Bug 51783 – [SOUP] Fallback to IPv4 does not work on dual-stack pages (IPv6 + IPv4)

• Component: Page Loading
• Status: RESOLVED FIXED

Created attachment 77729 [details]
strace logfile of luakit michael.stapelberg.de

When having not-working IPv6 connectivity (properly configured system but requests result in 'Network unreachable'), Webkit (I tried with luakit and midori) displays an error instead of retrying with IPv4.

To reproduce, test on a system with native IPv6 connectivity (or at least tunneled on a different host, so that the system you test on thinks it’s natively connected). Using getaddrinfo(3) to resolve an IPv6-enabled host (for example with the attached getaddrinfo.c example) should result in the following output:

michael.stapelberg.de resolved to 2001:4d88:1008:3::ae1
michael.stapelberg.de resolved to 212.86.204.165

Now configure your system to simulate not being able to reach the IPv6 address:

# ip6tables -I OUTPUT 1 -d 2001:4d88:1008:3::ae1 -j REJECT --reject-with icmp6-addr-unreachable

Then start 'luakit michael.stapelberg.de'. You will see an error page, but the site itself is still reachable via IPv4. Use for example 'wget http://michael.stapelberg.de -O/dev/null' to confirm that other tools handle this case correctly.

I’ve attached a complete strace file of 'luakit michael.stapelberg.de'.

Created attachment 77730 [details]
getaddrinfo.c (compile with cc -std=c99 -Wall -D_GNU_SOURCE    getaddrinfo.c   -o getaddrinfo)

[Alexey Proskuryakov]

THis is clearly an issue in network back-end, not in cross-platform WebKit code. Unless I'm mistaken, Midori uses Gtk port of WebKit, which uses SOUP fro networking.

[Dominik Röttsches (drott)]

It seems to me like we need a "happy eyeballs" [1][2] implementation in libsoup to fix that. 
What do you think, Dan? Should this be hidden from ResourceHandleSoup and just be in the library? Or does WebKit's libsoup backend need to know about the "happy eyeballs" algorithm state?

I filed libsoup bug: https://bugzilla.gnome.org/show_bug.cgi?id=687199

[1] http://tools.ietf.org/html/rfc6555
[2] https://bugzilla.mozilla.org/show_bug.cgi?id=684893

[Dan Winship]

Happy eyeballs is about *how fast* you fall back; here it appears that we aren't falling back at all, which is strange. I guess I need to get IPv6 working in my house again...

(And the happy eyeballs stuff can be implemented entirely in libsoup (and gio) without ResourceHandleSoup needing to do anything.)

[Dominik Röttsches (drott)]

(In reply to comment #4)
> Happy eyeballs is about *how fast* you fall back; here it appears that we aren't falling back at all, which is strange. I guess I need to get IPv6 working in my house again...

Fully agree. I meant to comment that if we fix this, we might as well fix it in the right way. Let me know, if you need any help with testing, I have access to native and tunneled IPv6 setups.

> (And the happy eyeballs stuff can be implemented entirely in libsoup (and gio) without ResourceHandleSoup needing to do anything.)

Okay, thanks.

[john]

I am here for the new update just follow me here and get the new post here https://onlinetechnicalhelps.com/snipping-tool-windows-10/ try this way once time i am sore you really like this website forever.

[Patrick Griffis]

This was fixed with GLib 2.60 and newer