Re: [PATCH 7/7] misc: syslog: Use RFC5424

[Paul Eggert <eggert@cs.ucla.edu>]

On 10/5/21 6:56 AM, Adhemerval Zanella via Libc-alpha wrote:

> +struct timebuf_t
> +{
> +  char b[sizeof ("YYYY-MM-DDThh:mm:ss.nnnnnnZ")];
> +};

This won't work if people fiddle with their system clocks and set the 
time far in the future or past. I suggest including <intprops.h> and 
using the following instead:

   struct timebuf
   {
     /* Leave room for outlandish but possible years.
        The "+ 1" is for strftime's adding 1900 to tm_year.  */
     char b[INT_STRLEN_BOUND (int) + 1
	   + sizeof "-MM-DDThh:mm:ss.nnnnnnZ"];
   };


> +  struct tm now_tm;
> +  __gmtime64_r (&ts.tv_sec, &now_tm);

This messes up if __gmtime64_r fails, which is possible when time_t is 
very large (or very negative).

> +  __snprintf (timebuf->b, sizeof (timebuf->b),
> +              "%04d-%02d-%02dT%02d:%02d:%02d.%06dZ",
> +              now_tm.tm_year + 1900, now_tm.tm_mon + 1, now_tm.tm_mday,
> +              now_tm.tm_hour, now_tm.tm_min, now_tm.tm_sec,
> +              (int32_t) ts.tv_nsec / 1000);

This messes up if now_tm.tm_year + 1900 overflows. (strftime has a 
similar bug; it should get fixed too but one thing at a time.) Also, 
that 'int32_t' should be plain 'int', to match the format. And it's a 
bit better to not use casts when it's easy, as is the case here.

I suggest something like this instead (I haven't tested it):

   struct tm *gmtm = __gmtime64_r (&ts.tv_sec, &now_tm);
   if (gmtm == NULL)
     strcpy (timebuf->b, "-");
   else
     {
       size_t datebytes = __strftime_l (timebuf->b, sizeof timebuf->b,
                                        "%FT%T", gmtm, _nl_C_locobj_ptr);
       int usec = ts.tv_nsec / 1000;
       __snprintf (timebuf->b + datebytes, sizeof timebuf->b - datebytes,
                   ".%06dZ", usec);
     }

The "-" is as per RFC 5434's NILVALUE.