Skip to contentSkip to conversation

libc umount and umount2 can't both be right

Linux-ia64 MONTGOMERY,BOB (HP-FtCollins,ex1) 5 participants 7 messages May 13, 2003 – May 15, 2003

lore
Share & tools
lore
#0 lore raw

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 r1544
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 r1544
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

#1 lore raw

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

Quoted 3 lines · On Tue, May 13, 2003 at 04:06:41PM -0400, 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
[email protected]
http://www.gelato.unsw.edu.au

sysdeps/unix/sysv/linux/ia64/syscalls.list 2003-05-14 09:09:40.000000000 +1000 renamed from sysdeps/unix/sysv/linux/ia64/syscalls.list.old 2003-05-14 09:10:04.000000000 +1000 +1 -1
@@ -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
#2 lore raw
In reply to MONTGOMERY,BOB (HP-FtCollins,ex1)

Ian Wienand <[email protected]> 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
|> [email protected]
|> 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.

Signature · 4 lines
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."
#4 lore raw
In reply to MONTGOMERY,BOB (HP-FtCollins,ex1)

"MONTGOMERY,BOB (HP-FtCollins,ex1)" <[email protected]> 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 <[email protected]>

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

unix/sysv/linux/ia64/umount.c new file +31 -0
@@ -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.

Signature · 4 lines
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."
#5 lore raw
In reply to MONTGOMERY,BOB (HP-FtCollins,ex1)
Quoted 9 lines
Quoted 3 lines

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
[email protected]
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

Quoted 18 lines

--- 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
(quoted diff) +0 -1
@@ -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

#6 lore raw
In reply to MONTGOMERY,BOB (HP-FtCollins,ex1)

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

Quoted 1 line · Andreas Schwab

[ 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-----