Once we distribute socket protocol processing requests to different
[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 $
e4700d00 35 * $DragonFly: src/sys/kern/uipc_usrreq.c,v 1.12 2004/03/05 16:57:15 hsu Exp $
984263bc
MD
36 */
37
38#include <sys/param.h>
39#include <sys/systm.h>
40#include <sys/kernel.h>
41#include <sys/domain.h>
42#include <sys/fcntl.h>
43#include <sys/malloc.h> /* XXX must be before <sys/file.h> */
dadab5e9 44#include <sys/proc.h>
984263bc
MD
45#include <sys/file.h>
46#include <sys/filedesc.h>
47#include <sys/mbuf.h>
48#include <sys/namei.h>
984263bc
MD
49#include <sys/protosw.h>
50#include <sys/socket.h>
51#include <sys/socketvar.h>
52#include <sys/resourcevar.h>
53#include <sys/stat.h>
54#include <sys/sysctl.h>
55#include <sys/un.h>
56#include <sys/unpcb.h>
57#include <sys/vnode.h>
dadab5e9 58#include <sys/file2.h>
984263bc
MD
59
60#include <vm/vm_zone.h>
61
62static struct vm_zone *unp_zone;
63static unp_gen_t unp_gencnt;
64static u_int unp_count;
65
66static struct unp_head unp_shead, unp_dhead;
67
68/*
69 * Unix communications domain.
70 *
71 * TODO:
72 * SEQPACKET, RDM
73 * rethink name space problems
74 * need a proper out-of-band
75 * lock pushdown
76 */
77static struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL };
78static ino_t unp_ino; /* prototype for fake inode numbers */
79
e4700d00 80static int unp_attach (struct socket *, struct pru_attach_info *);
402ed7e1
RG
81static void unp_detach (struct unpcb *);
82static int unp_bind (struct unpcb *,struct sockaddr *, struct thread *);
83static int unp_connect (struct socket *,struct sockaddr *,
84 struct thread *);
85static void unp_disconnect (struct unpcb *);
86static void unp_shutdown (struct unpcb *);
87static void unp_drop (struct unpcb *, int);
88static void unp_gc (void);
89static void unp_scan (struct mbuf *, void (*)(struct file *));
90static void unp_mark (struct file *);
91static void unp_discard (struct file *);
92static int unp_internalize (struct mbuf *, struct thread *);
93static int unp_listen (struct unpcb *, struct thread *);
984263bc
MD
94
95static int
96uipc_abort(struct socket *so)
97{
98 struct unpcb *unp = sotounpcb(so);
99
100 if (unp == 0)
101 return EINVAL;
102 unp_drop(unp, ECONNABORTED);
103 unp_detach(unp);
104 sofree(so);
105 return 0;
106}
107
108static int
109uipc_accept(struct socket *so, struct sockaddr **nam)
110{
111 struct unpcb *unp = sotounpcb(so);
112
113 if (unp == 0)
114 return EINVAL;
115
116 /*
117 * Pass back name of connected socket,
118 * if it was bound and we are still connected
119 * (our peer may have closed already!).
120 */
121 if (unp->unp_conn && unp->unp_conn->unp_addr) {
122 *nam = dup_sockaddr((struct sockaddr *)unp->unp_conn->unp_addr,
123 1);
124 } else {
125 *nam = dup_sockaddr((struct sockaddr *)&sun_noname, 1);
126 }
127 return 0;
128}
129
130static int
e4700d00 131uipc_attach(struct socket *so, int proto, struct pru_attach_info *ai)
984263bc
MD
132{
133 struct unpcb *unp = sotounpcb(so);
134
135 if (unp != 0)
136 return EISCONN;
e4700d00 137 return unp_attach(so, ai);
984263bc
MD
138}
139
140static int
dadab5e9 141uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
984263bc
MD
142{
143 struct unpcb *unp = sotounpcb(so);
144
145 if (unp == 0)
146 return EINVAL;
dadab5e9 147 return unp_bind(unp, nam, td);
984263bc
MD
148}
149
150static int
dadab5e9 151uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
984263bc
MD
152{
153 struct unpcb *unp = sotounpcb(so);
154
155 if (unp == 0)
156 return EINVAL;
dadab5e9 157 return unp_connect(so, nam, td);
984263bc
MD
158}
159
160static int
161uipc_connect2(struct socket *so1, struct socket *so2)
162{
163 struct unpcb *unp = sotounpcb(so1);
164
165 if (unp == 0)
166 return EINVAL;
167
168 return unp_connect2(so1, so2);
169}
170
171/* control is EOPNOTSUPP */
172
173static int
174uipc_detach(struct socket *so)
175{
176 struct unpcb *unp = sotounpcb(so);
177
178 if (unp == 0)
179 return EINVAL;
180
181 unp_detach(unp);
182 return 0;
183}
184
185static int
186uipc_disconnect(struct socket *so)
187{
188 struct unpcb *unp = sotounpcb(so);
189
190 if (unp == 0)
191 return EINVAL;
192 unp_disconnect(unp);
193 return 0;
194}
195
196static int
dadab5e9 197uipc_listen(struct socket *so, struct thread *td)
984263bc
MD
198{
199 struct unpcb *unp = sotounpcb(so);
200
201 if (unp == 0 || unp->unp_vnode == 0)
202 return EINVAL;
dadab5e9 203 return unp_listen(unp, td);
984263bc
MD
204}
205
206static int
207uipc_peeraddr(struct socket *so, struct sockaddr **nam)
208{
209 struct unpcb *unp = sotounpcb(so);
210
211 if (unp == 0)
212 return EINVAL;
213 if (unp->unp_conn && unp->unp_conn->unp_addr)
214 *nam = dup_sockaddr((struct sockaddr *)unp->unp_conn->unp_addr,
215 1);
216 else {
217 /*
218 * XXX: It seems that this test always fails even when
219 * connection is established. So, this else clause is
220 * added as workaround to return PF_LOCAL sockaddr.
221 */
222 *nam = dup_sockaddr((struct sockaddr *)&sun_noname, 1);
223 }
224 return 0;
225}
226
227static int
228uipc_rcvd(struct socket *so, int flags)
229{
230 struct unpcb *unp = sotounpcb(so);
231 struct socket *so2;
232 u_long newhiwat;
233
234 if (unp == 0)
235 return EINVAL;
236 switch (so->so_type) {
237 case SOCK_DGRAM:
238 panic("uipc_rcvd DGRAM?");
239 /*NOTREACHED*/
240
241 case SOCK_STREAM:
242 if (unp->unp_conn == 0)
243 break;
244 so2 = unp->unp_conn->unp_socket;
245 /*
246 * Adjust backpressure on sender
247 * and wakeup any waiting to write.
248 */
249 so2->so_snd.sb_mbmax += unp->unp_mbcnt - so->so_rcv.sb_mbcnt;
250 unp->unp_mbcnt = so->so_rcv.sb_mbcnt;
251 newhiwat = so2->so_snd.sb_hiwat + unp->unp_cc -
252 so->so_rcv.sb_cc;
253 (void)chgsbsize(so2->so_cred->cr_uidinfo, &so2->so_snd.sb_hiwat,
254 newhiwat, RLIM_INFINITY);
255 unp->unp_cc = so->so_rcv.sb_cc;
256 sowwakeup(so2);
257 break;
258
259 default:
260 panic("uipc_rcvd unknown socktype");
261 }
262 return 0;
263}
264
265/* pru_rcvoob is EOPNOTSUPP */
266
267static int
268uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
dadab5e9 269 struct mbuf *control, struct thread *td)
984263bc
MD
270{
271 int error = 0;
272 struct unpcb *unp = sotounpcb(so);
273 struct socket *so2;
274 u_long newhiwat;
275
276 if (unp == 0) {
277 error = EINVAL;
278 goto release;
279 }
280 if (flags & PRUS_OOB) {
281 error = EOPNOTSUPP;
282 goto release;
283 }
284
dadab5e9 285 if (control && (error = unp_internalize(control, td)))
984263bc
MD
286 goto release;
287
288 switch (so->so_type) {
289 case SOCK_DGRAM:
290 {
291 struct sockaddr *from;
292
293 if (nam) {
294 if (unp->unp_conn) {
295 error = EISCONN;
296 break;
297 }
dadab5e9 298 error = unp_connect(so, nam, td);
984263bc
MD
299 if (error)
300 break;
301 } else {
302 if (unp->unp_conn == 0) {
303 error = ENOTCONN;
304 break;
305 }
306 }
307 so2 = unp->unp_conn->unp_socket;
308 if (unp->unp_addr)
309 from = (struct sockaddr *)unp->unp_addr;
310 else
311 from = &sun_noname;
312 if (sbappendaddr(&so2->so_rcv, from, m, control)) {
313 sorwakeup(so2);
314 m = 0;
315 control = 0;
316 } else
317 error = ENOBUFS;
318 if (nam)
319 unp_disconnect(unp);
320 break;
321 }
322
323 case SOCK_STREAM:
324 /* Connect if not connected yet. */
325 /*
326 * Note: A better implementation would complain
327 * if not equal to the peer's address.
328 */
329 if ((so->so_state & SS_ISCONNECTED) == 0) {
330 if (nam) {
dadab5e9 331 error = unp_connect(so, nam, td);
984263bc
MD
332 if (error)
333 break; /* XXX */
334 } else {
335 error = ENOTCONN;
336 break;
337 }
338 }
339
340 if (so->so_state & SS_CANTSENDMORE) {
341 error = EPIPE;
342 break;
343 }
344 if (unp->unp_conn == 0)
345 panic("uipc_send connected but no connection?");
346 so2 = unp->unp_conn->unp_socket;
347 /*
348 * Send to paired receive port, and then reduce
349 * send buffer hiwater marks to maintain backpressure.
350 * Wake up readers.
351 */
352 if (control) {
353 if (sbappendcontrol(&so2->so_rcv, m, control))
354 control = 0;
355 } else
356 sbappend(&so2->so_rcv, m);
357 so->so_snd.sb_mbmax -=
358 so2->so_rcv.sb_mbcnt - unp->unp_conn->unp_mbcnt;
359 unp->unp_conn->unp_mbcnt = so2->so_rcv.sb_mbcnt;
360 newhiwat = so->so_snd.sb_hiwat -
361 (so2->so_rcv.sb_cc - unp->unp_conn->unp_cc);
362 (void)chgsbsize(so->so_cred->cr_uidinfo, &so->so_snd.sb_hiwat,
363 newhiwat, RLIM_INFINITY);
364 unp->unp_conn->unp_cc = so2->so_rcv.sb_cc;
365 sorwakeup(so2);
366 m = 0;
367 break;
368
369 default:
370 panic("uipc_send unknown socktype");
371 }
372
373 /*
374 * SEND_EOF is equivalent to a SEND followed by
375 * a SHUTDOWN.
376 */
377 if (flags & PRUS_EOF) {
378 socantsendmore(so);
379 unp_shutdown(unp);
380 }
381
382 if (control && error != 0)
383 unp_dispose(control);
384
385release:
386 if (control)
387 m_freem(control);
388 if (m)
389 m_freem(m);
390 return error;
391}
392
393static int
394uipc_sense(struct socket *so, struct stat *sb)
395{
396 struct unpcb *unp = sotounpcb(so);
397 struct socket *so2;
398
399 if (unp == 0)
400 return EINVAL;
401 sb->st_blksize = so->so_snd.sb_hiwat;
402 if (so->so_type == SOCK_STREAM && unp->unp_conn != 0) {
403 so2 = unp->unp_conn->unp_socket;
404 sb->st_blksize += so2->so_rcv.sb_cc;
405 }
406 sb->st_dev = NOUDEV;
407 if (unp->unp_ino == 0) /* make up a non-zero inode number */
408 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino;
409 sb->st_ino = unp->unp_ino;
410 return (0);
411}
412
413static int
414uipc_shutdown(struct socket *so)
415{
416 struct unpcb *unp = sotounpcb(so);
417
418 if (unp == 0)
419 return EINVAL;
420 socantsendmore(so);
421 unp_shutdown(unp);
422 return 0;
423}
424
425static int
426uipc_sockaddr(struct socket *so, struct sockaddr **nam)
427{
428 struct unpcb *unp = sotounpcb(so);
429
430 if (unp == 0)
431 return EINVAL;
432 if (unp->unp_addr)
433 *nam = dup_sockaddr((struct sockaddr *)unp->unp_addr, 1);
434 return 0;
435}
436
437struct pr_usrreqs uipc_usrreqs = {
438 uipc_abort, uipc_accept, uipc_attach, uipc_bind, uipc_connect,
439 uipc_connect2, pru_control_notsupp, uipc_detach, uipc_disconnect,
440 uipc_listen, uipc_peeraddr, uipc_rcvd, pru_rcvoob_notsupp,
441 uipc_send, uipc_sense, uipc_shutdown, uipc_sockaddr,
442 sosend, soreceive, sopoll
443};
444
445int
446uipc_ctloutput(so, sopt)
447 struct socket *so;
448 struct sockopt *sopt;
449{
450 struct unpcb *unp = sotounpcb(so);
451 int error;
452
453 switch (sopt->sopt_dir) {
454 case SOPT_GET:
455 switch (sopt->sopt_name) {
456 case LOCAL_PEERCRED:
457 if (unp->unp_flags & UNP_HAVEPC)
458 error = sooptcopyout(sopt, &unp->unp_peercred,
459 sizeof(unp->unp_peercred));
460 else {
461 if (so->so_type == SOCK_STREAM)
462 error = ENOTCONN;
463 else
464 error = EINVAL;
465 }
466 break;
467 default:
468 error = EOPNOTSUPP;
469 break;
470 }
471 break;
472 case SOPT_SET:
473 default:
474 error = EOPNOTSUPP;
475 break;
476 }
477 return (error);
478}
479
480/*
481 * Both send and receive buffers are allocated PIPSIZ bytes of buffering
482 * for stream sockets, although the total for sender and receiver is
483 * actually only PIPSIZ.
484 * Datagram sockets really use the sendspace as the maximum datagram size,
485 * and don't really want to reserve the sendspace. Their recvspace should
486 * be large enough for at least one max-size datagram plus address.
487 */
488#ifndef PIPSIZ
489#define PIPSIZ 8192
490#endif
491static u_long unpst_sendspace = PIPSIZ;
492static u_long unpst_recvspace = PIPSIZ;
493static u_long unpdg_sendspace = 2*1024; /* really max datagram size */
494static u_long unpdg_recvspace = 4*1024;
495
496static int unp_rights; /* file descriptors in flight */
497
498SYSCTL_DECL(_net_local_stream);
499SYSCTL_INT(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW,
500 &unpst_sendspace, 0, "");
501SYSCTL_INT(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW,
502 &unpst_recvspace, 0, "");
503SYSCTL_DECL(_net_local_dgram);
504SYSCTL_INT(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW,
505 &unpdg_sendspace, 0, "");
506SYSCTL_INT(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW,
507 &unpdg_recvspace, 0, "");
508SYSCTL_DECL(_net_local);
509SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, "");
510
511static int
e4700d00 512unp_attach(struct socket *so, struct pru_attach_info *ai)
984263bc 513{
1fd87d54 514 struct unpcb *unp;
984263bc
MD
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:
e4700d00
JH
521 error = soreserve(so, unpst_sendspace, unpst_recvspace,
522 ai->sb_rlimit);
984263bc
MD
523 break;
524
525 case SOCK_DGRAM:
e4700d00
JH
526 error = soreserve(so, unpdg_sendspace, unpdg_recvspace,
527 ai->sb_rlimit);
984263bc
MD
528 break;
529
530 default:
531 panic("unp_attach");
532 }
533 if (error)
534 return (error);
535 }
536 unp = zalloc(unp_zone);
537 if (unp == NULL)
538 return (ENOBUFS);
539 bzero(unp, sizeof *unp);
540 unp->unp_gencnt = ++unp_gencnt;
541 unp_count++;
542 LIST_INIT(&unp->unp_refs);
543 unp->unp_socket = so;
e4700d00 544 unp->unp_rvnode = ai->fd_rdir; /* jail cruft XXX JH */
984263bc
MD
545 LIST_INSERT_HEAD(so->so_type == SOCK_DGRAM ? &unp_dhead
546 : &unp_shead, unp, unp_link);
547 so->so_pcb = (caddr_t)unp;
548 return (0);
549}
550
551static void
552unp_detach(unp)
1fd87d54 553 struct unpcb *unp;
984263bc
MD
554{
555 LIST_REMOVE(unp, unp_link);
556 unp->unp_gencnt = ++unp_gencnt;
557 --unp_count;
558 if (unp->unp_vnode) {
559 unp->unp_vnode->v_socket = 0;
560 vrele(unp->unp_vnode);
561 unp->unp_vnode = 0;
562 }
563 if (unp->unp_conn)
564 unp_disconnect(unp);
565 while (!LIST_EMPTY(&unp->unp_refs))
566 unp_drop(LIST_FIRST(&unp->unp_refs), ECONNRESET);
567 soisdisconnected(unp->unp_socket);
568 unp->unp_socket->so_pcb = 0;
569 if (unp_rights) {
570 /*
571 * Normally the receive buffer is flushed later,
572 * in sofree, but if our receive buffer holds references
573 * to descriptors that are now garbage, we will dispose
574 * of those descriptor references after the garbage collector
575 * gets them (resulting in a "panic: closef: count < 0").
576 */
577 sorflush(unp->unp_socket);
578 unp_gc();
579 }
580 if (unp->unp_addr)
581 FREE(unp->unp_addr, M_SONAME);
582 zfree(unp_zone, unp);
583}
584
585static int
dadab5e9 586unp_bind(struct unpcb *unp, struct sockaddr *nam, struct thread *td)
984263bc 587{
dadab5e9 588 struct proc *p = td->td_proc;
984263bc 589 struct sockaddr_un *soun = (struct sockaddr_un *)nam;
dadab5e9 590 struct vnode *vp;
984263bc
MD
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 */
6fc0e645 603 NDINIT(&nd, NAMEI_CREATE, CNP_LOCKPARENT, UIO_SYSSPACE, buf, td);
984263bc
MD
604/* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
605 error = namei(&nd);
606 if (error)
607 return (error);
608 vp = nd.ni_vp;
609 if (vp != NULL) {
610 NDFREE(&nd, NDF_ONLY_PNBUF);
611 if (nd.ni_dvp == vp)
612 vrele(nd.ni_dvp);
613 else
614 vput(nd.ni_dvp);
615 vrele(vp);
616 return (EADDRINUSE);
617 }
618 VATTR_NULL(&vattr);
619 vattr.va_type = VSOCK;
620 vattr.va_mode = (ACCESSPERMS & ~p->p_fd->fd_cmask);
dadab5e9 621 VOP_LEASE(nd.ni_dvp, td, p->p_ucred, LEASE_WRITE);
bc0c094e 622 error = VOP_CREATE(nd.ni_dvp, NCPNULL, &nd.ni_vp, &nd.ni_cnd, &vattr);
984263bc
MD
623 NDFREE(&nd, NDF_ONLY_PNBUF);
624 vput(nd.ni_dvp);
625 if (error)
626 return (error);
627 vp = nd.ni_vp;
628 vp->v_socket = unp->unp_socket;
629 unp->unp_vnode = vp;
630 unp->unp_addr = (struct sockaddr_un *)dup_sockaddr(nam, 1);
41a01a4d 631 VOP_UNLOCK(vp, NULL, 0, td);
984263bc
MD
632 return (0);
633}
634
635static int
dadab5e9 636unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
984263bc 637{
dadab5e9
MD
638 struct proc *p = td->td_proc;
639 struct sockaddr_un *soun = (struct sockaddr_un *)nam;
640 struct vnode *vp;
641 struct socket *so2, *so3;
984263bc
MD
642 struct unpcb *unp, *unp2, *unp3;
643 int error, len;
644 struct nameidata nd;
645 char buf[SOCK_MAXADDRLEN];
646
dadab5e9
MD
647 KKASSERT(p);
648
984263bc
MD
649 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
650 if (len <= 0)
651 return EINVAL;
652 strncpy(buf, soun->sun_path, len);
653 buf[len] = 0;
654
2b69e610
MD
655 NDINIT(&nd, NAMEI_LOOKUP, CNP_FOLLOW | CNP_LOCKLEAF,
656 UIO_SYSSPACE, buf, td);
984263bc
MD
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 }
dadab5e9 666 error = VOP_ACCESS(vp, VWRITE, p->p_ucred, td);
984263bc
MD
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 ||
dadab5e9 680 (so3 = sonewconn(so2, 0)) == 0) {
984263bc
MD
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)
1fd87d54
RG
724 struct socket *so;
725 struct socket *so2;
984263bc 726{
1fd87d54
RG
727 struct unpcb *unp = sotounpcb(so);
728 struct unpcb *unp2;
984263bc
MD
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{
1fd87d54 757 struct unpcb *unp2 = unp->unp_conn;
984263bc
MD
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
dadab5e9 788prison_unpcb(struct thread *td, struct unpcb *unp)
984263bc 789{
dadab5e9
MD
790 struct proc *p;
791
792 if (td == NULL)
793 return (0);
794 if ((p = td->td_proc) == NULL)
795 return (0);
41c20dac 796 if (!p->p_ucred->cr_prison)
984263bc
MD
797 return (0);
798 if (p->p_fd->fd_rdir == unp->unp_rvnode)
799 return (0);
800 return (1);
801}
802
803static int
804unp_pcblist(SYSCTL_HANDLER_ARGS)
805{
806 int error, i, n;
807 struct unpcb *unp, **unp_list;
808 unp_gen_t gencnt;
809 struct xunpgen xug;
810 struct unp_head *head;
811
812 head = ((intptr_t)arg1 == SOCK_DGRAM ? &unp_dhead : &unp_shead);
813
41c20dac
MD
814 KKASSERT(curproc != NULL);
815
984263bc
MD
816 /*
817 * The process of preparing the PCB list is too time-consuming and
818 * resource-intensive to repeat twice on every request.
819 */
820 if (req->oldptr == 0) {
821 n = unp_count;
822 req->oldidx = 2 * (sizeof xug)
823 + (n + n/8) * sizeof(struct xunpcb);
824 return 0;
825 }
826
827 if (req->newptr != 0)
828 return EPERM;
829
830 /*
831 * OK, now we're committed to doing something.
832 */
833 gencnt = unp_gencnt;
834 n = unp_count;
835
836 xug.xug_len = sizeof xug;
837 xug.xug_count = n;
838 xug.xug_gen = gencnt;
839 xug.xug_sogen = so_gencnt;
840 error = SYSCTL_OUT(req, &xug, sizeof xug);
841 if (error)
842 return error;
843
844 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
845 if (unp_list == 0)
846 return ENOMEM;
847
848 for (unp = LIST_FIRST(head), i = 0; unp && i < n;
849 unp = LIST_NEXT(unp, unp_link)) {
dadab5e9 850 if (unp->unp_gencnt <= gencnt && !prison_unpcb(req->td, unp))
984263bc
MD
851 unp_list[i++] = unp;
852 }
853 n = i; /* in case we lost some during malloc */
854
855 error = 0;
856 for (i = 0; i < n; i++) {
857 unp = unp_list[i];
858 if (unp->unp_gencnt <= gencnt) {
859 struct xunpcb xu;
860 xu.xu_len = sizeof xu;
861 xu.xu_unpp = unp;
862 /*
863 * XXX - need more locking here to protect against
864 * connect/disconnect races for SMP.
865 */
866 if (unp->unp_addr)
867 bcopy(unp->unp_addr, &xu.xu_addr,
868 unp->unp_addr->sun_len);
869 if (unp->unp_conn && unp->unp_conn->unp_addr)
870 bcopy(unp->unp_conn->unp_addr,
871 &xu.xu_caddr,
872 unp->unp_conn->unp_addr->sun_len);
873 bcopy(unp, &xu.xu_unp, sizeof *unp);
874 sotoxsocket(unp->unp_socket, &xu.xu_socket);
875 error = SYSCTL_OUT(req, &xu, sizeof xu);
876 }
877 }
878 if (!error) {
879 /*
880 * Give the user an updated idea of our state.
881 * If the generation differs from what we told
882 * her before, she knows that something happened
883 * while we were processing this request, and it
884 * might be necessary to retry.
885 */
886 xug.xug_gen = unp_gencnt;
887 xug.xug_sogen = so_gencnt;
888 xug.xug_count = unp_count;
889 error = SYSCTL_OUT(req, &xug, sizeof xug);
890 }
891 free(unp_list, M_TEMP);
892 return error;
893}
894
895SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLFLAG_RD,
896 (caddr_t)(long)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
897 "List of active local datagram sockets");
898SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLFLAG_RD,
899 (caddr_t)(long)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
900 "List of active local stream sockets");
901
902static void
903unp_shutdown(unp)
904 struct unpcb *unp;
905{
906 struct socket *so;
907
908 if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn &&
909 (so = unp->unp_conn->unp_socket))
910 socantrcvmore(so);
911}
912
913static void
914unp_drop(unp, errno)
915 struct unpcb *unp;
916 int errno;
917{
918 struct socket *so = unp->unp_socket;
919
920 so->so_error = errno;
921 unp_disconnect(unp);
922}
923
924#ifdef notdef
925void
926unp_drain()
927{
928
929}
930#endif
931
932int
dadab5e9 933unp_externalize(struct mbuf *rights)
984263bc
MD
934{
935 struct proc *p = curproc; /* XXX */
dadab5e9
MD
936 int i;
937 struct cmsghdr *cm = mtod(rights, struct cmsghdr *);
938 int *fdp;
939 struct file **rp;
940 struct file *fp;
984263bc
MD
941 int newfds = (cm->cmsg_len - (CMSG_DATA(cm) - (u_char *)cm))
942 / sizeof (struct file *);
943 int f;
944
945 /*
946 * if the new FD's will not fit, then we free them all
947 */
948 if (!fdavail(p, newfds)) {
949 rp = (struct file **)CMSG_DATA(cm);
950 for (i = 0; i < newfds; i++) {
951 fp = *rp;
952 /*
953 * zero the pointer before calling unp_discard,
954 * since it may end up in unp_gc()..
955 */
956 *rp++ = 0;
957 unp_discard(fp);
958 }
959 return (EMSGSIZE);
960 }
961 /*
962 * now change each pointer to an fd in the global table to
963 * an integer that is the index to the local fd table entry
964 * that we set up to point to the global one we are transferring.
965 * If sizeof (struct file *) is bigger than or equal to sizeof int,
966 * then do it in forward order. In that case, an integer will
967 * always come in the same place or before its corresponding
968 * struct file pointer.
969 * If sizeof (struct file *) is smaller than sizeof int, then
970 * do it in reverse order.
971 */
972 if (sizeof (struct file *) >= sizeof (int)) {
973 fdp = (int *)(cm + 1);
974 rp = (struct file **)CMSG_DATA(cm);
975 for (i = 0; i < newfds; i++) {
976 if (fdalloc(p, 0, &f))
977 panic("unp_externalize");
978 fp = *rp++;
979 p->p_fd->fd_ofiles[f] = fp;
980 fp->f_msgcount--;
981 unp_rights--;
982 *fdp++ = f;
983 }
984 } else {
985 fdp = (int *)(cm + 1) + newfds - 1;
986 rp = (struct file **)CMSG_DATA(cm) + newfds - 1;
987 for (i = 0; i < newfds; i++) {
988 if (fdalloc(p, 0, &f))
989 panic("unp_externalize");
990 fp = *rp--;
991 p->p_fd->fd_ofiles[f] = fp;
992 fp->f_msgcount--;
993 unp_rights--;
994 *fdp-- = f;
995 }
996 }
997
998 /*
999 * Adjust length, in case sizeof(struct file *) and sizeof(int)
1000 * differs.
1001 */
1002 cm->cmsg_len = CMSG_LEN(newfds * sizeof(int));
1003 rights->m_len = cm->cmsg_len;
1004 return (0);
1005}
1006
1007void
1008unp_init(void)
1009{
1010 unp_zone = zinit("unpcb", sizeof(struct unpcb), nmbclusters, 0, 0);
1011 if (unp_zone == 0)
1012 panic("unp_init");
1013 LIST_INIT(&unp_dhead);
1014 LIST_INIT(&unp_shead);
1015}
1016
984263bc 1017static int
dadab5e9 1018unp_internalize(struct mbuf *control, struct thread *td)
984263bc 1019{
dadab5e9
MD
1020 struct proc *p = td->td_proc;
1021 struct filedesc *fdescp;
1022 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1023 struct file **rp;
1024 struct file *fp;
1025 int i, fd, *fdp;
1026 struct cmsgcred *cmcred;
984263bc
MD
1027 int oldfds;
1028 u_int newlen;
1029
dadab5e9
MD
1030 KKASSERT(p);
1031 fdescp = p->p_fd;
984263bc
MD
1032 if ((cm->cmsg_type != SCM_RIGHTS && cm->cmsg_type != SCM_CREDS) ||
1033 cm->cmsg_level != SOL_SOCKET || cm->cmsg_len != control->m_len)
1034 return (EINVAL);
1035
1036 /*
1037 * Fill in credential information.
1038 */
1039 if (cm->cmsg_type == SCM_CREDS) {
1040 cmcred = (struct cmsgcred *)(cm + 1);
1041 cmcred->cmcred_pid = p->p_pid;
41c20dac
MD
1042 cmcred->cmcred_uid = p->p_ucred->cr_ruid;
1043 cmcred->cmcred_gid = p->p_ucred->cr_rgid;
984263bc
MD
1044 cmcred->cmcred_euid = p->p_ucred->cr_uid;
1045 cmcred->cmcred_ngroups = MIN(p->p_ucred->cr_ngroups,
1046 CMGROUP_MAX);
1047 for (i = 0; i < cmcred->cmcred_ngroups; i++)
1048 cmcred->cmcred_groups[i] = p->p_ucred->cr_groups[i];
1049 return(0);
1050 }
1051
1052 oldfds = (cm->cmsg_len - sizeof (*cm)) / sizeof (int);
1053 /*
1054 * check that all the FDs passed in refer to legal OPEN files
1055 * If not, reject the entire operation.
1056 */
1057 fdp = (int *)(cm + 1);
1058 for (i = 0; i < oldfds; i++) {
1059 fd = *fdp++;
1060 if ((unsigned)fd >= fdescp->fd_nfiles ||
1061 fdescp->fd_ofiles[fd] == NULL)
1062 return (EBADF);
1063 if (fdescp->fd_ofiles[fd]->f_type == DTYPE_KQUEUE)
1064 return (EOPNOTSUPP);
1065 }
1066 /*
1067 * Now replace the integer FDs with pointers to
1068 * the associated global file table entry..
1069 * Allocate a bigger buffer as necessary. But if an cluster is not
1070 * enough, return E2BIG.
1071 */
1072 newlen = CMSG_LEN(oldfds * sizeof(struct file *));
1073 if (newlen > MCLBYTES)
1074 return (E2BIG);
1075 if (newlen - control->m_len > M_TRAILINGSPACE(control)) {
1076 if (control->m_flags & M_EXT)
1077 return (E2BIG);
1078 MCLGET(control, M_WAIT);
1079 if ((control->m_flags & M_EXT) == 0)
1080 return (ENOBUFS);
1081
1082 /* copy the data to the cluster */
1083 memcpy(mtod(control, char *), cm, cm->cmsg_len);
1084 cm = mtod(control, struct cmsghdr *);
1085 }
1086
1087 /*
1088 * Adjust length, in case sizeof(struct file *) and sizeof(int)
1089 * differs.
1090 */
1091 control->m_len = cm->cmsg_len = newlen;
1092
1093 /*
1094 * Transform the file descriptors into struct file pointers.
1095 * If sizeof (struct file *) is bigger than or equal to sizeof int,
1096 * then do it in reverse order so that the int won't get until
1097 * we're done.
1098 * If sizeof (struct file *) is smaller than sizeof int, then
1099 * do it in forward order.
1100 */
1101 if (sizeof (struct file *) >= sizeof (int)) {
1102 fdp = (int *)(cm + 1) + oldfds - 1;
1103 rp = (struct file **)CMSG_DATA(cm) + oldfds - 1;
1104 for (i = 0; i < oldfds; i++) {
1105 fp = fdescp->fd_ofiles[*fdp--];
1106 *rp-- = fp;
1107 fp->f_count++;
1108 fp->f_msgcount++;
1109 unp_rights++;
1110 }
1111 } else {
1112 fdp = (int *)(cm + 1);
1113 rp = (struct file **)CMSG_DATA(cm);
1114 for (i = 0; i < oldfds; i++) {
1115 fp = fdescp->fd_ofiles[*fdp++];
1116 *rp++ = fp;
1117 fp->f_count++;
1118 fp->f_msgcount++;
1119 unp_rights++;
1120 }
1121 }
1122 return (0);
1123}
1124
1125static int unp_defer, unp_gcing;
1126
1127static void
1128unp_gc()
1129{
1fd87d54
RG
1130 struct file *fp, *nextfp;
1131 struct socket *so;
984263bc
MD
1132 struct file **extra_ref, **fpp;
1133 int nunref, i;
1134
1135 if (unp_gcing)
1136 return;
1137 unp_gcing = 1;
1138 unp_defer = 0;
1139 /*
1140 * before going through all this, set all FDs to
1141 * be NOT defered and NOT externally accessible
1142 */
1143 LIST_FOREACH(fp, &filehead, f_list)
1144 fp->f_flag &= ~(FMARK|FDEFER);
1145 do {
1146 LIST_FOREACH(fp, &filehead, f_list) {
1147 /*
1148 * If the file is not open, skip it
1149 */
1150 if (fp->f_count == 0)
1151 continue;
1152 /*
1153 * If we already marked it as 'defer' in a
1154 * previous pass, then try process it this time
1155 * and un-mark it
1156 */
1157 if (fp->f_flag & FDEFER) {
1158 fp->f_flag &= ~FDEFER;
1159 unp_defer--;
1160 } else {
1161 /*
1162 * if it's not defered, then check if it's
1163 * already marked.. if so skip it
1164 */
1165 if (fp->f_flag & FMARK)
1166 continue;
1167 /*
1168 * If all references are from messages
1169 * in transit, then skip it. it's not
1170 * externally accessible.
1171 */
1172 if (fp->f_count == fp->f_msgcount)
1173 continue;
1174 /*
1175 * If it got this far then it must be
1176 * externally accessible.
1177 */
1178 fp->f_flag |= FMARK;
1179 }
1180 /*
1181 * either it was defered, or it is externally
1182 * accessible and not already marked so.
1183 * Now check if it is possibly one of OUR sockets.
1184 */
1185 if (fp->f_type != DTYPE_SOCKET ||
1186 (so = (struct socket *)fp->f_data) == 0)
1187 continue;
1188 if (so->so_proto->pr_domain != &localdomain ||
1189 (so->so_proto->pr_flags&PR_RIGHTS) == 0)
1190 continue;
1191#ifdef notdef
1192 if (so->so_rcv.sb_flags & SB_LOCK) {
1193 /*
1194 * This is problematical; it's not clear
1195 * we need to wait for the sockbuf to be
1196 * unlocked (on a uniprocessor, at least),
1197 * and it's also not clear what to do
1198 * if sbwait returns an error due to receipt
1199 * of a signal. If sbwait does return
1200 * an error, we'll go into an infinite
1201 * loop. Delete all of this for now.
1202 */
1203 (void) sbwait(&so->so_rcv);
1204 goto restart;
1205 }
1206#endif
1207 /*
1208 * So, Ok, it's one of our sockets and it IS externally
1209 * accessible (or was defered). Now we look
1210 * to see if we hold any file descriptors in its
1211 * message buffers. Follow those links and mark them
1212 * as accessible too.
1213 */
1214 unp_scan(so->so_rcv.sb_mb, unp_mark);
1215 }
1216 } while (unp_defer);
1217 /*
1218 * We grab an extra reference to each of the file table entries
1219 * that are not otherwise accessible and then free the rights
1220 * that are stored in messages on them.
1221 *
1222 * The bug in the orginal code is a little tricky, so I'll describe
1223 * what's wrong with it here.
1224 *
1225 * It is incorrect to simply unp_discard each entry for f_msgcount
1226 * times -- consider the case of sockets A and B that contain
1227 * references to each other. On a last close of some other socket,
1228 * we trigger a gc since the number of outstanding rights (unp_rights)
1229 * is non-zero. If during the sweep phase the gc code un_discards,
1230 * we end up doing a (full) closef on the descriptor. A closef on A
1231 * results in the following chain. Closef calls soo_close, which
1232 * calls soclose. Soclose calls first (through the switch
1233 * uipc_usrreq) unp_detach, which re-invokes unp_gc. Unp_gc simply
1234 * returns because the previous instance had set unp_gcing, and
1235 * we return all the way back to soclose, which marks the socket
1236 * with SS_NOFDREF, and then calls sofree. Sofree calls sorflush
1237 * to free up the rights that are queued in messages on the socket A,
1238 * i.e., the reference on B. The sorflush calls via the dom_dispose
1239 * switch unp_dispose, which unp_scans with unp_discard. This second
1240 * instance of unp_discard just calls closef on B.
1241 *
1242 * Well, a similar chain occurs on B, resulting in a sorflush on B,
1243 * which results in another closef on A. Unfortunately, A is already
1244 * being closed, and the descriptor has already been marked with
1245 * SS_NOFDREF, and soclose panics at this point.
1246 *
1247 * Here, we first take an extra reference to each inaccessible
1248 * descriptor. Then, we call sorflush ourself, since we know
1249 * it is a Unix domain socket anyhow. After we destroy all the
1250 * rights carried in messages, we do a last closef to get rid
1251 * of our extra reference. This is the last close, and the
1252 * unp_detach etc will shut down the socket.
1253 *
1254 * 91/09/19, bsy@cs.cmu.edu
1255 */
1256 extra_ref = malloc(nfiles * sizeof(struct file *), M_FILE, M_WAITOK);
1257 for (nunref = 0, fp = LIST_FIRST(&filehead), fpp = extra_ref; fp != 0;
1258 fp = nextfp) {
1259 nextfp = LIST_NEXT(fp, f_list);
1260 /*
1261 * If it's not open, skip it
1262 */
1263 if (fp->f_count == 0)
1264 continue;
1265 /*
1266 * If all refs are from msgs, and it's not marked accessible
1267 * then it must be referenced from some unreachable cycle
1268 * of (shut-down) FDs, so include it in our
1269 * list of FDs to remove
1270 */
1271 if (fp->f_count == fp->f_msgcount && !(fp->f_flag & FMARK)) {
1272 *fpp++ = fp;
1273 nunref++;
1274 fp->f_count++;
1275 }
1276 }
1277 /*
1278 * for each FD on our hit list, do the following two things
1279 */
1280 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) {
1281 struct file *tfp = *fpp;
1282 if (tfp->f_type == DTYPE_SOCKET && tfp->f_data != NULL)
1283 sorflush((struct socket *)(tfp->f_data));
1284 }
1285 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp)
dadab5e9 1286 closef(*fpp, NULL);
984263bc
MD
1287 free((caddr_t)extra_ref, M_FILE);
1288 unp_gcing = 0;
1289}
1290
1291void
dadab5e9 1292unp_dispose(struct mbuf *m)
984263bc 1293{
984263bc
MD
1294 if (m)
1295 unp_scan(m, unp_discard);
1296}
1297
1298static int
dadab5e9 1299unp_listen(struct unpcb *unp, struct thread *td)
984263bc 1300{
dadab5e9 1301 struct proc *p = td->td_proc;
984263bc 1302
dadab5e9 1303 KKASSERT(p);
984263bc
MD
1304 cru2x(p->p_ucred, &unp->unp_peercred);
1305 unp->unp_flags |= UNP_HAVEPCCACHED;
1306 return (0);
1307}
1308
1309static void
1310unp_scan(m0, op)
dadab5e9 1311 struct mbuf *m0;
402ed7e1 1312 void (*op) (struct file *);
984263bc 1313{
dadab5e9
MD
1314 struct mbuf *m;
1315 struct file **rp;
1316 struct cmsghdr *cm;
1317 int i;
984263bc
MD
1318 int qfds;
1319
1320 while (m0) {
1321 for (m = m0; m; m = m->m_next)
1322 if (m->m_type == MT_CONTROL &&
1323 m->m_len >= sizeof(*cm)) {
1324 cm = mtod(m, struct cmsghdr *);
1325 if (cm->cmsg_level != SOL_SOCKET ||
1326 cm->cmsg_type != SCM_RIGHTS)
1327 continue;
1328 qfds = (cm->cmsg_len -
1329 (CMSG_DATA(cm) - (u_char *)cm))
1330 / sizeof (struct file *);
1331 rp = (struct file **)CMSG_DATA(cm);
1332 for (i = 0; i < qfds; i++)
1333 (*op)(*rp++);
1334 break; /* XXX, but saves time */
1335 }
1336 m0 = m0->m_act;
1337 }
1338}
1339
1340static void
1341unp_mark(fp)
1342 struct file *fp;
1343{
1344
1345 if (fp->f_flag & FMARK)
1346 return;
1347 unp_defer++;
1348 fp->f_flag |= (FMARK|FDEFER);
1349}
1350
1351static void
1352unp_discard(fp)
1353 struct file *fp;
1354{
1355
1356 fp->f_msgcount--;
1357 unp_rights--;
dadab5e9 1358 (void) closef(fp, NULL);
984263bc 1359}