Fix crash with PPP interface with nil dst address #294
Conversation
|
Your landr site preview has been successfully deployed to https://landr-balena-os-repo-balena-engine-preview-294.netlify.app Deployed with Landr 9.4.14 |
|
@lmbarros could we cherry-pick vishvananda/netlink#665 for the time being? |
|
@alexgg I see I mentioned cherry picking myself, but that's not easily done. netlink is a dependency, not code we own. (We do keep a vendored copy of netlink within our repo -- changing it directly could technically work, but would be really bad practice and asking for trouble in the future.) |
|
We would need to fork netlink, cherry pick into our fork and point the dependency to the fork. It's doable, but I don't know... doesn't appear to be much simpler than pointing to the beta netlink and going through a round of Engine tests. |
|
hey @lmbarros using a beta version of netlink feels like looking for trouble - if it was production ready it wouldn't be beta. I understand forking a dependency is cumbersome, but once netlink releases a stable version we can just drop it. And having the engine crash when using primary cellular connectivity is something we need to address. |
|
@alexgg , not saying this is the right approach, but in their next-release branch Moby is already using the beta netlink for some time. |
@lmbarros but we also wouldn't use Moby's next release in production, would we? |
lmbarros
force-pushed
the
lmb/update-netlink
branch
from
c135dc3
to
c4b4b29
Compare
balenaEngine initialization would fail on devices connected to the
network via PPP (Point-to-Point Protocol) and with a nil destination
address (0.0.0.0):
panic: runtime error: invalid memory address or nil pointer dereference
This commit updates the netlink dependency to a fork where we
cherry-picked the correction. This correction wasn't available in any
stable release of netlink, so we opted to use this fork instead of
relying on a beta netlink version. We can obsolete our fork once Moby
starts using a netlink version that includes the fix.
Fixes #292
Signed-off-by: Leandro Motta Barros <leandro@balena.io>
Change-type: patch
lmbarros
force-pushed
the
lmb/update-netlink
branch
from
c4b4b29
to
67cc46a
Compare
lmbarros
changed the title
During the migration to Flowzone (the balena CI system) we brought several changes from Moby's HEAD Dockerfile. Among those, we replaced the `frozen-images` base image to a debian:bullseye-slim. This was a mistake: the slim image doesn't contain the `setcap` binary we need by the `TestBuildUserNamespaceValidateCapabilitiesAreV2` integration test. So, this PR simply brings back the full-blown `debian:bullseye`. Signed-off-by: Leandro Motta Barros <leandro@balena.io> Change-type: patch
|
Hey @alexgg ! This is the netlib update (with the fix cherry-picked into a temporary netlib fork, as we agreed). I slipped in a second commit fixing an integration test failure introduced by the migration to Flowzone. |
balenaEngine initialization would fail on devices connected to the
network via PPP (Point-to-Point Protocol) and with a nil destination
address (0.0.0.0):
This commit updates the netlink dependency to a fork where we
cherry-picked the correction. This correction wasn't available in any
stable release of netlink, so we opted to use this fork instead of
relying on a beta netlink version. We can obsolete our fork once Moby
starts using a netlink version that includes the fix.