This is the mail archive of the
cygwin
mailing list for the Cygwin project.
Re: autossh crash with 20050314 and earlier cygwin1.dll
On Wed, Mar 16, 2005 at 06:06:20PM -0000, Dave Korn wrote:
>----Original Message----
>>From: Andrew Schulman
>>Sent: 16 March 2005 17:08
>
>
>> Would the job be easier in my case, where I have a repeatable crash? I
>> provided strace output in the earlier thread about my autossh problems,
>> but haven't tried using any debug builds of cygwin1.dll. I'm not
>> conversant with debuggers, but I have used gdb before and could work my
>> way through it. Let me know if this would be useful, and how best to go
>> about it.
>
>
> Thanks, it's good of you to offer, but as you may have seen, cgf's already
>put his finger on it: the code in autossh calls free (...) on a pointer that
>it didn't get from one of the *alloc (...)-family functions. That's the
>problem (undefined behaviour) and his fix is the correct solution.
No, it is freeing something that *was* allocated using malloc by cygwin
but cygwin expects that it should be able to control the allocation
since the caller shouldn't be making any assumptions about where the
buffer lives. The buffer could be on the heap or it could be statically
allocated.
> GOK why it works on glibc-based systems, but I guess it must, or the
>problem would surely have cropped up before. Hmmmmm......
>
>dk@mace /gnu/glibc-2.3.4/resolv> cat -n gethnamaddr.c
> [........snip!........]
> 103
> 104 static struct hostent host;
> 105 static char *host_aliases[MAXALIASES];
> 106 static char hostbuf[8*1024];
> [........snip!........]
>
> Well, it also uses a static buffer. Guess glibc free (...) must be more
>robust in the presence of garbage input than newlib free (...), although
>that's odd, since they're both based on dl-malloc under the hood.
Cygwin doesn't use newlib's free.
cgf
--
Unsubscribe info: http://cygwin.com/ml/#unsubscribe-simple
Problem reports: http://cygwin.com/problems.html
Documentation: http://cygwin.com/docs.html
FAQ: http://cygwin.com/faq/