Re: [PATCH v5 15/15] libxl: Re-implement domain_suspend_device_model using libxl__ev_qmp

[Anthony PERARD <anthony.perard@citrix.com>]

On Tue, Oct 16, 2018 at 04:28:55PM +0100, Ian Jackson wrote:
> Anthony PERARD writes ("[PATCH v5 15/15] libxl: Re-implement domain_suspend_device_model using libxl__ev_qmp"):
> > +    if (rc)
> > +        goto error;
> > +
> > +    libxl__carefd_begin();
> > +    ev->cfd = libxl__carefd_opened(CTX,
> > +                                 open(filename, O_WRONLY | O_CREAT, 0600));
> 
> Does this statefile fd really need to be a carefd ?  Is it a pipe or a
> file ?  If it is a file, is it of nontrivial size ?

Yes, it needs to be caredfd, because that's what the libxl__ev_qmp API
wants. I don't know yet if it is a good idee to have the _ev_qmp API
only takes fd. Do you think it's fine to have libxl__ev_qmp API takes a
simple fd and let callers handle the fd the way they want?

(In previous version of the patch series, libxl__ev_qmp used to close
the carefd. That's not the case anymore, and that carefd is only read,
so I don't think it matter anymore which kind it is between int and
carefd.)

> If it's a file of a few kb, which I think is the case, then the worst
> result of (with very low probability) leaking this fd into another
> process is simply that the file might be kept alive for a while after
> it was deleted.


-- 
Anthony PERARD

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xenproject.org
https://lists.xenproject.org/mailman/listinfo/xen-devel