections on a single system.

A system administrator can set a system default stack. By default, all TCP connections will use the system default stack. Additionally, users can specify a particular stack to use on a per-connection basis. (See tcp(4) for details on setting the system default stack, or selecting a specific stack for a given connection.)

This man page treats "TCP stacks" as synonymous with "function blocks". This is intentional. A "TCP stack" is a collection of functions that implement a set of behavior. Therefore, an alternate "function block" defines an alternate "TCP stack".

modules must call the Fn register_tcp_functions function during initialization and successfully call the Fn deregister_tcp_functions function prior to allowing the module to be unloaded.

The Fn register_tcp_functions function requests that the system add a specified function block to the system.

The Fn deregister_tcp_functions function requests that the system remove a specified function block from the system. If the call fails because sockets are still using the specified function block, the system will mark the function block as being in the process of being removed. This will prevent additional sockets from using the specified function block. However, it will not impact sockets that are already using the function block.

The Fa blk argument is a pointer to a Vt struct tcp_function_block , which is explained below (see Sx Function Block Structure ) . The Fa wait argument is used as the Fa flags argument to malloc(9), and must be set to one of the valid values defined in that man page.

Function Block Structure

The Fa blk argument is a pointer to a Vt struct tcp_function_block , which has the following members:

struct tcp_function_block {
        char    tfb_tcp_block_name[TCP_FUNCTION_NAME_LEN_MAX];
        int     (*tfb_tcp_output)(struct tcpcb *);
        void    (*tfb_tcp_do_segment)(struct mbuf *, struct tcphdr *,
                            struct socket *, struct tcpcb *,
                            int, int, uint8_t,
                            int);
        int     (*tfb_tcp_ctloutput)(struct socket *so,
                            struct sockopt *sopt,
                            struct inpcb *inp, struct tcpcb *tp);
        /* Optional memory allocation/free routine */
        void    (*tfb_tcp_fb_init)(struct tcpcb *);
        void    (*tfb_tcp_fb_fini)(struct tcpcb *);
        /* Optional timers, must define all if you define one */
        int     (*tfb_tcp_timer_stop_all)(struct tcpcb *);
        void    (*tfb_tcp_timer_activate)(struct tcpcb *,
                            uint32_t, u_int);
        int     (*tfb_tcp_timer_active)(struct tcpcb *, uint32_t);
        void    (*tfb_tcp_timer_stop)(struct tcpcb *, uint32_t);
        void    (*tfb_tcp_rexmit_tmr)(struct tcpcb *);
        volatile uint32_t tfb_refcnt;
        uint32_t  tfb_flags;
};

The tfb_tcp_block_name field identifies the unique name of the TCP stack, and should be no longer than TCP_FUNCTION_NAME_LEN_MAX-1 characters in length.

The tfb_tcp_output tfb_tcp_do_segment and tfb_tcp_ctloutput fields are pointers to functions that perform the equivalent actions as the default Fn tcp_output , Fn tcp_do_segment , and Fn tcp_default_ctloutput functions, respectively. Each of these function pointers must be non-NULL.

If a TCP stack needs to initialize data when a socket first selects the TCP stack (or, when the socket is first opened), it should set a non-NULL pointer in the tfb_tcp_fb_init field. Likewise, if a TCP stack needs to cleanup data when a socket stops using the TCP stack (or, when the socket is closed), it should set a non-NULL pointer in the tfb_tcp_fb_fini field.

If the TCP stack implements additional timers, the TCP stack should set a non-NULL pointer in the tfb_tcp_timer_stop_all tfb_tcp_timer_activate tfb_tcp_timer_active and tfb_tcp_timer_stop fields. These fields should all be NULL or should all contain pointers to functions. The tfb_tcp_timer_activate tfb_tcp_timer_active and tfb_tcp_timer_stop functions will be called when the Fn tcp_timer_activate , Fn tcp_timer_active , and Fn tcp_timer_stop functions, respectively, are called with a timer type other than the standard types. The functions defined by the TCP stack have the same semantics (both for arguments and return values) as the normal timer functions they supplement.

Additionally, a stack may define its own actions to take when the retransmit timer fires by setting a non-NULL function pointer in the tfb_tcp_rexmit_tmr field. This function is called very early in the process of handling a retransmit timer. However, care must be taken to ensure the retransmit timer leaves the TCP control block in a valid state for the remainder of the retransmit timer logic.

The tfb_refcnt and tfb_flags fields are used by the kernel's TCP code and will be initialized when the TCP stack is registered.

Requirements for Alternate TCP Stacks

If the TCP stack needs to store data beyond what is stored in the default TCP control block, the TCP stack can initialize its own per-connection storage. The t_fb_ptr field in the Vt struct tcpcb control block structure has been reserved to hold a pointer to this per-connection storage. If the TCP stack uses this alternate storage, it should understand that the value of the t_fb_ptr pointer may not be initialized to NULL Therefore, it should use a tfb_tcp_fb_init function to initialize this field. Additionally, it should use a tfb_tcp_fb_fini function to deallocate storage when the socket is closed.

It is understood that alternate TCP stacks may keep different sets of data. However, in order to ensure that data is available to both the user and the rest of the system in a standardized format, alternate TCP stacks must update all fields in the TCP control block to the greatest extent practical.

RETURN VALUES

The Fn register_tcp_functions and Fn deregister_tcp_functions functions return zero on success and non-zero on failure. In particular, the Fn deregister_tcp_functions will return Er EBUSY until no more connections are using the specified TCP stack. A module calling Fn deregister_tcp_functions must be prepared to wait until all connections have stopped using the specified TCP stack.

ERRORS

The Fn register_tcp_functions function will fail if:

Bq Er EINVAL

Any of the members of the Fa blk argument are set incorrectly.

Bq Er ENOMEM

The function could not allocate memory for its internal data.

Bq Er EALREADY

A function block is already registered with the same name.

The Fn deregister_tcp_functions function will fail if:

Bq Er EPERM

The Fa blk argument references the kernel's compiled-in default function block.

Bq Er EBUSY

The function block is still in use by one or more sockets, or is defined as the current default function block.

Bq Er ENOENT

The Fa blk argument references a function block that is not currently registered.

SEE ALSO

malloc(9), tcp(4)

HISTORY

This framework first appeared in Fx 11.0 .

AUTHORS

An -nosplit The framework was written by An Randall Stewart Aq Mt rrs@FreeBSD.org .

This manual page was written by An Jonathan Looney Aq Mt jtl@FreeBSD.org .