xref: /illumos-gate/usr/src/uts/common/inet/tcp/tcp_fusion.c (revision 4e93fb0f6383eaac21897dcdae56b87118131e4d)
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  * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #pragma ident	"%Z%%M%	%I%	%E% SMI"
27 
28 #include <sys/types.h>
29 #include <sys/stream.h>
30 #include <sys/strsun.h>
31 #include <sys/strsubr.h>
32 #include <sys/debug.h>
33 #include <sys/sdt.h>
34 #include <sys/cmn_err.h>
35 #include <sys/tihdr.h>
36 
37 #include <inet/common.h>
38 #include <inet/ip.h>
39 #include <inet/ip_impl.h>
40 #include <inet/tcp.h>
41 #include <inet/tcp_impl.h>
42 #include <inet/ipsec_impl.h>
43 #include <inet/ipclassifier.h>
44 #include <inet/ipp_common.h>
45 
46 /*
47  * This file implements TCP fusion - a protocol-less data path for TCP
48  * loopback connections.  The fusion of two local TCP endpoints occurs
49  * at connection establishment time.  Various conditions (see details
50  * in tcp_fuse()) need to be met for fusion to be successful.  If it
51  * fails, we fall back to the regular TCP data path; if it succeeds,
52  * both endpoints proceed to use tcp_fuse_output() as the transmit path.
53  * tcp_fuse_output() enqueues application data directly onto the peer's
54  * receive queue; no protocol processing is involved.  After enqueueing
55  * the data, the sender can either push (putnext) data up the receiver's
56  * read queue; or the sender can simply return and let the receiver
57  * retrieve the enqueued data via the synchronous streams entry point
58  * tcp_fuse_rrw().  The latter path is taken if synchronous streams is
59  * enabled (the default).  It is disabled if sockfs no longer resides
60  * directly on top of tcp module due to a module insertion or removal.
61  * It also needs to be temporarily disabled when sending urgent data
62  * because the tcp_fuse_rrw() path bypasses the M_PROTO processing done
63  * by strsock_proto() hook.
64  *
65  * Sychronization is handled by squeue and the mutex tcp_non_sq_lock.
66  * One of the requirements for fusion to succeed is that both endpoints
67  * need to be using the same squeue.  This ensures that neither side
68  * can disappear while the other side is still sending data.  By itself,
69  * squeue is not sufficient for guaranteeing safety when synchronous
70  * streams is enabled.  The reason is that tcp_fuse_rrw() doesn't enter
71  * the squeue and its access to tcp_rcv_list and other fusion-related
72  * fields needs to be sychronized with the sender.  tcp_non_sq_lock is
73  * used for this purpose.  When there is urgent data, the sender needs
74  * to push the data up the receiver's streams read queue.  In order to
75  * avoid holding the tcp_non_sq_lock across putnext(), the sender sets
76  * the peer tcp's tcp_fuse_syncstr_plugged bit and releases tcp_non_sq_lock
77  * (see macro TCP_FUSE_SYNCSTR_PLUG_DRAIN()).  If tcp_fuse_rrw() enters
78  * after this point, it will see that synchronous streams is plugged and
79  * will wait on tcp_fuse_plugcv.  After the sender has finished pushing up
80  * all urgent data, it will clear the tcp_fuse_syncstr_plugged bit using
81  * TCP_FUSE_SYNCSTR_UNPLUG_DRAIN().  This will cause any threads waiting
82  * on tcp_fuse_plugcv to return EBUSY, and in turn cause strget() to call
83  * getq_noenab() to dequeue data from the stream head instead.  Once the
84  * data on the stream head has been consumed, tcp_fuse_rrw() may again
85  * be used to process tcp_rcv_list.  However, if TCP_FUSE_SYNCSTR_STOP()
86  * has been called, all future calls to tcp_fuse_rrw() will return EBUSY,
87  * effectively disabling synchronous streams.
88  *
89  * The following note applies only to the synchronous streams mode.
90  *
91  * Flow control is done by checking the size of receive buffer and
92  * the number of data blocks, both set to different limits.  This is
93  * different than regular streams flow control where cumulative size
94  * check dominates block count check -- streams queue high water mark
95  * typically represents bytes.  Each enqueue triggers notifications
96  * to the receiving process; a build up of data blocks indicates a
97  * slow receiver and the sender should be blocked or informed at the
98  * earliest moment instead of further wasting system resources.  In
99  * effect, this is equivalent to limiting the number of outstanding
100  * segments in flight.
101  */
102 
103 /*
104  * Setting this to false means we disable fusion altogether and
105  * loopback connections would go through the protocol paths.
106  */
107 boolean_t do_tcp_fusion = B_TRUE;
108 
109 /*
110  * Enabling this flag allows sockfs to retrieve data directly
111  * from a fused tcp endpoint using synchronous streams interface.
112  */
113 boolean_t do_tcp_direct_sockfs = B_TRUE;
114 
115 /*
116  * This is the minimum amount of outstanding writes allowed on
117  * a synchronous streams-enabled receiving endpoint before the
118  * sender gets flow-controlled.  Setting this value to 0 means
119  * that the data block limit is equivalent to the byte count
120  * limit, which essentially disables the check.
121  */
122 #define	TCP_FUSION_RCV_UNREAD_MIN	8
123 uint_t tcp_fusion_rcv_unread_min = TCP_FUSION_RCV_UNREAD_MIN;
124 
125 static void	tcp_fuse_syncstr_enable(tcp_t *);
126 static void	tcp_fuse_syncstr_disable(tcp_t *);
127 static void	strrput_sig(queue_t *, boolean_t);
128 
129 /*
130  * Return true if this connection needs some IP functionality
131  */
132 static boolean_t
133 tcp_loopback_needs_ip(tcp_t *tcp, netstack_t *ns)
134 {
135 	ipsec_stack_t	*ipss = ns->netstack_ipsec;
136 
137 	if (tcp->tcp_ipversion == IPV4_VERSION) {
138 		if (tcp->tcp_ip_hdr_len != IP_SIMPLE_HDR_LENGTH)
139 			return (B_TRUE);
140 		if (CONN_OUTBOUND_POLICY_PRESENT(tcp->tcp_connp, ipss))
141 			return (B_TRUE);
142 		if (CONN_INBOUND_POLICY_PRESENT(tcp->tcp_connp, ipss))
143 			return (B_TRUE);
144 	} else {
145 		if (tcp->tcp_ip_hdr_len != IPV6_HDR_LEN)
146 			return (B_TRUE);
147 		if (CONN_OUTBOUND_POLICY_PRESENT_V6(tcp->tcp_connp, ipss))
148 			return (B_TRUE);
149 		if (CONN_INBOUND_POLICY_PRESENT_V6(tcp->tcp_connp, ipss))
150 			return (B_TRUE);
151 	}
152 	if (!CONN_IS_LSO_MD_FASTPATH(tcp->tcp_connp))
153 		return (B_TRUE);
154 	return (B_FALSE);
155 }
156 
157 
158 /*
159  * This routine gets called by the eager tcp upon changing state from
160  * SYN_RCVD to ESTABLISHED.  It fuses a direct path between itself
161  * and the active connect tcp such that the regular tcp processings
162  * may be bypassed under allowable circumstances.  Because the fusion
163  * requires both endpoints to be in the same squeue, it does not work
164  * for simultaneous active connects because there is no easy way to
165  * switch from one squeue to another once the connection is created.
166  * This is different from the eager tcp case where we assign it the
167  * same squeue as the one given to the active connect tcp during open.
168  */
169 void
170 tcp_fuse(tcp_t *tcp, uchar_t *iphdr, tcph_t *tcph)
171 {
172 	conn_t *peer_connp, *connp = tcp->tcp_connp;
173 	tcp_t *peer_tcp;
174 	tcp_stack_t	*tcps = tcp->tcp_tcps;
175 	netstack_t	*ns;
176 	ip_stack_t	*ipst = tcps->tcps_netstack->netstack_ip;
177 
178 	ASSERT(!tcp->tcp_fused);
179 	ASSERT(tcp->tcp_loopback);
180 	ASSERT(tcp->tcp_loopback_peer == NULL);
181 	/*
182 	 * We need to inherit q_hiwat of the listener tcp, but we can't
183 	 * really use tcp_listener since we get here after sending up
184 	 * T_CONN_IND and tcp_wput_accept() may be called independently,
185 	 * at which point tcp_listener is cleared; this is why we use
186 	 * tcp_saved_listener.  The listener itself is guaranteed to be
187 	 * around until tcp_accept_finish() is called on this eager --
188 	 * this won't happen until we're done since we're inside the
189 	 * eager's perimeter now.
190 	 */
191 	ASSERT(tcp->tcp_saved_listener != NULL);
192 
193 	/*
194 	 * Lookup peer endpoint; search for the remote endpoint having
195 	 * the reversed address-port quadruplet in ESTABLISHED state,
196 	 * which is guaranteed to be unique in the system.  Zone check
197 	 * is applied accordingly for loopback address, but not for
198 	 * local address since we want fusion to happen across Zones.
199 	 */
200 	if (tcp->tcp_ipversion == IPV4_VERSION) {
201 		peer_connp = ipcl_conn_tcp_lookup_reversed_ipv4(connp,
202 		    (ipha_t *)iphdr, tcph, ipst);
203 	} else {
204 		peer_connp = ipcl_conn_tcp_lookup_reversed_ipv6(connp,
205 		    (ip6_t *)iphdr, tcph, ipst);
206 	}
207 
208 	/*
209 	 * We can only proceed if peer exists, resides in the same squeue
210 	 * as our conn and is not raw-socket.  The squeue assignment of
211 	 * this eager tcp was done earlier at the time of SYN processing
212 	 * in ip_fanout_tcp{_v6}.  Note that similar squeues by itself
213 	 * doesn't guarantee a safe condition to fuse, hence we perform
214 	 * additional tests below.
215 	 */
216 	ASSERT(peer_connp == NULL || peer_connp != connp);
217 	if (peer_connp == NULL || peer_connp->conn_sqp != connp->conn_sqp ||
218 	    !IPCL_IS_TCP(peer_connp)) {
219 		if (peer_connp != NULL) {
220 			TCP_STAT(tcps, tcp_fusion_unqualified);
221 			CONN_DEC_REF(peer_connp);
222 		}
223 		return;
224 	}
225 	peer_tcp = peer_connp->conn_tcp;	/* active connect tcp */
226 
227 	ASSERT(peer_tcp != NULL && peer_tcp != tcp && !peer_tcp->tcp_fused);
228 	ASSERT(peer_tcp->tcp_loopback && peer_tcp->tcp_loopback_peer == NULL);
229 	ASSERT(peer_connp->conn_sqp == connp->conn_sqp);
230 
231 	/*
232 	 * Fuse the endpoints; we perform further checks against both
233 	 * tcp endpoints to ensure that a fusion is allowed to happen.
234 	 * In particular we bail out for non-simple TCP/IP or if IPsec/
235 	 * IPQoS policy/kernel SSL exists.
236 	 */
237 	ns = tcps->tcps_netstack;
238 	ipst = ns->netstack_ip;
239 
240 	if (!tcp->tcp_unfusable && !peer_tcp->tcp_unfusable &&
241 	    !tcp_loopback_needs_ip(tcp, ns) &&
242 	    !tcp_loopback_needs_ip(peer_tcp, ns) &&
243 	    tcp->tcp_kssl_ent == NULL &&
244 	    !IPP_ENABLED(IPP_LOCAL_OUT|IPP_LOCAL_IN, ipst)) {
245 		mblk_t *mp;
246 		struct stroptions *stropt;
247 		queue_t *peer_rq = peer_tcp->tcp_rq;
248 
249 		ASSERT(!TCP_IS_DETACHED(peer_tcp) && peer_rq != NULL);
250 		ASSERT(tcp->tcp_fused_sigurg_mp == NULL);
251 		ASSERT(peer_tcp->tcp_fused_sigurg_mp == NULL);
252 		ASSERT(tcp->tcp_kssl_ctx == NULL);
253 
254 		/*
255 		 * We need to drain data on both endpoints during unfuse.
256 		 * If we need to send up SIGURG at the time of draining,
257 		 * we want to be sure that an mblk is readily available.
258 		 * This is why we pre-allocate the M_PCSIG mblks for both
259 		 * endpoints which will only be used during/after unfuse.
260 		 */
261 		if ((mp = allocb(1, BPRI_HI)) == NULL)
262 			goto failed;
263 
264 		tcp->tcp_fused_sigurg_mp = mp;
265 
266 		if ((mp = allocb(1, BPRI_HI)) == NULL)
267 			goto failed;
268 
269 		peer_tcp->tcp_fused_sigurg_mp = mp;
270 
271 		/* Allocate M_SETOPTS mblk */
272 		if ((mp = allocb(sizeof (*stropt), BPRI_HI)) == NULL)
273 			goto failed;
274 
275 		/* Fuse both endpoints */
276 		peer_tcp->tcp_loopback_peer = tcp;
277 		tcp->tcp_loopback_peer = peer_tcp;
278 		peer_tcp->tcp_fused = tcp->tcp_fused = B_TRUE;
279 
280 		/*
281 		 * We never use regular tcp paths in fusion and should
282 		 * therefore clear tcp_unsent on both endpoints.  Having
283 		 * them set to non-zero values means asking for trouble
284 		 * especially after unfuse, where we may end up sending
285 		 * through regular tcp paths which expect xmit_list and
286 		 * friends to be correctly setup.
287 		 */
288 		peer_tcp->tcp_unsent = tcp->tcp_unsent = 0;
289 
290 		tcp_timers_stop(tcp);
291 		tcp_timers_stop(peer_tcp);
292 
293 		/*
294 		 * At this point we are a detached eager tcp and therefore
295 		 * don't have a queue assigned to us until accept happens.
296 		 * In the mean time the peer endpoint may immediately send
297 		 * us data as soon as fusion is finished, and we need to be
298 		 * able to flow control it in case it sends down huge amount
299 		 * of data while we're still detached.  To prevent that we
300 		 * inherit the listener's q_hiwat value; this is temporary
301 		 * since we'll repeat the process in tcp_accept_finish().
302 		 */
303 		(void) tcp_fuse_set_rcv_hiwat(tcp,
304 		    tcp->tcp_saved_listener->tcp_rq->q_hiwat);
305 
306 		/*
307 		 * Set the stream head's write offset value to zero since we
308 		 * won't be needing any room for TCP/IP headers; tell it to
309 		 * not break up the writes (this would reduce the amount of
310 		 * work done by kmem); and configure our receive buffer.
311 		 * Note that we can only do this for the active connect tcp
312 		 * since our eager is still detached; it will be dealt with
313 		 * later in tcp_accept_finish().
314 		 */
315 		DB_TYPE(mp) = M_SETOPTS;
316 		mp->b_wptr += sizeof (*stropt);
317 
318 		stropt = (struct stroptions *)mp->b_rptr;
319 		stropt->so_flags = SO_MAXBLK | SO_WROFF | SO_HIWAT;
320 		stropt->so_maxblk = tcp_maxpsz_set(peer_tcp, B_FALSE);
321 		stropt->so_wroff = 0;
322 
323 		/*
324 		 * Record the stream head's high water mark for
325 		 * peer endpoint; this is used for flow-control
326 		 * purposes in tcp_fuse_output().
327 		 */
328 		stropt->so_hiwat = tcp_fuse_set_rcv_hiwat(peer_tcp,
329 		    peer_rq->q_hiwat);
330 
331 		/* Send the options up */
332 		putnext(peer_rq, mp);
333 	} else {
334 		TCP_STAT(tcps, tcp_fusion_unqualified);
335 	}
336 	CONN_DEC_REF(peer_connp);
337 	return;
338 
339 failed:
340 	if (tcp->tcp_fused_sigurg_mp != NULL) {
341 		freeb(tcp->tcp_fused_sigurg_mp);
342 		tcp->tcp_fused_sigurg_mp = NULL;
343 	}
344 	if (peer_tcp->tcp_fused_sigurg_mp != NULL) {
345 		freeb(peer_tcp->tcp_fused_sigurg_mp);
346 		peer_tcp->tcp_fused_sigurg_mp = NULL;
347 	}
348 	CONN_DEC_REF(peer_connp);
349 }
350 
351 /*
352  * Unfuse a previously-fused pair of tcp loopback endpoints.
353  */
354 void
355 tcp_unfuse(tcp_t *tcp)
356 {
357 	tcp_t *peer_tcp = tcp->tcp_loopback_peer;
358 
359 	ASSERT(tcp->tcp_fused && peer_tcp != NULL);
360 	ASSERT(peer_tcp->tcp_fused && peer_tcp->tcp_loopback_peer == tcp);
361 	ASSERT(tcp->tcp_connp->conn_sqp == peer_tcp->tcp_connp->conn_sqp);
362 	ASSERT(tcp->tcp_unsent == 0 && peer_tcp->tcp_unsent == 0);
363 	ASSERT(tcp->tcp_fused_sigurg_mp != NULL);
364 	ASSERT(peer_tcp->tcp_fused_sigurg_mp != NULL);
365 
366 	/*
367 	 * We disable synchronous streams, drain any queued data and
368 	 * clear tcp_direct_sockfs.  The synchronous streams entry
369 	 * points will become no-ops after this point.
370 	 */
371 	tcp_fuse_disable_pair(tcp, B_TRUE);
372 
373 	/*
374 	 * Update th_seq and th_ack in the header template
375 	 */
376 	U32_TO_ABE32(tcp->tcp_snxt, tcp->tcp_tcph->th_seq);
377 	U32_TO_ABE32(tcp->tcp_rnxt, tcp->tcp_tcph->th_ack);
378 	U32_TO_ABE32(peer_tcp->tcp_snxt, peer_tcp->tcp_tcph->th_seq);
379 	U32_TO_ABE32(peer_tcp->tcp_rnxt, peer_tcp->tcp_tcph->th_ack);
380 
381 	/* Unfuse the endpoints */
382 	peer_tcp->tcp_fused = tcp->tcp_fused = B_FALSE;
383 	peer_tcp->tcp_loopback_peer = tcp->tcp_loopback_peer = NULL;
384 }
385 
386 /*
387  * Fusion output routine for urgent data.  This routine is called by
388  * tcp_fuse_output() for handling non-M_DATA mblks.
389  */
390 void
391 tcp_fuse_output_urg(tcp_t *tcp, mblk_t *mp)
392 {
393 	mblk_t *mp1;
394 	struct T_exdata_ind *tei;
395 	tcp_t *peer_tcp = tcp->tcp_loopback_peer;
396 	mblk_t *head, *prev_head = NULL;
397 	tcp_stack_t	*tcps = tcp->tcp_tcps;
398 
399 	ASSERT(tcp->tcp_fused);
400 	ASSERT(peer_tcp != NULL && peer_tcp->tcp_loopback_peer == tcp);
401 	ASSERT(DB_TYPE(mp) == M_PROTO || DB_TYPE(mp) == M_PCPROTO);
402 	ASSERT(mp->b_cont != NULL && DB_TYPE(mp->b_cont) == M_DATA);
403 	ASSERT(MBLKL(mp) >= sizeof (*tei) && MBLKL(mp->b_cont) > 0);
404 
405 	/*
406 	 * Urgent data arrives in the form of T_EXDATA_REQ from above.
407 	 * Each occurence denotes a new urgent pointer.  For each new
408 	 * urgent pointer we signal (SIGURG) the receiving app to indicate
409 	 * that it needs to go into urgent mode.  This is similar to the
410 	 * urgent data handling in the regular tcp.  We don't need to keep
411 	 * track of where the urgent pointer is, because each T_EXDATA_REQ
412 	 * "advances" the urgent pointer for us.
413 	 *
414 	 * The actual urgent data carried by T_EXDATA_REQ is then prepended
415 	 * by a T_EXDATA_IND before being enqueued behind any existing data
416 	 * destined for the receiving app.  There is only a single urgent
417 	 * pointer (out-of-band mark) for a given tcp.  If the new urgent
418 	 * data arrives before the receiving app reads some existing urgent
419 	 * data, the previous marker is lost.  This behavior is emulated
420 	 * accordingly below, by removing any existing T_EXDATA_IND messages
421 	 * and essentially converting old urgent data into non-urgent.
422 	 */
423 	ASSERT(tcp->tcp_valid_bits & TCP_URG_VALID);
424 	/* Let sender get out of urgent mode */
425 	tcp->tcp_valid_bits &= ~TCP_URG_VALID;
426 
427 	/*
428 	 * This flag indicates that a signal needs to be sent up.
429 	 * This flag will only get cleared once SIGURG is delivered and
430 	 * is not affected by the tcp_fused flag -- delivery will still
431 	 * happen even after an endpoint is unfused, to handle the case
432 	 * where the sending endpoint immediately closes/unfuses after
433 	 * sending urgent data and the accept is not yet finished.
434 	 */
435 	peer_tcp->tcp_fused_sigurg = B_TRUE;
436 
437 	/* Reuse T_EXDATA_REQ mblk for T_EXDATA_IND */
438 	DB_TYPE(mp) = M_PROTO;
439 	tei = (struct T_exdata_ind *)mp->b_rptr;
440 	tei->PRIM_type = T_EXDATA_IND;
441 	tei->MORE_flag = 0;
442 	mp->b_wptr = (uchar_t *)&tei[1];
443 
444 	TCP_STAT(tcps, tcp_fusion_urg);
445 	BUMP_MIB(&tcps->tcps_mib, tcpOutUrg);
446 
447 	head = peer_tcp->tcp_rcv_list;
448 	while (head != NULL) {
449 		/*
450 		 * Remove existing T_EXDATA_IND, keep the data which follows
451 		 * it and relink our list.  Note that we don't modify the
452 		 * tcp_rcv_last_tail since it never points to T_EXDATA_IND.
453 		 */
454 		if (DB_TYPE(head) != M_DATA) {
455 			mp1 = head;
456 
457 			ASSERT(DB_TYPE(mp1->b_cont) == M_DATA);
458 			head = mp1->b_cont;
459 			mp1->b_cont = NULL;
460 			head->b_next = mp1->b_next;
461 			mp1->b_next = NULL;
462 			if (prev_head != NULL)
463 				prev_head->b_next = head;
464 			if (peer_tcp->tcp_rcv_list == mp1)
465 				peer_tcp->tcp_rcv_list = head;
466 			if (peer_tcp->tcp_rcv_last_head == mp1)
467 				peer_tcp->tcp_rcv_last_head = head;
468 			freeb(mp1);
469 		}
470 		prev_head = head;
471 		head = head->b_next;
472 	}
473 }
474 
475 /*
476  * Fusion output routine, called by tcp_output() and tcp_wput_proto().
477  * If we are modifying any member that can be changed outside the squeue,
478  * like tcp_flow_stopped, we need to take tcp_non_sq_lock.
479  */
480 boolean_t
481 tcp_fuse_output(tcp_t *tcp, mblk_t *mp, uint32_t send_size)
482 {
483 	tcp_t *peer_tcp = tcp->tcp_loopback_peer;
484 	uint_t max_unread;
485 	boolean_t flow_stopped;
486 	boolean_t urgent = (DB_TYPE(mp) != M_DATA);
487 	mblk_t *mp1 = mp;
488 	ill_t *ilp, *olp;
489 	ipha_t *ipha;
490 	ip6_t *ip6h;
491 	tcph_t *tcph;
492 	uint_t ip_hdr_len;
493 	uint32_t seq;
494 	uint32_t recv_size = send_size;
495 	tcp_stack_t	*tcps = tcp->tcp_tcps;
496 	netstack_t	*ns = tcps->tcps_netstack;
497 	ip_stack_t	*ipst = ns->netstack_ip;
498 
499 	ASSERT(tcp->tcp_fused);
500 	ASSERT(peer_tcp != NULL && peer_tcp->tcp_loopback_peer == tcp);
501 	ASSERT(tcp->tcp_connp->conn_sqp == peer_tcp->tcp_connp->conn_sqp);
502 	ASSERT(DB_TYPE(mp) == M_DATA || DB_TYPE(mp) == M_PROTO ||
503 	    DB_TYPE(mp) == M_PCPROTO);
504 
505 	max_unread = peer_tcp->tcp_fuse_rcv_unread_hiwater;
506 
507 	/* If this connection requires IP, unfuse and use regular path */
508 	if (tcp_loopback_needs_ip(tcp, ns) ||
509 	    tcp_loopback_needs_ip(peer_tcp, ns) ||
510 	    IPP_ENABLED(IPP_LOCAL_OUT|IPP_LOCAL_IN, ipst)) {
511 		TCP_STAT(tcps, tcp_fusion_aborted);
512 		goto unfuse;
513 	}
514 
515 	if (send_size == 0) {
516 		freemsg(mp);
517 		return (B_TRUE);
518 	}
519 
520 	/*
521 	 * Handle urgent data; we either send up SIGURG to the peer now
522 	 * or do it later when we drain, in case the peer is detached
523 	 * or if we're short of memory for M_PCSIG mblk.
524 	 */
525 	if (urgent) {
526 		/*
527 		 * We stop synchronous streams when we have urgent data
528 		 * queued to prevent tcp_fuse_rrw() from pulling it.  If
529 		 * for some reasons the urgent data can't be delivered
530 		 * below, synchronous streams will remain stopped until
531 		 * someone drains the tcp_rcv_list.
532 		 */
533 		TCP_FUSE_SYNCSTR_PLUG_DRAIN(peer_tcp);
534 		tcp_fuse_output_urg(tcp, mp);
535 
536 		mp1 = mp->b_cont;
537 	}
538 
539 	if (tcp->tcp_ipversion == IPV4_VERSION &&
540 	    (HOOKS4_INTERESTED_LOOPBACK_IN(ipst) ||
541 	    HOOKS4_INTERESTED_LOOPBACK_OUT(ipst)) ||
542 	    tcp->tcp_ipversion == IPV6_VERSION &&
543 	    (HOOKS6_INTERESTED_LOOPBACK_IN(ipst) ||
544 	    HOOKS6_INTERESTED_LOOPBACK_OUT(ipst))) {
545 		/*
546 		 * Build ip and tcp header to satisfy FW_HOOKS.
547 		 * We only build it when any hook is present.
548 		 */
549 		if ((mp1 = tcp_xmit_mp(tcp, mp1, tcp->tcp_mss, NULL, NULL,
550 		    tcp->tcp_snxt, B_TRUE, NULL, B_FALSE)) == NULL)
551 			/* If tcp_xmit_mp fails, use regular path */
552 			goto unfuse;
553 
554 		ASSERT(peer_tcp->tcp_connp->conn_ire_cache->ire_ipif != NULL);
555 		olp = peer_tcp->tcp_connp->conn_ire_cache->ire_ipif->ipif_ill;
556 		/* PFHooks: LOOPBACK_OUT */
557 		if (tcp->tcp_ipversion == IPV4_VERSION) {
558 			ipha = (ipha_t *)mp1->b_rptr;
559 
560 			DTRACE_PROBE4(ip4__loopback__out__start,
561 			    ill_t *, NULL, ill_t *, olp,
562 			    ipha_t *, ipha, mblk_t *, mp1);
563 			FW_HOOKS(ipst->ips_ip4_loopback_out_event,
564 			    ipst->ips_ipv4firewall_loopback_out,
565 			    NULL, olp, ipha, mp1, mp1, ipst);
566 			DTRACE_PROBE1(ip4__loopback__out__end, mblk_t *, mp1);
567 		} else {
568 			ip6h = (ip6_t *)mp1->b_rptr;
569 
570 			DTRACE_PROBE4(ip6__loopback__out__start,
571 			    ill_t *, NULL, ill_t *, olp,
572 			    ip6_t *, ip6h, mblk_t *, mp1);
573 			FW_HOOKS6(ipst->ips_ip6_loopback_out_event,
574 			    ipst->ips_ipv6firewall_loopback_out,
575 			    NULL, olp, ip6h, mp1, mp1, ipst);
576 			DTRACE_PROBE1(ip6__loopback__out__end, mblk_t *, mp1);
577 		}
578 		if (mp1 == NULL)
579 			goto unfuse;
580 
581 
582 		/* PFHooks: LOOPBACK_IN */
583 		ASSERT(tcp->tcp_connp->conn_ire_cache->ire_ipif != NULL);
584 		ilp = tcp->tcp_connp->conn_ire_cache->ire_ipif->ipif_ill;
585 
586 		if (tcp->tcp_ipversion == IPV4_VERSION) {
587 			DTRACE_PROBE4(ip4__loopback__in__start,
588 			    ill_t *, ilp, ill_t *, NULL,
589 			    ipha_t *, ipha, mblk_t *, mp1);
590 			FW_HOOKS(ipst->ips_ip4_loopback_in_event,
591 			    ipst->ips_ipv4firewall_loopback_in,
592 			    ilp, NULL, ipha, mp1, mp1, ipst);
593 			DTRACE_PROBE1(ip4__loopback__in__end, mblk_t *, mp1);
594 			if (mp1 == NULL)
595 				goto unfuse;
596 
597 			ip_hdr_len = IPH_HDR_LENGTH(ipha);
598 		} else {
599 			DTRACE_PROBE4(ip6__loopback__in__start,
600 			    ill_t *, ilp, ill_t *, NULL,
601 			    ip6_t *, ip6h, mblk_t *, mp1);
602 			FW_HOOKS6(ipst->ips_ip6_loopback_in_event,
603 			    ipst->ips_ipv6firewall_loopback_in,
604 			    ilp, NULL, ip6h, mp1, mp1, ipst);
605 			DTRACE_PROBE1(ip6__loopback__in__end, mblk_t *, mp1);
606 			if (mp1 == NULL)
607 				goto unfuse;
608 
609 			ip_hdr_len = ip_hdr_length_v6(mp1, ip6h);
610 		}
611 
612 		/* Data length might be changed by FW_HOOKS */
613 		tcph = (tcph_t *)&mp1->b_rptr[ip_hdr_len];
614 		seq = ABE32_TO_U32(tcph->th_seq);
615 		recv_size += seq - tcp->tcp_snxt;
616 
617 		/*
618 		 * The message duplicated by tcp_xmit_mp is freed.
619 		 * Note: the original message passed in remains unchanged.
620 		 */
621 		freemsg(mp1);
622 	}
623 
624 	mutex_enter(&peer_tcp->tcp_non_sq_lock);
625 	/*
626 	 * Wake up and signal the peer; it is okay to do this before
627 	 * enqueueing because we are holding the lock.  One of the
628 	 * advantages of synchronous streams is the ability for us to
629 	 * find out when the application performs a read on the socket,
630 	 * by way of tcp_fuse_rrw() entry point being called.  Every
631 	 * data that gets enqueued onto the receiver is treated as if
632 	 * it has arrived at the receiving endpoint, thus generating
633 	 * SIGPOLL/SIGIO for asynchronous socket just as in the strrput()
634 	 * case.  However, we only wake up the application when necessary,
635 	 * i.e. during the first enqueue.  When tcp_fuse_rrw() is called
636 	 * it will send everything upstream.
637 	 */
638 	if (peer_tcp->tcp_direct_sockfs && !urgent &&
639 	    !TCP_IS_DETACHED(peer_tcp)) {
640 		if (peer_tcp->tcp_rcv_list == NULL)
641 			STR_WAKEUP_SET(STREAM(peer_tcp->tcp_rq));
642 		/* Update poll events and send SIGPOLL/SIGIO if necessary */
643 		STR_SENDSIG(STREAM(peer_tcp->tcp_rq));
644 	}
645 
646 	/*
647 	 * Enqueue data into the peer's receive list; we may or may not
648 	 * drain the contents depending on the conditions below.
649 	 */
650 	tcp_rcv_enqueue(peer_tcp, mp, recv_size);
651 
652 	/* In case it wrapped around and also to keep it constant */
653 	peer_tcp->tcp_rwnd += recv_size;
654 
655 	/*
656 	 * Exercise flow-control when needed; we will get back-enabled
657 	 * in either tcp_accept_finish(), tcp_unfuse(), or tcp_fuse_rrw().
658 	 * If tcp_direct_sockfs is on or if the peer endpoint is detached,
659 	 * we emulate streams flow control by checking the peer's queue
660 	 * size and high water mark; otherwise we simply use canputnext()
661 	 * to decide if we need to stop our flow.
662 	 *
663 	 * The outstanding unread data block check does not apply for a
664 	 * detached receiver; this is to avoid unnecessary blocking of the
665 	 * sender while the accept is currently in progress and is quite
666 	 * similar to the regular tcp.
667 	 */
668 	if (TCP_IS_DETACHED(peer_tcp) || max_unread == 0)
669 		max_unread = UINT_MAX;
670 
671 	/*
672 	 * Since we are accessing our tcp_flow_stopped and might modify it,
673 	 * we need to take tcp->tcp_non_sq_lock. The lock for the highest
674 	 * address is held first. Dropping peer_tcp->tcp_non_sq_lock should
675 	 * not be an issue here since we are within the squeue and the peer
676 	 * won't disappear.
677 	 */
678 	if (tcp > peer_tcp) {
679 		mutex_exit(&peer_tcp->tcp_non_sq_lock);
680 		mutex_enter(&tcp->tcp_non_sq_lock);
681 		mutex_enter(&peer_tcp->tcp_non_sq_lock);
682 	} else {
683 		mutex_enter(&tcp->tcp_non_sq_lock);
684 	}
685 	flow_stopped = tcp->tcp_flow_stopped;
686 	if (!flow_stopped &&
687 	    (((peer_tcp->tcp_direct_sockfs || TCP_IS_DETACHED(peer_tcp)) &&
688 	    (peer_tcp->tcp_rcv_cnt >= peer_tcp->tcp_fuse_rcv_hiwater ||
689 	    ++peer_tcp->tcp_fuse_rcv_unread_cnt >= max_unread)) ||
690 	    (!peer_tcp->tcp_direct_sockfs &&
691 	    !TCP_IS_DETACHED(peer_tcp) && !canputnext(peer_tcp->tcp_rq)))) {
692 		tcp_setqfull(tcp);
693 		flow_stopped = B_TRUE;
694 		TCP_STAT(tcps, tcp_fusion_flowctl);
695 		DTRACE_PROBE4(tcp__fuse__output__flowctl, tcp_t *, tcp,
696 		    uint_t, send_size, uint_t, peer_tcp->tcp_rcv_cnt,
697 		    uint_t, peer_tcp->tcp_fuse_rcv_unread_cnt);
698 	} else if (flow_stopped &&
699 	    TCP_UNSENT_BYTES(tcp) <= tcp->tcp_xmit_lowater) {
700 		tcp_clrqfull(tcp);
701 		flow_stopped = B_FALSE;
702 	}
703 	mutex_exit(&tcp->tcp_non_sq_lock);
704 	ipst->ips_loopback_packets++;
705 	tcp->tcp_last_sent_len = send_size;
706 
707 	/* Need to adjust the following SNMP MIB-related variables */
708 	tcp->tcp_snxt += send_size;
709 	tcp->tcp_suna = tcp->tcp_snxt;
710 	peer_tcp->tcp_rnxt += recv_size;
711 	peer_tcp->tcp_rack = peer_tcp->tcp_rnxt;
712 
713 	BUMP_MIB(&tcps->tcps_mib, tcpOutDataSegs);
714 	UPDATE_MIB(&tcps->tcps_mib, tcpOutDataBytes, send_size);
715 
716 	BUMP_MIB(&tcps->tcps_mib, tcpInSegs);
717 	BUMP_MIB(&tcps->tcps_mib, tcpInDataInorderSegs);
718 	UPDATE_MIB(&tcps->tcps_mib, tcpInDataInorderBytes, send_size);
719 
720 	BUMP_LOCAL(tcp->tcp_obsegs);
721 	BUMP_LOCAL(peer_tcp->tcp_ibsegs);
722 
723 	mutex_exit(&peer_tcp->tcp_non_sq_lock);
724 
725 	DTRACE_PROBE2(tcp__fuse__output, tcp_t *, tcp, uint_t, send_size);
726 
727 	if (!TCP_IS_DETACHED(peer_tcp)) {
728 		/*
729 		 * Drain the peer's receive queue it has urgent data or if
730 		 * we're not flow-controlled.  There is no need for draining
731 		 * normal data when tcp_direct_sockfs is on because the peer
732 		 * will pull the data via tcp_fuse_rrw().
733 		 */
734 		if (urgent || (!flow_stopped && !peer_tcp->tcp_direct_sockfs)) {
735 			ASSERT(peer_tcp->tcp_rcv_list != NULL);
736 			/*
737 			 * For TLI-based streams, a thread in tcp_accept_swap()
738 			 * can race with us.  That thread will ensure that the
739 			 * correct peer_tcp->tcp_rq is globally visible before
740 			 * peer_tcp->tcp_detached is visible as clear, but we
741 			 * must also ensure that the load of tcp_rq cannot be
742 			 * reordered to be before the tcp_detached check.
743 			 */
744 			membar_consumer();
745 			(void) tcp_fuse_rcv_drain(peer_tcp->tcp_rq, peer_tcp,
746 			    NULL);
747 			/*
748 			 * If synchronous streams was stopped above due
749 			 * to the presence of urgent data, re-enable it.
750 			 */
751 			if (urgent)
752 				TCP_FUSE_SYNCSTR_UNPLUG_DRAIN(peer_tcp);
753 		}
754 	}
755 	return (B_TRUE);
756 unfuse:
757 	tcp_unfuse(tcp);
758 	return (B_FALSE);
759 }
760 
761 /*
762  * This routine gets called to deliver data upstream on a fused or
763  * previously fused tcp loopback endpoint; the latter happens only
764  * when there is a pending SIGURG signal plus urgent data that can't
765  * be sent upstream in the past.
766  */
767 boolean_t
768 tcp_fuse_rcv_drain(queue_t *q, tcp_t *tcp, mblk_t **sigurg_mpp)
769 {
770 	mblk_t *mp;
771 #ifdef DEBUG
772 	uint_t cnt = 0;
773 #endif
774 	tcp_stack_t	*tcps = tcp->tcp_tcps;
775 
776 	ASSERT(tcp->tcp_loopback);
777 	ASSERT(tcp->tcp_fused || tcp->tcp_fused_sigurg);
778 	ASSERT(!tcp->tcp_fused || tcp->tcp_loopback_peer != NULL);
779 	ASSERT(sigurg_mpp != NULL || tcp->tcp_fused);
780 
781 	/* No need for the push timer now, in case it was scheduled */
782 	if (tcp->tcp_push_tid != 0) {
783 		(void) TCP_TIMER_CANCEL(tcp, tcp->tcp_push_tid);
784 		tcp->tcp_push_tid = 0;
785 	}
786 	/*
787 	 * If there's urgent data sitting in receive list and we didn't
788 	 * get a chance to send up a SIGURG signal, make sure we send
789 	 * it first before draining in order to ensure that SIOCATMARK
790 	 * works properly.
791 	 */
792 	if (tcp->tcp_fused_sigurg) {
793 		/*
794 		 * sigurg_mpp is normally NULL, i.e. when we're still
795 		 * fused and didn't get here because of tcp_unfuse().
796 		 * In this case try hard to allocate the M_PCSIG mblk.
797 		 */
798 		if (sigurg_mpp == NULL &&
799 		    (mp = allocb(1, BPRI_HI)) == NULL &&
800 		    (mp = allocb_tryhard(1)) == NULL) {
801 			/* Alloc failed; try again next time */
802 			tcp->tcp_push_tid = TCP_TIMER(tcp, tcp_push_timer,
803 			    MSEC_TO_TICK(tcps->tcps_push_timer_interval));
804 			return (B_TRUE);
805 		} else if (sigurg_mpp != NULL) {
806 			/*
807 			 * Use the supplied M_PCSIG mblk; it means we're
808 			 * either unfused or in the process of unfusing,
809 			 * and the drain must happen now.
810 			 */
811 			mp = *sigurg_mpp;
812 			*sigurg_mpp = NULL;
813 		}
814 		ASSERT(mp != NULL);
815 
816 		tcp->tcp_fused_sigurg = B_FALSE;
817 		/* Send up the signal */
818 		DB_TYPE(mp) = M_PCSIG;
819 		*mp->b_wptr++ = (uchar_t)SIGURG;
820 		putnext(q, mp);
821 		/*
822 		 * Let the regular tcp_rcv_drain() path handle
823 		 * draining the data if we're no longer fused.
824 		 */
825 		if (!tcp->tcp_fused)
826 			return (B_FALSE);
827 	}
828 
829 	/*
830 	 * In the synchronous streams case, we generate SIGPOLL/SIGIO for
831 	 * each M_DATA that gets enqueued onto the receiver.  At this point
832 	 * we are about to drain any queued data via putnext().  In order
833 	 * to avoid extraneous signal generation from strrput(), we set
834 	 * STRGETINPROG flag at the stream head prior to the draining and
835 	 * restore it afterwards.  This masks out signal generation only
836 	 * for M_DATA messages and does not affect urgent data.
837 	 */
838 	if (tcp->tcp_direct_sockfs)
839 		strrput_sig(q, B_FALSE);
840 
841 	/* Drain the data */
842 	while ((mp = tcp->tcp_rcv_list) != NULL) {
843 		tcp->tcp_rcv_list = mp->b_next;
844 		mp->b_next = NULL;
845 #ifdef DEBUG
846 		cnt += msgdsize(mp);
847 #endif
848 		putnext(q, mp);
849 		TCP_STAT(tcps, tcp_fusion_putnext);
850 	}
851 
852 	if (tcp->tcp_direct_sockfs)
853 		strrput_sig(q, B_TRUE);
854 
855 	ASSERT(cnt == tcp->tcp_rcv_cnt);
856 	tcp->tcp_rcv_last_head = NULL;
857 	tcp->tcp_rcv_last_tail = NULL;
858 	tcp->tcp_rcv_cnt = 0;
859 	tcp->tcp_fuse_rcv_unread_cnt = 0;
860 	tcp->tcp_rwnd = q->q_hiwat;
861 
862 	return (B_TRUE);
863 }
864 
865 /*
866  * Synchronous stream entry point for sockfs to retrieve
867  * data directly from tcp_rcv_list.
868  * tcp_fuse_rrw() might end up modifying the peer's tcp_flow_stopped,
869  * for which it  must take the tcp_non_sq_lock of the peer as well
870  * making any change. The order of taking the locks is based on
871  * the TCP pointer itself. Before we get the peer we need to take
872  * our tcp_non_sq_lock so that the peer doesn't disappear. However,
873  * we cannot drop the lock if we have to grab the peer's lock (because
874  * of ordering), since the peer might disappear in the interim. So,
875  * we take our tcp_non_sq_lock, get the peer, increment the ref on the
876  * peer's conn, drop all the locks and then take the tcp_non_sq_lock in the
877  * desired order. Incrementing the conn ref on the peer means that the
878  * peer won't disappear when we drop our tcp_non_sq_lock.
879  */
880 int
881 tcp_fuse_rrw(queue_t *q, struiod_t *dp)
882 {
883 	tcp_t *tcp = Q_TO_CONN(q)->conn_tcp;
884 	mblk_t *mp;
885 	tcp_t *peer_tcp;
886 	tcp_stack_t	*tcps = tcp->tcp_tcps;
887 
888 	mutex_enter(&tcp->tcp_non_sq_lock);
889 
890 	/*
891 	 * If tcp_fuse_syncstr_plugged is set, then another thread is moving
892 	 * the underlying data to the stream head.  We need to wait until it's
893 	 * done, then return EBUSY so that strget() will dequeue data from the
894 	 * stream head to ensure data is drained in-order.
895 	 */
896 plugged:
897 	if (tcp->tcp_fuse_syncstr_plugged) {
898 		do {
899 			cv_wait(&tcp->tcp_fuse_plugcv, &tcp->tcp_non_sq_lock);
900 		} while (tcp->tcp_fuse_syncstr_plugged);
901 
902 		mutex_exit(&tcp->tcp_non_sq_lock);
903 		TCP_STAT(tcps, tcp_fusion_rrw_plugged);
904 		TCP_STAT(tcps, tcp_fusion_rrw_busy);
905 		return (EBUSY);
906 	}
907 
908 	peer_tcp = tcp->tcp_loopback_peer;
909 
910 	/*
911 	 * If someone had turned off tcp_direct_sockfs or if synchronous
912 	 * streams is stopped, we return EBUSY.  This causes strget() to
913 	 * dequeue data from the stream head instead.
914 	 */
915 	if (!tcp->tcp_direct_sockfs || tcp->tcp_fuse_syncstr_stopped) {
916 		mutex_exit(&tcp->tcp_non_sq_lock);
917 		TCP_STAT(tcps, tcp_fusion_rrw_busy);
918 		return (EBUSY);
919 	}
920 
921 	/*
922 	 * Grab lock in order. The highest addressed tcp is locked first.
923 	 * We don't do this within the tcp_rcv_list check since if we
924 	 * have to drop the lock, for ordering, then the tcp_rcv_list
925 	 * could change.
926 	 */
927 	if (peer_tcp > tcp) {
928 		CONN_INC_REF(peer_tcp->tcp_connp);
929 		mutex_exit(&tcp->tcp_non_sq_lock);
930 		mutex_enter(&peer_tcp->tcp_non_sq_lock);
931 		mutex_enter(&tcp->tcp_non_sq_lock);
932 		CONN_DEC_REF(peer_tcp->tcp_connp);
933 		/* This might have changed in the interim */
934 		if (tcp->tcp_fuse_syncstr_plugged) {
935 			mutex_exit(&peer_tcp->tcp_non_sq_lock);
936 			goto plugged;
937 		}
938 	} else {
939 		mutex_enter(&peer_tcp->tcp_non_sq_lock);
940 	}
941 
942 	if ((mp = tcp->tcp_rcv_list) != NULL) {
943 
944 		DTRACE_PROBE3(tcp__fuse__rrw, tcp_t *, tcp,
945 		    uint32_t, tcp->tcp_rcv_cnt, ssize_t, dp->d_uio.uio_resid);
946 
947 		tcp->tcp_rcv_list = NULL;
948 		TCP_STAT(tcps, tcp_fusion_rrw_msgcnt);
949 
950 		/*
951 		 * At this point nothing should be left in tcp_rcv_list.
952 		 * The only possible case where we would have a chain of
953 		 * b_next-linked messages is urgent data, but we wouldn't
954 		 * be here if that's true since urgent data is delivered
955 		 * via putnext() and synchronous streams is stopped until
956 		 * tcp_fuse_rcv_drain() is finished.
957 		 */
958 		ASSERT(DB_TYPE(mp) == M_DATA && mp->b_next == NULL);
959 
960 		tcp->tcp_rcv_last_head = NULL;
961 		tcp->tcp_rcv_last_tail = NULL;
962 		tcp->tcp_rcv_cnt = 0;
963 		tcp->tcp_fuse_rcv_unread_cnt = 0;
964 
965 		if (peer_tcp->tcp_flow_stopped) {
966 			tcp_clrqfull(peer_tcp);
967 			TCP_STAT(tcps, tcp_fusion_backenabled);
968 		}
969 	}
970 	mutex_exit(&peer_tcp->tcp_non_sq_lock);
971 	/*
972 	 * Either we just dequeued everything or we get here from sockfs
973 	 * and have nothing to return; in this case clear RSLEEP.
974 	 */
975 	ASSERT(tcp->tcp_rcv_last_head == NULL);
976 	ASSERT(tcp->tcp_rcv_last_tail == NULL);
977 	ASSERT(tcp->tcp_rcv_cnt == 0);
978 	ASSERT(tcp->tcp_fuse_rcv_unread_cnt == 0);
979 	STR_WAKEUP_CLEAR(STREAM(q));
980 
981 	mutex_exit(&tcp->tcp_non_sq_lock);
982 	dp->d_mp = mp;
983 	return (0);
984 }
985 
986 /*
987  * Synchronous stream entry point used by certain ioctls to retrieve
988  * information about or peek into the tcp_rcv_list.
989  */
990 int
991 tcp_fuse_rinfop(queue_t *q, infod_t *dp)
992 {
993 	tcp_t	*tcp = Q_TO_CONN(q)->conn_tcp;
994 	mblk_t	*mp;
995 	uint_t	cmd = dp->d_cmd;
996 	int	res = 0;
997 	int	error = 0;
998 	struct stdata *stp = STREAM(q);
999 
1000 	mutex_enter(&tcp->tcp_non_sq_lock);
1001 	/* If shutdown on read has happened, return nothing */
1002 	mutex_enter(&stp->sd_lock);
1003 	if (stp->sd_flag & STREOF) {
1004 		mutex_exit(&stp->sd_lock);
1005 		goto done;
1006 	}
1007 	mutex_exit(&stp->sd_lock);
1008 
1009 	/*
1010 	 * It is OK not to return an answer if tcp_rcv_list is
1011 	 * currently not accessible.
1012 	 */
1013 	if (!tcp->tcp_direct_sockfs || tcp->tcp_fuse_syncstr_stopped ||
1014 	    tcp->tcp_fuse_syncstr_plugged || (mp = tcp->tcp_rcv_list) == NULL)
1015 		goto done;
1016 
1017 	if (cmd & INFOD_COUNT) {
1018 		/*
1019 		 * We have at least one message and
1020 		 * could return only one at a time.
1021 		 */
1022 		dp->d_count++;
1023 		res |= INFOD_COUNT;
1024 	}
1025 	if (cmd & INFOD_BYTES) {
1026 		/*
1027 		 * Return size of all data messages.
1028 		 */
1029 		dp->d_bytes += tcp->tcp_rcv_cnt;
1030 		res |= INFOD_BYTES;
1031 	}
1032 	if (cmd & INFOD_FIRSTBYTES) {
1033 		/*
1034 		 * Return size of first data message.
1035 		 */
1036 		dp->d_bytes = msgdsize(mp);
1037 		res |= INFOD_FIRSTBYTES;
1038 		dp->d_cmd &= ~INFOD_FIRSTBYTES;
1039 	}
1040 	if (cmd & INFOD_COPYOUT) {
1041 		mblk_t *mp1;
1042 		int n;
1043 
1044 		if (DB_TYPE(mp) == M_DATA) {
1045 			mp1 = mp;
1046 		} else {
1047 			mp1 = mp->b_cont;
1048 			ASSERT(mp1 != NULL);
1049 		}
1050 
1051 		/*
1052 		 * Return data contents of first message.
1053 		 */
1054 		ASSERT(DB_TYPE(mp1) == M_DATA);
1055 		while (mp1 != NULL && dp->d_uiop->uio_resid > 0) {
1056 			n = MIN(dp->d_uiop->uio_resid, MBLKL(mp1));
1057 			if (n != 0 && (error = uiomove((char *)mp1->b_rptr, n,
1058 			    UIO_READ, dp->d_uiop)) != 0) {
1059 				goto done;
1060 			}
1061 			mp1 = mp1->b_cont;
1062 		}
1063 		res |= INFOD_COPYOUT;
1064 		dp->d_cmd &= ~INFOD_COPYOUT;
1065 	}
1066 done:
1067 	mutex_exit(&tcp->tcp_non_sq_lock);
1068 
1069 	dp->d_res |= res;
1070 
1071 	return (error);
1072 }
1073 
1074 /*
1075  * Enable synchronous streams on a fused tcp loopback endpoint.
1076  */
1077 static void
1078 tcp_fuse_syncstr_enable(tcp_t *tcp)
1079 {
1080 	queue_t *rq = tcp->tcp_rq;
1081 	struct stdata *stp = STREAM(rq);
1082 
1083 	/* We can only enable synchronous streams for sockfs mode */
1084 	tcp->tcp_direct_sockfs = tcp->tcp_issocket && do_tcp_direct_sockfs;
1085 
1086 	if (!tcp->tcp_direct_sockfs)
1087 		return;
1088 
1089 	mutex_enter(&stp->sd_lock);
1090 	mutex_enter(QLOCK(rq));
1091 
1092 	/*
1093 	 * We replace our q_qinfo with one that has the qi_rwp entry point.
1094 	 * Clear SR_SIGALLDATA because we generate the equivalent signal(s)
1095 	 * for every enqueued data in tcp_fuse_output().
1096 	 */
1097 	rq->q_qinfo = &tcp_loopback_rinit;
1098 	rq->q_struiot = tcp_loopback_rinit.qi_struiot;
1099 	stp->sd_struiordq = rq;
1100 	stp->sd_rput_opt &= ~SR_SIGALLDATA;
1101 
1102 	mutex_exit(QLOCK(rq));
1103 	mutex_exit(&stp->sd_lock);
1104 }
1105 
1106 /*
1107  * Disable synchronous streams on a fused tcp loopback endpoint.
1108  */
1109 static void
1110 tcp_fuse_syncstr_disable(tcp_t *tcp)
1111 {
1112 	queue_t *rq = tcp->tcp_rq;
1113 	struct stdata *stp = STREAM(rq);
1114 
1115 	if (!tcp->tcp_direct_sockfs)
1116 		return;
1117 
1118 	mutex_enter(&stp->sd_lock);
1119 	mutex_enter(QLOCK(rq));
1120 
1121 	/*
1122 	 * Reset q_qinfo to point to the default tcp entry points.
1123 	 * Also restore SR_SIGALLDATA so that strrput() can generate
1124 	 * the signals again for future M_DATA messages.
1125 	 */
1126 	rq->q_qinfo = &tcp_rinit;
1127 	rq->q_struiot = tcp_rinit.qi_struiot;
1128 	stp->sd_struiordq = NULL;
1129 	stp->sd_rput_opt |= SR_SIGALLDATA;
1130 	tcp->tcp_direct_sockfs = B_FALSE;
1131 
1132 	mutex_exit(QLOCK(rq));
1133 	mutex_exit(&stp->sd_lock);
1134 }
1135 
1136 /*
1137  * Enable synchronous streams on a pair of fused tcp endpoints.
1138  */
1139 void
1140 tcp_fuse_syncstr_enable_pair(tcp_t *tcp)
1141 {
1142 	tcp_t *peer_tcp = tcp->tcp_loopback_peer;
1143 
1144 	ASSERT(tcp->tcp_fused);
1145 	ASSERT(peer_tcp != NULL);
1146 
1147 	tcp_fuse_syncstr_enable(tcp);
1148 	tcp_fuse_syncstr_enable(peer_tcp);
1149 }
1150 
1151 /*
1152  * Allow or disallow signals to be generated by strrput().
1153  */
1154 static void
1155 strrput_sig(queue_t *q, boolean_t on)
1156 {
1157 	struct stdata *stp = STREAM(q);
1158 
1159 	mutex_enter(&stp->sd_lock);
1160 	if (on)
1161 		stp->sd_flag &= ~STRGETINPROG;
1162 	else
1163 		stp->sd_flag |= STRGETINPROG;
1164 	mutex_exit(&stp->sd_lock);
1165 }
1166 
1167 /*
1168  * Disable synchronous streams on a pair of fused tcp endpoints and drain
1169  * any queued data; called either during unfuse or upon transitioning from
1170  * a socket to a stream endpoint due to _SIOCSOCKFALLBACK.
1171  */
1172 void
1173 tcp_fuse_disable_pair(tcp_t *tcp, boolean_t unfusing)
1174 {
1175 	tcp_t *peer_tcp = tcp->tcp_loopback_peer;
1176 	tcp_stack_t	*tcps = tcp->tcp_tcps;
1177 
1178 	ASSERT(tcp->tcp_fused);
1179 	ASSERT(peer_tcp != NULL);
1180 
1181 	/*
1182 	 * Force any tcp_fuse_rrw() calls to block until we've moved the data
1183 	 * onto the stream head.
1184 	 */
1185 	TCP_FUSE_SYNCSTR_PLUG_DRAIN(tcp);
1186 	TCP_FUSE_SYNCSTR_PLUG_DRAIN(peer_tcp);
1187 
1188 	/*
1189 	 * Drain any pending data; the detached check is needed because
1190 	 * we may be called as a result of a tcp_unfuse() triggered by
1191 	 * tcp_fuse_output().  Note that in case of a detached tcp, the
1192 	 * draining will happen later after the tcp is unfused.  For non-
1193 	 * urgent data, this can be handled by the regular tcp_rcv_drain().
1194 	 * If we have urgent data sitting in the receive list, we will
1195 	 * need to send up a SIGURG signal first before draining the data.
1196 	 * All of these will be handled by the code in tcp_fuse_rcv_drain()
1197 	 * when called from tcp_rcv_drain().
1198 	 */
1199 	if (!TCP_IS_DETACHED(tcp)) {
1200 		(void) tcp_fuse_rcv_drain(tcp->tcp_rq, tcp,
1201 		    (unfusing ? &tcp->tcp_fused_sigurg_mp : NULL));
1202 	}
1203 	if (!TCP_IS_DETACHED(peer_tcp)) {
1204 		(void) tcp_fuse_rcv_drain(peer_tcp->tcp_rq, peer_tcp,
1205 		    (unfusing ? &peer_tcp->tcp_fused_sigurg_mp : NULL));
1206 	}
1207 
1208 	/*
1209 	 * Make all current and future tcp_fuse_rrw() calls fail with EBUSY.
1210 	 * To ensure threads don't sneak past the checks in tcp_fuse_rrw(),
1211 	 * a given stream must be stopped prior to being unplugged (but the
1212 	 * ordering of operations between the streams is unimportant).
1213 	 */
1214 	TCP_FUSE_SYNCSTR_STOP(tcp);
1215 	TCP_FUSE_SYNCSTR_STOP(peer_tcp);
1216 	TCP_FUSE_SYNCSTR_UNPLUG_DRAIN(tcp);
1217 	TCP_FUSE_SYNCSTR_UNPLUG_DRAIN(peer_tcp);
1218 
1219 	/* Lift up any flow-control conditions */
1220 	if (tcp->tcp_flow_stopped) {
1221 		tcp_clrqfull(tcp);
1222 		TCP_STAT(tcps, tcp_fusion_backenabled);
1223 	}
1224 	if (peer_tcp->tcp_flow_stopped) {
1225 		tcp_clrqfull(peer_tcp);
1226 		TCP_STAT(tcps, tcp_fusion_backenabled);
1227 	}
1228 
1229 	/* Disable synchronous streams */
1230 	tcp_fuse_syncstr_disable(tcp);
1231 	tcp_fuse_syncstr_disable(peer_tcp);
1232 }
1233 
1234 /*
1235  * Calculate the size of receive buffer for a fused tcp endpoint.
1236  */
1237 size_t
1238 tcp_fuse_set_rcv_hiwat(tcp_t *tcp, size_t rwnd)
1239 {
1240 	tcp_stack_t	*tcps = tcp->tcp_tcps;
1241 
1242 	ASSERT(tcp->tcp_fused);
1243 
1244 	/* Ensure that value is within the maximum upper bound */
1245 	if (rwnd > tcps->tcps_max_buf)
1246 		rwnd = tcps->tcps_max_buf;
1247 
1248 	/* Obey the absolute minimum tcp receive high water mark */
1249 	if (rwnd < tcps->tcps_sth_rcv_hiwat)
1250 		rwnd = tcps->tcps_sth_rcv_hiwat;
1251 
1252 	/*
1253 	 * Round up to system page size in case SO_RCVBUF is modified
1254 	 * after SO_SNDBUF; the latter is also similarly rounded up.
1255 	 */
1256 	rwnd = P2ROUNDUP_TYPED(rwnd, PAGESIZE, size_t);
1257 	tcp->tcp_fuse_rcv_hiwater = rwnd;
1258 	return (rwnd);
1259 }
1260 
1261 /*
1262  * Calculate the maximum outstanding unread data block for a fused tcp endpoint.
1263  */
1264 int
1265 tcp_fuse_maxpsz_set(tcp_t *tcp)
1266 {
1267 	tcp_t *peer_tcp = tcp->tcp_loopback_peer;
1268 	uint_t sndbuf = tcp->tcp_xmit_hiwater;
1269 	uint_t maxpsz = sndbuf;
1270 
1271 	ASSERT(tcp->tcp_fused);
1272 	ASSERT(peer_tcp != NULL);
1273 	ASSERT(peer_tcp->tcp_fuse_rcv_hiwater != 0);
1274 	/*
1275 	 * In the fused loopback case, we want the stream head to split
1276 	 * up larger writes into smaller chunks for a more accurate flow-
1277 	 * control accounting.  Our maxpsz is half of the sender's send
1278 	 * buffer or the receiver's receive buffer, whichever is smaller.
1279 	 * We round up the buffer to system page size due to the lack of
1280 	 * TCP MSS concept in Fusion.
1281 	 */
1282 	if (maxpsz > peer_tcp->tcp_fuse_rcv_hiwater)
1283 		maxpsz = peer_tcp->tcp_fuse_rcv_hiwater;
1284 	maxpsz = P2ROUNDUP_TYPED(maxpsz, PAGESIZE, uint_t) >> 1;
1285 
1286 	/*
1287 	 * Calculate the peer's limit for the number of outstanding unread
1288 	 * data block.  This is the amount of data blocks that are allowed
1289 	 * to reside in the receiver's queue before the sender gets flow
1290 	 * controlled.  It is used only in the synchronous streams mode as
1291 	 * a way to throttle the sender when it performs consecutive writes
1292 	 * faster than can be read.  The value is derived from SO_SNDBUF in
1293 	 * order to give the sender some control; we divide it with a large
1294 	 * value (16KB) to produce a fairly low initial limit.
1295 	 */
1296 	if (tcp_fusion_rcv_unread_min == 0) {
1297 		/* A value of 0 means that we disable the check */
1298 		peer_tcp->tcp_fuse_rcv_unread_hiwater = 0;
1299 	} else {
1300 		peer_tcp->tcp_fuse_rcv_unread_hiwater =
1301 		    MAX(sndbuf >> 14, tcp_fusion_rcv_unread_min);
1302 	}
1303 	return (maxpsz);
1304 }
1305