xref: /freebsd/sys/kern/uipc_usrreq.c (revision 9b16399225f8acfba675cf0a9312dd9eac563ce3)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1989, 1991, 1993
5  *	The Regents of the University of California. All Rights Reserved.
6  * Copyright (c) 2004-2009 Robert N. M. Watson All Rights Reserved.
7  * Copyright (c) 2018 Matthew Macy
8  * Copyright (c) 2022-2025 Gleb Smirnoff <glebius@FreeBSD.org>
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  */
34 
35 /*
36  * UNIX Domain (Local) Sockets
37  *
38  * This is an implementation of UNIX (local) domain sockets.  Each socket has
39  * an associated struct unpcb (UNIX protocol control block).  Stream sockets
40  * may be connected to 0 or 1 other socket.  Datagram sockets may be
41  * connected to 0, 1, or many other sockets.  Sockets may be created and
42  * connected in pairs (socketpair(2)), or bound/connected to using the file
43  * system name space.  For most purposes, only the receive socket buffer is
44  * used, as sending on one socket delivers directly to the receive socket
45  * buffer of a second socket.
46  *
47  * The implementation is substantially complicated by the fact that
48  * "ancillary data", such as file descriptors or credentials, may be passed
49  * across UNIX domain sockets.  The potential for passing UNIX domain sockets
50  * over other UNIX domain sockets requires the implementation of a simple
51  * garbage collector to find and tear down cycles of disconnected sockets.
52  *
53  * TODO:
54  *	RDM
55  *	rethink name space problems
56  *	need a proper out-of-band
57  */
58 
59 #include "opt_ddb.h"
60 
61 #include <sys/param.h>
62 #include <sys/capsicum.h>
63 #include <sys/domain.h>
64 #include <sys/eventhandler.h>
65 #include <sys/fcntl.h>
66 #include <sys/file.h>
67 #include <sys/filedesc.h>
68 #include <sys/jail.h>
69 #include <sys/kernel.h>
70 #include <sys/lock.h>
71 #include <sys/malloc.h>
72 #include <sys/mbuf.h>
73 #include <sys/mount.h>
74 #include <sys/mutex.h>
75 #include <sys/namei.h>
76 #include <sys/poll.h>
77 #include <sys/proc.h>
78 #include <sys/protosw.h>
79 #include <sys/queue.h>
80 #include <sys/resourcevar.h>
81 #include <sys/rwlock.h>
82 #include <sys/socket.h>
83 #include <sys/socketvar.h>
84 #include <sys/signalvar.h>
85 #include <sys/stat.h>
86 #include <sys/sysent.h>
87 #include <sys/sx.h>
88 #include <sys/sysctl.h>
89 #include <sys/systm.h>
90 #include <sys/taskqueue.h>
91 #include <sys/un.h>
92 #include <sys/unpcb.h>
93 #include <sys/vnode.h>
94 
95 #include <net/vnet.h>
96 
97 #ifdef DDB
98 #include <ddb/ddb.h>
99 #endif
100 
101 #include <security/mac/mac_framework.h>
102 
103 #include <vm/uma.h>
104 
105 MALLOC_DECLARE(M_FILECAPS);
106 
107 static struct domain localdomain;
108 
109 static uma_zone_t	unp_zone;
110 static unp_gen_t	unp_gencnt;	/* (l) */
111 static u_int		unp_count;	/* (l) Count of local sockets. */
112 static ino_t		unp_ino;	/* Prototype for fake inode numbers. */
113 static int		unp_rights;	/* (g) File descriptors in flight. */
114 static struct unp_head	unp_shead;	/* (l) List of stream sockets. */
115 static struct unp_head	unp_dhead;	/* (l) List of datagram sockets. */
116 static struct unp_head	unp_sphead;	/* (l) List of seqpacket sockets. */
117 static struct mtx_pool	*unp_vp_mtxpool;
118 
119 struct unp_defer {
120 	SLIST_ENTRY(unp_defer) ud_link;
121 	struct file *ud_fp;
122 };
123 static SLIST_HEAD(, unp_defer) unp_defers;
124 static int unp_defers_count;
125 
126 static const struct sockaddr	sun_noname = {
127 	.sa_len = sizeof(sun_noname),
128 	.sa_family = AF_LOCAL,
129 };
130 
131 /*
132  * Garbage collection of cyclic file descriptor/socket references occurs
133  * asynchronously in a taskqueue context in order to avoid recursion and
134  * reentrance in the UNIX domain socket, file descriptor, and socket layer
135  * code.  See unp_gc() for a full description.
136  */
137 static struct timeout_task unp_gc_task;
138 
139 /*
140  * The close of unix domain sockets attached as SCM_RIGHTS is
141  * postponed to the taskqueue, to avoid arbitrary recursion depth.
142  * The attached sockets might have another sockets attached.
143  */
144 static struct task	unp_defer_task;
145 
146 /*
147  * SOCK_STREAM and SOCK_SEQPACKET unix(4) sockets fully bypass the send buffer,
148  * however the notion of send buffer still makes sense with them.  Its size is
149  * the amount of space that a send(2) syscall may copyin(9) before checking
150  * with the receive buffer of a peer.  Although not linked anywhere yet,
151  * pointed to by a stack variable, effectively it is a buffer that needs to be
152  * sized.
153  *
154  * SOCK_DGRAM sockets really use the sendspace as the maximum datagram size,
155  * and don't really want to reserve the sendspace.  Their recvspace should be
156  * large enough for at least one max-size datagram plus address.
157  */
158 static u_long	unpst_sendspace = 64*1024;
159 static u_long	unpst_recvspace = 64*1024;
160 static u_long	unpdg_maxdgram = 8*1024;	/* support 8KB syslog msgs */
161 static u_long	unpdg_recvspace = 16*1024;
162 static u_long	unpsp_sendspace = 64*1024;
163 static u_long	unpsp_recvspace = 64*1024;
164 
165 static SYSCTL_NODE(_net, PF_LOCAL, local, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
166     "Local domain");
167 static SYSCTL_NODE(_net_local, SOCK_STREAM, stream,
168     CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
169     "SOCK_STREAM");
170 static SYSCTL_NODE(_net_local, SOCK_DGRAM, dgram,
171     CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
172     "SOCK_DGRAM");
173 static SYSCTL_NODE(_net_local, SOCK_SEQPACKET, seqpacket,
174     CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
175     "SOCK_SEQPACKET");
176 
177 SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW,
178 	   &unpst_sendspace, 0, "Default stream send space.");
179 SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW,
180 	   &unpst_recvspace, 0, "Default stream receive space.");
181 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW,
182 	   &unpdg_maxdgram, 0, "Maximum datagram size.");
183 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW,
184 	   &unpdg_recvspace, 0, "Default datagram receive space.");
185 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, maxseqpacket, CTLFLAG_RW,
186 	   &unpsp_sendspace, 0, "Default seqpacket send space.");
187 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, recvspace, CTLFLAG_RW,
188 	   &unpsp_recvspace, 0, "Default seqpacket receive space.");
189 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0,
190     "File descriptors in flight.");
191 SYSCTL_INT(_net_local, OID_AUTO, deferred, CTLFLAG_RD,
192     &unp_defers_count, 0,
193     "File descriptors deferred to taskqueue for close.");
194 
195 /*
196  * Locking and synchronization:
197  *
198  * Several types of locks exist in the local domain socket implementation:
199  * - a global linkage lock
200  * - a global connection list lock
201  * - the mtxpool lock
202  * - per-unpcb mutexes
203  *
204  * The linkage lock protects the global socket lists, the generation number
205  * counter and garbage collector state.
206  *
207  * The connection list lock protects the list of referring sockets in a datagram
208  * socket PCB.  This lock is also overloaded to protect a global list of
209  * sockets whose buffers contain socket references in the form of SCM_RIGHTS
210  * messages.  To avoid recursion, such references are released by a dedicated
211  * thread.
212  *
213  * The mtxpool lock protects the vnode from being modified while referenced.
214  * Lock ordering rules require that it be acquired before any PCB locks.
215  *
216  * The unpcb lock (unp_mtx) protects the most commonly referenced fields in the
217  * unpcb.  This includes the unp_conn field, which either links two connected
218  * PCBs together (for connected socket types) or points at the destination
219  * socket (for connectionless socket types).  The operations of creating or
220  * destroying a connection therefore involve locking multiple PCBs.  To avoid
221  * lock order reversals, in some cases this involves dropping a PCB lock and
222  * using a reference counter to maintain liveness.
223  *
224  * UNIX domain sockets each have an unpcb hung off of their so_pcb pointer,
225  * allocated in pr_attach() and freed in pr_detach().  The validity of that
226  * pointer is an invariant, so no lock is required to dereference the so_pcb
227  * pointer if a valid socket reference is held by the caller.  In practice,
228  * this is always true during operations performed on a socket.  Each unpcb
229  * has a back-pointer to its socket, unp_socket, which will be stable under
230  * the same circumstances.
231  *
232  * This pointer may only be safely dereferenced as long as a valid reference
233  * to the unpcb is held.  Typically, this reference will be from the socket,
234  * or from another unpcb when the referring unpcb's lock is held (in order
235  * that the reference not be invalidated during use).  For example, to follow
236  * unp->unp_conn->unp_socket, you need to hold a lock on unp_conn to guarantee
237  * that detach is not run clearing unp_socket.
238  *
239  * Blocking with UNIX domain sockets is a tricky issue: unlike most network
240  * protocols, bind() is a non-atomic operation, and connect() requires
241  * potential sleeping in the protocol, due to potentially waiting on local or
242  * distributed file systems.  We try to separate "lookup" operations, which
243  * may sleep, and the IPC operations themselves, which typically can occur
244  * with relative atomicity as locks can be held over the entire operation.
245  *
246  * Another tricky issue is simultaneous multi-threaded or multi-process
247  * access to a single UNIX domain socket.  These are handled by the flags
248  * UNP_CONNECTING and UNP_BINDING, which prevent concurrent connecting or
249  * binding, both of which involve dropping UNIX domain socket locks in order
250  * to perform namei() and other file system operations.
251  */
252 static struct rwlock	unp_link_rwlock;
253 static struct mtx	unp_defers_lock;
254 
255 #define	UNP_LINK_LOCK_INIT()		rw_init(&unp_link_rwlock,	\
256 					    "unp_link_rwlock")
257 
258 #define	UNP_LINK_LOCK_ASSERT()		rw_assert(&unp_link_rwlock,	\
259 					    RA_LOCKED)
260 #define	UNP_LINK_UNLOCK_ASSERT()	rw_assert(&unp_link_rwlock,	\
261 					    RA_UNLOCKED)
262 
263 #define	UNP_LINK_RLOCK()		rw_rlock(&unp_link_rwlock)
264 #define	UNP_LINK_RUNLOCK()		rw_runlock(&unp_link_rwlock)
265 #define	UNP_LINK_WLOCK()		rw_wlock(&unp_link_rwlock)
266 #define	UNP_LINK_WUNLOCK()		rw_wunlock(&unp_link_rwlock)
267 #define	UNP_LINK_WLOCK_ASSERT()		rw_assert(&unp_link_rwlock,	\
268 					    RA_WLOCKED)
269 #define	UNP_LINK_WOWNED()		rw_wowned(&unp_link_rwlock)
270 
271 #define	UNP_DEFERRED_LOCK_INIT()	mtx_init(&unp_defers_lock, \
272 					    "unp_defer", NULL, MTX_DEF)
273 #define	UNP_DEFERRED_LOCK()		mtx_lock(&unp_defers_lock)
274 #define	UNP_DEFERRED_UNLOCK()		mtx_unlock(&unp_defers_lock)
275 
276 #define UNP_REF_LIST_LOCK()		UNP_DEFERRED_LOCK();
277 #define UNP_REF_LIST_UNLOCK()		UNP_DEFERRED_UNLOCK();
278 
279 #define UNP_PCB_LOCK_INIT(unp)		mtx_init(&(unp)->unp_mtx,	\
280 					    "unp", "unp",	\
281 					    MTX_DUPOK|MTX_DEF)
282 #define	UNP_PCB_LOCK_DESTROY(unp)	mtx_destroy(&(unp)->unp_mtx)
283 #define	UNP_PCB_LOCKPTR(unp)		(&(unp)->unp_mtx)
284 #define	UNP_PCB_LOCK(unp)		mtx_lock(&(unp)->unp_mtx)
285 #define	UNP_PCB_TRYLOCK(unp)		mtx_trylock(&(unp)->unp_mtx)
286 #define	UNP_PCB_UNLOCK(unp)		mtx_unlock(&(unp)->unp_mtx)
287 #define	UNP_PCB_OWNED(unp)		mtx_owned(&(unp)->unp_mtx)
288 #define	UNP_PCB_LOCK_ASSERT(unp)	mtx_assert(&(unp)->unp_mtx, MA_OWNED)
289 #define	UNP_PCB_UNLOCK_ASSERT(unp)	mtx_assert(&(unp)->unp_mtx, MA_NOTOWNED)
290 
291 static int	uipc_connect2(struct socket *, struct socket *);
292 static int	uipc_ctloutput(struct socket *, struct sockopt *);
293 static int	unp_connect(struct socket *, struct sockaddr *,
294 		    struct thread *);
295 static int	unp_connectat(int, struct socket *, struct sockaddr *,
296 		    struct thread *, bool);
297 static void	unp_connect2(struct socket *, struct socket *, bool);
298 static void	unp_disconnect(struct unpcb *unp, struct unpcb *unp2);
299 static void	unp_dispose(struct socket *so);
300 static void	unp_drop(struct unpcb *);
301 static void	unp_gc(__unused void *, int);
302 static void	unp_scan(struct mbuf *, void (*)(struct filedescent **, int));
303 static void	unp_discard(struct file *);
304 static void	unp_freerights(struct filedescent **, int);
305 static int	unp_internalize(struct mbuf *, struct mchain *,
306 		    struct thread *, int *);
307 static void	unp_internalize_fp(struct file *);
308 static int	unp_externalize(const struct socket *, struct mbuf *,
309 		    struct mbuf **, int);
310 static int	unp_externalize_fp(struct file *);
311 static void	unp_addsockcred(struct thread *, struct mchain *, int);
312 static void	unp_process_defers(void * __unused, int);
313 
314 static void	uipc_wrknl_lock(void *);
315 static void	uipc_wrknl_unlock(void *);
316 static void	uipc_wrknl_assert_lock(void *, int);
317 
318 static void
319 unp_pcb_hold(struct unpcb *unp)
320 {
321 	u_int old __unused;
322 
323 	old = refcount_acquire(&unp->unp_refcount);
324 	KASSERT(old > 0, ("%s: unpcb %p has no references", __func__, unp));
325 }
326 
327 static __result_use_check bool
328 unp_pcb_rele(struct unpcb *unp)
329 {
330 	bool ret;
331 
332 	UNP_PCB_LOCK_ASSERT(unp);
333 
334 	if ((ret = refcount_release(&unp->unp_refcount))) {
335 		UNP_PCB_UNLOCK(unp);
336 		UNP_PCB_LOCK_DESTROY(unp);
337 		uma_zfree(unp_zone, unp);
338 	}
339 	return (ret);
340 }
341 
342 static void
343 unp_pcb_rele_notlast(struct unpcb *unp)
344 {
345 	bool ret __unused;
346 
347 	ret = refcount_release(&unp->unp_refcount);
348 	KASSERT(!ret, ("%s: unpcb %p has no references", __func__, unp));
349 }
350 
351 static void
352 unp_pcb_lock_pair(struct unpcb *unp, struct unpcb *unp2)
353 {
354 	UNP_PCB_UNLOCK_ASSERT(unp);
355 	UNP_PCB_UNLOCK_ASSERT(unp2);
356 
357 	if (unp == unp2) {
358 		UNP_PCB_LOCK(unp);
359 	} else if ((uintptr_t)unp2 > (uintptr_t)unp) {
360 		UNP_PCB_LOCK(unp);
361 		UNP_PCB_LOCK(unp2);
362 	} else {
363 		UNP_PCB_LOCK(unp2);
364 		UNP_PCB_LOCK(unp);
365 	}
366 }
367 
368 static void
369 unp_pcb_unlock_pair(struct unpcb *unp, struct unpcb *unp2)
370 {
371 	UNP_PCB_UNLOCK(unp);
372 	if (unp != unp2)
373 		UNP_PCB_UNLOCK(unp2);
374 }
375 
376 /*
377  * Try to lock the connected peer of an already locked socket.  In some cases
378  * this requires that we unlock the current socket.  The pairbusy counter is
379  * used to block concurrent connection attempts while the lock is dropped.  The
380  * caller must be careful to revalidate PCB state.
381  */
382 static struct unpcb *
383 unp_pcb_lock_peer(struct unpcb *unp)
384 {
385 	struct unpcb *unp2;
386 
387 	UNP_PCB_LOCK_ASSERT(unp);
388 	unp2 = unp->unp_conn;
389 	if (unp2 == NULL)
390 		return (NULL);
391 	if (__predict_false(unp == unp2))
392 		return (unp);
393 
394 	UNP_PCB_UNLOCK_ASSERT(unp2);
395 
396 	if (__predict_true(UNP_PCB_TRYLOCK(unp2)))
397 		return (unp2);
398 	if ((uintptr_t)unp2 > (uintptr_t)unp) {
399 		UNP_PCB_LOCK(unp2);
400 		return (unp2);
401 	}
402 	unp->unp_pairbusy++;
403 	unp_pcb_hold(unp2);
404 	UNP_PCB_UNLOCK(unp);
405 
406 	UNP_PCB_LOCK(unp2);
407 	UNP_PCB_LOCK(unp);
408 	KASSERT(unp->unp_conn == unp2 || unp->unp_conn == NULL,
409 	    ("%s: socket %p was reconnected", __func__, unp));
410 	if (--unp->unp_pairbusy == 0 && (unp->unp_flags & UNP_WAITING) != 0) {
411 		unp->unp_flags &= ~UNP_WAITING;
412 		wakeup(unp);
413 	}
414 	if (unp_pcb_rele(unp2)) {
415 		/* unp2 is unlocked. */
416 		return (NULL);
417 	}
418 	if (unp->unp_conn == NULL) {
419 		UNP_PCB_UNLOCK(unp2);
420 		return (NULL);
421 	}
422 	return (unp2);
423 }
424 
425 /*
426  * Try to lock peer of our socket for purposes of sending data to it.
427  */
428 static int
429 uipc_lock_peer(struct socket *so, struct unpcb **unp2)
430 {
431 	struct unpcb *unp;
432 	int error;
433 
434 	unp = sotounpcb(so);
435 	UNP_PCB_LOCK(unp);
436 	*unp2 = unp_pcb_lock_peer(unp);
437 	if (__predict_false(so->so_error != 0)) {
438 		error = so->so_error;
439 		so->so_error = 0;
440 		UNP_PCB_UNLOCK(unp);
441 		if (*unp2 != NULL)
442 			UNP_PCB_UNLOCK(*unp2);
443 		return (error);
444 	}
445 	if (__predict_false(*unp2 == NULL)) {
446 		/*
447 		 * Different error code for a previously connected socket and
448 		 * a never connected one.  The SS_ISDISCONNECTED is set in the
449 		 * unp_soisdisconnected() and is synchronized by the pcb lock.
450 		 */
451 		error = so->so_state & SS_ISDISCONNECTED ? EPIPE : ENOTCONN;
452 		UNP_PCB_UNLOCK(unp);
453 		return (error);
454 	}
455 	UNP_PCB_UNLOCK(unp);
456 
457 	return (0);
458 }
459 
460 static void
461 uipc_abort(struct socket *so)
462 {
463 	struct unpcb *unp, *unp2;
464 
465 	unp = sotounpcb(so);
466 	KASSERT(unp != NULL, ("uipc_abort: unp == NULL"));
467 	UNP_PCB_UNLOCK_ASSERT(unp);
468 
469 	UNP_PCB_LOCK(unp);
470 	unp2 = unp->unp_conn;
471 	if (unp2 != NULL) {
472 		unp_pcb_hold(unp2);
473 		UNP_PCB_UNLOCK(unp);
474 		unp_drop(unp2);
475 	} else
476 		UNP_PCB_UNLOCK(unp);
477 }
478 
479 static int
480 uipc_attach(struct socket *so, int proto, struct thread *td)
481 {
482 	u_long sendspace, recvspace;
483 	struct unpcb *unp;
484 	int error, rcvmtxopts;
485 	bool locked;
486 
487 	KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL"));
488 	switch (so->so_type) {
489 	case SOCK_DGRAM:
490 		STAILQ_INIT(&so->so_rcv.uxdg_mb);
491 		STAILQ_INIT(&so->so_snd.uxdg_mb);
492 		TAILQ_INIT(&so->so_rcv.uxdg_conns);
493 		/*
494 		 * Since send buffer is either bypassed or is a part
495 		 * of one-to-many receive buffer, we assign both space
496 		 * limits to unpdg_recvspace.
497 		 */
498 		sendspace = recvspace = unpdg_recvspace;
499 		rcvmtxopts = 0;
500 		break;
501 
502 	case SOCK_STREAM:
503 		sendspace = unpst_sendspace;
504 		recvspace = unpst_recvspace;
505 		goto common;
506 
507 	case SOCK_SEQPACKET:
508 		sendspace = unpsp_sendspace;
509 		recvspace = unpsp_recvspace;
510 common:
511 		rcvmtxopts = MTX_DUPOK;
512 		knlist_init(&so->so_wrsel.si_note, so, uipc_wrknl_lock,
513 		    uipc_wrknl_unlock, uipc_wrknl_assert_lock);
514 		STAILQ_INIT(&so->so_rcv.uxst_mbq);
515 		break;
516 	default:
517 		panic("uipc_attach");
518 	}
519 	mtx_init(&so->so_rcv_mtx, "unix so_rcv", NULL, MTX_DEF | rcvmtxopts);
520 	mtx_init(&so->so_snd_mtx, "unix so_snd", NULL, MTX_DEF);
521 	error = soreserve(so, sendspace, recvspace);
522 	if (error)
523 		return (error);
524 	unp = uma_zalloc(unp_zone, M_NOWAIT | M_ZERO);
525 	if (unp == NULL)
526 		return (ENOBUFS);
527 	LIST_INIT(&unp->unp_refs);
528 	UNP_PCB_LOCK_INIT(unp);
529 	unp->unp_socket = so;
530 	so->so_pcb = unp;
531 	so->so_options |= SO_PASSRIGHTS;
532 	refcount_init(&unp->unp_refcount, 1);
533 	unp->unp_mode = ACCESSPERMS;
534 
535 	if ((locked = UNP_LINK_WOWNED()) == false)
536 		UNP_LINK_WLOCK();
537 
538 	unp->unp_gencnt = ++unp_gencnt;
539 	unp->unp_ino = ++unp_ino;
540 	unp_count++;
541 	switch (so->so_type) {
542 	case SOCK_STREAM:
543 		LIST_INSERT_HEAD(&unp_shead, unp, unp_link);
544 		break;
545 
546 	case SOCK_DGRAM:
547 		LIST_INSERT_HEAD(&unp_dhead, unp, unp_link);
548 		break;
549 
550 	case SOCK_SEQPACKET:
551 		LIST_INSERT_HEAD(&unp_sphead, unp, unp_link);
552 		break;
553 
554 	default:
555 		panic("uipc_attach");
556 	}
557 
558 	if (locked == false)
559 		UNP_LINK_WUNLOCK();
560 
561 	return (0);
562 }
563 
564 static int
565 uipc_bindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
566 {
567 	struct sockaddr_un *soun = (struct sockaddr_un *)nam;
568 	struct vattr vattr;
569 	int error, namelen;
570 	struct nameidata nd;
571 	struct unpcb *unp;
572 	struct vnode *vp;
573 	struct mount *mp;
574 	cap_rights_t rights;
575 	char *buf;
576 	mode_t mode;
577 
578 	if (nam->sa_family != AF_UNIX)
579 		return (EAFNOSUPPORT);
580 
581 	unp = sotounpcb(so);
582 	KASSERT(unp != NULL, ("uipc_bind: unp == NULL"));
583 
584 	if (soun->sun_len > sizeof(struct sockaddr_un))
585 		return (EINVAL);
586 	namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path);
587 	if (namelen <= 0)
588 		return (EINVAL);
589 
590 	/*
591 	 * We don't allow simultaneous bind() calls on a single UNIX domain
592 	 * socket, so flag in-progress operations, and return an error if an
593 	 * operation is already in progress.
594 	 *
595 	 * Historically, we have not allowed a socket to be rebound, so this
596 	 * also returns an error.  Not allowing re-binding simplifies the
597 	 * implementation and avoids a great many possible failure modes.
598 	 */
599 	UNP_PCB_LOCK(unp);
600 	if (unp->unp_vnode != NULL) {
601 		UNP_PCB_UNLOCK(unp);
602 		return (EINVAL);
603 	}
604 	if (unp->unp_flags & UNP_BINDING) {
605 		UNP_PCB_UNLOCK(unp);
606 		return (EALREADY);
607 	}
608 	unp->unp_flags |= UNP_BINDING;
609 	mode = unp->unp_mode & ~td->td_proc->p_pd->pd_cmask;
610 	UNP_PCB_UNLOCK(unp);
611 
612 	buf = malloc(namelen + 1, M_TEMP, M_WAITOK);
613 	bcopy(soun->sun_path, buf, namelen);
614 	buf[namelen] = 0;
615 
616 restart:
617 	NDINIT_ATRIGHTS(&nd, CREATE, NOFOLLOW | LOCKPARENT | NOCACHE,
618 	    UIO_SYSSPACE, buf, fd, cap_rights_init_one(&rights, CAP_BINDAT));
619 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
620 	error = namei(&nd);
621 	if (error)
622 		goto error;
623 	vp = nd.ni_vp;
624 	if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
625 		NDFREE_PNBUF(&nd);
626 		if (nd.ni_dvp == vp)
627 			vrele(nd.ni_dvp);
628 		else
629 			vput(nd.ni_dvp);
630 		if (vp != NULL) {
631 			vrele(vp);
632 			error = EADDRINUSE;
633 			goto error;
634 		}
635 		error = vn_start_write(NULL, &mp, V_XSLEEP | V_PCATCH);
636 		if (error)
637 			goto error;
638 		goto restart;
639 	}
640 	VATTR_NULL(&vattr);
641 	vattr.va_type = VSOCK;
642 	vattr.va_mode = mode;
643 #ifdef MAC
644 	error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd,
645 	    &vattr);
646 #endif
647 	if (error == 0) {
648 		/*
649 		 * The prior lookup may have left LK_SHARED in cn_lkflags,
650 		 * and VOP_CREATE technically only requires the new vnode to
651 		 * be locked shared. Most filesystems will return the new vnode
652 		 * locked exclusive regardless, but we should explicitly
653 		 * specify that here since we require it and assert to that
654 		 * effect below.
655 		 */
656 		nd.ni_cnd.cn_lkflags = (nd.ni_cnd.cn_lkflags & ~LK_SHARED) |
657 		    LK_EXCLUSIVE;
658 		error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
659 	}
660 	NDFREE_PNBUF(&nd);
661 	if (error) {
662 		VOP_VPUT_PAIR(nd.ni_dvp, NULL, true);
663 		vn_finished_write(mp);
664 		if (error == ERELOOKUP)
665 			goto restart;
666 		goto error;
667 	}
668 	vp = nd.ni_vp;
669 	ASSERT_VOP_ELOCKED(vp, "uipc_bind");
670 	soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK);
671 
672 	UNP_PCB_LOCK(unp);
673 	VOP_UNP_BIND(vp, unp);
674 	unp->unp_vnode = vp;
675 	unp->unp_addr = soun;
676 	unp->unp_flags &= ~UNP_BINDING;
677 	UNP_PCB_UNLOCK(unp);
678 	vref(vp);
679 	VOP_VPUT_PAIR(nd.ni_dvp, &vp, true);
680 	vn_finished_write(mp);
681 	free(buf, M_TEMP);
682 	return (0);
683 
684 error:
685 	UNP_PCB_LOCK(unp);
686 	unp->unp_flags &= ~UNP_BINDING;
687 	UNP_PCB_UNLOCK(unp);
688 	free(buf, M_TEMP);
689 	return (error);
690 }
691 
692 static int
693 uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
694 {
695 
696 	return (uipc_bindat(AT_FDCWD, so, nam, td));
697 }
698 
699 static int
700 uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
701 {
702 	int error;
703 
704 	KASSERT(td == curthread, ("uipc_connect: td != curthread"));
705 	error = unp_connect(so, nam, td);
706 	return (error);
707 }
708 
709 static int
710 uipc_connectat(int fd, struct socket *so, struct sockaddr *nam,
711     struct thread *td)
712 {
713 	int error;
714 
715 	KASSERT(td == curthread, ("uipc_connectat: td != curthread"));
716 	error = unp_connectat(fd, so, nam, td, false);
717 	return (error);
718 }
719 
720 static void
721 uipc_close(struct socket *so)
722 {
723 	struct unpcb *unp, *unp2;
724 	struct vnode *vp = NULL;
725 	struct mtx *vplock;
726 
727 	unp = sotounpcb(so);
728 	KASSERT(unp != NULL, ("uipc_close: unp == NULL"));
729 
730 	vplock = NULL;
731 	if ((vp = unp->unp_vnode) != NULL) {
732 		vplock = mtx_pool_find(unp_vp_mtxpool, vp);
733 		mtx_lock(vplock);
734 	}
735 	UNP_PCB_LOCK(unp);
736 	if (vp && unp->unp_vnode == NULL) {
737 		mtx_unlock(vplock);
738 		vp = NULL;
739 	}
740 	if (vp != NULL) {
741 		VOP_UNP_DETACH(vp);
742 		unp->unp_vnode = NULL;
743 	}
744 	if ((unp2 = unp_pcb_lock_peer(unp)) != NULL)
745 		unp_disconnect(unp, unp2);
746 	else
747 		UNP_PCB_UNLOCK(unp);
748 	if (vp) {
749 		mtx_unlock(vplock);
750 		vrele(vp);
751 	}
752 }
753 
754 static int
755 uipc_chmod(struct socket *so, mode_t mode, struct ucred *cred __unused,
756     struct thread *td __unused)
757 {
758 	struct unpcb *unp;
759 	int error;
760 
761 	if ((mode & ~ACCESSPERMS) != 0)
762 		return (EINVAL);
763 
764 	error = 0;
765 	unp = sotounpcb(so);
766 	UNP_PCB_LOCK(unp);
767 	if (unp->unp_vnode != NULL || (unp->unp_flags & UNP_BINDING) != 0)
768 		error = EINVAL;
769 	else
770 		unp->unp_mode = mode;
771 	UNP_PCB_UNLOCK(unp);
772 	return (error);
773 }
774 
775 static int
776 uipc_connect2(struct socket *so1, struct socket *so2)
777 {
778 	struct unpcb *unp, *unp2;
779 
780 	if (so1->so_type != so2->so_type)
781 		return (EPROTOTYPE);
782 
783 	unp = so1->so_pcb;
784 	KASSERT(unp != NULL, ("uipc_connect2: unp == NULL"));
785 	unp2 = so2->so_pcb;
786 	KASSERT(unp2 != NULL, ("uipc_connect2: unp2 == NULL"));
787 	unp_pcb_lock_pair(unp, unp2);
788 	unp_connect2(so1, so2, false);
789 	unp_pcb_unlock_pair(unp, unp2);
790 
791 	return (0);
792 }
793 
794 static void
795 maybe_schedule_gc(void)
796 {
797 	if (atomic_load_int(&unp_rights) != 0)
798 		taskqueue_enqueue_timeout(taskqueue_thread, &unp_gc_task, -1);
799 }
800 
801 static void
802 uipc_detach(struct socket *so)
803 {
804 	struct unpcb *unp, *unp2;
805 	struct mtx *vplock;
806 	struct vnode *vp;
807 
808 	unp = sotounpcb(so);
809 	KASSERT(unp != NULL, ("uipc_detach: unp == NULL"));
810 
811 	vp = NULL;
812 	vplock = NULL;
813 
814 	if (!SOLISTENING(so))
815 		unp_dispose(so);
816 
817 	UNP_LINK_WLOCK();
818 	LIST_REMOVE(unp, unp_link);
819 	if (unp->unp_gcflag & UNPGC_DEAD)
820 		LIST_REMOVE(unp, unp_dead);
821 	unp->unp_gencnt = ++unp_gencnt;
822 	--unp_count;
823 	UNP_LINK_WUNLOCK();
824 
825 	UNP_PCB_UNLOCK_ASSERT(unp);
826  restart:
827 	if ((vp = unp->unp_vnode) != NULL) {
828 		vplock = mtx_pool_find(unp_vp_mtxpool, vp);
829 		mtx_lock(vplock);
830 	}
831 	UNP_PCB_LOCK(unp);
832 	if (unp->unp_vnode != vp && unp->unp_vnode != NULL) {
833 		if (vplock)
834 			mtx_unlock(vplock);
835 		UNP_PCB_UNLOCK(unp);
836 		goto restart;
837 	}
838 	if ((vp = unp->unp_vnode) != NULL) {
839 		VOP_UNP_DETACH(vp);
840 		unp->unp_vnode = NULL;
841 	}
842 	if ((unp2 = unp_pcb_lock_peer(unp)) != NULL)
843 		unp_disconnect(unp, unp2);
844 	else
845 		UNP_PCB_UNLOCK(unp);
846 
847 	UNP_REF_LIST_LOCK();
848 	while (!LIST_EMPTY(&unp->unp_refs)) {
849 		struct unpcb *ref = LIST_FIRST(&unp->unp_refs);
850 
851 		unp_pcb_hold(ref);
852 		UNP_REF_LIST_UNLOCK();
853 
854 		MPASS(ref != unp);
855 		UNP_PCB_UNLOCK_ASSERT(ref);
856 		unp_drop(ref);
857 		UNP_REF_LIST_LOCK();
858 	}
859 	UNP_REF_LIST_UNLOCK();
860 
861 	UNP_PCB_LOCK(unp);
862 	unp->unp_socket->so_pcb = NULL;
863 	unp->unp_socket = NULL;
864 	free(unp->unp_addr, M_SONAME);
865 	unp->unp_addr = NULL;
866 	if (!unp_pcb_rele(unp))
867 		UNP_PCB_UNLOCK(unp);
868 	if (vp) {
869 		mtx_unlock(vplock);
870 		vrele(vp);
871 	}
872 	maybe_schedule_gc();
873 
874 	switch (so->so_type) {
875 	case SOCK_STREAM:
876 	case SOCK_SEQPACKET:
877 		MPASS(SOLISTENING(so) || (STAILQ_EMPTY(&so->so_rcv.uxst_mbq) &&
878 		    so->so_rcv.uxst_peer == NULL));
879 		break;
880 	case SOCK_DGRAM:
881 		/*
882 		 * Everything should have been unlinked/freed by unp_dispose()
883 		 * and/or unp_disconnect().
884 		 */
885 		MPASS(so->so_rcv.uxdg_peeked == NULL);
886 		MPASS(STAILQ_EMPTY(&so->so_rcv.uxdg_mb));
887 		MPASS(TAILQ_EMPTY(&so->so_rcv.uxdg_conns));
888 		MPASS(STAILQ_EMPTY(&so->so_snd.uxdg_mb));
889 	}
890 
891 	mtx_destroy(&so->so_snd_mtx);
892 	mtx_destroy(&so->so_rcv_mtx);
893 }
894 
895 static int
896 uipc_disconnect(struct socket *so)
897 {
898 	struct unpcb *unp, *unp2;
899 
900 	unp = sotounpcb(so);
901 	KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL"));
902 
903 	UNP_PCB_LOCK(unp);
904 	if ((unp2 = unp_pcb_lock_peer(unp)) != NULL)
905 		unp_disconnect(unp, unp2);
906 	else
907 		UNP_PCB_UNLOCK(unp);
908 	return (0);
909 }
910 
911 static void
912 uipc_fdclose(struct socket *so __unused)
913 {
914 	/*
915 	 * Ensure that userspace can't create orphaned file descriptors without
916 	 * triggering garbage collection.  Triggering GC from uipc_detach() is
917 	 * not sufficient, since that's only closed once a socket reference
918 	 * count drops to zero.
919 	 */
920 	maybe_schedule_gc();
921 }
922 
923 static int
924 uipc_listen(struct socket *so, int backlog, struct thread *td)
925 {
926 	struct unpcb *unp;
927 	int error;
928 
929 	MPASS(so->so_type != SOCK_DGRAM);
930 
931 	/*
932 	 * Synchronize with concurrent connection attempts.
933 	 */
934 	error = 0;
935 	unp = sotounpcb(so);
936 	UNP_PCB_LOCK(unp);
937 	if (unp->unp_conn != NULL || (unp->unp_flags & UNP_CONNECTING) != 0)
938 		error = EINVAL;
939 	else if (unp->unp_vnode == NULL)
940 		error = EDESTADDRREQ;
941 	if (error != 0) {
942 		UNP_PCB_UNLOCK(unp);
943 		return (error);
944 	}
945 
946 	SOCK_LOCK(so);
947 	error = solisten_proto_check(so);
948 	if (error == 0) {
949 		cru2xt(td, &unp->unp_peercred);
950 		if (!SOLISTENING(so)) {
951 			(void)chgsbsize(so->so_cred->cr_uidinfo,
952 			    &so->so_snd.sb_hiwat, 0, RLIM_INFINITY);
953 			(void)chgsbsize(so->so_cred->cr_uidinfo,
954 			    &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY);
955 		}
956 		solisten_proto(so, backlog);
957 	}
958 	SOCK_UNLOCK(so);
959 	UNP_PCB_UNLOCK(unp);
960 	return (error);
961 }
962 
963 static int
964 uipc_peeraddr(struct socket *so, struct sockaddr *ret)
965 {
966 	struct unpcb *unp, *unp2;
967 	const struct sockaddr *sa;
968 
969 	unp = sotounpcb(so);
970 	KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL"));
971 
972 	UNP_PCB_LOCK(unp);
973 	unp2 = unp_pcb_lock_peer(unp);
974 	if (unp2 != NULL) {
975 		if (unp2->unp_addr != NULL)
976 			sa = (struct sockaddr *)unp2->unp_addr;
977 		else
978 			sa = &sun_noname;
979 		bcopy(sa, ret, sa->sa_len);
980 		unp_pcb_unlock_pair(unp, unp2);
981 	} else {
982 		UNP_PCB_UNLOCK(unp);
983 		sa = &sun_noname;
984 		bcopy(sa, ret, sa->sa_len);
985 	}
986 	return (0);
987 }
988 
989 /*
990  * pr_sosend() called with mbuf instead of uio is a kernel thread.  NFS,
991  * netgraph(4) and other subsystems can call into socket code.  The
992  * function will condition the mbuf so that it can be safely put onto socket
993  * buffer and calculate its char count and mbuf count.
994  *
995  * Note: we don't support receiving control data from a kernel thread.  Our
996  * pr_sosend methods have MPASS() to check that.  This may change.
997  */
998 static void
999 uipc_reset_kernel_mbuf(struct mbuf *m, struct mchain *mc)
1000 {
1001 
1002 	M_ASSERTPKTHDR(m);
1003 
1004 	m_clrprotoflags(m);
1005 	m_tag_delete_chain(m, NULL);
1006 	m->m_pkthdr.rcvif = NULL;
1007 	m->m_pkthdr.flowid = 0;
1008 	m->m_pkthdr.csum_flags = 0;
1009 	m->m_pkthdr.fibnum = 0;
1010 	m->m_pkthdr.rsstype = 0;
1011 
1012 	mc_init_m(mc, m);
1013 	MPASS(m->m_pkthdr.len == mc->mc_len);
1014 }
1015 
1016 #ifdef SOCKBUF_DEBUG
1017 static inline void
1018 uipc_stream_sbcheck(struct sockbuf *sb)
1019 {
1020 	struct mbuf *d;
1021 	u_int dacc, dccc, dctl, dmbcnt;
1022 	bool notready = false;
1023 
1024 	dacc = dccc = dctl = dmbcnt = 0;
1025 	STAILQ_FOREACH(d, &sb->uxst_mbq, m_stailq) {
1026 		if (d == sb->uxst_fnrdy) {
1027 			MPASS(d->m_flags & M_NOTREADY);
1028 			notready = true;
1029 		}
1030 		if (d->m_type == MT_CONTROL)
1031 			dctl += d->m_len;
1032 		else if (d->m_type == MT_DATA) {
1033 			dccc +=  d->m_len;
1034 			if (!notready)
1035 				dacc += d->m_len;
1036 		} else
1037 			MPASS(0);
1038 		dmbcnt += MSIZE;
1039 		if (d->m_flags & M_EXT)
1040 			dmbcnt += d->m_ext.ext_size;
1041 		if (d->m_stailq.stqe_next == NULL)
1042 			MPASS(sb->uxst_mbq.stqh_last == &d->m_stailq.stqe_next);
1043 	}
1044 	MPASS(sb->uxst_fnrdy == NULL || notready);
1045 	MPASS(dacc == sb->sb_acc);
1046 	MPASS(dccc == sb->sb_ccc);
1047 	MPASS(dctl == sb->sb_ctl);
1048 	MPASS(dmbcnt == sb->sb_mbcnt);
1049 	(void)STAILQ_EMPTY(&sb->uxst_mbq);
1050 }
1051 #define	UIPC_STREAM_SBCHECK(sb)	uipc_stream_sbcheck(sb)
1052 #else
1053 #define	UIPC_STREAM_SBCHECK(sb)	do {} while (0)
1054 #endif
1055 
1056 /*
1057  * uipc_stream_sbspace() returns how much a writer can send, limited by char
1058  * count or mbuf memory use, whatever ends first.
1059  *
1060  * An obvious and legitimate reason for a socket having more data than allowed,
1061  * is lowering the limit with setsockopt(SO_RCVBUF) on already full buffer.
1062  * Also, sb_mbcnt may overcommit sb_mbmax in case if previous write observed
1063  * 'space < mbspace', but mchain allocated to hold 'space' bytes of data ended
1064  * up with 'mc_mlen > mbspace'.  A typical scenario would be a full buffer with
1065  * writer trying to push in a large write, and a slow reader, that reads just
1066  * a few bytes at a time.  In that case writer will keep creating new mbufs
1067  * with mc_split().  These mbufs will carry little chars, but will all point at
1068  * the same cluster, thus each adding cluster size to sb_mbcnt.  This means we
1069  * will count same cluster many times potentially underutilizing socket buffer.
1070  * We aren't optimizing towards ineffective readers.  Classic socket buffer had
1071  * the same "feature".
1072  */
1073 static inline u_int
1074 uipc_stream_sbspace(struct sockbuf *sb)
1075 {
1076 	u_int space, mbspace;
1077 
1078 	if (__predict_true(sb->sb_hiwat >= sb->sb_ccc + sb->sb_ctl))
1079 		space = sb->sb_hiwat - sb->sb_ccc - sb->sb_ctl;
1080 	else
1081 		return (0);
1082 	if (__predict_true(sb->sb_mbmax >= sb->sb_mbcnt))
1083 		mbspace = sb->sb_mbmax - sb->sb_mbcnt;
1084 	else
1085 		return (0);
1086 
1087 	return (min(space, mbspace));
1088 }
1089 
1090 /*
1091  * UNIX version of generic sbwait() for writes.  We wait on peer's receive
1092  * buffer, using our timeout.
1093  */
1094 static int
1095 uipc_stream_sbwait(struct socket *so, sbintime_t timeo)
1096 {
1097 	struct sockbuf *sb = &so->so_rcv;
1098 
1099 	SOCK_RECVBUF_LOCK_ASSERT(so);
1100 	sb->sb_flags |= SB_WAIT;
1101 	return (msleep_sbt(&sb->sb_acc, SOCK_RECVBUF_MTX(so), PSOCK | PCATCH,
1102 	    "sbwait", timeo, 0, 0));
1103 }
1104 
1105 static int
1106 uipc_sosend_stream_or_seqpacket(struct socket *so, struct sockaddr *addr,
1107     struct uio *uio0, struct mbuf *m, struct mbuf *c, int flags,
1108     struct thread *td)
1109 {
1110 	struct unpcb *unp2;
1111 	struct socket *so2;
1112 	struct sockbuf *sb;
1113 	struct uio *uio;
1114 	struct mchain mc, cmc;
1115 	size_t resid, sent;
1116 	bool nonblock, eor, aio;
1117 	int error, needsopts;
1118 
1119 	MPASS((uio0 != NULL && m == NULL) || (m != NULL && uio0 == NULL));
1120 	MPASS(m == NULL || c == NULL);
1121 
1122 	if (__predict_false(flags & MSG_OOB))
1123 		return (EOPNOTSUPP);
1124 
1125 	nonblock = (so->so_state & SS_NBIO) ||
1126 	    (flags & (MSG_DONTWAIT | MSG_NBIO));
1127 	eor = flags & MSG_EOR;
1128 
1129 	mc = MCHAIN_INITIALIZER(&mc);
1130 	cmc = MCHAIN_INITIALIZER(&cmc);
1131 	sent = 0;
1132 	aio = false;
1133 	needsopts = 0;
1134 
1135 	if (m == NULL) {
1136 		if (c != NULL &&
1137 		    (error = unp_internalize(c, &cmc, td, &needsopts)))
1138 			goto out;
1139 		/*
1140 		 * This function may read more data from the uio than it would
1141 		 * then place on socket.  That would leave uio inconsistent
1142 		 * upon return.  Normally uio is allocated on the stack of the
1143 		 * syscall thread and we don't care about leaving it consistent.
1144 		 * However, aio(9) will allocate a uio as part of job and will
1145 		 * use it to track progress.  We detect aio(9) checking the
1146 		 * SB_AIO_RUNNING flag.  It is safe to check it without lock
1147 		 * cause it is set and cleared in the same taskqueue thread.
1148 		 *
1149 		 * This check can also produce a false positive: there is
1150 		 * aio(9) job and also there is a syscall we are serving now.
1151 		 * No sane software does that, it would leave to a mess in
1152 		 * the socket buffer, as aio(9) doesn't grab the I/O sx(9).
1153 		 * But syzkaller can create this mess.  For such false positive
1154 		 * our goal is just don't panic or leak memory.
1155 		 */
1156 		if (__predict_false(so->so_snd.sb_flags & SB_AIO_RUNNING)) {
1157 			uio = cloneuio(uio0);
1158 			aio = true;
1159 		} else {
1160 			uio = uio0;
1161 			resid = uio->uio_resid;
1162 		}
1163 		/*
1164 		 * Optimization for a case when our send fits into the receive
1165 		 * buffer - do the copyin before taking any locks, sized to our
1166 		 * send buffer.  Later copyins will also take into account
1167 		 * space in the peer's receive buffer.
1168 		 */
1169 		error = mc_uiotomc(&mc, uio, so->so_snd.sb_hiwat, 0, M_WAITOK,
1170 		    eor ? M_EOR : 0);
1171 		if (__predict_false(error))
1172 			goto out2;
1173 	} else
1174 		uipc_reset_kernel_mbuf(m, &mc);
1175 
1176 	error = SOCK_IO_SEND_LOCK(so, SBLOCKWAIT(flags));
1177 	if (error)
1178 		goto out2;
1179 
1180 	if (__predict_false((error = uipc_lock_peer(so, &unp2)) != 0))
1181 		goto out3;
1182 
1183 	/* Check for SO_PASS* flags */
1184 	so2 = unp2->unp_socket;
1185 	if ((atomic_load_int(&so2->so_options) & needsopts) != needsopts) {
1186 		error = EPERM;
1187 		UNP_PCB_UNLOCK(unp2);
1188 		goto out3;
1189 	}
1190 
1191 	if (unp2->unp_flags & UNP_WANTCRED_MASK) {
1192 		/*
1193 		 * Credentials are passed only once on SOCK_STREAM and
1194 		 * SOCK_SEQPACKET (LOCAL_CREDS => WANTCRED_ONESHOT), or
1195 		 * forever (LOCAL_CREDS_PERSISTENT => WANTCRED_ALWAYS).
1196 		 */
1197 		unp_addsockcred(td, &cmc, unp2->unp_flags);
1198 		unp2->unp_flags &= ~UNP_WANTCRED_ONESHOT;
1199 	}
1200 
1201 	/*
1202 	 * Cycle through the data to send and available space in the peer's
1203 	 * receive buffer.  Put a reference on the peer socket, so that it
1204 	 * doesn't get freed while we sbwait().  If peer goes away, we will
1205 	 * observe the SBS_CANTRCVMORE and our sorele() will finalize peer's
1206 	 * socket destruction.
1207 	 */
1208 	soref(so2);
1209 	UNP_PCB_UNLOCK(unp2);
1210 	sb = &so2->so_rcv;
1211 	while (mc.mc_len + cmc.mc_len > 0) {
1212 		struct mchain mcnext = MCHAIN_INITIALIZER(&mcnext);
1213 		u_int space;
1214 
1215 		SOCK_RECVBUF_LOCK(so2);
1216 restart:
1217 		UIPC_STREAM_SBCHECK(sb);
1218 		if (__predict_false(cmc.mc_len > sb->sb_hiwat)) {
1219 			SOCK_RECVBUF_UNLOCK(so2);
1220 			error = EMSGSIZE;
1221 			goto out4;
1222 		}
1223 		if (__predict_false(sb->sb_state & SBS_CANTRCVMORE)) {
1224 			SOCK_RECVBUF_UNLOCK(so2);
1225 			error = EPIPE;
1226 			goto out4;
1227 		}
1228 		/*
1229 		 * Wait on the peer socket receive buffer until we have enough
1230 		 * space to put at least control.  The data is a stream and can
1231 		 * be put partially, but control is really a datagram.
1232 		 */
1233 		space = uipc_stream_sbspace(sb);
1234 		if (space < sb->sb_lowat || space < cmc.mc_len) {
1235 			if (nonblock) {
1236 				if (aio)
1237 					sb->uxst_flags |= UXST_PEER_AIO;
1238 				SOCK_RECVBUF_UNLOCK(so2);
1239 				if (aio) {
1240 					SOCK_SENDBUF_LOCK(so);
1241 					so->so_snd.sb_ccc =
1242 					    so->so_snd.sb_hiwat - space;
1243 					SOCK_SENDBUF_UNLOCK(so);
1244 				}
1245 				error = EWOULDBLOCK;
1246 				goto out4;
1247 			}
1248 			if ((error = uipc_stream_sbwait(so2,
1249 			    so->so_snd.sb_timeo)) != 0) {
1250 				SOCK_RECVBUF_UNLOCK(so2);
1251 				goto out4;
1252 			} else
1253 				goto restart;
1254 		}
1255 		MPASS(space >= cmc.mc_len);
1256 		space -= cmc.mc_len;
1257 		if (space == 0) {
1258 			/* There is space only to send control. */
1259 			MPASS(!STAILQ_EMPTY(&cmc.mc_q));
1260 			mcnext = mc;
1261 			mc = MCHAIN_INITIALIZER(&mc);
1262 		} else if (space < mc.mc_len) {
1263 			/* Not enough space. */
1264 			if (__predict_false(mc_split(&mc, &mcnext, space,
1265 			    M_NOWAIT) == ENOMEM)) {
1266 				/*
1267 				 * If allocation failed use M_WAITOK and merge
1268 				 * the chain back.  Next time mc_split() will
1269 				 * easily split at the same place.  Only if we
1270 				 * race with setsockopt(SO_RCVBUF) shrinking
1271 				 * sb_hiwat can this happen more than once.
1272 				 */
1273 				SOCK_RECVBUF_UNLOCK(so2);
1274 				(void)mc_split(&mc, &mcnext, space, M_WAITOK);
1275 				mc_concat(&mc, &mcnext);
1276 				SOCK_RECVBUF_LOCK(so2);
1277 				goto restart;
1278 			}
1279 			MPASS(mc.mc_len == space);
1280 		}
1281 		if (!STAILQ_EMPTY(&cmc.mc_q)) {
1282 			STAILQ_CONCAT(&sb->uxst_mbq, &cmc.mc_q);
1283 			sb->sb_ctl += cmc.mc_len;
1284 			sb->sb_mbcnt += cmc.mc_mlen;
1285 			cmc.mc_len = 0;
1286 		}
1287 		sent += mc.mc_len;
1288 		if (sb->uxst_fnrdy == NULL)
1289 			sb->sb_acc += mc.mc_len;
1290 		sb->sb_ccc += mc.mc_len;
1291 		sb->sb_mbcnt += mc.mc_mlen;
1292 		STAILQ_CONCAT(&sb->uxst_mbq, &mc.mc_q);
1293 		UIPC_STREAM_SBCHECK(sb);
1294 		space = uipc_stream_sbspace(sb);
1295 		sorwakeup_locked(so2);
1296 		if (!STAILQ_EMPTY(&mcnext.mc_q)) {
1297 			/*
1298 			 * Such assignment is unsafe in general, but it is
1299 			 * safe with !STAILQ_EMPTY(&mcnext.mc_q).  In C++ we
1300 			 * could reload = for STAILQs :)
1301 			 */
1302 			mc = mcnext;
1303 		} else if (uio != NULL && uio->uio_resid > 0) {
1304 			/*
1305 			 * Copyin sum of peer's receive buffer space and our
1306 			 * sb_hiwat, which is our virtual send buffer size.
1307 			 * See comment above unpst_sendspace declaration.
1308 			 * We are reading sb_hiwat locklessly, cause a) we
1309 			 * don't care about an application that does send(2)
1310 			 * and setsockopt(2) racing internally, and for an
1311 			 * application that does this in sequence we will see
1312 			 * the correct value cause sbsetopt() uses buffer lock
1313 			 * and we also have already acquired it at least once.
1314 			 */
1315 			error = mc_uiotomc(&mc, uio, space +
1316 			    atomic_load_int(&so->so_snd.sb_hiwat), 0, M_WAITOK,
1317 			    eor ? M_EOR : 0);
1318 			if (__predict_false(error))
1319 				goto out4;
1320 		} else
1321 			mc = MCHAIN_INITIALIZER(&mc);
1322 	}
1323 
1324 	MPASS(STAILQ_EMPTY(&mc.mc_q));
1325 
1326 	td->td_ru.ru_msgsnd++;
1327 out4:
1328 	sorele(so2);
1329 out3:
1330 	SOCK_IO_SEND_UNLOCK(so);
1331 out2:
1332 	if (aio) {
1333 		freeuio(uio);
1334 		uioadvance(uio0, sent);
1335 	} else if (uio != NULL)
1336 		uio->uio_resid = resid - sent;
1337 	if (!mc_empty(&cmc))
1338 		unp_scan(mc_first(&cmc), unp_freerights);
1339 out:
1340 	mc_freem(&mc);
1341 	mc_freem(&cmc);
1342 
1343 	return (error);
1344 }
1345 
1346 /*
1347  * Wakeup a writer, used by recv(2) and shutdown(2).
1348  *
1349  * @param so	Points to a connected stream socket with receive buffer locked
1350  *
1351  * In a blocking mode peer is sleeping on our receive buffer, and we need just
1352  * wakeup(9) on it.  But to wake up various event engines, we need to reach
1353  * over to peer's selinfo.  This can be safely done as the socket buffer
1354  * receive lock is protecting us from the peer going away.
1355  */
1356 static void
1357 uipc_wakeup_writer(struct socket *so)
1358 {
1359 	struct sockbuf *sb = &so->so_rcv;
1360 	struct selinfo *sel;
1361 
1362 	SOCK_RECVBUF_LOCK_ASSERT(so);
1363 	MPASS(sb->uxst_peer != NULL);
1364 
1365 	sel = &sb->uxst_peer->so_wrsel;
1366 
1367 	if (sb->uxst_flags & UXST_PEER_SEL) {
1368 		selwakeuppri(sel, PSOCK);
1369 		/*
1370 		 * XXXGL: sowakeup() does SEL_WAITING() without locks.
1371 		 */
1372 		if (!SEL_WAITING(sel))
1373 			sb->uxst_flags &= ~UXST_PEER_SEL;
1374 	}
1375 	if (sb->sb_flags & SB_WAIT) {
1376 		sb->sb_flags &= ~SB_WAIT;
1377 		wakeup(&sb->sb_acc);
1378 	}
1379 	KNOTE_LOCKED(&sel->si_note, 0);
1380 	SOCK_RECVBUF_UNLOCK(so);
1381 }
1382 
1383 static void
1384 uipc_cantrcvmore(struct socket *so)
1385 {
1386 
1387 	SOCK_RECVBUF_LOCK(so);
1388 	so->so_rcv.sb_state |= SBS_CANTRCVMORE;
1389 	selwakeuppri(&so->so_rdsel, PSOCK);
1390 	KNOTE_LOCKED(&so->so_rdsel.si_note, 0);
1391 	if (so->so_rcv.uxst_peer != NULL)
1392 		uipc_wakeup_writer(so);
1393 	else
1394 		SOCK_RECVBUF_UNLOCK(so);
1395 }
1396 
1397 static int
1398 uipc_soreceive_stream_or_seqpacket(struct socket *so, struct sockaddr **psa,
1399     struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1400 {
1401 	struct sockbuf *sb = &so->so_rcv;
1402 	struct mbuf *control, *m, *first, *part, *next;
1403 	u_int ctl, space, datalen, mbcnt, partlen;
1404 	int error, flags;
1405 	bool nonblock, waitall, peek;
1406 
1407 	MPASS(mp0 == NULL);
1408 
1409 	if (psa != NULL)
1410 		*psa = NULL;
1411 	if (controlp != NULL)
1412 		*controlp = NULL;
1413 
1414 	flags = flagsp != NULL ? *flagsp : 0;
1415 	nonblock = (so->so_state & SS_NBIO) ||
1416 	    (flags & (MSG_DONTWAIT | MSG_NBIO));
1417 	peek = flags & MSG_PEEK;
1418 	waitall = (flags & MSG_WAITALL) && !peek;
1419 
1420 	/*
1421 	 * This check may fail only on a socket that never went through
1422 	 * connect(2).  We can check this locklessly, cause: a) for a new born
1423 	 * socket we don't care about applications that may race internally
1424 	 * between connect(2) and recv(2), and b) for a dying socket if we
1425 	 * miss update by unp_sosidisconnected(), we would still get the check
1426 	 * correct.  For dying socket we would observe SBS_CANTRCVMORE later.
1427 	 */
1428 	if (__predict_false((atomic_load_short(&so->so_state) &
1429 	    (SS_ISCONNECTED|SS_ISDISCONNECTED)) == 0))
1430 		return (ENOTCONN);
1431 
1432 	error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags));
1433 	if (__predict_false(error))
1434 		return (error);
1435 
1436 restart:
1437 	SOCK_RECVBUF_LOCK(so);
1438 	UIPC_STREAM_SBCHECK(sb);
1439 	while (sb->sb_acc < sb->sb_lowat &&
1440 	    (sb->sb_ctl == 0 || controlp == NULL)) {
1441 		if (so->so_error) {
1442 			error = so->so_error;
1443 			if (!peek)
1444 				so->so_error = 0;
1445 			SOCK_RECVBUF_UNLOCK(so);
1446 			SOCK_IO_RECV_UNLOCK(so);
1447 			return (error);
1448 		}
1449 		if (sb->sb_state & SBS_CANTRCVMORE) {
1450 			SOCK_RECVBUF_UNLOCK(so);
1451 			SOCK_IO_RECV_UNLOCK(so);
1452 			return (0);
1453 		}
1454 		if (nonblock) {
1455 			SOCK_RECVBUF_UNLOCK(so);
1456 			SOCK_IO_RECV_UNLOCK(so);
1457 			return (EWOULDBLOCK);
1458 		}
1459 		error = sbwait(so, SO_RCV);
1460 		if (error) {
1461 			SOCK_RECVBUF_UNLOCK(so);
1462 			SOCK_IO_RECV_UNLOCK(so);
1463 			return (error);
1464 		}
1465 	}
1466 
1467 	MPASS(STAILQ_FIRST(&sb->uxst_mbq));
1468 	MPASS(sb->sb_acc > 0 || sb->sb_ctl > 0);
1469 
1470 	mbcnt = 0;
1471 	ctl = 0;
1472 	first = STAILQ_FIRST(&sb->uxst_mbq);
1473 	if (first->m_type == MT_CONTROL) {
1474 		control = first;
1475 		STAILQ_FOREACH_FROM(first, &sb->uxst_mbq, m_stailq) {
1476 			if (first->m_type != MT_CONTROL)
1477 				break;
1478 			ctl += first->m_len;
1479 			mbcnt += MSIZE;
1480 			if (first->m_flags & M_EXT)
1481 				mbcnt += first->m_ext.ext_size;
1482 		}
1483 	} else
1484 		control = NULL;
1485 
1486 	/*
1487 	 * Find split point for the next copyout.  On exit from the loop,
1488 	 * 'next' points to the new head of the buffer STAILQ and 'datalen'
1489 	 * contains the amount of data we will copy out at the end.  The
1490 	 * copyout is protected by the I/O lock only, as writers can only
1491 	 * append to the buffer.  We need to record the socket buffer state
1492 	 * and do all length adjustments before dropping the socket buffer lock.
1493 	 */
1494 	for (space = uio->uio_resid, m = next = first, part = NULL, datalen = 0;
1495 	     space > 0 && m != sb->uxst_fnrdy && m->m_type == MT_DATA;
1496 	     m = STAILQ_NEXT(m, m_stailq)) {
1497 		if (space >= m->m_len) {
1498 			space -= m->m_len;
1499 			datalen += m->m_len;
1500 			mbcnt += MSIZE;
1501 			if (m->m_flags & M_EXT)
1502 				mbcnt += m->m_ext.ext_size;
1503 			if (m->m_flags & M_EOR) {
1504 				flags |= MSG_EOR;
1505 				next = STAILQ_NEXT(m, m_stailq);
1506 				break;
1507 			}
1508 		} else {
1509 			datalen += space;
1510 			partlen = space;
1511 			if (!peek) {
1512 				m->m_len -= partlen;
1513 				m->m_data += partlen;
1514 			}
1515 			next = part = m;
1516 			break;
1517 		}
1518 		next = STAILQ_NEXT(m, m_stailq);
1519 	}
1520 
1521 	if (!peek) {
1522 		if (next == NULL)
1523 			STAILQ_INIT(&sb->uxst_mbq);
1524 		else
1525 			STAILQ_FIRST(&sb->uxst_mbq) = next;
1526 		MPASS(sb->sb_acc >= datalen);
1527 		sb->sb_acc -= datalen;
1528 		sb->sb_ccc -= datalen;
1529 		MPASS(sb->sb_ctl >= ctl);
1530 		sb->sb_ctl -= ctl;
1531 		MPASS(sb->sb_mbcnt >= mbcnt);
1532 		sb->sb_mbcnt -= mbcnt;
1533 		UIPC_STREAM_SBCHECK(sb);
1534 		if (__predict_true(sb->uxst_peer != NULL)) {
1535 			struct unpcb *unp2;
1536 			bool aio;
1537 
1538 			if ((aio = sb->uxst_flags & UXST_PEER_AIO))
1539 				sb->uxst_flags &= ~UXST_PEER_AIO;
1540 
1541 			uipc_wakeup_writer(so);
1542 			/*
1543 			 * XXXGL: need to go through uipc_lock_peer() after
1544 			 * the receive buffer lock dropped, it was protecting
1545 			 * us from unp_soisdisconnected().  The aio workarounds
1546 			 * should be refactored to the aio(4) side.
1547 			 */
1548 			if (aio && uipc_lock_peer(so, &unp2) == 0) {
1549 				struct socket *so2 = unp2->unp_socket;
1550 
1551 				SOCK_SENDBUF_LOCK(so2);
1552 				so2->so_snd.sb_ccc -= datalen;
1553 				sowakeup_aio(so2, SO_SND);
1554 				SOCK_SENDBUF_UNLOCK(so2);
1555 				UNP_PCB_UNLOCK(unp2);
1556 			}
1557 		} else
1558 			SOCK_RECVBUF_UNLOCK(so);
1559 	} else
1560 		SOCK_RECVBUF_UNLOCK(so);
1561 
1562 	while (control != NULL && control->m_type == MT_CONTROL) {
1563 		if (!peek) {
1564 			/*
1565 			 * unp_externalize() failure must abort entire read(2).
1566 			 * Such failure should also free the problematic
1567 			 * control, but link back the remaining data to the head
1568 			 * of the buffer, so that socket is not left in a state
1569 			 * where it can't progress forward with reading.
1570 			 * Probability of such a failure is really low, so it
1571 			 * is fine that we need to perform pretty complex
1572 			 * operation here to reconstruct the buffer.
1573 			 */
1574 			error = unp_externalize(so, control, controlp, flags);
1575 			control = m_free(control);
1576 			if (__predict_false(error && control != NULL)) {
1577 				struct mchain cmc;
1578 
1579 				mc_init_m(&cmc, control);
1580 
1581 				SOCK_RECVBUF_LOCK(so);
1582 				if (__predict_false(
1583 				    (sb->sb_state & SBS_CANTRCVMORE) ||
1584 				    cmc.mc_len + sb->sb_ccc + sb->sb_ctl >
1585 				    sb->sb_hiwat)) {
1586 					/*
1587 					 * While the lock was dropped and we
1588 					 * were failing in unp_externalize(),
1589 					 * the peer could has a) disconnected,
1590 					 * b) filled the buffer so that we
1591 					 * can't prepend data back.
1592 					 * These are two edge conditions that
1593 					 * we just can't handle, so lose the
1594 					 * data and return the error.
1595 					 */
1596 					SOCK_RECVBUF_UNLOCK(so);
1597 					SOCK_IO_RECV_UNLOCK(so);
1598 					unp_scan(mc_first(&cmc),
1599 					    unp_freerights);
1600 					mc_freem(&cmc);
1601 					return (error);
1602 				}
1603 
1604 				UIPC_STREAM_SBCHECK(sb);
1605 				/* XXXGL: STAILQ_PREPEND */
1606 				STAILQ_CONCAT(&cmc.mc_q, &sb->uxst_mbq);
1607 				STAILQ_SWAP(&cmc.mc_q, &sb->uxst_mbq, mbuf);
1608 
1609 				sb->sb_ctl = sb->sb_acc = sb->sb_ccc =
1610 				    sb->sb_mbcnt = 0;
1611 				STAILQ_FOREACH(m, &sb->uxst_mbq, m_stailq) {
1612 					if (m->m_type == MT_DATA) {
1613 						sb->sb_acc += m->m_len;
1614 						sb->sb_ccc += m->m_len;
1615 					} else {
1616 						sb->sb_ctl += m->m_len;
1617 					}
1618 					sb->sb_mbcnt += MSIZE;
1619 					if (m->m_flags & M_EXT)
1620 						sb->sb_mbcnt +=
1621 						    m->m_ext.ext_size;
1622 				}
1623 				UIPC_STREAM_SBCHECK(sb);
1624 				SOCK_RECVBUF_UNLOCK(so);
1625 				SOCK_IO_RECV_UNLOCK(so);
1626 				return (error);
1627 			}
1628 			if (controlp != NULL) {
1629 				while (*controlp != NULL)
1630 					controlp = &(*controlp)->m_next;
1631 			}
1632 		} else {
1633 			/*
1634 			 * XXXGL
1635 			 *
1636 			 * In MSG_PEEK case control is not externalized.  This
1637 			 * means we are leaking some kernel pointers to the
1638 			 * userland.  They are useless to a law-abiding
1639 			 * application, but may be useful to a malware.  This
1640 			 * is what the historical implementation in the
1641 			 * soreceive_generic() did. To be improved?
1642 			 */
1643 			if (controlp != NULL) {
1644 				*controlp = m_copym(control, 0, control->m_len,
1645 				    M_WAITOK);
1646 				controlp = &(*controlp)->m_next;
1647 			}
1648 			control = STAILQ_NEXT(control, m_stailq);
1649 		}
1650 	}
1651 
1652 	for (m = first; datalen > 0; m = next) {
1653 		void *data;
1654 		u_int len;
1655 
1656 		next = STAILQ_NEXT(m, m_stailq);
1657 		if (m == part) {
1658 			data = peek ?
1659 			    mtod(m, char *) : mtod(m, char *) - partlen;
1660 			len = partlen;
1661 		} else {
1662 			data = mtod(m, char *);
1663 			len = m->m_len;
1664 		}
1665 		error = uiomove(data, len, uio);
1666 		if (__predict_false(error)) {
1667 			if (!peek)
1668 				for (; m != part && datalen > 0; m = next) {
1669 					next = STAILQ_NEXT(m, m_stailq);
1670 					MPASS(datalen >= m->m_len);
1671 					datalen -= m->m_len;
1672 					m_free(m);
1673 				}
1674 			SOCK_IO_RECV_UNLOCK(so);
1675 			return (error);
1676 		}
1677 		datalen -= len;
1678 		if (!peek && m != part)
1679 			m_free(m);
1680 	}
1681 	if (waitall && !(flags & MSG_EOR) && uio->uio_resid > 0)
1682 		goto restart;
1683 	SOCK_IO_RECV_UNLOCK(so);
1684 
1685 	if (flagsp != NULL)
1686 		*flagsp |= flags;
1687 
1688 	uio->uio_td->td_ru.ru_msgrcv++;
1689 
1690 	return (0);
1691 }
1692 
1693 static int
1694 uipc_sopoll_stream_or_seqpacket(struct socket *so, int events,
1695     struct thread *td)
1696 {
1697 	struct unpcb *unp = sotounpcb(so);
1698 	int revents;
1699 
1700 	UNP_PCB_LOCK(unp);
1701 	if (SOLISTENING(so)) {
1702 		/* The above check is safe, since conversion to listening uses
1703 		 * both protocol and socket lock.
1704 		 */
1705 		SOCK_LOCK(so);
1706 		if (!(events & (POLLIN | POLLRDNORM)))
1707 			revents = 0;
1708 		else if (!TAILQ_EMPTY(&so->sol_comp))
1709 			revents = events & (POLLIN | POLLRDNORM);
1710 		else if (so->so_error)
1711 			revents = (events & (POLLIN | POLLRDNORM)) | POLLHUP;
1712 		else {
1713 			selrecord(td, &so->so_rdsel);
1714 			revents = 0;
1715 		}
1716 		SOCK_UNLOCK(so);
1717 	} else {
1718 		if (so->so_state & SS_ISDISCONNECTED)
1719 			revents = POLLHUP;
1720 		else
1721 			revents = 0;
1722 		if (events & (POLLIN | POLLRDNORM | POLLRDHUP)) {
1723 			SOCK_RECVBUF_LOCK(so);
1724 			if (sbavail(&so->so_rcv) >= so->so_rcv.sb_lowat ||
1725 			    so->so_error || so->so_rerror)
1726 				revents |= events & (POLLIN | POLLRDNORM);
1727 			if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
1728 				revents |= events &
1729 				    (POLLIN | POLLRDNORM | POLLRDHUP);
1730 			if (!(revents & (POLLIN | POLLRDNORM | POLLRDHUP))) {
1731 				selrecord(td, &so->so_rdsel);
1732 				so->so_rcv.sb_flags |= SB_SEL;
1733 			}
1734 			SOCK_RECVBUF_UNLOCK(so);
1735 		}
1736 		if (events & (POLLOUT | POLLWRNORM)) {
1737 			struct socket *so2 = so->so_rcv.uxst_peer;
1738 
1739 			if (so2 != NULL) {
1740 				struct sockbuf *sb = &so2->so_rcv;
1741 
1742 				SOCK_RECVBUF_LOCK(so2);
1743 				if (uipc_stream_sbspace(sb) >= sb->sb_lowat)
1744 					revents |= events &
1745 					    (POLLOUT | POLLWRNORM);
1746 				if (sb->sb_state & SBS_CANTRCVMORE)
1747 					revents |= POLLHUP;
1748 				if (!(revents & (POLLOUT | POLLWRNORM))) {
1749 					so2->so_rcv.uxst_flags |= UXST_PEER_SEL;
1750 					selrecord(td, &so->so_wrsel);
1751 				}
1752 				SOCK_RECVBUF_UNLOCK(so2);
1753 			} else
1754 				selrecord(td, &so->so_wrsel);
1755 		}
1756 	}
1757 	UNP_PCB_UNLOCK(unp);
1758 	return (revents);
1759 }
1760 
1761 static void
1762 uipc_wrknl_lock(void *arg)
1763 {
1764 	struct socket *so = arg;
1765 	struct unpcb *unp = sotounpcb(so);
1766 
1767 retry:
1768 	if (SOLISTENING(so)) {
1769 		SOLISTEN_LOCK(so);
1770 	} else {
1771 		UNP_PCB_LOCK(unp);
1772 		if (__predict_false(SOLISTENING(so))) {
1773 			UNP_PCB_UNLOCK(unp);
1774 			goto retry;
1775 		}
1776 		if (so->so_rcv.uxst_peer != NULL)
1777 			SOCK_RECVBUF_LOCK(so->so_rcv.uxst_peer);
1778 	}
1779 }
1780 
1781 static void
1782 uipc_wrknl_unlock(void *arg)
1783 {
1784 	struct socket *so = arg;
1785 	struct unpcb *unp = sotounpcb(so);
1786 
1787 	if (SOLISTENING(so))
1788 		SOLISTEN_UNLOCK(so);
1789 	else {
1790 		if (so->so_rcv.uxst_peer != NULL)
1791 			SOCK_RECVBUF_UNLOCK(so->so_rcv.uxst_peer);
1792 		UNP_PCB_UNLOCK(unp);
1793 	}
1794 }
1795 
1796 static void
1797 uipc_wrknl_assert_lock(void *arg, int what)
1798 {
1799 	struct socket *so = arg;
1800 
1801 	if (SOLISTENING(so)) {
1802 		if (what == LA_LOCKED)
1803 			SOLISTEN_LOCK_ASSERT(so);
1804 		else
1805 			SOLISTEN_UNLOCK_ASSERT(so);
1806 	} else {
1807 		/*
1808 		 * The pr_soreceive method will put a note without owning the
1809 		 * unp lock, so we can't assert it here.  But we can safely
1810 		 * dereference uxst_peer pointer, since receive buffer lock
1811 		 * is assumed to be held here.
1812 		 */
1813 		if (what == LA_LOCKED && so->so_rcv.uxst_peer != NULL)
1814 			SOCK_RECVBUF_LOCK_ASSERT(so->so_rcv.uxst_peer);
1815 	}
1816 }
1817 
1818 static void
1819 uipc_filt_sowdetach(struct knote *kn)
1820 {
1821 	struct socket *so = kn->kn_fp->f_data;
1822 
1823 	uipc_wrknl_lock(so);
1824 	knlist_remove(&so->so_wrsel.si_note, kn, 1);
1825 	uipc_wrknl_unlock(so);
1826 }
1827 
1828 static int
1829 uipc_filt_sowrite(struct knote *kn, long hint)
1830 {
1831 	struct socket *so = kn->kn_fp->f_data, *so2;
1832 	struct unpcb *unp = sotounpcb(so), *unp2 = unp->unp_conn;
1833 
1834 	if (SOLISTENING(so))
1835 		return (0);
1836 
1837 	if (unp2 == NULL) {
1838 		if (so->so_state & SS_ISDISCONNECTED) {
1839 			kn->kn_flags |= EV_EOF;
1840 			kn->kn_fflags = so->so_error;
1841 			return (1);
1842 		} else
1843 			return (0);
1844 	}
1845 
1846 	so2 = unp2->unp_socket;
1847 	SOCK_RECVBUF_LOCK_ASSERT(so2);
1848 	kn->kn_data = uipc_stream_sbspace(&so2->so_rcv);
1849 
1850 	if (so2->so_rcv.sb_state & SBS_CANTRCVMORE) {
1851 		kn->kn_flags |= EV_EOF;
1852 		return (1);
1853 	} else if (kn->kn_sfflags & NOTE_LOWAT)
1854 		return (kn->kn_data >= kn->kn_sdata);
1855 	else
1856 		return (kn->kn_data >= so2->so_rcv.sb_lowat);
1857 }
1858 
1859 static int
1860 uipc_filt_soempty(struct knote *kn, long hint)
1861 {
1862 	struct socket *so = kn->kn_fp->f_data, *so2;
1863 	struct unpcb *unp = sotounpcb(so), *unp2 = unp->unp_conn;
1864 
1865 	if (SOLISTENING(so) || unp2 == NULL)
1866 		return (1);
1867 
1868 	so2 = unp2->unp_socket;
1869 	SOCK_RECVBUF_LOCK_ASSERT(so2);
1870 	kn->kn_data = uipc_stream_sbspace(&so2->so_rcv);
1871 
1872 	return (kn->kn_data == 0 ? 1 : 0);
1873 }
1874 
1875 static const struct filterops uipc_write_filtops = {
1876 	.f_isfd = 1,
1877 	.f_detach = uipc_filt_sowdetach,
1878 	.f_event = uipc_filt_sowrite,
1879 	.f_copy = knote_triv_copy,
1880 };
1881 static const struct filterops uipc_empty_filtops = {
1882 	.f_isfd = 1,
1883 	.f_detach = uipc_filt_sowdetach,
1884 	.f_event = uipc_filt_soempty,
1885 	.f_copy = knote_triv_copy,
1886 };
1887 
1888 static int
1889 uipc_kqfilter_stream_or_seqpacket(struct socket *so, struct knote *kn)
1890 {
1891 	struct unpcb *unp = sotounpcb(so);
1892 	struct knlist *knl;
1893 
1894 	switch (kn->kn_filter) {
1895 	case EVFILT_READ:
1896 		return (sokqfilter_generic(so, kn));
1897 	case EVFILT_WRITE:
1898 		kn->kn_fop = &uipc_write_filtops;
1899 		break;
1900 	case EVFILT_EMPTY:
1901 		kn->kn_fop = &uipc_empty_filtops;
1902 		break;
1903 	default:
1904 		return (EINVAL);
1905 	}
1906 
1907 	knl = &so->so_wrsel.si_note;
1908 	UNP_PCB_LOCK(unp);
1909 	if (SOLISTENING(so)) {
1910 		SOLISTEN_LOCK(so);
1911 		knlist_add(knl, kn, 1);
1912 		SOLISTEN_UNLOCK(so);
1913 	} else {
1914 		struct socket *so2 = so->so_rcv.uxst_peer;
1915 
1916 		if (so2 != NULL)
1917 			SOCK_RECVBUF_LOCK(so2);
1918 		knlist_add(knl, kn, 1);
1919 		if (so2 != NULL)
1920 			SOCK_RECVBUF_UNLOCK(so2);
1921 	}
1922 	UNP_PCB_UNLOCK(unp);
1923 	return (0);
1924 }
1925 
1926 /* PF_UNIX/SOCK_DGRAM version of sbspace() */
1927 static inline bool
1928 uipc_dgram_sbspace(struct sockbuf *sb, u_int cc, u_int mbcnt)
1929 {
1930 	u_int bleft, mleft;
1931 
1932 	/*
1933 	 * Negative space may happen if send(2) is followed by
1934 	 * setsockopt(SO_SNDBUF/SO_RCVBUF) that shrinks maximum.
1935 	 */
1936 	if (__predict_false(sb->sb_hiwat < sb->uxdg_cc ||
1937 	    sb->sb_mbmax < sb->uxdg_mbcnt))
1938 		return (false);
1939 
1940 	if (__predict_false(sb->sb_state & SBS_CANTRCVMORE))
1941 		return (false);
1942 
1943 	bleft = sb->sb_hiwat - sb->uxdg_cc;
1944 	mleft = sb->sb_mbmax - sb->uxdg_mbcnt;
1945 
1946 	return (bleft >= cc && mleft >= mbcnt);
1947 }
1948 
1949 /*
1950  * PF_UNIX/SOCK_DGRAM send
1951  *
1952  * Allocate a record consisting of 3 mbufs in the sequence of
1953  * from -> control -> data and append it to the socket buffer.
1954  *
1955  * The first mbuf carries sender's name and is a pkthdr that stores
1956  * overall length of datagram, its memory consumption and control length.
1957  */
1958 #define	ctllen	PH_loc.thirtytwo[1]
1959 _Static_assert(offsetof(struct pkthdr, memlen) + sizeof(u_int) <=
1960     offsetof(struct pkthdr, ctllen), "unix/dgram can not store ctllen");
1961 static int
1962 uipc_sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio,
1963     struct mbuf *m, struct mbuf *c, int flags, struct thread *td)
1964 {
1965 	struct unpcb *unp, *unp2;
1966 	const struct sockaddr *from;
1967 	struct socket *so2;
1968 	struct sockbuf *sb;
1969 	struct mchain cmc = MCHAIN_INITIALIZER(&cmc);
1970 	struct mbuf *f;
1971 	u_int cc, ctl, mbcnt;
1972 	u_int dcc __diagused, dctl __diagused, dmbcnt __diagused;
1973 	int error, needsopts;
1974 
1975 	MPASS((uio != NULL && m == NULL) || (m != NULL && uio == NULL));
1976 
1977 	error = needsopts = 0;
1978 	f = NULL;
1979 
1980 	if (__predict_false(flags & MSG_OOB)) {
1981 		error = EOPNOTSUPP;
1982 		goto out;
1983 	}
1984 	if (m == NULL) {
1985 		if (__predict_false(uio->uio_resid > unpdg_maxdgram)) {
1986 			error = EMSGSIZE;
1987 			goto out;
1988 		}
1989 		m = m_uiotombuf(uio, M_WAITOK, 0, max_hdr, M_PKTHDR);
1990 		if (__predict_false(m == NULL)) {
1991 			error = EFAULT;
1992 			goto out;
1993 		}
1994 		f = m_gethdr(M_WAITOK, MT_SONAME);
1995 		cc = m->m_pkthdr.len;
1996 		mbcnt = MSIZE + m->m_pkthdr.memlen;
1997 		if (c != NULL &&
1998 		    (error = unp_internalize(c, &cmc, td, &needsopts)))
1999 			goto out;
2000 	} else {
2001 		struct mchain mc;
2002 
2003 		uipc_reset_kernel_mbuf(m, &mc);
2004 		cc = mc.mc_len;
2005 		mbcnt = mc.mc_mlen;
2006 		if (__predict_false(m->m_pkthdr.len > unpdg_maxdgram)) {
2007 			error = EMSGSIZE;
2008 			goto out;
2009 		}
2010 		if ((f = m_gethdr(M_NOWAIT, MT_SONAME)) == NULL) {
2011 			error = ENOBUFS;
2012 			goto out;
2013 		}
2014 	}
2015 
2016 	unp = sotounpcb(so);
2017 	MPASS(unp);
2018 
2019 	/*
2020 	 * XXXGL: would be cool to fully remove so_snd out of the equation
2021 	 * and avoid this lock, which is not only extraneous, but also being
2022 	 * released, thus still leaving possibility for a race.  We can easily
2023 	 * handle SBS_CANTSENDMORE/SS_ISCONNECTED complement in unpcb, but it
2024 	 * is more difficult to invent something to handle so_error.
2025 	 */
2026 	error = SOCK_IO_SEND_LOCK(so, SBLOCKWAIT(flags));
2027 	if (error)
2028 		goto out2;
2029 	SOCK_SENDBUF_LOCK(so);
2030 	if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
2031 		SOCK_SENDBUF_UNLOCK(so);
2032 		error = EPIPE;
2033 		goto out3;
2034 	}
2035 	if (so->so_error != 0) {
2036 		error = so->so_error;
2037 		so->so_error = 0;
2038 		SOCK_SENDBUF_UNLOCK(so);
2039 		goto out3;
2040 	}
2041 	if (((so->so_state & SS_ISCONNECTED) == 0) && addr == NULL) {
2042 		SOCK_SENDBUF_UNLOCK(so);
2043 		error = EDESTADDRREQ;
2044 		goto out3;
2045 	}
2046 	SOCK_SENDBUF_UNLOCK(so);
2047 
2048 	if (addr != NULL) {
2049 		if ((error = unp_connectat(AT_FDCWD, so, addr, td, true)))
2050 			goto out3;
2051 		UNP_PCB_LOCK_ASSERT(unp);
2052 		unp2 = unp->unp_conn;
2053 		UNP_PCB_LOCK_ASSERT(unp2);
2054 	} else {
2055 		UNP_PCB_LOCK(unp);
2056 		unp2 = unp_pcb_lock_peer(unp);
2057 		if (unp2 == NULL) {
2058 			UNP_PCB_UNLOCK(unp);
2059 			error = ENOTCONN;
2060 			goto out3;
2061 		}
2062 	}
2063 
2064 	/* Check for SO_PASS* flags */
2065 	so2 = unp2->unp_socket;
2066 	if ((atomic_load_int(&so2->so_options) & needsopts) != needsopts) {
2067 		error = EPERM;
2068 		goto out4;
2069 	}
2070 
2071 	if (unp2->unp_flags & UNP_WANTCRED_MASK)
2072 		unp_addsockcred(td, &cmc, unp2->unp_flags);
2073 	if (unp->unp_addr != NULL)
2074 		from = (struct sockaddr *)unp->unp_addr;
2075 	else
2076 		from = &sun_noname;
2077 	f->m_len = from->sa_len;
2078 	MPASS(from->sa_len <= MLEN);
2079 	bcopy(from, mtod(f, void *), from->sa_len);
2080 
2081 	/*
2082 	 * Concatenate mbufs: from -> control -> data.
2083 	 * Save overall cc and mbcnt in "from" mbuf.
2084 	 */
2085 	if (!STAILQ_EMPTY(&cmc.mc_q)) {
2086 		f->m_next = mc_first(&cmc);
2087 		mc_last(&cmc)->m_next = m;
2088 		/* XXXGL: This is dirty as well as rollback after ENOBUFS. */
2089 		STAILQ_INIT(&cmc.mc_q);
2090 	} else
2091 		f->m_next = m;
2092 	m = NULL;
2093 	ctl = f->m_len + cmc.mc_len;
2094 	mbcnt += cmc.mc_mlen;
2095 #ifdef INVARIANTS
2096 	dcc = dctl = dmbcnt = 0;
2097 	for (struct mbuf *mb = f; mb != NULL; mb = mb->m_next) {
2098 		if (mb->m_type == MT_DATA)
2099 			dcc += mb->m_len;
2100 		else
2101 			dctl += mb->m_len;
2102 		dmbcnt += MSIZE;
2103 		if (mb->m_flags & M_EXT)
2104 			dmbcnt += mb->m_ext.ext_size;
2105 	}
2106 	MPASS(dcc == cc);
2107 	MPASS(dctl == ctl);
2108 	MPASS(dmbcnt == mbcnt);
2109 #endif
2110 	f->m_pkthdr.len = cc + ctl;
2111 	f->m_pkthdr.memlen = mbcnt;
2112 	f->m_pkthdr.ctllen = ctl;
2113 
2114 	/*
2115 	 * Destination socket buffer selection.
2116 	 *
2117 	 * Unconnected sends, when !(so->so_state & SS_ISCONNECTED) and the
2118 	 * destination address is supplied, create a temporary connection for
2119 	 * the run time of the function (see call to unp_connectat() above and
2120 	 * to unp_disconnect() below).  We distinguish them by condition of
2121 	 * (addr != NULL).  We intentionally avoid adding 'bool connected' for
2122 	 * that condition, since, again, through the run time of this code we
2123 	 * are always connected.  For such "unconnected" sends, the destination
2124 	 * buffer would be the receive buffer of destination socket so2.
2125 	 *
2126 	 * For connected sends, data lands on the send buffer of the sender's
2127 	 * socket "so".  Then, if we just added the very first datagram
2128 	 * on this send buffer, we need to add the send buffer on to the
2129 	 * receiving socket's buffer list.  We put ourselves on top of the
2130 	 * list.  Such logic gives infrequent senders priority over frequent
2131 	 * senders.
2132 	 *
2133 	 * Note on byte count management. As long as event methods kevent(2),
2134 	 * select(2) are not protocol specific (yet), we need to maintain
2135 	 * meaningful values on the receive buffer.  So, the receive buffer
2136 	 * would accumulate counters from all connected buffers potentially
2137 	 * having sb_ccc > sb_hiwat or sb_mbcnt > sb_mbmax.
2138 	 */
2139 	sb = (addr == NULL) ? &so->so_snd : &so2->so_rcv;
2140 	SOCK_RECVBUF_LOCK(so2);
2141 	if (uipc_dgram_sbspace(sb, cc + ctl, mbcnt)) {
2142 		if (addr == NULL && STAILQ_EMPTY(&sb->uxdg_mb))
2143 			TAILQ_INSERT_HEAD(&so2->so_rcv.uxdg_conns, &so->so_snd,
2144 			    uxdg_clist);
2145 		STAILQ_INSERT_TAIL(&sb->uxdg_mb, f, m_stailqpkt);
2146 		sb->uxdg_cc += cc + ctl;
2147 		sb->uxdg_ctl += ctl;
2148 		sb->uxdg_mbcnt += mbcnt;
2149 		so2->so_rcv.sb_acc += cc + ctl;
2150 		so2->so_rcv.sb_ccc += cc + ctl;
2151 		so2->so_rcv.sb_ctl += ctl;
2152 		so2->so_rcv.sb_mbcnt += mbcnt;
2153 		sorwakeup_locked(so2);
2154 		f = NULL;
2155 	} else {
2156 		soroverflow_locked(so2);
2157 		error = ENOBUFS;
2158 		if (f->m_next->m_type == MT_CONTROL) {
2159 			STAILQ_FIRST(&cmc.mc_q) = f->m_next;
2160 			f->m_next = NULL;
2161 		}
2162 	}
2163 
2164 out4:
2165 	if (addr != NULL)
2166 		unp_disconnect(unp, unp2);
2167 	else
2168 		unp_pcb_unlock_pair(unp, unp2);
2169 
2170 	td->td_ru.ru_msgsnd++;
2171 
2172 out3:
2173 	SOCK_IO_SEND_UNLOCK(so);
2174 out2:
2175 	if (!mc_empty(&cmc))
2176 		unp_scan(mc_first(&cmc), unp_freerights);
2177 out:
2178 	if (f)
2179 		m_freem(f);
2180 	mc_freem(&cmc);
2181 	if (m)
2182 		m_freem(m);
2183 
2184 	return (error);
2185 }
2186 
2187 /*
2188  * PF_UNIX/SOCK_DGRAM receive with MSG_PEEK.
2189  * The mbuf has already been unlinked from the uxdg_mb of socket buffer
2190  * and needs to be linked onto uxdg_peeked of receive socket buffer.
2191  */
2192 static int
2193 uipc_peek_dgram(struct socket *so, struct mbuf *m, struct sockaddr **psa,
2194     struct uio *uio, struct mbuf **controlp, int *flagsp)
2195 {
2196 	ssize_t len = 0;
2197 	int error;
2198 
2199 	so->so_rcv.uxdg_peeked = m;
2200 	so->so_rcv.uxdg_cc += m->m_pkthdr.len;
2201 	so->so_rcv.uxdg_ctl += m->m_pkthdr.ctllen;
2202 	so->so_rcv.uxdg_mbcnt += m->m_pkthdr.memlen;
2203 	SOCK_RECVBUF_UNLOCK(so);
2204 
2205 	KASSERT(m->m_type == MT_SONAME, ("m->m_type == %d", m->m_type));
2206 	if (psa != NULL)
2207 		*psa = sodupsockaddr(mtod(m, struct sockaddr *), M_WAITOK);
2208 
2209 	m = m->m_next;
2210 	KASSERT(m, ("%s: no data or control after soname", __func__));
2211 
2212 	/*
2213 	 * With MSG_PEEK the control isn't executed, just copied.
2214 	 */
2215 	while (m != NULL && m->m_type == MT_CONTROL) {
2216 		if (controlp != NULL) {
2217 			*controlp = m_copym(m, 0, m->m_len, M_WAITOK);
2218 			controlp = &(*controlp)->m_next;
2219 		}
2220 		m = m->m_next;
2221 	}
2222 	KASSERT(m == NULL || m->m_type == MT_DATA,
2223 	    ("%s: not MT_DATA mbuf %p", __func__, m));
2224 	while (m != NULL && uio->uio_resid > 0) {
2225 		len = uio->uio_resid;
2226 		if (len > m->m_len)
2227 			len = m->m_len;
2228 		error = uiomove(mtod(m, char *), (int)len, uio);
2229 		if (error) {
2230 			SOCK_IO_RECV_UNLOCK(so);
2231 			return (error);
2232 		}
2233 		if (len == m->m_len)
2234 			m = m->m_next;
2235 	}
2236 	SOCK_IO_RECV_UNLOCK(so);
2237 
2238 	if (flagsp != NULL) {
2239 		if (m != NULL) {
2240 			if (*flagsp & MSG_TRUNC) {
2241 				/* Report real length of the packet */
2242 				uio->uio_resid -= m_length(m, NULL) - len;
2243 			}
2244 			*flagsp |= MSG_TRUNC;
2245 		} else
2246 			*flagsp &= ~MSG_TRUNC;
2247 	}
2248 
2249 	return (0);
2250 }
2251 
2252 /*
2253  * PF_UNIX/SOCK_DGRAM receive
2254  */
2255 static int
2256 uipc_soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio,
2257     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2258 {
2259 	struct sockbuf *sb = NULL;
2260 	struct mbuf *m;
2261 	int flags, error;
2262 	ssize_t len = 0;
2263 	bool nonblock;
2264 
2265 	MPASS(mp0 == NULL);
2266 
2267 	if (psa != NULL)
2268 		*psa = NULL;
2269 	if (controlp != NULL)
2270 		*controlp = NULL;
2271 
2272 	flags = flagsp != NULL ? *flagsp : 0;
2273 	nonblock = (so->so_state & SS_NBIO) ||
2274 	    (flags & (MSG_DONTWAIT | MSG_NBIO));
2275 
2276 	error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags));
2277 	if (__predict_false(error))
2278 		return (error);
2279 
2280 	/*
2281 	 * Loop blocking while waiting for a datagram.  Prioritize connected
2282 	 * peers over unconnected sends.  Set sb to selected socket buffer
2283 	 * containing an mbuf on exit from the wait loop.  A datagram that
2284 	 * had already been peeked at has top priority.
2285 	 */
2286 	SOCK_RECVBUF_LOCK(so);
2287 	while ((m = so->so_rcv.uxdg_peeked) == NULL &&
2288 	    (sb = TAILQ_FIRST(&so->so_rcv.uxdg_conns)) == NULL &&
2289 	    (m = STAILQ_FIRST(&so->so_rcv.uxdg_mb)) == NULL) {
2290 		if (so->so_error) {
2291 			error = so->so_error;
2292 			if (!(flags & MSG_PEEK))
2293 				so->so_error = 0;
2294 			SOCK_RECVBUF_UNLOCK(so);
2295 			SOCK_IO_RECV_UNLOCK(so);
2296 			return (error);
2297 		}
2298 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE ||
2299 		    uio->uio_resid == 0) {
2300 			SOCK_RECVBUF_UNLOCK(so);
2301 			SOCK_IO_RECV_UNLOCK(so);
2302 			return (0);
2303 		}
2304 		if (nonblock) {
2305 			SOCK_RECVBUF_UNLOCK(so);
2306 			SOCK_IO_RECV_UNLOCK(so);
2307 			return (EWOULDBLOCK);
2308 		}
2309 		error = sbwait(so, SO_RCV);
2310 		if (error) {
2311 			SOCK_RECVBUF_UNLOCK(so);
2312 			SOCK_IO_RECV_UNLOCK(so);
2313 			return (error);
2314 		}
2315 	}
2316 
2317 	if (sb == NULL)
2318 		sb = &so->so_rcv;
2319 	else if (m == NULL)
2320 		m = STAILQ_FIRST(&sb->uxdg_mb);
2321 	else
2322 		MPASS(m == so->so_rcv.uxdg_peeked);
2323 
2324 	MPASS(sb->uxdg_cc > 0);
2325 	M_ASSERTPKTHDR(m);
2326 	KASSERT(m->m_type == MT_SONAME, ("m->m_type == %d", m->m_type));
2327 
2328 	if (uio->uio_td)
2329 		uio->uio_td->td_ru.ru_msgrcv++;
2330 
2331 	if (__predict_true(m != so->so_rcv.uxdg_peeked)) {
2332 		STAILQ_REMOVE_HEAD(&sb->uxdg_mb, m_stailqpkt);
2333 		if (STAILQ_EMPTY(&sb->uxdg_mb) && sb != &so->so_rcv)
2334 			TAILQ_REMOVE(&so->so_rcv.uxdg_conns, sb, uxdg_clist);
2335 	} else
2336 		so->so_rcv.uxdg_peeked = NULL;
2337 
2338 	sb->uxdg_cc -= m->m_pkthdr.len;
2339 	sb->uxdg_ctl -= m->m_pkthdr.ctllen;
2340 	sb->uxdg_mbcnt -= m->m_pkthdr.memlen;
2341 
2342 	if (__predict_false(flags & MSG_PEEK))
2343 		return (uipc_peek_dgram(so, m, psa, uio, controlp, flagsp));
2344 
2345 	so->so_rcv.sb_acc -= m->m_pkthdr.len;
2346 	so->so_rcv.sb_ccc -= m->m_pkthdr.len;
2347 	so->so_rcv.sb_ctl -= m->m_pkthdr.ctllen;
2348 	so->so_rcv.sb_mbcnt -= m->m_pkthdr.memlen;
2349 	SOCK_RECVBUF_UNLOCK(so);
2350 
2351 	if (psa != NULL)
2352 		*psa = sodupsockaddr(mtod(m, struct sockaddr *), M_WAITOK);
2353 	m = m_free(m);
2354 	KASSERT(m, ("%s: no data or control after soname", __func__));
2355 
2356 	/*
2357 	 * Packet to copyout() is now in 'm' and it is disconnected from the
2358 	 * queue.
2359 	 *
2360 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
2361 	 * in the first mbuf chain on the socket buffer.  We call into the
2362 	 * unp_externalize() to perform externalization (or freeing if
2363 	 * controlp == NULL). In some cases there can be only MT_CONTROL mbufs
2364 	 * without MT_DATA mbufs.
2365 	 */
2366 	while (m != NULL && m->m_type == MT_CONTROL) {
2367 		error = unp_externalize(so, m, controlp, flags);
2368 		m = m_free(m);
2369 		if (error != 0) {
2370 			SOCK_IO_RECV_UNLOCK(so);
2371 			unp_scan(m, unp_freerights);
2372 			m_freem(m);
2373 			return (error);
2374 		}
2375 		if (controlp != NULL) {
2376 			while (*controlp != NULL)
2377 				controlp = &(*controlp)->m_next;
2378 		}
2379 	}
2380 	KASSERT(m == NULL || m->m_type == MT_DATA,
2381 	    ("%s: not MT_DATA mbuf %p", __func__, m));
2382 	while (m != NULL && uio->uio_resid > 0) {
2383 		len = uio->uio_resid;
2384 		if (len > m->m_len)
2385 			len = m->m_len;
2386 		error = uiomove(mtod(m, char *), (int)len, uio);
2387 		if (error) {
2388 			SOCK_IO_RECV_UNLOCK(so);
2389 			m_freem(m);
2390 			return (error);
2391 		}
2392 		if (len == m->m_len)
2393 			m = m_free(m);
2394 		else {
2395 			m->m_data += len;
2396 			m->m_len -= len;
2397 		}
2398 	}
2399 	SOCK_IO_RECV_UNLOCK(so);
2400 
2401 	if (m != NULL) {
2402 		if (flagsp != NULL) {
2403 			if (flags & MSG_TRUNC) {
2404 				/* Report real length of the packet */
2405 				uio->uio_resid -= m_length(m, NULL);
2406 			}
2407 			*flagsp |= MSG_TRUNC;
2408 		}
2409 		m_freem(m);
2410 	} else if (flagsp != NULL)
2411 		*flagsp &= ~MSG_TRUNC;
2412 
2413 	return (0);
2414 }
2415 
2416 static int
2417 uipc_sendfile_wait(struct socket *so, off_t need, int *space)
2418 {
2419 	struct unpcb *unp2;
2420 	struct socket *so2;
2421 	struct sockbuf *sb;
2422 	bool nonblock, sockref;
2423 	int error;
2424 
2425 	MPASS(so->so_type == SOCK_STREAM);
2426 	MPASS(need > 0);
2427 	MPASS(space != NULL);
2428 
2429 	nonblock = so->so_state & SS_NBIO;
2430 	sockref = false;
2431 
2432 	if (__predict_false((so->so_state & SS_ISCONNECTED) == 0))
2433 		return (ENOTCONN);
2434 
2435 	if (__predict_false((error = uipc_lock_peer(so, &unp2)) != 0))
2436 		return (error);
2437 
2438 	so2 = unp2->unp_socket;
2439 	sb = &so2->so_rcv;
2440 	SOCK_RECVBUF_LOCK(so2);
2441 	UNP_PCB_UNLOCK(unp2);
2442 	while ((*space = uipc_stream_sbspace(sb)) < need &&
2443 	    (*space < so->so_snd.sb_hiwat / 2)) {
2444 		UIPC_STREAM_SBCHECK(sb);
2445 		if (nonblock) {
2446 			SOCK_RECVBUF_UNLOCK(so2);
2447 			return (EAGAIN);
2448 		}
2449 		if (!sockref)
2450 			soref(so2);
2451 		error = uipc_stream_sbwait(so2, so->so_snd.sb_timeo);
2452 		if (error == 0 &&
2453 		    __predict_false(sb->sb_state & SBS_CANTRCVMORE))
2454 			error = EPIPE;
2455 		if (error) {
2456 			SOCK_RECVBUF_UNLOCK(so2);
2457 			sorele(so2);
2458 			return (error);
2459 		}
2460 	}
2461 	UIPC_STREAM_SBCHECK(sb);
2462 	SOCK_RECVBUF_UNLOCK(so2);
2463 	if (sockref)
2464 		sorele(so2);
2465 
2466 	return (0);
2467 }
2468 
2469 /*
2470  * Although this is a pr_send method, for unix(4) it is called only via
2471  * sendfile(2) path.  This means we can be sure that mbufs are clear of
2472  * any extra flags and don't require any conditioning.
2473  */
2474 static int
2475 uipc_sendfile(struct socket *so, int flags, struct mbuf *m,
2476     struct sockaddr *from, struct mbuf *control, struct thread *td)
2477 {
2478 	struct mchain mc;
2479 	struct unpcb *unp2;
2480 	struct socket *so2;
2481 	struct sockbuf *sb;
2482 	bool notready, wakeup;
2483 	int error;
2484 
2485 	MPASS(so->so_type == SOCK_STREAM);
2486 	MPASS(from == NULL && control == NULL);
2487 	KASSERT(!(m->m_flags & M_EXTPG),
2488 	    ("unix(4): TLS sendfile(2) not supported"));
2489 
2490 	notready = flags & PRUS_NOTREADY;
2491 
2492 	if (__predict_false((so->so_state & SS_ISCONNECTED) == 0)) {
2493 		error = ENOTCONN;
2494 		goto out;
2495 	}
2496 
2497 	if (__predict_false((error = uipc_lock_peer(so, &unp2)) != 0))
2498 		goto out;
2499 
2500 	mc_init_m(&mc, m);
2501 
2502 	so2 = unp2->unp_socket;
2503 	sb = &so2->so_rcv;
2504 	SOCK_RECVBUF_LOCK(so2);
2505 	UNP_PCB_UNLOCK(unp2);
2506 	UIPC_STREAM_SBCHECK(sb);
2507 	sb->sb_ccc += mc.mc_len;
2508 	sb->sb_mbcnt += mc.mc_mlen;
2509 	if (sb->uxst_fnrdy == NULL) {
2510 		if (notready) {
2511 			wakeup = false;
2512 			STAILQ_FOREACH(m, &mc.mc_q, m_stailq) {
2513 				if (m->m_flags & M_NOTREADY) {
2514 					sb->uxst_fnrdy = m;
2515 					break;
2516 				} else {
2517 					sb->sb_acc += m->m_len;
2518 					wakeup = true;
2519 				}
2520 			}
2521 		} else {
2522 			wakeup = true;
2523 			sb->sb_acc += mc.mc_len;
2524 		}
2525 	} else {
2526 		wakeup = false;
2527 	}
2528 	STAILQ_CONCAT(&sb->uxst_mbq, &mc.mc_q);
2529 	UIPC_STREAM_SBCHECK(sb);
2530 	if (wakeup)
2531 		sorwakeup_locked(so2);
2532 	else
2533 		SOCK_RECVBUF_UNLOCK(so2);
2534 
2535 	return (0);
2536 out:
2537 	/*
2538 	 * In case of not ready data, uipc_ready() is responsible
2539 	 * for freeing memory.
2540 	 */
2541 	if (m != NULL && !notready)
2542 		m_freem(m);
2543 
2544 	return (error);
2545 }
2546 
2547 static int
2548 uipc_sbready(struct sockbuf *sb, struct mbuf *m, int count)
2549 {
2550 	bool blocker;
2551 
2552 	/* assert locked */
2553 
2554 	blocker = (sb->uxst_fnrdy == m);
2555 	STAILQ_FOREACH_FROM(m, &sb->uxst_mbq, m_stailq) {
2556 		if (count > 0) {
2557 			MPASS(m->m_flags & M_NOTREADY);
2558 			m->m_flags &= ~M_NOTREADY;
2559 			if (blocker)
2560 				sb->sb_acc += m->m_len;
2561 			count--;
2562 		} else if (m->m_flags & M_NOTREADY)
2563 			break;
2564 		else if (blocker)
2565 			sb->sb_acc += m->m_len;
2566 	}
2567 	if (blocker) {
2568 		sb->uxst_fnrdy = m;
2569 		return (0);
2570 	} else
2571 		return (EINPROGRESS);
2572 }
2573 
2574 static bool
2575 uipc_ready_scan(struct socket *so, struct mbuf *m, int count, int *errorp)
2576 {
2577 	struct mbuf *mb;
2578 	struct sockbuf *sb;
2579 
2580 	SOCK_LOCK(so);
2581 	if (SOLISTENING(so)) {
2582 		SOCK_UNLOCK(so);
2583 		return (false);
2584 	}
2585 	mb = NULL;
2586 	sb = &so->so_rcv;
2587 	SOCK_RECVBUF_LOCK(so);
2588 	if (sb->uxst_fnrdy != NULL) {
2589 		STAILQ_FOREACH(mb, &sb->uxst_mbq, m_stailq) {
2590 			if (mb == m) {
2591 				*errorp = uipc_sbready(sb, m, count);
2592 				break;
2593 			}
2594 		}
2595 	}
2596 	SOCK_RECVBUF_UNLOCK(so);
2597 	SOCK_UNLOCK(so);
2598 	return (mb != NULL);
2599 }
2600 
2601 static int
2602 uipc_ready(struct socket *so, struct mbuf *m, int count)
2603 {
2604 	struct unpcb *unp, *unp2;
2605 	int error;
2606 
2607 	MPASS(so->so_type == SOCK_STREAM);
2608 
2609 	if (__predict_true(uipc_lock_peer(so, &unp2) == 0)) {
2610 		struct socket *so2;
2611 		struct sockbuf *sb;
2612 
2613 		so2 = unp2->unp_socket;
2614 		sb = &so2->so_rcv;
2615 		SOCK_RECVBUF_LOCK(so2);
2616 		UNP_PCB_UNLOCK(unp2);
2617 		UIPC_STREAM_SBCHECK(sb);
2618 		error = uipc_sbready(sb, m, count);
2619 		UIPC_STREAM_SBCHECK(sb);
2620 		if (error == 0)
2621 			sorwakeup_locked(so2);
2622 		else
2623 			SOCK_RECVBUF_UNLOCK(so2);
2624 	} else {
2625 		/*
2626 		 * The receiving socket has been disconnected, but may still
2627 		 * be valid.  In this case, the not-ready mbufs are still
2628 		 * present in its socket buffer, so perform an exhaustive
2629 		 * search before giving up and freeing the mbufs.
2630 		 */
2631 		UNP_LINK_RLOCK();
2632 		LIST_FOREACH(unp, &unp_shead, unp_link) {
2633 			if (uipc_ready_scan(unp->unp_socket, m, count, &error))
2634 				break;
2635 		}
2636 		UNP_LINK_RUNLOCK();
2637 
2638 		if (unp == NULL) {
2639 			for (int i = 0; i < count; i++)
2640 				m = m_free(m);
2641 			return (ECONNRESET);
2642 		}
2643 	}
2644 	return (error);
2645 }
2646 
2647 static int
2648 uipc_sense(struct socket *so, struct stat *sb)
2649 {
2650 	struct unpcb *unp;
2651 
2652 	unp = sotounpcb(so);
2653 	KASSERT(unp != NULL, ("uipc_sense: unp == NULL"));
2654 
2655 	sb->st_blksize = so->so_snd.sb_hiwat;
2656 	sb->st_dev = NODEV;
2657 	sb->st_ino = unp->unp_ino;
2658 	return (0);
2659 }
2660 
2661 static int
2662 uipc_shutdown(struct socket *so, enum shutdown_how how)
2663 {
2664 	struct unpcb *unp = sotounpcb(so);
2665 	int error;
2666 
2667 	SOCK_LOCK(so);
2668 	if (SOLISTENING(so)) {
2669 		if (how != SHUT_WR) {
2670 			so->so_error = ECONNABORTED;
2671 			solisten_wakeup(so);    /* unlocks so */
2672 		} else
2673 			SOCK_UNLOCK(so);
2674 		return (ENOTCONN);
2675 	} else if ((so->so_state &
2676 	    (SS_ISCONNECTED | SS_ISCONNECTING | SS_ISDISCONNECTING)) == 0) {
2677 		/*
2678 		 * POSIX mandates us to just return ENOTCONN when shutdown(2) is
2679 		 * invoked on a datagram sockets, however historically we would
2680 		 * actually tear socket down.  This is known to be leveraged by
2681 		 * some applications to unblock process waiting in recv(2) by
2682 		 * other process that it shares that socket with.  Try to meet
2683 		 * both backward-compatibility and POSIX requirements by forcing
2684 		 * ENOTCONN but still flushing buffers and performing wakeup(9).
2685 		 *
2686 		 * XXXGL: it remains unknown what applications expect this
2687 		 * behavior and is this isolated to unix/dgram or inet/dgram or
2688 		 * both.  See: D10351, D3039.
2689 		 */
2690 		error = ENOTCONN;
2691 		if (so->so_type != SOCK_DGRAM) {
2692 			SOCK_UNLOCK(so);
2693 			return (error);
2694 		}
2695 	} else
2696 		error = 0;
2697 	SOCK_UNLOCK(so);
2698 
2699 	switch (how) {
2700 	case SHUT_RD:
2701 		if (so->so_type == SOCK_DGRAM)
2702 			socantrcvmore(so);
2703 		else
2704 			uipc_cantrcvmore(so);
2705 		unp_dispose(so);
2706 		break;
2707 	case SHUT_RDWR:
2708 		if (so->so_type == SOCK_DGRAM)
2709 			socantrcvmore(so);
2710 		else
2711 			uipc_cantrcvmore(so);
2712 		unp_dispose(so);
2713 		/* FALLTHROUGH */
2714 	case SHUT_WR:
2715 		if (so->so_type == SOCK_DGRAM) {
2716 			socantsendmore(so);
2717 		} else {
2718 			UNP_PCB_LOCK(unp);
2719 			if (unp->unp_conn != NULL)
2720 				uipc_cantrcvmore(unp->unp_conn->unp_socket);
2721 			UNP_PCB_UNLOCK(unp);
2722 		}
2723 	}
2724 	wakeup(&so->so_timeo);
2725 
2726 	return (error);
2727 }
2728 
2729 static int
2730 uipc_sockaddr(struct socket *so, struct sockaddr *ret)
2731 {
2732 	struct unpcb *unp;
2733 	const struct sockaddr *sa;
2734 
2735 	unp = sotounpcb(so);
2736 	KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL"));
2737 
2738 	UNP_PCB_LOCK(unp);
2739 	if (unp->unp_addr != NULL)
2740 		sa = (struct sockaddr *) unp->unp_addr;
2741 	else
2742 		sa = &sun_noname;
2743 	bcopy(sa, ret, sa->sa_len);
2744 	UNP_PCB_UNLOCK(unp);
2745 	return (0);
2746 }
2747 
2748 static int
2749 uipc_ctloutput(struct socket *so, struct sockopt *sopt)
2750 {
2751 	struct unpcb *unp;
2752 	struct xucred xu;
2753 	int error, optval;
2754 
2755 	if (sopt->sopt_level != SOL_LOCAL)
2756 		return (EINVAL);
2757 
2758 	unp = sotounpcb(so);
2759 	KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL"));
2760 	error = 0;
2761 	switch (sopt->sopt_dir) {
2762 	case SOPT_GET:
2763 		switch (sopt->sopt_name) {
2764 		case LOCAL_PEERCRED:
2765 			UNP_PCB_LOCK(unp);
2766 			if (unp->unp_flags & UNP_HAVEPC)
2767 				xu = unp->unp_peercred;
2768 			else {
2769 				if (so->so_proto->pr_flags & PR_CONNREQUIRED)
2770 					error = ENOTCONN;
2771 				else
2772 					error = EINVAL;
2773 			}
2774 			UNP_PCB_UNLOCK(unp);
2775 			if (error != 0)
2776 				break;
2777 #ifdef COMPAT_FREEBSD32
2778 			if (sopt->sopt_td &&
2779 			    SV_PROC_FLAG(sopt->sopt_td->td_proc, SV_ILP32))
2780 			{
2781 				struct xucred32 xu32 = {};
2782 				int i;
2783 
2784 				xu32.cr_version = xu.cr_version;
2785 				xu32.cr_uid = xu.cr_uid;
2786 				xu32.cr_ngroups = xu.cr_ngroups;
2787 				for (i = 0; i < XU_NGROUPS; i++)
2788 					xu32.cr_groups[i] = xu.cr_groups[i];
2789 				xu32.cr_pid = xu.cr_pid;
2790 				error = sooptcopyout(sopt, &xu32, sizeof(xu32));
2791 				break;
2792 			}
2793 #endif
2794 			error = sooptcopyout(sopt, &xu, sizeof(xu));
2795 			break;
2796 
2797 		case LOCAL_CREDS:
2798 			/* Unlocked read. */
2799 			optval = unp->unp_flags & UNP_WANTCRED_ONESHOT ? 1 : 0;
2800 			error = sooptcopyout(sopt, &optval, sizeof(optval));
2801 			break;
2802 
2803 		case LOCAL_CREDS_PERSISTENT:
2804 			/* Unlocked read. */
2805 			optval = unp->unp_flags & UNP_WANTCRED_ALWAYS ? 1 : 0;
2806 			error = sooptcopyout(sopt, &optval, sizeof(optval));
2807 			break;
2808 
2809 		default:
2810 			error = EOPNOTSUPP;
2811 			break;
2812 		}
2813 		break;
2814 
2815 	case SOPT_SET:
2816 		switch (sopt->sopt_name) {
2817 		case LOCAL_CREDS:
2818 		case LOCAL_CREDS_PERSISTENT:
2819 			error = sooptcopyin(sopt, &optval, sizeof(optval),
2820 					    sizeof(optval));
2821 			if (error)
2822 				break;
2823 
2824 #define	OPTSET(bit, exclusive) do {					\
2825 	UNP_PCB_LOCK(unp);						\
2826 	if (optval) {							\
2827 		if ((unp->unp_flags & (exclusive)) != 0) {		\
2828 			UNP_PCB_UNLOCK(unp);				\
2829 			error = EINVAL;					\
2830 			break;						\
2831 		}							\
2832 		unp->unp_flags |= (bit);				\
2833 	} else								\
2834 		unp->unp_flags &= ~(bit);				\
2835 	UNP_PCB_UNLOCK(unp);						\
2836 } while (0)
2837 
2838 			switch (sopt->sopt_name) {
2839 			case LOCAL_CREDS:
2840 				OPTSET(UNP_WANTCRED_ONESHOT, UNP_WANTCRED_ALWAYS);
2841 				break;
2842 
2843 			case LOCAL_CREDS_PERSISTENT:
2844 				OPTSET(UNP_WANTCRED_ALWAYS, UNP_WANTCRED_ONESHOT);
2845 				break;
2846 
2847 			default:
2848 				break;
2849 			}
2850 			break;
2851 #undef	OPTSET
2852 		default:
2853 			error = ENOPROTOOPT;
2854 			break;
2855 		}
2856 		break;
2857 
2858 	default:
2859 		error = EOPNOTSUPP;
2860 		break;
2861 	}
2862 	return (error);
2863 }
2864 
2865 static int
2866 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
2867 {
2868 
2869 	return (unp_connectat(AT_FDCWD, so, nam, td, false));
2870 }
2871 
2872 static int
2873 unp_connectat(int fd, struct socket *so, struct sockaddr *nam,
2874     struct thread *td, bool return_locked)
2875 {
2876 	struct mtx *vplock;
2877 	struct sockaddr_un *soun;
2878 	struct vnode *vp;
2879 	struct socket *so2;
2880 	struct unpcb *unp, *unp2, *unp3;
2881 	struct nameidata nd;
2882 	char buf[SOCK_MAXADDRLEN];
2883 	struct sockaddr *sa;
2884 	cap_rights_t rights;
2885 	int error, len;
2886 	bool connreq;
2887 
2888 	CURVNET_ASSERT_SET();
2889 
2890 	if (nam->sa_family != AF_UNIX)
2891 		return (EAFNOSUPPORT);
2892 	if (nam->sa_len > sizeof(struct sockaddr_un))
2893 		return (EINVAL);
2894 	len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
2895 	if (len <= 0)
2896 		return (EINVAL);
2897 	soun = (struct sockaddr_un *)nam;
2898 	bcopy(soun->sun_path, buf, len);
2899 	buf[len] = 0;
2900 
2901 	error = 0;
2902 	unp = sotounpcb(so);
2903 	UNP_PCB_LOCK(unp);
2904 	for (;;) {
2905 		/*
2906 		 * Wait for connection state to stabilize.  If a connection
2907 		 * already exists, give up.  For datagram sockets, which permit
2908 		 * multiple consecutive connect(2) calls, upper layers are
2909 		 * responsible for disconnecting in advance of a subsequent
2910 		 * connect(2), but this is not synchronized with PCB connection
2911 		 * state.
2912 		 *
2913 		 * Also make sure that no threads are currently attempting to
2914 		 * lock the peer socket, to ensure that unp_conn cannot
2915 		 * transition between two valid sockets while locks are dropped.
2916 		 */
2917 		if (SOLISTENING(so))
2918 			error = EOPNOTSUPP;
2919 		else if (unp->unp_conn != NULL)
2920 			error = EISCONN;
2921 		else if ((unp->unp_flags & UNP_CONNECTING) != 0) {
2922 			error = EALREADY;
2923 		}
2924 		if (error != 0) {
2925 			UNP_PCB_UNLOCK(unp);
2926 			return (error);
2927 		}
2928 		if (unp->unp_pairbusy > 0) {
2929 			unp->unp_flags |= UNP_WAITING;
2930 			mtx_sleep(unp, UNP_PCB_LOCKPTR(unp), 0, "unpeer", 0);
2931 			continue;
2932 		}
2933 		break;
2934 	}
2935 	unp->unp_flags |= UNP_CONNECTING;
2936 	UNP_PCB_UNLOCK(unp);
2937 
2938 	connreq = (so->so_proto->pr_flags & PR_CONNREQUIRED) != 0;
2939 	if (connreq)
2940 		sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
2941 	else
2942 		sa = NULL;
2943 	NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF |
2944 	    (fd == AT_FDCWD ? 0 : EMPTYPATH), UIO_SYSSPACE, buf, fd,
2945 	    cap_rights_init_one(&rights, CAP_CONNECTAT));
2946 	error = namei(&nd);
2947 	if (error)
2948 		vp = NULL;
2949 	else
2950 		vp = nd.ni_vp;
2951 	ASSERT_VOP_LOCKED(vp, "unp_connect");
2952 	if (error)
2953 		goto bad;
2954 	NDFREE_PNBUF(&nd);
2955 
2956 	if (vp->v_type != VSOCK) {
2957 		error = ENOTSOCK;
2958 		goto bad;
2959 	}
2960 #ifdef MAC
2961 	error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD);
2962 	if (error)
2963 		goto bad;
2964 #endif
2965 	error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td);
2966 	if (error)
2967 		goto bad;
2968 
2969 	unp = sotounpcb(so);
2970 	KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
2971 
2972 	vplock = mtx_pool_find(unp_vp_mtxpool, vp);
2973 	mtx_lock(vplock);
2974 	VOP_UNP_CONNECT(vp, &unp2);
2975 	if (unp2 == NULL) {
2976 		error = ECONNREFUSED;
2977 		goto bad2;
2978 	}
2979 	so2 = unp2->unp_socket;
2980 	if (so->so_type != so2->so_type) {
2981 		error = EPROTOTYPE;
2982 		goto bad2;
2983 	}
2984 	if (connreq) {
2985 		if (SOLISTENING(so2))
2986 			so2 = solisten_clone(so2);
2987 		else
2988 			so2 = NULL;
2989 		if (so2 == NULL) {
2990 			error = ECONNREFUSED;
2991 			goto bad2;
2992 		}
2993 		if ((error = uipc_attach(so2, 0, NULL)) != 0) {
2994 			sodealloc(so2);
2995 			goto bad2;
2996 		}
2997 		unp3 = sotounpcb(so2);
2998 		unp_pcb_lock_pair(unp2, unp3);
2999 		if (unp2->unp_addr != NULL) {
3000 			bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len);
3001 			unp3->unp_addr = (struct sockaddr_un *) sa;
3002 			sa = NULL;
3003 		}
3004 
3005 		unp_copy_peercred(td, unp3, unp, unp2);
3006 
3007 		UNP_PCB_UNLOCK(unp2);
3008 		unp2 = unp3;
3009 
3010 		/*
3011 		 * It is safe to block on the PCB lock here since unp2 is
3012 		 * nascent and cannot be connected to any other sockets.
3013 		 */
3014 		UNP_PCB_LOCK(unp);
3015 #ifdef MAC
3016 		mac_socketpeer_set_from_socket(so, so2);
3017 		mac_socketpeer_set_from_socket(so2, so);
3018 #endif
3019 	} else {
3020 		unp_pcb_lock_pair(unp, unp2);
3021 	}
3022 	KASSERT(unp2 != NULL && so2 != NULL && unp2->unp_socket == so2 &&
3023 	    sotounpcb(so2) == unp2,
3024 	    ("%s: unp2 %p so2 %p", __func__, unp2, so2));
3025 	unp_connect2(so, so2, connreq);
3026 	if (connreq)
3027 		(void)solisten_enqueue(so2, SS_ISCONNECTED);
3028 	KASSERT((unp->unp_flags & UNP_CONNECTING) != 0,
3029 	    ("%s: unp %p has UNP_CONNECTING clear", __func__, unp));
3030 	unp->unp_flags &= ~UNP_CONNECTING;
3031 	if (!return_locked)
3032 		unp_pcb_unlock_pair(unp, unp2);
3033 bad2:
3034 	mtx_unlock(vplock);
3035 bad:
3036 	if (vp != NULL) {
3037 		/*
3038 		 * If we are returning locked (called via uipc_sosend_dgram()),
3039 		 * we need to be sure that vput() won't sleep.  This is
3040 		 * guaranteed by VOP_UNP_CONNECT() call above and unp2 lock.
3041 		 * SOCK_STREAM/SEQPACKET can't request return_locked (yet).
3042 		 */
3043 		MPASS(!(return_locked && connreq));
3044 		vput(vp);
3045 	}
3046 	free(sa, M_SONAME);
3047 	if (__predict_false(error)) {
3048 		UNP_PCB_LOCK(unp);
3049 		KASSERT((unp->unp_flags & UNP_CONNECTING) != 0,
3050 		    ("%s: unp %p has UNP_CONNECTING clear", __func__, unp));
3051 		unp->unp_flags &= ~UNP_CONNECTING;
3052 		UNP_PCB_UNLOCK(unp);
3053 	}
3054 	return (error);
3055 }
3056 
3057 /*
3058  * Set socket peer credentials at connection time.
3059  *
3060  * The client's PCB credentials are copied from its process structure.  The
3061  * server's PCB credentials are copied from the socket on which it called
3062  * listen(2).  uipc_listen cached that process's credentials at the time.
3063  */
3064 void
3065 unp_copy_peercred(struct thread *td, struct unpcb *client_unp,
3066     struct unpcb *server_unp, struct unpcb *listen_unp)
3067 {
3068 	cru2xt(td, &client_unp->unp_peercred);
3069 	client_unp->unp_flags |= UNP_HAVEPC;
3070 
3071 	memcpy(&server_unp->unp_peercred, &listen_unp->unp_peercred,
3072 	    sizeof(server_unp->unp_peercred));
3073 	server_unp->unp_flags |= UNP_HAVEPC;
3074 	client_unp->unp_flags |= (listen_unp->unp_flags & UNP_WANTCRED_MASK);
3075 }
3076 
3077 /*
3078  * unix/stream & unix/seqpacket version of soisconnected().
3079  *
3080  * The crucial thing we are doing here is setting up the uxst_peer linkage,
3081  * holding unp and receive buffer locks of the both sockets.  The disconnect
3082  * procedure does the same.  This gives as a safe way to access the peer in the
3083  * send(2) and recv(2) during the socket lifetime.
3084  *
3085  * The less important thing is event notification of the fact that a socket is
3086  * now connected.  It is unusual for a software to put a socket into event
3087  * mechanism before connect(2), but is supposed to be supported.  Note that
3088  * there can not be any sleeping I/O on the socket, yet, only presence in the
3089  * select/poll/kevent.
3090  *
3091  * This function can be called via two call paths:
3092  * 1) socketpair(2) - in this case socket has not been yet reported to userland
3093  *    and just can't have any event notifications mechanisms set up.  The
3094  *    'wakeup' boolean is always false.
3095  * 2) connect(2) of existing socket to a recent clone of a listener:
3096  *   2.1) Socket that connect(2)s will have 'wakeup' true.  An application
3097  *        could have already put it into event mechanism, is it shall be
3098  *        reported as readable and as writable.
3099  *   2.2) Socket that was just cloned with solisten_clone().  Same as 1).
3100  */
3101 static void
3102 unp_soisconnected(struct socket *so, bool wakeup)
3103 {
3104 	struct socket *so2 = sotounpcb(so)->unp_conn->unp_socket;
3105 	struct sockbuf *sb;
3106 
3107 	SOCK_LOCK_ASSERT(so);
3108 	UNP_PCB_LOCK_ASSERT(sotounpcb(so));
3109 	UNP_PCB_LOCK_ASSERT(sotounpcb(so2));
3110 	SOCK_RECVBUF_LOCK_ASSERT(so);
3111 	SOCK_RECVBUF_LOCK_ASSERT(so2);
3112 
3113 	MPASS(so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET);
3114 	MPASS((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
3115 	    SS_ISDISCONNECTING)) == 0);
3116 	MPASS(so->so_qstate == SQ_NONE);
3117 
3118 	so->so_state &= ~SS_ISDISCONNECTED;
3119 	so->so_state |= SS_ISCONNECTED;
3120 
3121 	sb = &so2->so_rcv;
3122 	sb->uxst_peer = so;
3123 
3124 	if (wakeup) {
3125 		KNOTE_LOCKED(&sb->sb_sel->si_note, 0);
3126 		sb = &so->so_rcv;
3127 		selwakeuppri(sb->sb_sel, PSOCK);
3128 		SOCK_SENDBUF_LOCK_ASSERT(so);
3129 		sb = &so->so_snd;
3130 		selwakeuppri(sb->sb_sel, PSOCK);
3131 		SOCK_SENDBUF_UNLOCK(so);
3132 	}
3133 }
3134 
3135 static void
3136 unp_connect2(struct socket *so, struct socket *so2, bool wakeup)
3137 {
3138 	struct unpcb *unp;
3139 	struct unpcb *unp2;
3140 
3141 	MPASS(so2->so_type == so->so_type);
3142 	unp = sotounpcb(so);
3143 	KASSERT(unp != NULL, ("unp_connect2: unp == NULL"));
3144 	unp2 = sotounpcb(so2);
3145 	KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL"));
3146 
3147 	UNP_PCB_LOCK_ASSERT(unp);
3148 	UNP_PCB_LOCK_ASSERT(unp2);
3149 	KASSERT(unp->unp_conn == NULL,
3150 	    ("%s: socket %p is already connected", __func__, unp));
3151 
3152 	unp->unp_conn = unp2;
3153 	unp_pcb_hold(unp2);
3154 	unp_pcb_hold(unp);
3155 	switch (so->so_type) {
3156 	case SOCK_DGRAM:
3157 		UNP_REF_LIST_LOCK();
3158 		LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
3159 		UNP_REF_LIST_UNLOCK();
3160 		soisconnected(so);
3161 		break;
3162 
3163 	case SOCK_STREAM:
3164 	case SOCK_SEQPACKET:
3165 		KASSERT(unp2->unp_conn == NULL,
3166 		    ("%s: socket %p is already connected", __func__, unp2));
3167 		unp2->unp_conn = unp;
3168 		SOCK_LOCK(so);
3169 		SOCK_LOCK(so2);
3170 		if (wakeup)	/* Avoid LOR with receive buffer lock. */
3171 			SOCK_SENDBUF_LOCK(so);
3172 		SOCK_RECVBUF_LOCK(so);
3173 		SOCK_RECVBUF_LOCK(so2);
3174 		unp_soisconnected(so, wakeup);	/* Will unlock send buffer. */
3175 		unp_soisconnected(so2, false);
3176 		SOCK_RECVBUF_UNLOCK(so);
3177 		SOCK_RECVBUF_UNLOCK(so2);
3178 		SOCK_UNLOCK(so);
3179 		SOCK_UNLOCK(so2);
3180 		break;
3181 
3182 	default:
3183 		panic("unp_connect2");
3184 	}
3185 }
3186 
3187 static void
3188 unp_soisdisconnected(struct socket *so)
3189 {
3190 	SOCK_LOCK_ASSERT(so);
3191 	SOCK_RECVBUF_LOCK_ASSERT(so);
3192 	MPASS(so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET);
3193 	MPASS(!SOLISTENING(so));
3194 	MPASS((so->so_state & (SS_ISCONNECTING | SS_ISDISCONNECTING |
3195 	    SS_ISDISCONNECTED)) == 0);
3196 	MPASS(so->so_state & SS_ISCONNECTED);
3197 
3198 	so->so_state |= SS_ISDISCONNECTED;
3199 	so->so_state &= ~SS_ISCONNECTED;
3200 	so->so_rcv.uxst_peer = NULL;
3201 	selwakeuppri(&so->so_wrsel, PSOCK);
3202 	KNOTE_LOCKED(&so->so_snd.sb_sel->si_note, 0);
3203 	socantrcvmore_locked(so);
3204 }
3205 
3206 static void
3207 unp_disconnect(struct unpcb *unp, struct unpcb *unp2)
3208 {
3209 	struct socket *so, *so2;
3210 	struct mbuf *m = NULL;
3211 #ifdef INVARIANTS
3212 	struct unpcb *unptmp;
3213 #endif
3214 
3215 	UNP_PCB_LOCK_ASSERT(unp);
3216 	UNP_PCB_LOCK_ASSERT(unp2);
3217 	KASSERT(unp->unp_conn == unp2,
3218 	    ("%s: unpcb %p is not connected to %p", __func__, unp, unp2));
3219 
3220 	unp->unp_conn = NULL;
3221 	so = unp->unp_socket;
3222 	so2 = unp2->unp_socket;
3223 	switch (unp->unp_socket->so_type) {
3224 	case SOCK_DGRAM:
3225 		/*
3226 		 * Remove our send socket buffer from the peer's receive buffer.
3227 		 * Move the data to the receive buffer only if it is empty.
3228 		 * This is a protection against a scenario where a peer
3229 		 * connects, floods and disconnects, effectively blocking
3230 		 * sendto() from unconnected sockets.
3231 		 */
3232 		SOCK_RECVBUF_LOCK(so2);
3233 		if (!STAILQ_EMPTY(&so->so_snd.uxdg_mb)) {
3234 			TAILQ_REMOVE(&so2->so_rcv.uxdg_conns, &so->so_snd,
3235 			    uxdg_clist);
3236 			if (__predict_true((so2->so_rcv.sb_state &
3237 			    SBS_CANTRCVMORE) == 0) &&
3238 			    STAILQ_EMPTY(&so2->so_rcv.uxdg_mb)) {
3239 				STAILQ_CONCAT(&so2->so_rcv.uxdg_mb,
3240 				    &so->so_snd.uxdg_mb);
3241 				so2->so_rcv.uxdg_cc += so->so_snd.uxdg_cc;
3242 				so2->so_rcv.uxdg_ctl += so->so_snd.uxdg_ctl;
3243 				so2->so_rcv.uxdg_mbcnt += so->so_snd.uxdg_mbcnt;
3244 			} else {
3245 				m = STAILQ_FIRST(&so->so_snd.uxdg_mb);
3246 				STAILQ_INIT(&so->so_snd.uxdg_mb);
3247 				so2->so_rcv.sb_acc -= so->so_snd.uxdg_cc;
3248 				so2->so_rcv.sb_ccc -= so->so_snd.uxdg_cc;
3249 				so2->so_rcv.sb_ctl -= so->so_snd.uxdg_ctl;
3250 				so2->so_rcv.sb_mbcnt -= so->so_snd.uxdg_mbcnt;
3251 			}
3252 			/* Note: so may reconnect. */
3253 			so->so_snd.uxdg_cc = 0;
3254 			so->so_snd.uxdg_ctl = 0;
3255 			so->so_snd.uxdg_mbcnt = 0;
3256 		}
3257 		SOCK_RECVBUF_UNLOCK(so2);
3258 		UNP_REF_LIST_LOCK();
3259 #ifdef INVARIANTS
3260 		LIST_FOREACH(unptmp, &unp2->unp_refs, unp_reflink) {
3261 			if (unptmp == unp)
3262 				break;
3263 		}
3264 		KASSERT(unptmp != NULL,
3265 		    ("%s: %p not found in reflist of %p", __func__, unp, unp2));
3266 #endif
3267 		LIST_REMOVE(unp, unp_reflink);
3268 		UNP_REF_LIST_UNLOCK();
3269 		SOCK_LOCK(so);
3270 		so->so_state &= ~SS_ISCONNECTED;
3271 		SOCK_UNLOCK(so);
3272 		break;
3273 
3274 	case SOCK_STREAM:
3275 	case SOCK_SEQPACKET:
3276 		SOCK_LOCK(so);
3277 		SOCK_LOCK(so2);
3278 		SOCK_RECVBUF_LOCK(so);
3279 		SOCK_RECVBUF_LOCK(so2);
3280 		unp_soisdisconnected(so);
3281 		MPASS(unp2->unp_conn == unp);
3282 		unp2->unp_conn = NULL;
3283 		unp_soisdisconnected(so2);
3284 		SOCK_UNLOCK(so);
3285 		SOCK_UNLOCK(so2);
3286 		break;
3287 	}
3288 
3289 	if (unp == unp2) {
3290 		unp_pcb_rele_notlast(unp);
3291 		if (!unp_pcb_rele(unp))
3292 			UNP_PCB_UNLOCK(unp);
3293 	} else {
3294 		if (!unp_pcb_rele(unp))
3295 			UNP_PCB_UNLOCK(unp);
3296 		if (!unp_pcb_rele(unp2))
3297 			UNP_PCB_UNLOCK(unp2);
3298 	}
3299 
3300 	if (m != NULL) {
3301 		unp_scan(m, unp_freerights);
3302 		m_freemp(m);
3303 	}
3304 }
3305 
3306 /*
3307  * unp_pcblist() walks the global list of struct unpcb's to generate a
3308  * pointer list, bumping the refcount on each unpcb.  It then copies them out
3309  * sequentially, validating the generation number on each to see if it has
3310  * been detached.  All of this is necessary because copyout() may sleep on
3311  * disk I/O.
3312  */
3313 static int
3314 unp_pcblist(SYSCTL_HANDLER_ARGS)
3315 {
3316 	struct unpcb *unp, **unp_list;
3317 	unp_gen_t gencnt;
3318 	struct xunpgen *xug;
3319 	struct unp_head *head;
3320 	struct xunpcb *xu;
3321 	u_int i;
3322 	int error, n;
3323 
3324 	switch ((intptr_t)arg1) {
3325 	case SOCK_STREAM:
3326 		head = &unp_shead;
3327 		break;
3328 
3329 	case SOCK_DGRAM:
3330 		head = &unp_dhead;
3331 		break;
3332 
3333 	case SOCK_SEQPACKET:
3334 		head = &unp_sphead;
3335 		break;
3336 
3337 	default:
3338 		panic("unp_pcblist: arg1 %d", (int)(intptr_t)arg1);
3339 	}
3340 
3341 	/*
3342 	 * The process of preparing the PCB list is too time-consuming and
3343 	 * resource-intensive to repeat twice on every request.
3344 	 */
3345 	if (req->oldptr == NULL) {
3346 		n = unp_count;
3347 		req->oldidx = 2 * (sizeof *xug)
3348 			+ (n + n/8) * sizeof(struct xunpcb);
3349 		return (0);
3350 	}
3351 
3352 	if (req->newptr != NULL)
3353 		return (EPERM);
3354 
3355 	/*
3356 	 * OK, now we're committed to doing something.
3357 	 */
3358 	xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK | M_ZERO);
3359 	UNP_LINK_RLOCK();
3360 	gencnt = unp_gencnt;
3361 	n = unp_count;
3362 	UNP_LINK_RUNLOCK();
3363 
3364 	xug->xug_len = sizeof *xug;
3365 	xug->xug_count = n;
3366 	xug->xug_gen = gencnt;
3367 	xug->xug_sogen = so_gencnt;
3368 	error = SYSCTL_OUT(req, xug, sizeof *xug);
3369 	if (error) {
3370 		free(xug, M_TEMP);
3371 		return (error);
3372 	}
3373 
3374 	unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
3375 
3376 	UNP_LINK_RLOCK();
3377 	for (unp = LIST_FIRST(head), i = 0; unp && i < n;
3378 	     unp = LIST_NEXT(unp, unp_link)) {
3379 		UNP_PCB_LOCK(unp);
3380 		if (unp->unp_gencnt <= gencnt) {
3381 			if (cr_cansee(req->td->td_ucred,
3382 			    unp->unp_socket->so_cred)) {
3383 				UNP_PCB_UNLOCK(unp);
3384 				continue;
3385 			}
3386 			unp_list[i++] = unp;
3387 			unp_pcb_hold(unp);
3388 		}
3389 		UNP_PCB_UNLOCK(unp);
3390 	}
3391 	UNP_LINK_RUNLOCK();
3392 	n = i;			/* In case we lost some during malloc. */
3393 
3394 	error = 0;
3395 	xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO);
3396 	for (i = 0; i < n; i++) {
3397 		unp = unp_list[i];
3398 		UNP_PCB_LOCK(unp);
3399 		if (unp_pcb_rele(unp))
3400 			continue;
3401 
3402 		if (unp->unp_gencnt <= gencnt) {
3403 			xu->xu_len = sizeof *xu;
3404 			xu->xu_unpp = (uintptr_t)unp;
3405 			/*
3406 			 * XXX - need more locking here to protect against
3407 			 * connect/disconnect races for SMP.
3408 			 */
3409 			if (unp->unp_addr != NULL)
3410 				bcopy(unp->unp_addr, &xu->xu_addr,
3411 				      unp->unp_addr->sun_len);
3412 			else
3413 				bzero(&xu->xu_addr, sizeof(xu->xu_addr));
3414 			if (unp->unp_conn != NULL &&
3415 			    unp->unp_conn->unp_addr != NULL)
3416 				bcopy(unp->unp_conn->unp_addr,
3417 				      &xu->xu_caddr,
3418 				      unp->unp_conn->unp_addr->sun_len);
3419 			else
3420 				bzero(&xu->xu_caddr, sizeof(xu->xu_caddr));
3421 			xu->unp_vnode = (uintptr_t)unp->unp_vnode;
3422 			xu->unp_conn = (uintptr_t)unp->unp_conn;
3423 			xu->xu_firstref = (uintptr_t)LIST_FIRST(&unp->unp_refs);
3424 			xu->xu_nextref = (uintptr_t)LIST_NEXT(unp, unp_reflink);
3425 			xu->unp_gencnt = unp->unp_gencnt;
3426 			sotoxsocket(unp->unp_socket, &xu->xu_socket);
3427 			UNP_PCB_UNLOCK(unp);
3428 			error = SYSCTL_OUT(req, xu, sizeof *xu);
3429 		} else {
3430 			UNP_PCB_UNLOCK(unp);
3431 		}
3432 	}
3433 	free(xu, M_TEMP);
3434 	if (!error) {
3435 		/*
3436 		 * Give the user an updated idea of our state.  If the
3437 		 * generation differs from what we told her before, she knows
3438 		 * that something happened while we were processing this
3439 		 * request, and it might be necessary to retry.
3440 		 */
3441 		xug->xug_gen = unp_gencnt;
3442 		xug->xug_sogen = so_gencnt;
3443 		xug->xug_count = unp_count;
3444 		error = SYSCTL_OUT(req, xug, sizeof *xug);
3445 	}
3446 	free(unp_list, M_TEMP);
3447 	free(xug, M_TEMP);
3448 	return (error);
3449 }
3450 
3451 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist,
3452     CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
3453     (void *)(intptr_t)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
3454     "List of active local datagram sockets");
3455 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist,
3456     CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
3457     (void *)(intptr_t)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
3458     "List of active local stream sockets");
3459 SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist,
3460     CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
3461     (void *)(intptr_t)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb",
3462     "List of active local seqpacket sockets");
3463 
3464 static void
3465 unp_drop(struct unpcb *unp)
3466 {
3467 	struct socket *so;
3468 	struct unpcb *unp2;
3469 
3470 	/*
3471 	 * Regardless of whether the socket's peer dropped the connection
3472 	 * with this socket by aborting or disconnecting, POSIX requires
3473 	 * that ECONNRESET is returned on next connected send(2) in case of
3474 	 * a SOCK_DGRAM socket and EPIPE for SOCK_STREAM.
3475 	 */
3476 	UNP_PCB_LOCK(unp);
3477 	if ((so = unp->unp_socket) != NULL)
3478 		so->so_error =
3479 		    so->so_proto->pr_type == SOCK_DGRAM ? ECONNRESET : EPIPE;
3480 	if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) {
3481 		/* Last reference dropped in unp_disconnect(). */
3482 		unp_pcb_rele_notlast(unp);
3483 		unp_disconnect(unp, unp2);
3484 	} else if (!unp_pcb_rele(unp)) {
3485 		UNP_PCB_UNLOCK(unp);
3486 	}
3487 }
3488 
3489 static void
3490 unp_freerights(struct filedescent **fdep, int fdcount)
3491 {
3492 	struct file *fp;
3493 	int i;
3494 
3495 	KASSERT(fdcount > 0, ("%s: fdcount %d", __func__, fdcount));
3496 
3497 	for (i = 0; i < fdcount; i++) {
3498 		fp = fdep[i]->fde_file;
3499 		filecaps_free(&fdep[i]->fde_caps);
3500 		unp_discard(fp);
3501 	}
3502 	free(fdep[0], M_FILECAPS);
3503 }
3504 
3505 static bool
3506 restrict_rights(struct file *fp, struct thread *td)
3507 {
3508 	struct prison *prison1, *prison2;
3509 
3510 	prison1 = fp->f_cred->cr_prison;
3511 	prison2 = td->td_ucred->cr_prison;
3512 	return (prison1 != prison2 && prison1->pr_root != prison2->pr_root &&
3513 	    prison2 != &prison0);
3514 }
3515 
3516 static int
3517 unp_externalize(const struct socket *so, struct mbuf *control,
3518     struct mbuf **controlp, int flags)
3519 {
3520 	struct thread *td = curthread;		/* XXX */
3521 	struct cmsghdr *cm = mtod(control, struct cmsghdr *);
3522 	int *fdp;
3523 	struct filedesc *fdesc = td->td_proc->p_fd;
3524 	struct filedescent **fdep;
3525 	void *data;
3526 	socklen_t clen = control->m_len, datalen;
3527 	int error, fdflags, newfds;
3528 	u_int newlen;
3529 
3530 	UNP_LINK_UNLOCK_ASSERT();
3531 
3532 	fdflags = ((flags & MSG_CMSG_CLOEXEC) ? O_CLOEXEC : 0) |
3533 	    ((flags & MSG_CMSG_CLOFORK) ? O_CLOFORK : 0);
3534 
3535 	error = 0;
3536 	if (controlp != NULL) /* controlp == NULL => free control messages */
3537 		*controlp = NULL;
3538 	while (cm != NULL) {
3539 		MPASS(clen >= sizeof(*cm) && clen >= cm->cmsg_len);
3540 
3541 		data = CMSG_DATA(cm);
3542 		datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
3543 		if (cm->cmsg_level == SOL_SOCKET
3544 		    && cm->cmsg_type == SCM_RIGHTS) {
3545 			newfds = datalen / sizeof(*fdep);
3546 			if (newfds == 0)
3547 				goto next;
3548 			fdep = data;
3549 
3550 			/*
3551 			 * If we're not outputting the descriptors, free them.
3552 			 *
3553 			 * In the case of having revoked SCM_PASSRIGHTS, the
3554 			 * receiver must have toggled it before trying to
3555 			 * receive control messages- we'll take that as a signal
3556 			 * that they didn't want these, but they raced against
3557 			 * the sender trying to pass files anyways.
3558 			 */
3559 			if (error || controlp == NULL ||
3560 			    (atomic_load_int(&so->so_options) &
3561 			    SO_PASSRIGHTS) == 0) {
3562 				unp_freerights(fdep, newfds);
3563 				goto next;
3564 			}
3565 			FILEDESC_XLOCK(fdesc);
3566 
3567 			/*
3568 			 * Now change each pointer to an fd in the global
3569 			 * table to an integer that is the index to the local
3570 			 * fd table entry that we set up to point to the
3571 			 * global one we are transferring.
3572 			 */
3573 			newlen = newfds * sizeof(int);
3574 			*controlp = sbcreatecontrol(NULL, newlen,
3575 			    SCM_RIGHTS, SOL_SOCKET, M_WAITOK);
3576 
3577 			fdp = (int *)
3578 			    CMSG_DATA(mtod(*controlp, struct cmsghdr *));
3579 			if ((error = fdallocn(td, 0, fdp, newfds))) {
3580 				FILEDESC_XUNLOCK(fdesc);
3581 				unp_freerights(fdep, newfds);
3582 				m_freem(*controlp);
3583 				*controlp = NULL;
3584 				goto next;
3585 			}
3586 			for (int i = 0; i < newfds; i++, fdp++) {
3587 				struct file *fp;
3588 
3589 				fp = fdep[i]->fde_file;
3590 				_finstall(fdesc, fp, *fdp, fdflags |
3591 				    (restrict_rights(fp, td) ?
3592 				    O_RESOLVE_BENEATH : 0), &fdep[i]->fde_caps);
3593 				unp_externalize_fp(fp);
3594 			}
3595 
3596 			/*
3597 			 * The new type indicates that the mbuf data refers to
3598 			 * kernel resources that may need to be released before
3599 			 * the mbuf is freed.
3600 			 */
3601 			m_chtype(*controlp, MT_EXTCONTROL);
3602 			FILEDESC_XUNLOCK(fdesc);
3603 			free(fdep[0], M_FILECAPS);
3604 		} else {
3605 			/* We can just copy anything else across. */
3606 			if (error || controlp == NULL)
3607 				goto next;
3608 			*controlp = sbcreatecontrol(NULL, datalen,
3609 			    cm->cmsg_type, cm->cmsg_level, M_WAITOK);
3610 			bcopy(data,
3611 			    CMSG_DATA(mtod(*controlp, struct cmsghdr *)),
3612 			    datalen);
3613 		}
3614 		controlp = &(*controlp)->m_next;
3615 
3616 next:
3617 		if (CMSG_SPACE(datalen) < clen) {
3618 			clen -= CMSG_SPACE(datalen);
3619 			cm = (struct cmsghdr *)
3620 			    ((caddr_t)cm + CMSG_SPACE(datalen));
3621 		} else {
3622 			clen = 0;
3623 			cm = NULL;
3624 		}
3625 	}
3626 
3627 	return (error);
3628 }
3629 
3630 static void
3631 unp_zone_change(void *tag)
3632 {
3633 
3634 	uma_zone_set_max(unp_zone, maxsockets);
3635 }
3636 
3637 #ifdef INVARIANTS
3638 static void
3639 unp_zdtor(void *mem, int size __unused, void *arg __unused)
3640 {
3641 	struct unpcb *unp;
3642 
3643 	unp = mem;
3644 
3645 	KASSERT(LIST_EMPTY(&unp->unp_refs),
3646 	    ("%s: unpcb %p has lingering refs", __func__, unp));
3647 	KASSERT(unp->unp_socket == NULL,
3648 	    ("%s: unpcb %p has socket backpointer", __func__, unp));
3649 	KASSERT(unp->unp_vnode == NULL,
3650 	    ("%s: unpcb %p has vnode references", __func__, unp));
3651 	KASSERT(unp->unp_conn == NULL,
3652 	    ("%s: unpcb %p is still connected", __func__, unp));
3653 	KASSERT(unp->unp_addr == NULL,
3654 	    ("%s: unpcb %p has leaked addr", __func__, unp));
3655 }
3656 #endif
3657 
3658 static void
3659 unp_init(void *arg __unused)
3660 {
3661 	uma_dtor dtor;
3662 
3663 #ifdef INVARIANTS
3664 	dtor = unp_zdtor;
3665 #else
3666 	dtor = NULL;
3667 #endif
3668 	unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, dtor,
3669 	    NULL, NULL, UMA_ALIGN_CACHE, 0);
3670 	uma_zone_set_max(unp_zone, maxsockets);
3671 	uma_zone_set_warning(unp_zone, "kern.ipc.maxsockets limit reached");
3672 	EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change,
3673 	    NULL, EVENTHANDLER_PRI_ANY);
3674 	LIST_INIT(&unp_dhead);
3675 	LIST_INIT(&unp_shead);
3676 	LIST_INIT(&unp_sphead);
3677 	SLIST_INIT(&unp_defers);
3678 	TIMEOUT_TASK_INIT(taskqueue_thread, &unp_gc_task, 0, unp_gc, NULL);
3679 	TASK_INIT(&unp_defer_task, 0, unp_process_defers, NULL);
3680 	UNP_LINK_LOCK_INIT();
3681 	UNP_DEFERRED_LOCK_INIT();
3682 	unp_vp_mtxpool = mtx_pool_create("unp vp mtxpool", 32, MTX_DEF);
3683 }
3684 SYSINIT(unp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_SECOND, unp_init, NULL);
3685 
3686 static void
3687 unp_internalize_cleanup_rights(struct mbuf *control)
3688 {
3689 	struct cmsghdr *cp;
3690 	struct mbuf *m;
3691 	void *data;
3692 	socklen_t datalen;
3693 
3694 	for (m = control; m != NULL; m = m->m_next) {
3695 		cp = mtod(m, struct cmsghdr *);
3696 		if (cp->cmsg_level != SOL_SOCKET ||
3697 		    cp->cmsg_type != SCM_RIGHTS)
3698 			continue;
3699 		data = CMSG_DATA(cp);
3700 		datalen = (caddr_t)cp + cp->cmsg_len - (caddr_t)data;
3701 		unp_freerights(data, datalen / sizeof(struct filedesc *));
3702 	}
3703 }
3704 
3705 static int
3706 unp_internalize(struct mbuf *control, struct mchain *mc, struct thread *td,
3707     int *needsopts)
3708 {
3709 	struct proc *p;
3710 	struct filedesc *fdesc;
3711 	struct bintime *bt;
3712 	struct cmsghdr *cm;
3713 	struct cmsgcred *cmcred;
3714 	struct mbuf *m;
3715 	struct filedescent *fde, **fdep, *fdev;
3716 	struct file *fp;
3717 	struct timeval *tv;
3718 	struct timespec *ts;
3719 	void *data;
3720 	socklen_t clen, datalen;
3721 	int i, j, error, *fdp, oldfds;
3722 	u_int newlen;
3723 
3724 	MPASS(control->m_next == NULL); /* COMPAT_OLDSOCK may violate */
3725 	UNP_LINK_UNLOCK_ASSERT();
3726 
3727 	p = td->td_proc;
3728 	fdesc = p->p_fd;
3729 	error = 0;
3730 	*mc = MCHAIN_INITIALIZER(mc);
3731 	for (clen = control->m_len, cm = mtod(control, struct cmsghdr *),
3732 	    data = CMSG_DATA(cm);
3733 
3734 	    clen >= sizeof(*cm) && cm->cmsg_level == SOL_SOCKET &&
3735 	    clen >= cm->cmsg_len && cm->cmsg_len >= sizeof(*cm) &&
3736 	    (char *)cm + cm->cmsg_len >= (char *)data;
3737 
3738 	    clen -= min(CMSG_SPACE(datalen), clen),
3739 	    cm = (struct cmsghdr *) ((char *)cm + CMSG_SPACE(datalen)),
3740 	    data = CMSG_DATA(cm)) {
3741 		datalen = (char *)cm + cm->cmsg_len - (char *)data;
3742 		switch (cm->cmsg_type) {
3743 		case SCM_CREDS:
3744 			m = sbcreatecontrol(NULL, sizeof(*cmcred), SCM_CREDS,
3745 			    SOL_SOCKET, M_WAITOK);
3746 			cmcred = (struct cmsgcred *)
3747 			    CMSG_DATA(mtod(m, struct cmsghdr *));
3748 			cmcred->cmcred_pid = p->p_pid;
3749 			cmcred->cmcred_uid = td->td_ucred->cr_ruid;
3750 			cmcred->cmcred_gid = td->td_ucred->cr_rgid;
3751 			cmcred->cmcred_euid = td->td_ucred->cr_uid;
3752 			_Static_assert(CMGROUP_MAX >= 1,
3753 			    "Room needed for the effective GID.");
3754 			cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups + 1,
3755 			    CMGROUP_MAX);
3756 			cmcred->cmcred_groups[0] = td->td_ucred->cr_gid;
3757 			for (i = 1; i < cmcred->cmcred_ngroups; i++)
3758 				cmcred->cmcred_groups[i] =
3759 				    td->td_ucred->cr_groups[i - 1];
3760 			break;
3761 
3762 		case SCM_RIGHTS:
3763 			*needsopts |= SO_PASSRIGHTS;
3764 			oldfds = datalen / sizeof (int);
3765 			if (oldfds == 0)
3766 				continue;
3767 			/* On some machines sizeof pointer is bigger than
3768 			 * sizeof int, so we need to check if data fits into
3769 			 * single mbuf.  We could allocate several mbufs, and
3770 			 * unp_externalize() should even properly handle that.
3771 			 * But it is not worth to complicate the code for an
3772 			 * insane scenario of passing over 200 file descriptors
3773 			 * at once.
3774 			 */
3775 			newlen = oldfds * sizeof(fdep[0]);
3776 			if (CMSG_SPACE(newlen) > MCLBYTES) {
3777 				error = EMSGSIZE;
3778 				goto out;
3779 			}
3780 			/*
3781 			 * Check that all the FDs passed in refer to legal
3782 			 * files.  If not, reject the entire operation.
3783 			 */
3784 			fdp = data;
3785 			FILEDESC_SLOCK(fdesc);
3786 			for (i = 0; i < oldfds; i++, fdp++) {
3787 				fp = fget_noref(fdesc, *fdp);
3788 				if (fp == NULL) {
3789 					FILEDESC_SUNLOCK(fdesc);
3790 					error = EBADF;
3791 					goto out;
3792 				}
3793 				if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) {
3794 					FILEDESC_SUNLOCK(fdesc);
3795 					error = EOPNOTSUPP;
3796 					goto out;
3797 				}
3798 			}
3799 
3800 			/*
3801 			 * Now replace the integer FDs with pointers to the
3802 			 * file structure and capability rights.
3803 			 */
3804 			m = sbcreatecontrol(NULL, newlen, SCM_RIGHTS,
3805 			    SOL_SOCKET, M_WAITOK);
3806 			fdp = data;
3807 			for (i = 0; i < oldfds; i++, fdp++) {
3808 				if (!fhold(fdesc->fd_ofiles[*fdp].fde_file)) {
3809 					fdp = data;
3810 					for (j = 0; j < i; j++, fdp++) {
3811 						fdrop(fdesc->fd_ofiles[*fdp].
3812 						    fde_file, td);
3813 					}
3814 					FILEDESC_SUNLOCK(fdesc);
3815 					error = EBADF;
3816 					goto out;
3817 				}
3818 			}
3819 			fdp = data;
3820 			fdep = (struct filedescent **)
3821 			    CMSG_DATA(mtod(m, struct cmsghdr *));
3822 			fdev = malloc(sizeof(*fdev) * oldfds, M_FILECAPS,
3823 			    M_WAITOK);
3824 			for (i = 0; i < oldfds; i++, fdev++, fdp++) {
3825 				fde = &fdesc->fd_ofiles[*fdp];
3826 				fdep[i] = fdev;
3827 				fdep[i]->fde_file = fde->fde_file;
3828 				filecaps_copy(&fde->fde_caps,
3829 				    &fdep[i]->fde_caps, true);
3830 				unp_internalize_fp(fdep[i]->fde_file);
3831 			}
3832 			FILEDESC_SUNLOCK(fdesc);
3833 			break;
3834 
3835 		case SCM_TIMESTAMP:
3836 			m = sbcreatecontrol(NULL, sizeof(*tv), SCM_TIMESTAMP,
3837 			    SOL_SOCKET, M_WAITOK);
3838 			tv = (struct timeval *)
3839 			    CMSG_DATA(mtod(m, struct cmsghdr *));
3840 			microtime(tv);
3841 			break;
3842 
3843 		case SCM_BINTIME:
3844 			m = sbcreatecontrol(NULL, sizeof(*bt), SCM_BINTIME,
3845 			    SOL_SOCKET, M_WAITOK);
3846 			bt = (struct bintime *)
3847 			    CMSG_DATA(mtod(m, struct cmsghdr *));
3848 			bintime(bt);
3849 			break;
3850 
3851 		case SCM_REALTIME:
3852 			m = sbcreatecontrol(NULL, sizeof(*ts), SCM_REALTIME,
3853 			    SOL_SOCKET, M_WAITOK);
3854 			ts = (struct timespec *)
3855 			    CMSG_DATA(mtod(m, struct cmsghdr *));
3856 			nanotime(ts);
3857 			break;
3858 
3859 		case SCM_MONOTONIC:
3860 			m = sbcreatecontrol(NULL, sizeof(*ts), SCM_MONOTONIC,
3861 			    SOL_SOCKET, M_WAITOK);
3862 			ts = (struct timespec *)
3863 			    CMSG_DATA(mtod(m, struct cmsghdr *));
3864 			nanouptime(ts);
3865 			break;
3866 
3867 		default:
3868 			error = EINVAL;
3869 			goto out;
3870 		}
3871 
3872 		mc_append(mc, m);
3873 	}
3874 	if (clen > 0)
3875 		error = EINVAL;
3876 
3877 out:
3878 	if (error != 0)
3879 		unp_internalize_cleanup_rights(mc_first(mc));
3880 	m_freem(control);
3881 	return (error);
3882 }
3883 
3884 static void
3885 unp_addsockcred(struct thread *td, struct mchain *mc, int mode)
3886 {
3887 	struct mbuf *m, *n, *n_prev;
3888 	const struct cmsghdr *cm;
3889 	int ngroups, i, cmsgtype;
3890 	size_t ctrlsz;
3891 
3892 	ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX);
3893 	if (mode & UNP_WANTCRED_ALWAYS) {
3894 		ctrlsz = SOCKCRED2SIZE(ngroups);
3895 		cmsgtype = SCM_CREDS2;
3896 	} else {
3897 		ctrlsz = SOCKCREDSIZE(ngroups);
3898 		cmsgtype = SCM_CREDS;
3899 	}
3900 
3901 	/* XXXGL: uipc_sosend_*() need to be improved so that we can M_WAITOK */
3902 	m = sbcreatecontrol(NULL, ctrlsz, cmsgtype, SOL_SOCKET, M_NOWAIT);
3903 	if (m == NULL)
3904 		return;
3905 	MPASS((m->m_flags & M_EXT) == 0 && m->m_next == NULL);
3906 
3907 	if (mode & UNP_WANTCRED_ALWAYS) {
3908 		struct sockcred2 *sc;
3909 
3910 		sc = (void *)CMSG_DATA(mtod(m, struct cmsghdr *));
3911 		sc->sc_version = 0;
3912 		sc->sc_pid = td->td_proc->p_pid;
3913 		sc->sc_uid = td->td_ucred->cr_ruid;
3914 		sc->sc_euid = td->td_ucred->cr_uid;
3915 		sc->sc_gid = td->td_ucred->cr_rgid;
3916 		sc->sc_egid = td->td_ucred->cr_gid;
3917 		sc->sc_ngroups = ngroups;
3918 		for (i = 0; i < sc->sc_ngroups; i++)
3919 			sc->sc_groups[i] = td->td_ucred->cr_groups[i];
3920 	} else {
3921 		struct sockcred *sc;
3922 
3923 		sc = (void *)CMSG_DATA(mtod(m, struct cmsghdr *));
3924 		sc->sc_uid = td->td_ucred->cr_ruid;
3925 		sc->sc_euid = td->td_ucred->cr_uid;
3926 		sc->sc_gid = td->td_ucred->cr_rgid;
3927 		sc->sc_egid = td->td_ucred->cr_gid;
3928 		sc->sc_ngroups = ngroups;
3929 		for (i = 0; i < sc->sc_ngroups; i++)
3930 			sc->sc_groups[i] = td->td_ucred->cr_groups[i];
3931 	}
3932 
3933 	/*
3934 	 * Unlink SCM_CREDS control messages (struct cmsgcred), since just
3935 	 * created SCM_CREDS control message (struct sockcred) has another
3936 	 * format.
3937 	 */
3938 	if (!STAILQ_EMPTY(&mc->mc_q) && cmsgtype == SCM_CREDS)
3939 		STAILQ_FOREACH_SAFE(n, &mc->mc_q, m_stailq, n_prev) {
3940 			cm = mtod(n, struct cmsghdr *);
3941     			if (cm->cmsg_level == SOL_SOCKET &&
3942 			    cm->cmsg_type == SCM_CREDS) {
3943 				mc_remove(mc, n);
3944 				m_free(n);
3945 			}
3946 		}
3947 
3948 	/* Prepend it to the head. */
3949 	mc_prepend(mc, m);
3950 }
3951 
3952 static struct unpcb *
3953 fptounp(struct file *fp)
3954 {
3955 	struct socket *so;
3956 
3957 	if (fp->f_type != DTYPE_SOCKET)
3958 		return (NULL);
3959 	if ((so = fp->f_data) == NULL)
3960 		return (NULL);
3961 	if (so->so_proto->pr_domain != &localdomain)
3962 		return (NULL);
3963 	return sotounpcb(so);
3964 }
3965 
3966 static void
3967 unp_discard(struct file *fp)
3968 {
3969 	struct unp_defer *dr;
3970 
3971 	if (unp_externalize_fp(fp)) {
3972 		dr = malloc(sizeof(*dr), M_TEMP, M_WAITOK);
3973 		dr->ud_fp = fp;
3974 		UNP_DEFERRED_LOCK();
3975 		SLIST_INSERT_HEAD(&unp_defers, dr, ud_link);
3976 		UNP_DEFERRED_UNLOCK();
3977 		atomic_add_int(&unp_defers_count, 1);
3978 		taskqueue_enqueue(taskqueue_thread, &unp_defer_task);
3979 	} else
3980 		closef_nothread(fp);
3981 }
3982 
3983 static void
3984 unp_process_defers(void *arg __unused, int pending)
3985 {
3986 	struct unp_defer *dr;
3987 	SLIST_HEAD(, unp_defer) drl;
3988 	int count;
3989 
3990 	SLIST_INIT(&drl);
3991 	for (;;) {
3992 		UNP_DEFERRED_LOCK();
3993 		if (SLIST_FIRST(&unp_defers) == NULL) {
3994 			UNP_DEFERRED_UNLOCK();
3995 			break;
3996 		}
3997 		SLIST_SWAP(&unp_defers, &drl, unp_defer);
3998 		UNP_DEFERRED_UNLOCK();
3999 		count = 0;
4000 		while ((dr = SLIST_FIRST(&drl)) != NULL) {
4001 			SLIST_REMOVE_HEAD(&drl, ud_link);
4002 			closef_nothread(dr->ud_fp);
4003 			free(dr, M_TEMP);
4004 			count++;
4005 		}
4006 		atomic_add_int(&unp_defers_count, -count);
4007 	}
4008 }
4009 
4010 static void
4011 unp_internalize_fp(struct file *fp)
4012 {
4013 	struct unpcb *unp;
4014 
4015 	UNP_LINK_WLOCK();
4016 	if ((unp = fptounp(fp)) != NULL) {
4017 		unp->unp_file = fp;
4018 		unp->unp_msgcount++;
4019 	}
4020 	unp_rights++;
4021 	UNP_LINK_WUNLOCK();
4022 }
4023 
4024 static int
4025 unp_externalize_fp(struct file *fp)
4026 {
4027 	struct unpcb *unp;
4028 	int ret;
4029 
4030 	UNP_LINK_WLOCK();
4031 	if ((unp = fptounp(fp)) != NULL) {
4032 		unp->unp_msgcount--;
4033 		ret = 1;
4034 	} else
4035 		ret = 0;
4036 	unp_rights--;
4037 	UNP_LINK_WUNLOCK();
4038 	return (ret);
4039 }
4040 
4041 /*
4042  * unp_defer indicates whether additional work has been defered for a future
4043  * pass through unp_gc().  It is thread local and does not require explicit
4044  * synchronization.
4045  */
4046 static int	unp_marked;
4047 
4048 static void
4049 unp_remove_dead_ref(struct filedescent **fdep, int fdcount)
4050 {
4051 	struct unpcb *unp;
4052 	struct file *fp;
4053 	int i;
4054 
4055 	/*
4056 	 * This function can only be called from the gc task.
4057 	 */
4058 	KASSERT(taskqueue_member(taskqueue_thread, curthread) != 0,
4059 	    ("%s: not on gc callout", __func__));
4060 	UNP_LINK_LOCK_ASSERT();
4061 
4062 	for (i = 0; i < fdcount; i++) {
4063 		fp = fdep[i]->fde_file;
4064 		if ((unp = fptounp(fp)) == NULL)
4065 			continue;
4066 		if ((unp->unp_gcflag & UNPGC_DEAD) == 0)
4067 			continue;
4068 		unp->unp_gcrefs--;
4069 	}
4070 }
4071 
4072 static void
4073 unp_restore_undead_ref(struct filedescent **fdep, int fdcount)
4074 {
4075 	struct unpcb *unp;
4076 	struct file *fp;
4077 	int i;
4078 
4079 	/*
4080 	 * This function can only be called from the gc task.
4081 	 */
4082 	KASSERT(taskqueue_member(taskqueue_thread, curthread) != 0,
4083 	    ("%s: not on gc callout", __func__));
4084 	UNP_LINK_LOCK_ASSERT();
4085 
4086 	for (i = 0; i < fdcount; i++) {
4087 		fp = fdep[i]->fde_file;
4088 		if ((unp = fptounp(fp)) == NULL)
4089 			continue;
4090 		if ((unp->unp_gcflag & UNPGC_DEAD) == 0)
4091 			continue;
4092 		unp->unp_gcrefs++;
4093 		unp_marked++;
4094 	}
4095 }
4096 
4097 static void
4098 unp_scan_socket(struct socket *so, void (*op)(struct filedescent **, int))
4099 {
4100 	struct sockbuf *sb;
4101 
4102 	SOCK_LOCK_ASSERT(so);
4103 
4104 	if (sotounpcb(so)->unp_gcflag & UNPGC_IGNORE_RIGHTS)
4105 		return;
4106 
4107 	SOCK_RECVBUF_LOCK(so);
4108 	switch (so->so_type) {
4109 	case SOCK_DGRAM:
4110 		unp_scan(STAILQ_FIRST(&so->so_rcv.uxdg_mb), op);
4111 		unp_scan(so->so_rcv.uxdg_peeked, op);
4112 		TAILQ_FOREACH(sb, &so->so_rcv.uxdg_conns, uxdg_clist)
4113 			unp_scan(STAILQ_FIRST(&sb->uxdg_mb), op);
4114 		break;
4115 	case SOCK_STREAM:
4116 	case SOCK_SEQPACKET:
4117 		unp_scan(STAILQ_FIRST(&so->so_rcv.uxst_mbq), op);
4118 		break;
4119 	}
4120 	SOCK_RECVBUF_UNLOCK(so);
4121 }
4122 
4123 static void
4124 unp_gc_scan(struct unpcb *unp, void (*op)(struct filedescent **, int))
4125 {
4126 	struct socket *so, *soa;
4127 
4128 	so = unp->unp_socket;
4129 	SOCK_LOCK(so);
4130 	if (SOLISTENING(so)) {
4131 		/*
4132 		 * Mark all sockets in our accept queue.
4133 		 */
4134 		TAILQ_FOREACH(soa, &so->sol_comp, so_list)
4135 			unp_scan_socket(soa, op);
4136 	} else {
4137 		/*
4138 		 * Mark all sockets we reference with RIGHTS.
4139 		 */
4140 		unp_scan_socket(so, op);
4141 	}
4142 	SOCK_UNLOCK(so);
4143 }
4144 
4145 static int unp_recycled;
4146 SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0,
4147     "Number of unreachable sockets claimed by the garbage collector.");
4148 
4149 static int unp_taskcount;
4150 SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0,
4151     "Number of times the garbage collector has run.");
4152 
4153 SYSCTL_UINT(_net_local, OID_AUTO, sockcount, CTLFLAG_RD, &unp_count, 0,
4154     "Number of active local sockets.");
4155 
4156 static void
4157 unp_gc(__unused void *arg, int pending)
4158 {
4159 	struct unp_head *heads[] = { &unp_dhead, &unp_shead, &unp_sphead,
4160 				    NULL };
4161 	struct unp_head **head;
4162 	struct unp_head unp_deadhead;	/* List of potentially-dead sockets. */
4163 	struct file *f, **unref;
4164 	struct unpcb *unp, *unptmp;
4165 	int i, total, unp_unreachable;
4166 
4167 	LIST_INIT(&unp_deadhead);
4168 	unp_taskcount++;
4169 	UNP_LINK_RLOCK();
4170 	/*
4171 	 * First determine which sockets may be in cycles.
4172 	 */
4173 	unp_unreachable = 0;
4174 
4175 	for (head = heads; *head != NULL; head++)
4176 		LIST_FOREACH(unp, *head, unp_link) {
4177 			KASSERT((unp->unp_gcflag & ~UNPGC_IGNORE_RIGHTS) == 0,
4178 			    ("%s: unp %p has unexpected gc flags 0x%x",
4179 			    __func__, unp, (unsigned int)unp->unp_gcflag));
4180 
4181 			f = unp->unp_file;
4182 
4183 			/*
4184 			 * Check for an unreachable socket potentially in a
4185 			 * cycle.  It must be in a queue as indicated by
4186 			 * msgcount, and this must equal the file reference
4187 			 * count.  Note that when msgcount is 0 the file is
4188 			 * NULL.
4189 			 */
4190 			if (f != NULL && unp->unp_msgcount != 0 &&
4191 			    refcount_load(&f->f_count) == unp->unp_msgcount) {
4192 				LIST_INSERT_HEAD(&unp_deadhead, unp, unp_dead);
4193 				unp->unp_gcflag |= UNPGC_DEAD;
4194 				unp->unp_gcrefs = unp->unp_msgcount;
4195 				unp_unreachable++;
4196 			}
4197 		}
4198 
4199 	/*
4200 	 * Scan all sockets previously marked as potentially being in a cycle
4201 	 * and remove the references each socket holds on any UNPGC_DEAD
4202 	 * sockets in its queue.  After this step, all remaining references on
4203 	 * sockets marked UNPGC_DEAD should not be part of any cycle.
4204 	 */
4205 	LIST_FOREACH(unp, &unp_deadhead, unp_dead)
4206 		unp_gc_scan(unp, unp_remove_dead_ref);
4207 
4208 	/*
4209 	 * If a socket still has a non-negative refcount, it cannot be in a
4210 	 * cycle.  In this case increment refcount of all children iteratively.
4211 	 * Stop the scan once we do a complete loop without discovering
4212 	 * a new reachable socket.
4213 	 */
4214 	do {
4215 		unp_marked = 0;
4216 		LIST_FOREACH_SAFE(unp, &unp_deadhead, unp_dead, unptmp)
4217 			if (unp->unp_gcrefs > 0) {
4218 				unp->unp_gcflag &= ~UNPGC_DEAD;
4219 				LIST_REMOVE(unp, unp_dead);
4220 				KASSERT(unp_unreachable > 0,
4221 				    ("%s: unp_unreachable underflow.",
4222 				    __func__));
4223 				unp_unreachable--;
4224 				unp_gc_scan(unp, unp_restore_undead_ref);
4225 			}
4226 	} while (unp_marked);
4227 
4228 	UNP_LINK_RUNLOCK();
4229 
4230 	if (unp_unreachable == 0)
4231 		return;
4232 
4233 	/*
4234 	 * Allocate space for a local array of dead unpcbs.
4235 	 * TODO: can this path be simplified by instead using the local
4236 	 * dead list at unp_deadhead, after taking out references
4237 	 * on the file object and/or unpcb and dropping the link lock?
4238 	 */
4239 	unref = malloc(unp_unreachable * sizeof(struct file *),
4240 	    M_TEMP, M_WAITOK);
4241 
4242 	/*
4243 	 * Iterate looking for sockets which have been specifically marked
4244 	 * as unreachable and store them locally.
4245 	 */
4246 	UNP_LINK_RLOCK();
4247 	total = 0;
4248 	LIST_FOREACH(unp, &unp_deadhead, unp_dead) {
4249 		KASSERT((unp->unp_gcflag & UNPGC_DEAD) != 0,
4250 		    ("%s: unp %p not marked UNPGC_DEAD", __func__, unp));
4251 		unp->unp_gcflag &= ~UNPGC_DEAD;
4252 		f = unp->unp_file;
4253 		if (unp->unp_msgcount == 0 || f == NULL ||
4254 		    refcount_load(&f->f_count) != unp->unp_msgcount ||
4255 		    !fhold(f))
4256 			continue;
4257 		unref[total++] = f;
4258 		KASSERT(total <= unp_unreachable,
4259 		    ("%s: incorrect unreachable count.", __func__));
4260 	}
4261 	UNP_LINK_RUNLOCK();
4262 
4263 	/*
4264 	 * Now flush all sockets, free'ing rights.  This will free the
4265 	 * struct files associated with these sockets but leave each socket
4266 	 * with one remaining ref.
4267 	 */
4268 	for (i = 0; i < total; i++) {
4269 		struct socket *so;
4270 
4271 		so = unref[i]->f_data;
4272 		if (!SOLISTENING(so)) {
4273 			CURVNET_SET(so->so_vnet);
4274 			socantrcvmore(so);
4275 			unp_dispose(so);
4276 			CURVNET_RESTORE();
4277 		}
4278 	}
4279 
4280 	/*
4281 	 * And finally release the sockets so they can be reclaimed.
4282 	 */
4283 	for (i = 0; i < total; i++)
4284 		fdrop(unref[i], NULL);
4285 	unp_recycled += total;
4286 	free(unref, M_TEMP);
4287 }
4288 
4289 /*
4290  * Synchronize against unp_gc, which can trip over data as we are freeing it.
4291  */
4292 static void
4293 unp_dispose(struct socket *so)
4294 {
4295 	struct sockbuf *sb;
4296 	struct unpcb *unp;
4297 	struct mbuf *m;
4298 	int error __diagused;
4299 
4300 	MPASS(!SOLISTENING(so));
4301 
4302 	unp = sotounpcb(so);
4303 	UNP_LINK_WLOCK();
4304 	unp->unp_gcflag |= UNPGC_IGNORE_RIGHTS;
4305 	UNP_LINK_WUNLOCK();
4306 
4307 	/*
4308 	 * Grab our special mbufs before calling sbrelease().
4309 	 */
4310 	error = SOCK_IO_RECV_LOCK(so, SBL_WAIT | SBL_NOINTR);
4311 	MPASS(!error);
4312 	SOCK_RECVBUF_LOCK(so);
4313 	switch (so->so_type) {
4314 	case SOCK_DGRAM:
4315 		while ((sb = TAILQ_FIRST(&so->so_rcv.uxdg_conns)) != NULL) {
4316 			STAILQ_CONCAT(&so->so_rcv.uxdg_mb, &sb->uxdg_mb);
4317 			TAILQ_REMOVE(&so->so_rcv.uxdg_conns, sb, uxdg_clist);
4318 			/* Note: socket of sb may reconnect. */
4319 			sb->uxdg_cc = sb->uxdg_ctl = sb->uxdg_mbcnt = 0;
4320 		}
4321 		sb = &so->so_rcv;
4322 		if (sb->uxdg_peeked != NULL) {
4323 			STAILQ_INSERT_HEAD(&sb->uxdg_mb, sb->uxdg_peeked,
4324 			    m_stailqpkt);
4325 			sb->uxdg_peeked = NULL;
4326 		}
4327 		m = STAILQ_FIRST(&sb->uxdg_mb);
4328 		STAILQ_INIT(&sb->uxdg_mb);
4329 		break;
4330 	case SOCK_STREAM:
4331 	case SOCK_SEQPACKET:
4332 		sb = &so->so_rcv;
4333 		m = STAILQ_FIRST(&sb->uxst_mbq);
4334 		STAILQ_INIT(&sb->uxst_mbq);
4335 		sb->sb_acc = sb->sb_ccc = sb->sb_ctl = sb->sb_mbcnt = 0;
4336 		/*
4337 		 * Trim M_NOTREADY buffers from the free list.  They are
4338 		 * referenced by the I/O thread.
4339 		 */
4340 		if (sb->uxst_fnrdy != NULL) {
4341 			struct mbuf *n, *prev;
4342 
4343 			while (m != NULL && m->m_flags & M_NOTREADY)
4344 				m = m->m_next;
4345 			for (prev = n = m; n != NULL; n = n->m_next) {
4346 				if (n->m_flags & M_NOTREADY)
4347 					prev->m_next = n->m_next;
4348 				else
4349 					prev = n;
4350 			}
4351 			sb->uxst_fnrdy = NULL;
4352 		}
4353 		break;
4354 	}
4355 	/*
4356 	 * Mark sb with SBS_CANTRCVMORE.  This is needed to prevent
4357 	 * uipc_sosend_*() or unp_disconnect() adding more data to the socket.
4358 	 * We came here either through shutdown(2) or from the final sofree().
4359 	 * The sofree() case is simple as it guarantees that no more sends will
4360 	 * happen, however we can race with unp_disconnect() from our peer.
4361 	 * The shutdown(2) case is more exotic.  It would call into
4362 	 * unp_dispose() only if socket is SS_ISCONNECTED.  This is possible if
4363 	 * we did connect(2) on this socket and we also had it bound with
4364 	 * bind(2) and receive connections from other sockets.  Because
4365 	 * uipc_shutdown() violates POSIX (see comment there) this applies to
4366 	 * SOCK_DGRAM as well.  For SOCK_DGRAM this SBS_CANTRCVMORE will have
4367 	 * affect not only on the peer we connect(2)ed to, but also on all of
4368 	 * the peers who had connect(2)ed to us.  Their sends would end up
4369 	 * with ENOBUFS.
4370 	 */
4371 	sb->sb_state |= SBS_CANTRCVMORE;
4372 	(void)chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, 0,
4373 	    RLIM_INFINITY);
4374 	SOCK_RECVBUF_UNLOCK(so);
4375 	SOCK_IO_RECV_UNLOCK(so);
4376 
4377 	if (m != NULL) {
4378 		unp_scan(m, unp_freerights);
4379 		m_freemp(m);
4380 	}
4381 }
4382 
4383 static void
4384 unp_scan(struct mbuf *m0, void (*op)(struct filedescent **, int))
4385 {
4386 	struct mbuf *m;
4387 	struct cmsghdr *cm;
4388 	void *data;
4389 	socklen_t clen, datalen;
4390 
4391 	while (m0 != NULL) {
4392 		for (m = m0; m; m = m->m_next) {
4393 			if (m->m_type != MT_CONTROL)
4394 				continue;
4395 
4396 			cm = mtod(m, struct cmsghdr *);
4397 			clen = m->m_len;
4398 
4399 			while (cm != NULL) {
4400 				if (sizeof(*cm) > clen || cm->cmsg_len > clen)
4401 					break;
4402 
4403 				data = CMSG_DATA(cm);
4404 				datalen = (caddr_t)cm + cm->cmsg_len
4405 				    - (caddr_t)data;
4406 
4407 				if (cm->cmsg_level == SOL_SOCKET &&
4408 				    cm->cmsg_type == SCM_RIGHTS) {
4409 					(*op)(data, datalen /
4410 					    sizeof(struct filedescent *));
4411 				}
4412 
4413 				if (CMSG_SPACE(datalen) < clen) {
4414 					clen -= CMSG_SPACE(datalen);
4415 					cm = (struct cmsghdr *)
4416 					    ((caddr_t)cm + CMSG_SPACE(datalen));
4417 				} else {
4418 					clen = 0;
4419 					cm = NULL;
4420 				}
4421 			}
4422 		}
4423 		m0 = m0->m_nextpkt;
4424 	}
4425 }
4426 
4427 /*
4428  * Definitions of protocols supported in the LOCAL domain.
4429  */
4430 static struct protosw streamproto = {
4431 	.pr_type =		SOCK_STREAM,
4432 	.pr_flags =		PR_CONNREQUIRED | PR_CAPATTACH | PR_SOCKBUF,
4433 	.pr_ctloutput =		&uipc_ctloutput,
4434 	.pr_abort = 		uipc_abort,
4435 	.pr_accept =		uipc_peeraddr,
4436 	.pr_attach =		uipc_attach,
4437 	.pr_bind =		uipc_bind,
4438 	.pr_bindat =		uipc_bindat,
4439 	.pr_connect =		uipc_connect,
4440 	.pr_connectat =		uipc_connectat,
4441 	.pr_connect2 =		uipc_connect2,
4442 	.pr_detach =		uipc_detach,
4443 	.pr_disconnect =	uipc_disconnect,
4444 	.pr_fdclose =		uipc_fdclose,
4445 	.pr_listen =		uipc_listen,
4446 	.pr_peeraddr =		uipc_peeraddr,
4447 	.pr_send =		uipc_sendfile,
4448 	.pr_sendfile_wait =	uipc_sendfile_wait,
4449 	.pr_ready =		uipc_ready,
4450 	.pr_sense =		uipc_sense,
4451 	.pr_shutdown =		uipc_shutdown,
4452 	.pr_sockaddr =		uipc_sockaddr,
4453 	.pr_sosend = 		uipc_sosend_stream_or_seqpacket,
4454 	.pr_soreceive =		uipc_soreceive_stream_or_seqpacket,
4455 	.pr_sopoll =		uipc_sopoll_stream_or_seqpacket,
4456 	.pr_kqfilter =		uipc_kqfilter_stream_or_seqpacket,
4457 	.pr_close =		uipc_close,
4458 	.pr_chmod =		uipc_chmod,
4459 };
4460 
4461 static struct protosw dgramproto = {
4462 	.pr_type =		SOCK_DGRAM,
4463 	.pr_flags =		PR_ATOMIC | PR_ADDR | PR_CAPATTACH | PR_SOCKBUF,
4464 	.pr_ctloutput =		&uipc_ctloutput,
4465 	.pr_abort = 		uipc_abort,
4466 	.pr_accept =		uipc_peeraddr,
4467 	.pr_attach =		uipc_attach,
4468 	.pr_bind =		uipc_bind,
4469 	.pr_bindat =		uipc_bindat,
4470 	.pr_connect =		uipc_connect,
4471 	.pr_connectat =		uipc_connectat,
4472 	.pr_connect2 =		uipc_connect2,
4473 	.pr_detach =		uipc_detach,
4474 	.pr_disconnect =	uipc_disconnect,
4475 	.pr_fdclose =		uipc_fdclose,
4476 	.pr_peeraddr =		uipc_peeraddr,
4477 	.pr_sosend =		uipc_sosend_dgram,
4478 	.pr_sense =		uipc_sense,
4479 	.pr_shutdown =		uipc_shutdown,
4480 	.pr_sockaddr =		uipc_sockaddr,
4481 	.pr_soreceive =		uipc_soreceive_dgram,
4482 	.pr_close =		uipc_close,
4483 	.pr_chmod =		uipc_chmod,
4484 };
4485 
4486 static struct protosw seqpacketproto = {
4487 	.pr_type =		SOCK_SEQPACKET,
4488 	.pr_flags =		PR_CONNREQUIRED | PR_CAPATTACH | PR_SOCKBUF,
4489 	.pr_ctloutput =		&uipc_ctloutput,
4490 	.pr_abort =		uipc_abort,
4491 	.pr_accept =		uipc_peeraddr,
4492 	.pr_attach =		uipc_attach,
4493 	.pr_bind =		uipc_bind,
4494 	.pr_bindat =		uipc_bindat,
4495 	.pr_connect =		uipc_connect,
4496 	.pr_connectat =		uipc_connectat,
4497 	.pr_connect2 =		uipc_connect2,
4498 	.pr_detach =		uipc_detach,
4499 	.pr_disconnect =	uipc_disconnect,
4500 	.pr_fdclose =		uipc_fdclose,
4501 	.pr_listen =		uipc_listen,
4502 	.pr_peeraddr =		uipc_peeraddr,
4503 	.pr_sense =		uipc_sense,
4504 	.pr_shutdown =		uipc_shutdown,
4505 	.pr_sockaddr =		uipc_sockaddr,
4506 	.pr_sosend = 		uipc_sosend_stream_or_seqpacket,
4507 	.pr_soreceive =		uipc_soreceive_stream_or_seqpacket,
4508 	.pr_sopoll =		uipc_sopoll_stream_or_seqpacket,
4509 	.pr_kqfilter =		uipc_kqfilter_stream_or_seqpacket,
4510 	.pr_close =		uipc_close,
4511 	.pr_chmod =		uipc_chmod,
4512 };
4513 
4514 static struct domain localdomain = {
4515 	.dom_family =		AF_LOCAL,
4516 	.dom_name =		"local",
4517 	.dom_nprotosw =		3,
4518 	.dom_protosw =		{
4519 		&streamproto,
4520 		&dgramproto,
4521 		&seqpacketproto,
4522 	}
4523 };
4524 DOMAIN_SET(local);
4525 
4526 /*
4527  * A helper function called by VFS before socket-type vnode reclamation.
4528  * For an active vnode it clears unp_vnode pointer and decrements unp_vnode
4529  * use count.
4530  */
4531 void
4532 vfs_unp_reclaim(struct vnode *vp)
4533 {
4534 	struct unpcb *unp;
4535 	int active;
4536 	struct mtx *vplock;
4537 
4538 	ASSERT_VOP_ELOCKED(vp, "vfs_unp_reclaim");
4539 	KASSERT(vp->v_type == VSOCK,
4540 	    ("vfs_unp_reclaim: vp->v_type != VSOCK"));
4541 
4542 	active = 0;
4543 	vplock = mtx_pool_find(unp_vp_mtxpool, vp);
4544 	mtx_lock(vplock);
4545 	VOP_UNP_CONNECT(vp, &unp);
4546 	if (unp == NULL)
4547 		goto done;
4548 	UNP_PCB_LOCK(unp);
4549 	if (unp->unp_vnode == vp) {
4550 		VOP_UNP_DETACH(vp);
4551 		unp->unp_vnode = NULL;
4552 		active = 1;
4553 	}
4554 	UNP_PCB_UNLOCK(unp);
4555  done:
4556 	mtx_unlock(vplock);
4557 	if (active)
4558 		vunref(vp);
4559 }
4560 
4561 #ifdef DDB
4562 static void
4563 db_print_indent(int indent)
4564 {
4565 	int i;
4566 
4567 	for (i = 0; i < indent; i++)
4568 		db_printf(" ");
4569 }
4570 
4571 static void
4572 db_print_unpflags(int unp_flags)
4573 {
4574 	int comma;
4575 
4576 	comma = 0;
4577 	if (unp_flags & UNP_HAVEPC) {
4578 		db_printf("%sUNP_HAVEPC", comma ? ", " : "");
4579 		comma = 1;
4580 	}
4581 	if (unp_flags & UNP_WANTCRED_ALWAYS) {
4582 		db_printf("%sUNP_WANTCRED_ALWAYS", comma ? ", " : "");
4583 		comma = 1;
4584 	}
4585 	if (unp_flags & UNP_WANTCRED_ONESHOT) {
4586 		db_printf("%sUNP_WANTCRED_ONESHOT", comma ? ", " : "");
4587 		comma = 1;
4588 	}
4589 	if (unp_flags & UNP_CONNECTING) {
4590 		db_printf("%sUNP_CONNECTING", comma ? ", " : "");
4591 		comma = 1;
4592 	}
4593 	if (unp_flags & UNP_BINDING) {
4594 		db_printf("%sUNP_BINDING", comma ? ", " : "");
4595 		comma = 1;
4596 	}
4597 }
4598 
4599 static void
4600 db_print_xucred(int indent, struct xucred *xu)
4601 {
4602 	int comma, i;
4603 
4604 	db_print_indent(indent);
4605 	db_printf("cr_version: %u   cr_uid: %u   cr_pid: %d   cr_ngroups: %d\n",
4606 	    xu->cr_version, xu->cr_uid, xu->cr_pid, xu->cr_ngroups);
4607 	db_print_indent(indent);
4608 	db_printf("cr_groups: ");
4609 	comma = 0;
4610 	for (i = 0; i < xu->cr_ngroups; i++) {
4611 		db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]);
4612 		comma = 1;
4613 	}
4614 	db_printf("\n");
4615 }
4616 
4617 static void
4618 db_print_unprefs(int indent, struct unp_head *uh)
4619 {
4620 	struct unpcb *unp;
4621 	int counter;
4622 
4623 	counter = 0;
4624 	LIST_FOREACH(unp, uh, unp_reflink) {
4625 		if (counter % 4 == 0)
4626 			db_print_indent(indent);
4627 		db_printf("%p  ", unp);
4628 		if (counter % 4 == 3)
4629 			db_printf("\n");
4630 		counter++;
4631 	}
4632 	if (counter != 0 && counter % 4 != 0)
4633 		db_printf("\n");
4634 }
4635 
4636 DB_SHOW_COMMAND(unpcb, db_show_unpcb)
4637 {
4638 	struct unpcb *unp;
4639 
4640         if (!have_addr) {
4641                 db_printf("usage: show unpcb <addr>\n");
4642                 return;
4643         }
4644         unp = (struct unpcb *)addr;
4645 
4646 	db_printf("unp_socket: %p   unp_vnode: %p\n", unp->unp_socket,
4647 	    unp->unp_vnode);
4648 
4649 	db_printf("unp_ino: %ju   unp_conn: %p\n", (uintmax_t)unp->unp_ino,
4650 	    unp->unp_conn);
4651 
4652 	db_printf("unp_refs:\n");
4653 	db_print_unprefs(2, &unp->unp_refs);
4654 
4655 	/* XXXRW: Would be nice to print the full address, if any. */
4656 	db_printf("unp_addr: %p\n", unp->unp_addr);
4657 
4658 	db_printf("unp_gencnt: %llu\n",
4659 	    (unsigned long long)unp->unp_gencnt);
4660 
4661 	db_printf("unp_flags: %x (", unp->unp_flags);
4662 	db_print_unpflags(unp->unp_flags);
4663 	db_printf(")\n");
4664 
4665 	db_printf("unp_peercred:\n");
4666 	db_print_xucred(2, &unp->unp_peercred);
4667 
4668 	db_printf("unp_refcount: %u\n", unp->unp_refcount);
4669 }
4670 #endif
4671