xref: /illumos-gate/usr/src/uts/common/rpc/svc_rdma.c (revision 2695d4f4d1e2a6022c8a279d40c3cb750964974d)
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  * Copyright (c) 1983, 2010, Oracle and/or its affiliates. All rights reserved.
23  * Copyright 2013 Nexenta Systems, Inc.  All rights reserved.
24  */
25 /* Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T */
26 /* All Rights Reserved */
27 /*
28  * Portions of this source code were derived from Berkeley
29  * 4.3 BSD under license from the Regents of the University of
30  * California.
31  */
32 
33 /*
34  * Server side of RPC over RDMA in the kernel.
35  */
36 
37 #include <sys/param.h>
38 #include <sys/types.h>
39 #include <sys/user.h>
40 #include <sys/sysmacros.h>
41 #include <sys/proc.h>
42 #include <sys/file.h>
43 #include <sys/errno.h>
44 #include <sys/kmem.h>
45 #include <sys/debug.h>
46 #include <sys/systm.h>
47 #include <sys/cmn_err.h>
48 #include <sys/kstat.h>
49 #include <sys/vtrace.h>
50 #include <sys/debug.h>
51 
52 #include <rpc/types.h>
53 #include <rpc/xdr.h>
54 #include <rpc/auth.h>
55 #include <rpc/clnt.h>
56 #include <rpc/rpc_msg.h>
57 #include <rpc/svc.h>
58 #include <rpc/rpc_rdma.h>
59 #include <sys/ddi.h>
60 #include <sys/sunddi.h>
61 
62 #include <inet/common.h>
63 #include <inet/ip.h>
64 #include <inet/ip6.h>
65 
66 #include <nfs/nfs.h>
67 #include <sys/sdt.h>
68 
69 #define	SVC_RDMA_SUCCESS 0
70 #define	SVC_RDMA_FAIL -1
71 
72 #define	SVC_CREDIT_FACTOR (0.5)
73 
74 #define	MSG_IS_RPCSEC_GSS(msg)		\
75 	((msg)->rm_reply.rp_acpt.ar_verf.oa_flavor == RPCSEC_GSS)
76 
77 
78 uint32_t rdma_bufs_granted = RDMA_BUFS_GRANT;
79 
80 /*
81  * RDMA transport specific data associated with SVCMASTERXPRT
82  */
83 struct rdma_data {
84 	SVCMASTERXPRT 	*rd_xprt;	/* back ptr to SVCMASTERXPRT */
85 	struct rdma_svc_data rd_data;	/* rdma data */
86 	rdma_mod_t	*r_mod;		/* RDMA module containing ops ptr */
87 };
88 
89 /*
90  * Plugin connection specific data stashed away in clone SVCXPRT
91  */
92 struct clone_rdma_data {
93 	bool_t		cloned;		/* xprt cloned for thread processing */
94 	CONN		*conn;		/* RDMA connection */
95 	rdma_buf_t	rpcbuf;		/* RPC req/resp buffer */
96 	struct clist	*cl_reply;	/* reply chunk buffer info */
97 	struct clist	*cl_wlist;		/* write list clist */
98 };
99 
100 
101 #define	MAXADDRLEN	128	/* max length for address mask */
102 
103 /*
104  * Routines exported through ops vector.
105  */
106 static bool_t		svc_rdma_krecv(SVCXPRT *, mblk_t *, struct rpc_msg *);
107 static bool_t		svc_rdma_ksend(SVCXPRT *, struct rpc_msg *);
108 static bool_t		svc_rdma_kgetargs(SVCXPRT *, xdrproc_t, caddr_t);
109 static bool_t		svc_rdma_kfreeargs(SVCXPRT *, xdrproc_t, caddr_t);
110 void			svc_rdma_kdestroy(SVCMASTERXPRT *);
111 static int		svc_rdma_kdup(struct svc_req *, caddr_t, int,
112 				struct dupreq **, bool_t *);
113 static void		svc_rdma_kdupdone(struct dupreq *, caddr_t,
114 				void (*)(), int, int);
115 static int32_t		*svc_rdma_kgetres(SVCXPRT *, int);
116 static void		svc_rdma_kfreeres(SVCXPRT *);
117 static void		svc_rdma_kclone_destroy(SVCXPRT *);
118 static void		svc_rdma_kstart(SVCMASTERXPRT *);
119 void			svc_rdma_kstop(SVCMASTERXPRT *);
120 static void		svc_rdma_kclone_xprt(SVCXPRT *, SVCXPRT *);
121 static void		svc_rdma_ktattrs(SVCXPRT *, int, void **);
122 
123 static int	svc_process_long_reply(SVCXPRT *, xdrproc_t,
124 			caddr_t, struct rpc_msg *, bool_t, int *,
125 			int *, int *, unsigned int *);
126 
127 static int	svc_compose_rpcmsg(SVCXPRT *, CONN *, xdrproc_t,
128 			caddr_t, rdma_buf_t *, XDR **, struct rpc_msg *,
129 			bool_t, uint_t *);
130 static bool_t rpcmsg_length(xdrproc_t,
131 		caddr_t,
132 		struct rpc_msg *, bool_t, int);
133 
134 /*
135  * Server transport operations vector.
136  */
137 struct svc_ops rdma_svc_ops = {
138 	svc_rdma_krecv,		/* Get requests */
139 	svc_rdma_kgetargs,	/* Deserialize arguments */
140 	svc_rdma_ksend,		/* Send reply */
141 	svc_rdma_kfreeargs,	/* Free argument data space */
142 	svc_rdma_kdestroy,	/* Destroy transport handle */
143 	svc_rdma_kdup,		/* Check entry in dup req cache */
144 	svc_rdma_kdupdone,	/* Mark entry in dup req cache as done */
145 	svc_rdma_kgetres,	/* Get pointer to response buffer */
146 	svc_rdma_kfreeres,	/* Destroy pre-serialized response header */
147 	svc_rdma_kclone_destroy,	/* Destroy a clone xprt */
148 	svc_rdma_kstart,	/* Tell `ready-to-receive' to rpcmod */
149 	svc_rdma_kclone_xprt,	/* Transport specific clone xprt */
150 	svc_rdma_ktattrs	/* Get Transport Attributes */
151 };
152 
153 /*
154  * Server statistics
155  * NOTE: This structure type is duplicated in the NFS fast path.
156  */
157 struct {
158 	kstat_named_t	rscalls;
159 	kstat_named_t	rsbadcalls;
160 	kstat_named_t	rsnullrecv;
161 	kstat_named_t	rsbadlen;
162 	kstat_named_t	rsxdrcall;
163 	kstat_named_t	rsdupchecks;
164 	kstat_named_t	rsdupreqs;
165 	kstat_named_t	rslongrpcs;
166 	kstat_named_t	rstotalreplies;
167 	kstat_named_t	rstotallongreplies;
168 	kstat_named_t	rstotalinlinereplies;
169 } rdmarsstat = {
170 	{ "calls",	KSTAT_DATA_UINT64 },
171 	{ "badcalls",	KSTAT_DATA_UINT64 },
172 	{ "nullrecv",	KSTAT_DATA_UINT64 },
173 	{ "badlen",	KSTAT_DATA_UINT64 },
174 	{ "xdrcall",	KSTAT_DATA_UINT64 },
175 	{ "dupchecks",	KSTAT_DATA_UINT64 },
176 	{ "dupreqs",	KSTAT_DATA_UINT64 },
177 	{ "longrpcs",	KSTAT_DATA_UINT64 },
178 	{ "totalreplies",	KSTAT_DATA_UINT64 },
179 	{ "totallongreplies",	KSTAT_DATA_UINT64 },
180 	{ "totalinlinereplies",	KSTAT_DATA_UINT64 },
181 };
182 
183 kstat_named_t *rdmarsstat_ptr = (kstat_named_t *)&rdmarsstat;
184 uint_t rdmarsstat_ndata = sizeof (rdmarsstat) / sizeof (kstat_named_t);
185 
186 #define	RSSTAT_INCR(x)	atomic_inc_64(&rdmarsstat.x.value.ui64)
187 /*
188  * Create a transport record.
189  * The transport record, output buffer, and private data structure
190  * are allocated.  The output buffer is serialized into using xdrmem.
191  * There is one transport record per user process which implements a
192  * set of services.
193  */
194 /* ARGSUSED */
195 int
196 svc_rdma_kcreate(char *netid, SVC_CALLOUT_TABLE *sct, int id,
197     rdma_xprt_group_t *started_xprts)
198 {
199 	int error;
200 	SVCMASTERXPRT *xprt;
201 	struct rdma_data *rd;
202 	rdma_registry_t *rmod;
203 	rdma_xprt_record_t *xprt_rec;
204 	queue_t	*q;
205 	/*
206 	 * modload the RDMA plugins is not already done.
207 	 */
208 	if (!rdma_modloaded) {
209 		/*CONSTANTCONDITION*/
210 		ASSERT(sizeof (struct clone_rdma_data) <= SVC_P2LEN);
211 
212 		mutex_enter(&rdma_modload_lock);
213 		if (!rdma_modloaded) {
214 			error = rdma_modload();
215 		}
216 		mutex_exit(&rdma_modload_lock);
217 
218 		if (error)
219 			return (error);
220 	}
221 
222 	/*
223 	 * master_xprt_count is the count of master transport handles
224 	 * that were successfully created and are ready to recieve for
225 	 * RDMA based access.
226 	 */
227 	error = 0;
228 	xprt_rec = NULL;
229 	rw_enter(&rdma_lock, RW_READER);
230 	if (rdma_mod_head == NULL) {
231 		started_xprts->rtg_count = 0;
232 		rw_exit(&rdma_lock);
233 		if (rdma_dev_available)
234 			return (EPROTONOSUPPORT);
235 		else
236 			return (ENODEV);
237 	}
238 
239 	/*
240 	 * If we have reached here, then atleast one RDMA plugin has loaded.
241 	 * Create a master_xprt, make it start listenining on the device,
242 	 * if an error is generated, record it, we might need to shut
243 	 * the master_xprt.
244 	 * SVC_START() calls svc_rdma_kstart which calls plugin binding
245 	 * routines.
246 	 */
247 	for (rmod = rdma_mod_head; rmod != NULL; rmod = rmod->r_next) {
248 
249 		/*
250 		 * One SVCMASTERXPRT per RDMA plugin.
251 		 */
252 		xprt = kmem_zalloc(sizeof (*xprt), KM_SLEEP);
253 		xprt->xp_ops = &rdma_svc_ops;
254 		xprt->xp_sct = sct;
255 		xprt->xp_type = T_RDMA;
256 		mutex_init(&xprt->xp_req_lock, NULL, MUTEX_DEFAULT, NULL);
257 		mutex_init(&xprt->xp_thread_lock, NULL, MUTEX_DEFAULT, NULL);
258 		xprt->xp_req_head = (mblk_t *)0;
259 		xprt->xp_req_tail = (mblk_t *)0;
260 		xprt->xp_full = FALSE;
261 		xprt->xp_enable = FALSE;
262 		xprt->xp_reqs = 0;
263 		xprt->xp_size = 0;
264 		xprt->xp_threads = 0;
265 		xprt->xp_detached_threads = 0;
266 
267 		rd = kmem_zalloc(sizeof (*rd), KM_SLEEP);
268 		xprt->xp_p2 = (caddr_t)rd;
269 		rd->rd_xprt = xprt;
270 		rd->r_mod = rmod->r_mod;
271 
272 		q = &rd->rd_data.q;
273 		xprt->xp_wq = q;
274 		q->q_ptr = &rd->rd_xprt;
275 		xprt->xp_netid = NULL;
276 
277 		/*
278 		 * Each of the plugins will have their own Service ID
279 		 * to listener specific mapping, like port number for VI
280 		 * and service name for IB.
281 		 */
282 		rd->rd_data.svcid = id;
283 		error = svc_xprt_register(xprt, id);
284 		if (error) {
285 			DTRACE_PROBE(krpc__e__svcrdma__xprt__reg);
286 			goto cleanup;
287 		}
288 
289 		SVC_START(xprt);
290 		if (!rd->rd_data.active) {
291 			svc_xprt_unregister(xprt);
292 			error = rd->rd_data.err_code;
293 			goto cleanup;
294 		}
295 
296 		/*
297 		 * This is set only when there is atleast one or more
298 		 * transports successfully created. We insert the pointer
299 		 * to the created RDMA master xprt into a separately maintained
300 		 * list. This way we can easily reference it later to cleanup,
301 		 * when NFS kRPC service pool is going away/unregistered.
302 		 */
303 		started_xprts->rtg_count ++;
304 		xprt_rec = kmem_alloc(sizeof (*xprt_rec), KM_SLEEP);
305 		xprt_rec->rtr_xprt_ptr = xprt;
306 		xprt_rec->rtr_next = started_xprts->rtg_listhead;
307 		started_xprts->rtg_listhead = xprt_rec;
308 		continue;
309 cleanup:
310 		SVC_DESTROY(xprt);
311 		if (error == RDMA_FAILED)
312 			error = EPROTONOSUPPORT;
313 	}
314 
315 	rw_exit(&rdma_lock);
316 
317 	/*
318 	 * Don't return any error even if a single plugin was started
319 	 * successfully.
320 	 */
321 	if (started_xprts->rtg_count == 0)
322 		return (error);
323 	return (0);
324 }
325 
326 /*
327  * Cleanup routine for freeing up memory allocated by
328  * svc_rdma_kcreate()
329  */
330 void
331 svc_rdma_kdestroy(SVCMASTERXPRT *xprt)
332 {
333 	struct rdma_data *rd = (struct rdma_data *)xprt->xp_p2;
334 
335 
336 	mutex_destroy(&xprt->xp_req_lock);
337 	mutex_destroy(&xprt->xp_thread_lock);
338 	kmem_free(rd, sizeof (*rd));
339 	kmem_free(xprt, sizeof (*xprt));
340 }
341 
342 
343 static void
344 svc_rdma_kstart(SVCMASTERXPRT *xprt)
345 {
346 	struct rdma_svc_data *svcdata;
347 	rdma_mod_t *rmod;
348 
349 	svcdata = &((struct rdma_data *)xprt->xp_p2)->rd_data;
350 	rmod = ((struct rdma_data *)xprt->xp_p2)->r_mod;
351 
352 	/*
353 	 * Create a listener for  module at this port
354 	 */
355 
356 	if (rmod->rdma_count != 0)
357 		(*rmod->rdma_ops->rdma_svc_listen)(svcdata);
358 	else
359 		svcdata->err_code = RDMA_FAILED;
360 }
361 
362 void
363 svc_rdma_kstop(SVCMASTERXPRT *xprt)
364 {
365 	struct rdma_svc_data *svcdata;
366 	rdma_mod_t *rmod;
367 
368 	svcdata	= &((struct rdma_data *)xprt->xp_p2)->rd_data;
369 	rmod = ((struct rdma_data *)xprt->xp_p2)->r_mod;
370 
371 	/*
372 	 * Call the stop listener routine for each plugin. If rdma_count is
373 	 * already zero set active to zero.
374 	 */
375 	if (rmod->rdma_count != 0)
376 		(*rmod->rdma_ops->rdma_svc_stop)(svcdata);
377 	else
378 		svcdata->active = 0;
379 	if (svcdata->active)
380 		DTRACE_PROBE(krpc__e__svcrdma__kstop);
381 }
382 
383 /* ARGSUSED */
384 static void
385 svc_rdma_kclone_destroy(SVCXPRT *clone_xprt)
386 {
387 
388 	struct clone_rdma_data *cdrp;
389 	cdrp = (struct clone_rdma_data *)clone_xprt->xp_p2buf;
390 
391 	/*
392 	 * Only free buffers and release connection when cloned is set.
393 	 */
394 	if (cdrp->cloned != TRUE)
395 		return;
396 
397 	rdma_buf_free(cdrp->conn, &cdrp->rpcbuf);
398 	if (cdrp->cl_reply) {
399 		clist_free(cdrp->cl_reply);
400 		cdrp->cl_reply = NULL;
401 	}
402 	RDMA_REL_CONN(cdrp->conn);
403 
404 	cdrp->cloned = 0;
405 }
406 
407 /*
408  * Clone the xprt specific information.  It will be freed by
409  * SVC_CLONE_DESTROY.
410  */
411 static void
412 svc_rdma_kclone_xprt(SVCXPRT *src_xprt, SVCXPRT *dst_xprt)
413 {
414 	struct clone_rdma_data *srcp2;
415 	struct clone_rdma_data *dstp2;
416 
417 	srcp2 = (struct clone_rdma_data *)src_xprt->xp_p2buf;
418 	dstp2 = (struct clone_rdma_data *)dst_xprt->xp_p2buf;
419 
420 	if (srcp2->conn != NULL) {
421 		srcp2->cloned = TRUE;
422 		*dstp2 = *srcp2;
423 	}
424 }
425 
426 static void
427 svc_rdma_ktattrs(SVCXPRT *clone_xprt, int attrflag, void **tattr)
428 {
429 	CONN	*conn;
430 	*tattr = NULL;
431 
432 	switch (attrflag) {
433 	case SVC_TATTR_ADDRMASK:
434 		conn = ((struct clone_rdma_data *)clone_xprt->xp_p2buf)->conn;
435 		ASSERT(conn != NULL);
436 		if (conn)
437 			*tattr = (void *)&conn->c_addrmask;
438 	}
439 }
440 
441 static bool_t
442 svc_rdma_krecv(SVCXPRT *clone_xprt, mblk_t *mp, struct rpc_msg *msg)
443 {
444 	XDR	*xdrs;
445 	CONN	*conn;
446 	rdma_recv_data_t	*rdp = (rdma_recv_data_t *)mp->b_rptr;
447 	struct clone_rdma_data *crdp;
448 	struct clist	*cl = NULL;
449 	struct clist	*wcl = NULL;
450 	struct clist	*cllong = NULL;
451 
452 	rdma_stat	status;
453 	uint32_t vers, op, pos, xid;
454 	uint32_t rdma_credit;
455 	uint32_t wcl_total_length = 0;
456 	bool_t	wwl = FALSE;
457 
458 	crdp = (struct clone_rdma_data *)clone_xprt->xp_p2buf;
459 	RSSTAT_INCR(rscalls);
460 	conn = rdp->conn;
461 
462 	status = rdma_svc_postrecv(conn);
463 	if (status != RDMA_SUCCESS) {
464 		DTRACE_PROBE(krpc__e__svcrdma__krecv__postrecv);
465 		goto badrpc_call;
466 	}
467 
468 	xdrs = &clone_xprt->xp_xdrin;
469 	xdrmem_create(xdrs, rdp->rpcmsg.addr, rdp->rpcmsg.len, XDR_DECODE);
470 	xid = *(uint32_t *)rdp->rpcmsg.addr;
471 	XDR_SETPOS(xdrs, sizeof (uint32_t));
472 
473 	if (! xdr_u_int(xdrs, &vers) ||
474 	    ! xdr_u_int(xdrs, &rdma_credit) ||
475 	    ! xdr_u_int(xdrs, &op)) {
476 		DTRACE_PROBE(krpc__e__svcrdma__krecv__uint);
477 		goto xdr_err;
478 	}
479 
480 	/* Checking if the status of the recv operation was normal */
481 	if (rdp->status != 0) {
482 		DTRACE_PROBE1(krpc__e__svcrdma__krecv__invalid__status,
483 		    int, rdp->status);
484 		goto badrpc_call;
485 	}
486 
487 	if (! xdr_do_clist(xdrs, &cl)) {
488 		DTRACE_PROBE(krpc__e__svcrdma__krecv__do__clist);
489 		goto xdr_err;
490 	}
491 
492 	if (!xdr_decode_wlist_svc(xdrs, &wcl, &wwl, &wcl_total_length, conn)) {
493 		DTRACE_PROBE(krpc__e__svcrdma__krecv__decode__wlist);
494 		if (cl)
495 			clist_free(cl);
496 		goto xdr_err;
497 	}
498 	crdp->cl_wlist = wcl;
499 
500 	crdp->cl_reply = NULL;
501 	(void) xdr_decode_reply_wchunk(xdrs, &crdp->cl_reply);
502 
503 	/*
504 	 * A chunk at 0 offset indicates that the RPC call message
505 	 * is in a chunk. Get the RPC call message chunk.
506 	 */
507 	if (cl != NULL && op == RDMA_NOMSG) {
508 
509 		/* Remove RPC call message chunk from chunklist */
510 		cllong = cl;
511 		cl = cl->c_next;
512 		cllong->c_next = NULL;
513 
514 
515 		/* Allocate and register memory for the RPC call msg chunk */
516 		cllong->rb_longbuf.type = RDMA_LONG_BUFFER;
517 		cllong->rb_longbuf.len = cllong->c_len > LONG_REPLY_LEN ?
518 		    cllong->c_len : LONG_REPLY_LEN;
519 
520 		if (rdma_buf_alloc(conn, &cllong->rb_longbuf)) {
521 			clist_free(cllong);
522 			goto cll_malloc_err;
523 		}
524 
525 		cllong->u.c_daddr3 = cllong->rb_longbuf.addr;
526 
527 		if (cllong->u.c_daddr == NULL) {
528 			DTRACE_PROBE(krpc__e__svcrdma__krecv__nomem);
529 			rdma_buf_free(conn, &cllong->rb_longbuf);
530 			clist_free(cllong);
531 			goto cll_malloc_err;
532 		}
533 
534 		status = clist_register(conn, cllong, CLIST_REG_DST);
535 		if (status) {
536 			DTRACE_PROBE(krpc__e__svcrdma__krecv__clist__reg);
537 			rdma_buf_free(conn, &cllong->rb_longbuf);
538 			clist_free(cllong);
539 			goto cll_malloc_err;
540 		}
541 
542 		/*
543 		 * Now read the RPC call message in
544 		 */
545 		status = RDMA_READ(conn, cllong, WAIT);
546 		if (status) {
547 			DTRACE_PROBE(krpc__e__svcrdma__krecv__read);
548 			(void) clist_deregister(conn, cllong);
549 			rdma_buf_free(conn, &cllong->rb_longbuf);
550 			clist_free(cllong);
551 			goto cll_malloc_err;
552 		}
553 
554 		status = clist_syncmem(conn, cllong, CLIST_REG_DST);
555 		(void) clist_deregister(conn, cllong);
556 
557 		xdrrdma_create(xdrs, (caddr_t)(uintptr_t)cllong->u.c_daddr3,
558 		    cllong->c_len, 0, cl, XDR_DECODE, conn);
559 
560 		crdp->rpcbuf = cllong->rb_longbuf;
561 		crdp->rpcbuf.len = cllong->c_len;
562 		clist_free(cllong);
563 		RDMA_BUF_FREE(conn, &rdp->rpcmsg);
564 	} else {
565 		pos = XDR_GETPOS(xdrs);
566 		xdrrdma_create(xdrs, rdp->rpcmsg.addr + pos,
567 		    rdp->rpcmsg.len - pos, 0, cl, XDR_DECODE, conn);
568 		crdp->rpcbuf = rdp->rpcmsg;
569 
570 		/* Use xdrrdmablk_ops to indicate there is a read chunk list */
571 		if (cl != NULL) {
572 			int32_t flg = XDR_RDMA_RLIST_REG;
573 
574 			XDR_CONTROL(xdrs, XDR_RDMA_SET_FLAGS, &flg);
575 			xdrs->x_ops = &xdrrdmablk_ops;
576 		}
577 	}
578 
579 	if (crdp->cl_wlist) {
580 		int32_t flg = XDR_RDMA_WLIST_REG;
581 
582 		XDR_CONTROL(xdrs, XDR_RDMA_SET_WLIST, crdp->cl_wlist);
583 		XDR_CONTROL(xdrs, XDR_RDMA_SET_FLAGS, &flg);
584 	}
585 
586 	if (! xdr_callmsg(xdrs, msg)) {
587 		DTRACE_PROBE(krpc__e__svcrdma__krecv__callmsg);
588 		RSSTAT_INCR(rsxdrcall);
589 		goto callmsg_err;
590 	}
591 
592 	/*
593 	 * Point the remote transport address in the service_transport
594 	 * handle at the address in the request.
595 	 */
596 	clone_xprt->xp_rtaddr.buf = conn->c_raddr.buf;
597 	clone_xprt->xp_rtaddr.len = conn->c_raddr.len;
598 	clone_xprt->xp_rtaddr.maxlen = conn->c_raddr.len;
599 
600 	clone_xprt->xp_lcladdr.buf = conn->c_laddr.buf;
601 	clone_xprt->xp_lcladdr.len = conn->c_laddr.len;
602 	clone_xprt->xp_lcladdr.maxlen = conn->c_laddr.len;
603 
604 	/*
605 	 * In case of RDMA, connection management is
606 	 * entirely done in rpcib module and netid in the
607 	 * SVCMASTERXPRT is NULL. Initialize the clone netid
608 	 * from the connection.
609 	 */
610 
611 	clone_xprt->xp_netid = conn->c_netid;
612 
613 	clone_xprt->xp_xid = xid;
614 	crdp->conn = conn;
615 
616 	freeb(mp);
617 
618 	return (TRUE);
619 
620 callmsg_err:
621 	rdma_buf_free(conn, &crdp->rpcbuf);
622 
623 cll_malloc_err:
624 	if (cl)
625 		clist_free(cl);
626 xdr_err:
627 	XDR_DESTROY(xdrs);
628 
629 badrpc_call:
630 	RDMA_BUF_FREE(conn, &rdp->rpcmsg);
631 	RDMA_REL_CONN(conn);
632 	freeb(mp);
633 	RSSTAT_INCR(rsbadcalls);
634 	return (FALSE);
635 }
636 
637 static int
638 svc_process_long_reply(SVCXPRT * clone_xprt,
639     xdrproc_t xdr_results, caddr_t xdr_location,
640     struct rpc_msg *msg, bool_t has_args, int *msglen,
641     int *freelen, int *numchunks, unsigned int *final_len)
642 {
643 	int status;
644 	XDR xdrslong;
645 	struct clist *wcl = NULL;
646 	int count = 0;
647 	int alloc_len;
648 	char  *memp;
649 	rdma_buf_t long_rpc = {0};
650 	struct clone_rdma_data *crdp;
651 
652 	crdp = (struct clone_rdma_data *)clone_xprt->xp_p2buf;
653 
654 	bzero(&xdrslong, sizeof (xdrslong));
655 
656 	/* Choose a size for the long rpc response */
657 	if (MSG_IS_RPCSEC_GSS(msg)) {
658 		alloc_len = RNDUP(MAX_AUTH_BYTES + *msglen);
659 	} else {
660 		alloc_len = RNDUP(*msglen);
661 	}
662 
663 	if (alloc_len <= 64 * 1024) {
664 		if (alloc_len > 32 * 1024) {
665 			alloc_len = 64 * 1024;
666 		} else {
667 			if (alloc_len > 16 * 1024) {
668 				alloc_len = 32 * 1024;
669 			} else {
670 				alloc_len = 16 * 1024;
671 			}
672 		}
673 	}
674 
675 	long_rpc.type = RDMA_LONG_BUFFER;
676 	long_rpc.len = alloc_len;
677 	if (rdma_buf_alloc(crdp->conn, &long_rpc)) {
678 		return (SVC_RDMA_FAIL);
679 	}
680 
681 	memp = long_rpc.addr;
682 	xdrmem_create(&xdrslong, memp, alloc_len, XDR_ENCODE);
683 
684 	msg->rm_xid = clone_xprt->xp_xid;
685 
686 	if (!(xdr_replymsg(&xdrslong, msg) &&
687 	    (!has_args || SVCAUTH_WRAP(&clone_xprt->xp_auth, &xdrslong,
688 	    xdr_results, xdr_location)))) {
689 		rdma_buf_free(crdp->conn, &long_rpc);
690 		DTRACE_PROBE(krpc__e__svcrdma__longrep__authwrap);
691 		return (SVC_RDMA_FAIL);
692 	}
693 
694 	*final_len = XDR_GETPOS(&xdrslong);
695 
696 	DTRACE_PROBE1(krpc__i__replylen, uint_t, *final_len);
697 	*numchunks = 0;
698 	*freelen = 0;
699 
700 	wcl = crdp->cl_reply;
701 	wcl->rb_longbuf = long_rpc;
702 
703 	count = *final_len;
704 	while ((wcl != NULL) && (count > 0)) {
705 
706 		if (wcl->c_dmemhandle.mrc_rmr == 0)
707 			break;
708 
709 		DTRACE_PROBE2(krpc__i__write__chunks, uint32_t, count,
710 		    uint32_t, wcl->c_len);
711 
712 		if (wcl->c_len > count) {
713 			wcl->c_len = count;
714 		}
715 		wcl->w.c_saddr3 = (caddr_t)memp;
716 
717 		count -= wcl->c_len;
718 		*numchunks +=  1;
719 		memp += wcl->c_len;
720 		wcl = wcl->c_next;
721 	}
722 
723 	/*
724 	 * Make rest of the chunks 0-len
725 	 */
726 	while (wcl != NULL) {
727 		if (wcl->c_dmemhandle.mrc_rmr == 0)
728 			break;
729 		wcl->c_len = 0;
730 		wcl = wcl->c_next;
731 	}
732 
733 	wcl = crdp->cl_reply;
734 
735 	/*
736 	 * MUST fail if there are still more data
737 	 */
738 	if (count > 0) {
739 		rdma_buf_free(crdp->conn, &long_rpc);
740 		DTRACE_PROBE(krpc__e__svcrdma__longrep__dlen__clist);
741 		return (SVC_RDMA_FAIL);
742 	}
743 
744 	if (clist_register(crdp->conn, wcl, CLIST_REG_SOURCE) != RDMA_SUCCESS) {
745 		rdma_buf_free(crdp->conn, &long_rpc);
746 		DTRACE_PROBE(krpc__e__svcrdma__longrep__clistreg);
747 		return (SVC_RDMA_FAIL);
748 	}
749 
750 	status = clist_syncmem(crdp->conn, wcl, CLIST_REG_SOURCE);
751 
752 	if (status) {
753 		(void) clist_deregister(crdp->conn, wcl);
754 		rdma_buf_free(crdp->conn, &long_rpc);
755 		DTRACE_PROBE(krpc__e__svcrdma__longrep__syncmem);
756 		return (SVC_RDMA_FAIL);
757 	}
758 
759 	status = RDMA_WRITE(crdp->conn, wcl, WAIT);
760 
761 	(void) clist_deregister(crdp->conn, wcl);
762 	rdma_buf_free(crdp->conn, &wcl->rb_longbuf);
763 
764 	if (status != RDMA_SUCCESS) {
765 		DTRACE_PROBE(krpc__e__svcrdma__longrep__write);
766 		return (SVC_RDMA_FAIL);
767 	}
768 
769 	return (SVC_RDMA_SUCCESS);
770 }
771 
772 
773 static int
774 svc_compose_rpcmsg(SVCXPRT * clone_xprt, CONN * conn, xdrproc_t xdr_results,
775     caddr_t xdr_location, rdma_buf_t *rpcreply, XDR ** xdrs,
776     struct rpc_msg *msg, bool_t has_args, uint_t *len)
777 {
778 	/*
779 	 * Get a pre-allocated buffer for rpc reply
780 	 */
781 	rpcreply->type = SEND_BUFFER;
782 	if (rdma_buf_alloc(conn, rpcreply)) {
783 		DTRACE_PROBE(krpc__e__svcrdma__rpcmsg__reply__nofreebufs);
784 		return (SVC_RDMA_FAIL);
785 	}
786 
787 	xdrrdma_create(*xdrs, rpcreply->addr, rpcreply->len,
788 	    0, NULL, XDR_ENCODE, conn);
789 
790 	msg->rm_xid = clone_xprt->xp_xid;
791 
792 	if (has_args) {
793 		if (!(xdr_replymsg(*xdrs, msg) &&
794 		    (!has_args ||
795 		    SVCAUTH_WRAP(&clone_xprt->xp_auth, *xdrs,
796 		    xdr_results, xdr_location)))) {
797 			rdma_buf_free(conn, rpcreply);
798 			DTRACE_PROBE(
799 			    krpc__e__svcrdma__rpcmsg__reply__authwrap1);
800 			return (SVC_RDMA_FAIL);
801 		}
802 	} else {
803 		if (!xdr_replymsg(*xdrs, msg)) {
804 			rdma_buf_free(conn, rpcreply);
805 			DTRACE_PROBE(
806 			    krpc__e__svcrdma__rpcmsg__reply__authwrap2);
807 			return (SVC_RDMA_FAIL);
808 		}
809 	}
810 
811 	*len = XDR_GETPOS(*xdrs);
812 
813 	return (SVC_RDMA_SUCCESS);
814 }
815 
816 /*
817  * Send rpc reply.
818  */
819 static bool_t
820 svc_rdma_ksend(SVCXPRT * clone_xprt, struct rpc_msg *msg)
821 {
822 	XDR *xdrs_rpc = &(clone_xprt->xp_xdrout);
823 	XDR xdrs_rhdr;
824 	CONN *conn = NULL;
825 	rdma_buf_t rbuf_resp = {0}, rbuf_rpc_resp = {0};
826 
827 	struct clone_rdma_data *crdp;
828 	struct clist *cl_read = NULL;
829 	struct clist *cl_send = NULL;
830 	struct clist *cl_write = NULL;
831 	xdrproc_t xdr_results;		/* results XDR encoding function */
832 	caddr_t xdr_location;		/* response results pointer */
833 
834 	int retval = FALSE;
835 	int status, msglen, num_wreply_segments = 0;
836 	uint32_t rdma_credit = 0;
837 	int freelen = 0;
838 	bool_t has_args;
839 	uint_t  final_resp_len, rdma_response_op, vers;
840 
841 	bzero(&xdrs_rhdr, sizeof (XDR));
842 	crdp = (struct clone_rdma_data *)clone_xprt->xp_p2buf;
843 	conn = crdp->conn;
844 
845 	/*
846 	 * If there is a result procedure specified in the reply message,
847 	 * it will be processed in the xdr_replymsg and SVCAUTH_WRAP.
848 	 * We need to make sure it won't be processed twice, so we null
849 	 * it for xdr_replymsg here.
850 	 */
851 	has_args = FALSE;
852 	if (msg->rm_reply.rp_stat == MSG_ACCEPTED &&
853 	    msg->rm_reply.rp_acpt.ar_stat == SUCCESS) {
854 		if ((xdr_results = msg->acpted_rply.ar_results.proc) != NULL) {
855 			has_args = TRUE;
856 			xdr_location = msg->acpted_rply.ar_results.where;
857 			msg->acpted_rply.ar_results.proc = xdr_void;
858 			msg->acpted_rply.ar_results.where = NULL;
859 		}
860 	}
861 
862 	/*
863 	 * Given the limit on the inline response size (RPC_MSG_SZ),
864 	 * there is a need to make a guess as to the overall size of
865 	 * the response.  If the resultant size is beyond the inline
866 	 * size, then the server needs to use the "reply chunk list"
867 	 * provided by the client (if the client provided one).  An
868 	 * example of this type of response would be a READDIR
869 	 * response (e.g. a small directory read would fit in RPC_MSG_SZ
870 	 * and that is the preference but it may not fit)
871 	 *
872 	 * Combine the encoded size and the size of the true results
873 	 * and then make the decision about where to encode and send results.
874 	 *
875 	 * One important note, this calculation is ignoring the size
876 	 * of the encoding of the authentication overhead.  The reason
877 	 * for this is rooted in the complexities of access to the
878 	 * encoded size of RPCSEC_GSS related authentiation,
879 	 * integrity, and privacy.
880 	 *
881 	 * If it turns out that the encoded authentication bumps the
882 	 * response over the RPC_MSG_SZ limit, then it may need to
883 	 * attempt to encode for the reply chunk list.
884 	 */
885 
886 	/*
887 	 * Calculating the "sizeof" the RPC response header and the
888 	 * encoded results.
889 	 */
890 	msglen = xdr_sizeof(xdr_replymsg, msg);
891 
892 	if (msglen > 0) {
893 		RSSTAT_INCR(rstotalreplies);
894 	}
895 	if (has_args)
896 		msglen += xdrrdma_sizeof(xdr_results, xdr_location,
897 		    rdma_minchunk, NULL, NULL);
898 
899 	DTRACE_PROBE1(krpc__i__svcrdma__ksend__msglen, int, msglen);
900 
901 	status = SVC_RDMA_SUCCESS;
902 
903 	if (msglen < RPC_MSG_SZ) {
904 		/*
905 		 * Looks like the response will fit in the inline
906 		 * response; let's try
907 		 */
908 		RSSTAT_INCR(rstotalinlinereplies);
909 
910 		rdma_response_op = RDMA_MSG;
911 
912 		status = svc_compose_rpcmsg(clone_xprt, conn, xdr_results,
913 		    xdr_location, &rbuf_rpc_resp, &xdrs_rpc, msg,
914 		    has_args, &final_resp_len);
915 
916 		DTRACE_PROBE1(krpc__i__srdma__ksend__compose_status,
917 		    int, status);
918 		DTRACE_PROBE1(krpc__i__srdma__ksend__compose_len,
919 		    int, final_resp_len);
920 
921 		if (status == SVC_RDMA_SUCCESS && crdp->cl_reply) {
922 			clist_free(crdp->cl_reply);
923 			crdp->cl_reply = NULL;
924 		}
925 	}
926 
927 	/*
928 	 * If the encode failed (size?) or the message really is
929 	 * larger than what is allowed, try the response chunk list.
930 	 */
931 	if (status != SVC_RDMA_SUCCESS || msglen >= RPC_MSG_SZ) {
932 		/*
933 		 * attempting to use a reply chunk list when there
934 		 * isn't one won't get very far...
935 		 */
936 		if (crdp->cl_reply == NULL) {
937 			DTRACE_PROBE(krpc__e__svcrdma__ksend__noreplycl);
938 			goto out;
939 		}
940 
941 		RSSTAT_INCR(rstotallongreplies);
942 
943 		msglen = xdr_sizeof(xdr_replymsg, msg);
944 		msglen += xdrrdma_sizeof(xdr_results, xdr_location, 0,
945 		    NULL, NULL);
946 
947 		status = svc_process_long_reply(clone_xprt, xdr_results,
948 		    xdr_location, msg, has_args, &msglen, &freelen,
949 		    &num_wreply_segments, &final_resp_len);
950 
951 		DTRACE_PROBE1(krpc__i__svcrdma__ksend__longreplen,
952 		    int, final_resp_len);
953 
954 		if (status != SVC_RDMA_SUCCESS) {
955 			DTRACE_PROBE(krpc__e__svcrdma__ksend__compose__failed);
956 			goto out;
957 		}
958 
959 		rdma_response_op = RDMA_NOMSG;
960 	}
961 
962 	DTRACE_PROBE1(krpc__i__svcrdma__ksend__rdmamsg__len,
963 	    int, final_resp_len);
964 
965 	rbuf_resp.type = SEND_BUFFER;
966 	if (rdma_buf_alloc(conn, &rbuf_resp)) {
967 		rdma_buf_free(conn, &rbuf_rpc_resp);
968 		DTRACE_PROBE(krpc__e__svcrdma__ksend__nofreebufs);
969 		goto out;
970 	}
971 
972 	rdma_credit = rdma_bufs_granted;
973 
974 	vers = RPCRDMA_VERS;
975 	xdrmem_create(&xdrs_rhdr, rbuf_resp.addr, rbuf_resp.len, XDR_ENCODE);
976 	(*(uint32_t *)rbuf_resp.addr) = msg->rm_xid;
977 	/* Skip xid and set the xdr position accordingly. */
978 	XDR_SETPOS(&xdrs_rhdr, sizeof (uint32_t));
979 	if (!xdr_u_int(&xdrs_rhdr, &vers) ||
980 	    !xdr_u_int(&xdrs_rhdr, &rdma_credit) ||
981 	    !xdr_u_int(&xdrs_rhdr, &rdma_response_op)) {
982 		rdma_buf_free(conn, &rbuf_rpc_resp);
983 		rdma_buf_free(conn, &rbuf_resp);
984 		DTRACE_PROBE(krpc__e__svcrdma__ksend__uint);
985 		goto out;
986 	}
987 
988 	/*
989 	 * Now XDR the read chunk list, actually always NULL
990 	 */
991 	(void) xdr_encode_rlist_svc(&xdrs_rhdr, cl_read);
992 
993 	/*
994 	 * encode write list -- we already drove RDMA_WRITEs
995 	 */
996 	cl_write = crdp->cl_wlist;
997 	if (!xdr_encode_wlist(&xdrs_rhdr, cl_write)) {
998 		DTRACE_PROBE(krpc__e__svcrdma__ksend__enc__wlist);
999 		rdma_buf_free(conn, &rbuf_rpc_resp);
1000 		rdma_buf_free(conn, &rbuf_resp);
1001 		goto out;
1002 	}
1003 
1004 	/*
1005 	 * XDR encode the RDMA_REPLY write chunk
1006 	 */
1007 	if (!xdr_encode_reply_wchunk(&xdrs_rhdr, crdp->cl_reply,
1008 	    num_wreply_segments)) {
1009 		rdma_buf_free(conn, &rbuf_rpc_resp);
1010 		rdma_buf_free(conn, &rbuf_resp);
1011 		goto out;
1012 	}
1013 
1014 	clist_add(&cl_send, 0, XDR_GETPOS(&xdrs_rhdr), &rbuf_resp.handle,
1015 	    rbuf_resp.addr, NULL, NULL);
1016 
1017 	if (rdma_response_op == RDMA_MSG) {
1018 		clist_add(&cl_send, 0, final_resp_len, &rbuf_rpc_resp.handle,
1019 		    rbuf_rpc_resp.addr, NULL, NULL);
1020 	}
1021 
1022 	status = RDMA_SEND(conn, cl_send, msg->rm_xid);
1023 
1024 	if (status == RDMA_SUCCESS) {
1025 		retval = TRUE;
1026 	}
1027 
1028 out:
1029 	/*
1030 	 * Free up sendlist chunks
1031 	 */
1032 	if (cl_send != NULL)
1033 		clist_free(cl_send);
1034 
1035 	/*
1036 	 * Destroy private data for xdr rdma
1037 	 */
1038 	if (clone_xprt->xp_xdrout.x_ops != NULL) {
1039 		XDR_DESTROY(&(clone_xprt->xp_xdrout));
1040 	}
1041 
1042 	if (crdp->cl_reply) {
1043 		clist_free(crdp->cl_reply);
1044 		crdp->cl_reply = NULL;
1045 	}
1046 
1047 	/*
1048 	 * This is completely disgusting.  If public is set it is
1049 	 * a pointer to a structure whose first field is the address
1050 	 * of the function to free that structure and any related
1051 	 * stuff.  (see rrokfree in nfs_xdr.c).
1052 	 */
1053 	if (xdrs_rpc->x_public) {
1054 		/* LINTED pointer alignment */
1055 		(**((int (**)()) xdrs_rpc->x_public)) (xdrs_rpc->x_public);
1056 	}
1057 
1058 	if (xdrs_rhdr.x_ops != NULL) {
1059 		XDR_DESTROY(&xdrs_rhdr);
1060 	}
1061 
1062 	return (retval);
1063 }
1064 
1065 /*
1066  * Deserialize arguments.
1067  */
1068 static bool_t
1069 svc_rdma_kgetargs(SVCXPRT *clone_xprt, xdrproc_t xdr_args, caddr_t args_ptr)
1070 {
1071 	if ((SVCAUTH_UNWRAP(&clone_xprt->xp_auth, &clone_xprt->xp_xdrin,
1072 	    xdr_args, args_ptr)) != TRUE)
1073 		return (FALSE);
1074 	return (TRUE);
1075 }
1076 
1077 static bool_t
1078 svc_rdma_kfreeargs(SVCXPRT *clone_xprt, xdrproc_t xdr_args,
1079     caddr_t args_ptr)
1080 {
1081 	struct clone_rdma_data *crdp;
1082 	bool_t retval;
1083 
1084 	/*
1085 	 * If the cloned bit is true, then this transport specific
1086 	 * rmda data has been duplicated into another cloned xprt. Do
1087 	 * not free, or release the connection, it is still in use.  The
1088 	 * buffers will be freed and the connection released later by
1089 	 * SVC_CLONE_DESTROY().
1090 	 */
1091 	crdp = (struct clone_rdma_data *)clone_xprt->xp_p2buf;
1092 	if (crdp->cloned == TRUE) {
1093 		crdp->cloned = 0;
1094 		return (TRUE);
1095 	}
1096 
1097 	/*
1098 	 * Free the args if needed then XDR_DESTROY
1099 	 */
1100 	if (args_ptr) {
1101 		XDR	*xdrs = &clone_xprt->xp_xdrin;
1102 
1103 		xdrs->x_op = XDR_FREE;
1104 		retval = (*xdr_args)(xdrs, args_ptr);
1105 	}
1106 
1107 	XDR_DESTROY(&(clone_xprt->xp_xdrin));
1108 	rdma_buf_free(crdp->conn, &crdp->rpcbuf);
1109 	if (crdp->cl_reply) {
1110 		clist_free(crdp->cl_reply);
1111 		crdp->cl_reply = NULL;
1112 	}
1113 	RDMA_REL_CONN(crdp->conn);
1114 
1115 	return (retval);
1116 }
1117 
1118 /* ARGSUSED */
1119 static int32_t *
1120 svc_rdma_kgetres(SVCXPRT *clone_xprt, int size)
1121 {
1122 	return (NULL);
1123 }
1124 
1125 /* ARGSUSED */
1126 static void
1127 svc_rdma_kfreeres(SVCXPRT *clone_xprt)
1128 {
1129 }
1130 
1131 /*
1132  * the dup cacheing routines below provide a cache of non-failure
1133  * transaction id's.  rpc service routines can use this to detect
1134  * retransmissions and re-send a non-failure response.
1135  */
1136 
1137 /*
1138  * MAXDUPREQS is the number of cached items.  It should be adjusted
1139  * to the service load so that there is likely to be a response entry
1140  * when the first retransmission comes in.
1141  */
1142 #define	MAXDUPREQS	1024
1143 
1144 /*
1145  * This should be appropriately scaled to MAXDUPREQS.
1146  */
1147 #define	DRHASHSZ	257
1148 
1149 #if ((DRHASHSZ & (DRHASHSZ - 1)) == 0)
1150 #define	XIDHASH(xid)	((xid) & (DRHASHSZ - 1))
1151 #else
1152 #define	XIDHASH(xid)	((xid) % DRHASHSZ)
1153 #endif
1154 #define	DRHASH(dr)	XIDHASH((dr)->dr_xid)
1155 #define	REQTOXID(req)	((req)->rq_xprt->xp_xid)
1156 
1157 static int	rdmandupreqs = 0;
1158 int	rdmamaxdupreqs = MAXDUPREQS;
1159 static kmutex_t rdmadupreq_lock;
1160 static struct dupreq *rdmadrhashtbl[DRHASHSZ];
1161 static int	rdmadrhashstat[DRHASHSZ];
1162 
1163 static void unhash(struct dupreq *);
1164 
1165 /*
1166  * rdmadrmru points to the head of a circular linked list in lru order.
1167  * rdmadrmru->dr_next == drlru
1168  */
1169 struct dupreq *rdmadrmru;
1170 
1171 /*
1172  * svc_rdma_kdup searches the request cache and returns 0 if the
1173  * request is not found in the cache.  If it is found, then it
1174  * returns the state of the request (in progress or done) and
1175  * the status or attributes that were part of the original reply.
1176  */
1177 static int
1178 svc_rdma_kdup(struct svc_req *req, caddr_t res, int size, struct dupreq **drpp,
1179 	bool_t *dupcachedp)
1180 {
1181 	struct dupreq *dr;
1182 	uint32_t xid;
1183 	uint32_t drhash;
1184 	int status;
1185 
1186 	xid = REQTOXID(req);
1187 	mutex_enter(&rdmadupreq_lock);
1188 	RSSTAT_INCR(rsdupchecks);
1189 	/*
1190 	 * Check to see whether an entry already exists in the cache.
1191 	 */
1192 	dr = rdmadrhashtbl[XIDHASH(xid)];
1193 	while (dr != NULL) {
1194 		if (dr->dr_xid == xid &&
1195 		    dr->dr_proc == req->rq_proc &&
1196 		    dr->dr_prog == req->rq_prog &&
1197 		    dr->dr_vers == req->rq_vers &&
1198 		    dr->dr_addr.len == req->rq_xprt->xp_rtaddr.len &&
1199 		    bcmp((caddr_t)dr->dr_addr.buf,
1200 		    (caddr_t)req->rq_xprt->xp_rtaddr.buf,
1201 		    dr->dr_addr.len) == 0) {
1202 			status = dr->dr_status;
1203 			if (status == DUP_DONE) {
1204 				bcopy(dr->dr_resp.buf, res, size);
1205 				if (dupcachedp != NULL)
1206 					*dupcachedp = (dr->dr_resfree != NULL);
1207 			} else {
1208 				dr->dr_status = DUP_INPROGRESS;
1209 				*drpp = dr;
1210 			}
1211 			RSSTAT_INCR(rsdupreqs);
1212 			mutex_exit(&rdmadupreq_lock);
1213 			return (status);
1214 		}
1215 		dr = dr->dr_chain;
1216 	}
1217 
1218 	/*
1219 	 * There wasn't an entry, either allocate a new one or recycle
1220 	 * an old one.
1221 	 */
1222 	if (rdmandupreqs < rdmamaxdupreqs) {
1223 		dr = kmem_alloc(sizeof (*dr), KM_NOSLEEP);
1224 		if (dr == NULL) {
1225 			mutex_exit(&rdmadupreq_lock);
1226 			return (DUP_ERROR);
1227 		}
1228 		dr->dr_resp.buf = NULL;
1229 		dr->dr_resp.maxlen = 0;
1230 		dr->dr_addr.buf = NULL;
1231 		dr->dr_addr.maxlen = 0;
1232 		if (rdmadrmru) {
1233 			dr->dr_next = rdmadrmru->dr_next;
1234 			rdmadrmru->dr_next = dr;
1235 		} else {
1236 			dr->dr_next = dr;
1237 		}
1238 		rdmandupreqs++;
1239 	} else {
1240 		dr = rdmadrmru->dr_next;
1241 		while (dr->dr_status == DUP_INPROGRESS) {
1242 			dr = dr->dr_next;
1243 			if (dr == rdmadrmru->dr_next) {
1244 				mutex_exit(&rdmadupreq_lock);
1245 				return (DUP_ERROR);
1246 			}
1247 		}
1248 		unhash(dr);
1249 		if (dr->dr_resfree) {
1250 			(*dr->dr_resfree)(dr->dr_resp.buf);
1251 		}
1252 	}
1253 	dr->dr_resfree = NULL;
1254 	rdmadrmru = dr;
1255 
1256 	dr->dr_xid = REQTOXID(req);
1257 	dr->dr_prog = req->rq_prog;
1258 	dr->dr_vers = req->rq_vers;
1259 	dr->dr_proc = req->rq_proc;
1260 	if (dr->dr_addr.maxlen < req->rq_xprt->xp_rtaddr.len) {
1261 		if (dr->dr_addr.buf != NULL)
1262 			kmem_free(dr->dr_addr.buf, dr->dr_addr.maxlen);
1263 		dr->dr_addr.maxlen = req->rq_xprt->xp_rtaddr.len;
1264 		dr->dr_addr.buf = kmem_alloc(dr->dr_addr.maxlen, KM_NOSLEEP);
1265 		if (dr->dr_addr.buf == NULL) {
1266 			dr->dr_addr.maxlen = 0;
1267 			dr->dr_status = DUP_DROP;
1268 			mutex_exit(&rdmadupreq_lock);
1269 			return (DUP_ERROR);
1270 		}
1271 	}
1272 	dr->dr_addr.len = req->rq_xprt->xp_rtaddr.len;
1273 	bcopy(req->rq_xprt->xp_rtaddr.buf, dr->dr_addr.buf, dr->dr_addr.len);
1274 	if (dr->dr_resp.maxlen < size) {
1275 		if (dr->dr_resp.buf != NULL)
1276 			kmem_free(dr->dr_resp.buf, dr->dr_resp.maxlen);
1277 		dr->dr_resp.maxlen = (unsigned int)size;
1278 		dr->dr_resp.buf = kmem_alloc(size, KM_NOSLEEP);
1279 		if (dr->dr_resp.buf == NULL) {
1280 			dr->dr_resp.maxlen = 0;
1281 			dr->dr_status = DUP_DROP;
1282 			mutex_exit(&rdmadupreq_lock);
1283 			return (DUP_ERROR);
1284 		}
1285 	}
1286 	dr->dr_status = DUP_INPROGRESS;
1287 
1288 	drhash = (uint32_t)DRHASH(dr);
1289 	dr->dr_chain = rdmadrhashtbl[drhash];
1290 	rdmadrhashtbl[drhash] = dr;
1291 	rdmadrhashstat[drhash]++;
1292 	mutex_exit(&rdmadupreq_lock);
1293 	*drpp = dr;
1294 	return (DUP_NEW);
1295 }
1296 
1297 /*
1298  * svc_rdma_kdupdone marks the request done (DUP_DONE or DUP_DROP)
1299  * and stores the response.
1300  */
1301 static void
1302 svc_rdma_kdupdone(struct dupreq *dr, caddr_t res, void (*dis_resfree)(),
1303 	int size, int status)
1304 {
1305 	ASSERT(dr->dr_resfree == NULL);
1306 	if (status == DUP_DONE) {
1307 		bcopy(res, dr->dr_resp.buf, size);
1308 		dr->dr_resfree = dis_resfree;
1309 	}
1310 	dr->dr_status = status;
1311 }
1312 
1313 /*
1314  * This routine expects that the mutex, rdmadupreq_lock, is already held.
1315  */
1316 static void
1317 unhash(struct dupreq *dr)
1318 {
1319 	struct dupreq *drt;
1320 	struct dupreq *drtprev = NULL;
1321 	uint32_t drhash;
1322 
1323 	ASSERT(MUTEX_HELD(&rdmadupreq_lock));
1324 
1325 	drhash = (uint32_t)DRHASH(dr);
1326 	drt = rdmadrhashtbl[drhash];
1327 	while (drt != NULL) {
1328 		if (drt == dr) {
1329 			rdmadrhashstat[drhash]--;
1330 			if (drtprev == NULL) {
1331 				rdmadrhashtbl[drhash] = drt->dr_chain;
1332 			} else {
1333 				drtprev->dr_chain = drt->dr_chain;
1334 			}
1335 			return;
1336 		}
1337 		drtprev = drt;
1338 		drt = drt->dr_chain;
1339 	}
1340 }
1341 
1342 bool_t
1343 rdma_get_wchunk(struct svc_req *req, iovec_t *iov, struct clist *wlist)
1344 {
1345 	struct clist	*clist;
1346 	uint32_t	tlen;
1347 
1348 	if (req->rq_xprt->xp_type != T_RDMA) {
1349 		return (FALSE);
1350 	}
1351 
1352 	tlen = 0;
1353 	clist = wlist;
1354 	while (clist) {
1355 		tlen += clist->c_len;
1356 		clist = clist->c_next;
1357 	}
1358 
1359 	/*
1360 	 * set iov to addr+len of first segment of first wchunk of
1361 	 * wlist sent by client.  krecv() already malloc'd a buffer
1362 	 * large enough, but registration is deferred until we write
1363 	 * the buffer back to (NFS) client using RDMA_WRITE.
1364 	 */
1365 	iov->iov_base = (caddr_t)(uintptr_t)wlist->w.c_saddr;
1366 	iov->iov_len = tlen;
1367 
1368 	return (TRUE);
1369 }
1370 
1371 /*
1372  * routine to setup the read chunk lists
1373  */
1374 
1375 int
1376 rdma_setup_read_chunks(struct clist *wcl, uint32_t count, int *wcl_len)
1377 {
1378 	int		data_len, avail_len;
1379 	uint_t		round_len;
1380 
1381 	data_len = avail_len = 0;
1382 
1383 	while (wcl != NULL && count > 0) {
1384 		if (wcl->c_dmemhandle.mrc_rmr == 0)
1385 			break;
1386 
1387 		if (wcl->c_len < count) {
1388 			data_len += wcl->c_len;
1389 			avail_len = 0;
1390 		} else {
1391 			data_len += count;
1392 			avail_len = wcl->c_len - count;
1393 			wcl->c_len = count;
1394 		}
1395 		count -= wcl->c_len;
1396 
1397 		if (count == 0)
1398 			break;
1399 
1400 		wcl = wcl->c_next;
1401 	}
1402 
1403 	/*
1404 	 * MUST fail if there are still more data
1405 	 */
1406 	if (count > 0) {
1407 		DTRACE_PROBE2(krpc__e__rdma_setup_read_chunks_clist_len,
1408 		    int, data_len, int, count);
1409 		return (FALSE);
1410 	}
1411 
1412 	/*
1413 	 * Round up the last chunk to 4-byte boundary
1414 	 */
1415 	*wcl_len = roundup(data_len, BYTES_PER_XDR_UNIT);
1416 	round_len = *wcl_len - data_len;
1417 
1418 	if (round_len) {
1419 
1420 		/*
1421 		 * If there is space in the current chunk,
1422 		 * add the roundup to the chunk.
1423 		 */
1424 		if (avail_len >= round_len) {
1425 			wcl->c_len += round_len;
1426 		} else  {
1427 			/*
1428 			 * try the next one.
1429 			 */
1430 			wcl = wcl->c_next;
1431 			if ((wcl == NULL) || (wcl->c_len < round_len)) {
1432 				DTRACE_PROBE1(
1433 				    krpc__e__rdma_setup_read_chunks_rndup,
1434 				    int, round_len);
1435 				return (FALSE);
1436 			}
1437 			wcl->c_len = round_len;
1438 		}
1439 	}
1440 
1441 	wcl = wcl->c_next;
1442 
1443 	/*
1444 	 * Make rest of the chunks 0-len
1445 	 */
1446 
1447 	clist_zero_len(wcl);
1448 
1449 	return (TRUE);
1450 }
1451