Re: Patch to fixe Data Acess error in dup_fd

[Sharyathi Nagesh <sharyath@in.ibm.com>]

Excuse me for the late response

On Fri, 2006-11-17 at 18:38 +0530, Sharyathi Nagesh wrote:

> > I'm guessing that you've already tried this, but it never hurts to be
> > sure: does this machine pass memtest? :)
    Machine passes memtest without fail, even after running for many
hours.
    Since the problem looked genuine to me and may hit again in future
do let me know how I should improve the patch. Please let us know of the
opinion...
---------------------------------------------------------------------
To restate the problem:
Data access error was observed after testing IO and TCP Stress test for
more than 72 hrs over ppc-64 machine.

[c00000007ce2fa70] c000000000060d28 .dup_fd+0x1d8/0x39c (unreliable)
[c00000007ce2fb30] c000000000060f48 .copy_files+0x5c/0x88
[c00000007ce2fbd0] c000000000061f5c .copy_process+0x574/0x1520
[c00000007ce2fcd0] c000000000062f88 .do_fork+0x80/0x1c4
[c00000007ce2fdc0] c000000000011790 .sys_clone+0x5c/0x74
[c00000007ce2fe30] c000000000008950 .ppc_clone+0x8/0xc
--- Exception: c00 (System Call) at 000000000fee9c60

The location of the problem was 
-----------
for (i = open_files; i != 0; i--) {
                struct file *f = *old_fds++;
                if (f) {
                        get_file(f); <== Get file does increment of f_count of 
     
                                         struct file
                } else {
-----------
in kernel/fork.c.Accessing f_count gives Data access error as 'struct file' pointer f is 
pointing to invalid location

Regards
Sharyathi Nagesh