1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21
22 /*
23 * Copyright (c) 1999, 2010, Oracle and/or its affiliates. All rights reserved.
24 */
25
26 /*
27 * Copyright (c) 2014, Joyent, Inc. All rights reserved.
28 */
29
30 #include <sys/types.h>
31 #include <sys/param.h>
32 #include <sys/systm.h>
33 #include <sys/sysmacros.h>
34 #include <sys/debug.h>
35 #include <sys/cmn_err.h>
36
37 #include <sys/stropts.h>
38 #include <sys/socket.h>
39 #include <sys/socketvar.h>
40
41 #define _SUN_TPI_VERSION 2
42 #include <sys/tihdr.h>
43 #include <sys/sockio.h>
44 #include <sys/kmem_impl.h>
45
46 #include <sys/strsubr.h>
47 #include <sys/strsun.h>
48 #include <sys/ddi.h>
49 #include <netinet/in.h>
50 #include <inet/ip.h>
51
52 #include <fs/sockfs/sockcommon.h>
53 #include <fs/sockfs/sockfilter_impl.h>
54
55 #include <sys/socket_proto.h>
56
57 #include <fs/sockfs/socktpi_impl.h>
58 #include <fs/sockfs/sodirect.h>
59 #include <sys/tihdr.h>
60 #include <fs/sockfs/nl7c.h>
61
62 extern int xnet_skip_checks;
63 extern int xnet_check_print;
64
65 static void so_queue_oob(struct sonode *, mblk_t *, size_t);
66
67
68 /*ARGSUSED*/
69 int
70 so_accept_notsupp(struct sonode *lso, int fflag,
71 struct cred *cr, struct sonode **nsop)
72 {
73 return (EOPNOTSUPP);
74 }
75
76 /*ARGSUSED*/
77 int
78 so_listen_notsupp(struct sonode *so, int backlog, struct cred *cr)
79 {
80 return (EOPNOTSUPP);
81 }
82
83 /*ARGSUSED*/
84 int
85 so_getsockname_notsupp(struct sonode *so, struct sockaddr *sa,
86 socklen_t *len, struct cred *cr)
87 {
88 return (EOPNOTSUPP);
89 }
90
91 /*ARGSUSED*/
92 int
93 so_getpeername_notsupp(struct sonode *so, struct sockaddr *addr,
94 socklen_t *addrlen, boolean_t accept, struct cred *cr)
95 {
96 return (EOPNOTSUPP);
97 }
98
99 /*ARGSUSED*/
100 int
101 so_shutdown_notsupp(struct sonode *so, int how, struct cred *cr)
102 {
103 return (EOPNOTSUPP);
104 }
105
106 /*ARGSUSED*/
107 int
108 so_sendmblk_notsupp(struct sonode *so, struct msghdr *msg, int fflag,
109 struct cred *cr, mblk_t **mpp)
110 {
111 return (EOPNOTSUPP);
112 }
113
114 /*
115 * Generic Socket Ops
116 */
117
118 /* ARGSUSED */
119 int
120 so_init(struct sonode *so, struct sonode *pso, struct cred *cr, int flags)
121 {
122 return (socket_init_common(so, pso, flags, cr));
123 }
124
125 int
126 so_bind(struct sonode *so, struct sockaddr *name, socklen_t namelen,
127 int flags, struct cred *cr)
128 {
129 int error;
130
131 SO_BLOCK_FALLBACK(so, SOP_BIND(so, name, namelen, flags, cr));
132
133 ASSERT(flags == _SOBIND_XPG4_2 || flags == _SOBIND_SOCKBSD);
134
135 /* X/Open requires this check */
136 if ((so->so_state & SS_CANTSENDMORE) && !xnet_skip_checks) {
137 if (xnet_check_print) {
138 printf("sockfs: X/Open bind state check "
139 "caused EINVAL\n");
140 }
141 error = EINVAL;
142 goto done;
143 }
144
145 /*
146 * a bind to a NULL address is interpreted as unbind. So just
147 * do the downcall.
148 */
149 if (name == NULL)
150 goto dobind;
151
152 switch (so->so_family) {
153 case AF_INET:
154 if ((size_t)namelen != sizeof (sin_t)) {
155 error = name->sa_family != so->so_family ?
156 EAFNOSUPPORT : EINVAL;
157 eprintsoline(so, error);
158 goto done;
159 }
160
161 if ((flags & _SOBIND_XPG4_2) &&
162 (name->sa_family != so->so_family)) {
163 /*
164 * This check has to be made for X/Open
165 * sockets however application failures have
166 * been observed when it is applied to
167 * all sockets.
168 */
169 error = EAFNOSUPPORT;
170 eprintsoline(so, error);
171 goto done;
172 }
173 /*
174 * Force a zero sa_family to match so_family.
175 *
176 * Some programs like inetd(1M) don't set the
177 * family field. Other programs leave
178 * sin_family set to garbage - SunOS 4.X does
179 * not check the family field on a bind.
180 * We use the family field that
181 * was passed in to the socket() call.
182 */
183 name->sa_family = so->so_family;
184 break;
185
186 case AF_INET6: {
187 #ifdef DEBUG
188 sin6_t *sin6 = (sin6_t *)name;
189 #endif
190 if ((size_t)namelen != sizeof (sin6_t)) {
191 error = name->sa_family != so->so_family ?
192 EAFNOSUPPORT : EINVAL;
193 eprintsoline(so, error);
194 goto done;
195 }
196
197 if (name->sa_family != so->so_family) {
198 /*
199 * With IPv6 we require the family to match
200 * unlike in IPv4.
201 */
202 error = EAFNOSUPPORT;
203 eprintsoline(so, error);
204 goto done;
205 }
206 #ifdef DEBUG
207 /*
208 * Verify that apps don't forget to clear
209 * sin6_scope_id etc
210 */
211 if (sin6->sin6_scope_id != 0 &&
212 !IN6_IS_ADDR_LINKSCOPE(&sin6->sin6_addr)) {
213 zcmn_err(getzoneid(), CE_WARN,
214 "bind with uninitialized sin6_scope_id "
215 "(%d) on socket. Pid = %d\n",
216 (int)sin6->sin6_scope_id,
217 (int)curproc->p_pid);
218 }
219 if (sin6->__sin6_src_id != 0) {
220 zcmn_err(getzoneid(), CE_WARN,
221 "bind with uninitialized __sin6_src_id "
222 "(%d) on socket. Pid = %d\n",
223 (int)sin6->__sin6_src_id,
224 (int)curproc->p_pid);
225 }
226 #endif /* DEBUG */
227
228 break;
229 }
230 default:
231 /* Just pass the request to the protocol */
232 goto dobind;
233 }
234
235 /*
236 * First we check if either NCA or KSSL has been enabled for
237 * the requested address, and if so, we fall back to TPI.
238 * If neither of those two services are enabled, then we just
239 * pass the request to the protocol.
240 *
241 * Note that KSSL can only be enabled on a socket if NCA is NOT
242 * enabled for that socket, hence the else-statement below.
243 */
244 if (nl7c_enabled && ((so->so_family == AF_INET ||
245 so->so_family == AF_INET6) &&
246 nl7c_lookup_addr(name, namelen) != NULL)) {
247 /*
248 * NL7C is not supported in non-global zones,
249 * we enforce this restriction here.
250 */
251 if (so->so_zoneid == GLOBAL_ZONEID) {
252 /* NCA should be used, so fall back to TPI */
253 error = so_tpi_fallback(so, cr);
254 SO_UNBLOCK_FALLBACK(so);
255 if (error)
256 return (error);
257 else
258 return (SOP_BIND(so, name, namelen, flags, cr));
259 }
260 }
261
262 dobind:
263 if (so->so_filter_active == 0 ||
264 (error = sof_filter_bind(so, name, &namelen, cr)) < 0) {
265 error = (*so->so_downcalls->sd_bind)
266 (so->so_proto_handle, name, namelen, cr);
267 }
268 done:
269 SO_UNBLOCK_FALLBACK(so);
270
271 return (error);
272 }
273
274 int
275 so_listen(struct sonode *so, int backlog, struct cred *cr)
276 {
277 int error = 0;
278
279 ASSERT(MUTEX_NOT_HELD(&so->so_lock));
280 SO_BLOCK_FALLBACK(so, SOP_LISTEN(so, backlog, cr));
281
282 if ((so)->so_filter_active == 0 ||
283 (error = sof_filter_listen(so, &backlog, cr)) < 0)
284 error = (*so->so_downcalls->sd_listen)(so->so_proto_handle,
285 backlog, cr);
286
287 SO_UNBLOCK_FALLBACK(so);
288
289 return (error);
290 }
291
292
293 int
294 so_connect(struct sonode *so, struct sockaddr *name,
295 socklen_t namelen, int fflag, int flags, struct cred *cr)
296 {
297 int error = 0;
298 sock_connid_t id;
299
300 ASSERT(MUTEX_NOT_HELD(&so->so_lock));
301 SO_BLOCK_FALLBACK(so, SOP_CONNECT(so, name, namelen, fflag, flags, cr));
302
303 /*
304 * If there is a pending error, return error
305 * This can happen if a non blocking operation caused an error.
306 */
307
308 if (so->so_error != 0) {
309 mutex_enter(&so->so_lock);
310 error = sogeterr(so, B_TRUE);
311 mutex_exit(&so->so_lock);
312 if (error != 0)
313 goto done;
314 }
315
316 if (so->so_filter_active == 0 ||
317 (error = sof_filter_connect(so, (struct sockaddr *)name,
318 &namelen, cr)) < 0) {
319 error = (*so->so_downcalls->sd_connect)(so->so_proto_handle,
320 name, namelen, &id, cr);
321
322 if (error == EINPROGRESS)
323 error = so_wait_connected(so,
324 fflag & (FNONBLOCK|FNDELAY), id);
325 }
326 done:
327 SO_UNBLOCK_FALLBACK(so);
328 return (error);
329 }
330
331 /*ARGSUSED*/
332 int
333 so_accept(struct sonode *so, int fflag, struct cred *cr, struct sonode **nsop)
334 {
335 int error = 0;
336 struct sonode *nso;
337
338 *nsop = NULL;
339
340 SO_BLOCK_FALLBACK(so, SOP_ACCEPT(so, fflag, cr, nsop));
341 if ((so->so_state & SS_ACCEPTCONN) == 0) {
342 SO_UNBLOCK_FALLBACK(so);
343 return ((so->so_type == SOCK_DGRAM || so->so_type == SOCK_RAW) ?
344 EOPNOTSUPP : EINVAL);
345 }
346
347 if ((error = so_acceptq_dequeue(so, (fflag & (FNONBLOCK|FNDELAY)),
348 &nso)) == 0) {
349 ASSERT(nso != NULL);
350
351 /* finish the accept */
352 if ((so->so_filter_active > 0 &&
353 (error = sof_filter_accept(nso, cr)) > 0) ||
354 (error = (*so->so_downcalls->sd_accept)(so->so_proto_handle,
355 nso->so_proto_handle, (sock_upper_handle_t)nso, cr)) != 0) {
356 (void) socket_close(nso, 0, cr);
357 socket_destroy(nso);
358 } else {
359 *nsop = nso;
360 }
361 }
362
363 SO_UNBLOCK_FALLBACK(so);
364 return (error);
365 }
366
367 int
368 so_sendmsg(struct sonode *so, struct nmsghdr *msg, struct uio *uiop,
369 struct cred *cr)
370 {
371 int error, flags;
372 boolean_t dontblock;
373 ssize_t orig_resid;
374 mblk_t *mp;
375
376 SO_BLOCK_FALLBACK(so, SOP_SENDMSG(so, msg, uiop, cr));
377
378 flags = msg->msg_flags;
379 error = 0;
380 dontblock = (flags & MSG_DONTWAIT) ||
381 (uiop->uio_fmode & (FNONBLOCK|FNDELAY));
382
383 if (!(flags & MSG_XPG4_2) && msg->msg_controllen != 0) {
384 /*
385 * Old way of passing fd's is not supported
386 */
387 SO_UNBLOCK_FALLBACK(so);
388 return (EOPNOTSUPP);
389 }
390
391 if ((so->so_mode & SM_ATOMIC) &&
392 uiop->uio_resid > so->so_proto_props.sopp_maxpsz &&
393 so->so_proto_props.sopp_maxpsz != -1) {
394 SO_UNBLOCK_FALLBACK(so);
395 return (EMSGSIZE);
396 }
397
398 /*
399 * For atomic sends we will only do one iteration.
400 */
401 do {
402 if (so->so_state & SS_CANTSENDMORE) {
403 error = EPIPE;
404 break;
405 }
406
407 if (so->so_error != 0) {
408 mutex_enter(&so->so_lock);
409 error = sogeterr(so, B_TRUE);
410 mutex_exit(&so->so_lock);
411 if (error != 0)
412 break;
413 }
414
415 /*
416 * Send down OOB messages even if the send path is being
417 * flow controlled (assuming the protocol supports OOB data).
418 */
419 if (flags & MSG_OOB) {
420 if ((so->so_mode & SM_EXDATA) == 0) {
421 error = EOPNOTSUPP;
422 break;
423 }
424 } else if (SO_SND_FLOWCTRLD(so)) {
425 /*
426 * Need to wait until the protocol is ready to receive
427 * more data for transmission.
428 */
429 if ((error = so_snd_wait_qnotfull(so, dontblock)) != 0)
430 break;
431 }
432
433 /*
434 * Time to send data to the protocol. We either copy the
435 * data into mblks or pass the uio directly to the protocol.
436 * We decide what to do based on the available down calls.
437 */
438 if (so->so_downcalls->sd_send_uio != NULL) {
439 error = (*so->so_downcalls->sd_send_uio)
440 (so->so_proto_handle, uiop, msg, cr);
441 if (error != 0)
442 break;
443 } else {
444 /* save the resid in case of failure */
445 orig_resid = uiop->uio_resid;
446
447 if ((mp = socopyinuio(uiop,
448 so->so_proto_props.sopp_maxpsz,
449 so->so_proto_props.sopp_wroff,
450 so->so_proto_props.sopp_maxblk,
451 so->so_proto_props.sopp_tail, &error)) == NULL) {
452 break;
453 }
454 ASSERT(uiop->uio_resid >= 0);
455
456 if (so->so_filter_active > 0 &&
457 ((mp = SOF_FILTER_DATA_OUT(so, mp, msg, cr,
458 &error)) == NULL)) {
459 if (error != 0)
460 break;
461 continue;
462 }
463 error = (*so->so_downcalls->sd_send)
464 (so->so_proto_handle, mp, msg, cr);
465 if (error != 0) {
466 /*
467 * The send failed. We do not have to free the
468 * mblks, because that is the protocol's
469 * responsibility. However, uio_resid must
470 * remain accurate, so adjust that here.
471 */
472 uiop->uio_resid = orig_resid;
473 break;
474 }
475 }
476 } while (uiop->uio_resid > 0);
477
478 SO_UNBLOCK_FALLBACK(so);
479
480 return (error);
481 }
482
483 int
484 so_sendmblk_impl(struct sonode *so, struct nmsghdr *msg, int fflag,
485 struct cred *cr, mblk_t **mpp, sof_instance_t *fil,
486 boolean_t fil_inject)
487 {
488 int error;
489 boolean_t dontblock;
490 size_t size;
491 mblk_t *mp = *mpp;
492
493 if (so->so_downcalls->sd_send == NULL)
494 return (EOPNOTSUPP);
495
496 error = 0;
497 dontblock = (msg->msg_flags & MSG_DONTWAIT) ||
498 (fflag & (FNONBLOCK|FNDELAY));
499 size = msgdsize(mp);
500
501 if ((so->so_mode & SM_ATOMIC) &&
502 size > so->so_proto_props.sopp_maxpsz &&
503 so->so_proto_props.sopp_maxpsz != -1) {
504 SO_UNBLOCK_FALLBACK(so);
505 return (EMSGSIZE);
506 }
507
508 while (mp != NULL) {
509 mblk_t *nmp, *last_mblk;
510 size_t mlen;
511
512 if (so->so_state & SS_CANTSENDMORE) {
513 error = EPIPE;
514 break;
515 }
516 if (so->so_error != 0) {
517 mutex_enter(&so->so_lock);
518 error = sogeterr(so, B_TRUE);
519 mutex_exit(&so->so_lock);
520 if (error != 0)
521 break;
522 }
523 /* Socket filters are not flow controlled */
524 if (SO_SND_FLOWCTRLD(so) && !fil_inject) {
525 /*
526 * Need to wait until the protocol is ready to receive
527 * more data for transmission.
528 */
529 if ((error = so_snd_wait_qnotfull(so, dontblock)) != 0)
530 break;
531 }
532
533 /*
534 * We only allow so_maxpsz of data to be sent down to
535 * the protocol at time.
536 */
537 mlen = MBLKL(mp);
538 nmp = mp->b_cont;
539 last_mblk = mp;
540 while (nmp != NULL) {
541 mlen += MBLKL(nmp);
542 if (mlen > so->so_proto_props.sopp_maxpsz) {
543 last_mblk->b_cont = NULL;
544 break;
545 }
546 last_mblk = nmp;
547 nmp = nmp->b_cont;
548 }
549
550 if (so->so_filter_active > 0 &&
551 (mp = SOF_FILTER_DATA_OUT_FROM(so, fil, mp, msg,
552 cr, &error)) == NULL) {
553 *mpp = mp = nmp;
554 if (error != 0)
555 break;
556 continue;
557 }
558 error = (*so->so_downcalls->sd_send)
559 (so->so_proto_handle, mp, msg, cr);
560 if (error != 0) {
561 /*
562 * The send failed. The protocol will free the mblks
563 * that were sent down. Let the caller deal with the
564 * rest.
565 */
566 *mpp = nmp;
567 break;
568 }
569
570 *mpp = mp = nmp;
571 }
572 /* Let the filter know whether the protocol is flow controlled */
573 if (fil_inject && error == 0 && SO_SND_FLOWCTRLD(so))
574 error = ENOSPC;
575
576 return (error);
577 }
578
579 #pragma inline(so_sendmblk_impl)
580
581 int
582 so_sendmblk(struct sonode *so, struct nmsghdr *msg, int fflag,
583 struct cred *cr, mblk_t **mpp)
584 {
585 int error;
586
587 SO_BLOCK_FALLBACK(so, SOP_SENDMBLK(so, msg, fflag, cr, mpp));
588
589 if ((so->so_mode & SM_SENDFILESUPP) == 0) {
590 SO_UNBLOCK_FALLBACK(so);
591 return (EOPNOTSUPP);
592 }
593
594 error = so_sendmblk_impl(so, msg, fflag, cr, mpp, so->so_filter_top,
595 B_FALSE);
596
597 SO_UNBLOCK_FALLBACK(so);
598
599 return (error);
600 }
601
602 int
603 so_shutdown(struct sonode *so, int how, struct cred *cr)
604 {
605 int error;
606
607 SO_BLOCK_FALLBACK(so, SOP_SHUTDOWN(so, how, cr));
608
609 /*
610 * SunOS 4.X has no check for datagram sockets.
611 * 5.X checks that it is connected (ENOTCONN)
612 * X/Open requires that we check the connected state.
613 */
614 if (!(so->so_state & SS_ISCONNECTED)) {
615 if (!xnet_skip_checks) {
616 error = ENOTCONN;
617 if (xnet_check_print) {
618 printf("sockfs: X/Open shutdown check "
619 "caused ENOTCONN\n");
620 }
621 }
622 goto done;
623 }
624
625 if (so->so_filter_active == 0 ||
626 (error = sof_filter_shutdown(so, &how, cr)) < 0)
627 error = ((*so->so_downcalls->sd_shutdown)(so->so_proto_handle,
628 how, cr));
629
630 /*
631 * Protocol agreed to shutdown. We need to flush the
632 * receive buffer if the receive side is being shutdown.
633 */
634 if (error == 0 && how != SHUT_WR) {
635 mutex_enter(&so->so_lock);
636 /* wait for active reader to finish */
637 (void) so_lock_read(so, 0);
638
639 so_rcv_flush(so);
640
641 so_unlock_read(so);
642 mutex_exit(&so->so_lock);
643 }
644
645 done:
646 SO_UNBLOCK_FALLBACK(so);
647 return (error);
648 }
649
650 int
651 so_getsockname(struct sonode *so, struct sockaddr *addr,
652 socklen_t *addrlen, struct cred *cr)
653 {
654 int error;
655
656 SO_BLOCK_FALLBACK(so, SOP_GETSOCKNAME(so, addr, addrlen, cr));
657
658 if (so->so_filter_active == 0 ||
659 (error = sof_filter_getsockname(so, addr, addrlen, cr)) < 0)
660 error = (*so->so_downcalls->sd_getsockname)
661 (so->so_proto_handle, addr, addrlen, cr);
662
663 SO_UNBLOCK_FALLBACK(so);
664 return (error);
665 }
666
667 int
668 so_getpeername(struct sonode *so, struct sockaddr *addr,
669 socklen_t *addrlen, boolean_t accept, struct cred *cr)
670 {
671 int error;
672
673 SO_BLOCK_FALLBACK(so, SOP_GETPEERNAME(so, addr, addrlen, accept, cr));
674
675 if (accept) {
676 error = (*so->so_downcalls->sd_getpeername)
677 (so->so_proto_handle, addr, addrlen, cr);
678 } else if (!(so->so_state & SS_ISCONNECTED)) {
679 error = ENOTCONN;
680 } else if ((so->so_state & SS_CANTSENDMORE) && !xnet_skip_checks) {
681 /* Added this check for X/Open */
682 error = EINVAL;
683 if (xnet_check_print) {
684 printf("sockfs: X/Open getpeername check => EINVAL\n");
685 }
686 } else if (so->so_filter_active == 0 ||
687 (error = sof_filter_getpeername(so, addr, addrlen, cr)) < 0) {
688 error = (*so->so_downcalls->sd_getpeername)
689 (so->so_proto_handle, addr, addrlen, cr);
690 }
691
692 SO_UNBLOCK_FALLBACK(so);
693 return (error);
694 }
695
696 int
697 so_getsockopt(struct sonode *so, int level, int option_name,
698 void *optval, socklen_t *optlenp, int flags, struct cred *cr)
699 {
700 int error = 0;
701
702 if (level == SOL_FILTER)
703 return (sof_getsockopt(so, option_name, optval, optlenp, cr));
704
705 SO_BLOCK_FALLBACK(so,
706 SOP_GETSOCKOPT(so, level, option_name, optval, optlenp, flags, cr));
707
708 if ((so->so_filter_active == 0 ||
709 (error = sof_filter_getsockopt(so, level, option_name, optval,
710 optlenp, cr)) < 0) &&
711 (error = socket_getopt_common(so, level, option_name, optval,
712 optlenp, flags)) < 0) {
713 error = (*so->so_downcalls->sd_getsockopt)
714 (so->so_proto_handle, level, option_name, optval, optlenp,
715 cr);
716 if (error == ENOPROTOOPT) {
717 if (level == SOL_SOCKET) {
718 /*
719 * If a protocol does not support a particular
720 * socket option, set can fail (not allowed)
721 * but get can not fail. This is the previous
722 * sockfs bahvior.
723 */
724 switch (option_name) {
725 case SO_LINGER:
726 if (*optlenp < (t_uscalar_t)
727 sizeof (struct linger)) {
728 error = EINVAL;
729 break;
730 }
731 error = 0;
732 bzero(optval, sizeof (struct linger));
733 *optlenp = sizeof (struct linger);
734 break;
735 case SO_RCVTIMEO:
736 case SO_SNDTIMEO:
737 if (*optlenp < (t_uscalar_t)
738 sizeof (struct timeval)) {
739 error = EINVAL;
740 break;
741 }
742 error = 0;
743 bzero(optval, sizeof (struct timeval));
744 *optlenp = sizeof (struct timeval);
745 break;
746 case SO_SND_BUFINFO:
747 if (*optlenp < (t_uscalar_t)
748 sizeof (struct so_snd_bufinfo)) {
749 error = EINVAL;
750 break;
751 }
752 error = 0;
753 bzero(optval,
754 sizeof (struct so_snd_bufinfo));
755 *optlenp =
756 sizeof (struct so_snd_bufinfo);
757 break;
758 case SO_DEBUG:
759 case SO_REUSEADDR:
760 case SO_KEEPALIVE:
761 case SO_DONTROUTE:
762 case SO_BROADCAST:
763 case SO_USELOOPBACK:
764 case SO_OOBINLINE:
765 case SO_DGRAM_ERRIND:
766 case SO_SNDBUF:
767 case SO_RCVBUF:
768 error = 0;
769 *((int32_t *)optval) = 0;
770 *optlenp = sizeof (int32_t);
771 break;
772 default:
773 break;
774 }
775 }
776 }
777 }
778
779 SO_UNBLOCK_FALLBACK(so);
780 return (error);
781 }
782
783 int
784 so_setsockopt(struct sonode *so, int level, int option_name,
785 const void *optval, socklen_t optlen, struct cred *cr)
786 {
787 int error = 0;
788 struct timeval tl;
789 const void *opt = optval;
790
791 if (level == SOL_FILTER)
792 return (sof_setsockopt(so, option_name, optval, optlen, cr));
793
794 SO_BLOCK_FALLBACK(so,
795 SOP_SETSOCKOPT(so, level, option_name, optval, optlen, cr));
796
797 /* X/Open requires this check */
798 if (so->so_state & SS_CANTSENDMORE && !xnet_skip_checks) {
799 SO_UNBLOCK_FALLBACK(so);
800 if (xnet_check_print)
801 printf("sockfs: X/Open setsockopt check => EINVAL\n");
802 return (EINVAL);
803 }
804
805 if (so->so_filter_active > 0 &&
806 (error = sof_filter_setsockopt(so, level, option_name,
807 (void *)optval, &optlen, cr)) >= 0)
808 goto done;
809
810 if (level == SOL_SOCKET) {
811 switch (option_name) {
812 case SO_RCVTIMEO:
813 case SO_SNDTIMEO: {
814 /*
815 * We pass down these two options to protocol in order
816 * to support some third part protocols which need to
817 * know them. For those protocols which don't care
818 * these two options, simply return 0.
819 */
820 clock_t t_usec;
821
822 if (get_udatamodel() == DATAMODEL_NONE ||
823 get_udatamodel() == DATAMODEL_NATIVE) {
824 if (optlen != sizeof (struct timeval)) {
825 error = EINVAL;
826 goto done;
827 }
828 bcopy((struct timeval *)optval, &tl,
829 sizeof (struct timeval));
830 } else {
831 if (optlen != sizeof (struct timeval32)) {
832 error = EINVAL;
833 goto done;
834 }
835 TIMEVAL32_TO_TIMEVAL(&tl,
836 (struct timeval32 *)optval);
837 }
838 opt = &tl;
839 optlen = sizeof (tl);
840 t_usec = tl.tv_sec * 1000 * 1000 + tl.tv_usec;
841 mutex_enter(&so->so_lock);
842 if (option_name == SO_RCVTIMEO)
843 so->so_rcvtimeo = drv_usectohz(t_usec);
844 else
845 so->so_sndtimeo = drv_usectohz(t_usec);
846 mutex_exit(&so->so_lock);
847 break;
848 }
849 case SO_RCVBUF:
850 /*
851 * XXX XPG 4.2 applications retrieve SO_RCVBUF from
852 * sockfs since the transport might adjust the value
853 * and not return exactly what was set by the
854 * application.
855 */
856 so->so_xpg_rcvbuf = *(int32_t *)optval;
857 break;
858 }
859 }
860 error = (*so->so_downcalls->sd_setsockopt)
861 (so->so_proto_handle, level, option_name, opt, optlen, cr);
862 done:
863 SO_UNBLOCK_FALLBACK(so);
864 return (error);
865 }
866
867 int
868 so_ioctl(struct sonode *so, int cmd, intptr_t arg, int mode,
869 struct cred *cr, int32_t *rvalp)
870 {
871 int error = 0;
872
873 SO_BLOCK_FALLBACK(so, SOP_IOCTL(so, cmd, arg, mode, cr, rvalp));
874
875 /*
876 * If there is a pending error, return error
877 * This can happen if a non blocking operation caused an error.
878 */
879 if (so->so_error != 0) {
880 mutex_enter(&so->so_lock);
881 error = sogeterr(so, B_TRUE);
882 mutex_exit(&so->so_lock);
883 if (error != 0)
884 goto done;
885 }
886
887 /*
888 * calling strioc can result in the socket falling back to TPI,
889 * if that is supported.
890 */
891 if ((so->so_filter_active == 0 ||
892 (error = sof_filter_ioctl(so, cmd, arg, mode,
893 rvalp, cr)) < 0) &&
894 (error = socket_ioctl_common(so, cmd, arg, mode, cr, rvalp)) < 0 &&
895 (error = socket_strioc_common(so, cmd, arg, mode, cr, rvalp)) < 0) {
896 error = (*so->so_downcalls->sd_ioctl)(so->so_proto_handle,
897 cmd, arg, mode, rvalp, cr);
898 }
899
900 done:
901 SO_UNBLOCK_FALLBACK(so);
902
903 return (error);
904 }
905
906 int
907 so_poll(struct sonode *so, short events, int anyyet, short *reventsp,
908 struct pollhead **phpp)
909 {
910 int state = so->so_state, mask;
911 *reventsp = 0;
912
913 /*
914 * In sockets the errors are represented as input/output events
915 */
916 if (so->so_error != 0 &&
917 ((POLLIN|POLLRDNORM|POLLOUT) & events) != 0) {
918 *reventsp = (POLLIN|POLLRDNORM|POLLOUT) & events;
919 return (0);
920 }
921
922 /*
923 * If the socket is in a state where it can send data
924 * turn on POLLWRBAND and POLLOUT events.
925 */
926 if ((so->so_mode & SM_CONNREQUIRED) == 0 || (state & SS_ISCONNECTED)) {
927 /*
928 * out of band data is allowed even if the connection
929 * is flow controlled
930 */
931 *reventsp |= POLLWRBAND & events;
932 if (!SO_SND_FLOWCTRLD(so)) {
933 /*
934 * As long as there is buffer to send data
935 * turn on POLLOUT events
936 */
937 *reventsp |= POLLOUT & events;
938 }
939 }
940
941 /*
942 * Turn on POLLIN whenever there is data on the receive queue,
943 * or the socket is in a state where no more data will be received.
944 * Also, if the socket is accepting connections, flip the bit if
945 * there is something on the queue.
946 *
947 * We do an initial check for events without holding locks. However,
948 * if there are no event available, then we redo the check for POLLIN
949 * events under the lock.
950 */
951
952 /* Pending connections */
953 if (!list_is_empty(&so->so_acceptq_list))
954 *reventsp |= (POLLIN|POLLRDNORM) & events;
955
956 /*
957 * If we're looking for POLLRDHUP, indicate it if we have sent the
958 * last rx signal for the socket.
959 */
960 if ((events & POLLRDHUP) && (state & SS_SENTLASTREADSIG))
961 *reventsp |= POLLRDHUP;
962
963 /* Data */
964 /* so_downcalls is null for sctp */
965 if (so->so_downcalls != NULL && so->so_downcalls->sd_poll != NULL) {
966 *reventsp |= (*so->so_downcalls->sd_poll)
967 (so->so_proto_handle, events & SO_PROTO_POLLEV, anyyet,
968 CRED()) & events;
969 ASSERT((*reventsp & ~events) == 0);
970 /* do not recheck events */
971 events &= ~SO_PROTO_POLLEV;
972 } else {
973 if (SO_HAVE_DATA(so))
974 *reventsp |= (POLLIN|POLLRDNORM) & events;
975
976 /* Urgent data */
977 if ((state & SS_OOBPEND) != 0) {
978 *reventsp |= (POLLRDBAND | POLLPRI) & events;
979 }
980
981 /*
982 * If the socket has become disconnected, we set POLLHUP.
983 * Note that if we are in this state, we will have set POLLIN
984 * (SO_HAVE_DATA() is true on a disconnected socket), but not
985 * POLLOUT (SS_ISCONNECTED is false). This is in keeping with
986 * the semantics of POLLHUP, which is defined to be mutually
987 * exclusive with respect to POLLOUT but not POLLIN. We are
988 * therefore setting POLLHUP primarily for the benefit of
989 * those not polling on POLLIN, as they have no other way of
990 * knowing that the socket has been disconnected.
991 */
992 mask = SS_SENTLASTREADSIG | SS_SENTLASTWRITESIG;
993
994 if ((state & (mask | SS_ISCONNECTED)) == mask)
995 *reventsp |= POLLHUP;
996 }
997
998 if ((!*reventsp && !anyyet) || (events & POLLET)) {
999 /* Check for read events again, but this time under lock */
1000 if (events & (POLLIN|POLLRDNORM)) {
1001 mutex_enter(&so->so_lock);
1002 if (SO_HAVE_DATA(so) ||
1003 !list_is_empty(&so->so_acceptq_list)) {
1004 if (events & POLLET) {
1005 so->so_pollev |= SO_POLLEV_IN;
1006 *phpp = &so->so_poll_list;
1007 }
1008
1009 mutex_exit(&so->so_lock);
1010 *reventsp |= (POLLIN|POLLRDNORM) & events;
1011
1012 return (0);
1013 } else {
1014 so->so_pollev |= SO_POLLEV_IN;
1015 mutex_exit(&so->so_lock);
1016 }
1017 }
1018 *phpp = &so->so_poll_list;
1019 }
1020 return (0);
1021 }
1022
1023 /*
1024 * Generic Upcalls
1025 */
1026 void
1027 so_connected(sock_upper_handle_t sock_handle, sock_connid_t id,
1028 cred_t *peer_cred, pid_t peer_cpid)
1029 {
1030 struct sonode *so = (struct sonode *)sock_handle;
1031
1032 mutex_enter(&so->so_lock);
1033 ASSERT(so->so_proto_handle != NULL);
1034
1035 if (peer_cred != NULL) {
1036 if (so->so_peercred != NULL)
1037 crfree(so->so_peercred);
1038 crhold(peer_cred);
1039 so->so_peercred = peer_cred;
1040 so->so_cpid = peer_cpid;
1041 }
1042
1043 so->so_proto_connid = id;
1044 soisconnected(so);
1045 /*
1046 * Wake ones who're waiting for conn to become established.
1047 */
1048 so_notify_connected(so);
1049 }
1050
1051 int
1052 so_disconnected(sock_upper_handle_t sock_handle, sock_connid_t id, int error)
1053 {
1054 struct sonode *so = (struct sonode *)sock_handle;
1055 boolean_t connect_failed;
1056
1057 mutex_enter(&so->so_lock);
1058
1059 /*
1060 * If we aren't currently connected, then this isn't a disconnect but
1061 * rather a failure to connect.
1062 */
1063 connect_failed = !(so->so_state & SS_ISCONNECTED);
1064
1065 so->so_proto_connid = id;
1066 soisdisconnected(so, error);
1067 so_notify_disconnected(so, connect_failed, error);
1068
1069 return (0);
1070 }
1071
1072 void
1073 so_opctl(sock_upper_handle_t sock_handle, sock_opctl_action_t action,
1074 uintptr_t arg)
1075 {
1076 struct sonode *so = (struct sonode *)sock_handle;
1077
1078 switch (action) {
1079 case SOCK_OPCTL_SHUT_SEND:
1080 mutex_enter(&so->so_lock);
1081 socantsendmore(so);
1082 so_notify_disconnecting(so);
1083 break;
1084 case SOCK_OPCTL_SHUT_RECV: {
1085 mutex_enter(&so->so_lock);
1086 socantrcvmore(so);
1087 so_notify_eof(so);
1088 break;
1089 }
1090 case SOCK_OPCTL_ENAB_ACCEPT:
1091 mutex_enter(&so->so_lock);
1092 so->so_state |= SS_ACCEPTCONN;
1093 so->so_backlog = (unsigned int)arg;
1094 /*
1095 * The protocol can stop generating newconn upcalls when
1096 * the backlog is full, so to make sure the listener does
1097 * not end up with a queue full of deferred connections
1098 * we reduce the backlog by one. Thus the listener will
1099 * start closing deferred connections before the backlog
1100 * is full.
1101 */
1102 if (so->so_filter_active > 0)
1103 so->so_backlog = MAX(1, so->so_backlog - 1);
1104 mutex_exit(&so->so_lock);
1105 break;
1106 default:
1107 ASSERT(0);
1108 break;
1109 }
1110 }
1111
1112 void
1113 so_txq_full(sock_upper_handle_t sock_handle, boolean_t qfull)
1114 {
1115 struct sonode *so = (struct sonode *)sock_handle;
1116
1117 if (qfull) {
1118 so_snd_qfull(so);
1119 } else {
1120 so_snd_qnotfull(so);
1121 mutex_enter(&so->so_lock);
1122 /* so_notify_writable drops so_lock */
1123 so_notify_writable(so);
1124 }
1125 }
1126
1127 sock_upper_handle_t
1128 so_newconn(sock_upper_handle_t parenthandle,
1129 sock_lower_handle_t proto_handle, sock_downcalls_t *sock_downcalls,
1130 struct cred *peer_cred, pid_t peer_cpid, sock_upcalls_t **sock_upcallsp)
1131 {
1132 struct sonode *so = (struct sonode *)parenthandle;
1133 struct sonode *nso;
1134 int error;
1135
1136 ASSERT(proto_handle != NULL);
1137
1138 if ((so->so_state & SS_ACCEPTCONN) == 0 ||
1139 (so->so_acceptq_len >= so->so_backlog &&
1140 (so->so_filter_active == 0 || !sof_sonode_drop_deferred(so)))) {
1141 return (NULL);
1142 }
1143
1144 nso = socket_newconn(so, proto_handle, sock_downcalls, SOCKET_NOSLEEP,
1145 &error);
1146 if (nso == NULL)
1147 return (NULL);
1148
1149 if (peer_cred != NULL) {
1150 crhold(peer_cred);
1151 nso->so_peercred = peer_cred;
1152 nso->so_cpid = peer_cpid;
1153 }
1154 nso->so_listener = so;
1155
1156 /*
1157 * The new socket (nso), proto_handle and sock_upcallsp are all
1158 * valid at this point. But as soon as nso is placed in the accept
1159 * queue that can no longer be assumed (since an accept() thread may
1160 * pull it off the queue and close the socket).
1161 */
1162 *sock_upcallsp = &so_upcalls;
1163
1164 mutex_enter(&so->so_acceptq_lock);
1165 if (so->so_state & (SS_CLOSING|SS_FALLBACK_PENDING|SS_FALLBACK_COMP)) {
1166 mutex_exit(&so->so_acceptq_lock);
1167 ASSERT(nso->so_count == 1);
1168 nso->so_count--;
1169 nso->so_listener = NULL;
1170 /* drop proto ref */
1171 VN_RELE(SOTOV(nso));
1172 socket_destroy(nso);
1173 return (NULL);
1174 } else {
1175 so->so_acceptq_len++;
1176 if (nso->so_state & SS_FIL_DEFER) {
1177 list_insert_tail(&so->so_acceptq_defer, nso);
1178 mutex_exit(&so->so_acceptq_lock);
1179 } else {
1180 list_insert_tail(&so->so_acceptq_list, nso);
1181 cv_signal(&so->so_acceptq_cv);
1182 mutex_exit(&so->so_acceptq_lock);
1183 mutex_enter(&so->so_lock);
1184 so_notify_newconn(so);
1185 }
1186
1187 return ((sock_upper_handle_t)nso);
1188 }
1189 }
1190
1191 void
1192 so_set_prop(sock_upper_handle_t sock_handle, struct sock_proto_props *soppp)
1193 {
1194 struct sonode *so;
1195
1196 so = (struct sonode *)sock_handle;
1197
1198 mutex_enter(&so->so_lock);
1199
1200 if (soppp->sopp_flags & SOCKOPT_MAXBLK)
1201 so->so_proto_props.sopp_maxblk = soppp->sopp_maxblk;
1202 if (soppp->sopp_flags & SOCKOPT_WROFF)
1203 so->so_proto_props.sopp_wroff = soppp->sopp_wroff;
1204 if (soppp->sopp_flags & SOCKOPT_TAIL)
1205 so->so_proto_props.sopp_tail = soppp->sopp_tail;
1206 if (soppp->sopp_flags & SOCKOPT_RCVHIWAT)
1207 so->so_proto_props.sopp_rxhiwat = soppp->sopp_rxhiwat;
1208 if (soppp->sopp_flags & SOCKOPT_RCVLOWAT)
1209 so->so_proto_props.sopp_rxlowat = soppp->sopp_rxlowat;
1210 if (soppp->sopp_flags & SOCKOPT_MAXPSZ)
1211 so->so_proto_props.sopp_maxpsz = soppp->sopp_maxpsz;
1212 if (soppp->sopp_flags & SOCKOPT_MINPSZ)
1213 so->so_proto_props.sopp_minpsz = soppp->sopp_minpsz;
1214 if (soppp->sopp_flags & SOCKOPT_ZCOPY) {
1215 if (soppp->sopp_zcopyflag & ZCVMSAFE) {
1216 so->so_proto_props.sopp_zcopyflag |= STZCVMSAFE;
1217 so->so_proto_props.sopp_zcopyflag &= ~STZCVMUNSAFE;
1218 } else if (soppp->sopp_zcopyflag & ZCVMUNSAFE) {
1219 so->so_proto_props.sopp_zcopyflag |= STZCVMUNSAFE;
1220 so->so_proto_props.sopp_zcopyflag &= ~STZCVMSAFE;
1221 }
1222
1223 if (soppp->sopp_zcopyflag & COPYCACHED) {
1224 so->so_proto_props.sopp_zcopyflag |= STRCOPYCACHED;
1225 }
1226 }
1227 if (soppp->sopp_flags & SOCKOPT_OOBINLINE)
1228 so->so_proto_props.sopp_oobinline = soppp->sopp_oobinline;
1229 if (soppp->sopp_flags & SOCKOPT_RCVTIMER)
1230 so->so_proto_props.sopp_rcvtimer = soppp->sopp_rcvtimer;
1231 if (soppp->sopp_flags & SOCKOPT_RCVTHRESH)
1232 so->so_proto_props.sopp_rcvthresh = soppp->sopp_rcvthresh;
1233 if (soppp->sopp_flags & SOCKOPT_MAXADDRLEN)
1234 so->so_proto_props.sopp_maxaddrlen = soppp->sopp_maxaddrlen;
1235 if (soppp->sopp_flags & SOCKOPT_LOOPBACK)
1236 so->so_proto_props.sopp_loopback = soppp->sopp_loopback;
1237
1238 mutex_exit(&so->so_lock);
1239
1240 if (so->so_filter_active > 0) {
1241 sof_instance_t *inst;
1242 ssize_t maxblk;
1243 ushort_t wroff, tail;
1244 maxblk = so->so_proto_props.sopp_maxblk;
1245 wroff = so->so_proto_props.sopp_wroff;
1246 tail = so->so_proto_props.sopp_tail;
1247 for (inst = so->so_filter_bottom; inst != NULL;
1248 inst = inst->sofi_prev) {
1249 if (SOF_INTERESTED(inst, mblk_prop)) {
1250 (*inst->sofi_ops->sofop_mblk_prop)(
1251 (sof_handle_t)inst, inst->sofi_cookie,
1252 &maxblk, &wroff, &tail);
1253 }
1254 }
1255 mutex_enter(&so->so_lock);
1256 so->so_proto_props.sopp_maxblk = maxblk;
1257 so->so_proto_props.sopp_wroff = wroff;
1258 so->so_proto_props.sopp_tail = tail;
1259 mutex_exit(&so->so_lock);
1260 }
1261 #ifdef DEBUG
1262 soppp->sopp_flags &= ~(SOCKOPT_MAXBLK | SOCKOPT_WROFF | SOCKOPT_TAIL |
1263 SOCKOPT_RCVHIWAT | SOCKOPT_RCVLOWAT | SOCKOPT_MAXPSZ |
1264 SOCKOPT_ZCOPY | SOCKOPT_OOBINLINE | SOCKOPT_RCVTIMER |
1265 SOCKOPT_RCVTHRESH | SOCKOPT_MAXADDRLEN | SOCKOPT_MINPSZ |
1266 SOCKOPT_LOOPBACK);
1267 ASSERT(soppp->sopp_flags == 0);
1268 #endif
1269 }
1270
1271 /* ARGSUSED */
1272 ssize_t
1273 so_queue_msg_impl(struct sonode *so, mblk_t *mp,
1274 size_t msg_size, int flags, int *errorp, boolean_t *force_pushp,
1275 sof_instance_t *filter)
1276 {
1277 boolean_t force_push = B_TRUE;
1278 int space_left;
1279 sodirect_t *sodp = so->so_direct;
1280
1281 ASSERT(errorp != NULL);
1282 *errorp = 0;
1283 if (mp == NULL) {
1284 if (so->so_downcalls->sd_recv_uio != NULL) {
1285 mutex_enter(&so->so_lock);
1286 /* the notify functions will drop the lock */
1287 if (flags & MSG_OOB)
1288 so_notify_oobdata(so, IS_SO_OOB_INLINE(so));
1289 else
1290 so_notify_data(so, msg_size);
1291 return (0);
1292 }
1293 ASSERT(msg_size == 0);
1294 mutex_enter(&so->so_lock);
1295 goto space_check;
1296 }
1297
1298 ASSERT(mp->b_next == NULL);
1299 ASSERT(DB_TYPE(mp) == M_DATA || DB_TYPE(mp) == M_PROTO);
1300 ASSERT(msg_size == msgdsize(mp));
1301
1302 if (DB_TYPE(mp) == M_PROTO && !__TPI_PRIM_ISALIGNED(mp->b_rptr)) {
1303 /* The read pointer is not aligned correctly for TPI */
1304 zcmn_err(getzoneid(), CE_WARN,
1305 "sockfs: Unaligned TPI message received. rptr = %p\n",
1306 (void *)mp->b_rptr);
1307 freemsg(mp);
1308 mutex_enter(&so->so_lock);
1309 if (sodp != NULL)
1310 SOD_UIOAFINI(sodp);
1311 goto space_check;
1312 }
1313
1314 if (so->so_filter_active > 0) {
1315 for (; filter != NULL; filter = filter->sofi_prev) {
1316 if (!SOF_INTERESTED(filter, data_in))
1317 continue;
1318 mp = (*filter->sofi_ops->sofop_data_in)(
1319 (sof_handle_t)filter, filter->sofi_cookie, mp,
1320 flags, &msg_size);
1321 ASSERT(msgdsize(mp) == msg_size);
1322 DTRACE_PROBE2(filter__data, (sof_instance_t), filter,
1323 (mblk_t *), mp);
1324 /* Data was consumed/dropped, just do space check */
1325 if (msg_size == 0) {
1326 mutex_enter(&so->so_lock);
1327 goto space_check;
1328 }
1329 }
1330 }
1331
1332 if (flags & MSG_OOB) {
1333 so_queue_oob(so, mp, msg_size);
1334 mutex_enter(&so->so_lock);
1335 goto space_check;
1336 }
1337
1338 if (force_pushp != NULL)
1339 force_push = *force_pushp;
1340
1341 mutex_enter(&so->so_lock);
1342 if (so->so_state & (SS_FALLBACK_DRAIN | SS_FALLBACK_COMP)) {
1343 if (sodp != NULL)
1344 SOD_DISABLE(sodp);
1345 mutex_exit(&so->so_lock);
1346 *errorp = EOPNOTSUPP;
1347 return (-1);
1348 }
1349 if (so->so_state & (SS_CANTRCVMORE | SS_CLOSING)) {
1350 freemsg(mp);
1351 if (sodp != NULL)
1352 SOD_DISABLE(sodp);
1353 mutex_exit(&so->so_lock);
1354 return (0);
1355 }
1356
1357 /* process the mblk via I/OAT if capable */
1358 if (sodp != NULL && sodp->sod_enabled) {
1359 if (DB_TYPE(mp) == M_DATA) {
1360 sod_uioa_mblk_init(sodp, mp, msg_size);
1361 } else {
1362 SOD_UIOAFINI(sodp);
1363 }
1364 }
1365
1366 if (mp->b_next == NULL) {
1367 so_enqueue_msg(so, mp, msg_size);
1368 } else {
1369 do {
1370 mblk_t *nmp;
1371
1372 if ((nmp = mp->b_next) != NULL) {
1373 mp->b_next = NULL;
1374 }
1375 so_enqueue_msg(so, mp, msgdsize(mp));
1376 mp = nmp;
1377 } while (mp != NULL);
1378 }
1379
1380 space_left = so->so_rcvbuf - so->so_rcv_queued;
1381 if (space_left <= 0) {
1382 so->so_flowctrld = B_TRUE;
1383 *errorp = ENOSPC;
1384 space_left = -1;
1385 }
1386
1387 if (force_push || so->so_rcv_queued >= so->so_rcv_thresh ||
1388 so->so_rcv_queued >= so->so_rcv_wanted) {
1389 SOCKET_TIMER_CANCEL(so);
1390 /*
1391 * so_notify_data will release the lock
1392 */
1393 so_notify_data(so, so->so_rcv_queued);
1394
1395 if (force_pushp != NULL)
1396 *force_pushp = B_TRUE;
1397 goto done;
1398 } else if (so->so_rcv_timer_tid == 0) {
1399 /* Make sure the recv push timer is running */
1400 SOCKET_TIMER_START(so);
1401 }
1402
1403 done_unlock:
1404 mutex_exit(&so->so_lock);
1405 done:
1406 return (space_left);
1407
1408 space_check:
1409 space_left = so->so_rcvbuf - so->so_rcv_queued;
1410 if (space_left <= 0) {
1411 so->so_flowctrld = B_TRUE;
1412 *errorp = ENOSPC;
1413 space_left = -1;
1414 }
1415 goto done_unlock;
1416 }
1417
1418 #pragma inline(so_queue_msg_impl)
1419
1420 ssize_t
1421 so_queue_msg(sock_upper_handle_t sock_handle, mblk_t *mp,
1422 size_t msg_size, int flags, int *errorp, boolean_t *force_pushp)
1423 {
1424 struct sonode *so = (struct sonode *)sock_handle;
1425
1426 return (so_queue_msg_impl(so, mp, msg_size, flags, errorp, force_pushp,
1427 so->so_filter_bottom));
1428 }
1429
1430 /*
1431 * Set the offset of where the oob data is relative to the bytes in
1432 * queued. Also generate SIGURG
1433 */
1434 void
1435 so_signal_oob(sock_upper_handle_t sock_handle, ssize_t offset)
1436 {
1437 struct sonode *so;
1438
1439 ASSERT(offset >= 0);
1440 so = (struct sonode *)sock_handle;
1441 mutex_enter(&so->so_lock);
1442 if (so->so_direct != NULL)
1443 SOD_UIOAFINI(so->so_direct);
1444
1445 /*
1446 * New urgent data on the way so forget about any old
1447 * urgent data.
1448 */
1449 so->so_state &= ~(SS_HAVEOOBDATA|SS_HADOOBDATA);
1450
1451 /*
1452 * Record that urgent data is pending.
1453 */
1454 so->so_state |= SS_OOBPEND;
1455
1456 if (so->so_oobmsg != NULL) {
1457 dprintso(so, 1, ("sock: discarding old oob\n"));
1458 freemsg(so->so_oobmsg);
1459 so->so_oobmsg = NULL;
1460 }
1461
1462 /*
1463 * set the offset where the urgent byte is
1464 */
1465 so->so_oobmark = so->so_rcv_queued + offset;
1466 if (so->so_oobmark == 0)
1467 so->so_state |= SS_RCVATMARK;
1468 else
1469 so->so_state &= ~SS_RCVATMARK;
1470
1471 so_notify_oobsig(so);
1472 }
1473
1474 /*
1475 * Queue the OOB byte
1476 */
1477 static void
1478 so_queue_oob(struct sonode *so, mblk_t *mp, size_t len)
1479 {
1480 mutex_enter(&so->so_lock);
1481 if (so->so_direct != NULL)
1482 SOD_UIOAFINI(so->so_direct);
1483
1484 ASSERT(mp != NULL);
1485 if (!IS_SO_OOB_INLINE(so)) {
1486 so->so_oobmsg = mp;
1487 so->so_state |= SS_HAVEOOBDATA;
1488 } else {
1489 so_enqueue_msg(so, mp, len);
1490 }
1491
1492 so_notify_oobdata(so, IS_SO_OOB_INLINE(so));
1493 }
1494
1495 int
1496 so_close(struct sonode *so, int flag, struct cred *cr)
1497 {
1498 int error;
1499
1500 /*
1501 * No new data will be enqueued once the CLOSING flag is set.
1502 */
1503 mutex_enter(&so->so_lock);
1504 so->so_state |= SS_CLOSING;
1505 ASSERT(so_verify_oobstate(so));
1506 so_rcv_flush(so);
1507 mutex_exit(&so->so_lock);
1508
1509 if (so->so_filter_active > 0)
1510 sof_sonode_closing(so);
1511
1512 if (so->so_state & SS_ACCEPTCONN) {
1513 /*
1514 * We grab and release the accept lock to ensure that any
1515 * thread about to insert a socket in so_newconn completes
1516 * before we flush the queue. Any thread calling so_newconn
1517 * after we drop the lock will observe the SS_CLOSING flag,
1518 * which will stop it from inserting the socket in the queue.
1519 */
1520 mutex_enter(&so->so_acceptq_lock);
1521 mutex_exit(&so->so_acceptq_lock);
1522
1523 so_acceptq_flush(so, B_TRUE);
1524 }
1525
1526 error = (*so->so_downcalls->sd_close)(so->so_proto_handle, flag, cr);
1527 switch (error) {
1528 default:
1529 /* Protocol made a synchronous close; remove proto ref */
1530 VN_RELE(SOTOV(so));
1531 break;
1532 case EINPROGRESS:
1533 /*
1534 * Protocol is in the process of closing, it will make a
1535 * 'closed' upcall to remove the reference.
1536 */
1537 error = 0;
1538 break;
1539 }
1540
1541 return (error);
1542 }
1543
1544 /*
1545 * Upcall made by the protocol when it's doing an asynchronous close. It
1546 * will drop the protocol's reference on the socket.
1547 */
1548 void
1549 so_closed(sock_upper_handle_t sock_handle)
1550 {
1551 struct sonode *so = (struct sonode *)sock_handle;
1552
1553 VN_RELE(SOTOV(so));
1554 }
1555
1556 void
1557 so_zcopy_notify(sock_upper_handle_t sock_handle)
1558 {
1559 struct sonode *so = (struct sonode *)sock_handle;
1560
1561 mutex_enter(&so->so_lock);
1562 so->so_copyflag |= STZCNOTIFY;
1563 cv_broadcast(&so->so_copy_cv);
1564 mutex_exit(&so->so_lock);
1565 }
1566
1567 void
1568 so_set_error(sock_upper_handle_t sock_handle, int error)
1569 {
1570 struct sonode *so = (struct sonode *)sock_handle;
1571
1572 mutex_enter(&so->so_lock);
1573
1574 soseterror(so, error);
1575
1576 so_notify_error(so);
1577 }
1578
1579 /*
1580 * so_recvmsg - read data from the socket
1581 *
1582 * There are two ways of obtaining data; either we ask the protocol to
1583 * copy directly into the supplied buffer, or we copy data from the
1584 * sonode's receive queue. The decision which one to use depends on
1585 * whether the protocol has a sd_recv_uio down call.
1586 */
1587 int
1588 so_recvmsg(struct sonode *so, struct nmsghdr *msg, struct uio *uiop,
1589 struct cred *cr)
1590 {
1591 rval_t rval;
1592 int flags = 0;
1593 t_uscalar_t controllen, namelen;
1594 int error = 0;
1595 int ret;
1596 mblk_t *mctlp = NULL;
1597 union T_primitives *tpr;
1598 void *control;
1599 ssize_t saved_resid;
1600 struct uio *suiop;
1601
1602 SO_BLOCK_FALLBACK(so, SOP_RECVMSG(so, msg, uiop, cr));
1603
1604 if ((so->so_state & (SS_ISCONNECTED|SS_CANTRCVMORE)) == 0 &&
1605 (so->so_mode & SM_CONNREQUIRED)) {
1606 SO_UNBLOCK_FALLBACK(so);
1607 return (ENOTCONN);
1608 }
1609
1610 if (msg->msg_flags & MSG_PEEK)
1611 msg->msg_flags &= ~MSG_WAITALL;
1612
1613 if (so->so_mode & SM_ATOMIC)
1614 msg->msg_flags |= MSG_TRUNC;
1615
1616 if (msg->msg_flags & MSG_OOB) {
1617 if ((so->so_mode & SM_EXDATA) == 0) {
1618 error = EOPNOTSUPP;
1619 } else if (so->so_downcalls->sd_recv_uio != NULL) {
1620 error = (*so->so_downcalls->sd_recv_uio)
1621 (so->so_proto_handle, uiop, msg, cr);
1622 } else {
1623 error = sorecvoob(so, msg, uiop, msg->msg_flags,
1624 IS_SO_OOB_INLINE(so));
1625 }
1626 SO_UNBLOCK_FALLBACK(so);
1627 return (error);
1628 }
1629
1630 /*
1631 * If the protocol has the recv down call, then pass the request
1632 * down.
1633 */
1634 if (so->so_downcalls->sd_recv_uio != NULL) {
1635 error = (*so->so_downcalls->sd_recv_uio)
1636 (so->so_proto_handle, uiop, msg, cr);
1637 SO_UNBLOCK_FALLBACK(so);
1638 return (error);
1639 }
1640
1641 /*
1642 * Reading data from the socket buffer
1643 */
1644 flags = msg->msg_flags;
1645 msg->msg_flags = 0;
1646
1647 /*
1648 * Set msg_controllen and msg_namelen to zero here to make it
1649 * simpler in the cases that no control or name is returned.
1650 */
1651 controllen = msg->msg_controllen;
1652 namelen = msg->msg_namelen;
1653 msg->msg_controllen = 0;
1654 msg->msg_namelen = 0;
1655
1656 mutex_enter(&so->so_lock);
1657 /* Set SOREADLOCKED */
1658 error = so_lock_read_intr(so,
1659 uiop->uio_fmode | ((flags & MSG_DONTWAIT) ? FNONBLOCK : 0));
1660 mutex_exit(&so->so_lock);
1661 if (error) {
1662 SO_UNBLOCK_FALLBACK(so);
1663 return (error);
1664 }
1665
1666 suiop = sod_rcv_init(so, flags, &uiop);
1667 retry:
1668 saved_resid = uiop->uio_resid;
1669 error = so_dequeue_msg(so, &mctlp, uiop, &rval, flags);
1670 if (error != 0) {
1671 goto out;
1672 }
1673 /*
1674 * For datagrams the MOREDATA flag is used to set MSG_TRUNC.
1675 * For non-datagrams MOREDATA is used to set MSG_EOR.
1676 */
1677 ASSERT(!(rval.r_val1 & MORECTL));
1678 if ((rval.r_val1 & MOREDATA) && (so->so_mode & SM_ATOMIC))
1679 msg->msg_flags |= MSG_TRUNC;
1680 if (mctlp == NULL) {
1681 dprintso(so, 1, ("so_recvmsg: got M_DATA\n"));
1682
1683 mutex_enter(&so->so_lock);
1684 /* Set MSG_EOR based on MOREDATA */
1685 if (!(rval.r_val1 & MOREDATA)) {
1686 if (so->so_state & SS_SAVEDEOR) {
1687 msg->msg_flags |= MSG_EOR;
1688 so->so_state &= ~SS_SAVEDEOR;
1689 }
1690 }
1691 /*
1692 * If some data was received (i.e. not EOF) and the
1693 * read/recv* has not been satisfied wait for some more.
1694 */
1695 if ((flags & MSG_WAITALL) && !(msg->msg_flags & MSG_EOR) &&
1696 uiop->uio_resid != saved_resid && uiop->uio_resid > 0) {
1697 mutex_exit(&so->so_lock);
1698 flags |= MSG_NOMARK;
1699 goto retry;
1700 }
1701
1702 goto out_locked;
1703 }
1704 /* so_queue_msg has already verified length and alignment */
1705 tpr = (union T_primitives *)mctlp->b_rptr;
1706 dprintso(so, 1, ("so_recvmsg: type %d\n", tpr->type));
1707 switch (tpr->type) {
1708 case T_DATA_IND: {
1709 /*
1710 * Set msg_flags to MSG_EOR based on
1711 * MORE_flag and MOREDATA.
1712 */
1713 mutex_enter(&so->so_lock);
1714 so->so_state &= ~SS_SAVEDEOR;
1715 if (!(tpr->data_ind.MORE_flag & 1)) {
1716 if (!(rval.r_val1 & MOREDATA))
1717 msg->msg_flags |= MSG_EOR;
1718 else
1719 so->so_state |= SS_SAVEDEOR;
1720 }
1721 freemsg(mctlp);
1722 /*
1723 * If some data was received (i.e. not EOF) and the
1724 * read/recv* has not been satisfied wait for some more.
1725 */
1726 if ((flags & MSG_WAITALL) && !(msg->msg_flags & MSG_EOR) &&
1727 uiop->uio_resid != saved_resid && uiop->uio_resid > 0) {
1728 mutex_exit(&so->so_lock);
1729 flags |= MSG_NOMARK;
1730 goto retry;
1731 }
1732 goto out_locked;
1733 }
1734 case T_UNITDATA_IND: {
1735 void *addr;
1736 t_uscalar_t addrlen;
1737 void *abuf;
1738 t_uscalar_t optlen;
1739 void *opt;
1740
1741 if (namelen != 0) {
1742 /* Caller wants source address */
1743 addrlen = tpr->unitdata_ind.SRC_length;
1744 addr = sogetoff(mctlp, tpr->unitdata_ind.SRC_offset,
1745 addrlen, 1);
1746 if (addr == NULL) {
1747 freemsg(mctlp);
1748 error = EPROTO;
1749 eprintsoline(so, error);
1750 goto out;
1751 }
1752 ASSERT(so->so_family != AF_UNIX);
1753 }
1754 optlen = tpr->unitdata_ind.OPT_length;
1755 if (optlen != 0) {
1756 t_uscalar_t ncontrollen;
1757
1758 /*
1759 * Extract any source address option.
1760 * Determine how large cmsg buffer is needed.
1761 */
1762 opt = sogetoff(mctlp, tpr->unitdata_ind.OPT_offset,
1763 optlen, __TPI_ALIGN_SIZE);
1764
1765 if (opt == NULL) {
1766 freemsg(mctlp);
1767 error = EPROTO;
1768 eprintsoline(so, error);
1769 goto out;
1770 }
1771 if (so->so_family == AF_UNIX)
1772 so_getopt_srcaddr(opt, optlen, &addr, &addrlen);
1773 ncontrollen = so_cmsglen(mctlp, opt, optlen,
1774 !(flags & MSG_XPG4_2));
1775 if (controllen != 0)
1776 controllen = ncontrollen;
1777 else if (ncontrollen != 0)
1778 msg->msg_flags |= MSG_CTRUNC;
1779 } else {
1780 controllen = 0;
1781 }
1782
1783 if (namelen != 0) {
1784 /*
1785 * Return address to caller.
1786 * Caller handles truncation if length
1787 * exceeds msg_namelen.
1788 * NOTE: AF_UNIX NUL termination is ensured by
1789 * the sender's copyin_name().
1790 */
1791 abuf = kmem_alloc(addrlen, KM_SLEEP);
1792
1793 bcopy(addr, abuf, addrlen);
1794 msg->msg_name = abuf;
1795 msg->msg_namelen = addrlen;
1796 }
1797
1798 if (controllen != 0) {
1799 /*
1800 * Return control msg to caller.
1801 * Caller handles truncation if length
1802 * exceeds msg_controllen.
1803 */
1804 control = kmem_zalloc(controllen, KM_SLEEP);
1805
1806 error = so_opt2cmsg(mctlp, opt, optlen,
1807 !(flags & MSG_XPG4_2), control, controllen);
1808 if (error) {
1809 freemsg(mctlp);
1810 if (msg->msg_namelen != 0)
1811 kmem_free(msg->msg_name,
1812 msg->msg_namelen);
1813 kmem_free(control, controllen);
1814 eprintsoline(so, error);
1815 goto out;
1816 }
1817 msg->msg_control = control;
1818 msg->msg_controllen = controllen;
1819 }
1820
1821 freemsg(mctlp);
1822 goto out;
1823 }
1824 case T_OPTDATA_IND: {
1825 struct T_optdata_req *tdr;
1826 void *opt;
1827 t_uscalar_t optlen;
1828
1829 tdr = (struct T_optdata_req *)mctlp->b_rptr;
1830 optlen = tdr->OPT_length;
1831 if (optlen != 0) {
1832 t_uscalar_t ncontrollen;
1833 /*
1834 * Determine how large cmsg buffer is needed.
1835 */
1836 opt = sogetoff(mctlp,
1837 tpr->optdata_ind.OPT_offset, optlen,
1838 __TPI_ALIGN_SIZE);
1839
1840 if (opt == NULL) {
1841 freemsg(mctlp);
1842 error = EPROTO;
1843 eprintsoline(so, error);
1844 goto out;
1845 }
1846
1847 ncontrollen = so_cmsglen(mctlp, opt, optlen,
1848 !(flags & MSG_XPG4_2));
1849 if (controllen != 0)
1850 controllen = ncontrollen;
1851 else if (ncontrollen != 0)
1852 msg->msg_flags |= MSG_CTRUNC;
1853 } else {
1854 controllen = 0;
1855 }
1856
1857 if (controllen != 0) {
1858 /*
1859 * Return control msg to caller.
1860 * Caller handles truncation if length
1861 * exceeds msg_controllen.
1862 */
1863 control = kmem_zalloc(controllen, KM_SLEEP);
1864
1865 error = so_opt2cmsg(mctlp, opt, optlen,
1866 !(flags & MSG_XPG4_2), control, controllen);
1867 if (error) {
1868 freemsg(mctlp);
1869 kmem_free(control, controllen);
1870 eprintsoline(so, error);
1871 goto out;
1872 }
1873 msg->msg_control = control;
1874 msg->msg_controllen = controllen;
1875 }
1876
1877 /*
1878 * Set msg_flags to MSG_EOR based on
1879 * DATA_flag and MOREDATA.
1880 */
1881 mutex_enter(&so->so_lock);
1882 so->so_state &= ~SS_SAVEDEOR;
1883 if (!(tpr->data_ind.MORE_flag & 1)) {
1884 if (!(rval.r_val1 & MOREDATA))
1885 msg->msg_flags |= MSG_EOR;
1886 else
1887 so->so_state |= SS_SAVEDEOR;
1888 }
1889 freemsg(mctlp);
1890 /*
1891 * If some data was received (i.e. not EOF) and the
1892 * read/recv* has not been satisfied wait for some more.
1893 * Not possible to wait if control info was received.
1894 */
1895 if ((flags & MSG_WAITALL) && !(msg->msg_flags & MSG_EOR) &&
1896 controllen == 0 &&
1897 uiop->uio_resid != saved_resid && uiop->uio_resid > 0) {
1898 mutex_exit(&so->so_lock);
1899 flags |= MSG_NOMARK;
1900 goto retry;
1901 }
1902 goto out_locked;
1903 }
1904 default:
1905 cmn_err(CE_CONT, "so_recvmsg bad type %x \n",
1906 tpr->type);
1907 freemsg(mctlp);
1908 error = EPROTO;
1909 ASSERT(0);
1910 }
1911 out:
1912 mutex_enter(&so->so_lock);
1913 out_locked:
1914 ret = sod_rcv_done(so, suiop, uiop);
1915 if (ret != 0 && error == 0)
1916 error = ret;
1917
1918 so_unlock_read(so); /* Clear SOREADLOCKED */
1919 mutex_exit(&so->so_lock);
1920
1921 SO_UNBLOCK_FALLBACK(so);
1922
1923 return (error);
1924 }
1925
1926 sonodeops_t so_sonodeops = {
1927 so_init, /* sop_init */
1928 so_accept, /* sop_accept */
1929 so_bind, /* sop_bind */
1930 so_listen, /* sop_listen */
1931 so_connect, /* sop_connect */
1932 so_recvmsg, /* sop_recvmsg */
1933 so_sendmsg, /* sop_sendmsg */
1934 so_sendmblk, /* sop_sendmblk */
1935 so_getpeername, /* sop_getpeername */
1936 so_getsockname, /* sop_getsockname */
1937 so_shutdown, /* sop_shutdown */
1938 so_getsockopt, /* sop_getsockopt */
1939 so_setsockopt, /* sop_setsockopt */
1940 so_ioctl, /* sop_ioctl */
1941 so_poll, /* sop_poll */
1942 so_close, /* sop_close */
1943 };
1944
1945 sock_upcalls_t so_upcalls = {
1946 so_newconn,
1947 so_connected,
1948 so_disconnected,
1949 so_opctl,
1950 so_queue_msg,
1951 so_set_prop,
1952 so_txq_full,
1953 so_signal_oob,
1954 so_zcopy_notify,
1955 so_set_error,
1956 so_closed
1957 };