linux/include
Hans Verkuil ac34b79da1 media: v4l2-ctrls: fix reference to freed memory
When controls are used together with the Request API, then for
each request a v4l2_ctrl_handler struct is allocated. This contains
the controls that can be set in a request. If a control is *not* set in
the request, then the value used in the most recent previous request
must be used, or the current value if it is not found in any outstanding
requests.

The framework tried to find such a previous request and it would set
the 'req' pointer in struct v4l2_ctrl_ref to the v4l2_ctrl_ref of the
control in such a previous request. So far, so good. However, when that
previous request was applied to the hardware, returned to userspace, and
then userspace would re-init or free that request, any 'ref' pointer in
still-queued requests would suddenly point to freed memory.

This was not noticed before since the drivers that use this expected
that each request would always have the controls set, so there was
never any need to find a control in older requests. This requirement
was relaxed, and now this bug surfaced.

It was also made worse by changeset
2fae4d6aab ("media: v4l2-ctrls: v4l2_ctrl_request_complete() should always set ref->req")
which increased the chance of this happening.

The use of the 'req' pointer in v4l2_ctrl_ref was very fragile, so
drop this entirely. Instead add a valid_p_req bool to indicate that
p_req contains a valid value for this control. And if it is false,
then just use the current value of the control.

Note that VIDIOC_G_EXT_CTRLS will always return -EACCES when attempting
to get a control from a request until the request is completed. And in
that case, all controls in the request will have the control value set
(i.e. valid_p_req is true). This means that the whole 'find the most
recent previous request containing a control' idea is pointless, and
the code can be simplified considerably.

The v4l2_g_ext_ctrls_common() function was refactored a bit to make
it more understandable. It also avoids updating volatile controls
in a completed request since that was already done when the request
was completed.

Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl>
Fixes: 2fae4d6aab ("media: v4l2-ctrls: v4l2_ctrl_request_complete() should always set ref->req")
Fixes: 6fa6f831f0 ("media: v4l2-ctrls: add core request support")
Cc: <stable@vger.kernel.org>      # for v5.9 and up
Tested-by: Alexandre Courbot <acourbot@chromium.org>
Signed-off-by: Mauro Carvalho Chehab <mchehab+huawei@kernel.org>
2021-04-15 13:18:09 +02:00
..
acpi IOMMU Updates for Linux v5.12 2021-02-22 10:31:29 -08:00
asm-generic RISC-V Patches for the 5.12 Merge Window 2021-02-26 10:28:35 -08:00
clocksource
crypto Keyrings miscellany 2021-02-23 16:09:23 -08:00
drm drm/drm_vblank: set the dma-fence timestamp during send_vblank_event 2021-02-24 21:05:54 +05:30
dt-bindings RISC-V Patches for the 5.12 Merge Window 2021-02-26 10:28:35 -08:00
keys
kunit
kvm
linux dt-bindings:iio:dac: update microchip,mcp4725.yaml reference 2021-04-06 14:45:03 +02:00
math-emu
media media: v4l2-ctrls: fix reference to freed memory 2021-04-15 13:18:09 +02:00
memory
misc
net net: icmp: pass zeroed opts from icmp{,v6}_ndo_send before sending 2021-02-23 11:29:52 -08:00
pcmcia
ras
rdma RDMA/ipoib: Remove racy Subnet Manager sendonly join checks 2021-02-16 14:42:58 -04:00
scsi
soc RISC-V Patches for the 5.12 Merge Window 2021-02-26 10:28:35 -08:00
sound ALSA/ASoC/SOF/SoundWire: fix Kconfig issues 2021-03-02 18:30:07 +01:00
target scsi: target: core: Add cmd length set before cmd complete 2021-02-22 22:21:29 -05:00
trace block-5.12-2021-02-27 2021-02-28 11:23:38 -08:00
uapi media: v4l: Add HDR10 static metadata controls 2021-04-06 14:56:00 +02:00
vdso
video
xen xen/events: add per-xenbus device event statistics and settings 2021-02-23 10:07:11 -06:00