Please note that diffs are not public domain; they are subject to the
copyright notices on the relevant files.

===================================================================
RCS file: /ftp/cvs/cvsroot/src/sys/kern/uipc_socket.c,v
rcsdiff: /ftp/cvs/cvsroot/src/sys/kern/uipc_socket.c,v: warning: Unknown phrases like `commitid ...;' are present.
retrieving revision 1.132.2.3
retrieving revision 1.140.6.2
diff -u -p -r1.132.2.3 -r1.140.6.2
--- src/sys/kern/uipc_socket.c	2007/03/24 14:56:06	1.132.2.3
+++ src/sys/kern/uipc_socket.c	2007/10/02 18:29:06	1.140.6.2
@@ -1,4 +1,4 @@
-/*	$NetBSD: uipc_socket.c,v 1.132.2.3 2007/03/24 14:56:06 yamt Exp $	*/
+/*	$NetBSD: uipc_socket.c,v 1.140.6.2 2007/10/02 18:29:06 joerg Exp $	*/
 
 /*-
  * Copyright (c) 2002, 2007 The NetBSD Foundation, Inc.
@@ -68,7 +68,7 @@
  */
 
 #include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: uipc_socket.c,v 1.132.2.3 2007/03/24 14:56:06 yamt Exp $");
+__KERNEL_RCSID(0, "$NetBSD: uipc_socket.c,v 1.140.6.2 2007/10/02 18:29:06 joerg Exp $");
 
 #include "opt_sock_counters.h"
 #include "opt_sosend_loan.h"
@@ -79,6 +79,7 @@ __KERNEL_RCSID(0, "$NetBSD: uipc_socket.
 #include <sys/systm.h>
 #include <sys/proc.h>
 #include <sys/file.h>
+#include <sys/filedesc.h>
 #include <sys/malloc.h>
 #include <sys/mbuf.h>
 #include <sys/domain.h>
@@ -103,6 +104,8 @@ POOL_INIT(socket_pool, sizeof(struct soc
 MALLOC_DEFINE(M_SOOPTS, "soopts", "socket options");
 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
 
+extern const struct fileops socketops;
+
 extern int	somaxconn;			/* patchable (XXX sysctl) */
 int		somaxconn = SOMAXCONN;
 
@@ -427,6 +430,27 @@ sokva_reclaim_callback(struct callback_e
 	return CALLBACK_CHAIN_CONTINUE;
 }
 
+struct mbuf *
+getsombuf(struct socket *so)
+{
+	struct mbuf *m;
+
+	m = m_get(M_WAIT, MT_SONAME);
+	MCLAIM(m, so->so_mowner);
+	return m;
+}
+
+struct mbuf *
+m_intopt(struct socket *so, int val)
+{
+	struct mbuf *m;
+
+	m = getsombuf(so);
+	m->m_len = sizeof(int);
+	*mtod(m, int *) = val;
+	return m;
+}
+
 void
 soinit(void)
 {
@@ -461,29 +485,29 @@ socreate(int dom, struct socket **aso, i
 	error = kauth_authorize_network(l->l_cred, KAUTH_NETWORK_SOCKET,
 	    KAUTH_REQ_NETWORK_SOCKET_OPEN, KAUTH_ARG(dom), KAUTH_ARG(type),
 	    KAUTH_ARG(proto));
-	if (error)
-		return (error);
+	if (error != 0)
+		return error;
 
 	if (proto)
 		prp = pffindproto(dom, proto, type);
 	else
 		prp = pffindtype(dom, type);
-	if (prp == 0) {
+	if (prp == NULL) {
 		/* no support for domain */
 		if (pffinddomain(dom) == 0)
-			return (EAFNOSUPPORT);
+			return EAFNOSUPPORT;
 		/* no support for socket type */
 		if (proto == 0 && type != 0)
-			return (EPROTOTYPE);
-		return (EPROTONOSUPPORT);
+			return EPROTOTYPE;
+		return EPROTONOSUPPORT;
 	}
-	if (prp->pr_usrreq == 0)
-		return (EPROTONOSUPPORT);
+	if (prp->pr_usrreq == NULL)
+		return EPROTONOSUPPORT;
 	if (prp->pr_type != type)
-		return (EPROTOTYPE);
+		return EPROTOTYPE;
 	s = splsoftnet();
 	so = pool_get(&socket_pool, PR_WAITOK);
-	memset((void *)so, 0, sizeof(*so));
+	memset(so, 0, sizeof(*so));
 	TAILQ_INIT(&so->so_q0);
 	TAILQ_INIT(&so->so_q);
 	so->so_type = type;
@@ -495,23 +519,56 @@ socreate(int dom, struct socket **aso, i
 	so->so_snd.sb_mowner = &prp->pr_domain->dom_mowner;
 	so->so_mowner = &prp->pr_domain->dom_mowner;
 #endif
-	if (l != NULL) {
-		uid = kauth_cred_geteuid(l->l_cred);
-	} else {
-		uid = 0;
-	}
+	selinit(&so->so_rcv.sb_sel);
+	selinit(&so->so_snd.sb_sel);
+	uid = kauth_cred_geteuid(l->l_cred);
 	so->so_uidinfo = uid_find(uid);
-	error = (*prp->pr_usrreq)(so, PRU_ATTACH, (struct mbuf *)0,
-	    (struct mbuf *)(long)proto, (struct mbuf *)0, l);
-	if (error) {
+	error = (*prp->pr_usrreq)(so, PRU_ATTACH, NULL,
+	    (struct mbuf *)(long)proto, NULL, l);
+	if (error != 0) {
 		so->so_state |= SS_NOFDREF;
 		sofree(so);
 		splx(s);
-		return (error);
+		return error;
 	}
 	splx(s);
 	*aso = so;
-	return (0);
+	return 0;
+}
+
+/* On success, write file descriptor to fdout and return zero.  On
+ * failure, return non-zero; *fdout will be undefined.
+ */
+int
+fsocreate(int domain, struct socket **sop, int type, int protocol,
+    struct lwp *l, int *fdout)
+{
+	struct filedesc	*fdp;
+	struct socket	*so;
+	struct file	*fp;
+	int		fd, error;
+
+	fdp = l->l_proc->p_fd;
+	/* falloc() will use the desciptor for us */
+	if ((error = falloc(l, &fp, &fd)) != 0)
+		return (error);
+	fp->f_flag = FREAD|FWRITE;
+	fp->f_type = DTYPE_SOCKET;
+	fp->f_ops = &socketops;
+	error = socreate(domain, &so, type, protocol, l);
+	if (error != 0) {
+		FILE_UNUSE(fp, l);
+		fdremove(fdp, fd);
+		ffree(fp);
+	} else {
+		if (sop != NULL)
+			*sop = so;
+		fp->f_data = so;
+		FILE_SET_MATURE(fp);
+		FILE_UNUSE(fp, l);
+		*fdout = fd;
+	}
+	return error;
 }
 
 int
@@ -520,10 +577,9 @@ sobind(struct socket *so, struct mbuf *n
 	int	s, error;
 
 	s = splsoftnet();
-	error = (*so->so_proto->pr_usrreq)(so, PRU_BIND, (struct mbuf *)0,
-	    nam, (struct mbuf *)0, l);
+	error = (*so->so_proto->pr_usrreq)(so, PRU_BIND, NULL, nam, NULL, l);
 	splx(s);
-	return (error);
+	return error;
 }
 
 int
@@ -532,11 +588,11 @@ solisten(struct socket *so, int backlog)
 	int	s, error;
 
 	s = splsoftnet();
-	error = (*so->so_proto->pr_usrreq)(so, PRU_LISTEN, (struct mbuf *)0,
-	    (struct mbuf *)0, (struct mbuf *)0, (struct lwp *)0);
-	if (error) {
+	error = (*so->so_proto->pr_usrreq)(so, PRU_LISTEN, NULL,
+	    NULL, NULL, NULL);
+	if (error != 0) {
 		splx(s);
-		return (error);
+		return error;
 	}
 	if (TAILQ_EMPTY(&so->so_q))
 		so->so_options |= SO_ACCEPTCONN;
@@ -544,7 +600,7 @@ solisten(struct socket *so, int backlog)
 		backlog = 0;
 	so->so_qlimit = min(backlog, somaxconn);
 	splx(s);
-	return (0);
+	return 0;
 }
 
 void
@@ -570,6 +626,8 @@ sofree(struct socket *so)
 		    RLIM_INFINITY);
 	sbrelease(&so->so_snd, so);
 	sorflush(so);
+	seldestroy(&so->so_rcv.sb_sel);
+	seldestroy(&so->so_snd.sb_sel);
 	pool_put(&socket_pool, so);
 }
 
@@ -620,8 +678,7 @@ soclose(struct socket *so)
  drop:
 	if (so->so_pcb) {
 		int error2 = (*so->so_proto->pr_usrreq)(so, PRU_DETACH,
-		    (struct mbuf *)0, (struct mbuf *)0, (struct mbuf *)0,
-		    (struct lwp *)0);
+		    NULL, NULL, NULL, NULL);
 		if (error == 0)
 			error = error2;
 	}
@@ -640,9 +697,15 @@ soclose(struct socket *so)
 int
 soabort(struct socket *so)
 {
+	int error;
 
-	return (*so->so_proto->pr_usrreq)(so, PRU_ABORT, (struct mbuf *)0,
-	    (struct mbuf *)0, (struct mbuf *)0, (struct lwp *)0);
+	KASSERT(so->so_head == NULL);
+	error = (*so->so_proto->pr_usrreq)(so, PRU_ABORT, NULL,
+	    NULL, NULL, NULL);
+	if (error) {
+		sofree(so);
+	}
+	return error;
 }
 
 int
@@ -658,7 +721,7 @@ soaccept(struct socket *so, struct mbuf 
 	if ((so->so_state & SS_ISDISCONNECTED) == 0 ||
 	    (so->so_proto->pr_flags & PR_ABRTACPTDIS) == 0)
 		error = (*so->so_proto->pr_usrreq)(so, PRU_ACCEPT,
-		    (struct mbuf *)0, nam, (struct mbuf *)0, (struct lwp *)0);
+		    NULL, nam, NULL, NULL);
 	else
 		error = ECONNABORTED;
 
@@ -686,7 +749,7 @@ soconnect(struct socket *so, struct mbuf
 		error = EISCONN;
 	else
 		error = (*so->so_proto->pr_usrreq)(so, PRU_CONNECT,
-		    (struct mbuf *)0, nam, (struct mbuf *)0, l);
+		    NULL, nam, NULL, l);
 	splx(s);
 	return (error);
 }
@@ -698,8 +761,7 @@ soconnect2(struct socket *so1, struct so
 
 	s = splsoftnet();
 	error = (*so1->so_proto->pr_usrreq)(so1, PRU_CONNECT2,
-	    (struct mbuf *)0, (struct mbuf *)so2, (struct mbuf *)0,
-	    (struct lwp *)0);
+	    NULL, (struct mbuf *)so2, NULL, NULL);
 	splx(s);
 	return (error);
 }
@@ -719,8 +781,7 @@ sodisconnect(struct socket *so)
 		goto bad;
 	}
 	error = (*so->so_proto->pr_usrreq)(so, PRU_DISCONNECT,
-	    (struct mbuf *)0, (struct mbuf *)0, (struct mbuf *)0,
-	    (struct lwp *)0);
+	    NULL, NULL, NULL, NULL);
  bad:
 	splx(s);
 	sodopendfree();
@@ -837,7 +898,7 @@ sosend(struct socket *so, struct mbuf *a
 					m = m_gethdr(M_WAIT, MT_DATA);
 					mlen = MHLEN;
 					m->m_pkthdr.len = 0;
-					m->m_pkthdr.rcvif = (struct ifnet *)0;
+					m->m_pkthdr.rcvif = NULL;
 				} else {
 					m = m_get(M_WAIT, MT_DATA);
 					mlen = MLEN;
@@ -978,8 +1039,7 @@ soreceive(struct socket *so, struct mbuf
 	if (flags & MSG_OOB) {
 		m = m_get(M_WAIT, MT_DATA);
 		error = (*pr->pr_usrreq)(so, PRU_RCVOOB, m,
-		    (struct mbuf *)(long)(flags & MSG_PEEK),
-		    (struct mbuf *)0, l);
+		    (struct mbuf *)(long)(flags & MSG_PEEK), NULL, l);
 		if (error)
 			goto bad;
 		do {
@@ -993,10 +1053,9 @@ soreceive(struct socket *so, struct mbuf
 		return (error);
 	}
 	if (mp)
-		*mp = (struct mbuf *)0;
+		*mp = NULL;
 	if (so->so_state & SS_ISCONFIRMING && uio->uio_resid)
-		(*pr->pr_usrreq)(so, PRU_RCVD, (struct mbuf *)0,
-		    (struct mbuf *)0, (struct mbuf *)0, l);
+		(*pr->pr_usrreq)(so, PRU_RCVD, NULL, NULL, NULL, l);
 
  restart:
 	if ((error = sblock(&so->so_rcv, SBLOCKWAIT(flags))) != 0)
@@ -1232,7 +1291,7 @@ soreceive(struct socket *so, struct mbuf
 					*mp = m;
 					mp = &m->m_next;
 					so->so_rcv.sb_mb = m = m->m_next;
-					*mp = (struct mbuf *)0;
+					*mp = NULL;
 				} else {
 					MFREE(m, so->so_rcv.sb_mb);
 					m = so->so_rcv.sb_mb;
@@ -1303,9 +1362,7 @@ soreceive(struct socket *so, struct mbuf
 			 */
 			if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb)
 				(*pr->pr_usrreq)(so, PRU_RCVD,
-				    (struct mbuf *)0,
-				    (struct mbuf *)(long)flags,
-				    (struct mbuf *)0, l);
+				    NULL, (struct mbuf *)(long)flags, NULL, l);
 			SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2");
 			SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2");
 			error = sbwait(&so->so_rcv);
@@ -1341,8 +1398,8 @@ soreceive(struct socket *so, struct mbuf
 		SBLASTRECORDCHK(&so->so_rcv, "soreceive 4");
 		SBLASTMBUFCHK(&so->so_rcv, "soreceive 4");
 		if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
-			(*pr->pr_usrreq)(so, PRU_RCVD, (struct mbuf *)0,
-			    (struct mbuf *)(long)flags, (struct mbuf *)0, l);
+			(*pr->pr_usrreq)(so, PRU_RCVD, NULL,
+			    (struct mbuf *)(long)flags, NULL, l);
 	}
 	if (orig_resid == uio->uio_resid && orig_resid &&
 	    (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
@@ -1371,8 +1428,8 @@ soshutdown(struct socket *so, int how)
 	if (how == SHUT_RD || how == SHUT_RDWR)
 		sorflush(so);
 	if (how == SHUT_WR || how == SHUT_RDWR)
-		return (*pr->pr_usrreq)(so, PRU_SHUTDOWN, (struct mbuf *)0,
-		    (struct mbuf *)0, (struct mbuf *)0, (struct lwp *)0);
+		return (*pr->pr_usrreq)(so, PRU_SHUTDOWN, NULL,
+		    NULL, NULL, NULL);
 	return (0);
 }
 
@@ -1403,157 +1460,140 @@ sorflush(struct socket *so)
 	sbrelease(&asb, so);
 }
 
-int
-sosetopt(struct socket *so, int level, int optname, struct mbuf *m0)
+static int
+sosetopt1(struct socket *so, int level, int optname, struct mbuf *m)
 {
-	int		error;
-	struct mbuf	*m;
+	int optval, val;
 	struct linger	*l;
+	struct sockbuf	*sb;
+	struct timeval *tv;
 
-	error = 0;
-	m = m0;
-	if (level != SOL_SOCKET) {
-		if (so->so_proto && so->so_proto->pr_ctloutput)
-			return ((*so->so_proto->pr_ctloutput)
-				  (PRCO_SETOPT, so, level, optname, &m0));
-		error = ENOPROTOOPT;
-	} else {
-		switch (optname) {
+	switch (optname) {
 
-		case SO_LINGER:
-			if (m == NULL || m->m_len != sizeof(struct linger)) {
-				error = EINVAL;
-				goto bad;
-			}
-			l = mtod(m, struct linger *);
-			if (l->l_linger < 0 || l->l_linger > USHRT_MAX ||
-			    l->l_linger > (INT_MAX / hz)) {
-				error = EDOM;
-				goto bad;
-			}
-			so->so_linger = l->l_linger;
-			if (l->l_onoff)
-				so->so_options |= SO_LINGER;
-			else
-				so->so_options &= ~SO_LINGER;
-			break;
+	case SO_LINGER:
+		if (m == NULL || m->m_len != sizeof(struct linger))
+			return EINVAL;
+		l = mtod(m, struct linger *);
+		if (l->l_linger < 0 || l->l_linger > USHRT_MAX ||
+		    l->l_linger > (INT_MAX / hz))
+			return EDOM;
+		so->so_linger = l->l_linger;
+		if (l->l_onoff)
+			so->so_options |= SO_LINGER;
+		else
+			so->so_options &= ~SO_LINGER;
+		break;
 
-		case SO_DEBUG:
-		case SO_KEEPALIVE:
-		case SO_DONTROUTE:
-		case SO_USELOOPBACK:
-		case SO_BROADCAST:
-		case SO_REUSEADDR:
-		case SO_REUSEPORT:
-		case SO_OOBINLINE:
-		case SO_TIMESTAMP:
-			if (m == NULL || m->m_len < sizeof(int)) {
-				error = EINVAL;
-				goto bad;
-			}
-			if (*mtod(m, int *))
-				so->so_options |= optname;
-			else
-				so->so_options &= ~optname;
-			break;
+	case SO_DEBUG:
+	case SO_KEEPALIVE:
+	case SO_DONTROUTE:
+	case SO_USELOOPBACK:
+	case SO_BROADCAST:
+	case SO_REUSEADDR:
+	case SO_REUSEPORT:
+	case SO_OOBINLINE:
+	case SO_TIMESTAMP:
+		if (m == NULL || m->m_len < sizeof(int))
+			return EINVAL;
+		if (*mtod(m, int *))
+			so->so_options |= optname;
+		else
+			so->so_options &= ~optname;
+		break;
+
+	case SO_SNDBUF:
+	case SO_RCVBUF:
+	case SO_SNDLOWAT:
+	case SO_RCVLOWAT:
+		if (m == NULL || m->m_len < sizeof(int))
+			return EINVAL;
+
+		/*
+		 * Values < 1 make no sense for any of these
+		 * options, so disallow them.
+		 */
+		optval = *mtod(m, int *);
+		if (optval < 1)
+			return EINVAL;
+
+		switch (optname) {
 
 		case SO_SNDBUF:
 		case SO_RCVBUF:
+			sb = (optname == SO_SNDBUF) ?
+			    &so->so_snd : &so->so_rcv;
+			if (sbreserve(sb, (u_long)optval, so) == 0)
+				return ENOBUFS;
+			sb->sb_flags &= ~SB_AUTOSIZE;
+			break;
+
+		/*
+		 * Make sure the low-water is never greater than
+		 * the high-water.
+		 */
 		case SO_SNDLOWAT:
+			so->so_snd.sb_lowat =
+			    (optval > so->so_snd.sb_hiwat) ?
+			    so->so_snd.sb_hiwat : optval;
+			break;
 		case SO_RCVLOWAT:
-		    {
-			int optval;
-
-			if (m == NULL || m->m_len < sizeof(int)) {
-				error = EINVAL;
-				goto bad;
-			}
-
-			/*
-			 * Values < 1 make no sense for any of these
-			 * options, so disallow them.
-			 */
-			optval = *mtod(m, int *);
-			if (optval < 1) {
-				error = EINVAL;
-				goto bad;
-			}
-
-			switch (optname) {
+			so->so_rcv.sb_lowat =
+			    (optval > so->so_rcv.sb_hiwat) ?
+			    so->so_rcv.sb_hiwat : optval;
+			break;
+		}
+		break;
 
-			case SO_SNDBUF:
-			case SO_RCVBUF:
-				if (sbreserve(optname == SO_SNDBUF ?
-				    &so->so_snd : &so->so_rcv,
-				    (u_long) optval, so) == 0) {
-					error = ENOBUFS;
-					goto bad;
-				}
-				break;
+	case SO_SNDTIMEO:
+	case SO_RCVTIMEO:
+		if (m == NULL || m->m_len < sizeof(*tv))
+			return EINVAL;
+		tv = mtod(m, struct timeval *);
+		if (tv->tv_sec > (INT_MAX - tv->tv_usec / tick) / hz)
+			return EDOM;
+		val = tv->tv_sec * hz + tv->tv_usec / tick;
+		if (val == 0 && tv->tv_usec != 0)
+			val = 1;
 
-			/*
-			 * Make sure the low-water is never greater than
-			 * the high-water.
-			 */
-			case SO_SNDLOWAT:
-				so->so_snd.sb_lowat =
-				    (optval > so->so_snd.sb_hiwat) ?
-				    so->so_snd.sb_hiwat : optval;
-				break;
-			case SO_RCVLOWAT:
-				so->so_rcv.sb_lowat =
-				    (optval > so->so_rcv.sb_hiwat) ?
-				    so->so_rcv.sb_hiwat : optval;
-				break;
-			}
-			break;
-		    }
+		switch (optname) {
 
 		case SO_SNDTIMEO:
+			so->so_snd.sb_timeo = val;
+			break;
 		case SO_RCVTIMEO:
-		    {
-			struct timeval *tv;
-			int val;
+			so->so_rcv.sb_timeo = val;
+			break;
+		}
+		break;
 
-			if (m == NULL || m->m_len < sizeof(*tv)) {
-				error = EINVAL;
-				goto bad;
-			}
-			tv = mtod(m, struct timeval *);
-			if (tv->tv_sec > (INT_MAX - tv->tv_usec / tick) / hz) {
-				error = EDOM;
-				goto bad;
-			}
-			val = tv->tv_sec * hz + tv->tv_usec / tick;
-			if (val == 0 && tv->tv_usec != 0)
-				val = 1;
+	default:
+		return ENOPROTOOPT;
+	}
+	return 0;
+}
 
-			switch (optname) {
+int
+sosetopt(struct socket *so, int level, int optname, struct mbuf *m)
+{
+	int error, prerr;
 
-			case SO_SNDTIMEO:
-				so->so_snd.sb_timeo = val;
-				break;
-			case SO_RCVTIMEO:
-				so->so_rcv.sb_timeo = val;
-				break;
-			}
-			break;
-		    }
+	if (level == SOL_SOCKET)
+		error = sosetopt1(so, level, optname, m);
+	else
+		error = ENOPROTOOPT;
 
-		default:
-			error = ENOPROTOOPT;
-			break;
-		}
-		if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) {
-			(void) ((*so->so_proto->pr_ctloutput)
-				  (PRCO_SETOPT, so, level, optname, &m0));
-			m = NULL;	/* freed by protocol */
-		}
-	}
- bad:
-	if (m)
-		(void) m_free(m);
-	return (error);
+	if ((error == 0 || error == ENOPROTOOPT) &&
+	    so->so_proto != NULL && so->so_proto->pr_ctloutput != NULL) {
+		/* give the protocol stack a shot */
+		prerr = (*so->so_proto->pr_ctloutput)(PRCO_SETOPT, so, level,
+		    optname, &m);
+		if (prerr == 0)
+			error = 0;
+		else if (prerr != ENOPROTOOPT)
+			error = prerr;
+	} else if (m != NULL)
+		(void)m_free(m);
+	return error;
 }
 
 int