On 21/03/2022 12:53, Laurent Pinchart wrote:
> Hi Hans,
>
> Thank you for the patch.
>
> On Mon, Mar 21, 2022 at 12:03:30PM +0100, Hans Verkuil wrote:
>> If allocating array_buf fails, or copying data from userspace into that
>> buffer fails, then just free memory and return the error. Don't attempt
>> to call video_put_user() since there is no point.
>>
>> If writing the array back to userspace fails, then don't go to out_array_args,
>> instead just continue with the regular code that just returns the error
>> unless 'always_copy' is set. That flag was just completely ignored when
>> writing back the array data.
>>
>> Update the VIDIOC_G/S/TRY_EXT_CTRLS ioctls to set the ALWAYS_COPY flag
>> since they now need it.
>>
>> Signed-off-by: Hans Verkuil <hverkuil-cisco@xxxxxxxxx>
>> ---
>> Changes since v1:
>> - drop the now unused out_array_args label
>> - set the INFO_FL_ALWAYS_COPY for the VIDIOC_G/S/TRY_EXT_CTRLS ioctls
>> ---
>> drivers/media/v4l2-core/v4l2-ioctl.c | 12 +++++-------
>> 1 file changed, 5 insertions(+), 7 deletions(-)
>>
>> diff --git a/drivers/media/v4l2-core/v4l2-ioctl.c b/drivers/media/v4l2-core/v4l2-ioctl.c
>> index 96e307fe3aab..3eb8f5f0958d 100644
>> --- a/drivers/media/v4l2-core/v4l2-ioctl.c
>> +++ b/drivers/media/v4l2-core/v4l2-ioctl.c
>> @@ -2816,9 +2816,9 @@ static const struct v4l2_ioctl_info v4l2_ioctls[] = {
>> IOCTL_INFO(VIDIOC_S_PRIORITY, v4l_s_priority, v4l_print_u32, INFO_FL_PRIO),
>> IOCTL_INFO(VIDIOC_G_SLICED_VBI_CAP, v4l_g_sliced_vbi_cap, v4l_print_sliced_vbi_cap, INFO_FL_CLEAR(v4l2_sliced_vbi_cap, type)),
>> IOCTL_INFO(VIDIOC_LOG_STATUS, v4l_log_status, v4l_print_newline, 0),
>> - IOCTL_INFO(VIDIOC_G_EXT_CTRLS, v4l_g_ext_ctrls, v4l_print_ext_controls, INFO_FL_CTRL),
>> - IOCTL_INFO(VIDIOC_S_EXT_CTRLS, v4l_s_ext_ctrls, v4l_print_ext_controls, INFO_FL_PRIO | INFO_FL_CTRL),
>> - IOCTL_INFO(VIDIOC_TRY_EXT_CTRLS, v4l_try_ext_ctrls, v4l_print_ext_controls, INFO_FL_CTRL),
>> + IOCTL_INFO(VIDIOC_G_EXT_CTRLS, v4l_g_ext_ctrls, v4l_print_ext_controls, INFO_FL_CTRL | INFO_FL_ALWAYS_COPY),
>> + IOCTL_INFO(VIDIOC_S_EXT_CTRLS, v4l_s_ext_ctrls, v4l_print_ext_controls, INFO_FL_PRIO | INFO_FL_CTRL | INFO_FL_ALWAYS_COPY),
>> + IOCTL_INFO(VIDIOC_TRY_EXT_CTRLS, v4l_try_ext_ctrls, v4l_print_ext_controls, INFO_FL_CTRL | INFO_FL_ALWAYS_COPY),
>> IOCTL_INFO(VIDIOC_ENUM_FRAMESIZES, v4l_stub_enum_framesizes, v4l_print_frmsizeenum, INFO_FL_CLEAR(v4l2_frmsizeenum, pixel_format)),
>> IOCTL_INFO(VIDIOC_ENUM_FRAMEINTERVALS, v4l_stub_enum_frameintervals, v4l_print_frmivalenum, INFO_FL_CLEAR(v4l2_frmivalenum, height)),
>> IOCTL_INFO(VIDIOC_G_ENC_INDEX, v4l_stub_g_enc_index, v4l_print_enc_idx, 0),
>> @@ -3311,7 +3311,7 @@ video_usercopy(struct file *file, unsigned int orig_cmd, unsigned long arg,
>> array_buf = kvmalloc(array_size, GFP_KERNEL);
>> err = -ENOMEM;
>> if (array_buf == NULL)
>> - goto out_array_args;
>> + goto out;
>> err = -EFAULT;
>> if (in_compat_syscall())
>> err = v4l2_compat_get_array_args(file, array_buf,
>> @@ -3321,7 +3321,7 @@ video_usercopy(struct file *file, unsigned int orig_cmd, unsigned long arg,
>> err = copy_from_user(array_buf, user_ptr, array_size) ?
>> -EFAULT : 0;
>> if (err)
>> - goto out_array_args;
>> + goto out;
>> *kernel_ptr = array_buf;
>> }
>>
>> @@ -3353,7 +3353,6 @@ video_usercopy(struct file *file, unsigned int orig_cmd, unsigned long arg,
>> } else if (copy_to_user(user_ptr, array_buf, array_size)) {
>> err = -EFAULT;
>> }
>> - goto out_array_args;
>> }
>> /*
>> * Some ioctls can return an error, but still have valid
>> @@ -3362,7 +3361,6 @@ video_usercopy(struct file *file, unsigned int orig_cmd, unsigned long arg,
>> if (err < 0 && !always_copy)
>> goto out;
>>
>> -out_array_args:
>> if (video_put_user((void __user *)arg, parg, cmd, orig_cmd))
>> err = -EFAULT;
>
> Could we turn this into
>
> if (err >= 0 || always_copy) {
> if (video_put_user((void __user *)arg, parg, cmd, orig_cmd))
> err = -EFAULT;
> }
>
> to avoid the goto ?
>
> With this,
>
> Reviewed-by: Laurent Pinchart <laurent.pinchart@xxxxxxxxxxxxxxxx>
>
> I also wonder if we should skip the copy to user for the array args when
> an error occurs (unless always_copy is set). What do you think ? It's
> not a candidate for this patch of course.
Actually, that's a very good point and I think it belongs to this patch as well.
It's really wrong to copy back data on error unless always_copy is set.
But in that case I prefer to keep the goto: the 'if (err < 0 && !always_copy)'
just moves up to before the 'if (has_array_args) {':
That part of the patch will then look like this:
@@ -3339,6 +3339,13 @@ video_usercopy(struct file *file, unsigned int orig_cmd, unsigned long arg,
trace_v4l2_qbuf(video_devdata(file)->minor, parg);
}
+ /*
+ * Some ioctls can return an error, but still have valid
+ * results that must be returned.
+ */
+ if (err < 0 && !always_copy)
+ goto out;
+
if (has_array_args) {
*kernel_ptr = (void __force *)user_ptr;
if (in_compat_syscall()) {
@@ -3353,16 +3360,8 @@ video_usercopy(struct file *file, unsigned int orig_cmd, unsigned long arg,
} else if (copy_to_user(user_ptr, array_buf, array_size)) {
err = -EFAULT;
}
- goto out_array_args;
}
- /*
- * Some ioctls can return an error, but still have valid
- * results that must be returned.
- */
- if (err < 0 && !always_copy)
- goto out;
-out_array_args:
if (video_put_user((void __user *)arg, parg, cmd, orig_cmd))
err = -EFAULT;
out:
Regards,
Hans
