Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.kernel > #1721757 > unrolled thread

Re: [3/7,media] dvb: don't use 'time_t' in event ioctl

Back to article view | Back to linux.kernel

This discussion starts older than the indexed window; earlier articles aren't shown. The article labeled Started by below is the oldest one visible, not the original post.

Contents

  Re: [3/7,media] dvb: don't use 'time_t' in event ioctl Eugene Syromiatnikov <esyr@redhat.com> - 2017-08-28 17:40 +0200
    Re: [3/7,media] dvb: don't use 'time_t' in event ioctl Arnd Bergmann <arnd@arndb.de> - 2017-08-30 22:30 +0200
      Re: [3/7,media] dvb: don't use 'time_t' in event ioctl Eugene Syromiatnikov <esyr@redhat.com> - 2017-08-31 15:20 +0200

#1721757 — Re: [3/7,media] dvb: don't use 'time_t' in event ioctl

On Tue, Sep 15, 2015 at 05:49:04PM +0200, Arnd Bergmann wrote:
> 'struct video_event' is used for the VIDEO_GET_EVENT ioctl, implemented
> by drivers/media/pci/ivtv/ivtv-ioctl.c and
> drivers/media/pci/ttpci/av7110_av.c. The structure contains a 'time_t',
> which will be redefined in the future to be 64-bit wide, causing an
> incompatible ABI change for this ioctl.
> 
> As it turns out, neither of the drivers currently sets the timestamp
> field, and it is presumably useless anyway because of the limited
> resolutions (no sub-second times). This means we can simply change
> the structure definition to use a 'long' instead of 'time_t' and
> remain compatible with all existing user space binaries when time_t
> gets changed.
> 
> If anybody ever starts using this field, they have to make sure not
> to use 1970 based seconds in there, as those overflow in 2038.
> 
> Signed-off-by: Arnd Bergmann <arnd@arndb.de>
> ---
>  include/uapi/linux/dvb/video.h | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/include/uapi/linux/dvb/video.h b/include/uapi/linux/dvb/video.h
> index d3d14a59d2d5..6c7f9298d7c2 100644
> --- a/include/uapi/linux/dvb/video.h
> +++ b/include/uapi/linux/dvb/video.h
> @@ -135,7 +135,8 @@ struct video_event {
>  #define VIDEO_EVENT_FRAME_RATE_CHANGED	2
>  #define VIDEO_EVENT_DECODER_STOPPED 	3
>  #define VIDEO_EVENT_VSYNC 		4
> -	__kernel_time_t timestamp;
> +	/* unused, make sure to use atomic time for y2038 if it ever gets used */
> +	long timestamp;

This change breaks x32 ABI (and possibly MIPS n32 ABI), as __kernel_time_t
there is 64 bit already:
https://sourceforge.net/p/strace/mailman/message/36015326/

Note the change in structure size from 0x20 to 0x14 for VIDEO_GET_EVENT
command in linux/x32/ioctls_inc0.h.

>  	union {
>  		video_size_t size;
>  		unsigned int frame_rate;	/* in frames per 1000sec */

[toc] | [next] | [standalone]

#1723650

>> diff --git a/include/uapi/linux/dvb/video.h b/include/uapi/linux/dvb/video.h
>> index d3d14a59d2d5..6c7f9298d7c2 100644
>> --- a/include/uapi/linux/dvb/video.h
>> +++ b/include/uapi/linux/dvb/video.h
>> @@ -135,7 +135,8 @@ struct video_event {
>>  #define VIDEO_EVENT_FRAME_RATE_CHANGED       2
>>  #define VIDEO_EVENT_DECODER_STOPPED  3
>>  #define VIDEO_EVENT_VSYNC            4
>> -     __kernel_time_t timestamp;
>> +     /* unused, make sure to use atomic time for y2038 if it ever gets used */
>> +     long timestamp;
>
> This change breaks x32 ABI (and possibly MIPS n32 ABI), as __kernel_time_t
> there is 64 bit already:
> https://sourceforge.net/p/strace/mailman/message/36015326/
>
> Note the change in structure size from 0x20 to 0x14 for VIDEO_GET_EVENT
> command in linux/x32/ioctls_inc0.h.

Are you sure it worked before the change? I don't see any handler in the kernel
for the x32 compat ioctl call here, only the compat_video_event handling, so
my guess is that the change unintentionally fixes x32.

         Arnd

[toc] | [prev] | [next] | [standalone]

#1724215

On Wed, Aug 30, 2017 at 10:25:01PM +0200, Arnd Bergmann wrote:
> >> diff --git a/include/uapi/linux/dvb/video.h b/include/uapi/linux/dvb/video.h
> >> index d3d14a59d2d5..6c7f9298d7c2 100644
> >> --- a/include/uapi/linux/dvb/video.h
> >> +++ b/include/uapi/linux/dvb/video.h
> >> @@ -135,7 +135,8 @@ struct video_event {
> >>  #define VIDEO_EVENT_FRAME_RATE_CHANGED       2
> >>  #define VIDEO_EVENT_DECODER_STOPPED  3
> >>  #define VIDEO_EVENT_VSYNC            4
> >> -     __kernel_time_t timestamp;
> >> +     /* unused, make sure to use atomic time for y2038 if it ever gets used */
> >> +     long timestamp;
> >
> > This change breaks x32 ABI (and possibly MIPS n32 ABI), as __kernel_time_t
> > there is 64 bit already:
> > https://sourceforge.net/p/strace/mailman/message/36015326/
> >
> > Note the change in structure size from 0x20 to 0x14 for VIDEO_GET_EVENT
> > command in linux/x32/ioctls_inc0.h.
> 
> Are you sure it worked before the change? I don't see any handler in the kernel
> for the x32 compat ioctl call here, only the compat_video_event handling, so
> my guess is that the change unintentionally fixes x32.

Yes, you're right; unfortunately, I decided to check which ioctl handler
x32 code is using only after sending the e-mail, and now in the process
of preparing some RFC patch for the ioctl commands which have discrepancies
between x32 and compat sizes.

> 
>          Arnd

[toc] | [prev] | [standalone]

Back to top | Article view | linux.kernel

csiph-web