A preloadable library that will print a stacktrace of the program when it crashes (and received SEGV), or upon receiving any other relevant signal.
A lot of times the application resident memory size is so big that collecting core dumps is impractical. However being able to see the crash stacktrace is desirable. Without a stacktrace figuring our the reason of the crash might be very hard if not impossible.
Just link your program again libsegfault. Or use LD_PRELOAD, e.g.
% env LD_PRELOAD=/path/to/libsegfault.so ./app
- Install libunwind: pkg install libunwind
- make && make install
After starting your program with libsegfault preloaded or linked in send the SEGV signal to the program. You should see a stacktrace printed into the stderror.
% env LD_PRELOAD=/path/to/libsegfault.so ./app
% kill -SEGV ${PID}
You should see something along the lines of:
Caught signal 11 (SEGV) in program sleep [27759]
thread frame IP function
[100689] 0: 0x800afbe0a: __sys_nanosleep()+0xa
[100689] 1: 0x80120ebcc: _pthread_suspend_all_np()+0x10dc
[100689] 2: 0x000400a38: <unknown>
[100689] 3: 0x00040086f: <unknown>
[100689] 4: 0x800620000: <unknown>
Backtrace: 0x800afbe0a 0x80120ebcc 0x400a38 0x40086f 0x800620000
Segmentation fault (core dumped)
libsegfaults looks at the SEGFAULT_SIGNALS environment variable on startup. If that variable is not set, only SIGSEGV handler is enabled. If it's set to "all", libsegfault will intercept all the signals that usually result in a crash: SIGSEGV, SIGBUS, SIGILL, SIGABRT, SIGFPE and SIGSYS. Optionally, the variable can also be set to a custom list of space separated signals.
For example the following command will tell /libsegfault/ to intercept only SEGV and SIGILL (illegal instruction) conditions:
% env SEGFAULT_SIGNALS="SEGV ILL" LD_PRELOAD=/path/to/libsegfault.so ./app
To intercept all signals, use:
% env SEGFAULT_SIGNALS=all LD_PRELOAD=/path/to/libsegfault.so ./app