[RFA] Fix gdb/15827 (crash w/corrupt DWARF)

Message ID 532C6D4D.2050705@redhat.com
State Committed
Headers

Commit Message

Keith Seitz March 21, 2014, 4:48 p.m. UTC
  Hi,

I think most of us know that the dwarf2read.c pretty much assumes that 
the DWARF input is okay. The reporter of this bug used a binary fuzzer 
to corrupt the DWARF info:

<1><569>: Abbrev Number: 16 (DW_TAG_enumeration_type)
     <56a>   DW_AT_byte_size   : 4
     <56b>   DW_AT_decl_file   : 17
     <56c>   DW_AT_decl_line   : 73
     <56d>   DW_AT_sibling     : <0xb12>
  <2><571>: Abbrev Number: 17 (DW_TAG_enumerator)
     <572>   DW_AT_name        : (indirect string, offset: 0xf14): 
_SC_ARG_MAX
     <576>   DW_AT_const_value : 0

[snip]

<2><a8c>: Abbrev Number: 17 (DW_TAG_enumerator)
     <a8d>   DW_AT_name        : (indirect string, offset: 0x9b7): 
_SC_LEVEL3_CACHE_ASSOC
     <a91>   DW_AT_const_value : 2243
  <2><a93>: Abbrev Number: 12 (DW_TAG_array_type)
     <a94>   DW_AT_type        : <0x114>
     <a98>   DW_AT_sibling     : <0x761101c4>
  <3><a9c>: Abbrev Number: 10 (DW_TAG_member)
     <a9d>   DW_AT_name        : (indirect string, offset: 0x1c50000): 
<offset is
  too big>
     <aa1>   DW_AT_decl_file   : 17
     <aa2>   DW_AT_decl_line   : 2539
     <aa4>   DW_AT_type        : <0x1c60000>
     <aa8>   DW_AT_data_member_location: 17

As you can see, the sibling for DIE 0xa93 points of to la-la land. This 
causes skip_one_die to crash, since it never validates whether the 
offset of the sibling is contained within the current readers's input 
buffer. This function does check, however, whether the offset is negative.

This patch essentially adds the counterpoint check and a beginning at 
some tests for catch-all "corrupt" DWARF.

Keith

ChangeLog
2014-03-20  Keith Seitz  <keiths@redhat.com>

	PR gdb/15827
	* dwarf2read.c (skip_one_die): Check that all relative-offset
	sibling DIEs fall within range of the current reader's buffer.
	(read_partial_die): Likewise.

testsuite/ChangeLog
2014-03-20  Keith Seitz  <keiths@redhat.com>

	PR gdb/15827
	* gdb.dwarf2/corrupt.c: New file.
	* gdb.dwarf2/corrupt.exp: New file.
  

Comments

Joel Brobecker March 21, 2014, 5:14 p.m. UTC | #1
Hi Keith,

> ChangeLog
> 2014-03-20  Keith Seitz  <keiths@redhat.com>
> 
> 	PR gdb/15827
> 	* dwarf2read.c (skip_one_die): Check that all relative-offset
> 	sibling DIEs fall within range of the current reader's buffer.
> 	(read_partial_die): Likewise.
>
> testsuite/ChangeLog
> 2014-03-20  Keith Seitz  <keiths@redhat.com>
> 
> 	PR gdb/15827
> 	* gdb.dwarf2/corrupt.c: New file.
> 	* gdb.dwarf2/corrupt.exp: New file.


LGTM! Just one tiny nit...

> +/* Dummy main function.  */
> +
> +int
> +main()

Use "(void)" instead of "()". There was a recent policy clarification
regarding the CS to be using with testcases, and basically we decided
to try to follow the GCS as much as we reasonably could.

> +}
> +
> +# If we get here and gdb hasn't crashed, the tests pass.
> +pass "corrupt DWARF"

That's just me but I usually do a "print 1" test, just to make sure
that even if the testing framework did not detect the GDB process
dying, the "print 1" test definitely will. Not important on most,
if not all platforms, but ISTR some odd platforms where this helped.
That's just a suggestion, you don't have to follow it.
  
Doug Evans March 21, 2014, 7:27 p.m. UTC | #2
On Fri, Mar 21, 2014 at 10:14 AM, Joel Brobecker <brobecker@adacore.com> wrote:
>> +# If we get here and gdb hasn't crashed, the tests pass.
>> +pass "corrupt DWARF"
>
> That's just me but I usually do a "print 1" test, just to make sure
> that even if the testing framework did not detect the GDB process
> dying, the "print 1" test definitely will. Not important on most,
> if not all platforms, but ISTR some odd platforms where this helped.
> That's just a suggestion, you don't have to follow it.

Not that this needs to be added to this patch, but IWBN to codify this
in a utility routine, some tcl proc that does "print 1" or whatever
and verifies the expected result.  The point is to encapsulate the
intent in a suitably named function, verify_gdb_alive or some such.
  

Patch

diff --git a/gdb/dwarf2read.c b/gdb/dwarf2read.c
index 705dc2d..c30b1b3 100644
--- a/gdb/dwarf2read.c
+++ b/gdb/dwarf2read.c
@@ -7103,6 +7103,8 @@  skip_one_die (const struct die_reader_specs *reader, const gdb_byte *info_ptr,
 	      if (sibling_ptr < info_ptr)
 		complaint (&symfile_complaints,
 			   _("DW_AT_sibling points backwards"));
+	      else if (sibling_ptr > reader->buffer_end)
+		dwarf2_section_buffer_overflow_complaint (reader->die_section);
 	      else
 		return sibling_ptr;
 	    }
@@ -15416,6 +15418,8 @@  read_partial_die (const struct die_reader_specs *reader,
 	      if (sibling_ptr < info_ptr)
 		complaint (&symfile_complaints,
 			   _("DW_AT_sibling points backwards"));
+	      else if (sibling_ptr > reader->buffer_end)
+		dwarf2_section_buffer_overflow_complaint (reader->die_section);
 	      else
 		part_die->sibling = sibling_ptr;
 	    }
diff --git a/gdb/testsuite/gdb.dwarf2/corrupt.c b/gdb/testsuite/gdb.dwarf2/corrupt.c
new file mode 100644
index 0000000..4120f3d
--- /dev/null
+++ b/gdb/testsuite/gdb.dwarf2/corrupt.c
@@ -0,0 +1,24 @@ 
+/* This testcase is part of GDB, the GNU debugger.
+
+   Copyright 2014 Free Software Foundation, Inc.
+
+   This program is free software; you can redistribute it and/or modify
+   it under the terms of the GNU General Public License as published by
+   the Free Software Foundation; either version 3 of the License, or
+   (at your option) any later version.
+
+   This program is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+   GNU General Public License for more details.
+
+   You should have received a copy of the GNU General Public License
+   along with this program.  If not, see <http://www.gnu.org/licenses/>.  */
+
+/* Dummy main function.  */
+
+int
+main()
+{
+  return 0;
+}
diff --git a/gdb/testsuite/gdb.dwarf2/corrupt.exp b/gdb/testsuite/gdb.dwarf2/corrupt.exp
new file mode 100644
index 0000000..ba5e551
--- /dev/null
+++ b/gdb/testsuite/gdb.dwarf2/corrupt.exp
@@ -0,0 +1,78 @@ 
+# Copyright 2014 Free Software Foundation, Inc.
+
+# This program is free software; you can redistribute it and/or modify
+# it under the terms of the GNU General Public License as published by
+# the Free Software Foundation; either version 3 of the License, or
+# (at your option) any later version.
+#
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU General Public License
+# along with this program.  If not, see <http://www.gnu.org/licenses/>.
+
+# Test corrupt DWARF input
+# PR gdb/15827
+
+load_lib dwarf.exp
+
+if {![dwarf2_support]} {
+    return 0
+}
+
+standard_testfile corrupt.c corrupt.S
+
+# Make the DWARF used for the test.
+#
+# Here we put DW_AT_sibling DIEs into the output which
+# point off into la-la land.  The whole purpose is to simulate
+# corrupt DWARF information and make sure that GDB can handle it
+# without crashing.
+
+set asm_file [standard_output_file $srcfile2]
+Dwarf::assemble $asm_file {
+    cu {} {
+	compile_unit {} {
+	    declare_labels int_label
+
+	    int_label: base_type {
+		{byte_size 4}
+		{name "int"}
+	    }
+
+	    enumeration_type {
+		{name "ENUM"}
+		{byte_size 4}
+	    } {
+		enumerator {
+		    {name "A"}
+		    {const_value 0}
+		}
+		enumerator {
+		    {name "B"}
+		    {const_value 1}
+		    {sibling 12345678 DW_FORM_ref4}
+		} {
+		    base_type {
+			{byte_size 1}
+			{name "char"}
+		    }
+		}
+		array_type {
+		    {type :$int_label}
+		    {sibling 12345678 DW_FORM_ref4}
+		}
+	    }
+	}
+    }
+}
+
+if {[prepare_for_testing $testfile.exp $testfile \
+	 [list $srcfile $asm_file] {nodebug}]} {
+    return -1
+}
+
+# If we get here and gdb hasn't crashed, the tests pass.
+pass "corrupt DWARF"