Re: accpet(3n) under SOlaris 2.4
Casper Dik (casper@fwi.uva.nl)
Wed, 25 Jan 1995 12:01:50 +0100
>i recently tried to install identd on my new Solaris 2.4 machine.. i have
>since found out it has a problem with accept(3n). now.. im not sure why it
>spits the dummy at it, but here is the truss dump, so that those with either
>the source or more knowledge than myself could explain why it is barfing..
>here below it starts right at the accept.. lines beginning with *** are from
>the source code..
>
>*** fd = accept(0, &sad, &sadlen);
>2428: sigaction(SIGCLD, 0xEFFFFC50, 0xEFFFFD50) = 0
>2428: poll(0xEFFFDD90, 1, -1) = 1
>2428: sigfillset(0xEF683BB0) = 0
>2428: sigprocmask(SIG_SETMASK, 0xEFFFFC8C, 0xEFFFFCC0) = 0
>2428: ioctl(0, I_STR, 0xEFFFFB10) = 0
>2428: ioctl(0, I_GETSIG, 0xEFFFFB94) Err#22 EINVAL
>2428: sigprocmask(SIG_SETMASK, 0xEFFFFCC0, 0xEFFFFC6C) = 0
>*** if (fd == -1)
>*** ERROR1("main: accept. errno = %d", errno);
>
>anyway... the ioctl(0,I_GETSIG,xxxxx) seems to be the problem.. fd 0 is a
>socket (which will probably make a difference), just a straight normal tcp
>listen socket with no options. so anyway there is the problem... if anyone
>has a fix or knows what the problem is, it would be appreciated.
There are several reasons why you can't run identd 2.4.x in daemon or
wait mode.
First, there's a general bug in indentd. It uses a struct sockaddr
for accept, not a struct sockaddr_in. But this really shouldn't
matter as both have the same size. I did get EINVAL but after
chaging the type a no longer did. I'll look into it a bit more.
Secondly, in Solaris 2.4, identd needs to use kvm_setproc()/kvm_nextproc().
Unfortunately, kvm_setproc() is broken, and you'll never see processes
started after kvm_open.
I'm working on a fix that will either fix kvm_setproc() for identd
because it's cheaper or will put kvm_open in the child instead of the
parent.
So, identd under 2.4 will only run in nowait mode for now.
Casper