> > > > i don't know how read requests with B_VFLUSH happens
> > >
> > > Just try a newfs over the vnd device and you'll get into
> > > handle_with_rw with B_READ and B_VFLUSH set.
> >
> > i tried the following with the attached patch, but got no panics.
> >
> >                 mount_tmpfs a /mnt
> >                 cd /mnt
> >                 dd if=/dev/zero of=x bs=1m count=1
> >                 vnconfig vnd0 x
> >                 newfs -F /dev/rvnd0d
> 
> Of course not because that's what the increment of numoutput is for
> (or was supposed to be for).

i meant i didn't get the panic in a patch attached in my previous mail.

> 	printf("%s, %s\n", doread ? "read" : "write",
> 	    obp->b_flags & B_VFLUSH ? "flush" : "noflush");
> 	if (!doread)
> 		vp->v_numoutput++;
> 
> And it does not panic.  I'm sure it did before!  And I'm sure I got
> "read, sync" messages from the debug printf.  (But maybe it was
> because I still had some other problem somewhere in the handling of
> petitions.)

do you mean "read, flush" ?

> So... should I remove all the B_VFLUSH handling from the function?
> Should I change the manual v_numoutput increment to use the
> V_INCR_NUMOUTPUT macro?

yes and yes.  afaik, B_VFLUSH is not something which vnd should care.

YAMAMOTO Takashi