1/*
2 * Copyright (c) 1982, 1986, 1989, 1991, 1993
3 * The Regents of the University of California. All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 3. All advertising materials mentioning features or use of this software
14 * must display the following acknowledgement:
15 * This product includes software developed by the University of
16 * California, Berkeley and its contributors.
17 * 4. Neither the name of the University nor the names of its contributors
18 * may be used to endorse or promote products derived from this software
19 * without specific prior written permission.
20 *
21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * SUCH DAMAGE.
32 *
33 * From: @(#)uipc_usrreq.c 8.3 (Berkeley) 1/4/94
34 */
35
36#include <sys/cdefs.h>
37__FBSDID("$FreeBSD: head/sys/kern/uipc_usrreq.c 126103 2004-02-22 01:11:53Z cperciva $");
38
39#include "opt_mac.h"
40
41#include <sys/param.h>
42#include <sys/domain.h>
43#include <sys/fcntl.h>
44#include <sys/malloc.h> /* XXX must be before <sys/file.h> */
45#include <sys/file.h>
46#include <sys/filedesc.h>
47#include <sys/jail.h>
48#include <sys/kernel.h>
49#include <sys/lock.h>
50#include <sys/mac.h>
51#include <sys/mbuf.h>
52#include <sys/mutex.h>
53#include <sys/namei.h>
54#include <sys/proc.h>
55#include <sys/protosw.h>
56#include <sys/resourcevar.h>
57#include <sys/socket.h>
58#include <sys/socketvar.h>
59#include <sys/signalvar.h>
60#include <sys/stat.h>
61#include <sys/sx.h>
62#include <sys/sysctl.h>
63#include <sys/systm.h>
64#include <sys/un.h>
65#include <sys/unpcb.h>
66#include <sys/vnode.h>
67
68#include <vm/uma.h>
69
70static uma_zone_t unp_zone;
71static unp_gen_t unp_gencnt;
72static u_int unp_count;
73
74static struct unp_head unp_shead, unp_dhead;
75
76/*
77 * Unix communications domain.
78 *
79 * TODO:
80 * SEQPACKET, RDM
81 * rethink name space problems
82 * need a proper out-of-band
83 * lock pushdown
84 */
85static struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL };
86static ino_t unp_ino; /* prototype for fake inode numbers */
87
88static int unp_attach(struct socket *);
89static void unp_detach(struct unpcb *);
90static int unp_bind(struct unpcb *,struct sockaddr *, struct thread *);
91static int unp_connect(struct socket *,struct sockaddr *, struct thread *);
92static void unp_disconnect(struct unpcb *);
93static void unp_shutdown(struct unpcb *);
94static void unp_drop(struct unpcb *, int);
95static void unp_gc(void);
96static void unp_scan(struct mbuf *, void (*)(struct file *));
97static void unp_mark(struct file *);
98static void unp_discard(struct file *);
99static void unp_freerights(struct file **, int);
100static int unp_internalize(struct mbuf **, struct thread *);
101static int unp_listen(struct unpcb *, struct thread *);
102
103static int
104uipc_abort(struct socket *so)
105{
106 struct unpcb *unp = sotounpcb(so);
107
108 if (unp == 0)
109 return (EINVAL);
110 unp_drop(unp, ECONNABORTED);
111 unp_detach(unp);
112 sotryfree(so);
113 return (0);
114}
115
116static int
117uipc_accept(struct socket *so, struct sockaddr **nam)
118{
119 struct unpcb *unp = sotounpcb(so);
120
121 if (unp == 0)
122 return (EINVAL);
123
124 /*
125 * Pass back name of connected socket,
126 * if it was bound and we are still connected
127 * (our peer may have closed already!).
128 */
129 if (unp->unp_conn && unp->unp_conn->unp_addr) {
130 *nam = dup_sockaddr((struct sockaddr *)unp->unp_conn->unp_addr,
131 1);
132 } else {
133 *nam = dup_sockaddr((struct sockaddr *)&sun_noname, 1);
134 }
135 return (0);
136}
137
138static int
139uipc_attach(struct socket *so, int proto, struct thread *td)
140{
141 struct unpcb *unp = sotounpcb(so);
142
143 if (unp != 0)
144 return (EISCONN);
145 return (unp_attach(so));
146}
147
148static int
149uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
150{
151 struct unpcb *unp = sotounpcb(so);
152
153 if (unp == 0)
154 return (EINVAL);
155
156 return (unp_bind(unp, nam, td));
157}
158
159static int
160uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
161{
162 struct unpcb *unp = sotounpcb(so);
163
164 if (unp == 0)
165 return (EINVAL);
166 return (unp_connect(so, nam, curthread));
167}
168
169static int
170uipc_connect2(struct socket *so1, struct socket *so2)
171{
172 struct unpcb *unp = sotounpcb(so1);
173
174 if (unp == 0)
175 return (EINVAL);
176
177 return (unp_connect2(so1, so2));
178}
179
180/* control is EOPNOTSUPP */
181
182static int
183uipc_detach(struct socket *so)
184{
185 struct unpcb *unp = sotounpcb(so);
186
187 if (unp == 0)
188 return (EINVAL);
189
190 unp_detach(unp);
191 return (0);
192}
193
194static int
195uipc_disconnect(struct socket *so)
196{
197 struct unpcb *unp = sotounpcb(so);
198
199 if (unp == 0)
200 return (EINVAL);
201 unp_disconnect(unp);
202 return (0);
203}
204
205static int
206uipc_listen(struct socket *so, struct thread *td)
207{
208 struct unpcb *unp = sotounpcb(so);
209
210 if (unp == 0 || unp->unp_vnode == 0)
211 return (EINVAL);
212 return (unp_listen(unp, td));
213}
214
215static int
216uipc_peeraddr(struct socket *so, struct sockaddr **nam)
217{
218 struct unpcb *unp = sotounpcb(so);
219
220 if (unp == 0)
221 return (EINVAL);
222 if (unp->unp_conn && unp->unp_conn->unp_addr)
223 *nam = dup_sockaddr((struct sockaddr *)unp->unp_conn->unp_addr,
224 1);
225 else {
226 /*
227 * XXX: It seems that this test always fails even when
228 * connection is established. So, this else clause is
229 * added as workaround to return PF_LOCAL sockaddr.
230 */
231 *nam = dup_sockaddr((struct sockaddr *)&sun_noname, 1);
232 }
233 return (0);
234}
235
236static int
237uipc_rcvd(struct socket *so, int flags)
238{
239 struct unpcb *unp = sotounpcb(so);
240 struct socket *so2;
241 u_long newhiwat;
242
243 if (unp == 0)
244 return (EINVAL);
245 switch (so->so_type) {
246 case SOCK_DGRAM:
247 panic("uipc_rcvd DGRAM?");
248 /*NOTREACHED*/
249
250 case SOCK_STREAM:
251 if (unp->unp_conn == 0)
252 break;
253 so2 = unp->unp_conn->unp_socket;
254 /*
255 * Adjust backpressure on sender
256 * and wakeup any waiting to write.
257 */
258 so2->so_snd.sb_mbmax += unp->unp_mbcnt - so->so_rcv.sb_mbcnt;
259 unp->unp_mbcnt = so->so_rcv.sb_mbcnt;
260 newhiwat = so2->so_snd.sb_hiwat + unp->unp_cc -
261 so->so_rcv.sb_cc;
262 (void)chgsbsize(so2->so_cred->cr_uidinfo, &so2->so_snd.sb_hiwat,
263 newhiwat, RLIM_INFINITY);
264 unp->unp_cc = so->so_rcv.sb_cc;
265 sowwakeup(so2);
266 break;
267
268 default:
269 panic("uipc_rcvd unknown socktype");
270 }
271 return (0);
272}
273
274/* pru_rcvoob is EOPNOTSUPP */
275
276static int
277uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
278 struct mbuf *control, struct thread *td)
279{
280 int error = 0;
281 struct unpcb *unp = sotounpcb(so);
282 struct socket *so2;
283 u_long newhiwat;
284
285 if (unp == 0) {
286 error = EINVAL;
287 goto release;
288 }
289 if (flags & PRUS_OOB) {
290 error = EOPNOTSUPP;
291 goto release;
292 }
293
294 if (control && (error = unp_internalize(&control, td)))
295 goto release;
296
297 switch (so->so_type) {
298 case SOCK_DGRAM:
299 {
300 struct sockaddr *from;
301
302 if (nam) {
303 if (unp->unp_conn) {
304 error = EISCONN;
305 break;
306 }
307 error = unp_connect(so, nam, td);
308 if (error)
309 break;
310 } else {
311 if (unp->unp_conn == 0) {
312 error = ENOTCONN;
313 break;
314 }
315 }
316 so2 = unp->unp_conn->unp_socket;
317 if (unp->unp_addr)
318 from = (struct sockaddr *)unp->unp_addr;
319 else
320 from = &sun_noname;
321 if (sbappendaddr(&so2->so_rcv, from, m, control)) {
322 sorwakeup(so2);
323 m = 0;
324 control = 0;
325 } else {
326 error = ENOBUFS;
327 }
328 if (nam)
329 unp_disconnect(unp);
330 break;
331 }
332
333 case SOCK_STREAM:
334 /* Connect if not connected yet. */
335 /*
336 * Note: A better implementation would complain
337 * if not equal to the peer's address.
338 */
339 if ((so->so_state & SS_ISCONNECTED) == 0) {
340 if (nam) {
341 error = unp_connect(so, nam, td);
342 if (error)
343 break; /* XXX */
344 } else {
345 error = ENOTCONN;
346 break;
347 }
348 }
349
350 if (so->so_state & SS_CANTSENDMORE) {
351 error = EPIPE;
352 break;
353 }
354 if (unp->unp_conn == 0)
355 panic("uipc_send connected but no connection?");
356 so2 = unp->unp_conn->unp_socket;
357 /*
358 * Send to paired receive port, and then reduce
359 * send buffer hiwater marks to maintain backpressure.
360 * Wake up readers.
361 */
362 if (control) {
363 if (sbappendcontrol(&so2->so_rcv, m, control))
364 control = 0;
365 } else {
366 sbappend(&so2->so_rcv, m);
367 }
368 so->so_snd.sb_mbmax -=
369 so2->so_rcv.sb_mbcnt - unp->unp_conn->unp_mbcnt;
370 unp->unp_conn->unp_mbcnt = so2->so_rcv.sb_mbcnt;
371 newhiwat = so->so_snd.sb_hiwat -
372 (so2->so_rcv.sb_cc - unp->unp_conn->unp_cc);
373 (void)chgsbsize(so->so_cred->cr_uidinfo, &so->so_snd.sb_hiwat,
374 newhiwat, RLIM_INFINITY);
375 unp->unp_conn->unp_cc = so2->so_rcv.sb_cc;
376 sorwakeup(so2);
377 m = 0;
378 break;
379
380 default:
381 panic("uipc_send unknown socktype");
382 }
383
384 /*
385 * SEND_EOF is equivalent to a SEND followed by
386 * a SHUTDOWN.
387 */
388 if (flags & PRUS_EOF) {
389 socantsendmore(so);
390 unp_shutdown(unp);
391 }
392
393 if (control && error != 0)
394 unp_dispose(control);
395
396release:
397 if (control)
398 m_freem(control);
399 if (m)
400 m_freem(m);
401 return (error);
402}
403
404static int
405uipc_sense(struct socket *so, struct stat *sb)
406{
407 struct unpcb *unp = sotounpcb(so);
408 struct socket *so2;
409
410 if (unp == 0)
411 return (EINVAL);
412 sb->st_blksize = so->so_snd.sb_hiwat;
413 if (so->so_type == SOCK_STREAM && unp->unp_conn != 0) {
414 so2 = unp->unp_conn->unp_socket;
415 sb->st_blksize += so2->so_rcv.sb_cc;
416 }
417 sb->st_dev = NOUDEV;
418 if (unp->unp_ino == 0)
419 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino;
420 sb->st_ino = unp->unp_ino;
421 return (0);
422}
423
424static int
425uipc_shutdown(struct socket *so)
426{
427 struct unpcb *unp = sotounpcb(so);
428
429 if (unp == 0)
430 return (EINVAL);
431 socantsendmore(so);
432 unp_shutdown(unp);
433 return (0);
434}
435
436static int
437uipc_sockaddr(struct socket *so, struct sockaddr **nam)
438{
439 struct unpcb *unp = sotounpcb(so);
440
441 if (unp == 0)
442 return (EINVAL);
443 if (unp->unp_addr)
444 *nam = dup_sockaddr((struct sockaddr *)unp->unp_addr, 1);
445 else
446 *nam = dup_sockaddr((struct sockaddr *)&sun_noname, 1);
447 return (0);
448}
449
450struct pr_usrreqs uipc_usrreqs = {
451 uipc_abort, uipc_accept, uipc_attach, uipc_bind, uipc_connect,
452 uipc_connect2, pru_control_notsupp, uipc_detach, uipc_disconnect,
453 uipc_listen, uipc_peeraddr, uipc_rcvd, pru_rcvoob_notsupp,
454 uipc_send, uipc_sense, uipc_shutdown, uipc_sockaddr,
455 sosend, soreceive, sopoll, pru_sosetlabel_null
456};
457
458int
459uipc_ctloutput(so, sopt)
460 struct socket *so;
461 struct sockopt *sopt;
462{
463 struct unpcb *unp = sotounpcb(so);
464 int error;
465
466 switch (sopt->sopt_dir) {
467 case SOPT_GET:
468 switch (sopt->sopt_name) {
469 case LOCAL_PEERCRED:
470 if (unp->unp_flags & UNP_HAVEPC)
471 error = sooptcopyout(sopt, &unp->unp_peercred,
472 sizeof(unp->unp_peercred));
473 else {
474 if (so->so_type == SOCK_STREAM)
475 error = ENOTCONN;
476 else
477 error = EINVAL;
478 }
479 break;
480 default:
481 error = EOPNOTSUPP;
482 break;
483 }
484 break;
485 case SOPT_SET:
486 default:
487 error = EOPNOTSUPP;
488 break;
489 }
490 return (error);
491}
492
493/*
494 * Both send and receive buffers are allocated PIPSIZ bytes of buffering
495 * for stream sockets, although the total for sender and receiver is
496 * actually only PIPSIZ.
497 * Datagram sockets really use the sendspace as the maximum datagram size,
498 * and don't really want to reserve the sendspace. Their recvspace should
499 * be large enough for at least one max-size datagram plus address.
500 */
501#ifndef PIPSIZ
502#define PIPSIZ 8192
503#endif
504static u_long unpst_sendspace = PIPSIZ;
505static u_long unpst_recvspace = PIPSIZ;
506static u_long unpdg_sendspace = 2*1024; /* really max datagram size */
507static u_long unpdg_recvspace = 4*1024;
508
509static int unp_rights; /* file descriptors in flight */
510
511SYSCTL_DECL(_net_local_stream);
512SYSCTL_INT(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW,
513 &unpst_sendspace, 0, "");
514SYSCTL_INT(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW,
515 &unpst_recvspace, 0, "");
516SYSCTL_DECL(_net_local_dgram);
517SYSCTL_INT(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW,
518 &unpdg_sendspace, 0, "");
519SYSCTL_INT(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW,
520 &unpdg_recvspace, 0, "");
521SYSCTL_DECL(_net_local);
522SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, "");
523
524static int
525unp_attach(so)
526 struct socket *so;
527{
528 register struct unpcb *unp;
529 int error;
530
531 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
532 switch (so->so_type) {
533
534 case SOCK_STREAM:
535 error = soreserve(so, unpst_sendspace, unpst_recvspace);
536 break;
537
538 case SOCK_DGRAM:
539 error = soreserve(so, unpdg_sendspace, unpdg_recvspace);
540 break;
541
542 default:
543 panic("unp_attach");
544 }
545 if (error)
546 return (error);
547 }
548 unp = uma_zalloc(unp_zone, M_WAITOK);
549 if (unp == NULL)
550 return (ENOBUFS);
551 bzero(unp, sizeof *unp);
552 unp->unp_gencnt = ++unp_gencnt;
553 unp_count++;
554 LIST_INIT(&unp->unp_refs);
555 unp->unp_socket = so;
556 LIST_INSERT_HEAD(so->so_type == SOCK_DGRAM ? &unp_dhead
557 : &unp_shead, unp, unp_link);
558 so->so_pcb = unp;
559 return (0);
560}
561
562static void
563unp_detach(unp)
564 register struct unpcb *unp;
565{
566 LIST_REMOVE(unp, unp_link);
567 unp->unp_gencnt = ++unp_gencnt;
568 --unp_count;
569 if (unp->unp_vnode) {
570 unp->unp_vnode->v_socket = 0;
571 vrele(unp->unp_vnode);
572 unp->unp_vnode = 0;
573 }
574 if (unp->unp_conn)
575 unp_disconnect(unp);
576 while (!LIST_EMPTY(&unp->unp_refs))
577 unp_drop(LIST_FIRST(&unp->unp_refs), ECONNRESET);
578 soisdisconnected(unp->unp_socket);
579 unp->unp_socket->so_pcb = 0;
580 if (unp_rights) {
581 /*
582 * Normally the receive buffer is flushed later,
583 * in sofree, but if our receive buffer holds references
584 * to descriptors that are now garbage, we will dispose
585 * of those descriptor references after the garbage collector
586 * gets them (resulting in a "panic: closef: count < 0").
587 */
588 sorflush(unp->unp_socket);
589 unp_gc();
590 }
591 if (unp->unp_addr)
592 FREE(unp->unp_addr, M_SONAME);
593 uma_zfree(unp_zone, unp);
594}
595
596static int
597unp_bind(unp, nam, td)
598 struct unpcb *unp;
599 struct sockaddr *nam;
600 struct thread *td;
601{
602 struct sockaddr_un *soun = (struct sockaddr_un *)nam;
603 struct vnode *vp;
604 struct mount *mp;
605 struct vattr vattr;
606 int error, namelen;
607 struct nameidata nd;
608 char *buf;
609
610 if (unp->unp_vnode != NULL)
611 return (EINVAL);
612
613 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path);
614 if (namelen <= 0)
615 return (EINVAL);
616
617 buf = malloc(namelen + 1, M_TEMP, M_WAITOK);
618 strlcpy(buf, soun->sun_path, namelen + 1);
619
620restart:
621 NDINIT(&nd, CREATE, NOFOLLOW | LOCKPARENT | SAVENAME, UIO_SYSSPACE,
622 buf, td);
623/* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
624 error = namei(&nd);
625 if (error) {
626 free(buf, M_TEMP);
627 return (error);
628 }
629 vp = nd.ni_vp;
630 if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
631 NDFREE(&nd, NDF_ONLY_PNBUF);
632 if (nd.ni_dvp == vp)
633 vrele(nd.ni_dvp);
634 else
635 vput(nd.ni_dvp);
636 if (vp != NULL) {
637 vrele(vp);
638 free(buf, M_TEMP);
639 return (EADDRINUSE);
640 }
641 error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH);
642 if (error) {
643 free(buf, M_TEMP);
644 return (error);
645 }
646 goto restart;
647 }
648 VATTR_NULL(&vattr);
649 vattr.va_type = VSOCK;
650 vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask);
651#ifdef MAC
652 error = mac_check_vnode_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd,
653 &vattr);
654#endif
655 if (error == 0) {
656 VOP_LEASE(nd.ni_dvp, td, td->td_ucred, LEASE_WRITE);
657 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
658 }
659 NDFREE(&nd, NDF_ONLY_PNBUF);
660 vput(nd.ni_dvp);
661 if (error) {
662 free(buf, M_TEMP);
663 return (error);
664 }
665 vp = nd.ni_vp;
666 vp->v_socket = unp->unp_socket;
667 unp->unp_vnode = vp;
668 unp->unp_addr = (struct sockaddr_un *)dup_sockaddr(nam, 1);
669 VOP_UNLOCK(vp, 0, td);
670 vn_finished_write(mp);
671 free(buf, M_TEMP);
672 return (0);
673}
674
675static int
676unp_connect(so, nam, td)
677 struct socket *so;
678 struct sockaddr *nam;
679 struct thread *td;
680{
681 register struct sockaddr_un *soun = (struct sockaddr_un *)nam;
682 register struct vnode *vp;
683 register struct socket *so2, *so3;
684 struct unpcb *unp, *unp2, *unp3;
685 int error, len;
686 struct nameidata nd;
687 char buf[SOCK_MAXADDRLEN];
688
689 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
690 if (len <= 0)
691 return (EINVAL);
692 strlcpy(buf, soun->sun_path, len + 1);
693
694 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, buf, td);
695 error = namei(&nd);
696 if (error)
697 return (error);
698 vp = nd.ni_vp;
699 NDFREE(&nd, NDF_ONLY_PNBUF);
700 if (vp->v_type != VSOCK) {
701 error = ENOTSOCK;
702 goto bad;
703 }
704 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td);
705 if (error)
706 goto bad;
707 so2 = vp->v_socket;
708 if (so2 == 0) {
709 error = ECONNREFUSED;
710 goto bad;
711 }
712 if (so->so_type != so2->so_type) {
713 error = EPROTOTYPE;
714 goto bad;
715 }
716 if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
717 if ((so2->so_options & SO_ACCEPTCONN) == 0 ||
718 (so3 = sonewconn(so2, 0)) == 0) {
719 error = ECONNREFUSED;
720 goto bad;
721 }
722 unp = sotounpcb(so);
723 unp2 = sotounpcb(so2);
724 unp3 = sotounpcb(so3);
725 if (unp2->unp_addr)
726 unp3->unp_addr = (struct sockaddr_un *)
727 dup_sockaddr((struct sockaddr *)
728 unp2->unp_addr, 1);
729
730 /*
731 * unp_peercred management:
732 *
733 * The connecter's (client's) credentials are copied
734 * from its process structure at the time of connect()
735 * (which is now).
736 */
737 cru2x(td->td_ucred, &unp3->unp_peercred);
738 unp3->unp_flags |= UNP_HAVEPC;
739 /*
740 * The receiver's (server's) credentials are copied
741 * from the unp_peercred member of socket on which the
742 * former called listen(); unp_listen() cached that
743 * process's credentials at that time so we can use
744 * them now.
745 */
746 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED,
747 ("unp_connect: listener without cached peercred"));
748 memcpy(&unp->unp_peercred, &unp2->unp_peercred,
749 sizeof(unp->unp_peercred));
750 unp->unp_flags |= UNP_HAVEPC;
751#ifdef MAC
752 mac_set_socket_peer_from_socket(so, so3);
753 mac_set_socket_peer_from_socket(so3, so);
754#endif
755
756 so2 = so3;
757 }
758 error = unp_connect2(so, so2);
759bad:
760 vput(vp);
761 return (error);
762}
763
764int
765unp_connect2(so, so2)
766 register struct socket *so;
767 register struct socket *so2;
768{
769 register struct unpcb *unp = sotounpcb(so);
770 register struct unpcb *unp2;
771
772 if (so2->so_type != so->so_type)
773 return (EPROTOTYPE);
774 unp2 = sotounpcb(so2);
775 unp->unp_conn = unp2;
776 switch (so->so_type) {
777
778 case SOCK_DGRAM:
779 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
780 soisconnected(so);
781 break;
782
783 case SOCK_STREAM:
784 unp2->unp_conn = unp;
785 soisconnected(so);
786 soisconnected(so2);
787 break;
788
789 default:
790 panic("unp_connect2");
791 }
792 return (0);
793}
794
795static void
796unp_disconnect(unp)
797 struct unpcb *unp;
798{
799 register struct unpcb *unp2 = unp->unp_conn;
800
801 if (unp2 == 0)
802 return;
803 unp->unp_conn = 0;
804 switch (unp->unp_socket->so_type) {
805
806 case SOCK_DGRAM:
807 LIST_REMOVE(unp, unp_reflink);
808 unp->unp_socket->so_state &= ~SS_ISCONNECTED;
809 break;
810
811 case SOCK_STREAM:
812 soisdisconnected(unp->unp_socket);
813 unp2->unp_conn = 0;
814 soisdisconnected(unp2->unp_socket);
815 break;
816 }
817}
818
819#ifdef notdef
820void
821unp_abort(unp)
822 struct unpcb *unp;
823{
824
825 unp_detach(unp);
826}
827#endif
828
829static int
830unp_pcblist(SYSCTL_HANDLER_ARGS)
831{
832 int error, i, n;
833 struct unpcb *unp, **unp_list;
834 unp_gen_t gencnt;
835 struct xunpgen *xug;
836 struct unp_head *head;
837 struct xunpcb *xu;
838
839 head = ((intptr_t)arg1 == SOCK_DGRAM ? &unp_dhead : &unp_shead);
840
841 /*
842 * The process of preparing the PCB list is too time-consuming and
843 * resource-intensive to repeat twice on every request.
844 */
845 if (req->oldptr == 0) {
846 n = unp_count;
847 req->oldidx = 2 * (sizeof *xug)
848 + (n + n/8) * sizeof(struct xunpcb);
849 return (0);
850 }
851
852 if (req->newptr != 0)
853 return (EPERM);
854
855 /*
856 * OK, now we're committed to doing something.
857 */
858 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK);
859 gencnt = unp_gencnt;
860 n = unp_count;
861
862 xug->xug_len = sizeof *xug;
863 xug->xug_count = n;
864 xug->xug_gen = gencnt;
865 xug->xug_sogen = so_gencnt;
866 error = SYSCTL_OUT(req, xug, sizeof *xug);
867 if (error) {
868 free(xug, M_TEMP);
869 return (error);
870 }
871
872 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
873
874 for (unp = LIST_FIRST(head), i = 0; unp && i < n;
875 unp = LIST_NEXT(unp, unp_link)) {
876 if (unp->unp_gencnt <= gencnt) {
877 if (cr_cansee(req->td->td_ucred,
878 unp->unp_socket->so_cred))
879 continue;
880 unp_list[i++] = unp;
881 }
882 }
883 n = i; /* in case we lost some during malloc */
884
885 error = 0;
886 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK);
887 for (i = 0; i < n; i++) {
888 unp = unp_list[i];
889 if (unp->unp_gencnt <= gencnt) {
890 xu->xu_len = sizeof *xu;
891 xu->xu_unpp = unp;
892 /*
893 * XXX - need more locking here to protect against
894 * connect/disconnect races for SMP.
895 */
896 if (unp->unp_addr)
897 bcopy(unp->unp_addr, &xu->xu_addr,
898 unp->unp_addr->sun_len);
899 if (unp->unp_conn && unp->unp_conn->unp_addr)
900 bcopy(unp->unp_conn->unp_addr,
901 &xu->xu_caddr,
902 unp->unp_conn->unp_addr->sun_len);
903 bcopy(unp, &xu->xu_unp, sizeof *unp);
904 sotoxsocket(unp->unp_socket, &xu->xu_socket);
905 error = SYSCTL_OUT(req, xu, sizeof *xu);
906 }
907 }
908 free(xu, M_TEMP);
909 if (!error) {
910 /*
911 * Give the user an updated idea of our state.
912 * If the generation differs from what we told
913 * her before, she knows that something happened
914 * while we were processing this request, and it
915 * might be necessary to retry.
916 */
917 xug->xug_gen = unp_gencnt;
918 xug->xug_sogen = so_gencnt;
919 xug->xug_count = unp_count;
920 error = SYSCTL_OUT(req, xug, sizeof *xug);
921 }
922 free(unp_list, M_TEMP);
923 free(xug, M_TEMP);
924 return (error);
925}
926
927SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLFLAG_RD,
928 (caddr_t)(long)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
929 "List of active local datagram sockets");
930SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLFLAG_RD,
931 (caddr_t)(long)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
932 "List of active local stream sockets");
933
934static void
935unp_shutdown(unp)
936 struct unpcb *unp;
937{
938 struct socket *so;
939
940 if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn &&
941 (so = unp->unp_conn->unp_socket))
942 socantrcvmore(so);
943}
944
945static void
946unp_drop(unp, errno)
947 struct unpcb *unp;
948 int errno;
949{
950 struct socket *so = unp->unp_socket;
951
952 so->so_error = errno;
953 unp_disconnect(unp);
954}
955
956#ifdef notdef
957void
958unp_drain()
959{
960
961}
962#endif
963
964static void
965unp_freerights(rp, fdcount)
966 struct file **rp;
967 int fdcount;
968{
969 int i;
970 struct file *fp;
971
972 for (i = 0; i < fdcount; i++) {
973 fp = *rp;
974 /*
975 * zero the pointer before calling
976 * unp_discard since it may end up
977 * in unp_gc()..
978 */
979 *rp++ = 0;
980 unp_discard(fp);
981 }
982}
983
984int
985unp_externalize(control, controlp)
986 struct mbuf *control, **controlp;
987{
988 struct thread *td = curthread; /* XXX */
989 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
990 int i;
991 int *fdp;
992 struct file **rp;
993 struct file *fp;
994 void *data;
995 socklen_t clen = control->m_len, datalen;
996 int error, newfds;
997 int f;
998 u_int newlen;
999
1000 error = 0;
1001 if (controlp != NULL) /* controlp == NULL => free control messages */
1002 *controlp = NULL;
1003
1004 while (cm != NULL) {
1005 if (sizeof(*cm) > clen || cm->cmsg_len > clen) {
1006 error = EINVAL;
1007 break;
1008 }
1009
1010 data = CMSG_DATA(cm);
1011 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1012
1013 if (cm->cmsg_level == SOL_SOCKET
1014 && cm->cmsg_type == SCM_RIGHTS) {
1015 newfds = datalen / sizeof(struct file *);
1016 rp = data;
1017
1018 /* If we're not outputting the descriptors free them. */
1019 if (error || controlp == NULL) {
1020 unp_freerights(rp, newfds);
1021 goto next;
1022 }
1023 FILEDESC_LOCK(td->td_proc->p_fd);
1024 /* if the new FD's will not fit free them. */
1025 if (!fdavail(td, newfds)) {
1026 FILEDESC_UNLOCK(td->td_proc->p_fd);
1027 error = EMSGSIZE;
1028 unp_freerights(rp, newfds);
1029 goto next;
1030 }
1031 /*
1032 * now change each pointer to an fd in the global
1033 * table to an integer that is the index to the
1034 * local fd table entry that we set up to point
1035 * to the global one we are transferring.
1036 */
1037 newlen = newfds * sizeof(int);
1038 *controlp = sbcreatecontrol(NULL, newlen,
1039 SCM_RIGHTS, SOL_SOCKET);
1040 if (*controlp == NULL) {
1041 FILEDESC_UNLOCK(td->td_proc->p_fd);
1042 error = E2BIG;
1043 unp_freerights(rp, newfds);
1044 goto next;
1045 }
1046
1047 fdp = (int *)
1048 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1049 for (i = 0; i < newfds; i++) {
1050 if (fdalloc(td, 0, &f))
1051 panic("unp_externalize fdalloc failed");
1052 fp = *rp++;
1053 td->td_proc->p_fd->fd_ofiles[f] = fp;
1054 FILE_LOCK(fp);
1055 fp->f_msgcount--;
1056 FILE_UNLOCK(fp);
1057 unp_rights--;
1058 *fdp++ = f;
1059 }
1060 FILEDESC_UNLOCK(td->td_proc->p_fd);
1061 } else { /* We can just copy anything else across */
1062 if (error || controlp == NULL)
1063 goto next;
1064 *controlp = sbcreatecontrol(NULL, datalen,
1065 cm->cmsg_type, cm->cmsg_level);
1066 if (*controlp == NULL) {
1067 error = ENOBUFS;
1068 goto next;
1069 }
1070 bcopy(data,
1071 CMSG_DATA(mtod(*controlp, struct cmsghdr *)),
1072 datalen);
1073 }
1074
1075 controlp = &(*controlp)->m_next;
1076
1077next:
1078 if (CMSG_SPACE(datalen) < clen) {
1079 clen -= CMSG_SPACE(datalen);
1080 cm = (struct cmsghdr *)
1081 ((caddr_t)cm + CMSG_SPACE(datalen));
1082 } else {
1083 clen = 0;
1084 cm = NULL;
1085 }
1086 }
1087
1088 m_freem(control);
1089
1090 return (error);
1091}
1092
1093void
1094unp_init(void)
1095{
1096 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL,
1097 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
1098 if (unp_zone == 0)
1099 panic("unp_init");
1100 uma_zone_set_max(unp_zone, nmbclusters);
1101 LIST_INIT(&unp_dhead);
1102 LIST_INIT(&unp_shead);
1103}
1104
1105static int
1106unp_internalize(controlp, td)
1107 struct mbuf **controlp;
1108 struct thread *td;
1109{
1110 struct mbuf *control = *controlp;
1111 struct proc *p = td->td_proc;
1112 struct filedesc *fdescp = p->p_fd;
1113 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1114 struct cmsgcred *cmcred;
1115 struct file **rp;
1116 struct file *fp;
1117 struct timeval *tv;
1118 int i, fd, *fdp;
1119 void *data;
1120 socklen_t clen = control->m_len, datalen;
1121 int error, oldfds;
1122 u_int newlen;
1123
1124 error = 0;
1125 *controlp = NULL;
1126
1127 while (cm != NULL) {
1128 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET
1129 || cm->cmsg_len > clen) {
1130 error = EINVAL;
1131 goto out;
1132 }
1133
1134 data = CMSG_DATA(cm);
1135 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1136
1137 switch (cm->cmsg_type) {
1138 /*
1139 * Fill in credential information.
1140 */
1141 case SCM_CREDS:
1142 *controlp = sbcreatecontrol(NULL, sizeof(*cmcred),
1143 SCM_CREDS, SOL_SOCKET);
1144 if (*controlp == NULL) {
1145 error = ENOBUFS;
1146 goto out;
1147 }
1148
1149 cmcred = (struct cmsgcred *)
1150 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1151 cmcred->cmcred_pid = p->p_pid;
1152 cmcred->cmcred_uid = td->td_ucred->cr_ruid;
1153 cmcred->cmcred_gid = td->td_ucred->cr_rgid;
1154 cmcred->cmcred_euid = td->td_ucred->cr_uid;
1155 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups,
1156 CMGROUP_MAX);
1157 for (i = 0; i < cmcred->cmcred_ngroups; i++)
1158 cmcred->cmcred_groups[i] =
1159 td->td_ucred->cr_groups[i];
1160 break;
1161
1162 case SCM_RIGHTS:
1163 oldfds = datalen / sizeof (int);
1164 /*
1165 * check that all the FDs passed in refer to legal files
1166 * If not, reject the entire operation.
1167 */
1168 fdp = data;
1169 FILEDESC_LOCK(fdescp);
1170 for (i = 0; i < oldfds; i++) {
1171 fd = *fdp++;
1172 if ((unsigned)fd >= fdescp->fd_nfiles ||
1173 fdescp->fd_ofiles[fd] == NULL) {
1174 FILEDESC_UNLOCK(fdescp);
1175 error = EBADF;
1176 goto out;
1177 }
1178 fp = fdescp->fd_ofiles[fd];
1179 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) {
1180 FILEDESC_UNLOCK(fdescp);
1181 error = EOPNOTSUPP;
1182 goto out;
1183 }
1184
1185 }
1186 /*
1187 * Now replace the integer FDs with pointers to
1188 * the associated global file table entry..
1189 */
1190 newlen = oldfds * sizeof(struct file *);
1191 *controlp = sbcreatecontrol(NULL, newlen,
1192 SCM_RIGHTS, SOL_SOCKET);
1193 if (*controlp == NULL) {
1194 FILEDESC_UNLOCK(fdescp);
1195 error = E2BIG;
1196 goto out;
1197 }
1198
1199 fdp = data;
1200 rp = (struct file **)
1201 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1202 for (i = 0; i < oldfds; i++) {
1203 fp = fdescp->fd_ofiles[*fdp++];
1204 *rp++ = fp;
1205 FILE_LOCK(fp);
1206 fp->f_count++;
1207 fp->f_msgcount++;
1208 FILE_UNLOCK(fp);
1209 unp_rights++;
1210 }
1211 FILEDESC_UNLOCK(fdescp);
1212 break;
1213
1214 case SCM_TIMESTAMP:
1215 *controlp = sbcreatecontrol(NULL, sizeof(*tv),
1216 SCM_TIMESTAMP, SOL_SOCKET);
1217 if (*controlp == NULL) {
1218 error = ENOBUFS;
1219 goto out;
1220 }
1221 tv = (struct timeval *)
1222 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1223 microtime(tv);
1224 break;
1225
1226 default:
1227 error = EINVAL;
1228 goto out;
1229 }
1230
1231 controlp = &(*controlp)->m_next;
1232
1233 if (CMSG_SPACE(datalen) < clen) {
1234 clen -= CMSG_SPACE(datalen);
1235 cm = (struct cmsghdr *)
1236 ((caddr_t)cm + CMSG_SPACE(datalen));
1237 } else {
1238 clen = 0;
1239 cm = NULL;
1240 }
1241 }
1242
1243out:
1244 m_freem(control);
1245
1246 return (error);
1247}
1248
1249static int unp_defer, unp_gcing;
1250
1251static void
1252unp_gc()
1253{
1254 register struct file *fp, *nextfp;
1255 register struct socket *so;
1256 struct file **extra_ref, **fpp;
1257 int nunref, i;
1258
1259 if (unp_gcing)
1260 return;
1261 unp_gcing = 1;
1262 unp_defer = 0;
1263 /*
1264 * before going through all this, set all FDs to
1265 * be NOT defered and NOT externally accessible
1266 */
1267 sx_slock(&filelist_lock);
1268 LIST_FOREACH(fp, &filehead, f_list)
1269 fp->f_gcflag &= ~(FMARK|FDEFER);
1270 do {
1271 LIST_FOREACH(fp, &filehead, f_list) {
1272 FILE_LOCK(fp);
1273 /*
1274 * If the file is not open, skip it
1275 */
1276 if (fp->f_count == 0) {
1277 FILE_UNLOCK(fp);
1278 continue;
1279 }
1280 /*
1281 * If we already marked it as 'defer' in a
1282 * previous pass, then try process it this time
1283 * and un-mark it
1284 */
1285 if (fp->f_gcflag & FDEFER) {
1286 fp->f_gcflag &= ~FDEFER;
1287 unp_defer--;
1288 } else {
1289 /*
1290 * if it's not defered, then check if it's
1291 * already marked.. if so skip it
1292 */
1293 if (fp->f_gcflag & FMARK) {
1294 FILE_UNLOCK(fp);
1295 continue;
1296 }
1297 /*
1298 * If all references are from messages
1299 * in transit, then skip it. it's not
1300 * externally accessible.
1301 */
1302 if (fp->f_count == fp->f_msgcount) {
1303 FILE_UNLOCK(fp);
1304 continue;
1305 }
1306 /*
1307 * If it got this far then it must be
1308 * externally accessible.
1309 */
1310 fp->f_gcflag |= FMARK;
1311 }
1312 /*
1313 * either it was defered, or it is externally
1314 * accessible and not already marked so.
1315 * Now check if it is possibly one of OUR sockets.
1316 */
1317 if (fp->f_type != DTYPE_SOCKET ||
1318 (so = fp->f_data) == NULL) {
1319 FILE_UNLOCK(fp);
1320 continue;
1321 }
1322 FILE_UNLOCK(fp);
1323 if (so->so_proto->pr_domain != &localdomain ||
1324 (so->so_proto->pr_flags&PR_RIGHTS) == 0)
1325 continue;
1326#ifdef notdef
1327 if (so->so_rcv.sb_flags & SB_LOCK) {
1328 /*
1329 * This is problematical; it's not clear
1330 * we need to wait for the sockbuf to be
1331 * unlocked (on a uniprocessor, at least),
1332 * and it's also not clear what to do
1333 * if sbwait returns an error due to receipt
1334 * of a signal. If sbwait does return
1335 * an error, we'll go into an infinite
1336 * loop. Delete all of this for now.
1337 */
1338 (void) sbwait(&so->so_rcv);
1339 goto restart;
1340 }
1341#endif
1342 /*
1343 * So, Ok, it's one of our sockets and it IS externally
1344 * accessible (or was defered). Now we look
1345 * to see if we hold any file descriptors in its
1346 * message buffers. Follow those links and mark them
1347 * as accessible too.
1348 */
1349 unp_scan(so->so_rcv.sb_mb, unp_mark);
1350 }
1351 } while (unp_defer);
1352 sx_sunlock(&filelist_lock);
1353 /*
1354 * We grab an extra reference to each of the file table entries
1355 * that are not otherwise accessible and then free the rights
1356 * that are stored in messages on them.
1357 *
1358 * The bug in the orginal code is a little tricky, so I'll describe
1359 * what's wrong with it here.
1360 *
1361 * It is incorrect to simply unp_discard each entry for f_msgcount
1362 * times -- consider the case of sockets A and B that contain
1363 * references to each other. On a last close of some other socket,
1364 * we trigger a gc since the number of outstanding rights (unp_rights)
1365 * is non-zero. If during the sweep phase the gc code un_discards,
1366 * we end up doing a (full) closef on the descriptor. A closef on A
1367 * results in the following chain. Closef calls soo_close, which
1368 * calls soclose. Soclose calls first (through the switch
1369 * uipc_usrreq) unp_detach, which re-invokes unp_gc. Unp_gc simply
1370 * returns because the previous instance had set unp_gcing, and
1371 * we return all the way back to soclose, which marks the socket
1372 * with SS_NOFDREF, and then calls sofree. Sofree calls sorflush
1373 * to free up the rights that are queued in messages on the socket A,
1374 * i.e., the reference on B. The sorflush calls via the dom_dispose
1375 * switch unp_dispose, which unp_scans with unp_discard. This second
1376 * instance of unp_discard just calls closef on B.
1377 *
1378 * Well, a similar chain occurs on B, resulting in a sorflush on B,
1379 * which results in another closef on A. Unfortunately, A is already
1380 * being closed, and the descriptor has already been marked with
1381 * SS_NOFDREF, and soclose panics at this point.
1382 *
1383 * Here, we first take an extra reference to each inaccessible
1384 * descriptor. Then, we call sorflush ourself, since we know
1385 * it is a Unix domain socket anyhow. After we destroy all the
1386 * rights carried in messages, we do a last closef to get rid
1387 * of our extra reference. This is the last close, and the
1388 * unp_detach etc will shut down the socket.
1389 *
1390 * 91/09/19, bsy@cs.cmu.edu
1391 */
1392 extra_ref = malloc(nfiles * sizeof(struct file *), M_TEMP, M_WAITOK);
1393 sx_slock(&filelist_lock);
1394 for (nunref = 0, fp = LIST_FIRST(&filehead), fpp = extra_ref; fp != 0;
1395 fp = nextfp) {
1396 nextfp = LIST_NEXT(fp, f_list);
1397 FILE_LOCK(fp);
1398 /*
1399 * If it's not open, skip it
1400 */
1401 if (fp->f_count == 0) {
1402 FILE_UNLOCK(fp);
1403 continue;
1404 }
1405 /*
1406 * If all refs are from msgs, and it's not marked accessible
1407 * then it must be referenced from some unreachable cycle
1408 * of (shut-down) FDs, so include it in our
1409 * list of FDs to remove
1410 */
1411 if (fp->f_count == fp->f_msgcount && !(fp->f_gcflag & FMARK)) {
1412 *fpp++ = fp;
1413 nunref++;
1414 fp->f_count++;
1415 }
1416 FILE_UNLOCK(fp);
1417 }
1418 sx_sunlock(&filelist_lock);
1419 /*
1420 * for each FD on our hit list, do the following two things
1421 */
1422 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) {
1423 struct file *tfp = *fpp;
1424 FILE_LOCK(tfp);
1425 if (tfp->f_type == DTYPE_SOCKET &&
1426 tfp->f_data != NULL) {
1427 FILE_UNLOCK(tfp);
1428 sorflush(tfp->f_data);
1429 } else {
1430 FILE_UNLOCK(tfp);
1431 }
1432 }
1433 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp)
1434 closef(*fpp, (struct thread *) NULL);
1435 free(extra_ref, M_TEMP);
1436 unp_gcing = 0;
1437}
1438
1439void
1440unp_dispose(m)
1441 struct mbuf *m;
1442{
1443
1444 if (m)
1445 unp_scan(m, unp_discard);
1446}
1447
1448static int
1449unp_listen(unp, td)
1450 struct unpcb *unp;
1451 struct thread *td;
1452{
1453
1454 cru2x(td->td_ucred, &unp->unp_peercred);
1455 unp->unp_flags |= UNP_HAVEPCCACHED;
1456 return (0);
1457}
1458
1459static void
1460unp_scan(m0, op)
1461 register struct mbuf *m0;
1462 void (*op)(struct file *);
1463{
1464 struct mbuf *m;
1465 struct file **rp;
1466 struct cmsghdr *cm;
1467 void *data;
1468 int i;
1469 socklen_t clen, datalen;
1470 int qfds;
1471
1472 while (m0) {
1473 for (m = m0; m; m = m->m_next) {
1474 if (m->m_type != MT_CONTROL)
1475 continue;
1476
1477 cm = mtod(m, struct cmsghdr *);
1478 clen = m->m_len;
1479
1480 while (cm != NULL) {
1481 if (sizeof(*cm) > clen || cm->cmsg_len > clen)
1482 break;
1483
1484 data = CMSG_DATA(cm);
1485 datalen = (caddr_t)cm + cm->cmsg_len
1486 - (caddr_t)data;
1487
1488 if (cm->cmsg_level == SOL_SOCKET &&
1489 cm->cmsg_type == SCM_RIGHTS) {
1490 qfds = datalen / sizeof (struct file *);
1491 rp = data;
1492 for (i = 0; i < qfds; i++)
1493 (*op)(*rp++);
1494 }
1495
1496 if (CMSG_SPACE(datalen) < clen) {
1497 clen -= CMSG_SPACE(datalen);
1498 cm = (struct cmsghdr *)
1499 ((caddr_t)cm + CMSG_SPACE(datalen));
1500 } else {
1501 clen = 0;
1502 cm = NULL;
1503 }
1504 }
1505 }
1506 m0 = m0->m_act;
1507 }
1508}
1509
1510static void
1511unp_mark(fp)
1512 struct file *fp;
1513{
1514 if (fp->f_gcflag & FMARK)
1515 return;
1516 unp_defer++;
1517 fp->f_gcflag |= (FMARK|FDEFER);
1518}
1519
1520static void
1521unp_discard(fp)
1522 struct file *fp;
1523{
1524 FILE_LOCK(fp);
1525 fp->f_msgcount--;
1526 unp_rights--;
1527 FILE_UNLOCK(fp);
1528 (void) closef(fp, (struct thread *)NULL);
1529}
2 * Copyright (c) 1982, 1986, 1989, 1991, 1993
3 * The Regents of the University of California. All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 3. All advertising materials mentioning features or use of this software
14 * must display the following acknowledgement:
15 * This product includes software developed by the University of
16 * California, Berkeley and its contributors.
17 * 4. Neither the name of the University nor the names of its contributors
18 * may be used to endorse or promote products derived from this software
19 * without specific prior written permission.
20 *
21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * SUCH DAMAGE.
32 *
33 * From: @(#)uipc_usrreq.c 8.3 (Berkeley) 1/4/94
34 */
35
36#include <sys/cdefs.h>
37__FBSDID("$FreeBSD: head/sys/kern/uipc_usrreq.c 126103 2004-02-22 01:11:53Z cperciva $");
38
39#include "opt_mac.h"
40
41#include <sys/param.h>
42#include <sys/domain.h>
43#include <sys/fcntl.h>
44#include <sys/malloc.h> /* XXX must be before <sys/file.h> */
45#include <sys/file.h>
46#include <sys/filedesc.h>
47#include <sys/jail.h>
48#include <sys/kernel.h>
49#include <sys/lock.h>
50#include <sys/mac.h>
51#include <sys/mbuf.h>
52#include <sys/mutex.h>
53#include <sys/namei.h>
54#include <sys/proc.h>
55#include <sys/protosw.h>
56#include <sys/resourcevar.h>
57#include <sys/socket.h>
58#include <sys/socketvar.h>
59#include <sys/signalvar.h>
60#include <sys/stat.h>
61#include <sys/sx.h>
62#include <sys/sysctl.h>
63#include <sys/systm.h>
64#include <sys/un.h>
65#include <sys/unpcb.h>
66#include <sys/vnode.h>
67
68#include <vm/uma.h>
69
70static uma_zone_t unp_zone;
71static unp_gen_t unp_gencnt;
72static u_int unp_count;
73
74static struct unp_head unp_shead, unp_dhead;
75
76/*
77 * Unix communications domain.
78 *
79 * TODO:
80 * SEQPACKET, RDM
81 * rethink name space problems
82 * need a proper out-of-band
83 * lock pushdown
84 */
85static struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL };
86static ino_t unp_ino; /* prototype for fake inode numbers */
87
88static int unp_attach(struct socket *);
89static void unp_detach(struct unpcb *);
90static int unp_bind(struct unpcb *,struct sockaddr *, struct thread *);
91static int unp_connect(struct socket *,struct sockaddr *, struct thread *);
92static void unp_disconnect(struct unpcb *);
93static void unp_shutdown(struct unpcb *);
94static void unp_drop(struct unpcb *, int);
95static void unp_gc(void);
96static void unp_scan(struct mbuf *, void (*)(struct file *));
97static void unp_mark(struct file *);
98static void unp_discard(struct file *);
99static void unp_freerights(struct file **, int);
100static int unp_internalize(struct mbuf **, struct thread *);
101static int unp_listen(struct unpcb *, struct thread *);
102
103static int
104uipc_abort(struct socket *so)
105{
106 struct unpcb *unp = sotounpcb(so);
107
108 if (unp == 0)
109 return (EINVAL);
110 unp_drop(unp, ECONNABORTED);
111 unp_detach(unp);
112 sotryfree(so);
113 return (0);
114}
115
116static int
117uipc_accept(struct socket *so, struct sockaddr **nam)
118{
119 struct unpcb *unp = sotounpcb(so);
120
121 if (unp == 0)
122 return (EINVAL);
123
124 /*
125 * Pass back name of connected socket,
126 * if it was bound and we are still connected
127 * (our peer may have closed already!).
128 */
129 if (unp->unp_conn && unp->unp_conn->unp_addr) {
130 *nam = dup_sockaddr((struct sockaddr *)unp->unp_conn->unp_addr,
131 1);
132 } else {
133 *nam = dup_sockaddr((struct sockaddr *)&sun_noname, 1);
134 }
135 return (0);
136}
137
138static int
139uipc_attach(struct socket *so, int proto, struct thread *td)
140{
141 struct unpcb *unp = sotounpcb(so);
142
143 if (unp != 0)
144 return (EISCONN);
145 return (unp_attach(so));
146}
147
148static int
149uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
150{
151 struct unpcb *unp = sotounpcb(so);
152
153 if (unp == 0)
154 return (EINVAL);
155
156 return (unp_bind(unp, nam, td));
157}
158
159static int
160uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
161{
162 struct unpcb *unp = sotounpcb(so);
163
164 if (unp == 0)
165 return (EINVAL);
166 return (unp_connect(so, nam, curthread));
167}
168
169static int
170uipc_connect2(struct socket *so1, struct socket *so2)
171{
172 struct unpcb *unp = sotounpcb(so1);
173
174 if (unp == 0)
175 return (EINVAL);
176
177 return (unp_connect2(so1, so2));
178}
179
180/* control is EOPNOTSUPP */
181
182static int
183uipc_detach(struct socket *so)
184{
185 struct unpcb *unp = sotounpcb(so);
186
187 if (unp == 0)
188 return (EINVAL);
189
190 unp_detach(unp);
191 return (0);
192}
193
194static int
195uipc_disconnect(struct socket *so)
196{
197 struct unpcb *unp = sotounpcb(so);
198
199 if (unp == 0)
200 return (EINVAL);
201 unp_disconnect(unp);
202 return (0);
203}
204
205static int
206uipc_listen(struct socket *so, struct thread *td)
207{
208 struct unpcb *unp = sotounpcb(so);
209
210 if (unp == 0 || unp->unp_vnode == 0)
211 return (EINVAL);
212 return (unp_listen(unp, td));
213}
214
215static int
216uipc_peeraddr(struct socket *so, struct sockaddr **nam)
217{
218 struct unpcb *unp = sotounpcb(so);
219
220 if (unp == 0)
221 return (EINVAL);
222 if (unp->unp_conn && unp->unp_conn->unp_addr)
223 *nam = dup_sockaddr((struct sockaddr *)unp->unp_conn->unp_addr,
224 1);
225 else {
226 /*
227 * XXX: It seems that this test always fails even when
228 * connection is established. So, this else clause is
229 * added as workaround to return PF_LOCAL sockaddr.
230 */
231 *nam = dup_sockaddr((struct sockaddr *)&sun_noname, 1);
232 }
233 return (0);
234}
235
236static int
237uipc_rcvd(struct socket *so, int flags)
238{
239 struct unpcb *unp = sotounpcb(so);
240 struct socket *so2;
241 u_long newhiwat;
242
243 if (unp == 0)
244 return (EINVAL);
245 switch (so->so_type) {
246 case SOCK_DGRAM:
247 panic("uipc_rcvd DGRAM?");
248 /*NOTREACHED*/
249
250 case SOCK_STREAM:
251 if (unp->unp_conn == 0)
252 break;
253 so2 = unp->unp_conn->unp_socket;
254 /*
255 * Adjust backpressure on sender
256 * and wakeup any waiting to write.
257 */
258 so2->so_snd.sb_mbmax += unp->unp_mbcnt - so->so_rcv.sb_mbcnt;
259 unp->unp_mbcnt = so->so_rcv.sb_mbcnt;
260 newhiwat = so2->so_snd.sb_hiwat + unp->unp_cc -
261 so->so_rcv.sb_cc;
262 (void)chgsbsize(so2->so_cred->cr_uidinfo, &so2->so_snd.sb_hiwat,
263 newhiwat, RLIM_INFINITY);
264 unp->unp_cc = so->so_rcv.sb_cc;
265 sowwakeup(so2);
266 break;
267
268 default:
269 panic("uipc_rcvd unknown socktype");
270 }
271 return (0);
272}
273
274/* pru_rcvoob is EOPNOTSUPP */
275
276static int
277uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
278 struct mbuf *control, struct thread *td)
279{
280 int error = 0;
281 struct unpcb *unp = sotounpcb(so);
282 struct socket *so2;
283 u_long newhiwat;
284
285 if (unp == 0) {
286 error = EINVAL;
287 goto release;
288 }
289 if (flags & PRUS_OOB) {
290 error = EOPNOTSUPP;
291 goto release;
292 }
293
294 if (control && (error = unp_internalize(&control, td)))
295 goto release;
296
297 switch (so->so_type) {
298 case SOCK_DGRAM:
299 {
300 struct sockaddr *from;
301
302 if (nam) {
303 if (unp->unp_conn) {
304 error = EISCONN;
305 break;
306 }
307 error = unp_connect(so, nam, td);
308 if (error)
309 break;
310 } else {
311 if (unp->unp_conn == 0) {
312 error = ENOTCONN;
313 break;
314 }
315 }
316 so2 = unp->unp_conn->unp_socket;
317 if (unp->unp_addr)
318 from = (struct sockaddr *)unp->unp_addr;
319 else
320 from = &sun_noname;
321 if (sbappendaddr(&so2->so_rcv, from, m, control)) {
322 sorwakeup(so2);
323 m = 0;
324 control = 0;
325 } else {
326 error = ENOBUFS;
327 }
328 if (nam)
329 unp_disconnect(unp);
330 break;
331 }
332
333 case SOCK_STREAM:
334 /* Connect if not connected yet. */
335 /*
336 * Note: A better implementation would complain
337 * if not equal to the peer's address.
338 */
339 if ((so->so_state & SS_ISCONNECTED) == 0) {
340 if (nam) {
341 error = unp_connect(so, nam, td);
342 if (error)
343 break; /* XXX */
344 } else {
345 error = ENOTCONN;
346 break;
347 }
348 }
349
350 if (so->so_state & SS_CANTSENDMORE) {
351 error = EPIPE;
352 break;
353 }
354 if (unp->unp_conn == 0)
355 panic("uipc_send connected but no connection?");
356 so2 = unp->unp_conn->unp_socket;
357 /*
358 * Send to paired receive port, and then reduce
359 * send buffer hiwater marks to maintain backpressure.
360 * Wake up readers.
361 */
362 if (control) {
363 if (sbappendcontrol(&so2->so_rcv, m, control))
364 control = 0;
365 } else {
366 sbappend(&so2->so_rcv, m);
367 }
368 so->so_snd.sb_mbmax -=
369 so2->so_rcv.sb_mbcnt - unp->unp_conn->unp_mbcnt;
370 unp->unp_conn->unp_mbcnt = so2->so_rcv.sb_mbcnt;
371 newhiwat = so->so_snd.sb_hiwat -
372 (so2->so_rcv.sb_cc - unp->unp_conn->unp_cc);
373 (void)chgsbsize(so->so_cred->cr_uidinfo, &so->so_snd.sb_hiwat,
374 newhiwat, RLIM_INFINITY);
375 unp->unp_conn->unp_cc = so2->so_rcv.sb_cc;
376 sorwakeup(so2);
377 m = 0;
378 break;
379
380 default:
381 panic("uipc_send unknown socktype");
382 }
383
384 /*
385 * SEND_EOF is equivalent to a SEND followed by
386 * a SHUTDOWN.
387 */
388 if (flags & PRUS_EOF) {
389 socantsendmore(so);
390 unp_shutdown(unp);
391 }
392
393 if (control && error != 0)
394 unp_dispose(control);
395
396release:
397 if (control)
398 m_freem(control);
399 if (m)
400 m_freem(m);
401 return (error);
402}
403
404static int
405uipc_sense(struct socket *so, struct stat *sb)
406{
407 struct unpcb *unp = sotounpcb(so);
408 struct socket *so2;
409
410 if (unp == 0)
411 return (EINVAL);
412 sb->st_blksize = so->so_snd.sb_hiwat;
413 if (so->so_type == SOCK_STREAM && unp->unp_conn != 0) {
414 so2 = unp->unp_conn->unp_socket;
415 sb->st_blksize += so2->so_rcv.sb_cc;
416 }
417 sb->st_dev = NOUDEV;
418 if (unp->unp_ino == 0)
419 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino;
420 sb->st_ino = unp->unp_ino;
421 return (0);
422}
423
424static int
425uipc_shutdown(struct socket *so)
426{
427 struct unpcb *unp = sotounpcb(so);
428
429 if (unp == 0)
430 return (EINVAL);
431 socantsendmore(so);
432 unp_shutdown(unp);
433 return (0);
434}
435
436static int
437uipc_sockaddr(struct socket *so, struct sockaddr **nam)
438{
439 struct unpcb *unp = sotounpcb(so);
440
441 if (unp == 0)
442 return (EINVAL);
443 if (unp->unp_addr)
444 *nam = dup_sockaddr((struct sockaddr *)unp->unp_addr, 1);
445 else
446 *nam = dup_sockaddr((struct sockaddr *)&sun_noname, 1);
447 return (0);
448}
449
450struct pr_usrreqs uipc_usrreqs = {
451 uipc_abort, uipc_accept, uipc_attach, uipc_bind, uipc_connect,
452 uipc_connect2, pru_control_notsupp, uipc_detach, uipc_disconnect,
453 uipc_listen, uipc_peeraddr, uipc_rcvd, pru_rcvoob_notsupp,
454 uipc_send, uipc_sense, uipc_shutdown, uipc_sockaddr,
455 sosend, soreceive, sopoll, pru_sosetlabel_null
456};
457
458int
459uipc_ctloutput(so, sopt)
460 struct socket *so;
461 struct sockopt *sopt;
462{
463 struct unpcb *unp = sotounpcb(so);
464 int error;
465
466 switch (sopt->sopt_dir) {
467 case SOPT_GET:
468 switch (sopt->sopt_name) {
469 case LOCAL_PEERCRED:
470 if (unp->unp_flags & UNP_HAVEPC)
471 error = sooptcopyout(sopt, &unp->unp_peercred,
472 sizeof(unp->unp_peercred));
473 else {
474 if (so->so_type == SOCK_STREAM)
475 error = ENOTCONN;
476 else
477 error = EINVAL;
478 }
479 break;
480 default:
481 error = EOPNOTSUPP;
482 break;
483 }
484 break;
485 case SOPT_SET:
486 default:
487 error = EOPNOTSUPP;
488 break;
489 }
490 return (error);
491}
492
493/*
494 * Both send and receive buffers are allocated PIPSIZ bytes of buffering
495 * for stream sockets, although the total for sender and receiver is
496 * actually only PIPSIZ.
497 * Datagram sockets really use the sendspace as the maximum datagram size,
498 * and don't really want to reserve the sendspace. Their recvspace should
499 * be large enough for at least one max-size datagram plus address.
500 */
501#ifndef PIPSIZ
502#define PIPSIZ 8192
503#endif
504static u_long unpst_sendspace = PIPSIZ;
505static u_long unpst_recvspace = PIPSIZ;
506static u_long unpdg_sendspace = 2*1024; /* really max datagram size */
507static u_long unpdg_recvspace = 4*1024;
508
509static int unp_rights; /* file descriptors in flight */
510
511SYSCTL_DECL(_net_local_stream);
512SYSCTL_INT(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW,
513 &unpst_sendspace, 0, "");
514SYSCTL_INT(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW,
515 &unpst_recvspace, 0, "");
516SYSCTL_DECL(_net_local_dgram);
517SYSCTL_INT(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW,
518 &unpdg_sendspace, 0, "");
519SYSCTL_INT(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW,
520 &unpdg_recvspace, 0, "");
521SYSCTL_DECL(_net_local);
522SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, "");
523
524static int
525unp_attach(so)
526 struct socket *so;
527{
528 register struct unpcb *unp;
529 int error;
530
531 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
532 switch (so->so_type) {
533
534 case SOCK_STREAM:
535 error = soreserve(so, unpst_sendspace, unpst_recvspace);
536 break;
537
538 case SOCK_DGRAM:
539 error = soreserve(so, unpdg_sendspace, unpdg_recvspace);
540 break;
541
542 default:
543 panic("unp_attach");
544 }
545 if (error)
546 return (error);
547 }
548 unp = uma_zalloc(unp_zone, M_WAITOK);
549 if (unp == NULL)
550 return (ENOBUFS);
551 bzero(unp, sizeof *unp);
552 unp->unp_gencnt = ++unp_gencnt;
553 unp_count++;
554 LIST_INIT(&unp->unp_refs);
555 unp->unp_socket = so;
556 LIST_INSERT_HEAD(so->so_type == SOCK_DGRAM ? &unp_dhead
557 : &unp_shead, unp, unp_link);
558 so->so_pcb = unp;
559 return (0);
560}
561
562static void
563unp_detach(unp)
564 register struct unpcb *unp;
565{
566 LIST_REMOVE(unp, unp_link);
567 unp->unp_gencnt = ++unp_gencnt;
568 --unp_count;
569 if (unp->unp_vnode) {
570 unp->unp_vnode->v_socket = 0;
571 vrele(unp->unp_vnode);
572 unp->unp_vnode = 0;
573 }
574 if (unp->unp_conn)
575 unp_disconnect(unp);
576 while (!LIST_EMPTY(&unp->unp_refs))
577 unp_drop(LIST_FIRST(&unp->unp_refs), ECONNRESET);
578 soisdisconnected(unp->unp_socket);
579 unp->unp_socket->so_pcb = 0;
580 if (unp_rights) {
581 /*
582 * Normally the receive buffer is flushed later,
583 * in sofree, but if our receive buffer holds references
584 * to descriptors that are now garbage, we will dispose
585 * of those descriptor references after the garbage collector
586 * gets them (resulting in a "panic: closef: count < 0").
587 */
588 sorflush(unp->unp_socket);
589 unp_gc();
590 }
591 if (unp->unp_addr)
592 FREE(unp->unp_addr, M_SONAME);
593 uma_zfree(unp_zone, unp);
594}
595
596static int
597unp_bind(unp, nam, td)
598 struct unpcb *unp;
599 struct sockaddr *nam;
600 struct thread *td;
601{
602 struct sockaddr_un *soun = (struct sockaddr_un *)nam;
603 struct vnode *vp;
604 struct mount *mp;
605 struct vattr vattr;
606 int error, namelen;
607 struct nameidata nd;
608 char *buf;
609
610 if (unp->unp_vnode != NULL)
611 return (EINVAL);
612
613 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path);
614 if (namelen <= 0)
615 return (EINVAL);
616
617 buf = malloc(namelen + 1, M_TEMP, M_WAITOK);
618 strlcpy(buf, soun->sun_path, namelen + 1);
619
620restart:
621 NDINIT(&nd, CREATE, NOFOLLOW | LOCKPARENT | SAVENAME, UIO_SYSSPACE,
622 buf, td);
623/* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
624 error = namei(&nd);
625 if (error) {
626 free(buf, M_TEMP);
627 return (error);
628 }
629 vp = nd.ni_vp;
630 if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
631 NDFREE(&nd, NDF_ONLY_PNBUF);
632 if (nd.ni_dvp == vp)
633 vrele(nd.ni_dvp);
634 else
635 vput(nd.ni_dvp);
636 if (vp != NULL) {
637 vrele(vp);
638 free(buf, M_TEMP);
639 return (EADDRINUSE);
640 }
641 error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH);
642 if (error) {
643 free(buf, M_TEMP);
644 return (error);
645 }
646 goto restart;
647 }
648 VATTR_NULL(&vattr);
649 vattr.va_type = VSOCK;
650 vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask);
651#ifdef MAC
652 error = mac_check_vnode_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd,
653 &vattr);
654#endif
655 if (error == 0) {
656 VOP_LEASE(nd.ni_dvp, td, td->td_ucred, LEASE_WRITE);
657 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
658 }
659 NDFREE(&nd, NDF_ONLY_PNBUF);
660 vput(nd.ni_dvp);
661 if (error) {
662 free(buf, M_TEMP);
663 return (error);
664 }
665 vp = nd.ni_vp;
666 vp->v_socket = unp->unp_socket;
667 unp->unp_vnode = vp;
668 unp->unp_addr = (struct sockaddr_un *)dup_sockaddr(nam, 1);
669 VOP_UNLOCK(vp, 0, td);
670 vn_finished_write(mp);
671 free(buf, M_TEMP);
672 return (0);
673}
674
675static int
676unp_connect(so, nam, td)
677 struct socket *so;
678 struct sockaddr *nam;
679 struct thread *td;
680{
681 register struct sockaddr_un *soun = (struct sockaddr_un *)nam;
682 register struct vnode *vp;
683 register struct socket *so2, *so3;
684 struct unpcb *unp, *unp2, *unp3;
685 int error, len;
686 struct nameidata nd;
687 char buf[SOCK_MAXADDRLEN];
688
689 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
690 if (len <= 0)
691 return (EINVAL);
692 strlcpy(buf, soun->sun_path, len + 1);
693
694 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, buf, td);
695 error = namei(&nd);
696 if (error)
697 return (error);
698 vp = nd.ni_vp;
699 NDFREE(&nd, NDF_ONLY_PNBUF);
700 if (vp->v_type != VSOCK) {
701 error = ENOTSOCK;
702 goto bad;
703 }
704 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td);
705 if (error)
706 goto bad;
707 so2 = vp->v_socket;
708 if (so2 == 0) {
709 error = ECONNREFUSED;
710 goto bad;
711 }
712 if (so->so_type != so2->so_type) {
713 error = EPROTOTYPE;
714 goto bad;
715 }
716 if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
717 if ((so2->so_options & SO_ACCEPTCONN) == 0 ||
718 (so3 = sonewconn(so2, 0)) == 0) {
719 error = ECONNREFUSED;
720 goto bad;
721 }
722 unp = sotounpcb(so);
723 unp2 = sotounpcb(so2);
724 unp3 = sotounpcb(so3);
725 if (unp2->unp_addr)
726 unp3->unp_addr = (struct sockaddr_un *)
727 dup_sockaddr((struct sockaddr *)
728 unp2->unp_addr, 1);
729
730 /*
731 * unp_peercred management:
732 *
733 * The connecter's (client's) credentials are copied
734 * from its process structure at the time of connect()
735 * (which is now).
736 */
737 cru2x(td->td_ucred, &unp3->unp_peercred);
738 unp3->unp_flags |= UNP_HAVEPC;
739 /*
740 * The receiver's (server's) credentials are copied
741 * from the unp_peercred member of socket on which the
742 * former called listen(); unp_listen() cached that
743 * process's credentials at that time so we can use
744 * them now.
745 */
746 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED,
747 ("unp_connect: listener without cached peercred"));
748 memcpy(&unp->unp_peercred, &unp2->unp_peercred,
749 sizeof(unp->unp_peercred));
750 unp->unp_flags |= UNP_HAVEPC;
751#ifdef MAC
752 mac_set_socket_peer_from_socket(so, so3);
753 mac_set_socket_peer_from_socket(so3, so);
754#endif
755
756 so2 = so3;
757 }
758 error = unp_connect2(so, so2);
759bad:
760 vput(vp);
761 return (error);
762}
763
764int
765unp_connect2(so, so2)
766 register struct socket *so;
767 register struct socket *so2;
768{
769 register struct unpcb *unp = sotounpcb(so);
770 register struct unpcb *unp2;
771
772 if (so2->so_type != so->so_type)
773 return (EPROTOTYPE);
774 unp2 = sotounpcb(so2);
775 unp->unp_conn = unp2;
776 switch (so->so_type) {
777
778 case SOCK_DGRAM:
779 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
780 soisconnected(so);
781 break;
782
783 case SOCK_STREAM:
784 unp2->unp_conn = unp;
785 soisconnected(so);
786 soisconnected(so2);
787 break;
788
789 default:
790 panic("unp_connect2");
791 }
792 return (0);
793}
794
795static void
796unp_disconnect(unp)
797 struct unpcb *unp;
798{
799 register struct unpcb *unp2 = unp->unp_conn;
800
801 if (unp2 == 0)
802 return;
803 unp->unp_conn = 0;
804 switch (unp->unp_socket->so_type) {
805
806 case SOCK_DGRAM:
807 LIST_REMOVE(unp, unp_reflink);
808 unp->unp_socket->so_state &= ~SS_ISCONNECTED;
809 break;
810
811 case SOCK_STREAM:
812 soisdisconnected(unp->unp_socket);
813 unp2->unp_conn = 0;
814 soisdisconnected(unp2->unp_socket);
815 break;
816 }
817}
818
819#ifdef notdef
820void
821unp_abort(unp)
822 struct unpcb *unp;
823{
824
825 unp_detach(unp);
826}
827#endif
828
829static int
830unp_pcblist(SYSCTL_HANDLER_ARGS)
831{
832 int error, i, n;
833 struct unpcb *unp, **unp_list;
834 unp_gen_t gencnt;
835 struct xunpgen *xug;
836 struct unp_head *head;
837 struct xunpcb *xu;
838
839 head = ((intptr_t)arg1 == SOCK_DGRAM ? &unp_dhead : &unp_shead);
840
841 /*
842 * The process of preparing the PCB list is too time-consuming and
843 * resource-intensive to repeat twice on every request.
844 */
845 if (req->oldptr == 0) {
846 n = unp_count;
847 req->oldidx = 2 * (sizeof *xug)
848 + (n + n/8) * sizeof(struct xunpcb);
849 return (0);
850 }
851
852 if (req->newptr != 0)
853 return (EPERM);
854
855 /*
856 * OK, now we're committed to doing something.
857 */
858 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK);
859 gencnt = unp_gencnt;
860 n = unp_count;
861
862 xug->xug_len = sizeof *xug;
863 xug->xug_count = n;
864 xug->xug_gen = gencnt;
865 xug->xug_sogen = so_gencnt;
866 error = SYSCTL_OUT(req, xug, sizeof *xug);
867 if (error) {
868 free(xug, M_TEMP);
869 return (error);
870 }
871
872 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
873
874 for (unp = LIST_FIRST(head), i = 0; unp && i < n;
875 unp = LIST_NEXT(unp, unp_link)) {
876 if (unp->unp_gencnt <= gencnt) {
877 if (cr_cansee(req->td->td_ucred,
878 unp->unp_socket->so_cred))
879 continue;
880 unp_list[i++] = unp;
881 }
882 }
883 n = i; /* in case we lost some during malloc */
884
885 error = 0;
886 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK);
887 for (i = 0; i < n; i++) {
888 unp = unp_list[i];
889 if (unp->unp_gencnt <= gencnt) {
890 xu->xu_len = sizeof *xu;
891 xu->xu_unpp = unp;
892 /*
893 * XXX - need more locking here to protect against
894 * connect/disconnect races for SMP.
895 */
896 if (unp->unp_addr)
897 bcopy(unp->unp_addr, &xu->xu_addr,
898 unp->unp_addr->sun_len);
899 if (unp->unp_conn && unp->unp_conn->unp_addr)
900 bcopy(unp->unp_conn->unp_addr,
901 &xu->xu_caddr,
902 unp->unp_conn->unp_addr->sun_len);
903 bcopy(unp, &xu->xu_unp, sizeof *unp);
904 sotoxsocket(unp->unp_socket, &xu->xu_socket);
905 error = SYSCTL_OUT(req, xu, sizeof *xu);
906 }
907 }
908 free(xu, M_TEMP);
909 if (!error) {
910 /*
911 * Give the user an updated idea of our state.
912 * If the generation differs from what we told
913 * her before, she knows that something happened
914 * while we were processing this request, and it
915 * might be necessary to retry.
916 */
917 xug->xug_gen = unp_gencnt;
918 xug->xug_sogen = so_gencnt;
919 xug->xug_count = unp_count;
920 error = SYSCTL_OUT(req, xug, sizeof *xug);
921 }
922 free(unp_list, M_TEMP);
923 free(xug, M_TEMP);
924 return (error);
925}
926
927SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLFLAG_RD,
928 (caddr_t)(long)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
929 "List of active local datagram sockets");
930SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLFLAG_RD,
931 (caddr_t)(long)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
932 "List of active local stream sockets");
933
934static void
935unp_shutdown(unp)
936 struct unpcb *unp;
937{
938 struct socket *so;
939
940 if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn &&
941 (so = unp->unp_conn->unp_socket))
942 socantrcvmore(so);
943}
944
945static void
946unp_drop(unp, errno)
947 struct unpcb *unp;
948 int errno;
949{
950 struct socket *so = unp->unp_socket;
951
952 so->so_error = errno;
953 unp_disconnect(unp);
954}
955
956#ifdef notdef
957void
958unp_drain()
959{
960
961}
962#endif
963
964static void
965unp_freerights(rp, fdcount)
966 struct file **rp;
967 int fdcount;
968{
969 int i;
970 struct file *fp;
971
972 for (i = 0; i < fdcount; i++) {
973 fp = *rp;
974 /*
975 * zero the pointer before calling
976 * unp_discard since it may end up
977 * in unp_gc()..
978 */
979 *rp++ = 0;
980 unp_discard(fp);
981 }
982}
983
984int
985unp_externalize(control, controlp)
986 struct mbuf *control, **controlp;
987{
988 struct thread *td = curthread; /* XXX */
989 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
990 int i;
991 int *fdp;
992 struct file **rp;
993 struct file *fp;
994 void *data;
995 socklen_t clen = control->m_len, datalen;
996 int error, newfds;
997 int f;
998 u_int newlen;
999
1000 error = 0;
1001 if (controlp != NULL) /* controlp == NULL => free control messages */
1002 *controlp = NULL;
1003
1004 while (cm != NULL) {
1005 if (sizeof(*cm) > clen || cm->cmsg_len > clen) {
1006 error = EINVAL;
1007 break;
1008 }
1009
1010 data = CMSG_DATA(cm);
1011 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1012
1013 if (cm->cmsg_level == SOL_SOCKET
1014 && cm->cmsg_type == SCM_RIGHTS) {
1015 newfds = datalen / sizeof(struct file *);
1016 rp = data;
1017
1018 /* If we're not outputting the descriptors free them. */
1019 if (error || controlp == NULL) {
1020 unp_freerights(rp, newfds);
1021 goto next;
1022 }
1023 FILEDESC_LOCK(td->td_proc->p_fd);
1024 /* if the new FD's will not fit free them. */
1025 if (!fdavail(td, newfds)) {
1026 FILEDESC_UNLOCK(td->td_proc->p_fd);
1027 error = EMSGSIZE;
1028 unp_freerights(rp, newfds);
1029 goto next;
1030 }
1031 /*
1032 * now change each pointer to an fd in the global
1033 * table to an integer that is the index to the
1034 * local fd table entry that we set up to point
1035 * to the global one we are transferring.
1036 */
1037 newlen = newfds * sizeof(int);
1038 *controlp = sbcreatecontrol(NULL, newlen,
1039 SCM_RIGHTS, SOL_SOCKET);
1040 if (*controlp == NULL) {
1041 FILEDESC_UNLOCK(td->td_proc->p_fd);
1042 error = E2BIG;
1043 unp_freerights(rp, newfds);
1044 goto next;
1045 }
1046
1047 fdp = (int *)
1048 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1049 for (i = 0; i < newfds; i++) {
1050 if (fdalloc(td, 0, &f))
1051 panic("unp_externalize fdalloc failed");
1052 fp = *rp++;
1053 td->td_proc->p_fd->fd_ofiles[f] = fp;
1054 FILE_LOCK(fp);
1055 fp->f_msgcount--;
1056 FILE_UNLOCK(fp);
1057 unp_rights--;
1058 *fdp++ = f;
1059 }
1060 FILEDESC_UNLOCK(td->td_proc->p_fd);
1061 } else { /* We can just copy anything else across */
1062 if (error || controlp == NULL)
1063 goto next;
1064 *controlp = sbcreatecontrol(NULL, datalen,
1065 cm->cmsg_type, cm->cmsg_level);
1066 if (*controlp == NULL) {
1067 error = ENOBUFS;
1068 goto next;
1069 }
1070 bcopy(data,
1071 CMSG_DATA(mtod(*controlp, struct cmsghdr *)),
1072 datalen);
1073 }
1074
1075 controlp = &(*controlp)->m_next;
1076
1077next:
1078 if (CMSG_SPACE(datalen) < clen) {
1079 clen -= CMSG_SPACE(datalen);
1080 cm = (struct cmsghdr *)
1081 ((caddr_t)cm + CMSG_SPACE(datalen));
1082 } else {
1083 clen = 0;
1084 cm = NULL;
1085 }
1086 }
1087
1088 m_freem(control);
1089
1090 return (error);
1091}
1092
1093void
1094unp_init(void)
1095{
1096 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL,
1097 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
1098 if (unp_zone == 0)
1099 panic("unp_init");
1100 uma_zone_set_max(unp_zone, nmbclusters);
1101 LIST_INIT(&unp_dhead);
1102 LIST_INIT(&unp_shead);
1103}
1104
1105static int
1106unp_internalize(controlp, td)
1107 struct mbuf **controlp;
1108 struct thread *td;
1109{
1110 struct mbuf *control = *controlp;
1111 struct proc *p = td->td_proc;
1112 struct filedesc *fdescp = p->p_fd;
1113 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1114 struct cmsgcred *cmcred;
1115 struct file **rp;
1116 struct file *fp;
1117 struct timeval *tv;
1118 int i, fd, *fdp;
1119 void *data;
1120 socklen_t clen = control->m_len, datalen;
1121 int error, oldfds;
1122 u_int newlen;
1123
1124 error = 0;
1125 *controlp = NULL;
1126
1127 while (cm != NULL) {
1128 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET
1129 || cm->cmsg_len > clen) {
1130 error = EINVAL;
1131 goto out;
1132 }
1133
1134 data = CMSG_DATA(cm);
1135 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1136
1137 switch (cm->cmsg_type) {
1138 /*
1139 * Fill in credential information.
1140 */
1141 case SCM_CREDS:
1142 *controlp = sbcreatecontrol(NULL, sizeof(*cmcred),
1143 SCM_CREDS, SOL_SOCKET);
1144 if (*controlp == NULL) {
1145 error = ENOBUFS;
1146 goto out;
1147 }
1148
1149 cmcred = (struct cmsgcred *)
1150 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1151 cmcred->cmcred_pid = p->p_pid;
1152 cmcred->cmcred_uid = td->td_ucred->cr_ruid;
1153 cmcred->cmcred_gid = td->td_ucred->cr_rgid;
1154 cmcred->cmcred_euid = td->td_ucred->cr_uid;
1155 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups,
1156 CMGROUP_MAX);
1157 for (i = 0; i < cmcred->cmcred_ngroups; i++)
1158 cmcred->cmcred_groups[i] =
1159 td->td_ucred->cr_groups[i];
1160 break;
1161
1162 case SCM_RIGHTS:
1163 oldfds = datalen / sizeof (int);
1164 /*
1165 * check that all the FDs passed in refer to legal files
1166 * If not, reject the entire operation.
1167 */
1168 fdp = data;
1169 FILEDESC_LOCK(fdescp);
1170 for (i = 0; i < oldfds; i++) {
1171 fd = *fdp++;
1172 if ((unsigned)fd >= fdescp->fd_nfiles ||
1173 fdescp->fd_ofiles[fd] == NULL) {
1174 FILEDESC_UNLOCK(fdescp);
1175 error = EBADF;
1176 goto out;
1177 }
1178 fp = fdescp->fd_ofiles[fd];
1179 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) {
1180 FILEDESC_UNLOCK(fdescp);
1181 error = EOPNOTSUPP;
1182 goto out;
1183 }
1184
1185 }
1186 /*
1187 * Now replace the integer FDs with pointers to
1188 * the associated global file table entry..
1189 */
1190 newlen = oldfds * sizeof(struct file *);
1191 *controlp = sbcreatecontrol(NULL, newlen,
1192 SCM_RIGHTS, SOL_SOCKET);
1193 if (*controlp == NULL) {
1194 FILEDESC_UNLOCK(fdescp);
1195 error = E2BIG;
1196 goto out;
1197 }
1198
1199 fdp = data;
1200 rp = (struct file **)
1201 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1202 for (i = 0; i < oldfds; i++) {
1203 fp = fdescp->fd_ofiles[*fdp++];
1204 *rp++ = fp;
1205 FILE_LOCK(fp);
1206 fp->f_count++;
1207 fp->f_msgcount++;
1208 FILE_UNLOCK(fp);
1209 unp_rights++;
1210 }
1211 FILEDESC_UNLOCK(fdescp);
1212 break;
1213
1214 case SCM_TIMESTAMP:
1215 *controlp = sbcreatecontrol(NULL, sizeof(*tv),
1216 SCM_TIMESTAMP, SOL_SOCKET);
1217 if (*controlp == NULL) {
1218 error = ENOBUFS;
1219 goto out;
1220 }
1221 tv = (struct timeval *)
1222 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1223 microtime(tv);
1224 break;
1225
1226 default:
1227 error = EINVAL;
1228 goto out;
1229 }
1230
1231 controlp = &(*controlp)->m_next;
1232
1233 if (CMSG_SPACE(datalen) < clen) {
1234 clen -= CMSG_SPACE(datalen);
1235 cm = (struct cmsghdr *)
1236 ((caddr_t)cm + CMSG_SPACE(datalen));
1237 } else {
1238 clen = 0;
1239 cm = NULL;
1240 }
1241 }
1242
1243out:
1244 m_freem(control);
1245
1246 return (error);
1247}
1248
1249static int unp_defer, unp_gcing;
1250
1251static void
1252unp_gc()
1253{
1254 register struct file *fp, *nextfp;
1255 register struct socket *so;
1256 struct file **extra_ref, **fpp;
1257 int nunref, i;
1258
1259 if (unp_gcing)
1260 return;
1261 unp_gcing = 1;
1262 unp_defer = 0;
1263 /*
1264 * before going through all this, set all FDs to
1265 * be NOT defered and NOT externally accessible
1266 */
1267 sx_slock(&filelist_lock);
1268 LIST_FOREACH(fp, &filehead, f_list)
1269 fp->f_gcflag &= ~(FMARK|FDEFER);
1270 do {
1271 LIST_FOREACH(fp, &filehead, f_list) {
1272 FILE_LOCK(fp);
1273 /*
1274 * If the file is not open, skip it
1275 */
1276 if (fp->f_count == 0) {
1277 FILE_UNLOCK(fp);
1278 continue;
1279 }
1280 /*
1281 * If we already marked it as 'defer' in a
1282 * previous pass, then try process it this time
1283 * and un-mark it
1284 */
1285 if (fp->f_gcflag & FDEFER) {
1286 fp->f_gcflag &= ~FDEFER;
1287 unp_defer--;
1288 } else {
1289 /*
1290 * if it's not defered, then check if it's
1291 * already marked.. if so skip it
1292 */
1293 if (fp->f_gcflag & FMARK) {
1294 FILE_UNLOCK(fp);
1295 continue;
1296 }
1297 /*
1298 * If all references are from messages
1299 * in transit, then skip it. it's not
1300 * externally accessible.
1301 */
1302 if (fp->f_count == fp->f_msgcount) {
1303 FILE_UNLOCK(fp);
1304 continue;
1305 }
1306 /*
1307 * If it got this far then it must be
1308 * externally accessible.
1309 */
1310 fp->f_gcflag |= FMARK;
1311 }
1312 /*
1313 * either it was defered, or it is externally
1314 * accessible and not already marked so.
1315 * Now check if it is possibly one of OUR sockets.
1316 */
1317 if (fp->f_type != DTYPE_SOCKET ||
1318 (so = fp->f_data) == NULL) {
1319 FILE_UNLOCK(fp);
1320 continue;
1321 }
1322 FILE_UNLOCK(fp);
1323 if (so->so_proto->pr_domain != &localdomain ||
1324 (so->so_proto->pr_flags&PR_RIGHTS) == 0)
1325 continue;
1326#ifdef notdef
1327 if (so->so_rcv.sb_flags & SB_LOCK) {
1328 /*
1329 * This is problematical; it's not clear
1330 * we need to wait for the sockbuf to be
1331 * unlocked (on a uniprocessor, at least),
1332 * and it's also not clear what to do
1333 * if sbwait returns an error due to receipt
1334 * of a signal. If sbwait does return
1335 * an error, we'll go into an infinite
1336 * loop. Delete all of this for now.
1337 */
1338 (void) sbwait(&so->so_rcv);
1339 goto restart;
1340 }
1341#endif
1342 /*
1343 * So, Ok, it's one of our sockets and it IS externally
1344 * accessible (or was defered). Now we look
1345 * to see if we hold any file descriptors in its
1346 * message buffers. Follow those links and mark them
1347 * as accessible too.
1348 */
1349 unp_scan(so->so_rcv.sb_mb, unp_mark);
1350 }
1351 } while (unp_defer);
1352 sx_sunlock(&filelist_lock);
1353 /*
1354 * We grab an extra reference to each of the file table entries
1355 * that are not otherwise accessible and then free the rights
1356 * that are stored in messages on them.
1357 *
1358 * The bug in the orginal code is a little tricky, so I'll describe
1359 * what's wrong with it here.
1360 *
1361 * It is incorrect to simply unp_discard each entry for f_msgcount
1362 * times -- consider the case of sockets A and B that contain
1363 * references to each other. On a last close of some other socket,
1364 * we trigger a gc since the number of outstanding rights (unp_rights)
1365 * is non-zero. If during the sweep phase the gc code un_discards,
1366 * we end up doing a (full) closef on the descriptor. A closef on A
1367 * results in the following chain. Closef calls soo_close, which
1368 * calls soclose. Soclose calls first (through the switch
1369 * uipc_usrreq) unp_detach, which re-invokes unp_gc. Unp_gc simply
1370 * returns because the previous instance had set unp_gcing, and
1371 * we return all the way back to soclose, which marks the socket
1372 * with SS_NOFDREF, and then calls sofree. Sofree calls sorflush
1373 * to free up the rights that are queued in messages on the socket A,
1374 * i.e., the reference on B. The sorflush calls via the dom_dispose
1375 * switch unp_dispose, which unp_scans with unp_discard. This second
1376 * instance of unp_discard just calls closef on B.
1377 *
1378 * Well, a similar chain occurs on B, resulting in a sorflush on B,
1379 * which results in another closef on A. Unfortunately, A is already
1380 * being closed, and the descriptor has already been marked with
1381 * SS_NOFDREF, and soclose panics at this point.
1382 *
1383 * Here, we first take an extra reference to each inaccessible
1384 * descriptor. Then, we call sorflush ourself, since we know
1385 * it is a Unix domain socket anyhow. After we destroy all the
1386 * rights carried in messages, we do a last closef to get rid
1387 * of our extra reference. This is the last close, and the
1388 * unp_detach etc will shut down the socket.
1389 *
1390 * 91/09/19, bsy@cs.cmu.edu
1391 */
1392 extra_ref = malloc(nfiles * sizeof(struct file *), M_TEMP, M_WAITOK);
1393 sx_slock(&filelist_lock);
1394 for (nunref = 0, fp = LIST_FIRST(&filehead), fpp = extra_ref; fp != 0;
1395 fp = nextfp) {
1396 nextfp = LIST_NEXT(fp, f_list);
1397 FILE_LOCK(fp);
1398 /*
1399 * If it's not open, skip it
1400 */
1401 if (fp->f_count == 0) {
1402 FILE_UNLOCK(fp);
1403 continue;
1404 }
1405 /*
1406 * If all refs are from msgs, and it's not marked accessible
1407 * then it must be referenced from some unreachable cycle
1408 * of (shut-down) FDs, so include it in our
1409 * list of FDs to remove
1410 */
1411 if (fp->f_count == fp->f_msgcount && !(fp->f_gcflag & FMARK)) {
1412 *fpp++ = fp;
1413 nunref++;
1414 fp->f_count++;
1415 }
1416 FILE_UNLOCK(fp);
1417 }
1418 sx_sunlock(&filelist_lock);
1419 /*
1420 * for each FD on our hit list, do the following two things
1421 */
1422 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) {
1423 struct file *tfp = *fpp;
1424 FILE_LOCK(tfp);
1425 if (tfp->f_type == DTYPE_SOCKET &&
1426 tfp->f_data != NULL) {
1427 FILE_UNLOCK(tfp);
1428 sorflush(tfp->f_data);
1429 } else {
1430 FILE_UNLOCK(tfp);
1431 }
1432 }
1433 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp)
1434 closef(*fpp, (struct thread *) NULL);
1435 free(extra_ref, M_TEMP);
1436 unp_gcing = 0;
1437}
1438
1439void
1440unp_dispose(m)
1441 struct mbuf *m;
1442{
1443
1444 if (m)
1445 unp_scan(m, unp_discard);
1446}
1447
1448static int
1449unp_listen(unp, td)
1450 struct unpcb *unp;
1451 struct thread *td;
1452{
1453
1454 cru2x(td->td_ucred, &unp->unp_peercred);
1455 unp->unp_flags |= UNP_HAVEPCCACHED;
1456 return (0);
1457}
1458
1459static void
1460unp_scan(m0, op)
1461 register struct mbuf *m0;
1462 void (*op)(struct file *);
1463{
1464 struct mbuf *m;
1465 struct file **rp;
1466 struct cmsghdr *cm;
1467 void *data;
1468 int i;
1469 socklen_t clen, datalen;
1470 int qfds;
1471
1472 while (m0) {
1473 for (m = m0; m; m = m->m_next) {
1474 if (m->m_type != MT_CONTROL)
1475 continue;
1476
1477 cm = mtod(m, struct cmsghdr *);
1478 clen = m->m_len;
1479
1480 while (cm != NULL) {
1481 if (sizeof(*cm) > clen || cm->cmsg_len > clen)
1482 break;
1483
1484 data = CMSG_DATA(cm);
1485 datalen = (caddr_t)cm + cm->cmsg_len
1486 - (caddr_t)data;
1487
1488 if (cm->cmsg_level == SOL_SOCKET &&
1489 cm->cmsg_type == SCM_RIGHTS) {
1490 qfds = datalen / sizeof (struct file *);
1491 rp = data;
1492 for (i = 0; i < qfds; i++)
1493 (*op)(*rp++);
1494 }
1495
1496 if (CMSG_SPACE(datalen) < clen) {
1497 clen -= CMSG_SPACE(datalen);
1498 cm = (struct cmsghdr *)
1499 ((caddr_t)cm + CMSG_SPACE(datalen));
1500 } else {
1501 clen = 0;
1502 cm = NULL;
1503 }
1504 }
1505 }
1506 m0 = m0->m_act;
1507 }
1508}
1509
1510static void
1511unp_mark(fp)
1512 struct file *fp;
1513{
1514 if (fp->f_gcflag & FMARK)
1515 return;
1516 unp_defer++;
1517 fp->f_gcflag |= (FMARK|FDEFER);
1518}
1519
1520static void
1521unp_discard(fp)
1522 struct file *fp;
1523{
1524 FILE_LOCK(fp);
1525 fp->f_msgcount--;
1526 unp_rights--;
1527 FILE_UNLOCK(fp);
1528 (void) closef(fp, (struct thread *)NULL);
1529}