Bug #2664: scenario template: ndn::Name double free in __cxa_finalize - ndnSIM - NDN project issue tracking system

Actions

Copy link

Bug #2664

closed

scenario template: ndn::Name double free in __cxa_finalize

Added by Junxiao Shi over 9 years ago. Updated over 9 years ago.

Status:

Closed

Priority:

Normal

Assignee:

Junxiao Shi

Category:

build

Target version:

Start date:

03/22/2015

Due date:

% Done:

100%

Estimated time:

0.50 h

Description

Steps to reproduce:

install ndnSIM, and apply #2662 note-1 workaround
clone ndnSIM-scenario-template
create file ndnSIM-scenario-template/scenarios/hello.cpp with following code:
```
#include <ns3/ndnSIM-module.h>

int
main()
{
  return 0;
}
```
execute CXXFLAGS='-std=c++0x -Wall' ./waf configure --debug (#2663 note-1 workaround) followed by ./waf --run=hello

Expected: code builds and scenario terminates without error

Actual: code builds, but scenario fails with *** glibc detected *** build/hello: double free or corruption (!prev): 0x00000000014239d0 ***

======= Backtrace: =========
/lib/x86_64-linux-gnu/libc.so.6( 0x7de16)[0x7fdaaa543e16]
build/hello(__gnu_cxx::new_allocator<ndn::Block>::deallocate(ndn::Block*, unsigned long) 0x20)[0x4d6654]
build/hello(std::_Vector_base<ndn::Block, std::allocator<ndn::Block> >::_M_deallocate(ndn::Block*, unsigned long) 0x32)[0x4d53c4]
build/hello(std::_Vector_base<ndn::Block, std::allocator<ndn::Block> >::~_Vector_base() 0x56)[0x4d4e12]
build/hello(std::vector<ndn::Block, std::allocator<ndn::Block> >::~vector() 0x42)[0x4d3a54]
build/hello(ndn::Block::~Block() 0x1d)[0x4d06df]
build/hello(ndn::Name::~Name() 0x1d)[0x4d0d7b]
/lib/x86_64-linux-gnu/libc.so.6(__cxa_finalize 0x9d)[0x7fdaaa5019cd]
/usr/local/lib/libns3-dev-ndnSIM-debug.so( 0xb0f436)[0x7fdab0d95436]

Files

Download all files

2664-print-Name-dtor.patch (1.46 KB) 2664-print-Name-dtor.patch		Junxiao Shi, 03/23/2015 09:39 AM
patch_ndnsim_scenario.diff (2.18 KB) patch_ndnsim_scenario.diff		Christian Kreuzberger, 03/24/2015 12:57 AM

Related issues 1 (0 open — 1 closed)

Related to ChronoSync - Bug #2842: "double free" reported when smart pointer is created outside of the wrapper class

Closed

05/28/2015

Actions

Issue # Delay: days Cancel

History
Notes
Property changes

Actions

Copy link

Updated by Junxiao Shi over 9 years ago

Workaround:
At end of simulation scenario's main function, add std::abort().

This does not fix the double free problem, but can suppress the error because __cxa_finalize is not executed if program is aborted.

Actions

Copy link

Updated by Christian Kreuzberger over 9 years ago

I am getting the same error when trying ndn-simple via the scenario template.

Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1".
*** Error in `/home/ckreuz/ndnSIM/my-simulations/build/ndn-simple': double free or corruption (!prev): 0x000000000089ef70 ***

Program received signal SIGABRT, Aborted.
0x00007fffee58acc9 in __GI_raise (sig=sig@entry=6) at ../nptl/sysdeps/unix/sysv/linux/raise.c:56
56  ../nptl/sysdeps/unix/sysv/linux/raise.c: No such file or directory.
(gdb) bt
#0  0x00007fffee58acc9 in __GI_raise (sig=sig@entry=6) at ../nptl/sysdeps/unix/sysv/linux/raise.c:56
#1  0x00007fffee58e0d8 in __GI_abort () at abort.c:89
#2  0x00007fffee5c7394 in __libc_message (do_abort=do_abort@entry=1, fmt=fmt@entry=0x7fffee6d5b28 "*** Error in `%s': %s: 0x%s ***\n") at ../sysdeps/posix/libc_fatal.c:175
#3  0x00007fffee5d366e in malloc_printerr (ptr=<optimized out>, str=0x7fffee6d5c10 "double free or corruption (!prev)", action=1) at malloc.c:4996
#4  _int_free (av=<optimized out>, p=<optimized out>, have_lock=0) at malloc.c:3840
#5  0x00007ffff3bd41bf in deallocate (this=<optimized out>, __p=<optimized out>) at /usr/include/c++/4.8/ext/new_allocator.h:110
#6  _M_deallocate (this=<optimized out>, __n=<optimized out>, __p=<optimized out>) at /usr/include/c++/4.8/bits/stl_vector.h:174
#7  ~_Vector_base (this=0x79abd0 <ndn::nfd::CommandOptions::DEFAULT_PREFIX+80>, __in_chrg=<optimized out>) at /usr/include/c++/4.8/bits/stl_vector.h:160
#8  ~vector (this=0x79abd0 <ndn::nfd::CommandOptions::DEFAULT_PREFIX+80>, __in_chrg=<optimized out>) at /usr/include/c++/4.8/bits/stl_vector.h:416
#9  ndn::Block::~Block (this=0x79ab90 <ndn::nfd::CommandOptions::DEFAULT_PREFIX+16>, __in_chrg=<optimized out>) at /usr/local/include/ndn-cxx/encoding/block.hpp:43
#10 0x00007ffff3bd6bad in ndn::Name::~Name (this=0x79ab80 <ndn::nfd::CommandOptions::DEFAULT_PREFIX>, __in_chrg=<optimized out>) at /usr/local/include/ndn-cxx/name.hpp:39
#11 0x00007fffee5905ea in __cxa_finalize (d=0x7ffff4072360) at cxa_finalize.c:56
#12 0x00007ffff3bc5c83 in __do_global_dtors_aux () from /usr/local/lib/libns3-dev-ndnSIM-optimized.so
#13 0x00007fffffffde10 in ?? ()
#14 0x00007ffff7dea73a in _dl_fini () at dl-fini.c:252

Additionally, here is the output from valgrind for build/hello:
http://pastebin.com/y32vBm1j

It looks like there are several invalid free and delete commands, probably related to Name/Block. Though I don't have time to dig into this problem.

Actions

Copy link

Updated by Junxiao Shi over 9 years ago

File 2664-print-Name-dtor.patch 2664-print-Name-dtor.patch added

The Name being double freed is ndn::nfd::CommandOptions::DEFAULT_PREFIX. I found this out with the following method:

apply 2664-print-Name-dtor.patch to ndn-cxx

run scenario in gdb; I added std::cout << "hello" << std::endl; in the scenario to show where cleanup starts.

(gdb) run
Starting program: /home/shijunxiao/ndnSIM-scenario-template/build/hello 
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1".
Name::~Name(0x7fffffffe2a0 /)
Name::~Name(0x7fffffffdcf0 /)
Name::~Name(0x7fffffffdee0 /)
hello
Name::~Name(0xabbe60 /723821fd-f534-44b3-80d9-44bf5f58bbbb)
Name::~Name(0xabbb40 /localhost/nfd)
hello: ../src/name.cpp:87: ndn::Name::~Name(): Assertion `it != getAllNameSet().end()' failed.

Program received signal SIGABRT, Aborted.
0x00007fffeb4ba0d5 in __GI_raise (sig=<optimized out>) at ../nptl/sysdeps/unix/sysv/linux/raise.c:64
64  ../nptl/sysdeps/unix/sysv/linux/raise.c: No such file or directory.
(gdb) bt
#0  0x00007fffeb4ba0d5 in __GI_raise (sig=<optimized out>) at ../nptl/sysdeps/unix/sysv/linux/raise.c:64
#1  0x00007fffeb4bd83b in __GI_abort () at abort.c:91
#2  0x00007fffeb4b2d9e in __assert_fail_base (fmt=<optimized out>, assertion=0x7d6950 "it != getAllNameSet().end()", 
    file=0x7d6940 "../src/name.cpp", line=<optimized out>, function=<optimized out>) at assert.c:94
#3  0x00007fffeb4b2e42 in __GI___assert_fail (assertion=0x7d6950 "it != getAllNameSet().end()", file=0x7d6940 "../src/name.cpp", 
    line=87, function=0x7d6b10 "ndn::Name::~Name()") at assert.c:103
#4  0x00000000006c7862 in ndn::Name::~Name (this=0xabbb40, __in_chrg=<optimized out>) at ../src/name.cpp:87
#5  0x00007fffeb4bf9cd in __cxa_finalize (d=0x7ffff27df7e0) at cxa_finalize.c:56
#6  0x00007ffff1f21976 in __do_global_dtors_aux () from /home/shijunxiao/local/lib/libns3-dev-ndnSIM-debug.so

Stack frame 4 shows the Name being destructed is 0xabbb40, which has been destructed earlier and its value was /localhost/nfd.

The low address also indicates that it's a static constant.

There are only two Name constants of this value in the code: ndn-cxx/src/management/nfd-command-options.cpp and ndnSIM/model/ndn-l3-protocol.cpp.

Changing the value of ndn::nfd::CommandOptions::DEFAULT_PREFIX to something else, and gdb shows the Name changed, so that I can conclude ndn::nfd::CommandOptions::DEFAULT_PREFIX is the Name being double freed.

This Bug is caused by linking ndn-cxx static library twice: once in libns3-dev-ndnSIM-debug.so and again in the scenario binary.

After those analysis, I have found the solution. Please add me as a Developer of ndnSIM Redmine site, and assign this issue to me. I'll submit the solution afterwards.

Actions

Copy link

Updated by Christian Kreuzberger over 9 years ago

File patch_ndnsim_scenario.diff patch_ndnsim_scenario.diff added

As yous aid, the problem is caused by ndn-cxx STATIC library being linked twice. The fix is not to "fix" ndn::Data (though obviously this could be another problem), but to fix the build scripts.
The changes are simple:
i) (this is preliminary and just fixes some warnings in g++/gcc):

remove -pedantic from .waf-tools/default-compiler-flags.py

ii) add a waf-conf "check" to wscript, that checks that boost is at least at version 1.53 and in addition that boost lib system and iostreams are available

iii) fix waf-build by changing the deps from NDN_CXX + ... to BOOST BOOST_IOSTREAMS + ...

I've done that, and the patch is attached.

Actions

Copy link

Updated by Junxiao Shi over 9 years ago

My solution is much simpler than this. I'm waiting for this Bug to be assigned to me before uploading the solution.

Actions

Copy link

Updated by Junxiao Shi over 9 years ago

Tracker changed from Task to Bug
Category set to build
Status changed from New to In Progress
Assignee set to Junxiao Shi
Estimated time set to 0.50 h

Actions

Copy link

Updated by Junxiao Shi over 9 years ago

Status changed from In Progress to Code review
% Done changed from 0 to 100

https://github.com/cawka/ndnSIM-scenario-template/pull/1

Actions

Copy link

Updated by Junxiao Shi over 9 years ago

The PullRequest was rejected with the following reason:

This is not right. The library needs to be linked, as in some cases it is needed to use ndn-cxx directly from the extensions or scenarios.

I disagree with this reason.

Not linking ndn-cxx in scenario does not prevent the scenario to use ndn-cxx directly.

The compiler can find ndn-cxx headers because pkg-config --cflags libndn-cxx output is appended.
The linker can find ndn-cxx symbols because they appear in libns3*-ndnSIM-*.so.