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) 1989, 2010, Oracle and/or its affiliates. All rights reserved.
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 /*
36 * svc_clts.c
37 * Server side for RPC in the kernel.
38 *
39 */
40
41 #include <sys/param.h>
42 #include <sys/types.h>
43 #include <sys/sysmacros.h>
44 #include <sys/file.h>
45 #include <sys/stream.h>
46 #include <sys/strsun.h>
47 #include <sys/strsubr.h>
48 #include <sys/tihdr.h>
49 #include <sys/tiuser.h>
50 #include <sys/t_kuser.h>
51 #include <sys/fcntl.h>
52 #include <sys/errno.h>
53 #include <sys/kmem.h>
54 #include <sys/systm.h>
55 #include <sys/cmn_err.h>
56 #include <sys/kstat.h>
57 #include <sys/vtrace.h>
58 #include <sys/debug.h>
59
60 #include <rpc/types.h>
61 #include <rpc/xdr.h>
62 #include <rpc/auth.h>
63 #include <rpc/clnt.h>
64 #include <rpc/rpc_msg.h>
65 #include <rpc/svc.h>
66 #include <inet/ip.h>
67
68 /*
69 * Routines exported through ops vector.
70 */
71 static bool_t svc_clts_krecv(SVCXPRT *, mblk_t *, struct rpc_msg *);
72 static bool_t svc_clts_ksend(SVCXPRT *, struct rpc_msg *);
73 static bool_t svc_clts_kgetargs(SVCXPRT *, xdrproc_t, caddr_t);
74 static bool_t svc_clts_kfreeargs(SVCXPRT *, xdrproc_t, caddr_t);
75 static void svc_clts_kdestroy(SVCMASTERXPRT *);
76 static int svc_clts_kdup(struct svc_req *, caddr_t, int,
77 struct dupreq **, bool_t *);
78 static void svc_clts_kdupdone(struct dupreq *, caddr_t,
79 void (*)(), int, int);
80 static int32_t *svc_clts_kgetres(SVCXPRT *, int);
81 static void svc_clts_kclone_destroy(SVCXPRT *);
82 static void svc_clts_kfreeres(SVCXPRT *);
83 static void svc_clts_kstart(SVCMASTERXPRT *);
84 static void svc_clts_kclone_xprt(SVCXPRT *, SVCXPRT *);
85 static void svc_clts_ktattrs(SVCXPRT *, int, void **);
86
87 /*
88 * Server transport operations vector.
89 */
90 struct svc_ops svc_clts_op = {
91 svc_clts_krecv, /* Get requests */
92 svc_clts_kgetargs, /* Deserialize arguments */
93 svc_clts_ksend, /* Send reply */
94 svc_clts_kfreeargs, /* Free argument data space */
95 svc_clts_kdestroy, /* Destroy transport handle */
96 svc_clts_kdup, /* Check entry in dup req cache */
97 svc_clts_kdupdone, /* Mark entry in dup req cache as done */
98 svc_clts_kgetres, /* Get pointer to response buffer */
99 svc_clts_kfreeres, /* Destroy pre-serialized response header */
100 svc_clts_kclone_destroy, /* Destroy a clone xprt */
101 svc_clts_kstart, /* Tell `ready-to-receive' to rpcmod */
102 svc_clts_kclone_xprt, /* transport specific clone xprt function */
103 svc_clts_ktattrs /* Transport specific attributes. */
104 };
105
106 /*
107 * Transport private data.
108 * Kept in xprt->xp_p2buf.
109 */
110 struct udp_data {
111 mblk_t *ud_resp; /* buffer for response */
112 mblk_t *ud_inmp; /* mblk chain of request */
113 };
114
115 #define UD_MAXSIZE 8800
116 #define UD_INITSIZE 2048
117
118 /*
119 * Connectionless server statistics
120 */
121 static const struct rpc_clts_server {
122 kstat_named_t rscalls;
123 kstat_named_t rsbadcalls;
124 kstat_named_t rsnullrecv;
125 kstat_named_t rsbadlen;
126 kstat_named_t rsxdrcall;
127 kstat_named_t rsdupchecks;
128 kstat_named_t rsdupreqs;
129 } clts_rsstat_tmpl = {
130 { "calls", KSTAT_DATA_UINT64 },
131 { "badcalls", KSTAT_DATA_UINT64 },
132 { "nullrecv", KSTAT_DATA_UINT64 },
133 { "badlen", KSTAT_DATA_UINT64 },
134 { "xdrcall", KSTAT_DATA_UINT64 },
135 { "dupchecks", KSTAT_DATA_UINT64 },
136 { "dupreqs", KSTAT_DATA_UINT64 }
137 };
138
139 static uint_t clts_rsstat_ndata =
140 sizeof (clts_rsstat_tmpl) / sizeof (kstat_named_t);
141
142 #define CLONE2STATS(clone_xprt) \
143 (struct rpc_clts_server *)(clone_xprt)->xp_master->xp_p2
144
145 #define RSSTAT_INCR(stats, x) \
146 atomic_inc_64(&(stats)->x.value.ui64)
147
148 /*
149 * Create a transport record.
150 * The transport record, output buffer, and private data structure
151 * are allocated. The output buffer is serialized into using xdrmem.
152 * There is one transport record per user process which implements a
153 * set of services.
154 */
155 /* ARGSUSED */
156 int
svc_clts_kcreate(file_t * fp,uint_t sendsz,struct T_info_ack * tinfo,SVCMASTERXPRT ** nxprt)157 svc_clts_kcreate(file_t *fp, uint_t sendsz, struct T_info_ack *tinfo,
158 SVCMASTERXPRT **nxprt)
159 {
160 SVCMASTERXPRT *xprt;
161 struct rpcstat *rpcstat;
162
163 if (nxprt == NULL)
164 return (EINVAL);
165
166 rpcstat = zone_getspecific(rpcstat_zone_key, curproc->p_zone);
167 ASSERT(rpcstat != NULL);
168
169 xprt = kmem_zalloc(sizeof (*xprt), KM_SLEEP);
170 xprt->xp_lcladdr.buf = kmem_zalloc(sizeof (sin6_t), KM_SLEEP);
171 xprt->xp_p2 = (caddr_t)rpcstat->rpc_clts_server;
172 xprt->xp_ops = &svc_clts_op;
173 xprt->xp_msg_size = tinfo->TSDU_size;
174
175 xprt->xp_rtaddr.buf = NULL;
176 xprt->xp_rtaddr.maxlen = tinfo->ADDR_size;
177 xprt->xp_rtaddr.len = 0;
178
179 *nxprt = xprt;
180
181 return (0);
182 }
183
184 /*
185 * Destroy a transport record.
186 * Frees the space allocated for a transport record.
187 */
188 static void
svc_clts_kdestroy(SVCMASTERXPRT * xprt)189 svc_clts_kdestroy(SVCMASTERXPRT *xprt)
190 {
191 if (xprt->xp_netid)
192 kmem_free(xprt->xp_netid, strlen(xprt->xp_netid) + 1);
193 if (xprt->xp_addrmask.maxlen)
194 kmem_free(xprt->xp_addrmask.buf, xprt->xp_addrmask.maxlen);
195
196 mutex_destroy(&xprt->xp_req_lock);
197 mutex_destroy(&xprt->xp_thread_lock);
198
199 kmem_free(xprt->xp_lcladdr.buf, sizeof (sin6_t));
200 kmem_free(xprt, sizeof (SVCMASTERXPRT));
201 }
202
203 /*
204 * Transport-type specific part of svc_xprt_cleanup().
205 * Frees the message buffer space allocated for a clone of a transport record
206 */
207 static void
svc_clts_kclone_destroy(SVCXPRT * clone_xprt)208 svc_clts_kclone_destroy(SVCXPRT *clone_xprt)
209 {
210 /* LINTED pointer alignment */
211 struct udp_data *ud = (struct udp_data *)clone_xprt->xp_p2buf;
212
213 if (ud->ud_resp) {
214 /*
215 * There should not be any left over results buffer.
216 */
217 ASSERT(ud->ud_resp->b_cont == NULL);
218
219 /*
220 * Free the T_UNITDATA_{REQ/IND} that svc_clts_krecv
221 * saved.
222 */
223 freeb(ud->ud_resp);
224 }
225 if (ud->ud_inmp)
226 freemsg(ud->ud_inmp);
227 }
228
229 /*
230 * svc_tli_kcreate() calls this function at the end to tell
231 * rpcmod that the transport is ready to receive requests.
232 */
233 /* ARGSUSED */
234 static void
svc_clts_kstart(SVCMASTERXPRT * xprt)235 svc_clts_kstart(SVCMASTERXPRT *xprt)
236 {
237 }
238
239 static void
svc_clts_kclone_xprt(SVCXPRT * src_xprt,SVCXPRT * dst_xprt)240 svc_clts_kclone_xprt(SVCXPRT *src_xprt, SVCXPRT *dst_xprt)
241 {
242 struct udp_data *ud_src =
243 (struct udp_data *)src_xprt->xp_p2buf;
244 struct udp_data *ud_dst =
245 (struct udp_data *)dst_xprt->xp_p2buf;
246
247 if (ud_src->ud_resp)
248 ud_dst->ud_resp = dupb(ud_src->ud_resp);
249
250 }
251
252 static void
svc_clts_ktattrs(SVCXPRT * clone_xprt,int attrflag,void ** tattr)253 svc_clts_ktattrs(SVCXPRT *clone_xprt, int attrflag, void **tattr)
254 {
255 *tattr = NULL;
256
257 switch (attrflag) {
258 case SVC_TATTR_ADDRMASK:
259 *tattr = (void *)&clone_xprt->xp_master->xp_addrmask;
260 }
261 }
262
263 /*
264 * Receive rpc requests.
265 * Pulls a request in off the socket, checks if the packet is intact,
266 * and deserializes the call packet.
267 */
268 static bool_t
svc_clts_krecv(SVCXPRT * clone_xprt,mblk_t * mp,struct rpc_msg * msg)269 svc_clts_krecv(SVCXPRT *clone_xprt, mblk_t *mp, struct rpc_msg *msg)
270 {
271 /* LINTED pointer alignment */
272 struct udp_data *ud = (struct udp_data *)clone_xprt->xp_p2buf;
273 XDR *xdrs = &clone_xprt->xp_xdrin;
274 struct rpc_clts_server *stats = CLONE2STATS(clone_xprt);
275 union T_primitives *pptr;
276 int hdrsz;
277 cred_t *cr;
278
279 TRACE_0(TR_FAC_KRPC, TR_SVC_CLTS_KRECV_START,
280 "svc_clts_krecv_start:");
281
282 RSSTAT_INCR(stats, rscalls);
283
284 /*
285 * The incoming request should start with an M_PROTO message.
286 */
287 if (mp->b_datap->db_type != M_PROTO) {
288 goto bad;
289 }
290
291 /*
292 * The incoming request should be an T_UNITDTA_IND. There
293 * might be other messages coming up the stream, but we can
294 * ignore them.
295 */
296 pptr = (union T_primitives *)mp->b_rptr;
297 if (pptr->type != T_UNITDATA_IND) {
298 goto bad;
299 }
300 /*
301 * Do some checking to make sure that the header at least looks okay.
302 */
303 hdrsz = (int)(mp->b_wptr - mp->b_rptr);
304 if (hdrsz < TUNITDATAINDSZ ||
305 hdrsz < (pptr->unitdata_ind.OPT_offset +
306 pptr->unitdata_ind.OPT_length) ||
307 hdrsz < (pptr->unitdata_ind.SRC_offset +
308 pptr->unitdata_ind.SRC_length)) {
309 goto bad;
310 }
311
312 /*
313 * Make sure that the transport provided a usable address.
314 */
315 if (pptr->unitdata_ind.SRC_length <= 0) {
316 goto bad;
317 }
318 /*
319 * Point the remote transport address in the service_transport
320 * handle at the address in the request.
321 */
322 clone_xprt->xp_rtaddr.buf = (char *)mp->b_rptr +
323 pptr->unitdata_ind.SRC_offset;
324 clone_xprt->xp_rtaddr.len = pptr->unitdata_ind.SRC_length;
325
326 /*
327 * Copy the local transport address in the service_transport
328 * handle at the address in the request. We will have only
329 * the local IP address in options.
330 */
331 ((sin_t *)(clone_xprt->xp_lcladdr.buf))->sin_family = AF_UNSPEC;
332 if (pptr->unitdata_ind.OPT_length && pptr->unitdata_ind.OPT_offset) {
333 char *dstopt = (char *)mp->b_rptr +
334 pptr->unitdata_ind.OPT_offset;
335 struct T_opthdr *toh = (struct T_opthdr *)dstopt;
336
337 if (toh->level == IPPROTO_IPV6 && toh->status == 0 &&
338 toh->name == IPV6_PKTINFO) {
339 struct in6_pktinfo *pkti;
340
341 dstopt += sizeof (struct T_opthdr);
342 pkti = (struct in6_pktinfo *)dstopt;
343 ((sin6_t *)(clone_xprt->xp_lcladdr.buf))->sin6_addr
344 = pkti->ipi6_addr;
345 ((sin6_t *)(clone_xprt->xp_lcladdr.buf))->sin6_family
346 = AF_INET6;
347 } else if (toh->level == IPPROTO_IP && toh->status == 0 &&
348 toh->name == IP_RECVDSTADDR) {
349 dstopt += sizeof (struct T_opthdr);
350 ((sin_t *)(clone_xprt->xp_lcladdr.buf))->sin_addr
351 = *(struct in_addr *)dstopt;
352 ((sin_t *)(clone_xprt->xp_lcladdr.buf))->sin_family
353 = AF_INET;
354 }
355 }
356
357 /*
358 * Save the first mblk which contains the T_unidata_ind in
359 * ud_resp. It will be used to generate the T_unitdata_req
360 * during the reply.
361 * We reuse any options in the T_unitdata_ind for the T_unitdata_req
362 * since we must pass any SCM_UCRED across in order for TX to
363 * work. We also make sure any cred_t is carried across.
364 */
365 if (ud->ud_resp) {
366 if (ud->ud_resp->b_cont != NULL) {
367 cmn_err(CE_WARN, "svc_clts_krecv: ud_resp %p, "
368 "b_cont %p", (void *)ud->ud_resp,
369 (void *)ud->ud_resp->b_cont);
370 }
371 freeb(ud->ud_resp);
372 }
373 /* Move any cred_t to the first mblk in the message */
374 cr = msg_getcred(mp, NULL);
375 if (cr != NULL)
376 mblk_setcred(mp, cr, NOPID);
377
378 ud->ud_resp = mp;
379 mp = mp->b_cont;
380 ud->ud_resp->b_cont = NULL;
381
382 xdrmblk_init(xdrs, mp, XDR_DECODE, 0);
383
384 TRACE_0(TR_FAC_KRPC, TR_XDR_CALLMSG_START,
385 "xdr_callmsg_start:");
386 if (! xdr_callmsg(xdrs, msg)) {
387 XDR_DESTROY(xdrs);
388 TRACE_1(TR_FAC_KRPC, TR_XDR_CALLMSG_END,
389 "xdr_callmsg_end:(%S)", "bad");
390 RSSTAT_INCR(stats, rsxdrcall);
391 goto bad;
392 }
393 TRACE_1(TR_FAC_KRPC, TR_XDR_CALLMSG_END,
394 "xdr_callmsg_end:(%S)", "good");
395
396 clone_xprt->xp_xid = msg->rm_xid;
397 ud->ud_inmp = mp;
398
399 TRACE_1(TR_FAC_KRPC, TR_SVC_CLTS_KRECV_END,
400 "svc_clts_krecv_end:(%S)", "good");
401 return (TRUE);
402
403 bad:
404 freemsg(mp);
405 if (ud->ud_resp) {
406 /*
407 * There should not be any left over results buffer.
408 */
409 ASSERT(ud->ud_resp->b_cont == NULL);
410 freeb(ud->ud_resp);
411 ud->ud_resp = NULL;
412 }
413
414 RSSTAT_INCR(stats, rsbadcalls);
415 TRACE_1(TR_FAC_KRPC, TR_SVC_CLTS_KRECV_END,
416 "svc_clts_krecv_end:(%S)", "bad");
417 return (FALSE);
418 }
419
420 /*
421 * Send rpc reply.
422 * Serialize the reply packet into the output buffer then
423 * call t_ksndudata to send it.
424 */
425 static bool_t
svc_clts_ksend(SVCXPRT * clone_xprt,struct rpc_msg * msg)426 svc_clts_ksend(SVCXPRT *clone_xprt, struct rpc_msg *msg)
427 {
428 /* LINTED pointer alignment */
429 struct udp_data *ud = (struct udp_data *)clone_xprt->xp_p2buf;
430 XDR *xdrs = &clone_xprt->xp_xdrout;
431 int stat = FALSE;
432 mblk_t *mp;
433 int msgsz;
434 struct T_unitdata_req *udreq;
435 xdrproc_t xdr_results;
436 caddr_t xdr_location;
437 bool_t has_args;
438
439 TRACE_0(TR_FAC_KRPC, TR_SVC_CLTS_KSEND_START,
440 "svc_clts_ksend_start:");
441
442 ASSERT(ud->ud_resp != NULL);
443
444 /*
445 * If there is a result procedure specified in the reply message,
446 * it will be processed in the xdr_replymsg and SVCAUTH_WRAP.
447 * We need to make sure it won't be processed twice, so we null
448 * it for xdr_replymsg here.
449 */
450 has_args = FALSE;
451 if (msg->rm_reply.rp_stat == MSG_ACCEPTED &&
452 msg->rm_reply.rp_acpt.ar_stat == SUCCESS) {
453 if ((xdr_results = msg->acpted_rply.ar_results.proc) != NULL) {
454 has_args = TRUE;
455 xdr_location = msg->acpted_rply.ar_results.where;
456 msg->acpted_rply.ar_results.proc = xdr_void;
457 msg->acpted_rply.ar_results.where = NULL;
458 }
459 }
460
461 if (ud->ud_resp->b_cont == NULL) {
462 /*
463 * Allocate an initial mblk for the response data.
464 */
465 while ((mp = allocb(UD_INITSIZE, BPRI_LO)) == NULL) {
466 if (strwaitbuf(UD_INITSIZE, BPRI_LO)) {
467 TRACE_1(TR_FAC_KRPC, TR_SVC_CLTS_KSEND_END,
468 "svc_clts_ksend_end:(%S)", "strwaitbuf");
469 return (FALSE);
470 }
471 }
472
473 /*
474 * Initialize the XDR encode stream. Additional mblks
475 * will be allocated if necessary. They will be UD_MAXSIZE
476 * sized.
477 */
478 xdrmblk_init(xdrs, mp, XDR_ENCODE, UD_MAXSIZE);
479
480 /*
481 * Leave some space for protocol headers.
482 */
483 (void) XDR_SETPOS(xdrs, 512);
484 mp->b_rptr += 512;
485
486 msg->rm_xid = clone_xprt->xp_xid;
487
488 ud->ud_resp->b_cont = mp;
489
490 TRACE_0(TR_FAC_KRPC, TR_XDR_REPLYMSG_START,
491 "xdr_replymsg_start:");
492 if (!(xdr_replymsg(xdrs, msg) &&
493 (!has_args || SVCAUTH_WRAP(&clone_xprt->xp_auth, xdrs,
494 xdr_results, xdr_location)))) {
495 XDR_DESTROY(xdrs);
496 TRACE_1(TR_FAC_KRPC, TR_XDR_REPLYMSG_END,
497 "xdr_replymsg_end:(%S)", "bad");
498 RPCLOG0(1, "xdr_replymsg/SVCAUTH_WRAP failed\n");
499 goto out;
500 }
501 TRACE_1(TR_FAC_KRPC, TR_XDR_REPLYMSG_END,
502 "xdr_replymsg_end:(%S)", "good");
503
504 } else if (!(xdr_replymsg_body(xdrs, msg) &&
505 (!has_args || SVCAUTH_WRAP(&clone_xprt->xp_auth, xdrs,
506 xdr_results, xdr_location)))) {
507 XDR_DESTROY(xdrs);
508 RPCLOG0(1, "xdr_replymsg_body/SVCAUTH_WRAP failed\n");
509 goto out;
510 }
511
512 XDR_DESTROY(xdrs);
513
514 msgsz = (int)xmsgsize(ud->ud_resp->b_cont);
515
516 if (msgsz <= 0 || (clone_xprt->xp_msg_size != -1 &&
517 msgsz > clone_xprt->xp_msg_size)) {
518 #ifdef DEBUG
519 cmn_err(CE_NOTE,
520 "KRPC: server response message of %d bytes; transport limits are [0, %d]",
521 msgsz, clone_xprt->xp_msg_size);
522 #endif
523 goto out;
524 }
525
526 /*
527 * Construct the T_unitdata_req. We take advantage of the fact that
528 * T_unitdata_ind looks just like T_unitdata_req, except for the
529 * primitive type. Reusing it means we preserve the SCM_UCRED, and
530 * we must preserve it for TX to work.
531 *
532 * This has the side effect that we can also pass certain receive-side
533 * options like IPV6_PKTINFO back down the send side. This implies
534 * that we can not ASSERT on a non-NULL db_credp when we have send-side
535 * options in UDP.
536 */
537 ASSERT(MBLKL(ud->ud_resp) >= TUNITDATAREQSZ);
538 udreq = (struct T_unitdata_req *)ud->ud_resp->b_rptr;
539 ASSERT(udreq->PRIM_type == T_UNITDATA_IND);
540 udreq->PRIM_type = T_UNITDATA_REQ;
541
542 /*
543 * If the local IPv4 transport address is known use it as a source
544 * address for the outgoing UDP packet.
545 */
546 if (((sin_t *)(clone_xprt->xp_lcladdr.buf))->sin_family == AF_INET) {
547 struct T_opthdr *opthdr;
548 in_pktinfo_t *pktinfo;
549 size_t size;
550
551 if (udreq->DEST_length == 0)
552 udreq->OPT_offset = _TPI_ALIGN_TOPT(TUNITDATAREQSZ);
553 else
554 udreq->OPT_offset = _TPI_ALIGN_TOPT(udreq->DEST_offset +
555 udreq->DEST_length);
556
557 udreq->OPT_length = sizeof (struct T_opthdr) +
558 sizeof (in_pktinfo_t);
559
560 size = udreq->OPT_length + udreq->OPT_offset;
561
562 /* make sure we have enough space for the option data */
563 mp = reallocb(ud->ud_resp, size, 1);
564 if (mp == NULL)
565 goto out;
566 ud->ud_resp = mp;
567 udreq = (struct T_unitdata_req *)mp->b_rptr;
568
569 /* set desired option header */
570 opthdr = (struct T_opthdr *)(mp->b_rptr + udreq->OPT_offset);
571 opthdr->len = udreq->OPT_length;
572 opthdr->level = IPPROTO_IP;
573 opthdr->name = IP_PKTINFO;
574
575 /*
576 * 1. set source IP of outbound packet
577 * 2. value '0' for index means IP layer uses this as source
578 * address
579 */
580 pktinfo = (in_pktinfo_t *)(opthdr + 1);
581 (void) memset(pktinfo, 0, sizeof (in_pktinfo_t));
582 pktinfo->ipi_spec_dst.s_addr =
583 ((sin_t *)(clone_xprt->xp_lcladdr.buf))->sin_addr.s_addr;
584 pktinfo->ipi_ifindex = 0;
585
586 /* adjust the end of active data */
587 mp->b_wptr = mp->b_rptr + size;
588 }
589
590 put(clone_xprt->xp_wq, ud->ud_resp);
591 stat = TRUE;
592 ud->ud_resp = NULL;
593
594 out:
595 if (stat == FALSE) {
596 freemsg(ud->ud_resp);
597 ud->ud_resp = NULL;
598 }
599
600 /*
601 * This is completely disgusting. If public is set it is
602 * a pointer to a structure whose first field is the address
603 * of the function to free that structure and any related
604 * stuff. (see rrokfree in nfs_xdr.c).
605 */
606 if (xdrs->x_public) {
607 /* LINTED pointer alignment */
608 (**((int (**)())xdrs->x_public))(xdrs->x_public);
609 }
610
611 TRACE_1(TR_FAC_KRPC, TR_SVC_CLTS_KSEND_END,
612 "svc_clts_ksend_end:(%S)", "done");
613 return (stat);
614 }
615
616 /*
617 * Deserialize arguments.
618 */
619 static bool_t
svc_clts_kgetargs(SVCXPRT * clone_xprt,xdrproc_t xdr_args,caddr_t args_ptr)620 svc_clts_kgetargs(SVCXPRT *clone_xprt, xdrproc_t xdr_args,
621 caddr_t args_ptr)
622 {
623
624 /* LINTED pointer alignment */
625 return (SVCAUTH_UNWRAP(&clone_xprt->xp_auth, &clone_xprt->xp_xdrin,
626 xdr_args, args_ptr));
627
628 }
629
630 static bool_t
svc_clts_kfreeargs(SVCXPRT * clone_xprt,xdrproc_t xdr_args,caddr_t args_ptr)631 svc_clts_kfreeargs(SVCXPRT *clone_xprt, xdrproc_t xdr_args,
632 caddr_t args_ptr)
633 {
634 /* LINTED pointer alignment */
635 struct udp_data *ud = (struct udp_data *)clone_xprt->xp_p2buf;
636 XDR *xdrs = &clone_xprt->xp_xdrin;
637 bool_t retval;
638
639 if (args_ptr) {
640 xdrs->x_op = XDR_FREE;
641 retval = (*xdr_args)(xdrs, args_ptr);
642 } else
643 retval = TRUE;
644
645 XDR_DESTROY(xdrs);
646
647 if (ud->ud_inmp) {
648 freemsg(ud->ud_inmp);
649 ud->ud_inmp = NULL;
650 }
651
652 return (retval);
653 }
654
655 static int32_t *
svc_clts_kgetres(SVCXPRT * clone_xprt,int size)656 svc_clts_kgetres(SVCXPRT *clone_xprt, int size)
657 {
658 /* LINTED pointer alignment */
659 struct udp_data *ud = (struct udp_data *)clone_xprt->xp_p2buf;
660 XDR *xdrs = &clone_xprt->xp_xdrout;
661 mblk_t *mp;
662 int32_t *buf;
663 struct rpc_msg rply;
664
665 /*
666 * Allocate an initial mblk for the response data.
667 */
668 while ((mp = allocb(UD_INITSIZE, BPRI_LO)) == NULL) {
669 if (strwaitbuf(UD_INITSIZE, BPRI_LO)) {
670 return (NULL);
671 }
672 }
673
674 mp->b_cont = NULL;
675
676 /*
677 * Initialize the XDR encode stream. Additional mblks
678 * will be allocated if necessary. They will be UD_MAXSIZE
679 * sized.
680 */
681 xdrmblk_init(xdrs, mp, XDR_ENCODE, UD_MAXSIZE);
682
683 /*
684 * Leave some space for protocol headers.
685 */
686 (void) XDR_SETPOS(xdrs, 512);
687 mp->b_rptr += 512;
688
689 /*
690 * Assume a successful RPC since most of them are.
691 */
692 rply.rm_xid = clone_xprt->xp_xid;
693 rply.rm_direction = REPLY;
694 rply.rm_reply.rp_stat = MSG_ACCEPTED;
695 rply.acpted_rply.ar_verf = clone_xprt->xp_verf;
696 rply.acpted_rply.ar_stat = SUCCESS;
697
698 if (!xdr_replymsg_hdr(xdrs, &rply)) {
699 XDR_DESTROY(xdrs);
700 freeb(mp);
701 return (NULL);
702 }
703
704 buf = XDR_INLINE(xdrs, size);
705
706 if (buf == NULL) {
707 XDR_DESTROY(xdrs);
708 freeb(mp);
709 } else {
710 ud->ud_resp->b_cont = mp;
711 }
712
713 return (buf);
714 }
715
716 static void
svc_clts_kfreeres(SVCXPRT * clone_xprt)717 svc_clts_kfreeres(SVCXPRT *clone_xprt)
718 {
719 /* LINTED pointer alignment */
720 struct udp_data *ud = (struct udp_data *)clone_xprt->xp_p2buf;
721
722 if (ud->ud_resp == NULL || ud->ud_resp->b_cont == NULL)
723 return;
724
725 XDR_DESTROY(&clone_xprt->xp_xdrout);
726
727 /*
728 * SVC_FREERES() is called whenever the server decides not to
729 * send normal reply. Thus, we expect only one mblk to be allocated,
730 * because we have not attempted any XDR encoding.
731 * If we do any XDR encoding and we get an error, then SVC_REPLY()
732 * will freemsg(ud->ud_resp);
733 */
734 ASSERT(ud->ud_resp->b_cont->b_cont == NULL);
735 freeb(ud->ud_resp->b_cont);
736 ud->ud_resp->b_cont = NULL;
737 }
738
739 /*
740 * the dup cacheing routines below provide a cache of non-failure
741 * transaction id's. rpc service routines can use this to detect
742 * retransmissions and re-send a non-failure response.
743 */
744
745 /*
746 * MAXDUPREQS is the number of cached items. It should be adjusted
747 * to the service load so that there is likely to be a response entry
748 * when the first retransmission comes in.
749 */
750 #define MAXDUPREQS 1024
751
752 /*
753 * This should be appropriately scaled to MAXDUPREQS.
754 */
755 #define DRHASHSZ 257
756
757 #if ((DRHASHSZ & (DRHASHSZ - 1)) == 0)
758 #define XIDHASH(xid) ((xid) & (DRHASHSZ - 1))
759 #else
760 #define XIDHASH(xid) ((xid) % DRHASHSZ)
761 #endif
762 #define DRHASH(dr) XIDHASH((dr)->dr_xid)
763 #define REQTOXID(req) ((req)->rq_xprt->xp_xid)
764
765 static int ndupreqs = 0;
766 int maxdupreqs = MAXDUPREQS;
767 static kmutex_t dupreq_lock;
768 static struct dupreq *drhashtbl[DRHASHSZ];
769 static int drhashstat[DRHASHSZ];
770
771 static void unhash(struct dupreq *);
772
773 /*
774 * drmru points to the head of a circular linked list in lru order.
775 * drmru->dr_next == drlru
776 */
777 struct dupreq *drmru;
778
779 /*
780 * PSARC 2003/523 Contract Private Interface
781 * svc_clts_kdup
782 * Changes must be reviewed by Solaris File Sharing
783 * Changes must be communicated to contract-2003-523@sun.com
784 *
785 * svc_clts_kdup searches the request cache and returns 0 if the
786 * request is not found in the cache. If it is found, then it
787 * returns the state of the request (in progress or done) and
788 * the status or attributes that were part of the original reply.
789 *
790 * If DUP_DONE (there is a duplicate) svc_clts_kdup copies over the
791 * value of the response. In that case, also return in *dupcachedp
792 * whether the response free routine is cached in the dupreq - in which case
793 * the caller should not be freeing it, because it will be done later
794 * in the svc_clts_kdup code when the dupreq is reused.
795 */
796 static int
svc_clts_kdup(struct svc_req * req,caddr_t res,int size,struct dupreq ** drpp,bool_t * dupcachedp)797 svc_clts_kdup(struct svc_req *req, caddr_t res, int size, struct dupreq **drpp,
798 bool_t *dupcachedp)
799 {
800 struct rpc_clts_server *stats = CLONE2STATS(req->rq_xprt);
801 struct dupreq *dr;
802 uint32_t xid;
803 uint32_t drhash;
804 int status;
805
806 xid = REQTOXID(req);
807 mutex_enter(&dupreq_lock);
808 RSSTAT_INCR(stats, rsdupchecks);
809 /*
810 * Check to see whether an entry already exists in the cache.
811 */
812 dr = drhashtbl[XIDHASH(xid)];
813 while (dr != NULL) {
814 if (dr->dr_xid == xid &&
815 dr->dr_proc == req->rq_proc &&
816 dr->dr_prog == req->rq_prog &&
817 dr->dr_vers == req->rq_vers &&
818 dr->dr_addr.len == req->rq_xprt->xp_rtaddr.len &&
819 bcmp(dr->dr_addr.buf, req->rq_xprt->xp_rtaddr.buf,
820 dr->dr_addr.len) == 0) {
821 status = dr->dr_status;
822 if (status == DUP_DONE) {
823 bcopy(dr->dr_resp.buf, res, size);
824 if (dupcachedp != NULL)
825 *dupcachedp = (dr->dr_resfree != NULL);
826 } else {
827 dr->dr_status = DUP_INPROGRESS;
828 *drpp = dr;
829 }
830 RSSTAT_INCR(stats, rsdupreqs);
831 mutex_exit(&dupreq_lock);
832 return (status);
833 }
834 dr = dr->dr_chain;
835 }
836
837 /*
838 * There wasn't an entry, either allocate a new one or recycle
839 * an old one.
840 */
841 if (ndupreqs < maxdupreqs) {
842 dr = kmem_alloc(sizeof (*dr), KM_NOSLEEP);
843 if (dr == NULL) {
844 mutex_exit(&dupreq_lock);
845 return (DUP_ERROR);
846 }
847 dr->dr_resp.buf = NULL;
848 dr->dr_resp.maxlen = 0;
849 dr->dr_addr.buf = NULL;
850 dr->dr_addr.maxlen = 0;
851 if (drmru) {
852 dr->dr_next = drmru->dr_next;
853 drmru->dr_next = dr;
854 } else {
855 dr->dr_next = dr;
856 }
857 ndupreqs++;
858 } else {
859 dr = drmru->dr_next;
860 while (dr->dr_status == DUP_INPROGRESS) {
861 dr = dr->dr_next;
862 if (dr == drmru->dr_next) {
863 cmn_err(CE_WARN, "svc_clts_kdup no slots free");
864 mutex_exit(&dupreq_lock);
865 return (DUP_ERROR);
866 }
867 }
868 unhash(dr);
869 if (dr->dr_resfree) {
870 (*dr->dr_resfree)(dr->dr_resp.buf);
871 }
872 }
873 dr->dr_resfree = NULL;
874 drmru = dr;
875
876 dr->dr_xid = REQTOXID(req);
877 dr->dr_prog = req->rq_prog;
878 dr->dr_vers = req->rq_vers;
879 dr->dr_proc = req->rq_proc;
880 if (dr->dr_addr.maxlen < req->rq_xprt->xp_rtaddr.len) {
881 if (dr->dr_addr.buf != NULL)
882 kmem_free(dr->dr_addr.buf, dr->dr_addr.maxlen);
883 dr->dr_addr.maxlen = req->rq_xprt->xp_rtaddr.len;
884 dr->dr_addr.buf = kmem_alloc(dr->dr_addr.maxlen,
885 KM_NOSLEEP);
886 if (dr->dr_addr.buf == NULL) {
887 dr->dr_addr.maxlen = 0;
888 dr->dr_status = DUP_DROP;
889 mutex_exit(&dupreq_lock);
890 return (DUP_ERROR);
891 }
892 }
893 dr->dr_addr.len = req->rq_xprt->xp_rtaddr.len;
894 bcopy(req->rq_xprt->xp_rtaddr.buf, dr->dr_addr.buf, dr->dr_addr.len);
895 if (dr->dr_resp.maxlen < size) {
896 if (dr->dr_resp.buf != NULL)
897 kmem_free(dr->dr_resp.buf, dr->dr_resp.maxlen);
898 dr->dr_resp.maxlen = (unsigned int)size;
899 dr->dr_resp.buf = kmem_alloc(size, KM_NOSLEEP);
900 if (dr->dr_resp.buf == NULL) {
901 dr->dr_resp.maxlen = 0;
902 dr->dr_status = DUP_DROP;
903 mutex_exit(&dupreq_lock);
904 return (DUP_ERROR);
905 }
906 }
907 dr->dr_status = DUP_INPROGRESS;
908
909 drhash = (uint32_t)DRHASH(dr);
910 dr->dr_chain = drhashtbl[drhash];
911 drhashtbl[drhash] = dr;
912 drhashstat[drhash]++;
913 mutex_exit(&dupreq_lock);
914 *drpp = dr;
915 return (DUP_NEW);
916 }
917
918 /*
919 * PSARC 2003/523 Contract Private Interface
920 * svc_clts_kdupdone
921 * Changes must be reviewed by Solaris File Sharing
922 * Changes must be communicated to contract-2003-523@sun.com
923 *
924 * svc_clts_kdupdone marks the request done (DUP_DONE or DUP_DROP)
925 * and stores the response.
926 */
927 static void
svc_clts_kdupdone(struct dupreq * dr,caddr_t res,void (* dis_resfree)(),int size,int status)928 svc_clts_kdupdone(struct dupreq *dr, caddr_t res, void (*dis_resfree)(),
929 int size, int status)
930 {
931
932 ASSERT(dr->dr_resfree == NULL);
933 if (status == DUP_DONE) {
934 bcopy(res, dr->dr_resp.buf, size);
935 dr->dr_resfree = dis_resfree;
936 }
937 dr->dr_status = status;
938 }
939
940 /*
941 * This routine expects that the mutex, dupreq_lock, is already held.
942 */
943 static void
unhash(struct dupreq * dr)944 unhash(struct dupreq *dr)
945 {
946 struct dupreq *drt;
947 struct dupreq *drtprev = NULL;
948 uint32_t drhash;
949
950 ASSERT(MUTEX_HELD(&dupreq_lock));
951
952 drhash = (uint32_t)DRHASH(dr);
953 drt = drhashtbl[drhash];
954 while (drt != NULL) {
955 if (drt == dr) {
956 drhashstat[drhash]--;
957 if (drtprev == NULL) {
958 drhashtbl[drhash] = drt->dr_chain;
959 } else {
960 drtprev->dr_chain = drt->dr_chain;
961 }
962 return;
963 }
964 drtprev = drt;
965 drt = drt->dr_chain;
966 }
967 }
968
969 void
svc_clts_stats_init(zoneid_t zoneid,struct rpc_clts_server ** statsp)970 svc_clts_stats_init(zoneid_t zoneid, struct rpc_clts_server **statsp)
971 {
972 kstat_t *ksp;
973 kstat_named_t *knp;
974
975 knp = rpcstat_zone_init_common(zoneid, "unix", "rpc_clts_server",
976 (const kstat_named_t *)&clts_rsstat_tmpl,
977 sizeof (clts_rsstat_tmpl));
978 /*
979 * Backwards compatibility for old kstat clients
980 */
981 ksp = kstat_create_zone("unix", 0, "rpc_server", "rpc",
982 KSTAT_TYPE_NAMED, clts_rsstat_ndata,
983 KSTAT_FLAG_VIRTUAL | KSTAT_FLAG_WRITABLE, zoneid);
984 if (ksp) {
985 ksp->ks_data = knp;
986 kstat_install(ksp);
987 }
988 *statsp = (struct rpc_clts_server *)knp;
989 }
990
991 void
svc_clts_stats_fini(zoneid_t zoneid,struct rpc_clts_server ** statsp)992 svc_clts_stats_fini(zoneid_t zoneid, struct rpc_clts_server **statsp)
993 {
994 rpcstat_zone_fini_common(zoneid, "unix", "rpc_clts_server");
995 kstat_delete_byname_zone("unix", 0, "rpc_server", zoneid);
996 kmem_free(*statsp, sizeof (clts_rsstat_tmpl));
997 }
998
999 void
svc_clts_init()1000 svc_clts_init()
1001 {
1002 /*
1003 * Check to make sure that the clts private data will fit into
1004 * the stack buffer allocated by svc_run. The compiler should
1005 * remove this check, but it's a safety net if the udp_data
1006 * structure ever changes.
1007 */
1008 /*CONSTANTCONDITION*/
1009 ASSERT(sizeof (struct udp_data) <= SVC_P2LEN);
1010
1011 mutex_init(&dupreq_lock, NULL, MUTEX_DEFAULT, NULL);
1012 }
1013