Sockets not opened or closed unexpectedly

[ronald-scheckelhoff-trac]

This report pertains to a nightly based install, from the "hrev 49652" image.

Using Webpositive, I found that occasionally pages would not load. Thinking this may not be a Webpositive issue, but instead a network issue, I used Netsurf for a while, and got approximately the same result (occasionally pages would not load). Netsurf displayed the error ("socket not connected), and Webpositive did not, but I suspect the same problem applies to both. I've attached two png images (the Netsurf message, and the debugger screen).

[ronald-scheckelhoff-trac]

"Socket not connected" error

[ronald-scheckelhoff-trac]

Debugger screen that appeared after the "socket not connected" error was displayed.

[ronald-scheckelhoff-trac]

Drilling down further into debugger screen (curl error)

[ronald-scheckelhoff-trac]

I guess that since it was polling, it already had opened successfully, hence wrong title on bug report (head slap).

[ronald-scheckelhoff-trac]

This is from the Netsurf verbose output The error 56 from curl is unknown, perhaps caused by unexpectedly closed socket?

(201.814393) content/fetchers/curl.c fetch_curl_stop 748: fetch 0x18a7e740, url '​http://comments.gmane.org/gmane.os.plan9.general/71902' (279.875120) content/fetchers/curl.c fetch_curl_done 860: done ​http://comments.gmane.org/gmane.os.plan9.general/73371 (279.875191) content/fetchers/curl.c fetch_curl_done 897: Unknown cURL response code 56 (279.875227) content/fetchers/curl.c fetch_curl_stop 748: fetch 0x19306c70, url '​http://comments.gmane.org/gmane.os.plan9.general/73371' (279.900373) beos/gui.cpp void warn_user(const char *, const char *) 924: warn_user: Recv failure: Socket is not connected ((null))

[ronald-scheckelhoff-trac]

Curl error 56 = "connection reset by peer" which can happen, ... but I'm getting this on a number of sites, including once on google.com and multiple times on haiku-os.org and gmane.org. OK, maybe gmane.org is having problems. In such a case, should Webpositive report a reset? As it is, it does nothing, which makes it hard to figure out what's going on. Netsurf reports an error in a popup message, which may be overkill in the other direction.

[AlienSoldier]

I have the same problem, from quite a while. Webpositive just don't load anything, Qupzilla report "socket not connected" when it happen. If i go in the network pref and restart DHCP, once it mention "ready" for the connection, i can refresh the pages and they appear. Quite frustrating to download big or slow file that can't be resumed. Weirdly it only does that on one of my two internet connections (the satellite one, so perhaps it can be triggered more easily on slow connection). In my case it can happen 1 time or more each hour and it seem random, not caused by any site in particular.

[ronald-scheckelhoff-trac]

Replying to AlienSoldier:

I have the same problem, from quite a while. Webpositive just don't load anything, Qupzilla report "socket not connected" when it happen. If i go in the network pref and restart DHCP, once it mention "ready" for the connection, i can refresh the pages and they appear. Quite frustrating to download big or slow file that can't be resumed. Weirdly it only does that on one of my two internet connections (the satellite one, so perhaps it can be triggered more easily on slow connection). In my case it can happen 1 time or more each hour and it seem random, not caused by any site in particular.

I think, like Netsurf, Qupzilla uses cURL? Last night I could not access ​https://www.haiku-os.org in any browser, until after a reboot, but I could ping stuff readily enough. It seems not to be a browser problem, per se, or at least not any specific one. Ping is not TCP ... so I wonder if any TCP related network stuff has been changed recently?

[AlienSoldier]

Do you need absolutely a reboot or are you ready to go again with disabling and re-enabling DHCP?

[ronald-scheckelhoff-trac]

Replying to AlienSoldier:

Do you need absolutely a reboot or are you ready to go again with disabling and re-enabling DHCP?

Well, since I could ping anything I didn't think it should make any difference. But, I'll give it a try ...

[ronald-scheckelhoff-trac]

I guess another thing I could do is to put wireshark on one of the other boxes, to see if there really are any resets coming in, or whether they're being self generated by Haiku ...

[AlienSoldier]

Seem that when it happen i can't ping any site. It work again once i reactivate DHCP dectection that said.

[ronald-scheckelhoff-trac]

@AlienSoldier: Just curious, would you say it's been about two months since this started, or longer than that?

[AlienSoldier]

I would say more than that. Hard to say how long as i only have this problem at this location (still have one month here, after that i will only be able to check that again in June or jully of next year). Next time someone bring a computer here i will try to see if it is related to my hardware.

[pulkomandy]

NetSurf is indeed using cURL. If your error is "connection reset by peer" it probably means something on the line (possibly your ISP) is killing the connection because its not happy about our packets.

There were some recent changes in our TCP code, but it is still not 100% complete, only enough to get things to work.

A wireshark or tcpdump capture of the traffic on the link when you reproduce the issue may be helpful - so we can see the TCP streams, see who is resetting the connection, etc.

[ronald-scheckelhoff-trac]

This is fixed on hrev-49913. So, thanks for that. I'm curious to know what it was.

[waddlesplash]

Please retest with a recent nightly.

[waddlesplash]

No reply.