current/SOURCES/ghostscript-9.20-cve-2016-10220.patch

From daf85701dab05f17e924a48a81edc9195b4a04e8 Mon Sep 17 00:00:00 2001
From: Ken Sharp <ken.sharp@artifex.com>
Date: Wed, 21 Dec 2016 16:54:14 +0000
Subject: [PATCH] fix crash with bad data supplied to makeimagedevice

Bug #697450 "Null pointer dereference in gx_device_finalize()"

The problem here is that the code to finalise a device unconditionally
frees the icc_struct member of the device structure. However this
particular (weird) device is not setup as a normal device, probably
because its very, very ancient. Its possible for the initialisation
of the device to abort with an error before calling gs_make_mem_device()
which is where the icc_struct member gets allocated (or set to NULL).

If that happens, then the cleanup code tries to free the device, which
calls finalize() which tries to free a garbage pointer.

Setting the device memory to 0x00 after we allocate it means that the
icc_struct member will be NULL< and our memory manager allows for that
happily enough, which avoids the problem.
---
 base/gsdevmem.c | 12 ++++++++++++
 1 file changed, 12 insertions(+)

diff --git a/base/gsdevmem.c b/base/gsdevmem.c
index 97b9cf4..fe75bcc 100644
--- a/base/gsdevmem.c
+++ b/base/gsdevmem.c
@@ -225,6 +225,18 @@ gs_makewordimagedevice(gx_device ** pnew_dev, const gs_matrix * pmat,
 
     if (pnew == 0)
         return_error(gs_error_VMerror);
+
+    /* Bug #697450 "Null pointer dereference in gx_device_finalize()"
+     * If we have incorrect data passed to gs_initialise_wordimagedevice() then the
+     * initialisation will fail, crucially it will fail *before* it calls
+     * gs_make_mem_device() which initialises the device. This means that the
+     * icc_struct member will be uninitialsed, but the device finalise method
+     * will unconditionally free that memory. Since its a garbage pointer, bad things happen.
+     * Apparently we do still need makeimagedevice to be available from
+     * PostScript, so in here just zero the device memory, which means that
+     * the finalise routine won't have a problem.
+     */
+    memset(pnew, 0x00, st_device_memory.ssize);
     code = gs_initialize_wordimagedevice(pnew, pmat, width, height,
                                          colors, num_colors, word_oriented,
                                          page_device, mem);
-- 
2.9.3

1	From daf85701dab05f17e924a48a81edc9195b4a04e8 Mon Sep 17 00:00:00 2001
2	From: Ken Sharp <ken.sharp@artifex.com>
3	Date: Wed, 21 Dec 2016 16:54:14 +0000
4	Subject: [PATCH] fix crash with bad data supplied to makeimagedevice
5
6	Bug #697450 "Null pointer dereference in gx_device_finalize()"
7
8	The problem here is that the code to finalise a device unconditionally
9	frees the icc_struct member of the device structure. However this
10	particular (weird) device is not setup as a normal device, probably
11	because its very, very ancient. Its possible for the initialisation
12	of the device to abort with an error before calling gs_make_mem_device()
13	which is where the icc_struct member gets allocated (or set to NULL).
14
15	If that happens, then the cleanup code tries to free the device, which
16	calls finalize() which tries to free a garbage pointer.
17
18	Setting the device memory to 0x00 after we allocate it means that the
19	icc_struct member will be NULL< and our memory manager allows for that
20	happily enough, which avoids the problem.
21	---
22	base/gsdevmem.c \| 12 ++++++++++++
23	1 file changed, 12 insertions(+)
24
25	diff --git a/base/gsdevmem.c b/base/gsdevmem.c
26	index 97b9cf4..fe75bcc 100644
27	--- a/base/gsdevmem.c
28	+++ b/base/gsdevmem.c
29	@@ -225,6 +225,18 @@ gs_makewordimagedevice(gx_device ** pnew_dev, const gs_matrix * pmat,
30
31	if (pnew == 0)
32	return_error(gs_error_VMerror);
33	+
34	+ /* Bug #697450 "Null pointer dereference in gx_device_finalize()"
35	+ * If we have incorrect data passed to gs_initialise_wordimagedevice() then the
36	+ * initialisation will fail, crucially it will fail before it calls
37	+ * gs_make_mem_device() which initialises the device. This means that the
38	+ * icc_struct member will be uninitialsed, but the device finalise method
39	+ * will unconditionally free that memory. Since its a garbage pointer, bad things happen.
40	+ * Apparently we do still need makeimagedevice to be available from
41	+ * PostScript, so in here just zero the device memory, which means that
42	+ * the finalise routine won't have a problem.
43	+ */
44	+ memset(pnew, 0x00, st_device_memory.ssize);
45	code = gs_initialize_wordimagedevice(pnew, pmat, width, height,
46	colors, num_colors, word_oriented,
47	page_device, mem);
48	--
49	2.9.3
50