Re: Mbuf double-free guilty party detection patch

From: Peter Holm <peter_at_holm.cc>
Date: Sat, 25 Jun 2005 23:45:03 +0200
On Sat, Jun 25, 2005 at 03:52:01PM -0400, Mike Silbersack wrote:
> > On Sat, Jun 25, 2005 at 12:38:56PM -0400, Mike Silbersack wrote:
> >> > x.123e:This memory last freed by: 0xc2fa6c00
> >> > x.123e-panic: Memory modified after free 0xc2fa6a00(256) val=c2fa6c00
> > (kgdb) x/40x 0xc2fa6a00
> > 0xc2fa6a00:   0xc2fa6c00    0x00000000    0xc24b1ae0    0x00000520
> 
> I'm going to have to doublecheck my patch when I get home - the address
> returned by "This memory last freed by" should be a pointer to a function
> address, not a piece of data.
> 
> So that we get more results, why don't you change the panic in trash_ctor
> to a printf, then you can continue to run tests and see how many mbufs are
> being corrupted without the issue of having to reboot after a panic, etc.
> 
> Mike "Silby" Silbersack

Maybe a panic is better: I got quite a few before I had to reset the
box:

 71 This memory last freed by: 0
 18 This memory last freed by: 0x800
 17 This memory last freed by: 0xdeadc0df
 17 This memory last freed by: 0x1

I also added a backtrace and here's the high score:

 32 tcp_output(c1e4e564) at tcp_output+0xa42
 32 tcp_input(c1a1fc00,14,c1a1fc00,0,0) at tcp_input+0x2b0f
 32 ip_input(c1a1fc00) at ip_input+0x511
 29 vm_pageout_scan(2,c09ab0c0,0,c086e990,5c3) at vm_pageout_scan+0x107

-- 
Peter Holm
Received on Sat Jun 25 2005 - 19:45:09 UTC

This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:38:37 UTC