[PATCH 02/20] 32-bit ABI: introduce ARCH_32BIT_OFF_T config option

[Catalin Marinas]

On Tue, Jun 13, 2017 at 02:04:11PM +0300, Yury Norov wrote:
> On Thu, Jun 08, 2017 at 04:09:50PM +0100, Catalin Marinas wrote:
> > On Sun, Jun 04, 2017 at 02:59:51PM +0300, Yury Norov wrote:
> > > All new 32-bit architectures should have 64-bit off_t type, but existing
> > > architectures has 32-bit ones.
> > > 
> > > To handle it, new config option is added to arch/Kconfig that defaults
> > > ARCH_32BIT_OFF_T to be disabled for non-64 bit architectures. All existing
> > > 32-bit architectures enable it explicitly here.
> > > 
> > > New option affects force_o_largefile() behaviour. Namely, if off_t is
> > > 64-bits long, we have no reason to reject user to open big files.
> > > 
> > > Note that even if architectures has only 64-bit off_t in the kernel
> > > (arc, c6x, h8300, hexagon, metag, nios2, openrisc, tile32 and unicore32),
> > > a libc may use 32-bit off_t, and therefore want to limit the file size
> > > to 4GB unless specified differently in the open flags.
> > > 
> > > Signed-off-by: Yury Norov <ynorov at caviumnetworks.com>
> > > Acked-by: Arnd Bergmann <arnd at arndb.de>
> > [...]
> > > diff --git a/include/linux/fcntl.h b/include/linux/fcntl.h
> > > index 1b48d9c9a561..297993c92490 100644
> > > --- a/include/linux/fcntl.h
> > > +++ b/include/linux/fcntl.h
> > > @@ -11,7 +11,7 @@
> > >  	 O_NOATIME | O_CLOEXEC | O_PATH | __O_TMPFILE)
> > >  
> > >  #ifndef force_o_largefile
> > > -#define force_o_largefile() (BITS_PER_LONG != 32)
> > > +#define force_o_largefile() (!IS_ENABLED(CONFIG_ARCH_32BIT_OFF_T))
> > >  #endif
> > 
> > I may have confused myself with which off_t is 64-bit here for new
> > 32-bit architectures. Are we referring to the glibc definition, the
> > kernel one or simply that force_o_largefile() is true by default.
> > Because the type off_t for 32-bit kernel builds is still, well, 32-bit.
> > 
> > Otherwise it seems that the first paragraph in the description above
> > should read "all new 32-bit ABIs on a 64-bit kernel..." but then
> > AArch64/ILP32 is no longer the same as a new, pure 32-bit architecture.
>  
> This is all about userspace off_t types, like Arnd told in the comment
> to patch 13. I'll underline it in the comment to the patch. If it's
> not enough, I can also rename the config option to
> CONFIG_ARCH_32BIT_USER_OFF_T or similar. For me it's too much, but if
> you find it reasonable, I'll do it. Just let me know.

Thanks for clarification. I had the impression that it should match the
kernel's off_t (which is exported in the kernel headers as 32-bit) but
compiling with -mabi=ilp32 indeed shows sizeof(off_t) == 8. So that's
just a user decision to use loff_t instead and such port shouldn't use
any of the syscalls that pass the kernel's off_t.

I would rather see the comment in the arch/Kconfig help entry in this
patch for future reference.

-- 
Catalin