[Libguestfs] [PATCH 03/12] mountable: Implement Mountable support for all apis which take it

[Matthew Booth]

On Fri, 2013-02-08 at 11:56 +0000, Richard W.M. Jones wrote:
> On Thu, Feb 07, 2013 at 03:57:49PM +0000, Matthew Booth wrote:
> > A Mountable is passed from the library to the daemon as a string. The daemon
> > stub parses it into a mountable_t, which it passes to the implementation.
> > 
> > Update all implementations which now take a mountable_t.
> > ---
> >  daemon/blkid.c      | 33 +++++++++++++++++++++++-----
> >  daemon/daemon.h     | 41 ++++++++++++++++++++++++++++++++++
> >  daemon/ext2.c       | 15 ++++++++++---
> >  daemon/guestfsd.c   | 63 ++++++++++++++++++++++++++++++++++++++++++++++++++---
> >  daemon/labels.c     | 16 ++++++++++++--
> >  daemon/mount.c      | 49 +++++++++++++++++++++++++++++++++--------
> >  generator/c.ml      | 12 +++++++++-
> >  generator/daemon.ml | 11 +++++++---
> >  8 files changed, 213 insertions(+), 27 deletions(-)
> > 
> > diff --git a/daemon/blkid.c b/daemon/blkid.c
> > index 64919dd..d54ac22 100644
> > --- a/daemon/blkid.c
> > +++ b/daemon/blkid.c
> > @@ -67,21 +67,42 @@ get_blkid_tag (const char *device, const char *tag)
> >  }
> >  
> >  char *
> > -do_vfs_type (const char *device)
> > +do_vfs_type (const mountable_t *mountable)
> >  {
> [...]
> > +    default:
> > +      reply_with_error ("Invalid mountable type: %i", mountable->type);
> > +      return NULL;
> > +  }
> >  }
> 
> Lots of this code has these error cases.  However it would be much
> simpler if the error case was handled by the generator, and my reading
> of RESOLVE_MOUNTABLE is that it already *is* handled by the generator.
> So you could assume that mountable->type would never be anything else
> and replace the 'default' case with a call to abort () (even better if
> GCC gave us a way to say the default case never happens, but that
> doesn't seem to be possible).

They're certainly belt and braces. I'm happy to take them out.

> There's even a possibility that these error cases are wrong because
> they call reply_with_error twice along the error path (which results
> in protocol desynchronization).

I don't think so. I've was pretty careful about errors on error paths
(see extensive use of fprintf).

> > +    if (strncmp ((string), "btrfsvol:", strlen ("btrfsvol:")) == 0) {   \
> 
> Better to use STRPREFIX here.

ACK.

> 
> > +      RESOLVE_DEVICE((string), cancel_stmt, fail_stmt);                 \
> 
> Do we need parens around the *_stmts?  (I'm not sure ...)

I don't know either. I think I was copying other usage, tbh. I'm happy
to change this or not.

> > @@ -257,7 +262,7 @@ and generate_daemon_actions () =
> >              (function FileIn _ | FileOut _ -> false | _ -> true) args in
> >          let style = ret, args' @ args_of_optargs optargs, [] in
> >          pr "  r = do_%s " name;
> > -        generate_c_call_args style;
> > +        generate_c_call_args ~in_daemon:true style;
> 
> Is this a bug in existing code?  It may need to be separately
> backported if so.

No, I added in_daemon to generate_c_call_args specifically for mountable
support. The reason is that in the library the call is passed a string,
but in the daemon it's passed a parsed mountable, so it needs to
differentiate.

Matt