Peter Memishian wrote:
>  > > One thing I noticed that doesn't seem right in the UV bits is the
>  > > timestamps on /dev/net nodes.  They seem to be at time 0:
>  > > 
>  > >   # ls -l /dev/net
>  > >   total 0
>  > >   crw-rw-rw-   1 root     sys       58,  1 Dec 31  1969 net0
>  > > 
>  > > I'd think they should have the time they were created, like
>  > > the /dev/pts nodes do.  I suspect this is just a matter of
>  > > setting va_atime, va_mtime and va_ctime in devnet_create_rvp().
>  > > 
>  > See bug 6616725.
> 
> Thanks.  I don't completely follow that bug report, though.  In
> particular, it states:
> 
>   I could change devnet_create_rvp() to fill in the right timestamp in the
>   second argument. But that makes that every time SDEV_UPDATE is set, it
>   reset a new timestamp to this node, which is not what I'd expect.
> 
> ... but I don't see any SDEV_UPDATE flag -- are you talking about
> SDEV_ATTR_INVALID?  

Yes. It is the SDEV_ATTR_INVALID flag. I personally think that only means 
the dev_t has been changed, but not the timestamp. Further, we set that 
SDEV_ATTR_INVALID flag once there is no reference on this /dev/net node. The 
associated device could possibly detach, then attach, with the different or 
the same dev_t.

In the case where SDEV_ATTR_INVALID has been set and
> the dev_t has actually changed, it would seem reasonable to have a new
> timestamp, though I'm not sure if there is any precedent for it.  (Can we
> distinguish the case where the dev_t has changed from the one where it
> hasn't?)
> 
> Perhaps I should discuss this with the devname team?
> 
I tried to set va_atime, va_mtime and va_ctime in devnet_create_rvp() 
yesterday. But the result is not what I expected. It seems that the 
timestamp did not change after I forced the device to detach/reattach. I 
will need some time to investigate more what has happened there.

Thanks
- Cathy


Reply via email to