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 */
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_cots.c
37 * Server side for connection-oriented 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/strsubr.h>
47 #include <sys/strsun.h>
48 #include <sys/stropts.h>
49 #include <sys/tiuser.h>
50 #include <sys/timod.h>
51 #include <sys/tihdr.h>
52 #include <sys/fcntl.h>
53 #include <sys/errno.h>
54 #include <sys/kmem.h>
55 #include <sys/systm.h>
56 #include <sys/debug.h>
57 #include <sys/cmn_err.h>
58 #include <sys/kstat.h>
59 #include <sys/vtrace.h>
60
61 #include <rpc/types.h>
62 #include <rpc/xdr.h>
63 #include <rpc/auth.h>
64 #include <rpc/rpc_msg.h>
65 #include <rpc/svc.h>
66 #include <inet/ip.h>
67
68 #define COTS_MAX_ALLOCSIZE 2048
69 #define MSG_OFFSET 128 /* offset of call into the mblk */
70 #define RM_HDR_SIZE 4 /* record mark header size */
71
72 /*
73 * Routines exported through ops vector.
74 */
75 static bool_t svc_cots_krecv(SVCXPRT *, mblk_t *, struct rpc_msg *);
76 static bool_t svc_cots_ksend(SVCXPRT *, struct rpc_msg *);
77 static bool_t svc_cots_kgetargs(SVCXPRT *, xdrproc_t, caddr_t);
78 static bool_t svc_cots_kfreeargs(SVCXPRT *, xdrproc_t, caddr_t);
79 static void svc_cots_kdestroy(SVCMASTERXPRT *);
80 static int svc_cots_kdup(struct svc_req *, caddr_t, int,
81 struct dupreq **, bool_t *);
82 static void svc_cots_kdupdone(struct dupreq *, caddr_t,
83 void (*)(), int, int);
84 static int32_t *svc_cots_kgetres(SVCXPRT *, int);
85 static void svc_cots_kfreeres(SVCXPRT *);
86 static void svc_cots_kclone_destroy(SVCXPRT *);
87 static void svc_cots_kstart(SVCMASTERXPRT *);
88 static void svc_cots_ktattrs(SVCXPRT *, int, void **);
89
90 /*
91 * Server transport operations vector.
92 */
93 struct svc_ops svc_cots_op = {
94 svc_cots_krecv, /* Get requests */
95 svc_cots_kgetargs, /* Deserialize arguments */
96 svc_cots_ksend, /* Send reply */
97 svc_cots_kfreeargs, /* Free argument data space */
98 svc_cots_kdestroy, /* Destroy transport handle */
99 svc_cots_kdup, /* Check entry in dup req cache */
100 svc_cots_kdupdone, /* Mark entry in dup req cache as done */
101 svc_cots_kgetres, /* Get pointer to response buffer */
102 svc_cots_kfreeres, /* Destroy pre-serialized response header */
103 svc_cots_kclone_destroy, /* Destroy a clone xprt */
104 svc_cots_kstart, /* Tell `ready-to-receive' to rpcmod */
105 NULL, /* Transport specific clone xprt */
106 svc_cots_ktattrs /* Transport Attributes */
107 };
108
109 /*
110 * Master transport private data.
111 * Kept in xprt->xp_p2.
112 */
113 struct cots_master_data {
114 char *cmd_src_addr; /* client's address */
115 int cmd_xprt_started; /* flag for clone routine to call */
116 /* rpcmod's start routine. */
117 struct rpc_cots_server *cmd_stats; /* stats for zone */
118 };
119
120 /*
121 * Transport private data.
122 * Kept in clone_xprt->xp_p2buf.
123 */
124 typedef struct cots_data {
125 mblk_t *cd_mp; /* pre-allocated reply message */
126 mblk_t *cd_req_mp; /* request message */
127 } cots_data_t;
128
129 /*
130 * Server statistics
131 * NOTE: This structure type is duplicated in the NFS fast path.
132 */
133 static const struct rpc_cots_server {
134 kstat_named_t rscalls;
135 kstat_named_t rsbadcalls;
136 kstat_named_t rsnullrecv;
137 kstat_named_t rsbadlen;
138 kstat_named_t rsxdrcall;
139 kstat_named_t rsdupchecks;
140 kstat_named_t rsdupreqs;
141 } cots_rsstat_tmpl = {
142 { "calls", KSTAT_DATA_UINT64 },
143 { "badcalls", KSTAT_DATA_UINT64 },
144 { "nullrecv", KSTAT_DATA_UINT64 },
145 { "badlen", KSTAT_DATA_UINT64 },
146 { "xdrcall", KSTAT_DATA_UINT64 },
147 { "dupchecks", KSTAT_DATA_UINT64 },
148 { "dupreqs", KSTAT_DATA_UINT64 }
149 };
150
151 #define CLONE2STATS(clone_xprt) \
152 ((struct cots_master_data *)(clone_xprt)->xp_master->xp_p2)->cmd_stats
153 #define RSSTAT_INCR(s, x) \
154 atomic_inc_64(&(s)->x.value.ui64)
155
156 /*
157 * Pointer to a transport specific `ready to receive' function in rpcmod
158 * (set from rpcmod).
159 */
160 void (*mir_start)(queue_t *);
161 uint_t *svc_max_msg_sizep;
162
163 /*
164 * the address size of the underlying transport can sometimes be
165 * unknown (tinfo->ADDR_size == -1). For this case, it is
166 * necessary to figure out what the size is so the correct amount
167 * of data is allocated. This is an itterative process:
168 * 1. take a good guess (use T_MINADDRSIZE)
169 * 2. try it.
170 * 3. if it works then everything is ok
171 * 4. if the error is ENAMETOLONG, double the guess
172 * 5. go back to step 2.
173 */
174 #define T_UNKNOWNADDRSIZE (-1)
175 #define T_MINADDRSIZE 32
176
177 /*
178 * Create a transport record.
179 * The transport record, output buffer, and private data structure
180 * are allocated. The output buffer is serialized into using xdrmem.
181 * There is one transport record per user process which implements a
182 * set of services.
183 */
184 static kmutex_t cots_kcreate_lock;
185
186 int
187 svc_cots_kcreate(file_t *fp, uint_t max_msgsize, struct T_info_ack *tinfo,
188 SVCMASTERXPRT **nxprt)
189 {
190 struct cots_master_data *cmd;
191 int err, retval;
192 SVCMASTERXPRT *xprt;
193 struct rpcstat *rpcstat;
194 struct T_addr_ack *ack_p;
195 struct strioctl getaddr;
196
197 if (nxprt == NULL)
198 return (EINVAL);
199
200 rpcstat = zone_getspecific(rpcstat_zone_key, curproc->p_zone);
201 ASSERT(rpcstat != NULL);
202
203 xprt = kmem_zalloc(sizeof (SVCMASTERXPRT), KM_SLEEP);
204
205 cmd = kmem_zalloc(sizeof (*cmd) + sizeof (*ack_p)
206 + (2 * sizeof (sin6_t)), KM_SLEEP);
207
208 ack_p = (struct T_addr_ack *)&cmd[1];
209
210 if ((tinfo->TIDU_size > COTS_MAX_ALLOCSIZE) ||
211 (tinfo->TIDU_size <= 0))
212 xprt->xp_msg_size = COTS_MAX_ALLOCSIZE;
213 else {
214 xprt->xp_msg_size = tinfo->TIDU_size -
215 (tinfo->TIDU_size % BYTES_PER_XDR_UNIT);
216 }
217
218 xprt->xp_ops = &svc_cots_op;
219 xprt->xp_p2 = (caddr_t)cmd;
220 cmd->cmd_xprt_started = 0;
221 cmd->cmd_stats = rpcstat->rpc_cots_server;
222
223 getaddr.ic_cmd = TI_GETINFO;
224 getaddr.ic_timout = -1;
225 getaddr.ic_len = sizeof (*ack_p) + (2 * sizeof (sin6_t));
226 getaddr.ic_dp = (char *)ack_p;
227 ack_p->PRIM_type = T_ADDR_REQ;
228
229 err = strioctl(fp->f_vnode, I_STR, (intptr_t)&getaddr,
230 0, K_TO_K, CRED(), &retval);
231 if (err) {
232 kmem_free(cmd, sizeof (*cmd) + sizeof (*ack_p) +
233 (2 * sizeof (sin6_t)));
234 kmem_free(xprt, sizeof (SVCMASTERXPRT));
235 return (err);
236 }
237
238 xprt->xp_rtaddr.maxlen = ack_p->REMADDR_length;
239 xprt->xp_rtaddr.len = ack_p->REMADDR_length;
240 cmd->cmd_src_addr = xprt->xp_rtaddr.buf =
241 (char *)ack_p + ack_p->REMADDR_offset;
242
243 xprt->xp_lcladdr.maxlen = ack_p->LOCADDR_length;
244 xprt->xp_lcladdr.len = ack_p->LOCADDR_length;
245 xprt->xp_lcladdr.buf = (char *)ack_p + ack_p->LOCADDR_offset;
246
247 /*
248 * If the current sanity check size in rpcmod is smaller
249 * than the size needed for this xprt, then increase
250 * the sanity check.
251 */
252 if (max_msgsize != 0 && svc_max_msg_sizep &&
253 max_msgsize > *svc_max_msg_sizep) {
254
255 /* This check needs a lock */
256 mutex_enter(&cots_kcreate_lock);
257 if (svc_max_msg_sizep && max_msgsize > *svc_max_msg_sizep)
258 *svc_max_msg_sizep = max_msgsize;
259 mutex_exit(&cots_kcreate_lock);
260 }
261
262 *nxprt = xprt;
263
264 return (0);
265 }
266
267 /*
268 * Destroy a master transport record.
269 * Frees the space allocated for a transport record.
270 */
271 static void
272 svc_cots_kdestroy(SVCMASTERXPRT *xprt)
273 {
274 struct cots_master_data *cmd = (struct cots_master_data *)xprt->xp_p2;
275
276 ASSERT(cmd);
277
278 if (xprt->xp_netid)
279 kmem_free(xprt->xp_netid, strlen(xprt->xp_netid) + 1);
280 if (xprt->xp_addrmask.maxlen)
281 kmem_free(xprt->xp_addrmask.buf, xprt->xp_addrmask.maxlen);
282
283 mutex_destroy(&xprt->xp_req_lock);
284 mutex_destroy(&xprt->xp_thread_lock);
285
286 kmem_free(cmd, sizeof (*cmd) + sizeof (struct T_addr_ack) +
287 (2 * sizeof (sin6_t)));
288
289 kmem_free(xprt, sizeof (SVCMASTERXPRT));
290 }
291
292 /*
293 * svc_tli_kcreate() calls this function at the end to tell
294 * rpcmod that the transport is ready to receive requests.
295 */
296 static void
297 svc_cots_kstart(SVCMASTERXPRT *xprt)
298 {
299 struct cots_master_data *cmd = (struct cots_master_data *)xprt->xp_p2;
300
301 if (cmd->cmd_xprt_started == 0) {
302 /*
303 * Acquire the xp_req_lock in order to use xp_wq
304 * safely (we don't want to qenable a queue that has
305 * already been closed).
306 */
307 mutex_enter(&xprt->xp_req_lock);
308 if (cmd->cmd_xprt_started == 0 &&
309 xprt->xp_wq != NULL) {
310 (*mir_start)(xprt->xp_wq);
311 cmd->cmd_xprt_started = 1;
312 }
313 mutex_exit(&xprt->xp_req_lock);
314 }
315 }
316
317 /*
318 * Transport-type specific part of svc_xprt_cleanup().
319 */
320 static void
321 svc_cots_kclone_destroy(SVCXPRT *clone_xprt)
322 {
323 cots_data_t *cd = (cots_data_t *)clone_xprt->xp_p2buf;
324
325 if (cd->cd_req_mp) {
326 freemsg(cd->cd_req_mp);
327 cd->cd_req_mp = (mblk_t *)0;
328 }
329 ASSERT(cd->cd_mp == NULL);
330 }
331
332 /*
333 * Transport Attributes.
334 */
335 static void
336 svc_cots_ktattrs(SVCXPRT *clone_xprt, int attrflag, void **tattr)
337 {
338 *tattr = NULL;
339
340 switch (attrflag) {
341 case SVC_TATTR_ADDRMASK:
342 *tattr = (void *)&clone_xprt->xp_master->xp_addrmask;
343 }
344 }
345
346 /*
347 * Receive rpc requests.
348 * Checks if the message is intact, and deserializes the call packet.
349 */
350 static bool_t
351 svc_cots_krecv(SVCXPRT *clone_xprt, mblk_t *mp, struct rpc_msg *msg)
352 {
353 cots_data_t *cd = (cots_data_t *)clone_xprt->xp_p2buf;
354 XDR *xdrs = &clone_xprt->xp_xdrin;
355 struct rpc_cots_server *stats = CLONE2STATS(clone_xprt);
356
357 TRACE_0(TR_FAC_KRPC, TR_SVC_COTS_KRECV_START,
358 "svc_cots_krecv_start:");
359 RPCLOG(4, "svc_cots_krecv_start clone_xprt = %p:\n",
360 (void *)clone_xprt);
361
362 RSSTAT_INCR(stats, rscalls);
363
364 if (mp->b_datap->db_type != M_DATA) {
365 RPCLOG(16, "svc_cots_krecv bad db_type %d\n",
366 mp->b_datap->db_type);
367 goto bad;
368 }
369
370 xdrmblk_init(xdrs, mp, XDR_DECODE, 0);
371
372 TRACE_0(TR_FAC_KRPC, TR_XDR_CALLMSG_START,
373 "xdr_callmsg_start:");
374 RPCLOG0(4, "xdr_callmsg_start:\n");
375 if (!xdr_callmsg(xdrs, msg)) {
376 XDR_DESTROY(xdrs);
377 TRACE_1(TR_FAC_KRPC, TR_XDR_CALLMSG_END,
378 "xdr_callmsg_end:(%S)", "bad");
379 RPCLOG0(1, "svc_cots_krecv xdr_callmsg failure\n");
380 RSSTAT_INCR(stats, rsxdrcall);
381 goto bad;
382 }
383 TRACE_1(TR_FAC_KRPC, TR_XDR_CALLMSG_END,
384 "xdr_callmsg_end:(%S)", "good");
385
386 clone_xprt->xp_xid = msg->rm_xid;
387 cd->cd_req_mp = mp;
388
389 TRACE_1(TR_FAC_KRPC, TR_SVC_COTS_KRECV_END,
390 "svc_cots_krecv_end:(%S)", "good");
391 RPCLOG0(4, "svc_cots_krecv_end:good\n");
392 return (TRUE);
393
394 bad:
395 if (mp)
396 freemsg(mp);
397
398 RSSTAT_INCR(stats, rsbadcalls);
399 TRACE_1(TR_FAC_KRPC, TR_SVC_COTS_KRECV_END,
400 "svc_cots_krecv_end:(%S)", "bad");
401 return (FALSE);
402 }
403
404 /*
405 * Send rpc reply.
406 */
407 static bool_t
408 svc_cots_ksend(SVCXPRT *clone_xprt, struct rpc_msg *msg)
409 {
410 /* LINTED pointer alignment */
411 cots_data_t *cd = (cots_data_t *)clone_xprt->xp_p2buf;
412 XDR *xdrs = &(clone_xprt->xp_xdrout);
413 int retval = FALSE;
414 mblk_t *mp;
415 xdrproc_t xdr_results;
416 caddr_t xdr_location;
417 bool_t has_args;
418
419 TRACE_0(TR_FAC_KRPC, TR_SVC_COTS_KSEND_START,
420 "svc_cots_ksend_start:");
421
422 /*
423 * If there is a result procedure specified in the reply message,
424 * it will be processed in the xdr_replymsg and SVCAUTH_WRAP.
425 * We need to make sure it won't be processed twice, so we null
426 * it for xdr_replymsg here.
427 */
428 has_args = FALSE;
429 if (msg->rm_reply.rp_stat == MSG_ACCEPTED &&
430 msg->rm_reply.rp_acpt.ar_stat == SUCCESS) {
431 if ((xdr_results = msg->acpted_rply.ar_results.proc) != NULL) {
432 has_args = TRUE;
433 xdr_location = msg->acpted_rply.ar_results.where;
434 msg->acpted_rply.ar_results.proc = xdr_void;
435 msg->acpted_rply.ar_results.where = NULL;
436 }
437 }
438
439 mp = cd->cd_mp;
440 if (mp) {
441 /*
442 * The program above pre-allocated an mblk and put
443 * the data in place.
444 */
445 cd->cd_mp = (mblk_t *)NULL;
446 if (!(xdr_replymsg_body(xdrs, msg) &&
447 (!has_args || SVCAUTH_WRAP(&clone_xprt->xp_auth, xdrs,
448 xdr_results, xdr_location)))) {
449 XDR_DESTROY(xdrs);
450 RPCLOG0(1, "svc_cots_ksend: "
451 "xdr_replymsg_body/SVCAUTH_WRAP failed\n");
452 freemsg(mp);
453 goto out;
454 }
455 } else {
456 int len;
457 int mpsize;
458
459 /*
460 * Leave space for protocol headers.
461 */
462 len = MSG_OFFSET + clone_xprt->xp_msg_size;
463
464 /*
465 * Allocate an initial mblk for the response data.
466 */
467 while (!(mp = allocb(len, BPRI_LO))) {
468 RPCLOG0(16, "svc_cots_ksend: allocb failed failed\n");
469 if (strwaitbuf(len, BPRI_LO)) {
470 TRACE_1(TR_FAC_KRPC, TR_SVC_COTS_KSEND_END,
471 "svc_cots_ksend_end:(%S)", "strwaitbuf");
472 RPCLOG0(1,
473 "svc_cots_ksend: strwaitbuf failed\n");
474 goto out;
475 }
476 }
477
478 /*
479 * Initialize the XDR encode stream. Additional mblks
480 * will be allocated if necessary. They will be TIDU
481 * sized.
482 */
483 xdrmblk_init(xdrs, mp, XDR_ENCODE, clone_xprt->xp_msg_size);
484 mpsize = MBLKSIZE(mp);
485 ASSERT(mpsize >= len);
486 ASSERT(mp->b_rptr == mp->b_datap->db_base);
487
488 /*
489 * If the size of mblk is not appreciably larger than what we
490 * asked, then resize the mblk to exactly len bytes. Reason for
491 * this: suppose len is 1600 bytes, the tidu is 1460 bytes
492 * (from TCP over ethernet), and the arguments to RPC require
493 * 2800 bytes. Ideally we want the protocol to render two
494 * ~1400 byte segments over the wire. If allocb() gives us a 2k
495 * mblk, and we allocate a second mblk for the rest, the
496 * protocol module may generate 3 segments over the wire:
497 * 1460 bytes for the first, 448 (2048 - 1600) for the 2nd, and
498 * 892 for the 3rd. If we "waste" 448 bytes in the first mblk,
499 * the XDR encoding will generate two ~1400 byte mblks, and the
500 * protocol module is more likely to produce properly sized
501 * segments.
502 */
503 if ((mpsize >> 1) <= len) {
504 mp->b_rptr += (mpsize - len);
505 }
506
507 /*
508 * Adjust b_rptr to reserve space for the non-data protocol
509 * headers that any downstream modules might like to add, and
510 * for the record marking header.
511 */
512 mp->b_rptr += (MSG_OFFSET + RM_HDR_SIZE);
513
514 XDR_SETPOS(xdrs, (uint_t)(mp->b_rptr - mp->b_datap->db_base));
515 ASSERT(mp->b_wptr == mp->b_rptr);
516
517 msg->rm_xid = clone_xprt->xp_xid;
518
519 TRACE_0(TR_FAC_KRPC, TR_XDR_REPLYMSG_START,
520 "xdr_replymsg_start:");
521 if (!(xdr_replymsg(xdrs, msg) &&
522 (!has_args || SVCAUTH_WRAP(&clone_xprt->xp_auth, xdrs,
523 xdr_results, xdr_location)))) {
524 XDR_DESTROY(xdrs);
525 TRACE_1(TR_FAC_KRPC, TR_XDR_REPLYMSG_END,
526 "xdr_replymsg_end:(%S)", "bad");
527 freemsg(mp);
528 RPCLOG0(1, "svc_cots_ksend: xdr_replymsg/SVCAUTH_WRAP "
529 "failed\n");
530 goto out;
531 }
532 TRACE_1(TR_FAC_KRPC, TR_XDR_REPLYMSG_END,
533 "xdr_replymsg_end:(%S)", "good");
534 }
535
536 XDR_DESTROY(xdrs);
537
538 put(clone_xprt->xp_wq, mp);
539 retval = TRUE;
540
541 out:
542 /*
543 * This is completely disgusting. If public is set it is
544 * a pointer to a structure whose first field is the address
545 * of the function to free that structure and any related
546 * stuff. (see rrokfree in nfs_xdr.c).
547 */
548 if (xdrs->x_public) {
549 /* LINTED pointer alignment */
550 (**((int (**)())xdrs->x_public))(xdrs->x_public);
551 }
552
553 TRACE_1(TR_FAC_KRPC, TR_SVC_COTS_KSEND_END,
554 "svc_cots_ksend_end:(%S)", "done");
555 return (retval);
556 }
557
558 /*
559 * Deserialize arguments.
560 */
561 static bool_t
562 svc_cots_kgetargs(SVCXPRT *clone_xprt, xdrproc_t xdr_args,
563 caddr_t args_ptr)
564 {
565 return (SVCAUTH_UNWRAP(&clone_xprt->xp_auth, &clone_xprt->xp_xdrin,
566 xdr_args, args_ptr));
567 }
568
569 static bool_t
570 svc_cots_kfreeargs(SVCXPRT *clone_xprt, xdrproc_t xdr_args,
571 caddr_t args_ptr)
572 {
573 cots_data_t *cd = (cots_data_t *)clone_xprt->xp_p2buf;
574 /* LINTED pointer alignment */
575 XDR *xdrs = &clone_xprt->xp_xdrin;
576 mblk_t *mp;
577 bool_t retval;
578
579 /*
580 * It is important to call the XDR routine before
581 * freeing the request mblk. Structures in the
582 * XDR data may point into the mblk and require that
583 * the memory be intact during the free routine.
584 */
585 if (args_ptr) {
586 xdrs->x_op = XDR_FREE;
587 retval = (*xdr_args)(xdrs, args_ptr);
588 } else
589 retval = TRUE;
590
591 XDR_DESTROY(xdrs);
592
593 if ((mp = cd->cd_req_mp) != NULL) {
594 cd->cd_req_mp = (mblk_t *)0;
595 freemsg(mp);
596 }
597
598 return (retval);
599 }
600
601 static int32_t *
602 svc_cots_kgetres(SVCXPRT *clone_xprt, int size)
603 {
604 /* LINTED pointer alignment */
605 cots_data_t *cd = (cots_data_t *)clone_xprt->xp_p2buf;
606 XDR *xdrs = &clone_xprt->xp_xdrout;
607 mblk_t *mp;
608 int32_t *buf;
609 struct rpc_msg rply;
610 int len;
611 int mpsize;
612
613 /*
614 * Leave space for protocol headers.
615 */
616 len = MSG_OFFSET + clone_xprt->xp_msg_size;
617
618 /*
619 * Allocate an initial mblk for the response data.
620 */
621 while ((mp = allocb(len, BPRI_LO)) == NULL) {
622 if (strwaitbuf(len, BPRI_LO))
623 return (NULL);
624 }
625
626 /*
627 * Initialize the XDR encode stream. Additional mblks
628 * will be allocated if necessary. They will be TIDU
629 * sized.
630 */
631 xdrmblk_init(xdrs, mp, XDR_ENCODE, clone_xprt->xp_msg_size);
632 mpsize = MBLKSIZE(mp);
633 ASSERT(mpsize >= len);
634 ASSERT(mp->b_rptr == mp->b_datap->db_base);
635
636 /*
637 * If the size of mblk is not appreciably larger than what we
638 * asked, then resize the mblk to exactly len bytes. Reason for
639 * this: suppose len is 1600 bytes, the tidu is 1460 bytes
640 * (from TCP over ethernet), and the arguments to RPC require
641 * 2800 bytes. Ideally we want the protocol to render two
642 * ~1400 byte segments over the wire. If allocb() gives us a 2k
643 * mblk, and we allocate a second mblk for the rest, the
644 * protocol module may generate 3 segments over the wire:
645 * 1460 bytes for the first, 448 (2048 - 1600) for the 2nd, and
646 * 892 for the 3rd. If we "waste" 448 bytes in the first mblk,
647 * the XDR encoding will generate two ~1400 byte mblks, and the
648 * protocol module is more likely to produce properly sized
649 * segments.
650 */
651 if ((mpsize >> 1) <= len) {
652 mp->b_rptr += (mpsize - len);
653 }
654
655 /*
656 * Adjust b_rptr to reserve space for the non-data protocol
657 * headers that any downstream modules might like to add, and
658 * for the record marking header.
659 */
660 mp->b_rptr += (MSG_OFFSET + RM_HDR_SIZE);
661
662 XDR_SETPOS(xdrs, (uint_t)(mp->b_rptr - mp->b_datap->db_base));
663 ASSERT(mp->b_wptr == mp->b_rptr);
664
665 /*
666 * Assume a successful RPC since most of them are.
667 */
668 rply.rm_xid = clone_xprt->xp_xid;
669 rply.rm_direction = REPLY;
670 rply.rm_reply.rp_stat = MSG_ACCEPTED;
671 rply.acpted_rply.ar_verf = clone_xprt->xp_verf;
672 rply.acpted_rply.ar_stat = SUCCESS;
673
674 if (!xdr_replymsg_hdr(xdrs, &rply)) {
675 XDR_DESTROY(xdrs);
676 freeb(mp);
677 return (NULL);
678 }
679
680 buf = XDR_INLINE(xdrs, size);
681 if (buf == NULL) {
682 XDR_DESTROY(xdrs);
683 ASSERT(cd->cd_mp == NULL);
684 freemsg(mp);
685 } else {
686 cd->cd_mp = mp;
687 }
688 return (buf);
689 }
690
691 static void
692 svc_cots_kfreeres(SVCXPRT *clone_xprt)
693 {
694 cots_data_t *cd;
695 mblk_t *mp;
696
697 cd = (cots_data_t *)clone_xprt->xp_p2buf;
698 if ((mp = cd->cd_mp) != NULL) {
699 XDR_DESTROY(&clone_xprt->xp_xdrout);
700 cd->cd_mp = (mblk_t *)NULL;
701 freemsg(mp);
702 }
703 }
704
705 /*
706 * the dup cacheing routines below provide a cache of non-failure
707 * transaction id's. rpc service routines can use this to detect
708 * retransmissions and re-send a non-failure response.
709 */
710
711 /*
712 * MAXDUPREQS is the number of cached items. It should be adjusted
713 * to the service load so that there is likely to be a response entry
714 * when the first retransmission comes in.
715 */
716 #define MAXDUPREQS 1024
717
718 /*
719 * This should be appropriately scaled to MAXDUPREQS.
720 */
721 #define DRHASHSZ 257
722
723 #if ((DRHASHSZ & (DRHASHSZ - 1)) == 0)
724 #define XIDHASH(xid) ((xid) & (DRHASHSZ - 1))
725 #else
726 #define XIDHASH(xid) ((xid) % DRHASHSZ)
727 #endif
728 #define DRHASH(dr) XIDHASH((dr)->dr_xid)
729 #define REQTOXID(req) ((req)->rq_xprt->xp_xid)
730
731 static int cotsndupreqs = 0;
732 int cotsmaxdupreqs = MAXDUPREQS;
733 static kmutex_t cotsdupreq_lock;
734 static struct dupreq *cotsdrhashtbl[DRHASHSZ];
735 static int cotsdrhashstat[DRHASHSZ];
736
737 static void unhash(struct dupreq *);
738
739 /*
740 * cotsdrmru points to the head of a circular linked list in lru order.
741 * cotsdrmru->dr_next == drlru
742 */
743 struct dupreq *cotsdrmru;
744
745 /*
746 * PSARC 2003/523 Contract Private Interface
747 * svc_cots_kdup
748 * Changes must be reviewed by Solaris File Sharing
749 * Changes must be communicated to contract-2003-523@sun.com
750 *
751 * svc_cots_kdup searches the request cache and returns 0 if the
752 * request is not found in the cache. If it is found, then it
753 * returns the state of the request (in progress or done) and
754 * the status or attributes that were part of the original reply.
755 *
756 * If DUP_DONE (there is a duplicate) svc_cots_kdup copies over the
757 * value of the response. In that case, also return in *dupcachedp
758 * whether the response free routine is cached in the dupreq - in which case
759 * the caller should not be freeing it, because it will be done later
760 * in the svc_cots_kdup code when the dupreq is reused.
761 */
762 static int
763 svc_cots_kdup(struct svc_req *req, caddr_t res, int size, struct dupreq **drpp,
764 bool_t *dupcachedp)
765 {
766 struct rpc_cots_server *stats = CLONE2STATS(req->rq_xprt);
767 struct dupreq *dr;
768 uint32_t xid;
769 uint32_t drhash;
770 int status;
771
772 xid = REQTOXID(req);
773 mutex_enter(&cotsdupreq_lock);
774 RSSTAT_INCR(stats, rsdupchecks);
775 /*
776 * Check to see whether an entry already exists in the cache.
777 */
778 dr = cotsdrhashtbl[XIDHASH(xid)];
779 while (dr != NULL) {
780 if (dr->dr_xid == xid &&
781 dr->dr_proc == req->rq_proc &&
782 dr->dr_prog == req->rq_prog &&
783 dr->dr_vers == req->rq_vers &&
784 dr->dr_addr.len == req->rq_xprt->xp_rtaddr.len &&
785 bcmp((caddr_t)dr->dr_addr.buf,
786 (caddr_t)req->rq_xprt->xp_rtaddr.buf,
787 dr->dr_addr.len) == 0) {
788 status = dr->dr_status;
789 if (status == DUP_DONE) {
790 bcopy(dr->dr_resp.buf, res, size);
791 if (dupcachedp != NULL)
792 *dupcachedp = (dr->dr_resfree != NULL);
793 TRACE_0(TR_FAC_KRPC, TR_SVC_COTS_KDUP_DONE,
794 "svc_cots_kdup: DUP_DONE");
795 } else {
796 dr->dr_status = DUP_INPROGRESS;
797 *drpp = dr;
798 TRACE_0(TR_FAC_KRPC,
799 TR_SVC_COTS_KDUP_INPROGRESS,
800 "svc_cots_kdup: DUP_INPROGRESS");
801 }
802 RSSTAT_INCR(stats, rsdupreqs);
803 mutex_exit(&cotsdupreq_lock);
804 return (status);
805 }
806 dr = dr->dr_chain;
807 }
808
809 /*
810 * There wasn't an entry, either allocate a new one or recycle
811 * an old one.
812 */
813 if (cotsndupreqs < cotsmaxdupreqs) {
814 dr = kmem_alloc(sizeof (*dr), KM_NOSLEEP);
815 if (dr == NULL) {
816 mutex_exit(&cotsdupreq_lock);
817 return (DUP_ERROR);
818 }
819 dr->dr_resp.buf = NULL;
820 dr->dr_resp.maxlen = 0;
821 dr->dr_addr.buf = NULL;
822 dr->dr_addr.maxlen = 0;
823 if (cotsdrmru) {
824 dr->dr_next = cotsdrmru->dr_next;
825 cotsdrmru->dr_next = dr;
826 } else {
827 dr->dr_next = dr;
828 }
829 cotsndupreqs++;
830 } else {
831 dr = cotsdrmru->dr_next;
832 while (dr->dr_status == DUP_INPROGRESS) {
833 dr = dr->dr_next;
834 if (dr == cotsdrmru->dr_next) {
835 cmn_err(CE_WARN, "svc_cots_kdup no slots free");
836 mutex_exit(&cotsdupreq_lock);
837 return (DUP_ERROR);
838 }
839 }
840 unhash(dr);
841 if (dr->dr_resfree) {
842 (*dr->dr_resfree)(dr->dr_resp.buf);
843 }
844 }
845 dr->dr_resfree = NULL;
846 cotsdrmru = dr;
847
848 dr->dr_xid = REQTOXID(req);
849 dr->dr_prog = req->rq_prog;
850 dr->dr_vers = req->rq_vers;
851 dr->dr_proc = req->rq_proc;
852 if (dr->dr_addr.maxlen < req->rq_xprt->xp_rtaddr.len) {
853 if (dr->dr_addr.buf != NULL)
854 kmem_free(dr->dr_addr.buf, dr->dr_addr.maxlen);
855 dr->dr_addr.maxlen = req->rq_xprt->xp_rtaddr.len;
856 dr->dr_addr.buf = kmem_alloc(dr->dr_addr.maxlen, KM_NOSLEEP);
857 if (dr->dr_addr.buf == NULL) {
858 dr->dr_addr.maxlen = 0;
859 dr->dr_status = DUP_DROP;
860 mutex_exit(&cotsdupreq_lock);
861 return (DUP_ERROR);
862 }
863 }
864 dr->dr_addr.len = req->rq_xprt->xp_rtaddr.len;
865 bcopy(req->rq_xprt->xp_rtaddr.buf, dr->dr_addr.buf, dr->dr_addr.len);
866 if (dr->dr_resp.maxlen < size) {
867 if (dr->dr_resp.buf != NULL)
868 kmem_free(dr->dr_resp.buf, dr->dr_resp.maxlen);
869 dr->dr_resp.maxlen = (unsigned int)size;
870 dr->dr_resp.buf = kmem_alloc(size, KM_NOSLEEP);
871 if (dr->dr_resp.buf == NULL) {
872 dr->dr_resp.maxlen = 0;
873 dr->dr_status = DUP_DROP;
874 mutex_exit(&cotsdupreq_lock);
875 return (DUP_ERROR);
876 }
877 }
878 dr->dr_status = DUP_INPROGRESS;
879
880 drhash = (uint32_t)DRHASH(dr);
881 dr->dr_chain = cotsdrhashtbl[drhash];
882 cotsdrhashtbl[drhash] = dr;
883 cotsdrhashstat[drhash]++;
884 mutex_exit(&cotsdupreq_lock);
885 *drpp = dr;
886 return (DUP_NEW);
887 }
888
889 /*
890 * PSARC 2003/523 Contract Private Interface
891 * svc_cots_kdupdone
892 * Changes must be reviewed by Solaris File Sharing
893 * Changes must be communicated to contract-2003-523@sun.com
894 *
895 * svc_cots_kdupdone marks the request done (DUP_DONE or DUP_DROP)
896 * and stores the response.
897 */
898 static void
899 svc_cots_kdupdone(struct dupreq *dr, caddr_t res, void (*dis_resfree)(),
900 int size, int status)
901 {
902 ASSERT(dr->dr_resfree == NULL);
903 if (status == DUP_DONE) {
904 bcopy(res, dr->dr_resp.buf, size);
905 dr->dr_resfree = dis_resfree;
906 }
907 dr->dr_status = status;
908 }
909
910 /*
911 * This routine expects that the mutex, cotsdupreq_lock, is already held.
912 */
913 static void
914 unhash(struct dupreq *dr)
915 {
916 struct dupreq *drt;
917 struct dupreq *drtprev = NULL;
918 uint32_t drhash;
919
920 ASSERT(MUTEX_HELD(&cotsdupreq_lock));
921
922 drhash = (uint32_t)DRHASH(dr);
923 drt = cotsdrhashtbl[drhash];
924 while (drt != NULL) {
925 if (drt == dr) {
926 cotsdrhashstat[drhash]--;
927 if (drtprev == NULL) {
928 cotsdrhashtbl[drhash] = drt->dr_chain;
929 } else {
930 drtprev->dr_chain = drt->dr_chain;
931 }
932 return;
933 }
934 drtprev = drt;
935 drt = drt->dr_chain;
936 }
937 }
938
939 void
940 svc_cots_stats_init(zoneid_t zoneid, struct rpc_cots_server **statsp)
941 {
942 *statsp = (struct rpc_cots_server *)rpcstat_zone_init_common(zoneid,
943 "unix", "rpc_cots_server", (const kstat_named_t *)&cots_rsstat_tmpl,
944 sizeof (cots_rsstat_tmpl));
945 }
946
947 void
948 svc_cots_stats_fini(zoneid_t zoneid, struct rpc_cots_server **statsp)
949 {
950 rpcstat_zone_fini_common(zoneid, "unix", "rpc_cots_server");
951 kmem_free(*statsp, sizeof (cots_rsstat_tmpl));
952 }
953
954 void
955 svc_cots_init(void)
956 {
957 /*
958 * Check to make sure that the cots private data will fit into
959 * the stack buffer allocated by svc_run. The ASSERT is a safety
960 * net if the cots_data_t structure ever changes.
961 */
962 /*CONSTANTCONDITION*/
963 ASSERT(sizeof (cots_data_t) <= SVC_P2LEN);
964
965 mutex_init(&cots_kcreate_lock, NULL, MUTEX_DEFAULT, NULL);
966 mutex_init(&cotsdupreq_lock, NULL, MUTEX_DEFAULT, NULL);
967 }