uipc_usrreq.c revision 1.78 1 /* $NetBSD: uipc_usrreq.c,v 1.78 2004/05/22 22:52:13 jonathan Exp $ */
2
3 /*-
4 * Copyright (c) 1998, 2000, 2004 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.
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 * 3. All advertising materials mentioning features or use of this software
20 * must display the following acknowledgement:
21 * This product includes software developed by the NetBSD
22 * Foundation, Inc. and its contributors.
23 * 4. Neither the name of The NetBSD Foundation nor the names of its
24 * contributors may be used to endorse or promote products derived
25 * from this software without specific prior written permission.
26 *
27 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
28 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
29 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
30 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
31 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
32 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
33 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
34 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
35 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
36 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
37 * POSSIBILITY OF SUCH DAMAGE.
38 */
39
40 /*
41 * Copyright (c) 1982, 1986, 1989, 1991, 1993
42 * The Regents of the University of California. All rights reserved.
43 *
44 * Redistribution and use in source and binary forms, with or without
45 * modification, are permitted provided that the following conditions
46 * are met:
47 * 1. Redistributions of source code must retain the above copyright
48 * notice, this list of conditions and the following disclaimer.
49 * 2. Redistributions in binary form must reproduce the above copyright
50 * notice, this list of conditions and the following disclaimer in the
51 * documentation and/or other materials provided with the distribution.
52 * 3. Neither the name of the University nor the names of its contributors
53 * may be used to endorse or promote products derived from this software
54 * without specific prior written permission.
55 *
56 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
57 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
58 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
59 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
60 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
61 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
62 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
63 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
64 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
65 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
66 * SUCH DAMAGE.
67 *
68 * @(#)uipc_usrreq.c 8.9 (Berkeley) 5/14/95
69 */
70
71 /*
72 * Copyright (c) 1997 Christopher G. Demetriou. All rights reserved.
73 *
74 * Redistribution and use in source and binary forms, with or without
75 * modification, are permitted provided that the following conditions
76 * are met:
77 * 1. Redistributions of source code must retain the above copyright
78 * notice, this list of conditions and the following disclaimer.
79 * 2. Redistributions in binary form must reproduce the above copyright
80 * notice, this list of conditions and the following disclaimer in the
81 * documentation and/or other materials provided with the distribution.
82 * 3. All advertising materials mentioning features or use of this software
83 * must display the following acknowledgement:
84 * This product includes software developed by the University of
85 * California, Berkeley and its contributors.
86 * 4. Neither the name of the University nor the names of its contributors
87 * may be used to endorse or promote products derived from this software
88 * without specific prior written permission.
89 *
90 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
91 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
92 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
93 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
94 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
95 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
96 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
97 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
98 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
99 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
100 * SUCH DAMAGE.
101 *
102 * @(#)uipc_usrreq.c 8.9 (Berkeley) 5/14/95
103 */
104
105 #include <sys/cdefs.h>
106 __KERNEL_RCSID(0, "$NetBSD: uipc_usrreq.c,v 1.78 2004/05/22 22:52:13 jonathan Exp $");
107
108 #include <sys/param.h>
109 #include <sys/systm.h>
110 #include <sys/proc.h>
111 #include <sys/filedesc.h>
112 #include <sys/domain.h>
113 #include <sys/protosw.h>
114 #include <sys/socket.h>
115 #include <sys/socketvar.h>
116 #include <sys/unpcb.h>
117 #include <sys/un.h>
118 #include <sys/namei.h>
119 #include <sys/vnode.h>
120 #include <sys/file.h>
121 #include <sys/stat.h>
122 #include <sys/mbuf.h>
123
124 /*
125 * Unix communications domain.
126 *
127 * TODO:
128 * SEQPACKET, RDM
129 * rethink name space problems
130 * need a proper out-of-band
131 */
132 const struct sockaddr_un sun_noname = { sizeof(sun_noname), AF_LOCAL };
133 ino_t unp_ino; /* prototype for fake inode numbers */
134
135 struct mbuf *unp_addsockcred(struct proc *, struct mbuf *);
136
137 int
138 unp_output(struct mbuf *m, struct mbuf *control, struct unpcb *unp,
139 struct proc *p)
140 {
141 struct socket *so2;
142 const struct sockaddr_un *sun;
143
144 so2 = unp->unp_conn->unp_socket;
145 if (unp->unp_addr)
146 sun = unp->unp_addr;
147 else
148 sun = &sun_noname;
149 if (unp->unp_conn->unp_flags & UNP_WANTCRED)
150 control = unp_addsockcred(p, control);
151 if (sbappendaddr(&so2->so_rcv, (struct sockaddr *)sun, m,
152 control) == 0) {
153 m_freem(control);
154 m_freem(m);
155 return (ENOBUFS);
156 } else {
157 sorwakeup(so2);
158 return (0);
159 }
160 }
161
162 void
163 unp_setsockaddr(struct unpcb *unp, struct mbuf *nam)
164 {
165 const struct sockaddr_un *sun;
166
167 if (unp->unp_addr)
168 sun = unp->unp_addr;
169 else
170 sun = &sun_noname;
171 nam->m_len = sun->sun_len;
172 if (nam->m_len > MLEN)
173 MEXTMALLOC(nam, nam->m_len, M_WAITOK);
174 memcpy(mtod(nam, caddr_t), sun, (size_t)nam->m_len);
175 }
176
177 void
178 unp_setpeeraddr(struct unpcb *unp, struct mbuf *nam)
179 {
180 const struct sockaddr_un *sun;
181
182 if (unp->unp_conn && unp->unp_conn->unp_addr)
183 sun = unp->unp_conn->unp_addr;
184 else
185 sun = &sun_noname;
186 nam->m_len = sun->sun_len;
187 if (nam->m_len > MLEN)
188 MEXTMALLOC(nam, nam->m_len, M_WAITOK);
189 memcpy(mtod(nam, caddr_t), sun, (size_t)nam->m_len);
190 }
191
192 /*ARGSUSED*/
193 int
194 uipc_usrreq(struct socket *so, int req, struct mbuf *m, struct mbuf *nam,
195 struct mbuf *control, struct proc *p)
196 {
197 struct unpcb *unp = sotounpcb(so);
198 struct socket *so2;
199 u_int newhiwat;
200 int error = 0;
201
202 if (req == PRU_CONTROL)
203 return (EOPNOTSUPP);
204
205 #ifdef DIAGNOSTIC
206 if (req != PRU_SEND && req != PRU_SENDOOB && control)
207 panic("uipc_usrreq: unexpected control mbuf");
208 #endif
209 if (unp == 0 && req != PRU_ATTACH) {
210 error = EINVAL;
211 goto release;
212 }
213
214 switch (req) {
215
216 case PRU_ATTACH:
217 if (unp != 0) {
218 error = EISCONN;
219 break;
220 }
221 error = unp_attach(so);
222 break;
223
224 case PRU_DETACH:
225 unp_detach(unp);
226 break;
227
228 case PRU_BIND:
229 error = unp_bind(unp, nam, p);
230 break;
231
232 case PRU_LISTEN:
233 if (unp->unp_vnode == 0)
234 error = EINVAL;
235 break;
236
237 case PRU_CONNECT:
238 error = unp_connect(so, nam, p);
239 break;
240
241 case PRU_CONNECT2:
242 error = unp_connect2(so, (struct socket *)nam, PRU_CONNECT2);
243 break;
244
245 case PRU_DISCONNECT:
246 unp_disconnect(unp);
247 break;
248
249 case PRU_ACCEPT:
250 unp_setpeeraddr(unp, nam);
251 /*
252 * Mark the initiating STREAM socket as connected *ONLY*
253 * after it's been accepted. This prevents a client from
254 * overrunning a server and receiving ECONNREFUSED.
255 */
256 if (unp->unp_conn != NULL &&
257 (unp->unp_conn->unp_socket->so_state & SS_ISCONNECTING))
258 soisconnected(unp->unp_conn->unp_socket);
259 break;
260
261 case PRU_SHUTDOWN:
262 socantsendmore(so);
263 unp_shutdown(unp);
264 break;
265
266 case PRU_RCVD:
267 switch (so->so_type) {
268
269 case SOCK_DGRAM:
270 panic("uipc 1");
271 /*NOTREACHED*/
272
273 case SOCK_STREAM:
274 #define rcv (&so->so_rcv)
275 #define snd (&so2->so_snd)
276 if (unp->unp_conn == 0)
277 break;
278 so2 = unp->unp_conn->unp_socket;
279 /*
280 * Adjust backpressure on sender
281 * and wakeup any waiting to write.
282 */
283 snd->sb_mbmax += unp->unp_mbcnt - rcv->sb_mbcnt;
284 unp->unp_mbcnt = rcv->sb_mbcnt;
285 newhiwat = snd->sb_hiwat + unp->unp_cc - rcv->sb_cc;
286 (void)chgsbsize(so2->so_uid,
287 &snd->sb_hiwat, newhiwat, RLIM_INFINITY);
288 unp->unp_cc = rcv->sb_cc;
289 sowwakeup(so2);
290 #undef snd
291 #undef rcv
292 break;
293
294 default:
295 panic("uipc 2");
296 }
297 break;
298
299 case PRU_SEND:
300 /*
301 * Note: unp_internalize() rejects any control message
302 * other than SCM_RIGHTS, and only allows one. This
303 * has the side-effect of preventing a caller from
304 * forging SCM_CREDS.
305 */
306 if (control && (error = unp_internalize(control, p)))
307 break;
308 switch (so->so_type) {
309
310 case SOCK_DGRAM: {
311 if (nam) {
312 if ((so->so_state & SS_ISCONNECTED) != 0) {
313 error = EISCONN;
314 goto die;
315 }
316 error = unp_connect(so, nam, p);
317 if (error) {
318 die:
319 m_freem(control);
320 m_freem(m);
321 break;
322 }
323 } else {
324 if ((so->so_state & SS_ISCONNECTED) == 0) {
325 error = ENOTCONN;
326 goto die;
327 }
328 }
329 error = unp_output(m, control, unp, p);
330 if (nam)
331 unp_disconnect(unp);
332 break;
333 }
334
335 case SOCK_STREAM:
336 #define rcv (&so2->so_rcv)
337 #define snd (&so->so_snd)
338 if (unp->unp_conn == 0)
339 panic("uipc 3");
340 so2 = unp->unp_conn->unp_socket;
341 if (unp->unp_conn->unp_flags & UNP_WANTCRED) {
342 /*
343 * Credentials are passed only once on
344 * SOCK_STREAM.
345 */
346 unp->unp_conn->unp_flags &= ~UNP_WANTCRED;
347 control = unp_addsockcred(p, control);
348 }
349 /*
350 * Send to paired receive port, and then reduce
351 * send buffer hiwater marks to maintain backpressure.
352 * Wake up readers.
353 */
354 if (control) {
355 if (sbappendcontrol(rcv, m, control) == 0)
356 m_freem(control);
357 } else
358 sbappend(rcv, m);
359 snd->sb_mbmax -=
360 rcv->sb_mbcnt - unp->unp_conn->unp_mbcnt;
361 unp->unp_conn->unp_mbcnt = rcv->sb_mbcnt;
362 newhiwat = snd->sb_hiwat -
363 (rcv->sb_cc - unp->unp_conn->unp_cc);
364 (void)chgsbsize(so->so_uid,
365 &snd->sb_hiwat, newhiwat, RLIM_INFINITY);
366 unp->unp_conn->unp_cc = rcv->sb_cc;
367 sorwakeup(so2);
368 #undef snd
369 #undef rcv
370 break;
371
372 default:
373 panic("uipc 4");
374 }
375 break;
376
377 case PRU_ABORT:
378 unp_drop(unp, ECONNABORTED);
379
380 #ifdef DIAGNOSTIC
381 if (so->so_pcb == 0)
382 panic("uipc 5: drop killed pcb");
383 #endif
384 unp_detach(unp);
385 break;
386
387 case PRU_SENSE:
388 ((struct stat *) m)->st_blksize = so->so_snd.sb_hiwat;
389 if (so->so_type == SOCK_STREAM && unp->unp_conn != 0) {
390 so2 = unp->unp_conn->unp_socket;
391 ((struct stat *) m)->st_blksize += so2->so_rcv.sb_cc;
392 }
393 ((struct stat *) m)->st_dev = NODEV;
394 if (unp->unp_ino == 0)
395 unp->unp_ino = unp_ino++;
396 ((struct stat *) m)->st_atimespec =
397 ((struct stat *) m)->st_mtimespec =
398 ((struct stat *) m)->st_ctimespec = unp->unp_ctime;
399 ((struct stat *) m)->st_ino = unp->unp_ino;
400 return (0);
401
402 case PRU_RCVOOB:
403 error = EOPNOTSUPP;
404 break;
405
406 case PRU_SENDOOB:
407 m_freem(control);
408 m_freem(m);
409 error = EOPNOTSUPP;
410 break;
411
412 case PRU_SOCKADDR:
413 unp_setsockaddr(unp, nam);
414 break;
415
416 case PRU_PEERADDR:
417 unp_setpeeraddr(unp, nam);
418 break;
419
420 default:
421 panic("piusrreq");
422 }
423
424 release:
425 return (error);
426 }
427
428 /*
429 * Unix domain socket option processing.
430 */
431 int
432 uipc_ctloutput(int op, struct socket *so, int level, int optname,
433 struct mbuf **mp)
434 {
435 struct unpcb *unp = sotounpcb(so);
436 struct mbuf *m = *mp;
437 int optval = 0, error = 0;
438
439 if (level != 0) {
440 error = EINVAL;
441 if (op == PRCO_SETOPT && m)
442 (void) m_free(m);
443 } else switch (op) {
444
445 case PRCO_SETOPT:
446 switch (optname) {
447 case LOCAL_CREDS:
448 case LOCAL_CONNWAIT:
449 if (m == NULL || m->m_len != sizeof(int))
450 error = EINVAL;
451 else {
452 optval = *mtod(m, int *);
453 switch (optname) {
454 #define OPTSET(bit) \
455 if (optval) \
456 unp->unp_flags |= (bit); \
457 else \
458 unp->unp_flags &= ~(bit);
459
460 case LOCAL_CREDS:
461 OPTSET(UNP_WANTCRED);
462 break;
463 case LOCAL_CONNWAIT:
464 OPTSET(UNP_CONNWAIT);
465 break;
466 }
467 }
468 break;
469 #undef OPTSET
470
471 default:
472 error = ENOPROTOOPT;
473 break;
474 }
475 if (m)
476 (void) m_free(m);
477 break;
478
479 case PRCO_GETOPT:
480 switch (optname) {
481 case LOCAL_CREDS:
482 *mp = m = m_get(M_WAIT, MT_SOOPTS);
483 m->m_len = sizeof(int);
484 switch (optname) {
485
486 #define OPTBIT(bit) (unp->unp_flags & (bit) ? 1 : 0)
487
488 case LOCAL_CREDS:
489 optval = OPTBIT(UNP_WANTCRED);
490 break;
491 }
492 *mtod(m, int *) = optval;
493 break;
494 #undef OPTBIT
495
496 default:
497 error = ENOPROTOOPT;
498 break;
499 }
500 break;
501 }
502 return (error);
503 }
504
505 /*
506 * Both send and receive buffers are allocated PIPSIZ bytes of buffering
507 * for stream sockets, although the total for sender and receiver is
508 * actually only PIPSIZ.
509 * Datagram sockets really use the sendspace as the maximum datagram size,
510 * and don't really want to reserve the sendspace. Their recvspace should
511 * be large enough for at least one max-size datagram plus address.
512 */
513 #define PIPSIZ 4096
514 u_long unpst_sendspace = PIPSIZ;
515 u_long unpst_recvspace = PIPSIZ;
516 u_long unpdg_sendspace = 2*1024; /* really max datagram size */
517 u_long unpdg_recvspace = 4*1024;
518
519 int unp_rights; /* file descriptors in flight */
520
521 int
522 unp_attach(struct socket *so)
523 {
524 struct unpcb *unp;
525 struct timeval tv;
526 int error;
527
528 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
529 switch (so->so_type) {
530
531 case SOCK_STREAM:
532 error = soreserve(so, unpst_sendspace, unpst_recvspace);
533 break;
534
535 case SOCK_DGRAM:
536 error = soreserve(so, unpdg_sendspace, unpdg_recvspace);
537 break;
538
539 default:
540 panic("unp_attach");
541 }
542 if (error)
543 return (error);
544 }
545 unp = malloc(sizeof(*unp), M_PCB, M_NOWAIT);
546 if (unp == NULL)
547 return (ENOBUFS);
548 memset((caddr_t)unp, 0, sizeof(*unp));
549 unp->unp_socket = so;
550 so->so_pcb = unp;
551 microtime(&tv);
552 TIMEVAL_TO_TIMESPEC(&tv, &unp->unp_ctime);
553 return (0);
554 }
555
556 void
557 unp_detach(struct unpcb *unp)
558 {
559
560 if (unp->unp_vnode) {
561 unp->unp_vnode->v_socket = 0;
562 vrele(unp->unp_vnode);
563 unp->unp_vnode = 0;
564 }
565 if (unp->unp_conn)
566 unp_disconnect(unp);
567 while (unp->unp_refs)
568 unp_drop(unp->unp_refs, ECONNRESET);
569 soisdisconnected(unp->unp_socket);
570 unp->unp_socket->so_pcb = 0;
571 if (unp->unp_addr)
572 free(unp->unp_addr, M_SONAME);
573 if (unp_rights) {
574 /*
575 * Normally the receive buffer is flushed later,
576 * in sofree, but if our receive buffer holds references
577 * to descriptors that are now garbage, we will dispose
578 * of those descriptor references after the garbage collector
579 * gets them (resulting in a "panic: closef: count < 0").
580 */
581 sorflush(unp->unp_socket);
582 free(unp, M_PCB);
583 unp_gc();
584 } else
585 free(unp, M_PCB);
586 }
587
588 int
589 unp_bind(struct unpcb *unp, struct mbuf *nam, struct proc *p)
590 {
591 struct sockaddr_un *sun;
592 struct vnode *vp;
593 struct mount *mp;
594 struct vattr vattr;
595 size_t addrlen;
596 int error;
597 struct nameidata nd;
598
599 if (unp->unp_vnode != 0)
600 return (EINVAL);
601
602 /*
603 * Allocate the new sockaddr. We have to allocate one
604 * extra byte so that we can ensure that the pathname
605 * is nul-terminated.
606 */
607 addrlen = nam->m_len + 1;
608 sun = malloc(addrlen, M_SONAME, M_WAITOK);
609 m_copydata(nam, 0, nam->m_len, (caddr_t)sun);
610 *(((char *)sun) + nam->m_len) = '\0';
611
612 restart:
613 NDINIT(&nd, CREATE, FOLLOW | LOCKPARENT, UIO_SYSSPACE,
614 sun->sun_path, p);
615
616 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
617 if ((error = namei(&nd)) != 0)
618 goto bad;
619 vp = nd.ni_vp;
620 if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
621 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
622 if (nd.ni_dvp == vp)
623 vrele(nd.ni_dvp);
624 else
625 vput(nd.ni_dvp);
626 vrele(vp);
627 if (vp != NULL) {
628 error = EADDRINUSE;
629 goto bad;
630 }
631 error = vn_start_write(NULL, &mp,
632 V_WAIT | V_SLEEPONLY | V_PCATCH);
633 if (error)
634 goto bad;
635 goto restart;
636 }
637 VATTR_NULL(&vattr);
638 vattr.va_type = VSOCK;
639 vattr.va_mode = ACCESSPERMS;
640 VOP_LEASE(nd.ni_dvp, p, p->p_ucred, LEASE_WRITE);
641 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
642 vn_finished_write(mp, 0);
643 if (error)
644 goto bad;
645 vp = nd.ni_vp;
646 vp->v_socket = unp->unp_socket;
647 unp->unp_vnode = vp;
648 unp->unp_addrlen = addrlen;
649 unp->unp_addr = sun;
650 VOP_UNLOCK(vp, 0);
651 return (0);
652
653 bad:
654 free(sun, M_SONAME);
655 return (error);
656 }
657
658 int
659 unp_connect(struct socket *so, struct mbuf *nam, struct proc *p)
660 {
661 struct sockaddr_un *sun;
662 struct vnode *vp;
663 struct socket *so2, *so3;
664 struct unpcb *unp2, *unp3;
665 size_t addrlen;
666 int error;
667 struct nameidata nd;
668
669 /*
670 * Allocate a temporary sockaddr. We have to allocate one extra
671 * byte so that we can ensure that the pathname is nul-terminated.
672 * When we establish the connection, we copy the other PCB's
673 * sockaddr to our own.
674 */
675 addrlen = nam->m_len + 1;
676 sun = malloc(addrlen, M_SONAME, M_WAITOK);
677 m_copydata(nam, 0, nam->m_len, (caddr_t)sun);
678 *(((char *)sun) + nam->m_len) = '\0';
679
680 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, sun->sun_path, p);
681
682 if ((error = namei(&nd)) != 0)
683 goto bad2;
684 vp = nd.ni_vp;
685 if (vp->v_type != VSOCK) {
686 error = ENOTSOCK;
687 goto bad;
688 }
689 if ((error = VOP_ACCESS(vp, VWRITE, p->p_ucred, p)) != 0)
690 goto bad;
691 so2 = vp->v_socket;
692 if (so2 == 0) {
693 error = ECONNREFUSED;
694 goto bad;
695 }
696 if (so->so_type != so2->so_type) {
697 error = EPROTOTYPE;
698 goto bad;
699 }
700 if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
701 if ((so2->so_options & SO_ACCEPTCONN) == 0 ||
702 (so3 = sonewconn(so2, 0)) == 0) {
703 error = ECONNREFUSED;
704 goto bad;
705 }
706 unp2 = sotounpcb(so2);
707 unp3 = sotounpcb(so3);
708 if (unp2->unp_addr) {
709 unp3->unp_addr = malloc(unp2->unp_addrlen,
710 M_SONAME, M_WAITOK);
711 memcpy(unp3->unp_addr, unp2->unp_addr,
712 unp2->unp_addrlen);
713 unp3->unp_addrlen = unp2->unp_addrlen;
714 }
715 unp3->unp_flags = unp2->unp_flags;
716 so2 = so3;
717 }
718 error = unp_connect2(so, so2, PRU_CONNECT);
719 bad:
720 vput(vp);
721 bad2:
722 free(sun, M_SONAME);
723 return (error);
724 }
725
726 int
727 unp_connect2(struct socket *so, struct socket *so2, int req)
728 {
729 struct unpcb *unp = sotounpcb(so);
730 struct unpcb *unp2;
731
732 if (so2->so_type != so->so_type)
733 return (EPROTOTYPE);
734 unp2 = sotounpcb(so2);
735 unp->unp_conn = unp2;
736 switch (so->so_type) {
737
738 case SOCK_DGRAM:
739 unp->unp_nextref = unp2->unp_refs;
740 unp2->unp_refs = unp;
741 soisconnected(so);
742 break;
743
744 case SOCK_STREAM:
745 unp2->unp_conn = unp;
746 if (req == PRU_CONNECT &&
747 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT))
748 soisconnecting(so);
749 else
750 soisconnected(so);
751 soisconnected(so2);
752 break;
753
754 default:
755 panic("unp_connect2");
756 }
757 return (0);
758 }
759
760 void
761 unp_disconnect(struct unpcb *unp)
762 {
763 struct unpcb *unp2 = unp->unp_conn;
764
765 if (unp2 == 0)
766 return;
767 unp->unp_conn = 0;
768 switch (unp->unp_socket->so_type) {
769
770 case SOCK_DGRAM:
771 if (unp2->unp_refs == unp)
772 unp2->unp_refs = unp->unp_nextref;
773 else {
774 unp2 = unp2->unp_refs;
775 for (;;) {
776 if (unp2 == 0)
777 panic("unp_disconnect");
778 if (unp2->unp_nextref == unp)
779 break;
780 unp2 = unp2->unp_nextref;
781 }
782 unp2->unp_nextref = unp->unp_nextref;
783 }
784 unp->unp_nextref = 0;
785 unp->unp_socket->so_state &= ~SS_ISCONNECTED;
786 break;
787
788 case SOCK_STREAM:
789 soisdisconnected(unp->unp_socket);
790 unp2->unp_conn = 0;
791 soisdisconnected(unp2->unp_socket);
792 break;
793 }
794 }
795
796 #ifdef notdef
797 unp_abort(struct unpcb *unp)
798 {
799 unp_detach(unp);
800 }
801 #endif
802
803 void
804 unp_shutdown(struct unpcb *unp)
805 {
806 struct socket *so;
807
808 if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn &&
809 (so = unp->unp_conn->unp_socket))
810 socantrcvmore(so);
811 }
812
813 void
814 unp_drop(struct unpcb *unp, int errno)
815 {
816 struct socket *so = unp->unp_socket;
817
818 so->so_error = errno;
819 unp_disconnect(unp);
820 if (so->so_head) {
821 so->so_pcb = 0;
822 sofree(so);
823 if (unp->unp_addr)
824 free(unp->unp_addr, M_SONAME);
825 free(unp, M_PCB);
826 }
827 }
828
829 #ifdef notdef
830 unp_drain(void)
831 {
832
833 }
834 #endif
835
836 int
837 unp_externalize(struct mbuf *rights, struct proc *p)
838 {
839 struct cmsghdr *cm = mtod(rights, struct cmsghdr *);
840 int i, *fdp;
841 struct file **rp;
842 struct file *fp;
843 int nfds, error = 0;
844
845 nfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm))) /
846 sizeof(struct file *);
847 rp = (struct file **)CMSG_DATA(cm);
848
849 fdp = malloc(nfds * sizeof(int), M_TEMP, M_WAITOK);
850
851 /* Make sure the recipient should be able to see the descriptors.. */
852 if (p->p_cwdi->cwdi_rdir != NULL) {
853 rp = (struct file **)CMSG_DATA(cm);
854 for (i = 0; i < nfds; i++) {
855 fp = *rp++;
856 /*
857 * If we are in a chroot'ed directory, and
858 * someone wants to pass us a directory, make
859 * sure it's inside the subtree we're allowed
860 * to access.
861 */
862 if (fp->f_type == DTYPE_VNODE) {
863 struct vnode *vp = (struct vnode *)fp->f_data;
864 if ((vp->v_type == VDIR) &&
865 !vn_isunder(vp, p->p_cwdi->cwdi_rdir, p)) {
866 error = EPERM;
867 break;
868 }
869 }
870 }
871 }
872
873 restart:
874 rp = (struct file **)CMSG_DATA(cm);
875 if (error != 0) {
876 for (i = 0; i < nfds; i++) {
877 fp = *rp;
878 /*
879 * zero the pointer before calling unp_discard,
880 * since it may end up in unp_gc()..
881 */
882 *rp++ = 0;
883 unp_discard(fp);
884 }
885 goto out;
886 }
887
888 /*
889 * First loop -- allocate file descriptor table slots for the
890 * new descriptors.
891 */
892 for (i = 0; i < nfds; i++) {
893 fp = *rp++;
894 if ((error = fdalloc(p, 0, &fdp[i])) != 0) {
895 /*
896 * Back out what we've done so far.
897 */
898 for (--i; i >= 0; i--)
899 fdremove(p->p_fd, fdp[i]);
900
901 if (error == ENOSPC) {
902 fdexpand(p);
903 error = 0;
904 } else {
905 /*
906 * This is the error that has historically
907 * been returned, and some callers may
908 * expect it.
909 */
910 error = EMSGSIZE;
911 }
912 goto restart;
913 }
914
915 /*
916 * Make the slot reference the descriptor so that
917 * fdalloc() works properly.. We finalize it all
918 * in the loop below.
919 */
920 p->p_fd->fd_ofiles[fdp[i]] = fp;
921 }
922
923 /*
924 * Now that adding them has succeeded, update all of the
925 * descriptor passing state.
926 */
927 rp = (struct file **)CMSG_DATA(cm);
928 for (i = 0; i < nfds; i++) {
929 fp = *rp++;
930 fp->f_msgcount--;
931 unp_rights--;
932 }
933
934 /*
935 * Copy temporary array to message and adjust length, in case of
936 * transition from large struct file pointers to ints.
937 */
938 memcpy(CMSG_DATA(cm), fdp, nfds * sizeof(int));
939 cm->cmsg_len = CMSG_LEN(nfds * sizeof(int));
940 rights->m_len = CMSG_SPACE(nfds * sizeof(int));
941 out:
942 free(fdp, M_TEMP);
943 return (error);
944 }
945
946 int
947 unp_internalize(struct mbuf *control, struct proc *p)
948 {
949 struct filedesc *fdescp = p->p_fd;
950 struct cmsghdr *newcm, *cm = mtod(control, struct cmsghdr *);
951 struct file **rp, **files;
952 struct file *fp;
953 int i, fd, *fdp;
954 int nfds;
955 u_int neededspace;
956
957 /* Sanity check the control message header */
958 if (cm->cmsg_type != SCM_RIGHTS || cm->cmsg_level != SOL_SOCKET ||
959 cm->cmsg_len != control->m_len)
960 return (EINVAL);
961
962 /* Verify that the file descriptors are valid */
963 nfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm))) / sizeof(int);
964 fdp = (int *)CMSG_DATA(cm);
965 for (i = 0; i < nfds; i++) {
966 fd = *fdp++;
967 if ((fp = fd_getfile(fdescp, fd)) == NULL)
968 return (EBADF);
969 simple_unlock(&fp->f_slock);
970 }
971
972 /* Make sure we have room for the struct file pointers */
973 neededspace = CMSG_SPACE(nfds * sizeof(struct file *)) -
974 control->m_len;
975 if (neededspace > M_TRAILINGSPACE(control)) {
976
977 /* allocate new space and copy header into it */
978 newcm = malloc(
979 CMSG_SPACE(nfds * sizeof(struct file *)),
980 M_MBUF, M_WAITOK);
981 if (newcm == NULL)
982 return (E2BIG);
983 memcpy(newcm, cm, sizeof(struct cmsghdr));
984 files = (struct file **)CMSG_DATA(newcm);
985 } else {
986 /* we can convert in-place */
987 newcm = NULL;
988 files = (struct file **)CMSG_DATA(cm);
989 }
990
991 /*
992 * Transform the file descriptors into struct file pointers, in
993 * reverse order so that if pointers are bigger than ints, the
994 * int won't get until we're done.
995 */
996 fdp = (int *)CMSG_DATA(cm) + nfds - 1;
997 rp = files + nfds - 1;
998 for (i = 0; i < nfds; i++) {
999 fp = fdescp->fd_ofiles[*fdp--];
1000 simple_lock(&fp->f_slock);
1001 #ifdef DIAGNOSTIC
1002 if (fp->f_iflags & FIF_WANTCLOSE)
1003 panic("unp_internalize: file already closed");
1004 #endif
1005 *rp-- = fp;
1006 fp->f_count++;
1007 fp->f_msgcount++;
1008 simple_unlock(&fp->f_slock);
1009 unp_rights++;
1010 }
1011
1012 if (newcm) {
1013 if (control->m_flags & M_EXT)
1014 MEXTREMOVE(control);
1015 MEXTADD(control, newcm,
1016 CMSG_SPACE(nfds * sizeof(struct file *)),
1017 M_MBUF, NULL, NULL);
1018 cm = newcm;
1019 }
1020
1021 /* adjust message & mbuf to note amount of space actually used. */
1022 cm->cmsg_len = CMSG_LEN(nfds * sizeof(struct file *));
1023 control->m_len = CMSG_SPACE(nfds * sizeof(struct file *));
1024
1025 return (0);
1026 }
1027
1028 struct mbuf *
1029 unp_addsockcred(struct proc *p, struct mbuf *control)
1030 {
1031 struct cmsghdr *cmp;
1032 struct sockcred *sc;
1033 struct mbuf *m, *n;
1034 int len, space, i;
1035
1036 len = CMSG_LEN(SOCKCREDSIZE(p->p_ucred->cr_ngroups));
1037 space = CMSG_SPACE(SOCKCREDSIZE(p->p_ucred->cr_ngroups));
1038
1039 m = m_get(M_WAIT, MT_CONTROL);
1040 if (space > MLEN) {
1041 if (space > MCLBYTES)
1042 MEXTMALLOC(m, space, M_WAITOK);
1043 else
1044 m_clget(m, M_WAIT);
1045 if ((m->m_flags & M_EXT) == 0) {
1046 m_free(m);
1047 return (control);
1048 }
1049 }
1050
1051 m->m_len = space;
1052 m->m_next = NULL;
1053 cmp = mtod(m, struct cmsghdr *);
1054 sc = (struct sockcred *)CMSG_DATA(cmp);
1055 cmp->cmsg_len = len;
1056 cmp->cmsg_level = SOL_SOCKET;
1057 cmp->cmsg_type = SCM_CREDS;
1058 sc->sc_uid = p->p_cred->p_ruid;
1059 sc->sc_euid = p->p_ucred->cr_uid;
1060 sc->sc_gid = p->p_cred->p_rgid;
1061 sc->sc_egid = p->p_ucred->cr_gid;
1062 sc->sc_ngroups = p->p_ucred->cr_ngroups;
1063 for (i = 0; i < sc->sc_ngroups; i++)
1064 sc->sc_groups[i] = p->p_ucred->cr_groups[i];
1065
1066 /*
1067 * If a control message already exists, append us to the end.
1068 */
1069 if (control != NULL) {
1070 for (n = control; n->m_next != NULL; n = n->m_next)
1071 ;
1072 n->m_next = m;
1073 } else
1074 control = m;
1075
1076 return (control);
1077 }
1078
1079 int unp_defer, unp_gcing;
1080 extern struct domain unixdomain;
1081
1082 /*
1083 * Comment added long after the fact explaining what's going on here.
1084 * Do a mark-sweep GC of file descriptors on the system, to free up
1085 * any which are caught in flight to an about-to-be-closed socket.
1086 *
1087 * Traditional mark-sweep gc's start at the "root", and mark
1088 * everything reachable from the root (which, in our case would be the
1089 * process table). The mark bits are cleared during the sweep.
1090 *
1091 * XXX For some inexplicable reason (perhaps because the file
1092 * descriptor tables used to live in the u area which could be swapped
1093 * out and thus hard to reach), we do multiple scans over the set of
1094 * descriptors, using use *two* mark bits per object (DEFER and MARK).
1095 * Whenever we find a descriptor which references other descriptors,
1096 * the ones it references are marked with both bits, and we iterate
1097 * over the whole file table until there are no more DEFER bits set.
1098 * We also make an extra pass *before* the GC to clear the mark bits,
1099 * which could have been cleared at almost no cost during the previous
1100 * sweep.
1101 *
1102 * XXX MP: this needs to run with locks such that no other thread of
1103 * control can create or destroy references to file descriptors. it
1104 * may be necessary to defer the GC until later (when the locking
1105 * situation is more hospitable); it may be necessary to push this
1106 * into a separate thread.
1107 */
1108 void
1109 unp_gc(void)
1110 {
1111 struct file *fp, *nextfp;
1112 struct socket *so, *so1;
1113 struct file **extra_ref, **fpp;
1114 int nunref, i;
1115
1116 if (unp_gcing)
1117 return;
1118 unp_gcing = 1;
1119 unp_defer = 0;
1120
1121 /* Clear mark bits */
1122 LIST_FOREACH(fp, &filehead, f_list)
1123 fp->f_flag &= ~(FMARK|FDEFER);
1124
1125 /*
1126 * Iterate over the set of descriptors, marking ones believed
1127 * (based on refcount) to be referenced from a process, and
1128 * marking for rescan descriptors which are queued on a socket.
1129 */
1130 do {
1131 LIST_FOREACH(fp, &filehead, f_list) {
1132 if (fp->f_flag & FDEFER) {
1133 fp->f_flag &= ~FDEFER;
1134 unp_defer--;
1135 #ifdef DIAGNOSTIC
1136 if (fp->f_count == 0)
1137 panic("unp_gc: deferred unreferenced socket");
1138 #endif
1139 } else {
1140 if (fp->f_count == 0)
1141 continue;
1142 if (fp->f_flag & FMARK)
1143 continue;
1144 if (fp->f_count == fp->f_msgcount)
1145 continue;
1146 }
1147 fp->f_flag |= FMARK;
1148
1149 if (fp->f_type != DTYPE_SOCKET ||
1150 (so = (struct socket *)fp->f_data) == 0)
1151 continue;
1152 if (so->so_proto->pr_domain != &unixdomain ||
1153 (so->so_proto->pr_flags&PR_RIGHTS) == 0)
1154 continue;
1155 #ifdef notdef
1156 if (so->so_rcv.sb_flags & SB_LOCK) {
1157 /*
1158 * This is problematical; it's not clear
1159 * we need to wait for the sockbuf to be
1160 * unlocked (on a uniprocessor, at least),
1161 * and it's also not clear what to do
1162 * if sbwait returns an error due to receipt
1163 * of a signal. If sbwait does return
1164 * an error, we'll go into an infinite
1165 * loop. Delete all of this for now.
1166 */
1167 (void) sbwait(&so->so_rcv);
1168 goto restart;
1169 }
1170 #endif
1171 unp_scan(so->so_rcv.sb_mb, unp_mark, 0);
1172 /*
1173 * mark descriptors referenced from sockets queued on the accept queue as well.
1174 */
1175 if (so->so_options & SO_ACCEPTCONN) {
1176 TAILQ_FOREACH(so1, &so->so_q0, so_qe) {
1177 unp_scan(so1->so_rcv.sb_mb, unp_mark, 0);
1178 }
1179 TAILQ_FOREACH(so1, &so->so_q, so_qe) {
1180 unp_scan(so1->so_rcv.sb_mb, unp_mark, 0);
1181 }
1182 }
1183
1184 }
1185 } while (unp_defer);
1186 /*
1187 * Sweep pass. Find unmarked descriptors, and free them.
1188 *
1189 * We grab an extra reference to each of the file table entries
1190 * that are not otherwise accessible and then free the rights
1191 * that are stored in messages on them.
1192 *
1193 * The bug in the original code is a little tricky, so I'll describe
1194 * what's wrong with it here.
1195 *
1196 * It is incorrect to simply unp_discard each entry for f_msgcount
1197 * times -- consider the case of sockets A and B that contain
1198 * references to each other. On a last close of some other socket,
1199 * we trigger a gc since the number of outstanding rights (unp_rights)
1200 * is non-zero. If during the sweep phase the gc code un_discards,
1201 * we end up doing a (full) closef on the descriptor. A closef on A
1202 * results in the following chain. Closef calls soo_close, which
1203 * calls soclose. Soclose calls first (through the switch
1204 * uipc_usrreq) unp_detach, which re-invokes unp_gc. Unp_gc simply
1205 * returns because the previous instance had set unp_gcing, and
1206 * we return all the way back to soclose, which marks the socket
1207 * with SS_NOFDREF, and then calls sofree. Sofree calls sorflush
1208 * to free up the rights that are queued in messages on the socket A,
1209 * i.e., the reference on B. The sorflush calls via the dom_dispose
1210 * switch unp_dispose, which unp_scans with unp_discard. This second
1211 * instance of unp_discard just calls closef on B.
1212 *
1213 * Well, a similar chain occurs on B, resulting in a sorflush on B,
1214 * which results in another closef on A. Unfortunately, A is already
1215 * being closed, and the descriptor has already been marked with
1216 * SS_NOFDREF, and soclose panics at this point.
1217 *
1218 * Here, we first take an extra reference to each inaccessible
1219 * descriptor. Then, if the inaccessible descriptor is a
1220 * socket, we call sorflush in case it is a Unix domain
1221 * socket. After we destroy all the rights carried in
1222 * messages, we do a last closef to get rid of our extra
1223 * reference. This is the last close, and the unp_detach etc
1224 * will shut down the socket.
1225 *
1226 * 91/09/19, bsy (at) cs.cmu.edu
1227 */
1228 extra_ref = malloc(nfiles * sizeof(struct file *), M_FILE, M_WAITOK);
1229 for (nunref = 0, fp = LIST_FIRST(&filehead), fpp = extra_ref; fp != 0;
1230 fp = nextfp) {
1231 nextfp = LIST_NEXT(fp, f_list);
1232 simple_lock(&fp->f_slock);
1233 if (fp->f_count != 0 &&
1234 fp->f_count == fp->f_msgcount && !(fp->f_flag & FMARK)) {
1235 *fpp++ = fp;
1236 nunref++;
1237 fp->f_count++;
1238 }
1239 simple_unlock(&fp->f_slock);
1240 }
1241 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) {
1242 fp = *fpp;
1243 simple_lock(&fp->f_slock);
1244 FILE_USE(fp);
1245 if (fp->f_type == DTYPE_SOCKET)
1246 sorflush((struct socket *)fp->f_data);
1247 FILE_UNUSE(fp, NULL);
1248 }
1249 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) {
1250 fp = *fpp;
1251 simple_lock(&fp->f_slock);
1252 FILE_USE(fp);
1253 (void) closef(fp, (struct proc *)0);
1254 }
1255 free((caddr_t)extra_ref, M_FILE);
1256 unp_gcing = 0;
1257 }
1258
1259 void
1260 unp_dispose(struct mbuf *m)
1261 {
1262
1263 if (m)
1264 unp_scan(m, unp_discard, 1);
1265 }
1266
1267 void
1268 unp_scan(struct mbuf *m0, void (*op)(struct file *), int discard)
1269 {
1270 struct mbuf *m;
1271 struct file **rp;
1272 struct cmsghdr *cm;
1273 int i;
1274 int qfds;
1275
1276 while (m0) {
1277 for (m = m0; m; m = m->m_next) {
1278 if (m->m_type == MT_CONTROL &&
1279 m->m_len >= sizeof(*cm)) {
1280 cm = mtod(m, struct cmsghdr *);
1281 if (cm->cmsg_level != SOL_SOCKET ||
1282 cm->cmsg_type != SCM_RIGHTS)
1283 continue;
1284 qfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm)))
1285 / sizeof(struct file *);
1286 rp = (struct file **)CMSG_DATA(cm);
1287 for (i = 0; i < qfds; i++) {
1288 struct file *fp = *rp;
1289 if (discard)
1290 *rp = 0;
1291 (*op)(fp);
1292 rp++;
1293 }
1294 break; /* XXX, but saves time */
1295 }
1296 }
1297 m0 = m0->m_nextpkt;
1298 }
1299 }
1300
1301 void
1302 unp_mark(struct file *fp)
1303 {
1304 if (fp == NULL)
1305 return;
1306
1307 if (fp->f_flag & FMARK)
1308 return;
1309
1310 /* If we're already deferred, don't screw up the defer count */
1311 if (fp->f_flag & FDEFER)
1312 return;
1313
1314 /*
1315 * Minimize the number of deferrals... Sockets are the only
1316 * type of descriptor which can hold references to another
1317 * descriptor, so just mark other descriptors, and defer
1318 * unmarked sockets for the next pass.
1319 */
1320 if (fp->f_type == DTYPE_SOCKET) {
1321 unp_defer++;
1322 if (fp->f_count == 0)
1323 panic("unp_mark: queued unref");
1324 fp->f_flag |= FDEFER;
1325 } else {
1326 fp->f_flag |= FMARK;
1327 }
1328 return;
1329 }
1330
1331 void
1332 unp_discard(struct file *fp)
1333 {
1334 if (fp == NULL)
1335 return;
1336 simple_lock(&fp->f_slock);
1337 fp->f_usecount++; /* i.e. FILE_USE(fp) sans locking */
1338 fp->f_msgcount--;
1339 simple_unlock(&fp->f_slock);
1340 unp_rights--;
1341 (void) closef(fp, (struct proc *)0);
1342 }
1343