1 /* $NetBSD: uipc_usrreq.c,v 1.74.2.1 2005/07/18 04:03:05 riz Exp $ */
2
3 /*-
4 * Copyright (c) 1998, 2000 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 1.74.2.1 2005/07/18 04:03:05 riz 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 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(m, control, unp, p)
139 struct mbuf *m, *control;
140 struct unpcb *unp;
141 struct proc *p;
142 {
143 struct socket *so2;
144 struct sockaddr_un *sun;
145
146 so2 = unp->unp_conn->unp_socket;
147 if (unp->unp_addr)
148 sun = unp->unp_addr;
149 else
150 sun = &sun_noname;
151 if (unp->unp_conn->unp_flags & UNP_WANTCRED)
152 control = unp_addsockcred(p, control);
153 if (sbappendaddr(&so2->so_rcv, (struct sockaddr *)sun, m,
154 control) == 0) {
155 m_freem(control);
156 m_freem(m);
157 return (ENOBUFS);
158 } else {
159 sorwakeup(so2);
160 return (0);
161 }
162 }
163
164 void
165 unp_setsockaddr(unp, nam)
166 struct unpcb *unp;
167 struct mbuf *nam;
168 {
169 struct sockaddr_un *sun;
170
171 if (unp->unp_addr)
172 sun = unp->unp_addr;
173 else
174 sun = &sun_noname;
175 nam->m_len = sun->sun_len;
176 if (nam->m_len > MLEN)
177 MEXTMALLOC(nam, nam->m_len, M_WAITOK);
178 memcpy(mtod(nam, caddr_t), sun, (size_t)nam->m_len);
179 }
180
181 void
182 unp_setpeeraddr(unp, nam)
183 struct unpcb *unp;
184 struct mbuf *nam;
185 {
186 struct sockaddr_un *sun;
187
188 if (unp->unp_conn && unp->unp_conn->unp_addr)
189 sun = unp->unp_conn->unp_addr;
190 else
191 sun = &sun_noname;
192 nam->m_len = sun->sun_len;
193 if (nam->m_len > MLEN)
194 MEXTMALLOC(nam, nam->m_len, M_WAITOK);
195 memcpy(mtod(nam, caddr_t), sun, (size_t)nam->m_len);
196 }
197
198 /*ARGSUSED*/
199 int
200 uipc_usrreq(so, req, m, nam, control, p)
201 struct socket *so;
202 int req;
203 struct mbuf *m, *nam, *control;
204 struct proc *p;
205 {
206 struct unpcb *unp = sotounpcb(so);
207 struct socket *so2;
208 int error = 0;
209
210 if (req == PRU_CONTROL)
211 return (EOPNOTSUPP);
212
213 #ifdef DIAGNOSTIC
214 if (req != PRU_SEND && req != PRU_SENDOOB && control)
215 panic("uipc_usrreq: unexpected control mbuf");
216 #endif
217 if (unp == 0 && req != PRU_ATTACH) {
218 error = EINVAL;
219 goto release;
220 }
221
222 switch (req) {
223
224 case PRU_ATTACH:
225 if (unp != 0) {
226 error = EISCONN;
227 break;
228 }
229 error = unp_attach(so);
230 break;
231
232 case PRU_DETACH:
233 unp_detach(unp);
234 break;
235
236 case PRU_BIND:
237 error = unp_bind(unp, nam, p);
238 break;
239
240 case PRU_LISTEN:
241 if (unp->unp_vnode == 0)
242 error = EINVAL;
243 break;
244
245 case PRU_CONNECT:
246 error = unp_connect(so, nam, p);
247 break;
248
249 case PRU_CONNECT2:
250 error = unp_connect2(so, (struct socket *)nam, PRU_CONNECT2);
251 break;
252
253 case PRU_DISCONNECT:
254 unp_disconnect(unp);
255 break;
256
257 case PRU_ACCEPT:
258 unp_setpeeraddr(unp, nam);
259 /*
260 * Mark the initiating STREAM socket as connected *ONLY*
261 * after it's been accepted. This prevents a client from
262 * overrunning a server and receiving ECONNREFUSED.
263 */
264 if (unp->unp_conn != NULL &&
265 (unp->unp_conn->unp_socket->so_state & SS_ISCONNECTING))
266 soisconnected(unp->unp_conn->unp_socket);
267 break;
268
269 case PRU_SHUTDOWN:
270 socantsendmore(so);
271 unp_shutdown(unp);
272 break;
273
274 case PRU_RCVD:
275 switch (so->so_type) {
276
277 case SOCK_DGRAM:
278 panic("uipc 1");
279 /*NOTREACHED*/
280
281 case SOCK_STREAM:
282 #define rcv (&so->so_rcv)
283 #define snd (&so2->so_snd)
284 if (unp->unp_conn == 0)
285 break;
286 so2 = unp->unp_conn->unp_socket;
287 /*
288 * Adjust backpressure on sender
289 * and wakeup any waiting to write.
290 */
291 snd->sb_mbmax += unp->unp_mbcnt - rcv->sb_mbcnt;
292 unp->unp_mbcnt = rcv->sb_mbcnt;
293 snd->sb_hiwat += unp->unp_cc - rcv->sb_cc;
294 unp->unp_cc = rcv->sb_cc;
295 sowwakeup(so2);
296 #undef snd
297 #undef rcv
298 break;
299
300 default:
301 panic("uipc 2");
302 }
303 break;
304
305 case PRU_SEND:
306 /*
307 * Note: unp_internalize() rejects any control message
308 * other than SCM_RIGHTS, and only allows one. This
309 * has the side-effect of preventing a caller from
310 * forging SCM_CREDS.
311 */
312 if (control && (error = unp_internalize(control, p))) {
313 goto die;
314 }
315 switch (so->so_type) {
316
317 case SOCK_DGRAM: {
318 if (nam) {
319 if ((so->so_state & SS_ISCONNECTED) != 0) {
320 error = EISCONN;
321 goto die;
322 }
323 error = unp_connect(so, nam, p);
324 if (error) {
325 die:
326 m_freem(control);
327 m_freem(m);
328 break;
329 }
330 } else {
331 if ((so->so_state & SS_ISCONNECTED) == 0) {
332 error = ENOTCONN;
333 goto die;
334 }
335 }
336 error = unp_output(m, control, unp, p);
337 if (nam)
338 unp_disconnect(unp);
339 break;
340 }
341
342 case SOCK_STREAM:
343 #define rcv (&so2->so_rcv)
344 #define snd (&so->so_snd)
345 if (unp->unp_conn == 0)
346 panic("uipc 3");
347 so2 = unp->unp_conn->unp_socket;
348 if (unp->unp_conn->unp_flags & UNP_WANTCRED) {
349 /*
350 * Credentials are passed only once on
351 * SOCK_STREAM.
352 */
353 unp->unp_conn->unp_flags &= ~UNP_WANTCRED;
354 control = unp_addsockcred(p, control);
355 }
356 /*
357 * Send to paired receive port, and then reduce
358 * send buffer hiwater marks to maintain backpressure.
359 * Wake up readers.
360 */
361 if (control) {
362 if (sbappendcontrol(rcv, m, control) == 0)
363 m_freem(control);
364 } else
365 sbappend(rcv, m);
366 snd->sb_mbmax -=
367 rcv->sb_mbcnt - unp->unp_conn->unp_mbcnt;
368 unp->unp_conn->unp_mbcnt = rcv->sb_mbcnt;
369 snd->sb_hiwat -= rcv->sb_cc - unp->unp_conn->unp_cc;
370 unp->unp_conn->unp_cc = rcv->sb_cc;
371 sorwakeup(so2);
372 #undef snd
373 #undef rcv
374 break;
375
376 default:
377 panic("uipc 4");
378 }
379 break;
380
381 case PRU_ABORT:
382 unp_drop(unp, ECONNABORTED);
383
384 #ifdef DIAGNOSTIC
385 if (so->so_pcb == 0)
386 panic("uipc 5: drop killed pcb");
387 #endif
388 unp_detach(unp);
389 break;
390
391 case PRU_SENSE:
392 ((struct stat *) m)->st_blksize = so->so_snd.sb_hiwat;
393 if (so->so_type == SOCK_STREAM && unp->unp_conn != 0) {
394 so2 = unp->unp_conn->unp_socket;
395 ((struct stat *) m)->st_blksize += so2->so_rcv.sb_cc;
396 }
397 ((struct stat *) m)->st_dev = NODEV;
398 if (unp->unp_ino == 0)
399 unp->unp_ino = unp_ino++;
400 ((struct stat *) m)->st_atimespec =
401 ((struct stat *) m)->st_mtimespec =
402 ((struct stat *) m)->st_ctimespec = unp->unp_ctime;
403 ((struct stat *) m)->st_ino = unp->unp_ino;
404 return (0);
405
406 case PRU_RCVOOB:
407 error = EOPNOTSUPP;
408 break;
409
410 case PRU_SENDOOB:
411 m_freem(control);
412 m_freem(m);
413 error = EOPNOTSUPP;
414 break;
415
416 case PRU_SOCKADDR:
417 unp_setsockaddr(unp, nam);
418 break;
419
420 case PRU_PEERADDR:
421 unp_setpeeraddr(unp, nam);
422 break;
423
424 default:
425 panic("piusrreq");
426 }
427
428 release:
429 return (error);
430 }
431
432 /*
433 * Unix domain socket option processing.
434 */
435 int
436 uipc_ctloutput(op, so, level, optname, mp)
437 int op;
438 struct socket *so;
439 int level, optname;
440 struct mbuf **mp;
441 {
442 struct unpcb *unp = sotounpcb(so);
443 struct mbuf *m = *mp;
444 int optval = 0, error = 0;
445
446 if (level != 0) {
447 error = EINVAL;
448 if (op == PRCO_SETOPT && m)
449 (void) m_free(m);
450 } else switch (op) {
451
452 case PRCO_SETOPT:
453 switch (optname) {
454 case LOCAL_CREDS:
455 case LOCAL_CONNWAIT:
456 if (m == NULL || m->m_len != sizeof(int))
457 error = EINVAL;
458 else {
459 optval = *mtod(m, int *);
460 switch (optname) {
461 #define OPTSET(bit) \
462 if (optval) \
463 unp->unp_flags |= (bit); \
464 else \
465 unp->unp_flags &= ~(bit);
466
467 case LOCAL_CREDS:
468 OPTSET(UNP_WANTCRED);
469 break;
470 case LOCAL_CONNWAIT:
471 OPTSET(UNP_CONNWAIT);
472 break;
473 }
474 }
475 break;
476 #undef OPTSET
477
478 default:
479 error = ENOPROTOOPT;
480 break;
481 }
482 if (m)
483 (void) m_free(m);
484 break;
485
486 case PRCO_GETOPT:
487 switch (optname) {
488 case LOCAL_CREDS:
489 *mp = m = m_get(M_WAIT, MT_SOOPTS);
490 m->m_len = sizeof(int);
491 switch (optname) {
492
493 #define OPTBIT(bit) (unp->unp_flags & (bit) ? 1 : 0)
494
495 case LOCAL_CREDS:
496 optval = OPTBIT(UNP_WANTCRED);
497 break;
498 }
499 *mtod(m, int *) = optval;
500 break;
501 #undef OPTBIT
502
503 default:
504 error = ENOPROTOOPT;
505 break;
506 }
507 break;
508 }
509 return (error);
510 }
511
512 /*
513 * Both send and receive buffers are allocated PIPSIZ bytes of buffering
514 * for stream sockets, although the total for sender and receiver is
515 * actually only PIPSIZ.
516 * Datagram sockets really use the sendspace as the maximum datagram size,
517 * and don't really want to reserve the sendspace. Their recvspace should
518 * be large enough for at least one max-size datagram plus address.
519 */
520 #define PIPSIZ 4096
521 u_long unpst_sendspace = PIPSIZ;
522 u_long unpst_recvspace = PIPSIZ;
523 u_long unpdg_sendspace = 2*1024; /* really max datagram size */
524 u_long unpdg_recvspace = 4*1024;
525
526 int unp_rights; /* file descriptors in flight */
527
528 int
529 unp_attach(so)
530 struct socket *so;
531 {
532 struct unpcb *unp;
533 struct timeval tv;
534 int error;
535
536 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
537 switch (so->so_type) {
538
539 case SOCK_STREAM:
540 error = soreserve(so, unpst_sendspace, unpst_recvspace);
541 break;
542
543 case SOCK_DGRAM:
544 error = soreserve(so, unpdg_sendspace, unpdg_recvspace);
545 break;
546
547 default:
548 panic("unp_attach");
549 }
550 if (error)
551 return (error);
552 }
553 unp = malloc(sizeof(*unp), M_PCB, M_NOWAIT);
554 if (unp == NULL)
555 return (ENOBUFS);
556 memset((caddr_t)unp, 0, sizeof(*unp));
557 unp->unp_socket = so;
558 so->so_pcb = unp;
559 microtime(&tv);
560 TIMEVAL_TO_TIMESPEC(&tv, &unp->unp_ctime);
561 return (0);
562 }
563
564 void
565 unp_detach(unp)
566 struct unpcb *unp;
567 {
568
569 if (unp->unp_vnode) {
570 unp->unp_vnode->v_socket = 0;
571 vrele(unp->unp_vnode);
572 unp->unp_vnode = 0;
573 }
574 if (unp->unp_conn)
575 unp_disconnect(unp);
576 while (unp->unp_refs)
577 unp_drop(unp->unp_refs, ECONNRESET);
578 soisdisconnected(unp->unp_socket);
579 unp->unp_socket->so_pcb = 0;
580 if (unp->unp_addr)
581 free(unp->unp_addr, M_SONAME);
582 if (unp_rights) {
583 /*
584 * Normally the receive buffer is flushed later,
585 * in sofree, but if our receive buffer holds references
586 * to descriptors that are now garbage, we will dispose
587 * of those descriptor references after the garbage collector
588 * gets them (resulting in a "panic: closef: count < 0").
589 */
590 sorflush(unp->unp_socket);
591 free(unp, M_PCB);
592 unp_gc();
593 } else
594 free(unp, M_PCB);
595 }
596
597 int
598 unp_bind(unp, nam, p)
599 struct unpcb *unp;
600 struct mbuf *nam;
601 struct proc *p;
602 {
603 struct sockaddr_un *sun;
604 struct vnode *vp;
605 struct mount *mp;
606 struct vattr vattr;
607 size_t addrlen;
608 int error;
609 struct nameidata nd;
610
611 if (unp->unp_vnode != 0)
612 return (EINVAL);
613
614 /*
615 * Allocate the new sockaddr. We have to allocate one
616 * extra byte so that we can ensure that the pathname
617 * is nul-terminated.
618 */
619 addrlen = nam->m_len + 1;
620 sun = malloc(addrlen, M_SONAME, M_WAITOK);
621 m_copydata(nam, 0, nam->m_len, (caddr_t)sun);
622 *(((char *)sun) + nam->m_len) = '\0';
623
624 restart:
625 NDINIT(&nd, CREATE, FOLLOW | LOCKPARENT, UIO_SYSSPACE,
626 sun->sun_path, p);
627
628 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
629 if ((error = namei(&nd)) != 0)
630 goto bad;
631 vp = nd.ni_vp;
632 if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
633 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
634 if (nd.ni_dvp == vp)
635 vrele(nd.ni_dvp);
636 else
637 vput(nd.ni_dvp);
638 vrele(vp);
639 if (vp != NULL) {
640 error = EADDRINUSE;
641 goto bad;
642 }
643 error = vn_start_write(NULL, &mp,
644 V_WAIT | V_SLEEPONLY | V_PCATCH);
645 if (error)
646 goto bad;
647 goto restart;
648 }
649 VATTR_NULL(&vattr);
650 vattr.va_type = VSOCK;
651 vattr.va_mode = ACCESSPERMS;
652 VOP_LEASE(nd.ni_dvp, p, p->p_ucred, LEASE_WRITE);
653 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
654 vn_finished_write(mp, 0);
655 if (error)
656 goto bad;
657 vp = nd.ni_vp;
658 vp->v_socket = unp->unp_socket;
659 unp->unp_vnode = vp;
660 unp->unp_addrlen = addrlen;
661 unp->unp_addr = sun;
662 VOP_UNLOCK(vp, 0);
663 return (0);
664
665 bad:
666 free(sun, M_SONAME);
667 return (error);
668 }
669
670 int
671 unp_connect(so, nam, p)
672 struct socket *so;
673 struct mbuf *nam;
674 struct proc *p;
675 {
676 struct sockaddr_un *sun;
677 struct vnode *vp;
678 struct socket *so2, *so3;
679 struct unpcb *unp2, *unp3;
680 size_t addrlen;
681 int error;
682 struct nameidata nd;
683
684 /*
685 * Allocate a temporary sockaddr. We have to allocate one extra
686 * byte so that we can ensure that the pathname is nul-terminated.
687 * When we establish the connection, we copy the other PCB's
688 * sockaddr to our own.
689 */
690 addrlen = nam->m_len + 1;
691 sun = malloc(addrlen, M_SONAME, M_WAITOK);
692 m_copydata(nam, 0, nam->m_len, (caddr_t)sun);
693 *(((char *)sun) + nam->m_len) = '\0';
694
695 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, sun->sun_path, p);
696
697 if ((error = namei(&nd)) != 0)
698 goto bad2;
699 vp = nd.ni_vp;
700 if (vp->v_type != VSOCK) {
701 error = ENOTSOCK;
702 goto bad;
703 }
704 if ((error = VOP_ACCESS(vp, VWRITE, p->p_ucred, p)) != 0)
705 goto bad;
706 so2 = vp->v_socket;
707 if (so2 == 0) {
708 error = ECONNREFUSED;
709 goto bad;
710 }
711 if (so->so_type != so2->so_type) {
712 error = EPROTOTYPE;
713 goto bad;
714 }
715 if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
716 if ((so2->so_options & SO_ACCEPTCONN) == 0 ||
717 (so3 = sonewconn(so2, 0)) == 0) {
718 error = ECONNREFUSED;
719 goto bad;
720 }
721 unp2 = sotounpcb(so2);
722 unp3 = sotounpcb(so3);
723 if (unp2->unp_addr) {
724 unp3->unp_addr = malloc(unp2->unp_addrlen,
725 M_SONAME, M_WAITOK);
726 memcpy(unp3->unp_addr, unp2->unp_addr,
727 unp2->unp_addrlen);
728 unp3->unp_addrlen = unp2->unp_addrlen;
729 }
730 unp3->unp_flags = unp2->unp_flags;
731 so2 = so3;
732 }
733 error = unp_connect2(so, so2, PRU_CONNECT);
734 bad:
735 vput(vp);
736 bad2:
737 free(sun, M_SONAME);
738 return (error);
739 }
740
741 int
742 unp_connect2(so, so2, req)
743 struct socket *so;
744 struct socket *so2;
745 int req;
746 {
747 struct unpcb *unp = sotounpcb(so);
748 struct unpcb *unp2;
749
750 if (so2->so_type != so->so_type)
751 return (EPROTOTYPE);
752 unp2 = sotounpcb(so2);
753 unp->unp_conn = unp2;
754 switch (so->so_type) {
755
756 case SOCK_DGRAM:
757 unp->unp_nextref = unp2->unp_refs;
758 unp2->unp_refs = unp;
759 soisconnected(so);
760 break;
761
762 case SOCK_STREAM:
763 unp2->unp_conn = unp;
764 if (req == PRU_CONNECT &&
765 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT))
766 soisconnecting(so);
767 else
768 soisconnected(so);
769 soisconnected(so2);
770 break;
771
772 default:
773 panic("unp_connect2");
774 }
775 return (0);
776 }
777
778 void
779 unp_disconnect(unp)
780 struct unpcb *unp;
781 {
782 struct unpcb *unp2 = unp->unp_conn;
783
784 if (unp2 == 0)
785 return;
786 unp->unp_conn = 0;
787 switch (unp->unp_socket->so_type) {
788
789 case SOCK_DGRAM:
790 if (unp2->unp_refs == unp)
791 unp2->unp_refs = unp->unp_nextref;
792 else {
793 unp2 = unp2->unp_refs;
794 for (;;) {
795 if (unp2 == 0)
796 panic("unp_disconnect");
797 if (unp2->unp_nextref == unp)
798 break;
799 unp2 = unp2->unp_nextref;
800 }
801 unp2->unp_nextref = unp->unp_nextref;
802 }
803 unp->unp_nextref = 0;
804 unp->unp_socket->so_state &= ~SS_ISCONNECTED;
805 break;
806
807 case SOCK_STREAM:
808 soisdisconnected(unp->unp_socket);
809 unp2->unp_conn = 0;
810 soisdisconnected(unp2->unp_socket);
811 break;
812 }
813 }
814
815 #ifdef notdef
816 unp_abort(unp)
817 struct unpcb *unp;
818 {
819
820 unp_detach(unp);
821 }
822 #endif
823
824 void
825 unp_shutdown(unp)
826 struct unpcb *unp;
827 {
828 struct socket *so;
829
830 if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn &&
831 (so = unp->unp_conn->unp_socket))
832 socantrcvmore(so);
833 }
834
835 void
836 unp_drop(unp, errno)
837 struct unpcb *unp;
838 int errno;
839 {
840 struct socket *so = unp->unp_socket;
841
842 so->so_error = errno;
843 unp_disconnect(unp);
844 if (so->so_head) {
845 so->so_pcb = 0;
846 sofree(so);
847 if (unp->unp_addr)
848 free(unp->unp_addr, M_SONAME);
849 free(unp, M_PCB);
850 }
851 }
852
853 #ifdef notdef
854 unp_drain()
855 {
856
857 }
858 #endif
859
860 int
861 unp_externalize(rights)
862 struct mbuf *rights;
863 {
864 struct proc *p = curproc; /* XXX */
865 struct cmsghdr *cm = mtod(rights, struct cmsghdr *);
866 int i, *fdp;
867 struct file **rp;
868 struct file *fp;
869 int nfds, error = 0;
870
871 nfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm))) /
872 sizeof(struct file *);
873 rp = (struct file **)CMSG_DATA(cm);
874
875 fdp = malloc(nfds * sizeof(int), M_TEMP, M_WAITOK);
876
877 /* Make sure the recipient should be able to see the descriptors.. */
878 if (p->p_cwdi->cwdi_rdir != NULL) {
879 rp = (struct file **)CMSG_DATA(cm);
880 for (i = 0; i < nfds; i++) {
881 fp = *rp++;
882 /*
883 * If we are in a chroot'ed directory, and
884 * someone wants to pass us a directory, make
885 * sure it's inside the subtree we're allowed
886 * to access.
887 */
888 if (fp->f_type == DTYPE_VNODE) {
889 struct vnode *vp = (struct vnode *)fp->f_data;
890 if ((vp->v_type == VDIR) &&
891 !vn_isunder(vp, p->p_cwdi->cwdi_rdir, p)) {
892 error = EPERM;
893 break;
894 }
895 }
896 }
897 }
898
899 restart:
900 rp = (struct file **)CMSG_DATA(cm);
901 if (error != 0) {
902 for (i = 0; i < nfds; i++) {
903 fp = *rp;
904 /*
905 * zero the pointer before calling unp_discard,
906 * since it may end up in unp_gc()..
907 */
908 *rp++ = 0;
909 unp_discard(fp);
910 }
911 goto out;
912 }
913
914 /*
915 * First loop -- allocate file descriptor table slots for the
916 * new descriptors.
917 */
918 for (i = 0; i < nfds; i++) {
919 fp = *rp++;
920 if ((error = fdalloc(p, 0, &fdp[i])) != 0) {
921 /*
922 * Back out what we've done so far.
923 */
924 for (--i; i >= 0; i--)
925 fdremove(p->p_fd, fdp[i]);
926
927 if (error == ENOSPC) {
928 fdexpand(p);
929 error = 0;
930 } else {
931 /*
932 * This is the error that has historically
933 * been returned, and some callers may
934 * expect it.
935 */
936 error = EMSGSIZE;
937 }
938 goto restart;
939 }
940
941 /*
942 * Make the slot reference the descriptor so that
943 * fdalloc() works properly.. We finalize it all
944 * in the loop below.
945 */
946 p->p_fd->fd_ofiles[fdp[i]] = fp;
947 }
948
949 /*
950 * Now that adding them has succeeded, update all of the
951 * descriptor passing state.
952 */
953 rp = (struct file **)CMSG_DATA(cm);
954 for (i = 0; i < nfds; i++) {
955 fp = *rp++;
956 fp->f_msgcount--;
957 unp_rights--;
958 }
959
960 /*
961 * Copy temporary array to message and adjust length, in case of
962 * transition from large struct file pointers to ints.
963 */
964 memcpy(CMSG_DATA(cm), fdp, nfds * sizeof(int));
965 cm->cmsg_len = CMSG_LEN(nfds * sizeof(int));
966 rights->m_len = CMSG_SPACE(nfds * sizeof(int));
967 out:
968 free(fdp, M_TEMP);
969 return (error);
970 }
971
972 int
973 unp_internalize(control, p)
974 struct mbuf *control;
975 struct proc *p;
976 {
977 struct filedesc *fdescp = p->p_fd;
978 struct cmsghdr *newcm, *cm = mtod(control, struct cmsghdr *);
979 struct file **rp, **files;
980 struct file *fp;
981 int i, fd, *fdp;
982 int nfds;
983 u_int neededspace;
984
985 /* Sanity check the control message header */
986 if (cm->cmsg_type != SCM_RIGHTS || cm->cmsg_level != SOL_SOCKET ||
987 cm->cmsg_len != control->m_len)
988 return (EINVAL);
989
990 /* Verify that the file descriptors are valid */
991 nfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm))) / sizeof(int);
992 fdp = (int *)CMSG_DATA(cm);
993 for (i = 0; i < nfds; i++) {
994 fd = *fdp++;
995 if ((fp = fd_getfile(fdescp, fd)) == NULL)
996 return (EBADF);
997 simple_unlock(&fp->f_slock);
998 }
999
1000 /* Make sure we have room for the struct file pointers */
1001 neededspace = CMSG_SPACE(nfds * sizeof(struct file *)) -
1002 control->m_len;
1003 if (neededspace > M_TRAILINGSPACE(control)) {
1004
1005 /* allocate new space and copy header into it */
1006 newcm = malloc(
1007 CMSG_SPACE(nfds * sizeof(struct file *)),
1008 M_MBUF, M_WAITOK);
1009 if (newcm == NULL)
1010 return (E2BIG);
1011 memcpy(newcm, cm, sizeof(struct cmsghdr));
1012 files = (struct file **)CMSG_DATA(newcm);
1013 } else {
1014 /* we can convert in-place */
1015 newcm = NULL;
1016 files = (struct file **)CMSG_DATA(cm);
1017 }
1018
1019 /*
1020 * Transform the file descriptors into struct file pointers, in
1021 * reverse order so that if pointers are bigger than ints, the
1022 * int won't get until we're done.
1023 */
1024 fdp = (int *)CMSG_DATA(cm) + nfds - 1;
1025 rp = files + nfds - 1;
1026 for (i = 0; i < nfds; i++) {
1027 fp = fdescp->fd_ofiles[*fdp--];
1028 simple_lock(&fp->f_slock);
1029 #ifdef DIAGNOSTIC
1030 if (fp->f_iflags & FIF_WANTCLOSE)
1031 panic("unp_internalize: file already closed");
1032 #endif
1033 *rp-- = fp;
1034 fp->f_count++;
1035 fp->f_msgcount++;
1036 simple_unlock(&fp->f_slock);
1037 unp_rights++;
1038 }
1039
1040 if (newcm) {
1041 if (control->m_flags & M_EXT)
1042 MEXTREMOVE(control);
1043 MEXTADD(control, newcm,
1044 CMSG_SPACE(nfds * sizeof(struct file *)),
1045 M_MBUF, NULL, NULL);
1046 cm = newcm;
1047 }
1048
1049 /* adjust message & mbuf to note amount of space actually used. */
1050 cm->cmsg_len = CMSG_LEN(nfds * sizeof(struct file *));
1051 control->m_len = CMSG_SPACE(nfds * sizeof(struct file *));
1052
1053 return (0);
1054 }
1055
1056 struct mbuf *
1057 unp_addsockcred(p, control)
1058 struct proc *p;
1059 struct mbuf *control;
1060 {
1061 struct cmsghdr *cmp;
1062 struct sockcred *sc;
1063 struct mbuf *m, *n;
1064 int len, space, i;
1065
1066 len = CMSG_LEN(SOCKCREDSIZE(p->p_ucred->cr_ngroups));
1067 space = CMSG_SPACE(SOCKCREDSIZE(p->p_ucred->cr_ngroups));
1068
1069 m = m_get(M_WAIT, MT_CONTROL);
1070 if (space > MLEN) {
1071 if (space > MCLBYTES)
1072 MEXTMALLOC(m, space, M_WAITOK);
1073 else
1074 m_clget(m, M_WAIT);
1075 if ((m->m_flags & M_EXT) == 0) {
1076 m_free(m);
1077 return (control);
1078 }
1079 }
1080
1081 m->m_len = space;
1082 m->m_next = NULL;
1083 cmp = mtod(m, struct cmsghdr *);
1084 sc = (struct sockcred *)CMSG_DATA(cmp);
1085 cmp->cmsg_len = len;
1086 cmp->cmsg_level = SOL_SOCKET;
1087 cmp->cmsg_type = SCM_CREDS;
1088 sc->sc_uid = p->p_cred->p_ruid;
1089 sc->sc_euid = p->p_ucred->cr_uid;
1090 sc->sc_gid = p->p_cred->p_rgid;
1091 sc->sc_egid = p->p_ucred->cr_gid;
1092 sc->sc_ngroups = p->p_ucred->cr_ngroups;
1093 for (i = 0; i < sc->sc_ngroups; i++)
1094 sc->sc_groups[i] = p->p_ucred->cr_groups[i];
1095
1096 /*
1097 * If a control message already exists, append us to the end.
1098 */
1099 if (control != NULL) {
1100 for (n = control; n->m_next != NULL; n = n->m_next)
1101 ;
1102 n->m_next = m;
1103 } else
1104 control = m;
1105
1106 return (control);
1107 }
1108
1109 int unp_defer, unp_gcing;
1110 extern struct domain unixdomain;
1111
1112 /*
1113 * Comment added long after the fact explaining what's going on here.
1114 * Do a mark-sweep GC of file descriptors on the system, to free up
1115 * any which are caught in flight to an about-to-be-closed socket.
1116 *
1117 * Traditional mark-sweep gc's start at the "root", and mark
1118 * everything reachable from the root (which, in our case would be the
1119 * process table). The mark bits are cleared during the sweep.
1120 *
1121 * XXX For some inexplicable reason (perhaps because the file
1122 * descriptor tables used to live in the u area which could be swapped
1123 * out and thus hard to reach), we do multiple scans over the set of
1124 * descriptors, using use *two* mark bits per object (DEFER and MARK).
1125 * Whenever we find a descriptor which references other descriptors,
1126 * the ones it references are marked with both bits, and we iterate
1127 * over the whole file table until there are no more DEFER bits set.
1128 * We also make an extra pass *before* the GC to clear the mark bits,
1129 * which could have been cleared at almost no cost during the previous
1130 * sweep.
1131 *
1132 * XXX MP: this needs to run with locks such that no other thread of
1133 * control can create or destroy references to file descriptors. it
1134 * may be necessary to defer the GC until later (when the locking
1135 * situation is more hospitable); it may be necessary to push this
1136 * into a separate thread.
1137 */
1138 void
1139 unp_gc()
1140 {
1141 struct file *fp, *nextfp;
1142 struct socket *so, *so1;
1143 struct file **extra_ref, **fpp;
1144 int nunref, i;
1145
1146 if (unp_gcing)
1147 return;
1148 unp_gcing = 1;
1149 unp_defer = 0;
1150
1151 /* Clear mark bits */
1152 LIST_FOREACH(fp, &filehead, f_list)
1153 fp->f_flag &= ~(FMARK|FDEFER);
1154
1155 /*
1156 * Iterate over the set of descriptors, marking ones believed
1157 * (based on refcount) to be referenced from a process, and
1158 * marking for rescan descriptors which are queued on a socket.
1159 */
1160 do {
1161 LIST_FOREACH(fp, &filehead, f_list) {
1162 if (fp->f_flag & FDEFER) {
1163 fp->f_flag &= ~FDEFER;
1164 unp_defer--;
1165 #ifdef DIAGNOSTIC
1166 if (fp->f_count == 0)
1167 panic("unp_gc: deferred unreferenced socket");
1168 #endif
1169 } else {
1170 if (fp->f_count == 0)
1171 continue;
1172 if (fp->f_flag & FMARK)
1173 continue;
1174 if (fp->f_count == fp->f_msgcount)
1175 continue;
1176 }
1177 fp->f_flag |= FMARK;
1178
1179 if (fp->f_type != DTYPE_SOCKET ||
1180 (so = (struct socket *)fp->f_data) == 0)
1181 continue;
1182 if (so->so_proto->pr_domain != &unixdomain ||
1183 (so->so_proto->pr_flags&PR_RIGHTS) == 0)
1184 continue;
1185 #ifdef notdef
1186 if (so->so_rcv.sb_flags & SB_LOCK) {
1187 /*
1188 * This is problematical; it's not clear
1189 * we need to wait for the sockbuf to be
1190 * unlocked (on a uniprocessor, at least),
1191 * and it's also not clear what to do
1192 * if sbwait returns an error due to receipt
1193 * of a signal. If sbwait does return
1194 * an error, we'll go into an infinite
1195 * loop. Delete all of this for now.
1196 */
1197 (void) sbwait(&so->so_rcv);
1198 goto restart;
1199 }
1200 #endif
1201 unp_scan(so->so_rcv.sb_mb, unp_mark, 0);
1202 /*
1203 * mark descriptors referenced from sockets queued on the accept queue as well.
1204 */
1205 if (so->so_options & SO_ACCEPTCONN) {
1206 TAILQ_FOREACH(so1, &so->so_q0, so_qe) {
1207 unp_scan(so1->so_rcv.sb_mb, unp_mark, 0);
1208 }
1209 TAILQ_FOREACH(so1, &so->so_q, so_qe) {
1210 unp_scan(so1->so_rcv.sb_mb, unp_mark, 0);
1211 }
1212 }
1213
1214 }
1215 } while (unp_defer);
1216 /*
1217 * Sweep pass. Find unmarked descriptors, and free them.
1218 *
1219 * We grab an extra reference to each of the file table entries
1220 * that are not otherwise accessible and then free the rights
1221 * that are stored in messages on them.
1222 *
1223 * The bug in the original code is a little tricky, so I'll describe
1224 * what's wrong with it here.
1225 *
1226 * It is incorrect to simply unp_discard each entry for f_msgcount
1227 * times -- consider the case of sockets A and B that contain
1228 * references to each other. On a last close of some other socket,
1229 * we trigger a gc since the number of outstanding rights (unp_rights)
1230 * is non-zero. If during the sweep phase the gc code un_discards,
1231 * we end up doing a (full) closef on the descriptor. A closef on A
1232 * results in the following chain. Closef calls soo_close, which
1233 * calls soclose. Soclose calls first (through the switch
1234 * uipc_usrreq) unp_detach, which re-invokes unp_gc. Unp_gc simply
1235 * returns because the previous instance had set unp_gcing, and
1236 * we return all the way back to soclose, which marks the socket
1237 * with SS_NOFDREF, and then calls sofree. Sofree calls sorflush
1238 * to free up the rights that are queued in messages on the socket A,
1239 * i.e., the reference on B. The sorflush calls via the dom_dispose
1240 * switch unp_dispose, which unp_scans with unp_discard. This second
1241 * instance of unp_discard just calls closef on B.
1242 *
1243 * Well, a similar chain occurs on B, resulting in a sorflush on B,
1244 * which results in another closef on A. Unfortunately, A is already
1245 * being closed, and the descriptor has already been marked with
1246 * SS_NOFDREF, and soclose panics at this point.
1247 *
1248 * Here, we first take an extra reference to each inaccessible
1249 * descriptor. Then, if the inaccessible descriptor is a
1250 * socket, we call sorflush in case it is a Unix domain
1251 * socket. After we destroy all the rights carried in
1252 * messages, we do a last closef to get rid of our extra
1253 * reference. This is the last close, and the unp_detach etc
1254 * will shut down the socket.
1255 *
1256 * 91/09/19, bsy@cs.cmu.edu
1257 */
1258 extra_ref = malloc(nfiles * sizeof(struct file *), M_FILE, M_WAITOK);
1259 for (nunref = 0, fp = LIST_FIRST(&filehead), fpp = extra_ref; fp != 0;
1260 fp = nextfp) {
1261 nextfp = LIST_NEXT(fp, f_list);
1262 simple_lock(&fp->f_slock);
1263 if (fp->f_count != 0 &&
1264 fp->f_count == fp->f_msgcount && !(fp->f_flag & FMARK)) {
1265 *fpp++ = fp;
1266 nunref++;
1267 fp->f_count++;
1268 }
1269 simple_unlock(&fp->f_slock);
1270 }
1271 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) {
1272 fp = *fpp;
1273 simple_lock(&fp->f_slock);
1274 FILE_USE(fp);
1275 if (fp->f_type == DTYPE_SOCKET)
1276 sorflush((struct socket *)fp->f_data);
1277 FILE_UNUSE(fp, NULL);
1278 }
1279 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) {
1280 fp = *fpp;
1281 simple_lock(&fp->f_slock);
1282 FILE_USE(fp);
1283 (void) closef(fp, (struct proc *)0);
1284 }
1285 free((caddr_t)extra_ref, M_FILE);
1286 unp_gcing = 0;
1287 }
1288
1289 void
1290 unp_dispose(m)
1291 struct mbuf *m;
1292 {
1293
1294 if (m)
1295 unp_scan(m, unp_discard, 1);
1296 }
1297
1298 void
1299 unp_scan(m0, op, discard)
1300 struct mbuf *m0;
1301 void (*op)(struct file *);
1302 int discard;
1303 {
1304 struct mbuf *m;
1305 struct file **rp;
1306 struct cmsghdr *cm;
1307 int i;
1308 int qfds;
1309
1310 while (m0) {
1311 for (m = m0; m; m = m->m_next) {
1312 if (m->m_type == MT_CONTROL &&
1313 m->m_len >= sizeof(*cm)) {
1314 cm = mtod(m, struct cmsghdr *);
1315 if (cm->cmsg_level != SOL_SOCKET ||
1316 cm->cmsg_type != SCM_RIGHTS)
1317 continue;
1318 qfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm)))
1319 / sizeof(struct file *);
1320 rp = (struct file **)CMSG_DATA(cm);
1321 for (i = 0; i < qfds; i++) {
1322 struct file *fp = *rp;
1323 if (discard)
1324 *rp = 0;
1325 (*op)(fp);
1326 rp++;
1327 }
1328 break; /* XXX, but saves time */
1329 }
1330 }
1331 m0 = m0->m_nextpkt;
1332 }
1333 }
1334
1335 void
1336 unp_mark(fp)
1337 struct file *fp;
1338 {
1339 if (fp == NULL)
1340 return;
1341
1342 if (fp->f_flag & FMARK)
1343 return;
1344
1345 /* If we're already deferred, don't screw up the defer count */
1346 if (fp->f_flag & FDEFER)
1347 return;
1348
1349 /*
1350 * Minimize the number of deferrals... Sockets are the only
1351 * type of descriptor which can hold references to another
1352 * descriptor, so just mark other descriptors, and defer
1353 * unmarked sockets for the next pass.
1354 */
1355 if (fp->f_type == DTYPE_SOCKET) {
1356 unp_defer++;
1357 if (fp->f_count == 0)
1358 panic("unp_mark: queued unref");
1359 fp->f_flag |= FDEFER;
1360 } else {
1361 fp->f_flag |= FMARK;
1362 }
1363 return;
1364 }
1365
1366 void
1367 unp_discard(fp)
1368 struct file *fp;
1369 {
1370 if (fp == NULL)
1371 return;
1372 simple_lock(&fp->f_slock);
1373 fp->f_usecount++; /* i.e. FILE_USE(fp) sans locking */
1374 fp->f_msgcount--;
1375 simple_unlock(&fp->f_slock);
1376 unp_rights--;
1377 (void) closef(fp, (struct proc *)0);
1378 }
Cache object: 7f3e5e2c150687159a98c35a6ae02d72
|