Installer Problem: DCHP client terminates right after
Hi folks, yesterday I reported bug https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=757711, but failed to let the BTS properly report to this list. It is basically about the DHCP client terminating directly after network detection (after receiving signal 15), which interferes later with the mirror selection, as there is no Internet connection then. Is there anything recently changed that might cause this problem? -- Best Regrads Jan signature.asc Description: OpenPGP digital signature
Re: Bug#757711: netcfg: promptly kills dhclient, deconfigures interface
retitle 757711 netcfg: promptly kills dhclient, deconfigures interface reassign 757711 src:netcfg found 757711 netcfg/1.118 severity 757711 grave user debian-bsd@lists.debian.org usertags 757711 + kfreebsd thanks Hi, On 11/08/14 08:18, Jan Henke wrote: [...] DHCP client terminating directly after network detection (after receiving signal 15), which interferes later with the mirror selection, as there is no Internet connection then. More specifically, this is what I see happening: * dhclient correctly gets a DHCP lease and correctly sets the IPv4 address on the interface * if it doesn't know yet what hostname to use, netcfg prompts the user to enter one (and the DNS search domain) * right after entering those, dhclient receives SIGTERM, gives up the DHCP lease and deconfigures the interface. Aug 11 11:12:09 netcfg[1872]: WARNING **: Started DHCP client; PID is 1919 Aug 11 11:12:10 dhclient: DHCPDISCOVER on em0 to 255.255.255.255 port 67 interval 1 Aug 11 11:12:10 dhclient: DHCPREQUEST on em0 to 255.255.255.255 port 67 Aug 11 11:12:10 dhclient: DHCPOFFER from 10.0.2.2 Aug 11 11:12:10 dhclient: DHCPACK from 10.0.2.2 Aug 11 11:12:10 dhclient: bound to 10.0.2.15 -- renewal in 41223 seconds. Aug 11 11:12:12 netcfg[1872]: DEBUG: Reading domain name returned via DHCP Aug 11 11:12:12 netcfg[1872]: DEBUG: DHCP domain name is '' Aug 11 11:12:12 netcfg[1872]: DEBUG: Reading nameservers from /etc/resolv.conf Aug 11 11:12:12 netcfg[1872]: DEBUG: Read nameserver 10.0.2.3 Aug 11 11:12:12 netcfg[1872]: DEBUG: State is now 1 Aug 11 11:12:12 netcfg[1872]: DEBUG: State is now 2 Aug 11 11:12:12 netcfg[1872]: DEBUG: State is now 5 Aug 11 11:12:12 netcfg[1872]: DEBUG: Using DNS to try and obtain default hostname Aug 11 11:12:12 netcfg[1872]: DEBUG: Getting default hostname from rDNS lookup of autoconfigured address fe80::5054:ff:fe12:3456 Aug 11 11:12:12 netcfg[1872]: DEBUG: getnameinfo() returned -2 (Name or service not known) Aug 11 11:12:12 netcfg[1872]: DEBUG: Getting default hostname from rDNS lookup of autoconfigured address 10.0.2.15 Aug 11 11:12:12 netcfg[1872]: DEBUG: getnameinfo() returned -2 (Name or service not known) ...then hostname and domain are provided... Aug 11 11:12:22 netcfg[1872]: DEBUG: State is now 6 Aug 11 11:12:23 netcfg[1872]: DEBUG: Network config complete Aug 11 11:12:23 netcfg[1872]: DEBUG: No interface given; clearing /etc/network/interfaces Aug 11 11:12:23 netcfg[1872]: DEBUG: Writing informative header Aug 11 11:12:23 netcfg[1872]: DEBUG: Success! Aug 11 11:12:23 netcfg[1872]: DEBUG: Writing loopback interface Aug 11 11:12:23 netcfg[1872]: DEBUG: Success! Aug 11 11:12:23 netcfg[1872]: DEBUG: Writing DHCP stanza for em0 Aug 11 11:12:23 netcfg[1872]: INFO: No hotpluggable devices are present in the system. Aug 11 11:12:23 netcfg[1872]: DEBUG: Success! Aug 11 11:12:23 dhclient: Received signal 15, initiating shutdown. Aug 11 11:12:23 dhclient: DHCPRELEASE on em0 to 10.0.2.2 port 67 Incriminating code is at: http://anonscm.debian.org/cgit/d-i/netcfg.git/tree/dhcp.c#n617 Is it really intended that kill_dhcp_client will deconfigure the interface? Why was this not a problem for wheezy d-i? Is it supposed to be brought back up again in a later step? Regards, -- Steven Chamberlain ste...@pyro.eu.org signature.asc Description: OpenPGP digital signature
Re: Bug#757711: netcfg: promptly kills dhclient, deconfigures interface
Steven Chamberlain ste...@pyro.eu.org (2014-08-11): retitle 757711 netcfg: promptly kills dhclient, deconfigures interface reassign 757711 src:netcfg found 757711 netcfg/1.118 severity 757711 grave user debian-bsd@lists.debian.org usertags 757711 + kfreebsd thanks Steve, if you want to interact with the maintainer of a package you're reassigning a bug to, you need to add the said maintainer to Cc. Otherwise the received mail only contains the results of the control commands: https://lists.debian.org/debian-boot/2014/08/msg00215.html Now that -boot@ owns the bug report, keeping the bug report, -bsd@, and the submitter in the loop should be sufficient. Full quoting follows, for those not following -bsd@. Mraw, KiBi. Hi, On 11/08/14 08:18, Jan Henke wrote: [...] DHCP client terminating directly after network detection (after receiving signal 15), which interferes later with the mirror selection, as there is no Internet connection then. More specifically, this is what I see happening: * dhclient correctly gets a DHCP lease and correctly sets the IPv4 address on the interface * if it doesn't know yet what hostname to use, netcfg prompts the user to enter one (and the DNS search domain) * right after entering those, dhclient receives SIGTERM, gives up the DHCP lease and deconfigures the interface. Aug 11 11:12:09 netcfg[1872]: WARNING **: Started DHCP client; PID is 1919 Aug 11 11:12:10 dhclient: DHCPDISCOVER on em0 to 255.255.255.255 port 67 interval 1 Aug 11 11:12:10 dhclient: DHCPREQUEST on em0 to 255.255.255.255 port 67 Aug 11 11:12:10 dhclient: DHCPOFFER from 10.0.2.2 Aug 11 11:12:10 dhclient: DHCPACK from 10.0.2.2 Aug 11 11:12:10 dhclient: bound to 10.0.2.15 -- renewal in 41223 seconds. Aug 11 11:12:12 netcfg[1872]: DEBUG: Reading domain name returned via DHCP Aug 11 11:12:12 netcfg[1872]: DEBUG: DHCP domain name is '' Aug 11 11:12:12 netcfg[1872]: DEBUG: Reading nameservers from /etc/resolv.conf Aug 11 11:12:12 netcfg[1872]: DEBUG: Read nameserver 10.0.2.3 Aug 11 11:12:12 netcfg[1872]: DEBUG: State is now 1 Aug 11 11:12:12 netcfg[1872]: DEBUG: State is now 2 Aug 11 11:12:12 netcfg[1872]: DEBUG: State is now 5 Aug 11 11:12:12 netcfg[1872]: DEBUG: Using DNS to try and obtain default hostname Aug 11 11:12:12 netcfg[1872]: DEBUG: Getting default hostname from rDNS lookup of autoconfigured address fe80::5054:ff:fe12:3456 Aug 11 11:12:12 netcfg[1872]: DEBUG: getnameinfo() returned -2 (Name or service not known) Aug 11 11:12:12 netcfg[1872]: DEBUG: Getting default hostname from rDNS lookup of autoconfigured address 10.0.2.15 Aug 11 11:12:12 netcfg[1872]: DEBUG: getnameinfo() returned -2 (Name or service not known) ...then hostname and domain are provided... Aug 11 11:12:22 netcfg[1872]: DEBUG: State is now 6 Aug 11 11:12:23 netcfg[1872]: DEBUG: Network config complete Aug 11 11:12:23 netcfg[1872]: DEBUG: No interface given; clearing /etc/network/interfaces Aug 11 11:12:23 netcfg[1872]: DEBUG: Writing informative header Aug 11 11:12:23 netcfg[1872]: DEBUG: Success! Aug 11 11:12:23 netcfg[1872]: DEBUG: Writing loopback interface Aug 11 11:12:23 netcfg[1872]: DEBUG: Success! Aug 11 11:12:23 netcfg[1872]: DEBUG: Writing DHCP stanza for em0 Aug 11 11:12:23 netcfg[1872]: INFO: No hotpluggable devices are present in the system. Aug 11 11:12:23 netcfg[1872]: DEBUG: Success! Aug 11 11:12:23 dhclient: Received signal 15, initiating shutdown. Aug 11 11:12:23 dhclient: DHCPRELEASE on em0 to 10.0.2.2 port 67 Incriminating code is at: http://anonscm.debian.org/cgit/d-i/netcfg.git/tree/dhcp.c#n617 Is it really intended that kill_dhcp_client will deconfigure the interface? Why was this not a problem for wheezy d-i? Is it supposed to be brought back up again in a later step? Regards, -- Steven Chamberlain ste...@pyro.eu.org signature.asc Description: Digital signature
Re: Bug#757711: netcfg: promptly kills dhclient, deconfigures interface
Control: found -1 netcfg/1.117 Cyril Brulebois k...@debian.org (2014-08-11): Steven Chamberlain ste...@pyro.eu.org (2014-08-11): retitle 757711 netcfg: promptly kills dhclient, deconfigures interface reassign 757711 src:netcfg found 757711 netcfg/1.118 severity 757711 grave Eeww, I missed that. user debian-bsd@lists.debian.org usertags 757711 + kfreebsd thanks Steven, if you're going to raise severity to something RC, *please* check whether the version in testing is also affected. In this particular case you've effectively prevented netcfg's migration to testing while it was fixing a bug, not introducing one! The found command above should tell the BTS and britney that the bug was already in testing, so that it's considered for the next britney run. (I'm not sure I'm convinced by the severity anyway.) Jan, sending the whole /var/log/syslog could help, instead of just a screenshot of the last few lines. Mraw, KiBi. signature.asc Description: Digital signature
Re: Bug#757711: netcfg: promptly kills dhclient, deconfigures interface
AFAICT so far: * netcfg kills the running DHCP client after a hostname is set - this could be a bug * on kFreeBSD, isc-dhcp-client-udeb is used; killing it releases the DHCP lease and deconfigures the interface, which could be intentional * on Linux, Busybox udhcpc is used, killing it seems to have no effect Regards, -- Steven Chamberlain ste...@pyro.eu.org signature.asc Description: OpenPGP digital signature
Re: Bug#757711: netcfg: promptly kills dhclient, deconfigures interface
Steven Chamberlain ste...@pyro.eu.org (2014-08-12): found 757711 1.108+deb7u1 thanks On 11/08/14 23:57, Cyril Brulebois wrote: found 757711 netcfg/1.118 severity 757711 grave Eeww, I missed that. Sorry, assumed you'd seen this... though I did feel the severity was justified since DHCP is the most common use-case. I also don't know yet it if this is really kfreebsd-specific? Well, given the amount of testing I've been doing lately (mostly if not only with DHCP), I would have caught it on Linux if it hadn't been kfreebsd-specific… Steven, if you're going to raise severity to something RC, *please* check whether the version in testing is also affected. I really tried to do this before the ~1800UTC deadline, which I think is when Britney takes a snapshot of BTS state. I'm afraid it didn't go smoothly and I ran out of time. And so that left only two options: * cheat: downgrade the bug against my own judgement, or tentatively mark it as affecting testing - with a slight chance of introducing a new bug otherwise, Given the following changelog entry, I think it would have been a safe thing to do, even without looking at the actual changes: https://packages.qa.debian.org/n/netcfg/news/20140809T213533Z.html I wouldn't call it cheating, BTW. :) If you plan to perform some testing, all good, found/notfound versions can be fixed (no pun intended) afterwards. In the worst case, udeb freeze is in place anyway, so I can only yell at myself (as a maintainer in this particular case and/or as d-i release manager in the general case) if I unblocked packages introducing regressions: https://lists.debian.org/debian-devel-announce/2014/08/msg3.html * or leave it this way, even if it delays migration for a day, until someone else or I can figure this out for sure. Anyway, that's not the end of the world, I was only trying to give general directions to avoid such things in the future. See my summary to debian-{boot,cd,release}@ (towards the end of the mail): https://lists.debian.org/debian-boot/2014/08/msg00251.html Anyhow I've tested as far back as the wheezy version of netcfg and seemingly, that shows the same issue, suggesting it was a change somewhere else (like isc-dhcp-client-udeb) that caused this bug to appear. It still may be netcfg where this needs to be fixed, I don't know yet. That I don't know. Maybe try and compare with netcfg + dhcp client du jour on Linux, and see whether netcfg's behaviour is different in both cases, or the dhcp client's one, or the kernel's one. At least that's what I'd try out of the blue if I were to debug this. (I'm not volunteering. :p) Mraw, KiBi. signature.asc Description: Digital signature
Re: Bug#757711: netcfg: promptly kills dhclient, deconfigures interface
I hope someone can explain the intent of the code below from netcfg: why are DHCP clients killed here? Is it to avoid them clobbering a manual hostname/domain setting? And is that really an issue in practice, with whatever DHCP client[s] we use now? The other thing is, the whole idea of DHCP is that the client should stay running, to renew/retain its lease and/or change IP if asked to. And perhaps release the DHCP lease at the end of the install. If on Linux, udhcpc is killed off, it probably isn't doing any of those things currently? I'll be testing this change tomorrow to see if it helps on kfreebsd at least: --- a/dhcp.c +++ b/dhcp.c @@ -608,19 +608,18 @@ int netcfg_activate_dhcp (struct debconfclient *client, struct netcfg_interface case DOMAIN: if (!have_domain netcfg_get_domain (client, domain)) state = HOSTNAME; else { di_debug(Network config complete); netcfg_write_common(, hostname, domain); netcfg_write_loopback(); netcfg_write_interface(interface); netcfg_write_resolv(domain, interface); -kill_dhcp_client(); stop_rdnssd(); return 0; } break; case HOSTNAME_SANS_NETWORK: if (netcfg_get_hostname (client, netcfg/get_hostname, hostname, 0)) state = ASK_OPTIONS; Regards, -- Steven Chamberlain ste...@pyro.eu.org -- To UNSUBSCRIBE, email to debian-bsd-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org Archive: https://lists.debian.org/20140812011201.ga11...@squeeze.pyro.eu.org