Home | History | Annotate | Line # | Download | only in kern
      1 /*	$NetBSD: uipc_socket.c,v 1.314 2025/07/16 19:14:13 kre Exp $	*/
      2 
      3 /*
      4  * Copyright (c) 2002, 2007, 2008, 2009, 2023 The NetBSD Foundation, Inc.
      5  * All rights reserved.
      6  *
      7  * This code is derived from software contributed to The NetBSD Foundation
      8  * by Jason R. Thorpe of Wasabi Systems, Inc, and by Andrew Doran.
      9  *
     10  * Redistribution and use in source and binary forms, with or without
     11  * modification, are permitted provided that the following conditions
     12  * are met:
     13  * 1. Redistributions of source code must retain the above copyright
     14  *    notice, this list of conditions and the following disclaimer.
     15  * 2. Redistributions in binary form must reproduce the above copyright
     16  *    notice, this list of conditions and the following disclaimer in the
     17  *    documentation and/or other materials provided with the distribution.
     18  *
     19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     29  * POSSIBILITY OF SUCH DAMAGE.
     30  */
     31 
     32 /*
     33  * Copyright (c) 2004 The FreeBSD Foundation
     34  * Copyright (c) 2004 Robert Watson
     35  * Copyright (c) 1982, 1986, 1988, 1990, 1993
     36  *	The Regents of the University of California.  All rights reserved.
     37  *
     38  * Redistribution and use in source and binary forms, with or without
     39  * modification, are permitted provided that the following conditions
     40  * are met:
     41  * 1. Redistributions of source code must retain the above copyright
     42  *    notice, this list of conditions and the following disclaimer.
     43  * 2. Redistributions in binary form must reproduce the above copyright
     44  *    notice, this list of conditions and the following disclaimer in the
     45  *    documentation and/or other materials provided with the distribution.
     46  * 3. Neither the name of the University nor the names of its contributors
     47  *    may be used to endorse or promote products derived from this software
     48  *    without specific prior written permission.
     49  *
     50  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     51  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     52  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     53  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     54  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     55  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     56  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     57  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     58  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     59  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     60  * SUCH DAMAGE.
     61  *
     62  *	@(#)uipc_socket.c	8.6 (Berkeley) 5/2/95
     63  */
     64 
     65 /*
     66  * Socket operation routines.
     67  *
     68  * These routines are called by the routines in sys_socket.c or from a
     69  * system process, and implement the semantics of socket operations by
     70  * switching out to the protocol specific routines.
     71  */
     72 
     73 #include <sys/cdefs.h>
     74 __KERNEL_RCSID(0, "$NetBSD: uipc_socket.c,v 1.314 2025/07/16 19:14:13 kre Exp $");
     75 
     76 #ifdef _KERNEL_OPT
     77 #include "opt_compat_netbsd.h"
     78 #include "opt_mbuftrace.h"
     79 #include "opt_multiprocessor.h"	/* XXX */
     80 #include "opt_pipe.h"
     81 #include "opt_sctp.h"
     82 #include "opt_sock_counters.h"
     83 #include "opt_somaxkva.h"
     84 #include "opt_sosend_loan.h"
     85 #endif
     86 
     87 #include <sys/param.h>
     88 #include <sys/types.h>
     89 
     90 #include <sys/compat_stub.h>
     91 #include <sys/condvar.h>
     92 #include <sys/domain.h>
     93 #include <sys/event.h>
     94 #include <sys/file.h>
     95 #include <sys/filedesc.h>
     96 #include <sys/kauth.h>
     97 #include <sys/kernel.h>
     98 #include <sys/kmem.h>
     99 #include <sys/kthread.h>
    100 #include <sys/mbuf.h>
    101 #include <sys/mutex.h>
    102 #include <sys/poll.h>
    103 #include <sys/proc.h>
    104 #include <sys/protosw.h>
    105 #include <sys/resourcevar.h>
    106 #include <sys/sdt.h>
    107 #include <sys/signalvar.h>
    108 #include <sys/socket.h>
    109 #include <sys/socketvar.h>
    110 #include <sys/systm.h>
    111 #include <sys/uidinfo.h>
    112 
    113 #include <compat/sys/socket.h>
    114 #include <compat/sys/time.h>
    115 
    116 #include <uvm/uvm_extern.h>
    117 #include <uvm/uvm_loan.h>
    118 #include <uvm/uvm_page.h>
    119 
    120 #ifdef SCTP
    121 #include <netinet/sctp_route.h>
    122 #endif
    123 
    124 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
    125 
    126 extern const struct fileops socketops;
    127 
    128 static int	sooptions;
    129 extern int	somaxconn;			/* patchable (XXX sysctl) */
    130 int		somaxconn = SOMAXCONN;
    131 kmutex_t	*softnet_lock;
    132 
    133 #ifdef SOSEND_COUNTERS
    134 #include <sys/device.h>
    135 
    136 static struct evcnt sosend_loan_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
    137     NULL, "sosend", "loan big");
    138 static struct evcnt sosend_copy_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
    139     NULL, "sosend", "copy big");
    140 static struct evcnt sosend_copy_small = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
    141     NULL, "sosend", "copy small");
    142 static struct evcnt sosend_kvalimit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
    143     NULL, "sosend", "kva limit");
    144 
    145 #define	SOSEND_COUNTER_INCR(ev)		(ev)->ev_count++
    146 
    147 EVCNT_ATTACH_STATIC(sosend_loan_big);
    148 EVCNT_ATTACH_STATIC(sosend_copy_big);
    149 EVCNT_ATTACH_STATIC(sosend_copy_small);
    150 EVCNT_ATTACH_STATIC(sosend_kvalimit);
    151 #else
    152 
    153 #define	SOSEND_COUNTER_INCR(ev)		/* nothing */
    154 
    155 #endif /* SOSEND_COUNTERS */
    156 
    157 #if defined(SOSEND_NO_LOAN) || defined(MULTIPROCESSOR)
    158 int sock_loan_thresh = -1;
    159 #else
    160 int sock_loan_thresh = 4096;
    161 #endif
    162 
    163 static kmutex_t so_pendfree_lock;
    164 static struct mbuf *so_pendfree = NULL;
    165 
    166 #ifndef SOMAXKVA
    167 #define	SOMAXKVA (16 * 1024 * 1024)
    168 #endif
    169 int somaxkva = SOMAXKVA;
    170 static int socurkva;
    171 static kcondvar_t socurkva_cv;
    172 
    173 #ifndef SOFIXEDBUF
    174 #define SOFIXEDBUF true
    175 #endif
    176 bool sofixedbuf = SOFIXEDBUF;
    177 
    178 static kauth_listener_t socket_listener;
    179 
    180 #define	SOCK_LOAN_CHUNK		65536
    181 
    182 static void sopendfree_thread(void *);
    183 static kcondvar_t pendfree_thread_cv;
    184 static lwp_t *sopendfree_lwp;
    185 
    186 static void sysctl_kern_socket_setup(void);
    187 static struct sysctllog *socket_sysctllog;
    188 
    189 static vsize_t
    190 sokvareserve(struct socket *so, vsize_t len)
    191 {
    192 	int error;
    193 
    194 	mutex_enter(&so_pendfree_lock);
    195 	while (socurkva + len > somaxkva) {
    196 		SOSEND_COUNTER_INCR(&sosend_kvalimit);
    197 		error = cv_wait_sig(&socurkva_cv, &so_pendfree_lock);
    198 		if (error) {
    199 			len = 0;
    200 			break;
    201 		}
    202 	}
    203 	socurkva += len;
    204 	mutex_exit(&so_pendfree_lock);
    205 	return len;
    206 }
    207 
    208 static void
    209 sokvaunreserve(vsize_t len)
    210 {
    211 
    212 	mutex_enter(&so_pendfree_lock);
    213 	socurkva -= len;
    214 	cv_broadcast(&socurkva_cv);
    215 	mutex_exit(&so_pendfree_lock);
    216 }
    217 
    218 /*
    219  * sokvaalloc: allocate kva for loan.
    220  */
    221 vaddr_t
    222 sokvaalloc(vaddr_t sva, vsize_t len, struct socket *so)
    223 {
    224 	vaddr_t lva;
    225 
    226 	if (sokvareserve(so, len) == 0)
    227 		return 0;
    228 
    229 	lva = uvm_km_alloc(kernel_map, len, atop(sva) & uvmexp.colormask,
    230 	    UVM_KMF_COLORMATCH | UVM_KMF_VAONLY | UVM_KMF_WAITVA);
    231 	if (lva == 0) {
    232 		sokvaunreserve(len);
    233 		return 0;
    234 	}
    235 
    236 	return lva;
    237 }
    238 
    239 /*
    240  * sokvafree: free kva for loan.
    241  */
    242 void
    243 sokvafree(vaddr_t sva, vsize_t len)
    244 {
    245 
    246 	uvm_km_free(kernel_map, sva, len, UVM_KMF_VAONLY);
    247 	sokvaunreserve(len);
    248 }
    249 
    250 static void
    251 sodoloanfree(struct vm_page **pgs, void *buf, size_t size)
    252 {
    253 	vaddr_t sva, eva;
    254 	vsize_t len;
    255 	int npgs;
    256 
    257 	KASSERT(pgs != NULL);
    258 
    259 	eva = round_page((vaddr_t) buf + size);
    260 	sva = trunc_page((vaddr_t) buf);
    261 	len = eva - sva;
    262 	npgs = len >> PAGE_SHIFT;
    263 
    264 	pmap_kremove(sva, len);
    265 	pmap_update(pmap_kernel());
    266 	uvm_unloan(pgs, npgs, UVM_LOAN_TOPAGE);
    267 	sokvafree(sva, len);
    268 }
    269 
    270 /*
    271  * sopendfree_thread: free mbufs on "pendfree" list. Unlock and relock
    272  * so_pendfree_lock when freeing mbufs.
    273  */
    274 static void
    275 sopendfree_thread(void *v)
    276 {
    277 	struct mbuf *m, *next;
    278 	size_t rv;
    279 
    280 	mutex_enter(&so_pendfree_lock);
    281 
    282 	for (;;) {
    283 		rv = 0;
    284 		while (so_pendfree != NULL) {
    285 			m = so_pendfree;
    286 			so_pendfree = NULL;
    287 			mutex_exit(&so_pendfree_lock);
    288 
    289 			for (; m != NULL; m = next) {
    290 				next = m->m_next;
    291 				KASSERT((~m->m_flags & (M_EXT|M_EXT_PAGES)) ==
    292 				    0);
    293 				KASSERT(m->m_ext.ext_refcnt == 0);
    294 
    295 				rv += m->m_ext.ext_size;
    296 				sodoloanfree(m->m_ext.ext_pgs, m->m_ext.ext_buf,
    297 				    m->m_ext.ext_size);
    298 				pool_cache_put(mb_cache, m);
    299 			}
    300 
    301 			mutex_enter(&so_pendfree_lock);
    302 		}
    303 		if (rv)
    304 			cv_broadcast(&socurkva_cv);
    305 		cv_wait(&pendfree_thread_cv, &so_pendfree_lock);
    306 	}
    307 	panic("sopendfree_thread");
    308 	/* NOTREACHED */
    309 }
    310 
    311 void
    312 soloanfree(struct mbuf *m, void *buf, size_t size, void *arg)
    313 {
    314 
    315 	KASSERT(m != NULL);
    316 
    317 	/*
    318 	 * postpone freeing mbuf.
    319 	 *
    320 	 * we can't do it in interrupt context
    321 	 * because we need to put kva back to kernel_map.
    322 	 */
    323 
    324 	mutex_enter(&so_pendfree_lock);
    325 	m->m_next = so_pendfree;
    326 	so_pendfree = m;
    327 	cv_signal(&pendfree_thread_cv);
    328 	mutex_exit(&so_pendfree_lock);
    329 }
    330 
    331 static long
    332 sosend_loan(struct socket *so, struct uio *uio, struct mbuf *m, long space)
    333 {
    334 	struct iovec *iov = uio->uio_iov;
    335 	vaddr_t sva, eva;
    336 	vsize_t len;
    337 	vaddr_t lva;
    338 	int npgs, error;
    339 	vaddr_t va;
    340 	int i;
    341 
    342 	if (VMSPACE_IS_KERNEL_P(uio->uio_vmspace))
    343 		return 0;
    344 
    345 	if (iov->iov_len < (size_t) space)
    346 		space = iov->iov_len;
    347 	if (space > SOCK_LOAN_CHUNK)
    348 		space = SOCK_LOAN_CHUNK;
    349 
    350 	eva = round_page((vaddr_t) iov->iov_base + space);
    351 	sva = trunc_page((vaddr_t) iov->iov_base);
    352 	len = eva - sva;
    353 	npgs = len >> PAGE_SHIFT;
    354 
    355 	KASSERT(npgs <= M_EXT_MAXPAGES);
    356 
    357 	lva = sokvaalloc(sva, len, so);
    358 	if (lva == 0)
    359 		return 0;
    360 
    361 	error = uvm_loan(&uio->uio_vmspace->vm_map, sva, len,
    362 	    m->m_ext.ext_pgs, UVM_LOAN_TOPAGE);
    363 	if (error) {
    364 		sokvafree(lva, len);
    365 		return 0;
    366 	}
    367 
    368 	for (i = 0, va = lva; i < npgs; i++, va += PAGE_SIZE)
    369 		pmap_kenter_pa(va, VM_PAGE_TO_PHYS(m->m_ext.ext_pgs[i]),
    370 		    VM_PROT_READ, 0);
    371 	pmap_update(pmap_kernel());
    372 
    373 	lva += (vaddr_t) iov->iov_base & PAGE_MASK;
    374 
    375 	MEXTADD(m, (void *) lva, space, M_MBUF, soloanfree, so);
    376 	m->m_flags |= M_EXT_PAGES | M_EXT_ROMAP;
    377 
    378 	uio->uio_resid -= space;
    379 	/* uio_offset not updated, not set/used for write(2) */
    380 	uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + space;
    381 	uio->uio_iov->iov_len -= space;
    382 	if (uio->uio_iov->iov_len == 0) {
    383 		uio->uio_iov++;
    384 		uio->uio_iovcnt--;
    385 	}
    386 
    387 	return space;
    388 }
    389 
    390 static int
    391 socket_listener_cb(kauth_cred_t cred, kauth_action_t action, void *cookie,
    392     void *arg0, void *arg1, void *arg2, void *arg3)
    393 {
    394 	int result;
    395 	enum kauth_network_req req;
    396 
    397 	result = KAUTH_RESULT_DEFER;
    398 	req = (enum kauth_network_req)(uintptr_t)arg0;
    399 
    400 	if ((action != KAUTH_NETWORK_SOCKET) &&
    401 	    (action != KAUTH_NETWORK_BIND))
    402 		return result;
    403 
    404 	switch (req) {
    405 	case KAUTH_REQ_NETWORK_BIND_PORT:
    406 		result = KAUTH_RESULT_ALLOW;
    407 		break;
    408 
    409 	case KAUTH_REQ_NETWORK_SOCKET_DROP: {
    410 		/* Normal users can only drop their own connections. */
    411 		struct socket *so = (struct socket *)arg1;
    412 
    413 		if (so->so_cred && proc_uidmatch(cred, so->so_cred) == 0)
    414 			result = KAUTH_RESULT_ALLOW;
    415 
    416 		break;
    417 		}
    418 
    419 	case KAUTH_REQ_NETWORK_SOCKET_OPEN:
    420 		/* We allow "raw" routing/bluetooth sockets to anyone. */
    421 		switch ((u_long)arg1) {
    422 		case PF_ROUTE:
    423 		case PF_OROUTE:
    424 		case PF_BLUETOOTH:
    425 		case PF_CAN:
    426 			result = KAUTH_RESULT_ALLOW;
    427 			break;
    428 		default:
    429 			/* Privileged, let secmodel handle this. */
    430 			if ((u_long)arg2 == SOCK_RAW)
    431 				break;
    432 			result = KAUTH_RESULT_ALLOW;
    433 			break;
    434 		}
    435 		break;
    436 
    437 	case KAUTH_REQ_NETWORK_SOCKET_CANSEE:
    438 		result = KAUTH_RESULT_ALLOW;
    439 
    440 		break;
    441 
    442 	default:
    443 		break;
    444 	}
    445 
    446 	return result;
    447 }
    448 
    449 void
    450 soinit(void)
    451 {
    452 
    453 	sysctl_kern_socket_setup();
    454 
    455 #ifdef SCTP
    456 	/* Update the SCTP function hooks if necessary*/
    457 
    458         vec_sctp_add_ip_address = sctp_add_ip_address;
    459         vec_sctp_delete_ip_address = sctp_delete_ip_address;
    460 #endif
    461 
    462 	mutex_init(&so_pendfree_lock, MUTEX_DEFAULT, IPL_VM);
    463 	softnet_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
    464 	cv_init(&socurkva_cv, "sokva");
    465 	cv_init(&pendfree_thread_cv, "sopendfr");
    466 	soinit2();
    467 
    468 	/* Set the initial adjusted socket buffer size. */
    469 	if (sb_max_set(sb_max))
    470 		panic("bad initial sb_max value: %lu", sb_max);
    471 
    472 	socket_listener = kauth_listen_scope(KAUTH_SCOPE_NETWORK,
    473 	    socket_listener_cb, NULL);
    474 }
    475 
    476 void
    477 soinit1(void)
    478 {
    479 	int error = kthread_create(PRI_NONE, KTHREAD_MPSAFE, NULL,
    480 	    sopendfree_thread, NULL, &sopendfree_lwp, "sopendfree");
    481 	if (error)
    482 		panic("soinit1 %d", error);
    483 }
    484 
    485 /*
    486  * socreate: create a new socket of the specified type and the protocol.
    487  *
    488  * => Caller may specify another socket for lock sharing (must not be held).
    489  * => Returns the new socket without lock held.
    490  */
    491 int
    492 socreate(int dom, struct socket **aso, int type, int proto, struct lwp *l,
    493     struct socket *lockso)
    494 {
    495 	const struct protosw *prp;
    496 	struct socket *so;
    497 	uid_t uid;
    498 	int error;
    499 	kmutex_t *lock;
    500 
    501 	error = kauth_authorize_network(l->l_cred, KAUTH_NETWORK_SOCKET,
    502 	    KAUTH_REQ_NETWORK_SOCKET_OPEN, KAUTH_ARG(dom), KAUTH_ARG(type),
    503 	    KAUTH_ARG(proto));
    504 	if (error != 0)
    505 		return error;
    506 
    507 	if (proto)
    508 		prp = pffindproto(dom, proto, type);
    509 	else
    510 		prp = pffindtype(dom, type);
    511 	if (prp == NULL) {
    512 		/* no support for domain */
    513 		if (pffinddomain(dom) == 0)
    514 			return SET_ERROR(EAFNOSUPPORT);
    515 		/* no support for socket type */
    516 		if (proto == 0 && type != 0)
    517 			return SET_ERROR(EPROTOTYPE);
    518 		return SET_ERROR(EPROTONOSUPPORT);
    519 	}
    520 	if (prp->pr_usrreqs == NULL)
    521 		return SET_ERROR(EPROTONOSUPPORT);
    522 	if (prp->pr_type != type)
    523 		return SET_ERROR(EPROTOTYPE);
    524 
    525 	so = soget(true);
    526 	so->so_type = type;
    527 	so->so_proto = prp;
    528 	so->so_send = sosend;
    529 	so->so_receive = soreceive;
    530 	so->so_options = sooptions;
    531 #ifdef MBUFTRACE
    532 	so->so_rcv.sb_mowner = &prp->pr_domain->dom_mowner;
    533 	so->so_snd.sb_mowner = &prp->pr_domain->dom_mowner;
    534 	so->so_mowner = &prp->pr_domain->dom_mowner;
    535 #endif
    536 	uid = kauth_cred_geteuid(l->l_cred);
    537 	so->so_uidinfo = uid_find(uid);
    538 	so->so_egid = kauth_cred_getegid(l->l_cred);
    539 	so->so_cpid = l->l_proc->p_pid;
    540 
    541 	/*
    542 	 * Lock assigned and taken during PCB attach, unless we share
    543 	 * the lock with another socket, e.g. socketpair(2) case.
    544 	 */
    545 	if (lockso) {
    546 		/*
    547 		 * lockso->so_lock should be stable at this point, so
    548 		 * no need for atomic_load_*.
    549 		 */
    550 		lock = lockso->so_lock;
    551 		so->so_lock = lock;
    552 		mutex_obj_hold(lock);
    553 		mutex_enter(lock);
    554 	}
    555 
    556 	/* Attach the PCB (returns with the socket lock held). */
    557 	error = (*prp->pr_usrreqs->pr_attach)(so, proto);
    558 	KASSERT(solocked(so));
    559 
    560 	if (error) {
    561 		KASSERT(so->so_pcb == NULL);
    562 		so->so_state |= SS_NOFDREF;
    563 		sofree(so);
    564 		return error;
    565 	}
    566 	so->so_cred = kauth_cred_hold(l->l_cred);
    567 	sounlock(so);
    568 
    569 	*aso = so;
    570 	return 0;
    571 }
    572 
    573 /*
    574  * fsocreate: create a socket and a file descriptor associated with it.
    575  * Returns the allocated file structure in *fpp, but the descriptor
    576  * is not visible yet for the process.
    577  * Caller is responsible for calling fd_affix() for the returned *fpp once
    578  * it's socket initialization is finished successfully, or fd_abort() if it's
    579  * initialization fails.
    580  *
    581  *
    582  * => On success, write file descriptor to *fdout and *fpp and return zero.
    583  * => On failure, return non-zero; *fdout and *fpp will be undefined.
    584  */
    585 int
    586 fsocreate(int domain, struct socket **sop, int type, int proto, int *fdout,
    587     file_t **fpp, struct socket *lockso)
    588 {
    589 	lwp_t *l = curlwp;
    590 	int error, fd, flags;
    591 	struct socket *so;
    592 	file_t *fp;
    593 
    594 	flags = type & SOCK_FLAGS_MASK;
    595 	type &= ~SOCK_FLAGS_MASK;
    596 	error = socreate(domain, &so, type, proto, l, lockso);
    597 	if (error) {
    598 		return error;
    599 	}
    600 
    601 	if ((error = fd_allocfile(&fp, &fd)) != 0) {
    602 		soclose(so);
    603 		return error;
    604 	}
    605 	fd_set_exclose(l, fd, (flags & SOCK_CLOEXEC) != 0);
    606 	fd_set_foclose(l, fd, (flags & SOCK_CLOFORK) != 0);
    607 	fp->f_flag = FREAD|FWRITE|((flags & SOCK_NONBLOCK) ? FNONBLOCK : 0)|
    608 	    ((flags & SOCK_NOSIGPIPE) ? FNOSIGPIPE : 0);
    609 	fp->f_type = DTYPE_SOCKET;
    610 	fp->f_ops = &socketops;
    611 	if (flags & SOCK_NONBLOCK) {
    612 		so->so_state |= SS_NBIO;
    613 	}
    614 	fp->f_socket = so;
    615 
    616 	if (sop != NULL) {
    617 		*sop = so;
    618 	}
    619 	*fdout = fd;
    620 	*fpp = fp;
    621 	return error;
    622 }
    623 
    624 int
    625 sofamily(const struct socket *so)
    626 {
    627 	const struct protosw *pr;
    628 	const struct domain *dom;
    629 
    630 	if ((pr = so->so_proto) == NULL)
    631 		return AF_UNSPEC;
    632 	if ((dom = pr->pr_domain) == NULL)
    633 		return AF_UNSPEC;
    634 	return dom->dom_family;
    635 }
    636 
    637 int
    638 sobind(struct socket *so, struct sockaddr *nam, struct lwp *l)
    639 {
    640 	int error;
    641 
    642 	solock(so);
    643 	if (nam->sa_family != so->so_proto->pr_domain->dom_family) {
    644 		sounlock(so);
    645 		return SET_ERROR(EAFNOSUPPORT);
    646 	}
    647 	error = (*so->so_proto->pr_usrreqs->pr_bind)(so, nam, l);
    648 	sounlock(so);
    649 	return error;
    650 }
    651 
    652 int
    653 solisten(struct socket *so, int backlog, struct lwp *l)
    654 {
    655 	int error;
    656 	short oldopt, oldqlimit;
    657 
    658 	solock(so);
    659 	if ((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
    660 	    SS_ISDISCONNECTING)) != 0) {
    661 		sounlock(so);
    662 		return SET_ERROR(EINVAL);
    663 	}
    664 	oldopt = so->so_options;
    665 	oldqlimit = so->so_qlimit;
    666 	if (TAILQ_EMPTY(&so->so_q))
    667 		so->so_options |= SO_ACCEPTCONN;
    668 	if (backlog < 0)
    669 		backlog = 0;
    670 	so->so_qlimit = uimin(backlog, somaxconn);
    671 
    672 	error = (*so->so_proto->pr_usrreqs->pr_listen)(so, l);
    673 	if (error != 0) {
    674 		so->so_options = oldopt;
    675 		so->so_qlimit = oldqlimit;
    676 		sounlock(so);
    677 		return error;
    678 	}
    679 	sounlock(so);
    680 	return 0;
    681 }
    682 
    683 void
    684 sofree(struct socket *so)
    685 {
    686 	u_int refs;
    687 
    688 	KASSERT(solocked(so));
    689 
    690 	if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0) {
    691 		sounlock(so);
    692 		return;
    693 	}
    694 	if (so->so_head) {
    695 		/*
    696 		 * We must not decommission a socket that's on the accept(2)
    697 		 * queue.  If we do, then accept(2) may hang after select(2)
    698 		 * indicated that the listening socket was ready.
    699 		 */
    700 		if (!soqremque(so, 0)) {
    701 			sounlock(so);
    702 			return;
    703 		}
    704 	}
    705 	if (so->so_rcv.sb_hiwat)
    706 		(void)chgsbsize(so->so_uidinfo, &so->so_rcv.sb_hiwat, 0,
    707 		    RLIM_INFINITY);
    708 	if (so->so_snd.sb_hiwat)
    709 		(void)chgsbsize(so->so_uidinfo, &so->so_snd.sb_hiwat, 0,
    710 		    RLIM_INFINITY);
    711 	sbrelease(&so->so_snd, so);
    712 	KASSERT(!cv_has_waiters(&so->so_cv));
    713 	KASSERT(!cv_has_waiters(&so->so_rcv.sb_cv));
    714 	KASSERT(!cv_has_waiters(&so->so_snd.sb_cv));
    715 	sorflush(so);
    716 	refs = so->so_aborting;	/* XXX */
    717 	/* Remove accept filter if one is present. */
    718 	if (so->so_accf != NULL)
    719 		(void)accept_filt_clear(so);
    720 	sounlock(so);
    721 	if (refs == 0)		/* XXX */
    722 		soput(so);
    723 }
    724 
    725 /*
    726  * soclose: close a socket on last file table reference removal.
    727  * Initiate disconnect if connected.  Free socket when disconnect complete.
    728  */
    729 int
    730 soclose(struct socket *so)
    731 {
    732 	struct socket *so2;
    733 	int error = 0;
    734 
    735 	solock(so);
    736 	if (so->so_options & SO_ACCEPTCONN) {
    737 		for (;;) {
    738 			if ((so2 = TAILQ_FIRST(&so->so_q0)) != 0) {
    739 				KASSERT(solocked2(so, so2));
    740 				(void) soqremque(so2, 0);
    741 				/* soabort drops the lock. */
    742 				(void) soabort(so2);
    743 				solock(so);
    744 				continue;
    745 			}
    746 			if ((so2 = TAILQ_FIRST(&so->so_q)) != 0) {
    747 				KASSERT(solocked2(so, so2));
    748 				(void) soqremque(so2, 1);
    749 				/* soabort drops the lock. */
    750 				(void) soabort(so2);
    751 				solock(so);
    752 				continue;
    753 			}
    754 			break;
    755 		}
    756 	}
    757 	if (so->so_pcb == NULL)
    758 		goto discard;
    759 	if (so->so_state & SS_ISCONNECTED) {
    760 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
    761 			error = sodisconnect(so);
    762 			if (error)
    763 				goto drop;
    764 		}
    765 		if (so->so_options & SO_LINGER) {
    766 			if ((so->so_state & (SS_ISDISCONNECTING|SS_NBIO)) ==
    767 			    (SS_ISDISCONNECTING|SS_NBIO))
    768 				goto drop;
    769 			while (so->so_state & SS_ISCONNECTED) {
    770 				error = sowait(so, true, so->so_linger * hz);
    771 				if (error)
    772 					break;
    773 			}
    774 		}
    775 	}
    776  drop:
    777 	if (so->so_pcb) {
    778 		KASSERT(solocked(so));
    779 		(*so->so_proto->pr_usrreqs->pr_detach)(so);
    780 	}
    781  discard:
    782 	KASSERT((so->so_state & SS_NOFDREF) == 0);
    783 	kauth_cred_free(so->so_cred);
    784 	so->so_cred = NULL;
    785 	so->so_state |= SS_NOFDREF;
    786 	sofree(so);
    787 	return error;
    788 }
    789 
    790 /*
    791  * Must be called with the socket locked..  Will return with it unlocked.
    792  */
    793 int
    794 soabort(struct socket *so)
    795 {
    796 	u_int refs;
    797 	int error;
    798 
    799 	KASSERT(solocked(so));
    800 	KASSERT(so->so_head == NULL);
    801 
    802 	so->so_aborting++;		/* XXX */
    803 	error = (*so->so_proto->pr_usrreqs->pr_abort)(so);
    804 	refs = --so->so_aborting;	/* XXX */
    805 	if (error || (refs == 0)) {
    806 		sofree(so);
    807 	} else {
    808 		sounlock(so);
    809 	}
    810 	return error;
    811 }
    812 
    813 int
    814 soaccept(struct socket *so, struct sockaddr *nam)
    815 {
    816 	int error;
    817 
    818 	KASSERT(solocked(so));
    819 	KASSERT((so->so_state & SS_NOFDREF) != 0);
    820 
    821 	so->so_state &= ~SS_NOFDREF;
    822 	if ((so->so_state & SS_ISDISCONNECTED) == 0 ||
    823 	    (so->so_proto->pr_flags & PR_ABRTACPTDIS) == 0)
    824 		error = (*so->so_proto->pr_usrreqs->pr_accept)(so, nam);
    825 	else
    826 		error = SET_ERROR(ECONNABORTED);
    827 
    828 	return error;
    829 }
    830 
    831 int
    832 soconnect(struct socket *so, struct sockaddr *nam, struct lwp *l)
    833 {
    834 	int error;
    835 
    836 	KASSERT(solocked(so));
    837 
    838 	if (so->so_options & SO_ACCEPTCONN)
    839 		return SET_ERROR(EOPNOTSUPP);
    840 	/*
    841 	 * If protocol is connection-based, can only connect once.
    842 	 * Otherwise, if connected, try to disconnect first.
    843 	 * This allows user to disconnect by connecting to, e.g.,
    844 	 * a null address.
    845 	 */
    846 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
    847 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
    848 	    (error = sodisconnect(so)))) {
    849 		error = SET_ERROR(EISCONN);
    850 	} else {
    851 		if (nam->sa_family != so->so_proto->pr_domain->dom_family) {
    852 			return SET_ERROR(EAFNOSUPPORT);
    853 		}
    854 		error = (*so->so_proto->pr_usrreqs->pr_connect)(so, nam, l);
    855 	}
    856 
    857 	return error;
    858 }
    859 
    860 int
    861 soconnect2(struct socket *so1, struct socket *so2)
    862 {
    863 	KASSERT(solocked2(so1, so2));
    864 
    865 	return (*so1->so_proto->pr_usrreqs->pr_connect2)(so1, so2);
    866 }
    867 
    868 int
    869 sodisconnect(struct socket *so)
    870 {
    871 	int error;
    872 
    873 	KASSERT(solocked(so));
    874 
    875 	if ((so->so_state & SS_ISCONNECTED) == 0) {
    876 		error = SET_ERROR(ENOTCONN);
    877 	} else if (so->so_state & SS_ISDISCONNECTING) {
    878 		error = SET_ERROR(EALREADY);
    879 	} else {
    880 		error = (*so->so_proto->pr_usrreqs->pr_disconnect)(so);
    881 	}
    882 	return error;
    883 }
    884 
    885 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
    886 /*
    887  * Send on a socket.
    888  * If send must go all at once and message is larger than
    889  * send buffering, then hard error.
    890  * Lock against other senders.
    891  * If must go all at once and not enough room now, then
    892  * inform user that this would block and do nothing.
    893  * Otherwise, if nonblocking, send as much as possible.
    894  * The data to be sent is described by "uio" if nonzero,
    895  * otherwise by the mbuf chain "top" (which must be null
    896  * if uio is not).  Data provided in mbuf chain must be small
    897  * enough to send all at once.
    898  *
    899  * Returns nonzero on error, timeout or signal; callers
    900  * must check for short counts if EINTR/ERESTART are returned.
    901  * Data and control buffers are freed on return.
    902  */
    903 int
    904 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
    905 	struct mbuf *top, struct mbuf *control, int flags, struct lwp *l)
    906 {
    907 	struct mbuf **mp, *m;
    908 	long space, len, resid, clen, mlen;
    909 	int error, s, dontroute, atomic;
    910 	short wakeup_state = 0;
    911 
    912 	clen = 0;
    913 
    914 	/*
    915 	 * solock() provides atomicity of access.  splsoftnet() prevents
    916 	 * protocol processing soft interrupts from interrupting us and
    917 	 * blocking (expensive).
    918 	 */
    919 	s = splsoftnet();
    920 	solock(so);
    921 	atomic = sosendallatonce(so) || top;
    922 	if (uio)
    923 		resid = uio->uio_resid;
    924 	else
    925 		resid = top->m_pkthdr.len;
    926 	/*
    927 	 * In theory resid should be unsigned.
    928 	 * However, space must be signed, as it might be less than 0
    929 	 * if we over-committed, and we must use a signed comparison
    930 	 * of space and resid.  On the other hand, a negative resid
    931 	 * causes us to loop sending 0-length segments to the protocol.
    932 	 */
    933 	if (resid < 0) {
    934 		error = SET_ERROR(EINVAL);
    935 		goto out;
    936 	}
    937 	dontroute =
    938 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
    939 	    (so->so_proto->pr_flags & PR_ATOMIC);
    940 	l->l_ru.ru_msgsnd++;
    941 	if (control)
    942 		clen = control->m_len;
    943  restart:
    944 	if ((error = sblock(&so->so_snd, SBLOCKWAIT(flags))) != 0)
    945 		goto out;
    946 	do {
    947 		if (so->so_state & SS_CANTSENDMORE) {
    948 			error = SET_ERROR(EPIPE);
    949 			goto release;
    950 		}
    951 		if (so->so_error) {
    952 			error = SET_ERROR(so->so_error);
    953 			if ((flags & MSG_PEEK) == 0)
    954 				so->so_error = 0;
    955 			goto release;
    956 		}
    957 		if ((so->so_state & SS_ISCONNECTED) == 0) {
    958 			if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
    959 				if (resid || clen == 0) {
    960 					error = SET_ERROR(ENOTCONN);
    961 					goto release;
    962 				}
    963 			} else if (addr == NULL) {
    964 				error = SET_ERROR(EDESTADDRREQ);
    965 				goto release;
    966 			}
    967 		}
    968 		space = sbspace(&so->so_snd);
    969 		if (flags & MSG_OOB)
    970 			space += 1024;
    971 		if ((atomic && resid > so->so_snd.sb_hiwat) ||
    972 		    clen > so->so_snd.sb_hiwat) {
    973 			error = SET_ERROR(EMSGSIZE);
    974 			goto release;
    975 		}
    976 		if (space < resid + clen &&
    977 		    (atomic || space < so->so_snd.sb_lowat || space < clen)) {
    978 			if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO)) {
    979 				error = SET_ERROR(EWOULDBLOCK);
    980 				goto release;
    981 			}
    982 			sbunlock(&so->so_snd);
    983 			if (wakeup_state & SS_RESTARTSYS) {
    984 				error = SET_ERROR(ERESTART);
    985 				goto out;
    986 			}
    987 			error = sbwait(&so->so_snd);
    988 			if (error)
    989 				goto out;
    990 			wakeup_state = so->so_state;
    991 			goto restart;
    992 		}
    993 		wakeup_state = 0;
    994 		mp = &top;
    995 		space -= clen;
    996 		do {
    997 			if (uio == NULL) {
    998 				/*
    999 				 * Data is prepackaged in "top".
   1000 				 */
   1001 				resid = 0;
   1002 				if (flags & MSG_EOR)
   1003 					top->m_flags |= M_EOR;
   1004 			} else do {
   1005 				sounlock(so);
   1006 				splx(s);
   1007 				if (top == NULL) {
   1008 					m = m_gethdr(M_WAIT, MT_DATA);
   1009 					mlen = MHLEN;
   1010 					m->m_pkthdr.len = 0;
   1011 					m_reset_rcvif(m);
   1012 				} else {
   1013 					m = m_get(M_WAIT, MT_DATA);
   1014 					mlen = MLEN;
   1015 				}
   1016 				MCLAIM(m, so->so_snd.sb_mowner);
   1017 				if (sock_loan_thresh >= 0 &&
   1018 				    uio->uio_iov->iov_len >= sock_loan_thresh &&
   1019 				    space >= sock_loan_thresh &&
   1020 				    (len = sosend_loan(so, uio, m,
   1021 						       space)) != 0) {
   1022 					SOSEND_COUNTER_INCR(&sosend_loan_big);
   1023 					space -= len;
   1024 					goto have_data;
   1025 				}
   1026 				if (resid >= MINCLSIZE && space >= MCLBYTES) {
   1027 					SOSEND_COUNTER_INCR(&sosend_copy_big);
   1028 					m_clget(m, M_DONTWAIT);
   1029 					if ((m->m_flags & M_EXT) == 0)
   1030 						goto nopages;
   1031 					mlen = MCLBYTES;
   1032 					if (atomic && top == 0) {
   1033 						len = lmin(MCLBYTES - max_hdr,
   1034 						    resid);
   1035 						m->m_data += max_hdr;
   1036 					} else
   1037 						len = lmin(MCLBYTES, resid);
   1038 					space -= len;
   1039 				} else {
   1040  nopages:
   1041 					SOSEND_COUNTER_INCR(&sosend_copy_small);
   1042 					len = lmin(lmin(mlen, resid), space);
   1043 					space -= len;
   1044 					/*
   1045 					 * For datagram protocols, leave room
   1046 					 * for protocol headers in first mbuf.
   1047 					 */
   1048 					if (atomic && top == 0 && len < mlen)
   1049 						m_align(m, len);
   1050 				}
   1051 				error = uiomove(mtod(m, void *), (int)len, uio);
   1052  have_data:
   1053 				resid = uio->uio_resid;
   1054 				m->m_len = len;
   1055 				*mp = m;
   1056 				top->m_pkthdr.len += len;
   1057 				s = splsoftnet();
   1058 				solock(so);
   1059 				if (error != 0)
   1060 					goto release;
   1061 				mp = &m->m_next;
   1062 				if (resid <= 0) {
   1063 					if (flags & MSG_EOR)
   1064 						top->m_flags |= M_EOR;
   1065 					break;
   1066 				}
   1067 			} while (space > 0 && atomic);
   1068 
   1069 			if (so->so_state & SS_CANTSENDMORE) {
   1070 				error = SET_ERROR(EPIPE);
   1071 				goto release;
   1072 			}
   1073 			if (dontroute)
   1074 				so->so_options |= SO_DONTROUTE;
   1075 			if (resid > 0)
   1076 				so->so_state |= SS_MORETOCOME;
   1077 			if (flags & MSG_OOB) {
   1078 				error = (*so->so_proto->pr_usrreqs->pr_sendoob)(
   1079 				    so, top, control);
   1080 			} else {
   1081 				error = (*so->so_proto->pr_usrreqs->pr_send)(so,
   1082 				    top, addr, control, l);
   1083 			}
   1084 			if (dontroute)
   1085 				so->so_options &= ~SO_DONTROUTE;
   1086 			if (resid > 0)
   1087 				so->so_state &= ~SS_MORETOCOME;
   1088 			clen = 0;
   1089 			control = NULL;
   1090 			top = NULL;
   1091 			mp = &top;
   1092 			if (error != 0)
   1093 				goto release;
   1094 		} while (resid && space > 0);
   1095 	} while (resid);
   1096 
   1097  release:
   1098 	sbunlock(&so->so_snd);
   1099  out:
   1100 	sounlock(so);
   1101 	splx(s);
   1102 	m_freem(top);
   1103 	m_freem(control);
   1104 	return error;
   1105 }
   1106 
   1107 /*
   1108  * Following replacement or removal of the first mbuf on the first
   1109  * mbuf chain of a socket buffer, push necessary state changes back
   1110  * into the socket buffer so that other consumers see the values
   1111  * consistently.  'nextrecord' is the caller's locally stored value of
   1112  * the original value of sb->sb_mb->m_nextpkt which must be restored
   1113  * when the lead mbuf changes.  NOTE: 'nextrecord' may be NULL.
   1114  */
   1115 static void
   1116 sbsync(struct sockbuf *sb, struct mbuf *nextrecord)
   1117 {
   1118 
   1119 	KASSERT(solocked(sb->sb_so));
   1120 
   1121 	/*
   1122 	 * First, update for the new value of nextrecord.  If necessary,
   1123 	 * make it the first record.
   1124 	 */
   1125 	if (sb->sb_mb != NULL)
   1126 		sb->sb_mb->m_nextpkt = nextrecord;
   1127 	else
   1128 		sb->sb_mb = nextrecord;
   1129 
   1130         /*
   1131          * Now update any dependent socket buffer fields to reflect
   1132          * the new state.  This is an inline of SB_EMPTY_FIXUP, with
   1133          * the addition of a second clause that takes care of the
   1134          * case where sb_mb has been updated, but remains the last
   1135          * record.
   1136          */
   1137         if (sb->sb_mb == NULL) {
   1138                 sb->sb_mbtail = NULL;
   1139                 sb->sb_lastrecord = NULL;
   1140         } else if (sb->sb_mb->m_nextpkt == NULL)
   1141                 sb->sb_lastrecord = sb->sb_mb;
   1142 }
   1143 
   1144 /*
   1145  * Implement receive operations on a socket.
   1146  *
   1147  * We depend on the way that records are added to the sockbuf by sbappend*. In
   1148  * particular, each record (mbufs linked through m_next) must begin with an
   1149  * address if the protocol so specifies, followed by an optional mbuf or mbufs
   1150  * containing ancillary data, and then zero or more mbufs of data.
   1151  *
   1152  * In order to avoid blocking network interrupts for the entire time here, we
   1153  * splx() while doing the actual copy to user space. Although the sockbuf is
   1154  * locked, new data may still be appended, and thus we must maintain
   1155  * consistency of the sockbuf during that time.
   1156  *
   1157  * The caller may receive the data as a single mbuf chain by supplying an mbuf
   1158  * **mp0 for use in returning the chain. The uio is then used only for the
   1159  * count in uio_resid.
   1160  */
   1161 int
   1162 soreceive(struct socket *so, struct mbuf **paddr, struct uio *uio,
   1163     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
   1164 {
   1165 	struct lwp *l = curlwp;
   1166 	struct mbuf *m, **mp, *mt;
   1167 	size_t len, offset, moff, orig_resid;
   1168 	int atomic, flags, error, s, type;
   1169 	const struct protosw *pr;
   1170 	struct mbuf *nextrecord;
   1171 	int mbuf_removed = 0;
   1172 	const struct domain *dom;
   1173 	short wakeup_state = 0;
   1174 
   1175 	pr = so->so_proto;
   1176 	atomic = pr->pr_flags & PR_ATOMIC;
   1177 	dom = pr->pr_domain;
   1178 	mp = mp0;
   1179 	type = 0;
   1180 	orig_resid = uio->uio_resid;
   1181 
   1182 	if (paddr != NULL)
   1183 		*paddr = NULL;
   1184 	if (controlp != NULL)
   1185 		*controlp = NULL;
   1186 	if (flagsp != NULL)
   1187 		flags = *flagsp &~ MSG_EOR;
   1188 	else
   1189 		flags = 0;
   1190 
   1191 	if (flags & MSG_OOB) {
   1192 		m = m_get(M_WAIT, MT_DATA);
   1193 		solock(so);
   1194 		error = (*pr->pr_usrreqs->pr_recvoob)(so, m, flags & MSG_PEEK);
   1195 		sounlock(so);
   1196 		if (error)
   1197 			goto bad;
   1198 		do {
   1199 			error = uiomove(mtod(m, void *),
   1200 			    MIN(uio->uio_resid, m->m_len), uio);
   1201 			m = m_free(m);
   1202 		} while (uio->uio_resid > 0 && error == 0 && m);
   1203 bad:
   1204 		m_freem(m);
   1205 		return error;
   1206 	}
   1207 	if (mp != NULL)
   1208 		*mp = NULL;
   1209 
   1210 	/*
   1211 	 * solock() provides atomicity of access.  splsoftnet() prevents
   1212 	 * protocol processing soft interrupts from interrupting us and
   1213 	 * blocking (expensive).
   1214 	 */
   1215 	s = splsoftnet();
   1216 	solock(so);
   1217 restart:
   1218 	if ((error = sblock(&so->so_rcv, SBLOCKWAIT(flags))) != 0) {
   1219 		sounlock(so);
   1220 		splx(s);
   1221 		return error;
   1222 	}
   1223 	m = so->so_rcv.sb_mb;
   1224 
   1225 	/*
   1226 	 * If we have less data than requested, block awaiting more
   1227 	 * (subject to any timeout) if:
   1228 	 *   1. the current count is less than the low water mark,
   1229 	 *   2. MSG_WAITALL is set, and it is possible to do the entire
   1230 	 *	receive operation at once if we block (resid <= hiwat), or
   1231 	 *   3. MSG_DONTWAIT is not set.
   1232 	 * If MSG_WAITALL is set but resid is larger than the receive buffer,
   1233 	 * we have to do the receive in sections, and thus risk returning
   1234 	 * a short count if a timeout or signal occurs after we start.
   1235 	 */
   1236 	if (m == NULL ||
   1237 	    ((flags & MSG_DONTWAIT) == 0 &&
   1238 	     so->so_rcv.sb_cc < uio->uio_resid &&
   1239 	     (so->so_rcv.sb_cc < so->so_rcv.sb_lowat ||
   1240 	      ((flags & MSG_WAITALL) &&
   1241 	       uio->uio_resid <= so->so_rcv.sb_hiwat)) &&
   1242 	     m->m_nextpkt == NULL && !atomic)) {
   1243 #ifdef DIAGNOSTIC
   1244 		if (m == NULL && so->so_rcv.sb_cc)
   1245 			panic("receive 1");
   1246 #endif
   1247 		if (so->so_error || so->so_rerror) {
   1248 			u_short *e;
   1249 			if (m != NULL)
   1250 				goto dontblock;
   1251 			e = so->so_error ? &so->so_error : &so->so_rerror;
   1252 			error = SET_ERROR(*e);
   1253 			if ((flags & MSG_PEEK) == 0)
   1254 				*e = 0;
   1255 			goto release;
   1256 		}
   1257 		if (so->so_state & SS_CANTRCVMORE) {
   1258 			if (m != NULL)
   1259 				goto dontblock;
   1260 			else
   1261 				goto release;
   1262 		}
   1263 		for (; m != NULL; m = m->m_next)
   1264 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
   1265 				m = so->so_rcv.sb_mb;
   1266 				goto dontblock;
   1267 			}
   1268 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
   1269 		    (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
   1270 			error = SET_ERROR(ENOTCONN);
   1271 			goto release;
   1272 		}
   1273 		if (uio->uio_resid == 0)
   1274 			goto release;
   1275 		if ((so->so_state & SS_NBIO) ||
   1276 		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
   1277 			error = SET_ERROR(EWOULDBLOCK);
   1278 			goto release;
   1279 		}
   1280 		SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 1");
   1281 		SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 1");
   1282 		sbunlock(&so->so_rcv);
   1283 		if (wakeup_state & SS_RESTARTSYS)
   1284 			error = SET_ERROR(ERESTART);
   1285 		else
   1286 			error = sbwait(&so->so_rcv);
   1287 		if (error != 0) {
   1288 			sounlock(so);
   1289 			splx(s);
   1290 			return error;
   1291 		}
   1292 		wakeup_state = so->so_state;
   1293 		goto restart;
   1294 	}
   1295 
   1296 dontblock:
   1297 	/*
   1298 	 * On entry here, m points to the first record of the socket buffer.
   1299 	 * From this point onward, we maintain 'nextrecord' as a cache of the
   1300 	 * pointer to the next record in the socket buffer.  We must keep the
   1301 	 * various socket buffer pointers and local stack versions of the
   1302 	 * pointers in sync, pushing out modifications before dropping the
   1303 	 * socket lock, and re-reading them when picking it up.
   1304 	 *
   1305 	 * Otherwise, we will race with the network stack appending new data
   1306 	 * or records onto the socket buffer by using inconsistent/stale
   1307 	 * versions of the field, possibly resulting in socket buffer
   1308 	 * corruption.
   1309 	 *
   1310 	 * By holding the high-level sblock(), we prevent simultaneous
   1311 	 * readers from pulling off the front of the socket buffer.
   1312 	 */
   1313 	if (l != NULL)
   1314 		l->l_ru.ru_msgrcv++;
   1315 	KASSERT(m == so->so_rcv.sb_mb);
   1316 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 1");
   1317 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 1");
   1318 	nextrecord = m->m_nextpkt;
   1319 
   1320 	if (pr->pr_flags & PR_ADDR) {
   1321 		KASSERT(m->m_type == MT_SONAME);
   1322 		orig_resid = 0;
   1323 		if (flags & MSG_PEEK) {
   1324 			if (paddr)
   1325 				*paddr = m_copym(m, 0, m->m_len, M_DONTWAIT);
   1326 			m = m->m_next;
   1327 		} else {
   1328 			sbfree(&so->so_rcv, m);
   1329 			mbuf_removed = 1;
   1330 			if (paddr != NULL) {
   1331 				*paddr = m;
   1332 				so->so_rcv.sb_mb = m->m_next;
   1333 				m->m_next = NULL;
   1334 				m = so->so_rcv.sb_mb;
   1335 			} else {
   1336 				m = so->so_rcv.sb_mb = m_free(m);
   1337 			}
   1338 			sbsync(&so->so_rcv, nextrecord);
   1339 		}
   1340 	}
   1341 
   1342 	if (pr->pr_flags & PR_ADDR_OPT) {
   1343 		/*
   1344 		 * For SCTP we may be getting a whole message OR a partial
   1345 		 * delivery.
   1346 		 */
   1347 		if (m->m_type == MT_SONAME) {
   1348 			orig_resid = 0;
   1349 			if (flags & MSG_PEEK) {
   1350 				if (paddr)
   1351 					*paddr = m_copym(m, 0, m->m_len, M_DONTWAIT);
   1352 				m = m->m_next;
   1353 			} else {
   1354 				sbfree(&so->so_rcv, m);
   1355 				mbuf_removed = 1;
   1356 				if (paddr) {
   1357 					*paddr = m;
   1358 					so->so_rcv.sb_mb = m->m_next;
   1359 					m->m_next = 0;
   1360 					m = so->so_rcv.sb_mb;
   1361 				} else {
   1362 					m = so->so_rcv.sb_mb = m_free(m);
   1363 				}
   1364 				sbsync(&so->so_rcv, nextrecord);
   1365 			}
   1366 		}
   1367 	}
   1368 
   1369 	/*
   1370 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
   1371 	 * in the first mbuf chain on the socket buffer.  If MSG_PEEK, we
   1372 	 * just copy the data; if !MSG_PEEK, we call into the protocol to
   1373 	 * perform externalization (or freeing if controlp == NULL).
   1374 	 */
   1375 	if (__predict_false(m != NULL && m->m_type == MT_CONTROL)) {
   1376 		struct mbuf *cm = NULL, *cmn;
   1377 		struct mbuf **cme = &cm;
   1378 
   1379 		do {
   1380 			if (flags & MSG_PEEK) {
   1381 				if (controlp != NULL) {
   1382 					*controlp = m_copym(m, 0, m->m_len, M_DONTWAIT);
   1383 					controlp = (*controlp == NULL ? NULL :
   1384 					    &(*controlp)->m_next);
   1385 				}
   1386 				m = m->m_next;
   1387 			} else {
   1388 				sbfree(&so->so_rcv, m);
   1389 				so->so_rcv.sb_mb = m->m_next;
   1390 				m->m_next = NULL;
   1391 				*cme = m;
   1392 				cme = &(*cme)->m_next;
   1393 				m = so->so_rcv.sb_mb;
   1394 			}
   1395 		} while (m != NULL && m->m_type == MT_CONTROL);
   1396 		if ((flags & MSG_PEEK) == 0)
   1397 			sbsync(&so->so_rcv, nextrecord);
   1398 
   1399 		for (; cm != NULL; cm = cmn) {
   1400 			cmn = cm->m_next;
   1401 			cm->m_next = NULL;
   1402 			type = mtod(cm, struct cmsghdr *)->cmsg_type;
   1403 			if (controlp != NULL) {
   1404 				if (dom->dom_externalize != NULL &&
   1405 				    type == SCM_RIGHTS) {
   1406 					sounlock(so);
   1407 					splx(s);
   1408 					error = (*dom->dom_externalize)(cm, l,
   1409 					    ((flags & MSG_CMSG_CLOEXEC) ?
   1410 					    O_CLOEXEC : 0) |
   1411 					    ((flags & MSG_CMSG_CLOFORK) ?
   1412 					    O_CLOFORK : 0));
   1413 					s = splsoftnet();
   1414 					solock(so);
   1415 				}
   1416 				*controlp = cm;
   1417 				while (*controlp != NULL)
   1418 					controlp = &(*controlp)->m_next;
   1419 			} else {
   1420 				/*
   1421 				 * Dispose of any SCM_RIGHTS message that went
   1422 				 * through the read path rather than recv.
   1423 				 */
   1424 				if (dom->dom_dispose != NULL &&
   1425 				    type == SCM_RIGHTS) {
   1426 					sounlock(so);
   1427 					(*dom->dom_dispose)(cm);
   1428 					solock(so);
   1429 				}
   1430 				m_freem(cm);
   1431 			}
   1432 		}
   1433 		if (m != NULL)
   1434 			nextrecord = so->so_rcv.sb_mb->m_nextpkt;
   1435 		else
   1436 			nextrecord = so->so_rcv.sb_mb;
   1437 		orig_resid = 0;
   1438 	}
   1439 
   1440 	/* If m is non-NULL, we have some data to read. */
   1441 	if (__predict_true(m != NULL)) {
   1442 		type = m->m_type;
   1443 		if (type == MT_OOBDATA)
   1444 			flags |= MSG_OOB;
   1445 	}
   1446 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 2");
   1447 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 2");
   1448 
   1449 	moff = 0;
   1450 	offset = 0;
   1451 	while (m != NULL && uio->uio_resid > 0 && error == 0) {
   1452 		/*
   1453 		 * If the type of mbuf has changed, end the receive
   1454 		 * operation and do a short read.
   1455 		 */
   1456 		if (m->m_type == MT_OOBDATA) {
   1457 			if (type != MT_OOBDATA)
   1458 				break;
   1459 		} else if (type == MT_OOBDATA) {
   1460 			break;
   1461 		} else if (m->m_type == MT_CONTROL) {
   1462 			break;
   1463 		}
   1464 #ifdef DIAGNOSTIC
   1465 		else if (m->m_type != MT_DATA && m->m_type != MT_HEADER) {
   1466 			panic("%s: m_type=%d", __func__, m->m_type);
   1467 		}
   1468 #endif
   1469 
   1470 		so->so_state &= ~SS_RCVATMARK;
   1471 		wakeup_state = 0;
   1472 		len = uio->uio_resid;
   1473 		if (so->so_oobmark && len > so->so_oobmark - offset)
   1474 			len = so->so_oobmark - offset;
   1475 		if (len > m->m_len - moff)
   1476 			len = m->m_len - moff;
   1477 
   1478 		/*
   1479 		 * If mp is set, just pass back the mbufs.
   1480 		 * Otherwise copy them out via the uio, then free.
   1481 		 * Sockbuf must be consistent here (points to current mbuf,
   1482 		 * it points to next record) when we drop priority;
   1483 		 * we must note any additions to the sockbuf when we
   1484 		 * block interrupts again.
   1485 		 */
   1486 		if (mp == NULL) {
   1487 			SBLASTRECORDCHK(&so->so_rcv, "soreceive uiomove");
   1488 			SBLASTMBUFCHK(&so->so_rcv, "soreceive uiomove");
   1489 			sounlock(so);
   1490 			splx(s);
   1491 			error = uiomove(mtod(m, char *) + moff, len, uio);
   1492 			s = splsoftnet();
   1493 			solock(so);
   1494 			if (error != 0) {
   1495 				/*
   1496 				 * If any part of the record has been removed
   1497 				 * (such as the MT_SONAME mbuf, which will
   1498 				 * happen when PR_ADDR, and thus also
   1499 				 * PR_ATOMIC, is set), then drop the entire
   1500 				 * record to maintain the atomicity of the
   1501 				 * receive operation.
   1502 				 *
   1503 				 * This avoids a later panic("receive 1a")
   1504 				 * when compiled with DIAGNOSTIC.
   1505 				 */
   1506 				if (m && mbuf_removed && atomic)
   1507 					(void) sbdroprecord(&so->so_rcv);
   1508 
   1509 				goto release;
   1510 			}
   1511 		} else {
   1512 			uio->uio_resid -= len;
   1513 		}
   1514 
   1515 		if (len == m->m_len - moff) {
   1516 			if (m->m_flags & M_EOR)
   1517 				flags |= MSG_EOR;
   1518 #ifdef SCTP
   1519 			if (m->m_flags & M_NOTIFICATION)
   1520 				flags |= MSG_NOTIFICATION;
   1521 #endif
   1522 			if (flags & MSG_PEEK) {
   1523 				m = m->m_next;
   1524 				moff = 0;
   1525 			} else {
   1526 				nextrecord = m->m_nextpkt;
   1527 				sbfree(&so->so_rcv, m);
   1528 				if (mp) {
   1529 					*mp = m;
   1530 					mp = &m->m_next;
   1531 					so->so_rcv.sb_mb = m = m->m_next;
   1532 					*mp = NULL;
   1533 				} else {
   1534 					m = so->so_rcv.sb_mb = m_free(m);
   1535 				}
   1536 				/*
   1537 				 * If m != NULL, we also know that
   1538 				 * so->so_rcv.sb_mb != NULL.
   1539 				 */
   1540 				KASSERT(so->so_rcv.sb_mb == m);
   1541 				if (m) {
   1542 					m->m_nextpkt = nextrecord;
   1543 					if (nextrecord == NULL)
   1544 						so->so_rcv.sb_lastrecord = m;
   1545 				} else {
   1546 					so->so_rcv.sb_mb = nextrecord;
   1547 					SB_EMPTY_FIXUP(&so->so_rcv);
   1548 				}
   1549 				SBLASTRECORDCHK(&so->so_rcv, "soreceive 3");
   1550 				SBLASTMBUFCHK(&so->so_rcv, "soreceive 3");
   1551 			}
   1552 		} else if (flags & MSG_PEEK) {
   1553 			moff += len;
   1554 		} else {
   1555 			if (mp != NULL) {
   1556 				mt = m_copym(m, 0, len, M_NOWAIT);
   1557 				if (__predict_false(mt == NULL)) {
   1558 					sounlock(so);
   1559 					mt = m_copym(m, 0, len, M_WAIT);
   1560 					solock(so);
   1561 				}
   1562 				*mp = mt;
   1563 			}
   1564 			m->m_data += len;
   1565 			m->m_len -= len;
   1566 			so->so_rcv.sb_cc -= len;
   1567 		}
   1568 
   1569 		if (so->so_oobmark) {
   1570 			if ((flags & MSG_PEEK) == 0) {
   1571 				so->so_oobmark -= len;
   1572 				if (so->so_oobmark == 0) {
   1573 					so->so_state |= SS_RCVATMARK;
   1574 					break;
   1575 				}
   1576 			} else {
   1577 				offset += len;
   1578 				if (offset == so->so_oobmark)
   1579 					break;
   1580 			}
   1581 		} else {
   1582 			so->so_state &= ~SS_POLLRDBAND;
   1583 		}
   1584 		if (flags & MSG_EOR)
   1585 			break;
   1586 
   1587 		/*
   1588 		 * If the MSG_WAITALL flag is set (for non-atomic socket),
   1589 		 * we must not quit until "uio->uio_resid == 0" or an error
   1590 		 * termination.  If a signal/timeout occurs, return
   1591 		 * with a short count but without error.
   1592 		 * Keep sockbuf locked against other readers.
   1593 		 */
   1594 		while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
   1595 		    !sosendallatonce(so) && !nextrecord) {
   1596 			if (so->so_error || so->so_rerror ||
   1597 			    so->so_state & SS_CANTRCVMORE)
   1598 				break;
   1599 			/*
   1600 			 * If we are peeking and the socket receive buffer is
   1601 			 * full, stop since we can't get more data to peek at.
   1602 			 */
   1603 			if ((flags & MSG_PEEK) && sbspace(&so->so_rcv) <= 0)
   1604 				break;
   1605 			/*
   1606 			 * If we've drained the socket buffer, tell the
   1607 			 * protocol in case it needs to do something to
   1608 			 * get it filled again.
   1609 			 */
   1610 			if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb)
   1611 				(*pr->pr_usrreqs->pr_rcvd)(so, flags, l);
   1612 			SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2");
   1613 			SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2");
   1614 			if (wakeup_state & SS_RESTARTSYS)
   1615 				error = SET_ERROR(ERESTART);
   1616 			else
   1617 				error = sbwait(&so->so_rcv);
   1618 			if (error != 0) {
   1619 				sbunlock(&so->so_rcv);
   1620 				sounlock(so);
   1621 				splx(s);
   1622 				return 0;
   1623 			}
   1624 			if ((m = so->so_rcv.sb_mb) != NULL)
   1625 				nextrecord = m->m_nextpkt;
   1626 			wakeup_state = so->so_state;
   1627 		}
   1628 	}
   1629 
   1630 	if (m && atomic) {
   1631 		flags |= MSG_TRUNC;
   1632 		if ((flags & MSG_PEEK) == 0)
   1633 			(void) sbdroprecord(&so->so_rcv);
   1634 	}
   1635 	if ((flags & MSG_PEEK) == 0) {
   1636 		if (m == NULL) {
   1637 			/*
   1638 			 * First part is an inline SB_EMPTY_FIXUP().  Second
   1639 			 * part makes sure sb_lastrecord is up-to-date if
   1640 			 * there is still data in the socket buffer.
   1641 			 */
   1642 			so->so_rcv.sb_mb = nextrecord;
   1643 			if (so->so_rcv.sb_mb == NULL) {
   1644 				so->so_rcv.sb_mbtail = NULL;
   1645 				so->so_rcv.sb_lastrecord = NULL;
   1646 			} else if (nextrecord->m_nextpkt == NULL)
   1647 				so->so_rcv.sb_lastrecord = nextrecord;
   1648 		}
   1649 		SBLASTRECORDCHK(&so->so_rcv, "soreceive 4");
   1650 		SBLASTMBUFCHK(&so->so_rcv, "soreceive 4");
   1651 		if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
   1652 			(*pr->pr_usrreqs->pr_rcvd)(so, flags, l);
   1653 	}
   1654 	if (orig_resid == uio->uio_resid && orig_resid &&
   1655 	    (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
   1656 		sbunlock(&so->so_rcv);
   1657 		goto restart;
   1658 	}
   1659 
   1660 	if (flagsp != NULL)
   1661 		*flagsp |= flags;
   1662 release:
   1663 	sbunlock(&so->so_rcv);
   1664 	sounlock(so);
   1665 	splx(s);
   1666 	return error;
   1667 }
   1668 
   1669 int
   1670 soshutdown(struct socket *so, int how)
   1671 {
   1672 	const struct protosw *pr;
   1673 	int error;
   1674 
   1675 	KASSERT(solocked(so));
   1676 
   1677 	pr = so->so_proto;
   1678 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
   1679 		return SET_ERROR(EINVAL);
   1680 
   1681 	if (how == SHUT_RD || how == SHUT_RDWR) {
   1682 		sorflush(so);
   1683 		error = 0;
   1684 	}
   1685 	if (how == SHUT_WR || how == SHUT_RDWR)
   1686 		error = (*pr->pr_usrreqs->pr_shutdown)(so);
   1687 
   1688 	return error;
   1689 }
   1690 
   1691 void
   1692 sorestart(struct socket *so)
   1693 {
   1694 	/*
   1695 	 * An application has called close() on an fd on which another
   1696 	 * of its threads has called a socket system call.
   1697 	 * Mark this and wake everyone up, and code that would block again
   1698 	 * instead returns ERESTART.
   1699 	 * On system call re-entry the fd is validated and EBADF returned.
   1700 	 * Any other fd will block again on the 2nd syscall.
   1701 	 */
   1702 	solock(so);
   1703 	so->so_state |= SS_RESTARTSYS;
   1704 	cv_broadcast(&so->so_cv);
   1705 	cv_broadcast(&so->so_snd.sb_cv);
   1706 	cv_broadcast(&so->so_rcv.sb_cv);
   1707 	sounlock(so);
   1708 }
   1709 
   1710 void
   1711 sorflush(struct socket *so)
   1712 {
   1713 	struct sockbuf *sb, asb;
   1714 	const struct protosw *pr;
   1715 
   1716 	KASSERT(solocked(so));
   1717 
   1718 	sb = &so->so_rcv;
   1719 	pr = so->so_proto;
   1720 	socantrcvmore(so);
   1721 	sb->sb_flags |= SB_NOINTR;
   1722 	(void )sblock(sb, M_WAITOK);
   1723 	sbunlock(sb);
   1724 	asb = *sb;
   1725 	/*
   1726 	 * Clear most of the sockbuf structure, but leave some of the
   1727 	 * fields valid.
   1728 	 */
   1729 	memset(&sb->sb_startzero, 0,
   1730 	    sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
   1731 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose) {
   1732 		sounlock(so);
   1733 		(*pr->pr_domain->dom_dispose)(asb.sb_mb);
   1734 		solock(so);
   1735 	}
   1736 	sbrelease(&asb, so);
   1737 }
   1738 
   1739 /*
   1740  * internal set SOL_SOCKET options
   1741  */
   1742 static int
   1743 sosetopt1(struct socket *so, const struct sockopt *sopt)
   1744 {
   1745 	int error, opt;
   1746 	int optval = 0; /* XXX: gcc */
   1747 	struct linger l;
   1748 	struct timeval tv;
   1749 
   1750 	opt = sopt->sopt_name;
   1751 
   1752 	switch (opt) {
   1753 
   1754 	case SO_ACCEPTFILTER:
   1755 		error = accept_filt_setopt(so, sopt);
   1756 		KASSERT(solocked(so));
   1757 		break;
   1758 
   1759 	case SO_LINGER:
   1760 		error = sockopt_get(sopt, &l, sizeof(l));
   1761 		solock(so);
   1762 		if (error)
   1763 			break;
   1764 		if (l.l_linger < 0 || l.l_linger > USHRT_MAX ||
   1765 		    l.l_linger > (INT_MAX / hz)) {
   1766 			error = SET_ERROR(EDOM);
   1767 			break;
   1768 		}
   1769 		so->so_linger = l.l_linger;
   1770 		if (l.l_onoff)
   1771 			so->so_options |= SO_LINGER;
   1772 		else
   1773 			so->so_options &= ~SO_LINGER;
   1774 		break;
   1775 
   1776 	case SO_DEBUG:
   1777 	case SO_KEEPALIVE:
   1778 	case SO_DONTROUTE:
   1779 	case SO_USELOOPBACK:
   1780 	case SO_BROADCAST:
   1781 	case SO_REUSEADDR:
   1782 	case SO_REUSEPORT:
   1783 	case SO_OOBINLINE:
   1784 	case SO_TIMESTAMP:
   1785 	case SO_NOSIGPIPE:
   1786 	case SO_RERROR:
   1787 		error = sockopt_getint(sopt, &optval);
   1788 		solock(so);
   1789 		if (error)
   1790 			break;
   1791 		if (optval)
   1792 			so->so_options |= opt;
   1793 		else
   1794 			so->so_options &= ~opt;
   1795 		break;
   1796 
   1797 	case SO_SNDBUF:
   1798 	case SO_RCVBUF:
   1799 	case SO_SNDLOWAT:
   1800 	case SO_RCVLOWAT:
   1801 		error = sockopt_getint(sopt, &optval);
   1802 		solock(so);
   1803 		if (error)
   1804 			break;
   1805 
   1806 		/*
   1807 		 * Values < 1 make no sense for any of these
   1808 		 * options, so disallow them.
   1809 		 */
   1810 		if (optval < 1) {
   1811 			error = SET_ERROR(EINVAL);
   1812 			break;
   1813 		}
   1814 
   1815 		switch (opt) {
   1816 		case SO_SNDBUF:
   1817 			if (sbreserve(&so->so_snd, (u_long)optval, so) == 0) {
   1818 				error = SET_ERROR(ENOBUFS);
   1819 				break;
   1820 			}
   1821 			if (sofixedbuf)
   1822 				so->so_snd.sb_flags &= ~SB_AUTOSIZE;
   1823 			break;
   1824 
   1825 		case SO_RCVBUF:
   1826 			if (sbreserve(&so->so_rcv, (u_long)optval, so) == 0) {
   1827 				error = SET_ERROR(ENOBUFS);
   1828 				break;
   1829 			}
   1830 			if (sofixedbuf)
   1831 				so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
   1832 			break;
   1833 
   1834 		/*
   1835 		 * Make sure the low-water is never greater than
   1836 		 * the high-water.
   1837 		 */
   1838 		case SO_SNDLOWAT:
   1839 			if (optval > so->so_snd.sb_hiwat)
   1840 				optval = so->so_snd.sb_hiwat;
   1841 
   1842 			so->so_snd.sb_lowat = optval;
   1843 			break;
   1844 
   1845 		case SO_RCVLOWAT:
   1846 			if (optval > so->so_rcv.sb_hiwat)
   1847 				optval = so->so_rcv.sb_hiwat;
   1848 
   1849 			so->so_rcv.sb_lowat = optval;
   1850 			break;
   1851 		}
   1852 		break;
   1853 
   1854 	case SO_SNDTIMEO:
   1855 	case SO_RCVTIMEO:
   1856 		solock(so);
   1857 		error = sockopt_get(sopt, &tv, sizeof(tv));
   1858 		if (error)
   1859 			break;
   1860 
   1861 		if (tv.tv_sec < 0 || tv.tv_usec < 0 || tv.tv_usec >= 1000000) {
   1862 			error = SET_ERROR(EDOM);
   1863 			break;
   1864 		}
   1865 		if (tv.tv_sec > (INT_MAX - tv.tv_usec / tick) / hz) {
   1866 			error = SET_ERROR(EDOM);
   1867 			break;
   1868 		}
   1869 
   1870 		optval = tv.tv_sec * hz + tv.tv_usec / tick;
   1871 		if (optval == 0 && tv.tv_usec != 0)
   1872 			optval = 1;
   1873 
   1874 		switch (opt) {
   1875 		case SO_SNDTIMEO:
   1876 			so->so_snd.sb_timeo = optval;
   1877 			break;
   1878 		case SO_RCVTIMEO:
   1879 			so->so_rcv.sb_timeo = optval;
   1880 			break;
   1881 		}
   1882 		break;
   1883 
   1884 	default:
   1885 		MODULE_HOOK_CALL(uipc_socket_50_setopt1_hook,
   1886 		    (opt, so, sopt), enosys(), error);
   1887 		if (error == ENOSYS || error == EPASSTHROUGH) {
   1888 			solock(so);
   1889 			error = SET_ERROR(ENOPROTOOPT);
   1890 		}
   1891 		break;
   1892 	}
   1893 	KASSERT(solocked(so));
   1894 	return error;
   1895 }
   1896 
   1897 int
   1898 sosetopt(struct socket *so, struct sockopt *sopt)
   1899 {
   1900 	int error, prerr;
   1901 
   1902 	if (sopt->sopt_level == SOL_SOCKET) {
   1903 		error = sosetopt1(so, sopt);
   1904 		KASSERT(solocked(so));
   1905 	} else {
   1906 		error = SET_ERROR(ENOPROTOOPT);
   1907 		solock(so);
   1908 	}
   1909 
   1910 	if ((error == 0 || error == ENOPROTOOPT) &&
   1911 	    so->so_proto != NULL && so->so_proto->pr_ctloutput != NULL) {
   1912 		/* give the protocol stack a shot */
   1913 		prerr = (*so->so_proto->pr_ctloutput)(PRCO_SETOPT, so, sopt);
   1914 		if (prerr == 0)
   1915 			error = 0;
   1916 		else if (prerr != ENOPROTOOPT)
   1917 			error = prerr;
   1918 	}
   1919 	sounlock(so);
   1920 	return error;
   1921 }
   1922 
   1923 /*
   1924  * so_setsockopt() is a wrapper providing a sockopt structure for sosetopt()
   1925  */
   1926 int
   1927 so_setsockopt(struct lwp *l, struct socket *so, int level, int name,
   1928     const void *val, size_t valsize)
   1929 {
   1930 	struct sockopt sopt;
   1931 	int error;
   1932 
   1933 	KASSERT(valsize == 0 || val != NULL);
   1934 
   1935 	sockopt_init(&sopt, level, name, valsize);
   1936 	sockopt_set(&sopt, val, valsize);
   1937 
   1938 	error = sosetopt(so, &sopt);
   1939 
   1940 	sockopt_destroy(&sopt);
   1941 
   1942 	return error;
   1943 }
   1944 
   1945 /*
   1946  * internal get SOL_SOCKET options
   1947  */
   1948 static int
   1949 sogetopt1(struct socket *so, struct sockopt *sopt)
   1950 {
   1951 	int error, optval, opt;
   1952 	struct linger l;
   1953 	struct timeval tv;
   1954 
   1955 	switch ((opt = sopt->sopt_name)) {
   1956 
   1957 	case SO_ACCEPTFILTER:
   1958 		error = accept_filt_getopt(so, sopt);
   1959 		break;
   1960 
   1961 	case SO_LINGER:
   1962 		l.l_onoff = (so->so_options & SO_LINGER) ? 1 : 0;
   1963 		l.l_linger = so->so_linger;
   1964 
   1965 		error = sockopt_set(sopt, &l, sizeof(l));
   1966 		break;
   1967 
   1968 	case SO_USELOOPBACK:
   1969 	case SO_DONTROUTE:
   1970 	case SO_DEBUG:
   1971 	case SO_KEEPALIVE:
   1972 	case SO_REUSEADDR:
   1973 	case SO_REUSEPORT:
   1974 	case SO_BROADCAST:
   1975 	case SO_OOBINLINE:
   1976 	case SO_TIMESTAMP:
   1977 	case SO_NOSIGPIPE:
   1978 	case SO_RERROR:
   1979 	case SO_ACCEPTCONN:
   1980 		error = sockopt_setint(sopt, (so->so_options & opt) ? 1 : 0);
   1981 		break;
   1982 
   1983 	case SO_TYPE:
   1984 		error = sockopt_setint(sopt, so->so_type);
   1985 		break;
   1986 
   1987 	case SO_ERROR:
   1988 		if (so->so_error == 0) {
   1989 			so->so_error = so->so_rerror;
   1990 			so->so_rerror = 0;
   1991 		}
   1992 		error = sockopt_setint(sopt, so->so_error);
   1993 		so->so_error = 0;
   1994 		break;
   1995 
   1996 	case SO_SNDBUF:
   1997 		error = sockopt_setint(sopt, so->so_snd.sb_hiwat);
   1998 		break;
   1999 
   2000 	case SO_RCVBUF:
   2001 		error = sockopt_setint(sopt, so->so_rcv.sb_hiwat);
   2002 		break;
   2003 
   2004 	case SO_SNDLOWAT:
   2005 		error = sockopt_setint(sopt, so->so_snd.sb_lowat);
   2006 		break;
   2007 
   2008 	case SO_RCVLOWAT:
   2009 		error = sockopt_setint(sopt, so->so_rcv.sb_lowat);
   2010 		break;
   2011 
   2012 	case SO_SNDTIMEO:
   2013 	case SO_RCVTIMEO:
   2014 		optval = (opt == SO_SNDTIMEO ?
   2015 		     so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
   2016 
   2017 		memset(&tv, 0, sizeof(tv));
   2018 		tv.tv_sec = optval / hz;
   2019 		tv.tv_usec = (optval % hz) * tick;
   2020 
   2021 		error = sockopt_set(sopt, &tv, sizeof(tv));
   2022 		break;
   2023 
   2024 	case SO_OVERFLOWED:
   2025 		error = sockopt_setint(sopt, so->so_rcv.sb_overflowed);
   2026 		break;
   2027 
   2028 	default:
   2029 		MODULE_HOOK_CALL(uipc_socket_50_getopt1_hook,
   2030 		    (opt, so, sopt), enosys(), error);
   2031 		if (error)
   2032 			error = SET_ERROR(ENOPROTOOPT);
   2033 		break;
   2034 	}
   2035 
   2036 	return error;
   2037 }
   2038 
   2039 int
   2040 sogetopt(struct socket *so, struct sockopt *sopt)
   2041 {
   2042 	int error;
   2043 
   2044 	solock(so);
   2045 	if (sopt->sopt_level != SOL_SOCKET) {
   2046 		if (so->so_proto && so->so_proto->pr_ctloutput) {
   2047 			error = ((*so->so_proto->pr_ctloutput)
   2048 			    (PRCO_GETOPT, so, sopt));
   2049 		} else
   2050 			error = SET_ERROR(ENOPROTOOPT);
   2051 	} else {
   2052 		error = sogetopt1(so, sopt);
   2053 	}
   2054 	sounlock(so);
   2055 	return error;
   2056 }
   2057 
   2058 /*
   2059  * alloc sockopt data buffer buffer
   2060  *	- will be released at destroy
   2061  */
   2062 static int
   2063 sockopt_alloc(struct sockopt *sopt, size_t len, km_flag_t kmflag)
   2064 {
   2065 	void *data;
   2066 
   2067 	KASSERT(sopt->sopt_size == 0);
   2068 
   2069 	if (len > sizeof(sopt->sopt_buf)) {
   2070 		data = kmem_zalloc(len, kmflag);
   2071 		if (data == NULL)
   2072 			return SET_ERROR(ENOMEM);
   2073 		sopt->sopt_data = data;
   2074 	} else
   2075 		sopt->sopt_data = sopt->sopt_buf;
   2076 
   2077 	sopt->sopt_size = len;
   2078 	return 0;
   2079 }
   2080 
   2081 /*
   2082  * initialise sockopt storage
   2083  *	- MAY sleep during allocation
   2084  */
   2085 void
   2086 sockopt_init(struct sockopt *sopt, int level, int name, size_t size)
   2087 {
   2088 
   2089 	memset(sopt, 0, sizeof(*sopt));
   2090 
   2091 	sopt->sopt_level = level;
   2092 	sopt->sopt_name = name;
   2093 	(void)sockopt_alloc(sopt, size, KM_SLEEP);
   2094 }
   2095 
   2096 /*
   2097  * destroy sockopt storage
   2098  *	- will release any held memory references
   2099  */
   2100 void
   2101 sockopt_destroy(struct sockopt *sopt)
   2102 {
   2103 
   2104 	if (sopt->sopt_data != sopt->sopt_buf)
   2105 		kmem_free(sopt->sopt_data, sopt->sopt_size);
   2106 
   2107 	memset(sopt, 0, sizeof(*sopt));
   2108 }
   2109 
   2110 /*
   2111  * set sockopt value
   2112  *	- value is copied into sockopt
   2113  *	- memory is allocated when necessary, will not sleep
   2114  */
   2115 int
   2116 sockopt_set(struct sockopt *sopt, const void *buf, size_t len)
   2117 {
   2118 	int error;
   2119 
   2120 	if (sopt->sopt_size == 0) {
   2121 		error = sockopt_alloc(sopt, len, KM_NOSLEEP);
   2122 		if (error)
   2123 			return error;
   2124 	}
   2125 
   2126 	sopt->sopt_retsize = MIN(sopt->sopt_size, len);
   2127 	if (sopt->sopt_retsize > 0) {
   2128 		memcpy(sopt->sopt_data, buf, sopt->sopt_retsize);
   2129 	}
   2130 
   2131 	return 0;
   2132 }
   2133 
   2134 /*
   2135  * common case of set sockopt integer value
   2136  */
   2137 int
   2138 sockopt_setint(struct sockopt *sopt, int val)
   2139 {
   2140 
   2141 	return sockopt_set(sopt, &val, sizeof(int));
   2142 }
   2143 
   2144 /*
   2145  * get sockopt value
   2146  *	- correct size must be given
   2147  */
   2148 int
   2149 sockopt_get(const struct sockopt *sopt, void *buf, size_t len)
   2150 {
   2151 
   2152 	if (sopt->sopt_size != len)
   2153 		return SET_ERROR(EINVAL);
   2154 
   2155 	memcpy(buf, sopt->sopt_data, len);
   2156 	return 0;
   2157 }
   2158 
   2159 /*
   2160  * common case of get sockopt integer value
   2161  */
   2162 int
   2163 sockopt_getint(const struct sockopt *sopt, int *valp)
   2164 {
   2165 
   2166 	return sockopt_get(sopt, valp, sizeof(int));
   2167 }
   2168 
   2169 /*
   2170  * set sockopt value from mbuf
   2171  *	- ONLY for legacy code
   2172  *	- mbuf is released by sockopt
   2173  *	- will not sleep
   2174  */
   2175 int
   2176 sockopt_setmbuf(struct sockopt *sopt, struct mbuf *m)
   2177 {
   2178 	size_t len;
   2179 	int error;
   2180 
   2181 	len = m_length(m);
   2182 
   2183 	if (sopt->sopt_size == 0) {
   2184 		error = sockopt_alloc(sopt, len, KM_NOSLEEP);
   2185 		if (error)
   2186 			return error;
   2187 	}
   2188 
   2189 	sopt->sopt_retsize = MIN(sopt->sopt_size, len);
   2190 	m_copydata(m, 0, sopt->sopt_retsize, sopt->sopt_data);
   2191 	m_freem(m);
   2192 
   2193 	return 0;
   2194 }
   2195 
   2196 /*
   2197  * get sockopt value into mbuf
   2198  *	- ONLY for legacy code
   2199  *	- mbuf to be released by the caller
   2200  *	- will not sleep
   2201  */
   2202 struct mbuf *
   2203 sockopt_getmbuf(const struct sockopt *sopt)
   2204 {
   2205 	struct mbuf *m;
   2206 
   2207 	if (sopt->sopt_size > MCLBYTES)
   2208 		return NULL;
   2209 
   2210 	m = m_get(M_DONTWAIT, MT_SOOPTS);
   2211 	if (m == NULL)
   2212 		return NULL;
   2213 
   2214 	if (sopt->sopt_size > MLEN) {
   2215 		MCLGET(m, M_DONTWAIT);
   2216 		if ((m->m_flags & M_EXT) == 0) {
   2217 			m_free(m);
   2218 			return NULL;
   2219 		}
   2220 	}
   2221 
   2222 	memcpy(mtod(m, void *), sopt->sopt_data, sopt->sopt_size);
   2223 	m->m_len = sopt->sopt_size;
   2224 
   2225 	return m;
   2226 }
   2227 
   2228 void
   2229 sohasoutofband(struct socket *so)
   2230 {
   2231 
   2232 	so->so_state |= SS_POLLRDBAND;
   2233 	fownsignal(so->so_pgid, SIGURG, POLL_PRI, POLLPRI|POLLRDBAND, so);
   2234 	selnotify(&so->so_rcv.sb_sel, POLLPRI | POLLRDBAND, NOTE_SUBMIT);
   2235 }
   2236 
   2237 static void
   2238 filt_sordetach(struct knote *kn)
   2239 {
   2240 	struct socket *so;
   2241 
   2242 	so = ((file_t *)kn->kn_obj)->f_socket;
   2243 	solock(so);
   2244 	if (selremove_knote(&so->so_rcv.sb_sel, kn))
   2245 		so->so_rcv.sb_flags &= ~SB_KNOTE;
   2246 	sounlock(so);
   2247 }
   2248 
   2249 /*ARGSUSED*/
   2250 static int
   2251 filt_soread(struct knote *kn, long hint)
   2252 {
   2253 	struct socket *so;
   2254 	int rv;
   2255 
   2256 	so = ((file_t *)kn->kn_obj)->f_socket;
   2257 	if (hint != NOTE_SUBMIT)
   2258 		solock(so);
   2259 	kn->kn_data = so->so_rcv.sb_cc;
   2260 	if (so->so_state & SS_CANTRCVMORE) {
   2261 		knote_set_eof(kn, 0);
   2262 		kn->kn_fflags = so->so_error;
   2263 		rv = 1;
   2264 	} else if (so->so_error || so->so_rerror)
   2265 		rv = 1;
   2266 	else if (kn->kn_sfflags & NOTE_LOWAT)
   2267 		rv = (kn->kn_data >= kn->kn_sdata);
   2268 	else
   2269 		rv = (kn->kn_data >= so->so_rcv.sb_lowat);
   2270 	if (hint != NOTE_SUBMIT)
   2271 		sounlock(so);
   2272 	return rv;
   2273 }
   2274 
   2275 static void
   2276 filt_sowdetach(struct knote *kn)
   2277 {
   2278 	struct socket *so;
   2279 
   2280 	so = ((file_t *)kn->kn_obj)->f_socket;
   2281 	solock(so);
   2282 	if (selremove_knote(&so->so_snd.sb_sel, kn))
   2283 		so->so_snd.sb_flags &= ~SB_KNOTE;
   2284 	sounlock(so);
   2285 }
   2286 
   2287 /*ARGSUSED*/
   2288 static int
   2289 filt_sowrite(struct knote *kn, long hint)
   2290 {
   2291 	struct socket *so;
   2292 	int rv;
   2293 
   2294 	so = ((file_t *)kn->kn_obj)->f_socket;
   2295 	if (hint != NOTE_SUBMIT)
   2296 		solock(so);
   2297 	kn->kn_data = sbspace(&so->so_snd);
   2298 	if (so->so_state & SS_CANTSENDMORE) {
   2299 		knote_set_eof(kn, 0);
   2300 		kn->kn_fflags = so->so_error;
   2301 		rv = 1;
   2302 	} else if (so->so_error)
   2303 		rv = 1;
   2304 	else if (((so->so_state & SS_ISCONNECTED) == 0) &&
   2305 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
   2306 		rv = 0;
   2307 	else if (kn->kn_sfflags & NOTE_LOWAT)
   2308 		rv = (kn->kn_data >= kn->kn_sdata);
   2309 	else
   2310 		rv = (kn->kn_data >= so->so_snd.sb_lowat);
   2311 	if (hint != NOTE_SUBMIT)
   2312 		sounlock(so);
   2313 	return rv;
   2314 }
   2315 
   2316 static int
   2317 filt_soempty(struct knote *kn, long hint)
   2318 {
   2319 	struct socket *so;
   2320 	int rv;
   2321 
   2322 	so = ((file_t *)kn->kn_obj)->f_socket;
   2323 	if (hint != NOTE_SUBMIT)
   2324 		solock(so);
   2325 	rv = (kn->kn_data = sbused(&so->so_snd)) == 0 ||
   2326 	     (so->so_options & SO_ACCEPTCONN) != 0;
   2327 	if (hint != NOTE_SUBMIT)
   2328 		sounlock(so);
   2329 	return rv;
   2330 }
   2331 
   2332 /*ARGSUSED*/
   2333 static int
   2334 filt_solisten(struct knote *kn, long hint)
   2335 {
   2336 	struct socket *so;
   2337 	int rv;
   2338 
   2339 	so = ((file_t *)kn->kn_obj)->f_socket;
   2340 
   2341 	/*
   2342 	 * Set kn_data to number of incoming connections, not
   2343 	 * counting partial (incomplete) connections.
   2344 	 */
   2345 	if (hint != NOTE_SUBMIT)
   2346 		solock(so);
   2347 	kn->kn_data = so->so_qlen;
   2348 	rv = (kn->kn_data > 0);
   2349 	if (hint != NOTE_SUBMIT)
   2350 		sounlock(so);
   2351 	return rv;
   2352 }
   2353 
   2354 static const struct filterops solisten_filtops = {
   2355 	.f_flags = FILTEROP_ISFD | FILTEROP_MPSAFE,
   2356 	.f_attach = NULL,
   2357 	.f_detach = filt_sordetach,
   2358 	.f_event = filt_solisten,
   2359 };
   2360 
   2361 static const struct filterops soread_filtops = {
   2362 	.f_flags = FILTEROP_ISFD | FILTEROP_MPSAFE,
   2363 	.f_attach = NULL,
   2364 	.f_detach = filt_sordetach,
   2365 	.f_event = filt_soread,
   2366 };
   2367 
   2368 static const struct filterops sowrite_filtops = {
   2369 	.f_flags = FILTEROP_ISFD | FILTEROP_MPSAFE,
   2370 	.f_attach = NULL,
   2371 	.f_detach = filt_sowdetach,
   2372 	.f_event = filt_sowrite,
   2373 };
   2374 
   2375 static const struct filterops soempty_filtops = {
   2376 	.f_flags = FILTEROP_ISFD | FILTEROP_MPSAFE,
   2377 	.f_attach = NULL,
   2378 	.f_detach = filt_sowdetach,
   2379 	.f_event = filt_soempty,
   2380 };
   2381 
   2382 int
   2383 soo_kqfilter(struct file *fp, struct knote *kn)
   2384 {
   2385 	struct socket *so;
   2386 	struct sockbuf *sb;
   2387 
   2388 	so = ((file_t *)kn->kn_obj)->f_socket;
   2389 	solock(so);
   2390 	switch (kn->kn_filter) {
   2391 	case EVFILT_READ:
   2392 		if (so->so_options & SO_ACCEPTCONN)
   2393 			kn->kn_fop = &solisten_filtops;
   2394 		else
   2395 			kn->kn_fop = &soread_filtops;
   2396 		sb = &so->so_rcv;
   2397 		break;
   2398 	case EVFILT_WRITE:
   2399 		kn->kn_fop = &sowrite_filtops;
   2400 		sb = &so->so_snd;
   2401 
   2402 #ifdef PIPE_SOCKETPAIR
   2403 		if (so->so_state & SS_ISAPIPE) {
   2404 			/* Other end of pipe has been closed. */
   2405 			if (so->so_state & SS_ISDISCONNECTED) {
   2406 				sounlock(so);
   2407 				return SET_ERROR(EBADF);
   2408 			}
   2409 		}
   2410 #endif
   2411 		break;
   2412 	case EVFILT_EMPTY:
   2413 		kn->kn_fop = &soempty_filtops;
   2414 		sb = &so->so_snd;
   2415 		break;
   2416 	default:
   2417 		sounlock(so);
   2418 		return SET_ERROR(EINVAL);
   2419 	}
   2420 	selrecord_knote(&sb->sb_sel, kn);
   2421 	sb->sb_flags |= SB_KNOTE;
   2422 	sounlock(so);
   2423 	return 0;
   2424 }
   2425 
   2426 static int
   2427 sodopoll(struct socket *so, int events)
   2428 {
   2429 	int revents;
   2430 
   2431 	revents = 0;
   2432 
   2433 	if (events & (POLLIN | POLLRDNORM))
   2434 		if (soreadable(so))
   2435 			revents |= events & (POLLIN | POLLRDNORM);
   2436 
   2437 	if (events & (POLLOUT | POLLWRNORM))
   2438 		if (sowritable(so))
   2439 			revents |= events & (POLLOUT | POLLWRNORM);
   2440 
   2441 	if (events & (POLLPRI | POLLRDBAND))
   2442 		if (so->so_state & SS_POLLRDBAND)
   2443 			revents |= events & (POLLPRI | POLLRDBAND);
   2444 
   2445 	return revents;
   2446 }
   2447 
   2448 int
   2449 sopoll(struct socket *so, int events)
   2450 {
   2451 	int revents = 0;
   2452 
   2453 #ifndef DIAGNOSTIC
   2454 	/*
   2455 	 * Do a quick, unlocked check in expectation that the socket
   2456 	 * will be ready for I/O.  Don't do this check if DIAGNOSTIC,
   2457 	 * as the solocked() assertions will fail.
   2458 	 */
   2459 	if ((revents = sodopoll(so, events)) != 0)
   2460 		return revents;
   2461 #endif
   2462 
   2463 	solock(so);
   2464 	if ((revents = sodopoll(so, events)) == 0) {
   2465 		if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) {
   2466 			selrecord(curlwp, &so->so_rcv.sb_sel);
   2467 			so->so_rcv.sb_flags |= SB_NOTIFY;
   2468 		}
   2469 
   2470 		if (events & (POLLOUT | POLLWRNORM)) {
   2471 			selrecord(curlwp, &so->so_snd.sb_sel);
   2472 			so->so_snd.sb_flags |= SB_NOTIFY;
   2473 		}
   2474 	}
   2475 	sounlock(so);
   2476 
   2477 	return revents;
   2478 }
   2479 
   2480 struct mbuf **
   2481 sbsavetimestamp(int opt, struct mbuf **mp)
   2482 {
   2483 	struct timeval tv;
   2484 	int error;
   2485 
   2486 	memset(&tv, 0, sizeof(tv));
   2487 	microtime(&tv);
   2488 
   2489 	MODULE_HOOK_CALL(uipc_socket_50_sbts_hook, (opt, &mp), enosys(), error);
   2490 	if (error == 0)
   2491 		return mp;
   2492 
   2493 	if (opt & SO_TIMESTAMP) {
   2494 		*mp = sbcreatecontrol(&tv, sizeof(tv),
   2495 		    SCM_TIMESTAMP, SOL_SOCKET);
   2496 		if (*mp)
   2497 			mp = &(*mp)->m_next;
   2498 	}
   2499 	return mp;
   2500 }
   2501 
   2502 
   2503 #include <sys/sysctl.h>
   2504 
   2505 static int sysctl_kern_somaxkva(SYSCTLFN_PROTO);
   2506 static int sysctl_kern_sbmax(SYSCTLFN_PROTO);
   2507 
   2508 /*
   2509  * sysctl helper routine for kern.somaxkva.  ensures that the given
   2510  * value is not too small.
   2511  * (XXX should we maybe make sure it's not too large as well?)
   2512  */
   2513 static int
   2514 sysctl_kern_somaxkva(SYSCTLFN_ARGS)
   2515 {
   2516 	int error, new_somaxkva;
   2517 	struct sysctlnode node;
   2518 
   2519 	new_somaxkva = somaxkva;
   2520 	node = *rnode;
   2521 	node.sysctl_data = &new_somaxkva;
   2522 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
   2523 	if (error || newp == NULL)
   2524 		return error;
   2525 
   2526 	if (new_somaxkva < (16 * 1024 * 1024)) /* sanity */
   2527 		return SET_ERROR(EINVAL);
   2528 
   2529 	mutex_enter(&so_pendfree_lock);
   2530 	somaxkva = new_somaxkva;
   2531 	cv_broadcast(&socurkva_cv);
   2532 	mutex_exit(&so_pendfree_lock);
   2533 
   2534 	return error;
   2535 }
   2536 
   2537 /*
   2538  * sysctl helper routine for kern.sbmax. Basically just ensures that
   2539  * any new value is not too small.
   2540  */
   2541 static int
   2542 sysctl_kern_sbmax(SYSCTLFN_ARGS)
   2543 {
   2544 	int error, new_sbmax;
   2545 	struct sysctlnode node;
   2546 
   2547 	new_sbmax = sb_max;
   2548 	node = *rnode;
   2549 	node.sysctl_data = &new_sbmax;
   2550 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
   2551 	if (error || newp == NULL)
   2552 		return error;
   2553 
   2554 	KERNEL_LOCK(1, NULL);
   2555 	error = sb_max_set(new_sbmax);
   2556 	KERNEL_UNLOCK_ONE(NULL);
   2557 
   2558 	return error;
   2559 }
   2560 
   2561 /*
   2562  * sysctl helper routine for kern.sooptions. Ensures that only allowed
   2563  * options can be set.
   2564  */
   2565 static int
   2566 sysctl_kern_sooptions(SYSCTLFN_ARGS)
   2567 {
   2568 	int error, new_options;
   2569 	struct sysctlnode node;
   2570 
   2571 	new_options = sooptions;
   2572 	node = *rnode;
   2573 	node.sysctl_data = &new_options;
   2574 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
   2575 	if (error || newp == NULL)
   2576 		return error;
   2577 
   2578 	if (new_options & ~SO_DEFOPTS)
   2579 		return SET_ERROR(EINVAL);
   2580 
   2581 	sooptions = new_options;
   2582 
   2583 	return 0;
   2584 }
   2585 
   2586 static void
   2587 sysctl_kern_socket_setup(void)
   2588 {
   2589 
   2590 	KASSERT(socket_sysctllog == NULL);
   2591 
   2592 	sysctl_createv(&socket_sysctllog, 0, NULL, NULL,
   2593 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
   2594 		       CTLTYPE_INT, "somaxkva",
   2595 		       SYSCTL_DESCR("Maximum amount of kernel memory to be "
   2596 		                    "used for socket buffers"),
   2597 		       sysctl_kern_somaxkva, 0, NULL, 0,
   2598 		       CTL_KERN, KERN_SOMAXKVA, CTL_EOL);
   2599 
   2600 	sysctl_createv(&socket_sysctllog, 0, NULL, NULL,
   2601 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
   2602 		       CTLTYPE_BOOL, "sofixedbuf",
   2603 		       SYSCTL_DESCR("Prevent scaling of fixed socket buffers"),
   2604 		       NULL, 0, &sofixedbuf, 0,
   2605 		       CTL_KERN, KERN_SOFIXEDBUF, CTL_EOL);
   2606 
   2607 	sysctl_createv(&socket_sysctllog, 0, NULL, NULL,
   2608 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
   2609 		       CTLTYPE_INT, "sbmax",
   2610 		       SYSCTL_DESCR("Maximum socket buffer size"),
   2611 		       sysctl_kern_sbmax, 0, NULL, 0,
   2612 		       CTL_KERN, KERN_SBMAX, CTL_EOL);
   2613 
   2614 	sysctl_createv(&socket_sysctllog, 0, NULL, NULL,
   2615 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
   2616 		       CTLTYPE_INT, "sooptions",
   2617 		       SYSCTL_DESCR("Default socket options"),
   2618 		       sysctl_kern_sooptions, 0, NULL, 0,
   2619 		       CTL_KERN, CTL_CREATE, CTL_EOL);
   2620 }
   2621