xref: /src/sys/kern/uipc_socket.c (revision f1e8b1aca54312a2c7f0d22a83aaae44950dac6b)

/*-
 * SPDX-License-Identifier: BSD-3-Clause
 *
 * Copyright (c) 1982, 1986, 1988, 1990, 1993
 *	The Regents of the University of California.
 * Copyright (c) 2004 The FreeBSD Foundation
 * Copyright (c) 2004-2008 Robert N. M. Watson
 * All rights reserved.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 * 1. Redistributions of source code must retain the above copyright
 *    notice, this list of conditions and the following disclaimer.
 * 2. Redistributions in binary form must reproduce the above copyright
 *    notice, this list of conditions and the following disclaimer in the
 *    documentation and/or other materials provided with the distribution.
 * 3. Neither the name of the University nor the names of its contributors
 *    may be used to endorse or promote products derived from this software
 *    without specific prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
 * SUCH DAMAGE.
 */

/*
 * Comments on the socket life cycle:
 *
 * soalloc() sets of socket layer state for a socket, called only by
 * socreate() and sonewconn().  Socket layer private.
 *
 * sodealloc() tears down socket layer state for a socket, called only by
 * sofree() and sonewconn().  Socket layer private.
 *
 * pr_attach() associates protocol layer state with an allocated socket;
 * called only once, may fail, aborting socket allocation.  This is called
 * from socreate() and sonewconn().  Socket layer private.
 *
 * pr_detach() disassociates protocol layer state from an attached socket,
 * and will be called exactly once for sockets in which pr_attach() has
 * been successfully called.  If pr_attach() returned an error,
 * pr_detach() will not be called.  Socket layer private.
 *
 * pr_abort() and pr_close() notify the protocol layer that the last
 * consumer of a socket is starting to tear down the socket, and that the
 * protocol should terminate the connection.  Historically, pr_abort() also
 * detached protocol state from the socket state, but this is no longer the
 * case.  pr_fdclose() is called when userspace invokes close(2) on a socket
 * file descriptor.
 *
 * socreate() creates a socket and attaches protocol state.  This is a public
 * interface that may be used by socket layer consumers to create new
 * sockets.
 *
 * sonewconn() creates a socket and attaches protocol state.  This is a
 * public interface  that may be used by protocols to create new sockets when
 * a new connection is received and will be available for accept() on a
 * listen socket.
 *
 * soclose() destroys a socket after possibly waiting for it to disconnect.
 * This is a public interface that socket consumers should use to close and
 * release a socket when done with it.
 *
 * soabort() destroys a socket without waiting for it to disconnect (used
 * only for incoming connections that are already partially or fully
 * connected).  This is used internally by the socket layer when clearing
 * listen socket queues (due to overflow or close on the listen socket), but
 * is also a public interface protocols may use to abort connections in
 * their incomplete listen queues should they no longer be required.  Sockets
 * placed in completed connection listen queues should not be aborted for
 * reasons described in the comment above the soclose() implementation.  This
 * is not a general purpose close routine, and except in the specific
 * circumstances described here, should not be used.
 *
 * sofree() will free a socket and its protocol state if all references on
 * the socket have been released, and is the public interface to attempt to
 * free a socket when a reference is removed.  This is a socket layer private
 * interface.
 *
 * NOTE: In addition to socreate() and soclose(), which provide a single
 * socket reference to the consumer to be managed as required, there are two
 * calls to explicitly manage socket references, soref(), and sorele().
 * Currently, these are generally required only when transitioning a socket
 * from a listen queue to a file descriptor, in order to prevent garbage
 * collection of the socket at an untimely moment.  For a number of reasons,
 * these interfaces are not preferred, and should be avoided.
 *
 * NOTE: With regard to VNETs the general rule is that callers do not set
 * curvnet. Exceptions to this rule include soabort(), sodisconnect(),
 * sofree(), sorele(), sonewconn() and sorflush(), which are usually called
 * from a pre-set VNET context.  sopoll_generic() currently does not need a
 * VNET context to be set.
 */

#include <sys/cdefs.h>
#include "opt_inet.h"
#include "opt_inet6.h"
#include "opt_kern_tls.h"
#include "opt_ktrace.h"
#include "opt_sctp.h"

#include <sys/param.h>
#include <sys/systm.h>
#include <sys/capsicum.h>
#include <sys/fcntl.h>
#include <sys/limits.h>
#include <sys/lock.h>
#include <sys/mac.h>
#include <sys/malloc.h>
#include <sys/mbuf.h>
#include <sys/mutex.h>
#include <sys/domain.h>
#include <sys/file.h>			/* for struct knote */
#include <sys/hhook.h>
#include <sys/kernel.h>
#include <sys/khelp.h>
#include <sys/kthread.h>
#include <sys/ktls.h>
#include <sys/event.h>
#include <sys/eventhandler.h>
#include <sys/poll.h>
#include <sys/proc.h>
#include <sys/protosw.h>
#include <sys/sbuf.h>
#include <sys/socket.h>
#include <sys/socketvar.h>
#include <sys/resourcevar.h>
#include <net/route.h>
#include <sys/sched.h>
#include <sys/signalvar.h>
#include <sys/smp.h>
#include <sys/stat.h>
#include <sys/sx.h>
#include <sys/sysctl.h>
#include <sys/taskqueue.h>
#include <sys/uio.h>
#include <sys/un.h>
#include <sys/unpcb.h>
#include <sys/jail.h>
#include <sys/syslog.h>
#include <netinet/in.h>
#include <netinet/in_pcb.h>
#include <netinet/tcp.h>

#include <net/vnet.h>

#include <security/mac/mac_framework.h>
#include <security/mac/mac_internal.h>

#include <vm/uma.h>

#ifdef COMPAT_FREEBSD32
#include <sys/mount.h>
#include <sys/sysent.h>
#include <compat/freebsd32/freebsd32.h>
#endif

static int	soreceive_generic_locked(struct socket *so,
		    struct sockaddr **psa, struct uio *uio, struct mbuf **mp,
		    struct mbuf **controlp, int *flagsp);
static int	soreceive_rcvoob(struct socket *so, struct uio *uio,
		    int flags);
static int	soreceive_stream_locked(struct socket *so, struct sockbuf *sb,
		    struct sockaddr **psa, struct uio *uio, struct mbuf **mp,
		    struct mbuf **controlp, int flags);
static int	sosend_generic_locked(struct socket *so, struct sockaddr *addr,
		    struct uio *uio, struct mbuf *top, struct mbuf *control,
		    int flags, struct thread *td);
static void	so_rdknl_lock(void *);
static void	so_rdknl_unlock(void *);
static void	so_rdknl_assert_lock(void *, int);
static void	so_wrknl_lock(void *);
static void	so_wrknl_unlock(void *);
static void	so_wrknl_assert_lock(void *, int);

static void	filt_sordetach(struct knote *kn);
static int	filt_soread(struct knote *kn, long hint);
static void	filt_sowdetach(struct knote *kn);
static int	filt_sowrite(struct knote *kn, long hint);
static int	filt_soempty(struct knote *kn, long hint);

static const struct filterops soread_filtops = {
	.f_isfd = 1,
	.f_detach = filt_sordetach,
	.f_event = filt_soread,
	.f_copy = knote_triv_copy,
};
static const struct filterops sowrite_filtops = {
	.f_isfd = 1,
	.f_detach = filt_sowdetach,
	.f_event = filt_sowrite,
	.f_copy = knote_triv_copy,
};
static const struct filterops soempty_filtops = {
	.f_isfd = 1,
	.f_detach = filt_sowdetach,
	.f_event = filt_soempty,
	.f_copy = knote_triv_copy,
};

so_gen_t	so_gencnt;	/* generation count for sockets */

MALLOC_DEFINE(M_SONAME, "soname", "socket name");
MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");

#define	VNET_SO_ASSERT(so)						\
	VNET_ASSERT(curvnet != NULL,					\
	    ("%s:%d curvnet is NULL, so=%p", __func__, __LINE__, (so)));

#ifdef SOCKET_HHOOK
VNET_DEFINE(struct hhook_head *, socket_hhh[HHOOK_SOCKET_LAST + 1]);
#define	V_socket_hhh		VNET(socket_hhh)
static inline int hhook_run_socket(struct socket *, void *, int32_t);
#endif

#ifdef COMPAT_FREEBSD32
#ifdef __amd64__
/* off_t has 4-byte alignment on i386 but not on other 32-bit platforms. */
#define	__splice32_packed	__packed
#else
#define	__splice32_packed
#endif
struct splice32 {
	int32_t	sp_fd;
	int64_t sp_max;
	struct timeval32 sp_idle;
} __splice32_packed;
#undef __splice32_packed
#endif

/*
 * Limit on the number of connections in the listen queue waiting
 * for accept(2).
 * NB: The original sysctl somaxconn is still available but hidden
 * to prevent confusion about the actual purpose of this number.
 */
VNET_DEFINE_STATIC(u_int, somaxconn) = SOMAXCONN;
#define	V_somaxconn	VNET(somaxconn)

static int
sysctl_somaxconn(SYSCTL_HANDLER_ARGS)
{
	int error;
	u_int val;

	val = V_somaxconn;
	error = sysctl_handle_int(oidp, &val, 0, req);
	if (error || !req->newptr )
		return (error);

	/*
	 * The purpose of the UINT_MAX / 3 limit, is so that the formula
	 *   3 * sol_qlimit / 2
	 * below, will not overflow.
         */

	if (val < 1 || val > UINT_MAX / 3)
		return (EINVAL);

	V_somaxconn = val;
	return (0);
}
SYSCTL_PROC(_kern_ipc, OID_AUTO, soacceptqueue,
    CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_MPSAFE | CTLFLAG_VNET, 0, sizeof(u_int),
    sysctl_somaxconn, "IU",
    "Maximum listen socket pending connection accept queue size");
SYSCTL_PROC(_kern_ipc, KIPC_SOMAXCONN, somaxconn,
    CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_SKIP | CTLFLAG_MPSAFE | CTLFLAG_VNET, 0,
    sizeof(u_int), sysctl_somaxconn, "IU",
    "Maximum listen socket pending connection accept queue size (compat)");

static u_int numopensockets;
static int
sysctl_numopensockets(SYSCTL_HANDLER_ARGS)
{
	u_int val;

#ifdef VIMAGE
	if(!IS_DEFAULT_VNET(curvnet))
		val = curvnet->vnet_sockcnt;
	else
#endif
		val = numopensockets;
	return (sysctl_handle_int(oidp, &val, 0, req));
}
SYSCTL_PROC(_kern_ipc, OID_AUTO, numopensockets,
    CTLTYPE_UINT | CTLFLAG_RD | CTLFLAG_MPSAFE | CTLFLAG_VNET, 0, sizeof(u_int),
    sysctl_numopensockets, "IU", "Number of open sockets");

/*
 * so_global_mtx protects so_gencnt, numopensockets, and the per-socket
 * so_gencnt field.
 */
static struct mtx so_global_mtx;
MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF);

/*
 * General IPC sysctl name space, used by sockets and a variety of other IPC
 * types.
 */
SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
    "IPC");

/*
 * Initialize the socket subsystem and set up the socket
 * memory allocator.
 */
static uma_zone_t socket_zone;
int	maxsockets;

static void
socket_zone_change(void *tag)
{

	maxsockets = uma_zone_set_max(socket_zone, maxsockets);
}

static int splice_init_state;
static struct sx splice_init_lock;
SX_SYSINIT(splice_init_lock, &splice_init_lock, "splice_init");

static SYSCTL_NODE(_kern_ipc, OID_AUTO, splice, CTLFLAG_RW, 0,
    "Settings relating to the SO_SPLICE socket option");

static bool splice_receive_stream = true;
SYSCTL_BOOL(_kern_ipc_splice, OID_AUTO, receive_stream, CTLFLAG_RWTUN,
    &splice_receive_stream, 0,
    "Use soreceive_stream() for stream splices");

static int splice_num_wq = -1;
static int
sysctl_splice_num_wq(SYSCTL_HANDLER_ARGS)
{
	int error, new;

	new = splice_num_wq;
	error = sysctl_handle_int(oidp, &new, 0, req);
	if (error == 0 && req->newptr && new != splice_num_wq) {
		if (!cold)
			sx_xlock(&splice_init_lock);
		if (new < -1 || new > mp_ncpus ||
		    (new <= 0 && splice_init_state != 0)) {
			error = EINVAL;
		} else {
			splice_num_wq = new;
		}
		if (!cold)
			sx_xunlock(&splice_init_lock);
	}
	return (error);
}
SYSCTL_PROC(_kern_ipc_splice, OID_AUTO, num_wq,
    CTLTYPE_INT | CTLFLAG_RWTUN | CTLFLAG_MPSAFE,
    &splice_num_wq, 0, sysctl_splice_num_wq, "IU",
    "Number of splice worker queues");

static uma_zone_t splice_zone;
static struct proc *splice_proc;
struct splice_wq {
	struct mtx	mtx;
	STAILQ_HEAD(, so_splice) head;
	bool		running;
} __aligned(CACHE_LINE_SIZE);
static struct splice_wq *splice_wq;
static uint32_t splice_index = 0;

static void so_splice_timeout(void *arg, int pending);
static void so_splice_xfer(struct so_splice *s);
static int so_unsplice(struct socket *so, bool timeout);

static void
splice_work_thread(void *ctx)
{
	struct splice_wq *wq = ctx;
	struct so_splice *s, *s_temp;
	STAILQ_HEAD(, so_splice) local_head;
	int cpu;

	cpu = wq - splice_wq;
	if (bootverbose)
		printf("starting so_splice worker thread for CPU %d\n", cpu);

	for (;;) {
		mtx_lock(&wq->mtx);
		while (STAILQ_EMPTY(&wq->head)) {
			wq->running = false;
			mtx_sleep(wq, &wq->mtx, 0, "-", 0);
			wq->running = true;
		}
		STAILQ_INIT(&local_head);
		STAILQ_CONCAT(&local_head, &wq->head);
		STAILQ_INIT(&wq->head);
		mtx_unlock(&wq->mtx);
		STAILQ_FOREACH_SAFE(s, &local_head, next, s_temp) {
			mtx_lock(&s->mtx);
			CURVNET_SET(s->src->so_vnet);
			so_splice_xfer(s);
			CURVNET_RESTORE();
		}
	}
}

static void
so_splice_dispatch_async(struct so_splice *sp)
{
	struct splice_wq *wq;
	bool running;

	wq = &splice_wq[sp->wq_index];
	mtx_lock(&wq->mtx);
	STAILQ_INSERT_TAIL(&wq->head, sp, next);
	running = wq->running;
	mtx_unlock(&wq->mtx);
	if (!running)
		wakeup(wq);
}

void
so_splice_dispatch(struct so_splice *sp)
{
	mtx_assert(&sp->mtx, MA_OWNED);

	if (sp->state != SPLICE_IDLE) {
		mtx_unlock(&sp->mtx);
	} else {
		sp->state = SPLICE_QUEUED;
		mtx_unlock(&sp->mtx);
		so_splice_dispatch_async(sp);
	}
}

static int
splice_zinit(void *mem, int size __unused, int flags __unused)
{
	struct so_splice *s;

	s = (struct so_splice *)mem;
	mtx_init(&s->mtx, "so_splice", NULL, MTX_DEF);
	return (0);
}

static void
splice_zfini(void *mem, int size)
{
	struct so_splice *s;

	s = (struct so_splice *)mem;
	mtx_destroy(&s->mtx);
}

static int
splice_init(void)
{
	struct thread *td;
	int error, i, state;

	state = atomic_load_acq_int(&splice_init_state);
	if (__predict_true(state > 0))
		return (0);
	if (state < 0)
		return (ENXIO);
	sx_xlock(&splice_init_lock);
	if (splice_init_state != 0) {
		sx_xunlock(&splice_init_lock);
		return (0);
	}

	if (splice_num_wq == -1) {
		/* if no user preference, use all cores */
		splice_num_wq = mp_ncpus;
	} else if (splice_num_wq == 0) {
		/* allow user to disable */
		splice_init_state = -1;
		sx_xunlock(&splice_init_lock);
		return (ENXIO);
	} else if (splice_num_wq > mp_ncpus) {
		splice_num_wq = mp_ncpus;
	}

	splice_zone = uma_zcreate("splice", sizeof(struct so_splice), NULL,
	    NULL, splice_zinit, splice_zfini, UMA_ALIGN_CACHE, 0);

	splice_wq = mallocarray(mp_ncpus, sizeof(*splice_wq), M_TEMP,
	    M_WAITOK | M_ZERO);

	/*
	 * Initialize the workqueues to run the splice work.  We create a
	 * work queue for each CPU.
	 */
	for (i = 0; i < mp_ncpus; i++) {
		STAILQ_INIT(&splice_wq[i].head);
		mtx_init(&splice_wq[i].mtx, "splice work queue", NULL, MTX_DEF);
	}

	/* Start kthreads for each workqueue. */
	error = 0;
	for (i = 0; i < mp_ncpus; i++) {
		error = kproc_kthread_add(splice_work_thread, &splice_wq[i],
		    &splice_proc, &td, 0, 0, "so_splice", "thr_%d", i);
		if (error) {
			printf("Can't add so_splice thread %d error %d\n",
			    i, error);
			break;
		}

		/*
		 * It's possible to create loops with SO_SPLICE; ensure that
		 * worker threads aren't able to starve the system too easily.
		 */
		thread_lock(td);
		sched_prio(td, PUSER);
		thread_unlock(td);
	}

	splice_init_state = error != 0 ? -1 : 1;
	sx_xunlock(&splice_init_lock);

	return (error);
}

/*
 * Lock a pair of socket's I/O locks for splicing.  Avoid blocking while holding
 * one lock in order to avoid potential deadlocks in case there is some other
 * code path which acquires more than one I/O lock at a time.
 */
static void
splice_lock_pair(struct socket *so_src, struct socket *so_dst)
{
	int error;

	for (;;) {
		error = SOCK_IO_SEND_LOCK(so_dst, SBL_WAIT | SBL_NOINTR);
		KASSERT(error == 0,
		    ("%s: failed to lock send I/O lock: %d", __func__, error));
		error = SOCK_IO_RECV_LOCK(so_src, 0);
		KASSERT(error == 0 || error == EWOULDBLOCK,
		    ("%s: failed to lock recv I/O lock: %d", __func__, error));
		if (error == 0)
			break;
		SOCK_IO_SEND_UNLOCK(so_dst);

		error = SOCK_IO_RECV_LOCK(so_src, SBL_WAIT | SBL_NOINTR);
		KASSERT(error == 0,
		    ("%s: failed to lock recv I/O lock: %d", __func__, error));
		error = SOCK_IO_SEND_LOCK(so_dst, 0);
		KASSERT(error == 0 || error == EWOULDBLOCK,
		    ("%s: failed to lock send I/O lock: %d", __func__, error));
		if (error == 0)
			break;
		SOCK_IO_RECV_UNLOCK(so_src);
	}
}

static void
splice_unlock_pair(struct socket *so_src, struct socket *so_dst)
{
	SOCK_IO_RECV_UNLOCK(so_src);
	SOCK_IO_SEND_UNLOCK(so_dst);
}

/*
 * Move data from the source to the sink.  Assumes that both of the relevant
 * socket I/O locks are held.
 */
static int
so_splice_xfer_data(struct socket *so_src, struct socket *so_dst, off_t max,
    ssize_t *lenp)
{
	struct uio uio;
	struct mbuf *m;
	struct sockbuf *sb_src, *sb_dst;
	ssize_t len;
	long space;
	int error, flags;

	SOCK_IO_RECV_ASSERT_LOCKED(so_src);
	SOCK_IO_SEND_ASSERT_LOCKED(so_dst);

	error = 0;
	m = NULL;
	memset(&uio, 0, sizeof(uio));

	sb_src = &so_src->so_rcv;
	sb_dst = &so_dst->so_snd;

	space = sbspace(sb_dst);
	if (space < 0)
		space = 0;
	len = MIN(max, MIN(space, sbavail(sb_src)));
	if (len == 0) {
		SOCK_RECVBUF_LOCK(so_src);
		if ((sb_src->sb_state & SBS_CANTRCVMORE) != 0)
			error = EPIPE;
		SOCK_RECVBUF_UNLOCK(so_src);
	} else {
		flags = MSG_DONTWAIT;
		uio.uio_resid = len;
		if (splice_receive_stream && sb_src->sb_tls_info == NULL) {
			error = soreceive_stream_locked(so_src, sb_src, NULL,
			    &uio, &m, NULL, flags);
		} else {
			error = soreceive_generic_locked(so_src, NULL,
			    &uio, &m, NULL, &flags);
		}
		if (error != 0 && m != NULL) {
			m_freem(m);
			m = NULL;
		}
	}
	if (m != NULL) {
		len -= uio.uio_resid;
		error = sosend_generic_locked(so_dst, NULL, NULL, m, NULL,
		    MSG_DONTWAIT, curthread);
	} else if (error == 0) {
		len = 0;
		SOCK_SENDBUF_LOCK(so_dst);
		if ((sb_dst->sb_state & SBS_CANTSENDMORE) != 0)
			error = EPIPE;
		SOCK_SENDBUF_UNLOCK(so_dst);
	}
	if (error == 0)
		*lenp = len;
	return (error);
}

/*
 * Transfer data from the source to the sink.
 */
static void
so_splice_xfer(struct so_splice *sp)
{
	struct socket *so_src, *so_dst;
	off_t max;
	ssize_t len;
	int error;

	mtx_assert(&sp->mtx, MA_OWNED);
	KASSERT(sp->state == SPLICE_QUEUED || sp->state == SPLICE_CLOSING,
	    ("so_splice_xfer: invalid state %d", sp->state));
	KASSERT(sp->max != 0, ("so_splice_xfer: max == 0"));

	if (sp->state == SPLICE_CLOSING) {
		/* Userspace asked us to close the splice. */
		goto closing;
	}

	sp->state = SPLICE_RUNNING;
	so_src = sp->src;
	so_dst = sp->dst;
	max = sp->max > 0 ? sp->max - so_src->so_splice_sent : OFF_MAX;
	if (max < 0)
		max = 0;

	/*
	 * Lock the sockets in order to block userspace from doing anything
	 * sneaky.  If an error occurs or one of the sockets can no longer
	 * transfer data, we will automatically unsplice.
	 */
	mtx_unlock(&sp->mtx);
	splice_lock_pair(so_src, so_dst);

	error = so_splice_xfer_data(so_src, so_dst, max, &len);

	mtx_lock(&sp->mtx);

	/*
	 * Update our stats while still holding the socket locks.  This
	 * synchronizes with getsockopt(SO_SPLICE), see the comment there.
	 */
	if (error == 0) {
		KASSERT(len >= 0, ("%s: len %zd < 0", __func__, len));
		so_src->so_splice_sent += len;
	}
	splice_unlock_pair(so_src, so_dst);

	switch (sp->state) {
	case SPLICE_CLOSING:
closing:
		sp->state = SPLICE_CLOSED;
		wakeup(sp);
		mtx_unlock(&sp->mtx);
		break;
	case SPLICE_RUNNING:
		if (error != 0 ||
		    (sp->max > 0 && so_src->so_splice_sent >= sp->max)) {
			sp->state = SPLICE_EXCEPTION;
			soref(so_src);
			mtx_unlock(&sp->mtx);
			(void)so_unsplice(so_src, false);
			sorele(so_src);
		} else {
			/*
			 * Locklessly check for additional bytes in the source's
			 * receive buffer and queue more work if possible.  We
			 * may end up queuing needless work, but that's ok, and
			 * if we race with a thread inserting more data into the
			 * buffer and observe sbavail() == 0, the splice mutex
			 * ensures that splice_push() will queue more work for
			 * us.
			 */
			if (sbavail(&so_src->so_rcv) > 0 &&
			    sbspace(&so_dst->so_snd) > 0) {
				sp->state = SPLICE_QUEUED;
				mtx_unlock(&sp->mtx);
				so_splice_dispatch_async(sp);
			} else {
				sp->state = SPLICE_IDLE;
				mtx_unlock(&sp->mtx);
			}
		}
		break;
	default:
		__assert_unreachable();
	}
}

static void
socket_init(void *tag)
{

	socket_zone = uma_zcreate("socket", sizeof(struct socket), NULL, NULL,
	    NULL, NULL, UMA_ALIGN_PTR, 0);
	maxsockets = uma_zone_set_max(socket_zone, maxsockets);
	uma_zone_set_warning(socket_zone, "kern.ipc.maxsockets limit reached");
	EVENTHANDLER_REGISTER(maxsockets_change, socket_zone_change, NULL,
	    EVENTHANDLER_PRI_FIRST);
}
SYSINIT(socket, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, socket_init, NULL);

#ifdef SOCKET_HHOOK
static void
socket_hhook_register(int subtype)
{

	if (hhook_head_register(HHOOK_TYPE_SOCKET, subtype,
	    &V_socket_hhh[subtype],
	    HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
		printf("%s: WARNING: unable to register hook\n", __func__);
}

static void
socket_hhook_deregister(int subtype)
{

	if (hhook_head_deregister(V_socket_hhh[subtype]) != 0)
		printf("%s: WARNING: unable to deregister hook\n", __func__);
}

static void
socket_vnet_init(const void *unused __unused)
{
	int i;

	/* We expect a contiguous range */
	for (i = 0; i <= HHOOK_SOCKET_LAST; i++)
		socket_hhook_register(i);
}
VNET_SYSINIT(socket_vnet_init, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
    socket_vnet_init, NULL);

static void
socket_vnet_uninit(const void *unused __unused)
{
	int i;

	for (i = 0; i <= HHOOK_SOCKET_LAST; i++)
		socket_hhook_deregister(i);
}
VNET_SYSUNINIT(socket_vnet_uninit, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
    socket_vnet_uninit, NULL);
#endif	/* SOCKET_HHOOK */

/*
 * Initialise maxsockets.  This SYSINIT must be run after
 * tunable_mbinit().
 */
static void
init_maxsockets(void *ignored)
{

	TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets);
	maxsockets = imax(maxsockets, maxfiles);
}
SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL);

/*
 * Sysctl to get and set the maximum global sockets limit.  Notify protocols
 * of the change so that they can update their dependent limits as required.
 */
static int
sysctl_maxsockets(SYSCTL_HANDLER_ARGS)
{
	int error, newmaxsockets;

	newmaxsockets = maxsockets;
	error = sysctl_handle_int(oidp, &newmaxsockets, 0, req);
	if (error == 0 && req->newptr && newmaxsockets != maxsockets) {
		if (newmaxsockets > maxsockets &&
		    newmaxsockets <= maxfiles) {
			maxsockets = newmaxsockets;
			EVENTHANDLER_INVOKE(maxsockets_change);
		} else
			error = EINVAL;
	}
	return (error);
}
SYSCTL_PROC(_kern_ipc, OID_AUTO, maxsockets,
    CTLTYPE_INT | CTLFLAG_RWTUN | CTLFLAG_NOFETCH | CTLFLAG_MPSAFE,
    &maxsockets, 0, sysctl_maxsockets, "IU",
    "Maximum number of sockets available");

/*
 * Socket operation routines.  These routines are called by the routines in
 * sys_socket.c or from a system process, and implement the semantics of
 * socket operations by switching out to the protocol specific routines.
 */

/*
 * Get a socket structure from our zone, and initialize it.  Note that it
 * would probably be better to allocate socket and PCB at the same time, but
 * I'm not convinced that all the protocols can be easily modified to do
 * this.
 *
 * soalloc() returns a socket with a ref count of 0.
 */
static struct socket *
soalloc(struct vnet *vnet)
{
	struct socket *so;

	so = uma_zalloc(socket_zone, M_NOWAIT | M_ZERO);
	if (so == NULL)
		return (NULL);
#ifdef MAC
	if (mac_socket_init(so, M_NOWAIT) != 0) {
		uma_zfree(socket_zone, so);
		return (NULL);
	}
#endif
	if (khelp_init_osd(HELPER_CLASS_SOCKET, &so->osd)) {
		uma_zfree(socket_zone, so);
		return (NULL);
	}

	/*
	 * The socket locking protocol allows to lock 2 sockets at a time,
	 * however, the first one must be a listening socket.  WITNESS lacks
	 * a feature to change class of an existing lock, so we use DUPOK.
	 */
	mtx_init(&so->so_lock, "socket", NULL, MTX_DEF | MTX_DUPOK);
	so->so_rcv.sb_sel = &so->so_rdsel;
	so->so_snd.sb_sel = &so->so_wrsel;
	sx_init(&so->so_snd_sx, "so_snd_sx");
	sx_init(&so->so_rcv_sx, "so_rcv_sx");
	TAILQ_INIT(&so->so_snd.sb_aiojobq);
	TAILQ_INIT(&so->so_rcv.sb_aiojobq);
	TASK_INIT(&so->so_snd.sb_aiotask, 0, soaio_snd, so);
	TASK_INIT(&so->so_rcv.sb_aiotask, 0, soaio_rcv, so);
#ifdef VIMAGE
	VNET_ASSERT(vnet != NULL, ("%s:%d vnet is NULL, so=%p",
	    __func__, __LINE__, so));
	so->so_vnet = vnet;
#endif
#ifdef SOCKET_HHOOK
	/* We shouldn't need the so_global_mtx */
	if (hhook_run_socket(so, NULL, HHOOK_SOCKET_CREATE)) {
		/* Do we need more comprehensive error returns? */
		uma_zfree(socket_zone, so);
		return (NULL);
	}
#endif
	mtx_lock(&so_global_mtx);
	so->so_gencnt = ++so_gencnt;
	++numopensockets;
#ifdef VIMAGE
	vnet->vnet_sockcnt++;
#endif
	mtx_unlock(&so_global_mtx);

	return (so);
}

/*
 * Free the storage associated with a socket at the socket layer, tear down
 * locks, labels, etc.  All protocol state is assumed already to have been
 * torn down (and possibly never set up) by the caller.
 */
void
sodealloc(struct socket *so)
{

	KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count));
	KASSERT(so->so_pcb == NULL, ("sodealloc(): so_pcb != NULL"));

	mtx_lock(&so_global_mtx);
	so->so_gencnt = ++so_gencnt;
	--numopensockets;	/* Could be below, but faster here. */
#ifdef VIMAGE
	VNET_ASSERT(so->so_vnet != NULL, ("%s:%d so_vnet is NULL, so=%p",
	    __func__, __LINE__, so));
	so->so_vnet->vnet_sockcnt--;
#endif
	mtx_unlock(&so_global_mtx);
#ifdef MAC
	mac_socket_destroy(so);
#endif
#ifdef SOCKET_HHOOK
	hhook_run_socket(so, NULL, HHOOK_SOCKET_CLOSE);
#endif

	khelp_destroy_osd(&so->osd);
	if (SOLISTENING(so)) {
		if (so->sol_accept_filter != NULL)
			accept_filt_setopt(so, NULL);
	} else {
		if (so->so_rcv.sb_hiwat)
			(void)chgsbsize(so->so_cred->cr_uidinfo,
			    &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY);
		if (so->so_snd.sb_hiwat)
			(void)chgsbsize(so->so_cred->cr_uidinfo,
			    &so->so_snd.sb_hiwat, 0, RLIM_INFINITY);
		sx_destroy(&so->so_snd_sx);
		sx_destroy(&so->so_rcv_sx);
	}
	crfree(so->so_cred);
	mtx_destroy(&so->so_lock);
	uma_zfree(socket_zone, so);
}

/*
 * Shim to accomodate protocols that already do their own socket buffers
 * management (marked with PR_SOCKBUF) with protocols that yet do not.
 *
 * Attach via socket(2) is different from attach via accept(2).  In case of
 * normal socket(2) syscall it is the pr_attach that calls soreserve(), even
 * for protocols that don't yet do PR_SOCKBUF.  In case of accepted connection
 * it is our shim that calls soreserve() and the hiwat values are taken from
 * the parent socket.  The SCTP's sopeeloff() hands us a non-listening parent
 * socket.
 *
 * This whole shim should go away when all major protocols fully manage their
 * socket buffers.
 */
static int
soattach(struct socket *so, int proto, struct thread *td, struct socket *head)
{
	int error;

	VNET_ASSERT(curvnet == so->so_vnet,
	    ("%s: %p != %p", __func__, curvnet,  so->so_vnet));

	if ((so->so_proto->pr_flags & PR_SOCKBUF) == 0) {
		mtx_init(&so->so_snd_mtx, "so_snd", NULL, MTX_DEF);
		mtx_init(&so->so_rcv_mtx, "so_rcv", NULL, MTX_DEF);
		so->so_snd.sb_mtx = &so->so_snd_mtx;
		so->so_rcv.sb_mtx = &so->so_rcv_mtx;
	}
	if (head == NULL || (error = soreserve(so,
	    SOLISTENING(head) ? head->sol_sbsnd_hiwat : head->so_snd.sb_hiwat,
	    SOLISTENING(head) ? head->sol_sbrcv_hiwat : head->so_rcv.sb_hiwat))
	    == 0)
		error = so->so_proto->pr_attach(so, proto, td);
	if (error != 0 && (so->so_proto->pr_flags & PR_SOCKBUF) == 0) {
		mtx_destroy(&so->so_snd_mtx);
		mtx_destroy(&so->so_rcv_mtx);
	}

	return (error);
}

/*
 * socreate returns a socket with a ref count of 1 and a file descriptor
 * reference.  The socket should be closed with soclose().
 */
int
socreate(int dom, struct socket **aso, int type, int proto,
    struct ucred *cred, struct thread *td)
{
	struct protosw *prp;
	struct socket *so;
	int error;

	prp = pffindproto(dom, type, proto);
	if (prp == NULL) {
		/* No support for domain. */
		if (pffinddomain(dom) == NULL)
			return (EAFNOSUPPORT);
		/* No support for socket type. */
		if (proto == 0 && type != 0)
			return (EPROTOTYPE);
		return (EPROTONOSUPPORT);
	}

	MPASS(prp->pr_attach);

	if ((prp->pr_flags & PR_CAPATTACH) == 0) {
		if (CAP_TRACING(td))
			ktrcapfail(CAPFAIL_PROTO, &proto);
		if (IN_CAPABILITY_MODE(td))
			return (ECAPMODE);
	}

	if (prison_check_af(cred, prp->pr_domain->dom_family) != 0)
		return (EPROTONOSUPPORT);

	so = soalloc(CRED_TO_VNET(cred));
	if (so == NULL)
		return (ENOBUFS);

	so->so_type = type;
	so->so_cred = crhold(cred);
	if ((prp->pr_domain->dom_family == PF_INET) ||
	    (prp->pr_domain->dom_family == PF_INET6) ||
	    (prp->pr_domain->dom_family == PF_ROUTE))
		so->so_fibnum = td->td_proc->p_fibnum;
	else
		so->so_fibnum = 0;
	so->so_proto = prp;
#ifdef MAC
	mac_socket_create(cred, so);
#endif
	knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
	    so_rdknl_assert_lock);
	knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
	    so_wrknl_assert_lock);
	CURVNET_SET(so->so_vnet);
	error = soattach(so, proto, td, NULL);
	CURVNET_RESTORE();
	if (error) {
		sodealloc(so);
		return (error);
	}
	soref(so);
	*aso = so;
	return (0);
}

#ifdef REGRESSION
static int regression_sonewconn_earlytest = 1;
SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW,
    &regression_sonewconn_earlytest, 0, "Perform early sonewconn limit test");
#endif

static int sooverprio = LOG_DEBUG;
SYSCTL_INT(_kern_ipc, OID_AUTO, sooverprio, CTLFLAG_RW,
    &sooverprio, 0, "Log priority for listen socket overflows: 0..7 or -1 to disable");

static struct timeval overinterval = { 60, 0 };
SYSCTL_TIMEVAL_SEC(_kern_ipc, OID_AUTO, sooverinterval, CTLFLAG_RW,
    &overinterval,
    "Delay in seconds between warnings for listen socket overflows");

/*
 * When an attempt at a new connection is noted on a socket which supports
 * accept(2), the protocol has two options:
 * 1) Call legacy sonewconn() function, which would call protocol attach
 *    method, same as used for socket(2).
 * 2) Call solisten_clone(), do attach that is specific to a cloned connection,
 *    and then call solisten_enqueue().
 *
 * Note: the ref count on the socket is 0 on return.
 */
struct socket *
solisten_clone(struct socket *head)
{
	struct sbuf descrsb;
	struct socket *so;
	int len, overcount;
	u_int qlen;
	const char localprefix[] = "local:";
	char descrbuf[SUNPATHLEN + sizeof(localprefix)];
#if defined(INET6)
	char addrbuf[INET6_ADDRSTRLEN];
#elif defined(INET)
	char addrbuf[INET_ADDRSTRLEN];
#endif
	bool dolog, over;

	SOLISTEN_LOCK(head);
	over = (head->sol_qlen > 3 * head->sol_qlimit / 2);
#ifdef REGRESSION
	if (regression_sonewconn_earlytest && over) {
#else
	if (over) {
#endif
		head->sol_overcount++;
		dolog = (sooverprio >= 0) &&
			!!ratecheck(&head->sol_lastover, &overinterval);

		/*
		 * If we're going to log, copy the overflow count and queue
		 * length from the listen socket before dropping the lock.
		 * Also, reset the overflow count.
		 */
		if (dolog) {
			overcount = head->sol_overcount;
			head->sol_overcount = 0;
			qlen = head->sol_qlen;
		}
		SOLISTEN_UNLOCK(head);

		if (dolog) {
			/*
			 * Try to print something descriptive about the
			 * socket for the error message.
			 */
			sbuf_new(&descrsb, descrbuf, sizeof(descrbuf),
			    SBUF_FIXEDLEN);
			switch (head->so_proto->pr_domain->dom_family) {
#if defined(INET) || defined(INET6)
#ifdef INET
			case AF_INET:
#endif
#ifdef INET6
			case AF_INET6:
				if (head->so_proto->pr_domain->dom_family ==
				    AF_INET6 ||
				    (sotoinpcb(head)->inp_inc.inc_flags &
				    INC_ISIPV6)) {
					ip6_sprintf(addrbuf,
					    &sotoinpcb(head)->inp_inc.inc6_laddr);
					sbuf_printf(&descrsb, "[%s]", addrbuf);
				} else
#endif
				{
#ifdef INET
					inet_ntoa_r(
					    sotoinpcb(head)->inp_inc.inc_laddr,
					    addrbuf);
					sbuf_cat(&descrsb, addrbuf);
#endif
				}
				sbuf_printf(&descrsb, ":%hu (proto %u)",
				    ntohs(sotoinpcb(head)->inp_inc.inc_lport),
				    head->so_proto->pr_protocol);
				break;
#endif /* INET || INET6 */
			case AF_UNIX:
				sbuf_cat(&descrsb, localprefix);
				if (sotounpcb(head)->unp_addr != NULL)
					len =
					    sotounpcb(head)->unp_addr->sun_len -
					    offsetof(struct sockaddr_un,
					    sun_path);
				else
					len = 0;
				if (len > 0)
					sbuf_bcat(&descrsb,
					    sotounpcb(head)->unp_addr->sun_path,
					    len);
				else
					sbuf_cat(&descrsb, "(unknown)");
				break;
			}

			/*
			 * If we can't print something more specific, at least
			 * print the domain name.
			 */
			if (sbuf_finish(&descrsb) != 0 ||
			    sbuf_len(&descrsb) <= 0) {
				sbuf_clear(&descrsb);
				sbuf_cat(&descrsb,
				    head->so_proto->pr_domain->dom_name ?:
				    "unknown");
				sbuf_finish(&descrsb);
			}
			KASSERT(sbuf_len(&descrsb) > 0,
			    ("%s: sbuf creation failed", __func__));
			/*
			 * Preserve the historic listen queue overflow log
			 * message, that starts with "sonewconn:".  It has
			 * been known to sysadmins for years and also test
			 * sys/kern/sonewconn_overflow checks for it.
			 */
			if (head->so_cred == 0) {
				log(LOG_PRI(sooverprio),
				    "sonewconn: pcb %p (%s): "
				    "Listen queue overflow: %i already in "
				    "queue awaiting acceptance (%d "
				    "occurrences)\n", head->so_pcb,
				    sbuf_data(&descrsb),
			    	qlen, overcount);
			} else {
				log(LOG_PRI(sooverprio),
				    "sonewconn: pcb %p (%s): "
				    "Listen queue overflow: "
				    "%i already in queue awaiting acceptance "
				    "(%d occurrences), euid %d, rgid %d, jail %s\n",
				    head->so_pcb, sbuf_data(&descrsb), qlen,
				    overcount, head->so_cred->cr_uid,
				    head->so_cred->cr_rgid,
				    head->so_cred->cr_prison ?
					head->so_cred->cr_prison->pr_name :
					"not_jailed");
			}
			sbuf_delete(&descrsb);

			overcount = 0;
		}

		return (NULL);
	}
	SOLISTEN_UNLOCK(head);
	VNET_ASSERT(head->so_vnet != NULL, ("%s: so %p vnet is NULL",
	    __func__, head));
	so = soalloc(head->so_vnet);
	if (so == NULL) {
		log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: "
		    "limit reached or out of memory\n",
		    __func__, head->so_pcb);
		return (NULL);
	}
	so->so_listen = head;
	so->so_type = head->so_type;
	/*
	 * POSIX is ambiguous on what options an accept(2)ed socket should
	 * inherit from the listener.  Words "create a new socket" may be
	 * interpreted as not inheriting anything.  Best programming practice
	 * for application developers is to not rely on such inheritance.
	 * FreeBSD had historically inherited all so_options excluding
	 * SO_ACCEPTCONN, which virtually means all SOL_SOCKET level options,
	 * including those completely irrelevant to a new born socket.  For
	 * compatibility with older versions we will inherit a list of
	 * meaningful options.
	 * The crucial bit to inherit is SO_ACCEPTFILTER.  We need it present
	 * in the child socket for soisconnected() promoting socket from the
	 * incomplete queue to complete.  It will be cleared before the child
	 * gets available to accept(2).
	 */
	so->so_options = head->so_options & (SO_ACCEPTFILTER | SO_KEEPALIVE |
	    SO_DONTROUTE | SO_LINGER | SO_OOBINLINE | SO_NOSIGPIPE);
	so->so_linger = head->so_linger;
	so->so_state = head->so_state;
	so->so_fibnum = head->so_fibnum;
	so->so_proto = head->so_proto;
	so->so_cred = crhold(head->so_cred);
#ifdef SOCKET_HHOOK
	if (V_socket_hhh[HHOOK_SOCKET_NEWCONN]->hhh_nhooks > 0) {
		if (hhook_run_socket(so, head, HHOOK_SOCKET_NEWCONN)) {
			sodealloc(so);
			log(LOG_DEBUG, "%s: hhook run failed\n", __func__);
			return (NULL);
		}
	}
#endif
#ifdef MAC
	mac_socket_newconn(head, so);
#endif
	knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
	    so_rdknl_assert_lock);
	knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
	    so_wrknl_assert_lock);
	so->so_rcv.sb_lowat = head->sol_sbrcv_lowat;
	so->so_snd.sb_lowat = head->sol_sbsnd_lowat;
	so->so_rcv.sb_timeo = head->sol_sbrcv_timeo;
	so->so_snd.sb_timeo = head->sol_sbsnd_timeo;
	so->so_rcv.sb_flags = head->sol_sbrcv_flags & SB_AUTOSIZE;
	so->so_snd.sb_flags = head->sol_sbsnd_flags &
	    (SB_AUTOSIZE | SB_AUTOLOWAT);

	return (so);
}

/* Connstatus may be 0 or SS_ISCONNECTED. */
struct socket *
sonewconn(struct socket *head, int connstatus)
{
	struct socket *so;

	if ((so = solisten_clone(head)) == NULL)
		return (NULL);

	if (soattach(so, 0, NULL, head) != 0) {
		sodealloc(so);
		log(LOG_DEBUG, "%s: pcb %p: pr_attach() failed\n",
		    __func__, head->so_pcb);
		return (NULL);
	}

	(void)solisten_enqueue(so, connstatus);

	return (so);
}

/*
 * Enqueue socket cloned by solisten_clone() to the listen queue of the
 * listener it has been cloned from.
 *
 * Return 'true' if socket landed on complete queue, otherwise 'false'.
 */
bool
solisten_enqueue(struct socket *so, int connstatus)
{
	struct socket *head = so->so_listen;

	MPASS(refcount_load(&so->so_count) == 0);
	refcount_init(&so->so_count, 1);

	SOLISTEN_LOCK(head);
	if (head->sol_accept_filter != NULL)
		connstatus = 0;
	so->so_state |= connstatus;
	soref(head); /* A socket on (in)complete queue refs head. */
	if (connstatus) {
		TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list);
		so->so_qstate = SQ_COMP;
		head->sol_qlen++;
		solisten_wakeup(head);	/* unlocks */
		return (true);
	} else {
		/*
		 * Keep removing sockets from the head until there's room for
		 * us to insert on the tail.  In pre-locking revisions, this
		 * was a simple if(), but as we could be racing with other
		 * threads and soabort() requires dropping locks, we must
		 * loop waiting for the condition to be true.
		 */
		while (head->sol_incqlen > head->sol_qlimit) {
			struct socket *sp;

			sp = TAILQ_FIRST(&head->sol_incomp);
			TAILQ_REMOVE(&head->sol_incomp, sp, so_list);
			head->sol_incqlen--;
			SOCK_LOCK(sp);
			sp->so_qstate = SQ_NONE;
			sp->so_listen = NULL;
			SOCK_UNLOCK(sp);
			sorele_locked(head);	/* does SOLISTEN_UNLOCK, head stays */
			soabort(sp);
			SOLISTEN_LOCK(head);
		}
		TAILQ_INSERT_TAIL(&head->sol_incomp, so, so_list);
		so->so_qstate = SQ_INCOMP;
		head->sol_incqlen++;
		SOLISTEN_UNLOCK(head);
		return (false);
	}
}

#if defined(SCTP) || defined(SCTP_SUPPORT)
/*
 * Socket part of sctp_peeloff().  Create a new socket for an
 * association.  The new socket is returned with a reference.
 *
 * XXXGL: reduce copy-paste with solisten_clone().
 */
struct socket *
sopeeloff(struct socket *head, struct protosw *so_proto)
{
	struct socket *so;

	VNET_ASSERT(head->so_vnet != NULL, ("%s:%d so_vnet is NULL, head=%p",
	    __func__, __LINE__, head));
	KASSERT(head->so_type == SOCK_SEQPACKET,
	    ("%s: unexpecte so_type: %d", __func__, head->so_type));
	so = soalloc(head->so_vnet);
	if (so == NULL) {
		log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: "
		    "limit reached or out of memory\n",
		    __func__, head->so_pcb);
		return (NULL);
	}
	so->so_type = SOCK_STREAM;
	so->so_options = head->so_options;
	so->so_linger = head->so_linger;
	so->so_state = (head->so_state & SS_NBIO) | SS_ISCONNECTED;
	so->so_fibnum = head->so_fibnum;
	so->so_proto = so_proto;
	so->so_cred = crhold(head->so_cred);
#ifdef MAC
	mac_socket_newconn(head, so);
#endif
	knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
	    so_rdknl_assert_lock);
	knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
	    so_wrknl_assert_lock);
	if (soattach(so, 0, NULL, head)) {
		sodealloc(so);
		log(LOG_DEBUG, "%s: pcb %p: pr_attach() failed\n",
		    __func__, head->so_pcb);
		return (NULL);
	}
	so->so_rcv.sb_lowat = head->so_rcv.sb_lowat;
	so->so_snd.sb_lowat = head->so_snd.sb_lowat;
	so->so_rcv.sb_timeo = head->so_rcv.sb_timeo;
	so->so_snd.sb_timeo = head->so_snd.sb_timeo;
	so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE;
	so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE;

	soref(so);

	return (so);
}
#endif	/* SCTP */

int
sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
{
	int error;

	CURVNET_SET(so->so_vnet);
	error = so->so_proto->pr_bind(so, nam, td);
	CURVNET_RESTORE();
	return (error);
}

int
sobindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
{
	int error;

	CURVNET_SET(so->so_vnet);
	error = so->so_proto->pr_bindat(fd, so, nam, td);
	CURVNET_RESTORE();
	return (error);
}

/*
 * solisten() transitions a socket from a non-listening state to a listening
 * state, but can also be used to update the listen queue depth on an
 * existing listen socket.  The protocol will call back into the sockets
 * layer using solisten_proto_check() and solisten_proto() to check and set
 * socket-layer listen state.  Call backs are used so that the protocol can
 * acquire both protocol and socket layer locks in whatever order is required
 * by the protocol.
 *
 * Protocol implementors are advised to hold the socket lock across the
 * socket-layer test and set to avoid races at the socket layer.
 */
int
solisten(struct socket *so, int backlog, struct thread *td)
{
	int error;

	CURVNET_SET(so->so_vnet);
	error = so->so_proto->pr_listen(so, backlog, td);
	CURVNET_RESTORE();
	return (error);
}

/*
 * Prepare for a call to solisten_proto().  Acquire all socket buffer locks in
 * order to interlock with socket I/O.
 */
int
solisten_proto_check(struct socket *so)
{
	SOCK_LOCK_ASSERT(so);

	if ((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
	    SS_ISDISCONNECTING)) != 0)
		return (EINVAL);

	/*
	 * Sleeping is not permitted here, so simply fail if userspace is
	 * attempting to transmit or receive on the socket.  This kind of
	 * transient failure is not ideal, but it should occur only if userspace
	 * is misusing the socket interfaces.
	 */
	if (!sx_try_xlock(&so->so_snd_sx))
		return (EAGAIN);
	if (!sx_try_xlock(&so->so_rcv_sx)) {
		sx_xunlock(&so->so_snd_sx);
		return (EAGAIN);
	}
	mtx_lock(&so->so_snd_mtx);
	mtx_lock(&so->so_rcv_mtx);

	/* Interlock with soo_aio_queue() and KTLS. */
	if (!SOLISTENING(so)) {
		bool ktls;

#ifdef KERN_TLS
		ktls = so->so_snd.sb_tls_info != NULL ||
		    so->so_rcv.sb_tls_info != NULL;
#else
		ktls = false;
#endif
		if (ktls ||
		    (so->so_snd.sb_flags & (SB_AIO | SB_AIO_RUNNING)) != 0 ||
		    (so->so_rcv.sb_flags & (SB_AIO | SB_AIO_RUNNING)) != 0) {
			solisten_proto_abort(so);
			return (EINVAL);
		}
	}

	return (0);
}

/*
 * Undo the setup done by solisten_proto_check().
 */
void
solisten_proto_abort(struct socket *so)
{
	mtx_unlock(&so->so_snd_mtx);
	mtx_unlock(&so->so_rcv_mtx);
	sx_xunlock(&so->so_snd_sx);
	sx_xunlock(&so->so_rcv_sx);
}

void
solisten_proto(struct socket *so, int backlog)
{
	int sbrcv_lowat, sbsnd_lowat;
	u_int sbrcv_hiwat, sbsnd_hiwat;
	short sbrcv_flags, sbsnd_flags;
	sbintime_t sbrcv_timeo, sbsnd_timeo;

	SOCK_LOCK_ASSERT(so);
	KASSERT((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
	    SS_ISDISCONNECTING)) == 0,
	    ("%s: bad socket state %p", __func__, so));

	if (SOLISTENING(so))
		goto listening;

	/*
	 * Change this socket to listening state.
	 */
	sbrcv_lowat = so->so_rcv.sb_lowat;
	sbsnd_lowat = so->so_snd.sb_lowat;
	sbrcv_hiwat = so->so_rcv.sb_hiwat;
	sbsnd_hiwat = so->so_snd.sb_hiwat;
	sbrcv_flags = so->so_rcv.sb_flags;
	sbsnd_flags = so->so_snd.sb_flags;
	sbrcv_timeo = so->so_rcv.sb_timeo;
	sbsnd_timeo = so->so_snd.sb_timeo;

#ifdef MAC
	mac_socketpeer_label_free(so->so_peerlabel);
#endif

	if (!(so->so_proto->pr_flags & PR_SOCKBUF)) {
		sbdestroy(so, SO_SND);
		sbdestroy(so, SO_RCV);
	}

#ifdef INVARIANTS
	bzero(&so->so_rcv,
	    sizeof(struct socket) - offsetof(struct socket, so_rcv));
#endif

	so->sol_sbrcv_lowat = sbrcv_lowat;
	so->sol_sbsnd_lowat = sbsnd_lowat;
	so->sol_sbrcv_hiwat = sbrcv_hiwat;
	so->sol_sbsnd_hiwat = sbsnd_hiwat;
	so->sol_sbrcv_flags = sbrcv_flags;
	so->sol_sbsnd_flags = sbsnd_flags;
	so->sol_sbrcv_timeo = sbrcv_timeo;
	so->sol_sbsnd_timeo = sbsnd_timeo;

	so->sol_qlen = so->sol_incqlen = 0;
	TAILQ_INIT(&so->sol_incomp);
	TAILQ_INIT(&so->sol_comp);

	so->sol_accept_filter = NULL;
	so->sol_accept_filter_arg = NULL;
	so->sol_accept_filter_str = NULL;

	so->sol_upcall = NULL;
	so->sol_upcallarg = NULL;

	so->so_options |= SO_ACCEPTCONN;

listening:
	if (backlog < 0 || backlog > V_somaxconn)
		backlog = V_somaxconn;
	so->sol_qlimit = backlog;

	mtx_unlock(&so->so_snd_mtx);
	mtx_unlock(&so->so_rcv_mtx);
	sx_xunlock(&so->so_snd_sx);
	sx_xunlock(&so->so_rcv_sx);
}

/*
 * Wakeup listeners/subsystems once we have a complete connection.
 * Enters with lock, returns unlocked.
 */
void
solisten_wakeup(struct socket *sol)
{

	if (sol->sol_upcall != NULL)
		(void )sol->sol_upcall(sol, sol->sol_upcallarg, M_NOWAIT);
	else {
		selwakeuppri(&sol->so_rdsel, PSOCK);
		KNOTE_LOCKED(&sol->so_rdsel.si_note, 0);
	}
	SOLISTEN_UNLOCK(sol);
	wakeup_one(&sol->sol_comp);
	if ((sol->so_state & SS_ASYNC) && sol->so_sigio != NULL)
		pgsigio(&sol->so_sigio, SIGIO, 0);
}

/*
 * Return single connection off a listening socket queue.  Main consumer of
 * the function is kern_accept4().  Some modules, that do their own accept
 * management also use the function.  The socket reference held by the
 * listen queue is handed to the caller.
 *
 * Listening socket must be locked on entry and is returned unlocked on
 * return.
 * The flags argument is set of accept4(2) flags and ACCEPT4_INHERIT.
 */
int
solisten_dequeue(struct socket *head, struct socket **ret, int flags)
{
	struct socket *so;
	int error;

	SOLISTEN_LOCK_ASSERT(head);

	while (!(head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp) &&
	    head->so_error == 0) {
		error = msleep(&head->sol_comp, SOCK_MTX(head), PSOCK | PCATCH,
		    "accept", 0);
		if (error != 0) {
			SOLISTEN_UNLOCK(head);
			return (error);
		}
	}
	if (head->so_error) {
		error = head->so_error;
		head->so_error = 0;
	} else if ((head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp))
		error = EWOULDBLOCK;
	else
		error = 0;
	if (error) {
		SOLISTEN_UNLOCK(head);
		return (error);
	}
	so = TAILQ_FIRST(&head->sol_comp);
	SOCK_LOCK(so);
	KASSERT(so->so_qstate == SQ_COMP,
	    ("%s: so %p not SQ_COMP", __func__, so));
	head->sol_qlen--;
	so->so_qstate = SQ_NONE;
	so->so_listen = NULL;
	TAILQ_REMOVE(&head->sol_comp, so, so_list);
	if (flags & ACCEPT4_INHERIT)
		so->so_state |= (head->so_state & SS_NBIO);
	else
		so->so_state |= (flags & SOCK_NONBLOCK) ? SS_NBIO : 0;
	SOCK_UNLOCK(so);
	sorele_locked(head);

	*ret = so;
	return (0);
}

static struct so_splice *
so_splice_alloc(off_t max)
{
	struct so_splice *sp;

	sp = uma_zalloc(splice_zone, M_WAITOK);
	sp->src = NULL;
	sp->dst = NULL;
	sp->max = max > 0 ? max : -1;
	sp->wq_index = atomic_fetchadd_32(&splice_index, 1) % splice_num_wq;
	sp->state = SPLICE_INIT;
	TIMEOUT_TASK_INIT(taskqueue_thread, &sp->timeout, 0, so_splice_timeout,
	    sp);
	return (sp);
}

static void
so_splice_free(struct so_splice *sp)
{
	KASSERT(sp->state == SPLICE_CLOSED,
	    ("so_splice_free: sp %p not closed", sp));
	uma_zfree(splice_zone, sp);
}

static void
so_splice_timeout(void *arg, int pending __unused)
{
	struct so_splice *sp;

	sp = arg;
	(void)so_unsplice(sp->src, true);
}

/*
 * Splice the output from so to the input of so2.
 */
static int
so_splice(struct socket *so, struct socket *so2, struct splice *splice)
{
	struct so_splice *sp;
	int error;

	if (splice->sp_max < 0)
		return (EINVAL);
	/* Handle only TCP for now; TODO: other streaming protos */
	if (so->so_proto->pr_protocol != IPPROTO_TCP ||
	    so2->so_proto->pr_protocol != IPPROTO_TCP)
		return (EPROTONOSUPPORT);
	if (so->so_vnet != so2->so_vnet)
		return (EINVAL);

	/* so_splice_xfer() assumes that we're using these implementations. */
	KASSERT(so->so_proto->pr_sosend == sosend_generic,
	    ("so_splice: sosend not sosend_generic"));
	KASSERT(so2->so_proto->pr_soreceive == soreceive_generic ||
	    so2->so_proto->pr_soreceive == soreceive_stream,
	    ("so_splice: soreceive not soreceive_generic/stream"));

	sp = so_splice_alloc(splice->sp_max);
	so->so_splice_sent = 0;
	sp->src = so;
	sp->dst = so2;

	error = 0;
	SOCK_LOCK(so);
	if (SOLISTENING(so))
		error = EINVAL;
	else if ((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING)) == 0)
		error = ENOTCONN;
	else if (so->so_splice != NULL)
		error = EBUSY;
	if (error != 0) {
		SOCK_UNLOCK(so);
		uma_zfree(splice_zone, sp);
		return (error);
	}
	SOCK_RECVBUF_LOCK(so);
	if (so->so_rcv.sb_tls_info != NULL) {
		SOCK_RECVBUF_UNLOCK(so);
		SOCK_UNLOCK(so);
		uma_zfree(splice_zone, sp);
		return (EINVAL);
	}
	so->so_rcv.sb_flags |= SB_SPLICED;
	so->so_splice = sp;
	soref(so);
	SOCK_RECVBUF_UNLOCK(so);
	SOCK_UNLOCK(so);

	error = 0;
	SOCK_LOCK(so2);
	if (SOLISTENING(so2))
		error = EINVAL;
	else if ((so2->so_state & (SS_ISCONNECTED | SS_ISCONNECTING)) == 0)
		error = ENOTCONN;
	else if (so2->so_splice_back != NULL)
		error = EBUSY;
	if (error != 0) {
		SOCK_UNLOCK(so2);
		mtx_lock(&sp->mtx);
		sp->dst = NULL;
		sp->state = SPLICE_EXCEPTION;
		mtx_unlock(&sp->mtx);
		so_unsplice(so, false);
		return (error);
	}
	SOCK_SENDBUF_LOCK(so2);
	if (so->so_snd.sb_tls_info != NULL) {
		SOCK_SENDBUF_UNLOCK(so2);
		SOCK_UNLOCK(so2);
		mtx_lock(&sp->mtx);
		sp->dst = NULL;
		sp->state = SPLICE_EXCEPTION;
		mtx_unlock(&sp->mtx);
		so_unsplice(so, false);
		return (EINVAL);
	}
	so2->so_snd.sb_flags |= SB_SPLICED;
	so2->so_splice_back = sp;
	soref(so2);
	mtx_lock(&sp->mtx);
	SOCK_SENDBUF_UNLOCK(so2);
	SOCK_UNLOCK(so2);

	if (splice->sp_idle.tv_sec != 0 || splice->sp_idle.tv_usec != 0) {
		taskqueue_enqueue_timeout_sbt(taskqueue_thread, &sp->timeout,
		    tvtosbt(splice->sp_idle), 0, C_PREL(4));
	}

	/*
	 * Transfer any data already present in the socket buffer.
	 */
	KASSERT(sp->state == SPLICE_INIT,
	    ("so_splice: splice %p state %d", sp, sp->state));
	sp->state = SPLICE_QUEUED;
	so_splice_xfer(sp);
	return (0);
}

static int
so_unsplice(struct socket *so, bool timeout)
{
	struct socket *so2;
	struct so_splice *sp;
	bool drain, so2rele;

	/*
	 * First unset SB_SPLICED and hide the splice structure so that
	 * wakeup routines will stop enqueuing work.  This also ensures that
	 * a only a single thread will proceed with the unsplice.
	 */
	SOCK_LOCK(so);
	if (SOLISTENING(so)) {
		SOCK_UNLOCK(so);
		return (EINVAL);
	}
	SOCK_RECVBUF_LOCK(so);
	if ((so->so_rcv.sb_flags & SB_SPLICED) == 0) {
		SOCK_RECVBUF_UNLOCK(so);
		SOCK_UNLOCK(so);
		return (ENOTCONN);
	}
	sp = so->so_splice;
	mtx_lock(&sp->mtx);
	if (sp->state == SPLICE_INIT) {
		/*
		 * A splice is in the middle of being set up.
		 */
		mtx_unlock(&sp->mtx);
		SOCK_RECVBUF_UNLOCK(so);
		SOCK_UNLOCK(so);
		return (ENOTCONN);
	}
	mtx_unlock(&sp->mtx);
	so->so_rcv.sb_flags &= ~SB_SPLICED;
	so->so_splice = NULL;
	SOCK_RECVBUF_UNLOCK(so);
	SOCK_UNLOCK(so);

	so2 = sp->dst;
	if (so2 != NULL) {
		SOCK_LOCK(so2);
		KASSERT(!SOLISTENING(so2), ("%s: so2 is listening", __func__));
		SOCK_SENDBUF_LOCK(so2);
		KASSERT((so2->so_snd.sb_flags & SB_SPLICED) != 0,
		    ("%s: so2 is not spliced", __func__));
		KASSERT(so2->so_splice_back == sp,
		    ("%s: so_splice_back != sp", __func__));
		so2->so_snd.sb_flags &= ~SB_SPLICED;
		so2rele = so2->so_splice_back != NULL;
		so2->so_splice_back = NULL;
		SOCK_SENDBUF_UNLOCK(so2);
		SOCK_UNLOCK(so2);
	}

	/*
	 * No new work is being enqueued.  The worker thread might be
	 * splicing data right now, in which case we want to wait for it to
	 * finish before proceeding.
	 */
	mtx_lock(&sp->mtx);
	switch (sp->state) {
	case SPLICE_QUEUED:
	case SPLICE_RUNNING:
		sp->state = SPLICE_CLOSING;
		while (sp->state == SPLICE_CLOSING)
			msleep(sp, &sp->mtx, PSOCK, "unsplice", 0);
		break;
	case SPLICE_INIT:
	case SPLICE_IDLE:
	case SPLICE_EXCEPTION:
		sp->state = SPLICE_CLOSED;
		break;
	default:
		__assert_unreachable();
	}
	if (!timeout) {
		drain = taskqueue_cancel_timeout(taskqueue_thread, &sp->timeout,
		    NULL) != 0;
	} else {
		drain = false;
	}
	mtx_unlock(&sp->mtx);
	if (drain)
		taskqueue_drain_timeout(taskqueue_thread, &sp->timeout);

	/*
	 * Now we hold the sole reference to the splice structure.
	 * Clean up: signal userspace and release socket references.
	 */
	sorwakeup(so);
	CURVNET_SET(so->so_vnet);
	sorele(so);
	if (so2 != NULL) {
		sowwakeup(so2);
		if (so2rele)
			sorele(so2);
	}
	CURVNET_RESTORE();
	so_splice_free(sp);
	return (0);
}

/*
 * Free socket upon release of the very last reference.
 */
static void
sofree(struct socket *so)
{
	struct protosw *pr = so->so_proto;

	SOCK_LOCK_ASSERT(so);
	KASSERT(refcount_load(&so->so_count) == 0,
	    ("%s: so %p has references", __func__, so));
	KASSERT(SOLISTENING(so) || so->so_qstate == SQ_NONE,
	    ("%s: so %p is on listen queue", __func__, so));
	KASSERT(SOLISTENING(so) || (so->so_rcv.sb_flags & SB_SPLICED) == 0,
	    ("%s: so %p rcvbuf is spliced", __func__, so));
	KASSERT(SOLISTENING(so) || (so->so_snd.sb_flags & SB_SPLICED) == 0,
	    ("%s: so %p sndbuf is spliced", __func__, so));
	KASSERT(so->so_splice == NULL && so->so_splice_back == NULL,
	    ("%s: so %p has spliced data", __func__, so));

	SOCK_UNLOCK(so);

	if (so->so_dtor != NULL)
		so->so_dtor(so);

	VNET_SO_ASSERT(so);
	if (pr->pr_detach != NULL)
		pr->pr_detach(so);

	if (!(pr->pr_flags & PR_SOCKBUF) && !SOLISTENING(so)) {
		/*
		 * From this point on, we assume that no other references to
		 * this socket exist anywhere else in the stack.  Therefore,
		 * no locks need to be acquired or held.
		 */
#ifdef INVARIANTS
		SOCK_SENDBUF_LOCK(so);
		SOCK_RECVBUF_LOCK(so);
#endif
		sbdestroy(so, SO_SND);
		sbdestroy(so, SO_RCV);
#ifdef INVARIANTS
		SOCK_SENDBUF_UNLOCK(so);
		SOCK_RECVBUF_UNLOCK(so);
#endif
		mtx_destroy(&so->so_snd_mtx);
		mtx_destroy(&so->so_rcv_mtx);
	}
	seldrain(&so->so_rdsel);
	seldrain(&so->so_wrsel);
	knlist_destroy(&so->so_rdsel.si_note);
	knlist_destroy(&so->so_wrsel.si_note);
	sodealloc(so);
}

/*
 * Release a reference on a socket while holding the socket lock.
 * Unlocks the socket lock before returning.
 */
void
sorele_locked(struct socket *so)
{
	SOCK_LOCK_ASSERT(so);
	if (refcount_release(&so->so_count))
		sofree(so);
	else
		SOCK_UNLOCK(so);
}

/*
 * Close a socket on last file table reference removal.  Initiate disconnect
 * if connected.  Free socket when disconnect complete.
 *
 * This function will sorele() the socket.  Note that soclose() may be called
 * prior to the ref count reaching zero.  The actual socket structure will
 * not be freed until the ref count reaches zero.
 */
int
soclose(struct socket *so)
{
	struct accept_queue lqueue;
	int error = 0;
	bool listening, last __diagused;

	CURVNET_SET(so->so_vnet);
	funsetown(&so->so_sigio);
	if (so->so_state & SS_ISCONNECTED) {
		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
			error = sodisconnect(so);
			if (error) {
				if (error == ENOTCONN)
					error = 0;
				goto drop;
			}
		}

		if ((so->so_options & SO_LINGER) != 0 && so->so_linger != 0) {
			if ((so->so_state & SS_ISDISCONNECTING) &&
			    (so->so_state & SS_NBIO))
				goto drop;
			while (so->so_state & SS_ISCONNECTED) {
				error = tsleep(&so->so_timeo,
				    PSOCK | PCATCH, "soclos",
				    so->so_linger * hz);
				if (error)
					break;
			}
		}
	}

drop:
	if (so->so_proto->pr_close != NULL)
		so->so_proto->pr_close(so);

	SOCK_LOCK(so);
	if ((listening = SOLISTENING(so))) {
		struct socket *sp;

		TAILQ_INIT(&lqueue);
		TAILQ_SWAP(&lqueue, &so->sol_incomp, socket, so_list);
		TAILQ_CONCAT(&lqueue, &so->sol_comp, so_list);

		so->sol_qlen = so->sol_incqlen = 0;

		TAILQ_FOREACH(sp, &lqueue, so_list) {
			SOCK_LOCK(sp);
			sp->so_qstate = SQ_NONE;
			sp->so_listen = NULL;
			SOCK_UNLOCK(sp);
			last = refcount_release(&so->so_count);
			KASSERT(!last, ("%s: released last reference for %p",
			    __func__, so));
		}
	}
	sorele_locked(so);
	if (listening) {
		struct socket *sp, *tsp;

		TAILQ_FOREACH_SAFE(sp, &lqueue, so_list, tsp)
			soabort(sp);
	}
	CURVNET_RESTORE();
	return (error);
}

/*
 * soabort() is used to abruptly tear down a connection, such as when a
 * resource limit is reached (listen queue depth exceeded), or if a listen
 * socket is closed while there are sockets waiting to be accepted.
 *
 * This interface is tricky, because it is called on an unreferenced socket,
 * and must be called only by a thread that has actually removed the socket
 * from the listen queue it was on.  Likely this thread holds the last
 * reference on the socket and soabort() will proceed with sofree().  But
 * it might be not the last, as the sockets on the listen queues are seen
 * from the protocol side.
 *
 * This interface will call into the protocol code, so must not be called
 * with any socket locks held.  Protocols do call it while holding their own
 * recursible protocol mutexes, but this is something that should be subject
 * to review in the future.
 *
 * Usually socket should have a single reference left, but this is not a
 * requirement.  In the past, when we have had named references for file
 * descriptor and protocol, we asserted that none of them are being held.
 */
void
soabort(struct socket *so)
{

	VNET_SO_ASSERT(so);

	if (so->so_proto->pr_abort != NULL)
		so->so_proto->pr_abort(so);
	SOCK_LOCK(so);
	sorele_locked(so);
}

int
soaccept(struct socket *so, struct sockaddr *sa)
{
#ifdef INVARIANTS
	u_char len = sa->sa_len;
#endif
	int error;

	CURVNET_SET(so->so_vnet);
	error = so->so_proto->pr_accept(so, sa);
	KASSERT(sa->sa_len <= len,
	    ("%s: protocol %p sockaddr overflow", __func__, so->so_proto));
	CURVNET_RESTORE();
	return (error);
}

int
sopeeraddr(struct socket *so, struct sockaddr *sa)
{
#ifdef INVARIANTS
	u_char len = sa->sa_len;
#endif
	int error;

	CURVNET_ASSERT_SET();

	error = so->so_proto->pr_peeraddr(so, sa);
	KASSERT(sa->sa_len <= len,
	    ("%s: protocol %p sockaddr overflow", __func__, so->so_proto));

	return (error);
}

int
sosockaddr(struct socket *so, struct sockaddr *sa)
{
#ifdef INVARIANTS
	u_char len = sa->sa_len;
#endif
	int error;

	CURVNET_SET(so->so_vnet);
	error = so->so_proto->pr_sockaddr(so, sa);
	KASSERT(sa->sa_len <= len,
	    ("%s: protocol %p sockaddr overflow", __func__, so->so_proto));
	CURVNET_RESTORE();

	return (error);
}

int
soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
{

	return (soconnectat(AT_FDCWD, so, nam, td));
}

int
soconnectat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
{
	int error;

	CURVNET_SET(so->so_vnet);

	/*
	 * If protocol is connection-based, can only connect once.
	 * Otherwise, if connected, try to disconnect first.  This allows
	 * user to disconnect by connecting to, e.g., a null address.
	 *
	 * Note, this check is racy and may need to be re-evaluated at the
	 * protocol layer.
	 */
	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
	    (error = sodisconnect(so)))) {
		error = EISCONN;
	} else {
		/*
		 * Prevent accumulated error from previous connection from
		 * biting us.
		 */
		so->so_error = 0;
		if (fd == AT_FDCWD) {
			error = so->so_proto->pr_connect(so, nam, td);
		} else {
			error = so->so_proto->pr_connectat(fd, so, nam, td);
		}
	}
	CURVNET_RESTORE();

	return (error);
}

int
soconnect2(struct socket *so1, struct socket *so2)
{
	int error;

	CURVNET_SET(so1->so_vnet);
	error = so1->so_proto->pr_connect2(so1, so2);
	CURVNET_RESTORE();
	return (error);
}

int
sodisconnect(struct socket *so)
{
	int error;

	if ((so->so_state & SS_ISCONNECTED) == 0)
		return (ENOTCONN);
	if (so->so_state & SS_ISDISCONNECTING)
		return (EALREADY);
	VNET_SO_ASSERT(so);
	error = so->so_proto->pr_disconnect(so);
	return (error);
}

int
sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio,
    struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
{
	long space;
	ssize_t resid;
	int clen = 0, error, dontroute;

	KASSERT(so->so_type == SOCK_DGRAM, ("sosend_dgram: !SOCK_DGRAM"));
	KASSERT(so->so_proto->pr_flags & PR_ATOMIC,
	    ("sosend_dgram: !PR_ATOMIC"));

	if (uio != NULL)
		resid = uio->uio_resid;
	else
		resid = top->m_pkthdr.len;
	/*
	 * In theory resid should be unsigned.  However, space must be
	 * signed, as it might be less than 0 if we over-committed, and we
	 * must use a signed comparison of space and resid.  On the other
	 * hand, a negative resid causes us to loop sending 0-length
	 * segments to the protocol.
	 */
	if (resid < 0) {
		error = EINVAL;
		goto out;
	}

	dontroute =
	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0;
	if (td != NULL)
		td->td_ru.ru_msgsnd++;
	if (control != NULL)
		clen = control->m_len;

	SOCKBUF_LOCK(&so->so_snd);
	if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
		SOCKBUF_UNLOCK(&so->so_snd);
		error = EPIPE;
		goto out;
	}
	if (so->so_error) {
		error = so->so_error;
		so->so_error = 0;
		SOCKBUF_UNLOCK(&so->so_snd);
		goto out;
	}
	if ((so->so_state & SS_ISCONNECTED) == 0) {
		/*
		 * `sendto' and `sendmsg' is allowed on a connection-based
		 * socket if it supports implied connect.  Return ENOTCONN if
		 * not connected and no address is supplied.
		 */
		if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
		    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
			if (!(resid == 0 && clen != 0)) {
				SOCKBUF_UNLOCK(&so->so_snd);
				error = ENOTCONN;
				goto out;
			}
		} else if (addr == NULL) {
			if (so->so_proto->pr_flags & PR_CONNREQUIRED)
				error = ENOTCONN;
			else
				error = EDESTADDRREQ;
			SOCKBUF_UNLOCK(&so->so_snd);
			goto out;
		}
	}

	/*
	 * Do we need MSG_OOB support in SOCK_DGRAM?  Signs here may be a
	 * problem and need fixing.
	 */
	space = sbspace(&so->so_snd);
	if (flags & MSG_OOB)
		space += 1024;
	space -= clen;
	SOCKBUF_UNLOCK(&so->so_snd);
	if (resid > space) {
		error = EMSGSIZE;
		goto out;
	}
	if (uio == NULL) {
		resid = 0;
		if (flags & MSG_EOR)
			top->m_flags |= M_EOR;
	} else {
		/*
		 * Copy the data from userland into a mbuf chain.
		 * If no data is to be copied in, a single empty mbuf
		 * is returned.
		 */
		top = m_uiotombuf(uio, M_WAITOK, space, max_hdr,
		    (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0)));
		if (top == NULL) {
			error = EFAULT;	/* only possible error */
			goto out;
		}
		space -= resid - uio->uio_resid;
		resid = uio->uio_resid;
	}
	KASSERT(resid == 0, ("sosend_dgram: resid != 0"));
	/*
	 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock
	 * than with.
	 */
	if (dontroute) {
		SOCK_LOCK(so);
		so->so_options |= SO_DONTROUTE;
		SOCK_UNLOCK(so);
	}
	/*
	 * XXX all the SBS_CANTSENDMORE checks previously done could be out
	 * of date.  We could have received a reset packet in an interrupt or
	 * maybe we slept while doing page faults in uiomove() etc.  We could
	 * probably recheck again inside the locking protection here, but
	 * there are probably other places that this also happens.  We must
	 * rethink this.
	 */
	VNET_SO_ASSERT(so);
	error = so->so_proto->pr_send(so, (flags & MSG_OOB) ? PRUS_OOB :
	/*
	 * If the user set MSG_EOF, the protocol understands this flag and
	 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND.
	 */
	    ((flags & MSG_EOF) &&
	     (so->so_proto->pr_flags & PR_IMPLOPCL) &&
	     (resid <= 0)) ?
		PRUS_EOF :
		/* If there is more to send set PRUS_MORETOCOME */
		(flags & MSG_MORETOCOME) ||
		(resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
		top, addr, control, td);
	if (dontroute) {
		SOCK_LOCK(so);
		so->so_options &= ~SO_DONTROUTE;
		SOCK_UNLOCK(so);
	}
	clen = 0;
	control = NULL;
	top = NULL;
out:
	if (top != NULL)
		m_freem(top);
	if (control != NULL)
		m_freem(control);
	return (error);
}

/*
 * Send on a socket.  If send must go all at once and message is larger than
 * send buffering, then hard error.  Lock against other senders.  If must go
 * all at once and not enough room now, then inform user that this would
 * block and do nothing.  Otherwise, if nonblocking, send as much as
 * possible.  The data to be sent is described by "uio" if nonzero, otherwise
 * by the mbuf chain "top" (which must be null if uio is not).  Data provided
 * in mbuf chain must be small enough to send all at once.
 *
 * Returns nonzero on error, timeout or signal; callers must check for short
 * counts if EINTR/ERESTART are returned.  Data and control buffers are freed
 * on return.
 */
static int
sosend_generic_locked(struct socket *so, struct sockaddr *addr, struct uio *uio,
    struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
{
	long space;
	ssize_t resid;
	int clen = 0, error, dontroute;
	int atomic = sosendallatonce(so) || top;
	int pr_send_flag;
#ifdef KERN_TLS
	struct ktls_session *tls;
	int tls_enq_cnt, tls_send_flag;
	uint8_t tls_rtype;

	tls = NULL;
	tls_rtype = TLS_RLTYPE_APP;
#endif

	SOCK_IO_SEND_ASSERT_LOCKED(so);

	if (uio != NULL)
		resid = uio->uio_resid;
	else if ((top->m_flags & M_PKTHDR) != 0)
		resid = top->m_pkthdr.len;
	else
		resid = m_length(top, NULL);
	/*
	 * In theory resid should be unsigned.  However, space must be
	 * signed, as it might be less than 0 if we over-committed, and we
	 * must use a signed comparison of space and resid.  On the other
	 * hand, a negative resid causes us to loop sending 0-length
	 * segments to the protocol.
	 *
	 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
	 * type sockets since that's an error.
	 */
	if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
		error = EINVAL;
		goto out;
	}

	dontroute =
	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
	    (so->so_proto->pr_flags & PR_ATOMIC);
	if (td != NULL)
		td->td_ru.ru_msgsnd++;
	if (control != NULL)
		clen = control->m_len;

#ifdef KERN_TLS
	tls_send_flag = 0;
	tls = ktls_hold(so->so_snd.sb_tls_info);
	if (tls != NULL) {
		if (tls->mode == TCP_TLS_MODE_SW)
			tls_send_flag = PRUS_NOTREADY;

		if (control != NULL) {
			struct cmsghdr *cm = mtod(control, struct cmsghdr *);

			if (clen >= sizeof(*cm) &&
			    cm->cmsg_type == TLS_SET_RECORD_TYPE) {
				tls_rtype = *((uint8_t *)CMSG_DATA(cm));
				clen = 0;
				m_freem(control);
				control = NULL;
				atomic = 1;
			}
		}

		if (resid == 0 && !ktls_permit_empty_frames(tls)) {
			error = EINVAL;
			goto out;
		}
	}
#endif

restart:
	do {
		SOCKBUF_LOCK(&so->so_snd);
		if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
			SOCKBUF_UNLOCK(&so->so_snd);
			error = EPIPE;
			goto out;
		}
		if (so->so_error) {
			error = so->so_error;
			so->so_error = 0;
			SOCKBUF_UNLOCK(&so->so_snd);
			goto out;
		}
		if ((so->so_state & SS_ISCONNECTED) == 0) {
			/*
			 * `sendto' and `sendmsg' is allowed on a connection-
			 * based socket if it supports implied connect.
			 * Return ENOTCONN if not connected and no address is
			 * supplied.
			 */
			if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
			    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
				if (!(resid == 0 && clen != 0)) {
					SOCKBUF_UNLOCK(&so->so_snd);
					error = ENOTCONN;
					goto out;
				}
			} else if (addr == NULL) {
				SOCKBUF_UNLOCK(&so->so_snd);
				if (so->so_proto->pr_flags & PR_CONNREQUIRED)
					error = ENOTCONN;
				else
					error = EDESTADDRREQ;
				goto out;
			}
		}
		space = sbspace(&so->so_snd);
		if (flags & MSG_OOB)
			space += 1024;
		if ((atomic && resid > so->so_snd.sb_hiwat) ||
		    clen > so->so_snd.sb_hiwat) {
			SOCKBUF_UNLOCK(&so->so_snd);
			error = EMSGSIZE;
			goto out;
		}
		if (space < resid + clen &&
		    (atomic || space < so->so_snd.sb_lowat || space < clen)) {
			if ((so->so_state & SS_NBIO) ||
			    (flags & (MSG_NBIO | MSG_DONTWAIT)) != 0) {
				SOCKBUF_UNLOCK(&so->so_snd);
				error = EWOULDBLOCK;
				goto out;
			}
			error = sbwait(so, SO_SND);
			SOCKBUF_UNLOCK(&so->so_snd);
			if (error)
				goto out;
			goto restart;
		}
		SOCKBUF_UNLOCK(&so->so_snd);
		space -= clen;
		do {
			if (uio == NULL) {
				resid = 0;
				if (flags & MSG_EOR)
					top->m_flags |= M_EOR;
#ifdef KERN_TLS
				if (tls != NULL) {
					ktls_frame(top, tls, &tls_enq_cnt,
					    tls_rtype);
					tls_rtype = TLS_RLTYPE_APP;
				}
#endif
			} else {
				/*
				 * Copy the data from userland into a mbuf
				 * chain.  If resid is 0, which can happen
				 * only if we have control to send, then
				 * a single empty mbuf is returned.  This
				 * is a workaround to prevent protocol send
				 * methods to panic.
				 */
#ifdef KERN_TLS
				if (tls != NULL) {
					top = m_uiotombuf(uio, M_WAITOK, space,
					    tls->params.max_frame_len,
					    M_EXTPG |
					    ((flags & MSG_EOR) ? M_EOR : 0));
					if (top != NULL) {
						ktls_frame(top, tls,
						    &tls_enq_cnt, tls_rtype);
					}
					tls_rtype = TLS_RLTYPE_APP;
				} else
#endif
					top = m_uiotombuf(uio, M_WAITOK, space,
					    (atomic ? max_hdr : 0),
					    (atomic ? M_PKTHDR : 0) |
					    ((flags & MSG_EOR) ? M_EOR : 0));
				if (top == NULL) {
					error = EFAULT; /* only possible error */
					goto out;
				}
				space -= resid - uio->uio_resid;
				resid = uio->uio_resid;
			}
			if (dontroute) {
				SOCK_LOCK(so);
				so->so_options |= SO_DONTROUTE;
				SOCK_UNLOCK(so);
			}
			/*
			 * XXX all the SBS_CANTSENDMORE checks previously
			 * done could be out of date.  We could have received
			 * a reset packet in an interrupt or maybe we slept
			 * while doing page faults in uiomove() etc.  We
			 * could probably recheck again inside the locking
			 * protection here, but there are probably other
			 * places that this also happens.  We must rethink
			 * this.
			 */
			VNET_SO_ASSERT(so);

			pr_send_flag = (flags & MSG_OOB) ? PRUS_OOB :
			/*
			 * If the user set MSG_EOF, the protocol understands
			 * this flag and nothing left to send then use
			 * PRU_SEND_EOF instead of PRU_SEND.
			 */
			    ((flags & MSG_EOF) &&
			     (so->so_proto->pr_flags & PR_IMPLOPCL) &&
			     (resid <= 0)) ?
				PRUS_EOF :
			/* If there is more to send set PRUS_MORETOCOME. */
			    (flags & MSG_MORETOCOME) ||
			    (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0;

#ifdef KERN_TLS
			pr_send_flag |= tls_send_flag;
#endif

			error = so->so_proto->pr_send(so, pr_send_flag, top,
			    addr, control, td);

			if (dontroute) {
				SOCK_LOCK(so);
				so->so_options &= ~SO_DONTROUTE;
				SOCK_UNLOCK(so);
			}

#ifdef KERN_TLS
			if (tls != NULL && tls->mode == TCP_TLS_MODE_SW) {
				if (error != 0) {
					m_freem(top);
					top = NULL;
				} else {
					soref(so);
					ktls_enqueue(top, so, tls_enq_cnt);
				}
			}
#endif
			clen = 0;
			control = NULL;
			top = NULL;
			if (error)
				goto out;
		} while (resid && space > 0);
	} while (resid);

out:
#ifdef KERN_TLS
	if (tls != NULL)
		ktls_free(tls);
#endif
	if (top != NULL)
		m_freem(top);
	if (control != NULL)
		m_freem(control);
	return (error);
}

int
sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio,
    struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
{
	int error;

	error = SOCK_IO_SEND_LOCK(so, SBLOCKWAIT(flags));
	if (error)
		return (error);
	error = sosend_generic_locked(so, addr, uio, top, control, flags, td);
	SOCK_IO_SEND_UNLOCK(so);
	return (error);
}

/*
 * Send to a socket from a kernel thread.
 *
 * XXXGL: in almost all cases uio is NULL and the mbuf is supplied.
 * Exception is nfs/bootp_subr.c.  It is arguable that the VNET context needs
 * to be set at all.  This function should just boil down to a static inline
 * calling the protocol method.
 */
int
sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
    struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
{
	int error;

	CURVNET_SET(so->so_vnet);
	error = so->so_proto->pr_sosend(so, addr, uio,
	    top, control, flags, td);
	CURVNET_RESTORE();
	return (error);
}

/*
 * send(2), write(2) or aio_write(2) on a socket.
 */
int
sousrsend(struct socket *so, struct sockaddr *addr, struct uio *uio,
    struct mbuf *control, int flags, struct proc *userproc)
{
	struct thread *td;
	ssize_t len;
	int error;

	td = uio->uio_td;
	len = uio->uio_resid;
	CURVNET_SET(so->so_vnet);
	error = so->so_proto->pr_sosend(so, addr, uio, NULL, control, flags,
	    td);
	CURVNET_RESTORE();
	if (error != 0) {
		/*
		 * Clear transient errors for stream protocols if they made
		 * some progress.  Make exclusion for aio(4) that would
		 * schedule a new write in case of EWOULDBLOCK and clear
		 * error itself.  See soaio_process_job().
		 */
		if (uio->uio_resid != len &&
		    (so->so_proto->pr_flags & PR_ATOMIC) == 0 &&
		    userproc == NULL &&
		    (error == ERESTART || error == EINTR ||
		    error == EWOULDBLOCK))
			error = 0;
		/* Generation of SIGPIPE can be controlled per socket. */
		if (error == EPIPE && (so->so_options & SO_NOSIGPIPE) == 0 &&
		    (flags & MSG_NOSIGNAL) == 0) {
			if (userproc != NULL) {
				/* aio(4) job */
				PROC_LOCK(userproc);
				kern_psignal(userproc, SIGPIPE);
				PROC_UNLOCK(userproc);
			} else {
				PROC_LOCK(td->td_proc);
				tdsignal(td, SIGPIPE);
				PROC_UNLOCK(td->td_proc);
			}
		}
	}
	return (error);
}

/*
 * The part of soreceive() that implements reading non-inline out-of-band
 * data from a socket.  For more complete comments, see soreceive(), from
 * which this code originated.
 *
 * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is
 * unable to return an mbuf chain to the caller.
 */
static int
soreceive_rcvoob(struct socket *so, struct uio *uio, int flags)
{
	struct protosw *pr = so->so_proto;
	struct mbuf *m;
	int error;

	KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
	VNET_SO_ASSERT(so);

	m = m_get(M_WAITOK, MT_DATA);
	error = pr->pr_rcvoob(so, m, flags & MSG_PEEK);
	if (error)
		goto bad;
	do {
		error = uiomove(mtod(m, void *),
		    (int) min(uio->uio_resid, m->m_len), uio);
		m = m_free(m);
	} while (uio->uio_resid && error == 0 && m);
bad:
	if (m != NULL)
		m_freem(m);
	return (error);
}

/*
 * Following replacement or removal of the first mbuf on the first mbuf chain
 * of a socket buffer, push necessary state changes back into the socket
 * buffer so that other consumers see the values consistently.  'nextrecord'
 * is the callers locally stored value of the original value of
 * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes.
 * NOTE: 'nextrecord' may be NULL.
 */
static __inline void
sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord)
{

	SOCKBUF_LOCK_ASSERT(sb);
	/*
	 * First, update for the new value of nextrecord.  If necessary, make
	 * it the first record.
	 */
	if (sb->sb_mb != NULL)
		sb->sb_mb->m_nextpkt = nextrecord;
	else
		sb->sb_mb = nextrecord;

	/*
	 * Now update any dependent socket buffer fields to reflect the new
	 * state.  This is an expanded inline of SB_EMPTY_FIXUP(), with the
	 * addition of a second clause that takes care of the case where
	 * sb_mb has been updated, but remains the last record.
	 */
	if (sb->sb_mb == NULL) {
		sb->sb_mbtail = NULL;
		sb->sb_lastrecord = NULL;
	} else if (sb->sb_mb->m_nextpkt == NULL)
		sb->sb_lastrecord = sb->sb_mb;
}

/*
 * Implement receive operations on a socket.  We depend on the way that
 * records are added to the sockbuf by sbappend.  In particular, each record
 * (mbufs linked through m_next) must begin with an address if the protocol
 * so specifies, followed by an optional mbuf or mbufs containing ancillary
 * data, and then zero or more mbufs of data.  In order to allow parallelism
 * between network receive and copying to user space, as well as avoid
 * sleeping with a mutex held, we release the socket buffer mutex during the
 * user space copy.  Although the sockbuf is locked, new data may still be
 * appended, and thus we must maintain consistency of the sockbuf during that
 * time.
 *
 * The caller may receive the data as a single mbuf chain by supplying an
 * mbuf **mp for use in returning the chain.  The uio is then used only for
 * the count in uio_resid.
 */
static int
soreceive_generic_locked(struct socket *so, struct sockaddr **psa,
    struct uio *uio, struct mbuf **mp, struct mbuf **controlp, int *flagsp)
{
	struct mbuf *m;
	int flags, error, offset;
	ssize_t len;
	struct protosw *pr = so->so_proto;
	struct mbuf *nextrecord;
	int moff, type = 0;
	ssize_t orig_resid = uio->uio_resid;
	bool report_real_len = false;

	SOCK_IO_RECV_ASSERT_LOCKED(so);

	error = 0;
	if (flagsp != NULL) {
		report_real_len = *flagsp & MSG_TRUNC;
		*flagsp &= ~MSG_TRUNC;
		flags = *flagsp &~ MSG_EOR;
	} else
		flags = 0;

restart:
	SOCKBUF_LOCK(&so->so_rcv);
	m = so->so_rcv.sb_mb;
	/*
	 * If we have less data than requested, block awaiting more (subject
	 * to any timeout) if:
	 *   1. the current count is less than the low water mark, or
	 *   2. MSG_DONTWAIT is not set
	 */
	if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
	    sbavail(&so->so_rcv) < uio->uio_resid) &&
	    sbavail(&so->so_rcv) < so->so_rcv.sb_lowat &&
	    m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
		KASSERT(m != NULL || !sbavail(&so->so_rcv),
		    ("receive: m == %p sbavail == %u",
		    m, sbavail(&so->so_rcv)));
		if (so->so_error || so->so_rerror) {
			if (m != NULL)
				goto dontblock;
			if (so->so_error)
				error = so->so_error;
			else
				error = so->so_rerror;
			if ((flags & MSG_PEEK) == 0) {
				if (so->so_error)
					so->so_error = 0;
				else
					so->so_rerror = 0;
			}
			SOCKBUF_UNLOCK(&so->so_rcv);
			goto release;
		}
		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
		if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
			if (m != NULL)
				goto dontblock;
#ifdef KERN_TLS
			else if (so->so_rcv.sb_tlsdcc == 0 &&
			    so->so_rcv.sb_tlscc == 0) {
#else
			else {
#endif
				SOCKBUF_UNLOCK(&so->so_rcv);
				goto release;
			}
		}
		for (; m != NULL; m = m->m_next)
			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
				m = so->so_rcv.sb_mb;
				goto dontblock;
			}
		if ((so->so_state & (SS_ISCONNECTING | SS_ISCONNECTED |
		    SS_ISDISCONNECTING | SS_ISDISCONNECTED)) == 0 &&
		    (so->so_proto->pr_flags & PR_CONNREQUIRED) != 0) {
			SOCKBUF_UNLOCK(&so->so_rcv);
			error = ENOTCONN;
			goto release;
		}
		if (uio->uio_resid == 0 && !report_real_len) {
			SOCKBUF_UNLOCK(&so->so_rcv);
			goto release;
		}
		if ((so->so_state & SS_NBIO) ||
		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
			SOCKBUF_UNLOCK(&so->so_rcv);
			error = EWOULDBLOCK;
			goto release;
		}
		SBLASTRECORDCHK(&so->so_rcv);
		SBLASTMBUFCHK(&so->so_rcv);
		error = sbwait(so, SO_RCV);
		SOCKBUF_UNLOCK(&so->so_rcv);
		if (error)
			goto release;
		goto restart;
	}
dontblock:
	/*
	 * From this point onward, we maintain 'nextrecord' as a cache of the
	 * pointer to the next record in the socket buffer.  We must keep the
	 * various socket buffer pointers and local stack versions of the
	 * pointers in sync, pushing out modifications before dropping the
	 * socket buffer mutex, and re-reading them when picking it up.
	 *
	 * Otherwise, we will race with the network stack appending new data
	 * or records onto the socket buffer by using inconsistent/stale
	 * versions of the field, possibly resulting in socket buffer
	 * corruption.
	 *
	 * By holding the high-level sblock(), we prevent simultaneous
	 * readers from pulling off the front of the socket buffer.
	 */
	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
	if (uio->uio_td)
		uio->uio_td->td_ru.ru_msgrcv++;
	KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb"));
	SBLASTRECORDCHK(&so->so_rcv);
	SBLASTMBUFCHK(&so->so_rcv);
	nextrecord = m->m_nextpkt;
	if (pr->pr_flags & PR_ADDR) {
		KASSERT(m->m_type == MT_SONAME,
		    ("m->m_type == %d", m->m_type));
		orig_resid = 0;
		if (psa != NULL)
			*psa = sodupsockaddr(mtod(m, struct sockaddr *),
			    M_NOWAIT);
		if (flags & MSG_PEEK) {
			m = m->m_next;
		} else {
			sbfree(&so->so_rcv, m);
			so->so_rcv.sb_mb = m_free(m);
			m = so->so_rcv.sb_mb;
			sockbuf_pushsync(&so->so_rcv, nextrecord);
		}
	}

	/*
	 * Process one or more MT_CONTROL mbufs present before any data mbufs
	 * in the first mbuf chain on the socket buffer.  If MSG_PEEK, we
	 * just copy the data; if !MSG_PEEK, we call into the protocol to
	 * perform externalization (or freeing if controlp == NULL).
	 */
	if (m != NULL && m->m_type == MT_CONTROL) {
		struct mbuf *cm = NULL, *cmn;
		struct mbuf **cme = &cm;
#ifdef KERN_TLS
		struct cmsghdr *cmsg;
		struct tls_get_record tgr;

		/*
		 * For MSG_TLSAPPDATA, check for an alert record.
		 * If found, return ENXIO without removing
		 * it from the receive queue.  This allows a subsequent
		 * call without MSG_TLSAPPDATA to receive it.
		 * Note that, for TLS, there should only be a single
		 * control mbuf with the TLS_GET_RECORD message in it.
		 */
		if (flags & MSG_TLSAPPDATA) {
			cmsg = mtod(m, struct cmsghdr *);
			if (cmsg->cmsg_type == TLS_GET_RECORD &&
			    cmsg->cmsg_len == CMSG_LEN(sizeof(tgr))) {
				memcpy(&tgr, CMSG_DATA(cmsg), sizeof(tgr));
				if (__predict_false(tgr.tls_type ==
				    TLS_RLTYPE_ALERT)) {
					SOCKBUF_UNLOCK(&so->so_rcv);
					error = ENXIO;
					goto release;
				}
			}
		}
#endif

		do {
			if (flags & MSG_PEEK) {
				if (controlp != NULL) {
					*controlp = m_copym(m, 0, m->m_len,
					    M_NOWAIT);
					controlp = &(*controlp)->m_next;
				}
				m = m->m_next;
			} else {
				sbfree(&so->so_rcv, m);
				so->so_rcv.sb_mb = m->m_next;
				m->m_next = NULL;
				*cme = m;
				cme = &(*cme)->m_next;
				m = so->so_rcv.sb_mb;
			}
		} while (m != NULL && m->m_type == MT_CONTROL);
		if ((flags & MSG_PEEK) == 0)
			sockbuf_pushsync(&so->so_rcv, nextrecord);
		while (cm != NULL) {
			cmn = cm->m_next;
			cm->m_next = NULL;
			if (controlp != NULL)
				*controlp = cm;
			else
				m_freem(cm);
			if (controlp != NULL) {
				while (*controlp != NULL)
					controlp = &(*controlp)->m_next;
			}
			cm = cmn;
		}
		if (m != NULL)
			nextrecord = so->so_rcv.sb_mb->m_nextpkt;
		else
			nextrecord = so->so_rcv.sb_mb;
		orig_resid = 0;
	}
	if (m != NULL) {
		if ((flags & MSG_PEEK) == 0) {
			KASSERT(m->m_nextpkt == nextrecord,
			    ("soreceive: post-control, nextrecord !sync"));
			if (nextrecord == NULL) {
				KASSERT(so->so_rcv.sb_mb == m,
				    ("soreceive: post-control, sb_mb!=m"));
				KASSERT(so->so_rcv.sb_lastrecord == m,
				    ("soreceive: post-control, lastrecord!=m"));
			}
		}
		type = m->m_type;
		if (type == MT_OOBDATA)
			flags |= MSG_OOB;
	} else {
		if ((flags & MSG_PEEK) == 0) {
			KASSERT(so->so_rcv.sb_mb == nextrecord,
			    ("soreceive: sb_mb != nextrecord"));
			if (so->so_rcv.sb_mb == NULL) {
				KASSERT(so->so_rcv.sb_lastrecord == NULL,
				    ("soreceive: sb_lastercord != NULL"));
			}
		}
	}
	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
	SBLASTRECORDCHK(&so->so_rcv);
	SBLASTMBUFCHK(&so->so_rcv);

	/*
	 * Now continue to read any data mbufs off of the head of the socket
	 * buffer until the read request is satisfied.  Note that 'type' is
	 * used to store the type of any mbuf reads that have happened so far
	 * such that soreceive() can stop reading if the type changes, which
	 * causes soreceive() to return only one of regular data and inline
	 * out-of-band data in a single socket receive operation.
	 */
	moff = 0;
	offset = 0;
	while (m != NULL && !(m->m_flags & M_NOTREADY) && uio->uio_resid > 0 &&
	    error == 0) {
		/*
		 * If the type of mbuf has changed since the last mbuf
		 * examined ('type'), end the receive operation.
		 */
		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
		if (m->m_type == MT_OOBDATA || m->m_type == MT_CONTROL) {
			if (type != m->m_type)
				break;
		} else if (type == MT_OOBDATA)
			break;
		else
		    KASSERT(m->m_type == MT_DATA,
			("m->m_type == %d", m->m_type));
		so->so_rcv.sb_state &= ~SBS_RCVATMARK;
		len = uio->uio_resid;
		if (so->so_oobmark && len > so->so_oobmark - offset)
			len = so->so_oobmark - offset;
		if (len > m->m_len - moff)
			len = m->m_len - moff;
		/*
		 * If mp is set, just pass back the mbufs.  Otherwise copy
		 * them out via the uio, then free.  Sockbuf must be
		 * consistent here (points to current mbuf, it points to next
		 * record) when we drop priority; we must note any additions
		 * to the sockbuf when we block interrupts again.
		 */
		if (mp == NULL) {
			SOCKBUF_LOCK_ASSERT(&so->so_rcv);
			SBLASTRECORDCHK(&so->so_rcv);
			SBLASTMBUFCHK(&so->so_rcv);
			SOCKBUF_UNLOCK(&so->so_rcv);
			if ((m->m_flags & M_EXTPG) != 0)
				error = m_unmapped_uiomove(m, moff, uio,
				    (int)len);
			else
				error = uiomove(mtod(m, char *) + moff,
				    (int)len, uio);
			SOCKBUF_LOCK(&so->so_rcv);
			if (error) {
				/*
				 * The MT_SONAME mbuf has already been removed
				 * from the record, so it is necessary to
				 * remove the data mbufs, if any, to preserve
				 * the invariant in the case of PR_ADDR that
				 * requires MT_SONAME mbufs at the head of
				 * each record.
				 */
				if (pr->pr_flags & PR_ATOMIC &&
				    ((flags & MSG_PEEK) == 0))
					(void)sbdroprecord_locked(&so->so_rcv);
				SOCKBUF_UNLOCK(&so->so_rcv);
				goto release;
			}
		} else
			uio->uio_resid -= len;
		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
		if (len == m->m_len - moff) {
			if (m->m_flags & M_EOR)
				flags |= MSG_EOR;
			if (flags & MSG_PEEK) {
				m = m->m_next;
				moff = 0;
			} else {
				nextrecord = m->m_nextpkt;
				sbfree(&so->so_rcv, m);
				if (mp != NULL) {
					m->m_nextpkt = NULL;
					*mp = m;
					mp = &m->m_next;
					so->so_rcv.sb_mb = m = m->m_next;
					*mp = NULL;
				} else {
					so->so_rcv.sb_mb = m_free(m);
					m = so->so_rcv.sb_mb;
				}
				sockbuf_pushsync(&so->so_rcv, nextrecord);
				SBLASTRECORDCHK(&so->so_rcv);
				SBLASTMBUFCHK(&so->so_rcv);
			}
		} else {
			if (flags & MSG_PEEK)
				moff += len;
			else {
				if (mp != NULL) {
					if (flags & MSG_DONTWAIT) {
						*mp = m_copym(m, 0, len,
						    M_NOWAIT);
						if (*mp == NULL) {
							/*
							 * m_copym() couldn't
							 * allocate an mbuf.
							 * Adjust uio_resid back
							 * (it was adjusted
							 * down by len bytes,
							 * which we didn't end
							 * up "copying" over).
							 */
							uio->uio_resid += len;
							break;
						}
					} else {
						SOCKBUF_UNLOCK(&so->so_rcv);
						*mp = m_copym(m, 0, len,
						    M_WAITOK);
						SOCKBUF_LOCK(&so->so_rcv);
					}
				}
				sbcut_locked(&so->so_rcv, len);
			}
		}
		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
		if (so->so_oobmark) {
			if ((flags & MSG_PEEK) == 0) {
				so->so_oobmark -= len;
				if (so->so_oobmark == 0) {
					so->so_rcv.sb_state |= SBS_RCVATMARK;
					break;
				}
			} else {
				offset += len;
				if (offset == so->so_oobmark)
					break;
			}
		}
		if (flags & MSG_EOR)
			break;
		/*
		 * If the MSG_WAITALL flag is set (for non-atomic socket), we
		 * must not quit until "uio->uio_resid == 0" or an error
		 * termination.  If a signal/timeout occurs, return with a
		 * short count but without error.  Keep sockbuf locked
		 * against other readers.
		 */
		while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
		    !sosendallatonce(so) && nextrecord == NULL) {
			SOCKBUF_LOCK_ASSERT(&so->so_rcv);
			if (so->so_error || so->so_rerror ||
			    so->so_rcv.sb_state & SBS_CANTRCVMORE)
				break;
			/*
			 * Notify the protocol that some data has been
			 * drained before blocking.
			 */
			if (pr->pr_flags & PR_WANTRCVD) {
				SOCKBUF_UNLOCK(&so->so_rcv);
				VNET_SO_ASSERT(so);
				pr->pr_rcvd(so, flags);
				SOCKBUF_LOCK(&so->so_rcv);
				if (__predict_false(so->so_rcv.sb_mb == NULL &&
				    (so->so_error || so->so_rerror ||
				    so->so_rcv.sb_state & SBS_CANTRCVMORE)))
					break;
			}
			SBLASTRECORDCHK(&so->so_rcv);
			SBLASTMBUFCHK(&so->so_rcv);
			/*
			 * We could receive some data while was notifying
			 * the protocol. Skip blocking in this case.
			 */
			if (so->so_rcv.sb_mb == NULL) {
				error = sbwait(so, SO_RCV);
				if (error) {
					SOCKBUF_UNLOCK(&so->so_rcv);
					goto release;
				}
			}
			m = so->so_rcv.sb_mb;
			if (m != NULL)
				nextrecord = m->m_nextpkt;
		}
	}

	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
	if (m != NULL && pr->pr_flags & PR_ATOMIC) {
		if (report_real_len)
			uio->uio_resid -= m_length(m, NULL) - moff;
		flags |= MSG_TRUNC;
		if ((flags & MSG_PEEK) == 0)
			(void) sbdroprecord_locked(&so->so_rcv);
	}
	if ((flags & MSG_PEEK) == 0) {
		if (m == NULL) {
			/*
			 * First part is an inline SB_EMPTY_FIXUP().  Second
			 * part makes sure sb_lastrecord is up-to-date if
			 * there is still data in the socket buffer.
			 */
			so->so_rcv.sb_mb = nextrecord;
			if (so->so_rcv.sb_mb == NULL) {
				so->so_rcv.sb_mbtail = NULL;
				so->so_rcv.sb_lastrecord = NULL;
			} else if (nextrecord->m_nextpkt == NULL)
				so->so_rcv.sb_lastrecord = nextrecord;
		}
		SBLASTRECORDCHK(&so->so_rcv);
		SBLASTMBUFCHK(&so->so_rcv);
		/*
		 * If soreceive() is being done from the socket callback,
		 * then don't need to generate ACK to peer to update window,
		 * since ACK will be generated on return to TCP.
		 */
		if (!(flags & MSG_SOCALLBCK) &&
		    (pr->pr_flags & PR_WANTRCVD)) {
			SOCKBUF_UNLOCK(&so->so_rcv);
			VNET_SO_ASSERT(so);
			pr->pr_rcvd(so, flags);
			SOCKBUF_LOCK(&so->so_rcv);
		}
	}
	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
	if (orig_resid == uio->uio_resid && orig_resid &&
	    (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) {
		SOCKBUF_UNLOCK(&so->so_rcv);
		goto restart;
	}
	SOCKBUF_UNLOCK(&so->so_rcv);

	if (flagsp != NULL)
		*flagsp |= flags;
release:
	return (error);
}

int
soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio,
    struct mbuf **mp, struct mbuf **controlp, int *flagsp)
{
	int error, flags;

	if (psa != NULL)
		*psa = NULL;
	if (controlp != NULL)
		*controlp = NULL;
	if (flagsp != NULL) {
		flags = *flagsp;
		if ((flags & MSG_OOB) != 0)
			return (soreceive_rcvoob(so, uio, flags));
	} else {
		flags = 0;
	}
	if (mp != NULL)
		*mp = NULL;

	error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags));
	if (error)
		return (error);
	error = soreceive_generic_locked(so, psa, uio, mp, controlp, flagsp);
	SOCK_IO_RECV_UNLOCK(so);
	return (error);
}

/*
 * Optimized version of soreceive() for stream (TCP) sockets.
 */
static int
soreceive_stream_locked(struct socket *so, struct sockbuf *sb,
    struct sockaddr **psa, struct uio *uio, struct mbuf **mp0,
    struct mbuf **controlp, int flags)
{
	int len = 0, error = 0, oresid;
	struct mbuf *m, *n = NULL;

	SOCK_IO_RECV_ASSERT_LOCKED(so);

	/* Easy one, no space to copyout anything. */
	if (uio->uio_resid == 0)
		return (EINVAL);
	oresid = uio->uio_resid;

	SOCKBUF_LOCK(sb);
	/* We will never ever get anything unless we are or were connected. */
	if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) {
		error = ENOTCONN;
		goto out;
	}

restart:
	SOCKBUF_LOCK_ASSERT(&so->so_rcv);

	/* Abort if socket has reported problems. */
	if (so->so_error) {
		if (sbavail(sb) > 0)
			goto deliver;
		if (oresid > uio->uio_resid)
			goto out;
		error = so->so_error;
		if (!(flags & MSG_PEEK))
			so->so_error = 0;
		goto out;
	}

	/* Door is closed.  Deliver what is left, if any. */
	if (sb->sb_state & SBS_CANTRCVMORE) {
		if (sbavail(sb) > 0)
			goto deliver;
		else
			goto out;
	}

	/* Socket buffer is empty and we shall not block. */
	if (sbavail(sb) == 0 &&
	    ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) {
		error = EAGAIN;
		goto out;
	}

	/* Socket buffer got some data that we shall deliver now. */
	if (sbavail(sb) > 0 && !(flags & MSG_WAITALL) &&
	    ((so->so_state & SS_NBIO) ||
	     (flags & (MSG_DONTWAIT|MSG_NBIO)) ||
	     sbavail(sb) >= sb->sb_lowat ||
	     sbavail(sb) >= uio->uio_resid ||
	     sbavail(sb) >= sb->sb_hiwat) ) {
		goto deliver;
	}

	/* On MSG_WAITALL we must wait until all data or error arrives. */
	if ((flags & MSG_WAITALL) &&
	    (sbavail(sb) >= uio->uio_resid || sbavail(sb) >= sb->sb_hiwat))
		goto deliver;

	/*
	 * Wait and block until (more) data comes in.
	 * NB: Drops the sockbuf lock during wait.
	 */
	error = sbwait(so, SO_RCV);
	if (error)
		goto out;
	goto restart;

deliver:
	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
	KASSERT(sbavail(sb) > 0, ("%s: sockbuf empty", __func__));
	KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__));

	/* Statistics. */
	if (uio->uio_td)
		uio->uio_td->td_ru.ru_msgrcv++;

	/* Fill uio until full or current end of socket buffer is reached. */
	len = min(uio->uio_resid, sbavail(sb));
	if (mp0 != NULL) {
		/* Dequeue as many mbufs as possible. */
		if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) {
			if (*mp0 == NULL)
				*mp0 = sb->sb_mb;
			else
				m_cat(*mp0, sb->sb_mb);
			for (m = sb->sb_mb;
			     m != NULL && m->m_len <= len;
			     m = m->m_next) {
				KASSERT(!(m->m_flags & M_NOTREADY),
				    ("%s: m %p not available", __func__, m));
				len -= m->m_len;
				uio->uio_resid -= m->m_len;
				sbfree(sb, m);
				n = m;
			}
			n->m_next = NULL;
			sb->sb_mb = m;
			sb->sb_lastrecord = sb->sb_mb;
			if (sb->sb_mb == NULL)
				SB_EMPTY_FIXUP(sb);
		}
		/* Copy the remainder. */
		if (len > 0) {
			KASSERT(sb->sb_mb != NULL,
			    ("%s: len > 0 && sb->sb_mb empty", __func__));

			m = m_copym(sb->sb_mb, 0, len, M_NOWAIT);
			if (m == NULL)
				len = 0;	/* Don't flush data from sockbuf. */
			else
				uio->uio_resid -= len;
			if (*mp0 != NULL)
				m_cat(*mp0, m);
			else
				*mp0 = m;
			if (*mp0 == NULL) {
				error = ENOBUFS;
				goto out;
			}
		}
	} else {
		/* NB: Must unlock socket buffer as uiomove may sleep. */
		SOCKBUF_UNLOCK(sb);
		error = m_mbuftouio(uio, sb->sb_mb, len);
		SOCKBUF_LOCK(sb);
		if (error)
			goto out;
	}
	SBLASTRECORDCHK(sb);
	SBLASTMBUFCHK(sb);

	/*
	 * Remove the delivered data from the socket buffer unless we
	 * were only peeking.
	 */
	if (!(flags & MSG_PEEK)) {
		if (len > 0)
			sbdrop_locked(sb, len);

		/* Notify protocol that we drained some data. */
		if ((so->so_proto->pr_flags & PR_WANTRCVD) &&
		    (((flags & MSG_WAITALL) && uio->uio_resid > 0) ||
		     !(flags & MSG_SOCALLBCK))) {
			SOCKBUF_UNLOCK(sb);
			VNET_SO_ASSERT(so);
			so->so_proto->pr_rcvd(so, flags);
			SOCKBUF_LOCK(sb);
		}
	}

	/*
	 * For MSG_WAITALL we may have to loop again and wait for
	 * more data to come in.
	 */
	if ((flags & MSG_WAITALL) && uio->uio_resid > 0)
		goto restart;
out:
	SBLASTRECORDCHK(sb);
	SBLASTMBUFCHK(sb);
	SOCKBUF_UNLOCK(sb);
	return (error);
}

int
soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio,
    struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
{
	struct sockbuf *sb;
	int error, flags;

	sb = &so->so_rcv;

	/* We only do stream sockets. */
	if (so->so_type != SOCK_STREAM)
		return (EINVAL);
	if (psa != NULL)
		*psa = NULL;
	if (flagsp != NULL)
		flags = *flagsp & ~MSG_EOR;
	else
		flags = 0;
	if (controlp != NULL)
		*controlp = NULL;
	if (flags & MSG_OOB)
		return (soreceive_rcvoob(so, uio, flags));
	if (mp0 != NULL)
		*mp0 = NULL;

#ifdef KERN_TLS
	/*
	 * KTLS store TLS records as records with a control message to
	 * describe the framing.
	 *
	 * We check once here before acquiring locks to optimize the
	 * common case.
	 */
	if (sb->sb_tls_info != NULL)
		return (soreceive_generic(so, psa, uio, mp0, controlp,
		    flagsp));
#endif

	/*
	 * Prevent other threads from reading from the socket.  This lock may be
	 * dropped in order to sleep waiting for data to arrive.
	 */
	error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags));
	if (error)
		return (error);
#ifdef KERN_TLS
	if (__predict_false(sb->sb_tls_info != NULL)) {
		SOCK_IO_RECV_UNLOCK(so);
		return (soreceive_generic(so, psa, uio, mp0, controlp,
		    flagsp));
	}
#endif
	error = soreceive_stream_locked(so, sb, psa, uio, mp0, controlp, flags);
	SOCK_IO_RECV_UNLOCK(so);
	return (error);
}

/*
 * Optimized version of soreceive() for simple datagram cases from userspace.
 * Unlike in the stream case, we're able to drop a datagram if copyout()
 * fails, and because we handle datagrams atomically, we don't need to use a
 * sleep lock to prevent I/O interlacing.
 */
int
soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio,
    struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
{
	struct mbuf *m, *m2;
	int flags, error;
	ssize_t len;
	struct protosw *pr = so->so_proto;
	struct mbuf *nextrecord;

	if (psa != NULL)
		*psa = NULL;
	if (controlp != NULL)
		*controlp = NULL;
	if (flagsp != NULL)
		flags = *flagsp &~ MSG_EOR;
	else
		flags = 0;

	/*
	 * For any complicated cases, fall back to the full
	 * soreceive_generic().
	 */
	if (mp0 != NULL || (flags & (MSG_PEEK | MSG_OOB | MSG_TRUNC)))
		return (soreceive_generic(so, psa, uio, mp0, controlp,
		    flagsp));

	/*
	 * Enforce restrictions on use.
	 */
	KASSERT((pr->pr_flags & PR_WANTRCVD) == 0,
	    ("soreceive_dgram: wantrcvd"));
	KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic"));
	KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0,
	    ("soreceive_dgram: SBS_RCVATMARK"));
	KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0,
	    ("soreceive_dgram: P_CONNREQUIRED"));

	/*
	 * Loop blocking while waiting for a datagram.
	 */
	SOCKBUF_LOCK(&so->so_rcv);
	while ((m = so->so_rcv.sb_mb) == NULL) {
		KASSERT(sbavail(&so->so_rcv) == 0,
		    ("soreceive_dgram: sb_mb NULL but sbavail %u",
		    sbavail(&so->so_rcv)));
		if (so->so_error) {
			error = so->so_error;
			so->so_error = 0;
			SOCKBUF_UNLOCK(&so->so_rcv);
			return (error);
		}
		if (so->so_rcv.sb_state & SBS_CANTRCVMORE ||
		    uio->uio_resid == 0) {
			SOCKBUF_UNLOCK(&so->so_rcv);
			return (0);
		}
		if ((so->so_state & SS_NBIO) ||
		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
			SOCKBUF_UNLOCK(&so->so_rcv);
			return (EWOULDBLOCK);
		}
		SBLASTRECORDCHK(&so->so_rcv);
		SBLASTMBUFCHK(&so->so_rcv);
		error = sbwait(so, SO_RCV);
		if (error) {
			SOCKBUF_UNLOCK(&so->so_rcv);
			return (error);
		}
	}
	SOCKBUF_LOCK_ASSERT(&so->so_rcv);

	if (uio->uio_td)
		uio->uio_td->td_ru.ru_msgrcv++;
	SBLASTRECORDCHK(&so->so_rcv);
	SBLASTMBUFCHK(&so->so_rcv);
	nextrecord = m->m_nextpkt;
	if (nextrecord == NULL) {
		KASSERT(so->so_rcv.sb_lastrecord == m,
		    ("soreceive_dgram: lastrecord != m"));
	}

	KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord,
	    ("soreceive_dgram: m_nextpkt != nextrecord"));

	/*
	 * Pull 'm' and its chain off the front of the packet queue.
	 */
	so->so_rcv.sb_mb = NULL;
	sockbuf_pushsync(&so->so_rcv, nextrecord);

	/*
	 * Walk 'm's chain and free that many bytes from the socket buffer.
	 */
	for (m2 = m; m2 != NULL; m2 = m2->m_next)
		sbfree(&so->so_rcv, m2);

	/*
	 * Do a few last checks before we let go of the lock.
	 */
	SBLASTRECORDCHK(&so->so_rcv);
	SBLASTMBUFCHK(&so->so_rcv);
	SOCKBUF_UNLOCK(&so->so_rcv);

	if (pr->pr_flags & PR_ADDR) {
		KASSERT(m->m_type == MT_SONAME,
		    ("m->m_type == %d", m->m_type));
		if (psa != NULL)
			*psa = sodupsockaddr(mtod(m, struct sockaddr *),
			    M_WAITOK);
		m = m_free(m);
	}
	KASSERT(m, ("%s: no data or control after soname", __func__));

	/*
	 * Packet to copyout() is now in 'm' and it is disconnected from the
	 * queue.
	 *
	 * Process one or more MT_CONTROL mbufs present before any data mbufs
	 * in the first mbuf chain on the socket buffer.  We call into the
	 * protocol to perform externalization (or freeing if controlp ==
	 * NULL). In some cases there can be only MT_CONTROL mbufs without
	 * MT_DATA mbufs.
	 */
	if (m->m_type == MT_CONTROL) {
		struct mbuf *cm = NULL, *cmn;
		struct mbuf **cme = &cm;

		do {
			m2 = m->m_next;
			m->m_next = NULL;
			*cme = m;
			cme = &(*cme)->m_next;
			m = m2;
		} while (m != NULL && m->m_type == MT_CONTROL);
		while (cm != NULL) {
			cmn = cm->m_next;
			cm->m_next = NULL;
			if (controlp != NULL)
				*controlp = cm;
			else
				m_freem(cm);
			if (controlp != NULL) {
				while (*controlp != NULL)
					controlp = &(*controlp)->m_next;
			}
			cm = cmn;
		}
	}
	KASSERT(m == NULL || m->m_type == MT_DATA,
	    ("soreceive_dgram: !data"));
	while (m != NULL && uio->uio_resid > 0) {
		len = uio->uio_resid;
		if (len > m->m_len)
			len = m->m_len;
		error = uiomove(mtod(m, char *), (int)len, uio);
		if (error) {
			m_freem(m);
			return (error);
		}
		if (len == m->m_len)
			m = m_free(m);
		else {
			m->m_data += len;
			m->m_len -= len;
		}
	}
	if (m != NULL) {
		flags |= MSG_TRUNC;
		m_freem(m);
	}
	if (flagsp != NULL)
		*flagsp |= flags;
	return (0);
}

int
soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
    struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
{
	int error;

	CURVNET_SET(so->so_vnet);
	error = so->so_proto->pr_soreceive(so, psa, uio, mp0, controlp, flagsp);
	CURVNET_RESTORE();
	return (error);
}

int
soshutdown(struct socket *so, enum shutdown_how how)
{
	int error;

	CURVNET_SET(so->so_vnet);
	error = so->so_proto->pr_shutdown(so, how);
	CURVNET_RESTORE();

	return (error);
}

/*
 * Used by several pr_shutdown implementations that use generic socket buffers.
 */
void
sorflush(struct socket *so)
{
	int error;

	VNET_SO_ASSERT(so);

	/*
	 * Dislodge threads currently blocked in receive and wait to acquire
	 * a lock against other simultaneous readers before clearing the
	 * socket buffer.  Don't let our acquire be interrupted by a signal
	 * despite any existing socket disposition on interruptable waiting.
	 *
	 * The SOCK_IO_RECV_LOCK() is important here as there some pr_soreceive
	 * methods that read the top of the socket buffer without acquisition
	 * of the socket buffer mutex, assuming that top of the buffer
	 * exclusively belongs to the read(2) syscall.  This is handy when
	 * performing MSG_PEEK.
	 */
	socantrcvmore(so);

	error = SOCK_IO_RECV_LOCK(so, SBL_WAIT | SBL_NOINTR);
	if (error != 0) {
		KASSERT(SOLISTENING(so),
		    ("%s: soiolock(%p) failed", __func__, so));
		return;
	}

	sbrelease(so, SO_RCV);
	SOCK_IO_RECV_UNLOCK(so);

}

int
sosetfib(struct socket *so, int fibnum)
{
	if (fibnum < 0 || fibnum >= rt_numfibs)
		return (EINVAL);

	SOCK_LOCK(so);
	so->so_fibnum = fibnum;
	SOCK_UNLOCK(so);

	return (0);
}

#ifdef SOCKET_HHOOK
/*
 * Wrapper for Socket established helper hook.
 * Parameters: socket, context of the hook point, hook id.
 */
static inline int
hhook_run_socket(struct socket *so, void *hctx, int32_t h_id)
{
	struct socket_hhook_data hhook_data = {
		.so = so,
		.hctx = hctx,
		.m = NULL,
		.status = 0
	};

	CURVNET_SET(so->so_vnet);
	HHOOKS_RUN_IF(V_socket_hhh[h_id], &hhook_data, &so->osd);
	CURVNET_RESTORE();

	/* Ugly but needed, since hhooks return void for now */
	return (hhook_data.status);
}
#endif

/*
 * Perhaps this routine, and sooptcopyout(), below, ought to come in an
 * additional variant to handle the case where the option value needs to be
 * some kind of integer, but not a specific size.  In addition to their use
 * here, these functions are also called by the protocol-level pr_ctloutput()
 * routines.
 */
int
sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
{
	size_t	valsize;

	/*
	 * If the user gives us more than we wanted, we ignore it, but if we
	 * don't get the minimum length the caller wants, we return EINVAL.
	 * On success, sopt->sopt_valsize is set to however much we actually
	 * retrieved.
	 */
	if ((valsize = sopt->sopt_valsize) < minlen)
		return EINVAL;
	if (valsize > len)
		sopt->sopt_valsize = valsize = len;

	if (sopt->sopt_td != NULL)
		return (copyin(sopt->sopt_val, buf, valsize));

	bcopy(sopt->sopt_val, buf, valsize);
	return (0);
}

/*
 * Kernel version of setsockopt(2).
 *
 * XXX: optlen is size_t, not socklen_t
 */
int
so_setsockopt(struct socket *so, int level, int optname, void *optval,
    size_t optlen)
{
	struct sockopt sopt;

	sopt.sopt_level = level;
	sopt.sopt_name = optname;
	sopt.sopt_dir = SOPT_SET;
	sopt.sopt_val = optval;
	sopt.sopt_valsize = optlen;
	sopt.sopt_td = NULL;
	return (sosetopt(so, &sopt));
}

int
sosetopt(struct socket *so, struct sockopt *sopt)
{
	int	error, optval;
	struct	linger l;
	struct	timeval tv;
	sbintime_t val, *valp;
	uint32_t val32;
#ifdef MAC
	struct mac extmac;
#endif

	CURVNET_SET(so->so_vnet);
	error = 0;
	if (sopt->sopt_level != SOL_SOCKET) {
		error = so->so_proto->pr_ctloutput(so, sopt);
	} else {
		switch (sopt->sopt_name) {
		case SO_ACCEPTFILTER:
			error = accept_filt_setopt(so, sopt);
			if (error)
				goto bad;
			break;

		case SO_LINGER:
			error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
			if (error)
				goto bad;
			if (l.l_linger < 0 ||
			    l.l_linger > USHRT_MAX ||
			    l.l_linger > (INT_MAX / hz)) {
				error = EDOM;
				goto bad;
			}
			SOCK_LOCK(so);
			so->so_linger = l.l_linger;
			if (l.l_onoff)
				so->so_options |= SO_LINGER;
			else
				so->so_options &= ~SO_LINGER;
			SOCK_UNLOCK(so);
			break;

		case SO_DEBUG:
		case SO_KEEPALIVE:
		case SO_DONTROUTE:
		case SO_USELOOPBACK:
		case SO_BROADCAST:
		case SO_REUSEADDR:
		case SO_REUSEPORT:
		case SO_REUSEPORT_LB:
		case SO_OOBINLINE:
		case SO_TIMESTAMP:
		case SO_BINTIME:
		case SO_NOSIGPIPE:
		case SO_NO_DDP:
		case SO_NO_OFFLOAD:
		case SO_RERROR:
			error = sooptcopyin(sopt, &optval, sizeof optval,
			    sizeof optval);
			if (error)
				goto bad;
			SOCK_LOCK(so);
			if (optval)
				so->so_options |= sopt->sopt_name;
			else
				so->so_options &= ~sopt->sopt_name;
			SOCK_UNLOCK(so);
			break;

		case SO_SETFIB:
			error = so->so_proto->pr_ctloutput(so, sopt);
			break;

		case SO_USER_COOKIE:
			error = sooptcopyin(sopt, &val32, sizeof val32,
			    sizeof val32);
			if (error)
				goto bad;
			so->so_user_cookie = val32;
			break;

		case SO_SNDBUF:
		case SO_RCVBUF:
		case SO_SNDLOWAT:
		case SO_RCVLOWAT:
			error = so->so_proto->pr_setsbopt(so, sopt);
			if (error)
				goto bad;
			break;

		case SO_SNDTIMEO:
		case SO_RCVTIMEO:
#ifdef COMPAT_FREEBSD32
			if (SV_CURPROC_FLAG(SV_ILP32)) {
				struct timeval32 tv32;

				error = sooptcopyin(sopt, &tv32, sizeof tv32,
				    sizeof tv32);
				CP(tv32, tv, tv_sec);
				CP(tv32, tv, tv_usec);
			} else
#endif
				error = sooptcopyin(sopt, &tv, sizeof tv,
				    sizeof tv);
			if (error)
				goto bad;
			if (tv.tv_sec < 0 || tv.tv_usec < 0 ||
			    tv.tv_usec >= 1000000) {
				error = EDOM;
				goto bad;
			}
			if (tv.tv_sec > INT32_MAX)
				val = SBT_MAX;
			else
				val = tvtosbt(tv);
			SOCK_LOCK(so);
			valp = sopt->sopt_name == SO_SNDTIMEO ?
			    (SOLISTENING(so) ? &so->sol_sbsnd_timeo :
			    &so->so_snd.sb_timeo) :
			    (SOLISTENING(so) ? &so->sol_sbrcv_timeo :
			    &so->so_rcv.sb_timeo);
			*valp = val;
			SOCK_UNLOCK(so);
			break;

		case SO_LABEL:
#ifdef MAC
			error = sooptcopyin(sopt, &extmac, sizeof extmac,
			    sizeof extmac);
			if (error)
				goto bad;
			error = mac_setsockopt_label(sopt->sopt_td->td_ucred,
			    so, &extmac);
#else
			error = EOPNOTSUPP;
#endif
			break;

		case SO_TS_CLOCK:
			error = sooptcopyin(sopt, &optval, sizeof optval,
			    sizeof optval);
			if (error)
				goto bad;
			if (optval < 0 || optval > SO_TS_CLOCK_MAX) {
				error = EINVAL;
				goto bad;
			}
			so->so_ts_clock = optval;
			break;

		case SO_MAX_PACING_RATE:
			error = sooptcopyin(sopt, &val32, sizeof(val32),
			    sizeof(val32));
			if (error)
				goto bad;
			so->so_max_pacing_rate = val32;
			break;

		case SO_SPLICE: {
			struct splice splice;

#ifdef COMPAT_FREEBSD32
			if (SV_CURPROC_FLAG(SV_ILP32)) {
				struct splice32 splice32;

				error = sooptcopyin(sopt, &splice32,
				    sizeof(splice32), sizeof(splice32));
				if (error == 0) {
					splice.sp_fd = splice32.sp_fd;
					splice.sp_max = splice32.sp_max;
					CP(splice32.sp_idle, splice.sp_idle,
					    tv_sec);
					CP(splice32.sp_idle, splice.sp_idle,
					    tv_usec);
				}
			} else
#endif
			{
				error = sooptcopyin(sopt, &splice,
				    sizeof(splice), sizeof(splice));
			}
			if (error)
				goto bad;
#ifdef KTRACE
			if (KTRPOINT(curthread, KTR_STRUCT))
				ktrsplice(&splice);
#endif

			error = splice_init();
			if (error != 0)
				goto bad;

			if (splice.sp_fd >= 0) {
				struct file *fp;
				struct socket *so2;

				if (!cap_rights_contains(sopt->sopt_rights,
				    &cap_recv_rights)) {
					error = ENOTCAPABLE;
					goto bad;
				}
				error = getsock(sopt->sopt_td, splice.sp_fd,
				    &cap_send_rights, &fp);
				if (error != 0)
					goto bad;
				so2 = fp->f_data;

				error = so_splice(so, so2, &splice);
				fdrop(fp, sopt->sopt_td);
			} else {
				error = so_unsplice(so, false);
			}
			break;
		}
		default:
#ifdef SOCKET_HHOOK
			if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
				error = hhook_run_socket(so, sopt,
				    HHOOK_SOCKET_OPT);
			else
#endif
				error = ENOPROTOOPT;
			break;
		}
		if (error == 0)
			(void)so->so_proto->pr_ctloutput(so, sopt);
	}
bad:
	CURVNET_RESTORE();
	return (error);
}

/*
 * Helper routine for getsockopt.
 */
int
sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
{
	int	error;
	size_t	valsize;

	error = 0;

	/*
	 * Documented get behavior is that we always return a value, possibly
	 * truncated to fit in the user's buffer.  Traditional behavior is
	 * that we always tell the user precisely how much we copied, rather
	 * than something useful like the total amount we had available for
	 * her.  Note that this interface is not idempotent; the entire
	 * answer must be generated ahead of time.
	 */
	valsize = min(len, sopt->sopt_valsize);
	sopt->sopt_valsize = valsize;
	if (sopt->sopt_val != NULL) {
		if (sopt->sopt_td != NULL)
			error = copyout(buf, sopt->sopt_val, valsize);
		else
			bcopy(buf, sopt->sopt_val, valsize);
	}
	return (error);
}

int
sogetopt(struct socket *so, struct sockopt *sopt)
{
	int	error, optval;
	struct	linger l;
	struct	timeval tv;
#ifdef MAC
	struct mac extmac;
#endif

	CURVNET_SET(so->so_vnet);
	error = 0;
	if (sopt->sopt_level != SOL_SOCKET) {
		error = so->so_proto->pr_ctloutput(so, sopt);
		CURVNET_RESTORE();
		return (error);
	} else {
		switch (sopt->sopt_name) {
		case SO_ACCEPTFILTER:
			error = accept_filt_getopt(so, sopt);
			break;

		case SO_LINGER:
			SOCK_LOCK(so);
			l.l_onoff = so->so_options & SO_LINGER;
			l.l_linger = so->so_linger;
			SOCK_UNLOCK(so);
			error = sooptcopyout(sopt, &l, sizeof l);
			break;

		case SO_USELOOPBACK:
		case SO_DONTROUTE:
		case SO_DEBUG:
		case SO_KEEPALIVE:
		case SO_REUSEADDR:
		case SO_REUSEPORT:
		case SO_REUSEPORT_LB:
		case SO_BROADCAST:
		case SO_OOBINLINE:
		case SO_ACCEPTCONN:
		case SO_TIMESTAMP:
		case SO_BINTIME:
		case SO_NOSIGPIPE:
		case SO_NO_DDP:
		case SO_NO_OFFLOAD:
		case SO_RERROR:
			optval = so->so_options & sopt->sopt_name;
integer:
			error = sooptcopyout(sopt, &optval, sizeof optval);
			break;

		case SO_FIB:
			SOCK_LOCK(so);
			optval = so->so_fibnum;
			SOCK_UNLOCK(so);
			goto integer;

		case SO_DOMAIN:
			optval = so->so_proto->pr_domain->dom_family;
			goto integer;

		case SO_TYPE:
			optval = so->so_type;
			goto integer;

		case SO_PROTOCOL:
			optval = so->so_proto->pr_protocol;
			goto integer;

		case SO_ERROR:
			SOCK_LOCK(so);
			if (so->so_error) {
				optval = so->so_error;
				so->so_error = 0;
			} else {
				optval = so->so_rerror;
				so->so_rerror = 0;
			}
			SOCK_UNLOCK(so);
			goto integer;

		case SO_SNDBUF:
			SOCK_LOCK(so);
			optval = SOLISTENING(so) ? so->sol_sbsnd_hiwat :
			    so->so_snd.sb_hiwat;
			SOCK_UNLOCK(so);
			goto integer;

		case SO_RCVBUF:
			SOCK_LOCK(so);
			optval = SOLISTENING(so) ? so->sol_sbrcv_hiwat :
			    so->so_rcv.sb_hiwat;
			SOCK_UNLOCK(so);
			goto integer;

		case SO_SNDLOWAT:
			SOCK_LOCK(so);
			optval = SOLISTENING(so) ? so->sol_sbsnd_lowat :
			    so->so_snd.sb_lowat;
			SOCK_UNLOCK(so);
			goto integer;

		case SO_RCVLOWAT:
			SOCK_LOCK(so);
			optval = SOLISTENING(so) ? so->sol_sbrcv_lowat :
			    so->so_rcv.sb_lowat;
			SOCK_UNLOCK(so);
			goto integer;

		case SO_SNDTIMEO:
		case SO_RCVTIMEO:
			SOCK_LOCK(so);
			tv = sbttotv(sopt->sopt_name == SO_SNDTIMEO ?
			    (SOLISTENING(so) ? so->sol_sbsnd_timeo :
			    so->so_snd.sb_timeo) :
			    (SOLISTENING(so) ? so->sol_sbrcv_timeo :
			    so->so_rcv.sb_timeo));
			SOCK_UNLOCK(so);
#ifdef COMPAT_FREEBSD32
			if (SV_CURPROC_FLAG(SV_ILP32)) {
				struct timeval32 tv32;

				CP(tv, tv32, tv_sec);
				CP(tv, tv32, tv_usec);
				error = sooptcopyout(sopt, &tv32, sizeof tv32);
			} else
#endif
				error = sooptcopyout(sopt, &tv, sizeof tv);
			break;

		case SO_LABEL:
#ifdef MAC
			error = sooptcopyin(sopt, &extmac, sizeof(extmac),
			    sizeof(extmac));
			if (error)
				goto bad;
			error = mac_getsockopt_label(sopt->sopt_td->td_ucred,
			    so, &extmac);
			if (error)
				goto bad;
			/* Don't copy out extmac, it is unchanged. */
#else
			error = EOPNOTSUPP;
#endif
			break;

		case SO_PEERLABEL:
#ifdef MAC
			error = sooptcopyin(sopt, &extmac, sizeof(extmac),
			    sizeof(extmac));
			if (error)
				goto bad;
			error = mac_getsockopt_peerlabel(
			    sopt->sopt_td->td_ucred, so, &extmac);
			if (error)
				goto bad;
			/* Don't copy out extmac, it is unchanged. */
#else
			error = EOPNOTSUPP;
#endif
			break;

		case SO_LISTENQLIMIT:
			SOCK_LOCK(so);
			optval = SOLISTENING(so) ? so->sol_qlimit : 0;
			SOCK_UNLOCK(so);
			goto integer;

		case SO_LISTENQLEN:
			SOCK_LOCK(so);
			optval = SOLISTENING(so) ? so->sol_qlen : 0;
			SOCK_UNLOCK(so);
			goto integer;

		case SO_LISTENINCQLEN:
			SOCK_LOCK(so);
			optval = SOLISTENING(so) ? so->sol_incqlen : 0;
			SOCK_UNLOCK(so);
			goto integer;

		case SO_TS_CLOCK:
			optval = so->so_ts_clock;
			goto integer;

		case SO_MAX_PACING_RATE:
			optval = so->so_max_pacing_rate;
			goto integer;

		case SO_SPLICE: {
			off_t n;

			/*
			 * Acquire the I/O lock to serialize with
			 * so_splice_xfer().  This is not required for
			 * correctness, but makes testing simpler: once a byte
			 * has been transmitted to the sink and observed (e.g.,
			 * by reading from the socket to which the sink is
			 * connected), a subsequent getsockopt(SO_SPLICE) will
			 * return an up-to-date value.
			 */
			error = SOCK_IO_RECV_LOCK(so, SBL_WAIT);
			if (error != 0)
				goto bad;
			SOCK_LOCK(so);
			if (SOLISTENING(so)) {
				n = 0;
			} else {
				n = so->so_splice_sent;
			}
			SOCK_UNLOCK(so);
			SOCK_IO_RECV_UNLOCK(so);
			error = sooptcopyout(sopt, &n, sizeof(n));
			break;
		}

		default:
#ifdef SOCKET_HHOOK
			if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
				error = hhook_run_socket(so, sopt,
				    HHOOK_SOCKET_OPT);
			else
#endif
				error = ENOPROTOOPT;
			break;
		}
	}
bad:
	CURVNET_RESTORE();
	return (error);
}

int
soopt_getm(struct sockopt *sopt, struct mbuf **mp)
{
	struct mbuf *m, *m_prev;
	int sopt_size = sopt->sopt_valsize;

	MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
	if (m == NULL)
		return ENOBUFS;
	if (sopt_size > MLEN) {
		MCLGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT);
		if ((m->m_flags & M_EXT) == 0) {
			m_free(m);
			return ENOBUFS;
		}
		m->m_len = min(MCLBYTES, sopt_size);
	} else {
		m->m_len = min(MLEN, sopt_size);
	}
	sopt_size -= m->m_len;
	*mp = m;
	m_prev = m;

	while (sopt_size) {
		MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
		if (m == NULL) {
			m_freem(*mp);
			return ENOBUFS;
		}
		if (sopt_size > MLEN) {
			MCLGET(m, sopt->sopt_td != NULL ? M_WAITOK :
			    M_NOWAIT);
			if ((m->m_flags & M_EXT) == 0) {
				m_freem(m);
				m_freem(*mp);
				return ENOBUFS;
			}
			m->m_len = min(MCLBYTES, sopt_size);
		} else {
			m->m_len = min(MLEN, sopt_size);
		}
		sopt_size -= m->m_len;
		m_prev->m_next = m;
		m_prev = m;
	}
	return (0);
}

int
soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
{
	struct mbuf *m0 = m;

	if (sopt->sopt_val == NULL)
		return (0);
	while (m != NULL && sopt->sopt_valsize >= m->m_len) {
		if (sopt->sopt_td != NULL) {
			int error;

			error = copyin(sopt->sopt_val, mtod(m, char *),
			    m->m_len);
			if (error != 0) {
				m_freem(m0);
				return(error);
			}
		} else
			bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
		sopt->sopt_valsize -= m->m_len;
		sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
		m = m->m_next;
	}
	if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
		panic("ip6_sooptmcopyin");
	return (0);
}

int
soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
{
	struct mbuf *m0 = m;
	size_t valsize = 0;

	if (sopt->sopt_val == NULL)
		return (0);
	while (m != NULL && sopt->sopt_valsize >= m->m_len) {
		if (sopt->sopt_td != NULL) {
			int error;

			error = copyout(mtod(m, char *), sopt->sopt_val,
			    m->m_len);
			if (error != 0) {
				m_freem(m0);
				return(error);
			}
		} else
			bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
		sopt->sopt_valsize -= m->m_len;
		sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
		valsize += m->m_len;
		m = m->m_next;
	}
	if (m != NULL) {
		/* enough soopt buffer should be given from user-land */
		m_freem(m0);
		return(EINVAL);
	}
	sopt->sopt_valsize = valsize;
	return (0);
}

/*
 * sohasoutofband(): protocol notifies socket layer of the arrival of new
 * out-of-band data, which will then notify socket consumers.
 */
void
sohasoutofband(struct socket *so)
{

	if (so->so_sigio != NULL)
		pgsigio(&so->so_sigio, SIGURG, 0);
	selwakeuppri(&so->so_rdsel, PSOCK);
}

int
sopoll_generic(struct socket *so, int events, struct thread *td)
{
	int revents;

	SOCK_LOCK(so);
	if (SOLISTENING(so)) {
		if (!(events & (POLLIN | POLLRDNORM)))
			revents = 0;
		else if (!TAILQ_EMPTY(&so->sol_comp))
			revents = events & (POLLIN | POLLRDNORM);
		else if ((events & POLLINIGNEOF) == 0 && so->so_error)
			revents = (events & (POLLIN | POLLRDNORM)) | POLLHUP;
		else {
			selrecord(td, &so->so_rdsel);
			revents = 0;
		}
	} else {
		revents = 0;
		SOCK_SENDBUF_LOCK(so);
		SOCK_RECVBUF_LOCK(so);
		if (events & (POLLIN | POLLRDNORM))
			if (soreadabledata(so) && !isspliced(so))
				revents |= events & (POLLIN | POLLRDNORM);
		if (events & (POLLOUT | POLLWRNORM))
			if (sowriteable(so) && !issplicedback(so))
				revents |= events & (POLLOUT | POLLWRNORM);
		if (events & (POLLPRI | POLLRDBAND))
			if (so->so_oobmark ||
			    (so->so_rcv.sb_state & SBS_RCVATMARK))
				revents |= events & (POLLPRI | POLLRDBAND);
		if ((events & POLLINIGNEOF) == 0) {
			if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
				revents |= events & (POLLIN | POLLRDNORM);
				if (so->so_snd.sb_state & SBS_CANTSENDMORE)
					revents |= POLLHUP;
			}
		}
		if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
			revents |= events & POLLRDHUP;
		if (revents == 0) {
			if (events &
			    (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND | POLLRDHUP)) {
				selrecord(td, &so->so_rdsel);
				so->so_rcv.sb_flags |= SB_SEL;
			}
			if (events & (POLLOUT | POLLWRNORM)) {
				selrecord(td, &so->so_wrsel);
				so->so_snd.sb_flags |= SB_SEL;
			}
		}
		SOCK_RECVBUF_UNLOCK(so);
		SOCK_SENDBUF_UNLOCK(so);
	}
	SOCK_UNLOCK(so);
	return (revents);
}

int
sokqfilter_generic(struct socket *so, struct knote *kn)
{
	struct sockbuf *sb;
	sb_which which;
	struct knlist *knl;

	switch (kn->kn_filter) {
	case EVFILT_READ:
		kn->kn_fop = &soread_filtops;
		knl = &so->so_rdsel.si_note;
		sb = &so->so_rcv;
		which = SO_RCV;
		break;
	case EVFILT_WRITE:
		kn->kn_fop = &sowrite_filtops;
		knl = &so->so_wrsel.si_note;
		sb = &so->so_snd;
		which = SO_SND;
		break;
	case EVFILT_EMPTY:
		kn->kn_fop = &soempty_filtops;
		knl = &so->so_wrsel.si_note;
		sb = &so->so_snd;
		which = SO_SND;
		break;
	default:
		return (EINVAL);
	}

	SOCK_LOCK(so);
	if (SOLISTENING(so)) {
		knlist_add(knl, kn, 1);
	} else {
		SOCK_BUF_LOCK(so, which);
		knlist_add(knl, kn, 1);
		sb->sb_flags |= SB_KNOTE;
		if ((kn->kn_sfflags & NOTE_LOWAT) &&
		    (sb->sb_flags & SB_AUTOLOWAT))
			sb->sb_flags &= ~SB_AUTOLOWAT;
		SOCK_BUF_UNLOCK(so, which);
	}
	SOCK_UNLOCK(so);
	return (0);
}

static void
filt_sordetach(struct knote *kn)
{
	struct socket *so = kn->kn_fp->f_data;

	so_rdknl_lock(so);
	knlist_remove(&so->so_rdsel.si_note, kn, 1);
	if (!SOLISTENING(so) && knlist_empty(&so->so_rdsel.si_note))
		so->so_rcv.sb_flags &= ~SB_KNOTE;
	so_rdknl_unlock(so);
}

/*ARGSUSED*/
static int
filt_soread(struct knote *kn, long hint)
{
	struct socket *so;

	so = kn->kn_fp->f_data;

	if (SOLISTENING(so)) {
		SOCK_LOCK_ASSERT(so);
		kn->kn_data = so->sol_qlen;
		if (so->so_error) {
			kn->kn_flags |= EV_EOF;
			kn->kn_fflags = so->so_error;
			return (1);
		}
		return (!TAILQ_EMPTY(&so->sol_comp));
	}

	if ((so->so_rcv.sb_flags & SB_SPLICED) != 0)
		return (0);

	SOCK_RECVBUF_LOCK_ASSERT(so);

	kn->kn_data = sbavail(&so->so_rcv) - so->so_rcv.sb_ctl;
	if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
		kn->kn_flags |= EV_EOF;
		kn->kn_fflags = so->so_error;
		return (1);
	} else if (so->so_error || so->so_rerror)
		return (1);

	if (kn->kn_sfflags & NOTE_LOWAT) {
		if (kn->kn_data >= kn->kn_sdata)
			return (1);
	} else if (sbavail(&so->so_rcv) >= so->so_rcv.sb_lowat)
		return (1);

#ifdef SOCKET_HHOOK
	/* This hook returning non-zero indicates an event, not error */
	return (hhook_run_socket(so, NULL, HHOOK_FILT_SOREAD));
#else
	return (0);
#endif
}

static void
filt_sowdetach(struct knote *kn)
{
	struct socket *so = kn->kn_fp->f_data;

	so_wrknl_lock(so);
	knlist_remove(&so->so_wrsel.si_note, kn, 1);
	if (!SOLISTENING(so) && knlist_empty(&so->so_wrsel.si_note))
		so->so_snd.sb_flags &= ~SB_KNOTE;
	so_wrknl_unlock(so);
}

/*ARGSUSED*/
static int
filt_sowrite(struct knote *kn, long hint)
{
	struct socket *so;

	so = kn->kn_fp->f_data;

	if (SOLISTENING(so))
		return (0);

	SOCK_SENDBUF_LOCK_ASSERT(so);
	kn->kn_data = sbspace(&so->so_snd);

#ifdef SOCKET_HHOOK
	hhook_run_socket(so, kn, HHOOK_FILT_SOWRITE);
#endif

	if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
		kn->kn_flags |= EV_EOF;
		kn->kn_fflags = so->so_error;
		return (1);
	} else if (so->so_error)	/* temporary udp error */
		return (1);
	else if (((so->so_state & SS_ISCONNECTED) == 0) &&
	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
		return (0);
	else if (kn->kn_sfflags & NOTE_LOWAT)
		return (kn->kn_data >= kn->kn_sdata);
	else
		return (kn->kn_data >= so->so_snd.sb_lowat);
}

static int
filt_soempty(struct knote *kn, long hint)
{
	struct socket *so;

	so = kn->kn_fp->f_data;

	if (SOLISTENING(so))
		return (1);

	SOCK_SENDBUF_LOCK_ASSERT(so);
	kn->kn_data = sbused(&so->so_snd);

	if (kn->kn_data == 0)
		return (1);
	else
		return (0);
}

int
socheckuid(struct socket *so, uid_t uid)
{

	if (so == NULL)
		return (EPERM);
	if (so->so_cred->cr_uid != uid)
		return (EPERM);
	return (0);
}

/*
 * These functions are used by protocols to notify the socket layer (and its
 * consumers) of state changes in the sockets driven by protocol-side events.
 */

/*
 * Procedures to manipulate state flags of socket and do appropriate wakeups.
 *
 * Normal sequence from the active (originating) side is that
 * soisconnecting() is called during processing of connect() call, resulting
 * in an eventual call to soisconnected() if/when the connection is
 * established.  When the connection is torn down soisdisconnecting() is
 * called during processing of disconnect() call, and soisdisconnected() is
 * called when the connection to the peer is totally severed.  The semantics
 * of these routines are such that connectionless protocols can call
 * soisconnected() and soisdisconnected() only, bypassing the in-progress
 * calls when setting up a ``connection'' takes no time.
 *
 * From the passive side, a socket is created with two queues of sockets:
 * so_incomp for connections in progress and so_comp for connections already
 * made and awaiting user acceptance.  As a protocol is preparing incoming
 * connections, it creates a socket structure queued on so_incomp by calling
 * sonewconn().  When the connection is established, soisconnected() is
 * called, and transfers the socket structure to so_comp, making it available
 * to accept().
 *
 * If a socket is closed with sockets on either so_incomp or so_comp, these
 * sockets are dropped.
 *
 * If higher-level protocols are implemented in the kernel, the wakeups done
 * here will sometimes cause software-interrupt process scheduling.
 */
void
soisconnecting(struct socket *so)
{

	SOCK_LOCK(so);
	so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
	so->so_state |= SS_ISCONNECTING;
	SOCK_UNLOCK(so);
}

void
soisconnected(struct socket *so)
{
	bool last __diagused;

	SOCK_LOCK(so);
	so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING);
	so->so_state |= SS_ISCONNECTED;

	if (so->so_qstate == SQ_INCOMP) {
		struct socket *head = so->so_listen;
		int ret;

		KASSERT(head, ("%s: so %p on incomp of NULL", __func__, so));
		/*
		 * Promoting a socket from incomplete queue to complete, we
		 * need to go through reverse order of locking.  We first do
		 * trylock, and if that doesn't succeed, we go the hard way
		 * leaving a reference and rechecking consistency after proper
		 * locking.
		 */
		if (__predict_false(SOLISTEN_TRYLOCK(head) == 0)) {
			soref(head);
			SOCK_UNLOCK(so);
			SOLISTEN_LOCK(head);
			SOCK_LOCK(so);
			if (__predict_false(head != so->so_listen)) {
				/*
				 * The socket went off the listen queue,
				 * should be lost race to close(2) of sol.
				 * The socket is about to soabort().
				 */
				SOCK_UNLOCK(so);
				sorele_locked(head);
				return;
			}
			last = refcount_release(&head->so_count);
			KASSERT(!last, ("%s: released last reference for %p",
			    __func__, head));
		}
again:
		if ((so->so_options & SO_ACCEPTFILTER) == 0) {
			TAILQ_REMOVE(&head->sol_incomp, so, so_list);
			head->sol_incqlen--;
			TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list);
			head->sol_qlen++;
			so->so_qstate = SQ_COMP;
			SOCK_UNLOCK(so);
			solisten_wakeup(head);	/* unlocks */
		} else {
			SOCK_RECVBUF_LOCK(so);
			soupcall_set(so, SO_RCV,
			    head->sol_accept_filter->accf_callback,
			    head->sol_accept_filter_arg);
			so->so_options &= ~SO_ACCEPTFILTER;
			ret = head->sol_accept_filter->accf_callback(so,
			    head->sol_accept_filter_arg, M_NOWAIT);
			if (ret == SU_ISCONNECTED) {
				soupcall_clear(so, SO_RCV);
				SOCK_RECVBUF_UNLOCK(so);
				goto again;
			}
			SOCK_RECVBUF_UNLOCK(so);
			SOCK_UNLOCK(so);
			SOLISTEN_UNLOCK(head);
		}
		return;
	}
	SOCK_UNLOCK(so);
	wakeup(&so->so_timeo);
	sorwakeup(so);
	sowwakeup(so);
}

void
soisdisconnecting(struct socket *so)
{

	SOCK_LOCK(so);
	so->so_state &= ~SS_ISCONNECTING;
	so->so_state |= SS_ISDISCONNECTING;

	if (!SOLISTENING(so)) {
		SOCK_RECVBUF_LOCK(so);
		socantrcvmore_locked(so);
		SOCK_SENDBUF_LOCK(so);
		socantsendmore_locked(so);
	}
	SOCK_UNLOCK(so);
	wakeup(&so->so_timeo);
}

void
soisdisconnected(struct socket *so)
{

	SOCK_LOCK(so);

	/*
	 * There is at least one reader of so_state that does not
	 * acquire socket lock, namely soreceive_generic().  Ensure
	 * that it never sees all flags that track connection status
	 * cleared, by ordering the update with a barrier semantic of
	 * our release thread fence.
	 */
	so->so_state |= SS_ISDISCONNECTED;
	atomic_thread_fence_rel();
	so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);

	if (!SOLISTENING(so)) {
		SOCK_UNLOCK(so);
		SOCK_RECVBUF_LOCK(so);
		socantrcvmore_locked(so);
		SOCK_SENDBUF_LOCK(so);
		sbdrop_locked(&so->so_snd, sbused(&so->so_snd));
		socantsendmore_locked(so);
	} else
		SOCK_UNLOCK(so);
	wakeup(&so->so_timeo);
}

int
soiolock(struct socket *so, struct sx *sx, int flags)
{
	int error;

	KASSERT((flags & SBL_VALID) == flags,
	    ("soiolock: invalid flags %#x", flags));

	if ((flags & SBL_WAIT) != 0) {
		if ((flags & SBL_NOINTR) != 0) {
			sx_xlock(sx);
		} else {
			error = sx_xlock_sig(sx);
			if (error != 0)
				return (error);
		}
	} else if (!sx_try_xlock(sx)) {
		return (EWOULDBLOCK);
	}

	if (__predict_false(SOLISTENING(so))) {
		sx_xunlock(sx);
		return (ENOTCONN);
	}
	return (0);
}

void
soiounlock(struct sx *sx)
{
	sx_xunlock(sx);
}

/*
 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
 */
struct sockaddr *
sodupsockaddr(const struct sockaddr *sa, int mflags)
{
	struct sockaddr *sa2;

	sa2 = malloc(sa->sa_len, M_SONAME, mflags);
	if (sa2)
		bcopy(sa, sa2, sa->sa_len);
	return sa2;
}

/*
 * Register per-socket destructor.
 */
void
sodtor_set(struct socket *so, so_dtor_t *func)
{

	SOCK_LOCK_ASSERT(so);
	so->so_dtor = func;
}

/*
 * Register per-socket buffer upcalls.
 */
void
soupcall_set(struct socket *so, sb_which which, so_upcall_t func, void *arg)
{
	struct sockbuf *sb;

	KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so));

	switch (which) {
	case SO_RCV:
		sb = &so->so_rcv;
		break;
	case SO_SND:
		sb = &so->so_snd;
		break;
	}
	SOCK_BUF_LOCK_ASSERT(so, which);
	sb->sb_upcall = func;
	sb->sb_upcallarg = arg;
	sb->sb_flags |= SB_UPCALL;
}

void
soupcall_clear(struct socket *so, sb_which which)
{
	struct sockbuf *sb;

	KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so));

	switch (which) {
	case SO_RCV:
		sb = &so->so_rcv;
		break;
	case SO_SND:
		sb = &so->so_snd;
		break;
	}
	SOCK_BUF_LOCK_ASSERT(so, which);
	KASSERT(sb->sb_upcall != NULL,
	    ("%s: so %p no upcall to clear", __func__, so));
	sb->sb_upcall = NULL;
	sb->sb_upcallarg = NULL;
	sb->sb_flags &= ~SB_UPCALL;
}

void
solisten_upcall_set(struct socket *so, so_upcall_t func, void *arg)
{

	SOLISTEN_LOCK_ASSERT(so);
	so->sol_upcall = func;
	so->sol_upcallarg = arg;
}

static void
so_rdknl_lock(void *arg)
{
	struct socket *so = arg;

retry:
	if (SOLISTENING(so)) {
		SOLISTEN_LOCK(so);
	} else {
		SOCK_RECVBUF_LOCK(so);
		if (__predict_false(SOLISTENING(so))) {
			SOCK_RECVBUF_UNLOCK(so);
			goto retry;
		}
	}
}

static void
so_rdknl_unlock(void *arg)
{
	struct socket *so = arg;

	if (SOLISTENING(so))
		SOLISTEN_UNLOCK(so);
	else
		SOCK_RECVBUF_UNLOCK(so);
}

static void
so_rdknl_assert_lock(void *arg, int what)
{
	struct socket *so = arg;

	if (what == LA_LOCKED) {
		if (SOLISTENING(so))
			SOLISTEN_LOCK_ASSERT(so);
		else
			SOCK_RECVBUF_LOCK_ASSERT(so);
	} else {
		if (SOLISTENING(so))
			SOLISTEN_UNLOCK_ASSERT(so);
		else
			SOCK_RECVBUF_UNLOCK_ASSERT(so);
	}
}

static void
so_wrknl_lock(void *arg)
{
	struct socket *so = arg;

retry:
	if (SOLISTENING(so)) {
		SOLISTEN_LOCK(so);
	} else {
		SOCK_SENDBUF_LOCK(so);
		if (__predict_false(SOLISTENING(so))) {
			SOCK_SENDBUF_UNLOCK(so);
			goto retry;
		}
	}
}

static void
so_wrknl_unlock(void *arg)
{
	struct socket *so = arg;

	if (SOLISTENING(so))
		SOLISTEN_UNLOCK(so);
	else
		SOCK_SENDBUF_UNLOCK(so);
}

static void
so_wrknl_assert_lock(void *arg, int what)
{
	struct socket *so = arg;

	if (what == LA_LOCKED) {
		if (SOLISTENING(so))
			SOLISTEN_LOCK_ASSERT(so);
		else
			SOCK_SENDBUF_LOCK_ASSERT(so);
	} else {
		if (SOLISTENING(so))
			SOLISTEN_UNLOCK_ASSERT(so);
		else
			SOCK_SENDBUF_UNLOCK_ASSERT(so);
	}
}

/*
 * Create an external-format (``xsocket'') structure using the information in
 * the kernel-format socket structure pointed to by so.  This is done to
 * reduce the spew of irrelevant information over this interface, to isolate
 * user code from changes in the kernel structure, and potentially to provide
 * information-hiding if we decide that some of this information should be
 * hidden from users.
 */
void
sotoxsocket(struct socket *so, struct xsocket *xso)
{

	bzero(xso, sizeof(*xso));
	xso->xso_len = sizeof *xso;
	xso->xso_so = (uintptr_t)so;
	xso->so_type = so->so_type;
	xso->so_options = so->so_options;
	xso->so_linger = so->so_linger;
	xso->so_state = so->so_state;
	xso->so_pcb = (uintptr_t)so->so_pcb;
	xso->xso_protocol = so->so_proto->pr_protocol;
	xso->xso_family = so->so_proto->pr_domain->dom_family;
	xso->so_timeo = so->so_timeo;
	xso->so_error = so->so_error;
	xso->so_uid = so->so_cred->cr_uid;
	xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
	SOCK_LOCK(so);
	xso->so_fibnum = so->so_fibnum;
	if (SOLISTENING(so)) {
		xso->so_qlen = so->sol_qlen;
		xso->so_incqlen = so->sol_incqlen;
		xso->so_qlimit = so->sol_qlimit;
		xso->so_oobmark = 0;
	} else {
		xso->so_state |= so->so_qstate;
		xso->so_qlen = xso->so_incqlen = xso->so_qlimit = 0;
		xso->so_oobmark = so->so_oobmark;
		sbtoxsockbuf(&so->so_snd, &xso->so_snd);
		sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
		if ((so->so_rcv.sb_flags & SB_SPLICED) != 0)
			xso->so_splice_so = (uintptr_t)so->so_splice->dst;
	}
	SOCK_UNLOCK(so);
}

int
so_options_get(const struct socket *so)
{

	return (so->so_options);
}

void
so_options_set(struct socket *so, int val)
{

	so->so_options = val;
}

int
so_error_get(const struct socket *so)
{

	return (so->so_error);
}

void
so_error_set(struct socket *so, int val)
{

	so->so_error = val;
}