NetBSD Problem Report #28339

From Thilo.Manske@HEH.Uni-Oldenburg.DE  Wed Nov 17 08:55:17 2004
Return-Path: <Thilo.Manske@HEH.Uni-Oldenburg.DE>
Received: from Server.HEH.Uni-Oldenburg.DE (extern.hrz.uni-oldenburg.de [134.106.137.37])
	by narn.netbsd.org (Postfix) with ESMTP id 746A4252383
	for <gnats-bugs@gnats.NetBSD.org>; Wed, 17 Nov 2004 08:55:16 +0000 (UTC)
Message-Id: <200411170855.iAH8tErC001102@Voigt.T>
Date: Wed, 17 Nov 2004 09:55:14 +0100 (MET)
From: Thilo Manske <Thilo.Manske@HEH.Uni-Oldenburg.DE>
Reply-To: Thilo.Manske@HEH.Uni-Oldenburg.DE
To: gnats-bugs@netbsd.org
Subject: M_HASFCS not honored in bpf(4)
X-Send-Pr-Version: 3.95

>Number:         28339
>Category:       kern
>Synopsis:       M_HASFCS not honored in bpf(4)
>Confidential:   no
>Severity:       non-critical
>Priority:       medium
>Responsible:    kern-bug-people
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Wed Nov 17 08:56:00 +0000 2004
>Last-Modified:  Mon Apr 03 19:00:00 +0000 2017
>Originator:     Thilo Manske
>Release:        NetBSD 2.99.10 and below
>Organization:
>Environment:
System: NetBSD Voigt 2.99.10 NetBSD 2.99.10 (Voigt) #69: Sat Nov 13 14:17:08 MET 2004 root@Voigt:/sys/arch/i386/compile/Voigt i386
Architecture: <all>
Machine: <all>
>Description:

Some ethernet network drivers (bce,sip,rtk,vr,wm..) deliver frames including
the 4 byte CRC sum at the end of the buffer, some don't (de, epic, ex, txp,
..) Drivers set the M_LINK0 (=M_HASFCS) flag according to that, but the flag
is neither honored nor passed to the next layer by bpf(4)

>How-To-Repeat:
- tcpdump ping traffic and notice received frames being 4 bytes larger
- use ethereal with tcpdumps taken from different interfaces
- watch dhclient complain like this:
	"ip length 328 disagrees with bytes received 332"
>Fix:
This is a simple patch/workaround, it works with drivers
using bpf_mtap() (tested with rtk(4)). For drivers using bpf_mtap2() a
very similar piece of code could be used, but there's no way for me to test
it, so I left it out.

Index: bpf.c
===================================================================
RCS file: /cvsroot/src/sys/net/bpf.c,v
retrieving revision 1.104
diff -c -u -r1.104 bpf.c
--- bpf.c	19 Aug 2004 20:58:23 -0000	1.104
+++ bpf.c	17 Nov 2004 08:42:25 -0000
@@ -1271,12 +1271,32 @@
 {
 	void *(*cpfn)(void *, const void *, size_t);
 	struct bpf_if *bp = arg;
-	u_int pktlen, buflen;
+	u_int frmlen, pktlen, buflen;
 	void *marg;

-	pktlen = m_length(m);
+	frmlen = pktlen = m_length(m);
+	/* 
+	 * try to find out if the mbuf is comming from an ethernet interface,
+	 * since M_LINK0 (M_HASFCS) meaning depends on the type of the interface
+	 */
+	if (m->m_pkthdr.rcvif != NULL)
+	{
+		switch (m->m_pkthdr.rcvif->if_dlt)
+		{
+			case DLT_EN10MB:
+			case DLT_EN3MB:
+			case DLT_IEEE802:
+			case DLT_IEEE802_11:
+			case DLT_IEEE802_11_RADIO:
+				if (m->m_flags & M_HASFCS)
+					pktlen-=ETHER_CRC_LEN;
+				break;
+			default:
+				break;
+		}
+	}

-	if (pktlen == m->m_len) {
+	if (frmlen == m->m_len) {
 		cpfn = memcpy;
 		marg = mtod(m, void *);
 		buflen = pktlen;

>Audit-Trail:
From: Martin Husemann <martin@duskware.de>
To: gnats-bugs@NetBSD.org
Cc: 
Subject: Re: kern/28339
Date: Wed, 17 Nov 2004 11:13:16 +0100

 I like this change - but we should have a better way to test an interface
 pointer for the "I am ethernet" property - or at least factor that out
 as a if_is_ether() macro.

 Martin

From: Martin Husemann <martin@duskware.de>
To: gnats-bugs@NetBSD.org
Cc: 
Subject: Re: kern/28339: M_HASFCS not honored in bpf(4)
Date: Mon, 3 Apr 2017 20:56:47 +0200

 This seems still to be wrong - maybe we should decouple M_LINK0
 from M_HASFCS and make M_HASFCS a uniqe flag, then do what this
 PR suggests (extended to the other taps).

 Or has this been worked around in bpf consomers differently and we
 should just improve documentation?

 Martin

Home	PR Database Search