1 /*
2 * Copyright 2004 Robert N. M. Watson
3 * Copyright (c) 1982, 1986, 1989, 1991, 1993
4 * The Regents of the University of California. All rights reserved.
5 *
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
8 * are met:
9 * 1. Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
14 * 4. Neither the name of the University nor the names of its contributors
15 * may be used to endorse or promote products derived from this software
16 * without specific prior written permission.
17 *
18 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28 * SUCH DAMAGE.
29 *
30 * From: @(#)uipc_usrreq.c 8.3 (Berkeley) 1/4/94
31 */
32
33 #include <sys/cdefs.h>
34 __FBSDID("$FreeBSD: releng/5.3/sys/kern/uipc_usrreq.c 146003 2005-05-08 10:23:51Z cperciva $");
35
36 #include "opt_mac.h"
37
38 #include <sys/param.h>
39 #include <sys/domain.h>
40 #include <sys/fcntl.h>
41 #include <sys/malloc.h> /* XXX must be before <sys/file.h> */
42 #include <sys/file.h>
43 #include <sys/filedesc.h>
44 #include <sys/jail.h>
45 #include <sys/kernel.h>
46 #include <sys/lock.h>
47 #include <sys/mac.h>
48 #include <sys/mbuf.h>
49 #include <sys/mutex.h>
50 #include <sys/namei.h>
51 #include <sys/proc.h>
52 #include <sys/protosw.h>
53 #include <sys/resourcevar.h>
54 #include <sys/socket.h>
55 #include <sys/socketvar.h>
56 #include <sys/signalvar.h>
57 #include <sys/stat.h>
58 #include <sys/sx.h>
59 #include <sys/sysctl.h>
60 #include <sys/systm.h>
61 #include <sys/un.h>
62 #include <sys/unpcb.h>
63 #include <sys/vnode.h>
64
65 #include <vm/uma.h>
66
67 static uma_zone_t unp_zone;
68 static unp_gen_t unp_gencnt;
69 static u_int unp_count;
70
71 static struct unp_head unp_shead, unp_dhead;
72
73 /*
74 * Unix communications domain.
75 *
76 * TODO:
77 * SEQPACKET, RDM
78 * rethink name space problems
79 * need a proper out-of-band
80 * lock pushdown
81 */
82 static const struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL };
83 static ino_t unp_ino; /* prototype for fake inode numbers */
84
85 /*
86 * Currently, UNIX domain sockets are protected by a single subsystem lock,
87 * which covers global data structures and variables, the contents of each
88 * per-socket unpcb structure, and the so_pcb field in sockets attached to
89 * the UNIX domain. This provides for a moderate degree of paralellism, as
90 * receive operations on UNIX domain sockets do not need to acquire the
91 * subsystem lock. Finer grained locking to permit send() without acquiring
92 * a global lock would be a logical next step.
93 *
94 * The UNIX domain socket lock preceds all socket layer locks, including the
95 * socket lock and socket buffer lock, permitting UNIX domain socket code to
96 * call into socket support routines without releasing its locks.
97 *
98 * Some caution is required in areas where the UNIX domain socket code enters
99 * VFS in order to create or find rendezvous points. This results in
100 * dropping of the UNIX domain socket subsystem lock, acquisition of the
101 * Giant lock, and potential sleeping. This increases the chances of races,
102 * and exposes weaknesses in the socket->protocol API by offering poor
103 * failure modes.
104 */
105 static struct mtx unp_mtx;
106 #define UNP_LOCK_INIT() \
107 mtx_init(&unp_mtx, "unp", NULL, MTX_DEF)
108 #define UNP_LOCK() mtx_lock(&unp_mtx)
109 #define UNP_UNLOCK() mtx_unlock(&unp_mtx)
110 #define UNP_LOCK_ASSERT() mtx_assert(&unp_mtx, MA_OWNED)
111 #define UNP_UNLOCK_ASSERT() mtx_assert(&unp_mtx, MA_NOTOWNED)
112
113 static int unp_attach(struct socket *);
114 static void unp_detach(struct unpcb *);
115 static int unp_bind(struct unpcb *,struct sockaddr *, struct thread *);
116 static int unp_connect(struct socket *,struct sockaddr *, struct thread *);
117 static int unp_connect2(struct socket *so, struct socket *so2);
118 static void unp_disconnect(struct unpcb *);
119 static void unp_shutdown(struct unpcb *);
120 static void unp_drop(struct unpcb *, int);
121 static void unp_gc(void);
122 static void unp_scan(struct mbuf *, void (*)(struct file *));
123 static void unp_mark(struct file *);
124 static void unp_discard(struct file *);
125 static void unp_freerights(struct file **, int);
126 static int unp_internalize(struct mbuf **, struct thread *);
127 static int unp_listen(struct unpcb *, struct thread *);
128
129 static int
130 uipc_abort(struct socket *so)
131 {
132 struct unpcb *unp;
133
134 UNP_LOCK();
135 unp = sotounpcb(so);
136 if (unp == NULL) {
137 UNP_UNLOCK();
138 return (EINVAL);
139 }
140 unp_drop(unp, ECONNABORTED);
141 unp_detach(unp);
142 UNP_UNLOCK_ASSERT();
143 ACCEPT_LOCK();
144 SOCK_LOCK(so);
145 sotryfree(so);
146 return (0);
147 }
148
149 static int
150 uipc_accept(struct socket *so, struct sockaddr **nam)
151 {
152 struct unpcb *unp;
153 const struct sockaddr *sa;
154
155 /*
156 * Pass back name of connected socket,
157 * if it was bound and we are still connected
158 * (our peer may have closed already!).
159 */
160 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
161 UNP_LOCK();
162 unp = sotounpcb(so);
163 if (unp == NULL) {
164 UNP_UNLOCK();
165 free(*nam, M_SONAME);
166 *nam = NULL;
167 return (EINVAL);
168 }
169 if (unp->unp_conn != NULL && unp->unp_conn->unp_addr != NULL)
170 sa = (struct sockaddr *) unp->unp_conn->unp_addr;
171 else
172 sa = &sun_noname;
173 bcopy(sa, *nam, sa->sa_len);
174 UNP_UNLOCK();
175 return (0);
176 }
177
178 static int
179 uipc_attach(struct socket *so, int proto, struct thread *td)
180 {
181 struct unpcb *unp = sotounpcb(so);
182
183 if (unp != NULL)
184 return (EISCONN);
185 return (unp_attach(so));
186 }
187
188 static int
189 uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
190 {
191 struct unpcb *unp;
192 int error;
193
194 UNP_LOCK();
195 unp = sotounpcb(so);
196 if (unp == NULL) {
197 UNP_UNLOCK();
198 return (EINVAL);
199 }
200 error = unp_bind(unp, nam, td);
201 UNP_UNLOCK();
202 return (error);
203 }
204
205 static int
206 uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
207 {
208 struct unpcb *unp;
209 int error;
210
211 KASSERT(td == curthread, ("uipc_connect: td != curthread"));
212
213 UNP_LOCK();
214 unp = sotounpcb(so);
215 if (unp == NULL) {
216 UNP_UNLOCK();
217 return (EINVAL);
218 }
219 error = unp_connect(so, nam, td);
220 UNP_UNLOCK();
221 return (error);
222 }
223
224 int
225 uipc_connect2(struct socket *so1, struct socket *so2)
226 {
227 struct unpcb *unp;
228 int error;
229
230 UNP_LOCK();
231 unp = sotounpcb(so1);
232 if (unp == NULL) {
233 UNP_UNLOCK();
234 return (EINVAL);
235 }
236 error = unp_connect2(so1, so2);
237 UNP_UNLOCK();
238 return (error);
239 }
240
241 /* control is EOPNOTSUPP */
242
243 static int
244 uipc_detach(struct socket *so)
245 {
246 struct unpcb *unp;
247
248 UNP_LOCK();
249 unp = sotounpcb(so);
250 if (unp == NULL) {
251 UNP_UNLOCK();
252 return (EINVAL);
253 }
254 unp_detach(unp);
255 UNP_UNLOCK_ASSERT();
256 return (0);
257 }
258
259 static int
260 uipc_disconnect(struct socket *so)
261 {
262 struct unpcb *unp;
263
264 UNP_LOCK();
265 unp = sotounpcb(so);
266 if (unp == NULL) {
267 UNP_UNLOCK();
268 return (EINVAL);
269 }
270 unp_disconnect(unp);
271 UNP_UNLOCK();
272 return (0);
273 }
274
275 static int
276 uipc_listen(struct socket *so, struct thread *td)
277 {
278 struct unpcb *unp;
279 int error;
280
281 UNP_LOCK();
282 unp = sotounpcb(so);
283 if (unp == NULL || unp->unp_vnode == NULL) {
284 UNP_UNLOCK();
285 return (EINVAL);
286 }
287 error = unp_listen(unp, td);
288 UNP_UNLOCK();
289 return (error);
290 }
291
292 static int
293 uipc_peeraddr(struct socket *so, struct sockaddr **nam)
294 {
295 struct unpcb *unp;
296 const struct sockaddr *sa;
297
298 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
299 UNP_LOCK();
300 unp = sotounpcb(so);
301 if (unp == NULL) {
302 UNP_UNLOCK();
303 free(*nam, M_SONAME);
304 *nam = NULL;
305 return (EINVAL);
306 }
307 if (unp->unp_conn != NULL && unp->unp_conn->unp_addr!= NULL)
308 sa = (struct sockaddr *) unp->unp_conn->unp_addr;
309 else {
310 /*
311 * XXX: It seems that this test always fails even when
312 * connection is established. So, this else clause is
313 * added as workaround to return PF_LOCAL sockaddr.
314 */
315 sa = &sun_noname;
316 }
317 bcopy(sa, *nam, sa->sa_len);
318 UNP_UNLOCK();
319 return (0);
320 }
321
322 static int
323 uipc_rcvd(struct socket *so, int flags)
324 {
325 struct unpcb *unp;
326 struct socket *so2;
327 u_long newhiwat;
328
329 UNP_LOCK();
330 unp = sotounpcb(so);
331 if (unp == NULL) {
332 UNP_UNLOCK();
333 return (EINVAL);
334 }
335 switch (so->so_type) {
336 case SOCK_DGRAM:
337 panic("uipc_rcvd DGRAM?");
338 /*NOTREACHED*/
339
340 case SOCK_STREAM:
341 if (unp->unp_conn == NULL)
342 break;
343 so2 = unp->unp_conn->unp_socket;
344 SOCKBUF_LOCK(&so2->so_snd);
345 SOCKBUF_LOCK(&so->so_rcv);
346 /*
347 * Adjust backpressure on sender
348 * and wakeup any waiting to write.
349 */
350 so2->so_snd.sb_mbmax += unp->unp_mbcnt - so->so_rcv.sb_mbcnt;
351 unp->unp_mbcnt = so->so_rcv.sb_mbcnt;
352 newhiwat = so2->so_snd.sb_hiwat + unp->unp_cc -
353 so->so_rcv.sb_cc;
354 (void)chgsbsize(so2->so_cred->cr_uidinfo, &so2->so_snd.sb_hiwat,
355 newhiwat, RLIM_INFINITY);
356 unp->unp_cc = so->so_rcv.sb_cc;
357 SOCKBUF_UNLOCK(&so->so_rcv);
358 sowwakeup_locked(so2);
359 break;
360
361 default:
362 panic("uipc_rcvd unknown socktype");
363 }
364 UNP_UNLOCK();
365 return (0);
366 }
367
368 /* pru_rcvoob is EOPNOTSUPP */
369
370 static int
371 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
372 struct mbuf *control, struct thread *td)
373 {
374 int error = 0;
375 struct unpcb *unp;
376 struct socket *so2;
377 u_long newhiwat;
378
379 unp = sotounpcb(so);
380 if (unp == NULL) {
381 error = EINVAL;
382 goto release;
383 }
384 if (flags & PRUS_OOB) {
385 error = EOPNOTSUPP;
386 goto release;
387 }
388
389 if (control != NULL && (error = unp_internalize(&control, td)))
390 goto release;
391
392 UNP_LOCK();
393 unp = sotounpcb(so);
394 if (unp == NULL) {
395 UNP_UNLOCK();
396 error = EINVAL;
397 goto dispose_release;
398 }
399
400 switch (so->so_type) {
401 case SOCK_DGRAM:
402 {
403 const struct sockaddr *from;
404
405 if (nam != NULL) {
406 if (unp->unp_conn != NULL) {
407 error = EISCONN;
408 break;
409 }
410 error = unp_connect(so, nam, td);
411 if (error)
412 break;
413 } else {
414 if (unp->unp_conn == NULL) {
415 error = ENOTCONN;
416 break;
417 }
418 }
419 so2 = unp->unp_conn->unp_socket;
420 if (unp->unp_addr != NULL)
421 from = (struct sockaddr *)unp->unp_addr;
422 else
423 from = &sun_noname;
424 SOCKBUF_LOCK(&so2->so_rcv);
425 if (sbappendaddr_locked(&so2->so_rcv, from, m, control)) {
426 sorwakeup_locked(so2);
427 m = NULL;
428 control = NULL;
429 } else {
430 SOCKBUF_UNLOCK(&so2->so_rcv);
431 error = ENOBUFS;
432 }
433 if (nam != NULL)
434 unp_disconnect(unp);
435 break;
436 }
437
438 case SOCK_STREAM:
439 /* Connect if not connected yet. */
440 /*
441 * Note: A better implementation would complain
442 * if not equal to the peer's address.
443 */
444 if ((so->so_state & SS_ISCONNECTED) == 0) {
445 if (nam != NULL) {
446 error = unp_connect(so, nam, td);
447 if (error)
448 break; /* XXX */
449 } else {
450 error = ENOTCONN;
451 break;
452 }
453 }
454
455 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
456 error = EPIPE;
457 break;
458 }
459 if (unp->unp_conn == NULL)
460 panic("uipc_send connected but no connection?");
461 so2 = unp->unp_conn->unp_socket;
462 SOCKBUF_LOCK(&so2->so_rcv);
463 /*
464 * Send to paired receive port, and then reduce
465 * send buffer hiwater marks to maintain backpressure.
466 * Wake up readers.
467 */
468 if (control != NULL) {
469 if (sbappendcontrol_locked(&so2->so_rcv, m, control))
470 control = NULL;
471 } else {
472 sbappend_locked(&so2->so_rcv, m);
473 }
474 so->so_snd.sb_mbmax -=
475 so2->so_rcv.sb_mbcnt - unp->unp_conn->unp_mbcnt;
476 unp->unp_conn->unp_mbcnt = so2->so_rcv.sb_mbcnt;
477 newhiwat = so->so_snd.sb_hiwat -
478 (so2->so_rcv.sb_cc - unp->unp_conn->unp_cc);
479 (void)chgsbsize(so->so_cred->cr_uidinfo, &so->so_snd.sb_hiwat,
480 newhiwat, RLIM_INFINITY);
481 unp->unp_conn->unp_cc = so2->so_rcv.sb_cc;
482 sorwakeup_locked(so2);
483 m = NULL;
484 break;
485
486 default:
487 panic("uipc_send unknown socktype");
488 }
489
490 /*
491 * SEND_EOF is equivalent to a SEND followed by
492 * a SHUTDOWN.
493 */
494 if (flags & PRUS_EOF) {
495 socantsendmore(so);
496 unp_shutdown(unp);
497 }
498 UNP_UNLOCK();
499
500 dispose_release:
501 if (control != NULL && error != 0)
502 unp_dispose(control);
503
504 release:
505 if (control != NULL)
506 m_freem(control);
507 if (m != NULL)
508 m_freem(m);
509 return (error);
510 }
511
512 static int
513 uipc_sense(struct socket *so, struct stat *sb)
514 {
515 struct unpcb *unp;
516 struct socket *so2;
517
518 UNP_LOCK();
519 unp = sotounpcb(so);
520 if (unp == NULL) {
521 UNP_UNLOCK();
522 return (EINVAL);
523 }
524 sb->st_blksize = so->so_snd.sb_hiwat;
525 if (so->so_type == SOCK_STREAM && unp->unp_conn != NULL) {
526 so2 = unp->unp_conn->unp_socket;
527 sb->st_blksize += so2->so_rcv.sb_cc;
528 }
529 sb->st_dev = NODEV;
530 if (unp->unp_ino == 0)
531 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino;
532 sb->st_ino = unp->unp_ino;
533 UNP_UNLOCK();
534 return (0);
535 }
536
537 static int
538 uipc_shutdown(struct socket *so)
539 {
540 struct unpcb *unp;
541
542 UNP_LOCK();
543 unp = sotounpcb(so);
544 if (unp == NULL) {
545 UNP_UNLOCK();
546 return (EINVAL);
547 }
548 socantsendmore(so);
549 unp_shutdown(unp);
550 UNP_UNLOCK();
551 return (0);
552 }
553
554 static int
555 uipc_sockaddr(struct socket *so, struct sockaddr **nam)
556 {
557 struct unpcb *unp;
558 const struct sockaddr *sa;
559
560 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
561 UNP_LOCK();
562 unp = sotounpcb(so);
563 if (unp == NULL) {
564 UNP_UNLOCK();
565 free(*nam, M_SONAME);
566 *nam = NULL;
567 return (EINVAL);
568 }
569 if (unp->unp_addr != NULL)
570 sa = (struct sockaddr *) unp->unp_addr;
571 else
572 sa = &sun_noname;
573 bcopy(sa, *nam, sa->sa_len);
574 UNP_UNLOCK();
575 return (0);
576 }
577
578 struct pr_usrreqs uipc_usrreqs = {
579 uipc_abort, uipc_accept, uipc_attach, uipc_bind, uipc_connect,
580 uipc_connect2, pru_control_notsupp, uipc_detach, uipc_disconnect,
581 uipc_listen, uipc_peeraddr, uipc_rcvd, pru_rcvoob_notsupp,
582 uipc_send, uipc_sense, uipc_shutdown, uipc_sockaddr,
583 sosend, soreceive, sopoll, pru_sosetlabel_null
584 };
585
586 int
587 uipc_ctloutput(so, sopt)
588 struct socket *so;
589 struct sockopt *sopt;
590 {
591 struct unpcb *unp;
592 struct xucred xu;
593 int error;
594
595 switch (sopt->sopt_dir) {
596 case SOPT_GET:
597 switch (sopt->sopt_name) {
598 case LOCAL_PEERCRED:
599 error = 0;
600 UNP_LOCK();
601 unp = sotounpcb(so);
602 if (unp == NULL) {
603 UNP_UNLOCK();
604 error = EINVAL;
605 break;
606 }
607 if (unp->unp_flags & UNP_HAVEPC)
608 xu = unp->unp_peercred;
609 else {
610 if (so->so_type == SOCK_STREAM)
611 error = ENOTCONN;
612 else
613 error = EINVAL;
614 }
615 UNP_UNLOCK();
616 if (error == 0)
617 error = sooptcopyout(sopt, &xu, sizeof(xu));
618 break;
619 default:
620 error = EOPNOTSUPP;
621 break;
622 }
623 break;
624 case SOPT_SET:
625 default:
626 error = EOPNOTSUPP;
627 break;
628 }
629 return (error);
630 }
631
632 /*
633 * Both send and receive buffers are allocated PIPSIZ bytes of buffering
634 * for stream sockets, although the total for sender and receiver is
635 * actually only PIPSIZ.
636 * Datagram sockets really use the sendspace as the maximum datagram size,
637 * and don't really want to reserve the sendspace. Their recvspace should
638 * be large enough for at least one max-size datagram plus address.
639 */
640 #ifndef PIPSIZ
641 #define PIPSIZ 8192
642 #endif
643 static u_long unpst_sendspace = PIPSIZ;
644 static u_long unpst_recvspace = PIPSIZ;
645 static u_long unpdg_sendspace = 2*1024; /* really max datagram size */
646 static u_long unpdg_recvspace = 4*1024;
647
648 static int unp_rights; /* file descriptors in flight */
649
650 SYSCTL_DECL(_net_local_stream);
651 SYSCTL_INT(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW,
652 &unpst_sendspace, 0, "");
653 SYSCTL_INT(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW,
654 &unpst_recvspace, 0, "");
655 SYSCTL_DECL(_net_local_dgram);
656 SYSCTL_INT(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW,
657 &unpdg_sendspace, 0, "");
658 SYSCTL_INT(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW,
659 &unpdg_recvspace, 0, "");
660 SYSCTL_DECL(_net_local);
661 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, "");
662
663 static int
664 unp_attach(so)
665 struct socket *so;
666 {
667 register struct unpcb *unp;
668 int error;
669
670 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
671 switch (so->so_type) {
672
673 case SOCK_STREAM:
674 error = soreserve(so, unpst_sendspace, unpst_recvspace);
675 break;
676
677 case SOCK_DGRAM:
678 error = soreserve(so, unpdg_sendspace, unpdg_recvspace);
679 break;
680
681 default:
682 panic("unp_attach");
683 }
684 if (error)
685 return (error);
686 }
687 unp = uma_zalloc(unp_zone, M_WAITOK);
688 if (unp == NULL)
689 return (ENOBUFS);
690 bzero(unp, sizeof *unp);
691 LIST_INIT(&unp->unp_refs);
692 unp->unp_socket = so;
693
694 UNP_LOCK();
695 unp->unp_gencnt = ++unp_gencnt;
696 unp_count++;
697 LIST_INSERT_HEAD(so->so_type == SOCK_DGRAM ? &unp_dhead
698 : &unp_shead, unp, unp_link);
699 so->so_pcb = unp;
700 UNP_UNLOCK();
701
702 return (0);
703 }
704
705 static void
706 unp_detach(unp)
707 register struct unpcb *unp;
708 {
709 struct vnode *vp;
710
711 UNP_LOCK_ASSERT();
712
713 LIST_REMOVE(unp, unp_link);
714 unp->unp_gencnt = ++unp_gencnt;
715 --unp_count;
716 if ((vp = unp->unp_vnode) != NULL) {
717 /*
718 * XXXRW: should v_socket be frobbed only while holding
719 * Giant?
720 */
721 unp->unp_vnode->v_socket = NULL;
722 unp->unp_vnode = NULL;
723 }
724 if (unp->unp_conn != NULL)
725 unp_disconnect(unp);
726 while (!LIST_EMPTY(&unp->unp_refs)) {
727 struct unpcb *ref = LIST_FIRST(&unp->unp_refs);
728 unp_drop(ref, ECONNRESET);
729 }
730 soisdisconnected(unp->unp_socket);
731 unp->unp_socket->so_pcb = NULL;
732 if (unp_rights) {
733 /*
734 * Normally the receive buffer is flushed later,
735 * in sofree, but if our receive buffer holds references
736 * to descriptors that are now garbage, we will dispose
737 * of those descriptor references after the garbage collector
738 * gets them (resulting in a "panic: closef: count < 0").
739 */
740 sorflush(unp->unp_socket);
741 unp_gc(); /* Will unlock UNP. */
742 } else
743 UNP_UNLOCK();
744 UNP_UNLOCK_ASSERT();
745 if (unp->unp_addr != NULL)
746 FREE(unp->unp_addr, M_SONAME);
747 uma_zfree(unp_zone, unp);
748 if (vp) {
749 mtx_lock(&Giant);
750 vrele(vp);
751 mtx_unlock(&Giant);
752 }
753 }
754
755 static int
756 unp_bind(unp, nam, td)
757 struct unpcb *unp;
758 struct sockaddr *nam;
759 struct thread *td;
760 {
761 struct sockaddr_un *soun = (struct sockaddr_un *)nam;
762 struct vnode *vp;
763 struct mount *mp;
764 struct vattr vattr;
765 int error, namelen;
766 struct nameidata nd;
767 char *buf;
768
769 UNP_LOCK_ASSERT();
770
771 /*
772 * XXXRW: This test-and-set of unp_vnode is non-atomic; the
773 * unlocked read here is fine, but the value of unp_vnode needs
774 * to be tested again after we do all the lookups to see if the
775 * pcb is still unbound?
776 */
777 if (unp->unp_vnode != NULL)
778 return (EINVAL);
779
780 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path);
781 if (namelen <= 0)
782 return (EINVAL);
783
784 UNP_UNLOCK();
785
786 buf = malloc(namelen + 1, M_TEMP, M_WAITOK);
787 strlcpy(buf, soun->sun_path, namelen + 1);
788
789 mtx_lock(&Giant);
790 restart:
791 mtx_assert(&Giant, MA_OWNED);
792 NDINIT(&nd, CREATE, NOFOLLOW | LOCKPARENT | SAVENAME, UIO_SYSSPACE,
793 buf, td);
794 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
795 error = namei(&nd);
796 if (error)
797 goto done;
798 vp = nd.ni_vp;
799 if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
800 NDFREE(&nd, NDF_ONLY_PNBUF);
801 if (nd.ni_dvp == vp)
802 vrele(nd.ni_dvp);
803 else
804 vput(nd.ni_dvp);
805 if (vp != NULL) {
806 vrele(vp);
807 error = EADDRINUSE;
808 goto done;
809 }
810 error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH);
811 if (error)
812 goto done;
813 goto restart;
814 }
815 VATTR_NULL(&vattr);
816 vattr.va_type = VSOCK;
817 vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask);
818 #ifdef MAC
819 error = mac_check_vnode_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd,
820 &vattr);
821 #endif
822 if (error == 0) {
823 VOP_LEASE(nd.ni_dvp, td, td->td_ucred, LEASE_WRITE);
824 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
825 }
826 NDFREE(&nd, NDF_ONLY_PNBUF);
827 vput(nd.ni_dvp);
828 if (error)
829 goto done;
830 vp = nd.ni_vp;
831 ASSERT_VOP_LOCKED(vp, "unp_bind");
832 soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK);
833 UNP_LOCK();
834 vp->v_socket = unp->unp_socket;
835 unp->unp_vnode = vp;
836 unp->unp_addr = soun;
837 UNP_UNLOCK();
838 VOP_UNLOCK(vp, 0, td);
839 vn_finished_write(mp);
840 done:
841 mtx_unlock(&Giant);
842 free(buf, M_TEMP);
843 UNP_LOCK();
844 return (error);
845 }
846
847 static int
848 unp_connect(so, nam, td)
849 struct socket *so;
850 struct sockaddr *nam;
851 struct thread *td;
852 {
853 register struct sockaddr_un *soun = (struct sockaddr_un *)nam;
854 register struct vnode *vp;
855 register struct socket *so2, *so3;
856 struct unpcb *unp, *unp2, *unp3;
857 int error, len;
858 struct nameidata nd;
859 char buf[SOCK_MAXADDRLEN];
860 struct sockaddr *sa;
861
862 UNP_LOCK_ASSERT();
863 unp = sotounpcb(so);
864
865 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
866 if (len <= 0)
867 return (EINVAL);
868 strlcpy(buf, soun->sun_path, len + 1);
869 UNP_UNLOCK();
870 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
871 mtx_lock(&Giant);
872 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, buf, td);
873 error = namei(&nd);
874 if (error)
875 vp = NULL;
876 else
877 vp = nd.ni_vp;
878 ASSERT_VOP_LOCKED(vp, "unp_connect");
879 NDFREE(&nd, NDF_ONLY_PNBUF);
880 if (error)
881 goto bad;
882
883 if (vp->v_type != VSOCK) {
884 error = ENOTSOCK;
885 goto bad;
886 }
887 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td);
888 if (error)
889 goto bad;
890 mtx_unlock(&Giant);
891 UNP_LOCK();
892 unp = sotounpcb(so);
893 if (unp == NULL) {
894 /*
895 * XXXRW: Temporary debugging printf.
896 */
897 printf("unp_connect(): lost race to another thread\n");
898 error = EINVAL;
899 goto bad2;
900 }
901 so2 = vp->v_socket;
902 if (so2 == NULL) {
903 error = ECONNREFUSED;
904 goto bad2;
905 }
906 if (so->so_type != so2->so_type) {
907 error = EPROTOTYPE;
908 goto bad2;
909 }
910 if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
911 if (so2->so_options & SO_ACCEPTCONN) {
912 /*
913 * NB: drop locks here so unp_attach is entered
914 * w/o locks; this avoids a recursive lock
915 * of the head and holding sleep locks across
916 * a (potentially) blocking malloc.
917 */
918 UNP_UNLOCK();
919 so3 = sonewconn(so2, 0);
920 UNP_LOCK();
921 } else
922 so3 = NULL;
923 if (so3 == NULL) {
924 error = ECONNREFUSED;
925 goto bad2;
926 }
927 unp = sotounpcb(so);
928 unp2 = sotounpcb(so2);
929 unp3 = sotounpcb(so3);
930 if (unp2->unp_addr != NULL) {
931 bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len);
932 unp3->unp_addr = (struct sockaddr_un *) sa;
933 sa = NULL;
934 }
935 /*
936 * unp_peercred management:
937 *
938 * The connecter's (client's) credentials are copied
939 * from its process structure at the time of connect()
940 * (which is now).
941 */
942 cru2x(td->td_ucred, &unp3->unp_peercred);
943 unp3->unp_flags |= UNP_HAVEPC;
944 /*
945 * The receiver's (server's) credentials are copied
946 * from the unp_peercred member of socket on which the
947 * former called listen(); unp_listen() cached that
948 * process's credentials at that time so we can use
949 * them now.
950 */
951 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED,
952 ("unp_connect: listener without cached peercred"));
953 memcpy(&unp->unp_peercred, &unp2->unp_peercred,
954 sizeof(unp->unp_peercred));
955 unp->unp_flags |= UNP_HAVEPC;
956 #ifdef MAC
957 SOCK_LOCK(so);
958 mac_set_socket_peer_from_socket(so, so3);
959 mac_set_socket_peer_from_socket(so3, so);
960 SOCK_UNLOCK(so);
961 #endif
962
963 so2 = so3;
964 }
965 error = unp_connect2(so, so2);
966 bad2:
967 UNP_UNLOCK();
968 mtx_lock(&Giant);
969 bad:
970 mtx_assert(&Giant, MA_OWNED);
971 if (vp != NULL)
972 vput(vp);
973 mtx_unlock(&Giant);
974 free(sa, M_SONAME);
975 UNP_LOCK();
976 return (error);
977 }
978
979 static int
980 unp_connect2(so, so2)
981 register struct socket *so;
982 register struct socket *so2;
983 {
984 register struct unpcb *unp = sotounpcb(so);
985 register struct unpcb *unp2;
986
987 UNP_LOCK_ASSERT();
988
989 if (so2->so_type != so->so_type)
990 return (EPROTOTYPE);
991 unp2 = sotounpcb(so2);
992 unp->unp_conn = unp2;
993 switch (so->so_type) {
994
995 case SOCK_DGRAM:
996 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
997 soisconnected(so);
998 break;
999
1000 case SOCK_STREAM:
1001 unp2->unp_conn = unp;
1002 soisconnected(so);
1003 soisconnected(so2);
1004 break;
1005
1006 default:
1007 panic("unp_connect2");
1008 }
1009 return (0);
1010 }
1011
1012 static void
1013 unp_disconnect(unp)
1014 struct unpcb *unp;
1015 {
1016 register struct unpcb *unp2 = unp->unp_conn;
1017 struct socket *so;
1018
1019 UNP_LOCK_ASSERT();
1020
1021 if (unp2 == NULL)
1022 return;
1023 unp->unp_conn = NULL;
1024 switch (unp->unp_socket->so_type) {
1025
1026 case SOCK_DGRAM:
1027 LIST_REMOVE(unp, unp_reflink);
1028 so = unp->unp_socket;
1029 SOCK_LOCK(so);
1030 so->so_state &= ~SS_ISCONNECTED;
1031 SOCK_UNLOCK(so);
1032 break;
1033
1034 case SOCK_STREAM:
1035 soisdisconnected(unp->unp_socket);
1036 unp2->unp_conn = NULL;
1037 soisdisconnected(unp2->unp_socket);
1038 break;
1039 }
1040 }
1041
1042 #ifdef notdef
1043 void
1044 unp_abort(unp)
1045 struct unpcb *unp;
1046 {
1047
1048 unp_detach(unp);
1049 UNP_UNLOCK_ASSERT();
1050 }
1051 #endif
1052
1053 /*
1054 * unp_pcblist() assumes that UNIX domain socket memory is never reclaimed
1055 * by the zone (UMA_ZONE_NOFREE), and as such potentially stale pointers
1056 * are safe to reference. It first scans the list of struct unpcb's to
1057 * generate a pointer list, then it rescans its list one entry at a time to
1058 * externalize and copyout. It checks the generation number to see if a
1059 * struct unpcb has been reused, and will skip it if so.
1060 */
1061 static int
1062 unp_pcblist(SYSCTL_HANDLER_ARGS)
1063 {
1064 int error, i, n;
1065 struct unpcb *unp, **unp_list;
1066 unp_gen_t gencnt;
1067 struct xunpgen *xug;
1068 struct unp_head *head;
1069 struct xunpcb *xu;
1070
1071 head = ((intptr_t)arg1 == SOCK_DGRAM ? &unp_dhead : &unp_shead);
1072
1073 /*
1074 * The process of preparing the PCB list is too time-consuming and
1075 * resource-intensive to repeat twice on every request.
1076 */
1077 if (req->oldptr == NULL) {
1078 n = unp_count;
1079 req->oldidx = 2 * (sizeof *xug)
1080 + (n + n/8) * sizeof(struct xunpcb);
1081 return (0);
1082 }
1083
1084 if (req->newptr != NULL)
1085 return (EPERM);
1086
1087 /*
1088 * OK, now we're committed to doing something.
1089 */
1090 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK);
1091 UNP_LOCK();
1092 gencnt = unp_gencnt;
1093 n = unp_count;
1094 UNP_UNLOCK();
1095
1096 xug->xug_len = sizeof *xug;
1097 xug->xug_count = n;
1098 xug->xug_gen = gencnt;
1099 xug->xug_sogen = so_gencnt;
1100 error = SYSCTL_OUT(req, xug, sizeof *xug);
1101 if (error) {
1102 free(xug, M_TEMP);
1103 return (error);
1104 }
1105
1106 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
1107
1108 UNP_LOCK();
1109 for (unp = LIST_FIRST(head), i = 0; unp && i < n;
1110 unp = LIST_NEXT(unp, unp_link)) {
1111 if (unp->unp_gencnt <= gencnt) {
1112 if (cr_cansee(req->td->td_ucred,
1113 unp->unp_socket->so_cred))
1114 continue;
1115 unp_list[i++] = unp;
1116 }
1117 }
1118 UNP_UNLOCK();
1119 n = i; /* in case we lost some during malloc */
1120
1121 error = 0;
1122 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO);
1123 for (i = 0; i < n; i++) {
1124 unp = unp_list[i];
1125 if (unp->unp_gencnt <= gencnt) {
1126 xu->xu_len = sizeof *xu;
1127 xu->xu_unpp = unp;
1128 /*
1129 * XXX - need more locking here to protect against
1130 * connect/disconnect races for SMP.
1131 */
1132 if (unp->unp_addr != NULL)
1133 bcopy(unp->unp_addr, &xu->xu_addr,
1134 unp->unp_addr->sun_len);
1135 if (unp->unp_conn != NULL &&
1136 unp->unp_conn->unp_addr != NULL)
1137 bcopy(unp->unp_conn->unp_addr,
1138 &xu->xu_caddr,
1139 unp->unp_conn->unp_addr->sun_len);
1140 bcopy(unp, &xu->xu_unp, sizeof *unp);
1141 sotoxsocket(unp->unp_socket, &xu->xu_socket);
1142 error = SYSCTL_OUT(req, xu, sizeof *xu);
1143 }
1144 }
1145 free(xu, M_TEMP);
1146 if (!error) {
1147 /*
1148 * Give the user an updated idea of our state.
1149 * If the generation differs from what we told
1150 * her before, she knows that something happened
1151 * while we were processing this request, and it
1152 * might be necessary to retry.
1153 */
1154 xug->xug_gen = unp_gencnt;
1155 xug->xug_sogen = so_gencnt;
1156 xug->xug_count = unp_count;
1157 error = SYSCTL_OUT(req, xug, sizeof *xug);
1158 }
1159 free(unp_list, M_TEMP);
1160 free(xug, M_TEMP);
1161 return (error);
1162 }
1163
1164 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLFLAG_RD,
1165 (caddr_t)(long)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
1166 "List of active local datagram sockets");
1167 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLFLAG_RD,
1168 (caddr_t)(long)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
1169 "List of active local stream sockets");
1170
1171 static void
1172 unp_shutdown(unp)
1173 struct unpcb *unp;
1174 {
1175 struct socket *so;
1176
1177 UNP_LOCK_ASSERT();
1178
1179 if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn &&
1180 (so = unp->unp_conn->unp_socket))
1181 socantrcvmore(so);
1182 }
1183
1184 static void
1185 unp_drop(unp, errno)
1186 struct unpcb *unp;
1187 int errno;
1188 {
1189 struct socket *so = unp->unp_socket;
1190
1191 UNP_LOCK_ASSERT();
1192
1193 so->so_error = errno;
1194 unp_disconnect(unp);
1195 }
1196
1197 #ifdef notdef
1198 void
1199 unp_drain()
1200 {
1201
1202 }
1203 #endif
1204
1205 static void
1206 unp_freerights(rp, fdcount)
1207 struct file **rp;
1208 int fdcount;
1209 {
1210 int i;
1211 struct file *fp;
1212
1213 for (i = 0; i < fdcount; i++) {
1214 fp = *rp;
1215 /*
1216 * zero the pointer before calling
1217 * unp_discard since it may end up
1218 * in unp_gc()..
1219 */
1220 *rp++ = 0;
1221 unp_discard(fp);
1222 }
1223 }
1224
1225 int
1226 unp_externalize(control, controlp)
1227 struct mbuf *control, **controlp;
1228 {
1229 struct thread *td = curthread; /* XXX */
1230 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1231 int i;
1232 int *fdp;
1233 struct file **rp;
1234 struct file *fp;
1235 void *data;
1236 socklen_t clen = control->m_len, datalen;
1237 int error, newfds;
1238 int f;
1239 u_int newlen;
1240
1241 UNP_UNLOCK_ASSERT();
1242
1243 error = 0;
1244 if (controlp != NULL) /* controlp == NULL => free control messages */
1245 *controlp = NULL;
1246
1247 while (cm != NULL) {
1248 if (sizeof(*cm) > clen || cm->cmsg_len > clen) {
1249 error = EINVAL;
1250 break;
1251 }
1252
1253 data = CMSG_DATA(cm);
1254 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1255
1256 if (cm->cmsg_level == SOL_SOCKET
1257 && cm->cmsg_type == SCM_RIGHTS) {
1258 newfds = datalen / sizeof(struct file *);
1259 rp = data;
1260
1261 /* If we're not outputting the descriptors free them. */
1262 if (error || controlp == NULL) {
1263 unp_freerights(rp, newfds);
1264 goto next;
1265 }
1266 FILEDESC_LOCK(td->td_proc->p_fd);
1267 /* if the new FD's will not fit free them. */
1268 if (!fdavail(td, newfds)) {
1269 FILEDESC_UNLOCK(td->td_proc->p_fd);
1270 error = EMSGSIZE;
1271 unp_freerights(rp, newfds);
1272 goto next;
1273 }
1274 /*
1275 * now change each pointer to an fd in the global
1276 * table to an integer that is the index to the
1277 * local fd table entry that we set up to point
1278 * to the global one we are transferring.
1279 */
1280 newlen = newfds * sizeof(int);
1281 *controlp = sbcreatecontrol(NULL, newlen,
1282 SCM_RIGHTS, SOL_SOCKET);
1283 if (*controlp == NULL) {
1284 FILEDESC_UNLOCK(td->td_proc->p_fd);
1285 error = E2BIG;
1286 unp_freerights(rp, newfds);
1287 goto next;
1288 }
1289
1290 fdp = (int *)
1291 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1292 for (i = 0; i < newfds; i++) {
1293 if (fdalloc(td, 0, &f))
1294 panic("unp_externalize fdalloc failed");
1295 fp = *rp++;
1296 td->td_proc->p_fd->fd_ofiles[f] = fp;
1297 FILE_LOCK(fp);
1298 fp->f_msgcount--;
1299 FILE_UNLOCK(fp);
1300 unp_rights--;
1301 *fdp++ = f;
1302 }
1303 FILEDESC_UNLOCK(td->td_proc->p_fd);
1304 } else { /* We can just copy anything else across */
1305 if (error || controlp == NULL)
1306 goto next;
1307 *controlp = sbcreatecontrol(NULL, datalen,
1308 cm->cmsg_type, cm->cmsg_level);
1309 if (*controlp == NULL) {
1310 error = ENOBUFS;
1311 goto next;
1312 }
1313 bcopy(data,
1314 CMSG_DATA(mtod(*controlp, struct cmsghdr *)),
1315 datalen);
1316 }
1317
1318 controlp = &(*controlp)->m_next;
1319
1320 next:
1321 if (CMSG_SPACE(datalen) < clen) {
1322 clen -= CMSG_SPACE(datalen);
1323 cm = (struct cmsghdr *)
1324 ((caddr_t)cm + CMSG_SPACE(datalen));
1325 } else {
1326 clen = 0;
1327 cm = NULL;
1328 }
1329 }
1330
1331 m_freem(control);
1332
1333 return (error);
1334 }
1335
1336 void
1337 unp_init(void)
1338 {
1339 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL,
1340 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
1341 if (unp_zone == NULL)
1342 panic("unp_init");
1343 uma_zone_set_max(unp_zone, nmbclusters);
1344 LIST_INIT(&unp_dhead);
1345 LIST_INIT(&unp_shead);
1346
1347 UNP_LOCK_INIT();
1348 }
1349
1350 static int
1351 unp_internalize(controlp, td)
1352 struct mbuf **controlp;
1353 struct thread *td;
1354 {
1355 struct mbuf *control = *controlp;
1356 struct proc *p = td->td_proc;
1357 struct filedesc *fdescp = p->p_fd;
1358 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1359 struct cmsgcred *cmcred;
1360 struct file **rp;
1361 struct file *fp;
1362 struct timeval *tv;
1363 int i, fd, *fdp;
1364 void *data;
1365 socklen_t clen = control->m_len, datalen;
1366 int error, oldfds;
1367 u_int newlen;
1368
1369 UNP_UNLOCK_ASSERT();
1370
1371 error = 0;
1372 *controlp = NULL;
1373
1374 while (cm != NULL) {
1375 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET
1376 || cm->cmsg_len > clen) {
1377 error = EINVAL;
1378 goto out;
1379 }
1380
1381 data = CMSG_DATA(cm);
1382 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1383
1384 switch (cm->cmsg_type) {
1385 /*
1386 * Fill in credential information.
1387 */
1388 case SCM_CREDS:
1389 *controlp = sbcreatecontrol(NULL, sizeof(*cmcred),
1390 SCM_CREDS, SOL_SOCKET);
1391 if (*controlp == NULL) {
1392 error = ENOBUFS;
1393 goto out;
1394 }
1395
1396 cmcred = (struct cmsgcred *)
1397 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1398 cmcred->cmcred_pid = p->p_pid;
1399 cmcred->cmcred_uid = td->td_ucred->cr_ruid;
1400 cmcred->cmcred_gid = td->td_ucred->cr_rgid;
1401 cmcred->cmcred_euid = td->td_ucred->cr_uid;
1402 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups,
1403 CMGROUP_MAX);
1404 for (i = 0; i < cmcred->cmcred_ngroups; i++)
1405 cmcred->cmcred_groups[i] =
1406 td->td_ucred->cr_groups[i];
1407 break;
1408
1409 case SCM_RIGHTS:
1410 oldfds = datalen / sizeof (int);
1411 /*
1412 * check that all the FDs passed in refer to legal files
1413 * If not, reject the entire operation.
1414 */
1415 fdp = data;
1416 FILEDESC_LOCK(fdescp);
1417 for (i = 0; i < oldfds; i++) {
1418 fd = *fdp++;
1419 if ((unsigned)fd >= fdescp->fd_nfiles ||
1420 fdescp->fd_ofiles[fd] == NULL) {
1421 FILEDESC_UNLOCK(fdescp);
1422 error = EBADF;
1423 goto out;
1424 }
1425 fp = fdescp->fd_ofiles[fd];
1426 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) {
1427 FILEDESC_UNLOCK(fdescp);
1428 error = EOPNOTSUPP;
1429 goto out;
1430 }
1431
1432 }
1433 /*
1434 * Now replace the integer FDs with pointers to
1435 * the associated global file table entry..
1436 */
1437 newlen = oldfds * sizeof(struct file *);
1438 *controlp = sbcreatecontrol(NULL, newlen,
1439 SCM_RIGHTS, SOL_SOCKET);
1440 if (*controlp == NULL) {
1441 FILEDESC_UNLOCK(fdescp);
1442 error = E2BIG;
1443 goto out;
1444 }
1445
1446 fdp = data;
1447 rp = (struct file **)
1448 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1449 for (i = 0; i < oldfds; i++) {
1450 fp = fdescp->fd_ofiles[*fdp++];
1451 *rp++ = fp;
1452 FILE_LOCK(fp);
1453 fp->f_count++;
1454 fp->f_msgcount++;
1455 FILE_UNLOCK(fp);
1456 unp_rights++;
1457 }
1458 FILEDESC_UNLOCK(fdescp);
1459 break;
1460
1461 case SCM_TIMESTAMP:
1462 *controlp = sbcreatecontrol(NULL, sizeof(*tv),
1463 SCM_TIMESTAMP, SOL_SOCKET);
1464 if (*controlp == NULL) {
1465 error = ENOBUFS;
1466 goto out;
1467 }
1468 tv = (struct timeval *)
1469 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1470 microtime(tv);
1471 break;
1472
1473 default:
1474 error = EINVAL;
1475 goto out;
1476 }
1477
1478 controlp = &(*controlp)->m_next;
1479
1480 if (CMSG_SPACE(datalen) < clen) {
1481 clen -= CMSG_SPACE(datalen);
1482 cm = (struct cmsghdr *)
1483 ((caddr_t)cm + CMSG_SPACE(datalen));
1484 } else {
1485 clen = 0;
1486 cm = NULL;
1487 }
1488 }
1489
1490 out:
1491 m_freem(control);
1492
1493 return (error);
1494 }
1495
1496 /*
1497 * unp_defer is thread-local during garbage collection, and does not require
1498 * explicit synchronization. unp_gcing prevents other threads from entering
1499 * garbage collection, and perhaps should be an sx lock instead.
1500 */
1501 static int unp_defer, unp_gcing;
1502
1503 static void
1504 unp_gc()
1505 {
1506 register struct file *fp, *nextfp;
1507 register struct socket *so;
1508 struct file **extra_ref, **fpp;
1509 int nunref, i;
1510 int nfiles_snap;
1511 int nfiles_slack = 20;
1512
1513 UNP_LOCK_ASSERT();
1514
1515 if (unp_gcing) {
1516 UNP_UNLOCK();
1517 return;
1518 }
1519 unp_gcing = 1;
1520 unp_defer = 0;
1521 UNP_UNLOCK();
1522 /*
1523 * before going through all this, set all FDs to
1524 * be NOT defered and NOT externally accessible
1525 */
1526 sx_slock(&filelist_lock);
1527 LIST_FOREACH(fp, &filehead, f_list)
1528 fp->f_gcflag &= ~(FMARK|FDEFER);
1529 do {
1530 LIST_FOREACH(fp, &filehead, f_list) {
1531 FILE_LOCK(fp);
1532 /*
1533 * If the file is not open, skip it
1534 */
1535 if (fp->f_count == 0) {
1536 FILE_UNLOCK(fp);
1537 continue;
1538 }
1539 /*
1540 * If we already marked it as 'defer' in a
1541 * previous pass, then try process it this time
1542 * and un-mark it
1543 */
1544 if (fp->f_gcflag & FDEFER) {
1545 fp->f_gcflag &= ~FDEFER;
1546 unp_defer--;
1547 } else {
1548 /*
1549 * if it's not defered, then check if it's
1550 * already marked.. if so skip it
1551 */
1552 if (fp->f_gcflag & FMARK) {
1553 FILE_UNLOCK(fp);
1554 continue;
1555 }
1556 /*
1557 * If all references are from messages
1558 * in transit, then skip it. it's not
1559 * externally accessible.
1560 */
1561 if (fp->f_count == fp->f_msgcount) {
1562 FILE_UNLOCK(fp);
1563 continue;
1564 }
1565 /*
1566 * If it got this far then it must be
1567 * externally accessible.
1568 */
1569 fp->f_gcflag |= FMARK;
1570 }
1571 /*
1572 * either it was defered, or it is externally
1573 * accessible and not already marked so.
1574 * Now check if it is possibly one of OUR sockets.
1575 */
1576 if (fp->f_type != DTYPE_SOCKET ||
1577 (so = fp->f_data) == NULL) {
1578 FILE_UNLOCK(fp);
1579 continue;
1580 }
1581 FILE_UNLOCK(fp);
1582 if (so->so_proto->pr_domain != &localdomain ||
1583 (so->so_proto->pr_flags&PR_RIGHTS) == 0)
1584 continue;
1585 #ifdef notdef
1586 if (so->so_rcv.sb_flags & SB_LOCK) {
1587 /*
1588 * This is problematical; it's not clear
1589 * we need to wait for the sockbuf to be
1590 * unlocked (on a uniprocessor, at least),
1591 * and it's also not clear what to do
1592 * if sbwait returns an error due to receipt
1593 * of a signal. If sbwait does return
1594 * an error, we'll go into an infinite
1595 * loop. Delete all of this for now.
1596 */
1597 (void) sbwait(&so->so_rcv);
1598 goto restart;
1599 }
1600 #endif
1601 /*
1602 * So, Ok, it's one of our sockets and it IS externally
1603 * accessible (or was defered). Now we look
1604 * to see if we hold any file descriptors in its
1605 * message buffers. Follow those links and mark them
1606 * as accessible too.
1607 */
1608 SOCKBUF_LOCK(&so->so_rcv);
1609 unp_scan(so->so_rcv.sb_mb, unp_mark);
1610 SOCKBUF_UNLOCK(&so->so_rcv);
1611 }
1612 } while (unp_defer);
1613 sx_sunlock(&filelist_lock);
1614 /*
1615 * We grab an extra reference to each of the file table entries
1616 * that are not otherwise accessible and then free the rights
1617 * that are stored in messages on them.
1618 *
1619 * The bug in the orginal code is a little tricky, so I'll describe
1620 * what's wrong with it here.
1621 *
1622 * It is incorrect to simply unp_discard each entry for f_msgcount
1623 * times -- consider the case of sockets A and B that contain
1624 * references to each other. On a last close of some other socket,
1625 * we trigger a gc since the number of outstanding rights (unp_rights)
1626 * is non-zero. If during the sweep phase the gc code un_discards,
1627 * we end up doing a (full) closef on the descriptor. A closef on A
1628 * results in the following chain. Closef calls soo_close, which
1629 * calls soclose. Soclose calls first (through the switch
1630 * uipc_usrreq) unp_detach, which re-invokes unp_gc. Unp_gc simply
1631 * returns because the previous instance had set unp_gcing, and
1632 * we return all the way back to soclose, which marks the socket
1633 * with SS_NOFDREF, and then calls sofree. Sofree calls sorflush
1634 * to free up the rights that are queued in messages on the socket A,
1635 * i.e., the reference on B. The sorflush calls via the dom_dispose
1636 * switch unp_dispose, which unp_scans with unp_discard. This second
1637 * instance of unp_discard just calls closef on B.
1638 *
1639 * Well, a similar chain occurs on B, resulting in a sorflush on B,
1640 * which results in another closef on A. Unfortunately, A is already
1641 * being closed, and the descriptor has already been marked with
1642 * SS_NOFDREF, and soclose panics at this point.
1643 *
1644 * Here, we first take an extra reference to each inaccessible
1645 * descriptor. Then, we call sorflush ourself, since we know
1646 * it is a Unix domain socket anyhow. After we destroy all the
1647 * rights carried in messages, we do a last closef to get rid
1648 * of our extra reference. This is the last close, and the
1649 * unp_detach etc will shut down the socket.
1650 *
1651 * 91/09/19, bsy@cs.cmu.edu
1652 */
1653 again:
1654 nfiles_snap = nfiles + nfiles_slack; /* some slack */
1655 extra_ref = malloc(nfiles_snap * sizeof(struct file *), M_TEMP,
1656 M_WAITOK);
1657 sx_slock(&filelist_lock);
1658 if (nfiles_snap < nfiles) {
1659 sx_sunlock(&filelist_lock);
1660 free(extra_ref, M_TEMP);
1661 nfiles_slack += 20;
1662 goto again;
1663 }
1664 for (nunref = 0, fp = LIST_FIRST(&filehead), fpp = extra_ref;
1665 fp != NULL; fp = nextfp) {
1666 nextfp = LIST_NEXT(fp, f_list);
1667 FILE_LOCK(fp);
1668 /*
1669 * If it's not open, skip it
1670 */
1671 if (fp->f_count == 0) {
1672 FILE_UNLOCK(fp);
1673 continue;
1674 }
1675 /*
1676 * If all refs are from msgs, and it's not marked accessible
1677 * then it must be referenced from some unreachable cycle
1678 * of (shut-down) FDs, so include it in our
1679 * list of FDs to remove
1680 */
1681 if (fp->f_count == fp->f_msgcount && !(fp->f_gcflag & FMARK)) {
1682 *fpp++ = fp;
1683 nunref++;
1684 fp->f_count++;
1685 }
1686 FILE_UNLOCK(fp);
1687 }
1688 sx_sunlock(&filelist_lock);
1689 /*
1690 * for each FD on our hit list, do the following two things
1691 */
1692 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) {
1693 struct file *tfp = *fpp;
1694 FILE_LOCK(tfp);
1695 if (tfp->f_type == DTYPE_SOCKET &&
1696 tfp->f_data != NULL) {
1697 FILE_UNLOCK(tfp);
1698 sorflush(tfp->f_data);
1699 } else {
1700 FILE_UNLOCK(tfp);
1701 }
1702 }
1703 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp)
1704 closef(*fpp, (struct thread *) NULL);
1705 free(extra_ref, M_TEMP);
1706 unp_gcing = 0;
1707
1708 UNP_UNLOCK_ASSERT();
1709 }
1710
1711 void
1712 unp_dispose(m)
1713 struct mbuf *m;
1714 {
1715
1716 if (m)
1717 unp_scan(m, unp_discard);
1718 }
1719
1720 static int
1721 unp_listen(unp, td)
1722 struct unpcb *unp;
1723 struct thread *td;
1724 {
1725 UNP_LOCK_ASSERT();
1726
1727 /*
1728 * XXXRW: Why populate the local peer cred with our own credential?
1729 */
1730 cru2x(td->td_ucred, &unp->unp_peercred);
1731 unp->unp_flags |= UNP_HAVEPCCACHED;
1732 return (0);
1733 }
1734
1735 static void
1736 unp_scan(m0, op)
1737 register struct mbuf *m0;
1738 void (*op)(struct file *);
1739 {
1740 struct mbuf *m;
1741 struct file **rp;
1742 struct cmsghdr *cm;
1743 void *data;
1744 int i;
1745 socklen_t clen, datalen;
1746 int qfds;
1747
1748 while (m0 != NULL) {
1749 for (m = m0; m; m = m->m_next) {
1750 if (m->m_type != MT_CONTROL)
1751 continue;
1752
1753 cm = mtod(m, struct cmsghdr *);
1754 clen = m->m_len;
1755
1756 while (cm != NULL) {
1757 if (sizeof(*cm) > clen || cm->cmsg_len > clen)
1758 break;
1759
1760 data = CMSG_DATA(cm);
1761 datalen = (caddr_t)cm + cm->cmsg_len
1762 - (caddr_t)data;
1763
1764 if (cm->cmsg_level == SOL_SOCKET &&
1765 cm->cmsg_type == SCM_RIGHTS) {
1766 qfds = datalen / sizeof (struct file *);
1767 rp = data;
1768 for (i = 0; i < qfds; i++)
1769 (*op)(*rp++);
1770 }
1771
1772 if (CMSG_SPACE(datalen) < clen) {
1773 clen -= CMSG_SPACE(datalen);
1774 cm = (struct cmsghdr *)
1775 ((caddr_t)cm + CMSG_SPACE(datalen));
1776 } else {
1777 clen = 0;
1778 cm = NULL;
1779 }
1780 }
1781 }
1782 m0 = m0->m_act;
1783 }
1784 }
1785
1786 static void
1787 unp_mark(fp)
1788 struct file *fp;
1789 {
1790 if (fp->f_gcflag & FMARK)
1791 return;
1792 unp_defer++;
1793 fp->f_gcflag |= (FMARK|FDEFER);
1794 }
1795
1796 static void
1797 unp_discard(fp)
1798 struct file *fp;
1799 {
1800 FILE_LOCK(fp);
1801 fp->f_msgcount--;
1802 unp_rights--;
1803 FILE_UNLOCK(fp);
1804 (void) closef(fp, (struct thread *)NULL);
1805 }
Cache object: 0b0b3b54af566f355d856ec73be605de
|