Bug 1007614

*** Description of problem:

The infpy_read_memory() function [gdb/python/py-inferior.c] leaks memory: it leaks a reference to the buffer that is read from the debuggee and is passed to the Python script. When the value goes out of scope or becomes otherwise unreferenced in the Python script, it should be freed, but due to the leaked reference, it is never released.

This is a problem for memory-intensive python scripts. For a real example, see blocked blug 826266; my Python script there wants to read several gigabytes from a coredump. I'll paste a simple reproducer below.

*** Version-Release number of selected component (if applicable):
gdb-7.2-60.el6_4.1.x86_64 (RHEL-6.4.z)

*** How reproducible:
100%

*** Steps to Reproduce:

1. Save the following python code in the file "testprog.py":
--------[snip]--------
class HelloWorld (gdb.Command):
    """Greet the whole world."""

    def __init__ (self):
        super (HelloWorld, self).__init__ ("hello-world",
                                           gdb.COMMAND_OBSCURE)

    def invoke (self, arg, from_tty):
        px = gdb.parse_and_eval("px")
        core = gdb.inferiors()[0]
        for i in range(256 * 1024):
            chunk = core.read_memory(px, 4096)
        print "Hello, World!"

HelloWorld ()
--------[snip]--------

2. Save the following C test program in the file "testprog.c":
--------[snip]--------
#include <unistd.h>

static struct x {
  char unsigned u[4096];
} x, *px = &x;

int
main(void)
{
  pause();
  return 0;
}
--------[snip]--------

3. build it:
    gcc -g3 -o testprog testprog.c

4. Enable coredumps:
    ulimit -c unlimited

5. Disable abrt or at least reconfigure /proc/sys/kernel/core_pattern so that the coredump is written to a local file and not piped to abrt.

6. start the program:
    ./testprog

7. Send SIGQUIT to the program, pressing Ctrl-\, eliciting a coredump.

8. Open the core with gdb:
    gdb ./testprog core.PID

9. Load the python script:
    (gdb) source testprog.py

10. From another terminal, check the initial memory usage of gdb:
    ps -o pid,rsz,vsz,cmd -C gdb

11. Run the python script:
    (gdb) hello-world

12. From another terminal, verify the memory usage of gdb after the 1st run:
    ps -o pid,rsz,vsz,cmd -C gdb

13. Run the python script again:
    (gdb) hello-world

14. From another terminal, verify the memory usage of gdb after the 2nd run:
    ps -o pid,rsz,vsz,cmd -C gdb

*** Actual results:
From step 10 to step 12, 1 GB virtual memory is leaked.
From step 12 to step 14, another GB of virtual memory is leaked.

*** Expected results:
Memory footprint stays reasonably stable.

*** Additional info:
Apparently, this bug has been fixed by upstream gdb git commit

commit 655e820cf9a039ee55325d9e1f8423796d592b4b
Author: Tom Tromey <tromey>
Date:   Wed Mar 28 17:38:07 2012 +0000

        * python/py-inferior.c (infpy_read_memory): Remove cleanups and
        explicitly free 'buffer' on exit paths.  Decref 'membuf_object'
        before returning.

The second part looks relevant here (ie. the Py_DECREF (membuf_obj) call before returning.)

The bug seems to be present in gdb-7.2-64.el6 too (ie. the current RHEL-6.5 development version).