xref: /freebsd/sys/kern/uipc_syscalls.c (revision c4f6a2a9e1b1879b618c436ab4f56ff75c73a0f5)
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  * $FreeBSD$
38  */
39 
40 #include "opt_compat.h"
41 #include "opt_ktrace.h"
42 #include "opt_mac.h"
43 
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/kernel.h>
47 #include <sys/lock.h>
48 #include <sys/mac.h>
49 #include <sys/mutex.h>
50 #include <sys/sysproto.h>
51 #include <sys/malloc.h>
52 #include <sys/filedesc.h>
53 #include <sys/event.h>
54 #include <sys/proc.h>
55 #include <sys/fcntl.h>
56 #include <sys/file.h>
57 #include <sys/lock.h>
58 #include <sys/mount.h>
59 #include <sys/mbuf.h>
60 #include <sys/protosw.h>
61 #include <sys/socket.h>
62 #include <sys/socketvar.h>
63 #include <sys/signalvar.h>
64 #include <sys/uio.h>
65 #include <sys/vnode.h>
66 #ifdef KTRACE
67 #include <sys/ktrace.h>
68 #endif
69 
70 #include <vm/vm.h>
71 #include <vm/vm_object.h>
72 #include <vm/vm_page.h>
73 #include <vm/vm_pageout.h>
74 #include <vm/vm_kern.h>
75 #include <vm/vm_extern.h>
76 
77 static void sf_buf_init(void *arg);
78 SYSINIT(sock_sf, SI_SUB_MBUF, SI_ORDER_ANY, sf_buf_init, NULL)
79 
80 static int sendit(struct thread *td, int s, struct msghdr *mp, int flags);
81 static int recvit(struct thread *td, int s, struct msghdr *mp, void *namelenp);
82 
83 static int accept1(struct thread *td, struct accept_args *uap, int compat);
84 static int do_sendfile(struct thread *td, struct sendfile_args *uap, int compat);
85 static int getsockname1(struct thread *td, struct getsockname_args *uap,
86 			int compat);
87 static int getpeername1(struct thread *td, struct getpeername_args *uap,
88 			int compat);
89 
90 /*
91  * Expanded sf_freelist head. Really an SLIST_HEAD() in disguise, with the
92  * sf_freelist head with the sf_lock mutex.
93  */
94 static struct {
95 	SLIST_HEAD(, sf_buf) sf_head;
96 	struct mtx sf_lock;
97 } sf_freelist;
98 
99 vm_offset_t sf_base;
100 struct sf_buf *sf_bufs;
101 u_int sf_buf_alloc_want;
102 
103 /*
104  * System call interface to the socket abstraction.
105  */
106 #if defined(COMPAT_43) || defined(COMPAT_SUNOS)
107 #define COMPAT_OLDSOCK
108 #endif
109 
110 extern	struct fileops socketops;
111 
112 /*
113  * MPSAFE
114  */
115 int
116 socket(td, uap)
117 	struct thread *td;
118 	register struct socket_args /* {
119 		int	domain;
120 		int	type;
121 		int	protocol;
122 	} */ *uap;
123 {
124 	struct filedesc *fdp;
125 	struct socket *so;
126 	struct file *fp;
127 	int fd, error;
128 
129 	mtx_lock(&Giant);
130 	fdp = td->td_proc->p_fd;
131 	error = falloc(td, &fp, &fd);
132 	if (error)
133 		goto done2;
134 	fhold(fp);
135 	error = socreate(uap->domain, &so, uap->type, uap->protocol,
136 	    td->td_ucred, td);
137 	FILEDESC_LOCK(fdp);
138 	if (error) {
139 		if (fdp->fd_ofiles[fd] == fp) {
140 			fdp->fd_ofiles[fd] = NULL;
141 			FILEDESC_UNLOCK(fdp);
142 			fdrop(fp, td);
143 		} else
144 			FILEDESC_UNLOCK(fdp);
145 	} else {
146 		fp->f_data = so;	/* already has ref count */
147 		fp->f_flag = FREAD|FWRITE;
148 		fp->f_ops = &socketops;
149 		fp->f_type = DTYPE_SOCKET;
150 		FILEDESC_UNLOCK(fdp);
151 		td->td_retval[0] = fd;
152 	}
153 	fdrop(fp, td);
154 done2:
155 	mtx_unlock(&Giant);
156 	return (error);
157 }
158 
159 /*
160  * MPSAFE
161  */
162 /* ARGSUSED */
163 int
164 bind(td, uap)
165 	struct thread *td;
166 	register struct bind_args /* {
167 		int	s;
168 		caddr_t	name;
169 		int	namelen;
170 	} */ *uap;
171 {
172 	struct socket *so;
173 	struct sockaddr *sa;
174 	int error;
175 
176 	mtx_lock(&Giant);
177 	if ((error = fgetsock(td, uap->s, &so, NULL)) != 0)
178 		goto done2;
179 	if ((error = getsockaddr(&sa, uap->name, uap->namelen)) != 0)
180 		goto done1;
181 #ifdef MAC
182 	error = mac_check_socket_bind(td->td_ucred, so, sa);
183 	if (error) {
184 		FREE(sa, M_SONAME);
185 		goto done1;
186 	}
187 #endif
188 	error = sobind(so, sa, td);
189 	FREE(sa, M_SONAME);
190 done1:
191 	fputsock(so);
192 done2:
193 	mtx_unlock(&Giant);
194 	return (error);
195 }
196 
197 /*
198  * MPSAFE
199  */
200 /* ARGSUSED */
201 int
202 listen(td, uap)
203 	struct thread *td;
204 	register struct listen_args /* {
205 		int	s;
206 		int	backlog;
207 	} */ *uap;
208 {
209 	struct socket *so;
210 	int error;
211 
212 	mtx_lock(&Giant);
213 	if ((error = fgetsock(td, uap->s, &so, NULL)) == 0) {
214 #ifdef MAC
215 		error = mac_check_socket_listen(td->td_ucred, so);
216 		if (error)
217 			goto done;
218 #endif
219 		error = solisten(so, uap->backlog, td);
220 #ifdef MAC
221 done:
222 #endif
223 		fputsock(so);
224 	}
225 	mtx_unlock(&Giant);
226 	return(error);
227 }
228 
229 /*
230  * accept1()
231  * MPSAFE
232  */
233 static int
234 accept1(td, uap, compat)
235 	struct thread *td;
236 	register struct accept_args /* {
237 		int	s;
238 		caddr_t	name;
239 		int	*anamelen;
240 	} */ *uap;
241 	int compat;
242 {
243 	struct filedesc *fdp;
244 	struct file *nfp = NULL;
245 	struct sockaddr *sa;
246 	int namelen, error, s;
247 	struct socket *head, *so;
248 	int fd;
249 	u_int fflag;
250 
251 	mtx_lock(&Giant);
252 	fdp = td->td_proc->p_fd;
253 	if (uap->name) {
254 		error = copyin(uap->anamelen, &namelen, sizeof (namelen));
255 		if(error)
256 			goto done2;
257 		if (namelen < 0) {
258 			error = EINVAL;
259 			goto done2;
260 		}
261 	}
262 	error = fgetsock(td, uap->s, &head, &fflag);
263 	if (error)
264 		goto done2;
265 	s = splnet();
266 	if ((head->so_options & SO_ACCEPTCONN) == 0) {
267 		splx(s);
268 		error = EINVAL;
269 		goto done;
270 	}
271 	if ((head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->so_comp)) {
272 		splx(s);
273 		error = EWOULDBLOCK;
274 		goto done;
275 	}
276 	while (TAILQ_EMPTY(&head->so_comp) && head->so_error == 0) {
277 		if (head->so_state & SS_CANTRCVMORE) {
278 			head->so_error = ECONNABORTED;
279 			break;
280 		}
281 		error = tsleep(&head->so_timeo, PSOCK | PCATCH,
282 		    "accept", 0);
283 		if (error) {
284 			splx(s);
285 			goto done;
286 		}
287 	}
288 	if (head->so_error) {
289 		error = head->so_error;
290 		head->so_error = 0;
291 		splx(s);
292 		goto done;
293 	}
294 
295 	/*
296 	 * At this point we know that there is at least one connection
297 	 * ready to be accepted. Remove it from the queue prior to
298 	 * allocating the file descriptor for it since falloc() may
299 	 * block allowing another process to accept the connection
300 	 * instead.
301 	 */
302 	so = TAILQ_FIRST(&head->so_comp);
303 	TAILQ_REMOVE(&head->so_comp, so, so_list);
304 	head->so_qlen--;
305 
306 	error = falloc(td, &nfp, &fd);
307 	if (error) {
308 		/*
309 		 * Probably ran out of file descriptors. Put the
310 		 * unaccepted connection back onto the queue and
311 		 * do another wakeup so some other process might
312 		 * have a chance at it.
313 		 */
314 		TAILQ_INSERT_HEAD(&head->so_comp, so, so_list);
315 		head->so_qlen++;
316 		wakeup_one(&head->so_timeo);
317 		splx(s);
318 		goto done;
319 	}
320 	fhold(nfp);
321 	td->td_retval[0] = fd;
322 
323 	/* connection has been removed from the listen queue */
324 	KNOTE(&head->so_rcv.sb_sel.si_note, 0);
325 
326 	so->so_state &= ~SS_COMP;
327 	so->so_head = NULL;
328 	if (head->so_sigio != NULL)
329 		fsetown(fgetown(head->so_sigio), &so->so_sigio);
330 
331 	FILE_LOCK(nfp);
332 	soref(so);			/* file descriptor reference */
333 	nfp->f_data = so;		/* nfp has ref count from falloc */
334 	nfp->f_flag = fflag;
335 	nfp->f_ops = &socketops;
336 	nfp->f_type = DTYPE_SOCKET;
337 	FILE_UNLOCK(nfp);
338 	sa = 0;
339 	error = soaccept(so, &sa);
340 	if (error) {
341 		/*
342 		 * return a namelen of zero for older code which might
343 	 	 * ignore the return value from accept.
344 		 */
345 		if (uap->name != NULL) {
346 			namelen = 0;
347 			(void) copyout(&namelen,
348 			    uap->anamelen, sizeof(*uap->anamelen));
349 		}
350 		goto noconnection;
351 	}
352 	if (sa == NULL) {
353 		namelen = 0;
354 		if (uap->name)
355 			goto gotnoname;
356 		splx(s);
357 		error = 0;
358 		goto done;
359 	}
360 	if (uap->name) {
361 		/* check sa_len before it is destroyed */
362 		if (namelen > sa->sa_len)
363 			namelen = sa->sa_len;
364 #ifdef COMPAT_OLDSOCK
365 		if (compat)
366 			((struct osockaddr *)sa)->sa_family =
367 			    sa->sa_family;
368 #endif
369 		error = copyout(sa, uap->name, (u_int)namelen);
370 		if (!error)
371 gotnoname:
372 			error = copyout(&namelen,
373 			    uap->anamelen, sizeof (*uap->anamelen));
374 	}
375 noconnection:
376 	if (sa)
377 		FREE(sa, M_SONAME);
378 
379 	/*
380 	 * close the new descriptor, assuming someone hasn't ripped it
381 	 * out from under us.
382 	 */
383 	if (error) {
384 		FILEDESC_LOCK(fdp);
385 		if (fdp->fd_ofiles[fd] == nfp) {
386 			fdp->fd_ofiles[fd] = NULL;
387 			FILEDESC_UNLOCK(fdp);
388 			fdrop(nfp, td);
389 		} else {
390 			FILEDESC_UNLOCK(fdp);
391 		}
392 	}
393 	splx(s);
394 
395 	/*
396 	 * Release explicitly held references before returning.
397 	 */
398 done:
399 	if (nfp != NULL)
400 		fdrop(nfp, td);
401 	fputsock(head);
402 done2:
403 	mtx_unlock(&Giant);
404 	return (error);
405 }
406 
407 /*
408  * MPSAFE (accept1() is MPSAFE)
409  */
410 int
411 accept(td, uap)
412 	struct thread *td;
413 	struct accept_args *uap;
414 {
415 
416 	return (accept1(td, uap, 0));
417 }
418 
419 #ifdef COMPAT_OLDSOCK
420 /*
421  * MPSAFE (accept1() is MPSAFE)
422  */
423 int
424 oaccept(td, uap)
425 	struct thread *td;
426 	struct accept_args *uap;
427 {
428 
429 	return (accept1(td, uap, 1));
430 }
431 #endif /* COMPAT_OLDSOCK */
432 
433 /*
434  * MPSAFE
435  */
436 /* ARGSUSED */
437 int
438 connect(td, uap)
439 	struct thread *td;
440 	register struct connect_args /* {
441 		int	s;
442 		caddr_t	name;
443 		int	namelen;
444 	} */ *uap;
445 {
446 	struct socket *so;
447 	struct sockaddr *sa;
448 	int error, s;
449 
450 	mtx_lock(&Giant);
451 	if ((error = fgetsock(td, uap->s, &so, NULL)) != 0)
452 		goto done2;
453 	if ((so->so_state & SS_NBIO) && (so->so_state & SS_ISCONNECTING)) {
454 		error = EALREADY;
455 		goto done1;
456 	}
457 	error = getsockaddr(&sa, uap->name, uap->namelen);
458 	if (error)
459 		goto done1;
460 #ifdef MAC
461 	error = mac_check_socket_connect(td->td_ucred, so, sa);
462 	if (error)
463 		goto bad;
464 #endif
465 	error = soconnect(so, sa, td);
466 	if (error)
467 		goto bad;
468 	if ((so->so_state & SS_NBIO) && (so->so_state & SS_ISCONNECTING)) {
469 		FREE(sa, M_SONAME);
470 		error = EINPROGRESS;
471 		goto done1;
472 	}
473 	s = splnet();
474 	while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) {
475 		error = tsleep(&so->so_timeo, PSOCK | PCATCH, "connec", 0);
476 		if (error)
477 			break;
478 	}
479 	if (error == 0) {
480 		error = so->so_error;
481 		so->so_error = 0;
482 	}
483 	splx(s);
484 bad:
485 	so->so_state &= ~SS_ISCONNECTING;
486 	FREE(sa, M_SONAME);
487 	if (error == ERESTART)
488 		error = EINTR;
489 done1:
490 	fputsock(so);
491 done2:
492 	mtx_unlock(&Giant);
493 	return (error);
494 }
495 
496 /*
497  * MPSAFE
498  */
499 int
500 socketpair(td, uap)
501 	struct thread *td;
502 	register struct socketpair_args /* {
503 		int	domain;
504 		int	type;
505 		int	protocol;
506 		int	*rsv;
507 	} */ *uap;
508 {
509 	register struct filedesc *fdp = td->td_proc->p_fd;
510 	struct file *fp1, *fp2;
511 	struct socket *so1, *so2;
512 	int fd, error, sv[2];
513 
514 	mtx_lock(&Giant);
515 	error = socreate(uap->domain, &so1, uap->type, uap->protocol,
516 	    td->td_ucred, td);
517 	if (error)
518 		goto done2;
519 	error = socreate(uap->domain, &so2, uap->type, uap->protocol,
520 	    td->td_ucred, td);
521 	if (error)
522 		goto free1;
523 	error = falloc(td, &fp1, &fd);
524 	if (error)
525 		goto free2;
526 	fhold(fp1);
527 	sv[0] = fd;
528 	fp1->f_data = so1;		/* so1 already has ref count */
529 	error = falloc(td, &fp2, &fd);
530 	if (error)
531 		goto free3;
532 	fhold(fp2);
533 	fp2->f_data = so2;		/* so2 already has ref count */
534 	sv[1] = fd;
535 	error = soconnect2(so1, so2);
536 	if (error)
537 		goto free4;
538 	if (uap->type == SOCK_DGRAM) {
539 		/*
540 		 * Datagram socket connection is asymmetric.
541 		 */
542 		 error = soconnect2(so2, so1);
543 		 if (error)
544 			goto free4;
545 	}
546 	FILE_LOCK(fp1);
547 	fp1->f_flag = FREAD|FWRITE;
548 	fp1->f_ops = &socketops;
549 	fp1->f_type = DTYPE_SOCKET;
550 	FILE_UNLOCK(fp1);
551 	FILE_LOCK(fp2);
552 	fp2->f_flag = FREAD|FWRITE;
553 	fp2->f_ops = &socketops;
554 	fp2->f_type = DTYPE_SOCKET;
555 	FILE_UNLOCK(fp2);
556 	error = copyout(sv, uap->rsv, 2 * sizeof (int));
557 	fdrop(fp1, td);
558 	fdrop(fp2, td);
559 	goto done2;
560 free4:
561 	FILEDESC_LOCK(fdp);
562 	if (fdp->fd_ofiles[sv[1]] == fp2) {
563 		fdp->fd_ofiles[sv[1]] = NULL;
564 		FILEDESC_UNLOCK(fdp);
565 		fdrop(fp2, td);
566 	} else
567 		FILEDESC_UNLOCK(fdp);
568 	fdrop(fp2, td);
569 free3:
570 	FILEDESC_LOCK(fdp);
571 	if (fdp->fd_ofiles[sv[0]] == fp1) {
572 		fdp->fd_ofiles[sv[0]] = NULL;
573 		FILEDESC_UNLOCK(fdp);
574 		fdrop(fp1, td);
575 	} else
576 		FILEDESC_UNLOCK(fdp);
577 	fdrop(fp1, td);
578 free2:
579 	(void)soclose(so2);
580 free1:
581 	(void)soclose(so1);
582 done2:
583 	mtx_unlock(&Giant);
584 	return (error);
585 }
586 
587 static int
588 sendit(td, s, mp, flags)
589 	register struct thread *td;
590 	int s;
591 	register struct msghdr *mp;
592 	int flags;
593 {
594 	struct uio auio;
595 	register struct iovec *iov;
596 	register int i;
597 	struct mbuf *control;
598 	struct sockaddr *to = NULL;
599 	int len, error;
600 	struct socket *so;
601 #ifdef KTRACE
602 	struct iovec *ktriov = NULL;
603 	struct uio ktruio;
604 	int iovlen;
605 #endif
606 
607 	if ((error = fgetsock(td, s, &so, NULL)) != 0)
608 		return (error);
609 	auio.uio_iov = mp->msg_iov;
610 	auio.uio_iovcnt = mp->msg_iovlen;
611 	auio.uio_segflg = UIO_USERSPACE;
612 	auio.uio_rw = UIO_WRITE;
613 	auio.uio_td = td;
614 	auio.uio_offset = 0;			/* XXX */
615 	auio.uio_resid = 0;
616 	iov = mp->msg_iov;
617 	for (i = 0; i < mp->msg_iovlen; i++, iov++) {
618 		if ((auio.uio_resid += iov->iov_len) < 0) {
619 			error = EINVAL;
620 			goto bad;
621 		}
622 	}
623 	if (mp->msg_name) {
624 		error = getsockaddr(&to, mp->msg_name, mp->msg_namelen);
625 		if (error)
626 			goto bad;
627 	}
628 	if (mp->msg_control) {
629 		if (mp->msg_controllen < sizeof(struct cmsghdr)
630 #ifdef COMPAT_OLDSOCK
631 		    && mp->msg_flags != MSG_COMPAT
632 #endif
633 		) {
634 			error = EINVAL;
635 			goto bad;
636 		}
637 		error = sockargs(&control, mp->msg_control,
638 		    mp->msg_controllen, MT_CONTROL);
639 		if (error)
640 			goto bad;
641 #ifdef COMPAT_OLDSOCK
642 		if (mp->msg_flags == MSG_COMPAT) {
643 			register struct cmsghdr *cm;
644 
645 			M_PREPEND(control, sizeof(*cm), M_TRYWAIT);
646 			if (control == 0) {
647 				error = ENOBUFS;
648 				goto bad;
649 			} else {
650 				cm = mtod(control, struct cmsghdr *);
651 				cm->cmsg_len = control->m_len;
652 				cm->cmsg_level = SOL_SOCKET;
653 				cm->cmsg_type = SCM_RIGHTS;
654 			}
655 		}
656 #endif
657 	} else {
658 		control = 0;
659 	}
660 #ifdef KTRACE
661 	if (KTRPOINT(td, KTR_GENIO)) {
662 		iovlen = auio.uio_iovcnt * sizeof (struct iovec);
663 		MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK);
664 		bcopy(auio.uio_iov, ktriov, iovlen);
665 		ktruio = auio;
666 	}
667 #endif
668 	len = auio.uio_resid;
669 	error = so->so_proto->pr_usrreqs->pru_sosend(so, to, &auio, 0, control,
670 						     flags, td);
671 	if (error) {
672 		if (auio.uio_resid != len && (error == ERESTART ||
673 		    error == EINTR || error == EWOULDBLOCK))
674 			error = 0;
675 		/* Generation of SIGPIPE can be controlled per socket */
676 		if (error == EPIPE && !(so->so_options & SO_NOSIGPIPE)) {
677 			PROC_LOCK(td->td_proc);
678 			psignal(td->td_proc, SIGPIPE);
679 			PROC_UNLOCK(td->td_proc);
680 		}
681 	}
682 	if (error == 0)
683 		td->td_retval[0] = len - auio.uio_resid;
684 #ifdef KTRACE
685 	if (ktriov != NULL) {
686 		if (error == 0) {
687 			ktruio.uio_iov = ktriov;
688 			ktruio.uio_resid = td->td_retval[0];
689 			ktrgenio(s, UIO_WRITE, &ktruio, error);
690 		}
691 		FREE(ktriov, M_TEMP);
692 	}
693 #endif
694 bad:
695 	fputsock(so);
696 	if (to)
697 		FREE(to, M_SONAME);
698 	return (error);
699 }
700 
701 /*
702  * MPSAFE
703  */
704 int
705 sendto(td, uap)
706 	struct thread *td;
707 	register struct sendto_args /* {
708 		int	s;
709 		caddr_t	buf;
710 		size_t	len;
711 		int	flags;
712 		caddr_t	to;
713 		int	tolen;
714 	} */ *uap;
715 {
716 	struct msghdr msg;
717 	struct iovec aiov;
718 	int error;
719 
720 	msg.msg_name = uap->to;
721 	msg.msg_namelen = uap->tolen;
722 	msg.msg_iov = &aiov;
723 	msg.msg_iovlen = 1;
724 	msg.msg_control = 0;
725 #ifdef COMPAT_OLDSOCK
726 	msg.msg_flags = 0;
727 #endif
728 	aiov.iov_base = uap->buf;
729 	aiov.iov_len = uap->len;
730 	mtx_lock(&Giant);
731 	error = sendit(td, uap->s, &msg, uap->flags);
732 	mtx_unlock(&Giant);
733 	return (error);
734 }
735 
736 #ifdef COMPAT_OLDSOCK
737 /*
738  * MPSAFE
739  */
740 int
741 osend(td, uap)
742 	struct thread *td;
743 	register struct osend_args /* {
744 		int	s;
745 		caddr_t	buf;
746 		int	len;
747 		int	flags;
748 	} */ *uap;
749 {
750 	struct msghdr msg;
751 	struct iovec aiov;
752 	int error;
753 
754 	msg.msg_name = 0;
755 	msg.msg_namelen = 0;
756 	msg.msg_iov = &aiov;
757 	msg.msg_iovlen = 1;
758 	aiov.iov_base = uap->buf;
759 	aiov.iov_len = uap->len;
760 	msg.msg_control = 0;
761 	msg.msg_flags = 0;
762 	mtx_lock(&Giant);
763 	error = sendit(td, uap->s, &msg, uap->flags);
764 	mtx_unlock(&Giant);
765 	return (error);
766 }
767 
768 /*
769  * MPSAFE
770  */
771 int
772 osendmsg(td, uap)
773 	struct thread *td;
774 	register struct osendmsg_args /* {
775 		int	s;
776 		caddr_t	msg;
777 		int	flags;
778 	} */ *uap;
779 {
780 	struct msghdr msg;
781 	struct iovec aiov[UIO_SMALLIOV], *iov;
782 	int error;
783 
784 	mtx_lock(&Giant);
785 	error = copyin(uap->msg, &msg, sizeof (struct omsghdr));
786 	if (error)
787 		goto done2;
788 	if ((u_int)msg.msg_iovlen >= UIO_SMALLIOV) {
789 		if ((u_int)msg.msg_iovlen >= UIO_MAXIOV) {
790 			error = EMSGSIZE;
791 			goto done2;
792 		}
793 		MALLOC(iov, struct iovec *,
794 		      sizeof(struct iovec) * (u_int)msg.msg_iovlen, M_IOV,
795 		      M_WAITOK);
796 	} else {
797 		iov = aiov;
798 	}
799 	error = copyin(msg.msg_iov, iov,
800 	    (unsigned)(msg.msg_iovlen * sizeof (struct iovec)));
801 	if (error)
802 		goto done;
803 	msg.msg_flags = MSG_COMPAT;
804 	msg.msg_iov = iov;
805 	error = sendit(td, uap->s, &msg, uap->flags);
806 done:
807 	if (iov != aiov)
808 		FREE(iov, M_IOV);
809 done2:
810 	mtx_unlock(&Giant);
811 	return (error);
812 }
813 #endif
814 
815 /*
816  * MPSAFE
817  */
818 int
819 sendmsg(td, uap)
820 	struct thread *td;
821 	register struct sendmsg_args /* {
822 		int	s;
823 		caddr_t	msg;
824 		int	flags;
825 	} */ *uap;
826 {
827 	struct msghdr msg;
828 	struct iovec aiov[UIO_SMALLIOV], *iov;
829 	int error;
830 
831 	mtx_lock(&Giant);
832 	error = copyin(uap->msg, &msg, sizeof (msg));
833 	if (error)
834 		goto done2;
835 	if ((u_int)msg.msg_iovlen >= UIO_SMALLIOV) {
836 		if ((u_int)msg.msg_iovlen >= UIO_MAXIOV) {
837 			error = EMSGSIZE;
838 			goto done2;
839 		}
840 		MALLOC(iov, struct iovec *,
841 		       sizeof(struct iovec) * (u_int)msg.msg_iovlen, M_IOV,
842 		       M_WAITOK);
843 	} else {
844 		iov = aiov;
845 	}
846 	if (msg.msg_iovlen &&
847 	    (error = copyin(msg.msg_iov, iov,
848 	    (unsigned)(msg.msg_iovlen * sizeof (struct iovec)))))
849 		goto done;
850 	msg.msg_iov = iov;
851 #ifdef COMPAT_OLDSOCK
852 	msg.msg_flags = 0;
853 #endif
854 	error = sendit(td, uap->s, &msg, uap->flags);
855 done:
856 	if (iov != aiov)
857 		FREE(iov, M_IOV);
858 done2:
859 	mtx_unlock(&Giant);
860 	return (error);
861 }
862 
863 static int
864 recvit(td, s, mp, namelenp)
865 	register struct thread *td;
866 	int s;
867 	register struct msghdr *mp;
868 	void *namelenp;
869 {
870 	struct uio auio;
871 	register struct iovec *iov;
872 	register int i;
873 	int len, error;
874 	struct mbuf *m, *control = 0;
875 	caddr_t ctlbuf;
876 	struct socket *so;
877 	struct sockaddr *fromsa = 0;
878 #ifdef KTRACE
879 	struct iovec *ktriov = NULL;
880 	struct uio ktruio;
881 	int iovlen;
882 #endif
883 
884 	if ((error = fgetsock(td, s, &so, NULL)) != 0)
885 		return (error);
886 	auio.uio_iov = mp->msg_iov;
887 	auio.uio_iovcnt = mp->msg_iovlen;
888 	auio.uio_segflg = UIO_USERSPACE;
889 	auio.uio_rw = UIO_READ;
890 	auio.uio_td = td;
891 	auio.uio_offset = 0;			/* XXX */
892 	auio.uio_resid = 0;
893 	iov = mp->msg_iov;
894 	for (i = 0; i < mp->msg_iovlen; i++, iov++) {
895 		if ((auio.uio_resid += iov->iov_len) < 0) {
896 			fputsock(so);
897 			return (EINVAL);
898 		}
899 	}
900 #ifdef KTRACE
901 	if (KTRPOINT(td, KTR_GENIO)) {
902 		iovlen = auio.uio_iovcnt * sizeof (struct iovec);
903 		MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK);
904 		bcopy(auio.uio_iov, ktriov, iovlen);
905 		ktruio = auio;
906 	}
907 #endif
908 	len = auio.uio_resid;
909 	error = so->so_proto->pr_usrreqs->pru_soreceive(so, &fromsa, &auio,
910 	    (struct mbuf **)0, mp->msg_control ? &control : (struct mbuf **)0,
911 	    &mp->msg_flags);
912 	if (error) {
913 		if (auio.uio_resid != len && (error == ERESTART ||
914 		    error == EINTR || error == EWOULDBLOCK))
915 			error = 0;
916 	}
917 #ifdef KTRACE
918 	if (ktriov != NULL) {
919 		if (error == 0) {
920 			ktruio.uio_iov = ktriov;
921 			ktruio.uio_resid = len - auio.uio_resid;
922 			ktrgenio(s, UIO_READ, &ktruio, error);
923 		}
924 		FREE(ktriov, M_TEMP);
925 	}
926 #endif
927 	if (error)
928 		goto out;
929 	td->td_retval[0] = len - auio.uio_resid;
930 	if (mp->msg_name) {
931 		len = mp->msg_namelen;
932 		if (len <= 0 || fromsa == 0)
933 			len = 0;
934 		else {
935 #ifndef MIN
936 #define MIN(a,b) ((a)>(b)?(b):(a))
937 #endif
938 			/* save sa_len before it is destroyed by MSG_COMPAT */
939 			len = MIN(len, fromsa->sa_len);
940 #ifdef COMPAT_OLDSOCK
941 			if (mp->msg_flags & MSG_COMPAT)
942 				((struct osockaddr *)fromsa)->sa_family =
943 				    fromsa->sa_family;
944 #endif
945 			error = copyout(fromsa, mp->msg_name, (unsigned)len);
946 			if (error)
947 				goto out;
948 		}
949 		mp->msg_namelen = len;
950 		if (namelenp &&
951 		    (error = copyout(&len, namelenp, sizeof (int)))) {
952 #ifdef COMPAT_OLDSOCK
953 			if (mp->msg_flags & MSG_COMPAT)
954 				error = 0;	/* old recvfrom didn't check */
955 			else
956 #endif
957 			goto out;
958 		}
959 	}
960 	if (mp->msg_control) {
961 #ifdef COMPAT_OLDSOCK
962 		/*
963 		 * We assume that old recvmsg calls won't receive access
964 		 * rights and other control info, esp. as control info
965 		 * is always optional and those options didn't exist in 4.3.
966 		 * If we receive rights, trim the cmsghdr; anything else
967 		 * is tossed.
968 		 */
969 		if (control && mp->msg_flags & MSG_COMPAT) {
970 			if (mtod(control, struct cmsghdr *)->cmsg_level !=
971 			    SOL_SOCKET ||
972 			    mtod(control, struct cmsghdr *)->cmsg_type !=
973 			    SCM_RIGHTS) {
974 				mp->msg_controllen = 0;
975 				goto out;
976 			}
977 			control->m_len -= sizeof (struct cmsghdr);
978 			control->m_data += sizeof (struct cmsghdr);
979 		}
980 #endif
981 		len = mp->msg_controllen;
982 		m = control;
983 		mp->msg_controllen = 0;
984 		ctlbuf = mp->msg_control;
985 
986 		while (m && len > 0) {
987 			unsigned int tocopy;
988 
989 			if (len >= m->m_len)
990 				tocopy = m->m_len;
991 			else {
992 				mp->msg_flags |= MSG_CTRUNC;
993 				tocopy = len;
994 			}
995 
996 			if ((error = copyout(mtod(m, caddr_t),
997 					ctlbuf, tocopy)) != 0)
998 				goto out;
999 
1000 			ctlbuf += tocopy;
1001 			len -= tocopy;
1002 			m = m->m_next;
1003 		}
1004 		mp->msg_controllen = ctlbuf - (caddr_t)mp->msg_control;
1005 	}
1006 out:
1007 	fputsock(so);
1008 	if (fromsa)
1009 		FREE(fromsa, M_SONAME);
1010 	if (control)
1011 		m_freem(control);
1012 	return (error);
1013 }
1014 
1015 /*
1016  * MPSAFE
1017  */
1018 int
1019 recvfrom(td, uap)
1020 	struct thread *td;
1021 	register struct recvfrom_args /* {
1022 		int	s;
1023 		caddr_t	buf;
1024 		size_t	len;
1025 		int	flags;
1026 		caddr_t	from;
1027 		int	*fromlenaddr;
1028 	} */ *uap;
1029 {
1030 	struct msghdr msg;
1031 	struct iovec aiov;
1032 	int error;
1033 
1034 	mtx_lock(&Giant);
1035 	if (uap->fromlenaddr) {
1036 		error = copyin(uap->fromlenaddr,
1037 		    &msg.msg_namelen, sizeof (msg.msg_namelen));
1038 		if (error)
1039 			goto done2;
1040 	} else {
1041 		msg.msg_namelen = 0;
1042 	}
1043 	msg.msg_name = uap->from;
1044 	msg.msg_iov = &aiov;
1045 	msg.msg_iovlen = 1;
1046 	aiov.iov_base = uap->buf;
1047 	aiov.iov_len = uap->len;
1048 	msg.msg_control = 0;
1049 	msg.msg_flags = uap->flags;
1050 	error = recvit(td, uap->s, &msg, uap->fromlenaddr);
1051 done2:
1052 	mtx_unlock(&Giant);
1053 	return(error);
1054 }
1055 
1056 #ifdef COMPAT_OLDSOCK
1057 /*
1058  * MPSAFE
1059  */
1060 int
1061 orecvfrom(td, uap)
1062 	struct thread *td;
1063 	struct recvfrom_args *uap;
1064 {
1065 
1066 	uap->flags |= MSG_COMPAT;
1067 	return (recvfrom(td, uap));
1068 }
1069 #endif
1070 
1071 
1072 #ifdef COMPAT_OLDSOCK
1073 /*
1074  * MPSAFE
1075  */
1076 int
1077 orecv(td, uap)
1078 	struct thread *td;
1079 	register struct orecv_args /* {
1080 		int	s;
1081 		caddr_t	buf;
1082 		int	len;
1083 		int	flags;
1084 	} */ *uap;
1085 {
1086 	struct msghdr msg;
1087 	struct iovec aiov;
1088 	int error;
1089 
1090 	mtx_lock(&Giant);
1091 	msg.msg_name = 0;
1092 	msg.msg_namelen = 0;
1093 	msg.msg_iov = &aiov;
1094 	msg.msg_iovlen = 1;
1095 	aiov.iov_base = uap->buf;
1096 	aiov.iov_len = uap->len;
1097 	msg.msg_control = 0;
1098 	msg.msg_flags = uap->flags;
1099 	error = recvit(td, uap->s, &msg, NULL);
1100 	mtx_unlock(&Giant);
1101 	return (error);
1102 }
1103 
1104 /*
1105  * Old recvmsg.  This code takes advantage of the fact that the old msghdr
1106  * overlays the new one, missing only the flags, and with the (old) access
1107  * rights where the control fields are now.
1108  *
1109  * MPSAFE
1110  */
1111 int
1112 orecvmsg(td, uap)
1113 	struct thread *td;
1114 	register struct orecvmsg_args /* {
1115 		int	s;
1116 		struct	omsghdr *msg;
1117 		int	flags;
1118 	} */ *uap;
1119 {
1120 	struct msghdr msg;
1121 	struct iovec aiov[UIO_SMALLIOV], *iov;
1122 	int error;
1123 
1124 	error = copyin(uap->msg, &msg, sizeof (struct omsghdr));
1125 	if (error)
1126 		return (error);
1127 
1128 	mtx_lock(&Giant);
1129 	if ((u_int)msg.msg_iovlen >= UIO_SMALLIOV) {
1130 		if ((u_int)msg.msg_iovlen >= UIO_MAXIOV) {
1131 			error = EMSGSIZE;
1132 			goto done2;
1133 		}
1134 		MALLOC(iov, struct iovec *,
1135 		      sizeof(struct iovec) * (u_int)msg.msg_iovlen, M_IOV,
1136 		      M_WAITOK);
1137 	} else {
1138 		iov = aiov;
1139 	}
1140 	msg.msg_flags = uap->flags | MSG_COMPAT;
1141 	error = copyin(msg.msg_iov, iov,
1142 	    (unsigned)(msg.msg_iovlen * sizeof (struct iovec)));
1143 	if (error)
1144 		goto done;
1145 	msg.msg_iov = iov;
1146 	error = recvit(td, uap->s, &msg, &uap->msg->msg_namelen);
1147 
1148 	if (msg.msg_controllen && error == 0)
1149 		error = copyout(&msg.msg_controllen,
1150 		    &uap->msg->msg_accrightslen, sizeof (int));
1151 done:
1152 	if (iov != aiov)
1153 		FREE(iov, M_IOV);
1154 done2:
1155 	mtx_unlock(&Giant);
1156 	return (error);
1157 }
1158 #endif
1159 
1160 /*
1161  * MPSAFE
1162  */
1163 int
1164 recvmsg(td, uap)
1165 	struct thread *td;
1166 	register struct recvmsg_args /* {
1167 		int	s;
1168 		struct	msghdr *msg;
1169 		int	flags;
1170 	} */ *uap;
1171 {
1172 	struct msghdr msg;
1173 	struct iovec aiov[UIO_SMALLIOV], *uiov, *iov;
1174 	register int error;
1175 
1176 	mtx_lock(&Giant);
1177 	error = copyin(uap->msg, &msg, sizeof (msg));
1178 	if (error)
1179 		goto done2;
1180 	if ((u_int)msg.msg_iovlen >= UIO_SMALLIOV) {
1181 		if ((u_int)msg.msg_iovlen >= UIO_MAXIOV) {
1182 			error = EMSGSIZE;
1183 			goto done2;
1184 		}
1185 		MALLOC(iov, struct iovec *,
1186 		       sizeof(struct iovec) * (u_int)msg.msg_iovlen, M_IOV,
1187 		       M_WAITOK);
1188 	} else {
1189 		iov = aiov;
1190 	}
1191 #ifdef COMPAT_OLDSOCK
1192 	msg.msg_flags = uap->flags &~ MSG_COMPAT;
1193 #else
1194 	msg.msg_flags = uap->flags;
1195 #endif
1196 	uiov = msg.msg_iov;
1197 	msg.msg_iov = iov;
1198 	error = copyin(uiov, iov,
1199 	    (unsigned)(msg.msg_iovlen * sizeof (struct iovec)));
1200 	if (error)
1201 		goto done;
1202 	error = recvit(td, uap->s, &msg, NULL);
1203 	if (!error) {
1204 		msg.msg_iov = uiov;
1205 		error = copyout(&msg, uap->msg, sizeof(msg));
1206 	}
1207 done:
1208 	if (iov != aiov)
1209 		FREE(iov, M_IOV);
1210 done2:
1211 	mtx_unlock(&Giant);
1212 	return (error);
1213 }
1214 
1215 /*
1216  * MPSAFE
1217  */
1218 /* ARGSUSED */
1219 int
1220 shutdown(td, uap)
1221 	struct thread *td;
1222 	register struct shutdown_args /* {
1223 		int	s;
1224 		int	how;
1225 	} */ *uap;
1226 {
1227 	struct socket *so;
1228 	int error;
1229 
1230 	mtx_lock(&Giant);
1231 	if ((error = fgetsock(td, uap->s, &so, NULL)) == 0) {
1232 		error = soshutdown(so, uap->how);
1233 		fputsock(so);
1234 	}
1235 	mtx_unlock(&Giant);
1236 	return(error);
1237 }
1238 
1239 /*
1240  * MPSAFE
1241  */
1242 /* ARGSUSED */
1243 int
1244 setsockopt(td, uap)
1245 	struct thread *td;
1246 	register struct setsockopt_args /* {
1247 		int	s;
1248 		int	level;
1249 		int	name;
1250 		caddr_t	val;
1251 		int	valsize;
1252 	} */ *uap;
1253 {
1254 	struct socket *so;
1255 	struct sockopt sopt;
1256 	int error;
1257 
1258 	if (uap->val == 0 && uap->valsize != 0)
1259 		return (EFAULT);
1260 	if (uap->valsize < 0)
1261 		return (EINVAL);
1262 
1263 	mtx_lock(&Giant);
1264 	if ((error = fgetsock(td, uap->s, &so, NULL)) == 0) {
1265 		sopt.sopt_dir = SOPT_SET;
1266 		sopt.sopt_level = uap->level;
1267 		sopt.sopt_name = uap->name;
1268 		sopt.sopt_val = uap->val;
1269 		sopt.sopt_valsize = uap->valsize;
1270 		sopt.sopt_td = td;
1271 		error = sosetopt(so, &sopt);
1272 		fputsock(so);
1273 	}
1274 	mtx_unlock(&Giant);
1275 	return(error);
1276 }
1277 
1278 /*
1279  * MPSAFE
1280  */
1281 /* ARGSUSED */
1282 int
1283 getsockopt(td, uap)
1284 	struct thread *td;
1285 	register struct getsockopt_args /* {
1286 		int	s;
1287 		int	level;
1288 		int	name;
1289 		caddr_t	val;
1290 		int	*avalsize;
1291 	} */ *uap;
1292 {
1293 	int	valsize, error;
1294 	struct  socket *so;
1295 	struct	sockopt sopt;
1296 
1297 	mtx_lock(&Giant);
1298 	if ((error = fgetsock(td, uap->s, &so, NULL)) != 0)
1299 		goto done2;
1300 	if (uap->val) {
1301 		error = copyin(uap->avalsize, &valsize, sizeof (valsize));
1302 		if (error)
1303 			goto done1;
1304 		if (valsize < 0) {
1305 			error = EINVAL;
1306 			goto done1;
1307 		}
1308 	} else {
1309 		valsize = 0;
1310 	}
1311 
1312 	sopt.sopt_dir = SOPT_GET;
1313 	sopt.sopt_level = uap->level;
1314 	sopt.sopt_name = uap->name;
1315 	sopt.sopt_val = uap->val;
1316 	sopt.sopt_valsize = (size_t)valsize; /* checked non-negative above */
1317 	sopt.sopt_td = td;
1318 
1319 	error = sogetopt(so, &sopt);
1320 	if (error == 0) {
1321 		valsize = sopt.sopt_valsize;
1322 		error = copyout(&valsize, uap->avalsize, sizeof (valsize));
1323 	}
1324 done1:
1325 	fputsock(so);
1326 done2:
1327 	mtx_unlock(&Giant);
1328 	return (error);
1329 }
1330 
1331 /*
1332  * getsockname1() - Get socket name.
1333  *
1334  * MPSAFE
1335  */
1336 /* ARGSUSED */
1337 static int
1338 getsockname1(td, uap, compat)
1339 	struct thread *td;
1340 	register struct getsockname_args /* {
1341 		int	fdes;
1342 		caddr_t	asa;
1343 		int	*alen;
1344 	} */ *uap;
1345 	int compat;
1346 {
1347 	struct socket *so;
1348 	struct sockaddr *sa;
1349 	int len, error;
1350 
1351 	mtx_lock(&Giant);
1352 	if ((error = fgetsock(td, uap->fdes, &so, NULL)) != 0)
1353 		goto done2;
1354 	error = copyin(uap->alen, &len, sizeof (len));
1355 	if (error)
1356 		goto done1;
1357 	if (len < 0) {
1358 		error = EINVAL;
1359 		goto done1;
1360 	}
1361 	sa = 0;
1362 	error = (*so->so_proto->pr_usrreqs->pru_sockaddr)(so, &sa);
1363 	if (error)
1364 		goto bad;
1365 	if (sa == 0) {
1366 		len = 0;
1367 		goto gotnothing;
1368 	}
1369 
1370 	len = MIN(len, sa->sa_len);
1371 #ifdef COMPAT_OLDSOCK
1372 	if (compat)
1373 		((struct osockaddr *)sa)->sa_family = sa->sa_family;
1374 #endif
1375 	error = copyout(sa, uap->asa, (u_int)len);
1376 	if (error == 0)
1377 gotnothing:
1378 		error = copyout(&len, uap->alen, sizeof (len));
1379 bad:
1380 	if (sa)
1381 		FREE(sa, M_SONAME);
1382 done1:
1383 	fputsock(so);
1384 done2:
1385 	mtx_unlock(&Giant);
1386 	return (error);
1387 }
1388 
1389 /*
1390  * MPSAFE
1391  */
1392 int
1393 getsockname(td, uap)
1394 	struct thread *td;
1395 	struct getsockname_args *uap;
1396 {
1397 
1398 	return (getsockname1(td, uap, 0));
1399 }
1400 
1401 #ifdef COMPAT_OLDSOCK
1402 /*
1403  * MPSAFE
1404  */
1405 int
1406 ogetsockname(td, uap)
1407 	struct thread *td;
1408 	struct getsockname_args *uap;
1409 {
1410 
1411 	return (getsockname1(td, uap, 1));
1412 }
1413 #endif /* COMPAT_OLDSOCK */
1414 
1415 /*
1416  * getpeername1() - Get name of peer for connected socket.
1417  *
1418  * MPSAFE
1419  */
1420 /* ARGSUSED */
1421 static int
1422 getpeername1(td, uap, compat)
1423 	struct thread *td;
1424 	register struct getpeername_args /* {
1425 		int	fdes;
1426 		caddr_t	asa;
1427 		int	*alen;
1428 	} */ *uap;
1429 	int compat;
1430 {
1431 	struct socket *so;
1432 	struct sockaddr *sa;
1433 	int len, error;
1434 
1435 	mtx_lock(&Giant);
1436 	if ((error = fgetsock(td, uap->fdes, &so, NULL)) != 0)
1437 		goto done2;
1438 	if ((so->so_state & (SS_ISCONNECTED|SS_ISCONFIRMING)) == 0) {
1439 		error = ENOTCONN;
1440 		goto done1;
1441 	}
1442 	error = copyin(uap->alen, &len, sizeof (len));
1443 	if (error)
1444 		goto done1;
1445 	if (len < 0) {
1446 		error = EINVAL;
1447 		goto done1;
1448 	}
1449 	sa = 0;
1450 	error = (*so->so_proto->pr_usrreqs->pru_peeraddr)(so, &sa);
1451 	if (error)
1452 		goto bad;
1453 	if (sa == 0) {
1454 		len = 0;
1455 		goto gotnothing;
1456 	}
1457 	len = MIN(len, sa->sa_len);
1458 #ifdef COMPAT_OLDSOCK
1459 	if (compat)
1460 		((struct osockaddr *)sa)->sa_family =
1461 		    sa->sa_family;
1462 #endif
1463 	error = copyout(sa, uap->asa, (u_int)len);
1464 	if (error)
1465 		goto bad;
1466 gotnothing:
1467 	error = copyout(&len, uap->alen, sizeof (len));
1468 bad:
1469 	if (sa)
1470 		FREE(sa, M_SONAME);
1471 done1:
1472 	fputsock(so);
1473 done2:
1474 	mtx_unlock(&Giant);
1475 	return (error);
1476 }
1477 
1478 /*
1479  * MPSAFE
1480  */
1481 int
1482 getpeername(td, uap)
1483 	struct thread *td;
1484 	struct getpeername_args *uap;
1485 {
1486 
1487 	return (getpeername1(td, uap, 0));
1488 }
1489 
1490 #ifdef COMPAT_OLDSOCK
1491 /*
1492  * MPSAFE
1493  */
1494 int
1495 ogetpeername(td, uap)
1496 	struct thread *td;
1497 	struct ogetpeername_args *uap;
1498 {
1499 
1500 	/* XXX uap should have type `getpeername_args *' to begin with. */
1501 	return (getpeername1(td, (struct getpeername_args *)uap, 1));
1502 }
1503 #endif /* COMPAT_OLDSOCK */
1504 
1505 int
1506 sockargs(mp, buf, buflen, type)
1507 	struct mbuf **mp;
1508 	caddr_t buf;
1509 	int buflen, type;
1510 {
1511 	register struct sockaddr *sa;
1512 	register struct mbuf *m;
1513 	int error;
1514 
1515 	if ((u_int)buflen > MLEN) {
1516 #ifdef COMPAT_OLDSOCK
1517 		if (type == MT_SONAME && (u_int)buflen <= 112)
1518 			buflen = MLEN;		/* unix domain compat. hack */
1519 		else
1520 #endif
1521 		return (EINVAL);
1522 	}
1523 	m = m_get(M_TRYWAIT, type);
1524 	if (m == NULL)
1525 		return (ENOBUFS);
1526 	m->m_len = buflen;
1527 	error = copyin(buf, mtod(m, caddr_t), (u_int)buflen);
1528 	if (error)
1529 		(void) m_free(m);
1530 	else {
1531 		*mp = m;
1532 		if (type == MT_SONAME) {
1533 			sa = mtod(m, struct sockaddr *);
1534 
1535 #if defined(COMPAT_OLDSOCK) && BYTE_ORDER != BIG_ENDIAN
1536 			if (sa->sa_family == 0 && sa->sa_len < AF_MAX)
1537 				sa->sa_family = sa->sa_len;
1538 #endif
1539 			sa->sa_len = buflen;
1540 		}
1541 	}
1542 	return (error);
1543 }
1544 
1545 int
1546 getsockaddr(namp, uaddr, len)
1547 	struct sockaddr **namp;
1548 	caddr_t uaddr;
1549 	size_t len;
1550 {
1551 	struct sockaddr *sa;
1552 	int error;
1553 
1554 	if (len > SOCK_MAXADDRLEN)
1555 		return ENAMETOOLONG;
1556 	MALLOC(sa, struct sockaddr *, len, M_SONAME, M_WAITOK);
1557 	error = copyin(uaddr, sa, len);
1558 	if (error) {
1559 		FREE(sa, M_SONAME);
1560 	} else {
1561 #if defined(COMPAT_OLDSOCK) && BYTE_ORDER != BIG_ENDIAN
1562 		if (sa->sa_family == 0 && sa->sa_len < AF_MAX)
1563 			sa->sa_family = sa->sa_len;
1564 #endif
1565 		sa->sa_len = len;
1566 		*namp = sa;
1567 	}
1568 	return error;
1569 }
1570 
1571 /*
1572  * Allocate a pool of sf_bufs (sendfile(2) or "super-fast" if you prefer. :-))
1573  */
1574 static void
1575 sf_buf_init(void *arg)
1576 {
1577 	int i;
1578 
1579 	mtx_init(&sf_freelist.sf_lock, "sf_bufs list lock", NULL, MTX_DEF);
1580 	mtx_lock(&sf_freelist.sf_lock);
1581 	SLIST_INIT(&sf_freelist.sf_head);
1582 	sf_base = kmem_alloc_pageable(kernel_map, nsfbufs * PAGE_SIZE);
1583 	sf_bufs = malloc(nsfbufs * sizeof(struct sf_buf), M_TEMP,
1584 	    M_NOWAIT | M_ZERO);
1585 	for (i = 0; i < nsfbufs; i++) {
1586 		sf_bufs[i].kva = sf_base + i * PAGE_SIZE;
1587 		SLIST_INSERT_HEAD(&sf_freelist.sf_head, &sf_bufs[i], free_list);
1588 	}
1589 	sf_buf_alloc_want = 0;
1590 	mtx_unlock(&sf_freelist.sf_lock);
1591 }
1592 
1593 /*
1594  * Get an sf_buf from the freelist. Will block if none are available.
1595  */
1596 struct sf_buf *
1597 sf_buf_alloc()
1598 {
1599 	struct sf_buf *sf;
1600 	int error;
1601 
1602 	mtx_lock(&sf_freelist.sf_lock);
1603 	while ((sf = SLIST_FIRST(&sf_freelist.sf_head)) == NULL) {
1604 		sf_buf_alloc_want++;
1605 		error = msleep(&sf_freelist, &sf_freelist.sf_lock, PVM|PCATCH,
1606 		    "sfbufa", 0);
1607 		sf_buf_alloc_want--;
1608 
1609 		/*
1610 		 * If we got a signal, don't risk going back to sleep.
1611 		 */
1612 		if (error)
1613 			break;
1614 	}
1615 	if (sf != NULL)
1616 		SLIST_REMOVE_HEAD(&sf_freelist.sf_head, free_list);
1617 	mtx_unlock(&sf_freelist.sf_lock);
1618 	return (sf);
1619 }
1620 
1621 #define dtosf(x)	(&sf_bufs[((uintptr_t)(x) - (uintptr_t)sf_base) >> PAGE_SHIFT])
1622 
1623 /*
1624  * Detatch mapped page and release resources back to the system.
1625  */
1626 void
1627 sf_buf_free(void *addr, void *args)
1628 {
1629 	struct sf_buf *sf;
1630 	struct vm_page *m;
1631 
1632 	GIANT_REQUIRED;
1633 
1634 	sf = dtosf(addr);
1635 	pmap_qremove((vm_offset_t)addr, 1);
1636 	m = sf->m;
1637 	vm_page_lock_queues();
1638 	vm_page_unwire(m, 0);
1639 	/*
1640 	 * Check for the object going away on us. This can
1641 	 * happen since we don't hold a reference to it.
1642 	 * If so, we're responsible for freeing the page.
1643 	 */
1644 	if (m->wire_count == 0 && m->object == NULL)
1645 		vm_page_free(m);
1646 	vm_page_unlock_queues();
1647 	sf->m = NULL;
1648 	mtx_lock(&sf_freelist.sf_lock);
1649 	SLIST_INSERT_HEAD(&sf_freelist.sf_head, sf, free_list);
1650 	if (sf_buf_alloc_want > 0)
1651 		wakeup_one(&sf_freelist);
1652 	mtx_unlock(&sf_freelist.sf_lock);
1653 }
1654 
1655 /*
1656  * sendfile(2)
1657  *
1658  * MPSAFE
1659  *
1660  * int sendfile(int fd, int s, off_t offset, size_t nbytes,
1661  *	 struct sf_hdtr *hdtr, off_t *sbytes, int flags)
1662  *
1663  * Send a file specified by 'fd' and starting at 'offset' to a socket
1664  * specified by 's'. Send only 'nbytes' of the file or until EOF if
1665  * nbytes == 0. Optionally add a header and/or trailer to the socket
1666  * output. If specified, write the total number of bytes sent into *sbytes.
1667  *
1668  */
1669 int
1670 sendfile(struct thread *td, struct sendfile_args *uap)
1671 {
1672 
1673 	return (do_sendfile(td, uap, 0));
1674 }
1675 
1676 #ifdef COMPAT_FREEBSD4
1677 int
1678 freebsd4_sendfile(struct thread *td, struct freebsd4_sendfile_args *uap)
1679 {
1680 	struct sendfile_args args;
1681 
1682 	args.fd = uap->fd;
1683 	args.s = uap->s;
1684 	args.offset = uap->offset;
1685 	args.nbytes = uap->nbytes;
1686 	args.hdtr = uap->hdtr;
1687 	args.sbytes = uap->sbytes;
1688 	args.flags = uap->flags;
1689 
1690 	return (do_sendfile(td, &args, 1));
1691 }
1692 #endif /* COMPAT_FREEBSD4 */
1693 
1694 static int
1695 do_sendfile(struct thread *td, struct sendfile_args *uap, int compat)
1696 {
1697 	struct vnode *vp;
1698 	struct vm_object *obj;
1699 	struct socket *so = NULL;
1700 	struct mbuf *m;
1701 	struct sf_buf *sf;
1702 	struct vm_page *pg;
1703 	struct writev_args nuap;
1704 	struct sf_hdtr hdtr;
1705 	off_t off, xfsize, hdtr_size, sbytes = 0;
1706 	int error, s;
1707 
1708 	mtx_lock(&Giant);
1709 
1710 	hdtr_size = 0;
1711 
1712 	/*
1713 	 * The descriptor must be a regular file and have a backing VM object.
1714 	 */
1715 	if ((error = fgetvp_read(td, uap->fd, &vp)) != 0)
1716 		goto done;
1717 	if (vp->v_type != VREG || VOP_GETVOBJECT(vp, &obj) != 0) {
1718 		error = EINVAL;
1719 		goto done;
1720 	}
1721 	if ((error = fgetsock(td, uap->s, &so, NULL)) != 0)
1722 		goto done;
1723 	if (so->so_type != SOCK_STREAM) {
1724 		error = EINVAL;
1725 		goto done;
1726 	}
1727 	if ((so->so_state & SS_ISCONNECTED) == 0) {
1728 		error = ENOTCONN;
1729 		goto done;
1730 	}
1731 	if (uap->offset < 0) {
1732 		error = EINVAL;
1733 		goto done;
1734 	}
1735 
1736 	/*
1737 	 * If specified, get the pointer to the sf_hdtr struct for
1738 	 * any headers/trailers.
1739 	 */
1740 	if (uap->hdtr != NULL) {
1741 		error = copyin(uap->hdtr, &hdtr, sizeof(hdtr));
1742 		if (error)
1743 			goto done;
1744 		/*
1745 		 * Send any headers. Wimp out and use writev(2).
1746 		 */
1747 		if (hdtr.headers != NULL) {
1748 			nuap.fd = uap->s;
1749 			nuap.iovp = hdtr.headers;
1750 			nuap.iovcnt = hdtr.hdr_cnt;
1751 			error = writev(td, &nuap);
1752 			if (error)
1753 				goto done;
1754 			if (compat)
1755 				sbytes += td->td_retval[0];
1756 			else
1757 				hdtr_size += td->td_retval[0];
1758 		}
1759 	}
1760 
1761 	/*
1762 	 * Protect against multiple writers to the socket.
1763 	 */
1764 	(void) sblock(&so->so_snd, M_WAITOK);
1765 
1766 	/*
1767 	 * Loop through the pages in the file, starting with the requested
1768 	 * offset. Get a file page (do I/O if necessary), map the file page
1769 	 * into an sf_buf, attach an mbuf header to the sf_buf, and queue
1770 	 * it on the socket.
1771 	 */
1772 	for (off = uap->offset; ; off += xfsize, sbytes += xfsize) {
1773 		vm_pindex_t pindex;
1774 		vm_offset_t pgoff;
1775 
1776 		pindex = OFF_TO_IDX(off);
1777 retry_lookup:
1778 		/*
1779 		 * Calculate the amount to transfer. Not to exceed a page,
1780 		 * the EOF, or the passed in nbytes.
1781 		 */
1782 		xfsize = obj->un_pager.vnp.vnp_size - off;
1783 		if (xfsize > PAGE_SIZE)
1784 			xfsize = PAGE_SIZE;
1785 		pgoff = (vm_offset_t)(off & PAGE_MASK);
1786 		if (PAGE_SIZE - pgoff < xfsize)
1787 			xfsize = PAGE_SIZE - pgoff;
1788 		if (uap->nbytes && xfsize > (uap->nbytes - sbytes))
1789 			xfsize = uap->nbytes - sbytes;
1790 		if (xfsize <= 0)
1791 			break;
1792 		/*
1793 		 * Optimize the non-blocking case by looking at the socket space
1794 		 * before going to the extra work of constituting the sf_buf.
1795 		 */
1796 		if ((so->so_state & SS_NBIO) && sbspace(&so->so_snd) <= 0) {
1797 			if (so->so_state & SS_CANTSENDMORE)
1798 				error = EPIPE;
1799 			else
1800 				error = EAGAIN;
1801 			sbunlock(&so->so_snd);
1802 			goto done;
1803 		}
1804 		/*
1805 		 * Attempt to look up the page.
1806 		 *
1807 		 *	Allocate if not found
1808 		 *
1809 		 *	Wait and loop if busy.
1810 		 */
1811 		pg = vm_page_lookup(obj, pindex);
1812 
1813 		if (pg == NULL) {
1814 			pg = vm_page_alloc(obj, pindex,
1815 			    VM_ALLOC_NORMAL | VM_ALLOC_WIRED);
1816 			if (pg == NULL) {
1817 				VM_WAIT;
1818 				goto retry_lookup;
1819 			}
1820 			vm_page_lock_queues();
1821 			vm_page_wakeup(pg);
1822 		} else {
1823 			vm_page_lock_queues();
1824 			if (vm_page_sleep_if_busy(pg, TRUE, "sfpbsy"))
1825 				goto retry_lookup;
1826 			/*
1827 		 	 * Wire the page so it does not get ripped out from
1828 			 * under us.
1829 			 */
1830 			vm_page_wire(pg);
1831 		}
1832 
1833 		/*
1834 		 * If page is not valid for what we need, initiate I/O
1835 		 */
1836 
1837 		if (!pg->valid || !vm_page_is_valid(pg, pgoff, xfsize)) {
1838 			int bsize, resid;
1839 
1840 			/*
1841 			 * Ensure that our page is still around when the I/O
1842 			 * completes.
1843 			 */
1844 			vm_page_io_start(pg);
1845 			vm_page_unlock_queues();
1846 
1847 			/*
1848 			 * Get the page from backing store.
1849 			 */
1850 			bsize = vp->v_mount->mnt_stat.f_iosize;
1851 			vn_lock(vp, LK_SHARED | LK_NOPAUSE | LK_RETRY, td);
1852 			/*
1853 			 * XXXMAC: Because we don't have fp->f_cred here,
1854 			 * we pass in NOCRED.  This is probably wrong, but
1855 			 * is consistent with our original implementation.
1856 			 */
1857 			error = vn_rdwr(UIO_READ, vp, NULL, MAXBSIZE,
1858 			    trunc_page(off), UIO_NOCOPY, IO_NODELOCKED |
1859 			    IO_VMIO | ((MAXBSIZE / bsize) << 16),
1860 			    td->td_ucred, NOCRED, &resid, td);
1861 			VOP_UNLOCK(vp, 0, td);
1862 			vm_page_lock_queues();
1863 			vm_page_flag_clear(pg, PG_ZERO);
1864 			vm_page_io_finish(pg);
1865 			if (error) {
1866 				vm_page_unwire(pg, 0);
1867 				/*
1868 				 * See if anyone else might know about this page.
1869 				 * If not and it is not valid, then free it.
1870 				 */
1871 				if (pg->wire_count == 0 && pg->valid == 0 &&
1872 				    pg->busy == 0 && !(pg->flags & PG_BUSY) &&
1873 				    pg->hold_count == 0) {
1874 					vm_page_busy(pg);
1875 					vm_page_free(pg);
1876 				}
1877 				vm_page_unlock_queues();
1878 				sbunlock(&so->so_snd);
1879 				goto done;
1880 			}
1881 		}
1882 		vm_page_unlock_queues();
1883 
1884 		/*
1885 		 * Get a sendfile buf. We usually wait as long as necessary,
1886 		 * but this wait can be interrupted.
1887 		 */
1888 		if ((sf = sf_buf_alloc()) == NULL) {
1889 			vm_page_lock_queues();
1890 			vm_page_unwire(pg, 0);
1891 			if (pg->wire_count == 0 && pg->object == NULL)
1892 				vm_page_free(pg);
1893 			vm_page_unlock_queues();
1894 			sbunlock(&so->so_snd);
1895 			error = EINTR;
1896 			goto done;
1897 		}
1898 
1899 		/*
1900 		 * Allocate a kernel virtual page and insert the physical page
1901 		 * into it.
1902 		 */
1903 		sf->m = pg;
1904 		pmap_qenter(sf->kva, &pg, 1);
1905 		/*
1906 		 * Get an mbuf header and set it up as having external storage.
1907 		 */
1908 		MGETHDR(m, M_TRYWAIT, MT_DATA);
1909 		if (m == NULL) {
1910 			error = ENOBUFS;
1911 			sf_buf_free((void *)sf->kva, NULL);
1912 			sbunlock(&so->so_snd);
1913 			goto done;
1914 		}
1915 		/*
1916 		 * Setup external storage for mbuf.
1917 		 */
1918 		MEXTADD(m, sf->kva, PAGE_SIZE, sf_buf_free, NULL, M_RDONLY,
1919 		    EXT_SFBUF);
1920 		m->m_data = (char *) sf->kva + pgoff;
1921 		m->m_pkthdr.len = m->m_len = xfsize;
1922 		/*
1923 		 * Add the buffer to the socket buffer chain.
1924 		 */
1925 		s = splnet();
1926 retry_space:
1927 		/*
1928 		 * Make sure that the socket is still able to take more data.
1929 		 * CANTSENDMORE being true usually means that the connection
1930 		 * was closed. so_error is true when an error was sensed after
1931 		 * a previous send.
1932 		 * The state is checked after the page mapping and buffer
1933 		 * allocation above since those operations may block and make
1934 		 * any socket checks stale. From this point forward, nothing
1935 		 * blocks before the pru_send (or more accurately, any blocking
1936 		 * results in a loop back to here to re-check).
1937 		 */
1938 		if ((so->so_state & SS_CANTSENDMORE) || so->so_error) {
1939 			if (so->so_state & SS_CANTSENDMORE) {
1940 				error = EPIPE;
1941 			} else {
1942 				error = so->so_error;
1943 				so->so_error = 0;
1944 			}
1945 			m_freem(m);
1946 			sbunlock(&so->so_snd);
1947 			splx(s);
1948 			goto done;
1949 		}
1950 		/*
1951 		 * Wait for socket space to become available. We do this just
1952 		 * after checking the connection state above in order to avoid
1953 		 * a race condition with sbwait().
1954 		 */
1955 		if (sbspace(&so->so_snd) < so->so_snd.sb_lowat) {
1956 			if (so->so_state & SS_NBIO) {
1957 				m_freem(m);
1958 				sbunlock(&so->so_snd);
1959 				splx(s);
1960 				error = EAGAIN;
1961 				goto done;
1962 			}
1963 			error = sbwait(&so->so_snd);
1964 			/*
1965 			 * An error from sbwait usually indicates that we've
1966 			 * been interrupted by a signal. If we've sent anything
1967 			 * then return bytes sent, otherwise return the error.
1968 			 */
1969 			if (error) {
1970 				m_freem(m);
1971 				sbunlock(&so->so_snd);
1972 				splx(s);
1973 				goto done;
1974 			}
1975 			goto retry_space;
1976 		}
1977 		error = (*so->so_proto->pr_usrreqs->pru_send)(so, 0, m, 0, 0, td);
1978 		splx(s);
1979 		if (error) {
1980 			sbunlock(&so->so_snd);
1981 			goto done;
1982 		}
1983 	}
1984 	sbunlock(&so->so_snd);
1985 
1986 	/*
1987 	 * Send trailers. Wimp out and use writev(2).
1988 	 */
1989 	if (uap->hdtr != NULL && hdtr.trailers != NULL) {
1990 			nuap.fd = uap->s;
1991 			nuap.iovp = hdtr.trailers;
1992 			nuap.iovcnt = hdtr.trl_cnt;
1993 			error = writev(td, &nuap);
1994 			if (error)
1995 				goto done;
1996 			if (compat)
1997 				sbytes += td->td_retval[0];
1998 			else
1999 				hdtr_size += td->td_retval[0];
2000 	}
2001 
2002 done:
2003 	/*
2004 	 * If there was no error we have to clear td->td_retval[0]
2005 	 * because it may have been set by writev.
2006 	 */
2007 	if (error == 0) {
2008 		td->td_retval[0] = 0;
2009 	}
2010 	if (uap->sbytes != NULL) {
2011 		if (!compat)
2012 			sbytes += hdtr_size;
2013 		copyout(&sbytes, uap->sbytes, sizeof(off_t));
2014 	}
2015 	if (vp)
2016 		vrele(vp);
2017 	if (so)
2018 		fputsock(so);
2019 	mtx_unlock(&Giant);
2020 	return (error);
2021 }
2022