xref: /titanic_50/usr/src/uts/common/fs/nfs/nfs4_state.c (revision 2032ea7bd13069ba4884066ed18f83bf12b3e247)
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 2007 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #pragma ident	"%Z%%M%	%I%	%E% SMI"
27 
28 #include <sys/systm.h>
29 #include <sys/kmem.h>
30 #include <sys/cmn_err.h>
31 #include <sys/atomic.h>
32 #include <sys/clconf.h>
33 #include <sys/cladm.h>
34 #include <sys/flock.h>
35 #include <nfs/export.h>
36 #include <nfs/nfs.h>
37 #include <nfs/nfs4.h>
38 #include <nfs/nfssys.h>
39 #include <nfs/lm.h>
40 #include <sys/pathname.h>
41 #include <sys/sdt.h>
42 #include <sys/nvpair.h>
43 
44 
45 extern time_t rfs4_start_time;
46 extern uint_t nfs4_srv_vkey;
47 
48 stateid4 special0 = {
49 	0,
50 	{ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 }
51 };
52 
53 stateid4 special1 = {
54 	0xffffffff,
55 	{
56 		(char)0xff, (char)0xff, (char)0xff, (char)0xff,
57 		(char)0xff, (char)0xff, (char)0xff, (char)0xff,
58 		(char)0xff, (char)0xff, (char)0xff, (char)0xff
59 	}
60 };
61 
62 
63 #define	ISSPECIAL(id)  (stateid4_cmp(id, &special0) || \
64 			stateid4_cmp(id, &special1))
65 
66 /* For embedding the cluster nodeid into our clientid */
67 #define	CLUSTER_NODEID_SHIFT	24
68 #define	CLUSTER_MAX_NODEID	255
69 
70 #ifdef DEBUG
71 int rfs4_debug;
72 #endif
73 
74 static uint32_t rfs4_database_debug = 0x00;
75 
76 static void rfs4_ss_clid_write(rfs4_client_t *cp, char *leaf);
77 static void rfs4_ss_clid_write_one(rfs4_client_t *cp, char *dir, char *leaf);
78 static void rfs4_dss_clear_oldstate(rfs4_servinst_t *sip);
79 static void rfs4_ss_chkclid_sip(rfs4_client_t *cp, rfs4_servinst_t *sip);
80 
81 /*
82  * Couple of simple init/destroy functions for a general waiter
83  */
84 void
85 rfs4_sw_init(rfs4_state_wait_t *swp)
86 {
87 	mutex_init(swp->sw_cv_lock, NULL, MUTEX_DEFAULT, NULL);
88 	cv_init(swp->sw_cv, NULL, CV_DEFAULT, NULL);
89 	swp->sw_active = FALSE;
90 	swp->sw_wait_count = 0;
91 }
92 
93 void
94 rfs4_sw_destroy(rfs4_state_wait_t *swp)
95 {
96 	mutex_destroy(swp->sw_cv_lock);
97 	cv_destroy(swp->sw_cv);
98 }
99 
100 void
101 rfs4_sw_enter(rfs4_state_wait_t *swp)
102 {
103 	mutex_enter(swp->sw_cv_lock);
104 	while (swp->sw_active) {
105 		swp->sw_wait_count++;
106 		cv_wait(swp->sw_cv, swp->sw_cv_lock);
107 		swp->sw_wait_count--;
108 	}
109 	ASSERT(swp->sw_active == FALSE);
110 	swp->sw_active = TRUE;
111 	mutex_exit(swp->sw_cv_lock);
112 }
113 
114 void
115 rfs4_sw_exit(rfs4_state_wait_t *swp)
116 {
117 	mutex_enter(swp->sw_cv_lock);
118 	ASSERT(swp->sw_active == TRUE);
119 	swp->sw_active = FALSE;
120 	if (swp->sw_wait_count != 0)
121 		cv_broadcast(swp->sw_cv);
122 	mutex_exit(swp->sw_cv_lock);
123 }
124 
125 /*
126  * CPR callback id -- not related to v4 callbacks
127  */
128 static callb_id_t cpr_id = 0;
129 
130 static void
131 deep_lock_copy(LOCK4res *dres, LOCK4res *sres)
132 {
133 	lock_owner4 *slo = &sres->LOCK4res_u.denied.owner;
134 	lock_owner4 *dlo = &dres->LOCK4res_u.denied.owner;
135 
136 	if (sres->status == NFS4ERR_DENIED) {
137 		dlo->owner_val = kmem_alloc(slo->owner_len, KM_SLEEP);
138 		bcopy(slo->owner_val, dlo->owner_val, slo->owner_len);
139 	}
140 }
141 
142 static void
143 deep_lock_free(LOCK4res *res)
144 {
145 	lock_owner4 *lo = &res->LOCK4res_u.denied.owner;
146 
147 	if (res->status == NFS4ERR_DENIED)
148 		kmem_free(lo->owner_val, lo->owner_len);
149 }
150 
151 static void
152 deep_open_copy(OPEN4res *dres, OPEN4res *sres)
153 {
154 	nfsace4 *sacep, *dacep;
155 
156 	if (sres->status != NFS4_OK) {
157 		return;
158 	}
159 
160 	dres->attrset = sres->attrset;
161 
162 	switch (sres->delegation.delegation_type) {
163 	case OPEN_DELEGATE_NONE:
164 		return;
165 	case OPEN_DELEGATE_READ:
166 		sacep = &sres->delegation.open_delegation4_u.read.permissions;
167 		dacep = &dres->delegation.open_delegation4_u.read.permissions;
168 		break;
169 	case OPEN_DELEGATE_WRITE:
170 		sacep = &sres->delegation.open_delegation4_u.write.permissions;
171 		dacep = &dres->delegation.open_delegation4_u.write.permissions;
172 		break;
173 	}
174 	dacep->who.utf8string_val =
175 	    kmem_alloc(sacep->who.utf8string_len, KM_SLEEP);
176 	bcopy(sacep->who.utf8string_val, dacep->who.utf8string_val,
177 	    sacep->who.utf8string_len);
178 }
179 
180 static void
181 deep_open_free(OPEN4res *res)
182 {
183 	nfsace4 *acep;
184 	if (res->status != NFS4_OK)
185 		return;
186 
187 	switch (res->delegation.delegation_type) {
188 	case OPEN_DELEGATE_NONE:
189 		return;
190 	case OPEN_DELEGATE_READ:
191 		acep = &res->delegation.open_delegation4_u.read.permissions;
192 		break;
193 	case OPEN_DELEGATE_WRITE:
194 		acep = &res->delegation.open_delegation4_u.write.permissions;
195 		break;
196 	}
197 
198 	if (acep->who.utf8string_val) {
199 		kmem_free(acep->who.utf8string_val, acep->who.utf8string_len);
200 		acep->who.utf8string_val = NULL;
201 	}
202 }
203 
204 void
205 rfs4_free_reply(nfs_resop4 *rp)
206 {
207 	switch (rp->resop) {
208 	case OP_LOCK:
209 		deep_lock_free(&rp->nfs_resop4_u.oplock);
210 		break;
211 	case OP_OPEN:
212 		deep_open_free(&rp->nfs_resop4_u.opopen);
213 	default:
214 		break;
215 	}
216 }
217 
218 void
219 rfs4_copy_reply(nfs_resop4 *dst, nfs_resop4 *src)
220 {
221 	*dst = *src;
222 
223 	/* Handle responses that need deep copy */
224 	switch (src->resop) {
225 	case OP_LOCK:
226 		deep_lock_copy(&dst->nfs_resop4_u.oplock,
227 		    &src->nfs_resop4_u.oplock);
228 		break;
229 	case OP_OPEN:
230 		deep_open_copy(&dst->nfs_resop4_u.opopen,
231 		    &src->nfs_resop4_u.opopen);
232 		break;
233 	default:
234 		break;
235 	};
236 }
237 
238 /*
239  * This is the implementation of the underlying state engine. The
240  * public interface to this engine is described by
241  * nfs4_state.h. Callers to the engine should hold no state engine
242  * locks when they call in to it. If the protocol needs to lock data
243  * structures it should do so after acquiring all references to them
244  * first and then follow the following lock order:
245  *
246  *	client > openowner > state > lo_state > lockowner > file.
247  *
248  * Internally we only allow a thread to hold one hash bucket lock at a
249  * time and the lock is higher in the lock order (must be acquired
250  * first) than the data structure that is on that hash list.
251  *
252  * If a new reference was acquired by the caller, that reference needs
253  * to be released after releasing all acquired locks with the
254  * corresponding rfs4_*_rele routine.
255  */
256 
257 /*
258  * This code is some what prototypical for now. Its purpose currently is to
259  * implement the interfaces sufficiently to finish the higher protocol
260  * elements. This will be replaced by a dynamically resizeable tables
261  * backed by kmem_cache allocator. However synchronization is handled
262  * correctly (I hope) and will not change by much.  The mutexes for
263  * the hash buckets that can be used to create new instances of data
264  * structures  might be good candidates to evolve into reader writer
265  * locks. If it has to do a creation, it would be holding the
266  * mutex across a kmem_alloc with KM_SLEEP specified.
267  */
268 
269 #ifdef DEBUG
270 #define	TABSIZE 17
271 #else
272 #define	TABSIZE 2047
273 #endif
274 
275 #define	ADDRHASH(key) ((unsigned long)(key) >> 3)
276 
277 /* Used to serialize create/destroy of rfs4_server_state database */
278 kmutex_t	rfs4_state_lock;
279 static rfs4_database_t *rfs4_server_state = NULL;
280 
281 /* Used to serialize lookups of clientids */
282 static	krwlock_t	rfs4_findclient_lock;
283 
284 /*
285  * For now this "table" is exposed so that the CPR callback
286  * function can tromp through it..
287  */
288 rfs4_table_t *rfs4_client_tab;
289 
290 static rfs4_index_t *rfs4_clientid_idx;
291 static rfs4_index_t *rfs4_nfsclnt_idx;
292 static rfs4_table_t *rfs4_openowner_tab;
293 static rfs4_index_t *rfs4_openowner_idx;
294 static rfs4_table_t *rfs4_state_tab;
295 static rfs4_index_t *rfs4_state_idx;
296 static rfs4_index_t *rfs4_state_owner_file_idx;
297 static rfs4_index_t *rfs4_state_file_idx;
298 static rfs4_table_t *rfs4_lo_state_tab;
299 static rfs4_index_t *rfs4_lo_state_idx;
300 static rfs4_index_t *rfs4_lo_state_owner_idx;
301 static rfs4_table_t *rfs4_lockowner_tab;
302 static rfs4_index_t *rfs4_lockowner_idx;
303 static rfs4_index_t *rfs4_lockowner_pid_idx;
304 static rfs4_table_t *rfs4_file_tab;
305 static rfs4_index_t *rfs4_file_idx;
306 static rfs4_table_t *rfs4_deleg_state_tab;
307 static rfs4_index_t *rfs4_deleg_idx;
308 static rfs4_index_t *rfs4_deleg_state_idx;
309 
310 #define	MAXTABSZ 1024*1024
311 
312 /* The values below are rfs4_lease_time units */
313 
314 #ifdef DEBUG
315 #define	CLIENT_CACHE_TIME 1
316 #define	OPENOWNER_CACHE_TIME 1
317 #define	STATE_CACHE_TIME 1
318 #define	LO_STATE_CACHE_TIME 1
319 #define	LOCKOWNER_CACHE_TIME 1
320 #define	FILE_CACHE_TIME 3
321 #define	DELEG_STATE_CACHE_TIME 1
322 #else
323 #define	CLIENT_CACHE_TIME 10
324 #define	OPENOWNER_CACHE_TIME 5
325 #define	STATE_CACHE_TIME 1
326 #define	LO_STATE_CACHE_TIME 1
327 #define	LOCKOWNER_CACHE_TIME 3
328 #define	FILE_CACHE_TIME 40
329 #define	DELEG_STATE_CACHE_TIME 1
330 #endif
331 
332 
333 static time_t rfs4_client_cache_time = 0;
334 static time_t rfs4_openowner_cache_time = 0;
335 static time_t rfs4_state_cache_time = 0;
336 static time_t rfs4_lo_state_cache_time = 0;
337 static time_t rfs4_lockowner_cache_time = 0;
338 static time_t rfs4_file_cache_time = 0;
339 static time_t rfs4_deleg_state_cache_time = 0;
340 
341 static bool_t rfs4_client_create(rfs4_entry_t, void *);
342 static void rfs4_dss_remove_cpleaf(rfs4_client_t *);
343 static void rfs4_dss_remove_leaf(rfs4_servinst_t *, char *, char *);
344 static void rfs4_client_destroy(rfs4_entry_t);
345 static bool_t rfs4_client_expiry(rfs4_entry_t);
346 static uint32_t clientid_hash(void *);
347 static bool_t clientid_compare(rfs4_entry_t, void *);
348 static void *clientid_mkkey(rfs4_entry_t);
349 static uint32_t nfsclnt_hash(void *);
350 static bool_t nfsclnt_compare(rfs4_entry_t, void *);
351 static void *nfsclnt_mkkey(rfs4_entry_t);
352 static bool_t rfs4_openowner_create(rfs4_entry_t, void *);
353 static void rfs4_openowner_destroy(rfs4_entry_t);
354 static bool_t rfs4_openowner_expiry(rfs4_entry_t);
355 static uint32_t openowner_hash(void *);
356 static bool_t openowner_compare(rfs4_entry_t, void *);
357 static void *openowner_mkkey(rfs4_entry_t);
358 static bool_t rfs4_state_create(rfs4_entry_t, void *);
359 static void rfs4_state_destroy(rfs4_entry_t);
360 static bool_t rfs4_state_expiry(rfs4_entry_t);
361 static uint32_t state_hash(void *);
362 static bool_t state_compare(rfs4_entry_t, void *);
363 static void *state_mkkey(rfs4_entry_t);
364 static uint32_t state_owner_file_hash(void *);
365 static bool_t state_owner_file_compare(rfs4_entry_t, void *);
366 static void *state_owner_file_mkkey(rfs4_entry_t);
367 static uint32_t state_file_hash(void *);
368 static bool_t state_file_compare(rfs4_entry_t, void *);
369 static void *state_file_mkkey(rfs4_entry_t);
370 static bool_t rfs4_lo_state_create(rfs4_entry_t, void *);
371 static void rfs4_lo_state_destroy(rfs4_entry_t);
372 static bool_t rfs4_lo_state_expiry(rfs4_entry_t);
373 static uint32_t lo_state_hash(void *);
374 static bool_t lo_state_compare(rfs4_entry_t, void *);
375 static void *lo_state_mkkey(rfs4_entry_t);
376 static uint32_t lo_state_lo_hash(void *);
377 static bool_t lo_state_lo_compare(rfs4_entry_t, void *);
378 static void *lo_state_lo_mkkey(rfs4_entry_t);
379 static bool_t rfs4_lockowner_create(rfs4_entry_t, void *);
380 static void rfs4_lockowner_destroy(rfs4_entry_t);
381 static bool_t rfs4_lockowner_expiry(rfs4_entry_t);
382 static uint32_t lockowner_hash(void *);
383 static bool_t lockowner_compare(rfs4_entry_t, void *);
384 static void *lockowner_mkkey(rfs4_entry_t);
385 static uint32_t pid_hash(void *);
386 static bool_t pid_compare(rfs4_entry_t, void *);
387 static void *pid_mkkey(rfs4_entry_t);
388 static bool_t rfs4_file_create(rfs4_entry_t, void *);
389 static void rfs4_file_destroy(rfs4_entry_t);
390 static uint32_t file_hash(void *);
391 static bool_t file_compare(rfs4_entry_t, void *);
392 static void *file_mkkey(rfs4_entry_t);
393 static bool_t rfs4_deleg_state_create(rfs4_entry_t, void *);
394 static void rfs4_deleg_state_destroy(rfs4_entry_t);
395 static bool_t rfs4_deleg_state_expiry(rfs4_entry_t);
396 static uint32_t deleg_hash(void *);
397 static bool_t deleg_compare(rfs4_entry_t, void *);
398 static void *deleg_mkkey(rfs4_entry_t);
399 static uint32_t deleg_state_hash(void *);
400 static bool_t deleg_state_compare(rfs4_entry_t, void *);
401 static void *deleg_state_mkkey(rfs4_entry_t);
402 
403 static void rfs4_state_rele_nounlock(rfs4_state_t *);
404 
405 static int rfs4_ss_enabled = 0;
406 
407 extern void (*rfs4_client_clrst)(struct nfs4clrst_args *);
408 
409 void
410 rfs4_ss_pnfree(rfs4_ss_pn_t *ss_pn)
411 {
412 	kmem_free(ss_pn, sizeof (rfs4_ss_pn_t));
413 }
414 
415 static rfs4_ss_pn_t *
416 rfs4_ss_pnalloc(char *dir, char *leaf)
417 {
418 	rfs4_ss_pn_t *ss_pn;
419 	int 	dir_len, leaf_len;
420 
421 	/*
422 	 * validate we have a resonable path
423 	 * (account for the '/' and trailing null)
424 	 */
425 	if ((dir_len = strlen(dir)) > MAXPATHLEN ||
426 	    (leaf_len = strlen(leaf)) > MAXNAMELEN ||
427 	    (dir_len + leaf_len + 2) > MAXPATHLEN) {
428 		return (NULL);
429 	}
430 
431 	ss_pn = kmem_alloc(sizeof (rfs4_ss_pn_t), KM_SLEEP);
432 
433 	(void) snprintf(ss_pn->pn, MAXPATHLEN, "%s/%s", dir, leaf);
434 	/* Handy pointer to just the leaf name */
435 	ss_pn->leaf = ss_pn->pn + dir_len + 1;
436 	return (ss_pn);
437 }
438 
439 
440 /*
441  * Move the "leaf" filename from "sdir" directory
442  * to the "ddir" directory. Return the pathname of
443  * the destination unless the rename fails in which
444  * case we need to return the source pathname.
445  */
446 static rfs4_ss_pn_t *
447 rfs4_ss_movestate(char *sdir, char *ddir, char *leaf)
448 {
449 	rfs4_ss_pn_t *src, *dst;
450 
451 	if ((src = rfs4_ss_pnalloc(sdir, leaf)) == NULL)
452 		return (NULL);
453 
454 	if ((dst = rfs4_ss_pnalloc(ddir, leaf)) == NULL) {
455 		rfs4_ss_pnfree(src);
456 		return (NULL);
457 	}
458 
459 	/*
460 	 * If the rename fails we shall return the src
461 	 * pathname and free the dst. Otherwise we need
462 	 * to free the src and return the dst pathanme.
463 	 */
464 	if (vn_rename(src->pn, dst->pn, UIO_SYSSPACE)) {
465 		rfs4_ss_pnfree(dst);
466 		return (src);
467 	}
468 	rfs4_ss_pnfree(src);
469 	return (dst);
470 }
471 
472 
473 static rfs4_oldstate_t *
474 rfs4_ss_getstate(vnode_t *dvp, rfs4_ss_pn_t *ss_pn)
475 {
476 	struct uio uio;
477 	struct iovec iov[3];
478 
479 	rfs4_oldstate_t *cl_ss = NULL;
480 	vnode_t *vp;
481 	vattr_t va;
482 	uint_t id_len;
483 	int err, kill_file, file_vers;
484 
485 	if (ss_pn == NULL)
486 		return (NULL);
487 
488 	/*
489 	 * open the state file.
490 	 */
491 	if (vn_open(ss_pn->pn, UIO_SYSSPACE, FREAD, 0, &vp, 0, 0) != 0) {
492 		return (NULL);
493 	}
494 
495 	if (vp->v_type != VREG) {
496 		(void) VOP_CLOSE(vp, FREAD, 1, (offset_t)0, CRED());
497 		VN_RELE(vp);
498 		return (NULL);
499 	}
500 
501 	err = VOP_ACCESS(vp, VREAD, 0, CRED());
502 	if (err) {
503 		/*
504 		 * We don't have read access? better get the heck out.
505 		 */
506 		(void) VOP_CLOSE(vp, FREAD, 1, (offset_t)0, CRED());
507 		VN_RELE(vp);
508 		return (NULL);
509 	}
510 
511 	(void) VOP_RWLOCK(vp, V_WRITELOCK_FALSE, NULL);
512 	/*
513 	 * get the file size to do some basic validation
514 	 */
515 	va.va_mask = AT_SIZE;
516 	err = VOP_GETATTR(vp, &va, 0, CRED());
517 
518 	kill_file = (va.va_size == 0 || va.va_size <
519 	    (NFS4_VERIFIER_SIZE + sizeof (uint_t)+1));
520 
521 	if (err || kill_file) {
522 		VOP_RWUNLOCK(vp, V_WRITELOCK_FALSE, NULL);
523 		(void) VOP_CLOSE(vp, FREAD, 1, (offset_t)0, CRED());
524 		VN_RELE(vp);
525 		if (kill_file) {
526 			(void) VOP_REMOVE(dvp, ss_pn->leaf, CRED());
527 		}
528 		return (NULL);
529 	}
530 
531 	cl_ss = kmem_alloc(sizeof (rfs4_oldstate_t), KM_SLEEP);
532 
533 	/*
534 	 * build iovecs to read in the file_version, verifier and id_len
535 	 */
536 	iov[0].iov_base = (caddr_t)&file_vers;
537 	iov[0].iov_len = sizeof (int);
538 	iov[1].iov_base = (caddr_t)&cl_ss->cl_id4.verifier;
539 	iov[1].iov_len = NFS4_VERIFIER_SIZE;
540 	iov[2].iov_base = (caddr_t)&id_len;
541 	iov[2].iov_len = sizeof (uint_t);
542 
543 	uio.uio_iov = iov;
544 	uio.uio_iovcnt = 3;
545 	uio.uio_segflg = UIO_SYSSPACE;
546 	uio.uio_loffset = 0;
547 	uio.uio_resid = sizeof (int) + NFS4_VERIFIER_SIZE + sizeof (uint_t);
548 
549 	if (err = VOP_READ(vp, &uio, FREAD, CRED(), NULL)) {
550 		VOP_RWUNLOCK(vp, V_WRITELOCK_FALSE, NULL);
551 		(void) VOP_CLOSE(vp, FREAD, 1, (offset_t)0, CRED());
552 		VN_RELE(vp);
553 		kmem_free(cl_ss, sizeof (rfs4_oldstate_t));
554 		return (NULL);
555 	}
556 
557 	/*
558 	 * if the file_version doesn't match or if the
559 	 * id_len is zero or the combination of the verifier,
560 	 * id_len and id_val is bigger than the file we have
561 	 * a problem. If so ditch the file.
562 	 */
563 	kill_file = (file_vers != NFS4_SS_VERSION || id_len == 0 ||
564 	    (id_len + NFS4_VERIFIER_SIZE + sizeof (uint_t)) > va.va_size);
565 
566 	if (err || kill_file) {
567 		VOP_RWUNLOCK(vp, V_WRITELOCK_FALSE, NULL);
568 		(void) VOP_CLOSE(vp, FREAD, 1, (offset_t)0, CRED());
569 		VN_RELE(vp);
570 		kmem_free(cl_ss, sizeof (rfs4_oldstate_t));
571 		if (kill_file) {
572 			(void) VOP_REMOVE(dvp, ss_pn->leaf, CRED());
573 		}
574 		return (NULL);
575 	}
576 
577 	/*
578 	 * now get the client id value
579 	 */
580 	cl_ss->cl_id4.id_val = kmem_alloc(id_len, KM_SLEEP);
581 	iov[0].iov_base = cl_ss->cl_id4.id_val;
582 	iov[0].iov_len = id_len;
583 
584 	uio.uio_iov = iov;
585 	uio.uio_iovcnt = 1;
586 	uio.uio_segflg = UIO_SYSSPACE;
587 	uio.uio_resid = cl_ss->cl_id4.id_len = id_len;
588 
589 	if (err = VOP_READ(vp, &uio, FREAD, CRED(), NULL)) {
590 		VOP_RWUNLOCK(vp, V_WRITELOCK_FALSE, NULL);
591 		(void) VOP_CLOSE(vp, FREAD, 1, (offset_t)0, CRED());
592 		VN_RELE(vp);
593 		kmem_free(cl_ss->cl_id4.id_val, id_len);
594 		kmem_free(cl_ss, sizeof (rfs4_oldstate_t));
595 		return (NULL);
596 	}
597 
598 	VOP_RWUNLOCK(vp, V_WRITELOCK_FALSE, NULL);
599 	(void) VOP_CLOSE(vp, FREAD, 1, (offset_t)0, CRED());
600 	VN_RELE(vp);
601 	return (cl_ss);
602 }
603 
604 #ifdef	nextdp
605 #undef nextdp
606 #endif
607 #define	nextdp(dp)	((struct dirent64 *)((char *)(dp) + (dp)->d_reclen))
608 
609 /*
610  * Add entries from statedir to supplied oldstate list.
611  * Optionally, move all entries from statedir -> destdir.
612  */
613 void
614 rfs4_ss_oldstate(rfs4_oldstate_t *oldstate, char *statedir, char *destdir)
615 {
616 	rfs4_ss_pn_t *ss_pn;
617 	rfs4_oldstate_t *cl_ss = NULL;
618 	char	*dirt = NULL;
619 	int	err, dir_eof = 0, size = 0;
620 	vnode_t *dvp;
621 	struct iovec iov;
622 	struct uio uio;
623 	struct dirent64 *dep;
624 	offset_t dirchunk_offset = 0;
625 
626 	/*
627 	 * open the state directory
628 	 */
629 	if (vn_open(statedir, UIO_SYSSPACE, FREAD, 0, &dvp, 0, 0))
630 		return;
631 
632 	if (dvp->v_type != VDIR || VOP_ACCESS(dvp, VREAD, 0, CRED()))
633 		goto out;
634 
635 	dirt = kmem_alloc(RFS4_SS_DIRSIZE, KM_SLEEP);
636 
637 	/*
638 	 * Get and process the directory entries
639 	 */
640 	while (!dir_eof) {
641 		(void) VOP_RWLOCK(dvp, V_WRITELOCK_FALSE, NULL);
642 		iov.iov_base = dirt;
643 		iov.iov_len = RFS4_SS_DIRSIZE;
644 		uio.uio_iov = &iov;
645 		uio.uio_iovcnt = 1;
646 		uio.uio_segflg = UIO_SYSSPACE;
647 		uio.uio_loffset = dirchunk_offset;
648 		uio.uio_resid = RFS4_SS_DIRSIZE;
649 
650 		err = VOP_READDIR(dvp, &uio, CRED(), &dir_eof);
651 		VOP_RWUNLOCK(dvp, V_WRITELOCK_FALSE, NULL);
652 		if (err)
653 			goto out;
654 
655 		size = RFS4_SS_DIRSIZE - uio.uio_resid;
656 
657 		/*
658 		 * Process all the directory entries in this
659 		 * readdir chunk
660 		 */
661 		for (dep = (struct dirent64 *)dirt; size > 0;
662 		    dep = nextdp(dep)) {
663 
664 			size -= dep->d_reclen;
665 			dirchunk_offset = dep->d_off;
666 
667 			/*
668 			 * Skip '.' and '..'
669 			 */
670 			if (NFS_IS_DOTNAME(dep->d_name))
671 				continue;
672 
673 			ss_pn = rfs4_ss_pnalloc(statedir, dep->d_name);
674 			if (ss_pn == NULL)
675 				continue;
676 
677 			if (cl_ss = rfs4_ss_getstate(dvp, ss_pn)) {
678 				if (destdir != NULL) {
679 					rfs4_ss_pnfree(ss_pn);
680 					cl_ss->ss_pn = rfs4_ss_movestate(
681 					    statedir, destdir, dep->d_name);
682 				} else {
683 					cl_ss->ss_pn = ss_pn;
684 				}
685 				insque(cl_ss, oldstate);
686 			} else {
687 				rfs4_ss_pnfree(ss_pn);
688 			}
689 		}
690 	}
691 
692 out:
693 	(void) VOP_CLOSE(dvp, FREAD, 1, (offset_t)0, CRED());
694 	VN_RELE(dvp);
695 	if (dirt)
696 		kmem_free((caddr_t)dirt, RFS4_SS_DIRSIZE);
697 }
698 
699 static void
700 rfs4_ss_init(void)
701 {
702 	int npaths = 1;
703 	char *default_dss_path = NFS4_DSS_VAR_DIR;
704 
705 	/* read the default stable storage state */
706 	rfs4_dss_readstate(npaths, &default_dss_path);
707 
708 	rfs4_ss_enabled = 1;
709 }
710 
711 static void
712 rfs4_ss_fini(void)
713 {
714 	rfs4_servinst_t *sip;
715 
716 	mutex_enter(&rfs4_servinst_lock);
717 	sip = rfs4_cur_servinst;
718 	while (sip != NULL) {
719 		rfs4_dss_clear_oldstate(sip);
720 		sip = sip->next;
721 	}
722 	mutex_exit(&rfs4_servinst_lock);
723 }
724 
725 /*
726  * Remove all oldstate files referenced by this servinst.
727  */
728 static void
729 rfs4_dss_clear_oldstate(rfs4_servinst_t *sip)
730 {
731 	rfs4_oldstate_t *os_head, *osp;
732 
733 	rw_enter(&sip->oldstate_lock, RW_WRITER);
734 	os_head = sip->oldstate;
735 
736 	if (os_head == NULL)
737 		return;
738 
739 	/* skip dummy entry */
740 	osp = os_head->next;
741 	while (osp != os_head) {
742 		char *leaf = osp->ss_pn->leaf;
743 		rfs4_oldstate_t *os_next;
744 
745 		rfs4_dss_remove_leaf(sip, NFS4_DSS_OLDSTATE_LEAF, leaf);
746 
747 		if (osp->cl_id4.id_val)
748 			kmem_free(osp->cl_id4.id_val, osp->cl_id4.id_len);
749 		if (osp->ss_pn)
750 			kmem_free(osp->ss_pn, sizeof (rfs4_ss_pn_t));
751 
752 		os_next = osp->next;
753 		remque(osp);
754 		kmem_free(osp, sizeof (rfs4_oldstate_t));
755 		osp = os_next;
756 	}
757 
758 	/* free dummy entry */
759 	kmem_free(osp, sizeof (rfs4_oldstate_t));
760 
761 	sip->oldstate = NULL;
762 
763 	rw_exit(&sip->oldstate_lock);
764 }
765 
766 /*
767  * Form the state and oldstate paths, and read in the stable storage files.
768  */
769 void
770 rfs4_dss_readstate(int npaths, char **paths)
771 {
772 	int i;
773 	char *state, *oldstate;
774 
775 	state = kmem_alloc(MAXPATHLEN, KM_SLEEP);
776 	oldstate = kmem_alloc(MAXPATHLEN, KM_SLEEP);
777 
778 	for (i = 0; i < npaths; i++) {
779 		char *path = paths[i];
780 
781 		(void) sprintf(state, "%s/%s", path, NFS4_DSS_STATE_LEAF);
782 		(void) sprintf(oldstate, "%s/%s", path, NFS4_DSS_OLDSTATE_LEAF);
783 
784 		/*
785 		 * Populate the current server instance's oldstate list.
786 		 *
787 		 * 1. Read stable storage data from old state directory,
788 		 *    leaving its contents alone.
789 		 *
790 		 * 2. Read stable storage data from state directory,
791 		 *    and move the latter's contents to old state
792 		 *    directory.
793 		 */
794 		rfs4_ss_oldstate(rfs4_cur_servinst->oldstate, oldstate, NULL);
795 		rfs4_ss_oldstate(rfs4_cur_servinst->oldstate, state, oldstate);
796 	}
797 
798 	kmem_free(state, MAXPATHLEN);
799 	kmem_free(oldstate, MAXPATHLEN);
800 }
801 
802 
803 /*
804  * Check if we are still in grace and if the client can be
805  * granted permission to perform reclaims.
806  */
807 void
808 rfs4_ss_chkclid(rfs4_client_t *cp)
809 {
810 	rfs4_servinst_t *sip;
811 
812 	/*
813 	 * It should be sufficient to check the oldstate data for just
814 	 * this client's instance. However, since our per-instance
815 	 * client grouping is solely temporal, HA-NFSv4 RG failover
816 	 * might result in clients of the same RG being partitioned into
817 	 * separate instances.
818 	 *
819 	 * Until the client grouping is improved, we must check the
820 	 * oldstate data for all instances with an active grace period.
821 	 *
822 	 * This also serves as the mechanism to remove stale oldstate data.
823 	 * The first time we check an instance after its grace period has
824 	 * expired, the oldstate data should be cleared.
825 	 *
826 	 * Start at the current instance, and walk the list backwards
827 	 * to the first.
828 	 */
829 	mutex_enter(&rfs4_servinst_lock);
830 	for (sip = rfs4_cur_servinst; sip != NULL; sip = sip->prev) {
831 		rfs4_ss_chkclid_sip(cp, sip);
832 
833 		/* if the above check found this client, we're done */
834 		if (cp->can_reclaim)
835 			break;
836 	}
837 	mutex_exit(&rfs4_servinst_lock);
838 }
839 
840 static void
841 rfs4_ss_chkclid_sip(rfs4_client_t *cp, rfs4_servinst_t *sip)
842 {
843 	rfs4_oldstate_t *osp, *os_head;
844 
845 	/* short circuit everything if this server instance has no oldstate */
846 	rw_enter(&sip->oldstate_lock, RW_READER);
847 	os_head = sip->oldstate;
848 	rw_exit(&sip->oldstate_lock);
849 	if (os_head == NULL)
850 		return;
851 
852 	/*
853 	 * If this server instance is no longer in a grace period then
854 	 * the client won't be able to reclaim. No further need for this
855 	 * instance's oldstate data, so it can be cleared.
856 	 */
857 	if (!rfs4_servinst_in_grace(sip))
858 		return;
859 
860 	/* this instance is still in grace; search for the clientid */
861 
862 	rw_enter(&sip->oldstate_lock, RW_READER);
863 
864 	os_head = sip->oldstate;
865 	/* skip dummy entry */
866 	osp = os_head->next;
867 	while (osp != os_head) {
868 		if (osp->cl_id4.id_len == cp->nfs_client.id_len) {
869 			if (bcmp(osp->cl_id4.id_val, cp->nfs_client.id_val,
870 			    osp->cl_id4.id_len) == 0) {
871 				cp->can_reclaim = 1;
872 				break;
873 			}
874 		}
875 		osp = osp->next;
876 	}
877 
878 	rw_exit(&sip->oldstate_lock);
879 }
880 
881 /*
882  * Place client information into stable storage: 1/3.
883  * First, generate the leaf filename, from the client's IP address and
884  * the server-generated short-hand clientid.
885  */
886 void
887 rfs4_ss_clid(rfs4_client_t *cp, struct svc_req *req)
888 {
889 	const char *kinet_ntop6(uchar_t *, char *, size_t);
890 	char leaf[MAXNAMELEN], buf[INET6_ADDRSTRLEN];
891 	struct sockaddr *ca;
892 	uchar_t *b;
893 
894 	if (rfs4_ss_enabled == 0) {
895 		return;
896 	}
897 
898 	buf[0] = 0;
899 
900 
901 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
902 	if (ca == NULL) {
903 		return;
904 	}
905 
906 	/*
907 	 * Convert the caller's IP address to a dotted string
908 	 */
909 	if (ca->sa_family == AF_INET) {
910 
911 		bcopy(svc_getrpccaller(req->rq_xprt)->buf, &cp->cl_addr,
912 		    sizeof (struct sockaddr_in));
913 		b = (uchar_t *)&((struct sockaddr_in *)ca)->sin_addr;
914 		(void) sprintf(buf, "%03d.%03d.%03d.%03d", b[0] & 0xFF,
915 		    b[1] & 0xFF, b[2] & 0xFF, b[3] & 0xFF);
916 	} else if (ca->sa_family == AF_INET6) {
917 		struct sockaddr_in6 *sin6;
918 
919 		sin6 = (struct sockaddr_in6 *)ca;
920 		bcopy(svc_getrpccaller(req->rq_xprt)->buf, &cp->cl_addr,
921 		    sizeof (struct sockaddr_in6));
922 		(void) kinet_ntop6((uchar_t *)&sin6->sin6_addr,
923 		    buf, INET6_ADDRSTRLEN);
924 	}
925 
926 	(void) snprintf(leaf, MAXNAMELEN, "%s-%llx", buf,
927 	    (longlong_t)cp->clientid);
928 	rfs4_ss_clid_write(cp, leaf);
929 }
930 
931 /*
932  * Place client information into stable storage: 2/3.
933  * DSS: distributed stable storage: the file may need to be written to
934  * multiple directories.
935  */
936 static void
937 rfs4_ss_clid_write(rfs4_client_t *cp, char *leaf)
938 {
939 	rfs4_servinst_t *sip;
940 
941 	/*
942 	 * It should be sufficient to write the leaf file to (all) DSS paths
943 	 * associated with just this client's instance. However, since our
944 	 * per-instance client grouping is solely temporal, HA-NFSv4 RG
945 	 * failover might result in us losing DSS data.
946 	 *
947 	 * Until the client grouping is improved, we must write the DSS data
948 	 * to all instances' paths. Start at the current instance, and
949 	 * walk the list backwards to the first.
950 	 */
951 	mutex_enter(&rfs4_servinst_lock);
952 	for (sip = rfs4_cur_servinst; sip != NULL; sip = sip->prev) {
953 		int i, npaths = sip->dss_npaths;
954 
955 		/* write the leaf file to all DSS paths */
956 		for (i = 0; i < npaths; i++) {
957 			rfs4_dss_path_t *dss_path = sip->dss_paths[i];
958 
959 			/* HA-NFSv4 path might have been failed-away from us */
960 			if (dss_path == NULL)
961 				continue;
962 
963 			rfs4_ss_clid_write_one(cp, dss_path->path, leaf);
964 		}
965 	}
966 	mutex_exit(&rfs4_servinst_lock);
967 }
968 
969 /*
970  * Place client information into stable storage: 3/3.
971  * Write the stable storage data to the requested file.
972  */
973 static void
974 rfs4_ss_clid_write_one(rfs4_client_t *cp, char *dss_path, char *leaf)
975 {
976 	int ioflag;
977 	int file_vers = NFS4_SS_VERSION;
978 	size_t dirlen;
979 	struct uio uio;
980 	struct iovec iov[4];
981 	char *dir;
982 	rfs4_ss_pn_t *ss_pn;
983 	vnode_t *vp;
984 	nfs_client_id4 *cl_id4 = &(cp->nfs_client);
985 
986 	/* allow 2 extra bytes for '/' & NUL */
987 	dirlen = strlen(dss_path) + strlen(NFS4_DSS_STATE_LEAF) + 2;
988 	dir = kmem_alloc(dirlen, KM_SLEEP);
989 	(void) sprintf(dir, "%s/%s", dss_path, NFS4_DSS_STATE_LEAF);
990 
991 	ss_pn = rfs4_ss_pnalloc(dir, leaf);
992 	/* rfs4_ss_pnalloc takes its own copy */
993 	kmem_free(dir, dirlen);
994 	if (ss_pn == NULL)
995 		return;
996 
997 	if (vn_open(ss_pn->pn, UIO_SYSSPACE, FCREAT|FWRITE, 0600, &vp,
998 	    CRCREAT, 0)) {
999 		rfs4_ss_pnfree(ss_pn);
1000 		return;
1001 	}
1002 
1003 	/*
1004 	 * We need to record leaf - i.e. the filename - so that we know
1005 	 * what to remove, in the future. However, the dir part of cp->ss_pn
1006 	 * should never be referenced directly, since it's potentially only
1007 	 * one of several paths with this leaf in it.
1008 	 */
1009 	if (cp->ss_pn != NULL) {
1010 		if (strcmp(cp->ss_pn->leaf, leaf) == 0) {
1011 			/* we've already recorded *this* leaf */
1012 			rfs4_ss_pnfree(ss_pn);
1013 		} else {
1014 			/* replace with this leaf */
1015 			rfs4_ss_pnfree(cp->ss_pn);
1016 			cp->ss_pn = ss_pn;
1017 		}
1018 	} else {
1019 		cp->ss_pn = ss_pn;
1020 	}
1021 
1022 	/*
1023 	 * Build a scatter list that points to the nfs_client_id4
1024 	 */
1025 	iov[0].iov_base = (caddr_t)&file_vers;
1026 	iov[0].iov_len = sizeof (int);
1027 	iov[1].iov_base = (caddr_t)&(cl_id4->verifier);
1028 	iov[1].iov_len = NFS4_VERIFIER_SIZE;
1029 	iov[2].iov_base = (caddr_t)&(cl_id4->id_len);
1030 	iov[2].iov_len = sizeof (uint_t);
1031 	iov[3].iov_base = (caddr_t)cl_id4->id_val;
1032 	iov[3].iov_len = cl_id4->id_len;
1033 
1034 	uio.uio_iov = iov;
1035 	uio.uio_iovcnt = 4;
1036 	uio.uio_loffset = 0;
1037 	uio.uio_segflg = UIO_SYSSPACE;
1038 	uio.uio_llimit = (rlim64_t)MAXOFFSET_T;
1039 	uio.uio_resid = cl_id4->id_len + sizeof (int) +
1040 	    NFS4_VERIFIER_SIZE + sizeof (uint_t);
1041 
1042 	ioflag = uio.uio_fmode = (FWRITE|FSYNC);
1043 	uio.uio_extflg = UIO_COPY_DEFAULT;
1044 
1045 	(void) VOP_RWLOCK(vp, V_WRITELOCK_TRUE, NULL);
1046 	/* write the full client id to the file. */
1047 	(void) VOP_WRITE(vp, &uio, ioflag, CRED(), NULL);
1048 	VOP_RWUNLOCK(vp, V_WRITELOCK_TRUE, NULL);
1049 
1050 	(void) VOP_CLOSE(vp, FWRITE, 1, (offset_t)0, CRED());
1051 	VN_RELE(vp);
1052 }
1053 
1054 /*
1055  * DSS: distributed stable storage.
1056  * Unpack the list of paths passed by nfsd.
1057  * Use nvlist_alloc(9F) to manage the data.
1058  * The caller is responsible for allocating and freeing the buffer.
1059  */
1060 int
1061 rfs4_dss_setpaths(char *buf, size_t buflen)
1062 {
1063 	int error;
1064 
1065 	/*
1066 	 * If this is a "warm start", i.e. we previously had DSS paths,
1067 	 * preserve the old paths.
1068 	 */
1069 	if (rfs4_dss_paths != NULL) {
1070 		/*
1071 		 * Before we lose the ptr, destroy the nvlist and pathnames
1072 		 * array from the warm start before this one.
1073 		 */
1074 		if (rfs4_dss_oldpaths)
1075 			nvlist_free(rfs4_dss_oldpaths);
1076 		rfs4_dss_oldpaths = rfs4_dss_paths;
1077 	}
1078 
1079 	/* unpack the buffer into a searchable nvlist */
1080 	error = nvlist_unpack(buf, buflen, &rfs4_dss_paths, KM_SLEEP);
1081 	if (error)
1082 		return (error);
1083 
1084 	/*
1085 	 * Search the nvlist for the pathnames nvpair (which is the only nvpair
1086 	 * in the list, and record its location.
1087 	 */
1088 	error = nvlist_lookup_string_array(rfs4_dss_paths, NFS4_DSS_NVPAIR_NAME,
1089 	    &rfs4_dss_newpaths, &rfs4_dss_numnewpaths);
1090 	return (error);
1091 }
1092 
1093 /*
1094  * Ultimately the nfssys() call NFS4_CLR_STATE endsup here
1095  * to find and mark the client for forced expire.
1096  */
1097 static void
1098 rfs4_client_scrub(rfs4_entry_t ent, void *arg)
1099 {
1100 	rfs4_client_t *cp = (rfs4_client_t *)ent;
1101 	struct nfs4clrst_args *clr = arg;
1102 	struct sockaddr_in6 *ent_sin6;
1103 	struct in6_addr  clr_in6;
1104 	struct sockaddr_in  *ent_sin;
1105 	struct in_addr   clr_in;
1106 
1107 	if (clr->addr_type != cp->cl_addr.ss_family) {
1108 		return;
1109 	}
1110 
1111 	switch (clr->addr_type) {
1112 
1113 	case AF_INET6:
1114 		/* copyin the address from user space */
1115 		if (copyin(clr->ap, &clr_in6, sizeof (clr_in6))) {
1116 			break;
1117 		}
1118 
1119 		ent_sin6 = (struct sockaddr_in6 *)&cp->cl_addr;
1120 
1121 		/*
1122 		 * now compare, and if equivalent mark entry
1123 		 * for forced expiration
1124 		 */
1125 		if (IN6_ARE_ADDR_EQUAL(&ent_sin6->sin6_addr, &clr_in6)) {
1126 			cp->forced_expire = 1;
1127 		}
1128 		break;
1129 
1130 	case AF_INET:
1131 		/* copyin the address from user space */
1132 		if (copyin(clr->ap, &clr_in, sizeof (clr_in))) {
1133 			break;
1134 		}
1135 
1136 		ent_sin = (struct sockaddr_in *)&cp->cl_addr;
1137 
1138 		/*
1139 		 * now compare, and if equivalent mark entry
1140 		 * for forced expiration
1141 		 */
1142 		if (ent_sin->sin_addr.s_addr == clr_in.s_addr) {
1143 			cp->forced_expire = 1;
1144 		}
1145 		break;
1146 
1147 	default:
1148 		/* force this assert to fail */
1149 		ASSERT(clr->addr_type != clr->addr_type);
1150 	}
1151 }
1152 
1153 /*
1154  * This is called from nfssys() in order to clear server state
1155  * for the specified client IP Address.
1156  */
1157 void
1158 rfs4_clear_client_state(struct nfs4clrst_args *clr)
1159 {
1160 	(void) rfs4_dbe_walk(rfs4_client_tab, rfs4_client_scrub, clr);
1161 }
1162 
1163 /*
1164  * Used to initialize the NFSv4 server's state or database.  All of
1165  * the tables are created and timers are set. Only called when NFSv4
1166  * service is provided.
1167  */
1168 void
1169 rfs4_state_init()
1170 {
1171 	int start_grace;
1172 	extern boolean_t rfs4_cpr_callb(void *, int);
1173 	char *dss_path = NFS4_DSS_VAR_DIR;
1174 
1175 	mutex_enter(&rfs4_state_lock);
1176 
1177 	/*
1178 	 * If the server state database has already been initialized,
1179 	 * skip it
1180 	 */
1181 	if (rfs4_server_state != NULL) {
1182 		mutex_exit(&rfs4_state_lock);
1183 		return;
1184 	}
1185 
1186 	rw_init(&rfs4_findclient_lock, NULL, RW_DEFAULT, NULL);
1187 
1188 	/*
1189 	 * Set the boot time.  If the server
1190 	 * has been restarted quickly and has had the opportunity to
1191 	 * service clients, then the start_time needs to be bumped
1192 	 * regardless.  A small window but it exists...
1193 	 */
1194 	if (rfs4_start_time != gethrestime_sec())
1195 		rfs4_start_time = gethrestime_sec();
1196 	else
1197 		rfs4_start_time++;
1198 
1199 	/* DSS: distributed stable storage: initialise served paths list */
1200 	rfs4_dss_pathlist = NULL;
1201 
1202 	/*
1203 	 * Create the first server instance, or a new one if the server has
1204 	 * been restarted; see above comments on rfs4_start_time. Don't
1205 	 * start its grace period; that will be done later, to maximise the
1206 	 * clients' recovery window.
1207 	 */
1208 	start_grace = 0;
1209 	rfs4_servinst_create(start_grace, 1, &dss_path);
1210 
1211 	/* reset the "first NFSv4 request" status */
1212 	rfs4_seen_first_compound = 0;
1213 
1214 	/*
1215 	 * Add a CPR callback so that we can update client
1216 	 * access times to extend the lease after a suspend
1217 	 * and resume (using the same class as rpcmod/connmgr)
1218 	 */
1219 	cpr_id = callb_add(rfs4_cpr_callb, 0, CB_CL_CPR_RPC, "rfs4");
1220 
1221 	/* set the various cache timers for table creation */
1222 	if (rfs4_client_cache_time == 0)
1223 		rfs4_client_cache_time = CLIENT_CACHE_TIME;
1224 	if (rfs4_openowner_cache_time == 0)
1225 		rfs4_openowner_cache_time = OPENOWNER_CACHE_TIME;
1226 	if (rfs4_state_cache_time == 0)
1227 		rfs4_state_cache_time = STATE_CACHE_TIME;
1228 	if (rfs4_lo_state_cache_time == 0)
1229 		rfs4_lo_state_cache_time = LO_STATE_CACHE_TIME;
1230 	if (rfs4_lockowner_cache_time == 0)
1231 		rfs4_lockowner_cache_time = LOCKOWNER_CACHE_TIME;
1232 	if (rfs4_file_cache_time == 0)
1233 		rfs4_file_cache_time = FILE_CACHE_TIME;
1234 	if (rfs4_deleg_state_cache_time == 0)
1235 		rfs4_deleg_state_cache_time = DELEG_STATE_CACHE_TIME;
1236 
1237 	/* Create the overall database to hold all server state */
1238 	rfs4_server_state = rfs4_database_create(rfs4_database_debug);
1239 
1240 	/* Now create the individual tables */
1241 	rfs4_client_cache_time *= rfs4_lease_time;
1242 	rfs4_client_tab = rfs4_table_create(rfs4_server_state,
1243 	    "Client",
1244 	    rfs4_client_cache_time,
1245 	    2,
1246 	    rfs4_client_create,
1247 	    rfs4_client_destroy,
1248 	    rfs4_client_expiry,
1249 	    sizeof (rfs4_client_t),
1250 	    TABSIZE,
1251 	    MAXTABSZ/8, 100);
1252 	rfs4_nfsclnt_idx = rfs4_index_create(rfs4_client_tab,
1253 	    "nfs_client_id4", nfsclnt_hash,
1254 	    nfsclnt_compare, nfsclnt_mkkey,
1255 	    TRUE);
1256 	rfs4_clientid_idx = rfs4_index_create(rfs4_client_tab,
1257 	    "client_id", clientid_hash,
1258 	    clientid_compare, clientid_mkkey,
1259 	    FALSE);
1260 
1261 	rfs4_openowner_cache_time *= rfs4_lease_time;
1262 	rfs4_openowner_tab = rfs4_table_create(rfs4_server_state,
1263 	    "OpenOwner",
1264 	    rfs4_openowner_cache_time,
1265 	    1,
1266 	    rfs4_openowner_create,
1267 	    rfs4_openowner_destroy,
1268 	    rfs4_openowner_expiry,
1269 	    sizeof (rfs4_openowner_t),
1270 	    TABSIZE,
1271 	    MAXTABSZ, 100);
1272 	rfs4_openowner_idx = rfs4_index_create(rfs4_openowner_tab,
1273 	    "open_owner4", openowner_hash,
1274 	    openowner_compare,
1275 	    openowner_mkkey, TRUE);
1276 
1277 	rfs4_state_cache_time *= rfs4_lease_time;
1278 	rfs4_state_tab = rfs4_table_create(rfs4_server_state,
1279 	    "OpenStateID",
1280 	    rfs4_state_cache_time,
1281 	    3,
1282 	    rfs4_state_create,
1283 	    rfs4_state_destroy,
1284 	    rfs4_state_expiry,
1285 	    sizeof (rfs4_state_t),
1286 	    TABSIZE,
1287 	    MAXTABSZ, 100);
1288 
1289 	rfs4_state_owner_file_idx = rfs4_index_create(rfs4_state_tab,
1290 	    "Openowner-File",
1291 	    state_owner_file_hash,
1292 	    state_owner_file_compare,
1293 	    state_owner_file_mkkey, TRUE);
1294 
1295 	rfs4_state_idx = rfs4_index_create(rfs4_state_tab,
1296 	    "State-id", state_hash,
1297 	    state_compare, state_mkkey, FALSE);
1298 
1299 	rfs4_state_file_idx = rfs4_index_create(rfs4_state_tab,
1300 	    "File", state_file_hash,
1301 	    state_file_compare, state_file_mkkey,
1302 	    FALSE);
1303 
1304 	rfs4_lo_state_cache_time *= rfs4_lease_time;
1305 	rfs4_lo_state_tab = rfs4_table_create(rfs4_server_state,
1306 	    "LockStateID",
1307 	    rfs4_lo_state_cache_time,
1308 	    2,
1309 	    rfs4_lo_state_create,
1310 	    rfs4_lo_state_destroy,
1311 	    rfs4_lo_state_expiry,
1312 	    sizeof (rfs4_lo_state_t),
1313 	    TABSIZE,
1314 	    MAXTABSZ, 100);
1315 
1316 	rfs4_lo_state_owner_idx = rfs4_index_create(rfs4_lo_state_tab,
1317 	    "lockownerxstate",
1318 	    lo_state_lo_hash,
1319 	    lo_state_lo_compare,
1320 	    lo_state_lo_mkkey, TRUE);
1321 
1322 	rfs4_lo_state_idx = rfs4_index_create(rfs4_lo_state_tab,
1323 	    "State-id",
1324 	    lo_state_hash, lo_state_compare,
1325 	    lo_state_mkkey, FALSE);
1326 
1327 	rfs4_lockowner_cache_time *= rfs4_lease_time;
1328 
1329 	rfs4_lockowner_tab = rfs4_table_create(rfs4_server_state,
1330 	    "Lockowner",
1331 	    rfs4_lockowner_cache_time,
1332 	    2,
1333 	    rfs4_lockowner_create,
1334 	    rfs4_lockowner_destroy,
1335 	    rfs4_lockowner_expiry,
1336 	    sizeof (rfs4_lockowner_t),
1337 	    TABSIZE,
1338 	    MAXTABSZ, 100);
1339 
1340 	rfs4_lockowner_idx = rfs4_index_create(rfs4_lockowner_tab,
1341 	    "lock_owner4", lockowner_hash,
1342 	    lockowner_compare,
1343 	    lockowner_mkkey, TRUE);
1344 
1345 	rfs4_lockowner_pid_idx = rfs4_index_create(rfs4_lockowner_tab,
1346 	    "pid", pid_hash,
1347 	    pid_compare, pid_mkkey,
1348 	    FALSE);
1349 
1350 	rfs4_file_cache_time *= rfs4_lease_time;
1351 	rfs4_file_tab = rfs4_table_create(rfs4_server_state,
1352 	    "File",
1353 	    rfs4_file_cache_time,
1354 	    1,
1355 	    rfs4_file_create,
1356 	    rfs4_file_destroy,
1357 	    NULL,
1358 	    sizeof (rfs4_file_t),
1359 	    TABSIZE,
1360 	    MAXTABSZ, -1);
1361 
1362 	rfs4_file_idx = rfs4_index_create(rfs4_file_tab,
1363 	    "Filehandle", file_hash,
1364 	    file_compare, file_mkkey, TRUE);
1365 
1366 	rfs4_deleg_state_cache_time *= rfs4_lease_time;
1367 	rfs4_deleg_state_tab = rfs4_table_create(rfs4_server_state,
1368 	    "DelegStateID",
1369 	    rfs4_deleg_state_cache_time,
1370 	    2,
1371 	    rfs4_deleg_state_create,
1372 	    rfs4_deleg_state_destroy,
1373 	    rfs4_deleg_state_expiry,
1374 	    sizeof (rfs4_deleg_state_t),
1375 	    TABSIZE,
1376 	    MAXTABSZ, 100);
1377 	rfs4_deleg_idx = rfs4_index_create(rfs4_deleg_state_tab,
1378 	    "DelegByFileClient",
1379 	    deleg_hash,
1380 	    deleg_compare,
1381 	    deleg_mkkey, TRUE);
1382 
1383 	rfs4_deleg_state_idx = rfs4_index_create(rfs4_deleg_state_tab,
1384 	    "DelegState",
1385 	    deleg_state_hash,
1386 	    deleg_state_compare,
1387 	    deleg_state_mkkey, FALSE);
1388 
1389 	/*
1390 	 * Init the stable storage.
1391 	 */
1392 	rfs4_ss_init();
1393 
1394 	rfs4_client_clrst = rfs4_clear_client_state;
1395 
1396 	mutex_exit(&rfs4_state_lock);
1397 }
1398 
1399 
1400 /*
1401  * Used at server shutdown to cleanup all of the NFSv4 server's structures
1402  * and other state.
1403  */
1404 void
1405 rfs4_state_fini()
1406 {
1407 	rfs4_database_t *dbp;
1408 
1409 	mutex_enter(&rfs4_state_lock);
1410 
1411 	if (rfs4_server_state == NULL) {
1412 		mutex_exit(&rfs4_state_lock);
1413 		return;
1414 	}
1415 
1416 	rfs4_client_clrst = NULL;
1417 
1418 	rfs4_set_deleg_policy(SRV_NEVER_DELEGATE);
1419 	dbp = rfs4_server_state;
1420 	rfs4_server_state = NULL;
1421 
1422 	/*
1423 	 * Cleanup the CPR callback.
1424 	 */
1425 	if (cpr_id)
1426 		(void) callb_delete(cpr_id);
1427 
1428 	rw_destroy(&rfs4_findclient_lock);
1429 
1430 	/* First stop all of the reaper threads in the database */
1431 	rfs4_database_shutdown(dbp);
1432 	/* clean up any dangling stable storage structures */
1433 	rfs4_ss_fini();
1434 	/* Now actually destroy/release the database and its tables */
1435 	rfs4_database_destroy(dbp);
1436 
1437 	/* Reset the cache timers for next time */
1438 	rfs4_client_cache_time = 0;
1439 	rfs4_openowner_cache_time = 0;
1440 	rfs4_state_cache_time = 0;
1441 	rfs4_lo_state_cache_time = 0;
1442 	rfs4_lockowner_cache_time = 0;
1443 	rfs4_file_cache_time = 0;
1444 	rfs4_deleg_state_cache_time = 0;
1445 
1446 	mutex_exit(&rfs4_state_lock);
1447 
1448 	/* destroy server instances and current instance ptr */
1449 	rfs4_servinst_destroy_all();
1450 
1451 	/* reset the "first NFSv4 request" status */
1452 	rfs4_seen_first_compound = 0;
1453 
1454 	/* DSS: distributed stable storage */
1455 	if (rfs4_dss_oldpaths)
1456 		nvlist_free(rfs4_dss_oldpaths);
1457 	if (rfs4_dss_paths)
1458 		nvlist_free(rfs4_dss_paths);
1459 	rfs4_dss_paths = rfs4_dss_oldpaths = NULL;
1460 }
1461 
1462 typedef union {
1463 	struct {
1464 		uint32_t start_time;
1465 		uint32_t c_id;
1466 	} impl_id;
1467 	clientid4 id4;
1468 } cid;
1469 
1470 static int foreign_stateid(stateid_t *id);
1471 static int foreign_clientid(cid *cidp);
1472 static void embed_nodeid(cid *cidp);
1473 
1474 typedef union {
1475 	struct {
1476 		uint32_t c_id;
1477 		uint32_t gen_num;
1478 	} cv_impl;
1479 	verifier4	confirm_verf;
1480 } scid_confirm_verf;
1481 
1482 static uint32_t
1483 clientid_hash(void *key)
1484 {
1485 	cid *idp = key;
1486 
1487 	return (idp->impl_id.c_id);
1488 }
1489 
1490 static bool_t
1491 clientid_compare(rfs4_entry_t entry, void *key)
1492 {
1493 	rfs4_client_t *client = (rfs4_client_t *)entry;
1494 	clientid4 *idp = key;
1495 
1496 	return (*idp == client->clientid);
1497 }
1498 
1499 static void *
1500 clientid_mkkey(rfs4_entry_t entry)
1501 {
1502 	rfs4_client_t *client = (rfs4_client_t *)entry;
1503 
1504 	return (&client->clientid);
1505 }
1506 
1507 static uint32_t
1508 nfsclnt_hash(void *key)
1509 {
1510 	nfs_client_id4 *client = key;
1511 	int i;
1512 	uint32_t hash = 0;
1513 
1514 	for (i = 0; i < client->id_len; i++) {
1515 		hash <<= 1;
1516 		hash += (uint_t)client->id_val[i];
1517 	}
1518 	return (hash);
1519 }
1520 
1521 
1522 static bool_t
1523 nfsclnt_compare(rfs4_entry_t entry, void *key)
1524 {
1525 	rfs4_client_t *client = (rfs4_client_t *)entry;
1526 	nfs_client_id4 *nfs_client = key;
1527 
1528 	if (client->nfs_client.id_len != nfs_client->id_len)
1529 		return (FALSE);
1530 
1531 	return (bcmp(client->nfs_client.id_val, nfs_client->id_val,
1532 	    nfs_client->id_len) == 0);
1533 }
1534 
1535 static void *
1536 nfsclnt_mkkey(rfs4_entry_t entry)
1537 {
1538 	rfs4_client_t *client = (rfs4_client_t *)entry;
1539 
1540 	return (&client->nfs_client);
1541 }
1542 
1543 static bool_t
1544 rfs4_client_expiry(rfs4_entry_t u_entry)
1545 {
1546 	rfs4_client_t *cp = (rfs4_client_t *)u_entry;
1547 	bool_t cp_expired;
1548 
1549 	if (rfs4_dbe_is_invalid(cp->dbe))
1550 		return (TRUE);
1551 	/*
1552 	 * If the sysadmin has used clear_locks for this
1553 	 * entry then forced_expire will be set and we
1554 	 * want this entry to be reaped. Or the entry
1555 	 * has exceeded its lease period.
1556 	 */
1557 	cp_expired = (cp->forced_expire ||
1558 	    (gethrestime_sec() - cp->last_access
1559 	    > rfs4_lease_time));
1560 
1561 	if (!cp->ss_remove && cp_expired)
1562 		cp->ss_remove = 1;
1563 	return (cp_expired);
1564 }
1565 
1566 /*
1567  * Remove the leaf file from all distributed stable storage paths.
1568  */
1569 static void
1570 rfs4_dss_remove_cpleaf(rfs4_client_t *cp)
1571 {
1572 	char *leaf = cp->ss_pn->leaf;
1573 
1574 	rfs4_dss_remove_leaf(cp->server_instance, NFS4_DSS_STATE_LEAF, leaf);
1575 }
1576 
1577 static void
1578 rfs4_dss_remove_leaf(rfs4_servinst_t *sip, char *dir_leaf, char *leaf)
1579 {
1580 	int i, npaths = sip->dss_npaths;
1581 
1582 	for (i = 0; i < npaths; i++) {
1583 		rfs4_dss_path_t *dss_path = sip->dss_paths[i];
1584 		char *path, *dir;
1585 		size_t pathlen;
1586 
1587 		/* the HA-NFSv4 path might have been failed-over away from us */
1588 		if (dss_path == NULL)
1589 			continue;
1590 
1591 		dir = dss_path->path;
1592 
1593 		/* allow 3 extra bytes for two '/' & a NUL */
1594 		pathlen = strlen(dir) + strlen(dir_leaf) + strlen(leaf) + 3;
1595 		path = kmem_alloc(pathlen, KM_SLEEP);
1596 		(void) sprintf(path, "%s/%s/%s", dir, dir_leaf, leaf);
1597 
1598 		(void) vn_remove(path, UIO_SYSSPACE, RMFILE);
1599 
1600 		kmem_free(path, pathlen);
1601 	}
1602 }
1603 
1604 static void
1605 rfs4_client_destroy(rfs4_entry_t u_entry)
1606 {
1607 	rfs4_client_t *cp = (rfs4_client_t *)u_entry;
1608 
1609 	mutex_destroy(cp->cbinfo.cb_lock);
1610 	cv_destroy(cp->cbinfo.cb_cv);
1611 	cv_destroy(cp->cbinfo.cb_cv_nullcaller);
1612 
1613 	/* free callback info */
1614 	rfs4_cbinfo_free(&cp->cbinfo);
1615 
1616 	if (cp->cp_confirmed)
1617 		rfs4_client_rele(cp->cp_confirmed);
1618 
1619 	if (cp->ss_pn) {
1620 		/* check if the stable storage files need to be removed */
1621 		if (cp->ss_remove)
1622 			rfs4_dss_remove_cpleaf(cp);
1623 		rfs4_ss_pnfree(cp->ss_pn);
1624 	}
1625 
1626 	/* Free the client supplied client id */
1627 	kmem_free(cp->nfs_client.id_val, cp->nfs_client.id_len);
1628 
1629 	if (cp->sysidt != LM_NOSYSID)
1630 		lm_free_sysidt(cp->sysidt);
1631 }
1632 
1633 static bool_t
1634 rfs4_client_create(rfs4_entry_t u_entry, void *arg)
1635 {
1636 	rfs4_client_t *cp = (rfs4_client_t *)u_entry;
1637 	nfs_client_id4 *client = (nfs_client_id4 *)arg;
1638 	cid *cidp;
1639 	scid_confirm_verf *scvp;
1640 
1641 	/* Get a clientid to give to the client */
1642 	cidp = (cid *)&cp->clientid;
1643 	cidp->impl_id.start_time = rfs4_start_time;
1644 	cidp->impl_id.c_id = (uint32_t)rfs4_dbe_getid(cp->dbe);
1645 
1646 	/* If we are booted as a cluster node, embed our nodeid */
1647 	if (cluster_bootflags & CLUSTER_BOOTED)
1648 		embed_nodeid(cidp);
1649 
1650 	/* Allocate and copy client's client id value */
1651 	cp->nfs_client.id_val = kmem_alloc(client->id_len, KM_SLEEP);
1652 	cp->nfs_client.id_len = client->id_len;
1653 	bcopy(client->id_val, cp->nfs_client.id_val, client->id_len);
1654 	cp->nfs_client.verifier = client->verifier;
1655 
1656 	/* Init the value for the SETCLIENTID_CONFIRM verifier */
1657 	scvp = (scid_confirm_verf *)&cp->confirm_verf;
1658 	scvp->cv_impl.c_id = cidp->impl_id.c_id;
1659 	scvp->cv_impl.gen_num = 0;
1660 
1661 	/* An F_UNLKSYS has been done for this client */
1662 	cp->unlksys_completed = FALSE;
1663 
1664 	/* We need the client to ack us */
1665 	cp->need_confirm = TRUE;
1666 	cp->cp_confirmed = NULL;
1667 
1668 	/* TRUE all the time until the callback path actually fails */
1669 	cp->cbinfo.cb_notified_of_cb_path_down = TRUE;
1670 
1671 	/* Initialize the access time to now */
1672 	cp->last_access = gethrestime_sec();
1673 
1674 	cp->cr_set = NULL;
1675 	/* Initialize list for insque/remque */
1676 	cp->openownerlist.next = cp->openownerlist.prev = &cp->openownerlist;
1677 	cp->openownerlist.oop = NULL; /* This is not an openowner */
1678 
1679 	cp->sysidt = LM_NOSYSID;
1680 
1681 	cp->clientdeleglist.next = cp->clientdeleglist.prev =
1682 	    &cp->clientdeleglist;
1683 	cp->clientdeleglist.dsp = NULL;
1684 
1685 	/* set up the callback control structure */
1686 	cp->cbinfo.cb_state = CB_UNINIT;
1687 	mutex_init(cp->cbinfo.cb_lock, NULL, MUTEX_DEFAULT, NULL);
1688 	cv_init(cp->cbinfo.cb_cv, NULL, CV_DEFAULT, NULL);
1689 	cv_init(cp->cbinfo.cb_cv_nullcaller, NULL, CV_DEFAULT, NULL);
1690 
1691 	/*
1692 	 * Associate the client_t with the current server instance.
1693 	 * The hold is solely to satisfy the calling requirement of
1694 	 * rfs4_servinst_assign(). In this case it's not strictly necessary.
1695 	 */
1696 	rfs4_dbe_hold(cp->dbe);
1697 	rfs4_servinst_assign(cp, rfs4_cur_servinst);
1698 	rfs4_dbe_rele(cp->dbe);
1699 
1700 	return (TRUE);
1701 }
1702 
1703 /*
1704  * Caller wants to generate/update the setclientid_confirm verifier
1705  * associated with a client.  This is done during the SETCLIENTID
1706  * processing.
1707  */
1708 void
1709 rfs4_client_scv_next(rfs4_client_t *cp)
1710 {
1711 	scid_confirm_verf *scvp;
1712 
1713 	/* Init the value for the SETCLIENTID_CONFIRM verifier */
1714 	scvp = (scid_confirm_verf *)&cp->confirm_verf;
1715 	scvp->cv_impl.gen_num++;
1716 }
1717 
1718 void
1719 rfs4_client_rele(rfs4_client_t *cp)
1720 {
1721 	rfs4_dbe_rele(cp->dbe);
1722 }
1723 
1724 rfs4_client_t *
1725 rfs4_findclient(nfs_client_id4 *client, bool_t *create,	rfs4_client_t *oldcp)
1726 {
1727 	rfs4_client_t *cp;
1728 
1729 
1730 	if (oldcp) {
1731 		rw_enter(&rfs4_findclient_lock, RW_WRITER);
1732 		rfs4_dbe_hide(oldcp->dbe);
1733 	} else {
1734 		rw_enter(&rfs4_findclient_lock, RW_READER);
1735 	}
1736 
1737 	cp = (rfs4_client_t *)rfs4_dbsearch(rfs4_nfsclnt_idx, client,
1738 	    create, (void *)client, RFS4_DBS_VALID);
1739 
1740 	if (oldcp)
1741 		rfs4_dbe_unhide(oldcp->dbe);
1742 
1743 	rw_exit(&rfs4_findclient_lock);
1744 
1745 	return (cp);
1746 }
1747 
1748 rfs4_client_t *
1749 rfs4_findclient_by_id(clientid4 clientid, bool_t find_unconfirmed)
1750 {
1751 	rfs4_client_t *cp;
1752 	bool_t create = FALSE;
1753 	cid *cidp = (cid *)&clientid;
1754 
1755 	/* If we're a cluster and the nodeid isn't right, short-circuit */
1756 	if (cluster_bootflags & CLUSTER_BOOTED && foreign_clientid(cidp))
1757 		return (NULL);
1758 
1759 	rw_enter(&rfs4_findclient_lock, RW_READER);
1760 
1761 	cp = (rfs4_client_t *)rfs4_dbsearch(rfs4_clientid_idx, &clientid,
1762 	    &create, NULL, RFS4_DBS_VALID);
1763 
1764 	rw_exit(&rfs4_findclient_lock);
1765 
1766 	if (cp && cp->need_confirm && find_unconfirmed == FALSE) {
1767 		rfs4_client_rele(cp);
1768 		return (NULL);
1769 	} else {
1770 		return (cp);
1771 	}
1772 }
1773 
1774 bool_t
1775 rfs4_lease_expired(rfs4_client_t *cp)
1776 {
1777 	bool_t rc;
1778 
1779 	rfs4_dbe_lock(cp->dbe);
1780 
1781 	/*
1782 	 * If the admin has executed clear_locks for this
1783 	 * client id, force expire will be set, so no need
1784 	 * to calculate anything because it's "outa here".
1785 	 */
1786 	if (cp->forced_expire) {
1787 		rc = TRUE;
1788 	} else {
1789 		rc = (gethrestime_sec() - cp->last_access > rfs4_lease_time);
1790 	}
1791 
1792 	/*
1793 	 * If the lease has expired we will also want
1794 	 * to remove any stable storage state data. So
1795 	 * mark the client id accordingly.
1796 	 */
1797 	if (!cp->ss_remove)
1798 		cp->ss_remove = (rc == TRUE);
1799 
1800 	rfs4_dbe_unlock(cp->dbe);
1801 
1802 	return (rc);
1803 }
1804 
1805 void
1806 rfs4_update_lease(rfs4_client_t *cp)
1807 {
1808 	rfs4_dbe_lock(cp->dbe);
1809 	if (!cp->forced_expire)
1810 		cp->last_access = gethrestime_sec();
1811 	rfs4_dbe_unlock(cp->dbe);
1812 }
1813 
1814 
1815 static bool_t
1816 EQOPENOWNER(open_owner4 *a, open_owner4 *b)
1817 {
1818 	bool_t rc;
1819 
1820 	if (a->clientid != b->clientid)
1821 		return (FALSE);
1822 
1823 	if (a->owner_len != b->owner_len)
1824 		return (FALSE);
1825 
1826 	rc = (bcmp(a->owner_val, b->owner_val, a->owner_len) == 0);
1827 
1828 	return (rc);
1829 }
1830 
1831 static uint_t
1832 openowner_hash(void *key)
1833 {
1834 	int i;
1835 	open_owner4 *openowner = key;
1836 	uint_t hash = 0;
1837 
1838 	for (i = 0; i < openowner->owner_len; i++) {
1839 		hash <<= 4;
1840 		hash += (uint_t)openowner->owner_val[i];
1841 	}
1842 	hash += (uint_t)openowner->clientid;
1843 	hash |= (openowner->clientid >> 32);
1844 
1845 	return (hash);
1846 }
1847 
1848 static bool_t
1849 openowner_compare(rfs4_entry_t u_entry, void *key)
1850 {
1851 	rfs4_openowner_t *op = (rfs4_openowner_t *)u_entry;
1852 	open_owner4 *arg = key;
1853 
1854 	return (EQOPENOWNER(&op->owner, arg));
1855 }
1856 
1857 void *
1858 openowner_mkkey(rfs4_entry_t u_entry)
1859 {
1860 	rfs4_openowner_t *op = (rfs4_openowner_t *)u_entry;
1861 
1862 	return (&op->owner);
1863 }
1864 
1865 static bool_t
1866 rfs4_openowner_expiry(rfs4_entry_t u_entry)
1867 {
1868 	rfs4_openowner_t *op = (rfs4_openowner_t *)u_entry;
1869 
1870 	if (rfs4_dbe_is_invalid(op->dbe))
1871 		return (TRUE);
1872 	return ((gethrestime_sec() - op->client->last_access
1873 	    > rfs4_lease_time));
1874 }
1875 
1876 static void
1877 rfs4_openowner_destroy(rfs4_entry_t u_entry)
1878 {
1879 	rfs4_openowner_t *op = (rfs4_openowner_t *)u_entry;
1880 
1881 	rfs4_sw_destroy(&op->oo_sw);
1882 
1883 	/* Remove open owner from client's lists of open owners */
1884 	rfs4_dbe_lock(op->client->dbe);
1885 
1886 	remque(&op->openownerlist);
1887 	op->openownerlist.next = op->openownerlist.prev = &op->openownerlist;
1888 
1889 	rfs4_dbe_unlock(op->client->dbe);
1890 
1891 	/* One less reference to the client */
1892 	rfs4_client_rele(op->client);
1893 	op->client = NULL;
1894 
1895 	/* Free the last reply for this lock owner */
1896 	rfs4_free_reply(op->reply);
1897 
1898 	if (op->reply_fh.nfs_fh4_val) {
1899 		kmem_free(op->reply_fh.nfs_fh4_val, op->reply_fh.nfs_fh4_len);
1900 		op->reply_fh.nfs_fh4_val = NULL;
1901 		op->reply_fh.nfs_fh4_len = 0;
1902 	}
1903 
1904 	/* Free the lock owner id */
1905 	kmem_free(op->owner.owner_val, op->owner.owner_len);
1906 }
1907 
1908 void
1909 rfs4_openowner_rele(rfs4_openowner_t *op)
1910 {
1911 	rfs4_dbe_rele(op->dbe);
1912 }
1913 
1914 static bool_t
1915 rfs4_openowner_create(rfs4_entry_t u_entry, void *arg)
1916 {
1917 	rfs4_openowner_t *op = (rfs4_openowner_t *)u_entry;
1918 	rfs4_openowner_t *argp = (rfs4_openowner_t *)arg;
1919 	open_owner4 *openowner = &argp->owner;
1920 	seqid4 seqid = argp->open_seqid;
1921 	rfs4_client_t *cp;
1922 	bool_t create = FALSE;
1923 
1924 	rw_enter(&rfs4_findclient_lock, RW_READER);
1925 
1926 	cp = (rfs4_client_t *)rfs4_dbsearch(rfs4_clientid_idx,
1927 	    &openowner->clientid,
1928 	    &create, NULL, RFS4_DBS_VALID);
1929 
1930 	rw_exit(&rfs4_findclient_lock);
1931 
1932 	if (cp == NULL)
1933 		return (FALSE);
1934 
1935 	op->reply_fh.nfs_fh4_len = 0;
1936 	op->reply_fh.nfs_fh4_val = NULL;
1937 
1938 	op->owner.clientid = openowner->clientid;
1939 	op->owner.owner_val =
1940 	    kmem_alloc(openowner->owner_len, KM_SLEEP);
1941 
1942 	bcopy(openowner->owner_val,
1943 	    op->owner.owner_val, openowner->owner_len);
1944 
1945 	op->owner.owner_len = openowner->owner_len;
1946 
1947 	op->need_confirm = TRUE;
1948 
1949 	rfs4_sw_init(&op->oo_sw);
1950 
1951 	op->open_seqid = seqid;
1952 	bzero(op->reply, sizeof (nfs_resop4));
1953 	op->client = cp;
1954 	op->cr_set = NULL;
1955 	/* Init lists for remque/insque */
1956 	op->ownerstateids.next = op->ownerstateids.prev = &op->ownerstateids;
1957 	op->ownerstateids.sp = NULL; /* NULL since this is the state list */
1958 	op->openownerlist.next = op->openownerlist.prev = &op->openownerlist;
1959 	op->openownerlist.oop = op; /* ourselves */
1960 
1961 	/* Insert openowner into client's open owner list */
1962 	rfs4_dbe_lock(cp->dbe);
1963 
1964 	insque(&op->openownerlist, cp->openownerlist.prev);
1965 
1966 	rfs4_dbe_unlock(cp->dbe);
1967 
1968 	return (TRUE);
1969 }
1970 
1971 rfs4_openowner_t *
1972 rfs4_findopenowner(open_owner4 *openowner, bool_t *create, seqid4 seqid)
1973 {
1974 	rfs4_openowner_t *op;
1975 	rfs4_openowner_t arg;
1976 
1977 	arg.owner = *openowner;
1978 	arg.open_seqid = seqid;
1979 	op = (rfs4_openowner_t *)rfs4_dbsearch(rfs4_openowner_idx, openowner,
1980 	    create, &arg, RFS4_DBS_VALID);
1981 
1982 	return (op);
1983 }
1984 
1985 void
1986 rfs4_update_open_sequence(rfs4_openowner_t *op)
1987 {
1988 
1989 	rfs4_dbe_lock(op->dbe);
1990 
1991 	op->open_seqid++;
1992 
1993 	rfs4_dbe_unlock(op->dbe);
1994 }
1995 
1996 void
1997 rfs4_update_open_resp(rfs4_openowner_t *op, nfs_resop4 *resp, nfs_fh4 *fh)
1998 {
1999 
2000 	rfs4_dbe_lock(op->dbe);
2001 
2002 	rfs4_free_reply(op->reply);
2003 
2004 	rfs4_copy_reply(op->reply, resp);
2005 
2006 	/* Save the filehandle if provided and free if not used */
2007 	if (resp->nfs_resop4_u.opopen.status == NFS4_OK &&
2008 	    fh && fh->nfs_fh4_len) {
2009 		if (op->reply_fh.nfs_fh4_val == NULL)
2010 			op->reply_fh.nfs_fh4_val =
2011 			    kmem_alloc(fh->nfs_fh4_len, KM_SLEEP);
2012 		nfs_fh4_copy(fh, &op->reply_fh);
2013 	} else {
2014 		if (op->reply_fh.nfs_fh4_val) {
2015 			kmem_free(op->reply_fh.nfs_fh4_val,
2016 			    op->reply_fh.nfs_fh4_len);
2017 			op->reply_fh.nfs_fh4_val = NULL;
2018 			op->reply_fh.nfs_fh4_len = 0;
2019 		}
2020 	}
2021 
2022 	rfs4_dbe_unlock(op->dbe);
2023 }
2024 
2025 static bool_t
2026 lockowner_compare(rfs4_entry_t u_entry, void *key)
2027 {
2028 	rfs4_lockowner_t *lo = (rfs4_lockowner_t *)u_entry;
2029 	lock_owner4 *b = (lock_owner4 *)key;
2030 
2031 	if (lo->owner.clientid != b->clientid)
2032 		return (FALSE);
2033 
2034 	if (lo->owner.owner_len != b->owner_len)
2035 		return (FALSE);
2036 
2037 	return (bcmp(lo->owner.owner_val, b->owner_val,
2038 	    lo->owner.owner_len) == 0);
2039 }
2040 
2041 void *
2042 lockowner_mkkey(rfs4_entry_t u_entry)
2043 {
2044 	rfs4_lockowner_t *lo = (rfs4_lockowner_t *)u_entry;
2045 
2046 	return (&lo->owner);
2047 }
2048 
2049 static uint32_t
2050 lockowner_hash(void *key)
2051 {
2052 	int i;
2053 	lock_owner4 *lockowner = key;
2054 	uint_t hash = 0;
2055 
2056 	for (i = 0; i < lockowner->owner_len; i++) {
2057 		hash <<= 4;
2058 		hash += (uint_t)lockowner->owner_val[i];
2059 	}
2060 	hash += (uint_t)lockowner->clientid;
2061 	hash |= (lockowner->clientid >> 32);
2062 
2063 	return (hash);
2064 }
2065 
2066 static uint32_t
2067 pid_hash(void *key)
2068 {
2069 	return ((uint32_t)(uintptr_t)key);
2070 }
2071 
2072 static void *
2073 pid_mkkey(rfs4_entry_t u_entry)
2074 {
2075 	rfs4_lockowner_t *lo = (rfs4_lockowner_t *)u_entry;
2076 
2077 	return ((void *)(uintptr_t)lo->pid);
2078 }
2079 
2080 static bool_t
2081 pid_compare(rfs4_entry_t u_entry, void *key)
2082 {
2083 	rfs4_lockowner_t *lo = (rfs4_lockowner_t *)u_entry;
2084 
2085 	return (lo->pid == (pid_t)(uintptr_t)key);
2086 }
2087 
2088 static void
2089 rfs4_lockowner_destroy(rfs4_entry_t u_entry)
2090 {
2091 	rfs4_lockowner_t *lo = (rfs4_lockowner_t *)u_entry;
2092 
2093 	/* Free the lock owner id */
2094 	kmem_free(lo->owner.owner_val, lo->owner.owner_len);
2095 	rfs4_client_rele(lo->client);
2096 }
2097 
2098 void
2099 rfs4_lockowner_rele(rfs4_lockowner_t *lo)
2100 {
2101 	rfs4_dbe_rele(lo->dbe);
2102 }
2103 
2104 /* ARGSUSED */
2105 static bool_t
2106 rfs4_lockowner_expiry(rfs4_entry_t u_entry)
2107 {
2108 	/*
2109 	 * Since expiry is called with no other references on
2110 	 * this struct, go ahead and have it removed.
2111 	 */
2112 	return (TRUE);
2113 }
2114 
2115 static bool_t
2116 rfs4_lockowner_create(rfs4_entry_t u_entry, void *arg)
2117 {
2118 	rfs4_lockowner_t *lo = (rfs4_lockowner_t *)u_entry;
2119 	lock_owner4 *lockowner = (lock_owner4 *)arg;
2120 	rfs4_client_t *cp;
2121 	bool_t create = FALSE;
2122 
2123 	rw_enter(&rfs4_findclient_lock, RW_READER);
2124 
2125 	cp = (rfs4_client_t *)rfs4_dbsearch(rfs4_clientid_idx,
2126 	    &lockowner->clientid,
2127 	    &create, NULL, RFS4_DBS_VALID);
2128 
2129 	rw_exit(&rfs4_findclient_lock);
2130 
2131 	if (cp == NULL)
2132 		return (FALSE);
2133 
2134 	/* Reference client */
2135 	lo->client = cp;
2136 	lo->owner.clientid = lockowner->clientid;
2137 	lo->owner.owner_val = kmem_alloc(lockowner->owner_len, KM_SLEEP);
2138 	bcopy(lockowner->owner_val, lo->owner.owner_val, lockowner->owner_len);
2139 	lo->owner.owner_len = lockowner->owner_len;
2140 	lo->pid = rfs4_dbe_getid(lo->dbe);
2141 
2142 	return (TRUE);
2143 }
2144 
2145 rfs4_lockowner_t *
2146 rfs4_findlockowner(lock_owner4 *lockowner, bool_t *create)
2147 {
2148 	rfs4_lockowner_t *lo;
2149 
2150 	lo = (rfs4_lockowner_t *)rfs4_dbsearch(rfs4_lockowner_idx, lockowner,
2151 	    create, lockowner, RFS4_DBS_VALID);
2152 
2153 	return (lo);
2154 }
2155 
2156 rfs4_lockowner_t *
2157 rfs4_findlockowner_by_pid(pid_t pid)
2158 {
2159 	rfs4_lockowner_t *lo;
2160 	bool_t create = FALSE;
2161 
2162 	lo = (rfs4_lockowner_t *)rfs4_dbsearch(rfs4_lockowner_pid_idx,
2163 	    (void *)(uintptr_t)pid, &create, NULL, RFS4_DBS_VALID);
2164 
2165 	return (lo);
2166 }
2167 
2168 
2169 static uint32_t
2170 file_hash(void *key)
2171 {
2172 	return (ADDRHASH(key));
2173 }
2174 
2175 static void *
2176 file_mkkey(rfs4_entry_t u_entry)
2177 {
2178 	rfs4_file_t *fp = (rfs4_file_t *)u_entry;
2179 
2180 	return (fp->vp);
2181 }
2182 
2183 static bool_t
2184 file_compare(rfs4_entry_t u_entry, void *key)
2185 {
2186 	rfs4_file_t *fp = (rfs4_file_t *)u_entry;
2187 
2188 	return (fp->vp == (vnode_t *)key);
2189 }
2190 
2191 static void
2192 rfs4_file_destroy(rfs4_entry_t u_entry)
2193 {
2194 	rfs4_file_t *fp = (rfs4_file_t *)u_entry;
2195 
2196 	ASSERT(fp->delegationlist.next == &fp->delegationlist);
2197 	if (fp->filehandle.nfs_fh4_val)
2198 		kmem_free(fp->filehandle.nfs_fh4_val,
2199 		    fp->filehandle.nfs_fh4_len);
2200 	cv_destroy(fp->dinfo->recall_cv);
2201 	if (fp->vp) {
2202 		vnode_t *vp = fp->vp;
2203 
2204 		mutex_enter(&vp->v_lock);
2205 		(void) vsd_set(vp, nfs4_srv_vkey, NULL);
2206 		mutex_exit(&vp->v_lock);
2207 		VN_RELE(vp);
2208 		fp->vp = NULL;
2209 	}
2210 	rw_destroy(&fp->file_rwlock);
2211 }
2212 
2213 /*
2214  * Used to unlock the underlying dbe struct only
2215  */
2216 void
2217 rfs4_file_rele(rfs4_file_t *fp)
2218 {
2219 	rfs4_dbe_rele(fp->dbe);
2220 }
2221 
2222 /*
2223  * Used to unlock the file rw lock and the file's dbe entry
2224  * Only used to pair with rfs4_findfile_withlock()
2225  */
2226 void
2227 rfs4_file_rele_withunlock(rfs4_file_t *fp)
2228 {
2229 	rw_exit(&fp->file_rwlock);
2230 	rfs4_dbe_rele(fp->dbe);
2231 }
2232 
2233 typedef struct {
2234     vnode_t *vp;
2235     nfs_fh4 *fh;
2236 } rfs4_fcreate_arg;
2237 
2238 static bool_t
2239 rfs4_file_create(rfs4_entry_t u_entry, void *arg)
2240 {
2241 	rfs4_file_t *fp = (rfs4_file_t *)u_entry;
2242 	rfs4_fcreate_arg *ap = (rfs4_fcreate_arg *)arg;
2243 	vnode_t *vp = ap->vp;
2244 	nfs_fh4 *fh = ap->fh;
2245 
2246 	VN_HOLD(vp);
2247 
2248 	fp->filehandle.nfs_fh4_len = 0;
2249 	fp->filehandle.nfs_fh4_val = NULL;
2250 	ASSERT(fh && fh->nfs_fh4_len);
2251 	if (fh && fh->nfs_fh4_len) {
2252 		fp->filehandle.nfs_fh4_val =
2253 		    kmem_alloc(fh->nfs_fh4_len, KM_SLEEP);
2254 		nfs_fh4_copy(fh, &fp->filehandle);
2255 	}
2256 	fp->vp = vp;
2257 
2258 	/* Init list for remque/insque */
2259 	fp->delegationlist.next = fp->delegationlist.prev =
2260 	    &fp->delegationlist;
2261 	fp->delegationlist.dsp = NULL; /* NULL since this is state list */
2262 
2263 	fp->share_deny = fp->share_access = fp->access_read = 0;
2264 	fp->access_write = fp->deny_read = fp->deny_write = 0;
2265 
2266 	mutex_init(fp->dinfo->recall_lock, NULL, MUTEX_DEFAULT, NULL);
2267 	cv_init(fp->dinfo->recall_cv, NULL, CV_DEFAULT, NULL);
2268 
2269 	fp->dinfo->dtype = OPEN_DELEGATE_NONE;
2270 
2271 	rw_init(&fp->file_rwlock, NULL, RW_DEFAULT, NULL);
2272 
2273 	mutex_enter(&vp->v_lock);
2274 	if (vsd_set(vp, nfs4_srv_vkey, (void *)fp)) {
2275 		ASSERT(FALSE);
2276 		cmn_err(CE_WARN, "rfs4_file_create: vsd_set failed.");
2277 	}
2278 	mutex_exit(&vp->v_lock);
2279 
2280 	return (TRUE);
2281 }
2282 
2283 rfs4_file_t *
2284 rfs4_findfile(vnode_t *vp, nfs_fh4 *fh, bool_t *create)
2285 {
2286 	rfs4_file_t *fp;
2287 	rfs4_fcreate_arg arg;
2288 
2289 	arg.vp = vp;
2290 	arg.fh = fh;
2291 
2292 	if (*create == TRUE)
2293 		fp = (rfs4_file_t *)rfs4_dbsearch(rfs4_file_idx, vp, create,
2294 		    &arg, RFS4_DBS_VALID);
2295 	else {
2296 		mutex_enter(&vp->v_lock);
2297 		fp = (rfs4_file_t *)vsd_get(vp, nfs4_srv_vkey);
2298 		mutex_exit(&vp->v_lock);
2299 		if (fp) {
2300 			rfs4_dbe_lock(fp->dbe);
2301 			if (rfs4_dbe_is_invalid(fp->dbe) ||
2302 			    (rfs4_dbe_refcnt(fp->dbe) == 0)) {
2303 				rfs4_dbe_unlock(fp->dbe);
2304 				fp = NULL;
2305 			} else {
2306 				rfs4_dbe_hold(fp->dbe);
2307 				rfs4_dbe_unlock(fp->dbe);
2308 			}
2309 		}
2310 	}
2311 	return (fp);
2312 }
2313 
2314 /*
2315  * Find a file in the db and once it is located, take the rw lock.
2316  * Need to check the vnode pointer and if it does not exist (it was
2317  * removed between the db location and check) redo the find.  This
2318  * assumes that a file struct that has a NULL vnode pointer is marked
2319  * at 'invalid' and will not be found in the db the second time
2320  * around.
2321  */
2322 rfs4_file_t *
2323 rfs4_findfile_withlock(vnode_t *vp, nfs_fh4 *fh, bool_t *create)
2324 {
2325 	rfs4_file_t *fp;
2326 	rfs4_fcreate_arg arg;
2327 	bool_t screate = *create;
2328 
2329 	if (screate == FALSE) {
2330 		mutex_enter(&vp->v_lock);
2331 		fp = (rfs4_file_t *)vsd_get(vp, nfs4_srv_vkey);
2332 		mutex_exit(&vp->v_lock);
2333 		if (fp) {
2334 			rfs4_dbe_lock(fp->dbe);
2335 			if (rfs4_dbe_is_invalid(fp->dbe) ||
2336 			    (rfs4_dbe_refcnt(fp->dbe) == 0)) {
2337 				rfs4_dbe_unlock(fp->dbe);
2338 				fp = NULL;
2339 			} else {
2340 				rfs4_dbe_hold(fp->dbe);
2341 				rfs4_dbe_unlock(fp->dbe);
2342 				rw_enter(&fp->file_rwlock, RW_WRITER);
2343 			}
2344 		}
2345 	} else {
2346 retry:
2347 		arg.vp = vp;
2348 		arg.fh = fh;
2349 
2350 		fp = (rfs4_file_t *)rfs4_dbsearch(rfs4_file_idx, vp, create,
2351 		    &arg, RFS4_DBS_VALID);
2352 		if (fp != NULL) {
2353 			rw_enter(&fp->file_rwlock, RW_WRITER);
2354 			if (fp->vp == NULL) {
2355 				rw_exit(&fp->file_rwlock);
2356 				rfs4_file_rele(fp);
2357 				*create = screate;
2358 				goto retry;
2359 			}
2360 		}
2361 	}
2362 
2363 	return (fp);
2364 }
2365 
2366 static uint32_t
2367 lo_state_hash(void *key)
2368 {
2369 	stateid_t *id = key;
2370 
2371 	return (id->bits.ident+id->bits.pid);
2372 }
2373 
2374 static bool_t
2375 lo_state_compare(rfs4_entry_t u_entry, void *key)
2376 {
2377 	rfs4_lo_state_t *lop = (rfs4_lo_state_t *)u_entry;
2378 	stateid_t *id = key;
2379 	bool_t rc;
2380 
2381 	rc = (lop->lockid.bits.boottime == id->bits.boottime &&
2382 	    lop->lockid.bits.type == id->bits.type &&
2383 	    lop->lockid.bits.ident == id->bits.ident &&
2384 	    lop->lockid.bits.pid == id->bits.pid);
2385 
2386 	return (rc);
2387 }
2388 
2389 static void *
2390 lo_state_mkkey(rfs4_entry_t u_entry)
2391 {
2392 	rfs4_lo_state_t *lsp = (rfs4_lo_state_t *)u_entry;
2393 
2394 	return (&lsp->lockid);
2395 }
2396 
2397 static bool_t
2398 rfs4_lo_state_expiry(rfs4_entry_t u_entry)
2399 {
2400 	rfs4_lo_state_t *lsp = (rfs4_lo_state_t *)u_entry;
2401 
2402 	if (rfs4_dbe_is_invalid(lsp->dbe))
2403 		return (TRUE);
2404 	if (lsp->state->closed)
2405 		return (TRUE);
2406 	return ((gethrestime_sec() - lsp->state->owner->client->last_access
2407 	    > rfs4_lease_time));
2408 }
2409 
2410 static void
2411 rfs4_lo_state_destroy(rfs4_entry_t u_entry)
2412 {
2413 	rfs4_lo_state_t *lsp = (rfs4_lo_state_t *)u_entry;
2414 
2415 	rfs4_sw_destroy(&lsp->ls_sw);
2416 
2417 	/* Make sure to release the file locks */
2418 	if (lsp->locks_cleaned == FALSE) {
2419 		lsp->locks_cleaned = TRUE;
2420 		if (lsp->locker->client->sysidt != LM_NOSYSID) {
2421 			/* Is the PxFS kernel module loaded? */
2422 			if (lm_remove_file_locks != NULL) {
2423 				int new_sysid;
2424 
2425 				/* Encode the cluster nodeid in new sysid */
2426 				new_sysid = lsp->locker->client->sysidt;
2427 				lm_set_nlmid_flk(&new_sysid);
2428 
2429 				/*
2430 				 * This PxFS routine removes file locks for a
2431 				 * client over all nodes of a cluster.
2432 				 */
2433 				DTRACE_PROBE1(nfss_i_clust_rm_lck,
2434 				    int, new_sysid);
2435 				(*lm_remove_file_locks)(new_sysid);
2436 			} else {
2437 				(void) cleanlocks(lsp->state->finfo->vp,
2438 				    lsp->locker->pid,
2439 				    lsp->locker->client->sysidt);
2440 			}
2441 		}
2442 	}
2443 
2444 	rfs4_dbe_lock(lsp->state->dbe);
2445 
2446 	remque(&lsp->lockownerlist);
2447 	lsp->lockownerlist.next = lsp->lockownerlist.prev =
2448 	    &lsp->lockownerlist;
2449 
2450 	rfs4_dbe_unlock(lsp->state->dbe);
2451 
2452 	/* Free the last reply for this state */
2453 	rfs4_free_reply(lsp->reply);
2454 
2455 	rfs4_lockowner_rele(lsp->locker);
2456 	lsp->locker = NULL;
2457 
2458 	rfs4_state_rele_nounlock(lsp->state);
2459 	lsp->state = NULL;
2460 }
2461 
2462 static bool_t
2463 rfs4_lo_state_create(rfs4_entry_t u_entry, void *arg)
2464 {
2465 	rfs4_lo_state_t *lsp = (rfs4_lo_state_t *)u_entry;
2466 	rfs4_lo_state_t *argp = (rfs4_lo_state_t *)arg;
2467 	rfs4_lockowner_t *lo = argp->locker;
2468 	rfs4_state_t *sp = argp->state;
2469 
2470 	lsp->state = sp;
2471 
2472 	lsp->lockid = sp->stateid;
2473 	lsp->lockid.bits.type = LOCKID;
2474 	lsp->lockid.bits.chgseq = 0;
2475 	lsp->lockid.bits.pid = lo->pid;
2476 
2477 	lsp->locks_cleaned = FALSE;
2478 	lsp->lock_completed = FALSE;
2479 
2480 	rfs4_sw_init(&lsp->ls_sw);
2481 
2482 	/* Attached the supplied lock owner */
2483 	rfs4_dbe_hold(lo->dbe);
2484 	lsp->locker = lo;
2485 
2486 	lsp->lockownerlist.next = lsp->lockownerlist.prev =
2487 	    &lsp->lockownerlist;
2488 	lsp->lockownerlist.lsp = lsp;
2489 
2490 	rfs4_dbe_lock(sp->dbe);
2491 
2492 	insque(&lsp->lockownerlist, sp->lockownerlist.prev);
2493 
2494 	rfs4_dbe_hold(sp->dbe);
2495 
2496 	rfs4_dbe_unlock(sp->dbe);
2497 
2498 	return (TRUE);
2499 }
2500 
2501 void
2502 rfs4_lo_state_rele(rfs4_lo_state_t *lsp, bool_t unlock_fp)
2503 {
2504 	if (unlock_fp == TRUE)
2505 		rw_exit(&lsp->state->finfo->file_rwlock);
2506 	rfs4_dbe_rele(lsp->dbe);
2507 }
2508 
2509 static rfs4_lo_state_t *
2510 rfs4_findlo_state(stateid_t *id, bool_t lock_fp)
2511 {
2512 	rfs4_lo_state_t *lsp;
2513 	bool_t create = FALSE;
2514 
2515 	lsp = (rfs4_lo_state_t *)rfs4_dbsearch(rfs4_lo_state_idx, id,
2516 	    &create, NULL, RFS4_DBS_VALID);
2517 	if (lock_fp == TRUE && lsp != NULL)
2518 		rw_enter(&lsp->state->finfo->file_rwlock, RW_READER);
2519 
2520 	return (lsp);
2521 }
2522 
2523 
2524 static uint32_t
2525 lo_state_lo_hash(void *key)
2526 {
2527 	rfs4_lo_state_t *lop = key;
2528 
2529 	return (ADDRHASH(lop->locker) ^ ADDRHASH(lop->state));
2530 }
2531 
2532 static bool_t
2533 lo_state_lo_compare(rfs4_entry_t u_entry, void *key)
2534 {
2535 	rfs4_lo_state_t *lop = (rfs4_lo_state_t *)u_entry;
2536 	rfs4_lo_state_t *keyp = key;
2537 
2538 	return (keyp->locker == lop->locker && keyp->state == lop->state);
2539 }
2540 
2541 static void *
2542 lo_state_lo_mkkey(rfs4_entry_t u_entry)
2543 {
2544 	return (u_entry);
2545 }
2546 
2547 rfs4_lo_state_t *
2548 rfs4_findlo_state_by_owner(rfs4_lockowner_t *lo,
2549 			rfs4_state_t *sp, bool_t *create)
2550 {
2551 	rfs4_lo_state_t *lsp;
2552 	rfs4_lo_state_t arg;
2553 
2554 	arg.locker = lo;
2555 	arg.state = sp;
2556 
2557 	lsp = (rfs4_lo_state_t *)rfs4_dbsearch(rfs4_lo_state_owner_idx, &arg,
2558 	    create, &arg, RFS4_DBS_VALID);
2559 
2560 	return (lsp);
2561 }
2562 
2563 static stateid_t
2564 get_stateid(id_t eid)
2565 {
2566 	stateid_t id;
2567 
2568 	id.bits.boottime = rfs4_start_time;
2569 	id.bits.ident = eid;
2570 	id.bits.chgseq = 0;
2571 	id.bits.type = 0;
2572 	id.bits.pid = 0;
2573 
2574 	/*
2575 	 * If we are booted as a cluster node, embed our nodeid.
2576 	 * We've already done sanity checks in rfs4_client_create() so no
2577 	 * need to repeat them here.
2578 	 */
2579 	id.bits.clnodeid = (cluster_bootflags & CLUSTER_BOOTED) ?
2580 	    clconf_get_nodeid() : 0;
2581 
2582 	return (id);
2583 }
2584 
2585 /*
2586  * For use only when booted as a cluster node.
2587  * Returns TRUE if the embedded nodeid indicates that this stateid was
2588  * generated on another node.
2589  */
2590 static int
2591 foreign_stateid(stateid_t *id)
2592 {
2593 	ASSERT(cluster_bootflags & CLUSTER_BOOTED);
2594 	return (id->bits.clnodeid != (uint32_t)clconf_get_nodeid());
2595 }
2596 
2597 /*
2598  * For use only when booted as a cluster node.
2599  * Returns TRUE if the embedded nodeid indicates that this clientid was
2600  * generated on another node.
2601  */
2602 static int
2603 foreign_clientid(cid *cidp)
2604 {
2605 	ASSERT(cluster_bootflags & CLUSTER_BOOTED);
2606 	return (cidp->impl_id.c_id >> CLUSTER_NODEID_SHIFT !=
2607 	    (uint32_t)clconf_get_nodeid());
2608 }
2609 
2610 /*
2611  * For use only when booted as a cluster node.
2612  * Embed our cluster nodeid into the clientid.
2613  */
2614 static void
2615 embed_nodeid(cid *cidp)
2616 {
2617 	int clnodeid;
2618 	/*
2619 	 * Currently, our state tables are small enough that their
2620 	 * ids will leave enough bits free for the nodeid. If the
2621 	 * tables become larger, we mustn't overwrite the id.
2622 	 * Equally, we only have room for so many bits of nodeid, so
2623 	 * must check that too.
2624 	 */
2625 	ASSERT(cluster_bootflags & CLUSTER_BOOTED);
2626 	ASSERT(cidp->impl_id.c_id >> CLUSTER_NODEID_SHIFT == 0);
2627 	clnodeid = clconf_get_nodeid();
2628 	ASSERT(clnodeid <= CLUSTER_MAX_NODEID);
2629 	ASSERT(clnodeid != NODEID_UNKNOWN);
2630 	cidp->impl_id.c_id |= (clnodeid << CLUSTER_NODEID_SHIFT);
2631 }
2632 
2633 static uint32_t
2634 state_hash(void *key)
2635 {
2636 	stateid_t *ip = (stateid_t *)key;
2637 
2638 	return (ip->bits.ident);
2639 }
2640 
2641 static bool_t
2642 state_compare(rfs4_entry_t u_entry, void *key)
2643 {
2644 	rfs4_state_t *sp = (rfs4_state_t *)u_entry;
2645 	stateid_t *id = (stateid_t *)key;
2646 	bool_t rc;
2647 
2648 	rc = (sp->stateid.bits.boottime == id->bits.boottime &&
2649 	    sp->stateid.bits.ident == id->bits.ident);
2650 
2651 	return (rc);
2652 }
2653 
2654 static void *
2655 state_mkkey(rfs4_entry_t u_entry)
2656 {
2657 	rfs4_state_t *sp = (rfs4_state_t *)u_entry;
2658 
2659 	return (&sp->stateid);
2660 }
2661 
2662 static void
2663 rfs4_state_destroy(rfs4_entry_t u_entry)
2664 {
2665 	rfs4_state_t *sp = (rfs4_state_t *)u_entry;
2666 
2667 	ASSERT(&sp->lockownerlist == sp->lockownerlist.next);
2668 
2669 	/* release any share locks for this stateid if it's still open */
2670 	if (!sp->closed)
2671 		rfs4_unshare(sp);
2672 
2673 	/* Were done with the file */
2674 	rfs4_file_rele(sp->finfo);
2675 	sp->finfo = NULL;
2676 
2677 	/* And now with the openowner */
2678 	rfs4_dbe_lock(sp->owner->dbe);
2679 
2680 	remque(&sp->ownerstateids);
2681 	sp->ownerstateids.next = sp->ownerstateids.prev = &sp->ownerstateids;
2682 
2683 	rfs4_dbe_unlock(sp->owner->dbe);
2684 
2685 	rfs4_openowner_rele(sp->owner);
2686 	sp->owner = NULL;
2687 }
2688 
2689 static void
2690 rfs4_state_rele_nounlock(rfs4_state_t *sp)
2691 {
2692 	rfs4_dbe_rele(sp->dbe);
2693 }
2694 
2695 void
2696 rfs4_state_rele(rfs4_state_t *sp)
2697 {
2698 	rw_exit(&sp->finfo->file_rwlock);
2699 	rfs4_dbe_rele(sp->dbe);
2700 }
2701 
2702 static uint32_t
2703 deleg_hash(void *key)
2704 {
2705 	rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)key;
2706 
2707 	return (ADDRHASH(dsp->client) ^ ADDRHASH(dsp->finfo));
2708 }
2709 
2710 static bool_t
2711 deleg_compare(rfs4_entry_t u_entry, void *key)
2712 {
2713 	rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
2714 	rfs4_deleg_state_t *kdsp = (rfs4_deleg_state_t *)key;
2715 
2716 	return (dsp->client == kdsp->client && dsp->finfo == kdsp->finfo);
2717 }
2718 
2719 static void *
2720 deleg_mkkey(rfs4_entry_t u_entry)
2721 {
2722 	return (u_entry);
2723 }
2724 
2725 static uint32_t
2726 deleg_state_hash(void *key)
2727 {
2728 	stateid_t *ip = (stateid_t *)key;
2729 
2730 	return (ip->bits.ident);
2731 }
2732 
2733 static bool_t
2734 deleg_state_compare(rfs4_entry_t u_entry, void *key)
2735 {
2736 	rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
2737 	stateid_t *id = (stateid_t *)key;
2738 	bool_t rc;
2739 
2740 	if (id->bits.type != DELEGID)
2741 		return (FALSE);
2742 
2743 	rc = (dsp->delegid.bits.boottime == id->bits.boottime &&
2744 	    dsp->delegid.bits.ident == id->bits.ident);
2745 
2746 	return (rc);
2747 }
2748 
2749 static void *
2750 deleg_state_mkkey(rfs4_entry_t u_entry)
2751 {
2752 	rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
2753 
2754 	return (&dsp->delegid);
2755 }
2756 
2757 static bool_t
2758 rfs4_deleg_state_expiry(rfs4_entry_t u_entry)
2759 {
2760 	rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
2761 
2762 	if (rfs4_dbe_is_invalid(dsp->dbe))
2763 		return (TRUE);
2764 
2765 	if ((gethrestime_sec() - dsp->client->last_access
2766 	    > rfs4_lease_time)) {
2767 		rfs4_dbe_invalidate(dsp->dbe);
2768 		return (TRUE);
2769 	}
2770 
2771 	return (FALSE);
2772 }
2773 
2774 static bool_t
2775 rfs4_deleg_state_create(rfs4_entry_t u_entry, void *argp)
2776 {
2777 	rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
2778 	rfs4_file_t *fp = ((rfs4_deleg_state_t *)argp)->finfo;
2779 	rfs4_client_t *cp = ((rfs4_deleg_state_t *)argp)->client;
2780 
2781 	rfs4_dbe_hold(fp->dbe);
2782 	rfs4_dbe_hold(cp->dbe);
2783 
2784 	dsp->delegid = get_stateid(rfs4_dbe_getid(dsp->dbe));
2785 	dsp->delegid.bits.type = DELEGID;
2786 	dsp->finfo = fp;
2787 	dsp->client = cp;
2788 	dsp->dtype = OPEN_DELEGATE_NONE;
2789 
2790 	dsp->time_granted = gethrestime_sec();	/* observability */
2791 	dsp->time_revoked = 0;
2792 
2793 	/* Init lists for remque/insque */
2794 	dsp->delegationlist.next = dsp->delegationlist.prev =
2795 	    &dsp->delegationlist;
2796 	dsp->delegationlist.dsp = dsp;
2797 
2798 	dsp->clientdeleglist.next = dsp->clientdeleglist.prev =
2799 	    &dsp->clientdeleglist;
2800 	dsp->clientdeleglist.dsp = dsp;
2801 
2802 	/* Insert state on per open owner's list */
2803 	rfs4_dbe_lock(cp->dbe);
2804 
2805 	insque(&dsp->clientdeleglist, cp->clientdeleglist.prev);
2806 
2807 	rfs4_dbe_unlock(cp->dbe);
2808 
2809 	return (TRUE);
2810 }
2811 
2812 static void
2813 rfs4_deleg_state_destroy(rfs4_entry_t u_entry)
2814 {
2815 	rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
2816 
2817 	if (&dsp->delegationlist != dsp->delegationlist.next)
2818 		rfs4_return_deleg(dsp, FALSE);
2819 
2820 	/* Were done with the file */
2821 	rfs4_file_rele(dsp->finfo);
2822 	dsp->finfo = NULL;
2823 
2824 	/* And now with the openowner */
2825 	rfs4_dbe_lock(dsp->client->dbe);
2826 
2827 	remque(&dsp->clientdeleglist);
2828 	dsp->clientdeleglist.next = dsp->clientdeleglist.prev =
2829 	    &dsp->clientdeleglist;
2830 
2831 	rfs4_dbe_unlock(dsp->client->dbe);
2832 
2833 	rfs4_client_rele(dsp->client);
2834 	dsp->client = NULL;
2835 }
2836 
2837 rfs4_deleg_state_t *
2838 rfs4_finddeleg(rfs4_state_t *sp, bool_t *create)
2839 {
2840 	rfs4_deleg_state_t ds, *dsp;
2841 
2842 	ds.client = sp->owner->client;
2843 	ds.finfo = sp->finfo;
2844 
2845 	dsp = (rfs4_deleg_state_t *)rfs4_dbsearch(rfs4_deleg_idx, &ds,
2846 	    create, &ds, RFS4_DBS_VALID);
2847 
2848 	return (dsp);
2849 }
2850 
2851 rfs4_deleg_state_t *
2852 rfs4_finddelegstate(stateid_t *id)
2853 {
2854 	rfs4_deleg_state_t *dsp;
2855 	bool_t create = FALSE;
2856 
2857 	dsp = (rfs4_deleg_state_t *)rfs4_dbsearch(rfs4_deleg_state_idx, id,
2858 	    &create, NULL, RFS4_DBS_VALID);
2859 
2860 	return (dsp);
2861 }
2862 
2863 void
2864 rfs4_deleg_state_rele(rfs4_deleg_state_t *dsp)
2865 {
2866 	rfs4_dbe_rele(dsp->dbe);
2867 }
2868 
2869 void
2870 rfs4_update_lock_sequence(rfs4_lo_state_t *lsp)
2871 {
2872 
2873 	rfs4_dbe_lock(lsp->dbe);
2874 
2875 	/*
2876 	 * If we are skipping sequence id checking, this means that
2877 	 * this is the first lock request and therefore the sequence
2878 	 * id does not need to be updated.  This only happens on the
2879 	 * first lock request for a lockowner
2880 	 */
2881 	if (!lsp->skip_seqid_check)
2882 		lsp->seqid++;
2883 
2884 	rfs4_dbe_unlock(lsp->dbe);
2885 }
2886 
2887 void
2888 rfs4_update_lock_resp(rfs4_lo_state_t *lsp, nfs_resop4 *resp)
2889 {
2890 
2891 	rfs4_dbe_lock(lsp->dbe);
2892 
2893 	rfs4_free_reply(lsp->reply);
2894 
2895 	rfs4_copy_reply(lsp->reply, resp);
2896 
2897 	rfs4_dbe_unlock(lsp->dbe);
2898 }
2899 
2900 void
2901 rfs4_free_opens(rfs4_openowner_t *op, bool_t invalidate,
2902 	bool_t close_of_client)
2903 {
2904 	rfs4_state_t *sp;
2905 
2906 	rfs4_dbe_lock(op->dbe);
2907 
2908 	for (sp = op->ownerstateids.next->sp; sp != NULL;
2909 	    sp = sp->ownerstateids.next->sp) {
2910 		rfs4_state_close(sp, FALSE, close_of_client, CRED());
2911 		if (invalidate == TRUE)
2912 			rfs4_dbe_invalidate(sp->dbe);
2913 	}
2914 
2915 	rfs4_dbe_unlock(op->dbe);
2916 	rfs4_dbe_invalidate(op->dbe);
2917 }
2918 
2919 static uint32_t
2920 state_owner_file_hash(void *key)
2921 {
2922 	rfs4_state_t *sp = key;
2923 
2924 	return (ADDRHASH(sp->owner) ^ ADDRHASH(sp->finfo));
2925 }
2926 
2927 static bool_t
2928 state_owner_file_compare(rfs4_entry_t u_entry, void *key)
2929 {
2930 	rfs4_state_t *sp = (rfs4_state_t *)u_entry;
2931 	rfs4_state_t *arg = key;
2932 
2933 	if (sp->closed == TRUE)
2934 		return (FALSE);
2935 
2936 	return (arg->owner == sp->owner && arg->finfo == sp->finfo);
2937 }
2938 
2939 static void *
2940 state_owner_file_mkkey(rfs4_entry_t u_entry)
2941 {
2942 	return (u_entry);
2943 }
2944 
2945 static uint32_t
2946 state_file_hash(void *key)
2947 {
2948 	return (ADDRHASH(key));
2949 }
2950 
2951 static bool_t
2952 state_file_compare(rfs4_entry_t u_entry, void *key)
2953 {
2954 	rfs4_state_t *sp = (rfs4_state_t *)u_entry;
2955 	rfs4_file_t *fp = key;
2956 
2957 	if (sp->closed == TRUE)
2958 		return (FALSE);
2959 
2960 	return (fp == sp->finfo);
2961 }
2962 
2963 static void *
2964 state_file_mkkey(rfs4_entry_t u_entry)
2965 {
2966 	rfs4_state_t *sp = (rfs4_state_t *)u_entry;
2967 
2968 	return (sp->finfo);
2969 }
2970 
2971 rfs4_state_t *
2972 rfs4_findstate_by_owner_file(rfs4_openowner_t *op, rfs4_file_t *file,
2973 	bool_t *create)
2974 {
2975 	rfs4_state_t *sp;
2976 	rfs4_state_t key;
2977 
2978 	key.owner = op;
2979 	key.finfo = file;
2980 
2981 	sp = (rfs4_state_t *)rfs4_dbsearch(rfs4_state_owner_file_idx, &key,
2982 	    create, &key, RFS4_DBS_VALID);
2983 
2984 	return (sp);
2985 }
2986 
2987 /* This returns ANY state struct that refers to this file */
2988 static rfs4_state_t *
2989 rfs4_findstate_by_file(rfs4_file_t *fp)
2990 {
2991 	bool_t create = FALSE;
2992 
2993 	return ((rfs4_state_t *)rfs4_dbsearch(rfs4_state_file_idx, fp,
2994 	    &create, fp, RFS4_DBS_VALID));
2995 }
2996 
2997 static bool_t
2998 rfs4_state_expiry(rfs4_entry_t u_entry)
2999 {
3000 	rfs4_state_t *sp = (rfs4_state_t *)u_entry;
3001 
3002 	if (rfs4_dbe_is_invalid(sp->dbe))
3003 		return (TRUE);
3004 
3005 	if (sp->closed == TRUE &&
3006 	    ((gethrestime_sec() - rfs4_dbe_get_timerele(sp->dbe))
3007 	    > rfs4_lease_time))
3008 		return (TRUE);
3009 
3010 	return ((gethrestime_sec() - sp->owner->client->last_access
3011 	    > rfs4_lease_time));
3012 }
3013 
3014 static bool_t
3015 rfs4_state_create(rfs4_entry_t u_entry, void *argp)
3016 {
3017 	rfs4_state_t *sp = (rfs4_state_t *)u_entry;
3018 	rfs4_file_t *fp = ((rfs4_state_t *)argp)->finfo;
3019 	rfs4_openowner_t *op = ((rfs4_state_t *)argp)->owner;
3020 
3021 	rfs4_dbe_hold(fp->dbe);
3022 	rfs4_dbe_hold(op->dbe);
3023 	sp->stateid = get_stateid(rfs4_dbe_getid(sp->dbe));
3024 	sp->stateid.bits.type = OPENID;
3025 	sp->owner = op;
3026 	sp->finfo = fp;
3027 
3028 	/* Init lists for remque/insque */
3029 	sp->ownerstateids.next = sp->ownerstateids.prev = &sp->ownerstateids;
3030 	sp->ownerstateids.sp = sp;
3031 	sp->lockownerlist.next = sp->lockownerlist.prev = &sp->lockownerlist;
3032 	sp->lockownerlist.lsp = NULL;
3033 
3034 	/* Insert state on per open owner's list */
3035 	rfs4_dbe_lock(op->dbe);
3036 
3037 	insque(&sp->ownerstateids, op->ownerstateids.prev);
3038 
3039 	rfs4_dbe_unlock(op->dbe);
3040 
3041 	return (TRUE);
3042 }
3043 
3044 static rfs4_state_t *
3045 rfs4_findstate(stateid_t *id, rfs4_dbsearch_type_t find_invalid,
3046 		bool_t lock_fp)
3047 {
3048 	rfs4_state_t *sp;
3049 	bool_t create = FALSE;
3050 
3051 	sp = (rfs4_state_t *)rfs4_dbsearch(rfs4_state_idx, id,
3052 	    &create, NULL, find_invalid);
3053 	if (lock_fp == TRUE && sp != NULL)
3054 		rw_enter(&sp->finfo->file_rwlock, RW_READER);
3055 
3056 	return (sp);
3057 }
3058 
3059 void
3060 rfs4_state_close(rfs4_state_t *sp, bool_t lock_held,
3061 			bool_t close_of_client, cred_t *cr)
3062 {
3063 	/* Remove the associated lo_state owners */
3064 	if (!lock_held)
3065 		rfs4_dbe_lock(sp->dbe);
3066 
3067 	/*
3068 	 * If refcnt == 0, the dbe is about to be destroyed.
3069 	 * lock state will be released by the reaper thread.
3070 	 */
3071 
3072 	if (rfs4_dbe_refcnt(sp->dbe) > 0) {
3073 		if (sp->closed == FALSE) {
3074 			sp->closed = TRUE;
3075 
3076 			rfs4_release_share_lock_state(sp, cr, close_of_client);
3077 		}
3078 	}
3079 
3080 	if (!lock_held)
3081 		rfs4_dbe_unlock(sp->dbe);
3082 }
3083 
3084 /*
3085  * Remove all state associated with the given client.
3086  */
3087 void
3088 rfs4_client_state_remove(rfs4_client_t *cp)
3089 {
3090 	rfs4_openowner_t *oop;
3091 
3092 	rfs4_dbe_lock(cp->dbe);
3093 
3094 	for (oop = cp->openownerlist.next->oop;  oop != NULL;
3095 	    oop = oop->openownerlist.next->oop) {
3096 		rfs4_free_opens(oop, TRUE, TRUE);
3097 	}
3098 
3099 	rfs4_dbe_unlock(cp->dbe);
3100 }
3101 
3102 void
3103 rfs4_client_close(rfs4_client_t *cp)
3104 {
3105 	/* Mark client as going away. */
3106 	rfs4_dbe_lock(cp->dbe);
3107 	rfs4_dbe_invalidate(cp->dbe);
3108 	rfs4_dbe_unlock(cp->dbe);
3109 
3110 	rfs4_client_state_remove(cp);
3111 
3112 	/* Release the client */
3113 	rfs4_client_rele(cp);
3114 }
3115 
3116 nfsstat4
3117 rfs4_check_clientid(clientid4 *cp, int setclid_confirm)
3118 {
3119 	cid *cidp = (cid *) cp;
3120 
3121 	/*
3122 	 * If we are booted as a cluster node, check the embedded nodeid.
3123 	 * If it indicates that this clientid was generated on another node,
3124 	 * inform the client accordingly.
3125 	 */
3126 	if (cluster_bootflags & CLUSTER_BOOTED && foreign_clientid(cidp))
3127 		return (NFS4ERR_STALE_CLIENTID);
3128 
3129 	/*
3130 	 * If the server start time matches the time provided
3131 	 * by the client (via the clientid) and this is NOT a
3132 	 * setclientid_confirm then return EXPIRED.
3133 	 */
3134 	if (!setclid_confirm && cidp->impl_id.start_time == rfs4_start_time)
3135 		return (NFS4ERR_EXPIRED);
3136 
3137 	return (NFS4ERR_STALE_CLIENTID);
3138 }
3139 
3140 /*
3141  * This is used when a stateid has not been found amongst the
3142  * current server's state.  Check the stateid to see if it
3143  * was from this server instantiation or not.
3144  */
3145 static nfsstat4
3146 what_stateid_error(stateid_t *id, stateid_type_t type)
3147 {
3148 	/* If we are booted as a cluster node, was stateid locally generated? */
3149 	if ((cluster_bootflags & CLUSTER_BOOTED) && foreign_stateid(id))
3150 		return (NFS4ERR_STALE_STATEID);
3151 
3152 	/* If types don't match then no use checking further */
3153 	if (type != id->bits.type)
3154 		return (NFS4ERR_BAD_STATEID);
3155 
3156 	/* From a previous server instantiation, return STALE */
3157 	if (id->bits.boottime < rfs4_start_time)
3158 		return (NFS4ERR_STALE_STATEID);
3159 
3160 	/*
3161 	 * From this server but the state is most likely beyond lease
3162 	 * timeout: return NFS4ERR_EXPIRED.  However, there is the
3163 	 * case of a delegation stateid.  For delegations, there is a
3164 	 * case where the state can be removed without the client's
3165 	 * knowledge/consent: revocation.  In the case of delegation
3166 	 * revocation, the delegation state will be removed and will
3167 	 * not be found.  If the client does something like a
3168 	 * DELEGRETURN or even a READ/WRITE with a delegatoin stateid
3169 	 * that has been revoked, the server should return BAD_STATEID
3170 	 * instead of the more common EXPIRED error.
3171 	 */
3172 	if (id->bits.boottime == rfs4_start_time) {
3173 		if (type == DELEGID)
3174 			return (NFS4ERR_BAD_STATEID);
3175 		else
3176 			return (NFS4ERR_EXPIRED);
3177 	}
3178 
3179 	return (NFS4ERR_BAD_STATEID);
3180 }
3181 
3182 /*
3183  * Used later on to find the various state structs.  When called from
3184  * rfs4_check_stateid()->rfs4_get_all_state(), no file struct lock is
3185  * taken (it is not needed) and helps on the read/write path with
3186  * respect to performance.
3187  */
3188 static nfsstat4
3189 rfs4_get_state_lockit(stateid4 *stateid, rfs4_state_t **spp,
3190 		rfs4_dbsearch_type_t find_invalid, bool_t lock_fp)
3191 {
3192 	stateid_t *id = (stateid_t *)stateid;
3193 	rfs4_state_t *sp;
3194 
3195 	*spp = NULL;
3196 
3197 	/* If we are booted as a cluster node, was stateid locally generated? */
3198 	if ((cluster_bootflags & CLUSTER_BOOTED) && foreign_stateid(id))
3199 		return (NFS4ERR_STALE_STATEID);
3200 
3201 	sp = rfs4_findstate(id, find_invalid, lock_fp);
3202 	if (sp == NULL) {
3203 		return (what_stateid_error(id, OPENID));
3204 	}
3205 
3206 	if (rfs4_lease_expired(sp->owner->client)) {
3207 		if (lock_fp == TRUE)
3208 			rfs4_state_rele(sp);
3209 		else
3210 			rfs4_state_rele_nounlock(sp);
3211 		return (NFS4ERR_EXPIRED);
3212 	}
3213 
3214 	*spp = sp;
3215 
3216 	return (NFS4_OK);
3217 }
3218 
3219 nfsstat4
3220 rfs4_get_state(stateid4 *stateid, rfs4_state_t **spp,
3221 		rfs4_dbsearch_type_t find_invalid)
3222 {
3223 	return (rfs4_get_state_lockit(stateid, spp, find_invalid, TRUE));
3224 }
3225 
3226 int
3227 rfs4_check_stateid_seqid(rfs4_state_t *sp, stateid4 *stateid)
3228 {
3229 	stateid_t *id = (stateid_t *)stateid;
3230 
3231 	if (rfs4_lease_expired(sp->owner->client))
3232 		return (NFS4_CHECK_STATEID_EXPIRED);
3233 
3234 	/* Stateid is some time in the future - that's bad */
3235 	if (sp->stateid.bits.chgseq < id->bits.chgseq)
3236 		return (NFS4_CHECK_STATEID_BAD);
3237 
3238 	if (sp->stateid.bits.chgseq == id->bits.chgseq + 1)
3239 		return (NFS4_CHECK_STATEID_REPLAY);
3240 
3241 	/* Stateid is some time in the past - that's old */
3242 	if (sp->stateid.bits.chgseq > id->bits.chgseq)
3243 		return (NFS4_CHECK_STATEID_OLD);
3244 
3245 	/* Caller needs to know about confirmation before closure */
3246 	if (sp->owner->need_confirm)
3247 		return (NFS4_CHECK_STATEID_UNCONFIRMED);
3248 
3249 	if (sp->closed == TRUE)
3250 		return (NFS4_CHECK_STATEID_CLOSED);
3251 
3252 	return (NFS4_CHECK_STATEID_OKAY);
3253 }
3254 
3255 int
3256 rfs4_check_lo_stateid_seqid(rfs4_lo_state_t *lsp, stateid4 *stateid)
3257 {
3258 	stateid_t *id = (stateid_t *)stateid;
3259 
3260 	if (rfs4_lease_expired(lsp->state->owner->client))
3261 		return (NFS4_CHECK_STATEID_EXPIRED);
3262 
3263 	/* Stateid is some time in the future - that's bad */
3264 	if (lsp->lockid.bits.chgseq < id->bits.chgseq)
3265 		return (NFS4_CHECK_STATEID_BAD);
3266 
3267 	if (lsp->lockid.bits.chgseq == id->bits.chgseq + 1)
3268 		return (NFS4_CHECK_STATEID_REPLAY);
3269 
3270 	/* Stateid is some time in the past - that's old */
3271 	if (lsp->lockid.bits.chgseq > id->bits.chgseq)
3272 		return (NFS4_CHECK_STATEID_OLD);
3273 
3274 	return (NFS4_CHECK_STATEID_OKAY);
3275 }
3276 
3277 nfsstat4
3278 rfs4_get_deleg_state(stateid4 *stateid, rfs4_deleg_state_t **dspp)
3279 {
3280 	stateid_t *id = (stateid_t *)stateid;
3281 	rfs4_deleg_state_t *dsp;
3282 
3283 	*dspp = NULL;
3284 
3285 	/* If we are booted as a cluster node, was stateid locally generated? */
3286 	if ((cluster_bootflags & CLUSTER_BOOTED) && foreign_stateid(id))
3287 		return (NFS4ERR_STALE_STATEID);
3288 
3289 	dsp = rfs4_finddelegstate(id);
3290 	if (dsp == NULL) {
3291 		return (what_stateid_error(id, DELEGID));
3292 	}
3293 
3294 	if (rfs4_lease_expired(dsp->client)) {
3295 		rfs4_deleg_state_rele(dsp);
3296 		return (NFS4ERR_EXPIRED);
3297 	}
3298 
3299 	*dspp = dsp;
3300 
3301 	return (NFS4_OK);
3302 }
3303 
3304 nfsstat4
3305 rfs4_get_lo_state(stateid4 *stateid, rfs4_lo_state_t **lspp, bool_t lock_fp)
3306 {
3307 	stateid_t *id = (stateid_t *)stateid;
3308 	rfs4_lo_state_t *lsp;
3309 
3310 	*lspp = NULL;
3311 
3312 	/* If we are booted as a cluster node, was stateid locally generated? */
3313 	if ((cluster_bootflags & CLUSTER_BOOTED) && foreign_stateid(id))
3314 		return (NFS4ERR_STALE_STATEID);
3315 
3316 	lsp = rfs4_findlo_state(id, lock_fp);
3317 	if (lsp == NULL) {
3318 		return (what_stateid_error(id, LOCKID));
3319 	}
3320 
3321 	if (rfs4_lease_expired(lsp->state->owner->client)) {
3322 		rfs4_lo_state_rele(lsp, lock_fp);
3323 		return (NFS4ERR_EXPIRED);
3324 	}
3325 
3326 	*lspp = lsp;
3327 
3328 	return (NFS4_OK);
3329 }
3330 
3331 static nfsstat4
3332 rfs4_get_all_state(stateid4 *sid, rfs4_state_t **spp,
3333 	rfs4_deleg_state_t **dspp, rfs4_lo_state_t **lospp)
3334 {
3335 	rfs4_state_t *sp = NULL;
3336 	rfs4_deleg_state_t *dsp = NULL;
3337 	rfs4_lo_state_t *losp = NULL;
3338 	stateid_t *id;
3339 	nfsstat4 status;
3340 
3341 	*spp = NULL; *dspp = NULL; *lospp = NULL;
3342 
3343 	id = (stateid_t *)sid;
3344 	switch (id->bits.type) {
3345 	case OPENID:
3346 		status = rfs4_get_state_lockit(sid, &sp, FALSE, FALSE);
3347 		break;
3348 	case DELEGID:
3349 		status = rfs4_get_deleg_state(sid, &dsp);
3350 		break;
3351 	case LOCKID:
3352 		status = rfs4_get_lo_state(sid, &losp, FALSE);
3353 		if (status == NFS4_OK) {
3354 			sp = losp->state;
3355 			rfs4_dbe_hold(sp->dbe);
3356 		}
3357 		break;
3358 	default:
3359 		status = NFS4ERR_BAD_STATEID;
3360 	}
3361 
3362 	if (status == NFS4_OK) {
3363 		*spp = sp;
3364 		*dspp = dsp;
3365 		*lospp = losp;
3366 	}
3367 
3368 	return (status);
3369 }
3370 
3371 /*
3372  * Given the I/O mode (FREAD or FWRITE), this checks whether the
3373  * rfs4_state_t struct has access to do this operation and if so
3374  * return NFS4_OK; otherwise the proper NFSv4 error is returned.
3375  */
3376 nfsstat4
3377 rfs4_state_has_access(rfs4_state_t *sp, int mode, vnode_t *vp)
3378 {
3379 	nfsstat4 stat = NFS4_OK;
3380 	rfs4_file_t *fp;
3381 	bool_t create = FALSE;
3382 
3383 	rfs4_dbe_lock(sp->dbe);
3384 	if (mode == FWRITE) {
3385 		if (!(sp->share_access & OPEN4_SHARE_ACCESS_WRITE)) {
3386 			stat = NFS4ERR_OPENMODE;
3387 		}
3388 	} else if (mode == FREAD) {
3389 		if (!(sp->share_access & OPEN4_SHARE_ACCESS_READ)) {
3390 			/*
3391 			 * If we have OPENed the file with DENYing access
3392 			 * to both READ and WRITE then no one else could
3393 			 * have OPENed the file, hence no conflicting READ
3394 			 * deny.  This check is merely an optimization.
3395 			 */
3396 			if (sp->share_deny == OPEN4_SHARE_DENY_BOTH)
3397 				goto out;
3398 
3399 			/* Check against file struct's DENY mode */
3400 			fp = rfs4_findfile(vp, NULL, &create);
3401 			if (fp != NULL) {
3402 				int deny_read = 0;
3403 				rfs4_dbe_lock(fp->dbe);
3404 				/*
3405 				 * Check if any other open owner has the file
3406 				 * OPENed with deny READ.
3407 				 */
3408 				if (sp->share_deny & OPEN4_SHARE_DENY_READ)
3409 					deny_read = 1;
3410 				ASSERT(fp->deny_read - deny_read >= 0);
3411 				if (fp->deny_read - deny_read > 0)
3412 					stat = NFS4ERR_OPENMODE;
3413 				rfs4_dbe_unlock(fp->dbe);
3414 				rfs4_file_rele(fp);
3415 			}
3416 		}
3417 	} else {
3418 		/* Illegal I/O mode */
3419 		stat = NFS4ERR_INVAL;
3420 	}
3421 out:
3422 	rfs4_dbe_unlock(sp->dbe);
3423 	return (stat);
3424 }
3425 
3426 /*
3427  * Given the I/O mode (FREAD or FWRITE), the vnode, the stateid and whether
3428  * the file is being truncated, return NFS4_OK if allowed or approriate
3429  * V4 error if not. Note NFS4ERR_DELAY will be returned and a recall on
3430  * the associated file will be done if the I/O is not consistent with any
3431  * delegation in effect on the file. Should be holding VOP_RWLOCK, either
3432  * as reader or writer as appropriate. rfs4_op_open will accquire the
3433  * VOP_RWLOCK as writer when setting up delegation. If the stateid is bad
3434  * this routine will return NFS4ERR_BAD_STATEID. In addition, through the
3435  * deleg parameter, we will return whether a write delegation is held by
3436  * the client associated with this stateid.
3437  * If the server instance associated with the relevant client is in its
3438  * grace period, return NFS4ERR_GRACE.
3439  */
3440 
3441 nfsstat4
3442 rfs4_check_stateid(int mode, vnode_t *vp,
3443 		stateid4 *stateid, bool_t trunc, bool_t *deleg,
3444 		bool_t do_access)
3445 {
3446 	rfs4_file_t *fp;
3447 	bool_t create = FALSE;
3448 	rfs4_state_t *sp;
3449 	rfs4_deleg_state_t *dsp;
3450 	rfs4_lo_state_t *lsp;
3451 	stateid_t *id = (stateid_t *)stateid;
3452 	nfsstat4 stat = NFS4_OK;
3453 
3454 	if (ISSPECIAL(stateid)) {
3455 		fp = rfs4_findfile(vp, NULL, &create);
3456 		if (fp == NULL)
3457 			return (NFS4_OK);
3458 		if (fp->dinfo->dtype == OPEN_DELEGATE_NONE) {
3459 			rfs4_file_rele(fp);
3460 			return (NFS4_OK);
3461 		}
3462 		if (mode == FWRITE ||
3463 		    fp->dinfo->dtype == OPEN_DELEGATE_WRITE) {
3464 			rfs4_recall_deleg(fp, trunc, NULL);
3465 			rfs4_file_rele(fp);
3466 			return (NFS4ERR_DELAY);
3467 		}
3468 		rfs4_file_rele(fp);
3469 		return (NFS4_OK);
3470 	} else {
3471 		stat = rfs4_get_all_state(stateid, &sp, &dsp, &lsp);
3472 		if (stat != NFS4_OK)
3473 			return (stat);
3474 		if (lsp != NULL) {
3475 			/* Is associated server instance in its grace period? */
3476 			if (rfs4_clnt_in_grace(lsp->locker->client)) {
3477 				rfs4_lo_state_rele(lsp, FALSE);
3478 				if (sp != NULL)
3479 					rfs4_state_rele_nounlock(sp);
3480 				return (NFS4ERR_GRACE);
3481 			}
3482 			if (id->bits.type == LOCKID) {
3483 				/* Seqid in the future? - that's bad */
3484 				if (lsp->lockid.bits.chgseq <
3485 				    id->bits.chgseq) {
3486 					rfs4_lo_state_rele(lsp, FALSE);
3487 					if (sp != NULL)
3488 						rfs4_state_rele_nounlock(sp);
3489 					return (NFS4ERR_BAD_STATEID);
3490 				}
3491 				/* Seqid in the past? - that's old */
3492 				if (lsp->lockid.bits.chgseq >
3493 				    id->bits.chgseq) {
3494 					rfs4_lo_state_rele(lsp, FALSE);
3495 					if (sp != NULL)
3496 						rfs4_state_rele_nounlock(sp);
3497 					return (NFS4ERR_OLD_STATEID);
3498 				}
3499 				/* Ensure specified filehandle matches */
3500 				if (lsp->state->finfo->vp != vp) {
3501 					rfs4_lo_state_rele(lsp, FALSE);
3502 					if (sp != NULL)
3503 						rfs4_state_rele_nounlock(sp);
3504 					return (NFS4ERR_BAD_STATEID);
3505 				}
3506 			}
3507 			rfs4_lo_state_rele(lsp, FALSE);
3508 		}
3509 
3510 		/* Stateid provided was an "open" stateid */
3511 		if (sp != NULL) {
3512 			/* Is associated server instance in its grace period? */
3513 			if (rfs4_clnt_in_grace(sp->owner->client)) {
3514 				rfs4_state_rele_nounlock(sp);
3515 				return (NFS4ERR_GRACE);
3516 			}
3517 			if (id->bits.type == OPENID) {
3518 				/* Seqid in the future? - that's bad */
3519 				if (sp->stateid.bits.chgseq <
3520 				    id->bits.chgseq) {
3521 					rfs4_state_rele_nounlock(sp);
3522 					return (NFS4ERR_BAD_STATEID);
3523 				}
3524 				/* Seqid in the past - that's old */
3525 				if (sp->stateid.bits.chgseq >
3526 				    id->bits.chgseq) {
3527 					rfs4_state_rele_nounlock(sp);
3528 					return (NFS4ERR_OLD_STATEID);
3529 				}
3530 			}
3531 			/* Ensure specified filehandle matches */
3532 			if (sp->finfo->vp != vp) {
3533 				rfs4_state_rele_nounlock(sp);
3534 				return (NFS4ERR_BAD_STATEID);
3535 			}
3536 
3537 			if (sp->owner->need_confirm) {
3538 				rfs4_state_rele_nounlock(sp);
3539 				return (NFS4ERR_BAD_STATEID);
3540 			}
3541 
3542 			if (sp->closed == TRUE) {
3543 				rfs4_state_rele_nounlock(sp);
3544 				return (NFS4ERR_OLD_STATEID);
3545 			}
3546 
3547 			if (do_access)
3548 				stat = rfs4_state_has_access(sp, mode, vp);
3549 			else
3550 				stat = NFS4_OK;
3551 
3552 			/*
3553 			 * Return whether this state has write
3554 			 * delegation if desired
3555 			 */
3556 			if (deleg &&
3557 			    (sp->finfo->dinfo->dtype == OPEN_DELEGATE_WRITE))
3558 				*deleg = TRUE;
3559 
3560 			/*
3561 			 * We got a valid stateid, so we update the
3562 			 * lease on the client. Ideally we would like
3563 			 * to do this after the calling op succeeds,
3564 			 * but for now this will be good
3565 			 * enough. Callers of this routine are
3566 			 * currently insulated from the state stuff.
3567 			 */
3568 			rfs4_update_lease(sp->owner->client);
3569 
3570 			/*
3571 			 * If a delegation is present on this file and
3572 			 * this is a WRITE, then update the lastwrite
3573 			 * time to indicate that activity is present.
3574 			 */
3575 			if (sp->finfo->dinfo->dtype == OPEN_DELEGATE_WRITE &&
3576 			    mode == FWRITE) {
3577 				sp->finfo->dinfo->time_lastwrite =
3578 				    gethrestime_sec();
3579 			}
3580 
3581 			rfs4_state_rele_nounlock(sp);
3582 
3583 			return (stat);
3584 		}
3585 
3586 		if (dsp != NULL) {
3587 			/* Is associated server instance in its grace period? */
3588 			if (rfs4_clnt_in_grace(dsp->client)) {
3589 				rfs4_deleg_state_rele(dsp);
3590 				return (NFS4ERR_GRACE);
3591 			}
3592 			if (dsp->delegid.bits.chgseq !=	id->bits.chgseq) {
3593 				rfs4_deleg_state_rele(dsp);
3594 				return (NFS4ERR_BAD_STATEID);
3595 			}
3596 
3597 			/* Ensure specified filehandle matches */
3598 			if (dsp->finfo->vp != vp) {
3599 				rfs4_deleg_state_rele(dsp);
3600 				return (NFS4ERR_BAD_STATEID);
3601 			}
3602 			/*
3603 			 * Return whether this state has write
3604 			 * delegation if desired
3605 			 */
3606 			if (deleg &&
3607 			    (dsp->finfo->dinfo->dtype == OPEN_DELEGATE_WRITE))
3608 				*deleg = TRUE;
3609 
3610 			rfs4_update_lease(dsp->client);
3611 
3612 			/*
3613 			 * If a delegation is present on this file and
3614 			 * this is a WRITE, then update the lastwrite
3615 			 * time to indicate that activity is present.
3616 			 */
3617 			if (dsp->finfo->dinfo->dtype == OPEN_DELEGATE_WRITE &&
3618 			    mode == FWRITE) {
3619 				dsp->finfo->dinfo->time_lastwrite =
3620 				    gethrestime_sec();
3621 			}
3622 
3623 			/*
3624 			 * XXX - what happens if this is a WRITE and the
3625 			 * delegation type of for READ.
3626 			 */
3627 			rfs4_deleg_state_rele(dsp);
3628 
3629 			return (stat);
3630 		}
3631 		/*
3632 		 * If we got this far, something bad happened
3633 		 */
3634 		return (NFS4ERR_BAD_STATEID);
3635 	}
3636 }
3637 
3638 
3639 /*
3640  * This is a special function in that for the file struct provided the
3641  * server wants to remove/close all current state associated with the
3642  * file.  The prime use of this would be with OP_REMOVE to force the
3643  * release of state and particularly of file locks.
3644  *
3645  * There is an assumption that there is no delegations outstanding on
3646  * this file at this point.  The caller should have waited for those
3647  * to be returned or revoked.
3648  */
3649 void
3650 rfs4_close_all_state(rfs4_file_t *fp)
3651 {
3652 	rfs4_state_t *sp;
3653 
3654 	rfs4_dbe_lock(fp->dbe);
3655 
3656 #ifdef DEBUG
3657 	/* only applies when server is handing out delegations */
3658 	if (rfs4_deleg_policy != SRV_NEVER_DELEGATE)
3659 		ASSERT(fp->dinfo->hold_grant > 0);
3660 #endif
3661 
3662 	/* No delegations for this file */
3663 	ASSERT(fp->delegationlist.next == &fp->delegationlist);
3664 
3665 	/* Make sure that it can not be found */
3666 	rfs4_dbe_invalidate(fp->dbe);
3667 
3668 	if (fp->vp == NULL) {
3669 		rfs4_dbe_unlock(fp->dbe);
3670 		return;
3671 	}
3672 	rfs4_dbe_unlock(fp->dbe);
3673 
3674 	/*
3675 	 * Hold as writer to prevent other server threads from
3676 	 * processing requests related to the file while all state is
3677 	 * being removed.
3678 	 */
3679 	rw_enter(&fp->file_rwlock, RW_WRITER);
3680 
3681 	/* Remove ALL state from the file */
3682 	while (sp = rfs4_findstate_by_file(fp)) {
3683 		rfs4_state_close(sp, FALSE, FALSE, CRED());
3684 		rfs4_state_rele_nounlock(sp);
3685 	}
3686 
3687 	/*
3688 	 * This is only safe since there are no further references to
3689 	 * the file.
3690 	 */
3691 	rfs4_dbe_lock(fp->dbe);
3692 	if (fp->vp) {
3693 		vnode_t *vp = fp->vp;
3694 
3695 		mutex_enter(&vp->v_lock);
3696 		(void) vsd_set(vp, nfs4_srv_vkey, NULL);
3697 		mutex_exit(&vp->v_lock);
3698 		VN_RELE(vp);
3699 		fp->vp = NULL;
3700 	}
3701 	rfs4_dbe_unlock(fp->dbe);
3702 
3703 	/* Finally let other references to proceed */
3704 	rw_exit(&fp->file_rwlock);
3705 }
3706 
3707 /*
3708  * This function is used as a target for the rfs4_dbe_walk() call
3709  * below.  The purpose of this function is to see if the
3710  * lockowner_state refers to a file that resides within the exportinfo
3711  * export.  If so, then remove the lock_owner state (file locks and
3712  * share "locks") for this object since the intent is the server is
3713  * unexporting the specified directory.  Be sure to invalidate the
3714  * object after the state has been released
3715  */
3716 static void
3717 rfs4_lo_state_walk_callout(rfs4_entry_t u_entry, void *e)
3718 {
3719 	rfs4_lo_state_t *lsp = (rfs4_lo_state_t *)u_entry;
3720 	struct exportinfo *exi = (struct exportinfo *)e;
3721 	nfs_fh4_fmt_t   fhfmt4, *exi_fhp, *finfo_fhp;
3722 	fhandle_t *efhp;
3723 
3724 	efhp = (fhandle_t *)&exi->exi_fh;
3725 	exi_fhp = (nfs_fh4_fmt_t *)&fhfmt4;
3726 
3727 	FH_TO_FMT4(efhp, exi_fhp);
3728 
3729 	finfo_fhp =
3730 	    (nfs_fh4_fmt_t *)lsp->state->finfo->filehandle.nfs_fh4_val;
3731 
3732 	if (EQFSID(&finfo_fhp->fh4_fsid, &exi_fhp->fh4_fsid) &&
3733 	    bcmp(&finfo_fhp->fh4_xdata, &exi_fhp->fh4_xdata,
3734 	    exi_fhp->fh4_xlen) == 0) {
3735 		rfs4_state_close(lsp->state, FALSE, FALSE, CRED());
3736 		rfs4_dbe_invalidate(lsp->dbe);
3737 		rfs4_dbe_invalidate(lsp->state->dbe);
3738 	}
3739 }
3740 
3741 /*
3742  * This function is used as a target for the rfs4_dbe_walk() call
3743  * below.  The purpose of this function is to see if the state refers
3744  * to a file that resides within the exportinfo export.  If so, then
3745  * remove the open state for this object since the intent is the
3746  * server is unexporting the specified directory.  The main result for
3747  * this type of entry is to invalidate it such it will not be found in
3748  * the future.
3749  */
3750 static void
3751 rfs4_state_walk_callout(rfs4_entry_t u_entry, void *e)
3752 {
3753 	rfs4_state_t *sp = (rfs4_state_t *)u_entry;
3754 	struct exportinfo *exi = (struct exportinfo *)e;
3755 	nfs_fh4_fmt_t   fhfmt4, *exi_fhp, *finfo_fhp;
3756 	fhandle_t *efhp;
3757 
3758 	efhp = (fhandle_t *)&exi->exi_fh;
3759 	exi_fhp = (nfs_fh4_fmt_t *)&fhfmt4;
3760 
3761 	FH_TO_FMT4(efhp, exi_fhp);
3762 
3763 	finfo_fhp =
3764 	    (nfs_fh4_fmt_t *)sp->finfo->filehandle.nfs_fh4_val;
3765 
3766 	if (EQFSID(&finfo_fhp->fh4_fsid, &exi_fhp->fh4_fsid) &&
3767 	    bcmp(&finfo_fhp->fh4_xdata, &exi_fhp->fh4_xdata,
3768 	    exi_fhp->fh4_xlen) == 0) {
3769 		rfs4_state_close(sp, TRUE, FALSE, CRED());
3770 		rfs4_dbe_invalidate(sp->dbe);
3771 	}
3772 }
3773 
3774 /*
3775  * This function is used as a target for the rfs4_dbe_walk() call
3776  * below.  The purpose of this function is to see if the state refers
3777  * to a file that resides within the exportinfo export.  If so, then
3778  * remove the deleg state for this object since the intent is the
3779  * server is unexporting the specified directory.  The main result for
3780  * this type of entry is to invalidate it such it will not be found in
3781  * the future.
3782  */
3783 static void
3784 rfs4_deleg_state_walk_callout(rfs4_entry_t u_entry, void *e)
3785 {
3786 	rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
3787 	struct exportinfo *exi = (struct exportinfo *)e;
3788 	nfs_fh4_fmt_t   fhfmt4, *exi_fhp, *finfo_fhp;
3789 	fhandle_t *efhp;
3790 
3791 	efhp = (fhandle_t *)&exi->exi_fh;
3792 	exi_fhp = (nfs_fh4_fmt_t *)&fhfmt4;
3793 
3794 	FH_TO_FMT4(efhp, exi_fhp);
3795 
3796 	finfo_fhp =
3797 	    (nfs_fh4_fmt_t *)dsp->finfo->filehandle.nfs_fh4_val;
3798 
3799 	if (EQFSID(&finfo_fhp->fh4_fsid, &exi_fhp->fh4_fsid) &&
3800 	    bcmp(&finfo_fhp->fh4_xdata, &exi_fhp->fh4_xdata,
3801 	    exi_fhp->fh4_xlen) == 0) {
3802 		rfs4_dbe_invalidate(dsp->dbe);
3803 	}
3804 }
3805 
3806 /*
3807  * This function is used as a target for the rfs4_dbe_walk() call
3808  * below.  The purpose of this function is to see if the state refers
3809  * to a file that resides within the exportinfo export.  If so, then
3810  * release vnode hold for this object since the intent is the server
3811  * is unexporting the specified directory.  Invalidation will prevent
3812  * this struct from being found in the future.
3813  */
3814 static void
3815 rfs4_file_walk_callout(rfs4_entry_t u_entry, void *e)
3816 {
3817 	rfs4_file_t *fp = (rfs4_file_t *)u_entry;
3818 	struct exportinfo *exi = (struct exportinfo *)e;
3819 	nfs_fh4_fmt_t   fhfmt4, *exi_fhp, *finfo_fhp;
3820 	fhandle_t *efhp;
3821 
3822 	efhp = (fhandle_t *)&exi->exi_fh;
3823 	exi_fhp = (nfs_fh4_fmt_t *)&fhfmt4;
3824 
3825 	FH_TO_FMT4(efhp, exi_fhp);
3826 
3827 	finfo_fhp = (nfs_fh4_fmt_t *)fp->filehandle.nfs_fh4_val;
3828 
3829 	if (EQFSID(&finfo_fhp->fh4_fsid, &exi_fhp->fh4_fsid) &&
3830 	    bcmp(&finfo_fhp->fh4_xdata, &exi_fhp->fh4_xdata,
3831 	    exi_fhp->fh4_xlen) == 0) {
3832 		if (fp->vp) {
3833 			vnode_t *vp = fp->vp;
3834 
3835 			/* don't leak monitors */
3836 			if (fp->dinfo->dtype == OPEN_DELEGATE_READ)
3837 				(void) fem_uninstall(vp, deleg_rdops,
3838 				    (void *)fp);
3839 			else if (fp->dinfo->dtype == OPEN_DELEGATE_WRITE)
3840 				(void) fem_uninstall(vp, deleg_wrops,
3841 				    (void *)fp);
3842 			mutex_enter(&vp->v_lock);
3843 			(void) vsd_set(vp, nfs4_srv_vkey, NULL);
3844 			mutex_exit(&vp->v_lock);
3845 			VN_RELE(vp);
3846 			fp->vp = NULL;
3847 		}
3848 		rfs4_dbe_invalidate(fp->dbe);
3849 	}
3850 }
3851 
3852 /*
3853  * Given a directory that is being unexported, cleanup/release all
3854  * state in the server that refers to objects residing underneath this
3855  * particular export.  The ordering of the release is important.
3856  * Lock_owner, then state and then file.
3857  */
3858 void
3859 rfs4_clean_state_exi(struct exportinfo *exi)
3860 {
3861 	mutex_enter(&rfs4_state_lock);
3862 
3863 	if (rfs4_server_state == NULL) {
3864 		mutex_exit(&rfs4_state_lock);
3865 		return;
3866 	}
3867 
3868 	rfs4_dbe_walk(rfs4_lo_state_tab, rfs4_lo_state_walk_callout, exi);
3869 	rfs4_dbe_walk(rfs4_state_tab, rfs4_state_walk_callout, exi);
3870 	rfs4_dbe_walk(rfs4_deleg_state_tab, rfs4_deleg_state_walk_callout, exi);
3871 	rfs4_dbe_walk(rfs4_file_tab, rfs4_file_walk_callout, exi);
3872 
3873 	mutex_exit(&rfs4_state_lock);
3874 }
3875