xref: /illumos-gate/usr/src/uts/common/fs/nfs/nfs4_state.c (revision 3ba944265c4ae1fcf23ef758537c2e4f4feec16e)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
23  * Copyright 2014 Nexenta Systems, Inc.  All rights reserved.
24  */
25 
26 #include <sys/systm.h>
27 #include <sys/kmem.h>
28 #include <sys/cmn_err.h>
29 #include <sys/atomic.h>
30 #include <sys/clconf.h>
31 #include <sys/cladm.h>
32 #include <sys/flock.h>
33 #include <nfs/export.h>
34 #include <nfs/nfs.h>
35 #include <nfs/nfs4.h>
36 #include <nfs/nfssys.h>
37 #include <nfs/lm.h>
38 #include <sys/pathname.h>
39 #include <sys/sdt.h>
40 #include <sys/nvpair.h>
41 
42 extern u_longlong_t nfs4_srv_caller_id;
43 
44 extern time_t rfs4_start_time;
45 extern uint_t nfs4_srv_vkey;
46 
47 stateid4 special0 = {
48 	0,
49 	{ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 }
50 };
51 
52 stateid4 special1 = {
53 	0xffffffff,
54 	{
55 		(char)0xff, (char)0xff, (char)0xff, (char)0xff,
56 		(char)0xff, (char)0xff, (char)0xff, (char)0xff,
57 		(char)0xff, (char)0xff, (char)0xff, (char)0xff
58 	}
59 };
60 
61 
62 #define	ISSPECIAL(id)  (stateid4_cmp(id, &special0) || \
63 			stateid4_cmp(id, &special1))
64 
65 /* For embedding the cluster nodeid into our clientid */
66 #define	CLUSTER_NODEID_SHIFT	24
67 #define	CLUSTER_MAX_NODEID	255
68 
69 #ifdef DEBUG
70 int rfs4_debug;
71 #endif
72 
73 static uint32_t rfs4_database_debug = 0x00;
74 
75 static void rfs4_ss_clid_write(rfs4_client_t *cp, char *leaf);
76 static void rfs4_ss_clid_write_one(rfs4_client_t *cp, char *dir, char *leaf);
77 static void rfs4_dss_clear_oldstate(rfs4_servinst_t *sip);
78 static void rfs4_ss_chkclid_sip(rfs4_client_t *cp, rfs4_servinst_t *sip);
79 
80 /*
81  * Couple of simple init/destroy functions for a general waiter
82  */
83 void
84 rfs4_sw_init(rfs4_state_wait_t *swp)
85 {
86 	mutex_init(swp->sw_cv_lock, NULL, MUTEX_DEFAULT, NULL);
87 	cv_init(swp->sw_cv, NULL, CV_DEFAULT, NULL);
88 	swp->sw_active = FALSE;
89 	swp->sw_wait_count = 0;
90 }
91 
92 void
93 rfs4_sw_destroy(rfs4_state_wait_t *swp)
94 {
95 	mutex_destroy(swp->sw_cv_lock);
96 	cv_destroy(swp->sw_cv);
97 }
98 
99 void
100 rfs4_sw_enter(rfs4_state_wait_t *swp)
101 {
102 	mutex_enter(swp->sw_cv_lock);
103 	while (swp->sw_active) {
104 		swp->sw_wait_count++;
105 		cv_wait(swp->sw_cv, swp->sw_cv_lock);
106 		swp->sw_wait_count--;
107 	}
108 	ASSERT(swp->sw_active == FALSE);
109 	swp->sw_active = TRUE;
110 	mutex_exit(swp->sw_cv_lock);
111 }
112 
113 void
114 rfs4_sw_exit(rfs4_state_wait_t *swp)
115 {
116 	mutex_enter(swp->sw_cv_lock);
117 	ASSERT(swp->sw_active == TRUE);
118 	swp->sw_active = FALSE;
119 	if (swp->sw_wait_count != 0)
120 		cv_broadcast(swp->sw_cv);
121 	mutex_exit(swp->sw_cv_lock);
122 }
123 
124 /*
125  * CPR callback id -- not related to v4 callbacks
126  */
127 static callb_id_t cpr_id = 0;
128 
129 static void
130 deep_lock_copy(LOCK4res *dres, LOCK4res *sres)
131 {
132 	lock_owner4 *slo = &sres->LOCK4res_u.denied.owner;
133 	lock_owner4 *dlo = &dres->LOCK4res_u.denied.owner;
134 
135 	if (sres->status == NFS4ERR_DENIED) {
136 		dlo->owner_val = kmem_alloc(slo->owner_len, KM_SLEEP);
137 		bcopy(slo->owner_val, dlo->owner_val, slo->owner_len);
138 	}
139 }
140 
141 static void
142 deep_lock_free(LOCK4res *res)
143 {
144 	lock_owner4 *lo = &res->LOCK4res_u.denied.owner;
145 
146 	if (res->status == NFS4ERR_DENIED)
147 		kmem_free(lo->owner_val, lo->owner_len);
148 }
149 
150 static void
151 deep_open_copy(OPEN4res *dres, OPEN4res *sres)
152 {
153 	nfsace4 *sacep, *dacep;
154 
155 	if (sres->status != NFS4_OK) {
156 		return;
157 	}
158 
159 	dres->attrset = sres->attrset;
160 
161 	switch (sres->delegation.delegation_type) {
162 	case OPEN_DELEGATE_NONE:
163 		return;
164 	case OPEN_DELEGATE_READ:
165 		sacep = &sres->delegation.open_delegation4_u.read.permissions;
166 		dacep = &dres->delegation.open_delegation4_u.read.permissions;
167 		break;
168 	case OPEN_DELEGATE_WRITE:
169 		sacep = &sres->delegation.open_delegation4_u.write.permissions;
170 		dacep = &dres->delegation.open_delegation4_u.write.permissions;
171 		break;
172 	}
173 	dacep->who.utf8string_val =
174 	    kmem_alloc(sacep->who.utf8string_len, KM_SLEEP);
175 	bcopy(sacep->who.utf8string_val, dacep->who.utf8string_val,
176 	    sacep->who.utf8string_len);
177 }
178 
179 static void
180 deep_open_free(OPEN4res *res)
181 {
182 	nfsace4 *acep;
183 	if (res->status != NFS4_OK)
184 		return;
185 
186 	switch (res->delegation.delegation_type) {
187 	case OPEN_DELEGATE_NONE:
188 		return;
189 	case OPEN_DELEGATE_READ:
190 		acep = &res->delegation.open_delegation4_u.read.permissions;
191 		break;
192 	case OPEN_DELEGATE_WRITE:
193 		acep = &res->delegation.open_delegation4_u.write.permissions;
194 		break;
195 	}
196 
197 	if (acep->who.utf8string_val) {
198 		kmem_free(acep->who.utf8string_val, acep->who.utf8string_len);
199 		acep->who.utf8string_val = NULL;
200 	}
201 }
202 
203 void
204 rfs4_free_reply(nfs_resop4 *rp)
205 {
206 	switch (rp->resop) {
207 	case OP_LOCK:
208 		deep_lock_free(&rp->nfs_resop4_u.oplock);
209 		break;
210 	case OP_OPEN:
211 		deep_open_free(&rp->nfs_resop4_u.opopen);
212 	default:
213 		break;
214 	}
215 }
216 
217 void
218 rfs4_copy_reply(nfs_resop4 *dst, nfs_resop4 *src)
219 {
220 	*dst = *src;
221 
222 	/* Handle responses that need deep copy */
223 	switch (src->resop) {
224 	case OP_LOCK:
225 		deep_lock_copy(&dst->nfs_resop4_u.oplock,
226 		    &src->nfs_resop4_u.oplock);
227 		break;
228 	case OP_OPEN:
229 		deep_open_copy(&dst->nfs_resop4_u.opopen,
230 		    &src->nfs_resop4_u.opopen);
231 		break;
232 	default:
233 		break;
234 	};
235 }
236 
237 /*
238  * This is the implementation of the underlying state engine. The
239  * public interface to this engine is described by
240  * nfs4_state.h. Callers to the engine should hold no state engine
241  * locks when they call in to it. If the protocol needs to lock data
242  * structures it should do so after acquiring all references to them
243  * first and then follow the following lock order:
244  *
245  *	client > openowner > state > lo_state > lockowner > file.
246  *
247  * Internally we only allow a thread to hold one hash bucket lock at a
248  * time and the lock is higher in the lock order (must be acquired
249  * first) than the data structure that is on that hash list.
250  *
251  * If a new reference was acquired by the caller, that reference needs
252  * to be released after releasing all acquired locks with the
253  * corresponding rfs4_*_rele routine.
254  */
255 
256 /*
257  * This code is some what prototypical for now. Its purpose currently is to
258  * implement the interfaces sufficiently to finish the higher protocol
259  * elements. This will be replaced by a dynamically resizeable tables
260  * backed by kmem_cache allocator. However synchronization is handled
261  * correctly (I hope) and will not change by much.  The mutexes for
262  * the hash buckets that can be used to create new instances of data
263  * structures  might be good candidates to evolve into reader writer
264  * locks. If it has to do a creation, it would be holding the
265  * mutex across a kmem_alloc with KM_SLEEP specified.
266  */
267 
268 #ifdef DEBUG
269 #define	TABSIZE 17
270 #else
271 #define	TABSIZE 2047
272 #endif
273 
274 #define	ADDRHASH(key) ((unsigned long)(key) >> 3)
275 
276 /* Used to serialize create/destroy of rfs4_server_state database */
277 kmutex_t	rfs4_state_lock;
278 static rfs4_database_t *rfs4_server_state = NULL;
279 
280 /* Used to serialize lookups of clientids */
281 static	krwlock_t	rfs4_findclient_lock;
282 
283 /*
284  * For now this "table" is exposed so that the CPR callback
285  * function can tromp through it..
286  */
287 rfs4_table_t *rfs4_client_tab;
288 
289 static rfs4_index_t *rfs4_clientid_idx;
290 static rfs4_index_t *rfs4_nfsclnt_idx;
291 static rfs4_table_t *rfs4_clntip_tab;
292 static rfs4_index_t *rfs4_clntip_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_clntip_cache_time = 0;
336 static time_t rfs4_openowner_cache_time = 0;
337 static time_t rfs4_state_cache_time = 0;
338 static time_t rfs4_lo_state_cache_time = 0;
339 static time_t rfs4_lockowner_cache_time = 0;
340 static time_t rfs4_file_cache_time = 0;
341 static time_t rfs4_deleg_state_cache_time = 0;
342 
343 static bool_t rfs4_client_create(rfs4_entry_t, void *);
344 static void rfs4_dss_remove_cpleaf(rfs4_client_t *);
345 static void rfs4_dss_remove_leaf(rfs4_servinst_t *, char *, char *);
346 static void rfs4_client_destroy(rfs4_entry_t);
347 static bool_t rfs4_client_expiry(rfs4_entry_t);
348 static uint32_t clientid_hash(void *);
349 static bool_t clientid_compare(rfs4_entry_t, void *);
350 static void *clientid_mkkey(rfs4_entry_t);
351 static uint32_t nfsclnt_hash(void *);
352 static bool_t nfsclnt_compare(rfs4_entry_t, void *);
353 static void *nfsclnt_mkkey(rfs4_entry_t);
354 static bool_t rfs4_clntip_expiry(rfs4_entry_t);
355 static void rfs4_clntip_destroy(rfs4_entry_t);
356 static bool_t rfs4_clntip_create(rfs4_entry_t, void *);
357 static uint32_t clntip_hash(void *);
358 static bool_t clntip_compare(rfs4_entry_t, void *);
359 static void *clntip_mkkey(rfs4_entry_t);
360 static bool_t rfs4_openowner_create(rfs4_entry_t, void *);
361 static void rfs4_openowner_destroy(rfs4_entry_t);
362 static bool_t rfs4_openowner_expiry(rfs4_entry_t);
363 static uint32_t openowner_hash(void *);
364 static bool_t openowner_compare(rfs4_entry_t, void *);
365 static void *openowner_mkkey(rfs4_entry_t);
366 static bool_t rfs4_state_create(rfs4_entry_t, void *);
367 static void rfs4_state_destroy(rfs4_entry_t);
368 static bool_t rfs4_state_expiry(rfs4_entry_t);
369 static uint32_t state_hash(void *);
370 static bool_t state_compare(rfs4_entry_t, void *);
371 static void *state_mkkey(rfs4_entry_t);
372 static uint32_t state_owner_file_hash(void *);
373 static bool_t state_owner_file_compare(rfs4_entry_t, void *);
374 static void *state_owner_file_mkkey(rfs4_entry_t);
375 static uint32_t state_file_hash(void *);
376 static bool_t state_file_compare(rfs4_entry_t, void *);
377 static void *state_file_mkkey(rfs4_entry_t);
378 static bool_t rfs4_lo_state_create(rfs4_entry_t, void *);
379 static void rfs4_lo_state_destroy(rfs4_entry_t);
380 static bool_t rfs4_lo_state_expiry(rfs4_entry_t);
381 static uint32_t lo_state_hash(void *);
382 static bool_t lo_state_compare(rfs4_entry_t, void *);
383 static void *lo_state_mkkey(rfs4_entry_t);
384 static uint32_t lo_state_lo_hash(void *);
385 static bool_t lo_state_lo_compare(rfs4_entry_t, void *);
386 static void *lo_state_lo_mkkey(rfs4_entry_t);
387 static bool_t rfs4_lockowner_create(rfs4_entry_t, void *);
388 static void rfs4_lockowner_destroy(rfs4_entry_t);
389 static bool_t rfs4_lockowner_expiry(rfs4_entry_t);
390 static uint32_t lockowner_hash(void *);
391 static bool_t lockowner_compare(rfs4_entry_t, void *);
392 static void *lockowner_mkkey(rfs4_entry_t);
393 static uint32_t pid_hash(void *);
394 static bool_t pid_compare(rfs4_entry_t, void *);
395 static void *pid_mkkey(rfs4_entry_t);
396 static bool_t rfs4_file_create(rfs4_entry_t, void *);
397 static void rfs4_file_destroy(rfs4_entry_t);
398 static uint32_t file_hash(void *);
399 static bool_t file_compare(rfs4_entry_t, void *);
400 static void *file_mkkey(rfs4_entry_t);
401 static bool_t rfs4_deleg_state_create(rfs4_entry_t, void *);
402 static void rfs4_deleg_state_destroy(rfs4_entry_t);
403 static bool_t rfs4_deleg_state_expiry(rfs4_entry_t);
404 static uint32_t deleg_hash(void *);
405 static bool_t deleg_compare(rfs4_entry_t, void *);
406 static void *deleg_mkkey(rfs4_entry_t);
407 static uint32_t deleg_state_hash(void *);
408 static bool_t deleg_state_compare(rfs4_entry_t, void *);
409 static void *deleg_state_mkkey(rfs4_entry_t);
410 
411 static void rfs4_state_rele_nounlock(rfs4_state_t *);
412 
413 static int rfs4_ss_enabled = 0;
414 
415 extern void (*rfs4_client_clrst)(struct nfs4clrst_args *);
416 
417 void
418 rfs4_ss_pnfree(rfs4_ss_pn_t *ss_pn)
419 {
420 	kmem_free(ss_pn, sizeof (rfs4_ss_pn_t));
421 }
422 
423 static rfs4_ss_pn_t *
424 rfs4_ss_pnalloc(char *dir, char *leaf)
425 {
426 	rfs4_ss_pn_t *ss_pn;
427 	int 	dir_len, leaf_len;
428 
429 	/*
430 	 * validate we have a resonable path
431 	 * (account for the '/' and trailing null)
432 	 */
433 	if ((dir_len = strlen(dir)) > MAXPATHLEN ||
434 	    (leaf_len = strlen(leaf)) > MAXNAMELEN ||
435 	    (dir_len + leaf_len + 2) > MAXPATHLEN) {
436 		return (NULL);
437 	}
438 
439 	ss_pn = kmem_alloc(sizeof (rfs4_ss_pn_t), KM_SLEEP);
440 
441 	(void) snprintf(ss_pn->pn, MAXPATHLEN, "%s/%s", dir, leaf);
442 	/* Handy pointer to just the leaf name */
443 	ss_pn->leaf = ss_pn->pn + dir_len + 1;
444 	return (ss_pn);
445 }
446 
447 
448 /*
449  * Move the "leaf" filename from "sdir" directory
450  * to the "ddir" directory. Return the pathname of
451  * the destination unless the rename fails in which
452  * case we need to return the source pathname.
453  */
454 static rfs4_ss_pn_t *
455 rfs4_ss_movestate(char *sdir, char *ddir, char *leaf)
456 {
457 	rfs4_ss_pn_t *src, *dst;
458 
459 	if ((src = rfs4_ss_pnalloc(sdir, leaf)) == NULL)
460 		return (NULL);
461 
462 	if ((dst = rfs4_ss_pnalloc(ddir, leaf)) == NULL) {
463 		rfs4_ss_pnfree(src);
464 		return (NULL);
465 	}
466 
467 	/*
468 	 * If the rename fails we shall return the src
469 	 * pathname and free the dst. Otherwise we need
470 	 * to free the src and return the dst pathanme.
471 	 */
472 	if (vn_rename(src->pn, dst->pn, UIO_SYSSPACE)) {
473 		rfs4_ss_pnfree(dst);
474 		return (src);
475 	}
476 	rfs4_ss_pnfree(src);
477 	return (dst);
478 }
479 
480 
481 static rfs4_oldstate_t *
482 rfs4_ss_getstate(vnode_t *dvp, rfs4_ss_pn_t *ss_pn)
483 {
484 	struct uio uio;
485 	struct iovec iov[3];
486 
487 	rfs4_oldstate_t *cl_ss = NULL;
488 	vnode_t *vp;
489 	vattr_t va;
490 	uint_t id_len;
491 	int err, kill_file, file_vers;
492 
493 	if (ss_pn == NULL)
494 		return (NULL);
495 
496 	/*
497 	 * open the state file.
498 	 */
499 	if (vn_open(ss_pn->pn, UIO_SYSSPACE, FREAD, 0, &vp, 0, 0) != 0) {
500 		return (NULL);
501 	}
502 
503 	if (vp->v_type != VREG) {
504 		(void) VOP_CLOSE(vp, FREAD, 1, (offset_t)0, CRED(), NULL);
505 		VN_RELE(vp);
506 		return (NULL);
507 	}
508 
509 	err = VOP_ACCESS(vp, VREAD, 0, CRED(), NULL);
510 	if (err) {
511 		/*
512 		 * We don't have read access? better get the heck out.
513 		 */
514 		(void) VOP_CLOSE(vp, FREAD, 1, (offset_t)0, CRED(), NULL);
515 		VN_RELE(vp);
516 		return (NULL);
517 	}
518 
519 	(void) VOP_RWLOCK(vp, V_WRITELOCK_FALSE, NULL);
520 	/*
521 	 * get the file size to do some basic validation
522 	 */
523 	va.va_mask = AT_SIZE;
524 	err = VOP_GETATTR(vp, &va, 0, CRED(), NULL);
525 
526 	kill_file = (va.va_size == 0 || va.va_size <
527 	    (NFS4_VERIFIER_SIZE + sizeof (uint_t)+1));
528 
529 	if (err || kill_file) {
530 		VOP_RWUNLOCK(vp, V_WRITELOCK_FALSE, NULL);
531 		(void) VOP_CLOSE(vp, FREAD, 1, (offset_t)0, CRED(), NULL);
532 		VN_RELE(vp);
533 		if (kill_file) {
534 			(void) VOP_REMOVE(dvp, ss_pn->leaf, CRED(), NULL, 0);
535 		}
536 		return (NULL);
537 	}
538 
539 	cl_ss = kmem_alloc(sizeof (rfs4_oldstate_t), KM_SLEEP);
540 
541 	/*
542 	 * build iovecs to read in the file_version, verifier and id_len
543 	 */
544 	iov[0].iov_base = (caddr_t)&file_vers;
545 	iov[0].iov_len = sizeof (int);
546 	iov[1].iov_base = (caddr_t)&cl_ss->cl_id4.verifier;
547 	iov[1].iov_len = NFS4_VERIFIER_SIZE;
548 	iov[2].iov_base = (caddr_t)&id_len;
549 	iov[2].iov_len = sizeof (uint_t);
550 
551 	uio.uio_iov = iov;
552 	uio.uio_iovcnt = 3;
553 	uio.uio_segflg = UIO_SYSSPACE;
554 	uio.uio_loffset = 0;
555 	uio.uio_resid = sizeof (int) + NFS4_VERIFIER_SIZE + sizeof (uint_t);
556 
557 	if (err = VOP_READ(vp, &uio, FREAD, CRED(), NULL)) {
558 		VOP_RWUNLOCK(vp, V_WRITELOCK_FALSE, NULL);
559 		(void) VOP_CLOSE(vp, FREAD, 1, (offset_t)0, CRED(), NULL);
560 		VN_RELE(vp);
561 		kmem_free(cl_ss, sizeof (rfs4_oldstate_t));
562 		return (NULL);
563 	}
564 
565 	/*
566 	 * if the file_version doesn't match or if the
567 	 * id_len is zero or the combination of the verifier,
568 	 * id_len and id_val is bigger than the file we have
569 	 * a problem. If so ditch the file.
570 	 */
571 	kill_file = (file_vers != NFS4_SS_VERSION || id_len == 0 ||
572 	    (id_len + NFS4_VERIFIER_SIZE + sizeof (uint_t)) > va.va_size);
573 
574 	if (err || kill_file) {
575 		VOP_RWUNLOCK(vp, V_WRITELOCK_FALSE, NULL);
576 		(void) VOP_CLOSE(vp, FREAD, 1, (offset_t)0, CRED(), NULL);
577 		VN_RELE(vp);
578 		kmem_free(cl_ss, sizeof (rfs4_oldstate_t));
579 		if (kill_file) {
580 			(void) VOP_REMOVE(dvp, ss_pn->leaf, CRED(), NULL, 0);
581 		}
582 		return (NULL);
583 	}
584 
585 	/*
586 	 * now get the client id value
587 	 */
588 	cl_ss->cl_id4.id_val = kmem_alloc(id_len, KM_SLEEP);
589 	iov[0].iov_base = cl_ss->cl_id4.id_val;
590 	iov[0].iov_len = id_len;
591 
592 	uio.uio_iov = iov;
593 	uio.uio_iovcnt = 1;
594 	uio.uio_segflg = UIO_SYSSPACE;
595 	uio.uio_resid = cl_ss->cl_id4.id_len = id_len;
596 
597 	if (err = VOP_READ(vp, &uio, FREAD, CRED(), NULL)) {
598 		VOP_RWUNLOCK(vp, V_WRITELOCK_FALSE, NULL);
599 		(void) VOP_CLOSE(vp, FREAD, 1, (offset_t)0, CRED(), NULL);
600 		VN_RELE(vp);
601 		kmem_free(cl_ss->cl_id4.id_val, id_len);
602 		kmem_free(cl_ss, sizeof (rfs4_oldstate_t));
603 		return (NULL);
604 	}
605 
606 	VOP_RWUNLOCK(vp, V_WRITELOCK_FALSE, NULL);
607 	(void) VOP_CLOSE(vp, FREAD, 1, (offset_t)0, CRED(), NULL);
608 	VN_RELE(vp);
609 	return (cl_ss);
610 }
611 
612 #ifdef	nextdp
613 #undef nextdp
614 #endif
615 #define	nextdp(dp)	((struct dirent64 *)((char *)(dp) + (dp)->d_reclen))
616 
617 /*
618  * Add entries from statedir to supplied oldstate list.
619  * Optionally, move all entries from statedir -> destdir.
620  */
621 void
622 rfs4_ss_oldstate(rfs4_oldstate_t *oldstate, char *statedir, char *destdir)
623 {
624 	rfs4_ss_pn_t *ss_pn;
625 	rfs4_oldstate_t *cl_ss = NULL;
626 	char	*dirt = NULL;
627 	int	err, dir_eof = 0, size = 0;
628 	vnode_t *dvp;
629 	struct iovec iov;
630 	struct uio uio;
631 	struct dirent64 *dep;
632 	offset_t dirchunk_offset = 0;
633 
634 	/*
635 	 * open the state directory
636 	 */
637 	if (vn_open(statedir, UIO_SYSSPACE, FREAD, 0, &dvp, 0, 0))
638 		return;
639 
640 	if (dvp->v_type != VDIR || VOP_ACCESS(dvp, VREAD, 0, CRED(), NULL))
641 		goto out;
642 
643 	dirt = kmem_alloc(RFS4_SS_DIRSIZE, KM_SLEEP);
644 
645 	/*
646 	 * Get and process the directory entries
647 	 */
648 	while (!dir_eof) {
649 		(void) VOP_RWLOCK(dvp, V_WRITELOCK_FALSE, NULL);
650 		iov.iov_base = dirt;
651 		iov.iov_len = RFS4_SS_DIRSIZE;
652 		uio.uio_iov = &iov;
653 		uio.uio_iovcnt = 1;
654 		uio.uio_segflg = UIO_SYSSPACE;
655 		uio.uio_loffset = dirchunk_offset;
656 		uio.uio_resid = RFS4_SS_DIRSIZE;
657 
658 		err = VOP_READDIR(dvp, &uio, CRED(), &dir_eof, NULL, 0);
659 		VOP_RWUNLOCK(dvp, V_WRITELOCK_FALSE, NULL);
660 		if (err)
661 			goto out;
662 
663 		size = RFS4_SS_DIRSIZE - uio.uio_resid;
664 
665 		/*
666 		 * Process all the directory entries in this
667 		 * readdir chunk
668 		 */
669 		for (dep = (struct dirent64 *)dirt; size > 0;
670 		    dep = nextdp(dep)) {
671 
672 			size -= dep->d_reclen;
673 			dirchunk_offset = dep->d_off;
674 
675 			/*
676 			 * Skip '.' and '..'
677 			 */
678 			if (NFS_IS_DOTNAME(dep->d_name))
679 				continue;
680 
681 			ss_pn = rfs4_ss_pnalloc(statedir, dep->d_name);
682 			if (ss_pn == NULL)
683 				continue;
684 
685 			if (cl_ss = rfs4_ss_getstate(dvp, ss_pn)) {
686 				if (destdir != NULL) {
687 					rfs4_ss_pnfree(ss_pn);
688 					cl_ss->ss_pn = rfs4_ss_movestate(
689 					    statedir, destdir, dep->d_name);
690 				} else {
691 					cl_ss->ss_pn = ss_pn;
692 				}
693 				insque(cl_ss, oldstate);
694 			} else {
695 				rfs4_ss_pnfree(ss_pn);
696 			}
697 		}
698 	}
699 
700 out:
701 	(void) VOP_CLOSE(dvp, FREAD, 1, (offset_t)0, CRED(), NULL);
702 	VN_RELE(dvp);
703 	if (dirt)
704 		kmem_free((caddr_t)dirt, RFS4_SS_DIRSIZE);
705 }
706 
707 static void
708 rfs4_ss_init(void)
709 {
710 	int npaths = 1;
711 	char *default_dss_path = NFS4_DSS_VAR_DIR;
712 
713 	/* read the default stable storage state */
714 	rfs4_dss_readstate(npaths, &default_dss_path);
715 
716 	rfs4_ss_enabled = 1;
717 }
718 
719 static void
720 rfs4_ss_fini(void)
721 {
722 	rfs4_servinst_t *sip;
723 
724 	mutex_enter(&rfs4_servinst_lock);
725 	sip = rfs4_cur_servinst;
726 	while (sip != NULL) {
727 		rfs4_dss_clear_oldstate(sip);
728 		sip = sip->next;
729 	}
730 	mutex_exit(&rfs4_servinst_lock);
731 }
732 
733 /*
734  * Remove all oldstate files referenced by this servinst.
735  */
736 static void
737 rfs4_dss_clear_oldstate(rfs4_servinst_t *sip)
738 {
739 	rfs4_oldstate_t *os_head, *osp;
740 
741 	rw_enter(&sip->oldstate_lock, RW_WRITER);
742 	os_head = sip->oldstate;
743 
744 	if (os_head == NULL) {
745 		rw_exit(&sip->oldstate_lock);
746 		return;
747 	}
748 
749 	/* skip dummy entry */
750 	osp = os_head->next;
751 	while (osp != os_head) {
752 		char *leaf = osp->ss_pn->leaf;
753 		rfs4_oldstate_t *os_next;
754 
755 		rfs4_dss_remove_leaf(sip, NFS4_DSS_OLDSTATE_LEAF, leaf);
756 
757 		if (osp->cl_id4.id_val)
758 			kmem_free(osp->cl_id4.id_val, osp->cl_id4.id_len);
759 		rfs4_ss_pnfree(osp->ss_pn);
760 
761 		os_next = osp->next;
762 		remque(osp);
763 		kmem_free(osp, sizeof (rfs4_oldstate_t));
764 		osp = os_next;
765 	}
766 
767 	rw_exit(&sip->oldstate_lock);
768 }
769 
770 /*
771  * Form the state and oldstate paths, and read in the stable storage files.
772  */
773 void
774 rfs4_dss_readstate(int npaths, char **paths)
775 {
776 	int i;
777 	char *state, *oldstate;
778 
779 	state = kmem_alloc(MAXPATHLEN, KM_SLEEP);
780 	oldstate = kmem_alloc(MAXPATHLEN, KM_SLEEP);
781 
782 	for (i = 0; i < npaths; i++) {
783 		char *path = paths[i];
784 
785 		(void) sprintf(state, "%s/%s", path, NFS4_DSS_STATE_LEAF);
786 		(void) sprintf(oldstate, "%s/%s", path, NFS4_DSS_OLDSTATE_LEAF);
787 
788 		/*
789 		 * Populate the current server instance's oldstate list.
790 		 *
791 		 * 1. Read stable storage data from old state directory,
792 		 *    leaving its contents alone.
793 		 *
794 		 * 2. Read stable storage data from state directory,
795 		 *    and move the latter's contents to old state
796 		 *    directory.
797 		 */
798 		rfs4_ss_oldstate(rfs4_cur_servinst->oldstate, oldstate, NULL);
799 		rfs4_ss_oldstate(rfs4_cur_servinst->oldstate, state, oldstate);
800 	}
801 
802 	kmem_free(state, MAXPATHLEN);
803 	kmem_free(oldstate, MAXPATHLEN);
804 }
805 
806 
807 /*
808  * Check if we are still in grace and if the client can be
809  * granted permission to perform reclaims.
810  */
811 void
812 rfs4_ss_chkclid(rfs4_client_t *cp)
813 {
814 	rfs4_servinst_t *sip;
815 
816 	/*
817 	 * It should be sufficient to check the oldstate data for just
818 	 * this client's instance. However, since our per-instance
819 	 * client grouping is solely temporal, HA-NFSv4 RG failover
820 	 * might result in clients of the same RG being partitioned into
821 	 * separate instances.
822 	 *
823 	 * Until the client grouping is improved, we must check the
824 	 * oldstate data for all instances with an active grace period.
825 	 *
826 	 * This also serves as the mechanism to remove stale oldstate data.
827 	 * The first time we check an instance after its grace period has
828 	 * expired, the oldstate data should be cleared.
829 	 *
830 	 * Start at the current instance, and walk the list backwards
831 	 * to the first.
832 	 */
833 	mutex_enter(&rfs4_servinst_lock);
834 	for (sip = rfs4_cur_servinst; sip != NULL; sip = sip->prev) {
835 		rfs4_ss_chkclid_sip(cp, sip);
836 
837 		/* if the above check found this client, we're done */
838 		if (cp->rc_can_reclaim)
839 			break;
840 	}
841 	mutex_exit(&rfs4_servinst_lock);
842 }
843 
844 static void
845 rfs4_ss_chkclid_sip(rfs4_client_t *cp, rfs4_servinst_t *sip)
846 {
847 	rfs4_oldstate_t *osp, *os_head;
848 
849 	/* short circuit everything if this server instance has no oldstate */
850 	rw_enter(&sip->oldstate_lock, RW_READER);
851 	os_head = sip->oldstate;
852 	rw_exit(&sip->oldstate_lock);
853 	if (os_head == NULL)
854 		return;
855 
856 	/*
857 	 * If this server instance is no longer in a grace period then
858 	 * the client won't be able to reclaim. No further need for this
859 	 * instance's oldstate data, so it can be cleared.
860 	 */
861 	if (!rfs4_servinst_in_grace(sip))
862 		return;
863 
864 	/* this instance is still in grace; search for the clientid */
865 
866 	rw_enter(&sip->oldstate_lock, RW_READER);
867 
868 	os_head = sip->oldstate;
869 	/* skip dummy entry */
870 	osp = os_head->next;
871 	while (osp != os_head) {
872 		if (osp->cl_id4.id_len == cp->rc_nfs_client.id_len) {
873 			if (bcmp(osp->cl_id4.id_val, cp->rc_nfs_client.id_val,
874 			    osp->cl_id4.id_len) == 0) {
875 				cp->rc_can_reclaim = 1;
876 				break;
877 			}
878 		}
879 		osp = osp->next;
880 	}
881 
882 	rw_exit(&sip->oldstate_lock);
883 }
884 
885 /*
886  * Place client information into stable storage: 1/3.
887  * First, generate the leaf filename, from the client's IP address and
888  * the server-generated short-hand clientid.
889  */
890 void
891 rfs4_ss_clid(rfs4_client_t *cp)
892 {
893 	const char *kinet_ntop6(uchar_t *, char *, size_t);
894 	char leaf[MAXNAMELEN], buf[INET6_ADDRSTRLEN];
895 	struct sockaddr *ca;
896 	uchar_t *b;
897 
898 	if (rfs4_ss_enabled == 0) {
899 		return;
900 	}
901 
902 	buf[0] = 0;
903 
904 	ca = (struct sockaddr *)&cp->rc_addr;
905 
906 	/*
907 	 * Convert the caller's IP address to a dotted string
908 	 */
909 	if (ca->sa_family == AF_INET) {
910 		b = (uchar_t *)&((struct sockaddr_in *)ca)->sin_addr;
911 		(void) sprintf(buf, "%03d.%03d.%03d.%03d", b[0] & 0xFF,
912 		    b[1] & 0xFF, b[2] & 0xFF, b[3] & 0xFF);
913 	} else if (ca->sa_family == AF_INET6) {
914 		struct sockaddr_in6 *sin6;
915 
916 		sin6 = (struct sockaddr_in6 *)ca;
917 		(void) kinet_ntop6((uchar_t *)&sin6->sin6_addr,
918 		    buf, INET6_ADDRSTRLEN);
919 	}
920 
921 	(void) snprintf(leaf, MAXNAMELEN, "%s-%llx", buf,
922 	    (longlong_t)cp->rc_clientid);
923 	rfs4_ss_clid_write(cp, leaf);
924 }
925 
926 /*
927  * Place client information into stable storage: 2/3.
928  * DSS: distributed stable storage: the file may need to be written to
929  * multiple directories.
930  */
931 static void
932 rfs4_ss_clid_write(rfs4_client_t *cp, char *leaf)
933 {
934 	rfs4_servinst_t *sip;
935 
936 	/*
937 	 * It should be sufficient to write the leaf file to (all) DSS paths
938 	 * associated with just this client's instance. However, since our
939 	 * per-instance client grouping is solely temporal, HA-NFSv4 RG
940 	 * failover might result in us losing DSS data.
941 	 *
942 	 * Until the client grouping is improved, we must write the DSS data
943 	 * to all instances' paths. Start at the current instance, and
944 	 * walk the list backwards to the first.
945 	 */
946 	mutex_enter(&rfs4_servinst_lock);
947 	for (sip = rfs4_cur_servinst; sip != NULL; sip = sip->prev) {
948 		int i, npaths = sip->dss_npaths;
949 
950 		/* write the leaf file to all DSS paths */
951 		for (i = 0; i < npaths; i++) {
952 			rfs4_dss_path_t *dss_path = sip->dss_paths[i];
953 
954 			/* HA-NFSv4 path might have been failed-away from us */
955 			if (dss_path == NULL)
956 				continue;
957 
958 			rfs4_ss_clid_write_one(cp, dss_path->path, leaf);
959 		}
960 	}
961 	mutex_exit(&rfs4_servinst_lock);
962 }
963 
964 /*
965  * Place client information into stable storage: 3/3.
966  * Write the stable storage data to the requested file.
967  */
968 static void
969 rfs4_ss_clid_write_one(rfs4_client_t *cp, char *dss_path, char *leaf)
970 {
971 	int ioflag;
972 	int file_vers = NFS4_SS_VERSION;
973 	size_t dirlen;
974 	struct uio uio;
975 	struct iovec iov[4];
976 	char *dir;
977 	rfs4_ss_pn_t *ss_pn;
978 	vnode_t *vp;
979 	nfs_client_id4 *cl_id4 = &(cp->rc_nfs_client);
980 
981 	/* allow 2 extra bytes for '/' & NUL */
982 	dirlen = strlen(dss_path) + strlen(NFS4_DSS_STATE_LEAF) + 2;
983 	dir = kmem_alloc(dirlen, KM_SLEEP);
984 	(void) sprintf(dir, "%s/%s", dss_path, NFS4_DSS_STATE_LEAF);
985 
986 	ss_pn = rfs4_ss_pnalloc(dir, leaf);
987 	/* rfs4_ss_pnalloc takes its own copy */
988 	kmem_free(dir, dirlen);
989 	if (ss_pn == NULL)
990 		return;
991 
992 	if (vn_open(ss_pn->pn, UIO_SYSSPACE, FCREAT|FWRITE, 0600, &vp,
993 	    CRCREAT, 0)) {
994 		rfs4_ss_pnfree(ss_pn);
995 		return;
996 	}
997 
998 	/*
999 	 * We need to record leaf - i.e. the filename - so that we know
1000 	 * what to remove, in the future. However, the dir part of cp->ss_pn
1001 	 * should never be referenced directly, since it's potentially only
1002 	 * one of several paths with this leaf in it.
1003 	 */
1004 	if (cp->rc_ss_pn != NULL) {
1005 		if (strcmp(cp->rc_ss_pn->leaf, leaf) == 0) {
1006 			/* we've already recorded *this* leaf */
1007 			rfs4_ss_pnfree(ss_pn);
1008 		} else {
1009 			/* replace with this leaf */
1010 			rfs4_ss_pnfree(cp->rc_ss_pn);
1011 			cp->rc_ss_pn = ss_pn;
1012 		}
1013 	} else {
1014 		cp->rc_ss_pn = ss_pn;
1015 	}
1016 
1017 	/*
1018 	 * Build a scatter list that points to the nfs_client_id4
1019 	 */
1020 	iov[0].iov_base = (caddr_t)&file_vers;
1021 	iov[0].iov_len = sizeof (int);
1022 	iov[1].iov_base = (caddr_t)&(cl_id4->verifier);
1023 	iov[1].iov_len = NFS4_VERIFIER_SIZE;
1024 	iov[2].iov_base = (caddr_t)&(cl_id4->id_len);
1025 	iov[2].iov_len = sizeof (uint_t);
1026 	iov[3].iov_base = (caddr_t)cl_id4->id_val;
1027 	iov[3].iov_len = cl_id4->id_len;
1028 
1029 	uio.uio_iov = iov;
1030 	uio.uio_iovcnt = 4;
1031 	uio.uio_loffset = 0;
1032 	uio.uio_segflg = UIO_SYSSPACE;
1033 	uio.uio_llimit = (rlim64_t)MAXOFFSET_T;
1034 	uio.uio_resid = cl_id4->id_len + sizeof (int) +
1035 	    NFS4_VERIFIER_SIZE + sizeof (uint_t);
1036 
1037 	ioflag = uio.uio_fmode = (FWRITE|FSYNC);
1038 	uio.uio_extflg = UIO_COPY_DEFAULT;
1039 
1040 	(void) VOP_RWLOCK(vp, V_WRITELOCK_TRUE, NULL);
1041 	/* write the full client id to the file. */
1042 	(void) VOP_WRITE(vp, &uio, ioflag, CRED(), NULL);
1043 	VOP_RWUNLOCK(vp, V_WRITELOCK_TRUE, NULL);
1044 
1045 	(void) VOP_CLOSE(vp, FWRITE, 1, (offset_t)0, CRED(), NULL);
1046 	VN_RELE(vp);
1047 }
1048 
1049 /*
1050  * DSS: distributed stable storage.
1051  * Unpack the list of paths passed by nfsd.
1052  * Use nvlist_alloc(9F) to manage the data.
1053  * The caller is responsible for allocating and freeing the buffer.
1054  */
1055 int
1056 rfs4_dss_setpaths(char *buf, size_t buflen)
1057 {
1058 	int error;
1059 
1060 	/*
1061 	 * If this is a "warm start", i.e. we previously had DSS paths,
1062 	 * preserve the old paths.
1063 	 */
1064 	if (rfs4_dss_paths != NULL) {
1065 		/*
1066 		 * Before we lose the ptr, destroy the nvlist and pathnames
1067 		 * array from the warm start before this one.
1068 		 */
1069 		nvlist_free(rfs4_dss_oldpaths);
1070 		rfs4_dss_oldpaths = rfs4_dss_paths;
1071 	}
1072 
1073 	/* unpack the buffer into a searchable nvlist */
1074 	error = nvlist_unpack(buf, buflen, &rfs4_dss_paths, KM_SLEEP);
1075 	if (error)
1076 		return (error);
1077 
1078 	/*
1079 	 * Search the nvlist for the pathnames nvpair (which is the only nvpair
1080 	 * in the list, and record its location.
1081 	 */
1082 	error = nvlist_lookup_string_array(rfs4_dss_paths, NFS4_DSS_NVPAIR_NAME,
1083 	    &rfs4_dss_newpaths, &rfs4_dss_numnewpaths);
1084 	return (error);
1085 }
1086 
1087 /*
1088  * Ultimately the nfssys() call NFS4_CLR_STATE endsup here
1089  * to find and mark the client for forced expire.
1090  */
1091 static void
1092 rfs4_client_scrub(rfs4_entry_t ent, void *arg)
1093 {
1094 	rfs4_client_t *cp = (rfs4_client_t *)ent;
1095 	struct nfs4clrst_args *clr = arg;
1096 	struct sockaddr_in6 *ent_sin6;
1097 	struct in6_addr  clr_in6;
1098 	struct sockaddr_in  *ent_sin;
1099 	struct in_addr   clr_in;
1100 
1101 	if (clr->addr_type != cp->rc_addr.ss_family) {
1102 		return;
1103 	}
1104 
1105 	switch (clr->addr_type) {
1106 
1107 	case AF_INET6:
1108 		/* copyin the address from user space */
1109 		if (copyin(clr->ap, &clr_in6, sizeof (clr_in6))) {
1110 			break;
1111 		}
1112 
1113 		ent_sin6 = (struct sockaddr_in6 *)&cp->rc_addr;
1114 
1115 		/*
1116 		 * now compare, and if equivalent mark entry
1117 		 * for forced expiration
1118 		 */
1119 		if (IN6_ARE_ADDR_EQUAL(&ent_sin6->sin6_addr, &clr_in6)) {
1120 			cp->rc_forced_expire = 1;
1121 		}
1122 		break;
1123 
1124 	case AF_INET:
1125 		/* copyin the address from user space */
1126 		if (copyin(clr->ap, &clr_in, sizeof (clr_in))) {
1127 			break;
1128 		}
1129 
1130 		ent_sin = (struct sockaddr_in *)&cp->rc_addr;
1131 
1132 		/*
1133 		 * now compare, and if equivalent mark entry
1134 		 * for forced expiration
1135 		 */
1136 		if (ent_sin->sin_addr.s_addr == clr_in.s_addr) {
1137 			cp->rc_forced_expire = 1;
1138 		}
1139 		break;
1140 
1141 	default:
1142 		/* force this assert to fail */
1143 		ASSERT(clr->addr_type != clr->addr_type);
1144 	}
1145 }
1146 
1147 /*
1148  * This is called from nfssys() in order to clear server state
1149  * for the specified client IP Address.
1150  */
1151 void
1152 rfs4_clear_client_state(struct nfs4clrst_args *clr)
1153 {
1154 	(void) rfs4_dbe_walk(rfs4_client_tab, rfs4_client_scrub, clr);
1155 }
1156 
1157 /*
1158  * Used to initialize the NFSv4 server's state or database.  All of
1159  * the tables are created and timers are set. Only called when NFSv4
1160  * service is provided.
1161  */
1162 void
1163 rfs4_state_init()
1164 {
1165 	int start_grace;
1166 	extern boolean_t rfs4_cpr_callb(void *, int);
1167 	char *dss_path = NFS4_DSS_VAR_DIR;
1168 	time_t start_time;
1169 
1170 	mutex_enter(&rfs4_state_lock);
1171 
1172 	/*
1173 	 * If the server state database has already been initialized,
1174 	 * skip it
1175 	 */
1176 	if (rfs4_server_state != NULL) {
1177 		mutex_exit(&rfs4_state_lock);
1178 		return;
1179 	}
1180 
1181 	rw_init(&rfs4_findclient_lock, NULL, RW_DEFAULT, NULL);
1182 
1183 	/*
1184 	 * Set the boot time.  If the server
1185 	 * has been restarted quickly and has had the opportunity to
1186 	 * service clients, then the start_time needs to be bumped
1187 	 * regardless.  A small window but it exists...
1188 	 */
1189 	start_time = gethrestime_sec();
1190 	if (rfs4_start_time < start_time)
1191 		rfs4_start_time = start_time;
1192 	else
1193 		rfs4_start_time++;
1194 
1195 	/* DSS: distributed stable storage: initialise served paths list */
1196 	rfs4_dss_pathlist = NULL;
1197 
1198 	/*
1199 	 * Create the first server instance, or a new one if the server has
1200 	 * been restarted; see above comments on rfs4_start_time. Don't
1201 	 * start its grace period; that will be done later, to maximise the
1202 	 * clients' recovery window.
1203 	 */
1204 	start_grace = 0;
1205 	rfs4_servinst_create(start_grace, 1, &dss_path);
1206 
1207 	/* reset the "first NFSv4 request" status */
1208 	rfs4_seen_first_compound = 0;
1209 
1210 	/*
1211 	 * Add a CPR callback so that we can update client
1212 	 * access times to extend the lease after a suspend
1213 	 * and resume (using the same class as rpcmod/connmgr)
1214 	 */
1215 	cpr_id = callb_add(rfs4_cpr_callb, 0, CB_CL_CPR_RPC, "rfs4");
1216 
1217 	/* set the various cache timers for table creation */
1218 	if (rfs4_client_cache_time == 0)
1219 		rfs4_client_cache_time = CLIENT_CACHE_TIME;
1220 	if (rfs4_openowner_cache_time == 0)
1221 		rfs4_openowner_cache_time = OPENOWNER_CACHE_TIME;
1222 	if (rfs4_state_cache_time == 0)
1223 		rfs4_state_cache_time = STATE_CACHE_TIME;
1224 	if (rfs4_lo_state_cache_time == 0)
1225 		rfs4_lo_state_cache_time = LO_STATE_CACHE_TIME;
1226 	if (rfs4_lockowner_cache_time == 0)
1227 		rfs4_lockowner_cache_time = LOCKOWNER_CACHE_TIME;
1228 	if (rfs4_file_cache_time == 0)
1229 		rfs4_file_cache_time = FILE_CACHE_TIME;
1230 	if (rfs4_deleg_state_cache_time == 0)
1231 		rfs4_deleg_state_cache_time = DELEG_STATE_CACHE_TIME;
1232 
1233 	/* Create the overall database to hold all server state */
1234 	rfs4_server_state = rfs4_database_create(rfs4_database_debug);
1235 
1236 	/* Now create the individual tables */
1237 	rfs4_client_cache_time *= rfs4_lease_time;
1238 	rfs4_client_tab = rfs4_table_create(rfs4_server_state,
1239 	    "Client",
1240 	    rfs4_client_cache_time,
1241 	    2,
1242 	    rfs4_client_create,
1243 	    rfs4_client_destroy,
1244 	    rfs4_client_expiry,
1245 	    sizeof (rfs4_client_t),
1246 	    TABSIZE,
1247 	    MAXTABSZ/8, 100);
1248 	rfs4_nfsclnt_idx = rfs4_index_create(rfs4_client_tab,
1249 	    "nfs_client_id4", nfsclnt_hash,
1250 	    nfsclnt_compare, nfsclnt_mkkey,
1251 	    TRUE);
1252 	rfs4_clientid_idx = rfs4_index_create(rfs4_client_tab,
1253 	    "client_id", clientid_hash,
1254 	    clientid_compare, clientid_mkkey,
1255 	    FALSE);
1256 
1257 	rfs4_clntip_cache_time = 86400 * 365;	/* about a year */
1258 	rfs4_clntip_tab = rfs4_table_create(rfs4_server_state,
1259 	    "ClntIP",
1260 	    rfs4_clntip_cache_time,
1261 	    1,
1262 	    rfs4_clntip_create,
1263 	    rfs4_clntip_destroy,
1264 	    rfs4_clntip_expiry,
1265 	    sizeof (rfs4_clntip_t),
1266 	    TABSIZE,
1267 	    MAXTABSZ, 100);
1268 	rfs4_clntip_idx = rfs4_index_create(rfs4_clntip_tab,
1269 	    "client_ip", clntip_hash,
1270 	    clntip_compare, clntip_mkkey,
1271 	    TRUE);
1272 
1273 	rfs4_openowner_cache_time *= rfs4_lease_time;
1274 	rfs4_openowner_tab = rfs4_table_create(rfs4_server_state,
1275 	    "OpenOwner",
1276 	    rfs4_openowner_cache_time,
1277 	    1,
1278 	    rfs4_openowner_create,
1279 	    rfs4_openowner_destroy,
1280 	    rfs4_openowner_expiry,
1281 	    sizeof (rfs4_openowner_t),
1282 	    TABSIZE,
1283 	    MAXTABSZ, 100);
1284 	rfs4_openowner_idx = rfs4_index_create(rfs4_openowner_tab,
1285 	    "open_owner4", openowner_hash,
1286 	    openowner_compare,
1287 	    openowner_mkkey, TRUE);
1288 
1289 	rfs4_state_cache_time *= rfs4_lease_time;
1290 	rfs4_state_tab = rfs4_table_create(rfs4_server_state,
1291 	    "OpenStateID",
1292 	    rfs4_state_cache_time,
1293 	    3,
1294 	    rfs4_state_create,
1295 	    rfs4_state_destroy,
1296 	    rfs4_state_expiry,
1297 	    sizeof (rfs4_state_t),
1298 	    TABSIZE,
1299 	    MAXTABSZ, 100);
1300 
1301 	rfs4_state_owner_file_idx = rfs4_index_create(rfs4_state_tab,
1302 	    "Openowner-File",
1303 	    state_owner_file_hash,
1304 	    state_owner_file_compare,
1305 	    state_owner_file_mkkey, TRUE);
1306 
1307 	rfs4_state_idx = rfs4_index_create(rfs4_state_tab,
1308 	    "State-id", state_hash,
1309 	    state_compare, state_mkkey, FALSE);
1310 
1311 	rfs4_state_file_idx = rfs4_index_create(rfs4_state_tab,
1312 	    "File", state_file_hash,
1313 	    state_file_compare, state_file_mkkey,
1314 	    FALSE);
1315 
1316 	rfs4_lo_state_cache_time *= rfs4_lease_time;
1317 	rfs4_lo_state_tab = rfs4_table_create(rfs4_server_state,
1318 	    "LockStateID",
1319 	    rfs4_lo_state_cache_time,
1320 	    2,
1321 	    rfs4_lo_state_create,
1322 	    rfs4_lo_state_destroy,
1323 	    rfs4_lo_state_expiry,
1324 	    sizeof (rfs4_lo_state_t),
1325 	    TABSIZE,
1326 	    MAXTABSZ, 100);
1327 
1328 	rfs4_lo_state_owner_idx = rfs4_index_create(rfs4_lo_state_tab,
1329 	    "lockownerxstate",
1330 	    lo_state_lo_hash,
1331 	    lo_state_lo_compare,
1332 	    lo_state_lo_mkkey, TRUE);
1333 
1334 	rfs4_lo_state_idx = rfs4_index_create(rfs4_lo_state_tab,
1335 	    "State-id",
1336 	    lo_state_hash, lo_state_compare,
1337 	    lo_state_mkkey, FALSE);
1338 
1339 	rfs4_lockowner_cache_time *= rfs4_lease_time;
1340 
1341 	rfs4_lockowner_tab = rfs4_table_create(rfs4_server_state,
1342 	    "Lockowner",
1343 	    rfs4_lockowner_cache_time,
1344 	    2,
1345 	    rfs4_lockowner_create,
1346 	    rfs4_lockowner_destroy,
1347 	    rfs4_lockowner_expiry,
1348 	    sizeof (rfs4_lockowner_t),
1349 	    TABSIZE,
1350 	    MAXTABSZ, 100);
1351 
1352 	rfs4_lockowner_idx = rfs4_index_create(rfs4_lockowner_tab,
1353 	    "lock_owner4", lockowner_hash,
1354 	    lockowner_compare,
1355 	    lockowner_mkkey, TRUE);
1356 
1357 	rfs4_lockowner_pid_idx = rfs4_index_create(rfs4_lockowner_tab,
1358 	    "pid", pid_hash,
1359 	    pid_compare, pid_mkkey,
1360 	    FALSE);
1361 
1362 	rfs4_file_cache_time *= rfs4_lease_time;
1363 	rfs4_file_tab = rfs4_table_create(rfs4_server_state,
1364 	    "File",
1365 	    rfs4_file_cache_time,
1366 	    1,
1367 	    rfs4_file_create,
1368 	    rfs4_file_destroy,
1369 	    NULL,
1370 	    sizeof (rfs4_file_t),
1371 	    TABSIZE,
1372 	    MAXTABSZ, -1);
1373 
1374 	rfs4_file_idx = rfs4_index_create(rfs4_file_tab,
1375 	    "Filehandle", file_hash,
1376 	    file_compare, file_mkkey, TRUE);
1377 
1378 	rfs4_deleg_state_cache_time *= rfs4_lease_time;
1379 	rfs4_deleg_state_tab = rfs4_table_create(rfs4_server_state,
1380 	    "DelegStateID",
1381 	    rfs4_deleg_state_cache_time,
1382 	    2,
1383 	    rfs4_deleg_state_create,
1384 	    rfs4_deleg_state_destroy,
1385 	    rfs4_deleg_state_expiry,
1386 	    sizeof (rfs4_deleg_state_t),
1387 	    TABSIZE,
1388 	    MAXTABSZ, 100);
1389 	rfs4_deleg_idx = rfs4_index_create(rfs4_deleg_state_tab,
1390 	    "DelegByFileClient",
1391 	    deleg_hash,
1392 	    deleg_compare,
1393 	    deleg_mkkey, TRUE);
1394 
1395 	rfs4_deleg_state_idx = rfs4_index_create(rfs4_deleg_state_tab,
1396 	    "DelegState",
1397 	    deleg_state_hash,
1398 	    deleg_state_compare,
1399 	    deleg_state_mkkey, FALSE);
1400 
1401 	/*
1402 	 * Init the stable storage.
1403 	 */
1404 	rfs4_ss_init();
1405 
1406 	rfs4_client_clrst = rfs4_clear_client_state;
1407 
1408 	mutex_exit(&rfs4_state_lock);
1409 }
1410 
1411 
1412 /*
1413  * Used at server shutdown to cleanup all of the NFSv4 server's structures
1414  * and other state.
1415  */
1416 void
1417 rfs4_state_fini()
1418 {
1419 	rfs4_database_t *dbp;
1420 
1421 	mutex_enter(&rfs4_state_lock);
1422 
1423 	if (rfs4_server_state == NULL) {
1424 		mutex_exit(&rfs4_state_lock);
1425 		return;
1426 	}
1427 
1428 	rfs4_client_clrst = NULL;
1429 
1430 	rfs4_set_deleg_policy(SRV_NEVER_DELEGATE);
1431 	dbp = rfs4_server_state;
1432 	rfs4_server_state = NULL;
1433 
1434 	/*
1435 	 * Cleanup the CPR callback.
1436 	 */
1437 	if (cpr_id)
1438 		(void) callb_delete(cpr_id);
1439 
1440 	rw_destroy(&rfs4_findclient_lock);
1441 
1442 	/* First stop all of the reaper threads in the database */
1443 	rfs4_database_shutdown(dbp);
1444 	/* clean up any dangling stable storage structures */
1445 	rfs4_ss_fini();
1446 	/* Now actually destroy/release the database and its tables */
1447 	rfs4_database_destroy(dbp);
1448 
1449 	/* Reset the cache timers for next time */
1450 	rfs4_client_cache_time = 0;
1451 	rfs4_openowner_cache_time = 0;
1452 	rfs4_state_cache_time = 0;
1453 	rfs4_lo_state_cache_time = 0;
1454 	rfs4_lockowner_cache_time = 0;
1455 	rfs4_file_cache_time = 0;
1456 	rfs4_deleg_state_cache_time = 0;
1457 
1458 	mutex_exit(&rfs4_state_lock);
1459 
1460 	/* destroy server instances and current instance ptr */
1461 	rfs4_servinst_destroy_all();
1462 
1463 	/* reset the "first NFSv4 request" status */
1464 	rfs4_seen_first_compound = 0;
1465 
1466 	/* DSS: distributed stable storage */
1467 	nvlist_free(rfs4_dss_oldpaths);
1468 	nvlist_free(rfs4_dss_paths);
1469 	rfs4_dss_paths = rfs4_dss_oldpaths = NULL;
1470 }
1471 
1472 typedef union {
1473 	struct {
1474 		uint32_t start_time;
1475 		uint32_t c_id;
1476 	} impl_id;
1477 	clientid4 id4;
1478 } cid;
1479 
1480 static int foreign_stateid(stateid_t *id);
1481 static int foreign_clientid(cid *cidp);
1482 static void embed_nodeid(cid *cidp);
1483 
1484 typedef union {
1485 	struct {
1486 		uint32_t c_id;
1487 		uint32_t gen_num;
1488 	} cv_impl;
1489 	verifier4	confirm_verf;
1490 } scid_confirm_verf;
1491 
1492 static uint32_t
1493 clientid_hash(void *key)
1494 {
1495 	cid *idp = key;
1496 
1497 	return (idp->impl_id.c_id);
1498 }
1499 
1500 static bool_t
1501 clientid_compare(rfs4_entry_t entry, void *key)
1502 {
1503 	rfs4_client_t *cp = (rfs4_client_t *)entry;
1504 	clientid4 *idp = key;
1505 
1506 	return (*idp == cp->rc_clientid);
1507 }
1508 
1509 static void *
1510 clientid_mkkey(rfs4_entry_t entry)
1511 {
1512 	rfs4_client_t *cp = (rfs4_client_t *)entry;
1513 
1514 	return (&cp->rc_clientid);
1515 }
1516 
1517 static uint32_t
1518 nfsclnt_hash(void *key)
1519 {
1520 	nfs_client_id4 *client = key;
1521 	int i;
1522 	uint32_t hash = 0;
1523 
1524 	for (i = 0; i < client->id_len; i++) {
1525 		hash <<= 1;
1526 		hash += (uint_t)client->id_val[i];
1527 	}
1528 	return (hash);
1529 }
1530 
1531 
1532 static bool_t
1533 nfsclnt_compare(rfs4_entry_t entry, void *key)
1534 {
1535 	rfs4_client_t *cp = (rfs4_client_t *)entry;
1536 	nfs_client_id4 *nfs_client = key;
1537 
1538 	if (cp->rc_nfs_client.id_len != nfs_client->id_len)
1539 		return (FALSE);
1540 
1541 	return (bcmp(cp->rc_nfs_client.id_val, nfs_client->id_val,
1542 	    nfs_client->id_len) == 0);
1543 }
1544 
1545 static void *
1546 nfsclnt_mkkey(rfs4_entry_t entry)
1547 {
1548 	rfs4_client_t *cp = (rfs4_client_t *)entry;
1549 
1550 	return (&cp->rc_nfs_client);
1551 }
1552 
1553 static bool_t
1554 rfs4_client_expiry(rfs4_entry_t u_entry)
1555 {
1556 	rfs4_client_t *cp = (rfs4_client_t *)u_entry;
1557 	bool_t cp_expired;
1558 
1559 	if (rfs4_dbe_is_invalid(cp->rc_dbe)) {
1560 		cp->rc_ss_remove = 1;
1561 		return (TRUE);
1562 	}
1563 	/*
1564 	 * If the sysadmin has used clear_locks for this
1565 	 * entry then forced_expire will be set and we
1566 	 * want this entry to be reaped. Or the entry
1567 	 * has exceeded its lease period.
1568 	 */
1569 	cp_expired = (cp->rc_forced_expire ||
1570 	    (gethrestime_sec() - cp->rc_last_access
1571 	    > rfs4_lease_time));
1572 
1573 	if (!cp->rc_ss_remove && cp_expired)
1574 		cp->rc_ss_remove = 1;
1575 	return (cp_expired);
1576 }
1577 
1578 /*
1579  * Remove the leaf file from all distributed stable storage paths.
1580  */
1581 static void
1582 rfs4_dss_remove_cpleaf(rfs4_client_t *cp)
1583 {
1584 	rfs4_servinst_t *sip;
1585 	char *leaf = cp->rc_ss_pn->leaf;
1586 
1587 	/*
1588 	 * since the state files are written to all DSS
1589 	 * paths we must remove this leaf file instance
1590 	 * from all server instances.
1591 	 */
1592 
1593 	mutex_enter(&rfs4_servinst_lock);
1594 	for (sip = rfs4_cur_servinst; sip != NULL; sip = sip->prev) {
1595 		/* remove the leaf file associated with this server instance */
1596 		rfs4_dss_remove_leaf(sip, NFS4_DSS_STATE_LEAF, leaf);
1597 	}
1598 	mutex_exit(&rfs4_servinst_lock);
1599 }
1600 
1601 static void
1602 rfs4_dss_remove_leaf(rfs4_servinst_t *sip, char *dir_leaf, char *leaf)
1603 {
1604 	int i, npaths = sip->dss_npaths;
1605 
1606 	for (i = 0; i < npaths; i++) {
1607 		rfs4_dss_path_t *dss_path = sip->dss_paths[i];
1608 		char *path, *dir;
1609 		size_t pathlen;
1610 
1611 		/* the HA-NFSv4 path might have been failed-over away from us */
1612 		if (dss_path == NULL)
1613 			continue;
1614 
1615 		dir = dss_path->path;
1616 
1617 		/* allow 3 extra bytes for two '/' & a NUL */
1618 		pathlen = strlen(dir) + strlen(dir_leaf) + strlen(leaf) + 3;
1619 		path = kmem_alloc(pathlen, KM_SLEEP);
1620 		(void) sprintf(path, "%s/%s/%s", dir, dir_leaf, leaf);
1621 
1622 		(void) vn_remove(path, UIO_SYSSPACE, RMFILE);
1623 
1624 		kmem_free(path, pathlen);
1625 	}
1626 }
1627 
1628 static void
1629 rfs4_client_destroy(rfs4_entry_t u_entry)
1630 {
1631 	rfs4_client_t *cp = (rfs4_client_t *)u_entry;
1632 
1633 	mutex_destroy(cp->rc_cbinfo.cb_lock);
1634 	cv_destroy(cp->rc_cbinfo.cb_cv);
1635 	cv_destroy(cp->rc_cbinfo.cb_cv_nullcaller);
1636 	list_destroy(&cp->rc_openownerlist);
1637 
1638 	/* free callback info */
1639 	rfs4_cbinfo_free(&cp->rc_cbinfo);
1640 
1641 	if (cp->rc_cp_confirmed)
1642 		rfs4_client_rele(cp->rc_cp_confirmed);
1643 
1644 	if (cp->rc_ss_pn) {
1645 		/* check if the stable storage files need to be removed */
1646 		if (cp->rc_ss_remove)
1647 			rfs4_dss_remove_cpleaf(cp);
1648 		rfs4_ss_pnfree(cp->rc_ss_pn);
1649 	}
1650 
1651 	/* Free the client supplied client id */
1652 	kmem_free(cp->rc_nfs_client.id_val, cp->rc_nfs_client.id_len);
1653 
1654 	if (cp->rc_sysidt != LM_NOSYSID)
1655 		lm_free_sysidt(cp->rc_sysidt);
1656 }
1657 
1658 static bool_t
1659 rfs4_client_create(rfs4_entry_t u_entry, void *arg)
1660 {
1661 	rfs4_client_t *cp = (rfs4_client_t *)u_entry;
1662 	nfs_client_id4 *client = (nfs_client_id4 *)arg;
1663 	struct sockaddr *ca;
1664 	cid *cidp;
1665 	scid_confirm_verf *scvp;
1666 
1667 	/* Get a clientid to give to the client */
1668 	cidp = (cid *)&cp->rc_clientid;
1669 	cidp->impl_id.start_time = rfs4_start_time;
1670 	cidp->impl_id.c_id = (uint32_t)rfs4_dbe_getid(cp->rc_dbe);
1671 
1672 	/* If we are booted as a cluster node, embed our nodeid */
1673 	if (cluster_bootflags & CLUSTER_BOOTED)
1674 		embed_nodeid(cidp);
1675 
1676 	/* Allocate and copy client's client id value */
1677 	cp->rc_nfs_client.id_val = kmem_alloc(client->id_len, KM_SLEEP);
1678 	cp->rc_nfs_client.id_len = client->id_len;
1679 	bcopy(client->id_val, cp->rc_nfs_client.id_val, client->id_len);
1680 	cp->rc_nfs_client.verifier = client->verifier;
1681 
1682 	/* Copy client's IP address */
1683 	ca = client->cl_addr;
1684 	if (ca->sa_family == AF_INET)
1685 		bcopy(ca, &cp->rc_addr, sizeof (struct sockaddr_in));
1686 	else if (ca->sa_family == AF_INET6)
1687 		bcopy(ca, &cp->rc_addr, sizeof (struct sockaddr_in6));
1688 	cp->rc_nfs_client.cl_addr = (struct sockaddr *)&cp->rc_addr;
1689 
1690 	/* Init the value for the SETCLIENTID_CONFIRM verifier */
1691 	scvp = (scid_confirm_verf *)&cp->rc_confirm_verf;
1692 	scvp->cv_impl.c_id = cidp->impl_id.c_id;
1693 	scvp->cv_impl.gen_num = 0;
1694 
1695 	/* An F_UNLKSYS has been done for this client */
1696 	cp->rc_unlksys_completed = FALSE;
1697 
1698 	/* We need the client to ack us */
1699 	cp->rc_need_confirm = TRUE;
1700 	cp->rc_cp_confirmed = NULL;
1701 
1702 	/* TRUE all the time until the callback path actually fails */
1703 	cp->rc_cbinfo.cb_notified_of_cb_path_down = TRUE;
1704 
1705 	/* Initialize the access time to now */
1706 	cp->rc_last_access = gethrestime_sec();
1707 
1708 	cp->rc_cr_set = NULL;
1709 
1710 	cp->rc_sysidt = LM_NOSYSID;
1711 
1712 	list_create(&cp->rc_openownerlist, sizeof (rfs4_openowner_t),
1713 	    offsetof(rfs4_openowner_t, ro_node));
1714 
1715 	/* set up the callback control structure */
1716 	cp->rc_cbinfo.cb_state = CB_UNINIT;
1717 	mutex_init(cp->rc_cbinfo.cb_lock, NULL, MUTEX_DEFAULT, NULL);
1718 	cv_init(cp->rc_cbinfo.cb_cv, NULL, CV_DEFAULT, NULL);
1719 	cv_init(cp->rc_cbinfo.cb_cv_nullcaller, NULL, CV_DEFAULT, NULL);
1720 
1721 	/*
1722 	 * Associate the client_t with the current server instance.
1723 	 * The hold is solely to satisfy the calling requirement of
1724 	 * rfs4_servinst_assign(). In this case it's not strictly necessary.
1725 	 */
1726 	rfs4_dbe_hold(cp->rc_dbe);
1727 	rfs4_servinst_assign(cp, rfs4_cur_servinst);
1728 	rfs4_dbe_rele(cp->rc_dbe);
1729 
1730 	return (TRUE);
1731 }
1732 
1733 /*
1734  * Caller wants to generate/update the setclientid_confirm verifier
1735  * associated with a client.  This is done during the SETCLIENTID
1736  * processing.
1737  */
1738 void
1739 rfs4_client_scv_next(rfs4_client_t *cp)
1740 {
1741 	scid_confirm_verf *scvp;
1742 
1743 	/* Init the value for the SETCLIENTID_CONFIRM verifier */
1744 	scvp = (scid_confirm_verf *)&cp->rc_confirm_verf;
1745 	scvp->cv_impl.gen_num++;
1746 }
1747 
1748 void
1749 rfs4_client_rele(rfs4_client_t *cp)
1750 {
1751 	rfs4_dbe_rele(cp->rc_dbe);
1752 }
1753 
1754 rfs4_client_t *
1755 rfs4_findclient(nfs_client_id4 *client, bool_t *create,	rfs4_client_t *oldcp)
1756 {
1757 	rfs4_client_t *cp;
1758 
1759 
1760 	if (oldcp) {
1761 		rw_enter(&rfs4_findclient_lock, RW_WRITER);
1762 		rfs4_dbe_hide(oldcp->rc_dbe);
1763 	} else {
1764 		rw_enter(&rfs4_findclient_lock, RW_READER);
1765 	}
1766 
1767 	cp = (rfs4_client_t *)rfs4_dbsearch(rfs4_nfsclnt_idx, client,
1768 	    create, (void *)client, RFS4_DBS_VALID);
1769 
1770 	if (oldcp)
1771 		rfs4_dbe_unhide(oldcp->rc_dbe);
1772 
1773 	rw_exit(&rfs4_findclient_lock);
1774 
1775 	return (cp);
1776 }
1777 
1778 rfs4_client_t *
1779 rfs4_findclient_by_id(clientid4 clientid, bool_t find_unconfirmed)
1780 {
1781 	rfs4_client_t *cp;
1782 	bool_t create = FALSE;
1783 	cid *cidp = (cid *)&clientid;
1784 
1785 	/* If we're a cluster and the nodeid isn't right, short-circuit */
1786 	if (cluster_bootflags & CLUSTER_BOOTED && foreign_clientid(cidp))
1787 		return (NULL);
1788 
1789 	rw_enter(&rfs4_findclient_lock, RW_READER);
1790 
1791 	cp = (rfs4_client_t *)rfs4_dbsearch(rfs4_clientid_idx, &clientid,
1792 	    &create, NULL, RFS4_DBS_VALID);
1793 
1794 	rw_exit(&rfs4_findclient_lock);
1795 
1796 	if (cp && cp->rc_need_confirm && find_unconfirmed == FALSE) {
1797 		rfs4_client_rele(cp);
1798 		return (NULL);
1799 	} else {
1800 		return (cp);
1801 	}
1802 }
1803 
1804 static uint32_t
1805 clntip_hash(void *key)
1806 {
1807 	struct sockaddr *addr = key;
1808 	int i, len = 0;
1809 	uint32_t hash = 0;
1810 	char *ptr;
1811 
1812 	if (addr->sa_family == AF_INET) {
1813 		struct sockaddr_in *a = (struct sockaddr_in *)addr;
1814 		len = sizeof (struct in_addr);
1815 		ptr = (char *)&a->sin_addr;
1816 	} else if (addr->sa_family == AF_INET6) {
1817 		struct sockaddr_in6 *a = (struct sockaddr_in6 *)addr;
1818 		len = sizeof (struct in6_addr);
1819 		ptr = (char *)&a->sin6_addr;
1820 	} else
1821 		return (0);
1822 
1823 	for (i = 0; i < len; i++) {
1824 		hash <<= 1;
1825 		hash += (uint_t)ptr[i];
1826 	}
1827 	return (hash);
1828 }
1829 
1830 static bool_t
1831 clntip_compare(rfs4_entry_t entry, void *key)
1832 {
1833 	rfs4_clntip_t *cp = (rfs4_clntip_t *)entry;
1834 	struct sockaddr *addr = key;
1835 	int len = 0;
1836 	char *p1, *p2;
1837 
1838 	if (addr->sa_family == AF_INET) {
1839 		struct sockaddr_in *a1 = (struct sockaddr_in *)&cp->ri_addr;
1840 		struct sockaddr_in *a2 = (struct sockaddr_in *)addr;
1841 		len = sizeof (struct in_addr);
1842 		p1 = (char *)&a1->sin_addr;
1843 		p2 = (char *)&a2->sin_addr;
1844 	} else if (addr->sa_family == AF_INET6) {
1845 		struct sockaddr_in6 *a1 = (struct sockaddr_in6 *)&cp->ri_addr;
1846 		struct sockaddr_in6 *a2 = (struct sockaddr_in6 *)addr;
1847 		len = sizeof (struct in6_addr);
1848 		p1 = (char *)&a1->sin6_addr;
1849 		p2 = (char *)&a2->sin6_addr;
1850 	} else
1851 		return (0);
1852 
1853 	return (bcmp(p1, p2, len) == 0);
1854 }
1855 
1856 static void *
1857 clntip_mkkey(rfs4_entry_t entry)
1858 {
1859 	rfs4_clntip_t *cp = (rfs4_clntip_t *)entry;
1860 
1861 	return (&cp->ri_addr);
1862 }
1863 
1864 static bool_t
1865 rfs4_clntip_expiry(rfs4_entry_t u_entry)
1866 {
1867 	rfs4_clntip_t *cp = (rfs4_clntip_t *)u_entry;
1868 
1869 	if (rfs4_dbe_is_invalid(cp->ri_dbe))
1870 		return (TRUE);
1871 	return (FALSE);
1872 }
1873 
1874 /* ARGSUSED */
1875 static void
1876 rfs4_clntip_destroy(rfs4_entry_t u_entry)
1877 {
1878 }
1879 
1880 static bool_t
1881 rfs4_clntip_create(rfs4_entry_t u_entry, void *arg)
1882 {
1883 	rfs4_clntip_t *cp = (rfs4_clntip_t *)u_entry;
1884 	struct sockaddr *ca = (struct sockaddr *)arg;
1885 
1886 	/* Copy client's IP address */
1887 	if (ca->sa_family == AF_INET)
1888 		bcopy(ca, &cp->ri_addr, sizeof (struct sockaddr_in));
1889 	else if (ca->sa_family == AF_INET6)
1890 		bcopy(ca, &cp->ri_addr, sizeof (struct sockaddr_in6));
1891 	else
1892 		return (FALSE);
1893 	cp->ri_no_referrals = 1;
1894 
1895 	return (TRUE);
1896 }
1897 
1898 rfs4_clntip_t *
1899 rfs4_find_clntip(struct sockaddr *addr, bool_t *create)
1900 {
1901 	rfs4_clntip_t *cp;
1902 
1903 	rw_enter(&rfs4_findclient_lock, RW_READER);
1904 
1905 	cp = (rfs4_clntip_t *)rfs4_dbsearch(rfs4_clntip_idx, addr,
1906 	    create, addr, RFS4_DBS_VALID);
1907 
1908 	rw_exit(&rfs4_findclient_lock);
1909 
1910 	return (cp);
1911 }
1912 
1913 void
1914 rfs4_invalidate_clntip(struct sockaddr *addr)
1915 {
1916 	rfs4_clntip_t *cp;
1917 	bool_t create = FALSE;
1918 
1919 	rw_enter(&rfs4_findclient_lock, RW_READER);
1920 
1921 	cp = (rfs4_clntip_t *)rfs4_dbsearch(rfs4_clntip_idx, addr,
1922 	    &create, NULL, RFS4_DBS_VALID);
1923 	if (cp == NULL) {
1924 		rw_exit(&rfs4_findclient_lock);
1925 		return;
1926 	}
1927 	rfs4_dbe_invalidate(cp->ri_dbe);
1928 	rfs4_dbe_rele(cp->ri_dbe);
1929 
1930 	rw_exit(&rfs4_findclient_lock);
1931 }
1932 
1933 bool_t
1934 rfs4_lease_expired(rfs4_client_t *cp)
1935 {
1936 	bool_t rc;
1937 
1938 	rfs4_dbe_lock(cp->rc_dbe);
1939 
1940 	/*
1941 	 * If the admin has executed clear_locks for this
1942 	 * client id, force expire will be set, so no need
1943 	 * to calculate anything because it's "outa here".
1944 	 */
1945 	if (cp->rc_forced_expire) {
1946 		rc = TRUE;
1947 	} else {
1948 		rc = (gethrestime_sec() - cp->rc_last_access > rfs4_lease_time);
1949 	}
1950 
1951 	/*
1952 	 * If the lease has expired we will also want
1953 	 * to remove any stable storage state data. So
1954 	 * mark the client id accordingly.
1955 	 */
1956 	if (!cp->rc_ss_remove)
1957 		cp->rc_ss_remove = (rc == TRUE);
1958 
1959 	rfs4_dbe_unlock(cp->rc_dbe);
1960 
1961 	return (rc);
1962 }
1963 
1964 void
1965 rfs4_update_lease(rfs4_client_t *cp)
1966 {
1967 	rfs4_dbe_lock(cp->rc_dbe);
1968 	if (!cp->rc_forced_expire)
1969 		cp->rc_last_access = gethrestime_sec();
1970 	rfs4_dbe_unlock(cp->rc_dbe);
1971 }
1972 
1973 
1974 static bool_t
1975 EQOPENOWNER(open_owner4 *a, open_owner4 *b)
1976 {
1977 	bool_t rc;
1978 
1979 	if (a->clientid != b->clientid)
1980 		return (FALSE);
1981 
1982 	if (a->owner_len != b->owner_len)
1983 		return (FALSE);
1984 
1985 	rc = (bcmp(a->owner_val, b->owner_val, a->owner_len) == 0);
1986 
1987 	return (rc);
1988 }
1989 
1990 static uint_t
1991 openowner_hash(void *key)
1992 {
1993 	int i;
1994 	open_owner4 *openowner = key;
1995 	uint_t hash = 0;
1996 
1997 	for (i = 0; i < openowner->owner_len; i++) {
1998 		hash <<= 4;
1999 		hash += (uint_t)openowner->owner_val[i];
2000 	}
2001 	hash += (uint_t)openowner->clientid;
2002 	hash |= (openowner->clientid >> 32);
2003 
2004 	return (hash);
2005 }
2006 
2007 static bool_t
2008 openowner_compare(rfs4_entry_t u_entry, void *key)
2009 {
2010 	rfs4_openowner_t *oo = (rfs4_openowner_t *)u_entry;
2011 	open_owner4 *arg = key;
2012 
2013 	return (EQOPENOWNER(&oo->ro_owner, arg));
2014 }
2015 
2016 void *
2017 openowner_mkkey(rfs4_entry_t u_entry)
2018 {
2019 	rfs4_openowner_t *oo = (rfs4_openowner_t *)u_entry;
2020 
2021 	return (&oo->ro_owner);
2022 }
2023 
2024 /* ARGSUSED */
2025 static bool_t
2026 rfs4_openowner_expiry(rfs4_entry_t u_entry)
2027 {
2028 	/* openstateid held us and did all needed delay */
2029 	return (TRUE);
2030 }
2031 
2032 static void
2033 rfs4_openowner_destroy(rfs4_entry_t u_entry)
2034 {
2035 	rfs4_openowner_t *oo = (rfs4_openowner_t *)u_entry;
2036 
2037 	/* Remove open owner from client's lists of open owners */
2038 	rfs4_dbe_lock(oo->ro_client->rc_dbe);
2039 	list_remove(&oo->ro_client->rc_openownerlist, oo);
2040 	rfs4_dbe_unlock(oo->ro_client->rc_dbe);
2041 
2042 	/* One less reference to the client */
2043 	rfs4_client_rele(oo->ro_client);
2044 	oo->ro_client = NULL;
2045 
2046 	/* Free the last reply for this lock owner */
2047 	rfs4_free_reply(&oo->ro_reply);
2048 
2049 	if (oo->ro_reply_fh.nfs_fh4_val) {
2050 		kmem_free(oo->ro_reply_fh.nfs_fh4_val,
2051 		    oo->ro_reply_fh.nfs_fh4_len);
2052 		oo->ro_reply_fh.nfs_fh4_val = NULL;
2053 		oo->ro_reply_fh.nfs_fh4_len = 0;
2054 	}
2055 
2056 	rfs4_sw_destroy(&oo->ro_sw);
2057 	list_destroy(&oo->ro_statelist);
2058 
2059 	/* Free the lock owner id */
2060 	kmem_free(oo->ro_owner.owner_val, oo->ro_owner.owner_len);
2061 }
2062 
2063 void
2064 rfs4_openowner_rele(rfs4_openowner_t *oo)
2065 {
2066 	rfs4_dbe_rele(oo->ro_dbe);
2067 }
2068 
2069 static bool_t
2070 rfs4_openowner_create(rfs4_entry_t u_entry, void *arg)
2071 {
2072 	rfs4_openowner_t *oo = (rfs4_openowner_t *)u_entry;
2073 	rfs4_openowner_t *argp = (rfs4_openowner_t *)arg;
2074 	open_owner4 *openowner = &argp->ro_owner;
2075 	seqid4 seqid = argp->ro_open_seqid;
2076 	rfs4_client_t *cp;
2077 	bool_t create = FALSE;
2078 
2079 	rw_enter(&rfs4_findclient_lock, RW_READER);
2080 
2081 	cp = (rfs4_client_t *)rfs4_dbsearch(rfs4_clientid_idx,
2082 	    &openowner->clientid,
2083 	    &create, NULL, RFS4_DBS_VALID);
2084 
2085 	rw_exit(&rfs4_findclient_lock);
2086 
2087 	if (cp == NULL)
2088 		return (FALSE);
2089 
2090 	oo->ro_reply_fh.nfs_fh4_len = 0;
2091 	oo->ro_reply_fh.nfs_fh4_val = NULL;
2092 
2093 	oo->ro_owner.clientid = openowner->clientid;
2094 	oo->ro_owner.owner_val =
2095 	    kmem_alloc(openowner->owner_len, KM_SLEEP);
2096 
2097 	bcopy(openowner->owner_val,
2098 	    oo->ro_owner.owner_val, openowner->owner_len);
2099 
2100 	oo->ro_owner.owner_len = openowner->owner_len;
2101 
2102 	oo->ro_need_confirm = TRUE;
2103 
2104 	rfs4_sw_init(&oo->ro_sw);
2105 
2106 	oo->ro_open_seqid = seqid;
2107 	bzero(&oo->ro_reply, sizeof (nfs_resop4));
2108 	oo->ro_client = cp;
2109 	oo->ro_cr_set = NULL;
2110 
2111 	list_create(&oo->ro_statelist, sizeof (rfs4_state_t),
2112 	    offsetof(rfs4_state_t, rs_node));
2113 
2114 	/* Insert openowner into client's open owner list */
2115 	rfs4_dbe_lock(cp->rc_dbe);
2116 	list_insert_tail(&cp->rc_openownerlist, oo);
2117 	rfs4_dbe_unlock(cp->rc_dbe);
2118 
2119 	return (TRUE);
2120 }
2121 
2122 rfs4_openowner_t *
2123 rfs4_findopenowner(open_owner4 *openowner, bool_t *create, seqid4 seqid)
2124 {
2125 	rfs4_openowner_t *oo;
2126 	rfs4_openowner_t arg;
2127 
2128 	arg.ro_owner = *openowner;
2129 	arg.ro_open_seqid = seqid;
2130 	oo = (rfs4_openowner_t *)rfs4_dbsearch(rfs4_openowner_idx, openowner,
2131 	    create, &arg, RFS4_DBS_VALID);
2132 
2133 	return (oo);
2134 }
2135 
2136 void
2137 rfs4_update_open_sequence(rfs4_openowner_t *oo)
2138 {
2139 
2140 	rfs4_dbe_lock(oo->ro_dbe);
2141 
2142 	oo->ro_open_seqid++;
2143 
2144 	rfs4_dbe_unlock(oo->ro_dbe);
2145 }
2146 
2147 void
2148 rfs4_update_open_resp(rfs4_openowner_t *oo, nfs_resop4 *resp, nfs_fh4 *fh)
2149 {
2150 
2151 	rfs4_dbe_lock(oo->ro_dbe);
2152 
2153 	rfs4_free_reply(&oo->ro_reply);
2154 
2155 	rfs4_copy_reply(&oo->ro_reply, resp);
2156 
2157 	/* Save the filehandle if provided and free if not used */
2158 	if (resp->nfs_resop4_u.opopen.status == NFS4_OK &&
2159 	    fh && fh->nfs_fh4_len) {
2160 		if (oo->ro_reply_fh.nfs_fh4_val == NULL)
2161 			oo->ro_reply_fh.nfs_fh4_val =
2162 			    kmem_alloc(fh->nfs_fh4_len, KM_SLEEP);
2163 		nfs_fh4_copy(fh, &oo->ro_reply_fh);
2164 	} else {
2165 		if (oo->ro_reply_fh.nfs_fh4_val) {
2166 			kmem_free(oo->ro_reply_fh.nfs_fh4_val,
2167 			    oo->ro_reply_fh.nfs_fh4_len);
2168 			oo->ro_reply_fh.nfs_fh4_val = NULL;
2169 			oo->ro_reply_fh.nfs_fh4_len = 0;
2170 		}
2171 	}
2172 
2173 	rfs4_dbe_unlock(oo->ro_dbe);
2174 }
2175 
2176 static bool_t
2177 lockowner_compare(rfs4_entry_t u_entry, void *key)
2178 {
2179 	rfs4_lockowner_t *lo = (rfs4_lockowner_t *)u_entry;
2180 	lock_owner4 *b = (lock_owner4 *)key;
2181 
2182 	if (lo->rl_owner.clientid != b->clientid)
2183 		return (FALSE);
2184 
2185 	if (lo->rl_owner.owner_len != b->owner_len)
2186 		return (FALSE);
2187 
2188 	return (bcmp(lo->rl_owner.owner_val, b->owner_val,
2189 	    lo->rl_owner.owner_len) == 0);
2190 }
2191 
2192 void *
2193 lockowner_mkkey(rfs4_entry_t u_entry)
2194 {
2195 	rfs4_lockowner_t *lo = (rfs4_lockowner_t *)u_entry;
2196 
2197 	return (&lo->rl_owner);
2198 }
2199 
2200 static uint32_t
2201 lockowner_hash(void *key)
2202 {
2203 	int i;
2204 	lock_owner4 *lockowner = key;
2205 	uint_t hash = 0;
2206 
2207 	for (i = 0; i < lockowner->owner_len; i++) {
2208 		hash <<= 4;
2209 		hash += (uint_t)lockowner->owner_val[i];
2210 	}
2211 	hash += (uint_t)lockowner->clientid;
2212 	hash |= (lockowner->clientid >> 32);
2213 
2214 	return (hash);
2215 }
2216 
2217 static uint32_t
2218 pid_hash(void *key)
2219 {
2220 	return ((uint32_t)(uintptr_t)key);
2221 }
2222 
2223 static void *
2224 pid_mkkey(rfs4_entry_t u_entry)
2225 {
2226 	rfs4_lockowner_t *lo = (rfs4_lockowner_t *)u_entry;
2227 
2228 	return ((void *)(uintptr_t)lo->rl_pid);
2229 }
2230 
2231 static bool_t
2232 pid_compare(rfs4_entry_t u_entry, void *key)
2233 {
2234 	rfs4_lockowner_t *lo = (rfs4_lockowner_t *)u_entry;
2235 
2236 	return (lo->rl_pid == (pid_t)(uintptr_t)key);
2237 }
2238 
2239 static void
2240 rfs4_lockowner_destroy(rfs4_entry_t u_entry)
2241 {
2242 	rfs4_lockowner_t *lo = (rfs4_lockowner_t *)u_entry;
2243 
2244 	/* Free the lock owner id */
2245 	kmem_free(lo->rl_owner.owner_val, lo->rl_owner.owner_len);
2246 	rfs4_client_rele(lo->rl_client);
2247 }
2248 
2249 void
2250 rfs4_lockowner_rele(rfs4_lockowner_t *lo)
2251 {
2252 	rfs4_dbe_rele(lo->rl_dbe);
2253 }
2254 
2255 /* ARGSUSED */
2256 static bool_t
2257 rfs4_lockowner_expiry(rfs4_entry_t u_entry)
2258 {
2259 	/*
2260 	 * Since expiry is called with no other references on
2261 	 * this struct, go ahead and have it removed.
2262 	 */
2263 	return (TRUE);
2264 }
2265 
2266 static bool_t
2267 rfs4_lockowner_create(rfs4_entry_t u_entry, void *arg)
2268 {
2269 	rfs4_lockowner_t *lo = (rfs4_lockowner_t *)u_entry;
2270 	lock_owner4 *lockowner = (lock_owner4 *)arg;
2271 	rfs4_client_t *cp;
2272 	bool_t create = FALSE;
2273 
2274 	rw_enter(&rfs4_findclient_lock, RW_READER);
2275 
2276 	cp = (rfs4_client_t *)rfs4_dbsearch(rfs4_clientid_idx,
2277 	    &lockowner->clientid,
2278 	    &create, NULL, RFS4_DBS_VALID);
2279 
2280 	rw_exit(&rfs4_findclient_lock);
2281 
2282 	if (cp == NULL)
2283 		return (FALSE);
2284 
2285 	/* Reference client */
2286 	lo->rl_client = cp;
2287 	lo->rl_owner.clientid = lockowner->clientid;
2288 	lo->rl_owner.owner_val = kmem_alloc(lockowner->owner_len, KM_SLEEP);
2289 	bcopy(lockowner->owner_val, lo->rl_owner.owner_val,
2290 	    lockowner->owner_len);
2291 	lo->rl_owner.owner_len = lockowner->owner_len;
2292 	lo->rl_pid = rfs4_dbe_getid(lo->rl_dbe);
2293 
2294 	return (TRUE);
2295 }
2296 
2297 rfs4_lockowner_t *
2298 rfs4_findlockowner(lock_owner4 *lockowner, bool_t *create)
2299 {
2300 	rfs4_lockowner_t *lo;
2301 
2302 	lo = (rfs4_lockowner_t *)rfs4_dbsearch(rfs4_lockowner_idx, lockowner,
2303 	    create, lockowner, RFS4_DBS_VALID);
2304 
2305 	return (lo);
2306 }
2307 
2308 rfs4_lockowner_t *
2309 rfs4_findlockowner_by_pid(pid_t pid)
2310 {
2311 	rfs4_lockowner_t *lo;
2312 	bool_t create = FALSE;
2313 
2314 	lo = (rfs4_lockowner_t *)rfs4_dbsearch(rfs4_lockowner_pid_idx,
2315 	    (void *)(uintptr_t)pid, &create, NULL, RFS4_DBS_VALID);
2316 
2317 	return (lo);
2318 }
2319 
2320 
2321 static uint32_t
2322 file_hash(void *key)
2323 {
2324 	return (ADDRHASH(key));
2325 }
2326 
2327 static void *
2328 file_mkkey(rfs4_entry_t u_entry)
2329 {
2330 	rfs4_file_t *fp = (rfs4_file_t *)u_entry;
2331 
2332 	return (fp->rf_vp);
2333 }
2334 
2335 static bool_t
2336 file_compare(rfs4_entry_t u_entry, void *key)
2337 {
2338 	rfs4_file_t *fp = (rfs4_file_t *)u_entry;
2339 
2340 	return (fp->rf_vp == (vnode_t *)key);
2341 }
2342 
2343 static void
2344 rfs4_file_destroy(rfs4_entry_t u_entry)
2345 {
2346 	rfs4_file_t *fp = (rfs4_file_t *)u_entry;
2347 
2348 	list_destroy(&fp->rf_delegstatelist);
2349 
2350 	if (fp->rf_filehandle.nfs_fh4_val)
2351 		kmem_free(fp->rf_filehandle.nfs_fh4_val,
2352 		    fp->rf_filehandle.nfs_fh4_len);
2353 	cv_destroy(fp->rf_dinfo.rd_recall_cv);
2354 	if (fp->rf_vp) {
2355 		vnode_t *vp = fp->rf_vp;
2356 
2357 		mutex_enter(&vp->v_vsd_lock);
2358 		(void) vsd_set(vp, nfs4_srv_vkey, NULL);
2359 		mutex_exit(&vp->v_vsd_lock);
2360 		VN_RELE(vp);
2361 		fp->rf_vp = NULL;
2362 	}
2363 	rw_destroy(&fp->rf_file_rwlock);
2364 }
2365 
2366 /*
2367  * Used to unlock the underlying dbe struct only
2368  */
2369 void
2370 rfs4_file_rele(rfs4_file_t *fp)
2371 {
2372 	rfs4_dbe_rele(fp->rf_dbe);
2373 }
2374 
2375 typedef struct {
2376     vnode_t *vp;
2377     nfs_fh4 *fh;
2378 } rfs4_fcreate_arg;
2379 
2380 static bool_t
2381 rfs4_file_create(rfs4_entry_t u_entry, void *arg)
2382 {
2383 	rfs4_file_t *fp = (rfs4_file_t *)u_entry;
2384 	rfs4_fcreate_arg *ap = (rfs4_fcreate_arg *)arg;
2385 	vnode_t *vp = ap->vp;
2386 	nfs_fh4 *fh = ap->fh;
2387 
2388 	VN_HOLD(vp);
2389 
2390 	fp->rf_filehandle.nfs_fh4_len = 0;
2391 	fp->rf_filehandle.nfs_fh4_val = NULL;
2392 	ASSERT(fh && fh->nfs_fh4_len);
2393 	if (fh && fh->nfs_fh4_len) {
2394 		fp->rf_filehandle.nfs_fh4_val =
2395 		    kmem_alloc(fh->nfs_fh4_len, KM_SLEEP);
2396 		nfs_fh4_copy(fh, &fp->rf_filehandle);
2397 	}
2398 	fp->rf_vp = vp;
2399 
2400 	list_create(&fp->rf_delegstatelist, sizeof (rfs4_deleg_state_t),
2401 	    offsetof(rfs4_deleg_state_t, rds_node));
2402 
2403 	fp->rf_share_deny = fp->rf_share_access = fp->rf_access_read = 0;
2404 	fp->rf_access_write = fp->rf_deny_read = fp->rf_deny_write = 0;
2405 
2406 	mutex_init(fp->rf_dinfo.rd_recall_lock, NULL, MUTEX_DEFAULT, NULL);
2407 	cv_init(fp->rf_dinfo.rd_recall_cv, NULL, CV_DEFAULT, NULL);
2408 
2409 	fp->rf_dinfo.rd_dtype = OPEN_DELEGATE_NONE;
2410 
2411 	rw_init(&fp->rf_file_rwlock, NULL, RW_DEFAULT, NULL);
2412 
2413 	mutex_enter(&vp->v_vsd_lock);
2414 	VERIFY(vsd_set(vp, nfs4_srv_vkey, (void *)fp) == 0);
2415 	mutex_exit(&vp->v_vsd_lock);
2416 
2417 	return (TRUE);
2418 }
2419 
2420 rfs4_file_t *
2421 rfs4_findfile(vnode_t *vp, nfs_fh4 *fh, bool_t *create)
2422 {
2423 	rfs4_file_t *fp;
2424 	rfs4_fcreate_arg arg;
2425 
2426 	arg.vp = vp;
2427 	arg.fh = fh;
2428 
2429 	if (*create == TRUE)
2430 		fp = (rfs4_file_t *)rfs4_dbsearch(rfs4_file_idx, vp, create,
2431 		    &arg, RFS4_DBS_VALID);
2432 	else {
2433 		mutex_enter(&vp->v_vsd_lock);
2434 		fp = (rfs4_file_t *)vsd_get(vp, nfs4_srv_vkey);
2435 		if (fp) {
2436 			rfs4_dbe_lock(fp->rf_dbe);
2437 			if (rfs4_dbe_is_invalid(fp->rf_dbe) ||
2438 			    (rfs4_dbe_refcnt(fp->rf_dbe) == 0)) {
2439 				rfs4_dbe_unlock(fp->rf_dbe);
2440 				fp = NULL;
2441 			} else {
2442 				rfs4_dbe_hold(fp->rf_dbe);
2443 				rfs4_dbe_unlock(fp->rf_dbe);
2444 			}
2445 		}
2446 		mutex_exit(&vp->v_vsd_lock);
2447 	}
2448 	return (fp);
2449 }
2450 
2451 /*
2452  * Find a file in the db and once it is located, take the rw lock.
2453  * Need to check the vnode pointer and if it does not exist (it was
2454  * removed between the db location and check) redo the find.  This
2455  * assumes that a file struct that has a NULL vnode pointer is marked
2456  * at 'invalid' and will not be found in the db the second time
2457  * around.
2458  */
2459 rfs4_file_t *
2460 rfs4_findfile_withlock(vnode_t *vp, nfs_fh4 *fh, bool_t *create)
2461 {
2462 	rfs4_file_t *fp;
2463 	rfs4_fcreate_arg arg;
2464 	bool_t screate = *create;
2465 
2466 	if (screate == FALSE) {
2467 		mutex_enter(&vp->v_vsd_lock);
2468 		fp = (rfs4_file_t *)vsd_get(vp, nfs4_srv_vkey);
2469 		if (fp) {
2470 			rfs4_dbe_lock(fp->rf_dbe);
2471 			if (rfs4_dbe_is_invalid(fp->rf_dbe) ||
2472 			    (rfs4_dbe_refcnt(fp->rf_dbe) == 0)) {
2473 				rfs4_dbe_unlock(fp->rf_dbe);
2474 				mutex_exit(&vp->v_vsd_lock);
2475 				fp = NULL;
2476 			} else {
2477 				rfs4_dbe_hold(fp->rf_dbe);
2478 				rfs4_dbe_unlock(fp->rf_dbe);
2479 				mutex_exit(&vp->v_vsd_lock);
2480 				rw_enter(&fp->rf_file_rwlock, RW_WRITER);
2481 				if (fp->rf_vp == NULL) {
2482 					rw_exit(&fp->rf_file_rwlock);
2483 					rfs4_file_rele(fp);
2484 					fp = NULL;
2485 				}
2486 			}
2487 		} else {
2488 			mutex_exit(&vp->v_vsd_lock);
2489 		}
2490 	} else {
2491 retry:
2492 		arg.vp = vp;
2493 		arg.fh = fh;
2494 
2495 		fp = (rfs4_file_t *)rfs4_dbsearch(rfs4_file_idx, vp, create,
2496 		    &arg, RFS4_DBS_VALID);
2497 		if (fp != NULL) {
2498 			rw_enter(&fp->rf_file_rwlock, RW_WRITER);
2499 			if (fp->rf_vp == NULL) {
2500 				rw_exit(&fp->rf_file_rwlock);
2501 				rfs4_file_rele(fp);
2502 				*create = screate;
2503 				goto retry;
2504 			}
2505 		}
2506 	}
2507 
2508 	return (fp);
2509 }
2510 
2511 static uint32_t
2512 lo_state_hash(void *key)
2513 {
2514 	stateid_t *id = key;
2515 
2516 	return (id->bits.ident+id->bits.pid);
2517 }
2518 
2519 static bool_t
2520 lo_state_compare(rfs4_entry_t u_entry, void *key)
2521 {
2522 	rfs4_lo_state_t *lsp = (rfs4_lo_state_t *)u_entry;
2523 	stateid_t *id = key;
2524 	bool_t rc;
2525 
2526 	rc = (lsp->rls_lockid.bits.boottime == id->bits.boottime &&
2527 	    lsp->rls_lockid.bits.type == id->bits.type &&
2528 	    lsp->rls_lockid.bits.ident == id->bits.ident &&
2529 	    lsp->rls_lockid.bits.pid == id->bits.pid);
2530 
2531 	return (rc);
2532 }
2533 
2534 static void *
2535 lo_state_mkkey(rfs4_entry_t u_entry)
2536 {
2537 	rfs4_lo_state_t *lsp = (rfs4_lo_state_t *)u_entry;
2538 
2539 	return (&lsp->rls_lockid);
2540 }
2541 
2542 static bool_t
2543 rfs4_lo_state_expiry(rfs4_entry_t u_entry)
2544 {
2545 	rfs4_lo_state_t *lsp = (rfs4_lo_state_t *)u_entry;
2546 
2547 	if (rfs4_dbe_is_invalid(lsp->rls_dbe))
2548 		return (TRUE);
2549 	if (lsp->rls_state->rs_closed)
2550 		return (TRUE);
2551 	return ((gethrestime_sec() -
2552 	    lsp->rls_state->rs_owner->ro_client->rc_last_access
2553 	    > rfs4_lease_time));
2554 }
2555 
2556 static void
2557 rfs4_lo_state_destroy(rfs4_entry_t u_entry)
2558 {
2559 	rfs4_lo_state_t *lsp = (rfs4_lo_state_t *)u_entry;
2560 
2561 	rfs4_dbe_lock(lsp->rls_state->rs_dbe);
2562 	list_remove(&lsp->rls_state->rs_lostatelist, lsp);
2563 	rfs4_dbe_unlock(lsp->rls_state->rs_dbe);
2564 
2565 	rfs4_sw_destroy(&lsp->rls_sw);
2566 
2567 	/* Make sure to release the file locks */
2568 	if (lsp->rls_locks_cleaned == FALSE) {
2569 		lsp->rls_locks_cleaned = TRUE;
2570 		if (lsp->rls_locker->rl_client->rc_sysidt != LM_NOSYSID) {
2571 			/* Is the PxFS kernel module loaded? */
2572 			if (lm_remove_file_locks != NULL) {
2573 				int new_sysid;
2574 
2575 				/* Encode the cluster nodeid in new sysid */
2576 				new_sysid =
2577 				    lsp->rls_locker->rl_client->rc_sysidt;
2578 				lm_set_nlmid_flk(&new_sysid);
2579 
2580 				/*
2581 				 * This PxFS routine removes file locks for a
2582 				 * client over all nodes of a cluster.
2583 				 */
2584 				DTRACE_PROBE1(nfss_i_clust_rm_lck,
2585 				    int, new_sysid);
2586 				(*lm_remove_file_locks)(new_sysid);
2587 			} else {
2588 				(void) cleanlocks(
2589 				    lsp->rls_state->rs_finfo->rf_vp,
2590 				    lsp->rls_locker->rl_pid,
2591 				    lsp->rls_locker->rl_client->rc_sysidt);
2592 			}
2593 		}
2594 	}
2595 
2596 	/* Free the last reply for this state */
2597 	rfs4_free_reply(&lsp->rls_reply);
2598 
2599 	rfs4_lockowner_rele(lsp->rls_locker);
2600 	lsp->rls_locker = NULL;
2601 
2602 	rfs4_state_rele_nounlock(lsp->rls_state);
2603 	lsp->rls_state = NULL;
2604 }
2605 
2606 static bool_t
2607 rfs4_lo_state_create(rfs4_entry_t u_entry, void *arg)
2608 {
2609 	rfs4_lo_state_t *lsp = (rfs4_lo_state_t *)u_entry;
2610 	rfs4_lo_state_t *argp = (rfs4_lo_state_t *)arg;
2611 	rfs4_lockowner_t *lo = argp->rls_locker;
2612 	rfs4_state_t *sp = argp->rls_state;
2613 
2614 	lsp->rls_state = sp;
2615 
2616 	lsp->rls_lockid = sp->rs_stateid;
2617 	lsp->rls_lockid.bits.type = LOCKID;
2618 	lsp->rls_lockid.bits.chgseq = 0;
2619 	lsp->rls_lockid.bits.pid = lo->rl_pid;
2620 
2621 	lsp->rls_locks_cleaned = FALSE;
2622 	lsp->rls_lock_completed = FALSE;
2623 
2624 	rfs4_sw_init(&lsp->rls_sw);
2625 
2626 	/* Attached the supplied lock owner */
2627 	rfs4_dbe_hold(lo->rl_dbe);
2628 	lsp->rls_locker = lo;
2629 
2630 	rfs4_dbe_lock(sp->rs_dbe);
2631 	list_insert_tail(&sp->rs_lostatelist, lsp);
2632 	rfs4_dbe_hold(sp->rs_dbe);
2633 	rfs4_dbe_unlock(sp->rs_dbe);
2634 
2635 	return (TRUE);
2636 }
2637 
2638 void
2639 rfs4_lo_state_rele(rfs4_lo_state_t *lsp, bool_t unlock_fp)
2640 {
2641 	if (unlock_fp == TRUE)
2642 		rw_exit(&lsp->rls_state->rs_finfo->rf_file_rwlock);
2643 	rfs4_dbe_rele(lsp->rls_dbe);
2644 }
2645 
2646 static rfs4_lo_state_t *
2647 rfs4_findlo_state(stateid_t *id, bool_t lock_fp)
2648 {
2649 	rfs4_lo_state_t *lsp;
2650 	bool_t create = FALSE;
2651 
2652 	lsp = (rfs4_lo_state_t *)rfs4_dbsearch(rfs4_lo_state_idx, id,
2653 	    &create, NULL, RFS4_DBS_VALID);
2654 	if (lock_fp == TRUE && lsp != NULL)
2655 		rw_enter(&lsp->rls_state->rs_finfo->rf_file_rwlock, RW_READER);
2656 
2657 	return (lsp);
2658 }
2659 
2660 
2661 static uint32_t
2662 lo_state_lo_hash(void *key)
2663 {
2664 	rfs4_lo_state_t *lsp = key;
2665 
2666 	return (ADDRHASH(lsp->rls_locker) ^ ADDRHASH(lsp->rls_state));
2667 }
2668 
2669 static bool_t
2670 lo_state_lo_compare(rfs4_entry_t u_entry, void *key)
2671 {
2672 	rfs4_lo_state_t *lsp = (rfs4_lo_state_t *)u_entry;
2673 	rfs4_lo_state_t *keyp = key;
2674 
2675 	return (keyp->rls_locker == lsp->rls_locker &&
2676 	    keyp->rls_state == lsp->rls_state);
2677 }
2678 
2679 static void *
2680 lo_state_lo_mkkey(rfs4_entry_t u_entry)
2681 {
2682 	return (u_entry);
2683 }
2684 
2685 rfs4_lo_state_t *
2686 rfs4_findlo_state_by_owner(rfs4_lockowner_t *lo, rfs4_state_t *sp,
2687     bool_t *create)
2688 {
2689 	rfs4_lo_state_t *lsp;
2690 	rfs4_lo_state_t arg;
2691 
2692 	arg.rls_locker = lo;
2693 	arg.rls_state = sp;
2694 
2695 	lsp = (rfs4_lo_state_t *)rfs4_dbsearch(rfs4_lo_state_owner_idx, &arg,
2696 	    create, &arg, RFS4_DBS_VALID);
2697 
2698 	return (lsp);
2699 }
2700 
2701 static stateid_t
2702 get_stateid(id_t eid)
2703 {
2704 	stateid_t id;
2705 
2706 	id.bits.boottime = rfs4_start_time;
2707 	id.bits.ident = eid;
2708 	id.bits.chgseq = 0;
2709 	id.bits.type = 0;
2710 	id.bits.pid = 0;
2711 
2712 	/*
2713 	 * If we are booted as a cluster node, embed our nodeid.
2714 	 * We've already done sanity checks in rfs4_client_create() so no
2715 	 * need to repeat them here.
2716 	 */
2717 	id.bits.clnodeid = (cluster_bootflags & CLUSTER_BOOTED) ?
2718 	    clconf_get_nodeid() : 0;
2719 
2720 	return (id);
2721 }
2722 
2723 /*
2724  * For use only when booted as a cluster node.
2725  * Returns TRUE if the embedded nodeid indicates that this stateid was
2726  * generated on another node.
2727  */
2728 static int
2729 foreign_stateid(stateid_t *id)
2730 {
2731 	ASSERT(cluster_bootflags & CLUSTER_BOOTED);
2732 	return (id->bits.clnodeid != (uint32_t)clconf_get_nodeid());
2733 }
2734 
2735 /*
2736  * For use only when booted as a cluster node.
2737  * Returns TRUE if the embedded nodeid indicates that this clientid was
2738  * generated on another node.
2739  */
2740 static int
2741 foreign_clientid(cid *cidp)
2742 {
2743 	ASSERT(cluster_bootflags & CLUSTER_BOOTED);
2744 	return (cidp->impl_id.c_id >> CLUSTER_NODEID_SHIFT !=
2745 	    (uint32_t)clconf_get_nodeid());
2746 }
2747 
2748 /*
2749  * For use only when booted as a cluster node.
2750  * Embed our cluster nodeid into the clientid.
2751  */
2752 static void
2753 embed_nodeid(cid *cidp)
2754 {
2755 	int clnodeid;
2756 	/*
2757 	 * Currently, our state tables are small enough that their
2758 	 * ids will leave enough bits free for the nodeid. If the
2759 	 * tables become larger, we mustn't overwrite the id.
2760 	 * Equally, we only have room for so many bits of nodeid, so
2761 	 * must check that too.
2762 	 */
2763 	ASSERT(cluster_bootflags & CLUSTER_BOOTED);
2764 	ASSERT(cidp->impl_id.c_id >> CLUSTER_NODEID_SHIFT == 0);
2765 	clnodeid = clconf_get_nodeid();
2766 	ASSERT(clnodeid <= CLUSTER_MAX_NODEID);
2767 	ASSERT(clnodeid != NODEID_UNKNOWN);
2768 	cidp->impl_id.c_id |= (clnodeid << CLUSTER_NODEID_SHIFT);
2769 }
2770 
2771 static uint32_t
2772 state_hash(void *key)
2773 {
2774 	stateid_t *ip = (stateid_t *)key;
2775 
2776 	return (ip->bits.ident);
2777 }
2778 
2779 static bool_t
2780 state_compare(rfs4_entry_t u_entry, void *key)
2781 {
2782 	rfs4_state_t *sp = (rfs4_state_t *)u_entry;
2783 	stateid_t *id = (stateid_t *)key;
2784 	bool_t rc;
2785 
2786 	rc = (sp->rs_stateid.bits.boottime == id->bits.boottime &&
2787 	    sp->rs_stateid.bits.ident == id->bits.ident);
2788 
2789 	return (rc);
2790 }
2791 
2792 static void *
2793 state_mkkey(rfs4_entry_t u_entry)
2794 {
2795 	rfs4_state_t *sp = (rfs4_state_t *)u_entry;
2796 
2797 	return (&sp->rs_stateid);
2798 }
2799 
2800 static void
2801 rfs4_state_destroy(rfs4_entry_t u_entry)
2802 {
2803 	rfs4_state_t *sp = (rfs4_state_t *)u_entry;
2804 
2805 	/* remove from openowner list */
2806 	rfs4_dbe_lock(sp->rs_owner->ro_dbe);
2807 	list_remove(&sp->rs_owner->ro_statelist, sp);
2808 	rfs4_dbe_unlock(sp->rs_owner->ro_dbe);
2809 
2810 	list_destroy(&sp->rs_lostatelist);
2811 
2812 	/* release any share locks for this stateid if it's still open */
2813 	if (!sp->rs_closed) {
2814 		rfs4_dbe_lock(sp->rs_dbe);
2815 		(void) rfs4_unshare(sp);
2816 		rfs4_dbe_unlock(sp->rs_dbe);
2817 	}
2818 
2819 	/* Were done with the file */
2820 	rfs4_file_rele(sp->rs_finfo);
2821 	sp->rs_finfo = NULL;
2822 
2823 	/* And now with the openowner */
2824 	rfs4_openowner_rele(sp->rs_owner);
2825 	sp->rs_owner = NULL;
2826 }
2827 
2828 static void
2829 rfs4_state_rele_nounlock(rfs4_state_t *sp)
2830 {
2831 	rfs4_dbe_rele(sp->rs_dbe);
2832 }
2833 
2834 void
2835 rfs4_state_rele(rfs4_state_t *sp)
2836 {
2837 	rw_exit(&sp->rs_finfo->rf_file_rwlock);
2838 	rfs4_dbe_rele(sp->rs_dbe);
2839 }
2840 
2841 static uint32_t
2842 deleg_hash(void *key)
2843 {
2844 	rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)key;
2845 
2846 	return (ADDRHASH(dsp->rds_client) ^ ADDRHASH(dsp->rds_finfo));
2847 }
2848 
2849 static bool_t
2850 deleg_compare(rfs4_entry_t u_entry, void *key)
2851 {
2852 	rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
2853 	rfs4_deleg_state_t *kdsp = (rfs4_deleg_state_t *)key;
2854 
2855 	return (dsp->rds_client == kdsp->rds_client &&
2856 	    dsp->rds_finfo == kdsp->rds_finfo);
2857 }
2858 
2859 static void *
2860 deleg_mkkey(rfs4_entry_t u_entry)
2861 {
2862 	return (u_entry);
2863 }
2864 
2865 static uint32_t
2866 deleg_state_hash(void *key)
2867 {
2868 	stateid_t *ip = (stateid_t *)key;
2869 
2870 	return (ip->bits.ident);
2871 }
2872 
2873 static bool_t
2874 deleg_state_compare(rfs4_entry_t u_entry, void *key)
2875 {
2876 	rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
2877 	stateid_t *id = (stateid_t *)key;
2878 	bool_t rc;
2879 
2880 	if (id->bits.type != DELEGID)
2881 		return (FALSE);
2882 
2883 	rc = (dsp->rds_delegid.bits.boottime == id->bits.boottime &&
2884 	    dsp->rds_delegid.bits.ident == id->bits.ident);
2885 
2886 	return (rc);
2887 }
2888 
2889 static void *
2890 deleg_state_mkkey(rfs4_entry_t u_entry)
2891 {
2892 	rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
2893 
2894 	return (&dsp->rds_delegid);
2895 }
2896 
2897 static bool_t
2898 rfs4_deleg_state_expiry(rfs4_entry_t u_entry)
2899 {
2900 	rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
2901 
2902 	if (rfs4_dbe_is_invalid(dsp->rds_dbe))
2903 		return (TRUE);
2904 
2905 	if (dsp->rds_dtype == OPEN_DELEGATE_NONE)
2906 		return (TRUE);
2907 
2908 	if ((gethrestime_sec() - dsp->rds_client->rc_last_access
2909 	    > rfs4_lease_time)) {
2910 		rfs4_dbe_invalidate(dsp->rds_dbe);
2911 		return (TRUE);
2912 	}
2913 
2914 	return (FALSE);
2915 }
2916 
2917 static bool_t
2918 rfs4_deleg_state_create(rfs4_entry_t u_entry, void *argp)
2919 {
2920 	rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
2921 	rfs4_file_t *fp = ((rfs4_deleg_state_t *)argp)->rds_finfo;
2922 	rfs4_client_t *cp = ((rfs4_deleg_state_t *)argp)->rds_client;
2923 
2924 	rfs4_dbe_hold(fp->rf_dbe);
2925 	rfs4_dbe_hold(cp->rc_dbe);
2926 
2927 	dsp->rds_delegid = get_stateid(rfs4_dbe_getid(dsp->rds_dbe));
2928 	dsp->rds_delegid.bits.type = DELEGID;
2929 	dsp->rds_finfo = fp;
2930 	dsp->rds_client = cp;
2931 	dsp->rds_dtype = OPEN_DELEGATE_NONE;
2932 
2933 	dsp->rds_time_granted = gethrestime_sec();	/* observability */
2934 	dsp->rds_time_revoked = 0;
2935 
2936 	list_link_init(&dsp->rds_node);
2937 
2938 	return (TRUE);
2939 }
2940 
2941 static void
2942 rfs4_deleg_state_destroy(rfs4_entry_t u_entry)
2943 {
2944 	rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
2945 
2946 	/* return delegation if necessary */
2947 	rfs4_return_deleg(dsp, FALSE);
2948 
2949 	/* Were done with the file */
2950 	rfs4_file_rele(dsp->rds_finfo);
2951 	dsp->rds_finfo = NULL;
2952 
2953 	/* And now with the openowner */
2954 	rfs4_client_rele(dsp->rds_client);
2955 	dsp->rds_client = NULL;
2956 }
2957 
2958 rfs4_deleg_state_t *
2959 rfs4_finddeleg(rfs4_state_t *sp, bool_t *create)
2960 {
2961 	rfs4_deleg_state_t ds, *dsp;
2962 
2963 	ds.rds_client = sp->rs_owner->ro_client;
2964 	ds.rds_finfo = sp->rs_finfo;
2965 
2966 	dsp = (rfs4_deleg_state_t *)rfs4_dbsearch(rfs4_deleg_idx, &ds,
2967 	    create, &ds, RFS4_DBS_VALID);
2968 
2969 	return (dsp);
2970 }
2971 
2972 rfs4_deleg_state_t *
2973 rfs4_finddelegstate(stateid_t *id)
2974 {
2975 	rfs4_deleg_state_t *dsp;
2976 	bool_t create = FALSE;
2977 
2978 	dsp = (rfs4_deleg_state_t *)rfs4_dbsearch(rfs4_deleg_state_idx, id,
2979 	    &create, NULL, RFS4_DBS_VALID);
2980 
2981 	return (dsp);
2982 }
2983 
2984 void
2985 rfs4_deleg_state_rele(rfs4_deleg_state_t *dsp)
2986 {
2987 	rfs4_dbe_rele(dsp->rds_dbe);
2988 }
2989 
2990 void
2991 rfs4_update_lock_sequence(rfs4_lo_state_t *lsp)
2992 {
2993 
2994 	rfs4_dbe_lock(lsp->rls_dbe);
2995 
2996 	/*
2997 	 * If we are skipping sequence id checking, this means that
2998 	 * this is the first lock request and therefore the sequence
2999 	 * id does not need to be updated.  This only happens on the
3000 	 * first lock request for a lockowner
3001 	 */
3002 	if (!lsp->rls_skip_seqid_check)
3003 		lsp->rls_seqid++;
3004 
3005 	rfs4_dbe_unlock(lsp->rls_dbe);
3006 }
3007 
3008 void
3009 rfs4_update_lock_resp(rfs4_lo_state_t *lsp, nfs_resop4 *resp)
3010 {
3011 
3012 	rfs4_dbe_lock(lsp->rls_dbe);
3013 
3014 	rfs4_free_reply(&lsp->rls_reply);
3015 
3016 	rfs4_copy_reply(&lsp->rls_reply, resp);
3017 
3018 	rfs4_dbe_unlock(lsp->rls_dbe);
3019 }
3020 
3021 void
3022 rfs4_free_opens(rfs4_openowner_t *oo, bool_t invalidate,
3023     bool_t close_of_client)
3024 {
3025 	rfs4_state_t *sp;
3026 
3027 	rfs4_dbe_lock(oo->ro_dbe);
3028 
3029 	for (sp = list_head(&oo->ro_statelist); sp != NULL;
3030 	    sp = list_next(&oo->ro_statelist, sp)) {
3031 		rfs4_state_close(sp, FALSE, close_of_client, CRED());
3032 		if (invalidate == TRUE)
3033 			rfs4_dbe_invalidate(sp->rs_dbe);
3034 	}
3035 
3036 	rfs4_dbe_invalidate(oo->ro_dbe);
3037 	rfs4_dbe_unlock(oo->ro_dbe);
3038 }
3039 
3040 static uint32_t
3041 state_owner_file_hash(void *key)
3042 {
3043 	rfs4_state_t *sp = key;
3044 
3045 	return (ADDRHASH(sp->rs_owner) ^ ADDRHASH(sp->rs_finfo));
3046 }
3047 
3048 static bool_t
3049 state_owner_file_compare(rfs4_entry_t u_entry, void *key)
3050 {
3051 	rfs4_state_t *sp = (rfs4_state_t *)u_entry;
3052 	rfs4_state_t *arg = key;
3053 
3054 	if (sp->rs_closed == TRUE)
3055 		return (FALSE);
3056 
3057 	return (arg->rs_owner == sp->rs_owner && arg->rs_finfo == sp->rs_finfo);
3058 }
3059 
3060 static void *
3061 state_owner_file_mkkey(rfs4_entry_t u_entry)
3062 {
3063 	return (u_entry);
3064 }
3065 
3066 static uint32_t
3067 state_file_hash(void *key)
3068 {
3069 	return (ADDRHASH(key));
3070 }
3071 
3072 static bool_t
3073 state_file_compare(rfs4_entry_t u_entry, void *key)
3074 {
3075 	rfs4_state_t *sp = (rfs4_state_t *)u_entry;
3076 	rfs4_file_t *fp = key;
3077 
3078 	if (sp->rs_closed == TRUE)
3079 		return (FALSE);
3080 
3081 	return (fp == sp->rs_finfo);
3082 }
3083 
3084 static void *
3085 state_file_mkkey(rfs4_entry_t u_entry)
3086 {
3087 	rfs4_state_t *sp = (rfs4_state_t *)u_entry;
3088 
3089 	return (sp->rs_finfo);
3090 }
3091 
3092 rfs4_state_t *
3093 rfs4_findstate_by_owner_file(rfs4_openowner_t *oo, rfs4_file_t *fp,
3094 	bool_t *create)
3095 {
3096 	rfs4_state_t *sp;
3097 	rfs4_state_t key;
3098 
3099 	key.rs_owner = oo;
3100 	key.rs_finfo = fp;
3101 
3102 	sp = (rfs4_state_t *)rfs4_dbsearch(rfs4_state_owner_file_idx, &key,
3103 	    create, &key, RFS4_DBS_VALID);
3104 
3105 	return (sp);
3106 }
3107 
3108 /* This returns ANY state struct that refers to this file */
3109 static rfs4_state_t *
3110 rfs4_findstate_by_file(rfs4_file_t *fp)
3111 {
3112 	bool_t create = FALSE;
3113 
3114 	return ((rfs4_state_t *)rfs4_dbsearch(rfs4_state_file_idx, fp,
3115 	    &create, fp, RFS4_DBS_VALID));
3116 }
3117 
3118 static bool_t
3119 rfs4_state_expiry(rfs4_entry_t u_entry)
3120 {
3121 	rfs4_state_t *sp = (rfs4_state_t *)u_entry;
3122 
3123 	if (rfs4_dbe_is_invalid(sp->rs_dbe))
3124 		return (TRUE);
3125 
3126 	if (sp->rs_closed == TRUE &&
3127 	    ((gethrestime_sec() - rfs4_dbe_get_timerele(sp->rs_dbe))
3128 	    > rfs4_lease_time))
3129 		return (TRUE);
3130 
3131 	return ((gethrestime_sec() - sp->rs_owner->ro_client->rc_last_access
3132 	    > rfs4_lease_time));
3133 }
3134 
3135 static bool_t
3136 rfs4_state_create(rfs4_entry_t u_entry, void *argp)
3137 {
3138 	rfs4_state_t *sp = (rfs4_state_t *)u_entry;
3139 	rfs4_file_t *fp = ((rfs4_state_t *)argp)->rs_finfo;
3140 	rfs4_openowner_t *oo = ((rfs4_state_t *)argp)->rs_owner;
3141 
3142 	rfs4_dbe_hold(fp->rf_dbe);
3143 	rfs4_dbe_hold(oo->ro_dbe);
3144 	sp->rs_stateid = get_stateid(rfs4_dbe_getid(sp->rs_dbe));
3145 	sp->rs_stateid.bits.type = OPENID;
3146 	sp->rs_owner = oo;
3147 	sp->rs_finfo = fp;
3148 
3149 	list_create(&sp->rs_lostatelist, sizeof (rfs4_lo_state_t),
3150 	    offsetof(rfs4_lo_state_t, rls_node));
3151 
3152 	/* Insert state on per open owner's list */
3153 	rfs4_dbe_lock(oo->ro_dbe);
3154 	list_insert_tail(&oo->ro_statelist, sp);
3155 	rfs4_dbe_unlock(oo->ro_dbe);
3156 
3157 	return (TRUE);
3158 }
3159 
3160 static rfs4_state_t *
3161 rfs4_findstate(stateid_t *id, rfs4_dbsearch_type_t find_invalid, bool_t lock_fp)
3162 {
3163 	rfs4_state_t *sp;
3164 	bool_t create = FALSE;
3165 
3166 	sp = (rfs4_state_t *)rfs4_dbsearch(rfs4_state_idx, id,
3167 	    &create, NULL, find_invalid);
3168 	if (lock_fp == TRUE && sp != NULL)
3169 		rw_enter(&sp->rs_finfo->rf_file_rwlock, RW_READER);
3170 
3171 	return (sp);
3172 }
3173 
3174 void
3175 rfs4_state_close(rfs4_state_t *sp, bool_t lock_held, bool_t close_of_client,
3176     cred_t *cr)
3177 {
3178 	/* Remove the associated lo_state owners */
3179 	if (!lock_held)
3180 		rfs4_dbe_lock(sp->rs_dbe);
3181 
3182 	/*
3183 	 * If refcnt == 0, the dbe is about to be destroyed.
3184 	 * lock state will be released by the reaper thread.
3185 	 */
3186 
3187 	if (rfs4_dbe_refcnt(sp->rs_dbe) > 0) {
3188 		if (sp->rs_closed == FALSE) {
3189 			rfs4_release_share_lock_state(sp, cr, close_of_client);
3190 			sp->rs_closed = TRUE;
3191 		}
3192 	}
3193 
3194 	if (!lock_held)
3195 		rfs4_dbe_unlock(sp->rs_dbe);
3196 }
3197 
3198 /*
3199  * Remove all state associated with the given client.
3200  */
3201 void
3202 rfs4_client_state_remove(rfs4_client_t *cp)
3203 {
3204 	rfs4_openowner_t *oo;
3205 
3206 	rfs4_dbe_lock(cp->rc_dbe);
3207 
3208 	for (oo = list_head(&cp->rc_openownerlist); oo != NULL;
3209 	    oo = list_next(&cp->rc_openownerlist, oo)) {
3210 		rfs4_free_opens(oo, TRUE, TRUE);
3211 	}
3212 
3213 	rfs4_dbe_unlock(cp->rc_dbe);
3214 }
3215 
3216 void
3217 rfs4_client_close(rfs4_client_t *cp)
3218 {
3219 	/* Mark client as going away. */
3220 	rfs4_dbe_lock(cp->rc_dbe);
3221 	rfs4_dbe_invalidate(cp->rc_dbe);
3222 	rfs4_dbe_unlock(cp->rc_dbe);
3223 
3224 	rfs4_client_state_remove(cp);
3225 
3226 	/* Release the client */
3227 	rfs4_client_rele(cp);
3228 }
3229 
3230 nfsstat4
3231 rfs4_check_clientid(clientid4 *cp, int setclid_confirm)
3232 {
3233 	cid *cidp = (cid *) cp;
3234 
3235 	/*
3236 	 * If we are booted as a cluster node, check the embedded nodeid.
3237 	 * If it indicates that this clientid was generated on another node,
3238 	 * inform the client accordingly.
3239 	 */
3240 	if (cluster_bootflags & CLUSTER_BOOTED && foreign_clientid(cidp))
3241 		return (NFS4ERR_STALE_CLIENTID);
3242 
3243 	/*
3244 	 * If the server start time matches the time provided
3245 	 * by the client (via the clientid) and this is NOT a
3246 	 * setclientid_confirm then return EXPIRED.
3247 	 */
3248 	if (!setclid_confirm && cidp->impl_id.start_time == rfs4_start_time)
3249 		return (NFS4ERR_EXPIRED);
3250 
3251 	return (NFS4ERR_STALE_CLIENTID);
3252 }
3253 
3254 /*
3255  * This is used when a stateid has not been found amongst the
3256  * current server's state.  Check the stateid to see if it
3257  * was from this server instantiation or not.
3258  */
3259 static nfsstat4
3260 what_stateid_error(stateid_t *id, stateid_type_t type)
3261 {
3262 	/* If we are booted as a cluster node, was stateid locally generated? */
3263 	if ((cluster_bootflags & CLUSTER_BOOTED) && foreign_stateid(id))
3264 		return (NFS4ERR_STALE_STATEID);
3265 
3266 	/* If types don't match then no use checking further */
3267 	if (type != id->bits.type)
3268 		return (NFS4ERR_BAD_STATEID);
3269 
3270 	/* From a different server instantiation, return STALE */
3271 	if (id->bits.boottime != rfs4_start_time)
3272 		return (NFS4ERR_STALE_STATEID);
3273 
3274 	/*
3275 	 * From this server but the state is most likely beyond lease
3276 	 * timeout: return NFS4ERR_EXPIRED.  However, there is the
3277 	 * case of a delegation stateid.  For delegations, there is a
3278 	 * case where the state can be removed without the client's
3279 	 * knowledge/consent: revocation.  In the case of delegation
3280 	 * revocation, the delegation state will be removed and will
3281 	 * not be found.  If the client does something like a
3282 	 * DELEGRETURN or even a READ/WRITE with a delegatoin stateid
3283 	 * that has been revoked, the server should return BAD_STATEID
3284 	 * instead of the more common EXPIRED error.
3285 	 */
3286 	if (id->bits.boottime == rfs4_start_time) {
3287 		if (type == DELEGID)
3288 			return (NFS4ERR_BAD_STATEID);
3289 		else
3290 			return (NFS4ERR_EXPIRED);
3291 	}
3292 
3293 	return (NFS4ERR_BAD_STATEID);
3294 }
3295 
3296 /*
3297  * Used later on to find the various state structs.  When called from
3298  * rfs4_check_stateid()->rfs4_get_all_state(), no file struct lock is
3299  * taken (it is not needed) and helps on the read/write path with
3300  * respect to performance.
3301  */
3302 static nfsstat4
3303 rfs4_get_state_lockit(stateid4 *stateid, rfs4_state_t **spp,
3304     rfs4_dbsearch_type_t find_invalid, bool_t lock_fp)
3305 {
3306 	stateid_t *id = (stateid_t *)stateid;
3307 	rfs4_state_t *sp;
3308 
3309 	*spp = NULL;
3310 
3311 	/* If we are booted as a cluster node, was stateid locally generated? */
3312 	if ((cluster_bootflags & CLUSTER_BOOTED) && foreign_stateid(id))
3313 		return (NFS4ERR_STALE_STATEID);
3314 
3315 	sp = rfs4_findstate(id, find_invalid, lock_fp);
3316 	if (sp == NULL) {
3317 		return (what_stateid_error(id, OPENID));
3318 	}
3319 
3320 	if (rfs4_lease_expired(sp->rs_owner->ro_client)) {
3321 		if (lock_fp == TRUE)
3322 			rfs4_state_rele(sp);
3323 		else
3324 			rfs4_state_rele_nounlock(sp);
3325 		return (NFS4ERR_EXPIRED);
3326 	}
3327 
3328 	*spp = sp;
3329 
3330 	return (NFS4_OK);
3331 }
3332 
3333 nfsstat4
3334 rfs4_get_state(stateid4 *stateid, rfs4_state_t **spp,
3335     rfs4_dbsearch_type_t find_invalid)
3336 {
3337 	return (rfs4_get_state_lockit(stateid, spp, find_invalid, TRUE));
3338 }
3339 
3340 int
3341 rfs4_check_stateid_seqid(rfs4_state_t *sp, stateid4 *stateid)
3342 {
3343 	stateid_t *id = (stateid_t *)stateid;
3344 
3345 	if (rfs4_lease_expired(sp->rs_owner->ro_client))
3346 		return (NFS4_CHECK_STATEID_EXPIRED);
3347 
3348 	/* Stateid is some time in the future - that's bad */
3349 	if (sp->rs_stateid.bits.chgseq < id->bits.chgseq)
3350 		return (NFS4_CHECK_STATEID_BAD);
3351 
3352 	if (sp->rs_stateid.bits.chgseq == id->bits.chgseq + 1)
3353 		return (NFS4_CHECK_STATEID_REPLAY);
3354 
3355 	/* Stateid is some time in the past - that's old */
3356 	if (sp->rs_stateid.bits.chgseq > id->bits.chgseq)
3357 		return (NFS4_CHECK_STATEID_OLD);
3358 
3359 	/* Caller needs to know about confirmation before closure */
3360 	if (sp->rs_owner->ro_need_confirm)
3361 		return (NFS4_CHECK_STATEID_UNCONFIRMED);
3362 
3363 	if (sp->rs_closed == TRUE)
3364 		return (NFS4_CHECK_STATEID_CLOSED);
3365 
3366 	return (NFS4_CHECK_STATEID_OKAY);
3367 }
3368 
3369 int
3370 rfs4_check_lo_stateid_seqid(rfs4_lo_state_t *lsp, stateid4 *stateid)
3371 {
3372 	stateid_t *id = (stateid_t *)stateid;
3373 
3374 	if (rfs4_lease_expired(lsp->rls_state->rs_owner->ro_client))
3375 		return (NFS4_CHECK_STATEID_EXPIRED);
3376 
3377 	/* Stateid is some time in the future - that's bad */
3378 	if (lsp->rls_lockid.bits.chgseq < id->bits.chgseq)
3379 		return (NFS4_CHECK_STATEID_BAD);
3380 
3381 	if (lsp->rls_lockid.bits.chgseq == id->bits.chgseq + 1)
3382 		return (NFS4_CHECK_STATEID_REPLAY);
3383 
3384 	/* Stateid is some time in the past - that's old */
3385 	if (lsp->rls_lockid.bits.chgseq > id->bits.chgseq)
3386 		return (NFS4_CHECK_STATEID_OLD);
3387 
3388 	if (lsp->rls_state->rs_closed == TRUE)
3389 		return (NFS4_CHECK_STATEID_CLOSED);
3390 
3391 	return (NFS4_CHECK_STATEID_OKAY);
3392 }
3393 
3394 nfsstat4
3395 rfs4_get_deleg_state(stateid4 *stateid, rfs4_deleg_state_t **dspp)
3396 {
3397 	stateid_t *id = (stateid_t *)stateid;
3398 	rfs4_deleg_state_t *dsp;
3399 
3400 	*dspp = NULL;
3401 
3402 	/* If we are booted as a cluster node, was stateid locally generated? */
3403 	if ((cluster_bootflags & CLUSTER_BOOTED) && foreign_stateid(id))
3404 		return (NFS4ERR_STALE_STATEID);
3405 
3406 	dsp = rfs4_finddelegstate(id);
3407 	if (dsp == NULL) {
3408 		return (what_stateid_error(id, DELEGID));
3409 	}
3410 
3411 	if (rfs4_lease_expired(dsp->rds_client)) {
3412 		rfs4_deleg_state_rele(dsp);
3413 		return (NFS4ERR_EXPIRED);
3414 	}
3415 
3416 	*dspp = dsp;
3417 
3418 	return (NFS4_OK);
3419 }
3420 
3421 nfsstat4
3422 rfs4_get_lo_state(stateid4 *stateid, rfs4_lo_state_t **lspp, bool_t lock_fp)
3423 {
3424 	stateid_t *id = (stateid_t *)stateid;
3425 	rfs4_lo_state_t *lsp;
3426 
3427 	*lspp = NULL;
3428 
3429 	/* If we are booted as a cluster node, was stateid locally generated? */
3430 	if ((cluster_bootflags & CLUSTER_BOOTED) && foreign_stateid(id))
3431 		return (NFS4ERR_STALE_STATEID);
3432 
3433 	lsp = rfs4_findlo_state(id, lock_fp);
3434 	if (lsp == NULL) {
3435 		return (what_stateid_error(id, LOCKID));
3436 	}
3437 
3438 	if (rfs4_lease_expired(lsp->rls_state->rs_owner->ro_client)) {
3439 		rfs4_lo_state_rele(lsp, lock_fp);
3440 		return (NFS4ERR_EXPIRED);
3441 	}
3442 
3443 	*lspp = lsp;
3444 
3445 	return (NFS4_OK);
3446 }
3447 
3448 static nfsstat4
3449 rfs4_get_all_state(stateid4 *sid, rfs4_state_t **spp,
3450     rfs4_deleg_state_t **dspp, rfs4_lo_state_t **lspp)
3451 {
3452 	rfs4_state_t *sp = NULL;
3453 	rfs4_deleg_state_t *dsp = NULL;
3454 	rfs4_lo_state_t *lsp = NULL;
3455 	stateid_t *id;
3456 	nfsstat4 status;
3457 
3458 	*spp = NULL; *dspp = NULL; *lspp = NULL;
3459 
3460 	id = (stateid_t *)sid;
3461 	switch (id->bits.type) {
3462 	case OPENID:
3463 		status = rfs4_get_state_lockit(sid, &sp, FALSE, FALSE);
3464 		break;
3465 	case DELEGID:
3466 		status = rfs4_get_deleg_state(sid, &dsp);
3467 		break;
3468 	case LOCKID:
3469 		status = rfs4_get_lo_state(sid, &lsp, FALSE);
3470 		if (status == NFS4_OK) {
3471 			sp = lsp->rls_state;
3472 			rfs4_dbe_hold(sp->rs_dbe);
3473 		}
3474 		break;
3475 	default:
3476 		status = NFS4ERR_BAD_STATEID;
3477 	}
3478 
3479 	if (status == NFS4_OK) {
3480 		*spp = sp;
3481 		*dspp = dsp;
3482 		*lspp = lsp;
3483 	}
3484 
3485 	return (status);
3486 }
3487 
3488 /*
3489  * Given the I/O mode (FREAD or FWRITE), this checks whether the
3490  * rfs4_state_t struct has access to do this operation and if so
3491  * return NFS4_OK; otherwise the proper NFSv4 error is returned.
3492  */
3493 nfsstat4
3494 rfs4_state_has_access(rfs4_state_t *sp, int mode, vnode_t *vp)
3495 {
3496 	nfsstat4 stat = NFS4_OK;
3497 	rfs4_file_t *fp;
3498 	bool_t create = FALSE;
3499 
3500 	rfs4_dbe_lock(sp->rs_dbe);
3501 	if (mode == FWRITE) {
3502 		if (!(sp->rs_share_access & OPEN4_SHARE_ACCESS_WRITE)) {
3503 			stat = NFS4ERR_OPENMODE;
3504 		}
3505 	} else if (mode == FREAD) {
3506 		if (!(sp->rs_share_access & OPEN4_SHARE_ACCESS_READ)) {
3507 			/*
3508 			 * If we have OPENed the file with DENYing access
3509 			 * to both READ and WRITE then no one else could
3510 			 * have OPENed the file, hence no conflicting READ
3511 			 * deny.  This check is merely an optimization.
3512 			 */
3513 			if (sp->rs_share_deny == OPEN4_SHARE_DENY_BOTH)
3514 				goto out;
3515 
3516 			/* Check against file struct's DENY mode */
3517 			fp = rfs4_findfile(vp, NULL, &create);
3518 			if (fp != NULL) {
3519 				int deny_read = 0;
3520 				rfs4_dbe_lock(fp->rf_dbe);
3521 				/*
3522 				 * Check if any other open owner has the file
3523 				 * OPENed with deny READ.
3524 				 */
3525 				if (sp->rs_share_deny & OPEN4_SHARE_DENY_READ)
3526 					deny_read = 1;
3527 				ASSERT(fp->rf_deny_read >= deny_read);
3528 				if (fp->rf_deny_read > deny_read)
3529 					stat = NFS4ERR_OPENMODE;
3530 				rfs4_dbe_unlock(fp->rf_dbe);
3531 				rfs4_file_rele(fp);
3532 			}
3533 		}
3534 	} else {
3535 		/* Illegal I/O mode */
3536 		stat = NFS4ERR_INVAL;
3537 	}
3538 out:
3539 	rfs4_dbe_unlock(sp->rs_dbe);
3540 	return (stat);
3541 }
3542 
3543 /*
3544  * Given the I/O mode (FREAD or FWRITE), the vnode, the stateid and whether
3545  * the file is being truncated, return NFS4_OK if allowed or appropriate
3546  * V4 error if not. Note NFS4ERR_DELAY will be returned and a recall on
3547  * the associated file will be done if the I/O is not consistent with any
3548  * delegation in effect on the file. Should be holding VOP_RWLOCK, either
3549  * as reader or writer as appropriate. rfs4_op_open will acquire the
3550  * VOP_RWLOCK as writer when setting up delegation. If the stateid is bad
3551  * this routine will return NFS4ERR_BAD_STATEID. In addition, through the
3552  * deleg parameter, we will return whether a write delegation is held by
3553  * the client associated with this stateid.
3554  * If the server instance associated with the relevant client is in its
3555  * grace period, return NFS4ERR_GRACE.
3556  */
3557 
3558 nfsstat4
3559 rfs4_check_stateid(int mode, vnode_t *vp,
3560     stateid4 *stateid, bool_t trunc, bool_t *deleg,
3561     bool_t do_access, caller_context_t *ct)
3562 {
3563 	rfs4_file_t *fp;
3564 	bool_t create = FALSE;
3565 	rfs4_state_t *sp;
3566 	rfs4_deleg_state_t *dsp;
3567 	rfs4_lo_state_t *lsp;
3568 	stateid_t *id = (stateid_t *)stateid;
3569 	nfsstat4 stat = NFS4_OK;
3570 
3571 	if (ct != NULL) {
3572 		ct->cc_sysid = 0;
3573 		ct->cc_pid = 0;
3574 		ct->cc_caller_id = nfs4_srv_caller_id;
3575 		ct->cc_flags = CC_DONTBLOCK;
3576 	}
3577 
3578 	if (ISSPECIAL(stateid)) {
3579 		fp = rfs4_findfile(vp, NULL, &create);
3580 		if (fp == NULL)
3581 			return (NFS4_OK);
3582 		if (fp->rf_dinfo.rd_dtype == OPEN_DELEGATE_NONE) {
3583 			rfs4_file_rele(fp);
3584 			return (NFS4_OK);
3585 		}
3586 		if (mode == FWRITE ||
3587 		    fp->rf_dinfo.rd_dtype == OPEN_DELEGATE_WRITE) {
3588 			rfs4_recall_deleg(fp, trunc, NULL);
3589 			rfs4_file_rele(fp);
3590 			return (NFS4ERR_DELAY);
3591 		}
3592 		rfs4_file_rele(fp);
3593 		return (NFS4_OK);
3594 	} else {
3595 		stat = rfs4_get_all_state(stateid, &sp, &dsp, &lsp);
3596 		if (stat != NFS4_OK)
3597 			return (stat);
3598 		if (lsp != NULL) {
3599 			/* Is associated server instance in its grace period? */
3600 			if (rfs4_clnt_in_grace(lsp->rls_locker->rl_client)) {
3601 				rfs4_lo_state_rele(lsp, FALSE);
3602 				if (sp != NULL)
3603 					rfs4_state_rele_nounlock(sp);
3604 				return (NFS4ERR_GRACE);
3605 			}
3606 			if (id->bits.type == LOCKID) {
3607 				/* Seqid in the future? - that's bad */
3608 				if (lsp->rls_lockid.bits.chgseq <
3609 				    id->bits.chgseq) {
3610 					rfs4_lo_state_rele(lsp, FALSE);
3611 					if (sp != NULL)
3612 						rfs4_state_rele_nounlock(sp);
3613 					return (NFS4ERR_BAD_STATEID);
3614 				}
3615 				/* Seqid in the past? - that's old */
3616 				if (lsp->rls_lockid.bits.chgseq >
3617 				    id->bits.chgseq) {
3618 					rfs4_lo_state_rele(lsp, FALSE);
3619 					if (sp != NULL)
3620 						rfs4_state_rele_nounlock(sp);
3621 					return (NFS4ERR_OLD_STATEID);
3622 				}
3623 				/* Ensure specified filehandle matches */
3624 				if (lsp->rls_state->rs_finfo->rf_vp != vp) {
3625 					rfs4_lo_state_rele(lsp, FALSE);
3626 					if (sp != NULL)
3627 						rfs4_state_rele_nounlock(sp);
3628 					return (NFS4ERR_BAD_STATEID);
3629 				}
3630 			}
3631 			if (ct != NULL) {
3632 				ct->cc_sysid =
3633 				    lsp->rls_locker->rl_client->rc_sysidt;
3634 				ct->cc_pid = lsp->rls_locker->rl_pid;
3635 			}
3636 			rfs4_lo_state_rele(lsp, FALSE);
3637 		}
3638 
3639 		/* Stateid provided was an "open" stateid */
3640 		if (sp != NULL) {
3641 			/* Is associated server instance in its grace period? */
3642 			if (rfs4_clnt_in_grace(sp->rs_owner->ro_client)) {
3643 				rfs4_state_rele_nounlock(sp);
3644 				return (NFS4ERR_GRACE);
3645 			}
3646 			if (id->bits.type == OPENID) {
3647 				/* Seqid in the future? - that's bad */
3648 				if (sp->rs_stateid.bits.chgseq <
3649 				    id->bits.chgseq) {
3650 					rfs4_state_rele_nounlock(sp);
3651 					return (NFS4ERR_BAD_STATEID);
3652 				}
3653 				/* Seqid in the past - that's old */
3654 				if (sp->rs_stateid.bits.chgseq >
3655 				    id->bits.chgseq) {
3656 					rfs4_state_rele_nounlock(sp);
3657 					return (NFS4ERR_OLD_STATEID);
3658 				}
3659 			}
3660 			/* Ensure specified filehandle matches */
3661 			if (sp->rs_finfo->rf_vp != vp) {
3662 				rfs4_state_rele_nounlock(sp);
3663 				return (NFS4ERR_BAD_STATEID);
3664 			}
3665 
3666 			if (sp->rs_owner->ro_need_confirm) {
3667 				rfs4_state_rele_nounlock(sp);
3668 				return (NFS4ERR_BAD_STATEID);
3669 			}
3670 
3671 			if (sp->rs_closed == TRUE) {
3672 				rfs4_state_rele_nounlock(sp);
3673 				return (NFS4ERR_OLD_STATEID);
3674 			}
3675 
3676 			if (do_access)
3677 				stat = rfs4_state_has_access(sp, mode, vp);
3678 			else
3679 				stat = NFS4_OK;
3680 
3681 			/*
3682 			 * Return whether this state has write
3683 			 * delegation if desired
3684 			 */
3685 			if (deleg && (sp->rs_finfo->rf_dinfo.rd_dtype ==
3686 			    OPEN_DELEGATE_WRITE))
3687 				*deleg = TRUE;
3688 
3689 			/*
3690 			 * We got a valid stateid, so we update the
3691 			 * lease on the client. Ideally we would like
3692 			 * to do this after the calling op succeeds,
3693 			 * but for now this will be good
3694 			 * enough. Callers of this routine are
3695 			 * currently insulated from the state stuff.
3696 			 */
3697 			rfs4_update_lease(sp->rs_owner->ro_client);
3698 
3699 			/*
3700 			 * If a delegation is present on this file and
3701 			 * this is a WRITE, then update the lastwrite
3702 			 * time to indicate that activity is present.
3703 			 */
3704 			if (sp->rs_finfo->rf_dinfo.rd_dtype ==
3705 			    OPEN_DELEGATE_WRITE &&
3706 			    mode == FWRITE) {
3707 				sp->rs_finfo->rf_dinfo.rd_time_lastwrite =
3708 				    gethrestime_sec();
3709 			}
3710 
3711 			rfs4_state_rele_nounlock(sp);
3712 
3713 			return (stat);
3714 		}
3715 
3716 		if (dsp != NULL) {
3717 			/* Is associated server instance in its grace period? */
3718 			if (rfs4_clnt_in_grace(dsp->rds_client)) {
3719 				rfs4_deleg_state_rele(dsp);
3720 				return (NFS4ERR_GRACE);
3721 			}
3722 			if (dsp->rds_delegid.bits.chgseq != id->bits.chgseq) {
3723 				rfs4_deleg_state_rele(dsp);
3724 				return (NFS4ERR_BAD_STATEID);
3725 			}
3726 
3727 			/* Ensure specified filehandle matches */
3728 			if (dsp->rds_finfo->rf_vp != vp) {
3729 				rfs4_deleg_state_rele(dsp);
3730 				return (NFS4ERR_BAD_STATEID);
3731 			}
3732 			/*
3733 			 * Return whether this state has write
3734 			 * delegation if desired
3735 			 */
3736 			if (deleg && (dsp->rds_finfo->rf_dinfo.rd_dtype ==
3737 			    OPEN_DELEGATE_WRITE))
3738 				*deleg = TRUE;
3739 
3740 			rfs4_update_lease(dsp->rds_client);
3741 
3742 			/*
3743 			 * If a delegation is present on this file and
3744 			 * this is a WRITE, then update the lastwrite
3745 			 * time to indicate that activity is present.
3746 			 */
3747 			if (dsp->rds_finfo->rf_dinfo.rd_dtype ==
3748 			    OPEN_DELEGATE_WRITE && mode == FWRITE) {
3749 				dsp->rds_finfo->rf_dinfo.rd_time_lastwrite =
3750 				    gethrestime_sec();
3751 			}
3752 
3753 			/*
3754 			 * XXX - what happens if this is a WRITE and the
3755 			 * delegation type of for READ.
3756 			 */
3757 			rfs4_deleg_state_rele(dsp);
3758 
3759 			return (stat);
3760 		}
3761 		/*
3762 		 * If we got this far, something bad happened
3763 		 */
3764 		return (NFS4ERR_BAD_STATEID);
3765 	}
3766 }
3767 
3768 
3769 /*
3770  * This is a special function in that for the file struct provided the
3771  * server wants to remove/close all current state associated with the
3772  * file.  The prime use of this would be with OP_REMOVE to force the
3773  * release of state and particularly of file locks.
3774  *
3775  * There is an assumption that there is no delegations outstanding on
3776  * this file at this point.  The caller should have waited for those
3777  * to be returned or revoked.
3778  */
3779 void
3780 rfs4_close_all_state(rfs4_file_t *fp)
3781 {
3782 	rfs4_state_t *sp;
3783 
3784 	rfs4_dbe_lock(fp->rf_dbe);
3785 
3786 #ifdef DEBUG
3787 	/* only applies when server is handing out delegations */
3788 	if (rfs4_deleg_policy != SRV_NEVER_DELEGATE)
3789 		ASSERT(fp->rf_dinfo.rd_hold_grant > 0);
3790 #endif
3791 
3792 	/* No delegations for this file */
3793 	ASSERT(list_is_empty(&fp->rf_delegstatelist));
3794 
3795 	/* Make sure that it can not be found */
3796 	rfs4_dbe_invalidate(fp->rf_dbe);
3797 
3798 	if (fp->rf_vp == NULL) {
3799 		rfs4_dbe_unlock(fp->rf_dbe);
3800 		return;
3801 	}
3802 	rfs4_dbe_unlock(fp->rf_dbe);
3803 
3804 	/*
3805 	 * Hold as writer to prevent other server threads from
3806 	 * processing requests related to the file while all state is
3807 	 * being removed.
3808 	 */
3809 	rw_enter(&fp->rf_file_rwlock, RW_WRITER);
3810 
3811 	/* Remove ALL state from the file */
3812 	while (sp = rfs4_findstate_by_file(fp)) {
3813 		rfs4_state_close(sp, FALSE, FALSE, CRED());
3814 		rfs4_state_rele_nounlock(sp);
3815 	}
3816 
3817 	/*
3818 	 * This is only safe since there are no further references to
3819 	 * the file.
3820 	 */
3821 	rfs4_dbe_lock(fp->rf_dbe);
3822 	if (fp->rf_vp) {
3823 		vnode_t *vp = fp->rf_vp;
3824 
3825 		mutex_enter(&vp->v_vsd_lock);
3826 		(void) vsd_set(vp, nfs4_srv_vkey, NULL);
3827 		mutex_exit(&vp->v_vsd_lock);
3828 		VN_RELE(vp);
3829 		fp->rf_vp = NULL;
3830 	}
3831 	rfs4_dbe_unlock(fp->rf_dbe);
3832 
3833 	/* Finally let other references to proceed */
3834 	rw_exit(&fp->rf_file_rwlock);
3835 }
3836 
3837 /*
3838  * This function is used as a target for the rfs4_dbe_walk() call
3839  * below.  The purpose of this function is to see if the
3840  * lockowner_state refers to a file that resides within the exportinfo
3841  * export.  If so, then remove the lock_owner state (file locks and
3842  * share "locks") for this object since the intent is the server is
3843  * unexporting the specified directory.  Be sure to invalidate the
3844  * object after the state has been released
3845  */
3846 static void
3847 rfs4_lo_state_walk_callout(rfs4_entry_t u_entry, void *e)
3848 {
3849 	rfs4_lo_state_t *lsp = (rfs4_lo_state_t *)u_entry;
3850 	struct exportinfo *exi = (struct exportinfo *)e;
3851 	nfs_fh4_fmt_t   fhfmt4, *exi_fhp, *finfo_fhp;
3852 	fhandle_t *efhp;
3853 
3854 	efhp = (fhandle_t *)&exi->exi_fh;
3855 	exi_fhp = (nfs_fh4_fmt_t *)&fhfmt4;
3856 
3857 	FH_TO_FMT4(efhp, exi_fhp);
3858 
3859 	finfo_fhp = (nfs_fh4_fmt_t *)lsp->rls_state->rs_finfo->
3860 	    rf_filehandle.nfs_fh4_val;
3861 
3862 	if (EQFSID(&finfo_fhp->fh4_fsid, &exi_fhp->fh4_fsid) &&
3863 	    bcmp(&finfo_fhp->fh4_xdata, &exi_fhp->fh4_xdata,
3864 	    exi_fhp->fh4_xlen) == 0) {
3865 		rfs4_state_close(lsp->rls_state, FALSE, FALSE, CRED());
3866 		rfs4_dbe_invalidate(lsp->rls_dbe);
3867 		rfs4_dbe_invalidate(lsp->rls_state->rs_dbe);
3868 	}
3869 }
3870 
3871 /*
3872  * This function is used as a target for the rfs4_dbe_walk() call
3873  * below.  The purpose of this function is to see if the state refers
3874  * to a file that resides within the exportinfo export.  If so, then
3875  * remove the open state for this object since the intent is the
3876  * server is unexporting the specified directory.  The main result for
3877  * this type of entry is to invalidate it such it will not be found in
3878  * the future.
3879  */
3880 static void
3881 rfs4_state_walk_callout(rfs4_entry_t u_entry, void *e)
3882 {
3883 	rfs4_state_t *sp = (rfs4_state_t *)u_entry;
3884 	struct exportinfo *exi = (struct exportinfo *)e;
3885 	nfs_fh4_fmt_t   fhfmt4, *exi_fhp, *finfo_fhp;
3886 	fhandle_t *efhp;
3887 
3888 	efhp = (fhandle_t *)&exi->exi_fh;
3889 	exi_fhp = (nfs_fh4_fmt_t *)&fhfmt4;
3890 
3891 	FH_TO_FMT4(efhp, exi_fhp);
3892 
3893 	finfo_fhp =
3894 	    (nfs_fh4_fmt_t *)sp->rs_finfo->rf_filehandle.nfs_fh4_val;
3895 
3896 	if (EQFSID(&finfo_fhp->fh4_fsid, &exi_fhp->fh4_fsid) &&
3897 	    bcmp(&finfo_fhp->fh4_xdata, &exi_fhp->fh4_xdata,
3898 	    exi_fhp->fh4_xlen) == 0) {
3899 		rfs4_state_close(sp, TRUE, FALSE, CRED());
3900 		rfs4_dbe_invalidate(sp->rs_dbe);
3901 	}
3902 }
3903 
3904 /*
3905  * This function is used as a target for the rfs4_dbe_walk() call
3906  * below.  The purpose of this function is to see if the state refers
3907  * to a file that resides within the exportinfo export.  If so, then
3908  * remove the deleg state for this object since the intent is the
3909  * server is unexporting the specified directory.  The main result for
3910  * this type of entry is to invalidate it such it will not be found in
3911  * the future.
3912  */
3913 static void
3914 rfs4_deleg_state_walk_callout(rfs4_entry_t u_entry, void *e)
3915 {
3916 	rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
3917 	struct exportinfo *exi = (struct exportinfo *)e;
3918 	nfs_fh4_fmt_t   fhfmt4, *exi_fhp, *finfo_fhp;
3919 	fhandle_t *efhp;
3920 
3921 	efhp = (fhandle_t *)&exi->exi_fh;
3922 	exi_fhp = (nfs_fh4_fmt_t *)&fhfmt4;
3923 
3924 	FH_TO_FMT4(efhp, exi_fhp);
3925 
3926 	finfo_fhp =
3927 	    (nfs_fh4_fmt_t *)dsp->rds_finfo->rf_filehandle.nfs_fh4_val;
3928 
3929 	if (EQFSID(&finfo_fhp->fh4_fsid, &exi_fhp->fh4_fsid) &&
3930 	    bcmp(&finfo_fhp->fh4_xdata, &exi_fhp->fh4_xdata,
3931 	    exi_fhp->fh4_xlen) == 0) {
3932 		rfs4_dbe_invalidate(dsp->rds_dbe);
3933 	}
3934 }
3935 
3936 /*
3937  * This function is used as a target for the rfs4_dbe_walk() call
3938  * below.  The purpose of this function is to see if the state refers
3939  * to a file that resides within the exportinfo export.  If so, then
3940  * release vnode hold for this object since the intent is the server
3941  * is unexporting the specified directory.  Invalidation will prevent
3942  * this struct from being found in the future.
3943  */
3944 static void
3945 rfs4_file_walk_callout(rfs4_entry_t u_entry, void *e)
3946 {
3947 	rfs4_file_t *fp = (rfs4_file_t *)u_entry;
3948 	struct exportinfo *exi = (struct exportinfo *)e;
3949 	nfs_fh4_fmt_t   fhfmt4, *exi_fhp, *finfo_fhp;
3950 	fhandle_t *efhp;
3951 
3952 	efhp = (fhandle_t *)&exi->exi_fh;
3953 	exi_fhp = (nfs_fh4_fmt_t *)&fhfmt4;
3954 
3955 	FH_TO_FMT4(efhp, exi_fhp);
3956 
3957 	finfo_fhp = (nfs_fh4_fmt_t *)fp->rf_filehandle.nfs_fh4_val;
3958 
3959 	if (EQFSID(&finfo_fhp->fh4_fsid, &exi_fhp->fh4_fsid) &&
3960 	    bcmp(&finfo_fhp->fh4_xdata, &exi_fhp->fh4_xdata,
3961 	    exi_fhp->fh4_xlen) == 0) {
3962 		if (fp->rf_vp) {
3963 			vnode_t *vp = fp->rf_vp;
3964 
3965 			/*
3966 			 * don't leak monitors and remove the reference
3967 			 * put on the vnode when the delegation was granted.
3968 			 */
3969 			if (fp->rf_dinfo.rd_dtype == OPEN_DELEGATE_READ) {
3970 				(void) fem_uninstall(vp, deleg_rdops,
3971 				    (void *)fp);
3972 				vn_open_downgrade(vp, FREAD);
3973 			} else if (fp->rf_dinfo.rd_dtype ==
3974 			    OPEN_DELEGATE_WRITE) {
3975 				(void) fem_uninstall(vp, deleg_wrops,
3976 				    (void *)fp);
3977 				vn_open_downgrade(vp, FREAD|FWRITE);
3978 			}
3979 			mutex_enter(&vp->v_vsd_lock);
3980 			(void) vsd_set(vp, nfs4_srv_vkey, NULL);
3981 			mutex_exit(&vp->v_vsd_lock);
3982 			VN_RELE(vp);
3983 			fp->rf_vp = NULL;
3984 		}
3985 		rfs4_dbe_invalidate(fp->rf_dbe);
3986 	}
3987 }
3988 
3989 /*
3990  * Given a directory that is being unexported, cleanup/release all
3991  * state in the server that refers to objects residing underneath this
3992  * particular export.  The ordering of the release is important.
3993  * Lock_owner, then state and then file.
3994  */
3995 void
3996 rfs4_clean_state_exi(struct exportinfo *exi)
3997 {
3998 	mutex_enter(&rfs4_state_lock);
3999 
4000 	if (rfs4_server_state == NULL) {
4001 		mutex_exit(&rfs4_state_lock);
4002 		return;
4003 	}
4004 
4005 	rfs4_dbe_walk(rfs4_lo_state_tab, rfs4_lo_state_walk_callout, exi);
4006 	rfs4_dbe_walk(rfs4_state_tab, rfs4_state_walk_callout, exi);
4007 	rfs4_dbe_walk(rfs4_deleg_state_tab, rfs4_deleg_state_walk_callout, exi);
4008 	rfs4_dbe_walk(rfs4_file_tab, rfs4_file_walk_callout, exi);
4009 
4010 	mutex_exit(&rfs4_state_lock);
4011 }
4012