1 /*-
2 * Copyright 2004-2005 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/5.4/sys/kern/uipc_usrreq.c 145979 2005-05-07 03:58:25Z 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 socket *, 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(so, 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 SOCKBUF_LOCK(&so->so_snd);
456 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
457 SOCKBUF_UNLOCK(&so->so_snd);
458 error = EPIPE;
459 break;
460 }
461 if (unp->unp_conn == NULL)
462 panic("uipc_send connected but no connection?");
463 so2 = unp->unp_conn->unp_socket;
464 SOCKBUF_LOCK(&so2->so_rcv);
465 /*
466 * Send to paired receive port, and then reduce
467 * send buffer hiwater marks to maintain backpressure.
468 * Wake up readers.
469 */
470 if (control != NULL) {
471 if (sbappendcontrol_locked(&so2->so_rcv, m, control))
472 control = NULL;
473 } else {
474 sbappend_locked(&so2->so_rcv, m);
475 }
476 so->so_snd.sb_mbmax -=
477 so2->so_rcv.sb_mbcnt - unp->unp_conn->unp_mbcnt;
478 unp->unp_conn->unp_mbcnt = so2->so_rcv.sb_mbcnt;
479 newhiwat = so->so_snd.sb_hiwat -
480 (so2->so_rcv.sb_cc - unp->unp_conn->unp_cc);
481 (void)chgsbsize(so->so_cred->cr_uidinfo, &so->so_snd.sb_hiwat,
482 newhiwat, RLIM_INFINITY);
483 SOCKBUF_UNLOCK(&so->so_snd);
484 unp->unp_conn->unp_cc = so2->so_rcv.sb_cc;
485 sorwakeup_locked(so2);
486 m = NULL;
487 break;
488
489 default:
490 panic("uipc_send unknown socktype");
491 }
492
493 /*
494 * SEND_EOF is equivalent to a SEND followed by
495 * a SHUTDOWN.
496 */
497 if (flags & PRUS_EOF) {
498 socantsendmore(so);
499 unp_shutdown(unp);
500 }
501 UNP_UNLOCK();
502
503 dispose_release:
504 if (control != NULL && error != 0)
505 unp_dispose(control);
506
507 release:
508 if (control != NULL)
509 m_freem(control);
510 if (m != NULL)
511 m_freem(m);
512 return (error);
513 }
514
515 static int
516 uipc_sense(struct socket *so, struct stat *sb)
517 {
518 struct unpcb *unp;
519 struct socket *so2;
520
521 UNP_LOCK();
522 unp = sotounpcb(so);
523 if (unp == NULL) {
524 UNP_UNLOCK();
525 return (EINVAL);
526 }
527 sb->st_blksize = so->so_snd.sb_hiwat;
528 if (so->so_type == SOCK_STREAM && unp->unp_conn != NULL) {
529 so2 = unp->unp_conn->unp_socket;
530 sb->st_blksize += so2->so_rcv.sb_cc;
531 }
532 sb->st_dev = NODEV;
533 if (unp->unp_ino == 0)
534 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino;
535 sb->st_ino = unp->unp_ino;
536 UNP_UNLOCK();
537 return (0);
538 }
539
540 static int
541 uipc_shutdown(struct socket *so)
542 {
543 struct unpcb *unp;
544
545 UNP_LOCK();
546 unp = sotounpcb(so);
547 if (unp == NULL) {
548 UNP_UNLOCK();
549 return (EINVAL);
550 }
551 socantsendmore(so);
552 unp_shutdown(unp);
553 UNP_UNLOCK();
554 return (0);
555 }
556
557 static int
558 uipc_sockaddr(struct socket *so, struct sockaddr **nam)
559 {
560 struct unpcb *unp;
561 const struct sockaddr *sa;
562
563 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
564 UNP_LOCK();
565 unp = sotounpcb(so);
566 if (unp == NULL) {
567 UNP_UNLOCK();
568 free(*nam, M_SONAME);
569 *nam = NULL;
570 return (EINVAL);
571 }
572 if (unp->unp_addr != NULL)
573 sa = (struct sockaddr *) unp->unp_addr;
574 else
575 sa = &sun_noname;
576 bcopy(sa, *nam, sa->sa_len);
577 UNP_UNLOCK();
578 return (0);
579 }
580
581 struct pr_usrreqs uipc_usrreqs = {
582 uipc_abort, uipc_accept, uipc_attach, uipc_bind, uipc_connect,
583 uipc_connect2, pru_control_notsupp, uipc_detach, uipc_disconnect,
584 uipc_listen, uipc_peeraddr, uipc_rcvd, pru_rcvoob_notsupp,
585 uipc_send, uipc_sense, uipc_shutdown, uipc_sockaddr,
586 sosend, soreceive, sopoll, pru_sosetlabel_null
587 };
588
589 int
590 uipc_ctloutput(struct socket *so, struct sockopt *sopt)
591 {
592 struct unpcb *unp;
593 struct xucred xu;
594 int error;
595
596 switch (sopt->sopt_dir) {
597 case SOPT_GET:
598 switch (sopt->sopt_name) {
599 case LOCAL_PEERCRED:
600 error = 0;
601 UNP_LOCK();
602 unp = sotounpcb(so);
603 if (unp == NULL) {
604 UNP_UNLOCK();
605 error = EINVAL;
606 break;
607 }
608 if (unp->unp_flags & UNP_HAVEPC)
609 xu = unp->unp_peercred;
610 else {
611 if (so->so_type == SOCK_STREAM)
612 error = ENOTCONN;
613 else
614 error = EINVAL;
615 }
616 UNP_UNLOCK();
617 if (error == 0)
618 error = sooptcopyout(sopt, &xu, sizeof(xu));
619 break;
620 default:
621 error = EOPNOTSUPP;
622 break;
623 }
624 break;
625 case SOPT_SET:
626 default:
627 error = EOPNOTSUPP;
628 break;
629 }
630 return (error);
631 }
632
633 /*
634 * Both send and receive buffers are allocated PIPSIZ bytes of buffering
635 * for stream sockets, although the total for sender and receiver is
636 * actually only PIPSIZ.
637 * Datagram sockets really use the sendspace as the maximum datagram size,
638 * and don't really want to reserve the sendspace. Their recvspace should
639 * be large enough for at least one max-size datagram plus address.
640 */
641 #ifndef PIPSIZ
642 #define PIPSIZ 8192
643 #endif
644 static u_long unpst_sendspace = PIPSIZ;
645 static u_long unpst_recvspace = PIPSIZ;
646 static u_long unpdg_sendspace = 2*1024; /* really max datagram size */
647 static u_long unpdg_recvspace = 4*1024;
648
649 static int unp_rights; /* file descriptors in flight */
650
651 SYSCTL_DECL(_net_local_stream);
652 SYSCTL_INT(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW,
653 &unpst_sendspace, 0, "");
654 SYSCTL_INT(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW,
655 &unpst_recvspace, 0, "");
656 SYSCTL_DECL(_net_local_dgram);
657 SYSCTL_INT(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW,
658 &unpdg_sendspace, 0, "");
659 SYSCTL_INT(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW,
660 &unpdg_recvspace, 0, "");
661 SYSCTL_DECL(_net_local);
662 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, "");
663
664 static int
665 unp_attach(struct socket *so)
666 {
667 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 | M_ZERO);
688 if (unp == NULL)
689 return (ENOBUFS);
690 LIST_INIT(&unp->unp_refs);
691 unp->unp_socket = so;
692 so->so_pcb = unp;
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 UNP_UNLOCK();
700
701 return (0);
702 }
703
704 static void
705 unp_detach(struct unpcb *unp)
706 {
707 struct vnode *vp;
708
709 UNP_LOCK_ASSERT();
710
711 LIST_REMOVE(unp, unp_link);
712 unp->unp_gencnt = ++unp_gencnt;
713 --unp_count;
714 if ((vp = unp->unp_vnode) != NULL) {
715 /*
716 * XXXRW: should v_socket be frobbed only while holding
717 * Giant?
718 */
719 unp->unp_vnode->v_socket = NULL;
720 unp->unp_vnode = NULL;
721 }
722 if (unp->unp_conn != NULL)
723 unp_disconnect(unp);
724 while (!LIST_EMPTY(&unp->unp_refs)) {
725 struct unpcb *ref = LIST_FIRST(&unp->unp_refs);
726 unp_drop(ref, ECONNRESET);
727 }
728 soisdisconnected(unp->unp_socket);
729 unp->unp_socket->so_pcb = NULL;
730 if (unp_rights) {
731 /*
732 * Normally the receive buffer is flushed later,
733 * in sofree, but if our receive buffer holds references
734 * to descriptors that are now garbage, we will dispose
735 * of those descriptor references after the garbage collector
736 * gets them (resulting in a "panic: closef: count < 0").
737 */
738 sorflush(unp->unp_socket);
739 unp_gc(); /* Will unlock UNP. */
740 } else
741 UNP_UNLOCK();
742 UNP_UNLOCK_ASSERT();
743 if (unp->unp_addr != NULL)
744 FREE(unp->unp_addr, M_SONAME);
745 uma_zfree(unp_zone, unp);
746 if (vp) {
747 mtx_lock(&Giant);
748 vrele(vp);
749 mtx_unlock(&Giant);
750 }
751 }
752
753 static int
754 unp_bind(struct unpcb *unp, struct sockaddr *nam, struct thread *td)
755 {
756 struct sockaddr_un *soun = (struct sockaddr_un *)nam;
757 struct vnode *vp;
758 struct mount *mp;
759 struct vattr vattr;
760 int error, namelen;
761 struct nameidata nd;
762 char *buf;
763
764 UNP_LOCK_ASSERT();
765
766 /*
767 * XXXRW: This test-and-set of unp_vnode is non-atomic; the
768 * unlocked read here is fine, but the value of unp_vnode needs
769 * to be tested again after we do all the lookups to see if the
770 * pcb is still unbound?
771 */
772 if (unp->unp_vnode != NULL)
773 return (EINVAL);
774
775 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path);
776 if (namelen <= 0)
777 return (EINVAL);
778
779 UNP_UNLOCK();
780
781 buf = malloc(namelen + 1, M_TEMP, M_WAITOK);
782 strlcpy(buf, soun->sun_path, namelen + 1);
783
784 mtx_lock(&Giant);
785 restart:
786 mtx_assert(&Giant, MA_OWNED);
787 NDINIT(&nd, CREATE, NOFOLLOW | LOCKPARENT | SAVENAME, UIO_SYSSPACE,
788 buf, td);
789 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
790 error = namei(&nd);
791 if (error)
792 goto done;
793 vp = nd.ni_vp;
794 if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
795 NDFREE(&nd, NDF_ONLY_PNBUF);
796 if (nd.ni_dvp == vp)
797 vrele(nd.ni_dvp);
798 else
799 vput(nd.ni_dvp);
800 if (vp != NULL) {
801 vrele(vp);
802 error = EADDRINUSE;
803 goto done;
804 }
805 error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH);
806 if (error)
807 goto done;
808 goto restart;
809 }
810 VATTR_NULL(&vattr);
811 vattr.va_type = VSOCK;
812 vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask);
813 #ifdef MAC
814 error = mac_check_vnode_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd,
815 &vattr);
816 #endif
817 if (error == 0) {
818 VOP_LEASE(nd.ni_dvp, td, td->td_ucred, LEASE_WRITE);
819 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
820 }
821 NDFREE(&nd, NDF_ONLY_PNBUF);
822 vput(nd.ni_dvp);
823 if (error) {
824 vn_finished_write(mp);
825 goto done;
826 }
827 vp = nd.ni_vp;
828 ASSERT_VOP_LOCKED(vp, "unp_bind");
829 soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK);
830 UNP_LOCK();
831 vp->v_socket = unp->unp_socket;
832 unp->unp_vnode = vp;
833 unp->unp_addr = soun;
834 UNP_UNLOCK();
835 VOP_UNLOCK(vp, 0, td);
836 vn_finished_write(mp);
837 done:
838 mtx_unlock(&Giant);
839 free(buf, M_TEMP);
840 UNP_LOCK();
841 return (error);
842 }
843
844 static int
845 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
846 {
847 struct sockaddr_un *soun = (struct sockaddr_un *)nam;
848 struct vnode *vp;
849 struct socket *so2, *so3;
850 struct unpcb *unp, *unp2, *unp3;
851 int error, len;
852 struct nameidata nd;
853 char buf[SOCK_MAXADDRLEN];
854 struct sockaddr *sa;
855
856 UNP_LOCK_ASSERT();
857 unp = sotounpcb(so);
858
859 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
860 if (len <= 0)
861 return (EINVAL);
862 strlcpy(buf, soun->sun_path, len + 1);
863 UNP_UNLOCK();
864 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
865 mtx_lock(&Giant);
866 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, buf, td);
867 error = namei(&nd);
868 if (error)
869 vp = NULL;
870 else
871 vp = nd.ni_vp;
872 ASSERT_VOP_LOCKED(vp, "unp_connect");
873 NDFREE(&nd, NDF_ONLY_PNBUF);
874 if (error)
875 goto bad;
876
877 if (vp->v_type != VSOCK) {
878 error = ENOTSOCK;
879 goto bad;
880 }
881 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td);
882 if (error)
883 goto bad;
884 mtx_unlock(&Giant);
885 UNP_LOCK();
886 unp = sotounpcb(so);
887 if (unp == NULL) {
888 error = EINVAL;
889 goto bad2;
890 }
891 so2 = vp->v_socket;
892 if (so2 == NULL) {
893 error = ECONNREFUSED;
894 goto bad2;
895 }
896 if (so->so_type != so2->so_type) {
897 error = EPROTOTYPE;
898 goto bad2;
899 }
900 if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
901 if (so2->so_options & SO_ACCEPTCONN) {
902 /*
903 * NB: drop locks here so unp_attach is entered
904 * w/o locks; this avoids a recursive lock
905 * of the head and holding sleep locks across
906 * a (potentially) blocking malloc.
907 */
908 UNP_UNLOCK();
909 so3 = sonewconn(so2, 0);
910 UNP_LOCK();
911 } else
912 so3 = NULL;
913 if (so3 == NULL) {
914 error = ECONNREFUSED;
915 goto bad2;
916 }
917 unp = sotounpcb(so);
918 unp2 = sotounpcb(so2);
919 unp3 = sotounpcb(so3);
920 if (unp2->unp_addr != NULL) {
921 bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len);
922 unp3->unp_addr = (struct sockaddr_un *) sa;
923 sa = NULL;
924 }
925 /*
926 * unp_peercred management:
927 *
928 * The connecter's (client's) credentials are copied
929 * from its process structure at the time of connect()
930 * (which is now).
931 */
932 cru2x(td->td_ucred, &unp3->unp_peercred);
933 unp3->unp_flags |= UNP_HAVEPC;
934 /*
935 * The receiver's (server's) credentials are copied
936 * from the unp_peercred member of socket on which the
937 * former called listen(); unp_listen() cached that
938 * process's credentials at that time so we can use
939 * them now.
940 */
941 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED,
942 ("unp_connect: listener without cached peercred"));
943 memcpy(&unp->unp_peercred, &unp2->unp_peercred,
944 sizeof(unp->unp_peercred));
945 unp->unp_flags |= UNP_HAVEPC;
946 #ifdef MAC
947 SOCK_LOCK(so);
948 mac_set_socket_peer_from_socket(so, so3);
949 mac_set_socket_peer_from_socket(so3, so);
950 SOCK_UNLOCK(so);
951 #endif
952
953 so2 = so3;
954 }
955 error = unp_connect2(so, so2);
956 bad2:
957 UNP_UNLOCK();
958 mtx_lock(&Giant);
959 bad:
960 mtx_assert(&Giant, MA_OWNED);
961 if (vp != NULL)
962 vput(vp);
963 mtx_unlock(&Giant);
964 free(sa, M_SONAME);
965 UNP_LOCK();
966 return (error);
967 }
968
969 static int
970 unp_connect2(struct socket *so, struct socket *so2)
971 {
972 struct unpcb *unp = sotounpcb(so);
973 struct unpcb *unp2;
974
975 UNP_LOCK_ASSERT();
976
977 if (so2->so_type != so->so_type)
978 return (EPROTOTYPE);
979 unp2 = sotounpcb(so2);
980 unp->unp_conn = unp2;
981 switch (so->so_type) {
982
983 case SOCK_DGRAM:
984 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
985 soisconnected(so);
986 break;
987
988 case SOCK_STREAM:
989 unp2->unp_conn = unp;
990 soisconnected(so);
991 soisconnected(so2);
992 break;
993
994 default:
995 panic("unp_connect2");
996 }
997 return (0);
998 }
999
1000 static void
1001 unp_disconnect(struct unpcb *unp)
1002 {
1003 struct unpcb *unp2 = unp->unp_conn;
1004 struct socket *so;
1005
1006 UNP_LOCK_ASSERT();
1007
1008 if (unp2 == NULL)
1009 return;
1010 unp->unp_conn = NULL;
1011 switch (unp->unp_socket->so_type) {
1012
1013 case SOCK_DGRAM:
1014 LIST_REMOVE(unp, unp_reflink);
1015 so = unp->unp_socket;
1016 SOCK_LOCK(so);
1017 so->so_state &= ~SS_ISCONNECTED;
1018 SOCK_UNLOCK(so);
1019 break;
1020
1021 case SOCK_STREAM:
1022 soisdisconnected(unp->unp_socket);
1023 unp2->unp_conn = NULL;
1024 soisdisconnected(unp2->unp_socket);
1025 break;
1026 }
1027 }
1028
1029 #ifdef notdef
1030 void
1031 unp_abort(struct unpcb *unp)
1032 {
1033
1034 unp_detach(unp);
1035 UNP_UNLOCK_ASSERT();
1036 }
1037 #endif
1038
1039 /*
1040 * unp_pcblist() assumes that UNIX domain socket memory is never reclaimed
1041 * by the zone (UMA_ZONE_NOFREE), and as such potentially stale pointers
1042 * are safe to reference. It first scans the list of struct unpcb's to
1043 * generate a pointer list, then it rescans its list one entry at a time to
1044 * externalize and copyout. It checks the generation number to see if a
1045 * struct unpcb has been reused, and will skip it if so.
1046 */
1047 static int
1048 unp_pcblist(SYSCTL_HANDLER_ARGS)
1049 {
1050 int error, i, n;
1051 struct unpcb *unp, **unp_list;
1052 unp_gen_t gencnt;
1053 struct xunpgen *xug;
1054 struct unp_head *head;
1055 struct xunpcb *xu;
1056
1057 head = ((intptr_t)arg1 == SOCK_DGRAM ? &unp_dhead : &unp_shead);
1058
1059 /*
1060 * The process of preparing the PCB list is too time-consuming and
1061 * resource-intensive to repeat twice on every request.
1062 */
1063 if (req->oldptr == NULL) {
1064 n = unp_count;
1065 req->oldidx = 2 * (sizeof *xug)
1066 + (n + n/8) * sizeof(struct xunpcb);
1067 return (0);
1068 }
1069
1070 if (req->newptr != NULL)
1071 return (EPERM);
1072
1073 /*
1074 * OK, now we're committed to doing something.
1075 */
1076 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK);
1077 UNP_LOCK();
1078 gencnt = unp_gencnt;
1079 n = unp_count;
1080 UNP_UNLOCK();
1081
1082 xug->xug_len = sizeof *xug;
1083 xug->xug_count = n;
1084 xug->xug_gen = gencnt;
1085 xug->xug_sogen = so_gencnt;
1086 error = SYSCTL_OUT(req, xug, sizeof *xug);
1087 if (error) {
1088 free(xug, M_TEMP);
1089 return (error);
1090 }
1091
1092 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
1093
1094 UNP_LOCK();
1095 for (unp = LIST_FIRST(head), i = 0; unp && i < n;
1096 unp = LIST_NEXT(unp, unp_link)) {
1097 if (unp->unp_gencnt <= gencnt) {
1098 if (cr_cansee(req->td->td_ucred,
1099 unp->unp_socket->so_cred))
1100 continue;
1101 unp_list[i++] = unp;
1102 }
1103 }
1104 UNP_UNLOCK();
1105 n = i; /* in case we lost some during malloc */
1106
1107 error = 0;
1108 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO);
1109 for (i = 0; i < n; i++) {
1110 unp = unp_list[i];
1111 if (unp->unp_gencnt <= gencnt) {
1112 xu->xu_len = sizeof *xu;
1113 xu->xu_unpp = unp;
1114 /*
1115 * XXX - need more locking here to protect against
1116 * connect/disconnect races for SMP.
1117 */
1118 if (unp->unp_addr != NULL)
1119 bcopy(unp->unp_addr, &xu->xu_addr,
1120 unp->unp_addr->sun_len);
1121 if (unp->unp_conn != NULL &&
1122 unp->unp_conn->unp_addr != NULL)
1123 bcopy(unp->unp_conn->unp_addr,
1124 &xu->xu_caddr,
1125 unp->unp_conn->unp_addr->sun_len);
1126 bcopy(unp, &xu->xu_unp, sizeof *unp);
1127 sotoxsocket(unp->unp_socket, &xu->xu_socket);
1128 error = SYSCTL_OUT(req, xu, sizeof *xu);
1129 }
1130 }
1131 free(xu, M_TEMP);
1132 if (!error) {
1133 /*
1134 * Give the user an updated idea of our state.
1135 * If the generation differs from what we told
1136 * her before, she knows that something happened
1137 * while we were processing this request, and it
1138 * might be necessary to retry.
1139 */
1140 xug->xug_gen = unp_gencnt;
1141 xug->xug_sogen = so_gencnt;
1142 xug->xug_count = unp_count;
1143 error = SYSCTL_OUT(req, xug, sizeof *xug);
1144 }
1145 free(unp_list, M_TEMP);
1146 free(xug, M_TEMP);
1147 return (error);
1148 }
1149
1150 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLFLAG_RD,
1151 (caddr_t)(long)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
1152 "List of active local datagram sockets");
1153 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLFLAG_RD,
1154 (caddr_t)(long)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
1155 "List of active local stream sockets");
1156
1157 static void
1158 unp_shutdown(struct unpcb *unp)
1159 {
1160 struct socket *so;
1161
1162 UNP_LOCK_ASSERT();
1163
1164 if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn &&
1165 (so = unp->unp_conn->unp_socket))
1166 socantrcvmore(so);
1167 }
1168
1169 static void
1170 unp_drop(struct unpcb *unp, int errno)
1171 {
1172 struct socket *so = unp->unp_socket;
1173
1174 UNP_LOCK_ASSERT();
1175
1176 so->so_error = errno;
1177 unp_disconnect(unp);
1178 }
1179
1180 #ifdef notdef
1181 void
1182 unp_drain(void)
1183 {
1184
1185 }
1186 #endif
1187
1188 static void
1189 unp_freerights(struct file **rp, int fdcount)
1190 {
1191 int i;
1192 struct file *fp;
1193
1194 for (i = 0; i < fdcount; i++) {
1195 fp = *rp;
1196 /*
1197 * zero the pointer before calling
1198 * unp_discard since it may end up
1199 * in unp_gc()..
1200 */
1201 *rp++ = 0;
1202 unp_discard(fp);
1203 }
1204 }
1205
1206 int
1207 unp_externalize(struct mbuf *control, struct mbuf **controlp)
1208 {
1209 struct thread *td = curthread; /* XXX */
1210 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1211 int i;
1212 int *fdp;
1213 struct file **rp;
1214 struct file *fp;
1215 void *data;
1216 socklen_t clen = control->m_len, datalen;
1217 int error, newfds;
1218 int f;
1219 u_int newlen;
1220
1221 UNP_UNLOCK_ASSERT();
1222
1223 error = 0;
1224 if (controlp != NULL) /* controlp == NULL => free control messages */
1225 *controlp = NULL;
1226
1227 while (cm != NULL) {
1228 if (sizeof(*cm) > clen || cm->cmsg_len > clen) {
1229 error = EINVAL;
1230 break;
1231 }
1232
1233 data = CMSG_DATA(cm);
1234 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1235
1236 if (cm->cmsg_level == SOL_SOCKET
1237 && cm->cmsg_type == SCM_RIGHTS) {
1238 newfds = datalen / sizeof(struct file *);
1239 rp = data;
1240
1241 /* If we're not outputting the descriptors free them. */
1242 if (error || controlp == NULL) {
1243 unp_freerights(rp, newfds);
1244 goto next;
1245 }
1246 FILEDESC_LOCK(td->td_proc->p_fd);
1247 /* if the new FD's will not fit free them. */
1248 if (!fdavail(td, newfds)) {
1249 FILEDESC_UNLOCK(td->td_proc->p_fd);
1250 error = EMSGSIZE;
1251 unp_freerights(rp, newfds);
1252 goto next;
1253 }
1254 /*
1255 * now change each pointer to an fd in the global
1256 * table to an integer that is the index to the
1257 * local fd table entry that we set up to point
1258 * to the global one we are transferring.
1259 */
1260 newlen = newfds * sizeof(int);
1261 *controlp = sbcreatecontrol(NULL, newlen,
1262 SCM_RIGHTS, SOL_SOCKET);
1263 if (*controlp == NULL) {
1264 FILEDESC_UNLOCK(td->td_proc->p_fd);
1265 error = E2BIG;
1266 unp_freerights(rp, newfds);
1267 goto next;
1268 }
1269
1270 fdp = (int *)
1271 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1272 for (i = 0; i < newfds; i++) {
1273 if (fdalloc(td, 0, &f))
1274 panic("unp_externalize fdalloc failed");
1275 fp = *rp++;
1276 td->td_proc->p_fd->fd_ofiles[f] = fp;
1277 FILE_LOCK(fp);
1278 fp->f_msgcount--;
1279 FILE_UNLOCK(fp);
1280 unp_rights--;
1281 *fdp++ = f;
1282 }
1283 FILEDESC_UNLOCK(td->td_proc->p_fd);
1284 } else { /* We can just copy anything else across */
1285 if (error || controlp == NULL)
1286 goto next;
1287 *controlp = sbcreatecontrol(NULL, datalen,
1288 cm->cmsg_type, cm->cmsg_level);
1289 if (*controlp == NULL) {
1290 error = ENOBUFS;
1291 goto next;
1292 }
1293 bcopy(data,
1294 CMSG_DATA(mtod(*controlp, struct cmsghdr *)),
1295 datalen);
1296 }
1297
1298 controlp = &(*controlp)->m_next;
1299
1300 next:
1301 if (CMSG_SPACE(datalen) < clen) {
1302 clen -= CMSG_SPACE(datalen);
1303 cm = (struct cmsghdr *)
1304 ((caddr_t)cm + CMSG_SPACE(datalen));
1305 } else {
1306 clen = 0;
1307 cm = NULL;
1308 }
1309 }
1310
1311 m_freem(control);
1312
1313 return (error);
1314 }
1315
1316 void
1317 unp_init(void)
1318 {
1319 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL,
1320 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
1321 if (unp_zone == NULL)
1322 panic("unp_init");
1323 uma_zone_set_max(unp_zone, nmbclusters);
1324 LIST_INIT(&unp_dhead);
1325 LIST_INIT(&unp_shead);
1326
1327 UNP_LOCK_INIT();
1328 }
1329
1330 static int
1331 unp_internalize(struct mbuf **controlp, struct thread *td)
1332 {
1333 struct mbuf *control = *controlp;
1334 struct proc *p = td->td_proc;
1335 struct filedesc *fdescp = p->p_fd;
1336 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1337 struct cmsgcred *cmcred;
1338 struct file **rp;
1339 struct file *fp;
1340 struct timeval *tv;
1341 int i, fd, *fdp;
1342 void *data;
1343 socklen_t clen = control->m_len, datalen;
1344 int error, oldfds;
1345 u_int newlen;
1346
1347 UNP_UNLOCK_ASSERT();
1348
1349 error = 0;
1350 *controlp = NULL;
1351
1352 while (cm != NULL) {
1353 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET
1354 || cm->cmsg_len > clen) {
1355 error = EINVAL;
1356 goto out;
1357 }
1358
1359 data = CMSG_DATA(cm);
1360 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1361
1362 switch (cm->cmsg_type) {
1363 /*
1364 * Fill in credential information.
1365 */
1366 case SCM_CREDS:
1367 *controlp = sbcreatecontrol(NULL, sizeof(*cmcred),
1368 SCM_CREDS, SOL_SOCKET);
1369 if (*controlp == NULL) {
1370 error = ENOBUFS;
1371 goto out;
1372 }
1373
1374 cmcred = (struct cmsgcred *)
1375 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1376 cmcred->cmcred_pid = p->p_pid;
1377 cmcred->cmcred_uid = td->td_ucred->cr_ruid;
1378 cmcred->cmcred_gid = td->td_ucred->cr_rgid;
1379 cmcred->cmcred_euid = td->td_ucred->cr_uid;
1380 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups,
1381 CMGROUP_MAX);
1382 for (i = 0; i < cmcred->cmcred_ngroups; i++)
1383 cmcred->cmcred_groups[i] =
1384 td->td_ucred->cr_groups[i];
1385 break;
1386
1387 case SCM_RIGHTS:
1388 oldfds = datalen / sizeof (int);
1389 /*
1390 * check that all the FDs passed in refer to legal files
1391 * If not, reject the entire operation.
1392 */
1393 fdp = data;
1394 FILEDESC_LOCK(fdescp);
1395 for (i = 0; i < oldfds; i++) {
1396 fd = *fdp++;
1397 if ((unsigned)fd >= fdescp->fd_nfiles ||
1398 fdescp->fd_ofiles[fd] == NULL) {
1399 FILEDESC_UNLOCK(fdescp);
1400 error = EBADF;
1401 goto out;
1402 }
1403 fp = fdescp->fd_ofiles[fd];
1404 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) {
1405 FILEDESC_UNLOCK(fdescp);
1406 error = EOPNOTSUPP;
1407 goto out;
1408 }
1409
1410 }
1411 /*
1412 * Now replace the integer FDs with pointers to
1413 * the associated global file table entry..
1414 */
1415 newlen = oldfds * sizeof(struct file *);
1416 *controlp = sbcreatecontrol(NULL, newlen,
1417 SCM_RIGHTS, SOL_SOCKET);
1418 if (*controlp == NULL) {
1419 FILEDESC_UNLOCK(fdescp);
1420 error = E2BIG;
1421 goto out;
1422 }
1423
1424 fdp = data;
1425 rp = (struct file **)
1426 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1427 for (i = 0; i < oldfds; i++) {
1428 fp = fdescp->fd_ofiles[*fdp++];
1429 *rp++ = fp;
1430 FILE_LOCK(fp);
1431 fp->f_count++;
1432 fp->f_msgcount++;
1433 FILE_UNLOCK(fp);
1434 unp_rights++;
1435 }
1436 FILEDESC_UNLOCK(fdescp);
1437 break;
1438
1439 case SCM_TIMESTAMP:
1440 *controlp = sbcreatecontrol(NULL, sizeof(*tv),
1441 SCM_TIMESTAMP, SOL_SOCKET);
1442 if (*controlp == NULL) {
1443 error = ENOBUFS;
1444 goto out;
1445 }
1446 tv = (struct timeval *)
1447 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1448 microtime(tv);
1449 break;
1450
1451 default:
1452 error = EINVAL;
1453 goto out;
1454 }
1455
1456 controlp = &(*controlp)->m_next;
1457
1458 if (CMSG_SPACE(datalen) < clen) {
1459 clen -= CMSG_SPACE(datalen);
1460 cm = (struct cmsghdr *)
1461 ((caddr_t)cm + CMSG_SPACE(datalen));
1462 } else {
1463 clen = 0;
1464 cm = NULL;
1465 }
1466 }
1467
1468 out:
1469 m_freem(control);
1470
1471 return (error);
1472 }
1473
1474 /*
1475 * unp_defer is thread-local during garbage collection, and does not require
1476 * explicit synchronization. unp_gcing prevents other threads from entering
1477 * garbage collection, and perhaps should be an sx lock instead.
1478 */
1479 static int unp_defer, unp_gcing;
1480
1481 static void
1482 unp_gc(void)
1483 {
1484 struct file *fp, *nextfp;
1485 struct socket *so;
1486 struct file **extra_ref, **fpp;
1487 int nunref, i;
1488 int nfiles_snap;
1489 int nfiles_slack = 20;
1490
1491 UNP_LOCK_ASSERT();
1492
1493 if (unp_gcing) {
1494 UNP_UNLOCK();
1495 return;
1496 }
1497 unp_gcing = 1;
1498 unp_defer = 0;
1499 UNP_UNLOCK();
1500 /*
1501 * before going through all this, set all FDs to
1502 * be NOT defered and NOT externally accessible
1503 */
1504 sx_slock(&filelist_lock);
1505 LIST_FOREACH(fp, &filehead, f_list)
1506 fp->f_gcflag &= ~(FMARK|FDEFER);
1507 do {
1508 LIST_FOREACH(fp, &filehead, f_list) {
1509 FILE_LOCK(fp);
1510 /*
1511 * If the file is not open, skip it
1512 */
1513 if (fp->f_count == 0) {
1514 FILE_UNLOCK(fp);
1515 continue;
1516 }
1517 /*
1518 * If we already marked it as 'defer' in a
1519 * previous pass, then try process it this time
1520 * and un-mark it
1521 */
1522 if (fp->f_gcflag & FDEFER) {
1523 fp->f_gcflag &= ~FDEFER;
1524 unp_defer--;
1525 } else {
1526 /*
1527 * if it's not defered, then check if it's
1528 * already marked.. if so skip it
1529 */
1530 if (fp->f_gcflag & FMARK) {
1531 FILE_UNLOCK(fp);
1532 continue;
1533 }
1534 /*
1535 * If all references are from messages
1536 * in transit, then skip it. it's not
1537 * externally accessible.
1538 */
1539 if (fp->f_count == fp->f_msgcount) {
1540 FILE_UNLOCK(fp);
1541 continue;
1542 }
1543 /*
1544 * If it got this far then it must be
1545 * externally accessible.
1546 */
1547 fp->f_gcflag |= FMARK;
1548 }
1549 /*
1550 * either it was defered, or it is externally
1551 * accessible and not already marked so.
1552 * Now check if it is possibly one of OUR sockets.
1553 */
1554 if (fp->f_type != DTYPE_SOCKET ||
1555 (so = fp->f_data) == NULL) {
1556 FILE_UNLOCK(fp);
1557 continue;
1558 }
1559 FILE_UNLOCK(fp);
1560 if (so->so_proto->pr_domain != &localdomain ||
1561 (so->so_proto->pr_flags&PR_RIGHTS) == 0)
1562 continue;
1563 #ifdef notdef
1564 if (so->so_rcv.sb_flags & SB_LOCK) {
1565 /*
1566 * This is problematical; it's not clear
1567 * we need to wait for the sockbuf to be
1568 * unlocked (on a uniprocessor, at least),
1569 * and it's also not clear what to do
1570 * if sbwait returns an error due to receipt
1571 * of a signal. If sbwait does return
1572 * an error, we'll go into an infinite
1573 * loop. Delete all of this for now.
1574 */
1575 (void) sbwait(&so->so_rcv);
1576 goto restart;
1577 }
1578 #endif
1579 /*
1580 * So, Ok, it's one of our sockets and it IS externally
1581 * accessible (or was defered). Now we look
1582 * to see if we hold any file descriptors in its
1583 * message buffers. Follow those links and mark them
1584 * as accessible too.
1585 */
1586 SOCKBUF_LOCK(&so->so_rcv);
1587 unp_scan(so->so_rcv.sb_mb, unp_mark);
1588 SOCKBUF_UNLOCK(&so->so_rcv);
1589 }
1590 } while (unp_defer);
1591 sx_sunlock(&filelist_lock);
1592 /*
1593 * We grab an extra reference to each of the file table entries
1594 * that are not otherwise accessible and then free the rights
1595 * that are stored in messages on them.
1596 *
1597 * The bug in the orginal code is a little tricky, so I'll describe
1598 * what's wrong with it here.
1599 *
1600 * It is incorrect to simply unp_discard each entry for f_msgcount
1601 * times -- consider the case of sockets A and B that contain
1602 * references to each other. On a last close of some other socket,
1603 * we trigger a gc since the number of outstanding rights (unp_rights)
1604 * is non-zero. If during the sweep phase the gc code un_discards,
1605 * we end up doing a (full) closef on the descriptor. A closef on A
1606 * results in the following chain. Closef calls soo_close, which
1607 * calls soclose. Soclose calls first (through the switch
1608 * uipc_usrreq) unp_detach, which re-invokes unp_gc. Unp_gc simply
1609 * returns because the previous instance had set unp_gcing, and
1610 * we return all the way back to soclose, which marks the socket
1611 * with SS_NOFDREF, and then calls sofree. Sofree calls sorflush
1612 * to free up the rights that are queued in messages on the socket A,
1613 * i.e., the reference on B. The sorflush calls via the dom_dispose
1614 * switch unp_dispose, which unp_scans with unp_discard. This second
1615 * instance of unp_discard just calls closef on B.
1616 *
1617 * Well, a similar chain occurs on B, resulting in a sorflush on B,
1618 * which results in another closef on A. Unfortunately, A is already
1619 * being closed, and the descriptor has already been marked with
1620 * SS_NOFDREF, and soclose panics at this point.
1621 *
1622 * Here, we first take an extra reference to each inaccessible
1623 * descriptor. Then, we call sorflush ourself, since we know
1624 * it is a Unix domain socket anyhow. After we destroy all the
1625 * rights carried in messages, we do a last closef to get rid
1626 * of our extra reference. This is the last close, and the
1627 * unp_detach etc will shut down the socket.
1628 *
1629 * 91/09/19, bsy@cs.cmu.edu
1630 */
1631 again:
1632 nfiles_snap = nfiles + nfiles_slack; /* some slack */
1633 extra_ref = malloc(nfiles_snap * sizeof(struct file *), M_TEMP,
1634 M_WAITOK);
1635 sx_slock(&filelist_lock);
1636 if (nfiles_snap < nfiles) {
1637 sx_sunlock(&filelist_lock);
1638 free(extra_ref, M_TEMP);
1639 nfiles_slack += 20;
1640 goto again;
1641 }
1642 for (nunref = 0, fp = LIST_FIRST(&filehead), fpp = extra_ref;
1643 fp != NULL; fp = nextfp) {
1644 nextfp = LIST_NEXT(fp, f_list);
1645 FILE_LOCK(fp);
1646 /*
1647 * If it's not open, skip it
1648 */
1649 if (fp->f_count == 0) {
1650 FILE_UNLOCK(fp);
1651 continue;
1652 }
1653 /*
1654 * If all refs are from msgs, and it's not marked accessible
1655 * then it must be referenced from some unreachable cycle
1656 * of (shut-down) FDs, so include it in our
1657 * list of FDs to remove
1658 */
1659 if (fp->f_count == fp->f_msgcount && !(fp->f_gcflag & FMARK)) {
1660 *fpp++ = fp;
1661 nunref++;
1662 fp->f_count++;
1663 }
1664 FILE_UNLOCK(fp);
1665 }
1666 sx_sunlock(&filelist_lock);
1667 /*
1668 * for each FD on our hit list, do the following two things
1669 */
1670 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) {
1671 struct file *tfp = *fpp;
1672 FILE_LOCK(tfp);
1673 if (tfp->f_type == DTYPE_SOCKET &&
1674 tfp->f_data != NULL) {
1675 FILE_UNLOCK(tfp);
1676 sorflush(tfp->f_data);
1677 } else {
1678 FILE_UNLOCK(tfp);
1679 }
1680 }
1681 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp)
1682 closef(*fpp, (struct thread *) NULL);
1683 free(extra_ref, M_TEMP);
1684 unp_gcing = 0;
1685
1686 UNP_UNLOCK_ASSERT();
1687 }
1688
1689 void
1690 unp_dispose(struct mbuf *m)
1691 {
1692
1693 if (m)
1694 unp_scan(m, unp_discard);
1695 }
1696
1697 static int
1698 unp_listen(struct socket *so, struct unpcb *unp, struct thread *td)
1699 {
1700 int error;
1701
1702 UNP_LOCK_ASSERT();
1703
1704 SOCK_LOCK(so);
1705 error = solisten_proto_check(so);
1706 if (error == 0) {
1707 cru2x(td->td_ucred, &unp->unp_peercred);
1708 unp->unp_flags |= UNP_HAVEPCCACHED;
1709 solisten_proto(so);
1710 }
1711 SOCK_UNLOCK(so);
1712 return (error);
1713 }
1714
1715 static void
1716 unp_scan(struct mbuf *m0, void (*op)(struct file *))
1717 {
1718 struct mbuf *m;
1719 struct file **rp;
1720 struct cmsghdr *cm;
1721 void *data;
1722 int i;
1723 socklen_t clen, datalen;
1724 int qfds;
1725
1726 while (m0 != NULL) {
1727 for (m = m0; m; m = m->m_next) {
1728 if (m->m_type != MT_CONTROL)
1729 continue;
1730
1731 cm = mtod(m, struct cmsghdr *);
1732 clen = m->m_len;
1733
1734 while (cm != NULL) {
1735 if (sizeof(*cm) > clen || cm->cmsg_len > clen)
1736 break;
1737
1738 data = CMSG_DATA(cm);
1739 datalen = (caddr_t)cm + cm->cmsg_len
1740 - (caddr_t)data;
1741
1742 if (cm->cmsg_level == SOL_SOCKET &&
1743 cm->cmsg_type == SCM_RIGHTS) {
1744 qfds = datalen / sizeof (struct file *);
1745 rp = data;
1746 for (i = 0; i < qfds; i++)
1747 (*op)(*rp++);
1748 }
1749
1750 if (CMSG_SPACE(datalen) < clen) {
1751 clen -= CMSG_SPACE(datalen);
1752 cm = (struct cmsghdr *)
1753 ((caddr_t)cm + CMSG_SPACE(datalen));
1754 } else {
1755 clen = 0;
1756 cm = NULL;
1757 }
1758 }
1759 }
1760 m0 = m0->m_act;
1761 }
1762 }
1763
1764 static void
1765 unp_mark(struct file *fp)
1766 {
1767 if (fp->f_gcflag & FMARK)
1768 return;
1769 unp_defer++;
1770 fp->f_gcflag |= (FMARK|FDEFER);
1771 }
1772
1773 static void
1774 unp_discard(struct file *fp)
1775 {
1776 FILE_LOCK(fp);
1777 fp->f_msgcount--;
1778 unp_rights--;
1779 FILE_UNLOCK(fp);
1780 (void) closef(fp, (struct thread *)NULL);
1781 }
Cache object: fb50c37a5d7ad627be10aaa0285ab474
|