On Wed, Jun 20, 2018 at 5:51 PM, Coly Li [off-list ref] wrote:

On 2018/6/20 5:51 PM, Arnd Bergmann wrote:

quoted

bcache uses get_seconds() to read the current system time and store it in
the superblock as well as in uuid_entry structures that are user visible.

This changes over from the deprecated function to
ktime_get_real_seconds(), which returns a 64-bit timestamp as it
should. Unfortunately, the two structures are still limited to 32 bits,
so this won't fix any real problems. Let's at least document that
properly, in case we get an updated format in the future it can be
fixed. Until then, we still have some time, and checking the tools
at https://github.com/koverstreet/bcache-tools reveals no access to
any of them.

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

Hi Arnd,

Firstly thanks to your patch, especially the detailed information in
patch log, it helps me to understand the problem more easier.

From the information, it seems the problem is current 32bit time stamp
will be overflow in 2106. So it will be 88 years later, which I have to
say I don't care.

Also for get_seconds() which works well for current code as many other
places call it, I would like to keep it.

I'm currently in the process of removing all instances of get_seconds()
with patches like this. In many cases, we actually want to use
ktime_get_seconds() to return a monotonic time that is immune
to concurrent setttimeofday() calls, in others the code needs to be
changed to avoid the y2038 overflow. For bcache, we don't
really need either of them, but I'd still want to move over everything
to ktime_get_* based interfaces.

Should I clarify that motivation in the changelog text further?

I can also do a simple replacement of get_seconds() with
ktime_get_real_seconds() throughout bcache instead of
adding the intermediate helper function.

      Arnd