Segmentation Fault on "out of memory" in backtrace_free_locked

[piesarmin]

The following stacktrace snippet shows the problem:

#0  backtrace_free_locked (state=state@entry=0x7f2da76b2000, addr=addr@entry=0x0, size=size@entry=784) at mmap.c:104
#1  0x0000000000e88533 in backtrace_free_locked (size=784, addr=0x0, state=0x7f2da76b2000) at mmap.c:226
#2  backtrace_free (state=state@entry=0x7f2da76b2000, addr=0x0, size=784, error_callback=error_callback@entry=0xc704d0 <boost::stacktrace::detail::libbacktrace_error_callback(void*, char const*, int)>, data=data@entry=0x7f2e8b8765c0)
    at mmap.c:233
#3  0x0000000000e8bef3 in free_line_header (data=0x7f2e8b8765c0, error_callback=0xc704d0 <boost::stacktrace::detail::libbacktrace_error_callback(void*, char const*, int)>, hdr=0x7f2e8b876310, state=0x7f2da76b2000) at dwarf.c:2669
#4  read_line_info (lines_count=<synthetic pointer>, lines=<synthetic pointer>, hdr=0x7f2e8b876310, u=<optimized out>, data=0x7f2da7676e30,
    error_callback=0xc704d0 <boost::stacktrace::detail::libbacktrace_error_callback(void*, char const*, int)>, ddata=0x7f2e8b8765c0, state=0x7f2d8f8e41e0) at dwarf.c:3340

The parameter addr is NULL. That leads to a segmentation fault in mmap.c:104:

103  p = (struct backtrace_freelist_struct *) addr;
104  p->next = state->freelist;

Due to this segmentation fault, the actual problem, namely the 'out of memory' error, is obscured. With the following patch in the function free_line_header in dwarf.c, especially with the check "hdr->filenames != NULL", we were able to prevent the crash at this point:

static void
free_line_header (struct backtrace_state *state, struct line_header *hdr,
		  backtrace_error_callback error_callback, void *data)
{
  if ((hdr->dirs_count != 0) && (hdr->dirs != NULL))
    {
      backtrace_free(state,
                     hdr->dirs,
                     hdr->dirs_count * sizeof(const char *),
                     error_callback,
                     data);
    }
  if ((hdr->filenames_count != 0) && (hdr->filenames != NULL))
    {
      backtrace_free(state,
                     hdr->filenames,
                     hdr->filenames_count * sizeof(char *),
                     error_callback,
                     data);
    }
}

Please check our patch. It might also make more sense to perform the check directly in backtrace_free_locked? Thank you for your effort.