On Fri, Dec 11, 2020 at 9:29 AM Al Viro <viro@xxxxxxxxxxxxxxxxxx> wrote:
>
> Again, that part is trivial - what to do with do_open()/open_last_lookups()
> is where the nastiness will be. Basically, it makes sure we bail out in
> cases when lookup itself would've blocked, but it does *not* bail out
> when equally heavy (and unlikely) blocking sources hit past the complete_walk().
See my other email - while you are obviously correct from a "must
never sleep" standpoint, and from a general standpoint, from a
practical standpoint I suspect it really doesn't matter.
Exactly because it's not primarily a correctness issue, but a
performance issue - and because people wouldn't use this for things
like "open a named pipe that then blocks on the open side" use.
I do agree that maybe that LOOKUP_NONBLOCK might then want to be also
coupled with extra logic to also just fail if it turns out it's a
special device file or whatever - I just think that ends up being a
separate issue.
For example, in user space, we already have alternate methods for that
(ie O_NONBLOCK for fifo etc), and maybe io_uring wants that too.
Linus
