1/*-
2 * Copyright (c) 1982, 1986, 1989, 1991, 1993
3 * The Regents of the University of California.
4 * Copyright 2004-2006 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: head/sys/kern/uipc_usrreq.c 157370 2006-04-01 15:42:02Z rwatson $");
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/file.h>
44#include <sys/filedesc.h>
45#include <sys/jail.h>
46#include <sys/kernel.h>
47#include <sys/lock.h>
48#include <sys/mac.h>
49#include <sys/mbuf.h>
50#include <sys/mount.h>
51#include <sys/mutex.h>
52#include <sys/namei.h>
53#include <sys/proc.h>
54#include <sys/protosw.h>
55#include <sys/resourcevar.h>
56#include <sys/socket.h>
57#include <sys/socketvar.h>
58#include <sys/signalvar.h>
59#include <sys/stat.h>
60#include <sys/sx.h>
61#include <sys/sysctl.h>
62#include <sys/systm.h>
63#include <sys/taskqueue.h>
64#include <sys/un.h>
65#include <sys/unpcb.h>
66#include <sys/vnode.h>
67
68#include <vm/uma.h>
69
70static uma_zone_t unp_zone;
71static unp_gen_t unp_gencnt;
72static u_int unp_count;
73
74static struct unp_head unp_shead, unp_dhead;
75
76/*
77 * Unix communications domain.
78 *
79 * TODO:
80 * SEQPACKET, RDM
81 * rethink name space problems
82 * need a proper out-of-band
83 * lock pushdown
84 */
85static const struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL };
86static ino_t unp_ino; /* prototype for fake inode numbers */
87struct mbuf *unp_addsockcred(struct thread *, struct mbuf *);
88
89/*
90 * Currently, UNIX domain sockets are protected by a single subsystem lock,
91 * which covers global data structures and variables, the contents of each
92 * per-socket unpcb structure, and the so_pcb field in sockets attached to
93 * the UNIX domain. This provides for a moderate degree of paralellism, as
94 * receive operations on UNIX domain sockets do not need to acquire the
95 * subsystem lock. Finer grained locking to permit send() without acquiring
96 * a global lock would be a logical next step.
97 *
98 * The UNIX domain socket lock preceds all socket layer locks, including the
99 * socket lock and socket buffer lock, permitting UNIX domain socket code to
100 * call into socket support routines without releasing its locks.
101 *
102 * Some caution is required in areas where the UNIX domain socket code enters
103 * VFS in order to create or find rendezvous points. This results in
104 * dropping of the UNIX domain socket subsystem lock, acquisition of the
105 * Giant lock, and potential sleeping. This increases the chances of races,
106 * and exposes weaknesses in the socket->protocol API by offering poor
107 * failure modes.
108 */
109static struct mtx unp_mtx;
110#define UNP_LOCK_INIT() \
111 mtx_init(&unp_mtx, "unp", NULL, MTX_DEF)
112#define UNP_LOCK() mtx_lock(&unp_mtx)
113#define UNP_UNLOCK() mtx_unlock(&unp_mtx)
114#define UNP_LOCK_ASSERT() mtx_assert(&unp_mtx, MA_OWNED)
115#define UNP_UNLOCK_ASSERT() mtx_assert(&unp_mtx, MA_NOTOWNED)
116
117/*
118 * Garbage collection of cyclic file descriptor/socket references occurs
119 * asynchronously in a taskqueue context in order to avoid recursion and
120 * reentrance in the UNIX domain socket, file descriptor, and socket layer
121 * code. See unp_gc() for a full description.
122 */
123static struct task unp_gc_task;
124
125static int unp_attach(struct socket *);
126static void unp_detach(struct unpcb *);
127static int unp_bind(struct unpcb *,struct sockaddr *, struct thread *);
128static int unp_connect(struct socket *,struct sockaddr *, struct thread *);
129static int unp_connect2(struct socket *so, struct socket *so2, int);
130static void unp_disconnect(struct unpcb *);
131static void unp_shutdown(struct unpcb *);
132static void unp_drop(struct unpcb *, int);
133static void unp_gc(__unused void *, int);
134static void unp_scan(struct mbuf *, void (*)(struct file *));
135static void unp_mark(struct file *);
136static void unp_discard(struct file *);
137static void unp_freerights(struct file **, int);
138static int unp_internalize(struct mbuf **, struct thread *);
139static int unp_listen(struct socket *, struct unpcb *, int,
140 struct thread *);
141
142static void
143uipc_abort(struct socket *so)
144{
145 struct unpcb *unp;
146
147 unp = sotounpcb(so);
148 KASSERT(unp != NULL, ("uipc_abort: unp == NULL"));
149 UNP_LOCK();
150 unp_drop(unp, ECONNABORTED);
151 unp_detach(unp);
152 UNP_UNLOCK_ASSERT();
153}
154
155static int
156uipc_accept(struct socket *so, struct sockaddr **nam)
157{
158 struct unpcb *unp;
159 const struct sockaddr *sa;
160
161 /*
162 * Pass back name of connected socket,
163 * if it was bound and we are still connected
164 * (our peer may have closed already!).
165 */
166 unp = sotounpcb(so);
167 KASSERT(unp != NULL, ("uipc_accept: unp == NULL"));
168 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
169 UNP_LOCK();
170 if (unp->unp_conn != NULL && unp->unp_conn->unp_addr != NULL)
171 sa = (struct sockaddr *) unp->unp_conn->unp_addr;
172 else
173 sa = &sun_noname;
174 bcopy(sa, *nam, sa->sa_len);
175 UNP_UNLOCK();
176 return (0);
177}
178
179static int
180uipc_attach(struct socket *so, int proto, struct thread *td)
181{
182
183 return (unp_attach(so));
184}
185
186static int
187uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
188{
189 struct unpcb *unp;
190 int error;
191
192 unp = sotounpcb(so);
193 KASSERT(unp != NULL, ("uipc_bind: unp == NULL"));
194 UNP_LOCK();
195 error = unp_bind(unp, nam, td);
196 UNP_UNLOCK();
197 return (error);
198}
199
200static int
201uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
202{
203 struct unpcb *unp;
204 int error;
205
206 KASSERT(td == curthread, ("uipc_connect: td != curthread"));
207 unp = sotounpcb(so);
208 KASSERT(unp != NULL, ("uipc_connect: unp == NULL"));
209 UNP_LOCK();
210 error = unp_connect(so, nam, td);
211 UNP_UNLOCK();
212 return (error);
213}
214
215int
216uipc_connect2(struct socket *so1, struct socket *so2)
217{
218 struct unpcb *unp;
219 int error;
220
221 unp = sotounpcb(so1);
222 KASSERT(unp != NULL, ("uipc_connect2: unp == NULL"));
223 UNP_LOCK();
224 error = unp_connect2(so1, so2, PRU_CONNECT2);
225 UNP_UNLOCK();
226 return (error);
227}
228
229/* control is EOPNOTSUPP */
230
231static void
232uipc_detach(struct socket *so)
233{
234 struct unpcb *unp;
235
236 unp = sotounpcb(so);
237 KASSERT(unp != NULL, ("uipc_detach: unp == NULL"));
238 UNP_LOCK();
239 unp_detach(unp);
240 UNP_UNLOCK_ASSERT();
241}
242
243static int
244uipc_disconnect(struct socket *so)
245{
246 struct unpcb *unp;
247
248 unp = sotounpcb(so);
249 KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL"));
250 UNP_LOCK();
251 unp_disconnect(unp);
252 UNP_UNLOCK();
253 return (0);
254}
255
256static int
257uipc_listen(struct socket *so, int backlog, struct thread *td)
258{
259 struct unpcb *unp;
260 int error;
261
262 unp = sotounpcb(so);
263 KASSERT(unp != NULL, ("uipc_listen: unp == NULL"));
264 UNP_LOCK();
265 if (unp->unp_vnode == NULL) {
266 UNP_UNLOCK();
267 return (EINVAL);
268 }
269 error = unp_listen(so, unp, backlog, td);
270 UNP_UNLOCK();
271 return (error);
272}
273
274static int
275uipc_peeraddr(struct socket *so, struct sockaddr **nam)
276{
277 struct unpcb *unp;
278 const struct sockaddr *sa;
279
280 unp = sotounpcb(so);
281 KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL"));
282 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
283 UNP_LOCK();
284 if (unp->unp_conn != NULL && unp->unp_conn->unp_addr!= NULL)
285 sa = (struct sockaddr *) unp->unp_conn->unp_addr;
286 else {
287 /*
288 * XXX: It seems that this test always fails even when
289 * connection is established. So, this else clause is
290 * added as workaround to return PF_LOCAL sockaddr.
291 */
292 sa = &sun_noname;
293 }
294 bcopy(sa, *nam, sa->sa_len);
295 UNP_UNLOCK();
296 return (0);
297}
298
299static int
300uipc_rcvd(struct socket *so, int flags)
301{
302 struct unpcb *unp;
303 struct socket *so2;
304 u_long newhiwat;
305
306 unp = sotounpcb(so);
307 KASSERT(unp != NULL, ("uipc_rcvd: unp == NULL"));
308 UNP_LOCK();
309 switch (so->so_type) {
310 case SOCK_DGRAM:
311 panic("uipc_rcvd DGRAM?");
312 /*NOTREACHED*/
313
314 case SOCK_STREAM:
315 if (unp->unp_conn == NULL)
316 break;
317 so2 = unp->unp_conn->unp_socket;
318 SOCKBUF_LOCK(&so2->so_snd);
319 SOCKBUF_LOCK(&so->so_rcv);
320 /*
321 * Adjust backpressure on sender
322 * and wakeup any waiting to write.
323 */
324 so2->so_snd.sb_mbmax += unp->unp_mbcnt - so->so_rcv.sb_mbcnt;
325 unp->unp_mbcnt = so->so_rcv.sb_mbcnt;
326 newhiwat = so2->so_snd.sb_hiwat + unp->unp_cc -
327 so->so_rcv.sb_cc;
328 (void)chgsbsize(so2->so_cred->cr_uidinfo, &so2->so_snd.sb_hiwat,
329 newhiwat, RLIM_INFINITY);
330 unp->unp_cc = so->so_rcv.sb_cc;
331 SOCKBUF_UNLOCK(&so->so_rcv);
332 sowwakeup_locked(so2);
333 break;
334
335 default:
336 panic("uipc_rcvd unknown socktype");
337 }
338 UNP_UNLOCK();
339 return (0);
340}
341
342/* pru_rcvoob is EOPNOTSUPP */
343
344static int
345uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
346 struct mbuf *control, struct thread *td)
347{
348 int error = 0;
349 struct unpcb *unp;
350 struct socket *so2;
351 u_long newhiwat;
352
353 unp = sotounpcb(so);
354 KASSERT(unp != NULL, ("uipc_send: unp == NULL"));
355 if (flags & PRUS_OOB) {
356 error = EOPNOTSUPP;
357 goto release;
358 }
359
360 if (control != NULL && (error = unp_internalize(&control, td)))
361 goto release;
362
363 UNP_LOCK();
364 switch (so->so_type) {
365 case SOCK_DGRAM:
366 {
367 const struct sockaddr *from;
368
369 if (nam != NULL) {
370 if (unp->unp_conn != NULL) {
371 error = EISCONN;
372 break;
373 }
374 error = unp_connect(so, nam, td);
375 if (error)
376 break;
377 } else {
378 if (unp->unp_conn == NULL) {
379 error = ENOTCONN;
380 break;
381 }
382 }
383 so2 = unp->unp_conn->unp_socket;
384 if (unp->unp_addr != NULL)
385 from = (struct sockaddr *)unp->unp_addr;
386 else
387 from = &sun_noname;
388 if (unp->unp_conn->unp_flags & UNP_WANTCRED)
389 control = unp_addsockcred(td, control);
390 SOCKBUF_LOCK(&so2->so_rcv);
391 if (sbappendaddr_locked(&so2->so_rcv, from, m, control)) {
392 sorwakeup_locked(so2);
393 m = NULL;
394 control = NULL;
395 } else {
396 SOCKBUF_UNLOCK(&so2->so_rcv);
397 error = ENOBUFS;
398 }
399 if (nam != NULL)
400 unp_disconnect(unp);
401 break;
402 }
403
404 case SOCK_STREAM:
405 /* Connect if not connected yet. */
406 /*
407 * Note: A better implementation would complain
408 * if not equal to the peer's address.
409 */
410 if ((so->so_state & SS_ISCONNECTED) == 0) {
411 if (nam != NULL) {
412 error = unp_connect(so, nam, td);
413 if (error)
414 break; /* XXX */
415 } else {
416 error = ENOTCONN;
417 break;
418 }
419 }
420
421 SOCKBUF_LOCK(&so->so_snd);
422 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
423 SOCKBUF_UNLOCK(&so->so_snd);
424 error = EPIPE;
425 break;
426 }
427 if (unp->unp_conn == NULL)
428 panic("uipc_send connected but no connection?");
429 so2 = unp->unp_conn->unp_socket;
430 SOCKBUF_LOCK(&so2->so_rcv);
431 if (unp->unp_conn->unp_flags & UNP_WANTCRED) {
432 /*
433 * Credentials are passed only once on
434 * SOCK_STREAM.
435 */
436 unp->unp_conn->unp_flags &= ~UNP_WANTCRED;
437 control = unp_addsockcred(td, control);
438 }
439 /*
440 * Send to paired receive port, and then reduce
441 * send buffer hiwater marks to maintain backpressure.
442 * Wake up readers.
443 */
444 if (control != NULL) {
445 if (sbappendcontrol_locked(&so2->so_rcv, m, control))
446 control = NULL;
447 } else {
448 sbappend_locked(&so2->so_rcv, m);
449 }
450 so->so_snd.sb_mbmax -=
451 so2->so_rcv.sb_mbcnt - unp->unp_conn->unp_mbcnt;
452 unp->unp_conn->unp_mbcnt = so2->so_rcv.sb_mbcnt;
453 newhiwat = so->so_snd.sb_hiwat -
454 (so2->so_rcv.sb_cc - unp->unp_conn->unp_cc);
455 (void)chgsbsize(so->so_cred->cr_uidinfo, &so->so_snd.sb_hiwat,
456 newhiwat, RLIM_INFINITY);
457 SOCKBUF_UNLOCK(&so->so_snd);
458 unp->unp_conn->unp_cc = so2->so_rcv.sb_cc;
459 sorwakeup_locked(so2);
460 m = NULL;
461 break;
462
463 default:
464 panic("uipc_send unknown socktype");
465 }
466
467 /*
468 * SEND_EOF is equivalent to a SEND followed by
469 * a SHUTDOWN.
470 */
471 if (flags & PRUS_EOF) {
472 socantsendmore(so);
473 unp_shutdown(unp);
474 }
475 UNP_UNLOCK();
476
477 if (control != NULL && error != 0)
478 unp_dispose(control);
479
480release:
481 if (control != NULL)
482 m_freem(control);
483 if (m != NULL)
484 m_freem(m);
485 return (error);
486}
487
488static int
489uipc_sense(struct socket *so, struct stat *sb)
490{
491 struct unpcb *unp;
492 struct socket *so2;
493
494 unp = sotounpcb(so);
495 KASSERT(unp != NULL, ("uipc_sense: unp == NULL"));
496 UNP_LOCK();
497 sb->st_blksize = so->so_snd.sb_hiwat;
498 if (so->so_type == SOCK_STREAM && unp->unp_conn != NULL) {
499 so2 = unp->unp_conn->unp_socket;
500 sb->st_blksize += so2->so_rcv.sb_cc;
501 }
502 sb->st_dev = NODEV;
503 if (unp->unp_ino == 0)
504 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino;
505 sb->st_ino = unp->unp_ino;
506 UNP_UNLOCK();
507 return (0);
508}
509
510static int
511uipc_shutdown(struct socket *so)
512{
513 struct unpcb *unp;
514
515 unp = sotounpcb(so);
516 KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL"));
517 UNP_LOCK();
518 socantsendmore(so);
519 unp_shutdown(unp);
520 UNP_UNLOCK();
521 return (0);
522}
523
524static int
525uipc_sockaddr(struct socket *so, struct sockaddr **nam)
526{
527 struct unpcb *unp;
528 const struct sockaddr *sa;
529
530 unp = sotounpcb(so);
531 KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL"));
532 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
533 UNP_LOCK();
534 if (unp->unp_addr != NULL)
535 sa = (struct sockaddr *) unp->unp_addr;
536 else
537 sa = &sun_noname;
538 bcopy(sa, *nam, sa->sa_len);
539 UNP_UNLOCK();
540 return (0);
541}
542
543struct pr_usrreqs uipc_usrreqs = {
544 .pru_abort = uipc_abort,
545 .pru_accept = uipc_accept,
546 .pru_attach = uipc_attach,
547 .pru_bind = uipc_bind,
548 .pru_connect = uipc_connect,
549 .pru_connect2 = uipc_connect2,
550 .pru_detach = uipc_detach,
551 .pru_disconnect = uipc_disconnect,
552 .pru_listen = uipc_listen,
553 .pru_peeraddr = uipc_peeraddr,
554 .pru_rcvd = uipc_rcvd,
555 .pru_send = uipc_send,
556 .pru_sense = uipc_sense,
557 .pru_shutdown = uipc_shutdown,
558 .pru_sockaddr = uipc_sockaddr,
559 .pru_sosend = sosend,
560 .pru_soreceive = soreceive,
561 .pru_sopoll = sopoll,
562};
563
564int
565uipc_ctloutput(struct socket *so, struct sockopt *sopt)
566{
567 struct unpcb *unp;
568 struct xucred xu;
569 int error, optval;
570
571 if (sopt->sopt_level != 0)
572 return (EINVAL);
573
574 unp = sotounpcb(so);
575 KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL"));
576 UNP_LOCK();
577 error = 0;
578 switch (sopt->sopt_dir) {
579 case SOPT_GET:
580 switch (sopt->sopt_name) {
581 case LOCAL_PEERCRED:
582 if (unp->unp_flags & UNP_HAVEPC)
583 xu = unp->unp_peercred;
584 else {
585 if (so->so_type == SOCK_STREAM)
586 error = ENOTCONN;
587 else
588 error = EINVAL;
589 }
590 if (error == 0)
591 error = sooptcopyout(sopt, &xu, sizeof(xu));
592 break;
593 case LOCAL_CREDS:
594 optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0;
595 error = sooptcopyout(sopt, &optval, sizeof(optval));
596 break;
597 case LOCAL_CONNWAIT:
598 optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0;
599 error = sooptcopyout(sopt, &optval, sizeof(optval));
600 break;
601 default:
602 error = EOPNOTSUPP;
603 break;
604 }
605 break;
606 case SOPT_SET:
607 switch (sopt->sopt_name) {
608 case LOCAL_CREDS:
609 case LOCAL_CONNWAIT:
610 error = sooptcopyin(sopt, &optval, sizeof(optval),
611 sizeof(optval));
612 if (error)
613 break;
614
615#define OPTSET(bit) \
616 if (optval) \
617 unp->unp_flags |= bit; \
618 else \
619 unp->unp_flags &= ~bit;
620
621 switch (sopt->sopt_name) {
622 case LOCAL_CREDS:
623 OPTSET(UNP_WANTCRED);
624 break;
625 case LOCAL_CONNWAIT:
626 OPTSET(UNP_CONNWAIT);
627 break;
628 default:
629 break;
630 }
631 break;
632#undef OPTSET
633 default:
634 error = ENOPROTOOPT;
635 break;
636 }
637 break;
638 default:
639 error = EOPNOTSUPP;
640 break;
641 }
642 UNP_UNLOCK();
643 return (error);
644}
645
646/*
647 * Both send and receive buffers are allocated PIPSIZ bytes of buffering
648 * for stream sockets, although the total for sender and receiver is
649 * actually only PIPSIZ.
650 * Datagram sockets really use the sendspace as the maximum datagram size,
651 * and don't really want to reserve the sendspace. Their recvspace should
652 * be large enough for at least one max-size datagram plus address.
653 */
654#ifndef PIPSIZ
655#define PIPSIZ 8192
656#endif
657static u_long unpst_sendspace = PIPSIZ;
658static u_long unpst_recvspace = PIPSIZ;
659static u_long unpdg_sendspace = 2*1024; /* really max datagram size */
660static u_long unpdg_recvspace = 4*1024;
661
662static int unp_rights; /* file descriptors in flight */
663
664SYSCTL_DECL(_net_local_stream);
665SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW,
666 &unpst_sendspace, 0, "");
667SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW,
668 &unpst_recvspace, 0, "");
669SYSCTL_DECL(_net_local_dgram);
670SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW,
671 &unpdg_sendspace, 0, "");
672SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW,
673 &unpdg_recvspace, 0, "");
674SYSCTL_DECL(_net_local);
675SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, "");
676
677static int
678unp_attach(struct socket *so)
679{
680 struct unpcb *unp;
681 int error;
682
683 KASSERT(so->so_pcb == NULL, ("unp_attach: so_pcb != NULL"));
684 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
685 switch (so->so_type) {
686
687 case SOCK_STREAM:
688 error = soreserve(so, unpst_sendspace, unpst_recvspace);
689 break;
690
691 case SOCK_DGRAM:
692 error = soreserve(so, unpdg_sendspace, unpdg_recvspace);
693 break;
694
695 default:
696 panic("unp_attach");
697 }
698 if (error)
699 return (error);
700 }
701 unp = uma_zalloc(unp_zone, M_WAITOK | M_ZERO);
702 if (unp == NULL)
703 return (ENOBUFS);
704 LIST_INIT(&unp->unp_refs);
705 unp->unp_socket = so;
706 so->so_pcb = unp;
707
708 UNP_LOCK();
709 unp->unp_gencnt = ++unp_gencnt;
710 unp_count++;
711 LIST_INSERT_HEAD(so->so_type == SOCK_DGRAM ? &unp_dhead
712 : &unp_shead, unp, unp_link);
713 UNP_UNLOCK();
714
715 return (0);
716}
717
718static void
719unp_detach(struct unpcb *unp)
720{
721 struct vnode *vp;
722 int local_unp_rights;
723
724 UNP_LOCK_ASSERT();
725
726 LIST_REMOVE(unp, unp_link);
727 unp->unp_gencnt = ++unp_gencnt;
728 --unp_count;
729 if ((vp = unp->unp_vnode) != NULL) {
730 /*
731 * XXXRW: should v_socket be frobbed only while holding
732 * Giant?
733 */
734 unp->unp_vnode->v_socket = NULL;
735 unp->unp_vnode = NULL;
736 }
737 if (unp->unp_conn != NULL)
738 unp_disconnect(unp);
739 while (!LIST_EMPTY(&unp->unp_refs)) {
740 struct unpcb *ref = LIST_FIRST(&unp->unp_refs);
741 unp_drop(ref, ECONNRESET);
742 }
743 soisdisconnected(unp->unp_socket);
744 unp->unp_socket->so_pcb = NULL;
745 local_unp_rights = unp_rights;
746 UNP_UNLOCK();
747 if (unp->unp_addr != NULL)
748 FREE(unp->unp_addr, M_SONAME);
749 uma_zfree(unp_zone, unp);
750 if (vp) {
751 int vfslocked;
752
753 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
754 vrele(vp);
755 VFS_UNLOCK_GIANT(vfslocked);
756 }
757 if (local_unp_rights)
758 taskqueue_enqueue(taskqueue_thread, &unp_gc_task);
759}
760
761static int
762unp_bind(struct unpcb *unp, struct sockaddr *nam, struct thread *td)
763{
764 struct sockaddr_un *soun = (struct sockaddr_un *)nam;
765 struct vnode *vp;
766 struct mount *mp;
767 struct vattr vattr;
768 int error, namelen;
769 struct nameidata nd;
770 char *buf;
771
772 UNP_LOCK_ASSERT();
773
774 /*
775 * XXXRW: This test-and-set of unp_vnode is non-atomic; the
776 * unlocked read here is fine, but the value of unp_vnode needs
777 * to be tested again after we do all the lookups to see if the
778 * pcb is still unbound?
779 */
780 if (unp->unp_vnode != NULL)
781 return (EINVAL);
782
783 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path);
784 if (namelen <= 0)
785 return (EINVAL);
786
787 UNP_UNLOCK();
788
789 buf = malloc(namelen + 1, M_TEMP, M_WAITOK);
790 strlcpy(buf, soun->sun_path, namelen + 1);
791
792 mtx_lock(&Giant);
793restart:
794 mtx_assert(&Giant, MA_OWNED);
795 NDINIT(&nd, CREATE, NOFOLLOW | LOCKPARENT | SAVENAME, UIO_SYSSPACE,
796 buf, td);
797/* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
798 error = namei(&nd);
799 if (error)
800 goto done;
801 vp = nd.ni_vp;
802 if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
803 NDFREE(&nd, NDF_ONLY_PNBUF);
804 if (nd.ni_dvp == vp)
805 vrele(nd.ni_dvp);
806 else
807 vput(nd.ni_dvp);
808 if (vp != NULL) {
809 vrele(vp);
810 error = EADDRINUSE;
811 goto done;
812 }
813 error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH);
814 if (error)
815 goto done;
816 goto restart;
817 }
818 VATTR_NULL(&vattr);
819 vattr.va_type = VSOCK;
820 vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask);
821#ifdef MAC
822 error = mac_check_vnode_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd,
823 &vattr);
824#endif
825 if (error == 0) {
826 VOP_LEASE(nd.ni_dvp, td, td->td_ucred, LEASE_WRITE);
827 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
828 }
829 NDFREE(&nd, NDF_ONLY_PNBUF);
830 vput(nd.ni_dvp);
831 if (error) {
832 vn_finished_write(mp);
833 goto done;
834 }
835 vp = nd.ni_vp;
836 ASSERT_VOP_LOCKED(vp, "unp_bind");
837 soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK);
838 UNP_LOCK();
839 vp->v_socket = unp->unp_socket;
840 unp->unp_vnode = vp;
841 unp->unp_addr = soun;
842 UNP_UNLOCK();
843 VOP_UNLOCK(vp, 0, td);
844 vn_finished_write(mp);
845done:
846 mtx_unlock(&Giant);
847 free(buf, M_TEMP);
848 UNP_LOCK();
849 return (error);
850}
851
852static int
853unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
854{
855 struct sockaddr_un *soun = (struct sockaddr_un *)nam;
856 struct vnode *vp;
857 struct socket *so2, *so3;
858 struct unpcb *unp, *unp2, *unp3;
859 int error, len;
860 struct nameidata nd;
861 char buf[SOCK_MAXADDRLEN];
862 struct sockaddr *sa;
863
864 UNP_LOCK_ASSERT();
865
866 unp = sotounpcb(so);
867 KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
868 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
869 if (len <= 0)
870 return (EINVAL);
871 strlcpy(buf, soun->sun_path, len + 1);
872 UNP_UNLOCK();
873 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
874 mtx_lock(&Giant);
875 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, buf, td);
876 error = namei(&nd);
877 if (error)
878 vp = NULL;
879 else
880 vp = nd.ni_vp;
881 ASSERT_VOP_LOCKED(vp, "unp_connect");
882 NDFREE(&nd, NDF_ONLY_PNBUF);
883 if (error)
884 goto bad;
885
886 if (vp->v_type != VSOCK) {
887 error = ENOTSOCK;
888 goto bad;
889 }
890 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td);
891 if (error)
892 goto bad;
893 mtx_unlock(&Giant);
894 UNP_LOCK();
895 unp = sotounpcb(so);
896 KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
897 so2 = vp->v_socket;
898 if (so2 == NULL) {
899 error = ECONNREFUSED;
900 goto bad2;
901 }
902 if (so->so_type != so2->so_type) {
903 error = EPROTOTYPE;
904 goto bad2;
905 }
906 if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
907 if (so2->so_options & SO_ACCEPTCONN) {
908 /*
909 * NB: drop locks here so unp_attach is entered
910 * w/o locks; this avoids a recursive lock
911 * of the head and holding sleep locks across
912 * a (potentially) blocking malloc.
913 */
914 UNP_UNLOCK();
915 so3 = sonewconn(so2, 0);
916 UNP_LOCK();
917 } else
918 so3 = NULL;
919 if (so3 == NULL) {
920 error = ECONNREFUSED;
921 goto bad2;
922 }
923 unp = sotounpcb(so);
924 unp2 = sotounpcb(so2);
925 unp3 = sotounpcb(so3);
926 if (unp2->unp_addr != NULL) {
927 bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len);
928 unp3->unp_addr = (struct sockaddr_un *) sa;
929 sa = NULL;
930 }
931 /*
932 * unp_peercred management:
933 *
934 * The connecter's (client's) credentials are copied
935 * from its process structure at the time of connect()
936 * (which is now).
937 */
938 cru2x(td->td_ucred, &unp3->unp_peercred);
939 unp3->unp_flags |= UNP_HAVEPC;
940 /*
941 * The receiver's (server's) credentials are copied
942 * from the unp_peercred member of socket on which the
943 * former called listen(); unp_listen() cached that
944 * process's credentials at that time so we can use
945 * them now.
946 */
947 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED,
948 ("unp_connect: listener without cached peercred"));
949 memcpy(&unp->unp_peercred, &unp2->unp_peercred,
950 sizeof(unp->unp_peercred));
951 unp->unp_flags |= UNP_HAVEPC;
952#ifdef MAC
953 SOCK_LOCK(so);
954 mac_set_socket_peer_from_socket(so, so3);
955 mac_set_socket_peer_from_socket(so3, so);
956 SOCK_UNLOCK(so);
957#endif
958
959 so2 = so3;
960 }
961 error = unp_connect2(so, so2, PRU_CONNECT);
962bad2:
963 UNP_UNLOCK();
964 mtx_lock(&Giant);
965bad:
966 mtx_assert(&Giant, MA_OWNED);
967 if (vp != NULL)
968 vput(vp);
969 mtx_unlock(&Giant);
970 free(sa, M_SONAME);
971 UNP_LOCK();
972 return (error);
973}
974
975static int
976unp_connect2(struct socket *so, struct socket *so2, int req)
977{
978 struct unpcb *unp = sotounpcb(so);
979 struct unpcb *unp2;
980
981 UNP_LOCK_ASSERT();
982
983 if (so2->so_type != so->so_type)
984 return (EPROTOTYPE);
985 unp2 = sotounpcb(so2);
986 KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL"));
987 unp->unp_conn = unp2;
988 switch (so->so_type) {
989
990 case SOCK_DGRAM:
991 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
992 soisconnected(so);
993 break;
994
995 case SOCK_STREAM:
996 unp2->unp_conn = unp;
997 if (req == PRU_CONNECT &&
998 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT))
999 soisconnecting(so);
1000 else
1001 soisconnected(so);
1002 soisconnected(so2);
1003 break;
1004
1005 default:
1006 panic("unp_connect2");
1007 }
1008 return (0);
1009}
1010
1011static void
1012unp_disconnect(struct unpcb *unp)
1013{
1014 struct unpcb *unp2 = unp->unp_conn;
1015 struct socket *so;
1016
1017 UNP_LOCK_ASSERT();
1018
1019 if (unp2 == NULL)
1020 return;
1021 unp->unp_conn = NULL;
1022 switch (unp->unp_socket->so_type) {
1023 case SOCK_DGRAM:
1024 LIST_REMOVE(unp, unp_reflink);
1025 so = unp->unp_socket;
1026 SOCK_LOCK(so);
1027 so->so_state &= ~SS_ISCONNECTED;
1028 SOCK_UNLOCK(so);
1029 break;
1030
1031 case SOCK_STREAM:
1032 soisdisconnected(unp->unp_socket);
1033 unp2->unp_conn = NULL;
1034 soisdisconnected(unp2->unp_socket);
1035 break;
1036 }
1037}
1038
1039#ifdef notdef
1040void
1041unp_abort(struct unpcb *unp)
1042{
1043
1044 unp_detach(unp);
1045 UNP_UNLOCK_ASSERT();
1046}
1047#endif
1048
1049/*
1050 * unp_pcblist() assumes that UNIX domain socket memory is never reclaimed
1051 * by the zone (UMA_ZONE_NOFREE), and as such potentially stale pointers
1052 * are safe to reference. It first scans the list of struct unpcb's to
1053 * generate a pointer list, then it rescans its list one entry at a time to
1054 * externalize and copyout. It checks the generation number to see if a
1055 * struct unpcb has been reused, and will skip it if so.
1056 */
1057static int
1058unp_pcblist(SYSCTL_HANDLER_ARGS)
1059{
1060 int error, i, n;
1061 struct unpcb *unp, **unp_list;
1062 unp_gen_t gencnt;
1063 struct xunpgen *xug;
1064 struct unp_head *head;
1065 struct xunpcb *xu;
1066
1067 head = ((intptr_t)arg1 == SOCK_DGRAM ? &unp_dhead : &unp_shead);
1068
1069 /*
1070 * The process of preparing the PCB list is too time-consuming and
1071 * resource-intensive to repeat twice on every request.
1072 */
1073 if (req->oldptr == NULL) {
1074 n = unp_count;
1075 req->oldidx = 2 * (sizeof *xug)
1076 + (n + n/8) * sizeof(struct xunpcb);
1077 return (0);
1078 }
1079
1080 if (req->newptr != NULL)
1081 return (EPERM);
1082
1083 /*
1084 * OK, now we're committed to doing something.
1085 */
1086 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK);
1087 UNP_LOCK();
1088 gencnt = unp_gencnt;
1089 n = unp_count;
1090 UNP_UNLOCK();
1091
1092 xug->xug_len = sizeof *xug;
1093 xug->xug_count = n;
1094 xug->xug_gen = gencnt;
1095 xug->xug_sogen = so_gencnt;
1096 error = SYSCTL_OUT(req, xug, sizeof *xug);
1097 if (error) {
1098 free(xug, M_TEMP);
1099 return (error);
1100 }
1101
1102 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
1103
1104 UNP_LOCK();
1105 for (unp = LIST_FIRST(head), i = 0; unp && i < n;
1106 unp = LIST_NEXT(unp, unp_link)) {
1107 if (unp->unp_gencnt <= gencnt) {
1108 if (cr_cansee(req->td->td_ucred,
1109 unp->unp_socket->so_cred))
1110 continue;
1111 unp_list[i++] = unp;
1112 }
1113 }
1114 UNP_UNLOCK();
1115 n = i; /* in case we lost some during malloc */
1116
1117 error = 0;
1118 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO);
1119 for (i = 0; i < n; i++) {
1120 unp = unp_list[i];
1121 if (unp->unp_gencnt <= gencnt) {
1122 xu->xu_len = sizeof *xu;
1123 xu->xu_unpp = unp;
1124 /*
1125 * XXX - need more locking here to protect against
1126 * connect/disconnect races for SMP.
1127 */
1128 if (unp->unp_addr != NULL)
1129 bcopy(unp->unp_addr, &xu->xu_addr,
1130 unp->unp_addr->sun_len);
1131 if (unp->unp_conn != NULL &&
1132 unp->unp_conn->unp_addr != NULL)
1133 bcopy(unp->unp_conn->unp_addr,
1134 &xu->xu_caddr,
1135 unp->unp_conn->unp_addr->sun_len);
1136 bcopy(unp, &xu->xu_unp, sizeof *unp);
1137 sotoxsocket(unp->unp_socket, &xu->xu_socket);
1138 error = SYSCTL_OUT(req, xu, sizeof *xu);
1139 }
1140 }
1141 free(xu, M_TEMP);
1142 if (!error) {
1143 /*
1144 * Give the user an updated idea of our state.
1145 * If the generation differs from what we told
1146 * her before, she knows that something happened
1147 * while we were processing this request, and it
1148 * might be necessary to retry.
1149 */
1150 xug->xug_gen = unp_gencnt;
1151 xug->xug_sogen = so_gencnt;
1152 xug->xug_count = unp_count;
1153 error = SYSCTL_OUT(req, xug, sizeof *xug);
1154 }
1155 free(unp_list, M_TEMP);
1156 free(xug, M_TEMP);
1157 return (error);
1158}
1159
1160SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLFLAG_RD,
1161 (caddr_t)(long)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
1162 "List of active local datagram sockets");
1163SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLFLAG_RD,
1164 (caddr_t)(long)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
1165 "List of active local stream sockets");
1166
1167static void
1168unp_shutdown(struct unpcb *unp)
1169{
1170 struct socket *so;
1171
1172 UNP_LOCK_ASSERT();
1173
1174 if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn &&
1175 (so = unp->unp_conn->unp_socket))
1176 socantrcvmore(so);
1177}
1178
1179static void
1180unp_drop(struct unpcb *unp, int errno)
1181{
1182 struct socket *so = unp->unp_socket;
1183
1184 UNP_LOCK_ASSERT();
1185
1186 so->so_error = errno;
1187 unp_disconnect(unp);
1188}
1189
1190#ifdef notdef
1191void
1192unp_drain(void)
1193{
1194
1195}
1196#endif
1197
1198static void
1199unp_freerights(struct file **rp, int fdcount)
1200{
1201 int i;
1202 struct file *fp;
1203
1204 for (i = 0; i < fdcount; i++) {
1205 fp = *rp;
1206 /*
1207 * zero the pointer before calling
1208 * unp_discard since it may end up
1209 * in unp_gc()..
1210 *
1211 * XXXRW: This is less true than it used to be.
1212 */
1213 *rp++ = 0;
1214 unp_discard(fp);
1215 }
1216}
1217
1218int
1219unp_externalize(struct mbuf *control, struct mbuf **controlp)
1220{
1221 struct thread *td = curthread; /* XXX */
1222 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1223 int i;
1224 int *fdp;
1225 struct file **rp;
1226 struct file *fp;
1227 void *data;
1228 socklen_t clen = control->m_len, datalen;
1229 int error, newfds;
1230 int f;
1231 u_int newlen;
1232
1233 UNP_UNLOCK_ASSERT();
1234
1235 error = 0;
1236 if (controlp != NULL) /* controlp == NULL => free control messages */
1237 *controlp = NULL;
1238
1239 while (cm != NULL) {
1240 if (sizeof(*cm) > clen || cm->cmsg_len > clen) {
1241 error = EINVAL;
1242 break;
1243 }
1244
1245 data = CMSG_DATA(cm);
1246 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1247
1248 if (cm->cmsg_level == SOL_SOCKET
1249 && cm->cmsg_type == SCM_RIGHTS) {
1250 newfds = datalen / sizeof(struct file *);
1251 rp = data;
1252
1253 /* If we're not outputting the descriptors free them. */
1254 if (error || controlp == NULL) {
1255 unp_freerights(rp, newfds);
1256 goto next;
1257 }
1258 FILEDESC_LOCK(td->td_proc->p_fd);
1259 /* if the new FD's will not fit free them. */
1260 if (!fdavail(td, newfds)) {
1261 FILEDESC_UNLOCK(td->td_proc->p_fd);
1262 error = EMSGSIZE;
1263 unp_freerights(rp, newfds);
1264 goto next;
1265 }
1266 /*
1267 * now change each pointer to an fd in the global
1268 * table to an integer that is the index to the
1269 * local fd table entry that we set up to point
1270 * to the global one we are transferring.
1271 */
1272 newlen = newfds * sizeof(int);
1273 *controlp = sbcreatecontrol(NULL, newlen,
1274 SCM_RIGHTS, SOL_SOCKET);
1275 if (*controlp == NULL) {
1276 FILEDESC_UNLOCK(td->td_proc->p_fd);
1277 error = E2BIG;
1278 unp_freerights(rp, newfds);
1279 goto next;
1280 }
1281
1282 fdp = (int *)
1283 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1284 for (i = 0; i < newfds; i++) {
1285 if (fdalloc(td, 0, &f))
1286 panic("unp_externalize fdalloc failed");
1287 fp = *rp++;
1288 td->td_proc->p_fd->fd_ofiles[f] = fp;
1289 FILE_LOCK(fp);
1290 fp->f_msgcount--;
1291 FILE_UNLOCK(fp);
1292 unp_rights--;
1293 *fdp++ = f;
1294 }
1295 FILEDESC_UNLOCK(td->td_proc->p_fd);
1296 } else { /* We can just copy anything else across */
1297 if (error || controlp == NULL)
1298 goto next;
1299 *controlp = sbcreatecontrol(NULL, datalen,
1300 cm->cmsg_type, cm->cmsg_level);
1301 if (*controlp == NULL) {
1302 error = ENOBUFS;
1303 goto next;
1304 }
1305 bcopy(data,
1306 CMSG_DATA(mtod(*controlp, struct cmsghdr *)),
1307 datalen);
1308 }
1309
1310 controlp = &(*controlp)->m_next;
1311
1312next:
1313 if (CMSG_SPACE(datalen) < clen) {
1314 clen -= CMSG_SPACE(datalen);
1315 cm = (struct cmsghdr *)
1316 ((caddr_t)cm + CMSG_SPACE(datalen));
1317 } else {
1318 clen = 0;
1319 cm = NULL;
1320 }
1321 }
1322
1323 m_freem(control);
1324
1325 return (error);
1326}
1327
1328void
1329unp_init(void)
1330{
1331 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL,
1332 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
1333 if (unp_zone == NULL)
1334 panic("unp_init");
1335 uma_zone_set_max(unp_zone, nmbclusters);
1336 LIST_INIT(&unp_dhead);
1337 LIST_INIT(&unp_shead);
1338 TASK_INIT(&unp_gc_task, 0, unp_gc, NULL);
1339 UNP_LOCK_INIT();
1340}
1341
1342static int
1343unp_internalize(struct mbuf **controlp, struct thread *td)
1344{
1345 struct mbuf *control = *controlp;
1346 struct proc *p = td->td_proc;
1347 struct filedesc *fdescp = p->p_fd;
1348 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1349 struct cmsgcred *cmcred;
1350 struct file **rp;
1351 struct file *fp;
1352 struct timeval *tv;
1353 int i, fd, *fdp;
1354 void *data;
1355 socklen_t clen = control->m_len, datalen;
1356 int error, oldfds;
1357 u_int newlen;
1358
1359 UNP_UNLOCK_ASSERT();
1360
1361 error = 0;
1362 *controlp = NULL;
1363
1364 while (cm != NULL) {
1365 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET
1366 || cm->cmsg_len > clen) {
1367 error = EINVAL;
1368 goto out;
1369 }
1370
1371 data = CMSG_DATA(cm);
1372 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1373
1374 switch (cm->cmsg_type) {
1375 /*
1376 * Fill in credential information.
1377 */
1378 case SCM_CREDS:
1379 *controlp = sbcreatecontrol(NULL, sizeof(*cmcred),
1380 SCM_CREDS, SOL_SOCKET);
1381 if (*controlp == NULL) {
1382 error = ENOBUFS;
1383 goto out;
1384 }
1385
1386 cmcred = (struct cmsgcred *)
1387 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1388 cmcred->cmcred_pid = p->p_pid;
1389 cmcred->cmcred_uid = td->td_ucred->cr_ruid;
1390 cmcred->cmcred_gid = td->td_ucred->cr_rgid;
1391 cmcred->cmcred_euid = td->td_ucred->cr_uid;
1392 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups,
1393 CMGROUP_MAX);
1394 for (i = 0; i < cmcred->cmcred_ngroups; i++)
1395 cmcred->cmcred_groups[i] =
1396 td->td_ucred->cr_groups[i];
1397 break;
1398
1399 case SCM_RIGHTS:
1400 oldfds = datalen / sizeof (int);
1401 /*
1402 * check that all the FDs passed in refer to legal files
1403 * If not, reject the entire operation.
1404 */
1405 fdp = data;
1406 FILEDESC_LOCK(fdescp);
1407 for (i = 0; i < oldfds; i++) {
1408 fd = *fdp++;
1409 if ((unsigned)fd >= fdescp->fd_nfiles ||
1410 fdescp->fd_ofiles[fd] == NULL) {
1411 FILEDESC_UNLOCK(fdescp);
1412 error = EBADF;
1413 goto out;
1414 }
1415 fp = fdescp->fd_ofiles[fd];
1416 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) {
1417 FILEDESC_UNLOCK(fdescp);
1418 error = EOPNOTSUPP;
1419 goto out;
1420 }
1421
1422 }
1423 /*
1424 * Now replace the integer FDs with pointers to
1425 * the associated global file table entry..
1426 */
1427 newlen = oldfds * sizeof(struct file *);
1428 *controlp = sbcreatecontrol(NULL, newlen,
1429 SCM_RIGHTS, SOL_SOCKET);
1430 if (*controlp == NULL) {
1431 FILEDESC_UNLOCK(fdescp);
1432 error = E2BIG;
1433 goto out;
1434 }
1435
1436 fdp = data;
1437 rp = (struct file **)
1438 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1439 for (i = 0; i < oldfds; i++) {
1440 fp = fdescp->fd_ofiles[*fdp++];
1441 *rp++ = fp;
1442 FILE_LOCK(fp);
1443 fp->f_count++;
1444 fp->f_msgcount++;
1445 FILE_UNLOCK(fp);
1446 unp_rights++;
1447 }
1448 FILEDESC_UNLOCK(fdescp);
1449 break;
1450
1451 case SCM_TIMESTAMP:
1452 *controlp = sbcreatecontrol(NULL, sizeof(*tv),
1453 SCM_TIMESTAMP, SOL_SOCKET);
1454 if (*controlp == NULL) {
1455 error = ENOBUFS;
1456 goto out;
1457 }
1458 tv = (struct timeval *)
1459 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1460 microtime(tv);
1461 break;
1462
1463 default:
1464 error = EINVAL;
1465 goto out;
1466 }
1467
1468 controlp = &(*controlp)->m_next;
1469
1470 if (CMSG_SPACE(datalen) < clen) {
1471 clen -= CMSG_SPACE(datalen);
1472 cm = (struct cmsghdr *)
1473 ((caddr_t)cm + CMSG_SPACE(datalen));
1474 } else {
1475 clen = 0;
1476 cm = NULL;
1477 }
1478 }
1479
1480out:
1481 m_freem(control);
1482
1483 return (error);
1484}
1485
1486struct mbuf *
1487unp_addsockcred(struct thread *td, struct mbuf *control)
1488{
1489 struct mbuf *m, *n;
1490 struct sockcred *sc;
1491 int ngroups;
1492 int i;
1493
1494 ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX);
1495
1496 m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET);
1497 if (m == NULL)
1498 return (control);
1499 m->m_next = NULL;
1500
1501 sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *));
1502 sc->sc_uid = td->td_ucred->cr_ruid;
1503 sc->sc_euid = td->td_ucred->cr_uid;
1504 sc->sc_gid = td->td_ucred->cr_rgid;
1505 sc->sc_egid = td->td_ucred->cr_gid;
1506 sc->sc_ngroups = ngroups;
1507 for (i = 0; i < sc->sc_ngroups; i++)
1508 sc->sc_groups[i] = td->td_ucred->cr_groups[i];
1509
1510 /*
1511 * If a control message already exists, append us to the end.
1512 */
1513 if (control != NULL) {
1514 for (n = control; n->m_next != NULL; n = n->m_next)
1515 ;
1516 n->m_next = m;
1517 } else
1518 control = m;
1519
1520 return (control);
1521}
1522
1523/*
1524 * unp_defer indicates whether additional work has been defered for a future
1525 * pass through unp_gc(). It is thread local and does not require explicit
1526 * synchronization.
1527 */
1528static int unp_defer;
1529
1530static int unp_taskcount;
1531SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0, "");
1532
1533static int unp_recycled;
1534SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0, "");
1535
1536static void
1537unp_gc(__unused void *arg, int pending)
1538{
1539 struct file *fp, *nextfp;
1540 struct socket *so;
1541 struct file **extra_ref, **fpp;
1542 int nunref, i;
1543 int nfiles_snap;
1544 int nfiles_slack = 20;
1545
1546 unp_taskcount++;
1547 unp_defer = 0;
1548 /*
1549 * before going through all this, set all FDs to
1550 * be NOT defered and NOT externally accessible
1551 */
1552 sx_slock(&filelist_lock);
1553 LIST_FOREACH(fp, &filehead, f_list)
1554 fp->f_gcflag &= ~(FMARK|FDEFER);
1555 do {
1556 KASSERT(unp_defer >= 0, ("unp_gc: unp_defer %d", unp_defer));
1557 LIST_FOREACH(fp, &filehead, f_list) {
1558 FILE_LOCK(fp);
1559 /*
1560 * If the file is not open, skip it -- could be a
1561 * file in the process of being opened, or in the
1562 * process of being closed. If the file is
1563 * "closing", it may have been marked for deferred
1564 * consideration. Clear the flag now if so.
1565 */
1566 if (fp->f_count == 0) {
1567 if (fp->f_gcflag & FDEFER)
1568 unp_defer--;
1569 fp->f_gcflag &= ~(FMARK|FDEFER);
1570 FILE_UNLOCK(fp);
1571 continue;
1572 }
1573 /*
1574 * If we already marked it as 'defer' in a
1575 * previous pass, then try process it this time
1576 * and un-mark it
1577 */
1578 if (fp->f_gcflag & FDEFER) {
1579 fp->f_gcflag &= ~FDEFER;
1580 unp_defer--;
1581 } else {
1582 /*
1583 * if it's not defered, then check if it's
1584 * already marked.. if so skip it
1585 */
1586 if (fp->f_gcflag & FMARK) {
1587 FILE_UNLOCK(fp);
1588 continue;
1589 }
1590 /*
1591 * If all references are from messages
1592 * in transit, then skip it. it's not
1593 * externally accessible.
1594 */
1595 if (fp->f_count == fp->f_msgcount) {
1596 FILE_UNLOCK(fp);
1597 continue;
1598 }
1599 /*
1600 * If it got this far then it must be
1601 * externally accessible.
1602 */
1603 fp->f_gcflag |= FMARK;
1604 }
1605 /*
1606 * either it was defered, or it is externally
1607 * accessible and not already marked so.
1608 * Now check if it is possibly one of OUR sockets.
1609 */
1610 if (fp->f_type != DTYPE_SOCKET ||
1611 (so = fp->f_data) == NULL) {
1612 FILE_UNLOCK(fp);
1613 continue;
1614 }
1615 FILE_UNLOCK(fp);
1616 if (so->so_proto->pr_domain != &localdomain ||
1617 (so->so_proto->pr_flags&PR_RIGHTS) == 0)
1618 continue;
1619 /*
1620 * So, Ok, it's one of our sockets and it IS externally
1621 * accessible (or was defered). Now we look
1622 * to see if we hold any file descriptors in its
1623 * message buffers. Follow those links and mark them
1624 * as accessible too.
1625 */
1626 SOCKBUF_LOCK(&so->so_rcv);
1627 unp_scan(so->so_rcv.sb_mb, unp_mark);
1628 SOCKBUF_UNLOCK(&so->so_rcv);
1629 }
1630 } while (unp_defer);
1631 sx_sunlock(&filelist_lock);
1632 /*
1633 * XXXRW: The following comments need updating for a post-SMPng and
1634 * deferred unp_gc() world, but are still generally accurate.
1635 *
1636 * We grab an extra reference to each of the file table entries
1637 * that are not otherwise accessible and then free the rights
1638 * that are stored in messages on them.
1639 *
1640 * The bug in the orginal code is a little tricky, so I'll describe
1641 * what's wrong with it here.
1642 *
1643 * It is incorrect to simply unp_discard each entry for f_msgcount
1644 * times -- consider the case of sockets A and B that contain
1645 * references to each other. On a last close of some other socket,
1646 * we trigger a gc since the number of outstanding rights (unp_rights)
1647 * is non-zero. If during the sweep phase the gc code unp_discards,
1648 * we end up doing a (full) closef on the descriptor. A closef on A
1649 * results in the following chain. Closef calls soo_close, which
1650 * calls soclose. Soclose calls first (through the switch
1651 * uipc_usrreq) unp_detach, which re-invokes unp_gc. Unp_gc simply
1652 * returns because the previous instance had set unp_gcing, and
1653 * we return all the way back to soclose, which marks the socket
1654 * with SS_NOFDREF, and then calls sofree. Sofree calls sorflush
1655 * to free up the rights that are queued in messages on the socket A,
1656 * i.e., the reference on B. The sorflush calls via the dom_dispose
1657 * switch unp_dispose, which unp_scans with unp_discard. This second
1658 * instance of unp_discard just calls closef on B.
1659 *
1660 * Well, a similar chain occurs on B, resulting in a sorflush on B,
1661 * which results in another closef on A. Unfortunately, A is already
1662 * being closed, and the descriptor has already been marked with
1663 * SS_NOFDREF, and soclose panics at this point.
1664 *
1665 * Here, we first take an extra reference to each inaccessible
1666 * descriptor. Then, we call sorflush ourself, since we know
1667 * it is a Unix domain socket anyhow. After we destroy all the
1668 * rights carried in messages, we do a last closef to get rid
1669 * of our extra reference. This is the last close, and the
1670 * unp_detach etc will shut down the socket.
1671 *
1672 * 91/09/19, bsy@cs.cmu.edu
1673 */
1674again:
1675 nfiles_snap = openfiles + nfiles_slack; /* some slack */
1676 extra_ref = malloc(nfiles_snap * sizeof(struct file *), M_TEMP,
1677 M_WAITOK);
1678 sx_slock(&filelist_lock);
1679 if (nfiles_snap < openfiles) {
1680 sx_sunlock(&filelist_lock);
1681 free(extra_ref, M_TEMP);
1682 nfiles_slack += 20;
1683 goto again;
1684 }
1685 for (nunref = 0, fp = LIST_FIRST(&filehead), fpp = extra_ref;
1686 fp != NULL; fp = nextfp) {
1687 nextfp = LIST_NEXT(fp, f_list);
1688 FILE_LOCK(fp);
1689 /*
1690 * If it's not open, skip it
1691 */
1692 if (fp->f_count == 0) {
1693 FILE_UNLOCK(fp);
1694 continue;
1695 }
1696 /*
1697 * If all refs are from msgs, and it's not marked accessible
1698 * then it must be referenced from some unreachable cycle
1699 * of (shut-down) FDs, so include it in our
1700 * list of FDs to remove
1701 */
1702 if (fp->f_count == fp->f_msgcount && !(fp->f_gcflag & FMARK)) {
1703 *fpp++ = fp;
1704 nunref++;
1705 fp->f_count++;
1706 }
1707 FILE_UNLOCK(fp);
1708 }
1709 sx_sunlock(&filelist_lock);
1710 /*
1711 * for each FD on our hit list, do the following two things
1712 */
1713 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) {
1714 struct file *tfp = *fpp;
1715 FILE_LOCK(tfp);
1716 if (tfp->f_type == DTYPE_SOCKET &&
1717 tfp->f_data != NULL) {
1718 FILE_UNLOCK(tfp);
1719 sorflush(tfp->f_data);
1720 } else {
1721 FILE_UNLOCK(tfp);
1722 }
1723 }
1724 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) {
1725 closef(*fpp, (struct thread *) NULL);
1726 unp_recycled++;
1727 }
1728 free(extra_ref, M_TEMP);
1729}
1730
1731void
1732unp_dispose(struct mbuf *m)
1733{
1734
1735 if (m)
1736 unp_scan(m, unp_discard);
1737}
1738
1739static int
1740unp_listen(struct socket *so, struct unpcb *unp, int backlog,
1741 struct thread *td)
1742{
1743 int error;
1744
1745 UNP_LOCK_ASSERT();
1746
1747 SOCK_LOCK(so);
1748 error = solisten_proto_check(so);
1749 if (error == 0) {
1750 cru2x(td->td_ucred, &unp->unp_peercred);
1751 unp->unp_flags |= UNP_HAVEPCCACHED;
1752 solisten_proto(so, backlog);
1753 }
1754 SOCK_UNLOCK(so);
1755 return (error);
1756}
1757
1758static void
1759unp_scan(struct mbuf *m0, void (*op)(struct file *))
1760{
1761 struct mbuf *m;
1762 struct file **rp;
1763 struct cmsghdr *cm;
1764 void *data;
1765 int i;
1766 socklen_t clen, datalen;
1767 int qfds;
1768
1769 while (m0 != NULL) {
1770 for (m = m0; m; m = m->m_next) {
1771 if (m->m_type != MT_CONTROL)
1772 continue;
1773
1774 cm = mtod(m, struct cmsghdr *);
1775 clen = m->m_len;
1776
1777 while (cm != NULL) {
1778 if (sizeof(*cm) > clen || cm->cmsg_len > clen)
1779 break;
1780
1781 data = CMSG_DATA(cm);
1782 datalen = (caddr_t)cm + cm->cmsg_len
1783 - (caddr_t)data;
1784
1785 if (cm->cmsg_level == SOL_SOCKET &&
1786 cm->cmsg_type == SCM_RIGHTS) {
1787 qfds = datalen / sizeof (struct file *);
1788 rp = data;
1789 for (i = 0; i < qfds; i++)
1790 (*op)(*rp++);
1791 }
1792
1793 if (CMSG_SPACE(datalen) < clen) {
1794 clen -= CMSG_SPACE(datalen);
1795 cm = (struct cmsghdr *)
1796 ((caddr_t)cm + CMSG_SPACE(datalen));
1797 } else {
1798 clen = 0;
1799 cm = NULL;
1800 }
1801 }
1802 }
1803 m0 = m0->m_act;
1804 }
1805}
1806
1807static void
1808unp_mark(struct file *fp)
1809{
1810 if (fp->f_gcflag & FMARK)
1811 return;
1812 unp_defer++;
1813 fp->f_gcflag |= (FMARK|FDEFER);
1814}
1815
1816static void
1817unp_discard(struct file *fp)
1818{
1819 UNP_LOCK();
1820 FILE_LOCK(fp);
1821 fp->f_msgcount--;
1822 unp_rights--;
1823 FILE_UNLOCK(fp);
1824 UNP_UNLOCK();
1825 (void) closef(fp, (struct thread *)NULL);
1826}
2 * Copyright (c) 1982, 1986, 1989, 1991, 1993
3 * The Regents of the University of California.
4 * Copyright 2004-2006 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: head/sys/kern/uipc_usrreq.c 157370 2006-04-01 15:42:02Z rwatson $");
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/file.h>
44#include <sys/filedesc.h>
45#include <sys/jail.h>
46#include <sys/kernel.h>
47#include <sys/lock.h>
48#include <sys/mac.h>
49#include <sys/mbuf.h>
50#include <sys/mount.h>
51#include <sys/mutex.h>
52#include <sys/namei.h>
53#include <sys/proc.h>
54#include <sys/protosw.h>
55#include <sys/resourcevar.h>
56#include <sys/socket.h>
57#include <sys/socketvar.h>
58#include <sys/signalvar.h>
59#include <sys/stat.h>
60#include <sys/sx.h>
61#include <sys/sysctl.h>
62#include <sys/systm.h>
63#include <sys/taskqueue.h>
64#include <sys/un.h>
65#include <sys/unpcb.h>
66#include <sys/vnode.h>
67
68#include <vm/uma.h>
69
70static uma_zone_t unp_zone;
71static unp_gen_t unp_gencnt;
72static u_int unp_count;
73
74static struct unp_head unp_shead, unp_dhead;
75
76/*
77 * Unix communications domain.
78 *
79 * TODO:
80 * SEQPACKET, RDM
81 * rethink name space problems
82 * need a proper out-of-band
83 * lock pushdown
84 */
85static const struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL };
86static ino_t unp_ino; /* prototype for fake inode numbers */
87struct mbuf *unp_addsockcred(struct thread *, struct mbuf *);
88
89/*
90 * Currently, UNIX domain sockets are protected by a single subsystem lock,
91 * which covers global data structures and variables, the contents of each
92 * per-socket unpcb structure, and the so_pcb field in sockets attached to
93 * the UNIX domain. This provides for a moderate degree of paralellism, as
94 * receive operations on UNIX domain sockets do not need to acquire the
95 * subsystem lock. Finer grained locking to permit send() without acquiring
96 * a global lock would be a logical next step.
97 *
98 * The UNIX domain socket lock preceds all socket layer locks, including the
99 * socket lock and socket buffer lock, permitting UNIX domain socket code to
100 * call into socket support routines without releasing its locks.
101 *
102 * Some caution is required in areas where the UNIX domain socket code enters
103 * VFS in order to create or find rendezvous points. This results in
104 * dropping of the UNIX domain socket subsystem lock, acquisition of the
105 * Giant lock, and potential sleeping. This increases the chances of races,
106 * and exposes weaknesses in the socket->protocol API by offering poor
107 * failure modes.
108 */
109static struct mtx unp_mtx;
110#define UNP_LOCK_INIT() \
111 mtx_init(&unp_mtx, "unp", NULL, MTX_DEF)
112#define UNP_LOCK() mtx_lock(&unp_mtx)
113#define UNP_UNLOCK() mtx_unlock(&unp_mtx)
114#define UNP_LOCK_ASSERT() mtx_assert(&unp_mtx, MA_OWNED)
115#define UNP_UNLOCK_ASSERT() mtx_assert(&unp_mtx, MA_NOTOWNED)
116
117/*
118 * Garbage collection of cyclic file descriptor/socket references occurs
119 * asynchronously in a taskqueue context in order to avoid recursion and
120 * reentrance in the UNIX domain socket, file descriptor, and socket layer
121 * code. See unp_gc() for a full description.
122 */
123static struct task unp_gc_task;
124
125static int unp_attach(struct socket *);
126static void unp_detach(struct unpcb *);
127static int unp_bind(struct unpcb *,struct sockaddr *, struct thread *);
128static int unp_connect(struct socket *,struct sockaddr *, struct thread *);
129static int unp_connect2(struct socket *so, struct socket *so2, int);
130static void unp_disconnect(struct unpcb *);
131static void unp_shutdown(struct unpcb *);
132static void unp_drop(struct unpcb *, int);
133static void unp_gc(__unused void *, int);
134static void unp_scan(struct mbuf *, void (*)(struct file *));
135static void unp_mark(struct file *);
136static void unp_discard(struct file *);
137static void unp_freerights(struct file **, int);
138static int unp_internalize(struct mbuf **, struct thread *);
139static int unp_listen(struct socket *, struct unpcb *, int,
140 struct thread *);
141
142static void
143uipc_abort(struct socket *so)
144{
145 struct unpcb *unp;
146
147 unp = sotounpcb(so);
148 KASSERT(unp != NULL, ("uipc_abort: unp == NULL"));
149 UNP_LOCK();
150 unp_drop(unp, ECONNABORTED);
151 unp_detach(unp);
152 UNP_UNLOCK_ASSERT();
153}
154
155static int
156uipc_accept(struct socket *so, struct sockaddr **nam)
157{
158 struct unpcb *unp;
159 const struct sockaddr *sa;
160
161 /*
162 * Pass back name of connected socket,
163 * if it was bound and we are still connected
164 * (our peer may have closed already!).
165 */
166 unp = sotounpcb(so);
167 KASSERT(unp != NULL, ("uipc_accept: unp == NULL"));
168 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
169 UNP_LOCK();
170 if (unp->unp_conn != NULL && unp->unp_conn->unp_addr != NULL)
171 sa = (struct sockaddr *) unp->unp_conn->unp_addr;
172 else
173 sa = &sun_noname;
174 bcopy(sa, *nam, sa->sa_len);
175 UNP_UNLOCK();
176 return (0);
177}
178
179static int
180uipc_attach(struct socket *so, int proto, struct thread *td)
181{
182
183 return (unp_attach(so));
184}
185
186static int
187uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
188{
189 struct unpcb *unp;
190 int error;
191
192 unp = sotounpcb(so);
193 KASSERT(unp != NULL, ("uipc_bind: unp == NULL"));
194 UNP_LOCK();
195 error = unp_bind(unp, nam, td);
196 UNP_UNLOCK();
197 return (error);
198}
199
200static int
201uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
202{
203 struct unpcb *unp;
204 int error;
205
206 KASSERT(td == curthread, ("uipc_connect: td != curthread"));
207 unp = sotounpcb(so);
208 KASSERT(unp != NULL, ("uipc_connect: unp == NULL"));
209 UNP_LOCK();
210 error = unp_connect(so, nam, td);
211 UNP_UNLOCK();
212 return (error);
213}
214
215int
216uipc_connect2(struct socket *so1, struct socket *so2)
217{
218 struct unpcb *unp;
219 int error;
220
221 unp = sotounpcb(so1);
222 KASSERT(unp != NULL, ("uipc_connect2: unp == NULL"));
223 UNP_LOCK();
224 error = unp_connect2(so1, so2, PRU_CONNECT2);
225 UNP_UNLOCK();
226 return (error);
227}
228
229/* control is EOPNOTSUPP */
230
231static void
232uipc_detach(struct socket *so)
233{
234 struct unpcb *unp;
235
236 unp = sotounpcb(so);
237 KASSERT(unp != NULL, ("uipc_detach: unp == NULL"));
238 UNP_LOCK();
239 unp_detach(unp);
240 UNP_UNLOCK_ASSERT();
241}
242
243static int
244uipc_disconnect(struct socket *so)
245{
246 struct unpcb *unp;
247
248 unp = sotounpcb(so);
249 KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL"));
250 UNP_LOCK();
251 unp_disconnect(unp);
252 UNP_UNLOCK();
253 return (0);
254}
255
256static int
257uipc_listen(struct socket *so, int backlog, struct thread *td)
258{
259 struct unpcb *unp;
260 int error;
261
262 unp = sotounpcb(so);
263 KASSERT(unp != NULL, ("uipc_listen: unp == NULL"));
264 UNP_LOCK();
265 if (unp->unp_vnode == NULL) {
266 UNP_UNLOCK();
267 return (EINVAL);
268 }
269 error = unp_listen(so, unp, backlog, td);
270 UNP_UNLOCK();
271 return (error);
272}
273
274static int
275uipc_peeraddr(struct socket *so, struct sockaddr **nam)
276{
277 struct unpcb *unp;
278 const struct sockaddr *sa;
279
280 unp = sotounpcb(so);
281 KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL"));
282 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
283 UNP_LOCK();
284 if (unp->unp_conn != NULL && unp->unp_conn->unp_addr!= NULL)
285 sa = (struct sockaddr *) unp->unp_conn->unp_addr;
286 else {
287 /*
288 * XXX: It seems that this test always fails even when
289 * connection is established. So, this else clause is
290 * added as workaround to return PF_LOCAL sockaddr.
291 */
292 sa = &sun_noname;
293 }
294 bcopy(sa, *nam, sa->sa_len);
295 UNP_UNLOCK();
296 return (0);
297}
298
299static int
300uipc_rcvd(struct socket *so, int flags)
301{
302 struct unpcb *unp;
303 struct socket *so2;
304 u_long newhiwat;
305
306 unp = sotounpcb(so);
307 KASSERT(unp != NULL, ("uipc_rcvd: unp == NULL"));
308 UNP_LOCK();
309 switch (so->so_type) {
310 case SOCK_DGRAM:
311 panic("uipc_rcvd DGRAM?");
312 /*NOTREACHED*/
313
314 case SOCK_STREAM:
315 if (unp->unp_conn == NULL)
316 break;
317 so2 = unp->unp_conn->unp_socket;
318 SOCKBUF_LOCK(&so2->so_snd);
319 SOCKBUF_LOCK(&so->so_rcv);
320 /*
321 * Adjust backpressure on sender
322 * and wakeup any waiting to write.
323 */
324 so2->so_snd.sb_mbmax += unp->unp_mbcnt - so->so_rcv.sb_mbcnt;
325 unp->unp_mbcnt = so->so_rcv.sb_mbcnt;
326 newhiwat = so2->so_snd.sb_hiwat + unp->unp_cc -
327 so->so_rcv.sb_cc;
328 (void)chgsbsize(so2->so_cred->cr_uidinfo, &so2->so_snd.sb_hiwat,
329 newhiwat, RLIM_INFINITY);
330 unp->unp_cc = so->so_rcv.sb_cc;
331 SOCKBUF_UNLOCK(&so->so_rcv);
332 sowwakeup_locked(so2);
333 break;
334
335 default:
336 panic("uipc_rcvd unknown socktype");
337 }
338 UNP_UNLOCK();
339 return (0);
340}
341
342/* pru_rcvoob is EOPNOTSUPP */
343
344static int
345uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
346 struct mbuf *control, struct thread *td)
347{
348 int error = 0;
349 struct unpcb *unp;
350 struct socket *so2;
351 u_long newhiwat;
352
353 unp = sotounpcb(so);
354 KASSERT(unp != NULL, ("uipc_send: unp == NULL"));
355 if (flags & PRUS_OOB) {
356 error = EOPNOTSUPP;
357 goto release;
358 }
359
360 if (control != NULL && (error = unp_internalize(&control, td)))
361 goto release;
362
363 UNP_LOCK();
364 switch (so->so_type) {
365 case SOCK_DGRAM:
366 {
367 const struct sockaddr *from;
368
369 if (nam != NULL) {
370 if (unp->unp_conn != NULL) {
371 error = EISCONN;
372 break;
373 }
374 error = unp_connect(so, nam, td);
375 if (error)
376 break;
377 } else {
378 if (unp->unp_conn == NULL) {
379 error = ENOTCONN;
380 break;
381 }
382 }
383 so2 = unp->unp_conn->unp_socket;
384 if (unp->unp_addr != NULL)
385 from = (struct sockaddr *)unp->unp_addr;
386 else
387 from = &sun_noname;
388 if (unp->unp_conn->unp_flags & UNP_WANTCRED)
389 control = unp_addsockcred(td, control);
390 SOCKBUF_LOCK(&so2->so_rcv);
391 if (sbappendaddr_locked(&so2->so_rcv, from, m, control)) {
392 sorwakeup_locked(so2);
393 m = NULL;
394 control = NULL;
395 } else {
396 SOCKBUF_UNLOCK(&so2->so_rcv);
397 error = ENOBUFS;
398 }
399 if (nam != NULL)
400 unp_disconnect(unp);
401 break;
402 }
403
404 case SOCK_STREAM:
405 /* Connect if not connected yet. */
406 /*
407 * Note: A better implementation would complain
408 * if not equal to the peer's address.
409 */
410 if ((so->so_state & SS_ISCONNECTED) == 0) {
411 if (nam != NULL) {
412 error = unp_connect(so, nam, td);
413 if (error)
414 break; /* XXX */
415 } else {
416 error = ENOTCONN;
417 break;
418 }
419 }
420
421 SOCKBUF_LOCK(&so->so_snd);
422 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
423 SOCKBUF_UNLOCK(&so->so_snd);
424 error = EPIPE;
425 break;
426 }
427 if (unp->unp_conn == NULL)
428 panic("uipc_send connected but no connection?");
429 so2 = unp->unp_conn->unp_socket;
430 SOCKBUF_LOCK(&so2->so_rcv);
431 if (unp->unp_conn->unp_flags & UNP_WANTCRED) {
432 /*
433 * Credentials are passed only once on
434 * SOCK_STREAM.
435 */
436 unp->unp_conn->unp_flags &= ~UNP_WANTCRED;
437 control = unp_addsockcred(td, control);
438 }
439 /*
440 * Send to paired receive port, and then reduce
441 * send buffer hiwater marks to maintain backpressure.
442 * Wake up readers.
443 */
444 if (control != NULL) {
445 if (sbappendcontrol_locked(&so2->so_rcv, m, control))
446 control = NULL;
447 } else {
448 sbappend_locked(&so2->so_rcv, m);
449 }
450 so->so_snd.sb_mbmax -=
451 so2->so_rcv.sb_mbcnt - unp->unp_conn->unp_mbcnt;
452 unp->unp_conn->unp_mbcnt = so2->so_rcv.sb_mbcnt;
453 newhiwat = so->so_snd.sb_hiwat -
454 (so2->so_rcv.sb_cc - unp->unp_conn->unp_cc);
455 (void)chgsbsize(so->so_cred->cr_uidinfo, &so->so_snd.sb_hiwat,
456 newhiwat, RLIM_INFINITY);
457 SOCKBUF_UNLOCK(&so->so_snd);
458 unp->unp_conn->unp_cc = so2->so_rcv.sb_cc;
459 sorwakeup_locked(so2);
460 m = NULL;
461 break;
462
463 default:
464 panic("uipc_send unknown socktype");
465 }
466
467 /*
468 * SEND_EOF is equivalent to a SEND followed by
469 * a SHUTDOWN.
470 */
471 if (flags & PRUS_EOF) {
472 socantsendmore(so);
473 unp_shutdown(unp);
474 }
475 UNP_UNLOCK();
476
477 if (control != NULL && error != 0)
478 unp_dispose(control);
479
480release:
481 if (control != NULL)
482 m_freem(control);
483 if (m != NULL)
484 m_freem(m);
485 return (error);
486}
487
488static int
489uipc_sense(struct socket *so, struct stat *sb)
490{
491 struct unpcb *unp;
492 struct socket *so2;
493
494 unp = sotounpcb(so);
495 KASSERT(unp != NULL, ("uipc_sense: unp == NULL"));
496 UNP_LOCK();
497 sb->st_blksize = so->so_snd.sb_hiwat;
498 if (so->so_type == SOCK_STREAM && unp->unp_conn != NULL) {
499 so2 = unp->unp_conn->unp_socket;
500 sb->st_blksize += so2->so_rcv.sb_cc;
501 }
502 sb->st_dev = NODEV;
503 if (unp->unp_ino == 0)
504 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino;
505 sb->st_ino = unp->unp_ino;
506 UNP_UNLOCK();
507 return (0);
508}
509
510static int
511uipc_shutdown(struct socket *so)
512{
513 struct unpcb *unp;
514
515 unp = sotounpcb(so);
516 KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL"));
517 UNP_LOCK();
518 socantsendmore(so);
519 unp_shutdown(unp);
520 UNP_UNLOCK();
521 return (0);
522}
523
524static int
525uipc_sockaddr(struct socket *so, struct sockaddr **nam)
526{
527 struct unpcb *unp;
528 const struct sockaddr *sa;
529
530 unp = sotounpcb(so);
531 KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL"));
532 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
533 UNP_LOCK();
534 if (unp->unp_addr != NULL)
535 sa = (struct sockaddr *) unp->unp_addr;
536 else
537 sa = &sun_noname;
538 bcopy(sa, *nam, sa->sa_len);
539 UNP_UNLOCK();
540 return (0);
541}
542
543struct pr_usrreqs uipc_usrreqs = {
544 .pru_abort = uipc_abort,
545 .pru_accept = uipc_accept,
546 .pru_attach = uipc_attach,
547 .pru_bind = uipc_bind,
548 .pru_connect = uipc_connect,
549 .pru_connect2 = uipc_connect2,
550 .pru_detach = uipc_detach,
551 .pru_disconnect = uipc_disconnect,
552 .pru_listen = uipc_listen,
553 .pru_peeraddr = uipc_peeraddr,
554 .pru_rcvd = uipc_rcvd,
555 .pru_send = uipc_send,
556 .pru_sense = uipc_sense,
557 .pru_shutdown = uipc_shutdown,
558 .pru_sockaddr = uipc_sockaddr,
559 .pru_sosend = sosend,
560 .pru_soreceive = soreceive,
561 .pru_sopoll = sopoll,
562};
563
564int
565uipc_ctloutput(struct socket *so, struct sockopt *sopt)
566{
567 struct unpcb *unp;
568 struct xucred xu;
569 int error, optval;
570
571 if (sopt->sopt_level != 0)
572 return (EINVAL);
573
574 unp = sotounpcb(so);
575 KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL"));
576 UNP_LOCK();
577 error = 0;
578 switch (sopt->sopt_dir) {
579 case SOPT_GET:
580 switch (sopt->sopt_name) {
581 case LOCAL_PEERCRED:
582 if (unp->unp_flags & UNP_HAVEPC)
583 xu = unp->unp_peercred;
584 else {
585 if (so->so_type == SOCK_STREAM)
586 error = ENOTCONN;
587 else
588 error = EINVAL;
589 }
590 if (error == 0)
591 error = sooptcopyout(sopt, &xu, sizeof(xu));
592 break;
593 case LOCAL_CREDS:
594 optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0;
595 error = sooptcopyout(sopt, &optval, sizeof(optval));
596 break;
597 case LOCAL_CONNWAIT:
598 optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0;
599 error = sooptcopyout(sopt, &optval, sizeof(optval));
600 break;
601 default:
602 error = EOPNOTSUPP;
603 break;
604 }
605 break;
606 case SOPT_SET:
607 switch (sopt->sopt_name) {
608 case LOCAL_CREDS:
609 case LOCAL_CONNWAIT:
610 error = sooptcopyin(sopt, &optval, sizeof(optval),
611 sizeof(optval));
612 if (error)
613 break;
614
615#define OPTSET(bit) \
616 if (optval) \
617 unp->unp_flags |= bit; \
618 else \
619 unp->unp_flags &= ~bit;
620
621 switch (sopt->sopt_name) {
622 case LOCAL_CREDS:
623 OPTSET(UNP_WANTCRED);
624 break;
625 case LOCAL_CONNWAIT:
626 OPTSET(UNP_CONNWAIT);
627 break;
628 default:
629 break;
630 }
631 break;
632#undef OPTSET
633 default:
634 error = ENOPROTOOPT;
635 break;
636 }
637 break;
638 default:
639 error = EOPNOTSUPP;
640 break;
641 }
642 UNP_UNLOCK();
643 return (error);
644}
645
646/*
647 * Both send and receive buffers are allocated PIPSIZ bytes of buffering
648 * for stream sockets, although the total for sender and receiver is
649 * actually only PIPSIZ.
650 * Datagram sockets really use the sendspace as the maximum datagram size,
651 * and don't really want to reserve the sendspace. Their recvspace should
652 * be large enough for at least one max-size datagram plus address.
653 */
654#ifndef PIPSIZ
655#define PIPSIZ 8192
656#endif
657static u_long unpst_sendspace = PIPSIZ;
658static u_long unpst_recvspace = PIPSIZ;
659static u_long unpdg_sendspace = 2*1024; /* really max datagram size */
660static u_long unpdg_recvspace = 4*1024;
661
662static int unp_rights; /* file descriptors in flight */
663
664SYSCTL_DECL(_net_local_stream);
665SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW,
666 &unpst_sendspace, 0, "");
667SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW,
668 &unpst_recvspace, 0, "");
669SYSCTL_DECL(_net_local_dgram);
670SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW,
671 &unpdg_sendspace, 0, "");
672SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW,
673 &unpdg_recvspace, 0, "");
674SYSCTL_DECL(_net_local);
675SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, "");
676
677static int
678unp_attach(struct socket *so)
679{
680 struct unpcb *unp;
681 int error;
682
683 KASSERT(so->so_pcb == NULL, ("unp_attach: so_pcb != NULL"));
684 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
685 switch (so->so_type) {
686
687 case SOCK_STREAM:
688 error = soreserve(so, unpst_sendspace, unpst_recvspace);
689 break;
690
691 case SOCK_DGRAM:
692 error = soreserve(so, unpdg_sendspace, unpdg_recvspace);
693 break;
694
695 default:
696 panic("unp_attach");
697 }
698 if (error)
699 return (error);
700 }
701 unp = uma_zalloc(unp_zone, M_WAITOK | M_ZERO);
702 if (unp == NULL)
703 return (ENOBUFS);
704 LIST_INIT(&unp->unp_refs);
705 unp->unp_socket = so;
706 so->so_pcb = unp;
707
708 UNP_LOCK();
709 unp->unp_gencnt = ++unp_gencnt;
710 unp_count++;
711 LIST_INSERT_HEAD(so->so_type == SOCK_DGRAM ? &unp_dhead
712 : &unp_shead, unp, unp_link);
713 UNP_UNLOCK();
714
715 return (0);
716}
717
718static void
719unp_detach(struct unpcb *unp)
720{
721 struct vnode *vp;
722 int local_unp_rights;
723
724 UNP_LOCK_ASSERT();
725
726 LIST_REMOVE(unp, unp_link);
727 unp->unp_gencnt = ++unp_gencnt;
728 --unp_count;
729 if ((vp = unp->unp_vnode) != NULL) {
730 /*
731 * XXXRW: should v_socket be frobbed only while holding
732 * Giant?
733 */
734 unp->unp_vnode->v_socket = NULL;
735 unp->unp_vnode = NULL;
736 }
737 if (unp->unp_conn != NULL)
738 unp_disconnect(unp);
739 while (!LIST_EMPTY(&unp->unp_refs)) {
740 struct unpcb *ref = LIST_FIRST(&unp->unp_refs);
741 unp_drop(ref, ECONNRESET);
742 }
743 soisdisconnected(unp->unp_socket);
744 unp->unp_socket->so_pcb = NULL;
745 local_unp_rights = unp_rights;
746 UNP_UNLOCK();
747 if (unp->unp_addr != NULL)
748 FREE(unp->unp_addr, M_SONAME);
749 uma_zfree(unp_zone, unp);
750 if (vp) {
751 int vfslocked;
752
753 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
754 vrele(vp);
755 VFS_UNLOCK_GIANT(vfslocked);
756 }
757 if (local_unp_rights)
758 taskqueue_enqueue(taskqueue_thread, &unp_gc_task);
759}
760
761static int
762unp_bind(struct unpcb *unp, struct sockaddr *nam, struct thread *td)
763{
764 struct sockaddr_un *soun = (struct sockaddr_un *)nam;
765 struct vnode *vp;
766 struct mount *mp;
767 struct vattr vattr;
768 int error, namelen;
769 struct nameidata nd;
770 char *buf;
771
772 UNP_LOCK_ASSERT();
773
774 /*
775 * XXXRW: This test-and-set of unp_vnode is non-atomic; the
776 * unlocked read here is fine, but the value of unp_vnode needs
777 * to be tested again after we do all the lookups to see if the
778 * pcb is still unbound?
779 */
780 if (unp->unp_vnode != NULL)
781 return (EINVAL);
782
783 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path);
784 if (namelen <= 0)
785 return (EINVAL);
786
787 UNP_UNLOCK();
788
789 buf = malloc(namelen + 1, M_TEMP, M_WAITOK);
790 strlcpy(buf, soun->sun_path, namelen + 1);
791
792 mtx_lock(&Giant);
793restart:
794 mtx_assert(&Giant, MA_OWNED);
795 NDINIT(&nd, CREATE, NOFOLLOW | LOCKPARENT | SAVENAME, UIO_SYSSPACE,
796 buf, td);
797/* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
798 error = namei(&nd);
799 if (error)
800 goto done;
801 vp = nd.ni_vp;
802 if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
803 NDFREE(&nd, NDF_ONLY_PNBUF);
804 if (nd.ni_dvp == vp)
805 vrele(nd.ni_dvp);
806 else
807 vput(nd.ni_dvp);
808 if (vp != NULL) {
809 vrele(vp);
810 error = EADDRINUSE;
811 goto done;
812 }
813 error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH);
814 if (error)
815 goto done;
816 goto restart;
817 }
818 VATTR_NULL(&vattr);
819 vattr.va_type = VSOCK;
820 vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask);
821#ifdef MAC
822 error = mac_check_vnode_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd,
823 &vattr);
824#endif
825 if (error == 0) {
826 VOP_LEASE(nd.ni_dvp, td, td->td_ucred, LEASE_WRITE);
827 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
828 }
829 NDFREE(&nd, NDF_ONLY_PNBUF);
830 vput(nd.ni_dvp);
831 if (error) {
832 vn_finished_write(mp);
833 goto done;
834 }
835 vp = nd.ni_vp;
836 ASSERT_VOP_LOCKED(vp, "unp_bind");
837 soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK);
838 UNP_LOCK();
839 vp->v_socket = unp->unp_socket;
840 unp->unp_vnode = vp;
841 unp->unp_addr = soun;
842 UNP_UNLOCK();
843 VOP_UNLOCK(vp, 0, td);
844 vn_finished_write(mp);
845done:
846 mtx_unlock(&Giant);
847 free(buf, M_TEMP);
848 UNP_LOCK();
849 return (error);
850}
851
852static int
853unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
854{
855 struct sockaddr_un *soun = (struct sockaddr_un *)nam;
856 struct vnode *vp;
857 struct socket *so2, *so3;
858 struct unpcb *unp, *unp2, *unp3;
859 int error, len;
860 struct nameidata nd;
861 char buf[SOCK_MAXADDRLEN];
862 struct sockaddr *sa;
863
864 UNP_LOCK_ASSERT();
865
866 unp = sotounpcb(so);
867 KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
868 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
869 if (len <= 0)
870 return (EINVAL);
871 strlcpy(buf, soun->sun_path, len + 1);
872 UNP_UNLOCK();
873 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
874 mtx_lock(&Giant);
875 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, buf, td);
876 error = namei(&nd);
877 if (error)
878 vp = NULL;
879 else
880 vp = nd.ni_vp;
881 ASSERT_VOP_LOCKED(vp, "unp_connect");
882 NDFREE(&nd, NDF_ONLY_PNBUF);
883 if (error)
884 goto bad;
885
886 if (vp->v_type != VSOCK) {
887 error = ENOTSOCK;
888 goto bad;
889 }
890 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td);
891 if (error)
892 goto bad;
893 mtx_unlock(&Giant);
894 UNP_LOCK();
895 unp = sotounpcb(so);
896 KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
897 so2 = vp->v_socket;
898 if (so2 == NULL) {
899 error = ECONNREFUSED;
900 goto bad2;
901 }
902 if (so->so_type != so2->so_type) {
903 error = EPROTOTYPE;
904 goto bad2;
905 }
906 if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
907 if (so2->so_options & SO_ACCEPTCONN) {
908 /*
909 * NB: drop locks here so unp_attach is entered
910 * w/o locks; this avoids a recursive lock
911 * of the head and holding sleep locks across
912 * a (potentially) blocking malloc.
913 */
914 UNP_UNLOCK();
915 so3 = sonewconn(so2, 0);
916 UNP_LOCK();
917 } else
918 so3 = NULL;
919 if (so3 == NULL) {
920 error = ECONNREFUSED;
921 goto bad2;
922 }
923 unp = sotounpcb(so);
924 unp2 = sotounpcb(so2);
925 unp3 = sotounpcb(so3);
926 if (unp2->unp_addr != NULL) {
927 bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len);
928 unp3->unp_addr = (struct sockaddr_un *) sa;
929 sa = NULL;
930 }
931 /*
932 * unp_peercred management:
933 *
934 * The connecter's (client's) credentials are copied
935 * from its process structure at the time of connect()
936 * (which is now).
937 */
938 cru2x(td->td_ucred, &unp3->unp_peercred);
939 unp3->unp_flags |= UNP_HAVEPC;
940 /*
941 * The receiver's (server's) credentials are copied
942 * from the unp_peercred member of socket on which the
943 * former called listen(); unp_listen() cached that
944 * process's credentials at that time so we can use
945 * them now.
946 */
947 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED,
948 ("unp_connect: listener without cached peercred"));
949 memcpy(&unp->unp_peercred, &unp2->unp_peercred,
950 sizeof(unp->unp_peercred));
951 unp->unp_flags |= UNP_HAVEPC;
952#ifdef MAC
953 SOCK_LOCK(so);
954 mac_set_socket_peer_from_socket(so, so3);
955 mac_set_socket_peer_from_socket(so3, so);
956 SOCK_UNLOCK(so);
957#endif
958
959 so2 = so3;
960 }
961 error = unp_connect2(so, so2, PRU_CONNECT);
962bad2:
963 UNP_UNLOCK();
964 mtx_lock(&Giant);
965bad:
966 mtx_assert(&Giant, MA_OWNED);
967 if (vp != NULL)
968 vput(vp);
969 mtx_unlock(&Giant);
970 free(sa, M_SONAME);
971 UNP_LOCK();
972 return (error);
973}
974
975static int
976unp_connect2(struct socket *so, struct socket *so2, int req)
977{
978 struct unpcb *unp = sotounpcb(so);
979 struct unpcb *unp2;
980
981 UNP_LOCK_ASSERT();
982
983 if (so2->so_type != so->so_type)
984 return (EPROTOTYPE);
985 unp2 = sotounpcb(so2);
986 KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL"));
987 unp->unp_conn = unp2;
988 switch (so->so_type) {
989
990 case SOCK_DGRAM:
991 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
992 soisconnected(so);
993 break;
994
995 case SOCK_STREAM:
996 unp2->unp_conn = unp;
997 if (req == PRU_CONNECT &&
998 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT))
999 soisconnecting(so);
1000 else
1001 soisconnected(so);
1002 soisconnected(so2);
1003 break;
1004
1005 default:
1006 panic("unp_connect2");
1007 }
1008 return (0);
1009}
1010
1011static void
1012unp_disconnect(struct unpcb *unp)
1013{
1014 struct unpcb *unp2 = unp->unp_conn;
1015 struct socket *so;
1016
1017 UNP_LOCK_ASSERT();
1018
1019 if (unp2 == NULL)
1020 return;
1021 unp->unp_conn = NULL;
1022 switch (unp->unp_socket->so_type) {
1023 case SOCK_DGRAM:
1024 LIST_REMOVE(unp, unp_reflink);
1025 so = unp->unp_socket;
1026 SOCK_LOCK(so);
1027 so->so_state &= ~SS_ISCONNECTED;
1028 SOCK_UNLOCK(so);
1029 break;
1030
1031 case SOCK_STREAM:
1032 soisdisconnected(unp->unp_socket);
1033 unp2->unp_conn = NULL;
1034 soisdisconnected(unp2->unp_socket);
1035 break;
1036 }
1037}
1038
1039#ifdef notdef
1040void
1041unp_abort(struct unpcb *unp)
1042{
1043
1044 unp_detach(unp);
1045 UNP_UNLOCK_ASSERT();
1046}
1047#endif
1048
1049/*
1050 * unp_pcblist() assumes that UNIX domain socket memory is never reclaimed
1051 * by the zone (UMA_ZONE_NOFREE), and as such potentially stale pointers
1052 * are safe to reference. It first scans the list of struct unpcb's to
1053 * generate a pointer list, then it rescans its list one entry at a time to
1054 * externalize and copyout. It checks the generation number to see if a
1055 * struct unpcb has been reused, and will skip it if so.
1056 */
1057static int
1058unp_pcblist(SYSCTL_HANDLER_ARGS)
1059{
1060 int error, i, n;
1061 struct unpcb *unp, **unp_list;
1062 unp_gen_t gencnt;
1063 struct xunpgen *xug;
1064 struct unp_head *head;
1065 struct xunpcb *xu;
1066
1067 head = ((intptr_t)arg1 == SOCK_DGRAM ? &unp_dhead : &unp_shead);
1068
1069 /*
1070 * The process of preparing the PCB list is too time-consuming and
1071 * resource-intensive to repeat twice on every request.
1072 */
1073 if (req->oldptr == NULL) {
1074 n = unp_count;
1075 req->oldidx = 2 * (sizeof *xug)
1076 + (n + n/8) * sizeof(struct xunpcb);
1077 return (0);
1078 }
1079
1080 if (req->newptr != NULL)
1081 return (EPERM);
1082
1083 /*
1084 * OK, now we're committed to doing something.
1085 */
1086 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK);
1087 UNP_LOCK();
1088 gencnt = unp_gencnt;
1089 n = unp_count;
1090 UNP_UNLOCK();
1091
1092 xug->xug_len = sizeof *xug;
1093 xug->xug_count = n;
1094 xug->xug_gen = gencnt;
1095 xug->xug_sogen = so_gencnt;
1096 error = SYSCTL_OUT(req, xug, sizeof *xug);
1097 if (error) {
1098 free(xug, M_TEMP);
1099 return (error);
1100 }
1101
1102 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
1103
1104 UNP_LOCK();
1105 for (unp = LIST_FIRST(head), i = 0; unp && i < n;
1106 unp = LIST_NEXT(unp, unp_link)) {
1107 if (unp->unp_gencnt <= gencnt) {
1108 if (cr_cansee(req->td->td_ucred,
1109 unp->unp_socket->so_cred))
1110 continue;
1111 unp_list[i++] = unp;
1112 }
1113 }
1114 UNP_UNLOCK();
1115 n = i; /* in case we lost some during malloc */
1116
1117 error = 0;
1118 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO);
1119 for (i = 0; i < n; i++) {
1120 unp = unp_list[i];
1121 if (unp->unp_gencnt <= gencnt) {
1122 xu->xu_len = sizeof *xu;
1123 xu->xu_unpp = unp;
1124 /*
1125 * XXX - need more locking here to protect against
1126 * connect/disconnect races for SMP.
1127 */
1128 if (unp->unp_addr != NULL)
1129 bcopy(unp->unp_addr, &xu->xu_addr,
1130 unp->unp_addr->sun_len);
1131 if (unp->unp_conn != NULL &&
1132 unp->unp_conn->unp_addr != NULL)
1133 bcopy(unp->unp_conn->unp_addr,
1134 &xu->xu_caddr,
1135 unp->unp_conn->unp_addr->sun_len);
1136 bcopy(unp, &xu->xu_unp, sizeof *unp);
1137 sotoxsocket(unp->unp_socket, &xu->xu_socket);
1138 error = SYSCTL_OUT(req, xu, sizeof *xu);
1139 }
1140 }
1141 free(xu, M_TEMP);
1142 if (!error) {
1143 /*
1144 * Give the user an updated idea of our state.
1145 * If the generation differs from what we told
1146 * her before, she knows that something happened
1147 * while we were processing this request, and it
1148 * might be necessary to retry.
1149 */
1150 xug->xug_gen = unp_gencnt;
1151 xug->xug_sogen = so_gencnt;
1152 xug->xug_count = unp_count;
1153 error = SYSCTL_OUT(req, xug, sizeof *xug);
1154 }
1155 free(unp_list, M_TEMP);
1156 free(xug, M_TEMP);
1157 return (error);
1158}
1159
1160SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLFLAG_RD,
1161 (caddr_t)(long)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
1162 "List of active local datagram sockets");
1163SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLFLAG_RD,
1164 (caddr_t)(long)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
1165 "List of active local stream sockets");
1166
1167static void
1168unp_shutdown(struct unpcb *unp)
1169{
1170 struct socket *so;
1171
1172 UNP_LOCK_ASSERT();
1173
1174 if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn &&
1175 (so = unp->unp_conn->unp_socket))
1176 socantrcvmore(so);
1177}
1178
1179static void
1180unp_drop(struct unpcb *unp, int errno)
1181{
1182 struct socket *so = unp->unp_socket;
1183
1184 UNP_LOCK_ASSERT();
1185
1186 so->so_error = errno;
1187 unp_disconnect(unp);
1188}
1189
1190#ifdef notdef
1191void
1192unp_drain(void)
1193{
1194
1195}
1196#endif
1197
1198static void
1199unp_freerights(struct file **rp, int fdcount)
1200{
1201 int i;
1202 struct file *fp;
1203
1204 for (i = 0; i < fdcount; i++) {
1205 fp = *rp;
1206 /*
1207 * zero the pointer before calling
1208 * unp_discard since it may end up
1209 * in unp_gc()..
1210 *
1211 * XXXRW: This is less true than it used to be.
1212 */
1213 *rp++ = 0;
1214 unp_discard(fp);
1215 }
1216}
1217
1218int
1219unp_externalize(struct mbuf *control, struct mbuf **controlp)
1220{
1221 struct thread *td = curthread; /* XXX */
1222 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1223 int i;
1224 int *fdp;
1225 struct file **rp;
1226 struct file *fp;
1227 void *data;
1228 socklen_t clen = control->m_len, datalen;
1229 int error, newfds;
1230 int f;
1231 u_int newlen;
1232
1233 UNP_UNLOCK_ASSERT();
1234
1235 error = 0;
1236 if (controlp != NULL) /* controlp == NULL => free control messages */
1237 *controlp = NULL;
1238
1239 while (cm != NULL) {
1240 if (sizeof(*cm) > clen || cm->cmsg_len > clen) {
1241 error = EINVAL;
1242 break;
1243 }
1244
1245 data = CMSG_DATA(cm);
1246 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1247
1248 if (cm->cmsg_level == SOL_SOCKET
1249 && cm->cmsg_type == SCM_RIGHTS) {
1250 newfds = datalen / sizeof(struct file *);
1251 rp = data;
1252
1253 /* If we're not outputting the descriptors free them. */
1254 if (error || controlp == NULL) {
1255 unp_freerights(rp, newfds);
1256 goto next;
1257 }
1258 FILEDESC_LOCK(td->td_proc->p_fd);
1259 /* if the new FD's will not fit free them. */
1260 if (!fdavail(td, newfds)) {
1261 FILEDESC_UNLOCK(td->td_proc->p_fd);
1262 error = EMSGSIZE;
1263 unp_freerights(rp, newfds);
1264 goto next;
1265 }
1266 /*
1267 * now change each pointer to an fd in the global
1268 * table to an integer that is the index to the
1269 * local fd table entry that we set up to point
1270 * to the global one we are transferring.
1271 */
1272 newlen = newfds * sizeof(int);
1273 *controlp = sbcreatecontrol(NULL, newlen,
1274 SCM_RIGHTS, SOL_SOCKET);
1275 if (*controlp == NULL) {
1276 FILEDESC_UNLOCK(td->td_proc->p_fd);
1277 error = E2BIG;
1278 unp_freerights(rp, newfds);
1279 goto next;
1280 }
1281
1282 fdp = (int *)
1283 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1284 for (i = 0; i < newfds; i++) {
1285 if (fdalloc(td, 0, &f))
1286 panic("unp_externalize fdalloc failed");
1287 fp = *rp++;
1288 td->td_proc->p_fd->fd_ofiles[f] = fp;
1289 FILE_LOCK(fp);
1290 fp->f_msgcount--;
1291 FILE_UNLOCK(fp);
1292 unp_rights--;
1293 *fdp++ = f;
1294 }
1295 FILEDESC_UNLOCK(td->td_proc->p_fd);
1296 } else { /* We can just copy anything else across */
1297 if (error || controlp == NULL)
1298 goto next;
1299 *controlp = sbcreatecontrol(NULL, datalen,
1300 cm->cmsg_type, cm->cmsg_level);
1301 if (*controlp == NULL) {
1302 error = ENOBUFS;
1303 goto next;
1304 }
1305 bcopy(data,
1306 CMSG_DATA(mtod(*controlp, struct cmsghdr *)),
1307 datalen);
1308 }
1309
1310 controlp = &(*controlp)->m_next;
1311
1312next:
1313 if (CMSG_SPACE(datalen) < clen) {
1314 clen -= CMSG_SPACE(datalen);
1315 cm = (struct cmsghdr *)
1316 ((caddr_t)cm + CMSG_SPACE(datalen));
1317 } else {
1318 clen = 0;
1319 cm = NULL;
1320 }
1321 }
1322
1323 m_freem(control);
1324
1325 return (error);
1326}
1327
1328void
1329unp_init(void)
1330{
1331 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL,
1332 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
1333 if (unp_zone == NULL)
1334 panic("unp_init");
1335 uma_zone_set_max(unp_zone, nmbclusters);
1336 LIST_INIT(&unp_dhead);
1337 LIST_INIT(&unp_shead);
1338 TASK_INIT(&unp_gc_task, 0, unp_gc, NULL);
1339 UNP_LOCK_INIT();
1340}
1341
1342static int
1343unp_internalize(struct mbuf **controlp, struct thread *td)
1344{
1345 struct mbuf *control = *controlp;
1346 struct proc *p = td->td_proc;
1347 struct filedesc *fdescp = p->p_fd;
1348 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1349 struct cmsgcred *cmcred;
1350 struct file **rp;
1351 struct file *fp;
1352 struct timeval *tv;
1353 int i, fd, *fdp;
1354 void *data;
1355 socklen_t clen = control->m_len, datalen;
1356 int error, oldfds;
1357 u_int newlen;
1358
1359 UNP_UNLOCK_ASSERT();
1360
1361 error = 0;
1362 *controlp = NULL;
1363
1364 while (cm != NULL) {
1365 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET
1366 || cm->cmsg_len > clen) {
1367 error = EINVAL;
1368 goto out;
1369 }
1370
1371 data = CMSG_DATA(cm);
1372 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1373
1374 switch (cm->cmsg_type) {
1375 /*
1376 * Fill in credential information.
1377 */
1378 case SCM_CREDS:
1379 *controlp = sbcreatecontrol(NULL, sizeof(*cmcred),
1380 SCM_CREDS, SOL_SOCKET);
1381 if (*controlp == NULL) {
1382 error = ENOBUFS;
1383 goto out;
1384 }
1385
1386 cmcred = (struct cmsgcred *)
1387 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1388 cmcred->cmcred_pid = p->p_pid;
1389 cmcred->cmcred_uid = td->td_ucred->cr_ruid;
1390 cmcred->cmcred_gid = td->td_ucred->cr_rgid;
1391 cmcred->cmcred_euid = td->td_ucred->cr_uid;
1392 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups,
1393 CMGROUP_MAX);
1394 for (i = 0; i < cmcred->cmcred_ngroups; i++)
1395 cmcred->cmcred_groups[i] =
1396 td->td_ucred->cr_groups[i];
1397 break;
1398
1399 case SCM_RIGHTS:
1400 oldfds = datalen / sizeof (int);
1401 /*
1402 * check that all the FDs passed in refer to legal files
1403 * If not, reject the entire operation.
1404 */
1405 fdp = data;
1406 FILEDESC_LOCK(fdescp);
1407 for (i = 0; i < oldfds; i++) {
1408 fd = *fdp++;
1409 if ((unsigned)fd >= fdescp->fd_nfiles ||
1410 fdescp->fd_ofiles[fd] == NULL) {
1411 FILEDESC_UNLOCK(fdescp);
1412 error = EBADF;
1413 goto out;
1414 }
1415 fp = fdescp->fd_ofiles[fd];
1416 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) {
1417 FILEDESC_UNLOCK(fdescp);
1418 error = EOPNOTSUPP;
1419 goto out;
1420 }
1421
1422 }
1423 /*
1424 * Now replace the integer FDs with pointers to
1425 * the associated global file table entry..
1426 */
1427 newlen = oldfds * sizeof(struct file *);
1428 *controlp = sbcreatecontrol(NULL, newlen,
1429 SCM_RIGHTS, SOL_SOCKET);
1430 if (*controlp == NULL) {
1431 FILEDESC_UNLOCK(fdescp);
1432 error = E2BIG;
1433 goto out;
1434 }
1435
1436 fdp = data;
1437 rp = (struct file **)
1438 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1439 for (i = 0; i < oldfds; i++) {
1440 fp = fdescp->fd_ofiles[*fdp++];
1441 *rp++ = fp;
1442 FILE_LOCK(fp);
1443 fp->f_count++;
1444 fp->f_msgcount++;
1445 FILE_UNLOCK(fp);
1446 unp_rights++;
1447 }
1448 FILEDESC_UNLOCK(fdescp);
1449 break;
1450
1451 case SCM_TIMESTAMP:
1452 *controlp = sbcreatecontrol(NULL, sizeof(*tv),
1453 SCM_TIMESTAMP, SOL_SOCKET);
1454 if (*controlp == NULL) {
1455 error = ENOBUFS;
1456 goto out;
1457 }
1458 tv = (struct timeval *)
1459 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1460 microtime(tv);
1461 break;
1462
1463 default:
1464 error = EINVAL;
1465 goto out;
1466 }
1467
1468 controlp = &(*controlp)->m_next;
1469
1470 if (CMSG_SPACE(datalen) < clen) {
1471 clen -= CMSG_SPACE(datalen);
1472 cm = (struct cmsghdr *)
1473 ((caddr_t)cm + CMSG_SPACE(datalen));
1474 } else {
1475 clen = 0;
1476 cm = NULL;
1477 }
1478 }
1479
1480out:
1481 m_freem(control);
1482
1483 return (error);
1484}
1485
1486struct mbuf *
1487unp_addsockcred(struct thread *td, struct mbuf *control)
1488{
1489 struct mbuf *m, *n;
1490 struct sockcred *sc;
1491 int ngroups;
1492 int i;
1493
1494 ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX);
1495
1496 m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET);
1497 if (m == NULL)
1498 return (control);
1499 m->m_next = NULL;
1500
1501 sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *));
1502 sc->sc_uid = td->td_ucred->cr_ruid;
1503 sc->sc_euid = td->td_ucred->cr_uid;
1504 sc->sc_gid = td->td_ucred->cr_rgid;
1505 sc->sc_egid = td->td_ucred->cr_gid;
1506 sc->sc_ngroups = ngroups;
1507 for (i = 0; i < sc->sc_ngroups; i++)
1508 sc->sc_groups[i] = td->td_ucred->cr_groups[i];
1509
1510 /*
1511 * If a control message already exists, append us to the end.
1512 */
1513 if (control != NULL) {
1514 for (n = control; n->m_next != NULL; n = n->m_next)
1515 ;
1516 n->m_next = m;
1517 } else
1518 control = m;
1519
1520 return (control);
1521}
1522
1523/*
1524 * unp_defer indicates whether additional work has been defered for a future
1525 * pass through unp_gc(). It is thread local and does not require explicit
1526 * synchronization.
1527 */
1528static int unp_defer;
1529
1530static int unp_taskcount;
1531SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0, "");
1532
1533static int unp_recycled;
1534SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0, "");
1535
1536static void
1537unp_gc(__unused void *arg, int pending)
1538{
1539 struct file *fp, *nextfp;
1540 struct socket *so;
1541 struct file **extra_ref, **fpp;
1542 int nunref, i;
1543 int nfiles_snap;
1544 int nfiles_slack = 20;
1545
1546 unp_taskcount++;
1547 unp_defer = 0;
1548 /*
1549 * before going through all this, set all FDs to
1550 * be NOT defered and NOT externally accessible
1551 */
1552 sx_slock(&filelist_lock);
1553 LIST_FOREACH(fp, &filehead, f_list)
1554 fp->f_gcflag &= ~(FMARK|FDEFER);
1555 do {
1556 KASSERT(unp_defer >= 0, ("unp_gc: unp_defer %d", unp_defer));
1557 LIST_FOREACH(fp, &filehead, f_list) {
1558 FILE_LOCK(fp);
1559 /*
1560 * If the file is not open, skip it -- could be a
1561 * file in the process of being opened, or in the
1562 * process of being closed. If the file is
1563 * "closing", it may have been marked for deferred
1564 * consideration. Clear the flag now if so.
1565 */
1566 if (fp->f_count == 0) {
1567 if (fp->f_gcflag & FDEFER)
1568 unp_defer--;
1569 fp->f_gcflag &= ~(FMARK|FDEFER);
1570 FILE_UNLOCK(fp);
1571 continue;
1572 }
1573 /*
1574 * If we already marked it as 'defer' in a
1575 * previous pass, then try process it this time
1576 * and un-mark it
1577 */
1578 if (fp->f_gcflag & FDEFER) {
1579 fp->f_gcflag &= ~FDEFER;
1580 unp_defer--;
1581 } else {
1582 /*
1583 * if it's not defered, then check if it's
1584 * already marked.. if so skip it
1585 */
1586 if (fp->f_gcflag & FMARK) {
1587 FILE_UNLOCK(fp);
1588 continue;
1589 }
1590 /*
1591 * If all references are from messages
1592 * in transit, then skip it. it's not
1593 * externally accessible.
1594 */
1595 if (fp->f_count == fp->f_msgcount) {
1596 FILE_UNLOCK(fp);
1597 continue;
1598 }
1599 /*
1600 * If it got this far then it must be
1601 * externally accessible.
1602 */
1603 fp->f_gcflag |= FMARK;
1604 }
1605 /*
1606 * either it was defered, or it is externally
1607 * accessible and not already marked so.
1608 * Now check if it is possibly one of OUR sockets.
1609 */
1610 if (fp->f_type != DTYPE_SOCKET ||
1611 (so = fp->f_data) == NULL) {
1612 FILE_UNLOCK(fp);
1613 continue;
1614 }
1615 FILE_UNLOCK(fp);
1616 if (so->so_proto->pr_domain != &localdomain ||
1617 (so->so_proto->pr_flags&PR_RIGHTS) == 0)
1618 continue;
1619 /*
1620 * So, Ok, it's one of our sockets and it IS externally
1621 * accessible (or was defered). Now we look
1622 * to see if we hold any file descriptors in its
1623 * message buffers. Follow those links and mark them
1624 * as accessible too.
1625 */
1626 SOCKBUF_LOCK(&so->so_rcv);
1627 unp_scan(so->so_rcv.sb_mb, unp_mark);
1628 SOCKBUF_UNLOCK(&so->so_rcv);
1629 }
1630 } while (unp_defer);
1631 sx_sunlock(&filelist_lock);
1632 /*
1633 * XXXRW: The following comments need updating for a post-SMPng and
1634 * deferred unp_gc() world, but are still generally accurate.
1635 *
1636 * We grab an extra reference to each of the file table entries
1637 * that are not otherwise accessible and then free the rights
1638 * that are stored in messages on them.
1639 *
1640 * The bug in the orginal code is a little tricky, so I'll describe
1641 * what's wrong with it here.
1642 *
1643 * It is incorrect to simply unp_discard each entry for f_msgcount
1644 * times -- consider the case of sockets A and B that contain
1645 * references to each other. On a last close of some other socket,
1646 * we trigger a gc since the number of outstanding rights (unp_rights)
1647 * is non-zero. If during the sweep phase the gc code unp_discards,
1648 * we end up doing a (full) closef on the descriptor. A closef on A
1649 * results in the following chain. Closef calls soo_close, which
1650 * calls soclose. Soclose calls first (through the switch
1651 * uipc_usrreq) unp_detach, which re-invokes unp_gc. Unp_gc simply
1652 * returns because the previous instance had set unp_gcing, and
1653 * we return all the way back to soclose, which marks the socket
1654 * with SS_NOFDREF, and then calls sofree. Sofree calls sorflush
1655 * to free up the rights that are queued in messages on the socket A,
1656 * i.e., the reference on B. The sorflush calls via the dom_dispose
1657 * switch unp_dispose, which unp_scans with unp_discard. This second
1658 * instance of unp_discard just calls closef on B.
1659 *
1660 * Well, a similar chain occurs on B, resulting in a sorflush on B,
1661 * which results in another closef on A. Unfortunately, A is already
1662 * being closed, and the descriptor has already been marked with
1663 * SS_NOFDREF, and soclose panics at this point.
1664 *
1665 * Here, we first take an extra reference to each inaccessible
1666 * descriptor. Then, we call sorflush ourself, since we know
1667 * it is a Unix domain socket anyhow. After we destroy all the
1668 * rights carried in messages, we do a last closef to get rid
1669 * of our extra reference. This is the last close, and the
1670 * unp_detach etc will shut down the socket.
1671 *
1672 * 91/09/19, bsy@cs.cmu.edu
1673 */
1674again:
1675 nfiles_snap = openfiles + nfiles_slack; /* some slack */
1676 extra_ref = malloc(nfiles_snap * sizeof(struct file *), M_TEMP,
1677 M_WAITOK);
1678 sx_slock(&filelist_lock);
1679 if (nfiles_snap < openfiles) {
1680 sx_sunlock(&filelist_lock);
1681 free(extra_ref, M_TEMP);
1682 nfiles_slack += 20;
1683 goto again;
1684 }
1685 for (nunref = 0, fp = LIST_FIRST(&filehead), fpp = extra_ref;
1686 fp != NULL; fp = nextfp) {
1687 nextfp = LIST_NEXT(fp, f_list);
1688 FILE_LOCK(fp);
1689 /*
1690 * If it's not open, skip it
1691 */
1692 if (fp->f_count == 0) {
1693 FILE_UNLOCK(fp);
1694 continue;
1695 }
1696 /*
1697 * If all refs are from msgs, and it's not marked accessible
1698 * then it must be referenced from some unreachable cycle
1699 * of (shut-down) FDs, so include it in our
1700 * list of FDs to remove
1701 */
1702 if (fp->f_count == fp->f_msgcount && !(fp->f_gcflag & FMARK)) {
1703 *fpp++ = fp;
1704 nunref++;
1705 fp->f_count++;
1706 }
1707 FILE_UNLOCK(fp);
1708 }
1709 sx_sunlock(&filelist_lock);
1710 /*
1711 * for each FD on our hit list, do the following two things
1712 */
1713 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) {
1714 struct file *tfp = *fpp;
1715 FILE_LOCK(tfp);
1716 if (tfp->f_type == DTYPE_SOCKET &&
1717 tfp->f_data != NULL) {
1718 FILE_UNLOCK(tfp);
1719 sorflush(tfp->f_data);
1720 } else {
1721 FILE_UNLOCK(tfp);
1722 }
1723 }
1724 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) {
1725 closef(*fpp, (struct thread *) NULL);
1726 unp_recycled++;
1727 }
1728 free(extra_ref, M_TEMP);
1729}
1730
1731void
1732unp_dispose(struct mbuf *m)
1733{
1734
1735 if (m)
1736 unp_scan(m, unp_discard);
1737}
1738
1739static int
1740unp_listen(struct socket *so, struct unpcb *unp, int backlog,
1741 struct thread *td)
1742{
1743 int error;
1744
1745 UNP_LOCK_ASSERT();
1746
1747 SOCK_LOCK(so);
1748 error = solisten_proto_check(so);
1749 if (error == 0) {
1750 cru2x(td->td_ucred, &unp->unp_peercred);
1751 unp->unp_flags |= UNP_HAVEPCCACHED;
1752 solisten_proto(so, backlog);
1753 }
1754 SOCK_UNLOCK(so);
1755 return (error);
1756}
1757
1758static void
1759unp_scan(struct mbuf *m0, void (*op)(struct file *))
1760{
1761 struct mbuf *m;
1762 struct file **rp;
1763 struct cmsghdr *cm;
1764 void *data;
1765 int i;
1766 socklen_t clen, datalen;
1767 int qfds;
1768
1769 while (m0 != NULL) {
1770 for (m = m0; m; m = m->m_next) {
1771 if (m->m_type != MT_CONTROL)
1772 continue;
1773
1774 cm = mtod(m, struct cmsghdr *);
1775 clen = m->m_len;
1776
1777 while (cm != NULL) {
1778 if (sizeof(*cm) > clen || cm->cmsg_len > clen)
1779 break;
1780
1781 data = CMSG_DATA(cm);
1782 datalen = (caddr_t)cm + cm->cmsg_len
1783 - (caddr_t)data;
1784
1785 if (cm->cmsg_level == SOL_SOCKET &&
1786 cm->cmsg_type == SCM_RIGHTS) {
1787 qfds = datalen / sizeof (struct file *);
1788 rp = data;
1789 for (i = 0; i < qfds; i++)
1790 (*op)(*rp++);
1791 }
1792
1793 if (CMSG_SPACE(datalen) < clen) {
1794 clen -= CMSG_SPACE(datalen);
1795 cm = (struct cmsghdr *)
1796 ((caddr_t)cm + CMSG_SPACE(datalen));
1797 } else {
1798 clen = 0;
1799 cm = NULL;
1800 }
1801 }
1802 }
1803 m0 = m0->m_act;
1804 }
1805}
1806
1807static void
1808unp_mark(struct file *fp)
1809{
1810 if (fp->f_gcflag & FMARK)
1811 return;
1812 unp_defer++;
1813 fp->f_gcflag |= (FMARK|FDEFER);
1814}
1815
1816static void
1817unp_discard(struct file *fp)
1818{
1819 UNP_LOCK();
1820 FILE_LOCK(fp);
1821 fp->f_msgcount--;
1822 unp_rights--;
1823 FILE_UNLOCK(fp);
1824 UNP_UNLOCK();
1825 (void) closef(fp, (struct thread *)NULL);
1826}