illumos gate

In an smbfs mount with a misbehaving SMB server, df dumps core like this:

$ dbx ./df
(dbx) catch 8                                                                
(dbx) run /g
Running: df /g 
(process id 15229)
signal FPE (integer divide by zero) in number_to_string at line 1164 in file "df.c" 
 1164                               number / (unsigned long long)(unit_to / unit_from));
(dbx) where 
=>[1] number_to_string(buf = 0x8046e0c ",n^D^H\xf0,\xfc\xfeH^E\xf7\xfep^A", number = 0, unit_from = 0, unit_to = 512), line 1164 in "df.c" 
  [2] default_output(dfrp = 0x806b240, fsp = 0x8046e44), line 1629 in "df.c" 
  [3] do_df(argc = 1, argv = 0x8046f58), line 2032 in "df.c" 
  [4] main(argc = 2, argv = 0x8046f54), line 301 in "df.c" 
(dbx).
(dbx) up  
Current function is default_output
 1629               files_str);
(dbx) print *fsp
*fsp = {
    f_bsize    = 0
    f_frsize   = 0
    f_blocks   = 0
    f_bfree    = 0
    f_bavail   = 0
    f_files    = 0
    f_ffree    = 0
    f_favail   = 0
    f_fsid     = 152043522U
    f_basetype = "smbfs" 
    f_flag     = 2U
    f_namemax  = 255U
    f_fstr     = "" 
    f_filler   = (...)
}
(dbx) 

The SIGFPE core dump happens because f_frsize is zero.
That should not happen (see separate bug TBD about that).
Nonetheless, let's defend against that in df.