Re: tmpfs replacement

[Marco Gerards]

Roland McGrath <roland@frob.com> writes:

> There is no reason to replace tmpfs per se, just hack it.
> There is no problem with its directory management code, right?

I'm not sure, but it seems to work. There are two bugs in the default
pager (or GNUMach, I'm not sure about this) that make tmpfs unusable.
 
> A tmpfs that uses vm_allocate for file pages is perfectly doable, it just
> has a limit on total file space imposed by the address space of the tmpfs
> server process (so presumably about the same as the ext2fs disk mapping
> size limit, which is probably big enough for tmpfs uses), and it doesn't
> make it easy to support io_map.  If you wanted to support io_map by also
> using libpager, that might be a little more weird and suboptimal.

I prefer to fix the current tmpfs instead.io_map is more ugly when
using libnetfs I think because the memory has to be copied, right?

Thanks,
Marco