Initial import from FreeBSD RELENG_4:
[dragonfly.git] / sys / kern / uipc_usrreq.c
CommitLineData
984263bc
MD
1/*
2 * Copyright (c) 1982, 1986, 1989, 1991, 1993
3 * The Regents of the University of California. All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 3. All advertising materials mentioning features or use of this software
14 * must display the following acknowledgement:
15 * This product includes software developed by the University of
16 * California, Berkeley and its contributors.
17 * 4. Neither the name of the University nor the names of its contributors
18 * may be used to endorse or promote products derived from this software
19 * without specific prior written permission.
20 *
21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * SUCH DAMAGE.
32 *
33 * From: @(#)uipc_usrreq.c 8.3 (Berkeley) 1/4/94
34 * $FreeBSD: src/sys/kern/uipc_usrreq.c,v 1.54.2.10 2003/03/04 17:28:09 nectar Exp $
35 */
36
37#include <sys/param.h>
38#include <sys/systm.h>
39#include <sys/kernel.h>
40#include <sys/domain.h>
41#include <sys/fcntl.h>
42#include <sys/malloc.h> /* XXX must be before <sys/file.h> */
43#include <sys/file.h>
44#include <sys/filedesc.h>
45#include <sys/mbuf.h>
46#include <sys/namei.h>
47#include <sys/proc.h>
48#include <sys/protosw.h>
49#include <sys/socket.h>
50#include <sys/socketvar.h>
51#include <sys/resourcevar.h>
52#include <sys/stat.h>
53#include <sys/sysctl.h>
54#include <sys/un.h>
55#include <sys/unpcb.h>
56#include <sys/vnode.h>
57
58#include <vm/vm_zone.h>
59
60static struct vm_zone *unp_zone;
61static unp_gen_t unp_gencnt;
62static u_int unp_count;
63
64static struct unp_head unp_shead, unp_dhead;
65
66/*
67 * Unix communications domain.
68 *
69 * TODO:
70 * SEQPACKET, RDM
71 * rethink name space problems
72 * need a proper out-of-band
73 * lock pushdown
74 */
75static struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL };
76static ino_t unp_ino; /* prototype for fake inode numbers */
77
78static int unp_attach __P((struct socket *));
79static void unp_detach __P((struct unpcb *));
80static int unp_bind __P((struct unpcb *,struct sockaddr *, struct proc *));
81static int unp_connect __P((struct socket *,struct sockaddr *,
82 struct proc *));
83static void unp_disconnect __P((struct unpcb *));
84static void unp_shutdown __P((struct unpcb *));
85static void unp_drop __P((struct unpcb *, int));
86static void unp_gc __P((void));
87static void unp_scan __P((struct mbuf *, void (*)(struct file *)));
88static void unp_mark __P((struct file *));
89static void unp_discard __P((struct file *));
90static int unp_internalize __P((struct mbuf *, struct proc *));
91static int unp_listen __P((struct unpcb *, struct proc *));
92
93static int
94uipc_abort(struct socket *so)
95{
96 struct unpcb *unp = sotounpcb(so);
97
98 if (unp == 0)
99 return EINVAL;
100 unp_drop(unp, ECONNABORTED);
101 unp_detach(unp);
102 sofree(so);
103 return 0;
104}
105
106static int
107uipc_accept(struct socket *so, struct sockaddr **nam)
108{
109 struct unpcb *unp = sotounpcb(so);
110
111 if (unp == 0)
112 return EINVAL;
113
114 /*
115 * Pass back name of connected socket,
116 * if it was bound and we are still connected
117 * (our peer may have closed already!).
118 */
119 if (unp->unp_conn && unp->unp_conn->unp_addr) {
120 *nam = dup_sockaddr((struct sockaddr *)unp->unp_conn->unp_addr,
121 1);
122 } else {
123 *nam = dup_sockaddr((struct sockaddr *)&sun_noname, 1);
124 }
125 return 0;
126}
127
128static int
129uipc_attach(struct socket *so, int proto, struct proc *p)
130{
131 struct unpcb *unp = sotounpcb(so);
132
133 if (unp != 0)
134 return EISCONN;
135 return unp_attach(so);
136}
137
138static int
139uipc_bind(struct socket *so, struct sockaddr *nam, struct proc *p)
140{
141 struct unpcb *unp = sotounpcb(so);
142
143 if (unp == 0)
144 return EINVAL;
145
146 return unp_bind(unp, nam, p);
147}
148
149static int
150uipc_connect(struct socket *so, struct sockaddr *nam, struct proc *p)
151{
152 struct unpcb *unp = sotounpcb(so);
153
154 if (unp == 0)
155 return EINVAL;
156 return unp_connect(so, nam, curproc);
157}
158
159static int
160uipc_connect2(struct socket *so1, struct socket *so2)
161{
162 struct unpcb *unp = sotounpcb(so1);
163
164 if (unp == 0)
165 return EINVAL;
166
167 return unp_connect2(so1, so2);
168}
169
170/* control is EOPNOTSUPP */
171
172static int
173uipc_detach(struct socket *so)
174{
175 struct unpcb *unp = sotounpcb(so);
176
177 if (unp == 0)
178 return EINVAL;
179
180 unp_detach(unp);
181 return 0;
182}
183
184static int
185uipc_disconnect(struct socket *so)
186{
187 struct unpcb *unp = sotounpcb(so);
188
189 if (unp == 0)
190 return EINVAL;
191 unp_disconnect(unp);
192 return 0;
193}
194
195static int
196uipc_listen(struct socket *so, struct proc *p)
197{
198 struct unpcb *unp = sotounpcb(so);
199
200 if (unp == 0 || unp->unp_vnode == 0)
201 return EINVAL;
202 return unp_listen(unp, p);
203}
204
205static int
206uipc_peeraddr(struct socket *so, struct sockaddr **nam)
207{
208 struct unpcb *unp = sotounpcb(so);
209
210 if (unp == 0)
211 return EINVAL;
212 if (unp->unp_conn && unp->unp_conn->unp_addr)
213 *nam = dup_sockaddr((struct sockaddr *)unp->unp_conn->unp_addr,
214 1);
215 else {
216 /*
217 * XXX: It seems that this test always fails even when
218 * connection is established. So, this else clause is
219 * added as workaround to return PF_LOCAL sockaddr.
220 */
221 *nam = dup_sockaddr((struct sockaddr *)&sun_noname, 1);
222 }
223 return 0;
224}
225
226static int
227uipc_rcvd(struct socket *so, int flags)
228{
229 struct unpcb *unp = sotounpcb(so);
230 struct socket *so2;
231 u_long newhiwat;
232
233 if (unp == 0)
234 return EINVAL;
235 switch (so->so_type) {
236 case SOCK_DGRAM:
237 panic("uipc_rcvd DGRAM?");
238 /*NOTREACHED*/
239
240 case SOCK_STREAM:
241 if (unp->unp_conn == 0)
242 break;
243 so2 = unp->unp_conn->unp_socket;
244 /*
245 * Adjust backpressure on sender
246 * and wakeup any waiting to write.
247 */
248 so2->so_snd.sb_mbmax += unp->unp_mbcnt - so->so_rcv.sb_mbcnt;
249 unp->unp_mbcnt = so->so_rcv.sb_mbcnt;
250 newhiwat = so2->so_snd.sb_hiwat + unp->unp_cc -
251 so->so_rcv.sb_cc;
252 (void)chgsbsize(so2->so_cred->cr_uidinfo, &so2->so_snd.sb_hiwat,
253 newhiwat, RLIM_INFINITY);
254 unp->unp_cc = so->so_rcv.sb_cc;
255 sowwakeup(so2);
256 break;
257
258 default:
259 panic("uipc_rcvd unknown socktype");
260 }
261 return 0;
262}
263
264/* pru_rcvoob is EOPNOTSUPP */
265
266static int
267uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
268 struct mbuf *control, struct proc *p)
269{
270 int error = 0;
271 struct unpcb *unp = sotounpcb(so);
272 struct socket *so2;
273 u_long newhiwat;
274
275 if (unp == 0) {
276 error = EINVAL;
277 goto release;
278 }
279 if (flags & PRUS_OOB) {
280 error = EOPNOTSUPP;
281 goto release;
282 }
283
284 if (control && (error = unp_internalize(control, p)))
285 goto release;
286
287 switch (so->so_type) {
288 case SOCK_DGRAM:
289 {
290 struct sockaddr *from;
291
292 if (nam) {
293 if (unp->unp_conn) {
294 error = EISCONN;
295 break;
296 }
297 error = unp_connect(so, nam, p);
298 if (error)
299 break;
300 } else {
301 if (unp->unp_conn == 0) {
302 error = ENOTCONN;
303 break;
304 }
305 }
306 so2 = unp->unp_conn->unp_socket;
307 if (unp->unp_addr)
308 from = (struct sockaddr *)unp->unp_addr;
309 else
310 from = &sun_noname;
311 if (sbappendaddr(&so2->so_rcv, from, m, control)) {
312 sorwakeup(so2);
313 m = 0;
314 control = 0;
315 } else
316 error = ENOBUFS;
317 if (nam)
318 unp_disconnect(unp);
319 break;
320 }
321
322 case SOCK_STREAM:
323 /* Connect if not connected yet. */
324 /*
325 * Note: A better implementation would complain
326 * if not equal to the peer's address.
327 */
328 if ((so->so_state & SS_ISCONNECTED) == 0) {
329 if (nam) {
330 error = unp_connect(so, nam, p);
331 if (error)
332 break; /* XXX */
333 } else {
334 error = ENOTCONN;
335 break;
336 }
337 }
338
339 if (so->so_state & SS_CANTSENDMORE) {
340 error = EPIPE;
341 break;
342 }
343 if (unp->unp_conn == 0)
344 panic("uipc_send connected but no connection?");
345 so2 = unp->unp_conn->unp_socket;
346 /*
347 * Send to paired receive port, and then reduce
348 * send buffer hiwater marks to maintain backpressure.
349 * Wake up readers.
350 */
351 if (control) {
352 if (sbappendcontrol(&so2->so_rcv, m, control))
353 control = 0;
354 } else
355 sbappend(&so2->so_rcv, m);
356 so->so_snd.sb_mbmax -=
357 so2->so_rcv.sb_mbcnt - unp->unp_conn->unp_mbcnt;
358 unp->unp_conn->unp_mbcnt = so2->so_rcv.sb_mbcnt;
359 newhiwat = so->so_snd.sb_hiwat -
360 (so2->so_rcv.sb_cc - unp->unp_conn->unp_cc);
361 (void)chgsbsize(so->so_cred->cr_uidinfo, &so->so_snd.sb_hiwat,
362 newhiwat, RLIM_INFINITY);
363 unp->unp_conn->unp_cc = so2->so_rcv.sb_cc;
364 sorwakeup(so2);
365 m = 0;
366 break;
367
368 default:
369 panic("uipc_send unknown socktype");
370 }
371
372 /*
373 * SEND_EOF is equivalent to a SEND followed by
374 * a SHUTDOWN.
375 */
376 if (flags & PRUS_EOF) {
377 socantsendmore(so);
378 unp_shutdown(unp);
379 }
380
381 if (control && error != 0)
382 unp_dispose(control);
383
384release:
385 if (control)
386 m_freem(control);
387 if (m)
388 m_freem(m);
389 return error;
390}
391
392static int
393uipc_sense(struct socket *so, struct stat *sb)
394{
395 struct unpcb *unp = sotounpcb(so);
396 struct socket *so2;
397
398 if (unp == 0)
399 return EINVAL;
400 sb->st_blksize = so->so_snd.sb_hiwat;
401 if (so->so_type == SOCK_STREAM && unp->unp_conn != 0) {
402 so2 = unp->unp_conn->unp_socket;
403 sb->st_blksize += so2->so_rcv.sb_cc;
404 }
405 sb->st_dev = NOUDEV;
406 if (unp->unp_ino == 0) /* make up a non-zero inode number */
407 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino;
408 sb->st_ino = unp->unp_ino;
409 return (0);
410}
411
412static int
413uipc_shutdown(struct socket *so)
414{
415 struct unpcb *unp = sotounpcb(so);
416
417 if (unp == 0)
418 return EINVAL;
419 socantsendmore(so);
420 unp_shutdown(unp);
421 return 0;
422}
423
424static int
425uipc_sockaddr(struct socket *so, struct sockaddr **nam)
426{
427 struct unpcb *unp = sotounpcb(so);
428
429 if (unp == 0)
430 return EINVAL;
431 if (unp->unp_addr)
432 *nam = dup_sockaddr((struct sockaddr *)unp->unp_addr, 1);
433 return 0;
434}
435
436struct pr_usrreqs uipc_usrreqs = {
437 uipc_abort, uipc_accept, uipc_attach, uipc_bind, uipc_connect,
438 uipc_connect2, pru_control_notsupp, uipc_detach, uipc_disconnect,
439 uipc_listen, uipc_peeraddr, uipc_rcvd, pru_rcvoob_notsupp,
440 uipc_send, uipc_sense, uipc_shutdown, uipc_sockaddr,
441 sosend, soreceive, sopoll
442};
443
444int
445uipc_ctloutput(so, sopt)
446 struct socket *so;
447 struct sockopt *sopt;
448{
449 struct unpcb *unp = sotounpcb(so);
450 int error;
451
452 switch (sopt->sopt_dir) {
453 case SOPT_GET:
454 switch (sopt->sopt_name) {
455 case LOCAL_PEERCRED:
456 if (unp->unp_flags & UNP_HAVEPC)
457 error = sooptcopyout(sopt, &unp->unp_peercred,
458 sizeof(unp->unp_peercred));
459 else {
460 if (so->so_type == SOCK_STREAM)
461 error = ENOTCONN;
462 else
463 error = EINVAL;
464 }
465 break;
466 default:
467 error = EOPNOTSUPP;
468 break;
469 }
470 break;
471 case SOPT_SET:
472 default:
473 error = EOPNOTSUPP;
474 break;
475 }
476 return (error);
477}
478
479/*
480 * Both send and receive buffers are allocated PIPSIZ bytes of buffering
481 * for stream sockets, although the total for sender and receiver is
482 * actually only PIPSIZ.
483 * Datagram sockets really use the sendspace as the maximum datagram size,
484 * and don't really want to reserve the sendspace. Their recvspace should
485 * be large enough for at least one max-size datagram plus address.
486 */
487#ifndef PIPSIZ
488#define PIPSIZ 8192
489#endif
490static u_long unpst_sendspace = PIPSIZ;
491static u_long unpst_recvspace = PIPSIZ;
492static u_long unpdg_sendspace = 2*1024; /* really max datagram size */
493static u_long unpdg_recvspace = 4*1024;
494
495static int unp_rights; /* file descriptors in flight */
496
497SYSCTL_DECL(_net_local_stream);
498SYSCTL_INT(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW,
499 &unpst_sendspace, 0, "");
500SYSCTL_INT(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW,
501 &unpst_recvspace, 0, "");
502SYSCTL_DECL(_net_local_dgram);
503SYSCTL_INT(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW,
504 &unpdg_sendspace, 0, "");
505SYSCTL_INT(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW,
506 &unpdg_recvspace, 0, "");
507SYSCTL_DECL(_net_local);
508SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, "");
509
510static int
511unp_attach(so)
512 struct socket *so;
513{
514 register struct unpcb *unp;
515 int error;
516
517 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
518 switch (so->so_type) {
519
520 case SOCK_STREAM:
521 error = soreserve(so, unpst_sendspace, unpst_recvspace);
522 break;
523
524 case SOCK_DGRAM:
525 error = soreserve(so, unpdg_sendspace, unpdg_recvspace);
526 break;
527
528 default:
529 panic("unp_attach");
530 }
531 if (error)
532 return (error);
533 }
534 unp = zalloc(unp_zone);
535 if (unp == NULL)
536 return (ENOBUFS);
537 bzero(unp, sizeof *unp);
538 unp->unp_gencnt = ++unp_gencnt;
539 unp_count++;
540 LIST_INIT(&unp->unp_refs);
541 unp->unp_socket = so;
542 unp->unp_rvnode = curproc->p_fd->fd_rdir;
543 LIST_INSERT_HEAD(so->so_type == SOCK_DGRAM ? &unp_dhead
544 : &unp_shead, unp, unp_link);
545 so->so_pcb = (caddr_t)unp;
546 return (0);
547}
548
549static void
550unp_detach(unp)
551 register struct unpcb *unp;
552{
553 LIST_REMOVE(unp, unp_link);
554 unp->unp_gencnt = ++unp_gencnt;
555 --unp_count;
556 if (unp->unp_vnode) {
557 unp->unp_vnode->v_socket = 0;
558 vrele(unp->unp_vnode);
559 unp->unp_vnode = 0;
560 }
561 if (unp->unp_conn)
562 unp_disconnect(unp);
563 while (!LIST_EMPTY(&unp->unp_refs))
564 unp_drop(LIST_FIRST(&unp->unp_refs), ECONNRESET);
565 soisdisconnected(unp->unp_socket);
566 unp->unp_socket->so_pcb = 0;
567 if (unp_rights) {
568 /*
569 * Normally the receive buffer is flushed later,
570 * in sofree, but if our receive buffer holds references
571 * to descriptors that are now garbage, we will dispose
572 * of those descriptor references after the garbage collector
573 * gets them (resulting in a "panic: closef: count < 0").
574 */
575 sorflush(unp->unp_socket);
576 unp_gc();
577 }
578 if (unp->unp_addr)
579 FREE(unp->unp_addr, M_SONAME);
580 zfree(unp_zone, unp);
581}
582
583static int
584unp_bind(unp, nam, p)
585 struct unpcb *unp;
586 struct sockaddr *nam;
587 struct proc *p;
588{
589 struct sockaddr_un *soun = (struct sockaddr_un *)nam;
590 register struct vnode *vp;
591 struct vattr vattr;
592 int error, namelen;
593 struct nameidata nd;
594 char buf[SOCK_MAXADDRLEN];
595
596 if (unp->unp_vnode != NULL)
597 return (EINVAL);
598 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path);
599 if (namelen <= 0)
600 return EINVAL;
601 strncpy(buf, soun->sun_path, namelen);
602 buf[namelen] = 0; /* null-terminate the string */
603 NDINIT(&nd, CREATE, NOFOLLOW | LOCKPARENT, UIO_SYSSPACE,
604 buf, p);
605/* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
606 error = namei(&nd);
607 if (error)
608 return (error);
609 vp = nd.ni_vp;
610 if (vp != NULL) {
611 NDFREE(&nd, NDF_ONLY_PNBUF);
612 if (nd.ni_dvp == vp)
613 vrele(nd.ni_dvp);
614 else
615 vput(nd.ni_dvp);
616 vrele(vp);
617 return (EADDRINUSE);
618 }
619 VATTR_NULL(&vattr);
620 vattr.va_type = VSOCK;
621 vattr.va_mode = (ACCESSPERMS & ~p->p_fd->fd_cmask);
622 VOP_LEASE(nd.ni_dvp, p, p->p_ucred, LEASE_WRITE);
623 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
624 NDFREE(&nd, NDF_ONLY_PNBUF);
625 vput(nd.ni_dvp);
626 if (error)
627 return (error);
628 vp = nd.ni_vp;
629 vp->v_socket = unp->unp_socket;
630 unp->unp_vnode = vp;
631 unp->unp_addr = (struct sockaddr_un *)dup_sockaddr(nam, 1);
632 VOP_UNLOCK(vp, 0, p);
633 return (0);
634}
635
636static int
637unp_connect(so, nam, p)
638 struct socket *so;
639 struct sockaddr *nam;
640 struct proc *p;
641{
642 register struct sockaddr_un *soun = (struct sockaddr_un *)nam;
643 register struct vnode *vp;
644 register struct socket *so2, *so3;
645 struct unpcb *unp, *unp2, *unp3;
646 int error, len;
647 struct nameidata nd;
648 char buf[SOCK_MAXADDRLEN];
649
650 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
651 if (len <= 0)
652 return EINVAL;
653 strncpy(buf, soun->sun_path, len);
654 buf[len] = 0;
655
656 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, buf, p);
657 error = namei(&nd);
658 if (error)
659 return (error);
660 vp = nd.ni_vp;
661 NDFREE(&nd, NDF_ONLY_PNBUF);
662 if (vp->v_type != VSOCK) {
663 error = ENOTSOCK;
664 goto bad;
665 }
666 error = VOP_ACCESS(vp, VWRITE, p->p_ucred, p);
667 if (error)
668 goto bad;
669 so2 = vp->v_socket;
670 if (so2 == 0) {
671 error = ECONNREFUSED;
672 goto bad;
673 }
674 if (so->so_type != so2->so_type) {
675 error = EPROTOTYPE;
676 goto bad;
677 }
678 if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
679 if ((so2->so_options & SO_ACCEPTCONN) == 0 ||
680 (so3 = sonewconn3(so2, 0, p)) == 0) {
681 error = ECONNREFUSED;
682 goto bad;
683 }
684 unp = sotounpcb(so);
685 unp2 = sotounpcb(so2);
686 unp3 = sotounpcb(so3);
687 if (unp2->unp_addr)
688 unp3->unp_addr = (struct sockaddr_un *)
689 dup_sockaddr((struct sockaddr *)
690 unp2->unp_addr, 1);
691
692 /*
693 * unp_peercred management:
694 *
695 * The connecter's (client's) credentials are copied
696 * from its process structure at the time of connect()
697 * (which is now).
698 */
699 cru2x(p->p_ucred, &unp3->unp_peercred);
700 unp3->unp_flags |= UNP_HAVEPC;
701 /*
702 * The receiver's (server's) credentials are copied
703 * from the unp_peercred member of socket on which the
704 * former called listen(); unp_listen() cached that
705 * process's credentials at that time so we can use
706 * them now.
707 */
708 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED,
709 ("unp_connect: listener without cached peercred"));
710 memcpy(&unp->unp_peercred, &unp2->unp_peercred,
711 sizeof(unp->unp_peercred));
712 unp->unp_flags |= UNP_HAVEPC;
713
714 so2 = so3;
715 }
716 error = unp_connect2(so, so2);
717bad:
718 vput(vp);
719 return (error);
720}
721
722int
723unp_connect2(so, so2)
724 register struct socket *so;
725 register struct socket *so2;
726{
727 register struct unpcb *unp = sotounpcb(so);
728 register struct unpcb *unp2;
729
730 if (so2->so_type != so->so_type)
731 return (EPROTOTYPE);
732 unp2 = sotounpcb(so2);
733 unp->unp_conn = unp2;
734 switch (so->so_type) {
735
736 case SOCK_DGRAM:
737 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
738 soisconnected(so);
739 break;
740
741 case SOCK_STREAM:
742 unp2->unp_conn = unp;
743 soisconnected(so);
744 soisconnected(so2);
745 break;
746
747 default:
748 panic("unp_connect2");
749 }
750 return (0);
751}
752
753static void
754unp_disconnect(unp)
755 struct unpcb *unp;
756{
757 register struct unpcb *unp2 = unp->unp_conn;
758
759 if (unp2 == 0)
760 return;
761 unp->unp_conn = 0;
762 switch (unp->unp_socket->so_type) {
763
764 case SOCK_DGRAM:
765 LIST_REMOVE(unp, unp_reflink);
766 unp->unp_socket->so_state &= ~SS_ISCONNECTED;
767 break;
768
769 case SOCK_STREAM:
770 soisdisconnected(unp->unp_socket);
771 unp2->unp_conn = 0;
772 soisdisconnected(unp2->unp_socket);
773 break;
774 }
775}
776
777#ifdef notdef
778void
779unp_abort(unp)
780 struct unpcb *unp;
781{
782
783 unp_detach(unp);
784}
785#endif
786
787static int
788prison_unpcb(struct proc *p, struct unpcb *unp)
789{
790 if (!p->p_prison)
791 return (0);
792 if (p->p_fd->fd_rdir == unp->unp_rvnode)
793 return (0);
794 return (1);
795}
796
797static int
798unp_pcblist(SYSCTL_HANDLER_ARGS)
799{
800 int error, i, n;
801 struct unpcb *unp, **unp_list;
802 unp_gen_t gencnt;
803 struct xunpgen xug;
804 struct unp_head *head;
805
806 head = ((intptr_t)arg1 == SOCK_DGRAM ? &unp_dhead : &unp_shead);
807
808 /*
809 * The process of preparing the PCB list is too time-consuming and
810 * resource-intensive to repeat twice on every request.
811 */
812 if (req->oldptr == 0) {
813 n = unp_count;
814 req->oldidx = 2 * (sizeof xug)
815 + (n + n/8) * sizeof(struct xunpcb);
816 return 0;
817 }
818
819 if (req->newptr != 0)
820 return EPERM;
821
822 /*
823 * OK, now we're committed to doing something.
824 */
825 gencnt = unp_gencnt;
826 n = unp_count;
827
828 xug.xug_len = sizeof xug;
829 xug.xug_count = n;
830 xug.xug_gen = gencnt;
831 xug.xug_sogen = so_gencnt;
832 error = SYSCTL_OUT(req, &xug, sizeof xug);
833 if (error)
834 return error;
835
836 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
837 if (unp_list == 0)
838 return ENOMEM;
839
840 for (unp = LIST_FIRST(head), i = 0; unp && i < n;
841 unp = LIST_NEXT(unp, unp_link)) {
842 if (unp->unp_gencnt <= gencnt && !prison_unpcb(req->p, unp))
843 unp_list[i++] = unp;
844 }
845 n = i; /* in case we lost some during malloc */
846
847 error = 0;
848 for (i = 0; i < n; i++) {
849 unp = unp_list[i];
850 if (unp->unp_gencnt <= gencnt) {
851 struct xunpcb xu;
852 xu.xu_len = sizeof xu;
853 xu.xu_unpp = unp;
854 /*
855 * XXX - need more locking here to protect against
856 * connect/disconnect races for SMP.
857 */
858 if (unp->unp_addr)
859 bcopy(unp->unp_addr, &xu.xu_addr,
860 unp->unp_addr->sun_len);
861 if (unp->unp_conn && unp->unp_conn->unp_addr)
862 bcopy(unp->unp_conn->unp_addr,
863 &xu.xu_caddr,
864 unp->unp_conn->unp_addr->sun_len);
865 bcopy(unp, &xu.xu_unp, sizeof *unp);
866 sotoxsocket(unp->unp_socket, &xu.xu_socket);
867 error = SYSCTL_OUT(req, &xu, sizeof xu);
868 }
869 }
870 if (!error) {
871 /*
872 * Give the user an updated idea of our state.
873 * If the generation differs from what we told
874 * her before, she knows that something happened
875 * while we were processing this request, and it
876 * might be necessary to retry.
877 */
878 xug.xug_gen = unp_gencnt;
879 xug.xug_sogen = so_gencnt;
880 xug.xug_count = unp_count;
881 error = SYSCTL_OUT(req, &xug, sizeof xug);
882 }
883 free(unp_list, M_TEMP);
884 return error;
885}
886
887SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLFLAG_RD,
888 (caddr_t)(long)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
889 "List of active local datagram sockets");
890SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLFLAG_RD,
891 (caddr_t)(long)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
892 "List of active local stream sockets");
893
894static void
895unp_shutdown(unp)
896 struct unpcb *unp;
897{
898 struct socket *so;
899
900 if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn &&
901 (so = unp->unp_conn->unp_socket))
902 socantrcvmore(so);
903}
904
905static void
906unp_drop(unp, errno)
907 struct unpcb *unp;
908 int errno;
909{
910 struct socket *so = unp->unp_socket;
911
912 so->so_error = errno;
913 unp_disconnect(unp);
914}
915
916#ifdef notdef
917void
918unp_drain()
919{
920
921}
922#endif
923
924int
925unp_externalize(rights)
926 struct mbuf *rights;
927{
928 struct proc *p = curproc; /* XXX */
929 register int i;
930 register struct cmsghdr *cm = mtod(rights, struct cmsghdr *);
931 register int *fdp;
932 register struct file **rp;
933 register struct file *fp;
934 int newfds = (cm->cmsg_len - (CMSG_DATA(cm) - (u_char *)cm))
935 / sizeof (struct file *);
936 int f;
937
938 /*
939 * if the new FD's will not fit, then we free them all
940 */
941 if (!fdavail(p, newfds)) {
942 rp = (struct file **)CMSG_DATA(cm);
943 for (i = 0; i < newfds; i++) {
944 fp = *rp;
945 /*
946 * zero the pointer before calling unp_discard,
947 * since it may end up in unp_gc()..
948 */
949 *rp++ = 0;
950 unp_discard(fp);
951 }
952 return (EMSGSIZE);
953 }
954 /*
955 * now change each pointer to an fd in the global table to
956 * an integer that is the index to the local fd table entry
957 * that we set up to point to the global one we are transferring.
958 * If sizeof (struct file *) is bigger than or equal to sizeof int,
959 * then do it in forward order. In that case, an integer will
960 * always come in the same place or before its corresponding
961 * struct file pointer.
962 * If sizeof (struct file *) is smaller than sizeof int, then
963 * do it in reverse order.
964 */
965 if (sizeof (struct file *) >= sizeof (int)) {
966 fdp = (int *)(cm + 1);
967 rp = (struct file **)CMSG_DATA(cm);
968 for (i = 0; i < newfds; i++) {
969 if (fdalloc(p, 0, &f))
970 panic("unp_externalize");
971 fp = *rp++;
972 p->p_fd->fd_ofiles[f] = fp;
973 fp->f_msgcount--;
974 unp_rights--;
975 *fdp++ = f;
976 }
977 } else {
978 fdp = (int *)(cm + 1) + newfds - 1;
979 rp = (struct file **)CMSG_DATA(cm) + newfds - 1;
980 for (i = 0; i < newfds; i++) {
981 if (fdalloc(p, 0, &f))
982 panic("unp_externalize");
983 fp = *rp--;
984 p->p_fd->fd_ofiles[f] = fp;
985 fp->f_msgcount--;
986 unp_rights--;
987 *fdp-- = f;
988 }
989 }
990
991 /*
992 * Adjust length, in case sizeof(struct file *) and sizeof(int)
993 * differs.
994 */
995 cm->cmsg_len = CMSG_LEN(newfds * sizeof(int));
996 rights->m_len = cm->cmsg_len;
997 return (0);
998}
999
1000void
1001unp_init(void)
1002{
1003 unp_zone = zinit("unpcb", sizeof(struct unpcb), nmbclusters, 0, 0);
1004 if (unp_zone == 0)
1005 panic("unp_init");
1006 LIST_INIT(&unp_dhead);
1007 LIST_INIT(&unp_shead);
1008}
1009
1010#ifndef MIN
1011#define MIN(a,b) (((a)<(b))?(a):(b))
1012#endif
1013
1014static int
1015unp_internalize(control, p)
1016 struct mbuf *control;
1017 struct proc *p;
1018{
1019 struct filedesc *fdescp = p->p_fd;
1020 register struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1021 register struct file **rp;
1022 register struct file *fp;
1023 register int i, fd, *fdp;
1024 register struct cmsgcred *cmcred;
1025 int oldfds;
1026 u_int newlen;
1027
1028 if ((cm->cmsg_type != SCM_RIGHTS && cm->cmsg_type != SCM_CREDS) ||
1029 cm->cmsg_level != SOL_SOCKET || cm->cmsg_len != control->m_len)
1030 return (EINVAL);
1031
1032 /*
1033 * Fill in credential information.
1034 */
1035 if (cm->cmsg_type == SCM_CREDS) {
1036 cmcred = (struct cmsgcred *)(cm + 1);
1037 cmcred->cmcred_pid = p->p_pid;
1038 cmcred->cmcred_uid = p->p_cred->p_ruid;
1039 cmcred->cmcred_gid = p->p_cred->p_rgid;
1040 cmcred->cmcred_euid = p->p_ucred->cr_uid;
1041 cmcred->cmcred_ngroups = MIN(p->p_ucred->cr_ngroups,
1042 CMGROUP_MAX);
1043 for (i = 0; i < cmcred->cmcred_ngroups; i++)
1044 cmcred->cmcred_groups[i] = p->p_ucred->cr_groups[i];
1045 return(0);
1046 }
1047
1048 oldfds = (cm->cmsg_len - sizeof (*cm)) / sizeof (int);
1049 /*
1050 * check that all the FDs passed in refer to legal OPEN files
1051 * If not, reject the entire operation.
1052 */
1053 fdp = (int *)(cm + 1);
1054 for (i = 0; i < oldfds; i++) {
1055 fd = *fdp++;
1056 if ((unsigned)fd >= fdescp->fd_nfiles ||
1057 fdescp->fd_ofiles[fd] == NULL)
1058 return (EBADF);
1059 if (fdescp->fd_ofiles[fd]->f_type == DTYPE_KQUEUE)
1060 return (EOPNOTSUPP);
1061 }
1062 /*
1063 * Now replace the integer FDs with pointers to
1064 * the associated global file table entry..
1065 * Allocate a bigger buffer as necessary. But if an cluster is not
1066 * enough, return E2BIG.
1067 */
1068 newlen = CMSG_LEN(oldfds * sizeof(struct file *));
1069 if (newlen > MCLBYTES)
1070 return (E2BIG);
1071 if (newlen - control->m_len > M_TRAILINGSPACE(control)) {
1072 if (control->m_flags & M_EXT)
1073 return (E2BIG);
1074 MCLGET(control, M_WAIT);
1075 if ((control->m_flags & M_EXT) == 0)
1076 return (ENOBUFS);
1077
1078 /* copy the data to the cluster */
1079 memcpy(mtod(control, char *), cm, cm->cmsg_len);
1080 cm = mtod(control, struct cmsghdr *);
1081 }
1082
1083 /*
1084 * Adjust length, in case sizeof(struct file *) and sizeof(int)
1085 * differs.
1086 */
1087 control->m_len = cm->cmsg_len = newlen;
1088
1089 /*
1090 * Transform the file descriptors into struct file pointers.
1091 * If sizeof (struct file *) is bigger than or equal to sizeof int,
1092 * then do it in reverse order so that the int won't get until
1093 * we're done.
1094 * If sizeof (struct file *) is smaller than sizeof int, then
1095 * do it in forward order.
1096 */
1097 if (sizeof (struct file *) >= sizeof (int)) {
1098 fdp = (int *)(cm + 1) + oldfds - 1;
1099 rp = (struct file **)CMSG_DATA(cm) + oldfds - 1;
1100 for (i = 0; i < oldfds; i++) {
1101 fp = fdescp->fd_ofiles[*fdp--];
1102 *rp-- = fp;
1103 fp->f_count++;
1104 fp->f_msgcount++;
1105 unp_rights++;
1106 }
1107 } else {
1108 fdp = (int *)(cm + 1);
1109 rp = (struct file **)CMSG_DATA(cm);
1110 for (i = 0; i < oldfds; i++) {
1111 fp = fdescp->fd_ofiles[*fdp++];
1112 *rp++ = fp;
1113 fp->f_count++;
1114 fp->f_msgcount++;
1115 unp_rights++;
1116 }
1117 }
1118 return (0);
1119}
1120
1121static int unp_defer, unp_gcing;
1122
1123static void
1124unp_gc()
1125{
1126 register struct file *fp, *nextfp;
1127 register struct socket *so;
1128 struct file **extra_ref, **fpp;
1129 int nunref, i;
1130
1131 if (unp_gcing)
1132 return;
1133 unp_gcing = 1;
1134 unp_defer = 0;
1135 /*
1136 * before going through all this, set all FDs to
1137 * be NOT defered and NOT externally accessible
1138 */
1139 LIST_FOREACH(fp, &filehead, f_list)
1140 fp->f_flag &= ~(FMARK|FDEFER);
1141 do {
1142 LIST_FOREACH(fp, &filehead, f_list) {
1143 /*
1144 * If the file is not open, skip it
1145 */
1146 if (fp->f_count == 0)
1147 continue;
1148 /*
1149 * If we already marked it as 'defer' in a
1150 * previous pass, then try process it this time
1151 * and un-mark it
1152 */
1153 if (fp->f_flag & FDEFER) {
1154 fp->f_flag &= ~FDEFER;
1155 unp_defer--;
1156 } else {
1157 /*
1158 * if it's not defered, then check if it's
1159 * already marked.. if so skip it
1160 */
1161 if (fp->f_flag & FMARK)
1162 continue;
1163 /*
1164 * If all references are from messages
1165 * in transit, then skip it. it's not
1166 * externally accessible.
1167 */
1168 if (fp->f_count == fp->f_msgcount)
1169 continue;
1170 /*
1171 * If it got this far then it must be
1172 * externally accessible.
1173 */
1174 fp->f_flag |= FMARK;
1175 }
1176 /*
1177 * either it was defered, or it is externally
1178 * accessible and not already marked so.
1179 * Now check if it is possibly one of OUR sockets.
1180 */
1181 if (fp->f_type != DTYPE_SOCKET ||
1182 (so = (struct socket *)fp->f_data) == 0)
1183 continue;
1184 if (so->so_proto->pr_domain != &localdomain ||
1185 (so->so_proto->pr_flags&PR_RIGHTS) == 0)
1186 continue;
1187#ifdef notdef
1188 if (so->so_rcv.sb_flags & SB_LOCK) {
1189 /*
1190 * This is problematical; it's not clear
1191 * we need to wait for the sockbuf to be
1192 * unlocked (on a uniprocessor, at least),
1193 * and it's also not clear what to do
1194 * if sbwait returns an error due to receipt
1195 * of a signal. If sbwait does return
1196 * an error, we'll go into an infinite
1197 * loop. Delete all of this for now.
1198 */
1199 (void) sbwait(&so->so_rcv);
1200 goto restart;
1201 }
1202#endif
1203 /*
1204 * So, Ok, it's one of our sockets and it IS externally
1205 * accessible (or was defered). Now we look
1206 * to see if we hold any file descriptors in its
1207 * message buffers. Follow those links and mark them
1208 * as accessible too.
1209 */
1210 unp_scan(so->so_rcv.sb_mb, unp_mark);
1211 }
1212 } while (unp_defer);
1213 /*
1214 * We grab an extra reference to each of the file table entries
1215 * that are not otherwise accessible and then free the rights
1216 * that are stored in messages on them.
1217 *
1218 * The bug in the orginal code is a little tricky, so I'll describe
1219 * what's wrong with it here.
1220 *
1221 * It is incorrect to simply unp_discard each entry for f_msgcount
1222 * times -- consider the case of sockets A and B that contain
1223 * references to each other. On a last close of some other socket,
1224 * we trigger a gc since the number of outstanding rights (unp_rights)
1225 * is non-zero. If during the sweep phase the gc code un_discards,
1226 * we end up doing a (full) closef on the descriptor. A closef on A
1227 * results in the following chain. Closef calls soo_close, which
1228 * calls soclose. Soclose calls first (through the switch
1229 * uipc_usrreq) unp_detach, which re-invokes unp_gc. Unp_gc simply
1230 * returns because the previous instance had set unp_gcing, and
1231 * we return all the way back to soclose, which marks the socket
1232 * with SS_NOFDREF, and then calls sofree. Sofree calls sorflush
1233 * to free up the rights that are queued in messages on the socket A,
1234 * i.e., the reference on B. The sorflush calls via the dom_dispose
1235 * switch unp_dispose, which unp_scans with unp_discard. This second
1236 * instance of unp_discard just calls closef on B.
1237 *
1238 * Well, a similar chain occurs on B, resulting in a sorflush on B,
1239 * which results in another closef on A. Unfortunately, A is already
1240 * being closed, and the descriptor has already been marked with
1241 * SS_NOFDREF, and soclose panics at this point.
1242 *
1243 * Here, we first take an extra reference to each inaccessible
1244 * descriptor. Then, we call sorflush ourself, since we know
1245 * it is a Unix domain socket anyhow. After we destroy all the
1246 * rights carried in messages, we do a last closef to get rid
1247 * of our extra reference. This is the last close, and the
1248 * unp_detach etc will shut down the socket.
1249 *
1250 * 91/09/19, bsy@cs.cmu.edu
1251 */
1252 extra_ref = malloc(nfiles * sizeof(struct file *), M_FILE, M_WAITOK);
1253 for (nunref = 0, fp = LIST_FIRST(&filehead), fpp = extra_ref; fp != 0;
1254 fp = nextfp) {
1255 nextfp = LIST_NEXT(fp, f_list);
1256 /*
1257 * If it's not open, skip it
1258 */
1259 if (fp->f_count == 0)
1260 continue;
1261 /*
1262 * If all refs are from msgs, and it's not marked accessible
1263 * then it must be referenced from some unreachable cycle
1264 * of (shut-down) FDs, so include it in our
1265 * list of FDs to remove
1266 */
1267 if (fp->f_count == fp->f_msgcount && !(fp->f_flag & FMARK)) {
1268 *fpp++ = fp;
1269 nunref++;
1270 fp->f_count++;
1271 }
1272 }
1273 /*
1274 * for each FD on our hit list, do the following two things
1275 */
1276 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) {
1277 struct file *tfp = *fpp;
1278 if (tfp->f_type == DTYPE_SOCKET && tfp->f_data != NULL)
1279 sorflush((struct socket *)(tfp->f_data));
1280 }
1281 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp)
1282 closef(*fpp, (struct proc *) NULL);
1283 free((caddr_t)extra_ref, M_FILE);
1284 unp_gcing = 0;
1285}
1286
1287void
1288unp_dispose(m)
1289 struct mbuf *m;
1290{
1291
1292 if (m)
1293 unp_scan(m, unp_discard);
1294}
1295
1296static int
1297unp_listen(unp, p)
1298 struct unpcb *unp;
1299 struct proc *p;
1300{
1301
1302 cru2x(p->p_ucred, &unp->unp_peercred);
1303 unp->unp_flags |= UNP_HAVEPCCACHED;
1304 return (0);
1305}
1306
1307static void
1308unp_scan(m0, op)
1309 register struct mbuf *m0;
1310 void (*op) __P((struct file *));
1311{
1312 register struct mbuf *m;
1313 register struct file **rp;
1314 register struct cmsghdr *cm;
1315 register int i;
1316 int qfds;
1317
1318 while (m0) {
1319 for (m = m0; m; m = m->m_next)
1320 if (m->m_type == MT_CONTROL &&
1321 m->m_len >= sizeof(*cm)) {
1322 cm = mtod(m, struct cmsghdr *);
1323 if (cm->cmsg_level != SOL_SOCKET ||
1324 cm->cmsg_type != SCM_RIGHTS)
1325 continue;
1326 qfds = (cm->cmsg_len -
1327 (CMSG_DATA(cm) - (u_char *)cm))
1328 / sizeof (struct file *);
1329 rp = (struct file **)CMSG_DATA(cm);
1330 for (i = 0; i < qfds; i++)
1331 (*op)(*rp++);
1332 break; /* XXX, but saves time */
1333 }
1334 m0 = m0->m_act;
1335 }
1336}
1337
1338static void
1339unp_mark(fp)
1340 struct file *fp;
1341{
1342
1343 if (fp->f_flag & FMARK)
1344 return;
1345 unp_defer++;
1346 fp->f_flag |= (FMARK|FDEFER);
1347}
1348
1349static void
1350unp_discard(fp)
1351 struct file *fp;
1352{
1353
1354 fp->f_msgcount--;
1355 unp_rights--;
1356 (void) closef(fp, (struct proc *)NULL);
1357}