From mboxrd@z Thu Jan  1 00:00:00 1970
From: Dave Hansen <haveblue@us.ibm.com>
Subject: race with i_flock?
Date: Tue, 16 Jul 2002 19:27:00 -0700
Sender: linux-fsdevel-owner@vger.kernel.org
Message-ID: <3D34D5F4.2060008@us.ibm.com>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii; format=flowed
Content-Transfer-Encoding: 7bit
Cc: linux-fsdevel@vger.kernel.org
Return-path: <linux-fsdevel-owner@vger.kernel.org>
To: Matthew Wilcox <willy@debian.org>
List-Id: linux-fsdevel.vger.kernel.org

I've been getting some funny freezes while running Specweb.  I don't 
think this is the cause, but I did run into it.

Program received signal SIGSEGV, Segmentation fault.
0xc0147540 in may_open (nd=0xdaa69f78, acc_mode=6, flag=3)
     at /home/dave/oprofile/linux-2.5.25/include/linux/fs.h:1047

which is:
static inline int get_lease(struct inode *inode, unsigned int mode)
{
------->if (inode->i_flock && (inode->i_flock->fl_flags & FL_LEASE))
                 return __get_lease(inode, mode);
         return 0;
}

It appears that i_flock is NULL:

(gdb) print inode
$1 = (struct inode *) 0xe8d638f0
(gdb) print *inode
$2 = {...big snip
      i_flock = 0x0,
      ...}
(gdb) print &inode->i_flock
$3 = (struct file_lock **) 0xe8d6396c

But, there was a check for that just a second earlier.  Looks racy to 
me.  I noticed that there is use of i_sem in some of the calling 
functions, but not in get_lease()'s call sequence.  Is this a problem? 
    A quick grep for i_flock didn't show any obvious places where it 
was set back to NULL.

#0  0xc0147540 in may_open (nd=0xdaa69f78, acc_mode=6, flag=3)
     at /home/dave/oprofile/linux-2.5.25/include/linux/fs.h:1047
#1  0xc014786c in open_namei (pathname=0xe7292000 
"/mnt/sdc1/www/post.log", flag=3, mode=438,
     nd=0xdaa69f78) at namei.c:1294
#2  0xc013bcbb in filp_open (filename=0xe7292000 
"/mnt/sdc1/www/post.log", flags=2, mode=438)
     at open.c:616
#3  0xc013c0df in sys_open (filename=0x8130cd0 
"/mnt/sdc1/www/post.log", flags=2, mode=438)
     at open.c:806
#4  0xc0106fbb in syscall_call ()


-- 
Dave Hansen
haveblue@us.ibm.com