Disable temporary address generation when renderer is NetworkManager (LP: #1948027) – change in behavior #244
Conversation
…and ipv6-privacy is false
|
Thank you very much for the contribution @peleroux ! FTR: Relevant documentation: https://developer-old.gnome.org/NetworkManager/stable/settings-ipv6.html |
|
Thank you for your answer. |
Codecov Report
@@ Coverage Diff @@
## main #244 +/- ##
=======================================
Coverage 99.10% 99.10%
=======================================
Files 58 58
Lines 10035 10038 +3
=======================================
+ Hits 9945 9948 +3
Misses 90 90
Continue to review full report at Codecov.
|
There was a problem hiding this comment.
Thank you Pierre-Emmanuel for this contribution and bug fix! (Sorry for me taking so long to get around reviewing this)
The fix is perfectly valid and brings the NM renderer implementation in line with the networkd backend and what is suggested by our netplan reference documentation:
Enable IPv6 Privacy Extensions (RFC 4941) for the specified interface, and
prefer temporary addresses. Defaults to false - no privacy extensions. There
is currently no way to have a private address but prefer the public address.
networkd's IPv6PrivacyExtensions= setting defaults to false and does not generate privacy addresses: https://www.freedesktop.org/software/systemd/man/systemd.network.html#IPv6PrivacyExtensions=
NM OTOH has been using the fallback mechanism to check the NM's global ipv6.ip6-privacy setting or fall back to the kernel's setting in /proc/sys/net/ipv6/conf/default/use_tempaddr (that would be networkd's "kernel" config value). => Therefore it would fall back to generate privacy IPs anyways, regardless of the ipv6-privacy: false setting.
So this is a change in default behaviour for the NM renderer backend, but at the same time fixes a bug to bring it in line with what is suggested by the documentation (and actually be able to disable privacy IP generation).
I'll merge this, but we should make sure to mention the change in behaviour in the release notes.
PS: I had a tiny remark about your adopted tests in test_keyfile.py, we should mark the ip6-privacy=0 setting as a default value instead of passing it through the keyfile parser. I fixed that up myself and pushed to your branch.
slyon
changed the title
Commit 64e381e (#244) changed netplan's behaviour to always write out the ipv6-privacy value (default to 0=off), in order to comply with what is written in our documentation and how it is implemented in the networkd backend. This is a problem for our NetworkManager integration, as NM's default value is -1=unknown. Netplan only supports values 0=off and 2=on while NM supports -1, 0, 1 and 2. We need to handle unsupported cases (-1 and 1) via passthrough.
Commit 64e381e (#244) changed netplan's behaviour to always write out the ipv6-privacy value (default to 0=off), in order to comply with what is written in our documentation and how it is implemented in the networkd backend. This is a problem for our NetworkManager integration, as NM's default value is -1=unknown. Netplan only supports values 0=off and 2=on while NM supports -1, 0, 1 and 2. We need to handle unsupported cases (-1 and 1) via passthrough.
…and ipv6-privacy is false
Description
Fix for https://bugs.launchpad.net/bugs/1948027
Checklist
make checksuccessfully.make check-coverage).