xref: /titanic_51/usr/src/uts/common/fs/sockfs/sockcommon_subr.c (revision 0c240c64cf90f44c2fdf3439010f6e8b33d85e7d)
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 2009 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 #include <sys/types.h>
28 #include <sys/param.h>
29 #include <sys/signal.h>
30 #include <sys/cmn_err.h>
31 
32 #include <sys/stropts.h>
33 #include <sys/socket.h>
34 #include <sys/socketvar.h>
35 #include <sys/sockio.h>
36 #include <sys/strsubr.h>
37 #include <sys/strsun.h>
38 #include <sys/atomic.h>
39 #include <sys/tihdr.h>
40 
41 #include <fs/sockfs/sockcommon.h>
42 #include <fs/sockfs/socktpi.h>
43 #include <fs/sockfs/sodirect.h>
44 #include <sys/ddi.h>
45 #include <inet/ip.h>
46 #include <sys/time.h>
47 #include <sys/cmn_err.h>
48 
49 #ifdef SOCK_TEST
50 extern int do_useracc;
51 extern clock_t sock_test_timelimit;
52 #endif /* SOCK_TEST */
53 
54 #define	MBLK_PULL_LEN 64
55 uint32_t so_mblk_pull_len = MBLK_PULL_LEN;
56 
57 #ifdef DEBUG
58 boolean_t so_debug_length = B_FALSE;
59 static boolean_t so_check_length(sonode_t *so);
60 #endif
61 
62 int
63 so_acceptq_enqueue_locked(struct sonode *so, struct sonode *nso)
64 {
65 	ASSERT(MUTEX_HELD(&so->so_acceptq_lock));
66 	ASSERT(nso->so_acceptq_next == NULL);
67 
68 	*so->so_acceptq_tail = nso;
69 	so->so_acceptq_tail = &nso->so_acceptq_next;
70 	so->so_acceptq_len++;
71 
72 	/* Wakeup a single consumer */
73 	cv_signal(&so->so_acceptq_cv);
74 
75 	return (so->so_acceptq_len);
76 }
77 
78 /*
79  * int so_acceptq_enqueue(struct sonode *so, struct sonode *nso)
80  *
81  * Enqueue an incoming connection on a listening socket.
82  *
83  * Arguments:
84  *   so	  - listening socket
85  *   nso  - new connection
86  *
87  * Returns:
88  *   Number of queued connections, including the new connection
89  */
90 int
91 so_acceptq_enqueue(struct sonode *so, struct sonode *nso)
92 {
93 	int conns;
94 
95 	mutex_enter(&so->so_acceptq_lock);
96 	conns = so_acceptq_enqueue_locked(so, nso);
97 	mutex_exit(&so->so_acceptq_lock);
98 
99 	return (conns);
100 }
101 
102 static int
103 so_acceptq_dequeue_locked(struct sonode *so, boolean_t dontblock,
104     struct sonode **nsop)
105 {
106 	struct sonode *nso = NULL;
107 
108 	*nsop = NULL;
109 	ASSERT(MUTEX_HELD(&so->so_acceptq_lock));
110 	while ((nso = so->so_acceptq_head) == NULL) {
111 		/*
112 		 * No need to check so_error here, because it is not
113 		 * possible for a listening socket to be reset or otherwise
114 		 * disconnected.
115 		 *
116 		 * So now we just need check if it's ok to wait.
117 		 */
118 		if (dontblock)
119 			return (EWOULDBLOCK);
120 		if (so->so_state & (SS_CLOSING | SS_FALLBACK_PENDING))
121 			return (EINTR);
122 
123 		if (cv_wait_sig_swap(&so->so_acceptq_cv,
124 		    &so->so_acceptq_lock) == 0)
125 			return (EINTR);
126 	}
127 
128 	ASSERT(nso != NULL);
129 	so->so_acceptq_head = nso->so_acceptq_next;
130 	nso->so_acceptq_next = NULL;
131 
132 	if (so->so_acceptq_head == NULL) {
133 		ASSERT(so->so_acceptq_tail == &nso->so_acceptq_next);
134 		so->so_acceptq_tail = &so->so_acceptq_head;
135 	}
136 	ASSERT(so->so_acceptq_len > 0);
137 	--so->so_acceptq_len;
138 
139 	*nsop = nso;
140 
141 	return (0);
142 }
143 
144 /*
145  * int so_acceptq_dequeue(struct sonode *, boolean_t, struct sonode **)
146  *
147  * Pulls a connection off of the accept queue.
148  *
149  * Arguments:
150  *   so	       - listening socket
151  *   dontblock - indicate whether it's ok to sleep if there are no
152  *		 connections on the queue
153  *   nsop      - Value-return argument
154  *
155  * Return values:
156  *   0 when a connection is successfully dequeued, in which case nsop
157  *   is set to point to the new connection. Upon failure a non-zero
158  *   value is returned, and the value of nsop is set to NULL.
159  *
160  * Note:
161  *   so_acceptq_dequeue() may return prematurly if the socket is falling
162  *   back to TPI.
163  */
164 int
165 so_acceptq_dequeue(struct sonode *so, boolean_t dontblock,
166     struct sonode **nsop)
167 {
168 	int error;
169 
170 	mutex_enter(&so->so_acceptq_lock);
171 	error = so_acceptq_dequeue_locked(so, dontblock, nsop);
172 	mutex_exit(&so->so_acceptq_lock);
173 
174 	return (error);
175 }
176 
177 /*
178  * void so_acceptq_flush(struct sonode *so, boolean_t doclose)
179  *
180  * Removes all pending connections from a listening socket, and
181  * frees the associated resources.
182  *
183  * Arguments
184  *   so	     - listening socket
185  *   doclose - make a close downcall for each socket on the accept queue
186  *             (Note, only SCTP and SDP sockets rely on this)
187  *
188  * Return values:
189  *   None.
190  *
191  * Note:
192  *   The caller has to ensure that no calls to so_acceptq_enqueue() or
193  *   so_acceptq_dequeue() occur while the accept queue is being flushed.
194  *   So either the socket needs to be in a state where no operations
195  *   would come in, or so_lock needs to be obtained.
196  */
197 void
198 so_acceptq_flush(struct sonode *so, boolean_t doclose)
199 {
200 	struct sonode *nso;
201 
202 	while ((nso = so->so_acceptq_head) != NULL) {
203 		so->so_acceptq_head = nso->so_acceptq_next;
204 		nso->so_acceptq_next = NULL;
205 
206 		if (doclose) {
207 			(void) socket_close(nso, 0, CRED());
208 		} else {
209 			/*
210 			 * Since the socket is on the accept queue, there can
211 			 * only be one reference. We drop the reference and
212 			 * just blow off the socket.
213 			 */
214 			ASSERT(nso->so_count == 1);
215 			nso->so_count--;
216 		}
217 		socket_destroy(nso);
218 	}
219 
220 	so->so_acceptq_head = NULL;
221 	so->so_acceptq_tail = &so->so_acceptq_head;
222 	so->so_acceptq_len = 0;
223 }
224 
225 int
226 so_wait_connected_locked(struct sonode *so, boolean_t nonblock,
227     sock_connid_t id)
228 {
229 	ASSERT(MUTEX_HELD(&so->so_lock));
230 
231 	/*
232 	 * The protocol has notified us that a connection attempt is being
233 	 * made, so before we wait for a notification to arrive we must
234 	 * clear out any errors associated with earlier connection attempts.
235 	 */
236 	if (so->so_error != 0 && SOCK_CONNID_LT(so->so_proto_connid, id))
237 		so->so_error = 0;
238 
239 	while (SOCK_CONNID_LT(so->so_proto_connid, id)) {
240 		if (nonblock)
241 			return (EINPROGRESS);
242 
243 		if (so->so_state & (SS_CLOSING | SS_FALLBACK_PENDING))
244 			return (EINTR);
245 
246 		if (cv_wait_sig_swap(&so->so_state_cv, &so->so_lock) == 0)
247 			return (EINTR);
248 	}
249 
250 	if (so->so_error != 0)
251 		return (sogeterr(so, B_TRUE));
252 	/*
253 	 * Under normal circumstances, so_error should contain an error
254 	 * in case the connect failed. However, it is possible for another
255 	 * thread to come in a consume the error, so generate a sensible
256 	 * error in that case.
257 	 */
258 	if ((so->so_state & SS_ISCONNECTED) == 0)
259 		return (ECONNREFUSED);
260 
261 	return (0);
262 }
263 
264 /*
265  * int so_wait_connected(struct sonode *so, boolean_t nonblock,
266  *    sock_connid_t id)
267  *
268  * Wait until the socket is connected or an error has occured.
269  *
270  * Arguments:
271  *   so	      - socket
272  *   nonblock - indicate whether it's ok to sleep if the connection has
273  *		not yet been established
274  *   gen      - generation number that was returned by the protocol
275  *		when the operation was started
276  *
277  * Returns:
278  *   0 if the connection attempt was successful, or an error indicating why
279  *   the connection attempt failed.
280  */
281 int
282 so_wait_connected(struct sonode *so, boolean_t nonblock, sock_connid_t id)
283 {
284 	int error;
285 
286 	mutex_enter(&so->so_lock);
287 	error = so_wait_connected_locked(so, nonblock, id);
288 	mutex_exit(&so->so_lock);
289 
290 	return (error);
291 }
292 
293 int
294 so_snd_wait_qnotfull_locked(struct sonode *so, boolean_t dontblock)
295 {
296 	int error;
297 
298 	ASSERT(MUTEX_HELD(&so->so_lock));
299 	while (so->so_snd_qfull) {
300 		if (so->so_state & SS_CANTSENDMORE)
301 			return (EPIPE);
302 		if (dontblock)
303 			return (EWOULDBLOCK);
304 
305 		if (so->so_state & (SS_CLOSING | SS_FALLBACK_PENDING))
306 			return (EINTR);
307 
308 		if (so->so_sndtimeo == 0) {
309 			/*
310 			 * Zero means disable timeout.
311 			 */
312 			error = cv_wait_sig(&so->so_snd_cv, &so->so_lock);
313 		} else {
314 			clock_t now;
315 
316 			time_to_wait(&now, so->so_sndtimeo);
317 			error = cv_timedwait_sig(&so->so_snd_cv, &so->so_lock,
318 			    now);
319 		}
320 		if (error == 0)
321 			return (EINTR);
322 		else if (error == -1)
323 			return (EAGAIN);
324 	}
325 	return (0);
326 }
327 
328 /*
329  * int so_wait_sendbuf(struct sonode *so, boolean_t dontblock)
330  *
331  * Wait for the transport to notify us about send buffers becoming
332  * available.
333  */
334 int
335 so_snd_wait_qnotfull(struct sonode *so, boolean_t dontblock)
336 {
337 	int error = 0;
338 
339 	mutex_enter(&so->so_lock);
340 	if (so->so_snd_qfull) {
341 		so->so_snd_wakeup = B_TRUE;
342 		error = so_snd_wait_qnotfull_locked(so, dontblock);
343 		so->so_snd_wakeup = B_FALSE;
344 	}
345 	mutex_exit(&so->so_lock);
346 
347 	return (error);
348 }
349 
350 void
351 so_snd_qfull(struct sonode *so)
352 {
353 	mutex_enter(&so->so_lock);
354 	so->so_snd_qfull = B_TRUE;
355 	mutex_exit(&so->so_lock);
356 }
357 
358 void
359 so_snd_qnotfull(struct sonode *so)
360 {
361 	mutex_enter(&so->so_lock);
362 	so->so_snd_qfull = B_FALSE;
363 	/* wake up everyone waiting for buffers */
364 	cv_broadcast(&so->so_snd_cv);
365 	mutex_exit(&so->so_lock);
366 }
367 
368 /*
369  * Change the process/process group to which SIGIO is sent.
370  */
371 int
372 socket_chgpgrp(struct sonode *so, pid_t pid)
373 {
374 	int error;
375 
376 	ASSERT(MUTEX_HELD(&so->so_lock));
377 	if (pid != 0) {
378 		/*
379 		 * Permissions check by sending signal 0.
380 		 * Note that when kill fails it does a
381 		 * set_errno causing the system call to fail.
382 		 */
383 		error = kill(pid, 0);
384 		if (error != 0) {
385 			return (error);
386 		}
387 	}
388 	so->so_pgrp = pid;
389 	return (0);
390 }
391 
392 
393 /*
394  * Generate a SIGIO, for 'writable' events include siginfo structure,
395  * for read events just send the signal.
396  */
397 /*ARGSUSED*/
398 static void
399 socket_sigproc(proc_t *proc, int event)
400 {
401 	k_siginfo_t info;
402 
403 	ASSERT(event & (SOCKETSIG_WRITE | SOCKETSIG_READ | SOCKETSIG_URG));
404 
405 	if (event & SOCKETSIG_WRITE) {
406 		info.si_signo = SIGPOLL;
407 		info.si_code = POLL_OUT;
408 		info.si_errno = 0;
409 		info.si_fd = 0;
410 		info.si_band = 0;
411 		sigaddq(proc, NULL, &info, KM_NOSLEEP);
412 	}
413 	if (event & SOCKETSIG_READ) {
414 		sigtoproc(proc, NULL, SIGPOLL);
415 	}
416 	if (event & SOCKETSIG_URG) {
417 		sigtoproc(proc, NULL, SIGURG);
418 	}
419 }
420 
421 void
422 socket_sendsig(struct sonode *so, int event)
423 {
424 	proc_t *proc;
425 
426 	ASSERT(MUTEX_HELD(&so->so_lock));
427 
428 	if (so->so_pgrp == 0 || (!(so->so_state & SS_ASYNC) &&
429 	    event != SOCKETSIG_URG)) {
430 		return;
431 	}
432 
433 	dprint(3, ("sending sig %d to %d\n", event, so->so_pgrp));
434 
435 	if (so->so_pgrp > 0) {
436 		/*
437 		 * XXX This unfortunately still generates
438 		 * a signal when a fd is closed but
439 		 * the proc is active.
440 		 */
441 		mutex_enter(&pidlock);
442 		proc = prfind(so->so_pgrp);
443 		if (proc == NULL) {
444 			mutex_exit(&pidlock);
445 			return;
446 		}
447 		mutex_enter(&proc->p_lock);
448 		mutex_exit(&pidlock);
449 		socket_sigproc(proc, event);
450 		mutex_exit(&proc->p_lock);
451 	} else {
452 		/*
453 		 * Send to process group. Hold pidlock across
454 		 * calls to socket_sigproc().
455 		 */
456 		pid_t pgrp = -so->so_pgrp;
457 
458 		mutex_enter(&pidlock);
459 		proc = pgfind(pgrp);
460 		while (proc != NULL) {
461 			mutex_enter(&proc->p_lock);
462 			socket_sigproc(proc, event);
463 			mutex_exit(&proc->p_lock);
464 			proc = proc->p_pglink;
465 		}
466 		mutex_exit(&pidlock);
467 	}
468 }
469 
470 #define	MIN(a, b) ((a) < (b) ? (a) : (b))
471 /* Copy userdata into a new mblk_t */
472 mblk_t *
473 socopyinuio(uio_t *uiop, ssize_t iosize, size_t wroff, ssize_t maxblk,
474     size_t tail_len, int *errorp)
475 {
476 	mblk_t	*head = NULL, **tail = &head;
477 
478 	ASSERT(iosize == INFPSZ || iosize > 0);
479 
480 	if (iosize == INFPSZ || iosize > uiop->uio_resid)
481 		iosize = uiop->uio_resid;
482 
483 	if (maxblk == INFPSZ)
484 		maxblk = iosize;
485 
486 	/* Nothing to do in these cases, so we're done */
487 	if (iosize < 0 || maxblk < 0 || (maxblk == 0 && iosize > 0))
488 		goto done;
489 
490 	/*
491 	 * We will enter the loop below if iosize is 0; it will allocate an
492 	 * empty message block and call uiomove(9F) which will just return.
493 	 * We could avoid that with an extra check but would only slow
494 	 * down the much more likely case where iosize is larger than 0.
495 	 */
496 	do {
497 		ssize_t blocksize;
498 		mblk_t	*mp;
499 
500 		blocksize = MIN(iosize, maxblk);
501 		ASSERT(blocksize >= 0);
502 		mp = allocb(wroff + blocksize + tail_len, BPRI_MED);
503 		if (mp == NULL) {
504 			*errorp = ENOMEM;
505 			return (head);
506 		}
507 		mp->b_rptr += wroff;
508 		mp->b_wptr = mp->b_rptr + blocksize;
509 
510 		*tail = mp;
511 		tail = &mp->b_cont;
512 
513 		/* uiomove(9F) either returns 0 or EFAULT */
514 		if ((*errorp = uiomove(mp->b_rptr, (size_t)blocksize,
515 		    UIO_WRITE, uiop)) != 0) {
516 			ASSERT(*errorp != ENOMEM);
517 			freemsg(head);
518 			return (NULL);
519 		}
520 
521 		iosize -= blocksize;
522 	} while (iosize > 0);
523 
524 done:
525 	*errorp = 0;
526 	return (head);
527 }
528 
529 mblk_t *
530 socopyoutuio(mblk_t *mp, struct uio *uiop, ssize_t max_read, int *errorp)
531 {
532 	int error;
533 	ptrdiff_t n;
534 	mblk_t *nmp;
535 
536 	ASSERT(mp->b_wptr >= mp->b_rptr);
537 
538 	/*
539 	 * max_read is the offset of the oobmark and read can not go pass
540 	 * the oobmark.
541 	 */
542 	if (max_read == INFPSZ || max_read > uiop->uio_resid)
543 		max_read = uiop->uio_resid;
544 
545 	do {
546 		if ((n = MIN(max_read, MBLKL(mp))) != 0) {
547 			ASSERT(n > 0);
548 
549 			error = uiomove(mp->b_rptr, n, UIO_READ, uiop);
550 			if (error != 0) {
551 				freemsg(mp);
552 				*errorp = error;
553 				return (NULL);
554 			}
555 		}
556 
557 		mp->b_rptr += n;
558 		max_read -= n;
559 		while (mp != NULL && (mp->b_rptr >= mp->b_wptr)) {
560 			/*
561 			 * get rid of zero length mblks
562 			 */
563 			nmp = mp;
564 			mp = mp->b_cont;
565 			freeb(nmp);
566 		}
567 	} while (mp != NULL && max_read > 0);
568 
569 	*errorp = 0;
570 	return (mp);
571 }
572 
573 static void
574 so_prepend_msg(struct sonode *so, mblk_t *mp, mblk_t *last_tail)
575 {
576 	ASSERT(last_tail != NULL);
577 	mp->b_next = so->so_rcv_q_head;
578 	mp->b_prev = last_tail;
579 	ASSERT(!(DB_FLAGS(mp) & DBLK_UIOA));
580 
581 	if (so->so_rcv_q_head == NULL) {
582 		ASSERT(so->so_rcv_q_last_head == NULL);
583 		so->so_rcv_q_last_head = mp;
584 #ifdef DEBUG
585 	} else {
586 		ASSERT(!(DB_FLAGS(so->so_rcv_q_head) & DBLK_UIOA));
587 #endif
588 	}
589 	so->so_rcv_q_head = mp;
590 
591 #ifdef DEBUG
592 	if (so_debug_length) {
593 		mutex_enter(&so->so_lock);
594 		ASSERT(so_check_length(so));
595 		mutex_exit(&so->so_lock);
596 	}
597 #endif
598 }
599 
600 /*
601  * Move a mblk chain (mp_head, mp_last_head) to the sonode's rcv queue so it
602  * can be processed by so_dequeue_msg().
603  */
604 void
605 so_process_new_message(struct sonode *so, mblk_t *mp_head, mblk_t *mp_last_head)
606 {
607 	ASSERT(mp_head->b_prev != NULL);
608 	if (so->so_rcv_q_head  == NULL) {
609 		so->so_rcv_q_head = mp_head;
610 		so->so_rcv_q_last_head = mp_last_head;
611 		ASSERT(so->so_rcv_q_last_head->b_prev != NULL);
612 	} else {
613 		boolean_t flag_equal = ((DB_FLAGS(mp_head) & DBLK_UIOA) ==
614 		    (DB_FLAGS(so->so_rcv_q_last_head) & DBLK_UIOA));
615 
616 		if (mp_head->b_next == NULL &&
617 		    DB_TYPE(mp_head) == M_DATA &&
618 		    DB_TYPE(so->so_rcv_q_last_head) == M_DATA && flag_equal) {
619 			so->so_rcv_q_last_head->b_prev->b_cont = mp_head;
620 			so->so_rcv_q_last_head->b_prev = mp_head->b_prev;
621 			mp_head->b_prev = NULL;
622 		} else if (flag_equal && (DB_FLAGS(mp_head) & DBLK_UIOA)) {
623 			/*
624 			 * Append to last_head if more than one mblks, and both
625 			 * mp_head and last_head are I/OAT mblks.
626 			 */
627 			ASSERT(mp_head->b_next != NULL);
628 			so->so_rcv_q_last_head->b_prev->b_cont = mp_head;
629 			so->so_rcv_q_last_head->b_prev = mp_head->b_prev;
630 			mp_head->b_prev = NULL;
631 
632 			so->so_rcv_q_last_head->b_next = mp_head->b_next;
633 			mp_head->b_next = NULL;
634 			so->so_rcv_q_last_head = mp_last_head;
635 		} else {
636 #ifdef DEBUG
637 			{
638 				mblk_t *tmp_mblk;
639 				tmp_mblk = mp_head;
640 				while (tmp_mblk != NULL) {
641 					ASSERT(tmp_mblk->b_prev != NULL);
642 					tmp_mblk = tmp_mblk->b_next;
643 				}
644 			}
645 #endif
646 			so->so_rcv_q_last_head->b_next = mp_head;
647 			so->so_rcv_q_last_head = mp_last_head;
648 		}
649 	}
650 }
651 
652 /*
653  * Check flow control on a given sonode.  Must have so_lock held, and
654  * this function will release the hold.
655  */
656 
657 static void
658 so_check_flow_control(struct sonode *so)
659 {
660 	ASSERT(MUTEX_HELD(&so->so_lock));
661 
662 	if (so->so_flowctrld && so->so_rcv_queued < so->so_rcvlowat) {
663 		so->so_flowctrld = B_FALSE;
664 		mutex_exit(&so->so_lock);
665 		/*
666 		 * Open up flow control. SCTP does not have any downcalls, and
667 		 * it will clr flow ctrl in sosctp_recvmsg().
668 		 */
669 		if (so->so_downcalls != NULL &&
670 		    so->so_downcalls->sd_clr_flowctrl != NULL) {
671 			(*so->so_downcalls->sd_clr_flowctrl)
672 			    (so->so_proto_handle);
673 		}
674 	} else {
675 		mutex_exit(&so->so_lock);
676 	}
677 }
678 
679 int
680 so_dequeue_msg(struct sonode *so, mblk_t **mctlp, struct uio *uiop,
681     rval_t *rvalp, int flags)
682 {
683 	mblk_t	*mp, *nmp;
684 	mblk_t	*savemp, *savemptail;
685 	mblk_t	*new_msg_head;
686 	mblk_t	*new_msg_last_head;
687 	mblk_t	*last_tail;
688 	boolean_t partial_read;
689 	boolean_t reset_atmark = B_FALSE;
690 	int more = 0;
691 	int error;
692 	ssize_t oobmark;
693 	sodirect_t *sodp = so->so_direct;
694 
695 	partial_read = B_FALSE;
696 	*mctlp = NULL;
697 again:
698 	mutex_enter(&so->so_lock);
699 again1:
700 #ifdef DEBUG
701 	if (so_debug_length) {
702 		ASSERT(so_check_length(so));
703 	}
704 #endif
705 	if (so->so_state & SS_RCVATMARK) {
706 		/* Check whether the caller is OK to read past the mark */
707 		if (flags & MSG_NOMARK) {
708 			mutex_exit(&so->so_lock);
709 			return (EWOULDBLOCK);
710 		}
711 		reset_atmark = B_TRUE;
712 	}
713 	/*
714 	 * First move messages from the dump area to processing area
715 	 */
716 	if (sodp != NULL) {
717 		if (sodp->sod_enabled) {
718 			if (sodp->sod_uioa.uioa_state & UIOA_ALLOC) {
719 				/* nothing to uioamove */
720 				sodp = NULL;
721 			} else if (sodp->sod_uioa.uioa_state & UIOA_INIT) {
722 				sodp->sod_uioa.uioa_state &= UIOA_CLR;
723 				sodp->sod_uioa.uioa_state |= UIOA_ENABLED;
724 				/*
725 				 * try to uioamove() the data that
726 				 * has already queued.
727 				 */
728 				sod_uioa_so_init(so, sodp, uiop);
729 			}
730 		} else {
731 			sodp = NULL;
732 		}
733 	}
734 	new_msg_head = so->so_rcv_head;
735 	new_msg_last_head = so->so_rcv_last_head;
736 	so->so_rcv_head = NULL;
737 	so->so_rcv_last_head = NULL;
738 	oobmark = so->so_oobmark;
739 	/*
740 	 * We can release the lock as there can only be one reader
741 	 */
742 	mutex_exit(&so->so_lock);
743 
744 	if (new_msg_head != NULL) {
745 		so_process_new_message(so, new_msg_head, new_msg_last_head);
746 	}
747 	savemp = savemptail = NULL;
748 	rvalp->r_val1 = 0;
749 	error = 0;
750 	mp = so->so_rcv_q_head;
751 
752 	if (mp != NULL &&
753 	    (so->so_rcv_timer_tid == 0 ||
754 	    so->so_rcv_queued >= so->so_rcv_thresh)) {
755 		partial_read = B_FALSE;
756 
757 		if (flags & MSG_PEEK) {
758 			if ((nmp = dupmsg(mp)) == NULL &&
759 			    (nmp = copymsg(mp)) == NULL) {
760 				size_t size = msgsize(mp);
761 
762 				error = strwaitbuf(size, BPRI_HI);
763 				if (error) {
764 					return (error);
765 				}
766 				goto again;
767 			}
768 			mp = nmp;
769 		} else {
770 			ASSERT(mp->b_prev != NULL);
771 			last_tail = mp->b_prev;
772 			mp->b_prev = NULL;
773 			so->so_rcv_q_head = mp->b_next;
774 			if (so->so_rcv_q_head == NULL) {
775 				so->so_rcv_q_last_head = NULL;
776 			}
777 			mp->b_next = NULL;
778 		}
779 
780 		ASSERT(mctlp != NULL);
781 		/*
782 		 * First process PROTO or PCPROTO blocks, if any.
783 		 */
784 		if (DB_TYPE(mp) != M_DATA) {
785 			*mctlp = mp;
786 			savemp = mp;
787 			savemptail = mp;
788 			ASSERT(DB_TYPE(mp) == M_PROTO ||
789 			    DB_TYPE(mp) == M_PCPROTO);
790 			while (mp->b_cont != NULL &&
791 			    DB_TYPE(mp->b_cont) != M_DATA) {
792 				ASSERT(DB_TYPE(mp->b_cont) == M_PROTO ||
793 				    DB_TYPE(mp->b_cont) == M_PCPROTO);
794 				mp = mp->b_cont;
795 				savemptail = mp;
796 			}
797 			mp = savemptail->b_cont;
798 			savemptail->b_cont = NULL;
799 		}
800 
801 		ASSERT(DB_TYPE(mp) == M_DATA);
802 		/*
803 		 * Now process DATA blocks, if any. Note that for sodirect
804 		 * enabled socket, uio_resid can be 0.
805 		 */
806 		if (uiop->uio_resid >= 0) {
807 			ssize_t copied = 0;
808 
809 			if (sodp != NULL && (DB_FLAGS(mp) & DBLK_UIOA)) {
810 				mutex_enter(&so->so_lock);
811 				ASSERT(uiop == (uio_t *)&sodp->sod_uioa);
812 				copied = sod_uioa_mblk(so, mp);
813 				if (copied > 0)
814 					partial_read = B_TRUE;
815 				mutex_exit(&so->so_lock);
816 				/* mark this mblk as processed */
817 				mp = NULL;
818 			} else {
819 				ssize_t oldresid = uiop->uio_resid;
820 
821 				if (MBLKL(mp) < so_mblk_pull_len) {
822 					if (pullupmsg(mp, -1) == 1) {
823 						last_tail = mp;
824 					}
825 				}
826 				/*
827 				 * Can not read beyond the oobmark
828 				 */
829 				mp = socopyoutuio(mp, uiop,
830 				    oobmark == 0 ? INFPSZ : oobmark, &error);
831 				if (error != 0) {
832 					freemsg(*mctlp);
833 					*mctlp = NULL;
834 					more = 0;
835 					goto done;
836 				}
837 				ASSERT(oldresid >= uiop->uio_resid);
838 				copied = oldresid - uiop->uio_resid;
839 				if (oldresid > uiop->uio_resid)
840 					partial_read = B_TRUE;
841 			}
842 			ASSERT(copied >= 0);
843 			if (copied > 0 && !(flags & MSG_PEEK)) {
844 				mutex_enter(&so->so_lock);
845 				so->so_rcv_queued -= copied;
846 				ASSERT(so->so_oobmark >= 0);
847 				if (so->so_oobmark > 0) {
848 					so->so_oobmark -= copied;
849 					ASSERT(so->so_oobmark >= 0);
850 					if (so->so_oobmark == 0) {
851 						ASSERT(so->so_state &
852 						    SS_OOBPEND);
853 						so->so_oobmark = 0;
854 						so->so_state |= SS_RCVATMARK;
855 					}
856 				}
857 				/*
858 				 * so_check_flow_control() will drop
859 				 * so->so_lock.
860 				 */
861 				so_check_flow_control(so);
862 			}
863 		}
864 		if (mp != NULL) { /* more data blocks in msg */
865 			more |= MOREDATA;
866 			if ((flags & (MSG_PEEK|MSG_TRUNC))) {
867 				if (flags & MSG_PEEK) {
868 					freemsg(mp);
869 				} else {
870 					unsigned int msize = msgdsize(mp);
871 
872 					freemsg(mp);
873 					mutex_enter(&so->so_lock);
874 					so->so_rcv_queued -= msize;
875 					/*
876 					 * so_check_flow_control() will drop
877 					 * so->so_lock.
878 					 */
879 					so_check_flow_control(so);
880 				}
881 			} else if (partial_read && !somsghasdata(mp)) {
882 				/*
883 				 * Avoid queuing a zero-length tail part of
884 				 * a message. partial_read == 1 indicates that
885 				 * we read some of the message.
886 				 */
887 				freemsg(mp);
888 				more &= ~MOREDATA;
889 			} else {
890 				if (savemp != NULL &&
891 				    (flags & MSG_DUPCTRL)) {
892 					mblk_t *nmp;
893 					/*
894 					 * There should only be non data mblks
895 					 */
896 					ASSERT(DB_TYPE(savemp) != M_DATA &&
897 					    DB_TYPE(savemptail) != M_DATA);
898 try_again:
899 					if ((nmp = dupmsg(savemp)) == NULL &&
900 					    (nmp = copymsg(savemp)) == NULL) {
901 
902 						size_t size = msgsize(savemp);
903 
904 						error = strwaitbuf(size,
905 						    BPRI_HI);
906 						if (error != 0) {
907 							/*
908 							 * In case we
909 							 * cannot copy
910 							 * control data
911 							 * free the remaining
912 							 * data.
913 							 */
914 							freemsg(mp);
915 							goto done;
916 						}
917 						goto try_again;
918 					}
919 
920 					ASSERT(nmp != NULL);
921 					ASSERT(DB_TYPE(nmp) != M_DATA);
922 					savemptail->b_cont = mp;
923 					*mctlp = nmp;
924 					mp = savemp;
925 				}
926 				/*
927 				 * putback mp
928 				 */
929 				so_prepend_msg(so, mp, last_tail);
930 			}
931 		}
932 
933 		/* fast check so_rcv_head if there is more data */
934 		if (partial_read && !(so->so_state & SS_RCVATMARK) &&
935 		    *mctlp == NULL && uiop->uio_resid > 0 &&
936 		    !(flags & MSG_PEEK) && so->so_rcv_head != NULL) {
937 			goto again;
938 		}
939 	} else if (!partial_read) {
940 		mutex_enter(&so->so_lock);
941 		if (so->so_error != 0) {
942 			error = sogeterr(so, !(flags & MSG_PEEK));
943 			mutex_exit(&so->so_lock);
944 			return (error);
945 		}
946 		/*
947 		 * No pending data. Return right away for nonblocking
948 		 * socket, otherwise sleep waiting for data.
949 		 */
950 		if (!(so->so_state & SS_CANTRCVMORE) && uiop->uio_resid > 0) {
951 			if ((uiop->uio_fmode & (FNDELAY|FNONBLOCK)) ||
952 			    (flags & MSG_DONTWAIT)) {
953 				error = EWOULDBLOCK;
954 			} else {
955 				if (so->so_state & (SS_CLOSING |
956 				    SS_FALLBACK_PENDING)) {
957 					mutex_exit(&so->so_lock);
958 					error = EINTR;
959 					goto done;
960 				}
961 
962 				if (so->so_rcv_head != NULL) {
963 					goto again1;
964 				}
965 				so->so_rcv_wakeup = B_TRUE;
966 				so->so_rcv_wanted = uiop->uio_resid;
967 				if (so->so_rcvtimeo == 0) {
968 					/*
969 					 * Zero means disable timeout.
970 					 */
971 					error = cv_wait_sig(&so->so_rcv_cv,
972 					    &so->so_lock);
973 				} else {
974 					clock_t now;
975 					time_to_wait(&now, so->so_rcvtimeo);
976 					error = cv_timedwait_sig(&so->so_rcv_cv,
977 					    &so->so_lock, now);
978 				}
979 				so->so_rcv_wakeup = B_FALSE;
980 				so->so_rcv_wanted = 0;
981 
982 				if (error == 0) {
983 					error = EINTR;
984 				} else if (error == -1) {
985 					error = EAGAIN;
986 				} else {
987 					goto again1;
988 				}
989 			}
990 		}
991 		mutex_exit(&so->so_lock);
992 	}
993 	if (reset_atmark && partial_read && !(flags & MSG_PEEK)) {
994 		/*
995 		 * We are passed the mark, update state
996 		 * 4.3BSD and 4.4BSD clears the mark when peeking across it.
997 		 * The draft Posix socket spec states that the mark should
998 		 * not be cleared when peeking. We follow the latter.
999 		 */
1000 		mutex_enter(&so->so_lock);
1001 		ASSERT(so_verify_oobstate(so));
1002 		so->so_state &= ~(SS_OOBPEND|SS_HAVEOOBDATA|SS_RCVATMARK);
1003 		freemsg(so->so_oobmsg);
1004 		so->so_oobmsg = NULL;
1005 		ASSERT(so_verify_oobstate(so));
1006 		mutex_exit(&so->so_lock);
1007 	}
1008 	ASSERT(so->so_rcv_wakeup == B_FALSE);
1009 done:
1010 	if (sodp != NULL) {
1011 		mutex_enter(&so->so_lock);
1012 		if (sodp->sod_enabled &&
1013 		    (sodp->sod_uioa.uioa_state & UIOA_ENABLED)) {
1014 			SOD_UIOAFINI(sodp);
1015 			if (sodp->sod_uioa.uioa_mbytes > 0) {
1016 				ASSERT(so->so_rcv_q_head != NULL ||
1017 				    so->so_rcv_head != NULL);
1018 				so->so_rcv_queued -= sod_uioa_mblk(so, NULL);
1019 				if (error == EWOULDBLOCK)
1020 					error = 0;
1021 			}
1022 		}
1023 		mutex_exit(&so->so_lock);
1024 	}
1025 #ifdef DEBUG
1026 	if (so_debug_length) {
1027 		mutex_enter(&so->so_lock);
1028 		ASSERT(so_check_length(so));
1029 		mutex_exit(&so->so_lock);
1030 	}
1031 #endif
1032 	rvalp->r_val1 = more;
1033 	ASSERT(MUTEX_NOT_HELD(&so->so_lock));
1034 	return (error);
1035 }
1036 
1037 /*
1038  * Enqueue data from the protocol on the socket's rcv queue.
1039  *
1040  * We try to hook new M_DATA mblks onto an existing chain, however,
1041  * that cannot be done if the existing chain has already been
1042  * processed by I/OAT. Non-M_DATA mblks are just linked together via
1043  * b_next. In all cases the b_prev of the enqueued mblk is set to
1044  * point to the last mblk in its b_cont chain.
1045  */
1046 void
1047 so_enqueue_msg(struct sonode *so, mblk_t *mp, size_t msg_size)
1048 {
1049 	ASSERT(MUTEX_HELD(&so->so_lock));
1050 
1051 #ifdef DEBUG
1052 	if (so_debug_length) {
1053 		ASSERT(so_check_length(so));
1054 	}
1055 #endif
1056 	so->so_rcv_queued += msg_size;
1057 
1058 	if (so->so_rcv_head == NULL) {
1059 		ASSERT(so->so_rcv_last_head == NULL);
1060 		so->so_rcv_head = mp;
1061 		so->so_rcv_last_head = mp;
1062 	} else if ((DB_TYPE(mp) == M_DATA &&
1063 	    DB_TYPE(so->so_rcv_last_head) == M_DATA) &&
1064 	    ((DB_FLAGS(mp) & DBLK_UIOA) ==
1065 	    (DB_FLAGS(so->so_rcv_last_head) & DBLK_UIOA))) {
1066 		/* Added to the end */
1067 		ASSERT(so->so_rcv_last_head != NULL);
1068 		ASSERT(so->so_rcv_last_head->b_prev != NULL);
1069 		so->so_rcv_last_head->b_prev->b_cont = mp;
1070 	} else {
1071 		/* Start a new end */
1072 		so->so_rcv_last_head->b_next = mp;
1073 		so->so_rcv_last_head = mp;
1074 	}
1075 	while (mp->b_cont != NULL)
1076 		mp = mp->b_cont;
1077 
1078 	so->so_rcv_last_head->b_prev = mp;
1079 #ifdef DEBUG
1080 	if (so_debug_length) {
1081 		ASSERT(so_check_length(so));
1082 	}
1083 #endif
1084 }
1085 
1086 /*
1087  * Return B_TRUE if there is data in the message, B_FALSE otherwise.
1088  */
1089 boolean_t
1090 somsghasdata(mblk_t *mp)
1091 {
1092 	for (; mp; mp = mp->b_cont)
1093 		if (mp->b_datap->db_type == M_DATA) {
1094 			ASSERT(mp->b_wptr >= mp->b_rptr);
1095 			if (mp->b_wptr > mp->b_rptr)
1096 				return (B_TRUE);
1097 		}
1098 	return (B_FALSE);
1099 }
1100 
1101 /*
1102  * Flush the read side of sockfs.
1103  *
1104  * The caller must be sure that a reader is not already active when the
1105  * buffer is being flushed.
1106  */
1107 void
1108 so_rcv_flush(struct sonode *so)
1109 {
1110 	mblk_t  *mp;
1111 
1112 	ASSERT(MUTEX_HELD(&so->so_lock));
1113 
1114 	if (so->so_oobmsg != NULL) {
1115 		freemsg(so->so_oobmsg);
1116 		so->so_oobmsg = NULL;
1117 		so->so_oobmark = 0;
1118 		so->so_state &=
1119 		    ~(SS_OOBPEND|SS_HAVEOOBDATA|SS_HADOOBDATA|SS_RCVATMARK);
1120 	}
1121 
1122 	/*
1123 	 * Free messages sitting in the send and recv queue
1124 	 */
1125 	while (so->so_rcv_q_head != NULL) {
1126 		mp = so->so_rcv_q_head;
1127 		so->so_rcv_q_head = mp->b_next;
1128 		mp->b_next = mp->b_prev = NULL;
1129 		freemsg(mp);
1130 	}
1131 	while (so->so_rcv_head != NULL) {
1132 		mp = so->so_rcv_head;
1133 		so->so_rcv_head = mp->b_next;
1134 		mp->b_next = mp->b_prev = NULL;
1135 		freemsg(mp);
1136 	}
1137 	so->so_rcv_queued = 0;
1138 	so->so_rcv_q_head = NULL;
1139 	so->so_rcv_q_last_head = NULL;
1140 	so->so_rcv_head = NULL;
1141 	so->so_rcv_last_head = NULL;
1142 }
1143 
1144 /*
1145  * Handle recv* calls that set MSG_OOB or MSG_OOB together with MSG_PEEK.
1146  */
1147 int
1148 sorecvoob(struct sonode *so, struct nmsghdr *msg, struct uio *uiop, int flags,
1149     boolean_t oob_inline)
1150 {
1151 	mblk_t		*mp, *nmp;
1152 	int		error;
1153 
1154 	dprintso(so, 1, ("sorecvoob(%p, %p, 0x%x)\n", (void *)so, (void *)msg,
1155 	    flags));
1156 
1157 	if (msg != NULL) {
1158 		/*
1159 		 * There is never any oob data with addresses or control since
1160 		 * the T_EXDATA_IND does not carry any options.
1161 		 */
1162 		msg->msg_controllen = 0;
1163 		msg->msg_namelen = 0;
1164 		msg->msg_flags = 0;
1165 	}
1166 
1167 	mutex_enter(&so->so_lock);
1168 	ASSERT(so_verify_oobstate(so));
1169 	if (oob_inline ||
1170 	    (so->so_state & (SS_OOBPEND|SS_HADOOBDATA)) != SS_OOBPEND) {
1171 		dprintso(so, 1, ("sorecvoob: inline or data consumed\n"));
1172 		mutex_exit(&so->so_lock);
1173 		return (EINVAL);
1174 	}
1175 	if (!(so->so_state & SS_HAVEOOBDATA)) {
1176 		dprintso(so, 1, ("sorecvoob: no data yet\n"));
1177 		mutex_exit(&so->so_lock);
1178 		return (EWOULDBLOCK);
1179 	}
1180 	ASSERT(so->so_oobmsg != NULL);
1181 	mp = so->so_oobmsg;
1182 	if (flags & MSG_PEEK) {
1183 		/*
1184 		 * Since recv* can not return ENOBUFS we can not use dupmsg.
1185 		 * Instead we revert to the consolidation private
1186 		 * allocb_wait plus bcopy.
1187 		 */
1188 		mblk_t *mp1;
1189 
1190 		mp1 = allocb_wait(msgdsize(mp), BPRI_MED, STR_NOSIG, NULL);
1191 		ASSERT(mp1);
1192 
1193 		while (mp != NULL) {
1194 			ssize_t size;
1195 
1196 			size = MBLKL(mp);
1197 			bcopy(mp->b_rptr, mp1->b_wptr, size);
1198 			mp1->b_wptr += size;
1199 			ASSERT(mp1->b_wptr <= mp1->b_datap->db_lim);
1200 			mp = mp->b_cont;
1201 		}
1202 		mp = mp1;
1203 	} else {
1204 		/*
1205 		 * Update the state indicating that the data has been consumed.
1206 		 * Keep SS_OOBPEND set until data is consumed past the mark.
1207 		 */
1208 		so->so_oobmsg = NULL;
1209 		so->so_state ^= SS_HAVEOOBDATA|SS_HADOOBDATA;
1210 	}
1211 	ASSERT(so_verify_oobstate(so));
1212 	mutex_exit(&so->so_lock);
1213 
1214 	error = 0;
1215 	nmp = mp;
1216 	while (nmp != NULL && uiop->uio_resid > 0) {
1217 		ssize_t n = MBLKL(nmp);
1218 
1219 		n = MIN(n, uiop->uio_resid);
1220 		if (n > 0)
1221 			error = uiomove(nmp->b_rptr, n,
1222 			    UIO_READ, uiop);
1223 		if (error)
1224 			break;
1225 		nmp = nmp->b_cont;
1226 	}
1227 	ASSERT(mp->b_next == NULL && mp->b_prev == NULL);
1228 	freemsg(mp);
1229 	return (error);
1230 }
1231 
1232 /*
1233  * Allocate and initializ sonode
1234  */
1235 /* ARGSUSED */
1236 struct sonode *
1237 socket_sonode_create(struct sockparams *sp, int family, int type,
1238     int protocol, int version, int sflags, int *errorp, struct cred *cr)
1239 {
1240 	sonode_t *so;
1241 	int	kmflags;
1242 
1243 	/*
1244 	 * Choose the right set of sonodeops based on the upcall and
1245 	 * down call version that the protocol has provided
1246 	 */
1247 	if (SOCK_UC_VERSION != sp->sp_smod_info->smod_uc_version ||
1248 	    SOCK_DC_VERSION != sp->sp_smod_info->smod_dc_version) {
1249 		/*
1250 		 * mismatch
1251 		 */
1252 #ifdef DEBUG
1253 		cmn_err(CE_CONT, "protocol and socket module version mismatch");
1254 #endif
1255 		*errorp = EINVAL;
1256 		return (NULL);
1257 	}
1258 
1259 	kmflags = (sflags & SOCKET_NOSLEEP) ? KM_NOSLEEP : KM_SLEEP;
1260 
1261 	so = kmem_cache_alloc(socket_cache, kmflags);
1262 	if (so == NULL) {
1263 		*errorp = ENOMEM;
1264 		return (NULL);
1265 	}
1266 
1267 	sonode_init(so, sp, family, type, protocol, &so_sonodeops);
1268 
1269 	if (version == SOV_DEFAULT)
1270 		version = so_default_version;
1271 
1272 	so->so_version = (short)version;
1273 
1274 	/*
1275 	 * set the default values to be INFPSZ
1276 	 * if a protocol desires it can change the value later
1277 	 */
1278 	so->so_proto_props.sopp_rxhiwat = SOCKET_RECVHIWATER;
1279 	so->so_proto_props.sopp_rxlowat = SOCKET_RECVLOWATER;
1280 	so->so_proto_props.sopp_maxpsz = INFPSZ;
1281 	so->so_proto_props.sopp_maxblk = INFPSZ;
1282 
1283 	return (so);
1284 }
1285 
1286 int
1287 socket_init_common(struct sonode *so, struct sonode *pso, int flags, cred_t *cr)
1288 {
1289 	int error = 0;
1290 
1291 	if (pso != NULL) {
1292 		/*
1293 		 * We have a passive open, so inherit basic state from
1294 		 * the parent (listener).
1295 		 *
1296 		 * No need to grab the new sonode's lock, since there is no
1297 		 * one that can have a reference to it.
1298 		 */
1299 		mutex_enter(&pso->so_lock);
1300 
1301 		so->so_state |= SS_ISCONNECTED | (pso->so_state & SS_ASYNC);
1302 		so->so_pgrp = pso->so_pgrp;
1303 		so->so_rcvtimeo = pso->so_rcvtimeo;
1304 		so->so_sndtimeo = pso->so_sndtimeo;
1305 		so->so_xpg_rcvbuf = pso->so_xpg_rcvbuf;
1306 		/*
1307 		 * Make note of the socket level options. TCP and IP level
1308 		 * options are already inherited. We could do all this after
1309 		 * accept is successful but doing it here simplifies code and
1310 		 * no harm done for error case.
1311 		 */
1312 		so->so_options = pso->so_options & (SO_DEBUG|SO_REUSEADDR|
1313 		    SO_KEEPALIVE|SO_DONTROUTE|SO_BROADCAST|SO_USELOOPBACK|
1314 		    SO_OOBINLINE|SO_DGRAM_ERRIND|SO_LINGER);
1315 		so->so_proto_props = pso->so_proto_props;
1316 		so->so_mode = pso->so_mode;
1317 		so->so_pollev = pso->so_pollev & SO_POLLEV_ALWAYS;
1318 
1319 		mutex_exit(&pso->so_lock);
1320 	} else {
1321 		struct sockparams *sp = so->so_sockparams;
1322 		sock_upcalls_t *upcalls_to_use;
1323 
1324 		/*
1325 		 * Based on the version number select the right upcalls to
1326 		 * pass down. Currently we only have one version so choose
1327 		 * default
1328 		 */
1329 		upcalls_to_use = &so_upcalls;
1330 
1331 		/* active open, so create a lower handle */
1332 		so->so_proto_handle =
1333 		    sp->sp_smod_info->smod_proto_create_func(so->so_family,
1334 		    so->so_type, so->so_protocol, &so->so_downcalls,
1335 		    &so->so_mode, &error, flags, cr);
1336 
1337 		if (so->so_proto_handle == NULL) {
1338 			ASSERT(error != 0);
1339 			/*
1340 			 * To be safe; if a lower handle cannot be created, and
1341 			 * the proto does not give a reason why, assume there
1342 			 * was a lack of memory.
1343 			 */
1344 			return ((error == 0) ? ENOMEM : error);
1345 		}
1346 		ASSERT(so->so_downcalls != NULL);
1347 		ASSERT(so->so_downcalls->sd_send != NULL ||
1348 		    so->so_downcalls->sd_send_uio != NULL);
1349 		if (so->so_downcalls->sd_recv_uio != NULL) {
1350 			ASSERT(so->so_downcalls->sd_poll != NULL);
1351 			so->so_pollev |= SO_POLLEV_ALWAYS;
1352 		}
1353 
1354 		(*so->so_downcalls->sd_activate)(so->so_proto_handle,
1355 		    (sock_upper_handle_t)so, upcalls_to_use, 0, cr);
1356 
1357 		/* Wildcard */
1358 
1359 		/*
1360 		 * FIXME No need for this, the protocol can deal with it in
1361 		 * sd_create(). Should update ICMP.
1362 		 */
1363 		if (so->so_protocol != so->so_sockparams->sp_protocol) {
1364 			int protocol = so->so_protocol;
1365 			int error;
1366 			/*
1367 			 * Issue SO_PROTOTYPE setsockopt.
1368 			 */
1369 			error = socket_setsockopt(so, SOL_SOCKET, SO_PROTOTYPE,
1370 			    &protocol, (t_uscalar_t)sizeof (protocol), cr);
1371 			if (error) {
1372 				(void) (*so->so_downcalls->sd_close)
1373 				    (so->so_proto_handle, 0, cr);
1374 
1375 				mutex_enter(&so->so_lock);
1376 				so_rcv_flush(so);
1377 				mutex_exit(&so->so_lock);
1378 				/*
1379 				 * Setsockopt often fails with ENOPROTOOPT but
1380 				 * socket() should fail with
1381 				 * EPROTONOSUPPORT/EPROTOTYPE.
1382 				 */
1383 				return (EPROTONOSUPPORT);
1384 			}
1385 		}
1386 	}
1387 
1388 	if (uioasync.enabled)
1389 		sod_sock_init(so);
1390 
1391 	return (0);
1392 }
1393 
1394 /*
1395  * int socket_ioctl_common(struct sonode *so, int cmd, intptr_t arg, int mode,
1396  *         struct cred *cr, int32_t *rvalp)
1397  *
1398  * Handle ioctls that manipulate basic socket state; non-blocking,
1399  * async, etc.
1400  *
1401  * Returns:
1402  *   < 0  - ioctl was not handle
1403  *  >= 0  - ioctl was handled, if > 0, then it is an errno
1404  *
1405  * Notes:
1406  *   Assumes the standard receive buffer is used to obtain info for
1407  *   NREAD.
1408  */
1409 /* ARGSUSED */
1410 int
1411 socket_ioctl_common(struct sonode *so, int cmd, intptr_t arg, int mode,
1412     struct cred *cr, int32_t *rvalp)
1413 {
1414 	switch (cmd) {
1415 	case SIOCSQPTR:
1416 		/*
1417 		 * SIOCSQPTR is valid only when helper stream is created
1418 		 * by the protocol.
1419 		 */
1420 
1421 		return (EOPNOTSUPP);
1422 	case FIONBIO: {
1423 		int32_t value;
1424 
1425 		if (so_copyin((void *)arg, &value, sizeof (int32_t),
1426 		    (mode & (int)FKIOCTL)))
1427 			return (EFAULT);
1428 
1429 		mutex_enter(&so->so_lock);
1430 		if (value) {
1431 			so->so_state |= SS_NDELAY;
1432 		} else {
1433 			so->so_state &= ~SS_NDELAY;
1434 		}
1435 		mutex_exit(&so->so_lock);
1436 		return (0);
1437 	}
1438 	case FIOASYNC: {
1439 		int32_t value;
1440 
1441 		if (so_copyin((void *)arg, &value, sizeof (int32_t),
1442 		    (mode & (int)FKIOCTL)))
1443 			return (EFAULT);
1444 
1445 		mutex_enter(&so->so_lock);
1446 
1447 		if (value) {
1448 			/* Turn on SIGIO */
1449 			so->so_state |= SS_ASYNC;
1450 		} else {
1451 			/* Turn off SIGIO */
1452 			so->so_state &= ~SS_ASYNC;
1453 		}
1454 		mutex_exit(&so->so_lock);
1455 
1456 		return (0);
1457 	}
1458 
1459 	case SIOCSPGRP:
1460 	case FIOSETOWN: {
1461 		int error;
1462 		pid_t pid;
1463 
1464 		if (so_copyin((void *)arg, &pid, sizeof (pid_t),
1465 		    (mode & (int)FKIOCTL)))
1466 			return (EFAULT);
1467 
1468 		mutex_enter(&so->so_lock);
1469 		error = (pid != so->so_pgrp) ? socket_chgpgrp(so, pid) : 0;
1470 		mutex_exit(&so->so_lock);
1471 		return (error);
1472 	}
1473 	case SIOCGPGRP:
1474 	case FIOGETOWN:
1475 		if (so_copyout(&so->so_pgrp, (void *)arg,
1476 		    sizeof (pid_t), (mode & (int)FKIOCTL)))
1477 			return (EFAULT);
1478 
1479 		return (0);
1480 	case SIOCATMARK: {
1481 		int retval;
1482 
1483 		/*
1484 		 * Only protocols that support urgent data can handle ATMARK.
1485 		 */
1486 		if ((so->so_mode & SM_EXDATA) == 0)
1487 			return (EINVAL);
1488 
1489 		/*
1490 		 * If the protocol is maintaining its own buffer, then the
1491 		 * request must be passed down.
1492 		 */
1493 		if (so->so_downcalls->sd_recv_uio != NULL)
1494 			return (-1);
1495 
1496 		retval = (so->so_state & SS_RCVATMARK) != 0;
1497 
1498 		if (so_copyout(&retval, (void *)arg, sizeof (int),
1499 		    (mode & (int)FKIOCTL))) {
1500 			return (EFAULT);
1501 		}
1502 		return (0);
1503 	}
1504 
1505 	case FIONREAD: {
1506 		int retval;
1507 
1508 		/*
1509 		 * If the protocol is maintaining its own buffer, then the
1510 		 * request must be passed down.
1511 		 */
1512 		if (so->so_downcalls->sd_recv_uio != NULL)
1513 			return (-1);
1514 
1515 		retval = MIN(so->so_rcv_queued, INT_MAX);
1516 
1517 		if (so_copyout(&retval, (void *)arg,
1518 		    sizeof (retval), (mode & (int)FKIOCTL))) {
1519 			return (EFAULT);
1520 		}
1521 		return (0);
1522 	}
1523 
1524 	case _I_GETPEERCRED: {
1525 		int error = 0;
1526 
1527 		if ((mode & FKIOCTL) == 0)
1528 			return (EINVAL);
1529 
1530 		mutex_enter(&so->so_lock);
1531 		if ((so->so_mode & SM_CONNREQUIRED) == 0) {
1532 			error = ENOTSUP;
1533 		} else if ((so->so_state & SS_ISCONNECTED) == 0) {
1534 			error = ENOTCONN;
1535 		} else if (so->so_peercred != NULL) {
1536 			k_peercred_t *kp = (k_peercred_t *)arg;
1537 			kp->pc_cr = so->so_peercred;
1538 			kp->pc_cpid = so->so_cpid;
1539 			crhold(so->so_peercred);
1540 		} else {
1541 			error = EINVAL;
1542 		}
1543 		mutex_exit(&so->so_lock);
1544 		return (error);
1545 	}
1546 	default:
1547 		return (-1);
1548 	}
1549 }
1550 
1551 /*
1552  * Handle the I_NREAD STREAM ioctl.
1553  */
1554 static int
1555 so_strioc_nread(struct sonode *so, intptr_t arg, int mode, int32_t *rvalp)
1556 {
1557 	size_t size = 0;
1558 	int retval;
1559 	int count = 0;
1560 	mblk_t *mp;
1561 
1562 	if (so->so_downcalls == NULL ||
1563 	    so->so_downcalls->sd_recv_uio != NULL)
1564 		return (EINVAL);
1565 
1566 	mutex_enter(&so->so_lock);
1567 	/* Wait for reader to get out of the way. */
1568 	while (so->so_flag & SOREADLOCKED) {
1569 		/*
1570 		 * If reader is waiting for data, then there should be nothing
1571 		 * on the rcv queue.
1572 		 */
1573 		if (so->so_rcv_wakeup)
1574 			goto out;
1575 
1576 		so->so_flag |= SOWANT;
1577 		/* Do a timed sleep, in case the reader goes to sleep. */
1578 		(void) cv_timedwait(&so->so_state_cv, &so->so_lock,
1579 		    lbolt + drv_usectohz(10));
1580 	}
1581 
1582 	/*
1583 	 * Since we are holding so_lock no new reader will come in, and the
1584 	 * protocol will not be able to enqueue data. So it's safe to walk
1585 	 * both rcv queues.
1586 	 */
1587 	mp = so->so_rcv_q_head;
1588 	if (mp != NULL) {
1589 		size = msgdsize(so->so_rcv_q_head);
1590 		for (; mp != NULL; mp = mp->b_next)
1591 			count++;
1592 	} else {
1593 		/*
1594 		 * In case the processing list was empty, get the size of the
1595 		 * next msg in line.
1596 		 */
1597 		size = msgdsize(so->so_rcv_head);
1598 	}
1599 
1600 	for (mp = so->so_rcv_head; mp != NULL; mp = mp->b_next)
1601 		count++;
1602 out:
1603 	mutex_exit(&so->so_lock);
1604 
1605 	/*
1606 	 * Drop down from size_t to the "int" required by the
1607 	 * interface.  Cap at INT_MAX.
1608 	 */
1609 	retval = MIN(size, INT_MAX);
1610 	if (so_copyout(&retval, (void *)arg, sizeof (retval),
1611 	    (mode & (int)FKIOCTL))) {
1612 		return (EFAULT);
1613 	} else {
1614 		*rvalp = count;
1615 		return (0);
1616 	}
1617 }
1618 
1619 /*
1620  * Process STREAM ioctls.
1621  *
1622  * Returns:
1623  *   < 0  - ioctl was not handle
1624  *  >= 0  - ioctl was handled, if > 0, then it is an errno
1625  */
1626 int
1627 socket_strioc_common(struct sonode *so, int cmd, intptr_t arg, int mode,
1628     struct cred *cr, int32_t *rvalp)
1629 {
1630 	int retval;
1631 
1632 	/* Only STREAM iotcls are handled here */
1633 	if ((cmd & 0xffffff00U) != STR)
1634 		return (-1);
1635 
1636 	switch (cmd) {
1637 	case I_CANPUT:
1638 		/*
1639 		 * We return an error for I_CANPUT so that isastream(3C) will
1640 		 * not report the socket as being a STREAM.
1641 		 */
1642 		return (EOPNOTSUPP);
1643 	case I_NREAD:
1644 		/* Avoid doing a fallback for I_NREAD. */
1645 		return (so_strioc_nread(so, arg, mode, rvalp));
1646 	case I_LOOK:
1647 		/* Avoid doing a fallback for I_LOOK. */
1648 		if (so_copyout("sockmod", (void *)arg, strlen("sockmod") + 1,
1649 		    (mode & (int)FKIOCTL))) {
1650 			return (EFAULT);
1651 		}
1652 		return (0);
1653 	default:
1654 		break;
1655 	}
1656 
1657 	/*
1658 	 * Try to fall back to TPI, and if successful, reissue the ioctl.
1659 	 */
1660 	if ((retval = so_tpi_fallback(so, cr)) == 0) {
1661 		/* Reissue the ioctl */
1662 		ASSERT(so->so_rcv_q_head == NULL);
1663 		return (SOP_IOCTL(so, cmd, arg, mode, cr, rvalp));
1664 	} else {
1665 		return (retval);
1666 	}
1667 }
1668 
1669 /*
1670  * This is called for all socket types to verify that the buffer size is large
1671  * enough for the option, and if we can, handle the request as well. Most
1672  * options will be forwarded to the protocol.
1673  */
1674 int
1675 socket_getopt_common(struct sonode *so, int level, int option_name,
1676     void *optval, socklen_t *optlenp, int flags)
1677 {
1678 	if (level != SOL_SOCKET)
1679 		return (-1);
1680 
1681 	switch (option_name) {
1682 	case SO_ERROR:
1683 	case SO_DOMAIN:
1684 	case SO_TYPE:
1685 	case SO_ACCEPTCONN: {
1686 		int32_t value;
1687 		socklen_t optlen = *optlenp;
1688 
1689 		if (optlen < (t_uscalar_t)sizeof (int32_t)) {
1690 			return (EINVAL);
1691 		}
1692 
1693 		switch (option_name) {
1694 		case SO_ERROR:
1695 			mutex_enter(&so->so_lock);
1696 			value = sogeterr(so, B_TRUE);
1697 			mutex_exit(&so->so_lock);
1698 			break;
1699 		case SO_DOMAIN:
1700 			value = so->so_family;
1701 			break;
1702 		case SO_TYPE:
1703 			value = so->so_type;
1704 			break;
1705 		case SO_ACCEPTCONN:
1706 			if (so->so_state & SS_ACCEPTCONN)
1707 				value = SO_ACCEPTCONN;
1708 			else
1709 				value = 0;
1710 			break;
1711 		}
1712 
1713 		bcopy(&value, optval, sizeof (value));
1714 		*optlenp = sizeof (value);
1715 
1716 		return (0);
1717 	}
1718 	case SO_SNDTIMEO:
1719 	case SO_RCVTIMEO: {
1720 		clock_t value;
1721 		socklen_t optlen = *optlenp;
1722 
1723 		if (get_udatamodel() == DATAMODEL_NONE ||
1724 		    get_udatamodel() == DATAMODEL_NATIVE) {
1725 			if (optlen < sizeof (struct timeval))
1726 				return (EINVAL);
1727 		} else {
1728 			if (optlen < sizeof (struct timeval32))
1729 				return (EINVAL);
1730 		}
1731 		if (option_name == SO_RCVTIMEO)
1732 			value = drv_hztousec(so->so_rcvtimeo);
1733 		else
1734 			value = drv_hztousec(so->so_sndtimeo);
1735 
1736 		if (get_udatamodel() == DATAMODEL_NONE ||
1737 		    get_udatamodel() == DATAMODEL_NATIVE) {
1738 			((struct timeval *)(optval))->tv_sec =
1739 			    value / (1000 * 1000);
1740 			((struct timeval *)(optval))->tv_usec =
1741 			    value % (1000 * 1000);
1742 			*optlenp = sizeof (struct timeval);
1743 		} else {
1744 			((struct timeval32 *)(optval))->tv_sec =
1745 			    value / (1000 * 1000);
1746 			((struct timeval32 *)(optval))->tv_usec =
1747 			    value % (1000 * 1000);
1748 			*optlenp = sizeof (struct timeval32);
1749 		}
1750 		return (0);
1751 	}
1752 	case SO_DEBUG:
1753 	case SO_REUSEADDR:
1754 	case SO_KEEPALIVE:
1755 	case SO_DONTROUTE:
1756 	case SO_BROADCAST:
1757 	case SO_USELOOPBACK:
1758 	case SO_OOBINLINE:
1759 	case SO_SNDBUF:
1760 #ifdef notyet
1761 	case SO_SNDLOWAT:
1762 	case SO_RCVLOWAT:
1763 #endif /* notyet */
1764 	case SO_DGRAM_ERRIND: {
1765 		socklen_t optlen = *optlenp;
1766 
1767 		if (optlen < (t_uscalar_t)sizeof (int32_t))
1768 			return (EINVAL);
1769 		break;
1770 	}
1771 	case SO_RCVBUF: {
1772 		socklen_t optlen = *optlenp;
1773 
1774 		if (optlen < (t_uscalar_t)sizeof (int32_t))
1775 			return (EINVAL);
1776 
1777 		if ((flags & _SOGETSOCKOPT_XPG4_2) && so->so_xpg_rcvbuf != 0) {
1778 			/*
1779 			 * XXX If SO_RCVBUF has been set and this is an
1780 			 * XPG 4.2 application then do not ask the transport
1781 			 * since the transport might adjust the value and not
1782 			 * return exactly what was set by the application.
1783 			 * For non-XPG 4.2 application we return the value
1784 			 * that the transport is actually using.
1785 			 */
1786 			*(int32_t *)optval = so->so_xpg_rcvbuf;
1787 			*optlenp = sizeof (so->so_xpg_rcvbuf);
1788 			return (0);
1789 		}
1790 		/*
1791 		 * If the option has not been set then get a default
1792 		 * value from the transport.
1793 		 */
1794 		break;
1795 	}
1796 	case SO_LINGER: {
1797 		socklen_t optlen = *optlenp;
1798 
1799 		if (optlen < (t_uscalar_t)sizeof (struct linger))
1800 			return (EINVAL);
1801 		break;
1802 	}
1803 	case SO_SND_BUFINFO: {
1804 		socklen_t optlen = *optlenp;
1805 
1806 		if (optlen < (t_uscalar_t)sizeof (struct so_snd_bufinfo))
1807 			return (EINVAL);
1808 		((struct so_snd_bufinfo *)(optval))->sbi_wroff =
1809 		    (so->so_proto_props).sopp_wroff;
1810 		((struct so_snd_bufinfo *)(optval))->sbi_maxblk =
1811 		    (so->so_proto_props).sopp_maxblk;
1812 		((struct so_snd_bufinfo *)(optval))->sbi_maxpsz =
1813 		    (so->so_proto_props).sopp_maxpsz;
1814 		((struct so_snd_bufinfo *)(optval))->sbi_tail =
1815 		    (so->so_proto_props).sopp_tail;
1816 		*optlenp = sizeof (struct so_snd_bufinfo);
1817 		return (0);
1818 	}
1819 	default:
1820 		break;
1821 	}
1822 
1823 	/* Unknown Option */
1824 	return (-1);
1825 }
1826 
1827 void
1828 socket_sonode_destroy(struct sonode *so)
1829 {
1830 	sonode_fini(so);
1831 	kmem_cache_free(socket_cache, so);
1832 }
1833 
1834 int
1835 so_zcopy_wait(struct sonode *so)
1836 {
1837 	int error = 0;
1838 
1839 	mutex_enter(&so->so_lock);
1840 	while (!(so->so_copyflag & STZCNOTIFY)) {
1841 		if (so->so_state & SS_CLOSING) {
1842 			mutex_exit(&so->so_lock);
1843 			return (EINTR);
1844 		}
1845 		if (cv_wait_sig(&so->so_copy_cv, &so->so_lock) == 0) {
1846 			error = EINTR;
1847 			break;
1848 		}
1849 	}
1850 	so->so_copyflag &= ~STZCNOTIFY;
1851 	mutex_exit(&so->so_lock);
1852 	return (error);
1853 }
1854 
1855 void
1856 so_timer_callback(void *arg)
1857 {
1858 	struct sonode *so = (struct sonode *)arg;
1859 
1860 	mutex_enter(&so->so_lock);
1861 
1862 	so->so_rcv_timer_tid = 0;
1863 	if (so->so_rcv_queued > 0) {
1864 		so_notify_data(so, so->so_rcv_queued);
1865 	} else {
1866 		mutex_exit(&so->so_lock);
1867 	}
1868 }
1869 
1870 #ifdef DEBUG
1871 /*
1872  * Verify that the length stored in so_rcv_queued and the length of data blocks
1873  * queued is same.
1874  */
1875 static boolean_t
1876 so_check_length(sonode_t *so)
1877 {
1878 	mblk_t *mp = so->so_rcv_q_head;
1879 	int len = 0;
1880 
1881 	ASSERT(MUTEX_HELD(&so->so_lock));
1882 
1883 	if (mp != NULL) {
1884 		len = msgdsize(mp);
1885 		while ((mp = mp->b_next) != NULL)
1886 			len += msgdsize(mp);
1887 	}
1888 	mp = so->so_rcv_head;
1889 	if (mp != NULL) {
1890 		len += msgdsize(mp);
1891 		while ((mp = mp->b_next) != NULL)
1892 			len += msgdsize(mp);
1893 	}
1894 	return ((len == so->so_rcv_queued) ? B_TRUE : B_FALSE);
1895 }
1896 #endif
1897 
1898 int
1899 so_get_mod_version(struct sockparams *sp)
1900 {
1901 	ASSERT(sp != NULL && sp->sp_smod_info != NULL);
1902 	return (sp->sp_smod_info->smod_version);
1903 }
1904 
1905 /*
1906  * so_start_fallback()
1907  *
1908  * Block new socket operations from coming in, and wait for active operations
1909  * to complete. Threads that are sleeping will be woken up so they can get
1910  * out of the way.
1911  *
1912  * The caller must be a reader on so_fallback_rwlock.
1913  */
1914 static boolean_t
1915 so_start_fallback(struct sonode *so)
1916 {
1917 	ASSERT(RW_READ_HELD(&so->so_fallback_rwlock));
1918 
1919 	mutex_enter(&so->so_lock);
1920 	if (so->so_state & SS_FALLBACK_PENDING) {
1921 		mutex_exit(&so->so_lock);
1922 		return (B_FALSE);
1923 	}
1924 	so->so_state |= SS_FALLBACK_PENDING;
1925 	/*
1926 	 * Poke all threads that might be sleeping. Any operation that comes
1927 	 * in after the cv_broadcast will observe the fallback pending flag
1928 	 * which cause the call to return where it would normally sleep.
1929 	 */
1930 	cv_broadcast(&so->so_state_cv);		/* threads in connect() */
1931 	cv_broadcast(&so->so_rcv_cv);		/* threads in recvmsg() */
1932 	cv_broadcast(&so->so_snd_cv);		/* threads in sendmsg() */
1933 	mutex_enter(&so->so_acceptq_lock);
1934 	cv_broadcast(&so->so_acceptq_cv);	/* threads in accept() */
1935 	mutex_exit(&so->so_acceptq_lock);
1936 	mutex_exit(&so->so_lock);
1937 
1938 	/*
1939 	 * The main reason for the rw_tryupgrade call is to provide
1940 	 * observability during the fallback process. We want to
1941 	 * be able to see if there are pending operations.
1942 	 */
1943 	if (rw_tryupgrade(&so->so_fallback_rwlock) == 0) {
1944 		/*
1945 		 * It is safe to drop and reaquire the fallback lock, because
1946 		 * we are guaranteed that another fallback cannot take place.
1947 		 */
1948 		rw_exit(&so->so_fallback_rwlock);
1949 		DTRACE_PROBE1(pending__ops__wait, (struct sonode *), so);
1950 		rw_enter(&so->so_fallback_rwlock, RW_WRITER);
1951 		DTRACE_PROBE1(pending__ops__complete, (struct sonode *), so);
1952 	}
1953 
1954 	return (B_TRUE);
1955 }
1956 
1957 /*
1958  * so_end_fallback()
1959  *
1960  * Allow socket opertions back in.
1961  *
1962  * The caller must be a writer on so_fallback_rwlock.
1963  */
1964 static void
1965 so_end_fallback(struct sonode *so)
1966 {
1967 	ASSERT(RW_ISWRITER(&so->so_fallback_rwlock));
1968 
1969 	mutex_enter(&so->so_lock);
1970 	so->so_state &= ~(SS_FALLBACK_PENDING|SS_FALLBACK_DRAIN);
1971 	mutex_exit(&so->so_lock);
1972 
1973 	rw_downgrade(&so->so_fallback_rwlock);
1974 }
1975 
1976 /*
1977  * so_quiesced_cb()
1978  *
1979  * Callback passed to the protocol during fallback. It is called once
1980  * the endpoint is quiescent.
1981  *
1982  * No requests from the user, no notifications from the protocol, so it
1983  * is safe to synchronize the state. Data can also be moved without
1984  * risk for reordering.
1985  *
1986  * We do not need to hold so_lock, since there can be only one thread
1987  * operating on the sonode.
1988  */
1989 static void
1990 so_quiesced_cb(sock_upper_handle_t sock_handle, queue_t *q,
1991     struct T_capability_ack *tcap, struct sockaddr *laddr, socklen_t laddrlen,
1992     struct sockaddr *faddr, socklen_t faddrlen, short opts)
1993 {
1994 	struct sonode *so = (struct sonode *)sock_handle;
1995 	boolean_t atmark;
1996 
1997 	sotpi_update_state(so, tcap, laddr, laddrlen, faddr, faddrlen, opts);
1998 
1999 	/*
2000 	 * Some protocols do not quiece the data path during fallback. Once
2001 	 * we set the SS_FALLBACK_DRAIN flag any attempt to queue data will
2002 	 * fail and the protocol is responsible for saving the data for later
2003 	 * delivery (i.e., once the fallback has completed).
2004 	 */
2005 	mutex_enter(&so->so_lock);
2006 	so->so_state |= SS_FALLBACK_DRAIN;
2007 	SOCKET_TIMER_CANCEL(so);
2008 	mutex_exit(&so->so_lock);
2009 
2010 	if (so->so_rcv_head != NULL) {
2011 		if (so->so_rcv_q_last_head == NULL)
2012 			so->so_rcv_q_head = so->so_rcv_head;
2013 		else
2014 			so->so_rcv_q_last_head->b_next = so->so_rcv_head;
2015 		so->so_rcv_q_last_head = so->so_rcv_last_head;
2016 	}
2017 
2018 	atmark = (so->so_state & SS_RCVATMARK) != 0;
2019 	/*
2020 	 * Clear any OOB state having to do with pending data. The TPI
2021 	 * code path will set the appropriate oob state when we move the
2022 	 * oob data to the STREAM head. We leave SS_HADOOBDATA since the oob
2023 	 * data has already been consumed.
2024 	 */
2025 	so->so_state &= ~(SS_RCVATMARK|SS_OOBPEND|SS_HAVEOOBDATA);
2026 
2027 	ASSERT(so->so_oobmsg != NULL || so->so_oobmark <= so->so_rcv_queued);
2028 
2029 	/*
2030 	 * Move data to the STREAM head.
2031 	 */
2032 	while (so->so_rcv_q_head != NULL) {
2033 		mblk_t *mp = so->so_rcv_q_head;
2034 		size_t mlen = msgdsize(mp);
2035 
2036 		so->so_rcv_q_head = mp->b_next;
2037 		mp->b_next = NULL;
2038 		mp->b_prev = NULL;
2039 
2040 		/*
2041 		 * Send T_EXDATA_IND if we are at the oob mark.
2042 		 */
2043 		if (atmark) {
2044 			struct T_exdata_ind *tei;
2045 			mblk_t *mp1 = SOTOTPI(so)->sti_exdata_mp;
2046 
2047 			SOTOTPI(so)->sti_exdata_mp = NULL;
2048 			ASSERT(mp1 != NULL);
2049 			mp1->b_datap->db_type = M_PROTO;
2050 			tei = (struct T_exdata_ind *)mp1->b_rptr;
2051 			tei->PRIM_type = T_EXDATA_IND;
2052 			tei->MORE_flag = 0;
2053 			mp1->b_wptr = (uchar_t *)&tei[1];
2054 
2055 			if (IS_SO_OOB_INLINE(so)) {
2056 				mp1->b_cont = mp;
2057 			} else {
2058 				ASSERT(so->so_oobmsg != NULL);
2059 				mp1->b_cont = so->so_oobmsg;
2060 				so->so_oobmsg = NULL;
2061 
2062 				/* process current mp next time around */
2063 				mp->b_next = so->so_rcv_q_head;
2064 				so->so_rcv_q_head = mp;
2065 				mlen = 0;
2066 			}
2067 			mp = mp1;
2068 
2069 			/* we have consumed the oob mark */
2070 			atmark = B_FALSE;
2071 		} else if (so->so_oobmark > 0) {
2072 			/*
2073 			 * Check if the OOB mark is within the current
2074 			 * mblk chain. In that case we have to split it up.
2075 			 */
2076 			if (so->so_oobmark < mlen) {
2077 				mblk_t *urg_mp = mp;
2078 
2079 				atmark = B_TRUE;
2080 				mp = NULL;
2081 				mlen = so->so_oobmark;
2082 
2083 				/*
2084 				 * It is assumed that the OOB mark does
2085 				 * not land within a mblk.
2086 				 */
2087 				do {
2088 					so->so_oobmark -= MBLKL(urg_mp);
2089 					mp = urg_mp;
2090 					urg_mp = urg_mp->b_cont;
2091 				} while (so->so_oobmark > 0);
2092 				mp->b_cont = NULL;
2093 				if (urg_mp != NULL) {
2094 					urg_mp->b_next = so->so_rcv_q_head;
2095 					so->so_rcv_q_head = urg_mp;
2096 				}
2097 			} else {
2098 				so->so_oobmark -= mlen;
2099 				if (so->so_oobmark == 0)
2100 					atmark = B_TRUE;
2101 			}
2102 		}
2103 
2104 		/*
2105 		 * Queue data on the STREAM head.
2106 		 */
2107 		so->so_rcv_queued -= mlen;
2108 		putnext(q, mp);
2109 	}
2110 	so->so_rcv_head = NULL;
2111 	so->so_rcv_last_head = NULL;
2112 	so->so_rcv_q_head = NULL;
2113 	so->so_rcv_q_last_head = NULL;
2114 
2115 	/*
2116 	 * Check if the oob byte is at the end of the data stream, or if the
2117 	 * oob byte has not yet arrived. In the latter case we have to send a
2118 	 * SIGURG and a mark indicator to the STREAM head. The mark indicator
2119 	 * is needed to guarantee correct behavior for SIOCATMARK. See block
2120 	 * comment in socktpi.h for more details.
2121 	 */
2122 	if (atmark || so->so_oobmark > 0) {
2123 		mblk_t *mp;
2124 
2125 		if (atmark && so->so_oobmsg != NULL) {
2126 			struct T_exdata_ind *tei;
2127 
2128 			mp = SOTOTPI(so)->sti_exdata_mp;
2129 			SOTOTPI(so)->sti_exdata_mp = NULL;
2130 			ASSERT(mp != NULL);
2131 			mp->b_datap->db_type = M_PROTO;
2132 			tei = (struct T_exdata_ind *)mp->b_rptr;
2133 			tei->PRIM_type = T_EXDATA_IND;
2134 			tei->MORE_flag = 0;
2135 			mp->b_wptr = (uchar_t *)&tei[1];
2136 
2137 			mp->b_cont = so->so_oobmsg;
2138 			so->so_oobmsg = NULL;
2139 
2140 			putnext(q, mp);
2141 		} else {
2142 			/* Send up the signal */
2143 			mp = SOTOTPI(so)->sti_exdata_mp;
2144 			SOTOTPI(so)->sti_exdata_mp = NULL;
2145 			ASSERT(mp != NULL);
2146 			DB_TYPE(mp) = M_PCSIG;
2147 			*mp->b_wptr++ = (uchar_t)SIGURG;
2148 			putnext(q, mp);
2149 
2150 			/* Send up the mark indicator */
2151 			mp = SOTOTPI(so)->sti_urgmark_mp;
2152 			SOTOTPI(so)->sti_urgmark_mp = NULL;
2153 			mp->b_flag = atmark ? MSGMARKNEXT : MSGNOTMARKNEXT;
2154 			putnext(q, mp);
2155 
2156 			so->so_oobmark = 0;
2157 		}
2158 	}
2159 
2160 	if (SOTOTPI(so)->sti_exdata_mp != NULL) {
2161 		freeb(SOTOTPI(so)->sti_exdata_mp);
2162 		SOTOTPI(so)->sti_exdata_mp = NULL;
2163 	}
2164 
2165 	if (SOTOTPI(so)->sti_urgmark_mp != NULL) {
2166 		freeb(SOTOTPI(so)->sti_urgmark_mp);
2167 		SOTOTPI(so)->sti_urgmark_mp = NULL;
2168 	}
2169 
2170 	ASSERT(so->so_oobmark == 0);
2171 	ASSERT(so->so_rcv_queued == 0);
2172 }
2173 
2174 #ifdef DEBUG
2175 /*
2176  * Do an integrity check of the sonode. This should be done if a
2177  * fallback fails after sonode has initially been converted to use
2178  * TPI and subsequently have to be reverted.
2179  *
2180  * Failure to pass the integrity check will panic the system.
2181  */
2182 void
2183 so_integrity_check(struct sonode *cur, struct sonode *orig)
2184 {
2185 	VERIFY(cur->so_vnode == orig->so_vnode);
2186 	VERIFY(cur->so_ops == orig->so_ops);
2187 	/*
2188 	 * For so_state we can only VERIFY the state flags in CHECK_STATE.
2189 	 * The other state flags might be affected by a notification from the
2190 	 * protocol.
2191 	 */
2192 #define	CHECK_STATE	(SS_CANTRCVMORE|SS_CANTSENDMORE|SS_NDELAY|SS_NONBLOCK| \
2193 	SS_ASYNC|SS_ACCEPTCONN|SS_SAVEDEOR|SS_RCVATMARK|SS_OOBPEND| \
2194 	SS_HAVEOOBDATA|SS_HADOOBDATA|SS_SENTLASTREADSIG|SS_SENTLASTWRITESIG)
2195 	VERIFY((cur->so_state & (orig->so_state & CHECK_STATE)) ==
2196 	    (orig->so_state & CHECK_STATE));
2197 	VERIFY(cur->so_mode == orig->so_mode);
2198 	VERIFY(cur->so_flag == orig->so_flag);
2199 	VERIFY(cur->so_count == orig->so_count);
2200 	/* Cannot VERIFY so_proto_connid; proto can update it */
2201 	VERIFY(cur->so_sockparams == orig->so_sockparams);
2202 	/* an error might have been recorded, but it can not be lost */
2203 	VERIFY(cur->so_error != 0 || orig->so_error == 0);
2204 	VERIFY(cur->so_family == orig->so_family);
2205 	VERIFY(cur->so_type == orig->so_type);
2206 	VERIFY(cur->so_protocol == orig->so_protocol);
2207 	VERIFY(cur->so_version == orig->so_version);
2208 	/* New conns might have arrived, but none should have been lost */
2209 	VERIFY(cur->so_acceptq_len >= orig->so_acceptq_len);
2210 	VERIFY(cur->so_acceptq_head == orig->so_acceptq_head);
2211 	VERIFY(cur->so_backlog == orig->so_backlog);
2212 	/* New OOB migth have arrived, but mark should not have been lost */
2213 	VERIFY(cur->so_oobmark >= orig->so_oobmark);
2214 	/* Cannot VERIFY so_oobmsg; the proto might have sent up a new one */
2215 	VERIFY(cur->so_pgrp == orig->so_pgrp);
2216 	VERIFY(cur->so_peercred == orig->so_peercred);
2217 	VERIFY(cur->so_cpid == orig->so_cpid);
2218 	VERIFY(cur->so_zoneid == orig->so_zoneid);
2219 	/* New data migth have arrived, but none should have been lost */
2220 	VERIFY(cur->so_rcv_queued >= orig->so_rcv_queued);
2221 	VERIFY(cur->so_rcv_q_head == orig->so_rcv_q_head);
2222 	VERIFY(cur->so_rcv_head == orig->so_rcv_head);
2223 	VERIFY(cur->so_proto_handle == orig->so_proto_handle);
2224 	VERIFY(cur->so_downcalls == orig->so_downcalls);
2225 	/* Cannot VERIFY so_proto_props; they can be updated by proto */
2226 }
2227 #endif
2228 
2229 /*
2230  * so_tpi_fallback()
2231  *
2232  * This is the fallback initation routine; things start here.
2233  *
2234  * Basic strategy:
2235  *   o Block new socket operations from coming in
2236  *   o Allocate/initate info needed by TPI
2237  *   o Quiesce the connection, at which point we sync
2238  *     state and move data
2239  *   o Change operations (sonodeops) associated with the socket
2240  *   o Unblock threads waiting for the fallback to finish
2241  */
2242 int
2243 so_tpi_fallback(struct sonode *so, struct cred *cr)
2244 {
2245 	int error;
2246 	queue_t *q;
2247 	struct sockparams *sp;
2248 	struct sockparams *newsp = NULL;
2249 	so_proto_fallback_func_t fbfunc;
2250 	boolean_t direct;
2251 	struct sonode *nso;
2252 #ifdef DEBUG
2253 	struct sonode origso;
2254 #endif
2255 	error = 0;
2256 	sp = so->so_sockparams;
2257 	fbfunc = sp->sp_smod_info->smod_proto_fallback_func;
2258 
2259 	/*
2260 	 * Fallback can only happen if there is a device associated
2261 	 * with the sonode, and the socket module has a fallback function.
2262 	 */
2263 	if (!SOCKPARAMS_HAS_DEVICE(sp) || fbfunc == NULL)
2264 		return (EINVAL);
2265 
2266 	/*
2267 	 * Initiate fallback; upon success we know that no new requests
2268 	 * will come in from the user.
2269 	 */
2270 	if (!so_start_fallback(so))
2271 		return (EAGAIN);
2272 #ifdef DEBUG
2273 	/*
2274 	 * Make a copy of the sonode in case we need to make an integrity
2275 	 * check later on.
2276 	 */
2277 	bcopy(so, &origso, sizeof (*so));
2278 #endif
2279 
2280 	sp->sp_stats.sps_nfallback.value.ui64++;
2281 
2282 	newsp = sockparams_hold_ephemeral_bydev(so->so_family, so->so_type,
2283 	    so->so_protocol, so->so_sockparams->sp_sdev_info.sd_devpath,
2284 	    KM_SLEEP, &error);
2285 	if (error != 0)
2286 		goto out;
2287 
2288 	if (so->so_direct != NULL) {
2289 		sodirect_t *sodp = so->so_direct;
2290 		mutex_enter(&so->so_lock);
2291 
2292 		so->so_direct->sod_enabled = B_FALSE;
2293 		so->so_state &= ~SS_SODIRECT;
2294 		ASSERT(sodp->sod_uioafh == NULL);
2295 		mutex_exit(&so->so_lock);
2296 	}
2297 
2298 	/* Turn sonode into a TPI socket */
2299 	error = sotpi_convert_sonode(so, newsp, &direct, &q, cr);
2300 	if (error != 0)
2301 		goto out;
2302 
2303 
2304 	/*
2305 	 * Now tell the protocol to start using TPI. so_quiesced_cb be
2306 	 * called once it's safe to synchronize state.
2307 	 */
2308 	DTRACE_PROBE1(proto__fallback__begin, struct sonode *, so);
2309 	error = (*fbfunc)(so->so_proto_handle, q, direct, so_quiesced_cb);
2310 	DTRACE_PROBE1(proto__fallback__end, struct sonode *, so);
2311 
2312 	if (error != 0) {
2313 		/* protocol was unable to do a fallback, revert the sonode */
2314 		sotpi_revert_sonode(so, cr);
2315 		goto out;
2316 	}
2317 
2318 	/*
2319 	 * Walk the accept queue and notify the proto that they should
2320 	 * fall back to TPI. The protocol will send up the T_CONN_IND.
2321 	 */
2322 	nso = so->so_acceptq_head;
2323 	while (nso != NULL) {
2324 		int rval;
2325 
2326 		DTRACE_PROBE1(proto__fallback__begin, struct sonode *, nso);
2327 		rval = (*fbfunc)(nso->so_proto_handle, NULL, direct, NULL);
2328 		DTRACE_PROBE1(proto__fallback__end, struct sonode *, nso);
2329 		if (rval != 0) {
2330 			zcmn_err(getzoneid(), CE_WARN,
2331 			    "Failed to convert socket in accept queue to TPI. "
2332 			    "Pid = %d\n", curproc->p_pid);
2333 		}
2334 		nso = nso->so_acceptq_next;
2335 	}
2336 
2337 	/*
2338 	 * Now flush the acceptq, this will destroy all sockets. They will
2339 	 * be recreated in sotpi_accept().
2340 	 */
2341 	so_acceptq_flush(so, B_FALSE);
2342 
2343 	mutex_enter(&so->so_lock);
2344 	so->so_state |= SS_FALLBACK_COMP;
2345 	mutex_exit(&so->so_lock);
2346 
2347 	/*
2348 	 * Swap the sonode ops. Socket opertations that come in once this
2349 	 * is done will proceed without blocking.
2350 	 */
2351 	so->so_ops = &sotpi_sonodeops;
2352 
2353 	/*
2354 	 * Wake up any threads stuck in poll. This is needed since the poll
2355 	 * head changes when the fallback happens (moves from the sonode to
2356 	 * the STREAMS head).
2357 	 */
2358 	pollwakeup(&so->so_poll_list, POLLERR);
2359 out:
2360 	so_end_fallback(so);
2361 
2362 	if (error != 0) {
2363 #ifdef DEBUG
2364 		so_integrity_check(so, &origso);
2365 #endif
2366 		zcmn_err(getzoneid(), CE_WARN,
2367 		    "Failed to convert socket to TPI (err=%d). Pid = %d\n",
2368 		    error, curproc->p_pid);
2369 		if (newsp != NULL)
2370 			SOCKPARAMS_DEC_REF(newsp);
2371 	}
2372 
2373 	return (error);
2374 }
2375