Fix a multiple-pointers-to-single-loop-variable bug in EgressNetworkPolicy #12045
Conversation
|
I assume you're going to open a PR against release-1.4 as well? |
|
[test] |
|
Evaluated for origin test up to bbb79e3 |
|
continuous-integration/openshift-jenkins/test SUCCESS (https://ci.openshift.redhat.com/jenkins/job/test_pr_origin/11732/) (Base Commit: 9b1d477) |
1 similar comment
|
continuous-integration/openshift-jenkins/test SUCCESS (https://ci.openshift.redhat.com/jenkins/job/test_pr_origin/11732/) (Base Commit: 9b1d477) |
|
[merge] |
|
@danwinship as @smarterclayton suggested (and I am sure you are planning to do), this should go on the 1.4 branch too. |
|
[merge] flaked on #11662 |
|
Evaluated for origin merge up to bbb79e3 |
|
continuous-integration/openshift-jenkins/merge SUCCESS (https://ci.openshift.redhat.com/jenkins/job/test_pr_origin/11787/) (Base Commit: daef9e7) (Image: devenv-rhel7_5436) |
The EgressNetworkPolicy code was doing the same thing as in https://github.com/openshift/ose/pull/420 where we were trying to store pointers to various objects in an array but ended up storing multiple pointers to the same object instead.
(I guess I was imagining [and maybe others were too since this bug passed code review twice] that since
policywas a local variable, but&policysurvives after the function returns, then that must mean that&in this context copies the local variable to a new heap object. But in fact, all that happens is that the compiler sees that&policyneeds to survive after the function returns, and so allocatespolicyon the heap rather than the stack to begin with, and so&policyis the same address each time through the loop.)Fixes #12038.
@openshift/networking PTAL