Sergey Bugaev, le dim. 19 juil. 2026 22:23:16 +0200, a ecrit:
> On Sun, Jul 19, 2026 at 9:24 PM Samuel Thibault <[email protected]> 
> wrote:
> > I don't think we want to store the owner in the socket itself? As
> > mentioned in the comment, the "owner" is related to the file descriptor,
> > not to the socket. So it'd rather be the sock_user.
> 
> I don't fully understand this concept of "owner"ship

It's the process that should receive signals about availability of data
to read etc. for O_ASYNC processing. The signaling part is actually not
implemented at all, but servers apparently at least support storing the
information, so that testsuites are fine with it.

> (and storing a
> numeric PID is clearly broken, it should have been the message port or
> something), but netfs stores this in the node, and the comment in
> io.defs indicates that the related old-style async I/O flag is
> "per-object (not per-open)". So possibly it is the comment that needs
> changing rather.

Mmm, thinking about it, if it was per-open it wouldn't make sense to
tell which process should get the signal. So indeed it is per-object, I
have applied the patch with your fix, thanks!

Samuel

Reply via email to