[erlang-bugs] Memory leakage regarding enif_alloc_env() and iolists

[Sverker Eriksson]

Yes, you are right. The enif_inspect_io_list_as_binary() does not work
well with process independent environments. The same goes for
enif_inspect_binary(), but only if the input binary is not byte aligned.

I quick and dirty fix on top of your patch could be to replace the first
four occurrences of ERTS_ALC_T_TMP with ERTS_ALC_T_NIF
in erl_nif.c.

I'll write a ticket for R15.

Thanks for reporting
/Sverker, Erlang/OTP Ericsson

Masatake Daimon wrote:
> Hello,
>
> I found a memory leakage regarding enif_alloc_env() and iolists. The
> function enif_inspect_iolist_as_binary() (and sometimes
> enif_inspect_binary()) allocates a temporary object to be retained by
> ((ErlNifEnv*)env)->tmp_obj_list. This is OK for ordinary
> process-dependent environments as such objects will later be
> deallocated by either erts_post_nif() or post_nif_noproc(). But for
> process-independent ones allocated by enif_alloc_env(), neither
> enif_clear_env() nor enif_free_env() calls free_tmp_objs() so
> temporary objects will never be freed, hence the memory leak.
>
> So I attempted to fix the problem by applying the attached patch. The
> patch seemingly cleared away the leakage but unfortunately it induced
> the runtime system to randomly crash on the call of enif_free_env():
>
>> Program terminated with signal 11, Segmentation fault.
>> #0  0x000000000044275a in destroy_carrier (allctr=0x791440, 
>> blk=0x2b1776d98030)
>>     at beam/erl_alloc_util.c:636
>> 636     beam/erl_alloc_util.c: No such file or directory.
>>         in beam/erl_alloc_util.c
>> (gdb) bt
>> #0  0x000000000044275a in destroy_carrier (allctr=0x791440, 
>> blk=0x2b1776d98030)
>>     at beam/erl_alloc_util.c:636
>> #1  0x0000000000443d67 in erts_alcu_free_thr_spec (type=<value 
>> optimized out>,
>>     extra=<value optimized out>, p=0x2b1776d98038)
>>     at beam/erl_alloc_util.c:2898
>> #2  0x000000000051a6cd in enif_free_env (env=0x823c78) at 
>> beam/erl_nif.c:145
>> << snip >>
>> #7  0x000000000052b186 in process_main () at beam/beam_emu.c:3321
>> #8  0x000000000049d5db in sched_thread_func (vesdp=<value optimized 
>> out>)
>>     at beam/erl_process.c:3782
>> #9  0x000000000059f589 in thr_wrapper (vtwd=<value optimized out>)
>>     at pthread/ethread.c:106
>> #10 0x00002b1777345143 in start_thread () from /lib64/libpthread.so.0
>> #11 0x00002b177761e52d in clone () from /lib64/libc.so.6
>> #12 0x0000000000000000 in ?? ()
>
> The Erlang/OTP version I'm using is R14B04 while I highly suspect that
> the leakage still remains in the Git repository's master branch. I'd
> be glad if someone familiar with the internals of beam finds a proper
> way to fix the problem.
>
> Regards,
> Masatake Daimon
> ------------------------------------------------------------------------
>
> _______________________________________________
> erlang-bugs mailing list
> erlang-bugs@REDACTED
> http://erlang.org/mailman/listinfo/erlang-bugs
>