Re: tst-posix_fallocate test failure

[Renaud Marquet <rmarquet at gmail dot com>]

As suggested I did a little step into the failing test. But I cannot
figure out what's happening. I hope you'll get a clearer idea of the
problem with this. Here's a dump of the gdb session :

(gdb) break do_test
Breakpoint 1 at 0x10000758
(gdb) run
Starting program: /tmp/fstat/test 

Breakpoint 1, do_test () at test.c:25
25	  if (posix_fallocate (fd, 50, 1024) != 0)
(gdb) break posix_fallocate
Breakpoint 2 at 0x4000014342c:
file ../sysdeps/unix/sysv/linux/wordsize-64/posix_fallocate.c, line 38.
(gdb) cont
Continuing.

Breakpoint 2, posix_fallocate (fd=4, offset=50, len=1024)
at ../sysdeps/unix/sysv/linux/wordsize-64/posix_fallocate.c:38
38	  if (__builtin_expect (__have_fallocate >= 0, 1))
(gdb) s
35	{
(gdb) 
38	  if (__builtin_expect (__have_fallocate >= 0, 1))
(gdb) 
42	      int res = INTERNAL_SYSCALL (fallocate, err, 4, fd, 0, offset,
len);
(gdb) 
44	      if (! INTERNAL_SYSCALL_ERROR_P (res, err))
(gdb) 
42	      int res = INTERNAL_SYSCALL (fallocate, err, 4, fd, 0, offset,
len);
(gdb) 
44	      if (! INTERNAL_SYSCALL_ERROR_P (res, err))
(gdb) 
58	}
(gdb) 
do_test () at test.c:31
31	  if (fstat64 (fd, &st) != 0)
(gdb) cont
Continuing.
st.st_size : 0
file size changed in second posix_fallocate

Program exited with code 01.
(gdb) 

I don't understand why the code in lines 42 and 44 seems to be executed
twice (as I'm not familiar with gdb I don't know if it's related to gdb
or could be the source of the problem). I took a look at
INTERNAL_SYSCALL macro definition, it's mainly assembler code (in
sysdeps/unix/sysv/linux/powerpc/powerpc64/sysdep.h)

excerpt of sysdeps/unix/sysv/linux/wordsize-64/posix_fallocate.c :

32 /* Reserve storage for the data of the file associated with FD.  */
33 int
34 posix_fallocate (int fd, __off_t offset, __off_t len)
35 {
36 #ifdef __NR_fallocate
37 # ifndef __ASSUME_FALLOCATE
38   if (__builtin_expect (__have_fallocate >= 0, 1))
39 # endif
40     {
41       INTERNAL_SYSCALL_DECL (err);
42       int res = INTERNAL_SYSCALL (fallocate, err, 4, fd, 0, offset,
len);
43 
44       if (! INTERNAL_SYSCALL_ERROR_P (res, err))
45         return 0;
46 
47 # ifndef __ASSUME_FALLOCATE
48       if (__builtin_expect (INTERNAL_SYSCALL_ERRNO (res, err) ==
ENOSYS, 0))
49         __have_fallocate = -1;
50       else
51 # endif
52         if (INTERNAL_SYSCALL_ERRNO (res, err) != EOPNOTSUPP)
53           return INTERNAL_SYSCALL_ERRNO (res, err);
54     }
55 #endif
56 
57   return internal_fallocate (fd, offset, len);
58 }

Cheers,
Renaud

Le mardi 25 novembre 2008 Ã 09:10 -0500, Carlos O'Donell a Ãcrit :
> On Mon, Nov 24, 2008 at 2:00 PM, Renaud Marquet <rmarquet@gmail.com> wrote:
> > I'm compiling glibc (2.8-20080929) for playstation3 (PowerPC 64bits).
> > Building was ok but when I ran the test suite, there was only one
> > failure, concerning posix_fallocate :
> >
> > "file size after first posix_fallocate call is 0, expected 1280"
> >
> > It seems the function does nothing at all although do not send any error
> > code. I checked fstat64 function work correctly on another file, so the
> > problem comes from posix_fallocate.
> 
> This is very little information. Have you tried to debug the test?
> 
> Cheers,
> Carlos.