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_SAFE(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 && (so->so_mode & SM_DEFERERR) == 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 && (so->so_mode & SM_DEFERERR) == 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 && (so->so_mode & SM_DEFERERR) == 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 error = so_sendmblk_impl(so, msg, fflag, cr, mpp, so->so_filter_top,
590 B_FALSE);
591
592 SO_UNBLOCK_FALLBACK(so);
593
594 return (error);
595 }
596
597 int
598 so_shutdown(struct sonode *so, int how, struct cred *cr)
599 {
600 int error;
601
602 SO_BLOCK_FALLBACK(so, SOP_SHUTDOWN(so, how, cr));
603
604 /*
605 * SunOS 4.X has no check for datagram sockets.
606 * 5.X checks that it is connected (ENOTCONN)
607 * X/Open requires that we check the connected state.
608 */
609 if (!(so->so_state & SS_ISCONNECTED)) {
610 if (!xnet_skip_checks) {
611 error = ENOTCONN;
612 if (xnet_check_print) {
613 printf("sockfs: X/Open shutdown check "
614 "caused ENOTCONN\n");
615 }
616 }
617 goto done;
618 }
619
620 if (so->so_filter_active == 0 ||
621 (error = sof_filter_shutdown(so, &how, cr)) < 0)
622 error = ((*so->so_downcalls->sd_shutdown)(so->so_proto_handle,
623 how, cr));
624
625 /*
626 * Protocol agreed to shutdown. We need to flush the
627 * receive buffer if the receive side is being shutdown.
628 */
629 if (error == 0 && how != SHUT_WR) {
630 mutex_enter(&so->so_lock);
631 /* wait for active reader to finish */
632 (void) so_lock_read(so, 0);
633
634 so_rcv_flush(so);
635
636 so_unlock_read(so);
637 mutex_exit(&so->so_lock);
638 }
639
640 done:
641 SO_UNBLOCK_FALLBACK(so);
642 return (error);
643 }
644
645 int
646 so_getsockname(struct sonode *so, struct sockaddr *addr,
647 socklen_t *addrlen, struct cred *cr)
648 {
649 int error;
650
651 SO_BLOCK_FALLBACK_SAFE(so, SOP_GETSOCKNAME(so, addr, addrlen, cr));
652
653 if (so->so_filter_active == 0 ||
654 (error = sof_filter_getsockname(so, addr, addrlen, cr)) < 0)
655 error = (*so->so_downcalls->sd_getsockname)
656 (so->so_proto_handle, addr, addrlen, cr);
657
658 SO_UNBLOCK_FALLBACK(so);
659 return (error);
660 }
661
662 int
663 so_getpeername(struct sonode *so, struct sockaddr *addr,
664 socklen_t *addrlen, boolean_t accept, struct cred *cr)
665 {
666 int error;
667
668 SO_BLOCK_FALLBACK(so, SOP_GETPEERNAME(so, addr, addrlen, accept, cr));
669
670 if (accept) {
671 error = (*so->so_downcalls->sd_getpeername)
672 (so->so_proto_handle, addr, addrlen, cr);
673 } else if (!(so->so_state & SS_ISCONNECTED)) {
674 error = ENOTCONN;
675 } else if ((so->so_state & SS_CANTSENDMORE) && !xnet_skip_checks) {
676 /* Added this check for X/Open */
677 error = EINVAL;
678 if (xnet_check_print) {
679 printf("sockfs: X/Open getpeername check => EINVAL\n");
680 }
681 } else if (so->so_filter_active == 0 ||
682 (error = sof_filter_getpeername(so, addr, addrlen, cr)) < 0) {
683 error = (*so->so_downcalls->sd_getpeername)
684 (so->so_proto_handle, addr, addrlen, cr);
685 }
686
687 SO_UNBLOCK_FALLBACK(so);
688 return (error);
689 }
690
691 int
692 so_getsockopt(struct sonode *so, int level, int option_name,
693 void *optval, socklen_t *optlenp, int flags, struct cred *cr)
694 {
695 int error = 0;
696
697 if (level == SOL_FILTER)
698 return (sof_getsockopt(so, option_name, optval, optlenp, cr));
699
700 SO_BLOCK_FALLBACK_SAFE(so,
701 SOP_GETSOCKOPT(so, level, option_name, optval, optlenp, flags, cr));
702
703 if ((so->so_filter_active == 0 ||
704 (error = sof_filter_getsockopt(so, level, option_name, optval,
705 optlenp, cr)) < 0) &&
706 (error = socket_getopt_common(so, level, option_name, optval,
707 optlenp, flags)) < 0) {
708 error = (*so->so_downcalls->sd_getsockopt)
709 (so->so_proto_handle, level, option_name, optval, optlenp,
710 cr);
711 if (error == ENOPROTOOPT) {
712 if (level == SOL_SOCKET) {
713 /*
714 * If a protocol does not support a particular
715 * socket option, set can fail (not allowed)
716 * but get can not fail. This is the previous
717 * sockfs bahvior.
718 */
719 switch (option_name) {
720 case SO_LINGER:
721 if (*optlenp < (t_uscalar_t)
722 sizeof (struct linger)) {
723 error = EINVAL;
724 break;
725 }
726 error = 0;
727 bzero(optval, sizeof (struct linger));
728 *optlenp = sizeof (struct linger);
729 break;
730 case SO_RCVTIMEO:
731 case SO_SNDTIMEO:
732 if (*optlenp < (t_uscalar_t)
733 sizeof (struct timeval)) {
734 error = EINVAL;
735 break;
736 }
737 error = 0;
738 bzero(optval, sizeof (struct timeval));
739 *optlenp = sizeof (struct timeval);
740 break;
741 case SO_SND_BUFINFO:
742 if (*optlenp < (t_uscalar_t)
743 sizeof (struct so_snd_bufinfo)) {
744 error = EINVAL;
745 break;
746 }
747 error = 0;
748 bzero(optval,
749 sizeof (struct so_snd_bufinfo));
750 *optlenp =
751 sizeof (struct so_snd_bufinfo);
752 break;
753 case SO_DEBUG:
754 case SO_REUSEADDR:
755 case SO_KEEPALIVE:
756 case SO_DONTROUTE:
757 case SO_BROADCAST:
758 case SO_USELOOPBACK:
759 case SO_OOBINLINE:
760 case SO_DGRAM_ERRIND:
761 case SO_SNDBUF:
762 case SO_RCVBUF:
763 error = 0;
764 *((int32_t *)optval) = 0;
765 *optlenp = sizeof (int32_t);
766 break;
767 default:
768 break;
769 }
770 }
771 }
772 }
773
774 SO_UNBLOCK_FALLBACK(so);
775 return (error);
776 }
777
778 int
779 so_setsockopt(struct sonode *so, int level, int option_name,
780 const void *optval, socklen_t optlen, struct cred *cr)
781 {
782 int error = 0;
783 struct timeval tl;
784 const void *opt = optval;
785
786 if (level == SOL_FILTER)
787 return (sof_setsockopt(so, option_name, optval, optlen, cr));
788
789 SO_BLOCK_FALLBACK_SAFE(so,
790 SOP_SETSOCKOPT(so, level, option_name, optval, optlen, cr));
791
792 /* X/Open requires this check */
793 if (so->so_state & SS_CANTSENDMORE && !xnet_skip_checks) {
794 SO_UNBLOCK_FALLBACK(so);
795 if (xnet_check_print)
796 printf("sockfs: X/Open setsockopt check => EINVAL\n");
797 return (EINVAL);
798 }
799
800 if (so->so_filter_active > 0 &&
801 (error = sof_filter_setsockopt(so, level, option_name,
802 (void *)optval, &optlen, cr)) >= 0)
803 goto done;
804
805 if (level == SOL_SOCKET) {
806 switch (option_name) {
807 case SO_RCVTIMEO:
808 case SO_SNDTIMEO: {
809 /*
810 * We pass down these two options to protocol in order
811 * to support some third part protocols which need to
812 * know them. For those protocols which don't care
813 * these two options, simply return 0.
814 */
815 clock_t t_usec;
816
817 if (get_udatamodel() == DATAMODEL_NONE ||
818 get_udatamodel() == DATAMODEL_NATIVE) {
819 if (optlen != sizeof (struct timeval)) {
820 error = EINVAL;
821 goto done;
822 }
823 bcopy((struct timeval *)optval, &tl,
824 sizeof (struct timeval));
825 } else {
826 if (optlen != sizeof (struct timeval32)) {
827 error = EINVAL;
828 goto done;
829 }
830 TIMEVAL32_TO_TIMEVAL(&tl,
831 (struct timeval32 *)optval);
832 }
833 opt = &tl;
834 optlen = sizeof (tl);
835 t_usec = tl.tv_sec * 1000 * 1000 + tl.tv_usec;
836 mutex_enter(&so->so_lock);
837 if (option_name == SO_RCVTIMEO)
838 so->so_rcvtimeo = drv_usectohz(t_usec);
839 else
840 so->so_sndtimeo = drv_usectohz(t_usec);
841 mutex_exit(&so->so_lock);
842 break;
843 }
844 case SO_RCVBUF:
845 /*
846 * XXX XPG 4.2 applications retrieve SO_RCVBUF from
847 * sockfs since the transport might adjust the value
848 * and not return exactly what was set by the
849 * application.
850 */
851 so->so_xpg_rcvbuf = *(int32_t *)optval;
852 break;
853 }
854 }
855 error = (*so->so_downcalls->sd_setsockopt)
856 (so->so_proto_handle, level, option_name, opt, optlen, cr);
857 done:
858 SO_UNBLOCK_FALLBACK(so);
859 return (error);
860 }
861
862 int
863 so_ioctl(struct sonode *so, int cmd, intptr_t arg, int mode,
864 struct cred *cr, int32_t *rvalp)
865 {
866 int error = 0;
867
868 SO_BLOCK_FALLBACK(so, SOP_IOCTL(so, cmd, arg, mode, cr, rvalp));
869
870 /*
871 * If there is a pending error, return error
872 * This can happen if a non blocking operation caused an error.
873 */
874 if (so->so_error != 0 && (so->so_mode & SM_DEFERERR) == 0) {
875 mutex_enter(&so->so_lock);
876 error = sogeterr(so, B_TRUE);
877 mutex_exit(&so->so_lock);
878 if (error != 0)
879 goto done;
880 }
881
882 /*
883 * calling strioc can result in the socket falling back to TPI,
884 * if that is supported.
885 */
886 if ((so->so_filter_active == 0 ||
887 (error = sof_filter_ioctl(so, cmd, arg, mode,
888 rvalp, cr)) < 0) &&
889 (error = socket_ioctl_common(so, cmd, arg, mode, cr, rvalp)) < 0 &&
890 (error = socket_strioc_common(so, cmd, arg, mode, cr, rvalp)) < 0) {
891 error = (*so->so_downcalls->sd_ioctl)(so->so_proto_handle,
892 cmd, arg, mode, rvalp, cr);
893 }
894
895 done:
896 SO_UNBLOCK_FALLBACK(so);
897
898 return (error);
899 }
900
901 int
902 so_poll(struct sonode *so, short events, int anyyet, short *reventsp,
903 struct pollhead **phpp)
904 {
905 int state = so->so_state, mask;
906 *reventsp = 0;
907
908 /*
909 * In sockets the errors are represented as input/output events
910 */
911 if (so->so_error != 0 &&
912 ((POLLIN|POLLRDNORM|POLLOUT) & events) != 0) {
913 *reventsp = (POLLIN|POLLRDNORM|POLLOUT) & events;
914 return (0);
915 }
916
917 /*
918 * If the socket is in a state where it can send data
919 * turn on POLLWRBAND and POLLOUT events.
920 */
921 if ((so->so_mode & SM_CONNREQUIRED) == 0 || (state & SS_ISCONNECTED)) {
922 /*
923 * out of band data is allowed even if the connection
924 * is flow controlled
925 */
926 *reventsp |= POLLWRBAND & events;
927 if (!SO_SND_FLOWCTRLD(so)) {
928 /*
929 * As long as there is buffer to send data
930 * turn on POLLOUT events
931 */
932 *reventsp |= POLLOUT & events;
933 }
934 }
935
936 /*
937 * Turn on POLLIN whenever there is data on the receive queue,
938 * or the socket is in a state where no more data will be received.
939 * Also, if the socket is accepting connections, flip the bit if
940 * there is something on the queue.
941 *
942 * We do an initial check for events without holding locks. However,
943 * if there are no event available, then we redo the check for POLLIN
944 * events under the lock.
945 */
946
947 /* Pending connections */
948 if (!list_is_empty(&so->so_acceptq_list))
949 *reventsp |= (POLLIN|POLLRDNORM) & events;
950
951 /*
952 * If we're looking for POLLRDHUP, indicate it if we have sent the
953 * last rx signal for the socket.
954 */
955 if ((events & POLLRDHUP) && (state & SS_SENTLASTREADSIG))
956 *reventsp |= POLLRDHUP;
957
958 /* Data */
959 /* so_downcalls is null for sctp */
960 if (so->so_downcalls != NULL && so->so_downcalls->sd_poll != NULL) {
961 *reventsp |= (*so->so_downcalls->sd_poll)
962 (so->so_proto_handle, events & SO_PROTO_POLLEV, anyyet,
963 CRED()) & events;
964 ASSERT((*reventsp & ~events) == 0);
965 /* do not recheck events */
966 events &= ~SO_PROTO_POLLEV;
967 } else {
968 if (SO_HAVE_DATA(so))
969 *reventsp |= (POLLIN|POLLRDNORM) & events;
970
971 /* Urgent data */
972 if ((state & SS_OOBPEND) != 0) {
973 *reventsp |= (POLLRDBAND | POLLPRI) & events;
974 }
975
976 /*
977 * If the socket has become disconnected, we set POLLHUP.
978 * Note that if we are in this state, we will have set POLLIN
979 * (SO_HAVE_DATA() is true on a disconnected socket), but not
980 * POLLOUT (SS_ISCONNECTED is false). This is in keeping with
981 * the semantics of POLLHUP, which is defined to be mutually
982 * exclusive with respect to POLLOUT but not POLLIN. We are
983 * therefore setting POLLHUP primarily for the benefit of
984 * those not polling on POLLIN, as they have no other way of
985 * knowing that the socket has been disconnected.
986 */
987 mask = SS_SENTLASTREADSIG | SS_SENTLASTWRITESIG;
988
989 if ((state & (mask | SS_ISCONNECTED)) == mask)
990 *reventsp |= POLLHUP;
991 }
992
993 if ((!*reventsp && !anyyet) || (events & POLLET)) {
994 /* Check for read events again, but this time under lock */
995 if (events & (POLLIN|POLLRDNORM)) {
996 mutex_enter(&so->so_lock);
997 if (SO_HAVE_DATA(so) ||
998 !list_is_empty(&so->so_acceptq_list)) {
999 if (events & POLLET) {
1000 so->so_pollev |= SO_POLLEV_IN;
1001 *phpp = &so->so_poll_list;
1002 }
1003
1004 mutex_exit(&so->so_lock);
1005 *reventsp |= (POLLIN|POLLRDNORM) & events;
1006
1007 return (0);
1008 } else {
1009 so->so_pollev |= SO_POLLEV_IN;
1010 mutex_exit(&so->so_lock);
1011 }
1012 }
1013 *phpp = &so->so_poll_list;
1014 }
1015 return (0);
1016 }
1017
1018 /*
1019 * Generic Upcalls
1020 */
1021 void
1022 so_connected(sock_upper_handle_t sock_handle, sock_connid_t id,
1023 cred_t *peer_cred, pid_t peer_cpid)
1024 {
1025 struct sonode *so = (struct sonode *)sock_handle;
1026
1027 mutex_enter(&so->so_lock);
1028 ASSERT(so->so_proto_handle != NULL);
1029
1030 if (peer_cred != NULL) {
1031 if (so->so_peercred != NULL)
1032 crfree(so->so_peercred);
1033 crhold(peer_cred);
1034 so->so_peercred = peer_cred;
1035 so->so_cpid = peer_cpid;
1036 }
1037
1038 so->so_proto_connid = id;
1039 soisconnected(so);
1040 /*
1041 * Wake ones who're waiting for conn to become established.
1042 */
1043 so_notify_connected(so);
1044 }
1045
1046 int
1047 so_disconnected(sock_upper_handle_t sock_handle, sock_connid_t id, int error)
1048 {
1049 struct sonode *so = (struct sonode *)sock_handle;
1050 boolean_t connect_failed;
1051
1052 mutex_enter(&so->so_lock);
1053
1054 /*
1055 * If we aren't currently connected, then this isn't a disconnect but
1056 * rather a failure to connect.
1057 */
1058 connect_failed = !(so->so_state & SS_ISCONNECTED);
1059
1060 so->so_proto_connid = id;
1061 soisdisconnected(so, error);
1062 so_notify_disconnected(so, connect_failed, error);
1063
1064 return (0);
1065 }
1066
1067 void
1068 so_opctl(sock_upper_handle_t sock_handle, sock_opctl_action_t action,
1069 uintptr_t arg)
1070 {
1071 struct sonode *so = (struct sonode *)sock_handle;
1072
1073 switch (action) {
1074 case SOCK_OPCTL_SHUT_SEND:
1075 mutex_enter(&so->so_lock);
1076 socantsendmore(so);
1077 so_notify_disconnecting(so);
1078 break;
1079 case SOCK_OPCTL_SHUT_RECV: {
1080 mutex_enter(&so->so_lock);
1081 socantrcvmore(so);
1082 so_notify_eof(so);
1083 break;
1084 }
1085 case SOCK_OPCTL_ENAB_ACCEPT:
1086 mutex_enter(&so->so_lock);
1087 so->so_state |= SS_ACCEPTCONN;
1088 so->so_backlog = (unsigned int)arg;
1089 /*
1090 * The protocol can stop generating newconn upcalls when
1091 * the backlog is full, so to make sure the listener does
1092 * not end up with a queue full of deferred connections
1093 * we reduce the backlog by one. Thus the listener will
1094 * start closing deferred connections before the backlog
1095 * is full.
1096 */
1097 if (so->so_filter_active > 0)
1098 so->so_backlog = MAX(1, so->so_backlog - 1);
1099 mutex_exit(&so->so_lock);
1100 break;
1101 default:
1102 ASSERT(0);
1103 break;
1104 }
1105 }
1106
1107 void
1108 so_txq_full(sock_upper_handle_t sock_handle, boolean_t qfull)
1109 {
1110 struct sonode *so = (struct sonode *)sock_handle;
1111
1112 if (qfull) {
1113 so_snd_qfull(so);
1114 } else {
1115 so_snd_qnotfull(so);
1116 mutex_enter(&so->so_lock);
1117 /* so_notify_writable drops so_lock */
1118 so_notify_writable(so);
1119 }
1120 }
1121
1122 sock_upper_handle_t
1123 so_newconn(sock_upper_handle_t parenthandle,
1124 sock_lower_handle_t proto_handle, sock_downcalls_t *sock_downcalls,
1125 struct cred *peer_cred, pid_t peer_cpid, sock_upcalls_t **sock_upcallsp)
1126 {
1127 struct sonode *so = (struct sonode *)parenthandle;
1128 struct sonode *nso;
1129 int error;
1130
1131 ASSERT(proto_handle != NULL);
1132
1133 if ((so->so_state & SS_ACCEPTCONN) == 0 ||
1134 (so->so_acceptq_len >= so->so_backlog &&
1135 (so->so_filter_active == 0 || !sof_sonode_drop_deferred(so)))) {
1136 return (NULL);
1137 }
1138
1139 nso = socket_newconn(so, proto_handle, sock_downcalls, SOCKET_NOSLEEP,
1140 &error);
1141 if (nso == NULL)
1142 return (NULL);
1143
1144 if (peer_cred != NULL) {
1145 crhold(peer_cred);
1146 nso->so_peercred = peer_cred;
1147 nso->so_cpid = peer_cpid;
1148 }
1149 nso->so_listener = so;
1150
1151 /*
1152 * The new socket (nso), proto_handle and sock_upcallsp are all
1153 * valid at this point. But as soon as nso is placed in the accept
1154 * queue that can no longer be assumed (since an accept() thread may
1155 * pull it off the queue and close the socket).
1156 */
1157 *sock_upcallsp = &so_upcalls;
1158
1159 mutex_enter(&so->so_acceptq_lock);
1160 if (so->so_state & (SS_CLOSING|SS_FALLBACK_PENDING|SS_FALLBACK_COMP)) {
1161 mutex_exit(&so->so_acceptq_lock);
1162 ASSERT(nso->so_count == 1);
1163 nso->so_count--;
1164 nso->so_listener = NULL;
1165 /* drop proto ref */
1166 VN_RELE(SOTOV(nso));
1167 socket_destroy(nso);
1168 return (NULL);
1169 } else {
1170 so->so_acceptq_len++;
1171 if (nso->so_state & SS_FIL_DEFER) {
1172 list_insert_tail(&so->so_acceptq_defer, nso);
1173 mutex_exit(&so->so_acceptq_lock);
1174 } else {
1175 list_insert_tail(&so->so_acceptq_list, nso);
1176 cv_signal(&so->so_acceptq_cv);
1177 mutex_exit(&so->so_acceptq_lock);
1178 mutex_enter(&so->so_lock);
1179 so_notify_newconn(so);
1180 }
1181
1182 return ((sock_upper_handle_t)nso);
1183 }
1184 }
1185
1186 void
1187 so_set_prop(sock_upper_handle_t sock_handle, struct sock_proto_props *soppp)
1188 {
1189 struct sonode *so;
1190
1191 so = (struct sonode *)sock_handle;
1192
1193 mutex_enter(&so->so_lock);
1194
1195 if (soppp->sopp_flags & SOCKOPT_MAXBLK)
1196 so->so_proto_props.sopp_maxblk = soppp->sopp_maxblk;
1197 if (soppp->sopp_flags & SOCKOPT_WROFF)
1198 so->so_proto_props.sopp_wroff = soppp->sopp_wroff;
1199 if (soppp->sopp_flags & SOCKOPT_TAIL)
1200 so->so_proto_props.sopp_tail = soppp->sopp_tail;
1201 if (soppp->sopp_flags & SOCKOPT_RCVHIWAT)
1202 so->so_proto_props.sopp_rxhiwat = soppp->sopp_rxhiwat;
1203 if (soppp->sopp_flags & SOCKOPT_RCVLOWAT)
1204 so->so_proto_props.sopp_rxlowat = soppp->sopp_rxlowat;
1205 if (soppp->sopp_flags & SOCKOPT_MAXPSZ)
1206 so->so_proto_props.sopp_maxpsz = soppp->sopp_maxpsz;
1207 if (soppp->sopp_flags & SOCKOPT_MINPSZ)
1208 so->so_proto_props.sopp_minpsz = soppp->sopp_minpsz;
1209 if (soppp->sopp_flags & SOCKOPT_ZCOPY) {
1210 if (soppp->sopp_zcopyflag & ZCVMSAFE) {
1211 so->so_proto_props.sopp_zcopyflag |= STZCVMSAFE;
1212 so->so_proto_props.sopp_zcopyflag &= ~STZCVMUNSAFE;
1213 } else if (soppp->sopp_zcopyflag & ZCVMUNSAFE) {
1214 so->so_proto_props.sopp_zcopyflag |= STZCVMUNSAFE;
1215 so->so_proto_props.sopp_zcopyflag &= ~STZCVMSAFE;
1216 }
1217
1218 if (soppp->sopp_zcopyflag & COPYCACHED) {
1219 so->so_proto_props.sopp_zcopyflag |= STRCOPYCACHED;
1220 }
1221 }
1222 if (soppp->sopp_flags & SOCKOPT_OOBINLINE)
1223 so->so_proto_props.sopp_oobinline = soppp->sopp_oobinline;
1224 if (soppp->sopp_flags & SOCKOPT_RCVTIMER)
1225 so->so_proto_props.sopp_rcvtimer = soppp->sopp_rcvtimer;
1226 if (soppp->sopp_flags & SOCKOPT_RCVTHRESH)
1227 so->so_proto_props.sopp_rcvthresh = soppp->sopp_rcvthresh;
1228 if (soppp->sopp_flags & SOCKOPT_MAXADDRLEN)
1229 so->so_proto_props.sopp_maxaddrlen = soppp->sopp_maxaddrlen;
1230 if (soppp->sopp_flags & SOCKOPT_LOOPBACK)
1231 so->so_proto_props.sopp_loopback = soppp->sopp_loopback;
1232
1233 mutex_exit(&so->so_lock);
1234
1235 if (so->so_filter_active > 0) {
1236 sof_instance_t *inst;
1237 ssize_t maxblk;
1238 ushort_t wroff, tail;
1239 maxblk = so->so_proto_props.sopp_maxblk;
1240 wroff = so->so_proto_props.sopp_wroff;
1241 tail = so->so_proto_props.sopp_tail;
1242 for (inst = so->so_filter_bottom; inst != NULL;
1243 inst = inst->sofi_prev) {
1244 if (SOF_INTERESTED(inst, mblk_prop)) {
1245 (*inst->sofi_ops->sofop_mblk_prop)(
1246 (sof_handle_t)inst, inst->sofi_cookie,
1247 &maxblk, &wroff, &tail);
1248 }
1249 }
1250 mutex_enter(&so->so_lock);
1251 so->so_proto_props.sopp_maxblk = maxblk;
1252 so->so_proto_props.sopp_wroff = wroff;
1253 so->so_proto_props.sopp_tail = tail;
1254 mutex_exit(&so->so_lock);
1255 }
1256 #ifdef DEBUG
1257 soppp->sopp_flags &= ~(SOCKOPT_MAXBLK | SOCKOPT_WROFF | SOCKOPT_TAIL |
1258 SOCKOPT_RCVHIWAT | SOCKOPT_RCVLOWAT | SOCKOPT_MAXPSZ |
1259 SOCKOPT_ZCOPY | SOCKOPT_OOBINLINE | SOCKOPT_RCVTIMER |
1260 SOCKOPT_RCVTHRESH | SOCKOPT_MAXADDRLEN | SOCKOPT_MINPSZ |
1261 SOCKOPT_LOOPBACK);
1262 ASSERT(soppp->sopp_flags == 0);
1263 #endif
1264 }
1265
1266 /* ARGSUSED */
1267 ssize_t
1268 so_queue_msg_impl(struct sonode *so, mblk_t *mp,
1269 size_t msg_size, int flags, int *errorp, boolean_t *force_pushp,
1270 sof_instance_t *filter)
1271 {
1272 boolean_t force_push = B_TRUE;
1273 int space_left;
1274 sodirect_t *sodp = so->so_direct;
1275
1276 ASSERT(errorp != NULL);
1277 *errorp = 0;
1278 if (mp == NULL) {
1279 if (so->so_downcalls->sd_recv_uio != NULL) {
1280 mutex_enter(&so->so_lock);
1281 /* the notify functions will drop the lock */
1282 if (flags & MSG_OOB)
1283 so_notify_oobdata(so, IS_SO_OOB_INLINE(so));
1284 else
1285 so_notify_data(so, msg_size);
1286 return (0);
1287 }
1288 ASSERT(msg_size == 0);
1289 mutex_enter(&so->so_lock);
1290 goto space_check;
1291 }
1292
1293 ASSERT(mp->b_next == NULL);
1294 ASSERT(DB_TYPE(mp) == M_DATA || DB_TYPE(mp) == M_PROTO);
1295 ASSERT(msg_size == msgdsize(mp));
1296
1297 if (DB_TYPE(mp) == M_PROTO && !__TPI_PRIM_ISALIGNED(mp->b_rptr)) {
1298 /* The read pointer is not aligned correctly for TPI */
1299 zcmn_err(getzoneid(), CE_WARN,
1300 "sockfs: Unaligned TPI message received. rptr = %p\n",
1301 (void *)mp->b_rptr);
1302 freemsg(mp);
1303 mutex_enter(&so->so_lock);
1304 if (sodp != NULL)
1305 SOD_UIOAFINI(sodp);
1306 goto space_check;
1307 }
1308
1309 if (so->so_filter_active > 0) {
1310 for (; filter != NULL; filter = filter->sofi_prev) {
1311 if (!SOF_INTERESTED(filter, data_in))
1312 continue;
1313 mp = (*filter->sofi_ops->sofop_data_in)(
1314 (sof_handle_t)filter, filter->sofi_cookie, mp,
1315 flags, &msg_size);
1316 ASSERT(msgdsize(mp) == msg_size);
1317 DTRACE_PROBE2(filter__data, (sof_instance_t), filter,
1318 (mblk_t *), mp);
1319 /* Data was consumed/dropped, just do space check */
1320 if (msg_size == 0) {
1321 mutex_enter(&so->so_lock);
1322 goto space_check;
1323 }
1324 }
1325 }
1326
1327 if (flags & MSG_OOB) {
1328 so_queue_oob(so, mp, msg_size);
1329 mutex_enter(&so->so_lock);
1330 goto space_check;
1331 }
1332
1333 if (force_pushp != NULL)
1334 force_push = *force_pushp;
1335
1336 mutex_enter(&so->so_lock);
1337 if (so->so_state & (SS_FALLBACK_DRAIN | SS_FALLBACK_COMP)) {
1338 if (sodp != NULL)
1339 SOD_DISABLE(sodp);
1340 mutex_exit(&so->so_lock);
1341 *errorp = EOPNOTSUPP;
1342 return (-1);
1343 }
1344 if (so->so_state & (SS_CANTRCVMORE | SS_CLOSING)) {
1345 freemsg(mp);
1346 if (sodp != NULL)
1347 SOD_DISABLE(sodp);
1348 mutex_exit(&so->so_lock);
1349 return (0);
1350 }
1351
1352 /* process the mblk via I/OAT if capable */
1353 if (sodp != NULL && sodp->sod_enabled) {
1354 if (DB_TYPE(mp) == M_DATA) {
1355 sod_uioa_mblk_init(sodp, mp, msg_size);
1356 } else {
1357 SOD_UIOAFINI(sodp);
1358 }
1359 }
1360
1361 if (mp->b_next == NULL) {
1362 so_enqueue_msg(so, mp, msg_size);
1363 } else {
1364 do {
1365 mblk_t *nmp;
1366
1367 if ((nmp = mp->b_next) != NULL) {
1368 mp->b_next = NULL;
1369 }
1370 so_enqueue_msg(so, mp, msgdsize(mp));
1371 mp = nmp;
1372 } while (mp != NULL);
1373 }
1374
1375 space_left = so->so_rcvbuf - so->so_rcv_queued;
1376 if (space_left <= 0) {
1377 so->so_flowctrld = B_TRUE;
1378 *errorp = ENOSPC;
1379 space_left = -1;
1380 }
1381
1382 if (force_push || so->so_rcv_queued >= so->so_rcv_thresh ||
1383 so->so_rcv_queued >= so->so_rcv_wanted) {
1384 SOCKET_TIMER_CANCEL(so);
1385 /*
1386 * so_notify_data will release the lock
1387 */
1388 so_notify_data(so, so->so_rcv_queued);
1389
1390 if (force_pushp != NULL)
1391 *force_pushp = B_TRUE;
1392 goto done;
1393 } else if (so->so_rcv_timer_tid == 0) {
1394 /* Make sure the recv push timer is running */
1395 SOCKET_TIMER_START(so);
1396 }
1397
1398 done_unlock:
1399 mutex_exit(&so->so_lock);
1400 done:
1401 return (space_left);
1402
1403 space_check:
1404 space_left = so->so_rcvbuf - so->so_rcv_queued;
1405 if (space_left <= 0) {
1406 so->so_flowctrld = B_TRUE;
1407 *errorp = ENOSPC;
1408 space_left = -1;
1409 }
1410 goto done_unlock;
1411 }
1412
1413 #pragma inline(so_queue_msg_impl)
1414
1415 ssize_t
1416 so_queue_msg(sock_upper_handle_t sock_handle, mblk_t *mp,
1417 size_t msg_size, int flags, int *errorp, boolean_t *force_pushp)
1418 {
1419 struct sonode *so = (struct sonode *)sock_handle;
1420
1421 return (so_queue_msg_impl(so, mp, msg_size, flags, errorp, force_pushp,
1422 so->so_filter_bottom));
1423 }
1424
1425 /*
1426 * Set the offset of where the oob data is relative to the bytes in
1427 * queued. Also generate SIGURG
1428 */
1429 void
1430 so_signal_oob(sock_upper_handle_t sock_handle, ssize_t offset)
1431 {
1432 struct sonode *so;
1433
1434 ASSERT(offset >= 0);
1435 so = (struct sonode *)sock_handle;
1436 mutex_enter(&so->so_lock);
1437 if (so->so_direct != NULL)
1438 SOD_UIOAFINI(so->so_direct);
1439
1440 /*
1441 * New urgent data on the way so forget about any old
1442 * urgent data.
1443 */
1444 so->so_state &= ~(SS_HAVEOOBDATA|SS_HADOOBDATA);
1445
1446 /*
1447 * Record that urgent data is pending.
1448 */
1449 so->so_state |= SS_OOBPEND;
1450
1451 if (so->so_oobmsg != NULL) {
1452 dprintso(so, 1, ("sock: discarding old oob\n"));
1453 freemsg(so->so_oobmsg);
1454 so->so_oobmsg = NULL;
1455 }
1456
1457 /*
1458 * set the offset where the urgent byte is
1459 */
1460 so->so_oobmark = so->so_rcv_queued + offset;
1461 if (so->so_oobmark == 0)
1462 so->so_state |= SS_RCVATMARK;
1463 else
1464 so->so_state &= ~SS_RCVATMARK;
1465
1466 so_notify_oobsig(so);
1467 }
1468
1469 /*
1470 * Queue the OOB byte
1471 */
1472 static void
1473 so_queue_oob(struct sonode *so, mblk_t *mp, size_t len)
1474 {
1475 mutex_enter(&so->so_lock);
1476 if (so->so_direct != NULL)
1477 SOD_UIOAFINI(so->so_direct);
1478
1479 ASSERT(mp != NULL);
1480 if (!IS_SO_OOB_INLINE(so)) {
1481 so->so_oobmsg = mp;
1482 so->so_state |= SS_HAVEOOBDATA;
1483 } else {
1484 so_enqueue_msg(so, mp, len);
1485 }
1486
1487 so_notify_oobdata(so, IS_SO_OOB_INLINE(so));
1488 }
1489
1490 int
1491 so_close(struct sonode *so, int flag, struct cred *cr)
1492 {
1493 int error;
1494
1495 /*
1496 * No new data will be enqueued once the CLOSING flag is set.
1497 */
1498 mutex_enter(&so->so_lock);
1499 so->so_state |= SS_CLOSING;
1500 ASSERT(so_verify_oobstate(so));
1501 so_rcv_flush(so);
1502 mutex_exit(&so->so_lock);
1503
1504 if (so->so_filter_active > 0)
1505 sof_sonode_closing(so);
1506
1507 if (so->so_state & SS_ACCEPTCONN) {
1508 /*
1509 * We grab and release the accept lock to ensure that any
1510 * thread about to insert a socket in so_newconn completes
1511 * before we flush the queue. Any thread calling so_newconn
1512 * after we drop the lock will observe the SS_CLOSING flag,
1513 * which will stop it from inserting the socket in the queue.
1514 */
1515 mutex_enter(&so->so_acceptq_lock);
1516 mutex_exit(&so->so_acceptq_lock);
1517
1518 so_acceptq_flush(so, B_TRUE);
1519 }
1520
1521 error = (*so->so_downcalls->sd_close)(so->so_proto_handle, flag, cr);
1522 switch (error) {
1523 default:
1524 /* Protocol made a synchronous close; remove proto ref */
1525 VN_RELE(SOTOV(so));
1526 break;
1527 case EINPROGRESS:
1528 /*
1529 * Protocol is in the process of closing, it will make a
1530 * 'closed' upcall to remove the reference.
1531 */
1532 error = 0;
1533 break;
1534 }
1535
1536 return (error);
1537 }
1538
1539 /*
1540 * Upcall made by the protocol when it's doing an asynchronous close. It
1541 * will drop the protocol's reference on the socket.
1542 */
1543 void
1544 so_closed(sock_upper_handle_t sock_handle)
1545 {
1546 struct sonode *so = (struct sonode *)sock_handle;
1547
1548 VN_RELE(SOTOV(so));
1549 }
1550
1551 void
1552 so_zcopy_notify(sock_upper_handle_t sock_handle)
1553 {
1554 struct sonode *so = (struct sonode *)sock_handle;
1555
1556 mutex_enter(&so->so_lock);
1557 so->so_copyflag |= STZCNOTIFY;
1558 cv_broadcast(&so->so_copy_cv);
1559 mutex_exit(&so->so_lock);
1560 }
1561
1562 void
1563 so_set_error(sock_upper_handle_t sock_handle, int error)
1564 {
1565 struct sonode *so = (struct sonode *)sock_handle;
1566
1567 mutex_enter(&so->so_lock);
1568
1569 soseterror(so, error);
1570
1571 so_notify_error(so);
1572 }
1573
1574 /*
1575 * so_recvmsg - read data from the socket
1576 *
1577 * There are two ways of obtaining data; either we ask the protocol to
1578 * copy directly into the supplied buffer, or we copy data from the
1579 * sonode's receive queue. The decision which one to use depends on
1580 * whether the protocol has a sd_recv_uio down call.
1581 */
1582 int
1583 so_recvmsg(struct sonode *so, struct nmsghdr *msg, struct uio *uiop,
1584 struct cred *cr)
1585 {
1586 rval_t rval;
1587 int flags = 0;
1588 t_uscalar_t controllen, namelen;
1589 int error = 0;
1590 int ret;
1591 mblk_t *mctlp = NULL;
1592 union T_primitives *tpr;
1593 void *control;
1594 ssize_t saved_resid;
1595 struct uio *suiop;
1596
1597 SO_BLOCK_FALLBACK(so, SOP_RECVMSG(so, msg, uiop, cr));
1598
1599 if ((so->so_state & (SS_ISCONNECTED|SS_CANTRCVMORE)) == 0 &&
1600 (so->so_mode & SM_CONNREQUIRED)) {
1601 SO_UNBLOCK_FALLBACK(so);
1602 return (ENOTCONN);
1603 }
1604
1605 if (msg->msg_flags & MSG_PEEK)
1606 msg->msg_flags &= ~MSG_WAITALL;
1607
1608 if (so->so_mode & SM_ATOMIC)
1609 msg->msg_flags |= MSG_TRUNC;
1610
1611 if (msg->msg_flags & MSG_OOB) {
1612 if ((so->so_mode & SM_EXDATA) == 0) {
1613 error = EOPNOTSUPP;
1614 } else if (so->so_downcalls->sd_recv_uio != NULL) {
1615 error = (*so->so_downcalls->sd_recv_uio)
1616 (so->so_proto_handle, uiop, msg, cr);
1617 } else {
1618 error = sorecvoob(so, msg, uiop, msg->msg_flags,
1619 IS_SO_OOB_INLINE(so));
1620 }
1621 SO_UNBLOCK_FALLBACK(so);
1622 return (error);
1623 }
1624
1625 /*
1626 * If the protocol has the recv down call, then pass the request
1627 * down.
1628 */
1629 if (so->so_downcalls->sd_recv_uio != NULL) {
1630 error = (*so->so_downcalls->sd_recv_uio)
1631 (so->so_proto_handle, uiop, msg, cr);
1632 SO_UNBLOCK_FALLBACK(so);
1633 return (error);
1634 }
1635
1636 /*
1637 * Reading data from the socket buffer
1638 */
1639 flags = msg->msg_flags;
1640 msg->msg_flags = 0;
1641
1642 /*
1643 * Set msg_controllen and msg_namelen to zero here to make it
1644 * simpler in the cases that no control or name is returned.
1645 */
1646 controllen = msg->msg_controllen;
1647 namelen = msg->msg_namelen;
1648 msg->msg_controllen = 0;
1649 msg->msg_namelen = 0;
1650
1651 mutex_enter(&so->so_lock);
1652 /* Set SOREADLOCKED */
1653 error = so_lock_read_intr(so,
1654 uiop->uio_fmode | ((flags & MSG_DONTWAIT) ? FNONBLOCK : 0));
1655 mutex_exit(&so->so_lock);
1656 if (error) {
1657 SO_UNBLOCK_FALLBACK(so);
1658 return (error);
1659 }
1660
1661 suiop = sod_rcv_init(so, flags, &uiop);
1662 retry:
1663 saved_resid = uiop->uio_resid;
1664 error = so_dequeue_msg(so, &mctlp, uiop, &rval, flags);
1665 if (error != 0) {
1666 goto out;
1667 }
1668 /*
1669 * For datagrams the MOREDATA flag is used to set MSG_TRUNC.
1670 * For non-datagrams MOREDATA is used to set MSG_EOR.
1671 */
1672 ASSERT(!(rval.r_val1 & MORECTL));
1673 if ((rval.r_val1 & MOREDATA) && (so->so_mode & SM_ATOMIC))
1674 msg->msg_flags |= MSG_TRUNC;
1675 if (mctlp == NULL) {
1676 dprintso(so, 1, ("so_recvmsg: got M_DATA\n"));
1677
1678 mutex_enter(&so->so_lock);
1679 /* Set MSG_EOR based on MOREDATA */
1680 if (!(rval.r_val1 & MOREDATA)) {
1681 if (so->so_state & SS_SAVEDEOR) {
1682 msg->msg_flags |= MSG_EOR;
1683 so->so_state &= ~SS_SAVEDEOR;
1684 }
1685 }
1686 /*
1687 * If some data was received (i.e. not EOF) and the
1688 * read/recv* has not been satisfied wait for some more.
1689 */
1690 if ((flags & MSG_WAITALL) && !(msg->msg_flags & MSG_EOR) &&
1691 uiop->uio_resid != saved_resid && uiop->uio_resid > 0) {
1692 mutex_exit(&so->so_lock);
1693 flags |= MSG_NOMARK;
1694 goto retry;
1695 }
1696
1697 goto out_locked;
1698 }
1699 /* so_queue_msg has already verified length and alignment */
1700 tpr = (union T_primitives *)mctlp->b_rptr;
1701 dprintso(so, 1, ("so_recvmsg: type %d\n", tpr->type));
1702 switch (tpr->type) {
1703 case T_DATA_IND: {
1704 /*
1705 * Set msg_flags to MSG_EOR based on
1706 * MORE_flag and MOREDATA.
1707 */
1708 mutex_enter(&so->so_lock);
1709 so->so_state &= ~SS_SAVEDEOR;
1710 if (!(tpr->data_ind.MORE_flag & 1)) {
1711 if (!(rval.r_val1 & MOREDATA))
1712 msg->msg_flags |= MSG_EOR;
1713 else
1714 so->so_state |= SS_SAVEDEOR;
1715 }
1716 freemsg(mctlp);
1717 /*
1718 * If some data was received (i.e. not EOF) and the
1719 * read/recv* has not been satisfied wait for some more.
1720 */
1721 if ((flags & MSG_WAITALL) && !(msg->msg_flags & MSG_EOR) &&
1722 uiop->uio_resid != saved_resid && uiop->uio_resid > 0) {
1723 mutex_exit(&so->so_lock);
1724 flags |= MSG_NOMARK;
1725 goto retry;
1726 }
1727 goto out_locked;
1728 }
1729 case T_UNITDATA_IND: {
1730 void *addr;
1731 t_uscalar_t addrlen;
1732 void *abuf;
1733 t_uscalar_t optlen;
1734 void *opt;
1735
1736 if (namelen != 0) {
1737 /* Caller wants source address */
1738 addrlen = tpr->unitdata_ind.SRC_length;
1739 addr = sogetoff(mctlp, tpr->unitdata_ind.SRC_offset,
1740 addrlen, 1);
1741 if (addr == NULL) {
1742 freemsg(mctlp);
1743 error = EPROTO;
1744 eprintsoline(so, error);
1745 goto out;
1746 }
1747 ASSERT(so->so_family != AF_UNIX);
1748 }
1749 optlen = tpr->unitdata_ind.OPT_length;
1750 if (optlen != 0) {
1751 t_uscalar_t ncontrollen;
1752
1753 /*
1754 * Extract any source address option.
1755 * Determine how large cmsg buffer is needed.
1756 */
1757 opt = sogetoff(mctlp, tpr->unitdata_ind.OPT_offset,
1758 optlen, __TPI_ALIGN_SIZE);
1759
1760 if (opt == NULL) {
1761 freemsg(mctlp);
1762 error = EPROTO;
1763 eprintsoline(so, error);
1764 goto out;
1765 }
1766 if (so->so_family == AF_UNIX)
1767 so_getopt_srcaddr(opt, optlen, &addr, &addrlen);
1768 ncontrollen = so_cmsglen(mctlp, opt, optlen,
1769 !(flags & MSG_XPG4_2));
1770 if (controllen != 0)
1771 controllen = ncontrollen;
1772 else if (ncontrollen != 0)
1773 msg->msg_flags |= MSG_CTRUNC;
1774 } else {
1775 controllen = 0;
1776 }
1777
1778 if (namelen != 0) {
1779 /*
1780 * Return address to caller.
1781 * Caller handles truncation if length
1782 * exceeds msg_namelen.
1783 * NOTE: AF_UNIX NUL termination is ensured by
1784 * the sender's copyin_name().
1785 */
1786 abuf = kmem_alloc(addrlen, KM_SLEEP);
1787
1788 bcopy(addr, abuf, addrlen);
1789 msg->msg_name = abuf;
1790 msg->msg_namelen = addrlen;
1791 }
1792
1793 if (controllen != 0) {
1794 /*
1795 * Return control msg to caller.
1796 * Caller handles truncation if length
1797 * exceeds msg_controllen.
1798 */
1799 control = kmem_zalloc(controllen, KM_SLEEP);
1800
1801 error = so_opt2cmsg(mctlp, opt, optlen,
1802 !(flags & MSG_XPG4_2), control, controllen);
1803 if (error) {
1804 freemsg(mctlp);
1805 if (msg->msg_namelen != 0)
1806 kmem_free(msg->msg_name,
1807 msg->msg_namelen);
1808 kmem_free(control, controllen);
1809 eprintsoline(so, error);
1810 goto out;
1811 }
1812 msg->msg_control = control;
1813 msg->msg_controllen = controllen;
1814 }
1815
1816 freemsg(mctlp);
1817 goto out;
1818 }
1819 case T_OPTDATA_IND: {
1820 struct T_optdata_req *tdr;
1821 void *opt;
1822 t_uscalar_t optlen;
1823
1824 tdr = (struct T_optdata_req *)mctlp->b_rptr;
1825 optlen = tdr->OPT_length;
1826 if (optlen != 0) {
1827 t_uscalar_t ncontrollen;
1828 /*
1829 * Determine how large cmsg buffer is needed.
1830 */
1831 opt = sogetoff(mctlp,
1832 tpr->optdata_ind.OPT_offset, optlen,
1833 __TPI_ALIGN_SIZE);
1834
1835 if (opt == NULL) {
1836 freemsg(mctlp);
1837 error = EPROTO;
1838 eprintsoline(so, error);
1839 goto out;
1840 }
1841
1842 ncontrollen = so_cmsglen(mctlp, opt, optlen,
1843 !(flags & MSG_XPG4_2));
1844 if (controllen != 0)
1845 controllen = ncontrollen;
1846 else if (ncontrollen != 0)
1847 msg->msg_flags |= MSG_CTRUNC;
1848 } else {
1849 controllen = 0;
1850 }
1851
1852 if (controllen != 0) {
1853 /*
1854 * Return control msg to caller.
1855 * Caller handles truncation if length
1856 * exceeds msg_controllen.
1857 */
1858 control = kmem_zalloc(controllen, KM_SLEEP);
1859
1860 error = so_opt2cmsg(mctlp, opt, optlen,
1861 !(flags & MSG_XPG4_2), control, controllen);
1862 if (error) {
1863 freemsg(mctlp);
1864 kmem_free(control, controllen);
1865 eprintsoline(so, error);
1866 goto out;
1867 }
1868 msg->msg_control = control;
1869 msg->msg_controllen = controllen;
1870 }
1871
1872 /*
1873 * Set msg_flags to MSG_EOR based on
1874 * DATA_flag and MOREDATA.
1875 */
1876 mutex_enter(&so->so_lock);
1877 so->so_state &= ~SS_SAVEDEOR;
1878 if (!(tpr->data_ind.MORE_flag & 1)) {
1879 if (!(rval.r_val1 & MOREDATA))
1880 msg->msg_flags |= MSG_EOR;
1881 else
1882 so->so_state |= SS_SAVEDEOR;
1883 }
1884 freemsg(mctlp);
1885 /*
1886 * If some data was received (i.e. not EOF) and the
1887 * read/recv* has not been satisfied wait for some more.
1888 * Not possible to wait if control info was received.
1889 */
1890 if ((flags & MSG_WAITALL) && !(msg->msg_flags & MSG_EOR) &&
1891 controllen == 0 &&
1892 uiop->uio_resid != saved_resid && uiop->uio_resid > 0) {
1893 mutex_exit(&so->so_lock);
1894 flags |= MSG_NOMARK;
1895 goto retry;
1896 }
1897 goto out_locked;
1898 }
1899 default:
1900 cmn_err(CE_CONT, "so_recvmsg bad type %x \n",
1901 tpr->type);
1902 freemsg(mctlp);
1903 error = EPROTO;
1904 ASSERT(0);
1905 }
1906 out:
1907 mutex_enter(&so->so_lock);
1908 out_locked:
1909 ret = sod_rcv_done(so, suiop, uiop);
1910 if (ret != 0 && error == 0)
1911 error = ret;
1912
1913 so_unlock_read(so); /* Clear SOREADLOCKED */
1914 mutex_exit(&so->so_lock);
1915
1916 SO_UNBLOCK_FALLBACK(so);
1917
1918 return (error);
1919 }
1920
1921 sonodeops_t so_sonodeops = {
1922 so_init, /* sop_init */
1923 so_accept, /* sop_accept */
1924 so_bind, /* sop_bind */
1925 so_listen, /* sop_listen */
1926 so_connect, /* sop_connect */
1927 so_recvmsg, /* sop_recvmsg */
1928 so_sendmsg, /* sop_sendmsg */
1929 so_sendmblk, /* sop_sendmblk */
1930 so_getpeername, /* sop_getpeername */
1931 so_getsockname, /* sop_getsockname */
1932 so_shutdown, /* sop_shutdown */
1933 so_getsockopt, /* sop_getsockopt */
1934 so_setsockopt, /* sop_setsockopt */
1935 so_ioctl, /* sop_ioctl */
1936 so_poll, /* sop_poll */
1937 so_close, /* sop_close */
1938 };
1939
1940 sock_upcalls_t so_upcalls = {
1941 so_newconn,
1942 so_connected,
1943 so_disconnected,
1944 so_opctl,
1945 so_queue_msg,
1946 so_set_prop,
1947 so_txq_full,
1948 so_signal_oob,
1949 so_zcopy_notify,
1950 so_set_error,
1951 so_closed
1952 };