freebsd-dev/share/man/man9/tcp_functions.9
Jonathan T. Looney dc6a41b936 Add the infrastructure to support loading multiple versions of TCP
stack modules.

It adds support for mangling symbols exported by a module by prepending
a string to them. (This avoids overlapping symbols in the kernel linker.)

It allows the use of a macro as the module name in the DECLARE_MACRO()
and MACRO_VERSION() macros.

It allows the code to register stack aliases (e.g. both a generic name
["default"] and version-specific name ["default_10_3p1"]).

With these changes, it is trivial to compile TCP stack modules with
the name defined in the Makefile and to load multiple versions of the
same stack simultaneously. This functionality can be used to enable
side-by-side testing of an old and new version of the same TCP stack.
It also could support upgrading the TCP stack without a reboot.

Reviewed by:	gnn, sjg (makefiles only)
Sponsored by:	Netflix
Differential Revision:	https://reviews.freebsd.org/D11086
2017-06-08 20:41:28 +00:00

392 lines
13 KiB
Groff

.\"
.\" 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
.\" documentation and/or other materials provided with the distribution.
.\"
.\" THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
.\" ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
.\" IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
.\" ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE FOR
.\" ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
.\" DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
.\" OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
.\" HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
.\" LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
.\" OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
.\" SUCH DAMAGE.
.\"
.\" $FreeBSD$
.\"
.Dd March 10, 2017
.Dt TCP_FUNCTIONS 9
.Os
.Sh NAME
.Nm tcp_functions
.Nd Alternate TCP Stack Framework
.Sh 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 register_tcp_functions_as_name "struct tcp_function_block *blk" \
"const char *name" "int wait"
.Fn register_tcp_functions_as_names "struct tcp_function_block *blk" \
"int wait" "const char *names[]" "int *num_names"
.Ft int
.Fn deregister_tcp_functions "struct tcp_function_block *blk"
.Sh DESCRIPTION
The
.Nm
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.
.Pp
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
.Xr tcp 4
for details on setting the system default stack, or selecting a specific stack
for a given connection.)
.Pp
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".
.Pp
The
.Fn register_tcp_functions ,
.Fn register_tcp_functions_as_name ,
and
.Fn register_tcp_functions_as_names
functions request that the system add a specified function block
and register it for use with a given name.
Modules may register the same function block multiple times with different
names.
However, names must be globally unique among all registered function blocks.
Also, modules may not ever modify the contents of the function block (including
the name) after it has been registered, unless the module first successfully
de-registers the function block.
.Pp
The
.Fn register_tcp_functions
function requests that the system register the function block with the name
defined in the function block's
.Va tfb_tcp_block_name
field.
Note that this is the only one of the three registration functions that
automatically registers the function block using the name defined in the
function block's
.Va tfb_tcp_block_name
field.
If a module uses one of the other registration functions, it may request that
the system register the function block using the name defined in the
function block's
.Va tfb_tcp_block_name
field by explicitly providing that name.
.Pp
The
.Fn register_tcp_functions_as_name
function requests that the system register the function block with the name
provided in the
.Fa name
argument.
.Pp
The
.Fn register_tcp_functions_as_names
function requests that the system register the function block with all the
names provided in the
.Fa names
argument.
The
.Fa num_names
argument provides a pointer to the number of names.
This function will either succeed in registering all of the names in the array,
or none of the names in the array.
On failure, the
.Fa num_names
argument is updated with the index number of the entry in the
.Fa names
array which the system was processing when it encountered the error.
.Pp
The
.Fn deregister_tcp_functions
function requests that the system remove a specified function block from the
system.
If this call succeeds, it will completely deregister the function block,
regardless of the number of names used to register the function block.
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.
.Pp
.Nm
modules must call one or more of the registration functions during
initialization and successfully call the
.Fn deregister_tcp_functions
function prior to allowing the module to be unloaded.
.Pp
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
.Xr malloc 9 ,
and must be set to one of the valid values defined in that man page.
.Ss Function Block Structure
The
.Fa blk argument is a pointer to a
.Vt "struct tcp_function_block" ,
which has the following members:
.Bd -literal -offset indent
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 *, int);
/* 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);
/* Optional functions */
void (*tfb_tcp_rexmit_tmr)(struct tcpcb *);
void (*tfb_tcp_handoff_ok)(struct tcpcb *);
/* System use */
volatile uint32_t tfb_refcnt;
uint32_t tfb_flags;
};
.Ed
.Pp
The
.Va 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.
.Pp
The
.Va tfb_tcp_output ,
.Va tfb_tcp_do_segment ,
and
.Va 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.
.Pp
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
.Va 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
.Va tfb_tcp_fb_fini
field.
.Pp
If the
.Va tfb_tcp_fb_fini
argument is non-NULL, the function to which it points is called when the
kernel is destroying the TCP control block or when the socket is transitioning
to use a different TCP stack.
The function is called with arguments of the TCP control block and an integer
flag.
The flag will be zero if the socket is transitioning to use another TCP stack
or one if the TCP control block is being destroyed.
.Pp
If the TCP stack implements additional timers, the TCP stack should set a
non-NULL pointer in the
.Va tfb_tcp_timer_stop_all ,
.Va tfb_tcp_timer_activate ,
.Va tfb_tcp_timer_active ,
and
.Va tfb_tcp_timer_stop
fields.
These fields should all be
.Dv NULL
or should all contain pointers to functions.
The
.Va tfb_tcp_timer_activate ,
.Va tfb_tcp_timer_active ,
and
.Va 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.
.Pp
Additionally, a stack may define its own actions to take when the retransmit
timer fires by setting a non-NULL function pointer in the
.Va 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.
.Pp
A user may select a new TCP stack before calling
.Xr connect 2
or
.Xr listen 2 .
Optionally, a TCP stack may also allow a user to begin using the TCP stack for
a connection that is in a later state by setting a non-NULL function pointer in
the
.Va tfb_tcp_handoff_ok
field.
If this field is non-NULL and a user attempts to select that TCP stack after
calling
.Xr connect 2
or
.Xr listen 2
for that socket, the kernel will call the function pointed to by the
.Va tfb_tcp_handoff_ok
field.
The function should return 0 if the user is allowed to switch the socket to use
the TCP stack. Otherwise, the function should return an error code, which will
be returned to the user.
If the
.Va tfb_tcp_handoff_ok
field is
.Dv NULL
and a user attempts to select the TCP stack after calling
.Xr connect 2
or
.Xr listen 2
for that socket, the operation will fail and the kernel will return
.Er EINVAL .
.Pp
The
.Va tfb_refcnt
and
.Va tfb_flags
fields are used by the kernel's TCP code and will be initialized when the
TCP stack is registered.
.Ss 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
.Va 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
.Va t_fb_ptr
pointer may not be initialized to
.Dv NULL .
Therefore, it should use a
.Va tfb_tcp_fb_init
function to initialize this field.
Additionally, it should use a
.Va tfb_tcp_fb_fini
function to deallocate storage when the socket is closed.
.Pp
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.
.Sh RETURN VALUES
The
.Fn register_tcp_functions ,
.Fn register_tcp_functions_as_name ,
.Fn register_tcp_functions_as_names ,
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.
.Sh ERRORS
The
.Fn register_tcp_functions
function will fail if:
.Bl -tag -width Er
.It Bq Er EINVAL
Any of the members of the
.Fa blk
argument are set incorrectly.
.It Bq Er ENOMEM
The function could not allocate memory for its internal data.
.It Bq Er EALREADY
A function block is already registered with the same name.
.El
The
.Fn deregister_tcp_functions
function will fail if:
.Bl -tag -width Er
.It Bq Er EPERM
The
.Fa blk
argument references the kernel's compiled-in default function block.
.It Bq Er EBUSY
The function block is still in use by one or more sockets, or is defined as
the current default function block.
.It Bq Er ENOENT
The
.Fa blk
argument references a function block that is not currently registered.
.El
.Sh SEE ALSO
.Xr connect 2 ,
.Xr listen 2 ,
.Xr tcp 4 ,
.Xr malloc 9
.Sh HISTORY
This framework first appeared in
.Fx 11.0 .
.Sh AUTHORS
.An -nosplit
The
.Nm
framework was written by
.An Randall Stewart Aq Mt rrs@FreeBSD.org .
.Pp
This manual page was written by
.An Jonathan Looney Aq Mt jtl@FreeBSD.org .