xref: /netbsd-src/sys/compat/linux/common/linux_socket.c (revision 23c8222edbfb0f0932d88a8351d3a0cf817dfb9e)
1 /*	$NetBSD: linux_socket.c,v 1.53 2004/09/12 15:32:55 jdolecek Exp $	*/
2 
3 /*-
4  * Copyright (c) 1995, 1998 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  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the NetBSD
21  *	Foundation, Inc. and its contributors.
22  * 4. Neither the name of The NetBSD Foundation nor the names of its
23  *    contributors may be used to endorse or promote products derived
24  *    from this software without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
27  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
28  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
29  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
30  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36  * POSSIBILITY OF SUCH DAMAGE.
37  */
38 
39 /*
40  * Functions in multiarch:
41  *	linux_sys_socketcall		: linux_socketcall.c
42  */
43 
44 #include <sys/cdefs.h>
45 __KERNEL_RCSID(0, "$NetBSD: linux_socket.c,v 1.53 2004/09/12 15:32:55 jdolecek Exp $");
46 
47 #if defined(_KERNEL_OPT)
48 #include "opt_inet.h"
49 #endif
50 
51 #include <sys/param.h>
52 #include <sys/kernel.h>
53 #include <sys/systm.h>
54 #include <sys/buf.h>
55 #include <sys/malloc.h>
56 #include <sys/ioctl.h>
57 #include <sys/tty.h>
58 #include <sys/file.h>
59 #include <sys/filedesc.h>
60 #include <sys/select.h>
61 #include <sys/socket.h>
62 #include <sys/socketvar.h>
63 #include <sys/domain.h>
64 #include <net/if.h>
65 #include <net/if_dl.h>
66 #include <net/if_types.h>
67 #include <netinet/in.h>
68 #include <netinet/tcp.h>
69 #include <sys/mount.h>
70 #include <sys/proc.h>
71 #include <sys/vnode.h>
72 #include <sys/device.h>
73 #include <sys/protosw.h>
74 #include <sys/mbuf.h>
75 #include <sys/syslog.h>
76 #include <sys/exec.h>
77 
78 #include <sys/sa.h>
79 #include <sys/syscallargs.h>
80 
81 #ifdef INET6
82 #include <netinet/ip6.h>
83 #include <netinet6/ip6_var.h>
84 #endif
85 
86 #include <compat/linux/common/linux_types.h>
87 #include <compat/linux/common/linux_util.h>
88 #include <compat/linux/common/linux_signal.h>
89 #include <compat/linux/common/linux_ioctl.h>
90 #include <compat/linux/common/linux_socket.h>
91 #include <compat/linux/common/linux_socketcall.h>
92 #include <compat/linux/common/linux_sockio.h>
93 
94 #include <compat/linux/linux_syscallargs.h>
95 
96 #ifdef DEBUG_LINUX
97 #define DPRINTF(a) uprintf a
98 #else
99 #define DPRINTF(a)
100 #endif
101 
102 /*
103  * The calls in this file are entered either via the linux_socketcall()
104  * interface or, on the Alpha, as individual syscalls.  The
105  * linux_socketcall function does any massaging of arguments so that all
106  * the calls in here need not think that they are anything other
107  * than a normal syscall.
108  */
109 
110 static int linux_to_bsd_domain __P((int));
111 static int bsd_to_linux_domain __P((int));
112 int linux_to_bsd_sopt_level __P((int));
113 int linux_to_bsd_so_sockopt __P((int));
114 int linux_to_bsd_ip_sockopt __P((int));
115 int linux_to_bsd_tcp_sockopt __P((int));
116 int linux_to_bsd_udp_sockopt __P((int));
117 int linux_getifhwaddr __P((struct proc *, register_t *, u_int, void *));
118 static int linux_sa_get __P((struct proc *, int, caddr_t *, struct sockaddr **,
119 		const struct osockaddr *, int *));
120 static int linux_sa_put __P((struct osockaddr *osa));
121 static int linux_to_bsd_msg_flags __P((int));
122 static int bsd_to_linux_msg_flags __P((int));
123 
124 static const int linux_to_bsd_domain_[LINUX_AF_MAX] = {
125 	AF_UNSPEC,
126 	AF_UNIX,
127 	AF_INET,
128 	AF_CCITT,	/* LINUX_AF_AX25 */
129 	AF_IPX,
130 	AF_APPLETALK,
131 	-1,		/* LINUX_AF_NETROM */
132 	-1,		/* LINUX_AF_BRIDGE */
133 	-1,		/* LINUX_AF_ATMPVC */
134 	AF_CCITT,	/* LINUX_AF_X25 */
135 	AF_INET6,
136 	-1,		/* LINUX_AF_ROSE */
137 	AF_DECnet,
138 	-1,		/* LINUX_AF_NETBEUI */
139 	-1,		/* LINUX_AF_SECURITY */
140 	pseudo_AF_KEY,
141 	AF_ROUTE,	/* LINUX_AF_NETLINK */
142 	-1,		/* LINUX_AF_PACKET */
143 	-1,		/* LINUX_AF_ASH */
144 	-1,		/* LINUX_AF_ECONET */
145 	-1,		/* LINUX_AF_ATMSVC */
146 	AF_SNA,
147 	/* rest up to LINUX_AF_MAX-1 is not allocated */
148 	-1, -1, -1, -1, -1, -1, -1, -1, -1, -1,
149 };
150 
151 static const int bsd_to_linux_domain_[AF_MAX] = {
152 	LINUX_AF_UNSPEC,
153 	LINUX_AF_UNIX,
154 	LINUX_AF_INET,
155 	-1,		/* AF_IMPLINK */
156 	-1,		/* AF_PUP */
157 	-1,		/* AF_CHAOS */
158 	-1,		/* AF_NS */
159 	-1,		/* AF_ISO */
160 	-1,		/* AF_ECMA */
161 	-1,		/* AF_DATAKIT */
162 	LINUX_AF_AX25,	/* AF_CCITT */
163 	LINUX_AF_SNA,
164 	LINUX_AF_DECnet,
165 	-1,		/* AF_DLI */
166 	-1,		/* AF_LAT */
167 	-1,		/* AF_HYLINK */
168 	LINUX_AF_APPLETALK,
169 	LINUX_AF_NETLINK,
170 	-1,		/* AF_LINK */
171 	-1,		/* AF_XTP */
172 	-1,		/* AF_COIP */
173 	-1,		/* AF_CNT */
174 	-1,		/* pseudo_AF_RTIP */
175 	LINUX_AF_IPX,
176 	LINUX_AF_INET6,
177 	-1,		/* pseudo_AF_PIP */
178 	-1,		/* AF_ISDN */
179 	-1,		/* AF_NATM */
180 	-1,		/* AF_ARP */
181 	LINUX_pseudo_AF_KEY,
182 	-1,		/* pseudo_AF_HDRCMPLT */
183 };
184 
185 static const int bsd_to_linux_msg_flags_[] = {
186 	MSG_OOB,		LINUX_MSG_OOB,
187 	MSG_PEEK,		LINUX_MSG_PEEK,
188 	MSG_DONTROUTE,		LINUX_MSG_DONTROUTE,
189 	MSG_EOR,		LINUX_MSG_EOR,
190 	MSG_TRUNC,		LINUX_MSG_TRUNC,
191 	MSG_CTRUNC,		LINUX_MSG_CTRUNC,
192 	MSG_WAITALL,		LINUX_MSG_WAITALL,
193 	MSG_DONTWAIT,		LINUX_MSG_DONTWAIT,
194 	MSG_BCAST,		0,		/* not supported, clear */
195 	MSG_MCAST,		0,		/* not supported, clear */
196 	-1, /* not supp */	LINUX_MSG_PROBE,
197 	-1, /* not supp */	LINUX_MSG_FIN,
198 	-1, /* not supp */	LINUX_MSG_SYN,
199 	-1, /* not supp */	LINUX_MSG_CONFIRM,
200 	-1, /* not supp */	LINUX_MSG_RST,
201 	-1, /* not supp */	LINUX_MSG_ERRQUEUE,
202 	-1, /* not supp */	LINUX_MSG_NOSIGNAL,
203 	-1, /* not supp */	LINUX_MSG_MORE,
204 };
205 
206 /*
207  * Convert between Linux and BSD socket domain values
208  */
209 static int
210 linux_to_bsd_domain(ldom)
211 	int ldom;
212 {
213 	if (ldom < 0 || ldom >= LINUX_AF_MAX)
214 		return (-1);
215 
216 	return linux_to_bsd_domain_[ldom];
217 }
218 
219 /*
220  * Convert between BSD and Linux socket domain values
221  */
222 static int
223 bsd_to_linux_domain(bdom)
224 	int bdom;
225 {
226 	if (bdom < 0 || bdom >= AF_MAX)
227 		return (-1);
228 
229 	return bsd_to_linux_domain_[bdom];
230 }
231 
232 static int
233 linux_to_bsd_msg_flags(lflag)
234 	int lflag;
235 {
236 	int i, lfl, bfl;
237 	int bflag = 0;
238 
239 	if (lflag == 0)
240 		return (0);
241 
242 	for(i=0; i < sizeof(bsd_to_linux_msg_flags_)/2; i += 2) {
243 		bfl = bsd_to_linux_msg_flags_[i];
244 		lfl = bsd_to_linux_msg_flags_[i+1];
245 
246 		if (lfl == 0)
247 			continue;
248 
249 		if (lflag & lfl) {
250 			if (bfl < 0)
251 				return (-1);
252 
253 			bflag |= bfl;
254 		}
255 	}
256 
257 	return (bflag);
258 }
259 
260 static int
261 bsd_to_linux_msg_flags(bflag)
262 	int bflag;
263 {
264 	int i, lfl, bfl;
265 	int lflag = 0;
266 
267 	if (bflag == 0)
268 		return (0);
269 
270 	for(i=0; i < sizeof(bsd_to_linux_msg_flags_)/2; i += 2) {
271 		bfl = bsd_to_linux_msg_flags_[i];
272 		lfl = bsd_to_linux_msg_flags_[i+1];
273 
274 		if (bfl <= 0)
275 			continue;
276 
277 		if (bflag & bfl) {
278 			if (lfl < 0)
279 				return (-1);
280 
281 			lflag |= lfl;
282 		}
283 	}
284 
285 	return (lflag);
286 }
287 
288 int
289 linux_sys_socket(l, v, retval)
290 	struct lwp *l;
291 	void *v;
292 	register_t *retval;
293 {
294 	struct linux_sys_socket_args /* {
295 		syscallarg(int)	domain;
296 		syscallarg(int)	type;
297 		syscallarg(int) protocol;
298 	} */ *uap = v;
299 	struct sys_socket_args bsa;
300 	int error;
301 
302 	SCARG(&bsa, protocol) = SCARG(uap, protocol);
303 	SCARG(&bsa, type) = SCARG(uap, type);
304 	SCARG(&bsa, domain) = linux_to_bsd_domain(SCARG(uap, domain));
305 	if (SCARG(&bsa, domain) == -1)
306 		return EINVAL;
307 	error = sys_socket(l, &bsa, retval);
308 
309 #ifdef INET6
310 	/*
311 	 * Linux AF_INET6 socket has IPV6_V6ONLY setsockopt set to 0 by
312 	 * default and some apps depend on this. So, set V6ONLY to 0
313 	 * for Linux apps if the sysctl value is set to 1.
314 	 */
315 	if (!error && ip6_v6only && SCARG(&bsa, domain) == PF_INET6) {
316 		struct proc *p = l->l_proc;
317 		struct file *fp;
318 
319 		if (getsock(p->p_fd, *retval, &fp) == 0) {
320 			struct mbuf *m;
321 
322 			m = m_get(M_WAIT, MT_SOOPTS);
323 			m->m_len = sizeof(int);
324 			*mtod(m, int *) = 0;
325 
326 			/* ignore error */
327 			(void) sosetopt((struct socket *)fp->f_data,
328 				IPPROTO_IPV6, IPV6_V6ONLY, m);
329 
330 			FILE_UNUSE(fp, p);
331 		}
332 	}
333 #endif
334 
335 	return (error);
336 }
337 
338 int
339 linux_sys_socketpair(l, v, retval)
340 	struct lwp *l;
341 	void *v;
342 	register_t *retval;
343 {
344 	struct linux_sys_socketpair_args /* {
345 		syscallarg(int) domain;
346 		syscallarg(int) type;
347 		syscallarg(int) protocol;
348 		syscallarg(int *) rsv;
349 	} */ *uap = v;
350 	struct sys_socketpair_args bsa;
351 
352 	SCARG(&bsa, domain) = linux_to_bsd_domain(SCARG(uap, domain));
353 	if (SCARG(&bsa, domain) == -1)
354 		return EINVAL;
355 	SCARG(&bsa, type) = SCARG(uap, type);
356 	SCARG(&bsa, protocol) = SCARG(uap, protocol);
357 	SCARG(&bsa, rsv) = SCARG(uap, rsv);
358 
359 	return sys_socketpair(l, &bsa, retval);
360 }
361 
362 int
363 linux_sys_sendto(l, v, retval)
364 	struct lwp *l;
365 	void *v;
366 	register_t *retval;
367 {
368 	struct linux_sys_sendto_args /* {
369 		syscallarg(int)				s;
370 		syscallarg(void *)			msg;
371 		syscallarg(int)				len;
372 		syscallarg(int)				flags;
373 		syscallarg(struct osockaddr *)		to;
374 		syscallarg(int)				tolen;
375 	} */ *uap = v;
376 	struct proc *p = l->l_proc;
377 	struct sys_sendto_args bsa;
378 	int tolen;
379 
380 	SCARG(&bsa, s) = SCARG(uap, s);
381 	SCARG(&bsa, buf) = SCARG(uap, msg);
382 	SCARG(&bsa, len) = (size_t) SCARG(uap, len);
383 	SCARG(&bsa, flags) = SCARG(uap, flags);
384 	tolen = SCARG(uap, tolen);
385 	if (SCARG(uap, to)) {
386 		struct sockaddr *sa;
387 		int error;
388 		caddr_t sg = stackgap_init(p, 0);
389 
390 		error = linux_sa_get(p, SCARG(uap, s), &sg, &sa,
391 		    SCARG(uap, to), &tolen);
392 		if (error)
393 			return (error);
394 
395 		SCARG(&bsa, to) = sa;
396 	} else
397 		SCARG(&bsa, to) = NULL;
398 	SCARG(&bsa, tolen) = tolen;
399 
400 	return (sys_sendto(l, &bsa, retval));
401 }
402 
403 int
404 linux_sys_sendmsg(l, v, retval)
405 	struct lwp *l;
406 	void *v;
407 	register_t *retval;
408 {
409 	struct linux_sys_sendmsg_args /* {
410 		syscallarg(int) s;
411 		syscallarg(struct msghdr *) msg;
412 		syscallarg(u_int) flags;
413 	} */ *uap = v;
414 	struct proc *p = l->l_proc;
415 	struct msghdr	msg;
416 	int		error;
417 	struct iovec	aiov[UIO_SMALLIOV], *iov;
418 	caddr_t sg = 0;
419 	int		bflags;
420 	u_int8_t	*control=NULL;
421 
422 	error = copyin(SCARG(uap, msg), (caddr_t)&msg, sizeof(msg));
423 	if (error)
424 		return (error);
425 	if ((unsigned int)msg.msg_iovlen > UIO_SMALLIOV) {
426 		if ((unsigned int)msg.msg_iovlen > IOV_MAX)
427 			return (EMSGSIZE);
428 		iov = malloc(sizeof(struct iovec) * msg.msg_iovlen,
429 		    M_IOV, M_WAITOK);
430 	} else
431 		iov = aiov;
432 	if ((unsigned int)msg.msg_iovlen > 0) {
433 		error = copyin((caddr_t)msg.msg_iov, (caddr_t)iov,
434 		    (size_t)(msg.msg_iovlen * sizeof(struct iovec)));
435 		if (error)
436 			goto done;
437 	}
438 	msg.msg_iov = iov;
439 	msg.msg_flags = 0;
440 
441 	/* Convert the sockaddr if necessary */
442 	if (msg.msg_name) {
443 		struct sockaddr *sa;
444 		sg = stackgap_init(p, 0);
445 
446 		error = linux_sa_get(p, SCARG(uap, s), &sg, &sa,
447 		    (struct osockaddr *) msg.msg_name, &msg.msg_namelen);
448 		if (error)
449 			goto done;
450 		msg.msg_name = sa;
451 	}
452 
453 	/*
454 	 * Translate message flags.
455 	 */
456 	bflags = linux_to_bsd_msg_flags(SCARG(uap, flags));
457 	if (bflags < 0) {
458 		/* Some supported flag */
459 		error = EINVAL;
460 		goto done;
461 	}
462 
463 	/*
464 	 * Handle cmsg if there is any.
465 	 */
466 	if (CMSG_FIRSTHDR(&msg)) {
467 		struct cmsghdr cmsg, *cc;
468 		int changed = 0;
469 		ssize_t resid = msg.msg_controllen;
470 		size_t clen, cidx = 0, cspace;
471 
472 		/*
473 		 * Limit the size even more than what sockargs() would do,
474 		 * We need to fit into stackgap space.
475 		 */
476 		if (msg.msg_controllen > (STACKGAPLEN / 2)) {
477 			/* Sorry guys! */
478 			error = EMSGSIZE;
479 			goto done;
480 		}
481 
482 		control = malloc((clen = msg.msg_controllen), M_TEMP, M_WAITOK);
483 		if (!control) {
484 			error = ENOMEM;
485 			goto done;
486 		}
487 
488 		cc = CMSG_FIRSTHDR(&msg);
489 		do {
490 			error = copyin(cc, &cmsg, sizeof(cmsg));
491 			if (error)
492 				goto done;
493 
494 			/*
495 			 * Sanity check the control message length.
496 			 */
497 			if (cmsg.cmsg_len > resid
498 			    || cmsg.cmsg_len < sizeof(struct cmsghdr)) {
499 				error = EINVAL;
500 				goto done;
501 			}
502 
503 			/*
504 			 * Refuse unsupported control messages, and
505 			 * translate fields as appropriate.
506 			 */
507 			switch (cmsg.cmsg_level) {
508 			case LINUX_SOL_SOCKET:
509 				/* It only differs on some archs */
510 				if (LINUX_SOL_SOCKET != SOL_SOCKET) {
511 					cmsg.cmsg_level = SOL_SOCKET;
512 					changed = 1;
513 				}
514 
515 				switch(cmsg.cmsg_type) {
516 				case LINUX_SCM_RIGHTS:
517 					/* Linux SCM_RIGHTS is same as NetBSD */
518 					break;
519 
520 				default:
521 					/* other types not supported */
522 					error = EINVAL;
523 					goto done;
524 				}
525 				break;
526 			default:
527 				/* pray and leave intact */
528 				break;
529 			}
530 
531 			cspace = CMSG_SPACE(cmsg.cmsg_len - sizeof(cmsg));
532 
533 			/* Check the buffer is big enough */
534 			if (__predict_false(cidx + cspace > clen)) {
535 				u_int8_t *nc;
536 
537 				clen = cidx + cspace;
538 				nc = realloc(control, clen, M_TEMP, M_WAITOK);
539 				if (!nc) {
540 					error = ENOMEM;
541 					goto done;
542 				}
543 				control = nc;
544 			}
545 
546 			/* Copy header */
547 			memcpy(&control[cidx], &cmsg, sizeof(cmsg));
548 
549 			/* Zero are between header and data */
550 			memset(&control[cidx+sizeof(cmsg)], 0,
551 				CMSG_ALIGN(sizeof(cmsg)) - sizeof(cmsg));
552 
553 			/* Copyin the data */
554 			error = copyin(LINUX_CMSG_DATA(cc),
555 				CMSG_DATA(control),
556 				cmsg.cmsg_len - sizeof(cmsg));
557 			if (error)
558 				goto done;
559 
560 			/*
561 			 * If there is alignment difference, we changed
562 			 * layout of cmsg.
563 			 */
564 			if (LINUX_CMSG_ALIGNDIFF)
565 				changed = 1;
566 
567 			resid -= cspace;
568 			cidx += cspace;
569 		} while ((cc = LINUX_CMSG_NXTHDR(&msg, cc)) && resid > 0);
570 
571 		/*
572 		 * If any of the passed control message needed
573 		 * a change, put the changed data into stackgap
574 		 * and adjust msg appropriately.
575 		 */
576 		if (changed) {
577 			char *newc;
578 
579 			/*
580 			 * Check again the total len is maximum half of
581 			 * stackgap. The length might change if the
582 			 * alignment is different.
583 			 */
584 			if (clen > STACKGAPLEN/2) {
585 				error = EMSGSIZE;
586 				goto done;
587 			}
588 
589 			/*
590 			 * Allocate space on stack within stackgap, and
591 			 * copy changed data there.
592 			 */
593 			if (!sg)
594 				sg = stackgap_init(p, STACKGAPLEN/3);
595 			newc = stackgap_alloc(p, &sg, clen);
596 			if (!newc) {
597 				error = ENOMEM;
598 				goto done;
599 			}
600 
601 			error = copyout(control, newc, clen);
602 			if (error)
603 				goto done;
604 
605 			msg.msg_control = newc;
606 			msg.msg_controllen = clen;
607 		}
608 
609 		free(control, M_TEMP);
610 		control = NULL;
611 	}
612 
613 	error = sendit(p, SCARG(uap, s), &msg, bflags, retval);
614 
615 done:
616 	if (control)
617 		free(control, M_TEMP);
618 	if (iov != aiov)
619 		free(iov, M_IOV);
620 	return (error);
621 }
622 
623 int
624 linux_sys_recvfrom(l, v, retval)
625 	struct lwp *l;
626 	void *v;
627 	register_t *retval;
628 {
629 	struct linux_sys_recvfrom_args /* {
630 		syscallarg(int) s;
631 		syscallarg(void *) buf;
632 		syscallarg(int) len;
633 		syscallarg(int) flags;
634 		syscallarg(struct osockaddr *) from;
635 		syscallarg(int *) fromlenaddr;
636 	} */ *uap = v;
637 	int		error;
638 	struct sys_recvfrom_args bra;
639 
640 	SCARG(&bra, s) = SCARG(uap, s);
641 	SCARG(&bra, buf) = SCARG(uap, buf);
642 	SCARG(&bra, len) = SCARG(uap, len);
643 	SCARG(&bra, flags) = SCARG(uap, flags);
644 	SCARG(&bra, from) = (struct sockaddr *) SCARG(uap, from);
645 	SCARG(&bra, fromlenaddr) = SCARG(uap, fromlenaddr);
646 
647 	if ((error = sys_recvfrom(l, &bra, retval)))
648 		return (error);
649 
650 	if (SCARG(uap, from) && (error = linux_sa_put(SCARG(uap, from))))
651 		return (error);
652 
653 	return (0);
654 }
655 
656 int
657 linux_sys_recvmsg(l, v, retval)
658 	struct lwp *l;
659 	void *v;
660 	register_t *retval;
661 {
662 	struct linux_sys_recvmsg_args /* {
663 		syscallarg(int) s;
664 		syscallarg(struct msghdr *) msg;
665 		syscallarg(u_int) flags;
666 	} */ *uap = v;
667 	struct msghdr	msg;
668 	int		error;
669 	struct sys_recvmsg_args bsa;
670 	int lflags;
671 	u_int8_t *ocontrol = NULL; /* XXX: gcc */
672 	socklen_t ocontrollen = 0;
673 
674 	/*
675 	 * Data alignment is different on some architectures. If control
676 	 * message is expected, we must arrange for the control message
677 	 * to be initially put elsewhere, and copy to target place
678 	 * with Linux alignment.
679 	 */
680 	if (LINUX_CMSG_ALIGNDIFF) {
681 		error = copyin(SCARG(uap, msg), &msg, sizeof(msg));
682 		if (error)
683 			return (error);
684 
685 		if (CMSG_FIRSTHDR(&msg)) {
686 			caddr_t sg;
687 
688 			/* Need to fit within stackgap */
689 			if (msg.msg_controllen > STACKGAPLEN/2) {
690 				/* Sorry guys! */
691 				return (EINVAL);
692 			}
693 
694 			sg = stackgap_init(l->l_proc, STACKGAPLEN/3);
695 
696 			ocontrol = msg.msg_control;
697 			ocontrollen = msg.msg_controllen;
698 
699 			/* space for at least one message's worth align */
700 			msg.msg_controllen += CMSG_ALIGN(1);
701 
702 			msg.msg_control = stackgap_alloc(l->l_proc, &sg,
703 				msg.msg_controllen);
704 			if (!msg.msg_control)
705 				return (ENOMEM);
706 
707 			/*
708 			 * Okay to overwrite the original structure, it's
709 			 * supposed to be writable.
710 			 */
711 			error = copyout(&msg, SCARG(uap, msg), sizeof(msg));
712 			if (error)
713 				return (error);
714 		}
715 	}
716 
717 	SCARG(&bsa, s) = SCARG(uap, s);
718 	SCARG(&bsa, msg) = SCARG(uap, msg);
719 	SCARG(&bsa, flags) = linux_to_bsd_msg_flags(SCARG(uap, flags));
720 
721 	if (SCARG(&bsa, flags) < 0) {
722 		/* Some unsupported flag */
723 		return (EINVAL);
724 	}
725 
726 	if ((error = sys_recvmsg(l, &bsa, retval)))
727 		goto done;
728 
729 	/* Fixup sockaddr */
730 	error = copyin((caddr_t)SCARG(uap, msg), (caddr_t)&msg,
731 		       sizeof(msg));
732 	if (error)
733 		goto done;
734 
735 	if (msg.msg_name && msg.msg_namelen > 2) {
736 		if ((error = linux_sa_put(msg.msg_name)))
737 			goto done;
738 	}
739 
740 	/* Fixup msg flags */
741 	lflags = bsd_to_linux_msg_flags(msg.msg_flags);
742 	if (lflags < 0) {
743 		/* Some flag unsupported by Linux */
744 		error = EINVAL;
745 		goto done;
746 	}
747 	error = copyout(&lflags, (u_int8_t *) SCARG(uap, msg) +
748 			offsetof(struct msghdr, msg_flags), sizeof(lflags));
749 	if (error)
750 		goto done;
751 
752 	/*
753 	 * Fixup cmsg. We handle two things:
754 	 * 1. different values for level/type on some archs
755 	 * 2. different alignment of CMSG_DATA on some archs
756 	 */
757 	if (CMSG_FIRSTHDR(&msg)) {
758 		struct cmsghdr cmsg, *cc;
759 		int changed = 0;
760 		size_t resid = ocontrollen;
761 
762 		cc = CMSG_FIRSTHDR(&msg);
763 		do {
764 			error = copyin(cc, &cmsg, sizeof(cmsg));
765 			if (error)
766 				goto done;
767 
768 			switch (cmsg.cmsg_level) {
769 			case SOL_SOCKET:
770 				if (SOL_SOCKET != LINUX_SOL_SOCKET) {
771 					cmsg.cmsg_level = LINUX_SOL_SOCKET;
772 					changed = 1;
773 				}
774 
775 				switch (cmsg.cmsg_type) {
776 				case SCM_RIGHTS:
777 					/* Linux SCM_RIGHTS is same as NetBSD */
778 					break;
779 
780 				default:
781 					/* other types not supported */
782 					error = EINVAL;
783 					goto done;
784 				}
785 			default:
786 				/* pray and leave intact */
787 				break;
788 			}
789 
790 			if (LINUX_CMSG_ALIGNDIFF) {
791 				int i;
792 				u_int8_t d, *sd, *td;
793 
794 				/*
795 				 * Sanity check.
796 				 */
797 				if (cmsg.cmsg_len > resid
798 				    || cmsg.cmsg_len < sizeof(cmsg)) {
799 					error = EINVAL;
800 					goto done;
801 				}
802 
803 				/*
804 				 * Need to copy the cmsg from scratch area
805 				 * to the original place, converting data
806 				 * alignment from NetBSD to Linux one.
807 				 */
808 				error = copyout(&cmsg, ocontrol, sizeof(cmsg));
809 				if (error)
810 					goto done;
811 				/* zero pad */
812 				for(i=0; i < LINUX_CMSG_ALIGN(sizeof(cmsg)) - sizeof(cmsg); i++) {
813 					copyout("",&ocontrol[sizeof(cmsg)+i],1);
814 				}
815 
816 				sd = CMSG_DATA(cc);
817 				td = LINUX_CMSG_DATA(ocontrol);
818 
819 				/* This is not particularily effective, but ..*/
820 				d = '\0';
821 				for(i=0; i < cmsg.cmsg_len - sizeof(cmsg); i++){
822 					copyin(sd++, &d, 1);
823 					copyout(&d, td++, 1);
824 				}
825 
826 				resid -= (td - ocontrol);
827 				ocontrol = td;
828 			} else if (changed) {
829 				/* Update cmsghdr in-place */
830 				error = copyout(&cmsg, cc, sizeof(cmsg));
831 				if (error)
832 					goto done;
833 				changed = 0;
834 			}
835 		} while((cc = CMSG_NXTHDR(&msg, cc)));
836 	}
837 
838 done:
839 	return (error);
840 }
841 
842 /*
843  * Convert socket option level from Linux to NetBSD value. Only SOL_SOCKET
844  * is different, the rest matches IPPROTO_* on both systems.
845  */
846 int
847 linux_to_bsd_sopt_level(llevel)
848 	int llevel;
849 {
850 
851 	switch (llevel) {
852 	case LINUX_SOL_SOCKET:
853 		return SOL_SOCKET;
854 	case LINUX_SOL_IP:
855 		return IPPROTO_IP;
856 	case LINUX_SOL_TCP:
857 		return IPPROTO_TCP;
858 	case LINUX_SOL_UDP:
859 		return IPPROTO_UDP;
860 	default:
861 		return -1;
862 	}
863 }
864 
865 /*
866  * Convert Linux socket level socket option numbers to NetBSD values.
867  */
868 int
869 linux_to_bsd_so_sockopt(lopt)
870 	int lopt;
871 {
872 
873 	switch (lopt) {
874 	case LINUX_SO_DEBUG:
875 		return SO_DEBUG;
876 	case LINUX_SO_REUSEADDR:
877 		/*
878 		 * Linux does not implement SO_REUSEPORT, but allows reuse of a
879 		 * host:port pair through SO_REUSEADDR even if the address is not a
880 		 * multicast-address.  Effectively, this means that we should use
881 		 * SO_REUSEPORT to allow Linux applications to not exit with
882 		 * EADDRINUSE
883 		 */
884 		return SO_REUSEPORT;
885 	case LINUX_SO_TYPE:
886 		return SO_TYPE;
887 	case LINUX_SO_ERROR:
888 		return SO_ERROR;
889 	case LINUX_SO_DONTROUTE:
890 		return SO_DONTROUTE;
891 	case LINUX_SO_BROADCAST:
892 		return SO_BROADCAST;
893 	case LINUX_SO_SNDBUF:
894 		return SO_SNDBUF;
895 	case LINUX_SO_RCVBUF:
896 		return SO_RCVBUF;
897 	case LINUX_SO_KEEPALIVE:
898 		return SO_KEEPALIVE;
899 	case LINUX_SO_OOBINLINE:
900 		return SO_OOBINLINE;
901 	case LINUX_SO_LINGER:
902 		return SO_LINGER;
903 	case LINUX_SO_PRIORITY:
904 	case LINUX_SO_NO_CHECK:
905 	default:
906 		return -1;
907 	}
908 }
909 
910 /*
911  * Convert Linux IP level socket option number to NetBSD values.
912  */
913 int
914 linux_to_bsd_ip_sockopt(lopt)
915 	int lopt;
916 {
917 
918 	switch (lopt) {
919 	case LINUX_IP_TOS:
920 		return IP_TOS;
921 	case LINUX_IP_TTL:
922 		return IP_TTL;
923 	case LINUX_IP_MULTICAST_TTL:
924 		return IP_MULTICAST_TTL;
925 	case LINUX_IP_MULTICAST_LOOP:
926 		return IP_MULTICAST_LOOP;
927 	case LINUX_IP_MULTICAST_IF:
928 		return IP_MULTICAST_IF;
929 	case LINUX_IP_ADD_MEMBERSHIP:
930 		return IP_ADD_MEMBERSHIP;
931 	case LINUX_IP_DROP_MEMBERSHIP:
932 		return IP_DROP_MEMBERSHIP;
933 	default:
934 		return -1;
935 	}
936 }
937 
938 /*
939  * Convert Linux TCP level socket option number to NetBSD values.
940  */
941 int
942 linux_to_bsd_tcp_sockopt(lopt)
943 	int lopt;
944 {
945 
946 	switch (lopt) {
947 	case LINUX_TCP_NODELAY:
948 		return TCP_NODELAY;
949 	case LINUX_TCP_MAXSEG:
950 		return TCP_MAXSEG;
951 	default:
952 		return -1;
953 	}
954 }
955 
956 /*
957  * Convert Linux UDP level socket option number to NetBSD values.
958  */
959 int
960 linux_to_bsd_udp_sockopt(lopt)
961 	int lopt;
962 {
963 
964 	switch (lopt) {
965 	default:
966 		return -1;
967 	}
968 }
969 
970 /*
971  * Another reasonably straightforward function: setsockopt(2).
972  * The level and option numbers are converted; the values passed
973  * are not (yet) converted, the ones currently implemented don't
974  * need conversion, as they are the same on both systems.
975  */
976 int
977 linux_sys_setsockopt(l, v, retval)
978 	struct lwp *l;
979 	void *v;
980 	register_t *retval;
981 {
982 	struct linux_sys_setsockopt_args /* {
983 		syscallarg(int) s;
984 		syscallarg(int) level;
985 		syscallarg(int) optname;
986 		syscallarg(void *) optval;
987 		syscallarg(int) optlen;
988 	} */ *uap = v;
989 	struct sys_setsockopt_args bsa;
990 	int name;
991 
992 	SCARG(&bsa, s) = SCARG(uap, s);
993 	SCARG(&bsa, level) = linux_to_bsd_sopt_level(SCARG(uap, level));
994 	SCARG(&bsa, val) = SCARG(uap, optval);
995 	SCARG(&bsa, valsize) = SCARG(uap, optlen);
996 
997 	switch (SCARG(&bsa, level)) {
998 	case SOL_SOCKET:
999 		name = linux_to_bsd_so_sockopt(SCARG(uap, optname));
1000 		break;
1001 	case IPPROTO_IP:
1002 		name = linux_to_bsd_ip_sockopt(SCARG(uap, optname));
1003 		break;
1004 	case IPPROTO_TCP:
1005 		name = linux_to_bsd_tcp_sockopt(SCARG(uap, optname));
1006 		break;
1007 	case IPPROTO_UDP:
1008 		name = linux_to_bsd_udp_sockopt(SCARG(uap, optname));
1009 		break;
1010 	default:
1011 		return EINVAL;
1012 	}
1013 
1014 	if (name == -1)
1015 		return EINVAL;
1016 	SCARG(&bsa, name) = name;
1017 
1018 	return sys_setsockopt(l, &bsa, retval);
1019 }
1020 
1021 /*
1022  * getsockopt(2) is very much the same as setsockopt(2) (see above)
1023  */
1024 int
1025 linux_sys_getsockopt(l, v, retval)
1026 	struct lwp *l;
1027 	void *v;
1028 	register_t *retval;
1029 {
1030 	struct linux_sys_getsockopt_args /* {
1031 		syscallarg(int) s;
1032 		syscallarg(int) level;
1033 		syscallarg(int) optname;
1034 		syscallarg(void *) optval;
1035 		syscallarg(int *) optlen;
1036 	} */ *uap = v;
1037 	struct sys_getsockopt_args bga;
1038 	int name;
1039 
1040 	SCARG(&bga, s) = SCARG(uap, s);
1041 	SCARG(&bga, level) = linux_to_bsd_sopt_level(SCARG(uap, level));
1042 	SCARG(&bga, val) = SCARG(uap, optval);
1043 	SCARG(&bga, avalsize) = SCARG(uap, optlen);
1044 
1045 	switch (SCARG(&bga, level)) {
1046 	case SOL_SOCKET:
1047 		name = linux_to_bsd_so_sockopt(SCARG(uap, optname));
1048 		break;
1049 	case IPPROTO_IP:
1050 		name = linux_to_bsd_ip_sockopt(SCARG(uap, optname));
1051 		break;
1052 	case IPPROTO_TCP:
1053 		name = linux_to_bsd_tcp_sockopt(SCARG(uap, optname));
1054 		break;
1055 	case IPPROTO_UDP:
1056 		name = linux_to_bsd_udp_sockopt(SCARG(uap, optname));
1057 		break;
1058 	default:
1059 		return EINVAL;
1060 	}
1061 
1062 	if (name == -1)
1063 		return EINVAL;
1064 	SCARG(&bga, name) = name;
1065 
1066 	return sys_getsockopt(l, &bga, retval);
1067 }
1068 
1069 #define IF_NAME_LEN 16
1070 
1071 int
1072 linux_getifhwaddr(p, retval, fd, data)
1073 	struct proc *p;
1074 	register_t *retval;
1075 	u_int fd;
1076 	void *data;
1077 {
1078 	/* Not the full structure, just enough to map what we do here */
1079 	struct linux_ifreq {
1080 		char if_name[IF_NAME_LEN];
1081 		struct osockaddr hwaddr;
1082 	} lreq;
1083 	struct filedesc *fdp;
1084 	struct file *fp;
1085 	struct ifaddr *ifa;
1086 	struct ifnet *ifp;
1087 	struct sockaddr_dl *sadl;
1088 	int error, found;
1089 	int index, ifnum;
1090 
1091 	/*
1092 	 * We can't emulate this ioctl by calling sys_ioctl() to run
1093 	 * SIOCGIFCONF, because the user buffer is not of the right
1094 	 * type to take those results.  We can't use kernel buffers to
1095 	 * receive the results, as the implementation of sys_ioctl()
1096 	 * and ifconf() [which implements SIOCGIFCONF] use
1097 	 * copyin()/copyout() which will fail on kernel addresses.
1098 	 *
1099 	 * So, we must duplicate code from sys_ioctl() and ifconf().  Ugh.
1100 	 */
1101 
1102 	fdp = p->p_fd;
1103 	if ((fp = fd_getfile(fdp, fd)) == NULL)
1104 		return (EBADF);
1105 
1106 	FILE_USE(fp);
1107 	if ((fp->f_flag & (FREAD | FWRITE)) == 0) {
1108 		error = EBADF;
1109 		goto out;
1110 	}
1111 
1112 	error = copyin(data, (caddr_t)&lreq, sizeof(lreq));
1113 	if (error)
1114 		goto out;
1115 	lreq.if_name[IF_NAME_LEN-1] = '\0';		/* just in case */
1116 
1117 	/*
1118 	 * Try real interface name first, then fake "ethX"
1119 	 */
1120 	for (ifp = ifnet.tqh_first, found = 0;
1121 	     ifp != 0 && !found;
1122 	     ifp = ifp->if_list.tqe_next) {
1123 		if (strcmp(lreq.if_name, ifp->if_xname))
1124 			/* not this interface */
1125 			continue;
1126 		found=1;
1127 		if ((ifa = ifp->if_addrlist.tqh_first) != 0) {
1128 			for (; ifa != 0; ifa = ifa->ifa_list.tqe_next) {
1129 				sadl = (struct sockaddr_dl *)ifa->ifa_addr;
1130 				/* only return ethernet addresses */
1131 				/* XXX what about FDDI, etc. ? */
1132 				if (sadl->sdl_family != AF_LINK ||
1133 				    sadl->sdl_type != IFT_ETHER)
1134 					continue;
1135 				memcpy((caddr_t)&lreq.hwaddr.sa_data,
1136 				       LLADDR(sadl),
1137 				       MIN(sadl->sdl_alen,
1138 					   sizeof(lreq.hwaddr.sa_data)));
1139 				lreq.hwaddr.sa_family =
1140 					sadl->sdl_family;
1141 				error = copyout((caddr_t)&lreq, data,
1142 						sizeof(lreq));
1143 				goto out;
1144 			}
1145 		} else {
1146 			error = ENODEV;
1147 			goto out;
1148 		}
1149 	}
1150 
1151 	if (strncmp(lreq.if_name, "eth", 3) == 0) {
1152 		for (ifnum = 0, index = 3;
1153 		     lreq.if_name[index] != '\0' && index < IF_NAME_LEN;
1154 		     index++) {
1155 			ifnum *= 10;
1156 			ifnum += lreq.if_name[index] - '0';
1157 		}
1158 
1159 		error = EINVAL;			/* in case we don't find one */
1160 		for (ifp = ifnet.tqh_first, found = 0;
1161 		     ifp != 0 && !found;
1162 		     ifp = ifp->if_list.tqe_next) {
1163 			memcpy(lreq.if_name, ifp->if_xname,
1164 			       MIN(IF_NAME_LEN, IFNAMSIZ));
1165 			if ((ifa = ifp->if_addrlist.tqh_first) == 0)
1166 				/* no addresses on this interface */
1167 				continue;
1168 			else
1169 				for (; ifa != 0; ifa = ifa->ifa_list.tqe_next) {
1170 					sadl = (struct sockaddr_dl *)ifa->ifa_addr;
1171 					/* only return ethernet addresses */
1172 					/* XXX what about FDDI, etc. ? */
1173 					if (sadl->sdl_family != AF_LINK ||
1174 					    sadl->sdl_type != IFT_ETHER)
1175 						continue;
1176 					if (ifnum--)
1177 						/* not the reqested iface */
1178 						continue;
1179 					memcpy((caddr_t)&lreq.hwaddr.sa_data,
1180 					       LLADDR(sadl),
1181 					       MIN(sadl->sdl_alen,
1182 						   sizeof(lreq.hwaddr.sa_data)));
1183 					lreq.hwaddr.sa_family =
1184 						sadl->sdl_family;
1185 					error = copyout((caddr_t)&lreq, data,
1186 							sizeof(lreq));
1187 					found = 1;
1188 					break;
1189 				}
1190 		}
1191 	} else {
1192 		/* unknown interface, not even an "eth*" name */
1193 		error = ENODEV;
1194 	}
1195 
1196 out:
1197 	FILE_UNUSE(fp, p);
1198 	return error;
1199 }
1200 #undef IF_NAME_LEN
1201 
1202 int
1203 linux_ioctl_socket(p, uap, retval)
1204 	struct proc *p;
1205 	struct linux_sys_ioctl_args /* {
1206 		syscallarg(int) fd;
1207 		syscallarg(u_long) com;
1208 		syscallarg(caddr_t) data;
1209 	} */ *uap;
1210 	register_t *retval;
1211 {
1212 	u_long com;
1213 	int error = 0, isdev = 0, dosys = 1;
1214 	struct sys_ioctl_args ia;
1215 	struct file *fp;
1216 	struct filedesc *fdp;
1217 	struct vnode *vp;
1218 	int (*ioctlf)(struct file *, u_long, void *, struct proc *);
1219 	struct ioctl_pt pt;
1220 
1221         fdp = p->p_fd;
1222 	if ((fp = fd_getfile(fdp, SCARG(uap, fd))) == NULL)
1223 		return (EBADF);
1224 
1225 	FILE_USE(fp);
1226 
1227 	if (fp->f_type == DTYPE_VNODE) {
1228 		vp = (struct vnode *)fp->f_data;
1229 		isdev = vp->v_type == VCHR;
1230 	}
1231 
1232 	/*
1233 	 * Don't try to interpret socket ioctl calls that are done
1234 	 * on a device filedescriptor, just pass them through, to
1235 	 * emulate Linux behaviour. Use PTIOCLINUX so that the
1236 	 * device will only handle these if it's prepared to do
1237 	 * so, to avoid unexpected things from happening.
1238 	 */
1239 	if (isdev) {
1240 		dosys = 0;
1241 		ioctlf = fp->f_ops->fo_ioctl;
1242 		pt.com = SCARG(uap, com);
1243 		pt.data = SCARG(uap, data);
1244 		error = ioctlf(fp, PTIOCLINUX, (caddr_t)&pt, p);
1245 		/*
1246 		 * XXX hack: if the function returns EJUSTRETURN,
1247 		 * it has stuffed a sysctl return value in pt.data.
1248 		 */
1249 		if (error == EJUSTRETURN) {
1250 			retval[0] = (register_t)pt.data;
1251 			error = 0;
1252 		}
1253 		goto out;
1254 	}
1255 
1256 	com = SCARG(uap, com);
1257 	retval[0] = 0;
1258 
1259 	switch (com) {
1260 	case LINUX_SIOCGIFCONF:
1261 		SCARG(&ia, com) = OSIOCGIFCONF;
1262 		break;
1263 	case LINUX_SIOCGIFFLAGS:
1264 		SCARG(&ia, com) = SIOCGIFFLAGS;
1265 		break;
1266 	case LINUX_SIOCSIFFLAGS:
1267 		SCARG(&ia, com) = SIOCSIFFLAGS;
1268 		break;
1269 	case LINUX_SIOCGIFADDR:
1270 		SCARG(&ia, com) = OSIOCGIFADDR;
1271 		break;
1272 	case LINUX_SIOCGIFDSTADDR:
1273 		SCARG(&ia, com) = OSIOCGIFDSTADDR;
1274 		break;
1275 	case LINUX_SIOCGIFBRDADDR:
1276 		SCARG(&ia, com) = OSIOCGIFBRDADDR;
1277 		break;
1278 	case LINUX_SIOCGIFNETMASK:
1279 		SCARG(&ia, com) = OSIOCGIFNETMASK;
1280 		break;
1281 	case LINUX_SIOCADDMULTI:
1282 		SCARG(&ia, com) = SIOCADDMULTI;
1283 		break;
1284 	case LINUX_SIOCDELMULTI:
1285 		SCARG(&ia, com) = SIOCDELMULTI;
1286 		break;
1287 	case LINUX_SIOCGIFHWADDR:
1288 	        error = linux_getifhwaddr(p, retval, SCARG(uap, fd),
1289 					 SCARG(uap, data));
1290 		dosys = 0;
1291 		break;
1292 	default:
1293 		error = EINVAL;
1294 	}
1295 
1296 out:
1297 	FILE_UNUSE(fp, p);
1298 
1299 	if (error ==0 && dosys) {
1300 		SCARG(&ia, fd) = SCARG(uap, fd);
1301 		SCARG(&ia, data) = SCARG(uap, data);
1302 		/* XXX NJWLWP */
1303 		error = sys_ioctl(curlwp, &ia, retval);
1304 	}
1305 
1306 	return error;
1307 }
1308 
1309 int
1310 linux_sys_connect(l, v, retval)
1311 	struct lwp *l;
1312 	void *v;
1313 	register_t *retval;
1314 {
1315 	struct linux_sys_connect_args /* {
1316 		syscallarg(int) s;
1317 		syscallarg(const struct sockaddr *) name;
1318 		syscallarg(int) namelen;
1319 	} */ *uap = v;
1320 	struct proc *p = l->l_proc;
1321 	int		error;
1322 	struct sockaddr *sa;
1323 	struct sys_connect_args bca;
1324 	caddr_t sg = stackgap_init(p, 0);
1325 	int namlen;
1326 
1327 	namlen = SCARG(uap, namelen);
1328 	error = linux_sa_get(p, SCARG(uap, s), &sg, &sa,
1329 	    SCARG(uap, name), &namlen);
1330 	if (error)
1331 		return (error);
1332 
1333 	SCARG(&bca, s) = SCARG(uap, s);
1334 	SCARG(&bca, name) = sa;
1335 	SCARG(&bca, namelen) = (unsigned int) namlen;
1336 
1337 	error = sys_connect(l, &bca, retval);
1338 
1339 	if (error == EISCONN) {
1340 		struct file *fp;
1341 		struct socket *so;
1342 		int s, state, prflags;
1343 
1344 		/* getsock() will use the descriptor for us */
1345 	    	if (getsock(p->p_fd, SCARG(uap, s), &fp) != 0)
1346 		    	return EISCONN;
1347 
1348 		s = splsoftnet();
1349 		so = (struct socket *)fp->f_data;
1350 		state = so->so_state;
1351 		prflags = so->so_proto->pr_flags;
1352 		splx(s);
1353 		FILE_UNUSE(fp, p);
1354 		/*
1355 		 * We should only let this call succeed once per
1356 		 * non-blocking connect; however we don't have
1357 		 * a convenient place to keep that state..
1358 		 */
1359 		if ((state & SS_NBIO) && (state & SS_ISCONNECTED) &&
1360 		    (prflags & PR_CONNREQUIRED))
1361 			return 0;
1362 	}
1363 
1364 	return (error);
1365 }
1366 
1367 int
1368 linux_sys_bind(l, v, retval)
1369 	struct lwp *l;
1370 	void *v;
1371 	register_t *retval;
1372 {
1373 	struct linux_sys_bind_args /* {
1374 		syscallarg(int) s;
1375 		syscallarg(const struct osockaddr *) name;
1376 		syscallarg(int) namelen;
1377 	} */ *uap = v;
1378 	struct proc *p = l->l_proc;
1379 	int		error, namlen;
1380 	struct sys_bind_args bsa;
1381 
1382 	namlen = SCARG(uap, namelen);
1383 	SCARG(&bsa, s) = SCARG(uap, s);
1384 	if (SCARG(uap, name)) {
1385 		struct sockaddr *sa;
1386 		caddr_t sg = stackgap_init(p, 0);
1387 
1388 		error = linux_sa_get(p, SCARG(uap, s), &sg, &sa,
1389 		    SCARG(uap, name), &namlen);
1390 		if (error)
1391 			return (error);
1392 
1393 		SCARG(&bsa, name) = sa;
1394 	} else
1395 		SCARG(&bsa, name) = NULL;
1396 	SCARG(&bsa, namelen) = namlen;
1397 
1398 	return (sys_bind(l, &bsa, retval));
1399 }
1400 
1401 int
1402 linux_sys_getsockname(l, v, retval)
1403 	struct lwp *l;
1404 	void *v;
1405 	register_t *retval;
1406 {
1407 	struct linux_sys_getsockname_args /* {
1408 		syscallarg(int) fdes;
1409 		syscallarg(caddr_t) asa;
1410 		syscallarg(int *) alen;
1411 	} */ *uap = v;
1412 	int error;
1413 
1414 	if ((error = sys_getsockname(l, uap, retval)) != 0)
1415 		return (error);
1416 
1417 	if ((error = linux_sa_put((struct osockaddr *)SCARG(uap, asa))))
1418 		return (error);
1419 
1420 	return (0);
1421 }
1422 
1423 int
1424 linux_sys_getpeername(l, v, retval)
1425 	struct lwp *l;
1426 	void *v;
1427 	register_t *retval;
1428 {
1429 	struct sys_getpeername_args /* {
1430 		syscallarg(int) fdes;
1431 		syscallarg(caddr_t) asa;
1432 		syscallarg(int *) alen;
1433 	} */ *uap = v;
1434 	int error;
1435 
1436 	if ((error = sys_getpeername(l, uap, retval)) != 0)
1437 		return (error);
1438 
1439 	if ((error = linux_sa_put((struct osockaddr *)SCARG(uap, asa))))
1440 		return (error);
1441 
1442 	return (0);
1443 }
1444 
1445 /*
1446  * Copy the osockaddr structure pointed to by osa to kernel, adjust
1447  * family and convert to sockaddr, allocate stackgap and put the
1448  * the converted structure there, address on stackgap returned in sap.
1449  */
1450 static int
1451 linux_sa_get(p, s, sgp, sap, osa, osalen)
1452 	struct proc *p;
1453 	int s;
1454 	caddr_t *sgp;
1455 	struct sockaddr **sap;
1456 	const struct osockaddr *osa;
1457 	int *osalen;
1458 {
1459 	int error=0, bdom;
1460 	struct sockaddr *sa, *usa;
1461 	struct osockaddr *kosa = (struct osockaddr *) &sa;
1462 	int alloclen;
1463 #ifdef INET6
1464 	int oldv6size;
1465 	struct sockaddr_in6 *sin6;
1466 #endif
1467 
1468 	if (*osalen < 2 || *osalen > UCHAR_MAX || !osa) {
1469 		DPRINTF(("bad osa=%p osalen=%d\n", osa, *osalen));
1470 		return (EINVAL);
1471 	}
1472 
1473 	alloclen = *osalen;
1474 #ifdef INET6
1475 	oldv6size = 0;
1476 	/*
1477 	 * Check for old (pre-RFC2553) sockaddr_in6. We may accept it
1478 	 * if it's a v4-mapped address, so reserve the proper space
1479 	 * for it.
1480 	 */
1481 	if (alloclen == sizeof (struct sockaddr_in6) - sizeof (u_int32_t)) {
1482 		alloclen = sizeof (struct sockaddr_in6);
1483 		oldv6size = 1;
1484 	}
1485 #endif
1486 
1487 	kosa = (struct osockaddr *) malloc(alloclen, M_TEMP, M_WAITOK);
1488 
1489 	if ((error = copyin(osa, (caddr_t) kosa, *osalen))) {
1490 		DPRINTF(("error copying osa %d\n", error));
1491 		goto out;
1492 	}
1493 
1494 	bdom = linux_to_bsd_domain(kosa->sa_family);
1495 	if (bdom == -1) {
1496 		DPRINTF(("bad linux family=%d\n", kosa->sa_family));
1497 		error = EINVAL;
1498 		goto out;
1499 	}
1500 
1501 #ifdef INET6
1502 	/*
1503 	 * Older Linux IPv6 code uses obsolete RFC2133 struct sockaddr_in6,
1504 	 * which lacks the scope id compared with RFC2553 one. If we detect
1505 	 * the situation, reject the address and write a message to system log.
1506 	 *
1507 	 * Still accept addresses for which the scope id is not used.
1508 	 */
1509 	if (oldv6size && bdom == AF_INET6) {
1510 		sin6 = (struct sockaddr_in6 *)kosa;
1511 		if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr) ||
1512 		    (!IN6_IS_ADDR_LINKLOCAL(&sin6->sin6_addr) &&
1513 		     !IN6_IS_ADDR_SITELOCAL(&sin6->sin6_addr) &&
1514 		     !IN6_IS_ADDR_V4COMPAT(&sin6->sin6_addr) &&
1515 		     !IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr) &&
1516 		     !IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr))) {
1517 			sin6->sin6_scope_id = 0;
1518 		} else {
1519 			int uid = p->p_cred && p->p_ucred ?
1520 					p->p_ucred->cr_uid : -1;
1521 
1522 			log(LOG_DEBUG,
1523 			    "pid %d (%s), uid %d: obsolete pre-RFC2553 "
1524 			    "sockaddr_in6 rejected",
1525 			    p->p_pid, p->p_comm, uid);
1526 			error = EINVAL;
1527 			goto out;
1528 		}
1529 	}
1530 #endif
1531 
1532 	/*
1533 	 * If the family is unspecified, use address family of the
1534 	 * socket. This avoid triggering COMPAT_43 struct socket family check
1535 	 * in sockargs() on little-endian machines, and strict family checks
1536 	 * in netinet/in_pcb.c et.al.
1537 	 */
1538 	if (bdom == AF_UNSPEC) {
1539 		struct file *fp;
1540 		struct socket *so;
1541 
1542 		/* getsock() will use the descriptor for us */
1543 		if ((error = getsock(p->p_fd, s, &fp)) != 0)
1544 			goto out;
1545 
1546 		so = (struct socket *)fp->f_data;
1547 		bdom = so->so_proto->pr_domain->dom_family;
1548 
1549 		FILE_UNUSE(fp, p);
1550 
1551 		DPRINTF(("AF_UNSPEC family adjusted to %d\n", bdom));
1552 	}
1553 
1554 	if (bdom == AF_INET) {
1555 		alloclen = sizeof(struct sockaddr_in);
1556 	}
1557 
1558 	sa = (struct sockaddr *) kosa;
1559 	sa->sa_family = bdom;
1560 	sa->sa_len = alloclen;
1561 #ifdef DEBUG_LINUX
1562 	DPRINTF(("family %d, len = %d [ ", sa->sa_family, sa->sa_len));
1563 	for (bdom = 0; bdom < sizeof(sa->sa_data); bdom++)
1564 	    DPRINTF(("%02x ", (unsigned char) sa->sa_data[bdom]));
1565 	DPRINTF(("\n"));
1566 #endif
1567 
1568 	usa = (struct sockaddr *) stackgap_alloc(p, sgp, alloclen);
1569 	if (!usa) {
1570 		error = ENOMEM;
1571 		goto out;
1572 	}
1573 
1574 	if ((error = copyout(sa, usa, alloclen))) {
1575 		DPRINTF(("error copying out socket %d\n", error));
1576 		goto out;
1577 	}
1578 
1579 	*sap = usa;
1580 
1581     out:
1582 	*osalen = alloclen;
1583 	free(kosa, M_TEMP);
1584 	return (error);
1585 }
1586 
1587 static int
1588 linux_sa_put(osa)
1589 	struct osockaddr *osa;
1590 {
1591 	struct sockaddr sa;
1592 	struct osockaddr *kosa;
1593 	int error, bdom, len;
1594 
1595 	/*
1596 	 * Only read/write the sockaddr family and length part, the rest is
1597 	 * not changed.
1598 	 */
1599 	len = sizeof(sa.sa_len) + sizeof(sa.sa_family);
1600 
1601 	error = copyin((caddr_t) osa, (caddr_t) &sa, len);
1602 	if (error)
1603 		return (error);
1604 
1605 	bdom = bsd_to_linux_domain(sa.sa_family);
1606 	if (bdom == -1)
1607 		return (EINVAL);
1608 
1609 	/* Note: we convert from sockaddr to osockaddr here, too */
1610 	kosa = (struct osockaddr *) &sa;
1611 	kosa->sa_family = bdom;
1612 	error = copyout(kosa, osa, len);
1613 	if (error)
1614 		return (error);
1615 
1616 	return (0);
1617 }
1618 
1619 int
1620 linux_sys_recv(l, v, retval)
1621 	struct lwp *l;
1622 	void *v;
1623 	register_t *retval;
1624 {
1625 	struct linux_sys_recv_args /* {
1626 		syscallarg(int) s;
1627 		syscallarg(void *) buf;
1628 		syscallarg(int) len;
1629 		syscallarg(int) flags;
1630 	} */ *uap = v;
1631 	struct sys_recvfrom_args bra;
1632 
1633 
1634 	SCARG(&bra, s) = SCARG(uap, s);
1635 	SCARG(&bra, buf) = SCARG(uap, buf);
1636 	SCARG(&bra, len) = (size_t) SCARG(uap, len);
1637 	SCARG(&bra, flags) = SCARG(uap, flags);
1638 	SCARG(&bra, from) = NULL;
1639 	SCARG(&bra, fromlenaddr) = NULL;
1640 
1641 	return (sys_recvfrom(l, &bra, retval));
1642 }
1643 
1644 int
1645 linux_sys_send(l, v, retval)
1646 	struct lwp *l;
1647 	void *v;
1648 	register_t *retval;
1649 {
1650 	struct linux_sys_send_args /* {
1651 		syscallarg(int) s;
1652 		syscallarg(caddr_t) buf;
1653 		syscallarg(int) len;
1654 		syscallarg(int) flags;
1655 	} */ *uap = v;
1656 	struct sys_sendto_args bsa;
1657 
1658 	SCARG(&bsa, s)		= SCARG(uap, s);
1659 	SCARG(&bsa, buf)	= SCARG(uap, buf);
1660 	SCARG(&bsa, len)	= SCARG(uap, len);
1661 	SCARG(&bsa, flags)	= SCARG(uap, flags);
1662 	SCARG(&bsa, to)		= NULL;
1663 	SCARG(&bsa, tolen)	= 0;
1664 
1665 	return (sys_sendto(l, &bsa, retval));
1666 }
1667 
1668 int
1669 linux_sys_accept(l, v, retval)
1670 	struct lwp *l;
1671 	void *v;
1672 	register_t *retval;
1673 {
1674 	struct linux_sys_accept_args /* {
1675 		syscallarg(int) s;
1676 		syscallarg(struct osockaddr *) name;
1677 		syscallarg(int *) anamelen;
1678 	} */ *uap = v;
1679 	int error;
1680 	struct sys_accept_args baa;
1681 
1682 	SCARG(&baa, s)		= SCARG(uap, s);
1683 	SCARG(&baa, name)	= (struct sockaddr *) SCARG(uap, name);
1684 	SCARG(&baa, anamelen)	= (unsigned int *) SCARG(uap, anamelen);
1685 
1686 	if ((error = sys_accept(l, &baa, retval)))
1687 		return (error);
1688 
1689 	if (SCARG(uap, name) && (error = linux_sa_put(SCARG(uap, name))))
1690 		return (error);
1691 
1692 	return (0);
1693 }
1694