dwarves.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
* libdwarves memory leaks in the stealer interface
@ 2016-07-14 19:58 Tavis Ormandy
       [not found] ` <CABD==127dOrixY6KaHKBnycQV_+h1aMXPgT8h67Tkfka6GboGQ-JsoAwUIsXosN+BqQ9rBEUg@public.gmane.org>
  0 siblings, 1 reply; 2+ messages in thread
From: Tavis Ormandy @ 2016-07-14 19:58 UTC (permalink / raw)
  To: dwarves-u79uwXL29TY76Z2rM5mHXA

Hello, I have a fun project that I'm trying to use libdwarves with,
but had a couple of questions.

I'm using the "steal" callback interface to search for a specific cu.
If the cu matches the one I was looking for, I process it and then
return LSK__STOP_LOADING to abort processing this file. If it's not
the cu I wanted, I just return LSK__DELETE.

This seems to do what I want, but leaks memory.

1) LSK__DELETE doesn't seem to clean up the dwarf_cu obstack, I think
this wasn't intended.
2) If die__process_and_recode() fails in finalize_cu_immediately(),
the cu is leaked. This one is definitely a bug :)

I'm not 100% sure if I'm using LSK__STOP_LOADING correctly. It doesn't
cleanup the cu like LSK__DELETE does, and it doesn't call cus__add()
like LSK__KEEPIT does, so was the stealer supposed to clean it up?

If so, how do I clean up the dcu? I think I can pull it out of
cu->priv, but I think you didn't intend to expose that to callbacks?

The patch I'm using is below.

Thanks for libdwarves, it's a really great library!

Tavis.

diff --git a/src/struct/dwarf_loader.c b/src/struct/dwarf_loader.c
index 163cb51..ba715a9 100644
--- a/src/struct/dwarf_loader.c
+++ b/src/struct/dwarf_loader.c
@@ -2169,9 +2169,12 @@ static int finalize_cu_immediately(struct cus
*cus, struct cu *cu,
  int lsk = finalize_cu(cus, cu, dcu, conf);
  switch (lsk) {
  case LSK__DELETE:
+ obstack_free(&dcu->obstack, NULL);
  cu__delete(cu);
  break;
  case LSK__STOP_LOADING:
+ obstack_free(&dcu->obstack, NULL);
+ cu__delete(cu);
  break;
  case LSK__KEEPIT:
  if (!cu->extra_dbg_info)
@@ -2304,8 +2307,11 @@ static int cus__load_module(struct cus *cus,
struct conf_load *conf,
  cu->priv = &dcu;
  cu->dfops = &dwarf__ops;

- if (die__process_and_recode(cu_die, cu) != 0)
+ if (die__process_and_recode(cu_die, cu) != 0) {
+ obstack_free(&dcu.obstack, NULL);
+ cu__delete(cu);
  return DWARF_CB_ABORT;
+ }

  if (finalize_cu_immediately(cus, cu, &dcu, conf)
     == LSK__STOP_LOADING)
--
To unsubscribe from this list: send the line "unsubscribe dwarves" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

^ permalink raw reply related	[flat|nested] 2+ messages in thread

* Re: libdwarves memory leaks in the stealer interface
       [not found] ` <CABD==127dOrixY6KaHKBnycQV_+h1aMXPgT8h67Tkfka6GboGQ-JsoAwUIsXosN+BqQ9rBEUg@public.gmane.org>
@ 2016-07-15  0:04   ` Arnaldo Carvalho de Melo
  0 siblings, 0 replies; 2+ messages in thread
From: Arnaldo Carvalho de Melo @ 2016-07-15  0:04 UTC (permalink / raw)
  To: Tavis Ormandy; +Cc: dwarves-u79uwXL29TY76Z2rM5mHXA

Em Thu, Jul 14, 2016 at 12:58:07PM -0700, Tavis Ormandy escreveu:
> Hello, I have a fun project that I'm trying to use libdwarves with,
> but had a couple of questions.
> 
> I'm using the "steal" callback interface to search for a specific cu.
> If the cu matches the one I was looking for, I process it and then
> return LSK__STOP_LOADING to abort processing this file. If it's not
> the cu I wanted, I just return LSK__DELETE.
> 
> This seems to do what I want, but leaks memory.
> 
> 1) LSK__DELETE doesn't seem to clean up the dwarf_cu obstack, I think
> this wasn't intended.
> 2) If die__process_and_recode() fails in finalize_cu_immediately(),
> the cu is leaked. This one is definitely a bug :)
> 
> I'm not 100% sure if I'm using LSK__STOP_LOADING correctly. It doesn't
> cleanup the cu like LSK__DELETE does, and it doesn't call cus__add()
> like LSK__KEEPIT does, so was the stealer supposed to clean it up?
> 
> If so, how do I clean up the dcu? I think I can pull it out of
> cu->priv, but I think you didn't intend to expose that to callbacks?
> 
> The patch I'm using is below.

At first sight it is ok, i.e. it seems to fix the problem, but I think
the right place to do the free will be at:

  cu__delete()
     cu->dfops->cu__delete(cu);

Which for DWARF, maps to... nothing, I'll continue looking at this
tomorrow, its been a while I haven't touched this code :-\
 
> Thanks for libdwarves, it's a really great library!

Glad you liked it!
 
> Tavis.
> 
> diff --git a/src/struct/dwarf_loader.c b/src/struct/dwarf_loader.c
> index 163cb51..ba715a9 100644
> --- a/src/struct/dwarf_loader.c
> +++ b/src/struct/dwarf_loader.c
> @@ -2169,9 +2169,12 @@ static int finalize_cu_immediately(struct cus
> *cus, struct cu *cu,
>   int lsk = finalize_cu(cus, cu, dcu, conf);
>   switch (lsk) {
>   case LSK__DELETE:
> + obstack_free(&dcu->obstack, NULL);
>   cu__delete(cu);
>   break;
>   case LSK__STOP_LOADING:
> + obstack_free(&dcu->obstack, NULL);
> + cu__delete(cu);
>   break;
>   case LSK__KEEPIT:
>   if (!cu->extra_dbg_info)
> @@ -2304,8 +2307,11 @@ static int cus__load_module(struct cus *cus,
> struct conf_load *conf,
>   cu->priv = &dcu;
>   cu->dfops = &dwarf__ops;
> 
> - if (die__process_and_recode(cu_die, cu) != 0)
> + if (die__process_and_recode(cu_die, cu) != 0) {
> + obstack_free(&dcu.obstack, NULL);
> + cu__delete(cu);
>   return DWARF_CB_ABORT;
> + }
> 
>   if (finalize_cu_immediately(cus, cu, &dcu, conf)
>      == LSK__STOP_LOADING)
> --
> To unsubscribe from this list: send the line "unsubscribe dwarves" in
> the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
--
To unsubscribe from this list: send the line "unsubscribe dwarves" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2016-07-15  0:04 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2016-07-14 19:58 libdwarves memory leaks in the stealer interface Tavis Ormandy
     [not found] ` <CABD==127dOrixY6KaHKBnycQV_+h1aMXPgT8h67Tkfka6GboGQ-JsoAwUIsXosN+BqQ9rBEUg@public.gmane.org>
2016-07-15  0:04   ` Arnaldo Carvalho de Melo

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).