Re: [Qemu-devel] [Qemu-block] [PATCH] block.c: fix real cdrom detection

[Stefan Hajnoczi]

On Thu, Jun 25, 2015 at 07:19:14PM +0200, Laurent Vivier wrote:
> 
> 
> On 25/06/2015 18:16, Paolo Bonzini wrote:
> > 
> > 
> > On 25/06/2015 18:12, Laurent Vivier wrote:
> >>
> >>
> >> On 25/06/2015 17:48, Paolo Bonzini wrote:
> >>>
> >>> On 25/06/2015 17:32, Programmingkid wrote:
> >>>>> I think we are going to have to agree to disagree. I have never
> >>>>> used the /dev/sr(0 | 1) devices and don't see how they would be
> >>>>> effected by this patch. Are you trying to say the /dev/sr(0 | 1)
> >>>>> devices *should* be handled by this patch?
> >>>>
> >>>> Thinking about your question some more, I see what you mean. On Linux
> >>>> /dev/sr0 refers to the cdrom drive. Also on Linux, the /dev/cdrom
> >>>> link refers to the /dev/sr0 device file. So if you just use
> >>>> /dev/cdrom, you are good.
> >>>
> >>> Well, that's not how things work.
> >>>
> >>> If you do things like that, you end up with a bunch of hacks, not with a
> >>> decent piece of software.
> >>>
> >>> There is support for CD-ROM passthrough on Linux and FreeBSD in
> >>> block/raw-posix.c.  Perhaps the FreeBSD support can be extended to OS X
> >>> as well.
> >>>
> >>
> >> In fact, programmingkid, you should fix it in hdev_open() where there is
> >> already a #if __APPLE__ .
> >>
> >> Paolo, I agree with you but :
> >>
> >> hdev_open()
> >>
> >> #if defined(__linux__)
> >>     {
> >>         char resolved_path[ MAXPATHLEN ], *temp;
> >>
> >>         temp = realpath(filename, resolved_path);
> >>         if (temp && strstart(temp, "/dev/sg", NULL)) {
> >>             bs->sg = 1;
> >>         }
> >> #endif
> >>
> >> I'm wondering who had this strange idea... :)
> > 
> > I was very scared to type "git blame" here. :)  But the question is also
> 
> http://geek-and-poke.com/2013/11/24/simply-explained
> 
> BTW, it is a legacy from 2006:
> 
> 19cb373 better support of host drives
> 
> coming from MacOS X (again!):
> 
> 3b0d4f6 OS X: support for the built in CD-ROM drive (Mike Kronenberg)
> 
> > where to put the checks.  Putting it at a random place in block.c is not
> > a good idea.
> > 
> > But yes, this is also bad.  It should use stat and check the major/minor
> > numbers.
> 
> Yes, we should check if major is SCSI_GENERIC_MAJOR (21) (on linux).

That would be too specific since there are other drivers that support SG
ioctls, like block/bsg.c.

> We can also try to send an SG command like in cdrom_probe_device().
> Something like in scsi_generic_realize():
> 
>     rc = blk_ioctl(s->conf.blk, SG_GET_VERSION_NUM, &sg_version);
>     if (rc < 0) {
>         error_setg(errp, "cannot get SG_IO version number: %s.  "
>                          "Is this a SCSI device?",
>                          strerror(-rc));
>         return;
>     }

That was recently done in:

commit 3307ed7b3fac5ba99eb3b84904b0b7cdc3592a61
Author: Dimitris Aragiorgis <address@hidden>
Date:   Tue Jun 23 13:45:00 2015 +0300

    raw-posix: Introduce hdev_is_sg()

pgp4B7gQ33ORr.pgp
Description: PGP signature