uipc_usrreq.c revision 1.192 1 /* $NetBSD: uipc_usrreq.c,v 1.192 2019/03/01 11:06:57 pgoyette 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 1.192 2019/03/01 11:06:57 pgoyette Exp $");
100
101 #ifdef _KERNEL_OPT
102 #include "opt_compat_netbsd.h"
103 #endif
104
105 #include <sys/param.h>
106 #include <sys/systm.h>
107 #include <sys/proc.h>
108 #include <sys/filedesc.h>
109 #include <sys/domain.h>
110 #include <sys/protosw.h>
111 #include <sys/socket.h>
112 #include <sys/socketvar.h>
113 #include <sys/unpcb.h>
114 #include <sys/un.h>
115 #include <sys/namei.h>
116 #include <sys/vnode.h>
117 #include <sys/file.h>
118 #include <sys/stat.h>
119 #include <sys/mbuf.h>
120 #include <sys/kauth.h>
121 #include <sys/kmem.h>
122 #include <sys/atomic.h>
123 #include <sys/uidinfo.h>
124 #include <sys/kernel.h>
125 #include <sys/kthread.h>
126 #include <sys/compat_stub.h>
127
128 #include <compat/sys/socket.h>
129 #include <compat/net/route_70.h>
130
131 /*
132 * Unix communications domain.
133 *
134 * TODO:
135 * RDM
136 * rethink name space problems
137 * need a proper out-of-band
138 *
139 * Notes on locking:
140 *
141 * The generic rules noted in uipc_socket2.c apply. In addition:
142 *
143 * o We have a global lock, uipc_lock.
144 *
145 * o All datagram sockets are locked by uipc_lock.
146 *
147 * o For stream socketpairs, the two endpoints are created sharing the same
148 * independent lock. Sockets presented to PRU_CONNECT2 must already have
149 * matching locks.
150 *
151 * o Stream sockets created via socket() start life with their own
152 * independent lock.
153 *
154 * o Stream connections to a named endpoint are slightly more complicated.
155 * Sockets that have called listen() have their lock pointer mutated to
156 * the global uipc_lock. When establishing a connection, the connecting
157 * socket also has its lock mutated to uipc_lock, which matches the head
158 * (listening socket). We create a new socket for accept() to return, and
159 * that also shares the head's lock. Until the connection is completely
160 * done on both ends, all three sockets are locked by uipc_lock. Once the
161 * connection is complete, the association with the head's lock is broken.
162 * The connecting socket and the socket returned from accept() have their
163 * lock pointers mutated away from uipc_lock, and back to the connecting
164 * socket's original, independent lock. The head continues to be locked
165 * by uipc_lock.
166 *
167 * o If uipc_lock is determined to be a significant source of contention,
168 * it could easily be hashed out. It is difficult to simply make it an
169 * independent lock because of visibility / garbage collection issues:
170 * if a socket has been associated with a lock at any point, that lock
171 * must remain valid until the socket is no longer visible in the system.
172 * The lock must not be freed or otherwise destroyed until any sockets
173 * that had referenced it have also been destroyed.
174 */
175 const struct sockaddr_un sun_noname = {
176 .sun_len = offsetof(struct sockaddr_un, sun_path),
177 .sun_family = AF_LOCAL,
178 };
179 ino_t unp_ino; /* prototype for fake inode numbers */
180
181 static struct mbuf * unp_addsockcred(struct lwp *, struct mbuf *);
182 static void unp_discard_later(file_t *);
183 static void unp_discard_now(file_t *);
184 static void unp_disconnect1(struct unpcb *);
185 static bool unp_drop(struct unpcb *, int);
186 static int unp_internalize(struct mbuf **);
187 static void unp_mark(file_t *);
188 static void unp_scan(struct mbuf *, void (*)(file_t *), int);
189 static void unp_shutdown1(struct unpcb *);
190 static void unp_thread(void *);
191 static void unp_thread_kick(void);
192
193 static kmutex_t *uipc_lock;
194
195 static kcondvar_t unp_thread_cv;
196 static lwp_t *unp_thread_lwp;
197 static SLIST_HEAD(,file) unp_thread_discard;
198 static int unp_defer;
199
200 /* Compat interface */
201
202 struct mbuf * stub_compat_70_unp_addsockcred(lwp_t *, struct mbuf *);
203
204 struct mbuf * stub_compat_70_unp_addsockcred(struct lwp *lwp,
205 struct mbuf *control)
206 {
207
208 /* just copy our initial argument */
209 return control;
210 }
211
212 bool compat70_ocreds_valid = false;
213
214 /*
215 * Initialize Unix protocols.
216 */
217 void
218 uipc_init(void)
219 {
220 int error;
221
222 uipc_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
223 cv_init(&unp_thread_cv, "unpgc");
224
225 error = kthread_create(PRI_NONE, KTHREAD_MPSAFE, NULL, unp_thread,
226 NULL, &unp_thread_lwp, "unpgc");
227 if (error != 0)
228 panic("uipc_init %d", error);
229 }
230
231 static void
232 unp_connid(struct lwp *l, struct unpcb *unp, int flags)
233 {
234 unp->unp_connid.unp_pid = l->l_proc->p_pid;
235 unp->unp_connid.unp_euid = kauth_cred_geteuid(l->l_cred);
236 unp->unp_connid.unp_egid = kauth_cred_getegid(l->l_cred);
237 unp->unp_flags |= flags;
238 }
239
240 /*
241 * A connection succeeded: disassociate both endpoints from the head's
242 * lock, and make them share their own lock. There is a race here: for
243 * a very brief time one endpoint will be locked by a different lock
244 * than the other end. However, since the current thread holds the old
245 * lock (the listening socket's lock, the head) access can still only be
246 * made to one side of the connection.
247 */
248 static void
249 unp_setpeerlocks(struct socket *so, struct socket *so2)
250 {
251 struct unpcb *unp;
252 kmutex_t *lock;
253
254 KASSERT(solocked2(so, so2));
255
256 /*
257 * Bail out if either end of the socket is not yet fully
258 * connected or accepted. We only break the lock association
259 * with the head when the pair of sockets stand completely
260 * on their own.
261 */
262 KASSERT(so->so_head == NULL);
263 if (so2->so_head != NULL)
264 return;
265
266 /*
267 * Drop references to old lock. A third reference (from the
268 * queue head) must be held as we still hold its lock. Bonus:
269 * we don't need to worry about garbage collecting the lock.
270 */
271 lock = so->so_lock;
272 KASSERT(lock == uipc_lock);
273 mutex_obj_free(lock);
274 mutex_obj_free(lock);
275
276 /*
277 * Grab stream lock from the initiator and share between the two
278 * endpoints. Issue memory barrier to ensure all modifications
279 * become globally visible before the lock change. so2 is
280 * assumed not to have a stream lock, because it was created
281 * purely for the server side to accept this connection and
282 * started out life using the domain-wide lock.
283 */
284 unp = sotounpcb(so);
285 KASSERT(unp->unp_streamlock != NULL);
286 KASSERT(sotounpcb(so2)->unp_streamlock == NULL);
287 lock = unp->unp_streamlock;
288 unp->unp_streamlock = NULL;
289 mutex_obj_hold(lock);
290 membar_exit();
291 /*
292 * possible race if lock is not held - see comment in
293 * uipc_usrreq(PRU_ACCEPT).
294 */
295 KASSERT(mutex_owned(lock));
296 solockreset(so, lock);
297 solockreset(so2, lock);
298 }
299
300 /*
301 * Reset a socket's lock back to the domain-wide lock.
302 */
303 static void
304 unp_resetlock(struct socket *so)
305 {
306 kmutex_t *olock, *nlock;
307 struct unpcb *unp;
308
309 KASSERT(solocked(so));
310
311 olock = so->so_lock;
312 nlock = uipc_lock;
313 if (olock == nlock)
314 return;
315 unp = sotounpcb(so);
316 KASSERT(unp->unp_streamlock == NULL);
317 unp->unp_streamlock = olock;
318 mutex_obj_hold(nlock);
319 mutex_enter(nlock);
320 solockreset(so, nlock);
321 mutex_exit(olock);
322 }
323
324 static void
325 unp_free(struct unpcb *unp)
326 {
327 if (unp->unp_addr)
328 free(unp->unp_addr, M_SONAME);
329 if (unp->unp_streamlock != NULL)
330 mutex_obj_free(unp->unp_streamlock);
331 kmem_free(unp, sizeof(*unp));
332 }
333
334 static int
335 unp_output(struct mbuf *m, struct mbuf *control, struct unpcb *unp)
336 {
337 struct socket *so2;
338 const struct sockaddr_un *sun;
339
340 /* XXX: server side closed the socket */
341 if (unp->unp_conn == NULL)
342 return ECONNREFUSED;
343 so2 = unp->unp_conn->unp_socket;
344
345 KASSERT(solocked(so2));
346
347 if (unp->unp_addr)
348 sun = unp->unp_addr;
349 else
350 sun = &sun_noname;
351 if (unp->unp_conn->unp_flags & UNP_WANTCRED)
352 control = unp_addsockcred(curlwp, control);
353 if (unp->unp_conn->unp_flags & UNP_OWANTCRED)
354 MODULE_HOOK_CALL(uipc_unp_70_hook, (curlwp, control),
355 stub_compat_70_unp_addsockcred(curlwp, control), control);
356 if (sbappendaddr(&so2->so_rcv, (const struct sockaddr *)sun, m,
357 control) == 0) {
358 unp_dispose(control);
359 m_freem(control);
360 m_freem(m);
361 /* Don't call soroverflow because we're returning this
362 * error directly to the sender. */
363 so2->so_rcv.sb_overflowed++;
364 return ENOBUFS;
365 } else {
366 sorwakeup(so2);
367 return 0;
368 }
369 }
370
371 static void
372 unp_setaddr(struct socket *so, struct sockaddr *nam, bool peeraddr)
373 {
374 const struct sockaddr_un *sun = NULL;
375 struct unpcb *unp;
376
377 KASSERT(solocked(so));
378 unp = sotounpcb(so);
379
380 if (peeraddr) {
381 if (unp->unp_conn && unp->unp_conn->unp_addr)
382 sun = unp->unp_conn->unp_addr;
383 } else {
384 if (unp->unp_addr)
385 sun = unp->unp_addr;
386 }
387 if (sun == NULL)
388 sun = &sun_noname;
389
390 memcpy(nam, sun, sun->sun_len);
391 }
392
393 static int
394 unp_rcvd(struct socket *so, int flags, struct lwp *l)
395 {
396 struct unpcb *unp = sotounpcb(so);
397 struct socket *so2;
398 u_int newhiwat;
399
400 KASSERT(solocked(so));
401 KASSERT(unp != NULL);
402
403 switch (so->so_type) {
404
405 case SOCK_DGRAM:
406 panic("uipc 1");
407 /*NOTREACHED*/
408
409 case SOCK_SEQPACKET: /* FALLTHROUGH */
410 case SOCK_STREAM:
411 #define rcv (&so->so_rcv)
412 #define snd (&so2->so_snd)
413 if (unp->unp_conn == 0)
414 break;
415 so2 = unp->unp_conn->unp_socket;
416 KASSERT(solocked2(so, so2));
417 /*
418 * Adjust backpressure on sender
419 * and wakeup any waiting to write.
420 */
421 snd->sb_mbmax += unp->unp_mbcnt - rcv->sb_mbcnt;
422 unp->unp_mbcnt = rcv->sb_mbcnt;
423 newhiwat = snd->sb_hiwat + unp->unp_cc - rcv->sb_cc;
424 (void)chgsbsize(so2->so_uidinfo,
425 &snd->sb_hiwat, newhiwat, RLIM_INFINITY);
426 unp->unp_cc = rcv->sb_cc;
427 sowwakeup(so2);
428 #undef snd
429 #undef rcv
430 break;
431
432 default:
433 panic("uipc 2");
434 }
435
436 return 0;
437 }
438
439 static int
440 unp_recvoob(struct socket *so, struct mbuf *m, int flags)
441 {
442 KASSERT(solocked(so));
443
444 return EOPNOTSUPP;
445 }
446
447 static int
448 unp_send(struct socket *so, struct mbuf *m, struct sockaddr *nam,
449 struct mbuf *control, struct lwp *l)
450 {
451 struct unpcb *unp = sotounpcb(so);
452 int error = 0;
453 u_int newhiwat;
454 struct socket *so2;
455
456 KASSERT(solocked(so));
457 KASSERT(unp != NULL);
458 KASSERT(m != NULL);
459
460 /*
461 * Note: unp_internalize() rejects any control message
462 * other than SCM_RIGHTS, and only allows one. This
463 * has the side-effect of preventing a caller from
464 * forging SCM_CREDS.
465 */
466 if (control) {
467 sounlock(so);
468 error = unp_internalize(&control);
469 solock(so);
470 if (error != 0) {
471 m_freem(control);
472 m_freem(m);
473 return error;
474 }
475 }
476
477 switch (so->so_type) {
478
479 case SOCK_DGRAM: {
480 KASSERT(so->so_lock == uipc_lock);
481 if (nam) {
482 if ((so->so_state & SS_ISCONNECTED) != 0)
483 error = EISCONN;
484 else {
485 /*
486 * Note: once connected, the
487 * socket's lock must not be
488 * dropped until we have sent
489 * the message and disconnected.
490 * This is necessary to prevent
491 * intervening control ops, like
492 * another connection.
493 */
494 error = unp_connect(so, nam, l);
495 }
496 } else {
497 if ((so->so_state & SS_ISCONNECTED) == 0)
498 error = ENOTCONN;
499 }
500 if (error) {
501 unp_dispose(control);
502 m_freem(control);
503 m_freem(m);
504 return error;
505 }
506 error = unp_output(m, control, unp);
507 if (nam)
508 unp_disconnect1(unp);
509 break;
510 }
511
512 case SOCK_SEQPACKET: /* FALLTHROUGH */
513 case SOCK_STREAM:
514 #define rcv (&so2->so_rcv)
515 #define snd (&so->so_snd)
516 if (unp->unp_conn == NULL) {
517 error = ENOTCONN;
518 break;
519 }
520 so2 = unp->unp_conn->unp_socket;
521 KASSERT(solocked2(so, so2));
522 if (unp->unp_conn->unp_flags & UNP_WANTCRED) {
523 /*
524 * Credentials are passed only once on
525 * SOCK_STREAM and SOCK_SEQPACKET.
526 */
527 unp->unp_conn->unp_flags &= ~UNP_WANTCRED;
528 control = unp_addsockcred(l, control);
529 }
530 if (unp->unp_conn->unp_flags & UNP_OWANTCRED) {
531 /*
532 * Credentials are passed only once on
533 * SOCK_STREAM and SOCK_SEQPACKET.
534 */
535 unp->unp_conn->unp_flags &= ~UNP_OWANTCRED;
536 MODULE_HOOK_CALL(uipc_unp_70_hook, (curlwp, control),
537 stub_compat_70_unp_addsockcred(curlwp, control),
538 control);
539 }
540 /*
541 * Send to paired receive port, and then reduce
542 * send buffer hiwater marks to maintain backpressure.
543 * Wake up readers.
544 */
545 if (control) {
546 if (sbappendcontrol(rcv, m, control) != 0)
547 control = NULL;
548 } else {
549 switch(so->so_type) {
550 case SOCK_SEQPACKET:
551 sbappendrecord(rcv, m);
552 break;
553 case SOCK_STREAM:
554 sbappend(rcv, m);
555 break;
556 default:
557 panic("uipc_usrreq");
558 break;
559 }
560 }
561 snd->sb_mbmax -=
562 rcv->sb_mbcnt - unp->unp_conn->unp_mbcnt;
563 unp->unp_conn->unp_mbcnt = rcv->sb_mbcnt;
564 newhiwat = snd->sb_hiwat -
565 (rcv->sb_cc - unp->unp_conn->unp_cc);
566 (void)chgsbsize(so->so_uidinfo,
567 &snd->sb_hiwat, newhiwat, RLIM_INFINITY);
568 unp->unp_conn->unp_cc = rcv->sb_cc;
569 sorwakeup(so2);
570 #undef snd
571 #undef rcv
572 if (control != NULL) {
573 unp_dispose(control);
574 m_freem(control);
575 }
576 break;
577
578 default:
579 panic("uipc 4");
580 }
581
582 return error;
583 }
584
585 static int
586 unp_sendoob(struct socket *so, struct mbuf *m, struct mbuf * control)
587 {
588 KASSERT(solocked(so));
589
590 m_freem(m);
591 m_freem(control);
592
593 return EOPNOTSUPP;
594 }
595
596 /*
597 * Unix domain socket option processing.
598 */
599 int
600 uipc_ctloutput(int op, struct socket *so, struct sockopt *sopt)
601 {
602 struct unpcb *unp = sotounpcb(so);
603 int optval = 0, error = 0;
604
605 KASSERT(solocked(so));
606
607 if (sopt->sopt_level != 0) {
608 error = ENOPROTOOPT;
609 } else switch (op) {
610
611 case PRCO_SETOPT:
612 switch (sopt->sopt_name) {
613 case LOCAL_OCREDS:
614 if (!compat70_ocreds_valid) {
615 error = ENOPROTOOPT;
616 break;
617 }
618 /* FALLTHROUGH */
619 case LOCAL_CREDS:
620 case LOCAL_CONNWAIT:
621 error = sockopt_getint(sopt, &optval);
622 if (error)
623 break;
624 switch (sopt->sopt_name) {
625 #define OPTSET(bit) \
626 if (optval) \
627 unp->unp_flags |= (bit); \
628 else \
629 unp->unp_flags &= ~(bit);
630
631 case LOCAL_CREDS:
632 OPTSET(UNP_WANTCRED);
633 break;
634 case LOCAL_CONNWAIT:
635 OPTSET(UNP_CONNWAIT);
636 break;
637 case LOCAL_OCREDS:
638 OPTSET(UNP_OWANTCRED);
639 break;
640 }
641 break;
642 #undef OPTSET
643
644 default:
645 error = ENOPROTOOPT;
646 break;
647 }
648 break;
649
650 case PRCO_GETOPT:
651 sounlock(so);
652 switch (sopt->sopt_name) {
653 case LOCAL_PEEREID:
654 if (unp->unp_flags & UNP_EIDSVALID) {
655 error = sockopt_set(sopt, &unp->unp_connid,
656 sizeof(unp->unp_connid));
657 } else {
658 error = EINVAL;
659 }
660 break;
661 case LOCAL_CREDS:
662 #define OPTBIT(bit) (unp->unp_flags & (bit) ? 1 : 0)
663
664 optval = OPTBIT(UNP_WANTCRED);
665 error = sockopt_setint(sopt, optval);
666 break;
667 case LOCAL_OCREDS:
668 if (compat70_ocreds_valid) {
669 optval = OPTBIT(UNP_OWANTCRED);
670 error = sockopt_setint(sopt, optval);
671 break;
672 }
673 #undef OPTBIT
674 /* FALLTHROUGH */
675 default:
676 error = ENOPROTOOPT;
677 break;
678 }
679 solock(so);
680 break;
681 }
682 return (error);
683 }
684
685 /*
686 * Both send and receive buffers are allocated PIPSIZ bytes of buffering
687 * for stream sockets, although the total for sender and receiver is
688 * actually only PIPSIZ.
689 * Datagram sockets really use the sendspace as the maximum datagram size,
690 * and don't really want to reserve the sendspace. Their recvspace should
691 * be large enough for at least one max-size datagram plus address.
692 */
693 #ifndef PIPSIZ
694 #define PIPSIZ 8192
695 #endif
696 u_long unpst_sendspace = PIPSIZ;
697 u_long unpst_recvspace = PIPSIZ;
698 u_long unpdg_sendspace = 2*1024; /* really max datagram size */
699 u_long unpdg_recvspace = 16*1024;
700
701 u_int unp_rights; /* files in flight */
702 u_int unp_rights_ratio = 2; /* limit, fraction of maxfiles */
703
704 static int
705 unp_attach(struct socket *so, int proto)
706 {
707 struct unpcb *unp = sotounpcb(so);
708 u_long sndspc, rcvspc;
709 int error;
710
711 KASSERT(unp == NULL);
712
713 switch (so->so_type) {
714 case SOCK_SEQPACKET:
715 /* FALLTHROUGH */
716 case SOCK_STREAM:
717 if (so->so_lock == NULL) {
718 so->so_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
719 solock(so);
720 }
721 sndspc = unpst_sendspace;
722 rcvspc = unpst_recvspace;
723 break;
724
725 case SOCK_DGRAM:
726 if (so->so_lock == NULL) {
727 mutex_obj_hold(uipc_lock);
728 so->so_lock = uipc_lock;
729 solock(so);
730 }
731 sndspc = unpdg_sendspace;
732 rcvspc = unpdg_recvspace;
733 break;
734
735 default:
736 panic("unp_attach");
737 }
738
739 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
740 error = soreserve(so, sndspc, rcvspc);
741 if (error) {
742 return error;
743 }
744 }
745
746 unp = kmem_zalloc(sizeof(*unp), KM_SLEEP);
747 nanotime(&unp->unp_ctime);
748 unp->unp_socket = so;
749 so->so_pcb = unp;
750
751 KASSERT(solocked(so));
752 return 0;
753 }
754
755 static void
756 unp_detach(struct socket *so)
757 {
758 struct unpcb *unp;
759 vnode_t *vp;
760
761 unp = sotounpcb(so);
762 KASSERT(unp != NULL);
763 KASSERT(solocked(so));
764 retry:
765 if ((vp = unp->unp_vnode) != NULL) {
766 sounlock(so);
767 /* Acquire v_interlock to protect against unp_connect(). */
768 /* XXXAD racy */
769 mutex_enter(vp->v_interlock);
770 vp->v_socket = NULL;
771 mutex_exit(vp->v_interlock);
772 vrele(vp);
773 solock(so);
774 unp->unp_vnode = NULL;
775 }
776 if (unp->unp_conn)
777 unp_disconnect1(unp);
778 while (unp->unp_refs) {
779 KASSERT(solocked2(so, unp->unp_refs->unp_socket));
780 if (unp_drop(unp->unp_refs, ECONNRESET)) {
781 solock(so);
782 goto retry;
783 }
784 }
785 soisdisconnected(so);
786 so->so_pcb = NULL;
787 if (unp_rights) {
788 /*
789 * Normally the receive buffer is flushed later, in sofree,
790 * but if our receive buffer holds references to files that
791 * are now garbage, we will enqueue those file references to
792 * the garbage collector and kick it into action.
793 */
794 sorflush(so);
795 unp_free(unp);
796 unp_thread_kick();
797 } else
798 unp_free(unp);
799 }
800
801 static int
802 unp_accept(struct socket *so, struct sockaddr *nam)
803 {
804 struct unpcb *unp = sotounpcb(so);
805 struct socket *so2;
806
807 KASSERT(solocked(so));
808 KASSERT(nam != NULL);
809
810 /* XXX code review required to determine if unp can ever be NULL */
811 if (unp == NULL)
812 return EINVAL;
813
814 KASSERT(so->so_lock == uipc_lock);
815 /*
816 * Mark the initiating STREAM socket as connected *ONLY*
817 * after it's been accepted. This prevents a client from
818 * overrunning a server and receiving ECONNREFUSED.
819 */
820 if (unp->unp_conn == NULL) {
821 /*
822 * This will use the empty socket and will not
823 * allocate.
824 */
825 unp_setaddr(so, nam, true);
826 return 0;
827 }
828 so2 = unp->unp_conn->unp_socket;
829 if (so2->so_state & SS_ISCONNECTING) {
830 KASSERT(solocked2(so, so->so_head));
831 KASSERT(solocked2(so2, so->so_head));
832 soisconnected(so2);
833 }
834 /*
835 * If the connection is fully established, break the
836 * association with uipc_lock and give the connected
837 * pair a separate lock to share.
838 * There is a race here: sotounpcb(so2)->unp_streamlock
839 * is not locked, so when changing so2->so_lock
840 * another thread can grab it while so->so_lock is still
841 * pointing to the (locked) uipc_lock.
842 * this should be harmless, except that this makes
843 * solocked2() and solocked() unreliable.
844 * Another problem is that unp_setaddr() expects the
845 * the socket locked. Grabing sotounpcb(so2)->unp_streamlock
846 * fixes both issues.
847 */
848 mutex_enter(sotounpcb(so2)->unp_streamlock);
849 unp_setpeerlocks(so2, so);
850 /*
851 * Only now return peer's address, as we may need to
852 * block in order to allocate memory.
853 *
854 * XXX Minor race: connection can be broken while
855 * lock is dropped in unp_setaddr(). We will return
856 * error == 0 and sun_noname as the peer address.
857 */
858 unp_setaddr(so, nam, true);
859 /* so_lock now points to unp_streamlock */
860 mutex_exit(so2->so_lock);
861 return 0;
862 }
863
864 static int
865 unp_ioctl(struct socket *so, u_long cmd, void *nam, struct ifnet *ifp)
866 {
867 return EOPNOTSUPP;
868 }
869
870 static int
871 unp_stat(struct socket *so, struct stat *ub)
872 {
873 struct unpcb *unp;
874 struct socket *so2;
875
876 KASSERT(solocked(so));
877
878 unp = sotounpcb(so);
879 if (unp == NULL)
880 return EINVAL;
881
882 ub->st_blksize = so->so_snd.sb_hiwat;
883 switch (so->so_type) {
884 case SOCK_SEQPACKET: /* FALLTHROUGH */
885 case SOCK_STREAM:
886 if (unp->unp_conn == 0)
887 break;
888
889 so2 = unp->unp_conn->unp_socket;
890 KASSERT(solocked2(so, so2));
891 ub->st_blksize += so2->so_rcv.sb_cc;
892 break;
893 default:
894 break;
895 }
896 ub->st_dev = NODEV;
897 if (unp->unp_ino == 0)
898 unp->unp_ino = unp_ino++;
899 ub->st_atimespec = ub->st_mtimespec = ub->st_ctimespec = unp->unp_ctime;
900 ub->st_ino = unp->unp_ino;
901 return (0);
902 }
903
904 static int
905 unp_peeraddr(struct socket *so, struct sockaddr *nam)
906 {
907 KASSERT(solocked(so));
908 KASSERT(sotounpcb(so) != NULL);
909 KASSERT(nam != NULL);
910
911 unp_setaddr(so, nam, true);
912 return 0;
913 }
914
915 static int
916 unp_sockaddr(struct socket *so, struct sockaddr *nam)
917 {
918 KASSERT(solocked(so));
919 KASSERT(sotounpcb(so) != NULL);
920 KASSERT(nam != NULL);
921
922 unp_setaddr(so, nam, false);
923 return 0;
924 }
925
926 /*
927 * we only need to perform this allocation until syscalls other than
928 * bind are adjusted to use sockaddr_big.
929 */
930 static struct sockaddr_un *
931 makeun_sb(struct sockaddr *nam, size_t *addrlen)
932 {
933 struct sockaddr_un *sun;
934
935 *addrlen = nam->sa_len + 1;
936 sun = malloc(*addrlen, M_SONAME, M_WAITOK);
937 memcpy(sun, nam, nam->sa_len);
938 *(((char *)sun) + nam->sa_len) = '\0';
939 return sun;
940 }
941
942 static int
943 unp_bind(struct socket *so, struct sockaddr *nam, struct lwp *l)
944 {
945 struct sockaddr_un *sun;
946 struct unpcb *unp;
947 vnode_t *vp;
948 struct vattr vattr;
949 size_t addrlen;
950 int error;
951 struct pathbuf *pb;
952 struct nameidata nd;
953 proc_t *p;
954
955 unp = sotounpcb(so);
956
957 KASSERT(solocked(so));
958 KASSERT(unp != NULL);
959 KASSERT(nam != NULL);
960
961 if (unp->unp_vnode != NULL)
962 return (EINVAL);
963 if ((unp->unp_flags & UNP_BUSY) != 0) {
964 /*
965 * EALREADY may not be strictly accurate, but since this
966 * is a major application error it's hardly a big deal.
967 */
968 return (EALREADY);
969 }
970 unp->unp_flags |= UNP_BUSY;
971 sounlock(so);
972
973 p = l->l_proc;
974 sun = makeun_sb(nam, &addrlen);
975
976 pb = pathbuf_create(sun->sun_path);
977 if (pb == NULL) {
978 error = ENOMEM;
979 goto bad;
980 }
981 NDINIT(&nd, CREATE, FOLLOW | LOCKPARENT | TRYEMULROOT, pb);
982
983 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
984 if ((error = namei(&nd)) != 0) {
985 pathbuf_destroy(pb);
986 goto bad;
987 }
988 vp = nd.ni_vp;
989 if (vp != NULL) {
990 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
991 if (nd.ni_dvp == vp)
992 vrele(nd.ni_dvp);
993 else
994 vput(nd.ni_dvp);
995 vrele(vp);
996 pathbuf_destroy(pb);
997 error = EADDRINUSE;
998 goto bad;
999 }
1000 vattr_null(&vattr);
1001 vattr.va_type = VSOCK;
1002 vattr.va_mode = ACCESSPERMS & ~(p->p_cwdi->cwdi_cmask);
1003 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
1004 if (error) {
1005 vput(nd.ni_dvp);
1006 pathbuf_destroy(pb);
1007 goto bad;
1008 }
1009 vp = nd.ni_vp;
1010 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1011 solock(so);
1012 vp->v_socket = unp->unp_socket;
1013 unp->unp_vnode = vp;
1014 unp->unp_addrlen = addrlen;
1015 unp->unp_addr = sun;
1016 VOP_UNLOCK(vp);
1017 vput(nd.ni_dvp);
1018 unp->unp_flags &= ~UNP_BUSY;
1019 pathbuf_destroy(pb);
1020 return (0);
1021
1022 bad:
1023 free(sun, M_SONAME);
1024 solock(so);
1025 unp->unp_flags &= ~UNP_BUSY;
1026 return (error);
1027 }
1028
1029 static int
1030 unp_listen(struct socket *so, struct lwp *l)
1031 {
1032 struct unpcb *unp = sotounpcb(so);
1033
1034 KASSERT(solocked(so));
1035 KASSERT(unp != NULL);
1036
1037 /*
1038 * If the socket can accept a connection, it must be
1039 * locked by uipc_lock.
1040 */
1041 unp_resetlock(so);
1042 if (unp->unp_vnode == NULL)
1043 return EINVAL;
1044
1045 unp_connid(l, unp, UNP_EIDSBIND);
1046 return 0;
1047 }
1048
1049 static int
1050 unp_disconnect(struct socket *so)
1051 {
1052 KASSERT(solocked(so));
1053 KASSERT(sotounpcb(so) != NULL);
1054
1055 unp_disconnect1(sotounpcb(so));
1056 return 0;
1057 }
1058
1059 static int
1060 unp_shutdown(struct socket *so)
1061 {
1062 KASSERT(solocked(so));
1063 KASSERT(sotounpcb(so) != NULL);
1064
1065 socantsendmore(so);
1066 unp_shutdown1(sotounpcb(so));
1067 return 0;
1068 }
1069
1070 static int
1071 unp_abort(struct socket *so)
1072 {
1073 KASSERT(solocked(so));
1074 KASSERT(sotounpcb(so) != NULL);
1075
1076 (void)unp_drop(sotounpcb(so), ECONNABORTED);
1077 KASSERT(so->so_head == NULL);
1078 KASSERT(so->so_pcb != NULL);
1079 unp_detach(so);
1080 return 0;
1081 }
1082
1083 static int
1084 unp_connect1(struct socket *so, struct socket *so2, struct lwp *l)
1085 {
1086 struct unpcb *unp = sotounpcb(so);
1087 struct unpcb *unp2;
1088
1089 if (so2->so_type != so->so_type)
1090 return EPROTOTYPE;
1091
1092 /*
1093 * All three sockets involved must be locked by same lock:
1094 *
1095 * local endpoint (so)
1096 * remote endpoint (so2)
1097 * queue head (so2->so_head, only if PR_CONNREQUIRED)
1098 */
1099 KASSERT(solocked2(so, so2));
1100 KASSERT(so->so_head == NULL);
1101 if (so2->so_head != NULL) {
1102 KASSERT(so2->so_lock == uipc_lock);
1103 KASSERT(solocked2(so2, so2->so_head));
1104 }
1105
1106 unp2 = sotounpcb(so2);
1107 unp->unp_conn = unp2;
1108
1109 switch (so->so_type) {
1110
1111 case SOCK_DGRAM:
1112 unp->unp_nextref = unp2->unp_refs;
1113 unp2->unp_refs = unp;
1114 soisconnected(so);
1115 break;
1116
1117 case SOCK_SEQPACKET: /* FALLTHROUGH */
1118 case SOCK_STREAM:
1119
1120 /*
1121 * SOCK_SEQPACKET and SOCK_STREAM cases are handled by callers
1122 * which are unp_connect() or unp_connect2().
1123 */
1124
1125 break;
1126
1127 default:
1128 panic("unp_connect1");
1129 }
1130
1131 return 0;
1132 }
1133
1134 int
1135 unp_connect(struct socket *so, struct sockaddr *nam, struct lwp *l)
1136 {
1137 struct sockaddr_un *sun;
1138 vnode_t *vp;
1139 struct socket *so2, *so3;
1140 struct unpcb *unp, *unp2, *unp3;
1141 size_t addrlen;
1142 int error;
1143 struct pathbuf *pb;
1144 struct nameidata nd;
1145
1146 unp = sotounpcb(so);
1147 if ((unp->unp_flags & UNP_BUSY) != 0) {
1148 /*
1149 * EALREADY may not be strictly accurate, but since this
1150 * is a major application error it's hardly a big deal.
1151 */
1152 return (EALREADY);
1153 }
1154 unp->unp_flags |= UNP_BUSY;
1155 sounlock(so);
1156
1157 sun = makeun_sb(nam, &addrlen);
1158 pb = pathbuf_create(sun->sun_path);
1159 if (pb == NULL) {
1160 error = ENOMEM;
1161 goto bad2;
1162 }
1163
1164 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | TRYEMULROOT, pb);
1165
1166 if ((error = namei(&nd)) != 0) {
1167 pathbuf_destroy(pb);
1168 goto bad2;
1169 }
1170 vp = nd.ni_vp;
1171 pathbuf_destroy(pb);
1172 if (vp->v_type != VSOCK) {
1173 error = ENOTSOCK;
1174 goto bad;
1175 }
1176 if ((error = VOP_ACCESS(vp, VWRITE, l->l_cred)) != 0)
1177 goto bad;
1178 /* Acquire v_interlock to protect against unp_detach(). */
1179 mutex_enter(vp->v_interlock);
1180 so2 = vp->v_socket;
1181 if (so2 == NULL) {
1182 mutex_exit(vp->v_interlock);
1183 error = ECONNREFUSED;
1184 goto bad;
1185 }
1186 if (so->so_type != so2->so_type) {
1187 mutex_exit(vp->v_interlock);
1188 error = EPROTOTYPE;
1189 goto bad;
1190 }
1191 solock(so);
1192 unp_resetlock(so);
1193 mutex_exit(vp->v_interlock);
1194 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) != 0) {
1195 /*
1196 * This may seem somewhat fragile but is OK: if we can
1197 * see SO_ACCEPTCONN set on the endpoint, then it must
1198 * be locked by the domain-wide uipc_lock.
1199 */
1200 KASSERT((so2->so_options & SO_ACCEPTCONN) == 0 ||
1201 so2->so_lock == uipc_lock);
1202 if ((so2->so_options & SO_ACCEPTCONN) == 0 ||
1203 (so3 = sonewconn(so2, false)) == NULL) {
1204 error = ECONNREFUSED;
1205 sounlock(so);
1206 goto bad;
1207 }
1208 unp2 = sotounpcb(so2);
1209 unp3 = sotounpcb(so3);
1210 if (unp2->unp_addr) {
1211 unp3->unp_addr = malloc(unp2->unp_addrlen,
1212 M_SONAME, M_WAITOK);
1213 memcpy(unp3->unp_addr, unp2->unp_addr,
1214 unp2->unp_addrlen);
1215 unp3->unp_addrlen = unp2->unp_addrlen;
1216 }
1217 unp3->unp_flags = unp2->unp_flags;
1218 so2 = so3;
1219 /*
1220 * The connector's (client's) credentials are copied from its
1221 * process structure at the time of connect() (which is now).
1222 */
1223 unp_connid(l, unp3, UNP_EIDSVALID);
1224 /*
1225 * The receiver's (server's) credentials are copied from the
1226 * unp_peercred member of socket on which the former called
1227 * listen(); unp_listen() cached that process's credentials
1228 * at that time so we can use them now.
1229 */
1230 if (unp2->unp_flags & UNP_EIDSBIND) {
1231 memcpy(&unp->unp_connid, &unp2->unp_connid,
1232 sizeof(unp->unp_connid));
1233 unp->unp_flags |= UNP_EIDSVALID;
1234 }
1235 }
1236 error = unp_connect1(so, so2, l);
1237 if (error) {
1238 sounlock(so);
1239 goto bad;
1240 }
1241 unp2 = sotounpcb(so2);
1242 switch (so->so_type) {
1243
1244 /*
1245 * SOCK_DGRAM and default cases are handled in prior call to
1246 * unp_connect1(), do not add a default case without fixing
1247 * unp_connect1().
1248 */
1249
1250 case SOCK_SEQPACKET: /* FALLTHROUGH */
1251 case SOCK_STREAM:
1252 unp2->unp_conn = unp;
1253 if ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT)
1254 soisconnecting(so);
1255 else
1256 soisconnected(so);
1257 soisconnected(so2);
1258 /*
1259 * If the connection is fully established, break the
1260 * association with uipc_lock and give the connected
1261 * pair a seperate lock to share.
1262 */
1263 KASSERT(so2->so_head != NULL);
1264 unp_setpeerlocks(so, so2);
1265 break;
1266
1267 }
1268 sounlock(so);
1269 bad:
1270 vput(vp);
1271 bad2:
1272 free(sun, M_SONAME);
1273 solock(so);
1274 unp->unp_flags &= ~UNP_BUSY;
1275 return (error);
1276 }
1277
1278 int
1279 unp_connect2(struct socket *so, struct socket *so2)
1280 {
1281 struct unpcb *unp = sotounpcb(so);
1282 struct unpcb *unp2;
1283 int error = 0;
1284
1285 KASSERT(solocked2(so, so2));
1286
1287 error = unp_connect1(so, so2, curlwp);
1288 if (error)
1289 return error;
1290
1291 unp2 = sotounpcb(so2);
1292 switch (so->so_type) {
1293
1294 /*
1295 * SOCK_DGRAM and default cases are handled in prior call to
1296 * unp_connect1(), do not add a default case without fixing
1297 * unp_connect1().
1298 */
1299
1300 case SOCK_SEQPACKET: /* FALLTHROUGH */
1301 case SOCK_STREAM:
1302 unp2->unp_conn = unp;
1303 soisconnected(so);
1304 soisconnected(so2);
1305 break;
1306
1307 }
1308 return error;
1309 }
1310
1311 static void
1312 unp_disconnect1(struct unpcb *unp)
1313 {
1314 struct unpcb *unp2 = unp->unp_conn;
1315 struct socket *so;
1316
1317 if (unp2 == 0)
1318 return;
1319 unp->unp_conn = 0;
1320 so = unp->unp_socket;
1321 switch (so->so_type) {
1322 case SOCK_DGRAM:
1323 if (unp2->unp_refs == unp)
1324 unp2->unp_refs = unp->unp_nextref;
1325 else {
1326 unp2 = unp2->unp_refs;
1327 for (;;) {
1328 KASSERT(solocked2(so, unp2->unp_socket));
1329 if (unp2 == 0)
1330 panic("unp_disconnect1");
1331 if (unp2->unp_nextref == unp)
1332 break;
1333 unp2 = unp2->unp_nextref;
1334 }
1335 unp2->unp_nextref = unp->unp_nextref;
1336 }
1337 unp->unp_nextref = 0;
1338 so->so_state &= ~SS_ISCONNECTED;
1339 break;
1340
1341 case SOCK_SEQPACKET: /* FALLTHROUGH */
1342 case SOCK_STREAM:
1343 KASSERT(solocked2(so, unp2->unp_socket));
1344 soisdisconnected(so);
1345 unp2->unp_conn = 0;
1346 soisdisconnected(unp2->unp_socket);
1347 break;
1348 }
1349 }
1350
1351 static void
1352 unp_shutdown1(struct unpcb *unp)
1353 {
1354 struct socket *so;
1355
1356 switch(unp->unp_socket->so_type) {
1357 case SOCK_SEQPACKET: /* FALLTHROUGH */
1358 case SOCK_STREAM:
1359 if (unp->unp_conn && (so = unp->unp_conn->unp_socket))
1360 socantrcvmore(so);
1361 break;
1362 default:
1363 break;
1364 }
1365 }
1366
1367 static bool
1368 unp_drop(struct unpcb *unp, int errno)
1369 {
1370 struct socket *so = unp->unp_socket;
1371
1372 KASSERT(solocked(so));
1373
1374 so->so_error = errno;
1375 unp_disconnect1(unp);
1376 if (so->so_head) {
1377 so->so_pcb = NULL;
1378 /* sofree() drops the socket lock */
1379 sofree(so);
1380 unp_free(unp);
1381 return true;
1382 }
1383 return false;
1384 }
1385
1386 #ifdef notdef
1387 unp_drain(void)
1388 {
1389
1390 }
1391 #endif
1392
1393 int
1394 unp_externalize(struct mbuf *rights, struct lwp *l, int flags)
1395 {
1396 struct cmsghdr * const cm = mtod(rights, struct cmsghdr *);
1397 struct proc * const p = l->l_proc;
1398 file_t **rp;
1399 int error = 0;
1400
1401 const size_t nfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm))) /
1402 sizeof(file_t *);
1403 if (nfds == 0)
1404 goto noop;
1405
1406 int * const fdp = kmem_alloc(nfds * sizeof(int), KM_SLEEP);
1407 rw_enter(&p->p_cwdi->cwdi_lock, RW_READER);
1408
1409 /* Make sure the recipient should be able to see the files.. */
1410 rp = (file_t **)CMSG_DATA(cm);
1411 for (size_t i = 0; i < nfds; i++) {
1412 file_t * const fp = *rp++;
1413 if (fp == NULL) {
1414 error = EINVAL;
1415 goto out;
1416 }
1417 /*
1418 * If we are in a chroot'ed directory, and
1419 * someone wants to pass us a directory, make
1420 * sure it's inside the subtree we're allowed
1421 * to access.
1422 */
1423 if (p->p_cwdi->cwdi_rdir != NULL && fp->f_type == DTYPE_VNODE) {
1424 vnode_t *vp = fp->f_vnode;
1425 if ((vp->v_type == VDIR) &&
1426 !vn_isunder(vp, p->p_cwdi->cwdi_rdir, l)) {
1427 error = EPERM;
1428 goto out;
1429 }
1430 }
1431 }
1432
1433 restart:
1434 /*
1435 * First loop -- allocate file descriptor table slots for the
1436 * new files.
1437 */
1438 for (size_t i = 0; i < nfds; i++) {
1439 if ((error = fd_alloc(p, 0, &fdp[i])) != 0) {
1440 /*
1441 * Back out what we've done so far.
1442 */
1443 while (i-- > 0) {
1444 fd_abort(p, NULL, fdp[i]);
1445 }
1446 if (error == ENOSPC) {
1447 fd_tryexpand(p);
1448 error = 0;
1449 goto restart;
1450 }
1451 /*
1452 * This is the error that has historically
1453 * been returned, and some callers may
1454 * expect it.
1455 */
1456 error = EMSGSIZE;
1457 goto out;
1458 }
1459 }
1460
1461 /*
1462 * Now that adding them has succeeded, update all of the
1463 * file passing state and affix the descriptors.
1464 */
1465 rp = (file_t **)CMSG_DATA(cm);
1466 int *ofdp = (int *)CMSG_DATA(cm);
1467 for (size_t i = 0; i < nfds; i++) {
1468 file_t * const fp = *rp++;
1469 const int fd = fdp[i];
1470 atomic_dec_uint(&unp_rights);
1471 fd_set_exclose(l, fd, (flags & O_CLOEXEC) != 0);
1472 fd_affix(p, fp, fd);
1473 /*
1474 * Done with this file pointer, replace it with a fd;
1475 */
1476 *ofdp++ = fd;
1477 mutex_enter(&fp->f_lock);
1478 fp->f_msgcount--;
1479 mutex_exit(&fp->f_lock);
1480 /*
1481 * Note that fd_affix() adds a reference to the file.
1482 * The file may already have been closed by another
1483 * LWP in the process, so we must drop the reference
1484 * added by unp_internalize() with closef().
1485 */
1486 closef(fp);
1487 }
1488
1489 /*
1490 * Adjust length, in case of transition from large file_t
1491 * pointers to ints.
1492 */
1493 if (sizeof(file_t *) != sizeof(int)) {
1494 cm->cmsg_len = CMSG_LEN(nfds * sizeof(int));
1495 rights->m_len = CMSG_SPACE(nfds * sizeof(int));
1496 }
1497 out:
1498 if (__predict_false(error != 0)) {
1499 file_t **const fpp = (file_t **)CMSG_DATA(cm);
1500 for (size_t i = 0; i < nfds; i++)
1501 unp_discard_now(fpp[i]);
1502 /*
1503 * Truncate the array so that nobody will try to interpret
1504 * what is now garbage in it.
1505 */
1506 cm->cmsg_len = CMSG_LEN(0);
1507 rights->m_len = CMSG_SPACE(0);
1508 }
1509 rw_exit(&p->p_cwdi->cwdi_lock);
1510 kmem_free(fdp, nfds * sizeof(int));
1511
1512 noop:
1513 /*
1514 * Don't disclose kernel memory in the alignment space.
1515 */
1516 KASSERT(cm->cmsg_len <= rights->m_len);
1517 memset(&mtod(rights, char *)[cm->cmsg_len], 0, rights->m_len -
1518 cm->cmsg_len);
1519 return error;
1520 }
1521
1522 static int
1523 unp_internalize(struct mbuf **controlp)
1524 {
1525 filedesc_t *fdescp = curlwp->l_fd;
1526 struct mbuf *control = *controlp;
1527 struct cmsghdr *newcm, *cm = mtod(control, struct cmsghdr *);
1528 file_t **rp, **files;
1529 file_t *fp;
1530 int i, fd, *fdp;
1531 int nfds, error;
1532 u_int maxmsg;
1533
1534 error = 0;
1535 newcm = NULL;
1536
1537 /* Sanity check the control message header. */
1538 if (cm->cmsg_type != SCM_RIGHTS || cm->cmsg_level != SOL_SOCKET ||
1539 cm->cmsg_len > control->m_len ||
1540 cm->cmsg_len < CMSG_ALIGN(sizeof(*cm)))
1541 return (EINVAL);
1542
1543 /*
1544 * Verify that the file descriptors are valid, and acquire
1545 * a reference to each.
1546 */
1547 nfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm))) / sizeof(int);
1548 fdp = (int *)CMSG_DATA(cm);
1549 maxmsg = maxfiles / unp_rights_ratio;
1550 for (i = 0; i < nfds; i++) {
1551 fd = *fdp++;
1552 if (atomic_inc_uint_nv(&unp_rights) > maxmsg) {
1553 atomic_dec_uint(&unp_rights);
1554 nfds = i;
1555 error = EAGAIN;
1556 goto out;
1557 }
1558 if ((fp = fd_getfile(fd)) == NULL
1559 || fp->f_type == DTYPE_KQUEUE) {
1560 if (fp)
1561 fd_putfile(fd);
1562 atomic_dec_uint(&unp_rights);
1563 nfds = i;
1564 error = EBADF;
1565 goto out;
1566 }
1567 }
1568
1569 /* Allocate new space and copy header into it. */
1570 newcm = malloc(CMSG_SPACE(nfds * sizeof(file_t *)), M_MBUF, M_WAITOK);
1571 if (newcm == NULL) {
1572 error = E2BIG;
1573 goto out;
1574 }
1575 memcpy(newcm, cm, sizeof(struct cmsghdr));
1576 files = (file_t **)CMSG_DATA(newcm);
1577
1578 /*
1579 * Transform the file descriptors into file_t pointers, in
1580 * reverse order so that if pointers are bigger than ints, the
1581 * int won't get until we're done. No need to lock, as we have
1582 * already validated the descriptors with fd_getfile().
1583 */
1584 fdp = (int *)CMSG_DATA(cm) + nfds;
1585 rp = files + nfds;
1586 for (i = 0; i < nfds; i++) {
1587 fp = fdescp->fd_dt->dt_ff[*--fdp]->ff_file;
1588 KASSERT(fp != NULL);
1589 mutex_enter(&fp->f_lock);
1590 *--rp = fp;
1591 fp->f_count++;
1592 fp->f_msgcount++;
1593 mutex_exit(&fp->f_lock);
1594 }
1595
1596 out:
1597 /* Release descriptor references. */
1598 fdp = (int *)CMSG_DATA(cm);
1599 for (i = 0; i < nfds; i++) {
1600 fd_putfile(*fdp++);
1601 if (error != 0) {
1602 atomic_dec_uint(&unp_rights);
1603 }
1604 }
1605
1606 if (error == 0) {
1607 if (control->m_flags & M_EXT) {
1608 m_freem(control);
1609 *controlp = control = m_get(M_WAIT, MT_CONTROL);
1610 }
1611 MEXTADD(control, newcm, CMSG_SPACE(nfds * sizeof(file_t *)),
1612 M_MBUF, NULL, NULL);
1613 cm = newcm;
1614 /*
1615 * Adjust message & mbuf to note amount of space
1616 * actually used.
1617 */
1618 cm->cmsg_len = CMSG_LEN(nfds * sizeof(file_t *));
1619 control->m_len = CMSG_SPACE(nfds * sizeof(file_t *));
1620 }
1621
1622 return error;
1623 }
1624
1625 struct mbuf *
1626 unp_addsockcred(struct lwp *l, struct mbuf *control)
1627 {
1628 struct sockcred *sc;
1629 struct mbuf *m;
1630 void *p;
1631
1632 m = sbcreatecontrol1(&p, SOCKCREDSIZE(kauth_cred_ngroups(l->l_cred)),
1633 SCM_CREDS, SOL_SOCKET, M_WAITOK);
1634 if (m == NULL)
1635 return control;
1636
1637 sc = p;
1638 sc->sc_pid = l->l_proc->p_pid;
1639 sc->sc_uid = kauth_cred_getuid(l->l_cred);
1640 sc->sc_euid = kauth_cred_geteuid(l->l_cred);
1641 sc->sc_gid = kauth_cred_getgid(l->l_cred);
1642 sc->sc_egid = kauth_cred_getegid(l->l_cred);
1643 sc->sc_ngroups = kauth_cred_ngroups(l->l_cred);
1644
1645 for (int i = 0; i < sc->sc_ngroups; i++)
1646 sc->sc_groups[i] = kauth_cred_group(l->l_cred, i);
1647
1648 return m_add(control, m);
1649 }
1650
1651 /*
1652 * Do a mark-sweep GC of files in the system, to free up any which are
1653 * caught in flight to an about-to-be-closed socket. Additionally,
1654 * process deferred file closures.
1655 */
1656 static void
1657 unp_gc(file_t *dp)
1658 {
1659 extern struct domain unixdomain;
1660 file_t *fp, *np;
1661 struct socket *so, *so1;
1662 u_int i, oflags, rflags;
1663 bool didwork;
1664
1665 KASSERT(curlwp == unp_thread_lwp);
1666 KASSERT(mutex_owned(&filelist_lock));
1667
1668 /*
1669 * First, process deferred file closures.
1670 */
1671 while (!SLIST_EMPTY(&unp_thread_discard)) {
1672 fp = SLIST_FIRST(&unp_thread_discard);
1673 KASSERT(fp->f_unpcount > 0);
1674 KASSERT(fp->f_count > 0);
1675 KASSERT(fp->f_msgcount > 0);
1676 KASSERT(fp->f_count >= fp->f_unpcount);
1677 KASSERT(fp->f_count >= fp->f_msgcount);
1678 KASSERT(fp->f_msgcount >= fp->f_unpcount);
1679 SLIST_REMOVE_HEAD(&unp_thread_discard, f_unplist);
1680 i = fp->f_unpcount;
1681 fp->f_unpcount = 0;
1682 mutex_exit(&filelist_lock);
1683 for (; i != 0; i--) {
1684 unp_discard_now(fp);
1685 }
1686 mutex_enter(&filelist_lock);
1687 }
1688
1689 /*
1690 * Clear mark bits. Ensure that we don't consider new files
1691 * entering the file table during this loop (they will not have
1692 * FSCAN set).
1693 */
1694 unp_defer = 0;
1695 LIST_FOREACH(fp, &filehead, f_list) {
1696 for (oflags = fp->f_flag;; oflags = rflags) {
1697 rflags = atomic_cas_uint(&fp->f_flag, oflags,
1698 (oflags | FSCAN) & ~(FMARK|FDEFER));
1699 if (__predict_true(oflags == rflags)) {
1700 break;
1701 }
1702 }
1703 }
1704
1705 /*
1706 * Iterate over the set of sockets, marking ones believed (based on
1707 * refcount) to be referenced from a process, and marking for rescan
1708 * sockets which are queued on a socket. Recan continues descending
1709 * and searching for sockets referenced by sockets (FDEFER), until
1710 * there are no more socket->socket references to be discovered.
1711 */
1712 do {
1713 didwork = false;
1714 for (fp = LIST_FIRST(&filehead); fp != NULL; fp = np) {
1715 KASSERT(mutex_owned(&filelist_lock));
1716 np = LIST_NEXT(fp, f_list);
1717 mutex_enter(&fp->f_lock);
1718 if ((fp->f_flag & FDEFER) != 0) {
1719 atomic_and_uint(&fp->f_flag, ~FDEFER);
1720 unp_defer--;
1721 if (fp->f_count == 0) {
1722 /*
1723 * XXX: closef() doesn't pay attention
1724 * to FDEFER
1725 */
1726 mutex_exit(&fp->f_lock);
1727 continue;
1728 }
1729 } else {
1730 if (fp->f_count == 0 ||
1731 (fp->f_flag & FMARK) != 0 ||
1732 fp->f_count == fp->f_msgcount ||
1733 fp->f_unpcount != 0) {
1734 mutex_exit(&fp->f_lock);
1735 continue;
1736 }
1737 }
1738 atomic_or_uint(&fp->f_flag, FMARK);
1739
1740 if (fp->f_type != DTYPE_SOCKET ||
1741 (so = fp->f_socket) == NULL ||
1742 so->so_proto->pr_domain != &unixdomain ||
1743 (so->so_proto->pr_flags & PR_RIGHTS) == 0) {
1744 mutex_exit(&fp->f_lock);
1745 continue;
1746 }
1747
1748 /* Gain file ref, mark our position, and unlock. */
1749 didwork = true;
1750 LIST_INSERT_AFTER(fp, dp, f_list);
1751 fp->f_count++;
1752 mutex_exit(&fp->f_lock);
1753 mutex_exit(&filelist_lock);
1754
1755 /*
1756 * Mark files referenced from sockets queued on the
1757 * accept queue as well.
1758 */
1759 solock(so);
1760 unp_scan(so->so_rcv.sb_mb, unp_mark, 0);
1761 if ((so->so_options & SO_ACCEPTCONN) != 0) {
1762 TAILQ_FOREACH(so1, &so->so_q0, so_qe) {
1763 unp_scan(so1->so_rcv.sb_mb, unp_mark, 0);
1764 }
1765 TAILQ_FOREACH(so1, &so->so_q, so_qe) {
1766 unp_scan(so1->so_rcv.sb_mb, unp_mark, 0);
1767 }
1768 }
1769 sounlock(so);
1770
1771 /* Re-lock and restart from where we left off. */
1772 closef(fp);
1773 mutex_enter(&filelist_lock);
1774 np = LIST_NEXT(dp, f_list);
1775 LIST_REMOVE(dp, f_list);
1776 }
1777 /*
1778 * Bail early if we did nothing in the loop above. Could
1779 * happen because of concurrent activity causing unp_defer
1780 * to get out of sync.
1781 */
1782 } while (unp_defer != 0 && didwork);
1783
1784 /*
1785 * Sweep pass.
1786 *
1787 * We grab an extra reference to each of the files that are
1788 * not otherwise accessible and then free the rights that are
1789 * stored in messages on them.
1790 */
1791 for (fp = LIST_FIRST(&filehead); fp != NULL; fp = np) {
1792 KASSERT(mutex_owned(&filelist_lock));
1793 np = LIST_NEXT(fp, f_list);
1794 mutex_enter(&fp->f_lock);
1795
1796 /*
1797 * Ignore non-sockets.
1798 * Ignore dead sockets, or sockets with pending close.
1799 * Ignore sockets obviously referenced elsewhere.
1800 * Ignore sockets marked as referenced by our scan.
1801 * Ignore new sockets that did not exist during the scan.
1802 */
1803 if (fp->f_type != DTYPE_SOCKET ||
1804 fp->f_count == 0 || fp->f_unpcount != 0 ||
1805 fp->f_count != fp->f_msgcount ||
1806 (fp->f_flag & (FMARK | FSCAN)) != FSCAN) {
1807 mutex_exit(&fp->f_lock);
1808 continue;
1809 }
1810
1811 /* Gain file ref, mark our position, and unlock. */
1812 LIST_INSERT_AFTER(fp, dp, f_list);
1813 fp->f_count++;
1814 mutex_exit(&fp->f_lock);
1815 mutex_exit(&filelist_lock);
1816
1817 /*
1818 * Flush all data from the socket's receive buffer.
1819 * This will cause files referenced only by the
1820 * socket to be queued for close.
1821 */
1822 so = fp->f_socket;
1823 solock(so);
1824 sorflush(so);
1825 sounlock(so);
1826
1827 /* Re-lock and restart from where we left off. */
1828 closef(fp);
1829 mutex_enter(&filelist_lock);
1830 np = LIST_NEXT(dp, f_list);
1831 LIST_REMOVE(dp, f_list);
1832 }
1833 }
1834
1835 /*
1836 * Garbage collector thread. While SCM_RIGHTS messages are in transit,
1837 * wake once per second to garbage collect. Run continually while we
1838 * have deferred closes to process.
1839 */
1840 static void
1841 unp_thread(void *cookie)
1842 {
1843 file_t *dp;
1844
1845 /* Allocate a dummy file for our scans. */
1846 if ((dp = fgetdummy()) == NULL) {
1847 panic("unp_thread");
1848 }
1849
1850 mutex_enter(&filelist_lock);
1851 for (;;) {
1852 KASSERT(mutex_owned(&filelist_lock));
1853 if (SLIST_EMPTY(&unp_thread_discard)) {
1854 if (unp_rights != 0) {
1855 (void)cv_timedwait(&unp_thread_cv,
1856 &filelist_lock, hz);
1857 } else {
1858 cv_wait(&unp_thread_cv, &filelist_lock);
1859 }
1860 }
1861 unp_gc(dp);
1862 }
1863 /* NOTREACHED */
1864 }
1865
1866 /*
1867 * Kick the garbage collector into action if there is something for
1868 * it to process.
1869 */
1870 static void
1871 unp_thread_kick(void)
1872 {
1873
1874 if (!SLIST_EMPTY(&unp_thread_discard) || unp_rights != 0) {
1875 mutex_enter(&filelist_lock);
1876 cv_signal(&unp_thread_cv);
1877 mutex_exit(&filelist_lock);
1878 }
1879 }
1880
1881 void
1882 unp_dispose(struct mbuf *m)
1883 {
1884
1885 if (m)
1886 unp_scan(m, unp_discard_later, 1);
1887 }
1888
1889 void
1890 unp_scan(struct mbuf *m0, void (*op)(file_t *), int discard)
1891 {
1892 struct mbuf *m;
1893 file_t **rp, *fp;
1894 struct cmsghdr *cm;
1895 int i, qfds;
1896
1897 while (m0) {
1898 for (m = m0; m; m = m->m_next) {
1899 if (m->m_type != MT_CONTROL ||
1900 m->m_len < sizeof(*cm)) {
1901 continue;
1902 }
1903 cm = mtod(m, struct cmsghdr *);
1904 if (cm->cmsg_level != SOL_SOCKET ||
1905 cm->cmsg_type != SCM_RIGHTS)
1906 continue;
1907 qfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm)))
1908 / sizeof(file_t *);
1909 rp = (file_t **)CMSG_DATA(cm);
1910 for (i = 0; i < qfds; i++) {
1911 fp = *rp;
1912 if (discard) {
1913 *rp = 0;
1914 }
1915 (*op)(fp);
1916 rp++;
1917 }
1918 }
1919 m0 = m0->m_nextpkt;
1920 }
1921 }
1922
1923 void
1924 unp_mark(file_t *fp)
1925 {
1926
1927 if (fp == NULL)
1928 return;
1929
1930 /* If we're already deferred, don't screw up the defer count */
1931 mutex_enter(&fp->f_lock);
1932 if (fp->f_flag & (FMARK | FDEFER)) {
1933 mutex_exit(&fp->f_lock);
1934 return;
1935 }
1936
1937 /*
1938 * Minimize the number of deferrals... Sockets are the only type of
1939 * file which can hold references to another file, so just mark
1940 * other files, and defer unmarked sockets for the next pass.
1941 */
1942 if (fp->f_type == DTYPE_SOCKET) {
1943 unp_defer++;
1944 KASSERT(fp->f_count != 0);
1945 atomic_or_uint(&fp->f_flag, FDEFER);
1946 } else {
1947 atomic_or_uint(&fp->f_flag, FMARK);
1948 }
1949 mutex_exit(&fp->f_lock);
1950 }
1951
1952 static void
1953 unp_discard_now(file_t *fp)
1954 {
1955
1956 if (fp == NULL)
1957 return;
1958
1959 KASSERT(fp->f_count > 0);
1960 KASSERT(fp->f_msgcount > 0);
1961
1962 mutex_enter(&fp->f_lock);
1963 fp->f_msgcount--;
1964 mutex_exit(&fp->f_lock);
1965 atomic_dec_uint(&unp_rights);
1966 (void)closef(fp);
1967 }
1968
1969 static void
1970 unp_discard_later(file_t *fp)
1971 {
1972
1973 if (fp == NULL)
1974 return;
1975
1976 KASSERT(fp->f_count > 0);
1977 KASSERT(fp->f_msgcount > 0);
1978
1979 mutex_enter(&filelist_lock);
1980 if (fp->f_unpcount++ == 0) {
1981 SLIST_INSERT_HEAD(&unp_thread_discard, fp, f_unplist);
1982 }
1983 mutex_exit(&filelist_lock);
1984 }
1985
1986 void
1987 unp_sysctl_create(struct sysctllog **clog)
1988 {
1989 sysctl_createv(clog, 0, NULL, NULL,
1990 CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1991 CTLTYPE_LONG, "sendspace",
1992 SYSCTL_DESCR("Default stream send space"),
1993 NULL, 0, &unpst_sendspace, 0,
1994 CTL_NET, PF_LOCAL, SOCK_STREAM, CTL_CREATE, CTL_EOL);
1995 sysctl_createv(clog, 0, NULL, NULL,
1996 CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1997 CTLTYPE_LONG, "recvspace",
1998 SYSCTL_DESCR("Default stream recv space"),
1999 NULL, 0, &unpst_recvspace, 0,
2000 CTL_NET, PF_LOCAL, SOCK_STREAM, CTL_CREATE, CTL_EOL);
2001 sysctl_createv(clog, 0, NULL, NULL,
2002 CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2003 CTLTYPE_LONG, "sendspace",
2004 SYSCTL_DESCR("Default datagram send space"),
2005 NULL, 0, &unpdg_sendspace, 0,
2006 CTL_NET, PF_LOCAL, SOCK_DGRAM, CTL_CREATE, CTL_EOL);
2007 sysctl_createv(clog, 0, NULL, NULL,
2008 CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2009 CTLTYPE_LONG, "recvspace",
2010 SYSCTL_DESCR("Default datagram recv space"),
2011 NULL, 0, &unpdg_recvspace, 0,
2012 CTL_NET, PF_LOCAL, SOCK_DGRAM, CTL_CREATE, CTL_EOL);
2013 sysctl_createv(clog, 0, NULL, NULL,
2014 CTLFLAG_PERMANENT|CTLFLAG_READONLY,
2015 CTLTYPE_INT, "inflight",
2016 SYSCTL_DESCR("File descriptors in flight"),
2017 NULL, 0, &unp_rights, 0,
2018 CTL_NET, PF_LOCAL, CTL_CREATE, CTL_EOL);
2019 sysctl_createv(clog, 0, NULL, NULL,
2020 CTLFLAG_PERMANENT|CTLFLAG_READONLY,
2021 CTLTYPE_INT, "deferred",
2022 SYSCTL_DESCR("File descriptors deferred for close"),
2023 NULL, 0, &unp_defer, 0,
2024 CTL_NET, PF_LOCAL, CTL_CREATE, CTL_EOL);
2025 }
2026
2027 const struct pr_usrreqs unp_usrreqs = {
2028 .pr_attach = unp_attach,
2029 .pr_detach = unp_detach,
2030 .pr_accept = unp_accept,
2031 .pr_bind = unp_bind,
2032 .pr_listen = unp_listen,
2033 .pr_connect = unp_connect,
2034 .pr_connect2 = unp_connect2,
2035 .pr_disconnect = unp_disconnect,
2036 .pr_shutdown = unp_shutdown,
2037 .pr_abort = unp_abort,
2038 .pr_ioctl = unp_ioctl,
2039 .pr_stat = unp_stat,
2040 .pr_peeraddr = unp_peeraddr,
2041 .pr_sockaddr = unp_sockaddr,
2042 .pr_rcvd = unp_rcvd,
2043 .pr_recvoob = unp_recvoob,
2044 .pr_send = unp_send,
2045 .pr_sendoob = unp_sendoob,
2046 };
2047