[Linux-ia64] libc umount and umount2 can't both be right

[Linux-ia64] libc umount and umount2 can't both be right

[MONTGOMERY,BOB (HP-FtCollins,ex1)]

The problem:  On IA64, the implementation of umount(2) in glibc
and of sys_umount in the kernel results in the kernel performing
umounts with essentially random flag words.

$ dpkg -l | grep libc6
ii libc6.1       2.3.1-17  GNU C Library: Shared libraries and Timezone
ii libc6.1-dev   2.3.1-17  GNU C Library: Development Libraries and Hea

In libc, umount and umount2 are the same code.  They both use
syscall number 1044 without introducing a call layer. 
   
$ gdb /lib/libc.so.6.1
...
(gdb) disas umount
Dump of assembler code for function umount:
0x1a63a0 <umount>:      [MII]       mov r15\x1044
0x1a63a1 <umount+1>:                break.i 0x100000;;
0x1a63a2 <umount+2>:                cmp.eq p6,p0=-1,r10;;
0x1a63b0 <umount+16>:
    [BBB] (p06) br.cond.spnt.few 0x2d820 <__libc_start_main+640>
0x1a63b1 <umount+17>:               br.ret.sptk.few b0
0x1a63b2 <umount+18>:               nop.b 0x0;;
End of assembler dump.

(gdb) disas umount2
Dump of assembler code for function umount2:
0x1a63c0 <umount2>:     [MII]       mov r15\x1044
0x1a63c1 <umount2+1>:               break.i 0x100000;;
0x1a63c2 <umount2+2>:               cmp.eq p6,p0=-1,r10;;
0x1a63d0 <umount2+16>:
    [BBB] (p06) br.cond.spnt.few 0x2d820 <__libc_start_main+640>
0x1a63d1 <umount2+17>:              br.ret.sptk.few b0
0x1a63d2 <umount2+18>:              nop.b 0x0;;
End of assembler dump.


And from fs/namespace.c, the interface for sys_umount is:

   asmlinkage long sys_umount(char * name, int flags)

The problem is that sys_umount expects two parameters, the sys/umount.h
header only allows the program to specify one parameter for umount(),
and the kernel picks up whatever is in the next output register of the
caller as the flags word.  So umount calls are executed with essentially
random flag words.

The problem could be solved in libc by implementing a wrapper for
umount that looks something like:

extern int umount2 (__const char *__special_file, int __flags) ;

int umount (const char *target)
{
	umount2(target, 0);
}

In this case, umount2 could use the same lightweight method that
it uses now.  

Bob Montgomery, HP

Re: [Linux-ia64] libc umount and umount2 can't both be right

[Ian Wienand]

[originally cc: libc-alpha but screwed up the cc line so it was rejected here]

>On Tue, May 13, 2003 at 04:06:41PM -0400, MONTGOMERY,BOB (HP-FtCollins,ex1) wrote:
> In libc, umount and umount2 are the same code.  They both use
> syscall number 1044 without introducing a call layer. 

looks like a typo in libc to me

-i
ianw@gelato.unsw.edu.au
http://www.gelato.unsw.edu.au

--- libc/sysdeps/unix/sysv/linux/ia64/syscalls.list.old 2003-05-14 09:10:04.000000000 +1000
+++ libc/sysdeps/unix/sysv/linux/ia64/syscalls.list     2003-05-14 09:09:40.000000000 +1000
@@ -1,6 +1,6 @@
 # File name    Caller  Syscall name    # args  Strong name     Weak names
  
-umount2                -       umount          2       __umount2       umount2
+umount2                -       umount2         2       __umount2       umount2
  
 # Whee! 64-bit systems naturally implement llseek.
 llseek         EXTRA   lseek           C:3     __libc_lseek    __lseek lseek __libc_lseek64 __llseek llseek __lseek64 lseek64

Re: [Linux-ia64] libc umount and umount2 can't both be right

[Andreas Schwab]

Ian Wienand <ianw@gelato.unsw.edu.au> writes:

|> [originally cc: libc-alpha but screwed up the cc line so it was rejected here]
|> 
|> >On Tue, May 13, 2003 at 04:06:41PM -0400, MONTGOMERY,BOB (HP-FtCollins,ex1) wrote:
|> > In libc, umount and umount2 are the same code.  They both use
|> > syscall number 1044 without introducing a call layer. 
|> 
|> looks like a typo in libc to me
|> 
|> -i
|> ianw@gelato.unsw.edu.au
|> http://www.gelato.unsw.edu.au
|> 
|> --- libc/sysdeps/unix/sysv/linux/ia64/syscalls.list.old 2003-05-14 09:10:04.000000000 +1000
|> +++ libc/sysdeps/unix/sysv/linux/ia64/syscalls.list     2003-05-14 09:09:40.000000000 +1000
|> @@ -1,6 +1,6 @@
|>  # File name    Caller  Syscall name    # args  Strong name     Weak names
|>   
|> -umount2                -       umount          2       __umount2       umount2
|> +umount2                -       umount2         2       __umount2       umount2

Except there is no umount2 syscall on ia64.

Andreas.

-- 
Andreas Schwab, SuSE Labs, schwab@suse.de
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."

Re: [Linux-ia64] libc umount and umount2 can't both be right

[Ian Wienand]

> Except there is no umount2 syscall on ia64.

yep sorry didn't read the email properly and jumped to an assumption.
appologies.

-i

Re: [Linux-ia64] libc umount and umount2 can't both be right

[Andreas Schwab]

"MONTGOMERY,BOB (HP-FtCollins,ex1)" <bob.montgomery@hp.com> writes:

|> The problem:  On IA64, the implementation of umount(2) in glibc
|> and of sys_umount in the kernel results in the kernel performing
|> umounts with essentially random flag words.

[ SYS_umount is actually umount2 on ia64 ]

Here is a patch:

2003-05-14  Andreas Schwab  <schwab@suse.de>

	* sysdeps/unix/sysv/linux/ia64/umount.c: New file.

--- /dev/null	2003-03-21 07:28:43.000000000 +0100
+++ sysdeps/unix/sysv/linux/ia64/umount.c	2003-05-14 14:33:40.000000000 +0200
@@ -0,0 +1,31 @@
+/* umount system call for Linux/ia64.
+   Copyright (C) 2003 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
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, write to the Free
+   Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
+   02111-1307 USA.  */
+
+#include <sys/mount.h>
+#include <errno.h>
+
+#include <sysdep.h>
+#include <bp-checks.h>
+
+/* Unmount a filesystem.  */
+int
+umount (const char *special_file)
+{
+  return INLINE_SYSCALL (umount, 2, CHECK_STRING (special_file), 0);
+}

Andreas.

-- 
Andreas Schwab, SuSE Labs, schwab@suse.de
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."

RE: [Linux-ia64] libc umount and umount2 can't both be right

[MONTGOMERY,BOB (HP-FtCollins,ex1)]

> > In libc, umount and umount2 are the same code.  They both use
> > syscall number 1044 without introducing a call layer. 
> 
> looks like a typo in libc to me
> 
> -i
> ianw@gelato.unsw.edu.au
> http://www.gelato.unsw.edu.au

I don't think it was a typo.

The problem is that 2.4 IA64 kernels don't have a umount2
entry point.  They have a single entry point to provide
functionality for both umount and umount2 in the library.

Since the single kernel entry point expects two parameters,
the library needs to supply the second parameter (0) in the
case of umount, where the user only supplies the first
one.

Bob Montgomery, HP

> 
> --- libc/sysdeps/unix/sysv/linux/ia64/syscalls.list.old 
> 2003-05-14 09:10:04.000000000 +1000
> +++ libc/sysdeps/unix/sysv/linux/ia64/syscalls.list     
> 2003-05-14 09:09:40.000000000 +1000
> @@ -1,6 +1,6 @@
>  # File name    Caller  Syscall name    # args  Strong name   
>   Weak names
>   
> -umount2                -       umount          2       
> __umount2       umount2
> +umount2                -       umount2         2       
> __umount2       umount2
>   
>  # Whee! 64-bit systems naturally implement llseek.
>  llseek         EXTRA   lseek           C:3     __libc_lseek  
>   __lseek lseek __libc_lseek64 __llseek llseek __lseek64 lseek64
> 

Re: [Linux-ia64] libc umount and umount2 can't both be right

[Ulrich Drepper]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Andreas Schwab wrote:

> [ SYS_umount is actually umount2 on ia64 ]

Indeed.  This is the right patch, I've applied it.  Thanks,

- -- 
- --------------.                        ,-.            444 Castro Street
Ulrich Drepper \    ,-----------------'   \ Mountain View, CA 94041 USA
Red Hat         `--' drepper at redhat.com `---------------------------
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.1 (GNU/Linux)

iD8DBQE+wt7F2ijCOnn/RHQRAk3CAKCtCGqcjD3nhWuJenbeKKfWRWnHFACfdT/D
XdI87pwOvJw8IdGtnhQ0mhY=M4/U
-----END PGP SIGNATURE-----