Description of problem:

We got a customer report of a command going through PAM crashing for a given user.
It appears that the pam_lastlog code doesn't check the result of localtime_r(), which leads to crashing in glibc's strftime():

~~~
494 static int
495 last_login_failed(pam_handle_t *pamh, int announce, const char *user, time_t lltime)
496 {
 :
502     char the_time[256];
 :
535     if (failed) {
536         /* we want the date? */
537         if (announce & LASTLOG_DATE) {
538             struct tm *tm, tm_buf;
539             time_t lf_time;
540 
541             lf_time = utuser.ut_tv.tv_sec;
542             tm = localtime_r (&lf_time, &tm_buf);
543             strftime (the_time, sizeof (the_time),
544                 /* TRANSLATORS: "strftime options for date of last login" */
545                 _(" %a %b %e %H:%M:%S %Z %Y"), tm);
546 
547             date = the_time;
548         }
~~~

Here above, assuming "lf_time" is very large, due to a corruption in btmp database, this leads to:
1. having "tm" = NULL
2. calling strftime(..., NULL), which crashes

Hence, checking the result is mandatory.


Version-Release number of selected component (if applicable):

PAM from RHEL7 and later


How reproducible:

Don't know, need to have a corrupted "btmp" entry