1/* $NetBSD: uipc_usrreq.c,v 1.181 2016/10/31 15:05:05 maxv Exp $ */
2
3/*-
4 * Copyright (c) 1998, 2000, 2004, 2008, 2009 The NetBSD Foundation, Inc.
5 * All rights reserved.
6 *
7 * This code is derived from software contributed to The NetBSD Foundation
8 * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility,
9 * NASA Ames Research Center, and by Andrew Doran.
10 *
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
13 * are met:
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 *
20 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
22 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
23 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
24 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
27 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
29 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
30 * POSSIBILITY OF SUCH DAMAGE.
31 */
32
33/*
34 * Copyright (c) 1982, 1986, 1989, 1991, 1993
35 * The Regents of the University of California. All rights reserved.
36 *
37 * Redistribution and use in source and binary forms, with or without
38 * modification, are permitted provided that the following conditions
39 * are met:
40 * 1. Redistributions of source code must retain the above copyright
41 * notice, this list of conditions and the following disclaimer.
42 * 2. Redistributions in binary form must reproduce the above copyright
43 * notice, this list of conditions and the following disclaimer in the
44 * documentation and/or other materials provided with the distribution.
45 * 3. Neither the name of the University nor the names of its contributors
46 * may be used to endorse or promote products derived from this software
47 * without specific prior written permission.
48 *
49 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
50 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
51 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
52 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
53 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
54 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
55 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
56 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
57 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
58 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
59 * SUCH DAMAGE.
60 *
61 * @(#)uipc_usrreq.c 8.9 (Berkeley) 5/14/95
62 */
63
64/*
65 * Copyright (c) 1997 Christopher G. Demetriou. All rights reserved.
66 *
67 * Redistribution and use in source and binary forms, with or without
68 * modification, are permitted provided that the following conditions
69 * are met:
70 * 1. Redistributions of source code must retain the above copyright
71 * notice, this list of conditions and the following disclaimer.
72 * 2. Redistributions in binary form must reproduce the above copyright
73 * notice, this list of conditions and the following disclaimer in the
74 * documentation and/or other materials provided with the distribution.
75 * 3. All advertising materials mentioning features or use of this software
76 * must display the following acknowledgement:
77 * This product includes software developed by the University of
78 * California, Berkeley and its contributors.
79 * 4. Neither the name of the University nor the names of its contributors
80 * may be used to endorse or promote products derived from this software
81 * without specific prior written permission.
82 *
83 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
84 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
85 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
86 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
87 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
88 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
89 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
90 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
91 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
92 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
93 * SUCH DAMAGE.
94 *
95 * @(#)uipc_usrreq.c 8.9 (Berkeley) 5/14/95
96 */
97
98#include <sys/cdefs.h>
99__KERNEL_RCSID(0, "$NetBSD: uipc_usrreq.c,v 1.181 2016/10/31 15:05:05 maxv Exp $");
100
101#include <sys/param.h>
102#include <sys/systm.h>
103#include <sys/proc.h>
104#include <sys/filedesc.h>
105#include <sys/domain.h>
106#include <sys/protosw.h>
107#include <sys/socket.h>
108#include <sys/socketvar.h>
109#include <sys/unpcb.h>
110#include <sys/un.h>
111#include <sys/namei.h>
112#include <sys/vnode.h>
113#include <sys/file.h>
114#include <sys/stat.h>
115#include <sys/mbuf.h>
116#include <sys/kauth.h>
117#include <sys/kmem.h>
118#include <sys/atomic.h>
119#include <sys/uidinfo.h>
120#include <sys/kernel.h>
121#include <sys/kthread.h>
122
123#ifdef COMPAT_70
124#include <compat/sys/socket.h>
125#endif
126
127/*
128 * Unix communications domain.
129 *
130 * TODO:
131 * RDM
132 * rethink name space problems
133 * need a proper out-of-band
134 *
135 * Notes on locking:
136 *
137 * The generic rules noted in uipc_socket2.c apply. In addition:
138 *
139 * o We have a global lock, uipc_lock.
140 *
141 * o All datagram sockets are locked by uipc_lock.
142 *
143 * o For stream socketpairs, the two endpoints are created sharing the same
144 * independent lock. Sockets presented to PRU_CONNECT2 must already have
145 * matching locks.
146 *
147 * o Stream sockets created via socket() start life with their own
148 * independent lock.
149 *
150 * o Stream connections to a named endpoint are slightly more complicated.
151 * Sockets that have called listen() have their lock pointer mutated to
152 * the global uipc_lock. When establishing a connection, the connecting
153 * socket also has its lock mutated to uipc_lock, which matches the head
154 * (listening socket). We create a new socket for accept() to return, and
155 * that also shares the head's lock. Until the connection is completely
156 * done on both ends, all three sockets are locked by uipc_lock. Once the
157 * connection is complete, the association with the head's lock is broken.
158 * The connecting socket and the socket returned from accept() have their
159 * lock pointers mutated away from uipc_lock, and back to the connecting
160 * socket's original, independent lock. The head continues to be locked
161 * by uipc_lock.
162 *
163 * o If uipc_lock is determined to be a significant source of contention,
164 * it could easily be hashed out. It is difficult to simply make it an
165 * independent lock because of visibility / garbage collection issues:
166 * if a socket has been associated with a lock at any point, that lock
167 * must remain valid until the socket is no longer visible in the system.
168 * The lock must not be freed or otherwise destroyed until any sockets
169 * that had referenced it have also been destroyed.
170 */
171const struct sockaddr_un sun_noname = {
172 .sun_len = offsetof(struct sockaddr_un, sun_path),
173 .sun_family = AF_LOCAL,
174};
175ino_t unp_ino; /* prototype for fake inode numbers */
176
177static struct mbuf * unp_addsockcred(struct lwp *, struct mbuf *);
178static void unp_discard_later(file_t *);
179static void unp_discard_now(file_t *);
180static void unp_disconnect1(struct unpcb *);
181static bool unp_drop(struct unpcb *, int);
182static int unp_internalize(struct mbuf **);
183static void unp_mark(file_t *);
184static void unp_scan(struct mbuf *, void (*)(file_t *), int);
185static void unp_shutdown1(struct unpcb *);
186static void unp_thread(void *);
187static void unp_thread_kick(void);
188
189static kmutex_t *uipc_lock;
190
191static kcondvar_t unp_thread_cv;
192static lwp_t *unp_thread_lwp;
193static SLIST_HEAD(,file) unp_thread_discard;
194static int unp_defer;
195
196/*
197 * Initialize Unix protocols.
198 */
199void
200uipc_init(void)
201{
202 int error;
203
204 uipc_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
205 cv_init(&unp_thread_cv, "unpgc");
206
207 error = kthread_create(PRI_NONE, KTHREAD_MPSAFE, NULL, unp_thread,
208 NULL, &unp_thread_lwp, "unpgc");
209 if (error != 0)
210 panic("uipc_init %d", error);
211}
212
213/*
214 * A connection succeeded: disassociate both endpoints from the head's
215 * lock, and make them share their own lock. There is a race here: for
216 * a very brief time one endpoint will be locked by a different lock
217 * than the other end. However, since the current thread holds the old
218 * lock (the listening socket's lock, the head) access can still only be
219 * made to one side of the connection.
220 */
221static void
222unp_setpeerlocks(struct socket *so, struct socket *so2)
223{
224 struct unpcb *unp;
225 kmutex_t *lock;
226
227 KASSERT(solocked2(so, so2));
228
229 /*
230 * Bail out if either end of the socket is not yet fully
231 * connected or accepted. We only break the lock association
232 * with the head when the pair of sockets stand completely
233 * on their own.
234 */
235 KASSERT(so->so_head == NULL);
236 if (so2->so_head != NULL)
237 return;
238
239 /*
240 * Drop references to old lock. A third reference (from the
241 * queue head) must be held as we still hold its lock. Bonus:
242 * we don't need to worry about garbage collecting the lock.
243 */
244 lock = so->so_lock;
245 KASSERT(lock == uipc_lock);
246 mutex_obj_free(lock);
247 mutex_obj_free(lock);
248
249 /*
250 * Grab stream lock from the initiator and share between the two
251 * endpoints. Issue memory barrier to ensure all modifications
252 * become globally visible before the lock change. so2 is
253 * assumed not to have a stream lock, because it was created
254 * purely for the server side to accept this connection and
255 * started out life using the domain-wide lock.
256 */
257 unp = sotounpcb(so);
258 KASSERT(unp->unp_streamlock != NULL);
259 KASSERT(sotounpcb(so2)->unp_streamlock == NULL);
260 lock = unp->unp_streamlock;
261 unp->unp_streamlock = NULL;
262 mutex_obj_hold(lock);
263 membar_exit();
264 /*
265 * possible race if lock is not held - see comment in
266 * uipc_usrreq(PRU_ACCEPT).
267 */
268 KASSERT(mutex_owned(lock));
269 solockreset(so, lock);
270 solockreset(so2, lock);
271}
272
273/*
274 * Reset a socket's lock back to the domain-wide lock.
275 */
276static void
277unp_resetlock(struct socket *so)
278{
279 kmutex_t *olock, *nlock;
280 struct unpcb *unp;
281
282 KASSERT(solocked(so));
283
284 olock = so->so_lock;
285 nlock = uipc_lock;
286 if (olock == nlock)
287 return;
288 unp = sotounpcb(so);
289 KASSERT(unp->unp_streamlock == NULL);
290 unp->unp_streamlock = olock;
291 mutex_obj_hold(nlock);
292 mutex_enter(nlock);
293 solockreset(so, nlock);
294 mutex_exit(olock);
295}
296
297static void
298unp_free(struct unpcb *unp)
299{
300 if (unp->unp_addr)
301 free(unp->unp_addr, M_SONAME);
302 if (unp->unp_streamlock != NULL)
303 mutex_obj_free(unp->unp_streamlock);
304 kmem_free(unp, sizeof(*unp));
305}
306
307static int
308unp_output(struct mbuf *m, struct mbuf *control, struct unpcb *unp)
309{
310 struct socket *so2;
311 const struct sockaddr_un *sun;
312
313 /* XXX: server side closed the socket */
314 if (unp->unp_conn == NULL)
315 return ECONNREFUSED;
316 so2 = unp->unp_conn->unp_socket;
317
318 KASSERT(solocked(so2));
319
320 if (unp->unp_addr)
321 sun = unp->unp_addr;
322 else
323 sun = &sun_noname;
324 if (unp->unp_conn->unp_flags & UNP_WANTCRED)
325 control = unp_addsockcred(curlwp, control);
326#ifdef COMPAT_SOCKCRED70
327 if (unp->unp_conn->unp_flags & UNP_OWANTCRED)
328 control = compat_70_unp_addsockcred(curlwp, control);
329#endif
330 if (sbappendaddr(&so2->so_rcv, (const struct sockaddr *)sun, m,
331 control) == 0) {
332 so2->so_rcv.sb_overflowed++;
333 unp_dispose(control);
334 m_freem(control);
335 m_freem(m);
336 return (ENOBUFS);
337 } else {
338 sorwakeup(so2);
339 return (0);
340 }
341}
342
343static void
344unp_setaddr(struct socket *so, struct sockaddr *nam, bool peeraddr)
345{
346 const struct sockaddr_un *sun = NULL;
347 struct unpcb *unp;
348
349 KASSERT(solocked(so));
350 unp = sotounpcb(so);
351
352 if (peeraddr) {
353 if (unp->unp_conn && unp->unp_conn->unp_addr)
354 sun = unp->unp_conn->unp_addr;
355 } else {
356 if (unp->unp_addr)
357 sun = unp->unp_addr;
358 }
359 if (sun == NULL)
360 sun = &sun_noname;
361
362 memcpy(nam, sun, sun->sun_len);
363}
364
365static int
366unp_rcvd(struct socket *so, int flags, struct lwp *l)
367{
368 struct unpcb *unp = sotounpcb(so);
369 struct socket *so2;
370 u_int newhiwat;
371
372 KASSERT(solocked(so));
373 KASSERT(unp != NULL);
374
375 switch (so->so_type) {
376
377 case SOCK_DGRAM:
378 panic("uipc 1");
379 /*NOTREACHED*/
380
381 case SOCK_SEQPACKET: /* FALLTHROUGH */
382 case SOCK_STREAM:
383#define rcv (&so->so_rcv)
384#define snd (&so2->so_snd)
385 if (unp->unp_conn == 0)
386 break;
387 so2 = unp->unp_conn->unp_socket;
388 KASSERT(solocked2(so, so2));
389 /*
390 * Adjust backpressure on sender
391 * and wakeup any waiting to write.
392 */
393 snd->sb_mbmax += unp->unp_mbcnt - rcv->sb_mbcnt;
394 unp->unp_mbcnt = rcv->sb_mbcnt;
395 newhiwat = snd->sb_hiwat + unp->unp_cc - rcv->sb_cc;
396 (void)chgsbsize(so2->so_uidinfo,
397 &snd->sb_hiwat, newhiwat, RLIM_INFINITY);
398 unp->unp_cc = rcv->sb_cc;
399 sowwakeup(so2);
400#undef snd
401#undef rcv
402 break;
403
404 default:
405 panic("uipc 2");
406 }
407
408 return 0;
409}
410
411static int
412unp_recvoob(struct socket *so, struct mbuf *m, int flags)
413{
414 KASSERT(solocked(so));
415
416 return EOPNOTSUPP;
417}
418
419static int
420unp_send(struct socket *so, struct mbuf *m, struct sockaddr *nam,
421 struct mbuf *control, struct lwp *l)
422{
423 struct unpcb *unp = sotounpcb(so);
424 int error = 0;
425 u_int newhiwat;
426 struct socket *so2;
427
428 KASSERT(solocked(so));
429 KASSERT(unp != NULL);
430 KASSERT(m != NULL);
431
432 /*
433 * Note: unp_internalize() rejects any control message
434 * other than SCM_RIGHTS, and only allows one. This
435 * has the side-effect of preventing a caller from
436 * forging SCM_CREDS.
437 */
438 if (control) {
439 sounlock(so);
440 error = unp_internalize(&control);
441 solock(so);
442 if (error != 0) {
443 m_freem(control);
444 m_freem(m);
445 return error;
446 }
447 }
448
449 switch (so->so_type) {
450
451 case SOCK_DGRAM: {
452 KASSERT(so->so_lock == uipc_lock);
453 if (nam) {
454 if ((so->so_state & SS_ISCONNECTED) != 0)
455 error = EISCONN;
456 else {
457 /*
458 * Note: once connected, the
459 * socket's lock must not be
460 * dropped until we have sent
461 * the message and disconnected.
462 * This is necessary to prevent
463 * intervening control ops, like
464 * another connection.
465 */
466 error = unp_connect(so, nam, l);
467 }
468 } else {
469 if ((so->so_state & SS_ISCONNECTED) == 0)
470 error = ENOTCONN;
471 }
472 if (error) {
473 unp_dispose(control);
474 m_freem(control);
475 m_freem(m);
476 return error;
477 }
478 error = unp_output(m, control, unp);
479 if (nam)
480 unp_disconnect1(unp);
481 break;
482 }
483
484 case SOCK_SEQPACKET: /* FALLTHROUGH */
485 case SOCK_STREAM:
486#define rcv (&so2->so_rcv)
487#define snd (&so->so_snd)
488 if (unp->unp_conn == NULL) {
489 error = ENOTCONN;
490 break;
491 }
492 so2 = unp->unp_conn->unp_socket;
493 KASSERT(solocked2(so, so2));
494 if (unp->unp_conn->unp_flags & UNP_WANTCRED) {
495 /*
496 * Credentials are passed only once on
497 * SOCK_STREAM and SOCK_SEQPACKET.
498 */
499 unp->unp_conn->unp_flags &= ~UNP_WANTCRED;
500 control = unp_addsockcred(l, control);
501 }
502#ifdef COMPAT_SOCKCRED70
503 if (unp->unp_conn->unp_flags & UNP_OWANTCRED) {
504 /*
505 * Credentials are passed only once on
506 * SOCK_STREAM and SOCK_SEQPACKET.
507 */
508 unp->unp_conn->unp_flags &= ~UNP_OWANTCRED;
509 control = compat_70_unp_addsockcred(l, control);
510 }
511#endif
512 /*
513 * Send to paired receive port, and then reduce
514 * send buffer hiwater marks to maintain backpressure.
515 * Wake up readers.
516 */
517 if (control) {
518 if (sbappendcontrol(rcv, m, control) != 0)
519 control = NULL;
520 } else {
521 switch(so->so_type) {
522 case SOCK_SEQPACKET:
523 sbappendrecord(rcv, m);
524 break;
525 case SOCK_STREAM:
526 sbappend(rcv, m);
527 break;
528 default:
529 panic("uipc_usrreq");
530 break;
531 }
532 }
533 snd->sb_mbmax -=
534 rcv->sb_mbcnt - unp->unp_conn->unp_mbcnt;
535 unp->unp_conn->unp_mbcnt = rcv->sb_mbcnt;
536 newhiwat = snd->sb_hiwat -
537 (rcv->sb_cc - unp->unp_conn->unp_cc);
538 (void)chgsbsize(so->so_uidinfo,
539 &snd->sb_hiwat, newhiwat, RLIM_INFINITY);
540 unp->unp_conn->unp_cc = rcv->sb_cc;
541 sorwakeup(so2);
542#undef snd
543#undef rcv
544 if (control != NULL) {
545 unp_dispose(control);
546 m_freem(control);
547 }
548 break;
549
550 default:
551 panic("uipc 4");
552 }
553
554 return error;
555}
556
557static int
558unp_sendoob(struct socket *so, struct mbuf *m, struct mbuf * control)
559{
560 KASSERT(solocked(so));
561
562 m_freem(m);
563 m_freem(control);
564
565 return EOPNOTSUPP;
566}
567
568/*
569 * Unix domain socket option processing.
570 */
571int
572uipc_ctloutput(int op, struct socket *so, struct sockopt *sopt)
573{
574 struct unpcb *unp = sotounpcb(so);
575 int optval = 0, error = 0;
576
577 KASSERT(solocked(so));
578
579 if (sopt->sopt_level != 0) {
580 error = ENOPROTOOPT;
581 } else switch (op) {
582
583 case PRCO_SETOPT:
584 switch (sopt->sopt_name) {
585 case LOCAL_CREDS:
586 case LOCAL_CONNWAIT:
587#ifdef COMPAT_SOCKCRED70
588 case LOCAL_OCREDS:
589#endif
590 error = sockopt_getint(sopt, &optval);
591 if (error)
592 break;
593 switch (sopt->sopt_name) {
594#define OPTSET(bit) \
595 if (optval) \
596 unp->unp_flags |= (bit); \
597 else \
598 unp->unp_flags &= ~(bit);
599
600 case LOCAL_CREDS:
601 OPTSET(UNP_WANTCRED);
602 break;
603 case LOCAL_CONNWAIT:
604 OPTSET(UNP_CONNWAIT);
605 break;
606#ifdef COMPAT_SOCKCRED70
607 case LOCAL_OCREDS:
608 OPTSET(UNP_OWANTCRED);
609 break;
610#endif
611 }
612 break;
613#undef OPTSET
614
615 default:
616 error = ENOPROTOOPT;
617 break;
618 }
619 break;
620
621 case PRCO_GETOPT:
622 sounlock(so);
623 switch (sopt->sopt_name) {
624 case LOCAL_PEEREID:
625 if (unp->unp_flags & UNP_EIDSVALID) {
626 error = sockopt_set(sopt,
627 &unp->unp_connid, sizeof(unp->unp_connid));
628 } else {
629 error = EINVAL;
630 }
631 break;
632 case LOCAL_CREDS:
633#define OPTBIT(bit) (unp->unp_flags & (bit) ? 1 : 0)
634
635 optval = OPTBIT(UNP_WANTCRED);
636 error = sockopt_setint(sopt, optval);
637 break;
638#ifdef COMPAT_SOCKCRED70
639 case LOCAL_OCREDS:
640 optval = OPTBIT(UNP_OWANTCRED);
641 error = sockopt_setint(sopt, optval);
642 break;
643#endif
644#undef OPTBIT
645
646 default:
647 error = ENOPROTOOPT;
648 break;
649 }
650 solock(so);
651 break;
652 }
653 return (error);
654}
655
656/*
657 * Both send and receive buffers are allocated PIPSIZ bytes of buffering
658 * for stream sockets, although the total for sender and receiver is
659 * actually only PIPSIZ.
660 * Datagram sockets really use the sendspace as the maximum datagram size,
661 * and don't really want to reserve the sendspace. Their recvspace should
662 * be large enough for at least one max-size datagram plus address.
663 */
664#define PIPSIZ 4096
665u_long unpst_sendspace = PIPSIZ;
666u_long unpst_recvspace = PIPSIZ;
667u_long unpdg_sendspace = 2*1024; /* really max datagram size */
668u_long unpdg_recvspace = 4*1024;
669
670u_int unp_rights; /* files in flight */
671u_int unp_rights_ratio = 2; /* limit, fraction of maxfiles */
672
673static int
674unp_attach(struct socket *so, int proto)
675{
676 struct unpcb *unp = sotounpcb(so);
677 u_long sndspc, rcvspc;
678 int error;
679
680 KASSERT(unp == NULL);
681
682 switch (so->so_type) {
683 case SOCK_SEQPACKET:
684 /* FALLTHROUGH */
685 case SOCK_STREAM:
686 if (so->so_lock == NULL) {
687 so->so_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
688 solock(so);
689 }
690 sndspc = unpst_sendspace;
691 rcvspc = unpst_recvspace;
692 break;
693
694 case SOCK_DGRAM:
695 if (so->so_lock == NULL) {
696 mutex_obj_hold(uipc_lock);
697 so->so_lock = uipc_lock;
698 solock(so);
699 }
700 sndspc = unpdg_sendspace;
701 rcvspc = unpdg_recvspace;
702 break;
703
704 default:
705 panic("unp_attach");
706 }
707
708 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
709 error = soreserve(so, sndspc, rcvspc);
710 if (error) {
711 return error;
712 }
713 }
714
715 unp = kmem_zalloc(sizeof(*unp), KM_SLEEP);
716 nanotime(&unp->unp_ctime);
717 unp->unp_socket = so;
718 so->so_pcb = unp;
719
720 KASSERT(solocked(so));
721 return 0;
722}
723
724static void
725unp_detach(struct socket *so)
726{
727 struct unpcb *unp;
728 vnode_t *vp;
729
730 unp = sotounpcb(so);
731 KASSERT(unp != NULL);
732 KASSERT(solocked(so));
733 retry:
734 if ((vp = unp->unp_vnode) != NULL) {
735 sounlock(so);
736 /* Acquire v_interlock to protect against unp_connect(). */
737 /* XXXAD racy */
738 mutex_enter(vp->v_interlock);
739 vp->v_socket = NULL;
740 mutex_exit(vp->v_interlock);
741 vrele(vp);
742 solock(so);
743 unp->unp_vnode = NULL;
744 }
745 if (unp->unp_conn)
746 unp_disconnect1(unp);
747 while (unp->unp_refs) {
748 KASSERT(solocked2(so, unp->unp_refs->unp_socket));
749 if (unp_drop(unp->unp_refs, ECONNRESET)) {
750 solock(so);
751 goto retry;
752 }
753 }
754 soisdisconnected(so);
755 so->so_pcb = NULL;
756 if (unp_rights) {
757 /*
758 * Normally the receive buffer is flushed later, in sofree,
759 * but if our receive buffer holds references to files that
760 * are now garbage, we will enqueue those file references to
761 * the garbage collector and kick it into action.
762 */
763 sorflush(so);
764 unp_free(unp);
765 unp_thread_kick();
766 } else
767 unp_free(unp);
768}
769
770static int
771unp_accept(struct socket *so, struct sockaddr *nam)
772{
773 struct unpcb *unp = sotounpcb(so);
774 struct socket *so2;
775
776 KASSERT(solocked(so));
777 KASSERT(nam != NULL);
778
779 /* XXX code review required to determine if unp can ever be NULL */
780 if (unp == NULL)
781 return EINVAL;
782
783 KASSERT(so->so_lock == uipc_lock);
784 /*
785 * Mark the initiating STREAM socket as connected *ONLY*
786 * after it's been accepted. This prevents a client from
787 * overrunning a server and receiving ECONNREFUSED.
788 */
789 if (unp->unp_conn == NULL) {
790 /*
791 * This will use the empty socket and will not
792 * allocate.
793 */
794 unp_setaddr(so, nam, true);
795 return 0;
796 }
797 so2 = unp->unp_conn->unp_socket;
798 if (so2->so_state & SS_ISCONNECTING) {
799 KASSERT(solocked2(so, so->so_head));
800 KASSERT(solocked2(so2, so->so_head));
801 soisconnected(so2);
802 }
803 /*
804 * If the connection is fully established, break the
805 * association with uipc_lock and give the connected
806 * pair a separate lock to share.
807 * There is a race here: sotounpcb(so2)->unp_streamlock
808 * is not locked, so when changing so2->so_lock
809 * another thread can grab it while so->so_lock is still
810 * pointing to the (locked) uipc_lock.
811 * this should be harmless, except that this makes
812 * solocked2() and solocked() unreliable.
813 * Another problem is that unp_setaddr() expects the
814 * the socket locked. Grabing sotounpcb(so2)->unp_streamlock
815 * fixes both issues.
816 */
817 mutex_enter(sotounpcb(so2)->unp_streamlock);
818 unp_setpeerlocks(so2, so);
819 /*
820 * Only now return peer's address, as we may need to
821 * block in order to allocate memory.
822 *
823 * XXX Minor race: connection can be broken while
824 * lock is dropped in unp_setaddr(). We will return
825 * error == 0 and sun_noname as the peer address.
826 */
827 unp_setaddr(so, nam, true);
828 /* so_lock now points to unp_streamlock */
829 mutex_exit(so2->so_lock);
830 return 0;
831}
832
833static int
834unp_ioctl(struct socket *so, u_long cmd, void *nam, struct ifnet *ifp)
835{
836 return EOPNOTSUPP;
837}
838
839static int
840unp_stat(struct socket *so, struct stat *ub)
841{
842 struct unpcb *unp;
843 struct socket *so2;
844
845 KASSERT(solocked(so));
846
847 unp = sotounpcb(so);
848 if (unp == NULL)
849 return EINVAL;
850
851 ub->st_blksize = so->so_snd.sb_hiwat;
852 switch (so->so_type) {
853 case SOCK_SEQPACKET: /* FALLTHROUGH */
854 case SOCK_STREAM:
855 if (unp->unp_conn == 0)
856 break;
857
858 so2 = unp->unp_conn->unp_socket;
859 KASSERT(solocked2(so, so2));
860 ub->st_blksize += so2->so_rcv.sb_cc;
861 break;
862 default:
863 break;
864 }
865 ub->st_dev = NODEV;
866 if (unp->unp_ino == 0)
867 unp->unp_ino = unp_ino++;
868 ub->st_atimespec = ub->st_mtimespec = ub->st_ctimespec = unp->unp_ctime;
869 ub->st_ino = unp->unp_ino;
870 return (0);
871}
872
873static int
874unp_peeraddr(struct socket *so, struct sockaddr *nam)
875{
876 KASSERT(solocked(so));
877 KASSERT(sotounpcb(so) != NULL);
878 KASSERT(nam != NULL);
879
880 unp_setaddr(so, nam, true);
881 return 0;
882}
883
884static int
885unp_sockaddr(struct socket *so, struct sockaddr *nam)
886{
887 KASSERT(solocked(so));
888 KASSERT(sotounpcb(so) != NULL);
889 KASSERT(nam != NULL);
890
891 unp_setaddr(so, nam, false);
892 return 0;
893}
894
895/*
896 * we only need to perform this allocation until syscalls other than
897 * bind are adjusted to use sockaddr_big.
898 */
899static struct sockaddr_un *
900makeun_sb(struct sockaddr *nam, size_t *addrlen)
901{
902 struct sockaddr_un *sun;
903
904 *addrlen = nam->sa_len + 1;
905 sun = malloc(*addrlen, M_SONAME, M_WAITOK);
906 memcpy(sun, nam, nam->sa_len);
907 *(((char *)sun) + nam->sa_len) = '\0';
908 return sun;
909}
910
911static int
912unp_bind(struct socket *so, struct sockaddr *nam, struct lwp *l)
913{
914 struct sockaddr_un *sun;
915 struct unpcb *unp;
916 vnode_t *vp;
917 struct vattr vattr;
918 size_t addrlen;
919 int error;
920 struct pathbuf *pb;
921 struct nameidata nd;
922 proc_t *p;
923
924 unp = sotounpcb(so);
925
926 KASSERT(solocked(so));
927 KASSERT(unp != NULL);
928 KASSERT(nam != NULL);
929
930 if (unp->unp_vnode != NULL)
931 return (EINVAL);
932 if ((unp->unp_flags & UNP_BUSY) != 0) {
933 /*
934 * EALREADY may not be strictly accurate, but since this
935 * is a major application error it's hardly a big deal.
936 */
937 return (EALREADY);
938 }
939 unp->unp_flags |= UNP_BUSY;
940 sounlock(so);
941
942 p = l->l_proc;
943 sun = makeun_sb(nam, &addrlen);
944
945 pb = pathbuf_create(sun->sun_path);
946 if (pb == NULL) {
947 error = ENOMEM;
948 goto bad;
949 }
950 NDINIT(&nd, CREATE, FOLLOW | LOCKPARENT | TRYEMULROOT, pb);
951
952/* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
953 if ((error = namei(&nd)) != 0) {
954 pathbuf_destroy(pb);
955 goto bad;
956 }
957 vp = nd.ni_vp;
958 if (vp != NULL) {
959 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
960 if (nd.ni_dvp == vp)
961 vrele(nd.ni_dvp);
962 else
963 vput(nd.ni_dvp);
964 vrele(vp);
965 pathbuf_destroy(pb);
966 error = EADDRINUSE;
967 goto bad;
968 }
969 vattr_null(&vattr);
970 vattr.va_type = VSOCK;
971 vattr.va_mode = ACCESSPERMS & ~(p->p_cwdi->cwdi_cmask);
972 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
973 if (error) {
974 vput(nd.ni_dvp);
975 pathbuf_destroy(pb);
976 goto bad;
977 }
978 vp = nd.ni_vp;
979 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
980 solock(so);
981 vp->v_socket = unp->unp_socket;
982 unp->unp_vnode = vp;
983 unp->unp_addrlen = addrlen;
984 unp->unp_addr = sun;
985 unp->unp_connid.unp_pid = p->p_pid;
986 unp->unp_connid.unp_euid = kauth_cred_geteuid(l->l_cred);
987 unp->unp_connid.unp_egid = kauth_cred_getegid(l->l_cred);
988 unp->unp_flags |= UNP_EIDSBIND;
989 VOP_UNLOCK(vp);
990 vput(nd.ni_dvp);
991 unp->unp_flags &= ~UNP_BUSY;
992 pathbuf_destroy(pb);
993 return (0);
994
995 bad:
996 free(sun, M_SONAME);
997 solock(so);
998 unp->unp_flags &= ~UNP_BUSY;
999 return (error);
1000}
1001
1002static int
1003unp_listen(struct socket *so, struct lwp *l)
1004{
1005 struct unpcb *unp = sotounpcb(so);
1006
1007 KASSERT(solocked(so));
1008 KASSERT(unp != NULL);
1009
1010 /*
1011 * If the socket can accept a connection, it must be
1012 * locked by uipc_lock.
1013 */
1014 unp_resetlock(so);
1015 if (unp->unp_vnode == NULL)
1016 return EINVAL;
1017
1018 return 0;
1019}
1020
1021static int
1022unp_disconnect(struct socket *so)
1023{
1024 KASSERT(solocked(so));
1025 KASSERT(sotounpcb(so) != NULL);
1026
1027 unp_disconnect1(sotounpcb(so));
1028 return 0;
1029}
1030
1031static int
1032unp_shutdown(struct socket *so)
1033{
1034 KASSERT(solocked(so));
1035 KASSERT(sotounpcb(so) != NULL);
1036
1037 socantsendmore(so);
1038 unp_shutdown1(sotounpcb(so));
1039 return 0;
1040}
1041
1042static int
1043unp_abort(struct socket *so)
1044{
1045 KASSERT(solocked(so));
1046 KASSERT(sotounpcb(so) != NULL);
1047
1048 (void)unp_drop(sotounpcb(so), ECONNABORTED);
1049 KASSERT(so->so_head == NULL);
1050 KASSERT(so->so_pcb != NULL);
1051 unp_detach(so);
1052 return 0;
1053}
1054
1055static int
1056unp_connect1(struct socket *so, struct socket *so2, struct lwp *l)
1057{
1058 struct unpcb *unp = sotounpcb(so);
1059 struct unpcb *unp2;
1060
1061 if (so2->so_type != so->so_type)
1062 return EPROTOTYPE;
1063
1064 /*
1065 * All three sockets involved must be locked by same lock:
1066 *
1067 * local endpoint (so)
1068 * remote endpoint (so2)
1069 * queue head (so2->so_head, only if PR_CONNREQUIRED)
1070 */
1071 KASSERT(solocked2(so, so2));
1072 KASSERT(so->so_head == NULL);
1073 if (so2->so_head != NULL) {
1074 KASSERT(so2->so_lock == uipc_lock);
1075 KASSERT(solocked2(so2, so2->so_head));
1076 }
1077
1078 unp2 = sotounpcb(so2);
1079 unp->unp_conn = unp2;
1080
1081 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) != 0) {
1082 unp2->unp_connid.unp_pid = l->l_proc->p_pid;
1083 unp2->unp_connid.unp_euid = kauth_cred_geteuid(l->l_cred);
1084 unp2->unp_connid.unp_egid = kauth_cred_getegid(l->l_cred);
1085 unp2->unp_flags |= UNP_EIDSVALID;
1086 if (unp2->unp_flags & UNP_EIDSBIND) {
1087 unp->unp_connid = unp2->unp_connid;
1088 unp->unp_flags |= UNP_EIDSVALID;
1089 }
1090 }
1091
1092 switch (so->so_type) {
1093
1094 case SOCK_DGRAM:
1095 unp->unp_nextref = unp2->unp_refs;
1096 unp2->unp_refs = unp;
1097 soisconnected(so);
1098 break;
1099
1100 case SOCK_SEQPACKET: /* FALLTHROUGH */
1101 case SOCK_STREAM:
1102
1103 /*
1104 * SOCK_SEQPACKET and SOCK_STREAM cases are handled by callers
1105 * which are unp_connect() or unp_connect2().
1106 */
1107
1108 break;
1109
1110 default:
1111 panic("unp_connect1");
1112 }
1113
1114 return 0;
1115}
1116
1117int
1118unp_connect(struct socket *so, struct sockaddr *nam, struct lwp *l)
1119{
1120 struct sockaddr_un *sun;
1121 vnode_t *vp;
1122 struct socket *so2, *so3;
1123 struct unpcb *unp, *unp2, *unp3;
1124 size_t addrlen;
1125 int error;
1126 struct pathbuf *pb;
1127 struct nameidata nd;
1128
1129 unp = sotounpcb(so);
1130 if ((unp->unp_flags & UNP_BUSY) != 0) {
1131 /*
1132 * EALREADY may not be strictly accurate, but since this
1133 * is a major application error it's hardly a big deal.
1134 */
1135 return (EALREADY);
1136 }
1137 unp->unp_flags |= UNP_BUSY;
1138 sounlock(so);
1139
1140 sun = makeun_sb(nam, &addrlen);
1141 pb = pathbuf_create(sun->sun_path);
1142 if (pb == NULL) {
1143 error = ENOMEM;
1144 goto bad2;
1145 }
1146
1147 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | TRYEMULROOT, pb);
1148
1149 if ((error = namei(&nd)) != 0) {
1150 pathbuf_destroy(pb);
1151 goto bad2;
1152 }
1153 vp = nd.ni_vp;
1154 pathbuf_destroy(pb);
1155 if (vp->v_type != VSOCK) {
1156 error = ENOTSOCK;
1157 goto bad;
1158 }
1159 if ((error = VOP_ACCESS(vp, VWRITE, l->l_cred)) != 0)
1160 goto bad;
1161 /* Acquire v_interlock to protect against unp_detach(). */
1162 mutex_enter(vp->v_interlock);
1163 so2 = vp->v_socket;
1164 if (so2 == NULL) {
1165 mutex_exit(vp->v_interlock);
1166 error = ECONNREFUSED;
1167 goto bad;
1168 }
1169 if (so->so_type != so2->so_type) {
1170 mutex_exit(vp->v_interlock);
1171 error = EPROTOTYPE;
1172 goto bad;
1173 }
1174 solock(so);
1175 unp_resetlock(so);
1176 mutex_exit(vp->v_interlock);
1177 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) != 0) {
1178 /*
1179 * This may seem somewhat fragile but is OK: if we can
1180 * see SO_ACCEPTCONN set on the endpoint, then it must
1181 * be locked by the domain-wide uipc_lock.
1182 */
1183 KASSERT((so2->so_options & SO_ACCEPTCONN) == 0 ||
1184 so2->so_lock == uipc_lock);
1185 if ((so2->so_options & SO_ACCEPTCONN) == 0 ||
1186 (so3 = sonewconn(so2, false)) == NULL) {
1187 error = ECONNREFUSED;
1188 sounlock(so);
1189 goto bad;
1190 }
1191 unp2 = sotounpcb(so2);
1192 unp3 = sotounpcb(so3);
1193 if (unp2->unp_addr) {
1194 unp3->unp_addr = malloc(unp2->unp_addrlen,
1195 M_SONAME, M_WAITOK);
1196 memcpy(unp3->unp_addr, unp2->unp_addr,
1197 unp2->unp_addrlen);
1198 unp3->unp_addrlen = unp2->unp_addrlen;
1199 }
1200 unp3->unp_flags = unp2->unp_flags;
1201 so2 = so3;
1202 }
1203 error = unp_connect1(so, so2, l);
1204 if (error) {
1205 sounlock(so);
1206 goto bad;
1207 }
1208 unp2 = sotounpcb(so2);
1209 switch (so->so_type) {
1210
1211 /*
1212 * SOCK_DGRAM and default cases are handled in prior call to
1213 * unp_connect1(), do not add a default case without fixing
1214 * unp_connect1().
1215 */
1216
1217 case SOCK_SEQPACKET: /* FALLTHROUGH */
1218 case SOCK_STREAM:
1219 unp2->unp_conn = unp;
1220 if ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT)
1221 soisconnecting(so);
1222 else
1223 soisconnected(so);
1224 soisconnected(so2);
1225 /*
1226 * If the connection is fully established, break the
1227 * association with uipc_lock and give the connected
1228 * pair a seperate lock to share.
1229 */
1230 KASSERT(so2->so_head != NULL);
1231 unp_setpeerlocks(so, so2);
1232 break;
1233
1234 }
1235 sounlock(so);
1236 bad:
1237 vput(vp);
1238 bad2:
1239 free(sun, M_SONAME);
1240 solock(so);
1241 unp->unp_flags &= ~UNP_BUSY;
1242 return (error);
1243}
1244
1245int
1246unp_connect2(struct socket *so, struct socket *so2)
1247{
1248 struct unpcb *unp = sotounpcb(so);
1249 struct unpcb *unp2;
1250 int error = 0;
1251
1252 KASSERT(solocked2(so, so2));
1253
1254 error = unp_connect1(so, so2, curlwp);
1255 if (error)
1256 return error;
1257
1258 unp2 = sotounpcb(so2);
1259 switch (so->so_type) {
1260
1261 /*
1262 * SOCK_DGRAM and default cases are handled in prior call to
1263 * unp_connect1(), do not add a default case without fixing
1264 * unp_connect1().
1265 */
1266
1267 case SOCK_SEQPACKET: /* FALLTHROUGH */
1268 case SOCK_STREAM:
1269 unp2->unp_conn = unp;
1270 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) != 0) {
1271 unp->unp_connid = unp2->unp_connid;
1272 unp->unp_flags |= UNP_EIDSVALID;
1273 }
1274 soisconnected(so);
1275 soisconnected(so2);
1276 break;
1277
1278 }
1279 return error;
1280}
1281
1282static void
1283unp_disconnect1(struct unpcb *unp)
1284{
1285 struct unpcb *unp2 = unp->unp_conn;
1286 struct socket *so;
1287
1288 if (unp2 == 0)
1289 return;
1290 unp->unp_conn = 0;
1291 so = unp->unp_socket;
1292 switch (so->so_type) {
1293 case SOCK_DGRAM:
1294 if (unp2->unp_refs == unp)
1295 unp2->unp_refs = unp->unp_nextref;
1296 else {
1297 unp2 = unp2->unp_refs;
1298 for (;;) {
1299 KASSERT(solocked2(so, unp2->unp_socket));
1300 if (unp2 == 0)
1301 panic("unp_disconnect1");
1302 if (unp2->unp_nextref == unp)
1303 break;
1304 unp2 = unp2->unp_nextref;
1305 }
1306 unp2->unp_nextref = unp->unp_nextref;
1307 }
1308 unp->unp_nextref = 0;
1309 so->so_state &= ~SS_ISCONNECTED;
1310 break;
1311
1312 case SOCK_SEQPACKET: /* FALLTHROUGH */
1313 case SOCK_STREAM:
1314 KASSERT(solocked2(so, unp2->unp_socket));
1315 soisdisconnected(so);
1316 unp2->unp_conn = 0;
1317 soisdisconnected(unp2->unp_socket);
1318 break;
1319 }
1320}
1321
1322static void
1323unp_shutdown1(struct unpcb *unp)
1324{
1325 struct socket *so;
1326
1327 switch(unp->unp_socket->so_type) {
1328 case SOCK_SEQPACKET: /* FALLTHROUGH */
1329 case SOCK_STREAM:
1330 if (unp->unp_conn && (so = unp->unp_conn->unp_socket))
1331 socantrcvmore(so);
1332 break;
1333 default:
1334 break;
1335 }
1336}
1337
1338static bool
1339unp_drop(struct unpcb *unp, int errno)
1340{
1341 struct socket *so = unp->unp_socket;
1342
1343 KASSERT(solocked(so));
1344
1345 so->so_error = errno;
1346 unp_disconnect1(unp);
1347 if (so->so_head) {
1348 so->so_pcb = NULL;
1349 /* sofree() drops the socket lock */
1350 sofree(so);
1351 unp_free(unp);
1352 return true;
1353 }
1354 return false;
1355}
1356
1357#ifdef notdef
1358unp_drain(void)
1359{
1360
1361}
1362#endif
1363
1364int
1365unp_externalize(struct mbuf *rights, struct lwp *l, int flags)
1366{
1367 struct cmsghdr * const cm = mtod(rights, struct cmsghdr *);
1368 struct proc * const p = l->l_proc;
1369 file_t **rp;
1370 int error = 0;
1371
1372 const size_t nfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm))) /
1373 sizeof(file_t *);
1374 if (nfds == 0)
1375 goto noop;
1376
1377 int * const fdp = kmem_alloc(nfds * sizeof(int), KM_SLEEP);
1378 rw_enter(&p->p_cwdi->cwdi_lock, RW_READER);
1379
1380 /* Make sure the recipient should be able to see the files.. */
1381 rp = (file_t **)CMSG_DATA(cm);
1382 for (size_t i = 0; i < nfds; i++) {
1383 file_t * const fp = *rp++;
1384 if (fp == NULL) {
1385 error = EINVAL;
1386 goto out;
1387 }
1388 /*
1389 * If we are in a chroot'ed directory, and
1390 * someone wants to pass us a directory, make
1391 * sure it's inside the subtree we're allowed
1392 * to access.
1393 */
1394 if (p->p_cwdi->cwdi_rdir != NULL && fp->f_type == DTYPE_VNODE) {
1395 vnode_t *vp = fp->f_vnode;
1396 if ((vp->v_type == VDIR) &&
1397 !vn_isunder(vp, p->p_cwdi->cwdi_rdir, l)) {
1398 error = EPERM;
1399 goto out;
1400 }
1401 }
1402 }
1403
1404 restart:
1405 /*
1406 * First loop -- allocate file descriptor table slots for the
1407 * new files.
1408 */
1409 for (size_t i = 0; i < nfds; i++) {
1410 if ((error = fd_alloc(p, 0, &fdp[i])) != 0) {
1411 /*
1412 * Back out what we've done so far.
1413 */
1414 while (i-- > 0) {
1415 fd_abort(p, NULL, fdp[i]);
1416 }
1417 if (error == ENOSPC) {
1418 fd_tryexpand(p);
1419 error = 0;
1420 goto restart;
1421 }
1422 /*
1423 * This is the error that has historically
1424 * been returned, and some callers may
1425 * expect it.
1426 */
1427 error = EMSGSIZE;
1428 goto out;
1429 }
1430 }
1431
1432 /*
1433 * Now that adding them has succeeded, update all of the
1434 * file passing state and affix the descriptors.
1435 */
1436 rp = (file_t **)CMSG_DATA(cm);
1437 int *ofdp = (int *)CMSG_DATA(cm);
1438 for (size_t i = 0; i < nfds; i++) {
1439 file_t * const fp = *rp++;
1440 const int fd = fdp[i];
1441 atomic_dec_uint(&unp_rights);
1442 fd_set_exclose(l, fd, (flags & O_CLOEXEC) != 0);
1443 fd_affix(p, fp, fd);
1444 /*
1445 * Done with this file pointer, replace it with a fd;
1446 */
1447 *ofdp++ = fd;
1448 mutex_enter(&fp->f_lock);
1449 fp->f_msgcount--;
1450 mutex_exit(&fp->f_lock);
1451 /*
1452 * Note that fd_affix() adds a reference to the file.
1453 * The file may already have been closed by another
1454 * LWP in the process, so we must drop the reference
1455 * added by unp_internalize() with closef().
1456 */
1457 closef(fp);
1458 }
1459
1460 /*
1461 * Adjust length, in case of transition from large file_t
1462 * pointers to ints.
1463 */
1464 if (sizeof(file_t *) != sizeof(int)) {
1465 cm->cmsg_len = CMSG_LEN(nfds * sizeof(int));
1466 rights->m_len = CMSG_SPACE(nfds * sizeof(int));
1467 }
1468 out:
1469 if (__predict_false(error != 0)) {
1470 file_t **const fpp = (file_t **)CMSG_DATA(cm);
1471 for (size_t i = 0; i < nfds; i++)
1472 unp_discard_now(fpp[i]);
1473 /*
1474 * Truncate the array so that nobody will try to interpret
1475 * what is now garbage in it.
1476 */
1477 cm->cmsg_len = CMSG_LEN(0);
1478 rights->m_len = CMSG_SPACE(0);
1479 }
1480 rw_exit(&p->p_cwdi->cwdi_lock);
1481 kmem_free(fdp, nfds * sizeof(int));
1482
1483 noop:
1484 /*
1485 * Don't disclose kernel memory in the alignment space.
1486 */
1487 KASSERT(cm->cmsg_len <= rights->m_len);
1488 memset(&mtod(rights, char *)[cm->cmsg_len], 0, rights->m_len -
1489 cm->cmsg_len);
1490 return error;
1491}
1492
1493static int
1494unp_internalize(struct mbuf **controlp)
1495{
1496 filedesc_t *fdescp = curlwp->l_fd;
1497 struct mbuf *control = *controlp;
1498 struct cmsghdr *newcm, *cm = mtod(control, struct cmsghdr *);
1499 file_t **rp, **files;
1500 file_t *fp;
1501 int i, fd, *fdp;
1502 int nfds, error;
1503 u_int maxmsg;
1504
1505 error = 0;
1506 newcm = NULL;
1507
1508 /* Sanity check the control message header. */
1509 if (cm->cmsg_type != SCM_RIGHTS || cm->cmsg_level != SOL_SOCKET ||
1510 cm->cmsg_len > control->m_len ||
1511 cm->cmsg_len < CMSG_ALIGN(sizeof(*cm)))
1512 return (EINVAL);
1513
1514 /*
1515 * Verify that the file descriptors are valid, and acquire
1516 * a reference to each.
1517 */
1518 nfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm))) / sizeof(int);
1519 fdp = (int *)CMSG_DATA(cm);
1520 maxmsg = maxfiles / unp_rights_ratio;
1521 for (i = 0; i < nfds; i++) {
1522 fd = *fdp++;
1523 if (atomic_inc_uint_nv(&unp_rights) > maxmsg) {
1524 atomic_dec_uint(&unp_rights);
1525 nfds = i;
1526 error = EAGAIN;
1527 goto out;
1528 }
1529 if ((fp = fd_getfile(fd)) == NULL
1530 || fp->f_type == DTYPE_KQUEUE) {
1531 if (fp)
1532 fd_putfile(fd);
1533 atomic_dec_uint(&unp_rights);
1534 nfds = i;
1535 error = EBADF;
1536 goto out;
1537 }
1538 }
1539
1540 /* Allocate new space and copy header into it. */
1541 newcm = malloc(CMSG_SPACE(nfds * sizeof(file_t *)), M_MBUF, M_WAITOK);
1542 if (newcm == NULL) {
1543 error = E2BIG;
1544 goto out;
1545 }
1546 memcpy(newcm, cm, sizeof(struct cmsghdr));
1547 files = (file_t **)CMSG_DATA(newcm);
1548
1549 /*
1550 * Transform the file descriptors into file_t pointers, in
1551 * reverse order so that if pointers are bigger than ints, the
1552 * int won't get until we're done. No need to lock, as we have
1553 * already validated the descriptors with fd_getfile().
1554 */
1555 fdp = (int *)CMSG_DATA(cm) + nfds;
1556 rp = files + nfds;
1557 for (i = 0; i < nfds; i++) {
1558 fp = fdescp->fd_dt->dt_ff[*--fdp]->ff_file;
1559 KASSERT(fp != NULL);
1560 mutex_enter(&fp->f_lock);
1561 *--rp = fp;
1562 fp->f_count++;
1563 fp->f_msgcount++;
1564 mutex_exit(&fp->f_lock);
1565 }
1566
1567 out:
1568 /* Release descriptor references. */
1569 fdp = (int *)CMSG_DATA(cm);
1570 for (i = 0; i < nfds; i++) {
1571 fd_putfile(*fdp++);
1572 if (error != 0) {
1573 atomic_dec_uint(&unp_rights);
1574 }
1575 }
1576
1577 if (error == 0) {
1578 if (control->m_flags & M_EXT) {
1579 m_freem(control);
1580 *controlp = control = m_get(M_WAIT, MT_CONTROL);
1581 }
1582 MEXTADD(control, newcm, CMSG_SPACE(nfds * sizeof(file_t *)),
1583 M_MBUF, NULL, NULL);
1584 cm = newcm;
1585 /*
1586 * Adjust message & mbuf to note amount of space
1587 * actually used.
1588 */
1589 cm->cmsg_len = CMSG_LEN(nfds * sizeof(file_t *));
1590 control->m_len = CMSG_SPACE(nfds * sizeof(file_t *));
1591 }
1592
1593 return error;
1594}
1595
1596struct mbuf *
1597unp_addsockcred(struct lwp *l, struct mbuf *control)
1598{
1599 struct sockcred *sc;
1600 struct mbuf *m;
1601 void *p;
1602
1603 m = sbcreatecontrol1(&p, SOCKCREDSIZE(kauth_cred_ngroups(l->l_cred)),
1604 SCM_CREDS, SOL_SOCKET, M_WAITOK);
1605 if (m == NULL)
1606 return control;
1607
1608 sc = p;
1609 sc->sc_pid = l->l_proc->p_pid;
1610 sc->sc_uid = kauth_cred_getuid(l->l_cred);
1611 sc->sc_euid = kauth_cred_geteuid(l->l_cred);
1612 sc->sc_gid = kauth_cred_getgid(l->l_cred);
1613 sc->sc_egid = kauth_cred_getegid(l->l_cred);
1614 sc->sc_ngroups = kauth_cred_ngroups(l->l_cred);
1615
1616 for (int i = 0; i < sc->sc_ngroups; i++)
1617 sc->sc_groups[i] = kauth_cred_group(l->l_cred, i);
1618
1619 return m_add(control, m);
1620}
1621
1622/*
1623 * Do a mark-sweep GC of files in the system, to free up any which are
1624 * caught in flight to an about-to-be-closed socket. Additionally,
1625 * process deferred file closures.
1626 */
1627static void
1628unp_gc(file_t *dp)
1629{
1630 extern struct domain unixdomain;
1631 file_t *fp, *np;
1632 struct socket *so, *so1;
1633 u_int i, oflags, rflags;
1634 bool didwork;
1635
1636 KASSERT(curlwp == unp_thread_lwp);
1637 KASSERT(mutex_owned(&filelist_lock));
1638
1639 /*
1640 * First, process deferred file closures.
1641 */
1642 while (!SLIST_EMPTY(&unp_thread_discard)) {
1643 fp = SLIST_FIRST(&unp_thread_discard);
1644 KASSERT(fp->f_unpcount > 0);
1645 KASSERT(fp->f_count > 0);
1646 KASSERT(fp->f_msgcount > 0);
1647 KASSERT(fp->f_count >= fp->f_unpcount);
1648 KASSERT(fp->f_count >= fp->f_msgcount);
1649 KASSERT(fp->f_msgcount >= fp->f_unpcount);
1650 SLIST_REMOVE_HEAD(&unp_thread_discard, f_unplist);
1651 i = fp->f_unpcount;
1652 fp->f_unpcount = 0;
1653 mutex_exit(&filelist_lock);
1654 for (; i != 0; i--) {
1655 unp_discard_now(fp);
1656 }
1657 mutex_enter(&filelist_lock);
1658 }
1659
1660 /*
1661 * Clear mark bits. Ensure that we don't consider new files
1662 * entering the file table during this loop (they will not have
1663 * FSCAN set).
1664 */
1665 unp_defer = 0;
1666 LIST_FOREACH(fp, &filehead, f_list) {
1667 for (oflags = fp->f_flag;; oflags = rflags) {
1668 rflags = atomic_cas_uint(&fp->f_flag, oflags,
1669 (oflags | FSCAN) & ~(FMARK|FDEFER));
1670 if (__predict_true(oflags == rflags)) {
1671 break;
1672 }
1673 }
1674 }
1675
1676 /*
1677 * Iterate over the set of sockets, marking ones believed (based on
1678 * refcount) to be referenced from a process, and marking for rescan
1679 * sockets which are queued on a socket. Recan continues descending
1680 * and searching for sockets referenced by sockets (FDEFER), until
1681 * there are no more socket->socket references to be discovered.
1682 */
1683 do {
1684 didwork = false;
1685 for (fp = LIST_FIRST(&filehead); fp != NULL; fp = np) {
1686 KASSERT(mutex_owned(&filelist_lock));
1687 np = LIST_NEXT(fp, f_list);
1688 mutex_enter(&fp->f_lock);
1689 if ((fp->f_flag & FDEFER) != 0) {
1690 atomic_and_uint(&fp->f_flag, ~FDEFER);
1691 unp_defer--;
1692 if (fp->f_count == 0) {
1693 /*
1694 * XXX: closef() doesn't pay attention
1695 * to FDEFER
1696 */
1697 mutex_exit(&fp->f_lock);
1698 continue;
1699 }
1700 } else {
1701 if (fp->f_count == 0 ||
1702 (fp->f_flag & FMARK) != 0 ||
1703 fp->f_count == fp->f_msgcount ||
1704 fp->f_unpcount != 0) {
1705 mutex_exit(&fp->f_lock);
1706 continue;
1707 }
1708 }
1709 atomic_or_uint(&fp->f_flag, FMARK);
1710
1711 if (fp->f_type != DTYPE_SOCKET ||
1712 (so = fp->f_socket) == NULL ||
1713 so->so_proto->pr_domain != &unixdomain ||
1714 (so->so_proto->pr_flags & PR_RIGHTS) == 0) {
1715 mutex_exit(&fp->f_lock);
1716 continue;
1717 }
1718
1719 /* Gain file ref, mark our position, and unlock. */
1720 didwork = true;
1721 LIST_INSERT_AFTER(fp, dp, f_list);
1722 fp->f_count++;
1723 mutex_exit(&fp->f_lock);
1724 mutex_exit(&filelist_lock);
1725
1726 /*
1727 * Mark files referenced from sockets queued on the
1728 * accept queue as well.
1729 */
1730 solock(so);
1731 unp_scan(so->so_rcv.sb_mb, unp_mark, 0);
1732 if ((so->so_options & SO_ACCEPTCONN) != 0) {
1733 TAILQ_FOREACH(so1, &so->so_q0, so_qe) {
1734 unp_scan(so1->so_rcv.sb_mb, unp_mark, 0);
1735 }
1736 TAILQ_FOREACH(so1, &so->so_q, so_qe) {
1737 unp_scan(so1->so_rcv.sb_mb, unp_mark, 0);
1738 }
1739 }
1740 sounlock(so);
1741
1742 /* Re-lock and restart from where we left off. */
1743 closef(fp);
1744 mutex_enter(&filelist_lock);
1745 np = LIST_NEXT(dp, f_list);
1746 LIST_REMOVE(dp, f_list);
1747 }
1748 /*
1749 * Bail early if we did nothing in the loop above. Could
1750 * happen because of concurrent activity causing unp_defer
1751 * to get out of sync.
1752 */
1753 } while (unp_defer != 0 && didwork);
1754
1755 /*
1756 * Sweep pass.
1757 *
1758 * We grab an extra reference to each of the files that are
1759 * not otherwise accessible and then free the rights that are
1760 * stored in messages on them.
1761 */
1762 for (fp = LIST_FIRST(&filehead); fp != NULL; fp = np) {
1763 KASSERT(mutex_owned(&filelist_lock));
1764 np = LIST_NEXT(fp, f_list);
1765 mutex_enter(&fp->f_lock);
1766
1767 /*
1768 * Ignore non-sockets.
1769 * Ignore dead sockets, or sockets with pending close.
1770 * Ignore sockets obviously referenced elsewhere.
1771 * Ignore sockets marked as referenced by our scan.
1772 * Ignore new sockets that did not exist during the scan.
1773 */
1774 if (fp->f_type != DTYPE_SOCKET ||
1775 fp->f_count == 0 || fp->f_unpcount != 0 ||
1776 fp->f_count != fp->f_msgcount ||
1777 (fp->f_flag & (FMARK | FSCAN)) != FSCAN) {
1778 mutex_exit(&fp->f_lock);
1779 continue;
1780 }
1781
1782 /* Gain file ref, mark our position, and unlock. */
1783 LIST_INSERT_AFTER(fp, dp, f_list);
1784 fp->f_count++;
1785 mutex_exit(&fp->f_lock);
1786 mutex_exit(&filelist_lock);
1787
1788 /*
1789 * Flush all data from the socket's receive buffer.
1790 * This will cause files referenced only by the
1791 * socket to be queued for close.
1792 */
1793 so = fp->f_socket;
1794 solock(so);
1795 sorflush(so);
1796 sounlock(so);
1797
1798 /* Re-lock and restart from where we left off. */
1799 closef(fp);
1800 mutex_enter(&filelist_lock);
1801 np = LIST_NEXT(dp, f_list);
1802 LIST_REMOVE(dp, f_list);
1803 }
1804}
1805
1806/*
1807 * Garbage collector thread. While SCM_RIGHTS messages are in transit,
1808 * wake once per second to garbage collect. Run continually while we
1809 * have deferred closes to process.
1810 */
1811static void
1812unp_thread(void *cookie)
1813{
1814 file_t *dp;
1815
1816 /* Allocate a dummy file for our scans. */
1817 if ((dp = fgetdummy()) == NULL) {
1818 panic("unp_thread");
1819 }
1820
1821 mutex_enter(&filelist_lock);
1822 for (;;) {
1823 KASSERT(mutex_owned(&filelist_lock));
1824 if (SLIST_EMPTY(&unp_thread_discard)) {
1825 if (unp_rights != 0) {
1826 (void)cv_timedwait(&unp_thread_cv,
1827 &filelist_lock, hz);
1828 } else {
1829 cv_wait(&unp_thread_cv, &filelist_lock);
1830 }
1831 }
1832 unp_gc(dp);
1833 }
1834 /* NOTREACHED */
1835}
1836
1837/*
1838 * Kick the garbage collector into action if there is something for
1839 * it to process.
1840 */
1841static void
1842unp_thread_kick(void)
1843{
1844
1845 if (!SLIST_EMPTY(&unp_thread_discard) || unp_rights != 0) {
1846 mutex_enter(&filelist_lock);
1847 cv_signal(&unp_thread_cv);
1848 mutex_exit(&filelist_lock);
1849 }
1850}
1851
1852void
1853unp_dispose(struct mbuf *m)
1854{
1855
1856 if (m)
1857 unp_scan(m, unp_discard_later, 1);
1858}
1859
1860void
1861unp_scan(struct mbuf *m0, void (*op)(file_t *), int discard)
1862{
1863 struct mbuf *m;
1864 file_t **rp, *fp;
1865 struct cmsghdr *cm;
1866 int i, qfds;
1867
1868 while (m0) {
1869 for (m = m0; m; m = m->m_next) {
1870 if (m->m_type != MT_CONTROL ||
1871 m->m_len < sizeof(*cm)) {
1872 continue;
1873 }
1874 cm = mtod(m, struct cmsghdr *);
1875 if (cm->cmsg_level != SOL_SOCKET ||
1876 cm->cmsg_type != SCM_RIGHTS)
1877 continue;
1878 qfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm)))
1879 / sizeof(file_t *);
1880 rp = (file_t **)CMSG_DATA(cm);
1881 for (i = 0; i < qfds; i++) {
1882 fp = *rp;
1883 if (discard) {
1884 *rp = 0;
1885 }
1886 (*op)(fp);
1887 rp++;
1888 }
1889 }
1890 m0 = m0->m_nextpkt;
1891 }
1892}
1893
1894void
1895unp_mark(file_t *fp)
1896{
1897
1898 if (fp == NULL)
1899 return;
1900
1901 /* If we're already deferred, don't screw up the defer count */
1902 mutex_enter(&fp->f_lock);
1903 if (fp->f_flag & (FMARK | FDEFER)) {
1904 mutex_exit(&fp->f_lock);
1905 return;
1906 }
1907
1908 /*
1909 * Minimize the number of deferrals... Sockets are the only type of
1910 * file which can hold references to another file, so just mark
1911 * other files, and defer unmarked sockets for the next pass.
1912 */
1913 if (fp->f_type == DTYPE_SOCKET) {
1914 unp_defer++;
1915 KASSERT(fp->f_count != 0);
1916 atomic_or_uint(&fp->f_flag, FDEFER);
1917 } else {
1918 atomic_or_uint(&fp->f_flag, FMARK);
1919 }
1920 mutex_exit(&fp->f_lock);
1921}
1922
1923static void
1924unp_discard_now(file_t *fp)
1925{
1926
1927 if (fp == NULL)
1928 return;
1929
1930 KASSERT(fp->f_count > 0);
1931 KASSERT(fp->f_msgcount > 0);
1932
1933 mutex_enter(&fp->f_lock);
1934 fp->f_msgcount--;
1935 mutex_exit(&fp->f_lock);
1936 atomic_dec_uint(&unp_rights);
1937 (void)closef(fp);
1938}
1939
1940static void
1941unp_discard_later(file_t *fp)
1942{
1943
1944 if (fp == NULL)
1945 return;
1946
1947 KASSERT(fp->f_count > 0);
1948 KASSERT(fp->f_msgcount > 0);
1949
1950 mutex_enter(&filelist_lock);
1951 if (fp->f_unpcount++ == 0) {
1952 SLIST_INSERT_HEAD(&unp_thread_discard, fp, f_unplist);
1953 }
1954 mutex_exit(&filelist_lock);
1955}
1956
1957const struct pr_usrreqs unp_usrreqs = {
1958 .pr_attach = unp_attach,
1959 .pr_detach = unp_detach,
1960 .pr_accept = unp_accept,
1961 .pr_bind = unp_bind,
1962 .pr_listen = unp_listen,
1963 .pr_connect = unp_connect,
1964 .pr_connect2 = unp_connect2,
1965 .pr_disconnect = unp_disconnect,
1966 .pr_shutdown = unp_shutdown,
1967 .pr_abort = unp_abort,
1968 .pr_ioctl = unp_ioctl,
1969 .pr_stat = unp_stat,
1970 .pr_peeraddr = unp_peeraddr,
1971 .pr_sockaddr = unp_sockaddr,
1972 .pr_rcvd = unp_rcvd,
1973 .pr_recvoob = unp_recvoob,
1974 .pr_send = unp_send,
1975 .pr_sendoob = unp_sendoob,
1976};
1977