On Tue, Mar 29, 2022 at 6:48 PM Linus Torvalds
<torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:
>
> I'd be even happier if we could actually have some common helper
> starting at that
>
> trap = lock_rename(...);
>
> because that whole locking and 'trap' logic is kind of a big deal, and
> the locking is the only thing that makes __lookup_hash() work.
Hmm. The cachefiles and ecryptfs code seems to have a fair amount of
this pattern.
None of it is exactly the same, and maybe it's hard to really have a
common helper, but it does look like that sequence from
trap = lock_rename(...)
leading up to the eventual
ret = vfs_rename(&renamedata);
would be really nice to capture some way.
But maybe there just isn't any actual code commonality outside of the
general pattern.
I was curious, and that "trap = lock_rename()" pattern including those
"exit5" labels is _ancient_, btw.
It goes back to a commit by Al back in 2002, best seen in Thomas' BK
history conversion:
https://git.kernel.org/pub/scm/linux/kernel/git/tglx/history.git/commit/?id=1b3d7c93c6d1927540120bafb08fa60c0d96cbee
so it's not like this is a pattern that has seen a lot of changes.
Maybe it's ok to just have various copies of it, but let's see if Al
has any ideas.
Linus
