SIGSEGV in strcat, strlen, ...

Question

I am trying to debug a core file, and am having problems. The app cores with a SIGSEGV after days of running. All of the core files end up crashing in some libc.so.1 string function (like strcat, strcpy, strlen). The backtrace is always screwed up, so that only the first function shows in the bt (I see that the framepointer=0x0). Viewing the assembler code, I cannot understand how a core occurs: e.g. in the strcat case, it appears to be trying to find the end of the string "/usr/share/lib/zoneinfo" to append a "/", and crashes on the line comparing the "e" of "zoneinfo" to null. (Is there some pipelining going on so that the crash address reported in the core is inaccurate?)
Another thing: I'm pretty sure that the code is trying to do some internationalization of a time. The functions localtime() and _tzload() appear to be in the backtrace.
Any hints on how I should approach this?
Bob

Answer 1

Hallo Bob,
I had a similar problem in localtime() in Solaris 2.6
My circumvention is:
a) use localtime_r with supply of my own buffer.
b) trying not to call it too often by saving my last conversion in a static
together with a static time_t when I did it.
Erich