xref: /titanic_41/usr/src/uts/common/fs/nfs/nfs4_state.c (revision f8919bdadda3ebb97bd55cc14a16e0271ed57615)
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 extern u_longlong_t nfs4_srv_caller_id;
45 
46 extern time_t rfs4_start_time;
47 extern uint_t nfs4_srv_vkey;
48 
49 stateid4 special0 = {
50 	0,
51 	{ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 }
52 };
53 
54 stateid4 special1 = {
55 	0xffffffff,
56 	{
57 		(char)0xff, (char)0xff, (char)0xff, (char)0xff,
58 		(char)0xff, (char)0xff, (char)0xff, (char)0xff,
59 		(char)0xff, (char)0xff, (char)0xff, (char)0xff
60 	}
61 };
62 
63 
64 #define	ISSPECIAL(id)  (stateid4_cmp(id, &special0) || \
65 			stateid4_cmp(id, &special1))
66 
67 /* For embedding the cluster nodeid into our clientid */
68 #define	CLUSTER_NODEID_SHIFT	24
69 #define	CLUSTER_MAX_NODEID	255
70 
71 #ifdef DEBUG
72 int rfs4_debug;
73 #endif
74 
75 static uint32_t rfs4_database_debug = 0x00;
76 
77 static void rfs4_ss_clid_write(rfs4_client_t *cp, char *leaf);
78 static void rfs4_ss_clid_write_one(rfs4_client_t *cp, char *dir, char *leaf);
79 static void rfs4_dss_clear_oldstate(rfs4_servinst_t *sip);
80 static void rfs4_ss_chkclid_sip(rfs4_client_t *cp, rfs4_servinst_t *sip);
81 
82 /*
83  * Couple of simple init/destroy functions for a general waiter
84  */
85 void
86 rfs4_sw_init(rfs4_state_wait_t *swp)
87 {
88 	mutex_init(swp->sw_cv_lock, NULL, MUTEX_DEFAULT, NULL);
89 	cv_init(swp->sw_cv, NULL, CV_DEFAULT, NULL);
90 	swp->sw_active = FALSE;
91 	swp->sw_wait_count = 0;
92 }
93 
94 void
95 rfs4_sw_destroy(rfs4_state_wait_t *swp)
96 {
97 	mutex_destroy(swp->sw_cv_lock);
98 	cv_destroy(swp->sw_cv);
99 }
100 
101 void
102 rfs4_sw_enter(rfs4_state_wait_t *swp)
103 {
104 	mutex_enter(swp->sw_cv_lock);
105 	while (swp->sw_active) {
106 		swp->sw_wait_count++;
107 		cv_wait(swp->sw_cv, swp->sw_cv_lock);
108 		swp->sw_wait_count--;
109 	}
110 	ASSERT(swp->sw_active == FALSE);
111 	swp->sw_active = TRUE;
112 	mutex_exit(swp->sw_cv_lock);
113 }
114 
115 void
116 rfs4_sw_exit(rfs4_state_wait_t *swp)
117 {
118 	mutex_enter(swp->sw_cv_lock);
119 	ASSERT(swp->sw_active == TRUE);
120 	swp->sw_active = FALSE;
121 	if (swp->sw_wait_count != 0)
122 		cv_broadcast(swp->sw_cv);
123 	mutex_exit(swp->sw_cv_lock);
124 }
125 
126 /*
127  * CPR callback id -- not related to v4 callbacks
128  */
129 static callb_id_t cpr_id = 0;
130 
131 static void
132 deep_lock_copy(LOCK4res *dres, LOCK4res *sres)
133 {
134 	lock_owner4 *slo = &sres->LOCK4res_u.denied.owner;
135 	lock_owner4 *dlo = &dres->LOCK4res_u.denied.owner;
136 
137 	if (sres->status == NFS4ERR_DENIED) {
138 		dlo->owner_val = kmem_alloc(slo->owner_len, KM_SLEEP);
139 		bcopy(slo->owner_val, dlo->owner_val, slo->owner_len);
140 	}
141 }
142 
143 static void
144 deep_lock_free(LOCK4res *res)
145 {
146 	lock_owner4 *lo = &res->LOCK4res_u.denied.owner;
147 
148 	if (res->status == NFS4ERR_DENIED)
149 		kmem_free(lo->owner_val, lo->owner_len);
150 }
151 
152 static void
153 deep_open_copy(OPEN4res *dres, OPEN4res *sres)
154 {
155 	nfsace4 *sacep, *dacep;
156 
157 	if (sres->status != NFS4_OK) {
158 		return;
159 	}
160 
161 	dres->attrset = sres->attrset;
162 
163 	switch (sres->delegation.delegation_type) {
164 	case OPEN_DELEGATE_NONE:
165 		return;
166 	case OPEN_DELEGATE_READ:
167 		sacep = &sres->delegation.open_delegation4_u.read.permissions;
168 		dacep = &dres->delegation.open_delegation4_u.read.permissions;
169 		break;
170 	case OPEN_DELEGATE_WRITE:
171 		sacep = &sres->delegation.open_delegation4_u.write.permissions;
172 		dacep = &dres->delegation.open_delegation4_u.write.permissions;
173 		break;
174 	}
175 	dacep->who.utf8string_val =
176 	    kmem_alloc(sacep->who.utf8string_len, KM_SLEEP);
177 	bcopy(sacep->who.utf8string_val, dacep->who.utf8string_val,
178 	    sacep->who.utf8string_len);
179 }
180 
181 static void
182 deep_open_free(OPEN4res *res)
183 {
184 	nfsace4 *acep;
185 	if (res->status != NFS4_OK)
186 		return;
187 
188 	switch (res->delegation.delegation_type) {
189 	case OPEN_DELEGATE_NONE:
190 		return;
191 	case OPEN_DELEGATE_READ:
192 		acep = &res->delegation.open_delegation4_u.read.permissions;
193 		break;
194 	case OPEN_DELEGATE_WRITE:
195 		acep = &res->delegation.open_delegation4_u.write.permissions;
196 		break;
197 	}
198 
199 	if (acep->who.utf8string_val) {
200 		kmem_free(acep->who.utf8string_val, acep->who.utf8string_len);
201 		acep->who.utf8string_val = NULL;
202 	}
203 }
204 
205 void
206 rfs4_free_reply(nfs_resop4 *rp)
207 {
208 	switch (rp->resop) {
209 	case OP_LOCK:
210 		deep_lock_free(&rp->nfs_resop4_u.oplock);
211 		break;
212 	case OP_OPEN:
213 		deep_open_free(&rp->nfs_resop4_u.opopen);
214 	default:
215 		break;
216 	}
217 }
218 
219 void
220 rfs4_copy_reply(nfs_resop4 *dst, nfs_resop4 *src)
221 {
222 	*dst = *src;
223 
224 	/* Handle responses that need deep copy */
225 	switch (src->resop) {
226 	case OP_LOCK:
227 		deep_lock_copy(&dst->nfs_resop4_u.oplock,
228 		    &src->nfs_resop4_u.oplock);
229 		break;
230 	case OP_OPEN:
231 		deep_open_copy(&dst->nfs_resop4_u.opopen,
232 		    &src->nfs_resop4_u.opopen);
233 		break;
234 	default:
235 		break;
236 	};
237 }
238 
239 /*
240  * This is the implementation of the underlying state engine. The
241  * public interface to this engine is described by
242  * nfs4_state.h. Callers to the engine should hold no state engine
243  * locks when they call in to it. If the protocol needs to lock data
244  * structures it should do so after acquiring all references to them
245  * first and then follow the following lock order:
246  *
247  *	client > openowner > state > lo_state > lockowner > file.
248  *
249  * Internally we only allow a thread to hold one hash bucket lock at a
250  * time and the lock is higher in the lock order (must be acquired
251  * first) than the data structure that is on that hash list.
252  *
253  * If a new reference was acquired by the caller, that reference needs
254  * to be released after releasing all acquired locks with the
255  * corresponding rfs4_*_rele routine.
256  */
257 
258 /*
259  * This code is some what prototypical for now. Its purpose currently is to
260  * implement the interfaces sufficiently to finish the higher protocol
261  * elements. This will be replaced by a dynamically resizeable tables
262  * backed by kmem_cache allocator. However synchronization is handled
263  * correctly (I hope) and will not change by much.  The mutexes for
264  * the hash buckets that can be used to create new instances of data
265  * structures  might be good candidates to evolve into reader writer
266  * locks. If it has to do a creation, it would be holding the
267  * mutex across a kmem_alloc with KM_SLEEP specified.
268  */
269 
270 #ifdef DEBUG
271 #define	TABSIZE 17
272 #else
273 #define	TABSIZE 2047
274 #endif
275 
276 #define	ADDRHASH(key) ((unsigned long)(key) >> 3)
277 
278 /* Used to serialize create/destroy of rfs4_server_state database */
279 kmutex_t	rfs4_state_lock;
280 static rfs4_database_t *rfs4_server_state = NULL;
281 
282 /* Used to serialize lookups of clientids */
283 static	krwlock_t	rfs4_findclient_lock;
284 
285 /*
286  * For now this "table" is exposed so that the CPR callback
287  * function can tromp through it..
288  */
289 rfs4_table_t *rfs4_client_tab;
290 
291 static rfs4_index_t *rfs4_clientid_idx;
292 static rfs4_index_t *rfs4_nfsclnt_idx;
293 static rfs4_table_t *rfs4_openowner_tab;
294 static rfs4_index_t *rfs4_openowner_idx;
295 static rfs4_table_t *rfs4_state_tab;
296 static rfs4_index_t *rfs4_state_idx;
297 static rfs4_index_t *rfs4_state_owner_file_idx;
298 static rfs4_index_t *rfs4_state_file_idx;
299 static rfs4_table_t *rfs4_lo_state_tab;
300 static rfs4_index_t *rfs4_lo_state_idx;
301 static rfs4_index_t *rfs4_lo_state_owner_idx;
302 static rfs4_table_t *rfs4_lockowner_tab;
303 static rfs4_index_t *rfs4_lockowner_idx;
304 static rfs4_index_t *rfs4_lockowner_pid_idx;
305 static rfs4_table_t *rfs4_file_tab;
306 static rfs4_index_t *rfs4_file_idx;
307 static rfs4_table_t *rfs4_deleg_state_tab;
308 static rfs4_index_t *rfs4_deleg_idx;
309 static rfs4_index_t *rfs4_deleg_state_idx;
310 
311 #define	MAXTABSZ 1024*1024
312 
313 /* The values below are rfs4_lease_time units */
314 
315 #ifdef DEBUG
316 #define	CLIENT_CACHE_TIME 1
317 #define	OPENOWNER_CACHE_TIME 1
318 #define	STATE_CACHE_TIME 1
319 #define	LO_STATE_CACHE_TIME 1
320 #define	LOCKOWNER_CACHE_TIME 1
321 #define	FILE_CACHE_TIME 3
322 #define	DELEG_STATE_CACHE_TIME 1
323 #else
324 #define	CLIENT_CACHE_TIME 10
325 #define	OPENOWNER_CACHE_TIME 5
326 #define	STATE_CACHE_TIME 1
327 #define	LO_STATE_CACHE_TIME 1
328 #define	LOCKOWNER_CACHE_TIME 3
329 #define	FILE_CACHE_TIME 40
330 #define	DELEG_STATE_CACHE_TIME 1
331 #endif
332 
333 
334 static time_t rfs4_client_cache_time = 0;
335 static time_t rfs4_openowner_cache_time = 0;
336 static time_t rfs4_state_cache_time = 0;
337 static time_t rfs4_lo_state_cache_time = 0;
338 static time_t rfs4_lockowner_cache_time = 0;
339 static time_t rfs4_file_cache_time = 0;
340 static time_t rfs4_deleg_state_cache_time = 0;
341 
342 static bool_t rfs4_client_create(rfs4_entry_t, void *);
343 static void rfs4_dss_remove_cpleaf(rfs4_client_t *);
344 static void rfs4_dss_remove_leaf(rfs4_servinst_t *, char *, char *);
345 static void rfs4_client_destroy(rfs4_entry_t);
346 static bool_t rfs4_client_expiry(rfs4_entry_t);
347 static uint32_t clientid_hash(void *);
348 static bool_t clientid_compare(rfs4_entry_t, void *);
349 static void *clientid_mkkey(rfs4_entry_t);
350 static uint32_t nfsclnt_hash(void *);
351 static bool_t nfsclnt_compare(rfs4_entry_t, void *);
352 static void *nfsclnt_mkkey(rfs4_entry_t);
353 static bool_t rfs4_openowner_create(rfs4_entry_t, void *);
354 static void rfs4_openowner_destroy(rfs4_entry_t);
355 static bool_t rfs4_openowner_expiry(rfs4_entry_t);
356 static uint32_t openowner_hash(void *);
357 static bool_t openowner_compare(rfs4_entry_t, void *);
358 static void *openowner_mkkey(rfs4_entry_t);
359 static bool_t rfs4_state_create(rfs4_entry_t, void *);
360 static void rfs4_state_destroy(rfs4_entry_t);
361 static bool_t rfs4_state_expiry(rfs4_entry_t);
362 static uint32_t state_hash(void *);
363 static bool_t state_compare(rfs4_entry_t, void *);
364 static void *state_mkkey(rfs4_entry_t);
365 static uint32_t state_owner_file_hash(void *);
366 static bool_t state_owner_file_compare(rfs4_entry_t, void *);
367 static void *state_owner_file_mkkey(rfs4_entry_t);
368 static uint32_t state_file_hash(void *);
369 static bool_t state_file_compare(rfs4_entry_t, void *);
370 static void *state_file_mkkey(rfs4_entry_t);
371 static bool_t rfs4_lo_state_create(rfs4_entry_t, void *);
372 static void rfs4_lo_state_destroy(rfs4_entry_t);
373 static bool_t rfs4_lo_state_expiry(rfs4_entry_t);
374 static uint32_t lo_state_hash(void *);
375 static bool_t lo_state_compare(rfs4_entry_t, void *);
376 static void *lo_state_mkkey(rfs4_entry_t);
377 static uint32_t lo_state_lo_hash(void *);
378 static bool_t lo_state_lo_compare(rfs4_entry_t, void *);
379 static void *lo_state_lo_mkkey(rfs4_entry_t);
380 static bool_t rfs4_lockowner_create(rfs4_entry_t, void *);
381 static void rfs4_lockowner_destroy(rfs4_entry_t);
382 static bool_t rfs4_lockowner_expiry(rfs4_entry_t);
383 static uint32_t lockowner_hash(void *);
384 static bool_t lockowner_compare(rfs4_entry_t, void *);
385 static void *lockowner_mkkey(rfs4_entry_t);
386 static uint32_t pid_hash(void *);
387 static bool_t pid_compare(rfs4_entry_t, void *);
388 static void *pid_mkkey(rfs4_entry_t);
389 static bool_t rfs4_file_create(rfs4_entry_t, void *);
390 static void rfs4_file_destroy(rfs4_entry_t);
391 static uint32_t file_hash(void *);
392 static bool_t file_compare(rfs4_entry_t, void *);
393 static void *file_mkkey(rfs4_entry_t);
394 static bool_t rfs4_deleg_state_create(rfs4_entry_t, void *);
395 static void rfs4_deleg_state_destroy(rfs4_entry_t);
396 static bool_t rfs4_deleg_state_expiry(rfs4_entry_t);
397 static uint32_t deleg_hash(void *);
398 static bool_t deleg_compare(rfs4_entry_t, void *);
399 static void *deleg_mkkey(rfs4_entry_t);
400 static uint32_t deleg_state_hash(void *);
401 static bool_t deleg_state_compare(rfs4_entry_t, void *);
402 static void *deleg_state_mkkey(rfs4_entry_t);
403 
404 static void rfs4_state_rele_nounlock(rfs4_state_t *);
405 
406 static int rfs4_ss_enabled = 0;
407 
408 extern void (*rfs4_client_clrst)(struct nfs4clrst_args *);
409 
410 void
411 rfs4_ss_pnfree(rfs4_ss_pn_t *ss_pn)
412 {
413 	kmem_free(ss_pn, sizeof (rfs4_ss_pn_t));
414 }
415 
416 static rfs4_ss_pn_t *
417 rfs4_ss_pnalloc(char *dir, char *leaf)
418 {
419 	rfs4_ss_pn_t *ss_pn;
420 	int 	dir_len, leaf_len;
421 
422 	/*
423 	 * validate we have a resonable path
424 	 * (account for the '/' and trailing null)
425 	 */
426 	if ((dir_len = strlen(dir)) > MAXPATHLEN ||
427 	    (leaf_len = strlen(leaf)) > MAXNAMELEN ||
428 	    (dir_len + leaf_len + 2) > MAXPATHLEN) {
429 		return (NULL);
430 	}
431 
432 	ss_pn = kmem_alloc(sizeof (rfs4_ss_pn_t), KM_SLEEP);
433 
434 	(void) snprintf(ss_pn->pn, MAXPATHLEN, "%s/%s", dir, leaf);
435 	/* Handy pointer to just the leaf name */
436 	ss_pn->leaf = ss_pn->pn + dir_len + 1;
437 	return (ss_pn);
438 }
439 
440 
441 /*
442  * Move the "leaf" filename from "sdir" directory
443  * to the "ddir" directory. Return the pathname of
444  * the destination unless the rename fails in which
445  * case we need to return the source pathname.
446  */
447 static rfs4_ss_pn_t *
448 rfs4_ss_movestate(char *sdir, char *ddir, char *leaf)
449 {
450 	rfs4_ss_pn_t *src, *dst;
451 
452 	if ((src = rfs4_ss_pnalloc(sdir, leaf)) == NULL)
453 		return (NULL);
454 
455 	if ((dst = rfs4_ss_pnalloc(ddir, leaf)) == NULL) {
456 		rfs4_ss_pnfree(src);
457 		return (NULL);
458 	}
459 
460 	/*
461 	 * If the rename fails we shall return the src
462 	 * pathname and free the dst. Otherwise we need
463 	 * to free the src and return the dst pathanme.
464 	 */
465 	if (vn_rename(src->pn, dst->pn, UIO_SYSSPACE)) {
466 		rfs4_ss_pnfree(dst);
467 		return (src);
468 	}
469 	rfs4_ss_pnfree(src);
470 	return (dst);
471 }
472 
473 
474 static rfs4_oldstate_t *
475 rfs4_ss_getstate(vnode_t *dvp, rfs4_ss_pn_t *ss_pn)
476 {
477 	struct uio uio;
478 	struct iovec iov[3];
479 
480 	rfs4_oldstate_t *cl_ss = NULL;
481 	vnode_t *vp;
482 	vattr_t va;
483 	uint_t id_len;
484 	int err, kill_file, file_vers;
485 
486 	if (ss_pn == NULL)
487 		return (NULL);
488 
489 	/*
490 	 * open the state file.
491 	 */
492 	if (vn_open(ss_pn->pn, UIO_SYSSPACE, FREAD, 0, &vp, 0, 0) != 0) {
493 		return (NULL);
494 	}
495 
496 	if (vp->v_type != VREG) {
497 		(void) VOP_CLOSE(vp, FREAD, 1, (offset_t)0, CRED(), NULL);
498 		VN_RELE(vp);
499 		return (NULL);
500 	}
501 
502 	err = VOP_ACCESS(vp, VREAD, 0, CRED(), NULL);
503 	if (err) {
504 		/*
505 		 * We don't have read access? better get the heck out.
506 		 */
507 		(void) VOP_CLOSE(vp, FREAD, 1, (offset_t)0, CRED(), NULL);
508 		VN_RELE(vp);
509 		return (NULL);
510 	}
511 
512 	(void) VOP_RWLOCK(vp, V_WRITELOCK_FALSE, NULL);
513 	/*
514 	 * get the file size to do some basic validation
515 	 */
516 	va.va_mask = AT_SIZE;
517 	err = VOP_GETATTR(vp, &va, 0, CRED(), NULL);
518 
519 	kill_file = (va.va_size == 0 || va.va_size <
520 	    (NFS4_VERIFIER_SIZE + sizeof (uint_t)+1));
521 
522 	if (err || kill_file) {
523 		VOP_RWUNLOCK(vp, V_WRITELOCK_FALSE, NULL);
524 		(void) VOP_CLOSE(vp, FREAD, 1, (offset_t)0, CRED(), NULL);
525 		VN_RELE(vp);
526 		if (kill_file) {
527 			(void) VOP_REMOVE(dvp, ss_pn->leaf, CRED(), NULL, 0);
528 		}
529 		return (NULL);
530 	}
531 
532 	cl_ss = kmem_alloc(sizeof (rfs4_oldstate_t), KM_SLEEP);
533 
534 	/*
535 	 * build iovecs to read in the file_version, verifier and id_len
536 	 */
537 	iov[0].iov_base = (caddr_t)&file_vers;
538 	iov[0].iov_len = sizeof (int);
539 	iov[1].iov_base = (caddr_t)&cl_ss->cl_id4.verifier;
540 	iov[1].iov_len = NFS4_VERIFIER_SIZE;
541 	iov[2].iov_base = (caddr_t)&id_len;
542 	iov[2].iov_len = sizeof (uint_t);
543 
544 	uio.uio_iov = iov;
545 	uio.uio_iovcnt = 3;
546 	uio.uio_segflg = UIO_SYSSPACE;
547 	uio.uio_loffset = 0;
548 	uio.uio_resid = sizeof (int) + NFS4_VERIFIER_SIZE + sizeof (uint_t);
549 
550 	if (err = VOP_READ(vp, &uio, FREAD, CRED(), NULL)) {
551 		VOP_RWUNLOCK(vp, V_WRITELOCK_FALSE, NULL);
552 		(void) VOP_CLOSE(vp, FREAD, 1, (offset_t)0, CRED(), NULL);
553 		VN_RELE(vp);
554 		kmem_free(cl_ss, sizeof (rfs4_oldstate_t));
555 		return (NULL);
556 	}
557 
558 	/*
559 	 * if the file_version doesn't match or if the
560 	 * id_len is zero or the combination of the verifier,
561 	 * id_len and id_val is bigger than the file we have
562 	 * a problem. If so ditch the file.
563 	 */
564 	kill_file = (file_vers != NFS4_SS_VERSION || id_len == 0 ||
565 	    (id_len + NFS4_VERIFIER_SIZE + sizeof (uint_t)) > va.va_size);
566 
567 	if (err || kill_file) {
568 		VOP_RWUNLOCK(vp, V_WRITELOCK_FALSE, NULL);
569 		(void) VOP_CLOSE(vp, FREAD, 1, (offset_t)0, CRED(), NULL);
570 		VN_RELE(vp);
571 		kmem_free(cl_ss, sizeof (rfs4_oldstate_t));
572 		if (kill_file) {
573 			(void) VOP_REMOVE(dvp, ss_pn->leaf, CRED(), NULL, 0);
574 		}
575 		return (NULL);
576 	}
577 
578 	/*
579 	 * now get the client id value
580 	 */
581 	cl_ss->cl_id4.id_val = kmem_alloc(id_len, KM_SLEEP);
582 	iov[0].iov_base = cl_ss->cl_id4.id_val;
583 	iov[0].iov_len = id_len;
584 
585 	uio.uio_iov = iov;
586 	uio.uio_iovcnt = 1;
587 	uio.uio_segflg = UIO_SYSSPACE;
588 	uio.uio_resid = cl_ss->cl_id4.id_len = id_len;
589 
590 	if (err = VOP_READ(vp, &uio, FREAD, CRED(), NULL)) {
591 		VOP_RWUNLOCK(vp, V_WRITELOCK_FALSE, NULL);
592 		(void) VOP_CLOSE(vp, FREAD, 1, (offset_t)0, CRED(), NULL);
593 		VN_RELE(vp);
594 		kmem_free(cl_ss->cl_id4.id_val, id_len);
595 		kmem_free(cl_ss, sizeof (rfs4_oldstate_t));
596 		return (NULL);
597 	}
598 
599 	VOP_RWUNLOCK(vp, V_WRITELOCK_FALSE, NULL);
600 	(void) VOP_CLOSE(vp, FREAD, 1, (offset_t)0, CRED(), NULL);
601 	VN_RELE(vp);
602 	return (cl_ss);
603 }
604 
605 #ifdef	nextdp
606 #undef nextdp
607 #endif
608 #define	nextdp(dp)	((struct dirent64 *)((char *)(dp) + (dp)->d_reclen))
609 
610 /*
611  * Add entries from statedir to supplied oldstate list.
612  * Optionally, move all entries from statedir -> destdir.
613  */
614 void
615 rfs4_ss_oldstate(rfs4_oldstate_t *oldstate, char *statedir, char *destdir)
616 {
617 	rfs4_ss_pn_t *ss_pn;
618 	rfs4_oldstate_t *cl_ss = NULL;
619 	char	*dirt = NULL;
620 	int	err, dir_eof = 0, size = 0;
621 	vnode_t *dvp;
622 	struct iovec iov;
623 	struct uio uio;
624 	struct dirent64 *dep;
625 	offset_t dirchunk_offset = 0;
626 
627 	/*
628 	 * open the state directory
629 	 */
630 	if (vn_open(statedir, UIO_SYSSPACE, FREAD, 0, &dvp, 0, 0))
631 		return;
632 
633 	if (dvp->v_type != VDIR || VOP_ACCESS(dvp, VREAD, 0, CRED(), NULL))
634 		goto out;
635 
636 	dirt = kmem_alloc(RFS4_SS_DIRSIZE, KM_SLEEP);
637 
638 	/*
639 	 * Get and process the directory entries
640 	 */
641 	while (!dir_eof) {
642 		(void) VOP_RWLOCK(dvp, V_WRITELOCK_FALSE, NULL);
643 		iov.iov_base = dirt;
644 		iov.iov_len = RFS4_SS_DIRSIZE;
645 		uio.uio_iov = &iov;
646 		uio.uio_iovcnt = 1;
647 		uio.uio_segflg = UIO_SYSSPACE;
648 		uio.uio_loffset = dirchunk_offset;
649 		uio.uio_resid = RFS4_SS_DIRSIZE;
650 
651 		err = VOP_READDIR(dvp, &uio, CRED(), &dir_eof, NULL, 0);
652 		VOP_RWUNLOCK(dvp, V_WRITELOCK_FALSE, NULL);
653 		if (err)
654 			goto out;
655 
656 		size = RFS4_SS_DIRSIZE - uio.uio_resid;
657 
658 		/*
659 		 * Process all the directory entries in this
660 		 * readdir chunk
661 		 */
662 		for (dep = (struct dirent64 *)dirt; size > 0;
663 		    dep = nextdp(dep)) {
664 
665 			size -= dep->d_reclen;
666 			dirchunk_offset = dep->d_off;
667 
668 			/*
669 			 * Skip '.' and '..'
670 			 */
671 			if (NFS_IS_DOTNAME(dep->d_name))
672 				continue;
673 
674 			ss_pn = rfs4_ss_pnalloc(statedir, dep->d_name);
675 			if (ss_pn == NULL)
676 				continue;
677 
678 			if (cl_ss = rfs4_ss_getstate(dvp, ss_pn)) {
679 				if (destdir != NULL) {
680 					rfs4_ss_pnfree(ss_pn);
681 					cl_ss->ss_pn = rfs4_ss_movestate(
682 					    statedir, destdir, dep->d_name);
683 				} else {
684 					cl_ss->ss_pn = ss_pn;
685 				}
686 				insque(cl_ss, oldstate);
687 			} else {
688 				rfs4_ss_pnfree(ss_pn);
689 			}
690 		}
691 	}
692 
693 out:
694 	(void) VOP_CLOSE(dvp, FREAD, 1, (offset_t)0, CRED(), NULL);
695 	VN_RELE(dvp);
696 	if (dirt)
697 		kmem_free((caddr_t)dirt, RFS4_SS_DIRSIZE);
698 }
699 
700 static void
701 rfs4_ss_init(void)
702 {
703 	int npaths = 1;
704 	char *default_dss_path = NFS4_DSS_VAR_DIR;
705 
706 	/* read the default stable storage state */
707 	rfs4_dss_readstate(npaths, &default_dss_path);
708 
709 	rfs4_ss_enabled = 1;
710 }
711 
712 static void
713 rfs4_ss_fini(void)
714 {
715 	rfs4_servinst_t *sip;
716 
717 	mutex_enter(&rfs4_servinst_lock);
718 	sip = rfs4_cur_servinst;
719 	while (sip != NULL) {
720 		rfs4_dss_clear_oldstate(sip);
721 		sip = sip->next;
722 	}
723 	mutex_exit(&rfs4_servinst_lock);
724 }
725 
726 /*
727  * Remove all oldstate files referenced by this servinst.
728  */
729 static void
730 rfs4_dss_clear_oldstate(rfs4_servinst_t *sip)
731 {
732 	rfs4_oldstate_t *os_head, *osp;
733 
734 	rw_enter(&sip->oldstate_lock, RW_WRITER);
735 	os_head = sip->oldstate;
736 
737 	if (os_head == NULL)
738 		return;
739 
740 	/* skip dummy entry */
741 	osp = os_head->next;
742 	while (osp != os_head) {
743 		char *leaf = osp->ss_pn->leaf;
744 		rfs4_oldstate_t *os_next;
745 
746 		rfs4_dss_remove_leaf(sip, NFS4_DSS_OLDSTATE_LEAF, leaf);
747 
748 		if (osp->cl_id4.id_val)
749 			kmem_free(osp->cl_id4.id_val, osp->cl_id4.id_len);
750 		if (osp->ss_pn)
751 			kmem_free(osp->ss_pn, sizeof (rfs4_ss_pn_t));
752 
753 		os_next = osp->next;
754 		remque(osp);
755 		kmem_free(osp, sizeof (rfs4_oldstate_t));
756 		osp = os_next;
757 	}
758 
759 	/* free dummy entry */
760 	kmem_free(osp, sizeof (rfs4_oldstate_t));
761 
762 	sip->oldstate = NULL;
763 
764 	rw_exit(&sip->oldstate_lock);
765 }
766 
767 /*
768  * Form the state and oldstate paths, and read in the stable storage files.
769  */
770 void
771 rfs4_dss_readstate(int npaths, char **paths)
772 {
773 	int i;
774 	char *state, *oldstate;
775 
776 	state = kmem_alloc(MAXPATHLEN, KM_SLEEP);
777 	oldstate = kmem_alloc(MAXPATHLEN, KM_SLEEP);
778 
779 	for (i = 0; i < npaths; i++) {
780 		char *path = paths[i];
781 
782 		(void) sprintf(state, "%s/%s", path, NFS4_DSS_STATE_LEAF);
783 		(void) sprintf(oldstate, "%s/%s", path, NFS4_DSS_OLDSTATE_LEAF);
784 
785 		/*
786 		 * Populate the current server instance's oldstate list.
787 		 *
788 		 * 1. Read stable storage data from old state directory,
789 		 *    leaving its contents alone.
790 		 *
791 		 * 2. Read stable storage data from state directory,
792 		 *    and move the latter's contents to old state
793 		 *    directory.
794 		 */
795 		rfs4_ss_oldstate(rfs4_cur_servinst->oldstate, oldstate, NULL);
796 		rfs4_ss_oldstate(rfs4_cur_servinst->oldstate, state, oldstate);
797 	}
798 
799 	kmem_free(state, MAXPATHLEN);
800 	kmem_free(oldstate, MAXPATHLEN);
801 }
802 
803 
804 /*
805  * Check if we are still in grace and if the client can be
806  * granted permission to perform reclaims.
807  */
808 void
809 rfs4_ss_chkclid(rfs4_client_t *cp)
810 {
811 	rfs4_servinst_t *sip;
812 
813 	/*
814 	 * It should be sufficient to check the oldstate data for just
815 	 * this client's instance. However, since our per-instance
816 	 * client grouping is solely temporal, HA-NFSv4 RG failover
817 	 * might result in clients of the same RG being partitioned into
818 	 * separate instances.
819 	 *
820 	 * Until the client grouping is improved, we must check the
821 	 * oldstate data for all instances with an active grace period.
822 	 *
823 	 * This also serves as the mechanism to remove stale oldstate data.
824 	 * The first time we check an instance after its grace period has
825 	 * expired, the oldstate data should be cleared.
826 	 *
827 	 * Start at the current instance, and walk the list backwards
828 	 * to the first.
829 	 */
830 	mutex_enter(&rfs4_servinst_lock);
831 	for (sip = rfs4_cur_servinst; sip != NULL; sip = sip->prev) {
832 		rfs4_ss_chkclid_sip(cp, sip);
833 
834 		/* if the above check found this client, we're done */
835 		if (cp->can_reclaim)
836 			break;
837 	}
838 	mutex_exit(&rfs4_servinst_lock);
839 }
840 
841 static void
842 rfs4_ss_chkclid_sip(rfs4_client_t *cp, rfs4_servinst_t *sip)
843 {
844 	rfs4_oldstate_t *osp, *os_head;
845 
846 	/* short circuit everything if this server instance has no oldstate */
847 	rw_enter(&sip->oldstate_lock, RW_READER);
848 	os_head = sip->oldstate;
849 	rw_exit(&sip->oldstate_lock);
850 	if (os_head == NULL)
851 		return;
852 
853 	/*
854 	 * If this server instance is no longer in a grace period then
855 	 * the client won't be able to reclaim. No further need for this
856 	 * instance's oldstate data, so it can be cleared.
857 	 */
858 	if (!rfs4_servinst_in_grace(sip))
859 		return;
860 
861 	/* this instance is still in grace; search for the clientid */
862 
863 	rw_enter(&sip->oldstate_lock, RW_READER);
864 
865 	os_head = sip->oldstate;
866 	/* skip dummy entry */
867 	osp = os_head->next;
868 	while (osp != os_head) {
869 		if (osp->cl_id4.id_len == cp->nfs_client.id_len) {
870 			if (bcmp(osp->cl_id4.id_val, cp->nfs_client.id_val,
871 			    osp->cl_id4.id_len) == 0) {
872 				cp->can_reclaim = 1;
873 				break;
874 			}
875 		}
876 		osp = osp->next;
877 	}
878 
879 	rw_exit(&sip->oldstate_lock);
880 }
881 
882 /*
883  * Place client information into stable storage: 1/3.
884  * First, generate the leaf filename, from the client's IP address and
885  * the server-generated short-hand clientid.
886  */
887 void
888 rfs4_ss_clid(rfs4_client_t *cp, struct svc_req *req)
889 {
890 	const char *kinet_ntop6(uchar_t *, char *, size_t);
891 	char leaf[MAXNAMELEN], buf[INET6_ADDRSTRLEN];
892 	struct sockaddr *ca;
893 	uchar_t *b;
894 
895 	if (rfs4_ss_enabled == 0) {
896 		return;
897 	}
898 
899 	buf[0] = 0;
900 
901 
902 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
903 	if (ca == NULL) {
904 		return;
905 	}
906 
907 	/*
908 	 * Convert the caller's IP address to a dotted string
909 	 */
910 	if (ca->sa_family == AF_INET) {
911 
912 		bcopy(svc_getrpccaller(req->rq_xprt)->buf, &cp->cl_addr,
913 		    sizeof (struct sockaddr_in));
914 		b = (uchar_t *)&((struct sockaddr_in *)ca)->sin_addr;
915 		(void) sprintf(buf, "%03d.%03d.%03d.%03d", b[0] & 0xFF,
916 		    b[1] & 0xFF, b[2] & 0xFF, b[3] & 0xFF);
917 	} else if (ca->sa_family == AF_INET6) {
918 		struct sockaddr_in6 *sin6;
919 
920 		sin6 = (struct sockaddr_in6 *)ca;
921 		bcopy(svc_getrpccaller(req->rq_xprt)->buf, &cp->cl_addr,
922 		    sizeof (struct sockaddr_in6));
923 		(void) kinet_ntop6((uchar_t *)&sin6->sin6_addr,
924 		    buf, INET6_ADDRSTRLEN);
925 	}
926 
927 	(void) snprintf(leaf, MAXNAMELEN, "%s-%llx", buf,
928 	    (longlong_t)cp->clientid);
929 	rfs4_ss_clid_write(cp, leaf);
930 }
931 
932 /*
933  * Place client information into stable storage: 2/3.
934  * DSS: distributed stable storage: the file may need to be written to
935  * multiple directories.
936  */
937 static void
938 rfs4_ss_clid_write(rfs4_client_t *cp, char *leaf)
939 {
940 	rfs4_servinst_t *sip;
941 
942 	/*
943 	 * It should be sufficient to write the leaf file to (all) DSS paths
944 	 * associated with just this client's instance. However, since our
945 	 * per-instance client grouping is solely temporal, HA-NFSv4 RG
946 	 * failover might result in us losing DSS data.
947 	 *
948 	 * Until the client grouping is improved, we must write the DSS data
949 	 * to all instances' paths. Start at the current instance, and
950 	 * walk the list backwards to the first.
951 	 */
952 	mutex_enter(&rfs4_servinst_lock);
953 	for (sip = rfs4_cur_servinst; sip != NULL; sip = sip->prev) {
954 		int i, npaths = sip->dss_npaths;
955 
956 		/* write the leaf file to all DSS paths */
957 		for (i = 0; i < npaths; i++) {
958 			rfs4_dss_path_t *dss_path = sip->dss_paths[i];
959 
960 			/* HA-NFSv4 path might have been failed-away from us */
961 			if (dss_path == NULL)
962 				continue;
963 
964 			rfs4_ss_clid_write_one(cp, dss_path->path, leaf);
965 		}
966 	}
967 	mutex_exit(&rfs4_servinst_lock);
968 }
969 
970 /*
971  * Place client information into stable storage: 3/3.
972  * Write the stable storage data to the requested file.
973  */
974 static void
975 rfs4_ss_clid_write_one(rfs4_client_t *cp, char *dss_path, char *leaf)
976 {
977 	int ioflag;
978 	int file_vers = NFS4_SS_VERSION;
979 	size_t dirlen;
980 	struct uio uio;
981 	struct iovec iov[4];
982 	char *dir;
983 	rfs4_ss_pn_t *ss_pn;
984 	vnode_t *vp;
985 	nfs_client_id4 *cl_id4 = &(cp->nfs_client);
986 
987 	/* allow 2 extra bytes for '/' & NUL */
988 	dirlen = strlen(dss_path) + strlen(NFS4_DSS_STATE_LEAF) + 2;
989 	dir = kmem_alloc(dirlen, KM_SLEEP);
990 	(void) sprintf(dir, "%s/%s", dss_path, NFS4_DSS_STATE_LEAF);
991 
992 	ss_pn = rfs4_ss_pnalloc(dir, leaf);
993 	/* rfs4_ss_pnalloc takes its own copy */
994 	kmem_free(dir, dirlen);
995 	if (ss_pn == NULL)
996 		return;
997 
998 	if (vn_open(ss_pn->pn, UIO_SYSSPACE, FCREAT|FWRITE, 0600, &vp,
999 	    CRCREAT, 0)) {
1000 		rfs4_ss_pnfree(ss_pn);
1001 		return;
1002 	}
1003 
1004 	/*
1005 	 * We need to record leaf - i.e. the filename - so that we know
1006 	 * what to remove, in the future. However, the dir part of cp->ss_pn
1007 	 * should never be referenced directly, since it's potentially only
1008 	 * one of several paths with this leaf in it.
1009 	 */
1010 	if (cp->ss_pn != NULL) {
1011 		if (strcmp(cp->ss_pn->leaf, leaf) == 0) {
1012 			/* we've already recorded *this* leaf */
1013 			rfs4_ss_pnfree(ss_pn);
1014 		} else {
1015 			/* replace with this leaf */
1016 			rfs4_ss_pnfree(cp->ss_pn);
1017 			cp->ss_pn = ss_pn;
1018 		}
1019 	} else {
1020 		cp->ss_pn = ss_pn;
1021 	}
1022 
1023 	/*
1024 	 * Build a scatter list that points to the nfs_client_id4
1025 	 */
1026 	iov[0].iov_base = (caddr_t)&file_vers;
1027 	iov[0].iov_len = sizeof (int);
1028 	iov[1].iov_base = (caddr_t)&(cl_id4->verifier);
1029 	iov[1].iov_len = NFS4_VERIFIER_SIZE;
1030 	iov[2].iov_base = (caddr_t)&(cl_id4->id_len);
1031 	iov[2].iov_len = sizeof (uint_t);
1032 	iov[3].iov_base = (caddr_t)cl_id4->id_val;
1033 	iov[3].iov_len = cl_id4->id_len;
1034 
1035 	uio.uio_iov = iov;
1036 	uio.uio_iovcnt = 4;
1037 	uio.uio_loffset = 0;
1038 	uio.uio_segflg = UIO_SYSSPACE;
1039 	uio.uio_llimit = (rlim64_t)MAXOFFSET_T;
1040 	uio.uio_resid = cl_id4->id_len + sizeof (int) +
1041 	    NFS4_VERIFIER_SIZE + sizeof (uint_t);
1042 
1043 	ioflag = uio.uio_fmode = (FWRITE|FSYNC);
1044 	uio.uio_extflg = UIO_COPY_DEFAULT;
1045 
1046 	(void) VOP_RWLOCK(vp, V_WRITELOCK_TRUE, NULL);
1047 	/* write the full client id to the file. */
1048 	(void) VOP_WRITE(vp, &uio, ioflag, CRED(), NULL);
1049 	VOP_RWUNLOCK(vp, V_WRITELOCK_TRUE, NULL);
1050 
1051 	(void) VOP_CLOSE(vp, FWRITE, 1, (offset_t)0, CRED(), NULL);
1052 	VN_RELE(vp);
1053 }
1054 
1055 /*
1056  * DSS: distributed stable storage.
1057  * Unpack the list of paths passed by nfsd.
1058  * Use nvlist_alloc(9F) to manage the data.
1059  * The caller is responsible for allocating and freeing the buffer.
1060  */
1061 int
1062 rfs4_dss_setpaths(char *buf, size_t buflen)
1063 {
1064 	int error;
1065 
1066 	/*
1067 	 * If this is a "warm start", i.e. we previously had DSS paths,
1068 	 * preserve the old paths.
1069 	 */
1070 	if (rfs4_dss_paths != NULL) {
1071 		/*
1072 		 * Before we lose the ptr, destroy the nvlist and pathnames
1073 		 * array from the warm start before this one.
1074 		 */
1075 		if (rfs4_dss_oldpaths)
1076 			nvlist_free(rfs4_dss_oldpaths);
1077 		rfs4_dss_oldpaths = rfs4_dss_paths;
1078 	}
1079 
1080 	/* unpack the buffer into a searchable nvlist */
1081 	error = nvlist_unpack(buf, buflen, &rfs4_dss_paths, KM_SLEEP);
1082 	if (error)
1083 		return (error);
1084 
1085 	/*
1086 	 * Search the nvlist for the pathnames nvpair (which is the only nvpair
1087 	 * in the list, and record its location.
1088 	 */
1089 	error = nvlist_lookup_string_array(rfs4_dss_paths, NFS4_DSS_NVPAIR_NAME,
1090 	    &rfs4_dss_newpaths, &rfs4_dss_numnewpaths);
1091 	return (error);
1092 }
1093 
1094 /*
1095  * Ultimately the nfssys() call NFS4_CLR_STATE endsup here
1096  * to find and mark the client for forced expire.
1097  */
1098 static void
1099 rfs4_client_scrub(rfs4_entry_t ent, void *arg)
1100 {
1101 	rfs4_client_t *cp = (rfs4_client_t *)ent;
1102 	struct nfs4clrst_args *clr = arg;
1103 	struct sockaddr_in6 *ent_sin6;
1104 	struct in6_addr  clr_in6;
1105 	struct sockaddr_in  *ent_sin;
1106 	struct in_addr   clr_in;
1107 
1108 	if (clr->addr_type != cp->cl_addr.ss_family) {
1109 		return;
1110 	}
1111 
1112 	switch (clr->addr_type) {
1113 
1114 	case AF_INET6:
1115 		/* copyin the address from user space */
1116 		if (copyin(clr->ap, &clr_in6, sizeof (clr_in6))) {
1117 			break;
1118 		}
1119 
1120 		ent_sin6 = (struct sockaddr_in6 *)&cp->cl_addr;
1121 
1122 		/*
1123 		 * now compare, and if equivalent mark entry
1124 		 * for forced expiration
1125 		 */
1126 		if (IN6_ARE_ADDR_EQUAL(&ent_sin6->sin6_addr, &clr_in6)) {
1127 			cp->forced_expire = 1;
1128 		}
1129 		break;
1130 
1131 	case AF_INET:
1132 		/* copyin the address from user space */
1133 		if (copyin(clr->ap, &clr_in, sizeof (clr_in))) {
1134 			break;
1135 		}
1136 
1137 		ent_sin = (struct sockaddr_in *)&cp->cl_addr;
1138 
1139 		/*
1140 		 * now compare, and if equivalent mark entry
1141 		 * for forced expiration
1142 		 */
1143 		if (ent_sin->sin_addr.s_addr == clr_in.s_addr) {
1144 			cp->forced_expire = 1;
1145 		}
1146 		break;
1147 
1148 	default:
1149 		/* force this assert to fail */
1150 		ASSERT(clr->addr_type != clr->addr_type);
1151 	}
1152 }
1153 
1154 /*
1155  * This is called from nfssys() in order to clear server state
1156  * for the specified client IP Address.
1157  */
1158 void
1159 rfs4_clear_client_state(struct nfs4clrst_args *clr)
1160 {
1161 	(void) rfs4_dbe_walk(rfs4_client_tab, rfs4_client_scrub, clr);
1162 }
1163 
1164 /*
1165  * Used to initialize the NFSv4 server's state or database.  All of
1166  * the tables are created and timers are set. Only called when NFSv4
1167  * service is provided.
1168  */
1169 void
1170 rfs4_state_init()
1171 {
1172 	int start_grace;
1173 	extern boolean_t rfs4_cpr_callb(void *, int);
1174 	char *dss_path = NFS4_DSS_VAR_DIR;
1175 
1176 	mutex_enter(&rfs4_state_lock);
1177 
1178 	/*
1179 	 * If the server state database has already been initialized,
1180 	 * skip it
1181 	 */
1182 	if (rfs4_server_state != NULL) {
1183 		mutex_exit(&rfs4_state_lock);
1184 		return;
1185 	}
1186 
1187 	rw_init(&rfs4_findclient_lock, NULL, RW_DEFAULT, NULL);
1188 
1189 	/*
1190 	 * Set the boot time.  If the server
1191 	 * has been restarted quickly and has had the opportunity to
1192 	 * service clients, then the start_time needs to be bumped
1193 	 * regardless.  A small window but it exists...
1194 	 */
1195 	if (rfs4_start_time != gethrestime_sec())
1196 		rfs4_start_time = gethrestime_sec();
1197 	else
1198 		rfs4_start_time++;
1199 
1200 	/* DSS: distributed stable storage: initialise served paths list */
1201 	rfs4_dss_pathlist = NULL;
1202 
1203 	/*
1204 	 * Create the first server instance, or a new one if the server has
1205 	 * been restarted; see above comments on rfs4_start_time. Don't
1206 	 * start its grace period; that will be done later, to maximise the
1207 	 * clients' recovery window.
1208 	 */
1209 	start_grace = 0;
1210 	rfs4_servinst_create(start_grace, 1, &dss_path);
1211 
1212 	/* reset the "first NFSv4 request" status */
1213 	rfs4_seen_first_compound = 0;
1214 
1215 	/*
1216 	 * Add a CPR callback so that we can update client
1217 	 * access times to extend the lease after a suspend
1218 	 * and resume (using the same class as rpcmod/connmgr)
1219 	 */
1220 	cpr_id = callb_add(rfs4_cpr_callb, 0, CB_CL_CPR_RPC, "rfs4");
1221 
1222 	/* set the various cache timers for table creation */
1223 	if (rfs4_client_cache_time == 0)
1224 		rfs4_client_cache_time = CLIENT_CACHE_TIME;
1225 	if (rfs4_openowner_cache_time == 0)
1226 		rfs4_openowner_cache_time = OPENOWNER_CACHE_TIME;
1227 	if (rfs4_state_cache_time == 0)
1228 		rfs4_state_cache_time = STATE_CACHE_TIME;
1229 	if (rfs4_lo_state_cache_time == 0)
1230 		rfs4_lo_state_cache_time = LO_STATE_CACHE_TIME;
1231 	if (rfs4_lockowner_cache_time == 0)
1232 		rfs4_lockowner_cache_time = LOCKOWNER_CACHE_TIME;
1233 	if (rfs4_file_cache_time == 0)
1234 		rfs4_file_cache_time = FILE_CACHE_TIME;
1235 	if (rfs4_deleg_state_cache_time == 0)
1236 		rfs4_deleg_state_cache_time = DELEG_STATE_CACHE_TIME;
1237 
1238 	/* Create the overall database to hold all server state */
1239 	rfs4_server_state = rfs4_database_create(rfs4_database_debug);
1240 
1241 	/* Now create the individual tables */
1242 	rfs4_client_cache_time *= rfs4_lease_time;
1243 	rfs4_client_tab = rfs4_table_create(rfs4_server_state,
1244 	    "Client",
1245 	    rfs4_client_cache_time,
1246 	    2,
1247 	    rfs4_client_create,
1248 	    rfs4_client_destroy,
1249 	    rfs4_client_expiry,
1250 	    sizeof (rfs4_client_t),
1251 	    TABSIZE,
1252 	    MAXTABSZ/8, 100);
1253 	rfs4_nfsclnt_idx = rfs4_index_create(rfs4_client_tab,
1254 	    "nfs_client_id4", nfsclnt_hash,
1255 	    nfsclnt_compare, nfsclnt_mkkey,
1256 	    TRUE);
1257 	rfs4_clientid_idx = rfs4_index_create(rfs4_client_tab,
1258 	    "client_id", clientid_hash,
1259 	    clientid_compare, clientid_mkkey,
1260 	    FALSE);
1261 
1262 	rfs4_openowner_cache_time *= rfs4_lease_time;
1263 	rfs4_openowner_tab = rfs4_table_create(rfs4_server_state,
1264 	    "OpenOwner",
1265 	    rfs4_openowner_cache_time,
1266 	    1,
1267 	    rfs4_openowner_create,
1268 	    rfs4_openowner_destroy,
1269 	    rfs4_openowner_expiry,
1270 	    sizeof (rfs4_openowner_t),
1271 	    TABSIZE,
1272 	    MAXTABSZ, 100);
1273 	rfs4_openowner_idx = rfs4_index_create(rfs4_openowner_tab,
1274 	    "open_owner4", openowner_hash,
1275 	    openowner_compare,
1276 	    openowner_mkkey, TRUE);
1277 
1278 	rfs4_state_cache_time *= rfs4_lease_time;
1279 	rfs4_state_tab = rfs4_table_create(rfs4_server_state,
1280 	    "OpenStateID",
1281 	    rfs4_state_cache_time,
1282 	    3,
1283 	    rfs4_state_create,
1284 	    rfs4_state_destroy,
1285 	    rfs4_state_expiry,
1286 	    sizeof (rfs4_state_t),
1287 	    TABSIZE,
1288 	    MAXTABSZ, 100);
1289 
1290 	rfs4_state_owner_file_idx = rfs4_index_create(rfs4_state_tab,
1291 	    "Openowner-File",
1292 	    state_owner_file_hash,
1293 	    state_owner_file_compare,
1294 	    state_owner_file_mkkey, TRUE);
1295 
1296 	rfs4_state_idx = rfs4_index_create(rfs4_state_tab,
1297 	    "State-id", state_hash,
1298 	    state_compare, state_mkkey, FALSE);
1299 
1300 	rfs4_state_file_idx = rfs4_index_create(rfs4_state_tab,
1301 	    "File", state_file_hash,
1302 	    state_file_compare, state_file_mkkey,
1303 	    FALSE);
1304 
1305 	rfs4_lo_state_cache_time *= rfs4_lease_time;
1306 	rfs4_lo_state_tab = rfs4_table_create(rfs4_server_state,
1307 	    "LockStateID",
1308 	    rfs4_lo_state_cache_time,
1309 	    2,
1310 	    rfs4_lo_state_create,
1311 	    rfs4_lo_state_destroy,
1312 	    rfs4_lo_state_expiry,
1313 	    sizeof (rfs4_lo_state_t),
1314 	    TABSIZE,
1315 	    MAXTABSZ, 100);
1316 
1317 	rfs4_lo_state_owner_idx = rfs4_index_create(rfs4_lo_state_tab,
1318 	    "lockownerxstate",
1319 	    lo_state_lo_hash,
1320 	    lo_state_lo_compare,
1321 	    lo_state_lo_mkkey, TRUE);
1322 
1323 	rfs4_lo_state_idx = rfs4_index_create(rfs4_lo_state_tab,
1324 	    "State-id",
1325 	    lo_state_hash, lo_state_compare,
1326 	    lo_state_mkkey, FALSE);
1327 
1328 	rfs4_lockowner_cache_time *= rfs4_lease_time;
1329 
1330 	rfs4_lockowner_tab = rfs4_table_create(rfs4_server_state,
1331 	    "Lockowner",
1332 	    rfs4_lockowner_cache_time,
1333 	    2,
1334 	    rfs4_lockowner_create,
1335 	    rfs4_lockowner_destroy,
1336 	    rfs4_lockowner_expiry,
1337 	    sizeof (rfs4_lockowner_t),
1338 	    TABSIZE,
1339 	    MAXTABSZ, 100);
1340 
1341 	rfs4_lockowner_idx = rfs4_index_create(rfs4_lockowner_tab,
1342 	    "lock_owner4", lockowner_hash,
1343 	    lockowner_compare,
1344 	    lockowner_mkkey, TRUE);
1345 
1346 	rfs4_lockowner_pid_idx = rfs4_index_create(rfs4_lockowner_tab,
1347 	    "pid", pid_hash,
1348 	    pid_compare, pid_mkkey,
1349 	    FALSE);
1350 
1351 	rfs4_file_cache_time *= rfs4_lease_time;
1352 	rfs4_file_tab = rfs4_table_create(rfs4_server_state,
1353 	    "File",
1354 	    rfs4_file_cache_time,
1355 	    1,
1356 	    rfs4_file_create,
1357 	    rfs4_file_destroy,
1358 	    NULL,
1359 	    sizeof (rfs4_file_t),
1360 	    TABSIZE,
1361 	    MAXTABSZ, -1);
1362 
1363 	rfs4_file_idx = rfs4_index_create(rfs4_file_tab,
1364 	    "Filehandle", file_hash,
1365 	    file_compare, file_mkkey, TRUE);
1366 
1367 	rfs4_deleg_state_cache_time *= rfs4_lease_time;
1368 	rfs4_deleg_state_tab = rfs4_table_create(rfs4_server_state,
1369 	    "DelegStateID",
1370 	    rfs4_deleg_state_cache_time,
1371 	    2,
1372 	    rfs4_deleg_state_create,
1373 	    rfs4_deleg_state_destroy,
1374 	    rfs4_deleg_state_expiry,
1375 	    sizeof (rfs4_deleg_state_t),
1376 	    TABSIZE,
1377 	    MAXTABSZ, 100);
1378 	rfs4_deleg_idx = rfs4_index_create(rfs4_deleg_state_tab,
1379 	    "DelegByFileClient",
1380 	    deleg_hash,
1381 	    deleg_compare,
1382 	    deleg_mkkey, TRUE);
1383 
1384 	rfs4_deleg_state_idx = rfs4_index_create(rfs4_deleg_state_tab,
1385 	    "DelegState",
1386 	    deleg_state_hash,
1387 	    deleg_state_compare,
1388 	    deleg_state_mkkey, FALSE);
1389 
1390 	/*
1391 	 * Init the stable storage.
1392 	 */
1393 	rfs4_ss_init();
1394 
1395 	rfs4_client_clrst = rfs4_clear_client_state;
1396 
1397 	mutex_exit(&rfs4_state_lock);
1398 }
1399 
1400 
1401 /*
1402  * Used at server shutdown to cleanup all of the NFSv4 server's structures
1403  * and other state.
1404  */
1405 void
1406 rfs4_state_fini()
1407 {
1408 	rfs4_database_t *dbp;
1409 
1410 	mutex_enter(&rfs4_state_lock);
1411 
1412 	if (rfs4_server_state == NULL) {
1413 		mutex_exit(&rfs4_state_lock);
1414 		return;
1415 	}
1416 
1417 	rfs4_client_clrst = NULL;
1418 
1419 	rfs4_set_deleg_policy(SRV_NEVER_DELEGATE);
1420 	dbp = rfs4_server_state;
1421 	rfs4_server_state = NULL;
1422 
1423 	/*
1424 	 * Cleanup the CPR callback.
1425 	 */
1426 	if (cpr_id)
1427 		(void) callb_delete(cpr_id);
1428 
1429 	rw_destroy(&rfs4_findclient_lock);
1430 
1431 	/* First stop all of the reaper threads in the database */
1432 	rfs4_database_shutdown(dbp);
1433 	/* clean up any dangling stable storage structures */
1434 	rfs4_ss_fini();
1435 	/* Now actually destroy/release the database and its tables */
1436 	rfs4_database_destroy(dbp);
1437 
1438 	/* Reset the cache timers for next time */
1439 	rfs4_client_cache_time = 0;
1440 	rfs4_openowner_cache_time = 0;
1441 	rfs4_state_cache_time = 0;
1442 	rfs4_lo_state_cache_time = 0;
1443 	rfs4_lockowner_cache_time = 0;
1444 	rfs4_file_cache_time = 0;
1445 	rfs4_deleg_state_cache_time = 0;
1446 
1447 	mutex_exit(&rfs4_state_lock);
1448 
1449 	/* destroy server instances and current instance ptr */
1450 	rfs4_servinst_destroy_all();
1451 
1452 	/* reset the "first NFSv4 request" status */
1453 	rfs4_seen_first_compound = 0;
1454 
1455 	/* DSS: distributed stable storage */
1456 	if (rfs4_dss_oldpaths)
1457 		nvlist_free(rfs4_dss_oldpaths);
1458 	if (rfs4_dss_paths)
1459 		nvlist_free(rfs4_dss_paths);
1460 	rfs4_dss_paths = rfs4_dss_oldpaths = NULL;
1461 }
1462 
1463 typedef union {
1464 	struct {
1465 		uint32_t start_time;
1466 		uint32_t c_id;
1467 	} impl_id;
1468 	clientid4 id4;
1469 } cid;
1470 
1471 static int foreign_stateid(stateid_t *id);
1472 static int foreign_clientid(cid *cidp);
1473 static void embed_nodeid(cid *cidp);
1474 
1475 typedef union {
1476 	struct {
1477 		uint32_t c_id;
1478 		uint32_t gen_num;
1479 	} cv_impl;
1480 	verifier4	confirm_verf;
1481 } scid_confirm_verf;
1482 
1483 static uint32_t
1484 clientid_hash(void *key)
1485 {
1486 	cid *idp = key;
1487 
1488 	return (idp->impl_id.c_id);
1489 }
1490 
1491 static bool_t
1492 clientid_compare(rfs4_entry_t entry, void *key)
1493 {
1494 	rfs4_client_t *client = (rfs4_client_t *)entry;
1495 	clientid4 *idp = key;
1496 
1497 	return (*idp == client->clientid);
1498 }
1499 
1500 static void *
1501 clientid_mkkey(rfs4_entry_t entry)
1502 {
1503 	rfs4_client_t *client = (rfs4_client_t *)entry;
1504 
1505 	return (&client->clientid);
1506 }
1507 
1508 static uint32_t
1509 nfsclnt_hash(void *key)
1510 {
1511 	nfs_client_id4 *client = key;
1512 	int i;
1513 	uint32_t hash = 0;
1514 
1515 	for (i = 0; i < client->id_len; i++) {
1516 		hash <<= 1;
1517 		hash += (uint_t)client->id_val[i];
1518 	}
1519 	return (hash);
1520 }
1521 
1522 
1523 static bool_t
1524 nfsclnt_compare(rfs4_entry_t entry, void *key)
1525 {
1526 	rfs4_client_t *client = (rfs4_client_t *)entry;
1527 	nfs_client_id4 *nfs_client = key;
1528 
1529 	if (client->nfs_client.id_len != nfs_client->id_len)
1530 		return (FALSE);
1531 
1532 	return (bcmp(client->nfs_client.id_val, nfs_client->id_val,
1533 	    nfs_client->id_len) == 0);
1534 }
1535 
1536 static void *
1537 nfsclnt_mkkey(rfs4_entry_t entry)
1538 {
1539 	rfs4_client_t *client = (rfs4_client_t *)entry;
1540 
1541 	return (&client->nfs_client);
1542 }
1543 
1544 static bool_t
1545 rfs4_client_expiry(rfs4_entry_t u_entry)
1546 {
1547 	rfs4_client_t *cp = (rfs4_client_t *)u_entry;
1548 	bool_t cp_expired;
1549 
1550 	if (rfs4_dbe_is_invalid(cp->dbe))
1551 		return (TRUE);
1552 	/*
1553 	 * If the sysadmin has used clear_locks for this
1554 	 * entry then forced_expire will be set and we
1555 	 * want this entry to be reaped. Or the entry
1556 	 * has exceeded its lease period.
1557 	 */
1558 	cp_expired = (cp->forced_expire ||
1559 	    (gethrestime_sec() - cp->last_access
1560 	    > rfs4_lease_time));
1561 
1562 	if (!cp->ss_remove && cp_expired)
1563 		cp->ss_remove = 1;
1564 	return (cp_expired);
1565 }
1566 
1567 /*
1568  * Remove the leaf file from all distributed stable storage paths.
1569  */
1570 static void
1571 rfs4_dss_remove_cpleaf(rfs4_client_t *cp)
1572 {
1573 	char *leaf = cp->ss_pn->leaf;
1574 
1575 	rfs4_dss_remove_leaf(cp->server_instance, NFS4_DSS_STATE_LEAF, leaf);
1576 }
1577 
1578 static void
1579 rfs4_dss_remove_leaf(rfs4_servinst_t *sip, char *dir_leaf, char *leaf)
1580 {
1581 	int i, npaths = sip->dss_npaths;
1582 
1583 	for (i = 0; i < npaths; i++) {
1584 		rfs4_dss_path_t *dss_path = sip->dss_paths[i];
1585 		char *path, *dir;
1586 		size_t pathlen;
1587 
1588 		/* the HA-NFSv4 path might have been failed-over away from us */
1589 		if (dss_path == NULL)
1590 			continue;
1591 
1592 		dir = dss_path->path;
1593 
1594 		/* allow 3 extra bytes for two '/' & a NUL */
1595 		pathlen = strlen(dir) + strlen(dir_leaf) + strlen(leaf) + 3;
1596 		path = kmem_alloc(pathlen, KM_SLEEP);
1597 		(void) sprintf(path, "%s/%s/%s", dir, dir_leaf, leaf);
1598 
1599 		(void) vn_remove(path, UIO_SYSSPACE, RMFILE);
1600 
1601 		kmem_free(path, pathlen);
1602 	}
1603 }
1604 
1605 static void
1606 rfs4_client_destroy(rfs4_entry_t u_entry)
1607 {
1608 	rfs4_client_t *cp = (rfs4_client_t *)u_entry;
1609 
1610 	mutex_destroy(cp->cbinfo.cb_lock);
1611 	cv_destroy(cp->cbinfo.cb_cv);
1612 	cv_destroy(cp->cbinfo.cb_cv_nullcaller);
1613 
1614 	/* free callback info */
1615 	rfs4_cbinfo_free(&cp->cbinfo);
1616 
1617 	if (cp->cp_confirmed)
1618 		rfs4_client_rele(cp->cp_confirmed);
1619 
1620 	if (cp->ss_pn) {
1621 		/* check if the stable storage files need to be removed */
1622 		if (cp->ss_remove)
1623 			rfs4_dss_remove_cpleaf(cp);
1624 		rfs4_ss_pnfree(cp->ss_pn);
1625 	}
1626 
1627 	/* Free the client supplied client id */
1628 	kmem_free(cp->nfs_client.id_val, cp->nfs_client.id_len);
1629 
1630 	if (cp->sysidt != LM_NOSYSID)
1631 		lm_free_sysidt(cp->sysidt);
1632 }
1633 
1634 static bool_t
1635 rfs4_client_create(rfs4_entry_t u_entry, void *arg)
1636 {
1637 	rfs4_client_t *cp = (rfs4_client_t *)u_entry;
1638 	nfs_client_id4 *client = (nfs_client_id4 *)arg;
1639 	cid *cidp;
1640 	scid_confirm_verf *scvp;
1641 
1642 	/* Get a clientid to give to the client */
1643 	cidp = (cid *)&cp->clientid;
1644 	cidp->impl_id.start_time = rfs4_start_time;
1645 	cidp->impl_id.c_id = (uint32_t)rfs4_dbe_getid(cp->dbe);
1646 
1647 	/* If we are booted as a cluster node, embed our nodeid */
1648 	if (cluster_bootflags & CLUSTER_BOOTED)
1649 		embed_nodeid(cidp);
1650 
1651 	/* Allocate and copy client's client id value */
1652 	cp->nfs_client.id_val = kmem_alloc(client->id_len, KM_SLEEP);
1653 	cp->nfs_client.id_len = client->id_len;
1654 	bcopy(client->id_val, cp->nfs_client.id_val, client->id_len);
1655 	cp->nfs_client.verifier = client->verifier;
1656 
1657 	/* Init the value for the SETCLIENTID_CONFIRM verifier */
1658 	scvp = (scid_confirm_verf *)&cp->confirm_verf;
1659 	scvp->cv_impl.c_id = cidp->impl_id.c_id;
1660 	scvp->cv_impl.gen_num = 0;
1661 
1662 	/* An F_UNLKSYS has been done for this client */
1663 	cp->unlksys_completed = FALSE;
1664 
1665 	/* We need the client to ack us */
1666 	cp->need_confirm = TRUE;
1667 	cp->cp_confirmed = NULL;
1668 
1669 	/* TRUE all the time until the callback path actually fails */
1670 	cp->cbinfo.cb_notified_of_cb_path_down = TRUE;
1671 
1672 	/* Initialize the access time to now */
1673 	cp->last_access = gethrestime_sec();
1674 
1675 	cp->cr_set = NULL;
1676 	/* Initialize list for insque/remque */
1677 	cp->openownerlist.next = cp->openownerlist.prev = &cp->openownerlist;
1678 	cp->openownerlist.oop = NULL; /* This is not an openowner */
1679 
1680 	cp->sysidt = LM_NOSYSID;
1681 
1682 	cp->clientdeleglist.next = cp->clientdeleglist.prev =
1683 	    &cp->clientdeleglist;
1684 	cp->clientdeleglist.dsp = NULL;
1685 
1686 	/* set up the callback control structure */
1687 	cp->cbinfo.cb_state = CB_UNINIT;
1688 	mutex_init(cp->cbinfo.cb_lock, NULL, MUTEX_DEFAULT, NULL);
1689 	cv_init(cp->cbinfo.cb_cv, NULL, CV_DEFAULT, NULL);
1690 	cv_init(cp->cbinfo.cb_cv_nullcaller, NULL, CV_DEFAULT, NULL);
1691 
1692 	/*
1693 	 * Associate the client_t with the current server instance.
1694 	 * The hold is solely to satisfy the calling requirement of
1695 	 * rfs4_servinst_assign(). In this case it's not strictly necessary.
1696 	 */
1697 	rfs4_dbe_hold(cp->dbe);
1698 	rfs4_servinst_assign(cp, rfs4_cur_servinst);
1699 	rfs4_dbe_rele(cp->dbe);
1700 
1701 	return (TRUE);
1702 }
1703 
1704 /*
1705  * Caller wants to generate/update the setclientid_confirm verifier
1706  * associated with a client.  This is done during the SETCLIENTID
1707  * processing.
1708  */
1709 void
1710 rfs4_client_scv_next(rfs4_client_t *cp)
1711 {
1712 	scid_confirm_verf *scvp;
1713 
1714 	/* Init the value for the SETCLIENTID_CONFIRM verifier */
1715 	scvp = (scid_confirm_verf *)&cp->confirm_verf;
1716 	scvp->cv_impl.gen_num++;
1717 }
1718 
1719 void
1720 rfs4_client_rele(rfs4_client_t *cp)
1721 {
1722 	rfs4_dbe_rele(cp->dbe);
1723 }
1724 
1725 rfs4_client_t *
1726 rfs4_findclient(nfs_client_id4 *client, bool_t *create,	rfs4_client_t *oldcp)
1727 {
1728 	rfs4_client_t *cp;
1729 
1730 
1731 	if (oldcp) {
1732 		rw_enter(&rfs4_findclient_lock, RW_WRITER);
1733 		rfs4_dbe_hide(oldcp->dbe);
1734 	} else {
1735 		rw_enter(&rfs4_findclient_lock, RW_READER);
1736 	}
1737 
1738 	cp = (rfs4_client_t *)rfs4_dbsearch(rfs4_nfsclnt_idx, client,
1739 	    create, (void *)client, RFS4_DBS_VALID);
1740 
1741 	if (oldcp)
1742 		rfs4_dbe_unhide(oldcp->dbe);
1743 
1744 	rw_exit(&rfs4_findclient_lock);
1745 
1746 	return (cp);
1747 }
1748 
1749 rfs4_client_t *
1750 rfs4_findclient_by_id(clientid4 clientid, bool_t find_unconfirmed)
1751 {
1752 	rfs4_client_t *cp;
1753 	bool_t create = FALSE;
1754 	cid *cidp = (cid *)&clientid;
1755 
1756 	/* If we're a cluster and the nodeid isn't right, short-circuit */
1757 	if (cluster_bootflags & CLUSTER_BOOTED && foreign_clientid(cidp))
1758 		return (NULL);
1759 
1760 	rw_enter(&rfs4_findclient_lock, RW_READER);
1761 
1762 	cp = (rfs4_client_t *)rfs4_dbsearch(rfs4_clientid_idx, &clientid,
1763 	    &create, NULL, RFS4_DBS_VALID);
1764 
1765 	rw_exit(&rfs4_findclient_lock);
1766 
1767 	if (cp && cp->need_confirm && find_unconfirmed == FALSE) {
1768 		rfs4_client_rele(cp);
1769 		return (NULL);
1770 	} else {
1771 		return (cp);
1772 	}
1773 }
1774 
1775 bool_t
1776 rfs4_lease_expired(rfs4_client_t *cp)
1777 {
1778 	bool_t rc;
1779 
1780 	rfs4_dbe_lock(cp->dbe);
1781 
1782 	/*
1783 	 * If the admin has executed clear_locks for this
1784 	 * client id, force expire will be set, so no need
1785 	 * to calculate anything because it's "outa here".
1786 	 */
1787 	if (cp->forced_expire) {
1788 		rc = TRUE;
1789 	} else {
1790 		rc = (gethrestime_sec() - cp->last_access > rfs4_lease_time);
1791 	}
1792 
1793 	/*
1794 	 * If the lease has expired we will also want
1795 	 * to remove any stable storage state data. So
1796 	 * mark the client id accordingly.
1797 	 */
1798 	if (!cp->ss_remove)
1799 		cp->ss_remove = (rc == TRUE);
1800 
1801 	rfs4_dbe_unlock(cp->dbe);
1802 
1803 	return (rc);
1804 }
1805 
1806 void
1807 rfs4_update_lease(rfs4_client_t *cp)
1808 {
1809 	rfs4_dbe_lock(cp->dbe);
1810 	if (!cp->forced_expire)
1811 		cp->last_access = gethrestime_sec();
1812 	rfs4_dbe_unlock(cp->dbe);
1813 }
1814 
1815 
1816 static bool_t
1817 EQOPENOWNER(open_owner4 *a, open_owner4 *b)
1818 {
1819 	bool_t rc;
1820 
1821 	if (a->clientid != b->clientid)
1822 		return (FALSE);
1823 
1824 	if (a->owner_len != b->owner_len)
1825 		return (FALSE);
1826 
1827 	rc = (bcmp(a->owner_val, b->owner_val, a->owner_len) == 0);
1828 
1829 	return (rc);
1830 }
1831 
1832 static uint_t
1833 openowner_hash(void *key)
1834 {
1835 	int i;
1836 	open_owner4 *openowner = key;
1837 	uint_t hash = 0;
1838 
1839 	for (i = 0; i < openowner->owner_len; i++) {
1840 		hash <<= 4;
1841 		hash += (uint_t)openowner->owner_val[i];
1842 	}
1843 	hash += (uint_t)openowner->clientid;
1844 	hash |= (openowner->clientid >> 32);
1845 
1846 	return (hash);
1847 }
1848 
1849 static bool_t
1850 openowner_compare(rfs4_entry_t u_entry, void *key)
1851 {
1852 	rfs4_openowner_t *op = (rfs4_openowner_t *)u_entry;
1853 	open_owner4 *arg = key;
1854 
1855 	return (EQOPENOWNER(&op->owner, arg));
1856 }
1857 
1858 void *
1859 openowner_mkkey(rfs4_entry_t u_entry)
1860 {
1861 	rfs4_openowner_t *op = (rfs4_openowner_t *)u_entry;
1862 
1863 	return (&op->owner);
1864 }
1865 
1866 static bool_t
1867 rfs4_openowner_expiry(rfs4_entry_t u_entry)
1868 {
1869 	rfs4_openowner_t *op = (rfs4_openowner_t *)u_entry;
1870 
1871 	if (rfs4_dbe_is_invalid(op->dbe))
1872 		return (TRUE);
1873 	return ((gethrestime_sec() - op->client->last_access
1874 	    > rfs4_lease_time));
1875 }
1876 
1877 static void
1878 rfs4_openowner_destroy(rfs4_entry_t u_entry)
1879 {
1880 	rfs4_openowner_t *op = (rfs4_openowner_t *)u_entry;
1881 
1882 	rfs4_sw_destroy(&op->oo_sw);
1883 
1884 	/* Remove open owner from client's lists of open owners */
1885 	rfs4_dbe_lock(op->client->dbe);
1886 
1887 	remque(&op->openownerlist);
1888 	op->openownerlist.next = op->openownerlist.prev = &op->openownerlist;
1889 
1890 	rfs4_dbe_unlock(op->client->dbe);
1891 
1892 	/* One less reference to the client */
1893 	rfs4_client_rele(op->client);
1894 	op->client = NULL;
1895 
1896 	/* Free the last reply for this lock owner */
1897 	rfs4_free_reply(op->reply);
1898 
1899 	if (op->reply_fh.nfs_fh4_val) {
1900 		kmem_free(op->reply_fh.nfs_fh4_val, op->reply_fh.nfs_fh4_len);
1901 		op->reply_fh.nfs_fh4_val = NULL;
1902 		op->reply_fh.nfs_fh4_len = 0;
1903 	}
1904 
1905 	/* Free the lock owner id */
1906 	kmem_free(op->owner.owner_val, op->owner.owner_len);
1907 }
1908 
1909 void
1910 rfs4_openowner_rele(rfs4_openowner_t *op)
1911 {
1912 	rfs4_dbe_rele(op->dbe);
1913 }
1914 
1915 static bool_t
1916 rfs4_openowner_create(rfs4_entry_t u_entry, void *arg)
1917 {
1918 	rfs4_openowner_t *op = (rfs4_openowner_t *)u_entry;
1919 	rfs4_openowner_t *argp = (rfs4_openowner_t *)arg;
1920 	open_owner4 *openowner = &argp->owner;
1921 	seqid4 seqid = argp->open_seqid;
1922 	rfs4_client_t *cp;
1923 	bool_t create = FALSE;
1924 
1925 	rw_enter(&rfs4_findclient_lock, RW_READER);
1926 
1927 	cp = (rfs4_client_t *)rfs4_dbsearch(rfs4_clientid_idx,
1928 	    &openowner->clientid,
1929 	    &create, NULL, RFS4_DBS_VALID);
1930 
1931 	rw_exit(&rfs4_findclient_lock);
1932 
1933 	if (cp == NULL)
1934 		return (FALSE);
1935 
1936 	op->reply_fh.nfs_fh4_len = 0;
1937 	op->reply_fh.nfs_fh4_val = NULL;
1938 
1939 	op->owner.clientid = openowner->clientid;
1940 	op->owner.owner_val =
1941 	    kmem_alloc(openowner->owner_len, KM_SLEEP);
1942 
1943 	bcopy(openowner->owner_val,
1944 	    op->owner.owner_val, openowner->owner_len);
1945 
1946 	op->owner.owner_len = openowner->owner_len;
1947 
1948 	op->need_confirm = TRUE;
1949 
1950 	rfs4_sw_init(&op->oo_sw);
1951 
1952 	op->open_seqid = seqid;
1953 	bzero(op->reply, sizeof (nfs_resop4));
1954 	op->client = cp;
1955 	op->cr_set = NULL;
1956 	/* Init lists for remque/insque */
1957 	op->ownerstateids.next = op->ownerstateids.prev = &op->ownerstateids;
1958 	op->ownerstateids.sp = NULL; /* NULL since this is the state list */
1959 	op->openownerlist.next = op->openownerlist.prev = &op->openownerlist;
1960 	op->openownerlist.oop = op; /* ourselves */
1961 
1962 	/* Insert openowner into client's open owner list */
1963 	rfs4_dbe_lock(cp->dbe);
1964 
1965 	insque(&op->openownerlist, cp->openownerlist.prev);
1966 
1967 	rfs4_dbe_unlock(cp->dbe);
1968 
1969 	return (TRUE);
1970 }
1971 
1972 rfs4_openowner_t *
1973 rfs4_findopenowner(open_owner4 *openowner, bool_t *create, seqid4 seqid)
1974 {
1975 	rfs4_openowner_t *op;
1976 	rfs4_openowner_t arg;
1977 
1978 	arg.owner = *openowner;
1979 	arg.open_seqid = seqid;
1980 	op = (rfs4_openowner_t *)rfs4_dbsearch(rfs4_openowner_idx, openowner,
1981 	    create, &arg, RFS4_DBS_VALID);
1982 
1983 	return (op);
1984 }
1985 
1986 void
1987 rfs4_update_open_sequence(rfs4_openowner_t *op)
1988 {
1989 
1990 	rfs4_dbe_lock(op->dbe);
1991 
1992 	op->open_seqid++;
1993 
1994 	rfs4_dbe_unlock(op->dbe);
1995 }
1996 
1997 void
1998 rfs4_update_open_resp(rfs4_openowner_t *op, nfs_resop4 *resp, nfs_fh4 *fh)
1999 {
2000 
2001 	rfs4_dbe_lock(op->dbe);
2002 
2003 	rfs4_free_reply(op->reply);
2004 
2005 	rfs4_copy_reply(op->reply, resp);
2006 
2007 	/* Save the filehandle if provided and free if not used */
2008 	if (resp->nfs_resop4_u.opopen.status == NFS4_OK &&
2009 	    fh && fh->nfs_fh4_len) {
2010 		if (op->reply_fh.nfs_fh4_val == NULL)
2011 			op->reply_fh.nfs_fh4_val =
2012 			    kmem_alloc(fh->nfs_fh4_len, KM_SLEEP);
2013 		nfs_fh4_copy(fh, &op->reply_fh);
2014 	} else {
2015 		if (op->reply_fh.nfs_fh4_val) {
2016 			kmem_free(op->reply_fh.nfs_fh4_val,
2017 			    op->reply_fh.nfs_fh4_len);
2018 			op->reply_fh.nfs_fh4_val = NULL;
2019 			op->reply_fh.nfs_fh4_len = 0;
2020 		}
2021 	}
2022 
2023 	rfs4_dbe_unlock(op->dbe);
2024 }
2025 
2026 static bool_t
2027 lockowner_compare(rfs4_entry_t u_entry, void *key)
2028 {
2029 	rfs4_lockowner_t *lo = (rfs4_lockowner_t *)u_entry;
2030 	lock_owner4 *b = (lock_owner4 *)key;
2031 
2032 	if (lo->owner.clientid != b->clientid)
2033 		return (FALSE);
2034 
2035 	if (lo->owner.owner_len != b->owner_len)
2036 		return (FALSE);
2037 
2038 	return (bcmp(lo->owner.owner_val, b->owner_val,
2039 	    lo->owner.owner_len) == 0);
2040 }
2041 
2042 void *
2043 lockowner_mkkey(rfs4_entry_t u_entry)
2044 {
2045 	rfs4_lockowner_t *lo = (rfs4_lockowner_t *)u_entry;
2046 
2047 	return (&lo->owner);
2048 }
2049 
2050 static uint32_t
2051 lockowner_hash(void *key)
2052 {
2053 	int i;
2054 	lock_owner4 *lockowner = key;
2055 	uint_t hash = 0;
2056 
2057 	for (i = 0; i < lockowner->owner_len; i++) {
2058 		hash <<= 4;
2059 		hash += (uint_t)lockowner->owner_val[i];
2060 	}
2061 	hash += (uint_t)lockowner->clientid;
2062 	hash |= (lockowner->clientid >> 32);
2063 
2064 	return (hash);
2065 }
2066 
2067 static uint32_t
2068 pid_hash(void *key)
2069 {
2070 	return ((uint32_t)(uintptr_t)key);
2071 }
2072 
2073 static void *
2074 pid_mkkey(rfs4_entry_t u_entry)
2075 {
2076 	rfs4_lockowner_t *lo = (rfs4_lockowner_t *)u_entry;
2077 
2078 	return ((void *)(uintptr_t)lo->pid);
2079 }
2080 
2081 static bool_t
2082 pid_compare(rfs4_entry_t u_entry, void *key)
2083 {
2084 	rfs4_lockowner_t *lo = (rfs4_lockowner_t *)u_entry;
2085 
2086 	return (lo->pid == (pid_t)(uintptr_t)key);
2087 }
2088 
2089 static void
2090 rfs4_lockowner_destroy(rfs4_entry_t u_entry)
2091 {
2092 	rfs4_lockowner_t *lo = (rfs4_lockowner_t *)u_entry;
2093 
2094 	/* Free the lock owner id */
2095 	kmem_free(lo->owner.owner_val, lo->owner.owner_len);
2096 	rfs4_client_rele(lo->client);
2097 }
2098 
2099 void
2100 rfs4_lockowner_rele(rfs4_lockowner_t *lo)
2101 {
2102 	rfs4_dbe_rele(lo->dbe);
2103 }
2104 
2105 /* ARGSUSED */
2106 static bool_t
2107 rfs4_lockowner_expiry(rfs4_entry_t u_entry)
2108 {
2109 	/*
2110 	 * Since expiry is called with no other references on
2111 	 * this struct, go ahead and have it removed.
2112 	 */
2113 	return (TRUE);
2114 }
2115 
2116 static bool_t
2117 rfs4_lockowner_create(rfs4_entry_t u_entry, void *arg)
2118 {
2119 	rfs4_lockowner_t *lo = (rfs4_lockowner_t *)u_entry;
2120 	lock_owner4 *lockowner = (lock_owner4 *)arg;
2121 	rfs4_client_t *cp;
2122 	bool_t create = FALSE;
2123 
2124 	rw_enter(&rfs4_findclient_lock, RW_READER);
2125 
2126 	cp = (rfs4_client_t *)rfs4_dbsearch(rfs4_clientid_idx,
2127 	    &lockowner->clientid,
2128 	    &create, NULL, RFS4_DBS_VALID);
2129 
2130 	rw_exit(&rfs4_findclient_lock);
2131 
2132 	if (cp == NULL)
2133 		return (FALSE);
2134 
2135 	/* Reference client */
2136 	lo->client = cp;
2137 	lo->owner.clientid = lockowner->clientid;
2138 	lo->owner.owner_val = kmem_alloc(lockowner->owner_len, KM_SLEEP);
2139 	bcopy(lockowner->owner_val, lo->owner.owner_val, lockowner->owner_len);
2140 	lo->owner.owner_len = lockowner->owner_len;
2141 	lo->pid = rfs4_dbe_getid(lo->dbe);
2142 
2143 	return (TRUE);
2144 }
2145 
2146 rfs4_lockowner_t *
2147 rfs4_findlockowner(lock_owner4 *lockowner, bool_t *create)
2148 {
2149 	rfs4_lockowner_t *lo;
2150 
2151 	lo = (rfs4_lockowner_t *)rfs4_dbsearch(rfs4_lockowner_idx, lockowner,
2152 	    create, lockowner, RFS4_DBS_VALID);
2153 
2154 	return (lo);
2155 }
2156 
2157 rfs4_lockowner_t *
2158 rfs4_findlockowner_by_pid(pid_t pid)
2159 {
2160 	rfs4_lockowner_t *lo;
2161 	bool_t create = FALSE;
2162 
2163 	lo = (rfs4_lockowner_t *)rfs4_dbsearch(rfs4_lockowner_pid_idx,
2164 	    (void *)(uintptr_t)pid, &create, NULL, RFS4_DBS_VALID);
2165 
2166 	return (lo);
2167 }
2168 
2169 
2170 static uint32_t
2171 file_hash(void *key)
2172 {
2173 	return (ADDRHASH(key));
2174 }
2175 
2176 static void *
2177 file_mkkey(rfs4_entry_t u_entry)
2178 {
2179 	rfs4_file_t *fp = (rfs4_file_t *)u_entry;
2180 
2181 	return (fp->vp);
2182 }
2183 
2184 static bool_t
2185 file_compare(rfs4_entry_t u_entry, void *key)
2186 {
2187 	rfs4_file_t *fp = (rfs4_file_t *)u_entry;
2188 
2189 	return (fp->vp == (vnode_t *)key);
2190 }
2191 
2192 static void
2193 rfs4_file_destroy(rfs4_entry_t u_entry)
2194 {
2195 	rfs4_file_t *fp = (rfs4_file_t *)u_entry;
2196 
2197 	ASSERT(fp->delegationlist.next == &fp->delegationlist);
2198 	if (fp->filehandle.nfs_fh4_val)
2199 		kmem_free(fp->filehandle.nfs_fh4_val,
2200 		    fp->filehandle.nfs_fh4_len);
2201 	cv_destroy(fp->dinfo->recall_cv);
2202 	if (fp->vp) {
2203 		vnode_t *vp = fp->vp;
2204 
2205 		mutex_enter(&vp->v_lock);
2206 		(void) vsd_set(vp, nfs4_srv_vkey, NULL);
2207 		mutex_exit(&vp->v_lock);
2208 		VN_RELE(vp);
2209 		fp->vp = NULL;
2210 	}
2211 	rw_destroy(&fp->file_rwlock);
2212 }
2213 
2214 /*
2215  * Used to unlock the underlying dbe struct only
2216  */
2217 void
2218 rfs4_file_rele(rfs4_file_t *fp)
2219 {
2220 	rfs4_dbe_rele(fp->dbe);
2221 }
2222 
2223 /*
2224  * Used to unlock the file rw lock and the file's dbe entry
2225  * Only used to pair with rfs4_findfile_withlock()
2226  */
2227 void
2228 rfs4_file_rele_withunlock(rfs4_file_t *fp)
2229 {
2230 	rw_exit(&fp->file_rwlock);
2231 	rfs4_dbe_rele(fp->dbe);
2232 }
2233 
2234 typedef struct {
2235     vnode_t *vp;
2236     nfs_fh4 *fh;
2237 } rfs4_fcreate_arg;
2238 
2239 static bool_t
2240 rfs4_file_create(rfs4_entry_t u_entry, void *arg)
2241 {
2242 	rfs4_file_t *fp = (rfs4_file_t *)u_entry;
2243 	rfs4_fcreate_arg *ap = (rfs4_fcreate_arg *)arg;
2244 	vnode_t *vp = ap->vp;
2245 	nfs_fh4 *fh = ap->fh;
2246 
2247 	VN_HOLD(vp);
2248 
2249 	fp->filehandle.nfs_fh4_len = 0;
2250 	fp->filehandle.nfs_fh4_val = NULL;
2251 	ASSERT(fh && fh->nfs_fh4_len);
2252 	if (fh && fh->nfs_fh4_len) {
2253 		fp->filehandle.nfs_fh4_val =
2254 		    kmem_alloc(fh->nfs_fh4_len, KM_SLEEP);
2255 		nfs_fh4_copy(fh, &fp->filehandle);
2256 	}
2257 	fp->vp = vp;
2258 
2259 	/* Init list for remque/insque */
2260 	fp->delegationlist.next = fp->delegationlist.prev =
2261 	    &fp->delegationlist;
2262 	fp->delegationlist.dsp = NULL; /* NULL since this is state list */
2263 
2264 	fp->share_deny = fp->share_access = fp->access_read = 0;
2265 	fp->access_write = fp->deny_read = fp->deny_write = 0;
2266 
2267 	mutex_init(fp->dinfo->recall_lock, NULL, MUTEX_DEFAULT, NULL);
2268 	cv_init(fp->dinfo->recall_cv, NULL, CV_DEFAULT, NULL);
2269 
2270 	fp->dinfo->dtype = OPEN_DELEGATE_NONE;
2271 
2272 	rw_init(&fp->file_rwlock, NULL, RW_DEFAULT, NULL);
2273 
2274 	mutex_enter(&vp->v_lock);
2275 	if (vsd_set(vp, nfs4_srv_vkey, (void *)fp)) {
2276 		ASSERT(FALSE);
2277 		cmn_err(CE_WARN, "rfs4_file_create: vsd_set failed.");
2278 	}
2279 	mutex_exit(&vp->v_lock);
2280 
2281 	return (TRUE);
2282 }
2283 
2284 rfs4_file_t *
2285 rfs4_findfile(vnode_t *vp, nfs_fh4 *fh, bool_t *create)
2286 {
2287 	rfs4_file_t *fp;
2288 	rfs4_fcreate_arg arg;
2289 
2290 	arg.vp = vp;
2291 	arg.fh = fh;
2292 
2293 	if (*create == TRUE)
2294 		fp = (rfs4_file_t *)rfs4_dbsearch(rfs4_file_idx, vp, create,
2295 		    &arg, RFS4_DBS_VALID);
2296 	else {
2297 		mutex_enter(&vp->v_lock);
2298 		fp = (rfs4_file_t *)vsd_get(vp, nfs4_srv_vkey);
2299 		mutex_exit(&vp->v_lock);
2300 		if (fp) {
2301 			rfs4_dbe_lock(fp->dbe);
2302 			if (rfs4_dbe_is_invalid(fp->dbe) ||
2303 			    (rfs4_dbe_refcnt(fp->dbe) == 0)) {
2304 				rfs4_dbe_unlock(fp->dbe);
2305 				fp = NULL;
2306 			} else {
2307 				rfs4_dbe_hold(fp->dbe);
2308 				rfs4_dbe_unlock(fp->dbe);
2309 			}
2310 		}
2311 	}
2312 	return (fp);
2313 }
2314 
2315 /*
2316  * Find a file in the db and once it is located, take the rw lock.
2317  * Need to check the vnode pointer and if it does not exist (it was
2318  * removed between the db location and check) redo the find.  This
2319  * assumes that a file struct that has a NULL vnode pointer is marked
2320  * at 'invalid' and will not be found in the db the second time
2321  * around.
2322  */
2323 rfs4_file_t *
2324 rfs4_findfile_withlock(vnode_t *vp, nfs_fh4 *fh, bool_t *create)
2325 {
2326 	rfs4_file_t *fp;
2327 	rfs4_fcreate_arg arg;
2328 	bool_t screate = *create;
2329 
2330 	if (screate == FALSE) {
2331 		mutex_enter(&vp->v_lock);
2332 		fp = (rfs4_file_t *)vsd_get(vp, nfs4_srv_vkey);
2333 		mutex_exit(&vp->v_lock);
2334 		if (fp) {
2335 			rfs4_dbe_lock(fp->dbe);
2336 			if (rfs4_dbe_is_invalid(fp->dbe) ||
2337 			    (rfs4_dbe_refcnt(fp->dbe) == 0)) {
2338 				rfs4_dbe_unlock(fp->dbe);
2339 				fp = NULL;
2340 			} else {
2341 				rfs4_dbe_hold(fp->dbe);
2342 				rfs4_dbe_unlock(fp->dbe);
2343 				rw_enter(&fp->file_rwlock, RW_WRITER);
2344 			}
2345 		}
2346 	} else {
2347 retry:
2348 		arg.vp = vp;
2349 		arg.fh = fh;
2350 
2351 		fp = (rfs4_file_t *)rfs4_dbsearch(rfs4_file_idx, vp, create,
2352 		    &arg, RFS4_DBS_VALID);
2353 		if (fp != NULL) {
2354 			rw_enter(&fp->file_rwlock, RW_WRITER);
2355 			if (fp->vp == NULL) {
2356 				rw_exit(&fp->file_rwlock);
2357 				rfs4_file_rele(fp);
2358 				*create = screate;
2359 				goto retry;
2360 			}
2361 		}
2362 	}
2363 
2364 	return (fp);
2365 }
2366 
2367 static uint32_t
2368 lo_state_hash(void *key)
2369 {
2370 	stateid_t *id = key;
2371 
2372 	return (id->bits.ident+id->bits.pid);
2373 }
2374 
2375 static bool_t
2376 lo_state_compare(rfs4_entry_t u_entry, void *key)
2377 {
2378 	rfs4_lo_state_t *lop = (rfs4_lo_state_t *)u_entry;
2379 	stateid_t *id = key;
2380 	bool_t rc;
2381 
2382 	rc = (lop->lockid.bits.boottime == id->bits.boottime &&
2383 	    lop->lockid.bits.type == id->bits.type &&
2384 	    lop->lockid.bits.ident == id->bits.ident &&
2385 	    lop->lockid.bits.pid == id->bits.pid);
2386 
2387 	return (rc);
2388 }
2389 
2390 static void *
2391 lo_state_mkkey(rfs4_entry_t u_entry)
2392 {
2393 	rfs4_lo_state_t *lsp = (rfs4_lo_state_t *)u_entry;
2394 
2395 	return (&lsp->lockid);
2396 }
2397 
2398 static bool_t
2399 rfs4_lo_state_expiry(rfs4_entry_t u_entry)
2400 {
2401 	rfs4_lo_state_t *lsp = (rfs4_lo_state_t *)u_entry;
2402 
2403 	if (rfs4_dbe_is_invalid(lsp->dbe))
2404 		return (TRUE);
2405 	if (lsp->state->closed)
2406 		return (TRUE);
2407 	return ((gethrestime_sec() - lsp->state->owner->client->last_access
2408 	    > rfs4_lease_time));
2409 }
2410 
2411 static void
2412 rfs4_lo_state_destroy(rfs4_entry_t u_entry)
2413 {
2414 	rfs4_lo_state_t *lsp = (rfs4_lo_state_t *)u_entry;
2415 
2416 	rfs4_sw_destroy(&lsp->ls_sw);
2417 
2418 	/* Make sure to release the file locks */
2419 	if (lsp->locks_cleaned == FALSE) {
2420 		lsp->locks_cleaned = TRUE;
2421 		if (lsp->locker->client->sysidt != LM_NOSYSID) {
2422 			/* Is the PxFS kernel module loaded? */
2423 			if (lm_remove_file_locks != NULL) {
2424 				int new_sysid;
2425 
2426 				/* Encode the cluster nodeid in new sysid */
2427 				new_sysid = lsp->locker->client->sysidt;
2428 				lm_set_nlmid_flk(&new_sysid);
2429 
2430 				/*
2431 				 * This PxFS routine removes file locks for a
2432 				 * client over all nodes of a cluster.
2433 				 */
2434 				DTRACE_PROBE1(nfss_i_clust_rm_lck,
2435 				    int, new_sysid);
2436 				(*lm_remove_file_locks)(new_sysid);
2437 			} else {
2438 				(void) cleanlocks(lsp->state->finfo->vp,
2439 				    lsp->locker->pid,
2440 				    lsp->locker->client->sysidt);
2441 			}
2442 		}
2443 	}
2444 
2445 	rfs4_dbe_lock(lsp->state->dbe);
2446 
2447 	remque(&lsp->lockownerlist);
2448 	lsp->lockownerlist.next = lsp->lockownerlist.prev =
2449 	    &lsp->lockownerlist;
2450 
2451 	rfs4_dbe_unlock(lsp->state->dbe);
2452 
2453 	/* Free the last reply for this state */
2454 	rfs4_free_reply(lsp->reply);
2455 
2456 	rfs4_lockowner_rele(lsp->locker);
2457 	lsp->locker = NULL;
2458 
2459 	rfs4_state_rele_nounlock(lsp->state);
2460 	lsp->state = NULL;
2461 }
2462 
2463 static bool_t
2464 rfs4_lo_state_create(rfs4_entry_t u_entry, void *arg)
2465 {
2466 	rfs4_lo_state_t *lsp = (rfs4_lo_state_t *)u_entry;
2467 	rfs4_lo_state_t *argp = (rfs4_lo_state_t *)arg;
2468 	rfs4_lockowner_t *lo = argp->locker;
2469 	rfs4_state_t *sp = argp->state;
2470 
2471 	lsp->state = sp;
2472 
2473 	lsp->lockid = sp->stateid;
2474 	lsp->lockid.bits.type = LOCKID;
2475 	lsp->lockid.bits.chgseq = 0;
2476 	lsp->lockid.bits.pid = lo->pid;
2477 
2478 	lsp->locks_cleaned = FALSE;
2479 	lsp->lock_completed = FALSE;
2480 
2481 	rfs4_sw_init(&lsp->ls_sw);
2482 
2483 	/* Attached the supplied lock owner */
2484 	rfs4_dbe_hold(lo->dbe);
2485 	lsp->locker = lo;
2486 
2487 	lsp->lockownerlist.next = lsp->lockownerlist.prev =
2488 	    &lsp->lockownerlist;
2489 	lsp->lockownerlist.lsp = lsp;
2490 
2491 	rfs4_dbe_lock(sp->dbe);
2492 
2493 	insque(&lsp->lockownerlist, sp->lockownerlist.prev);
2494 
2495 	rfs4_dbe_hold(sp->dbe);
2496 
2497 	rfs4_dbe_unlock(sp->dbe);
2498 
2499 	return (TRUE);
2500 }
2501 
2502 void
2503 rfs4_lo_state_rele(rfs4_lo_state_t *lsp, bool_t unlock_fp)
2504 {
2505 	if (unlock_fp == TRUE)
2506 		rw_exit(&lsp->state->finfo->file_rwlock);
2507 	rfs4_dbe_rele(lsp->dbe);
2508 }
2509 
2510 static rfs4_lo_state_t *
2511 rfs4_findlo_state(stateid_t *id, bool_t lock_fp)
2512 {
2513 	rfs4_lo_state_t *lsp;
2514 	bool_t create = FALSE;
2515 
2516 	lsp = (rfs4_lo_state_t *)rfs4_dbsearch(rfs4_lo_state_idx, id,
2517 	    &create, NULL, RFS4_DBS_VALID);
2518 	if (lock_fp == TRUE && lsp != NULL)
2519 		rw_enter(&lsp->state->finfo->file_rwlock, RW_READER);
2520 
2521 	return (lsp);
2522 }
2523 
2524 
2525 static uint32_t
2526 lo_state_lo_hash(void *key)
2527 {
2528 	rfs4_lo_state_t *lop = key;
2529 
2530 	return (ADDRHASH(lop->locker) ^ ADDRHASH(lop->state));
2531 }
2532 
2533 static bool_t
2534 lo_state_lo_compare(rfs4_entry_t u_entry, void *key)
2535 {
2536 	rfs4_lo_state_t *lop = (rfs4_lo_state_t *)u_entry;
2537 	rfs4_lo_state_t *keyp = key;
2538 
2539 	return (keyp->locker == lop->locker && keyp->state == lop->state);
2540 }
2541 
2542 static void *
2543 lo_state_lo_mkkey(rfs4_entry_t u_entry)
2544 {
2545 	return (u_entry);
2546 }
2547 
2548 rfs4_lo_state_t *
2549 rfs4_findlo_state_by_owner(rfs4_lockowner_t *lo,
2550 			rfs4_state_t *sp, bool_t *create)
2551 {
2552 	rfs4_lo_state_t *lsp;
2553 	rfs4_lo_state_t arg;
2554 
2555 	arg.locker = lo;
2556 	arg.state = sp;
2557 
2558 	lsp = (rfs4_lo_state_t *)rfs4_dbsearch(rfs4_lo_state_owner_idx, &arg,
2559 	    create, &arg, RFS4_DBS_VALID);
2560 
2561 	return (lsp);
2562 }
2563 
2564 static stateid_t
2565 get_stateid(id_t eid)
2566 {
2567 	stateid_t id;
2568 
2569 	id.bits.boottime = rfs4_start_time;
2570 	id.bits.ident = eid;
2571 	id.bits.chgseq = 0;
2572 	id.bits.type = 0;
2573 	id.bits.pid = 0;
2574 
2575 	/*
2576 	 * If we are booted as a cluster node, embed our nodeid.
2577 	 * We've already done sanity checks in rfs4_client_create() so no
2578 	 * need to repeat them here.
2579 	 */
2580 	id.bits.clnodeid = (cluster_bootflags & CLUSTER_BOOTED) ?
2581 	    clconf_get_nodeid() : 0;
2582 
2583 	return (id);
2584 }
2585 
2586 /*
2587  * For use only when booted as a cluster node.
2588  * Returns TRUE if the embedded nodeid indicates that this stateid was
2589  * generated on another node.
2590  */
2591 static int
2592 foreign_stateid(stateid_t *id)
2593 {
2594 	ASSERT(cluster_bootflags & CLUSTER_BOOTED);
2595 	return (id->bits.clnodeid != (uint32_t)clconf_get_nodeid());
2596 }
2597 
2598 /*
2599  * For use only when booted as a cluster node.
2600  * Returns TRUE if the embedded nodeid indicates that this clientid was
2601  * generated on another node.
2602  */
2603 static int
2604 foreign_clientid(cid *cidp)
2605 {
2606 	ASSERT(cluster_bootflags & CLUSTER_BOOTED);
2607 	return (cidp->impl_id.c_id >> CLUSTER_NODEID_SHIFT !=
2608 	    (uint32_t)clconf_get_nodeid());
2609 }
2610 
2611 /*
2612  * For use only when booted as a cluster node.
2613  * Embed our cluster nodeid into the clientid.
2614  */
2615 static void
2616 embed_nodeid(cid *cidp)
2617 {
2618 	int clnodeid;
2619 	/*
2620 	 * Currently, our state tables are small enough that their
2621 	 * ids will leave enough bits free for the nodeid. If the
2622 	 * tables become larger, we mustn't overwrite the id.
2623 	 * Equally, we only have room for so many bits of nodeid, so
2624 	 * must check that too.
2625 	 */
2626 	ASSERT(cluster_bootflags & CLUSTER_BOOTED);
2627 	ASSERT(cidp->impl_id.c_id >> CLUSTER_NODEID_SHIFT == 0);
2628 	clnodeid = clconf_get_nodeid();
2629 	ASSERT(clnodeid <= CLUSTER_MAX_NODEID);
2630 	ASSERT(clnodeid != NODEID_UNKNOWN);
2631 	cidp->impl_id.c_id |= (clnodeid << CLUSTER_NODEID_SHIFT);
2632 }
2633 
2634 static uint32_t
2635 state_hash(void *key)
2636 {
2637 	stateid_t *ip = (stateid_t *)key;
2638 
2639 	return (ip->bits.ident);
2640 }
2641 
2642 static bool_t
2643 state_compare(rfs4_entry_t u_entry, void *key)
2644 {
2645 	rfs4_state_t *sp = (rfs4_state_t *)u_entry;
2646 	stateid_t *id = (stateid_t *)key;
2647 	bool_t rc;
2648 
2649 	rc = (sp->stateid.bits.boottime == id->bits.boottime &&
2650 	    sp->stateid.bits.ident == id->bits.ident);
2651 
2652 	return (rc);
2653 }
2654 
2655 static void *
2656 state_mkkey(rfs4_entry_t u_entry)
2657 {
2658 	rfs4_state_t *sp = (rfs4_state_t *)u_entry;
2659 
2660 	return (&sp->stateid);
2661 }
2662 
2663 static void
2664 rfs4_state_destroy(rfs4_entry_t u_entry)
2665 {
2666 	rfs4_state_t *sp = (rfs4_state_t *)u_entry;
2667 
2668 	ASSERT(&sp->lockownerlist == sp->lockownerlist.next);
2669 
2670 	/* release any share locks for this stateid if it's still open */
2671 	if (!sp->closed)
2672 		rfs4_unshare(sp);
2673 
2674 	/* Were done with the file */
2675 	rfs4_file_rele(sp->finfo);
2676 	sp->finfo = NULL;
2677 
2678 	/* And now with the openowner */
2679 	rfs4_dbe_lock(sp->owner->dbe);
2680 
2681 	remque(&sp->ownerstateids);
2682 	sp->ownerstateids.next = sp->ownerstateids.prev = &sp->ownerstateids;
2683 
2684 	rfs4_dbe_unlock(sp->owner->dbe);
2685 
2686 	rfs4_openowner_rele(sp->owner);
2687 	sp->owner = NULL;
2688 }
2689 
2690 static void
2691 rfs4_state_rele_nounlock(rfs4_state_t *sp)
2692 {
2693 	rfs4_dbe_rele(sp->dbe);
2694 }
2695 
2696 void
2697 rfs4_state_rele(rfs4_state_t *sp)
2698 {
2699 	rw_exit(&sp->finfo->file_rwlock);
2700 	rfs4_dbe_rele(sp->dbe);
2701 }
2702 
2703 static uint32_t
2704 deleg_hash(void *key)
2705 {
2706 	rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)key;
2707 
2708 	return (ADDRHASH(dsp->client) ^ ADDRHASH(dsp->finfo));
2709 }
2710 
2711 static bool_t
2712 deleg_compare(rfs4_entry_t u_entry, void *key)
2713 {
2714 	rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
2715 	rfs4_deleg_state_t *kdsp = (rfs4_deleg_state_t *)key;
2716 
2717 	return (dsp->client == kdsp->client && dsp->finfo == kdsp->finfo);
2718 }
2719 
2720 static void *
2721 deleg_mkkey(rfs4_entry_t u_entry)
2722 {
2723 	return (u_entry);
2724 }
2725 
2726 static uint32_t
2727 deleg_state_hash(void *key)
2728 {
2729 	stateid_t *ip = (stateid_t *)key;
2730 
2731 	return (ip->bits.ident);
2732 }
2733 
2734 static bool_t
2735 deleg_state_compare(rfs4_entry_t u_entry, void *key)
2736 {
2737 	rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
2738 	stateid_t *id = (stateid_t *)key;
2739 	bool_t rc;
2740 
2741 	if (id->bits.type != DELEGID)
2742 		return (FALSE);
2743 
2744 	rc = (dsp->delegid.bits.boottime == id->bits.boottime &&
2745 	    dsp->delegid.bits.ident == id->bits.ident);
2746 
2747 	return (rc);
2748 }
2749 
2750 static void *
2751 deleg_state_mkkey(rfs4_entry_t u_entry)
2752 {
2753 	rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
2754 
2755 	return (&dsp->delegid);
2756 }
2757 
2758 static bool_t
2759 rfs4_deleg_state_expiry(rfs4_entry_t u_entry)
2760 {
2761 	rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
2762 
2763 	if (rfs4_dbe_is_invalid(dsp->dbe))
2764 		return (TRUE);
2765 
2766 	if ((gethrestime_sec() - dsp->client->last_access
2767 	    > rfs4_lease_time)) {
2768 		rfs4_dbe_invalidate(dsp->dbe);
2769 		return (TRUE);
2770 	}
2771 
2772 	return (FALSE);
2773 }
2774 
2775 static bool_t
2776 rfs4_deleg_state_create(rfs4_entry_t u_entry, void *argp)
2777 {
2778 	rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
2779 	rfs4_file_t *fp = ((rfs4_deleg_state_t *)argp)->finfo;
2780 	rfs4_client_t *cp = ((rfs4_deleg_state_t *)argp)->client;
2781 
2782 	rfs4_dbe_hold(fp->dbe);
2783 	rfs4_dbe_hold(cp->dbe);
2784 
2785 	dsp->delegid = get_stateid(rfs4_dbe_getid(dsp->dbe));
2786 	dsp->delegid.bits.type = DELEGID;
2787 	dsp->finfo = fp;
2788 	dsp->client = cp;
2789 	dsp->dtype = OPEN_DELEGATE_NONE;
2790 
2791 	dsp->time_granted = gethrestime_sec();	/* observability */
2792 	dsp->time_revoked = 0;
2793 
2794 	/* Init lists for remque/insque */
2795 	dsp->delegationlist.next = dsp->delegationlist.prev =
2796 	    &dsp->delegationlist;
2797 	dsp->delegationlist.dsp = dsp;
2798 
2799 	dsp->clientdeleglist.next = dsp->clientdeleglist.prev =
2800 	    &dsp->clientdeleglist;
2801 	dsp->clientdeleglist.dsp = dsp;
2802 
2803 	/* Insert state on per open owner's list */
2804 	rfs4_dbe_lock(cp->dbe);
2805 
2806 	insque(&dsp->clientdeleglist, cp->clientdeleglist.prev);
2807 
2808 	rfs4_dbe_unlock(cp->dbe);
2809 
2810 	return (TRUE);
2811 }
2812 
2813 static void
2814 rfs4_deleg_state_destroy(rfs4_entry_t u_entry)
2815 {
2816 	rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
2817 
2818 	if (&dsp->delegationlist != dsp->delegationlist.next)
2819 		rfs4_return_deleg(dsp, FALSE);
2820 
2821 	/* Were done with the file */
2822 	rfs4_file_rele(dsp->finfo);
2823 	dsp->finfo = NULL;
2824 
2825 	/* And now with the openowner */
2826 	rfs4_dbe_lock(dsp->client->dbe);
2827 
2828 	remque(&dsp->clientdeleglist);
2829 	dsp->clientdeleglist.next = dsp->clientdeleglist.prev =
2830 	    &dsp->clientdeleglist;
2831 
2832 	rfs4_dbe_unlock(dsp->client->dbe);
2833 
2834 	rfs4_client_rele(dsp->client);
2835 	dsp->client = NULL;
2836 }
2837 
2838 rfs4_deleg_state_t *
2839 rfs4_finddeleg(rfs4_state_t *sp, bool_t *create)
2840 {
2841 	rfs4_deleg_state_t ds, *dsp;
2842 
2843 	ds.client = sp->owner->client;
2844 	ds.finfo = sp->finfo;
2845 
2846 	dsp = (rfs4_deleg_state_t *)rfs4_dbsearch(rfs4_deleg_idx, &ds,
2847 	    create, &ds, RFS4_DBS_VALID);
2848 
2849 	return (dsp);
2850 }
2851 
2852 rfs4_deleg_state_t *
2853 rfs4_finddelegstate(stateid_t *id)
2854 {
2855 	rfs4_deleg_state_t *dsp;
2856 	bool_t create = FALSE;
2857 
2858 	dsp = (rfs4_deleg_state_t *)rfs4_dbsearch(rfs4_deleg_state_idx, id,
2859 	    &create, NULL, RFS4_DBS_VALID);
2860 
2861 	return (dsp);
2862 }
2863 
2864 void
2865 rfs4_deleg_state_rele(rfs4_deleg_state_t *dsp)
2866 {
2867 	rfs4_dbe_rele(dsp->dbe);
2868 }
2869 
2870 void
2871 rfs4_update_lock_sequence(rfs4_lo_state_t *lsp)
2872 {
2873 
2874 	rfs4_dbe_lock(lsp->dbe);
2875 
2876 	/*
2877 	 * If we are skipping sequence id checking, this means that
2878 	 * this is the first lock request and therefore the sequence
2879 	 * id does not need to be updated.  This only happens on the
2880 	 * first lock request for a lockowner
2881 	 */
2882 	if (!lsp->skip_seqid_check)
2883 		lsp->seqid++;
2884 
2885 	rfs4_dbe_unlock(lsp->dbe);
2886 }
2887 
2888 void
2889 rfs4_update_lock_resp(rfs4_lo_state_t *lsp, nfs_resop4 *resp)
2890 {
2891 
2892 	rfs4_dbe_lock(lsp->dbe);
2893 
2894 	rfs4_free_reply(lsp->reply);
2895 
2896 	rfs4_copy_reply(lsp->reply, resp);
2897 
2898 	rfs4_dbe_unlock(lsp->dbe);
2899 }
2900 
2901 void
2902 rfs4_free_opens(rfs4_openowner_t *op, bool_t invalidate,
2903 	bool_t close_of_client)
2904 {
2905 	rfs4_state_t *sp;
2906 
2907 	rfs4_dbe_lock(op->dbe);
2908 
2909 	for (sp = op->ownerstateids.next->sp; sp != NULL;
2910 	    sp = sp->ownerstateids.next->sp) {
2911 		rfs4_state_close(sp, FALSE, close_of_client, CRED());
2912 		if (invalidate == TRUE)
2913 			rfs4_dbe_invalidate(sp->dbe);
2914 	}
2915 
2916 	rfs4_dbe_unlock(op->dbe);
2917 	rfs4_dbe_invalidate(op->dbe);
2918 }
2919 
2920 static uint32_t
2921 state_owner_file_hash(void *key)
2922 {
2923 	rfs4_state_t *sp = key;
2924 
2925 	return (ADDRHASH(sp->owner) ^ ADDRHASH(sp->finfo));
2926 }
2927 
2928 static bool_t
2929 state_owner_file_compare(rfs4_entry_t u_entry, void *key)
2930 {
2931 	rfs4_state_t *sp = (rfs4_state_t *)u_entry;
2932 	rfs4_state_t *arg = key;
2933 
2934 	if (sp->closed == TRUE)
2935 		return (FALSE);
2936 
2937 	return (arg->owner == sp->owner && arg->finfo == sp->finfo);
2938 }
2939 
2940 static void *
2941 state_owner_file_mkkey(rfs4_entry_t u_entry)
2942 {
2943 	return (u_entry);
2944 }
2945 
2946 static uint32_t
2947 state_file_hash(void *key)
2948 {
2949 	return (ADDRHASH(key));
2950 }
2951 
2952 static bool_t
2953 state_file_compare(rfs4_entry_t u_entry, void *key)
2954 {
2955 	rfs4_state_t *sp = (rfs4_state_t *)u_entry;
2956 	rfs4_file_t *fp = key;
2957 
2958 	if (sp->closed == TRUE)
2959 		return (FALSE);
2960 
2961 	return (fp == sp->finfo);
2962 }
2963 
2964 static void *
2965 state_file_mkkey(rfs4_entry_t u_entry)
2966 {
2967 	rfs4_state_t *sp = (rfs4_state_t *)u_entry;
2968 
2969 	return (sp->finfo);
2970 }
2971 
2972 rfs4_state_t *
2973 rfs4_findstate_by_owner_file(rfs4_openowner_t *op, rfs4_file_t *file,
2974 	bool_t *create)
2975 {
2976 	rfs4_state_t *sp;
2977 	rfs4_state_t key;
2978 
2979 	key.owner = op;
2980 	key.finfo = file;
2981 
2982 	sp = (rfs4_state_t *)rfs4_dbsearch(rfs4_state_owner_file_idx, &key,
2983 	    create, &key, RFS4_DBS_VALID);
2984 
2985 	return (sp);
2986 }
2987 
2988 /* This returns ANY state struct that refers to this file */
2989 static rfs4_state_t *
2990 rfs4_findstate_by_file(rfs4_file_t *fp)
2991 {
2992 	bool_t create = FALSE;
2993 
2994 	return ((rfs4_state_t *)rfs4_dbsearch(rfs4_state_file_idx, fp,
2995 	    &create, fp, RFS4_DBS_VALID));
2996 }
2997 
2998 static bool_t
2999 rfs4_state_expiry(rfs4_entry_t u_entry)
3000 {
3001 	rfs4_state_t *sp = (rfs4_state_t *)u_entry;
3002 
3003 	if (rfs4_dbe_is_invalid(sp->dbe))
3004 		return (TRUE);
3005 
3006 	if (sp->closed == TRUE &&
3007 	    ((gethrestime_sec() - rfs4_dbe_get_timerele(sp->dbe))
3008 	    > rfs4_lease_time))
3009 		return (TRUE);
3010 
3011 	return ((gethrestime_sec() - sp->owner->client->last_access
3012 	    > rfs4_lease_time));
3013 }
3014 
3015 static bool_t
3016 rfs4_state_create(rfs4_entry_t u_entry, void *argp)
3017 {
3018 	rfs4_state_t *sp = (rfs4_state_t *)u_entry;
3019 	rfs4_file_t *fp = ((rfs4_state_t *)argp)->finfo;
3020 	rfs4_openowner_t *op = ((rfs4_state_t *)argp)->owner;
3021 
3022 	rfs4_dbe_hold(fp->dbe);
3023 	rfs4_dbe_hold(op->dbe);
3024 	sp->stateid = get_stateid(rfs4_dbe_getid(sp->dbe));
3025 	sp->stateid.bits.type = OPENID;
3026 	sp->owner = op;
3027 	sp->finfo = fp;
3028 
3029 	/* Init lists for remque/insque */
3030 	sp->ownerstateids.next = sp->ownerstateids.prev = &sp->ownerstateids;
3031 	sp->ownerstateids.sp = sp;
3032 	sp->lockownerlist.next = sp->lockownerlist.prev = &sp->lockownerlist;
3033 	sp->lockownerlist.lsp = NULL;
3034 
3035 	/* Insert state on per open owner's list */
3036 	rfs4_dbe_lock(op->dbe);
3037 
3038 	insque(&sp->ownerstateids, op->ownerstateids.prev);
3039 
3040 	rfs4_dbe_unlock(op->dbe);
3041 
3042 	return (TRUE);
3043 }
3044 
3045 static rfs4_state_t *
3046 rfs4_findstate(stateid_t *id, rfs4_dbsearch_type_t find_invalid,
3047 		bool_t lock_fp)
3048 {
3049 	rfs4_state_t *sp;
3050 	bool_t create = FALSE;
3051 
3052 	sp = (rfs4_state_t *)rfs4_dbsearch(rfs4_state_idx, id,
3053 	    &create, NULL, find_invalid);
3054 	if (lock_fp == TRUE && sp != NULL)
3055 		rw_enter(&sp->finfo->file_rwlock, RW_READER);
3056 
3057 	return (sp);
3058 }
3059 
3060 void
3061 rfs4_state_close(rfs4_state_t *sp, bool_t lock_held,
3062 			bool_t close_of_client, cred_t *cr)
3063 {
3064 	/* Remove the associated lo_state owners */
3065 	if (!lock_held)
3066 		rfs4_dbe_lock(sp->dbe);
3067 
3068 	/*
3069 	 * If refcnt == 0, the dbe is about to be destroyed.
3070 	 * lock state will be released by the reaper thread.
3071 	 */
3072 
3073 	if (rfs4_dbe_refcnt(sp->dbe) > 0) {
3074 		if (sp->closed == FALSE) {
3075 			sp->closed = TRUE;
3076 
3077 			rfs4_release_share_lock_state(sp, cr, close_of_client);
3078 		}
3079 	}
3080 
3081 	if (!lock_held)
3082 		rfs4_dbe_unlock(sp->dbe);
3083 }
3084 
3085 /*
3086  * Remove all state associated with the given client.
3087  */
3088 void
3089 rfs4_client_state_remove(rfs4_client_t *cp)
3090 {
3091 	rfs4_openowner_t *oop;
3092 
3093 	rfs4_dbe_lock(cp->dbe);
3094 
3095 	for (oop = cp->openownerlist.next->oop;  oop != NULL;
3096 	    oop = oop->openownerlist.next->oop) {
3097 		rfs4_free_opens(oop, TRUE, TRUE);
3098 	}
3099 
3100 	rfs4_dbe_unlock(cp->dbe);
3101 }
3102 
3103 void
3104 rfs4_client_close(rfs4_client_t *cp)
3105 {
3106 	/* Mark client as going away. */
3107 	rfs4_dbe_lock(cp->dbe);
3108 	rfs4_dbe_invalidate(cp->dbe);
3109 	rfs4_dbe_unlock(cp->dbe);
3110 
3111 	rfs4_client_state_remove(cp);
3112 
3113 	/* Release the client */
3114 	rfs4_client_rele(cp);
3115 }
3116 
3117 nfsstat4
3118 rfs4_check_clientid(clientid4 *cp, int setclid_confirm)
3119 {
3120 	cid *cidp = (cid *) cp;
3121 
3122 	/*
3123 	 * If we are booted as a cluster node, check the embedded nodeid.
3124 	 * If it indicates that this clientid was generated on another node,
3125 	 * inform the client accordingly.
3126 	 */
3127 	if (cluster_bootflags & CLUSTER_BOOTED && foreign_clientid(cidp))
3128 		return (NFS4ERR_STALE_CLIENTID);
3129 
3130 	/*
3131 	 * If the server start time matches the time provided
3132 	 * by the client (via the clientid) and this is NOT a
3133 	 * setclientid_confirm then return EXPIRED.
3134 	 */
3135 	if (!setclid_confirm && cidp->impl_id.start_time == rfs4_start_time)
3136 		return (NFS4ERR_EXPIRED);
3137 
3138 	return (NFS4ERR_STALE_CLIENTID);
3139 }
3140 
3141 /*
3142  * This is used when a stateid has not been found amongst the
3143  * current server's state.  Check the stateid to see if it
3144  * was from this server instantiation or not.
3145  */
3146 static nfsstat4
3147 what_stateid_error(stateid_t *id, stateid_type_t type)
3148 {
3149 	/* If we are booted as a cluster node, was stateid locally generated? */
3150 	if ((cluster_bootflags & CLUSTER_BOOTED) && foreign_stateid(id))
3151 		return (NFS4ERR_STALE_STATEID);
3152 
3153 	/* If types don't match then no use checking further */
3154 	if (type != id->bits.type)
3155 		return (NFS4ERR_BAD_STATEID);
3156 
3157 	/* From a previous server instantiation, return STALE */
3158 	if (id->bits.boottime < rfs4_start_time)
3159 		return (NFS4ERR_STALE_STATEID);
3160 
3161 	/*
3162 	 * From this server but the state is most likely beyond lease
3163 	 * timeout: return NFS4ERR_EXPIRED.  However, there is the
3164 	 * case of a delegation stateid.  For delegations, there is a
3165 	 * case where the state can be removed without the client's
3166 	 * knowledge/consent: revocation.  In the case of delegation
3167 	 * revocation, the delegation state will be removed and will
3168 	 * not be found.  If the client does something like a
3169 	 * DELEGRETURN or even a READ/WRITE with a delegatoin stateid
3170 	 * that has been revoked, the server should return BAD_STATEID
3171 	 * instead of the more common EXPIRED error.
3172 	 */
3173 	if (id->bits.boottime == rfs4_start_time) {
3174 		if (type == DELEGID)
3175 			return (NFS4ERR_BAD_STATEID);
3176 		else
3177 			return (NFS4ERR_EXPIRED);
3178 	}
3179 
3180 	return (NFS4ERR_BAD_STATEID);
3181 }
3182 
3183 /*
3184  * Used later on to find the various state structs.  When called from
3185  * rfs4_check_stateid()->rfs4_get_all_state(), no file struct lock is
3186  * taken (it is not needed) and helps on the read/write path with
3187  * respect to performance.
3188  */
3189 static nfsstat4
3190 rfs4_get_state_lockit(stateid4 *stateid, rfs4_state_t **spp,
3191 		rfs4_dbsearch_type_t find_invalid, bool_t lock_fp)
3192 {
3193 	stateid_t *id = (stateid_t *)stateid;
3194 	rfs4_state_t *sp;
3195 
3196 	*spp = NULL;
3197 
3198 	/* If we are booted as a cluster node, was stateid locally generated? */
3199 	if ((cluster_bootflags & CLUSTER_BOOTED) && foreign_stateid(id))
3200 		return (NFS4ERR_STALE_STATEID);
3201 
3202 	sp = rfs4_findstate(id, find_invalid, lock_fp);
3203 	if (sp == NULL) {
3204 		return (what_stateid_error(id, OPENID));
3205 	}
3206 
3207 	if (rfs4_lease_expired(sp->owner->client)) {
3208 		if (lock_fp == TRUE)
3209 			rfs4_state_rele(sp);
3210 		else
3211 			rfs4_state_rele_nounlock(sp);
3212 		return (NFS4ERR_EXPIRED);
3213 	}
3214 
3215 	*spp = sp;
3216 
3217 	return (NFS4_OK);
3218 }
3219 
3220 nfsstat4
3221 rfs4_get_state(stateid4 *stateid, rfs4_state_t **spp,
3222 		rfs4_dbsearch_type_t find_invalid)
3223 {
3224 	return (rfs4_get_state_lockit(stateid, spp, find_invalid, TRUE));
3225 }
3226 
3227 int
3228 rfs4_check_stateid_seqid(rfs4_state_t *sp, stateid4 *stateid)
3229 {
3230 	stateid_t *id = (stateid_t *)stateid;
3231 
3232 	if (rfs4_lease_expired(sp->owner->client))
3233 		return (NFS4_CHECK_STATEID_EXPIRED);
3234 
3235 	/* Stateid is some time in the future - that's bad */
3236 	if (sp->stateid.bits.chgseq < id->bits.chgseq)
3237 		return (NFS4_CHECK_STATEID_BAD);
3238 
3239 	if (sp->stateid.bits.chgseq == id->bits.chgseq + 1)
3240 		return (NFS4_CHECK_STATEID_REPLAY);
3241 
3242 	/* Stateid is some time in the past - that's old */
3243 	if (sp->stateid.bits.chgseq > id->bits.chgseq)
3244 		return (NFS4_CHECK_STATEID_OLD);
3245 
3246 	/* Caller needs to know about confirmation before closure */
3247 	if (sp->owner->need_confirm)
3248 		return (NFS4_CHECK_STATEID_UNCONFIRMED);
3249 
3250 	if (sp->closed == TRUE)
3251 		return (NFS4_CHECK_STATEID_CLOSED);
3252 
3253 	return (NFS4_CHECK_STATEID_OKAY);
3254 }
3255 
3256 int
3257 rfs4_check_lo_stateid_seqid(rfs4_lo_state_t *lsp, stateid4 *stateid)
3258 {
3259 	stateid_t *id = (stateid_t *)stateid;
3260 
3261 	if (rfs4_lease_expired(lsp->state->owner->client))
3262 		return (NFS4_CHECK_STATEID_EXPIRED);
3263 
3264 	/* Stateid is some time in the future - that's bad */
3265 	if (lsp->lockid.bits.chgseq < id->bits.chgseq)
3266 		return (NFS4_CHECK_STATEID_BAD);
3267 
3268 	if (lsp->lockid.bits.chgseq == id->bits.chgseq + 1)
3269 		return (NFS4_CHECK_STATEID_REPLAY);
3270 
3271 	/* Stateid is some time in the past - that's old */
3272 	if (lsp->lockid.bits.chgseq > id->bits.chgseq)
3273 		return (NFS4_CHECK_STATEID_OLD);
3274 
3275 	return (NFS4_CHECK_STATEID_OKAY);
3276 }
3277 
3278 nfsstat4
3279 rfs4_get_deleg_state(stateid4 *stateid, rfs4_deleg_state_t **dspp)
3280 {
3281 	stateid_t *id = (stateid_t *)stateid;
3282 	rfs4_deleg_state_t *dsp;
3283 
3284 	*dspp = NULL;
3285 
3286 	/* If we are booted as a cluster node, was stateid locally generated? */
3287 	if ((cluster_bootflags & CLUSTER_BOOTED) && foreign_stateid(id))
3288 		return (NFS4ERR_STALE_STATEID);
3289 
3290 	dsp = rfs4_finddelegstate(id);
3291 	if (dsp == NULL) {
3292 		return (what_stateid_error(id, DELEGID));
3293 	}
3294 
3295 	if (rfs4_lease_expired(dsp->client)) {
3296 		rfs4_deleg_state_rele(dsp);
3297 		return (NFS4ERR_EXPIRED);
3298 	}
3299 
3300 	*dspp = dsp;
3301 
3302 	return (NFS4_OK);
3303 }
3304 
3305 nfsstat4
3306 rfs4_get_lo_state(stateid4 *stateid, rfs4_lo_state_t **lspp, bool_t lock_fp)
3307 {
3308 	stateid_t *id = (stateid_t *)stateid;
3309 	rfs4_lo_state_t *lsp;
3310 
3311 	*lspp = NULL;
3312 
3313 	/* If we are booted as a cluster node, was stateid locally generated? */
3314 	if ((cluster_bootflags & CLUSTER_BOOTED) && foreign_stateid(id))
3315 		return (NFS4ERR_STALE_STATEID);
3316 
3317 	lsp = rfs4_findlo_state(id, lock_fp);
3318 	if (lsp == NULL) {
3319 		return (what_stateid_error(id, LOCKID));
3320 	}
3321 
3322 	if (rfs4_lease_expired(lsp->state->owner->client)) {
3323 		rfs4_lo_state_rele(lsp, lock_fp);
3324 		return (NFS4ERR_EXPIRED);
3325 	}
3326 
3327 	*lspp = lsp;
3328 
3329 	return (NFS4_OK);
3330 }
3331 
3332 static nfsstat4
3333 rfs4_get_all_state(stateid4 *sid, rfs4_state_t **spp,
3334 	rfs4_deleg_state_t **dspp, rfs4_lo_state_t **lospp)
3335 {
3336 	rfs4_state_t *sp = NULL;
3337 	rfs4_deleg_state_t *dsp = NULL;
3338 	rfs4_lo_state_t *losp = NULL;
3339 	stateid_t *id;
3340 	nfsstat4 status;
3341 
3342 	*spp = NULL; *dspp = NULL; *lospp = NULL;
3343 
3344 	id = (stateid_t *)sid;
3345 	switch (id->bits.type) {
3346 	case OPENID:
3347 		status = rfs4_get_state_lockit(sid, &sp, FALSE, FALSE);
3348 		break;
3349 	case DELEGID:
3350 		status = rfs4_get_deleg_state(sid, &dsp);
3351 		break;
3352 	case LOCKID:
3353 		status = rfs4_get_lo_state(sid, &losp, FALSE);
3354 		if (status == NFS4_OK) {
3355 			sp = losp->state;
3356 			rfs4_dbe_hold(sp->dbe);
3357 		}
3358 		break;
3359 	default:
3360 		status = NFS4ERR_BAD_STATEID;
3361 	}
3362 
3363 	if (status == NFS4_OK) {
3364 		*spp = sp;
3365 		*dspp = dsp;
3366 		*lospp = losp;
3367 	}
3368 
3369 	return (status);
3370 }
3371 
3372 /*
3373  * Given the I/O mode (FREAD or FWRITE), this checks whether the
3374  * rfs4_state_t struct has access to do this operation and if so
3375  * return NFS4_OK; otherwise the proper NFSv4 error is returned.
3376  */
3377 nfsstat4
3378 rfs4_state_has_access(rfs4_state_t *sp, int mode, vnode_t *vp)
3379 {
3380 	nfsstat4 stat = NFS4_OK;
3381 	rfs4_file_t *fp;
3382 	bool_t create = FALSE;
3383 
3384 	rfs4_dbe_lock(sp->dbe);
3385 	if (mode == FWRITE) {
3386 		if (!(sp->share_access & OPEN4_SHARE_ACCESS_WRITE)) {
3387 			stat = NFS4ERR_OPENMODE;
3388 		}
3389 	} else if (mode == FREAD) {
3390 		if (!(sp->share_access & OPEN4_SHARE_ACCESS_READ)) {
3391 			/*
3392 			 * If we have OPENed the file with DENYing access
3393 			 * to both READ and WRITE then no one else could
3394 			 * have OPENed the file, hence no conflicting READ
3395 			 * deny.  This check is merely an optimization.
3396 			 */
3397 			if (sp->share_deny == OPEN4_SHARE_DENY_BOTH)
3398 				goto out;
3399 
3400 			/* Check against file struct's DENY mode */
3401 			fp = rfs4_findfile(vp, NULL, &create);
3402 			if (fp != NULL) {
3403 				int deny_read = 0;
3404 				rfs4_dbe_lock(fp->dbe);
3405 				/*
3406 				 * Check if any other open owner has the file
3407 				 * OPENed with deny READ.
3408 				 */
3409 				if (sp->share_deny & OPEN4_SHARE_DENY_READ)
3410 					deny_read = 1;
3411 				ASSERT(fp->deny_read - deny_read >= 0);
3412 				if (fp->deny_read - deny_read > 0)
3413 					stat = NFS4ERR_OPENMODE;
3414 				rfs4_dbe_unlock(fp->dbe);
3415 				rfs4_file_rele(fp);
3416 			}
3417 		}
3418 	} else {
3419 		/* Illegal I/O mode */
3420 		stat = NFS4ERR_INVAL;
3421 	}
3422 out:
3423 	rfs4_dbe_unlock(sp->dbe);
3424 	return (stat);
3425 }
3426 
3427 /*
3428  * Given the I/O mode (FREAD or FWRITE), the vnode, the stateid and whether
3429  * the file is being truncated, return NFS4_OK if allowed or appropriate
3430  * V4 error if not. Note NFS4ERR_DELAY will be returned and a recall on
3431  * the associated file will be done if the I/O is not consistent with any
3432  * delegation in effect on the file. Should be holding VOP_RWLOCK, either
3433  * as reader or writer as appropriate. rfs4_op_open will acquire the
3434  * VOP_RWLOCK as writer when setting up delegation. If the stateid is bad
3435  * this routine will return NFS4ERR_BAD_STATEID. In addition, through the
3436  * deleg parameter, we will return whether a write delegation is held by
3437  * the client associated with this stateid.
3438  * If the server instance associated with the relevant client is in its
3439  * grace period, return NFS4ERR_GRACE.
3440  */
3441 
3442 nfsstat4
3443 rfs4_check_stateid(int mode, vnode_t *vp,
3444 		stateid4 *stateid, bool_t trunc, bool_t *deleg,
3445 		bool_t do_access, caller_context_t *ct)
3446 {
3447 	rfs4_file_t *fp;
3448 	bool_t create = FALSE;
3449 	rfs4_state_t *sp;
3450 	rfs4_deleg_state_t *dsp;
3451 	rfs4_lo_state_t *lsp;
3452 	stateid_t *id = (stateid_t *)stateid;
3453 	nfsstat4 stat = NFS4_OK;
3454 
3455 	if (ct != NULL) {
3456 		ct->cc_sysid = 0;
3457 		ct->cc_pid = 0;
3458 		ct->cc_caller_id = nfs4_srv_caller_id;
3459 	}
3460 
3461 	if (ISSPECIAL(stateid)) {
3462 		fp = rfs4_findfile(vp, NULL, &create);
3463 		if (fp == NULL)
3464 			return (NFS4_OK);
3465 		if (fp->dinfo->dtype == OPEN_DELEGATE_NONE) {
3466 			rfs4_file_rele(fp);
3467 			return (NFS4_OK);
3468 		}
3469 		if (mode == FWRITE ||
3470 		    fp->dinfo->dtype == OPEN_DELEGATE_WRITE) {
3471 			rfs4_recall_deleg(fp, trunc, NULL);
3472 			rfs4_file_rele(fp);
3473 			return (NFS4ERR_DELAY);
3474 		}
3475 		rfs4_file_rele(fp);
3476 		return (NFS4_OK);
3477 	} else {
3478 		stat = rfs4_get_all_state(stateid, &sp, &dsp, &lsp);
3479 		if (stat != NFS4_OK)
3480 			return (stat);
3481 		if (lsp != NULL) {
3482 			/* Is associated server instance in its grace period? */
3483 			if (rfs4_clnt_in_grace(lsp->locker->client)) {
3484 				rfs4_lo_state_rele(lsp, FALSE);
3485 				if (sp != NULL)
3486 					rfs4_state_rele_nounlock(sp);
3487 				return (NFS4ERR_GRACE);
3488 			}
3489 			if (id->bits.type == LOCKID) {
3490 				/* Seqid in the future? - that's bad */
3491 				if (lsp->lockid.bits.chgseq <
3492 				    id->bits.chgseq) {
3493 					rfs4_lo_state_rele(lsp, FALSE);
3494 					if (sp != NULL)
3495 						rfs4_state_rele_nounlock(sp);
3496 					return (NFS4ERR_BAD_STATEID);
3497 				}
3498 				/* Seqid in the past? - that's old */
3499 				if (lsp->lockid.bits.chgseq >
3500 				    id->bits.chgseq) {
3501 					rfs4_lo_state_rele(lsp, FALSE);
3502 					if (sp != NULL)
3503 						rfs4_state_rele_nounlock(sp);
3504 					return (NFS4ERR_OLD_STATEID);
3505 				}
3506 				/* Ensure specified filehandle matches */
3507 				if (lsp->state->finfo->vp != vp) {
3508 					rfs4_lo_state_rele(lsp, FALSE);
3509 					if (sp != NULL)
3510 						rfs4_state_rele_nounlock(sp);
3511 					return (NFS4ERR_BAD_STATEID);
3512 				}
3513 			}
3514 			if (ct != NULL) {
3515 				ct->cc_sysid = lsp->locker->client->sysidt;
3516 				ct->cc_pid = lsp->locker->pid;
3517 			}
3518 			rfs4_lo_state_rele(lsp, FALSE);
3519 		}
3520 
3521 		/* Stateid provided was an "open" stateid */
3522 		if (sp != NULL) {
3523 			/* Is associated server instance in its grace period? */
3524 			if (rfs4_clnt_in_grace(sp->owner->client)) {
3525 				rfs4_state_rele_nounlock(sp);
3526 				return (NFS4ERR_GRACE);
3527 			}
3528 			if (id->bits.type == OPENID) {
3529 				/* Seqid in the future? - that's bad */
3530 				if (sp->stateid.bits.chgseq <
3531 				    id->bits.chgseq) {
3532 					rfs4_state_rele_nounlock(sp);
3533 					return (NFS4ERR_BAD_STATEID);
3534 				}
3535 				/* Seqid in the past - that's old */
3536 				if (sp->stateid.bits.chgseq >
3537 				    id->bits.chgseq) {
3538 					rfs4_state_rele_nounlock(sp);
3539 					return (NFS4ERR_OLD_STATEID);
3540 				}
3541 			}
3542 			/* Ensure specified filehandle matches */
3543 			if (sp->finfo->vp != vp) {
3544 				rfs4_state_rele_nounlock(sp);
3545 				return (NFS4ERR_BAD_STATEID);
3546 			}
3547 
3548 			if (sp->owner->need_confirm) {
3549 				rfs4_state_rele_nounlock(sp);
3550 				return (NFS4ERR_BAD_STATEID);
3551 			}
3552 
3553 			if (sp->closed == TRUE) {
3554 				rfs4_state_rele_nounlock(sp);
3555 				return (NFS4ERR_OLD_STATEID);
3556 			}
3557 
3558 			if (do_access)
3559 				stat = rfs4_state_has_access(sp, mode, vp);
3560 			else
3561 				stat = NFS4_OK;
3562 
3563 			/*
3564 			 * Return whether this state has write
3565 			 * delegation if desired
3566 			 */
3567 			if (deleg &&
3568 			    (sp->finfo->dinfo->dtype == OPEN_DELEGATE_WRITE))
3569 				*deleg = TRUE;
3570 
3571 			/*
3572 			 * We got a valid stateid, so we update the
3573 			 * lease on the client. Ideally we would like
3574 			 * to do this after the calling op succeeds,
3575 			 * but for now this will be good
3576 			 * enough. Callers of this routine are
3577 			 * currently insulated from the state stuff.
3578 			 */
3579 			rfs4_update_lease(sp->owner->client);
3580 
3581 			/*
3582 			 * If a delegation is present on this file and
3583 			 * this is a WRITE, then update the lastwrite
3584 			 * time to indicate that activity is present.
3585 			 */
3586 			if (sp->finfo->dinfo->dtype == OPEN_DELEGATE_WRITE &&
3587 			    mode == FWRITE) {
3588 				sp->finfo->dinfo->time_lastwrite =
3589 				    gethrestime_sec();
3590 			}
3591 
3592 			rfs4_state_rele_nounlock(sp);
3593 
3594 			return (stat);
3595 		}
3596 
3597 		if (dsp != NULL) {
3598 			/* Is associated server instance in its grace period? */
3599 			if (rfs4_clnt_in_grace(dsp->client)) {
3600 				rfs4_deleg_state_rele(dsp);
3601 				return (NFS4ERR_GRACE);
3602 			}
3603 			if (dsp->delegid.bits.chgseq !=	id->bits.chgseq) {
3604 				rfs4_deleg_state_rele(dsp);
3605 				return (NFS4ERR_BAD_STATEID);
3606 			}
3607 
3608 			/* Ensure specified filehandle matches */
3609 			if (dsp->finfo->vp != vp) {
3610 				rfs4_deleg_state_rele(dsp);
3611 				return (NFS4ERR_BAD_STATEID);
3612 			}
3613 			/*
3614 			 * Return whether this state has write
3615 			 * delegation if desired
3616 			 */
3617 			if (deleg &&
3618 			    (dsp->finfo->dinfo->dtype == OPEN_DELEGATE_WRITE))
3619 				*deleg = TRUE;
3620 
3621 			rfs4_update_lease(dsp->client);
3622 
3623 			/*
3624 			 * If a delegation is present on this file and
3625 			 * this is a WRITE, then update the lastwrite
3626 			 * time to indicate that activity is present.
3627 			 */
3628 			if (dsp->finfo->dinfo->dtype == OPEN_DELEGATE_WRITE &&
3629 			    mode == FWRITE) {
3630 				dsp->finfo->dinfo->time_lastwrite =
3631 				    gethrestime_sec();
3632 			}
3633 
3634 			/*
3635 			 * XXX - what happens if this is a WRITE and the
3636 			 * delegation type of for READ.
3637 			 */
3638 			rfs4_deleg_state_rele(dsp);
3639 
3640 			return (stat);
3641 		}
3642 		/*
3643 		 * If we got this far, something bad happened
3644 		 */
3645 		return (NFS4ERR_BAD_STATEID);
3646 	}
3647 }
3648 
3649 
3650 /*
3651  * This is a special function in that for the file struct provided the
3652  * server wants to remove/close all current state associated with the
3653  * file.  The prime use of this would be with OP_REMOVE to force the
3654  * release of state and particularly of file locks.
3655  *
3656  * There is an assumption that there is no delegations outstanding on
3657  * this file at this point.  The caller should have waited for those
3658  * to be returned or revoked.
3659  */
3660 void
3661 rfs4_close_all_state(rfs4_file_t *fp)
3662 {
3663 	rfs4_state_t *sp;
3664 
3665 	rfs4_dbe_lock(fp->dbe);
3666 
3667 #ifdef DEBUG
3668 	/* only applies when server is handing out delegations */
3669 	if (rfs4_deleg_policy != SRV_NEVER_DELEGATE)
3670 		ASSERT(fp->dinfo->hold_grant > 0);
3671 #endif
3672 
3673 	/* No delegations for this file */
3674 	ASSERT(fp->delegationlist.next == &fp->delegationlist);
3675 
3676 	/* Make sure that it can not be found */
3677 	rfs4_dbe_invalidate(fp->dbe);
3678 
3679 	if (fp->vp == NULL) {
3680 		rfs4_dbe_unlock(fp->dbe);
3681 		return;
3682 	}
3683 	rfs4_dbe_unlock(fp->dbe);
3684 
3685 	/*
3686 	 * Hold as writer to prevent other server threads from
3687 	 * processing requests related to the file while all state is
3688 	 * being removed.
3689 	 */
3690 	rw_enter(&fp->file_rwlock, RW_WRITER);
3691 
3692 	/* Remove ALL state from the file */
3693 	while (sp = rfs4_findstate_by_file(fp)) {
3694 		rfs4_state_close(sp, FALSE, FALSE, CRED());
3695 		rfs4_state_rele_nounlock(sp);
3696 	}
3697 
3698 	/*
3699 	 * This is only safe since there are no further references to
3700 	 * the file.
3701 	 */
3702 	rfs4_dbe_lock(fp->dbe);
3703 	if (fp->vp) {
3704 		vnode_t *vp = fp->vp;
3705 
3706 		mutex_enter(&vp->v_lock);
3707 		(void) vsd_set(vp, nfs4_srv_vkey, NULL);
3708 		mutex_exit(&vp->v_lock);
3709 		VN_RELE(vp);
3710 		fp->vp = NULL;
3711 	}
3712 	rfs4_dbe_unlock(fp->dbe);
3713 
3714 	/* Finally let other references to proceed */
3715 	rw_exit(&fp->file_rwlock);
3716 }
3717 
3718 /*
3719  * This function is used as a target for the rfs4_dbe_walk() call
3720  * below.  The purpose of this function is to see if the
3721  * lockowner_state refers to a file that resides within the exportinfo
3722  * export.  If so, then remove the lock_owner state (file locks and
3723  * share "locks") for this object since the intent is the server is
3724  * unexporting the specified directory.  Be sure to invalidate the
3725  * object after the state has been released
3726  */
3727 static void
3728 rfs4_lo_state_walk_callout(rfs4_entry_t u_entry, void *e)
3729 {
3730 	rfs4_lo_state_t *lsp = (rfs4_lo_state_t *)u_entry;
3731 	struct exportinfo *exi = (struct exportinfo *)e;
3732 	nfs_fh4_fmt_t   fhfmt4, *exi_fhp, *finfo_fhp;
3733 	fhandle_t *efhp;
3734 
3735 	efhp = (fhandle_t *)&exi->exi_fh;
3736 	exi_fhp = (nfs_fh4_fmt_t *)&fhfmt4;
3737 
3738 	FH_TO_FMT4(efhp, exi_fhp);
3739 
3740 	finfo_fhp =
3741 	    (nfs_fh4_fmt_t *)lsp->state->finfo->filehandle.nfs_fh4_val;
3742 
3743 	if (EQFSID(&finfo_fhp->fh4_fsid, &exi_fhp->fh4_fsid) &&
3744 	    bcmp(&finfo_fhp->fh4_xdata, &exi_fhp->fh4_xdata,
3745 	    exi_fhp->fh4_xlen) == 0) {
3746 		rfs4_state_close(lsp->state, FALSE, FALSE, CRED());
3747 		rfs4_dbe_invalidate(lsp->dbe);
3748 		rfs4_dbe_invalidate(lsp->state->dbe);
3749 	}
3750 }
3751 
3752 /*
3753  * This function is used as a target for the rfs4_dbe_walk() call
3754  * below.  The purpose of this function is to see if the state refers
3755  * to a file that resides within the exportinfo export.  If so, then
3756  * remove the open state for this object since the intent is the
3757  * server is unexporting the specified directory.  The main result for
3758  * this type of entry is to invalidate it such it will not be found in
3759  * the future.
3760  */
3761 static void
3762 rfs4_state_walk_callout(rfs4_entry_t u_entry, void *e)
3763 {
3764 	rfs4_state_t *sp = (rfs4_state_t *)u_entry;
3765 	struct exportinfo *exi = (struct exportinfo *)e;
3766 	nfs_fh4_fmt_t   fhfmt4, *exi_fhp, *finfo_fhp;
3767 	fhandle_t *efhp;
3768 
3769 	efhp = (fhandle_t *)&exi->exi_fh;
3770 	exi_fhp = (nfs_fh4_fmt_t *)&fhfmt4;
3771 
3772 	FH_TO_FMT4(efhp, exi_fhp);
3773 
3774 	finfo_fhp =
3775 	    (nfs_fh4_fmt_t *)sp->finfo->filehandle.nfs_fh4_val;
3776 
3777 	if (EQFSID(&finfo_fhp->fh4_fsid, &exi_fhp->fh4_fsid) &&
3778 	    bcmp(&finfo_fhp->fh4_xdata, &exi_fhp->fh4_xdata,
3779 	    exi_fhp->fh4_xlen) == 0) {
3780 		rfs4_state_close(sp, TRUE, FALSE, CRED());
3781 		rfs4_dbe_invalidate(sp->dbe);
3782 	}
3783 }
3784 
3785 /*
3786  * This function is used as a target for the rfs4_dbe_walk() call
3787  * below.  The purpose of this function is to see if the state refers
3788  * to a file that resides within the exportinfo export.  If so, then
3789  * remove the deleg state for this object since the intent is the
3790  * server is unexporting the specified directory.  The main result for
3791  * this type of entry is to invalidate it such it will not be found in
3792  * the future.
3793  */
3794 static void
3795 rfs4_deleg_state_walk_callout(rfs4_entry_t u_entry, void *e)
3796 {
3797 	rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
3798 	struct exportinfo *exi = (struct exportinfo *)e;
3799 	nfs_fh4_fmt_t   fhfmt4, *exi_fhp, *finfo_fhp;
3800 	fhandle_t *efhp;
3801 
3802 	efhp = (fhandle_t *)&exi->exi_fh;
3803 	exi_fhp = (nfs_fh4_fmt_t *)&fhfmt4;
3804 
3805 	FH_TO_FMT4(efhp, exi_fhp);
3806 
3807 	finfo_fhp =
3808 	    (nfs_fh4_fmt_t *)dsp->finfo->filehandle.nfs_fh4_val;
3809 
3810 	if (EQFSID(&finfo_fhp->fh4_fsid, &exi_fhp->fh4_fsid) &&
3811 	    bcmp(&finfo_fhp->fh4_xdata, &exi_fhp->fh4_xdata,
3812 	    exi_fhp->fh4_xlen) == 0) {
3813 		rfs4_dbe_invalidate(dsp->dbe);
3814 	}
3815 }
3816 
3817 /*
3818  * This function is used as a target for the rfs4_dbe_walk() call
3819  * below.  The purpose of this function is to see if the state refers
3820  * to a file that resides within the exportinfo export.  If so, then
3821  * release vnode hold for this object since the intent is the server
3822  * is unexporting the specified directory.  Invalidation will prevent
3823  * this struct from being found in the future.
3824  */
3825 static void
3826 rfs4_file_walk_callout(rfs4_entry_t u_entry, void *e)
3827 {
3828 	rfs4_file_t *fp = (rfs4_file_t *)u_entry;
3829 	struct exportinfo *exi = (struct exportinfo *)e;
3830 	nfs_fh4_fmt_t   fhfmt4, *exi_fhp, *finfo_fhp;
3831 	fhandle_t *efhp;
3832 
3833 	efhp = (fhandle_t *)&exi->exi_fh;
3834 	exi_fhp = (nfs_fh4_fmt_t *)&fhfmt4;
3835 
3836 	FH_TO_FMT4(efhp, exi_fhp);
3837 
3838 	finfo_fhp = (nfs_fh4_fmt_t *)fp->filehandle.nfs_fh4_val;
3839 
3840 	if (EQFSID(&finfo_fhp->fh4_fsid, &exi_fhp->fh4_fsid) &&
3841 	    bcmp(&finfo_fhp->fh4_xdata, &exi_fhp->fh4_xdata,
3842 	    exi_fhp->fh4_xlen) == 0) {
3843 		if (fp->vp) {
3844 			vnode_t *vp = fp->vp;
3845 
3846 			/*
3847 			 * don't leak monitors and remove the reference
3848 			 * put on the vnode when the delegation was granted.
3849 			 */
3850 			if (fp->dinfo->dtype == OPEN_DELEGATE_READ) {
3851 				(void) fem_uninstall(vp, deleg_rdops,
3852 				    (void *)fp);
3853 				vn_open_downgrade(vp, FREAD);
3854 			} else if (fp->dinfo->dtype == OPEN_DELEGATE_WRITE) {
3855 				(void) fem_uninstall(vp, deleg_wrops,
3856 				    (void *)fp);
3857 				vn_open_downgrade(vp, FREAD|FWRITE);
3858 			}
3859 			mutex_enter(&vp->v_lock);
3860 			(void) vsd_set(vp, nfs4_srv_vkey, NULL);
3861 			mutex_exit(&vp->v_lock);
3862 			VN_RELE(vp);
3863 			fp->vp = NULL;
3864 		}
3865 		rfs4_dbe_invalidate(fp->dbe);
3866 	}
3867 }
3868 
3869 /*
3870  * Given a directory that is being unexported, cleanup/release all
3871  * state in the server that refers to objects residing underneath this
3872  * particular export.  The ordering of the release is important.
3873  * Lock_owner, then state and then file.
3874  */
3875 void
3876 rfs4_clean_state_exi(struct exportinfo *exi)
3877 {
3878 	mutex_enter(&rfs4_state_lock);
3879 
3880 	if (rfs4_server_state == NULL) {
3881 		mutex_exit(&rfs4_state_lock);
3882 		return;
3883 	}
3884 
3885 	rfs4_dbe_walk(rfs4_lo_state_tab, rfs4_lo_state_walk_callout, exi);
3886 	rfs4_dbe_walk(rfs4_state_tab, rfs4_state_walk_callout, exi);
3887 	rfs4_dbe_walk(rfs4_deleg_state_tab, rfs4_deleg_state_walk_callout, exi);
3888 	rfs4_dbe_walk(rfs4_file_tab, rfs4_file_walk_callout, exi);
3889 
3890 	mutex_exit(&rfs4_state_lock);
3891 }
3892