RFA: libiberty: Add a limit on demangling qualifiers (PR 87241)

Message ID 87pnu7j85i.fsf@redhat.com
State Superseded
Headers show
Series
  • RFA: libiberty: Add a limit on demangling qualifiers (PR 87241)
Related show

Commit Message

Nick Clifton Dec. 12, 2018, 11:28 a.m.
Hi Ian,

  Sorry to bother you, but I have another libiberty demangler resource
  exhaustion prevention patch to present.  This one is for:

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=87241

  Jonathan Wakely reported that __cxa_demanlge() was returning a -2
  result, but I did not see this.  Instead I found that
  consume_count_with_underscores() is returning a very large number
  (because a very large value is encoded in the mangled string) and this
  is resulting in many calls to remember_Ktype() which eventually
  exhaust the amount of memory available.

  The attached patch is a simplistic approach to solving this problem by
  adding a hard upper limit on the number of qualifiers that will be
  allowed by the demangler.  I am not sure if this is the best approach
  to solving the problem, but it is a simple one, and I would think one
  that would not prevent the demangling of any real mangled names.  The
  limit does not have to be DEMANGLE_RECURSE_LIMIT of course.  I just
  chose that value because it was convenient and of a size that I
  thought was appropriate.

  I also did run the libiberty testsuite this time, with no failures
  reported. :-)

  OK to apply ?

Cheers
  Nick

libiberty/ChangeLog
2018-12-12  Nick Clifton  <nickc@redhat.com>

	* cplus-dem.c (demangle_qualified): Add an upper limit on the
	number of qualifiers supported, based upon the value of
	DEMANGLE_RECURSE_LIMIT.

Comments

Jason Merrill Dec. 12, 2018, 8:30 p.m. | #1
On Wed, Dec 12, 2018 at 6:29 AM Nick Clifton <nickc@redhat.com> wrote:
>

>   Sorry to bother you, but I have another libiberty demangler resource

>   exhaustion prevention patch to present.  This one is for:

>

> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=87241

>

>   Jonathan Wakely reported that __cxa_demanlge() was returning a -2

>   result, but I did not see this.  Instead I found that

>   consume_count_with_underscores() is returning a very large number

>   (because a very large value is encoded in the mangled string) and this

>   is resulting in many calls to remember_Ktype() which eventually

>   exhaust the amount of memory available.

>

>   The attached patch is a simplistic approach to solving this problem by

>   adding a hard upper limit on the number of qualifiers that will be

>   allowed by the demangler.  I am not sure if this is the best approach

>   to solving the problem, but it is a simple one, and I would think one

>   that would not prevent the demangling of any real mangled names.  The

>   limit does not have to be DEMANGLE_RECURSE_LIMIT of course.  I just

>   chose that value because it was convenient and of a size that I

>   thought was appropriate.

>

>   I also did run the libiberty testsuite this time, with no failures

>   reported. :-)

>

>   OK to apply ?

>

> Cheers

>   Nick

>

> libiberty/ChangeLog

> 2018-12-12  Nick Clifton  <nickc@redhat.com>

>

>         * cplus-dem.c (demangle_qualified): Add an upper limit on the

>         number of qualifiers supported, based upon the value of

>         DEMANGLE_RECURSE_LIMIT.


This issue also will be resolved by disabling or removing the old
demangling code, which I haven't seen anyone argue against.

Jason
Nick Clifton Dec. 13, 2018, 9:47 a.m. | #2
Hi Jason,

> This issue also will be resolved by disabling or removing the old

> demangling code, which I haven't seen anyone argue against.


Doh - of course.  I withdraw my patch and I hope that yours will go in soon.

Cheers
  Nick

Patch

Index: libiberty/cplus-dem.c
===================================================================
--- libiberty/cplus-dem.c	(revision 267043)
+++ libiberty/cplus-dem.c	(working copy)
@@ -3443,6 +3443,17 @@ 
       success = 0;
     }
 
+  /* PR 87241: Catch malicious input that will try to trick this code into
+     allocating a ridiculous amount of memory via the remember_Ktype()
+     function.
+     The choice of DEMANGLE_RECURSION_LIMIT is somewhat arbitrary.  Possibly
+     a better solution would be to track how much memory remember_Ktype
+     allocates and abort when some upper limit is reached.  */
+  if (qualifiers > DEMANGLE_RECURSION_LIMIT)
+    /* FIXME: We ought to have some way to tell the user that
+       this limit has been reached.  */
+    success = 0;
+
   if (!success)
     return success;