xref: /illumos-gate/usr/src/uts/common/rpc/svc_cots.c (revision c65ebfc7045424bd04a6c7719a27b0ad3399ad54)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright 2015 Nexenta Systems, Inc.  All rights reserved.
24  * Copyright (c) 1993, 2010, Oracle and/or its affiliates. All rights reserved.
25  * Copyright (c) 2012 by Delphix. All rights reserved.
26  * Copyright 2012 Marcel Telka <marcel@telka.sk>
27  * Copyright 2018 OmniOS Community Edition (OmniOSce) Association.
28  */
29 
30 /*	Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T	*/
31 /*	All Rights Reserved	*/
32 
33 /*
34  * Portions of this source code were derived from Berkeley 4.3 BSD
35  * under license from the Regents of the University of California.
36  */
37 
38 /*
39  * svc_cots.c
40  * Server side for connection-oriented RPC in the kernel.
41  *
42  */
43 
44 #include <sys/param.h>
45 #include <sys/types.h>
46 #include <sys/sysmacros.h>
47 #include <sys/file.h>
48 #include <sys/stream.h>
49 #include <sys/strsubr.h>
50 #include <sys/strsun.h>
51 #include <sys/stropts.h>
52 #include <sys/tiuser.h>
53 #include <sys/timod.h>
54 #include <sys/tihdr.h>
55 #include <sys/fcntl.h>
56 #include <sys/errno.h>
57 #include <sys/kmem.h>
58 #include <sys/systm.h>
59 #include <sys/debug.h>
60 #include <sys/cmn_err.h>
61 #include <sys/kstat.h>
62 #include <sys/vtrace.h>
63 
64 #include <rpc/types.h>
65 #include <rpc/xdr.h>
66 #include <rpc/auth.h>
67 #include <rpc/rpc_msg.h>
68 #include <rpc/svc.h>
69 #include <inet/ip.h>
70 
71 #define	COTS_MAX_ALLOCSIZE	2048
72 #define	MSG_OFFSET		128	/* offset of call into the mblk */
73 #define	RM_HDR_SIZE		4	/* record mark header size */
74 
75 /*
76  * Routines exported through ops vector.
77  */
78 static bool_t		svc_cots_krecv(SVCXPRT *, mblk_t *, struct rpc_msg *);
79 static bool_t		svc_cots_ksend(SVCXPRT *, struct rpc_msg *);
80 static bool_t		svc_cots_kgetargs(SVCXPRT *, xdrproc_t, caddr_t);
81 static bool_t		svc_cots_kfreeargs(SVCXPRT *, xdrproc_t, caddr_t);
82 static void		svc_cots_kdestroy(SVCMASTERXPRT *);
83 static int		svc_cots_kdup(struct svc_req *, caddr_t, int,
84 				struct dupreq **, bool_t *);
85 static void		svc_cots_kdupdone(struct dupreq *, caddr_t,
86 				void (*)(), int, int);
87 static int32_t		*svc_cots_kgetres(SVCXPRT *, int);
88 static void		svc_cots_kfreeres(SVCXPRT *);
89 static void		svc_cots_kclone_destroy(SVCXPRT *);
90 static void		svc_cots_kstart(SVCMASTERXPRT *);
91 static void		svc_cots_ktattrs(SVCXPRT *, int, void **);
92 
93 /*
94  * Server transport operations vector.
95  */
96 struct svc_ops svc_cots_op = {
97 	svc_cots_krecv,		/* Get requests */
98 	svc_cots_kgetargs,	/* Deserialize arguments */
99 	svc_cots_ksend,		/* Send reply */
100 	svc_cots_kfreeargs,	/* Free argument data space */
101 	svc_cots_kdestroy,	/* Destroy transport handle */
102 	svc_cots_kdup,		/* Check entry in dup req cache */
103 	svc_cots_kdupdone,	/* Mark entry in dup req cache as done */
104 	svc_cots_kgetres,	/* Get pointer to response buffer */
105 	svc_cots_kfreeres,	/* Destroy pre-serialized response header */
106 	svc_cots_kclone_destroy, /* Destroy a clone xprt */
107 	svc_cots_kstart,	/* Tell `ready-to-receive' to rpcmod */
108 	NULL,			/* Transport specific clone xprt */
109 	svc_cots_ktattrs,	/* Transport Attributes */
110 	mir_svc_hold,		/* Increment transport reference count */
111 	mir_svc_release		/* Decrement transport reference count */
112 };
113 
114 /*
115  * Master transport private data.
116  * Kept in xprt->xp_p2.
117  */
118 struct cots_master_data {
119 	char	*cmd_src_addr;	/* client's address */
120 	int	cmd_xprt_started; /* flag for clone routine to call */
121 				/* rpcmod's start routine. */
122 	struct rpc_cots_server *cmd_stats;	/* stats for zone */
123 };
124 
125 /*
126  * Transport private data.
127  * Kept in clone_xprt->xp_p2buf.
128  */
129 typedef struct cots_data {
130 	mblk_t	*cd_mp;		/* pre-allocated reply message */
131 	mblk_t	*cd_req_mp;	/* request message */
132 } cots_data_t;
133 
134 /*
135  * Server statistics
136  * NOTE: This structure type is duplicated in the NFS fast path.
137  */
138 static const struct rpc_cots_server {
139 	kstat_named_t	rscalls;
140 	kstat_named_t	rsbadcalls;
141 	kstat_named_t	rsnullrecv;
142 	kstat_named_t	rsbadlen;
143 	kstat_named_t	rsxdrcall;
144 	kstat_named_t	rsdupchecks;
145 	kstat_named_t	rsdupreqs;
146 } cots_rsstat_tmpl = {
147 	{ "calls",	KSTAT_DATA_UINT64 },
148 	{ "badcalls",	KSTAT_DATA_UINT64 },
149 	{ "nullrecv",	KSTAT_DATA_UINT64 },
150 	{ "badlen",	KSTAT_DATA_UINT64 },
151 	{ "xdrcall",	KSTAT_DATA_UINT64 },
152 	{ "dupchecks",	KSTAT_DATA_UINT64 },
153 	{ "dupreqs",	KSTAT_DATA_UINT64 }
154 };
155 
156 #define	CLONE2STATS(clone_xprt)	\
157 	((struct cots_master_data *)(clone_xprt)->xp_master->xp_p2)->cmd_stats
158 #define	RSSTAT_INCR(s, x)	\
159 	atomic_inc_64(&(s)->x.value.ui64)
160 
161 /*
162  * Pointer to a transport specific `ready to receive' function in rpcmod
163  * (set from rpcmod).
164  */
165 void    (*mir_start)(queue_t *);
166 uint_t	*svc_max_msg_sizep;
167 
168 /*
169  * the address size of the underlying transport can sometimes be
170  * unknown (tinfo->ADDR_size == -1).  For this case, it is
171  * necessary to figure out what the size is so the correct amount
172  * of data is allocated.  This is an itterative process:
173  *	1. take a good guess (use T_MINADDRSIZE)
174  *	2. try it.
175  *	3. if it works then everything is ok
176  *	4. if the error is ENAMETOLONG, double the guess
177  *	5. go back to step 2.
178  */
179 #define	T_UNKNOWNADDRSIZE	(-1)
180 #define	T_MINADDRSIZE	32
181 
182 /*
183  * Create a transport record.
184  * The transport record, output buffer, and private data structure
185  * are allocated.  The output buffer is serialized into using xdrmem.
186  * There is one transport record per user process which implements a
187  * set of services.
188  */
189 static kmutex_t cots_kcreate_lock;
190 
191 int
192 svc_cots_kcreate(file_t *fp, uint_t max_msgsize, struct T_info_ack *tinfo,
193     SVCMASTERXPRT **nxprt)
194 {
195 	struct cots_master_data *cmd;
196 	int err, retval;
197 	SVCMASTERXPRT *xprt;
198 	struct rpcstat *rpcstat;
199 	struct T_addr_ack *ack_p;
200 	struct strioctl getaddr;
201 
202 	if (nxprt == NULL)
203 		return (EINVAL);
204 
205 	rpcstat = zone_getspecific(rpcstat_zone_key, curproc->p_zone);
206 	ASSERT(rpcstat != NULL);
207 
208 	xprt = kmem_zalloc(sizeof (SVCMASTERXPRT), KM_SLEEP);
209 
210 	cmd = kmem_zalloc(sizeof (*cmd) + sizeof (*ack_p)
211 	    + (2 * sizeof (sin6_t)), KM_SLEEP);
212 
213 	ack_p = (struct T_addr_ack *)&cmd[1];
214 
215 	if ((tinfo->TIDU_size > COTS_MAX_ALLOCSIZE) ||
216 	    (tinfo->TIDU_size <= 0))
217 		xprt->xp_msg_size = COTS_MAX_ALLOCSIZE;
218 	else {
219 		xprt->xp_msg_size = tinfo->TIDU_size -
220 		    (tinfo->TIDU_size % BYTES_PER_XDR_UNIT);
221 	}
222 
223 	xprt->xp_ops = &svc_cots_op;
224 	xprt->xp_p2 = (caddr_t)cmd;
225 	cmd->cmd_xprt_started = 0;
226 	cmd->cmd_stats = rpcstat->rpc_cots_server;
227 
228 	getaddr.ic_cmd = TI_GETINFO;
229 	getaddr.ic_timout = -1;
230 	getaddr.ic_len = sizeof (*ack_p) + (2 * sizeof (sin6_t));
231 	getaddr.ic_dp = (char *)ack_p;
232 	ack_p->PRIM_type = T_ADDR_REQ;
233 
234 	err = strioctl(fp->f_vnode, I_STR, (intptr_t)&getaddr,
235 	    0, K_TO_K, CRED(), &retval);
236 	if (err) {
237 		kmem_free(cmd, sizeof (*cmd) + sizeof (*ack_p) +
238 		    (2 * sizeof (sin6_t)));
239 		kmem_free(xprt, sizeof (SVCMASTERXPRT));
240 		return (err);
241 	}
242 
243 	xprt->xp_rtaddr.maxlen = ack_p->REMADDR_length;
244 	xprt->xp_rtaddr.len = ack_p->REMADDR_length;
245 	cmd->cmd_src_addr = xprt->xp_rtaddr.buf =
246 	    (char *)ack_p + ack_p->REMADDR_offset;
247 
248 	xprt->xp_lcladdr.maxlen = ack_p->LOCADDR_length;
249 	xprt->xp_lcladdr.len = ack_p->LOCADDR_length;
250 	xprt->xp_lcladdr.buf = (char *)ack_p + ack_p->LOCADDR_offset;
251 
252 	/*
253 	 * If the current sanity check size in rpcmod is smaller
254 	 * than the size needed for this xprt, then increase
255 	 * the sanity check.
256 	 */
257 	if (max_msgsize != 0 && svc_max_msg_sizep &&
258 	    max_msgsize > *svc_max_msg_sizep) {
259 
260 		/* This check needs a lock */
261 		mutex_enter(&cots_kcreate_lock);
262 		if (svc_max_msg_sizep && max_msgsize > *svc_max_msg_sizep)
263 			*svc_max_msg_sizep = max_msgsize;
264 		mutex_exit(&cots_kcreate_lock);
265 	}
266 
267 	*nxprt = xprt;
268 
269 	return (0);
270 }
271 
272 /*
273  * Destroy a master transport record.
274  * Frees the space allocated for a transport record.
275  */
276 static void
277 svc_cots_kdestroy(SVCMASTERXPRT *xprt)
278 {
279 	struct cots_master_data *cmd = (struct cots_master_data *)xprt->xp_p2;
280 
281 	ASSERT(cmd);
282 
283 	if (xprt->xp_netid)
284 		kmem_free(xprt->xp_netid, strlen(xprt->xp_netid) + 1);
285 	if (xprt->xp_addrmask.maxlen)
286 		kmem_free(xprt->xp_addrmask.buf, xprt->xp_addrmask.maxlen);
287 
288 	mutex_destroy(&xprt->xp_req_lock);
289 	mutex_destroy(&xprt->xp_thread_lock);
290 
291 	kmem_free(cmd, sizeof (*cmd) + sizeof (struct T_addr_ack) +
292 	    (2 * sizeof (sin6_t)));
293 
294 	kmem_free(xprt, sizeof (SVCMASTERXPRT));
295 }
296 
297 /*
298  * svc_tli_kcreate() calls this function at the end to tell
299  * rpcmod that the transport is ready to receive requests.
300  */
301 static void
302 svc_cots_kstart(SVCMASTERXPRT *xprt)
303 {
304 	struct cots_master_data *cmd = (struct cots_master_data *)xprt->xp_p2;
305 
306 	if (cmd->cmd_xprt_started == 0) {
307 		/*
308 		 * Acquire the xp_req_lock in order to use xp_wq
309 		 * safely (we don't want to qenable a queue that has
310 		 * already been closed).
311 		 */
312 		mutex_enter(&xprt->xp_req_lock);
313 		if (cmd->cmd_xprt_started == 0 &&
314 		    xprt->xp_wq != NULL) {
315 			(*mir_start)(xprt->xp_wq);
316 			cmd->cmd_xprt_started = 1;
317 		}
318 		mutex_exit(&xprt->xp_req_lock);
319 	}
320 }
321 
322 /*
323  * Transport-type specific part of svc_xprt_cleanup().
324  */
325 static void
326 svc_cots_kclone_destroy(SVCXPRT *clone_xprt)
327 {
328 	cots_data_t *cd = (cots_data_t *)clone_xprt->xp_p2buf;
329 
330 	if (cd->cd_req_mp) {
331 		freemsg(cd->cd_req_mp);
332 		cd->cd_req_mp = (mblk_t *)0;
333 	}
334 	ASSERT(cd->cd_mp == NULL);
335 }
336 
337 /*
338  * Transport Attributes.
339  */
340 static void
341 svc_cots_ktattrs(SVCXPRT *clone_xprt, int attrflag, void **tattr)
342 {
343 	*tattr = NULL;
344 
345 	switch (attrflag) {
346 	case SVC_TATTR_ADDRMASK:
347 		*tattr = (void *)&clone_xprt->xp_master->xp_addrmask;
348 	}
349 }
350 
351 /*
352  * Receive rpc requests.
353  * Checks if the message is intact, and deserializes the call packet.
354  */
355 static bool_t
356 svc_cots_krecv(SVCXPRT *clone_xprt, mblk_t *mp, struct rpc_msg *msg)
357 {
358 	cots_data_t *cd = (cots_data_t *)clone_xprt->xp_p2buf;
359 	XDR *xdrs = &clone_xprt->xp_xdrin;
360 	struct rpc_cots_server *stats = CLONE2STATS(clone_xprt);
361 
362 	TRACE_0(TR_FAC_KRPC, TR_SVC_COTS_KRECV_START,
363 	    "svc_cots_krecv_start:");
364 	RPCLOG(4, "svc_cots_krecv_start clone_xprt = %p:\n",
365 	    (void *)clone_xprt);
366 
367 	RSSTAT_INCR(stats, rscalls);
368 
369 	if (mp->b_datap->db_type != M_DATA) {
370 		RPCLOG(16, "svc_cots_krecv bad db_type %d\n",
371 		    mp->b_datap->db_type);
372 		goto bad;
373 	}
374 
375 	xdrmblk_init(xdrs, mp, XDR_DECODE, 0);
376 
377 	TRACE_0(TR_FAC_KRPC, TR_XDR_CALLMSG_START,
378 	    "xdr_callmsg_start:");
379 	RPCLOG0(4, "xdr_callmsg_start:\n");
380 	if (!xdr_callmsg(xdrs, msg)) {
381 		XDR_DESTROY(xdrs);
382 		TRACE_1(TR_FAC_KRPC, TR_XDR_CALLMSG_END,
383 		    "xdr_callmsg_end:(%S)", "bad");
384 		RPCLOG0(1, "svc_cots_krecv xdr_callmsg failure\n");
385 		RSSTAT_INCR(stats, rsxdrcall);
386 		goto bad;
387 	}
388 	TRACE_1(TR_FAC_KRPC, TR_XDR_CALLMSG_END,
389 	    "xdr_callmsg_end:(%S)", "good");
390 
391 	clone_xprt->xp_xid = msg->rm_xid;
392 	cd->cd_req_mp = mp;
393 
394 	TRACE_1(TR_FAC_KRPC, TR_SVC_COTS_KRECV_END,
395 	    "svc_cots_krecv_end:(%S)", "good");
396 	RPCLOG0(4, "svc_cots_krecv_end:good\n");
397 	return (TRUE);
398 
399 bad:
400 	if (mp)
401 		freemsg(mp);
402 
403 	RSSTAT_INCR(stats, rsbadcalls);
404 	TRACE_1(TR_FAC_KRPC, TR_SVC_COTS_KRECV_END,
405 	    "svc_cots_krecv_end:(%S)", "bad");
406 	return (FALSE);
407 }
408 
409 /*
410  * Send rpc reply.
411  */
412 static bool_t
413 svc_cots_ksend(SVCXPRT *clone_xprt, struct rpc_msg *msg)
414 {
415 	/* LINTED pointer alignment */
416 	cots_data_t *cd = (cots_data_t *)clone_xprt->xp_p2buf;
417 	XDR *xdrs = &(clone_xprt->xp_xdrout);
418 	int retval = FALSE;
419 	mblk_t *mp;
420 	xdrproc_t xdr_results;
421 	caddr_t xdr_location;
422 	bool_t has_args;
423 
424 	TRACE_0(TR_FAC_KRPC, TR_SVC_COTS_KSEND_START,
425 	    "svc_cots_ksend_start:");
426 
427 	/*
428 	 * If there is a result procedure specified in the reply message,
429 	 * it will be processed in the xdr_replymsg and SVCAUTH_WRAP.
430 	 * We need to make sure it won't be processed twice, so we null
431 	 * it for xdr_replymsg here.
432 	 */
433 	has_args = FALSE;
434 	if (msg->rm_reply.rp_stat == MSG_ACCEPTED &&
435 	    msg->rm_reply.rp_acpt.ar_stat == SUCCESS) {
436 		if ((xdr_results = msg->acpted_rply.ar_results.proc) != NULL) {
437 			has_args = TRUE;
438 			xdr_location = msg->acpted_rply.ar_results.where;
439 			msg->acpted_rply.ar_results.proc = xdr_void;
440 			msg->acpted_rply.ar_results.where = NULL;
441 		}
442 	}
443 
444 	mp = cd->cd_mp;
445 	if (mp) {
446 		/*
447 		 * The program above pre-allocated an mblk and put
448 		 * the data in place.
449 		 */
450 		cd->cd_mp = (mblk_t *)NULL;
451 		if (!(xdr_replymsg_body(xdrs, msg) &&
452 		    (!has_args || SVCAUTH_WRAP(&clone_xprt->xp_auth, xdrs,
453 		    xdr_results, xdr_location)))) {
454 			XDR_DESTROY(xdrs);
455 			RPCLOG0(1, "svc_cots_ksend: "
456 			    "xdr_replymsg_body/SVCAUTH_WRAP failed\n");
457 			freemsg(mp);
458 			goto out;
459 		}
460 	} else {
461 		int	len;
462 		int	mpsize;
463 
464 		/*
465 		 * Leave space for protocol headers.
466 		 */
467 		len = MSG_OFFSET + clone_xprt->xp_msg_size;
468 
469 		/*
470 		 * Allocate an initial mblk for the response data.
471 		 */
472 		while (!(mp = allocb(len, BPRI_LO))) {
473 			RPCLOG0(16, "svc_cots_ksend: allocb failed failed\n");
474 			if (strwaitbuf(len, BPRI_LO)) {
475 				TRACE_1(TR_FAC_KRPC, TR_SVC_COTS_KSEND_END,
476 				    "svc_cots_ksend_end:(%S)", "strwaitbuf");
477 				RPCLOG0(1,
478 				    "svc_cots_ksend: strwaitbuf failed\n");
479 				goto out;
480 			}
481 		}
482 
483 		/*
484 		 * Initialize the XDR encode stream.  Additional mblks
485 		 * will be allocated if necessary.  They will be TIDU
486 		 * sized.
487 		 */
488 		xdrmblk_init(xdrs, mp, XDR_ENCODE, clone_xprt->xp_msg_size);
489 		mpsize = MBLKSIZE(mp);
490 		ASSERT(mpsize >= len);
491 		ASSERT(mp->b_rptr == mp->b_datap->db_base);
492 
493 		/*
494 		 * If the size of mblk is not appreciably larger than what we
495 		 * asked, then resize the mblk to exactly len bytes. Reason for
496 		 * this: suppose len is 1600 bytes, the tidu is 1460 bytes
497 		 * (from TCP over ethernet), and the arguments to RPC require
498 		 * 2800 bytes. Ideally we want the protocol to render two
499 		 * ~1400 byte segments over the wire. If allocb() gives us a 2k
500 		 * mblk, and we allocate a second mblk for the rest, the
501 		 * protocol module may generate 3 segments over the wire:
502 		 * 1460 bytes for the first, 448 (2048 - 1600) for the 2nd, and
503 		 * 892 for the 3rd. If we "waste" 448 bytes in the first mblk,
504 		 * the XDR encoding will generate two ~1400 byte mblks, and the
505 		 * protocol module is more likely to produce properly sized
506 		 * segments.
507 		 */
508 		if ((mpsize >> 1) <= len) {
509 			mp->b_rptr += (mpsize - len);
510 		}
511 
512 		/*
513 		 * Adjust b_rptr to reserve space for the non-data protocol
514 		 * headers that any downstream modules might like to add, and
515 		 * for the record marking header.
516 		 */
517 		mp->b_rptr += (MSG_OFFSET + RM_HDR_SIZE);
518 
519 		XDR_SETPOS(xdrs, (uint_t)(mp->b_rptr - mp->b_datap->db_base));
520 		ASSERT(mp->b_wptr == mp->b_rptr);
521 
522 		msg->rm_xid = clone_xprt->xp_xid;
523 
524 		TRACE_0(TR_FAC_KRPC, TR_XDR_REPLYMSG_START,
525 		    "xdr_replymsg_start:");
526 		if (!(xdr_replymsg(xdrs, msg) &&
527 		    (!has_args || SVCAUTH_WRAP(&clone_xprt->xp_auth, xdrs,
528 		    xdr_results, xdr_location)))) {
529 			XDR_DESTROY(xdrs);
530 			TRACE_1(TR_FAC_KRPC, TR_XDR_REPLYMSG_END,
531 			    "xdr_replymsg_end:(%S)", "bad");
532 			freemsg(mp);
533 			RPCLOG0(1, "svc_cots_ksend: xdr_replymsg/SVCAUTH_WRAP "
534 			    "failed\n");
535 			goto out;
536 		}
537 		TRACE_1(TR_FAC_KRPC, TR_XDR_REPLYMSG_END,
538 		    "xdr_replymsg_end:(%S)", "good");
539 	}
540 
541 	XDR_DESTROY(xdrs);
542 
543 	put(clone_xprt->xp_wq, mp);
544 	retval = TRUE;
545 
546 out:
547 	/*
548 	 * This is completely disgusting.  If public is set it is
549 	 * a pointer to a structure whose first field is the address
550 	 * of the function to free that structure and any related
551 	 * stuff.  (see rrokfree in nfs_xdr.c).
552 	 */
553 	if (xdrs->x_public) {
554 		/* LINTED pointer alignment */
555 		(**((int (**)())xdrs->x_public))(xdrs->x_public);
556 	}
557 
558 	TRACE_1(TR_FAC_KRPC, TR_SVC_COTS_KSEND_END,
559 	    "svc_cots_ksend_end:(%S)", "done");
560 	return (retval);
561 }
562 
563 /*
564  * Deserialize arguments.
565  */
566 static bool_t
567 svc_cots_kgetargs(SVCXPRT *clone_xprt, xdrproc_t xdr_args,
568     caddr_t args_ptr)
569 {
570 	return (SVCAUTH_UNWRAP(&clone_xprt->xp_auth, &clone_xprt->xp_xdrin,
571 	    xdr_args, args_ptr));
572 }
573 
574 static bool_t
575 svc_cots_kfreeargs(SVCXPRT *clone_xprt, xdrproc_t xdr_args,
576     caddr_t args_ptr)
577 {
578 	cots_data_t *cd = (cots_data_t *)clone_xprt->xp_p2buf;
579 	/* LINTED pointer alignment */
580 	XDR *xdrs = &clone_xprt->xp_xdrin;
581 	mblk_t *mp;
582 	bool_t retval;
583 
584 	/*
585 	 * It is important to call the XDR routine before
586 	 * freeing the request mblk.  Structures in the
587 	 * XDR data may point into the mblk and require that
588 	 * the memory be intact during the free routine.
589 	 */
590 	if (args_ptr) {
591 		xdrs->x_op = XDR_FREE;
592 		retval = (*xdr_args)(xdrs, args_ptr);
593 	} else
594 		retval = TRUE;
595 
596 	XDR_DESTROY(xdrs);
597 
598 	if ((mp = cd->cd_req_mp) != NULL) {
599 		cd->cd_req_mp = (mblk_t *)0;
600 		freemsg(mp);
601 	}
602 
603 	return (retval);
604 }
605 
606 static int32_t *
607 svc_cots_kgetres(SVCXPRT *clone_xprt, int size)
608 {
609 	/* LINTED pointer alignment */
610 	cots_data_t *cd = (cots_data_t *)clone_xprt->xp_p2buf;
611 	XDR *xdrs = &clone_xprt->xp_xdrout;
612 	mblk_t *mp;
613 	int32_t *buf;
614 	struct rpc_msg rply;
615 	int len;
616 	int mpsize;
617 
618 	/*
619 	 * Leave space for protocol headers.
620 	 */
621 	len = MSG_OFFSET + clone_xprt->xp_msg_size;
622 
623 	/*
624 	 * Allocate an initial mblk for the response data.
625 	 */
626 	while ((mp = allocb(len, BPRI_LO)) == NULL) {
627 		if (strwaitbuf(len, BPRI_LO))
628 			return (NULL);
629 	}
630 
631 	/*
632 	 * Initialize the XDR encode stream.  Additional mblks
633 	 * will be allocated if necessary.  They will be TIDU
634 	 * sized.
635 	 */
636 	xdrmblk_init(xdrs, mp, XDR_ENCODE, clone_xprt->xp_msg_size);
637 	mpsize = MBLKSIZE(mp);
638 	ASSERT(mpsize >= len);
639 	ASSERT(mp->b_rptr == mp->b_datap->db_base);
640 
641 	/*
642 	 * If the size of mblk is not appreciably larger than what we
643 	 * asked, then resize the mblk to exactly len bytes. Reason for
644 	 * this: suppose len is 1600 bytes, the tidu is 1460 bytes
645 	 * (from TCP over ethernet), and the arguments to RPC require
646 	 * 2800 bytes. Ideally we want the protocol to render two
647 	 * ~1400 byte segments over the wire. If allocb() gives us a 2k
648 	 * mblk, and we allocate a second mblk for the rest, the
649 	 * protocol module may generate 3 segments over the wire:
650 	 * 1460 bytes for the first, 448 (2048 - 1600) for the 2nd, and
651 	 * 892 for the 3rd. If we "waste" 448 bytes in the first mblk,
652 	 * the XDR encoding will generate two ~1400 byte mblks, and the
653 	 * protocol module is more likely to produce properly sized
654 	 * segments.
655 	 */
656 	if ((mpsize >> 1) <= len) {
657 		mp->b_rptr += (mpsize - len);
658 	}
659 
660 	/*
661 	 * Adjust b_rptr to reserve space for the non-data protocol
662 	 * headers that any downstream modules might like to add, and
663 	 * for the record marking header.
664 	 */
665 	mp->b_rptr += (MSG_OFFSET + RM_HDR_SIZE);
666 
667 	XDR_SETPOS(xdrs, (uint_t)(mp->b_rptr - mp->b_datap->db_base));
668 	ASSERT(mp->b_wptr == mp->b_rptr);
669 
670 	/*
671 	 * Assume a successful RPC since most of them are.
672 	 */
673 	rply.rm_xid = clone_xprt->xp_xid;
674 	rply.rm_direction = REPLY;
675 	rply.rm_reply.rp_stat = MSG_ACCEPTED;
676 	rply.acpted_rply.ar_verf = clone_xprt->xp_verf;
677 	rply.acpted_rply.ar_stat = SUCCESS;
678 
679 	if (!xdr_replymsg_hdr(xdrs, &rply)) {
680 		XDR_DESTROY(xdrs);
681 		freeb(mp);
682 		return (NULL);
683 	}
684 
685 	buf = XDR_INLINE(xdrs, size);
686 	if (buf == NULL) {
687 		XDR_DESTROY(xdrs);
688 		ASSERT(cd->cd_mp == NULL);
689 		freemsg(mp);
690 	} else {
691 		cd->cd_mp = mp;
692 	}
693 	return (buf);
694 }
695 
696 static void
697 svc_cots_kfreeres(SVCXPRT *clone_xprt)
698 {
699 	cots_data_t *cd;
700 	mblk_t *mp;
701 
702 	cd = (cots_data_t *)clone_xprt->xp_p2buf;
703 	if ((mp = cd->cd_mp) != NULL) {
704 		XDR_DESTROY(&clone_xprt->xp_xdrout);
705 		cd->cd_mp = (mblk_t *)NULL;
706 		freemsg(mp);
707 	}
708 }
709 
710 /*
711  * the dup cacheing routines below provide a cache of non-failure
712  * transaction id's.  rpc service routines can use this to detect
713  * retransmissions and re-send a non-failure response.
714  */
715 
716 /*
717  * MAXDUPREQS is the number of cached items.  It should be adjusted
718  * to the service load so that there is likely to be a response entry
719  * when the first retransmission comes in.
720  */
721 #define	MAXDUPREQS	8192
722 
723 /*
724  * This should be appropriately scaled to MAXDUPREQS.  To produce as less as
725  * possible collisions it is suggested to set this to a prime.
726  */
727 #define	DRHASHSZ	2053
728 
729 #define	XIDHASH(xid)	((xid) % DRHASHSZ)
730 #define	DRHASH(dr)	XIDHASH((dr)->dr_xid)
731 #define	REQTOXID(req)	((req)->rq_xprt->xp_xid)
732 
733 static int	cotsndupreqs = 0;
734 int	cotsmaxdupreqs = MAXDUPREQS;
735 static kmutex_t cotsdupreq_lock;
736 static struct dupreq *cotsdrhashtbl[DRHASHSZ];
737 static int	cotsdrhashstat[DRHASHSZ];
738 
739 static void unhash(struct dupreq *);
740 
741 /*
742  * cotsdrmru points to the head of a circular linked list in lru order.
743  * cotsdrmru->dr_next == drlru
744  */
745 struct dupreq *cotsdrmru;
746 
747 /*
748  * PSARC 2003/523 Contract Private Interface
749  * svc_cots_kdup
750  * Changes must be reviewed by Solaris File Sharing
751  * Changes must be communicated to contract-2003-523@sun.com
752  *
753  * svc_cots_kdup searches the request cache and returns 0 if the
754  * request is not found in the cache.  If it is found, then it
755  * returns the state of the request (in progress or done) and
756  * the status or attributes that were part of the original reply.
757  *
758  * If DUP_DONE (there is a duplicate) svc_cots_kdup copies over the
759  * value of the response. In that case, also return in *dupcachedp
760  * whether the response free routine is cached in the dupreq - in which case
761  * the caller should not be freeing it, because it will be done later
762  * in the svc_cots_kdup code when the dupreq is reused.
763  */
764 static int
765 svc_cots_kdup(struct svc_req *req, caddr_t res, int size, struct dupreq **drpp,
766     bool_t *dupcachedp)
767 {
768 	struct rpc_cots_server *stats = CLONE2STATS(req->rq_xprt);
769 	struct dupreq *dr;
770 	uint32_t xid;
771 	uint32_t drhash;
772 	int status;
773 
774 	xid = REQTOXID(req);
775 	mutex_enter(&cotsdupreq_lock);
776 	RSSTAT_INCR(stats, rsdupchecks);
777 	/*
778 	 * Check to see whether an entry already exists in the cache.
779 	 */
780 	dr = cotsdrhashtbl[XIDHASH(xid)];
781 	while (dr != NULL) {
782 		if (dr->dr_xid == xid &&
783 		    dr->dr_proc == req->rq_proc &&
784 		    dr->dr_prog == req->rq_prog &&
785 		    dr->dr_vers == req->rq_vers &&
786 		    dr->dr_addr.len == req->rq_xprt->xp_rtaddr.len &&
787 		    bcmp((caddr_t)dr->dr_addr.buf,
788 		    (caddr_t)req->rq_xprt->xp_rtaddr.buf,
789 		    dr->dr_addr.len) == 0) {
790 			status = dr->dr_status;
791 			if (status == DUP_DONE) {
792 				bcopy(dr->dr_resp.buf, res, size);
793 				if (dupcachedp != NULL)
794 					*dupcachedp = (dr->dr_resfree != NULL);
795 				TRACE_0(TR_FAC_KRPC, TR_SVC_COTS_KDUP_DONE,
796 				    "svc_cots_kdup: DUP_DONE");
797 			} else {
798 				dr->dr_status = DUP_INPROGRESS;
799 				*drpp = dr;
800 				TRACE_0(TR_FAC_KRPC,
801 				    TR_SVC_COTS_KDUP_INPROGRESS,
802 				    "svc_cots_kdup: DUP_INPROGRESS");
803 			}
804 			RSSTAT_INCR(stats, rsdupreqs);
805 			mutex_exit(&cotsdupreq_lock);
806 			return (status);
807 		}
808 		dr = dr->dr_chain;
809 	}
810 
811 	/*
812 	 * There wasn't an entry, either allocate a new one or recycle
813 	 * an old one.
814 	 */
815 	if (cotsndupreqs < cotsmaxdupreqs) {
816 		dr = kmem_alloc(sizeof (*dr), KM_NOSLEEP);
817 		if (dr == NULL) {
818 			mutex_exit(&cotsdupreq_lock);
819 			return (DUP_ERROR);
820 		}
821 		dr->dr_resp.buf = NULL;
822 		dr->dr_resp.maxlen = 0;
823 		dr->dr_addr.buf = NULL;
824 		dr->dr_addr.maxlen = 0;
825 		if (cotsdrmru) {
826 			dr->dr_next = cotsdrmru->dr_next;
827 			cotsdrmru->dr_next = dr;
828 		} else {
829 			dr->dr_next = dr;
830 		}
831 		cotsndupreqs++;
832 	} else {
833 		dr = cotsdrmru->dr_next;
834 		while (dr->dr_status == DUP_INPROGRESS) {
835 			dr = dr->dr_next;
836 			if (dr == cotsdrmru->dr_next) {
837 				cmn_err(CE_WARN, "svc_cots_kdup no slots free");
838 				mutex_exit(&cotsdupreq_lock);
839 				return (DUP_ERROR);
840 			}
841 		}
842 		unhash(dr);
843 		if (dr->dr_resfree) {
844 			(*dr->dr_resfree)(dr->dr_resp.buf);
845 		}
846 	}
847 	dr->dr_resfree = NULL;
848 	cotsdrmru = dr;
849 
850 	dr->dr_xid = REQTOXID(req);
851 	dr->dr_prog = req->rq_prog;
852 	dr->dr_vers = req->rq_vers;
853 	dr->dr_proc = req->rq_proc;
854 	if (dr->dr_addr.maxlen < req->rq_xprt->xp_rtaddr.len) {
855 		if (dr->dr_addr.buf != NULL)
856 			kmem_free(dr->dr_addr.buf, dr->dr_addr.maxlen);
857 		dr->dr_addr.maxlen = req->rq_xprt->xp_rtaddr.len;
858 		dr->dr_addr.buf = kmem_alloc(dr->dr_addr.maxlen, KM_NOSLEEP);
859 		if (dr->dr_addr.buf == NULL) {
860 			dr->dr_addr.maxlen = 0;
861 			dr->dr_status = DUP_DROP;
862 			mutex_exit(&cotsdupreq_lock);
863 			return (DUP_ERROR);
864 		}
865 	}
866 	dr->dr_addr.len = req->rq_xprt->xp_rtaddr.len;
867 	bcopy(req->rq_xprt->xp_rtaddr.buf, dr->dr_addr.buf, dr->dr_addr.len);
868 	if (dr->dr_resp.maxlen < size) {
869 		if (dr->dr_resp.buf != NULL)
870 			kmem_free(dr->dr_resp.buf, dr->dr_resp.maxlen);
871 		dr->dr_resp.maxlen = (unsigned int)size;
872 		dr->dr_resp.buf = kmem_alloc(size, KM_NOSLEEP);
873 		if (dr->dr_resp.buf == NULL) {
874 			dr->dr_resp.maxlen = 0;
875 			dr->dr_status = DUP_DROP;
876 			mutex_exit(&cotsdupreq_lock);
877 			return (DUP_ERROR);
878 		}
879 	}
880 	dr->dr_status = DUP_INPROGRESS;
881 
882 	drhash = (uint32_t)DRHASH(dr);
883 	dr->dr_chain = cotsdrhashtbl[drhash];
884 	cotsdrhashtbl[drhash] = dr;
885 	cotsdrhashstat[drhash]++;
886 	mutex_exit(&cotsdupreq_lock);
887 	*drpp = dr;
888 	return (DUP_NEW);
889 }
890 
891 /*
892  * PSARC 2003/523 Contract Private Interface
893  * svc_cots_kdupdone
894  * Changes must be reviewed by Solaris File Sharing
895  * Changes must be communicated to contract-2003-523@sun.com
896  *
897  * svc_cots_kdupdone marks the request done (DUP_DONE or DUP_DROP)
898  * and stores the response.
899  */
900 static void
901 svc_cots_kdupdone(struct dupreq *dr, caddr_t res, void (*dis_resfree)(),
902     int size, int status)
903 {
904 	ASSERT(dr->dr_resfree == NULL);
905 	if (status == DUP_DONE) {
906 		bcopy(res, dr->dr_resp.buf, size);
907 		dr->dr_resfree = dis_resfree;
908 	}
909 	dr->dr_status = status;
910 }
911 
912 /*
913  * This routine expects that the mutex, cotsdupreq_lock, is already held.
914  */
915 static void
916 unhash(struct dupreq *dr)
917 {
918 	struct dupreq *drt;
919 	struct dupreq *drtprev = NULL;
920 	uint32_t drhash;
921 
922 	ASSERT(MUTEX_HELD(&cotsdupreq_lock));
923 
924 	drhash = (uint32_t)DRHASH(dr);
925 	drt = cotsdrhashtbl[drhash];
926 	while (drt != NULL) {
927 		if (drt == dr) {
928 			cotsdrhashstat[drhash]--;
929 			if (drtprev == NULL) {
930 				cotsdrhashtbl[drhash] = drt->dr_chain;
931 			} else {
932 				drtprev->dr_chain = drt->dr_chain;
933 			}
934 			return;
935 		}
936 		drtprev = drt;
937 		drt = drt->dr_chain;
938 	}
939 }
940 
941 void
942 svc_cots_stats_init(zoneid_t zoneid, struct rpc_cots_server **statsp)
943 {
944 	*statsp = (struct rpc_cots_server *)rpcstat_zone_init_common(zoneid,
945 	    "unix", "rpc_cots_server", (const kstat_named_t *)&cots_rsstat_tmpl,
946 	    sizeof (cots_rsstat_tmpl));
947 }
948 
949 void
950 svc_cots_stats_fini(zoneid_t zoneid, struct rpc_cots_server **statsp)
951 {
952 	rpcstat_zone_fini_common(zoneid, "unix", "rpc_cots_server");
953 	kmem_free(*statsp, sizeof (cots_rsstat_tmpl));
954 }
955 
956 void
957 svc_cots_init(void)
958 {
959 	/*
960 	 * Check to make sure that the cots private data will fit into
961 	 * the stack buffer allocated by svc_run.  The ASSERT is a safety
962 	 * net if the cots_data_t structure ever changes.
963 	 */
964 	/*CONSTANTCONDITION*/
965 	ASSERT(sizeof (cots_data_t) <= SVC_P2LEN);
966 
967 	mutex_init(&cots_kcreate_lock, NULL, MUTEX_DEFAULT, NULL);
968 	mutex_init(&cotsdupreq_lock, NULL, MUTEX_DEFAULT, NULL);
969 }
970