On Mon, 23 Sept 2024 at 19:26, Linus Torvalds
<torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:
>
> And I had missed the issue with PREEMPT_RT and the fact that right now
> the inode hash lock is outside the inode lock, which is problematic.
.. although I guess that could be solved by just making the spinlock
be external to the hash list, rather than part of the list like
list_bh.
You wouldn't need to do one lock per list head, you could just do some
grouping of the hash.
That said, the vfs inode hash code really is pretty disgusting and has
been accumulating these warts for a long time. So maybe a "filesystems
do their own thing" together with some helper infrastructure (like
using rhashtable) is actually the right thing to do.
Linus
