| Message ID | 4BFECA65.7030102@hp.com |
|---|---|
| State | RFC, archived |
| Delegated to: | David Miller |
| Headers | show |
Hi, Brian Haley <brian.haley@hp.com> writes: >> In previous debug outputs, the content of the fl->oif is ok, i.e. it is >> set to the interface on which the CoA is configured, i.e. the output >> interface. But the commit results in flags |= RT6_LOOKUP_F_IFACE. >> Later, in rt6_score_route(), the call to rt6_check_dev() returns 0 >> (dev->ifindex is ip6tnl1 but oif is wlan0). Because of the change to flags >> flags, we quickly return -1 in rt6_score_route(): > > Ok, so the call to ip6_route_output() was from the tunnel code, which is > using it's cached flowi, which has oif set to the tunnel. The XFRM code > swaps the addresses, which should invalidate the oif, but it doesn't. > >> static int rt6_score_route(struct rt6_info *rt, int oif, >> int strict) >> { >> int m, n; >> >> m = rt6_check_dev(rt, oif); >> if (!m && (strict & RT6_LOOKUP_F_IFACE)) >> return -1; >> ... >> >> Now, I wonder if the following is correct. Don't hesitate to correct me >> if I am wrong: >> >> Initially (before f4f914b58019f0), the purpose of the test using >> rt6_need_strict() in ip6_route_output() (introduced by c71099ac) was to >> allow the multiple routing table logic to be applied to all global >> addresses but to preserve the addresses for which it would not make >> sense (link-local, multicast, ). The change introduced by f4f914b58019f0 >> basically reduces the ability to route traffic as you want and forces >> the traffic to leave the device by the interface on which it is >> configured (if fl->oif is set). > > The problem is we assumed the caller's would only set fl->oif if they > wanted it enforced (multicast, link-local, SO_BINDTODEVICE), but it > didn't take into account the tunnel code. I guess the easy answer > would be to revert this until we can fix it correctly. Nothing against it but maybe Jiri or David have other ideas. >> From my (very limited and possibly wrong) understanding, the change >> introduced by f4f914b58019f0 looks like a workaround for the >> SO_BINDTODEVICE issue. Looking at the code, there is something I don't >> understand: if SO_BINDTODEVICE has been used on a socket, the socket >> should have its sk_bound_dev_if attribute set to the correct ifindex >> value. Hence the following (naive) question: why is that information not >> used to inflect the selection of the route cached for the socket? And >> why would the fix be at the adress level instead of being at the >> interface level (ifindex)? > > I guess I always believed setting SO_BINDTODEVICE should always force > traffic out that interface, but from Yoshifuji's email it seems that > maybe wasn't the intention, at least for things that don't meet > the rt_need_strict() criteria like globals. I don't know the history > behind the setsockopt. The behavior I would expect from a combination of RFC 4191 and SO_BINDTODEVICE sockopt would be the use of the interface as outgoing interface and then the use of the best router (using router preference info, reachability, ...) available on the subnet. IIRC, the router preference info is per default router list in the RFC, i.e. per interface. > The below might actually be what was actually intended, triggering > on what the user forced, rather than assuming all callers require > strict behavior. > > -Brian > > > diff --git a/net/ipv6/route.c b/net/ipv6/route.c > index 294cbe8..252d761 100644 > --- a/net/ipv6/route.c > +++ b/net/ipv6/route.c > @@ -814,7 +814,7 @@ struct dst_entry * ip6_route_output(struct net *net, struct sock *sk, > { > int flags = 0; > > - if (fl->oif || rt6_need_strict(&fl->fl6_dst)) > + if ((sk && sk->sk_bound_dev_if) || rt6_need_strict(&fl->fl6_dst)) > flags |= RT6_LOOKUP_F_IFACE; > > if (!ipv6_addr_any(&fl->fl6_src)) Just for fun, I'll test it tomorrow. I think this would be a better alternative to a simple revert. Cheers, a+ -- 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
All, Thanks for looking into this. The behavior of SO_BINDTODEVICE, certainly with IPV4, is to identify a specific interface to use for sending/receiving AF_INET packets upon. And that's how it has behaved with IPV4. Tools like ping, traceroute (and ping6, traceroute6) make use of it with the -i (interface) options. We ran into this issue when updating an IPV4 application to IPV6. The report provided one example of the issue. The original change was based on the semantics of flowi.oif used in IPV4. There flowi.oif (as it is with IPV6 as well) is also set to sk->sk_bound_dev_if when using ip_route_connect() and routing simply took a non-zero oif to enforce an interface. Looking at IPV4 tunneling, flowi.oif is set the same way from tunnel's parms.link as with IPV6 so would follow those semantics as well. Obviously, with IPV6, the semantics of flowi.oif are different and perhaps even vary with the users of IPV6. If that variability is desired, the proposed change from Brian (using sk->sk_bound_dev_if) would be an equivalent means of supporting SO_BINDTODEVICE while limiting the impact. Scott On 5/27/2010 2:39 PM, Brian Haley wrote: > Hi Arnoud, > > On 05/27/2010 11:14 AM, Arnaud Ebalard wrote: > >>Hi, >> >>Thanks for your reply Brian and sorry for the length of this response. If >>Hideaki and David can comment on the IPv6/XFRM and SO_BINDTODEVICE >>aspects discussed below that would be helpful, IMHO. > > > Thanks for all the analysis, comments below. > > >>>On 05/26/2010 01:01 PM, Arnaud Ebalard wrote: >>> >>>>Hi, >>>> >>>>I just updated my laptop's kernel to 2.6.34 (previously running .33 and >>>>configured to act as an IPsec/IKE-protected MIPv6 Mobile Node using >>>>racoon and umip): after rebooting on the new kernel, the transport mode >>>>SA protecting MIPv6 signaling traffic are missing. >>>> >>>>I bisected the issue down to f4f914b58019f0e50d521bbbadfaee260d766f95 >>>>(net: ipv6 bind to device issue) which was added after 2.6.34-rc5: >>>> >>>>diff --git a/net/ipv6/route.c b/net/ipv6/route.c >>>>index c2438e8..05ebd78 100644 >>>>--- a/net/ipv6/route.c >>>>+++ b/net/ipv6/route.c >>>>@@ -815,7 +815,7 @@ struct dst_entry * ip6_route_output(struct net *net, struct sock *sk, >>>> { >>>> int flags = 0; >>>> >>>>- if (rt6_need_strict(&fl->fl6_dst)) >>>>+ if (fl->oif || rt6_need_strict(&fl->fl6_dst)) >>>> flags |= RT6_LOOKUP_F_IFACE; >>> >>>Can you see if fl->oif is at least a sane value here? Maybe there's some >>>partially un-initialized flowi getting passed-in, a quick source code check >>>didn't find anything obvious. >> >>When it's not 0, fl->oif is a sane value: it is set to the index of the >>interface on which the current *Care-of Address* is configured. All the >>traffic is expected to leave the host via this interface. > > > Ok, so it's not un-initialized data causing this. > > >>In previous debug outputs, the content of the fl->oif is ok, i.e. it is >>set to the interface on which the CoA is configured, i.e. the output >>interface. But the commit results in flags |= RT6_LOOKUP_F_IFACE. >>Later, in rt6_score_route(), the call to rt6_check_dev() returns 0 >>(dev->ifindex is ip6tnl1 but oif is wlan0). Because of the change to flags >>flags, we quickly return -1 in rt6_score_route(): > > > Ok, so the call to ip6_route_output() was from the tunnel code, which is > using it's cached flowi, which has oif set to the tunnel. The XFRM code > swaps the addresses, which should invalidate the oif, but it doesn't. > > >>static int rt6_score_route(struct rt6_info *rt, int oif, >> int strict) >>{ >> int m, n; >> >> m = rt6_check_dev(rt, oif); >> if (!m && (strict & RT6_LOOKUP_F_IFACE)) >> return -1; >> ... >> >>Now, I wonder if the following is correct. Don't hesitate to correct me >>if I am wrong: >> >>Initially (before f4f914b58019f0), the purpose of the test using >>rt6_need_strict() in ip6_route_output() (introduced by c71099ac) was to >>allow the multiple routing table logic to be applied to all global >>addresses but to preserve the addresses for which it would not make >>sense (link-local, multicast, ). The change introduced by f4f914b58019f0 >>basically reduces the ability to route traffic as you want and forces >>the traffic to leave the device by the interface on which it is >>configured (if fl->oif is set). > > > The problem is we assumed the caller's would only set fl->oif if they > wanted it enforced (multicast, link-local, SO_BINDTODEVICE), but it > didn't take into account the tunnel code. I guess the easy answer > would be to revert this until we can fix it correctly. > > >>From my (very limited and possibly wrong) understanding, the change >>introduced by f4f914b58019f0 looks like a workaround for the >>SO_BINDTODEVICE issue. Looking at the code, there is something I don't >>understand: if SO_BINDTODEVICE has been used on a socket, the socket >>should have its sk_bound_dev_if attribute set to the correct ifindex >>value. Hence the following (naive) question: why is that information not >>used to inflect the selection of the route cached for the socket? And >>why would the fix be at the adress level instead of being at the >>interface level (ifindex)? > > > I guess I always believed setting SO_BINDTODEVICE should always force > traffic out that interface, but from Yoshifuji's email it seems that > maybe wasn't the intention, at least for things that don't meet > the rt_need_strict() criteria like globals. I don't know the history > behind the setsockopt. > > The below might actually be what was actually intended, triggering > on what the user forced, rather than assuming all callers require > strict behavior. > > -Brian > > > diff --git a/net/ipv6/route.c b/net/ipv6/route.c > index 294cbe8..252d761 100644 > --- a/net/ipv6/route.c > +++ b/net/ipv6/route.c > @@ -814,7 +814,7 @@ struct dst_entry * ip6_route_output(struct net *net, struct sock *sk, > { > int flags = 0; > > - if (fl->oif || rt6_need_strict(&fl->fl6_dst)) > + if ((sk && sk->sk_bound_dev_if) || rt6_need_strict(&fl->fl6_dst)) > flags |= RT6_LOOKUP_F_IFACE; > > if (!ipv6_addr_any(&fl->fl6_src)) -- 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, Scott C Otto <otts@alcatel-lucent.com> writes: > All, > Thanks for looking into this. > > The behavior of SO_BINDTODEVICE, certainly with IPV4, is to identify a specific > interface to use for sending/receiving AF_INET packets upon. And that's how it > has behaved with IPV4. Tools like ping, traceroute (and ping6, traceroute6) > make use of it with the -i (interface) options. > > We ran into this issue when updating an IPV4 application to IPV6. The report > provided one example of the issue. > > The original change was based on the semantics of flowi.oif used in IPV4. There > flowi.oif (as it is with IPV6 as well) is also set to sk->sk_bound_dev_if when > using ip_route_connect() and routing simply took a non-zero oif to enforce an > interface. Looking at IPV4 tunneling, flowi.oif is set the same way from > tunnel's parms.link as with IPV6 so would follow those semantics as well. > > Obviously, with IPV6, the semantics of flowi.oif are different and perhaps even > vary with the users of IPV6. > > If that variability is desired, the proposed change from Brian (using > sk->sk_bound_dev_if) would be an equivalent means of supporting SO_BINDTODEVICE > while limiting the impact. ok. Thanks for the feedback. Can you comment on what is below? >> The below might actually be what was actually intended, triggering >> on what the user forced, rather than assuming all callers require >> strict behavior. >> >> -Brian >> >> >> diff --git a/net/ipv6/route.c b/net/ipv6/route.c >> index 294cbe8..252d761 100644 >> --- a/net/ipv6/route.c >> +++ b/net/ipv6/route.c >> @@ -814,7 +814,7 @@ struct dst_entry * ip6_route_output(struct net *net, struct sock *sk, >> { >> int flags = 0; >> >> - if (fl->oif || rt6_need_strict(&fl->fl6_dst)) >> + if ((sk && sk->sk_bound_dev_if) || rt6_need_strict(&fl->fl6_dst)) >> flags |= RT6_LOOKUP_F_IFACE; >> >> if (!ipv6_addr_any(&fl->fl6_src)) Brian, I tested the patch on my Mobile Node: it fixes the regression. I also updated the kernel on my Home Agent to a 2.6.34 with that fix and everything works as expected. *For that aspect* and fwiw, you get my Tested-by: Arnaud Ebalard <arno@natisbad.org> For the SO_BINDTODEVICE aspect, I don't have code at hand to test if the fix works as expected. We should also double check that this will not break other paths which use the sk->sk_bound_dev_if with a different semantic: $ grep -R sk_bound_dev_if net/ | wc -l 125 $ grep -R 'sk_bound_dev_if = ' net/ net/ieee802154/raw.c: sk->sk_bound_dev_if = dev->ifindex; net/core/sock.c: sk->sk_bound_dev_if = index; net/ipv6/datagram.c: sk->sk_bound_dev_if = usin->sin6_scope_id; net/ipv6/datagram.c: sk->sk_bound_dev_if = np->mcast_oif; net/ipv6/af_inet6.c: sk->sk_bound_dev_if = addr->sin6_scope_id; net/ipv6/tcp_ipv6.c: sk->sk_bound_dev_if = usin->sin6_scope_id; net/ipv6/tcp_ipv6.c: newsk->sk_bound_dev_if = treq->iif; net/ipv6/raw.c: sk->sk_bound_dev_if = addr->sin6_scope_id; net/sctp/socket.c: newsk->sk_bound_dev_if = sk->sk_bound_dev_if; net/ipv4/ip_output.c: sk->sk_bound_dev_if = arg->bound_dev_if; net/ipv4/udp.c: sk->sk_bound_dev_if = 0; net/dccp/ipv6.c: newsk->sk_bound_dev_if = ireq6->iif; net/dccp/ipv6.c: sk->sk_bound_dev_if = usin->sin6_scope_id; Cheers, a+ -- 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
On 05/28/2010 04:51 AM, Arnaud Ebalard wrote: >>> The below might actually be what was actually intended, triggering >>> on what the user forced, rather than assuming all callers require >>> strict behavior. >>> >>> -Brian >>> >>> >>> diff --git a/net/ipv6/route.c b/net/ipv6/route.c >>> index 294cbe8..252d761 100644 >>> --- a/net/ipv6/route.c >>> +++ b/net/ipv6/route.c >>> @@ -814,7 +814,7 @@ struct dst_entry * ip6_route_output(struct net *net, struct sock *sk, >>> { >>> int flags = 0; >>> >>> - if (fl->oif || rt6_need_strict(&fl->fl6_dst)) >>> + if ((sk && sk->sk_bound_dev_if) || rt6_need_strict(&fl->fl6_dst)) >>> flags |= RT6_LOOKUP_F_IFACE; >>> >>> if (!ipv6_addr_any(&fl->fl6_src)) > > Brian, I tested the patch on my Mobile Node: it fixes the regression. I > also updated the kernel on my Home Agent to a 2.6.34 with that fix and > everything works as expected. *For that aspect* and fwiw, you get my > > Tested-by: Arnaud Ebalard <arno@natisbad.org> Ok, thanks for testing, I'll send out an updated patch, with the caveat there still might be a DCCP regression, see below. > For the SO_BINDTODEVICE aspect, I don't have code at hand to test if the > fix works as expected. We should also double check that this will not > break other paths which use the sk->sk_bound_dev_if with a different > semantic: > net/ieee802154/raw.c: sk->sk_bound_dev_if = dev->ifindex; Isn't AF_INET6, OK. > net/core/sock.c: sk->sk_bound_dev_if = index; The actual setsockopt() call, OK. > net/ipv6/datagram.c: sk->sk_bound_dev_if = usin->sin6_scope_id; > net/ipv6/datagram.c: sk->sk_bound_dev_if = np->mcast_oif; > net/ipv6/af_inet6.c: sk->sk_bound_dev_if = addr->sin6_scope_id; > net/ipv6/tcp_ipv6.c: sk->sk_bound_dev_if = usin->sin6_scope_id; > net/ipv6/tcp_ipv6.c: newsk->sk_bound_dev_if = treq->iif; > net/ipv6/raw.c: sk->sk_bound_dev_if = addr->sin6_scope_id; This are all in link-local or multicast code, caller passing-in scopeid, would trigger rt6_need_strict() regardless, so are OK. > net/sctp/socket.c: newsk->sk_bound_dev_if = sk->sk_bound_dev_if; SCTP peeling-off a socket, cloning parent, OK. > net/ipv4/ip_output.c: sk->sk_bound_dev_if = arg->bound_dev_if; > net/ipv4/udp.c: sk->sk_bound_dev_if = 0; IPv4, shouldn't be affected by an IPv6 route change, OK. > net/dccp/ipv6.c: newsk->sk_bound_dev_if = ireq6->iif; This is the only mystery in this list. Looks like the DCCP accept() codepath, and it's getting bound to the interface the request was received on. I'm not sure if the intention here was to force this strict behavior or not. > net/dccp/ipv6.c: sk->sk_bound_dev_if = usin->sin6_scope_id; In link-local code, caller passing-in scopeid, would trigger rt6_need_strict() regardless, so OK. -Brian -- 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
Hello. (2010/05/28 6:01), Arnaud Ebalard wrote: > Hi, > > Brian Haley<brian.haley@hp.com> writes: > >>> In previous debug outputs, the content of the fl->oif is ok, i.e. it is >>> set to the interface on which the CoA is configured, i.e. the output >>> interface. But the commit results in flags |= RT6_LOOKUP_F_IFACE. >>> Later, in rt6_score_route(), the call to rt6_check_dev() returns 0 >>> (dev->ifindex is ip6tnl1 but oif is wlan0). Because of the change to flags >>> flags, we quickly return -1 in rt6_score_route(): >> >> Ok, so the call to ip6_route_output() was from the tunnel code, which is >> using it's cached flowi, which has oif set to the tunnel. The XFRM code >> swaps the addresses, which should invalidate the oif, but it doesn't. >> >>> static int rt6_score_route(struct rt6_info *rt, int oif, >>> int strict) >>> { >>> int m, n; >>> >>> m = rt6_check_dev(rt, oif); >>> if (!m&& (strict& RT6_LOOKUP_F_IFACE)) >>> return -1; >>> ... >>> >>> Now, I wonder if the following is correct. Don't hesitate to correct me >>> if I am wrong: >>> >>> Initially (before f4f914b58019f0), the purpose of the test using >>> rt6_need_strict() in ip6_route_output() (introduced by c71099ac) was to >>> allow the multiple routing table logic to be applied to all global >>> addresses but to preserve the addresses for which it would not make >>> sense (link-local, multicast, ). The change introduced by f4f914b58019f0 >>> basically reduces the ability to route traffic as you want and forces >>> the traffic to leave the device by the interface on which it is >>> configured (if fl->oif is set). >> >> The problem is we assumed the caller's would only set fl->oif if they >> wanted it enforced (multicast, link-local, SO_BINDTODEVICE), but it >> didn't take into account the tunnel code. I guess the easy answer >> would be to revert this until we can fix it correctly. > > Nothing against it but maybe Jiri or David have other ideas. > > >>> From my (very limited and possibly wrong) understanding, the change >>> introduced by f4f914b58019f0 looks like a workaround for the >>> SO_BINDTODEVICE issue. Looking at the code, there is something I don't >>> understand: if SO_BINDTODEVICE has been used on a socket, the socket >>> should have its sk_bound_dev_if attribute set to the correct ifindex >>> value. Hence the following (naive) question: why is that information not >>> used to inflect the selection of the route cached for the socket? And >>> why would the fix be at the adress level instead of being at the >>> interface level (ifindex)? >> >> I guess I always believed setting SO_BINDTODEVICE should always force >> traffic out that interface, but from Yoshifuji's email it seems that >> maybe wasn't the intention, at least for things that don't meet >> the rt_need_strict() criteria like globals. I don't know the history >> behind the setsockopt. > > The behavior I would expect from a combination of RFC 4191 and > SO_BINDTODEVICE sockopt would be the use of the interface as outgoing > interface and then the use of the best router (using router preference > info, reachability, ...) available on the subnet. IIRC, the router > preference info is per default router list in the RFC, i.e. per > interface. Good point. Whatever our original intention/thought was, current RFC says that we should honor outgoing interface specified by user (by IPV6_PKTINFO etc.), as we do for SO_BINDTODEVICE in IPv4 as well. In this sense, checking sk->sk_bound_dev_if in ip6_route_output() is not enough because we need to take outgoing interface specified in ancillary data into account, which is set to fl->oif. How about adding additional "flags" parameter for ip6_route_output()? Thoughts? --yoshfuji -- 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, YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org> writes: >>> I guess I always believed setting SO_BINDTODEVICE should always force >>> traffic out that interface, but from Yoshifuji's email it seems that >>> maybe wasn't the intention, at least for things that don't meet >>> the rt_need_strict() criteria like globals. I don't know the history >>> behind the setsockopt. >> >> The behavior I would expect from a combination of RFC 4191 and >> SO_BINDTODEVICE sockopt would be the use of the interface as outgoing >> interface and then the use of the best router (using router preference >> info, reachability, ...) available on the subnet. IIRC, the router >> preference info is per default router list in the RFC, i.e. per >> interface. > > Good point. > > Whatever our original intention/thought was, > current RFC says that we should honor outgoing interface > specified by user (by IPV6_PKTINFO etc.), as we do for > SO_BINDTODEVICE in IPv4 as well. > > In this sense, checking sk->sk_bound_dev_if in > ip6_route_output() is not enough because we need to > take outgoing interface specified in ancillary data > into account, which is set to fl->oif. > > How about adding additional "flags" parameter > for ip6_route_output()? I think this may provide a better long term solution but getting all combinations of cases (SO_BINDTODEVICE and other IPv6 sockopts) work together (possibly with external info like RFC 4191 ones gathered from RA or specific local routing config) will be a bit tricky. Meanwhile, regarding the regression, as Brian's fix handles most cases, I think it would be useful to apply it and push it to the stable team. Cheers, a+ -- 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/ipv6/route.c b/net/ipv6/route.c index 294cbe8..252d761 100644 --- a/net/ipv6/route.c +++ b/net/ipv6/route.c @@ -814,7 +814,7 @@ struct dst_entry * ip6_route_output(struct net *net, struct sock *sk, { int flags = 0; - if (fl->oif || rt6_need_strict(&fl->fl6_dst)) + if ((sk && sk->sk_bound_dev_if) || rt6_need_strict(&fl->fl6_dst)) flags |= RT6_LOOKUP_F_IFACE; if (!ipv6_addr_any(&fl->fl6_src))