utimensat fails to update ctime

[Eric Blake <ebb9@byu.net>]

POSIX requires that utimensat/futimens must update ctime in all cases
where any change is made (it only exempts when both atime and mtime were
requested as UTIME_OMIT, where the file must exist but no change is made).
 Unfortunately, when atime is specified and mtime is UTIME_OMIT, the
kernel mistakenly behaves like read(), by updating atime but not ctime.
This in turn caused a regression in coreutils 8.2, visible through 'touch -a':
http://lists.gnu.org/archive/html/bug-coreutils/2009-12/msg00171.html

Here is a simple program demonstrating the failure:
$ cat foo.c
#include <fcntl.h>
#include <unistd.h>
#include <sys/stat.h>
int
main ()
{
  int fd = creat ("file", 0600);
  struct stat st1, st2;
  struct timespec t[2] = { { 1000000000, 0 }, { 0, UTIME_OMIT } };
  fstat (fd, &st1);
  sleep (1);
  futimens (fd, t);
  fstat (fd, &st2);
  return st1.st_ctime == st2.st_ctime;
}
$ gcc -o foo foo.c -D_GNU_SOURCE
$ ./foo; echo $?
1

The exit status should have been 0.

GNU coreutils will end up working around the bug by calling fstat/[l]stat
prior to futimens/utimensat, and populating the mtime field with the
desired value rather than using UTIME_OMIT.  But this is a pointless stat
call, which could be avoided if the kernel were fixed to comply with POSIX
by updating ctime even when mtime is UTIME_OMIT.

-- 
Don't work too hard, make some time for fun as well!

Eric Blake             ebb9@byu.net