Re: [GIT PULL] siginfo fix for v4.16-rc5

From: Eugene Syromiatnikov
Date: Sat Mar 31 2018 - 06:56:46 EST


On Tue, Mar 06, 2018 at 01:11:03AM -0600, Eric W. Biederman wrote:
> Linus,
>
> Please pull the siginfo-linus branch from the git tree:
>
> git://git.kernel.org/pub/scm/linux/kernel/git/ebiederm/user-namespace.git siginfo-linus
>
> HEAD: f6a015498dcaee72f80283cb7873d88deb07129c signal/x86: Include the field offsets in the build time checks
>
> The kbuild test robot found that I accidentally moved si_pkey when I was
> cleaning up siginfo_t. A short followed by an int with the int having 8
> byte alignment. Sheesh siginfo_t is a weird structure.
>
> I have now corrected it and added build time checks that with a little
> luck will catch any similar future mistakes. The build time checks were
> sufficient for me to verify the bug and to verify my fix. So they are
> at least useful this once.
>
> Eric W. Biederman (2):
> signal: Correct the offset of si_pkey in struct siginfo

Looks like this commit changes layout of the siginfo struct on m68k:

pts/0, esyr@fedora: /tmp % cat si.c
#include <stddef.h>
#include "linux/signal.h"

static const size_t lower_offset = offsetof(struct siginfo, si_lower);
pts/0, esyr@fedora: /tmp % m68k-linux-gnu-gcc -Ikhdr-v4.16-rc1\~159\^2\~20/include -g -c si.c -o si-orig.o
pts/0, esyr@fedora: /tmp % m68k-linux-gnu-gcc -Ikhdr-v4.16-rc3\~17\^2/include -g -c si.c -o si-1.o
pts/0, esyr@fedora: /tmp % m68k-linux-gnu-gcc -Ikhdr-v4.16-rc7-194-g29d9d38/include -g -c si.c -o si-2.o
pts/0, esyr@fedora: /tmp % for i in si-orig.o si-1.o si-2.o; do echo -------- $i; objdump -t -j .rodata $i; objdump -s -j .rodata $i; done
-------- si-orig.o

si-orig.o: file format elf32-big

SYMBOL TABLE:
00000000 l d .rodata 00000000 .rodata
00000000 l O .rodata 00000004 lower_offset



si-orig.o: file format elf32-big

Contents of section .rodata:
0000 00000012 ....
-------- si-1.o

si-1.o: file format elf32-big

SYMBOL TABLE:
00000000 l d .rodata 00000000 .rodata
00000000 l O .rodata 00000004 lower_offset



si-1.o: file format elf32-big

Contents of section .rodata:
0000 00000012 ....
-------- si-2.o

si-2.o: file format elf32-big

SYMBOL TABLE:
00000000 l d .rodata 00000000 .rodata
00000000 l O .rodata 00000004 lower_offset



si-2.o: file format elf32-big

Contents of section .rodata:
0000 00000014 ....

So, the offset of the si_lower field is 20 at the current HEAD and was 18 at
commits v4.16-rc3~17^2 and v4.16-rc1~159^2~20. I believe this is due to
the fact that m68k uses 2-byte default alignment and not 4-byte.

> signal/x86: Include the field offsets in the build time checks
>
> arch/x86/kernel/signal_compat.c | 65 ++++++++++++++++++++++++++++++++++++++
> include/linux/compat.h | 4 +--
> include/uapi/asm-generic/siginfo.h | 4 +--
> 3 files changed, 69 insertions(+), 4 deletions(-)
>