In the Linux kernel, the following vulnerability has been resolved:
media: mediatek: vcodec: bound AV1 tile-start copy to the array capacity
vdec_av1_slice_setup_tile() copies tile_cols + 1 / tile_rows + 1 entries
into mi_col_starts[] / mi_row_starts[] from the bitstream tile_info. Bound
the copy to the array capacity.
CVSS Vector: CVSS:3.1/AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H
CVSS Score: 7.8
AV:L - The bad tile_info.tile_cols/tile_rows reach vdec_av1_slice_setup_tile() from a local process that sets V4L2_CID_STATELESS_AV1_FRAME via VIDIOC_S_EXT_CTRLS on the mtk-vcodec decoder /dev/videoN and queues a request. A spec-conforming AV1 parser caps tile counts at 64, so a remote bitstream does not produce the malformed values.
AC:L - The attacker sets tile_cols or tile_rows to any value from 65 to 255. validate_av1_frame() never checks them, and the driver's num_tiles check in vdec_av1_slice_setup_tile_group() runs after vdec_av1_slice_setup_pfc() has already done the copy, so the overflow happens on every such request.
PR:L - It needs only the ability to open the decoder video node, which is normally granted to the video group or to media/GPU processes. No capability check lies on the ioctl or request-queue path to vdec_av1_slice_lat_decode().
UI:N - The attacker supplies the control payload and queues the request themselves. No other user has to do anything.
S:U - This is a heap overflow in kernel memory. It gives kernel-level impact within the same security authority, not a VM or IOMMU boundary crossing.
C:H - Writing up to 764 bytes past mi_row_starts[] runs off the end of the kzalloc'd vdec_av1_slice_pfc into the neighbouring slab object. That corruption can be used to build read primitives and leak kernel memory.
I:H - The written values are ALIGN(ctrl_tile->mi_*_starts[i]) >> mib_size_log2, taken from the attacker's own control payload. This gives an attacker-shaped heap out-of-bounds write that can corrupt neighbouring objects.
A:H - Corrupting the pfc fields past the tile arrays (frame, state, ref_idx) and the neighbouring slab object can oops the kernel or hang the decoder, and any local user with device access can repeat it.
| Attack Vector |
Local |
Scope |
Unchanged |
| Attack Complexity |
Low |
Confidentiality Impact |
High |
| Privileges Required |
Low |
Integrity Impact |
High |
| User Interaction |
None |
Availability Impact |
High |
AV:L - The bad tile_info.tile_cols/tile_rows reach vdec_av1_slice_setup_tile() from a local process that sets V4L2_CID_STATELESS_AV1_FRAME via VIDIOC_S_EXT_CTRLS on the mtk-vcodec decoder /dev/videoN and queues a request. A spec-conforming AV1 parser caps tile counts at 64, so a remote bitstream does not produce the malformed values.
AC:L - The attacker sets tile_cols or tile_rows to any value from 65 to 255. validate_av1_frame() never checks them, and the driver's num_tiles check in vdec_av1_slice_setup_tile_group() runs after vdec_av1_slice_setup_pfc() has already done the copy, so the overflow happens on every such request.
PR:L - It needs only the ability to open the decoder video node, which is normally granted to the video group or to media/GPU processes. No capability check lies on the ioctl or request-queue path to vdec_av1_slice_lat_decode().
UI:N - The attacker supplies the control payload and queues the request themselves. No other user has to do anything.
S:U - This is a heap overflow in kernel memory. It gives kernel-level impact within the same security authority, not a VM or IOMMU boundary crossing.
C:H - Writing up to 764 bytes past mi_row_starts[] runs off the end of the kzalloc'd vdec_av1_slice_pfc into the neighbouring slab object. That corruption can be used to build read primitives and leak kernel memory.
I:H - The written values are ALIGN(ctrl_tile->mi_*_starts[i]) >> mib_size_log2, taken from the attacker's own control payload. This gives an attacker-shaped heap out-of-bounds write that can corrupt neighbouring objects.
A:H - Corrupting the pfc fields past the tile arrays (frame, state, ref_idx) and the neighbouring slab object can oops the kernel or hang the decoder, and any local user with device access can repeat it.
CVSS 3.1