xref: /netbsd-src/sys/compat/linux/common/linux_socket.c (revision 181254a7b1bdde6873432bffef2d2decc4b5c22f)
1 /*	$NetBSD: linux_socket.c,v 1.150 2020/07/16 15:02:08 msaitoh Exp $	*/
2 
3 /*-
4  * Copyright (c) 1995, 1998, 2008 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Frank van der Linden and Eric Haszlakiewicz.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29  * POSSIBILITY OF SUCH DAMAGE.
30  */
31 
32 /*
33  * Functions in multiarch:
34  *	linux_sys_socketcall		: linux_socketcall.c
35  */
36 
37 #include <sys/cdefs.h>
38 __KERNEL_RCSID(0, "$NetBSD: linux_socket.c,v 1.150 2020/07/16 15:02:08 msaitoh Exp $");
39 
40 #if defined(_KERNEL_OPT)
41 #include "opt_inet.h"
42 #endif /* defined(_KERNEL_OPT) */
43 
44 #include <sys/param.h>
45 #include <sys/kernel.h>
46 #include <sys/systm.h>
47 #include <sys/buf.h>
48 #include <sys/ioctl.h>
49 #include <sys/tty.h>
50 #include <sys/file.h>
51 #include <sys/filedesc.h>
52 #include <sys/select.h>
53 #include <sys/socket.h>
54 #include <sys/socketvar.h>
55 #include <sys/domain.h>
56 #include <net/if.h>
57 #include <net/if_dl.h>
58 #include <net/if_types.h>
59 #include <netinet/in.h>
60 #include <netinet/tcp.h>
61 #include <sys/mount.h>
62 #include <sys/proc.h>
63 #include <sys/vnode.h>
64 #include <sys/device.h>
65 #include <sys/protosw.h>
66 #include <sys/mbuf.h>
67 #include <sys/syslog.h>
68 #include <sys/exec.h>
69 #include <sys/kauth.h>
70 #include <sys/syscallargs.h>
71 #include <sys/ktrace.h>
72 
73 #include <lib/libkern/libkern.h>
74 
75 #include <netinet/ip6.h>
76 #include <netinet6/ip6_var.h>
77 
78 #include <compat/sys/socket.h>
79 #include <compat/sys/sockio.h>
80 
81 #include <compat/linux/common/linux_types.h>
82 #include <compat/linux/common/linux_util.h>
83 #include <compat/linux/common/linux_signal.h>
84 #include <compat/linux/common/linux_ioctl.h>
85 #include <compat/linux/common/linux_sched.h>
86 #include <compat/linux/common/linux_socket.h>
87 #include <compat/linux/common/linux_fcntl.h>
88 #if !defined(__alpha__) && !defined(__amd64__)
89 #include <compat/linux/common/linux_socketcall.h>
90 #endif
91 #include <compat/linux/common/linux_sockio.h>
92 #include <compat/linux/common/linux_ipc.h>
93 #include <compat/linux/common/linux_sem.h>
94 
95 #include <compat/linux/linux_syscallargs.h>
96 
97 #ifdef DEBUG_LINUX
98 #define DPRINTF(a) uprintf a
99 #else
100 #define DPRINTF(a)
101 #endif
102 
103 /*
104  * The calls in this file are entered either via the linux_socketcall()
105  * interface or, on the Alpha, as individual syscalls.  The
106  * linux_socketcall function does any massaging of arguments so that all
107  * the calls in here need not think that they are anything other
108  * than a normal syscall.
109  */
110 
111 static int linux_to_bsd_domain(int);
112 static int bsd_to_linux_domain(int);
113 static int linux_to_bsd_type(int);
114 int linux_to_bsd_sopt_level(int);
115 int linux_to_bsd_so_sockopt(int);
116 int linux_to_bsd_ip_sockopt(int);
117 int linux_to_bsd_ipv6_sockopt(int);
118 int linux_to_bsd_tcp_sockopt(int);
119 int linux_to_bsd_udp_sockopt(int);
120 int linux_getifname(struct lwp *, register_t *, void *);
121 int linux_getifconf(struct lwp *, register_t *, void *);
122 int linux_getifhwaddr(struct lwp *, register_t *, u_int, void *);
123 static int linux_get_sa(struct lwp *, int, struct sockaddr_big *,
124 		const struct osockaddr *, socklen_t);
125 static int linux_sa_put(struct osockaddr *osa);
126 static int linux_to_bsd_msg_flags(int);
127 static int bsd_to_linux_msg_flags(int);
128 static void linux_to_bsd_msghdr(const struct linux_msghdr *, struct msghdr *);
129 static void bsd_to_linux_msghdr(const struct msghdr *, struct linux_msghdr *);
130 
131 static const int linux_to_bsd_domain_[LINUX_AF_MAX] = {
132 	AF_UNSPEC,
133 	AF_UNIX,
134 	AF_INET,
135 	AF_CCITT,	/* LINUX_AF_AX25 */
136 	AF_IPX,
137 	AF_APPLETALK,
138 	-1,		/* LINUX_AF_NETROM */
139 	-1,		/* LINUX_AF_BRIDGE */
140 	-1,		/* LINUX_AF_ATMPVC */
141 	AF_CCITT,	/* LINUX_AF_X25 */
142 	AF_INET6,
143 	-1,		/* LINUX_AF_ROSE */
144 	AF_DECnet,
145 	-1,		/* LINUX_AF_NETBEUI */
146 	-1,		/* LINUX_AF_SECURITY */
147 	pseudo_AF_KEY,
148 	AF_ROUTE,	/* LINUX_AF_NETLINK */
149 	-1,		/* LINUX_AF_PACKET */
150 	-1,		/* LINUX_AF_ASH */
151 	-1,		/* LINUX_AF_ECONET */
152 	-1,		/* LINUX_AF_ATMSVC */
153 	AF_SNA,
154 	/* rest up to LINUX_AF_MAX-1 is not allocated */
155 	-1, -1, -1, -1, -1, -1, -1, -1, -1, -1,
156 };
157 
158 static const int bsd_to_linux_domain_[AF_MAX] = {
159 	LINUX_AF_UNSPEC,
160 	LINUX_AF_UNIX,
161 	LINUX_AF_INET,
162 	-1,		/* AF_IMPLINK */
163 	-1,		/* AF_PUP */
164 	-1,		/* AF_CHAOS */
165 	-1,		/* AF_NS */
166 	-1,		/* AF_ISO */
167 	-1,		/* AF_ECMA */
168 	-1,		/* AF_DATAKIT */
169 	LINUX_AF_AX25,	/* AF_CCITT */
170 	LINUX_AF_SNA,
171 	LINUX_AF_DECnet,
172 	-1,		/* AF_DLI */
173 	-1,		/* AF_LAT */
174 	-1,		/* AF_HYLINK */
175 	LINUX_AF_APPLETALK,
176 	LINUX_AF_NETLINK,
177 	-1,		/* AF_LINK */
178 	-1,		/* AF_XTP */
179 	-1,		/* AF_COIP */
180 	-1,		/* AF_CNT */
181 	-1,		/* pseudo_AF_RTIP */
182 	LINUX_AF_IPX,
183 	LINUX_AF_INET6,
184 	-1,		/* pseudo_AF_PIP */
185 	-1,		/* AF_ISDN */
186 	-1,		/* AF_NATM */
187 	-1,		/* AF_ARP */
188 	LINUX_pseudo_AF_KEY,
189 	-1,		/* pseudo_AF_HDRCMPLT */
190 };
191 
192 static const struct {
193 	int bfl;
194 	int lfl;
195 } bsd_to_linux_msg_flags_[] = {
196 	{MSG_OOB,		LINUX_MSG_OOB},
197 	{MSG_PEEK,		LINUX_MSG_PEEK},
198 	{MSG_DONTROUTE,		LINUX_MSG_DONTROUTE},
199 	{MSG_EOR,		LINUX_MSG_EOR},
200 	{MSG_TRUNC,		LINUX_MSG_TRUNC},
201 	{MSG_CTRUNC,		LINUX_MSG_CTRUNC},
202 	{MSG_WAITALL,		LINUX_MSG_WAITALL},
203 	{MSG_DONTWAIT,		LINUX_MSG_DONTWAIT},
204 	{MSG_BCAST,		0},		/* not supported, clear */
205 	{MSG_MCAST,		0},		/* not supported, clear */
206 	{MSG_NOSIGNAL,		LINUX_MSG_NOSIGNAL},
207 	{-1, /* not supp */	LINUX_MSG_PROBE},
208 	{-1, /* not supp */	LINUX_MSG_FIN},
209 	{-1, /* not supp */	LINUX_MSG_SYN},
210 	{-1, /* not supp */	LINUX_MSG_CONFIRM},
211 	{-1, /* not supp */	LINUX_MSG_RST},
212 	{-1, /* not supp */	LINUX_MSG_ERRQUEUE},
213 	{-1, /* not supp */	LINUX_MSG_MORE},
214 };
215 
216 /*
217  * Convert between Linux and BSD socket domain values
218  */
219 static int
220 linux_to_bsd_domain(int ldom)
221 {
222 	if (ldom < 0 || ldom >= LINUX_AF_MAX)
223 		return (-1);
224 
225 	return linux_to_bsd_domain_[ldom];
226 }
227 
228 /*
229  * Convert between BSD and Linux socket domain values
230  */
231 static int
232 bsd_to_linux_domain(int bdom)
233 {
234 	if (bdom < 0 || bdom >= AF_MAX)
235 		return (-1);
236 
237 	return bsd_to_linux_domain_[bdom];
238 }
239 
240 static int
241 linux_to_bsd_type(int ltype)
242 {
243 	int type, flags;
244 
245 	/* Real types are identical between Linux and NetBSD */
246 	type = ltype & LINUX_SOCK_TYPE_MASK;
247 
248 	/* But flags are not .. */
249 	flags = ltype & ~LINUX_SOCK_TYPE_MASK;
250 	if (flags & ~(LINUX_SOCK_CLOEXEC|LINUX_SOCK_NONBLOCK))
251 		return -1;
252 
253 	if (flags & LINUX_SOCK_CLOEXEC)
254 		type |= SOCK_CLOEXEC;
255 	if (flags & LINUX_SOCK_NONBLOCK)
256 		type |= SOCK_NONBLOCK;
257 
258 	return type;
259 }
260 
261 static int
262 linux_to_bsd_msg_flags(int lflag)
263 {
264 	int i, lfl, bfl;
265 	int bflag = 0;
266 
267 	if (lflag == 0)
268 		return (0);
269 
270 	for(i = 0; i < __arraycount(bsd_to_linux_msg_flags_); i++) {
271 		bfl = bsd_to_linux_msg_flags_[i].bfl;
272 		lfl = bsd_to_linux_msg_flags_[i].lfl;
273 
274 		if (lfl == 0)
275 			continue;
276 
277 		if (lflag & lfl) {
278 			if (bfl < 0)
279 				return (-1);
280 
281 			bflag |= bfl;
282 		}
283 	}
284 
285 	return (bflag);
286 }
287 
288 static int
289 bsd_to_linux_msg_flags(int bflag)
290 {
291 	int i, lfl, bfl;
292 	int lflag = 0;
293 
294 	if (bflag == 0)
295 		return (0);
296 
297 	for(i = 0; i < __arraycount(bsd_to_linux_msg_flags_); i++) {
298 		bfl = bsd_to_linux_msg_flags_[i].bfl;
299 		lfl = bsd_to_linux_msg_flags_[i].lfl;
300 
301 		if (bfl <= 0)
302 			continue;
303 
304 		if (bflag & bfl) {
305 			if (lfl < 0)
306 				return (-1);
307 
308 			lflag |= lfl;
309 		}
310 	}
311 
312 	return (lflag);
313 }
314 
315 int
316 linux_sys_socket(struct lwp *l, const struct linux_sys_socket_args *uap, register_t *retval)
317 {
318 	/* {
319 		syscallarg(int)	domain;
320 		syscallarg(int)	type;
321 		syscallarg(int) protocol;
322 	} */
323 	struct sys___socket30_args bsa;
324 	int error;
325 
326 
327 	SCARG(&bsa, protocol) = SCARG(uap, protocol);
328 	SCARG(&bsa, domain) = linux_to_bsd_domain(SCARG(uap, domain));
329 	if (SCARG(&bsa, domain) == -1)
330 		return EINVAL;
331 	SCARG(&bsa, type) = linux_to_bsd_type(SCARG(uap, type));
332 	if (SCARG(&bsa, type) == -1)
333 		return EINVAL;
334 	/*
335 	 * Apparently linux uses this to talk to ISDN sockets. If we fail
336 	 * now programs seems to handle it, but if we don't we are going
337 	 * to fail when we bind and programs don't handle this well.
338 	 */
339 	if (SCARG(&bsa, domain) == AF_ROUTE && SCARG(&bsa, type) == SOCK_RAW)
340 		return ENOTSUP;
341 	error = sys___socket30(l, &bsa, retval);
342 
343 #ifdef INET6
344 	/*
345 	 * Linux AF_INET6 socket has IPV6_V6ONLY setsockopt set to 0 by
346 	 * default and some apps depend on this. So, set V6ONLY to 0
347 	 * for Linux apps if the sysctl value is set to 1.
348 	 */
349 	if (!error && ip6_v6only && SCARG(&bsa, domain) == PF_INET6) {
350 		struct socket *so;
351 
352 		if (fd_getsock(*retval, &so) == 0) {
353 			int val = 0;
354 
355 			/* ignore error */
356 			(void)so_setsockopt(l, so, IPPROTO_IPV6, IPV6_V6ONLY,
357 			    &val, sizeof(val));
358 
359 			fd_putfile(*retval);
360 		}
361 	}
362 #endif
363 
364 	return (error);
365 }
366 
367 int
368 linux_sys_socketpair(struct lwp *l, const struct linux_sys_socketpair_args *uap, register_t *retval)
369 {
370 	/* {
371 		syscallarg(int) domain;
372 		syscallarg(int) type;
373 		syscallarg(int) protocol;
374 		syscallarg(int *) rsv;
375 	} */
376 	struct sys_socketpair_args bsa;
377 
378 	SCARG(&bsa, domain) = linux_to_bsd_domain(SCARG(uap, domain));
379 	if (SCARG(&bsa, domain) == -1)
380 		return EINVAL;
381 	SCARG(&bsa, type) = linux_to_bsd_type(SCARG(uap, type));
382 	if (SCARG(&bsa, type) == -1)
383 		return EINVAL;
384 	SCARG(&bsa, protocol) = SCARG(uap, protocol);
385 	SCARG(&bsa, rsv) = SCARG(uap, rsv);
386 
387 	return sys_socketpair(l, &bsa, retval);
388 }
389 
390 int
391 linux_sys_sendto(struct lwp *l, const struct linux_sys_sendto_args *uap, register_t *retval)
392 {
393 	/* {
394 		syscallarg(int)				s;
395 		syscallarg(void *)			msg;
396 		syscallarg(int)				len;
397 		syscallarg(int)				flags;
398 		syscallarg(struct osockaddr *)		to;
399 		syscallarg(int)				tolen;
400 	} */
401 	struct msghdr   msg;
402 	struct iovec    aiov;
403 	struct sockaddr_big nam;
404 	struct mbuf *m;
405 	int bflags;
406 	int error;
407 
408 	/* Translate message flags.  */
409 	bflags = linux_to_bsd_msg_flags(SCARG(uap, flags));
410 	if (bflags < 0)
411 		/* Some supported flag */
412 		return EINVAL;
413 
414 	msg.msg_flags = 0;
415 	msg.msg_name = NULL;
416 	msg.msg_control = NULL;
417 
418 	if (SCARG(uap, tolen)) {
419 		/* Read in and convert the sockaddr */
420 		error = linux_get_sa(l, SCARG(uap, s), &nam, SCARG(uap, to),
421 		    SCARG(uap, tolen));
422 		if (error)
423 			return error;
424 		error = sockargs(&m, &nam, nam.sb_len, UIO_SYSSPACE, MT_SONAME);
425 		if (error)
426 			return error;
427 		msg.msg_flags |= MSG_NAMEMBUF;
428 		msg.msg_name = m;
429 		msg.msg_namelen = nam.sb_len;
430 	}
431 
432 	msg.msg_iov = &aiov;
433 	msg.msg_iovlen = 1;
434 	aiov.iov_base = __UNCONST(SCARG(uap, msg));
435 	aiov.iov_len = SCARG(uap, len);
436 
437 	return do_sys_sendmsg(l, SCARG(uap, s), &msg, bflags, retval);
438 }
439 
440 static void
441 linux_to_bsd_msghdr(const struct linux_msghdr *lmsg, struct msghdr *bmsg)
442 {
443 	bmsg->msg_name = lmsg->msg_name;
444 	bmsg->msg_namelen = lmsg->msg_namelen;
445 	bmsg->msg_iov = lmsg->msg_iov;
446 	bmsg->msg_iovlen = lmsg->msg_iovlen;
447 	bmsg->msg_control = lmsg->msg_control;
448 	bmsg->msg_controllen = lmsg->msg_controllen;
449 	bmsg->msg_flags = lmsg->msg_flags;
450 }
451 
452 static void
453 bsd_to_linux_msghdr(const struct msghdr *bmsg, struct linux_msghdr *lmsg)
454 {
455 	lmsg->msg_name = bmsg->msg_name;
456 	lmsg->msg_namelen = bmsg->msg_namelen;
457 	lmsg->msg_iov = bmsg->msg_iov;
458 	lmsg->msg_iovlen = bmsg->msg_iovlen;
459 	lmsg->msg_control = bmsg->msg_control;
460 	lmsg->msg_controllen = bmsg->msg_controllen;
461 	lmsg->msg_flags = bmsg->msg_flags;
462 }
463 
464 int
465 linux_sys_sendmsg(struct lwp *l, const struct linux_sys_sendmsg_args *uap, register_t *retval)
466 {
467 	/* {
468 		syscallarg(int) s;
469 		syscallarg(struct linux_msghdr *) msg;
470 		syscallarg(u_int) flags;
471 	} */
472 	struct msghdr	msg;
473 	struct linux_msghdr lmsg;
474 	int		error;
475 	int		bflags;
476 	struct sockaddr_big nam;
477 	u_int8_t	*control;
478 	struct mbuf     *ctl_mbuf = NULL;
479 
480 	error = copyin(SCARG(uap, msg), &lmsg, sizeof(lmsg));
481 	if (error)
482 		return error;
483 	linux_to_bsd_msghdr(&lmsg, &msg);
484 
485 	msg.msg_flags = MSG_IOVUSRSPACE;
486 
487 	/*
488 	 * Translate message flags.
489 	 */
490 	bflags = linux_to_bsd_msg_flags(SCARG(uap, flags));
491 	if (bflags < 0)
492 		/* Some supported flag */
493 		return EINVAL;
494 
495 	if (lmsg.msg_name) {
496 		/* Read in and convert the sockaddr */
497 		error = linux_get_sa(l, SCARG(uap, s), &nam, msg.msg_name,
498 		    msg.msg_namelen);
499 		if (error)
500 			return (error);
501 		msg.msg_name = &nam;
502 	}
503 
504 	/*
505 	 * Handle cmsg if there is any.
506 	 */
507 	if (LINUX_CMSG_FIRSTHDR(&lmsg)) {
508 		struct linux_cmsghdr l_cmsg, *l_cc;
509 		struct cmsghdr *cmsg;
510 		ssize_t resid = msg.msg_controllen;
511 		size_t clen, cidx = 0, cspace;
512 
513 		ctl_mbuf = m_get(M_WAIT, MT_CONTROL);
514 		clen = MLEN;
515 		control = mtod(ctl_mbuf, void *);
516 
517 		l_cc = LINUX_CMSG_FIRSTHDR(&lmsg);
518 		do {
519 			error = copyin(l_cc, &l_cmsg, sizeof(l_cmsg));
520 			if (error)
521 				goto done;
522 
523 			/*
524 			 * Sanity check the control message length.
525 			 */
526 			if (l_cmsg.cmsg_len > resid
527 			    || l_cmsg.cmsg_len < sizeof l_cmsg) {
528 				error = EINVAL;
529 				goto done;
530 			}
531 
532 			/*
533 			 * Refuse unsupported control messages, and
534 			 * translate fields as appropriate.
535 			 */
536 			switch (l_cmsg.cmsg_level) {
537 			case LINUX_SOL_SOCKET:
538 				/* It only differs on some archs */
539 				if (LINUX_SOL_SOCKET != SOL_SOCKET)
540 					l_cmsg.cmsg_level = SOL_SOCKET;
541 
542 				switch(l_cmsg.cmsg_type) {
543 				case LINUX_SCM_RIGHTS:
544 					/* Linux SCM_RIGHTS is same as NetBSD */
545 					break;
546 
547 				case LINUX_SCM_CREDENTIALS:
548 					/* no native equivalent, just drop it */
549 					if (control != mtod(ctl_mbuf, void *))
550 						free(control, M_MBUF);
551 					m_free(ctl_mbuf);
552 					ctl_mbuf = NULL;
553 					msg.msg_control = NULL;
554 					msg.msg_controllen = 0;
555 					goto skipcmsg;
556 
557 				default:
558 					/* other types not supported */
559 					error = EINVAL;
560 					goto done;
561 				}
562 				break;
563 			default:
564 				/* pray and leave intact */
565 				break;
566 			}
567 
568 			cspace = CMSG_SPACE(l_cmsg.cmsg_len - sizeof(l_cmsg));
569 
570 			/* Check the buffer is big enough */
571 			if (__predict_false(cidx + cspace > clen)) {
572 				u_int8_t *nc;
573 				size_t nclen;
574 
575 				nclen = cidx + cspace;
576 				if (nclen >= PAGE_SIZE) {
577 					error = EINVAL;
578 					goto done;
579 				}
580 				nc = realloc(clen <= MLEN ? NULL : control,
581 						nclen, M_TEMP, M_WAITOK);
582 				if (!nc) {
583 					error = ENOMEM;
584 					goto done;
585 				}
586 				if (cidx <= MLEN)
587 					/* Old buffer was in mbuf... */
588 					memcpy(nc, control, cidx);
589 				control = nc;
590 				clen = nclen;
591 			}
592 
593 			/* Copy header */
594 			cmsg = (void *)&control[cidx];
595 			cmsg->cmsg_len = l_cmsg.cmsg_len + LINUX_CMSG_ALIGN_DELTA;
596 			cmsg->cmsg_level = l_cmsg.cmsg_level;
597 			cmsg->cmsg_type = l_cmsg.cmsg_type;
598 
599 			/* Zero area between header and data */
600 			memset(cmsg + 1, 0,
601 				CMSG_ALIGN(sizeof(*cmsg)) - sizeof(*cmsg));
602 
603 			/* Copyin the data */
604 			error = copyin(LINUX_CMSG_DATA(l_cc),
605 				CMSG_DATA(cmsg),
606 				l_cmsg.cmsg_len - sizeof(l_cmsg));
607 			if (error)
608 				goto done;
609 
610 			resid -= LINUX_CMSG_ALIGN(l_cmsg.cmsg_len);
611 			cidx += cspace;
612 		} while ((l_cc = LINUX_CMSG_NXTHDR(&msg, l_cc, &l_cmsg)) && resid > 0);
613 
614 		/* If we allocated a buffer, attach to mbuf */
615 		if (cidx > MLEN) {
616 			MEXTADD(ctl_mbuf, control, clen, M_MBUF, NULL, NULL);
617 			ctl_mbuf->m_flags |= M_EXT_RW;
618 		}
619 		control = NULL;
620 		ctl_mbuf->m_len = cidx;
621 
622 		msg.msg_control = ctl_mbuf;
623 		msg.msg_flags |= MSG_CONTROLMBUF;
624 
625 		ktrkuser("mbcontrol", mtod(ctl_mbuf, void *),
626 		    msg.msg_controllen);
627 	}
628 
629 skipcmsg:
630 	error = do_sys_sendmsg(l, SCARG(uap, s), &msg, bflags, retval);
631 	/* Freed internally */
632 	ctl_mbuf = NULL;
633 
634 done:
635 	if (ctl_mbuf != NULL) {
636 		if (control != NULL && control != mtod(ctl_mbuf, void *))
637 			free(control, M_MBUF);
638 		m_free(ctl_mbuf);
639 	}
640 	return (error);
641 }
642 
643 int
644 linux_sys_recvfrom(struct lwp *l, const struct linux_sys_recvfrom_args *uap, register_t *retval)
645 {
646 	/* {
647 		syscallarg(int) s;
648 		syscallarg(void *) buf;
649 		syscallarg(int) len;
650 		syscallarg(int) flags;
651 		syscallarg(struct osockaddr *) from;
652 		syscallarg(int *) fromlenaddr;
653 	} */
654 	int		error;
655 	struct sys_recvfrom_args bra;
656 
657 	SCARG(&bra, s) = SCARG(uap, s);
658 	SCARG(&bra, buf) = SCARG(uap, buf);
659 	SCARG(&bra, len) = SCARG(uap, len);
660 	SCARG(&bra, flags) = SCARG(uap, flags);
661 	SCARG(&bra, from) = (struct sockaddr *) SCARG(uap, from);
662 	SCARG(&bra, fromlenaddr) = (socklen_t *)SCARG(uap, fromlenaddr);
663 
664 	if ((error = sys_recvfrom(l, &bra, retval)))
665 		return (error);
666 
667 	if (SCARG(uap, from) && (error = linux_sa_put(SCARG(uap, from))))
668 		return (error);
669 
670 	return (0);
671 }
672 
673 static int
674 linux_copyout_msg_control(struct lwp *l, struct msghdr *mp, struct mbuf *control)
675 {
676 	int dlen, error = 0;
677 	struct cmsghdr *cmsg;
678 	struct linux_cmsghdr linux_cmsg;
679 	struct mbuf *m;
680 	char *q, *q_end;
681 
682 	if (mp->msg_controllen <= 0 || control == 0) {
683 		mp->msg_controllen = 0;
684 		free_control_mbuf(l, control, control);
685 		return 0;
686 	}
687 
688 	ktrkuser("msgcontrol", mtod(control, void *), mp->msg_controllen);
689 
690 	q = (char *)mp->msg_control;
691 	q_end = q + mp->msg_controllen;
692 
693 	for (m = control; m != NULL; ) {
694 		cmsg = mtod(m, struct cmsghdr *);
695 
696 		/*
697 		 * Fixup cmsg. We handle two things:
698 		 * 0. different sizeof cmsg_len.
699 		 * 1. different values for level/type on some archs
700 		 * 2. different alignment of CMSG_DATA on some archs
701 		 */
702 		linux_cmsg.cmsg_len = cmsg->cmsg_len - LINUX_CMSG_ALIGN_DELTA;
703 		linux_cmsg.cmsg_level = cmsg->cmsg_level;
704 		linux_cmsg.cmsg_type = cmsg->cmsg_type;
705 
706 		dlen = q_end - q;
707 		if (linux_cmsg.cmsg_len > dlen) {
708 			/* Not enough room for the parameter */
709 			dlen -= sizeof linux_cmsg;
710 			if (dlen <= 0)
711 				/* Discard if header wont fit */
712 				break;
713 			mp->msg_flags |= MSG_CTRUNC;
714 			if (linux_cmsg.cmsg_level == SOL_SOCKET
715 			    && linux_cmsg.cmsg_type == SCM_RIGHTS)
716 				/* Do not truncate me ... */
717 				break;
718 		} else
719 			dlen = linux_cmsg.cmsg_len - sizeof linux_cmsg;
720 
721 		switch (linux_cmsg.cmsg_level) {
722 		case SOL_SOCKET:
723 			linux_cmsg.cmsg_level = LINUX_SOL_SOCKET;
724 			switch (linux_cmsg.cmsg_type) {
725 			case SCM_RIGHTS:
726 				/* Linux SCM_RIGHTS is same as NetBSD */
727 				break;
728 
729 			default:
730 				/* other types not supported */
731 				error = EINVAL;
732 				goto done;
733 			}
734 			/* machine dependent ! */
735 			break;
736 		default:
737 			/* pray and leave intact */
738 			break;
739 		}
740 
741 		/* There can be padding between the header and data... */
742 		error = copyout(&linux_cmsg, q, sizeof linux_cmsg);
743 		if (error != 0) {
744 			error = copyout(CCMSG_DATA(cmsg), q + sizeof linux_cmsg,
745 			    dlen);
746 		}
747 		if (error != 0) {
748 			/* We must free all the SCM_RIGHTS */
749 			m = control;
750 			break;
751 		}
752 		m = m->m_next;
753 		if (m == NULL || q + LINUX_CMSG_SPACE(dlen) > q_end) {
754 			q += LINUX_CMSG_LEN(dlen);
755 			break;
756 		}
757 		q += LINUX_CMSG_SPACE(dlen);
758 	}
759 
760   done:
761 	free_control_mbuf(l, control, m);
762 
763 	mp->msg_controllen = q - (char *)mp->msg_control;
764 	return error;
765 }
766 
767 int
768 linux_sys_recvmsg(struct lwp *l, const struct linux_sys_recvmsg_args *uap, register_t *retval)
769 {
770 	/* {
771 		syscallarg(int) s;
772 		syscallarg(struct linux_msghdr *) msg;
773 		syscallarg(u_int) flags;
774 	} */
775 	struct msghdr	msg;
776 	struct linux_msghdr lmsg;
777 	int		error;
778 	struct mbuf	*from, *control;
779 
780 	error = copyin(SCARG(uap, msg), &lmsg, sizeof(lmsg));
781 	if (error)
782 		return (error);
783 	linux_to_bsd_msghdr(&lmsg, &msg);
784 
785 	msg.msg_flags = linux_to_bsd_msg_flags(SCARG(uap, flags));
786 	if (msg.msg_flags < 0) {
787 		/* Some unsupported flag */
788 		return (EINVAL);
789 	}
790 	msg.msg_flags |= MSG_IOVUSRSPACE;
791 
792 	error = do_sys_recvmsg(l, SCARG(uap, s), &msg, &from,
793 	    msg.msg_control != NULL ? &control : NULL, retval);
794 	if (error != 0)
795 		return error;
796 
797 	if (msg.msg_control != NULL)
798 		error = linux_copyout_msg_control(l, &msg, control);
799 
800 	if (error == 0 && from != 0) {
801 		mtod(from, struct osockaddr *)->sa_family =
802 		    bsd_to_linux_domain(mtod(from, struct sockaddr *)->sa_family);
803 		error = copyout_sockname(msg.msg_name, &msg.msg_namelen, 0,
804 			from);
805 	} else
806 		msg.msg_namelen = 0;
807 
808 	if (from != NULL)
809 		m_free(from);
810 
811 	if (error == 0) {
812 		msg.msg_flags = bsd_to_linux_msg_flags(msg.msg_flags);
813 		if (msg.msg_flags < 0)
814 			/* Some flag unsupported by Linux */
815 			error = EINVAL;
816 		else {
817 			ktrkuser("msghdr", &msg, sizeof(msg));
818 			bsd_to_linux_msghdr(&msg, &lmsg);
819 			error = copyout(&lmsg, SCARG(uap, msg), sizeof(lmsg));
820 		}
821 	}
822 
823 	return (error);
824 }
825 
826 /*
827  * Convert socket option level from Linux to NetBSD value. Only SOL_SOCKET
828  * is different, the rest matches IPPROTO_* on both systems.
829  */
830 int
831 linux_to_bsd_sopt_level(int llevel)
832 {
833 
834 	switch (llevel) {
835 	case LINUX_SOL_SOCKET:
836 		return SOL_SOCKET;
837 	case LINUX_SOL_IP:
838 		return IPPROTO_IP;
839 #ifdef INET6
840 	case LINUX_SOL_IPV6:
841 		return IPPROTO_IPV6;
842 #endif
843 	case LINUX_SOL_TCP:
844 		return IPPROTO_TCP;
845 	case LINUX_SOL_UDP:
846 		return IPPROTO_UDP;
847 	default:
848 		return -1;
849 	}
850 }
851 
852 /*
853  * Convert Linux socket level socket option numbers to NetBSD values.
854  */
855 int
856 linux_to_bsd_so_sockopt(int lopt)
857 {
858 
859 	switch (lopt) {
860 	case LINUX_SO_DEBUG:
861 		return SO_DEBUG;
862 	case LINUX_SO_REUSEADDR:
863 		/*
864 		 * Linux does not implement SO_REUSEPORT, but allows reuse of a
865 		 * host:port pair through SO_REUSEADDR even if the address is not a
866 		 * multicast-address.  Effectively, this means that we should use
867 		 * SO_REUSEPORT to allow Linux applications to not exit with
868 		 * EADDRINUSE
869 		 */
870 		return SO_REUSEPORT;
871 	case LINUX_SO_TYPE:
872 		return SO_TYPE;
873 	case LINUX_SO_ERROR:
874 		return SO_ERROR;
875 	case LINUX_SO_DONTROUTE:
876 		return SO_DONTROUTE;
877 	case LINUX_SO_BROADCAST:
878 		return SO_BROADCAST;
879 	case LINUX_SO_SNDBUF:
880 		return SO_SNDBUF;
881 	case LINUX_SO_RCVBUF:
882 		return SO_RCVBUF;
883 	case LINUX_SO_SNDLOWAT:
884 		return SO_SNDLOWAT;
885 	case LINUX_SO_RCVLOWAT:
886 		return SO_RCVLOWAT;
887 	case LINUX_SO_KEEPALIVE:
888 		return SO_KEEPALIVE;
889 	case LINUX_SO_OOBINLINE:
890 		return SO_OOBINLINE;
891 	case LINUX_SO_LINGER:
892 		return SO_LINGER;
893 	case LINUX_SO_ACCEPTCONN:
894 		return SO_ACCEPTCONN;
895 	case LINUX_SO_PRIORITY:
896 	case LINUX_SO_NO_CHECK:
897 	default:
898 		return -1;
899 	}
900 }
901 
902 /*
903  * Convert Linux IP level socket option number to NetBSD values.
904  */
905 int
906 linux_to_bsd_ip_sockopt(int lopt)
907 {
908 
909 	switch (lopt) {
910 	case LINUX_IP_TOS:
911 		return IP_TOS;
912 	case LINUX_IP_TTL:
913 		return IP_TTL;
914 	case LINUX_IP_HDRINCL:
915 		return IP_HDRINCL;
916 	case LINUX_IP_MULTICAST_TTL:
917 		return IP_MULTICAST_TTL;
918 	case LINUX_IP_MULTICAST_LOOP:
919 		return IP_MULTICAST_LOOP;
920 	case LINUX_IP_MULTICAST_IF:
921 		return IP_MULTICAST_IF;
922 	case LINUX_IP_ADD_MEMBERSHIP:
923 		return IP_ADD_MEMBERSHIP;
924 	case LINUX_IP_DROP_MEMBERSHIP:
925 		return IP_DROP_MEMBERSHIP;
926 	default:
927 		return -1;
928 	}
929 }
930 
931 /*
932  * Convert Linux IPV6 level socket option number to NetBSD values.
933  */
934 #ifdef INET6
935 int
936 linux_to_bsd_ipv6_sockopt(int lopt)
937 {
938 
939 	switch (lopt) {
940 	case LINUX_IPV6_V6ONLY:
941 		return IPV6_V6ONLY;
942 	default:
943 		return -1;
944 	}
945 }
946 #endif
947 
948 /*
949  * Convert Linux TCP level socket option number to NetBSD values.
950  */
951 int
952 linux_to_bsd_tcp_sockopt(int lopt)
953 {
954 
955 	switch (lopt) {
956 	case LINUX_TCP_NODELAY:
957 		return TCP_NODELAY;
958 	case LINUX_TCP_MAXSEG:
959 		return TCP_MAXSEG;
960 	default:
961 		return -1;
962 	}
963 }
964 
965 /*
966  * Convert Linux UDP level socket option number to NetBSD values.
967  */
968 int
969 linux_to_bsd_udp_sockopt(int lopt)
970 {
971 
972 	switch (lopt) {
973 	default:
974 		return -1;
975 	}
976 }
977 
978 /*
979  * Another reasonably straightforward function: setsockopt(2).
980  * The level and option numbers are converted; the values passed
981  * are not (yet) converted, the ones currently implemented don't
982  * need conversion, as they are the same on both systems.
983  */
984 int
985 linux_sys_setsockopt(struct lwp *l, const struct linux_sys_setsockopt_args *uap, register_t *retval)
986 {
987 	/* {
988 		syscallarg(int) s;
989 		syscallarg(int) level;
990 		syscallarg(int) optname;
991 		syscallarg(void *) optval;
992 		syscallarg(int) optlen;
993 	} */
994 	struct sys_setsockopt_args bsa;
995 	int name;
996 
997 	SCARG(&bsa, s) = SCARG(uap, s);
998 	SCARG(&bsa, level) = linux_to_bsd_sopt_level(SCARG(uap, level));
999 	SCARG(&bsa, val) = SCARG(uap, optval);
1000 	SCARG(&bsa, valsize) = SCARG(uap, optlen);
1001 
1002 	/*
1003 	 * Linux supports only SOL_SOCKET for AF_LOCAL domain sockets
1004 	 * and returns EOPNOTSUPP for other levels
1005 	 */
1006 	if (SCARG(&bsa, level) != SOL_SOCKET) {
1007 		struct socket *so;
1008 		int error, family;
1009 
1010 		/* fd_getsock() will use the descriptor for us */
1011 	    	if ((error = fd_getsock(SCARG(&bsa, s), &so)) != 0)
1012 		    	return error;
1013 		family = so->so_proto->pr_domain->dom_family;
1014 		fd_putfile(SCARG(&bsa, s));
1015 
1016 		if (family == AF_LOCAL)
1017 			return EOPNOTSUPP;
1018 	}
1019 
1020 	switch (SCARG(&bsa, level)) {
1021 	case SOL_SOCKET:
1022 		name = linux_to_bsd_so_sockopt(SCARG(uap, optname));
1023 		break;
1024 	case IPPROTO_IP:
1025 		name = linux_to_bsd_ip_sockopt(SCARG(uap, optname));
1026 		break;
1027 #ifdef INET6
1028 	case IPPROTO_IPV6:
1029 		name = linux_to_bsd_ipv6_sockopt(SCARG(uap, optname));
1030 		break;
1031 #endif
1032 	case IPPROTO_TCP:
1033 		name = linux_to_bsd_tcp_sockopt(SCARG(uap, optname));
1034 		break;
1035 	case IPPROTO_UDP:
1036 		name = linux_to_bsd_udp_sockopt(SCARG(uap, optname));
1037 		break;
1038 	default:
1039 		return EINVAL;
1040 	}
1041 
1042 	if (name == -1)
1043 		return EINVAL;
1044 	SCARG(&bsa, name) = name;
1045 
1046 	return sys_setsockopt(l, &bsa, retval);
1047 }
1048 
1049 /*
1050  * getsockopt(2) is very much the same as setsockopt(2) (see above)
1051  */
1052 int
1053 linux_sys_getsockopt(struct lwp *l, const struct linux_sys_getsockopt_args *uap, register_t *retval)
1054 {
1055 	/* {
1056 		syscallarg(int) s;
1057 		syscallarg(int) level;
1058 		syscallarg(int) optname;
1059 		syscallarg(void *) optval;
1060 		syscallarg(int *) optlen;
1061 	} */
1062 	struct sys_getsockopt_args bga;
1063 	int name;
1064 
1065 	SCARG(&bga, s) = SCARG(uap, s);
1066 	SCARG(&bga, level) = linux_to_bsd_sopt_level(SCARG(uap, level));
1067 	SCARG(&bga, val) = SCARG(uap, optval);
1068 	SCARG(&bga, avalsize) = (socklen_t *)SCARG(uap, optlen);
1069 
1070 	switch (SCARG(&bga, level)) {
1071 	case SOL_SOCKET:
1072 		name = linux_to_bsd_so_sockopt(SCARG(uap, optname));
1073 		break;
1074 	case IPPROTO_IP:
1075 		name = linux_to_bsd_ip_sockopt(SCARG(uap, optname));
1076 		break;
1077 #ifdef INET6
1078 	case IPPROTO_IPV6:
1079 		name = linux_to_bsd_ipv6_sockopt(SCARG(uap, optname));
1080 		break;
1081 #endif
1082 	case IPPROTO_TCP:
1083 		name = linux_to_bsd_tcp_sockopt(SCARG(uap, optname));
1084 		break;
1085 	case IPPROTO_UDP:
1086 		name = linux_to_bsd_udp_sockopt(SCARG(uap, optname));
1087 		break;
1088 	default:
1089 		return EINVAL;
1090 	}
1091 
1092 	if (name == -1)
1093 		return EINVAL;
1094 	SCARG(&bga, name) = name;
1095 
1096 	return sys_getsockopt(l, &bga, retval);
1097 }
1098 
1099 int
1100 linux_getifname(struct lwp *l, register_t *retval, void *data)
1101 {
1102 	struct ifnet *ifp;
1103 	struct linux_ifreq ifr;
1104 	int error;
1105 	int s;
1106 
1107 	error = copyin(data, &ifr, sizeof(ifr));
1108 	if (error)
1109 		return error;
1110 
1111 	s = pserialize_read_enter();
1112 	ifp = if_byindex(ifr.ifr_ifru.ifru_ifindex);
1113 	if (ifp == NULL) {
1114 		pserialize_read_exit(s);
1115 		return ENODEV;
1116 	}
1117 
1118 	strncpy(ifr.ifr_name, ifp->if_xname, sizeof(ifr.ifr_name));
1119 	pserialize_read_exit(s);
1120 
1121 	return copyout(&ifr, data, sizeof(ifr));
1122 }
1123 
1124 int
1125 linux_getifconf(struct lwp *l, register_t *retval, void *data)
1126 {
1127 	struct linux_ifreq ifr, *ifrp = NULL;
1128 	struct linux_ifconf ifc;
1129 	struct ifnet *ifp;
1130 	struct sockaddr *sa;
1131 	struct osockaddr *osa;
1132 	int space = 0, error;
1133 	const int sz = (int)sizeof(ifr);
1134 	bool docopy;
1135 	int s;
1136 	int bound;
1137 	struct psref psref;
1138 
1139 	error = copyin(data, &ifc, sizeof(ifc));
1140 	if (error)
1141 		return error;
1142 
1143 	docopy = ifc.ifc_req != NULL;
1144 	if (docopy) {
1145 		if (ifc.ifc_len < 0)
1146 			return EINVAL;
1147 
1148 		space = ifc.ifc_len;
1149 		ifrp = ifc.ifc_req;
1150 	}
1151 	memset(&ifr, 0, sizeof(ifr));
1152 
1153 	bound = curlwp_bind();
1154 	s = pserialize_read_enter();
1155 	IFNET_READER_FOREACH(ifp) {
1156 		struct ifaddr *ifa;
1157 		if_acquire(ifp, &psref);
1158 		pserialize_read_exit(s);
1159 
1160 		(void)strncpy(ifr.ifr_name, ifp->if_xname,
1161 		    sizeof(ifr.ifr_name));
1162 		if (ifr.ifr_name[sizeof(ifr.ifr_name) - 1] != '\0') {
1163 			error = ENAMETOOLONG;
1164 			goto release_exit;
1165 		}
1166 
1167 		s = pserialize_read_enter();
1168 		IFADDR_READER_FOREACH(ifa, ifp) {
1169 			struct psref psref_ifa;
1170 			ifa_acquire(ifa, &psref_ifa);
1171 			pserialize_read_exit(s);
1172 
1173 			sa = ifa->ifa_addr;
1174 			if (sa->sa_family != AF_INET ||
1175 			    sa->sa_len > sizeof(*osa))
1176 				goto next;
1177 			memcpy(&ifr.ifr_addr, sa, sa->sa_len);
1178 			osa = (struct osockaddr *)&ifr.ifr_addr;
1179 			osa->sa_family = sa->sa_family;
1180 			if (space >= sz) {
1181 				error = copyout(&ifr, ifrp, sz);
1182 				if (error != 0) {
1183 					ifa_release(ifa, &psref_ifa);
1184 					goto release_exit;
1185 				}
1186 				ifrp++;
1187 			}
1188 			space -= sz;
1189 		next:
1190 			s = pserialize_read_enter();
1191 			ifa_release(ifa, &psref_ifa);
1192 		}
1193 
1194 		KASSERT(pserialize_in_read_section());
1195 		if_release(ifp, &psref);
1196 	}
1197 	pserialize_read_exit(s);
1198 	curlwp_bindx(bound);
1199 
1200 	if (docopy)
1201 		ifc.ifc_len -= space;
1202 	else
1203 		ifc.ifc_len = -space;
1204 
1205 	return copyout(&ifc, data, sizeof(ifc));
1206 
1207 release_exit:
1208 	if_release(ifp, &psref);
1209 	curlwp_bindx(bound);
1210 	return error;
1211 }
1212 
1213 int
1214 linux_getifhwaddr(struct lwp *l, register_t *retval, u_int fd,
1215     void *data)
1216 {
1217 	/* Not the full structure, just enough to map what we do here */
1218 	struct linux_ifreq lreq;
1219 	file_t *fp;
1220 	struct ifaddr *ifa;
1221 	struct ifnet *ifp;
1222 	struct sockaddr_dl *sadl;
1223 	int error, found;
1224 	int index, ifnum;
1225 	int s;
1226 
1227 	/*
1228 	 * We can't emulate this ioctl by calling sys_ioctl() to run
1229 	 * SIOCGIFCONF, because the user buffer is not of the right
1230 	 * type to take those results.  We can't use kernel buffers to
1231 	 * receive the results, as the implementation of sys_ioctl()
1232 	 * and ifconf() [which implements SIOCGIFCONF] use
1233 	 * copyin()/copyout() which will fail on kernel addresses.
1234 	 *
1235 	 * So, we must duplicate code from sys_ioctl() and ifconf().  Ugh.
1236 	 */
1237 
1238 	if ((fp = fd_getfile(fd)) == NULL)
1239 		return (EBADF);
1240 
1241 	KERNEL_LOCK(1, NULL);
1242 
1243 	if ((fp->f_flag & (FREAD | FWRITE)) == 0) {
1244 		error = EBADF;
1245 		goto out;
1246 	}
1247 
1248 	error = copyin(data, &lreq, sizeof(lreq));
1249 	if (error)
1250 		goto out;
1251 	lreq.ifr_name[LINUX_IFNAMSIZ-1] = '\0';		/* just in case */
1252 
1253 	/*
1254 	 * Try real interface name first, then fake "ethX"
1255 	 */
1256 	found = 0;
1257 	s = pserialize_read_enter();
1258 	IFNET_READER_FOREACH(ifp) {
1259 		if (found)
1260 			break;
1261 		if (strcmp(lreq.ifr_name, ifp->if_xname))
1262 			/* not this interface */
1263 			continue;
1264 
1265 		found=1;
1266 		if (IFADDR_READER_EMPTY(ifp)) {
1267 			pserialize_read_exit(s);
1268 			error = ENODEV;
1269 			goto out;
1270 		}
1271 		IFADDR_READER_FOREACH(ifa, ifp) {
1272 			sadl = satosdl(ifa->ifa_addr);
1273 			/* only return ethernet addresses */
1274 			/* XXX what about FDDI, etc. ? */
1275 			if (sadl->sdl_family != AF_LINK ||
1276 			    sadl->sdl_type != IFT_ETHER)
1277 				continue;
1278 			memcpy(&lreq.ifr_hwaddr.sa_data, CLLADDR(sadl),
1279 			       MIN(sadl->sdl_alen,
1280 				   sizeof(lreq.ifr_hwaddr.sa_data)));
1281 			lreq.ifr_hwaddr.sa_family =
1282 				sadl->sdl_family;
1283 			pserialize_read_exit(s);
1284 
1285 			error = copyout(&lreq, data, sizeof(lreq));
1286 			goto out;
1287 		}
1288 	}
1289 	pserialize_read_exit(s);
1290 
1291 	if (strncmp(lreq.ifr_name, "eth", 3) != 0) {
1292 		/* unknown interface, not even an "eth*" name */
1293 		error = ENODEV;
1294 		goto out;
1295 	}
1296 
1297 	for (ifnum = 0, index = 3;
1298 	     index < LINUX_IFNAMSIZ && lreq.ifr_name[index] != '\0';
1299 	     index++) {
1300 		ifnum *= 10;
1301 		ifnum += lreq.ifr_name[index] - '0';
1302 	}
1303 
1304 	error = EINVAL;			/* in case we don't find one */
1305 	s = pserialize_read_enter();
1306 	IFNET_READER_FOREACH(ifp) {
1307 		memcpy(lreq.ifr_name, ifp->if_xname,
1308 		       MIN(LINUX_IFNAMSIZ, IFNAMSIZ));
1309 		IFADDR_READER_FOREACH(ifa, ifp) {
1310 			sadl = satosdl(ifa->ifa_addr);
1311 			/* only return ethernet addresses */
1312 			/* XXX what about FDDI, etc. ? */
1313 			if (sadl->sdl_family != AF_LINK ||
1314 			    sadl->sdl_type != IFT_ETHER)
1315 				continue;
1316 			if (ifnum--)
1317 				/* not the reqested iface */
1318 				continue;
1319 			memcpy(&lreq.ifr_hwaddr.sa_data,
1320 			       CLLADDR(sadl),
1321 			       MIN(sadl->sdl_alen,
1322 				   sizeof(lreq.ifr_hwaddr.sa_data)));
1323 			lreq.ifr_hwaddr.sa_family =
1324 				sadl->sdl_family;
1325 			pserialize_read_exit(s);
1326 
1327 			error = copyout(&lreq, data, sizeof(lreq));
1328 			goto out;
1329 		}
1330 	}
1331 	pserialize_read_exit(s);
1332 
1333 out:
1334 	KERNEL_UNLOCK_ONE(NULL);
1335 	fd_putfile(fd);
1336 	return error;
1337 }
1338 
1339 int
1340 linux_ioctl_socket(struct lwp *l, const struct linux_sys_ioctl_args *uap, register_t *retval)
1341 {
1342 	/* {
1343 		syscallarg(int) fd;
1344 		syscallarg(u_long) com;
1345 		syscallarg(void *) data;
1346 	} */
1347 	u_long com;
1348 	int error = 0, isdev = 0, dosys = 1;
1349 	struct sys_ioctl_args ia;
1350 	file_t *fp;
1351 	struct vnode *vp;
1352 	int (*ioctlf)(file_t *, u_long, void *);
1353 	struct ioctl_pt pt;
1354 
1355 	if ((fp = fd_getfile(SCARG(uap, fd))) == NULL)
1356 		return (EBADF);
1357 
1358 	if (fp->f_type == DTYPE_VNODE) {
1359 		vp = (struct vnode *)fp->f_data;
1360 		isdev = vp->v_type == VCHR;
1361 	}
1362 
1363 	/*
1364 	 * Don't try to interpret socket ioctl calls that are done
1365 	 * on a device filedescriptor, just pass them through, to
1366 	 * emulate Linux behaviour. Use PTIOCLINUX so that the
1367 	 * device will only handle these if it's prepared to do
1368 	 * so, to avoid unexpected things from happening.
1369 	 */
1370 	if (isdev) {
1371 		dosys = 0;
1372 		ioctlf = fp->f_ops->fo_ioctl;
1373 		pt.com = SCARG(uap, com);
1374 		pt.data = SCARG(uap, data);
1375 		error = ioctlf(fp, PTIOCLINUX, &pt);
1376 		/*
1377 		 * XXX hack: if the function returns EJUSTRETURN,
1378 		 * it has stuffed a sysctl return value in pt.data.
1379 		 */
1380 		if (error == EJUSTRETURN) {
1381 			retval[0] = (register_t)pt.data;
1382 			error = 0;
1383 		}
1384 		goto out;
1385 	}
1386 
1387 	com = SCARG(uap, com);
1388 	retval[0] = 0;
1389 
1390 	switch (com) {
1391 	case LINUX_SIOCGIFNAME:
1392 		error = linux_getifname(l, retval, SCARG(uap, data));
1393 		dosys = 0;
1394 		break;
1395 	case LINUX_SIOCGIFCONF:
1396 		error = linux_getifconf(l, retval, SCARG(uap, data));
1397 		dosys = 0;
1398 		break;
1399 	case LINUX_SIOCGIFFLAGS:
1400 		SCARG(&ia, com) = OSIOCGIFFLAGS;
1401 		break;
1402 	case LINUX_SIOCSIFFLAGS:
1403 		SCARG(&ia, com) = OSIOCSIFFLAGS;
1404 		break;
1405 	case LINUX_SIOCGIFADDR:
1406 		SCARG(&ia, com) = OOSIOCGIFADDR;
1407 		break;
1408 	case LINUX_SIOCGIFDSTADDR:
1409 		SCARG(&ia, com) = OOSIOCGIFDSTADDR;
1410 		break;
1411 	case LINUX_SIOCGIFBRDADDR:
1412 		SCARG(&ia, com) = OOSIOCGIFBRDADDR;
1413 		break;
1414 	case LINUX_SIOCGIFNETMASK:
1415 		SCARG(&ia, com) = OOSIOCGIFNETMASK;
1416 		break;
1417 	case LINUX_SIOCGIFMTU:
1418 		SCARG(&ia, com) = OSIOCGIFMTU;
1419 		break;
1420 	case LINUX_SIOCADDMULTI:
1421 		SCARG(&ia, com) = OSIOCADDMULTI;
1422 		break;
1423 	case LINUX_SIOCDELMULTI:
1424 		SCARG(&ia, com) = OSIOCDELMULTI;
1425 		break;
1426 	case LINUX_SIOCGIFHWADDR:
1427 		error = linux_getifhwaddr(l, retval, SCARG(uap, fd),
1428 		    SCARG(uap, data));
1429 		dosys = 0;
1430 		break;
1431 	default:
1432 		error = EINVAL;
1433 	}
1434 
1435  out:
1436  	fd_putfile(SCARG(uap, fd));
1437 
1438 	if (error ==0 && dosys) {
1439 		SCARG(&ia, fd) = SCARG(uap, fd);
1440 		SCARG(&ia, data) = SCARG(uap, data);
1441 		error = sys_ioctl(curlwp, &ia, retval);
1442 	}
1443 
1444 	return error;
1445 }
1446 
1447 int
1448 linux_sys_connect(struct lwp *l, const struct linux_sys_connect_args *uap, register_t *retval)
1449 {
1450 	/* {
1451 		syscallarg(int) s;
1452 		syscallarg(const struct sockaddr *) name;
1453 		syscallarg(int) namelen;
1454 	} */
1455 	int		error;
1456 	struct sockaddr_big sb;
1457 
1458 	error = linux_get_sa(l, SCARG(uap, s), &sb, SCARG(uap, name),
1459 	    SCARG(uap, namelen));
1460 	if (error)
1461 		return (error);
1462 
1463 	error = do_sys_connect(l, SCARG(uap, s), (struct sockaddr *)&sb);
1464 
1465 	if (error == EISCONN) {
1466 		struct socket *so;
1467 		int state, prflags;
1468 
1469 		/* fd_getsock() will use the descriptor for us */
1470 	    	if (fd_getsock(SCARG(uap, s), &so) != 0)
1471 		    	return EISCONN;
1472 
1473 		solock(so);
1474 		state = so->so_state;
1475 		prflags = so->so_proto->pr_flags;
1476 		sounlock(so);
1477 		fd_putfile(SCARG(uap, s));
1478 		/*
1479 		 * We should only let this call succeed once per
1480 		 * non-blocking connect; however we don't have
1481 		 * a convenient place to keep that state..
1482 		 */
1483 		if ((state & (SS_ISCONNECTED|SS_NBIO)) ==
1484 		    (SS_ISCONNECTED|SS_NBIO) &&
1485 		    (prflags & PR_CONNREQUIRED))
1486 			return 0;
1487 	}
1488 
1489 	return (error);
1490 }
1491 
1492 int
1493 linux_sys_bind(struct lwp *l, const struct linux_sys_bind_args *uap, register_t *retval)
1494 {
1495 	/* {
1496 		syscallarg(int) s;
1497 		syscallarg(const struct osockaddr *) name;
1498 		syscallarg(int) namelen;
1499 	} */
1500 	int		error;
1501 	struct sockaddr_big sb;
1502 
1503 	error = linux_get_sa(l, SCARG(uap, s), &sb, SCARG(uap, name),
1504 	    SCARG(uap, namelen));
1505 	if (error)
1506 		return (error);
1507 
1508 	return do_sys_bind(l, SCARG(uap, s), (struct sockaddr *)&sb);
1509 }
1510 
1511 int
1512 linux_sys_getsockname(struct lwp *l, const struct linux_sys_getsockname_args *uap, register_t *retval)
1513 {
1514 	/* {
1515 		syscallarg(int) fdes;
1516 		syscallarg(void *) asa;
1517 		syscallarg(int *) alen;
1518 	} */
1519 	int error;
1520 
1521 	if ((error = sys_getsockname(l, (const void *)uap, retval)) != 0)
1522 		return (error);
1523 
1524 	if ((error = linux_sa_put((struct osockaddr *)SCARG(uap, asa))))
1525 		return (error);
1526 
1527 	return (0);
1528 }
1529 
1530 int
1531 linux_sys_getpeername(struct lwp *l, const struct linux_sys_getpeername_args *uap, register_t *retval)
1532 {
1533 	/* {
1534 		syscallarg(int) fdes;
1535 		syscallarg(void *) asa;
1536 		syscallarg(int *) alen;
1537 	} */
1538 	int error;
1539 
1540 	if ((error = sys_getpeername(l, (const void *)uap, retval)) != 0)
1541 		return (error);
1542 
1543 	if ((error = linux_sa_put((struct osockaddr *)SCARG(uap, asa))))
1544 		return (error);
1545 
1546 	return (0);
1547 }
1548 
1549 /*
1550  * Copy the osockaddr structure pointed to by name to sb, adjust
1551  * family and convert to sockaddr.
1552  */
1553 static int
1554 linux_get_sa(struct lwp *l, int s, struct sockaddr_big *sb,
1555     const struct osockaddr *name, socklen_t namelen)
1556 {
1557 	int error, bdom;
1558 
1559 	if (namelen > UCHAR_MAX ||
1560 	    namelen <= offsetof(struct sockaddr_big, sb_data))
1561 		return EINVAL;
1562 
1563 	error = copyin(name, sb, namelen);
1564 	if (error)
1565 		return error;
1566 
1567 	bdom = linux_to_bsd_domain(sb->sb_family);
1568 	if (bdom == -1)
1569 		return EINVAL;
1570 
1571 	/*
1572 	 * If the family is unspecified, use address family of the socket.
1573 	 * This avoid triggering strict family checks in netinet/in_pcb.c et.al.
1574 	 */
1575 	if (bdom == AF_UNSPEC) {
1576 		struct socket *so;
1577 
1578 		/* fd_getsock() will use the descriptor for us */
1579 		if ((error = fd_getsock(s, &so)) != 0)
1580 			return error;
1581 
1582 		bdom = so->so_proto->pr_domain->dom_family;
1583 		fd_putfile(s);
1584 	}
1585 
1586 	/*
1587 	 * Older Linux IPv6 code uses obsolete RFC2133 struct sockaddr_in6,
1588 	 * which lacks the scope id compared with RFC2553 one. If we detect
1589 	 * the situation, reject the address and write a message to system log.
1590 	 *
1591 	 * Still accept addresses for which the scope id is not used.
1592 	 */
1593 	if (bdom == AF_INET6 &&
1594 	    namelen == sizeof(struct sockaddr_in6) - sizeof(uint32_t)) {
1595 		struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sb;
1596 		if (!IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr) &&
1597 		    (IN6_IS_ADDR_LINKLOCAL(&sin6->sin6_addr) ||
1598 		     IN6_IS_ADDR_SITELOCAL(&sin6->sin6_addr) ||
1599 		     IN6_IS_ADDR_V4COMPAT(&sin6->sin6_addr) ||
1600 		     IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr) ||
1601 		     IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr))) {
1602 			struct proc *p = l->l_proc;
1603 			int uid = l->l_cred ? kauth_cred_geteuid(l->l_cred) : -1;
1604 
1605 			log(LOG_DEBUG,
1606 			    "pid %d (%s), uid %d: obsolete pre-RFC2553 "
1607 			    "sockaddr_in6 rejected",
1608 			    p->p_pid, p->p_comm, uid);
1609 			return EINVAL;
1610 		}
1611 		namelen = sizeof(struct sockaddr_in6);
1612 		sin6->sin6_scope_id = 0;
1613 	}
1614 
1615 	sb->sb_family = bdom;
1616 	sb->sb_len = namelen;
1617 	ktrkuser("mbsoname", sb, namelen);
1618 	return 0;
1619 }
1620 
1621 static int
1622 linux_sa_put(struct osockaddr *osa)
1623 {
1624 	struct sockaddr sa;
1625 	struct osockaddr *kosa;
1626 	int error, bdom, len;
1627 
1628 	/*
1629 	 * Only read/write the sockaddr family and length part, the rest is
1630 	 * not changed.
1631 	 */
1632 	len = sizeof(sa.sa_len) + sizeof(sa.sa_family);
1633 
1634 	error = copyin(osa, &sa, len);
1635 	if (error)
1636 		return (error);
1637 
1638 	bdom = bsd_to_linux_domain(sa.sa_family);
1639 	if (bdom == -1)
1640 		return (EINVAL);
1641 
1642 	/* Note: we convert from sockaddr to osockaddr here, too */
1643 	kosa = (struct osockaddr *) &sa;
1644 	kosa->sa_family = bdom;
1645 	error = copyout(kosa, osa, len);
1646 	if (error)
1647 		return (error);
1648 
1649 	return (0);
1650 }
1651 
1652 #ifndef __amd64__
1653 int
1654 linux_sys_recv(struct lwp *l, const struct linux_sys_recv_args *uap, register_t *retval)
1655 {
1656 	/* {
1657 		syscallarg(int) s;
1658 		syscallarg(void *) buf;
1659 		syscallarg(int) len;
1660 		syscallarg(int) flags;
1661 	} */
1662 	struct sys_recvfrom_args bra;
1663 
1664 
1665 	SCARG(&bra, s) = SCARG(uap, s);
1666 	SCARG(&bra, buf) = SCARG(uap, buf);
1667 	SCARG(&bra, len) = (size_t) SCARG(uap, len);
1668 	SCARG(&bra, flags) = SCARG(uap, flags);
1669 	SCARG(&bra, from) = NULL;
1670 	SCARG(&bra, fromlenaddr) = NULL;
1671 
1672 	return (sys_recvfrom(l, &bra, retval));
1673 }
1674 
1675 int
1676 linux_sys_send(struct lwp *l, const struct linux_sys_send_args *uap, register_t *retval)
1677 {
1678 	/* {
1679 		syscallarg(int) s;
1680 		syscallarg(void *) buf;
1681 		syscallarg(int) len;
1682 		syscallarg(int) flags;
1683 	} */
1684 	struct sys_sendto_args bsa;
1685 
1686 	SCARG(&bsa, s)		= SCARG(uap, s);
1687 	SCARG(&bsa, buf)	= SCARG(uap, buf);
1688 	SCARG(&bsa, len)	= SCARG(uap, len);
1689 	SCARG(&bsa, flags)	= SCARG(uap, flags);
1690 	SCARG(&bsa, to)		= NULL;
1691 	SCARG(&bsa, tolen)	= 0;
1692 
1693 	return (sys_sendto(l, &bsa, retval));
1694 }
1695 #endif
1696 
1697 int
1698 linux_sys_accept(struct lwp *l, const struct linux_sys_accept_args *uap, register_t *retval)
1699 {
1700 	/* {
1701 		syscallarg(int) s;
1702 		syscallarg(struct osockaddr *) name;
1703 		syscallarg(int *) anamelen;
1704 	} */
1705 	int error;
1706 	struct sys_accept_args baa;
1707 
1708 	SCARG(&baa, s)		= SCARG(uap, s);
1709 	SCARG(&baa, name)	= (struct sockaddr *) SCARG(uap, name);
1710 	SCARG(&baa, anamelen)	= (unsigned int *) SCARG(uap, anamelen);
1711 
1712 	if ((error = sys_accept(l, &baa, retval)))
1713 		return (error);
1714 
1715 	if (SCARG(uap, name) && (error = linux_sa_put(SCARG(uap, name))))
1716 		return (error);
1717 
1718 	return (0);
1719 }
1720 
1721 int
1722 linux_sys_accept4(struct lwp *l, const struct linux_sys_accept4_args *uap, register_t *retval)
1723 {
1724 	/* {
1725 		syscallarg(int) s;
1726 		syscallarg(struct osockaddr *) name;
1727 		syscallarg(int *) anamelen;
1728 		syscallarg(int) flags;
1729 	} */
1730 	int error, flags;
1731 	struct sockaddr_big name;
1732 
1733 	if ((flags = linux_to_bsd_type(SCARG(uap, flags))) == -1)
1734 		return EINVAL;
1735 
1736 	name.sb_len = UCHAR_MAX;
1737 	error = do_sys_accept(l, SCARG(uap, s), (struct sockaddr *)&name,
1738 	    retval, NULL, flags, 0);
1739 	if (error != 0)
1740 		return error;
1741 
1742 	error = copyout_sockname_sb((struct sockaddr *)SCARG(uap, name),
1743 	    SCARG(uap, anamelen), MSG_LENUSRSPACE, &name);
1744 	if (error != 0) {
1745 		int fd = (int)*retval;
1746 		if (fd_getfile(fd) != NULL)
1747 			(void)fd_close(fd);
1748 		return error;
1749 	}
1750 	if (SCARG(uap, name) && (error = linux_sa_put(SCARG(uap, name))))
1751 		return error;
1752 
1753 	return 0;
1754 }
1755 
1756 int
1757 linux_sys_sendmmsg(struct lwp *l, const struct linux_sys_sendmmsg_args *uap,
1758     register_t *retval)
1759 {
1760 	/* {
1761 		syscallarg(int) s;
1762 		syscallarg(struct linux_mmsghdr *) msgvec;
1763 		syscallarg(unsigned int) vlen;
1764 		syscallarg(unsigned int) flags;
1765 	} */
1766 	struct linux_mmsghdr lmsg;
1767 	struct mmsghdr bmsg;
1768 	struct socket *so;
1769 	file_t *fp;
1770 	struct msghdr *msg = &bmsg.msg_hdr;
1771 	int error, s;
1772 	unsigned int vlen, flags, dg;
1773 
1774 	if ((flags = linux_to_bsd_msg_flags(SCARG(uap, flags))) == -1)
1775 		return EINVAL;
1776 
1777 	flags = (flags & MSG_USERFLAGS) | MSG_IOVUSRSPACE;
1778 
1779 	s = SCARG(uap, s);
1780 	if ((error = fd_getsock1(s, &so, &fp)) != 0)
1781 		return error;
1782 
1783 	vlen = SCARG(uap, vlen);
1784 	if (vlen > 1024)
1785 		vlen = 1024;
1786 
1787 	for (dg = 0; dg < vlen;) {
1788 		error = copyin(SCARG(uap, msgvec) + dg, &lmsg, sizeof(lmsg));
1789 		if (error)
1790 			break;
1791 		linux_to_bsd_msghdr(&lmsg.msg_hdr, &bmsg.msg_hdr);
1792 
1793 		msg->msg_flags = flags;
1794 
1795 		error = do_sys_sendmsg_so(l, s, so, fp, msg, flags, retval);
1796 		if (error)
1797 			break;
1798 
1799 		ktrkuser("msghdr", msg, sizeof *msg);
1800 		lmsg.msg_len = *retval;
1801 		error = copyout(&lmsg, SCARG(uap, msgvec) + dg, sizeof(lmsg));
1802 		if (error)
1803 			break;
1804 		dg++;
1805 
1806 	}
1807 
1808 	*retval = dg;
1809 
1810 	fd_putfile(s);
1811 
1812 	/*
1813 	 * If we succeeded at least once, return 0.
1814 	 */
1815 	if (dg)
1816 		return 0;
1817 	return error;
1818 }
1819 
1820 int
1821 linux_sys_recvmmsg(struct lwp *l, const struct linux_sys_recvmmsg_args *uap,
1822     register_t *retval)
1823 {
1824 	/* {
1825 		syscallarg(int) s;
1826 		syscallarg(struct linux_mmsghdr *) msgvec;
1827 		syscallarg(unsigned int) vlen;
1828 		syscallarg(unsigned int) flags;
1829 		syscallarg(struct linux_timespec *) timeout;
1830 	} */
1831 	struct linux_mmsghdr lmsg;
1832 	struct mmsghdr bmsg;
1833 	struct socket *so;
1834 	struct msghdr *msg = &bmsg.msg_hdr;
1835 	int error, s;
1836 	struct mbuf *from, *control;
1837 	struct timespec ts = {0}, now;
1838 	struct linux_timespec lts;
1839 	unsigned int vlen, flags, dg;
1840 
1841 	if (SCARG(uap, timeout)) {
1842 		error = copyin(SCARG(uap, timeout), &lts, sizeof(lts));
1843 			return error;
1844 		ts.tv_sec = lts.tv_sec;
1845 		ts.tv_nsec = lts.tv_nsec;
1846 		getnanotime(&now);
1847 		timespecadd(&now, &ts, &ts);
1848 	}
1849 
1850 	s = SCARG(uap, s);
1851 	if ((error = fd_getsock(s, &so)) != 0)
1852 		return error;
1853 
1854 	/*
1855 	 * If so->so_rerror holds a deferred error return it now.
1856 	 */
1857 	if (so->so_rerror) {
1858 		error = so->so_rerror;
1859 		so->so_rerror = 0;
1860 		fd_putfile(s);
1861 		return error;
1862 	}
1863 
1864 	vlen = SCARG(uap, vlen);
1865 	if (vlen > 1024)
1866 		vlen = 1024;
1867 
1868 	from = NULL;
1869 	flags = (SCARG(uap, flags) & MSG_USERFLAGS) | MSG_IOVUSRSPACE;
1870 
1871 	for (dg = 0; dg < vlen;) {
1872 		error = copyin(SCARG(uap, msgvec) + dg, &lmsg, sizeof(lmsg));
1873 		if (error)
1874 			break;
1875 		linux_to_bsd_msghdr(&lmsg.msg_hdr, &bmsg.msg_hdr);
1876 		msg->msg_flags = flags & ~MSG_WAITFORONE;
1877 
1878 		if (from != NULL) {
1879 			m_free(from);
1880 			from = NULL;
1881 		}
1882 
1883 		error = do_sys_recvmsg_so(l, s, so, msg, &from,
1884 		    msg->msg_control != NULL ? &control : NULL, retval);
1885 		if (error) {
1886 			if (error == EAGAIN && dg > 0)
1887 				error = 0;
1888 			break;
1889 		}
1890 
1891 		if (msg->msg_control != NULL)
1892 			error = linux_copyout_msg_control(l, msg, control);
1893 		if (error)
1894 			break;
1895 
1896 		if (from != NULL) {
1897 			mtod(from, struct osockaddr *)->sa_family =
1898 			    bsd_to_linux_domain(mtod(from,
1899 			    struct sockaddr *)->sa_family);
1900 			error = copyout_sockname(msg->msg_name,
1901 			    &msg->msg_namelen, 0, from);
1902 			if (error)
1903 				break;
1904 		}
1905 
1906 
1907 		lmsg.msg_len = *retval;
1908 		ktrkuser("msghdr", msg, sizeof(*msg));
1909 		bsd_to_linux_msghdr(msg, &lmsg.msg_hdr);
1910 		error = copyout(&lmsg, SCARG(uap, msgvec) + dg, sizeof(lmsg));
1911 		if (error)
1912 			break;
1913 
1914 		dg++;
1915 		if (msg->msg_flags & MSG_OOB)
1916 			break;
1917 
1918 		if (SCARG(uap, timeout)) {
1919 			getnanotime(&now);
1920 			timespecsub(&now, &ts, &now);
1921 			if (now.tv_sec > 0)
1922 				break;
1923 		}
1924 
1925 		if (flags & MSG_WAITFORONE)
1926 			flags |= MSG_DONTWAIT;
1927 
1928 	}
1929 
1930 	if (from != NULL)
1931 		m_free(from);
1932 
1933 	*retval = dg;
1934 
1935 	/*
1936 	 * If we succeeded at least once, return 0, hopefully so->so_rerror
1937 	 * will catch it next time.
1938 	 */
1939 	if (error && dg > 0) {
1940 		so->so_rerror = error;
1941 		error = 0;
1942 	}
1943 
1944 	fd_putfile(s);
1945 
1946 	return error;
1947 }
1948