GNOME Bugzilla – Bug 786432
Connectivity Checker reports limited connectivity when it should be full
Last modified: 2017-08-27 10:54:22 UTC
I'm testing this on Ubuntu Artful running NetworkManager 1.8.2. The uri for portal checking is currently set to http://network-test.debian.org/nm Usually n-m reports that the connectivity check has timed out. I can ping the server, and curl returns the page correctly with the required headers visible. Sometimes though, it works as expected. But this is rare. Typical logs: # /usr/sbin/NetworkManager -d -n --log-level=TRACE --log-domains=CONCHECK <debug> [1502985413.6138] connectivity: (wlp3s0) sending request to 'http://network-test.debian.org/nm' <trace> [1502985413.6412] connectivity: timeout elapsed - multi_socket_action (1 conn remaining) <trace> [1502985413.6933] connectivity: timeout elapsed - multi_socket_action (1 conn remaining) <info> [1502985443.8223] connectivity: (wlp3s0) timed out <debug> [1502985443.8224] device[0xe5432a0e0] (wlp3s0): state changed from UNKNOWN to LIMITED <debug> [1502985711.8235] connectivity: (wlp3s0) sending request to 'http://network-test.debian.org/nm' <trace> [1502985711.8238] connectivity: timeout elapsed - multi_socket_action (1 conn remaining) <trace> [1502985711.8239] connectivity: timeout elapsed - multi_socket_action (1 conn remaining) <trace> [1502985712.8255] connectivity: timeout elapsed - multi_socket_action (1 conn remaining) <info> [1502985741.8229] connectivity: (wlp3s0) timed out <debug> [1502986011.8242] connectivity: (wlp3s0) sending request to 'http://network-test.debian.org/nm' <trace> [1502986011.8244] connectivity: timeout elapsed - multi_socket_action (1 conn remaining) <trace> [1502986011.8245] connectivity: timeout elapsed - multi_socket_action (1 conn remaining) <trace> [1502986012.8260] connectivity: timeout elapsed - multi_socket_action (1 conn remaining) <info> [1502986041.8234] connectivity: (wlp3s0) timed out Any tips on further troubleshooting much appreciated.
Created attachment 357825 [details] Packet capture of a curl first, and then n-m trying I used Wireshark to do a packet capture to see if I can get some more info on what's going on here. A curl fetch is done first, then 30 seconds later or so n-m does it's thing. You can see that the n-m attempt is seemingly malformed, and never actually GETs the page.
I also see this on Fedora 27 with NetworkManager-1.8.2-3.fc27.2.x86_64 I keep getting those timeout errors as well. For example, 3 minutes ago: Aug 25 13:51:45 classic NetworkManager[908]: <info> [1503661905.5660] connectivity: (wlp2s0) timed out $ journalctl -b | grep "timed out" | wc -l 744 $ uptime 13:54:30 up 3 days, 19:23, 5 users, load average: 0.48, 0.51, 0.47
This seems to be the right fix. In some initial testing here it seems to have done the trick. https://git.archlinux.org/svntogit/packages.git/commit/trunk?h=packages/curl&id=058482ae218dca2475db23046bf513ab4e9a6601
Upstream patch reference: https://github.com/curl/curl/pull/1803 Thomas, can you make sure that the patch gets added to the Fedora package? Thanks! (I'll assume that Will will report the bug for Ubuntu ;)
(In reply to Bastien Nocera from comment #4) > Upstream patch reference: > https://github.com/curl/curl/pull/1803 > > Thomas, can you make sure that the patch gets added to the Fedora package? > Thanks! (I'll assume that Will will report the bug for Ubuntu ;) Reference to downstream Fedora bug: https://bugzilla.redhat.com/show_bug.cgi?id=1485702