Re: darwin works with small tweak (long)

[Rick Robino]

> On Thu, Sep 18, 2003 at 11:41:29AM +0200, Martin Pala wrote:
> Jan-Henrik Haukeland wrote:
> 
> >Rick Robino <address@hidden> writes:
> >
> >>As I mentioned in that past letter I can make the error go away by
> >>adjusting just one expression in socket.c; 
> >
> >Ah, but using the address of and address must be wrong, and by doing
> >this, the host variable must point to wonderland (i.e. the host
> >variable cannot contain sensible data in create_socket()):
> >
> >Socket_T socket_new(const char *host, int port, int type, int use_ssl) {
> >....
> >if((s= create_socket((char *)&host, port, proto, NET_TIMEOUT)) != -1)
> >
> > 
> >
> socket_new() is called from remote_status() by:
> 
>  Socket_T sock= socket_new(Run.bind_addr?Run.bind_addr:"localhost", 
> Run.httpdport, SOCKET_TCP, Run.httpdssl);
> 
> => in the case that you have not specified which address is monit httpd 
> binded to, "monit status" will connect to localhost to obtain status 
> from running daemon (where the string "localhost" is static).
> 
> Does it fail if you specify monit httpd address in monit configuration 
> file, such as:
> 
>  set httpd port 2812 and
>     use address 127.0.0.1
>     allow 127.0.0.1
> 
> as well as in the case that you have not specified the interface?:
> 
>  set httpd port 2812 and
>     allow 127.0.0.1
> 
> (Note: in the first case value of Run.bind_addr will be used, in the 
> second case "localhost" will be used).

Sorry I didn't reply earlier, had to sleep and work.  I think you
are on to something.  Here are a few cases and results, varying
only use address and allow.

base monitrc:
set logfile syslog
set mailserver spiff.local
set httpd port 1443
     and use address 192.168.1.2
#     allow 127.0.0.1
#     allow rrobino:******
#     ssl enable
#     pemfile /System/Library/OpenSSL/certs/monit.pem

check apache with pidfile /var/run/httpd.pid
    start program = "/usr/bin/sudo /usr/sbin/apachectl start"
    stop program  = "/usr/bin/sudo /usr/sbin/apachectl stop"

check syslogd with pidfile /var/run/syslog.pid
    start program = "/usr/bin/sudo 
/System/Library/StartupItems/SystemLog/SystemLog StartServic>
    stop program = "/usr/bin/sudo 
/System/Library/StartupItems/SystemLog/SystemLog StopService"

case1:
use address 127.0.0.1
allow 127.0.0.1
        result1: malloc warning

case2:
allow 127.0.0.1
        result2: malloc warning

case3:
use address 127.0.0.1
        result3: no error
        [ same result using 'localhost' instead of loopback address ]

case4:
use address 10.0.1.65 (unresolvable by dns, but assigned and up)
        result4: "The monit server did not return a process, device, ..." error
        [ same result using hostname of 10.1.1.65, as in /etc/hosts and in 
netinfo ]

case5:
use address 10.0.1.65 
allow 10.0.1.65
        result5: both the malloc and the recently fixed "did not return a 
process" error

case6: (unresolvable address, _not_ assigned to any local interface)
use address 192.168.1.2
        result6: no error, slow return
        [ same result using bogus name instead of IP address ]

case7: 
use address 192.168.1.2
allow 192.168.1.2
        result7: malloc error, slow return

Summary:  allow <host|ip> yields the malloc warning, and use address
is responsible for the 'monit server did not return a...' error.
Bogus addresses that aren't really bound to local interfaces
apparently cause a resolution attempt once per service and then
return without displaying an error.

My ssl lines are commented out not only to enforce control on the
test but also because I get an error now; but I haven't updated
my CVS since last night.  Am updating now.

Does this give you any clues?  Any other tests you can think of?

--Rick