On Monday, August 15, 2016 5:30:28 PM CEST Yury Norov wrote:

On Mon, Jun 27, 2016 at 09:47:38AM +0200, Andreas Schwab wrote:

quoted

Yury Norov [off-list ref] writes:

quoted

The only difference is that non-compat version forces O_LARGEFILE,
and it should be the default behaviour for all architectures, as
we don't support 32-bit off_t. The only exception is tile32, that
continues with compat version of syscalls.

Signed-off-by: Yury Norov <redacted>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Acked-by: Chris Metcalf <redacted> [for tile]
---
 arch/tile/kernel/compat.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/arch/tile/kernel/compat.c b/arch/tile/kernel/compat.c
index 4912084..489ae19 100644
--- a/arch/tile/kernel/compat.c
+++ b/arch/tile/kernel/compat.c

@@ -94,6 +94,9 @@ COMPAT_SYSCALL_DEFINE5(llseek, unsigned int, fd, unsigned int, offset_high,
 #define compat_sys_readahead sys32_readahead
 #define sys_llseek compat_sys_llseek
 
+#define sys_openat		compat_sys_openat
+#define sys_open_by_handle_at	compat_sys_open_by_handle_at
+
 /* Call the assembly trampolines where necessary. */
 #define compat_sys_rt_sigreturn _compat_sys_rt_sigreturn
 #define sys_clone _sys_clone

This is a no-op.  Did you mean to add this?  Without that the testsuite
of tar fails on ILP32.

The change you suggest below is what I originally sent at the end of
December, but that chunk was lost after one of numerous rebases probably.

http://lkml.iu.edu/hypermail/linux/kernel/1512.3/00510.html

IIRC, the only difference between this couple of syscalls and their
compat versions is that non-compat syscalls set O_LARGEFILE depending
on (BITS_PER_LONG != 32). Next patch of this series introduces
ARCH_32BIT_OFF_T config option to force O_LARGEFILE for 64-bit arches,
and all new 32-bit arches, as ARCH_32BIT_OFF_T is enabled for existing
32-bit arches explicitly.

O_LARGEFILE is controlled like this:
        -#define force_o_largefile() (BITS_PER_LONG != 32)
        +#define force_o_largefile() (!IS_ENABLED(CONFIG_ARCH_32BIT_OFF_T))

So if I restore original patch, everything will work. But my question
is: if ARCH_32BIT_OFF_T disables O_LARGEFILE for tile32 anyway, what
for we redirect native syscall handlers to compat ones? They are
looking identical...

I've lost context here, but let me try to recall what we had
discussed in the past:

- native 32-bit architectures are handled using CONFIG_ARCH_32BIT_OFF_T
- native 64-bit architectures keep setting the flag.
- compat 32-bit architectures need special cases, but the only one
  that exists is tile, which has to keep not setting O_LARGEFILE
  while new compat architectures (ilp32) should set it.
 

If my understanding is correct, we can drop the change in
arch/tile/kernel/compat.c and join this patch with next one in this
series.

Chris, Arnd?

The change in arch/tile/kernel/compat.c can either come before the
below change, or in the same patch, but not later.

	Arnd

quoted

diff --git a/include/uapi/asm-generic/unistd.h b/include/uapi/asm-generic/unistd.h
index a26415b..4dcc38d 100644
--- a/include/uapi/asm-generic/unistd.h
+++ b/include/uapi/asm-generic/unistd.h

@@ -178,7 +178,7 @@ __SYSCALL(__NR_fchownat, sys_fchownat)
 #define __NR_fchown 55
 __SYSCALL(__NR_fchown, sys_fchown)
 #define __NR_openat 56
-__SC_COMP(__NR_openat, sys_openat, compat_sys_openat)
+__SYSCALL(__NR_openat, sys_openat)
 #define __NR_close 57
 __SYSCALL(__NR_close, sys_close)
 #define __NR_vhangup 58

@@ -676,8 +676,7 @@ __SYSCALL(__NR_fanotify_mark, sys_fanotify_mark)
 #define __NR_name_to_handle_at         264
 __SYSCALL(__NR_name_to_handle_at, sys_name_to_handle_at)
 #define __NR_open_by_handle_at         265
-__SC_COMP(__NR_open_by_handle_at, sys_open_by_handle_at, \
-	  compat_sys_open_by_handle_at)
+__SYSCALL(__NR_open_by_handle_at, sys_open_by_handle_at)
 #define __NR_clock_adjtime 266
 __SC_COMP(__NR_clock_adjtime, sys_clock_adjtime, compat_sys_clock_adjtime)
 #define __NR_syncfs 267