From mboxrd@z Thu Jan 1 00:00:00 1970 From: "James R. Leu" Subject: Re: send to self patch Date: Thu, 28 Aug 2003 14:22:47 -0500 Sender: netdev-bounce@oss.sgi.com Message-ID: <20030828142246.A6499@mindspring.com> References: <20030828174859.GG12541@krispykreme> Reply-To: jleu@mindspring.com Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: netdev@oss.sgi.com Return-path: To: Anton Blanchard Content-Disposition: inline In-Reply-To: <20030828174859.GG12541@krispykreme>; from anton@samba.org on Fri, Aug 29, 2003 at 03:48:59AM +1000 Errors-to: netdev-bounce@oss.sgi.com List-Id: netdev.vger.kernel.org While the various 'send-to-self' patches are useful for exactly what you are trying to do, I think a more generic solution, which has as one of it's features the ability to 'send-to-self', should be integrated into the kernel (I think it might be too late for anything new to be added to 2.6). Since I'm not the type to criticize without providing an alternate solution, take a look at http://linux-vrf.sf.net/ for my attempt at solving a larger problem (virtual routing and forwarding), which also solves the 'send-to-self' problem (by creating a vrf and place one of the 'send-to-self' interfaces in that vrf). The patch is against a 2.4.19 kernel, but should apply to 2.4.20-21. On Fri, Aug 29, 2003 at 03:48:59AM +1000, Anton Blanchard wrote: > > Hi, > > Here in the lab we often want to send TCP/IP data out over looped back > networks. We do have tools to send and receive raw packets but sometimes > thats not enough. > > eg When running netperf, you dont always need two machines. If you are > just trying to stress a box (eg to find a bug like the NAPI one we hit a > week or two ago) then running loopback is fine. > > We also have exercisers that run over TCP/IP and checksum sent data vs > received data. > > I stumbled upon a patch by Julian Anastasov, and have found it works well. > Is there any chance of getting this (or something like it) into 2.6? > > http://www.ssi.bg/~ja/#loop > > Anton > > diff -ur v2.5.73/linux/Documentation/filesystems/proc.txt linux/Documentation/filesystems/proc.txt > --- v2.5.73/linux/Documentation/filesystems/proc.txt Sun Jun 15 10:23:55 2003 > +++ linux/Documentation/filesystems/proc.txt Tue Jul 1 23:29:02 2003 > @@ -1502,6 +1502,15 @@ > > Log packets with source addresses with no known route to kernel log. > > +loop > +---- > + > +By default (loop=0) the traffic between local IP addresses > +is routed via interface "lo". Setting this flag for two > +interfaces allows traffic between their IP addresses to > +be looped externally. This is useful for setups where the > +interfaces are attached to same broadcast medium. > + > mc_forwarding > ------------- > > diff -ur v2.5.73/linux/Documentation/networking/ip-sysctl.txt linux/Documentation/networking/ip-sysctl.txt > --- v2.5.73/linux/Documentation/networking/ip-sysctl.txt Sun Jun 15 10:23:56 2003 > +++ linux/Documentation/networking/ip-sysctl.txt Tue Jul 1 23:27:02 2003 > @@ -390,6 +390,13 @@ > forwarding - BOOLEAN > Enable IP forwarding on this interface. > > +loop - BOOLEAN > + By default (loop=0) the traffic between local IP addresses > + is routed via interface "lo". Setting this flag for two > + interfaces allows traffic between their IP addresses to > + be looped externally. This is useful for setups where the > + interfaces are attached to same broadcast medium. > + > mc_forwarding - BOOLEAN > Do multicast routing. The kernel needs to be compiled with CONFIG_MROUTE > and a multicast routing daemon is required. > diff -ur v2.5.73/linux/include/linux/inetdevice.h linux/include/linux/inetdevice.h > --- v2.5.73/linux/include/linux/inetdevice.h Sun Apr 20 11:02:16 2003 > +++ linux/include/linux/inetdevice.h Mon Jun 30 23:46:20 2003 > @@ -21,6 +21,7 @@ > int medium_id; > int no_xfrm; > int no_policy; > + int loop; > void *sysctl; > }; > > @@ -70,6 +71,7 @@ > (ipv4_devconf.accept_redirects || (in_dev)->cnf.accept_redirects))) > > #define IN_DEV_ARPFILTER(in_dev) (ipv4_devconf.arp_filter || (in_dev)->cnf.arp_filter) > +#define IN_DEV_LOOP(in_dev) ((in_dev)->cnf.loop) > > struct in_ifaddr > { > diff -ur v2.5.73/linux/include/linux/sysctl.h linux/include/linux/sysctl.h > --- v2.5.73/linux/include/linux/sysctl.h Sun Jun 15 10:25:03 2003 > +++ linux/include/linux/sysctl.h Mon Jun 30 23:47:14 2003 > @@ -359,6 +359,7 @@ > NET_IPV4_CONF_MEDIUM_ID=14, > NET_IPV4_CONF_NOXFRM=15, > NET_IPV4_CONF_NOPOLICY=16, > + NET_IPV4_CONF_LOOP=17, > }; > > /* /proc/sys/net/ipv6 */ > diff -ur v2.5.73/linux/net/ipv4/devinet.c linux/net/ipv4/devinet.c > --- v2.5.73/linux/net/ipv4/devinet.c Sun Jun 15 10:25:07 2003 > +++ linux/net/ipv4/devinet.c Mon Jun 30 23:49:26 2003 > @@ -1122,7 +1122,7 @@ > > static struct devinet_sysctl_table { > struct ctl_table_header *sysctl_header; > - ctl_table devinet_vars[17]; > + ctl_table devinet_vars[18]; > ctl_table devinet_dev[2]; > ctl_table devinet_conf_dir[2]; > ctl_table devinet_proto_dir[2]; > @@ -1258,6 +1258,14 @@ > .mode = 0644, > .proc_handler = &ipv4_doint_and_flush, > .strategy = &ipv4_doint_and_flush_strategy, > + }, > + { > + .ctl_name = NET_IPV4_CONF_LOOP, > + .procname = "loop", > + .data = &ipv4_devconf.loop, > + .maxlen = sizeof(int), > + .mode = 0644, > + .proc_handler = &proc_dointvec, > }, > }, > .devinet_dev = { > diff -ur v2.5.73/linux/net/ipv4/fib_frontend.c linux/net/ipv4/fib_frontend.c > --- v2.5.73/linux/net/ipv4/fib_frontend.c Wed May 28 00:02:50 2003 > +++ linux/net/ipv4/fib_frontend.c Mon Jun 30 23:52:38 2003 > @@ -167,15 +167,16 @@ > .tos = tos } }, > .iif = oif }; > struct fib_result res; > - int no_addr, rpf; > + int no_addr, rpf, loop; > int ret; > > - no_addr = rpf = 0; > + no_addr = rpf = loop = 0; > read_lock(&inetdev_lock); > in_dev = __in_dev_get(dev); > if (in_dev) { > no_addr = in_dev->ifa_list == NULL; > rpf = IN_DEV_RPFILTER(in_dev); > + loop = IN_DEV_LOOP(in_dev); > } > read_unlock(&inetdev_lock); > > @@ -184,6 +185,11 @@ > > if (fib_lookup(&fl, &res)) > goto last_resort; > + if (loop && res.type == RTN_LOCAL) { > + *spec_dst = FIB_RES_PREFSRC(res); > + fib_res_put(&res); > + return 0; > + } > if (res.type != RTN_UNICAST) > goto e_inval_res; > *spec_dst = FIB_RES_PREFSRC(res); > diff -ur v2.5.73/linux/net/ipv4/route.c linux/net/ipv4/route.c > --- v2.5.73/linux/net/ipv4/route.c Tue Jun 17 21:39:39 2003 > +++ linux/net/ipv4/route.c Tue Jul 1 00:29:05 2003 > @@ -1967,6 +1967,11 @@ > dev_put(dev_out); > goto out; /* Wrong error code */ > } > + err = -ENETDOWN; > + if (!(dev_out->flags&IFF_UP)) { > + dev_put(dev_out); > + goto out; > + } > > if (LOCAL_MCAST(oldflp->fl4_dst) || oldflp->fl4_dst == 0xFFFFFFFF) { > if (!fl.fl4_src) > @@ -2036,10 +2041,41 @@ > goto e_inval; > > if (res.type == RTN_LOCAL) { > - if (!fl.fl4_src) > - fl.fl4_src = fl.fl4_dst; > + struct in_device *in_dev; > + u32 src; > + > if (dev_out) > dev_put(dev_out); > + dev_out = FIB_RES_DEV(res); > + in_dev = in_dev_get(dev_out); > + src = fl.fl4_src? : FIB_RES_PREFSRC(res); > + if (in_dev && IN_DEV_LOOP(in_dev) && src) { > + struct net_device *dev_src; > + > + in_dev_put(in_dev); > + in_dev = NULL; > + dev_src = ip_dev_find(src); > + if (dev_src && dev_src != dev_out && > + (in_dev = in_dev_get(dev_src)) && > + IN_DEV_LOOP(in_dev)) { > + in_dev_put(in_dev); > + dev_out = dev_src; > + fl.fl4_src = src; > + fl.oif = dev_out->ifindex; > + res.type = RTN_UNICAST; > + if (res.fi) { > + fib_info_put(res.fi); > + res.fi = NULL; > + } > + goto make_route; > + } > + if (dev_src) > + dev_put(dev_src); > + } > + if (in_dev) > + in_dev_put(in_dev); > + if (!fl.fl4_src) > + fl.fl4_src = fl.fl4_dst; > dev_out = &loopback_dev; > dev_hold(dev_out); > fl.oif = dev_out->ifindex; -- James R. Leu