On Sat, Jun 16, 2001 at 10:10:35PM -0400, Andrew Brown wrote:
> ...but this is where i disagree.  perhaps netstat should just be
> changed not to used sethostent().  this would be better than altering
> the semantics of sethostent(3).

that would be ok, but not my preference.   i did a recursive grep on 
the entire source tree and the only two programs that actually use 
sethostent(1) are netstat and systat.


> what i'm saying (and badly, i might add, since i'm not sure of a few
> details), is that if udp sockets persist (do they?), then netstat
> could possibly get into a loop whereby each lookup it did caused one
> more lookup to be necessary.  certainly this would happen if netstat
> was using one tcp for connection for each dns lookup.  udp...i dunno.
> i beg ignorance of that.
 
i'm still not sure what you are saying.   there is no recursion at
the resolver-level.   it simply opens /etc/resolv.conf and reads the 
list of IP addresses of local DNS servers from that file.   it contacts
them using either UDP (the default) or TCP (if USEVC flag is set).



> > 3. it seems that sethostent(stayopen=1) should keep /etc/hosts open
> >	if it is using it.   but it closes it anyway.
 
> well...for the next query, it really ought to start again at the
> beginning, no?  i guess that could be reduced from a close()/open() to
> a lseek(?,0,0)?
 
yes.   i think that was the idea of the stayopen flag in the first place?
maybe people don't care so much any more about a few extra opens.   i
thought the system call trace was quite interesting... it shows how simple
operations get more complex over time (due to lib changes).



> > 4. resolver(3) documents RES_STAYOPEN like this:
> >    i read this to say that RES_STAYOPEN is only used with TCP (RES_USEVC).
> >    a system call trace shows that it works with UDP too (saves a couple
> >    of syscalls).
 
> that's what i thought.  i wonder how much it's actually worth, though.

if you are going to do repeated lookups, it save the kernel from 
having to setup and tear down a UDP protocol control block for 
each lookup.   depends on how much you care about optimization...


chuck