Copyright (c) 2010 The FreeBSD Foundation
All rights reserved.
This documentation was written by CK Software GmbH under sponsorship from
the FreeBSD Foundation.
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 December 10, 2020 .Dt VNET 9 .Os .Sh NAME .Nm VNET .Nd "network subsystem virtualization infrastructure" .Sh SYNOPSIS .Cd "options VIMAGE" .Cd "options VNET_DEBUG"
p
n net/vnet.h ------------------------------------------------------------
.Ss "Constants and Global Variables"
.Dv VNET_SETNAME
"set_vnet"
.Dv VNET_SYMPREFIX
"vnet_entry_"
.Vt extern struct vnet *vnet0;
------------------------------------------------------------
.Ss "Variable Declaration"
.Fo VNET
.Fa "name"
.Fc
.Fo VNET_NAME
.Fa "name"
.Fc
.Fo VNET_DECLARE
.Fa "type" "name"
.Fc
.Fo VNET_DEFINE
.Fa "type" "name"
.Fc
.Fo VNET_DEFINE_STATIC
.Fa "type" "name"
.Fc
d -literal #define V_name VNET(name)
.Ed
------------------------------------------------------------
.Ss "Virtual Instance Selection"
.Fo CRED_TO_VNET
.Fa "struct ucred *"
.Fc
.Fo TD_TO_VNET
.Fa "struct thread *"
.Fc
.Fo P_TO_VNET
.Fa "struct proc *"
.Fc
.Fo IS_DEFAULT_VNET
.Fa "struct vnet *"
.Fc
.Fo VNET_ASSERT
.Fa exp msg
.Fc
.Fo CURVNET_SET
.Fa "struct vnet *"
.Fc
.Fo CURVNET_SET_QUIET
.Fa "struct vnet *"
.Fc
.Fn CURVNET_RESTORE
.Fo VNET_ITERATOR_DECL
.Fa "struct vnet *"
.Fc
.Fo VNET_FOREACH
.Fa "struct vnet *"
.Fc
------------------------------------------------------------
.Ss "Locking"
.Fn VNET_LIST_RLOCK
.Fn VNET_LIST_RUNLOCK
.Fn VNET_LIST_RLOCK_NOSLEEP
.Fn VNET_LIST_RUNLOCK_NOSLEEP
------------------------------------------------------------
.Ss "Startup and Teardown Functions"
.Ft "struct vnet *"
.Fo vnet_alloc
.Fa void
.Fc
.Ft void
.Fo vnet_destroy
.Fa "struct vnet *"
.Fc
.Fo VNET_SYSINIT
.Fa ident
.Fa "enum sysinit_sub_id subsystem"
.Fa "enum sysinit_elem_order order"
.Fa "sysinit_cfunc_t func"
.Fa "const void *arg"
.Fc
.Fo VNET_SYSUNINIT
.Fa ident
.Fa "enum sysinit_sub_id subsystem"
.Fa "enum sysinit_elem_order order"
.Fa "sysinit_cfunc_t func"
.Fa "const void *arg"
.Fc
------------------------------------------------------------
.Ss "Eventhandlers"
.Fo VNET_GLOBAL_EVENTHANDLER_REGISTER
.Fa "const char *name"
.Fa "void *func"
.Fa "void *arg"
.Fa "int priority"
.Fc
.Fo VNET_GLOBAL_EVENTHANDLER_REGISTER_TAG
.Fa "eventhandler_tag tag"
.Fa "const char *name"
.Fa "void *func"
.Fa "void *arg"
.Fa "int priority"
.Fc
------------------------------------------------------------
.Ss "Sysctl Handling"
.Fo SYSCTL_VNET_INT
.Fa parent nbr name access ptr val descr
.Fc
.Fo SYSCTL_VNET_PROC
.Fa parent nbr name access ptr arg handler fmt descr
.Fc
.Fo SYSCTL_VNET_STRING
.Fa parent nbr name access arg len descr
.Fc
.Fo SYSCTL_VNET_STRUCT
.Fa parent nbr name access ptr type descr
.Fc
.Fo SYSCTL_VNET_UINT
.Fa parent nbr name access ptr val descr
.Fc
.Fo VNET_SYSCTL_ARG
.Fa req arg1
.Fc
------------------------------------------------------------
.Sh DESCRIPTION
.Nm
is the name of a technique to virtualize the network stack.
The basic idea is to change global resources most notably variables into
per network stack resources and have functions, sysctls, eventhandlers,
etc. access and handle them in the context of the correct instance.
Each (virtual) network stack is attached to a
.Em prison ,
with
.Vt vnet0
being the unrestricted default network stack of the base system.
p
The global defines for
.Dv VNET_SETNAME
and
.Dv VNET_SYMPREFIX
are shared with
.Xr kvm 3
to access internals for debugging reasons.
------------------------------------------------------------
.Ss "Variable Declaration"
Variables are virtualized by using the
.Fn VNET_DEFINE
macro rather than writing them out as
.Em type name .
One can still use static initialization, e.g.,
p .Dl Li VNET_DEFINE(int, foo) = 1;
p Variables declared with the static keyword can use the .Fn VNET_DEFINE_STATIC macro, e.g.,
p .Dl Li VNET_DEFINE_STATIC(SLIST_HEAD(, bar), bars);
p Static initialization is not possible when the virtualized variable would need to be referenced, e.g., with .Dq TAILQ_HEAD_INITIALIZER() . In that case a .Fn VNET_SYSINIT based initialization function must be used.
p External variables have to be declared using the .Fn VNET_DECLARE macro. In either case the convention is to define another macro, that is then used throughout the implementation to access that variable. The variable name is usually prefixed by .Em V_ to express that it is virtualized. The .Fn VNET macro will then translate accesses to that variable to the copy of the currently selected instance (see the .Sx "Virtual instance selection" section):
p .Dl Li #define V_name VNET(name)
p .Em NOTE: Do not confuse this with the convention used by .Xr VFS 9 .
p
The
.Fn VNET_NAME
macro returns the offset within the memory region of the virtual network
stack instance.
It is usually only used with
.Fn SYSCTL_VNET_*
macros.
------------------------------------------------------------
.Ss "Virtual Instance Selection"
There are three different places where the current virtual
network stack pointer is stored and can be taken from:
l -enum -offset indent t a
.Em prison :
.Dl "(struct prison *)->pr_vnet"
p For convenience the following macros are provided: d -literal -compact -offset indent .Fn CRED_TO_VNET "struct ucred *" .Fn TD_TO_VNET "struct thread *" .Fn P_TO_VNET "struct proc *" .Ed t a .Em socket : .Dl "(struct socket *)->so_vnet" t an .Em interface : .Dl "(struct ifnet *)->if_vnet" .El
p
In addition the currently active instance is cached in
.Dq "curthread->td_vnet"
which is usually only accessed through the
.Dv curvnet
macro.
p
To set the correct context of the current virtual network instance, use the
.Fn CURVNET_SET
or
.Fn CURVNET_SET_QUIET
macros.
The
.Fn CURVNET_SET_QUIET
version will not record vnet recursions in case the kernel was compiled
with
.Cd "options VNET_DEBUG"
and should thus only be used in well known cases, where recursion is
unavoidable.
Both macros will save the previous state on the stack and it must be restored
with the
.Fn CURVNET_RESTORE
macro.
p .Em NOTE: As the previous state is saved on the stack, you cannot have multiple .Fn CURVNET_SET calls in the same block.
p .Em NOTE: As the previous state is saved on the stack, a .Fn CURVNET_RESTORE call has to be in the same block as the .Fn CURVNET_SET call or in a subblock with the same idea of the saved instances as the outer block.
p .Em NOTE: As each macro is a set of operations and, as previously explained, cannot be put into its own block when defined, one cannot conditionally set the current vnet context. The following will .Em not work: d -literal -offset indent if (condition) CURVNET_SET(vnet); .Ed
p nor would this work: d -literal -offset indent if (condition) { CURVNET_SET(vnet); } CURVNET_RESTORE(); .Ed
p
Sometimes one needs to loop over all virtual instances, for example to update
virtual from global state, to run a function from a
.Xr callout 9
for each instance, etc.
For those cases the
.Fn VNET_ITERATOR_DECL
and
.Fn VNET_FOREACH
macros are provided.
The former macro defines the variable that iterates over the loop,
and the latter loops over all of the virtual network stack instances.
See
.Sx "Locking"
for how to savely traverse the list of all virtual instances.
p
The
.Fn IS_DEFAULT_VNET
macro provides a safe way to check whether the currently active instance is the
unrestricted default network stack of the base system
q Vt vnet0 .
p
The
.Fn VNET_ASSERT
macro provides a way to conditionally add assertions that are only active with
.Cd "options VIMAGE"
compiled in and either
.Cd "options VNET_DEBUG"
or
.Cd "options INVARIANTS"
enabled as well.
It uses the same semantics as
.Xr KASSERT 9 .
------------------------------------------------------------
.Ss "Locking"
For public access to the list of virtual network stack instances
e.g., by the
.Fn VNET_FOREACH
macro, read locks are provided.
Macros are used to abstract from the actual type of the locks.
If a caller may sleep while traversing the list, it must use the
.Fn VNET_LIST_RLOCK
and
.Fn VNET_LIST_RUNLOCK
macros.
Otherwise, the caller can use
.Fn VNET_LIST_RLOCK_NOSLEEP
and
.Fn VNET_LIST_RUNLOCK_NOSLEEP .
------------------------------------------------------------
.Ss "Startup and Teardown Functions"
To start or tear down a virtual network stack instance the internal
functions
.Fn vnet_alloc
and
.Fn vnet_destroy
are provided and called from the jail framework.
They run the publicly provided methods to handle network stack
startup and teardown.
p
For public control, the system startup interface has been enhanced
to not only handle a system boot but to also handle a virtual
network stack startup and teardown.
To the base system the
.Fn VNET_SYSINIT
and
.Fn VNET_SYSUNINIT
macros look exactly as if there were no virtual network stack.
In fact, if
.Cd "options VIMAGE"
is not compiled in they are compiled to the standard
.Fn SYSINIT
macros.
In addition to that they are run for each virtual network stack
when starting or, in reverse order, when shutting down.
------------------------------------------------------------
.Ss "Eventhandlers"
Eventhandlers can be handled in two ways:
p l -enum -offset indent -compact t save the .Em tags returned in each virtual instance and properly free the eventhandlers on teardown using those, or t use one eventhandler that will iterate over all virtual network stack instances. .El
p
For the first case one can just use the normal
.Xr EVENTHANDLER 9
functions, while for the second case the
.Fn VNET_GLOBAL_EVENTHANDLER_REGISTER
and
.Fn VNET_GLOBAL_EVENTHANDLER_REGISTER_TAG
macros are provided.
These differ in that
.Fn VNET_GLOBAL_EVENTHANDLER_REGISTER_TAG
takes an extra first argument that will carry the
.Fa "tag"
upon return.
Eventhandlers registered with either of these will not run
.Fa func
directly but
.Fa func
will be called from an internal iterator function for each vnet.
Both macros can only be used for eventhandlers that do not take
additional arguments, as the variadic arguments from an
.Xr EVENTHANDLER_INVOKE 9
call will be ignored.
------------------------------------------------------------
.Ss "Sysctl Handling"
A
.Xr sysctl 9
can be virtualized by using one of the
.Fn SYSCTL_VNET_*
macros.
p They take the same arguments as the standard .Xr sysctl 9 functions, with the only difference, that the .Fa ptr argument has to be passed as .Ql &VNET_NAME(foo) instead of .Ql &foo so that the variable can be selected from the correct memory region of the virtual network stack instance of the caller.
p
For the very rare case a sysctl handler function would want to
handle
.Fa arg1
itself the
.Fn VNET_SYSCTL_ARG req arg1
is provided that will translate the
.Fa arg1
argument to the correct memory address in the virtual network stack
context of the caller.
------------------------------------------------------------
.Sh SEE ALSO
.Xr jail 2 ,
.Xr kvm 3 ,
.Xr EVENTHANDLER 9 ,
.Xr pcpu 9 ,
.Xr KASSERT 9 ,
.Xr sysctl 9
.Xr SYSINIT 9
p Marko Zec, Implementing a Clonable Network Stack in the FreeBSD Kernel, USENIX ATC'03, June 2003, Boston .Sh HISTORY The virtual network stack implementation first appeared in .Fx 8.0 . .Sh AUTHORS .An -nosplit The .Nm framework was designed and implemented at the University of Zagreb by .An Marko Zec under sponsorship of the FreeBSD Foundation and NLnet Foundation, and later extended and refined by .An Bjoern A. Zeeb (also under FreeBSD Foundation sponsorship), and .An Robert Watson .
p This manual page was written by .An Bjoern A. Zeeb, CK Software GmbH, under sponsorship from the FreeBSD Foundation.