Segfault when closing a db

[RcrdBrt]

On arrmv7h (Raspberry Pi) with archlinux, I get a segmentation fault when executing db.close() or del db.

Steps to reproduce the issue:

import plyvel
db = plyvel.DB('/tmp/testdb/', create_if_missing=True)
db.close() # segfault

Plyvel version --> 0.9-4 LevelDB version --> 1.20-1

[wbolster]

without a stack trace this is very hard to debug. please try running via gdb and type bt after the segfault.

[RcrdBrt]

Here's the trace.

segfault_leveldb.py content is just the 3 lines in the first post.

pi@rpi3 ~ $ gdb -ex r --args python segfault_leveldb.py 
GNU gdb (GDB) 8.0
[...]
This GDB was configured as "armv7l-unknown-linux-gnueabihf".
[...]
Program received signal SIGSEGV, Segmentation fault.
0x766e9890 in ?? () from /usr/lib/python3.6/site-packages/plyvel/_plyvel.cpython-36m-arm-linux-gnueabihf.so
(gdb) bt
#0  0x766e9890 in ?? () from /usr/lib/python3.6/site-packages/plyvel/_plyvel.cpython-36m-arm-linux-gnueabihf.so
#1  0x766ea4ec in ?? () from /usr/lib/python3.6/site-packages/plyvel/_plyvel.cpython-36m-arm-linux-gnueabihf.so
#2  0x76d520f8 in _PyCFunction_FastCallDict () from /usr/lib/libpython3.6m.so.1.0
#3  0x76e1f3f8 in ?? () from /usr/lib/libpython3.6m.so.1.0
Backtrace stopped: previous frame identical to this frame (corrupt stack?)
(gdb)

[wbolster]

any chance you can write a small c++ program that does the same as the python code above and try that? i cannot reproduce this issue with python (3.6 and other versions), but i only have access to linux amd64 right now.

[RcrdBrt]

segfault_leveldb.cpp content:

#include "leveldb/db.h"
#include "leveldb/options.h"

int main() {
        leveldb::DB* db;
        leveldb::Options options;
        options.create_if_missing = true;
        leveldb::Status status = leveldb::DB::Open(options, "/tmp/testdb", &db);
        delete db;
}

compilation directives:

• g++ -c segfault_leveldb.cpp -o segfault_leveldb.o
• g++ -pthread segfault_leveldb.o /usr/lib/libleveldb.so -o segfault_leveldb

No segfault happened althought in both the python and this latest C++ program there is some noisy valgrind output:

pi@rpi3 ~/segfault_leveldb $ valgrind ./segfault_leveldb --leak-check=full --track-origins=yes 
==17857== Memcheck, a memory error detector
==17857== Copyright (C) 2002-2017, and GNU GPL'd, by Julian Seward et al.
==17857== Using Valgrind-3.13.0 and LibVEX; rerun with -h for copyright info
==17857== Command: ./segfault_leveldb --leak-check=full --track-origins=yes
==17857== 
==17857== Conditional jump or move depends on uninitialised value(s)
==17857==    at 0x401A254: index (in /usr/lib/ld-2.25.so)
==17857==    by 0x40082CF: expand_dynamic_string_token (in /usr/lib/ld-2.25.so)
==17857== 
==17857== Conditional jump or move depends on uninitialised value(s)
==17857==    at 0x401A258: index (in /usr/lib/ld-2.25.so)
==17857==    by 0x40082CF: expand_dynamic_string_token (in /usr/lib/ld-2.25.so)
==17857== 
==17857== Conditional jump or move depends on uninitialised value(s)
==17857==    at 0x40082D4: expand_dynamic_string_token (in /usr/lib/ld-2.25.so)
==17857== 
==17857== Invalid write of size 4
==17857==    at 0x108AA4: main (in /home/pi/segfault_leveldb/segfault_leveldb)
==17857==  Address 0x7d8e828c is on thread 1's stack
==17857==  12400 bytes below stack pointer
==17857== 
==17857== Invalid write of size 4
==17857==    at 0x108BC8: leveldb::Status::~Status() (in /home/pi/segfault_leveldb/segfault_leveldb)
==17857==  Address 0x7d8e8280 is on thread 1's stack
==17857==  12304 bytes below stack pointer
==17857== 
==17857== Invalid write of size 4
==17857==    at 0x4C46F4C: ??? (in /usr/lib/libsnappy.so.1.3.1)
==17857==  Address 0x7d8ea248 is on thread 1's stack
==17857==  4104 bytes below stack pointer
==17857== 
==17857== 
==17857== HEAP SUMMARY:
==17857==     in use at exit: 816 bytes in 7 blocks
==17857==   total heap usage: 157 allocs, 150 frees, 206,865 bytes allocated
==17857== 
==17857== LEAK SUMMARY:
==17857==    definitely lost: 4 bytes in 1 blocks
==17857==    indirectly lost: 0 bytes in 0 blocks
==17857==      possibly lost: 0 bytes in 0 blocks
==17857==    still reachable: 812 bytes in 6 blocks
==17857==         suppressed: 0 bytes in 0 blocks
==17857== Rerun with --leak-check=full to see details of leaked memory
==17857== 
==17857== For counts of detected and suppressed errors, rerun with: -v
==17857== Use --track-origins=yes to see where uninitialised values come from
==17857== ERROR SUMMARY: 9 errors from 6 contexts (suppressed: 0 from 0)

this looks like an out of boundaries pointer.

[wbolster]

ok so this might have been caused by a bug in leveldb... unfortunately there's not much plyvel can do about that.

[hqmonaro]

I had the same segfault problem recently on x86_64 when using the Arch community/python-plyvel 0.9-4 and community/python2-plyvel 0.9-4 with both Python 2 & 3 against leveldb 1.20. I removed them and installed via pip and pip2 and both work fine now. I expect the Arch community packages have been compiled against a different version of the leveldb libs.