xref: /illumos-gate/usr/src/stand/lib/fs/nfs/clnt_btcp.c (revision 4de2612967d06c4fdbf524a62556a1e8118a006f)
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, Version 1.0 only
6  * (the "License").  You may not use this file except in compliance
7  * with the License.
8  *
9  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10  * or http://www.opensolaris.org/os/licensing.
11  * See the License for the specific language governing permissions
12  * and limitations under the License.
13  *
14  * When distributing Covered Code, include this CDDL HEADER in each
15  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16  * If applicable, add the following below this CDDL HEADER, with the
17  * fields enclosed by brackets "[]" replaced with your own identifying
18  * information: Portions Copyright [yyyy] [name of copyright owner]
19  *
20  * CDDL HEADER END
21  */
22 /*
23  * Copyright 2004 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 /* Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T */
28 /* All Rights Reserved */
29 
30 /*
31  * Portions of this source code were derived from Berkeley 4.3 BSD
32  * under license from the Regents of the University of California.
33  */
34 
35 #pragma ident	"%Z%%M%	%I%	%E% SMI"
36 
37 /*
38  * Boot subsystem client side rpc (TCP)
39  */
40 
41 #include <sys/salib.h>
42 #include <sys/errno.h>
43 #include <rpc/types.h>
44 #include <sys/socket.h>
45 #include <netinet/in.h>
46 #include "socket_inet.h"
47 #include "ipv4.h"
48 #include "clnt.h"
49 #include <rpc/rpc.h>
50 #include "brpc.h"
51 #include "pmap.h"
52 #include <sys/promif.h>
53 #include <rpc/xdr.h>
54 #include <rpc/auth.h>
55 #include <rpc/auth_sys.h>
56 #include "auth_inet.h"
57 #include <rpc/rpc_msg.h>
58 #include <sys/bootdebug.h>
59 
60 #define	dprintf if (boothowto & RB_DEBUG) printf
61 
62 #define	MCALL_MSG_SIZE 24
63 
64 extern int errno;
65 
66 extern void xdrrec_create();
67 extern bool_t xdrrec_endofrecord();
68 extern bool_t xdrrec_skiprecord();
69 
70 /*
71  * If we create another clnt type this should be
72  * moved to a common file
73  */
74 struct rpc_createerr rpc_createerr;
75 
76 static int readtcp();
77 static int writetcp();
78 
79 static struct clnt_ops *clntbtcp_ops();
80 
81 /*
82  * Private data kept per client handle
83  */
84 struct ct_data {
85 	int			ct_sock;
86 	bool_t			ct_closeit;
87 	struct sockaddr_in	ct_raddr;
88 	uint_t			ct_wait_msec;
89 	struct timeval		ct_total;
90 	struct rpc_err		ct_error;
91 	XDR			ct_xdrs;
92 	char			ct_mcall[MCALL_MSG_SIZE];
93 	uint_t			ct_mpos;
94 	uint_t			ct_xdrpos;
95 };
96 
97 /*
98  * Create a TCP based client handle.
99  * If *sockp<0, *sockp is set to a newly created TCP socket.
100  * If raddr->sin_port is 0 a binder on the remote machine
101  * is consulted for the correct port number.
102  * NB: It is the clients responsibility to close *sockp.
103  * NB: The rpch->cl_auth is initialized to null authentication.
104  *	Caller may wish to set this something more useful.
105  *
106  * wait is the amount of time used between retransmitting a call if
107  * no response has been heard;  retransmition occurs until the actual
108  * rpc call times out.
109  *
110  * sendsz and recvsz are the maximum allowable packet sizes that can be
111  * sent and received.
112  */
113 CLIENT *
114 clntbtcp_create(
115 	struct sockaddr_in *raddr,
116 	rpcprog_t program,
117 	rpcvers_t version,
118 	struct timeval wait,
119 	int *sockp,
120 	uint_t sendsz,
121 	uint_t recvsz)
122 {
123 	CLIENT *cl;
124 	struct ct_data *ct;
125 	struct rpc_msg call_msg;
126 #if 0	/* XXX not yet */
127 	int min_buf_sz;
128 	int pref_buf_sz = 64 * 1024; /* 64 KB */
129 	socklen_t optlen;
130 #endif /* not yet */
131 	cl = (CLIENT *)bkmem_alloc(sizeof (CLIENT));
132 	if (cl == NULL) {
133 		errno = ENOMEM;
134 		rpc_createerr.cf_stat = RPC_SYSTEMERROR;
135 		rpc_createerr.cf_error.re_errno = errno;
136 		return ((CLIENT *)NULL);
137 	}
138 
139 	ct = (struct ct_data *)bkmem_alloc(sizeof (*ct));
140 	if (ct == NULL) {
141 		errno = ENOMEM;
142 		rpc_createerr.cf_stat = RPC_SYSTEMERROR;
143 		rpc_createerr.cf_error.re_errno = errno;
144 		goto fooy;
145 	}
146 
147 	if (raddr->sin_port == 0) {
148 		ushort_t port;
149 		if ((port = bpmap_getport(program, version,
150 				&(rpc_createerr.cf_stat), raddr, NULL)) == 0) {
151 			goto fooy;
152 		}
153 		raddr->sin_port = htons(port);
154 	}
155 
156 	if (*sockp < 0) {
157 		struct sockaddr_in from;
158 
159 		*sockp = socket(PF_INET, SOCK_STREAM, IPPROTO_TCP);
160 		if (*sockp < 0) {
161 			rpc_createerr.cf_stat = RPC_SYSTEMERROR;
162 			rpc_createerr.cf_error.re_errno = errno;
163 			goto fooy;
164 		}
165 		/*
166 		 * Bootparams assumes a local net, so be sure to let lower
167 		 * layer protocols know not to route.
168 		 */
169 		if (dontroute) {
170 			(void) setsockopt(*sockp, SOL_SOCKET, SO_DONTROUTE,
171 				(const void *)&dontroute, sizeof (dontroute));
172 		}
173 
174 		/* attempt to bind to priv port */
175 		from.sin_family = AF_INET;
176 		ipv4_getipaddr(&from.sin_addr);
177 		from.sin_addr.s_addr = htonl(from.sin_addr.s_addr);
178 		from.sin_port = get_source_port(TRUE);
179 
180 		if (bind(*sockp, (struct sockaddr *)&from, sizeof (from)) < 0) {
181 			rpc_createerr.cf_stat = RPC_SYSTEMERROR;
182 			rpc_createerr.cf_error.re_errno = errno;
183 			if (*sockp > 0)
184 				close(*sockp);
185 			goto fooy;
186 		}
187 
188 		if (connect(*sockp, (struct sockaddr *)raddr,
189 			    sizeof (struct sockaddr_in)) < 0) {
190 			rpc_createerr.cf_stat = RPC_SYSTEMERROR;
191 			rpc_createerr.cf_error.re_errno = errno;
192 			if (*sockp > 0)
193 				close(*sockp);
194 			goto fooy;
195 		}
196 
197 #if 0 /* XXX not yet */
198 		/*
199 		 * In the future we may want RPC to use larger transfer sizes
200 		 * over TCP.  In this case we will want to increase the
201 		 * window size.
202 		 */
203 		/*
204 		 * Resize the receive window if possible
205 		 */
206 		optlen = sizeof (int);
207 		if (getsockopt(*sockp, SOL_SOCKET, SO_RCVBUF,
208 				(void *)&min_buf_sz, &optlen) != 0)
209 			goto keep_going;
210 
211 		if (min_buf_sz < pref_buf_sz)
212 			(void) setsockopt(*sockp, SOL_SOCKET, SO_RCVBUF,
213 				(const void *)&pref_buf_sz, sizeof (int));
214 
215 keep_going:
216 #endif		/* not yet */
217 		ct->ct_closeit = TRUE;
218 	} else
219 		ct->ct_closeit = FALSE;
220 
221 	/*
222 	 * Set up the private data
223 	 */
224 	ct->ct_sock = *sockp;
225 	ct->ct_wait_msec = 0;
226 	ct->ct_total.tv_sec = wait.tv_sec;
227 	ct->ct_total.tv_usec = -1;
228 	ct->ct_raddr = *raddr;
229 
230 	/*
231 	 * Initialize the call message
232 	 */
233 
234 	/*
235 	 * XXX - The xid might need to be randomized more.  Imagine if there
236 	 * are a rack of blade servers all booting at the same time.  They
237 	 * may cause havoc on the server with xid replays.
238 	 */
239 	call_msg.rm_xid = (uint_t)prom_gettime() + 1;
240 	call_msg.rm_direction = CALL;
241 	call_msg.rm_call.cb_rpcvers = RPC_MSG_VERSION;
242 	call_msg.rm_call.cb_prog = program;
243 	call_msg.rm_call.cb_vers = version;
244 
245 	/*
246 	 * pre-serialize the static part of the call msg and stash it away
247 	 */
248 	xdrmem_create(&(ct->ct_xdrs), ct->ct_mcall, MCALL_MSG_SIZE,
249 			XDR_ENCODE);
250 	if (! xdr_callhdr(&(ct->ct_xdrs), &call_msg)) {
251 		if (ct->ct_closeit)
252 			(void) close(*sockp);
253 		goto fooy;
254 	}
255 	ct->ct_mpos = XDR_GETPOS(&(ct->ct_xdrs));
256 	XDR_DESTROY(&(ct->ct_xdrs));
257 
258 	/*
259 	 * XXX - Memory allocations can fail in xdrrec_create, so we need to
260 	 * be able to catch those errors.
261 	 */
262 	xdrrec_create(&(ct->ct_xdrs), sendsz, recvsz, (caddr_t)ct, readtcp,
263 			writetcp);
264 
265 	cl->cl_ops = clntbtcp_ops();
266 	cl->cl_private = (caddr_t)ct;
267 	cl->cl_auth = authnone_create();
268 	return (cl);
269 
270 fooy:
271 	if (ct)
272 		bkmem_free((caddr_t)ct, sizeof (*ct));
273 	if (cl)
274 		bkmem_free((caddr_t)cl, sizeof (CLIENT));
275 	return ((CLIENT *)NULL);
276 }
277 
278 static enum clnt_stat
279 clntbtcp_call(
280 	CLIENT *cl,
281 	rpcproc_t proc,
282 	xdrproc_t xargs,
283 	caddr_t argsp,
284 	xdrproc_t xdr_results,
285 	caddr_t resultsp,
286 	struct timeval utimeout)
287 {
288 	struct ct_data *ct;
289 	XDR *xdrs;
290 	struct rpc_msg reply_msg;
291 	uint32_t x_id;
292 	uint32_t *msg_x_id;
293 	bool_t shipnow;
294 	int nrefreshes = 2;	/* number of times to refresh cred */
295 	struct timeval timeout;
296 
297 	ct = (struct ct_data *)cl->cl_private;
298 	msg_x_id = (uint32_t *)ct->ct_mcall;
299 
300 	xdrs = &(ct->ct_xdrs);
301 
302 	ct->ct_total = utimeout;
303 
304 	/*
305 	 * We have to be able to wait for some non-zero period of time, so
306 	 * use a default timeout.
307 	 */
308 	if (ct->ct_total.tv_sec == 0)
309 		ct->ct_total.tv_sec = RPC_RCVWAIT_MSEC / 1000;
310 
311 	ct->ct_wait_msec = ct->ct_total.tv_sec * 1000 +
312 		ct->ct_total.tv_usec / 1000;
313 
314 	timeout = ct->ct_total;
315 
316 	shipnow = (xdr_results == (xdrproc_t)0 && timeout.tv_sec == 0 &&
317 			timeout.tv_usec == 0) ? FALSE : TRUE;
318 
319 call_again:
320 	xdrs->x_op = XDR_ENCODE;
321 	ct->ct_error.re_status = RPC_SUCCESS;
322 	x_id = ntohl(++(*msg_x_id));
323 	if ((! XDR_PUTBYTES(xdrs, ct->ct_mcall, ct->ct_mpos)) ||
324 	    (! XDR_PUTINT32(xdrs, (int32_t *)&proc)) ||
325 	    (! AUTH_MARSHALL(cl->cl_auth, xdrs, NULL)) ||
326 	    (! (*xargs)(xdrs, argsp))) {
327 		(void) xdrrec_endofrecord(xdrs, TRUE);
328 		ct->ct_error.re_status = RPC_CANTENCODEARGS;
329 		printf("clntbtcp_call: xdr encode args failed\n");
330 		return (ct->ct_error.re_status);
331 	}
332 
333 	if (!xdrrec_endofrecord(xdrs, shipnow)) {
334 		printf("clntbtcp_call: rpc cansend error\n");
335 		ct->ct_error.re_status = RPC_CANTSEND;
336 		return (ct->ct_error.re_status);
337 	}
338 
339 	if (!shipnow)
340 		return (RPC_SUCCESS);
341 
342 	if (timeout.tv_sec == 0 && timeout.tv_usec == 0) {
343 		ct->ct_error.re_status = RPC_TIMEDOUT;
344 		return (ct->ct_error.re_status);
345 	}
346 
347 	xdrs->x_op = XDR_DECODE;
348 
349 	/* CONSTCOND */
350 	while (TRUE) {
351 		reply_msg.acpted_rply.ar_verf = _null_auth;
352 		reply_msg.acpted_rply.ar_results.where = NULL;
353 		reply_msg.acpted_rply.ar_results.proc = xdr_void;
354 		if (!xdrrec_skiprecord(xdrs)) {
355 			return (ct->ct_error.re_status);
356 		}
357 
358 		if (!xdr_replymsg(xdrs, &reply_msg)) {
359 			if (ct->ct_error.re_status == RPC_SUCCESS)
360 				continue;
361 			return (ct->ct_error.re_status);
362 		}
363 		if (reply_msg.rm_xid == x_id) {
364 			break;
365 		}
366 	}
367 
368 	/*
369 	 * process header
370 	 */
371 	_seterr_reply(&reply_msg, &(ct->ct_error));
372 	if (ct->ct_error.re_status == RPC_SUCCESS) {
373 		if (!AUTH_VALIDATE(cl->cl_auth,
374 				&reply_msg.acpted_rply.ar_verf)) {
375 			ct->ct_error.re_status = RPC_AUTHERROR;
376 			ct->ct_error.re_why = AUTH_INVALIDRESP;
377 		} else if (!(*xdr_results)(xdrs, resultsp)) {
378 			if (ct->ct_error.re_status == RPC_SUCCESS) {
379 				ct->ct_error.re_status = RPC_CANTDECODERES;
380 			}
381 		}
382 		if (reply_msg.acpted_rply.ar_verf.oa_base != NULL) {
383 			xdrs->x_op = XDR_FREE;
384 			(void) xdr_opaque_auth(xdrs,
385 				&(reply_msg.acpted_rply.ar_verf));
386 		}
387 	} else {
388 		if (nrefreshes-- && AUTH_REFRESH(cl->cl_auth, &reply_msg,
389 						NULL)) {
390 			goto call_again;
391 		}
392 	}
393 	return (ct->ct_error.re_status);
394 }
395 
396 /*
397  * Interface between xdr serializer and tcp connection.
398  * Behaves like the system calls, read & write, but keeps some error state
399  * around for the rpc level.
400  */
401 static int
402 readtcp(struct ct_data *ct,
403 	caddr_t buf,
404 	int len)
405 {
406 	int inlen = 0;
407 	uint_t start, diff;
408 	struct sockaddr from;
409 	uint_t fromlen = sizeof (from);
410 
411 	if (len <= 0)
412 		return (0);
413 
414 	/*
415 	 * Do non-blocking reads here until we get some data or timeout
416 	 */
417 	start = prom_gettime();
418 	while ((inlen = recvfrom(ct->ct_sock, buf, len, 0, &from,
419 					&fromlen)) == 0) {
420 		diff = (uint_t)(prom_gettime() - start);
421 		if (diff > ct->ct_wait_msec) {
422 			errno = ETIMEDOUT;
423 			inlen = -1;
424 			break;
425 		}
426 	}
427 #ifdef DEBUG
428 	printf("readtcp: inlen = %d\n", inlen);
429 #endif
430 	switch (inlen) {
431 	case 0:
432 		/* premature eof */
433 		ct->ct_error.re_errno = ECONNRESET;
434 		ct->ct_error.re_status = RPC_CANTRECV;
435 		inlen = -1;  /* it's really an error */
436 		break;
437 	case -1:
438 		ct->ct_error.re_errno = errno;
439 		ct->ct_error.re_status = RPC_CANTRECV;
440 		break;
441 	}
442 
443 	return (inlen);
444 }
445 
446 static int
447 writetcp(ct, buf, len)
448 	struct ct_data *ct;
449 	caddr_t buf;
450 	int len;
451 {
452 	register int i, cnt;
453 
454 	for (cnt = len; cnt > 0; cnt -= i, buf += i) {
455 		if ((i = sendto(ct->ct_sock, (void *)buf, cnt, 0,
456 				(struct sockaddr *)&(ct->ct_raddr),
457 				sizeof (ct->ct_raddr))) == -1) {
458 			ct->ct_error.re_errno = errno;
459 			ct->ct_error.re_status = RPC_CANTSEND;
460 			return (-1);
461 		}
462 	}
463 	return (len);
464 }
465 
466 static void
467 clntbtcp_geterr(
468 	CLIENT *cl,
469 	struct rpc_err *errp)
470 {
471 	struct ct_data *ct = (struct ct_data *)cl->cl_private;
472 
473 	*errp = ct->ct_error;
474 }
475 
476 
477 static bool_t
478 clntbtcp_freeres(
479 	CLIENT *cl,
480 	xdrproc_t xdr_res,
481 	caddr_t res_ptr)
482 {
483 	struct ct_data *ct = (struct ct_data *)cl->cl_private;
484 	XDR *xdrs = &(ct->ct_xdrs);
485 
486 	xdrs->x_op = XDR_FREE;
487 	return ((*xdr_res)(xdrs, res_ptr));
488 }
489 
490 static void
491 clntbtcp_abort()
492 	/* CLIENT *h; */
493 {
494 }
495 
496 /* ARGSUSED */
497 static bool_t
498 clntbtcp_control(
499 	CLIENT *cl,
500 	int request,
501 	char *info)
502 {
503 	/* Not implemented in boot */
504 	return (FALSE);
505 }
506 
507 static void
508 clntbtcp_destroy(CLIENT *cl)
509 {
510 	struct ct_data *ct = (struct ct_data *)cl->cl_private;
511 
512 	if (ct->ct_closeit) {
513 		(void) socket_close(ct->ct_sock);
514 	}
515 	XDR_DESTROY(&(ct->ct_xdrs));
516 	bkmem_free((caddr_t)ct, (sizeof (struct ct_data)));
517 	bkmem_free((caddr_t)cl, sizeof (CLIENT));
518 }
519 
520 static struct clnt_ops *
521 clntbtcp_ops()
522 {
523 	static struct clnt_ops ops;
524 
525 	if (ops.cl_call == NULL) {
526 		ops.cl_call = clntbtcp_call;
527 		ops.cl_abort = clntbtcp_abort;
528 		ops.cl_geterr = clntbtcp_geterr;
529 		ops.cl_freeres = clntbtcp_freeres;
530 		ops.cl_destroy = clntbtcp_destroy;
531 		ops.cl_control = clntbtcp_control;
532 	}
533 	return (&ops);
534 }
535