Fix OOB read in SgiRleDecode.c

* From Pillow 4.3.0->8.1.0
* CVE-2021-25293
This commit is contained in:
Eric Soroos 2021-01-02 16:07:36 +01:00 committed by Andrew Murray
parent 3fee28eb94
commit 4853e522bd
9 changed files with 84 additions and 14 deletions

View File

@ -11,6 +11,13 @@ from PIL import Image
"Tests/images/sgi_crash.bin",
"Tests/images/crash-6b7f2244da6d0ae297ee0754a424213444e92778.sgi",
"Tests/images/ossfuzz-5730089102868480.sgi",
"Tests/images/crash-754d9c7ec485ffb76a90eeaab191ef69a2a3a3cd.sgi",
"Tests/images/crash-465703f71a0f0094873a3e0e82c9f798161171b8.sgi",
"Tests/images/crash-64834657ee604b8797bf99eac6a194c124a9a8ba.sgi",
"Tests/images/crash-abcf1c97b8fe42a6c68f1fb0b978530c98d57ced.sgi",
"Tests/images/crash-b82e64d4f3f76d7465b6af535283029eda211259.sgi",
"Tests/images/crash-c1b2595b8b0b92cc5f38b6635e98e3a119ade807.sgi",
"Tests/images/crash-db8bfa78b19721225425530c5946217720d7df4e.sgi",
],
)
def test_crashes(test_file):

View File

@ -25,12 +25,58 @@ read4B(UINT32 *dest, UINT8 *buf) {
*dest = (UINT32)((buf[0] << 24) | (buf[1] << 16) | (buf[2] << 8) | buf[3]);
}
/*
SgiRleDecoding is done in a single channel row oriented set of RLE chunks.
* The file is arranged as
- SGI Header
- Rle Offset Table
- Rle Length Table
- Scanline Data
* Each RLE atom is c->bpc bytes wide (1 or 2)
* Each RLE Chunk is [specifier atom] [ 1 or n data atoms ]
* Copy Atoms are a byte with the high bit set, and the low 7 are
the number of bytes to copy from the source to the
destination. e.g.
CBBBBBBBB or 0CHLHLHLHLHLHL (B=byte, H/L = Hi low bytes)
* Run atoms do not have the high bit set, and the low 7 bits are
the number of copies of the next atom to copy to the
destination. e.g.:
RB -> BBBBB or RHL -> HLHLHLHLHL
The upshot of this is, there is no way to determine the required
length of the input buffer from reloffset and rlelength without
going through the data at that scan line.
Furthermore, there's no requirement that individual scan lines
pointed to from the rleoffset table are in any sort of order or
used only once, or even disjoint. There's also no requirement that
all of the data in the scan line area of the image file be used
*/
static int
expandrow(UINT8 *dest, UINT8 *src, int n, int z, int xsize) {
expandrow(UINT8 *dest, UINT8 *src, int n, int z, int xsize, UINT8 *end_of_buffer) {
/*
* n here is the number of rlechunks
* z is the number of channels, for calculating the interleave
* offset to go to RGBA style pixels
* xsize is the row width
* end_of_buffer is the address of the end of the input buffer
*/
UINT8 pixel, count;
int x = 0;
for (; n > 0; n--) {
if (src > end_of_buffer) {
return -1;
}
pixel = *src++;
if (n == 1 && pixel != 0) {
return n;
@ -44,12 +90,18 @@ expandrow(UINT8 *dest, UINT8 *src, int n, int z, int xsize) {
}
x += count;
if (pixel & RLE_COPY_FLAG) {
if (src + count > end_of_buffer) {
return -1;
}
while (count--) {
*dest = *src++;
dest += z;
}
} else {
if (src > end_of_buffer) {
return -1;
}
pixel = *src++;
while (count--) {
*dest = pixel;
@ -61,12 +113,14 @@ expandrow(UINT8 *dest, UINT8 *src, int n, int z, int xsize) {
}
static int
expandrow2(UINT8 *dest, const UINT8 *src, int n, int z, int xsize) {
expandrow2(UINT8 *dest, const UINT8 *src, int n, int z, int xsize, UINT8 *end_of_buffer) {
UINT8 pixel, count;
int x = 0;
for (; n > 0; n--) {
if (src + 1 > end_of_buffer) {
return -1;
}
pixel = src[1];
src += 2;
if (n == 1 && pixel != 0) {
@ -81,12 +135,18 @@ expandrow2(UINT8 *dest, const UINT8 *src, int n, int z, int xsize) {
}
x += count;
if (pixel & RLE_COPY_FLAG) {
if (src + 2 * count > end_of_buffer) {
return -1;
}
while (count--) {
memcpy(dest, src, 2);
src += 2;
dest += z * 2;
}
} else {
if (src + 2 > end_of_buffer) {
return -1;
}
while (count--) {
memcpy(dest, src, 2);
dest += z * 2;
@ -132,7 +192,11 @@ ImagingSgiRleDecode(Imaging im, ImagingCodecState state, UINT8 *buf, Py_ssize_t
return -1;
}
_imaging_seek_pyFd(state->fd, SGI_HEADER_SIZE, SEEK_SET);
_imaging_read_pyFd(state->fd, (char *)ptr, c->bufsize);
if (_imaging_read_pyFd(state->fd, (char *)ptr, c->bufsize) != c->bufsize) {
state->errcode = IMAGING_CODEC_UNKNOWN;
return -1;
}
/* decoder initialization */
state->count = 0;
@ -166,20 +230,20 @@ ImagingSgiRleDecode(Imaging im, ImagingCodecState state, UINT8 *buf, Py_ssize_t
read4B(&c->lengthtab[c->tabindex], &ptr[c->bufindex]);
}
state->count += c->tablen * sizeof(UINT32) * 2;
/* read compressed rows */
for (c->rowno = 0; c->rowno < im->ysize; c->rowno++, state->y += state->ystep) {
for (c->channo = 0; c->channo < im->bands; c->channo++) {
c->rleoffset = c->starttab[c->rowno + c->channo * im->ysize];
c->rlelength = c->lengthtab[c->rowno + c->channo * im->ysize];
c->rleoffset -= SGI_HEADER_SIZE;
if (c->rleoffset + c->rlelength > c->bufsize) {
// Check for underflow of rleoffset-SGI_HEADER_SIZE
if (c->rleoffset < SGI_HEADER_SIZE) {
state->errcode = IMAGING_CODEC_OVERRUN;
goto sgi_finish_decode;
}
c->rleoffset -= SGI_HEADER_SIZE;
/* row decompression */
if (c->bpc == 1) {
status = expandrow(
@ -187,14 +251,16 @@ ImagingSgiRleDecode(Imaging im, ImagingCodecState state, UINT8 *buf, Py_ssize_t
&ptr[c->rleoffset],
c->rlelength,
im->bands,
im->xsize);
im->xsize,
&ptr[c->bufsize-1]);
} else {
status = expandrow2(
&state->buffer[c->channo * 2],
&ptr[c->rleoffset],
c->rlelength,
im->bands,
im->xsize);
im->xsize,
&ptr[c->bufsize-1]);
}
if (status == -1) {
state->errcode = IMAGING_CODEC_OVERRUN;
@ -203,15 +269,12 @@ ImagingSgiRleDecode(Imaging im, ImagingCodecState state, UINT8 *buf, Py_ssize_t
goto sgi_finish_decode;
}
state->count += c->rlelength;
}
/* store decompressed data in image */
state->shuffle((UINT8 *)im->image[state->y], state->buffer, im->xsize);
}
c->bufsize++;
sgi_finish_decode:;
free(c->starttab);
@ -221,5 +284,5 @@ sgi_finish_decode:;
state->errcode = err;
return -1;
}
return state->count - c->bufsize;
return 0;
}