xref: /titanic_41/usr/src/uts/common/fs/sockfs/sockcommon_subr.c (revision 3fb517f786391b507780c78aabb8d98bfea9efe9)
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 2010 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 			error = cv_reltimedwait_sig(&so->so_snd_cv,
315 			    &so->so_lock, so->so_sndtimeo, TR_CLOCK_TICK);
316 		}
317 		if (error == 0)
318 			return (EINTR);
319 		else if (error == -1)
320 			return (EAGAIN);
321 	}
322 	return (0);
323 }
324 
325 /*
326  * int so_wait_sendbuf(struct sonode *so, boolean_t dontblock)
327  *
328  * Wait for the transport to notify us about send buffers becoming
329  * available.
330  */
331 int
332 so_snd_wait_qnotfull(struct sonode *so, boolean_t dontblock)
333 {
334 	int error = 0;
335 
336 	mutex_enter(&so->so_lock);
337 	if (so->so_snd_qfull) {
338 		so->so_snd_wakeup = B_TRUE;
339 		error = so_snd_wait_qnotfull_locked(so, dontblock);
340 		so->so_snd_wakeup = B_FALSE;
341 	}
342 	mutex_exit(&so->so_lock);
343 
344 	return (error);
345 }
346 
347 void
348 so_snd_qfull(struct sonode *so)
349 {
350 	mutex_enter(&so->so_lock);
351 	so->so_snd_qfull = B_TRUE;
352 	mutex_exit(&so->so_lock);
353 }
354 
355 void
356 so_snd_qnotfull(struct sonode *so)
357 {
358 	mutex_enter(&so->so_lock);
359 	so->so_snd_qfull = B_FALSE;
360 	/* wake up everyone waiting for buffers */
361 	cv_broadcast(&so->so_snd_cv);
362 	mutex_exit(&so->so_lock);
363 }
364 
365 /*
366  * Change the process/process group to which SIGIO is sent.
367  */
368 int
369 socket_chgpgrp(struct sonode *so, pid_t pid)
370 {
371 	int error;
372 
373 	ASSERT(MUTEX_HELD(&so->so_lock));
374 	if (pid != 0) {
375 		/*
376 		 * Permissions check by sending signal 0.
377 		 * Note that when kill fails it does a
378 		 * set_errno causing the system call to fail.
379 		 */
380 		error = kill(pid, 0);
381 		if (error != 0) {
382 			return (error);
383 		}
384 	}
385 	so->so_pgrp = pid;
386 	return (0);
387 }
388 
389 
390 /*
391  * Generate a SIGIO, for 'writable' events include siginfo structure,
392  * for read events just send the signal.
393  */
394 /*ARGSUSED*/
395 static void
396 socket_sigproc(proc_t *proc, int event)
397 {
398 	k_siginfo_t info;
399 
400 	ASSERT(event & (SOCKETSIG_WRITE | SOCKETSIG_READ | SOCKETSIG_URG));
401 
402 	if (event & SOCKETSIG_WRITE) {
403 		info.si_signo = SIGPOLL;
404 		info.si_code = POLL_OUT;
405 		info.si_errno = 0;
406 		info.si_fd = 0;
407 		info.si_band = 0;
408 		sigaddq(proc, NULL, &info, KM_NOSLEEP);
409 	}
410 	if (event & SOCKETSIG_READ) {
411 		sigtoproc(proc, NULL, SIGPOLL);
412 	}
413 	if (event & SOCKETSIG_URG) {
414 		sigtoproc(proc, NULL, SIGURG);
415 	}
416 }
417 
418 void
419 socket_sendsig(struct sonode *so, int event)
420 {
421 	proc_t *proc;
422 
423 	ASSERT(MUTEX_HELD(&so->so_lock));
424 
425 	if (so->so_pgrp == 0 || (!(so->so_state & SS_ASYNC) &&
426 	    event != SOCKETSIG_URG)) {
427 		return;
428 	}
429 
430 	dprint(3, ("sending sig %d to %d\n", event, so->so_pgrp));
431 
432 	if (so->so_pgrp > 0) {
433 		/*
434 		 * XXX This unfortunately still generates
435 		 * a signal when a fd is closed but
436 		 * the proc is active.
437 		 */
438 		mutex_enter(&pidlock);
439 		proc = prfind(so->so_pgrp);
440 		if (proc == NULL) {
441 			mutex_exit(&pidlock);
442 			return;
443 		}
444 		mutex_enter(&proc->p_lock);
445 		mutex_exit(&pidlock);
446 		socket_sigproc(proc, event);
447 		mutex_exit(&proc->p_lock);
448 	} else {
449 		/*
450 		 * Send to process group. Hold pidlock across
451 		 * calls to socket_sigproc().
452 		 */
453 		pid_t pgrp = -so->so_pgrp;
454 
455 		mutex_enter(&pidlock);
456 		proc = pgfind(pgrp);
457 		while (proc != NULL) {
458 			mutex_enter(&proc->p_lock);
459 			socket_sigproc(proc, event);
460 			mutex_exit(&proc->p_lock);
461 			proc = proc->p_pglink;
462 		}
463 		mutex_exit(&pidlock);
464 	}
465 }
466 
467 #define	MIN(a, b) ((a) < (b) ? (a) : (b))
468 /* Copy userdata into a new mblk_t */
469 mblk_t *
470 socopyinuio(uio_t *uiop, ssize_t iosize, size_t wroff, ssize_t maxblk,
471     size_t tail_len, int *errorp)
472 {
473 	mblk_t	*head = NULL, **tail = &head;
474 
475 	ASSERT(iosize == INFPSZ || iosize > 0);
476 
477 	if (iosize == INFPSZ || iosize > uiop->uio_resid)
478 		iosize = uiop->uio_resid;
479 
480 	if (maxblk == INFPSZ)
481 		maxblk = iosize;
482 
483 	/* Nothing to do in these cases, so we're done */
484 	if (iosize < 0 || maxblk < 0 || (maxblk == 0 && iosize > 0))
485 		goto done;
486 
487 	/*
488 	 * We will enter the loop below if iosize is 0; it will allocate an
489 	 * empty message block and call uiomove(9F) which will just return.
490 	 * We could avoid that with an extra check but would only slow
491 	 * down the much more likely case where iosize is larger than 0.
492 	 */
493 	do {
494 		ssize_t blocksize;
495 		mblk_t	*mp;
496 
497 		blocksize = MIN(iosize, maxblk);
498 		ASSERT(blocksize >= 0);
499 		mp = allocb(wroff + blocksize + tail_len, BPRI_MED);
500 		if (mp == NULL) {
501 			*errorp = ENOMEM;
502 			return (head);
503 		}
504 		mp->b_rptr += wroff;
505 		mp->b_wptr = mp->b_rptr + blocksize;
506 
507 		*tail = mp;
508 		tail = &mp->b_cont;
509 
510 		/* uiomove(9F) either returns 0 or EFAULT */
511 		if ((*errorp = uiomove(mp->b_rptr, (size_t)blocksize,
512 		    UIO_WRITE, uiop)) != 0) {
513 			ASSERT(*errorp != ENOMEM);
514 			freemsg(head);
515 			return (NULL);
516 		}
517 
518 		iosize -= blocksize;
519 	} while (iosize > 0);
520 
521 done:
522 	*errorp = 0;
523 	return (head);
524 }
525 
526 mblk_t *
527 socopyoutuio(mblk_t *mp, struct uio *uiop, ssize_t max_read, int *errorp)
528 {
529 	int error;
530 	ptrdiff_t n;
531 	mblk_t *nmp;
532 
533 	ASSERT(mp->b_wptr >= mp->b_rptr);
534 
535 	/*
536 	 * max_read is the offset of the oobmark and read can not go pass
537 	 * the oobmark.
538 	 */
539 	if (max_read == INFPSZ || max_read > uiop->uio_resid)
540 		max_read = uiop->uio_resid;
541 
542 	do {
543 		if ((n = MIN(max_read, MBLKL(mp))) != 0) {
544 			ASSERT(n > 0);
545 
546 			error = uiomove(mp->b_rptr, n, UIO_READ, uiop);
547 			if (error != 0) {
548 				freemsg(mp);
549 				*errorp = error;
550 				return (NULL);
551 			}
552 		}
553 
554 		mp->b_rptr += n;
555 		max_read -= n;
556 		while (mp != NULL && (mp->b_rptr >= mp->b_wptr)) {
557 			/*
558 			 * get rid of zero length mblks
559 			 */
560 			nmp = mp;
561 			mp = mp->b_cont;
562 			freeb(nmp);
563 		}
564 	} while (mp != NULL && max_read > 0);
565 
566 	*errorp = 0;
567 	return (mp);
568 }
569 
570 static void
571 so_prepend_msg(struct sonode *so, mblk_t *mp, mblk_t *last_tail)
572 {
573 	ASSERT(last_tail != NULL);
574 	mp->b_next = so->so_rcv_q_head;
575 	mp->b_prev = last_tail;
576 	ASSERT(!(DB_FLAGS(mp) & DBLK_UIOA));
577 
578 	if (so->so_rcv_q_head == NULL) {
579 		ASSERT(so->so_rcv_q_last_head == NULL);
580 		so->so_rcv_q_last_head = mp;
581 #ifdef DEBUG
582 	} else {
583 		ASSERT(!(DB_FLAGS(so->so_rcv_q_head) & DBLK_UIOA));
584 #endif
585 	}
586 	so->so_rcv_q_head = mp;
587 
588 #ifdef DEBUG
589 	if (so_debug_length) {
590 		mutex_enter(&so->so_lock);
591 		ASSERT(so_check_length(so));
592 		mutex_exit(&so->so_lock);
593 	}
594 #endif
595 }
596 
597 /*
598  * Move a mblk chain (mp_head, mp_last_head) to the sonode's rcv queue so it
599  * can be processed by so_dequeue_msg().
600  */
601 void
602 so_process_new_message(struct sonode *so, mblk_t *mp_head, mblk_t *mp_last_head)
603 {
604 	ASSERT(mp_head->b_prev != NULL);
605 	if (so->so_rcv_q_head  == NULL) {
606 		so->so_rcv_q_head = mp_head;
607 		so->so_rcv_q_last_head = mp_last_head;
608 		ASSERT(so->so_rcv_q_last_head->b_prev != NULL);
609 	} else {
610 		boolean_t flag_equal = ((DB_FLAGS(mp_head) & DBLK_UIOA) ==
611 		    (DB_FLAGS(so->so_rcv_q_last_head) & DBLK_UIOA));
612 
613 		if (mp_head->b_next == NULL &&
614 		    DB_TYPE(mp_head) == M_DATA &&
615 		    DB_TYPE(so->so_rcv_q_last_head) == M_DATA && flag_equal) {
616 			so->so_rcv_q_last_head->b_prev->b_cont = mp_head;
617 			so->so_rcv_q_last_head->b_prev = mp_head->b_prev;
618 			mp_head->b_prev = NULL;
619 		} else if (flag_equal && (DB_FLAGS(mp_head) & DBLK_UIOA)) {
620 			/*
621 			 * Append to last_head if more than one mblks, and both
622 			 * mp_head and last_head are I/OAT mblks.
623 			 */
624 			ASSERT(mp_head->b_next != NULL);
625 			so->so_rcv_q_last_head->b_prev->b_cont = mp_head;
626 			so->so_rcv_q_last_head->b_prev = mp_head->b_prev;
627 			mp_head->b_prev = NULL;
628 
629 			so->so_rcv_q_last_head->b_next = mp_head->b_next;
630 			mp_head->b_next = NULL;
631 			so->so_rcv_q_last_head = mp_last_head;
632 		} else {
633 #ifdef DEBUG
634 			{
635 				mblk_t *tmp_mblk;
636 				tmp_mblk = mp_head;
637 				while (tmp_mblk != NULL) {
638 					ASSERT(tmp_mblk->b_prev != NULL);
639 					tmp_mblk = tmp_mblk->b_next;
640 				}
641 			}
642 #endif
643 			so->so_rcv_q_last_head->b_next = mp_head;
644 			so->so_rcv_q_last_head = mp_last_head;
645 		}
646 	}
647 }
648 
649 /*
650  * Check flow control on a given sonode.  Must have so_lock held, and
651  * this function will release the hold.
652  */
653 
654 static void
655 so_check_flow_control(struct sonode *so)
656 {
657 	ASSERT(MUTEX_HELD(&so->so_lock));
658 
659 	if (so->so_flowctrld && so->so_rcv_queued < so->so_rcvlowat) {
660 		so->so_flowctrld = B_FALSE;
661 		mutex_exit(&so->so_lock);
662 		/*
663 		 * Open up flow control. SCTP does not have any downcalls, and
664 		 * it will clr flow ctrl in sosctp_recvmsg().
665 		 */
666 		if (so->so_downcalls != NULL &&
667 		    so->so_downcalls->sd_clr_flowctrl != NULL) {
668 			(*so->so_downcalls->sd_clr_flowctrl)
669 			    (so->so_proto_handle);
670 		}
671 	} else {
672 		mutex_exit(&so->so_lock);
673 	}
674 }
675 
676 int
677 so_dequeue_msg(struct sonode *so, mblk_t **mctlp, struct uio *uiop,
678     rval_t *rvalp, int flags)
679 {
680 	mblk_t	*mp, *nmp;
681 	mblk_t	*savemp, *savemptail;
682 	mblk_t	*new_msg_head;
683 	mblk_t	*new_msg_last_head;
684 	mblk_t	*last_tail;
685 	boolean_t partial_read;
686 	boolean_t reset_atmark = B_FALSE;
687 	int more = 0;
688 	int error;
689 	ssize_t oobmark;
690 	sodirect_t *sodp = so->so_direct;
691 
692 	partial_read = B_FALSE;
693 	*mctlp = NULL;
694 again:
695 	mutex_enter(&so->so_lock);
696 again1:
697 #ifdef DEBUG
698 	if (so_debug_length) {
699 		ASSERT(so_check_length(so));
700 	}
701 #endif
702 	if (so->so_state & SS_RCVATMARK) {
703 		/* Check whether the caller is OK to read past the mark */
704 		if (flags & MSG_NOMARK) {
705 			mutex_exit(&so->so_lock);
706 			return (EWOULDBLOCK);
707 		}
708 		reset_atmark = B_TRUE;
709 	}
710 	/*
711 	 * First move messages from the dump area to processing area
712 	 */
713 	if (sodp != NULL) {
714 		if (sodp->sod_enabled) {
715 			if (sodp->sod_uioa.uioa_state & UIOA_ALLOC) {
716 				/* nothing to uioamove */
717 				sodp = NULL;
718 			} else if (sodp->sod_uioa.uioa_state & UIOA_INIT) {
719 				sodp->sod_uioa.uioa_state &= UIOA_CLR;
720 				sodp->sod_uioa.uioa_state |= UIOA_ENABLED;
721 				/*
722 				 * try to uioamove() the data that
723 				 * has already queued.
724 				 */
725 				sod_uioa_so_init(so, sodp, uiop);
726 			}
727 		} else {
728 			sodp = NULL;
729 		}
730 	}
731 	new_msg_head = so->so_rcv_head;
732 	new_msg_last_head = so->so_rcv_last_head;
733 	so->so_rcv_head = NULL;
734 	so->so_rcv_last_head = NULL;
735 	oobmark = so->so_oobmark;
736 	/*
737 	 * We can release the lock as there can only be one reader
738 	 */
739 	mutex_exit(&so->so_lock);
740 
741 	if (new_msg_head != NULL) {
742 		so_process_new_message(so, new_msg_head, new_msg_last_head);
743 	}
744 	savemp = savemptail = NULL;
745 	rvalp->r_val1 = 0;
746 	error = 0;
747 	mp = so->so_rcv_q_head;
748 
749 	if (mp != NULL &&
750 	    (so->so_rcv_timer_tid == 0 ||
751 	    so->so_rcv_queued >= so->so_rcv_thresh)) {
752 		partial_read = B_FALSE;
753 
754 		if (flags & MSG_PEEK) {
755 			if ((nmp = dupmsg(mp)) == NULL &&
756 			    (nmp = copymsg(mp)) == NULL) {
757 				size_t size = msgsize(mp);
758 
759 				error = strwaitbuf(size, BPRI_HI);
760 				if (error) {
761 					return (error);
762 				}
763 				goto again;
764 			}
765 			mp = nmp;
766 		} else {
767 			ASSERT(mp->b_prev != NULL);
768 			last_tail = mp->b_prev;
769 			mp->b_prev = NULL;
770 			so->so_rcv_q_head = mp->b_next;
771 			if (so->so_rcv_q_head == NULL) {
772 				so->so_rcv_q_last_head = NULL;
773 			}
774 			mp->b_next = NULL;
775 		}
776 
777 		ASSERT(mctlp != NULL);
778 		/*
779 		 * First process PROTO or PCPROTO blocks, if any.
780 		 */
781 		if (DB_TYPE(mp) != M_DATA) {
782 			*mctlp = mp;
783 			savemp = mp;
784 			savemptail = mp;
785 			ASSERT(DB_TYPE(mp) == M_PROTO ||
786 			    DB_TYPE(mp) == M_PCPROTO);
787 			while (mp->b_cont != NULL &&
788 			    DB_TYPE(mp->b_cont) != M_DATA) {
789 				ASSERT(DB_TYPE(mp->b_cont) == M_PROTO ||
790 				    DB_TYPE(mp->b_cont) == M_PCPROTO);
791 				mp = mp->b_cont;
792 				savemptail = mp;
793 			}
794 			mp = savemptail->b_cont;
795 			savemptail->b_cont = NULL;
796 		}
797 
798 		ASSERT(DB_TYPE(mp) == M_DATA);
799 		/*
800 		 * Now process DATA blocks, if any. Note that for sodirect
801 		 * enabled socket, uio_resid can be 0.
802 		 */
803 		if (uiop->uio_resid >= 0) {
804 			ssize_t copied = 0;
805 
806 			if (sodp != NULL && (DB_FLAGS(mp) & DBLK_UIOA)) {
807 				mutex_enter(&so->so_lock);
808 				ASSERT(uiop == (uio_t *)&sodp->sod_uioa);
809 				copied = sod_uioa_mblk(so, mp);
810 				if (copied > 0)
811 					partial_read = B_TRUE;
812 				mutex_exit(&so->so_lock);
813 				/* mark this mblk as processed */
814 				mp = NULL;
815 			} else {
816 				ssize_t oldresid = uiop->uio_resid;
817 
818 				if (MBLKL(mp) < so_mblk_pull_len) {
819 					if (pullupmsg(mp, -1) == 1) {
820 						last_tail = mp;
821 					}
822 				}
823 				/*
824 				 * Can not read beyond the oobmark
825 				 */
826 				mp = socopyoutuio(mp, uiop,
827 				    oobmark == 0 ? INFPSZ : oobmark, &error);
828 				if (error != 0) {
829 					freemsg(*mctlp);
830 					*mctlp = NULL;
831 					more = 0;
832 					goto done;
833 				}
834 				ASSERT(oldresid >= uiop->uio_resid);
835 				copied = oldresid - uiop->uio_resid;
836 				if (oldresid > uiop->uio_resid)
837 					partial_read = B_TRUE;
838 			}
839 			ASSERT(copied >= 0);
840 			if (copied > 0 && !(flags & MSG_PEEK)) {
841 				mutex_enter(&so->so_lock);
842 				so->so_rcv_queued -= copied;
843 				ASSERT(so->so_oobmark >= 0);
844 				if (so->so_oobmark > 0) {
845 					so->so_oobmark -= copied;
846 					ASSERT(so->so_oobmark >= 0);
847 					if (so->so_oobmark == 0) {
848 						ASSERT(so->so_state &
849 						    SS_OOBPEND);
850 						so->so_oobmark = 0;
851 						so->so_state |= SS_RCVATMARK;
852 					}
853 				}
854 				/*
855 				 * so_check_flow_control() will drop
856 				 * so->so_lock.
857 				 */
858 				so_check_flow_control(so);
859 			}
860 		}
861 		if (mp != NULL) { /* more data blocks in msg */
862 			more |= MOREDATA;
863 			if ((flags & (MSG_PEEK|MSG_TRUNC))) {
864 				if (flags & MSG_PEEK) {
865 					freemsg(mp);
866 				} else {
867 					unsigned int msize = msgdsize(mp);
868 
869 					freemsg(mp);
870 					mutex_enter(&so->so_lock);
871 					so->so_rcv_queued -= msize;
872 					/*
873 					 * so_check_flow_control() will drop
874 					 * so->so_lock.
875 					 */
876 					so_check_flow_control(so);
877 				}
878 			} else if (partial_read && !somsghasdata(mp)) {
879 				/*
880 				 * Avoid queuing a zero-length tail part of
881 				 * a message. partial_read == 1 indicates that
882 				 * we read some of the message.
883 				 */
884 				freemsg(mp);
885 				more &= ~MOREDATA;
886 			} else {
887 				if (savemp != NULL &&
888 				    (flags & MSG_DUPCTRL)) {
889 					mblk_t *nmp;
890 					/*
891 					 * There should only be non data mblks
892 					 */
893 					ASSERT(DB_TYPE(savemp) != M_DATA &&
894 					    DB_TYPE(savemptail) != M_DATA);
895 try_again:
896 					if ((nmp = dupmsg(savemp)) == NULL &&
897 					    (nmp = copymsg(savemp)) == NULL) {
898 
899 						size_t size = msgsize(savemp);
900 
901 						error = strwaitbuf(size,
902 						    BPRI_HI);
903 						if (error != 0) {
904 							/*
905 							 * In case we
906 							 * cannot copy
907 							 * control data
908 							 * free the remaining
909 							 * data.
910 							 */
911 							freemsg(mp);
912 							goto done;
913 						}
914 						goto try_again;
915 					}
916 
917 					ASSERT(nmp != NULL);
918 					ASSERT(DB_TYPE(nmp) != M_DATA);
919 					savemptail->b_cont = mp;
920 					*mctlp = nmp;
921 					mp = savemp;
922 				}
923 				/*
924 				 * putback mp
925 				 */
926 				so_prepend_msg(so, mp, last_tail);
927 			}
928 		}
929 
930 		/* fast check so_rcv_head if there is more data */
931 		if (partial_read && !(so->so_state & SS_RCVATMARK) &&
932 		    *mctlp == NULL && uiop->uio_resid > 0 &&
933 		    !(flags & MSG_PEEK) && so->so_rcv_head != NULL) {
934 			goto again;
935 		}
936 	} else if (!partial_read) {
937 		mutex_enter(&so->so_lock);
938 		if (so->so_error != 0) {
939 			error = sogeterr(so, !(flags & MSG_PEEK));
940 			mutex_exit(&so->so_lock);
941 			return (error);
942 		}
943 		/*
944 		 * No pending data. Return right away for nonblocking
945 		 * socket, otherwise sleep waiting for data.
946 		 */
947 		if (!(so->so_state & SS_CANTRCVMORE) && uiop->uio_resid > 0) {
948 			if ((uiop->uio_fmode & (FNDELAY|FNONBLOCK)) ||
949 			    (flags & MSG_DONTWAIT)) {
950 				error = EWOULDBLOCK;
951 			} else {
952 				if (so->so_state & (SS_CLOSING |
953 				    SS_FALLBACK_PENDING)) {
954 					mutex_exit(&so->so_lock);
955 					error = EINTR;
956 					goto done;
957 				}
958 
959 				if (so->so_rcv_head != NULL) {
960 					goto again1;
961 				}
962 				so->so_rcv_wakeup = B_TRUE;
963 				so->so_rcv_wanted = uiop->uio_resid;
964 				if (so->so_rcvtimeo == 0) {
965 					/*
966 					 * Zero means disable timeout.
967 					 */
968 					error = cv_wait_sig(&so->so_rcv_cv,
969 					    &so->so_lock);
970 				} else {
971 					error = cv_reltimedwait_sig(
972 					    &so->so_rcv_cv, &so->so_lock,
973 					    so->so_rcvtimeo, TR_CLOCK_TICK);
974 				}
975 				so->so_rcv_wakeup = B_FALSE;
976 				so->so_rcv_wanted = 0;
977 
978 				if (error == 0) {
979 					error = EINTR;
980 				} else if (error == -1) {
981 					error = EAGAIN;
982 				} else {
983 					goto again1;
984 				}
985 			}
986 		}
987 		mutex_exit(&so->so_lock);
988 	}
989 	if (reset_atmark && partial_read && !(flags & MSG_PEEK)) {
990 		/*
991 		 * We are passed the mark, update state
992 		 * 4.3BSD and 4.4BSD clears the mark when peeking across it.
993 		 * The draft Posix socket spec states that the mark should
994 		 * not be cleared when peeking. We follow the latter.
995 		 */
996 		mutex_enter(&so->so_lock);
997 		ASSERT(so_verify_oobstate(so));
998 		so->so_state &= ~(SS_OOBPEND|SS_HAVEOOBDATA|SS_RCVATMARK);
999 		freemsg(so->so_oobmsg);
1000 		so->so_oobmsg = NULL;
1001 		ASSERT(so_verify_oobstate(so));
1002 		mutex_exit(&so->so_lock);
1003 	}
1004 	ASSERT(so->so_rcv_wakeup == B_FALSE);
1005 done:
1006 	if (sodp != NULL) {
1007 		mutex_enter(&so->so_lock);
1008 		if (sodp->sod_enabled &&
1009 		    (sodp->sod_uioa.uioa_state & UIOA_ENABLED)) {
1010 			SOD_UIOAFINI(sodp);
1011 			if (sodp->sod_uioa.uioa_mbytes > 0) {
1012 				ASSERT(so->so_rcv_q_head != NULL ||
1013 				    so->so_rcv_head != NULL);
1014 				so->so_rcv_queued -= sod_uioa_mblk(so, NULL);
1015 				if (error == EWOULDBLOCK)
1016 					error = 0;
1017 			}
1018 		}
1019 		mutex_exit(&so->so_lock);
1020 	}
1021 #ifdef DEBUG
1022 	if (so_debug_length) {
1023 		mutex_enter(&so->so_lock);
1024 		ASSERT(so_check_length(so));
1025 		mutex_exit(&so->so_lock);
1026 	}
1027 #endif
1028 	rvalp->r_val1 = more;
1029 	ASSERT(MUTEX_NOT_HELD(&so->so_lock));
1030 	return (error);
1031 }
1032 
1033 /*
1034  * Enqueue data from the protocol on the socket's rcv queue.
1035  *
1036  * We try to hook new M_DATA mblks onto an existing chain, however,
1037  * that cannot be done if the existing chain has already been
1038  * processed by I/OAT. Non-M_DATA mblks are just linked together via
1039  * b_next. In all cases the b_prev of the enqueued mblk is set to
1040  * point to the last mblk in its b_cont chain.
1041  */
1042 void
1043 so_enqueue_msg(struct sonode *so, mblk_t *mp, size_t msg_size)
1044 {
1045 	ASSERT(MUTEX_HELD(&so->so_lock));
1046 
1047 #ifdef DEBUG
1048 	if (so_debug_length) {
1049 		ASSERT(so_check_length(so));
1050 	}
1051 #endif
1052 	so->so_rcv_queued += msg_size;
1053 
1054 	if (so->so_rcv_head == NULL) {
1055 		ASSERT(so->so_rcv_last_head == NULL);
1056 		so->so_rcv_head = mp;
1057 		so->so_rcv_last_head = mp;
1058 	} else if ((DB_TYPE(mp) == M_DATA &&
1059 	    DB_TYPE(so->so_rcv_last_head) == M_DATA) &&
1060 	    ((DB_FLAGS(mp) & DBLK_UIOA) ==
1061 	    (DB_FLAGS(so->so_rcv_last_head) & DBLK_UIOA))) {
1062 		/* Added to the end */
1063 		ASSERT(so->so_rcv_last_head != NULL);
1064 		ASSERT(so->so_rcv_last_head->b_prev != NULL);
1065 		so->so_rcv_last_head->b_prev->b_cont = mp;
1066 	} else {
1067 		/* Start a new end */
1068 		so->so_rcv_last_head->b_next = mp;
1069 		so->so_rcv_last_head = mp;
1070 	}
1071 	while (mp->b_cont != NULL)
1072 		mp = mp->b_cont;
1073 
1074 	so->so_rcv_last_head->b_prev = mp;
1075 #ifdef DEBUG
1076 	if (so_debug_length) {
1077 		ASSERT(so_check_length(so));
1078 	}
1079 #endif
1080 }
1081 
1082 /*
1083  * Return B_TRUE if there is data in the message, B_FALSE otherwise.
1084  */
1085 boolean_t
1086 somsghasdata(mblk_t *mp)
1087 {
1088 	for (; mp; mp = mp->b_cont)
1089 		if (mp->b_datap->db_type == M_DATA) {
1090 			ASSERT(mp->b_wptr >= mp->b_rptr);
1091 			if (mp->b_wptr > mp->b_rptr)
1092 				return (B_TRUE);
1093 		}
1094 	return (B_FALSE);
1095 }
1096 
1097 /*
1098  * Flush the read side of sockfs.
1099  *
1100  * The caller must be sure that a reader is not already active when the
1101  * buffer is being flushed.
1102  */
1103 void
1104 so_rcv_flush(struct sonode *so)
1105 {
1106 	mblk_t  *mp;
1107 
1108 	ASSERT(MUTEX_HELD(&so->so_lock));
1109 
1110 	if (so->so_oobmsg != NULL) {
1111 		freemsg(so->so_oobmsg);
1112 		so->so_oobmsg = NULL;
1113 		so->so_oobmark = 0;
1114 		so->so_state &=
1115 		    ~(SS_OOBPEND|SS_HAVEOOBDATA|SS_HADOOBDATA|SS_RCVATMARK);
1116 	}
1117 
1118 	/*
1119 	 * Free messages sitting in the send and recv queue
1120 	 */
1121 	while (so->so_rcv_q_head != NULL) {
1122 		mp = so->so_rcv_q_head;
1123 		so->so_rcv_q_head = mp->b_next;
1124 		mp->b_next = mp->b_prev = NULL;
1125 		freemsg(mp);
1126 	}
1127 	while (so->so_rcv_head != NULL) {
1128 		mp = so->so_rcv_head;
1129 		so->so_rcv_head = mp->b_next;
1130 		mp->b_next = mp->b_prev = NULL;
1131 		freemsg(mp);
1132 	}
1133 	so->so_rcv_queued = 0;
1134 	so->so_rcv_q_head = NULL;
1135 	so->so_rcv_q_last_head = NULL;
1136 	so->so_rcv_head = NULL;
1137 	so->so_rcv_last_head = NULL;
1138 }
1139 
1140 /*
1141  * Handle recv* calls that set MSG_OOB or MSG_OOB together with MSG_PEEK.
1142  */
1143 int
1144 sorecvoob(struct sonode *so, struct nmsghdr *msg, struct uio *uiop, int flags,
1145     boolean_t oob_inline)
1146 {
1147 	mblk_t		*mp, *nmp;
1148 	int		error;
1149 
1150 	dprintso(so, 1, ("sorecvoob(%p, %p, 0x%x)\n", (void *)so, (void *)msg,
1151 	    flags));
1152 
1153 	if (msg != NULL) {
1154 		/*
1155 		 * There is never any oob data with addresses or control since
1156 		 * the T_EXDATA_IND does not carry any options.
1157 		 */
1158 		msg->msg_controllen = 0;
1159 		msg->msg_namelen = 0;
1160 		msg->msg_flags = 0;
1161 	}
1162 
1163 	mutex_enter(&so->so_lock);
1164 	ASSERT(so_verify_oobstate(so));
1165 	if (oob_inline ||
1166 	    (so->so_state & (SS_OOBPEND|SS_HADOOBDATA)) != SS_OOBPEND) {
1167 		dprintso(so, 1, ("sorecvoob: inline or data consumed\n"));
1168 		mutex_exit(&so->so_lock);
1169 		return (EINVAL);
1170 	}
1171 	if (!(so->so_state & SS_HAVEOOBDATA)) {
1172 		dprintso(so, 1, ("sorecvoob: no data yet\n"));
1173 		mutex_exit(&so->so_lock);
1174 		return (EWOULDBLOCK);
1175 	}
1176 	ASSERT(so->so_oobmsg != NULL);
1177 	mp = so->so_oobmsg;
1178 	if (flags & MSG_PEEK) {
1179 		/*
1180 		 * Since recv* can not return ENOBUFS we can not use dupmsg.
1181 		 * Instead we revert to the consolidation private
1182 		 * allocb_wait plus bcopy.
1183 		 */
1184 		mblk_t *mp1;
1185 
1186 		mp1 = allocb_wait(msgdsize(mp), BPRI_MED, STR_NOSIG, NULL);
1187 		ASSERT(mp1);
1188 
1189 		while (mp != NULL) {
1190 			ssize_t size;
1191 
1192 			size = MBLKL(mp);
1193 			bcopy(mp->b_rptr, mp1->b_wptr, size);
1194 			mp1->b_wptr += size;
1195 			ASSERT(mp1->b_wptr <= mp1->b_datap->db_lim);
1196 			mp = mp->b_cont;
1197 		}
1198 		mp = mp1;
1199 	} else {
1200 		/*
1201 		 * Update the state indicating that the data has been consumed.
1202 		 * Keep SS_OOBPEND set until data is consumed past the mark.
1203 		 */
1204 		so->so_oobmsg = NULL;
1205 		so->so_state ^= SS_HAVEOOBDATA|SS_HADOOBDATA;
1206 	}
1207 	ASSERT(so_verify_oobstate(so));
1208 	mutex_exit(&so->so_lock);
1209 
1210 	error = 0;
1211 	nmp = mp;
1212 	while (nmp != NULL && uiop->uio_resid > 0) {
1213 		ssize_t n = MBLKL(nmp);
1214 
1215 		n = MIN(n, uiop->uio_resid);
1216 		if (n > 0)
1217 			error = uiomove(nmp->b_rptr, n,
1218 			    UIO_READ, uiop);
1219 		if (error)
1220 			break;
1221 		nmp = nmp->b_cont;
1222 	}
1223 	ASSERT(mp->b_next == NULL && mp->b_prev == NULL);
1224 	freemsg(mp);
1225 	return (error);
1226 }
1227 
1228 /*
1229  * Allocate and initializ sonode
1230  */
1231 /* ARGSUSED */
1232 struct sonode *
1233 socket_sonode_create(struct sockparams *sp, int family, int type,
1234     int protocol, int version, int sflags, int *errorp, struct cred *cr)
1235 {
1236 	sonode_t *so;
1237 	int	kmflags;
1238 
1239 	/*
1240 	 * Choose the right set of sonodeops based on the upcall and
1241 	 * down call version that the protocol has provided
1242 	 */
1243 	if (SOCK_UC_VERSION != sp->sp_smod_info->smod_uc_version ||
1244 	    SOCK_DC_VERSION != sp->sp_smod_info->smod_dc_version) {
1245 		/*
1246 		 * mismatch
1247 		 */
1248 #ifdef DEBUG
1249 		cmn_err(CE_CONT, "protocol and socket module version mismatch");
1250 #endif
1251 		*errorp = EINVAL;
1252 		return (NULL);
1253 	}
1254 
1255 	kmflags = (sflags & SOCKET_NOSLEEP) ? KM_NOSLEEP : KM_SLEEP;
1256 
1257 	so = kmem_cache_alloc(socket_cache, kmflags);
1258 	if (so == NULL) {
1259 		*errorp = ENOMEM;
1260 		return (NULL);
1261 	}
1262 
1263 	sonode_init(so, sp, family, type, protocol, &so_sonodeops);
1264 
1265 	if (version == SOV_DEFAULT)
1266 		version = so_default_version;
1267 
1268 	so->so_version = (short)version;
1269 
1270 	/*
1271 	 * set the default values to be INFPSZ
1272 	 * if a protocol desires it can change the value later
1273 	 */
1274 	so->so_proto_props.sopp_rxhiwat = SOCKET_RECVHIWATER;
1275 	so->so_proto_props.sopp_rxlowat = SOCKET_RECVLOWATER;
1276 	so->so_proto_props.sopp_maxpsz = INFPSZ;
1277 	so->so_proto_props.sopp_maxblk = INFPSZ;
1278 
1279 	return (so);
1280 }
1281 
1282 int
1283 socket_init_common(struct sonode *so, struct sonode *pso, int flags, cred_t *cr)
1284 {
1285 	int error = 0;
1286 
1287 	if (pso != NULL) {
1288 		/*
1289 		 * We have a passive open, so inherit basic state from
1290 		 * the parent (listener).
1291 		 *
1292 		 * No need to grab the new sonode's lock, since there is no
1293 		 * one that can have a reference to it.
1294 		 */
1295 		mutex_enter(&pso->so_lock);
1296 
1297 		so->so_state |= SS_ISCONNECTED | (pso->so_state & SS_ASYNC);
1298 		so->so_pgrp = pso->so_pgrp;
1299 		so->so_rcvtimeo = pso->so_rcvtimeo;
1300 		so->so_sndtimeo = pso->so_sndtimeo;
1301 		so->so_xpg_rcvbuf = pso->so_xpg_rcvbuf;
1302 		/*
1303 		 * Make note of the socket level options. TCP and IP level
1304 		 * options are already inherited. We could do all this after
1305 		 * accept is successful but doing it here simplifies code and
1306 		 * no harm done for error case.
1307 		 */
1308 		so->so_options = pso->so_options & (SO_DEBUG|SO_REUSEADDR|
1309 		    SO_KEEPALIVE|SO_DONTROUTE|SO_BROADCAST|SO_USELOOPBACK|
1310 		    SO_OOBINLINE|SO_DGRAM_ERRIND|SO_LINGER);
1311 		so->so_proto_props = pso->so_proto_props;
1312 		so->so_mode = pso->so_mode;
1313 		so->so_pollev = pso->so_pollev & SO_POLLEV_ALWAYS;
1314 
1315 		mutex_exit(&pso->so_lock);
1316 	} else {
1317 		struct sockparams *sp = so->so_sockparams;
1318 		sock_upcalls_t *upcalls_to_use;
1319 
1320 		/*
1321 		 * Based on the version number select the right upcalls to
1322 		 * pass down. Currently we only have one version so choose
1323 		 * default
1324 		 */
1325 		upcalls_to_use = &so_upcalls;
1326 
1327 		/* active open, so create a lower handle */
1328 		so->so_proto_handle =
1329 		    sp->sp_smod_info->smod_proto_create_func(so->so_family,
1330 		    so->so_type, so->so_protocol, &so->so_downcalls,
1331 		    &so->so_mode, &error, flags, cr);
1332 
1333 		if (so->so_proto_handle == NULL) {
1334 			ASSERT(error != 0);
1335 			/*
1336 			 * To be safe; if a lower handle cannot be created, and
1337 			 * the proto does not give a reason why, assume there
1338 			 * was a lack of memory.
1339 			 */
1340 			return ((error == 0) ? ENOMEM : error);
1341 		}
1342 		ASSERT(so->so_downcalls != NULL);
1343 		ASSERT(so->so_downcalls->sd_send != NULL ||
1344 		    so->so_downcalls->sd_send_uio != NULL);
1345 		if (so->so_downcalls->sd_recv_uio != NULL) {
1346 			ASSERT(so->so_downcalls->sd_poll != NULL);
1347 			so->so_pollev |= SO_POLLEV_ALWAYS;
1348 		}
1349 
1350 		(*so->so_downcalls->sd_activate)(so->so_proto_handle,
1351 		    (sock_upper_handle_t)so, upcalls_to_use, 0, cr);
1352 
1353 		/* Wildcard */
1354 
1355 		/*
1356 		 * FIXME No need for this, the protocol can deal with it in
1357 		 * sd_create(). Should update ICMP.
1358 		 */
1359 		if (so->so_protocol != so->so_sockparams->sp_protocol) {
1360 			int protocol = so->so_protocol;
1361 			int error;
1362 			/*
1363 			 * Issue SO_PROTOTYPE setsockopt.
1364 			 */
1365 			error = socket_setsockopt(so, SOL_SOCKET, SO_PROTOTYPE,
1366 			    &protocol, (t_uscalar_t)sizeof (protocol), cr);
1367 			if (error) {
1368 				(void) (*so->so_downcalls->sd_close)
1369 				    (so->so_proto_handle, 0, cr);
1370 
1371 				mutex_enter(&so->so_lock);
1372 				so_rcv_flush(so);
1373 				mutex_exit(&so->so_lock);
1374 				/*
1375 				 * Setsockopt often fails with ENOPROTOOPT but
1376 				 * socket() should fail with
1377 				 * EPROTONOSUPPORT/EPROTOTYPE.
1378 				 */
1379 				return (EPROTONOSUPPORT);
1380 			}
1381 		}
1382 	}
1383 
1384 	if (uioasync.enabled)
1385 		sod_sock_init(so);
1386 
1387 	return (0);
1388 }
1389 
1390 /*
1391  * int socket_ioctl_common(struct sonode *so, int cmd, intptr_t arg, int mode,
1392  *         struct cred *cr, int32_t *rvalp)
1393  *
1394  * Handle ioctls that manipulate basic socket state; non-blocking,
1395  * async, etc.
1396  *
1397  * Returns:
1398  *   < 0  - ioctl was not handle
1399  *  >= 0  - ioctl was handled, if > 0, then it is an errno
1400  *
1401  * Notes:
1402  *   Assumes the standard receive buffer is used to obtain info for
1403  *   NREAD.
1404  */
1405 /* ARGSUSED */
1406 int
1407 socket_ioctl_common(struct sonode *so, int cmd, intptr_t arg, int mode,
1408     struct cred *cr, int32_t *rvalp)
1409 {
1410 	switch (cmd) {
1411 	case SIOCSQPTR:
1412 		/*
1413 		 * SIOCSQPTR is valid only when helper stream is created
1414 		 * by the protocol.
1415 		 */
1416 
1417 		return (EOPNOTSUPP);
1418 	case FIONBIO: {
1419 		int32_t value;
1420 
1421 		if (so_copyin((void *)arg, &value, sizeof (int32_t),
1422 		    (mode & (int)FKIOCTL)))
1423 			return (EFAULT);
1424 
1425 		mutex_enter(&so->so_lock);
1426 		if (value) {
1427 			so->so_state |= SS_NDELAY;
1428 		} else {
1429 			so->so_state &= ~SS_NDELAY;
1430 		}
1431 		mutex_exit(&so->so_lock);
1432 		return (0);
1433 	}
1434 	case FIOASYNC: {
1435 		int32_t value;
1436 
1437 		if (so_copyin((void *)arg, &value, sizeof (int32_t),
1438 		    (mode & (int)FKIOCTL)))
1439 			return (EFAULT);
1440 
1441 		mutex_enter(&so->so_lock);
1442 
1443 		if (value) {
1444 			/* Turn on SIGIO */
1445 			so->so_state |= SS_ASYNC;
1446 		} else {
1447 			/* Turn off SIGIO */
1448 			so->so_state &= ~SS_ASYNC;
1449 		}
1450 		mutex_exit(&so->so_lock);
1451 
1452 		return (0);
1453 	}
1454 
1455 	case SIOCSPGRP:
1456 	case FIOSETOWN: {
1457 		int error;
1458 		pid_t pid;
1459 
1460 		if (so_copyin((void *)arg, &pid, sizeof (pid_t),
1461 		    (mode & (int)FKIOCTL)))
1462 			return (EFAULT);
1463 
1464 		mutex_enter(&so->so_lock);
1465 		error = (pid != so->so_pgrp) ? socket_chgpgrp(so, pid) : 0;
1466 		mutex_exit(&so->so_lock);
1467 		return (error);
1468 	}
1469 	case SIOCGPGRP:
1470 	case FIOGETOWN:
1471 		if (so_copyout(&so->so_pgrp, (void *)arg,
1472 		    sizeof (pid_t), (mode & (int)FKIOCTL)))
1473 			return (EFAULT);
1474 
1475 		return (0);
1476 	case SIOCATMARK: {
1477 		int retval;
1478 
1479 		/*
1480 		 * Only protocols that support urgent data can handle ATMARK.
1481 		 */
1482 		if ((so->so_mode & SM_EXDATA) == 0)
1483 			return (EINVAL);
1484 
1485 		/*
1486 		 * If the protocol is maintaining its own buffer, then the
1487 		 * request must be passed down.
1488 		 */
1489 		if (so->so_downcalls->sd_recv_uio != NULL)
1490 			return (-1);
1491 
1492 		retval = (so->so_state & SS_RCVATMARK) != 0;
1493 
1494 		if (so_copyout(&retval, (void *)arg, sizeof (int),
1495 		    (mode & (int)FKIOCTL))) {
1496 			return (EFAULT);
1497 		}
1498 		return (0);
1499 	}
1500 
1501 	case FIONREAD: {
1502 		int retval;
1503 
1504 		/*
1505 		 * If the protocol is maintaining its own buffer, then the
1506 		 * request must be passed down.
1507 		 */
1508 		if (so->so_downcalls->sd_recv_uio != NULL)
1509 			return (-1);
1510 
1511 		retval = MIN(so->so_rcv_queued, INT_MAX);
1512 
1513 		if (so_copyout(&retval, (void *)arg,
1514 		    sizeof (retval), (mode & (int)FKIOCTL))) {
1515 			return (EFAULT);
1516 		}
1517 		return (0);
1518 	}
1519 
1520 	case _I_GETPEERCRED: {
1521 		int error = 0;
1522 
1523 		if ((mode & FKIOCTL) == 0)
1524 			return (EINVAL);
1525 
1526 		mutex_enter(&so->so_lock);
1527 		if ((so->so_mode & SM_CONNREQUIRED) == 0) {
1528 			error = ENOTSUP;
1529 		} else if ((so->so_state & SS_ISCONNECTED) == 0) {
1530 			error = ENOTCONN;
1531 		} else if (so->so_peercred != NULL) {
1532 			k_peercred_t *kp = (k_peercred_t *)arg;
1533 			kp->pc_cr = so->so_peercred;
1534 			kp->pc_cpid = so->so_cpid;
1535 			crhold(so->so_peercred);
1536 		} else {
1537 			error = EINVAL;
1538 		}
1539 		mutex_exit(&so->so_lock);
1540 		return (error);
1541 	}
1542 	default:
1543 		return (-1);
1544 	}
1545 }
1546 
1547 /*
1548  * Handle the I_NREAD STREAM ioctl.
1549  */
1550 static int
1551 so_strioc_nread(struct sonode *so, intptr_t arg, int mode, int32_t *rvalp)
1552 {
1553 	size_t size = 0;
1554 	int retval;
1555 	int count = 0;
1556 	mblk_t *mp;
1557 	clock_t wakeup = drv_usectohz(10);
1558 
1559 	if (so->so_downcalls == NULL ||
1560 	    so->so_downcalls->sd_recv_uio != NULL)
1561 		return (EINVAL);
1562 
1563 	mutex_enter(&so->so_lock);
1564 	/* Wait for reader to get out of the way. */
1565 	while (so->so_flag & SOREADLOCKED) {
1566 		/*
1567 		 * If reader is waiting for data, then there should be nothing
1568 		 * on the rcv queue.
1569 		 */
1570 		if (so->so_rcv_wakeup)
1571 			goto out;
1572 
1573 		/* Do a timed sleep, in case the reader goes to sleep. */
1574 		(void) cv_reltimedwait(&so->so_read_cv, &so->so_lock, wakeup,
1575 		    TR_CLOCK_TICK);
1576 	}
1577 
1578 	/*
1579 	 * Since we are holding so_lock no new reader will come in, and the
1580 	 * protocol will not be able to enqueue data. So it's safe to walk
1581 	 * both rcv queues.
1582 	 */
1583 	mp = so->so_rcv_q_head;
1584 	if (mp != NULL) {
1585 		size = msgdsize(so->so_rcv_q_head);
1586 		for (; mp != NULL; mp = mp->b_next)
1587 			count++;
1588 	} else {
1589 		/*
1590 		 * In case the processing list was empty, get the size of the
1591 		 * next msg in line.
1592 		 */
1593 		size = msgdsize(so->so_rcv_head);
1594 	}
1595 
1596 	for (mp = so->so_rcv_head; mp != NULL; mp = mp->b_next)
1597 		count++;
1598 out:
1599 	mutex_exit(&so->so_lock);
1600 
1601 	/*
1602 	 * Drop down from size_t to the "int" required by the
1603 	 * interface.  Cap at INT_MAX.
1604 	 */
1605 	retval = MIN(size, INT_MAX);
1606 	if (so_copyout(&retval, (void *)arg, sizeof (retval),
1607 	    (mode & (int)FKIOCTL))) {
1608 		return (EFAULT);
1609 	} else {
1610 		*rvalp = count;
1611 		return (0);
1612 	}
1613 }
1614 
1615 /*
1616  * Process STREAM ioctls.
1617  *
1618  * Returns:
1619  *   < 0  - ioctl was not handle
1620  *  >= 0  - ioctl was handled, if > 0, then it is an errno
1621  */
1622 int
1623 socket_strioc_common(struct sonode *so, int cmd, intptr_t arg, int mode,
1624     struct cred *cr, int32_t *rvalp)
1625 {
1626 	int retval;
1627 
1628 	/* Only STREAM iotcls are handled here */
1629 	if ((cmd & 0xffffff00U) != STR)
1630 		return (-1);
1631 
1632 	switch (cmd) {
1633 	case I_CANPUT:
1634 		/*
1635 		 * We return an error for I_CANPUT so that isastream(3C) will
1636 		 * not report the socket as being a STREAM.
1637 		 */
1638 		return (EOPNOTSUPP);
1639 	case I_NREAD:
1640 		/* Avoid doing a fallback for I_NREAD. */
1641 		return (so_strioc_nread(so, arg, mode, rvalp));
1642 	case I_LOOK:
1643 		/* Avoid doing a fallback for I_LOOK. */
1644 		if (so_copyout("sockmod", (void *)arg, strlen("sockmod") + 1,
1645 		    (mode & (int)FKIOCTL))) {
1646 			return (EFAULT);
1647 		}
1648 		return (0);
1649 	default:
1650 		break;
1651 	}
1652 
1653 	/*
1654 	 * Try to fall back to TPI, and if successful, reissue the ioctl.
1655 	 */
1656 	if ((retval = so_tpi_fallback(so, cr)) == 0) {
1657 		/* Reissue the ioctl */
1658 		ASSERT(so->so_rcv_q_head == NULL);
1659 		return (SOP_IOCTL(so, cmd, arg, mode, cr, rvalp));
1660 	} else {
1661 		return (retval);
1662 	}
1663 }
1664 
1665 /*
1666  * This is called for all socket types to verify that the buffer size is large
1667  * enough for the option, and if we can, handle the request as well. Most
1668  * options will be forwarded to the protocol.
1669  */
1670 int
1671 socket_getopt_common(struct sonode *so, int level, int option_name,
1672     void *optval, socklen_t *optlenp, int flags)
1673 {
1674 	if (level != SOL_SOCKET)
1675 		return (-1);
1676 
1677 	switch (option_name) {
1678 	case SO_ERROR:
1679 	case SO_DOMAIN:
1680 	case SO_TYPE:
1681 	case SO_ACCEPTCONN: {
1682 		int32_t value;
1683 		socklen_t optlen = *optlenp;
1684 
1685 		if (optlen < (t_uscalar_t)sizeof (int32_t)) {
1686 			return (EINVAL);
1687 		}
1688 
1689 		switch (option_name) {
1690 		case SO_ERROR:
1691 			mutex_enter(&so->so_lock);
1692 			value = sogeterr(so, B_TRUE);
1693 			mutex_exit(&so->so_lock);
1694 			break;
1695 		case SO_DOMAIN:
1696 			value = so->so_family;
1697 			break;
1698 		case SO_TYPE:
1699 			value = so->so_type;
1700 			break;
1701 		case SO_ACCEPTCONN:
1702 			if (so->so_state & SS_ACCEPTCONN)
1703 				value = SO_ACCEPTCONN;
1704 			else
1705 				value = 0;
1706 			break;
1707 		}
1708 
1709 		bcopy(&value, optval, sizeof (value));
1710 		*optlenp = sizeof (value);
1711 
1712 		return (0);
1713 	}
1714 	case SO_SNDTIMEO:
1715 	case SO_RCVTIMEO: {
1716 		clock_t value;
1717 		socklen_t optlen = *optlenp;
1718 
1719 		if (get_udatamodel() == DATAMODEL_NONE ||
1720 		    get_udatamodel() == DATAMODEL_NATIVE) {
1721 			if (optlen < sizeof (struct timeval))
1722 				return (EINVAL);
1723 		} else {
1724 			if (optlen < sizeof (struct timeval32))
1725 				return (EINVAL);
1726 		}
1727 		if (option_name == SO_RCVTIMEO)
1728 			value = drv_hztousec(so->so_rcvtimeo);
1729 		else
1730 			value = drv_hztousec(so->so_sndtimeo);
1731 
1732 		if (get_udatamodel() == DATAMODEL_NONE ||
1733 		    get_udatamodel() == DATAMODEL_NATIVE) {
1734 			((struct timeval *)(optval))->tv_sec =
1735 			    value / (1000 * 1000);
1736 			((struct timeval *)(optval))->tv_usec =
1737 			    value % (1000 * 1000);
1738 			*optlenp = sizeof (struct timeval);
1739 		} else {
1740 			((struct timeval32 *)(optval))->tv_sec =
1741 			    value / (1000 * 1000);
1742 			((struct timeval32 *)(optval))->tv_usec =
1743 			    value % (1000 * 1000);
1744 			*optlenp = sizeof (struct timeval32);
1745 		}
1746 		return (0);
1747 	}
1748 	case SO_DEBUG:
1749 	case SO_REUSEADDR:
1750 	case SO_KEEPALIVE:
1751 	case SO_DONTROUTE:
1752 	case SO_BROADCAST:
1753 	case SO_USELOOPBACK:
1754 	case SO_OOBINLINE:
1755 	case SO_SNDBUF:
1756 #ifdef notyet
1757 	case SO_SNDLOWAT:
1758 	case SO_RCVLOWAT:
1759 #endif /* notyet */
1760 	case SO_DGRAM_ERRIND: {
1761 		socklen_t optlen = *optlenp;
1762 
1763 		if (optlen < (t_uscalar_t)sizeof (int32_t))
1764 			return (EINVAL);
1765 		break;
1766 	}
1767 	case SO_RCVBUF: {
1768 		socklen_t optlen = *optlenp;
1769 
1770 		if (optlen < (t_uscalar_t)sizeof (int32_t))
1771 			return (EINVAL);
1772 
1773 		if ((flags & _SOGETSOCKOPT_XPG4_2) && so->so_xpg_rcvbuf != 0) {
1774 			/*
1775 			 * XXX If SO_RCVBUF has been set and this is an
1776 			 * XPG 4.2 application then do not ask the transport
1777 			 * since the transport might adjust the value and not
1778 			 * return exactly what was set by the application.
1779 			 * For non-XPG 4.2 application we return the value
1780 			 * that the transport is actually using.
1781 			 */
1782 			*(int32_t *)optval = so->so_xpg_rcvbuf;
1783 			*optlenp = sizeof (so->so_xpg_rcvbuf);
1784 			return (0);
1785 		}
1786 		/*
1787 		 * If the option has not been set then get a default
1788 		 * value from the transport.
1789 		 */
1790 		break;
1791 	}
1792 	case SO_LINGER: {
1793 		socklen_t optlen = *optlenp;
1794 
1795 		if (optlen < (t_uscalar_t)sizeof (struct linger))
1796 			return (EINVAL);
1797 		break;
1798 	}
1799 	case SO_SND_BUFINFO: {
1800 		socklen_t optlen = *optlenp;
1801 
1802 		if (optlen < (t_uscalar_t)sizeof (struct so_snd_bufinfo))
1803 			return (EINVAL);
1804 		((struct so_snd_bufinfo *)(optval))->sbi_wroff =
1805 		    (so->so_proto_props).sopp_wroff;
1806 		((struct so_snd_bufinfo *)(optval))->sbi_maxblk =
1807 		    (so->so_proto_props).sopp_maxblk;
1808 		((struct so_snd_bufinfo *)(optval))->sbi_maxpsz =
1809 		    (so->so_proto_props).sopp_maxpsz;
1810 		((struct so_snd_bufinfo *)(optval))->sbi_tail =
1811 		    (so->so_proto_props).sopp_tail;
1812 		*optlenp = sizeof (struct so_snd_bufinfo);
1813 		return (0);
1814 	}
1815 	default:
1816 		break;
1817 	}
1818 
1819 	/* Unknown Option */
1820 	return (-1);
1821 }
1822 
1823 void
1824 socket_sonode_destroy(struct sonode *so)
1825 {
1826 	sonode_fini(so);
1827 	kmem_cache_free(socket_cache, so);
1828 }
1829 
1830 int
1831 so_zcopy_wait(struct sonode *so)
1832 {
1833 	int error = 0;
1834 
1835 	mutex_enter(&so->so_lock);
1836 	while (!(so->so_copyflag & STZCNOTIFY)) {
1837 		if (so->so_state & SS_CLOSING) {
1838 			mutex_exit(&so->so_lock);
1839 			return (EINTR);
1840 		}
1841 		if (cv_wait_sig(&so->so_copy_cv, &so->so_lock) == 0) {
1842 			error = EINTR;
1843 			break;
1844 		}
1845 	}
1846 	so->so_copyflag &= ~STZCNOTIFY;
1847 	mutex_exit(&so->so_lock);
1848 	return (error);
1849 }
1850 
1851 void
1852 so_timer_callback(void *arg)
1853 {
1854 	struct sonode *so = (struct sonode *)arg;
1855 
1856 	mutex_enter(&so->so_lock);
1857 
1858 	so->so_rcv_timer_tid = 0;
1859 	if (so->so_rcv_queued > 0) {
1860 		so_notify_data(so, so->so_rcv_queued);
1861 	} else {
1862 		mutex_exit(&so->so_lock);
1863 	}
1864 }
1865 
1866 #ifdef DEBUG
1867 /*
1868  * Verify that the length stored in so_rcv_queued and the length of data blocks
1869  * queued is same.
1870  */
1871 static boolean_t
1872 so_check_length(sonode_t *so)
1873 {
1874 	mblk_t *mp = so->so_rcv_q_head;
1875 	int len = 0;
1876 
1877 	ASSERT(MUTEX_HELD(&so->so_lock));
1878 
1879 	if (mp != NULL) {
1880 		len = msgdsize(mp);
1881 		while ((mp = mp->b_next) != NULL)
1882 			len += msgdsize(mp);
1883 	}
1884 	mp = so->so_rcv_head;
1885 	if (mp != NULL) {
1886 		len += msgdsize(mp);
1887 		while ((mp = mp->b_next) != NULL)
1888 			len += msgdsize(mp);
1889 	}
1890 	return ((len == so->so_rcv_queued) ? B_TRUE : B_FALSE);
1891 }
1892 #endif
1893 
1894 int
1895 so_get_mod_version(struct sockparams *sp)
1896 {
1897 	ASSERT(sp != NULL && sp->sp_smod_info != NULL);
1898 	return (sp->sp_smod_info->smod_version);
1899 }
1900 
1901 /*
1902  * so_start_fallback()
1903  *
1904  * Block new socket operations from coming in, and wait for active operations
1905  * to complete. Threads that are sleeping will be woken up so they can get
1906  * out of the way.
1907  *
1908  * The caller must be a reader on so_fallback_rwlock.
1909  */
1910 static boolean_t
1911 so_start_fallback(struct sonode *so)
1912 {
1913 	ASSERT(RW_READ_HELD(&so->so_fallback_rwlock));
1914 
1915 	mutex_enter(&so->so_lock);
1916 	if (so->so_state & SS_FALLBACK_PENDING) {
1917 		mutex_exit(&so->so_lock);
1918 		return (B_FALSE);
1919 	}
1920 	so->so_state |= SS_FALLBACK_PENDING;
1921 	/*
1922 	 * Poke all threads that might be sleeping. Any operation that comes
1923 	 * in after the cv_broadcast will observe the fallback pending flag
1924 	 * which cause the call to return where it would normally sleep.
1925 	 */
1926 	cv_broadcast(&so->so_state_cv);		/* threads in connect() */
1927 	cv_broadcast(&so->so_rcv_cv);		/* threads in recvmsg() */
1928 	cv_broadcast(&so->so_snd_cv);		/* threads in sendmsg() */
1929 	mutex_enter(&so->so_acceptq_lock);
1930 	cv_broadcast(&so->so_acceptq_cv);	/* threads in accept() */
1931 	mutex_exit(&so->so_acceptq_lock);
1932 	mutex_exit(&so->so_lock);
1933 
1934 	/*
1935 	 * The main reason for the rw_tryupgrade call is to provide
1936 	 * observability during the fallback process. We want to
1937 	 * be able to see if there are pending operations.
1938 	 */
1939 	if (rw_tryupgrade(&so->so_fallback_rwlock) == 0) {
1940 		/*
1941 		 * It is safe to drop and reaquire the fallback lock, because
1942 		 * we are guaranteed that another fallback cannot take place.
1943 		 */
1944 		rw_exit(&so->so_fallback_rwlock);
1945 		DTRACE_PROBE1(pending__ops__wait, (struct sonode *), so);
1946 		rw_enter(&so->so_fallback_rwlock, RW_WRITER);
1947 		DTRACE_PROBE1(pending__ops__complete, (struct sonode *), so);
1948 	}
1949 
1950 	return (B_TRUE);
1951 }
1952 
1953 /*
1954  * so_end_fallback()
1955  *
1956  * Allow socket opertions back in.
1957  *
1958  * The caller must be a writer on so_fallback_rwlock.
1959  */
1960 static void
1961 so_end_fallback(struct sonode *so)
1962 {
1963 	ASSERT(RW_ISWRITER(&so->so_fallback_rwlock));
1964 
1965 	mutex_enter(&so->so_lock);
1966 	so->so_state &= ~(SS_FALLBACK_PENDING|SS_FALLBACK_DRAIN);
1967 	mutex_exit(&so->so_lock);
1968 
1969 	rw_downgrade(&so->so_fallback_rwlock);
1970 }
1971 
1972 /*
1973  * so_quiesced_cb()
1974  *
1975  * Callback passed to the protocol during fallback. It is called once
1976  * the endpoint is quiescent.
1977  *
1978  * No requests from the user, no notifications from the protocol, so it
1979  * is safe to synchronize the state. Data can also be moved without
1980  * risk for reordering.
1981  *
1982  * We do not need to hold so_lock, since there can be only one thread
1983  * operating on the sonode.
1984  */
1985 static void
1986 so_quiesced_cb(sock_upper_handle_t sock_handle, queue_t *q,
1987     struct T_capability_ack *tcap, struct sockaddr *laddr, socklen_t laddrlen,
1988     struct sockaddr *faddr, socklen_t faddrlen, short opts)
1989 {
1990 	struct sonode *so = (struct sonode *)sock_handle;
1991 	boolean_t atmark;
1992 
1993 	sotpi_update_state(so, tcap, laddr, laddrlen, faddr, faddrlen, opts);
1994 
1995 	/*
1996 	 * Some protocols do not quiece the data path during fallback. Once
1997 	 * we set the SS_FALLBACK_DRAIN flag any attempt to queue data will
1998 	 * fail and the protocol is responsible for saving the data for later
1999 	 * delivery (i.e., once the fallback has completed).
2000 	 */
2001 	mutex_enter(&so->so_lock);
2002 	so->so_state |= SS_FALLBACK_DRAIN;
2003 	SOCKET_TIMER_CANCEL(so);
2004 	mutex_exit(&so->so_lock);
2005 
2006 	if (so->so_rcv_head != NULL) {
2007 		if (so->so_rcv_q_last_head == NULL)
2008 			so->so_rcv_q_head = so->so_rcv_head;
2009 		else
2010 			so->so_rcv_q_last_head->b_next = so->so_rcv_head;
2011 		so->so_rcv_q_last_head = so->so_rcv_last_head;
2012 	}
2013 
2014 	atmark = (so->so_state & SS_RCVATMARK) != 0;
2015 	/*
2016 	 * Clear any OOB state having to do with pending data. The TPI
2017 	 * code path will set the appropriate oob state when we move the
2018 	 * oob data to the STREAM head. We leave SS_HADOOBDATA since the oob
2019 	 * data has already been consumed.
2020 	 */
2021 	so->so_state &= ~(SS_RCVATMARK|SS_OOBPEND|SS_HAVEOOBDATA);
2022 
2023 	ASSERT(so->so_oobmsg != NULL || so->so_oobmark <= so->so_rcv_queued);
2024 
2025 	/*
2026 	 * Move data to the STREAM head.
2027 	 */
2028 	while (so->so_rcv_q_head != NULL) {
2029 		mblk_t *mp = so->so_rcv_q_head;
2030 		size_t mlen = msgdsize(mp);
2031 
2032 		so->so_rcv_q_head = mp->b_next;
2033 		mp->b_next = NULL;
2034 		mp->b_prev = NULL;
2035 
2036 		/*
2037 		 * Send T_EXDATA_IND if we are at the oob mark.
2038 		 */
2039 		if (atmark) {
2040 			struct T_exdata_ind *tei;
2041 			mblk_t *mp1 = SOTOTPI(so)->sti_exdata_mp;
2042 
2043 			SOTOTPI(so)->sti_exdata_mp = NULL;
2044 			ASSERT(mp1 != NULL);
2045 			mp1->b_datap->db_type = M_PROTO;
2046 			tei = (struct T_exdata_ind *)mp1->b_rptr;
2047 			tei->PRIM_type = T_EXDATA_IND;
2048 			tei->MORE_flag = 0;
2049 			mp1->b_wptr = (uchar_t *)&tei[1];
2050 
2051 			if (IS_SO_OOB_INLINE(so)) {
2052 				mp1->b_cont = mp;
2053 			} else {
2054 				ASSERT(so->so_oobmsg != NULL);
2055 				mp1->b_cont = so->so_oobmsg;
2056 				so->so_oobmsg = NULL;
2057 
2058 				/* process current mp next time around */
2059 				mp->b_next = so->so_rcv_q_head;
2060 				so->so_rcv_q_head = mp;
2061 				mlen = 0;
2062 			}
2063 			mp = mp1;
2064 
2065 			/* we have consumed the oob mark */
2066 			atmark = B_FALSE;
2067 		} else if (so->so_oobmark > 0) {
2068 			/*
2069 			 * Check if the OOB mark is within the current
2070 			 * mblk chain. In that case we have to split it up.
2071 			 */
2072 			if (so->so_oobmark < mlen) {
2073 				mblk_t *urg_mp = mp;
2074 
2075 				atmark = B_TRUE;
2076 				mp = NULL;
2077 				mlen = so->so_oobmark;
2078 
2079 				/*
2080 				 * It is assumed that the OOB mark does
2081 				 * not land within a mblk.
2082 				 */
2083 				do {
2084 					so->so_oobmark -= MBLKL(urg_mp);
2085 					mp = urg_mp;
2086 					urg_mp = urg_mp->b_cont;
2087 				} while (so->so_oobmark > 0);
2088 				mp->b_cont = NULL;
2089 				if (urg_mp != NULL) {
2090 					urg_mp->b_next = so->so_rcv_q_head;
2091 					so->so_rcv_q_head = urg_mp;
2092 				}
2093 			} else {
2094 				so->so_oobmark -= mlen;
2095 				if (so->so_oobmark == 0)
2096 					atmark = B_TRUE;
2097 			}
2098 		}
2099 
2100 		/*
2101 		 * Queue data on the STREAM head.
2102 		 */
2103 		so->so_rcv_queued -= mlen;
2104 		putnext(q, mp);
2105 	}
2106 	so->so_rcv_head = NULL;
2107 	so->so_rcv_last_head = NULL;
2108 	so->so_rcv_q_head = NULL;
2109 	so->so_rcv_q_last_head = NULL;
2110 
2111 	/*
2112 	 * Check if the oob byte is at the end of the data stream, or if the
2113 	 * oob byte has not yet arrived. In the latter case we have to send a
2114 	 * SIGURG and a mark indicator to the STREAM head. The mark indicator
2115 	 * is needed to guarantee correct behavior for SIOCATMARK. See block
2116 	 * comment in socktpi.h for more details.
2117 	 */
2118 	if (atmark || so->so_oobmark > 0) {
2119 		mblk_t *mp;
2120 
2121 		if (atmark && so->so_oobmsg != NULL) {
2122 			struct T_exdata_ind *tei;
2123 
2124 			mp = SOTOTPI(so)->sti_exdata_mp;
2125 			SOTOTPI(so)->sti_exdata_mp = NULL;
2126 			ASSERT(mp != NULL);
2127 			mp->b_datap->db_type = M_PROTO;
2128 			tei = (struct T_exdata_ind *)mp->b_rptr;
2129 			tei->PRIM_type = T_EXDATA_IND;
2130 			tei->MORE_flag = 0;
2131 			mp->b_wptr = (uchar_t *)&tei[1];
2132 
2133 			mp->b_cont = so->so_oobmsg;
2134 			so->so_oobmsg = NULL;
2135 
2136 			putnext(q, mp);
2137 		} else {
2138 			/* Send up the signal */
2139 			mp = SOTOTPI(so)->sti_exdata_mp;
2140 			SOTOTPI(so)->sti_exdata_mp = NULL;
2141 			ASSERT(mp != NULL);
2142 			DB_TYPE(mp) = M_PCSIG;
2143 			*mp->b_wptr++ = (uchar_t)SIGURG;
2144 			putnext(q, mp);
2145 
2146 			/* Send up the mark indicator */
2147 			mp = SOTOTPI(so)->sti_urgmark_mp;
2148 			SOTOTPI(so)->sti_urgmark_mp = NULL;
2149 			mp->b_flag = atmark ? MSGMARKNEXT : MSGNOTMARKNEXT;
2150 			putnext(q, mp);
2151 
2152 			so->so_oobmark = 0;
2153 		}
2154 	}
2155 
2156 	if (SOTOTPI(so)->sti_exdata_mp != NULL) {
2157 		freeb(SOTOTPI(so)->sti_exdata_mp);
2158 		SOTOTPI(so)->sti_exdata_mp = NULL;
2159 	}
2160 
2161 	if (SOTOTPI(so)->sti_urgmark_mp != NULL) {
2162 		freeb(SOTOTPI(so)->sti_urgmark_mp);
2163 		SOTOTPI(so)->sti_urgmark_mp = NULL;
2164 	}
2165 
2166 	ASSERT(so->so_oobmark == 0);
2167 	ASSERT(so->so_rcv_queued == 0);
2168 }
2169 
2170 #ifdef DEBUG
2171 /*
2172  * Do an integrity check of the sonode. This should be done if a
2173  * fallback fails after sonode has initially been converted to use
2174  * TPI and subsequently have to be reverted.
2175  *
2176  * Failure to pass the integrity check will panic the system.
2177  */
2178 void
2179 so_integrity_check(struct sonode *cur, struct sonode *orig)
2180 {
2181 	VERIFY(cur->so_vnode == orig->so_vnode);
2182 	VERIFY(cur->so_ops == orig->so_ops);
2183 	/*
2184 	 * For so_state we can only VERIFY the state flags in CHECK_STATE.
2185 	 * The other state flags might be affected by a notification from the
2186 	 * protocol.
2187 	 */
2188 #define	CHECK_STATE	(SS_CANTRCVMORE|SS_CANTSENDMORE|SS_NDELAY|SS_NONBLOCK| \
2189 	SS_ASYNC|SS_ACCEPTCONN|SS_SAVEDEOR|SS_RCVATMARK|SS_OOBPEND| \
2190 	SS_HAVEOOBDATA|SS_HADOOBDATA|SS_SENTLASTREADSIG|SS_SENTLASTWRITESIG)
2191 	VERIFY((cur->so_state & (orig->so_state & CHECK_STATE)) ==
2192 	    (orig->so_state & CHECK_STATE));
2193 	VERIFY(cur->so_mode == orig->so_mode);
2194 	VERIFY(cur->so_flag == orig->so_flag);
2195 	VERIFY(cur->so_count == orig->so_count);
2196 	/* Cannot VERIFY so_proto_connid; proto can update it */
2197 	VERIFY(cur->so_sockparams == orig->so_sockparams);
2198 	/* an error might have been recorded, but it can not be lost */
2199 	VERIFY(cur->so_error != 0 || orig->so_error == 0);
2200 	VERIFY(cur->so_family == orig->so_family);
2201 	VERIFY(cur->so_type == orig->so_type);
2202 	VERIFY(cur->so_protocol == orig->so_protocol);
2203 	VERIFY(cur->so_version == orig->so_version);
2204 	/* New conns might have arrived, but none should have been lost */
2205 	VERIFY(cur->so_acceptq_len >= orig->so_acceptq_len);
2206 	VERIFY(cur->so_acceptq_head == orig->so_acceptq_head);
2207 	VERIFY(cur->so_backlog == orig->so_backlog);
2208 	/* New OOB migth have arrived, but mark should not have been lost */
2209 	VERIFY(cur->so_oobmark >= orig->so_oobmark);
2210 	/* Cannot VERIFY so_oobmsg; the proto might have sent up a new one */
2211 	VERIFY(cur->so_pgrp == orig->so_pgrp);
2212 	VERIFY(cur->so_peercred == orig->so_peercred);
2213 	VERIFY(cur->so_cpid == orig->so_cpid);
2214 	VERIFY(cur->so_zoneid == orig->so_zoneid);
2215 	/* New data migth have arrived, but none should have been lost */
2216 	VERIFY(cur->so_rcv_queued >= orig->so_rcv_queued);
2217 	VERIFY(cur->so_rcv_q_head == orig->so_rcv_q_head);
2218 	VERIFY(cur->so_rcv_head == orig->so_rcv_head);
2219 	VERIFY(cur->so_proto_handle == orig->so_proto_handle);
2220 	VERIFY(cur->so_downcalls == orig->so_downcalls);
2221 	/* Cannot VERIFY so_proto_props; they can be updated by proto */
2222 }
2223 #endif
2224 
2225 /*
2226  * so_tpi_fallback()
2227  *
2228  * This is the fallback initation routine; things start here.
2229  *
2230  * Basic strategy:
2231  *   o Block new socket operations from coming in
2232  *   o Allocate/initate info needed by TPI
2233  *   o Quiesce the connection, at which point we sync
2234  *     state and move data
2235  *   o Change operations (sonodeops) associated with the socket
2236  *   o Unblock threads waiting for the fallback to finish
2237  */
2238 int
2239 so_tpi_fallback(struct sonode *so, struct cred *cr)
2240 {
2241 	int error;
2242 	queue_t *q;
2243 	struct sockparams *sp;
2244 	struct sockparams *newsp = NULL;
2245 	so_proto_fallback_func_t fbfunc;
2246 	boolean_t direct;
2247 	struct sonode *nso;
2248 #ifdef DEBUG
2249 	struct sonode origso;
2250 #endif
2251 	error = 0;
2252 	sp = so->so_sockparams;
2253 	fbfunc = sp->sp_smod_info->smod_proto_fallback_func;
2254 
2255 	/*
2256 	 * Fallback can only happen if there is a device associated
2257 	 * with the sonode, and the socket module has a fallback function.
2258 	 */
2259 	if (!SOCKPARAMS_HAS_DEVICE(sp) || fbfunc == NULL)
2260 		return (EINVAL);
2261 
2262 	/*
2263 	 * Initiate fallback; upon success we know that no new requests
2264 	 * will come in from the user.
2265 	 */
2266 	if (!so_start_fallback(so))
2267 		return (EAGAIN);
2268 #ifdef DEBUG
2269 	/*
2270 	 * Make a copy of the sonode in case we need to make an integrity
2271 	 * check later on.
2272 	 */
2273 	bcopy(so, &origso, sizeof (*so));
2274 #endif
2275 
2276 	sp->sp_stats.sps_nfallback.value.ui64++;
2277 
2278 	newsp = sockparams_hold_ephemeral_bydev(so->so_family, so->so_type,
2279 	    so->so_protocol, so->so_sockparams->sp_sdev_info.sd_devpath,
2280 	    KM_SLEEP, &error);
2281 	if (error != 0)
2282 		goto out;
2283 
2284 	if (so->so_direct != NULL) {
2285 		sodirect_t *sodp = so->so_direct;
2286 		mutex_enter(&so->so_lock);
2287 
2288 		so->so_direct->sod_enabled = B_FALSE;
2289 		so->so_state &= ~SS_SODIRECT;
2290 		ASSERT(sodp->sod_uioafh == NULL);
2291 		mutex_exit(&so->so_lock);
2292 	}
2293 
2294 	/* Turn sonode into a TPI socket */
2295 	error = sotpi_convert_sonode(so, newsp, &direct, &q, cr);
2296 	if (error != 0)
2297 		goto out;
2298 
2299 
2300 	/*
2301 	 * Now tell the protocol to start using TPI. so_quiesced_cb be
2302 	 * called once it's safe to synchronize state.
2303 	 */
2304 	DTRACE_PROBE1(proto__fallback__begin, struct sonode *, so);
2305 	error = (*fbfunc)(so->so_proto_handle, q, direct, so_quiesced_cb);
2306 	DTRACE_PROBE1(proto__fallback__end, struct sonode *, so);
2307 
2308 	if (error != 0) {
2309 		/* protocol was unable to do a fallback, revert the sonode */
2310 		sotpi_revert_sonode(so, cr);
2311 		goto out;
2312 	}
2313 
2314 	/*
2315 	 * Walk the accept queue and notify the proto that they should
2316 	 * fall back to TPI. The protocol will send up the T_CONN_IND.
2317 	 */
2318 	nso = so->so_acceptq_head;
2319 	while (nso != NULL) {
2320 		int rval;
2321 
2322 		DTRACE_PROBE1(proto__fallback__begin, struct sonode *, nso);
2323 		rval = (*fbfunc)(nso->so_proto_handle, NULL, direct, NULL);
2324 		DTRACE_PROBE1(proto__fallback__end, struct sonode *, nso);
2325 		if (rval != 0) {
2326 			zcmn_err(getzoneid(), CE_WARN,
2327 			    "Failed to convert socket in accept queue to TPI. "
2328 			    "Pid = %d\n", curproc->p_pid);
2329 		}
2330 		nso = nso->so_acceptq_next;
2331 	}
2332 
2333 	/*
2334 	 * Now flush the acceptq, this will destroy all sockets. They will
2335 	 * be recreated in sotpi_accept().
2336 	 */
2337 	so_acceptq_flush(so, B_FALSE);
2338 
2339 	mutex_enter(&so->so_lock);
2340 	so->so_state |= SS_FALLBACK_COMP;
2341 	mutex_exit(&so->so_lock);
2342 
2343 	/*
2344 	 * Swap the sonode ops. Socket opertations that come in once this
2345 	 * is done will proceed without blocking.
2346 	 */
2347 	so->so_ops = &sotpi_sonodeops;
2348 
2349 	/*
2350 	 * Wake up any threads stuck in poll. This is needed since the poll
2351 	 * head changes when the fallback happens (moves from the sonode to
2352 	 * the STREAMS head).
2353 	 */
2354 	pollwakeup(&so->so_poll_list, POLLERR);
2355 out:
2356 	so_end_fallback(so);
2357 
2358 	if (error != 0) {
2359 #ifdef DEBUG
2360 		so_integrity_check(so, &origso);
2361 #endif
2362 		zcmn_err(getzoneid(), CE_WARN,
2363 		    "Failed to convert socket to TPI (err=%d). Pid = %d\n",
2364 		    error, curproc->p_pid);
2365 		if (newsp != NULL)
2366 			SOCKPARAMS_DEC_REF(newsp);
2367 	}
2368 
2369 	return (error);
2370 }
2371