xref: /freebsd/sys/kern/uipc_usrreq.c (revision aed557087269cd052aa76cc15af4a1fd70cbbf24)
19454b2d8SWarner Losh /*-
2df8bae1dSRodney W. Grimes  * Copyright (c) 1982, 1986, 1989, 1991, 1993
3e1ac28e2SRobert Watson  *	The Regents of the University of California.
44d4b555eSRobert Watson  * Copyright 2004-2006 Robert N. M. Watson
5e1ac28e2SRobert Watson  * All rights reserved.
6df8bae1dSRodney W. Grimes  *
7df8bae1dSRodney W. Grimes  * Redistribution and use in source and binary forms, with or without
8df8bae1dSRodney W. Grimes  * modification, are permitted provided that the following conditions
9df8bae1dSRodney W. Grimes  * are met:
10df8bae1dSRodney W. Grimes  * 1. Redistributions of source code must retain the above copyright
11df8bae1dSRodney W. Grimes  *    notice, this list of conditions and the following disclaimer.
12df8bae1dSRodney W. Grimes  * 2. Redistributions in binary form must reproduce the above copyright
13df8bae1dSRodney W. Grimes  *    notice, this list of conditions and the following disclaimer in the
14df8bae1dSRodney W. Grimes  *    documentation and/or other materials provided with the distribution.
15df8bae1dSRodney W. Grimes  * 4. Neither the name of the University nor the names of its contributors
16df8bae1dSRodney W. Grimes  *    may be used to endorse or promote products derived from this software
17df8bae1dSRodney W. Grimes  *    without specific prior written permission.
18df8bae1dSRodney W. Grimes  *
19df8bae1dSRodney W. Grimes  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20df8bae1dSRodney W. Grimes  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21df8bae1dSRodney W. Grimes  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22df8bae1dSRodney W. Grimes  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23df8bae1dSRodney W. Grimes  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24df8bae1dSRodney W. Grimes  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25df8bae1dSRodney W. Grimes  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26df8bae1dSRodney W. Grimes  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27df8bae1dSRodney W. Grimes  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28df8bae1dSRodney W. Grimes  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29df8bae1dSRodney W. Grimes  * SUCH DAMAGE.
30df8bae1dSRodney W. Grimes  *
31748e0b0aSGarrett Wollman  *	From: @(#)uipc_usrreq.c	8.3 (Berkeley) 1/4/94
32df8bae1dSRodney W. Grimes  */
33df8bae1dSRodney W. Grimes 
34f23929fbSRobert Watson /*
35f23929fbSRobert Watson  * UNIX Domain (Local) Sockets
36f23929fbSRobert Watson  *
37f23929fbSRobert Watson  * This is an implementation of UNIX (local) domain sockets.  Each socket has
38f23929fbSRobert Watson  * an associated struct unpcb (UNIX protocol control block).  Stream sockets
39f23929fbSRobert Watson  * may be connected to 0 or 1 other socket.  Datagram sockets may be
40f23929fbSRobert Watson  * connected to 0, 1, or many other sockets.  Sockets may be created and
41f23929fbSRobert Watson  * connected in pairs (socketpair(2)), or bound/connected to using the file
42f23929fbSRobert Watson  * system name space.  For most purposes, only the receive socket buffer is
43f23929fbSRobert Watson  * used, as sending on one socket delivers directly to the receive socket
44f23929fbSRobert Watson  * buffer of a second socket.  The implementation is substantially
45f23929fbSRobert Watson  * complicated by the fact that "ancillary data", such as file descriptors or
46ca948c5eSRobert Watson  * credentials, may be passed across UNIX domain sockets.  The potential for
47ca948c5eSRobert Watson  * passing UNIX domain sockets over other UNIX domain sockets requires the
48ca948c5eSRobert Watson  * implementation of a simple garbage collector to find and tear down cycles
49ca948c5eSRobert Watson  * of disconnected sockets.
50f23929fbSRobert Watson  */
51f23929fbSRobert Watson 
52677b542eSDavid E. O'Brien #include <sys/cdefs.h>
53677b542eSDavid E. O'Brien __FBSDID("$FreeBSD$");
54677b542eSDavid E. O'Brien 
55335654d7SRobert Watson #include "opt_mac.h"
56335654d7SRobert Watson 
57df8bae1dSRodney W. Grimes #include <sys/param.h>
58fb919e4dSMark Murray #include <sys/domain.h>
59960ed29cSSeigo Tanimura #include <sys/fcntl.h>
60d826c479SBruce Evans #include <sys/malloc.h>		/* XXX must be before <sys/file.h> */
614f590175SPaul Saab #include <sys/eventhandler.h>
62639acc13SGarrett Wollman #include <sys/file.h>
63960ed29cSSeigo Tanimura #include <sys/filedesc.h>
64960ed29cSSeigo Tanimura #include <sys/jail.h>
65960ed29cSSeigo Tanimura #include <sys/kernel.h>
66960ed29cSSeigo Tanimura #include <sys/lock.h>
67639acc13SGarrett Wollman #include <sys/mbuf.h>
68033eb86eSJeff Roberson #include <sys/mount.h>
69960ed29cSSeigo Tanimura #include <sys/mutex.h>
70639acc13SGarrett Wollman #include <sys/namei.h>
71639acc13SGarrett Wollman #include <sys/proc.h>
72df8bae1dSRodney W. Grimes #include <sys/protosw.h>
73960ed29cSSeigo Tanimura #include <sys/resourcevar.h>
74df8bae1dSRodney W. Grimes #include <sys/socket.h>
75df8bae1dSRodney W. Grimes #include <sys/socketvar.h>
76960ed29cSSeigo Tanimura #include <sys/signalvar.h>
77df8bae1dSRodney W. Grimes #include <sys/stat.h>
78960ed29cSSeigo Tanimura #include <sys/sx.h>
79639acc13SGarrett Wollman #include <sys/sysctl.h>
80960ed29cSSeigo Tanimura #include <sys/systm.h>
81a0ec558aSRobert Watson #include <sys/taskqueue.h>
82639acc13SGarrett Wollman #include <sys/un.h>
8398271db4SGarrett Wollman #include <sys/unpcb.h>
84639acc13SGarrett Wollman #include <sys/vnode.h>
85df8bae1dSRodney W. Grimes 
86aed55708SRobert Watson #include <security/mac/mac_framework.h>
87aed55708SRobert Watson 
889e9d298aSJeff Roberson #include <vm/uma.h>
8998271db4SGarrett Wollman 
909e9d298aSJeff Roberson static uma_zone_t unp_zone;
9198271db4SGarrett Wollman static	unp_gen_t unp_gencnt;
9298271db4SGarrett Wollman static	u_int unp_count;
9398271db4SGarrett Wollman 
9498271db4SGarrett Wollman static	struct unp_head unp_shead, unp_dhead;
9598271db4SGarrett Wollman 
96df8bae1dSRodney W. Grimes /*
97df8bae1dSRodney W. Grimes  * Unix communications domain.
98df8bae1dSRodney W. Grimes  *
99df8bae1dSRodney W. Grimes  * TODO:
100df8bae1dSRodney W. Grimes  *	SEQPACKET, RDM
101df8bae1dSRodney W. Grimes  *	rethink name space problems
102df8bae1dSRodney W. Grimes  *	need a proper out-of-band
10398271db4SGarrett Wollman  *	lock pushdown
104df8bae1dSRodney W. Grimes  */
105e7dd9a10SRobert Watson static const struct	sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL };
106f708ef1bSPoul-Henning Kamp static ino_t	unp_ino;		/* prototype for fake inode numbers */
1076a2989fdSMatthew N. Dodd struct mbuf *unp_addsockcred(struct thread *, struct mbuf *);
108f708ef1bSPoul-Henning Kamp 
109ce5f32deSRobert Watson /*
1107e711c3aSRobert Watson  * Both send and receive buffers are allocated PIPSIZ bytes of buffering for
1117e711c3aSRobert Watson  * stream sockets, although the total for sender and receiver is actually
1127e711c3aSRobert Watson  * only PIPSIZ.
1137e711c3aSRobert Watson  *
1147e711c3aSRobert Watson  * Datagram sockets really use the sendspace as the maximum datagram size,
1157e711c3aSRobert Watson  * and don't really want to reserve the sendspace.  Their recvspace should be
1167e711c3aSRobert Watson  * large enough for at least one max-size datagram plus address.
1177e711c3aSRobert Watson  */
1187e711c3aSRobert Watson #ifndef PIPSIZ
1197e711c3aSRobert Watson #define	PIPSIZ	8192
1207e711c3aSRobert Watson #endif
1217e711c3aSRobert Watson static u_long	unpst_sendspace = PIPSIZ;
1227e711c3aSRobert Watson static u_long	unpst_recvspace = PIPSIZ;
1237e711c3aSRobert Watson static u_long	unpdg_sendspace = 2*1024;	/* really max datagram size */
1247e711c3aSRobert Watson static u_long	unpdg_recvspace = 4*1024;
1257e711c3aSRobert Watson 
1267e711c3aSRobert Watson static int	unp_rights;			/* file descriptors in flight */
1277e711c3aSRobert Watson 
128e4445a03SRobert Watson SYSCTL_NODE(_net, PF_LOCAL, local, CTLFLAG_RW, 0, "Local domain");
129e4445a03SRobert Watson SYSCTL_NODE(_net_local, SOCK_STREAM, stream, CTLFLAG_RW, 0, "SOCK_STREAM");
130e4445a03SRobert Watson SYSCTL_NODE(_net_local, SOCK_DGRAM, dgram, CTLFLAG_RW, 0, "SOCK_DGRAM");
131e4445a03SRobert Watson 
1327e711c3aSRobert Watson SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW,
1337e711c3aSRobert Watson 	   &unpst_sendspace, 0, "");
1347e711c3aSRobert Watson SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW,
1357e711c3aSRobert Watson 	   &unpst_recvspace, 0, "");
1367e711c3aSRobert Watson SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW,
1377e711c3aSRobert Watson 	   &unpdg_sendspace, 0, "");
1387e711c3aSRobert Watson SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW,
1397e711c3aSRobert Watson 	   &unpdg_recvspace, 0, "");
1407e711c3aSRobert Watson SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, "");
1417e711c3aSRobert Watson 
1427e711c3aSRobert Watson /*
143ce5f32deSRobert Watson  * Currently, UNIX domain sockets are protected by a single subsystem lock,
144ce5f32deSRobert Watson  * which covers global data structures and variables, the contents of each
145ce5f32deSRobert Watson  * per-socket unpcb structure, and the so_pcb field in sockets attached to
146ce5f32deSRobert Watson  * the UNIX domain.  This provides for a moderate degree of paralellism, as
147ce5f32deSRobert Watson  * receive operations on UNIX domain sockets do not need to acquire the
148ce5f32deSRobert Watson  * subsystem lock.  Finer grained locking to permit send() without acquiring
149ce5f32deSRobert Watson  * a global lock would be a logical next step.
150ce5f32deSRobert Watson  *
151ce5f32deSRobert Watson  * The UNIX domain socket lock preceds all socket layer locks, including the
152ce5f32deSRobert Watson  * socket lock and socket buffer lock, permitting UNIX domain socket code to
153ce5f32deSRobert Watson  * call into socket support routines without releasing its locks.
154ce5f32deSRobert Watson  *
155ce5f32deSRobert Watson  * Some caution is required in areas where the UNIX domain socket code enters
156ce5f32deSRobert Watson  * VFS in order to create or find rendezvous points.  This results in
157ce5f32deSRobert Watson  * dropping of the UNIX domain socket subsystem lock, acquisition of the
158ce5f32deSRobert Watson  * Giant lock, and potential sleeping.  This increases the chances of races,
159ce5f32deSRobert Watson  * and exposes weaknesses in the socket->protocol API by offering poor
160ce5f32deSRobert Watson  * failure modes.
161ce5f32deSRobert Watson  */
1620d9ce3a1SRobert Watson static struct mtx unp_mtx;
1630d9ce3a1SRobert Watson #define	UNP_LOCK_INIT() \
1640d9ce3a1SRobert Watson 	mtx_init(&unp_mtx, "unp", NULL, MTX_DEF)
1650d9ce3a1SRobert Watson #define	UNP_LOCK()		mtx_lock(&unp_mtx)
1660d9ce3a1SRobert Watson #define	UNP_UNLOCK()		mtx_unlock(&unp_mtx)
1670d9ce3a1SRobert Watson #define	UNP_LOCK_ASSERT()	mtx_assert(&unp_mtx, MA_OWNED)
1684c5bc1caSRobert Watson #define	UNP_UNLOCK_ASSERT()	mtx_assert(&unp_mtx, MA_NOTOWNED)
1690d9ce3a1SRobert Watson 
170a0ec558aSRobert Watson /*
171a0ec558aSRobert Watson  * Garbage collection of cyclic file descriptor/socket references occurs
172a0ec558aSRobert Watson  * asynchronously in a taskqueue context in order to avoid recursion and
173a0ec558aSRobert Watson  * reentrance in the UNIX domain socket, file descriptor, and socket layer
174a0ec558aSRobert Watson  * code.  See unp_gc() for a full description.
175a0ec558aSRobert Watson  */
176a0ec558aSRobert Watson static struct task	unp_gc_task;
177a0ec558aSRobert Watson 
17870f52b48SBruce Evans static int     unp_connect(struct socket *,struct sockaddr *, struct thread *);
1796a2989fdSMatthew N. Dodd static int     unp_connect2(struct socket *so, struct socket *so2, int);
1804d77a549SAlfred Perlstein static void    unp_disconnect(struct unpcb *);
1814d77a549SAlfred Perlstein static void    unp_shutdown(struct unpcb *);
1824d77a549SAlfred Perlstein static void    unp_drop(struct unpcb *, int);
183a0ec558aSRobert Watson static void    unp_gc(__unused void *, int);
1844d77a549SAlfred Perlstein static void    unp_scan(struct mbuf *, void (*)(struct file *));
1854d77a549SAlfred Perlstein static void    unp_mark(struct file *);
1864d77a549SAlfred Perlstein static void    unp_discard(struct file *);
1874d77a549SAlfred Perlstein static void    unp_freerights(struct file **, int);
1884d77a549SAlfred Perlstein static int     unp_internalize(struct mbuf **, struct thread *);
189d374e81eSRobert Watson static int     unp_listen(struct socket *, struct unpcb *, int,
190d374e81eSRobert Watson 		   struct thread *);
191f708ef1bSPoul-Henning Kamp 
192e4445a03SRobert Watson /*
193e4445a03SRobert Watson  * Definitions of protocols supported in the LOCAL domain.
194e4445a03SRobert Watson  */
195e4445a03SRobert Watson static struct domain localdomain;
196e4445a03SRobert Watson static struct protosw localsw[] = {
197e4445a03SRobert Watson {
198e4445a03SRobert Watson 	.pr_type =		SOCK_STREAM,
199e4445a03SRobert Watson 	.pr_domain =		&localdomain,
200e4445a03SRobert Watson 	.pr_flags =		PR_CONNREQUIRED|PR_WANTRCVD|PR_RIGHTS,
201e4445a03SRobert Watson 	.pr_ctloutput =		&uipc_ctloutput,
202e4445a03SRobert Watson 	.pr_usrreqs =		&uipc_usrreqs
203e4445a03SRobert Watson },
204e4445a03SRobert Watson {
205e4445a03SRobert Watson 	.pr_type =		SOCK_DGRAM,
206e4445a03SRobert Watson 	.pr_domain =		&localdomain,
207e4445a03SRobert Watson 	.pr_flags =		PR_ATOMIC|PR_ADDR|PR_RIGHTS,
208e4445a03SRobert Watson 	.pr_usrreqs =		&uipc_usrreqs
209e4445a03SRobert Watson },
210e4445a03SRobert Watson };
211e4445a03SRobert Watson 
212e4445a03SRobert Watson static struct domain localdomain = {
213e4445a03SRobert Watson 	.dom_family =		AF_LOCAL,
214e4445a03SRobert Watson 	.dom_name =		"local",
215e4445a03SRobert Watson 	.dom_init =		unp_init,
216e4445a03SRobert Watson 	.dom_externalize =	unp_externalize,
217e4445a03SRobert Watson 	.dom_dispose =		unp_dispose,
218e4445a03SRobert Watson 	.dom_protosw =		localsw,
219e4445a03SRobert Watson 	.dom_protoswNPROTOSW =	&localsw[sizeof(localsw)/sizeof(localsw[0])]
220e4445a03SRobert Watson };
221e4445a03SRobert Watson DOMAIN_SET(local);
222e4445a03SRobert Watson 
223ac45e92fSRobert Watson static void
224a29f300eSGarrett Wollman uipc_abort(struct socket *so)
225df8bae1dSRodney W. Grimes {
22640f2ac28SRobert Watson 	struct unpcb *unp;
227df8bae1dSRodney W. Grimes 
22840f2ac28SRobert Watson 	unp = sotounpcb(so);
2294d4b555eSRobert Watson 	KASSERT(unp != NULL, ("uipc_abort: unp == NULL"));
2304d4b555eSRobert Watson 	UNP_LOCK();
231a29f300eSGarrett Wollman 	unp_drop(unp, ECONNABORTED);
232a152f8a3SRobert Watson 	UNP_UNLOCK();
233df8bae1dSRodney W. Grimes }
234df8bae1dSRodney W. Grimes 
235a29f300eSGarrett Wollman static int
23657bf258eSGarrett Wollman uipc_accept(struct socket *so, struct sockaddr **nam)
237a29f300eSGarrett Wollman {
23840f2ac28SRobert Watson 	struct unpcb *unp;
2390d9ce3a1SRobert Watson 	const struct sockaddr *sa;
240df8bae1dSRodney W. Grimes 
241df8bae1dSRodney W. Grimes 	/*
2421c381b19SRobert Watson 	 * Pass back name of connected socket, if it was bound and we are
2431c381b19SRobert Watson 	 * still connected (our peer may have closed already!).
244df8bae1dSRodney W. Grimes 	 */
2454d4b555eSRobert Watson 	unp = sotounpcb(so);
2464d4b555eSRobert Watson 	KASSERT(unp != NULL, ("uipc_accept: unp == NULL"));
2470d9ce3a1SRobert Watson 	*nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
2480d9ce3a1SRobert Watson 	UNP_LOCK();
2490d9ce3a1SRobert Watson 	if (unp->unp_conn != NULL && unp->unp_conn->unp_addr != NULL)
2500d9ce3a1SRobert Watson 		sa = (struct sockaddr *) unp->unp_conn->unp_addr;
2510d9ce3a1SRobert Watson 	else
2520d9ce3a1SRobert Watson 		sa = &sun_noname;
2530d9ce3a1SRobert Watson 	bcopy(sa, *nam, sa->sa_len);
2540d9ce3a1SRobert Watson 	UNP_UNLOCK();
255e5aeaa0cSDag-Erling Smørgrav 	return (0);
256a29f300eSGarrett Wollman }
257df8bae1dSRodney W. Grimes 
258a29f300eSGarrett Wollman static int
259b40ce416SJulian Elischer uipc_attach(struct socket *so, int proto, struct thread *td)
260a29f300eSGarrett Wollman {
2616d32873cSRobert Watson 	struct unpcb *unp;
2626d32873cSRobert Watson 	int error;
263df8bae1dSRodney W. Grimes 
2646d32873cSRobert Watson 	KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL"));
2656d32873cSRobert Watson 	if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
2666d32873cSRobert Watson 		switch (so->so_type) {
2676d32873cSRobert Watson 		case SOCK_STREAM:
2686d32873cSRobert Watson 			error = soreserve(so, unpst_sendspace, unpst_recvspace);
2696d32873cSRobert Watson 			break;
2706d32873cSRobert Watson 
2716d32873cSRobert Watson 		case SOCK_DGRAM:
2726d32873cSRobert Watson 			error = soreserve(so, unpdg_sendspace, unpdg_recvspace);
2736d32873cSRobert Watson 			break;
2746d32873cSRobert Watson 
2756d32873cSRobert Watson 		default:
2766d32873cSRobert Watson 			panic("unp_attach");
2776d32873cSRobert Watson 		}
2786d32873cSRobert Watson 		if (error)
2796d32873cSRobert Watson 			return (error);
2806d32873cSRobert Watson 	}
2816d32873cSRobert Watson 	unp = uma_zalloc(unp_zone, M_WAITOK | M_ZERO);
2826d32873cSRobert Watson 	if (unp == NULL)
2836d32873cSRobert Watson 		return (ENOBUFS);
2846d32873cSRobert Watson 	LIST_INIT(&unp->unp_refs);
2856d32873cSRobert Watson 	unp->unp_socket = so;
2866d32873cSRobert Watson 	so->so_pcb = unp;
2876d32873cSRobert Watson 
2886d32873cSRobert Watson 	UNP_LOCK();
2896d32873cSRobert Watson 	unp->unp_gencnt = ++unp_gencnt;
2906d32873cSRobert Watson 	unp_count++;
291b7e2f3ecSRobert Watson 	LIST_INSERT_HEAD(so->so_type == SOCK_DGRAM ? &unp_dhead : &unp_shead,
292b7e2f3ecSRobert Watson 	    unp, unp_link);
2936d32873cSRobert Watson 	UNP_UNLOCK();
2946d32873cSRobert Watson 
2956d32873cSRobert Watson 	return (0);
296a29f300eSGarrett Wollman }
297a29f300eSGarrett Wollman 
298a29f300eSGarrett Wollman static int
299b40ce416SJulian Elischer uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
300a29f300eSGarrett Wollman {
301dd47f5caSRobert Watson 	struct sockaddr_un *soun = (struct sockaddr_un *)nam;
302dd47f5caSRobert Watson 	struct vattr vattr;
303dd47f5caSRobert Watson 	int error, namelen;
304dd47f5caSRobert Watson 	struct nameidata nd;
30540f2ac28SRobert Watson 	struct unpcb *unp;
306dd47f5caSRobert Watson 	struct vnode *vp;
307dd47f5caSRobert Watson 	struct mount *mp;
308dd47f5caSRobert Watson 	char *buf;
309a29f300eSGarrett Wollman 
31040f2ac28SRobert Watson 	unp = sotounpcb(so);
3114d4b555eSRobert Watson 	KASSERT(unp != NULL, ("uipc_bind: unp == NULL"));
3124f1f0ef5SRobert Watson 
3134f1f0ef5SRobert Watson 	namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path);
3144f1f0ef5SRobert Watson 	if (namelen <= 0)
3154f1f0ef5SRobert Watson 		return (EINVAL);
316dd47f5caSRobert Watson 
317dd47f5caSRobert Watson 	/*
3184f1f0ef5SRobert Watson 	 * We don't allow simultaneous bind() calls on a single UNIX domain
3194f1f0ef5SRobert Watson 	 * socket, so flag in-progress operations, and return an error if an
3204f1f0ef5SRobert Watson 	 * operation is already in progress.
3214f1f0ef5SRobert Watson 	 *
3224f1f0ef5SRobert Watson 	 * Historically, we have not allowed a socket to be rebound, so this
3234f1f0ef5SRobert Watson 	 * also returns an error.  Not allowing re-binding certainly
3244f1f0ef5SRobert Watson 	 * simplifies the implementation and avoids a great many possible
3254f1f0ef5SRobert Watson 	 * failure modes.
326dd47f5caSRobert Watson 	 */
3274f1f0ef5SRobert Watson 	UNP_LOCK();
328dd47f5caSRobert Watson 	if (unp->unp_vnode != NULL) {
32940f2ac28SRobert Watson 		UNP_UNLOCK();
330dd47f5caSRobert Watson 		return (EINVAL);
331dd47f5caSRobert Watson 	}
3324f1f0ef5SRobert Watson 	if (unp->unp_flags & UNP_BINDING) {
333dd47f5caSRobert Watson 		UNP_UNLOCK();
3344f1f0ef5SRobert Watson 		return (EALREADY);
335dd47f5caSRobert Watson 	}
3364f1f0ef5SRobert Watson 	unp->unp_flags |= UNP_BINDING;
337dd47f5caSRobert Watson 	UNP_UNLOCK();
338dd47f5caSRobert Watson 
339dd47f5caSRobert Watson 	buf = malloc(namelen + 1, M_TEMP, M_WAITOK);
340dd47f5caSRobert Watson 	strlcpy(buf, soun->sun_path, namelen + 1);
341dd47f5caSRobert Watson 
342dd47f5caSRobert Watson 	mtx_lock(&Giant);
343dd47f5caSRobert Watson restart:
344dd47f5caSRobert Watson 	mtx_assert(&Giant, MA_OWNED);
345dd47f5caSRobert Watson 	NDINIT(&nd, CREATE, NOFOLLOW | LOCKPARENT | SAVENAME, UIO_SYSSPACE,
346dd47f5caSRobert Watson 	    buf, td);
347dd47f5caSRobert Watson /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
348dd47f5caSRobert Watson 	error = namei(&nd);
349dd47f5caSRobert Watson 	if (error)
3504f1f0ef5SRobert Watson 		goto error;
351dd47f5caSRobert Watson 	vp = nd.ni_vp;
352dd47f5caSRobert Watson 	if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
353dd47f5caSRobert Watson 		NDFREE(&nd, NDF_ONLY_PNBUF);
354dd47f5caSRobert Watson 		if (nd.ni_dvp == vp)
355dd47f5caSRobert Watson 			vrele(nd.ni_dvp);
356dd47f5caSRobert Watson 		else
357dd47f5caSRobert Watson 			vput(nd.ni_dvp);
358dd47f5caSRobert Watson 		if (vp != NULL) {
359dd47f5caSRobert Watson 			vrele(vp);
360dd47f5caSRobert Watson 			error = EADDRINUSE;
3614f1f0ef5SRobert Watson 			goto error;
362dd47f5caSRobert Watson 		}
363dd47f5caSRobert Watson 		error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH);
364dd47f5caSRobert Watson 		if (error)
3654f1f0ef5SRobert Watson 			goto error;
366dd47f5caSRobert Watson 		goto restart;
367dd47f5caSRobert Watson 	}
368dd47f5caSRobert Watson 	VATTR_NULL(&vattr);
369dd47f5caSRobert Watson 	vattr.va_type = VSOCK;
370dd47f5caSRobert Watson 	vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask);
371dd47f5caSRobert Watson #ifdef MAC
372dd47f5caSRobert Watson 	error = mac_check_vnode_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd,
373dd47f5caSRobert Watson 	    &vattr);
374dd47f5caSRobert Watson #endif
375dd47f5caSRobert Watson 	if (error == 0) {
376dd47f5caSRobert Watson 		VOP_LEASE(nd.ni_dvp, td, td->td_ucred, LEASE_WRITE);
377dd47f5caSRobert Watson 		error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
378dd47f5caSRobert Watson 	}
379dd47f5caSRobert Watson 	NDFREE(&nd, NDF_ONLY_PNBUF);
380dd47f5caSRobert Watson 	vput(nd.ni_dvp);
381dd47f5caSRobert Watson 	if (error) {
382dd47f5caSRobert Watson 		vn_finished_write(mp);
3834f1f0ef5SRobert Watson 		goto error;
384dd47f5caSRobert Watson 	}
385dd47f5caSRobert Watson 	vp = nd.ni_vp;
3864f1f0ef5SRobert Watson 	ASSERT_VOP_LOCKED(vp, "uipc_bind");
387dd47f5caSRobert Watson 	soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK);
388dd47f5caSRobert Watson 	UNP_LOCK();
389dd47f5caSRobert Watson 	vp->v_socket = unp->unp_socket;
390dd47f5caSRobert Watson 	unp->unp_vnode = vp;
391dd47f5caSRobert Watson 	unp->unp_addr = soun;
3924f1f0ef5SRobert Watson 	unp->unp_flags &= ~UNP_BINDING;
393dd47f5caSRobert Watson 	UNP_UNLOCK();
394dd47f5caSRobert Watson 	VOP_UNLOCK(vp, 0, td);
395dd47f5caSRobert Watson 	vn_finished_write(mp);
3964f1f0ef5SRobert Watson 	mtx_unlock(&Giant);
3974f1f0ef5SRobert Watson 	free(buf, M_TEMP);
3984f1f0ef5SRobert Watson 	return (0);
3994f1f0ef5SRobert Watson error:
4004f1f0ef5SRobert Watson 	UNP_LOCK();
4014f1f0ef5SRobert Watson 	unp->unp_flags &= ~UNP_BINDING;
4024f1f0ef5SRobert Watson 	UNP_UNLOCK();
403dd47f5caSRobert Watson 	mtx_unlock(&Giant);
404dd47f5caSRobert Watson 	free(buf, M_TEMP);
40540f2ac28SRobert Watson 	return (error);
406a29f300eSGarrett Wollman }
407a29f300eSGarrett Wollman 
408a29f300eSGarrett Wollman static int
409b40ce416SJulian Elischer uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
410a29f300eSGarrett Wollman {
4110d9ce3a1SRobert Watson 	int error;
412a29f300eSGarrett Wollman 
413fd179ee9SRobert Watson 	KASSERT(td == curthread, ("uipc_connect: td != curthread"));
4144d4b555eSRobert Watson 	UNP_LOCK();
415fd179ee9SRobert Watson 	error = unp_connect(so, nam, td);
4160d9ce3a1SRobert Watson 	UNP_UNLOCK();
4170d9ce3a1SRobert Watson 	return (error);
418a29f300eSGarrett Wollman }
419a29f300eSGarrett Wollman 
420a152f8a3SRobert Watson /*
421a152f8a3SRobert Watson  * XXXRW: Should also unbind?
422a152f8a3SRobert Watson  */
423a152f8a3SRobert Watson static void
424a152f8a3SRobert Watson uipc_close(struct socket *so)
425a152f8a3SRobert Watson {
426a152f8a3SRobert Watson 	struct unpcb *unp;
427a152f8a3SRobert Watson 
428a152f8a3SRobert Watson 	unp = sotounpcb(so);
429a152f8a3SRobert Watson 	KASSERT(unp != NULL, ("uipc_close: unp == NULL"));
430a152f8a3SRobert Watson 	UNP_LOCK();
431a152f8a3SRobert Watson 	unp_disconnect(unp);
432a152f8a3SRobert Watson 	UNP_UNLOCK();
433a152f8a3SRobert Watson }
434a152f8a3SRobert Watson 
435db48c0d2SRobert Watson int
436a29f300eSGarrett Wollman uipc_connect2(struct socket *so1, struct socket *so2)
437a29f300eSGarrett Wollman {
43840f2ac28SRobert Watson 	struct unpcb *unp;
4390d9ce3a1SRobert Watson 	int error;
440a29f300eSGarrett Wollman 
44140f2ac28SRobert Watson 	unp = sotounpcb(so1);
4424d4b555eSRobert Watson 	KASSERT(unp != NULL, ("uipc_connect2: unp == NULL"));
4434d4b555eSRobert Watson 	UNP_LOCK();
4446a2989fdSMatthew N. Dodd 	error = unp_connect2(so1, so2, PRU_CONNECT2);
4450d9ce3a1SRobert Watson 	UNP_UNLOCK();
4460d9ce3a1SRobert Watson 	return (error);
447a29f300eSGarrett Wollman }
448a29f300eSGarrett Wollman 
449a29f300eSGarrett Wollman /* control is EOPNOTSUPP */
450a29f300eSGarrett Wollman 
451bc725eafSRobert Watson static void
452a29f300eSGarrett Wollman uipc_detach(struct socket *so)
453a29f300eSGarrett Wollman {
4546d32873cSRobert Watson 	int local_unp_rights;
45540f2ac28SRobert Watson 	struct unpcb *unp;
4566d32873cSRobert Watson 	struct vnode *vp;
457a29f300eSGarrett Wollman 
45840f2ac28SRobert Watson 	unp = sotounpcb(so);
4594d4b555eSRobert Watson 	KASSERT(unp != NULL, ("uipc_detach: unp == NULL"));
4604d4b555eSRobert Watson 	UNP_LOCK();
4616d32873cSRobert Watson 	LIST_REMOVE(unp, unp_link);
4626d32873cSRobert Watson 	unp->unp_gencnt = ++unp_gencnt;
4636d32873cSRobert Watson 	--unp_count;
4646d32873cSRobert Watson 	if ((vp = unp->unp_vnode) != NULL) {
4656d32873cSRobert Watson 		/*
4666d32873cSRobert Watson 		 * XXXRW: should v_socket be frobbed only while holding
4676d32873cSRobert Watson 		 * Giant?
4686d32873cSRobert Watson 		 */
4696d32873cSRobert Watson 		unp->unp_vnode->v_socket = NULL;
4706d32873cSRobert Watson 		unp->unp_vnode = NULL;
4716d32873cSRobert Watson 	}
4726d32873cSRobert Watson 	if (unp->unp_conn != NULL)
4736d32873cSRobert Watson 		unp_disconnect(unp);
4746d32873cSRobert Watson 	while (!LIST_EMPTY(&unp->unp_refs)) {
4756d32873cSRobert Watson 		struct unpcb *ref = LIST_FIRST(&unp->unp_refs);
4766d32873cSRobert Watson 		unp_drop(ref, ECONNRESET);
4776d32873cSRobert Watson 	}
4786d32873cSRobert Watson 	unp->unp_socket->so_pcb = NULL;
4796d32873cSRobert Watson 	local_unp_rights = unp_rights;
4806d32873cSRobert Watson 	UNP_UNLOCK();
4816d32873cSRobert Watson 	if (unp->unp_addr != NULL)
4826d32873cSRobert Watson 		FREE(unp->unp_addr, M_SONAME);
4836d32873cSRobert Watson 	uma_zfree(unp_zone, unp);
4846d32873cSRobert Watson 	if (vp) {
4856d32873cSRobert Watson 		int vfslocked;
4866d32873cSRobert Watson 
4876d32873cSRobert Watson 		vfslocked = VFS_LOCK_GIANT(vp->v_mount);
4886d32873cSRobert Watson 		vrele(vp);
4896d32873cSRobert Watson 		VFS_UNLOCK_GIANT(vfslocked);
4906d32873cSRobert Watson 	}
4916d32873cSRobert Watson 	if (local_unp_rights)
4926d32873cSRobert Watson 		taskqueue_enqueue(taskqueue_thread, &unp_gc_task);
493a29f300eSGarrett Wollman }
494a29f300eSGarrett Wollman 
495a29f300eSGarrett Wollman static int
496a29f300eSGarrett Wollman uipc_disconnect(struct socket *so)
497a29f300eSGarrett Wollman {
49840f2ac28SRobert Watson 	struct unpcb *unp;
499a29f300eSGarrett Wollman 
50040f2ac28SRobert Watson 	unp = sotounpcb(so);
5014d4b555eSRobert Watson 	KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL"));
5024d4b555eSRobert Watson 	UNP_LOCK();
503a29f300eSGarrett Wollman 	unp_disconnect(unp);
5040d9ce3a1SRobert Watson 	UNP_UNLOCK();
505e5aeaa0cSDag-Erling Smørgrav 	return (0);
506a29f300eSGarrett Wollman }
507a29f300eSGarrett Wollman 
508a29f300eSGarrett Wollman static int
509d374e81eSRobert Watson uipc_listen(struct socket *so, int backlog, struct thread *td)
510a29f300eSGarrett Wollman {
51140f2ac28SRobert Watson 	struct unpcb *unp;
5120d9ce3a1SRobert Watson 	int error;
513a29f300eSGarrett Wollman 
51440f2ac28SRobert Watson 	unp = sotounpcb(so);
5154d4b555eSRobert Watson 	KASSERT(unp != NULL, ("uipc_listen: unp == NULL"));
5164d4b555eSRobert Watson 	UNP_LOCK();
5174d4b555eSRobert Watson 	if (unp->unp_vnode == NULL) {
51840f2ac28SRobert Watson 		UNP_UNLOCK();
51940f2ac28SRobert Watson 		return (EINVAL);
52040f2ac28SRobert Watson 	}
521d374e81eSRobert Watson 	error = unp_listen(so, unp, backlog, td);
5220d9ce3a1SRobert Watson 	UNP_UNLOCK();
5230d9ce3a1SRobert Watson 	return (error);
524a29f300eSGarrett Wollman }
525a29f300eSGarrett Wollman 
526a29f300eSGarrett Wollman static int
52757bf258eSGarrett Wollman uipc_peeraddr(struct socket *so, struct sockaddr **nam)
528a29f300eSGarrett Wollman {
52940f2ac28SRobert Watson 	struct unpcb *unp;
5300d9ce3a1SRobert Watson 	const struct sockaddr *sa;
531a29f300eSGarrett Wollman 
5324d4b555eSRobert Watson 	unp = sotounpcb(so);
5334d4b555eSRobert Watson 	KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL"));
5340d9ce3a1SRobert Watson 	*nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
5350d9ce3a1SRobert Watson 	UNP_LOCK();
536fc3fcacfSRobert Watson 	if (unp->unp_conn != NULL && unp->unp_conn->unp_addr!= NULL)
5370d9ce3a1SRobert Watson 		sa = (struct sockaddr *) unp->unp_conn->unp_addr;
538bdc5f6a3SHajimu UMEMOTO 	else {
539bdc5f6a3SHajimu UMEMOTO 		/*
540bdc5f6a3SHajimu UMEMOTO 		 * XXX: It seems that this test always fails even when
541bdc5f6a3SHajimu UMEMOTO 		 * connection is established.  So, this else clause is
542bdc5f6a3SHajimu UMEMOTO 		 * added as workaround to return PF_LOCAL sockaddr.
543bdc5f6a3SHajimu UMEMOTO 		 */
5440d9ce3a1SRobert Watson 		sa = &sun_noname;
545bdc5f6a3SHajimu UMEMOTO 	}
5460d9ce3a1SRobert Watson 	bcopy(sa, *nam, sa->sa_len);
5470d9ce3a1SRobert Watson 	UNP_UNLOCK();
548e5aeaa0cSDag-Erling Smørgrav 	return (0);
549a29f300eSGarrett Wollman }
550a29f300eSGarrett Wollman 
551a29f300eSGarrett Wollman static int
552a29f300eSGarrett Wollman uipc_rcvd(struct socket *so, int flags)
553a29f300eSGarrett Wollman {
55440f2ac28SRobert Watson 	struct unpcb *unp;
555a29f300eSGarrett Wollman 	struct socket *so2;
556337cc6b6SRobert Watson 	u_int mbcnt, sbcc;
5576aef685fSBrian Feldman 	u_long newhiwat;
558a29f300eSGarrett Wollman 
55940f2ac28SRobert Watson 	unp = sotounpcb(so);
5604d4b555eSRobert Watson 	KASSERT(unp != NULL, ("uipc_rcvd: unp == NULL"));
561df8bae1dSRodney W. Grimes 	switch (so->so_type) {
562df8bae1dSRodney W. Grimes 	case SOCK_DGRAM:
563a29f300eSGarrett Wollman 		panic("uipc_rcvd DGRAM?");
564df8bae1dSRodney W. Grimes 		/*NOTREACHED*/
565df8bae1dSRodney W. Grimes 
566df8bae1dSRodney W. Grimes 	case SOCK_STREAM:
567df8bae1dSRodney W. Grimes 		/*
5681c381b19SRobert Watson 		 * Adjust backpressure on sender and wakeup any waiting to
5691c381b19SRobert Watson 		 * write.
570df8bae1dSRodney W. Grimes 		 */
571337cc6b6SRobert Watson 		SOCKBUF_LOCK(&so->so_rcv);
572337cc6b6SRobert Watson 		mbcnt = so->so_rcv.sb_mbcnt;
573337cc6b6SRobert Watson 		sbcc = so->so_rcv.sb_cc;
574337cc6b6SRobert Watson 		SOCKBUF_UNLOCK(&so->so_rcv);
575337cc6b6SRobert Watson 		UNP_LOCK();
576337cc6b6SRobert Watson 		if (unp->unp_conn == NULL) {
577337cc6b6SRobert Watson 			UNP_UNLOCK();
578337cc6b6SRobert Watson 			break;
579337cc6b6SRobert Watson 		}
580337cc6b6SRobert Watson 		so2 = unp->unp_conn->unp_socket;
581337cc6b6SRobert Watson 		SOCKBUF_LOCK(&so2->so_snd);
582337cc6b6SRobert Watson 		so2->so_snd.sb_mbmax += unp->unp_mbcnt - mbcnt;
583337cc6b6SRobert Watson 		newhiwat = so2->so_snd.sb_hiwat + unp->unp_cc - sbcc;
584f535380cSDon Lewis 		(void)chgsbsize(so2->so_cred->cr_uidinfo, &so2->so_snd.sb_hiwat,
5856aef685fSBrian Feldman 		    newhiwat, RLIM_INFINITY);
5861e4d7da7SRobert Watson 		sowwakeup_locked(so2);
587337cc6b6SRobert Watson 		unp->unp_mbcnt = mbcnt;
588337cc6b6SRobert Watson 		unp->unp_cc = sbcc;
589337cc6b6SRobert Watson 		UNP_UNLOCK();
590df8bae1dSRodney W. Grimes 		break;
591df8bae1dSRodney W. Grimes 
592df8bae1dSRodney W. Grimes 	default:
593a29f300eSGarrett Wollman 		panic("uipc_rcvd unknown socktype");
594df8bae1dSRodney W. Grimes 	}
595e5aeaa0cSDag-Erling Smørgrav 	return (0);
596a29f300eSGarrett Wollman }
597df8bae1dSRodney W. Grimes 
598a29f300eSGarrett Wollman /* pru_rcvoob is EOPNOTSUPP */
599a29f300eSGarrett Wollman 
600a29f300eSGarrett Wollman static int
60157bf258eSGarrett Wollman uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
602b40ce416SJulian Elischer     struct mbuf *control, struct thread *td)
603a29f300eSGarrett Wollman {
604f3f49bbbSRobert Watson 	struct unpcb *unp, *unp2;
605a29f300eSGarrett Wollman 	struct socket *so2;
606337cc6b6SRobert Watson 	u_int mbcnt, sbcc;
6076aef685fSBrian Feldman 	u_long newhiwat;
608f3f49bbbSRobert Watson 	int error = 0;
609a29f300eSGarrett Wollman 
61040f2ac28SRobert Watson 	unp = sotounpcb(so);
6114d4b555eSRobert Watson 	KASSERT(unp != NULL, ("uipc_send: unp == NULL"));
612a29f300eSGarrett Wollman 	if (flags & PRUS_OOB) {
613a29f300eSGarrett Wollman 		error = EOPNOTSUPP;
614a29f300eSGarrett Wollman 		goto release;
615a29f300eSGarrett Wollman 	}
616a29f300eSGarrett Wollman 
617fc3fcacfSRobert Watson 	if (control != NULL && (error = unp_internalize(&control, td)))
618a29f300eSGarrett Wollman 		goto release;
619df8bae1dSRodney W. Grimes 
6200d9ce3a1SRobert Watson 	UNP_LOCK();
621a29f300eSGarrett Wollman 	switch (so->so_type) {
622a29f300eSGarrett Wollman 	case SOCK_DGRAM:
623a29f300eSGarrett Wollman 	{
624e7dd9a10SRobert Watson 		const struct sockaddr *from;
625df8bae1dSRodney W. Grimes 
626fc3fcacfSRobert Watson 		if (nam != NULL) {
627fc3fcacfSRobert Watson 			if (unp->unp_conn != NULL) {
628df8bae1dSRodney W. Grimes 				error = EISCONN;
629df8bae1dSRodney W. Grimes 				break;
630df8bae1dSRodney W. Grimes 			}
631b40ce416SJulian Elischer 			error = unp_connect(so, nam, td);
632df8bae1dSRodney W. Grimes 			if (error)
633df8bae1dSRodney W. Grimes 				break;
634df8bae1dSRodney W. Grimes 		}
635b5ff0914SRobert Watson 		/*
636b5ff0914SRobert Watson 		 * Because connect() and send() are non-atomic in a sendto()
637b5ff0914SRobert Watson 		 * with a target address, it's possible that the socket will
638b5ff0914SRobert Watson 		 * have disconnected before the send() can run.  In that case
639b5ff0914SRobert Watson 		 * return the slightly counter-intuitive but otherwise
640b5ff0914SRobert Watson 		 * correct error that the socket is not connected.
641b5ff0914SRobert Watson 		 */
642f3f49bbbSRobert Watson 		unp2 = unp->unp_conn;
643b5ff0914SRobert Watson 		if (unp2 == NULL) {
644b5ff0914SRobert Watson 			error = ENOTCONN;
645b5ff0914SRobert Watson 			break;
646b5ff0914SRobert Watson 		}
647f3f49bbbSRobert Watson 		so2 = unp2->unp_socket;
648fc3fcacfSRobert Watson 		if (unp->unp_addr != NULL)
64957bf258eSGarrett Wollman 			from = (struct sockaddr *)unp->unp_addr;
650df8bae1dSRodney W. Grimes 		else
651df8bae1dSRodney W. Grimes 			from = &sun_noname;
652f3f49bbbSRobert Watson 		if (unp2->unp_flags & UNP_WANTCRED)
6536a2989fdSMatthew N. Dodd 			control = unp_addsockcred(td, control);
654a34b7046SRobert Watson 		SOCKBUF_LOCK(&so2->so_rcv);
655a34b7046SRobert Watson 		if (sbappendaddr_locked(&so2->so_rcv, from, m, control)) {
6561e4d7da7SRobert Watson 			sorwakeup_locked(so2);
657fc3fcacfSRobert Watson 			m = NULL;
658fc3fcacfSRobert Watson 			control = NULL;
659e5aeaa0cSDag-Erling Smørgrav 		} else {
660a34b7046SRobert Watson 			SOCKBUF_UNLOCK(&so2->so_rcv);
661df8bae1dSRodney W. Grimes 			error = ENOBUFS;
662e5aeaa0cSDag-Erling Smørgrav 		}
663fc3fcacfSRobert Watson 		if (nam != NULL)
664df8bae1dSRodney W. Grimes 			unp_disconnect(unp);
665df8bae1dSRodney W. Grimes 		break;
666df8bae1dSRodney W. Grimes 	}
667df8bae1dSRodney W. Grimes 
668df8bae1dSRodney W. Grimes 	case SOCK_STREAM:
6696b8fda4dSGarrett Wollman 		/*
6701c381b19SRobert Watson 		 * Connect if not connected yet.
6711c381b19SRobert Watson 		 *
6721c381b19SRobert Watson 		 * Note: A better implementation would complain if not equal
6731c381b19SRobert Watson 		 * to the peer's address.
6746b8fda4dSGarrett Wollman 		 */
675402cc72dSDavid Greenman 		if ((so->so_state & SS_ISCONNECTED) == 0) {
676fc3fcacfSRobert Watson 			if (nam != NULL) {
677b40ce416SJulian Elischer 				error = unp_connect(so, nam, td);
678402cc72dSDavid Greenman 				if (error)
6796b8fda4dSGarrett Wollman 					break;	/* XXX */
680402cc72dSDavid Greenman 			} else {
681402cc72dSDavid Greenman 				error = ENOTCONN;
682402cc72dSDavid Greenman 				break;
683402cc72dSDavid Greenman 			}
684402cc72dSDavid Greenman 		}
685402cc72dSDavid Greenman 
686337cc6b6SRobert Watson 		/* Lockless read. */
687c0b99ffaSRobert Watson 		if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
688df8bae1dSRodney W. Grimes 			error = EPIPE;
689df8bae1dSRodney W. Grimes 			break;
690df8bae1dSRodney W. Grimes 		}
691b5ff0914SRobert Watson 		/*
692b5ff0914SRobert Watson 		 * Because connect() and send() are non-atomic in a sendto()
693b5ff0914SRobert Watson 		 * with a target address, it's possible that the socket will
694b5ff0914SRobert Watson 		 * have disconnected before the send() can run.  In that case
695b5ff0914SRobert Watson 		 * return the slightly counter-intuitive but otherwise
696b5ff0914SRobert Watson 		 * correct error that the socket is not connected.
697b5ff0914SRobert Watson 		 */
698f3f49bbbSRobert Watson 		unp2 = unp->unp_conn;
699b5ff0914SRobert Watson 		if (unp2 == NULL) {
700b5ff0914SRobert Watson 			error = ENOTCONN;
701b5ff0914SRobert Watson 			break;
702b5ff0914SRobert Watson 		}
703f3f49bbbSRobert Watson 		so2 = unp2->unp_socket;
704a34b7046SRobert Watson 		SOCKBUF_LOCK(&so2->so_rcv);
705f3f49bbbSRobert Watson 		if (unp2->unp_flags & UNP_WANTCRED) {
7066a2989fdSMatthew N. Dodd 			/*
7076a2989fdSMatthew N. Dodd 			 * Credentials are passed only once on
7086a2989fdSMatthew N. Dodd 			 * SOCK_STREAM.
7096a2989fdSMatthew N. Dodd 			 */
710f3f49bbbSRobert Watson 			unp2->unp_flags &= ~UNP_WANTCRED;
7116a2989fdSMatthew N. Dodd 			control = unp_addsockcred(td, control);
7126a2989fdSMatthew N. Dodd 		}
713df8bae1dSRodney W. Grimes 		/*
7141c381b19SRobert Watson 		 * Send to paired receive port, and then reduce send buffer
7151c381b19SRobert Watson 		 * hiwater marks to maintain backpressure.  Wake up readers.
716df8bae1dSRodney W. Grimes 		 */
717fc3fcacfSRobert Watson 		if (control != NULL) {
718a34b7046SRobert Watson 			if (sbappendcontrol_locked(&so2->so_rcv, m, control))
719fc3fcacfSRobert Watson 				control = NULL;
720e5aeaa0cSDag-Erling Smørgrav 		} else {
721a34b7046SRobert Watson 			sbappend_locked(&so2->so_rcv, m);
722e5aeaa0cSDag-Erling Smørgrav 		}
723f3f49bbbSRobert Watson 		mbcnt = so2->so_rcv.sb_mbcnt - unp2->unp_mbcnt;
724f3f49bbbSRobert Watson 		unp2->unp_mbcnt = so2->so_rcv.sb_mbcnt;
725337cc6b6SRobert Watson 		sbcc = so2->so_rcv.sb_cc;
726337cc6b6SRobert Watson 		sorwakeup_locked(so2);
727337cc6b6SRobert Watson 
728337cc6b6SRobert Watson 		SOCKBUF_LOCK(&so->so_snd);
729f3f49bbbSRobert Watson 		newhiwat = so->so_snd.sb_hiwat - (sbcc - unp2->unp_cc);
730f535380cSDon Lewis 		(void)chgsbsize(so->so_cred->cr_uidinfo, &so->so_snd.sb_hiwat,
7316aef685fSBrian Feldman 		    newhiwat, RLIM_INFINITY);
732337cc6b6SRobert Watson 		so->so_snd.sb_mbmax -= mbcnt;
7337abe2ac2SAlan Cox 		SOCKBUF_UNLOCK(&so->so_snd);
734337cc6b6SRobert Watson 
735f3f49bbbSRobert Watson 		unp2->unp_cc = sbcc;
736fc3fcacfSRobert Watson 		m = NULL;
737df8bae1dSRodney W. Grimes 		break;
738df8bae1dSRodney W. Grimes 
739df8bae1dSRodney W. Grimes 	default:
740a29f300eSGarrett Wollman 		panic("uipc_send unknown socktype");
741df8bae1dSRodney W. Grimes 	}
742a29f300eSGarrett Wollman 
7436b8fda4dSGarrett Wollman 	/*
7446b8fda4dSGarrett Wollman 	 * SEND_EOF is equivalent to a SEND followed by
7456b8fda4dSGarrett Wollman 	 * a SHUTDOWN.
7466b8fda4dSGarrett Wollman 	 */
747a29f300eSGarrett Wollman 	if (flags & PRUS_EOF) {
7486b8fda4dSGarrett Wollman 		socantsendmore(so);
7496b8fda4dSGarrett Wollman 		unp_shutdown(unp);
7506b8fda4dSGarrett Wollman 	}
7510d9ce3a1SRobert Watson 	UNP_UNLOCK();
752df8bae1dSRodney W. Grimes 
753fc3fcacfSRobert Watson 	if (control != NULL && error != 0)
754bd508d39SDon Lewis 		unp_dispose(control);
755bd508d39SDon Lewis 
756a29f300eSGarrett Wollman release:
757fc3fcacfSRobert Watson 	if (control != NULL)
758a29f300eSGarrett Wollman 		m_freem(control);
759fc3fcacfSRobert Watson 	if (m != NULL)
760a29f300eSGarrett Wollman 		m_freem(m);
761e5aeaa0cSDag-Erling Smørgrav 	return (error);
762a29f300eSGarrett Wollman }
763df8bae1dSRodney W. Grimes 
764a29f300eSGarrett Wollman static int
765a29f300eSGarrett Wollman uipc_sense(struct socket *so, struct stat *sb)
766a29f300eSGarrett Wollman {
76740f2ac28SRobert Watson 	struct unpcb *unp;
768a29f300eSGarrett Wollman 	struct socket *so2;
769a29f300eSGarrett Wollman 
77040f2ac28SRobert Watson 	unp = sotounpcb(so);
7714d4b555eSRobert Watson 	KASSERT(unp != NULL, ("uipc_sense: unp == NULL"));
7724d4b555eSRobert Watson 	UNP_LOCK();
773a29f300eSGarrett Wollman 	sb->st_blksize = so->so_snd.sb_hiwat;
774fc3fcacfSRobert Watson 	if (so->so_type == SOCK_STREAM && unp->unp_conn != NULL) {
775df8bae1dSRodney W. Grimes 		so2 = unp->unp_conn->unp_socket;
776a29f300eSGarrett Wollman 		sb->st_blksize += so2->so_rcv.sb_cc;
777df8bae1dSRodney W. Grimes 	}
778f3732fd1SPoul-Henning Kamp 	sb->st_dev = NODEV;
779df8bae1dSRodney W. Grimes 	if (unp->unp_ino == 0)
7806f782c46SJeffrey Hsu 		unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino;
781a29f300eSGarrett Wollman 	sb->st_ino = unp->unp_ino;
7820d9ce3a1SRobert Watson 	UNP_UNLOCK();
783df8bae1dSRodney W. Grimes 	return (0);
784a29f300eSGarrett Wollman }
785df8bae1dSRodney W. Grimes 
786a29f300eSGarrett Wollman static int
787a29f300eSGarrett Wollman uipc_shutdown(struct socket *so)
788a29f300eSGarrett Wollman {
78940f2ac28SRobert Watson 	struct unpcb *unp;
790df8bae1dSRodney W. Grimes 
79140f2ac28SRobert Watson 	unp = sotounpcb(so);
7924d4b555eSRobert Watson 	KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL"));
7934d4b555eSRobert Watson 	UNP_LOCK();
794a29f300eSGarrett Wollman 	socantsendmore(so);
795a29f300eSGarrett Wollman 	unp_shutdown(unp);
7960d9ce3a1SRobert Watson 	UNP_UNLOCK();
797e5aeaa0cSDag-Erling Smørgrav 	return (0);
798a29f300eSGarrett Wollman }
799df8bae1dSRodney W. Grimes 
800a29f300eSGarrett Wollman static int
80157bf258eSGarrett Wollman uipc_sockaddr(struct socket *so, struct sockaddr **nam)
802a29f300eSGarrett Wollman {
80340f2ac28SRobert Watson 	struct unpcb *unp;
8040d9ce3a1SRobert Watson 	const struct sockaddr *sa;
805a29f300eSGarrett Wollman 
8064d4b555eSRobert Watson 	unp = sotounpcb(so);
8074d4b555eSRobert Watson 	KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL"));
8080d9ce3a1SRobert Watson 	*nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
8090d9ce3a1SRobert Watson 	UNP_LOCK();
810fc3fcacfSRobert Watson 	if (unp->unp_addr != NULL)
8110d9ce3a1SRobert Watson 		sa = (struct sockaddr *) unp->unp_addr;
81283f3198bSThomas Moestl 	else
8130d9ce3a1SRobert Watson 		sa = &sun_noname;
8140d9ce3a1SRobert Watson 	bcopy(sa, *nam, sa->sa_len);
8150d9ce3a1SRobert Watson 	UNP_UNLOCK();
816e5aeaa0cSDag-Erling Smørgrav 	return (0);
817df8bae1dSRodney W. Grimes }
818a29f300eSGarrett Wollman 
819a29f300eSGarrett Wollman struct pr_usrreqs uipc_usrreqs = {
820756d52a1SPoul-Henning Kamp 	.pru_abort = 		uipc_abort,
821756d52a1SPoul-Henning Kamp 	.pru_accept =		uipc_accept,
822756d52a1SPoul-Henning Kamp 	.pru_attach =		uipc_attach,
823756d52a1SPoul-Henning Kamp 	.pru_bind =		uipc_bind,
824756d52a1SPoul-Henning Kamp 	.pru_connect =		uipc_connect,
825756d52a1SPoul-Henning Kamp 	.pru_connect2 =		uipc_connect2,
826756d52a1SPoul-Henning Kamp 	.pru_detach =		uipc_detach,
827756d52a1SPoul-Henning Kamp 	.pru_disconnect =	uipc_disconnect,
828756d52a1SPoul-Henning Kamp 	.pru_listen =		uipc_listen,
829756d52a1SPoul-Henning Kamp 	.pru_peeraddr =		uipc_peeraddr,
830756d52a1SPoul-Henning Kamp 	.pru_rcvd =		uipc_rcvd,
831756d52a1SPoul-Henning Kamp 	.pru_send =		uipc_send,
832756d52a1SPoul-Henning Kamp 	.pru_sense =		uipc_sense,
833756d52a1SPoul-Henning Kamp 	.pru_shutdown =		uipc_shutdown,
834756d52a1SPoul-Henning Kamp 	.pru_sockaddr =		uipc_sockaddr,
835a152f8a3SRobert Watson 	.pru_close =		uipc_close,
836a29f300eSGarrett Wollman };
837df8bae1dSRodney W. Grimes 
8380c1bb4fbSDima Dorfman int
839892af6b9SRobert Watson uipc_ctloutput(struct socket *so, struct sockopt *sopt)
8400c1bb4fbSDima Dorfman {
84140f2ac28SRobert Watson 	struct unpcb *unp;
8420d9ce3a1SRobert Watson 	struct xucred xu;
8436a2989fdSMatthew N. Dodd 	int error, optval;
8446a2989fdSMatthew N. Dodd 
84596a041b5SMatthew N. Dodd 	if (sopt->sopt_level != 0)
84696a041b5SMatthew N. Dodd 		return (EINVAL);
84796a041b5SMatthew N. Dodd 
8486a2989fdSMatthew N. Dodd 	unp = sotounpcb(so);
8494d4b555eSRobert Watson 	KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL"));
8504d4b555eSRobert Watson 	UNP_LOCK();
8516a2989fdSMatthew N. Dodd 	error = 0;
8520c1bb4fbSDima Dorfman 	switch (sopt->sopt_dir) {
8530c1bb4fbSDima Dorfman 	case SOPT_GET:
8540c1bb4fbSDima Dorfman 		switch (sopt->sopt_name) {
8550c1bb4fbSDima Dorfman 		case LOCAL_PEERCRED:
8560c1bb4fbSDima Dorfman 			if (unp->unp_flags & UNP_HAVEPC)
8570d9ce3a1SRobert Watson 				xu = unp->unp_peercred;
8580c1bb4fbSDima Dorfman 			else {
8590c1bb4fbSDima Dorfman 				if (so->so_type == SOCK_STREAM)
8600c1bb4fbSDima Dorfman 					error = ENOTCONN;
8610c1bb4fbSDima Dorfman 				else
8620c1bb4fbSDima Dorfman 					error = EINVAL;
8630c1bb4fbSDima Dorfman 			}
8640d9ce3a1SRobert Watson 			if (error == 0)
8650d9ce3a1SRobert Watson 				error = sooptcopyout(sopt, &xu, sizeof(xu));
8660c1bb4fbSDima Dorfman 			break;
8676a2989fdSMatthew N. Dodd 		case LOCAL_CREDS:
8686a2989fdSMatthew N. Dodd 			optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0;
8696a2989fdSMatthew N. Dodd 			error = sooptcopyout(sopt, &optval, sizeof(optval));
8706a2989fdSMatthew N. Dodd 			break;
8716a2989fdSMatthew N. Dodd 		case LOCAL_CONNWAIT:
8726a2989fdSMatthew N. Dodd 			optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0;
8736a2989fdSMatthew N. Dodd 			error = sooptcopyout(sopt, &optval, sizeof(optval));
8746a2989fdSMatthew N. Dodd 			break;
8750c1bb4fbSDima Dorfman 		default:
8760c1bb4fbSDima Dorfman 			error = EOPNOTSUPP;
8770c1bb4fbSDima Dorfman 			break;
8780c1bb4fbSDima Dorfman 		}
8790c1bb4fbSDima Dorfman 		break;
8800c1bb4fbSDima Dorfman 	case SOPT_SET:
8816a2989fdSMatthew N. Dodd 		switch (sopt->sopt_name) {
8826a2989fdSMatthew N. Dodd 		case LOCAL_CREDS:
8836a2989fdSMatthew N. Dodd 		case LOCAL_CONNWAIT:
8846a2989fdSMatthew N. Dodd 			error = sooptcopyin(sopt, &optval, sizeof(optval),
8856a2989fdSMatthew N. Dodd 					    sizeof(optval));
8866a2989fdSMatthew N. Dodd 			if (error)
8876a2989fdSMatthew N. Dodd 				break;
8886a2989fdSMatthew N. Dodd 
8896a2989fdSMatthew N. Dodd #define	OPTSET(bit) \
8906a2989fdSMatthew N. Dodd 	if (optval) \
8916a2989fdSMatthew N. Dodd 		unp->unp_flags |= bit; \
8926a2989fdSMatthew N. Dodd 	else \
8936a2989fdSMatthew N. Dodd 		unp->unp_flags &= ~bit;
8946a2989fdSMatthew N. Dodd 
8956a2989fdSMatthew N. Dodd 			switch (sopt->sopt_name) {
8966a2989fdSMatthew N. Dodd 			case LOCAL_CREDS:
8976a2989fdSMatthew N. Dodd 				OPTSET(UNP_WANTCRED);
8986a2989fdSMatthew N. Dodd 				break;
8996a2989fdSMatthew N. Dodd 			case LOCAL_CONNWAIT:
9006a2989fdSMatthew N. Dodd 				OPTSET(UNP_CONNWAIT);
9016a2989fdSMatthew N. Dodd 				break;
9026a2989fdSMatthew N. Dodd 			default:
9036a2989fdSMatthew N. Dodd 				break;
9046a2989fdSMatthew N. Dodd 			}
9056a2989fdSMatthew N. Dodd 			break;
9066a2989fdSMatthew N. Dodd #undef	OPTSET
9076a2989fdSMatthew N. Dodd 		default:
9086a2989fdSMatthew N. Dodd 			error = ENOPROTOOPT;
9096a2989fdSMatthew N. Dodd 			break;
9106a2989fdSMatthew N. Dodd 		}
911abb886faSMatthew N. Dodd 		break;
9120c1bb4fbSDima Dorfman 	default:
9130c1bb4fbSDima Dorfman 		error = EOPNOTSUPP;
9140c1bb4fbSDima Dorfman 		break;
9150c1bb4fbSDima Dorfman 	}
9166a2989fdSMatthew N. Dodd 	UNP_UNLOCK();
9170c1bb4fbSDima Dorfman 	return (error);
9180c1bb4fbSDima Dorfman }
9190c1bb4fbSDima Dorfman 
920f708ef1bSPoul-Henning Kamp static int
921892af6b9SRobert Watson unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
922df8bae1dSRodney W. Grimes {
923892af6b9SRobert Watson 	struct sockaddr_un *soun = (struct sockaddr_un *)nam;
924892af6b9SRobert Watson 	struct vnode *vp;
925892af6b9SRobert Watson 	struct socket *so2, *so3;
926b295bdcdSRobert Watson 	struct unpcb *unp, *unp2, *unp3;
92757bf258eSGarrett Wollman 	int error, len;
928df8bae1dSRodney W. Grimes 	struct nameidata nd;
92957bf258eSGarrett Wollman 	char buf[SOCK_MAXADDRLEN];
9300d9ce3a1SRobert Watson 	struct sockaddr *sa;
9310d9ce3a1SRobert Watson 
9320d9ce3a1SRobert Watson 	UNP_LOCK_ASSERT();
933df8bae1dSRodney W. Grimes 
9344d4b555eSRobert Watson 	unp = sotounpcb(so);
9354d4b555eSRobert Watson 	KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
93657bf258eSGarrett Wollman 	len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
93757bf258eSGarrett Wollman 	if (len <= 0)
938e5aeaa0cSDag-Erling Smørgrav 		return (EINVAL);
93955c85568SRobert Drehmel 	strlcpy(buf, soun->sun_path, len + 1);
9404f1f0ef5SRobert Watson 	if (unp->unp_flags & UNP_CONNECTING) {
9414f1f0ef5SRobert Watson 		UNP_UNLOCK();
9424f1f0ef5SRobert Watson 		return (EALREADY);
9434f1f0ef5SRobert Watson 	}
9440d9ce3a1SRobert Watson 	UNP_UNLOCK();
9450d9ce3a1SRobert Watson 	sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
9460d9ce3a1SRobert Watson 	mtx_lock(&Giant);
947b40ce416SJulian Elischer 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, buf, td);
948797f2d22SPoul-Henning Kamp 	error = namei(&nd);
949797f2d22SPoul-Henning Kamp 	if (error)
9500d9ce3a1SRobert Watson 		vp = NULL;
9510d9ce3a1SRobert Watson 	else
952df8bae1dSRodney W. Grimes 		vp = nd.ni_vp;
9530d9ce3a1SRobert Watson 	ASSERT_VOP_LOCKED(vp, "unp_connect");
954762e6b85SEivind Eklund 	NDFREE(&nd, NDF_ONLY_PNBUF);
9550d9ce3a1SRobert Watson 	if (error)
9560d9ce3a1SRobert Watson 		goto bad;
9570d9ce3a1SRobert Watson 
958df8bae1dSRodney W. Grimes 	if (vp->v_type != VSOCK) {
959df8bae1dSRodney W. Grimes 		error = ENOTSOCK;
960df8bae1dSRodney W. Grimes 		goto bad;
961df8bae1dSRodney W. Grimes 	}
962a854ed98SJohn Baldwin 	error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td);
963797f2d22SPoul-Henning Kamp 	if (error)
964df8bae1dSRodney W. Grimes 		goto bad;
9652260c03dSRobert Watson 	mtx_unlock(&Giant);
9662260c03dSRobert Watson 	UNP_LOCK();
967b295bdcdSRobert Watson 	unp = sotounpcb(so);
9684d4b555eSRobert Watson 	KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
969df8bae1dSRodney W. Grimes 	so2 = vp->v_socket;
970fc3fcacfSRobert Watson 	if (so2 == NULL) {
971df8bae1dSRodney W. Grimes 		error = ECONNREFUSED;
9722260c03dSRobert Watson 		goto bad2;
973df8bae1dSRodney W. Grimes 	}
974df8bae1dSRodney W. Grimes 	if (so->so_type != so2->so_type) {
975df8bae1dSRodney W. Grimes 		error = EPROTOTYPE;
9762260c03dSRobert Watson 		goto bad2;
977df8bae1dSRodney W. Grimes 	}
978df8bae1dSRodney W. Grimes 	if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
9790d9ce3a1SRobert Watson 		if (so2->so_options & SO_ACCEPTCONN) {
9800d9ce3a1SRobert Watson 			/*
9811c381b19SRobert Watson 			 * NB: drop locks here so unp_attach is entered w/o
9821c381b19SRobert Watson 			 * locks; this avoids a recursive lock of the head
9831c381b19SRobert Watson 			 * and holding sleep locks across a (potentially)
9841c381b19SRobert Watson 			 * blocking malloc.
9850d9ce3a1SRobert Watson 			 */
9860d9ce3a1SRobert Watson 			UNP_UNLOCK();
9870d9ce3a1SRobert Watson 			so3 = sonewconn(so2, 0);
9880d9ce3a1SRobert Watson 			UNP_LOCK();
9890d9ce3a1SRobert Watson 		} else
9900d9ce3a1SRobert Watson 			so3 = NULL;
9910d9ce3a1SRobert Watson 		if (so3 == NULL) {
992df8bae1dSRodney W. Grimes 			error = ECONNREFUSED;
9930d9ce3a1SRobert Watson 			goto bad2;
994df8bae1dSRodney W. Grimes 		}
9950c1bb4fbSDima Dorfman 		unp = sotounpcb(so);
996df8bae1dSRodney W. Grimes 		unp2 = sotounpcb(so2);
997df8bae1dSRodney W. Grimes 		unp3 = sotounpcb(so3);
9980d9ce3a1SRobert Watson 		if (unp2->unp_addr != NULL) {
9990d9ce3a1SRobert Watson 			bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len);
10000d9ce3a1SRobert Watson 			unp3->unp_addr = (struct sockaddr_un *) sa;
10010d9ce3a1SRobert Watson 			sa = NULL;
10020d9ce3a1SRobert Watson 		}
10030c1bb4fbSDima Dorfman 		/*
10040c1bb4fbSDima Dorfman 		 * unp_peercred management:
10050c1bb4fbSDima Dorfman 		 *
10061c381b19SRobert Watson 		 * The connecter's (client's) credentials are copied from its
10071c381b19SRobert Watson 		 * process structure at the time of connect() (which is now).
10080c1bb4fbSDima Dorfman 		 */
1009a854ed98SJohn Baldwin 		cru2x(td->td_ucred, &unp3->unp_peercred);
10100c1bb4fbSDima Dorfman 		unp3->unp_flags |= UNP_HAVEPC;
10110c1bb4fbSDima Dorfman 		/*
10121c381b19SRobert Watson 		 * The receiver's (server's) credentials are copied from the
10131c381b19SRobert Watson 		 * unp_peercred member of socket on which the former called
10141c381b19SRobert Watson 		 * listen(); unp_listen() cached that process's credentials
10151c381b19SRobert Watson 		 * at that time so we can use them now.
10160c1bb4fbSDima Dorfman 		 */
10170c1bb4fbSDima Dorfman 		KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED,
10180c1bb4fbSDima Dorfman 		    ("unp_connect: listener without cached peercred"));
10190c1bb4fbSDima Dorfman 		memcpy(&unp->unp_peercred, &unp2->unp_peercred,
10200c1bb4fbSDima Dorfman 		    sizeof(unp->unp_peercred));
10210c1bb4fbSDima Dorfman 		unp->unp_flags |= UNP_HAVEPC;
1022481f8fe8SMaxim Konovalov 		if (unp2->unp_flags & UNP_WANTCRED)
1023481f8fe8SMaxim Konovalov 			unp3->unp_flags |= UNP_WANTCRED;
1024335654d7SRobert Watson #ifdef MAC
1025310e7cebSRobert Watson 		SOCK_LOCK(so);
1026335654d7SRobert Watson 		mac_set_socket_peer_from_socket(so, so3);
1027335654d7SRobert Watson 		mac_set_socket_peer_from_socket(so3, so);
1028310e7cebSRobert Watson 		SOCK_UNLOCK(so);
1029335654d7SRobert Watson #endif
10300c1bb4fbSDima Dorfman 
1031df8bae1dSRodney W. Grimes 		so2 = so3;
1032df8bae1dSRodney W. Grimes 	}
10336a2989fdSMatthew N. Dodd 	error = unp_connect2(so, so2, PRU_CONNECT);
10340d9ce3a1SRobert Watson bad2:
10350d9ce3a1SRobert Watson 	UNP_UNLOCK();
10360d9ce3a1SRobert Watson 	mtx_lock(&Giant);
1037df8bae1dSRodney W. Grimes bad:
10380d9ce3a1SRobert Watson 	mtx_assert(&Giant, MA_OWNED);
10390d9ce3a1SRobert Watson 	if (vp != NULL)
1040df8bae1dSRodney W. Grimes 		vput(vp);
10410d9ce3a1SRobert Watson 	mtx_unlock(&Giant);
10420d9ce3a1SRobert Watson 	free(sa, M_SONAME);
10430d9ce3a1SRobert Watson 	UNP_LOCK();
10444f1f0ef5SRobert Watson 	unp->unp_flags &= ~UNP_CONNECTING;
1045df8bae1dSRodney W. Grimes 	return (error);
1046df8bae1dSRodney W. Grimes }
1047df8bae1dSRodney W. Grimes 
1048db48c0d2SRobert Watson static int
10496a2989fdSMatthew N. Dodd unp_connect2(struct socket *so, struct socket *so2, int req)
1050df8bae1dSRodney W. Grimes {
1051892af6b9SRobert Watson 	struct unpcb *unp = sotounpcb(so);
1052892af6b9SRobert Watson 	struct unpcb *unp2;
1053df8bae1dSRodney W. Grimes 
10540d9ce3a1SRobert Watson 	UNP_LOCK_ASSERT();
10550d9ce3a1SRobert Watson 
1056df8bae1dSRodney W. Grimes 	if (so2->so_type != so->so_type)
1057df8bae1dSRodney W. Grimes 		return (EPROTOTYPE);
1058df8bae1dSRodney W. Grimes 	unp2 = sotounpcb(so2);
10594d4b555eSRobert Watson 	KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL"));
1060df8bae1dSRodney W. Grimes 	unp->unp_conn = unp2;
1061df8bae1dSRodney W. Grimes 	switch (so->so_type) {
1062df8bae1dSRodney W. Grimes 	case SOCK_DGRAM:
106398271db4SGarrett Wollman 		LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
1064df8bae1dSRodney W. Grimes 		soisconnected(so);
1065df8bae1dSRodney W. Grimes 		break;
1066df8bae1dSRodney W. Grimes 
1067df8bae1dSRodney W. Grimes 	case SOCK_STREAM:
1068df8bae1dSRodney W. Grimes 		unp2->unp_conn = unp;
10696a2989fdSMatthew N. Dodd 		if (req == PRU_CONNECT &&
10706a2989fdSMatthew N. Dodd 		    ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT))
10716a2989fdSMatthew N. Dodd 			soisconnecting(so);
10726a2989fdSMatthew N. Dodd 		else
1073df8bae1dSRodney W. Grimes 			soisconnected(so);
1074df8bae1dSRodney W. Grimes 		soisconnected(so2);
1075df8bae1dSRodney W. Grimes 		break;
1076df8bae1dSRodney W. Grimes 
1077df8bae1dSRodney W. Grimes 	default:
1078df8bae1dSRodney W. Grimes 		panic("unp_connect2");
1079df8bae1dSRodney W. Grimes 	}
1080df8bae1dSRodney W. Grimes 	return (0);
1081df8bae1dSRodney W. Grimes }
1082df8bae1dSRodney W. Grimes 
1083f708ef1bSPoul-Henning Kamp static void
1084892af6b9SRobert Watson unp_disconnect(struct unpcb *unp)
1085df8bae1dSRodney W. Grimes {
1086892af6b9SRobert Watson 	struct unpcb *unp2 = unp->unp_conn;
10871b2e3b4bSRobert Watson 	struct socket *so;
1088df8bae1dSRodney W. Grimes 
10890d9ce3a1SRobert Watson 	UNP_LOCK_ASSERT();
10900d9ce3a1SRobert Watson 
1091fc3fcacfSRobert Watson 	if (unp2 == NULL)
1092df8bae1dSRodney W. Grimes 		return;
1093fc3fcacfSRobert Watson 	unp->unp_conn = NULL;
1094df8bae1dSRodney W. Grimes 	switch (unp->unp_socket->so_type) {
1095df8bae1dSRodney W. Grimes 	case SOCK_DGRAM:
109698271db4SGarrett Wollman 		LIST_REMOVE(unp, unp_reflink);
10971b2e3b4bSRobert Watson 		so = unp->unp_socket;
10981b2e3b4bSRobert Watson 		SOCK_LOCK(so);
10991b2e3b4bSRobert Watson 		so->so_state &= ~SS_ISCONNECTED;
11001b2e3b4bSRobert Watson 		SOCK_UNLOCK(so);
1101df8bae1dSRodney W. Grimes 		break;
1102df8bae1dSRodney W. Grimes 
1103df8bae1dSRodney W. Grimes 	case SOCK_STREAM:
1104df8bae1dSRodney W. Grimes 		soisdisconnected(unp->unp_socket);
1105fc3fcacfSRobert Watson 		unp2->unp_conn = NULL;
1106df8bae1dSRodney W. Grimes 		soisdisconnected(unp2->unp_socket);
1107df8bae1dSRodney W. Grimes 		break;
1108df8bae1dSRodney W. Grimes 	}
1109df8bae1dSRodney W. Grimes }
1110df8bae1dSRodney W. Grimes 
11110d9ce3a1SRobert Watson /*
11121c381b19SRobert Watson  * unp_pcblist() assumes that UNIX domain socket memory is never reclaimed by
11131c381b19SRobert Watson  * the zone (UMA_ZONE_NOFREE), and as such potentially stale pointers are
11141c381b19SRobert Watson  * safe to reference.  It first scans the list of struct unpcb's to generate
11151c381b19SRobert Watson  * a pointer list, then it rescans its list one entry at a time to
11160d9ce3a1SRobert Watson  * externalize and copyout.  It checks the generation number to see if a
11170d9ce3a1SRobert Watson  * struct unpcb has been reused, and will skip it if so.
11180d9ce3a1SRobert Watson  */
111998271db4SGarrett Wollman static int
112082d9ae4eSPoul-Henning Kamp unp_pcblist(SYSCTL_HANDLER_ARGS)
112198271db4SGarrett Wollman {
1122f5ef029eSPoul-Henning Kamp 	int error, i, n;
112398271db4SGarrett Wollman 	struct unpcb *unp, **unp_list;
112498271db4SGarrett Wollman 	unp_gen_t gencnt;
11258f364875SJulian Elischer 	struct xunpgen *xug;
112698271db4SGarrett Wollman 	struct unp_head *head;
11278f364875SJulian Elischer 	struct xunpcb *xu;
112898271db4SGarrett Wollman 
1129a23d65bfSBruce Evans 	head = ((intptr_t)arg1 == SOCK_DGRAM ? &unp_dhead : &unp_shead);
113098271db4SGarrett Wollman 
113198271db4SGarrett Wollman 	/*
113298271db4SGarrett Wollman 	 * The process of preparing the PCB list is too time-consuming and
113398271db4SGarrett Wollman 	 * resource-intensive to repeat twice on every request.
113498271db4SGarrett Wollman 	 */
1135fc3fcacfSRobert Watson 	if (req->oldptr == NULL) {
113698271db4SGarrett Wollman 		n = unp_count;
11378f364875SJulian Elischer 		req->oldidx = 2 * (sizeof *xug)
113898271db4SGarrett Wollman 			+ (n + n/8) * sizeof(struct xunpcb);
1139e5aeaa0cSDag-Erling Smørgrav 		return (0);
114098271db4SGarrett Wollman 	}
114198271db4SGarrett Wollman 
1142fc3fcacfSRobert Watson 	if (req->newptr != NULL)
1143e5aeaa0cSDag-Erling Smørgrav 		return (EPERM);
114498271db4SGarrett Wollman 
114598271db4SGarrett Wollman 	/*
114698271db4SGarrett Wollman 	 * OK, now we're committed to doing something.
114798271db4SGarrett Wollman 	 */
1148a163d034SWarner Losh 	xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK);
11490d9ce3a1SRobert Watson 	UNP_LOCK();
115098271db4SGarrett Wollman 	gencnt = unp_gencnt;
115198271db4SGarrett Wollman 	n = unp_count;
11520d9ce3a1SRobert Watson 	UNP_UNLOCK();
115398271db4SGarrett Wollman 
11548f364875SJulian Elischer 	xug->xug_len = sizeof *xug;
11558f364875SJulian Elischer 	xug->xug_count = n;
11568f364875SJulian Elischer 	xug->xug_gen = gencnt;
11578f364875SJulian Elischer 	xug->xug_sogen = so_gencnt;
11588f364875SJulian Elischer 	error = SYSCTL_OUT(req, xug, sizeof *xug);
11598f364875SJulian Elischer 	if (error) {
11608f364875SJulian Elischer 		free(xug, M_TEMP);
1161e5aeaa0cSDag-Erling Smørgrav 		return (error);
11628f364875SJulian Elischer 	}
116398271db4SGarrett Wollman 
1164a163d034SWarner Losh 	unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
116598271db4SGarrett Wollman 
11660d9ce3a1SRobert Watson 	UNP_LOCK();
11672e3c8fcbSPoul-Henning Kamp 	for (unp = LIST_FIRST(head), i = 0; unp && i < n;
11682e3c8fcbSPoul-Henning Kamp 	     unp = LIST_NEXT(unp, unp_link)) {
11698a7d8cc6SRobert Watson 		if (unp->unp_gencnt <= gencnt) {
1170a854ed98SJohn Baldwin 			if (cr_cansee(req->td->td_ucred,
11718a7d8cc6SRobert Watson 			    unp->unp_socket->so_cred))
11724787fd37SPaul Saab 				continue;
117398271db4SGarrett Wollman 			unp_list[i++] = unp;
117498271db4SGarrett Wollman 		}
11754787fd37SPaul Saab 	}
11760d9ce3a1SRobert Watson 	UNP_UNLOCK();
11771c381b19SRobert Watson 	n = i;			/* In case we lost some during malloc. */
117898271db4SGarrett Wollman 
117998271db4SGarrett Wollman 	error = 0;
1180fe2eee82SColin Percival 	xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO);
118198271db4SGarrett Wollman 	for (i = 0; i < n; i++) {
118298271db4SGarrett Wollman 		unp = unp_list[i];
118398271db4SGarrett Wollman 		if (unp->unp_gencnt <= gencnt) {
11848f364875SJulian Elischer 			xu->xu_len = sizeof *xu;
11858f364875SJulian Elischer 			xu->xu_unpp = unp;
118698271db4SGarrett Wollman 			/*
118798271db4SGarrett Wollman 			 * XXX - need more locking here to protect against
118898271db4SGarrett Wollman 			 * connect/disconnect races for SMP.
118998271db4SGarrett Wollman 			 */
1190fc3fcacfSRobert Watson 			if (unp->unp_addr != NULL)
11918f364875SJulian Elischer 				bcopy(unp->unp_addr, &xu->xu_addr,
119298271db4SGarrett Wollman 				      unp->unp_addr->sun_len);
1193fc3fcacfSRobert Watson 			if (unp->unp_conn != NULL &&
1194fc3fcacfSRobert Watson 			    unp->unp_conn->unp_addr != NULL)
119598271db4SGarrett Wollman 				bcopy(unp->unp_conn->unp_addr,
11968f364875SJulian Elischer 				      &xu->xu_caddr,
119798271db4SGarrett Wollman 				      unp->unp_conn->unp_addr->sun_len);
11988f364875SJulian Elischer 			bcopy(unp, &xu->xu_unp, sizeof *unp);
11998f364875SJulian Elischer 			sotoxsocket(unp->unp_socket, &xu->xu_socket);
12008f364875SJulian Elischer 			error = SYSCTL_OUT(req, xu, sizeof *xu);
120198271db4SGarrett Wollman 		}
120298271db4SGarrett Wollman 	}
12038f364875SJulian Elischer 	free(xu, M_TEMP);
120498271db4SGarrett Wollman 	if (!error) {
120598271db4SGarrett Wollman 		/*
12061c381b19SRobert Watson 		 * Give the user an updated idea of our state.  If the
12071c381b19SRobert Watson 		 * generation differs from what we told her before, she knows
12081c381b19SRobert Watson 		 * that something happened while we were processing this
12091c381b19SRobert Watson 		 * request, and it might be necessary to retry.
121098271db4SGarrett Wollman 		 */
12118f364875SJulian Elischer 		xug->xug_gen = unp_gencnt;
12128f364875SJulian Elischer 		xug->xug_sogen = so_gencnt;
12138f364875SJulian Elischer 		xug->xug_count = unp_count;
12148f364875SJulian Elischer 		error = SYSCTL_OUT(req, xug, sizeof *xug);
121598271db4SGarrett Wollman 	}
121698271db4SGarrett Wollman 	free(unp_list, M_TEMP);
12178f364875SJulian Elischer 	free(xug, M_TEMP);
1218e5aeaa0cSDag-Erling Smørgrav 	return (error);
121998271db4SGarrett Wollman }
122098271db4SGarrett Wollman 
122198271db4SGarrett Wollman SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLFLAG_RD,
122298271db4SGarrett Wollman 	    (caddr_t)(long)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
122398271db4SGarrett Wollman 	    "List of active local datagram sockets");
122498271db4SGarrett Wollman SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLFLAG_RD,
122598271db4SGarrett Wollman 	    (caddr_t)(long)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
122698271db4SGarrett Wollman 	    "List of active local stream sockets");
122798271db4SGarrett Wollman 
1228f708ef1bSPoul-Henning Kamp static void
1229892af6b9SRobert Watson unp_shutdown(struct unpcb *unp)
1230df8bae1dSRodney W. Grimes {
1231df8bae1dSRodney W. Grimes 	struct socket *so;
1232df8bae1dSRodney W. Grimes 
12330d9ce3a1SRobert Watson 	UNP_LOCK_ASSERT();
12340d9ce3a1SRobert Watson 
1235df8bae1dSRodney W. Grimes 	if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn &&
1236df8bae1dSRodney W. Grimes 	    (so = unp->unp_conn->unp_socket))
1237df8bae1dSRodney W. Grimes 		socantrcvmore(so);
1238df8bae1dSRodney W. Grimes }
1239df8bae1dSRodney W. Grimes 
1240f708ef1bSPoul-Henning Kamp static void
1241892af6b9SRobert Watson unp_drop(struct unpcb *unp, int errno)
1242df8bae1dSRodney W. Grimes {
1243df8bae1dSRodney W. Grimes 	struct socket *so = unp->unp_socket;
1244df8bae1dSRodney W. Grimes 
12450d9ce3a1SRobert Watson 	UNP_LOCK_ASSERT();
12460d9ce3a1SRobert Watson 
1247df8bae1dSRodney W. Grimes 	so->so_error = errno;
1248df8bae1dSRodney W. Grimes 	unp_disconnect(unp);
1249df8bae1dSRodney W. Grimes }
1250df8bae1dSRodney W. Grimes 
12512bc21ed9SDavid Malone static void
1252892af6b9SRobert Watson unp_freerights(struct file **rp, int fdcount)
1253df8bae1dSRodney W. Grimes {
12542bc21ed9SDavid Malone 	int i;
12552bc21ed9SDavid Malone 	struct file *fp;
1256df8bae1dSRodney W. Grimes 
12572bc21ed9SDavid Malone 	for (i = 0; i < fdcount; i++) {
1258df8bae1dSRodney W. Grimes 		fp = *rp;
12598692c025SYoshinobu Inoue 		/*
12601c381b19SRobert Watson 		 * Zero the pointer before calling unp_discard since it may
12611c381b19SRobert Watson 		 * end up in unp_gc()..
1262d7dca903SRobert Watson 		 *
1263d7dca903SRobert Watson 		 * XXXRW: This is less true than it used to be.
12648692c025SYoshinobu Inoue 		 */
1265df8bae1dSRodney W. Grimes 		*rp++ = 0;
12668692c025SYoshinobu Inoue 		unp_discard(fp);
1267df8bae1dSRodney W. Grimes 	}
12682bc21ed9SDavid Malone }
12692bc21ed9SDavid Malone 
12702bc21ed9SDavid Malone int
1271892af6b9SRobert Watson unp_externalize(struct mbuf *control, struct mbuf **controlp)
12722bc21ed9SDavid Malone {
12732bc21ed9SDavid Malone 	struct thread *td = curthread;		/* XXX */
12742bc21ed9SDavid Malone 	struct cmsghdr *cm = mtod(control, struct cmsghdr *);
12752bc21ed9SDavid Malone 	int i;
12762bc21ed9SDavid Malone 	int *fdp;
12772bc21ed9SDavid Malone 	struct file **rp;
12782bc21ed9SDavid Malone 	struct file *fp;
12792bc21ed9SDavid Malone 	void *data;
12802bc21ed9SDavid Malone 	socklen_t clen = control->m_len, datalen;
12812bc21ed9SDavid Malone 	int error, newfds;
12822bc21ed9SDavid Malone 	int f;
12832bc21ed9SDavid Malone 	u_int newlen;
12842bc21ed9SDavid Malone 
12854c5bc1caSRobert Watson 	UNP_UNLOCK_ASSERT();
12864c5bc1caSRobert Watson 
12872bc21ed9SDavid Malone 	error = 0;
12882bc21ed9SDavid Malone 	if (controlp != NULL) /* controlp == NULL => free control messages */
12892bc21ed9SDavid Malone 		*controlp = NULL;
12902bc21ed9SDavid Malone 
12912bc21ed9SDavid Malone 	while (cm != NULL) {
12922bc21ed9SDavid Malone 		if (sizeof(*cm) > clen || cm->cmsg_len > clen) {
12932bc21ed9SDavid Malone 			error = EINVAL;
12942bc21ed9SDavid Malone 			break;
12952bc21ed9SDavid Malone 		}
12962bc21ed9SDavid Malone 
12972bc21ed9SDavid Malone 		data = CMSG_DATA(cm);
12982bc21ed9SDavid Malone 		datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
12992bc21ed9SDavid Malone 
13002bc21ed9SDavid Malone 		if (cm->cmsg_level == SOL_SOCKET
13012bc21ed9SDavid Malone 		    && cm->cmsg_type == SCM_RIGHTS) {
13022bc21ed9SDavid Malone 			newfds = datalen / sizeof(struct file *);
13032bc21ed9SDavid Malone 			rp = data;
13042bc21ed9SDavid Malone 
1305e2f9a08bSOlivier Houchard 			/* If we're not outputting the descriptors free them. */
13062bc21ed9SDavid Malone 			if (error || controlp == NULL) {
13072bc21ed9SDavid Malone 				unp_freerights(rp, newfds);
13082bc21ed9SDavid Malone 				goto next;
13092bc21ed9SDavid Malone 			}
1310426da3bcSAlfred Perlstein 			FILEDESC_LOCK(td->td_proc->p_fd);
13112bc21ed9SDavid Malone 			/* if the new FD's will not fit free them.  */
13122bc21ed9SDavid Malone 			if (!fdavail(td, newfds)) {
1313426da3bcSAlfred Perlstein 				FILEDESC_UNLOCK(td->td_proc->p_fd);
13142bc21ed9SDavid Malone 				error = EMSGSIZE;
13152bc21ed9SDavid Malone 				unp_freerights(rp, newfds);
13162bc21ed9SDavid Malone 				goto next;
1317df8bae1dSRodney W. Grimes 			}
1318ed5b7817SJulian Elischer 			/*
13191c381b19SRobert Watson 			 * Now change each pointer to an fd in the global
13201c381b19SRobert Watson 			 * table to an integer that is the index to the local
13211c381b19SRobert Watson 			 * fd table entry that we set up to point to the
13221c381b19SRobert Watson 			 * global one we are transferring.
1323ed5b7817SJulian Elischer 			 */
13242bc21ed9SDavid Malone 			newlen = newfds * sizeof(int);
13252bc21ed9SDavid Malone 			*controlp = sbcreatecontrol(NULL, newlen,
13262bc21ed9SDavid Malone 			    SCM_RIGHTS, SOL_SOCKET);
13272bc21ed9SDavid Malone 			if (*controlp == NULL) {
1328426da3bcSAlfred Perlstein 				FILEDESC_UNLOCK(td->td_proc->p_fd);
13292bc21ed9SDavid Malone 				error = E2BIG;
13302bc21ed9SDavid Malone 				unp_freerights(rp, newfds);
13312bc21ed9SDavid Malone 				goto next;
13322bc21ed9SDavid Malone 			}
13332bc21ed9SDavid Malone 
13342bc21ed9SDavid Malone 			fdp = (int *)
13352bc21ed9SDavid Malone 			    CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1336df8bae1dSRodney W. Grimes 			for (i = 0; i < newfds; i++) {
1337a6d4491cSDag-Erling Smørgrav 				if (fdalloc(td, 0, &f))
13382bc21ed9SDavid Malone 					panic("unp_externalize fdalloc failed");
13398692c025SYoshinobu Inoue 				fp = *rp++;
1340b40ce416SJulian Elischer 				td->td_proc->p_fd->fd_ofiles[f] = fp;
1341426da3bcSAlfred Perlstein 				FILE_LOCK(fp);
1342df8bae1dSRodney W. Grimes 				fp->f_msgcount--;
1343426da3bcSAlfred Perlstein 				FILE_UNLOCK(fp);
1344df8bae1dSRodney W. Grimes 				unp_rights--;
13458692c025SYoshinobu Inoue 				*fdp++ = f;
1346df8bae1dSRodney W. Grimes 			}
1347426da3bcSAlfred Perlstein 			FILEDESC_UNLOCK(td->td_proc->p_fd);
13481c381b19SRobert Watson 		} else {
13491c381b19SRobert Watson 			/* We can just copy anything else across. */
13502bc21ed9SDavid Malone 			if (error || controlp == NULL)
13512bc21ed9SDavid Malone 				goto next;
13522bc21ed9SDavid Malone 			*controlp = sbcreatecontrol(NULL, datalen,
13532bc21ed9SDavid Malone 			    cm->cmsg_type, cm->cmsg_level);
13542bc21ed9SDavid Malone 			if (*controlp == NULL) {
13552bc21ed9SDavid Malone 				error = ENOBUFS;
13562bc21ed9SDavid Malone 				goto next;
13572bc21ed9SDavid Malone 			}
13582bc21ed9SDavid Malone 			bcopy(data,
13592bc21ed9SDavid Malone 			    CMSG_DATA(mtod(*controlp, struct cmsghdr *)),
13602bc21ed9SDavid Malone 			    datalen);
13612bc21ed9SDavid Malone 		}
13622bc21ed9SDavid Malone 
13632bc21ed9SDavid Malone 		controlp = &(*controlp)->m_next;
13642bc21ed9SDavid Malone 
13652bc21ed9SDavid Malone next:
13662bc21ed9SDavid Malone 		if (CMSG_SPACE(datalen) < clen) {
13672bc21ed9SDavid Malone 			clen -= CMSG_SPACE(datalen);
13682bc21ed9SDavid Malone 			cm = (struct cmsghdr *)
13692bc21ed9SDavid Malone 			    ((caddr_t)cm + CMSG_SPACE(datalen));
13708692c025SYoshinobu Inoue 		} else {
13712bc21ed9SDavid Malone 			clen = 0;
13722bc21ed9SDavid Malone 			cm = NULL;
13738692c025SYoshinobu Inoue 		}
13748692c025SYoshinobu Inoue 	}
13758692c025SYoshinobu Inoue 
13762bc21ed9SDavid Malone 	m_freem(control);
13772bc21ed9SDavid Malone 
13782bc21ed9SDavid Malone 	return (error);
1379df8bae1dSRodney W. Grimes }
1380df8bae1dSRodney W. Grimes 
13814f590175SPaul Saab static void
13824f590175SPaul Saab unp_zone_change(void *tag)
13834f590175SPaul Saab {
13844f590175SPaul Saab 
13854f590175SPaul Saab 	uma_zone_set_max(unp_zone, maxsockets);
13864f590175SPaul Saab }
13874f590175SPaul Saab 
138898271db4SGarrett Wollman void
138998271db4SGarrett Wollman unp_init(void)
139098271db4SGarrett Wollman {
13911c381b19SRobert Watson 
13929e9d298aSJeff Roberson 	unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL,
13939e9d298aSJeff Roberson 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
1394fc3fcacfSRobert Watson 	if (unp_zone == NULL)
139598271db4SGarrett Wollman 		panic("unp_init");
13964f590175SPaul Saab 	uma_zone_set_max(unp_zone, maxsockets);
13974f590175SPaul Saab 	EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change,
13984f590175SPaul Saab 	    NULL, EVENTHANDLER_PRI_ANY);
139998271db4SGarrett Wollman 	LIST_INIT(&unp_dhead);
140098271db4SGarrett Wollman 	LIST_INIT(&unp_shead);
1401a0ec558aSRobert Watson 	TASK_INIT(&unp_gc_task, 0, unp_gc, NULL);
14020d9ce3a1SRobert Watson 	UNP_LOCK_INIT();
140398271db4SGarrett Wollman }
140498271db4SGarrett Wollman 
1405f708ef1bSPoul-Henning Kamp static int
1406892af6b9SRobert Watson unp_internalize(struct mbuf **controlp, struct thread *td)
1407df8bae1dSRodney W. Grimes {
14082bc21ed9SDavid Malone 	struct mbuf *control = *controlp;
1409b40ce416SJulian Elischer 	struct proc *p = td->td_proc;
14108692c025SYoshinobu Inoue 	struct filedesc *fdescp = p->p_fd;
14112bc21ed9SDavid Malone 	struct cmsghdr *cm = mtod(control, struct cmsghdr *);
14122bc21ed9SDavid Malone 	struct cmsgcred *cmcred;
14132bc21ed9SDavid Malone 	struct file **rp;
14142bc21ed9SDavid Malone 	struct file *fp;
14152bc21ed9SDavid Malone 	struct timeval *tv;
14162bc21ed9SDavid Malone 	int i, fd, *fdp;
14172bc21ed9SDavid Malone 	void *data;
14182bc21ed9SDavid Malone 	socklen_t clen = control->m_len, datalen;
14192bc21ed9SDavid Malone 	int error, oldfds;
14208692c025SYoshinobu Inoue 	u_int newlen;
1421df8bae1dSRodney W. Grimes 
14224c5bc1caSRobert Watson 	UNP_UNLOCK_ASSERT();
14234c5bc1caSRobert Watson 
14242bc21ed9SDavid Malone 	error = 0;
14252bc21ed9SDavid Malone 	*controlp = NULL;
14260b788fa1SBill Paul 
14272bc21ed9SDavid Malone 	while (cm != NULL) {
14282bc21ed9SDavid Malone 		if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET
14292bc21ed9SDavid Malone 		    || cm->cmsg_len > clen) {
14302bc21ed9SDavid Malone 			error = EINVAL;
14312bc21ed9SDavid Malone 			goto out;
14322bc21ed9SDavid Malone 		}
14332bc21ed9SDavid Malone 
14342bc21ed9SDavid Malone 		data = CMSG_DATA(cm);
14352bc21ed9SDavid Malone 		datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
14362bc21ed9SDavid Malone 
14372bc21ed9SDavid Malone 		switch (cm->cmsg_type) {
14380b788fa1SBill Paul 		/*
14390b788fa1SBill Paul 		 * Fill in credential information.
14400b788fa1SBill Paul 		 */
14412bc21ed9SDavid Malone 		case SCM_CREDS:
14422bc21ed9SDavid Malone 			*controlp = sbcreatecontrol(NULL, sizeof(*cmcred),
14432bc21ed9SDavid Malone 			    SCM_CREDS, SOL_SOCKET);
14442bc21ed9SDavid Malone 			if (*controlp == NULL) {
14452bc21ed9SDavid Malone 				error = ENOBUFS;
14462bc21ed9SDavid Malone 				goto out;
14472bc21ed9SDavid Malone 			}
14482bc21ed9SDavid Malone 
14492bc21ed9SDavid Malone 			cmcred = (struct cmsgcred *)
14502bc21ed9SDavid Malone 			    CMSG_DATA(mtod(*controlp, struct cmsghdr *));
14510b788fa1SBill Paul 			cmcred->cmcred_pid = p->p_pid;
1452a854ed98SJohn Baldwin 			cmcred->cmcred_uid = td->td_ucred->cr_ruid;
1453a854ed98SJohn Baldwin 			cmcred->cmcred_gid = td->td_ucred->cr_rgid;
1454a854ed98SJohn Baldwin 			cmcred->cmcred_euid = td->td_ucred->cr_uid;
1455a854ed98SJohn Baldwin 			cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups,
14560b788fa1SBill Paul 							CMGROUP_MAX);
14570b788fa1SBill Paul 			for (i = 0; i < cmcred->cmcred_ngroups; i++)
14582bc21ed9SDavid Malone 				cmcred->cmcred_groups[i] =
1459a854ed98SJohn Baldwin 				    td->td_ucred->cr_groups[i];
14602bc21ed9SDavid Malone 			break;
14610b788fa1SBill Paul 
14622bc21ed9SDavid Malone 		case SCM_RIGHTS:
14632bc21ed9SDavid Malone 			oldfds = datalen / sizeof (int);
1464ed5b7817SJulian Elischer 			/*
14651c381b19SRobert Watson 			 * Check that all the FDs passed in refer to legal
14661c381b19SRobert Watson 			 * files.  If not, reject the entire operation.
1467ed5b7817SJulian Elischer 			 */
14682bc21ed9SDavid Malone 			fdp = data;
1469426da3bcSAlfred Perlstein 			FILEDESC_LOCK(fdescp);
1470df8bae1dSRodney W. Grimes 			for (i = 0; i < oldfds; i++) {
14718692c025SYoshinobu Inoue 				fd = *fdp++;
14728692c025SYoshinobu Inoue 				if ((unsigned)fd >= fdescp->fd_nfiles ||
14732bc21ed9SDavid Malone 				    fdescp->fd_ofiles[fd] == NULL) {
1474426da3bcSAlfred Perlstein 					FILEDESC_UNLOCK(fdescp);
14752bc21ed9SDavid Malone 					error = EBADF;
14762bc21ed9SDavid Malone 					goto out;
14772bc21ed9SDavid Malone 				}
1478e7d6662fSAlfred Perlstein 				fp = fdescp->fd_ofiles[fd];
1479e7d6662fSAlfred Perlstein 				if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) {
1480e7d6662fSAlfred Perlstein 					FILEDESC_UNLOCK(fdescp);
1481e7d6662fSAlfred Perlstein 					error = EOPNOTSUPP;
1482e7d6662fSAlfred Perlstein 					goto out;
1483e7d6662fSAlfred Perlstein 				}
1484e7d6662fSAlfred Perlstein 
1485df8bae1dSRodney W. Grimes 			}
1486ed5b7817SJulian Elischer 			/*
14871c381b19SRobert Watson 			 * Now replace the integer FDs with pointers to the
14881c381b19SRobert Watson 			 * associated global file table entry..
1489ed5b7817SJulian Elischer 			 */
14902bc21ed9SDavid Malone 			newlen = oldfds * sizeof(struct file *);
14912bc21ed9SDavid Malone 			*controlp = sbcreatecontrol(NULL, newlen,
14922bc21ed9SDavid Malone 			    SCM_RIGHTS, SOL_SOCKET);
14932bc21ed9SDavid Malone 			if (*controlp == NULL) {
1494426da3bcSAlfred Perlstein 				FILEDESC_UNLOCK(fdescp);
14952bc21ed9SDavid Malone 				error = E2BIG;
14962bc21ed9SDavid Malone 				goto out;
14978692c025SYoshinobu Inoue 			}
14988692c025SYoshinobu Inoue 
14992bc21ed9SDavid Malone 			fdp = data;
15002bc21ed9SDavid Malone 			rp = (struct file **)
15012bc21ed9SDavid Malone 			    CMSG_DATA(mtod(*controlp, struct cmsghdr *));
15028692c025SYoshinobu Inoue 			for (i = 0; i < oldfds; i++) {
15038692c025SYoshinobu Inoue 				fp = fdescp->fd_ofiles[*fdp++];
1504df8bae1dSRodney W. Grimes 				*rp++ = fp;
1505426da3bcSAlfred Perlstein 				FILE_LOCK(fp);
1506df8bae1dSRodney W. Grimes 				fp->f_count++;
1507df8bae1dSRodney W. Grimes 				fp->f_msgcount++;
1508426da3bcSAlfred Perlstein 				FILE_UNLOCK(fp);
1509df8bae1dSRodney W. Grimes 				unp_rights++;
1510df8bae1dSRodney W. Grimes 			}
1511426da3bcSAlfred Perlstein 			FILEDESC_UNLOCK(fdescp);
15122bc21ed9SDavid Malone 			break;
15132bc21ed9SDavid Malone 
15142bc21ed9SDavid Malone 		case SCM_TIMESTAMP:
15152bc21ed9SDavid Malone 			*controlp = sbcreatecontrol(NULL, sizeof(*tv),
15162bc21ed9SDavid Malone 			    SCM_TIMESTAMP, SOL_SOCKET);
15172bc21ed9SDavid Malone 			if (*controlp == NULL) {
15182bc21ed9SDavid Malone 				error = ENOBUFS;
15192bc21ed9SDavid Malone 				goto out;
15208692c025SYoshinobu Inoue 			}
15212bc21ed9SDavid Malone 			tv = (struct timeval *)
15222bc21ed9SDavid Malone 			    CMSG_DATA(mtod(*controlp, struct cmsghdr *));
15232bc21ed9SDavid Malone 			microtime(tv);
15242bc21ed9SDavid Malone 			break;
15252bc21ed9SDavid Malone 
15262bc21ed9SDavid Malone 		default:
15272bc21ed9SDavid Malone 			error = EINVAL;
15282bc21ed9SDavid Malone 			goto out;
15292bc21ed9SDavid Malone 		}
15302bc21ed9SDavid Malone 
15312bc21ed9SDavid Malone 		controlp = &(*controlp)->m_next;
15322bc21ed9SDavid Malone 
15332bc21ed9SDavid Malone 		if (CMSG_SPACE(datalen) < clen) {
15342bc21ed9SDavid Malone 			clen -= CMSG_SPACE(datalen);
15352bc21ed9SDavid Malone 			cm = (struct cmsghdr *)
15362bc21ed9SDavid Malone 			    ((caddr_t)cm + CMSG_SPACE(datalen));
15372bc21ed9SDavid Malone 		} else {
15382bc21ed9SDavid Malone 			clen = 0;
15392bc21ed9SDavid Malone 			cm = NULL;
15402bc21ed9SDavid Malone 		}
15412bc21ed9SDavid Malone 	}
15422bc21ed9SDavid Malone 
15432bc21ed9SDavid Malone out:
15442bc21ed9SDavid Malone 	m_freem(control);
15452bc21ed9SDavid Malone 
15462bc21ed9SDavid Malone 	return (error);
1547df8bae1dSRodney W. Grimes }
1548df8bae1dSRodney W. Grimes 
15496a2989fdSMatthew N. Dodd struct mbuf *
15506a2989fdSMatthew N. Dodd unp_addsockcred(struct thread *td, struct mbuf *control)
15516a2989fdSMatthew N. Dodd {
155270df31f4SMaxim Konovalov 	struct mbuf *m, *n, *n_prev;
15536a2989fdSMatthew N. Dodd 	struct sockcred *sc;
155470df31f4SMaxim Konovalov 	const struct cmsghdr *cm;
15556a2989fdSMatthew N. Dodd 	int ngroups;
15566a2989fdSMatthew N. Dodd 	int i;
15576a2989fdSMatthew N. Dodd 
15586a2989fdSMatthew N. Dodd 	ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX);
15596a2989fdSMatthew N. Dodd 
15606a2989fdSMatthew N. Dodd 	m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET);
15616a2989fdSMatthew N. Dodd 	if (m == NULL)
15626a2989fdSMatthew N. Dodd 		return (control);
15636a2989fdSMatthew N. Dodd 
15646a2989fdSMatthew N. Dodd 	sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *));
15656a2989fdSMatthew N. Dodd 	sc->sc_uid = td->td_ucred->cr_ruid;
15666a2989fdSMatthew N. Dodd 	sc->sc_euid = td->td_ucred->cr_uid;
15676a2989fdSMatthew N. Dodd 	sc->sc_gid = td->td_ucred->cr_rgid;
15686a2989fdSMatthew N. Dodd 	sc->sc_egid = td->td_ucred->cr_gid;
15696a2989fdSMatthew N. Dodd 	sc->sc_ngroups = ngroups;
15706a2989fdSMatthew N. Dodd 	for (i = 0; i < sc->sc_ngroups; i++)
15716a2989fdSMatthew N. Dodd 		sc->sc_groups[i] = td->td_ucred->cr_groups[i];
15726a2989fdSMatthew N. Dodd 
15736a2989fdSMatthew N. Dodd 	/*
15741c381b19SRobert Watson 	 * Unlink SCM_CREDS control messages (struct cmsgcred), since just
15751c381b19SRobert Watson 	 * created SCM_CREDS control message (struct sockcred) has another
15761c381b19SRobert Watson 	 * format.
15776a2989fdSMatthew N. Dodd 	 */
157870df31f4SMaxim Konovalov 	if (control != NULL)
157970df31f4SMaxim Konovalov 		for (n = control, n_prev = NULL; n != NULL;) {
158070df31f4SMaxim Konovalov 			cm = mtod(n, struct cmsghdr *);
158170df31f4SMaxim Konovalov     			if (cm->cmsg_level == SOL_SOCKET &&
158270df31f4SMaxim Konovalov 			    cm->cmsg_type == SCM_CREDS) {
158370df31f4SMaxim Konovalov     				if (n_prev == NULL)
158470df31f4SMaxim Konovalov 					control = n->m_next;
158570df31f4SMaxim Konovalov 				else
158670df31f4SMaxim Konovalov 					n_prev->m_next = n->m_next;
158770df31f4SMaxim Konovalov 				n = m_free(n);
158870df31f4SMaxim Konovalov 			} else {
158970df31f4SMaxim Konovalov 				n_prev = n;
159070df31f4SMaxim Konovalov 				n = n->m_next;
159170df31f4SMaxim Konovalov 			}
159270df31f4SMaxim Konovalov 		}
15936a2989fdSMatthew N. Dodd 
159470df31f4SMaxim Konovalov 	/* Prepend it to the head. */
159570df31f4SMaxim Konovalov 	m->m_next = control;
159670df31f4SMaxim Konovalov 
159770df31f4SMaxim Konovalov 	return (m);
15986a2989fdSMatthew N. Dodd }
15996a2989fdSMatthew N. Dodd 
1600161a0c7cSRobert Watson /*
1601a0ec558aSRobert Watson  * unp_defer indicates whether additional work has been defered for a future
1602a0ec558aSRobert Watson  * pass through unp_gc().  It is thread local and does not require explicit
1603a0ec558aSRobert Watson  * synchronization.
1604161a0c7cSRobert Watson  */
1605a0ec558aSRobert Watson static int	unp_defer;
1606a0ec558aSRobert Watson 
1607a0ec558aSRobert Watson static int unp_taskcount;
1608a0ec558aSRobert Watson SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0, "");
1609a0ec558aSRobert Watson 
1610a0ec558aSRobert Watson static int unp_recycled;
1611a0ec558aSRobert Watson SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0, "");
1612df8bae1dSRodney W. Grimes 
1613f708ef1bSPoul-Henning Kamp static void
1614a0ec558aSRobert Watson unp_gc(__unused void *arg, int pending)
1615df8bae1dSRodney W. Grimes {
1616892af6b9SRobert Watson 	struct file *fp, *nextfp;
1617892af6b9SRobert Watson 	struct socket *so;
1618df8bae1dSRodney W. Grimes 	struct file **extra_ref, **fpp;
1619df8bae1dSRodney W. Grimes 	int nunref, i;
162095f004dcSAlfred Perlstein 	int nfiles_snap;
162195f004dcSAlfred Perlstein 	int nfiles_slack = 20;
1622df8bae1dSRodney W. Grimes 
1623a0ec558aSRobert Watson 	unp_taskcount++;
1624df8bae1dSRodney W. Grimes 	unp_defer = 0;
1625ed5b7817SJulian Elischer 	/*
16261c381b19SRobert Watson 	 * Before going through all this, set all FDs to be NOT defered and
16271c381b19SRobert Watson 	 * NOT externally accessible.
1628ed5b7817SJulian Elischer 	 */
1629426da3bcSAlfred Perlstein 	sx_slock(&filelist_lock);
16302e3c8fcbSPoul-Henning Kamp 	LIST_FOREACH(fp, &filehead, f_list)
1631426da3bcSAlfred Perlstein 		fp->f_gcflag &= ~(FMARK|FDEFER);
1632df8bae1dSRodney W. Grimes 	do {
16335bb84bc8SRobert Watson 		KASSERT(unp_defer >= 0, ("unp_gc: unp_defer %d", unp_defer));
16342e3c8fcbSPoul-Henning Kamp 		LIST_FOREACH(fp, &filehead, f_list) {
1635426da3bcSAlfred Perlstein 			FILE_LOCK(fp);
1636ed5b7817SJulian Elischer 			/*
1637a0ec558aSRobert Watson 			 * If the file is not open, skip it -- could be a
1638a0ec558aSRobert Watson 			 * file in the process of being opened, or in the
1639a0ec558aSRobert Watson 			 * process of being closed.  If the file is
1640a0ec558aSRobert Watson 			 * "closing", it may have been marked for deferred
1641a0ec558aSRobert Watson 			 * consideration.  Clear the flag now if so.
1642ed5b7817SJulian Elischer 			 */
1643426da3bcSAlfred Perlstein 			if (fp->f_count == 0) {
1644a0ec558aSRobert Watson 				if (fp->f_gcflag & FDEFER)
1645a0ec558aSRobert Watson 					unp_defer--;
1646a0ec558aSRobert Watson 				fp->f_gcflag &= ~(FMARK|FDEFER);
1647426da3bcSAlfred Perlstein 				FILE_UNLOCK(fp);
1648df8bae1dSRodney W. Grimes 				continue;
1649426da3bcSAlfred Perlstein 			}
1650ed5b7817SJulian Elischer 			/*
16511c381b19SRobert Watson 			 * If we already marked it as 'defer' in a previous
16521c381b19SRobert Watson 			 * pass, then try process it this time and un-mark
16531c381b19SRobert Watson 			 * it.
1654ed5b7817SJulian Elischer 			 */
1655426da3bcSAlfred Perlstein 			if (fp->f_gcflag & FDEFER) {
1656426da3bcSAlfred Perlstein 				fp->f_gcflag &= ~FDEFER;
1657df8bae1dSRodney W. Grimes 				unp_defer--;
1658df8bae1dSRodney W. Grimes 			} else {
1659ed5b7817SJulian Elischer 				/*
1660ed5b7817SJulian Elischer 				 * if it's not defered, then check if it's
1661ed5b7817SJulian Elischer 				 * already marked.. if so skip it
1662ed5b7817SJulian Elischer 				 */
1663426da3bcSAlfred Perlstein 				if (fp->f_gcflag & FMARK) {
1664426da3bcSAlfred Perlstein 					FILE_UNLOCK(fp);
1665df8bae1dSRodney W. Grimes 					continue;
1666426da3bcSAlfred Perlstein 				}
1667ed5b7817SJulian Elischer 				/*
16681c381b19SRobert Watson 				 * If all references are from messages in
16691c381b19SRobert Watson 				 * transit, then skip it. it's not externally
16701c381b19SRobert Watson 				 * accessible.
1671ed5b7817SJulian Elischer 				 */
1672426da3bcSAlfred Perlstein 				if (fp->f_count == fp->f_msgcount) {
1673426da3bcSAlfred Perlstein 					FILE_UNLOCK(fp);
1674df8bae1dSRodney W. Grimes 					continue;
1675426da3bcSAlfred Perlstein 				}
1676ed5b7817SJulian Elischer 				/*
1677ed5b7817SJulian Elischer 				 * If it got this far then it must be
1678ed5b7817SJulian Elischer 				 * externally accessible.
1679ed5b7817SJulian Elischer 				 */
1680426da3bcSAlfred Perlstein 				fp->f_gcflag |= FMARK;
1681df8bae1dSRodney W. Grimes 			}
1682ed5b7817SJulian Elischer 			/*
16831c381b19SRobert Watson 			 * Either it was defered, or it is externally
16841c381b19SRobert Watson 			 * accessible and not already marked so.  Now check
16851c381b19SRobert Watson 			 * if it is possibly one of OUR sockets.
1686ed5b7817SJulian Elischer 			 */
1687df8bae1dSRodney W. Grimes 			if (fp->f_type != DTYPE_SOCKET ||
168848e3128bSMatthew Dillon 			    (so = fp->f_data) == NULL) {
1689426da3bcSAlfred Perlstein 				FILE_UNLOCK(fp);
1690df8bae1dSRodney W. Grimes 				continue;
1691426da3bcSAlfred Perlstein 			}
1692426da3bcSAlfred Perlstein 			FILE_UNLOCK(fp);
1693748e0b0aSGarrett Wollman 			if (so->so_proto->pr_domain != &localdomain ||
1694df8bae1dSRodney W. Grimes 			    (so->so_proto->pr_flags&PR_RIGHTS) == 0)
1695df8bae1dSRodney W. Grimes 				continue;
1696ed5b7817SJulian Elischer 			/*
16971c381b19SRobert Watson 			 * So, Ok, it's one of our sockets and it IS
16981c381b19SRobert Watson 			 * externally accessible (or was defered).  Now we
16991c381b19SRobert Watson 			 * look to see if we hold any file descriptors in its
1700ed5b7817SJulian Elischer 			 * message buffers. Follow those links and mark them
1701ed5b7817SJulian Elischer 			 * as accessible too.
1702ed5b7817SJulian Elischer 			 */
17037717cf07SRobert Watson 			SOCKBUF_LOCK(&so->so_rcv);
1704df8bae1dSRodney W. Grimes 			unp_scan(so->so_rcv.sb_mb, unp_mark);
17057717cf07SRobert Watson 			SOCKBUF_UNLOCK(&so->so_rcv);
1706df8bae1dSRodney W. Grimes 		}
1707df8bae1dSRodney W. Grimes 	} while (unp_defer);
1708426da3bcSAlfred Perlstein 	sx_sunlock(&filelist_lock);
1709df8bae1dSRodney W. Grimes 	/*
1710a0ec558aSRobert Watson 	 * XXXRW: The following comments need updating for a post-SMPng and
1711a0ec558aSRobert Watson 	 * deferred unp_gc() world, but are still generally accurate.
1712a0ec558aSRobert Watson 	 *
17131c381b19SRobert Watson 	 * We grab an extra reference to each of the file table entries that
17141c381b19SRobert Watson 	 * are not otherwise accessible and then free the rights that are
17151c381b19SRobert Watson 	 * stored in messages on them.
1716df8bae1dSRodney W. Grimes 	 *
1717df8bae1dSRodney W. Grimes 	 * The bug in the orginal code is a little tricky, so I'll describe
1718df8bae1dSRodney W. Grimes 	 * what's wrong with it here.
1719df8bae1dSRodney W. Grimes 	 *
1720df8bae1dSRodney W. Grimes 	 * It is incorrect to simply unp_discard each entry for f_msgcount
1721df8bae1dSRodney W. Grimes 	 * times -- consider the case of sockets A and B that contain
1722df8bae1dSRodney W. Grimes 	 * references to each other.  On a last close of some other socket,
1723df8bae1dSRodney W. Grimes 	 * we trigger a gc since the number of outstanding rights (unp_rights)
1724a0ec558aSRobert Watson 	 * is non-zero.  If during the sweep phase the gc code unp_discards,
1725df8bae1dSRodney W. Grimes 	 * we end up doing a (full) closef on the descriptor.  A closef on A
1726df8bae1dSRodney W. Grimes 	 * results in the following chain.  Closef calls soo_close, which
1727df8bae1dSRodney W. Grimes 	 * calls soclose.   Soclose calls first (through the switch
1728df8bae1dSRodney W. Grimes 	 * uipc_usrreq) unp_detach, which re-invokes unp_gc.  Unp_gc simply
17291c381b19SRobert Watson 	 * returns because the previous instance had set unp_gcing, and we
17301c381b19SRobert Watson 	 * return all the way back to soclose, which marks the socket with
17311c381b19SRobert Watson 	 * SS_NOFDREF, and then calls sofree.  Sofree calls sorflush to free
17321c381b19SRobert Watson 	 * up the rights that are queued in messages on the socket A, i.e.,
17331c381b19SRobert Watson 	 * the reference on B.  The sorflush calls via the dom_dispose switch
17341c381b19SRobert Watson 	 * unp_dispose, which unp_scans with unp_discard.  This second
1735df8bae1dSRodney W. Grimes 	 * instance of unp_discard just calls closef on B.
1736df8bae1dSRodney W. Grimes 	 *
1737df8bae1dSRodney W. Grimes 	 * Well, a similar chain occurs on B, resulting in a sorflush on B,
1738df8bae1dSRodney W. Grimes 	 * which results in another closef on A.  Unfortunately, A is already
1739df8bae1dSRodney W. Grimes 	 * being closed, and the descriptor has already been marked with
1740df8bae1dSRodney W. Grimes 	 * SS_NOFDREF, and soclose panics at this point.
1741df8bae1dSRodney W. Grimes 	 *
1742df8bae1dSRodney W. Grimes 	 * Here, we first take an extra reference to each inaccessible
17431c381b19SRobert Watson 	 * descriptor.  Then, we call sorflush ourself, since we know it is a
17441c381b19SRobert Watson 	 * Unix domain socket anyhow.  After we destroy all the rights
17451c381b19SRobert Watson 	 * carried in messages, we do a last closef to get rid of our extra
17461c381b19SRobert Watson 	 * reference.  This is the last close, and the unp_detach etc will
17471c381b19SRobert Watson 	 * shut down the socket.
1748df8bae1dSRodney W. Grimes 	 *
1749df8bae1dSRodney W. Grimes 	 * 91/09/19, bsy@cs.cmu.edu
1750df8bae1dSRodney W. Grimes 	 */
175195f004dcSAlfred Perlstein again:
1752e4643c73SPoul-Henning Kamp 	nfiles_snap = openfiles + nfiles_slack;	/* some slack */
175395f004dcSAlfred Perlstein 	extra_ref = malloc(nfiles_snap * sizeof(struct file *), M_TEMP,
175495f004dcSAlfred Perlstein 	    M_WAITOK);
1755426da3bcSAlfred Perlstein 	sx_slock(&filelist_lock);
1756e4643c73SPoul-Henning Kamp 	if (nfiles_snap < openfiles) {
175795f004dcSAlfred Perlstein 		sx_sunlock(&filelist_lock);
175895f004dcSAlfred Perlstein 		free(extra_ref, M_TEMP);
175995f004dcSAlfred Perlstein 		nfiles_slack += 20;
176095f004dcSAlfred Perlstein 		goto again;
176195f004dcSAlfred Perlstein 	}
1762fc3fcacfSRobert Watson 	for (nunref = 0, fp = LIST_FIRST(&filehead), fpp = extra_ref;
1763fc3fcacfSRobert Watson 	    fp != NULL; fp = nextfp) {
17642e3c8fcbSPoul-Henning Kamp 		nextfp = LIST_NEXT(fp, f_list);
1765426da3bcSAlfred Perlstein 		FILE_LOCK(fp);
1766ed5b7817SJulian Elischer 		/*
1767ed5b7817SJulian Elischer 		 * If it's not open, skip it
1768ed5b7817SJulian Elischer 		 */
1769426da3bcSAlfred Perlstein 		if (fp->f_count == 0) {
1770426da3bcSAlfred Perlstein 			FILE_UNLOCK(fp);
1771df8bae1dSRodney W. Grimes 			continue;
1772426da3bcSAlfred Perlstein 		}
1773ed5b7817SJulian Elischer 		/*
1774ed5b7817SJulian Elischer 		 * If all refs are from msgs, and it's not marked accessible
17751c381b19SRobert Watson 		 * then it must be referenced from some unreachable cycle of
17761c381b19SRobert Watson 		 * (shut-down) FDs, so include it in our list of FDs to
17771c381b19SRobert Watson 		 * remove.
1778ed5b7817SJulian Elischer 		 */
1779426da3bcSAlfred Perlstein 		if (fp->f_count == fp->f_msgcount && !(fp->f_gcflag & FMARK)) {
1780df8bae1dSRodney W. Grimes 			*fpp++ = fp;
1781df8bae1dSRodney W. Grimes 			nunref++;
1782df8bae1dSRodney W. Grimes 			fp->f_count++;
1783df8bae1dSRodney W. Grimes 		}
1784426da3bcSAlfred Perlstein 		FILE_UNLOCK(fp);
1785df8bae1dSRodney W. Grimes 	}
1786426da3bcSAlfred Perlstein 	sx_sunlock(&filelist_lock);
1787ed5b7817SJulian Elischer 	/*
17881c381b19SRobert Watson 	 * For each FD on our hit list, do the following two things:
1789ed5b7817SJulian Elischer 	 */
17901c7c3c6aSMatthew Dillon 	for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) {
17911c7c3c6aSMatthew Dillon 		struct file *tfp = *fpp;
1792426da3bcSAlfred Perlstein 		FILE_LOCK(tfp);
1793cd72f218SMatthew Dillon 		if (tfp->f_type == DTYPE_SOCKET &&
179448e3128bSMatthew Dillon 		    tfp->f_data != NULL) {
1795426da3bcSAlfred Perlstein 			FILE_UNLOCK(tfp);
179648e3128bSMatthew Dillon 			sorflush(tfp->f_data);
1797e5aeaa0cSDag-Erling Smørgrav 		} else {
1798426da3bcSAlfred Perlstein 			FILE_UNLOCK(tfp);
17991c7c3c6aSMatthew Dillon 		}
1800e5aeaa0cSDag-Erling Smørgrav 	}
1801a0ec558aSRobert Watson 	for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) {
1802b40ce416SJulian Elischer 		closef(*fpp, (struct thread *) NULL);
1803a0ec558aSRobert Watson 		unp_recycled++;
1804a0ec558aSRobert Watson 	}
1805210a5a71SAlfred Perlstein 	free(extra_ref, M_TEMP);
1806df8bae1dSRodney W. Grimes }
1807df8bae1dSRodney W. Grimes 
180826f9a767SRodney W. Grimes void
1809892af6b9SRobert Watson unp_dispose(struct mbuf *m)
1810df8bae1dSRodney W. Grimes {
1811996c772fSJohn Dyson 
1812df8bae1dSRodney W. Grimes 	if (m)
1813df8bae1dSRodney W. Grimes 		unp_scan(m, unp_discard);
1814df8bae1dSRodney W. Grimes }
1815df8bae1dSRodney W. Grimes 
18160c1bb4fbSDima Dorfman static int
1817d374e81eSRobert Watson unp_listen(struct socket *so, struct unpcb *unp, int backlog,
1818d374e81eSRobert Watson     struct thread *td)
18190c1bb4fbSDima Dorfman {
18200daccb9cSRobert Watson 	int error;
18210daccb9cSRobert Watson 
18220d9ce3a1SRobert Watson 	UNP_LOCK_ASSERT();
18230c1bb4fbSDima Dorfman 
18240daccb9cSRobert Watson 	SOCK_LOCK(so);
18250daccb9cSRobert Watson 	error = solisten_proto_check(so);
18260daccb9cSRobert Watson 	if (error == 0) {
18276f105b34SJohn Baldwin 		cru2x(td->td_ucred, &unp->unp_peercred);
18280c1bb4fbSDima Dorfman 		unp->unp_flags |= UNP_HAVEPCCACHED;
1829d374e81eSRobert Watson 		solisten_proto(so, backlog);
18300daccb9cSRobert Watson 	}
18310daccb9cSRobert Watson 	SOCK_UNLOCK(so);
18320daccb9cSRobert Watson 	return (error);
18330c1bb4fbSDima Dorfman }
18340c1bb4fbSDima Dorfman 
1835f708ef1bSPoul-Henning Kamp static void
1836892af6b9SRobert Watson unp_scan(struct mbuf *m0, void (*op)(struct file *))
1837df8bae1dSRodney W. Grimes {
18382bc21ed9SDavid Malone 	struct mbuf *m;
18392bc21ed9SDavid Malone 	struct file **rp;
18402bc21ed9SDavid Malone 	struct cmsghdr *cm;
18412bc21ed9SDavid Malone 	void *data;
18422bc21ed9SDavid Malone 	int i;
18432bc21ed9SDavid Malone 	socklen_t clen, datalen;
1844df8bae1dSRodney W. Grimes 	int qfds;
1845df8bae1dSRodney W. Grimes 
1846fc3fcacfSRobert Watson 	while (m0 != NULL) {
18472bc21ed9SDavid Malone 		for (m = m0; m; m = m->m_next) {
184812396bdcSDavid Malone 			if (m->m_type != MT_CONTROL)
1849df8bae1dSRodney W. Grimes 				continue;
18502bc21ed9SDavid Malone 
18512bc21ed9SDavid Malone 			cm = mtod(m, struct cmsghdr *);
18522bc21ed9SDavid Malone 			clen = m->m_len;
18532bc21ed9SDavid Malone 
18542bc21ed9SDavid Malone 			while (cm != NULL) {
18552bc21ed9SDavid Malone 				if (sizeof(*cm) > clen || cm->cmsg_len > clen)
18562bc21ed9SDavid Malone 					break;
18572bc21ed9SDavid Malone 
18582bc21ed9SDavid Malone 				data = CMSG_DATA(cm);
18592bc21ed9SDavid Malone 				datalen = (caddr_t)cm + cm->cmsg_len
18602bc21ed9SDavid Malone 				    - (caddr_t)data;
18612bc21ed9SDavid Malone 
18622bc21ed9SDavid Malone 				if (cm->cmsg_level == SOL_SOCKET &&
18632bc21ed9SDavid Malone 				    cm->cmsg_type == SCM_RIGHTS) {
18642bc21ed9SDavid Malone 					qfds = datalen / sizeof (struct file *);
18652bc21ed9SDavid Malone 					rp = data;
1866df8bae1dSRodney W. Grimes 					for (i = 0; i < qfds; i++)
1867df8bae1dSRodney W. Grimes 						(*op)(*rp++);
18682bc21ed9SDavid Malone 				}
18692bc21ed9SDavid Malone 
18702bc21ed9SDavid Malone 				if (CMSG_SPACE(datalen) < clen) {
18712bc21ed9SDavid Malone 					clen -= CMSG_SPACE(datalen);
18722bc21ed9SDavid Malone 					cm = (struct cmsghdr *)
18732bc21ed9SDavid Malone 					    ((caddr_t)cm + CMSG_SPACE(datalen));
18742bc21ed9SDavid Malone 				} else {
18752bc21ed9SDavid Malone 					clen = 0;
18762bc21ed9SDavid Malone 					cm = NULL;
18772bc21ed9SDavid Malone 				}
18782bc21ed9SDavid Malone 			}
1879df8bae1dSRodney W. Grimes 		}
1880df8bae1dSRodney W. Grimes 		m0 = m0->m_act;
1881df8bae1dSRodney W. Grimes 	}
1882df8bae1dSRodney W. Grimes }
1883df8bae1dSRodney W. Grimes 
1884f708ef1bSPoul-Henning Kamp static void
1885892af6b9SRobert Watson unp_mark(struct file *fp)
1886df8bae1dSRodney W. Grimes {
1887426da3bcSAlfred Perlstein 	if (fp->f_gcflag & FMARK)
1888df8bae1dSRodney W. Grimes 		return;
1889df8bae1dSRodney W. Grimes 	unp_defer++;
1890426da3bcSAlfred Perlstein 	fp->f_gcflag |= (FMARK|FDEFER);
1891df8bae1dSRodney W. Grimes }
1892df8bae1dSRodney W. Grimes 
1893f708ef1bSPoul-Henning Kamp static void
1894892af6b9SRobert Watson unp_discard(struct file *fp)
1895df8bae1dSRodney W. Grimes {
1896a0ec558aSRobert Watson 	UNP_LOCK();
1897426da3bcSAlfred Perlstein 	FILE_LOCK(fp);
1898df8bae1dSRodney W. Grimes 	fp->f_msgcount--;
1899df8bae1dSRodney W. Grimes 	unp_rights--;
1900426da3bcSAlfred Perlstein 	FILE_UNLOCK(fp);
1901a0ec558aSRobert Watson 	UNP_UNLOCK();
1902b40ce416SJulian Elischer 	(void) closef(fp, (struct thread *)NULL);
1903df8bae1dSRodney W. Grimes }
1904