xref: /csrg-svn/sys/kern/uipc_usrreq.c (revision 47540)
1 /*
2  * Copyright (c) 1982, 1986, 1989 Regents of the University of California.
3  * All rights reserved.
4  *
5  * %sccs.include.redist.c%
6  *
7  *	@(#)uipc_usrreq.c	7.23 (Berkeley) 03/17/91
8  */
9 
10 #include "param.h"
11 #include "user.h"
12 #include "proc.h"
13 #include "filedesc.h"
14 #include "domain.h"
15 #include "protosw.h"
16 #include "socket.h"
17 #include "socketvar.h"
18 #include "unpcb.h"
19 #include "un.h"
20 #include "vnode.h"
21 #include "file.h"
22 #include "stat.h"
23 #include "mbuf.h"
24 
25 /*
26  * Unix communications domain.
27  *
28  * TODO:
29  *	SEQPACKET, RDM
30  *	rethink name space problems
31  *	need a proper out-of-band
32  */
33 struct	sockaddr sun_noname = { sizeof(sun_noname), AF_UNIX };
34 ino_t	unp_ino;			/* prototype for fake inode numbers */
35 
36 /*ARGSUSED*/
37 uipc_usrreq(so, req, m, nam, control)
38 	struct socket *so;
39 	int req;
40 	struct mbuf *m, *nam, *control;
41 {
42 	struct unpcb *unp = sotounpcb(so);
43 	register struct socket *so2;
44 	register int error = 0;
45 
46 	if (req == PRU_CONTROL)
47 		return (EOPNOTSUPP);
48 	if (req != PRU_SEND && control && control->m_len) {
49 		error = EOPNOTSUPP;
50 		goto release;
51 	}
52 	if (unp == 0 && req != PRU_ATTACH) {
53 		error = EINVAL;
54 		goto release;
55 	}
56 	switch (req) {
57 
58 	case PRU_ATTACH:
59 		if (unp) {
60 			error = EISCONN;
61 			break;
62 		}
63 		error = unp_attach(so);
64 		break;
65 
66 	case PRU_DETACH:
67 		unp_detach(unp);
68 		break;
69 
70 	case PRU_BIND:
71 		error = unp_bind(unp, nam);
72 		break;
73 
74 	case PRU_LISTEN:
75 		if (unp->unp_vnode == 0)
76 			error = EINVAL;
77 		break;
78 
79 	case PRU_CONNECT:
80 		error = unp_connect(so, nam);
81 		break;
82 
83 	case PRU_CONNECT2:
84 		error = unp_connect2(so, (struct socket *)nam);
85 		break;
86 
87 	case PRU_DISCONNECT:
88 		unp_disconnect(unp);
89 		break;
90 
91 	case PRU_ACCEPT:
92 		/*
93 		 * Pass back name of connected socket,
94 		 * if it was bound and we are still connected
95 		 * (our peer may have closed already!).
96 		 */
97 		if (unp->unp_conn && unp->unp_conn->unp_addr) {
98 			nam->m_len = unp->unp_conn->unp_addr->m_len;
99 			bcopy(mtod(unp->unp_conn->unp_addr, caddr_t),
100 			    mtod(nam, caddr_t), (unsigned)nam->m_len);
101 		} else {
102 			nam->m_len = sizeof(sun_noname);
103 			*(mtod(nam, struct sockaddr *)) = sun_noname;
104 		}
105 		break;
106 
107 	case PRU_SHUTDOWN:
108 		socantsendmore(so);
109 		unp_shutdown(unp);
110 		break;
111 
112 	case PRU_RCVD:
113 		switch (so->so_type) {
114 
115 		case SOCK_DGRAM:
116 			panic("uipc 1");
117 			/*NOTREACHED*/
118 
119 		case SOCK_STREAM:
120 #define	rcv (&so->so_rcv)
121 #define snd (&so2->so_snd)
122 			if (unp->unp_conn == 0)
123 				break;
124 			so2 = unp->unp_conn->unp_socket;
125 			/*
126 			 * Adjust backpressure on sender
127 			 * and wakeup any waiting to write.
128 			 */
129 			snd->sb_mbmax += unp->unp_mbcnt - rcv->sb_mbcnt;
130 			unp->unp_mbcnt = rcv->sb_mbcnt;
131 			snd->sb_hiwat += unp->unp_cc - rcv->sb_cc;
132 			unp->unp_cc = rcv->sb_cc;
133 			sowwakeup(so2);
134 #undef snd
135 #undef rcv
136 			break;
137 
138 		default:
139 			panic("uipc 2");
140 		}
141 		break;
142 
143 	case PRU_SEND:
144 		if (control && (error = unp_internalize(control)))
145 			break;
146 		switch (so->so_type) {
147 
148 		case SOCK_DGRAM: {
149 			struct sockaddr *from;
150 
151 			if (nam) {
152 				if (unp->unp_conn) {
153 					error = EISCONN;
154 					break;
155 				}
156 				error = unp_connect(so, nam);
157 				if (error)
158 					break;
159 			} else {
160 				if (unp->unp_conn == 0) {
161 					error = ENOTCONN;
162 					break;
163 				}
164 			}
165 			so2 = unp->unp_conn->unp_socket;
166 			if (unp->unp_addr)
167 				from = mtod(unp->unp_addr, struct sockaddr *);
168 			else
169 				from = &sun_noname;
170 			if (sbappendaddr(&so2->so_rcv, from, m, control)) {
171 				sorwakeup(so2);
172 				m = 0;
173 				control = 0;
174 			} else
175 				error = ENOBUFS;
176 			if (nam)
177 				unp_disconnect(unp);
178 			break;
179 		}
180 
181 		case SOCK_STREAM:
182 #define	rcv (&so2->so_rcv)
183 #define	snd (&so->so_snd)
184 			if (so->so_state & SS_CANTSENDMORE) {
185 				error = EPIPE;
186 				break;
187 			}
188 			if (unp->unp_conn == 0)
189 				panic("uipc 3");
190 			so2 = unp->unp_conn->unp_socket;
191 			/*
192 			 * Send to paired receive port, and then reduce
193 			 * send buffer hiwater marks to maintain backpressure.
194 			 * Wake up readers.
195 			 */
196 			if (control) {
197 				if (sbappendcontrol(rcv, m, control))
198 					control = 0;
199 			} else
200 				sbappend(rcv, m);
201 			snd->sb_mbmax -=
202 			    rcv->sb_mbcnt - unp->unp_conn->unp_mbcnt;
203 			unp->unp_conn->unp_mbcnt = rcv->sb_mbcnt;
204 			snd->sb_hiwat -= rcv->sb_cc - unp->unp_conn->unp_cc;
205 			unp->unp_conn->unp_cc = rcv->sb_cc;
206 			sorwakeup(so2);
207 			m = 0;
208 #undef snd
209 #undef rcv
210 			break;
211 
212 		default:
213 			panic("uipc 4");
214 		}
215 		break;
216 
217 	case PRU_ABORT:
218 		unp_drop(unp, ECONNABORTED);
219 		break;
220 
221 	case PRU_SENSE:
222 		((struct stat *) m)->st_blksize = so->so_snd.sb_hiwat;
223 		if (so->so_type == SOCK_STREAM && unp->unp_conn != 0) {
224 			so2 = unp->unp_conn->unp_socket;
225 			((struct stat *) m)->st_blksize += so2->so_rcv.sb_cc;
226 		}
227 		((struct stat *) m)->st_dev = NODEV;
228 		if (unp->unp_ino == 0)
229 			unp->unp_ino = unp_ino++;
230 		((struct stat *) m)->st_ino = unp->unp_ino;
231 		return (0);
232 
233 	case PRU_RCVOOB:
234 		return (EOPNOTSUPP);
235 
236 	case PRU_SENDOOB:
237 		error = EOPNOTSUPP;
238 		break;
239 
240 	case PRU_SOCKADDR:
241 		if (unp->unp_addr) {
242 			nam->m_len = unp->unp_addr->m_len;
243 			bcopy(mtod(unp->unp_addr, caddr_t),
244 			    mtod(nam, caddr_t), (unsigned)nam->m_len);
245 		} else
246 			nam->m_len = 0;
247 		break;
248 
249 	case PRU_PEERADDR:
250 		if (unp->unp_conn && unp->unp_conn->unp_addr) {
251 			nam->m_len = unp->unp_conn->unp_addr->m_len;
252 			bcopy(mtod(unp->unp_conn->unp_addr, caddr_t),
253 			    mtod(nam, caddr_t), (unsigned)nam->m_len);
254 		} else
255 			nam->m_len = 0;
256 		break;
257 
258 	case PRU_SLOWTIMO:
259 		break;
260 
261 	default:
262 		panic("piusrreq");
263 	}
264 release:
265 	if (control)
266 		m_freem(control);
267 	if (m)
268 		m_freem(m);
269 	return (error);
270 }
271 
272 /*
273  * Both send and receive buffers are allocated PIPSIZ bytes of buffering
274  * for stream sockets, although the total for sender and receiver is
275  * actually only PIPSIZ.
276  * Datagram sockets really use the sendspace as the maximum datagram size,
277  * and don't really want to reserve the sendspace.  Their recvspace should
278  * be large enough for at least one max-size datagram plus address.
279  */
280 #define	PIPSIZ	4096
281 u_long	unpst_sendspace = PIPSIZ;
282 u_long	unpst_recvspace = PIPSIZ;
283 u_long	unpdg_sendspace = 2*1024;	/* really max datagram size */
284 u_long	unpdg_recvspace = 4*1024;
285 
286 int	unp_rights;			/* file descriptors in flight */
287 
288 unp_attach(so)
289 	struct socket *so;
290 {
291 	register struct mbuf *m;
292 	register struct unpcb *unp;
293 	int error;
294 
295 	if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
296 		switch (so->so_type) {
297 
298 		case SOCK_STREAM:
299 			error = soreserve(so, unpst_sendspace, unpst_recvspace);
300 			break;
301 
302 		case SOCK_DGRAM:
303 			error = soreserve(so, unpdg_sendspace, unpdg_recvspace);
304 			break;
305 		}
306 		if (error)
307 			return (error);
308 	}
309 	m = m_getclr(M_DONTWAIT, MT_PCB);
310 	if (m == NULL)
311 		return (ENOBUFS);
312 	unp = mtod(m, struct unpcb *);
313 	so->so_pcb = (caddr_t)unp;
314 	unp->unp_socket = so;
315 	return (0);
316 }
317 
318 unp_detach(unp)
319 	register struct unpcb *unp;
320 {
321 
322 	if (unp->unp_vnode) {
323 		unp->unp_vnode->v_socket = 0;
324 		vrele(unp->unp_vnode);
325 		unp->unp_vnode = 0;
326 	}
327 	if (unp->unp_conn)
328 		unp_disconnect(unp);
329 	while (unp->unp_refs)
330 		unp_drop(unp->unp_refs, ECONNRESET);
331 	soisdisconnected(unp->unp_socket);
332 	unp->unp_socket->so_pcb = 0;
333 	m_freem(unp->unp_addr);
334 	(void) m_free(dtom(unp));
335 	if (unp_rights)
336 		unp_gc();
337 }
338 
339 unp_bind(unp, nam)
340 	struct unpcb *unp;
341 	struct mbuf *nam;
342 {
343 	struct sockaddr_un *soun = mtod(nam, struct sockaddr_un *);
344 	register struct vnode *vp;
345 	register struct nameidata *ndp;
346 	struct vattr vattr;
347 	int error;
348 	struct nameidata nd;
349 
350 	ndp = &nd;
351 	ndp->ni_dirp = soun->sun_path;
352 	if (unp->unp_vnode != NULL)
353 		return (EINVAL);
354 	if (nam->m_len == MLEN) {
355 		if (*(mtod(nam, caddr_t) + nam->m_len - 1) != 0)
356 			return (EINVAL);
357 	} else
358 		*(mtod(nam, caddr_t) + nam->m_len) = 0;
359 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
360 	ndp->ni_nameiop = CREATE | FOLLOW | LOCKPARENT;
361 	ndp->ni_segflg = UIO_SYSSPACE;
362 	if (error = namei(ndp, curproc)) 		/* XXX */
363 		return (error);
364 	vp = ndp->ni_vp;
365 	if (vp != NULL) {
366 		VOP_ABORTOP(ndp);
367 		if (ndp->ni_dvp == vp)
368 			vrele(ndp->ni_dvp);
369 		else
370 			vput(ndp->ni_dvp);
371 		vrele(vp);
372 		return (EADDRINUSE);
373 	}
374 	VATTR_NULL(&vattr);
375 	vattr.va_type = VSOCK;
376 	vattr.va_mode = 0777;
377 	if (error = VOP_CREATE(ndp, &vattr))
378 		return (error);
379 	vp = ndp->ni_vp;
380 	vp->v_socket = unp->unp_socket;
381 	unp->unp_vnode = vp;
382 	unp->unp_addr = m_copy(nam, 0, (int)M_COPYALL);
383 	VOP_UNLOCK(vp);
384 	return (0);
385 }
386 
387 unp_connect(so, nam)
388 	struct socket *so;
389 	struct mbuf *nam;
390 {
391 	register struct sockaddr_un *soun = mtod(nam, struct sockaddr_un *);
392 	register struct vnode *vp;
393 	register struct socket *so2, *so3;
394 	register struct nameidata *ndp;
395 	struct unpcb *unp2, *unp3;
396 	int error;
397 	struct nameidata nd;
398 
399 	ndp = &nd;
400 	ndp->ni_dirp = soun->sun_path;
401 	if (nam->m_data + nam->m_len == &nam->m_dat[MLEN]) {	/* XXX */
402 		if (*(mtod(nam, caddr_t) + nam->m_len - 1) != 0)
403 			return (EMSGSIZE);
404 	} else
405 		*(mtod(nam, caddr_t) + nam->m_len) = 0;
406 	ndp->ni_nameiop = LOOKUP | FOLLOW | LOCKLEAF;
407 	ndp->ni_segflg = UIO_SYSSPACE;
408 	if (error = namei(ndp, curproc))		/* XXX */
409 		return (error);
410 	vp = ndp->ni_vp;
411 	if (vp->v_type != VSOCK) {
412 		error = ENOTSOCK;
413 		goto bad;
414 	}
415 	if (error = VOP_ACCESS(vp, VWRITE, curproc->p_ucred))
416 		goto bad;
417 	so2 = vp->v_socket;
418 	if (so2 == 0) {
419 		error = ECONNREFUSED;
420 		goto bad;
421 	}
422 	if (so->so_type != so2->so_type) {
423 		error = EPROTOTYPE;
424 		goto bad;
425 	}
426 	if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
427 		if ((so2->so_options & SO_ACCEPTCONN) == 0 ||
428 		    (so3 = sonewconn(so2, 0)) == 0) {
429 			error = ECONNREFUSED;
430 			goto bad;
431 		}
432 		unp2 = sotounpcb(so2);
433 		unp3 = sotounpcb(so3);
434 		if (unp2->unp_addr)
435 			unp3->unp_addr =
436 				  m_copy(unp2->unp_addr, 0, (int)M_COPYALL);
437 		so2 = so3;
438 	}
439 	error = unp_connect2(so, so2);
440 bad:
441 	vput(vp);
442 	return (error);
443 }
444 
445 unp_connect2(so, so2)
446 	register struct socket *so;
447 	register struct socket *so2;
448 {
449 	register struct unpcb *unp = sotounpcb(so);
450 	register struct unpcb *unp2;
451 
452 	if (so2->so_type != so->so_type)
453 		return (EPROTOTYPE);
454 	unp2 = sotounpcb(so2);
455 	unp->unp_conn = unp2;
456 	switch (so->so_type) {
457 
458 	case SOCK_DGRAM:
459 		unp->unp_nextref = unp2->unp_refs;
460 		unp2->unp_refs = unp;
461 		soisconnected(so);
462 		break;
463 
464 	case SOCK_STREAM:
465 		unp2->unp_conn = unp;
466 		soisconnected(so);
467 		soisconnected(so2);
468 		break;
469 
470 	default:
471 		panic("unp_connect2");
472 	}
473 	return (0);
474 }
475 
476 unp_disconnect(unp)
477 	struct unpcb *unp;
478 {
479 	register struct unpcb *unp2 = unp->unp_conn;
480 
481 	if (unp2 == 0)
482 		return;
483 	unp->unp_conn = 0;
484 	switch (unp->unp_socket->so_type) {
485 
486 	case SOCK_DGRAM:
487 		if (unp2->unp_refs == unp)
488 			unp2->unp_refs = unp->unp_nextref;
489 		else {
490 			unp2 = unp2->unp_refs;
491 			for (;;) {
492 				if (unp2 == 0)
493 					panic("unp_disconnect");
494 				if (unp2->unp_nextref == unp)
495 					break;
496 				unp2 = unp2->unp_nextref;
497 			}
498 			unp2->unp_nextref = unp->unp_nextref;
499 		}
500 		unp->unp_nextref = 0;
501 		unp->unp_socket->so_state &= ~SS_ISCONNECTED;
502 		break;
503 
504 	case SOCK_STREAM:
505 		soisdisconnected(unp->unp_socket);
506 		unp2->unp_conn = 0;
507 		soisdisconnected(unp2->unp_socket);
508 		break;
509 	}
510 }
511 
512 #ifdef notdef
513 unp_abort(unp)
514 	struct unpcb *unp;
515 {
516 
517 	unp_detach(unp);
518 }
519 #endif
520 
521 unp_shutdown(unp)
522 	struct unpcb *unp;
523 {
524 	struct socket *so;
525 
526 	if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn &&
527 	    (so = unp->unp_conn->unp_socket))
528 		socantrcvmore(so);
529 }
530 
531 unp_drop(unp, errno)
532 	struct unpcb *unp;
533 	int errno;
534 {
535 	struct socket *so = unp->unp_socket;
536 
537 	so->so_error = errno;
538 	unp_disconnect(unp);
539 	if (so->so_head) {
540 		so->so_pcb = (caddr_t) 0;
541 		m_freem(unp->unp_addr);
542 		(void) m_free(dtom(unp));
543 		sofree(so);
544 	}
545 }
546 
547 #ifdef notdef
548 unp_drain()
549 {
550 
551 }
552 #endif
553 
554 unp_externalize(rights)
555 	struct mbuf *rights;
556 {
557 	struct proc *p = curproc;		/* XXX */
558 	register int i;
559 	register struct cmsghdr *cm = mtod(rights, struct cmsghdr *);
560 	register struct file **rp = (struct file **)(cm + 1);
561 	register struct file *fp;
562 	int newfds = (cm->cmsg_len - sizeof(*cm)) / sizeof (int);
563 	int f;
564 
565 	if (fdavail(p, newfds)) {
566 		for (i = 0; i < newfds; i++) {
567 			fp = *rp;
568 			unp_discard(fp);
569 			*rp++ = 0;
570 		}
571 		return (EMSGSIZE);
572 	}
573 	for (i = 0; i < newfds; i++) {
574 		if (fdalloc(p, 0, &f))
575 			panic("unp_externalize");
576 		fp = *rp;
577 		OFILE(p->p_fd, f) = fp;
578 		fp->f_msgcount--;
579 		unp_rights--;
580 		*(int *)rp++ = f;
581 	}
582 	return (0);
583 }
584 
585 unp_internalize(control)
586 	struct mbuf *control;
587 {
588 	struct filedesc *fdp = curproc->p_fd;		/* XXX */
589 	register struct cmsghdr *cm = mtod(control, struct cmsghdr *);
590 	register struct file **rp;
591 	register struct file *fp;
592 	register int i, fd;
593 	int oldfds;
594 
595 	if (cm->cmsg_type != SCM_RIGHTS || cm->cmsg_level != SOL_SOCKET ||
596 	    cm->cmsg_len != control->m_len)
597 		return (EINVAL);
598 	oldfds = (cm->cmsg_len - sizeof (*cm)) / sizeof (int);
599 	rp = (struct file **)(cm + 1);
600 	for (i = 0; i < oldfds; i++) {
601 		fd = *(int *)rp++;
602 		if ((unsigned)fd >= fdp->fd_nfiles || OFILE(fdp, fd) == NULL)
603 			return (EBADF);
604 	}
605 	rp = (struct file **)(cm + 1);
606 	for (i = 0; i < oldfds; i++) {
607 		fp = OFILE(fdp, *(int *)rp);
608 		*rp++ = fp;
609 		fp->f_count++;
610 		fp->f_msgcount++;
611 		unp_rights++;
612 	}
613 	return (0);
614 }
615 
616 int	unp_defer, unp_gcing;
617 int	unp_mark();
618 extern	struct domain unixdomain;
619 
620 unp_gc()
621 {
622 	register struct file *fp;
623 	register struct socket *so;
624 
625 	if (unp_gcing)
626 		return;
627 	unp_gcing = 1;
628 restart:
629 	unp_defer = 0;
630 	for (fp = file; fp < fileNFILE; fp++)
631 		fp->f_flag &= ~(FMARK|FDEFER);
632 	do {
633 		for (fp = file; fp < fileNFILE; fp++) {
634 			if (fp->f_count == 0)
635 				continue;
636 			if (fp->f_flag & FDEFER) {
637 				fp->f_flag &= ~FDEFER;
638 				unp_defer--;
639 			} else {
640 				if (fp->f_flag & FMARK)
641 					continue;
642 				if (fp->f_count == fp->f_msgcount)
643 					continue;
644 				fp->f_flag |= FMARK;
645 			}
646 			if (fp->f_type != DTYPE_SOCKET ||
647 			    (so = (struct socket *)fp->f_data) == 0)
648 				continue;
649 			if (so->so_proto->pr_domain != &unixdomain ||
650 			    (so->so_proto->pr_flags&PR_RIGHTS) == 0)
651 				continue;
652 #ifdef notdef
653 			if (so->so_rcv.sb_flags & SB_LOCK) {
654 				/*
655 				 * This is problematical; it's not clear
656 				 * we need to wait for the sockbuf to be
657 				 * unlocked (on a uniprocessor, at least),
658 				 * and it's also not clear what to do
659 				 * if sbwait returns an error due to receipt
660 				 * of a signal.  If sbwait does return
661 				 * an error, we'll go into an infinite
662 				 * loop.  Delete all of this for now.
663 				 */
664 				(void) sbwait(&so->so_rcv);
665 				goto restart;
666 			}
667 #endif
668 			unp_scan(so->so_rcv.sb_mb, unp_mark);
669 		}
670 	} while (unp_defer);
671 	for (fp = file; fp < fileNFILE; fp++) {
672 		if (fp->f_count == 0)
673 			continue;
674 		if (fp->f_count == fp->f_msgcount && (fp->f_flag & FMARK) == 0)
675 			while (fp->f_msgcount)
676 				unp_discard(fp);
677 	}
678 	unp_gcing = 0;
679 }
680 
681 unp_dispose(m)
682 	struct mbuf *m;
683 {
684 	int unp_discard();
685 
686 	if (m)
687 		unp_scan(m, unp_discard);
688 }
689 
690 unp_scan(m0, op)
691 	register struct mbuf *m0;
692 	int (*op)();
693 {
694 	register struct mbuf *m;
695 	register struct file **rp;
696 	register struct cmsghdr *cm;
697 	register int i;
698 	int qfds;
699 
700 	while (m0) {
701 		for (m = m0; m; m = m->m_next)
702 			if (m->m_type == MT_CONTROL &&
703 			    m->m_len >= sizeof(*cm)) {
704 				cm = mtod(m, struct cmsghdr *);
705 				if (cm->cmsg_level != SOL_SOCKET ||
706 				    cm->cmsg_type != SCM_RIGHTS)
707 					continue;
708 				qfds = (cm->cmsg_len - sizeof *cm)
709 						/ sizeof (struct file *);
710 				rp = (struct file **)(cm + 1);
711 				for (i = 0; i < qfds; i++)
712 					(*op)(*rp++);
713 				break;		/* XXX, but saves time */
714 			}
715 		m0 = m0->m_act;
716 	}
717 }
718 
719 unp_mark(fp)
720 	struct file *fp;
721 {
722 
723 	if (fp->f_flag & FMARK)
724 		return;
725 	unp_defer++;
726 	fp->f_flag |= (FMARK|FDEFER);
727 }
728 
729 unp_discard(fp)
730 	struct file *fp;
731 {
732 
733 	fp->f_msgcount--;
734 	unp_rights--;
735 	(void) closef(fp);
736 }
737