1/*	$NetBSD: clnt_vc.c,v 1.4 2000/07/14 08:40:42 fvdl Exp $	*/
2
3/*-
4 * Copyright (c) 2009, Sun Microsystems, Inc.
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions are met:
9 * - Redistributions of source code must retain the above copyright notice,
10 *   this list of conditions and the following disclaimer.
11 * - Redistributions in binary form must reproduce the above copyright notice,
12 *   this list of conditions and the following disclaimer in the documentation
13 *   and/or other materials provided with the distribution.
14 * - Neither the name of Sun Microsystems, Inc. nor the names of its
15 *   contributors may be used to endorse or promote products derived
16 *   from this software without specific prior written permission.
17 *
18 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
19 * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21 * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
22 * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
23 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
24 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
25 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
26 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
27 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
28 * POSSIBILITY OF SUCH DAMAGE.
29 */
30
31#if defined(LIBC_SCCS) && !defined(lint)
32static char *sccsid2 = "@(#)clnt_tcp.c 1.37 87/10/05 Copyr 1984 Sun Micro";
33static char *sccsid = "@(#)clnt_tcp.c	2.2 88/08/01 4.0 RPCSRC";
34static char sccsid3[] = "@(#)clnt_vc.c 1.19 89/03/16 Copyr 1988 Sun Micro";
35#endif
36#include <sys/cdefs.h>
37__FBSDID("$FreeBSD$");
38
39/*
40 * clnt_tcp.c, Implements a TCP/IP based, client side RPC.
41 *
42 * Copyright (C) 1984, Sun Microsystems, Inc.
43 *
44 * TCP based RPC supports 'batched calls'.
45 * A sequence of calls may be batched-up in a send buffer.  The rpc call
46 * return immediately to the client even though the call was not necessarily
47 * sent.  The batching occurs if the results' xdr routine is NULL (0) AND
48 * the rpc timeout value is zero (see clnt.h, rpc).
49 *
50 * Clients should NOT casually batch calls that in fact return results; that is,
51 * the server side should be aware that a call is batched and not produce any
52 * return message.  Batched calls that produce many result messages can
53 * deadlock (netlock) the client and the server....
54 *
55 * Now go hang yourself.
56 */
57
58/*
59 * This code handles the special case of a NFSv4.n backchannel for
60 * callback RPCs. It is similar to clnt_vc.c, but uses the TCP
61 * connection provided by the client to the server.
62 */
63
64#include <sys/param.h>
65#include <sys/systm.h>
66#include <sys/lock.h>
67#include <sys/malloc.h>
68#include <sys/mbuf.h>
69#include <sys/mutex.h>
70#include <sys/pcpu.h>
71#include <sys/proc.h>
72#include <sys/protosw.h>
73#include <sys/socket.h>
74#include <sys/socketvar.h>
75#include <sys/sx.h>
76#include <sys/syslog.h>
77#include <sys/time.h>
78#include <sys/uio.h>
79
80#include <net/vnet.h>
81
82#include <netinet/tcp.h>
83
84#include <rpc/rpc.h>
85#include <rpc/rpc_com.h>
86#include <rpc/krpc.h>
87
88struct cmessage {
89        struct cmsghdr cmsg;
90        struct cmsgcred cmcred;
91};
92
93static void clnt_bck_geterr(CLIENT *, struct rpc_err *);
94static bool_t clnt_bck_freeres(CLIENT *, xdrproc_t, void *);
95static void clnt_bck_abort(CLIENT *);
96static bool_t clnt_bck_control(CLIENT *, u_int, void *);
97static void clnt_bck_close(CLIENT *);
98static void clnt_bck_destroy(CLIENT *);
99
100static struct clnt_ops clnt_bck_ops = {
101	.cl_abort =	clnt_bck_abort,
102	.cl_geterr =	clnt_bck_geterr,
103	.cl_freeres =	clnt_bck_freeres,
104	.cl_close =	clnt_bck_close,
105	.cl_destroy =	clnt_bck_destroy,
106	.cl_control =	clnt_bck_control
107};
108
109/*
110 * Create a client handle for a connection.
111 * Default options are set, which the user can change using clnt_control()'s.
112 * This code handles the special case of an NFSv4.1 session backchannel
113 * call, which is sent on a TCP connection created against the server
114 * by a client.
115 */
116void *
117clnt_bck_create(
118	struct socket *so,		/* Server transport socket. */
119	const rpcprog_t prog,		/* program number */
120	const rpcvers_t vers)		/* version number */
121{
122	CLIENT *cl;			/* client handle */
123	struct ct_data *ct = NULL;	/* client handle */
124	struct timeval now;
125	struct rpc_msg call_msg;
126	static uint32_t disrupt;
127	XDR xdrs;
128
129	if (disrupt == 0)
130		disrupt = (uint32_t)(long)so;
131
132	cl = (CLIENT *)mem_alloc(sizeof (*cl));
133	ct = (struct ct_data *)mem_alloc(sizeof (*ct));
134
135	mtx_init(&ct->ct_lock, "ct->ct_lock", NULL, MTX_DEF);
136	ct->ct_threads = 0;
137	ct->ct_closing = FALSE;
138	ct->ct_closed = FALSE;
139	ct->ct_upcallrefs = 0;
140	ct->ct_closeit = FALSE;
141
142	/*
143	 * Set up private data struct
144	 */
145	ct->ct_wait.tv_sec = -1;
146	ct->ct_wait.tv_usec = -1;
147
148	/*
149	 * Initialize call message
150	 */
151	getmicrotime(&now);
152	ct->ct_xid = ((uint32_t)++disrupt) ^ __RPC_GETXID(&now);
153	call_msg.rm_xid = ct->ct_xid;
154	call_msg.rm_direction = CALL;
155	call_msg.rm_call.cb_rpcvers = RPC_MSG_VERSION;
156	call_msg.rm_call.cb_prog = (uint32_t)prog;
157	call_msg.rm_call.cb_vers = (uint32_t)vers;
158
159	/*
160	 * pre-serialize the static part of the call msg and stash it away
161	 */
162	xdrmem_create(&xdrs, ct->ct_mcallc, MCALL_MSG_SIZE,
163	    XDR_ENCODE);
164	if (!xdr_callhdr(&xdrs, &call_msg))
165		goto err;
166	ct->ct_mpos = XDR_GETPOS(&xdrs);
167	XDR_DESTROY(&xdrs);
168	ct->ct_waitchan = "rpcbck";
169	ct->ct_waitflag = 0;
170	cl->cl_refs = 1;
171	cl->cl_ops = &clnt_bck_ops;
172	cl->cl_private = ct;
173	cl->cl_auth = authnone_create();
174	TAILQ_INIT(&ct->ct_pending);
175	return (cl);
176
177err:
178	mtx_destroy(&ct->ct_lock);
179	mem_free(ct, sizeof (struct ct_data));
180	mem_free(cl, sizeof (CLIENT));
181	return (NULL);
182}
183
184enum clnt_stat
185clnt_bck_call(
186	CLIENT		*cl,		/* client handle */
187	struct rpc_callextra *ext,	/* call metadata */
188	rpcproc_t	proc,		/* procedure number */
189	struct mbuf	*args,		/* pointer to args */
190	struct mbuf	**resultsp,	/* pointer to results */
191	struct timeval	utimeout,
192	SVCXPRT		*xprt)
193{
194	struct ct_data *ct = (struct ct_data *) cl->cl_private;
195	AUTH *auth;
196	struct rpc_err *errp;
197	enum clnt_stat stat;
198	XDR xdrs;
199	struct rpc_msg reply_msg;
200	bool_t ok;
201	int nrefreshes = 2;		/* number of times to refresh cred */
202	struct timeval timeout;
203	uint32_t xid;
204	struct mbuf *mreq = NULL, *results;
205	struct ct_request *cr;
206	int error;
207
208	cr = malloc(sizeof(struct ct_request), M_RPC, M_WAITOK);
209
210	mtx_lock(&ct->ct_lock);
211
212	if (ct->ct_closing || ct->ct_closed) {
213		mtx_unlock(&ct->ct_lock);
214		free(cr, M_RPC);
215		return (RPC_CANTSEND);
216	}
217	ct->ct_threads++;
218
219	if (ext) {
220		auth = ext->rc_auth;
221		errp = &ext->rc_err;
222	} else {
223		auth = cl->cl_auth;
224		errp = &ct->ct_error;
225	}
226
227	cr->cr_mrep = NULL;
228	cr->cr_error = 0;
229
230	if (ct->ct_wait.tv_usec == -1)
231		timeout = utimeout;	/* use supplied timeout */
232	else
233		timeout = ct->ct_wait;	/* use default timeout */
234
235call_again:
236	mtx_assert(&ct->ct_lock, MA_OWNED);
237
238	ct->ct_xid++;
239	xid = ct->ct_xid;
240
241	mtx_unlock(&ct->ct_lock);
242
243	/*
244	 * Leave space to pre-pend the record mark.
245	 */
246	mreq = m_gethdr(M_WAITOK, MT_DATA);
247	mreq->m_data += sizeof(uint32_t);
248	KASSERT(ct->ct_mpos + sizeof(uint32_t) <= MHLEN,
249	    ("RPC header too big"));
250	bcopy(ct->ct_mcallc, mreq->m_data, ct->ct_mpos);
251	mreq->m_len = ct->ct_mpos;
252
253	/*
254	 * The XID is the first thing in the request.
255	 */
256	*mtod(mreq, uint32_t *) = htonl(xid);
257
258	xdrmbuf_create(&xdrs, mreq, XDR_ENCODE);
259
260	errp->re_status = stat = RPC_SUCCESS;
261
262	if ((!XDR_PUTINT32(&xdrs, &proc)) ||
263	    (!AUTH_MARSHALL(auth, xid, &xdrs,
264	     m_copym(args, 0, M_COPYALL, M_WAITOK)))) {
265		errp->re_status = stat = RPC_CANTENCODEARGS;
266		mtx_lock(&ct->ct_lock);
267		goto out;
268	}
269	mreq->m_pkthdr.len = m_length(mreq, NULL);
270
271	/*
272	 * Prepend a record marker containing the packet length.
273	 */
274	M_PREPEND(mreq, sizeof(uint32_t), M_WAITOK);
275	*mtod(mreq, uint32_t *) =
276	    htonl(0x80000000 | (mreq->m_pkthdr.len - sizeof(uint32_t)));
277
278	cr->cr_xid = xid;
279	mtx_lock(&ct->ct_lock);
280	/*
281	 * Check to see if the client end has already started to close down
282	 * the connection. The svc code will have set ct_error.re_status
283	 * to RPC_CANTRECV if this is the case.
284	 * If the client starts to close down the connection after this
285	 * point, it will be detected later when cr_error is checked,
286	 * since the request is in the ct_pending queue.
287	 */
288	if (ct->ct_error.re_status == RPC_CANTRECV) {
289		if (errp != &ct->ct_error) {
290			errp->re_errno = ct->ct_error.re_errno;
291			errp->re_status = RPC_CANTRECV;
292		}
293		stat = RPC_CANTRECV;
294		goto out;
295	}
296	TAILQ_INSERT_TAIL(&ct->ct_pending, cr, cr_link);
297	mtx_unlock(&ct->ct_lock);
298
299	/*
300	 * sosend consumes mreq.
301	 */
302	sx_xlock(&xprt->xp_lock);
303	error = sosend(xprt->xp_socket, NULL, NULL, mreq, NULL, 0, curthread);
304if (error != 0) printf("sosend=%d\n", error);
305	mreq = NULL;
306	if (error == EMSGSIZE) {
307printf("emsgsize\n");
308		SOCKBUF_LOCK(&xprt->xp_socket->so_snd);
309		sbwait(&xprt->xp_socket->so_snd);
310		SOCKBUF_UNLOCK(&xprt->xp_socket->so_snd);
311		sx_xunlock(&xprt->xp_lock);
312		AUTH_VALIDATE(auth, xid, NULL, NULL);
313		mtx_lock(&ct->ct_lock);
314		TAILQ_REMOVE(&ct->ct_pending, cr, cr_link);
315		goto call_again;
316	}
317	sx_xunlock(&xprt->xp_lock);
318
319	reply_msg.acpted_rply.ar_verf.oa_flavor = AUTH_NULL;
320	reply_msg.acpted_rply.ar_verf.oa_base = cr->cr_verf;
321	reply_msg.acpted_rply.ar_verf.oa_length = 0;
322	reply_msg.acpted_rply.ar_results.where = NULL;
323	reply_msg.acpted_rply.ar_results.proc = (xdrproc_t)xdr_void;
324
325	mtx_lock(&ct->ct_lock);
326	if (error) {
327		TAILQ_REMOVE(&ct->ct_pending, cr, cr_link);
328		errp->re_errno = error;
329		errp->re_status = stat = RPC_CANTSEND;
330		goto out;
331	}
332
333	/*
334	 * Check to see if we got an upcall while waiting for the
335	 * lock. In both these cases, the request has been removed
336	 * from ct->ct_pending.
337	 */
338	if (cr->cr_error) {
339		TAILQ_REMOVE(&ct->ct_pending, cr, cr_link);
340		errp->re_errno = cr->cr_error;
341		errp->re_status = stat = RPC_CANTRECV;
342		goto out;
343	}
344	if (cr->cr_mrep) {
345		TAILQ_REMOVE(&ct->ct_pending, cr, cr_link);
346		goto got_reply;
347	}
348
349	/*
350	 * Hack to provide rpc-based message passing
351	 */
352	if (timeout.tv_sec == 0 && timeout.tv_usec == 0) {
353		TAILQ_REMOVE(&ct->ct_pending, cr, cr_link);
354		errp->re_status = stat = RPC_TIMEDOUT;
355		goto out;
356	}
357
358	error = msleep(cr, &ct->ct_lock, ct->ct_waitflag, ct->ct_waitchan,
359	    tvtohz(&timeout));
360
361	TAILQ_REMOVE(&ct->ct_pending, cr, cr_link);
362
363	if (error) {
364		/*
365		 * The sleep returned an error so our request is still
366		 * on the list. Turn the error code into an
367		 * appropriate client status.
368		 */
369		errp->re_errno = error;
370		switch (error) {
371		case EINTR:
372			stat = RPC_INTR;
373			break;
374		case EWOULDBLOCK:
375			stat = RPC_TIMEDOUT;
376			break;
377		default:
378			stat = RPC_CANTRECV;
379		};
380		errp->re_status = stat;
381		goto out;
382	} else {
383		/*
384		 * We were woken up by the svc thread.  If the
385		 * upcall had a receive error, report that,
386		 * otherwise we have a reply.
387		 */
388		if (cr->cr_error) {
389			errp->re_errno = cr->cr_error;
390			errp->re_status = stat = RPC_CANTRECV;
391			goto out;
392		}
393	}
394
395got_reply:
396	/*
397	 * Now decode and validate the response. We need to drop the
398	 * lock since xdr_replymsg may end up sleeping in malloc.
399	 */
400	mtx_unlock(&ct->ct_lock);
401
402	if (ext && ext->rc_feedback)
403		ext->rc_feedback(FEEDBACK_OK, proc, ext->rc_feedback_arg);
404
405	xdrmbuf_create(&xdrs, cr->cr_mrep, XDR_DECODE);
406	ok = xdr_replymsg(&xdrs, &reply_msg);
407	cr->cr_mrep = NULL;
408
409	if (ok) {
410		if ((reply_msg.rm_reply.rp_stat == MSG_ACCEPTED) &&
411		    (reply_msg.acpted_rply.ar_stat == SUCCESS))
412			errp->re_status = stat = RPC_SUCCESS;
413		else
414			stat = _seterr_reply(&reply_msg, errp);
415
416		if (stat == RPC_SUCCESS) {
417			results = xdrmbuf_getall(&xdrs);
418			if (!AUTH_VALIDATE(auth, xid,
419			    &reply_msg.acpted_rply.ar_verf, &results)) {
420				errp->re_status = stat = RPC_AUTHERROR;
421				errp->re_why = AUTH_INVALIDRESP;
422			} else {
423				KASSERT(results,
424				    ("auth validated but no result"));
425				*resultsp = results;
426			}
427		}		/* end successful completion */
428		/*
429		 * If unsuccesful AND error is an authentication error
430		 * then refresh credentials and try again, else break
431		 */
432		else if (stat == RPC_AUTHERROR)
433			/* maybe our credentials need to be refreshed ... */
434			if (nrefreshes > 0 && AUTH_REFRESH(auth, &reply_msg)) {
435				nrefreshes--;
436				XDR_DESTROY(&xdrs);
437				mtx_lock(&ct->ct_lock);
438				goto call_again;
439			}
440			/* end of unsuccessful completion */
441		/* end of valid reply message */
442	} else
443		errp->re_status = stat = RPC_CANTDECODERES;
444	XDR_DESTROY(&xdrs);
445	mtx_lock(&ct->ct_lock);
446out:
447	mtx_assert(&ct->ct_lock, MA_OWNED);
448
449	KASSERT(stat != RPC_SUCCESS || *resultsp,
450	    ("RPC_SUCCESS without reply"));
451
452	if (mreq != NULL)
453		m_freem(mreq);
454	if (cr->cr_mrep != NULL)
455		m_freem(cr->cr_mrep);
456
457	ct->ct_threads--;
458	if (ct->ct_closing)
459		wakeup(ct);
460
461	mtx_unlock(&ct->ct_lock);
462
463	if (auth && stat != RPC_SUCCESS)
464		AUTH_VALIDATE(auth, xid, NULL, NULL);
465
466	free(cr, M_RPC);
467
468	return (stat);
469}
470
471static void
472clnt_bck_geterr(CLIENT *cl, struct rpc_err *errp)
473{
474	struct ct_data *ct = (struct ct_data *) cl->cl_private;
475
476	*errp = ct->ct_error;
477}
478
479static bool_t
480clnt_bck_freeres(CLIENT *cl, xdrproc_t xdr_res, void *res_ptr)
481{
482	XDR xdrs;
483	bool_t dummy;
484
485	xdrs.x_op = XDR_FREE;
486	dummy = (*xdr_res)(&xdrs, res_ptr);
487
488	return (dummy);
489}
490
491/*ARGSUSED*/
492static void
493clnt_bck_abort(CLIENT *cl)
494{
495}
496
497static bool_t
498clnt_bck_control(CLIENT *cl, u_int request, void *info)
499{
500
501	return (TRUE);
502}
503
504static void
505clnt_bck_close(CLIENT *cl)
506{
507	struct ct_data *ct = (struct ct_data *) cl->cl_private;
508
509	mtx_lock(&ct->ct_lock);
510
511	if (ct->ct_closed) {
512		mtx_unlock(&ct->ct_lock);
513		return;
514	}
515
516	if (ct->ct_closing) {
517		while (ct->ct_closing)
518			msleep(ct, &ct->ct_lock, 0, "rpcclose", 0);
519		KASSERT(ct->ct_closed, ("client should be closed"));
520		mtx_unlock(&ct->ct_lock);
521		return;
522	}
523
524	ct->ct_closing = FALSE;
525	ct->ct_closed = TRUE;
526	mtx_unlock(&ct->ct_lock);
527	wakeup(ct);
528}
529
530static void
531clnt_bck_destroy(CLIENT *cl)
532{
533	struct ct_data *ct = (struct ct_data *) cl->cl_private;
534
535	clnt_bck_close(cl);
536
537	mtx_destroy(&ct->ct_lock);
538	mem_free(ct, sizeof(struct ct_data));
539	if (cl->cl_netid && cl->cl_netid[0])
540		mem_free(cl->cl_netid, strlen(cl->cl_netid) +1);
541	if (cl->cl_tp && cl->cl_tp[0])
542		mem_free(cl->cl_tp, strlen(cl->cl_tp) +1);
543	mem_free(cl, sizeof(CLIENT));
544}
545
546/*
547 * This call is done by the svc code when a backchannel RPC reply is
548 * received.
549 */
550void
551clnt_bck_svccall(void *arg, struct mbuf *mrep, uint32_t xid)
552{
553	struct ct_data *ct = (struct ct_data *)arg;
554	struct ct_request *cr;
555	int foundreq;
556
557	mtx_lock(&ct->ct_lock);
558	ct->ct_upcallrefs++;
559	/*
560	 * See if we can match this reply to a request.
561	 */
562	foundreq = 0;
563	TAILQ_FOREACH(cr, &ct->ct_pending, cr_link) {
564		if (cr->cr_xid == xid) {
565			/*
566			 * This one matches. We leave the reply mbuf list in
567			 * cr->cr_mrep. Set the XID to zero so that we will
568			 * ignore any duplicated replies.
569			 */
570			cr->cr_xid = 0;
571			cr->cr_mrep = mrep;
572			cr->cr_error = 0;
573			foundreq = 1;
574			wakeup(cr);
575			break;
576		}
577	}
578
579	ct->ct_upcallrefs--;
580	if (ct->ct_upcallrefs < 0)
581		panic("rpcvc svccall refcnt");
582	if (ct->ct_upcallrefs == 0)
583		wakeup(&ct->ct_upcallrefs);
584	mtx_unlock(&ct->ct_lock);
585	if (foundreq == 0)
586		m_freem(mrep);
587}
588
589