xref: /freebsd/sys/rpc/svc_vc.c (revision 315ee00f)
1 /*	$NetBSD: svc_vc.c,v 1.7 2000/08/03 00:01:53 fvdl Exp $	*/
2 
3 /*-
4  * SPDX-License-Identifier: BSD-3-Clause
5  *
6  * Copyright (c) 2009, Sun Microsystems, Inc.
7  * All rights reserved.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions are met:
11  * - Redistributions of source code must retain the above copyright notice,
12  *   this list of conditions and the following disclaimer.
13  * - Redistributions in binary form must reproduce the above copyright notice,
14  *   this list of conditions and the following disclaimer in the documentation
15  *   and/or other materials provided with the distribution.
16  * - Neither the name of Sun Microsystems, Inc. nor the names of its
17  *   contributors may be used to endorse or promote products derived
18  *   from this software without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
21  * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
24  * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
27  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
29  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
30  * POSSIBILITY OF SUCH DAMAGE.
31  */
32 
33 #if defined(LIBC_SCCS) && !defined(lint)
34 static char *sccsid2 = "@(#)svc_tcp.c 1.21 87/08/11 Copyr 1984 Sun Micro";
35 static char *sccsid = "@(#)svc_tcp.c	2.2 88/08/01 4.0 RPCSRC";
36 #endif
37 #include <sys/cdefs.h>
38 /*
39  * svc_vc.c, Server side for Connection Oriented based RPC.
40  *
41  * Actually implements two flavors of transporter -
42  * a tcp rendezvouser (a listner and connection establisher)
43  * and a record/tcp stream.
44  */
45 
46 #include "opt_kern_tls.h"
47 
48 #include <sys/param.h>
49 #include <sys/limits.h>
50 #include <sys/lock.h>
51 #include <sys/kernel.h>
52 #include <sys/ktls.h>
53 #include <sys/malloc.h>
54 #include <sys/mbuf.h>
55 #include <sys/mutex.h>
56 #include <sys/proc.h>
57 #include <sys/protosw.h>
58 #include <sys/queue.h>
59 #include <sys/socket.h>
60 #include <sys/socketvar.h>
61 #include <sys/sx.h>
62 #include <sys/systm.h>
63 #include <sys/uio.h>
64 
65 #include <net/vnet.h>
66 
67 #include <netinet/tcp.h>
68 
69 #include <rpc/rpc.h>
70 #include <rpc/rpcsec_tls.h>
71 
72 #include <rpc/krpc.h>
73 #include <rpc/rpc_com.h>
74 
75 #include <security/mac/mac_framework.h>
76 
77 static bool_t svc_vc_rendezvous_recv(SVCXPRT *, struct rpc_msg *,
78     struct sockaddr **, struct mbuf **);
79 static enum xprt_stat svc_vc_rendezvous_stat(SVCXPRT *);
80 static void svc_vc_rendezvous_destroy(SVCXPRT *);
81 static bool_t svc_vc_null(void);
82 static void svc_vc_destroy(SVCXPRT *);
83 static enum xprt_stat svc_vc_stat(SVCXPRT *);
84 static bool_t svc_vc_ack(SVCXPRT *, uint32_t *);
85 static bool_t svc_vc_recv(SVCXPRT *, struct rpc_msg *,
86     struct sockaddr **, struct mbuf **);
87 static bool_t svc_vc_reply(SVCXPRT *, struct rpc_msg *,
88     struct sockaddr *, struct mbuf *, uint32_t *seq);
89 static bool_t svc_vc_control(SVCXPRT *xprt, const u_int rq, void *in);
90 static bool_t svc_vc_rendezvous_control (SVCXPRT *xprt, const u_int rq,
91     void *in);
92 static void svc_vc_backchannel_destroy(SVCXPRT *);
93 static enum xprt_stat svc_vc_backchannel_stat(SVCXPRT *);
94 static bool_t svc_vc_backchannel_recv(SVCXPRT *, struct rpc_msg *,
95     struct sockaddr **, struct mbuf **);
96 static bool_t svc_vc_backchannel_reply(SVCXPRT *, struct rpc_msg *,
97     struct sockaddr *, struct mbuf *, uint32_t *);
98 static bool_t svc_vc_backchannel_control(SVCXPRT *xprt, const u_int rq,
99     void *in);
100 static SVCXPRT *svc_vc_create_conn(SVCPOOL *pool, struct socket *so,
101     struct sockaddr *raddr);
102 static int svc_vc_accept(struct socket *head, struct socket **sop);
103 static int svc_vc_soupcall(struct socket *so, void *arg, int waitflag);
104 static int svc_vc_rendezvous_soupcall(struct socket *, void *, int);
105 
106 static const struct xp_ops svc_vc_rendezvous_ops = {
107 	.xp_recv =	svc_vc_rendezvous_recv,
108 	.xp_stat =	svc_vc_rendezvous_stat,
109 	.xp_reply =	(bool_t (*)(SVCXPRT *, struct rpc_msg *,
110 		struct sockaddr *, struct mbuf *, uint32_t *))svc_vc_null,
111 	.xp_destroy =	svc_vc_rendezvous_destroy,
112 	.xp_control =	svc_vc_rendezvous_control
113 };
114 
115 static const struct xp_ops svc_vc_ops = {
116 	.xp_recv =	svc_vc_recv,
117 	.xp_stat =	svc_vc_stat,
118 	.xp_ack =	svc_vc_ack,
119 	.xp_reply =	svc_vc_reply,
120 	.xp_destroy =	svc_vc_destroy,
121 	.xp_control =	svc_vc_control
122 };
123 
124 static const struct xp_ops svc_vc_backchannel_ops = {
125 	.xp_recv =	svc_vc_backchannel_recv,
126 	.xp_stat =	svc_vc_backchannel_stat,
127 	.xp_reply =	svc_vc_backchannel_reply,
128 	.xp_destroy =	svc_vc_backchannel_destroy,
129 	.xp_control =	svc_vc_backchannel_control
130 };
131 
132 /*
133  * Usage:
134  *	xprt = svc_vc_create(sock, send_buf_size, recv_buf_size);
135  *
136  * Creates, registers, and returns a (rpc) tcp based transporter.
137  * Once *xprt is initialized, it is registered as a transporter
138  * see (svc.h, xprt_register).  This routine returns
139  * a NULL if a problem occurred.
140  *
141  * The filedescriptor passed in is expected to refer to a bound, but
142  * not yet connected socket.
143  *
144  * Since streams do buffered io similar to stdio, the caller can specify
145  * how big the send and receive buffers are via the second and third parms;
146  * 0 => use the system default.
147  */
148 SVCXPRT *
149 svc_vc_create(SVCPOOL *pool, struct socket *so, size_t sendsize,
150     size_t recvsize)
151 {
152 	SVCXPRT *xprt;
153 	struct sockaddr* sa;
154 	int error;
155 
156 	SOCK_LOCK(so);
157 	if (so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED)) {
158 		SOCK_UNLOCK(so);
159 		CURVNET_SET(so->so_vnet);
160 		error = so->so_proto->pr_peeraddr(so, &sa);
161 		CURVNET_RESTORE();
162 		if (error)
163 			return (NULL);
164 		xprt = svc_vc_create_conn(pool, so, sa);
165 		free(sa, M_SONAME);
166 		return (xprt);
167 	}
168 	SOCK_UNLOCK(so);
169 
170 	xprt = svc_xprt_alloc();
171 	sx_init(&xprt->xp_lock, "xprt->xp_lock");
172 	xprt->xp_pool = pool;
173 	xprt->xp_socket = so;
174 	xprt->xp_p1 = NULL;
175 	xprt->xp_p2 = NULL;
176 	xprt->xp_ops = &svc_vc_rendezvous_ops;
177 
178 	CURVNET_SET(so->so_vnet);
179 	error = so->so_proto->pr_sockaddr(so, &sa);
180 	CURVNET_RESTORE();
181 	if (error) {
182 		goto cleanup_svc_vc_create;
183 	}
184 
185 	memcpy(&xprt->xp_ltaddr, sa, sa->sa_len);
186 	free(sa, M_SONAME);
187 
188 	xprt_register(xprt);
189 
190 	solisten(so, -1, curthread);
191 
192 	SOLISTEN_LOCK(so);
193 	xprt->xp_upcallset = 1;
194 	solisten_upcall_set(so, svc_vc_rendezvous_soupcall, xprt);
195 	SOLISTEN_UNLOCK(so);
196 
197 	return (xprt);
198 
199 cleanup_svc_vc_create:
200 	sx_destroy(&xprt->xp_lock);
201 	svc_xprt_free(xprt);
202 
203 	return (NULL);
204 }
205 
206 /*
207  * Create a new transport for a socket optained via soaccept().
208  */
209 SVCXPRT *
210 svc_vc_create_conn(SVCPOOL *pool, struct socket *so, struct sockaddr *raddr)
211 {
212 	SVCXPRT *xprt;
213 	struct cf_conn *cd;
214 	struct sockaddr* sa = NULL;
215 	struct sockopt opt;
216 	int one = 1;
217 	int error;
218 
219 	bzero(&opt, sizeof(struct sockopt));
220 	opt.sopt_dir = SOPT_SET;
221 	opt.sopt_level = SOL_SOCKET;
222 	opt.sopt_name = SO_KEEPALIVE;
223 	opt.sopt_val = &one;
224 	opt.sopt_valsize = sizeof(one);
225 	error = sosetopt(so, &opt);
226 	if (error) {
227 		return (NULL);
228 	}
229 
230 	if (so->so_proto->pr_protocol == IPPROTO_TCP) {
231 		bzero(&opt, sizeof(struct sockopt));
232 		opt.sopt_dir = SOPT_SET;
233 		opt.sopt_level = IPPROTO_TCP;
234 		opt.sopt_name = TCP_NODELAY;
235 		opt.sopt_val = &one;
236 		opt.sopt_valsize = sizeof(one);
237 		error = sosetopt(so, &opt);
238 		if (error) {
239 			return (NULL);
240 		}
241 	}
242 
243 	cd = mem_alloc(sizeof(*cd));
244 	cd->strm_stat = XPRT_IDLE;
245 
246 	xprt = svc_xprt_alloc();
247 	sx_init(&xprt->xp_lock, "xprt->xp_lock");
248 	xprt->xp_pool = pool;
249 	xprt->xp_socket = so;
250 	xprt->xp_p1 = cd;
251 	xprt->xp_p2 = NULL;
252 	xprt->xp_ops = &svc_vc_ops;
253 
254 	/*
255 	 * See http://www.connectathon.org/talks96/nfstcp.pdf - client
256 	 * has a 5 minute timer, server has a 6 minute timer.
257 	 */
258 	xprt->xp_idletimeout = 6 * 60;
259 
260 	memcpy(&xprt->xp_rtaddr, raddr, raddr->sa_len);
261 
262 	CURVNET_SET(so->so_vnet);
263 	error = so->so_proto->pr_sockaddr(so, &sa);
264 	CURVNET_RESTORE();
265 	if (error)
266 		goto cleanup_svc_vc_create;
267 
268 	memcpy(&xprt->xp_ltaddr, sa, sa->sa_len);
269 	free(sa, M_SONAME);
270 
271 	xprt_register(xprt);
272 
273 	SOCKBUF_LOCK(&so->so_rcv);
274 	xprt->xp_upcallset = 1;
275 	soupcall_set(so, SO_RCV, svc_vc_soupcall, xprt);
276 	SOCKBUF_UNLOCK(&so->so_rcv);
277 
278 	/*
279 	 * Throw the transport into the active list in case it already
280 	 * has some data buffered.
281 	 */
282 	sx_xlock(&xprt->xp_lock);
283 	xprt_active(xprt);
284 	sx_xunlock(&xprt->xp_lock);
285 
286 	return (xprt);
287 cleanup_svc_vc_create:
288 	sx_destroy(&xprt->xp_lock);
289 	svc_xprt_free(xprt);
290 	mem_free(cd, sizeof(*cd));
291 
292 	return (NULL);
293 }
294 
295 /*
296  * Create a new transport for a backchannel on a clnt_vc socket.
297  */
298 SVCXPRT *
299 svc_vc_create_backchannel(SVCPOOL *pool)
300 {
301 	SVCXPRT *xprt = NULL;
302 	struct cf_conn *cd = NULL;
303 
304 	cd = mem_alloc(sizeof(*cd));
305 	cd->strm_stat = XPRT_IDLE;
306 
307 	xprt = svc_xprt_alloc();
308 	sx_init(&xprt->xp_lock, "xprt->xp_lock");
309 	xprt->xp_pool = pool;
310 	xprt->xp_socket = NULL;
311 	xprt->xp_p1 = cd;
312 	xprt->xp_p2 = NULL;
313 	xprt->xp_ops = &svc_vc_backchannel_ops;
314 	return (xprt);
315 }
316 
317 /*
318  * This does all of the accept except the final call to soaccept. The
319  * caller will call soaccept after dropping its locks (soaccept may
320  * call malloc).
321  */
322 int
323 svc_vc_accept(struct socket *head, struct socket **sop)
324 {
325 	struct socket *so;
326 	int error = 0;
327 	short nbio;
328 
329 	KASSERT(SOLISTENING(head),
330 	    ("%s: socket %p is not listening", __func__, head));
331 
332 #ifdef MAC
333 	error = mac_socket_check_accept(curthread->td_ucred, head);
334 	if (error != 0)
335 		goto done;
336 #endif
337 	/*
338 	 * XXXGL: we want non-blocking semantics.  The socket could be a
339 	 * socket created by kernel as well as socket shared with userland,
340 	 * so we can't be sure about presense of SS_NBIO.  We also shall not
341 	 * toggle it on the socket, since that may surprise userland.  So we
342 	 * set SS_NBIO only temporarily.
343 	 */
344 	SOLISTEN_LOCK(head);
345 	nbio = head->so_state & SS_NBIO;
346 	head->so_state |= SS_NBIO;
347 	error = solisten_dequeue(head, &so, 0);
348 	head->so_state &= (nbio & ~SS_NBIO);
349 	if (error)
350 		goto done;
351 
352 	so->so_state |= nbio;
353 	*sop = so;
354 
355 	/* connection has been removed from the listen queue */
356 	KNOTE_UNLOCKED(&head->so_rdsel.si_note, 0);
357 done:
358 	return (error);
359 }
360 
361 /*ARGSUSED*/
362 static bool_t
363 svc_vc_rendezvous_recv(SVCXPRT *xprt, struct rpc_msg *msg,
364     struct sockaddr **addrp, struct mbuf **mp)
365 {
366 	struct socket *so = NULL;
367 	struct sockaddr *sa = NULL;
368 	int error;
369 	SVCXPRT *new_xprt;
370 
371 	/*
372 	 * The socket upcall calls xprt_active() which will eventually
373 	 * cause the server to call us here. We attempt to accept a
374 	 * connection from the socket and turn it into a new
375 	 * transport. If the accept fails, we have drained all pending
376 	 * connections so we call xprt_inactive().
377 	 */
378 	sx_xlock(&xprt->xp_lock);
379 
380 	error = svc_vc_accept(xprt->xp_socket, &so);
381 
382 	if (error == EWOULDBLOCK) {
383 		/*
384 		 * We must re-test for new connections after taking
385 		 * the lock to protect us in the case where a new
386 		 * connection arrives after our call to accept fails
387 		 * with EWOULDBLOCK.
388 		 */
389 		SOLISTEN_LOCK(xprt->xp_socket);
390 		if (TAILQ_EMPTY(&xprt->xp_socket->sol_comp))
391 			xprt_inactive_self(xprt);
392 		SOLISTEN_UNLOCK(xprt->xp_socket);
393 		sx_xunlock(&xprt->xp_lock);
394 		return (FALSE);
395 	}
396 
397 	if (error) {
398 		SOLISTEN_LOCK(xprt->xp_socket);
399 		if (xprt->xp_upcallset) {
400 			xprt->xp_upcallset = 0;
401 			soupcall_clear(xprt->xp_socket, SO_RCV);
402 		}
403 		SOLISTEN_UNLOCK(xprt->xp_socket);
404 		xprt_inactive_self(xprt);
405 		sx_xunlock(&xprt->xp_lock);
406 		return (FALSE);
407 	}
408 
409 	sx_xunlock(&xprt->xp_lock);
410 
411 	sa = NULL;
412 	error = soaccept(so, &sa);
413 
414 	if (error) {
415 		/*
416 		 * XXX not sure if I need to call sofree or soclose here.
417 		 */
418 		if (sa)
419 			free(sa, M_SONAME);
420 		return (FALSE);
421 	}
422 
423 	/*
424 	 * svc_vc_create_conn will call xprt_register - we don't need
425 	 * to do anything with the new connection except derefence it.
426 	 */
427 	new_xprt = svc_vc_create_conn(xprt->xp_pool, so, sa);
428 	if (!new_xprt) {
429 		soclose(so);
430 	} else {
431 		SVC_RELEASE(new_xprt);
432 	}
433 
434 	free(sa, M_SONAME);
435 
436 	return (FALSE); /* there is never an rpc msg to be processed */
437 }
438 
439 /*ARGSUSED*/
440 static enum xprt_stat
441 svc_vc_rendezvous_stat(SVCXPRT *xprt)
442 {
443 
444 	return (XPRT_IDLE);
445 }
446 
447 static void
448 svc_vc_destroy_common(SVCXPRT *xprt)
449 {
450 	uint32_t reterr;
451 
452 	if (xprt->xp_socket) {
453 		if ((xprt->xp_tls & (RPCTLS_FLAGS_HANDSHAKE |
454 		    RPCTLS_FLAGS_HANDSHFAIL)) != 0) {
455 			if ((xprt->xp_tls & RPCTLS_FLAGS_HANDSHAKE) != 0) {
456 				/*
457 				 * If the upcall fails, the socket has
458 				 * probably been closed via the rpctlssd
459 				 * daemon having crashed or been
460 				 * restarted, so just ignore returned stat.
461 				 */
462 				rpctls_srv_disconnect(xprt->xp_sslsec,
463 				    xprt->xp_sslusec, xprt->xp_sslrefno,
464 				    xprt->xp_sslproc, &reterr);
465 			}
466 			/* Must sorele() to get rid of reference. */
467 			CURVNET_SET(xprt->xp_socket->so_vnet);
468 			sorele(xprt->xp_socket);
469 			CURVNET_RESTORE();
470 		} else
471 			(void)soclose(xprt->xp_socket);
472 	}
473 
474 	if (xprt->xp_netid)
475 		(void) mem_free(xprt->xp_netid, strlen(xprt->xp_netid) + 1);
476 	svc_xprt_free(xprt);
477 }
478 
479 static void
480 svc_vc_rendezvous_destroy(SVCXPRT *xprt)
481 {
482 
483 	SOLISTEN_LOCK(xprt->xp_socket);
484 	if (xprt->xp_upcallset) {
485 		xprt->xp_upcallset = 0;
486 		solisten_upcall_set(xprt->xp_socket, NULL, NULL);
487 	}
488 	SOLISTEN_UNLOCK(xprt->xp_socket);
489 
490 	svc_vc_destroy_common(xprt);
491 }
492 
493 static void
494 svc_vc_destroy(SVCXPRT *xprt)
495 {
496 	struct cf_conn *cd = (struct cf_conn *)xprt->xp_p1;
497 	CLIENT *cl = (CLIENT *)xprt->xp_p2;
498 
499 	SOCKBUF_LOCK(&xprt->xp_socket->so_rcv);
500 	if (xprt->xp_upcallset) {
501 		xprt->xp_upcallset = 0;
502 		if (xprt->xp_socket->so_rcv.sb_upcall != NULL)
503 			soupcall_clear(xprt->xp_socket, SO_RCV);
504 	}
505 	SOCKBUF_UNLOCK(&xprt->xp_socket->so_rcv);
506 
507 	if (cl != NULL)
508 		CLNT_RELEASE(cl);
509 
510 	svc_vc_destroy_common(xprt);
511 
512 	if (cd->mreq)
513 		m_freem(cd->mreq);
514 	if (cd->mpending)
515 		m_freem(cd->mpending);
516 	mem_free(cd, sizeof(*cd));
517 }
518 
519 static void
520 svc_vc_backchannel_destroy(SVCXPRT *xprt)
521 {
522 	struct cf_conn *cd = (struct cf_conn *)xprt->xp_p1;
523 	struct mbuf *m, *m2;
524 
525 	svc_xprt_free(xprt);
526 	m = cd->mreq;
527 	while (m != NULL) {
528 		m2 = m;
529 		m = m->m_nextpkt;
530 		m_freem(m2);
531 	}
532 	mem_free(cd, sizeof(*cd));
533 }
534 
535 /*ARGSUSED*/
536 static bool_t
537 svc_vc_control(SVCXPRT *xprt, const u_int rq, void *in)
538 {
539 	return (FALSE);
540 }
541 
542 static bool_t
543 svc_vc_rendezvous_control(SVCXPRT *xprt, const u_int rq, void *in)
544 {
545 
546 	return (FALSE);
547 }
548 
549 static bool_t
550 svc_vc_backchannel_control(SVCXPRT *xprt, const u_int rq, void *in)
551 {
552 
553 	return (FALSE);
554 }
555 
556 static enum xprt_stat
557 svc_vc_stat(SVCXPRT *xprt)
558 {
559 	struct cf_conn *cd;
560 
561 	cd = (struct cf_conn *)(xprt->xp_p1);
562 
563 	if (cd->strm_stat == XPRT_DIED)
564 		return (XPRT_DIED);
565 
566 	if (cd->mreq != NULL && cd->resid == 0 && cd->eor)
567 		return (XPRT_MOREREQS);
568 
569 	if (soreadable(xprt->xp_socket))
570 		return (XPRT_MOREREQS);
571 
572 	return (XPRT_IDLE);
573 }
574 
575 static bool_t
576 svc_vc_ack(SVCXPRT *xprt, uint32_t *ack)
577 {
578 
579 	*ack = atomic_load_acq_32(&xprt->xp_snt_cnt);
580 	*ack -= sbused(&xprt->xp_socket->so_snd);
581 	return (TRUE);
582 }
583 
584 static enum xprt_stat
585 svc_vc_backchannel_stat(SVCXPRT *xprt)
586 {
587 	struct cf_conn *cd;
588 
589 	cd = (struct cf_conn *)(xprt->xp_p1);
590 
591 	if (cd->mreq != NULL)
592 		return (XPRT_MOREREQS);
593 
594 	return (XPRT_IDLE);
595 }
596 
597 /*
598  * If we have an mbuf chain in cd->mpending, try to parse a record from it,
599  * leaving the result in cd->mreq. If we don't have a complete record, leave
600  * the partial result in cd->mreq and try to read more from the socket.
601  */
602 static int
603 svc_vc_process_pending(SVCXPRT *xprt)
604 {
605 	struct cf_conn *cd = (struct cf_conn *) xprt->xp_p1;
606 	struct socket *so = xprt->xp_socket;
607 	struct mbuf *m;
608 
609 	/*
610 	 * If cd->resid is non-zero, we have part of the
611 	 * record already, otherwise we are expecting a record
612 	 * marker.
613 	 */
614 	if (!cd->resid && cd->mpending) {
615 		/*
616 		 * See if there is enough data buffered to
617 		 * make up a record marker. Make sure we can
618 		 * handle the case where the record marker is
619 		 * split across more than one mbuf.
620 		 */
621 		size_t n = 0;
622 		uint32_t header;
623 
624 		m = cd->mpending;
625 		while (n < sizeof(uint32_t) && m) {
626 			n += m->m_len;
627 			m = m->m_next;
628 		}
629 		if (n < sizeof(uint32_t)) {
630 			so->so_rcv.sb_lowat = sizeof(uint32_t) - n;
631 			return (FALSE);
632 		}
633 		m_copydata(cd->mpending, 0, sizeof(header),
634 		    (char *)&header);
635 		header = ntohl(header);
636 		cd->eor = (header & 0x80000000) != 0;
637 		cd->resid = header & 0x7fffffff;
638 		m_adj(cd->mpending, sizeof(uint32_t));
639 	}
640 
641 	/*
642 	 * Start pulling off mbufs from cd->mpending
643 	 * until we either have a complete record or
644 	 * we run out of data. We use m_split to pull
645 	 * data - it will pull as much as possible and
646 	 * split the last mbuf if necessary.
647 	 */
648 	while (cd->mpending && cd->resid) {
649 		m = cd->mpending;
650 		if (cd->mpending->m_next
651 		    || cd->mpending->m_len > cd->resid)
652 			cd->mpending = m_split(cd->mpending,
653 			    cd->resid, M_WAITOK);
654 		else
655 			cd->mpending = NULL;
656 		if (cd->mreq)
657 			m_last(cd->mreq)->m_next = m;
658 		else
659 			cd->mreq = m;
660 		while (m) {
661 			cd->resid -= m->m_len;
662 			m = m->m_next;
663 		}
664 	}
665 
666 	/*
667 	 * Block receive upcalls if we have more data pending,
668 	 * otherwise report our need.
669 	 */
670 	if (cd->mpending)
671 		so->so_rcv.sb_lowat = INT_MAX;
672 	else
673 		so->so_rcv.sb_lowat =
674 		    imax(1, imin(cd->resid, so->so_rcv.sb_hiwat / 2));
675 	return (TRUE);
676 }
677 
678 static bool_t
679 svc_vc_recv(SVCXPRT *xprt, struct rpc_msg *msg,
680     struct sockaddr **addrp, struct mbuf **mp)
681 {
682 	struct cf_conn *cd = (struct cf_conn *) xprt->xp_p1;
683 	struct uio uio;
684 	struct mbuf *m, *ctrl;
685 	struct socket* so = xprt->xp_socket;
686 	XDR xdrs;
687 	int error, rcvflag;
688 	uint32_t reterr, xid_plus_direction[2];
689 	struct cmsghdr *cmsg;
690 	struct tls_get_record tgr;
691 	enum clnt_stat ret;
692 
693 	/*
694 	 * Serialise access to the socket and our own record parsing
695 	 * state.
696 	 */
697 	sx_xlock(&xprt->xp_lock);
698 
699 	for (;;) {
700 		/* If we have no request ready, check pending queue. */
701 		while (cd->mpending &&
702 		    (cd->mreq == NULL || cd->resid != 0 || !cd->eor)) {
703 			if (!svc_vc_process_pending(xprt))
704 				break;
705 		}
706 
707 		/* Process and return complete request in cd->mreq. */
708 		if (cd->mreq != NULL && cd->resid == 0 && cd->eor) {
709 
710 			/*
711 			 * Now, check for a backchannel reply.
712 			 * The XID is in the first uint32_t of the reply
713 			 * and the message direction is the second one.
714 			 */
715 			if ((cd->mreq->m_len >= sizeof(xid_plus_direction) ||
716 			    m_length(cd->mreq, NULL) >=
717 			    sizeof(xid_plus_direction)) &&
718 			    xprt->xp_p2 != NULL) {
719 				m_copydata(cd->mreq, 0,
720 				    sizeof(xid_plus_direction),
721 				    (char *)xid_plus_direction);
722 				xid_plus_direction[0] =
723 				    ntohl(xid_plus_direction[0]);
724 				xid_plus_direction[1] =
725 				    ntohl(xid_plus_direction[1]);
726 				/* Check message direction. */
727 				if (xid_plus_direction[1] == REPLY) {
728 					clnt_bck_svccall(xprt->xp_p2,
729 					    cd->mreq,
730 					    xid_plus_direction[0]);
731 					cd->mreq = NULL;
732 					continue;
733 				}
734 			}
735 
736 			xdrmbuf_create(&xdrs, cd->mreq, XDR_DECODE);
737 			cd->mreq = NULL;
738 
739 			/* Check for next request in a pending queue. */
740 			svc_vc_process_pending(xprt);
741 			if (cd->mreq == NULL || cd->resid != 0) {
742 				SOCKBUF_LOCK(&so->so_rcv);
743 				if (!soreadable(so))
744 					xprt_inactive_self(xprt);
745 				SOCKBUF_UNLOCK(&so->so_rcv);
746 			}
747 
748 			sx_xunlock(&xprt->xp_lock);
749 
750 			if (! xdr_callmsg(&xdrs, msg)) {
751 				XDR_DESTROY(&xdrs);
752 				return (FALSE);
753 			}
754 
755 			*addrp = NULL;
756 			*mp = xdrmbuf_getall(&xdrs);
757 			XDR_DESTROY(&xdrs);
758 
759 			return (TRUE);
760 		}
761 
762 		/*
763 		 * If receiving is disabled so that a TLS handshake can be
764 		 * done by the rpctlssd daemon, return FALSE here.
765 		 */
766 		rcvflag = MSG_DONTWAIT;
767 		if ((xprt->xp_tls & RPCTLS_FLAGS_HANDSHAKE) != 0)
768 			rcvflag |= MSG_TLSAPPDATA;
769 tryagain:
770 		if (xprt->xp_dontrcv) {
771 			sx_xunlock(&xprt->xp_lock);
772 			return (FALSE);
773 		}
774 
775 		/*
776 		 * The socket upcall calls xprt_active() which will eventually
777 		 * cause the server to call us here. We attempt to
778 		 * read as much as possible from the socket and put
779 		 * the result in cd->mpending. If the read fails,
780 		 * we have drained both cd->mpending and the socket so
781 		 * we can call xprt_inactive().
782 		 */
783 		uio.uio_resid = 1000000000;
784 		uio.uio_td = curthread;
785 		ctrl = m = NULL;
786 		error = soreceive(so, NULL, &uio, &m, &ctrl, &rcvflag);
787 
788 		if (error == EWOULDBLOCK) {
789 			/*
790 			 * We must re-test for readability after
791 			 * taking the lock to protect us in the case
792 			 * where a new packet arrives on the socket
793 			 * after our call to soreceive fails with
794 			 * EWOULDBLOCK.
795 			 */
796 			SOCKBUF_LOCK(&so->so_rcv);
797 			if (!soreadable(so))
798 				xprt_inactive_self(xprt);
799 			SOCKBUF_UNLOCK(&so->so_rcv);
800 			sx_xunlock(&xprt->xp_lock);
801 			return (FALSE);
802 		}
803 
804 		/*
805 		 * A return of ENXIO indicates that there is an
806 		 * alert record at the head of the
807 		 * socket's receive queue, for TLS connections.
808 		 * This record needs to be handled in userland
809 		 * via an SSL_read() call, so do an upcall to the daemon.
810 		 */
811 		if ((xprt->xp_tls & RPCTLS_FLAGS_HANDSHAKE) != 0 &&
812 		    error == ENXIO) {
813 			/* Disable reception. */
814 			xprt->xp_dontrcv = TRUE;
815 			sx_xunlock(&xprt->xp_lock);
816 			ret = rpctls_srv_handlerecord(xprt->xp_sslsec,
817 			    xprt->xp_sslusec, xprt->xp_sslrefno,
818 			    xprt->xp_sslproc, &reterr);
819 			sx_xlock(&xprt->xp_lock);
820 			xprt->xp_dontrcv = FALSE;
821 			if (ret != RPC_SUCCESS || reterr != RPCTLSERR_OK) {
822 				/*
823 				 * All we can do is soreceive() it and
824 				 * then toss it.
825 				 */
826 				rcvflag = MSG_DONTWAIT;
827 				goto tryagain;
828 			}
829 			sx_xunlock(&xprt->xp_lock);
830 			xprt_active(xprt);   /* Harmless if already active. */
831 			return (FALSE);
832 		}
833 
834 		if (error) {
835 			SOCKBUF_LOCK(&so->so_rcv);
836 			if (xprt->xp_upcallset) {
837 				xprt->xp_upcallset = 0;
838 				soupcall_clear(so, SO_RCV);
839 			}
840 			SOCKBUF_UNLOCK(&so->so_rcv);
841 			xprt_inactive_self(xprt);
842 			cd->strm_stat = XPRT_DIED;
843 			sx_xunlock(&xprt->xp_lock);
844 			return (FALSE);
845 		}
846 
847 		if (!m) {
848 			/*
849 			 * EOF - the other end has closed the socket.
850 			 */
851 			xprt_inactive_self(xprt);
852 			cd->strm_stat = XPRT_DIED;
853 			sx_xunlock(&xprt->xp_lock);
854 			return (FALSE);
855 		}
856 
857 		/* Process any record header(s). */
858 		if (ctrl != NULL) {
859 			cmsg = mtod(ctrl, struct cmsghdr *);
860 			if (cmsg->cmsg_type == TLS_GET_RECORD &&
861 			    cmsg->cmsg_len == CMSG_LEN(sizeof(tgr))) {
862 				memcpy(&tgr, CMSG_DATA(cmsg), sizeof(tgr));
863 				/*
864 				 * TLS_RLTYPE_ALERT records should be handled
865 				 * since soreceive() would have returned
866 				 * ENXIO.  Just throw any other
867 				 * non-TLS_RLTYPE_APP records away.
868 				 */
869 				if (tgr.tls_type != TLS_RLTYPE_APP) {
870 					m_freem(m);
871 					m_free(ctrl);
872 					rcvflag = MSG_DONTWAIT | MSG_TLSAPPDATA;
873 					goto tryagain;
874 				}
875 			}
876 			m_free(ctrl);
877 		}
878 
879 		if (cd->mpending)
880 			m_last(cd->mpending)->m_next = m;
881 		else
882 			cd->mpending = m;
883 	}
884 }
885 
886 static bool_t
887 svc_vc_backchannel_recv(SVCXPRT *xprt, struct rpc_msg *msg,
888     struct sockaddr **addrp, struct mbuf **mp)
889 {
890 	struct cf_conn *cd = (struct cf_conn *) xprt->xp_p1;
891 	struct ct_data *ct;
892 	struct mbuf *m;
893 	XDR xdrs;
894 
895 	sx_xlock(&xprt->xp_lock);
896 	ct = (struct ct_data *)xprt->xp_p2;
897 	if (ct == NULL) {
898 		sx_xunlock(&xprt->xp_lock);
899 		return (FALSE);
900 	}
901 	mtx_lock(&ct->ct_lock);
902 	m = cd->mreq;
903 	if (m == NULL) {
904 		xprt_inactive_self(xprt);
905 		mtx_unlock(&ct->ct_lock);
906 		sx_xunlock(&xprt->xp_lock);
907 		return (FALSE);
908 	}
909 	cd->mreq = m->m_nextpkt;
910 	mtx_unlock(&ct->ct_lock);
911 	sx_xunlock(&xprt->xp_lock);
912 
913 	xdrmbuf_create(&xdrs, m, XDR_DECODE);
914 	if (! xdr_callmsg(&xdrs, msg)) {
915 		XDR_DESTROY(&xdrs);
916 		return (FALSE);
917 	}
918 	*addrp = NULL;
919 	*mp = xdrmbuf_getall(&xdrs);
920 	XDR_DESTROY(&xdrs);
921 	return (TRUE);
922 }
923 
924 static bool_t
925 svc_vc_reply(SVCXPRT *xprt, struct rpc_msg *msg,
926     struct sockaddr *addr, struct mbuf *m, uint32_t *seq)
927 {
928 	XDR xdrs;
929 	struct mbuf *mrep;
930 	bool_t stat = TRUE;
931 	int error, len, maxextsiz;
932 #ifdef KERN_TLS
933 	u_int maxlen;
934 #endif
935 
936 	/*
937 	 * Leave space for record mark.
938 	 */
939 	mrep = m_gethdr(M_WAITOK, MT_DATA);
940 	mrep->m_data += sizeof(uint32_t);
941 
942 	xdrmbuf_create(&xdrs, mrep, XDR_ENCODE);
943 
944 	if (msg->rm_reply.rp_stat == MSG_ACCEPTED &&
945 	    msg->rm_reply.rp_acpt.ar_stat == SUCCESS) {
946 		if (!xdr_replymsg(&xdrs, msg))
947 			stat = FALSE;
948 		else
949 			xdrmbuf_append(&xdrs, m);
950 	} else {
951 		stat = xdr_replymsg(&xdrs, msg);
952 	}
953 
954 	if (stat) {
955 		m_fixhdr(mrep);
956 
957 		/*
958 		 * Prepend a record marker containing the reply length.
959 		 */
960 		M_PREPEND(mrep, sizeof(uint32_t), M_WAITOK);
961 		len = mrep->m_pkthdr.len;
962 		*mtod(mrep, uint32_t *) =
963 			htonl(0x80000000 | (len - sizeof(uint32_t)));
964 
965 		/* For RPC-over-TLS, copy mrep to a chain of ext_pgs. */
966 		if ((xprt->xp_tls & RPCTLS_FLAGS_HANDSHAKE) != 0) {
967 			/*
968 			 * Copy the mbuf chain to a chain of
969 			 * ext_pgs mbuf(s) as required by KERN_TLS.
970 			 */
971 			maxextsiz = TLS_MAX_MSG_SIZE_V10_2;
972 #ifdef KERN_TLS
973 			if (rpctls_getinfo(&maxlen, false, false))
974 				maxextsiz = min(maxextsiz, maxlen);
975 #endif
976 			mrep = _rpc_copym_into_ext_pgs(mrep, maxextsiz);
977 		}
978 		atomic_add_32(&xprt->xp_snd_cnt, len);
979 		/*
980 		 * sosend consumes mreq.
981 		 */
982 		error = sosend(xprt->xp_socket, NULL, NULL, mrep, NULL,
983 		    0, curthread);
984 		if (!error) {
985 			atomic_add_rel_32(&xprt->xp_snt_cnt, len);
986 			if (seq)
987 				*seq = xprt->xp_snd_cnt;
988 			stat = TRUE;
989 		} else
990 			atomic_subtract_32(&xprt->xp_snd_cnt, len);
991 	} else {
992 		m_freem(mrep);
993 	}
994 
995 	XDR_DESTROY(&xdrs);
996 
997 	return (stat);
998 }
999 
1000 static bool_t
1001 svc_vc_backchannel_reply(SVCXPRT *xprt, struct rpc_msg *msg,
1002     struct sockaddr *addr, struct mbuf *m, uint32_t *seq)
1003 {
1004 	struct ct_data *ct;
1005 	XDR xdrs;
1006 	struct mbuf *mrep;
1007 	bool_t stat = TRUE;
1008 	int error, maxextsiz;
1009 #ifdef KERN_TLS
1010 	u_int maxlen;
1011 #endif
1012 
1013 	/*
1014 	 * Leave space for record mark.
1015 	 */
1016 	mrep = m_gethdr(M_WAITOK, MT_DATA);
1017 	mrep->m_data += sizeof(uint32_t);
1018 
1019 	xdrmbuf_create(&xdrs, mrep, XDR_ENCODE);
1020 
1021 	if (msg->rm_reply.rp_stat == MSG_ACCEPTED &&
1022 	    msg->rm_reply.rp_acpt.ar_stat == SUCCESS) {
1023 		if (!xdr_replymsg(&xdrs, msg))
1024 			stat = FALSE;
1025 		else
1026 			xdrmbuf_append(&xdrs, m);
1027 	} else {
1028 		stat = xdr_replymsg(&xdrs, msg);
1029 	}
1030 
1031 	if (stat) {
1032 		m_fixhdr(mrep);
1033 
1034 		/*
1035 		 * Prepend a record marker containing the reply length.
1036 		 */
1037 		M_PREPEND(mrep, sizeof(uint32_t), M_WAITOK);
1038 		*mtod(mrep, uint32_t *) =
1039 			htonl(0x80000000 | (mrep->m_pkthdr.len
1040 				- sizeof(uint32_t)));
1041 
1042 		/* For RPC-over-TLS, copy mrep to a chain of ext_pgs. */
1043 		if ((xprt->xp_tls & RPCTLS_FLAGS_HANDSHAKE) != 0) {
1044 			/*
1045 			 * Copy the mbuf chain to a chain of
1046 			 * ext_pgs mbuf(s) as required by KERN_TLS.
1047 			 */
1048 			maxextsiz = TLS_MAX_MSG_SIZE_V10_2;
1049 #ifdef KERN_TLS
1050 			if (rpctls_getinfo(&maxlen, false, false))
1051 				maxextsiz = min(maxextsiz, maxlen);
1052 #endif
1053 			mrep = _rpc_copym_into_ext_pgs(mrep, maxextsiz);
1054 		}
1055 		sx_xlock(&xprt->xp_lock);
1056 		ct = (struct ct_data *)xprt->xp_p2;
1057 		if (ct != NULL)
1058 			error = sosend(ct->ct_socket, NULL, NULL, mrep, NULL,
1059 			    0, curthread);
1060 		else
1061 			error = EPIPE;
1062 		sx_xunlock(&xprt->xp_lock);
1063 		if (!error) {
1064 			stat = TRUE;
1065 		}
1066 	} else {
1067 		m_freem(mrep);
1068 	}
1069 
1070 	XDR_DESTROY(&xdrs);
1071 
1072 	return (stat);
1073 }
1074 
1075 static bool_t
1076 svc_vc_null(void)
1077 {
1078 
1079 	return (FALSE);
1080 }
1081 
1082 static int
1083 svc_vc_soupcall(struct socket *so, void *arg, int waitflag)
1084 {
1085 	SVCXPRT *xprt = (SVCXPRT *) arg;
1086 
1087 	if (soreadable(xprt->xp_socket))
1088 		xprt_active(xprt);
1089 	return (SU_OK);
1090 }
1091 
1092 static int
1093 svc_vc_rendezvous_soupcall(struct socket *head, void *arg, int waitflag)
1094 {
1095 	SVCXPRT *xprt = (SVCXPRT *) arg;
1096 
1097 	if (!TAILQ_EMPTY(&head->sol_comp))
1098 		xprt_active(xprt);
1099 	return (SU_OK);
1100 }
1101 
1102 #if 0
1103 /*
1104  * Get the effective UID of the sending process. Used by rpcbind, keyserv
1105  * and rpc.yppasswdd on AF_LOCAL.
1106  */
1107 int
1108 __rpc_get_local_uid(SVCXPRT *transp, uid_t *uid) {
1109 	int sock, ret;
1110 	gid_t egid;
1111 	uid_t euid;
1112 	struct sockaddr *sa;
1113 
1114 	sock = transp->xp_fd;
1115 	sa = (struct sockaddr *)transp->xp_rtaddr;
1116 	if (sa->sa_family == AF_LOCAL) {
1117 		ret = getpeereid(sock, &euid, &egid);
1118 		if (ret == 0)
1119 			*uid = euid;
1120 		return (ret);
1121 	} else
1122 		return (-1);
1123 }
1124 #endif
1125