GDB: find value in whole memory of dump

Question

I have a process coredump on a line like

// ptr is boost::shared_ptr<Whatever>
assert(ptr.unique())

That is there are two shared_ptr referencing same object, but program logically expects exclusive ownership. It's logic problem, not memory corruption.

Using gdb I can see address (e.g. 0x001234567890) of pointer, contained in ptr and verify that it's use_count == 2.

Using hexdump on something like it I can easily find other occurrences:

$ xxd core2 | fgrep '9078 5634 1200'
114e3e1e0109c 002c 307f 0000 9078 5634 1200 0000  ...
15b8b2ba000d7 ffa2 307f 0000 9078 5634 1200 0000  ...
1618b644000e7 7fa3 307f 0000 9078 5634 1200 0000  ...

There is a command find in gdb than can search specified region for specified value, but it need correct allocated memory region.

There is a command info mem which shows info about memory regions, but it doesn't work for coredumps.

Are there other ways to find, where this address/value is stored?

phd · Accepted Answer · 2017-08-07 18:24:25Z

0

To find the pointers pointing at a piece of memory, you can use valgrind and gdb together.

Then from gdb, you can use the monitor command

(gdb)  monitor who_points_at <addr> [<len>]

to find where are these pointers.

So, if you have 2 references to a piece of memory, who_points_at should be able to return them.

Note that who_points_at is searching for any 'word aligned' data that points at the range [addr, addr + len( So, you might find more occurences as e.g. an integer might 'look' like it is pointing at this memory.

See http://www.valgrind.org/docs/manual/manual-core-adv.html#manual-core-adv.gdbserver and http://www.valgrind.org/docs/manual/mc-manual.html#mc-manual.monitor-commands

for more info.

answered Aug 7, 2017 at 18:24

phd

3,8821 gold badge15 silver badges14 bronze badges

Sign up to request clarification or add additional context in comments.

1 Comment

Gizmo Over a year ago

not a real solution when debugging core dumps. There is nothing to monitor. "monitor" command not supported by this target.

MadRunner · Accepted Answer · 2017-08-03 10:50:40Z

0

You should consider using valgrind or some address sanitizer (e.g. compile with -fsanitize=address instrumentation option to GCC). It probably is the easiest way to help finding your bug (assuming your bug is reproducible).

Regarding your original question, recent GDB are extendable in Python (and perhaps in Guile). You might write a script for your needs.

You could also put a watchpoint on the use_count field of your smart pointer or pointed object (you may need to disable ASLR to ease debugging).

edited Aug 3, 2017 at 10:50

MadRunner

6509 silver badges18 bronze badges

answered Aug 3, 2017 at 7:13

Basile Starynkevitch

231k18 gold badges323 silver badges578 bronze badges

1 Comment

MadRunner Over a year ago

Sorry for misguidance, but it's not a memory problem. It seems everything GDB provides for python available in plain GDB so it's not an answer. Watchpoint may solve the problem, but there are many dynamically created pointers in runtime and problem arise very seldom.

Collectives™ on Stack Overflow

GDB: find value in whole memory of dump

2 Answers 2

1 Comment

1 Comment

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

2 Answers 2

1 Comment

1 Comment

Your Answer

Sign up or log in

Post as a guest

Related