[Qemu-devel] [PATCH] native vde support

[Qemu-devel] [PATCH] native vde support

[Luca Bigliardi]

[-- Attachment #1: Type: text/plain, Size: 2444 bytes --]

Hi!

The patch I'm proposing here adds native support to Virtual Distributed
Ethernet using libvdeplug.

A small example:
- patch latest svn with the attached patch
- configure qemu to use vde support (disabled by default) using
  "./configure --enable-vde"
- start a vde switch
    vde_switch -F -sock /tmp/my_qemu_test
- start qemu (full syntax explanation is in the qemu-doc and appears
  into qemu online help if it's compiled with vde support)
    qemu -net nic -net vde,sock=/tmp/my_qemu_test linux-0.2.img

What is VDE?
Virtual Distributed Ethernet is a virtual network that can be spawned over
a set of physical computer over the Internet and that can be shared
between physical hosts and various kinds of virtual machines.
vde_switch, core app of VDE framework, is an evolution of uml_switch.
The project was originally written by Renzo Davoli and now is maintained
and developed by Virtualsquare Team (which I'm part of).
For more information and simple examples of vde usage visit:
http://wiki.virtualsquare.org/index.php/VDE and
http://wiki.virtualsquare.org/index.php/VDE_Basic_Networking
	 
Why add native support to VDE?
Nowadays a number of qemu instances already use vde networking, to give
you an idea simply search "qemu vde" on google, or think that vde is the
recommended networking tool for qemu in major distributions as debian or
ubuntu (and maybe others..), so in my opinion native vde networking will
be widely used!

What's wrong with the present solution?
What are you doing with your patch?
Unfortunately at the moment qemu is connected to a vde_switch through
a dirty hack called 'vdeq' which places itself in the middle between the
two applications.
The life of a vdeq is quite simple:
- do some tricky syntax parsing and mangling
- connect to vde_switch using libvdeplug
- open socketpair
- fork qemu giving him one end of socketpair
- enter a loop to forward data between vde_switch and qemu
- when qemu exits close the connection with vde_switch
This is of course far from being optimal for obvious reasons, so with my
patch I simply put into qemu the part of vdeq which connects/disconnects
to vde_switch and send/receive data to it.


Please tell me what do you think about this idea and the attached patch,
thank you!

Regards,

Luca

-- 
Beware of programmers who carry screwdrivers.
                        -- Leonard Brandwein

http://shammash.homelinux.org/ - http://www.artha.org/ - http://www.yue.it/

[-- Attachment #2: qemu_vde_0.1.diff --]
[-- Type: text/x-diff, Size: 7868 bytes --]

Index: Makefile.target
===================================================================
--- Makefile.target	(revision 4815)
+++ Makefile.target	(working copy)
@@ -652,7 +652,7 @@
 endif
 
 $(QEMU_PROG): $(OBJS) ../libqemu_common.a libqemu.a
-	$(CC) $(LDFLAGS) -o $@ $^ $(LIBS) $(SDL_LIBS) $(COCOA_LIBS) $(CURSES_LIBS) $(BRLAPI_LIBS)
+	$(CC) $(LDFLAGS) -o $@ $^ $(LIBS) $(SDL_LIBS) $(COCOA_LIBS) $(CURSES_LIBS) $(BRLAPI_LIBS) $(VDE_LIBS)
 
 endif # !CONFIG_USER_ONLY
 
Index: vl.c
===================================================================
--- vl.c	(revision 4815)
+++ vl.c	(working copy)
@@ -106,6 +106,10 @@
 #include "libslirp.h"
 #endif
 
+#if defined(CONFIG_VDE)
+#include <libvdeplug.h>
+#endif
+
 #ifdef _WIN32
 #include <malloc.h>
 #include <sys/timeb.h>
@@ -4404,6 +4408,66 @@
 
 #endif /* !_WIN32 */
 
+#if defined(CONFIG_VDE)
+typedef struct VDEState {
+    VLANClientState *vc;
+    VDECONN *vde;
+} VDEState;
+
+static void vde_to_qemu(void *opaque)
+{
+    VDEState *s = opaque;
+    uint8_t buf[4096];
+    int size;
+
+    size = vde_recv(s->vde, buf, sizeof(buf), 0);
+    if (size > 0) {
+        qemu_send_packet(s->vc, buf, size);
+    }
+}
+
+static void vde_from_qemu(void *opaque, const uint8_t *buf, int size)
+{
+    VDEState *s = opaque;
+    int ret;
+    for(;;) {
+        ret = vde_send(s->vde, buf, size, 0);
+        if (ret < 0 && errno == EINTR) {
+        } else {
+            break;
+        }
+    }
+}
+
+static int net_vde_init(VLANState *vlan, const char *sock, int port,
+                        const char *group, int mode)
+{
+    VDEState *s;
+    char *init_group = strlen(group) ? (char *)group : NULL;
+    char *init_sock = strlen(sock) ? (char *)sock : NULL;
+
+    struct vde_open_args args = {
+        .port = port,
+        .group = init_group,
+        .mode = mode,
+    };
+
+    s = qemu_mallocz(sizeof(VDEState));
+    if (!s)
+        return -1;
+    s->vde = vde_open(init_sock, "QEMU", &args);
+    if (!s->vde){
+        free(s);
+        return -1;
+    }
+    s->vc = qemu_new_vlan_client(vlan, vde_from_qemu, NULL, s);
+    qemu_set_fd_handler(vde_datafd(s->vde), vde_to_qemu, NULL, s);
+    snprintf(s->vc->info_str, sizeof(s->vc->info_str), "vde: sock=%s fd=%d",
+             sock, vde_datafd(s->vde));
+    return 0;
+}
+#endif
+
 /* network connection */
 typedef struct NetSocketState {
     VLANClientState *vc;
@@ -5033,6 +5097,30 @@
         }
         vlan->nb_host_devs++;
     } else
+#ifdef CONFIG_VDE
+    if (!strcmp(device, "vde")) {
+        char vde_sock[1024], vde_group[512];
+	int vde_port, vde_mode;
+        vlan->nb_host_devs++;
+        if (get_param_value(vde_sock, sizeof(vde_sock), "sock", p) <= 0) {
+	    vde_sock[0] = '\0';
+	}
+	if (get_param_value(buf, sizeof(buf), "port", p) > 0) {
+	    vde_port = strtol(buf, NULL, 10);
+	} else {
+	    vde_port = 0;
+	}
+	if (get_param_value(vde_group, sizeof(vde_group), "group", p) <= 0) {
+	    vde_group[0] = '\0';
+	}
+	if (get_param_value(buf, sizeof(buf), "mode", p) > 0) {
+	    vde_mode = strtol(buf, NULL, 8);
+	} else {
+	    vde_mode = 0700;
+	}
+	ret = net_vde_init(vlan, vde_sock, vde_port, vde_group, vde_mode);
+    } else
+#endif
     {
         fprintf(stderr, "Unknown network device: %s\n", device);
         return -1;
@@ -7389,6 +7477,13 @@
            "                connect the vlan 'n' to another VLAN using a socket connection\n"
            "-net socket[,vlan=n][,fd=h][,mcast=maddr:port]\n"
            "                connect the vlan 'n' to multicast maddr and port\n"
+#ifdef CONFIG_VDE
+           "-net vde[,vlan=n][,sock=socketpath][,port=n][,group=groupname][,mode=octalmode]\n"
+           "                connect the vlan 'n' to port 'n' of a vde switch running\n"
+           "                on host and listening for incoming connections on 'socketpath'.\n"
+           "                Use group 'groupname' and mode 'octalmode' to change default\n"
+           "                ownership and permissions for communication port.\n"
+#endif
            "-net none       use it alone to have zero network devices; if no -net option\n"
            "                is provided, the default is '-net nic -net user'\n"
            "\n"
@@ -8868,6 +8963,12 @@
                     s->down_script[0])
                     launch_script(s->down_script, ifname, s->fd);
             }
+#if defined(CONFIG_VDE)
+            if (vc->fd_read == vde_from_qemu) {
+                VDEState *s = vc->opaque;
+                vde_close(s->vde);
+            }
+#endif
         }
     }
 #endif
Index: Makefile
===================================================================
--- Makefile	(revision 4815)
+++ Makefile	(working copy)
@@ -127,6 +127,10 @@
 OBJS+=$(addprefix slirp/, $(SLIRP_OBJS))
 endif
 
+ifdef CONFIG_VDE
+LIBS+=-lvdeplug
+endif
+
 cocoa.o: cocoa.m
 	$(CC) $(CFLAGS) $(CPPFLAGS) -c -o $@ $<
 
Index: qemu-doc.texi
===================================================================
--- qemu-doc.texi	(revision 4815)
+++ qemu-doc.texi	(working copy)
@@ -672,6 +672,21 @@
 /path/to/linux ubd0=/path/to/root_fs eth0=mcast
 @end example
 
+@item -net vde[,vlan=@var{n}][,sock=@var{socketpath}][,port=@var{n}][,group=@var{groupname}][,mode=@var{octalmode}]
+Connect VLAN @var{n} to PORT @var{n} of a vde switch running on host and
+listening for incoming connections on @var{socketpath}. Use GROUP @var{groupname}
+and MODE @var{octalmode} to change default ownership and permissions for
+communication port. This option is available only if QEMU has been compiled
+with vde support enabled.
+
+Example:
+@example
+# launch vde switch
+vde_switch -F -sock /tmp/myswitch
+# launch QEMU instance
+qemu linux.img -net nic -net vde,sock=/tmp/myswitch
+@end example
+
 @item -net none
 Indicate that no network devices should be configured. It is used to
 override the default configuration (@option{-net nic -net user}) which
Index: configure
===================================================================
--- configure	(revision 4815)
+++ configure	(working copy)
@@ -89,6 +89,7 @@
 EXESUF=""
 gdbstub="yes"
 slirp="yes"
+vde="no"
 fmod_lib=""
 fmod_inc=""
 vnc_tls="yes"
@@ -280,6 +281,8 @@
   ;;
   --disable-slirp) slirp="no"
   ;;
+  --enable-vde) vde="yes"
+  ;;
   --disable-kqemu) kqemu="no"
   ;;
   --disable-brlapi) brlapi="no"
@@ -432,6 +435,7 @@
 echo "  --fmod-inc               path to FMOD includes"
 echo "  --enable-uname-release=R Return R for uname -r in usermode emulation"
 echo "  --sparc_cpu=V            Build qemu for Sparc architecture v7, v8, v8plus, v8plusa, v9"
+echo "  --enable-vde             enable support for vde network [$vde]"
 echo ""
 echo "NOTE: The object files are built at the place where configure is launched"
 exit 1
@@ -722,6 +726,19 @@
 fi
 
 ##########################################
+# vde libraries probe
+if test "$vde" = "yes" ; then
+  vde=no
+  cat > $TMPC << EOF
+#include <libvdeplug.h>
+int main(void) { struct vde_open_args a = {0, 0, 0} ; return 0;}
+EOF
+  if $cc $ARCH_CFLAGS -o $TMPE $TMPC 2> /dev/null ; then
+    vde=yes
+  fi
+fi
+
+##########################################
 # Sound support libraries probe
 
 audio_drv_probe()
@@ -868,6 +885,7 @@
 [ ! -z "$uname_release" ] && \
 echo "uname -r          $uname_release"
 echo "NPTL support      $nptl"
+echo "vde support       $vde"
 
 if test $sdl_too_old = "yes"; then
 echo "-> Your SDL version is too old - please upgrade to have SDL support"
@@ -1032,6 +1050,11 @@
   echo "CONFIG_SLIRP=yes" >> $config_mak
   echo "#define CONFIG_SLIRP 1" >> $config_h
 fi
+if test "$vde" = "yes" ; then
+  echo "CONFIG_VDE=yes" >> $config_mak
+  echo "#define CONFIG_VDE 1" >> $config_h
+  echo "VDE_LIBS=-lvdeplug" >> $config_mak
+fi
 for card in $audio_card_list; do
     def=CONFIG_`echo $card | tr '[:lower:]' '[:upper:]'`
     echo "$def=yes" >> $config_mak

Re: [Qemu-devel] [PATCH] native vde support

[Erik de Castro Lopo]

Luca Bigliardi wrote:

> The patch I'm proposing here adds native support to Virtual Distributed
> Ethernet using libvdeplug.

Yes, yes, yes, please!

> Why add native support to VDE?
> Nowadays a number of qemu instances already use vde networking, to give
> you an idea simply search "qemu vde" on google, or think that vde is the
> recommended networking tool for qemu in major distributions as debian or
> ubuntu (and maybe others..), so in my opinion native vde networking will
> be widely used!

I tried a number of different networking options when I first started
using qemu and VDE is the one that works best for me. I'd love to
see this go in.

Cheers,
Erik
-- 
-----------------------------------------------------------------
Erik de Castro Lopo
-----------------------------------------------------------------
"Anyone who considers arithmetical methods of producing random
digits is, of course, in a state of sin." - John Von Neumann (1951)

Re: [Qemu-devel] [PATCH] native vde support

[Erik de Castro Lopo]

Luca Bigliardi wrote:

> The patch I'm proposing here adds native support to Virtual Distributed
> Ethernet using libvdeplug.

Is someone with commit rights able to have a look at this patch?

If its not up to scratch now, please point out the problems so it
can be fixed and re-submitted.

Cheers,
Erik
-- 
-----------------------------------------------------------------
Erik de Castro Lopo
-----------------------------------------------------------------
Seen on usenet (possibly a quote from an IBM exec):
"Each large company needs its Vietnam, and Microsoft will
 experience it with NT..."

Re: [Qemu-devel] [PATCH] native vde support

[Thiemo Seufer]

Luca Bigliardi wrote:
> Hi!
> 
> The patch I'm proposing here adds native support to Virtual Distributed
> Ethernet using libvdeplug.
> 
> A small example:
> - patch latest svn with the attached patch
> - configure qemu to use vde support (disabled by default) using
>   "./configure --enable-vde"

When no libvdeplug is available it silently ignores the user selection.
I changed that, removed a hardcoded -lvdeplug in Makefile, and committed
the result.


Thiemo

Re: [Qemu-devel] [PATCH] native vde support

[Erik de Castro Lopo]

Thiemo Seufer wrote:

> When no libvdeplug is available it silently ignores the user selection.
> I changed that, removed a hardcoded -lvdeplug in Makefile, and committed
> the result.

Awesome Thiemo. Thanks. I'll test that some time over the next couple
of days.

Erik
-- 
-----------------------------------------------------------------
Erik de Castro Lopo
-----------------------------------------------------------------
Everything that I've learned about computers I have boiled
down into three principles:
  PC/Windows: You think it won't work, and it won't.
  Macintosh: You think it will work, but it won't.
  Unix: You think it won't work, but if you find the right
        guru, you can make it work.

Re: [Qemu-devel] [PATCH] native vde support

[Anthony Liguori]

Thiemo Seufer wrote:
> Luca Bigliardi wrote:
>   
>> Hi!
>>
>> The patch I'm proposing here adds native support to Virtual Distributed
>> Ethernet using libvdeplug.
>>
>> A small example:
>> - patch latest svn with the attached patch
>> - configure qemu to use vde support (disabled by default) using
>>   "./configure --enable-vde"
>>     
>
> When no libvdeplug is available it silently ignores the user selection.
> I changed that, removed a hardcoded -lvdeplug in Makefile, and committed
> the result.
>   

I've changed it to enable vde support by default if the library is 
present and changed --enable-vde to --disable-vde.

This is how we use other external libraries (like gnutls).  Since the 
user has to explicitly ask to use libvde, there seems to be no harm in 
enabling it by default.

Regards,

Anthony Liguori

> Thiemo
>
>
>   

Re: [Qemu-devel] [PATCH] native vde support

[Erik de Castro Lopo]

Anthony Liguori wrote:

> I've changed it to enable vde support by default if the library is 
> present and changed --enable-vde to --disable-vde.
> 
> This is how we use other external libraries (like gnutls).  Since the 
> user has to explicitly ask to use libvde, there seems to be no harm in 
> enabling it by default.

Even better. Thanks Anthony.

Cheers,
Erik
-- 
-----------------------------------------------------------------
Erik de Castro Lopo
-----------------------------------------------------------------
"There is no satisfactory substitute for excellence."
-- Dr. Arnold O. Beckman