On Thu, Jul 13, 2000 at 05:27:32PM -0700, Konrad Schroder wrote:

 > Passing zero to wait forever is fine, the cleaner only needs to wake up
 > when lfs_avail goes too low or a new segment is acquired for writing, or
 > the fs is unmounted.
 > 
 > But I definitely don't understand what the point is of adding "time"
 > into the value (though it's been that way since 4.4Lite)...do I
 > misunderstand something, or should it just be

I think you misunderstand something.  If you look at hzto(), it subtracts
`time' from the passed-in timeval to compute the returned ticks value.

It seems like the right way to get "forever" is to pass NULL for the
system call's `tv' argument, rather than to pass {0,0}.  Otherwise,
there's no way to distinguish between "passed {0,0} to wait forever" and
"timeout expired already for whatever reason, and we should not wait
at all".

 > 
 >         if (SCARG(uap, tv)) { 
 >                 error = copyin(SCARG(uap, tv), &atv, sizeof(struct timeval));
 >                 if (error)
 >                         return (error);
 >                 if (itimerfix(&atv))  
 >                         return (EINVAL);
 >                 timeout = hzto(&atv);
 >         } else
 >                 timeout = 0;
 >    
 >         error = tsleep(addr, PCATCH | PUSER, "segment", timeout);
 > 
 > ?
 > 						Konrad Schroder
 > 						perseant@hhhh.org
 > 
 > 

-- 
        -- Jason R. Thorpe <thorpej@zembu.com>