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 (c) 2012 by Delphix. All rights reserved.
24 * Copyright 2013 Nexenta Systems, Inc. All rights reserved.
25 */
26 /* Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T */
27 /* All Rights Reserved */
28 /*
29 * Portions of this source code were derived from Berkeley
30 * 4.3 BSD under license from the Regents of the University of
31 * California.
32 */
33
34 /*
35 * Server side of RPC over RDMA in the kernel.
36 */
37
38 #include <sys/param.h>
39 #include <sys/types.h>
40 #include <sys/user.h>
41 #include <sys/sysmacros.h>
42 #include <sys/proc.h>
43 #include <sys/file.h>
44 #include <sys/errno.h>
45 #include <sys/kmem.h>
46 #include <sys/debug.h>
47 #include <sys/systm.h>
48 #include <sys/cmn_err.h>
49 #include <sys/kstat.h>
50 #include <sys/vtrace.h>
51 #include <sys/debug.h>
52
53 #include <rpc/types.h>
54 #include <rpc/xdr.h>
55 #include <rpc/auth.h>
56 #include <rpc/clnt.h>
57 #include <rpc/rpc_msg.h>
58 #include <rpc/svc.h>
59 #include <rpc/rpc_rdma.h>
60 #include <sys/ddi.h>
61 #include <sys/sunddi.h>
62
63 #include <inet/common.h>
64 #include <inet/ip.h>
65 #include <inet/ip6.h>
66
67 #include <nfs/nfs.h>
68 #include <sys/sdt.h>
69
70 #define SVC_RDMA_SUCCESS 0
71 #define SVC_RDMA_FAIL -1
72
73 #define SVC_CREDIT_FACTOR (0.5)
74
75 #define MSG_IS_RPCSEC_GSS(msg) \
76 ((msg)->rm_reply.rp_acpt.ar_verf.oa_flavor == RPCSEC_GSS)
77
78
79 uint32_t rdma_bufs_granted = RDMA_BUFS_GRANT;
80
81 /*
82 * RDMA transport specific data associated with SVCMASTERXPRT
83 */
84 struct rdma_data {
85 SVCMASTERXPRT *rd_xprt; /* back ptr to SVCMASTERXPRT */
86 struct rdma_svc_data rd_data; /* rdma data */
87 rdma_mod_t *r_mod; /* RDMA module containing ops ptr */
88 };
89
90 /*
91 * Plugin connection specific data stashed away in clone SVCXPRT
92 */
93 struct clone_rdma_data {
94 bool_t cloned; /* xprt cloned for thread processing */
95 CONN *conn; /* RDMA connection */
96 rdma_buf_t rpcbuf; /* RPC req/resp buffer */
97 struct clist *cl_reply; /* reply chunk buffer info */
98 struct clist *cl_wlist; /* write list clist */
99 };
100
101
102 #define MAXADDRLEN 128 /* max length for address mask */
103
104 /*
105 * Routines exported through ops vector.
106 */
107 static bool_t svc_rdma_krecv(SVCXPRT *, mblk_t *, struct rpc_msg *);
108 static bool_t svc_rdma_ksend(SVCXPRT *, struct rpc_msg *);
109 static bool_t svc_rdma_kgetargs(SVCXPRT *, xdrproc_t, caddr_t);
110 static bool_t svc_rdma_kfreeargs(SVCXPRT *, xdrproc_t, caddr_t);
111 void svc_rdma_kdestroy(SVCMASTERXPRT *);
112 static int svc_rdma_kdup(struct svc_req *, caddr_t, int,
113 struct dupreq **, bool_t *);
114 static void svc_rdma_kdupdone(struct dupreq *, caddr_t,
115 void (*)(), int, int);
116 static int32_t *svc_rdma_kgetres(SVCXPRT *, int);
117 static void svc_rdma_kfreeres(SVCXPRT *);
118 static void svc_rdma_kclone_destroy(SVCXPRT *);
119 static void svc_rdma_kstart(SVCMASTERXPRT *);
120 void svc_rdma_kstop(SVCMASTERXPRT *);
121 static void svc_rdma_kclone_xprt(SVCXPRT *, SVCXPRT *);
122 static void svc_rdma_ktattrs(SVCXPRT *, int, void **);
123
124 static int svc_process_long_reply(SVCXPRT *, xdrproc_t,
125 caddr_t, struct rpc_msg *, bool_t, int *,
126 int *, int *, unsigned int *);
127
128 static int svc_compose_rpcmsg(SVCXPRT *, CONN *, xdrproc_t,
129 caddr_t, rdma_buf_t *, XDR **, struct rpc_msg *,
130 bool_t, uint_t *);
131 static bool_t rpcmsg_length(xdrproc_t,
132 caddr_t,
133 struct rpc_msg *, bool_t, int);
134
135 /*
136 * Server transport operations vector.
137 */
138 struct svc_ops rdma_svc_ops = {
139 svc_rdma_krecv, /* Get requests */
140 svc_rdma_kgetargs, /* Deserialize arguments */
141 svc_rdma_ksend, /* Send reply */
142 svc_rdma_kfreeargs, /* Free argument data space */
143 svc_rdma_kdestroy, /* Destroy transport handle */
144 svc_rdma_kdup, /* Check entry in dup req cache */
145 svc_rdma_kdupdone, /* Mark entry in dup req cache as done */
146 svc_rdma_kgetres, /* Get pointer to response buffer */
147 svc_rdma_kfreeres, /* Destroy pre-serialized response header */
148 svc_rdma_kclone_destroy, /* Destroy a clone xprt */
149 svc_rdma_kstart, /* Tell `ready-to-receive' to rpcmod */
150 svc_rdma_kclone_xprt, /* Transport specific clone xprt */
151 svc_rdma_ktattrs /* Get Transport Attributes */
152 };
153
154 /*
155 * Server statistics
156 * NOTE: This structure type is duplicated in the NFS fast path.
157 */
158 struct {
159 kstat_named_t rscalls;
160 kstat_named_t rsbadcalls;
161 kstat_named_t rsnullrecv;
162 kstat_named_t rsbadlen;
163 kstat_named_t rsxdrcall;
164 kstat_named_t rsdupchecks;
165 kstat_named_t rsdupreqs;
166 kstat_named_t rslongrpcs;
167 kstat_named_t rstotalreplies;
168 kstat_named_t rstotallongreplies;
169 kstat_named_t rstotalinlinereplies;
170 } rdmarsstat = {
171 { "calls", KSTAT_DATA_UINT64 },
172 { "badcalls", KSTAT_DATA_UINT64 },
173 { "nullrecv", KSTAT_DATA_UINT64 },
174 { "badlen", KSTAT_DATA_UINT64 },
175 { "xdrcall", KSTAT_DATA_UINT64 },
176 { "dupchecks", KSTAT_DATA_UINT64 },
177 { "dupreqs", KSTAT_DATA_UINT64 },
178 { "longrpcs", KSTAT_DATA_UINT64 },
179 { "totalreplies", KSTAT_DATA_UINT64 },
180 { "totallongreplies", KSTAT_DATA_UINT64 },
181 { "totalinlinereplies", KSTAT_DATA_UINT64 },
182 };
183
184 kstat_named_t *rdmarsstat_ptr = (kstat_named_t *)&rdmarsstat;
185 uint_t rdmarsstat_ndata = sizeof (rdmarsstat) / sizeof (kstat_named_t);
186
187 #define RSSTAT_INCR(x) atomic_inc_64(&rdmarsstat.x.value.ui64)
188 /*
189 * Create a transport record.
190 * The transport record, output buffer, and private data structure
191 * are allocated. The output buffer is serialized into using xdrmem.
192 * There is one transport record per user process which implements a
193 * set of services.
194 */
195 /* ARGSUSED */
196 int
197 svc_rdma_kcreate(char *netid, SVC_CALLOUT_TABLE *sct, int id,
198 rdma_xprt_group_t *started_xprts)
199 {
200 int error;
201 SVCMASTERXPRT *xprt;
202 struct rdma_data *rd;
203 rdma_registry_t *rmod;
204 rdma_xprt_record_t *xprt_rec;
205 queue_t *q;
206 /*
207 * modload the RDMA plugins is not already done.
208 */
209 if (!rdma_modloaded) {
210 /*CONSTANTCONDITION*/
211 ASSERT(sizeof (struct clone_rdma_data) <= SVC_P2LEN);
212
213 mutex_enter(&rdma_modload_lock);
214 if (!rdma_modloaded) {
215 error = rdma_modload();
216 }
217 mutex_exit(&rdma_modload_lock);
218
219 if (error)
220 return (error);
221 }
222
223 /*
224 * master_xprt_count is the count of master transport handles
225 * that were successfully created and are ready to recieve for
226 * RDMA based access.
227 */
228 error = 0;
229 xprt_rec = NULL;
230 rw_enter(&rdma_lock, RW_READER);
231 if (rdma_mod_head == NULL) {
232 started_xprts->rtg_count = 0;
233 rw_exit(&rdma_lock);
234 if (rdma_dev_available)
235 return (EPROTONOSUPPORT);
236 else
237 return (ENODEV);
238 }
239
240 /*
241 * If we have reached here, then atleast one RDMA plugin has loaded.
242 * Create a master_xprt, make it start listenining on the device,
243 * if an error is generated, record it, we might need to shut
244 * the master_xprt.
245 * SVC_START() calls svc_rdma_kstart which calls plugin binding
246 * routines.
247 */
248 for (rmod = rdma_mod_head; rmod != NULL; rmod = rmod->r_next) {
249
250 /*
251 * One SVCMASTERXPRT per RDMA plugin.
252 */
253 xprt = kmem_zalloc(sizeof (*xprt), KM_SLEEP);
254 xprt->xp_ops = &rdma_svc_ops;
255 xprt->xp_sct = sct;
256 xprt->xp_type = T_RDMA;
257 mutex_init(&xprt->xp_req_lock, NULL, MUTEX_DEFAULT, NULL);
258 mutex_init(&xprt->xp_thread_lock, NULL, MUTEX_DEFAULT, NULL);
259 xprt->xp_req_head = (mblk_t *)0;
260 xprt->xp_req_tail = (mblk_t *)0;
261 xprt->xp_full = FALSE;
262 xprt->xp_enable = FALSE;
263 xprt->xp_reqs = 0;
264 xprt->xp_size = 0;
265 xprt->xp_threads = 0;
266 xprt->xp_detached_threads = 0;
267
268 rd = kmem_zalloc(sizeof (*rd), KM_SLEEP);
269 xprt->xp_p2 = (caddr_t)rd;
270 rd->rd_xprt = xprt;
271 rd->r_mod = rmod->r_mod;
272
273 q = &rd->rd_data.q;
274 xprt->xp_wq = q;
275 q->q_ptr = &rd->rd_xprt;
276 xprt->xp_netid = NULL;
277
278 /*
279 * Each of the plugins will have their own Service ID
280 * to listener specific mapping, like port number for VI
281 * and service name for IB.
282 */
283 rd->rd_data.svcid = id;
284 error = svc_xprt_register(xprt, id);
285 if (error) {
286 DTRACE_PROBE(krpc__e__svcrdma__xprt__reg);
287 goto cleanup;
288 }
289
290 SVC_START(xprt);
291 if (!rd->rd_data.active) {
292 svc_xprt_unregister(xprt);
293 error = rd->rd_data.err_code;
294 goto cleanup;
295 }
296
297 /*
298 * This is set only when there is atleast one or more
299 * transports successfully created. We insert the pointer
300 * to the created RDMA master xprt into a separately maintained
301 * list. This way we can easily reference it later to cleanup,
302 * when NFS kRPC service pool is going away/unregistered.
303 */
304 started_xprts->rtg_count ++;
305 xprt_rec = kmem_alloc(sizeof (*xprt_rec), KM_SLEEP);
306 xprt_rec->rtr_xprt_ptr = xprt;
307 xprt_rec->rtr_next = started_xprts->rtg_listhead;
308 started_xprts->rtg_listhead = xprt_rec;
309 continue;
310 cleanup:
311 SVC_DESTROY(xprt);
312 if (error == RDMA_FAILED)
313 error = EPROTONOSUPPORT;
314 }
315
316 rw_exit(&rdma_lock);
317
318 /*
319 * Don't return any error even if a single plugin was started
320 * successfully.
321 */
322 if (started_xprts->rtg_count == 0)
323 return (error);
324 return (0);
325 }
326
327 /*
328 * Cleanup routine for freeing up memory allocated by
329 * svc_rdma_kcreate()
330 */
331 void
332 svc_rdma_kdestroy(SVCMASTERXPRT *xprt)
333 {
334 struct rdma_data *rd = (struct rdma_data *)xprt->xp_p2;
335
336
337 mutex_destroy(&xprt->xp_req_lock);
338 mutex_destroy(&xprt->xp_thread_lock);
339 kmem_free(rd, sizeof (*rd));
340 kmem_free(xprt, sizeof (*xprt));
341 }
342
343
344 static void
345 svc_rdma_kstart(SVCMASTERXPRT *xprt)
346 {
347 struct rdma_svc_data *svcdata;
348 rdma_mod_t *rmod;
349
350 svcdata = &((struct rdma_data *)xprt->xp_p2)->rd_data;
351 rmod = ((struct rdma_data *)xprt->xp_p2)->r_mod;
352
353 /*
354 * Create a listener for module at this port
355 */
356
357 if (rmod->rdma_count != 0)
358 (*rmod->rdma_ops->rdma_svc_listen)(svcdata);
359 else
360 svcdata->err_code = RDMA_FAILED;
361 }
362
363 void
364 svc_rdma_kstop(SVCMASTERXPRT *xprt)
365 {
366 struct rdma_svc_data *svcdata;
367 rdma_mod_t *rmod;
368
369 svcdata = &((struct rdma_data *)xprt->xp_p2)->rd_data;
370 rmod = ((struct rdma_data *)xprt->xp_p2)->r_mod;
371
372 /*
373 * Call the stop listener routine for each plugin. If rdma_count is
374 * already zero set active to zero.
375 */
376 if (rmod->rdma_count != 0)
377 (*rmod->rdma_ops->rdma_svc_stop)(svcdata);
378 else
379 svcdata->active = 0;
380 if (svcdata->active)
381 DTRACE_PROBE(krpc__e__svcrdma__kstop);
382 }
383
384 /* ARGSUSED */
385 static void
386 svc_rdma_kclone_destroy(SVCXPRT *clone_xprt)
387 {
388
389 struct clone_rdma_data *cdrp;
390 cdrp = (struct clone_rdma_data *)clone_xprt->xp_p2buf;
391
392 /*
393 * Only free buffers and release connection when cloned is set.
394 */
395 if (cdrp->cloned != TRUE)
396 return;
397
398 rdma_buf_free(cdrp->conn, &cdrp->rpcbuf);
399 if (cdrp->cl_reply) {
400 clist_free(cdrp->cl_reply);
401 cdrp->cl_reply = NULL;
402 }
403 RDMA_REL_CONN(cdrp->conn);
404
405 cdrp->cloned = 0;
406 }
407
408 /*
409 * Clone the xprt specific information. It will be freed by
410 * SVC_CLONE_DESTROY.
411 */
412 static void
413 svc_rdma_kclone_xprt(SVCXPRT *src_xprt, SVCXPRT *dst_xprt)
414 {
415 struct clone_rdma_data *srcp2;
416 struct clone_rdma_data *dstp2;
417
418 srcp2 = (struct clone_rdma_data *)src_xprt->xp_p2buf;
419 dstp2 = (struct clone_rdma_data *)dst_xprt->xp_p2buf;
420
421 if (srcp2->conn != NULL) {
422 srcp2->cloned = TRUE;
423 *dstp2 = *srcp2;
424 }
425 }
426
427 static void
428 svc_rdma_ktattrs(SVCXPRT *clone_xprt, int attrflag, void **tattr)
429 {
430 CONN *conn;
431 *tattr = NULL;
432
433 switch (attrflag) {
434 case SVC_TATTR_ADDRMASK:
435 conn = ((struct clone_rdma_data *)clone_xprt->xp_p2buf)->conn;
436 ASSERT(conn != NULL);
437 if (conn)
438 *tattr = (void *)&conn->c_addrmask;
439 }
440 }
441
442 static bool_t
443 svc_rdma_krecv(SVCXPRT *clone_xprt, mblk_t *mp, struct rpc_msg *msg)
444 {
445 XDR *xdrs;
446 CONN *conn;
447 rdma_recv_data_t *rdp = (rdma_recv_data_t *)mp->b_rptr;
448 struct clone_rdma_data *crdp;
449 struct clist *cl = NULL;
450 struct clist *wcl = NULL;
451 struct clist *cllong = NULL;
452
453 rdma_stat status;
454 uint32_t vers, op, pos, xid;
455 uint32_t rdma_credit;
456 uint32_t wcl_total_length = 0;
457 bool_t wwl = FALSE;
458
459 crdp = (struct clone_rdma_data *)clone_xprt->xp_p2buf;
460 RSSTAT_INCR(rscalls);
461 conn = rdp->conn;
462
463 status = rdma_svc_postrecv(conn);
464 if (status != RDMA_SUCCESS) {
465 DTRACE_PROBE(krpc__e__svcrdma__krecv__postrecv);
466 goto badrpc_call;
467 }
468
469 xdrs = &clone_xprt->xp_xdrin;
470 xdrmem_create(xdrs, rdp->rpcmsg.addr, rdp->rpcmsg.len, XDR_DECODE);
471 xid = *(uint32_t *)rdp->rpcmsg.addr;
472 XDR_SETPOS(xdrs, sizeof (uint32_t));
473
474 if (! xdr_u_int(xdrs, &vers) ||
475 ! xdr_u_int(xdrs, &rdma_credit) ||
476 ! xdr_u_int(xdrs, &op)) {
477 DTRACE_PROBE(krpc__e__svcrdma__krecv__uint);
478 goto xdr_err;
479 }
480
481 /* Checking if the status of the recv operation was normal */
482 if (rdp->status != 0) {
483 DTRACE_PROBE1(krpc__e__svcrdma__krecv__invalid__status,
484 int, rdp->status);
485 goto badrpc_call;
486 }
487
488 if (! xdr_do_clist(xdrs, &cl)) {
489 DTRACE_PROBE(krpc__e__svcrdma__krecv__do__clist);
490 goto xdr_err;
491 }
492
493 if (!xdr_decode_wlist_svc(xdrs, &wcl, &wwl, &wcl_total_length, conn)) {
494 DTRACE_PROBE(krpc__e__svcrdma__krecv__decode__wlist);
495 if (cl)
496 clist_free(cl);
497 goto xdr_err;
498 }
499 crdp->cl_wlist = wcl;
500
501 crdp->cl_reply = NULL;
502 (void) xdr_decode_reply_wchunk(xdrs, &crdp->cl_reply);
503
504 /*
505 * A chunk at 0 offset indicates that the RPC call message
506 * is in a chunk. Get the RPC call message chunk.
507 */
508 if (cl != NULL && op == RDMA_NOMSG) {
509
510 /* Remove RPC call message chunk from chunklist */
511 cllong = cl;
512 cl = cl->c_next;
513 cllong->c_next = NULL;
514
515
516 /* Allocate and register memory for the RPC call msg chunk */
517 cllong->rb_longbuf.type = RDMA_LONG_BUFFER;
518 cllong->rb_longbuf.len = cllong->c_len > LONG_REPLY_LEN ?
519 cllong->c_len : LONG_REPLY_LEN;
520
521 if (rdma_buf_alloc(conn, &cllong->rb_longbuf)) {
522 clist_free(cllong);
523 goto cll_malloc_err;
524 }
525
526 cllong->u.c_daddr3 = cllong->rb_longbuf.addr;
527
528 if (cllong->u.c_daddr == NULL) {
529 DTRACE_PROBE(krpc__e__svcrdma__krecv__nomem);
530 rdma_buf_free(conn, &cllong->rb_longbuf);
531 clist_free(cllong);
532 goto cll_malloc_err;
533 }
534
535 status = clist_register(conn, cllong, CLIST_REG_DST);
536 if (status) {
537 DTRACE_PROBE(krpc__e__svcrdma__krecv__clist__reg);
538 rdma_buf_free(conn, &cllong->rb_longbuf);
539 clist_free(cllong);
540 goto cll_malloc_err;
541 }
542
543 /*
544 * Now read the RPC call message in
545 */
546 status = RDMA_READ(conn, cllong, WAIT);
547 if (status) {
548 DTRACE_PROBE(krpc__e__svcrdma__krecv__read);
549 (void) clist_deregister(conn, cllong);
550 rdma_buf_free(conn, &cllong->rb_longbuf);
551 clist_free(cllong);
552 goto cll_malloc_err;
553 }
554
555 status = clist_syncmem(conn, cllong, CLIST_REG_DST);
556 (void) clist_deregister(conn, cllong);
557
558 xdrrdma_create(xdrs, (caddr_t)(uintptr_t)cllong->u.c_daddr3,
559 cllong->c_len, 0, cl, XDR_DECODE, conn);
560
561 crdp->rpcbuf = cllong->rb_longbuf;
562 crdp->rpcbuf.len = cllong->c_len;
563 clist_free(cllong);
564 RDMA_BUF_FREE(conn, &rdp->rpcmsg);
565 } else {
566 pos = XDR_GETPOS(xdrs);
567 xdrrdma_create(xdrs, rdp->rpcmsg.addr + pos,
568 rdp->rpcmsg.len - pos, 0, cl, XDR_DECODE, conn);
569 crdp->rpcbuf = rdp->rpcmsg;
570
571 /* Use xdrrdmablk_ops to indicate there is a read chunk list */
572 if (cl != NULL) {
573 int32_t flg = XDR_RDMA_RLIST_REG;
574
575 XDR_CONTROL(xdrs, XDR_RDMA_SET_FLAGS, &flg);
576 xdrs->x_ops = &xdrrdmablk_ops;
577 }
578 }
579
580 if (crdp->cl_wlist) {
581 int32_t flg = XDR_RDMA_WLIST_REG;
582
583 XDR_CONTROL(xdrs, XDR_RDMA_SET_WLIST, crdp->cl_wlist);
584 XDR_CONTROL(xdrs, XDR_RDMA_SET_FLAGS, &flg);
585 }
586
587 if (! xdr_callmsg(xdrs, msg)) {
588 DTRACE_PROBE(krpc__e__svcrdma__krecv__callmsg);
589 RSSTAT_INCR(rsxdrcall);
590 goto callmsg_err;
591 }
592
593 /*
594 * Point the remote transport address in the service_transport
595 * handle at the address in the request.
596 */
597 clone_xprt->xp_rtaddr.buf = conn->c_raddr.buf;
598 clone_xprt->xp_rtaddr.len = conn->c_raddr.len;
599 clone_xprt->xp_rtaddr.maxlen = conn->c_raddr.len;
600
601 clone_xprt->xp_lcladdr.buf = conn->c_laddr.buf;
602 clone_xprt->xp_lcladdr.len = conn->c_laddr.len;
603 clone_xprt->xp_lcladdr.maxlen = conn->c_laddr.len;
604
605 /*
606 * In case of RDMA, connection management is
607 * entirely done in rpcib module and netid in the
608 * SVCMASTERXPRT is NULL. Initialize the clone netid
609 * from the connection.
610 */
611
612 clone_xprt->xp_netid = conn->c_netid;
613
614 clone_xprt->xp_xid = xid;
615 crdp->conn = conn;
616
617 freeb(mp);
618
619 return (TRUE);
620
621 callmsg_err:
622 rdma_buf_free(conn, &crdp->rpcbuf);
623
624 cll_malloc_err:
625 if (cl)
626 clist_free(cl);
627 xdr_err:
628 XDR_DESTROY(xdrs);
629
630 badrpc_call:
631 RDMA_BUF_FREE(conn, &rdp->rpcmsg);
632 RDMA_REL_CONN(conn);
633 freeb(mp);
634 RSSTAT_INCR(rsbadcalls);
635 return (FALSE);
636 }
637
638 static int
639 svc_process_long_reply(SVCXPRT * clone_xprt,
640 xdrproc_t xdr_results, caddr_t xdr_location,
641 struct rpc_msg *msg, bool_t has_args, int *msglen,
642 int *freelen, int *numchunks, unsigned int *final_len)
643 {
644 int status;
645 XDR xdrslong;
646 struct clist *wcl = NULL;
647 int count = 0;
648 int alloc_len;
649 char *memp;
650 rdma_buf_t long_rpc = {0};
651 struct clone_rdma_data *crdp;
652
653 crdp = (struct clone_rdma_data *)clone_xprt->xp_p2buf;
654
655 bzero(&xdrslong, sizeof (xdrslong));
656
657 /* Choose a size for the long rpc response */
658 if (MSG_IS_RPCSEC_GSS(msg)) {
659 alloc_len = RNDUP(MAX_AUTH_BYTES + *msglen);
660 } else {
661 alloc_len = RNDUP(*msglen);
662 }
663
664 if (alloc_len <= 64 * 1024) {
665 if (alloc_len > 32 * 1024) {
666 alloc_len = 64 * 1024;
667 } else {
668 if (alloc_len > 16 * 1024) {
669 alloc_len = 32 * 1024;
670 } else {
671 alloc_len = 16 * 1024;
672 }
673 }
674 }
675
676 long_rpc.type = RDMA_LONG_BUFFER;
677 long_rpc.len = alloc_len;
678 if (rdma_buf_alloc(crdp->conn, &long_rpc)) {
679 return (SVC_RDMA_FAIL);
680 }
681
682 memp = long_rpc.addr;
683 xdrmem_create(&xdrslong, memp, alloc_len, XDR_ENCODE);
684
685 msg->rm_xid = clone_xprt->xp_xid;
686
687 if (!(xdr_replymsg(&xdrslong, msg) &&
688 (!has_args || SVCAUTH_WRAP(&clone_xprt->xp_auth, &xdrslong,
689 xdr_results, xdr_location)))) {
690 rdma_buf_free(crdp->conn, &long_rpc);
691 DTRACE_PROBE(krpc__e__svcrdma__longrep__authwrap);
692 return (SVC_RDMA_FAIL);
693 }
694
695 *final_len = XDR_GETPOS(&xdrslong);
696
697 DTRACE_PROBE1(krpc__i__replylen, uint_t, *final_len);
698 *numchunks = 0;
699 *freelen = 0;
700
701 wcl = crdp->cl_reply;
702 wcl->rb_longbuf = long_rpc;
703
704 count = *final_len;
705 while ((wcl != NULL) && (count > 0)) {
706
707 if (wcl->c_dmemhandle.mrc_rmr == 0)
708 break;
709
710 DTRACE_PROBE2(krpc__i__write__chunks, uint32_t, count,
711 uint32_t, wcl->c_len);
712
713 if (wcl->c_len > count) {
714 wcl->c_len = count;
715 }
716 wcl->w.c_saddr3 = (caddr_t)memp;
717
718 count -= wcl->c_len;
719 *numchunks += 1;
720 memp += wcl->c_len;
721 wcl = wcl->c_next;
722 }
723
724 /*
725 * Make rest of the chunks 0-len
726 */
727 while (wcl != NULL) {
728 if (wcl->c_dmemhandle.mrc_rmr == 0)
729 break;
730 wcl->c_len = 0;
731 wcl = wcl->c_next;
732 }
733
734 wcl = crdp->cl_reply;
735
736 /*
737 * MUST fail if there are still more data
738 */
739 if (count > 0) {
740 rdma_buf_free(crdp->conn, &long_rpc);
741 DTRACE_PROBE(krpc__e__svcrdma__longrep__dlen__clist);
742 return (SVC_RDMA_FAIL);
743 }
744
745 if (clist_register(crdp->conn, wcl, CLIST_REG_SOURCE) != RDMA_SUCCESS) {
746 rdma_buf_free(crdp->conn, &long_rpc);
747 DTRACE_PROBE(krpc__e__svcrdma__longrep__clistreg);
748 return (SVC_RDMA_FAIL);
749 }
750
751 status = clist_syncmem(crdp->conn, wcl, CLIST_REG_SOURCE);
752
753 if (status) {
754 (void) clist_deregister(crdp->conn, wcl);
755 rdma_buf_free(crdp->conn, &long_rpc);
756 DTRACE_PROBE(krpc__e__svcrdma__longrep__syncmem);
757 return (SVC_RDMA_FAIL);
758 }
759
760 status = RDMA_WRITE(crdp->conn, wcl, WAIT);
761
762 (void) clist_deregister(crdp->conn, wcl);
763 rdma_buf_free(crdp->conn, &wcl->rb_longbuf);
764
765 if (status != RDMA_SUCCESS) {
766 DTRACE_PROBE(krpc__e__svcrdma__longrep__write);
767 return (SVC_RDMA_FAIL);
768 }
769
770 return (SVC_RDMA_SUCCESS);
771 }
772
773
774 static int
775 svc_compose_rpcmsg(SVCXPRT * clone_xprt, CONN * conn, xdrproc_t xdr_results,
776 caddr_t xdr_location, rdma_buf_t *rpcreply, XDR ** xdrs,
777 struct rpc_msg *msg, bool_t has_args, uint_t *len)
778 {
779 /*
780 * Get a pre-allocated buffer for rpc reply
781 */
782 rpcreply->type = SEND_BUFFER;
783 if (rdma_buf_alloc(conn, rpcreply)) {
784 DTRACE_PROBE(krpc__e__svcrdma__rpcmsg__reply__nofreebufs);
785 return (SVC_RDMA_FAIL);
786 }
787
788 xdrrdma_create(*xdrs, rpcreply->addr, rpcreply->len,
789 0, NULL, XDR_ENCODE, conn);
790
791 msg->rm_xid = clone_xprt->xp_xid;
792
793 if (has_args) {
794 if (!(xdr_replymsg(*xdrs, msg) &&
795 (!has_args ||
796 SVCAUTH_WRAP(&clone_xprt->xp_auth, *xdrs,
797 xdr_results, xdr_location)))) {
798 rdma_buf_free(conn, rpcreply);
799 DTRACE_PROBE(
800 krpc__e__svcrdma__rpcmsg__reply__authwrap1);
801 return (SVC_RDMA_FAIL);
802 }
803 } else {
804 if (!xdr_replymsg(*xdrs, msg)) {
805 rdma_buf_free(conn, rpcreply);
806 DTRACE_PROBE(
807 krpc__e__svcrdma__rpcmsg__reply__authwrap2);
808 return (SVC_RDMA_FAIL);
809 }
810 }
811
812 *len = XDR_GETPOS(*xdrs);
813
814 return (SVC_RDMA_SUCCESS);
815 }
816
817 /*
818 * Send rpc reply.
819 */
820 static bool_t
821 svc_rdma_ksend(SVCXPRT * clone_xprt, struct rpc_msg *msg)
822 {
823 XDR *xdrs_rpc = &(clone_xprt->xp_xdrout);
824 XDR xdrs_rhdr;
825 CONN *conn = NULL;
826 rdma_buf_t rbuf_resp = {0}, rbuf_rpc_resp = {0};
827
828 struct clone_rdma_data *crdp;
829 struct clist *cl_read = NULL;
830 struct clist *cl_send = NULL;
831 struct clist *cl_write = NULL;
832 xdrproc_t xdr_results; /* results XDR encoding function */
833 caddr_t xdr_location; /* response results pointer */
834
835 int retval = FALSE;
836 int status, msglen, num_wreply_segments = 0;
837 uint32_t rdma_credit = 0;
838 int freelen = 0;
839 bool_t has_args;
840 uint_t final_resp_len, rdma_response_op, vers;
841
842 bzero(&xdrs_rhdr, sizeof (XDR));
843 crdp = (struct clone_rdma_data *)clone_xprt->xp_p2buf;
844 conn = crdp->conn;
845
846 /*
847 * If there is a result procedure specified in the reply message,
848 * it will be processed in the xdr_replymsg and SVCAUTH_WRAP.
849 * We need to make sure it won't be processed twice, so we null
850 * it for xdr_replymsg here.
851 */
852 has_args = FALSE;
853 if (msg->rm_reply.rp_stat == MSG_ACCEPTED &&
854 msg->rm_reply.rp_acpt.ar_stat == SUCCESS) {
855 if ((xdr_results = msg->acpted_rply.ar_results.proc) != NULL) {
856 has_args = TRUE;
857 xdr_location = msg->acpted_rply.ar_results.where;
858 msg->acpted_rply.ar_results.proc = xdr_void;
859 msg->acpted_rply.ar_results.where = NULL;
860 }
861 }
862
863 /*
864 * Given the limit on the inline response size (RPC_MSG_SZ),
865 * there is a need to make a guess as to the overall size of
866 * the response. If the resultant size is beyond the inline
867 * size, then the server needs to use the "reply chunk list"
868 * provided by the client (if the client provided one). An
869 * example of this type of response would be a READDIR
870 * response (e.g. a small directory read would fit in RPC_MSG_SZ
871 * and that is the preference but it may not fit)
872 *
873 * Combine the encoded size and the size of the true results
874 * and then make the decision about where to encode and send results.
875 *
876 * One important note, this calculation is ignoring the size
877 * of the encoding of the authentication overhead. The reason
878 * for this is rooted in the complexities of access to the
879 * encoded size of RPCSEC_GSS related authentiation,
880 * integrity, and privacy.
881 *
882 * If it turns out that the encoded authentication bumps the
883 * response over the RPC_MSG_SZ limit, then it may need to
884 * attempt to encode for the reply chunk list.
885 */
886
887 /*
888 * Calculating the "sizeof" the RPC response header and the
889 * encoded results.
890 */
891 msglen = xdr_sizeof(xdr_replymsg, msg);
892
893 if (msglen > 0) {
894 RSSTAT_INCR(rstotalreplies);
895 }
896 if (has_args)
897 msglen += xdrrdma_sizeof(xdr_results, xdr_location,
898 rdma_minchunk, NULL, NULL);
899
900 DTRACE_PROBE1(krpc__i__svcrdma__ksend__msglen, int, msglen);
901
902 status = SVC_RDMA_SUCCESS;
903
904 if (msglen < RPC_MSG_SZ) {
905 /*
906 * Looks like the response will fit in the inline
907 * response; let's try
908 */
909 RSSTAT_INCR(rstotalinlinereplies);
910
911 rdma_response_op = RDMA_MSG;
912
913 status = svc_compose_rpcmsg(clone_xprt, conn, xdr_results,
914 xdr_location, &rbuf_rpc_resp, &xdrs_rpc, msg,
915 has_args, &final_resp_len);
916
917 DTRACE_PROBE1(krpc__i__srdma__ksend__compose_status,
918 int, status);
919 DTRACE_PROBE1(krpc__i__srdma__ksend__compose_len,
920 int, final_resp_len);
921
922 if (status == SVC_RDMA_SUCCESS && crdp->cl_reply) {
923 clist_free(crdp->cl_reply);
924 crdp->cl_reply = NULL;
925 }
926 }
927
928 /*
929 * If the encode failed (size?) or the message really is
930 * larger than what is allowed, try the response chunk list.
931 */
932 if (status != SVC_RDMA_SUCCESS || msglen >= RPC_MSG_SZ) {
933 /*
934 * attempting to use a reply chunk list when there
935 * isn't one won't get very far...
936 */
937 if (crdp->cl_reply == NULL) {
938 DTRACE_PROBE(krpc__e__svcrdma__ksend__noreplycl);
939 goto out;
940 }
941
942 RSSTAT_INCR(rstotallongreplies);
943
944 msglen = xdr_sizeof(xdr_replymsg, msg);
945 msglen += xdrrdma_sizeof(xdr_results, xdr_location, 0,
946 NULL, NULL);
947
948 status = svc_process_long_reply(clone_xprt, xdr_results,
949 xdr_location, msg, has_args, &msglen, &freelen,
950 &num_wreply_segments, &final_resp_len);
951
952 DTRACE_PROBE1(krpc__i__svcrdma__ksend__longreplen,
953 int, final_resp_len);
954
955 if (status != SVC_RDMA_SUCCESS) {
956 DTRACE_PROBE(krpc__e__svcrdma__ksend__compose__failed);
957 goto out;
958 }
959
960 rdma_response_op = RDMA_NOMSG;
961 }
962
963 DTRACE_PROBE1(krpc__i__svcrdma__ksend__rdmamsg__len,
964 int, final_resp_len);
965
966 rbuf_resp.type = SEND_BUFFER;
967 if (rdma_buf_alloc(conn, &rbuf_resp)) {
968 rdma_buf_free(conn, &rbuf_rpc_resp);
969 DTRACE_PROBE(krpc__e__svcrdma__ksend__nofreebufs);
970 goto out;
971 }
972
973 rdma_credit = rdma_bufs_granted;
974
975 vers = RPCRDMA_VERS;
976 xdrmem_create(&xdrs_rhdr, rbuf_resp.addr, rbuf_resp.len, XDR_ENCODE);
977 (*(uint32_t *)rbuf_resp.addr) = msg->rm_xid;
978 /* Skip xid and set the xdr position accordingly. */
979 XDR_SETPOS(&xdrs_rhdr, sizeof (uint32_t));
980 if (!xdr_u_int(&xdrs_rhdr, &vers) ||
981 !xdr_u_int(&xdrs_rhdr, &rdma_credit) ||
982 !xdr_u_int(&xdrs_rhdr, &rdma_response_op)) {
983 rdma_buf_free(conn, &rbuf_rpc_resp);
984 rdma_buf_free(conn, &rbuf_resp);
985 DTRACE_PROBE(krpc__e__svcrdma__ksend__uint);
986 goto out;
987 }
988
989 /*
990 * Now XDR the read chunk list, actually always NULL
991 */
992 (void) xdr_encode_rlist_svc(&xdrs_rhdr, cl_read);
993
994 /*
995 * encode write list -- we already drove RDMA_WRITEs
996 */
997 cl_write = crdp->cl_wlist;
998 if (!xdr_encode_wlist(&xdrs_rhdr, cl_write)) {
999 DTRACE_PROBE(krpc__e__svcrdma__ksend__enc__wlist);
1000 rdma_buf_free(conn, &rbuf_rpc_resp);
1001 rdma_buf_free(conn, &rbuf_resp);
1002 goto out;
1003 }
1004
1005 /*
1006 * XDR encode the RDMA_REPLY write chunk
1007 */
1008 if (!xdr_encode_reply_wchunk(&xdrs_rhdr, crdp->cl_reply,
1009 num_wreply_segments)) {
1010 rdma_buf_free(conn, &rbuf_rpc_resp);
1011 rdma_buf_free(conn, &rbuf_resp);
1012 goto out;
1013 }
1014
1015 clist_add(&cl_send, 0, XDR_GETPOS(&xdrs_rhdr), &rbuf_resp.handle,
1016 rbuf_resp.addr, NULL, NULL);
1017
1018 if (rdma_response_op == RDMA_MSG) {
1019 clist_add(&cl_send, 0, final_resp_len, &rbuf_rpc_resp.handle,
1020 rbuf_rpc_resp.addr, NULL, NULL);
1021 }
1022
1023 status = RDMA_SEND(conn, cl_send, msg->rm_xid);
1024
1025 if (status == RDMA_SUCCESS) {
1026 retval = TRUE;
1027 }
1028
1029 out:
1030 /*
1031 * Free up sendlist chunks
1032 */
1033 if (cl_send != NULL)
1034 clist_free(cl_send);
1035
1036 /*
1037 * Destroy private data for xdr rdma
1038 */
1039 if (clone_xprt->xp_xdrout.x_ops != NULL) {
1040 XDR_DESTROY(&(clone_xprt->xp_xdrout));
1041 }
1042
1043 if (crdp->cl_reply) {
1044 clist_free(crdp->cl_reply);
1045 crdp->cl_reply = NULL;
1046 }
1047
1048 /*
1049 * This is completely disgusting. If public is set it is
1050 * a pointer to a structure whose first field is the address
1051 * of the function to free that structure and any related
1052 * stuff. (see rrokfree in nfs_xdr.c).
1053 */
1054 if (xdrs_rpc->x_public) {
1055 /* LINTED pointer alignment */
1056 (**((int (**)()) xdrs_rpc->x_public)) (xdrs_rpc->x_public);
1057 }
1058
1059 if (xdrs_rhdr.x_ops != NULL) {
1060 XDR_DESTROY(&xdrs_rhdr);
1061 }
1062
1063 return (retval);
1064 }
1065
1066 /*
1067 * Deserialize arguments.
1068 */
1069 static bool_t
1070 svc_rdma_kgetargs(SVCXPRT *clone_xprt, xdrproc_t xdr_args, caddr_t args_ptr)
1071 {
1072 if ((SVCAUTH_UNWRAP(&clone_xprt->xp_auth, &clone_xprt->xp_xdrin,
1073 xdr_args, args_ptr)) != TRUE)
1074 return (FALSE);
1075 return (TRUE);
1076 }
1077
1078 static bool_t
1079 svc_rdma_kfreeargs(SVCXPRT *clone_xprt, xdrproc_t xdr_args,
1080 caddr_t args_ptr)
1081 {
1082 struct clone_rdma_data *crdp;
1083 bool_t retval;
1084
1085 /*
1086 * If the cloned bit is true, then this transport specific
1087 * rmda data has been duplicated into another cloned xprt. Do
1088 * not free, or release the connection, it is still in use. The
1089 * buffers will be freed and the connection released later by
1090 * SVC_CLONE_DESTROY().
1091 */
1092 crdp = (struct clone_rdma_data *)clone_xprt->xp_p2buf;
1093 if (crdp->cloned == TRUE) {
1094 crdp->cloned = 0;
1095 return (TRUE);
1096 }
1097
1098 /*
1099 * Free the args if needed then XDR_DESTROY
1100 */
1101 if (args_ptr) {
1102 XDR *xdrs = &clone_xprt->xp_xdrin;
1103
1104 xdrs->x_op = XDR_FREE;
1105 retval = (*xdr_args)(xdrs, args_ptr);
1106 }
1107
1108 XDR_DESTROY(&(clone_xprt->xp_xdrin));
1109 rdma_buf_free(crdp->conn, &crdp->rpcbuf);
1110 if (crdp->cl_reply) {
1111 clist_free(crdp->cl_reply);
1112 crdp->cl_reply = NULL;
1113 }
1114 RDMA_REL_CONN(crdp->conn);
1115
1116 return (retval);
1117 }
1118
1119 /* ARGSUSED */
1120 static int32_t *
1121 svc_rdma_kgetres(SVCXPRT *clone_xprt, int size)
1122 {
1123 return (NULL);
1124 }
1125
1126 /* ARGSUSED */
1127 static void
1128 svc_rdma_kfreeres(SVCXPRT *clone_xprt)
1129 {
1130 }
1131
1132 /*
1133 * the dup cacheing routines below provide a cache of non-failure
1134 * transaction id's. rpc service routines can use this to detect
1135 * retransmissions and re-send a non-failure response.
1136 */
1137
1138 /*
1139 * MAXDUPREQS is the number of cached items. It should be adjusted
1140 * to the service load so that there is likely to be a response entry
1141 * when the first retransmission comes in.
1142 */
1143 #define MAXDUPREQS 8192
1144
1145 /*
1146 * This should be appropriately scaled to MAXDUPREQS. To produce as less as
1147 * possible collisions it is suggested to set this to a prime.
1148 */
1149 #define DRHASHSZ 2053
1150
1151 #define XIDHASH(xid) ((xid) % DRHASHSZ)
1152 #define DRHASH(dr) XIDHASH((dr)->dr_xid)
1153 #define REQTOXID(req) ((req)->rq_xprt->xp_xid)
1154
1155 static int rdmandupreqs = 0;
1156 int rdmamaxdupreqs = MAXDUPREQS;
1157 static kmutex_t rdmadupreq_lock;
1158 static struct dupreq *rdmadrhashtbl[DRHASHSZ];
1159 static int rdmadrhashstat[DRHASHSZ];
1160
1161 static void unhash(struct dupreq *);
1162
1163 /*
1164 * rdmadrmru points to the head of a circular linked list in lru order.
1165 * rdmadrmru->dr_next == drlru
1166 */
1167 struct dupreq *rdmadrmru;
1168
1169 /*
1170 * svc_rdma_kdup searches the request cache and returns 0 if the
1171 * request is not found in the cache. If it is found, then it
1172 * returns the state of the request (in progress or done) and
1173 * the status or attributes that were part of the original reply.
1174 */
1175 static int
1176 svc_rdma_kdup(struct svc_req *req, caddr_t res, int size, struct dupreq **drpp,
1177 bool_t *dupcachedp)
1178 {
1179 struct dupreq *dr;
1180 uint32_t xid;
1181 uint32_t drhash;
1182 int status;
1183
1184 xid = REQTOXID(req);
1185 mutex_enter(&rdmadupreq_lock);
1186 RSSTAT_INCR(rsdupchecks);
1187 /*
1188 * Check to see whether an entry already exists in the cache.
1189 */
1190 dr = rdmadrhashtbl[XIDHASH(xid)];
1191 while (dr != NULL) {
1192 if (dr->dr_xid == xid &&
1193 dr->dr_proc == req->rq_proc &&
1194 dr->dr_prog == req->rq_prog &&
1195 dr->dr_vers == req->rq_vers &&
1196 dr->dr_addr.len == req->rq_xprt->xp_rtaddr.len &&
1197 bcmp((caddr_t)dr->dr_addr.buf,
1198 (caddr_t)req->rq_xprt->xp_rtaddr.buf,
1199 dr->dr_addr.len) == 0) {
1200 status = dr->dr_status;
1201 if (status == DUP_DONE) {
1202 bcopy(dr->dr_resp.buf, res, size);
1203 if (dupcachedp != NULL)
1204 *dupcachedp = (dr->dr_resfree != NULL);
1205 } else {
1206 dr->dr_status = DUP_INPROGRESS;
1207 *drpp = dr;
1208 }
1209 RSSTAT_INCR(rsdupreqs);
1210 mutex_exit(&rdmadupreq_lock);
1211 return (status);
1212 }
1213 dr = dr->dr_chain;
1214 }
1215
1216 /*
1217 * There wasn't an entry, either allocate a new one or recycle
1218 * an old one.
1219 */
1220 if (rdmandupreqs < rdmamaxdupreqs) {
1221 dr = kmem_alloc(sizeof (*dr), KM_NOSLEEP);
1222 if (dr == NULL) {
1223 mutex_exit(&rdmadupreq_lock);
1224 return (DUP_ERROR);
1225 }
1226 dr->dr_resp.buf = NULL;
1227 dr->dr_resp.maxlen = 0;
1228 dr->dr_addr.buf = NULL;
1229 dr->dr_addr.maxlen = 0;
1230 if (rdmadrmru) {
1231 dr->dr_next = rdmadrmru->dr_next;
1232 rdmadrmru->dr_next = dr;
1233 } else {
1234 dr->dr_next = dr;
1235 }
1236 rdmandupreqs++;
1237 } else {
1238 dr = rdmadrmru->dr_next;
1239 while (dr->dr_status == DUP_INPROGRESS) {
1240 dr = dr->dr_next;
1241 if (dr == rdmadrmru->dr_next) {
1242 mutex_exit(&rdmadupreq_lock);
1243 return (DUP_ERROR);
1244 }
1245 }
1246 unhash(dr);
1247 if (dr->dr_resfree) {
1248 (*dr->dr_resfree)(dr->dr_resp.buf);
1249 }
1250 }
1251 dr->dr_resfree = NULL;
1252 rdmadrmru = dr;
1253
1254 dr->dr_xid = REQTOXID(req);
1255 dr->dr_prog = req->rq_prog;
1256 dr->dr_vers = req->rq_vers;
1257 dr->dr_proc = req->rq_proc;
1258 if (dr->dr_addr.maxlen < req->rq_xprt->xp_rtaddr.len) {
1259 if (dr->dr_addr.buf != NULL)
1260 kmem_free(dr->dr_addr.buf, dr->dr_addr.maxlen);
1261 dr->dr_addr.maxlen = req->rq_xprt->xp_rtaddr.len;
1262 dr->dr_addr.buf = kmem_alloc(dr->dr_addr.maxlen, KM_NOSLEEP);
1263 if (dr->dr_addr.buf == NULL) {
1264 dr->dr_addr.maxlen = 0;
1265 dr->dr_status = DUP_DROP;
1266 mutex_exit(&rdmadupreq_lock);
1267 return (DUP_ERROR);
1268 }
1269 }
1270 dr->dr_addr.len = req->rq_xprt->xp_rtaddr.len;
1271 bcopy(req->rq_xprt->xp_rtaddr.buf, dr->dr_addr.buf, dr->dr_addr.len);
1272 if (dr->dr_resp.maxlen < size) {
1273 if (dr->dr_resp.buf != NULL)
1274 kmem_free(dr->dr_resp.buf, dr->dr_resp.maxlen);
1275 dr->dr_resp.maxlen = (unsigned int)size;
1276 dr->dr_resp.buf = kmem_alloc(size, KM_NOSLEEP);
1277 if (dr->dr_resp.buf == NULL) {
1278 dr->dr_resp.maxlen = 0;
1279 dr->dr_status = DUP_DROP;
1280 mutex_exit(&rdmadupreq_lock);
1281 return (DUP_ERROR);
1282 }
1283 }
1284 dr->dr_status = DUP_INPROGRESS;
1285
1286 drhash = (uint32_t)DRHASH(dr);
1287 dr->dr_chain = rdmadrhashtbl[drhash];
1288 rdmadrhashtbl[drhash] = dr;
1289 rdmadrhashstat[drhash]++;
1290 mutex_exit(&rdmadupreq_lock);
1291 *drpp = dr;
1292 return (DUP_NEW);
1293 }
1294
1295 /*
1296 * svc_rdma_kdupdone marks the request done (DUP_DONE or DUP_DROP)
1297 * and stores the response.
1298 */
1299 static void
1300 svc_rdma_kdupdone(struct dupreq *dr, caddr_t res, void (*dis_resfree)(),
1301 int size, int status)
1302 {
1303 ASSERT(dr->dr_resfree == NULL);
1304 if (status == DUP_DONE) {
1305 bcopy(res, dr->dr_resp.buf, size);
1306 dr->dr_resfree = dis_resfree;
1307 }
1308 dr->dr_status = status;
1309 }
1310
1311 /*
1312 * This routine expects that the mutex, rdmadupreq_lock, is already held.
1313 */
1314 static void
1315 unhash(struct dupreq *dr)
1316 {
1317 struct dupreq *drt;
1318 struct dupreq *drtprev = NULL;
1319 uint32_t drhash;
1320
1321 ASSERT(MUTEX_HELD(&rdmadupreq_lock));
1322
1323 drhash = (uint32_t)DRHASH(dr);
1324 drt = rdmadrhashtbl[drhash];
1325 while (drt != NULL) {
1326 if (drt == dr) {
1327 rdmadrhashstat[drhash]--;
1328 if (drtprev == NULL) {
1329 rdmadrhashtbl[drhash] = drt->dr_chain;
1330 } else {
1331 drtprev->dr_chain = drt->dr_chain;
1332 }
1333 return;
1334 }
1335 drtprev = drt;
1336 drt = drt->dr_chain;
1337 }
1338 }
1339
1340 bool_t
1341 rdma_get_wchunk(struct svc_req *req, iovec_t *iov, struct clist *wlist)
1342 {
1343 struct clist *clist;
1344 uint32_t tlen;
1345
1346 if (req->rq_xprt->xp_type != T_RDMA) {
1347 return (FALSE);
1348 }
1349
1350 tlen = 0;
1351 clist = wlist;
1352 while (clist) {
1353 tlen += clist->c_len;
1354 clist = clist->c_next;
1355 }
1356
1357 /*
1358 * set iov to addr+len of first segment of first wchunk of
1359 * wlist sent by client. krecv() already malloc'd a buffer
1360 * large enough, but registration is deferred until we write
1361 * the buffer back to (NFS) client using RDMA_WRITE.
1362 */
1363 iov->iov_base = (caddr_t)(uintptr_t)wlist->w.c_saddr;
1364 iov->iov_len = tlen;
1365
1366 return (TRUE);
1367 }
1368
1369 /*
1370 * routine to setup the read chunk lists
1371 */
1372
1373 int
1374 rdma_setup_read_chunks(struct clist *wcl, uint32_t count, int *wcl_len)
1375 {
1376 int data_len, avail_len;
1377 uint_t round_len;
1378
1379 data_len = avail_len = 0;
1380
1381 while (wcl != NULL && count > 0) {
1382 if (wcl->c_dmemhandle.mrc_rmr == 0)
1383 break;
1384
1385 if (wcl->c_len < count) {
1386 data_len += wcl->c_len;
1387 avail_len = 0;
1388 } else {
1389 data_len += count;
1390 avail_len = wcl->c_len - count;
1391 wcl->c_len = count;
1392 }
1393 count -= wcl->c_len;
1394
1395 if (count == 0)
1396 break;
1397
1398 wcl = wcl->c_next;
1399 }
1400
1401 /*
1402 * MUST fail if there are still more data
1403 */
1404 if (count > 0) {
1405 DTRACE_PROBE2(krpc__e__rdma_setup_read_chunks_clist_len,
1406 int, data_len, int, count);
1407 return (FALSE);
1408 }
1409
1410 /*
1411 * Round up the last chunk to 4-byte boundary
1412 */
1413 *wcl_len = roundup(data_len, BYTES_PER_XDR_UNIT);
1414 round_len = *wcl_len - data_len;
1415
1416 if (round_len) {
1417
1418 /*
1419 * If there is space in the current chunk,
1420 * add the roundup to the chunk.
1421 */
1422 if (avail_len >= round_len) {
1423 wcl->c_len += round_len;
1424 } else {
1425 /*
1426 * try the next one.
1427 */
1428 wcl = wcl->c_next;
1429 if ((wcl == NULL) || (wcl->c_len < round_len)) {
1430 DTRACE_PROBE1(
1431 krpc__e__rdma_setup_read_chunks_rndup,
1432 int, round_len);
1433 return (FALSE);
1434 }
1435 wcl->c_len = round_len;
1436 }
1437 }
1438
1439 wcl = wcl->c_next;
1440
1441 /*
1442 * Make rest of the chunks 0-len
1443 */
1444
1445 clist_zero_len(wcl);
1446
1447 return (TRUE);
1448 }