xref: /dflybsd-src/sys/kern/uipc_syscalls.c (revision 93bffecadc0caefc46f12b736eab0e62c2b6f42e)
1 /*
2  * Copyright (c) 1982, 1986, 1989, 1990, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * sendfile(2) and related extensions:
6  * Copyright (c) 1998, David Greenman. All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. All advertising materials mentioning features or use of this software
17  *    must display the following acknowledgement:
18  *	This product includes software developed by the University of
19  *	California, Berkeley and its contributors.
20  * 4. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)uipc_syscalls.c	8.4 (Berkeley) 2/21/94
37  * $FreeBSD: src/sys/kern/uipc_syscalls.c,v 1.65.2.17 2003/04/04 17:11:16 tegge Exp $
38  * $DragonFly: src/sys/kern/uipc_syscalls.c,v 1.92 2008/11/26 13:10:56 sephe Exp $
39  */
40 
41 #include "opt_ktrace.h"
42 #include "opt_sctp.h"
43 
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/kernel.h>
47 #include <sys/sysproto.h>
48 #include <sys/malloc.h>
49 #include <sys/filedesc.h>
50 #include <sys/event.h>
51 #include <sys/proc.h>
52 #include <sys/fcntl.h>
53 #include <sys/file.h>
54 #include <sys/filio.h>
55 #include <sys/kern_syscall.h>
56 #include <sys/mbuf.h>
57 #include <sys/protosw.h>
58 #include <sys/sfbuf.h>
59 #include <sys/socket.h>
60 #include <sys/socketvar.h>
61 #include <sys/socketops.h>
62 #include <sys/uio.h>
63 #include <sys/vnode.h>
64 #include <sys/lock.h>
65 #include <sys/mount.h>
66 #ifdef KTRACE
67 #include <sys/ktrace.h>
68 #endif
69 #include <vm/vm.h>
70 #include <vm/vm_object.h>
71 #include <vm/vm_page.h>
72 #include <vm/vm_pageout.h>
73 #include <vm/vm_kern.h>
74 #include <vm/vm_extern.h>
75 #include <sys/file2.h>
76 #include <sys/signalvar.h>
77 #include <sys/serialize.h>
78 
79 #include <sys/thread2.h>
80 #include <sys/msgport2.h>
81 #include <sys/socketvar2.h>
82 #include <sys/mplock2.h>
83 #include <net/netmsg2.h>
84 
85 #ifdef SCTP
86 #include <netinet/sctp_peeloff.h>
87 #endif /* SCTP */
88 
89 /*
90  * System call interface to the socket abstraction.
91  */
92 
93 extern	struct fileops socketops;
94 
95 /*
96  * socket_args(int domain, int type, int protocol)
97  */
98 int
99 kern_socket(int domain, int type, int protocol, int *res)
100 {
101 	struct thread *td = curthread;
102 	struct filedesc *fdp = td->td_proc->p_fd;
103 	struct socket *so;
104 	struct file *fp;
105 	int fd, error;
106 
107 	KKASSERT(td->td_lwp);
108 
109 	error = falloc(td->td_lwp, &fp, &fd);
110 	if (error)
111 		return (error);
112 	error = socreate(domain, &so, type, protocol, td);
113 	if (error) {
114 		fsetfd(fdp, NULL, fd);
115 	} else {
116 		fp->f_type = DTYPE_SOCKET;
117 		fp->f_flag = FREAD | FWRITE;
118 		fp->f_ops = &socketops;
119 		fp->f_data = so;
120 		*res = fd;
121 		fsetfd(fdp, fp, fd);
122 	}
123 	fdrop(fp);
124 	return (error);
125 }
126 
127 /*
128  * MPALMOSTSAFE
129  */
130 int
131 sys_socket(struct socket_args *uap)
132 {
133 	int error;
134 
135 	get_mplock();
136 	error = kern_socket(uap->domain, uap->type, uap->protocol,
137 			    &uap->sysmsg_iresult);
138 	rel_mplock();
139 
140 	return (error);
141 }
142 
143 int
144 kern_bind(int s, struct sockaddr *sa)
145 {
146 	struct thread *td = curthread;
147 	struct proc *p = td->td_proc;
148 	struct file *fp;
149 	int error;
150 
151 	KKASSERT(p);
152 	error = holdsock(p->p_fd, s, &fp);
153 	if (error)
154 		return (error);
155 	error = sobind((struct socket *)fp->f_data, sa, td);
156 	fdrop(fp);
157 	return (error);
158 }
159 
160 /*
161  * bind_args(int s, caddr_t name, int namelen)
162  *
163  * MPALMOSTSAFE
164  */
165 int
166 sys_bind(struct bind_args *uap)
167 {
168 	struct sockaddr *sa;
169 	int error;
170 
171 	error = getsockaddr(&sa, uap->name, uap->namelen);
172 	if (error)
173 		return (error);
174 	get_mplock();
175 	error = kern_bind(uap->s, sa);
176 	rel_mplock();
177 	FREE(sa, M_SONAME);
178 
179 	return (error);
180 }
181 
182 int
183 kern_listen(int s, int backlog)
184 {
185 	struct thread *td = curthread;
186 	struct proc *p = td->td_proc;
187 	struct file *fp;
188 	int error;
189 
190 	KKASSERT(p);
191 	error = holdsock(p->p_fd, s, &fp);
192 	if (error)
193 		return (error);
194 	error = solisten((struct socket *)fp->f_data, backlog, td);
195 	fdrop(fp);
196 	return(error);
197 }
198 
199 /*
200  * listen_args(int s, int backlog)
201  *
202  * MPALMOSTSAFE
203  */
204 int
205 sys_listen(struct listen_args *uap)
206 {
207 	int error;
208 
209 	get_mplock();
210 	error = kern_listen(uap->s, uap->backlog);
211 	rel_mplock();
212 	return (error);
213 }
214 
215 /*
216  * Returns the accepted socket as well.
217  */
218 static boolean_t
219 soaccept_predicate(struct netmsg *msg0)
220 {
221 	struct netmsg_so_notify *msg = (struct netmsg_so_notify *)msg0;
222 	struct socket *head = msg->nm_so;
223 
224 	if (head->so_error != 0) {
225 		msg->nm_netmsg.nm_lmsg.ms_error = head->so_error;
226 		return (TRUE);
227 	}
228 	lwkt_gettoken(&head->so_rcv.ssb_token);
229 	if (!TAILQ_EMPTY(&head->so_comp)) {
230 		/* Abuse nm_so field as copy in/copy out parameter. XXX JH */
231 		msg->nm_so = TAILQ_FIRST(&head->so_comp);
232 		TAILQ_REMOVE(&head->so_comp, msg->nm_so, so_list);
233 		head->so_qlen--;
234 
235 		msg->nm_netmsg.nm_lmsg.ms_error = 0;
236 		lwkt_reltoken(&head->so_rcv.ssb_token);
237 		return (TRUE);
238 	}
239 	lwkt_reltoken(&head->so_rcv.ssb_token);
240 	if (head->so_state & SS_CANTRCVMORE) {
241 		msg->nm_netmsg.nm_lmsg.ms_error = ECONNABORTED;
242 		return (TRUE);
243 	}
244 	if (msg->nm_fflags & FNONBLOCK) {
245 		msg->nm_netmsg.nm_lmsg.ms_error = EWOULDBLOCK;
246 		return (TRUE);
247 	}
248 
249 	return (FALSE);
250 }
251 
252 /*
253  * The second argument to kern_accept() is a handle to a struct sockaddr.
254  * This allows kern_accept() to return a pointer to an allocated struct
255  * sockaddr which must be freed later with FREE().  The caller must
256  * initialize *name to NULL.
257  */
258 int
259 kern_accept(int s, int fflags, struct sockaddr **name, int *namelen, int *res)
260 {
261 	struct thread *td = curthread;
262 	struct filedesc *fdp = td->td_proc->p_fd;
263 	struct file *lfp = NULL;
264 	struct file *nfp = NULL;
265 	struct sockaddr *sa;
266 	struct socket *head, *so;
267 	struct netmsg_so_notify msg;
268 	int fd;
269 	u_int fflag;		/* type must match fp->f_flag */
270 	int error, tmp;
271 
272 	*res = -1;
273 	if (name && namelen && *namelen < 0)
274 		return (EINVAL);
275 
276 	error = holdsock(td->td_proc->p_fd, s, &lfp);
277 	if (error)
278 		return (error);
279 
280 	error = falloc(td->td_lwp, &nfp, &fd);
281 	if (error) {		/* Probably ran out of file descriptors. */
282 		fdrop(lfp);
283 		return (error);
284 	}
285 	head = (struct socket *)lfp->f_data;
286 	if ((head->so_options & SO_ACCEPTCONN) == 0) {
287 		error = EINVAL;
288 		goto done;
289 	}
290 
291 	if (fflags & O_FBLOCKING)
292 		fflags |= lfp->f_flag & ~FNONBLOCK;
293 	else if (fflags & O_FNONBLOCKING)
294 		fflags |= lfp->f_flag | FNONBLOCK;
295 	else
296 		fflags = lfp->f_flag;
297 
298 	/* optimize for uniprocessor case later XXX JH */
299 	netmsg_init_abortable(&msg.nm_netmsg, head, &curthread->td_msgport,
300 			      0, netmsg_so_notify, netmsg_so_notify_doabort);
301 	msg.nm_predicate = soaccept_predicate;
302 	msg.nm_fflags = fflags;
303 	msg.nm_so = head;
304 	msg.nm_etype = NM_REVENT;
305 	error = lwkt_domsg(head->so_port, &msg.nm_netmsg.nm_lmsg, PCATCH);
306 	if (error)
307 		goto done;
308 
309 	/*
310 	 * At this point we have the connection that's ready to be accepted.
311 	 */
312 	so = msg.nm_so;
313 
314 	fflag = lfp->f_flag;
315 
316 	/* connection has been removed from the listen queue */
317 	KNOTE(&head->so_rcv.ssb_kq.ki_note, 0);
318 
319 	soclrstate(so, SS_COMP);
320 	so->so_head = NULL;
321 	if (head->so_sigio != NULL)
322 		fsetown(fgetown(head->so_sigio), &so->so_sigio);
323 
324 	nfp->f_type = DTYPE_SOCKET;
325 	nfp->f_flag = fflag;
326 	nfp->f_ops = &socketops;
327 	nfp->f_data = so;
328 	/* Sync socket nonblocking/async state with file flags */
329 	tmp = fflag & FNONBLOCK;
330 	fo_ioctl(nfp, FIONBIO, (caddr_t)&tmp, td->td_ucred, NULL);
331 	tmp = fflag & FASYNC;
332 	fo_ioctl(nfp, FIOASYNC, (caddr_t)&tmp, td->td_ucred, NULL);
333 
334 	sa = NULL;
335 	error = soaccept(so, &sa);
336 
337 	/*
338 	 * Set the returned name and namelen as applicable.  Set the returned
339 	 * namelen to 0 for older code which might ignore the return value
340 	 * from accept.
341 	 */
342 	if (error == 0) {
343 		if (sa && name && namelen) {
344 			if (*namelen > sa->sa_len)
345 				*namelen = sa->sa_len;
346 			*name = sa;
347 		} else {
348 			if (sa)
349 				FREE(sa, M_SONAME);
350 		}
351 	}
352 
353 done:
354 	/*
355 	 * If an error occured clear the reserved descriptor, else associate
356 	 * nfp with it.
357 	 *
358 	 * Note that *res is normally ignored if an error is returned but
359 	 * a syscall message will still have access to the result code.
360 	 */
361 	if (error) {
362 		fsetfd(fdp, NULL, fd);
363 	} else {
364 		*res = fd;
365 		fsetfd(fdp, nfp, fd);
366 	}
367 	fdrop(nfp);
368 	fdrop(lfp);
369 	return (error);
370 }
371 
372 /*
373  * accept(int s, caddr_t name, int *anamelen)
374  *
375  * MPALMOSTSAFE
376  */
377 int
378 sys_accept(struct accept_args *uap)
379 {
380 	struct sockaddr *sa = NULL;
381 	int sa_len;
382 	int error;
383 
384 	if (uap->name) {
385 		error = copyin(uap->anamelen, &sa_len, sizeof(sa_len));
386 		if (error)
387 			return (error);
388 
389 		get_mplock();
390 		error = kern_accept(uap->s, 0, &sa, &sa_len,
391 				    &uap->sysmsg_iresult);
392 		rel_mplock();
393 
394 		if (error == 0)
395 			error = copyout(sa, uap->name, sa_len);
396 		if (error == 0) {
397 			error = copyout(&sa_len, uap->anamelen,
398 			    sizeof(*uap->anamelen));
399 		}
400 		if (sa)
401 			FREE(sa, M_SONAME);
402 	} else {
403 		get_mplock();
404 		error = kern_accept(uap->s, 0, NULL, 0,
405 				    &uap->sysmsg_iresult);
406 		rel_mplock();
407 	}
408 	return (error);
409 }
410 
411 /*
412  * extaccept(int s, int fflags, caddr_t name, int *anamelen)
413  *
414  * MPALMOSTSAFE
415  */
416 int
417 sys_extaccept(struct extaccept_args *uap)
418 {
419 	struct sockaddr *sa = NULL;
420 	int sa_len;
421 	int error;
422 	int fflags = uap->flags & O_FMASK;
423 
424 	if (uap->name) {
425 		error = copyin(uap->anamelen, &sa_len, sizeof(sa_len));
426 		if (error)
427 			return (error);
428 
429 		get_mplock();
430 		error = kern_accept(uap->s, fflags, &sa, &sa_len,
431 				    &uap->sysmsg_iresult);
432 		rel_mplock();
433 
434 		if (error == 0)
435 			error = copyout(sa, uap->name, sa_len);
436 		if (error == 0) {
437 			error = copyout(&sa_len, uap->anamelen,
438 			    sizeof(*uap->anamelen));
439 		}
440 		if (sa)
441 			FREE(sa, M_SONAME);
442 	} else {
443 		get_mplock();
444 		error = kern_accept(uap->s, fflags, NULL, 0,
445 				    &uap->sysmsg_iresult);
446 		rel_mplock();
447 	}
448 	return (error);
449 }
450 
451 
452 /*
453  * Returns TRUE if predicate satisfied.
454  */
455 static boolean_t
456 soconnected_predicate(struct netmsg *msg0)
457 {
458 	struct netmsg_so_notify *msg = (struct netmsg_so_notify *)msg0;
459 	struct socket *so = msg->nm_so;
460 
461 	/* check predicate */
462 	if (!(so->so_state & SS_ISCONNECTING) || so->so_error != 0) {
463 		msg->nm_netmsg.nm_lmsg.ms_error = so->so_error;
464 		return (TRUE);
465 	}
466 
467 	return (FALSE);
468 }
469 
470 int
471 kern_connect(int s, int fflags, struct sockaddr *sa)
472 {
473 	struct thread *td = curthread;
474 	struct proc *p = td->td_proc;
475 	struct file *fp;
476 	struct socket *so;
477 	int error, interrupted = 0;
478 
479 	error = holdsock(p->p_fd, s, &fp);
480 	if (error)
481 		return (error);
482 	so = (struct socket *)fp->f_data;
483 
484 	if (fflags & O_FBLOCKING)
485 		/* fflags &= ~FNONBLOCK; */;
486 	else if (fflags & O_FNONBLOCKING)
487 		fflags |= FNONBLOCK;
488 	else
489 		fflags = fp->f_flag;
490 
491 	if (so->so_state & SS_ISCONNECTING) {
492 		error = EALREADY;
493 		goto done;
494 	}
495 	error = soconnect(so, sa, td);
496 	if (error)
497 		goto bad;
498 	if ((fflags & FNONBLOCK) && (so->so_state & SS_ISCONNECTING)) {
499 		error = EINPROGRESS;
500 		goto done;
501 	}
502 	if ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) {
503 		struct netmsg_so_notify msg;
504 
505 		netmsg_init_abortable(&msg.nm_netmsg, so,
506 				      &curthread->td_msgport,
507 				      0,
508 				      netmsg_so_notify,
509 				      netmsg_so_notify_doabort);
510 		msg.nm_predicate = soconnected_predicate;
511 		msg.nm_so = so;
512 		msg.nm_etype = NM_REVENT;
513 		error = lwkt_domsg(so->so_port, &msg.nm_netmsg.nm_lmsg, PCATCH);
514 		if (error == EINTR || error == ERESTART)
515 			interrupted = 1;
516 	}
517 	if (error == 0) {
518 		error = so->so_error;
519 		so->so_error = 0;
520 	}
521 bad:
522 	if (!interrupted)
523 		soclrstate(so, SS_ISCONNECTING);
524 	if (error == ERESTART)
525 		error = EINTR;
526 done:
527 	fdrop(fp);
528 	return (error);
529 }
530 
531 /*
532  * connect_args(int s, caddr_t name, int namelen)
533  *
534  * MPALMOSTSAFE
535  */
536 int
537 sys_connect(struct connect_args *uap)
538 {
539 	struct sockaddr *sa;
540 	int error;
541 
542 	error = getsockaddr(&sa, uap->name, uap->namelen);
543 	if (error)
544 		return (error);
545 	get_mplock();
546 	error = kern_connect(uap->s, 0, sa);
547 	rel_mplock();
548 	FREE(sa, M_SONAME);
549 
550 	return (error);
551 }
552 
553 /*
554  * connect_args(int s, int fflags, caddr_t name, int namelen)
555  *
556  * MPALMOSTSAFE
557  */
558 int
559 sys_extconnect(struct extconnect_args *uap)
560 {
561 	struct sockaddr *sa;
562 	int error;
563 	int fflags = uap->flags & O_FMASK;
564 
565 	error = getsockaddr(&sa, uap->name, uap->namelen);
566 	if (error)
567 		return (error);
568 	get_mplock();
569 	error = kern_connect(uap->s, fflags, sa);
570 	rel_mplock();
571 	FREE(sa, M_SONAME);
572 
573 	return (error);
574 }
575 
576 int
577 kern_socketpair(int domain, int type, int protocol, int *sv)
578 {
579 	struct thread *td = curthread;
580 	struct filedesc *fdp;
581 	struct file *fp1, *fp2;
582 	struct socket *so1, *so2;
583 	int fd1, fd2, error;
584 
585 	fdp = td->td_proc->p_fd;
586 	error = socreate(domain, &so1, type, protocol, td);
587 	if (error)
588 		return (error);
589 	error = socreate(domain, &so2, type, protocol, td);
590 	if (error)
591 		goto free1;
592 	error = falloc(td->td_lwp, &fp1, &fd1);
593 	if (error)
594 		goto free2;
595 	sv[0] = fd1;
596 	fp1->f_data = so1;
597 	error = falloc(td->td_lwp, &fp2, &fd2);
598 	if (error)
599 		goto free3;
600 	fp2->f_data = so2;
601 	sv[1] = fd2;
602 	error = soconnect2(so1, so2);
603 	if (error)
604 		goto free4;
605 	if (type == SOCK_DGRAM) {
606 		/*
607 		 * Datagram socket connection is asymmetric.
608 		 */
609 		 error = soconnect2(so2, so1);
610 		 if (error)
611 			goto free4;
612 	}
613 	fp1->f_type = fp2->f_type = DTYPE_SOCKET;
614 	fp1->f_flag = fp2->f_flag = FREAD|FWRITE;
615 	fp1->f_ops = fp2->f_ops = &socketops;
616 	fsetfd(fdp, fp1, fd1);
617 	fsetfd(fdp, fp2, fd2);
618 	fdrop(fp1);
619 	fdrop(fp2);
620 	return (error);
621 free4:
622 	fsetfd(fdp, NULL, fd2);
623 	fdrop(fp2);
624 free3:
625 	fsetfd(fdp, NULL, fd1);
626 	fdrop(fp1);
627 free2:
628 	(void)soclose(so2, 0);
629 free1:
630 	(void)soclose(so1, 0);
631 	return (error);
632 }
633 
634 /*
635  * socketpair(int domain, int type, int protocol, int *rsv)
636  *
637  * MPALMOSTSAFE
638  */
639 int
640 sys_socketpair(struct socketpair_args *uap)
641 {
642 	int error, sockv[2];
643 
644 	get_mplock();
645 	error = kern_socketpair(uap->domain, uap->type, uap->protocol, sockv);
646 	rel_mplock();
647 
648 	if (error == 0)
649 		error = copyout(sockv, uap->rsv, sizeof(sockv));
650 	return (error);
651 }
652 
653 int
654 kern_sendmsg(int s, struct sockaddr *sa, struct uio *auio,
655 	     struct mbuf *control, int flags, size_t *res)
656 {
657 	struct thread *td = curthread;
658 	struct lwp *lp = td->td_lwp;
659 	struct proc *p = td->td_proc;
660 	struct file *fp;
661 	size_t len;
662 	int error;
663 	struct socket *so;
664 #ifdef KTRACE
665 	struct iovec *ktriov = NULL;
666 	struct uio ktruio;
667 #endif
668 
669 	error = holdsock(p->p_fd, s, &fp);
670 	if (error)
671 		return (error);
672 #ifdef KTRACE
673 	if (KTRPOINT(td, KTR_GENIO)) {
674 		int iovlen = auio->uio_iovcnt * sizeof (struct iovec);
675 
676 		MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK);
677 		bcopy((caddr_t)auio->uio_iov, (caddr_t)ktriov, iovlen);
678 		ktruio = *auio;
679 	}
680 #endif
681 	len = auio->uio_resid;
682 	so = (struct socket *)fp->f_data;
683 	if ((flags & (MSG_FNONBLOCKING|MSG_FBLOCKING)) == 0) {
684 		if (fp->f_flag & FNONBLOCK)
685 			flags |= MSG_FNONBLOCKING;
686 	}
687 	error = so_pru_sosend(so, sa, auio, NULL, control, flags, td);
688 	if (error) {
689 		if (auio->uio_resid != len && (error == ERESTART ||
690 		    error == EINTR || error == EWOULDBLOCK))
691 			error = 0;
692 		if (error == EPIPE)
693 			lwpsignal(p, lp, SIGPIPE);
694 	}
695 #ifdef KTRACE
696 	if (ktriov != NULL) {
697 		if (error == 0) {
698 			ktruio.uio_iov = ktriov;
699 			ktruio.uio_resid = len - auio->uio_resid;
700 			ktrgenio(lp, s, UIO_WRITE, &ktruio, error);
701 		}
702 		FREE(ktriov, M_TEMP);
703 	}
704 #endif
705 	if (error == 0)
706 		*res  = len - auio->uio_resid;
707 	fdrop(fp);
708 	return (error);
709 }
710 
711 /*
712  * sendto_args(int s, caddr_t buf, size_t len, int flags, caddr_t to, int tolen)
713  *
714  * MPALMOSTSAFE
715  */
716 int
717 sys_sendto(struct sendto_args *uap)
718 {
719 	struct thread *td = curthread;
720 	struct uio auio;
721 	struct iovec aiov;
722 	struct sockaddr *sa = NULL;
723 	int error;
724 
725 	if (uap->to) {
726 		error = getsockaddr(&sa, uap->to, uap->tolen);
727 		if (error)
728 			return (error);
729 	}
730 	aiov.iov_base = uap->buf;
731 	aiov.iov_len = uap->len;
732 	auio.uio_iov = &aiov;
733 	auio.uio_iovcnt = 1;
734 	auio.uio_offset = 0;
735 	auio.uio_resid = uap->len;
736 	auio.uio_segflg = UIO_USERSPACE;
737 	auio.uio_rw = UIO_WRITE;
738 	auio.uio_td = td;
739 
740 	get_mplock();
741 	error = kern_sendmsg(uap->s, sa, &auio, NULL, uap->flags,
742 			     &uap->sysmsg_szresult);
743 	rel_mplock();
744 
745 	if (sa)
746 		FREE(sa, M_SONAME);
747 	return (error);
748 }
749 
750 /*
751  * sendmsg_args(int s, caddr_t msg, int flags)
752  *
753  * MPALMOSTSAFE
754  */
755 int
756 sys_sendmsg(struct sendmsg_args *uap)
757 {
758 	struct thread *td = curthread;
759 	struct msghdr msg;
760 	struct uio auio;
761 	struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
762 	struct sockaddr *sa = NULL;
763 	struct mbuf *control = NULL;
764 	int error;
765 
766 	error = copyin(uap->msg, (caddr_t)&msg, sizeof(msg));
767 	if (error)
768 		return (error);
769 
770 	/*
771 	 * Conditionally copyin msg.msg_name.
772 	 */
773 	if (msg.msg_name) {
774 		error = getsockaddr(&sa, msg.msg_name, msg.msg_namelen);
775 		if (error)
776 			return (error);
777 	}
778 
779 	/*
780 	 * Populate auio.
781 	 */
782 	error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen,
783 			     &auio.uio_resid);
784 	if (error)
785 		goto cleanup2;
786 	auio.uio_iov = iov;
787 	auio.uio_iovcnt = msg.msg_iovlen;
788 	auio.uio_offset = 0;
789 	auio.uio_segflg = UIO_USERSPACE;
790 	auio.uio_rw = UIO_WRITE;
791 	auio.uio_td = td;
792 
793 	/*
794 	 * Conditionally copyin msg.msg_control.
795 	 */
796 	if (msg.msg_control) {
797 		if (msg.msg_controllen < sizeof(struct cmsghdr) ||
798 		    msg.msg_controllen > MLEN) {
799 			error = EINVAL;
800 			goto cleanup;
801 		}
802 		control = m_get(MB_WAIT, MT_CONTROL);
803 		if (control == NULL) {
804 			error = ENOBUFS;
805 			goto cleanup;
806 		}
807 		control->m_len = msg.msg_controllen;
808 		error = copyin(msg.msg_control, mtod(control, caddr_t),
809 			       msg.msg_controllen);
810 		if (error) {
811 			m_free(control);
812 			goto cleanup;
813 		}
814 	}
815 
816 	get_mplock();
817 	error = kern_sendmsg(uap->s, sa, &auio, control, uap->flags,
818 			     &uap->sysmsg_szresult);
819 	rel_mplock();
820 
821 cleanup:
822 	iovec_free(&iov, aiov);
823 cleanup2:
824 	if (sa)
825 		FREE(sa, M_SONAME);
826 	return (error);
827 }
828 
829 /*
830  * kern_recvmsg() takes a handle to sa and control.  If the handle is non-
831  * null, it returns a dynamically allocated struct sockaddr and an mbuf.
832  * Don't forget to FREE() and m_free() these if they are returned.
833  */
834 int
835 kern_recvmsg(int s, struct sockaddr **sa, struct uio *auio,
836 	     struct mbuf **control, int *flags, size_t *res)
837 {
838 	struct thread *td = curthread;
839 	struct proc *p = td->td_proc;
840 	struct file *fp;
841 	size_t len;
842 	int error;
843 	int lflags;
844 	struct socket *so;
845 #ifdef KTRACE
846 	struct iovec *ktriov = NULL;
847 	struct uio ktruio;
848 #endif
849 
850 	error = holdsock(p->p_fd, s, &fp);
851 	if (error)
852 		return (error);
853 #ifdef KTRACE
854 	if (KTRPOINT(td, KTR_GENIO)) {
855 		int iovlen = auio->uio_iovcnt * sizeof (struct iovec);
856 
857 		MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK);
858 		bcopy(auio->uio_iov, ktriov, iovlen);
859 		ktruio = *auio;
860 	}
861 #endif
862 	len = auio->uio_resid;
863 	so = (struct socket *)fp->f_data;
864 
865 	if (flags == NULL || (*flags & (MSG_FNONBLOCKING|MSG_FBLOCKING)) == 0) {
866 		if (fp->f_flag & FNONBLOCK) {
867 			if (flags) {
868 				*flags |= MSG_FNONBLOCKING;
869 			} else {
870 				lflags = MSG_FNONBLOCKING;
871 				flags = &lflags;
872 			}
873 		}
874 	}
875 
876 	error = so_pru_soreceive(so, sa, auio, NULL, control, flags);
877 	if (error) {
878 		if (auio->uio_resid != len && (error == ERESTART ||
879 		    error == EINTR || error == EWOULDBLOCK))
880 			error = 0;
881 	}
882 #ifdef KTRACE
883 	if (ktriov != NULL) {
884 		if (error == 0) {
885 			ktruio.uio_iov = ktriov;
886 			ktruio.uio_resid = len - auio->uio_resid;
887 			ktrgenio(td->td_lwp, s, UIO_READ, &ktruio, error);
888 		}
889 		FREE(ktriov, M_TEMP);
890 	}
891 #endif
892 	if (error == 0)
893 		*res = len - auio->uio_resid;
894 	fdrop(fp);
895 	return (error);
896 }
897 
898 /*
899  * recvfrom_args(int s, caddr_t buf, size_t len, int flags,
900  *			caddr_t from, int *fromlenaddr)
901  *
902  * MPALMOSTSAFE
903  */
904 int
905 sys_recvfrom(struct recvfrom_args *uap)
906 {
907 	struct thread *td = curthread;
908 	struct uio auio;
909 	struct iovec aiov;
910 	struct sockaddr *sa = NULL;
911 	int error, fromlen;
912 
913 	if (uap->from && uap->fromlenaddr) {
914 		error = copyin(uap->fromlenaddr, &fromlen, sizeof(fromlen));
915 		if (error)
916 			return (error);
917 		if (fromlen < 0)
918 			return (EINVAL);
919 	} else {
920 		fromlen = 0;
921 	}
922 	aiov.iov_base = uap->buf;
923 	aiov.iov_len = uap->len;
924 	auio.uio_iov = &aiov;
925 	auio.uio_iovcnt = 1;
926 	auio.uio_offset = 0;
927 	auio.uio_resid = uap->len;
928 	auio.uio_segflg = UIO_USERSPACE;
929 	auio.uio_rw = UIO_READ;
930 	auio.uio_td = td;
931 
932 	get_mplock();
933 	error = kern_recvmsg(uap->s, uap->from ? &sa : NULL, &auio, NULL,
934 			     &uap->flags, &uap->sysmsg_szresult);
935 	rel_mplock();
936 
937 	if (error == 0 && uap->from) {
938 		/* note: sa may still be NULL */
939 		if (sa) {
940 			fromlen = MIN(fromlen, sa->sa_len);
941 			error = copyout(sa, uap->from, fromlen);
942 		} else {
943 			fromlen = 0;
944 		}
945 		if (error == 0) {
946 			error = copyout(&fromlen, uap->fromlenaddr,
947 					sizeof(fromlen));
948 		}
949 	}
950 	if (sa)
951 		FREE(sa, M_SONAME);
952 
953 	return (error);
954 }
955 
956 /*
957  * recvmsg_args(int s, struct msghdr *msg, int flags)
958  *
959  * MPALMOSTSAFE
960  */
961 int
962 sys_recvmsg(struct recvmsg_args *uap)
963 {
964 	struct thread *td = curthread;
965 	struct msghdr msg;
966 	struct uio auio;
967 	struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
968 	struct mbuf *m, *control = NULL;
969 	struct sockaddr *sa = NULL;
970 	caddr_t ctlbuf;
971 	socklen_t *ufromlenp, *ucontrollenp;
972 	int error, fromlen, controllen, len, flags, *uflagsp;
973 
974 	/*
975 	 * This copyin handles everything except the iovec.
976 	 */
977 	error = copyin(uap->msg, &msg, sizeof(msg));
978 	if (error)
979 		return (error);
980 
981 	if (msg.msg_name && msg.msg_namelen < 0)
982 		return (EINVAL);
983 	if (msg.msg_control && msg.msg_controllen < 0)
984 		return (EINVAL);
985 
986 	ufromlenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr,
987 		    msg_namelen));
988 	ucontrollenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr,
989 		       msg_controllen));
990 	uflagsp = (int *)((caddr_t)uap->msg + offsetof(struct msghdr,
991 							msg_flags));
992 
993 	/*
994 	 * Populate auio.
995 	 */
996 	error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen,
997 			     &auio.uio_resid);
998 	if (error)
999 		return (error);
1000 	auio.uio_iov = iov;
1001 	auio.uio_iovcnt = msg.msg_iovlen;
1002 	auio.uio_offset = 0;
1003 	auio.uio_segflg = UIO_USERSPACE;
1004 	auio.uio_rw = UIO_READ;
1005 	auio.uio_td = td;
1006 
1007 	flags = uap->flags;
1008 
1009 	get_mplock();
1010 	error = kern_recvmsg(uap->s,
1011 			     (msg.msg_name ? &sa : NULL), &auio,
1012 			     (msg.msg_control ? &control : NULL), &flags,
1013 			     &uap->sysmsg_szresult);
1014 	rel_mplock();
1015 
1016 	/*
1017 	 * Conditionally copyout the name and populate the namelen field.
1018 	 */
1019 	if (error == 0 && msg.msg_name) {
1020 		/* note: sa may still be NULL */
1021 		if (sa != NULL) {
1022 			fromlen = MIN(msg.msg_namelen, sa->sa_len);
1023 			error = copyout(sa, msg.msg_name, fromlen);
1024 		} else {
1025 			fromlen = 0;
1026 		}
1027 		if (error == 0)
1028 			error = copyout(&fromlen, ufromlenp,
1029 			    sizeof(*ufromlenp));
1030 	}
1031 
1032 	/*
1033 	 * Copyout msg.msg_control and msg.msg_controllen.
1034 	 */
1035 	if (error == 0 && msg.msg_control) {
1036 		len = msg.msg_controllen;
1037 		m = control;
1038 		ctlbuf = (caddr_t)msg.msg_control;
1039 
1040 		while(m && len > 0) {
1041 			unsigned int tocopy;
1042 
1043 			if (len >= m->m_len) {
1044 				tocopy = m->m_len;
1045 			} else {
1046 				msg.msg_flags |= MSG_CTRUNC;
1047 				tocopy = len;
1048 			}
1049 
1050 			error = copyout(mtod(m, caddr_t), ctlbuf, tocopy);
1051 			if (error)
1052 				goto cleanup;
1053 
1054 			ctlbuf += tocopy;
1055 			len -= tocopy;
1056 			m = m->m_next;
1057 		}
1058 		controllen = ctlbuf - (caddr_t)msg.msg_control;
1059 		error = copyout(&controllen, ucontrollenp,
1060 		    sizeof(*ucontrollenp));
1061 	}
1062 
1063 	if (error == 0)
1064 		error = copyout(&flags, uflagsp, sizeof(*uflagsp));
1065 
1066 cleanup:
1067 	if (sa)
1068 		FREE(sa, M_SONAME);
1069 	iovec_free(&iov, aiov);
1070 	if (control)
1071 		m_freem(control);
1072 	return (error);
1073 }
1074 
1075 /*
1076  * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an
1077  * in kernel pointer instead of a userland pointer.  This allows us
1078  * to manipulate socket options in the emulation code.
1079  */
1080 int
1081 kern_setsockopt(int s, struct sockopt *sopt)
1082 {
1083 	struct thread *td = curthread;
1084 	struct proc *p = td->td_proc;
1085 	struct file *fp;
1086 	int error;
1087 
1088 	if (sopt->sopt_val == NULL && sopt->sopt_valsize != 0)
1089 		return (EFAULT);
1090 	if (sopt->sopt_val != NULL && sopt->sopt_valsize == 0)
1091 		return (EINVAL);
1092 	if (sopt->sopt_valsize < 0)
1093 		return (EINVAL);
1094 
1095 	error = holdsock(p->p_fd, s, &fp);
1096 	if (error)
1097 		return (error);
1098 
1099 	error = sosetopt((struct socket *)fp->f_data, sopt);
1100 	fdrop(fp);
1101 	return (error);
1102 }
1103 
1104 /*
1105  * setsockopt_args(int s, int level, int name, caddr_t val, int valsize)
1106  *
1107  * MPALMOSTSAFE
1108  */
1109 int
1110 sys_setsockopt(struct setsockopt_args *uap)
1111 {
1112 	struct thread *td = curthread;
1113 	struct sockopt sopt;
1114 	int error;
1115 
1116 	sopt.sopt_level = uap->level;
1117 	sopt.sopt_name = uap->name;
1118 	sopt.sopt_valsize = uap->valsize;
1119 	sopt.sopt_td = td;
1120 	sopt.sopt_val = NULL;
1121 
1122 	if (sopt.sopt_valsize < 0 || sopt.sopt_valsize > SOMAXOPT_SIZE)
1123 		return (EINVAL);
1124 	if (uap->val) {
1125 		sopt.sopt_val = kmalloc(sopt.sopt_valsize, M_TEMP, M_WAITOK);
1126 		error = copyin(uap->val, sopt.sopt_val, sopt.sopt_valsize);
1127 		if (error)
1128 			goto out;
1129 	}
1130 
1131 	get_mplock();
1132 	error = kern_setsockopt(uap->s, &sopt);
1133 	rel_mplock();
1134 out:
1135 	if (uap->val)
1136 		kfree(sopt.sopt_val, M_TEMP);
1137 	return(error);
1138 }
1139 
1140 /*
1141  * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an
1142  * in kernel pointer instead of a userland pointer.  This allows us
1143  * to manipulate socket options in the emulation code.
1144  */
1145 int
1146 kern_getsockopt(int s, struct sockopt *sopt)
1147 {
1148 	struct thread *td = curthread;
1149 	struct proc *p = td->td_proc;
1150 	struct file *fp;
1151 	int error;
1152 
1153 	if (sopt->sopt_val == NULL && sopt->sopt_valsize != 0)
1154 		return (EFAULT);
1155 	if (sopt->sopt_val != NULL && sopt->sopt_valsize == 0)
1156 		return (EINVAL);
1157 	if (sopt->sopt_valsize < 0 || sopt->sopt_valsize > SOMAXOPT_SIZE)
1158 		return (EINVAL);
1159 
1160 	error = holdsock(p->p_fd, s, &fp);
1161 	if (error)
1162 		return (error);
1163 
1164 	error = sogetopt((struct socket *)fp->f_data, sopt);
1165 	fdrop(fp);
1166 	return (error);
1167 }
1168 
1169 /*
1170  * getsockopt_args(int s, int level, int name, caddr_t val, int *avalsize)
1171  *
1172  * MPALMOSTSAFE
1173  */
1174 int
1175 sys_getsockopt(struct getsockopt_args *uap)
1176 {
1177 	struct thread *td = curthread;
1178 	struct	sockopt sopt;
1179 	int	error, valsize;
1180 
1181 	if (uap->val) {
1182 		error = copyin(uap->avalsize, &valsize, sizeof(valsize));
1183 		if (error)
1184 			return (error);
1185 	} else {
1186 		valsize = 0;
1187 	}
1188 
1189 	sopt.sopt_level = uap->level;
1190 	sopt.sopt_name = uap->name;
1191 	sopt.sopt_valsize = valsize;
1192 	sopt.sopt_td = td;
1193 	sopt.sopt_val = NULL;
1194 
1195 	if (sopt.sopt_valsize < 0 || sopt.sopt_valsize > SOMAXOPT_SIZE)
1196 		return (EINVAL);
1197 	if (uap->val) {
1198 		sopt.sopt_val = kmalloc(sopt.sopt_valsize, M_TEMP, M_WAITOK);
1199 		error = copyin(uap->val, sopt.sopt_val, sopt.sopt_valsize);
1200 		if (error)
1201 			goto out;
1202 	}
1203 
1204 	get_mplock();
1205 	error = kern_getsockopt(uap->s, &sopt);
1206 	rel_mplock();
1207 	if (error)
1208 		goto out;
1209 	valsize = sopt.sopt_valsize;
1210 	error = copyout(&valsize, uap->avalsize, sizeof(valsize));
1211 	if (error)
1212 		goto out;
1213 	if (uap->val)
1214 		error = copyout(sopt.sopt_val, uap->val, sopt.sopt_valsize);
1215 out:
1216 	if (uap->val)
1217 		kfree(sopt.sopt_val, M_TEMP);
1218 	return (error);
1219 }
1220 
1221 /*
1222  * The second argument to kern_getsockname() is a handle to a struct sockaddr.
1223  * This allows kern_getsockname() to return a pointer to an allocated struct
1224  * sockaddr which must be freed later with FREE().  The caller must
1225  * initialize *name to NULL.
1226  */
1227 int
1228 kern_getsockname(int s, struct sockaddr **name, int *namelen)
1229 {
1230 	struct thread *td = curthread;
1231 	struct proc *p = td->td_proc;
1232 	struct file *fp;
1233 	struct socket *so;
1234 	struct sockaddr *sa = NULL;
1235 	int error;
1236 
1237 	error = holdsock(p->p_fd, s, &fp);
1238 	if (error)
1239 		return (error);
1240 	if (*namelen < 0) {
1241 		fdrop(fp);
1242 		return (EINVAL);
1243 	}
1244 	so = (struct socket *)fp->f_data;
1245 	error = so_pru_sockaddr(so, &sa);
1246 	if (error == 0) {
1247 		if (sa == NULL) {
1248 			*namelen = 0;
1249 		} else {
1250 			*namelen = MIN(*namelen, sa->sa_len);
1251 			*name = sa;
1252 		}
1253 	}
1254 
1255 	fdrop(fp);
1256 	return (error);
1257 }
1258 
1259 /*
1260  * getsockname_args(int fdes, caddr_t asa, int *alen)
1261  *
1262  * Get socket name.
1263  *
1264  * MPALMOSTSAFE
1265  */
1266 int
1267 sys_getsockname(struct getsockname_args *uap)
1268 {
1269 	struct sockaddr *sa = NULL;
1270 	int error, sa_len;
1271 
1272 	error = copyin(uap->alen, &sa_len, sizeof(sa_len));
1273 	if (error)
1274 		return (error);
1275 
1276 	get_mplock();
1277 	error = kern_getsockname(uap->fdes, &sa, &sa_len);
1278 	rel_mplock();
1279 
1280 	if (error == 0)
1281 		error = copyout(sa, uap->asa, sa_len);
1282 	if (error == 0)
1283 		error = copyout(&sa_len, uap->alen, sizeof(*uap->alen));
1284 	if (sa)
1285 		FREE(sa, M_SONAME);
1286 	return (error);
1287 }
1288 
1289 /*
1290  * The second argument to kern_getpeername() is a handle to a struct sockaddr.
1291  * This allows kern_getpeername() to return a pointer to an allocated struct
1292  * sockaddr which must be freed later with FREE().  The caller must
1293  * initialize *name to NULL.
1294  */
1295 int
1296 kern_getpeername(int s, struct sockaddr **name, int *namelen)
1297 {
1298 	struct thread *td = curthread;
1299 	struct proc *p = td->td_proc;
1300 	struct file *fp;
1301 	struct socket *so;
1302 	struct sockaddr *sa = NULL;
1303 	int error;
1304 
1305 	error = holdsock(p->p_fd, s, &fp);
1306 	if (error)
1307 		return (error);
1308 	if (*namelen < 0) {
1309 		fdrop(fp);
1310 		return (EINVAL);
1311 	}
1312 	so = (struct socket *)fp->f_data;
1313 	if ((so->so_state & (SS_ISCONNECTED|SS_ISCONFIRMING)) == 0) {
1314 		fdrop(fp);
1315 		return (ENOTCONN);
1316 	}
1317 	error = so_pru_peeraddr(so, &sa);
1318 	if (error == 0) {
1319 		if (sa == NULL) {
1320 			*namelen = 0;
1321 		} else {
1322 			*namelen = MIN(*namelen, sa->sa_len);
1323 			*name = sa;
1324 		}
1325 	}
1326 
1327 	fdrop(fp);
1328 	return (error);
1329 }
1330 
1331 /*
1332  * getpeername_args(int fdes, caddr_t asa, int *alen)
1333  *
1334  * Get name of peer for connected socket.
1335  *
1336  * MPALMOSTSAFE
1337  */
1338 int
1339 sys_getpeername(struct getpeername_args *uap)
1340 {
1341 	struct sockaddr *sa = NULL;
1342 	int error, sa_len;
1343 
1344 	error = copyin(uap->alen, &sa_len, sizeof(sa_len));
1345 	if (error)
1346 		return (error);
1347 
1348 	get_mplock();
1349 	error = kern_getpeername(uap->fdes, &sa, &sa_len);
1350 	rel_mplock();
1351 
1352 	if (error == 0)
1353 		error = copyout(sa, uap->asa, sa_len);
1354 	if (error == 0)
1355 		error = copyout(&sa_len, uap->alen, sizeof(*uap->alen));
1356 	if (sa)
1357 		FREE(sa, M_SONAME);
1358 	return (error);
1359 }
1360 
1361 int
1362 getsockaddr(struct sockaddr **namp, caddr_t uaddr, size_t len)
1363 {
1364 	struct sockaddr *sa;
1365 	int error;
1366 
1367 	*namp = NULL;
1368 	if (len > SOCK_MAXADDRLEN)
1369 		return ENAMETOOLONG;
1370 	if (len < offsetof(struct sockaddr, sa_data[0]))
1371 		return EDOM;
1372 	MALLOC(sa, struct sockaddr *, len, M_SONAME, M_WAITOK);
1373 	error = copyin(uaddr, sa, len);
1374 	if (error) {
1375 		FREE(sa, M_SONAME);
1376 	} else {
1377 #if BYTE_ORDER != BIG_ENDIAN
1378 		/*
1379 		 * The bind(), connect(), and sendto() syscalls were not
1380 		 * versioned for COMPAT_43.  Thus, this check must stay.
1381 		 */
1382 		if (sa->sa_family == 0 && sa->sa_len < AF_MAX)
1383 			sa->sa_family = sa->sa_len;
1384 #endif
1385 		sa->sa_len = len;
1386 		*namp = sa;
1387 	}
1388 	return error;
1389 }
1390 
1391 /*
1392  * Detach a mapped page and release resources back to the system.
1393  * We must release our wiring and if the object is ripped out
1394  * from under the vm_page we become responsible for freeing the
1395  * page.  These routines must be MPSAFE.
1396  *
1397  * XXX HACK XXX TEMPORARY UNTIL WE IMPLEMENT EXT MBUF REFERENCE COUNTING
1398  *
1399  * XXX vm_page_*() routines are not MPSAFE yet, the MP lock is required.
1400  */
1401 static void
1402 sf_buf_mfree(void *arg)
1403 {
1404 	struct sf_buf *sf = arg;
1405 	vm_page_t m;
1406 
1407 	/*
1408 	 * XXX vm_page_*() and SFBUF routines not MPSAFE yet.
1409 	 */
1410 	get_mplock();
1411 	crit_enter();
1412 	m = sf_buf_page(sf);
1413 	if (sf_buf_free(sf) == 0) {
1414 		vm_page_unwire(m, 0);
1415 		if (m->wire_count == 0 && m->object == NULL)
1416 			vm_page_try_to_free(m);
1417 	}
1418 	crit_exit();
1419 	rel_mplock();
1420 }
1421 
1422 /*
1423  * sendfile(2).
1424  * int sendfile(int fd, int s, off_t offset, size_t nbytes,
1425  *	 struct sf_hdtr *hdtr, off_t *sbytes, int flags)
1426  *
1427  * Send a file specified by 'fd' and starting at 'offset' to a socket
1428  * specified by 's'. Send only 'nbytes' of the file or until EOF if
1429  * nbytes == 0. Optionally add a header and/or trailer to the socket
1430  * output. If specified, write the total number of bytes sent into *sbytes.
1431  *
1432  * In FreeBSD kern/uipc_syscalls.c,v 1.103, a bug was fixed that caused
1433  * the headers to count against the remaining bytes to be sent from
1434  * the file descriptor.  We may wish to implement a compatibility syscall
1435  * in the future.
1436  *
1437  * MPALMOSTSAFE
1438  */
1439 int
1440 sys_sendfile(struct sendfile_args *uap)
1441 {
1442 	struct thread *td = curthread;
1443 	struct proc *p = td->td_proc;
1444 	struct file *fp;
1445 	struct vnode *vp = NULL;
1446 	struct sf_hdtr hdtr;
1447 	struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
1448 	struct uio auio;
1449 	struct mbuf *mheader = NULL;
1450 	size_t hbytes = 0;
1451 	size_t tbytes;
1452 	off_t hdtr_size = 0;
1453 	off_t sbytes;
1454 	int error;
1455 
1456 	KKASSERT(p);
1457 
1458 	/*
1459 	 * Do argument checking. Must be a regular file in, stream
1460 	 * type and connected socket out, positive offset.
1461 	 */
1462 	fp = holdfp(p->p_fd, uap->fd, FREAD);
1463 	if (fp == NULL) {
1464 		return (EBADF);
1465 	}
1466 	if (fp->f_type != DTYPE_VNODE) {
1467 		fdrop(fp);
1468 		return (EINVAL);
1469 	}
1470 	get_mplock();
1471 	vp = (struct vnode *)fp->f_data;
1472 	vref(vp);
1473 	fdrop(fp);
1474 
1475 	/*
1476 	 * If specified, get the pointer to the sf_hdtr struct for
1477 	 * any headers/trailers.
1478 	 */
1479 	if (uap->hdtr) {
1480 		error = copyin(uap->hdtr, &hdtr, sizeof(hdtr));
1481 		if (error)
1482 			goto done;
1483 		/*
1484 		 * Send any headers.
1485 		 */
1486 		if (hdtr.headers) {
1487 			error = iovec_copyin(hdtr.headers, &iov, aiov,
1488 					     hdtr.hdr_cnt, &hbytes);
1489 			if (error)
1490 				goto done;
1491 			auio.uio_iov = iov;
1492 			auio.uio_iovcnt = hdtr.hdr_cnt;
1493 			auio.uio_offset = 0;
1494 			auio.uio_segflg = UIO_USERSPACE;
1495 			auio.uio_rw = UIO_WRITE;
1496 			auio.uio_td = td;
1497 			auio.uio_resid = hbytes;
1498 
1499 			mheader = m_uiomove(&auio);
1500 
1501 			iovec_free(&iov, aiov);
1502 			if (mheader == NULL)
1503 				goto done;
1504 		}
1505 	}
1506 
1507 	error = kern_sendfile(vp, uap->s, uap->offset, uap->nbytes, mheader,
1508 			      &sbytes, uap->flags);
1509 	if (error)
1510 		goto done;
1511 
1512 	/*
1513 	 * Send trailers. Wimp out and use writev(2).
1514 	 */
1515 	if (uap->hdtr != NULL && hdtr.trailers != NULL) {
1516 		error = iovec_copyin(hdtr.trailers, &iov, aiov,
1517 				     hdtr.trl_cnt, &auio.uio_resid);
1518 		if (error)
1519 			goto done;
1520 		auio.uio_iov = iov;
1521 		auio.uio_iovcnt = hdtr.trl_cnt;
1522 		auio.uio_offset = 0;
1523 		auio.uio_segflg = UIO_USERSPACE;
1524 		auio.uio_rw = UIO_WRITE;
1525 		auio.uio_td = td;
1526 
1527 		error = kern_sendmsg(uap->s, NULL, &auio, NULL, 0, &tbytes);
1528 
1529 		iovec_free(&iov, aiov);
1530 		if (error)
1531 			goto done;
1532 		hdtr_size += tbytes;	/* trailer bytes successfully sent */
1533 	}
1534 
1535 done:
1536 	if (vp)
1537 		vrele(vp);
1538 	rel_mplock();
1539 	if (uap->sbytes != NULL) {
1540 		sbytes += hdtr_size;
1541 		copyout(&sbytes, uap->sbytes, sizeof(off_t));
1542 	}
1543 	return (error);
1544 }
1545 
1546 int
1547 kern_sendfile(struct vnode *vp, int sfd, off_t offset, size_t nbytes,
1548     struct mbuf *mheader, off_t *sbytes, int flags)
1549 {
1550 	struct thread *td = curthread;
1551 	struct proc *p = td->td_proc;
1552 	struct vm_object *obj;
1553 	struct socket *so;
1554 	struct file *fp;
1555 	struct mbuf *m;
1556 	struct sf_buf *sf;
1557 	struct vm_page *pg;
1558 	off_t off, xfsize;
1559 	off_t hbytes = 0;
1560 	int error = 0;
1561 
1562 	if (vp->v_type != VREG) {
1563 		error = EINVAL;
1564 		goto done0;
1565 	}
1566 	if ((obj = vp->v_object) == NULL) {
1567 		error = EINVAL;
1568 		goto done0;
1569 	}
1570 	error = holdsock(p->p_fd, sfd, &fp);
1571 	if (error)
1572 		goto done0;
1573 	so = (struct socket *)fp->f_data;
1574 	if (so->so_type != SOCK_STREAM) {
1575 		error = EINVAL;
1576 		goto done;
1577 	}
1578 	if ((so->so_state & SS_ISCONNECTED) == 0) {
1579 		error = ENOTCONN;
1580 		goto done;
1581 	}
1582 	if (offset < 0) {
1583 		error = EINVAL;
1584 		goto done;
1585 	}
1586 
1587 	*sbytes = 0;
1588 	/*
1589 	 * Protect against multiple writers to the socket.
1590 	 */
1591 	ssb_lock(&so->so_snd, M_WAITOK);
1592 
1593 	/*
1594 	 * Loop through the pages in the file, starting with the requested
1595 	 * offset. Get a file page (do I/O if necessary), map the file page
1596 	 * into an sf_buf, attach an mbuf header to the sf_buf, and queue
1597 	 * it on the socket.
1598 	 */
1599 	for (off = offset; ; off += xfsize, *sbytes += xfsize + hbytes) {
1600 		vm_pindex_t pindex;
1601 		vm_offset_t pgoff;
1602 
1603 		pindex = OFF_TO_IDX(off);
1604 retry_lookup:
1605 		/*
1606 		 * Calculate the amount to transfer. Not to exceed a page,
1607 		 * the EOF, or the passed in nbytes.
1608 		 */
1609 		xfsize = vp->v_filesize - off;
1610 		if (xfsize > PAGE_SIZE)
1611 			xfsize = PAGE_SIZE;
1612 		pgoff = (vm_offset_t)(off & PAGE_MASK);
1613 		if (PAGE_SIZE - pgoff < xfsize)
1614 			xfsize = PAGE_SIZE - pgoff;
1615 		if (nbytes && xfsize > (nbytes - *sbytes))
1616 			xfsize = nbytes - *sbytes;
1617 		if (xfsize <= 0)
1618 			break;
1619 		/*
1620 		 * Optimize the non-blocking case by looking at the socket space
1621 		 * before going to the extra work of constituting the sf_buf.
1622 		 */
1623 		if ((fp->f_flag & FNONBLOCK) && ssb_space(&so->so_snd) <= 0) {
1624 			if (so->so_state & SS_CANTSENDMORE)
1625 				error = EPIPE;
1626 			else
1627 				error = EAGAIN;
1628 			ssb_unlock(&so->so_snd);
1629 			goto done;
1630 		}
1631 		/*
1632 		 * Attempt to look up the page.
1633 		 *
1634 		 *	Allocate if not found, wait and loop if busy, then
1635 		 *	wire the page.  critical section protection is
1636 		 * 	required to maintain the object association (an
1637 		 *	interrupt can free the page) through to the
1638 		 *	vm_page_wire() call.
1639 		 */
1640 		crit_enter();
1641 		lwkt_gettoken(&vm_token);
1642 		pg = vm_page_lookup(obj, pindex);
1643 		if (pg == NULL) {
1644 			pg = vm_page_alloc(obj, pindex, VM_ALLOC_NORMAL);
1645 			if (pg == NULL) {
1646 				vm_wait(0);
1647 				lwkt_reltoken(&vm_token);
1648 				crit_exit();
1649 				goto retry_lookup;
1650 			}
1651 			vm_page_wakeup(pg);
1652 		} else if (vm_page_sleep_busy(pg, TRUE, "sfpbsy")) {
1653 			lwkt_reltoken(&vm_token);
1654 			crit_exit();
1655 			goto retry_lookup;
1656 		}
1657 		vm_page_wire(pg);
1658 		lwkt_reltoken(&vm_token);
1659 		crit_exit();
1660 
1661 		/*
1662 		 * If page is not valid for what we need, initiate I/O
1663 		 */
1664 
1665 		if (!pg->valid || !vm_page_is_valid(pg, pgoff, xfsize)) {
1666 			struct uio auio;
1667 			struct iovec aiov;
1668 			int bsize;
1669 
1670 			/*
1671 			 * Ensure that our page is still around when the I/O
1672 			 * completes.
1673 			 */
1674 			vm_page_io_start(pg);
1675 
1676 			/*
1677 			 * Get the page from backing store.
1678 			 */
1679 			bsize = vp->v_mount->mnt_stat.f_iosize;
1680 			auio.uio_iov = &aiov;
1681 			auio.uio_iovcnt = 1;
1682 			aiov.iov_base = 0;
1683 			aiov.iov_len = MAXBSIZE;
1684 			auio.uio_resid = MAXBSIZE;
1685 			auio.uio_offset = trunc_page(off);
1686 			auio.uio_segflg = UIO_NOCOPY;
1687 			auio.uio_rw = UIO_READ;
1688 			auio.uio_td = td;
1689 			vn_lock(vp, LK_SHARED | LK_RETRY);
1690 			error = VOP_READ(vp, &auio,
1691 				    IO_VMIO | ((MAXBSIZE / bsize) << 16),
1692 				    td->td_ucred);
1693 			vn_unlock(vp);
1694 			vm_page_flag_clear(pg, PG_ZERO);
1695 			vm_page_io_finish(pg);
1696 			if (error) {
1697 				crit_enter();
1698 				vm_page_unwire(pg, 0);
1699 				vm_page_try_to_free(pg);
1700 				crit_exit();
1701 				ssb_unlock(&so->so_snd);
1702 				goto done;
1703 			}
1704 		}
1705 
1706 
1707 		/*
1708 		 * Get a sendfile buf. We usually wait as long as necessary,
1709 		 * but this wait can be interrupted.
1710 		 */
1711 		if ((sf = sf_buf_alloc(pg)) == NULL) {
1712 			crit_enter();
1713 			vm_page_unwire(pg, 0);
1714 			vm_page_try_to_free(pg);
1715 			crit_exit();
1716 			ssb_unlock(&so->so_snd);
1717 			error = EINTR;
1718 			goto done;
1719 		}
1720 
1721 		/*
1722 		 * Get an mbuf header and set it up as having external storage.
1723 		 */
1724 		MGETHDR(m, MB_WAIT, MT_DATA);
1725 		if (m == NULL) {
1726 			error = ENOBUFS;
1727 			sf_buf_free(sf);
1728 			ssb_unlock(&so->so_snd);
1729 			goto done;
1730 		}
1731 
1732 		m->m_ext.ext_free = sf_buf_mfree;
1733 		m->m_ext.ext_ref = sf_buf_ref;
1734 		m->m_ext.ext_arg = sf;
1735 		m->m_ext.ext_buf = (void *)sf_buf_kva(sf);
1736 		m->m_ext.ext_size = PAGE_SIZE;
1737 		m->m_data = (char *)sf_buf_kva(sf) + pgoff;
1738 		m->m_flags |= M_EXT;
1739 		m->m_pkthdr.len = m->m_len = xfsize;
1740 		KKASSERT((m->m_flags & (M_EXT_CLUSTER)) == 0);
1741 
1742 		if (mheader != NULL) {
1743 			hbytes = mheader->m_pkthdr.len;
1744 			mheader->m_pkthdr.len += m->m_pkthdr.len;
1745 			m_cat(mheader, m);
1746 			m = mheader;
1747 			mheader = NULL;
1748 		} else
1749 			hbytes = 0;
1750 
1751 		/*
1752 		 * Add the buffer to the socket buffer chain.
1753 		 */
1754 		crit_enter();
1755 retry_space:
1756 		/*
1757 		 * Make sure that the socket is still able to take more data.
1758 		 * CANTSENDMORE being true usually means that the connection
1759 		 * was closed. so_error is true when an error was sensed after
1760 		 * a previous send.
1761 		 * The state is checked after the page mapping and buffer
1762 		 * allocation above since those operations may block and make
1763 		 * any socket checks stale. From this point forward, nothing
1764 		 * blocks before the pru_send (or more accurately, any blocking
1765 		 * results in a loop back to here to re-check).
1766 		 */
1767 		if ((so->so_state & SS_CANTSENDMORE) || so->so_error) {
1768 			if (so->so_state & SS_CANTSENDMORE) {
1769 				error = EPIPE;
1770 			} else {
1771 				error = so->so_error;
1772 				so->so_error = 0;
1773 			}
1774 			m_freem(m);
1775 			ssb_unlock(&so->so_snd);
1776 			crit_exit();
1777 			goto done;
1778 		}
1779 		/*
1780 		 * Wait for socket space to become available. We do this just
1781 		 * after checking the connection state above in order to avoid
1782 		 * a race condition with ssb_wait().
1783 		 */
1784 		if (ssb_space(&so->so_snd) < so->so_snd.ssb_lowat) {
1785 			if (fp->f_flag & FNONBLOCK) {
1786 				m_freem(m);
1787 				ssb_unlock(&so->so_snd);
1788 				crit_exit();
1789 				error = EAGAIN;
1790 				goto done;
1791 			}
1792 			error = ssb_wait(&so->so_snd);
1793 			/*
1794 			 * An error from ssb_wait usually indicates that we've
1795 			 * been interrupted by a signal. If we've sent anything
1796 			 * then return bytes sent, otherwise return the error.
1797 			 */
1798 			if (error) {
1799 				m_freem(m);
1800 				ssb_unlock(&so->so_snd);
1801 				crit_exit();
1802 				goto done;
1803 			}
1804 			goto retry_space;
1805 		}
1806 		error = so_pru_send(so, 0, m, NULL, NULL, td);
1807 		crit_exit();
1808 		if (error) {
1809 			ssb_unlock(&so->so_snd);
1810 			goto done;
1811 		}
1812 	}
1813 	if (mheader != NULL) {
1814 		*sbytes += mheader->m_pkthdr.len;
1815 		error = so_pru_send(so, 0, mheader, NULL, NULL, td);
1816 		mheader = NULL;
1817 	}
1818 	ssb_unlock(&so->so_snd);
1819 
1820 done:
1821 	fdrop(fp);
1822 done0:
1823 	if (mheader != NULL)
1824 		m_freem(mheader);
1825 	return (error);
1826 }
1827 
1828 /*
1829  * MPALMOSTSAFE
1830  */
1831 int
1832 sys_sctp_peeloff(struct sctp_peeloff_args *uap)
1833 {
1834 #ifdef SCTP
1835 	struct thread *td = curthread;
1836 	struct filedesc *fdp = td->td_proc->p_fd;
1837 	struct file *lfp = NULL;
1838 	struct file *nfp = NULL;
1839 	int error;
1840 	struct socket *head, *so;
1841 	caddr_t assoc_id;
1842 	int fd;
1843 	short fflag;		/* type must match fp->f_flag */
1844 
1845 	assoc_id = uap->name;
1846 	error = holdsock(td->td_proc->p_fd, uap->sd, &lfp);
1847 	if (error)
1848 		return (error);
1849 
1850 	get_mplock();
1851 	crit_enter();
1852 	head = (struct socket *)lfp->f_data;
1853 	error = sctp_can_peel_off(head, assoc_id);
1854 	if (error) {
1855 		crit_exit();
1856 		goto done;
1857 	}
1858 	/*
1859 	 * At this point we know we do have a assoc to pull
1860 	 * we proceed to get the fd setup. This may block
1861 	 * but that is ok.
1862 	 */
1863 
1864 	fflag = lfp->f_flag;
1865 	error = falloc(td->td_lwp, &nfp, &fd);
1866 	if (error) {
1867 		/*
1868 		 * Probably ran out of file descriptors. Put the
1869 		 * unaccepted connection back onto the queue and
1870 		 * do another wakeup so some other process might
1871 		 * have a chance at it.
1872 		 */
1873 		crit_exit();
1874 		goto done;
1875 	}
1876 	uap->sysmsg_iresult = fd;
1877 
1878 	so = sctp_get_peeloff(head, assoc_id, &error);
1879 	if (so == NULL) {
1880 		/*
1881 		 * Either someone else peeled it off OR
1882 		 * we can't get a socket.
1883 		 */
1884 		goto noconnection;
1885 	}
1886 	soreference(so);			/* reference needed */
1887 	soclrstate(so, SS_NOFDREF | SS_COMP);	/* when clearing NOFDREF */
1888 	so->so_head = NULL;
1889 	if (head->so_sigio != NULL)
1890 		fsetown(fgetown(head->so_sigio), &so->so_sigio);
1891 
1892 	nfp->f_type = DTYPE_SOCKET;
1893 	nfp->f_flag = fflag;
1894 	nfp->f_ops = &socketops;
1895 	nfp->f_data = so;
1896 
1897 noconnection:
1898 	/*
1899 	 * Assign the file pointer to the reserved descriptor, or clear
1900 	 * the reserved descriptor if an error occured.
1901 	 */
1902 	if (error)
1903 		fsetfd(fdp, NULL, fd);
1904 	else
1905 		fsetfd(fdp, nfp, fd);
1906 	crit_exit();
1907 	/*
1908 	 * Release explicitly held references before returning.
1909 	 */
1910 done:
1911 	rel_mplock();
1912 	if (nfp != NULL)
1913 		fdrop(nfp);
1914 	fdrop(lfp);
1915 	return (error);
1916 #else /* SCTP */
1917 	return(EOPNOTSUPP);
1918 #endif /* SCTP */
1919 }
1920