[Qemu-devel] user networking - win xp quest, amd64 linux host

[Qemu-devel] user networking - win xp quest, amd64 linux host

[Adrian Smarzewski]

When I set 10.0.2.15 for win xp guest I can ping 10.0.2.2
but DHCP doesn't work and routing too. Any ideas?
Something in host-kernel configuration?

Re: [Qemu-devel] user networking - win xp quest, amd64 linux host

[Ben Taylor]

> 
> From: Adrian Smarzewski <adrian@kadu.net>
> Date: 2005/05/08 Sun PM 06:22:27 EDT
> To: qemu-devel@nongnu.org
> Subject: [Qemu-devel] user networking - win xp quest, amd64 linux host
> 
> When I set 10.0.2.15 for win xp guest I can ping 10.0.2.2
> but DHCP doesn't work and routing too. Any ideas?
> Something in host-kernel configuration?

When you set a fixed address for a win xp guest, you won't
get the default router or dns.  Why are you manually
setting the IP addes in XP?

Re: [Qemu-devel] user networking - win xp quest, amd64 linux host

[Adrian Smarzewski]

Ben Taylor wrote:
>>From: Adrian Smarzewski <adrian@kadu.net>
>>Date: 2005/05/08 Sun PM 06:22:27 EDT
>>To: qemu-devel@nongnu.org
>>Subject: [Qemu-devel] user networking - win xp quest, amd64 linux host
>>
>>When I set 10.0.2.15 for win xp guest I can ping 10.0.2.2
>>but DHCP doesn't work and routing too. Any ideas?
>>Something in host-kernel configuration?
> 
> 
> When you set a fixed address for a win xp guest, you won't
> get the default router or dns.  Why are you manually
> setting the IP addes in XP?

DHCP is not working just like I wrote. I set DNS and default gateway
using QEMU manual for user-networking. No success. Does user-mode
networking need routing enabled in host kernel?

Re: [Qemu-devel] user networking - win xp quest, amd64 linux host

[Henrik Nordstrom]

On Mon, 9 May 2005, Adrian Smarzewski wrote:

> DHCP is not working just like I wrote. I set DNS and default gateway
> using QEMU manual for user-networking. No success.

Should work..

> Does user-mode networking need routing enabled in host kernel?

No.

Regards
Henrik

Re: [Qemu-devel] user networking - win xp quest, amd64 linux host

[Adrian Smarzewski]

Henrik Nordstrom wrote:
> Should work..

how can I debug it? Any logs for user-networking? qemu 0.7.0.

Re: [Qemu-devel] user networking - win xp quest, amd64 linux host

[Adrian Smarzewski]

Ben Taylor wrote:
> suspect the amd64 is biting you. What is your host OS
> environment like, and what version of qemu are you
> compiling?

bash-2.05b$ uname -a
Linux localhost 2.6.11-gentoo-r7 #1 Sun May 8 06:59:56 CEST 2005 x86_64
AMD Athlon(tm) 64 Processor 3400+ AuthenticAMD GNU/Linux

bash-2.05b$ qemu -hda win_xp_home_sp2.raw -user-net -localtime

Qemu 0.7.0 without any patches.

Re: [Qemu-devel] user networking - win xp quest, amd64 linux host

[Adrian Smarzewski]

Adrian Smarzewski wrote:
> bash-2.05b$ uname -a
> Linux localhost 2.6.11-gentoo-r7 #1 Sun May 8 06:59:56 CEST 2005 x86_64
> AMD Athlon(tm) 64 Processor 3400+ AuthenticAMD GNU/Linux
> 
> bash-2.05b$ qemu -hda win_xp_home_sp2.raw -user-net -localtime
> 
> Qemu 0.7.0 without any patches.

localhost root # ifconfig
eth0      Link encap:Ethernet  HWaddr 00:0F:B0:0C:7F:0C
          inet addr:10.42.1.129  Bcast:10.42.1.255  Mask:255.255.255.0
          UP BROADCAST NOTRAILERS RUNNING MULTICAST  MTU:1500  Metric:1
          RX packets:2848615 errors:0 dropped:0 overruns:0 frame:0
          TX packets:1600599 errors:0 dropped:0 overruns:0 carrier:0
          collisions:3 txqueuelen:1000
          RX bytes:3674369478 (3504.1 Mb)  TX bytes:133994216 (127.7 Mb)
          Interrupt:19 Base address:0x800

lo        Link encap:Local Loopback
          inet addr:127.0.0.1  Mask:255.0.0.0
          UP LOOPBACK RUNNING  MTU:16436  Metric:1
          RX packets:121 errors:0 dropped:0 overruns:0 frame:0
          TX packets:121 errors:0 dropped:0 overruns:0 carrier:0
          collisions:0 txqueuelen:0
          RX bytes:7678 (7.4 Kb)  TX bytes:7678 (7.4 Kb)

Re: [Qemu-devel] user networking - win xp quest, amd64 linux host

[Leonardo E. Reiter]

We have seen problems on the AMD64 with -user-net as well.  While we 
have not been able to debug further yet, I suspect what is happening is 
that some 32-bit values in packet headers that SLIRP is decoding are 
actually represented as 64-bit types on 64-bit processors.  Again, this 
is just a guess, but it's the first place we'll look once we get to it, 
since we think it's the most likely cause.

- Leo Reiter

Adrian Smarzewski wrote:
> Adrian Smarzewski wrote:
> 
>>bash-2.05b$ uname -a
>>Linux localhost 2.6.11-gentoo-r7 #1 Sun May 8 06:59:56 CEST 2005 x86_64
>>AMD Athlon(tm) 64 Processor 3400+ AuthenticAMD GNU/Linux

-- 
Leonardo E. Reiter
Vice President of Product Development, CTO

Win4Lin, Inc.
Virtual Computing from Desktop to Data Center
Main: +1 512 339 7979
Fax: +1 512 532 6501
http://www.win4lin.com

Re: [Qemu-devel] user networking - win xp quest, amd64 linux host

[Adrian Smarzewski]

On Mon, 09 May 2005 10:37:04 -0400, Leonardo E. Reiter wrote
> We have seen problems on the AMD64 with -user-net as well.  While we 
> have not been able to debug further yet, I suspect what is happening 
> is that some 32-bit values in packet headers that SLIRP is decoding 
> are actually represented as 64-bit types on 64-bit processors. 
>  Again, this is just a guess, but it's the first place we'll look 
> once we get to it, since we think it's the most likely cause.

If you need some help... I'm a programmer and I have amd64 machine.
I don't know anything about user-network code and don't have a lot
of time but It's important for me to have a qemu networking. Just write
me what to do.

Any success with qemu networking on amd64 host? Somebody?

Re: [Qemu-devel] user networking - win xp quest, amd64 linux host

[Leonardo E. Reiter]

Adrian,

we will be working on this shortly.  In the meantime, where I would 
start is by putting a breakpoint in udp_input() in the file slirp/udp.c, 
and following the flow until it calls bootp_input().  If it gets that 
far, it means it's probably decoding the UDP packet correctly and 
detects it as a DHCP/BOOTP request.  You might want to also study 
slirp/ip.h and slirp/udp.h to make sure all the values are represented 
in the correct bit length.

- Leo Reiter

P.S. we also noticed the same problem on the IBM iSeries PowerPC host 
(which is the "new" AS/400 running Linux), so perhaps it is indeed 
something to do specifically with 64-bit hosts.  But again, this is just 
a guess, until we can debug and pinpoint the issue.

Adrian Smarzewski wrote:
> If you need some help... I'm a programmer and I have amd64 machine.
> I don't know anything about user-network code and don't have a lot
> of time but It's important for me to have a qemu networking. Just write
> me what to do.
> 
> Any success with qemu networking on amd64 host? Somebody?
> 

-- 
Leonardo E. Reiter
Vice President of Product Development, CTO

Win4Lin, Inc.
Virtual Computing from Desktop to Data Center
Main: +1 512 339 7979
Fax: +1 512 532 6501
http://www.win4lin.com

Re: [Qemu-devel] user networking - win xp quest, amd64 linux host

[Adrian Smarzewski]

On Mon, 9 May 2005 17:19:57 +0200, Adrian Smarzewski wrote
> If you need some help... I'm a programmer and I have amd64 machine.
> I don't know anything about user-network code and don't have a lot
> of time but It's important for me to have a qemu networking. Just write
> me what to do.

I will look for "long int" types in code later and try to change
them to "int", maybe it will help.

Re: [Qemu-devel] user networking - win xp quest, amd64 linux host

[Adrian Smarzewski]

On Mon, 09 May 2005 11:50:09 -0400, Leonardo E. Reiter wrote
> correctly and detects it as a DHCP/BOOTP request.  You might want to 
> also study slirp/ip.h and slirp/udp.h to make sure all the values 
> are represented in the correct bit length.

If you have this problem too I will try to fix 64-bit compatibility today.
If I manage to do it I will send a patch.

Re: [Qemu-devel] user networking - win xp quest, amd64 linux host

[Juergen Lock]

On Mon, May 09, 2005 at 10:37:04AM -0400, Leonardo E. Reiter wrote:
> We have seen problems on the AMD64 with -user-net as well.  While we 
> have not been able to debug further yet, I suspect what is happening is 
> that some 32-bit values in packet headers that SLIRP is decoding are 
> actually represented as 64-bit types on 64-bit processors.  Again, this 
> is just a guess, but it's the first place we'll look once we get to it, 
> since we think it's the most likely cause.
> 
> - Leo Reiter

There are a few cases of `long' being used when apparently (u)int32_t was
meant.  Not sure i catched them all, but try this patch:  (Unfortunately
I don't have an amd64 box here so I cant test it myself...)

Index: slirp/bootp.h
===================================================================
RCS file: /cvsroot/qemu/qemu/slirp/bootp.h,v
retrieving revision 1.1
diff -u -r1.1 bootp.h
--- slirp/bootp.h	22 Apr 2004 00:10:47 -0000	1.1
+++ slirp/bootp.h	9 May 2005 17:11:55 -0000
@@ -97,7 +98,7 @@
     uint8_t bp_htype;
     uint8_t bp_hlen;
     uint8_t bp_hops;
-    unsigned long bp_xid;
+    uint32_t bp_xid;
     unsigned short bp_secs;
     unsigned short unused;
     struct in_addr bp_ciaddr;
Index: slirp/ip_icmp.h
===================================================================
RCS file: /cvsroot/qemu/qemu/slirp/ip_icmp.h,v
retrieving revision 1.1
diff -u -r1.1 ip_icmp.h
--- slirp/ip_icmp.h	22 Apr 2004 00:10:47 -0000	1.1
+++ slirp/ip_icmp.h	9 May 2005 17:11:56 -0000
@@ -83,7 +83,7 @@
 			struct ip idi_ip;
 			/* options and then 64 bits of data */
 		} id_ip;
-		u_long	id_mask;
+		uint32_t	id_mask;
 		char	id_data[1];
 	} icmp_dun;
 #define	icmp_otime	icmp_dun.id_ts.its_otime
Index: slirp/udp.c
===================================================================
RCS file: /cvsroot/qemu/qemu/slirp/udp.c,v
retrieving revision 1.4
diff -u -r1.4 udp.c
--- slirp/udp.c	7 Oct 2004 23:27:35 -0000	1.4
+++ slirp/udp.c	9 May 2005 17:11:56 -0000
@@ -423,7 +423,7 @@
 	char 	dest[8];
 	short 	family;
 	u_short	port;
-	u_long	addr;
+	uint32_t	addr;
 } *cu_head;
 
 	switch(so->so_emu) {
@@ -611,7 +611,7 @@
 				return;
 			cu_head = mtod(m, struct cu_header *);
 			cu_head->port = addr.sin_port;
-			cu_head->addr = (u_long) our_addr.s_addr;
+			cu_head->addr = (uint32_t) our_addr.s_addr;
 		}
 		
 		return;

Re: [Qemu-devel] user networking - win xp quest, amd64 linux host

[Adrian Smarzewski]

On Mon, 9 May 2005 19:26:33 +0200, Juergen Lock wrote
> There are a few cases of `long' being used when apparently (u)
> int32_t was meant.  Not sure i catched them all, but try this patch: 
>  (Unfortunately I don't have an amd64 box here so I cant test it 
> myself...)

I can try it later (4 to 5 hours). I see that you changed two things
in udp so I hope it will work :)

Re: [Qemu-devel] user networking - win xp quest, amd64 linux host

[Jung-uk Kim]

On Monday 09 May 2005 01:26 pm, Juergen Lock wrote:
> On Mon, May 09, 2005 at 10:37:04AM -0400, Leonardo E. Reiter wrote:
> > We have seen problems on the AMD64 with -user-net as well.  While
> > we have not been able to debug further yet, I suspect what is
> > happening is that some 32-bit values in packet headers that SLIRP
> > is decoding are actually represented as 64-bit types on 64-bit
> > processors.  Again, this is just a guess, but it's the first
> > place we'll look once we get to it, since we think it's the most
> > likely cause.
> >
> > - Leo Reiter
>
> There are a few cases of `long' being used when apparently
> (u)int32_t was meant.  Not sure i catched them all, but try this
> patch:  (Unfortunately I don't have an amd64 box here so I cant
> test it myself...)

Juergen,

sorry but it doesn't seem to work.  Tried on FreeBSD/amd64 -CURRENT 
host and FreeBSD/i386 -STABLE guest.

Jung-uk Kim

> Index: slirp/bootp.h
> ===================================================================
> RCS file: /cvsroot/qemu/qemu/slirp/bootp.h,v
> retrieving revision 1.1
> diff -u -r1.1 bootp.h
> --- slirp/bootp.h	22 Apr 2004 00:10:47 -0000	1.1
> +++ slirp/bootp.h	9 May 2005 17:11:55 -0000
> @@ -97,7 +98,7 @@
>      uint8_t bp_htype;
>      uint8_t bp_hlen;
>      uint8_t bp_hops;
> -    unsigned long bp_xid;
> +    uint32_t bp_xid;
>      unsigned short bp_secs;
>      unsigned short unused;
>      struct in_addr bp_ciaddr;
> Index: slirp/ip_icmp.h
> ===================================================================
> RCS file: /cvsroot/qemu/qemu/slirp/ip_icmp.h,v
> retrieving revision 1.1
> diff -u -r1.1 ip_icmp.h
> --- slirp/ip_icmp.h	22 Apr 2004 00:10:47 -0000	1.1
> +++ slirp/ip_icmp.h	9 May 2005 17:11:56 -0000
> @@ -83,7 +83,7 @@
>  			struct ip idi_ip;
>  			/* options and then 64 bits of data */
>  		} id_ip;
> -		u_long	id_mask;
> +		uint32_t	id_mask;
>  		char	id_data[1];
>  	} icmp_dun;
>  #define	icmp_otime	icmp_dun.id_ts.its_otime
> Index: slirp/udp.c
> ===================================================================
> RCS file: /cvsroot/qemu/qemu/slirp/udp.c,v
> retrieving revision 1.4
> diff -u -r1.4 udp.c
> --- slirp/udp.c	7 Oct 2004 23:27:35 -0000	1.4
> +++ slirp/udp.c	9 May 2005 17:11:56 -0000
> @@ -423,7 +423,7 @@
>  	char 	dest[8];
>  	short 	family;
>  	u_short	port;
> -	u_long	addr;
> +	uint32_t	addr;
>  } *cu_head;
>
>  	switch(so->so_emu) {
> @@ -611,7 +611,7 @@
>  				return;
>  			cu_head = mtod(m, struct cu_header *);
>  			cu_head->port = addr.sin_port;
> -			cu_head->addr = (u_long) our_addr.s_addr;
> +			cu_head->addr = (uint32_t) our_addr.s_addr;
>  		}
>
>  		return;

Re: [Qemu-devel] user networking - win xp quest, amd64 linux host

[Adrian Smarzewski]

Adrian Smarzewski wrote:
> When I set 10.0.2.15 for win xp guest I can ping 10.0.2.2
> but DHCP doesn't work and routing too. Any ideas?
> Something in host-kernel configuration?

Looks like user-mode networking doesn't work on AMD64 host.
I can give you access via ssh if someone has any ideas.