xref: /freebsd/sys/compat/linux/linux_socket.c (revision 7abf30d339336cef02a293352e0a82181e75c2fb)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 1995 Søren Schmidt
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  */
28 
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 /* XXX we use functions that might not exist. */
33 #include "opt_compat.h"
34 #include "opt_inet6.h"
35 
36 #include <sys/param.h>
37 #include <sys/proc.h>
38 #include <sys/systm.h>
39 #include <sys/sysproto.h>
40 #include <sys/capsicum.h>
41 #include <sys/fcntl.h>
42 #include <sys/file.h>
43 #include <sys/filedesc.h>
44 #include <sys/limits.h>
45 #include <sys/lock.h>
46 #include <sys/malloc.h>
47 #include <sys/mutex.h>
48 #include <sys/mbuf.h>
49 #include <sys/socket.h>
50 #include <sys/socketvar.h>
51 #include <sys/syscallsubr.h>
52 #include <sys/uio.h>
53 #include <sys/stat.h>
54 #include <sys/syslog.h>
55 #include <sys/un.h>
56 #include <sys/unistd.h>
57 
58 #include <security/audit/audit.h>
59 
60 #include <net/if.h>
61 #include <net/vnet.h>
62 #include <netinet/in.h>
63 #include <netinet/in_systm.h>
64 #include <netinet/ip.h>
65 #include <netinet/tcp.h>
66 #ifdef INET6
67 #include <netinet/ip6.h>
68 #include <netinet6/ip6_var.h>
69 #endif
70 
71 #ifdef COMPAT_LINUX32
72 #include <machine/../linux32/linux.h>
73 #include <machine/../linux32/linux32_proto.h>
74 #else
75 #include <machine/../linux/linux.h>
76 #include <machine/../linux/linux_proto.h>
77 #endif
78 #include <compat/linux/linux_common.h>
79 #include <compat/linux/linux_file.h>
80 #include <compat/linux/linux_mib.h>
81 #include <compat/linux/linux_socket.h>
82 #include <compat/linux/linux_timer.h>
83 #include <compat/linux/linux_util.h>
84 
85 static int linux_sendmsg_common(struct thread *, l_int, struct l_msghdr *,
86 					l_uint);
87 static int linux_recvmsg_common(struct thread *, l_int, struct l_msghdr *,
88 					l_uint, struct msghdr *);
89 static int linux_set_socket_flags(int, int *);
90 
91 static int
92 linux_to_bsd_sockopt_level(int level)
93 {
94 
95 	if (level == LINUX_SOL_SOCKET)
96 		return (SOL_SOCKET);
97 	/* Remaining values are RFC-defined protocol numbers. */
98 	return (level);
99 }
100 
101 static int
102 bsd_to_linux_sockopt_level(int level)
103 {
104 
105 	if (level == SOL_SOCKET)
106 		return (LINUX_SOL_SOCKET);
107 	return (level);
108 }
109 
110 static int
111 linux_to_bsd_ip_sockopt(int opt)
112 {
113 
114 	switch (opt) {
115 	case LINUX_IP_TOS:
116 		return (IP_TOS);
117 	case LINUX_IP_TTL:
118 		return (IP_TTL);
119 	case LINUX_IP_OPTIONS:
120 		return (IP_OPTIONS);
121 	case LINUX_IP_MULTICAST_IF:
122 		return (IP_MULTICAST_IF);
123 	case LINUX_IP_MULTICAST_TTL:
124 		return (IP_MULTICAST_TTL);
125 	case LINUX_IP_MULTICAST_LOOP:
126 		return (IP_MULTICAST_LOOP);
127 	case LINUX_IP_ADD_MEMBERSHIP:
128 		return (IP_ADD_MEMBERSHIP);
129 	case LINUX_IP_DROP_MEMBERSHIP:
130 		return (IP_DROP_MEMBERSHIP);
131 	case LINUX_IP_HDRINCL:
132 		return (IP_HDRINCL);
133 	}
134 	return (-1);
135 }
136 
137 static int
138 linux_to_bsd_ip6_sockopt(int opt)
139 {
140 
141 	switch (opt) {
142 	case LINUX_IPV6_NEXTHOP:
143 		return (IPV6_NEXTHOP);
144 	case LINUX_IPV6_UNICAST_HOPS:
145 		return (IPV6_UNICAST_HOPS);
146 	case LINUX_IPV6_MULTICAST_IF:
147 		return (IPV6_MULTICAST_IF);
148 	case LINUX_IPV6_MULTICAST_HOPS:
149 		return (IPV6_MULTICAST_HOPS);
150 	case LINUX_IPV6_MULTICAST_LOOP:
151 		return (IPV6_MULTICAST_LOOP);
152 	case LINUX_IPV6_ADD_MEMBERSHIP:
153 		return (IPV6_JOIN_GROUP);
154 	case LINUX_IPV6_DROP_MEMBERSHIP:
155 		return (IPV6_LEAVE_GROUP);
156 	case LINUX_IPV6_V6ONLY:
157 		return (IPV6_V6ONLY);
158 	case LINUX_IPV6_DONTFRAG:
159 		return (IPV6_DONTFRAG);
160 #if 0
161 	case LINUX_IPV6_CHECKSUM:
162 		return (IPV6_CHECKSUM);
163 	case LINUX_IPV6_RECVPKTINFO:
164 		return (IPV6_RECVPKTINFO);
165 	case LINUX_IPV6_PKTINFO:
166 		return (IPV6_PKTINFO);
167 	case LINUX_IPV6_RECVHOPLIMIT:
168 		return (IPV6_RECVHOPLIMIT);
169 	case LINUX_IPV6_HOPLIMIT:
170 		return (IPV6_HOPLIMIT);
171 	case LINUX_IPV6_RECVHOPOPTS:
172 		return (IPV6_RECVHOPOPTS);
173 	case LINUX_IPV6_HOPOPTS:
174 		return (IPV6_HOPOPTS);
175 	case LINUX_IPV6_RTHDRDSTOPTS:
176 		return (IPV6_RTHDRDSTOPTS);
177 	case LINUX_IPV6_RECVRTHDR:
178 		return (IPV6_RECVRTHDR);
179 	case LINUX_IPV6_RTHDR:
180 		return (IPV6_RTHDR);
181 	case LINUX_IPV6_RECVDSTOPTS:
182 		return (IPV6_RECVDSTOPTS);
183 	case LINUX_IPV6_DSTOPTS:
184 		return (IPV6_DSTOPTS);
185 	case LINUX_IPV6_RECVPATHMTU:
186 		return (IPV6_RECVPATHMTU);
187 	case LINUX_IPV6_PATHMTU:
188 		return (IPV6_PATHMTU);
189 #endif
190 	}
191 	return (-1);
192 }
193 
194 static int
195 linux_to_bsd_so_sockopt(int opt)
196 {
197 
198 	switch (opt) {
199 	case LINUX_SO_DEBUG:
200 		return (SO_DEBUG);
201 	case LINUX_SO_REUSEADDR:
202 		return (SO_REUSEADDR);
203 	case LINUX_SO_TYPE:
204 		return (SO_TYPE);
205 	case LINUX_SO_ERROR:
206 		return (SO_ERROR);
207 	case LINUX_SO_DONTROUTE:
208 		return (SO_DONTROUTE);
209 	case LINUX_SO_BROADCAST:
210 		return (SO_BROADCAST);
211 	case LINUX_SO_SNDBUF:
212 	case LINUX_SO_SNDBUFFORCE:
213 		return (SO_SNDBUF);
214 	case LINUX_SO_RCVBUF:
215 	case LINUX_SO_RCVBUFFORCE:
216 		return (SO_RCVBUF);
217 	case LINUX_SO_KEEPALIVE:
218 		return (SO_KEEPALIVE);
219 	case LINUX_SO_OOBINLINE:
220 		return (SO_OOBINLINE);
221 	case LINUX_SO_LINGER:
222 		return (SO_LINGER);
223 	case LINUX_SO_REUSEPORT:
224 		return (SO_REUSEPORT_LB);
225 	case LINUX_SO_PASSCRED:
226 		return (LOCAL_CREDS_PERSISTENT);
227 	case LINUX_SO_PEERCRED:
228 		return (LOCAL_PEERCRED);
229 	case LINUX_SO_RCVLOWAT:
230 		return (SO_RCVLOWAT);
231 	case LINUX_SO_SNDLOWAT:
232 		return (SO_SNDLOWAT);
233 	case LINUX_SO_RCVTIMEO:
234 		return (SO_RCVTIMEO);
235 	case LINUX_SO_SNDTIMEO:
236 		return (SO_SNDTIMEO);
237 	case LINUX_SO_TIMESTAMP:
238 		return (SO_TIMESTAMP);
239 	case LINUX_SO_ACCEPTCONN:
240 		return (SO_ACCEPTCONN);
241 	case LINUX_SO_PROTOCOL:
242 		return (SO_PROTOCOL);
243 	}
244 	return (-1);
245 }
246 
247 static int
248 linux_to_bsd_tcp_sockopt(int opt)
249 {
250 
251 	switch (opt) {
252 	case LINUX_TCP_NODELAY:
253 		return (TCP_NODELAY);
254 	case LINUX_TCP_MAXSEG:
255 		return (TCP_MAXSEG);
256 	case LINUX_TCP_CORK:
257 		return (TCP_NOPUSH);
258 	case LINUX_TCP_KEEPIDLE:
259 		return (TCP_KEEPIDLE);
260 	case LINUX_TCP_KEEPINTVL:
261 		return (TCP_KEEPINTVL);
262 	case LINUX_TCP_KEEPCNT:
263 		return (TCP_KEEPCNT);
264 	case LINUX_TCP_MD5SIG:
265 		return (TCP_MD5SIG);
266 	}
267 	return (-1);
268 }
269 
270 static int
271 linux_to_bsd_msg_flags(int flags)
272 {
273 	int ret_flags = 0;
274 
275 	if (flags & LINUX_MSG_OOB)
276 		ret_flags |= MSG_OOB;
277 	if (flags & LINUX_MSG_PEEK)
278 		ret_flags |= MSG_PEEK;
279 	if (flags & LINUX_MSG_DONTROUTE)
280 		ret_flags |= MSG_DONTROUTE;
281 	if (flags & LINUX_MSG_CTRUNC)
282 		ret_flags |= MSG_CTRUNC;
283 	if (flags & LINUX_MSG_TRUNC)
284 		ret_flags |= MSG_TRUNC;
285 	if (flags & LINUX_MSG_DONTWAIT)
286 		ret_flags |= MSG_DONTWAIT;
287 	if (flags & LINUX_MSG_EOR)
288 		ret_flags |= MSG_EOR;
289 	if (flags & LINUX_MSG_WAITALL)
290 		ret_flags |= MSG_WAITALL;
291 	if (flags & LINUX_MSG_NOSIGNAL)
292 		ret_flags |= MSG_NOSIGNAL;
293 #if 0 /* not handled */
294 	if (flags & LINUX_MSG_PROXY)
295 		;
296 	if (flags & LINUX_MSG_FIN)
297 		;
298 	if (flags & LINUX_MSG_SYN)
299 		;
300 	if (flags & LINUX_MSG_CONFIRM)
301 		;
302 	if (flags & LINUX_MSG_RST)
303 		;
304 	if (flags & LINUX_MSG_ERRQUEUE)
305 		;
306 #endif
307 	return (ret_flags);
308 }
309 
310 static int
311 linux_to_bsd_cmsg_type(int cmsg_type)
312 {
313 
314 	switch (cmsg_type) {
315 	case LINUX_SCM_RIGHTS:
316 		return (SCM_RIGHTS);
317 	case LINUX_SCM_CREDENTIALS:
318 		return (SCM_CREDS);
319 	}
320 	return (-1);
321 }
322 
323 static int
324 bsd_to_linux_cmsg_type(int cmsg_type)
325 {
326 
327 	switch (cmsg_type) {
328 	case SCM_RIGHTS:
329 		return (LINUX_SCM_RIGHTS);
330 	case SCM_CREDS:
331 		return (LINUX_SCM_CREDENTIALS);
332 	case SCM_TIMESTAMP:
333 		return (LINUX_SCM_TIMESTAMP);
334 	}
335 	return (-1);
336 }
337 
338 static int
339 linux_to_bsd_msghdr(struct msghdr *bhdr, const struct l_msghdr *lhdr)
340 {
341 	if (lhdr->msg_controllen > INT_MAX)
342 		return (ENOBUFS);
343 
344 	bhdr->msg_name		= PTRIN(lhdr->msg_name);
345 	bhdr->msg_namelen	= lhdr->msg_namelen;
346 	bhdr->msg_iov		= PTRIN(lhdr->msg_iov);
347 	bhdr->msg_iovlen	= lhdr->msg_iovlen;
348 	bhdr->msg_control	= PTRIN(lhdr->msg_control);
349 
350 	/*
351 	 * msg_controllen is skipped since BSD and LINUX control messages
352 	 * are potentially different sizes (e.g. the cred structure used
353 	 * by SCM_CREDS is different between the two operating system).
354 	 *
355 	 * The caller can set it (if necessary) after converting all the
356 	 * control messages.
357 	 */
358 
359 	bhdr->msg_flags		= linux_to_bsd_msg_flags(lhdr->msg_flags);
360 	return (0);
361 }
362 
363 static int
364 bsd_to_linux_msghdr(const struct msghdr *bhdr, struct l_msghdr *lhdr)
365 {
366 	lhdr->msg_name		= PTROUT(bhdr->msg_name);
367 	lhdr->msg_namelen	= bhdr->msg_namelen;
368 	lhdr->msg_iov		= PTROUT(bhdr->msg_iov);
369 	lhdr->msg_iovlen	= bhdr->msg_iovlen;
370 	lhdr->msg_control	= PTROUT(bhdr->msg_control);
371 
372 	/*
373 	 * msg_controllen is skipped since BSD and LINUX control messages
374 	 * are potentially different sizes (e.g. the cred structure used
375 	 * by SCM_CREDS is different between the two operating system).
376 	 *
377 	 * The caller can set it (if necessary) after converting all the
378 	 * control messages.
379 	 */
380 
381 	/* msg_flags skipped */
382 	return (0);
383 }
384 
385 static int
386 linux_set_socket_flags(int lflags, int *flags)
387 {
388 
389 	if (lflags & ~(LINUX_SOCK_CLOEXEC | LINUX_SOCK_NONBLOCK))
390 		return (EINVAL);
391 	if (lflags & LINUX_SOCK_NONBLOCK)
392 		*flags |= SOCK_NONBLOCK;
393 	if (lflags & LINUX_SOCK_CLOEXEC)
394 		*flags |= SOCK_CLOEXEC;
395 	return (0);
396 }
397 
398 static int
399 linux_copyout_sockaddr(const struct sockaddr *sa, void *uaddr, size_t len)
400 {
401 	struct l_sockaddr *lsa;
402 	int error;
403 
404 	error = bsd_to_linux_sockaddr(sa, &lsa, len);
405 	if (error != 0)
406 		return (error);
407 
408 	error = copyout(lsa, uaddr, len);
409 	free(lsa, M_SONAME);
410 
411 	return (error);
412 }
413 
414 static int
415 linux_sendit(struct thread *td, int s, struct msghdr *mp, int flags,
416     struct mbuf *control, enum uio_seg segflg)
417 {
418 	struct sockaddr *to;
419 	int error, len;
420 
421 	if (mp->msg_name != NULL) {
422 		len = mp->msg_namelen;
423 		error = linux_to_bsd_sockaddr(mp->msg_name, &to, &len);
424 		if (error != 0)
425 			return (error);
426 		mp->msg_name = to;
427 	} else
428 		to = NULL;
429 
430 	error = kern_sendit(td, s, mp, linux_to_bsd_msg_flags(flags), control,
431 	    segflg);
432 
433 	if (to)
434 		free(to, M_SONAME);
435 	return (error);
436 }
437 
438 /* Return 0 if IP_HDRINCL is set for the given socket. */
439 static int
440 linux_check_hdrincl(struct thread *td, int s)
441 {
442 	int error, optval;
443 	socklen_t size_val;
444 
445 	size_val = sizeof(optval);
446 	error = kern_getsockopt(td, s, IPPROTO_IP, IP_HDRINCL,
447 	    &optval, UIO_SYSSPACE, &size_val);
448 	if (error != 0)
449 		return (error);
450 
451 	return (optval == 0);
452 }
453 
454 /*
455  * Updated sendto() when IP_HDRINCL is set:
456  * tweak endian-dependent fields in the IP packet.
457  */
458 static int
459 linux_sendto_hdrincl(struct thread *td, struct linux_sendto_args *linux_args)
460 {
461 /*
462  * linux_ip_copysize defines how many bytes we should copy
463  * from the beginning of the IP packet before we customize it for BSD.
464  * It should include all the fields we modify (ip_len and ip_off).
465  */
466 #define linux_ip_copysize	8
467 
468 	struct ip *packet;
469 	struct msghdr msg;
470 	struct iovec aiov[1];
471 	int error;
472 
473 	/* Check that the packet isn't too big or too small. */
474 	if (linux_args->len < linux_ip_copysize ||
475 	    linux_args->len > IP_MAXPACKET)
476 		return (EINVAL);
477 
478 	packet = (struct ip *)malloc(linux_args->len, M_LINUX, M_WAITOK);
479 
480 	/* Make kernel copy of the packet to be sent */
481 	if ((error = copyin(PTRIN(linux_args->msg), packet,
482 	    linux_args->len)))
483 		goto goout;
484 
485 	/* Convert fields from Linux to BSD raw IP socket format */
486 	packet->ip_len = linux_args->len;
487 	packet->ip_off = ntohs(packet->ip_off);
488 
489 	/* Prepare the msghdr and iovec structures describing the new packet */
490 	msg.msg_name = PTRIN(linux_args->to);
491 	msg.msg_namelen = linux_args->tolen;
492 	msg.msg_iov = aiov;
493 	msg.msg_iovlen = 1;
494 	msg.msg_control = NULL;
495 	msg.msg_flags = 0;
496 	aiov[0].iov_base = (char *)packet;
497 	aiov[0].iov_len = linux_args->len;
498 	error = linux_sendit(td, linux_args->s, &msg, linux_args->flags,
499 	    NULL, UIO_SYSSPACE);
500 goout:
501 	free(packet, M_LINUX);
502 	return (error);
503 }
504 
505 int
506 linux_socket(struct thread *td, struct linux_socket_args *args)
507 {
508 	int domain, retval_socket, type;
509 
510 	type = args->type & LINUX_SOCK_TYPE_MASK;
511 	if (type < 0 || type > LINUX_SOCK_MAX)
512 		return (EINVAL);
513 	retval_socket = linux_set_socket_flags(args->type & ~LINUX_SOCK_TYPE_MASK,
514 		&type);
515 	if (retval_socket != 0)
516 		return (retval_socket);
517 	domain = linux_to_bsd_domain(args->domain);
518 	if (domain == -1) {
519 		if (args->domain == LINUX_AF_NETLINK &&
520 		    args->protocol == LINUX_NETLINK_ROUTE) {
521 			linux_msg(curthread,
522 			    "unsupported socket(AF_NETLINK, %d, NETLINK_ROUTE)", type);
523 			return (EAFNOSUPPORT);
524 		}
525 
526 		if (args->domain == LINUX_AF_NETLINK &&
527 		    args->protocol == LINUX_NETLINK_UEVENT) {
528 			linux_msg(curthread,
529 			    "unsupported socket(AF_NETLINK, %d, NETLINK_UEVENT)", type);
530 			return (EAFNOSUPPORT);
531 		}
532 
533 		linux_msg(curthread, "unsupported socket domain %d, type %d, protocol %d",
534 		    args->domain, args->type & LINUX_SOCK_TYPE_MASK, args->protocol);
535 		return (EAFNOSUPPORT);
536 	}
537 
538 	retval_socket = kern_socket(td, domain, type, args->protocol);
539 	if (retval_socket)
540 		return (retval_socket);
541 
542 	if (type == SOCK_RAW
543 	    && (args->protocol == IPPROTO_RAW || args->protocol == 0)
544 	    && domain == PF_INET) {
545 		/* It's a raw IP socket: set the IP_HDRINCL option. */
546 		int hdrincl;
547 
548 		hdrincl = 1;
549 		/* We ignore any error returned by kern_setsockopt() */
550 		kern_setsockopt(td, td->td_retval[0], IPPROTO_IP, IP_HDRINCL,
551 		    &hdrincl, UIO_SYSSPACE, sizeof(hdrincl));
552 	}
553 #ifdef INET6
554 	/*
555 	 * Linux AF_INET6 socket has IPV6_V6ONLY setsockopt set to 0 by default
556 	 * and some apps depend on this. So, set V6ONLY to 0 for Linux apps.
557 	 * For simplicity we do this unconditionally of the net.inet6.ip6.v6only
558 	 * sysctl value.
559 	 */
560 	if (domain == PF_INET6) {
561 		int v6only;
562 
563 		v6only = 0;
564 		/* We ignore any error returned by setsockopt() */
565 		kern_setsockopt(td, td->td_retval[0], IPPROTO_IPV6, IPV6_V6ONLY,
566 		    &v6only, UIO_SYSSPACE, sizeof(v6only));
567 	}
568 #endif
569 
570 	return (retval_socket);
571 }
572 
573 int
574 linux_bind(struct thread *td, struct linux_bind_args *args)
575 {
576 	struct sockaddr *sa;
577 	int error;
578 
579 	error = linux_to_bsd_sockaddr(PTRIN(args->name), &sa,
580 	    &args->namelen);
581 	if (error != 0)
582 		return (error);
583 
584 	error = kern_bindat(td, AT_FDCWD, args->s, sa);
585 	free(sa, M_SONAME);
586 
587 	/* XXX */
588 	if (error == EADDRNOTAVAIL && args->namelen != sizeof(struct sockaddr_in))
589 		return (EINVAL);
590 	return (error);
591 }
592 
593 int
594 linux_connect(struct thread *td, struct linux_connect_args *args)
595 {
596 	struct socket *so;
597 	struct sockaddr *sa;
598 	struct file *fp;
599 	u_int fflag;
600 	int error;
601 
602 	error = linux_to_bsd_sockaddr(PTRIN(args->name), &sa,
603 	    &args->namelen);
604 	if (error != 0)
605 		return (error);
606 
607 	error = kern_connectat(td, AT_FDCWD, args->s, sa);
608 	free(sa, M_SONAME);
609 	if (error != EISCONN)
610 		return (error);
611 
612 	/*
613 	 * Linux doesn't return EISCONN the first time it occurs,
614 	 * when on a non-blocking socket. Instead it returns the
615 	 * error getsockopt(SOL_SOCKET, SO_ERROR) would return on BSD.
616 	 */
617 	error = getsock_cap(td, args->s, &cap_connect_rights,
618 	    &fp, &fflag, NULL);
619 	if (error != 0)
620 		return (error);
621 
622 	error = EISCONN;
623 	so = fp->f_data;
624 	if (fflag & FNONBLOCK) {
625 		SOCK_LOCK(so);
626 		if (so->so_emuldata == 0)
627 			error = so->so_error;
628 		so->so_emuldata = (void *)1;
629 		SOCK_UNLOCK(so);
630 	}
631 	fdrop(fp, td);
632 
633 	return (error);
634 }
635 
636 int
637 linux_listen(struct thread *td, struct linux_listen_args *args)
638 {
639 
640 	return (kern_listen(td, args->s, args->backlog));
641 }
642 
643 static int
644 linux_accept_common(struct thread *td, int s, l_uintptr_t addr,
645     l_uintptr_t namelen, int flags)
646 {
647 	struct sockaddr *sa;
648 	struct file *fp, *fp1;
649 	int bflags, len;
650 	struct socket *so;
651 	int error, error1;
652 
653 	bflags = 0;
654 	fp = NULL;
655 	sa = NULL;
656 
657 	error = linux_set_socket_flags(flags, &bflags);
658 	if (error != 0)
659 		return (error);
660 
661 	if (PTRIN(addr) == NULL) {
662 		len = 0;
663 		error = kern_accept4(td, s, NULL, NULL, bflags, NULL);
664 	} else {
665 		error = copyin(PTRIN(namelen), &len, sizeof(len));
666 		if (error != 0)
667 			return (error);
668 		if (len < 0)
669 			return (EINVAL);
670 		error = kern_accept4(td, s, &sa, &len, bflags, &fp);
671 	}
672 
673 	/*
674 	 * Translate errno values into ones used by Linux.
675 	 */
676 	if (error != 0) {
677 		/*
678 		 * XXX. This is wrong, different sockaddr structures
679 		 * have different sizes.
680 		 */
681 		switch (error) {
682 		case EFAULT:
683 			if (namelen != sizeof(struct sockaddr_in))
684 				error = EINVAL;
685 			break;
686 		case EINVAL:
687 			error1 = getsock_cap(td, s, &cap_accept_rights, &fp1, NULL, NULL);
688 			if (error1 != 0) {
689 				error = error1;
690 				break;
691 			}
692 			so = fp1->f_data;
693 			if (so->so_type == SOCK_DGRAM)
694 				error = EOPNOTSUPP;
695 			fdrop(fp1, td);
696 			break;
697 		}
698 		return (error);
699 	}
700 
701 	if (len != 0) {
702 		error = linux_copyout_sockaddr(sa, PTRIN(addr), len);
703 
704 		/*
705 		 * XXX: We should also copyout the len, shouldn't we?
706 		 */
707 
708 		if (error != 0) {
709 			fdclose(td, fp, td->td_retval[0]);
710 			td->td_retval[0] = 0;
711 		}
712 	}
713 	if (fp != NULL)
714 		fdrop(fp, td);
715 	free(sa, M_SONAME);
716 	return (error);
717 }
718 
719 int
720 linux_accept(struct thread *td, struct linux_accept_args *args)
721 {
722 
723 	return (linux_accept_common(td, args->s, args->addr,
724 	    args->namelen, 0));
725 }
726 
727 int
728 linux_accept4(struct thread *td, struct linux_accept4_args *args)
729 {
730 
731 	return (linux_accept_common(td, args->s, args->addr,
732 	    args->namelen, args->flags));
733 }
734 
735 int
736 linux_getsockname(struct thread *td, struct linux_getsockname_args *args)
737 {
738 	struct sockaddr *sa;
739 	int len, error;
740 
741 	error = copyin(PTRIN(args->namelen), &len, sizeof(len));
742 	if (error != 0)
743 		return (error);
744 
745 	error = kern_getsockname(td, args->s, &sa, &len);
746 	if (error != 0)
747 		return (error);
748 
749 	if (len != 0)
750 		error = linux_copyout_sockaddr(sa, PTRIN(args->addr), len);
751 
752 	free(sa, M_SONAME);
753 	if (error == 0)
754 		error = copyout(&len, PTRIN(args->namelen), sizeof(len));
755 	return (error);
756 }
757 
758 int
759 linux_getpeername(struct thread *td, struct linux_getpeername_args *args)
760 {
761 	struct sockaddr *sa;
762 	int len, error;
763 
764 	error = copyin(PTRIN(args->namelen), &len, sizeof(len));
765 	if (error != 0)
766 		return (error);
767 	if (len < 0)
768 		return (EINVAL);
769 
770 	error = kern_getpeername(td, args->s, &sa, &len);
771 	if (error != 0)
772 		return (error);
773 
774 	if (len != 0)
775 		error = linux_copyout_sockaddr(sa, PTRIN(args->addr), len);
776 
777 	free(sa, M_SONAME);
778 	if (error == 0)
779 		error = copyout(&len, PTRIN(args->namelen), sizeof(len));
780 	return (error);
781 }
782 
783 int
784 linux_socketpair(struct thread *td, struct linux_socketpair_args *args)
785 {
786 	int domain, error, sv[2], type;
787 
788 	domain = linux_to_bsd_domain(args->domain);
789 	if (domain != PF_LOCAL)
790 		return (EAFNOSUPPORT);
791 	type = args->type & LINUX_SOCK_TYPE_MASK;
792 	if (type < 0 || type > LINUX_SOCK_MAX)
793 		return (EINVAL);
794 	error = linux_set_socket_flags(args->type & ~LINUX_SOCK_TYPE_MASK,
795 	    &type);
796 	if (error != 0)
797 		return (error);
798 	if (args->protocol != 0 && args->protocol != PF_UNIX) {
799 		/*
800 		 * Use of PF_UNIX as protocol argument is not right,
801 		 * but Linux does it.
802 		 * Do not map PF_UNIX as its Linux value is identical
803 		 * to FreeBSD one.
804 		 */
805 		return (EPROTONOSUPPORT);
806 	}
807 	error = kern_socketpair(td, domain, type, 0, sv);
808 	if (error != 0)
809                 return (error);
810         error = copyout(sv, PTRIN(args->rsv), 2 * sizeof(int));
811         if (error != 0) {
812                 (void)kern_close(td, sv[0]);
813                 (void)kern_close(td, sv[1]);
814         }
815 	return (error);
816 }
817 
818 #if defined(__i386__) || (defined(__amd64__) && defined(COMPAT_LINUX32))
819 struct linux_send_args {
820 	register_t s;
821 	register_t msg;
822 	register_t len;
823 	register_t flags;
824 };
825 
826 static int
827 linux_send(struct thread *td, struct linux_send_args *args)
828 {
829 	struct sendto_args /* {
830 		int s;
831 		caddr_t buf;
832 		int len;
833 		int flags;
834 		caddr_t to;
835 		int tolen;
836 	} */ bsd_args;
837 	struct file *fp;
838 	int error, fflag;
839 
840 	bsd_args.s = args->s;
841 	bsd_args.buf = (caddr_t)PTRIN(args->msg);
842 	bsd_args.len = args->len;
843 	bsd_args.flags = args->flags;
844 	bsd_args.to = NULL;
845 	bsd_args.tolen = 0;
846 	error = sys_sendto(td, &bsd_args);
847 	if (error == ENOTCONN) {
848 		/*
849 		 * Linux doesn't return ENOTCONN for non-blocking sockets.
850 		 * Instead it returns the EAGAIN.
851 		 */
852 		error = getsock_cap(td, args->s, &cap_send_rights, &fp,
853 		    &fflag, NULL);
854 		if (error == 0) {
855 			if (fflag & FNONBLOCK)
856 				error = EAGAIN;
857 			fdrop(fp, td);
858 		}
859 	}
860 	return (error);
861 }
862 
863 struct linux_recv_args {
864 	register_t s;
865 	register_t msg;
866 	register_t len;
867 	register_t flags;
868 };
869 
870 static int
871 linux_recv(struct thread *td, struct linux_recv_args *args)
872 {
873 	struct recvfrom_args /* {
874 		int s;
875 		caddr_t buf;
876 		int len;
877 		int flags;
878 		struct sockaddr *from;
879 		socklen_t fromlenaddr;
880 	} */ bsd_args;
881 
882 	bsd_args.s = args->s;
883 	bsd_args.buf = (caddr_t)PTRIN(args->msg);
884 	bsd_args.len = args->len;
885 	bsd_args.flags = linux_to_bsd_msg_flags(args->flags);
886 	bsd_args.from = NULL;
887 	bsd_args.fromlenaddr = 0;
888 	return (sys_recvfrom(td, &bsd_args));
889 }
890 #endif /* __i386__ || (__amd64__ && COMPAT_LINUX32) */
891 
892 int
893 linux_sendto(struct thread *td, struct linux_sendto_args *args)
894 {
895 	struct msghdr msg;
896 	struct iovec aiov;
897 
898 	if (linux_check_hdrincl(td, args->s) == 0)
899 		/* IP_HDRINCL set, tweak the packet before sending */
900 		return (linux_sendto_hdrincl(td, args));
901 
902 	msg.msg_name = PTRIN(args->to);
903 	msg.msg_namelen = args->tolen;
904 	msg.msg_iov = &aiov;
905 	msg.msg_iovlen = 1;
906 	msg.msg_control = NULL;
907 	msg.msg_flags = 0;
908 	aiov.iov_base = PTRIN(args->msg);
909 	aiov.iov_len = args->len;
910 	return (linux_sendit(td, args->s, &msg, args->flags, NULL,
911 	    UIO_USERSPACE));
912 }
913 
914 int
915 linux_recvfrom(struct thread *td, struct linux_recvfrom_args *args)
916 {
917 	struct sockaddr *sa;
918 	struct msghdr msg;
919 	struct iovec aiov;
920 	int error, fromlen;
921 
922 	if (PTRIN(args->fromlen) != NULL) {
923 		error = copyin(PTRIN(args->fromlen), &fromlen,
924 		    sizeof(fromlen));
925 		if (error != 0)
926 			return (error);
927 		if (fromlen < 0)
928 			return (EINVAL);
929 		sa = malloc(fromlen, M_SONAME, M_WAITOK);
930 	} else {
931 		fromlen = 0;
932 		sa = NULL;
933 	}
934 
935 	msg.msg_name = sa;
936 	msg.msg_namelen = fromlen;
937 	msg.msg_iov = &aiov;
938 	msg.msg_iovlen = 1;
939 	aiov.iov_base = PTRIN(args->buf);
940 	aiov.iov_len = args->len;
941 	msg.msg_control = 0;
942 	msg.msg_flags = linux_to_bsd_msg_flags(args->flags);
943 
944 	error = kern_recvit(td, args->s, &msg, UIO_SYSSPACE, NULL);
945 	if (error != 0)
946 		goto out;
947 
948 	if (PTRIN(args->from) != NULL)
949 		error = linux_copyout_sockaddr(sa, PTRIN(args->from), msg.msg_namelen);
950 
951 	if (error == 0 && PTRIN(args->fromlen) != NULL)
952 		error = copyout(&msg.msg_namelen, PTRIN(args->fromlen),
953 		    sizeof(msg.msg_namelen));
954 out:
955 	free(sa, M_SONAME);
956 	return (error);
957 }
958 
959 static int
960 linux_sendmsg_common(struct thread *td, l_int s, struct l_msghdr *msghdr,
961     l_uint flags)
962 {
963 	struct cmsghdr *cmsg;
964 	struct mbuf *control;
965 	struct msghdr msg;
966 	struct l_cmsghdr linux_cmsg;
967 	struct l_cmsghdr *ptr_cmsg;
968 	struct l_msghdr linux_msghdr;
969 	struct iovec *iov;
970 	socklen_t datalen;
971 	struct sockaddr *sa;
972 	struct socket *so;
973 	sa_family_t sa_family;
974 	struct file *fp;
975 	void *data;
976 	l_size_t len;
977 	l_size_t clen;
978 	int error, fflag;
979 
980 	error = copyin(msghdr, &linux_msghdr, sizeof(linux_msghdr));
981 	if (error != 0)
982 		return (error);
983 
984 	/*
985 	 * Some Linux applications (ping) define a non-NULL control data
986 	 * pointer, but a msg_controllen of 0, which is not allowed in the
987 	 * FreeBSD system call interface.  NULL the msg_control pointer in
988 	 * order to handle this case.  This should be checked, but allows the
989 	 * Linux ping to work.
990 	 */
991 	if (PTRIN(linux_msghdr.msg_control) != NULL &&
992 	    linux_msghdr.msg_controllen == 0)
993 		linux_msghdr.msg_control = PTROUT(NULL);
994 
995 	error = linux_to_bsd_msghdr(&msg, &linux_msghdr);
996 	if (error != 0)
997 		return (error);
998 
999 #ifdef COMPAT_LINUX32
1000 	error = linux32_copyiniov(PTRIN(msg.msg_iov), msg.msg_iovlen,
1001 	    &iov, EMSGSIZE);
1002 #else
1003 	error = copyiniov(msg.msg_iov, msg.msg_iovlen, &iov, EMSGSIZE);
1004 #endif
1005 	if (error != 0)
1006 		return (error);
1007 
1008 	control = NULL;
1009 
1010 	error = kern_getsockname(td, s, &sa, &datalen);
1011 	if (error != 0)
1012 		goto bad;
1013 	sa_family = sa->sa_family;
1014 	free(sa, M_SONAME);
1015 
1016 	if (flags & LINUX_MSG_OOB) {
1017 		error = EOPNOTSUPP;
1018 		if (sa_family == AF_UNIX)
1019 			goto bad;
1020 
1021 		error = getsock_cap(td, s, &cap_send_rights, &fp,
1022 		    &fflag, NULL);
1023 		if (error != 0)
1024 			goto bad;
1025 		so = fp->f_data;
1026 		if (so->so_type != SOCK_STREAM)
1027 			error = EOPNOTSUPP;
1028 		fdrop(fp, td);
1029 		if (error != 0)
1030 			goto bad;
1031 	}
1032 
1033 	if (linux_msghdr.msg_controllen >= sizeof(struct l_cmsghdr)) {
1034 		error = ENOBUFS;
1035 		control = m_get(M_WAITOK, MT_CONTROL);
1036 		MCLGET(control, M_WAITOK);
1037 		data = mtod(control, void *);
1038 		datalen = 0;
1039 
1040 		ptr_cmsg = PTRIN(linux_msghdr.msg_control);
1041 		clen = linux_msghdr.msg_controllen;
1042 		do {
1043 			error = copyin(ptr_cmsg, &linux_cmsg,
1044 			    sizeof(struct l_cmsghdr));
1045 			if (error != 0)
1046 				goto bad;
1047 
1048 			error = EINVAL;
1049 			if (linux_cmsg.cmsg_len < sizeof(struct l_cmsghdr) ||
1050 			    linux_cmsg.cmsg_len > clen)
1051 				goto bad;
1052 
1053 			if (datalen + CMSG_HDRSZ > MCLBYTES)
1054 				goto bad;
1055 
1056 			/*
1057 			 * Now we support only SCM_RIGHTS and SCM_CRED,
1058 			 * so return EINVAL in any other cmsg_type
1059 			 */
1060 			cmsg = data;
1061 			cmsg->cmsg_type =
1062 			    linux_to_bsd_cmsg_type(linux_cmsg.cmsg_type);
1063 			cmsg->cmsg_level =
1064 			    linux_to_bsd_sockopt_level(linux_cmsg.cmsg_level);
1065 			if (cmsg->cmsg_type == -1
1066 			    || cmsg->cmsg_level != SOL_SOCKET) {
1067 				linux_msg(curthread,
1068 				    "unsupported sendmsg cmsg level %d type %d",
1069 				    linux_cmsg.cmsg_level, linux_cmsg.cmsg_type);
1070 				goto bad;
1071 			}
1072 
1073 			/*
1074 			 * Some applications (e.g. pulseaudio) attempt to
1075 			 * send ancillary data even if the underlying protocol
1076 			 * doesn't support it which is not allowed in the
1077 			 * FreeBSD system call interface.
1078 			 */
1079 			if (sa_family != AF_UNIX)
1080 				goto next;
1081 
1082 			if (cmsg->cmsg_type == SCM_CREDS) {
1083 				len = sizeof(struct cmsgcred);
1084 				if (datalen + CMSG_SPACE(len) > MCLBYTES)
1085 					goto bad;
1086 
1087 				/*
1088 				 * The lower levels will fill in the structure
1089 				 */
1090 				memset(CMSG_DATA(data), 0, len);
1091 			} else {
1092 				len = linux_cmsg.cmsg_len - L_CMSG_HDRSZ;
1093 				if (datalen + CMSG_SPACE(len) < datalen ||
1094 				    datalen + CMSG_SPACE(len) > MCLBYTES)
1095 					goto bad;
1096 
1097 				error = copyin(LINUX_CMSG_DATA(ptr_cmsg),
1098 				    CMSG_DATA(data), len);
1099 				if (error != 0)
1100 					goto bad;
1101 			}
1102 
1103 			cmsg->cmsg_len = CMSG_LEN(len);
1104 			data = (char *)data + CMSG_SPACE(len);
1105 			datalen += CMSG_SPACE(len);
1106 
1107 next:
1108 			if (clen <= LINUX_CMSG_ALIGN(linux_cmsg.cmsg_len))
1109 				break;
1110 
1111 			clen -= LINUX_CMSG_ALIGN(linux_cmsg.cmsg_len);
1112 			ptr_cmsg = (struct l_cmsghdr *)((char *)ptr_cmsg +
1113 			    LINUX_CMSG_ALIGN(linux_cmsg.cmsg_len));
1114 		} while(clen >= sizeof(struct l_cmsghdr));
1115 
1116 		control->m_len = datalen;
1117 		if (datalen == 0) {
1118 			m_freem(control);
1119 			control = NULL;
1120 		}
1121 	}
1122 
1123 	msg.msg_iov = iov;
1124 	msg.msg_flags = 0;
1125 	error = linux_sendit(td, s, &msg, flags, control, UIO_USERSPACE);
1126 	control = NULL;
1127 
1128 bad:
1129 	m_freem(control);
1130 	free(iov, M_IOV);
1131 	return (error);
1132 }
1133 
1134 int
1135 linux_sendmsg(struct thread *td, struct linux_sendmsg_args *args)
1136 {
1137 
1138 	return (linux_sendmsg_common(td, args->s, PTRIN(args->msg),
1139 	    args->flags));
1140 }
1141 
1142 int
1143 linux_sendmmsg(struct thread *td, struct linux_sendmmsg_args *args)
1144 {
1145 	struct l_mmsghdr *msg;
1146 	l_uint retval;
1147 	int error, datagrams;
1148 
1149 	if (args->vlen > UIO_MAXIOV)
1150 		args->vlen = UIO_MAXIOV;
1151 
1152 	msg = PTRIN(args->msg);
1153 	datagrams = 0;
1154 	while (datagrams < args->vlen) {
1155 		error = linux_sendmsg_common(td, args->s, &msg->msg_hdr,
1156 		    args->flags);
1157 		if (error != 0)
1158 			break;
1159 
1160 		retval = td->td_retval[0];
1161 		error = copyout(&retval, &msg->msg_len, sizeof(msg->msg_len));
1162 		if (error != 0)
1163 			break;
1164 		++msg;
1165 		++datagrams;
1166 	}
1167 	if (error == 0)
1168 		td->td_retval[0] = datagrams;
1169 	return (error);
1170 }
1171 
1172 static int
1173 linux_recvmsg_common(struct thread *td, l_int s, struct l_msghdr *msghdr,
1174     l_uint flags, struct msghdr *msg)
1175 {
1176 	struct cmsghdr *cm;
1177 	struct cmsgcred *cmcred;
1178 	struct l_cmsghdr *linux_cmsg = NULL;
1179 	struct l_ucred linux_ucred;
1180 	socklen_t datalen, maxlen, outlen;
1181 	struct l_msghdr linux_msghdr;
1182 	struct iovec *iov, *uiov;
1183 	struct mbuf *control = NULL;
1184 	struct mbuf **controlp;
1185 	struct timeval *ftmvl;
1186 	struct sockaddr *sa;
1187 	l_timeval ltmvl;
1188 	caddr_t outbuf;
1189 	void *data;
1190 	int error, i, fd, fds, *fdp;
1191 
1192 	error = copyin(msghdr, &linux_msghdr, sizeof(linux_msghdr));
1193 	if (error != 0)
1194 		return (error);
1195 
1196 	error = linux_to_bsd_msghdr(msg, &linux_msghdr);
1197 	if (error != 0)
1198 		return (error);
1199 
1200 #ifdef COMPAT_LINUX32
1201 	error = linux32_copyiniov(PTRIN(msg->msg_iov), msg->msg_iovlen,
1202 	    &iov, EMSGSIZE);
1203 #else
1204 	error = copyiniov(msg->msg_iov, msg->msg_iovlen, &iov, EMSGSIZE);
1205 #endif
1206 	if (error != 0)
1207 		return (error);
1208 
1209 	if (msg->msg_name != NULL && msg->msg_namelen > 0) {
1210 		msg->msg_namelen = min(msg->msg_namelen, SOCK_MAXADDRLEN);
1211 		sa = malloc(msg->msg_namelen, M_SONAME, M_WAITOK);
1212 		msg->msg_name = sa;
1213 	} else {
1214 		sa = NULL;
1215 		msg->msg_name = NULL;
1216 	}
1217 
1218 	uiov = msg->msg_iov;
1219 	msg->msg_iov = iov;
1220 	controlp = (msg->msg_control != NULL) ? &control : NULL;
1221 	error = kern_recvit(td, s, msg, UIO_SYSSPACE, controlp);
1222 	msg->msg_iov = uiov;
1223 	if (error != 0)
1224 		goto bad;
1225 
1226 	/*
1227 	 * Note that kern_recvit() updates msg->msg_namelen.
1228 	 */
1229 	if (msg->msg_name != NULL && msg->msg_namelen > 0) {
1230 		msg->msg_name = PTRIN(linux_msghdr.msg_name);
1231 		error = linux_copyout_sockaddr(sa,
1232 		    PTRIN(msg->msg_name), msg->msg_namelen);
1233 		if (error != 0)
1234 			goto bad;
1235 	}
1236 
1237 	error = bsd_to_linux_msghdr(msg, &linux_msghdr);
1238 	if (error != 0)
1239 		goto bad;
1240 
1241 	maxlen = linux_msghdr.msg_controllen;
1242 	linux_msghdr.msg_controllen = 0;
1243 	if (control) {
1244 		linux_cmsg = malloc(L_CMSG_HDRSZ, M_LINUX, M_WAITOK | M_ZERO);
1245 
1246 		msg->msg_control = mtod(control, struct cmsghdr *);
1247 		msg->msg_controllen = control->m_len;
1248 
1249 		cm = CMSG_FIRSTHDR(msg);
1250 		outbuf = PTRIN(linux_msghdr.msg_control);
1251 		outlen = 0;
1252 		while (cm != NULL) {
1253 			linux_cmsg->cmsg_type =
1254 			    bsd_to_linux_cmsg_type(cm->cmsg_type);
1255 			linux_cmsg->cmsg_level =
1256 			    bsd_to_linux_sockopt_level(cm->cmsg_level);
1257 			if (linux_cmsg->cmsg_type == -1 ||
1258 			    cm->cmsg_level != SOL_SOCKET) {
1259 				linux_msg(curthread,
1260 				    "unsupported recvmsg cmsg level %d type %d",
1261 				    cm->cmsg_level, cm->cmsg_type);
1262 				error = EINVAL;
1263 				goto bad;
1264 			}
1265 
1266 			data = CMSG_DATA(cm);
1267 			datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1268 
1269 			switch (cm->cmsg_type) {
1270 			case SCM_RIGHTS:
1271 				if (flags & LINUX_MSG_CMSG_CLOEXEC) {
1272 					fds = datalen / sizeof(int);
1273 					fdp = data;
1274 					for (i = 0; i < fds; i++) {
1275 						fd = *fdp++;
1276 						(void)kern_fcntl(td, fd,
1277 						    F_SETFD, FD_CLOEXEC);
1278 					}
1279 				}
1280 				break;
1281 
1282 			case SCM_CREDS:
1283 				/*
1284 				 * Currently LOCAL_CREDS is never in
1285 				 * effect for Linux so no need to worry
1286 				 * about sockcred
1287 				 */
1288 				if (datalen != sizeof(*cmcred)) {
1289 					error = EMSGSIZE;
1290 					goto bad;
1291 				}
1292 				cmcred = (struct cmsgcred *)data;
1293 				bzero(&linux_ucred, sizeof(linux_ucred));
1294 				linux_ucred.pid = cmcred->cmcred_pid;
1295 				linux_ucred.uid = cmcred->cmcred_uid;
1296 				linux_ucred.gid = cmcred->cmcred_gid;
1297 				data = &linux_ucred;
1298 				datalen = sizeof(linux_ucred);
1299 				break;
1300 
1301 			case SCM_TIMESTAMP:
1302 				if (datalen != sizeof(struct timeval)) {
1303 					error = EMSGSIZE;
1304 					goto bad;
1305 				}
1306 				ftmvl = (struct timeval *)data;
1307 				ltmvl.tv_sec = ftmvl->tv_sec;
1308 				ltmvl.tv_usec = ftmvl->tv_usec;
1309 				data = &ltmvl;
1310 				datalen = sizeof(ltmvl);
1311 				break;
1312 			}
1313 
1314 			if (outlen + LINUX_CMSG_LEN(datalen) > maxlen) {
1315 				if (outlen == 0) {
1316 					error = EMSGSIZE;
1317 					goto bad;
1318 				} else {
1319 					linux_msghdr.msg_flags |= LINUX_MSG_CTRUNC;
1320 					m_dispose_extcontrolm(control);
1321 					goto out;
1322 				}
1323 			}
1324 
1325 			linux_cmsg->cmsg_len = LINUX_CMSG_LEN(datalen);
1326 
1327 			error = copyout(linux_cmsg, outbuf, L_CMSG_HDRSZ);
1328 			if (error != 0)
1329 				goto bad;
1330 			outbuf += L_CMSG_HDRSZ;
1331 
1332 			error = copyout(data, outbuf, datalen);
1333 			if (error != 0)
1334 				goto bad;
1335 
1336 			outbuf += LINUX_CMSG_ALIGN(datalen);
1337 			outlen += LINUX_CMSG_LEN(datalen);
1338 
1339 			cm = CMSG_NXTHDR(msg, cm);
1340 		}
1341 		linux_msghdr.msg_controllen = outlen;
1342 	}
1343 
1344 out:
1345 	error = copyout(&linux_msghdr, msghdr, sizeof(linux_msghdr));
1346 
1347 bad:
1348 	if (control != NULL) {
1349 		if (error != 0)
1350 			m_dispose_extcontrolm(control);
1351 		m_freem(control);
1352 	}
1353 	free(iov, M_IOV);
1354 	free(linux_cmsg, M_LINUX);
1355 	free(sa, M_SONAME);
1356 
1357 	return (error);
1358 }
1359 
1360 int
1361 linux_recvmsg(struct thread *td, struct linux_recvmsg_args *args)
1362 {
1363 	struct msghdr bsd_msg;
1364 
1365 	return (linux_recvmsg_common(td, args->s, PTRIN(args->msg),
1366 	    args->flags, &bsd_msg));
1367 }
1368 
1369 int
1370 linux_recvmmsg(struct thread *td, struct linux_recvmmsg_args *args)
1371 {
1372 	struct l_mmsghdr *msg;
1373 	struct msghdr bsd_msg;
1374 	struct l_timespec lts;
1375 	struct timespec ts, tts;
1376 	l_uint retval;
1377 	int error, datagrams;
1378 
1379 	if (args->timeout) {
1380 		error = copyin(args->timeout, &lts, sizeof(struct l_timespec));
1381 		if (error != 0)
1382 			return (error);
1383 		error = linux_to_native_timespec(&ts, &lts);
1384 		if (error != 0)
1385 			return (error);
1386 		getnanotime(&tts);
1387 		timespecadd(&tts, &ts, &tts);
1388 	}
1389 
1390 	msg = PTRIN(args->msg);
1391 	datagrams = 0;
1392 	while (datagrams < args->vlen) {
1393 		error = linux_recvmsg_common(td, args->s, &msg->msg_hdr,
1394 		    args->flags & ~LINUX_MSG_WAITFORONE, &bsd_msg);
1395 		if (error != 0)
1396 			break;
1397 
1398 		retval = td->td_retval[0];
1399 		error = copyout(&retval, &msg->msg_len, sizeof(msg->msg_len));
1400 		if (error != 0)
1401 			break;
1402 		++msg;
1403 		++datagrams;
1404 
1405 		/*
1406 		 * MSG_WAITFORONE turns on MSG_DONTWAIT after one packet.
1407 		 */
1408 		if (args->flags & LINUX_MSG_WAITFORONE)
1409 			args->flags |= LINUX_MSG_DONTWAIT;
1410 
1411 		/*
1412 		 * See BUGS section of recvmmsg(2).
1413 		 */
1414 		if (args->timeout) {
1415 			getnanotime(&ts);
1416 			timespecsub(&ts, &tts, &ts);
1417 			if (!timespecisset(&ts) || ts.tv_sec > 0)
1418 				break;
1419 		}
1420 		/* Out of band data, return right away. */
1421 		if (bsd_msg.msg_flags & MSG_OOB)
1422 			break;
1423 	}
1424 	if (error == 0)
1425 		td->td_retval[0] = datagrams;
1426 	return (error);
1427 }
1428 
1429 int
1430 linux_shutdown(struct thread *td, struct linux_shutdown_args *args)
1431 {
1432 
1433 	return (kern_shutdown(td, args->s, args->how));
1434 }
1435 
1436 int
1437 linux_setsockopt(struct thread *td, struct linux_setsockopt_args *args)
1438 {
1439 	l_timeval linux_tv;
1440 	struct sockaddr *sa;
1441 	struct timeval tv;
1442 	socklen_t len;
1443 	int error, level, name;
1444 
1445 	level = linux_to_bsd_sockopt_level(args->level);
1446 	switch (level) {
1447 	case SOL_SOCKET:
1448 		name = linux_to_bsd_so_sockopt(args->optname);
1449 		switch (name) {
1450 		case LOCAL_CREDS_PERSISTENT:
1451 			level = SOL_LOCAL;
1452 			break;
1453 		case SO_RCVTIMEO:
1454 			/* FALLTHROUGH */
1455 		case SO_SNDTIMEO:
1456 			error = copyin(PTRIN(args->optval), &linux_tv,
1457 			    sizeof(linux_tv));
1458 			if (error != 0)
1459 				return (error);
1460 			tv.tv_sec = linux_tv.tv_sec;
1461 			tv.tv_usec = linux_tv.tv_usec;
1462 			return (kern_setsockopt(td, args->s, level,
1463 			    name, &tv, UIO_SYSSPACE, sizeof(tv)));
1464 			/* NOTREACHED */
1465 		default:
1466 			break;
1467 		}
1468 		break;
1469 	case IPPROTO_IP:
1470 		if (args->optname == LINUX_IP_RECVERR &&
1471 		    linux_ignore_ip_recverr) {
1472 			/*
1473 			 * XXX: This is a hack to unbreak DNS resolution
1474 			 *	with glibc 2.30 and above.
1475 			 */
1476 			return (0);
1477 		}
1478 		name = linux_to_bsd_ip_sockopt(args->optname);
1479 		break;
1480 	case IPPROTO_IPV6:
1481 		name = linux_to_bsd_ip6_sockopt(args->optname);
1482 		break;
1483 	case IPPROTO_TCP:
1484 		name = linux_to_bsd_tcp_sockopt(args->optname);
1485 		break;
1486 	default:
1487 		name = -1;
1488 		break;
1489 	}
1490 	if (name == -1) {
1491 		linux_msg(curthread,
1492 		    "unsupported setsockopt level %d optname %d",
1493 		    args->level, args->optname);
1494 		return (ENOPROTOOPT);
1495 	}
1496 
1497 	if (name == IPV6_NEXTHOP) {
1498 		len = args->optlen;
1499 		error = linux_to_bsd_sockaddr(PTRIN(args->optval), &sa, &len);
1500 		if (error != 0)
1501 			return (error);
1502 
1503 		error = kern_setsockopt(td, args->s, level,
1504 		    name, sa, UIO_SYSSPACE, len);
1505 		free(sa, M_SONAME);
1506 	} else {
1507 		error = kern_setsockopt(td, args->s, level,
1508 		    name, PTRIN(args->optval), UIO_USERSPACE, args->optlen);
1509 	}
1510 
1511 	return (error);
1512 }
1513 
1514 int
1515 linux_getsockopt(struct thread *td, struct linux_getsockopt_args *args)
1516 {
1517 	l_timeval linux_tv;
1518 	struct timeval tv;
1519 	socklen_t tv_len, xulen, len;
1520 	struct sockaddr *sa;
1521 	struct xucred xu;
1522 	struct l_ucred lxu;
1523 	int error, level, name, newval;
1524 
1525 	level = linux_to_bsd_sockopt_level(args->level);
1526 	switch (level) {
1527 	case SOL_SOCKET:
1528 		name = linux_to_bsd_so_sockopt(args->optname);
1529 		switch (name) {
1530 		case LOCAL_CREDS_PERSISTENT:
1531 			level = SOL_LOCAL;
1532 			break;
1533 		case SO_RCVTIMEO:
1534 			/* FALLTHROUGH */
1535 		case SO_SNDTIMEO:
1536 			tv_len = sizeof(tv);
1537 			error = kern_getsockopt(td, args->s, level,
1538 			    name, &tv, UIO_SYSSPACE, &tv_len);
1539 			if (error != 0)
1540 				return (error);
1541 			linux_tv.tv_sec = tv.tv_sec;
1542 			linux_tv.tv_usec = tv.tv_usec;
1543 			return (copyout(&linux_tv, PTRIN(args->optval),
1544 			    sizeof(linux_tv)));
1545 			/* NOTREACHED */
1546 		case LOCAL_PEERCRED:
1547 			if (args->optlen < sizeof(lxu))
1548 				return (EINVAL);
1549 			/*
1550 			 * LOCAL_PEERCRED is not served at the SOL_SOCKET level,
1551 			 * but by the Unix socket's level 0.
1552 			 */
1553 			level = 0;
1554 			xulen = sizeof(xu);
1555 			error = kern_getsockopt(td, args->s, level,
1556 			    name, &xu, UIO_SYSSPACE, &xulen);
1557 			if (error != 0)
1558 				return (error);
1559 			lxu.pid = xu.cr_pid;
1560 			lxu.uid = xu.cr_uid;
1561 			lxu.gid = xu.cr_gid;
1562 			return (copyout(&lxu, PTRIN(args->optval), sizeof(lxu)));
1563 			/* NOTREACHED */
1564 		case SO_ERROR:
1565 			len = sizeof(newval);
1566 			error = kern_getsockopt(td, args->s, level,
1567 			    name, &newval, UIO_SYSSPACE, &len);
1568 			if (error != 0)
1569 				return (error);
1570 			newval = -bsd_to_linux_errno(newval);
1571 			return (copyout(&newval, PTRIN(args->optval), len));
1572 			/* NOTREACHED */
1573 		default:
1574 			break;
1575 		}
1576 		break;
1577 	case IPPROTO_IP:
1578 		name = linux_to_bsd_ip_sockopt(args->optname);
1579 		break;
1580 	case IPPROTO_IPV6:
1581 		name = linux_to_bsd_ip6_sockopt(args->optname);
1582 		break;
1583 	case IPPROTO_TCP:
1584 		name = linux_to_bsd_tcp_sockopt(args->optname);
1585 		break;
1586 	default:
1587 		name = -1;
1588 		break;
1589 	}
1590 	if (name == -1) {
1591 		linux_msg(curthread,
1592 		    "unsupported getsockopt level %d optname %d",
1593 		    args->level, args->optname);
1594 		return (EINVAL);
1595 	}
1596 
1597 	if (name == IPV6_NEXTHOP) {
1598 		error = copyin(PTRIN(args->optlen), &len, sizeof(len));
1599                 if (error != 0)
1600                         return (error);
1601 		sa = malloc(len, M_SONAME, M_WAITOK);
1602 
1603 		error = kern_getsockopt(td, args->s, level,
1604 		    name, sa, UIO_SYSSPACE, &len);
1605 		if (error != 0)
1606 			goto out;
1607 
1608 		error = linux_copyout_sockaddr(sa, PTRIN(args->optval), len);
1609 		if (error == 0)
1610 			error = copyout(&len, PTRIN(args->optlen),
1611 			    sizeof(len));
1612 out:
1613 		free(sa, M_SONAME);
1614 	} else {
1615 		if (args->optval) {
1616 			error = copyin(PTRIN(args->optlen), &len, sizeof(len));
1617 			if (error != 0)
1618 				return (error);
1619 		}
1620 		error = kern_getsockopt(td, args->s, level,
1621 		    name, PTRIN(args->optval), UIO_USERSPACE, &len);
1622 		if (error == 0)
1623 			error = copyout(&len, PTRIN(args->optlen),
1624 			    sizeof(len));
1625 	}
1626 
1627 	return (error);
1628 }
1629 
1630 static int
1631 linux_sendfile_common(struct thread *td, l_int out, l_int in,
1632     l_loff_t *offset, l_size_t count)
1633 {
1634 	off_t bytes_read;
1635 	int error;
1636 	l_loff_t current_offset;
1637 	struct file *fp;
1638 
1639 	AUDIT_ARG_FD(in);
1640 	error = fget_read(td, in, &cap_pread_rights, &fp);
1641 	if (error != 0)
1642 		return (error);
1643 
1644 	if (offset != NULL) {
1645 		current_offset = *offset;
1646 	} else {
1647 		error = (fp->f_ops->fo_flags & DFLAG_SEEKABLE) != 0 ?
1648 		    fo_seek(fp, 0, SEEK_CUR, td) : ESPIPE;
1649 		if (error != 0)
1650 			goto drop;
1651 		current_offset = td->td_uretoff.tdu_off;
1652 	}
1653 
1654 	bytes_read = 0;
1655 
1656 	/* Linux cannot have 0 count. */
1657 	if (count <= 0 || current_offset < 0) {
1658 		error = EINVAL;
1659 		goto drop;
1660 	}
1661 
1662 	error = fo_sendfile(fp, out, NULL, NULL, current_offset, count,
1663 	    &bytes_read, 0, td);
1664 	if (error != 0)
1665 		goto drop;
1666 	current_offset += bytes_read;
1667 
1668 	if (offset != NULL) {
1669 		*offset = current_offset;
1670 	} else {
1671 		error = fo_seek(fp, current_offset, SEEK_SET, td);
1672 		if (error != 0)
1673 			goto drop;
1674 	}
1675 
1676 	td->td_retval[0] = (ssize_t)bytes_read;
1677 drop:
1678 	fdrop(fp, td);
1679 	return (error);
1680 }
1681 
1682 int
1683 linux_sendfile(struct thread *td, struct linux_sendfile_args *arg)
1684 {
1685 	/*
1686 	 * Differences between FreeBSD and Linux sendfile:
1687 	 * - Linux doesn't send anything when count is 0 (FreeBSD uses 0 to
1688 	 *   mean send the whole file.)  In linux_sendfile given fds are still
1689 	 *   checked for validity when the count is 0.
1690 	 * - Linux can send to any fd whereas FreeBSD only supports sockets.
1691 	 *   The same restriction follows for linux_sendfile.
1692 	 * - Linux doesn't have an equivalent for FreeBSD's flags and sf_hdtr.
1693 	 * - Linux takes an offset pointer and updates it to the read location.
1694 	 *   FreeBSD takes in an offset and a 'bytes read' parameter which is
1695 	 *   only filled if it isn't NULL.  We use this parameter to update the
1696 	 *   offset pointer if it exists.
1697 	 * - Linux sendfile returns bytes read on success while FreeBSD
1698 	 *   returns 0.  We use the 'bytes read' parameter to get this value.
1699 	 */
1700 
1701 	l_loff_t offset64;
1702 	l_long offset;
1703 	int ret;
1704 	int error;
1705 
1706 	if (arg->offset != NULL) {
1707 		error = copyin(arg->offset, &offset, sizeof(offset));
1708 		if (error != 0)
1709 			return (error);
1710 		offset64 = (l_loff_t)offset;
1711 	}
1712 
1713 	ret = linux_sendfile_common(td, arg->out, arg->in,
1714 	    arg->offset != NULL ? &offset64 : NULL, arg->count);
1715 
1716 	if (arg->offset != NULL) {
1717 #if defined(__i386__) || defined(__arm__) || \
1718     (defined(__amd64__) && defined(COMPAT_LINUX32))
1719 		if (offset64 > INT32_MAX)
1720 			return (EOVERFLOW);
1721 #endif
1722 		offset = (l_long)offset64;
1723 		error = copyout(&offset, arg->offset, sizeof(offset));
1724 		if (error != 0)
1725 			return (error);
1726 	}
1727 
1728 	return (ret);
1729 }
1730 
1731 #if defined(__i386__) || defined(__arm__) || \
1732     (defined(__amd64__) && defined(COMPAT_LINUX32))
1733 
1734 int
1735 linux_sendfile64(struct thread *td, struct linux_sendfile64_args *arg)
1736 {
1737 	l_loff_t offset;
1738 	int ret;
1739 	int error;
1740 
1741 	if (arg->offset != NULL) {
1742 		error = copyin(arg->offset, &offset, sizeof(offset));
1743 		if (error != 0)
1744 			return (error);
1745 	}
1746 
1747 	ret = linux_sendfile_common(td, arg->out, arg->in,
1748 		arg->offset != NULL ? &offset : NULL, arg->count);
1749 
1750 	if (arg->offset != NULL) {
1751 		error = copyout(&offset, arg->offset, sizeof(offset));
1752 		if (error != 0)
1753 			return (error);
1754 	}
1755 
1756 	return (ret);
1757 }
1758 
1759 /* Argument list sizes for linux_socketcall */
1760 static const unsigned char lxs_args_cnt[] = {
1761 	0 /* unused*/,		3 /* socket */,
1762 	3 /* bind */,		3 /* connect */,
1763 	2 /* listen */,		3 /* accept */,
1764 	3 /* getsockname */,	3 /* getpeername */,
1765 	4 /* socketpair */,	4 /* send */,
1766 	4 /* recv */,		6 /* sendto */,
1767 	6 /* recvfrom */,	2 /* shutdown */,
1768 	5 /* setsockopt */,	5 /* getsockopt */,
1769 	3 /* sendmsg */,	3 /* recvmsg */,
1770 	4 /* accept4 */,	5 /* recvmmsg */,
1771 	4 /* sendmmsg */,	4 /* sendfile */
1772 };
1773 #define	LINUX_ARGS_CNT		(nitems(lxs_args_cnt) - 1)
1774 #define	LINUX_ARG_SIZE(x)	(lxs_args_cnt[x] * sizeof(l_ulong))
1775 
1776 int
1777 linux_socketcall(struct thread *td, struct linux_socketcall_args *args)
1778 {
1779 	l_ulong a[6];
1780 #if defined(__amd64__) && defined(COMPAT_LINUX32)
1781 	register_t l_args[6];
1782 #endif
1783 	void *arg;
1784 	int error;
1785 
1786 	if (args->what < LINUX_SOCKET || args->what > LINUX_ARGS_CNT)
1787 		return (EINVAL);
1788 	error = copyin(PTRIN(args->args), a, LINUX_ARG_SIZE(args->what));
1789 	if (error != 0)
1790 		return (error);
1791 
1792 #if defined(__amd64__) && defined(COMPAT_LINUX32)
1793 	for (int i = 0; i < lxs_args_cnt[args->what]; ++i)
1794 		l_args[i] = a[i];
1795 	arg = l_args;
1796 #else
1797 	arg = a;
1798 #endif
1799 	switch (args->what) {
1800 	case LINUX_SOCKET:
1801 		return (linux_socket(td, arg));
1802 	case LINUX_BIND:
1803 		return (linux_bind(td, arg));
1804 	case LINUX_CONNECT:
1805 		return (linux_connect(td, arg));
1806 	case LINUX_LISTEN:
1807 		return (linux_listen(td, arg));
1808 	case LINUX_ACCEPT:
1809 		return (linux_accept(td, arg));
1810 	case LINUX_GETSOCKNAME:
1811 		return (linux_getsockname(td, arg));
1812 	case LINUX_GETPEERNAME:
1813 		return (linux_getpeername(td, arg));
1814 	case LINUX_SOCKETPAIR:
1815 		return (linux_socketpair(td, arg));
1816 	case LINUX_SEND:
1817 		return (linux_send(td, arg));
1818 	case LINUX_RECV:
1819 		return (linux_recv(td, arg));
1820 	case LINUX_SENDTO:
1821 		return (linux_sendto(td, arg));
1822 	case LINUX_RECVFROM:
1823 		return (linux_recvfrom(td, arg));
1824 	case LINUX_SHUTDOWN:
1825 		return (linux_shutdown(td, arg));
1826 	case LINUX_SETSOCKOPT:
1827 		return (linux_setsockopt(td, arg));
1828 	case LINUX_GETSOCKOPT:
1829 		return (linux_getsockopt(td, arg));
1830 	case LINUX_SENDMSG:
1831 		return (linux_sendmsg(td, arg));
1832 	case LINUX_RECVMSG:
1833 		return (linux_recvmsg(td, arg));
1834 	case LINUX_ACCEPT4:
1835 		return (linux_accept4(td, arg));
1836 	case LINUX_RECVMMSG:
1837 		return (linux_recvmmsg(td, arg));
1838 	case LINUX_SENDMMSG:
1839 		return (linux_sendmmsg(td, arg));
1840 	case LINUX_SENDFILE:
1841 		return (linux_sendfile(td, arg));
1842 	}
1843 
1844 	linux_msg(td, "socket type %d not implemented", args->what);
1845 	return (ENOSYS);
1846 }
1847 #endif /* __i386__ || __arm__ || (__amd64__ && COMPAT_LINUX32) */
1848