[PATCH] socket: Add AF_NETLINK support.

[PATCH] socket: Add AF_NETLINK support.

[David Thompson]

[-- Attachment #1: Type: text/plain, Size: 502 bytes --]

Hello Guile hackers,

This is my first attempt at providing support for AF_NETLINK sockets in
Guile, based on a preliminary patch that Ludovic sent me.  Netlink
sockets are supported only on Linux.  The ultimate goal of this work is
to be able to create virtual ethernet devices from Scheme code.

There are no tests yet and I'm not sure if I went overboard by defining
all of the numerous netlink families and multicast group constants.
Feedback requested! :)

This patch is for stable-2.0.

Thanks,


[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: 0001-socket-Add-AF_NETLINK-support.patch --]
[-- Type: text/x-diff, Size: 13028 bytes --]

From 3fed7633ee4556907408c55ae29c12641ad6d80c Mon Sep 17 00:00:00 2001
From: David Thompson <dthompson2@worcester.edu>
Date: Sun, 3 May 2015 09:35:49 -0400
Subject: [PATCH] socket: Add AF_NETLINK support.

* libguile/socket.c (scm_fill_sockaddr, _scm_from_sockaddr,
  scm_to_sockaddr): Add AF_NETLINK cases.
  (AF_NETLINK, PF_NETLINK, NETLINK_ROUTE, NETLINK_USERSOCK,
  NETLINK_FIREWALL, NETLINK_SOCK_DIAG, NETLINK_NFLOG, NETLINK_XFRM,
  NETLINK_SELINUX, NETLINK_ISCSI, NETLINK_AUDIT, NETLINK_FIB_LOOKUP,
  NETLINK_CONNECTOR, NETLINK_NETFILTER, NETLINK_IP6_FW, NETLINK_DNRTMSG,
  NETLINK_UEVENT, NETLINK_GENERIC, NETLINK_SCSITRANSPORT,
  NETLINK_ECRYPTFS, NETLINK_RDMA, NETLINK_CRYPTO, RTNLGRP_NONE,
  RTNGRP_LINK, RTNGRP_NOTIFY, RTNGRP_NEIGH, RTNGRP_TC,
  RTNGRP_IPV4_IFADDR, RTNGRP_IPV4_MROUTE, RTNGRP_IPV4_ROUTE,
  RTNGRP_IPV4_RULE, RTNGRP_IPV6_IFADDR, RTNGRP_IPV6_MROUTE,
  RTNGRP_IPV6_ROUTE, RTNGRP_IPV6_IFINFO, RTNGRP_DECnet_IFADDR,
  RTNGRP_DECnet_ROUTE, RTNGRP_DECnet_RULE, RTNGRP_IPV6_PREFIX,
  RTNGRP_IPV6_RULE, RTNGRP_ND_USEROPT, RTNGRP_PHONET_IFADDR,
  RTNGRP_PHONET_ROUTE, RTNGRP_DCB, RTNGRP_IPV4_NETCONF,
  RTNGRP_IPV6_NETCONF, RTNGRP_MDB): New Scheme variables.
* module/ice-9/networking.scm (sockaddr:pid, sockaddr:groups): New
  procedures.
* doc/ref/posix.texi (make-socket-address, connect, bind): Document
  AF_NETLINK usage.
  (sockaddr:pid, sockaddr:groups): Document new procedures.
  (socket): Document PF_NETLINK family constant.
---
 doc/ref/posix.texi          |  17 ++++
 libguile/socket.c           | 203 ++++++++++++++++++++++++++++++++++++++++++++
 module/ice-9/networking.scm |   2 +
 3 files changed, 222 insertions(+)

diff --git a/doc/ref/posix.texi b/doc/ref/posix.texi
index ad5460c..33fe185 100644
--- a/doc/ref/posix.texi
+++ b/doc/ref/posix.texi
@@ -2950,6 +2950,7 @@ created with,
 @deffn {Scheme Procedure} make-socket-address AF_INET ipv4addr port
 @deffnx {Scheme Procedure} make-socket-address AF_INET6 ipv6addr port [flowinfo [scopeid]]
 @deffnx {Scheme Procedure} make-socket-address AF_UNIX path
+@deffnx {Scheme Procedure} make-socket-address AF_NETLINK pid groups
 @deffnx {C Function} scm_make_socket_address (family, address, arglist)
 Return a new socket address object.  The first argument is the address
 family, one of the @code{AF} constants, then the arguments vary
@@ -2964,6 +2965,9 @@ arguments may be given (both integers, default 0).
 
 For @code{AF_UNIX} the argument is a filename (a string).
 
+For @code{AF_NETLINK}, the arguments are a process ID and a multicast
+groups bitmask.  This socket type is only available on Linux.
+
 The C function @code{scm_make_socket_address} takes the @var{family}
 and @var{address} arguments directly, then @var{arglist} is a list of
 further arguments, being the port for IPv4, port and optional flowinfo
@@ -3004,6 +3008,16 @@ For an @code{AF_INET6} socket address object @var{sa}, return the
 scope ID value.
 @end deffn
 
+@deffn {Scheme Procedure} sockaddr:pid sa
+For an @code{AF_NETLINK} socket address object @var{sa}, return the
+process ID number.
+@end deffn
+
+@deffn {Scheme Procedure} sockaddr:groups sa
+For an @code{AF_NETLINK} socket address object @var{sa}, return the
+multicast groups bitmask.
+@end deffn
+
 @tpindex @code{struct sockaddr}
 @tpindex @code{sockaddr}
 The functions below convert to and from the C @code{struct sockaddr}
@@ -3075,6 +3089,7 @@ the system,
 @defvar PF_UNIX
 @defvarx PF_INET
 @defvarx PF_INET6
+@defvarx PF_NETLINK
 @end defvar
 
 The possible values for @var{style} are as follows, again where
@@ -3213,6 +3228,7 @@ The return value is unspecified.
 @deffnx {Scheme Procedure} connect sock AF_INET ipv4addr port
 @deffnx {Scheme Procedure} connect sock AF_INET6 ipv6addr port [flowinfo [scopeid]]
 @deffnx {Scheme Procedure} connect sock AF_UNIX path
+@deffnx {Scheme Procedure} connect sock AF_NETLINK pid groups
 @deffnx {C Function} scm_connect (sock, fam, address, args)
 Initiate a connection on socket port @var{sock} to a given address.
 The destination is either a socket address object, or arguments the
@@ -3229,6 +3245,7 @@ same as @code{make-socket-address} would take to make such an object
 @deffnx {Scheme Procedure} bind sock AF_INET ipv4addr port
 @deffnx {Scheme Procedure} bind sock AF_INET6 ipv6addr port [flowinfo [scopeid]]
 @deffnx {Scheme Procedure} bind sock AF_UNIX path
+@deffnx {Scheme Procedure} bind sock AF_NETLINK pid groups
 @deffnx {C Function} scm_bind (sock, fam, address, args)
 Bind socket port @var{sock} to the given address.  The address is
 either a socket address object, or arguments the same as
diff --git a/libguile/socket.c b/libguile/socket.c
index 5b17a74..3501646 100644
--- a/libguile/socket.c
+++ b/libguile/socket.c
@@ -43,6 +43,11 @@
 #include <netdb.h>
 #include <arpa/inet.h>
 
+#ifdef __linux__
+#include <linux/netlink.h>
+#include <linux/rtnetlink.h>
+#endif
+
 #include <gmp.h>
 
 #include "libguile/_scm.h"
@@ -802,6 +807,22 @@ scm_fill_sockaddr (int fam, SCM address, SCM *args, int which_arg,
 	return (struct sockaddr *) soka;
       }
 #endif
+#ifdef __linux__
+    case AF_NETLINK:
+      {
+        struct sockaddr_nl *soka;
+
+        soka = (struct sockaddr_nl *) scm_malloc (sizeof (struct sockaddr_nl));
+
+        soka->nl_family = AF_NETLINK;
+        soka->nl_pad    = 0;
+        soka->nl_pid    = 0;
+        soka->nl_groups = 0;
+
+        *size = sizeof (struct sockaddr_nl);
+        return (struct sockaddr *) soka;
+      }
+#endif
 #ifdef HAVE_UNIX_DOMAIN_SOCKETS
     case AF_UNIX:
       {
@@ -1024,6 +1045,19 @@ _scm_from_sockaddr (const scm_t_max_sockaddr *address, unsigned addr_size,
       }
       break;
 #endif
+#ifdef AF_NETLINK
+    case AF_NETLINK:
+      {
+        const struct sockaddr_nl *nad = (struct sockaddr_nl *) address;
+
+        result = scm_c_make_vector (3, SCM_UNSPECIFIED);
+        SCM_SIMPLE_VECTOR_SET(result, 0, scm_from_short (fam));
+        SCM_SIMPLE_VECTOR_SET(result, 1, scm_from_uint32 (nad->nl_pid));
+        SCM_SIMPLE_VECTOR_SET(result, 2, scm_from_uint32 (nad->nl_groups));
+
+        break;
+      }
+#endif
 #ifdef HAVE_UNIX_DOMAIN_SOCKETS
     case AF_UNIX:
       {
@@ -1140,6 +1174,30 @@ scm_to_sockaddr (SCM address, size_t *address_size)
       }
 #endif
 
+#ifdef __linux__
+    case AF_NETLINK:
+      {
+	if (SCM_SIMPLE_VECTOR_LENGTH (address) != 3)
+	  scm_misc_error (FUNC_NAME, "invalid netlink address representation: ~A",
+			  scm_list_1 (address));
+	else
+	  {
+	    struct sockaddr_nl c_nl;
+
+            c_nl.nl_family = AF_NETLINK;
+            c_nl.nl_pad = 0;
+            c_nl.nl_pid = scm_to_uint32 (SCM_SIMPLE_VECTOR_REF (address, 1));
+            c_nl.nl_groups = scm_to_uint32 (SCM_SIMPLE_VECTOR_REF (address, 2));
+
+	    *address_size = sizeof (c_nl);
+	    c_address = scm_malloc (sizeof (c_nl));
+	    memcpy (c_address, &c_nl, sizeof (c_nl));
+	  }
+
+	break;
+      }
+#endif
+
 #ifdef HAVE_UNIX_DOMAIN_SOCKETS
     case AF_UNIX:
       {
@@ -1699,6 +1757,9 @@ scm_init_socket ()
 #ifdef AF_INET6
   scm_c_define ("AF_INET6", scm_from_int (AF_INET6));
 #endif
+#ifdef AF_NETLINK
+  scm_c_define ("AF_NETLINK", scm_from_int (AF_NETLINK));
+#endif
 
 #ifdef PF_UNSPEC
   scm_c_define ("PF_UNSPEC", scm_from_int (PF_UNSPEC));
@@ -1712,6 +1773,71 @@ scm_init_socket ()
 #ifdef PF_INET6
   scm_c_define ("PF_INET6", scm_from_int (PF_INET6));
 #endif
+#ifdef PF_NETLINK
+  scm_c_define ("PF_NETLINK", scm_from_int (PF_NETLINK));
+#endif
+
+  /* netlink families.  */
+#ifdef NETLINK_ROUTE
+  scm_c_define ("NETLINK_ROUTE", scm_from_int (NETLINK_ROUTE));
+#endif
+#ifdef NETLINK_USERSOCK
+  scm_c_define ("NETLINK_USERSOCK", scm_from_int (NETLINK_USERSOCK));
+#endif
+#ifdef NETLINK_FIREWALL
+  scm_c_define ("NETLINK_FIREWALL", scm_from_int (NETLINK_FIREWALL));
+#endif
+#ifdef NETLINK_SOCK_DIAG
+  scm_c_define ("NETLINK_SOCK_DIAG", scm_from_int (NETLINK_SOCK_DIAG));
+#endif
+#ifdef NETLINK_NFLOG
+  scm_c_define ("NETLINK_NFLOG", scm_from_int (NETLINK_NFLOG));
+#endif
+#ifdef NETLINK_XFRM
+  scm_c_define ("NETLINK_XFRM", scm_from_int (NETLINK_XFRM));
+#endif
+#ifdef NETLINK_SELINUX
+  scm_c_define ("NETLINK_SELINUX", scm_from_int (NETLINK_SELINUX));
+#endif
+#ifdef NETLINK_ISCSI
+  scm_c_define ("NETLINK_ISCSI", scm_from_int (NETLINK_ISCSI));
+#endif
+#ifdef NETLINK_AUDIT
+  scm_c_define ("NETLINK_AUDIT", scm_from_int (NETLINK_AUDIT));
+#endif
+#ifdef NETLINK_FIB_LOOKUP
+  scm_c_define ("NETLINK_FIB_LOOKUP", scm_from_int (NETLINK_FIB_LOOKUP));
+#endif
+#ifdef NETLINK_CONNECTOR
+  scm_c_define ("NETLINK_CONNECTOR", scm_from_int (NETLINK_CONNECTOR));
+#endif
+#ifdef NETLINK_NETFILTER
+  scm_c_define ("NETLINK_NETFILTER", scm_from_int (NETLINK_NETFILTER));
+#endif
+#ifdef NETLINK_IP6_FW
+  scm_c_define ("NETLINK_IP6_FW", scm_from_int (NETLINK_IP6_FW));
+#endif
+#ifdef NETLINK_DNRTMSG
+  scm_c_define ("NETLINK_DNRTMSG", scm_from_int (NETLINK_DNRTMSG));
+#endif
+#ifdef NETLINK_KOBJECT_UEVENT
+  scm_c_define ("NETLINK_KOBJECT_UEVENT", scm_from_int (NETLINK_KOBJECT_UEVENT));
+#endif
+#ifdef NETLINK_GENERIC
+  scm_c_define ("NETLINK_GENERIC", scm_from_int (NETLINK_GENERIC));
+#endif
+#ifdef NETLINK_SCSITRANSPORT
+  scm_c_define ("NETLINK_SCSITRANSPORT", scm_from_int (NETLINK_SCSITRANSPORT));
+#endif
+#ifdef NETLINK_ECRYPTFS
+  scm_c_define ("NETLINK_ECRYPTFS", scm_from_int (NETLINK_ECRYPTFS));
+#endif
+#ifdef NETLINK_RDMA
+  scm_c_define ("NETLINK_RDMA", scm_from_int (NETLINK_RDMA));
+#endif
+#ifdef NETLINK_CRYPTO
+  scm_c_define ("NETLINK_CRYPTO", scm_from_int (NETLINK_CRYPTO));
+#endif
 
   /* standard addresses.  */
 #ifdef INADDR_ANY
@@ -1843,6 +1969,83 @@ scm_init_socket ()
   scm_c_define ("IP_MULTICAST_IF", scm_from_int ( IP_MULTICAST_IF));
 #endif
 
+  /* netlink multicast groups */
+#ifdef RTNLGRP_NONE
+  scm_c_define ("RTNLGRP_NONE", scm_from_int (RTNLGRP_NONE));
+#endif
+#ifdef RTNLGRP_LINK
+  scm_c_define ("RTNLGRP_LINK", scm_from_int (RTNLGRP_LINK));
+#endif
+#ifdef RTNLGRP_NOTIFY
+  scm_c_define ("RTNLGRP_NOTIFY", scm_from_int (RTNLGRP_NOTIFY));
+#endif
+#ifdef RTNLGRP_NEIGH
+  scm_c_define ("RTNLGRP_NEIGH", scm_from_int (RTNLGRP_NEIGH));
+#endif
+#ifdef RTNLGRP_TC
+  scm_c_define ("RTNLGRP_TC", scm_from_int (RTNLGRP_TC));
+#endif
+#ifdef RTNLGRP_IPV4_IFADDR
+  scm_c_define ("RTNLGRP_IPV4_IFADDR", scm_from_int (RTNLGRP_IPV4_IFADDR));
+#endif
+#ifdef RTNLGRP_IPV4_MROUTE
+  scm_c_define ("RTNLGRP_IPV4_MROUTE", scm_from_int (RTNLGRP_IPV4_MROUTE));
+#endif
+#ifdef RTNLGRP_IPV4_ROUTE
+  scm_c_define ("RTNLGRP_IPV4_ROUTE", scm_from_int (RTNLGRP_IPV4_ROUTE));
+#endif
+#ifdef RTNLGRP_IPV4_RULE
+  scm_c_define ("RTNLGRP_IPV4_RULE", scm_from_int (RTNLGRP_IPV4_RULE));
+#endif
+#ifdef RTNLGRP_IPV6_IFADDR
+  scm_c_define ("RTNLGRP_IPV6_IFADDR", scm_from_int (RTNLGRP_IPV6_IFADDR));
+#endif
+#ifdef RTNLGRP_IPV6_MROUTE
+  scm_c_define ("RTNLGRP_IPV6_MROUTE", scm_from_int (RTNLGRP_IPV6_MROUTE));
+#endif
+#ifdef RTNLGRP_IPV6_ROUTE
+  scm_c_define ("RTNLGRP_IPV6_ROUTE", scm_from_int (RTNLGRP_IPV6_ROUTE));
+#endif
+#ifdef RTNLGRP_IPV6_IFINFO
+  scm_c_define ("RTNLGRP_IPV6_IFINFO", scm_from_int (RTNLGRP_IPV6_IFINFO));
+#endif
+#ifdef RTNLGRP_DECnet_IFADDR
+  scm_c_define ("RTNLGRP_DECnet_IFADDR", scm_from_int (RTNLGRP_DECnet_IFADDR));
+#endif
+#ifdef RTNLGRP_DECnet_ROUTE
+  scm_c_define ("RTNLGRP_DECnet_ROUTE", scm_from_int (RTNLGRP_DECnet_ROUTE));
+#endif
+#ifdef RTNLGRP_DECnet_RULE
+  scm_c_define ("RTNLGRP_DECnet_RULE", scm_from_int (RTNLGRP_DECnet_RULE));
+#endif
+#ifdef RTNLGRP_IPV6_PREFIX
+  scm_c_define ("RTNLGRP_IPV6_PREFIX", scm_from_int (RTNLGRP_IPV6_PREFIX));
+#endif
+#ifdef RTNLGRP_IPV6_RULE
+  scm_c_define ("RTNLGRP_IPV6_RULE", scm_from_int (RTNLGRP_IPV6_RULE));
+#endif
+#ifdef RTNLGRP_ND_USEROPT
+  scm_c_define ("RTNLGRP_ND_USEROPT", scm_from_int (RTNLGRP_ND_USEROPT));
+#endif
+#ifdef RTNLGRP_PHONET_IFADDR
+  scm_c_define ("RTNLGRP_PHONET_IFADDR", scm_from_int (RTNLGRP_PHONET_IFADDR));
+#endif
+#ifdef RTNLGRP_PHONET_ROUTE
+  scm_c_define ("RTNLGRP_PHONET_ROUTE", scm_from_int (RTNLGRP_PHONET_ROUTE));
+#endif
+#ifdef RTNLGRP_DCB
+  scm_c_define ("RTNLGRP_DCB", scm_from_int (RTNLGRP_DCB));
+#endif
+#ifdef RTNLGRP_IPV4_NETCONF
+  scm_c_define ("RTNLGRP_IPV4_NETCONF", scm_from_int (RTNLGRP_IPV4_NETCONF));
+#endif
+#ifdef RTNLGRP_IPV6_NETCONF
+  scm_c_define ("RTNLGRP_IPV6_NETCONF", scm_from_int (RTNLGRP_IPV6_NETCONF));
+#endif
+#ifdef RTNLGRP_MDB
+  scm_c_define ("RTNLGRP_MDB", scm_from_int (RTNLGRP_MDB));
+#endif
+
   scm_add_feature ("socket");
 
 #include "libguile/socket.x"
diff --git a/module/ice-9/networking.scm b/module/ice-9/networking.scm
index f9ff394..1146d80 100644
--- a/module/ice-9/networking.scm
+++ b/module/ice-9/networking.scm
@@ -85,6 +85,8 @@
 (define (sockaddr:port obj) (vector-ref obj 2))
 (define (sockaddr:flowinfo obj) (vector-ref obj 3))
 (define (sockaddr:scopeid obj) (vector-ref obj 4))
+(define (sockaddr:pid obj) (vector-ref obj 1))
+(define (sockaddr:groups obj) (vector-ref obj 2))
 
 (define (addrinfo:flags obj) (vector-ref obj 0))
 (define (addrinfo:fam obj) (vector-ref obj 1))
-- 
2.1.4


[-- Attachment #3: Type: text/plain, Size: 136 bytes --]


-- 
David Thompson
Web Developer - Free Software Foundation - http://fsf.org
GPG Key: 0FF1D807
Support the FSF: https://fsf.org/donate

Re: [PATCH] socket: Add AF_NETLINK support.

[Ludovic Courtès]

Hi!

David Thompson <dthompson2@worcester.edu> skribis:

> This is my first attempt at providing support for AF_NETLINK sockets in
> Guile, based on a preliminary patch that Ludovic sent me.  Netlink
> sockets are supported only on Linux.  The ultimate goal of this work is
> to be able to create virtual ethernet devices from Scheme code.
>
> There are no tests yet and I'm not sure if I went overboard by defining
> all of the numerous netlink families and multicast group constants.
> Feedback requested! :)

One concern we discussed on IRC is precisely that it’s Linux-only, and
so far we’ve tried to avoid promoting Linux-only interfaces.

An argument in favor of having it in libguile proper is that it cannot
be implemented outside since we need ‘make-socket-address’ to do the
right thing.

Another one is that it just provides bindings to the existing socket
interface, and like other parts of the socket interface, libguile would
just mirror what the OS provides.

OTOH, it could be argued that libnl is more appropriate than the socket
interface anyway, and that one can write FFI bindings to libnl.

What do people think?

Ludo’.

Re: [PATCH] socket: Add AF_NETLINK support.

[Greg Troxel]

[-- Attachment #1: Type: text/plain, Size: 1334 bytes --]


ludo@gnu.org (Ludovic Courtès) writes:

> One concern we discussed on IRC is precisely that it’s Linux-only, and
> so far we’ve tried to avoid promoting Linux-only interfaces.
>
> An argument in favor of having it in libguile proper is that it cannot
> be implemented outside since we need ‘make-socket-address’ to do the
> right thing.
>
> Another one is that it just provides bindings to the existing socket
> interface, and like other parts of the socket interface, libguile would
> just mirror what the OS provides.
>
> OTOH, it could be argued that libnl is more appropriate than the socket
> interface anyway, and that one can write FFI bindings to libnl.

I think Linux-only interfaces is a real concern, as it will lead to
guile programs that work only on Linux.

An alternative is to look at something like quagga, which has a glue
layer for various OS support for netlink/rtsock/?.   That's more work,
but would let people implment a non-OS-specific interface on multiple
systems and allow portable scheme code.

But, that's much harder.


Until then, why can't the netlink code not be part of guile, and just
another package to build and install and then access via use-modules,
sort of how guile-pg is separate?  Other than wanting to avoid an extra
package, what's the downside?

[-- Attachment #2: Type: application/pgp-signature, Size: 180 bytes --]

Re: [PATCH] socket: Add AF_NETLINK support.

[Ludovic Courtès]

Greg Troxel <gdt@ir.bbn.com> skribis:

> I think Linux-only interfaces is a real concern, as it will lead to
> guile programs that work only on Linux.
>
> An alternative is to look at something like quagga, which has a glue
> layer for various OS support for netlink/rtsock/?.   That's more work,
> but would let people implment a non-OS-specific interface on multiple
> systems and allow portable scheme code.

I just checked and Quagga’s documentation doesn’t say anything about its
library.  Is it an internal library or something really exposed to
users?

(BTW, I should note that this patch of David’s is part of his work to
support Linux containers, with a longer-term goal to support GNU/Hurd,
but that’ll be different interfaces anyway.)

> Until then, why can't the netlink code not be part of guile, and just
> another package to build and install and then access via use-modules,
> sort of how guile-pg is separate?  Other than wanting to avoid an extra
> package, what's the downside?

As I wrote, the low-level socket interface cannot be outside of libguile
because at least ‘make-socket-address’ and friends need to cooperate.
However, libnl bindings could indeed be outside of Guile.

Thanks,
Ludo’.