1 /*-
2 * Copyright (c) 1982, 1986, 1989, 1991, 1993
3 * The Regents of the University of California.
4 * Copyright 2004-2005 Robert N. M. Watson
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 4. Neither the name of the University nor the names of its contributors
16 * may be used to endorse or promote products derived from this software
17 * without specific prior written permission.
18 *
19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * SUCH DAMAGE.
30 *
31 * From: @(#)uipc_usrreq.c 8.3 (Berkeley) 1/4/94
32 */
33
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD: releng/6.2/sys/kern/uipc_usrreq.c 164286 2006-11-14 20:42:41Z cvs2svn $");
36
37 #include "opt_mac.h"
38
39 #include <sys/param.h>
40 #include <sys/domain.h>
41 #include <sys/fcntl.h>
42 #include <sys/malloc.h> /* XXX must be before <sys/file.h> */
43 #include <sys/eventhandler.h>
44 #include <sys/file.h>
45 #include <sys/filedesc.h>
46 #include <sys/jail.h>
47 #include <sys/kernel.h>
48 #include <sys/lock.h>
49 #include <sys/mac.h>
50 #include <sys/mbuf.h>
51 #include <sys/mount.h>
52 #include <sys/mutex.h>
53 #include <sys/namei.h>
54 #include <sys/proc.h>
55 #include <sys/protosw.h>
56 #include <sys/resourcevar.h>
57 #include <sys/socket.h>
58 #include <sys/socketvar.h>
59 #include <sys/signalvar.h>
60 #include <sys/stat.h>
61 #include <sys/sx.h>
62 #include <sys/sysctl.h>
63 #include <sys/systm.h>
64 #include <sys/taskqueue.h>
65 #include <sys/un.h>
66 #include <sys/unpcb.h>
67 #include <sys/vnode.h>
68
69 #include <vm/uma.h>
70
71 static uma_zone_t unp_zone;
72 static unp_gen_t unp_gencnt;
73 static u_int unp_count;
74
75 static struct unp_head unp_shead, unp_dhead;
76
77 /*
78 * Unix communications domain.
79 *
80 * TODO:
81 * SEQPACKET, RDM
82 * rethink name space problems
83 * need a proper out-of-band
84 * lock pushdown
85 */
86 static const struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL };
87 static ino_t unp_ino; /* prototype for fake inode numbers */
88 struct mbuf *unp_addsockcred(struct thread *, struct mbuf *);
89
90 /*
91 * Currently, UNIX domain sockets are protected by a single subsystem lock,
92 * which covers global data structures and variables, the contents of each
93 * per-socket unpcb structure, and the so_pcb field in sockets attached to
94 * the UNIX domain. This provides for a moderate degree of paralellism, as
95 * receive operations on UNIX domain sockets do not need to acquire the
96 * subsystem lock. Finer grained locking to permit send() without acquiring
97 * a global lock would be a logical next step.
98 *
99 * The UNIX domain socket lock preceds all socket layer locks, including the
100 * socket lock and socket buffer lock, permitting UNIX domain socket code to
101 * call into socket support routines without releasing its locks.
102 *
103 * Some caution is required in areas where the UNIX domain socket code enters
104 * VFS in order to create or find rendezvous points. This results in
105 * dropping of the UNIX domain socket subsystem lock, acquisition of the
106 * Giant lock, and potential sleeping. This increases the chances of races,
107 * and exposes weaknesses in the socket->protocol API by offering poor
108 * failure modes.
109 */
110 static struct mtx unp_mtx;
111 #define UNP_LOCK_INIT() \
112 mtx_init(&unp_mtx, "unp", NULL, MTX_DEF)
113 #define UNP_LOCK() mtx_lock(&unp_mtx)
114 #define UNP_UNLOCK() mtx_unlock(&unp_mtx)
115 #define UNP_LOCK_ASSERT() mtx_assert(&unp_mtx, MA_OWNED)
116 #define UNP_UNLOCK_ASSERT() mtx_assert(&unp_mtx, MA_NOTOWNED)
117
118 /*
119 * Garbage collection of cyclic file descriptor/socket references occurs
120 * asynchronously in a taskqueue context in order to avoid recursion and
121 * reentrance in the UNIX domain socket, file descriptor, and socket layer
122 * code. See unp_gc() for a full description.
123 */
124 static struct task unp_gc_task;
125
126 static int unp_attach(struct socket *);
127 static void unp_detach(struct unpcb *);
128 static int unp_bind(struct unpcb *,struct sockaddr *, struct thread *);
129 static int unp_connect(struct socket *,struct sockaddr *, struct thread *);
130 static int unp_connect2(struct socket *so, struct socket *so2, int);
131 static void unp_disconnect(struct unpcb *);
132 static void unp_shutdown(struct unpcb *);
133 static void unp_drop(struct unpcb *, int);
134 static void unp_gc(__unused void *, int);
135 static void unp_scan(struct mbuf *, void (*)(struct file *));
136 static void unp_mark(struct file *);
137 static void unp_discard(struct file *);
138 static void unp_freerights(struct file **, int);
139 static int unp_internalize(struct mbuf **, struct thread *);
140 static int unp_listen(struct socket *, struct unpcb *, struct thread *);
141
142 static int
143 uipc_abort(struct socket *so)
144 {
145 struct unpcb *unp;
146
147 UNP_LOCK();
148 unp = sotounpcb(so);
149 if (unp == NULL) {
150 UNP_UNLOCK();
151 return (EINVAL);
152 }
153 unp_drop(unp, ECONNABORTED);
154 unp_detach(unp);
155 UNP_UNLOCK_ASSERT();
156 ACCEPT_LOCK();
157 SOCK_LOCK(so);
158 sotryfree(so);
159 return (0);
160 }
161
162 static int
163 uipc_accept(struct socket *so, struct sockaddr **nam)
164 {
165 struct unpcb *unp;
166 const struct sockaddr *sa;
167
168 /*
169 * Pass back name of connected socket,
170 * if it was bound and we are still connected
171 * (our peer may have closed already!).
172 */
173 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
174 UNP_LOCK();
175 unp = sotounpcb(so);
176 if (unp == NULL) {
177 UNP_UNLOCK();
178 free(*nam, M_SONAME);
179 *nam = NULL;
180 return (EINVAL);
181 }
182 if (unp->unp_conn != NULL && unp->unp_conn->unp_addr != NULL)
183 sa = (struct sockaddr *) unp->unp_conn->unp_addr;
184 else
185 sa = &sun_noname;
186 bcopy(sa, *nam, sa->sa_len);
187 UNP_UNLOCK();
188 return (0);
189 }
190
191 static int
192 uipc_attach(struct socket *so, int proto, struct thread *td)
193 {
194 struct unpcb *unp = sotounpcb(so);
195
196 if (unp != NULL)
197 return (EISCONN);
198 return (unp_attach(so));
199 }
200
201 static int
202 uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
203 {
204 struct unpcb *unp;
205 int error;
206
207 UNP_LOCK();
208 unp = sotounpcb(so);
209 if (unp == NULL) {
210 UNP_UNLOCK();
211 return (EINVAL);
212 }
213 error = unp_bind(unp, nam, td);
214 UNP_UNLOCK();
215 return (error);
216 }
217
218 static int
219 uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
220 {
221 struct unpcb *unp;
222 int error;
223
224 KASSERT(td == curthread, ("uipc_connect: td != curthread"));
225
226 UNP_LOCK();
227 unp = sotounpcb(so);
228 if (unp == NULL) {
229 UNP_UNLOCK();
230 return (EINVAL);
231 }
232 error = unp_connect(so, nam, td);
233 UNP_UNLOCK();
234 return (error);
235 }
236
237 int
238 uipc_connect2(struct socket *so1, struct socket *so2)
239 {
240 struct unpcb *unp;
241 int error;
242
243 UNP_LOCK();
244 unp = sotounpcb(so1);
245 if (unp == NULL) {
246 UNP_UNLOCK();
247 return (EINVAL);
248 }
249 error = unp_connect2(so1, so2, PRU_CONNECT2);
250 UNP_UNLOCK();
251 return (error);
252 }
253
254 /* control is EOPNOTSUPP */
255
256 static int
257 uipc_detach(struct socket *so)
258 {
259 struct unpcb *unp;
260
261 UNP_LOCK();
262 unp = sotounpcb(so);
263 if (unp == NULL) {
264 UNP_UNLOCK();
265 return (EINVAL);
266 }
267 unp_detach(unp);
268 UNP_UNLOCK_ASSERT();
269 return (0);
270 }
271
272 static int
273 uipc_disconnect(struct socket *so)
274 {
275 struct unpcb *unp;
276
277 UNP_LOCK();
278 unp = sotounpcb(so);
279 if (unp == NULL) {
280 UNP_UNLOCK();
281 return (EINVAL);
282 }
283 unp_disconnect(unp);
284 UNP_UNLOCK();
285 return (0);
286 }
287
288 static int
289 uipc_listen(struct socket *so, struct thread *td)
290 {
291 struct unpcb *unp;
292 int error;
293
294 UNP_LOCK();
295 unp = sotounpcb(so);
296 if (unp == NULL || unp->unp_vnode == NULL) {
297 UNP_UNLOCK();
298 return (EINVAL);
299 }
300 error = unp_listen(so, unp, td);
301 UNP_UNLOCK();
302 return (error);
303 }
304
305 static int
306 uipc_peeraddr(struct socket *so, struct sockaddr **nam)
307 {
308 struct unpcb *unp;
309 const struct sockaddr *sa;
310
311 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
312 UNP_LOCK();
313 unp = sotounpcb(so);
314 if (unp == NULL) {
315 UNP_UNLOCK();
316 free(*nam, M_SONAME);
317 *nam = NULL;
318 return (EINVAL);
319 }
320 if (unp->unp_conn != NULL && unp->unp_conn->unp_addr!= NULL)
321 sa = (struct sockaddr *) unp->unp_conn->unp_addr;
322 else {
323 /*
324 * XXX: It seems that this test always fails even when
325 * connection is established. So, this else clause is
326 * added as workaround to return PF_LOCAL sockaddr.
327 */
328 sa = &sun_noname;
329 }
330 bcopy(sa, *nam, sa->sa_len);
331 UNP_UNLOCK();
332 return (0);
333 }
334
335 static int
336 uipc_rcvd(struct socket *so, int flags)
337 {
338 struct unpcb *unp;
339 struct socket *so2;
340 u_long newhiwat;
341
342 UNP_LOCK();
343 unp = sotounpcb(so);
344 if (unp == NULL) {
345 UNP_UNLOCK();
346 return (EINVAL);
347 }
348 switch (so->so_type) {
349 case SOCK_DGRAM:
350 panic("uipc_rcvd DGRAM?");
351 /*NOTREACHED*/
352
353 case SOCK_STREAM:
354 if (unp->unp_conn == NULL)
355 break;
356 so2 = unp->unp_conn->unp_socket;
357 SOCKBUF_LOCK(&so2->so_snd);
358 SOCKBUF_LOCK(&so->so_rcv);
359 /*
360 * Adjust backpressure on sender
361 * and wakeup any waiting to write.
362 */
363 so2->so_snd.sb_mbmax += unp->unp_mbcnt - so->so_rcv.sb_mbcnt;
364 unp->unp_mbcnt = so->so_rcv.sb_mbcnt;
365 newhiwat = so2->so_snd.sb_hiwat + unp->unp_cc -
366 so->so_rcv.sb_cc;
367 (void)chgsbsize(so2->so_cred->cr_uidinfo, &so2->so_snd.sb_hiwat,
368 newhiwat, RLIM_INFINITY);
369 unp->unp_cc = so->so_rcv.sb_cc;
370 SOCKBUF_UNLOCK(&so->so_rcv);
371 sowwakeup_locked(so2);
372 break;
373
374 default:
375 panic("uipc_rcvd unknown socktype");
376 }
377 UNP_UNLOCK();
378 return (0);
379 }
380
381 /* pru_rcvoob is EOPNOTSUPP */
382
383 static int
384 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
385 struct mbuf *control, struct thread *td)
386 {
387 int error = 0;
388 struct unpcb *unp;
389 struct socket *so2;
390 u_long newhiwat;
391
392 unp = sotounpcb(so);
393 if (unp == NULL) {
394 error = EINVAL;
395 goto release;
396 }
397 if (flags & PRUS_OOB) {
398 error = EOPNOTSUPP;
399 goto release;
400 }
401
402 if (control != NULL && (error = unp_internalize(&control, td)))
403 goto release;
404
405 UNP_LOCK();
406 unp = sotounpcb(so);
407 if (unp == NULL) {
408 UNP_UNLOCK();
409 error = EINVAL;
410 goto dispose_release;
411 }
412
413 switch (so->so_type) {
414 case SOCK_DGRAM:
415 {
416 const struct sockaddr *from;
417
418 if (nam != NULL) {
419 if (unp->unp_conn != NULL) {
420 error = EISCONN;
421 break;
422 }
423 error = unp_connect(so, nam, td);
424 if (error)
425 break;
426 } else {
427 if (unp->unp_conn == NULL) {
428 error = ENOTCONN;
429 break;
430 }
431 }
432 so2 = unp->unp_conn->unp_socket;
433 if (unp->unp_addr != NULL)
434 from = (struct sockaddr *)unp->unp_addr;
435 else
436 from = &sun_noname;
437 if (unp->unp_conn->unp_flags & UNP_WANTCRED)
438 control = unp_addsockcred(td, control);
439 SOCKBUF_LOCK(&so2->so_rcv);
440 if (sbappendaddr_locked(&so2->so_rcv, from, m, control)) {
441 sorwakeup_locked(so2);
442 m = NULL;
443 control = NULL;
444 } else {
445 SOCKBUF_UNLOCK(&so2->so_rcv);
446 error = ENOBUFS;
447 }
448 if (nam != NULL)
449 unp_disconnect(unp);
450 break;
451 }
452
453 case SOCK_STREAM:
454 /* Connect if not connected yet. */
455 /*
456 * Note: A better implementation would complain
457 * if not equal to the peer's address.
458 */
459 if ((so->so_state & SS_ISCONNECTED) == 0) {
460 if (nam != NULL) {
461 error = unp_connect(so, nam, td);
462 if (error)
463 break; /* XXX */
464 } else {
465 error = ENOTCONN;
466 break;
467 }
468 }
469
470 SOCKBUF_LOCK(&so->so_snd);
471 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
472 SOCKBUF_UNLOCK(&so->so_snd);
473 error = EPIPE;
474 break;
475 }
476 if (unp->unp_conn == NULL)
477 panic("uipc_send connected but no connection?");
478 so2 = unp->unp_conn->unp_socket;
479 SOCKBUF_LOCK(&so2->so_rcv);
480 if (unp->unp_conn->unp_flags & UNP_WANTCRED) {
481 /*
482 * Credentials are passed only once on
483 * SOCK_STREAM.
484 */
485 unp->unp_conn->unp_flags &= ~UNP_WANTCRED;
486 control = unp_addsockcred(td, control);
487 }
488 /*
489 * Send to paired receive port, and then reduce
490 * send buffer hiwater marks to maintain backpressure.
491 * Wake up readers.
492 */
493 if (control != NULL) {
494 if (sbappendcontrol_locked(&so2->so_rcv, m, control))
495 control = NULL;
496 } else {
497 sbappend_locked(&so2->so_rcv, m);
498 }
499 so->so_snd.sb_mbmax -=
500 so2->so_rcv.sb_mbcnt - unp->unp_conn->unp_mbcnt;
501 unp->unp_conn->unp_mbcnt = so2->so_rcv.sb_mbcnt;
502 newhiwat = so->so_snd.sb_hiwat -
503 (so2->so_rcv.sb_cc - unp->unp_conn->unp_cc);
504 (void)chgsbsize(so->so_cred->cr_uidinfo, &so->so_snd.sb_hiwat,
505 newhiwat, RLIM_INFINITY);
506 SOCKBUF_UNLOCK(&so->so_snd);
507 unp->unp_conn->unp_cc = so2->so_rcv.sb_cc;
508 sorwakeup_locked(so2);
509 m = NULL;
510 break;
511
512 default:
513 panic("uipc_send unknown socktype");
514 }
515
516 /*
517 * SEND_EOF is equivalent to a SEND followed by
518 * a SHUTDOWN.
519 */
520 if (flags & PRUS_EOF) {
521 socantsendmore(so);
522 unp_shutdown(unp);
523 }
524 UNP_UNLOCK();
525
526 dispose_release:
527 if (control != NULL && error != 0)
528 unp_dispose(control);
529
530 release:
531 if (control != NULL)
532 m_freem(control);
533 if (m != NULL)
534 m_freem(m);
535 return (error);
536 }
537
538 static int
539 uipc_sense(struct socket *so, struct stat *sb)
540 {
541 struct unpcb *unp;
542 struct socket *so2;
543
544 UNP_LOCK();
545 unp = sotounpcb(so);
546 if (unp == NULL) {
547 UNP_UNLOCK();
548 return (EINVAL);
549 }
550 sb->st_blksize = so->so_snd.sb_hiwat;
551 if (so->so_type == SOCK_STREAM && unp->unp_conn != NULL) {
552 so2 = unp->unp_conn->unp_socket;
553 sb->st_blksize += so2->so_rcv.sb_cc;
554 }
555 sb->st_dev = NODEV;
556 if (unp->unp_ino == 0)
557 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino;
558 sb->st_ino = unp->unp_ino;
559 UNP_UNLOCK();
560 return (0);
561 }
562
563 static int
564 uipc_shutdown(struct socket *so)
565 {
566 struct unpcb *unp;
567
568 UNP_LOCK();
569 unp = sotounpcb(so);
570 if (unp == NULL) {
571 UNP_UNLOCK();
572 return (EINVAL);
573 }
574 socantsendmore(so);
575 unp_shutdown(unp);
576 UNP_UNLOCK();
577 return (0);
578 }
579
580 static int
581 uipc_sockaddr(struct socket *so, struct sockaddr **nam)
582 {
583 struct unpcb *unp;
584 const struct sockaddr *sa;
585
586 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
587 UNP_LOCK();
588 unp = sotounpcb(so);
589 if (unp == NULL) {
590 UNP_UNLOCK();
591 free(*nam, M_SONAME);
592 *nam = NULL;
593 return (EINVAL);
594 }
595 if (unp->unp_addr != NULL)
596 sa = (struct sockaddr *) unp->unp_addr;
597 else
598 sa = &sun_noname;
599 bcopy(sa, *nam, sa->sa_len);
600 UNP_UNLOCK();
601 return (0);
602 }
603
604 struct pr_usrreqs uipc_usrreqs = {
605 .pru_abort = uipc_abort,
606 .pru_accept = uipc_accept,
607 .pru_attach = uipc_attach,
608 .pru_bind = uipc_bind,
609 .pru_connect = uipc_connect,
610 .pru_connect2 = uipc_connect2,
611 .pru_detach = uipc_detach,
612 .pru_disconnect = uipc_disconnect,
613 .pru_listen = uipc_listen,
614 .pru_peeraddr = uipc_peeraddr,
615 .pru_rcvd = uipc_rcvd,
616 .pru_send = uipc_send,
617 .pru_sense = uipc_sense,
618 .pru_shutdown = uipc_shutdown,
619 .pru_sockaddr = uipc_sockaddr,
620 .pru_sosend = sosend,
621 .pru_soreceive = soreceive,
622 .pru_sopoll = sopoll,
623 };
624
625 int
626 uipc_ctloutput(struct socket *so, struct sockopt *sopt)
627 {
628 struct unpcb *unp;
629 struct xucred xu;
630 int error, optval;
631
632 if (sopt->sopt_level != 0)
633 return (EINVAL);
634
635 UNP_LOCK();
636 unp = sotounpcb(so);
637 if (unp == NULL) {
638 UNP_UNLOCK();
639 return (EINVAL);
640 }
641 error = 0;
642
643 switch (sopt->sopt_dir) {
644 case SOPT_GET:
645 switch (sopt->sopt_name) {
646 case LOCAL_PEERCRED:
647 if (unp->unp_flags & UNP_HAVEPC)
648 xu = unp->unp_peercred;
649 else {
650 if (so->so_type == SOCK_STREAM)
651 error = ENOTCONN;
652 else
653 error = EINVAL;
654 }
655 if (error == 0)
656 error = sooptcopyout(sopt, &xu, sizeof(xu));
657 break;
658 case LOCAL_CREDS:
659 optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0;
660 error = sooptcopyout(sopt, &optval, sizeof(optval));
661 break;
662 case LOCAL_CONNWAIT:
663 optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0;
664 error = sooptcopyout(sopt, &optval, sizeof(optval));
665 break;
666 default:
667 error = EOPNOTSUPP;
668 break;
669 }
670 break;
671 case SOPT_SET:
672 switch (sopt->sopt_name) {
673 case LOCAL_CREDS:
674 case LOCAL_CONNWAIT:
675 error = sooptcopyin(sopt, &optval, sizeof(optval),
676 sizeof(optval));
677 if (error)
678 break;
679
680 #define OPTSET(bit) \
681 if (optval) \
682 unp->unp_flags |= bit; \
683 else \
684 unp->unp_flags &= ~bit;
685
686 switch (sopt->sopt_name) {
687 case LOCAL_CREDS:
688 OPTSET(UNP_WANTCRED);
689 break;
690 case LOCAL_CONNWAIT:
691 OPTSET(UNP_CONNWAIT);
692 break;
693 default:
694 break;
695 }
696 break;
697 #undef OPTSET
698 default:
699 error = ENOPROTOOPT;
700 break;
701 }
702 break;
703 default:
704 error = EOPNOTSUPP;
705 break;
706 }
707 UNP_UNLOCK();
708 return (error);
709 }
710
711 /*
712 * Both send and receive buffers are allocated PIPSIZ bytes of buffering
713 * for stream sockets, although the total for sender and receiver is
714 * actually only PIPSIZ.
715 * Datagram sockets really use the sendspace as the maximum datagram size,
716 * and don't really want to reserve the sendspace. Their recvspace should
717 * be large enough for at least one max-size datagram plus address.
718 */
719 #ifndef PIPSIZ
720 #define PIPSIZ 8192
721 #endif
722 static u_long unpst_sendspace = PIPSIZ;
723 static u_long unpst_recvspace = PIPSIZ;
724 static u_long unpdg_sendspace = 2*1024; /* really max datagram size */
725 static u_long unpdg_recvspace = 4*1024;
726
727 static int unp_rights; /* file descriptors in flight */
728
729 SYSCTL_DECL(_net_local_stream);
730 SYSCTL_INT(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW,
731 &unpst_sendspace, 0, "");
732 SYSCTL_INT(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW,
733 &unpst_recvspace, 0, "");
734 SYSCTL_DECL(_net_local_dgram);
735 SYSCTL_INT(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW,
736 &unpdg_sendspace, 0, "");
737 SYSCTL_INT(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW,
738 &unpdg_recvspace, 0, "");
739 SYSCTL_DECL(_net_local);
740 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, "");
741
742 static int
743 unp_attach(struct socket *so)
744 {
745 struct unpcb *unp;
746 int error;
747
748 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
749 switch (so->so_type) {
750
751 case SOCK_STREAM:
752 error = soreserve(so, unpst_sendspace, unpst_recvspace);
753 break;
754
755 case SOCK_DGRAM:
756 error = soreserve(so, unpdg_sendspace, unpdg_recvspace);
757 break;
758
759 default:
760 panic("unp_attach");
761 }
762 if (error)
763 return (error);
764 }
765 unp = uma_zalloc(unp_zone, M_WAITOK | M_ZERO);
766 if (unp == NULL)
767 return (ENOBUFS);
768 LIST_INIT(&unp->unp_refs);
769 unp->unp_socket = so;
770 so->so_pcb = unp;
771
772 UNP_LOCK();
773 unp->unp_gencnt = ++unp_gencnt;
774 unp_count++;
775 LIST_INSERT_HEAD(so->so_type == SOCK_DGRAM ? &unp_dhead
776 : &unp_shead, unp, unp_link);
777 UNP_UNLOCK();
778
779 return (0);
780 }
781
782 static void
783 unp_detach(struct unpcb *unp)
784 {
785 struct vnode *vp;
786 int local_unp_rights;
787
788 UNP_LOCK_ASSERT();
789
790 LIST_REMOVE(unp, unp_link);
791 unp->unp_gencnt = ++unp_gencnt;
792 --unp_count;
793 if ((vp = unp->unp_vnode) != NULL) {
794 /*
795 * XXXRW: should v_socket be frobbed only while holding
796 * Giant?
797 */
798 unp->unp_vnode->v_socket = NULL;
799 unp->unp_vnode = NULL;
800 }
801 if (unp->unp_conn != NULL)
802 unp_disconnect(unp);
803 while (!LIST_EMPTY(&unp->unp_refs)) {
804 struct unpcb *ref = LIST_FIRST(&unp->unp_refs);
805 unp_drop(ref, ECONNRESET);
806 }
807 soisdisconnected(unp->unp_socket);
808 unp->unp_socket->so_pcb = NULL;
809 local_unp_rights = unp_rights;
810 UNP_UNLOCK();
811 if (unp->unp_addr != NULL)
812 FREE(unp->unp_addr, M_SONAME);
813 uma_zfree(unp_zone, unp);
814 if (vp) {
815 int vfslocked;
816
817 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
818 vrele(vp);
819 VFS_UNLOCK_GIANT(vfslocked);
820 }
821 if (local_unp_rights)
822 taskqueue_enqueue(taskqueue_thread, &unp_gc_task);
823 }
824
825 static int
826 unp_bind(struct unpcb *unp, struct sockaddr *nam, struct thread *td)
827 {
828 struct sockaddr_un *soun = (struct sockaddr_un *)nam;
829 struct vnode *vp;
830 struct mount *mp;
831 struct vattr vattr;
832 int error, namelen;
833 struct nameidata nd;
834 char *buf;
835
836 UNP_LOCK_ASSERT();
837
838 /*
839 * XXXRW: This test-and-set of unp_vnode is non-atomic; the
840 * unlocked read here is fine, but the value of unp_vnode needs
841 * to be tested again after we do all the lookups to see if the
842 * pcb is still unbound?
843 */
844 if (unp->unp_vnode != NULL)
845 return (EINVAL);
846
847 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path);
848 if (namelen <= 0)
849 return (EINVAL);
850
851 UNP_UNLOCK();
852
853 buf = malloc(namelen + 1, M_TEMP, M_WAITOK);
854 strlcpy(buf, soun->sun_path, namelen + 1);
855
856 mtx_lock(&Giant);
857 restart:
858 mtx_assert(&Giant, MA_OWNED);
859 NDINIT(&nd, CREATE, NOFOLLOW | LOCKPARENT | SAVENAME, UIO_SYSSPACE,
860 buf, td);
861 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
862 error = namei(&nd);
863 if (error)
864 goto done;
865 vp = nd.ni_vp;
866 if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
867 NDFREE(&nd, NDF_ONLY_PNBUF);
868 if (nd.ni_dvp == vp)
869 vrele(nd.ni_dvp);
870 else
871 vput(nd.ni_dvp);
872 if (vp != NULL) {
873 vrele(vp);
874 error = EADDRINUSE;
875 goto done;
876 }
877 error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH);
878 if (error)
879 goto done;
880 goto restart;
881 }
882 VATTR_NULL(&vattr);
883 vattr.va_type = VSOCK;
884 vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask);
885 #ifdef MAC
886 error = mac_check_vnode_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd,
887 &vattr);
888 #endif
889 if (error == 0) {
890 VOP_LEASE(nd.ni_dvp, td, td->td_ucred, LEASE_WRITE);
891 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
892 }
893 NDFREE(&nd, NDF_ONLY_PNBUF);
894 vput(nd.ni_dvp);
895 if (error) {
896 vn_finished_write(mp);
897 goto done;
898 }
899 vp = nd.ni_vp;
900 ASSERT_VOP_LOCKED(vp, "unp_bind");
901 soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK);
902 UNP_LOCK();
903 vp->v_socket = unp->unp_socket;
904 unp->unp_vnode = vp;
905 unp->unp_addr = soun;
906 UNP_UNLOCK();
907 VOP_UNLOCK(vp, 0, td);
908 vn_finished_write(mp);
909 done:
910 mtx_unlock(&Giant);
911 free(buf, M_TEMP);
912 UNP_LOCK();
913 return (error);
914 }
915
916 static int
917 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
918 {
919 struct sockaddr_un *soun = (struct sockaddr_un *)nam;
920 struct vnode *vp;
921 struct socket *so2, *so3;
922 struct unpcb *unp, *unp2, *unp3;
923 int error, len;
924 struct nameidata nd;
925 char buf[SOCK_MAXADDRLEN];
926 struct sockaddr *sa;
927
928 UNP_LOCK_ASSERT();
929 unp = sotounpcb(so);
930
931 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
932 if (len <= 0)
933 return (EINVAL);
934 strlcpy(buf, soun->sun_path, len + 1);
935 UNP_UNLOCK();
936 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
937 mtx_lock(&Giant);
938 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, buf, td);
939 error = namei(&nd);
940 if (error)
941 vp = NULL;
942 else
943 vp = nd.ni_vp;
944 ASSERT_VOP_LOCKED(vp, "unp_connect");
945 NDFREE(&nd, NDF_ONLY_PNBUF);
946 if (error)
947 goto bad;
948
949 if (vp->v_type != VSOCK) {
950 error = ENOTSOCK;
951 goto bad;
952 }
953 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td);
954 if (error)
955 goto bad;
956 mtx_unlock(&Giant);
957 UNP_LOCK();
958 unp = sotounpcb(so);
959 if (unp == NULL) {
960 error = EINVAL;
961 goto bad2;
962 }
963 so2 = vp->v_socket;
964 if (so2 == NULL) {
965 error = ECONNREFUSED;
966 goto bad2;
967 }
968 if (so->so_type != so2->so_type) {
969 error = EPROTOTYPE;
970 goto bad2;
971 }
972 if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
973 if (so2->so_options & SO_ACCEPTCONN) {
974 /*
975 * NB: drop locks here so unp_attach is entered
976 * w/o locks; this avoids a recursive lock
977 * of the head and holding sleep locks across
978 * a (potentially) blocking malloc.
979 */
980 UNP_UNLOCK();
981 so3 = sonewconn(so2, 0);
982 UNP_LOCK();
983 } else
984 so3 = NULL;
985 if (so3 == NULL) {
986 error = ECONNREFUSED;
987 goto bad2;
988 }
989 unp = sotounpcb(so);
990 unp2 = sotounpcb(so2);
991 unp3 = sotounpcb(so3);
992 if (unp2->unp_addr != NULL) {
993 bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len);
994 unp3->unp_addr = (struct sockaddr_un *) sa;
995 sa = NULL;
996 }
997 /*
998 * unp_peercred management:
999 *
1000 * The connecter's (client's) credentials are copied
1001 * from its process structure at the time of connect()
1002 * (which is now).
1003 */
1004 cru2x(td->td_ucred, &unp3->unp_peercred);
1005 unp3->unp_flags |= UNP_HAVEPC;
1006 /*
1007 * The receiver's (server's) credentials are copied
1008 * from the unp_peercred member of socket on which the
1009 * former called listen(); unp_listen() cached that
1010 * process's credentials at that time so we can use
1011 * them now.
1012 */
1013 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED,
1014 ("unp_connect: listener without cached peercred"));
1015 memcpy(&unp->unp_peercred, &unp2->unp_peercred,
1016 sizeof(unp->unp_peercred));
1017 unp->unp_flags |= UNP_HAVEPC;
1018 if (unp2->unp_flags & UNP_WANTCRED)
1019 unp3->unp_flags |= UNP_WANTCRED;
1020 #ifdef MAC
1021 SOCK_LOCK(so);
1022 mac_set_socket_peer_from_socket(so, so3);
1023 mac_set_socket_peer_from_socket(so3, so);
1024 SOCK_UNLOCK(so);
1025 #endif
1026
1027 so2 = so3;
1028 }
1029 error = unp_connect2(so, so2, PRU_CONNECT);
1030 bad2:
1031 UNP_UNLOCK();
1032 mtx_lock(&Giant);
1033 bad:
1034 mtx_assert(&Giant, MA_OWNED);
1035 if (vp != NULL)
1036 vput(vp);
1037 mtx_unlock(&Giant);
1038 free(sa, M_SONAME);
1039 UNP_LOCK();
1040 return (error);
1041 }
1042
1043 static int
1044 unp_connect2(struct socket *so, struct socket *so2, int req)
1045 {
1046 struct unpcb *unp = sotounpcb(so);
1047 struct unpcb *unp2;
1048
1049 UNP_LOCK_ASSERT();
1050
1051 if (so2->so_type != so->so_type)
1052 return (EPROTOTYPE);
1053 unp2 = sotounpcb(so2);
1054 unp->unp_conn = unp2;
1055 switch (so->so_type) {
1056
1057 case SOCK_DGRAM:
1058 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
1059 soisconnected(so);
1060 break;
1061
1062 case SOCK_STREAM:
1063 unp2->unp_conn = unp;
1064 if (req == PRU_CONNECT &&
1065 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT))
1066 soisconnecting(so);
1067 else
1068 soisconnected(so);
1069 soisconnected(so2);
1070 break;
1071
1072 default:
1073 panic("unp_connect2");
1074 }
1075 return (0);
1076 }
1077
1078 static void
1079 unp_disconnect(struct unpcb *unp)
1080 {
1081 struct unpcb *unp2 = unp->unp_conn;
1082 struct socket *so;
1083
1084 UNP_LOCK_ASSERT();
1085
1086 if (unp2 == NULL)
1087 return;
1088 unp->unp_conn = NULL;
1089 switch (unp->unp_socket->so_type) {
1090
1091 case SOCK_DGRAM:
1092 LIST_REMOVE(unp, unp_reflink);
1093 so = unp->unp_socket;
1094 SOCK_LOCK(so);
1095 so->so_state &= ~SS_ISCONNECTED;
1096 SOCK_UNLOCK(so);
1097 break;
1098
1099 case SOCK_STREAM:
1100 soisdisconnected(unp->unp_socket);
1101 unp2->unp_conn = NULL;
1102 soisdisconnected(unp2->unp_socket);
1103 break;
1104 }
1105 }
1106
1107 #ifdef notdef
1108 void
1109 unp_abort(struct unpcb *unp)
1110 {
1111
1112 unp_detach(unp);
1113 UNP_UNLOCK_ASSERT();
1114 }
1115 #endif
1116
1117 /*
1118 * unp_pcblist() assumes that UNIX domain socket memory is never reclaimed
1119 * by the zone (UMA_ZONE_NOFREE), and as such potentially stale pointers
1120 * are safe to reference. It first scans the list of struct unpcb's to
1121 * generate a pointer list, then it rescans its list one entry at a time to
1122 * externalize and copyout. It checks the generation number to see if a
1123 * struct unpcb has been reused, and will skip it if so.
1124 */
1125 static int
1126 unp_pcblist(SYSCTL_HANDLER_ARGS)
1127 {
1128 int error, i, n;
1129 struct unpcb *unp, **unp_list;
1130 unp_gen_t gencnt;
1131 struct xunpgen *xug;
1132 struct unp_head *head;
1133 struct xunpcb *xu;
1134
1135 head = ((intptr_t)arg1 == SOCK_DGRAM ? &unp_dhead : &unp_shead);
1136
1137 /*
1138 * The process of preparing the PCB list is too time-consuming and
1139 * resource-intensive to repeat twice on every request.
1140 */
1141 if (req->oldptr == NULL) {
1142 n = unp_count;
1143 req->oldidx = 2 * (sizeof *xug)
1144 + (n + n/8) * sizeof(struct xunpcb);
1145 return (0);
1146 }
1147
1148 if (req->newptr != NULL)
1149 return (EPERM);
1150
1151 /*
1152 * OK, now we're committed to doing something.
1153 */
1154 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK);
1155 UNP_LOCK();
1156 gencnt = unp_gencnt;
1157 n = unp_count;
1158 UNP_UNLOCK();
1159
1160 xug->xug_len = sizeof *xug;
1161 xug->xug_count = n;
1162 xug->xug_gen = gencnt;
1163 xug->xug_sogen = so_gencnt;
1164 error = SYSCTL_OUT(req, xug, sizeof *xug);
1165 if (error) {
1166 free(xug, M_TEMP);
1167 return (error);
1168 }
1169
1170 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
1171
1172 UNP_LOCK();
1173 for (unp = LIST_FIRST(head), i = 0; unp && i < n;
1174 unp = LIST_NEXT(unp, unp_link)) {
1175 if (unp->unp_gencnt <= gencnt) {
1176 if (cr_cansee(req->td->td_ucred,
1177 unp->unp_socket->so_cred))
1178 continue;
1179 unp_list[i++] = unp;
1180 }
1181 }
1182 UNP_UNLOCK();
1183 n = i; /* in case we lost some during malloc */
1184
1185 error = 0;
1186 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO);
1187 for (i = 0; i < n; i++) {
1188 unp = unp_list[i];
1189 if (unp->unp_gencnt <= gencnt) {
1190 xu->xu_len = sizeof *xu;
1191 xu->xu_unpp = unp;
1192 /*
1193 * XXX - need more locking here to protect against
1194 * connect/disconnect races for SMP.
1195 */
1196 if (unp->unp_addr != NULL)
1197 bcopy(unp->unp_addr, &xu->xu_addr,
1198 unp->unp_addr->sun_len);
1199 if (unp->unp_conn != NULL &&
1200 unp->unp_conn->unp_addr != NULL)
1201 bcopy(unp->unp_conn->unp_addr,
1202 &xu->xu_caddr,
1203 unp->unp_conn->unp_addr->sun_len);
1204 bcopy(unp, &xu->xu_unp, sizeof *unp);
1205 sotoxsocket(unp->unp_socket, &xu->xu_socket);
1206 error = SYSCTL_OUT(req, xu, sizeof *xu);
1207 }
1208 }
1209 free(xu, M_TEMP);
1210 if (!error) {
1211 /*
1212 * Give the user an updated idea of our state.
1213 * If the generation differs from what we told
1214 * her before, she knows that something happened
1215 * while we were processing this request, and it
1216 * might be necessary to retry.
1217 */
1218 xug->xug_gen = unp_gencnt;
1219 xug->xug_sogen = so_gencnt;
1220 xug->xug_count = unp_count;
1221 error = SYSCTL_OUT(req, xug, sizeof *xug);
1222 }
1223 free(unp_list, M_TEMP);
1224 free(xug, M_TEMP);
1225 return (error);
1226 }
1227
1228 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLFLAG_RD,
1229 (caddr_t)(long)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
1230 "List of active local datagram sockets");
1231 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLFLAG_RD,
1232 (caddr_t)(long)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
1233 "List of active local stream sockets");
1234
1235 static void
1236 unp_shutdown(struct unpcb *unp)
1237 {
1238 struct socket *so;
1239
1240 UNP_LOCK_ASSERT();
1241
1242 if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn &&
1243 (so = unp->unp_conn->unp_socket))
1244 socantrcvmore(so);
1245 }
1246
1247 static void
1248 unp_drop(struct unpcb *unp, int errno)
1249 {
1250 struct socket *so = unp->unp_socket;
1251
1252 UNP_LOCK_ASSERT();
1253
1254 so->so_error = errno;
1255 unp_disconnect(unp);
1256 }
1257
1258 #ifdef notdef
1259 void
1260 unp_drain(void)
1261 {
1262
1263 }
1264 #endif
1265
1266 static void
1267 unp_freerights(struct file **rp, int fdcount)
1268 {
1269 int i;
1270 struct file *fp;
1271
1272 for (i = 0; i < fdcount; i++) {
1273 fp = *rp;
1274 /*
1275 * zero the pointer before calling
1276 * unp_discard since it may end up
1277 * in unp_gc()..
1278 */
1279 *rp++ = 0;
1280 unp_discard(fp);
1281 }
1282 }
1283
1284 int
1285 unp_externalize(struct mbuf *control, struct mbuf **controlp)
1286 {
1287 struct thread *td = curthread; /* XXX */
1288 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1289 int i;
1290 int *fdp;
1291 struct file **rp;
1292 struct file *fp;
1293 void *data;
1294 socklen_t clen = control->m_len, datalen;
1295 int error, newfds;
1296 int f;
1297 u_int newlen;
1298
1299 UNP_UNLOCK_ASSERT();
1300
1301 error = 0;
1302 if (controlp != NULL) /* controlp == NULL => free control messages */
1303 *controlp = NULL;
1304
1305 while (cm != NULL) {
1306 if (sizeof(*cm) > clen || cm->cmsg_len > clen) {
1307 error = EINVAL;
1308 break;
1309 }
1310
1311 data = CMSG_DATA(cm);
1312 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1313
1314 if (cm->cmsg_level == SOL_SOCKET
1315 && cm->cmsg_type == SCM_RIGHTS) {
1316 newfds = datalen / sizeof(struct file *);
1317 rp = data;
1318
1319 /* If we're not outputting the descriptors free them. */
1320 if (error || controlp == NULL) {
1321 unp_freerights(rp, newfds);
1322 goto next;
1323 }
1324 FILEDESC_LOCK(td->td_proc->p_fd);
1325 /* if the new FD's will not fit free them. */
1326 if (!fdavail(td, newfds)) {
1327 FILEDESC_UNLOCK(td->td_proc->p_fd);
1328 error = EMSGSIZE;
1329 unp_freerights(rp, newfds);
1330 goto next;
1331 }
1332 /*
1333 * now change each pointer to an fd in the global
1334 * table to an integer that is the index to the
1335 * local fd table entry that we set up to point
1336 * to the global one we are transferring.
1337 */
1338 newlen = newfds * sizeof(int);
1339 *controlp = sbcreatecontrol(NULL, newlen,
1340 SCM_RIGHTS, SOL_SOCKET);
1341 if (*controlp == NULL) {
1342 FILEDESC_UNLOCK(td->td_proc->p_fd);
1343 error = E2BIG;
1344 unp_freerights(rp, newfds);
1345 goto next;
1346 }
1347
1348 fdp = (int *)
1349 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1350 for (i = 0; i < newfds; i++) {
1351 if (fdalloc(td, 0, &f))
1352 panic("unp_externalize fdalloc failed");
1353 fp = *rp++;
1354 td->td_proc->p_fd->fd_ofiles[f] = fp;
1355 FILE_LOCK(fp);
1356 fp->f_msgcount--;
1357 FILE_UNLOCK(fp);
1358 unp_rights--;
1359 *fdp++ = f;
1360 }
1361 FILEDESC_UNLOCK(td->td_proc->p_fd);
1362 } else { /* We can just copy anything else across */
1363 if (error || controlp == NULL)
1364 goto next;
1365 *controlp = sbcreatecontrol(NULL, datalen,
1366 cm->cmsg_type, cm->cmsg_level);
1367 if (*controlp == NULL) {
1368 error = ENOBUFS;
1369 goto next;
1370 }
1371 bcopy(data,
1372 CMSG_DATA(mtod(*controlp, struct cmsghdr *)),
1373 datalen);
1374 }
1375
1376 controlp = &(*controlp)->m_next;
1377
1378 next:
1379 if (CMSG_SPACE(datalen) < clen) {
1380 clen -= CMSG_SPACE(datalen);
1381 cm = (struct cmsghdr *)
1382 ((caddr_t)cm + CMSG_SPACE(datalen));
1383 } else {
1384 clen = 0;
1385 cm = NULL;
1386 }
1387 }
1388
1389 m_freem(control);
1390
1391 return (error);
1392 }
1393
1394 static void
1395 unp_zone_change(void *tag)
1396 {
1397
1398 uma_zone_set_max(unp_zone, maxsockets);
1399 }
1400
1401 void
1402 unp_init(void)
1403 {
1404 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL,
1405 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
1406 if (unp_zone == NULL)
1407 panic("unp_init");
1408 uma_zone_set_max(unp_zone, maxsockets);
1409 EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change,
1410 NULL, EVENTHANDLER_PRI_ANY);
1411 LIST_INIT(&unp_dhead);
1412 LIST_INIT(&unp_shead);
1413 TASK_INIT(&unp_gc_task, 0, unp_gc, NULL);
1414 UNP_LOCK_INIT();
1415 }
1416
1417 static int
1418 unp_internalize(struct mbuf **controlp, struct thread *td)
1419 {
1420 struct mbuf *control = *controlp;
1421 struct proc *p = td->td_proc;
1422 struct filedesc *fdescp = p->p_fd;
1423 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1424 struct cmsgcred *cmcred;
1425 struct file **rp;
1426 struct file *fp;
1427 struct timeval *tv;
1428 int i, fd, *fdp;
1429 void *data;
1430 socklen_t clen = control->m_len, datalen;
1431 int error, oldfds;
1432 u_int newlen;
1433
1434 UNP_UNLOCK_ASSERT();
1435
1436 error = 0;
1437 *controlp = NULL;
1438
1439 while (cm != NULL) {
1440 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET
1441 || cm->cmsg_len > clen) {
1442 error = EINVAL;
1443 goto out;
1444 }
1445
1446 data = CMSG_DATA(cm);
1447 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1448
1449 switch (cm->cmsg_type) {
1450 /*
1451 * Fill in credential information.
1452 */
1453 case SCM_CREDS:
1454 *controlp = sbcreatecontrol(NULL, sizeof(*cmcred),
1455 SCM_CREDS, SOL_SOCKET);
1456 if (*controlp == NULL) {
1457 error = ENOBUFS;
1458 goto out;
1459 }
1460
1461 cmcred = (struct cmsgcred *)
1462 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1463 cmcred->cmcred_pid = p->p_pid;
1464 cmcred->cmcred_uid = td->td_ucred->cr_ruid;
1465 cmcred->cmcred_gid = td->td_ucred->cr_rgid;
1466 cmcred->cmcred_euid = td->td_ucred->cr_uid;
1467 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups,
1468 CMGROUP_MAX);
1469 for (i = 0; i < cmcred->cmcred_ngroups; i++)
1470 cmcred->cmcred_groups[i] =
1471 td->td_ucred->cr_groups[i];
1472 break;
1473
1474 case SCM_RIGHTS:
1475 oldfds = datalen / sizeof (int);
1476 /*
1477 * check that all the FDs passed in refer to legal files
1478 * If not, reject the entire operation.
1479 */
1480 fdp = data;
1481 FILEDESC_LOCK(fdescp);
1482 for (i = 0; i < oldfds; i++) {
1483 fd = *fdp++;
1484 if ((unsigned)fd >= fdescp->fd_nfiles ||
1485 fdescp->fd_ofiles[fd] == NULL) {
1486 FILEDESC_UNLOCK(fdescp);
1487 error = EBADF;
1488 goto out;
1489 }
1490 fp = fdescp->fd_ofiles[fd];
1491 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) {
1492 FILEDESC_UNLOCK(fdescp);
1493 error = EOPNOTSUPP;
1494 goto out;
1495 }
1496
1497 }
1498 /*
1499 * Now replace the integer FDs with pointers to
1500 * the associated global file table entry..
1501 */
1502 newlen = oldfds * sizeof(struct file *);
1503 *controlp = sbcreatecontrol(NULL, newlen,
1504 SCM_RIGHTS, SOL_SOCKET);
1505 if (*controlp == NULL) {
1506 FILEDESC_UNLOCK(fdescp);
1507 error = E2BIG;
1508 goto out;
1509 }
1510
1511 fdp = data;
1512 rp = (struct file **)
1513 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1514 for (i = 0; i < oldfds; i++) {
1515 fp = fdescp->fd_ofiles[*fdp++];
1516 *rp++ = fp;
1517 FILE_LOCK(fp);
1518 fp->f_count++;
1519 fp->f_msgcount++;
1520 FILE_UNLOCK(fp);
1521 unp_rights++;
1522 }
1523 FILEDESC_UNLOCK(fdescp);
1524 break;
1525
1526 case SCM_TIMESTAMP:
1527 *controlp = sbcreatecontrol(NULL, sizeof(*tv),
1528 SCM_TIMESTAMP, SOL_SOCKET);
1529 if (*controlp == NULL) {
1530 error = ENOBUFS;
1531 goto out;
1532 }
1533 tv = (struct timeval *)
1534 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1535 microtime(tv);
1536 break;
1537
1538 default:
1539 error = EINVAL;
1540 goto out;
1541 }
1542
1543 controlp = &(*controlp)->m_next;
1544
1545 if (CMSG_SPACE(datalen) < clen) {
1546 clen -= CMSG_SPACE(datalen);
1547 cm = (struct cmsghdr *)
1548 ((caddr_t)cm + CMSG_SPACE(datalen));
1549 } else {
1550 clen = 0;
1551 cm = NULL;
1552 }
1553 }
1554
1555 out:
1556 m_freem(control);
1557
1558 return (error);
1559 }
1560
1561 struct mbuf *
1562 unp_addsockcred(struct thread *td, struct mbuf *control)
1563 {
1564 struct mbuf *m, *n, *n_prev;
1565 struct sockcred *sc;
1566 const struct cmsghdr *cm;
1567 int ngroups;
1568 int i;
1569
1570 ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX);
1571
1572 m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET);
1573 if (m == NULL)
1574 return (control);
1575
1576 sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *));
1577 sc->sc_uid = td->td_ucred->cr_ruid;
1578 sc->sc_euid = td->td_ucred->cr_uid;
1579 sc->sc_gid = td->td_ucred->cr_rgid;
1580 sc->sc_egid = td->td_ucred->cr_gid;
1581 sc->sc_ngroups = ngroups;
1582 for (i = 0; i < sc->sc_ngroups; i++)
1583 sc->sc_groups[i] = td->td_ucred->cr_groups[i];
1584
1585 /*
1586 * Unlink SCM_CREDS control messages (struct cmsgcred), since
1587 * just created SCM_CREDS control message (struct sockcred) has
1588 * another format.
1589 */
1590 if (control != NULL)
1591 for (n = control, n_prev = NULL; n != NULL;) {
1592 cm = mtod(n, struct cmsghdr *);
1593 if (cm->cmsg_level == SOL_SOCKET &&
1594 cm->cmsg_type == SCM_CREDS) {
1595 if (n_prev == NULL)
1596 control = n->m_next;
1597 else
1598 n_prev->m_next = n->m_next;
1599 n = m_free(n);
1600 } else {
1601 n_prev = n;
1602 n = n->m_next;
1603 }
1604 }
1605
1606 /* Prepend it to the head. */
1607 m->m_next = control;
1608
1609 return (m);
1610 }
1611
1612 /*
1613 * unp_defer indicates whether additional work has been defered for a future
1614 * pass through unp_gc(). It is thread local and does not require explicit
1615 * synchronization.
1616 */
1617 static int unp_defer;
1618
1619 static int unp_taskcount;
1620 SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0, "");
1621
1622 static int unp_recycled;
1623 SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0, "");
1624
1625 static void
1626 unp_gc(__unused void *arg, int pending)
1627 {
1628 struct file *fp, *nextfp;
1629 struct socket *so;
1630 struct file **extra_ref, **fpp;
1631 int nunref, i;
1632 int nfiles_snap;
1633 int nfiles_slack = 20;
1634
1635 unp_taskcount++;
1636 unp_defer = 0;
1637 /*
1638 * before going through all this, set all FDs to
1639 * be NOT defered and NOT externally accessible
1640 */
1641 sx_slock(&filelist_lock);
1642 LIST_FOREACH(fp, &filehead, f_list)
1643 fp->f_gcflag &= ~(FMARK|FDEFER);
1644 do {
1645 LIST_FOREACH(fp, &filehead, f_list) {
1646 FILE_LOCK(fp);
1647 /*
1648 * If the file is not open, skip it -- could be a
1649 * file in the process of being opened, or in the
1650 * process of being closed. If the file is
1651 * "closing", it may have been marked for deferred
1652 * consideration. Clear the flag now if so.
1653 */
1654 if (fp->f_count == 0) {
1655 if (fp->f_gcflag & FDEFER)
1656 unp_defer--;
1657 fp->f_gcflag &= ~(FMARK|FDEFER);
1658 FILE_UNLOCK(fp);
1659 continue;
1660 }
1661 /*
1662 * If we already marked it as 'defer' in a
1663 * previous pass, then try process it this time
1664 * and un-mark it
1665 */
1666 if (fp->f_gcflag & FDEFER) {
1667 fp->f_gcflag &= ~FDEFER;
1668 unp_defer--;
1669 } else {
1670 /*
1671 * if it's not defered, then check if it's
1672 * already marked.. if so skip it
1673 */
1674 if (fp->f_gcflag & FMARK) {
1675 FILE_UNLOCK(fp);
1676 continue;
1677 }
1678 /*
1679 * If all references are from messages
1680 * in transit, then skip it. it's not
1681 * externally accessible.
1682 */
1683 if (fp->f_count == fp->f_msgcount) {
1684 FILE_UNLOCK(fp);
1685 continue;
1686 }
1687 /*
1688 * If it got this far then it must be
1689 * externally accessible.
1690 */
1691 fp->f_gcflag |= FMARK;
1692 }
1693 /*
1694 * either it was defered, or it is externally
1695 * accessible and not already marked so.
1696 * Now check if it is possibly one of OUR sockets.
1697 */
1698 if (fp->f_type != DTYPE_SOCKET ||
1699 (so = fp->f_data) == NULL) {
1700 FILE_UNLOCK(fp);
1701 continue;
1702 }
1703 FILE_UNLOCK(fp);
1704 if (so->so_proto->pr_domain != &localdomain ||
1705 (so->so_proto->pr_flags&PR_RIGHTS) == 0)
1706 continue;
1707 /*
1708 * So, Ok, it's one of our sockets and it IS externally
1709 * accessible (or was defered). Now we look
1710 * to see if we hold any file descriptors in its
1711 * message buffers. Follow those links and mark them
1712 * as accessible too.
1713 */
1714 SOCKBUF_LOCK(&so->so_rcv);
1715 unp_scan(so->so_rcv.sb_mb, unp_mark);
1716 SOCKBUF_UNLOCK(&so->so_rcv);
1717 }
1718 } while (unp_defer);
1719 sx_sunlock(&filelist_lock);
1720 /*
1721 * XXXRW: The following comments need updating for a post-SMPng and
1722 * deferred unp_gc() world, but are still generally accurate.
1723 *
1724 * We grab an extra reference to each of the file table entries
1725 * that are not otherwise accessible and then free the rights
1726 * that are stored in messages on them.
1727 *
1728 * The bug in the orginal code is a little tricky, so I'll describe
1729 * what's wrong with it here.
1730 *
1731 * It is incorrect to simply unp_discard each entry for f_msgcount
1732 * times -- consider the case of sockets A and B that contain
1733 * references to each other. On a last close of some other socket,
1734 * we trigger a gc since the number of outstanding rights (unp_rights)
1735 * is non-zero. If during the sweep phase the gc code unp_discards,
1736 * we end up doing a (full) closef on the descriptor. A closef on A
1737 * results in the following chain. Closef calls soo_close, which
1738 * calls soclose. Soclose calls first (through the switch
1739 * uipc_usrreq) unp_detach, which re-invokes unp_gc. Unp_gc simply
1740 * returns because the previous instance had set unp_gcing, and
1741 * we return all the way back to soclose, which marks the socket
1742 * with SS_NOFDREF, and then calls sofree. Sofree calls sorflush
1743 * to free up the rights that are queued in messages on the socket A,
1744 * i.e., the reference on B. The sorflush calls via the dom_dispose
1745 * switch unp_dispose, which unp_scans with unp_discard. This second
1746 * instance of unp_discard just calls closef on B.
1747 *
1748 * Well, a similar chain occurs on B, resulting in a sorflush on B,
1749 * which results in another closef on A. Unfortunately, A is already
1750 * being closed, and the descriptor has already been marked with
1751 * SS_NOFDREF, and soclose panics at this point.
1752 *
1753 * Here, we first take an extra reference to each inaccessible
1754 * descriptor. Then, we call sorflush ourself, since we know
1755 * it is a Unix domain socket anyhow. After we destroy all the
1756 * rights carried in messages, we do a last closef to get rid
1757 * of our extra reference. This is the last close, and the
1758 * unp_detach etc will shut down the socket.
1759 *
1760 * 91/09/19, bsy@cs.cmu.edu
1761 */
1762 again:
1763 nfiles_snap = openfiles + nfiles_slack; /* some slack */
1764 extra_ref = malloc(nfiles_snap * sizeof(struct file *), M_TEMP,
1765 M_WAITOK);
1766 sx_slock(&filelist_lock);
1767 if (nfiles_snap < openfiles) {
1768 sx_sunlock(&filelist_lock);
1769 free(extra_ref, M_TEMP);
1770 nfiles_slack += 20;
1771 goto again;
1772 }
1773 for (nunref = 0, fp = LIST_FIRST(&filehead), fpp = extra_ref;
1774 fp != NULL; fp = nextfp) {
1775 nextfp = LIST_NEXT(fp, f_list);
1776 FILE_LOCK(fp);
1777 /*
1778 * If it's not open, skip it
1779 */
1780 if (fp->f_count == 0) {
1781 FILE_UNLOCK(fp);
1782 continue;
1783 }
1784 /*
1785 * If all refs are from msgs, and it's not marked accessible
1786 * then it must be referenced from some unreachable cycle
1787 * of (shut-down) FDs, so include it in our
1788 * list of FDs to remove
1789 */
1790 if (fp->f_count == fp->f_msgcount && !(fp->f_gcflag & FMARK)) {
1791 *fpp++ = fp;
1792 nunref++;
1793 fp->f_count++;
1794 }
1795 FILE_UNLOCK(fp);
1796 }
1797 sx_sunlock(&filelist_lock);
1798 /*
1799 * for each FD on our hit list, do the following two things
1800 */
1801 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) {
1802 struct file *tfp = *fpp;
1803 FILE_LOCK(tfp);
1804 if (tfp->f_type == DTYPE_SOCKET &&
1805 tfp->f_data != NULL) {
1806 FILE_UNLOCK(tfp);
1807 sorflush(tfp->f_data);
1808 } else {
1809 FILE_UNLOCK(tfp);
1810 }
1811 }
1812 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) {
1813 closef(*fpp, (struct thread *) NULL);
1814 unp_recycled++;
1815 }
1816 free(extra_ref, M_TEMP);
1817 }
1818
1819 void
1820 unp_dispose(struct mbuf *m)
1821 {
1822
1823 if (m)
1824 unp_scan(m, unp_discard);
1825 }
1826
1827 static int
1828 unp_listen(struct socket *so, struct unpcb *unp, struct thread *td)
1829 {
1830 int error;
1831
1832 UNP_LOCK_ASSERT();
1833
1834 SOCK_LOCK(so);
1835 error = solisten_proto_check(so);
1836 if (error == 0) {
1837 cru2x(td->td_ucred, &unp->unp_peercred);
1838 unp->unp_flags |= UNP_HAVEPCCACHED;
1839 solisten_proto(so);
1840 }
1841 SOCK_UNLOCK(so);
1842 return (error);
1843 }
1844
1845 static void
1846 unp_scan(struct mbuf *m0, void (*op)(struct file *))
1847 {
1848 struct mbuf *m;
1849 struct file **rp;
1850 struct cmsghdr *cm;
1851 void *data;
1852 int i;
1853 socklen_t clen, datalen;
1854 int qfds;
1855
1856 while (m0 != NULL) {
1857 for (m = m0; m; m = m->m_next) {
1858 if (m->m_type != MT_CONTROL)
1859 continue;
1860
1861 cm = mtod(m, struct cmsghdr *);
1862 clen = m->m_len;
1863
1864 while (cm != NULL) {
1865 if (sizeof(*cm) > clen || cm->cmsg_len > clen)
1866 break;
1867
1868 data = CMSG_DATA(cm);
1869 datalen = (caddr_t)cm + cm->cmsg_len
1870 - (caddr_t)data;
1871
1872 if (cm->cmsg_level == SOL_SOCKET &&
1873 cm->cmsg_type == SCM_RIGHTS) {
1874 qfds = datalen / sizeof (struct file *);
1875 rp = data;
1876 for (i = 0; i < qfds; i++)
1877 (*op)(*rp++);
1878 }
1879
1880 if (CMSG_SPACE(datalen) < clen) {
1881 clen -= CMSG_SPACE(datalen);
1882 cm = (struct cmsghdr *)
1883 ((caddr_t)cm + CMSG_SPACE(datalen));
1884 } else {
1885 clen = 0;
1886 cm = NULL;
1887 }
1888 }
1889 }
1890 m0 = m0->m_act;
1891 }
1892 }
1893
1894 static void
1895 unp_mark(struct file *fp)
1896 {
1897 if (fp->f_gcflag & FMARK)
1898 return;
1899 unp_defer++;
1900 fp->f_gcflag |= (FMARK|FDEFER);
1901 }
1902
1903 static void
1904 unp_discard(struct file *fp)
1905 {
1906 UNP_LOCK();
1907 FILE_LOCK(fp);
1908 fp->f_msgcount--;
1909 unp_rights--;
1910 FILE_UNLOCK(fp);
1911 UNP_UNLOCK();
1912 (void) closef(fp, (struct thread *)NULL);
1913 }
Cache object: 9119daa1bd96110191579f8c194f0d72
|