xref: /dragonfly/sys/kern/uipc_syscalls.c (revision 3641b7ca)
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 1.85 2008/04/14 12:01:50 dillon 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 <net/netmsg2.h>
83 
84 #ifdef SCTP
85 #include <netinet/sctp_peeloff.h>
86 #endif /* SCTP */
87 
88 struct sfbuf_mref {
89 	struct sf_buf	*sf;
90 	int		mref_count;
91 	struct lwkt_serialize serializer;
92 };
93 
94 static MALLOC_DEFINE(M_SENDFILE, "sendfile", "sendfile sfbuf ref structures");
95 
96 /*
97  * System call interface to the socket abstraction.
98  */
99 
100 extern	struct fileops socketops;
101 
102 /*
103  * socket_args(int domain, int type, int protocol)
104  */
105 int
106 kern_socket(int domain, int type, int protocol, int *res)
107 {
108 	struct thread *td = curthread;
109 	struct proc *p = td->td_proc;
110 	struct socket *so;
111 	struct file *fp;
112 	int fd, error;
113 
114 	KKASSERT(p);
115 
116 	error = falloc(p, &fp, &fd);
117 	if (error)
118 		return (error);
119 	error = socreate(domain, &so, type, protocol, td);
120 	if (error) {
121 		fsetfd(p, NULL, fd);
122 	} else {
123 		fp->f_type = DTYPE_SOCKET;
124 		fp->f_flag = FREAD | FWRITE;
125 		fp->f_ops = &socketops;
126 		fp->f_data = so;
127 		*res = fd;
128 		fsetfd(p, fp, fd);
129 	}
130 	fdrop(fp);
131 	return (error);
132 }
133 
134 int
135 sys_socket(struct socket_args *uap)
136 {
137 	int error;
138 
139 	error = kern_socket(uap->domain, uap->type, uap->protocol,
140 	    &uap->sysmsg_result);
141 
142 	return (error);
143 }
144 
145 int
146 kern_bind(int s, struct sockaddr *sa)
147 {
148 	struct thread *td = curthread;
149 	struct proc *p = td->td_proc;
150 	struct file *fp;
151 	int error;
152 
153 	KKASSERT(p);
154 	error = holdsock(p->p_fd, s, &fp);
155 	if (error)
156 		return (error);
157 	error = sobind((struct socket *)fp->f_data, sa, td);
158 	fdrop(fp);
159 	return (error);
160 }
161 
162 /*
163  * bind_args(int s, caddr_t name, int namelen)
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 	error = kern_bind(uap->s, sa);
175 	FREE(sa, M_SONAME);
176 
177 	return (error);
178 }
179 
180 int
181 kern_listen(int s, int backlog)
182 {
183 	struct thread *td = curthread;
184 	struct proc *p = td->td_proc;
185 	struct file *fp;
186 	int error;
187 
188 	KKASSERT(p);
189 	error = holdsock(p->p_fd, s, &fp);
190 	if (error)
191 		return (error);
192 	error = solisten((struct socket *)fp->f_data, backlog, td);
193 	fdrop(fp);
194 	return(error);
195 }
196 
197 /*
198  * listen_args(int s, int backlog)
199  */
200 int
201 sys_listen(struct listen_args *uap)
202 {
203 	int error;
204 
205 	error = kern_listen(uap->s, uap->backlog);
206 	return (error);
207 }
208 
209 /*
210  * Returns the accepted socket as well.
211  */
212 static boolean_t
213 soaccept_predicate(struct netmsg *msg0)
214 {
215 	struct netmsg_so_notify *msg = (struct netmsg_so_notify *)msg0;
216 	struct socket *head = msg->nm_so;
217 
218 	if (head->so_error != 0) {
219 		msg->nm_netmsg.nm_lmsg.ms_error = head->so_error;
220 		return (TRUE);
221 	}
222 	if (!TAILQ_EMPTY(&head->so_comp)) {
223 		/* Abuse nm_so field as copy in/copy out parameter. XXX JH */
224 		msg->nm_so = TAILQ_FIRST(&head->so_comp);
225 		TAILQ_REMOVE(&head->so_comp, msg->nm_so, so_list);
226 		head->so_qlen--;
227 
228 		msg->nm_netmsg.nm_lmsg.ms_error = 0;
229 		return (TRUE);
230 	}
231 	if (head->so_state & SS_CANTRCVMORE) {
232 		msg->nm_netmsg.nm_lmsg.ms_error = ECONNABORTED;
233 		return (TRUE);
234 	}
235 	if (msg->nm_fflags & FNONBLOCK) {
236 		msg->nm_netmsg.nm_lmsg.ms_error = EWOULDBLOCK;
237 		return (TRUE);
238 	}
239 
240 	return (FALSE);
241 }
242 
243 /*
244  * The second argument to kern_accept() is a handle to a struct sockaddr.
245  * This allows kern_accept() to return a pointer to an allocated struct
246  * sockaddr which must be freed later with FREE().  The caller must
247  * initialize *name to NULL.
248  */
249 int
250 kern_accept(int s, int fflags, struct sockaddr **name, int *namelen, int *res)
251 {
252 	struct thread *td = curthread;
253 	struct proc *p = td->td_proc;
254 	struct file *lfp = NULL;
255 	struct file *nfp = NULL;
256 	struct sockaddr *sa;
257 	struct socket *head, *so;
258 	struct netmsg_so_notify msg;
259 	lwkt_port_t port;
260 	int fd;
261 	u_int fflag;		/* type must match fp->f_flag */
262 	int error, tmp;
263 
264 	*res = -1;
265 	if (name && namelen && *namelen < 0)
266 		return (EINVAL);
267 
268 	error = holdsock(p->p_fd, s, &lfp);
269 	if (error)
270 		return (error);
271 
272 	error = falloc(p, &nfp, &fd);
273 	if (error) {		/* Probably ran out of file descriptors. */
274 		fdrop(lfp);
275 		return (error);
276 	}
277 	head = (struct socket *)lfp->f_data;
278 	if ((head->so_options & SO_ACCEPTCONN) == 0) {
279 		error = EINVAL;
280 		goto done;
281 	}
282 
283 	if (fflags & O_FBLOCKING)
284 		fflags |= lfp->f_flag & ~FNONBLOCK;
285 	else if (fflags & O_FNONBLOCKING)
286 		fflags |= lfp->f_flag | FNONBLOCK;
287 	else
288 		fflags = lfp->f_flag;
289 
290 	/* optimize for uniprocessor case later XXX JH */
291 	port = head->so_proto->pr_mport(head, NULL, NULL, PRU_PRED);
292 	netmsg_init_abortable(&msg.nm_netmsg, &curthread->td_msgport,
293 			      0,
294 			      netmsg_so_notify,
295 			      netmsg_so_notify_doabort);
296 	msg.nm_predicate = soaccept_predicate;
297 	msg.nm_fflags = fflags;
298 	msg.nm_so = head;
299 	msg.nm_etype = NM_REVENT;
300 	error = lwkt_domsg(port, &msg.nm_netmsg.nm_lmsg, PCATCH);
301 	if (error)
302 		goto done;
303 
304 	/*
305 	 * At this point we have the connection that's ready to be accepted.
306 	 */
307 	so = msg.nm_so;
308 
309 	fflag = lfp->f_flag;
310 
311 	/* connection has been removed from the listen queue */
312 	KNOTE(&head->so_rcv.ssb_sel.si_note, 0);
313 
314 	so->so_state &= ~SS_COMP;
315 	so->so_head = NULL;
316 	if (head->so_sigio != NULL)
317 		fsetown(fgetown(head->so_sigio), &so->so_sigio);
318 
319 	nfp->f_type = DTYPE_SOCKET;
320 	nfp->f_flag = fflag;
321 	nfp->f_ops = &socketops;
322 	nfp->f_data = so;
323 	/* Sync socket nonblocking/async state with file flags */
324 	tmp = fflag & FNONBLOCK;
325 	(void) fo_ioctl(nfp, FIONBIO, (caddr_t)&tmp, p->p_ucred);
326 	tmp = fflag & FASYNC;
327 	(void) fo_ioctl(nfp, FIOASYNC, (caddr_t)&tmp, p->p_ucred);
328 
329 	sa = NULL;
330 	error = soaccept(so, &sa);
331 
332 	/*
333 	 * Set the returned name and namelen as applicable.  Set the returned
334 	 * namelen to 0 for older code which might ignore the return value
335 	 * from accept.
336 	 */
337 	if (error == 0) {
338 		if (sa && name && namelen) {
339 			if (*namelen > sa->sa_len)
340 				*namelen = sa->sa_len;
341 			*name = sa;
342 		} else {
343 			if (sa)
344 				FREE(sa, M_SONAME);
345 		}
346 	}
347 
348 done:
349 	/*
350 	 * If an error occured clear the reserved descriptor, else associate
351 	 * nfp with it.
352 	 *
353 	 * Note that *res is normally ignored if an error is returned but
354 	 * a syscall message will still have access to the result code.
355 	 */
356 	if (error) {
357 		fsetfd(p, NULL, fd);
358 	} else {
359 		*res = fd;
360 		fsetfd(p, nfp, fd);
361 	}
362 	fdrop(nfp);
363 	fdrop(lfp);
364 	return (error);
365 }
366 
367 /*
368  * accept(int s, caddr_t name, int *anamelen)
369  */
370 int
371 sys_accept(struct accept_args *uap)
372 {
373 	struct sockaddr *sa = NULL;
374 	int sa_len;
375 	int error;
376 
377 	if (uap->name) {
378 		error = copyin(uap->anamelen, &sa_len, sizeof(sa_len));
379 		if (error)
380 			return (error);
381 
382 		error = kern_accept(uap->s, 0, &sa, &sa_len, &uap->sysmsg_result);
383 
384 		if (error == 0)
385 			error = copyout(sa, uap->name, sa_len);
386 		if (error == 0) {
387 			error = copyout(&sa_len, uap->anamelen,
388 			    sizeof(*uap->anamelen));
389 		}
390 		if (sa)
391 			FREE(sa, M_SONAME);
392 	} else {
393 		error = kern_accept(uap->s, 0, NULL, 0, &uap->sysmsg_result);
394 	}
395 	return (error);
396 }
397 
398 /*
399  * extaccept(int s, int fflags, caddr_t name, int *anamelen)
400  */
401 int
402 sys_extaccept(struct extaccept_args *uap)
403 {
404 	struct sockaddr *sa = NULL;
405 	int sa_len;
406 	int error;
407 	int fflags = uap->flags & O_FMASK;
408 
409 	if (uap->name) {
410 		error = copyin(uap->anamelen, &sa_len, sizeof(sa_len));
411 		if (error)
412 			return (error);
413 
414 		error = kern_accept(uap->s, fflags, &sa, &sa_len, &uap->sysmsg_result);
415 
416 		if (error == 0)
417 			error = copyout(sa, uap->name, sa_len);
418 		if (error == 0) {
419 			error = copyout(&sa_len, uap->anamelen,
420 			    sizeof(*uap->anamelen));
421 		}
422 		if (sa)
423 			FREE(sa, M_SONAME);
424 	} else {
425 		error = kern_accept(uap->s, fflags, NULL, 0, &uap->sysmsg_result);
426 	}
427 	return (error);
428 }
429 
430 
431 /*
432  * Returns TRUE if predicate satisfied.
433  */
434 static boolean_t
435 soconnected_predicate(struct netmsg *msg0)
436 {
437 	struct netmsg_so_notify *msg = (struct netmsg_so_notify *)msg0;
438 	struct socket *so = msg->nm_so;
439 
440 	/* check predicate */
441 	if (!(so->so_state & SS_ISCONNECTING) || so->so_error != 0) {
442 		msg->nm_netmsg.nm_lmsg.ms_error = so->so_error;
443 		return (TRUE);
444 	}
445 
446 	return (FALSE);
447 }
448 
449 int
450 kern_connect(int s, int fflags, struct sockaddr *sa)
451 {
452 	struct thread *td = curthread;
453 	struct proc *p = td->td_proc;
454 	struct file *fp;
455 	struct socket *so;
456 	int error;
457 
458 	error = holdsock(p->p_fd, s, &fp);
459 	if (error)
460 		return (error);
461 	so = (struct socket *)fp->f_data;
462 
463 	if (fflags & O_FBLOCKING)
464 		/* fflags &= ~FNONBLOCK; */;
465 	else if (fflags & O_FNONBLOCKING)
466 		fflags |= FNONBLOCK;
467 	else
468 		fflags = fp->f_flag;
469 
470 	if ((fflags & FNONBLOCK) && (so->so_state & SS_ISCONNECTING)) {
471 		error = EALREADY;
472 		goto done;
473 	}
474 	error = soconnect(so, sa, td);
475 	if (error)
476 		goto bad;
477 	if ((fflags & FNONBLOCK) && (so->so_state & SS_ISCONNECTING)) {
478 		error = EINPROGRESS;
479 		goto done;
480 	}
481 	if ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) {
482 		struct netmsg_so_notify msg;
483 		lwkt_port_t port;
484 
485 		port = so->so_proto->pr_mport(so, sa, NULL, PRU_PRED);
486 		netmsg_init_abortable(&msg.nm_netmsg,
487 				      &curthread->td_msgport,
488 				      0,
489 				      netmsg_so_notify,
490 				      netmsg_so_notify_doabort);
491 		msg.nm_predicate = soconnected_predicate;
492 		msg.nm_so = so;
493 		msg.nm_etype = NM_REVENT;
494 		error = lwkt_domsg(port, &msg.nm_netmsg.nm_lmsg, PCATCH);
495 	}
496 	if (error == 0) {
497 		error = so->so_error;
498 		so->so_error = 0;
499 	}
500 bad:
501 	so->so_state &= ~SS_ISCONNECTING;
502 	if (error == ERESTART)
503 		error = EINTR;
504 done:
505 	fdrop(fp);
506 	return (error);
507 }
508 
509 /*
510  * connect_args(int s, caddr_t name, int namelen)
511  */
512 int
513 sys_connect(struct connect_args *uap)
514 {
515 	struct sockaddr *sa;
516 	int error;
517 
518 	error = getsockaddr(&sa, uap->name, uap->namelen);
519 	if (error)
520 		return (error);
521 	error = kern_connect(uap->s, 0, sa);
522 	FREE(sa, M_SONAME);
523 
524 	return (error);
525 }
526 
527 /*
528  * connect_args(int s, int fflags, caddr_t name, int namelen)
529  */
530 int
531 sys_extconnect(struct extconnect_args *uap)
532 {
533 	struct sockaddr *sa;
534 	int error;
535 	int fflags = uap->flags & O_FMASK;
536 
537 	error = getsockaddr(&sa, uap->name, uap->namelen);
538 	if (error)
539 		return (error);
540 	error = kern_connect(uap->s, fflags, sa);
541 	FREE(sa, M_SONAME);
542 
543 	return (error);
544 }
545 
546 int
547 kern_socketpair(int domain, int type, int protocol, int *sv)
548 {
549 	struct thread *td = curthread;
550 	struct proc *p = td->td_proc;
551 	struct file *fp1, *fp2;
552 	struct socket *so1, *so2;
553 	int fd1, fd2, error;
554 
555 	KKASSERT(p);
556 	error = socreate(domain, &so1, type, protocol, td);
557 	if (error)
558 		return (error);
559 	error = socreate(domain, &so2, type, protocol, td);
560 	if (error)
561 		goto free1;
562 	error = falloc(p, &fp1, &fd1);
563 	if (error)
564 		goto free2;
565 	sv[0] = fd1;
566 	fp1->f_data = so1;
567 	error = falloc(p, &fp2, &fd2);
568 	if (error)
569 		goto free3;
570 	fp2->f_data = so2;
571 	sv[1] = fd2;
572 	error = soconnect2(so1, so2);
573 	if (error)
574 		goto free4;
575 	if (type == SOCK_DGRAM) {
576 		/*
577 		 * Datagram socket connection is asymmetric.
578 		 */
579 		 error = soconnect2(so2, so1);
580 		 if (error)
581 			goto free4;
582 	}
583 	fp1->f_type = fp2->f_type = DTYPE_SOCKET;
584 	fp1->f_flag = fp2->f_flag = FREAD|FWRITE;
585 	fp1->f_ops = fp2->f_ops = &socketops;
586 	fsetfd(p, fp1, fd1);
587 	fsetfd(p, fp2, fd2);
588 	fdrop(fp1);
589 	fdrop(fp2);
590 	return (error);
591 free4:
592 	fsetfd(p, NULL, fd2);
593 	fdrop(fp2);
594 free3:
595 	fsetfd(p, NULL, fd1);
596 	fdrop(fp1);
597 free2:
598 	(void)soclose(so2, 0);
599 free1:
600 	(void)soclose(so1, 0);
601 	return (error);
602 }
603 
604 /*
605  * socketpair(int domain, int type, int protocol, int *rsv)
606  */
607 int
608 sys_socketpair(struct socketpair_args *uap)
609 {
610 	int error, sockv[2];
611 
612 	error = kern_socketpair(uap->domain, uap->type, uap->protocol, sockv);
613 
614 	if (error == 0)
615 		error = copyout(sockv, uap->rsv, sizeof(sockv));
616 	return (error);
617 }
618 
619 int
620 kern_sendmsg(int s, struct sockaddr *sa, struct uio *auio,
621     struct mbuf *control, int flags, int *res)
622 {
623 	struct thread *td = curthread;
624 	struct lwp *lp = td->td_lwp;
625 	struct proc *p = td->td_proc;
626 	struct file *fp;
627 	int len, error;
628 	struct socket *so;
629 #ifdef KTRACE
630 	struct iovec *ktriov = NULL;
631 	struct uio ktruio;
632 #endif
633 
634 	error = holdsock(p->p_fd, s, &fp);
635 	if (error)
636 		return (error);
637 	if (auio->uio_resid < 0) {
638 		error = EINVAL;
639 		goto done;
640 	}
641 #ifdef KTRACE
642 	if (KTRPOINT(td, KTR_GENIO)) {
643 		int iovlen = auio->uio_iovcnt * sizeof (struct iovec);
644 
645 		MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK);
646 		bcopy((caddr_t)auio->uio_iov, (caddr_t)ktriov, iovlen);
647 		ktruio = *auio;
648 	}
649 #endif
650 	len = auio->uio_resid;
651 	so = (struct socket *)fp->f_data;
652 	if ((flags & (MSG_FNONBLOCKING|MSG_FBLOCKING)) == 0) {
653 		if (fp->f_flag & FNONBLOCK)
654 			flags |= MSG_FNONBLOCKING;
655 	}
656 	error = so_pru_sosend(so, sa, auio, NULL, control, flags, td);
657 	if (error) {
658 		if (auio->uio_resid != len && (error == ERESTART ||
659 		    error == EINTR || error == EWOULDBLOCK))
660 			error = 0;
661 		if (error == EPIPE)
662 			lwpsignal(p, lp, SIGPIPE);
663 	}
664 #ifdef KTRACE
665 	if (ktriov != NULL) {
666 		if (error == 0) {
667 			ktruio.uio_iov = ktriov;
668 			ktruio.uio_resid = len - auio->uio_resid;
669 			ktrgenio(lp, s, UIO_WRITE, &ktruio, error);
670 		}
671 		FREE(ktriov, M_TEMP);
672 	}
673 #endif
674 	if (error == 0)
675 		*res  = len - auio->uio_resid;
676 done:
677 	fdrop(fp);
678 	return (error);
679 }
680 
681 /*
682  * sendto_args(int s, caddr_t buf, size_t len, int flags, caddr_t to, int tolen)
683  */
684 int
685 sys_sendto(struct sendto_args *uap)
686 {
687 	struct thread *td = curthread;
688 	struct uio auio;
689 	struct iovec aiov;
690 	struct sockaddr *sa = NULL;
691 	int error;
692 
693 	if (uap->to) {
694 		error = getsockaddr(&sa, uap->to, uap->tolen);
695 		if (error)
696 			return (error);
697 	}
698 	aiov.iov_base = uap->buf;
699 	aiov.iov_len = uap->len;
700 	auio.uio_iov = &aiov;
701 	auio.uio_iovcnt = 1;
702 	auio.uio_offset = 0;
703 	auio.uio_resid = uap->len;
704 	auio.uio_segflg = UIO_USERSPACE;
705 	auio.uio_rw = UIO_WRITE;
706 	auio.uio_td = td;
707 
708 	error = kern_sendmsg(uap->s, sa, &auio, NULL, uap->flags,
709 	    &uap->sysmsg_result);
710 
711 	if (sa)
712 		FREE(sa, M_SONAME);
713 	return (error);
714 }
715 
716 /*
717  * sendmsg_args(int s, caddr_t msg, int flags)
718  */
719 int
720 sys_sendmsg(struct sendmsg_args *uap)
721 {
722 	struct thread *td = curthread;
723 	struct msghdr msg;
724 	struct uio auio;
725 	struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
726 	struct sockaddr *sa = NULL;
727 	struct mbuf *control = NULL;
728 	int error;
729 
730 	error = copyin(uap->msg, (caddr_t)&msg, sizeof(msg));
731 	if (error)
732 		return (error);
733 
734 	/*
735 	 * Conditionally copyin msg.msg_name.
736 	 */
737 	if (msg.msg_name) {
738 		error = getsockaddr(&sa, msg.msg_name, msg.msg_namelen);
739 		if (error)
740 			return (error);
741 	}
742 
743 	/*
744 	 * Populate auio.
745 	 */
746 	error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen,
747 			     &auio.uio_resid);
748 	if (error)
749 		goto cleanup2;
750 	auio.uio_iov = iov;
751 	auio.uio_iovcnt = msg.msg_iovlen;
752 	auio.uio_offset = 0;
753 	auio.uio_segflg = UIO_USERSPACE;
754 	auio.uio_rw = UIO_WRITE;
755 	auio.uio_td = td;
756 
757 	/*
758 	 * Conditionally copyin msg.msg_control.
759 	 */
760 	if (msg.msg_control) {
761 		if (msg.msg_controllen < sizeof(struct cmsghdr) ||
762 		    msg.msg_controllen > MLEN) {
763 			error = EINVAL;
764 			goto cleanup;
765 		}
766 		control = m_get(MB_WAIT, MT_CONTROL);
767 		if (control == NULL) {
768 			error = ENOBUFS;
769 			goto cleanup;
770 		}
771 		control->m_len = msg.msg_controllen;
772 		error = copyin(msg.msg_control, mtod(control, caddr_t),
773 		    msg.msg_controllen);
774 		if (error) {
775 			m_free(control);
776 			goto cleanup;
777 		}
778 	}
779 
780 	error = kern_sendmsg(uap->s, sa, &auio, control, uap->flags,
781 	    &uap->sysmsg_result);
782 
783 cleanup:
784 	iovec_free(&iov, aiov);
785 cleanup2:
786 	if (sa)
787 		FREE(sa, M_SONAME);
788 	return (error);
789 }
790 
791 /*
792  * kern_recvmsg() takes a handle to sa and control.  If the handle is non-
793  * null, it returns a dynamically allocated struct sockaddr and an mbuf.
794  * Don't forget to FREE() and m_free() these if they are returned.
795  */
796 int
797 kern_recvmsg(int s, struct sockaddr **sa, struct uio *auio,
798     struct mbuf **control, int *flags, int *res)
799 {
800 	struct thread *td = curthread;
801 	struct proc *p = td->td_proc;
802 	struct file *fp;
803 	int len, error;
804 	int lflags;
805 	struct socket *so;
806 #ifdef KTRACE
807 	struct iovec *ktriov = NULL;
808 	struct uio ktruio;
809 #endif
810 
811 	error = holdsock(p->p_fd, s, &fp);
812 	if (error)
813 		return (error);
814 	if (auio->uio_resid < 0) {
815 		error = EINVAL;
816 		goto done;
817 	}
818 #ifdef KTRACE
819 	if (KTRPOINT(td, KTR_GENIO)) {
820 		int iovlen = auio->uio_iovcnt * sizeof (struct iovec);
821 
822 		MALLOC(ktriov, struct iovec *, iovlen, M_TEMP, M_WAITOK);
823 		bcopy(auio->uio_iov, ktriov, iovlen);
824 		ktruio = *auio;
825 	}
826 #endif
827 	len = auio->uio_resid;
828 	so = (struct socket *)fp->f_data;
829 
830 	if (flags == NULL || (*flags & (MSG_FNONBLOCKING|MSG_FBLOCKING)) == 0) {
831 		if (fp->f_flag & FNONBLOCK) {
832 			if (flags) {
833 				*flags |= MSG_FNONBLOCKING;
834 			} else {
835 				lflags = MSG_FNONBLOCKING;
836 				flags = &lflags;
837 			}
838 		}
839 	}
840 
841 	error = so_pru_soreceive(so, sa, auio, NULL, control, flags);
842 	if (error) {
843 		if (auio->uio_resid != len && (error == ERESTART ||
844 		    error == EINTR || error == EWOULDBLOCK))
845 			error = 0;
846 	}
847 #ifdef KTRACE
848 	if (ktriov != NULL) {
849 		if (error == 0) {
850 			ktruio.uio_iov = ktriov;
851 			ktruio.uio_resid = len - auio->uio_resid;
852 			ktrgenio(td->td_lwp, s, UIO_READ, &ktruio, error);
853 		}
854 		FREE(ktriov, M_TEMP);
855 	}
856 #endif
857 	if (error == 0)
858 		*res = len - auio->uio_resid;
859 done:
860 	fdrop(fp);
861 	return (error);
862 }
863 
864 /*
865  * recvfrom_args(int s, caddr_t buf, size_t len, int flags,
866  *			caddr_t from, int *fromlenaddr)
867  */
868 int
869 sys_recvfrom(struct recvfrom_args *uap)
870 {
871 	struct thread *td = curthread;
872 	struct uio auio;
873 	struct iovec aiov;
874 	struct sockaddr *sa = NULL;
875 	int error, fromlen;
876 
877 	if (uap->from && uap->fromlenaddr) {
878 		error = copyin(uap->fromlenaddr, &fromlen, sizeof(fromlen));
879 		if (error)
880 			return (error);
881 		if (fromlen < 0)
882 			return (EINVAL);
883 	} else {
884 		fromlen = 0;
885 	}
886 	aiov.iov_base = uap->buf;
887 	aiov.iov_len = uap->len;
888 	auio.uio_iov = &aiov;
889 	auio.uio_iovcnt = 1;
890 	auio.uio_offset = 0;
891 	auio.uio_resid = uap->len;
892 	auio.uio_segflg = UIO_USERSPACE;
893 	auio.uio_rw = UIO_READ;
894 	auio.uio_td = td;
895 
896 	error = kern_recvmsg(uap->s, uap->from ? &sa : NULL, &auio, NULL,
897 	    &uap->flags, &uap->sysmsg_result);
898 
899 	if (error == 0 && uap->from) {
900 		/* note: sa may still be NULL */
901 		if (sa) {
902 			fromlen = MIN(fromlen, sa->sa_len);
903 			error = copyout(sa, uap->from, fromlen);
904 		} else {
905 			fromlen = 0;
906 		}
907 		if (error == 0) {
908 			error = copyout(&fromlen, uap->fromlenaddr,
909 					sizeof(fromlen));
910 		}
911 	}
912 	if (sa)
913 		FREE(sa, M_SONAME);
914 
915 	return (error);
916 }
917 
918 /*
919  * recvmsg_args(int s, struct msghdr *msg, int flags)
920  */
921 int
922 sys_recvmsg(struct recvmsg_args *uap)
923 {
924 	struct thread *td = curthread;
925 	struct msghdr msg;
926 	struct uio auio;
927 	struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
928 	struct mbuf *m, *control = NULL;
929 	struct sockaddr *sa = NULL;
930 	caddr_t ctlbuf;
931 	socklen_t *ufromlenp, *ucontrollenp;
932 	int error, fromlen, controllen, len, flags, *uflagsp;
933 
934 	/*
935 	 * This copyin handles everything except the iovec.
936 	 */
937 	error = copyin(uap->msg, &msg, sizeof(msg));
938 	if (error)
939 		return (error);
940 
941 	if (msg.msg_name && msg.msg_namelen < 0)
942 		return (EINVAL);
943 	if (msg.msg_control && msg.msg_controllen < 0)
944 		return (EINVAL);
945 
946 	ufromlenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr,
947 	    msg_namelen));
948 	ucontrollenp = (socklen_t *)((caddr_t)uap->msg + offsetof(struct msghdr,
949 	    msg_controllen));
950 	uflagsp = (int *)((caddr_t)uap->msg + offsetof(struct msghdr,
951 	    msg_flags));
952 
953 	/*
954 	 * Populate auio.
955 	 */
956 	error = iovec_copyin(msg.msg_iov, &iov, aiov, msg.msg_iovlen,
957 			     &auio.uio_resid);
958 	if (error)
959 		return (error);
960 	auio.uio_iov = iov;
961 	auio.uio_iovcnt = msg.msg_iovlen;
962 	auio.uio_offset = 0;
963 	auio.uio_segflg = UIO_USERSPACE;
964 	auio.uio_rw = UIO_READ;
965 	auio.uio_td = td;
966 
967 	flags = uap->flags;
968 
969 	error = kern_recvmsg(uap->s, msg.msg_name ? &sa : NULL, &auio,
970 	    msg.msg_control ? &control : NULL, &flags, &uap->sysmsg_result);
971 
972 	/*
973 	 * Conditionally copyout the name and populate the namelen field.
974 	 */
975 	if (error == 0 && msg.msg_name) {
976 		/* note: sa may still be NULL */
977 		if (sa != NULL) {
978 			fromlen = MIN(msg.msg_namelen, sa->sa_len);
979 			error = copyout(sa, msg.msg_name, fromlen);
980 		} else
981 			fromlen = 0;
982 		if (error == 0)
983 			error = copyout(&fromlen, ufromlenp,
984 			    sizeof(*ufromlenp));
985 	}
986 
987 	/*
988 	 * Copyout msg.msg_control and msg.msg_controllen.
989 	 */
990 	if (error == 0 && msg.msg_control) {
991 		len = msg.msg_controllen;
992 		m = control;
993 		ctlbuf = (caddr_t)msg.msg_control;
994 
995 		while(m && len > 0) {
996 			unsigned int tocopy;
997 
998 			if (len >= m->m_len) {
999 				tocopy = m->m_len;
1000 			} else {
1001 				msg.msg_flags |= MSG_CTRUNC;
1002 				tocopy = len;
1003 			}
1004 
1005 			error = copyout(mtod(m, caddr_t), ctlbuf, tocopy);
1006 			if (error)
1007 				goto cleanup;
1008 
1009 			ctlbuf += tocopy;
1010 			len -= tocopy;
1011 			m = m->m_next;
1012 		}
1013 		controllen = ctlbuf - (caddr_t)msg.msg_control;
1014 		error = copyout(&controllen, ucontrollenp,
1015 		    sizeof(*ucontrollenp));
1016 	}
1017 
1018 	if (error == 0)
1019 		error = copyout(&flags, uflagsp, sizeof(*uflagsp));
1020 
1021 cleanup:
1022 	if (sa)
1023 		FREE(sa, M_SONAME);
1024 	iovec_free(&iov, aiov);
1025 	if (control)
1026 		m_freem(control);
1027 	return (error);
1028 }
1029 
1030 /*
1031  * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an
1032  * in kernel pointer instead of a userland pointer.  This allows us
1033  * to manipulate socket options in the emulation code.
1034  */
1035 int
1036 kern_setsockopt(int s, struct sockopt *sopt)
1037 {
1038 	struct thread *td = curthread;
1039 	struct proc *p = td->td_proc;
1040 	struct file *fp;
1041 	int error;
1042 
1043 	if (sopt->sopt_val == 0 && sopt->sopt_valsize != 0)
1044 		return (EFAULT);
1045 	if (sopt->sopt_valsize < 0)
1046 		return (EINVAL);
1047 
1048 	error = holdsock(p->p_fd, s, &fp);
1049 	if (error)
1050 		return (error);
1051 
1052 	error = sosetopt((struct socket *)fp->f_data, sopt);
1053 	fdrop(fp);
1054 	return (error);
1055 }
1056 
1057 /*
1058  * setsockopt_args(int s, int level, int name, caddr_t val, int valsize)
1059  */
1060 int
1061 sys_setsockopt(struct setsockopt_args *uap)
1062 {
1063 	struct thread *td = curthread;
1064 	struct sockopt sopt;
1065 	int error;
1066 
1067 	sopt.sopt_level = uap->level;
1068 	sopt.sopt_name = uap->name;
1069 	sopt.sopt_val = uap->val;
1070 	sopt.sopt_valsize = uap->valsize;
1071 	sopt.sopt_td = td;
1072 
1073 	error = kern_setsockopt(uap->s, &sopt);
1074 	return(error);
1075 }
1076 
1077 /*
1078  * If sopt->sopt_td == NULL, then sopt->sopt_val is treated as an
1079  * in kernel pointer instead of a userland pointer.  This allows us
1080  * to manipulate socket options in the emulation code.
1081  */
1082 int
1083 kern_getsockopt(int s, struct sockopt *sopt)
1084 {
1085 	struct thread *td = curthread;
1086 	struct proc *p = td->td_proc;
1087 	struct file *fp;
1088 	int error;
1089 
1090 	if (sopt->sopt_val == 0 && sopt->sopt_valsize != 0)
1091 		return (EFAULT);
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 = sogetopt((struct socket *)fp->f_data, sopt);
1100 	fdrop(fp);
1101 	return (error);
1102 }
1103 
1104 /*
1105  * getsockopt_Args(int s, int level, int name, caddr_t val, int *avalsize)
1106  */
1107 int
1108 sys_getsockopt(struct getsockopt_args *uap)
1109 {
1110 	struct thread *td = curthread;
1111 	struct	sockopt sopt;
1112 	int	error, valsize;
1113 
1114 	if (uap->val) {
1115 		error = copyin(uap->avalsize, &valsize, sizeof(valsize));
1116 		if (error)
1117 			return (error);
1118 		if (valsize < 0)
1119 			return (EINVAL);
1120 	} else {
1121 		valsize = 0;
1122 	}
1123 
1124 	sopt.sopt_level = uap->level;
1125 	sopt.sopt_name = uap->name;
1126 	sopt.sopt_val = uap->val;
1127 	sopt.sopt_valsize = valsize;
1128 	sopt.sopt_td = td;
1129 
1130 	error = kern_getsockopt(uap->s, &sopt);
1131 	if (error == 0) {
1132 		valsize = sopt.sopt_valsize;
1133 		error = copyout(&valsize, uap->avalsize, sizeof(valsize));
1134 	}
1135 	return (error);
1136 }
1137 
1138 /*
1139  * The second argument to kern_getsockname() is a handle to a struct sockaddr.
1140  * This allows kern_getsockname() to return a pointer to an allocated struct
1141  * sockaddr which must be freed later with FREE().  The caller must
1142  * initialize *name to NULL.
1143  */
1144 int
1145 kern_getsockname(int s, struct sockaddr **name, int *namelen)
1146 {
1147 	struct thread *td = curthread;
1148 	struct proc *p = td->td_proc;
1149 	struct file *fp;
1150 	struct socket *so;
1151 	struct sockaddr *sa = NULL;
1152 	int error;
1153 
1154 	error = holdsock(p->p_fd, s, &fp);
1155 	if (error)
1156 		return (error);
1157 	if (*namelen < 0) {
1158 		fdrop(fp);
1159 		return (EINVAL);
1160 	}
1161 	so = (struct socket *)fp->f_data;
1162 	error = so_pru_sockaddr(so, &sa);
1163 	if (error == 0) {
1164 		if (sa == 0) {
1165 			*namelen = 0;
1166 		} else {
1167 			*namelen = MIN(*namelen, sa->sa_len);
1168 			*name = sa;
1169 		}
1170 	}
1171 
1172 	fdrop(fp);
1173 	return (error);
1174 }
1175 
1176 /*
1177  * getsockname_args(int fdes, caddr_t asa, int *alen)
1178  *
1179  * Get socket name.
1180  */
1181 int
1182 sys_getsockname(struct getsockname_args *uap)
1183 {
1184 	struct sockaddr *sa = NULL;
1185 	int error, sa_len;
1186 
1187 	error = copyin(uap->alen, &sa_len, sizeof(sa_len));
1188 	if (error)
1189 		return (error);
1190 
1191 	error = kern_getsockname(uap->fdes, &sa, &sa_len);
1192 
1193 	if (error == 0)
1194 		error = copyout(sa, uap->asa, sa_len);
1195 	if (error == 0)
1196 		error = copyout(&sa_len, uap->alen, sizeof(*uap->alen));
1197 	if (sa)
1198 		FREE(sa, M_SONAME);
1199 	return (error);
1200 }
1201 
1202 /*
1203  * The second argument to kern_getpeername() is a handle to a struct sockaddr.
1204  * This allows kern_getpeername() to return a pointer to an allocated struct
1205  * sockaddr which must be freed later with FREE().  The caller must
1206  * initialize *name to NULL.
1207  */
1208 int
1209 kern_getpeername(int s, struct sockaddr **name, int *namelen)
1210 {
1211 	struct thread *td = curthread;
1212 	struct proc *p = td->td_proc;
1213 	struct file *fp;
1214 	struct socket *so;
1215 	struct sockaddr *sa = NULL;
1216 	int error;
1217 
1218 	error = holdsock(p->p_fd, s, &fp);
1219 	if (error)
1220 		return (error);
1221 	if (*namelen < 0) {
1222 		fdrop(fp);
1223 		return (EINVAL);
1224 	}
1225 	so = (struct socket *)fp->f_data;
1226 	if ((so->so_state & (SS_ISCONNECTED|SS_ISCONFIRMING)) == 0) {
1227 		fdrop(fp);
1228 		return (ENOTCONN);
1229 	}
1230 	error = so_pru_peeraddr(so, &sa);
1231 	if (error == 0) {
1232 		if (sa == 0) {
1233 			*namelen = 0;
1234 		} else {
1235 			*namelen = MIN(*namelen, sa->sa_len);
1236 			*name = sa;
1237 		}
1238 	}
1239 
1240 	fdrop(fp);
1241 	return (error);
1242 }
1243 
1244 /*
1245  * getpeername_args(int fdes, caddr_t asa, int *alen)
1246  *
1247  * Get name of peer for connected socket.
1248  */
1249 int
1250 sys_getpeername(struct getpeername_args *uap)
1251 {
1252 	struct sockaddr *sa = NULL;
1253 	int error, sa_len;
1254 
1255 	error = copyin(uap->alen, &sa_len, sizeof(sa_len));
1256 	if (error)
1257 		return (error);
1258 
1259 	error = kern_getpeername(uap->fdes, &sa, &sa_len);
1260 
1261 	if (error == 0)
1262 		error = copyout(sa, uap->asa, sa_len);
1263 	if (error == 0)
1264 		error = copyout(&sa_len, uap->alen, sizeof(*uap->alen));
1265 	if (sa)
1266 		FREE(sa, M_SONAME);
1267 	return (error);
1268 }
1269 
1270 int
1271 getsockaddr(struct sockaddr **namp, caddr_t uaddr, size_t len)
1272 {
1273 	struct sockaddr *sa;
1274 	int error;
1275 
1276 	*namp = NULL;
1277 	if (len > SOCK_MAXADDRLEN)
1278 		return ENAMETOOLONG;
1279 	if (len < offsetof(struct sockaddr, sa_data[0]))
1280 		return EDOM;
1281 	MALLOC(sa, struct sockaddr *, len, M_SONAME, M_WAITOK);
1282 	error = copyin(uaddr, sa, len);
1283 	if (error) {
1284 		FREE(sa, M_SONAME);
1285 	} else {
1286 #if BYTE_ORDER != BIG_ENDIAN
1287 		/*
1288 		 * The bind(), connect(), and sendto() syscalls were not
1289 		 * versioned for COMPAT_43.  Thus, this check must stay.
1290 		 */
1291 		if (sa->sa_family == 0 && sa->sa_len < AF_MAX)
1292 			sa->sa_family = sa->sa_len;
1293 #endif
1294 		sa->sa_len = len;
1295 		*namp = sa;
1296 	}
1297 	return error;
1298 }
1299 
1300 /*
1301  * Detach a mapped page and release resources back to the system.
1302  * We must release our wiring and if the object is ripped out
1303  * from under the vm_page we become responsible for freeing the
1304  * page.  These routines must be MPSAFE.
1305  *
1306  * XXX HACK XXX TEMPORARY UNTIL WE IMPLEMENT EXT MBUF REFERENCE COUNTING
1307  *
1308  * XXX vm_page_*() routines are not MPSAFE yet, the MP lock is required.
1309  */
1310 static void
1311 sf_buf_mref(void *arg)
1312 {
1313 	struct sfbuf_mref *sfm = arg;
1314 
1315 	/*
1316 	 * We must already hold a ref so there is no race to 0, just
1317 	 * atomically increment the count.
1318 	 */
1319 	atomic_add_int(&sfm->mref_count, 1);
1320 }
1321 
1322 static void
1323 sf_buf_mfree(void *arg)
1324 {
1325 	struct sfbuf_mref *sfm = arg;
1326 	vm_page_t m;
1327 
1328 	KKASSERT(sfm->mref_count > 0);
1329 	if (sfm->mref_count == 1) {
1330 		/*
1331 		 * We are the only holder so no further locking is required,
1332 		 * the sfbuf can simply be freed.
1333 		 */
1334 		sfm->mref_count = 0;
1335 		goto freeit;
1336 	} else {
1337 		/*
1338 		 * There may be other holders, we must obtain the serializer
1339 		 * to protect against a sf_buf_mfree() race to 0.  An atomic
1340 		 * operation is still required for races against
1341 		 * sf_buf_mref().
1342 		 *
1343 		 * XXX vm_page_*() and SFBUF routines not MPSAFE yet.
1344 		 */
1345 		lwkt_serialize_enter(&sfm->serializer);
1346 		atomic_subtract_int(&sfm->mref_count, 1);
1347 		if (sfm->mref_count == 0) {
1348 			lwkt_serialize_exit(&sfm->serializer);
1349 freeit:
1350 			get_mplock();
1351 			crit_enter();
1352 			m = sf_buf_page(sfm->sf);
1353 			sf_buf_free(sfm->sf);
1354 			vm_page_unwire(m, 0);
1355 			if (m->wire_count == 0 && m->object == NULL)
1356 				vm_page_try_to_free(m);
1357 			crit_exit();
1358 			rel_mplock();
1359 			kfree(sfm, M_SENDFILE);
1360 		} else {
1361 			lwkt_serialize_exit(&sfm->serializer);
1362 		}
1363 	}
1364 }
1365 
1366 /*
1367  * sendfile(2).
1368  * int sendfile(int fd, int s, off_t offset, size_t nbytes,
1369  *	 struct sf_hdtr *hdtr, off_t *sbytes, int flags)
1370  *
1371  * Send a file specified by 'fd' and starting at 'offset' to a socket
1372  * specified by 's'. Send only 'nbytes' of the file or until EOF if
1373  * nbytes == 0. Optionally add a header and/or trailer to the socket
1374  * output. If specified, write the total number of bytes sent into *sbytes.
1375  *
1376  * In FreeBSD kern/uipc_syscalls.c,v 1.103, a bug was fixed that caused
1377  * the headers to count against the remaining bytes to be sent from
1378  * the file descriptor.  We may wish to implement a compatibility syscall
1379  * in the future.
1380  */
1381 int
1382 sys_sendfile(struct sendfile_args *uap)
1383 {
1384 	struct thread *td = curthread;
1385 	struct proc *p = td->td_proc;
1386 	struct file *fp;
1387 	struct vnode *vp = NULL;
1388 	struct sf_hdtr hdtr;
1389 	struct iovec aiov[UIO_SMALLIOV], *iov = NULL;
1390 	struct uio auio;
1391 	struct mbuf *mheader = NULL;
1392 	off_t hdtr_size = 0, sbytes;
1393 	int error, hbytes = 0, tbytes;
1394 
1395 	KKASSERT(p);
1396 
1397 	/*
1398 	 * Do argument checking. Must be a regular file in, stream
1399 	 * type and connected socket out, positive offset.
1400 	 */
1401 	fp = holdfp(p->p_fd, uap->fd, FREAD);
1402 	if (fp == NULL) {
1403 		return (EBADF);
1404 	}
1405 	if (fp->f_type != DTYPE_VNODE) {
1406 		fdrop(fp);
1407 		return (EINVAL);
1408 	}
1409 	vp = (struct vnode *)fp->f_data;
1410 	vref(vp);
1411 	fdrop(fp);
1412 
1413 	/*
1414 	 * If specified, get the pointer to the sf_hdtr struct for
1415 	 * any headers/trailers.
1416 	 */
1417 	if (uap->hdtr) {
1418 		error = copyin(uap->hdtr, &hdtr, sizeof(hdtr));
1419 		if (error)
1420 			goto done;
1421 		/*
1422 		 * Send any headers.
1423 		 */
1424 		if (hdtr.headers) {
1425 			error = iovec_copyin(hdtr.headers, &iov, aiov,
1426 					     hdtr.hdr_cnt, &hbytes);
1427 			if (error)
1428 				goto done;
1429 			auio.uio_iov = iov;
1430 			auio.uio_iovcnt = hdtr.hdr_cnt;
1431 			auio.uio_offset = 0;
1432 			auio.uio_segflg = UIO_USERSPACE;
1433 			auio.uio_rw = UIO_WRITE;
1434 			auio.uio_td = td;
1435 			auio.uio_resid = hbytes;
1436 
1437 			mheader = m_uiomove(&auio);
1438 
1439 			iovec_free(&iov, aiov);
1440 			if (mheader == NULL)
1441 				goto done;
1442 		}
1443 	}
1444 
1445 	error = kern_sendfile(vp, uap->s, uap->offset, uap->nbytes, mheader,
1446 	    &sbytes, uap->flags);
1447 	if (error)
1448 		goto done;
1449 
1450 	/*
1451 	 * Send trailers. Wimp out and use writev(2).
1452 	 */
1453 	if (uap->hdtr != NULL && hdtr.trailers != NULL) {
1454 		error = iovec_copyin(hdtr.trailers, &iov, aiov,
1455 				     hdtr.trl_cnt, &auio.uio_resid);
1456 		if (error)
1457 			goto done;
1458 		auio.uio_iov = iov;
1459 		auio.uio_iovcnt = hdtr.trl_cnt;
1460 		auio.uio_offset = 0;
1461 		auio.uio_segflg = UIO_USERSPACE;
1462 		auio.uio_rw = UIO_WRITE;
1463 		auio.uio_td = td;
1464 
1465 		error = kern_sendmsg(uap->s, NULL, &auio, NULL, 0, &tbytes);
1466 
1467 		iovec_free(&iov, aiov);
1468 		if (error)
1469 			goto done;
1470 		hdtr_size += tbytes;	/* trailer bytes successfully sent */
1471 	}
1472 
1473 done:
1474 	if (uap->sbytes != NULL) {
1475 		sbytes += hdtr_size;
1476 		copyout(&sbytes, uap->sbytes, sizeof(off_t));
1477 	}
1478 	if (vp)
1479 		vrele(vp);
1480 	return (error);
1481 }
1482 
1483 int
1484 kern_sendfile(struct vnode *vp, int sfd, off_t offset, size_t nbytes,
1485     struct mbuf *mheader, off_t *sbytes, int flags)
1486 {
1487 	struct thread *td = curthread;
1488 	struct proc *p = td->td_proc;
1489 	struct vm_object *obj;
1490 	struct socket *so;
1491 	struct file *fp;
1492 	struct mbuf *m;
1493 	struct sf_buf *sf;
1494 	struct sfbuf_mref *sfm;
1495 	struct vm_page *pg;
1496 	off_t off, xfsize;
1497 	off_t hbytes = 0;
1498 	int error = 0;
1499 
1500 	if (vp->v_type != VREG) {
1501 		error = EINVAL;
1502 		goto done0;
1503 	}
1504 	if ((obj = vp->v_object) == NULL) {
1505 		error = EINVAL;
1506 		goto done0;
1507 	}
1508 	error = holdsock(p->p_fd, sfd, &fp);
1509 	if (error)
1510 		goto done0;
1511 	so = (struct socket *)fp->f_data;
1512 	if (so->so_type != SOCK_STREAM) {
1513 		error = EINVAL;
1514 		goto done;
1515 	}
1516 	if ((so->so_state & SS_ISCONNECTED) == 0) {
1517 		error = ENOTCONN;
1518 		goto done;
1519 	}
1520 	if (offset < 0) {
1521 		error = EINVAL;
1522 		goto done;
1523 	}
1524 
1525 	*sbytes = 0;
1526 	/*
1527 	 * Protect against multiple writers to the socket.
1528 	 */
1529 	ssb_lock(&so->so_snd, M_WAITOK);
1530 
1531 	/*
1532 	 * Loop through the pages in the file, starting with the requested
1533 	 * offset. Get a file page (do I/O if necessary), map the file page
1534 	 * into an sf_buf, attach an mbuf header to the sf_buf, and queue
1535 	 * it on the socket.
1536 	 */
1537 	for (off = offset; ; off += xfsize, *sbytes += xfsize + hbytes) {
1538 		vm_pindex_t pindex;
1539 		vm_offset_t pgoff;
1540 
1541 		pindex = OFF_TO_IDX(off);
1542 retry_lookup:
1543 		/*
1544 		 * Calculate the amount to transfer. Not to exceed a page,
1545 		 * the EOF, or the passed in nbytes.
1546 		 */
1547 		xfsize = vp->v_filesize - off;
1548 		if (xfsize > PAGE_SIZE)
1549 			xfsize = PAGE_SIZE;
1550 		pgoff = (vm_offset_t)(off & PAGE_MASK);
1551 		if (PAGE_SIZE - pgoff < xfsize)
1552 			xfsize = PAGE_SIZE - pgoff;
1553 		if (nbytes && xfsize > (nbytes - *sbytes))
1554 			xfsize = nbytes - *sbytes;
1555 		if (xfsize <= 0)
1556 			break;
1557 		/*
1558 		 * Optimize the non-blocking case by looking at the socket space
1559 		 * before going to the extra work of constituting the sf_buf.
1560 		 */
1561 		if ((fp->f_flag & FNONBLOCK) && ssb_space(&so->so_snd) <= 0) {
1562 			if (so->so_state & SS_CANTSENDMORE)
1563 				error = EPIPE;
1564 			else
1565 				error = EAGAIN;
1566 			ssb_unlock(&so->so_snd);
1567 			goto done;
1568 		}
1569 		/*
1570 		 * Attempt to look up the page.
1571 		 *
1572 		 *	Allocate if not found, wait and loop if busy, then
1573 		 *	wire the page.  critical section protection is
1574 		 * 	required to maintain the object association (an
1575 		 *	interrupt can free the page) through to the
1576 		 *	vm_page_wire() call.
1577 		 */
1578 		crit_enter();
1579 		pg = vm_page_lookup(obj, pindex);
1580 		if (pg == NULL) {
1581 			pg = vm_page_alloc(obj, pindex, VM_ALLOC_NORMAL);
1582 			if (pg == NULL) {
1583 				vm_wait();
1584 				crit_exit();
1585 				goto retry_lookup;
1586 			}
1587 			vm_page_wakeup(pg);
1588 		} else if (vm_page_sleep_busy(pg, TRUE, "sfpbsy")) {
1589 			crit_exit();
1590 			goto retry_lookup;
1591 		}
1592 		vm_page_wire(pg);
1593 		crit_exit();
1594 
1595 		/*
1596 		 * If page is not valid for what we need, initiate I/O
1597 		 */
1598 
1599 		if (!pg->valid || !vm_page_is_valid(pg, pgoff, xfsize)) {
1600 			struct uio auio;
1601 			struct iovec aiov;
1602 			int bsize;
1603 
1604 			/*
1605 			 * Ensure that our page is still around when the I/O
1606 			 * completes.
1607 			 */
1608 			vm_page_io_start(pg);
1609 
1610 			/*
1611 			 * Get the page from backing store.
1612 			 */
1613 			bsize = vp->v_mount->mnt_stat.f_iosize;
1614 			auio.uio_iov = &aiov;
1615 			auio.uio_iovcnt = 1;
1616 			aiov.iov_base = 0;
1617 			aiov.iov_len = MAXBSIZE;
1618 			auio.uio_resid = MAXBSIZE;
1619 			auio.uio_offset = trunc_page(off);
1620 			auio.uio_segflg = UIO_NOCOPY;
1621 			auio.uio_rw = UIO_READ;
1622 			auio.uio_td = td;
1623 			vn_lock(vp, LK_SHARED | LK_RETRY);
1624 			error = VOP_READ(vp, &auio,
1625 				    IO_VMIO | ((MAXBSIZE / bsize) << 16),
1626 				    p->p_ucred);
1627 			vn_unlock(vp);
1628 			vm_page_flag_clear(pg, PG_ZERO);
1629 			vm_page_io_finish(pg);
1630 			if (error) {
1631 				crit_enter();
1632 				vm_page_unwire(pg, 0);
1633 				vm_page_try_to_free(pg);
1634 				crit_exit();
1635 				ssb_unlock(&so->so_snd);
1636 				goto done;
1637 			}
1638 		}
1639 
1640 
1641 		/*
1642 		 * Get a sendfile buf. We usually wait as long as necessary,
1643 		 * but this wait can be interrupted.
1644 		 */
1645 		if ((sf = sf_buf_alloc(pg, SFB_CATCH)) == NULL) {
1646 			crit_enter();
1647 			vm_page_unwire(pg, 0);
1648 			vm_page_try_to_free(pg);
1649 			crit_exit();
1650 			ssb_unlock(&so->so_snd);
1651 			error = EINTR;
1652 			goto done;
1653 		}
1654 
1655 		/*
1656 		 * Get an mbuf header and set it up as having external storage.
1657 		 */
1658 		MGETHDR(m, MB_WAIT, MT_DATA);
1659 		if (m == NULL) {
1660 			error = ENOBUFS;
1661 			sf_buf_free(sf);
1662 			ssb_unlock(&so->so_snd);
1663 			goto done;
1664 		}
1665 
1666 		/*
1667 		 * sfm is a temporary hack, use a per-cpu cache for this.
1668 		 */
1669 		sfm = kmalloc(sizeof(struct sfbuf_mref), M_SENDFILE, M_WAITOK);
1670 		sfm->sf = sf;
1671 		sfm->mref_count = 1;
1672 		lwkt_serialize_init(&sfm->serializer);
1673 
1674 		m->m_ext.ext_free = sf_buf_mfree;
1675 		m->m_ext.ext_ref = sf_buf_mref;
1676 		m->m_ext.ext_arg = sfm;
1677 		m->m_ext.ext_buf = (void *)sf->kva;
1678 		m->m_ext.ext_size = PAGE_SIZE;
1679 		m->m_data = (char *) sf->kva + pgoff;
1680 		m->m_flags |= M_EXT;
1681 		m->m_pkthdr.len = m->m_len = xfsize;
1682 		KKASSERT((m->m_flags & (M_EXT_CLUSTER)) == 0);
1683 
1684 		if (mheader != NULL) {
1685 			hbytes = mheader->m_pkthdr.len;
1686 			mheader->m_pkthdr.len += m->m_pkthdr.len;
1687 			m_cat(mheader, m);
1688 			m = mheader;
1689 			mheader = NULL;
1690 		} else
1691 			hbytes = 0;
1692 
1693 		/*
1694 		 * Add the buffer to the socket buffer chain.
1695 		 */
1696 		crit_enter();
1697 retry_space:
1698 		/*
1699 		 * Make sure that the socket is still able to take more data.
1700 		 * CANTSENDMORE being true usually means that the connection
1701 		 * was closed. so_error is true when an error was sensed after
1702 		 * a previous send.
1703 		 * The state is checked after the page mapping and buffer
1704 		 * allocation above since those operations may block and make
1705 		 * any socket checks stale. From this point forward, nothing
1706 		 * blocks before the pru_send (or more accurately, any blocking
1707 		 * results in a loop back to here to re-check).
1708 		 */
1709 		if ((so->so_state & SS_CANTSENDMORE) || so->so_error) {
1710 			if (so->so_state & SS_CANTSENDMORE) {
1711 				error = EPIPE;
1712 			} else {
1713 				error = so->so_error;
1714 				so->so_error = 0;
1715 			}
1716 			m_freem(m);
1717 			ssb_unlock(&so->so_snd);
1718 			crit_exit();
1719 			goto done;
1720 		}
1721 		/*
1722 		 * Wait for socket space to become available. We do this just
1723 		 * after checking the connection state above in order to avoid
1724 		 * a race condition with ssb_wait().
1725 		 */
1726 		if (ssb_space(&so->so_snd) < so->so_snd.ssb_lowat) {
1727 			if (fp->f_flag & FNONBLOCK) {
1728 				m_freem(m);
1729 				ssb_unlock(&so->so_snd);
1730 				crit_exit();
1731 				error = EAGAIN;
1732 				goto done;
1733 			}
1734 			error = ssb_wait(&so->so_snd);
1735 			/*
1736 			 * An error from ssb_wait usually indicates that we've
1737 			 * been interrupted by a signal. If we've sent anything
1738 			 * then return bytes sent, otherwise return the error.
1739 			 */
1740 			if (error) {
1741 				m_freem(m);
1742 				ssb_unlock(&so->so_snd);
1743 				crit_exit();
1744 				goto done;
1745 			}
1746 			goto retry_space;
1747 		}
1748 		error = so_pru_send(so, 0, m, NULL, NULL, td);
1749 		crit_exit();
1750 		if (error) {
1751 			ssb_unlock(&so->so_snd);
1752 			goto done;
1753 		}
1754 	}
1755 	if (mheader != NULL) {
1756 		*sbytes += mheader->m_pkthdr.len;
1757 		error = so_pru_send(so, 0, mheader, NULL, NULL, td);
1758 		mheader = NULL;
1759 	}
1760 	ssb_unlock(&so->so_snd);
1761 
1762 done:
1763 	fdrop(fp);
1764 done0:
1765 	if (mheader != NULL)
1766 		m_freem(mheader);
1767 	return (error);
1768 }
1769 
1770 int
1771 sys_sctp_peeloff(struct sctp_peeloff_args *uap)
1772 {
1773 #ifdef SCTP
1774 	struct thread *td = curthread;
1775 	struct proc *p = td->td_proc;
1776 	struct file *lfp = NULL;
1777 	struct file *nfp = NULL;
1778 	int error;
1779 	struct socket *head, *so;
1780 	caddr_t assoc_id;
1781 	int fd;
1782 	short fflag;		/* type must match fp->f_flag */
1783 
1784 	assoc_id = uap->name;
1785 	error = holdsock(p->p_fd, uap->sd, &lfp);
1786 	if (error) {
1787 		return (error);
1788 	}
1789 	crit_enter();
1790 	head = (struct socket *)lfp->f_data;
1791 	error = sctp_can_peel_off(head, assoc_id);
1792 	if (error) {
1793 		crit_exit();
1794 		goto done;
1795 	}
1796 	/*
1797 	 * At this point we know we do have a assoc to pull
1798 	 * we proceed to get the fd setup. This may block
1799 	 * but that is ok.
1800 	 */
1801 
1802 	fflag = lfp->f_flag;
1803 	error = falloc(p, &nfp, &fd);
1804 	if (error) {
1805 		/*
1806 		 * Probably ran out of file descriptors. Put the
1807 		 * unaccepted connection back onto the queue and
1808 		 * do another wakeup so some other process might
1809 		 * have a chance at it.
1810 		 */
1811 		crit_exit();
1812 		goto done;
1813 	}
1814 	uap->sysmsg_result = fd;
1815 
1816 	so = sctp_get_peeloff(head, assoc_id, &error);
1817 	if (so == NULL) {
1818 		/*
1819 		 * Either someone else peeled it off OR
1820 		 * we can't get a socket.
1821 		 */
1822 		goto noconnection;
1823 	}
1824 	so->so_state &= ~SS_COMP;
1825 	so->so_state &= ~SS_NOFDREF;
1826 	so->so_head = NULL;
1827 	if (head->so_sigio != NULL)
1828 		fsetown(fgetown(head->so_sigio), &so->so_sigio);
1829 
1830 	nfp->f_type = DTYPE_SOCKET;
1831 	nfp->f_flag = fflag;
1832 	nfp->f_ops = &socketops;
1833 	nfp->f_data = so;
1834 
1835 noconnection:
1836 	/*
1837 	 * Assign the file pointer to the reserved descriptor, or clear
1838 	 * the reserved descriptor if an error occured.
1839 	 */
1840 	if (error)
1841 		fsetfd(p, NULL, fd);
1842 	else
1843 		fsetfd(p, nfp, fd);
1844 	crit_exit();
1845 	/*
1846 	 * Release explicitly held references before returning.
1847 	 */
1848 done:
1849 	if (nfp != NULL)
1850 		fdrop(nfp);
1851 	fdrop(lfp);
1852 	return (error);
1853 #else /* SCTP */
1854 	return(EOPNOTSUPP);
1855 #endif /* SCTP */
1856 }
1857