Message ID | 1442822409-9799-1-git-send-email-wen.gang.wang@oracle.com |
---|---|
State | Changes Requested, archived |
Delegated to: | David Miller |
Headers | show |
From: Wengang Wang <wen.gang.wang@oracle.com> Date: Mon, 21 Sep 2015 16:00:09 +0800 > This is the v2, comparing the v1, the changes is: > * for loopback outbound device, it continue skipping cached route; > for others, it goes through the cached route. There are many things I want you to clean up in your commit message. > For multi-cast, we should find valid route(thus get the meaniful pmtu) for > the package on the socket which is not bound to a device(sk_bound_dev_if > being 0) too. Please use the term "packet" rather than "package". Gifts you receive from a friend might be in a "package", but data is sent onto a network inside of "packets". > A problem is hit that all multi-cast packages dropped by kernel(from sender "multicast" is a single word. > host). The lower layer is IPoIB with MTU being 7000. And I was sending 4096 > length multi-cast package. In side IPoIB the first send is dropped because "Inside" is one word. -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Hi David, 在 2015年09月25日 05:22, David Miller 写道: > From: Wengang Wang <wen.gang.wang@oracle.com> > Date: Mon, 21 Sep 2015 16:00:09 +0800 > >> This is the v2, comparing the v1, the changes is: >> * for loopback outbound device, it continue skipping cached route; >> for others, it goes through the cached route. > There are many things I want you to clean up in your commit message. > >> For multi-cast, we should find valid route(thus get the meaniful pmtu) for >> the package on the socket which is not bound to a device(sk_bound_dev_if >> being 0) too. > Please use the term "packet" rather than "package". Gifts you receive > from a friend might be in a "package", but data is sent onto a network > inside of "packets". > >> A problem is hit that all multi-cast packages dropped by kernel(from sender > "multicast" is a single word. > >> host). The lower layer is IPoIB with MTU being 7000. And I was sending 4096 >> length multi-cast package. In side IPoIB the first send is dropped because > "Inside" is one word. Accepted, sorry for my bad English. I will re-post as you suggested. thanks, wengang -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
diff --git a/net/ipv4/route.c b/net/ipv4/route.c index 5f4a556..c0534c2 100644 --- a/net/ipv4/route.c +++ b/net/ipv4/route.c @@ -2097,7 +2097,10 @@ struct rtable *__ip_route_output_key(struct net *net, struct flowi4 *fl4) */ fl4->flowi4_oif = dev_out->ifindex; - goto make_route; + if (dev_out->flags & IFF_LOOPBACK) + goto make_route; + else + goto lookup; } if (!(fl4->flowi4_flags & FLOWI_FLAG_ANYSRC)) { @@ -2153,6 +2156,7 @@ struct rtable *__ip_route_output_key(struct net *net, struct flowi4 *fl4) goto make_route; } +lookup: if (fib_lookup(net, fl4, &res, 0)) { res.fi = NULL; res.table = NULL;
This is the v2, comparing the v1, the changes is: * for loopback outbound device, it continue skipping cached route; for others, it goes through the cached route. For multi-cast, we should find valid route(thus get the meaniful pmtu) for the package on the socket which is not bound to a device(sk_bound_dev_if being 0) too. From man page of socket(7) SO_BINDTODEVICE Bind this socket to a particular device like “eth0”, as specified in the passed interface name. If the name is an empty string or the option length is zero, the socket device binding is removed. The passed option is a variable-length null-terminated interface name string with the maximum size of IFNAMSIZ. If a socket is bound to an interface, only packets received from that particular interface are processed by the socket. Note that this works only for some socket types, particularly AF_INET sockets. It is not supported for packet sockets (use normal bind(2) there). The man page doesn't say when socket not bound packages won't be routed. A problem is hit that all multi-cast packages dropped by kernel(from sender host). The lower layer is IPoIB with MTU being 7000. And I was sending 4096 length multi-cast package. In side IPoIB the first send is dropped because is exeeding the internal package size limitation mcast_mtu which is 2044. So IPoIB calls ip_rt_update_pmtu (indirectly) trying to set path mtu. A correct route is configured for the multi-cast, so the setting of pmtu cucceeded and the next multi-cast package(to the same target) is expected to succeed(it would be well fragmented accroding to the pmtu I just set). But actually the second and later multi-cast packages got dropped too. And the reason is that the neighor looking up(fib_lookup) is skipped because of the socket is not bound to device(sk_bound_dev_if being 0). After applied the patch I proposed here, it works fine. Signed-off-by: Wengang Wang <wen.gang.wang@oracle.com> --- net/ipv4/route.c | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-)