How to automatically generate a stacktrace when my program crashes

For Linux and I believe Mac OS X, if you're using gcc, or any compiler that uses glibc, you can use the backtrace() functions in execinfo.h to print a stacktrace and exit gracefully when you get a segmentation fault. Documentation can be found in the libc manual.

Here's an example program that installs a SIGSEGV handler and prints a stacktrace to stderr when it segfaults. The baz() function here causes the segfault that triggers the handler:

#include <stdio.h>
#include <execinfo.h>
#include <signal.h>
#include <stdlib.h>
#include <unistd.h>


void handler(int sig) {
  void *array[10];
  size_t size;

  // get void*'s for all entries on the stack
  size = backtrace(array, 10);

  // print out all the frames to stderr
  fprintf(stderr, "Error: signal %d:\n", sig);
  backtrace_symbols_fd(array, size, STDERR_FILENO);
  exit(1);
}

void baz() {
 int *foo = (int*)-1; // make a bad pointer
  printf("%d\n", *foo);       // causes segfault
}

void bar() { baz(); }
void foo() { bar(); }


int main(int argc, char **argv) {
  signal(SIGSEGV, handler);   // install our handler
  foo(); // this will call foo, bar, and baz.  baz segfaults.
}

Compiling with -g -rdynamic gets you symbol info in your output, which glibc can use to make a nice stacktrace:

$ gcc -g -rdynamic ./test.c -o test

Executing this gets you this output:

$ ./test
Error: signal 11:
./test(handler+0x19)[0x400911]
/lib64/tls/libc.so.6[0x3a9b92e380]
./test(baz+0x14)[0x400962]
./test(bar+0xe)[0x400983]
./test(foo+0xe)[0x400993]
./test(main+0x28)[0x4009bd]
/lib64/tls/libc.so.6(__libc_start_main+0xdb)[0x3a9b91c4bb]
./test[0x40086a]

This shows the load module, offset, and function that each frame in the stack came from. Here you can see the signal handler on top of the stack, and the libc functions before main in addition to main, foo, bar, and baz.


It's even easier than "man backtrace", there's a little-documented library (GNU specific) distributed with glibc as libSegFault.so, which was I believe was written by Ulrich Drepper to support the program catchsegv (see "man catchsegv").

This gives us 3 possibilities. Instead of running "program -o hai":

  1. Run within catchsegv:

    $ catchsegv program -o hai
    
  2. Link with libSegFault at runtime:

    $ LD_PRELOAD=/lib/libSegFault.so program -o hai
    
  3. Link with libSegFault at compile time:

    $ gcc -g1 -lSegFault -o program program.cc
    $ program -o hai
    

In all 3 cases, you will get clearer backtraces with less optimization (gcc -O0 or -O1) and debugging symbols (gcc -g). Otherwise, you may just end up with a pile of memory addresses.

You can also catch more signals for stack traces with something like:

$ export SEGFAULT_SIGNALS="all"       # "all" signals
$ export SEGFAULT_SIGNALS="bus abrt"  # SIGBUS and SIGABRT

The output will look something like this (notice the backtrace at the bottom):

*** Segmentation fault Register dump:

 EAX: 0000000c   EBX: 00000080   ECX:
00000000   EDX: 0000000c  ESI:
bfdbf080   EDI: 080497e0   EBP:
bfdbee38   ESP: bfdbee20

 EIP: 0805640f   EFLAGS: 00010282

 CS: 0073   DS: 007b   ES: 007b   FS:
0000   GS: 0033   SS: 007b

 Trap: 0000000e   Error: 00000004  
OldMask: 00000000  ESP/signal:
bfdbee20   CR2: 00000024

 FPUCW: ffff037f   FPUSW: ffff0000  
TAG: ffffffff  IPOFF: 00000000  
CSSEL: 0000   DATAOFF: 00000000  
DATASEL: 0000

 ST(0) 0000 0000000000000000   ST(1)
0000 0000000000000000  ST(2) 0000
0000000000000000   ST(3) 0000
0000000000000000  ST(4) 0000
0000000000000000   ST(5) 0000
0000000000000000  ST(6) 0000
0000000000000000   ST(7) 0000
0000000000000000

Backtrace:
/lib/libSegFault.so[0xb7f9e100]
??:0(??)[0xb7fa3400]
/usr/include/c++/4.3/bits/stl_queue.h:226(_ZNSt5queueISsSt5dequeISsSaISsEEE4pushERKSs)[0x805647a]
/home/dbingham/src/middle-earth-mud/alpha6/src/engine/player.cpp:73(_ZN6Player5inputESs)[0x805377c]
/home/dbingham/src/middle-earth-mud/alpha6/src/engine/socket.cpp:159(_ZN6Socket4ReadEv)[0x8050698]
/home/dbingham/src/middle-earth-mud/alpha6/src/engine/socket.cpp:413(_ZN12ServerSocket4ReadEv)[0x80507ad]
/home/dbingham/src/middle-earth-mud/alpha6/src/engine/socket.cpp:300(_ZN12ServerSocket4pollEv)[0x8050b44]
/home/dbingham/src/middle-earth-mud/alpha6/src/engine/main.cpp:34(main)[0x8049a72]
/lib/tls/i686/cmov/libc.so.6(__libc_start_main+0xe5)[0xb7d1b775]
/build/buildd/glibc-2.9/csu/../sysdeps/i386/elf/start.S:122(_start)[0x8049801]

If you want to know the gory details, the best source is unfortunately the source: See http://sourceware.org/git/?p=glibc.git;a=blob;f=debug/segfault.c and its parent directory http://sourceware.org/git/?p=glibc.git;a=tree;f=debug


Linux

While the use of the backtrace() functions in execinfo.h to print a stacktrace and exit gracefully when you get a segmentation fault has already been suggested, I see no mention of the intricacies necessary to ensure the resulting backtrace points to the actual location of the fault (at least for some architectures - x86 & ARM).

The first two entries in the stack frame chain when you get into the signal handler contain a return address inside the signal handler and one inside sigaction() in libc. The stack frame of the last function called before the signal (which is the location of the fault) is lost.

Code

#ifndef _GNU_SOURCE
#define _GNU_SOURCE
#endif
#ifndef __USE_GNU
#define __USE_GNU
#endif

#include <execinfo.h>
#include <signal.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <ucontext.h>
#include <unistd.h>

/* This structure mirrors the one found in /usr/include/asm/ucontext.h */
typedef struct _sig_ucontext {
 unsigned long     uc_flags;
 ucontext_t        *uc_link;
 stack_t           uc_stack;
 sigcontext_t      uc_mcontext;
 sigset_t          uc_sigmask;
} sig_ucontext_t;

void crit_err_hdlr(int sig_num, siginfo_t * info, void * ucontext)
{
 void *             array[50];
 void *             caller_address;
 char **            messages;
 int                size, i;
 sig_ucontext_t *   uc;

 uc = (sig_ucontext_t *)ucontext;

 /* Get the address at the time the signal was raised */
#if defined(__i386__) // gcc specific
 caller_address = (void *) uc->uc_mcontext.eip; // EIP: x86 specific
#elif defined(__x86_64__) // gcc specific
 caller_address = (void *) uc->uc_mcontext.rip; // RIP: x86_64 specific
#else
#error Unsupported architecture. // TODO: Add support for other arch.
#endif

 fprintf(stderr, "signal %d (%s), address is %p from %p\n", 
  sig_num, strsignal(sig_num), info->si_addr, 
  (void *)caller_address);

 size = backtrace(array, 50);

 /* overwrite sigaction with caller's address */
 array[1] = caller_address;

 messages = backtrace_symbols(array, size);

 /* skip first stack frame (points here) */
 for (i = 1; i < size && messages != NULL; ++i)
 {
  fprintf(stderr, "[bt]: (%d) %s\n", i, messages[i]);
 }

 free(messages);

 exit(EXIT_FAILURE);
}

int crash()
{
 char * p = NULL;
 *p = 0;
 return 0;
}

int foo4()
{
 crash();
 return 0;
}

int foo3()
{
 foo4();
 return 0;
}

int foo2()
{
 foo3();
 return 0;
}

int foo1()
{
 foo2();
 return 0;
}

int main(int argc, char ** argv)
{
 struct sigaction sigact;

 sigact.sa_sigaction = crit_err_hdlr;
 sigact.sa_flags = SA_RESTART | SA_SIGINFO;

 if (sigaction(SIGSEGV, &sigact, (struct sigaction *)NULL) != 0)
 {
  fprintf(stderr, "error setting signal handler for %d (%s)\n",
    SIGSEGV, strsignal(SIGSEGV));

  exit(EXIT_FAILURE);
 }

 foo1();

 exit(EXIT_SUCCESS);
}

Output

signal 11 (Segmentation fault), address is (nil) from 0x8c50
[bt]: (1) ./test(crash+0x24) [0x8c50]
[bt]: (2) ./test(foo4+0x10) [0x8c70]
[bt]: (3) ./test(foo3+0x10) [0x8c8c]
[bt]: (4) ./test(foo2+0x10) [0x8ca8]
[bt]: (5) ./test(foo1+0x10) [0x8cc4]
[bt]: (6) ./test(main+0x74) [0x8d44]
[bt]: (7) /lib/libc.so.6(__libc_start_main+0xa8) [0x40032e44]

All the hazards of calling the backtrace() functions in a signal handler still exist and should not be overlooked, but I find the functionality I described here quite helpful in debugging crashes.

It is important to note that the example I provided is developed/tested on Linux for x86. I have also successfully implemented this on ARM using uc_mcontext.arm_pc instead of uc_mcontext.eip.

Here's a link to the article where I learned the details for this implementation: http://www.linuxjournal.com/article/6391


Even though a correct answer has been provided that describes how to use the GNU libc backtrace() function1 and I provided my own answer that describes how to ensure a backtrace from a signal handler points to the actual location of the fault2, I don't see any mention of demangling C++ symbols output from the backtrace.

When obtaining backtraces from a C++ program, the output can be run through c++filt1 to demangle the symbols or by using abi::__cxa_demangle1 directly.

  • 1 Linux & OS X Note that c++filt and __cxa_demangle are GCC specific
  • 2 Linux

The following C++ Linux example uses the same signal handler as my other answer and demonstrates how c++filt can be used to demangle the symbols.

Code:

class foo
{
public:
    foo() { foo1(); }

private:
    void foo1() { foo2(); }
    void foo2() { foo3(); }
    void foo3() { foo4(); }
    void foo4() { crash(); }
    void crash() { char * p = NULL; *p = 0; }
};

int main(int argc, char ** argv)
{
    // Setup signal handler for SIGSEGV
    ...

    foo * f = new foo();
    return 0;
}

Output (./test):

signal 11 (Segmentation fault), address is (nil) from 0x8048e07
[bt]: (1) ./test(crash__3foo+0x13) [0x8048e07]
[bt]: (2) ./test(foo4__3foo+0x12) [0x8048dee]
[bt]: (3) ./test(foo3__3foo+0x12) [0x8048dd6]
[bt]: (4) ./test(foo2__3foo+0x12) [0x8048dbe]
[bt]: (5) ./test(foo1__3foo+0x12) [0x8048da6]
[bt]: (6) ./test(__3foo+0x12) [0x8048d8e]
[bt]: (7) ./test(main+0xe0) [0x8048d18]
[bt]: (8) ./test(__libc_start_main+0x95) [0x42017589]
[bt]: (9) ./test(__register_frame_info+0x3d) [0x8048981]

Demangled Output (./test 2>&1 | c++filt):

signal 11 (Segmentation fault), address is (nil) from 0x8048e07
[bt]: (1) ./test(foo::crash(void)+0x13) [0x8048e07]
[bt]: (2) ./test(foo::foo4(void)+0x12) [0x8048dee]
[bt]: (3) ./test(foo::foo3(void)+0x12) [0x8048dd6]
[bt]: (4) ./test(foo::foo2(void)+0x12) [0x8048dbe]
[bt]: (5) ./test(foo::foo1(void)+0x12) [0x8048da6]
[bt]: (6) ./test(foo::foo(void)+0x12) [0x8048d8e]
[bt]: (7) ./test(main+0xe0) [0x8048d18]
[bt]: (8) ./test(__libc_start_main+0x95) [0x42017589]
[bt]: (9) ./test(__register_frame_info+0x3d) [0x8048981]

The following builds on the signal handler from my original answer and can replace the signal handler in the above example to demonstrate how abi::__cxa_demangle can be used to demangle the symbols. This signal handler produces the same demangled output as the above example.

Code:

void crit_err_hdlr(int sig_num, siginfo_t * info, void * ucontext)
{
    sig_ucontext_t * uc = (sig_ucontext_t *)ucontext;

    void * caller_address = (void *) uc->uc_mcontext.eip; // x86 specific

    std::cerr << "signal " << sig_num 
              << " (" << strsignal(sig_num) << "), address is " 
              << info->si_addr << " from " << caller_address 
              << std::endl << std::endl;

    void * array[50];
    int size = backtrace(array, 50);

    array[1] = caller_address;

    char ** messages = backtrace_symbols(array, size);    

    // skip first stack frame (points here)
    for (int i = 1; i < size && messages != NULL; ++i)
    {
        char *mangled_name = 0, *offset_begin = 0, *offset_end = 0;

        // find parantheses and +address offset surrounding mangled name
        for (char *p = messages[i]; *p; ++p)
        {
            if (*p == '(') 
            {
                mangled_name = p; 
            }
            else if (*p == '+') 
            {
                offset_begin = p;
            }
            else if (*p == ')')
            {
                offset_end = p;
                break;
            }
        }

        // if the line could be processed, attempt to demangle the symbol
        if (mangled_name && offset_begin && offset_end && 
            mangled_name < offset_begin)
        {
            *mangled_name++ = '\0';
            *offset_begin++ = '\0';
            *offset_end++ = '\0';

            int status;
            char * real_name = abi::__cxa_demangle(mangled_name, 0, 0, &status);

            // if demangling is successful, output the demangled function name
            if (status == 0)
            {    
                std::cerr << "[bt]: (" << i << ") " << messages[i] << " : " 
                          << real_name << "+" << offset_begin << offset_end 
                          << std::endl;

            }
            // otherwise, output the mangled function name
            else
            {
                std::cerr << "[bt]: (" << i << ") " << messages[i] << " : " 
                          << mangled_name << "+" << offset_begin << offset_end 
                          << std::endl;
            }
            free(real_name);
        }
        // otherwise, print the whole line
        else
        {
            std::cerr << "[bt]: (" << i << ") " << messages[i] << std::endl;
        }
    }
    std::cerr << std::endl;

    free(messages);

    exit(EXIT_FAILURE);
}

Might be worth looking at Google Breakpad, a cross-platform crash dump generator and tools to process the dumps.