On Sun, 6 Aug 2023 at 06:26, Christian Brauner brauner@kernel.org wrote:
We got sent a fix for a wrong check for O_TMPFILE during RESOLVE_CACHED lookup which I've put on vfs.fixes yesterday:
git@gitolite.kernel.org:pub/scm/linux/kernel/git/vfs/vfs tags/v6.5-rc5.vfs.resolve_cached.fix
But in case you planned on applying this directly instead of waiting for next cycle I've added your two appended patches on top of it and my earlier patch for massaging the file_needs_f_pos_lock() check that triggered this whole thing:
git@gitolite.kernel.org:pub/scm/linux/kernel/git/vfs/vfs tags/v6.5-rc5.vfs.fixes
I had actually planned on just waiting for the 6.6 merge window, but then you made this _so_ easy for me that I ended up taking these things right now.
The timing may not be entirely right, but I'm very comfortable with the "get rid of '->iterate' op" change since it (a) would clearly fail the build on a missed conversion and (b) doesn't touch any core filesystems anyway.
And now that file_needs_f_pos_lock() does look better, and as you say in teh commit, makes it clearer why that locking rule exists.
Linus
Linus