When netcfg is finished configuring the interface, it will call kill_dhcp_client() (from netcfg_activate_dhcp()), which will send SIGTERM and then SIGKILL to dhcp6c. When dhcp6c gets a SIGTERM, it will send a DHCP6 Release message and remove the IP from the interface, which means that the installer will have no (public) IP anymore, and installation from network will fail. This happens both in an IPv6-only and in a dual-stack configuration (though in a dual-stack configuration there still will be an IPv4 address, which means the installation will probably succeed). A possible workaround is to switch to another console with Alt+F3 after the DHCP6 configuration was done but before netcfg finishes and kill the dhcp6c process with -9 (to prevent it from releasing the IP).
Hi, Just submitted a fix on Salsa: https://salsa.debian.org/installer-team/netcfg/-/merge_requests/18
Dear Maintainer,
I would like to add my own anecdotal experience with this bug and the trixie
net installer iso,
and also add the keywords that should cause this report to show up in web
search results (for those affected):
The mirror does not support the specified release (trixie).
* What led up to the situation?
I agree with the original report that this situation is most prominent
when the system
can have connectivity via IPv6 only (and DHCP6).
* What exactly did you do (or not do) that was effective (or ineffective)?
Effective: Applying the IPv6 manually to the system after the network
detection step.
Effective: Running the installer with IPv4 connectivity.
Ineffective: Trying different debian mirrors.
Ineffective: Tampering with the APT sources (as web search results
often recommended).
* What was the outcome of this action?
The net installer reaches the step where it would be downloading
packages to set up the
system, but then it fails with the message:
The mirror does not support the specified release (trixie).
* What outcome did you expect instead?
The installer would show the progress of downloading and installing
system packages.
Note: This bug does not exist in Debian 12 (Bookworm)'s installer when testing
with the same exact hardware configuration.
Hi, As noted in <https://lists.debian.org/msgid-search/angHe8EQxyEqN3Yi@mail.bitfolk.com> I just wanted to confirm that this problem still exists in the trixie installer (also tested with the daily installer image). On an IPv6-only network with no SLAAC, gateway provided by router solicitation, d-i attempts DHCPv6 which succeeds and networking is initially functional. After the host name and domain name questions are answered, netcfg does still issue a kill -TERM to the dhcp6c process. This causes dhcp6c to send a DHCP RELEASE packet and de-configure the interface. d-i does not request a lease again later, so it's not possible to continue the install. The environment is a bit unusual in that there's no SLAAC and no router advertisements; the prefix is /127 (just this host and its gateway). This is a valid configuration though and it does work fine from an installed Debian using the "dhcp" method in /etc/network/interfaces. The patch provided earlier in this bug, at <https://salsa.debian.org/installer-team/netcfg/-/merge_requests/18> does avoid the issue by doing kill -KILL. A workaround is to switch to the installer shell after dhcp6c has been killed and launch it again: # dhcp6c -c /var/lib/netcfg/dhcp6c.conf eno1 Or, one can hack in the patch's -KILL behaviour in kill-all-dhcp before network configuration happens: # sed -i 's/-TERM/-KILL/' /usr/bin/kill-all-dhcp Possibly placing a patched /usr/bin/kill-all-dhcp file there could be done in a preseeded early_cmd if that's easier than deploying the patch. Without this, DHCPv6 is not feasible; install on an IPv6-only network can only be carried out when there is SLAAC or static networking configuration. Thanks, Andy