On Thu, Feb 26, 2015 at 9:58 PM, Han Changzhe <hcz@xxxxxxxxxxx> wrote: > Thanks Dave! Please see below. > > Changzhe > > On 2015-02-26 15:30, Dave Taht wrote: >> >> This is not really the best place for a routing question. If you raise >> your question on the quagga list, you might be better off. >> > > I couldn't find a better place for the question and thought guys in LARTC > should be professional enough to help. And thanks for your suggestion on > the quagga list! I'll try it. > >> >> On Wed, Feb 25, 2015 at 10:39 PM, Han Changzhe <hcz@xxxxxxxxxxx> wrote: >>> >>> Hello experts, >>> >>> I'm setting up a routing server on Linux with following links >>> >>> 1. An Ethernet link (eth0) to the 1st internet link (fast, but can't >>> access some sites); >>> 2. A VPN link (tun0) to provide services to local users; >>> 3. A VPN link (tun1) to a proxy server as the 2nd internet link (slow, >>> free). >>> >>> My target is: >>> * for common internet access, routing the packets through eth0; >>> * for the sites can't be accessed through eth0, routing them through >>> tun1. >> >> Well, one of the things we have been working on in the homenet working >> group is >> source specific routing, which could possibly help here, but it is >> non-deterministic. > > > As for the project you mentioned, are there any public materials for me to > follow up? It is one of the products of the ietf homenet working group and mostly oriented around IPv6, but can be used with ipv4. https://plus.google.com/107942175615993706558/posts/PtnPVES2KRd The one, published, RFC https://tools.ietf.org/html/rfc7368 More stuff from homenet that is in progress: https://tools.ietf.org/wg/homenet/ The idea behind source specific routing is, you have multiple uplinks to the internet with distinct IP addresses and delegated subnets, and hosts get more than one IP address - so that when a host sources a query from a different IP address it goes out via a different interface on the router(s). This works well with what is known as "happy eyeballs" - where a browser tries multiple addresses to get out to the web site on the internet - helps on failover, etc. There is some working code - already in openwrt - and available as open source in mainline linux - but it could use some love. "babels" is quite mature - (referenced above) - and hnetd is coming along but not quite baked yet. A lovely use for the idea - that would work in your case were it in more protocols, is in multipath mosh. That version of mosh attempts connections via every way possible, and measures the path with the shortest RTT and uses that. So, in your case, it would probably choose the non-vpn path if not blocked, and the vpn path if it was. https://github.com/boutier/mosh Incidentally most of my vpns are routed over ipv6, over tinc. IPv4 got to be a PITA, now that so many random 10 networks exist in double/triple natted situations. > >>> By now, I set the routing table manually for serveral sites and it works >>> fine. Because there are thousands of them and the sites change with time, >>> so >>> I want a better solution. >>> >>> My idea is like this: setting up more than one default routes for >>> internet >>> access, then dynamically change the route table (or route table cache) >>> with >>> some software according to the internet access results. >>> >>> For example, if we get a timeout from https://www.google.com through >>> eth0, >>> the software should try it through tun1 link and, when succeed, adding >>> the >>> later route to current route table. >> >> Well you are conflating several layers of the protocol here. >> >> It is hard to recognise a timeout, for example, without sniffing for >> syns/syn_acks >> on the gateway. That sniffer could simultaneously try a syn out one of >> the vpn interfaces and if a syn/ack is not received from the main >> interface, and one IS received from the vpn, insert a route for it. > > Yes, a sniffer like that should work. Is it possible to allow the kernel > to raise an exception when TCP connection time-out happens so users can > handle the exception with the hock to try the vpn interface and manipulate > the route table? > >> You would still need to clean out that table periodically. > > Yes. With the old Linux routing cache, it might be easier. Then we only > add routes to the cache. > >> Then you would to insert and delete rules for each ip (or more likely >> network) you wish to reroute >> based on your measurements of what is working or not, and to otherwise >> fall back to the default ethernet route. >> >> Say for example you could not get dns from 8.8.8.8 locally. >> >> ip route add 8.8.8.8 dev tun0 >> >> This doesnt help you on any protocols except tcp. udp apps are >> different. so is quic, etc. >> >> a bulk method would be to go through the alexa top 1 million to see >> what you could and could not access, and set up routes for each (but >> this does not handle your desire for 2 tunnels) > > Actually I prepared a routing list of around 3K items which should > work by now. It's just not beautiful. > > >>> I don't know if any routing software on Linux work as I expected. I tried >>> quagga with zebra + ospf but not successful. >> >> ospf? oy, no.... >> >>> FYI, it's not a common case for link based fail-over/load balance. >>> >>> >>> Please give me suggestions! >> >> Well, my way would probably involve a squid or polipo web proxy to >> make the failover case easier. A lot of users would not dig that... > > It should work for web accessing. But we still need to support ports > for email/ftp and other applications. > > >>> Thanks in advance, >>> >>> Changzhe >>> >>> -- >>> To unsubscribe from this list: send the line "unsubscribe lartc" in >>> the body of a message to majordomo@xxxxxxxxxxxxxxx >>> More majordomo info at http://vger.kernel.org/majordomo-info.html >> >> >> > > -- > To unsubscribe from this list: send the line "unsubscribe lartc" in > the body of a message to majordomo@xxxxxxxxxxxxxxx > More majordomo info at http://vger.kernel.org/majordomo-info.html -- Dave Täht Let's make wifi fast, less jittery and reliable again! https://plus.google.com/u/0/107942175615993706558/posts/TVX3o84jjmb -- To unsubscribe from this list: send the line "unsubscribe lartc" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html