Created attachment 1182164 [details] ifcfg.log Description of problem: Default route, name server not set up during network installation. Version-Release number of selected component (if applicable): NetworkManager-1.2.2-2.fc25 anaconda-25.20-1 How reproducible: Everytime. Steps to Reproduce: 1. Download netinstall iso or pxe and attempt install 2. Fails - Error setting up software source Additional info: Add default route, name server for local network, installation completed.
Created attachment 1182165 [details] anaconda.log
Created attachment 1182166 [details] dnf.librepo.log
Created attachment 1182167 [details] packaging.log
Added as Alpha blocker.
I reproduced the issue with Fedora-Everything-netinst-x86_64-Rawhide-20160720.n.0.iso.
Discussed at 2016-07-20 blocker review meeting: [1]. This bug was accepted as Alpha blocker: Malfunctioning networking violates the alpha criterion: "When using a release-blocking dedicated installer image, the installer must be able to use either HTTP or FTP repositories (or both) as package sources. Release-blocking network install images must default to a valid publicly-accessible package source." [1] https://meetbot.fedoraproject.org/fedora-blocker-review/2016-07-20/f25-blocker-review.2016-07-20-16.00.html
This bug appears to have been reported against 'rawhide' during the Fedora 25 development cycle. Changing version to '25'.
This bug appears in Fedora-Server-netinst-x86_64-Rawhide-20160803.n.0.iso.
I'm running into this problem also and it seems to vary among composes because for me Fedora-Workstation-netinst-x86_64-25-20160730.n.0.iso fails to have any network access; whereas Server and Everything were working. ip a says the network is up, ncmcli says the network is up, but ping won't work to any IP address, and nothing resolves; somehow networking works enough for it to get a DHCP assigned IP, but then doesn't work well enough to resolve anything, or even ping an IP address directly.
Created attachment 1187302 [details] journal from workstation20160803 Attached journal based on Fedora-Workstation-netinst-x86_64-25-20160803.n.0.iso in qemu-kvm. All the 20160803 netinst composes have network failure, Installation Source = Error setting up base repository. Happens with enforcing=0. ping gets "connect: Network is unreachable" But nmcli c says it has a connection, and ip a reports an ip address assigned.
Fedora-Workstation-netinst-x86_64-25-20160803.n.0.iso does not exhibit the problem. The only difference I'm spotting is, netinst "state is now CONNECTED_LOCAL", whereas live I see both "state is now CONNECTED_SITE" followed by "state is now CONNECTED_GLOBAL".
*sigh*, sorry, bad paste /s/Fedora-Workstation-netinst-x86_64-25-20160803.n.0.iso/Fedora-KDE-Live-x86_64-25-20160803.n.0.iso. Live works, netinst's do not.
Created attachment 1187413 [details] syslog with dhclient log messages I've reproduced with some rawhide compose. DNS and gateway configuration from dhcp does not happen, same as reported above. I'm attaching /tmp/syslog containing dhclient log messages, suspecting: 08:48:06,014 ERR dhclient: /var/lib/NetworkManager/dhclient-ens3.conf line 7: expecting semicolon. 08:48:06,016 ERR dhclient: root-path, interface-mtu classless-routes; 08:48:06,016 ERR dhclient: ^ I'll attach also NetworkManager from the reproducer.
Created attachment 1187414 [details] NetworkManager log for comment #13
Reassigning to NM for debugging (or passing forward).
Fixed in dracut master: https://github.com/dracutdevs/dracut/commit/7e51abc81f53c08e464decd4103e8c4ec25fef87
*** Bug 1364281 has been marked as a duplicate of this bug. ***
this is breaking some openQA tests (see dupe).
Just to fill in a blank here, the change that triggered this is (I think): https://github.com/rhinstaller/anaconda/commit/69acea283e7c58ab22581a4d78947747c76cf7af the commit message claims "if inst.dhcpclass= was given on the command line, make sure the /etc/dhclient.conf file is copied [from the dracut environment] over to the anaconda environment", but it actually seems to copy it any time it exists. So prior to that commit anaconda wouldn't copy dracut's busted dhclient.conf into its environment, now it always or often does. (I'm now a bit curious how some of the openQA network tests *do* still work, but hey...after a bit of tinkering around it seems the syntax error isn't fatal in *all* cases, I think it depends what other directives NM adds to the composite dhclient config file it creates.) I have sent a dracut build (044-76) for both Rawhide and F25 with the relevant upstream patch backported, so we can get this urgent bug fixed ASAP.
Confirmed this is fixed with today's F25 nightly, openQA is no longer running into it either.