Re: [PATCH] y2038: Remove newstat family from default syscall set

[Palmer Dabbelt]

On Sat, 01 Sep 2018 10:43:53 PDT (-0700), linux@xxxxxxxxxxxx wrote:
> Hi Arnd,
>
> On Fri, Apr 13, 2018 at 11:50:12AM +0200, Arnd Bergmann wrote:
> > We have four generations of stat() syscalls:
> > - the oldstat syscalls that are only used on the older architectures
> > - the newstat family that is used on all 64-bit architectures but
> >   lacked support for large files on 32-bit architectures.
> > - the stat64 family that is used mostly on 32-bit architectures to
> >   replace newstat
> > - statx() to replace all of the above, adding 64-bit timestamps among
> >   other things.
> >
> > We already compile stat64 only on those architectures that need it,
> > but newstat is always built, including on those that don't reference
> > it. This adds a new __ARCH_WANT_NEW_STAT symbol along the lines of
> > __ARCH_WANT_OLD_STAT and __ARCH_WANT_STAT64 to control compilation of
> > newstat. All architectures that need it use an explict define, the
> > others now get a little bit smaller, and future architecture (including
> > 64-bit targets) won't ever see it.
>
> This patch causes my riscv boot tests to crash in -next

Ah, thanks for running these!

> sbin/init: error while loading shared libraries: libc.so.6: cannot stat shared object: Error 38
> Kernel panic - not syncing: Attempted to kill init! exitcode=0x00007f00
>
> The following change fixes the problem for me, but of course I have no idea
> if it is correct. Copying RISC-V maintainers for input.
>
> Guenter
>
> ---
> diff --git a/arch/riscv/include/asm/unistd.h b/arch/riscv/include/asm/unistd.h
> index 0caea01d5cca..eff7aa9aa163 100644
> --- a/arch/riscv/include/asm/unistd.h
> +++ b/arch/riscv/include/asm/unistd.h
> @@ -16,6 +16,7 @@
>   * be included multiple times.  See uapi/asm/syscalls.h for more info.
>   */
>
> +#define __ARCH_WANT_NEW_STAT
>  #define __ARCH_WANT_SYS_CLONE
>  #include <uapi/asm/unistd.h>
>  #include <uapi/asm/syscalls.h>

I'm afraid I'm not sure what the right thing to do here is either, but from the 
patch description it does seem like we should have this guarded by an "#ifdef 
CONFIG_32BIT" so we can keep it out of our 32-bit ABI (which isn't in glibc 
yet, so isn't stable) in favor of statx() (or maybe stat64()?).  The one 
problem here is that I can't find "newstat" anywhere in glibc to verify it's 
actually supposed to be part of our 64-bit ABI, though I can find a bunch of 
references to "statx" that seem to indicate it's meant to be present.

That said, assuming you don't have anything wacky going on in userspace if this 
breaks the ABI then it breaks the ABI, so however newstat got into a binary we 
still need to keep it around.  Poking around my Fedora glibc image I see

   000000000009b040 <__xstat>:
      9b040:       e51d                    bnez    a0,9b06e <__xstat+0x2e>
      9b042:       04f00893                li      a7,79
      9b046:       f9c00513                li      a0,-100
      9b04a:       4681                    li      a3,0
      9b04c:       00000073                ecall

which seems to coorespond with sys_newfstatat, which indicates to me we should 
have it in the 64-bit ABI.