[GXemul-devel] [PATCH] Update TAP networking to let it work on a Linux host

[Jan-Benedict G. <jb...@lu...>]

Hi!

There is probably already a (to-be-pushed-to-upstream-SVN) patch
available to enable TAP networking unter Linux, but nonetheless,
here's what I have locally for the time being.

  I'm using it this way (here as an example for a DECstation booting
the NetBSD installer: There are CI builds for NetBSD HEAD running,
storing the ISO and sets (along with the kernel) in tarballs. This is
extracted and GXemul started with it, on a local TAP interface bridged
with a physical wired NIC:

tar xzf /var/cache/laminar/netbsd-rel-pmax-mipsel.tar.gz
dd bs=1M count=1024 if=/dev/zero of=netbsd.img
ISO="$(ls images/*.iso | head -1)"

TAP=threemax
tunctl -t "${TAP}"       || true
ip link set "${TAP}" up  || true
brctl addif br0 "${TAP}" || true


# gxemul @3maxconfig

gxemul  -E decstation -e 3max                           \
        -L threemax                                     \
        -l 10:11:12:13:14:16                            \
        -d netbsd.img                                   \
        -d "6c:${ISO}"                                  \
        -o "-a"                                         \
        -j netbsd pmax/binary/kernel/netbsd-INSTALL.gz


The plan is to prepare as many simulated systems as possible to
exercise as many NetBSD variants as possible. DECstations are just one
example, SIMH is already running (VAX), as well as Qemu (amd64). Seems
that more archs are reachable (alpha with a small Qemu patch, aarch64,
...) These simulators are used to just do test installations (and
throw away the resulting disk images), but my goal is to also install
"master" disk images and run these with overlays. All three simulators
(Qemu, SIMH, GXemul) support this type of configuration, so that's
ideal to start a nice farm of build hosts. The only difference is that
they're getting a fresh overlay and a different MAC address.


  Oh, and here's the Linux TAP patch. There are others around that
also set the IFF_ONE_QUEUE if available (cf.
http://svn.exactcode.de/t2/trunk/package/emulators/gxemul/linux-tap.patch
for example), but I'm not sure if that's needed or really beneficial.

diff --git a/src/net/net_tap.c b/src/net/net_tap.c
index 188276e..d920695 100644
--- a/src/net/net_tap.c
+++ b/src/net/net_tap.c
@@ -48,6 +48,11 @@
 #include "misc.h"
 #include "net.h"
 
+#ifdef __linux__
+#  include <linux/if.h>
+#  include <linux/if_tun.h>
+#endif
+
 /*
  *  net_tap_rx_for_nic():
  *
@@ -166,7 +171,11 @@ bool net_tap_init(struct net *net, const char *tapdev)
 	int fd;
 	int one = 1;
 
+#ifdef __linux__
+	fd = open("/dev/net/tun", O_RDWR);
+#else
 	fd = open(tapdev, O_RDWR);
+#endif
 	if (fd < 0) {
 		debugmsg(SUBSYS_NET, "tap", VERBOSITY_ERROR,
 		    "unable to open tap device '%s': %s",
@@ -174,6 +183,22 @@ bool net_tap_init(struct net *net, const char *tapdev)
 		return false;
 	}
 
+#ifdef __linux__
+	struct ifreq ifr;
+
+	memset(&ifr, 0, sizeof(ifr));
+	ifr.ifr_flags = IFF_TAP | IFF_NO_PI;
+	strncpy(ifr.ifr_name, tapdev, IFNAMSIZ);
+
+	if (ioctl(fd, TUNSETIFF, &ifr) < 0) {
+		debugmsg(SUBSYS_NET, "tap", VERBOSITY_ERROR,
+		    "unable to attach to "
+		    "tap device '%s': %s", tapdev, strerror(errno));
+		close(fd);
+		return false;
+	}
+#endif
+
 	if (ioctl(fd, FIONBIO, &one) < 0) {
 		debugmsg(SUBSYS_NET, "tap", VERBOSITY_ERROR,
 		    "unable to set non-blocking mode on "


MfG, JBG

--