Re: [Qemu-devel] build regression on osx in block layer

[Kevin Wolf <kwolf@redhat.com>]

Am 29.09.2014 um 15:58 hat Peter Maydell geschrieben:
> On 29 September 2014 11:10, Igor Mammedov <imammedo@redhat.com> wrote:
> > Build fails with:
> >
> > $ make
> >   LINK  qemu-nbd
> > Undefined symbols for architecture x86_64:
> >   "_posix_fallocate", referenced from:
> >       _raw_create in raw-posix.o
> > ld: symbol(s) not found for architecture x86_64
> >
> > introduced by
> > 06247428be raw-posix: Add falloc and full preallocation
> 
> I think the MacOSX interface to this involves
> fcntl(F_PREALLOCATE). This is what Mozilla uses:
> 
>   fstore_t store = {F_ALLOCATECONTIG, F_PEOFPOSMODE, 0, aLength};
>   // Try to get a continous chunk of disk space
>   int ret = fcntl(fd, F_PREALLOCATE, &store);
>     if(-1 == ret){
>     // OK, perhaps we are too fragmented, allocate non-continuous
>     store.fst_flags = F_ALLOCATEALL;
>     ret = fcntl(fd, F_PREALLOCATE, &store);
>     if (-1 == ret)
>       return false;
>   }
>   return 0 == ftruncate(fd, aLength);
> 
> This has probably broken compilation on the BSDs
> as well, since I don't suppose they all have
> posix_fallocate(). In extremis we can implement
> it with truncate-and-write, cf:
> http://stackoverflow.com/questions/11497567/fallocate-command-equivalent-in-os-x
> 
> We probably need a util/qemu-fallocate.c...

Sent a patch to make the build of that code conditional. Having some
util/qemu-fallocate.c probably wouldn't hurt, but it needs to be done by
someone who has a system that uses a different path (i.e. not me).

Kevin