Page MenuHomeFreeBSD

Don't fail disk state change if g_mirror_clear_metadata() fails.
ClosedPublic

Authored by markj on Sep 1 2016, 9:24 PM.
Tags
None
Referenced Files
Unknown Object (File)
Tue, Sep 30, 4:17 PM
Unknown Object (File)
Fri, Sep 26, 8:26 PM
Unknown Object (File)
Sep 3 2025, 7:58 PM
Unknown Object (File)
Aug 30 2025, 1:39 AM
Unknown Object (File)
Aug 25 2025, 1:31 PM
Unknown Object (File)
Aug 19 2025, 1:39 PM
Unknown Object (File)
Aug 6 2025, 6:55 AM
Unknown Object (File)
Jul 16 2025, 12:55 PM
Subscribers

Details

Summary

We may fail to clear the metadata if we attempt to remove the disk from
the mirror as it is being destroyed, e.g. because of a concurrent hot
swap. This is not a fatal error since a failure from
g_mirror_clear_metadata() will result in the mirror's generation ID
being bumped, so the disk in question will not be added back into the
mirror upon a retaste.

Without this change, INVARIANTS kernels will panic in this case
because the mirror worker won't handle errors.

Diff Detail

Repository
rS FreeBSD src repository - subversion
Lint
Lint Not Applicable
Unit
Tests Not Applicable

Event Timeline

markj retitled this revision from to Don't fail disk state change if g_mirror_clear_metadata() fails..
markj edited the test plan for this revision. (Show Details)
markj updated this object.

Do we have a good way to unit test this so I can add a relevant gmirror testcase?

This revision was automatically updated to reflect the committed changes.