utime emulation

I'm finding that utime(filename, NULL) doesn't follow specs - this comes up in plodding through the LSB certification test suite.

For example, if the process does not own the file but has write permission, the above referenced call should succeed but fails with EPERM.

utime() is emulated on Itanium: in the kernel, fs/open.c doesn't have a sys_utime routine if __ia64__ or alpha; the emulation comes from glibc's sysdeps/unix/utime.c but appears to be bogus. The problem is that if the second argument to utime is NULL the emulation code does some work to build up a "struct timeval" array as expected by utimes(), and passes that off, instead of passing NULL... and so the proper checks in the NULL case don't get done by the kernel.

I guess this is a query to see if anyone on this list knows anything about this code. Is this just a glibc problem, or should Itanium go back to providing the non-emulated utime routine like nearly all the other arch's do (and so this emulation code is not run).

Mats Wichmann

Mats> I'm finding that utime(filename, NULL) doesn't follow specs - Mats> this comes up in plodding through the LSB certification test Mats> suite.

Mats> For example, if the process does not own the file but has Mats> write permission, the above referenced call should succeed but Mats> fails with EPERM.

Mats> utime() is emulated on Itanium: in the kernel, fs/open.c Mats> doesn't have a sys_utime routine if __ia64__ or alpha; the Mats> emulation comes from glibc's sysdeps/unix/utime.c but appears Mats> to be bogus. The problem is that if the second argument to Mats> utime is NULL the emulation code does some work to build up a Mats> "struct timeval" array as expected by utimes(), and passes Mats> that off, instead of passing NULL... and so the proper checks Mats> in the NULL case don't get done by the kernel.

Mats> I guess this is a query to see if anyone on this list knows Mats> anything about this code. Is this just a glibc problem, or Mats> should Itanium go back to providing the non-emulated utime Mats> routine like nearly all the other arch's do (and so this Mats> emulation code is not run).

Please report this as a glibc bug.

	--david

"Wichmann, Mats D" <[email protected]> writes:

|> I'm finding that utime(filename, NULL) doesn't
|> follow specs - this comes up in plodding through
|> the LSB certification test suite.
|> 
|> For example, if the process does not own the
|> file but has write permission, the above
|> referenced call should succeed but fails with EPERM.
|> 
|> utime() is emulated on Itanium: in the kernel,
|> fs/open.c doesn't have a sys_utime routine 
|> if __ia64__ or alpha; the emulation comes from
|> glibc's sysdeps/unix/utime.c but appears to
|> be bogus.  The problem is that if the second
|> argument to utime is NULL the emulation code
|> does some work to build up a "struct timeval"
|> array as expected by utimes(), and passes that
|> off, instead of passing NULL... and so the
|> proper checks in the NULL case don't get done
|> by the kernel.
|> 
|> I guess this is a query to see if anyone
|> on this list knows anything about this code.
|> Is this just a glibc problem,

Yes. Here is a patch:

2002-08-06 Andreas Schwab <[email protected]>

* sysdeps/unix/utime.c: If TIMES is NULL pass it through to utimes.

@@ -1,4 +1,4 @@
-/* Copyright (C) 1991, 1994, 1997 Free Software Foundation, Inc.
+/* Copyright (C) 1991, 1994, 1997, 2002 Free Software Foundation, Inc.
    This file is part of the GNU C Library.
 
    The GNU C Library is free software; you can redistribute it and/or

@@ -31,7 +31,7 @@ utime (file, times)
      const char *file;
      const struct utimbuf *times;
 {
-  struct timeval timevals[2];
+  struct timeval timevals[2], *tvp;
 
   if (times != NULL)
     {

@@ -39,13 +39,10 @@ utime (file, times)
       timevals[0].tv_usec = 0L;
       timevals[1].tv_sec = (long int) times->modtime;
       timevals[1].tv_usec = 0L;
+      tvp = timevals;
     }
   else
-    {
-      if (__gettimeofday (&timevals[0], NULL) < 0)
-	return -1;
-      timevals[1] = timevals[0];
-    }
+    tvp = NULL;
 
-  return __utimes (file, timevals);
+  return __utimes (file, tvp);
 }

Andreas.

Andreas Schwab, SuSE Labs, [email protected]
SuSE Linux AG, Deutschherrnstr. 15-19, D-90429 Nürnberg
Key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5
"And now for something completely different."