Re: strace of io_uring events?

[Andres Freund <andres@anarazel.de>]

Hi,

On 2020-07-21 10:23:22 -0700, Andy Lutomirski wrote:
> Andres, how final is your Postgres branch?

Not final at all. Some of the constraints like needing to submit/receive
completions to/from multiple processes are pretty immovable though.


> I'm wondering if we could get away with requiring a special flag when
> creating an io_uring to indicate that you intend to submit IO from
> outside the creating mm.

Perhaps. It'd need to be clear when we need to do so, as we certainly
won't want to move the minimal kernel version further up.


But I think postgres is far from the only use case for wanting the
submitting mm to be the relevant one, not the creating one. It seems far
more dangerous to use the creating mm than the submitting mm. Makes
things like passing a uring fd with a few pre-opened files to another
process impossible.

Greetings,

Andres Freund