xref: /freebsd/sys/kern/uipc_socket.c (revision fa38579f317d5c2ff2926fab9b12ee6d429bd155)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1988, 1990, 1993
5  *	The Regents of the University of California.
6  * Copyright (c) 2004 The FreeBSD Foundation
7  * Copyright (c) 2004-2008 Robert N. M. Watson
8  * All rights reserved.
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  * Comments on the socket life cycle:
37  *
38  * soalloc() sets of socket layer state for a socket, called only by
39  * socreate() and sonewconn().  Socket layer private.
40  *
41  * sodealloc() tears down socket layer state for a socket, called only by
42  * sofree() and sonewconn().  Socket layer private.
43  *
44  * pru_attach() associates protocol layer state with an allocated socket;
45  * called only once, may fail, aborting socket allocation.  This is called
46  * from socreate() and sonewconn().  Socket layer private.
47  *
48  * pru_detach() disassociates protocol layer state from an attached socket,
49  * and will be called exactly once for sockets in which pru_attach() has
50  * been successfully called.  If pru_attach() returned an error,
51  * pru_detach() will not be called.  Socket layer private.
52  *
53  * pru_abort() and pru_close() notify the protocol layer that the last
54  * consumer of a socket is starting to tear down the socket, and that the
55  * protocol should terminate the connection.  Historically, pru_abort() also
56  * detached protocol state from the socket state, but this is no longer the
57  * case.
58  *
59  * socreate() creates a socket and attaches protocol state.  This is a public
60  * interface that may be used by socket layer consumers to create new
61  * sockets.
62  *
63  * sonewconn() creates a socket and attaches protocol state.  This is a
64  * public interface  that may be used by protocols to create new sockets when
65  * a new connection is received and will be available for accept() on a
66  * listen socket.
67  *
68  * soclose() destroys a socket after possibly waiting for it to disconnect.
69  * This is a public interface that socket consumers should use to close and
70  * release a socket when done with it.
71  *
72  * soabort() destroys a socket without waiting for it to disconnect (used
73  * only for incoming connections that are already partially or fully
74  * connected).  This is used internally by the socket layer when clearing
75  * listen socket queues (due to overflow or close on the listen socket), but
76  * is also a public interface protocols may use to abort connections in
77  * their incomplete listen queues should they no longer be required.  Sockets
78  * placed in completed connection listen queues should not be aborted for
79  * reasons described in the comment above the soclose() implementation.  This
80  * is not a general purpose close routine, and except in the specific
81  * circumstances described here, should not be used.
82  *
83  * sofree() will free a socket and its protocol state if all references on
84  * the socket have been released, and is the public interface to attempt to
85  * free a socket when a reference is removed.  This is a socket layer private
86  * interface.
87  *
88  * NOTE: In addition to socreate() and soclose(), which provide a single
89  * socket reference to the consumer to be managed as required, there are two
90  * calls to explicitly manage socket references, soref(), and sorele().
91  * Currently, these are generally required only when transitioning a socket
92  * from a listen queue to a file descriptor, in order to prevent garbage
93  * collection of the socket at an untimely moment.  For a number of reasons,
94  * these interfaces are not preferred, and should be avoided.
95  *
96  * NOTE: With regard to VNETs the general rule is that callers do not set
97  * curvnet. Exceptions to this rule include soabort(), sodisconnect(),
98  * sofree(), sorele(), sonewconn() and sorflush(), which are usually called
99  * from a pre-set VNET context.  sopoll_generic() currently does not need a
100  * VNET context to be set.
101  */
102 
103 #include <sys/cdefs.h>
104 #include "opt_inet.h"
105 #include "opt_inet6.h"
106 #include "opt_kern_tls.h"
107 #include "opt_ktrace.h"
108 #include "opt_sctp.h"
109 
110 #include <sys/param.h>
111 #include <sys/systm.h>
112 #include <sys/capsicum.h>
113 #include <sys/fcntl.h>
114 #include <sys/limits.h>
115 #include <sys/lock.h>
116 #include <sys/mac.h>
117 #include <sys/malloc.h>
118 #include <sys/mbuf.h>
119 #include <sys/mutex.h>
120 #include <sys/domain.h>
121 #include <sys/file.h>			/* for struct knote */
122 #include <sys/hhook.h>
123 #include <sys/kernel.h>
124 #include <sys/khelp.h>
125 #include <sys/kthread.h>
126 #include <sys/ktls.h>
127 #include <sys/event.h>
128 #include <sys/eventhandler.h>
129 #include <sys/poll.h>
130 #include <sys/proc.h>
131 #include <sys/protosw.h>
132 #include <sys/sbuf.h>
133 #include <sys/socket.h>
134 #include <sys/socketvar.h>
135 #include <sys/resourcevar.h>
136 #include <net/route.h>
137 #include <sys/sched.h>
138 #include <sys/signalvar.h>
139 #include <sys/smp.h>
140 #include <sys/stat.h>
141 #include <sys/sx.h>
142 #include <sys/sysctl.h>
143 #include <sys/taskqueue.h>
144 #include <sys/uio.h>
145 #include <sys/un.h>
146 #include <sys/unpcb.h>
147 #include <sys/jail.h>
148 #include <sys/syslog.h>
149 #include <netinet/in.h>
150 #include <netinet/in_pcb.h>
151 #include <netinet/tcp.h>
152 
153 #include <net/vnet.h>
154 
155 #include <security/mac/mac_framework.h>
156 #include <security/mac/mac_internal.h>
157 
158 #include <vm/uma.h>
159 
160 #ifdef COMPAT_FREEBSD32
161 #include <sys/mount.h>
162 #include <sys/sysent.h>
163 #include <compat/freebsd32/freebsd32.h>
164 #endif
165 
166 static int	soreceive_generic_locked(struct socket *so,
167 		    struct sockaddr **psa, struct uio *uio, struct mbuf **mp,
168 		    struct mbuf **controlp, int *flagsp);
169 static int	soreceive_rcvoob(struct socket *so, struct uio *uio,
170 		    int flags);
171 static int	soreceive_stream_locked(struct socket *so, struct sockbuf *sb,
172 		    struct sockaddr **psa, struct uio *uio, struct mbuf **mp,
173 		    struct mbuf **controlp, int flags);
174 static int	sosend_generic_locked(struct socket *so, struct sockaddr *addr,
175 		    struct uio *uio, struct mbuf *top, struct mbuf *control,
176 		    int flags, struct thread *td);
177 static void	so_rdknl_lock(void *);
178 static void	so_rdknl_unlock(void *);
179 static void	so_rdknl_assert_lock(void *, int);
180 static void	so_wrknl_lock(void *);
181 static void	so_wrknl_unlock(void *);
182 static void	so_wrknl_assert_lock(void *, int);
183 
184 static void	filt_sordetach(struct knote *kn);
185 static int	filt_soread(struct knote *kn, long hint);
186 static void	filt_sowdetach(struct knote *kn);
187 static int	filt_sowrite(struct knote *kn, long hint);
188 static int	filt_soempty(struct knote *kn, long hint);
189 fo_kqfilter_t	soo_kqfilter;
190 
191 static const struct filterops soread_filtops = {
192 	.f_isfd = 1,
193 	.f_detach = filt_sordetach,
194 	.f_event = filt_soread,
195 };
196 static const struct filterops sowrite_filtops = {
197 	.f_isfd = 1,
198 	.f_detach = filt_sowdetach,
199 	.f_event = filt_sowrite,
200 };
201 static const struct filterops soempty_filtops = {
202 	.f_isfd = 1,
203 	.f_detach = filt_sowdetach,
204 	.f_event = filt_soempty,
205 };
206 
207 so_gen_t	so_gencnt;	/* generation count for sockets */
208 
209 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
210 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
211 
212 #define	VNET_SO_ASSERT(so)						\
213 	VNET_ASSERT(curvnet != NULL,					\
214 	    ("%s:%d curvnet is NULL, so=%p", __func__, __LINE__, (so)));
215 
216 #ifdef SOCKET_HHOOK
217 VNET_DEFINE(struct hhook_head *, socket_hhh[HHOOK_SOCKET_LAST + 1]);
218 #define	V_socket_hhh		VNET(socket_hhh)
219 static inline int hhook_run_socket(struct socket *, void *, int32_t);
220 #endif
221 
222 #ifdef COMPAT_FREEBSD32
223 #ifdef __amd64__
224 /* off_t has 4-byte alignment on i386 but not on other 32-bit platforms. */
225 #define	__splice32_packed	__packed
226 #else
227 #define	__splice32_packed
228 #endif
229 struct splice32 {
230 	int32_t	sp_fd;
231 	int64_t sp_max;
232 	struct timeval32 sp_idle;
233 } __splice32_packed;
234 #undef __splice32_packed
235 #endif
236 
237 /*
238  * Limit on the number of connections in the listen queue waiting
239  * for accept(2).
240  * NB: The original sysctl somaxconn is still available but hidden
241  * to prevent confusion about the actual purpose of this number.
242  */
243 VNET_DEFINE_STATIC(u_int, somaxconn) = SOMAXCONN;
244 #define	V_somaxconn	VNET(somaxconn)
245 
246 static int
247 sysctl_somaxconn(SYSCTL_HANDLER_ARGS)
248 {
249 	int error;
250 	u_int val;
251 
252 	val = V_somaxconn;
253 	error = sysctl_handle_int(oidp, &val, 0, req);
254 	if (error || !req->newptr )
255 		return (error);
256 
257 	/*
258 	 * The purpose of the UINT_MAX / 3 limit, is so that the formula
259 	 *   3 * sol_qlimit / 2
260 	 * below, will not overflow.
261          */
262 
263 	if (val < 1 || val > UINT_MAX / 3)
264 		return (EINVAL);
265 
266 	V_somaxconn = val;
267 	return (0);
268 }
269 SYSCTL_PROC(_kern_ipc, OID_AUTO, soacceptqueue,
270     CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_MPSAFE | CTLFLAG_VNET, 0, sizeof(u_int),
271     sysctl_somaxconn, "IU",
272     "Maximum listen socket pending connection accept queue size");
273 SYSCTL_PROC(_kern_ipc, KIPC_SOMAXCONN, somaxconn,
274     CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_SKIP | CTLFLAG_MPSAFE | CTLFLAG_VNET, 0,
275     sizeof(u_int), sysctl_somaxconn, "IU",
276     "Maximum listen socket pending connection accept queue size (compat)");
277 
278 static u_int numopensockets;
279 static int
280 sysctl_numopensockets(SYSCTL_HANDLER_ARGS)
281 {
282 	u_int val;
283 
284 #ifdef VIMAGE
285 	if(!IS_DEFAULT_VNET(curvnet))
286 		val = curvnet->vnet_sockcnt;
287 	else
288 #endif
289 		val = numopensockets;
290 	return (sysctl_handle_int(oidp, &val, 0, req));
291 }
292 SYSCTL_PROC(_kern_ipc, OID_AUTO, numopensockets,
293     CTLTYPE_UINT | CTLFLAG_RD | CTLFLAG_MPSAFE | CTLFLAG_VNET, 0, sizeof(u_int),
294     sysctl_numopensockets, "IU", "Number of open sockets");
295 
296 /*
297  * so_global_mtx protects so_gencnt, numopensockets, and the per-socket
298  * so_gencnt field.
299  */
300 static struct mtx so_global_mtx;
301 MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF);
302 
303 /*
304  * General IPC sysctl name space, used by sockets and a variety of other IPC
305  * types.
306  */
307 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
308     "IPC");
309 
310 /*
311  * Initialize the socket subsystem and set up the socket
312  * memory allocator.
313  */
314 static uma_zone_t socket_zone;
315 int	maxsockets;
316 
317 static void
318 socket_zone_change(void *tag)
319 {
320 
321 	maxsockets = uma_zone_set_max(socket_zone, maxsockets);
322 }
323 
324 static int splice_init_state;
325 static struct sx splice_init_lock;
326 SX_SYSINIT(splice_init_lock, &splice_init_lock, "splice_init");
327 
328 static SYSCTL_NODE(_kern_ipc, OID_AUTO, splice, CTLFLAG_RW, 0,
329     "Settings relating to the SO_SPLICE socket option");
330 
331 static bool splice_receive_stream = true;
332 SYSCTL_BOOL(_kern_ipc_splice, OID_AUTO, receive_stream, CTLFLAG_RWTUN,
333     &splice_receive_stream, 0,
334     "Use soreceive_stream() for stream splices");
335 
336 static uma_zone_t splice_zone;
337 static struct proc *splice_proc;
338 struct splice_wq {
339 	struct mtx	mtx;
340 	STAILQ_HEAD(, so_splice) head;
341 	bool		running;
342 } __aligned(CACHE_LINE_SIZE);
343 static struct splice_wq *splice_wq;
344 static uint32_t splice_index = 0;
345 
346 static void so_splice_timeout(void *arg, int pending);
347 static void so_splice_xfer(struct so_splice *s);
348 static int so_unsplice(struct socket *so, bool timeout);
349 
350 static void
351 splice_work_thread(void *ctx)
352 {
353 	struct splice_wq *wq = ctx;
354 	struct so_splice *s, *s_temp;
355 	STAILQ_HEAD(, so_splice) local_head;
356 	int cpu;
357 
358 	cpu = wq - splice_wq;
359 	if (bootverbose)
360 		printf("starting so_splice worker thread for CPU %d\n", cpu);
361 
362 	for (;;) {
363 		mtx_lock(&wq->mtx);
364 		while (STAILQ_EMPTY(&wq->head)) {
365 			wq->running = false;
366 			mtx_sleep(wq, &wq->mtx, 0, "-", 0);
367 			wq->running = true;
368 		}
369 		STAILQ_INIT(&local_head);
370 		STAILQ_CONCAT(&local_head, &wq->head);
371 		STAILQ_INIT(&wq->head);
372 		mtx_unlock(&wq->mtx);
373 		STAILQ_FOREACH_SAFE(s, &local_head, next, s_temp) {
374 			mtx_lock(&s->mtx);
375 			CURVNET_SET(s->src->so_vnet);
376 			so_splice_xfer(s);
377 			CURVNET_RESTORE();
378 		}
379 	}
380 }
381 
382 static void
383 so_splice_dispatch_async(struct so_splice *sp)
384 {
385 	struct splice_wq *wq;
386 	bool running;
387 
388 	wq = &splice_wq[sp->wq_index];
389 	mtx_lock(&wq->mtx);
390 	STAILQ_INSERT_TAIL(&wq->head, sp, next);
391 	running = wq->running;
392 	mtx_unlock(&wq->mtx);
393 	if (!running)
394 		wakeup(wq);
395 }
396 
397 void
398 so_splice_dispatch(struct so_splice *sp)
399 {
400 	mtx_assert(&sp->mtx, MA_OWNED);
401 
402 	if (sp->state != SPLICE_IDLE) {
403 		mtx_unlock(&sp->mtx);
404 	} else {
405 		sp->state = SPLICE_QUEUED;
406 		mtx_unlock(&sp->mtx);
407 		so_splice_dispatch_async(sp);
408 	}
409 }
410 
411 static int
412 splice_zinit(void *mem, int size __unused, int flags __unused)
413 {
414 	struct so_splice *s;
415 
416 	s = (struct so_splice *)mem;
417 	mtx_init(&s->mtx, "so_splice", NULL, MTX_DEF);
418 	return (0);
419 }
420 
421 static void
422 splice_zfini(void *mem, int size)
423 {
424 	struct so_splice *s;
425 
426 	s = (struct so_splice *)mem;
427 	mtx_destroy(&s->mtx);
428 }
429 
430 static int
431 splice_init(void)
432 {
433 	struct thread *td;
434 	int error, i, state;
435 
436 	state = atomic_load_acq_int(&splice_init_state);
437 	if (__predict_true(state > 0))
438 		return (0);
439 	if (state < 0)
440 		return (ENXIO);
441 	sx_xlock(&splice_init_lock);
442 	if (splice_init_state != 0) {
443 		sx_xunlock(&splice_init_lock);
444 		return (0);
445 	}
446 
447 	splice_zone = uma_zcreate("splice", sizeof(struct so_splice), NULL,
448 	    NULL, splice_zinit, splice_zfini, UMA_ALIGN_CACHE, 0);
449 
450 	splice_wq = mallocarray(mp_maxid + 1, sizeof(*splice_wq), M_TEMP,
451 	    M_WAITOK | M_ZERO);
452 
453 	/*
454 	 * Initialize the workqueues to run the splice work.  We create a
455 	 * work queue for each CPU.
456 	 */
457 	CPU_FOREACH(i) {
458 		STAILQ_INIT(&splice_wq[i].head);
459 		mtx_init(&splice_wq[i].mtx, "splice work queue", NULL, MTX_DEF);
460 	}
461 
462 	/* Start kthreads for each workqueue. */
463 	error = 0;
464 	CPU_FOREACH(i) {
465 		error = kproc_kthread_add(splice_work_thread, &splice_wq[i],
466 		    &splice_proc, &td, 0, 0, "so_splice", "thr_%d", i);
467 		if (error) {
468 			printf("Can't add so_splice thread %d error %d\n",
469 			    i, error);
470 			break;
471 		}
472 
473 		/*
474 		 * It's possible to create loops with SO_SPLICE; ensure that
475 		 * worker threads aren't able to starve the system too easily.
476 		 */
477 		thread_lock(td);
478 		sched_prio(td, PUSER);
479 		thread_unlock(td);
480 	}
481 
482 	splice_init_state = error != 0 ? -1 : 1;
483 	sx_xunlock(&splice_init_lock);
484 
485 	return (error);
486 }
487 
488 /*
489  * Lock a pair of socket's I/O locks for splicing.  Avoid blocking while holding
490  * one lock in order to avoid potential deadlocks in case there is some other
491  * code path which acquires more than one I/O lock at a time.
492  */
493 static void
494 splice_lock_pair(struct socket *so_src, struct socket *so_dst)
495 {
496 	int error;
497 
498 	for (;;) {
499 		error = SOCK_IO_SEND_LOCK(so_dst, SBL_WAIT | SBL_NOINTR);
500 		KASSERT(error == 0,
501 		    ("%s: failed to lock send I/O lock: %d", __func__, error));
502 		error = SOCK_IO_RECV_LOCK(so_src, 0);
503 		KASSERT(error == 0 || error == EWOULDBLOCK,
504 		    ("%s: failed to lock recv I/O lock: %d", __func__, error));
505 		if (error == 0)
506 			break;
507 		SOCK_IO_SEND_UNLOCK(so_dst);
508 
509 		error = SOCK_IO_RECV_LOCK(so_src, SBL_WAIT | SBL_NOINTR);
510 		KASSERT(error == 0,
511 		    ("%s: failed to lock recv I/O lock: %d", __func__, error));
512 		error = SOCK_IO_SEND_LOCK(so_dst, 0);
513 		KASSERT(error == 0 || error == EWOULDBLOCK,
514 		    ("%s: failed to lock send I/O lock: %d", __func__, error));
515 		if (error == 0)
516 			break;
517 		SOCK_IO_RECV_UNLOCK(so_src);
518 	}
519 }
520 
521 static void
522 splice_unlock_pair(struct socket *so_src, struct socket *so_dst)
523 {
524 	SOCK_IO_RECV_UNLOCK(so_src);
525 	SOCK_IO_SEND_UNLOCK(so_dst);
526 }
527 
528 /*
529  * Move data from the source to the sink.  Assumes that both of the relevant
530  * socket I/O locks are held.
531  */
532 static int
533 so_splice_xfer_data(struct socket *so_src, struct socket *so_dst, off_t max,
534     ssize_t *lenp)
535 {
536 	struct uio uio;
537 	struct mbuf *m;
538 	struct sockbuf *sb_src, *sb_dst;
539 	ssize_t len;
540 	long space;
541 	int error, flags;
542 
543 	SOCK_IO_RECV_ASSERT_LOCKED(so_src);
544 	SOCK_IO_SEND_ASSERT_LOCKED(so_dst);
545 
546 	error = 0;
547 	m = NULL;
548 	memset(&uio, 0, sizeof(uio));
549 
550 	sb_src = &so_src->so_rcv;
551 	sb_dst = &so_dst->so_snd;
552 
553 	space = sbspace(sb_dst);
554 	if (space < 0)
555 		space = 0;
556 	len = MIN(max, MIN(space, sbavail(sb_src)));
557 	if (len == 0) {
558 		SOCK_RECVBUF_LOCK(so_src);
559 		if ((sb_src->sb_state & SBS_CANTRCVMORE) != 0)
560 			error = EPIPE;
561 		SOCK_RECVBUF_UNLOCK(so_src);
562 	} else {
563 		flags = MSG_DONTWAIT;
564 		uio.uio_resid = len;
565 		if (splice_receive_stream && sb_src->sb_tls_info == NULL) {
566 			error = soreceive_stream_locked(so_src, sb_src, NULL,
567 			    &uio, &m, NULL, flags);
568 		} else {
569 			error = soreceive_generic_locked(so_src, NULL,
570 			    &uio, &m, NULL, &flags);
571 		}
572 		if (error != 0 && m != NULL) {
573 			m_freem(m);
574 			m = NULL;
575 		}
576 	}
577 	if (m != NULL) {
578 		len -= uio.uio_resid;
579 		error = sosend_generic_locked(so_dst, NULL, NULL, m, NULL,
580 		    MSG_DONTWAIT, curthread);
581 	} else if (error == 0) {
582 		len = 0;
583 		SOCK_SENDBUF_LOCK(so_dst);
584 		if ((sb_dst->sb_state & SBS_CANTSENDMORE) != 0)
585 			error = EPIPE;
586 		SOCK_SENDBUF_UNLOCK(so_dst);
587 	}
588 	if (error == 0)
589 		*lenp = len;
590 	return (error);
591 }
592 
593 /*
594  * Transfer data from the source to the sink.
595  */
596 static void
597 so_splice_xfer(struct so_splice *sp)
598 {
599 	struct socket *so_src, *so_dst;
600 	off_t max;
601 	ssize_t len;
602 	int error;
603 
604 	mtx_assert(&sp->mtx, MA_OWNED);
605 	KASSERT(sp->state == SPLICE_QUEUED || sp->state == SPLICE_CLOSING,
606 	    ("so_splice_xfer: invalid state %d", sp->state));
607 	KASSERT(sp->max != 0, ("so_splice_xfer: max == 0"));
608 
609 	if (sp->state == SPLICE_CLOSING) {
610 		/* Userspace asked us to close the splice. */
611 		goto closing;
612 	}
613 
614 	sp->state = SPLICE_RUNNING;
615 	so_src = sp->src;
616 	so_dst = sp->dst;
617 	max = sp->max > 0 ? sp->max - so_src->so_splice_sent : OFF_MAX;
618 	if (max < 0)
619 		max = 0;
620 
621 	/*
622 	 * Lock the sockets in order to block userspace from doing anything
623 	 * sneaky.  If an error occurs or one of the sockets can no longer
624 	 * transfer data, we will automatically unsplice.
625 	 */
626 	mtx_unlock(&sp->mtx);
627 	splice_lock_pair(so_src, so_dst);
628 
629 	error = so_splice_xfer_data(so_src, so_dst, max, &len);
630 
631 	mtx_lock(&sp->mtx);
632 
633 	/*
634 	 * Update our stats while still holding the socket locks.  This
635 	 * synchronizes with getsockopt(SO_SPLICE), see the comment there.
636 	 */
637 	if (error == 0) {
638 		KASSERT(len >= 0, ("%s: len %zd < 0", __func__, len));
639 		so_src->so_splice_sent += len;
640 	}
641 	splice_unlock_pair(so_src, so_dst);
642 
643 	switch (sp->state) {
644 	case SPLICE_CLOSING:
645 closing:
646 		sp->state = SPLICE_CLOSED;
647 		wakeup(sp);
648 		mtx_unlock(&sp->mtx);
649 		break;
650 	case SPLICE_RUNNING:
651 		if (error != 0 ||
652 		    (sp->max > 0 && so_src->so_splice_sent >= sp->max)) {
653 			sp->state = SPLICE_EXCEPTION;
654 			soref(so_src);
655 			mtx_unlock(&sp->mtx);
656 			(void)so_unsplice(so_src, false);
657 			sorele(so_src);
658 		} else {
659 			/*
660 			 * Locklessly check for additional bytes in the source's
661 			 * receive buffer and queue more work if possible.  We
662 			 * may end up queuing needless work, but that's ok, and
663 			 * if we race with a thread inserting more data into the
664 			 * buffer and observe sbavail() == 0, the splice mutex
665 			 * ensures that splice_push() will queue more work for
666 			 * us.
667 			 */
668 			if (sbavail(&so_src->so_rcv) > 0 &&
669 			    sbspace(&so_dst->so_snd) > 0) {
670 				sp->state = SPLICE_QUEUED;
671 				mtx_unlock(&sp->mtx);
672 				so_splice_dispatch_async(sp);
673 			} else {
674 				sp->state = SPLICE_IDLE;
675 				mtx_unlock(&sp->mtx);
676 			}
677 		}
678 		break;
679 	default:
680 		__assert_unreachable();
681 	}
682 }
683 
684 static void
685 socket_init(void *tag)
686 {
687 
688 	socket_zone = uma_zcreate("socket", sizeof(struct socket), NULL, NULL,
689 	    NULL, NULL, UMA_ALIGN_PTR, 0);
690 	maxsockets = uma_zone_set_max(socket_zone, maxsockets);
691 	uma_zone_set_warning(socket_zone, "kern.ipc.maxsockets limit reached");
692 	EVENTHANDLER_REGISTER(maxsockets_change, socket_zone_change, NULL,
693 	    EVENTHANDLER_PRI_FIRST);
694 }
695 SYSINIT(socket, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, socket_init, NULL);
696 
697 #ifdef SOCKET_HHOOK
698 static void
699 socket_hhook_register(int subtype)
700 {
701 
702 	if (hhook_head_register(HHOOK_TYPE_SOCKET, subtype,
703 	    &V_socket_hhh[subtype],
704 	    HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
705 		printf("%s: WARNING: unable to register hook\n", __func__);
706 }
707 
708 static void
709 socket_hhook_deregister(int subtype)
710 {
711 
712 	if (hhook_head_deregister(V_socket_hhh[subtype]) != 0)
713 		printf("%s: WARNING: unable to deregister hook\n", __func__);
714 }
715 
716 static void
717 socket_vnet_init(const void *unused __unused)
718 {
719 	int i;
720 
721 	/* We expect a contiguous range */
722 	for (i = 0; i <= HHOOK_SOCKET_LAST; i++)
723 		socket_hhook_register(i);
724 }
725 VNET_SYSINIT(socket_vnet_init, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
726     socket_vnet_init, NULL);
727 
728 static void
729 socket_vnet_uninit(const void *unused __unused)
730 {
731 	int i;
732 
733 	for (i = 0; i <= HHOOK_SOCKET_LAST; i++)
734 		socket_hhook_deregister(i);
735 }
736 VNET_SYSUNINIT(socket_vnet_uninit, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
737     socket_vnet_uninit, NULL);
738 #endif	/* SOCKET_HHOOK */
739 
740 /*
741  * Initialise maxsockets.  This SYSINIT must be run after
742  * tunable_mbinit().
743  */
744 static void
745 init_maxsockets(void *ignored)
746 {
747 
748 	TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets);
749 	maxsockets = imax(maxsockets, maxfiles);
750 }
751 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL);
752 
753 /*
754  * Sysctl to get and set the maximum global sockets limit.  Notify protocols
755  * of the change so that they can update their dependent limits as required.
756  */
757 static int
758 sysctl_maxsockets(SYSCTL_HANDLER_ARGS)
759 {
760 	int error, newmaxsockets;
761 
762 	newmaxsockets = maxsockets;
763 	error = sysctl_handle_int(oidp, &newmaxsockets, 0, req);
764 	if (error == 0 && req->newptr && newmaxsockets != maxsockets) {
765 		if (newmaxsockets > maxsockets &&
766 		    newmaxsockets <= maxfiles) {
767 			maxsockets = newmaxsockets;
768 			EVENTHANDLER_INVOKE(maxsockets_change);
769 		} else
770 			error = EINVAL;
771 	}
772 	return (error);
773 }
774 SYSCTL_PROC(_kern_ipc, OID_AUTO, maxsockets,
775     CTLTYPE_INT | CTLFLAG_RWTUN | CTLFLAG_NOFETCH | CTLFLAG_MPSAFE,
776     &maxsockets, 0, sysctl_maxsockets, "IU",
777     "Maximum number of sockets available");
778 
779 /*
780  * Socket operation routines.  These routines are called by the routines in
781  * sys_socket.c or from a system process, and implement the semantics of
782  * socket operations by switching out to the protocol specific routines.
783  */
784 
785 /*
786  * Get a socket structure from our zone, and initialize it.  Note that it
787  * would probably be better to allocate socket and PCB at the same time, but
788  * I'm not convinced that all the protocols can be easily modified to do
789  * this.
790  *
791  * soalloc() returns a socket with a ref count of 0.
792  */
793 static struct socket *
794 soalloc(struct vnet *vnet)
795 {
796 	struct socket *so;
797 
798 	so = uma_zalloc(socket_zone, M_NOWAIT | M_ZERO);
799 	if (so == NULL)
800 		return (NULL);
801 #ifdef MAC
802 	if (mac_socket_init(so, M_NOWAIT) != 0) {
803 		uma_zfree(socket_zone, so);
804 		return (NULL);
805 	}
806 #endif
807 	if (khelp_init_osd(HELPER_CLASS_SOCKET, &so->osd)) {
808 		uma_zfree(socket_zone, so);
809 		return (NULL);
810 	}
811 
812 	/*
813 	 * The socket locking protocol allows to lock 2 sockets at a time,
814 	 * however, the first one must be a listening socket.  WITNESS lacks
815 	 * a feature to change class of an existing lock, so we use DUPOK.
816 	 */
817 	mtx_init(&so->so_lock, "socket", NULL, MTX_DEF | MTX_DUPOK);
818 	mtx_init(&so->so_snd_mtx, "so_snd", NULL, MTX_DEF);
819 	mtx_init(&so->so_rcv_mtx, "so_rcv", NULL, MTX_DEF);
820 	so->so_rcv.sb_sel = &so->so_rdsel;
821 	so->so_snd.sb_sel = &so->so_wrsel;
822 	sx_init(&so->so_snd_sx, "so_snd_sx");
823 	sx_init(&so->so_rcv_sx, "so_rcv_sx");
824 	TAILQ_INIT(&so->so_snd.sb_aiojobq);
825 	TAILQ_INIT(&so->so_rcv.sb_aiojobq);
826 	TASK_INIT(&so->so_snd.sb_aiotask, 0, soaio_snd, so);
827 	TASK_INIT(&so->so_rcv.sb_aiotask, 0, soaio_rcv, so);
828 #ifdef VIMAGE
829 	VNET_ASSERT(vnet != NULL, ("%s:%d vnet is NULL, so=%p",
830 	    __func__, __LINE__, so));
831 	so->so_vnet = vnet;
832 #endif
833 #ifdef SOCKET_HHOOK
834 	/* We shouldn't need the so_global_mtx */
835 	if (hhook_run_socket(so, NULL, HHOOK_SOCKET_CREATE)) {
836 		/* Do we need more comprehensive error returns? */
837 		uma_zfree(socket_zone, so);
838 		return (NULL);
839 	}
840 #endif
841 	mtx_lock(&so_global_mtx);
842 	so->so_gencnt = ++so_gencnt;
843 	++numopensockets;
844 #ifdef VIMAGE
845 	vnet->vnet_sockcnt++;
846 #endif
847 	mtx_unlock(&so_global_mtx);
848 
849 	return (so);
850 }
851 
852 /*
853  * Free the storage associated with a socket at the socket layer, tear down
854  * locks, labels, etc.  All protocol state is assumed already to have been
855  * torn down (and possibly never set up) by the caller.
856  */
857 void
858 sodealloc(struct socket *so)
859 {
860 
861 	KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count));
862 	KASSERT(so->so_pcb == NULL, ("sodealloc(): so_pcb != NULL"));
863 
864 	mtx_lock(&so_global_mtx);
865 	so->so_gencnt = ++so_gencnt;
866 	--numopensockets;	/* Could be below, but faster here. */
867 #ifdef VIMAGE
868 	VNET_ASSERT(so->so_vnet != NULL, ("%s:%d so_vnet is NULL, so=%p",
869 	    __func__, __LINE__, so));
870 	so->so_vnet->vnet_sockcnt--;
871 #endif
872 	mtx_unlock(&so_global_mtx);
873 #ifdef MAC
874 	mac_socket_destroy(so);
875 #endif
876 #ifdef SOCKET_HHOOK
877 	hhook_run_socket(so, NULL, HHOOK_SOCKET_CLOSE);
878 #endif
879 
880 	khelp_destroy_osd(&so->osd);
881 	if (SOLISTENING(so)) {
882 		if (so->sol_accept_filter != NULL)
883 			accept_filt_setopt(so, NULL);
884 	} else {
885 		if (so->so_rcv.sb_hiwat)
886 			(void)chgsbsize(so->so_cred->cr_uidinfo,
887 			    &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY);
888 		if (so->so_snd.sb_hiwat)
889 			(void)chgsbsize(so->so_cred->cr_uidinfo,
890 			    &so->so_snd.sb_hiwat, 0, RLIM_INFINITY);
891 		sx_destroy(&so->so_snd_sx);
892 		sx_destroy(&so->so_rcv_sx);
893 		mtx_destroy(&so->so_snd_mtx);
894 		mtx_destroy(&so->so_rcv_mtx);
895 	}
896 	crfree(so->so_cred);
897 	mtx_destroy(&so->so_lock);
898 	uma_zfree(socket_zone, so);
899 }
900 
901 /*
902  * socreate returns a socket with a ref count of 1 and a file descriptor
903  * reference.  The socket should be closed with soclose().
904  */
905 int
906 socreate(int dom, struct socket **aso, int type, int proto,
907     struct ucred *cred, struct thread *td)
908 {
909 	struct protosw *prp;
910 	struct socket *so;
911 	int error;
912 
913 	/*
914 	 * XXX: divert(4) historically abused PF_INET.  Keep this compatibility
915 	 * shim until all applications have been updated.
916 	 */
917 	if (__predict_false(dom == PF_INET && type == SOCK_RAW &&
918 	    proto == IPPROTO_DIVERT)) {
919 		dom = PF_DIVERT;
920 		printf("%s uses obsolete way to create divert(4) socket\n",
921 		    td->td_proc->p_comm);
922 	}
923 
924 	prp = pffindproto(dom, type, proto);
925 	if (prp == NULL) {
926 		/* No support for domain. */
927 		if (pffinddomain(dom) == NULL)
928 			return (EAFNOSUPPORT);
929 		/* No support for socket type. */
930 		if (proto == 0 && type != 0)
931 			return (EPROTOTYPE);
932 		return (EPROTONOSUPPORT);
933 	}
934 
935 	MPASS(prp->pr_attach);
936 
937 	if ((prp->pr_flags & PR_CAPATTACH) == 0) {
938 		if (CAP_TRACING(td))
939 			ktrcapfail(CAPFAIL_PROTO, &proto);
940 		if (IN_CAPABILITY_MODE(td))
941 			return (ECAPMODE);
942 	}
943 
944 	if (prison_check_af(cred, prp->pr_domain->dom_family) != 0)
945 		return (EPROTONOSUPPORT);
946 
947 	so = soalloc(CRED_TO_VNET(cred));
948 	if (so == NULL)
949 		return (ENOBUFS);
950 
951 	so->so_type = type;
952 	so->so_cred = crhold(cred);
953 	if ((prp->pr_domain->dom_family == PF_INET) ||
954 	    (prp->pr_domain->dom_family == PF_INET6) ||
955 	    (prp->pr_domain->dom_family == PF_ROUTE))
956 		so->so_fibnum = td->td_proc->p_fibnum;
957 	else
958 		so->so_fibnum = 0;
959 	so->so_proto = prp;
960 #ifdef MAC
961 	mac_socket_create(cred, so);
962 #endif
963 	knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
964 	    so_rdknl_assert_lock);
965 	knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
966 	    so_wrknl_assert_lock);
967 	if ((prp->pr_flags & PR_SOCKBUF) == 0) {
968 		so->so_snd.sb_mtx = &so->so_snd_mtx;
969 		so->so_rcv.sb_mtx = &so->so_rcv_mtx;
970 	}
971 	/*
972 	 * Auto-sizing of socket buffers is managed by the protocols and
973 	 * the appropriate flags must be set in the pru_attach function.
974 	 */
975 	CURVNET_SET(so->so_vnet);
976 	error = prp->pr_attach(so, proto, td);
977 	CURVNET_RESTORE();
978 	if (error) {
979 		sodealloc(so);
980 		return (error);
981 	}
982 	soref(so);
983 	*aso = so;
984 	return (0);
985 }
986 
987 #ifdef REGRESSION
988 static int regression_sonewconn_earlytest = 1;
989 SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW,
990     &regression_sonewconn_earlytest, 0, "Perform early sonewconn limit test");
991 #endif
992 
993 static int sooverprio = LOG_DEBUG;
994 SYSCTL_INT(_kern_ipc, OID_AUTO, sooverprio, CTLFLAG_RW,
995     &sooverprio, 0, "Log priority for listen socket overflows: 0..7 or -1 to disable");
996 
997 static struct timeval overinterval = { 60, 0 };
998 SYSCTL_TIMEVAL_SEC(_kern_ipc, OID_AUTO, sooverinterval, CTLFLAG_RW,
999     &overinterval,
1000     "Delay in seconds between warnings for listen socket overflows");
1001 
1002 /*
1003  * When an attempt at a new connection is noted on a socket which supports
1004  * accept(2), the protocol has two options:
1005  * 1) Call legacy sonewconn() function, which would call protocol attach
1006  *    method, same as used for socket(2).
1007  * 2) Call solisten_clone(), do attach that is specific to a cloned connection,
1008  *    and then call solisten_enqueue().
1009  *
1010  * Note: the ref count on the socket is 0 on return.
1011  */
1012 struct socket *
1013 solisten_clone(struct socket *head)
1014 {
1015 	struct sbuf descrsb;
1016 	struct socket *so;
1017 	int len, overcount;
1018 	u_int qlen;
1019 	const char localprefix[] = "local:";
1020 	char descrbuf[SUNPATHLEN + sizeof(localprefix)];
1021 #if defined(INET6)
1022 	char addrbuf[INET6_ADDRSTRLEN];
1023 #elif defined(INET)
1024 	char addrbuf[INET_ADDRSTRLEN];
1025 #endif
1026 	bool dolog, over;
1027 
1028 	SOLISTEN_LOCK(head);
1029 	over = (head->sol_qlen > 3 * head->sol_qlimit / 2);
1030 #ifdef REGRESSION
1031 	if (regression_sonewconn_earlytest && over) {
1032 #else
1033 	if (over) {
1034 #endif
1035 		head->sol_overcount++;
1036 		dolog = (sooverprio >= 0) &&
1037 			!!ratecheck(&head->sol_lastover, &overinterval);
1038 
1039 		/*
1040 		 * If we're going to log, copy the overflow count and queue
1041 		 * length from the listen socket before dropping the lock.
1042 		 * Also, reset the overflow count.
1043 		 */
1044 		if (dolog) {
1045 			overcount = head->sol_overcount;
1046 			head->sol_overcount = 0;
1047 			qlen = head->sol_qlen;
1048 		}
1049 		SOLISTEN_UNLOCK(head);
1050 
1051 		if (dolog) {
1052 			/*
1053 			 * Try to print something descriptive about the
1054 			 * socket for the error message.
1055 			 */
1056 			sbuf_new(&descrsb, descrbuf, sizeof(descrbuf),
1057 			    SBUF_FIXEDLEN);
1058 			switch (head->so_proto->pr_domain->dom_family) {
1059 #if defined(INET) || defined(INET6)
1060 #ifdef INET
1061 			case AF_INET:
1062 #endif
1063 #ifdef INET6
1064 			case AF_INET6:
1065 				if (head->so_proto->pr_domain->dom_family ==
1066 				    AF_INET6 ||
1067 				    (sotoinpcb(head)->inp_inc.inc_flags &
1068 				    INC_ISIPV6)) {
1069 					ip6_sprintf(addrbuf,
1070 					    &sotoinpcb(head)->inp_inc.inc6_laddr);
1071 					sbuf_printf(&descrsb, "[%s]", addrbuf);
1072 				} else
1073 #endif
1074 				{
1075 #ifdef INET
1076 					inet_ntoa_r(
1077 					    sotoinpcb(head)->inp_inc.inc_laddr,
1078 					    addrbuf);
1079 					sbuf_cat(&descrsb, addrbuf);
1080 #endif
1081 				}
1082 				sbuf_printf(&descrsb, ":%hu (proto %u)",
1083 				    ntohs(sotoinpcb(head)->inp_inc.inc_lport),
1084 				    head->so_proto->pr_protocol);
1085 				break;
1086 #endif /* INET || INET6 */
1087 			case AF_UNIX:
1088 				sbuf_cat(&descrsb, localprefix);
1089 				if (sotounpcb(head)->unp_addr != NULL)
1090 					len =
1091 					    sotounpcb(head)->unp_addr->sun_len -
1092 					    offsetof(struct sockaddr_un,
1093 					    sun_path);
1094 				else
1095 					len = 0;
1096 				if (len > 0)
1097 					sbuf_bcat(&descrsb,
1098 					    sotounpcb(head)->unp_addr->sun_path,
1099 					    len);
1100 				else
1101 					sbuf_cat(&descrsb, "(unknown)");
1102 				break;
1103 			}
1104 
1105 			/*
1106 			 * If we can't print something more specific, at least
1107 			 * print the domain name.
1108 			 */
1109 			if (sbuf_finish(&descrsb) != 0 ||
1110 			    sbuf_len(&descrsb) <= 0) {
1111 				sbuf_clear(&descrsb);
1112 				sbuf_cat(&descrsb,
1113 				    head->so_proto->pr_domain->dom_name ?:
1114 				    "unknown");
1115 				sbuf_finish(&descrsb);
1116 			}
1117 			KASSERT(sbuf_len(&descrsb) > 0,
1118 			    ("%s: sbuf creation failed", __func__));
1119 			/*
1120 			 * Preserve the historic listen queue overflow log
1121 			 * message, that starts with "sonewconn:".  It has
1122 			 * been known to sysadmins for years and also test
1123 			 * sys/kern/sonewconn_overflow checks for it.
1124 			 */
1125 			if (head->so_cred == 0) {
1126 				log(LOG_PRI(sooverprio),
1127 				    "sonewconn: pcb %p (%s): "
1128 				    "Listen queue overflow: %i already in "
1129 				    "queue awaiting acceptance (%d "
1130 				    "occurrences)\n", head->so_pcb,
1131 				    sbuf_data(&descrsb),
1132 			    	qlen, overcount);
1133 			} else {
1134 				log(LOG_PRI(sooverprio),
1135 				    "sonewconn: pcb %p (%s): "
1136 				    "Listen queue overflow: "
1137 				    "%i already in queue awaiting acceptance "
1138 				    "(%d occurrences), euid %d, rgid %d, jail %s\n",
1139 				    head->so_pcb, sbuf_data(&descrsb), qlen,
1140 				    overcount, head->so_cred->cr_uid,
1141 				    head->so_cred->cr_rgid,
1142 				    head->so_cred->cr_prison ?
1143 					head->so_cred->cr_prison->pr_name :
1144 					"not_jailed");
1145 			}
1146 			sbuf_delete(&descrsb);
1147 
1148 			overcount = 0;
1149 		}
1150 
1151 		return (NULL);
1152 	}
1153 	SOLISTEN_UNLOCK(head);
1154 	VNET_ASSERT(head->so_vnet != NULL, ("%s: so %p vnet is NULL",
1155 	    __func__, head));
1156 	so = soalloc(head->so_vnet);
1157 	if (so == NULL) {
1158 		log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: "
1159 		    "limit reached or out of memory\n",
1160 		    __func__, head->so_pcb);
1161 		return (NULL);
1162 	}
1163 	so->so_listen = head;
1164 	so->so_type = head->so_type;
1165 	/*
1166 	 * POSIX is ambiguous on what options an accept(2)ed socket should
1167 	 * inherit from the listener.  Words "create a new socket" may be
1168 	 * interpreted as not inheriting anything.  Best programming practice
1169 	 * for application developers is to not rely on such inheritance.
1170 	 * FreeBSD had historically inherited all so_options excluding
1171 	 * SO_ACCEPTCONN, which virtually means all SOL_SOCKET level options,
1172 	 * including those completely irrelevant to a new born socket.  For
1173 	 * compatibility with older versions we will inherit a list of
1174 	 * meaningful options.
1175 	 * The crucial bit to inherit is SO_ACCEPTFILTER.  We need it present
1176 	 * in the child socket for soisconnected() promoting socket from the
1177 	 * incomplete queue to complete.  It will be cleared before the child
1178 	 * gets available to accept(2).
1179 	 */
1180 	so->so_options = head->so_options & (SO_ACCEPTFILTER | SO_KEEPALIVE |
1181 	    SO_DONTROUTE | SO_LINGER | SO_OOBINLINE | SO_NOSIGPIPE);
1182 	so->so_linger = head->so_linger;
1183 	so->so_state = head->so_state;
1184 	so->so_fibnum = head->so_fibnum;
1185 	so->so_proto = head->so_proto;
1186 	so->so_cred = crhold(head->so_cred);
1187 #ifdef SOCKET_HHOOK
1188 	if (V_socket_hhh[HHOOK_SOCKET_NEWCONN]->hhh_nhooks > 0) {
1189 		if (hhook_run_socket(so, head, HHOOK_SOCKET_NEWCONN)) {
1190 			sodealloc(so);
1191 			log(LOG_DEBUG, "%s: hhook run failed\n", __func__);
1192 			return (NULL);
1193 		}
1194 	}
1195 #endif
1196 #ifdef MAC
1197 	mac_socket_newconn(head, so);
1198 #endif
1199 	knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
1200 	    so_rdknl_assert_lock);
1201 	knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
1202 	    so_wrknl_assert_lock);
1203 	VNET_SO_ASSERT(head);
1204 	if (soreserve(so, head->sol_sbsnd_hiwat, head->sol_sbrcv_hiwat)) {
1205 		sodealloc(so);
1206 		log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n",
1207 		    __func__, head->so_pcb);
1208 		return (NULL);
1209 	}
1210 	so->so_rcv.sb_lowat = head->sol_sbrcv_lowat;
1211 	so->so_snd.sb_lowat = head->sol_sbsnd_lowat;
1212 	so->so_rcv.sb_timeo = head->sol_sbrcv_timeo;
1213 	so->so_snd.sb_timeo = head->sol_sbsnd_timeo;
1214 	so->so_rcv.sb_flags = head->sol_sbrcv_flags & SB_AUTOSIZE;
1215 	so->so_snd.sb_flags = head->sol_sbsnd_flags & SB_AUTOSIZE;
1216 	if ((so->so_proto->pr_flags & PR_SOCKBUF) == 0) {
1217 		so->so_snd.sb_mtx = &so->so_snd_mtx;
1218 		so->so_rcv.sb_mtx = &so->so_rcv_mtx;
1219 	}
1220 
1221 	return (so);
1222 }
1223 
1224 /* Connstatus may be 0 or SS_ISCONNECTED. */
1225 struct socket *
1226 sonewconn(struct socket *head, int connstatus)
1227 {
1228 	struct socket *so;
1229 
1230 	if ((so = solisten_clone(head)) == NULL)
1231 		return (NULL);
1232 
1233 	if (so->so_proto->pr_attach(so, 0, NULL) != 0) {
1234 		sodealloc(so);
1235 		log(LOG_DEBUG, "%s: pcb %p: pr_attach() failed\n",
1236 		    __func__, head->so_pcb);
1237 		return (NULL);
1238 	}
1239 
1240 	(void)solisten_enqueue(so, connstatus);
1241 
1242 	return (so);
1243 }
1244 
1245 /*
1246  * Enqueue socket cloned by solisten_clone() to the listen queue of the
1247  * listener it has been cloned from.
1248  *
1249  * Return 'true' if socket landed on complete queue, otherwise 'false'.
1250  */
1251 bool
1252 solisten_enqueue(struct socket *so, int connstatus)
1253 {
1254 	struct socket *head = so->so_listen;
1255 
1256 	MPASS(refcount_load(&so->so_count) == 0);
1257 	refcount_init(&so->so_count, 1);
1258 
1259 	SOLISTEN_LOCK(head);
1260 	if (head->sol_accept_filter != NULL)
1261 		connstatus = 0;
1262 	so->so_state |= connstatus;
1263 	soref(head); /* A socket on (in)complete queue refs head. */
1264 	if (connstatus) {
1265 		TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list);
1266 		so->so_qstate = SQ_COMP;
1267 		head->sol_qlen++;
1268 		solisten_wakeup(head);	/* unlocks */
1269 		return (true);
1270 	} else {
1271 		/*
1272 		 * Keep removing sockets from the head until there's room for
1273 		 * us to insert on the tail.  In pre-locking revisions, this
1274 		 * was a simple if(), but as we could be racing with other
1275 		 * threads and soabort() requires dropping locks, we must
1276 		 * loop waiting for the condition to be true.
1277 		 */
1278 		while (head->sol_incqlen > head->sol_qlimit) {
1279 			struct socket *sp;
1280 
1281 			sp = TAILQ_FIRST(&head->sol_incomp);
1282 			TAILQ_REMOVE(&head->sol_incomp, sp, so_list);
1283 			head->sol_incqlen--;
1284 			SOCK_LOCK(sp);
1285 			sp->so_qstate = SQ_NONE;
1286 			sp->so_listen = NULL;
1287 			SOCK_UNLOCK(sp);
1288 			sorele_locked(head);	/* does SOLISTEN_UNLOCK, head stays */
1289 			soabort(sp);
1290 			SOLISTEN_LOCK(head);
1291 		}
1292 		TAILQ_INSERT_TAIL(&head->sol_incomp, so, so_list);
1293 		so->so_qstate = SQ_INCOMP;
1294 		head->sol_incqlen++;
1295 		SOLISTEN_UNLOCK(head);
1296 		return (false);
1297 	}
1298 }
1299 
1300 #if defined(SCTP) || defined(SCTP_SUPPORT)
1301 /*
1302  * Socket part of sctp_peeloff().  Detach a new socket from an
1303  * association.  The new socket is returned with a reference.
1304  *
1305  * XXXGL: reduce copy-paste with solisten_clone().
1306  */
1307 struct socket *
1308 sopeeloff(struct socket *head)
1309 {
1310 	struct socket *so;
1311 
1312 	VNET_ASSERT(head->so_vnet != NULL, ("%s:%d so_vnet is NULL, head=%p",
1313 	    __func__, __LINE__, head));
1314 	so = soalloc(head->so_vnet);
1315 	if (so == NULL) {
1316 		log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: "
1317 		    "limit reached or out of memory\n",
1318 		    __func__, head->so_pcb);
1319 		return (NULL);
1320 	}
1321 	so->so_type = head->so_type;
1322 	so->so_options = head->so_options;
1323 	so->so_linger = head->so_linger;
1324 	so->so_state = (head->so_state & SS_NBIO) | SS_ISCONNECTED;
1325 	so->so_fibnum = head->so_fibnum;
1326 	so->so_proto = head->so_proto;
1327 	so->so_cred = crhold(head->so_cred);
1328 #ifdef MAC
1329 	mac_socket_newconn(head, so);
1330 #endif
1331 	knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
1332 	    so_rdknl_assert_lock);
1333 	knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
1334 	    so_wrknl_assert_lock);
1335 	VNET_SO_ASSERT(head);
1336 	if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat)) {
1337 		sodealloc(so);
1338 		log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n",
1339 		    __func__, head->so_pcb);
1340 		return (NULL);
1341 	}
1342 	if ((*so->so_proto->pr_attach)(so, 0, NULL)) {
1343 		sodealloc(so);
1344 		log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n",
1345 		    __func__, head->so_pcb);
1346 		return (NULL);
1347 	}
1348 	so->so_rcv.sb_lowat = head->so_rcv.sb_lowat;
1349 	so->so_snd.sb_lowat = head->so_snd.sb_lowat;
1350 	so->so_rcv.sb_timeo = head->so_rcv.sb_timeo;
1351 	so->so_snd.sb_timeo = head->so_snd.sb_timeo;
1352 	so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE;
1353 	so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE;
1354 	if ((so->so_proto->pr_flags & PR_SOCKBUF) == 0) {
1355 		so->so_snd.sb_mtx = &so->so_snd_mtx;
1356 		so->so_rcv.sb_mtx = &so->so_rcv_mtx;
1357 	}
1358 
1359 	soref(so);
1360 
1361 	return (so);
1362 }
1363 #endif	/* SCTP */
1364 
1365 int
1366 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
1367 {
1368 	int error;
1369 
1370 	CURVNET_SET(so->so_vnet);
1371 	error = so->so_proto->pr_bind(so, nam, td);
1372 	CURVNET_RESTORE();
1373 	return (error);
1374 }
1375 
1376 int
1377 sobindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
1378 {
1379 	int error;
1380 
1381 	CURVNET_SET(so->so_vnet);
1382 	error = so->so_proto->pr_bindat(fd, so, nam, td);
1383 	CURVNET_RESTORE();
1384 	return (error);
1385 }
1386 
1387 /*
1388  * solisten() transitions a socket from a non-listening state to a listening
1389  * state, but can also be used to update the listen queue depth on an
1390  * existing listen socket.  The protocol will call back into the sockets
1391  * layer using solisten_proto_check() and solisten_proto() to check and set
1392  * socket-layer listen state.  Call backs are used so that the protocol can
1393  * acquire both protocol and socket layer locks in whatever order is required
1394  * by the protocol.
1395  *
1396  * Protocol implementors are advised to hold the socket lock across the
1397  * socket-layer test and set to avoid races at the socket layer.
1398  */
1399 int
1400 solisten(struct socket *so, int backlog, struct thread *td)
1401 {
1402 	int error;
1403 
1404 	CURVNET_SET(so->so_vnet);
1405 	error = so->so_proto->pr_listen(so, backlog, td);
1406 	CURVNET_RESTORE();
1407 	return (error);
1408 }
1409 
1410 /*
1411  * Prepare for a call to solisten_proto().  Acquire all socket buffer locks in
1412  * order to interlock with socket I/O.
1413  */
1414 int
1415 solisten_proto_check(struct socket *so)
1416 {
1417 	SOCK_LOCK_ASSERT(so);
1418 
1419 	if ((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
1420 	    SS_ISDISCONNECTING)) != 0)
1421 		return (EINVAL);
1422 
1423 	/*
1424 	 * Sleeping is not permitted here, so simply fail if userspace is
1425 	 * attempting to transmit or receive on the socket.  This kind of
1426 	 * transient failure is not ideal, but it should occur only if userspace
1427 	 * is misusing the socket interfaces.
1428 	 */
1429 	if (!sx_try_xlock(&so->so_snd_sx))
1430 		return (EAGAIN);
1431 	if (!sx_try_xlock(&so->so_rcv_sx)) {
1432 		sx_xunlock(&so->so_snd_sx);
1433 		return (EAGAIN);
1434 	}
1435 	mtx_lock(&so->so_snd_mtx);
1436 	mtx_lock(&so->so_rcv_mtx);
1437 
1438 	/* Interlock with soo_aio_queue() and KTLS. */
1439 	if (!SOLISTENING(so)) {
1440 		bool ktls;
1441 
1442 #ifdef KERN_TLS
1443 		ktls = so->so_snd.sb_tls_info != NULL ||
1444 		    so->so_rcv.sb_tls_info != NULL;
1445 #else
1446 		ktls = false;
1447 #endif
1448 		if (ktls ||
1449 		    (so->so_snd.sb_flags & (SB_AIO | SB_AIO_RUNNING)) != 0 ||
1450 		    (so->so_rcv.sb_flags & (SB_AIO | SB_AIO_RUNNING)) != 0) {
1451 			solisten_proto_abort(so);
1452 			return (EINVAL);
1453 		}
1454 	}
1455 
1456 	return (0);
1457 }
1458 
1459 /*
1460  * Undo the setup done by solisten_proto_check().
1461  */
1462 void
1463 solisten_proto_abort(struct socket *so)
1464 {
1465 	mtx_unlock(&so->so_snd_mtx);
1466 	mtx_unlock(&so->so_rcv_mtx);
1467 	sx_xunlock(&so->so_snd_sx);
1468 	sx_xunlock(&so->so_rcv_sx);
1469 }
1470 
1471 void
1472 solisten_proto(struct socket *so, int backlog)
1473 {
1474 	int sbrcv_lowat, sbsnd_lowat;
1475 	u_int sbrcv_hiwat, sbsnd_hiwat;
1476 	short sbrcv_flags, sbsnd_flags;
1477 	sbintime_t sbrcv_timeo, sbsnd_timeo;
1478 
1479 	SOCK_LOCK_ASSERT(so);
1480 	KASSERT((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
1481 	    SS_ISDISCONNECTING)) == 0,
1482 	    ("%s: bad socket state %p", __func__, so));
1483 
1484 	if (SOLISTENING(so))
1485 		goto listening;
1486 
1487 	/*
1488 	 * Change this socket to listening state.
1489 	 */
1490 	sbrcv_lowat = so->so_rcv.sb_lowat;
1491 	sbsnd_lowat = so->so_snd.sb_lowat;
1492 	sbrcv_hiwat = so->so_rcv.sb_hiwat;
1493 	sbsnd_hiwat = so->so_snd.sb_hiwat;
1494 	sbrcv_flags = so->so_rcv.sb_flags;
1495 	sbsnd_flags = so->so_snd.sb_flags;
1496 	sbrcv_timeo = so->so_rcv.sb_timeo;
1497 	sbsnd_timeo = so->so_snd.sb_timeo;
1498 
1499 #ifdef MAC
1500 	mac_socketpeer_label_free(so->so_peerlabel);
1501 #endif
1502 
1503 	if (!(so->so_proto->pr_flags & PR_SOCKBUF)) {
1504 		sbdestroy(so, SO_SND);
1505 		sbdestroy(so, SO_RCV);
1506 	}
1507 
1508 #ifdef INVARIANTS
1509 	bzero(&so->so_rcv,
1510 	    sizeof(struct socket) - offsetof(struct socket, so_rcv));
1511 #endif
1512 
1513 	so->sol_sbrcv_lowat = sbrcv_lowat;
1514 	so->sol_sbsnd_lowat = sbsnd_lowat;
1515 	so->sol_sbrcv_hiwat = sbrcv_hiwat;
1516 	so->sol_sbsnd_hiwat = sbsnd_hiwat;
1517 	so->sol_sbrcv_flags = sbrcv_flags;
1518 	so->sol_sbsnd_flags = sbsnd_flags;
1519 	so->sol_sbrcv_timeo = sbrcv_timeo;
1520 	so->sol_sbsnd_timeo = sbsnd_timeo;
1521 
1522 	so->sol_qlen = so->sol_incqlen = 0;
1523 	TAILQ_INIT(&so->sol_incomp);
1524 	TAILQ_INIT(&so->sol_comp);
1525 
1526 	so->sol_accept_filter = NULL;
1527 	so->sol_accept_filter_arg = NULL;
1528 	so->sol_accept_filter_str = NULL;
1529 
1530 	so->sol_upcall = NULL;
1531 	so->sol_upcallarg = NULL;
1532 
1533 	so->so_options |= SO_ACCEPTCONN;
1534 
1535 listening:
1536 	if (backlog < 0 || backlog > V_somaxconn)
1537 		backlog = V_somaxconn;
1538 	so->sol_qlimit = backlog;
1539 
1540 	mtx_unlock(&so->so_snd_mtx);
1541 	mtx_unlock(&so->so_rcv_mtx);
1542 	sx_xunlock(&so->so_snd_sx);
1543 	sx_xunlock(&so->so_rcv_sx);
1544 }
1545 
1546 /*
1547  * Wakeup listeners/subsystems once we have a complete connection.
1548  * Enters with lock, returns unlocked.
1549  */
1550 void
1551 solisten_wakeup(struct socket *sol)
1552 {
1553 
1554 	if (sol->sol_upcall != NULL)
1555 		(void )sol->sol_upcall(sol, sol->sol_upcallarg, M_NOWAIT);
1556 	else {
1557 		selwakeuppri(&sol->so_rdsel, PSOCK);
1558 		KNOTE_LOCKED(&sol->so_rdsel.si_note, 0);
1559 	}
1560 	SOLISTEN_UNLOCK(sol);
1561 	wakeup_one(&sol->sol_comp);
1562 	if ((sol->so_state & SS_ASYNC) && sol->so_sigio != NULL)
1563 		pgsigio(&sol->so_sigio, SIGIO, 0);
1564 }
1565 
1566 /*
1567  * Return single connection off a listening socket queue.  Main consumer of
1568  * the function is kern_accept4().  Some modules, that do their own accept
1569  * management also use the function.  The socket reference held by the
1570  * listen queue is handed to the caller.
1571  *
1572  * Listening socket must be locked on entry and is returned unlocked on
1573  * return.
1574  * The flags argument is set of accept4(2) flags and ACCEPT4_INHERIT.
1575  */
1576 int
1577 solisten_dequeue(struct socket *head, struct socket **ret, int flags)
1578 {
1579 	struct socket *so;
1580 	int error;
1581 
1582 	SOLISTEN_LOCK_ASSERT(head);
1583 
1584 	while (!(head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp) &&
1585 	    head->so_error == 0) {
1586 		error = msleep(&head->sol_comp, SOCK_MTX(head), PSOCK | PCATCH,
1587 		    "accept", 0);
1588 		if (error != 0) {
1589 			SOLISTEN_UNLOCK(head);
1590 			return (error);
1591 		}
1592 	}
1593 	if (head->so_error) {
1594 		error = head->so_error;
1595 		head->so_error = 0;
1596 	} else if ((head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp))
1597 		error = EWOULDBLOCK;
1598 	else
1599 		error = 0;
1600 	if (error) {
1601 		SOLISTEN_UNLOCK(head);
1602 		return (error);
1603 	}
1604 	so = TAILQ_FIRST(&head->sol_comp);
1605 	SOCK_LOCK(so);
1606 	KASSERT(so->so_qstate == SQ_COMP,
1607 	    ("%s: so %p not SQ_COMP", __func__, so));
1608 	head->sol_qlen--;
1609 	so->so_qstate = SQ_NONE;
1610 	so->so_listen = NULL;
1611 	TAILQ_REMOVE(&head->sol_comp, so, so_list);
1612 	if (flags & ACCEPT4_INHERIT)
1613 		so->so_state |= (head->so_state & SS_NBIO);
1614 	else
1615 		so->so_state |= (flags & SOCK_NONBLOCK) ? SS_NBIO : 0;
1616 	SOCK_UNLOCK(so);
1617 	sorele_locked(head);
1618 
1619 	*ret = so;
1620 	return (0);
1621 }
1622 
1623 static struct so_splice *
1624 so_splice_alloc(off_t max)
1625 {
1626 	struct so_splice *sp;
1627 
1628 	sp = uma_zalloc(splice_zone, M_WAITOK);
1629 	sp->src = NULL;
1630 	sp->dst = NULL;
1631 	sp->max = max > 0 ? max : -1;
1632 	do {
1633 		sp->wq_index = atomic_fetchadd_32(&splice_index, 1) %
1634 		    (mp_maxid + 1);
1635 	} while (CPU_ABSENT(sp->wq_index));
1636 	sp->state = SPLICE_INIT;
1637 	TIMEOUT_TASK_INIT(taskqueue_thread, &sp->timeout, 0, so_splice_timeout,
1638 	    sp);
1639 	return (sp);
1640 }
1641 
1642 static void
1643 so_splice_free(struct so_splice *sp)
1644 {
1645 	KASSERT(sp->state == SPLICE_CLOSED,
1646 	    ("so_splice_free: sp %p not closed", sp));
1647 	uma_zfree(splice_zone, sp);
1648 }
1649 
1650 static void
1651 so_splice_timeout(void *arg, int pending __unused)
1652 {
1653 	struct so_splice *sp;
1654 
1655 	sp = arg;
1656 	(void)so_unsplice(sp->src, true);
1657 }
1658 
1659 /*
1660  * Splice the output from so to the input of so2.
1661  */
1662 static int
1663 so_splice(struct socket *so, struct socket *so2, struct splice *splice)
1664 {
1665 	struct so_splice *sp;
1666 	int error;
1667 
1668 	if (splice->sp_max < 0)
1669 		return (EINVAL);
1670 	/* Handle only TCP for now; TODO: other streaming protos */
1671 	if (so->so_proto->pr_protocol != IPPROTO_TCP ||
1672 	    so2->so_proto->pr_protocol != IPPROTO_TCP)
1673 		return (EPROTONOSUPPORT);
1674 	if (so->so_vnet != so2->so_vnet)
1675 		return (EINVAL);
1676 
1677 	/* so_splice_xfer() assumes that we're using these implementations. */
1678 	KASSERT(so->so_proto->pr_sosend == sosend_generic,
1679 	    ("so_splice: sosend not sosend_generic"));
1680 	KASSERT(so2->so_proto->pr_soreceive == soreceive_generic ||
1681 	    so2->so_proto->pr_soreceive == soreceive_stream,
1682 	    ("so_splice: soreceive not soreceive_generic/stream"));
1683 
1684 	sp = so_splice_alloc(splice->sp_max);
1685 	so->so_splice_sent = 0;
1686 	sp->src = so;
1687 	sp->dst = so2;
1688 
1689 	error = 0;
1690 	SOCK_LOCK(so);
1691 	if (SOLISTENING(so))
1692 		error = EINVAL;
1693 	else if ((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING)) == 0)
1694 		error = ENOTCONN;
1695 	else if (so->so_splice != NULL)
1696 		error = EBUSY;
1697 	if (error != 0) {
1698 		SOCK_UNLOCK(so);
1699 		uma_zfree(splice_zone, sp);
1700 		return (error);
1701 	}
1702 	SOCK_RECVBUF_LOCK(so);
1703 	if (so->so_rcv.sb_tls_info != NULL) {
1704 		SOCK_RECVBUF_UNLOCK(so);
1705 		SOCK_UNLOCK(so);
1706 		uma_zfree(splice_zone, sp);
1707 		return (EINVAL);
1708 	}
1709 	so->so_rcv.sb_flags |= SB_SPLICED;
1710 	so->so_splice = sp;
1711 	soref(so);
1712 	SOCK_RECVBUF_UNLOCK(so);
1713 	SOCK_UNLOCK(so);
1714 
1715 	error = 0;
1716 	SOCK_LOCK(so2);
1717 	if (SOLISTENING(so2))
1718 		error = EINVAL;
1719 	else if ((so2->so_state & (SS_ISCONNECTED | SS_ISCONNECTING)) == 0)
1720 		error = ENOTCONN;
1721 	else if (so2->so_splice_back != NULL)
1722 		error = EBUSY;
1723 	if (error != 0) {
1724 		SOCK_UNLOCK(so2);
1725 		so_unsplice(so, false);
1726 		return (error);
1727 	}
1728 	SOCK_SENDBUF_LOCK(so2);
1729 	if (so->so_snd.sb_tls_info != NULL) {
1730 		SOCK_SENDBUF_UNLOCK(so2);
1731 		SOCK_UNLOCK(so2);
1732 		so_unsplice(so, false);
1733 		return (EINVAL);
1734 	}
1735 	so2->so_snd.sb_flags |= SB_SPLICED;
1736 	so2->so_splice_back = sp;
1737 	soref(so2);
1738 	mtx_lock(&sp->mtx);
1739 	SOCK_SENDBUF_UNLOCK(so2);
1740 	SOCK_UNLOCK(so2);
1741 
1742 	if (splice->sp_idle.tv_sec != 0 || splice->sp_idle.tv_usec != 0) {
1743 		taskqueue_enqueue_timeout_sbt(taskqueue_thread, &sp->timeout,
1744 		    tvtosbt(splice->sp_idle), 0, C_PREL(4));
1745 	}
1746 
1747 	/*
1748 	 * Transfer any data already present in the socket buffer.
1749 	 */
1750 	KASSERT(sp->state == SPLICE_INIT,
1751 	    ("so_splice: splice %p state %d", sp, sp->state));
1752 	sp->state = SPLICE_QUEUED;
1753 	so_splice_xfer(sp);
1754 	return (0);
1755 }
1756 
1757 static int
1758 so_unsplice(struct socket *so, bool timeout)
1759 {
1760 	struct socket *so2;
1761 	struct so_splice *sp;
1762 	bool drain, so2rele;
1763 
1764 	/*
1765 	 * First unset SB_SPLICED and hide the splice structure so that
1766 	 * wakeup routines will stop enqueuing work.  This also ensures that
1767 	 * a only a single thread will proceed with the unsplice.
1768 	 */
1769 	SOCK_LOCK(so);
1770 	if (SOLISTENING(so)) {
1771 		SOCK_UNLOCK(so);
1772 		return (EINVAL);
1773 	}
1774 	SOCK_RECVBUF_LOCK(so);
1775 	if ((so->so_rcv.sb_flags & SB_SPLICED) == 0) {
1776 		SOCK_RECVBUF_UNLOCK(so);
1777 		SOCK_UNLOCK(so);
1778 		return (ENOTCONN);
1779 	}
1780 	sp = so->so_splice;
1781 	mtx_lock(&sp->mtx);
1782 	if (sp->state == SPLICE_INIT) {
1783 		/*
1784 		 * A splice is in the middle of being set up.
1785 		 */
1786 		mtx_unlock(&sp->mtx);
1787 		SOCK_RECVBUF_UNLOCK(so);
1788 		SOCK_UNLOCK(so);
1789 		return (ENOTCONN);
1790 	}
1791 	mtx_unlock(&sp->mtx);
1792 	so->so_rcv.sb_flags &= ~SB_SPLICED;
1793 	so->so_splice = NULL;
1794 	SOCK_RECVBUF_UNLOCK(so);
1795 	SOCK_UNLOCK(so);
1796 
1797 	so2 = sp->dst;
1798 	SOCK_LOCK(so2);
1799 	KASSERT(!SOLISTENING(so2), ("%s: so2 is listening", __func__));
1800 	SOCK_SENDBUF_LOCK(so2);
1801 	KASSERT(sp->state == SPLICE_INIT ||
1802 	    (so2->so_snd.sb_flags & SB_SPLICED) != 0,
1803 	    ("%s: so2 is not spliced", __func__));
1804 	KASSERT(sp->state == SPLICE_INIT ||
1805 	    so2->so_splice_back == sp,
1806 	    ("%s: so_splice_back != sp", __func__));
1807 	so2->so_snd.sb_flags &= ~SB_SPLICED;
1808 	so2rele = so2->so_splice_back != NULL;
1809 	so2->so_splice_back = NULL;
1810 	SOCK_SENDBUF_UNLOCK(so2);
1811 	SOCK_UNLOCK(so2);
1812 
1813 	/*
1814 	 * No new work is being enqueued.  The worker thread might be
1815 	 * splicing data right now, in which case we want to wait for it to
1816 	 * finish before proceeding.
1817 	 */
1818 	mtx_lock(&sp->mtx);
1819 	switch (sp->state) {
1820 	case SPLICE_QUEUED:
1821 	case SPLICE_RUNNING:
1822 		sp->state = SPLICE_CLOSING;
1823 		while (sp->state == SPLICE_CLOSING)
1824 			msleep(sp, &sp->mtx, PSOCK, "unsplice", 0);
1825 		break;
1826 	case SPLICE_INIT:
1827 	case SPLICE_IDLE:
1828 	case SPLICE_EXCEPTION:
1829 		sp->state = SPLICE_CLOSED;
1830 		break;
1831 	default:
1832 		__assert_unreachable();
1833 	}
1834 	if (!timeout) {
1835 		drain = taskqueue_cancel_timeout(taskqueue_thread, &sp->timeout,
1836 		    NULL) != 0;
1837 	} else {
1838 		drain = false;
1839 	}
1840 	mtx_unlock(&sp->mtx);
1841 	if (drain)
1842 		taskqueue_drain_timeout(taskqueue_thread, &sp->timeout);
1843 
1844 	/*
1845 	 * Now we hold the sole reference to the splice structure.
1846 	 * Clean up: signal userspace and release socket references.
1847 	 */
1848 	sorwakeup(so);
1849 	CURVNET_SET(so->so_vnet);
1850 	sorele(so);
1851 	sowwakeup(so2);
1852 	if (so2rele)
1853 		sorele(so2);
1854 	CURVNET_RESTORE();
1855 	so_splice_free(sp);
1856 	return (0);
1857 }
1858 
1859 /*
1860  * Free socket upon release of the very last reference.
1861  */
1862 static void
1863 sofree(struct socket *so)
1864 {
1865 	struct protosw *pr = so->so_proto;
1866 
1867 	SOCK_LOCK_ASSERT(so);
1868 	KASSERT(refcount_load(&so->so_count) == 0,
1869 	    ("%s: so %p has references", __func__, so));
1870 	KASSERT(SOLISTENING(so) || so->so_qstate == SQ_NONE,
1871 	    ("%s: so %p is on listen queue", __func__, so));
1872 	KASSERT(SOLISTENING(so) || (so->so_rcv.sb_flags & SB_SPLICED) == 0,
1873 	    ("%s: so %p rcvbuf is spliced", __func__, so));
1874 	KASSERT(SOLISTENING(so) || (so->so_snd.sb_flags & SB_SPLICED) == 0,
1875 	    ("%s: so %p sndbuf is spliced", __func__, so));
1876 	KASSERT(so->so_splice == NULL && so->so_splice_back == NULL,
1877 	    ("%s: so %p has spliced data", __func__, so));
1878 
1879 	SOCK_UNLOCK(so);
1880 
1881 	if (so->so_dtor != NULL)
1882 		so->so_dtor(so);
1883 
1884 	VNET_SO_ASSERT(so);
1885 	if (pr->pr_detach != NULL)
1886 		pr->pr_detach(so);
1887 
1888 	if (!(pr->pr_flags & PR_SOCKBUF) && !SOLISTENING(so)) {
1889 		/*
1890 		 * From this point on, we assume that no other references to
1891 		 * this socket exist anywhere else in the stack.  Therefore,
1892 		 * no locks need to be acquired or held.
1893 		 */
1894 #ifdef INVARIANTS
1895 		SOCK_SENDBUF_LOCK(so);
1896 		SOCK_RECVBUF_LOCK(so);
1897 #endif
1898 		sbdestroy(so, SO_SND);
1899 		sbdestroy(so, SO_RCV);
1900 #ifdef INVARIANTS
1901 		SOCK_SENDBUF_UNLOCK(so);
1902 		SOCK_RECVBUF_UNLOCK(so);
1903 #endif
1904 	}
1905 	seldrain(&so->so_rdsel);
1906 	seldrain(&so->so_wrsel);
1907 	knlist_destroy(&so->so_rdsel.si_note);
1908 	knlist_destroy(&so->so_wrsel.si_note);
1909 	sodealloc(so);
1910 }
1911 
1912 /*
1913  * Release a reference on a socket while holding the socket lock.
1914  * Unlocks the socket lock before returning.
1915  */
1916 void
1917 sorele_locked(struct socket *so)
1918 {
1919 	SOCK_LOCK_ASSERT(so);
1920 	if (refcount_release(&so->so_count))
1921 		sofree(so);
1922 	else
1923 		SOCK_UNLOCK(so);
1924 }
1925 
1926 /*
1927  * Close a socket on last file table reference removal.  Initiate disconnect
1928  * if connected.  Free socket when disconnect complete.
1929  *
1930  * This function will sorele() the socket.  Note that soclose() may be called
1931  * prior to the ref count reaching zero.  The actual socket structure will
1932  * not be freed until the ref count reaches zero.
1933  */
1934 int
1935 soclose(struct socket *so)
1936 {
1937 	struct accept_queue lqueue;
1938 	int error = 0;
1939 	bool listening, last __diagused;
1940 
1941 	CURVNET_SET(so->so_vnet);
1942 	funsetown(&so->so_sigio);
1943 	if (so->so_state & SS_ISCONNECTED) {
1944 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
1945 			error = sodisconnect(so);
1946 			if (error) {
1947 				if (error == ENOTCONN)
1948 					error = 0;
1949 				goto drop;
1950 			}
1951 		}
1952 
1953 		if ((so->so_options & SO_LINGER) != 0 && so->so_linger != 0) {
1954 			if ((so->so_state & SS_ISDISCONNECTING) &&
1955 			    (so->so_state & SS_NBIO))
1956 				goto drop;
1957 			while (so->so_state & SS_ISCONNECTED) {
1958 				error = tsleep(&so->so_timeo,
1959 				    PSOCK | PCATCH, "soclos",
1960 				    so->so_linger * hz);
1961 				if (error)
1962 					break;
1963 			}
1964 		}
1965 	}
1966 
1967 drop:
1968 	if (so->so_proto->pr_close != NULL)
1969 		so->so_proto->pr_close(so);
1970 
1971 	SOCK_LOCK(so);
1972 	if ((listening = SOLISTENING(so))) {
1973 		struct socket *sp;
1974 
1975 		TAILQ_INIT(&lqueue);
1976 		TAILQ_SWAP(&lqueue, &so->sol_incomp, socket, so_list);
1977 		TAILQ_CONCAT(&lqueue, &so->sol_comp, so_list);
1978 
1979 		so->sol_qlen = so->sol_incqlen = 0;
1980 
1981 		TAILQ_FOREACH(sp, &lqueue, so_list) {
1982 			SOCK_LOCK(sp);
1983 			sp->so_qstate = SQ_NONE;
1984 			sp->so_listen = NULL;
1985 			SOCK_UNLOCK(sp);
1986 			last = refcount_release(&so->so_count);
1987 			KASSERT(!last, ("%s: released last reference for %p",
1988 			    __func__, so));
1989 		}
1990 	}
1991 	sorele_locked(so);
1992 	if (listening) {
1993 		struct socket *sp, *tsp;
1994 
1995 		TAILQ_FOREACH_SAFE(sp, &lqueue, so_list, tsp)
1996 			soabort(sp);
1997 	}
1998 	CURVNET_RESTORE();
1999 	return (error);
2000 }
2001 
2002 /*
2003  * soabort() is used to abruptly tear down a connection, such as when a
2004  * resource limit is reached (listen queue depth exceeded), or if a listen
2005  * socket is closed while there are sockets waiting to be accepted.
2006  *
2007  * This interface is tricky, because it is called on an unreferenced socket,
2008  * and must be called only by a thread that has actually removed the socket
2009  * from the listen queue it was on.  Likely this thread holds the last
2010  * reference on the socket and soabort() will proceed with sofree().  But
2011  * it might be not the last, as the sockets on the listen queues are seen
2012  * from the protocol side.
2013  *
2014  * This interface will call into the protocol code, so must not be called
2015  * with any socket locks held.  Protocols do call it while holding their own
2016  * recursible protocol mutexes, but this is something that should be subject
2017  * to review in the future.
2018  *
2019  * Usually socket should have a single reference left, but this is not a
2020  * requirement.  In the past, when we have had named references for file
2021  * descriptor and protocol, we asserted that none of them are being held.
2022  */
2023 void
2024 soabort(struct socket *so)
2025 {
2026 
2027 	VNET_SO_ASSERT(so);
2028 
2029 	if (so->so_proto->pr_abort != NULL)
2030 		so->so_proto->pr_abort(so);
2031 	SOCK_LOCK(so);
2032 	sorele_locked(so);
2033 }
2034 
2035 int
2036 soaccept(struct socket *so, struct sockaddr *sa)
2037 {
2038 #ifdef INVARIANTS
2039 	u_char len = sa->sa_len;
2040 #endif
2041 	int error;
2042 
2043 	CURVNET_SET(so->so_vnet);
2044 	error = so->so_proto->pr_accept(so, sa);
2045 	KASSERT(sa->sa_len <= len,
2046 	    ("%s: protocol %p sockaddr overflow", __func__, so->so_proto));
2047 	CURVNET_RESTORE();
2048 	return (error);
2049 }
2050 
2051 int
2052 sopeeraddr(struct socket *so, struct sockaddr *sa)
2053 {
2054 #ifdef INVARIANTS
2055 	u_char len = sa->sa_len;
2056 #endif
2057 	int error;
2058 
2059 	CURVNET_ASSERT_SET();
2060 
2061 	error = so->so_proto->pr_peeraddr(so, sa);
2062 	KASSERT(sa->sa_len <= len,
2063 	    ("%s: protocol %p sockaddr overflow", __func__, so->so_proto));
2064 
2065 	return (error);
2066 }
2067 
2068 int
2069 sosockaddr(struct socket *so, struct sockaddr *sa)
2070 {
2071 #ifdef INVARIANTS
2072 	u_char len = sa->sa_len;
2073 #endif
2074 	int error;
2075 
2076 	CURVNET_SET(so->so_vnet);
2077 	error = so->so_proto->pr_sockaddr(so, sa);
2078 	KASSERT(sa->sa_len <= len,
2079 	    ("%s: protocol %p sockaddr overflow", __func__, so->so_proto));
2080 	CURVNET_RESTORE();
2081 
2082 	return (error);
2083 }
2084 
2085 int
2086 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
2087 {
2088 
2089 	return (soconnectat(AT_FDCWD, so, nam, td));
2090 }
2091 
2092 int
2093 soconnectat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
2094 {
2095 	int error;
2096 
2097 	CURVNET_SET(so->so_vnet);
2098 
2099 	/*
2100 	 * If protocol is connection-based, can only connect once.
2101 	 * Otherwise, if connected, try to disconnect first.  This allows
2102 	 * user to disconnect by connecting to, e.g., a null address.
2103 	 *
2104 	 * Note, this check is racy and may need to be re-evaluated at the
2105 	 * protocol layer.
2106 	 */
2107 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
2108 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
2109 	    (error = sodisconnect(so)))) {
2110 		error = EISCONN;
2111 	} else {
2112 		/*
2113 		 * Prevent accumulated error from previous connection from
2114 		 * biting us.
2115 		 */
2116 		so->so_error = 0;
2117 		if (fd == AT_FDCWD) {
2118 			error = so->so_proto->pr_connect(so, nam, td);
2119 		} else {
2120 			error = so->so_proto->pr_connectat(fd, so, nam, td);
2121 		}
2122 	}
2123 	CURVNET_RESTORE();
2124 
2125 	return (error);
2126 }
2127 
2128 int
2129 soconnect2(struct socket *so1, struct socket *so2)
2130 {
2131 	int error;
2132 
2133 	CURVNET_SET(so1->so_vnet);
2134 	error = so1->so_proto->pr_connect2(so1, so2);
2135 	CURVNET_RESTORE();
2136 	return (error);
2137 }
2138 
2139 int
2140 sodisconnect(struct socket *so)
2141 {
2142 	int error;
2143 
2144 	if ((so->so_state & SS_ISCONNECTED) == 0)
2145 		return (ENOTCONN);
2146 	if (so->so_state & SS_ISDISCONNECTING)
2147 		return (EALREADY);
2148 	VNET_SO_ASSERT(so);
2149 	error = so->so_proto->pr_disconnect(so);
2150 	return (error);
2151 }
2152 
2153 int
2154 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio,
2155     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
2156 {
2157 	long space;
2158 	ssize_t resid;
2159 	int clen = 0, error, dontroute;
2160 
2161 	KASSERT(so->so_type == SOCK_DGRAM, ("sosend_dgram: !SOCK_DGRAM"));
2162 	KASSERT(so->so_proto->pr_flags & PR_ATOMIC,
2163 	    ("sosend_dgram: !PR_ATOMIC"));
2164 
2165 	if (uio != NULL)
2166 		resid = uio->uio_resid;
2167 	else
2168 		resid = top->m_pkthdr.len;
2169 	/*
2170 	 * In theory resid should be unsigned.  However, space must be
2171 	 * signed, as it might be less than 0 if we over-committed, and we
2172 	 * must use a signed comparison of space and resid.  On the other
2173 	 * hand, a negative resid causes us to loop sending 0-length
2174 	 * segments to the protocol.
2175 	 */
2176 	if (resid < 0) {
2177 		error = EINVAL;
2178 		goto out;
2179 	}
2180 
2181 	dontroute =
2182 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0;
2183 	if (td != NULL)
2184 		td->td_ru.ru_msgsnd++;
2185 	if (control != NULL)
2186 		clen = control->m_len;
2187 
2188 	SOCKBUF_LOCK(&so->so_snd);
2189 	if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
2190 		SOCKBUF_UNLOCK(&so->so_snd);
2191 		error = EPIPE;
2192 		goto out;
2193 	}
2194 	if (so->so_error) {
2195 		error = so->so_error;
2196 		so->so_error = 0;
2197 		SOCKBUF_UNLOCK(&so->so_snd);
2198 		goto out;
2199 	}
2200 	if ((so->so_state & SS_ISCONNECTED) == 0) {
2201 		/*
2202 		 * `sendto' and `sendmsg' is allowed on a connection-based
2203 		 * socket if it supports implied connect.  Return ENOTCONN if
2204 		 * not connected and no address is supplied.
2205 		 */
2206 		if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
2207 		    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
2208 			if (!(resid == 0 && clen != 0)) {
2209 				SOCKBUF_UNLOCK(&so->so_snd);
2210 				error = ENOTCONN;
2211 				goto out;
2212 			}
2213 		} else if (addr == NULL) {
2214 			if (so->so_proto->pr_flags & PR_CONNREQUIRED)
2215 				error = ENOTCONN;
2216 			else
2217 				error = EDESTADDRREQ;
2218 			SOCKBUF_UNLOCK(&so->so_snd);
2219 			goto out;
2220 		}
2221 	}
2222 
2223 	/*
2224 	 * Do we need MSG_OOB support in SOCK_DGRAM?  Signs here may be a
2225 	 * problem and need fixing.
2226 	 */
2227 	space = sbspace(&so->so_snd);
2228 	if (flags & MSG_OOB)
2229 		space += 1024;
2230 	space -= clen;
2231 	SOCKBUF_UNLOCK(&so->so_snd);
2232 	if (resid > space) {
2233 		error = EMSGSIZE;
2234 		goto out;
2235 	}
2236 	if (uio == NULL) {
2237 		resid = 0;
2238 		if (flags & MSG_EOR)
2239 			top->m_flags |= M_EOR;
2240 	} else {
2241 		/*
2242 		 * Copy the data from userland into a mbuf chain.
2243 		 * If no data is to be copied in, a single empty mbuf
2244 		 * is returned.
2245 		 */
2246 		top = m_uiotombuf(uio, M_WAITOK, space, max_hdr,
2247 		    (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0)));
2248 		if (top == NULL) {
2249 			error = EFAULT;	/* only possible error */
2250 			goto out;
2251 		}
2252 		space -= resid - uio->uio_resid;
2253 		resid = uio->uio_resid;
2254 	}
2255 	KASSERT(resid == 0, ("sosend_dgram: resid != 0"));
2256 	/*
2257 	 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock
2258 	 * than with.
2259 	 */
2260 	if (dontroute) {
2261 		SOCK_LOCK(so);
2262 		so->so_options |= SO_DONTROUTE;
2263 		SOCK_UNLOCK(so);
2264 	}
2265 	/*
2266 	 * XXX all the SBS_CANTSENDMORE checks previously done could be out
2267 	 * of date.  We could have received a reset packet in an interrupt or
2268 	 * maybe we slept while doing page faults in uiomove() etc.  We could
2269 	 * probably recheck again inside the locking protection here, but
2270 	 * there are probably other places that this also happens.  We must
2271 	 * rethink this.
2272 	 */
2273 	VNET_SO_ASSERT(so);
2274 	error = so->so_proto->pr_send(so, (flags & MSG_OOB) ? PRUS_OOB :
2275 	/*
2276 	 * If the user set MSG_EOF, the protocol understands this flag and
2277 	 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND.
2278 	 */
2279 	    ((flags & MSG_EOF) &&
2280 	     (so->so_proto->pr_flags & PR_IMPLOPCL) &&
2281 	     (resid <= 0)) ?
2282 		PRUS_EOF :
2283 		/* If there is more to send set PRUS_MORETOCOME */
2284 		(flags & MSG_MORETOCOME) ||
2285 		(resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
2286 		top, addr, control, td);
2287 	if (dontroute) {
2288 		SOCK_LOCK(so);
2289 		so->so_options &= ~SO_DONTROUTE;
2290 		SOCK_UNLOCK(so);
2291 	}
2292 	clen = 0;
2293 	control = NULL;
2294 	top = NULL;
2295 out:
2296 	if (top != NULL)
2297 		m_freem(top);
2298 	if (control != NULL)
2299 		m_freem(control);
2300 	return (error);
2301 }
2302 
2303 /*
2304  * Send on a socket.  If send must go all at once and message is larger than
2305  * send buffering, then hard error.  Lock against other senders.  If must go
2306  * all at once and not enough room now, then inform user that this would
2307  * block and do nothing.  Otherwise, if nonblocking, send as much as
2308  * possible.  The data to be sent is described by "uio" if nonzero, otherwise
2309  * by the mbuf chain "top" (which must be null if uio is not).  Data provided
2310  * in mbuf chain must be small enough to send all at once.
2311  *
2312  * Returns nonzero on error, timeout or signal; callers must check for short
2313  * counts if EINTR/ERESTART are returned.  Data and control buffers are freed
2314  * on return.
2315  */
2316 static int
2317 sosend_generic_locked(struct socket *so, struct sockaddr *addr, struct uio *uio,
2318     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
2319 {
2320 	long space;
2321 	ssize_t resid;
2322 	int clen = 0, error, dontroute;
2323 	int atomic = sosendallatonce(so) || top;
2324 	int pr_send_flag;
2325 #ifdef KERN_TLS
2326 	struct ktls_session *tls;
2327 	int tls_enq_cnt, tls_send_flag;
2328 	uint8_t tls_rtype;
2329 
2330 	tls = NULL;
2331 	tls_rtype = TLS_RLTYPE_APP;
2332 #endif
2333 
2334 	SOCK_IO_SEND_ASSERT_LOCKED(so);
2335 
2336 	if (uio != NULL)
2337 		resid = uio->uio_resid;
2338 	else if ((top->m_flags & M_PKTHDR) != 0)
2339 		resid = top->m_pkthdr.len;
2340 	else
2341 		resid = m_length(top, NULL);
2342 	/*
2343 	 * In theory resid should be unsigned.  However, space must be
2344 	 * signed, as it might be less than 0 if we over-committed, and we
2345 	 * must use a signed comparison of space and resid.  On the other
2346 	 * hand, a negative resid causes us to loop sending 0-length
2347 	 * segments to the protocol.
2348 	 *
2349 	 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
2350 	 * type sockets since that's an error.
2351 	 */
2352 	if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
2353 		error = EINVAL;
2354 		goto out;
2355 	}
2356 
2357 	dontroute =
2358 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
2359 	    (so->so_proto->pr_flags & PR_ATOMIC);
2360 	if (td != NULL)
2361 		td->td_ru.ru_msgsnd++;
2362 	if (control != NULL)
2363 		clen = control->m_len;
2364 
2365 #ifdef KERN_TLS
2366 	tls_send_flag = 0;
2367 	tls = ktls_hold(so->so_snd.sb_tls_info);
2368 	if (tls != NULL) {
2369 		if (tls->mode == TCP_TLS_MODE_SW)
2370 			tls_send_flag = PRUS_NOTREADY;
2371 
2372 		if (control != NULL) {
2373 			struct cmsghdr *cm = mtod(control, struct cmsghdr *);
2374 
2375 			if (clen >= sizeof(*cm) &&
2376 			    cm->cmsg_type == TLS_SET_RECORD_TYPE) {
2377 				tls_rtype = *((uint8_t *)CMSG_DATA(cm));
2378 				clen = 0;
2379 				m_freem(control);
2380 				control = NULL;
2381 				atomic = 1;
2382 			}
2383 		}
2384 
2385 		if (resid == 0 && !ktls_permit_empty_frames(tls)) {
2386 			error = EINVAL;
2387 			goto out;
2388 		}
2389 	}
2390 #endif
2391 
2392 restart:
2393 	do {
2394 		SOCKBUF_LOCK(&so->so_snd);
2395 		if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
2396 			SOCKBUF_UNLOCK(&so->so_snd);
2397 			error = EPIPE;
2398 			goto out;
2399 		}
2400 		if (so->so_error) {
2401 			error = so->so_error;
2402 			so->so_error = 0;
2403 			SOCKBUF_UNLOCK(&so->so_snd);
2404 			goto out;
2405 		}
2406 		if ((so->so_state & SS_ISCONNECTED) == 0) {
2407 			/*
2408 			 * `sendto' and `sendmsg' is allowed on a connection-
2409 			 * based socket if it supports implied connect.
2410 			 * Return ENOTCONN if not connected and no address is
2411 			 * supplied.
2412 			 */
2413 			if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
2414 			    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
2415 				if (!(resid == 0 && clen != 0)) {
2416 					SOCKBUF_UNLOCK(&so->so_snd);
2417 					error = ENOTCONN;
2418 					goto out;
2419 				}
2420 			} else if (addr == NULL) {
2421 				SOCKBUF_UNLOCK(&so->so_snd);
2422 				if (so->so_proto->pr_flags & PR_CONNREQUIRED)
2423 					error = ENOTCONN;
2424 				else
2425 					error = EDESTADDRREQ;
2426 				goto out;
2427 			}
2428 		}
2429 		space = sbspace(&so->so_snd);
2430 		if (flags & MSG_OOB)
2431 			space += 1024;
2432 		if ((atomic && resid > so->so_snd.sb_hiwat) ||
2433 		    clen > so->so_snd.sb_hiwat) {
2434 			SOCKBUF_UNLOCK(&so->so_snd);
2435 			error = EMSGSIZE;
2436 			goto out;
2437 		}
2438 		if (space < resid + clen &&
2439 		    (atomic || space < so->so_snd.sb_lowat || space < clen)) {
2440 			if ((so->so_state & SS_NBIO) ||
2441 			    (flags & (MSG_NBIO | MSG_DONTWAIT)) != 0) {
2442 				SOCKBUF_UNLOCK(&so->so_snd);
2443 				error = EWOULDBLOCK;
2444 				goto out;
2445 			}
2446 			error = sbwait(so, SO_SND);
2447 			SOCKBUF_UNLOCK(&so->so_snd);
2448 			if (error)
2449 				goto out;
2450 			goto restart;
2451 		}
2452 		SOCKBUF_UNLOCK(&so->so_snd);
2453 		space -= clen;
2454 		do {
2455 			if (uio == NULL) {
2456 				resid = 0;
2457 				if (flags & MSG_EOR)
2458 					top->m_flags |= M_EOR;
2459 #ifdef KERN_TLS
2460 				if (tls != NULL) {
2461 					ktls_frame(top, tls, &tls_enq_cnt,
2462 					    tls_rtype);
2463 					tls_rtype = TLS_RLTYPE_APP;
2464 				}
2465 #endif
2466 			} else {
2467 				/*
2468 				 * Copy the data from userland into a mbuf
2469 				 * chain.  If resid is 0, which can happen
2470 				 * only if we have control to send, then
2471 				 * a single empty mbuf is returned.  This
2472 				 * is a workaround to prevent protocol send
2473 				 * methods to panic.
2474 				 */
2475 #ifdef KERN_TLS
2476 				if (tls != NULL) {
2477 					top = m_uiotombuf(uio, M_WAITOK, space,
2478 					    tls->params.max_frame_len,
2479 					    M_EXTPG |
2480 					    ((flags & MSG_EOR) ? M_EOR : 0));
2481 					if (top != NULL) {
2482 						ktls_frame(top, tls,
2483 						    &tls_enq_cnt, tls_rtype);
2484 					}
2485 					tls_rtype = TLS_RLTYPE_APP;
2486 				} else
2487 #endif
2488 					top = m_uiotombuf(uio, M_WAITOK, space,
2489 					    (atomic ? max_hdr : 0),
2490 					    (atomic ? M_PKTHDR : 0) |
2491 					    ((flags & MSG_EOR) ? M_EOR : 0));
2492 				if (top == NULL) {
2493 					error = EFAULT; /* only possible error */
2494 					goto out;
2495 				}
2496 				space -= resid - uio->uio_resid;
2497 				resid = uio->uio_resid;
2498 			}
2499 			if (dontroute) {
2500 				SOCK_LOCK(so);
2501 				so->so_options |= SO_DONTROUTE;
2502 				SOCK_UNLOCK(so);
2503 			}
2504 			/*
2505 			 * XXX all the SBS_CANTSENDMORE checks previously
2506 			 * done could be out of date.  We could have received
2507 			 * a reset packet in an interrupt or maybe we slept
2508 			 * while doing page faults in uiomove() etc.  We
2509 			 * could probably recheck again inside the locking
2510 			 * protection here, but there are probably other
2511 			 * places that this also happens.  We must rethink
2512 			 * this.
2513 			 */
2514 			VNET_SO_ASSERT(so);
2515 
2516 			pr_send_flag = (flags & MSG_OOB) ? PRUS_OOB :
2517 			/*
2518 			 * If the user set MSG_EOF, the protocol understands
2519 			 * this flag and nothing left to send then use
2520 			 * PRU_SEND_EOF instead of PRU_SEND.
2521 			 */
2522 			    ((flags & MSG_EOF) &&
2523 			     (so->so_proto->pr_flags & PR_IMPLOPCL) &&
2524 			     (resid <= 0)) ?
2525 				PRUS_EOF :
2526 			/* If there is more to send set PRUS_MORETOCOME. */
2527 			    (flags & MSG_MORETOCOME) ||
2528 			    (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0;
2529 
2530 #ifdef KERN_TLS
2531 			pr_send_flag |= tls_send_flag;
2532 #endif
2533 
2534 			error = so->so_proto->pr_send(so, pr_send_flag, top,
2535 			    addr, control, td);
2536 
2537 			if (dontroute) {
2538 				SOCK_LOCK(so);
2539 				so->so_options &= ~SO_DONTROUTE;
2540 				SOCK_UNLOCK(so);
2541 			}
2542 
2543 #ifdef KERN_TLS
2544 			if (tls != NULL && tls->mode == TCP_TLS_MODE_SW) {
2545 				if (error != 0) {
2546 					m_freem(top);
2547 					top = NULL;
2548 				} else {
2549 					soref(so);
2550 					ktls_enqueue(top, so, tls_enq_cnt);
2551 				}
2552 			}
2553 #endif
2554 			clen = 0;
2555 			control = NULL;
2556 			top = NULL;
2557 			if (error)
2558 				goto out;
2559 		} while (resid && space > 0);
2560 	} while (resid);
2561 
2562 out:
2563 #ifdef KERN_TLS
2564 	if (tls != NULL)
2565 		ktls_free(tls);
2566 #endif
2567 	if (top != NULL)
2568 		m_freem(top);
2569 	if (control != NULL)
2570 		m_freem(control);
2571 	return (error);
2572 }
2573 
2574 int
2575 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio,
2576     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
2577 {
2578 	int error;
2579 
2580 	error = SOCK_IO_SEND_LOCK(so, SBLOCKWAIT(flags));
2581 	if (error)
2582 		return (error);
2583 	error = sosend_generic_locked(so, addr, uio, top, control, flags, td);
2584 	SOCK_IO_SEND_UNLOCK(so);
2585 	return (error);
2586 }
2587 
2588 /*
2589  * Send to a socket from a kernel thread.
2590  *
2591  * XXXGL: in almost all cases uio is NULL and the mbuf is supplied.
2592  * Exception is nfs/bootp_subr.c.  It is arguable that the VNET context needs
2593  * to be set at all.  This function should just boil down to a static inline
2594  * calling the protocol method.
2595  */
2596 int
2597 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
2598     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
2599 {
2600 	int error;
2601 
2602 	CURVNET_SET(so->so_vnet);
2603 	error = so->so_proto->pr_sosend(so, addr, uio,
2604 	    top, control, flags, td);
2605 	CURVNET_RESTORE();
2606 	return (error);
2607 }
2608 
2609 /*
2610  * send(2), write(2) or aio_write(2) on a socket.
2611  */
2612 int
2613 sousrsend(struct socket *so, struct sockaddr *addr, struct uio *uio,
2614     struct mbuf *control, int flags, struct proc *userproc)
2615 {
2616 	struct thread *td;
2617 	ssize_t len;
2618 	int error;
2619 
2620 	td = uio->uio_td;
2621 	len = uio->uio_resid;
2622 	CURVNET_SET(so->so_vnet);
2623 	error = so->so_proto->pr_sosend(so, addr, uio, NULL, control, flags,
2624 	    td);
2625 	CURVNET_RESTORE();
2626 	if (error != 0) {
2627 		/*
2628 		 * Clear transient errors for stream protocols if they made
2629 		 * some progress.  Make exclusion for aio(4) that would
2630 		 * schedule a new write in case of EWOULDBLOCK and clear
2631 		 * error itself.  See soaio_process_job().
2632 		 */
2633 		if (uio->uio_resid != len &&
2634 		    (so->so_proto->pr_flags & PR_ATOMIC) == 0 &&
2635 		    userproc == NULL &&
2636 		    (error == ERESTART || error == EINTR ||
2637 		    error == EWOULDBLOCK))
2638 			error = 0;
2639 		/* Generation of SIGPIPE can be controlled per socket. */
2640 		if (error == EPIPE && (so->so_options & SO_NOSIGPIPE) == 0 &&
2641 		    (flags & MSG_NOSIGNAL) == 0) {
2642 			if (userproc != NULL) {
2643 				/* aio(4) job */
2644 				PROC_LOCK(userproc);
2645 				kern_psignal(userproc, SIGPIPE);
2646 				PROC_UNLOCK(userproc);
2647 			} else {
2648 				PROC_LOCK(td->td_proc);
2649 				tdsignal(td, SIGPIPE);
2650 				PROC_UNLOCK(td->td_proc);
2651 			}
2652 		}
2653 	}
2654 	return (error);
2655 }
2656 
2657 /*
2658  * The part of soreceive() that implements reading non-inline out-of-band
2659  * data from a socket.  For more complete comments, see soreceive(), from
2660  * which this code originated.
2661  *
2662  * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is
2663  * unable to return an mbuf chain to the caller.
2664  */
2665 static int
2666 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags)
2667 {
2668 	struct protosw *pr = so->so_proto;
2669 	struct mbuf *m;
2670 	int error;
2671 
2672 	KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
2673 	VNET_SO_ASSERT(so);
2674 
2675 	m = m_get(M_WAITOK, MT_DATA);
2676 	error = pr->pr_rcvoob(so, m, flags & MSG_PEEK);
2677 	if (error)
2678 		goto bad;
2679 	do {
2680 		error = uiomove(mtod(m, void *),
2681 		    (int) min(uio->uio_resid, m->m_len), uio);
2682 		m = m_free(m);
2683 	} while (uio->uio_resid && error == 0 && m);
2684 bad:
2685 	if (m != NULL)
2686 		m_freem(m);
2687 	return (error);
2688 }
2689 
2690 /*
2691  * Following replacement or removal of the first mbuf on the first mbuf chain
2692  * of a socket buffer, push necessary state changes back into the socket
2693  * buffer so that other consumers see the values consistently.  'nextrecord'
2694  * is the callers locally stored value of the original value of
2695  * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes.
2696  * NOTE: 'nextrecord' may be NULL.
2697  */
2698 static __inline void
2699 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord)
2700 {
2701 
2702 	SOCKBUF_LOCK_ASSERT(sb);
2703 	/*
2704 	 * First, update for the new value of nextrecord.  If necessary, make
2705 	 * it the first record.
2706 	 */
2707 	if (sb->sb_mb != NULL)
2708 		sb->sb_mb->m_nextpkt = nextrecord;
2709 	else
2710 		sb->sb_mb = nextrecord;
2711 
2712 	/*
2713 	 * Now update any dependent socket buffer fields to reflect the new
2714 	 * state.  This is an expanded inline of SB_EMPTY_FIXUP(), with the
2715 	 * addition of a second clause that takes care of the case where
2716 	 * sb_mb has been updated, but remains the last record.
2717 	 */
2718 	if (sb->sb_mb == NULL) {
2719 		sb->sb_mbtail = NULL;
2720 		sb->sb_lastrecord = NULL;
2721 	} else if (sb->sb_mb->m_nextpkt == NULL)
2722 		sb->sb_lastrecord = sb->sb_mb;
2723 }
2724 
2725 /*
2726  * Implement receive operations on a socket.  We depend on the way that
2727  * records are added to the sockbuf by sbappend.  In particular, each record
2728  * (mbufs linked through m_next) must begin with an address if the protocol
2729  * so specifies, followed by an optional mbuf or mbufs containing ancillary
2730  * data, and then zero or more mbufs of data.  In order to allow parallelism
2731  * between network receive and copying to user space, as well as avoid
2732  * sleeping with a mutex held, we release the socket buffer mutex during the
2733  * user space copy.  Although the sockbuf is locked, new data may still be
2734  * appended, and thus we must maintain consistency of the sockbuf during that
2735  * time.
2736  *
2737  * The caller may receive the data as a single mbuf chain by supplying an
2738  * mbuf **mp0 for use in returning the chain.  The uio is then used only for
2739  * the count in uio_resid.
2740  */
2741 static int
2742 soreceive_generic_locked(struct socket *so, struct sockaddr **psa,
2743     struct uio *uio, struct mbuf **mp, struct mbuf **controlp, int *flagsp)
2744 {
2745 	struct mbuf *m;
2746 	int flags, error, offset;
2747 	ssize_t len;
2748 	struct protosw *pr = so->so_proto;
2749 	struct mbuf *nextrecord;
2750 	int moff, type = 0;
2751 	ssize_t orig_resid = uio->uio_resid;
2752 	bool report_real_len = false;
2753 
2754 	SOCK_IO_RECV_ASSERT_LOCKED(so);
2755 
2756 	error = 0;
2757 	if (flagsp != NULL) {
2758 		report_real_len = *flagsp & MSG_TRUNC;
2759 		*flagsp &= ~MSG_TRUNC;
2760 		flags = *flagsp &~ MSG_EOR;
2761 	} else
2762 		flags = 0;
2763 
2764 restart:
2765 	SOCKBUF_LOCK(&so->so_rcv);
2766 	m = so->so_rcv.sb_mb;
2767 	/*
2768 	 * If we have less data than requested, block awaiting more (subject
2769 	 * to any timeout) if:
2770 	 *   1. the current count is less than the low water mark, or
2771 	 *   2. MSG_DONTWAIT is not set
2772 	 */
2773 	if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
2774 	    sbavail(&so->so_rcv) < uio->uio_resid) &&
2775 	    sbavail(&so->so_rcv) < so->so_rcv.sb_lowat &&
2776 	    m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
2777 		KASSERT(m != NULL || !sbavail(&so->so_rcv),
2778 		    ("receive: m == %p sbavail == %u",
2779 		    m, sbavail(&so->so_rcv)));
2780 		if (so->so_error || so->so_rerror) {
2781 			if (m != NULL)
2782 				goto dontblock;
2783 			if (so->so_error)
2784 				error = so->so_error;
2785 			else
2786 				error = so->so_rerror;
2787 			if ((flags & MSG_PEEK) == 0) {
2788 				if (so->so_error)
2789 					so->so_error = 0;
2790 				else
2791 					so->so_rerror = 0;
2792 			}
2793 			SOCKBUF_UNLOCK(&so->so_rcv);
2794 			goto release;
2795 		}
2796 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2797 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2798 			if (m != NULL)
2799 				goto dontblock;
2800 #ifdef KERN_TLS
2801 			else if (so->so_rcv.sb_tlsdcc == 0 &&
2802 			    so->so_rcv.sb_tlscc == 0) {
2803 #else
2804 			else {
2805 #endif
2806 				SOCKBUF_UNLOCK(&so->so_rcv);
2807 				goto release;
2808 			}
2809 		}
2810 		for (; m != NULL; m = m->m_next)
2811 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
2812 				m = so->so_rcv.sb_mb;
2813 				goto dontblock;
2814 			}
2815 		if ((so->so_state & (SS_ISCONNECTING | SS_ISCONNECTED |
2816 		    SS_ISDISCONNECTING | SS_ISDISCONNECTED)) == 0 &&
2817 		    (so->so_proto->pr_flags & PR_CONNREQUIRED) != 0) {
2818 			SOCKBUF_UNLOCK(&so->so_rcv);
2819 			error = ENOTCONN;
2820 			goto release;
2821 		}
2822 		if (uio->uio_resid == 0 && !report_real_len) {
2823 			SOCKBUF_UNLOCK(&so->so_rcv);
2824 			goto release;
2825 		}
2826 		if ((so->so_state & SS_NBIO) ||
2827 		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
2828 			SOCKBUF_UNLOCK(&so->so_rcv);
2829 			error = EWOULDBLOCK;
2830 			goto release;
2831 		}
2832 		SBLASTRECORDCHK(&so->so_rcv);
2833 		SBLASTMBUFCHK(&so->so_rcv);
2834 		error = sbwait(so, SO_RCV);
2835 		SOCKBUF_UNLOCK(&so->so_rcv);
2836 		if (error)
2837 			goto release;
2838 		goto restart;
2839 	}
2840 dontblock:
2841 	/*
2842 	 * From this point onward, we maintain 'nextrecord' as a cache of the
2843 	 * pointer to the next record in the socket buffer.  We must keep the
2844 	 * various socket buffer pointers and local stack versions of the
2845 	 * pointers in sync, pushing out modifications before dropping the
2846 	 * socket buffer mutex, and re-reading them when picking it up.
2847 	 *
2848 	 * Otherwise, we will race with the network stack appending new data
2849 	 * or records onto the socket buffer by using inconsistent/stale
2850 	 * versions of the field, possibly resulting in socket buffer
2851 	 * corruption.
2852 	 *
2853 	 * By holding the high-level sblock(), we prevent simultaneous
2854 	 * readers from pulling off the front of the socket buffer.
2855 	 */
2856 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2857 	if (uio->uio_td)
2858 		uio->uio_td->td_ru.ru_msgrcv++;
2859 	KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb"));
2860 	SBLASTRECORDCHK(&so->so_rcv);
2861 	SBLASTMBUFCHK(&so->so_rcv);
2862 	nextrecord = m->m_nextpkt;
2863 	if (pr->pr_flags & PR_ADDR) {
2864 		KASSERT(m->m_type == MT_SONAME,
2865 		    ("m->m_type == %d", m->m_type));
2866 		orig_resid = 0;
2867 		if (psa != NULL)
2868 			*psa = sodupsockaddr(mtod(m, struct sockaddr *),
2869 			    M_NOWAIT);
2870 		if (flags & MSG_PEEK) {
2871 			m = m->m_next;
2872 		} else {
2873 			sbfree(&so->so_rcv, m);
2874 			so->so_rcv.sb_mb = m_free(m);
2875 			m = so->so_rcv.sb_mb;
2876 			sockbuf_pushsync(&so->so_rcv, nextrecord);
2877 		}
2878 	}
2879 
2880 	/*
2881 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
2882 	 * in the first mbuf chain on the socket buffer.  If MSG_PEEK, we
2883 	 * just copy the data; if !MSG_PEEK, we call into the protocol to
2884 	 * perform externalization (or freeing if controlp == NULL).
2885 	 */
2886 	if (m != NULL && m->m_type == MT_CONTROL) {
2887 		struct mbuf *cm = NULL, *cmn;
2888 		struct mbuf **cme = &cm;
2889 #ifdef KERN_TLS
2890 		struct cmsghdr *cmsg;
2891 		struct tls_get_record tgr;
2892 
2893 		/*
2894 		 * For MSG_TLSAPPDATA, check for an alert record.
2895 		 * If found, return ENXIO without removing
2896 		 * it from the receive queue.  This allows a subsequent
2897 		 * call without MSG_TLSAPPDATA to receive it.
2898 		 * Note that, for TLS, there should only be a single
2899 		 * control mbuf with the TLS_GET_RECORD message in it.
2900 		 */
2901 		if (flags & MSG_TLSAPPDATA) {
2902 			cmsg = mtod(m, struct cmsghdr *);
2903 			if (cmsg->cmsg_type == TLS_GET_RECORD &&
2904 			    cmsg->cmsg_len == CMSG_LEN(sizeof(tgr))) {
2905 				memcpy(&tgr, CMSG_DATA(cmsg), sizeof(tgr));
2906 				if (__predict_false(tgr.tls_type ==
2907 				    TLS_RLTYPE_ALERT)) {
2908 					SOCKBUF_UNLOCK(&so->so_rcv);
2909 					error = ENXIO;
2910 					goto release;
2911 				}
2912 			}
2913 		}
2914 #endif
2915 
2916 		do {
2917 			if (flags & MSG_PEEK) {
2918 				if (controlp != NULL) {
2919 					*controlp = m_copym(m, 0, m->m_len,
2920 					    M_NOWAIT);
2921 					controlp = &(*controlp)->m_next;
2922 				}
2923 				m = m->m_next;
2924 			} else {
2925 				sbfree(&so->so_rcv, m);
2926 				so->so_rcv.sb_mb = m->m_next;
2927 				m->m_next = NULL;
2928 				*cme = m;
2929 				cme = &(*cme)->m_next;
2930 				m = so->so_rcv.sb_mb;
2931 			}
2932 		} while (m != NULL && m->m_type == MT_CONTROL);
2933 		if ((flags & MSG_PEEK) == 0)
2934 			sockbuf_pushsync(&so->so_rcv, nextrecord);
2935 		while (cm != NULL) {
2936 			cmn = cm->m_next;
2937 			cm->m_next = NULL;
2938 			if (pr->pr_domain->dom_externalize != NULL) {
2939 				SOCKBUF_UNLOCK(&so->so_rcv);
2940 				VNET_SO_ASSERT(so);
2941 				error = (*pr->pr_domain->dom_externalize)
2942 				    (cm, controlp, flags);
2943 				SOCKBUF_LOCK(&so->so_rcv);
2944 			} else if (controlp != NULL)
2945 				*controlp = cm;
2946 			else
2947 				m_freem(cm);
2948 			if (controlp != NULL) {
2949 				while (*controlp != NULL)
2950 					controlp = &(*controlp)->m_next;
2951 			}
2952 			cm = cmn;
2953 		}
2954 		if (m != NULL)
2955 			nextrecord = so->so_rcv.sb_mb->m_nextpkt;
2956 		else
2957 			nextrecord = so->so_rcv.sb_mb;
2958 		orig_resid = 0;
2959 	}
2960 	if (m != NULL) {
2961 		if ((flags & MSG_PEEK) == 0) {
2962 			KASSERT(m->m_nextpkt == nextrecord,
2963 			    ("soreceive: post-control, nextrecord !sync"));
2964 			if (nextrecord == NULL) {
2965 				KASSERT(so->so_rcv.sb_mb == m,
2966 				    ("soreceive: post-control, sb_mb!=m"));
2967 				KASSERT(so->so_rcv.sb_lastrecord == m,
2968 				    ("soreceive: post-control, lastrecord!=m"));
2969 			}
2970 		}
2971 		type = m->m_type;
2972 		if (type == MT_OOBDATA)
2973 			flags |= MSG_OOB;
2974 	} else {
2975 		if ((flags & MSG_PEEK) == 0) {
2976 			KASSERT(so->so_rcv.sb_mb == nextrecord,
2977 			    ("soreceive: sb_mb != nextrecord"));
2978 			if (so->so_rcv.sb_mb == NULL) {
2979 				KASSERT(so->so_rcv.sb_lastrecord == NULL,
2980 				    ("soreceive: sb_lastercord != NULL"));
2981 			}
2982 		}
2983 	}
2984 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2985 	SBLASTRECORDCHK(&so->so_rcv);
2986 	SBLASTMBUFCHK(&so->so_rcv);
2987 
2988 	/*
2989 	 * Now continue to read any data mbufs off of the head of the socket
2990 	 * buffer until the read request is satisfied.  Note that 'type' is
2991 	 * used to store the type of any mbuf reads that have happened so far
2992 	 * such that soreceive() can stop reading if the type changes, which
2993 	 * causes soreceive() to return only one of regular data and inline
2994 	 * out-of-band data in a single socket receive operation.
2995 	 */
2996 	moff = 0;
2997 	offset = 0;
2998 	while (m != NULL && !(m->m_flags & M_NOTAVAIL) && uio->uio_resid > 0
2999 	    && error == 0) {
3000 		/*
3001 		 * If the type of mbuf has changed since the last mbuf
3002 		 * examined ('type'), end the receive operation.
3003 		 */
3004 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
3005 		if (m->m_type == MT_OOBDATA || m->m_type == MT_CONTROL) {
3006 			if (type != m->m_type)
3007 				break;
3008 		} else if (type == MT_OOBDATA)
3009 			break;
3010 		else
3011 		    KASSERT(m->m_type == MT_DATA,
3012 			("m->m_type == %d", m->m_type));
3013 		so->so_rcv.sb_state &= ~SBS_RCVATMARK;
3014 		len = uio->uio_resid;
3015 		if (so->so_oobmark && len > so->so_oobmark - offset)
3016 			len = so->so_oobmark - offset;
3017 		if (len > m->m_len - moff)
3018 			len = m->m_len - moff;
3019 		/*
3020 		 * If mp is set, just pass back the mbufs.  Otherwise copy
3021 		 * them out via the uio, then free.  Sockbuf must be
3022 		 * consistent here (points to current mbuf, it points to next
3023 		 * record) when we drop priority; we must note any additions
3024 		 * to the sockbuf when we block interrupts again.
3025 		 */
3026 		if (mp == NULL) {
3027 			SOCKBUF_LOCK_ASSERT(&so->so_rcv);
3028 			SBLASTRECORDCHK(&so->so_rcv);
3029 			SBLASTMBUFCHK(&so->so_rcv);
3030 			SOCKBUF_UNLOCK(&so->so_rcv);
3031 			if ((m->m_flags & M_EXTPG) != 0)
3032 				error = m_unmapped_uiomove(m, moff, uio,
3033 				    (int)len);
3034 			else
3035 				error = uiomove(mtod(m, char *) + moff,
3036 				    (int)len, uio);
3037 			SOCKBUF_LOCK(&so->so_rcv);
3038 			if (error) {
3039 				/*
3040 				 * The MT_SONAME mbuf has already been removed
3041 				 * from the record, so it is necessary to
3042 				 * remove the data mbufs, if any, to preserve
3043 				 * the invariant in the case of PR_ADDR that
3044 				 * requires MT_SONAME mbufs at the head of
3045 				 * each record.
3046 				 */
3047 				if (pr->pr_flags & PR_ATOMIC &&
3048 				    ((flags & MSG_PEEK) == 0))
3049 					(void)sbdroprecord_locked(&so->so_rcv);
3050 				SOCKBUF_UNLOCK(&so->so_rcv);
3051 				goto release;
3052 			}
3053 		} else
3054 			uio->uio_resid -= len;
3055 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
3056 		if (len == m->m_len - moff) {
3057 			if (m->m_flags & M_EOR)
3058 				flags |= MSG_EOR;
3059 			if (flags & MSG_PEEK) {
3060 				m = m->m_next;
3061 				moff = 0;
3062 			} else {
3063 				nextrecord = m->m_nextpkt;
3064 				sbfree(&so->so_rcv, m);
3065 				if (mp != NULL) {
3066 					m->m_nextpkt = NULL;
3067 					*mp = m;
3068 					mp = &m->m_next;
3069 					so->so_rcv.sb_mb = m = m->m_next;
3070 					*mp = NULL;
3071 				} else {
3072 					so->so_rcv.sb_mb = m_free(m);
3073 					m = so->so_rcv.sb_mb;
3074 				}
3075 				sockbuf_pushsync(&so->so_rcv, nextrecord);
3076 				SBLASTRECORDCHK(&so->so_rcv);
3077 				SBLASTMBUFCHK(&so->so_rcv);
3078 			}
3079 		} else {
3080 			if (flags & MSG_PEEK)
3081 				moff += len;
3082 			else {
3083 				if (mp != NULL) {
3084 					if (flags & MSG_DONTWAIT) {
3085 						*mp = m_copym(m, 0, len,
3086 						    M_NOWAIT);
3087 						if (*mp == NULL) {
3088 							/*
3089 							 * m_copym() couldn't
3090 							 * allocate an mbuf.
3091 							 * Adjust uio_resid back
3092 							 * (it was adjusted
3093 							 * down by len bytes,
3094 							 * which we didn't end
3095 							 * up "copying" over).
3096 							 */
3097 							uio->uio_resid += len;
3098 							break;
3099 						}
3100 					} else {
3101 						SOCKBUF_UNLOCK(&so->so_rcv);
3102 						*mp = m_copym(m, 0, len,
3103 						    M_WAITOK);
3104 						SOCKBUF_LOCK(&so->so_rcv);
3105 					}
3106 				}
3107 				sbcut_locked(&so->so_rcv, len);
3108 			}
3109 		}
3110 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
3111 		if (so->so_oobmark) {
3112 			if ((flags & MSG_PEEK) == 0) {
3113 				so->so_oobmark -= len;
3114 				if (so->so_oobmark == 0) {
3115 					so->so_rcv.sb_state |= SBS_RCVATMARK;
3116 					break;
3117 				}
3118 			} else {
3119 				offset += len;
3120 				if (offset == so->so_oobmark)
3121 					break;
3122 			}
3123 		}
3124 		if (flags & MSG_EOR)
3125 			break;
3126 		/*
3127 		 * If the MSG_WAITALL flag is set (for non-atomic socket), we
3128 		 * must not quit until "uio->uio_resid == 0" or an error
3129 		 * termination.  If a signal/timeout occurs, return with a
3130 		 * short count but without error.  Keep sockbuf locked
3131 		 * against other readers.
3132 		 */
3133 		while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
3134 		    !sosendallatonce(so) && nextrecord == NULL) {
3135 			SOCKBUF_LOCK_ASSERT(&so->so_rcv);
3136 			if (so->so_error || so->so_rerror ||
3137 			    so->so_rcv.sb_state & SBS_CANTRCVMORE)
3138 				break;
3139 			/*
3140 			 * Notify the protocol that some data has been
3141 			 * drained before blocking.
3142 			 */
3143 			if (pr->pr_flags & PR_WANTRCVD) {
3144 				SOCKBUF_UNLOCK(&so->so_rcv);
3145 				VNET_SO_ASSERT(so);
3146 				pr->pr_rcvd(so, flags);
3147 				SOCKBUF_LOCK(&so->so_rcv);
3148 				if (__predict_false(so->so_rcv.sb_mb == NULL &&
3149 				    (so->so_error || so->so_rerror ||
3150 				    so->so_rcv.sb_state & SBS_CANTRCVMORE)))
3151 					break;
3152 			}
3153 			SBLASTRECORDCHK(&so->so_rcv);
3154 			SBLASTMBUFCHK(&so->so_rcv);
3155 			/*
3156 			 * We could receive some data while was notifying
3157 			 * the protocol. Skip blocking in this case.
3158 			 */
3159 			if (so->so_rcv.sb_mb == NULL) {
3160 				error = sbwait(so, SO_RCV);
3161 				if (error) {
3162 					SOCKBUF_UNLOCK(&so->so_rcv);
3163 					goto release;
3164 				}
3165 			}
3166 			m = so->so_rcv.sb_mb;
3167 			if (m != NULL)
3168 				nextrecord = m->m_nextpkt;
3169 		}
3170 	}
3171 
3172 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
3173 	if (m != NULL && pr->pr_flags & PR_ATOMIC) {
3174 		if (report_real_len)
3175 			uio->uio_resid -= m_length(m, NULL) - moff;
3176 		flags |= MSG_TRUNC;
3177 		if ((flags & MSG_PEEK) == 0)
3178 			(void) sbdroprecord_locked(&so->so_rcv);
3179 	}
3180 	if ((flags & MSG_PEEK) == 0) {
3181 		if (m == NULL) {
3182 			/*
3183 			 * First part is an inline SB_EMPTY_FIXUP().  Second
3184 			 * part makes sure sb_lastrecord is up-to-date if
3185 			 * there is still data in the socket buffer.
3186 			 */
3187 			so->so_rcv.sb_mb = nextrecord;
3188 			if (so->so_rcv.sb_mb == NULL) {
3189 				so->so_rcv.sb_mbtail = NULL;
3190 				so->so_rcv.sb_lastrecord = NULL;
3191 			} else if (nextrecord->m_nextpkt == NULL)
3192 				so->so_rcv.sb_lastrecord = nextrecord;
3193 		}
3194 		SBLASTRECORDCHK(&so->so_rcv);
3195 		SBLASTMBUFCHK(&so->so_rcv);
3196 		/*
3197 		 * If soreceive() is being done from the socket callback,
3198 		 * then don't need to generate ACK to peer to update window,
3199 		 * since ACK will be generated on return to TCP.
3200 		 */
3201 		if (!(flags & MSG_SOCALLBCK) &&
3202 		    (pr->pr_flags & PR_WANTRCVD)) {
3203 			SOCKBUF_UNLOCK(&so->so_rcv);
3204 			VNET_SO_ASSERT(so);
3205 			pr->pr_rcvd(so, flags);
3206 			SOCKBUF_LOCK(&so->so_rcv);
3207 		}
3208 	}
3209 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
3210 	if (orig_resid == uio->uio_resid && orig_resid &&
3211 	    (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) {
3212 		SOCKBUF_UNLOCK(&so->so_rcv);
3213 		goto restart;
3214 	}
3215 	SOCKBUF_UNLOCK(&so->so_rcv);
3216 
3217 	if (flagsp != NULL)
3218 		*flagsp |= flags;
3219 release:
3220 	return (error);
3221 }
3222 
3223 int
3224 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio,
3225     struct mbuf **mp, struct mbuf **controlp, int *flagsp)
3226 {
3227 	int error, flags;
3228 
3229 	if (psa != NULL)
3230 		*psa = NULL;
3231 	if (controlp != NULL)
3232 		*controlp = NULL;
3233 	if (flagsp != NULL) {
3234 		flags = *flagsp;
3235 		if ((flags & MSG_OOB) != 0)
3236 			return (soreceive_rcvoob(so, uio, flags));
3237 	} else {
3238 		flags = 0;
3239 	}
3240 	if (mp != NULL)
3241 		*mp = NULL;
3242 
3243 	error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags));
3244 	if (error)
3245 		return (error);
3246 	error = soreceive_generic_locked(so, psa, uio, mp, controlp, flagsp);
3247 	SOCK_IO_RECV_UNLOCK(so);
3248 	return (error);
3249 }
3250 
3251 /*
3252  * Optimized version of soreceive() for stream (TCP) sockets.
3253  */
3254 static int
3255 soreceive_stream_locked(struct socket *so, struct sockbuf *sb,
3256     struct sockaddr **psa, struct uio *uio, struct mbuf **mp0,
3257     struct mbuf **controlp, int flags)
3258 {
3259 	int len = 0, error = 0, oresid;
3260 	struct mbuf *m, *n = NULL;
3261 
3262 	SOCK_IO_RECV_ASSERT_LOCKED(so);
3263 
3264 	/* Easy one, no space to copyout anything. */
3265 	if (uio->uio_resid == 0)
3266 		return (EINVAL);
3267 	oresid = uio->uio_resid;
3268 
3269 	SOCKBUF_LOCK(sb);
3270 	/* We will never ever get anything unless we are or were connected. */
3271 	if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) {
3272 		error = ENOTCONN;
3273 		goto out;
3274 	}
3275 
3276 restart:
3277 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
3278 
3279 	/* Abort if socket has reported problems. */
3280 	if (so->so_error) {
3281 		if (sbavail(sb) > 0)
3282 			goto deliver;
3283 		if (oresid > uio->uio_resid)
3284 			goto out;
3285 		error = so->so_error;
3286 		if (!(flags & MSG_PEEK))
3287 			so->so_error = 0;
3288 		goto out;
3289 	}
3290 
3291 	/* Door is closed.  Deliver what is left, if any. */
3292 	if (sb->sb_state & SBS_CANTRCVMORE) {
3293 		if (sbavail(sb) > 0)
3294 			goto deliver;
3295 		else
3296 			goto out;
3297 	}
3298 
3299 	/* Socket buffer is empty and we shall not block. */
3300 	if (sbavail(sb) == 0 &&
3301 	    ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) {
3302 		error = EAGAIN;
3303 		goto out;
3304 	}
3305 
3306 	/* Socket buffer got some data that we shall deliver now. */
3307 	if (sbavail(sb) > 0 && !(flags & MSG_WAITALL) &&
3308 	    ((so->so_state & SS_NBIO) ||
3309 	     (flags & (MSG_DONTWAIT|MSG_NBIO)) ||
3310 	     sbavail(sb) >= sb->sb_lowat ||
3311 	     sbavail(sb) >= uio->uio_resid ||
3312 	     sbavail(sb) >= sb->sb_hiwat) ) {
3313 		goto deliver;
3314 	}
3315 
3316 	/* On MSG_WAITALL we must wait until all data or error arrives. */
3317 	if ((flags & MSG_WAITALL) &&
3318 	    (sbavail(sb) >= uio->uio_resid || sbavail(sb) >= sb->sb_hiwat))
3319 		goto deliver;
3320 
3321 	/*
3322 	 * Wait and block until (more) data comes in.
3323 	 * NB: Drops the sockbuf lock during wait.
3324 	 */
3325 	error = sbwait(so, SO_RCV);
3326 	if (error)
3327 		goto out;
3328 	goto restart;
3329 
3330 deliver:
3331 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
3332 	KASSERT(sbavail(sb) > 0, ("%s: sockbuf empty", __func__));
3333 	KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__));
3334 
3335 	/* Statistics. */
3336 	if (uio->uio_td)
3337 		uio->uio_td->td_ru.ru_msgrcv++;
3338 
3339 	/* Fill uio until full or current end of socket buffer is reached. */
3340 	len = min(uio->uio_resid, sbavail(sb));
3341 	if (mp0 != NULL) {
3342 		/* Dequeue as many mbufs as possible. */
3343 		if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) {
3344 			if (*mp0 == NULL)
3345 				*mp0 = sb->sb_mb;
3346 			else
3347 				m_cat(*mp0, sb->sb_mb);
3348 			for (m = sb->sb_mb;
3349 			     m != NULL && m->m_len <= len;
3350 			     m = m->m_next) {
3351 				KASSERT(!(m->m_flags & M_NOTAVAIL),
3352 				    ("%s: m %p not available", __func__, m));
3353 				len -= m->m_len;
3354 				uio->uio_resid -= m->m_len;
3355 				sbfree(sb, m);
3356 				n = m;
3357 			}
3358 			n->m_next = NULL;
3359 			sb->sb_mb = m;
3360 			sb->sb_lastrecord = sb->sb_mb;
3361 			if (sb->sb_mb == NULL)
3362 				SB_EMPTY_FIXUP(sb);
3363 		}
3364 		/* Copy the remainder. */
3365 		if (len > 0) {
3366 			KASSERT(sb->sb_mb != NULL,
3367 			    ("%s: len > 0 && sb->sb_mb empty", __func__));
3368 
3369 			m = m_copym(sb->sb_mb, 0, len, M_NOWAIT);
3370 			if (m == NULL)
3371 				len = 0;	/* Don't flush data from sockbuf. */
3372 			else
3373 				uio->uio_resid -= len;
3374 			if (*mp0 != NULL)
3375 				m_cat(*mp0, m);
3376 			else
3377 				*mp0 = m;
3378 			if (*mp0 == NULL) {
3379 				error = ENOBUFS;
3380 				goto out;
3381 			}
3382 		}
3383 	} else {
3384 		/* NB: Must unlock socket buffer as uiomove may sleep. */
3385 		SOCKBUF_UNLOCK(sb);
3386 		error = m_mbuftouio(uio, sb->sb_mb, len);
3387 		SOCKBUF_LOCK(sb);
3388 		if (error)
3389 			goto out;
3390 	}
3391 	SBLASTRECORDCHK(sb);
3392 	SBLASTMBUFCHK(sb);
3393 
3394 	/*
3395 	 * Remove the delivered data from the socket buffer unless we
3396 	 * were only peeking.
3397 	 */
3398 	if (!(flags & MSG_PEEK)) {
3399 		if (len > 0)
3400 			sbdrop_locked(sb, len);
3401 
3402 		/* Notify protocol that we drained some data. */
3403 		if ((so->so_proto->pr_flags & PR_WANTRCVD) &&
3404 		    (((flags & MSG_WAITALL) && uio->uio_resid > 0) ||
3405 		     !(flags & MSG_SOCALLBCK))) {
3406 			SOCKBUF_UNLOCK(sb);
3407 			VNET_SO_ASSERT(so);
3408 			so->so_proto->pr_rcvd(so, flags);
3409 			SOCKBUF_LOCK(sb);
3410 		}
3411 	}
3412 
3413 	/*
3414 	 * For MSG_WAITALL we may have to loop again and wait for
3415 	 * more data to come in.
3416 	 */
3417 	if ((flags & MSG_WAITALL) && uio->uio_resid > 0)
3418 		goto restart;
3419 out:
3420 	SBLASTRECORDCHK(sb);
3421 	SBLASTMBUFCHK(sb);
3422 	SOCKBUF_UNLOCK(sb);
3423 	return (error);
3424 }
3425 
3426 int
3427 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio,
3428     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
3429 {
3430 	struct sockbuf *sb;
3431 	int error, flags;
3432 
3433 	sb = &so->so_rcv;
3434 
3435 	/* We only do stream sockets. */
3436 	if (so->so_type != SOCK_STREAM)
3437 		return (EINVAL);
3438 	if (psa != NULL)
3439 		*psa = NULL;
3440 	if (flagsp != NULL)
3441 		flags = *flagsp & ~MSG_EOR;
3442 	else
3443 		flags = 0;
3444 	if (controlp != NULL)
3445 		*controlp = NULL;
3446 	if (flags & MSG_OOB)
3447 		return (soreceive_rcvoob(so, uio, flags));
3448 	if (mp0 != NULL)
3449 		*mp0 = NULL;
3450 
3451 #ifdef KERN_TLS
3452 	/*
3453 	 * KTLS store TLS records as records with a control message to
3454 	 * describe the framing.
3455 	 *
3456 	 * We check once here before acquiring locks to optimize the
3457 	 * common case.
3458 	 */
3459 	if (sb->sb_tls_info != NULL)
3460 		return (soreceive_generic(so, psa, uio, mp0, controlp,
3461 		    flagsp));
3462 #endif
3463 
3464 	/*
3465 	 * Prevent other threads from reading from the socket.  This lock may be
3466 	 * dropped in order to sleep waiting for data to arrive.
3467 	 */
3468 	error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags));
3469 	if (error)
3470 		return (error);
3471 #ifdef KERN_TLS
3472 	if (__predict_false(sb->sb_tls_info != NULL)) {
3473 		SOCK_IO_RECV_UNLOCK(so);
3474 		return (soreceive_generic(so, psa, uio, mp0, controlp,
3475 		    flagsp));
3476 	}
3477 #endif
3478 	error = soreceive_stream_locked(so, sb, psa, uio, mp0, controlp, flags);
3479 	SOCK_IO_RECV_UNLOCK(so);
3480 	return (error);
3481 }
3482 
3483 /*
3484  * Optimized version of soreceive() for simple datagram cases from userspace.
3485  * Unlike in the stream case, we're able to drop a datagram if copyout()
3486  * fails, and because we handle datagrams atomically, we don't need to use a
3487  * sleep lock to prevent I/O interlacing.
3488  */
3489 int
3490 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio,
3491     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
3492 {
3493 	struct mbuf *m, *m2;
3494 	int flags, error;
3495 	ssize_t len;
3496 	struct protosw *pr = so->so_proto;
3497 	struct mbuf *nextrecord;
3498 
3499 	if (psa != NULL)
3500 		*psa = NULL;
3501 	if (controlp != NULL)
3502 		*controlp = NULL;
3503 	if (flagsp != NULL)
3504 		flags = *flagsp &~ MSG_EOR;
3505 	else
3506 		flags = 0;
3507 
3508 	/*
3509 	 * For any complicated cases, fall back to the full
3510 	 * soreceive_generic().
3511 	 */
3512 	if (mp0 != NULL || (flags & (MSG_PEEK | MSG_OOB | MSG_TRUNC)))
3513 		return (soreceive_generic(so, psa, uio, mp0, controlp,
3514 		    flagsp));
3515 
3516 	/*
3517 	 * Enforce restrictions on use.
3518 	 */
3519 	KASSERT((pr->pr_flags & PR_WANTRCVD) == 0,
3520 	    ("soreceive_dgram: wantrcvd"));
3521 	KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic"));
3522 	KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0,
3523 	    ("soreceive_dgram: SBS_RCVATMARK"));
3524 	KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0,
3525 	    ("soreceive_dgram: P_CONNREQUIRED"));
3526 
3527 	/*
3528 	 * Loop blocking while waiting for a datagram.
3529 	 */
3530 	SOCKBUF_LOCK(&so->so_rcv);
3531 	while ((m = so->so_rcv.sb_mb) == NULL) {
3532 		KASSERT(sbavail(&so->so_rcv) == 0,
3533 		    ("soreceive_dgram: sb_mb NULL but sbavail %u",
3534 		    sbavail(&so->so_rcv)));
3535 		if (so->so_error) {
3536 			error = so->so_error;
3537 			so->so_error = 0;
3538 			SOCKBUF_UNLOCK(&so->so_rcv);
3539 			return (error);
3540 		}
3541 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE ||
3542 		    uio->uio_resid == 0) {
3543 			SOCKBUF_UNLOCK(&so->so_rcv);
3544 			return (0);
3545 		}
3546 		if ((so->so_state & SS_NBIO) ||
3547 		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
3548 			SOCKBUF_UNLOCK(&so->so_rcv);
3549 			return (EWOULDBLOCK);
3550 		}
3551 		SBLASTRECORDCHK(&so->so_rcv);
3552 		SBLASTMBUFCHK(&so->so_rcv);
3553 		error = sbwait(so, SO_RCV);
3554 		if (error) {
3555 			SOCKBUF_UNLOCK(&so->so_rcv);
3556 			return (error);
3557 		}
3558 	}
3559 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
3560 
3561 	if (uio->uio_td)
3562 		uio->uio_td->td_ru.ru_msgrcv++;
3563 	SBLASTRECORDCHK(&so->so_rcv);
3564 	SBLASTMBUFCHK(&so->so_rcv);
3565 	nextrecord = m->m_nextpkt;
3566 	if (nextrecord == NULL) {
3567 		KASSERT(so->so_rcv.sb_lastrecord == m,
3568 		    ("soreceive_dgram: lastrecord != m"));
3569 	}
3570 
3571 	KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord,
3572 	    ("soreceive_dgram: m_nextpkt != nextrecord"));
3573 
3574 	/*
3575 	 * Pull 'm' and its chain off the front of the packet queue.
3576 	 */
3577 	so->so_rcv.sb_mb = NULL;
3578 	sockbuf_pushsync(&so->so_rcv, nextrecord);
3579 
3580 	/*
3581 	 * Walk 'm's chain and free that many bytes from the socket buffer.
3582 	 */
3583 	for (m2 = m; m2 != NULL; m2 = m2->m_next)
3584 		sbfree(&so->so_rcv, m2);
3585 
3586 	/*
3587 	 * Do a few last checks before we let go of the lock.
3588 	 */
3589 	SBLASTRECORDCHK(&so->so_rcv);
3590 	SBLASTMBUFCHK(&so->so_rcv);
3591 	SOCKBUF_UNLOCK(&so->so_rcv);
3592 
3593 	if (pr->pr_flags & PR_ADDR) {
3594 		KASSERT(m->m_type == MT_SONAME,
3595 		    ("m->m_type == %d", m->m_type));
3596 		if (psa != NULL)
3597 			*psa = sodupsockaddr(mtod(m, struct sockaddr *),
3598 			    M_WAITOK);
3599 		m = m_free(m);
3600 	}
3601 	KASSERT(m, ("%s: no data or control after soname", __func__));
3602 
3603 	/*
3604 	 * Packet to copyout() is now in 'm' and it is disconnected from the
3605 	 * queue.
3606 	 *
3607 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
3608 	 * in the first mbuf chain on the socket buffer.  We call into the
3609 	 * protocol to perform externalization (or freeing if controlp ==
3610 	 * NULL). In some cases there can be only MT_CONTROL mbufs without
3611 	 * MT_DATA mbufs.
3612 	 */
3613 	if (m->m_type == MT_CONTROL) {
3614 		struct mbuf *cm = NULL, *cmn;
3615 		struct mbuf **cme = &cm;
3616 
3617 		do {
3618 			m2 = m->m_next;
3619 			m->m_next = NULL;
3620 			*cme = m;
3621 			cme = &(*cme)->m_next;
3622 			m = m2;
3623 		} while (m != NULL && m->m_type == MT_CONTROL);
3624 		while (cm != NULL) {
3625 			cmn = cm->m_next;
3626 			cm->m_next = NULL;
3627 			if (pr->pr_domain->dom_externalize != NULL) {
3628 				error = (*pr->pr_domain->dom_externalize)
3629 				    (cm, controlp, flags);
3630 			} else if (controlp != NULL)
3631 				*controlp = cm;
3632 			else
3633 				m_freem(cm);
3634 			if (controlp != NULL) {
3635 				while (*controlp != NULL)
3636 					controlp = &(*controlp)->m_next;
3637 			}
3638 			cm = cmn;
3639 		}
3640 	}
3641 	KASSERT(m == NULL || m->m_type == MT_DATA,
3642 	    ("soreceive_dgram: !data"));
3643 	while (m != NULL && uio->uio_resid > 0) {
3644 		len = uio->uio_resid;
3645 		if (len > m->m_len)
3646 			len = m->m_len;
3647 		error = uiomove(mtod(m, char *), (int)len, uio);
3648 		if (error) {
3649 			m_freem(m);
3650 			return (error);
3651 		}
3652 		if (len == m->m_len)
3653 			m = m_free(m);
3654 		else {
3655 			m->m_data += len;
3656 			m->m_len -= len;
3657 		}
3658 	}
3659 	if (m != NULL) {
3660 		flags |= MSG_TRUNC;
3661 		m_freem(m);
3662 	}
3663 	if (flagsp != NULL)
3664 		*flagsp |= flags;
3665 	return (0);
3666 }
3667 
3668 int
3669 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
3670     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
3671 {
3672 	int error;
3673 
3674 	CURVNET_SET(so->so_vnet);
3675 	error = so->so_proto->pr_soreceive(so, psa, uio, mp0, controlp, flagsp);
3676 	CURVNET_RESTORE();
3677 	return (error);
3678 }
3679 
3680 int
3681 soshutdown(struct socket *so, enum shutdown_how how)
3682 {
3683 	int error;
3684 
3685 	CURVNET_SET(so->so_vnet);
3686 	error = so->so_proto->pr_shutdown(so, how);
3687 	CURVNET_RESTORE();
3688 
3689 	return (error);
3690 }
3691 
3692 /*
3693  * Used by several pr_shutdown implementations that use generic socket buffers.
3694  */
3695 void
3696 sorflush(struct socket *so)
3697 {
3698 	int error;
3699 
3700 	VNET_SO_ASSERT(so);
3701 
3702 	/*
3703 	 * Dislodge threads currently blocked in receive and wait to acquire
3704 	 * a lock against other simultaneous readers before clearing the
3705 	 * socket buffer.  Don't let our acquire be interrupted by a signal
3706 	 * despite any existing socket disposition on interruptable waiting.
3707 	 *
3708 	 * The SOCK_IO_RECV_LOCK() is important here as there some pr_soreceive
3709 	 * methods that read the top of the socket buffer without acquisition
3710 	 * of the socket buffer mutex, assuming that top of the buffer
3711 	 * exclusively belongs to the read(2) syscall.  This is handy when
3712 	 * performing MSG_PEEK.
3713 	 */
3714 	socantrcvmore(so);
3715 
3716 	error = SOCK_IO_RECV_LOCK(so, SBL_WAIT | SBL_NOINTR);
3717 	if (error != 0) {
3718 		KASSERT(SOLISTENING(so),
3719 		    ("%s: soiolock(%p) failed", __func__, so));
3720 		return;
3721 	}
3722 
3723 	sbrelease(so, SO_RCV);
3724 	SOCK_IO_RECV_UNLOCK(so);
3725 
3726 }
3727 
3728 int
3729 sosetfib(struct socket *so, int fibnum)
3730 {
3731 	if (fibnum < 0 || fibnum >= rt_numfibs)
3732 		return (EINVAL);
3733 
3734 	SOCK_LOCK(so);
3735 	so->so_fibnum = fibnum;
3736 	SOCK_UNLOCK(so);
3737 
3738 	return (0);
3739 }
3740 
3741 #ifdef SOCKET_HHOOK
3742 /*
3743  * Wrapper for Socket established helper hook.
3744  * Parameters: socket, context of the hook point, hook id.
3745  */
3746 static inline int
3747 hhook_run_socket(struct socket *so, void *hctx, int32_t h_id)
3748 {
3749 	struct socket_hhook_data hhook_data = {
3750 		.so = so,
3751 		.hctx = hctx,
3752 		.m = NULL,
3753 		.status = 0
3754 	};
3755 
3756 	CURVNET_SET(so->so_vnet);
3757 	HHOOKS_RUN_IF(V_socket_hhh[h_id], &hhook_data, &so->osd);
3758 	CURVNET_RESTORE();
3759 
3760 	/* Ugly but needed, since hhooks return void for now */
3761 	return (hhook_data.status);
3762 }
3763 #endif
3764 
3765 /*
3766  * Perhaps this routine, and sooptcopyout(), below, ought to come in an
3767  * additional variant to handle the case where the option value needs to be
3768  * some kind of integer, but not a specific size.  In addition to their use
3769  * here, these functions are also called by the protocol-level pr_ctloutput()
3770  * routines.
3771  */
3772 int
3773 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
3774 {
3775 	size_t	valsize;
3776 
3777 	/*
3778 	 * If the user gives us more than we wanted, we ignore it, but if we
3779 	 * don't get the minimum length the caller wants, we return EINVAL.
3780 	 * On success, sopt->sopt_valsize is set to however much we actually
3781 	 * retrieved.
3782 	 */
3783 	if ((valsize = sopt->sopt_valsize) < minlen)
3784 		return EINVAL;
3785 	if (valsize > len)
3786 		sopt->sopt_valsize = valsize = len;
3787 
3788 	if (sopt->sopt_td != NULL)
3789 		return (copyin(sopt->sopt_val, buf, valsize));
3790 
3791 	bcopy(sopt->sopt_val, buf, valsize);
3792 	return (0);
3793 }
3794 
3795 /*
3796  * Kernel version of setsockopt(2).
3797  *
3798  * XXX: optlen is size_t, not socklen_t
3799  */
3800 int
3801 so_setsockopt(struct socket *so, int level, int optname, void *optval,
3802     size_t optlen)
3803 {
3804 	struct sockopt sopt;
3805 
3806 	sopt.sopt_level = level;
3807 	sopt.sopt_name = optname;
3808 	sopt.sopt_dir = SOPT_SET;
3809 	sopt.sopt_val = optval;
3810 	sopt.sopt_valsize = optlen;
3811 	sopt.sopt_td = NULL;
3812 	return (sosetopt(so, &sopt));
3813 }
3814 
3815 int
3816 sosetopt(struct socket *so, struct sockopt *sopt)
3817 {
3818 	int	error, optval;
3819 	struct	linger l;
3820 	struct	timeval tv;
3821 	sbintime_t val, *valp;
3822 	uint32_t val32;
3823 #ifdef MAC
3824 	struct mac extmac;
3825 #endif
3826 
3827 	CURVNET_SET(so->so_vnet);
3828 	error = 0;
3829 	if (sopt->sopt_level != SOL_SOCKET) {
3830 		error = (*so->so_proto->pr_ctloutput)(so, sopt);
3831 	} else {
3832 		switch (sopt->sopt_name) {
3833 		case SO_ACCEPTFILTER:
3834 			error = accept_filt_setopt(so, sopt);
3835 			if (error)
3836 				goto bad;
3837 			break;
3838 
3839 		case SO_LINGER:
3840 			error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
3841 			if (error)
3842 				goto bad;
3843 			if (l.l_linger < 0 ||
3844 			    l.l_linger > USHRT_MAX ||
3845 			    l.l_linger > (INT_MAX / hz)) {
3846 				error = EDOM;
3847 				goto bad;
3848 			}
3849 			SOCK_LOCK(so);
3850 			so->so_linger = l.l_linger;
3851 			if (l.l_onoff)
3852 				so->so_options |= SO_LINGER;
3853 			else
3854 				so->so_options &= ~SO_LINGER;
3855 			SOCK_UNLOCK(so);
3856 			break;
3857 
3858 		case SO_DEBUG:
3859 		case SO_KEEPALIVE:
3860 		case SO_DONTROUTE:
3861 		case SO_USELOOPBACK:
3862 		case SO_BROADCAST:
3863 		case SO_REUSEADDR:
3864 		case SO_REUSEPORT:
3865 		case SO_REUSEPORT_LB:
3866 		case SO_OOBINLINE:
3867 		case SO_TIMESTAMP:
3868 		case SO_BINTIME:
3869 		case SO_NOSIGPIPE:
3870 		case SO_NO_DDP:
3871 		case SO_NO_OFFLOAD:
3872 		case SO_RERROR:
3873 			error = sooptcopyin(sopt, &optval, sizeof optval,
3874 			    sizeof optval);
3875 			if (error)
3876 				goto bad;
3877 			SOCK_LOCK(so);
3878 			if (optval)
3879 				so->so_options |= sopt->sopt_name;
3880 			else
3881 				so->so_options &= ~sopt->sopt_name;
3882 			SOCK_UNLOCK(so);
3883 			break;
3884 
3885 		case SO_SETFIB:
3886 			error = so->so_proto->pr_ctloutput(so, sopt);
3887 			break;
3888 
3889 		case SO_USER_COOKIE:
3890 			error = sooptcopyin(sopt, &val32, sizeof val32,
3891 			    sizeof val32);
3892 			if (error)
3893 				goto bad;
3894 			so->so_user_cookie = val32;
3895 			break;
3896 
3897 		case SO_SNDBUF:
3898 		case SO_RCVBUF:
3899 		case SO_SNDLOWAT:
3900 		case SO_RCVLOWAT:
3901 			error = so->so_proto->pr_setsbopt(so, sopt);
3902 			if (error)
3903 				goto bad;
3904 			break;
3905 
3906 		case SO_SNDTIMEO:
3907 		case SO_RCVTIMEO:
3908 #ifdef COMPAT_FREEBSD32
3909 			if (SV_CURPROC_FLAG(SV_ILP32)) {
3910 				struct timeval32 tv32;
3911 
3912 				error = sooptcopyin(sopt, &tv32, sizeof tv32,
3913 				    sizeof tv32);
3914 				CP(tv32, tv, tv_sec);
3915 				CP(tv32, tv, tv_usec);
3916 			} else
3917 #endif
3918 				error = sooptcopyin(sopt, &tv, sizeof tv,
3919 				    sizeof tv);
3920 			if (error)
3921 				goto bad;
3922 			if (tv.tv_sec < 0 || tv.tv_usec < 0 ||
3923 			    tv.tv_usec >= 1000000) {
3924 				error = EDOM;
3925 				goto bad;
3926 			}
3927 			if (tv.tv_sec > INT32_MAX)
3928 				val = SBT_MAX;
3929 			else
3930 				val = tvtosbt(tv);
3931 			SOCK_LOCK(so);
3932 			valp = sopt->sopt_name == SO_SNDTIMEO ?
3933 			    (SOLISTENING(so) ? &so->sol_sbsnd_timeo :
3934 			    &so->so_snd.sb_timeo) :
3935 			    (SOLISTENING(so) ? &so->sol_sbrcv_timeo :
3936 			    &so->so_rcv.sb_timeo);
3937 			*valp = val;
3938 			SOCK_UNLOCK(so);
3939 			break;
3940 
3941 		case SO_LABEL:
3942 #ifdef MAC
3943 			error = sooptcopyin(sopt, &extmac, sizeof extmac,
3944 			    sizeof extmac);
3945 			if (error)
3946 				goto bad;
3947 			error = mac_setsockopt_label(sopt->sopt_td->td_ucred,
3948 			    so, &extmac);
3949 #else
3950 			error = EOPNOTSUPP;
3951 #endif
3952 			break;
3953 
3954 		case SO_TS_CLOCK:
3955 			error = sooptcopyin(sopt, &optval, sizeof optval,
3956 			    sizeof optval);
3957 			if (error)
3958 				goto bad;
3959 			if (optval < 0 || optval > SO_TS_CLOCK_MAX) {
3960 				error = EINVAL;
3961 				goto bad;
3962 			}
3963 			so->so_ts_clock = optval;
3964 			break;
3965 
3966 		case SO_MAX_PACING_RATE:
3967 			error = sooptcopyin(sopt, &val32, sizeof(val32),
3968 			    sizeof(val32));
3969 			if (error)
3970 				goto bad;
3971 			so->so_max_pacing_rate = val32;
3972 			break;
3973 
3974 		case SO_SPLICE: {
3975 			struct splice splice;
3976 
3977 #ifdef COMPAT_FREEBSD32
3978 			if (SV_CURPROC_FLAG(SV_ILP32)) {
3979 				struct splice32 splice32;
3980 
3981 				error = sooptcopyin(sopt, &splice32,
3982 				    sizeof(splice32), sizeof(splice32));
3983 				if (error == 0) {
3984 					splice.sp_fd = splice32.sp_fd;
3985 					splice.sp_max = splice32.sp_max;
3986 					CP(splice32.sp_idle, splice.sp_idle,
3987 					    tv_sec);
3988 					CP(splice32.sp_idle, splice.sp_idle,
3989 					    tv_usec);
3990 				}
3991 			} else
3992 #endif
3993 			{
3994 				error = sooptcopyin(sopt, &splice,
3995 				    sizeof(splice), sizeof(splice));
3996 			}
3997 			if (error)
3998 				goto bad;
3999 #ifdef KTRACE
4000 			if (KTRPOINT(curthread, KTR_STRUCT))
4001 				ktrsplice(&splice);
4002 #endif
4003 
4004 			error = splice_init();
4005 			if (error != 0)
4006 				goto bad;
4007 
4008 			if (splice.sp_fd >= 0) {
4009 				struct file *fp;
4010 				struct socket *so2;
4011 
4012 				if (!cap_rights_contains(sopt->sopt_rights,
4013 				    &cap_recv_rights)) {
4014 					error = ENOTCAPABLE;
4015 					goto bad;
4016 				}
4017 				error = getsock(sopt->sopt_td, splice.sp_fd,
4018 				    &cap_send_rights, &fp);
4019 				if (error != 0)
4020 					goto bad;
4021 				so2 = fp->f_data;
4022 
4023 				error = so_splice(so, so2, &splice);
4024 				fdrop(fp, sopt->sopt_td);
4025 			} else {
4026 				error = so_unsplice(so, false);
4027 			}
4028 			break;
4029 		}
4030 		default:
4031 #ifdef SOCKET_HHOOK
4032 			if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
4033 				error = hhook_run_socket(so, sopt,
4034 				    HHOOK_SOCKET_OPT);
4035 			else
4036 #endif
4037 				error = ENOPROTOOPT;
4038 			break;
4039 		}
4040 		if (error == 0)
4041 			(void)(*so->so_proto->pr_ctloutput)(so, sopt);
4042 	}
4043 bad:
4044 	CURVNET_RESTORE();
4045 	return (error);
4046 }
4047 
4048 /*
4049  * Helper routine for getsockopt.
4050  */
4051 int
4052 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
4053 {
4054 	int	error;
4055 	size_t	valsize;
4056 
4057 	error = 0;
4058 
4059 	/*
4060 	 * Documented get behavior is that we always return a value, possibly
4061 	 * truncated to fit in the user's buffer.  Traditional behavior is
4062 	 * that we always tell the user precisely how much we copied, rather
4063 	 * than something useful like the total amount we had available for
4064 	 * her.  Note that this interface is not idempotent; the entire
4065 	 * answer must be generated ahead of time.
4066 	 */
4067 	valsize = min(len, sopt->sopt_valsize);
4068 	sopt->sopt_valsize = valsize;
4069 	if (sopt->sopt_val != NULL) {
4070 		if (sopt->sopt_td != NULL)
4071 			error = copyout(buf, sopt->sopt_val, valsize);
4072 		else
4073 			bcopy(buf, sopt->sopt_val, valsize);
4074 	}
4075 	return (error);
4076 }
4077 
4078 int
4079 sogetopt(struct socket *so, struct sockopt *sopt)
4080 {
4081 	int	error, optval;
4082 	struct	linger l;
4083 	struct	timeval tv;
4084 #ifdef MAC
4085 	struct mac extmac;
4086 #endif
4087 
4088 	CURVNET_SET(so->so_vnet);
4089 	error = 0;
4090 	if (sopt->sopt_level != SOL_SOCKET) {
4091 		error = (*so->so_proto->pr_ctloutput)(so, sopt);
4092 		CURVNET_RESTORE();
4093 		return (error);
4094 	} else {
4095 		switch (sopt->sopt_name) {
4096 		case SO_ACCEPTFILTER:
4097 			error = accept_filt_getopt(so, sopt);
4098 			break;
4099 
4100 		case SO_LINGER:
4101 			SOCK_LOCK(so);
4102 			l.l_onoff = so->so_options & SO_LINGER;
4103 			l.l_linger = so->so_linger;
4104 			SOCK_UNLOCK(so);
4105 			error = sooptcopyout(sopt, &l, sizeof l);
4106 			break;
4107 
4108 		case SO_USELOOPBACK:
4109 		case SO_DONTROUTE:
4110 		case SO_DEBUG:
4111 		case SO_KEEPALIVE:
4112 		case SO_REUSEADDR:
4113 		case SO_REUSEPORT:
4114 		case SO_REUSEPORT_LB:
4115 		case SO_BROADCAST:
4116 		case SO_OOBINLINE:
4117 		case SO_ACCEPTCONN:
4118 		case SO_TIMESTAMP:
4119 		case SO_BINTIME:
4120 		case SO_NOSIGPIPE:
4121 		case SO_NO_DDP:
4122 		case SO_NO_OFFLOAD:
4123 		case SO_RERROR:
4124 			optval = so->so_options & sopt->sopt_name;
4125 integer:
4126 			error = sooptcopyout(sopt, &optval, sizeof optval);
4127 			break;
4128 
4129 		case SO_FIB:
4130 			SOCK_LOCK(so);
4131 			optval = so->so_fibnum;
4132 			SOCK_UNLOCK(so);
4133 			goto integer;
4134 
4135 		case SO_DOMAIN:
4136 			optval = so->so_proto->pr_domain->dom_family;
4137 			goto integer;
4138 
4139 		case SO_TYPE:
4140 			optval = so->so_type;
4141 			goto integer;
4142 
4143 		case SO_PROTOCOL:
4144 			optval = so->so_proto->pr_protocol;
4145 			goto integer;
4146 
4147 		case SO_ERROR:
4148 			SOCK_LOCK(so);
4149 			if (so->so_error) {
4150 				optval = so->so_error;
4151 				so->so_error = 0;
4152 			} else {
4153 				optval = so->so_rerror;
4154 				so->so_rerror = 0;
4155 			}
4156 			SOCK_UNLOCK(so);
4157 			goto integer;
4158 
4159 		case SO_SNDBUF:
4160 			SOCK_LOCK(so);
4161 			optval = SOLISTENING(so) ? so->sol_sbsnd_hiwat :
4162 			    so->so_snd.sb_hiwat;
4163 			SOCK_UNLOCK(so);
4164 			goto integer;
4165 
4166 		case SO_RCVBUF:
4167 			SOCK_LOCK(so);
4168 			optval = SOLISTENING(so) ? so->sol_sbrcv_hiwat :
4169 			    so->so_rcv.sb_hiwat;
4170 			SOCK_UNLOCK(so);
4171 			goto integer;
4172 
4173 		case SO_SNDLOWAT:
4174 			SOCK_LOCK(so);
4175 			optval = SOLISTENING(so) ? so->sol_sbsnd_lowat :
4176 			    so->so_snd.sb_lowat;
4177 			SOCK_UNLOCK(so);
4178 			goto integer;
4179 
4180 		case SO_RCVLOWAT:
4181 			SOCK_LOCK(so);
4182 			optval = SOLISTENING(so) ? so->sol_sbrcv_lowat :
4183 			    so->so_rcv.sb_lowat;
4184 			SOCK_UNLOCK(so);
4185 			goto integer;
4186 
4187 		case SO_SNDTIMEO:
4188 		case SO_RCVTIMEO:
4189 			SOCK_LOCK(so);
4190 			tv = sbttotv(sopt->sopt_name == SO_SNDTIMEO ?
4191 			    (SOLISTENING(so) ? so->sol_sbsnd_timeo :
4192 			    so->so_snd.sb_timeo) :
4193 			    (SOLISTENING(so) ? so->sol_sbrcv_timeo :
4194 			    so->so_rcv.sb_timeo));
4195 			SOCK_UNLOCK(so);
4196 #ifdef COMPAT_FREEBSD32
4197 			if (SV_CURPROC_FLAG(SV_ILP32)) {
4198 				struct timeval32 tv32;
4199 
4200 				CP(tv, tv32, tv_sec);
4201 				CP(tv, tv32, tv_usec);
4202 				error = sooptcopyout(sopt, &tv32, sizeof tv32);
4203 			} else
4204 #endif
4205 				error = sooptcopyout(sopt, &tv, sizeof tv);
4206 			break;
4207 
4208 		case SO_LABEL:
4209 #ifdef MAC
4210 			error = sooptcopyin(sopt, &extmac, sizeof(extmac),
4211 			    sizeof(extmac));
4212 			if (error)
4213 				goto bad;
4214 			error = mac_getsockopt_label(sopt->sopt_td->td_ucred,
4215 			    so, &extmac);
4216 			if (error)
4217 				goto bad;
4218 			/* Don't copy out extmac, it is unchanged. */
4219 #else
4220 			error = EOPNOTSUPP;
4221 #endif
4222 			break;
4223 
4224 		case SO_PEERLABEL:
4225 #ifdef MAC
4226 			error = sooptcopyin(sopt, &extmac, sizeof(extmac),
4227 			    sizeof(extmac));
4228 			if (error)
4229 				goto bad;
4230 			error = mac_getsockopt_peerlabel(
4231 			    sopt->sopt_td->td_ucred, so, &extmac);
4232 			if (error)
4233 				goto bad;
4234 			/* Don't copy out extmac, it is unchanged. */
4235 #else
4236 			error = EOPNOTSUPP;
4237 #endif
4238 			break;
4239 
4240 		case SO_LISTENQLIMIT:
4241 			SOCK_LOCK(so);
4242 			optval = SOLISTENING(so) ? so->sol_qlimit : 0;
4243 			SOCK_UNLOCK(so);
4244 			goto integer;
4245 
4246 		case SO_LISTENQLEN:
4247 			SOCK_LOCK(so);
4248 			optval = SOLISTENING(so) ? so->sol_qlen : 0;
4249 			SOCK_UNLOCK(so);
4250 			goto integer;
4251 
4252 		case SO_LISTENINCQLEN:
4253 			SOCK_LOCK(so);
4254 			optval = SOLISTENING(so) ? so->sol_incqlen : 0;
4255 			SOCK_UNLOCK(so);
4256 			goto integer;
4257 
4258 		case SO_TS_CLOCK:
4259 			optval = so->so_ts_clock;
4260 			goto integer;
4261 
4262 		case SO_MAX_PACING_RATE:
4263 			optval = so->so_max_pacing_rate;
4264 			goto integer;
4265 
4266 		case SO_SPLICE: {
4267 			off_t n;
4268 
4269 			/*
4270 			 * Acquire the I/O lock to serialize with
4271 			 * so_splice_xfer().  This is not required for
4272 			 * correctness, but makes testing simpler: once a byte
4273 			 * has been transmitted to the sink and observed (e.g.,
4274 			 * by reading from the socket to which the sink is
4275 			 * connected), a subsequent getsockopt(SO_SPLICE) will
4276 			 * return an up-to-date value.
4277 			 */
4278 			error = SOCK_IO_RECV_LOCK(so, SBL_WAIT);
4279 			if (error != 0)
4280 				goto bad;
4281 			SOCK_LOCK(so);
4282 			if (SOLISTENING(so)) {
4283 				n = 0;
4284 			} else {
4285 				n = so->so_splice_sent;
4286 			}
4287 			SOCK_UNLOCK(so);
4288 			SOCK_IO_RECV_UNLOCK(so);
4289 			error = sooptcopyout(sopt, &n, sizeof(n));
4290 			break;
4291 		}
4292 
4293 		default:
4294 #ifdef SOCKET_HHOOK
4295 			if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
4296 				error = hhook_run_socket(so, sopt,
4297 				    HHOOK_SOCKET_OPT);
4298 			else
4299 #endif
4300 				error = ENOPROTOOPT;
4301 			break;
4302 		}
4303 	}
4304 bad:
4305 	CURVNET_RESTORE();
4306 	return (error);
4307 }
4308 
4309 int
4310 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
4311 {
4312 	struct mbuf *m, *m_prev;
4313 	int sopt_size = sopt->sopt_valsize;
4314 
4315 	MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
4316 	if (m == NULL)
4317 		return ENOBUFS;
4318 	if (sopt_size > MLEN) {
4319 		MCLGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT);
4320 		if ((m->m_flags & M_EXT) == 0) {
4321 			m_free(m);
4322 			return ENOBUFS;
4323 		}
4324 		m->m_len = min(MCLBYTES, sopt_size);
4325 	} else {
4326 		m->m_len = min(MLEN, sopt_size);
4327 	}
4328 	sopt_size -= m->m_len;
4329 	*mp = m;
4330 	m_prev = m;
4331 
4332 	while (sopt_size) {
4333 		MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
4334 		if (m == NULL) {
4335 			m_freem(*mp);
4336 			return ENOBUFS;
4337 		}
4338 		if (sopt_size > MLEN) {
4339 			MCLGET(m, sopt->sopt_td != NULL ? M_WAITOK :
4340 			    M_NOWAIT);
4341 			if ((m->m_flags & M_EXT) == 0) {
4342 				m_freem(m);
4343 				m_freem(*mp);
4344 				return ENOBUFS;
4345 			}
4346 			m->m_len = min(MCLBYTES, sopt_size);
4347 		} else {
4348 			m->m_len = min(MLEN, sopt_size);
4349 		}
4350 		sopt_size -= m->m_len;
4351 		m_prev->m_next = m;
4352 		m_prev = m;
4353 	}
4354 	return (0);
4355 }
4356 
4357 int
4358 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
4359 {
4360 	struct mbuf *m0 = m;
4361 
4362 	if (sopt->sopt_val == NULL)
4363 		return (0);
4364 	while (m != NULL && sopt->sopt_valsize >= m->m_len) {
4365 		if (sopt->sopt_td != NULL) {
4366 			int error;
4367 
4368 			error = copyin(sopt->sopt_val, mtod(m, char *),
4369 			    m->m_len);
4370 			if (error != 0) {
4371 				m_freem(m0);
4372 				return(error);
4373 			}
4374 		} else
4375 			bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
4376 		sopt->sopt_valsize -= m->m_len;
4377 		sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
4378 		m = m->m_next;
4379 	}
4380 	if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
4381 		panic("ip6_sooptmcopyin");
4382 	return (0);
4383 }
4384 
4385 int
4386 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
4387 {
4388 	struct mbuf *m0 = m;
4389 	size_t valsize = 0;
4390 
4391 	if (sopt->sopt_val == NULL)
4392 		return (0);
4393 	while (m != NULL && sopt->sopt_valsize >= m->m_len) {
4394 		if (sopt->sopt_td != NULL) {
4395 			int error;
4396 
4397 			error = copyout(mtod(m, char *), sopt->sopt_val,
4398 			    m->m_len);
4399 			if (error != 0) {
4400 				m_freem(m0);
4401 				return(error);
4402 			}
4403 		} else
4404 			bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
4405 		sopt->sopt_valsize -= m->m_len;
4406 		sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
4407 		valsize += m->m_len;
4408 		m = m->m_next;
4409 	}
4410 	if (m != NULL) {
4411 		/* enough soopt buffer should be given from user-land */
4412 		m_freem(m0);
4413 		return(EINVAL);
4414 	}
4415 	sopt->sopt_valsize = valsize;
4416 	return (0);
4417 }
4418 
4419 /*
4420  * sohasoutofband(): protocol notifies socket layer of the arrival of new
4421  * out-of-band data, which will then notify socket consumers.
4422  */
4423 void
4424 sohasoutofband(struct socket *so)
4425 {
4426 
4427 	if (so->so_sigio != NULL)
4428 		pgsigio(&so->so_sigio, SIGURG, 0);
4429 	selwakeuppri(&so->so_rdsel, PSOCK);
4430 }
4431 
4432 int
4433 sopoll_generic(struct socket *so, int events, struct thread *td)
4434 {
4435 	int revents;
4436 
4437 	SOCK_LOCK(so);
4438 	if (SOLISTENING(so)) {
4439 		if (!(events & (POLLIN | POLLRDNORM)))
4440 			revents = 0;
4441 		else if (!TAILQ_EMPTY(&so->sol_comp))
4442 			revents = events & (POLLIN | POLLRDNORM);
4443 		else if ((events & POLLINIGNEOF) == 0 && so->so_error)
4444 			revents = (events & (POLLIN | POLLRDNORM)) | POLLHUP;
4445 		else {
4446 			selrecord(td, &so->so_rdsel);
4447 			revents = 0;
4448 		}
4449 	} else {
4450 		revents = 0;
4451 		SOCK_SENDBUF_LOCK(so);
4452 		SOCK_RECVBUF_LOCK(so);
4453 		if (events & (POLLIN | POLLRDNORM))
4454 			if (soreadabledata(so) && !isspliced(so))
4455 				revents |= events & (POLLIN | POLLRDNORM);
4456 		if (events & (POLLOUT | POLLWRNORM))
4457 			if (sowriteable(so) && !issplicedback(so))
4458 				revents |= events & (POLLOUT | POLLWRNORM);
4459 		if (events & (POLLPRI | POLLRDBAND))
4460 			if (so->so_oobmark ||
4461 			    (so->so_rcv.sb_state & SBS_RCVATMARK))
4462 				revents |= events & (POLLPRI | POLLRDBAND);
4463 		if ((events & POLLINIGNEOF) == 0) {
4464 			if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
4465 				revents |= events & (POLLIN | POLLRDNORM);
4466 				if (so->so_snd.sb_state & SBS_CANTSENDMORE)
4467 					revents |= POLLHUP;
4468 			}
4469 		}
4470 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
4471 			revents |= events & POLLRDHUP;
4472 		if (revents == 0) {
4473 			if (events &
4474 			    (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND | POLLRDHUP)) {
4475 				selrecord(td, &so->so_rdsel);
4476 				so->so_rcv.sb_flags |= SB_SEL;
4477 			}
4478 			if (events & (POLLOUT | POLLWRNORM)) {
4479 				selrecord(td, &so->so_wrsel);
4480 				so->so_snd.sb_flags |= SB_SEL;
4481 			}
4482 		}
4483 		SOCK_RECVBUF_UNLOCK(so);
4484 		SOCK_SENDBUF_UNLOCK(so);
4485 	}
4486 	SOCK_UNLOCK(so);
4487 	return (revents);
4488 }
4489 
4490 int
4491 soo_kqfilter(struct file *fp, struct knote *kn)
4492 {
4493 	struct socket *so = kn->kn_fp->f_data;
4494 	struct sockbuf *sb;
4495 	sb_which which;
4496 	struct knlist *knl;
4497 
4498 	switch (kn->kn_filter) {
4499 	case EVFILT_READ:
4500 		kn->kn_fop = &soread_filtops;
4501 		knl = &so->so_rdsel.si_note;
4502 		sb = &so->so_rcv;
4503 		which = SO_RCV;
4504 		break;
4505 	case EVFILT_WRITE:
4506 		kn->kn_fop = &sowrite_filtops;
4507 		knl = &so->so_wrsel.si_note;
4508 		sb = &so->so_snd;
4509 		which = SO_SND;
4510 		break;
4511 	case EVFILT_EMPTY:
4512 		kn->kn_fop = &soempty_filtops;
4513 		knl = &so->so_wrsel.si_note;
4514 		sb = &so->so_snd;
4515 		which = SO_SND;
4516 		break;
4517 	default:
4518 		return (EINVAL);
4519 	}
4520 
4521 	SOCK_LOCK(so);
4522 	if (SOLISTENING(so)) {
4523 		knlist_add(knl, kn, 1);
4524 	} else {
4525 		SOCK_BUF_LOCK(so, which);
4526 		knlist_add(knl, kn, 1);
4527 		sb->sb_flags |= SB_KNOTE;
4528 		SOCK_BUF_UNLOCK(so, which);
4529 	}
4530 	SOCK_UNLOCK(so);
4531 	return (0);
4532 }
4533 
4534 static void
4535 filt_sordetach(struct knote *kn)
4536 {
4537 	struct socket *so = kn->kn_fp->f_data;
4538 
4539 	so_rdknl_lock(so);
4540 	knlist_remove(&so->so_rdsel.si_note, kn, 1);
4541 	if (!SOLISTENING(so) && knlist_empty(&so->so_rdsel.si_note))
4542 		so->so_rcv.sb_flags &= ~SB_KNOTE;
4543 	so_rdknl_unlock(so);
4544 }
4545 
4546 /*ARGSUSED*/
4547 static int
4548 filt_soread(struct knote *kn, long hint)
4549 {
4550 	struct socket *so;
4551 
4552 	so = kn->kn_fp->f_data;
4553 
4554 	if (SOLISTENING(so)) {
4555 		SOCK_LOCK_ASSERT(so);
4556 		kn->kn_data = so->sol_qlen;
4557 		if (so->so_error) {
4558 			kn->kn_flags |= EV_EOF;
4559 			kn->kn_fflags = so->so_error;
4560 			return (1);
4561 		}
4562 		return (!TAILQ_EMPTY(&so->sol_comp));
4563 	}
4564 
4565 	if ((so->so_rcv.sb_flags & SB_SPLICED) != 0)
4566 		return (0);
4567 
4568 	SOCK_RECVBUF_LOCK_ASSERT(so);
4569 
4570 	kn->kn_data = sbavail(&so->so_rcv) - so->so_rcv.sb_ctl;
4571 	if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
4572 		kn->kn_flags |= EV_EOF;
4573 		kn->kn_fflags = so->so_error;
4574 		return (1);
4575 	} else if (so->so_error || so->so_rerror)
4576 		return (1);
4577 
4578 	if (kn->kn_sfflags & NOTE_LOWAT) {
4579 		if (kn->kn_data >= kn->kn_sdata)
4580 			return (1);
4581 	} else if (sbavail(&so->so_rcv) >= so->so_rcv.sb_lowat)
4582 		return (1);
4583 
4584 #ifdef SOCKET_HHOOK
4585 	/* This hook returning non-zero indicates an event, not error */
4586 	return (hhook_run_socket(so, NULL, HHOOK_FILT_SOREAD));
4587 #else
4588 	return (0);
4589 #endif
4590 }
4591 
4592 static void
4593 filt_sowdetach(struct knote *kn)
4594 {
4595 	struct socket *so = kn->kn_fp->f_data;
4596 
4597 	so_wrknl_lock(so);
4598 	knlist_remove(&so->so_wrsel.si_note, kn, 1);
4599 	if (!SOLISTENING(so) && knlist_empty(&so->so_wrsel.si_note))
4600 		so->so_snd.sb_flags &= ~SB_KNOTE;
4601 	so_wrknl_unlock(so);
4602 }
4603 
4604 /*ARGSUSED*/
4605 static int
4606 filt_sowrite(struct knote *kn, long hint)
4607 {
4608 	struct socket *so;
4609 
4610 	so = kn->kn_fp->f_data;
4611 
4612 	if (SOLISTENING(so))
4613 		return (0);
4614 
4615 	SOCK_SENDBUF_LOCK_ASSERT(so);
4616 	kn->kn_data = sbspace(&so->so_snd);
4617 
4618 #ifdef SOCKET_HHOOK
4619 	hhook_run_socket(so, kn, HHOOK_FILT_SOWRITE);
4620 #endif
4621 
4622 	if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
4623 		kn->kn_flags |= EV_EOF;
4624 		kn->kn_fflags = so->so_error;
4625 		return (1);
4626 	} else if (so->so_error)	/* temporary udp error */
4627 		return (1);
4628 	else if (((so->so_state & SS_ISCONNECTED) == 0) &&
4629 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
4630 		return (0);
4631 	else if (kn->kn_sfflags & NOTE_LOWAT)
4632 		return (kn->kn_data >= kn->kn_sdata);
4633 	else
4634 		return (kn->kn_data >= so->so_snd.sb_lowat);
4635 }
4636 
4637 static int
4638 filt_soempty(struct knote *kn, long hint)
4639 {
4640 	struct socket *so;
4641 
4642 	so = kn->kn_fp->f_data;
4643 
4644 	if (SOLISTENING(so))
4645 		return (1);
4646 
4647 	SOCK_SENDBUF_LOCK_ASSERT(so);
4648 	kn->kn_data = sbused(&so->so_snd);
4649 
4650 	if (kn->kn_data == 0)
4651 		return (1);
4652 	else
4653 		return (0);
4654 }
4655 
4656 int
4657 socheckuid(struct socket *so, uid_t uid)
4658 {
4659 
4660 	if (so == NULL)
4661 		return (EPERM);
4662 	if (so->so_cred->cr_uid != uid)
4663 		return (EPERM);
4664 	return (0);
4665 }
4666 
4667 /*
4668  * These functions are used by protocols to notify the socket layer (and its
4669  * consumers) of state changes in the sockets driven by protocol-side events.
4670  */
4671 
4672 /*
4673  * Procedures to manipulate state flags of socket and do appropriate wakeups.
4674  *
4675  * Normal sequence from the active (originating) side is that
4676  * soisconnecting() is called during processing of connect() call, resulting
4677  * in an eventual call to soisconnected() if/when the connection is
4678  * established.  When the connection is torn down soisdisconnecting() is
4679  * called during processing of disconnect() call, and soisdisconnected() is
4680  * called when the connection to the peer is totally severed.  The semantics
4681  * of these routines are such that connectionless protocols can call
4682  * soisconnected() and soisdisconnected() only, bypassing the in-progress
4683  * calls when setting up a ``connection'' takes no time.
4684  *
4685  * From the passive side, a socket is created with two queues of sockets:
4686  * so_incomp for connections in progress and so_comp for connections already
4687  * made and awaiting user acceptance.  As a protocol is preparing incoming
4688  * connections, it creates a socket structure queued on so_incomp by calling
4689  * sonewconn().  When the connection is established, soisconnected() is
4690  * called, and transfers the socket structure to so_comp, making it available
4691  * to accept().
4692  *
4693  * If a socket is closed with sockets on either so_incomp or so_comp, these
4694  * sockets are dropped.
4695  *
4696  * If higher-level protocols are implemented in the kernel, the wakeups done
4697  * here will sometimes cause software-interrupt process scheduling.
4698  */
4699 void
4700 soisconnecting(struct socket *so)
4701 {
4702 
4703 	SOCK_LOCK(so);
4704 	so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
4705 	so->so_state |= SS_ISCONNECTING;
4706 	SOCK_UNLOCK(so);
4707 }
4708 
4709 void
4710 soisconnected(struct socket *so)
4711 {
4712 	bool last __diagused;
4713 
4714 	SOCK_LOCK(so);
4715 	so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING);
4716 	so->so_state |= SS_ISCONNECTED;
4717 
4718 	if (so->so_qstate == SQ_INCOMP) {
4719 		struct socket *head = so->so_listen;
4720 		int ret;
4721 
4722 		KASSERT(head, ("%s: so %p on incomp of NULL", __func__, so));
4723 		/*
4724 		 * Promoting a socket from incomplete queue to complete, we
4725 		 * need to go through reverse order of locking.  We first do
4726 		 * trylock, and if that doesn't succeed, we go the hard way
4727 		 * leaving a reference and rechecking consistency after proper
4728 		 * locking.
4729 		 */
4730 		if (__predict_false(SOLISTEN_TRYLOCK(head) == 0)) {
4731 			soref(head);
4732 			SOCK_UNLOCK(so);
4733 			SOLISTEN_LOCK(head);
4734 			SOCK_LOCK(so);
4735 			if (__predict_false(head != so->so_listen)) {
4736 				/*
4737 				 * The socket went off the listen queue,
4738 				 * should be lost race to close(2) of sol.
4739 				 * The socket is about to soabort().
4740 				 */
4741 				SOCK_UNLOCK(so);
4742 				sorele_locked(head);
4743 				return;
4744 			}
4745 			last = refcount_release(&head->so_count);
4746 			KASSERT(!last, ("%s: released last reference for %p",
4747 			    __func__, head));
4748 		}
4749 again:
4750 		if ((so->so_options & SO_ACCEPTFILTER) == 0) {
4751 			TAILQ_REMOVE(&head->sol_incomp, so, so_list);
4752 			head->sol_incqlen--;
4753 			TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list);
4754 			head->sol_qlen++;
4755 			so->so_qstate = SQ_COMP;
4756 			SOCK_UNLOCK(so);
4757 			solisten_wakeup(head);	/* unlocks */
4758 		} else {
4759 			SOCK_RECVBUF_LOCK(so);
4760 			soupcall_set(so, SO_RCV,
4761 			    head->sol_accept_filter->accf_callback,
4762 			    head->sol_accept_filter_arg);
4763 			so->so_options &= ~SO_ACCEPTFILTER;
4764 			ret = head->sol_accept_filter->accf_callback(so,
4765 			    head->sol_accept_filter_arg, M_NOWAIT);
4766 			if (ret == SU_ISCONNECTED) {
4767 				soupcall_clear(so, SO_RCV);
4768 				SOCK_RECVBUF_UNLOCK(so);
4769 				goto again;
4770 			}
4771 			SOCK_RECVBUF_UNLOCK(so);
4772 			SOCK_UNLOCK(so);
4773 			SOLISTEN_UNLOCK(head);
4774 		}
4775 		return;
4776 	}
4777 	SOCK_UNLOCK(so);
4778 	wakeup(&so->so_timeo);
4779 	sorwakeup(so);
4780 	sowwakeup(so);
4781 }
4782 
4783 void
4784 soisdisconnecting(struct socket *so)
4785 {
4786 
4787 	SOCK_LOCK(so);
4788 	so->so_state &= ~SS_ISCONNECTING;
4789 	so->so_state |= SS_ISDISCONNECTING;
4790 
4791 	if (!SOLISTENING(so)) {
4792 		SOCK_RECVBUF_LOCK(so);
4793 		socantrcvmore_locked(so);
4794 		SOCK_SENDBUF_LOCK(so);
4795 		socantsendmore_locked(so);
4796 	}
4797 	SOCK_UNLOCK(so);
4798 	wakeup(&so->so_timeo);
4799 }
4800 
4801 void
4802 soisdisconnected(struct socket *so)
4803 {
4804 
4805 	SOCK_LOCK(so);
4806 
4807 	/*
4808 	 * There is at least one reader of so_state that does not
4809 	 * acquire socket lock, namely soreceive_generic().  Ensure
4810 	 * that it never sees all flags that track connection status
4811 	 * cleared, by ordering the update with a barrier semantic of
4812 	 * our release thread fence.
4813 	 */
4814 	so->so_state |= SS_ISDISCONNECTED;
4815 	atomic_thread_fence_rel();
4816 	so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
4817 
4818 	if (!SOLISTENING(so)) {
4819 		SOCK_UNLOCK(so);
4820 		SOCK_RECVBUF_LOCK(so);
4821 		socantrcvmore_locked(so);
4822 		SOCK_SENDBUF_LOCK(so);
4823 		sbdrop_locked(&so->so_snd, sbused(&so->so_snd));
4824 		socantsendmore_locked(so);
4825 	} else
4826 		SOCK_UNLOCK(so);
4827 	wakeup(&so->so_timeo);
4828 }
4829 
4830 int
4831 soiolock(struct socket *so, struct sx *sx, int flags)
4832 {
4833 	int error;
4834 
4835 	KASSERT((flags & SBL_VALID) == flags,
4836 	    ("soiolock: invalid flags %#x", flags));
4837 
4838 	if ((flags & SBL_WAIT) != 0) {
4839 		if ((flags & SBL_NOINTR) != 0) {
4840 			sx_xlock(sx);
4841 		} else {
4842 			error = sx_xlock_sig(sx);
4843 			if (error != 0)
4844 				return (error);
4845 		}
4846 	} else if (!sx_try_xlock(sx)) {
4847 		return (EWOULDBLOCK);
4848 	}
4849 
4850 	if (__predict_false(SOLISTENING(so))) {
4851 		sx_xunlock(sx);
4852 		return (ENOTCONN);
4853 	}
4854 	return (0);
4855 }
4856 
4857 void
4858 soiounlock(struct sx *sx)
4859 {
4860 	sx_xunlock(sx);
4861 }
4862 
4863 /*
4864  * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
4865  */
4866 struct sockaddr *
4867 sodupsockaddr(const struct sockaddr *sa, int mflags)
4868 {
4869 	struct sockaddr *sa2;
4870 
4871 	sa2 = malloc(sa->sa_len, M_SONAME, mflags);
4872 	if (sa2)
4873 		bcopy(sa, sa2, sa->sa_len);
4874 	return sa2;
4875 }
4876 
4877 /*
4878  * Register per-socket destructor.
4879  */
4880 void
4881 sodtor_set(struct socket *so, so_dtor_t *func)
4882 {
4883 
4884 	SOCK_LOCK_ASSERT(so);
4885 	so->so_dtor = func;
4886 }
4887 
4888 /*
4889  * Register per-socket buffer upcalls.
4890  */
4891 void
4892 soupcall_set(struct socket *so, sb_which which, so_upcall_t func, void *arg)
4893 {
4894 	struct sockbuf *sb;
4895 
4896 	KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so));
4897 
4898 	switch (which) {
4899 	case SO_RCV:
4900 		sb = &so->so_rcv;
4901 		break;
4902 	case SO_SND:
4903 		sb = &so->so_snd;
4904 		break;
4905 	}
4906 	SOCK_BUF_LOCK_ASSERT(so, which);
4907 	sb->sb_upcall = func;
4908 	sb->sb_upcallarg = arg;
4909 	sb->sb_flags |= SB_UPCALL;
4910 }
4911 
4912 void
4913 soupcall_clear(struct socket *so, sb_which which)
4914 {
4915 	struct sockbuf *sb;
4916 
4917 	KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so));
4918 
4919 	switch (which) {
4920 	case SO_RCV:
4921 		sb = &so->so_rcv;
4922 		break;
4923 	case SO_SND:
4924 		sb = &so->so_snd;
4925 		break;
4926 	}
4927 	SOCK_BUF_LOCK_ASSERT(so, which);
4928 	KASSERT(sb->sb_upcall != NULL,
4929 	    ("%s: so %p no upcall to clear", __func__, so));
4930 	sb->sb_upcall = NULL;
4931 	sb->sb_upcallarg = NULL;
4932 	sb->sb_flags &= ~SB_UPCALL;
4933 }
4934 
4935 void
4936 solisten_upcall_set(struct socket *so, so_upcall_t func, void *arg)
4937 {
4938 
4939 	SOLISTEN_LOCK_ASSERT(so);
4940 	so->sol_upcall = func;
4941 	so->sol_upcallarg = arg;
4942 }
4943 
4944 static void
4945 so_rdknl_lock(void *arg)
4946 {
4947 	struct socket *so = arg;
4948 
4949 retry:
4950 	if (SOLISTENING(so)) {
4951 		SOLISTEN_LOCK(so);
4952 	} else {
4953 		SOCK_RECVBUF_LOCK(so);
4954 		if (__predict_false(SOLISTENING(so))) {
4955 			SOCK_RECVBUF_UNLOCK(so);
4956 			goto retry;
4957 		}
4958 	}
4959 }
4960 
4961 static void
4962 so_rdknl_unlock(void *arg)
4963 {
4964 	struct socket *so = arg;
4965 
4966 	if (SOLISTENING(so))
4967 		SOLISTEN_UNLOCK(so);
4968 	else
4969 		SOCK_RECVBUF_UNLOCK(so);
4970 }
4971 
4972 static void
4973 so_rdknl_assert_lock(void *arg, int what)
4974 {
4975 	struct socket *so = arg;
4976 
4977 	if (what == LA_LOCKED) {
4978 		if (SOLISTENING(so))
4979 			SOLISTEN_LOCK_ASSERT(so);
4980 		else
4981 			SOCK_RECVBUF_LOCK_ASSERT(so);
4982 	} else {
4983 		if (SOLISTENING(so))
4984 			SOLISTEN_UNLOCK_ASSERT(so);
4985 		else
4986 			SOCK_RECVBUF_UNLOCK_ASSERT(so);
4987 	}
4988 }
4989 
4990 static void
4991 so_wrknl_lock(void *arg)
4992 {
4993 	struct socket *so = arg;
4994 
4995 retry:
4996 	if (SOLISTENING(so)) {
4997 		SOLISTEN_LOCK(so);
4998 	} else {
4999 		SOCK_SENDBUF_LOCK(so);
5000 		if (__predict_false(SOLISTENING(so))) {
5001 			SOCK_SENDBUF_UNLOCK(so);
5002 			goto retry;
5003 		}
5004 	}
5005 }
5006 
5007 static void
5008 so_wrknl_unlock(void *arg)
5009 {
5010 	struct socket *so = arg;
5011 
5012 	if (SOLISTENING(so))
5013 		SOLISTEN_UNLOCK(so);
5014 	else
5015 		SOCK_SENDBUF_UNLOCK(so);
5016 }
5017 
5018 static void
5019 so_wrknl_assert_lock(void *arg, int what)
5020 {
5021 	struct socket *so = arg;
5022 
5023 	if (what == LA_LOCKED) {
5024 		if (SOLISTENING(so))
5025 			SOLISTEN_LOCK_ASSERT(so);
5026 		else
5027 			SOCK_SENDBUF_LOCK_ASSERT(so);
5028 	} else {
5029 		if (SOLISTENING(so))
5030 			SOLISTEN_UNLOCK_ASSERT(so);
5031 		else
5032 			SOCK_SENDBUF_UNLOCK_ASSERT(so);
5033 	}
5034 }
5035 
5036 /*
5037  * Create an external-format (``xsocket'') structure using the information in
5038  * the kernel-format socket structure pointed to by so.  This is done to
5039  * reduce the spew of irrelevant information over this interface, to isolate
5040  * user code from changes in the kernel structure, and potentially to provide
5041  * information-hiding if we decide that some of this information should be
5042  * hidden from users.
5043  */
5044 void
5045 sotoxsocket(struct socket *so, struct xsocket *xso)
5046 {
5047 
5048 	bzero(xso, sizeof(*xso));
5049 	xso->xso_len = sizeof *xso;
5050 	xso->xso_so = (uintptr_t)so;
5051 	xso->so_type = so->so_type;
5052 	xso->so_options = so->so_options;
5053 	xso->so_linger = so->so_linger;
5054 	xso->so_state = so->so_state;
5055 	xso->so_pcb = (uintptr_t)so->so_pcb;
5056 	xso->xso_protocol = so->so_proto->pr_protocol;
5057 	xso->xso_family = so->so_proto->pr_domain->dom_family;
5058 	xso->so_timeo = so->so_timeo;
5059 	xso->so_error = so->so_error;
5060 	xso->so_uid = so->so_cred->cr_uid;
5061 	xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
5062 	SOCK_LOCK(so);
5063 	xso->so_fibnum = so->so_fibnum;
5064 	if (SOLISTENING(so)) {
5065 		xso->so_qlen = so->sol_qlen;
5066 		xso->so_incqlen = so->sol_incqlen;
5067 		xso->so_qlimit = so->sol_qlimit;
5068 		xso->so_oobmark = 0;
5069 	} else {
5070 		xso->so_state |= so->so_qstate;
5071 		xso->so_qlen = xso->so_incqlen = xso->so_qlimit = 0;
5072 		xso->so_oobmark = so->so_oobmark;
5073 		sbtoxsockbuf(&so->so_snd, &xso->so_snd);
5074 		sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
5075 		if ((so->so_rcv.sb_flags & SB_SPLICED) != 0)
5076 			xso->so_splice_so = (uintptr_t)so->so_splice->dst;
5077 	}
5078 	SOCK_UNLOCK(so);
5079 }
5080 
5081 int
5082 so_options_get(const struct socket *so)
5083 {
5084 
5085 	return (so->so_options);
5086 }
5087 
5088 void
5089 so_options_set(struct socket *so, int val)
5090 {
5091 
5092 	so->so_options = val;
5093 }
5094 
5095 int
5096 so_error_get(const struct socket *so)
5097 {
5098 
5099 	return (so->so_error);
5100 }
5101 
5102 void
5103 so_error_set(struct socket *so, int val)
5104 {
5105 
5106 	so->so_error = val;
5107 }
5108