xref: /netbsd-src/sys/kern/uipc_socket.c (revision 326b2259b73e878289ebd80cd9d20bc5aee35e99)
1 /*	$NetBSD: uipc_socket.c,v 1.84 2003/07/02 20:07:45 ragge Exp $	*/
2 
3 /*-
4  * Copyright (c) 2002 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.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the NetBSD
21  *	Foundation, Inc. and its contributors.
22  * 4. Neither the name of The NetBSD Foundation nor the names of its
23  *    contributors may be used to endorse or promote products derived
24  *    from this software without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
27  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
28  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
29  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
30  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36  * POSSIBILITY OF SUCH DAMAGE.
37  */
38 
39 /*
40  * Copyright (c) 1982, 1986, 1988, 1990, 1993
41  *	The Regents of the University of California.  All rights reserved.
42  *
43  * Redistribution and use in source and binary forms, with or without
44  * modification, are permitted provided that the following conditions
45  * are met:
46  * 1. Redistributions of source code must retain the above copyright
47  *    notice, this list of conditions and the following disclaimer.
48  * 2. Redistributions in binary form must reproduce the above copyright
49  *    notice, this list of conditions and the following disclaimer in the
50  *    documentation and/or other materials provided with the distribution.
51  * 3. All advertising materials mentioning features or use of this software
52  *    must display the following acknowledgement:
53  *	This product includes software developed by the University of
54  *	California, Berkeley and its contributors.
55  * 4. Neither the name of the University nor the names of its contributors
56  *    may be used to endorse or promote products derived from this software
57  *    without specific prior written permission.
58  *
59  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
60  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
61  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
62  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
63  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
64  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
65  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
66  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
67  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
68  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
69  * SUCH DAMAGE.
70  *
71  *	@(#)uipc_socket.c	8.6 (Berkeley) 5/2/95
72  */
73 
74 #include <sys/cdefs.h>
75 __KERNEL_RCSID(0, "$NetBSD: uipc_socket.c,v 1.84 2003/07/02 20:07:45 ragge Exp $");
76 
77 #include "opt_sock_counters.h"
78 #include "opt_sosend_loan.h"
79 #include "opt_mbuftrace.h"
80 #include "opt_somaxkva.h"
81 
82 #include <sys/param.h>
83 #include <sys/systm.h>
84 #include <sys/proc.h>
85 #include <sys/file.h>
86 #include <sys/malloc.h>
87 #include <sys/mbuf.h>
88 #include <sys/domain.h>
89 #include <sys/kernel.h>
90 #include <sys/protosw.h>
91 #include <sys/socket.h>
92 #include <sys/socketvar.h>
93 #include <sys/signalvar.h>
94 #include <sys/resourcevar.h>
95 #include <sys/pool.h>
96 #include <sys/event.h>
97 
98 #include <uvm/uvm.h>
99 
100 struct pool	socket_pool;
101 
102 MALLOC_DEFINE(M_SOOPTS, "soopts", "socket options");
103 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
104 
105 extern int	somaxconn;			/* patchable (XXX sysctl) */
106 int		somaxconn = SOMAXCONN;
107 
108 #ifdef SOSEND_COUNTERS
109 #include <sys/device.h>
110 
111 struct evcnt sosend_loan_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
112     NULL, "sosend", "loan big");
113 struct evcnt sosend_copy_big = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
114     NULL, "sosend", "copy big");
115 struct evcnt sosend_copy_small = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
116     NULL, "sosend", "copy small");
117 struct evcnt sosend_kvalimit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
118     NULL, "sosend", "kva limit");
119 
120 #define	SOSEND_COUNTER_INCR(ev)		(ev)->ev_count++
121 
122 #else
123 
124 #define	SOSEND_COUNTER_INCR(ev)		/* nothing */
125 
126 #endif /* SOSEND_COUNTERS */
127 
128 void
129 soinit(void)
130 {
131 
132 	pool_init(&socket_pool, sizeof(struct socket), 0, 0, 0,
133 	    "sockpl", NULL);
134 
135 #ifdef SOSEND_COUNTERS
136 	evcnt_attach_static(&sosend_loan_big);
137 	evcnt_attach_static(&sosend_copy_big);
138 	evcnt_attach_static(&sosend_copy_small);
139 	evcnt_attach_static(&sosend_kvalimit);
140 #endif /* SOSEND_COUNTERS */
141 }
142 
143 #ifdef SOSEND_NO_LOAN
144 int use_sosend_loan = 0;
145 #else
146 int use_sosend_loan = 1;
147 #endif
148 
149 struct mbuf *so_pendfree;
150 
151 #ifndef SOMAXKVA
152 #define	SOMAXKVA (16 * 1024 * 1024)
153 #endif
154 int somaxkva = SOMAXKVA;
155 int socurkva;
156 int sokvawaiters;
157 
158 #define	SOCK_LOAN_THRESH	4096
159 #define	SOCK_LOAN_CHUNK		65536
160 
161 static size_t sodopendfree(struct socket *);
162 
163 vaddr_t
164 sokvaalloc(vsize_t len, struct socket *so)
165 {
166 	vaddr_t lva;
167 	int s;
168 
169 	while (socurkva + len > somaxkva) {
170 		if (sodopendfree(so))
171 			continue;
172 		SOSEND_COUNTER_INCR(&sosend_kvalimit);
173 		s = splvm();
174 		sokvawaiters++;
175 		(void) tsleep(&socurkva, PVM, "sokva", 0);
176 		sokvawaiters--;
177 		splx(s);
178 	}
179 
180 	lva = uvm_km_valloc_wait(kernel_map, len);
181 	if (lva == 0)
182 		return (0);
183 	socurkva += len;
184 
185 	return lva;
186 }
187 
188 void
189 sokvafree(vaddr_t sva, vsize_t len)
190 {
191 
192 	uvm_km_free(kernel_map, sva, len);
193 	socurkva -= len;
194 	if (sokvawaiters)
195 		wakeup(&socurkva);
196 }
197 
198 static void
199 sodoloanfree(struct vm_page **pgs, caddr_t buf, size_t size)
200 {
201 	vaddr_t va, sva, eva;
202 	vsize_t len;
203 	paddr_t pa;
204 	int i, npgs;
205 
206 	eva = round_page((vaddr_t) buf + size);
207 	sva = trunc_page((vaddr_t) buf);
208 	len = eva - sva;
209 	npgs = len >> PAGE_SHIFT;
210 
211 	if (__predict_false(pgs == NULL)) {
212 		pgs = alloca(npgs * sizeof(*pgs));
213 
214 		for (i = 0, va = sva; va < eva; i++, va += PAGE_SIZE) {
215 			if (pmap_extract(pmap_kernel(), va, &pa) == FALSE)
216 				panic("sodoloanfree: va 0x%lx not mapped", va);
217 			pgs[i] = PHYS_TO_VM_PAGE(pa);
218 		}
219 	}
220 
221 	pmap_kremove(sva, len);
222 	pmap_update(pmap_kernel());
223 	uvm_unloan(pgs, npgs, UVM_LOAN_TOPAGE);
224 	sokvafree(sva, len);
225 }
226 
227 static size_t
228 sodopendfree(struct socket *so)
229 {
230 	struct mbuf *m;
231 	size_t rv = 0;
232 	int s;
233 
234 	s = splvm();
235 
236 	for (;;) {
237 		m = so_pendfree;
238 		if (m == NULL)
239 			break;
240 		so_pendfree = m->m_next;
241 		splx(s);
242 
243 		rv += m->m_ext.ext_size;
244 		sodoloanfree((m->m_flags & M_EXT_PAGES) ?
245 		    m->m_ext.ext_pgs : NULL, m->m_ext.ext_buf,
246 		    m->m_ext.ext_size);
247 		s = splvm();
248 		pool_cache_put(&mbpool_cache, m);
249 	}
250 
251 	for (;;) {
252 		m = so->so_pendfree;
253 		if (m == NULL)
254 			break;
255 		so->so_pendfree = m->m_next;
256 		splx(s);
257 
258 		rv += m->m_ext.ext_size;
259 		sodoloanfree((m->m_flags & M_EXT_PAGES) ?
260 		    m->m_ext.ext_pgs : NULL, m->m_ext.ext_buf,
261 		    m->m_ext.ext_size);
262 		s = splvm();
263 		pool_cache_put(&mbpool_cache, m);
264 	}
265 
266 	splx(s);
267 	return (rv);
268 }
269 
270 void
271 soloanfree(struct mbuf *m, caddr_t buf, size_t size, void *arg)
272 {
273 	struct socket *so = arg;
274 	int s;
275 
276 	if (m == NULL) {
277 		sodoloanfree(NULL, buf, size);
278 		return;
279 	}
280 
281 	s = splvm();
282 	m->m_next = so->so_pendfree;
283 	so->so_pendfree = m;
284 	splx(s);
285 	if (sokvawaiters)
286 		wakeup(&socurkva);
287 }
288 
289 static long
290 sosend_loan(struct socket *so, struct uio *uio, struct mbuf *m, long space)
291 {
292 	struct iovec *iov = uio->uio_iov;
293 	vaddr_t sva, eva;
294 	vsize_t len;
295 	vaddr_t lva, va;
296 	int npgs, i, error;
297 
298 	if (uio->uio_segflg != UIO_USERSPACE)
299 		return (0);
300 
301 	if (iov->iov_len < (size_t) space)
302 		space = iov->iov_len;
303 	if (space > SOCK_LOAN_CHUNK)
304 		space = SOCK_LOAN_CHUNK;
305 
306 	eva = round_page((vaddr_t) iov->iov_base + space);
307 	sva = trunc_page((vaddr_t) iov->iov_base);
308 	len = eva - sva;
309 	npgs = len >> PAGE_SHIFT;
310 
311 	/* XXX KDASSERT */
312 	KASSERT(npgs <= M_EXT_MAXPAGES);
313 
314 	lva = sokvaalloc(len, so);
315 	if (lva == 0)
316 		return 0;
317 
318 	error = uvm_loan(&uio->uio_procp->p_vmspace->vm_map, sva, len,
319 	    m->m_ext.ext_pgs, UVM_LOAN_TOPAGE);
320 	if (error) {
321 		sokvafree(lva, len);
322 		return (0);
323 	}
324 
325 	for (i = 0, va = lva; i < npgs; i++, va += PAGE_SIZE)
326 		pmap_kenter_pa(va, VM_PAGE_TO_PHYS(m->m_ext.ext_pgs[i]),
327 		    VM_PROT_READ);
328 	pmap_update(pmap_kernel());
329 
330 	lva += (vaddr_t) iov->iov_base & PAGE_MASK;
331 
332 	MEXTADD(m, (caddr_t) lva, space, M_MBUF, soloanfree, so);
333 	m->m_flags |= M_EXT_PAGES | M_EXT_ROMAP;
334 
335 	uio->uio_resid -= space;
336 	/* uio_offset not updated, not set/used for write(2) */
337 	uio->uio_iov->iov_base = (caddr_t) uio->uio_iov->iov_base + space;
338 	uio->uio_iov->iov_len -= space;
339 	if (uio->uio_iov->iov_len == 0) {
340 		uio->uio_iov++;
341 		uio->uio_iovcnt--;
342 	}
343 
344 	return (space);
345 }
346 
347 /*
348  * Socket operation routines.
349  * These routines are called by the routines in
350  * sys_socket.c or from a system process, and
351  * implement the semantics of socket operations by
352  * switching out to the protocol specific routines.
353  */
354 /*ARGSUSED*/
355 int
356 socreate(int dom, struct socket **aso, int type, int proto)
357 {
358 	struct proc	*p;
359 	struct protosw	*prp;
360 	struct socket	*so;
361 	int		error, s;
362 
363 	p = curproc;		/* XXX */
364 	if (proto)
365 		prp = pffindproto(dom, proto, type);
366 	else
367 		prp = pffindtype(dom, type);
368 	if (prp == 0 || prp->pr_usrreq == 0)
369 		return (EPROTONOSUPPORT);
370 	if (prp->pr_type != type)
371 		return (EPROTOTYPE);
372 	s = splsoftnet();
373 	so = pool_get(&socket_pool, PR_WAITOK);
374 	memset((caddr_t)so, 0, sizeof(*so));
375 	TAILQ_INIT(&so->so_q0);
376 	TAILQ_INIT(&so->so_q);
377 	so->so_type = type;
378 	so->so_proto = prp;
379 	so->so_send = sosend;
380 	so->so_receive = soreceive;
381 #ifdef MBUFTRACE
382 	so->so_rcv.sb_mowner = &prp->pr_domain->dom_mowner;
383 	so->so_snd.sb_mowner = &prp->pr_domain->dom_mowner;
384 	so->so_mowner = &prp->pr_domain->dom_mowner;
385 #endif
386 	if (p != 0)
387 		so->so_uid = p->p_ucred->cr_uid;
388 	error = (*prp->pr_usrreq)(so, PRU_ATTACH, (struct mbuf *)0,
389 	    (struct mbuf *)(long)proto, (struct mbuf *)0, p);
390 	if (error) {
391 		so->so_state |= SS_NOFDREF;
392 		sofree(so);
393 		splx(s);
394 		return (error);
395 	}
396 	splx(s);
397 	*aso = so;
398 	return (0);
399 }
400 
401 int
402 sobind(struct socket *so, struct mbuf *nam, struct proc *p)
403 {
404 	int	s, error;
405 
406 	s = splsoftnet();
407 	error = (*so->so_proto->pr_usrreq)(so, PRU_BIND, (struct mbuf *)0,
408 	    nam, (struct mbuf *)0, p);
409 	splx(s);
410 	return (error);
411 }
412 
413 int
414 solisten(struct socket *so, int backlog)
415 {
416 	int	s, error;
417 
418 	s = splsoftnet();
419 	error = (*so->so_proto->pr_usrreq)(so, PRU_LISTEN, (struct mbuf *)0,
420 	    (struct mbuf *)0, (struct mbuf *)0, (struct proc *)0);
421 	if (error) {
422 		splx(s);
423 		return (error);
424 	}
425 	if (TAILQ_EMPTY(&so->so_q))
426 		so->so_options |= SO_ACCEPTCONN;
427 	if (backlog < 0)
428 		backlog = 0;
429 	so->so_qlimit = min(backlog, somaxconn);
430 	splx(s);
431 	return (0);
432 }
433 
434 void
435 sofree(struct socket *so)
436 {
437 	struct mbuf *m;
438 
439 	if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0)
440 		return;
441 	if (so->so_head) {
442 		/*
443 		 * We must not decommission a socket that's on the accept(2)
444 		 * queue.  If we do, then accept(2) may hang after select(2)
445 		 * indicated that the listening socket was ready.
446 		 */
447 		if (!soqremque(so, 0))
448 			return;
449 	}
450 	sbrelease(&so->so_snd);
451 	sorflush(so);
452 	while ((m = so->so_pendfree) != NULL) {
453 		so->so_pendfree = m->m_next;
454 		m->m_next = so_pendfree;
455 		so_pendfree = m;
456 	}
457 	pool_put(&socket_pool, so);
458 }
459 
460 /*
461  * Close a socket on last file table reference removal.
462  * Initiate disconnect if connected.
463  * Free socket when disconnect complete.
464  */
465 int
466 soclose(struct socket *so)
467 {
468 	struct socket	*so2;
469 	int		s, error;
470 
471 	error = 0;
472 	s = splsoftnet();		/* conservative */
473 	if (so->so_options & SO_ACCEPTCONN) {
474 		while ((so2 = TAILQ_FIRST(&so->so_q0)) != 0) {
475 			(void) soqremque(so2, 0);
476 			(void) soabort(so2);
477 		}
478 		while ((so2 = TAILQ_FIRST(&so->so_q)) != 0) {
479 			(void) soqremque(so2, 1);
480 			(void) soabort(so2);
481 		}
482 	}
483 	if (so->so_pcb == 0)
484 		goto discard;
485 	if (so->so_state & SS_ISCONNECTED) {
486 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
487 			error = sodisconnect(so);
488 			if (error)
489 				goto drop;
490 		}
491 		if (so->so_options & SO_LINGER) {
492 			if ((so->so_state & SS_ISDISCONNECTING) &&
493 			    (so->so_state & SS_NBIO))
494 				goto drop;
495 			while (so->so_state & SS_ISCONNECTED) {
496 				error = tsleep((caddr_t)&so->so_timeo,
497 					       PSOCK | PCATCH, netcls,
498 					       so->so_linger * hz);
499 				if (error)
500 					break;
501 			}
502 		}
503 	}
504  drop:
505 	if (so->so_pcb) {
506 		int error2 = (*so->so_proto->pr_usrreq)(so, PRU_DETACH,
507 		    (struct mbuf *)0, (struct mbuf *)0, (struct mbuf *)0,
508 		    (struct proc *)0);
509 		if (error == 0)
510 			error = error2;
511 	}
512  discard:
513 	if (so->so_state & SS_NOFDREF)
514 		panic("soclose: NOFDREF");
515 	so->so_state |= SS_NOFDREF;
516 	sofree(so);
517 	splx(s);
518 	return (error);
519 }
520 
521 /*
522  * Must be called at splsoftnet...
523  */
524 int
525 soabort(struct socket *so)
526 {
527 
528 	return (*so->so_proto->pr_usrreq)(so, PRU_ABORT, (struct mbuf *)0,
529 	    (struct mbuf *)0, (struct mbuf *)0, (struct proc *)0);
530 }
531 
532 int
533 soaccept(struct socket *so, struct mbuf *nam)
534 {
535 	int	s, error;
536 
537 	error = 0;
538 	s = splsoftnet();
539 	if ((so->so_state & SS_NOFDREF) == 0)
540 		panic("soaccept: !NOFDREF");
541 	so->so_state &= ~SS_NOFDREF;
542 	if ((so->so_state & SS_ISDISCONNECTED) == 0 ||
543 	    (so->so_proto->pr_flags & PR_ABRTACPTDIS) == 0)
544 		error = (*so->so_proto->pr_usrreq)(so, PRU_ACCEPT,
545 		    (struct mbuf *)0, nam, (struct mbuf *)0, (struct proc *)0);
546 	else
547 		error = ECONNABORTED;
548 
549 	splx(s);
550 	return (error);
551 }
552 
553 int
554 soconnect(struct socket *so, struct mbuf *nam)
555 {
556 	struct proc	*p;
557 	int		s, error;
558 
559 	p = curproc;		/* XXX */
560 	if (so->so_options & SO_ACCEPTCONN)
561 		return (EOPNOTSUPP);
562 	s = splsoftnet();
563 	/*
564 	 * If protocol is connection-based, can only connect once.
565 	 * Otherwise, if connected, try to disconnect first.
566 	 * This allows user to disconnect by connecting to, e.g.,
567 	 * a null address.
568 	 */
569 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
570 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
571 	    (error = sodisconnect(so))))
572 		error = EISCONN;
573 	else
574 		error = (*so->so_proto->pr_usrreq)(so, PRU_CONNECT,
575 		    (struct mbuf *)0, nam, (struct mbuf *)0, p);
576 	splx(s);
577 	return (error);
578 }
579 
580 int
581 soconnect2(struct socket *so1, struct socket *so2)
582 {
583 	int	s, error;
584 
585 	s = splsoftnet();
586 	error = (*so1->so_proto->pr_usrreq)(so1, PRU_CONNECT2,
587 	    (struct mbuf *)0, (struct mbuf *)so2, (struct mbuf *)0,
588 	    (struct proc *)0);
589 	splx(s);
590 	return (error);
591 }
592 
593 int
594 sodisconnect(struct socket *so)
595 {
596 	int	s, error;
597 
598 	s = splsoftnet();
599 	if ((so->so_state & SS_ISCONNECTED) == 0) {
600 		error = ENOTCONN;
601 		goto bad;
602 	}
603 	if (so->so_state & SS_ISDISCONNECTING) {
604 		error = EALREADY;
605 		goto bad;
606 	}
607 	error = (*so->so_proto->pr_usrreq)(so, PRU_DISCONNECT,
608 	    (struct mbuf *)0, (struct mbuf *)0, (struct mbuf *)0,
609 	    (struct proc *)0);
610  bad:
611 	splx(s);
612 	sodopendfree(so);
613 	return (error);
614 }
615 
616 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
617 /*
618  * Send on a socket.
619  * If send must go all at once and message is larger than
620  * send buffering, then hard error.
621  * Lock against other senders.
622  * If must go all at once and not enough room now, then
623  * inform user that this would block and do nothing.
624  * Otherwise, if nonblocking, send as much as possible.
625  * The data to be sent is described by "uio" if nonzero,
626  * otherwise by the mbuf chain "top" (which must be null
627  * if uio is not).  Data provided in mbuf chain must be small
628  * enough to send all at once.
629  *
630  * Returns nonzero on error, timeout or signal; callers
631  * must check for short counts if EINTR/ERESTART are returned.
632  * Data and control buffers are freed on return.
633  */
634 int
635 sosend(struct socket *so, struct mbuf *addr, struct uio *uio, struct mbuf *top,
636 	struct mbuf *control, int flags)
637 {
638 	struct proc	*p;
639 	struct mbuf	**mp, *m;
640 	long		space, len, resid, clen, mlen;
641 	int		error, s, dontroute, atomic;
642 
643 	sodopendfree(so);
644 
645 	p = curproc;		/* XXX */
646 	clen = 0;
647 	atomic = sosendallatonce(so) || top;
648 	if (uio)
649 		resid = uio->uio_resid;
650 	else
651 		resid = top->m_pkthdr.len;
652 	/*
653 	 * In theory resid should be unsigned.
654 	 * However, space must be signed, as it might be less than 0
655 	 * if we over-committed, and we must use a signed comparison
656 	 * of space and resid.  On the other hand, a negative resid
657 	 * causes us to loop sending 0-length segments to the protocol.
658 	 */
659 	if (resid < 0) {
660 		error = EINVAL;
661 		goto out;
662 	}
663 	dontroute =
664 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
665 	    (so->so_proto->pr_flags & PR_ATOMIC);
666 	p->p_stats->p_ru.ru_msgsnd++;
667 	if (control)
668 		clen = control->m_len;
669 #define	snderr(errno)	{ error = errno; splx(s); goto release; }
670 
671  restart:
672 	if ((error = sblock(&so->so_snd, SBLOCKWAIT(flags))) != 0)
673 		goto out;
674 	do {
675 		s = splsoftnet();
676 		if (so->so_state & SS_CANTSENDMORE)
677 			snderr(EPIPE);
678 		if (so->so_error) {
679 			error = so->so_error;
680 			so->so_error = 0;
681 			splx(s);
682 			goto release;
683 		}
684 		if ((so->so_state & SS_ISCONNECTED) == 0) {
685 			if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
686 				if ((so->so_state & SS_ISCONFIRMING) == 0 &&
687 				    !(resid == 0 && clen != 0))
688 					snderr(ENOTCONN);
689 			} else if (addr == 0)
690 				snderr(EDESTADDRREQ);
691 		}
692 		space = sbspace(&so->so_snd);
693 		if (flags & MSG_OOB)
694 			space += 1024;
695 		if ((atomic && resid > so->so_snd.sb_hiwat) ||
696 		    clen > so->so_snd.sb_hiwat)
697 			snderr(EMSGSIZE);
698 		if (space < resid + clen && uio &&
699 		    (atomic || space < so->so_snd.sb_lowat || space < clen)) {
700 			if (so->so_state & SS_NBIO)
701 				snderr(EWOULDBLOCK);
702 			sbunlock(&so->so_snd);
703 			error = sbwait(&so->so_snd);
704 			splx(s);
705 			if (error)
706 				goto out;
707 			goto restart;
708 		}
709 		splx(s);
710 		mp = &top;
711 		space -= clen;
712 		do {
713 			if (uio == NULL) {
714 				/*
715 				 * Data is prepackaged in "top".
716 				 */
717 				resid = 0;
718 				if (flags & MSG_EOR)
719 					top->m_flags |= M_EOR;
720 			} else do {
721 				if (top == 0) {
722 					m = m_gethdr(M_WAIT, MT_DATA);
723 					mlen = MHLEN;
724 					m->m_pkthdr.len = 0;
725 					m->m_pkthdr.rcvif = (struct ifnet *)0;
726 				} else {
727 					m = m_get(M_WAIT, MT_DATA);
728 					mlen = MLEN;
729 				}
730 				MCLAIM(m, so->so_snd.sb_mowner);
731 				if (use_sosend_loan &&
732 				    uio->uio_iov->iov_len >= SOCK_LOAN_THRESH &&
733 				    space >= SOCK_LOAN_THRESH &&
734 				    (len = sosend_loan(so, uio, m,
735 						       space)) != 0) {
736 					SOSEND_COUNTER_INCR(&sosend_loan_big);
737 					space -= len;
738 					goto have_data;
739 				}
740 				if (resid >= MINCLSIZE && space >= MCLBYTES) {
741 					SOSEND_COUNTER_INCR(&sosend_copy_big);
742 					m_clget(m, M_WAIT);
743 					if ((m->m_flags & M_EXT) == 0)
744 						goto nopages;
745 					mlen = MCLBYTES;
746 					if (atomic && top == 0) {
747 						len = lmin(MCLBYTES - max_hdr,
748 						    resid);
749 						m->m_data += max_hdr;
750 					} else
751 						len = lmin(MCLBYTES, resid);
752 					space -= len;
753 				} else {
754  nopages:
755 					SOSEND_COUNTER_INCR(&sosend_copy_small);
756 					len = lmin(lmin(mlen, resid), space);
757 					space -= len;
758 					/*
759 					 * For datagram protocols, leave room
760 					 * for protocol headers in first mbuf.
761 					 */
762 					if (atomic && top == 0 && len < mlen)
763 						MH_ALIGN(m, len);
764 				}
765 				error = uiomove(mtod(m, caddr_t), (int)len,
766 				    uio);
767  have_data:
768 				resid = uio->uio_resid;
769 				m->m_len = len;
770 				*mp = m;
771 				top->m_pkthdr.len += len;
772 				if (error)
773 					goto release;
774 				mp = &m->m_next;
775 				if (resid <= 0) {
776 					if (flags & MSG_EOR)
777 						top->m_flags |= M_EOR;
778 					break;
779 				}
780 			} while (space > 0 && atomic);
781 
782 			s = splsoftnet();
783 
784 			if (so->so_state & SS_CANTSENDMORE)
785 				snderr(EPIPE);
786 
787 			if (dontroute)
788 				so->so_options |= SO_DONTROUTE;
789 			if (resid > 0)
790 				so->so_state |= SS_MORETOCOME;
791 			error = (*so->so_proto->pr_usrreq)(so,
792 			    (flags & MSG_OOB) ? PRU_SENDOOB : PRU_SEND,
793 			    top, addr, control, p);
794 			if (dontroute)
795 				so->so_options &= ~SO_DONTROUTE;
796 			if (resid > 0)
797 				so->so_state &= ~SS_MORETOCOME;
798 			splx(s);
799 
800 			clen = 0;
801 			control = 0;
802 			top = 0;
803 			mp = &top;
804 			if (error)
805 				goto release;
806 		} while (resid && space > 0);
807 	} while (resid);
808 
809  release:
810 	sbunlock(&so->so_snd);
811  out:
812 	if (top)
813 		m_freem(top);
814 	if (control)
815 		m_freem(control);
816 	return (error);
817 }
818 
819 /*
820  * Implement receive operations on a socket.
821  * We depend on the way that records are added to the sockbuf
822  * by sbappend*.  In particular, each record (mbufs linked through m_next)
823  * must begin with an address if the protocol so specifies,
824  * followed by an optional mbuf or mbufs containing ancillary data,
825  * and then zero or more mbufs of data.
826  * In order to avoid blocking network interrupts for the entire time here,
827  * we splx() while doing the actual copy to user space.
828  * Although the sockbuf is locked, new data may still be appended,
829  * and thus we must maintain consistency of the sockbuf during that time.
830  *
831  * The caller may receive the data as a single mbuf chain by supplying
832  * an mbuf **mp0 for use in returning the chain.  The uio is then used
833  * only for the count in uio_resid.
834  */
835 int
836 soreceive(struct socket *so, struct mbuf **paddr, struct uio *uio,
837 	struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
838 {
839 	struct mbuf	*m, **mp;
840 	int		flags, len, error, s, offset, moff, type, orig_resid;
841 	struct protosw	*pr;
842 	struct mbuf	*nextrecord;
843 	int		mbuf_removed = 0;
844 
845 	pr = so->so_proto;
846 	mp = mp0;
847 	type = 0;
848 	orig_resid = uio->uio_resid;
849 	if (paddr)
850 		*paddr = 0;
851 	if (controlp)
852 		*controlp = 0;
853 	if (flagsp)
854 		flags = *flagsp &~ MSG_EOR;
855 	else
856 		flags = 0;
857 
858 	if ((flags & MSG_DONTWAIT) == 0)
859 		sodopendfree(so);
860 
861 	if (flags & MSG_OOB) {
862 		m = m_get(M_WAIT, MT_DATA);
863 		error = (*pr->pr_usrreq)(so, PRU_RCVOOB, m,
864 		    (struct mbuf *)(long)(flags & MSG_PEEK), (struct mbuf *)0,
865 		    (struct proc *)0);
866 		if (error)
867 			goto bad;
868 		do {
869 			error = uiomove(mtod(m, caddr_t),
870 			    (int) min(uio->uio_resid, m->m_len), uio);
871 			m = m_free(m);
872 		} while (uio->uio_resid && error == 0 && m);
873  bad:
874 		if (m)
875 			m_freem(m);
876 		return (error);
877 	}
878 	if (mp)
879 		*mp = (struct mbuf *)0;
880 	if (so->so_state & SS_ISCONFIRMING && uio->uio_resid)
881 		(*pr->pr_usrreq)(so, PRU_RCVD, (struct mbuf *)0,
882 		    (struct mbuf *)0, (struct mbuf *)0, (struct proc *)0);
883 
884  restart:
885 	if ((error = sblock(&so->so_rcv, SBLOCKWAIT(flags))) != 0)
886 		return (error);
887 	s = splsoftnet();
888 
889 	m = so->so_rcv.sb_mb;
890 	/*
891 	 * If we have less data than requested, block awaiting more
892 	 * (subject to any timeout) if:
893 	 *   1. the current count is less than the low water mark,
894 	 *   2. MSG_WAITALL is set, and it is possible to do the entire
895 	 *	receive operation at once if we block (resid <= hiwat), or
896 	 *   3. MSG_DONTWAIT is not set.
897 	 * If MSG_WAITALL is set but resid is larger than the receive buffer,
898 	 * we have to do the receive in sections, and thus risk returning
899 	 * a short count if a timeout or signal occurs after we start.
900 	 */
901 	if (m == 0 || (((flags & MSG_DONTWAIT) == 0 &&
902 	    so->so_rcv.sb_cc < uio->uio_resid) &&
903 	    (so->so_rcv.sb_cc < so->so_rcv.sb_lowat ||
904 	    ((flags & MSG_WAITALL) && uio->uio_resid <= so->so_rcv.sb_hiwat)) &&
905 	    m->m_nextpkt == 0 && (pr->pr_flags & PR_ATOMIC) == 0)) {
906 #ifdef DIAGNOSTIC
907 		if (m == 0 && so->so_rcv.sb_cc)
908 			panic("receive 1");
909 #endif
910 		if (so->so_error) {
911 			if (m)
912 				goto dontblock;
913 			error = so->so_error;
914 			if ((flags & MSG_PEEK) == 0)
915 				so->so_error = 0;
916 			goto release;
917 		}
918 		if (so->so_state & SS_CANTRCVMORE) {
919 			if (m)
920 				goto dontblock;
921 			else
922 				goto release;
923 		}
924 		for (; m; m = m->m_next)
925 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
926 				m = so->so_rcv.sb_mb;
927 				goto dontblock;
928 			}
929 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
930 		    (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
931 			error = ENOTCONN;
932 			goto release;
933 		}
934 		if (uio->uio_resid == 0)
935 			goto release;
936 		if ((so->so_state & SS_NBIO) || (flags & MSG_DONTWAIT)) {
937 			error = EWOULDBLOCK;
938 			goto release;
939 		}
940 		SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 1");
941 		SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 1");
942 		sbunlock(&so->so_rcv);
943 		error = sbwait(&so->so_rcv);
944 		splx(s);
945 		if (error)
946 			return (error);
947 		goto restart;
948 	}
949  dontblock:
950 	/*
951 	 * On entry here, m points to the first record of the socket buffer.
952 	 * While we process the initial mbufs containing address and control
953 	 * info, we save a copy of m->m_nextpkt into nextrecord.
954 	 */
955 #ifdef notyet /* XXXX */
956 	if (uio->uio_procp)
957 		uio->uio_procp->p_stats->p_ru.ru_msgrcv++;
958 #endif
959 	KASSERT(m == so->so_rcv.sb_mb);
960 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 1");
961 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 1");
962 	nextrecord = m->m_nextpkt;
963 	if (pr->pr_flags & PR_ADDR) {
964 #ifdef DIAGNOSTIC
965 		if (m->m_type != MT_SONAME)
966 			panic("receive 1a");
967 #endif
968 		orig_resid = 0;
969 		if (flags & MSG_PEEK) {
970 			if (paddr)
971 				*paddr = m_copy(m, 0, m->m_len);
972 			m = m->m_next;
973 		} else {
974 			sbfree(&so->so_rcv, m);
975 			mbuf_removed = 1;
976 			if (paddr) {
977 				*paddr = m;
978 				so->so_rcv.sb_mb = m->m_next;
979 				m->m_next = 0;
980 				m = so->so_rcv.sb_mb;
981 			} else {
982 				MFREE(m, so->so_rcv.sb_mb);
983 				m = so->so_rcv.sb_mb;
984 			}
985 		}
986 	}
987 	while (m && m->m_type == MT_CONTROL && error == 0) {
988 		if (flags & MSG_PEEK) {
989 			if (controlp)
990 				*controlp = m_copy(m, 0, m->m_len);
991 			m = m->m_next;
992 		} else {
993 			sbfree(&so->so_rcv, m);
994 			mbuf_removed = 1;
995 			if (controlp) {
996 				if (pr->pr_domain->dom_externalize &&
997 				    mtod(m, struct cmsghdr *)->cmsg_type ==
998 				    SCM_RIGHTS)
999 					error = (*pr->pr_domain->dom_externalize)(m);
1000 				*controlp = m;
1001 				so->so_rcv.sb_mb = m->m_next;
1002 				m->m_next = 0;
1003 				m = so->so_rcv.sb_mb;
1004 			} else {
1005 				MFREE(m, so->so_rcv.sb_mb);
1006 				m = so->so_rcv.sb_mb;
1007 			}
1008 		}
1009 		if (controlp) {
1010 			orig_resid = 0;
1011 			controlp = &(*controlp)->m_next;
1012 		}
1013 	}
1014 
1015 	/*
1016 	 * If m is non-NULL, we have some data to read.  From now on,
1017 	 * make sure to keep sb_lastrecord consistent when working on
1018 	 * the last packet on the chain (nextrecord == NULL) and we
1019 	 * change m->m_nextpkt.
1020 	 */
1021 	if (m) {
1022 		if ((flags & MSG_PEEK) == 0) {
1023 			m->m_nextpkt = nextrecord;
1024 			/*
1025 			 * If nextrecord == NULL (this is a single chain),
1026 			 * then sb_lastrecord may not be valid here if m
1027 			 * was changed earlier.
1028 			 */
1029 			if (nextrecord == NULL) {
1030 				KASSERT(so->so_rcv.sb_mb == m);
1031 				so->so_rcv.sb_lastrecord = m;
1032 			}
1033 		}
1034 		type = m->m_type;
1035 		if (type == MT_OOBDATA)
1036 			flags |= MSG_OOB;
1037 	} else {
1038 		if ((flags & MSG_PEEK) == 0) {
1039 			KASSERT(so->so_rcv.sb_mb == m);
1040 			so->so_rcv.sb_mb = nextrecord;
1041 			SB_EMPTY_FIXUP(&so->so_rcv);
1042 		}
1043 	}
1044 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 2");
1045 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 2");
1046 
1047 	moff = 0;
1048 	offset = 0;
1049 	while (m && uio->uio_resid > 0 && error == 0) {
1050 		if (m->m_type == MT_OOBDATA) {
1051 			if (type != MT_OOBDATA)
1052 				break;
1053 		} else if (type == MT_OOBDATA)
1054 			break;
1055 #ifdef DIAGNOSTIC
1056 		else if (m->m_type != MT_DATA && m->m_type != MT_HEADER)
1057 			panic("receive 3");
1058 #endif
1059 		so->so_state &= ~SS_RCVATMARK;
1060 		len = uio->uio_resid;
1061 		if (so->so_oobmark && len > so->so_oobmark - offset)
1062 			len = so->so_oobmark - offset;
1063 		if (len > m->m_len - moff)
1064 			len = m->m_len - moff;
1065 		/*
1066 		 * If mp is set, just pass back the mbufs.
1067 		 * Otherwise copy them out via the uio, then free.
1068 		 * Sockbuf must be consistent here (points to current mbuf,
1069 		 * it points to next record) when we drop priority;
1070 		 * we must note any additions to the sockbuf when we
1071 		 * block interrupts again.
1072 		 */
1073 		if (mp == 0) {
1074 			SBLASTRECORDCHK(&so->so_rcv, "soreceive uiomove");
1075 			SBLASTMBUFCHK(&so->so_rcv, "soreceive uiomove");
1076 			splx(s);
1077 			error = uiomove(mtod(m, caddr_t) + moff, (int)len, uio);
1078 			s = splsoftnet();
1079 			if (error) {
1080 				/*
1081 				 * If any part of the record has been removed
1082 				 * (such as the MT_SONAME mbuf, which will
1083 				 * happen when PR_ADDR, and thus also
1084 				 * PR_ATOMIC, is set), then drop the entire
1085 				 * record to maintain the atomicity of the
1086 				 * receive operation.
1087 				 *
1088 				 * This avoids a later panic("receive 1a")
1089 				 * when compiled with DIAGNOSTIC.
1090 				 */
1091 				if (m && mbuf_removed
1092 				    && (pr->pr_flags & PR_ATOMIC))
1093 					(void) sbdroprecord(&so->so_rcv);
1094 
1095 				goto release;
1096 			}
1097 		} else
1098 			uio->uio_resid -= len;
1099 		if (len == m->m_len - moff) {
1100 			if (m->m_flags & M_EOR)
1101 				flags |= MSG_EOR;
1102 			if (flags & MSG_PEEK) {
1103 				m = m->m_next;
1104 				moff = 0;
1105 			} else {
1106 				nextrecord = m->m_nextpkt;
1107 				sbfree(&so->so_rcv, m);
1108 				if (mp) {
1109 					*mp = m;
1110 					mp = &m->m_next;
1111 					so->so_rcv.sb_mb = m = m->m_next;
1112 					*mp = (struct mbuf *)0;
1113 				} else {
1114 					MFREE(m, so->so_rcv.sb_mb);
1115 					m = so->so_rcv.sb_mb;
1116 				}
1117 				/*
1118 				 * If m != NULL, we also know that
1119 				 * so->so_rcv.sb_mb != NULL.
1120 				 */
1121 				KASSERT(so->so_rcv.sb_mb == m);
1122 				if (m) {
1123 					m->m_nextpkt = nextrecord;
1124 					if (nextrecord == NULL)
1125 						so->so_rcv.sb_lastrecord = m;
1126 				} else {
1127 					so->so_rcv.sb_mb = nextrecord;
1128 					SB_EMPTY_FIXUP(&so->so_rcv);
1129 				}
1130 				SBLASTRECORDCHK(&so->so_rcv, "soreceive 3");
1131 				SBLASTMBUFCHK(&so->so_rcv, "soreceive 3");
1132 			}
1133 		} else {
1134 			if (flags & MSG_PEEK)
1135 				moff += len;
1136 			else {
1137 				if (mp)
1138 					*mp = m_copym(m, 0, len, M_WAIT);
1139 				m->m_data += len;
1140 				m->m_len -= len;
1141 				so->so_rcv.sb_cc -= len;
1142 			}
1143 		}
1144 		if (so->so_oobmark) {
1145 			if ((flags & MSG_PEEK) == 0) {
1146 				so->so_oobmark -= len;
1147 				if (so->so_oobmark == 0) {
1148 					so->so_state |= SS_RCVATMARK;
1149 					break;
1150 				}
1151 			} else {
1152 				offset += len;
1153 				if (offset == so->so_oobmark)
1154 					break;
1155 			}
1156 		}
1157 		if (flags & MSG_EOR)
1158 			break;
1159 		/*
1160 		 * If the MSG_WAITALL flag is set (for non-atomic socket),
1161 		 * we must not quit until "uio->uio_resid == 0" or an error
1162 		 * termination.  If a signal/timeout occurs, return
1163 		 * with a short count but without error.
1164 		 * Keep sockbuf locked against other readers.
1165 		 */
1166 		while (flags & MSG_WAITALL && m == 0 && uio->uio_resid > 0 &&
1167 		    !sosendallatonce(so) && !nextrecord) {
1168 			if (so->so_error || so->so_state & SS_CANTRCVMORE)
1169 				break;
1170 			/*
1171 			 * If we are peeking and the socket receive buffer is
1172 			 * full, stop since we can't get more data to peek at.
1173 			 */
1174 			if ((flags & MSG_PEEK) && sbspace(&so->so_rcv) <= 0)
1175 				break;
1176 			/*
1177 			 * If we've drained the socket buffer, tell the
1178 			 * protocol in case it needs to do something to
1179 			 * get it filled again.
1180 			 */
1181 			if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb)
1182 				(*pr->pr_usrreq)(so, PRU_RCVD,
1183 				    (struct mbuf *)0,
1184 				    (struct mbuf *)(long)flags,
1185 				    (struct mbuf *)0,
1186 				    (struct proc *)0);
1187 			SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2");
1188 			SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2");
1189 			error = sbwait(&so->so_rcv);
1190 			if (error) {
1191 				sbunlock(&so->so_rcv);
1192 				splx(s);
1193 				return (0);
1194 			}
1195 			if ((m = so->so_rcv.sb_mb) != NULL)
1196 				nextrecord = m->m_nextpkt;
1197 		}
1198 	}
1199 
1200 	if (m && pr->pr_flags & PR_ATOMIC) {
1201 		flags |= MSG_TRUNC;
1202 		if ((flags & MSG_PEEK) == 0)
1203 			(void) sbdroprecord(&so->so_rcv);
1204 	}
1205 	if ((flags & MSG_PEEK) == 0) {
1206 		if (m == 0) {
1207 			/*
1208 			 * First part is an inline SB_EMPTY_FIXUP().  Second
1209 			 * part makes sure sb_lastrecord is up-to-date if
1210 			 * there is still data in the socket buffer.
1211 			 */
1212 			so->so_rcv.sb_mb = nextrecord;
1213 			if (so->so_rcv.sb_mb == NULL) {
1214 				so->so_rcv.sb_mbtail = NULL;
1215 				so->so_rcv.sb_lastrecord = NULL;
1216 			} else if (nextrecord->m_nextpkt == NULL)
1217 				so->so_rcv.sb_lastrecord = nextrecord;
1218 		}
1219 		SBLASTRECORDCHK(&so->so_rcv, "soreceive 4");
1220 		SBLASTMBUFCHK(&so->so_rcv, "soreceive 4");
1221 		if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1222 			(*pr->pr_usrreq)(so, PRU_RCVD, (struct mbuf *)0,
1223 			    (struct mbuf *)(long)flags, (struct mbuf *)0,
1224 			    (struct proc *)0);
1225 	}
1226 	if (orig_resid == uio->uio_resid && orig_resid &&
1227 	    (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1228 		sbunlock(&so->so_rcv);
1229 		splx(s);
1230 		goto restart;
1231 	}
1232 
1233 	if (flagsp)
1234 		*flagsp |= flags;
1235  release:
1236 	sbunlock(&so->so_rcv);
1237 	splx(s);
1238 	return (error);
1239 }
1240 
1241 int
1242 soshutdown(struct socket *so, int how)
1243 {
1244 	struct protosw	*pr;
1245 
1246 	pr = so->so_proto;
1247 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1248 		return (EINVAL);
1249 
1250 	if (how == SHUT_RD || how == SHUT_RDWR)
1251 		sorflush(so);
1252 	if (how == SHUT_WR || how == SHUT_RDWR)
1253 		return (*pr->pr_usrreq)(so, PRU_SHUTDOWN, (struct mbuf *)0,
1254 		    (struct mbuf *)0, (struct mbuf *)0, (struct proc *)0);
1255 	return (0);
1256 }
1257 
1258 void
1259 sorflush(struct socket *so)
1260 {
1261 	struct sockbuf	*sb, asb;
1262 	struct protosw	*pr;
1263 	int		s;
1264 
1265 	sb = &so->so_rcv;
1266 	pr = so->so_proto;
1267 	sb->sb_flags |= SB_NOINTR;
1268 	(void) sblock(sb, M_WAITOK);
1269 	s = splnet();
1270 	socantrcvmore(so);
1271 	sbunlock(sb);
1272 	asb = *sb;
1273 	memset((caddr_t)sb, 0, sizeof(*sb));
1274 	splx(s);
1275 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose)
1276 		(*pr->pr_domain->dom_dispose)(asb.sb_mb);
1277 	sbrelease(&asb);
1278 }
1279 
1280 int
1281 sosetopt(struct socket *so, int level, int optname, struct mbuf *m0)
1282 {
1283 	int		error;
1284 	struct mbuf	*m;
1285 
1286 	error = 0;
1287 	m = m0;
1288 	if (level != SOL_SOCKET) {
1289 		if (so->so_proto && so->so_proto->pr_ctloutput)
1290 			return ((*so->so_proto->pr_ctloutput)
1291 				  (PRCO_SETOPT, so, level, optname, &m0));
1292 		error = ENOPROTOOPT;
1293 	} else {
1294 		switch (optname) {
1295 
1296 		case SO_LINGER:
1297 			if (m == NULL || m->m_len != sizeof(struct linger)) {
1298 				error = EINVAL;
1299 				goto bad;
1300 			}
1301 			so->so_linger = mtod(m, struct linger *)->l_linger;
1302 			/* fall thru... */
1303 
1304 		case SO_DEBUG:
1305 		case SO_KEEPALIVE:
1306 		case SO_DONTROUTE:
1307 		case SO_USELOOPBACK:
1308 		case SO_BROADCAST:
1309 		case SO_REUSEADDR:
1310 		case SO_REUSEPORT:
1311 		case SO_OOBINLINE:
1312 		case SO_TIMESTAMP:
1313 			if (m == NULL || m->m_len < sizeof(int)) {
1314 				error = EINVAL;
1315 				goto bad;
1316 			}
1317 			if (*mtod(m, int *))
1318 				so->so_options |= optname;
1319 			else
1320 				so->so_options &= ~optname;
1321 			break;
1322 
1323 		case SO_SNDBUF:
1324 		case SO_RCVBUF:
1325 		case SO_SNDLOWAT:
1326 		case SO_RCVLOWAT:
1327 		    {
1328 			int optval;
1329 
1330 			if (m == NULL || m->m_len < sizeof(int)) {
1331 				error = EINVAL;
1332 				goto bad;
1333 			}
1334 
1335 			/*
1336 			 * Values < 1 make no sense for any of these
1337 			 * options, so disallow them.
1338 			 */
1339 			optval = *mtod(m, int *);
1340 			if (optval < 1) {
1341 				error = EINVAL;
1342 				goto bad;
1343 			}
1344 
1345 			switch (optname) {
1346 
1347 			case SO_SNDBUF:
1348 			case SO_RCVBUF:
1349 				if (sbreserve(optname == SO_SNDBUF ?
1350 				    &so->so_snd : &so->so_rcv,
1351 				    (u_long) optval) == 0) {
1352 					error = ENOBUFS;
1353 					goto bad;
1354 				}
1355 				break;
1356 
1357 			/*
1358 			 * Make sure the low-water is never greater than
1359 			 * the high-water.
1360 			 */
1361 			case SO_SNDLOWAT:
1362 				so->so_snd.sb_lowat =
1363 				    (optval > so->so_snd.sb_hiwat) ?
1364 				    so->so_snd.sb_hiwat : optval;
1365 				break;
1366 			case SO_RCVLOWAT:
1367 				so->so_rcv.sb_lowat =
1368 				    (optval > so->so_rcv.sb_hiwat) ?
1369 				    so->so_rcv.sb_hiwat : optval;
1370 				break;
1371 			}
1372 			break;
1373 		    }
1374 
1375 		case SO_SNDTIMEO:
1376 		case SO_RCVTIMEO:
1377 		    {
1378 			struct timeval *tv;
1379 			short val;
1380 
1381 			if (m == NULL || m->m_len < sizeof(*tv)) {
1382 				error = EINVAL;
1383 				goto bad;
1384 			}
1385 			tv = mtod(m, struct timeval *);
1386 			if (tv->tv_sec > (SHRT_MAX - tv->tv_usec / tick) / hz) {
1387 				error = EDOM;
1388 				goto bad;
1389 			}
1390 			val = tv->tv_sec * hz + tv->tv_usec / tick;
1391 			if (val == 0 && tv->tv_usec != 0)
1392 				val = 1;
1393 
1394 			switch (optname) {
1395 
1396 			case SO_SNDTIMEO:
1397 				so->so_snd.sb_timeo = val;
1398 				break;
1399 			case SO_RCVTIMEO:
1400 				so->so_rcv.sb_timeo = val;
1401 				break;
1402 			}
1403 			break;
1404 		    }
1405 
1406 		default:
1407 			error = ENOPROTOOPT;
1408 			break;
1409 		}
1410 		if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) {
1411 			(void) ((*so->so_proto->pr_ctloutput)
1412 				  (PRCO_SETOPT, so, level, optname, &m0));
1413 			m = NULL;	/* freed by protocol */
1414 		}
1415 	}
1416  bad:
1417 	if (m)
1418 		(void) m_free(m);
1419 	return (error);
1420 }
1421 
1422 int
1423 sogetopt(struct socket *so, int level, int optname, struct mbuf **mp)
1424 {
1425 	struct mbuf	*m;
1426 
1427 	if (level != SOL_SOCKET) {
1428 		if (so->so_proto && so->so_proto->pr_ctloutput) {
1429 			return ((*so->so_proto->pr_ctloutput)
1430 				  (PRCO_GETOPT, so, level, optname, mp));
1431 		} else
1432 			return (ENOPROTOOPT);
1433 	} else {
1434 		m = m_get(M_WAIT, MT_SOOPTS);
1435 		m->m_len = sizeof(int);
1436 
1437 		switch (optname) {
1438 
1439 		case SO_LINGER:
1440 			m->m_len = sizeof(struct linger);
1441 			mtod(m, struct linger *)->l_onoff =
1442 				so->so_options & SO_LINGER;
1443 			mtod(m, struct linger *)->l_linger = so->so_linger;
1444 			break;
1445 
1446 		case SO_USELOOPBACK:
1447 		case SO_DONTROUTE:
1448 		case SO_DEBUG:
1449 		case SO_KEEPALIVE:
1450 		case SO_REUSEADDR:
1451 		case SO_REUSEPORT:
1452 		case SO_BROADCAST:
1453 		case SO_OOBINLINE:
1454 		case SO_TIMESTAMP:
1455 			*mtod(m, int *) = so->so_options & optname;
1456 			break;
1457 
1458 		case SO_TYPE:
1459 			*mtod(m, int *) = so->so_type;
1460 			break;
1461 
1462 		case SO_ERROR:
1463 			*mtod(m, int *) = so->so_error;
1464 			so->so_error = 0;
1465 			break;
1466 
1467 		case SO_SNDBUF:
1468 			*mtod(m, int *) = so->so_snd.sb_hiwat;
1469 			break;
1470 
1471 		case SO_RCVBUF:
1472 			*mtod(m, int *) = so->so_rcv.sb_hiwat;
1473 			break;
1474 
1475 		case SO_SNDLOWAT:
1476 			*mtod(m, int *) = so->so_snd.sb_lowat;
1477 			break;
1478 
1479 		case SO_RCVLOWAT:
1480 			*mtod(m, int *) = so->so_rcv.sb_lowat;
1481 			break;
1482 
1483 		case SO_SNDTIMEO:
1484 		case SO_RCVTIMEO:
1485 		    {
1486 			int val = (optname == SO_SNDTIMEO ?
1487 			     so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
1488 
1489 			m->m_len = sizeof(struct timeval);
1490 			mtod(m, struct timeval *)->tv_sec = val / hz;
1491 			mtod(m, struct timeval *)->tv_usec =
1492 			    (val % hz) * tick;
1493 			break;
1494 		    }
1495 
1496 		default:
1497 			(void)m_free(m);
1498 			return (ENOPROTOOPT);
1499 		}
1500 		*mp = m;
1501 		return (0);
1502 	}
1503 }
1504 
1505 void
1506 sohasoutofband(struct socket *so)
1507 {
1508 	struct proc *p;
1509 
1510 	if (so->so_pgid < 0)
1511 		gsignal(-so->so_pgid, SIGURG);
1512 	else if (so->so_pgid > 0 && (p = pfind(so->so_pgid)) != 0)
1513 		psignal(p, SIGURG);
1514 	selwakeup(&so->so_rcv.sb_sel);
1515 }
1516 
1517 static void
1518 filt_sordetach(struct knote *kn)
1519 {
1520 	struct socket	*so;
1521 
1522 	so = (struct socket *)kn->kn_fp->f_data;
1523 	SLIST_REMOVE(&so->so_rcv.sb_sel.sel_klist, kn, knote, kn_selnext);
1524 	if (SLIST_EMPTY(&so->so_rcv.sb_sel.sel_klist))
1525 		so->so_rcv.sb_flags &= ~SB_KNOTE;
1526 }
1527 
1528 /*ARGSUSED*/
1529 static int
1530 filt_soread(struct knote *kn, long hint)
1531 {
1532 	struct socket	*so;
1533 
1534 	so = (struct socket *)kn->kn_fp->f_data;
1535 	kn->kn_data = so->so_rcv.sb_cc;
1536 	if (so->so_state & SS_CANTRCVMORE) {
1537 		kn->kn_flags |= EV_EOF;
1538 		kn->kn_fflags = so->so_error;
1539 		return (1);
1540 	}
1541 	if (so->so_error)	/* temporary udp error */
1542 		return (1);
1543 	if (kn->kn_sfflags & NOTE_LOWAT)
1544 		return (kn->kn_data >= kn->kn_sdata);
1545 	return (kn->kn_data >= so->so_rcv.sb_lowat);
1546 }
1547 
1548 static void
1549 filt_sowdetach(struct knote *kn)
1550 {
1551 	struct socket	*so;
1552 
1553 	so = (struct socket *)kn->kn_fp->f_data;
1554 	SLIST_REMOVE(&so->so_snd.sb_sel.sel_klist, kn, knote, kn_selnext);
1555 	if (SLIST_EMPTY(&so->so_snd.sb_sel.sel_klist))
1556 		so->so_snd.sb_flags &= ~SB_KNOTE;
1557 }
1558 
1559 /*ARGSUSED*/
1560 static int
1561 filt_sowrite(struct knote *kn, long hint)
1562 {
1563 	struct socket	*so;
1564 
1565 	so = (struct socket *)kn->kn_fp->f_data;
1566 	kn->kn_data = sbspace(&so->so_snd);
1567 	if (so->so_state & SS_CANTSENDMORE) {
1568 		kn->kn_flags |= EV_EOF;
1569 		kn->kn_fflags = so->so_error;
1570 		return (1);
1571 	}
1572 	if (so->so_error)	/* temporary udp error */
1573 		return (1);
1574 	if (((so->so_state & SS_ISCONNECTED) == 0) &&
1575 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
1576 		return (0);
1577 	if (kn->kn_sfflags & NOTE_LOWAT)
1578 		return (kn->kn_data >= kn->kn_sdata);
1579 	return (kn->kn_data >= so->so_snd.sb_lowat);
1580 }
1581 
1582 /*ARGSUSED*/
1583 static int
1584 filt_solisten(struct knote *kn, long hint)
1585 {
1586 	struct socket	*so;
1587 
1588 	so = (struct socket *)kn->kn_fp->f_data;
1589 
1590 	/*
1591 	 * Set kn_data to number of incoming connections, not
1592 	 * counting partial (incomplete) connections.
1593 	 */
1594 	kn->kn_data = so->so_qlen;
1595 	return (kn->kn_data > 0);
1596 }
1597 
1598 static const struct filterops solisten_filtops =
1599 	{ 1, NULL, filt_sordetach, filt_solisten };
1600 static const struct filterops soread_filtops =
1601 	{ 1, NULL, filt_sordetach, filt_soread };
1602 static const struct filterops sowrite_filtops =
1603 	{ 1, NULL, filt_sowdetach, filt_sowrite };
1604 
1605 int
1606 soo_kqfilter(struct file *fp, struct knote *kn)
1607 {
1608 	struct socket	*so;
1609 	struct sockbuf	*sb;
1610 
1611 	so = (struct socket *)kn->kn_fp->f_data;
1612 	switch (kn->kn_filter) {
1613 	case EVFILT_READ:
1614 		if (so->so_options & SO_ACCEPTCONN)
1615 			kn->kn_fop = &solisten_filtops;
1616 		else
1617 			kn->kn_fop = &soread_filtops;
1618 		sb = &so->so_rcv;
1619 		break;
1620 	case EVFILT_WRITE:
1621 		kn->kn_fop = &sowrite_filtops;
1622 		sb = &so->so_snd;
1623 		break;
1624 	default:
1625 		return (1);
1626 	}
1627 	SLIST_INSERT_HEAD(&sb->sb_sel.sel_klist, kn, kn_selnext);
1628 	sb->sb_flags |= SB_KNOTE;
1629 	return (0);
1630 }
1631 
1632