On Wed, Jan 8, 2020 at 7:03 PM Christoph Hellwig <hch@xxxxxx> wrote:
>
> Arnd, can you review the exfat time handling, especially vs y2038
> related issues?
Sure, thanks for adding me to the loop
> On Thu, Jan 02, 2020 at 10:19:02AM +0100, Pali Rohár wrote:
> > On Thursday 02 January 2020 16:20:32 Namjae Jeon wrote:
> > > +#define TIMEZONE_CUR_OFFSET() ((sys_tz.tz_minuteswest / (-15)) & 0x7F)
> > > +/* Convert linear UNIX date to a FAT time/date pair. */
> > > +void exfat_time_unix2fat(struct exfat_sb_info *sbi, struct timespec64 *ts,
> > > + struct exfat_date_time *tp)
> > > +{
> > > + time_t second = ts->tv_sec;
> > > + time_t day, month, year;
> > > + time_t ld; /* leap day */
> >
> > Question for other maintainers: Has kernel code already time_t defined
> > as 64bit? Or it is still just 32bit and 32bit systems and some time64_t
> > needs to be used? I remember that there was discussion about these
> > problems, but do not know if it was changed/fixed or not... Just a
> > pointer for possible Y2038 problem. As "ts" is of type timespec64, but
> > "second" of type time_t.
I am actually very close to sending the patches to remove the time_t
definition from the kernel, at least in yesterday's version there were no
users.
exfat_time_unix2fat() seems to be a copy of the old fat_time_unix2fat()
that we fixed a while ago, please have a look at that implementation
based on time64_to_tm(), which avoids time_t.
Arnd
