On Fri, Feb 09, 2024 at 11:08:45AM -0800, Jordan Rife wrote:
On Fri, Feb 9, 2024 at 10:52 AM Sasha Levin sashal@kernel.org wrote:
On Fri, Feb 09, 2024 at 10:26:57AM -0600, Jordan Rife wrote:
Backport e11dea8 ("dlm: use kernel_connect() and kernel_bind()") to Linux stable 6.1 caused a regression. The original patch expected dlm_local_addrs[0] to be of type sockaddr_storage, because c51c9cd ("fs: dlm: don't put dlm_local_addrs on heap") changed its type from sockaddr_storage* to sockaddr_storage in Linux 6.5+ while in older Linux versions this is still the original sockaddr_storage*.
Or we can just take c51c9cd8addc ("fs: dlm: don't put dlm_local_addrs on heap") into the relevant trees?
-- Thanks, Sasha
Hi Sasha,
Just my 2c, but backporting c51c9cd8addc ("fs: dlm: don't put dlm_local_addrs on
heap") feels a bit riskier than just correcting the call to kernel_bind(), as it's a much
bigger change. Maybe someone more familiar with the dlm codebase can chime in and say whether or not they are confident with backporting this change.
It's a bigger change, but in our experience it's the small fixups that end up carrying the bigger risk.
Backporting the original change also has the advantage of preventing similar issues from happening in the future.