xref: /illumos-gate/usr/src/uts/common/fs/nfs/nfs4_state.c (revision 382f00c9)
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  */
24 
25 #include <sys/systm.h>
26 #include <sys/kmem.h>
27 #include <sys/cmn_err.h>
28 #include <sys/atomic.h>
29 #include <sys/clconf.h>
30 #include <sys/cladm.h>
31 #include <sys/flock.h>
32 #include <nfs/export.h>
33 #include <nfs/nfs.h>
34 #include <nfs/nfs4.h>
35 #include <nfs/nfssys.h>
36 #include <nfs/lm.h>
37 #include <sys/pathname.h>
38 #include <sys/sdt.h>
39 #include <sys/nvpair.h>
40 
41 extern u_longlong_t nfs4_srv_caller_id;
42 
43 extern time_t rfs4_start_time;
44 extern uint_t nfs4_srv_vkey;
45 
46 stateid4 special0 = {
47 	0,
48 	{ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 }
49 };
50 
51 stateid4 special1 = {
52 	0xffffffff,
53 	{
54 		(char)0xff, (char)0xff, (char)0xff, (char)0xff,
55 		(char)0xff, (char)0xff, (char)0xff, (char)0xff,
56 		(char)0xff, (char)0xff, (char)0xff, (char)0xff
57 	}
58 };
59 
60 
61 #define	ISSPECIAL(id)  (stateid4_cmp(id, &special0) || \
62 			stateid4_cmp(id, &special1))
63 
64 /* For embedding the cluster nodeid into our clientid */
65 #define	CLUSTER_NODEID_SHIFT	24
66 #define	CLUSTER_MAX_NODEID	255
67 
68 #ifdef DEBUG
69 int rfs4_debug;
70 #endif
71 
72 static uint32_t rfs4_database_debug = 0x00;
73 
74 static void rfs4_ss_clid_write(rfs4_client_t *cp, char *leaf);
75 static void rfs4_ss_clid_write_one(rfs4_client_t *cp, char *dir, char *leaf);
76 static void rfs4_dss_clear_oldstate(rfs4_servinst_t *sip);
77 static void rfs4_ss_chkclid_sip(rfs4_client_t *cp, rfs4_servinst_t *sip);
78 
79 /*
80  * Couple of simple init/destroy functions for a general waiter
81  */
82 void
83 rfs4_sw_init(rfs4_state_wait_t *swp)
84 {
85 	mutex_init(swp->sw_cv_lock, NULL, MUTEX_DEFAULT, NULL);
86 	cv_init(swp->sw_cv, NULL, CV_DEFAULT, NULL);
87 	swp->sw_active = FALSE;
88 	swp->sw_wait_count = 0;
89 }
90 
91 void
92 rfs4_sw_destroy(rfs4_state_wait_t *swp)
93 {
94 	mutex_destroy(swp->sw_cv_lock);
95 	cv_destroy(swp->sw_cv);
96 }
97 
98 void
99 rfs4_sw_enter(rfs4_state_wait_t *swp)
100 {
101 	mutex_enter(swp->sw_cv_lock);
102 	while (swp->sw_active) {
103 		swp->sw_wait_count++;
104 		cv_wait(swp->sw_cv, swp->sw_cv_lock);
105 		swp->sw_wait_count--;
106 	}
107 	ASSERT(swp->sw_active == FALSE);
108 	swp->sw_active = TRUE;
109 	mutex_exit(swp->sw_cv_lock);
110 }
111 
112 void
113 rfs4_sw_exit(rfs4_state_wait_t *swp)
114 {
115 	mutex_enter(swp->sw_cv_lock);
116 	ASSERT(swp->sw_active == TRUE);
117 	swp->sw_active = FALSE;
118 	if (swp->sw_wait_count != 0)
119 		cv_broadcast(swp->sw_cv);
120 	mutex_exit(swp->sw_cv_lock);
121 }
122 
123 /*
124  * CPR callback id -- not related to v4 callbacks
125  */
126 static callb_id_t cpr_id = 0;
127 
128 static void
129 deep_lock_copy(LOCK4res *dres, LOCK4res *sres)
130 {
131 	lock_owner4 *slo = &sres->LOCK4res_u.denied.owner;
132 	lock_owner4 *dlo = &dres->LOCK4res_u.denied.owner;
133 
134 	if (sres->status == NFS4ERR_DENIED) {
135 		dlo->owner_val = kmem_alloc(slo->owner_len, KM_SLEEP);
136 		bcopy(slo->owner_val, dlo->owner_val, slo->owner_len);
137 	}
138 }
139 
140 static void
141 deep_lock_free(LOCK4res *res)
142 {
143 	lock_owner4 *lo = &res->LOCK4res_u.denied.owner;
144 
145 	if (res->status == NFS4ERR_DENIED)
146 		kmem_free(lo->owner_val, lo->owner_len);
147 }
148 
149 static void
150 deep_open_copy(OPEN4res *dres, OPEN4res *sres)
151 {
152 	nfsace4 *sacep, *dacep;
153 
154 	if (sres->status != NFS4_OK) {
155 		return;
156 	}
157 
158 	dres->attrset = sres->attrset;
159 
160 	switch (sres->delegation.delegation_type) {
161 	case OPEN_DELEGATE_NONE:
162 		return;
163 	case OPEN_DELEGATE_READ:
164 		sacep = &sres->delegation.open_delegation4_u.read.permissions;
165 		dacep = &dres->delegation.open_delegation4_u.read.permissions;
166 		break;
167 	case OPEN_DELEGATE_WRITE:
168 		sacep = &sres->delegation.open_delegation4_u.write.permissions;
169 		dacep = &dres->delegation.open_delegation4_u.write.permissions;
170 		break;
171 	}
172 	dacep->who.utf8string_val =
173 	    kmem_alloc(sacep->who.utf8string_len, KM_SLEEP);
174 	bcopy(sacep->who.utf8string_val, dacep->who.utf8string_val,
175 	    sacep->who.utf8string_len);
176 }
177 
178 static void
179 deep_open_free(OPEN4res *res)
180 {
181 	nfsace4 *acep;
182 	if (res->status != NFS4_OK)
183 		return;
184 
185 	switch (res->delegation.delegation_type) {
186 	case OPEN_DELEGATE_NONE:
187 		return;
188 	case OPEN_DELEGATE_READ:
189 		acep = &res->delegation.open_delegation4_u.read.permissions;
190 		break;
191 	case OPEN_DELEGATE_WRITE:
192 		acep = &res->delegation.open_delegation4_u.write.permissions;
193 		break;
194 	}
195 
196 	if (acep->who.utf8string_val) {
197 		kmem_free(acep->who.utf8string_val, acep->who.utf8string_len);
198 		acep->who.utf8string_val = NULL;
199 	}
200 }
201 
202 void
203 rfs4_free_reply(nfs_resop4 *rp)
204 {
205 	switch (rp->resop) {
206 	case OP_LOCK:
207 		deep_lock_free(&rp->nfs_resop4_u.oplock);
208 		break;
209 	case OP_OPEN:
210 		deep_open_free(&rp->nfs_resop4_u.opopen);
211 	default:
212 		break;
213 	}
214 }
215 
216 void
217 rfs4_copy_reply(nfs_resop4 *dst, nfs_resop4 *src)
218 {
219 	*dst = *src;
220 
221 	/* Handle responses that need deep copy */
222 	switch (src->resop) {
223 	case OP_LOCK:
224 		deep_lock_copy(&dst->nfs_resop4_u.oplock,
225 		    &src->nfs_resop4_u.oplock);
226 		break;
227 	case OP_OPEN:
228 		deep_open_copy(&dst->nfs_resop4_u.opopen,
229 		    &src->nfs_resop4_u.opopen);
230 		break;
231 	default:
232 		break;
233 	};
234 }
235 
236 /*
237  * This is the implementation of the underlying state engine. The
238  * public interface to this engine is described by
239  * nfs4_state.h. Callers to the engine should hold no state engine
240  * locks when they call in to it. If the protocol needs to lock data
241  * structures it should do so after acquiring all references to them
242  * first and then follow the following lock order:
243  *
244  *	client > openowner > state > lo_state > lockowner > file.
245  *
246  * Internally we only allow a thread to hold one hash bucket lock at a
247  * time and the lock is higher in the lock order (must be acquired
248  * first) than the data structure that is on that hash list.
249  *
250  * If a new reference was acquired by the caller, that reference needs
251  * to be released after releasing all acquired locks with the
252  * corresponding rfs4_*_rele routine.
253  */
254 
255 /*
256  * This code is some what prototypical for now. Its purpose currently is to
257  * implement the interfaces sufficiently to finish the higher protocol
258  * elements. This will be replaced by a dynamically resizeable tables
259  * backed by kmem_cache allocator. However synchronization is handled
260  * correctly (I hope) and will not change by much.  The mutexes for
261  * the hash buckets that can be used to create new instances of data
262  * structures  might be good candidates to evolve into reader writer
263  * locks. If it has to do a creation, it would be holding the
264  * mutex across a kmem_alloc with KM_SLEEP specified.
265  */
266 
267 #ifdef DEBUG
268 #define	TABSIZE 17
269 #else
270 #define	TABSIZE 2047
271 #endif
272 
273 #define	ADDRHASH(key) ((unsigned long)(key) >> 3)
274 
275 /* Used to serialize create/destroy of rfs4_server_state database */
276 kmutex_t	rfs4_state_lock;
277 static rfs4_database_t *rfs4_server_state = NULL;
278 
279 /* Used to serialize lookups of clientids */
280 static	krwlock_t	rfs4_findclient_lock;
281 
282 /*
283  * For now this "table" is exposed so that the CPR callback
284  * function can tromp through it..
285  */
286 rfs4_table_t *rfs4_client_tab;
287 
288 static rfs4_index_t *rfs4_clientid_idx;
289 static rfs4_index_t *rfs4_nfsclnt_idx;
290 static rfs4_table_t *rfs4_clntip_tab;
291 static rfs4_index_t *rfs4_clntip_idx;
292 static rfs4_table_t *rfs4_openowner_tab;
293 static rfs4_index_t *rfs4_openowner_idx;
294 static rfs4_table_t *rfs4_state_tab;
295 static rfs4_index_t *rfs4_state_idx;
296 static rfs4_index_t *rfs4_state_owner_file_idx;
297 static rfs4_index_t *rfs4_state_file_idx;
298 static rfs4_table_t *rfs4_lo_state_tab;
299 static rfs4_index_t *rfs4_lo_state_idx;
300 static rfs4_index_t *rfs4_lo_state_owner_idx;
301 static rfs4_table_t *rfs4_lockowner_tab;
302 static rfs4_index_t *rfs4_lockowner_idx;
303 static rfs4_index_t *rfs4_lockowner_pid_idx;
304 static rfs4_table_t *rfs4_file_tab;
305 static rfs4_index_t *rfs4_file_idx;
306 static rfs4_table_t *rfs4_deleg_state_tab;
307 static rfs4_index_t *rfs4_deleg_idx;
308 static rfs4_index_t *rfs4_deleg_state_idx;
309 
310 #define	MAXTABSZ 1024*1024
311 
312 /* The values below are rfs4_lease_time units */
313 
314 #ifdef DEBUG
315 #define	CLIENT_CACHE_TIME 1
316 #define	OPENOWNER_CACHE_TIME 1
317 #define	STATE_CACHE_TIME 1
318 #define	LO_STATE_CACHE_TIME 1
319 #define	LOCKOWNER_CACHE_TIME 1
320 #define	FILE_CACHE_TIME 3
321 #define	DELEG_STATE_CACHE_TIME 1
322 #else
323 #define	CLIENT_CACHE_TIME 10
324 #define	OPENOWNER_CACHE_TIME 5
325 #define	STATE_CACHE_TIME 1
326 #define	LO_STATE_CACHE_TIME 1
327 #define	LOCKOWNER_CACHE_TIME 3
328 #define	FILE_CACHE_TIME 40
329 #define	DELEG_STATE_CACHE_TIME 1
330 #endif
331 
332 
333 static time_t rfs4_client_cache_time = 0;
334 static time_t rfs4_clntip_cache_time = 0;
335 static time_t rfs4_openowner_cache_time = 0;
336 static time_t rfs4_state_cache_time = 0;
337 static time_t rfs4_lo_state_cache_time = 0;
338 static time_t rfs4_lockowner_cache_time = 0;
339 static time_t rfs4_file_cache_time = 0;
340 static time_t rfs4_deleg_state_cache_time = 0;
341 
342 static bool_t rfs4_client_create(rfs4_entry_t, void *);
343 static void rfs4_dss_remove_cpleaf(rfs4_client_t *);
344 static void rfs4_dss_remove_leaf(rfs4_servinst_t *, char *, char *);
345 static void rfs4_client_destroy(rfs4_entry_t);
346 static bool_t rfs4_client_expiry(rfs4_entry_t);
347 static uint32_t clientid_hash(void *);
348 static bool_t clientid_compare(rfs4_entry_t, void *);
349 static void *clientid_mkkey(rfs4_entry_t);
350 static uint32_t nfsclnt_hash(void *);
351 static bool_t nfsclnt_compare(rfs4_entry_t, void *);
352 static void *nfsclnt_mkkey(rfs4_entry_t);
353 static bool_t rfs4_clntip_expiry(rfs4_entry_t);
354 static void rfs4_clntip_destroy(rfs4_entry_t);
355 static bool_t rfs4_clntip_create(rfs4_entry_t, void *);
356 static uint32_t clntip_hash(void *);
357 static bool_t clntip_compare(rfs4_entry_t, void *);
358 static void *clntip_mkkey(rfs4_entry_t);
359 static bool_t rfs4_openowner_create(rfs4_entry_t, void *);
360 static void rfs4_openowner_destroy(rfs4_entry_t);
361 static bool_t rfs4_openowner_expiry(rfs4_entry_t);
362 static uint32_t openowner_hash(void *);
363 static bool_t openowner_compare(rfs4_entry_t, void *);
364 static void *openowner_mkkey(rfs4_entry_t);
365 static bool_t rfs4_state_create(rfs4_entry_t, void *);
366 static void rfs4_state_destroy(rfs4_entry_t);
367 static bool_t rfs4_state_expiry(rfs4_entry_t);
368 static uint32_t state_hash(void *);
369 static bool_t state_compare(rfs4_entry_t, void *);
370 static void *state_mkkey(rfs4_entry_t);
371 static uint32_t state_owner_file_hash(void *);
372 static bool_t state_owner_file_compare(rfs4_entry_t, void *);
373 static void *state_owner_file_mkkey(rfs4_entry_t);
374 static uint32_t state_file_hash(void *);
375 static bool_t state_file_compare(rfs4_entry_t, void *);
376 static void *state_file_mkkey(rfs4_entry_t);
377 static bool_t rfs4_lo_state_create(rfs4_entry_t, void *);
378 static void rfs4_lo_state_destroy(rfs4_entry_t);
379 static bool_t rfs4_lo_state_expiry(rfs4_entry_t);
380 static uint32_t lo_state_hash(void *);
381 static bool_t lo_state_compare(rfs4_entry_t, void *);
382 static void *lo_state_mkkey(rfs4_entry_t);
383 static uint32_t lo_state_lo_hash(void *);
384 static bool_t lo_state_lo_compare(rfs4_entry_t, void *);
385 static void *lo_state_lo_mkkey(rfs4_entry_t);
386 static bool_t rfs4_lockowner_create(rfs4_entry_t, void *);
387 static void rfs4_lockowner_destroy(rfs4_entry_t);
388 static bool_t rfs4_lockowner_expiry(rfs4_entry_t);
389 static uint32_t lockowner_hash(void *);
390 static bool_t lockowner_compare(rfs4_entry_t, void *);
391 static void *lockowner_mkkey(rfs4_entry_t);
392 static uint32_t pid_hash(void *);
393 static bool_t pid_compare(rfs4_entry_t, void *);
394 static void *pid_mkkey(rfs4_entry_t);
395 static bool_t rfs4_file_create(rfs4_entry_t, void *);
396 static void rfs4_file_destroy(rfs4_entry_t);
397 static uint32_t file_hash(void *);
398 static bool_t file_compare(rfs4_entry_t, void *);
399 static void *file_mkkey(rfs4_entry_t);
400 static bool_t rfs4_deleg_state_create(rfs4_entry_t, void *);
401 static void rfs4_deleg_state_destroy(rfs4_entry_t);
402 static bool_t rfs4_deleg_state_expiry(rfs4_entry_t);
403 static uint32_t deleg_hash(void *);
404 static bool_t deleg_compare(rfs4_entry_t, void *);
405 static void *deleg_mkkey(rfs4_entry_t);
406 static uint32_t deleg_state_hash(void *);
407 static bool_t deleg_state_compare(rfs4_entry_t, void *);
408 static void *deleg_state_mkkey(rfs4_entry_t);
409 
410 static void rfs4_state_rele_nounlock(rfs4_state_t *);
411 
412 static int rfs4_ss_enabled = 0;
413 
414 extern void (*rfs4_client_clrst)(struct nfs4clrst_args *);
415 
416 void
417 rfs4_ss_pnfree(rfs4_ss_pn_t *ss_pn)
418 {
419 	kmem_free(ss_pn, sizeof (rfs4_ss_pn_t));
420 }
421 
422 static rfs4_ss_pn_t *
423 rfs4_ss_pnalloc(char *dir, char *leaf)
424 {
425 	rfs4_ss_pn_t *ss_pn;
426 	int 	dir_len, leaf_len;
427 
428 	/*
429 	 * validate we have a resonable path
430 	 * (account for the '/' and trailing null)
431 	 */
432 	if ((dir_len = strlen(dir)) > MAXPATHLEN ||
433 	    (leaf_len = strlen(leaf)) > MAXNAMELEN ||
434 	    (dir_len + leaf_len + 2) > MAXPATHLEN) {
435 		return (NULL);
436 	}
437 
438 	ss_pn = kmem_alloc(sizeof (rfs4_ss_pn_t), KM_SLEEP);
439 
440 	(void) snprintf(ss_pn->pn, MAXPATHLEN, "%s/%s", dir, leaf);
441 	/* Handy pointer to just the leaf name */
442 	ss_pn->leaf = ss_pn->pn + dir_len + 1;
443 	return (ss_pn);
444 }
445 
446 
447 /*
448  * Move the "leaf" filename from "sdir" directory
449  * to the "ddir" directory. Return the pathname of
450  * the destination unless the rename fails in which
451  * case we need to return the source pathname.
452  */
453 static rfs4_ss_pn_t *
454 rfs4_ss_movestate(char *sdir, char *ddir, char *leaf)
455 {
456 	rfs4_ss_pn_t *src, *dst;
457 
458 	if ((src = rfs4_ss_pnalloc(sdir, leaf)) == NULL)
459 		return (NULL);
460 
461 	if ((dst = rfs4_ss_pnalloc(ddir, leaf)) == NULL) {
462 		rfs4_ss_pnfree(src);
463 		return (NULL);
464 	}
465 
466 	/*
467 	 * If the rename fails we shall return the src
468 	 * pathname and free the dst. Otherwise we need
469 	 * to free the src and return the dst pathanme.
470 	 */
471 	if (vn_rename(src->pn, dst->pn, UIO_SYSSPACE)) {
472 		rfs4_ss_pnfree(dst);
473 		return (src);
474 	}
475 	rfs4_ss_pnfree(src);
476 	return (dst);
477 }
478 
479 
480 static rfs4_oldstate_t *
481 rfs4_ss_getstate(vnode_t *dvp, rfs4_ss_pn_t *ss_pn)
482 {
483 	struct uio uio;
484 	struct iovec iov[3];
485 
486 	rfs4_oldstate_t *cl_ss = NULL;
487 	vnode_t *vp;
488 	vattr_t va;
489 	uint_t id_len;
490 	int err, kill_file, file_vers;
491 
492 	if (ss_pn == NULL)
493 		return (NULL);
494 
495 	/*
496 	 * open the state file.
497 	 */
498 	if (vn_open(ss_pn->pn, UIO_SYSSPACE, FREAD, 0, &vp, 0, 0) != 0) {
499 		return (NULL);
500 	}
501 
502 	if (vp->v_type != VREG) {
503 		(void) VOP_CLOSE(vp, FREAD, 1, (offset_t)0, CRED(), NULL);
504 		VN_RELE(vp);
505 		return (NULL);
506 	}
507 
508 	err = VOP_ACCESS(vp, VREAD, 0, CRED(), NULL);
509 	if (err) {
510 		/*
511 		 * We don't have read access? better get the heck out.
512 		 */
513 		(void) VOP_CLOSE(vp, FREAD, 1, (offset_t)0, CRED(), NULL);
514 		VN_RELE(vp);
515 		return (NULL);
516 	}
517 
518 	(void) VOP_RWLOCK(vp, V_WRITELOCK_FALSE, NULL);
519 	/*
520 	 * get the file size to do some basic validation
521 	 */
522 	va.va_mask = AT_SIZE;
523 	err = VOP_GETATTR(vp, &va, 0, CRED(), NULL);
524 
525 	kill_file = (va.va_size == 0 || va.va_size <
526 	    (NFS4_VERIFIER_SIZE + sizeof (uint_t)+1));
527 
528 	if (err || kill_file) {
529 		VOP_RWUNLOCK(vp, V_WRITELOCK_FALSE, NULL);
530 		(void) VOP_CLOSE(vp, FREAD, 1, (offset_t)0, CRED(), NULL);
531 		VN_RELE(vp);
532 		if (kill_file) {
533 			(void) VOP_REMOVE(dvp, ss_pn->leaf, CRED(), NULL, 0);
534 		}
535 		return (NULL);
536 	}
537 
538 	cl_ss = kmem_alloc(sizeof (rfs4_oldstate_t), KM_SLEEP);
539 
540 	/*
541 	 * build iovecs to read in the file_version, verifier and id_len
542 	 */
543 	iov[0].iov_base = (caddr_t)&file_vers;
544 	iov[0].iov_len = sizeof (int);
545 	iov[1].iov_base = (caddr_t)&cl_ss->cl_id4.verifier;
546 	iov[1].iov_len = NFS4_VERIFIER_SIZE;
547 	iov[2].iov_base = (caddr_t)&id_len;
548 	iov[2].iov_len = sizeof (uint_t);
549 
550 	uio.uio_iov = iov;
551 	uio.uio_iovcnt = 3;
552 	uio.uio_segflg = UIO_SYSSPACE;
553 	uio.uio_loffset = 0;
554 	uio.uio_resid = sizeof (int) + NFS4_VERIFIER_SIZE + sizeof (uint_t);
555 
556 	if (err = VOP_READ(vp, &uio, FREAD, CRED(), NULL)) {
557 		VOP_RWUNLOCK(vp, V_WRITELOCK_FALSE, NULL);
558 		(void) VOP_CLOSE(vp, FREAD, 1, (offset_t)0, CRED(), NULL);
559 		VN_RELE(vp);
560 		kmem_free(cl_ss, sizeof (rfs4_oldstate_t));
561 		return (NULL);
562 	}
563 
564 	/*
565 	 * if the file_version doesn't match or if the
566 	 * id_len is zero or the combination of the verifier,
567 	 * id_len and id_val is bigger than the file we have
568 	 * a problem. If so ditch the file.
569 	 */
570 	kill_file = (file_vers != NFS4_SS_VERSION || id_len == 0 ||
571 	    (id_len + NFS4_VERIFIER_SIZE + sizeof (uint_t)) > va.va_size);
572 
573 	if (err || kill_file) {
574 		VOP_RWUNLOCK(vp, V_WRITELOCK_FALSE, NULL);
575 		(void) VOP_CLOSE(vp, FREAD, 1, (offset_t)0, CRED(), NULL);
576 		VN_RELE(vp);
577 		kmem_free(cl_ss, sizeof (rfs4_oldstate_t));
578 		if (kill_file) {
579 			(void) VOP_REMOVE(dvp, ss_pn->leaf, CRED(), NULL, 0);
580 		}
581 		return (NULL);
582 	}
583 
584 	/*
585 	 * now get the client id value
586 	 */
587 	cl_ss->cl_id4.id_val = kmem_alloc(id_len, KM_SLEEP);
588 	iov[0].iov_base = cl_ss->cl_id4.id_val;
589 	iov[0].iov_len = id_len;
590 
591 	uio.uio_iov = iov;
592 	uio.uio_iovcnt = 1;
593 	uio.uio_segflg = UIO_SYSSPACE;
594 	uio.uio_resid = cl_ss->cl_id4.id_len = id_len;
595 
596 	if (err = VOP_READ(vp, &uio, FREAD, CRED(), NULL)) {
597 		VOP_RWUNLOCK(vp, V_WRITELOCK_FALSE, NULL);
598 		(void) VOP_CLOSE(vp, FREAD, 1, (offset_t)0, CRED(), NULL);
599 		VN_RELE(vp);
600 		kmem_free(cl_ss->cl_id4.id_val, id_len);
601 		kmem_free(cl_ss, sizeof (rfs4_oldstate_t));
602 		return (NULL);
603 	}
604 
605 	VOP_RWUNLOCK(vp, V_WRITELOCK_FALSE, NULL);
606 	(void) VOP_CLOSE(vp, FREAD, 1, (offset_t)0, CRED(), NULL);
607 	VN_RELE(vp);
608 	return (cl_ss);
609 }
610 
611 #ifdef	nextdp
612 #undef nextdp
613 #endif
614 #define	nextdp(dp)	((struct dirent64 *)((char *)(dp) + (dp)->d_reclen))
615 
616 /*
617  * Add entries from statedir to supplied oldstate list.
618  * Optionally, move all entries from statedir -> destdir.
619  */
620 void
621 rfs4_ss_oldstate(rfs4_oldstate_t *oldstate, char *statedir, char *destdir)
622 {
623 	rfs4_ss_pn_t *ss_pn;
624 	rfs4_oldstate_t *cl_ss = NULL;
625 	char	*dirt = NULL;
626 	int	err, dir_eof = 0, size = 0;
627 	vnode_t *dvp;
628 	struct iovec iov;
629 	struct uio uio;
630 	struct dirent64 *dep;
631 	offset_t dirchunk_offset = 0;
632 
633 	/*
634 	 * open the state directory
635 	 */
636 	if (vn_open(statedir, UIO_SYSSPACE, FREAD, 0, &dvp, 0, 0))
637 		return;
638 
639 	if (dvp->v_type != VDIR || VOP_ACCESS(dvp, VREAD, 0, CRED(), NULL))
640 		goto out;
641 
642 	dirt = kmem_alloc(RFS4_SS_DIRSIZE, KM_SLEEP);
643 
644 	/*
645 	 * Get and process the directory entries
646 	 */
647 	while (!dir_eof) {
648 		(void) VOP_RWLOCK(dvp, V_WRITELOCK_FALSE, NULL);
649 		iov.iov_base = dirt;
650 		iov.iov_len = RFS4_SS_DIRSIZE;
651 		uio.uio_iov = &iov;
652 		uio.uio_iovcnt = 1;
653 		uio.uio_segflg = UIO_SYSSPACE;
654 		uio.uio_loffset = dirchunk_offset;
655 		uio.uio_resid = RFS4_SS_DIRSIZE;
656 
657 		err = VOP_READDIR(dvp, &uio, CRED(), &dir_eof, NULL, 0);
658 		VOP_RWUNLOCK(dvp, V_WRITELOCK_FALSE, NULL);
659 		if (err)
660 			goto out;
661 
662 		size = RFS4_SS_DIRSIZE - uio.uio_resid;
663 
664 		/*
665 		 * Process all the directory entries in this
666 		 * readdir chunk
667 		 */
668 		for (dep = (struct dirent64 *)dirt; size > 0;
669 		    dep = nextdp(dep)) {
670 
671 			size -= dep->d_reclen;
672 			dirchunk_offset = dep->d_off;
673 
674 			/*
675 			 * Skip '.' and '..'
676 			 */
677 			if (NFS_IS_DOTNAME(dep->d_name))
678 				continue;
679 
680 			ss_pn = rfs4_ss_pnalloc(statedir, dep->d_name);
681 			if (ss_pn == NULL)
682 				continue;
683 
684 			if (cl_ss = rfs4_ss_getstate(dvp, ss_pn)) {
685 				if (destdir != NULL) {
686 					rfs4_ss_pnfree(ss_pn);
687 					cl_ss->ss_pn = rfs4_ss_movestate(
688 					    statedir, destdir, dep->d_name);
689 				} else {
690 					cl_ss->ss_pn = ss_pn;
691 				}
692 				insque(cl_ss, oldstate);
693 			} else {
694 				rfs4_ss_pnfree(ss_pn);
695 			}
696 		}
697 	}
698 
699 out:
700 	(void) VOP_CLOSE(dvp, FREAD, 1, (offset_t)0, CRED(), NULL);
701 	VN_RELE(dvp);
702 	if (dirt)
703 		kmem_free((caddr_t)dirt, RFS4_SS_DIRSIZE);
704 }
705 
706 static void
707 rfs4_ss_init(void)
708 {
709 	int npaths = 1;
710 	char *default_dss_path = NFS4_DSS_VAR_DIR;
711 
712 	/* read the default stable storage state */
713 	rfs4_dss_readstate(npaths, &default_dss_path);
714 
715 	rfs4_ss_enabled = 1;
716 }
717 
718 static void
719 rfs4_ss_fini(void)
720 {
721 	rfs4_servinst_t *sip;
722 
723 	mutex_enter(&rfs4_servinst_lock);
724 	sip = rfs4_cur_servinst;
725 	while (sip != NULL) {
726 		rfs4_dss_clear_oldstate(sip);
727 		sip = sip->next;
728 	}
729 	mutex_exit(&rfs4_servinst_lock);
730 }
731 
732 /*
733  * Remove all oldstate files referenced by this servinst.
734  */
735 static void
736 rfs4_dss_clear_oldstate(rfs4_servinst_t *sip)
737 {
738 	rfs4_oldstate_t *os_head, *osp;
739 
740 	rw_enter(&sip->oldstate_lock, RW_WRITER);
741 	os_head = sip->oldstate;
742 
743 	if (os_head == NULL) {
744 		rw_exit(&sip->oldstate_lock);
745 		return;
746 	}
747 
748 	/* skip dummy entry */
749 	osp = os_head->next;
750 	while (osp != os_head) {
751 		char *leaf = osp->ss_pn->leaf;
752 		rfs4_oldstate_t *os_next;
753 
754 		rfs4_dss_remove_leaf(sip, NFS4_DSS_OLDSTATE_LEAF, leaf);
755 
756 		if (osp->cl_id4.id_val)
757 			kmem_free(osp->cl_id4.id_val, osp->cl_id4.id_len);
758 		rfs4_ss_pnfree(osp->ss_pn);
759 
760 		os_next = osp->next;
761 		remque(osp);
762 		kmem_free(osp, sizeof (rfs4_oldstate_t));
763 		osp = os_next;
764 	}
765 
766 	rw_exit(&sip->oldstate_lock);
767 }
768 
769 /*
770  * Form the state and oldstate paths, and read in the stable storage files.
771  */
772 void
773 rfs4_dss_readstate(int npaths, char **paths)
774 {
775 	int i;
776 	char *state, *oldstate;
777 
778 	state = kmem_alloc(MAXPATHLEN, KM_SLEEP);
779 	oldstate = kmem_alloc(MAXPATHLEN, KM_SLEEP);
780 
781 	for (i = 0; i < npaths; i++) {
782 		char *path = paths[i];
783 
784 		(void) sprintf(state, "%s/%s", path, NFS4_DSS_STATE_LEAF);
785 		(void) sprintf(oldstate, "%s/%s", path, NFS4_DSS_OLDSTATE_LEAF);
786 
787 		/*
788 		 * Populate the current server instance's oldstate list.
789 		 *
790 		 * 1. Read stable storage data from old state directory,
791 		 *    leaving its contents alone.
792 		 *
793 		 * 2. Read stable storage data from state directory,
794 		 *    and move the latter's contents to old state
795 		 *    directory.
796 		 */
797 		rfs4_ss_oldstate(rfs4_cur_servinst->oldstate, oldstate, NULL);
798 		rfs4_ss_oldstate(rfs4_cur_servinst->oldstate, state, oldstate);
799 	}
800 
801 	kmem_free(state, MAXPATHLEN);
802 	kmem_free(oldstate, MAXPATHLEN);
803 }
804 
805 
806 /*
807  * Check if we are still in grace and if the client can be
808  * granted permission to perform reclaims.
809  */
810 void
811 rfs4_ss_chkclid(rfs4_client_t *cp)
812 {
813 	rfs4_servinst_t *sip;
814 
815 	/*
816 	 * It should be sufficient to check the oldstate data for just
817 	 * this client's instance. However, since our per-instance
818 	 * client grouping is solely temporal, HA-NFSv4 RG failover
819 	 * might result in clients of the same RG being partitioned into
820 	 * separate instances.
821 	 *
822 	 * Until the client grouping is improved, we must check the
823 	 * oldstate data for all instances with an active grace period.
824 	 *
825 	 * This also serves as the mechanism to remove stale oldstate data.
826 	 * The first time we check an instance after its grace period has
827 	 * expired, the oldstate data should be cleared.
828 	 *
829 	 * Start at the current instance, and walk the list backwards
830 	 * to the first.
831 	 */
832 	mutex_enter(&rfs4_servinst_lock);
833 	for (sip = rfs4_cur_servinst; sip != NULL; sip = sip->prev) {
834 		rfs4_ss_chkclid_sip(cp, sip);
835 
836 		/* if the above check found this client, we're done */
837 		if (cp->rc_can_reclaim)
838 			break;
839 	}
840 	mutex_exit(&rfs4_servinst_lock);
841 }
842 
843 static void
844 rfs4_ss_chkclid_sip(rfs4_client_t *cp, rfs4_servinst_t *sip)
845 {
846 	rfs4_oldstate_t *osp, *os_head;
847 
848 	/* short circuit everything if this server instance has no oldstate */
849 	rw_enter(&sip->oldstate_lock, RW_READER);
850 	os_head = sip->oldstate;
851 	rw_exit(&sip->oldstate_lock);
852 	if (os_head == NULL)
853 		return;
854 
855 	/*
856 	 * If this server instance is no longer in a grace period then
857 	 * the client won't be able to reclaim. No further need for this
858 	 * instance's oldstate data, so it can be cleared.
859 	 */
860 	if (!rfs4_servinst_in_grace(sip))
861 		return;
862 
863 	/* this instance is still in grace; search for the clientid */
864 
865 	rw_enter(&sip->oldstate_lock, RW_READER);
866 
867 	os_head = sip->oldstate;
868 	/* skip dummy entry */
869 	osp = os_head->next;
870 	while (osp != os_head) {
871 		if (osp->cl_id4.id_len == cp->rc_nfs_client.id_len) {
872 			if (bcmp(osp->cl_id4.id_val, cp->rc_nfs_client.id_val,
873 			    osp->cl_id4.id_len) == 0) {
874 				cp->rc_can_reclaim = 1;
875 				break;
876 			}
877 		}
878 		osp = osp->next;
879 	}
880 
881 	rw_exit(&sip->oldstate_lock);
882 }
883 
884 /*
885  * Place client information into stable storage: 1/3.
886  * First, generate the leaf filename, from the client's IP address and
887  * the server-generated short-hand clientid.
888  */
889 void
890 rfs4_ss_clid(rfs4_client_t *cp)
891 {
892 	const char *kinet_ntop6(uchar_t *, char *, size_t);
893 	char leaf[MAXNAMELEN], buf[INET6_ADDRSTRLEN];
894 	struct sockaddr *ca;
895 	uchar_t *b;
896 
897 	if (rfs4_ss_enabled == 0) {
898 		return;
899 	}
900 
901 	buf[0] = 0;
902 
903 	ca = (struct sockaddr *)&cp->rc_addr;
904 
905 	/*
906 	 * Convert the caller's IP address to a dotted string
907 	 */
908 	if (ca->sa_family == AF_INET) {
909 		b = (uchar_t *)&((struct sockaddr_in *)ca)->sin_addr;
910 		(void) sprintf(buf, "%03d.%03d.%03d.%03d", b[0] & 0xFF,
911 		    b[1] & 0xFF, b[2] & 0xFF, b[3] & 0xFF);
912 	} else if (ca->sa_family == AF_INET6) {
913 		struct sockaddr_in6 *sin6;
914 
915 		sin6 = (struct sockaddr_in6 *)ca;
916 		(void) kinet_ntop6((uchar_t *)&sin6->sin6_addr,
917 		    buf, INET6_ADDRSTRLEN);
918 	}
919 
920 	(void) snprintf(leaf, MAXNAMELEN, "%s-%llx", buf,
921 	    (longlong_t)cp->rc_clientid);
922 	rfs4_ss_clid_write(cp, leaf);
923 }
924 
925 /*
926  * Place client information into stable storage: 2/3.
927  * DSS: distributed stable storage: the file may need to be written to
928  * multiple directories.
929  */
930 static void
931 rfs4_ss_clid_write(rfs4_client_t *cp, char *leaf)
932 {
933 	rfs4_servinst_t *sip;
934 
935 	/*
936 	 * It should be sufficient to write the leaf file to (all) DSS paths
937 	 * associated with just this client's instance. However, since our
938 	 * per-instance client grouping is solely temporal, HA-NFSv4 RG
939 	 * failover might result in us losing DSS data.
940 	 *
941 	 * Until the client grouping is improved, we must write the DSS data
942 	 * to all instances' paths. Start at the current instance, and
943 	 * walk the list backwards to the first.
944 	 */
945 	mutex_enter(&rfs4_servinst_lock);
946 	for (sip = rfs4_cur_servinst; sip != NULL; sip = sip->prev) {
947 		int i, npaths = sip->dss_npaths;
948 
949 		/* write the leaf file to all DSS paths */
950 		for (i = 0; i < npaths; i++) {
951 			rfs4_dss_path_t *dss_path = sip->dss_paths[i];
952 
953 			/* HA-NFSv4 path might have been failed-away from us */
954 			if (dss_path == NULL)
955 				continue;
956 
957 			rfs4_ss_clid_write_one(cp, dss_path->path, leaf);
958 		}
959 	}
960 	mutex_exit(&rfs4_servinst_lock);
961 }
962 
963 /*
964  * Place client information into stable storage: 3/3.
965  * Write the stable storage data to the requested file.
966  */
967 static void
968 rfs4_ss_clid_write_one(rfs4_client_t *cp, char *dss_path, char *leaf)
969 {
970 	int ioflag;
971 	int file_vers = NFS4_SS_VERSION;
972 	size_t dirlen;
973 	struct uio uio;
974 	struct iovec iov[4];
975 	char *dir;
976 	rfs4_ss_pn_t *ss_pn;
977 	vnode_t *vp;
978 	nfs_client_id4 *cl_id4 = &(cp->rc_nfs_client);
979 
980 	/* allow 2 extra bytes for '/' & NUL */
981 	dirlen = strlen(dss_path) + strlen(NFS4_DSS_STATE_LEAF) + 2;
982 	dir = kmem_alloc(dirlen, KM_SLEEP);
983 	(void) sprintf(dir, "%s/%s", dss_path, NFS4_DSS_STATE_LEAF);
984 
985 	ss_pn = rfs4_ss_pnalloc(dir, leaf);
986 	/* rfs4_ss_pnalloc takes its own copy */
987 	kmem_free(dir, dirlen);
988 	if (ss_pn == NULL)
989 		return;
990 
991 	if (vn_open(ss_pn->pn, UIO_SYSSPACE, FCREAT|FWRITE, 0600, &vp,
992 	    CRCREAT, 0)) {
993 		rfs4_ss_pnfree(ss_pn);
994 		return;
995 	}
996 
997 	/*
998 	 * We need to record leaf - i.e. the filename - so that we know
999 	 * what to remove, in the future. However, the dir part of cp->ss_pn
1000 	 * should never be referenced directly, since it's potentially only
1001 	 * one of several paths with this leaf in it.
1002 	 */
1003 	if (cp->rc_ss_pn != NULL) {
1004 		if (strcmp(cp->rc_ss_pn->leaf, leaf) == 0) {
1005 			/* we've already recorded *this* leaf */
1006 			rfs4_ss_pnfree(ss_pn);
1007 		} else {
1008 			/* replace with this leaf */
1009 			rfs4_ss_pnfree(cp->rc_ss_pn);
1010 			cp->rc_ss_pn = ss_pn;
1011 		}
1012 	} else {
1013 		cp->rc_ss_pn = ss_pn;
1014 	}
1015 
1016 	/*
1017 	 * Build a scatter list that points to the nfs_client_id4
1018 	 */
1019 	iov[0].iov_base = (caddr_t)&file_vers;
1020 	iov[0].iov_len = sizeof (int);
1021 	iov[1].iov_base = (caddr_t)&(cl_id4->verifier);
1022 	iov[1].iov_len = NFS4_VERIFIER_SIZE;
1023 	iov[2].iov_base = (caddr_t)&(cl_id4->id_len);
1024 	iov[2].iov_len = sizeof (uint_t);
1025 	iov[3].iov_base = (caddr_t)cl_id4->id_val;
1026 	iov[3].iov_len = cl_id4->id_len;
1027 
1028 	uio.uio_iov = iov;
1029 	uio.uio_iovcnt = 4;
1030 	uio.uio_loffset = 0;
1031 	uio.uio_segflg = UIO_SYSSPACE;
1032 	uio.uio_llimit = (rlim64_t)MAXOFFSET_T;
1033 	uio.uio_resid = cl_id4->id_len + sizeof (int) +
1034 	    NFS4_VERIFIER_SIZE + sizeof (uint_t);
1035 
1036 	ioflag = uio.uio_fmode = (FWRITE|FSYNC);
1037 	uio.uio_extflg = UIO_COPY_DEFAULT;
1038 
1039 	(void) VOP_RWLOCK(vp, V_WRITELOCK_TRUE, NULL);
1040 	/* write the full client id to the file. */
1041 	(void) VOP_WRITE(vp, &uio, ioflag, CRED(), NULL);
1042 	VOP_RWUNLOCK(vp, V_WRITELOCK_TRUE, NULL);
1043 
1044 	(void) VOP_CLOSE(vp, FWRITE, 1, (offset_t)0, CRED(), NULL);
1045 	VN_RELE(vp);
1046 }
1047 
1048 /*
1049  * DSS: distributed stable storage.
1050  * Unpack the list of paths passed by nfsd.
1051  * Use nvlist_alloc(9F) to manage the data.
1052  * The caller is responsible for allocating and freeing the buffer.
1053  */
1054 int
1055 rfs4_dss_setpaths(char *buf, size_t buflen)
1056 {
1057 	int error;
1058 
1059 	/*
1060 	 * If this is a "warm start", i.e. we previously had DSS paths,
1061 	 * preserve the old paths.
1062 	 */
1063 	if (rfs4_dss_paths != NULL) {
1064 		/*
1065 		 * Before we lose the ptr, destroy the nvlist and pathnames
1066 		 * array from the warm start before this one.
1067 		 */
1068 		if (rfs4_dss_oldpaths)
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 
1169 	mutex_enter(&rfs4_state_lock);
1170 
1171 	/*
1172 	 * If the server state database has already been initialized,
1173 	 * skip it
1174 	 */
1175 	if (rfs4_server_state != NULL) {
1176 		mutex_exit(&rfs4_state_lock);
1177 		return;
1178 	}
1179 
1180 	rw_init(&rfs4_findclient_lock, NULL, RW_DEFAULT, NULL);
1181 
1182 	/*
1183 	 * Set the boot time.  If the server
1184 	 * has been restarted quickly and has had the opportunity to
1185 	 * service clients, then the start_time needs to be bumped
1186 	 * regardless.  A small window but it exists...
1187 	 */
1188 	if (rfs4_start_time != gethrestime_sec())
1189 		rfs4_start_time = gethrestime_sec();
1190 	else
1191 		rfs4_start_time++;
1192 
1193 	/* DSS: distributed stable storage: initialise served paths list */
1194 	rfs4_dss_pathlist = NULL;
1195 
1196 	/*
1197 	 * Create the first server instance, or a new one if the server has
1198 	 * been restarted; see above comments on rfs4_start_time. Don't
1199 	 * start its grace period; that will be done later, to maximise the
1200 	 * clients' recovery window.
1201 	 */
1202 	start_grace = 0;
1203 	rfs4_servinst_create(start_grace, 1, &dss_path);
1204 
1205 	/* reset the "first NFSv4 request" status */
1206 	rfs4_seen_first_compound = 0;
1207 
1208 	/*
1209 	 * Add a CPR callback so that we can update client
1210 	 * access times to extend the lease after a suspend
1211 	 * and resume (using the same class as rpcmod/connmgr)
1212 	 */
1213 	cpr_id = callb_add(rfs4_cpr_callb, 0, CB_CL_CPR_RPC, "rfs4");
1214 
1215 	/* set the various cache timers for table creation */
1216 	if (rfs4_client_cache_time == 0)
1217 		rfs4_client_cache_time = CLIENT_CACHE_TIME;
1218 	if (rfs4_openowner_cache_time == 0)
1219 		rfs4_openowner_cache_time = OPENOWNER_CACHE_TIME;
1220 	if (rfs4_state_cache_time == 0)
1221 		rfs4_state_cache_time = STATE_CACHE_TIME;
1222 	if (rfs4_lo_state_cache_time == 0)
1223 		rfs4_lo_state_cache_time = LO_STATE_CACHE_TIME;
1224 	if (rfs4_lockowner_cache_time == 0)
1225 		rfs4_lockowner_cache_time = LOCKOWNER_CACHE_TIME;
1226 	if (rfs4_file_cache_time == 0)
1227 		rfs4_file_cache_time = FILE_CACHE_TIME;
1228 	if (rfs4_deleg_state_cache_time == 0)
1229 		rfs4_deleg_state_cache_time = DELEG_STATE_CACHE_TIME;
1230 
1231 	/* Create the overall database to hold all server state */
1232 	rfs4_server_state = rfs4_database_create(rfs4_database_debug);
1233 
1234 	/* Now create the individual tables */
1235 	rfs4_client_cache_time *= rfs4_lease_time;
1236 	rfs4_client_tab = rfs4_table_create(rfs4_server_state,
1237 	    "Client",
1238 	    rfs4_client_cache_time,
1239 	    2,
1240 	    rfs4_client_create,
1241 	    rfs4_client_destroy,
1242 	    rfs4_client_expiry,
1243 	    sizeof (rfs4_client_t),
1244 	    TABSIZE,
1245 	    MAXTABSZ/8, 100);
1246 	rfs4_nfsclnt_idx = rfs4_index_create(rfs4_client_tab,
1247 	    "nfs_client_id4", nfsclnt_hash,
1248 	    nfsclnt_compare, nfsclnt_mkkey,
1249 	    TRUE);
1250 	rfs4_clientid_idx = rfs4_index_create(rfs4_client_tab,
1251 	    "client_id", clientid_hash,
1252 	    clientid_compare, clientid_mkkey,
1253 	    FALSE);
1254 
1255 	rfs4_clntip_cache_time = 86400 * 365;	/* about a year */
1256 	rfs4_clntip_tab = rfs4_table_create(rfs4_server_state,
1257 	    "ClntIP",
1258 	    rfs4_clntip_cache_time,
1259 	    1,
1260 	    rfs4_clntip_create,
1261 	    rfs4_clntip_destroy,
1262 	    rfs4_clntip_expiry,
1263 	    sizeof (rfs4_clntip_t),
1264 	    TABSIZE,
1265 	    MAXTABSZ, 100);
1266 	rfs4_clntip_idx = rfs4_index_create(rfs4_clntip_tab,
1267 	    "client_ip", clntip_hash,
1268 	    clntip_compare, clntip_mkkey,
1269 	    TRUE);
1270 
1271 	rfs4_openowner_cache_time *= rfs4_lease_time;
1272 	rfs4_openowner_tab = rfs4_table_create(rfs4_server_state,
1273 	    "OpenOwner",
1274 	    rfs4_openowner_cache_time,
1275 	    1,
1276 	    rfs4_openowner_create,
1277 	    rfs4_openowner_destroy,
1278 	    rfs4_openowner_expiry,
1279 	    sizeof (rfs4_openowner_t),
1280 	    TABSIZE,
1281 	    MAXTABSZ, 100);
1282 	rfs4_openowner_idx = rfs4_index_create(rfs4_openowner_tab,
1283 	    "open_owner4", openowner_hash,
1284 	    openowner_compare,
1285 	    openowner_mkkey, TRUE);
1286 
1287 	rfs4_state_cache_time *= rfs4_lease_time;
1288 	rfs4_state_tab = rfs4_table_create(rfs4_server_state,
1289 	    "OpenStateID",
1290 	    rfs4_state_cache_time,
1291 	    3,
1292 	    rfs4_state_create,
1293 	    rfs4_state_destroy,
1294 	    rfs4_state_expiry,
1295 	    sizeof (rfs4_state_t),
1296 	    TABSIZE,
1297 	    MAXTABSZ, 100);
1298 
1299 	rfs4_state_owner_file_idx = rfs4_index_create(rfs4_state_tab,
1300 	    "Openowner-File",
1301 	    state_owner_file_hash,
1302 	    state_owner_file_compare,
1303 	    state_owner_file_mkkey, TRUE);
1304 
1305 	rfs4_state_idx = rfs4_index_create(rfs4_state_tab,
1306 	    "State-id", state_hash,
1307 	    state_compare, state_mkkey, FALSE);
1308 
1309 	rfs4_state_file_idx = rfs4_index_create(rfs4_state_tab,
1310 	    "File", state_file_hash,
1311 	    state_file_compare, state_file_mkkey,
1312 	    FALSE);
1313 
1314 	rfs4_lo_state_cache_time *= rfs4_lease_time;
1315 	rfs4_lo_state_tab = rfs4_table_create(rfs4_server_state,
1316 	    "LockStateID",
1317 	    rfs4_lo_state_cache_time,
1318 	    2,
1319 	    rfs4_lo_state_create,
1320 	    rfs4_lo_state_destroy,
1321 	    rfs4_lo_state_expiry,
1322 	    sizeof (rfs4_lo_state_t),
1323 	    TABSIZE,
1324 	    MAXTABSZ, 100);
1325 
1326 	rfs4_lo_state_owner_idx = rfs4_index_create(rfs4_lo_state_tab,
1327 	    "lockownerxstate",
1328 	    lo_state_lo_hash,
1329 	    lo_state_lo_compare,
1330 	    lo_state_lo_mkkey, TRUE);
1331 
1332 	rfs4_lo_state_idx = rfs4_index_create(rfs4_lo_state_tab,
1333 	    "State-id",
1334 	    lo_state_hash, lo_state_compare,
1335 	    lo_state_mkkey, FALSE);
1336 
1337 	rfs4_lockowner_cache_time *= rfs4_lease_time;
1338 
1339 	rfs4_lockowner_tab = rfs4_table_create(rfs4_server_state,
1340 	    "Lockowner",
1341 	    rfs4_lockowner_cache_time,
1342 	    2,
1343 	    rfs4_lockowner_create,
1344 	    rfs4_lockowner_destroy,
1345 	    rfs4_lockowner_expiry,
1346 	    sizeof (rfs4_lockowner_t),
1347 	    TABSIZE,
1348 	    MAXTABSZ, 100);
1349 
1350 	rfs4_lockowner_idx = rfs4_index_create(rfs4_lockowner_tab,
1351 	    "lock_owner4", lockowner_hash,
1352 	    lockowner_compare,
1353 	    lockowner_mkkey, TRUE);
1354 
1355 	rfs4_lockowner_pid_idx = rfs4_index_create(rfs4_lockowner_tab,
1356 	    "pid", pid_hash,
1357 	    pid_compare, pid_mkkey,
1358 	    FALSE);
1359 
1360 	rfs4_file_cache_time *= rfs4_lease_time;
1361 	rfs4_file_tab = rfs4_table_create(rfs4_server_state,
1362 	    "File",
1363 	    rfs4_file_cache_time,
1364 	    1,
1365 	    rfs4_file_create,
1366 	    rfs4_file_destroy,
1367 	    NULL,
1368 	    sizeof (rfs4_file_t),
1369 	    TABSIZE,
1370 	    MAXTABSZ, -1);
1371 
1372 	rfs4_file_idx = rfs4_index_create(rfs4_file_tab,
1373 	    "Filehandle", file_hash,
1374 	    file_compare, file_mkkey, TRUE);
1375 
1376 	rfs4_deleg_state_cache_time *= rfs4_lease_time;
1377 	rfs4_deleg_state_tab = rfs4_table_create(rfs4_server_state,
1378 	    "DelegStateID",
1379 	    rfs4_deleg_state_cache_time,
1380 	    2,
1381 	    rfs4_deleg_state_create,
1382 	    rfs4_deleg_state_destroy,
1383 	    rfs4_deleg_state_expiry,
1384 	    sizeof (rfs4_deleg_state_t),
1385 	    TABSIZE,
1386 	    MAXTABSZ, 100);
1387 	rfs4_deleg_idx = rfs4_index_create(rfs4_deleg_state_tab,
1388 	    "DelegByFileClient",
1389 	    deleg_hash,
1390 	    deleg_compare,
1391 	    deleg_mkkey, TRUE);
1392 
1393 	rfs4_deleg_state_idx = rfs4_index_create(rfs4_deleg_state_tab,
1394 	    "DelegState",
1395 	    deleg_state_hash,
1396 	    deleg_state_compare,
1397 	    deleg_state_mkkey, FALSE);
1398 
1399 	/*
1400 	 * Init the stable storage.
1401 	 */
1402 	rfs4_ss_init();
1403 
1404 	rfs4_client_clrst = rfs4_clear_client_state;
1405 
1406 	mutex_exit(&rfs4_state_lock);
1407 }
1408 
1409 
1410 /*
1411  * Used at server shutdown to cleanup all of the NFSv4 server's structures
1412  * and other state.
1413  */
1414 void
1415 rfs4_state_fini()
1416 {
1417 	rfs4_database_t *dbp;
1418 
1419 	mutex_enter(&rfs4_state_lock);
1420 
1421 	if (rfs4_server_state == NULL) {
1422 		mutex_exit(&rfs4_state_lock);
1423 		return;
1424 	}
1425 
1426 	rfs4_client_clrst = NULL;
1427 
1428 	rfs4_set_deleg_policy(SRV_NEVER_DELEGATE);
1429 	dbp = rfs4_server_state;
1430 	rfs4_server_state = NULL;
1431 
1432 	/*
1433 	 * Cleanup the CPR callback.
1434 	 */
1435 	if (cpr_id)
1436 		(void) callb_delete(cpr_id);
1437 
1438 	rw_destroy(&rfs4_findclient_lock);
1439 
1440 	/* First stop all of the reaper threads in the database */
1441 	rfs4_database_shutdown(dbp);
1442 	/* clean up any dangling stable storage structures */
1443 	rfs4_ss_fini();
1444 	/* Now actually destroy/release the database and its tables */
1445 	rfs4_database_destroy(dbp);
1446 
1447 	/* Reset the cache timers for next time */
1448 	rfs4_client_cache_time = 0;
1449 	rfs4_openowner_cache_time = 0;
1450 	rfs4_state_cache_time = 0;
1451 	rfs4_lo_state_cache_time = 0;
1452 	rfs4_lockowner_cache_time = 0;
1453 	rfs4_file_cache_time = 0;
1454 	rfs4_deleg_state_cache_time = 0;
1455 
1456 	mutex_exit(&rfs4_state_lock);
1457 
1458 	/* destroy server instances and current instance ptr */
1459 	rfs4_servinst_destroy_all();
1460 
1461 	/* reset the "first NFSv4 request" status */
1462 	rfs4_seen_first_compound = 0;
1463 
1464 	/* DSS: distributed stable storage */
1465 	if (rfs4_dss_oldpaths)
1466 		nvlist_free(rfs4_dss_oldpaths);
1467 	if (rfs4_dss_paths)
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 static bool_t
2025 rfs4_openowner_expiry(rfs4_entry_t u_entry)
2026 {
2027 	rfs4_openowner_t *oo = (rfs4_openowner_t *)u_entry;
2028 
2029 	if (rfs4_dbe_is_invalid(oo->ro_dbe))
2030 		return (TRUE);
2031 	return ((gethrestime_sec() - oo->ro_client->rc_last_access
2032 	    > rfs4_lease_time));
2033 }
2034 
2035 static void
2036 rfs4_openowner_destroy(rfs4_entry_t u_entry)
2037 {
2038 	rfs4_openowner_t *oo = (rfs4_openowner_t *)u_entry;
2039 
2040 	/* Remove open owner from client's lists of open owners */
2041 	rfs4_dbe_lock(oo->ro_client->rc_dbe);
2042 	list_remove(&oo->ro_client->rc_openownerlist, oo);
2043 	rfs4_dbe_unlock(oo->ro_client->rc_dbe);
2044 
2045 	/* One less reference to the client */
2046 	rfs4_client_rele(oo->ro_client);
2047 	oo->ro_client = NULL;
2048 
2049 	/* Free the last reply for this lock owner */
2050 	rfs4_free_reply(&oo->ro_reply);
2051 
2052 	if (oo->ro_reply_fh.nfs_fh4_val) {
2053 		kmem_free(oo->ro_reply_fh.nfs_fh4_val,
2054 		    oo->ro_reply_fh.nfs_fh4_len);
2055 		oo->ro_reply_fh.nfs_fh4_val = NULL;
2056 		oo->ro_reply_fh.nfs_fh4_len = 0;
2057 	}
2058 
2059 	rfs4_sw_destroy(&oo->ro_sw);
2060 	list_destroy(&oo->ro_statelist);
2061 
2062 	/* Free the lock owner id */
2063 	kmem_free(oo->ro_owner.owner_val, oo->ro_owner.owner_len);
2064 }
2065 
2066 void
2067 rfs4_openowner_rele(rfs4_openowner_t *oo)
2068 {
2069 	rfs4_dbe_rele(oo->ro_dbe);
2070 }
2071 
2072 static bool_t
2073 rfs4_openowner_create(rfs4_entry_t u_entry, void *arg)
2074 {
2075 	rfs4_openowner_t *oo = (rfs4_openowner_t *)u_entry;
2076 	rfs4_openowner_t *argp = (rfs4_openowner_t *)arg;
2077 	open_owner4 *openowner = &argp->ro_owner;
2078 	seqid4 seqid = argp->ro_open_seqid;
2079 	rfs4_client_t *cp;
2080 	bool_t create = FALSE;
2081 
2082 	rw_enter(&rfs4_findclient_lock, RW_READER);
2083 
2084 	cp = (rfs4_client_t *)rfs4_dbsearch(rfs4_clientid_idx,
2085 	    &openowner->clientid,
2086 	    &create, NULL, RFS4_DBS_VALID);
2087 
2088 	rw_exit(&rfs4_findclient_lock);
2089 
2090 	if (cp == NULL)
2091 		return (FALSE);
2092 
2093 	oo->ro_reply_fh.nfs_fh4_len = 0;
2094 	oo->ro_reply_fh.nfs_fh4_val = NULL;
2095 
2096 	oo->ro_owner.clientid = openowner->clientid;
2097 	oo->ro_owner.owner_val =
2098 	    kmem_alloc(openowner->owner_len, KM_SLEEP);
2099 
2100 	bcopy(openowner->owner_val,
2101 	    oo->ro_owner.owner_val, openowner->owner_len);
2102 
2103 	oo->ro_owner.owner_len = openowner->owner_len;
2104 
2105 	oo->ro_need_confirm = TRUE;
2106 
2107 	rfs4_sw_init(&oo->ro_sw);
2108 
2109 	oo->ro_open_seqid = seqid;
2110 	bzero(&oo->ro_reply, sizeof (nfs_resop4));
2111 	oo->ro_client = cp;
2112 	oo->ro_cr_set = NULL;
2113 
2114 	list_create(&oo->ro_statelist, sizeof (rfs4_state_t),
2115 	    offsetof(rfs4_state_t, rs_node));
2116 
2117 	/* Insert openowner into client's open owner list */
2118 	rfs4_dbe_lock(cp->rc_dbe);
2119 	list_insert_tail(&cp->rc_openownerlist, oo);
2120 	rfs4_dbe_unlock(cp->rc_dbe);
2121 
2122 	return (TRUE);
2123 }
2124 
2125 rfs4_openowner_t *
2126 rfs4_findopenowner(open_owner4 *openowner, bool_t *create, seqid4 seqid)
2127 {
2128 	rfs4_openowner_t *oo;
2129 	rfs4_openowner_t arg;
2130 
2131 	arg.ro_owner = *openowner;
2132 	arg.ro_open_seqid = seqid;
2133 	oo = (rfs4_openowner_t *)rfs4_dbsearch(rfs4_openowner_idx, openowner,
2134 	    create, &arg, RFS4_DBS_VALID);
2135 
2136 	return (oo);
2137 }
2138 
2139 void
2140 rfs4_update_open_sequence(rfs4_openowner_t *oo)
2141 {
2142 
2143 	rfs4_dbe_lock(oo->ro_dbe);
2144 
2145 	oo->ro_open_seqid++;
2146 
2147 	rfs4_dbe_unlock(oo->ro_dbe);
2148 }
2149 
2150 void
2151 rfs4_update_open_resp(rfs4_openowner_t *oo, nfs_resop4 *resp, nfs_fh4 *fh)
2152 {
2153 
2154 	rfs4_dbe_lock(oo->ro_dbe);
2155 
2156 	rfs4_free_reply(&oo->ro_reply);
2157 
2158 	rfs4_copy_reply(&oo->ro_reply, resp);
2159 
2160 	/* Save the filehandle if provided and free if not used */
2161 	if (resp->nfs_resop4_u.opopen.status == NFS4_OK &&
2162 	    fh && fh->nfs_fh4_len) {
2163 		if (oo->ro_reply_fh.nfs_fh4_val == NULL)
2164 			oo->ro_reply_fh.nfs_fh4_val =
2165 			    kmem_alloc(fh->nfs_fh4_len, KM_SLEEP);
2166 		nfs_fh4_copy(fh, &oo->ro_reply_fh);
2167 	} else {
2168 		if (oo->ro_reply_fh.nfs_fh4_val) {
2169 			kmem_free(oo->ro_reply_fh.nfs_fh4_val,
2170 			    oo->ro_reply_fh.nfs_fh4_len);
2171 			oo->ro_reply_fh.nfs_fh4_val = NULL;
2172 			oo->ro_reply_fh.nfs_fh4_len = 0;
2173 		}
2174 	}
2175 
2176 	rfs4_dbe_unlock(oo->ro_dbe);
2177 }
2178 
2179 static bool_t
2180 lockowner_compare(rfs4_entry_t u_entry, void *key)
2181 {
2182 	rfs4_lockowner_t *lo = (rfs4_lockowner_t *)u_entry;
2183 	lock_owner4 *b = (lock_owner4 *)key;
2184 
2185 	if (lo->rl_owner.clientid != b->clientid)
2186 		return (FALSE);
2187 
2188 	if (lo->rl_owner.owner_len != b->owner_len)
2189 		return (FALSE);
2190 
2191 	return (bcmp(lo->rl_owner.owner_val, b->owner_val,
2192 	    lo->rl_owner.owner_len) == 0);
2193 }
2194 
2195 void *
2196 lockowner_mkkey(rfs4_entry_t u_entry)
2197 {
2198 	rfs4_lockowner_t *lo = (rfs4_lockowner_t *)u_entry;
2199 
2200 	return (&lo->rl_owner);
2201 }
2202 
2203 static uint32_t
2204 lockowner_hash(void *key)
2205 {
2206 	int i;
2207 	lock_owner4 *lockowner = key;
2208 	uint_t hash = 0;
2209 
2210 	for (i = 0; i < lockowner->owner_len; i++) {
2211 		hash <<= 4;
2212 		hash += (uint_t)lockowner->owner_val[i];
2213 	}
2214 	hash += (uint_t)lockowner->clientid;
2215 	hash |= (lockowner->clientid >> 32);
2216 
2217 	return (hash);
2218 }
2219 
2220 static uint32_t
2221 pid_hash(void *key)
2222 {
2223 	return ((uint32_t)(uintptr_t)key);
2224 }
2225 
2226 static void *
2227 pid_mkkey(rfs4_entry_t u_entry)
2228 {
2229 	rfs4_lockowner_t *lo = (rfs4_lockowner_t *)u_entry;
2230 
2231 	return ((void *)(uintptr_t)lo->rl_pid);
2232 }
2233 
2234 static bool_t
2235 pid_compare(rfs4_entry_t u_entry, void *key)
2236 {
2237 	rfs4_lockowner_t *lo = (rfs4_lockowner_t *)u_entry;
2238 
2239 	return (lo->rl_pid == (pid_t)(uintptr_t)key);
2240 }
2241 
2242 static void
2243 rfs4_lockowner_destroy(rfs4_entry_t u_entry)
2244 {
2245 	rfs4_lockowner_t *lo = (rfs4_lockowner_t *)u_entry;
2246 
2247 	/* Free the lock owner id */
2248 	kmem_free(lo->rl_owner.owner_val, lo->rl_owner.owner_len);
2249 	rfs4_client_rele(lo->rl_client);
2250 }
2251 
2252 void
2253 rfs4_lockowner_rele(rfs4_lockowner_t *lo)
2254 {
2255 	rfs4_dbe_rele(lo->rl_dbe);
2256 }
2257 
2258 /* ARGSUSED */
2259 static bool_t
2260 rfs4_lockowner_expiry(rfs4_entry_t u_entry)
2261 {
2262 	/*
2263 	 * Since expiry is called with no other references on
2264 	 * this struct, go ahead and have it removed.
2265 	 */
2266 	return (TRUE);
2267 }
2268 
2269 static bool_t
2270 rfs4_lockowner_create(rfs4_entry_t u_entry, void *arg)
2271 {
2272 	rfs4_lockowner_t *lo = (rfs4_lockowner_t *)u_entry;
2273 	lock_owner4 *lockowner = (lock_owner4 *)arg;
2274 	rfs4_client_t *cp;
2275 	bool_t create = FALSE;
2276 
2277 	rw_enter(&rfs4_findclient_lock, RW_READER);
2278 
2279 	cp = (rfs4_client_t *)rfs4_dbsearch(rfs4_clientid_idx,
2280 	    &lockowner->clientid,
2281 	    &create, NULL, RFS4_DBS_VALID);
2282 
2283 	rw_exit(&rfs4_findclient_lock);
2284 
2285 	if (cp == NULL)
2286 		return (FALSE);
2287 
2288 	/* Reference client */
2289 	lo->rl_client = cp;
2290 	lo->rl_owner.clientid = lockowner->clientid;
2291 	lo->rl_owner.owner_val = kmem_alloc(lockowner->owner_len, KM_SLEEP);
2292 	bcopy(lockowner->owner_val, lo->rl_owner.owner_val,
2293 	    lockowner->owner_len);
2294 	lo->rl_owner.owner_len = lockowner->owner_len;
2295 	lo->rl_pid = rfs4_dbe_getid(lo->rl_dbe);
2296 
2297 	return (TRUE);
2298 }
2299 
2300 rfs4_lockowner_t *
2301 rfs4_findlockowner(lock_owner4 *lockowner, bool_t *create)
2302 {
2303 	rfs4_lockowner_t *lo;
2304 
2305 	lo = (rfs4_lockowner_t *)rfs4_dbsearch(rfs4_lockowner_idx, lockowner,
2306 	    create, lockowner, RFS4_DBS_VALID);
2307 
2308 	return (lo);
2309 }
2310 
2311 rfs4_lockowner_t *
2312 rfs4_findlockowner_by_pid(pid_t pid)
2313 {
2314 	rfs4_lockowner_t *lo;
2315 	bool_t create = FALSE;
2316 
2317 	lo = (rfs4_lockowner_t *)rfs4_dbsearch(rfs4_lockowner_pid_idx,
2318 	    (void *)(uintptr_t)pid, &create, NULL, RFS4_DBS_VALID);
2319 
2320 	return (lo);
2321 }
2322 
2323 
2324 static uint32_t
2325 file_hash(void *key)
2326 {
2327 	return (ADDRHASH(key));
2328 }
2329 
2330 static void *
2331 file_mkkey(rfs4_entry_t u_entry)
2332 {
2333 	rfs4_file_t *fp = (rfs4_file_t *)u_entry;
2334 
2335 	return (fp->rf_vp);
2336 }
2337 
2338 static bool_t
2339 file_compare(rfs4_entry_t u_entry, void *key)
2340 {
2341 	rfs4_file_t *fp = (rfs4_file_t *)u_entry;
2342 
2343 	return (fp->rf_vp == (vnode_t *)key);
2344 }
2345 
2346 static void
2347 rfs4_file_destroy(rfs4_entry_t u_entry)
2348 {
2349 	rfs4_file_t *fp = (rfs4_file_t *)u_entry;
2350 
2351 	list_destroy(&fp->rf_delegstatelist);
2352 
2353 	if (fp->rf_filehandle.nfs_fh4_val)
2354 		kmem_free(fp->rf_filehandle.nfs_fh4_val,
2355 		    fp->rf_filehandle.nfs_fh4_len);
2356 	cv_destroy(fp->rf_dinfo.rd_recall_cv);
2357 	if (fp->rf_vp) {
2358 		vnode_t *vp = fp->rf_vp;
2359 
2360 		mutex_enter(&vp->v_vsd_lock);
2361 		(void) vsd_set(vp, nfs4_srv_vkey, NULL);
2362 		mutex_exit(&vp->v_vsd_lock);
2363 		VN_RELE(vp);
2364 		fp->rf_vp = NULL;
2365 	}
2366 	rw_destroy(&fp->rf_file_rwlock);
2367 }
2368 
2369 /*
2370  * Used to unlock the underlying dbe struct only
2371  */
2372 void
2373 rfs4_file_rele(rfs4_file_t *fp)
2374 {
2375 	rfs4_dbe_rele(fp->rf_dbe);
2376 }
2377 
2378 typedef struct {
2379     vnode_t *vp;
2380     nfs_fh4 *fh;
2381 } rfs4_fcreate_arg;
2382 
2383 static bool_t
2384 rfs4_file_create(rfs4_entry_t u_entry, void *arg)
2385 {
2386 	rfs4_file_t *fp = (rfs4_file_t *)u_entry;
2387 	rfs4_fcreate_arg *ap = (rfs4_fcreate_arg *)arg;
2388 	vnode_t *vp = ap->vp;
2389 	nfs_fh4 *fh = ap->fh;
2390 
2391 	VN_HOLD(vp);
2392 
2393 	fp->rf_filehandle.nfs_fh4_len = 0;
2394 	fp->rf_filehandle.nfs_fh4_val = NULL;
2395 	ASSERT(fh && fh->nfs_fh4_len);
2396 	if (fh && fh->nfs_fh4_len) {
2397 		fp->rf_filehandle.nfs_fh4_val =
2398 		    kmem_alloc(fh->nfs_fh4_len, KM_SLEEP);
2399 		nfs_fh4_copy(fh, &fp->rf_filehandle);
2400 	}
2401 	fp->rf_vp = vp;
2402 
2403 	list_create(&fp->rf_delegstatelist, sizeof (rfs4_deleg_state_t),
2404 	    offsetof(rfs4_deleg_state_t, rds_node));
2405 
2406 	fp->rf_share_deny = fp->rf_share_access = fp->rf_access_read = 0;
2407 	fp->rf_access_write = fp->rf_deny_read = fp->rf_deny_write = 0;
2408 
2409 	mutex_init(fp->rf_dinfo.rd_recall_lock, NULL, MUTEX_DEFAULT, NULL);
2410 	cv_init(fp->rf_dinfo.rd_recall_cv, NULL, CV_DEFAULT, NULL);
2411 
2412 	fp->rf_dinfo.rd_dtype = OPEN_DELEGATE_NONE;
2413 
2414 	rw_init(&fp->rf_file_rwlock, NULL, RW_DEFAULT, NULL);
2415 
2416 	mutex_enter(&vp->v_vsd_lock);
2417 	VERIFY(vsd_set(vp, nfs4_srv_vkey, (void *)fp) == 0);
2418 	mutex_exit(&vp->v_vsd_lock);
2419 
2420 	return (TRUE);
2421 }
2422 
2423 rfs4_file_t *
2424 rfs4_findfile(vnode_t *vp, nfs_fh4 *fh, bool_t *create)
2425 {
2426 	rfs4_file_t *fp;
2427 	rfs4_fcreate_arg arg;
2428 
2429 	arg.vp = vp;
2430 	arg.fh = fh;
2431 
2432 	if (*create == TRUE)
2433 		fp = (rfs4_file_t *)rfs4_dbsearch(rfs4_file_idx, vp, create,
2434 		    &arg, RFS4_DBS_VALID);
2435 	else {
2436 		mutex_enter(&vp->v_vsd_lock);
2437 		fp = (rfs4_file_t *)vsd_get(vp, nfs4_srv_vkey);
2438 		if (fp) {
2439 			rfs4_dbe_lock(fp->rf_dbe);
2440 			if (rfs4_dbe_is_invalid(fp->rf_dbe) ||
2441 			    (rfs4_dbe_refcnt(fp->rf_dbe) == 0)) {
2442 				rfs4_dbe_unlock(fp->rf_dbe);
2443 				fp = NULL;
2444 			} else {
2445 				rfs4_dbe_hold(fp->rf_dbe);
2446 				rfs4_dbe_unlock(fp->rf_dbe);
2447 			}
2448 		}
2449 		mutex_exit(&vp->v_vsd_lock);
2450 	}
2451 	return (fp);
2452 }
2453 
2454 /*
2455  * Find a file in the db and once it is located, take the rw lock.
2456  * Need to check the vnode pointer and if it does not exist (it was
2457  * removed between the db location and check) redo the find.  This
2458  * assumes that a file struct that has a NULL vnode pointer is marked
2459  * at 'invalid' and will not be found in the db the second time
2460  * around.
2461  */
2462 rfs4_file_t *
2463 rfs4_findfile_withlock(vnode_t *vp, nfs_fh4 *fh, bool_t *create)
2464 {
2465 	rfs4_file_t *fp;
2466 	rfs4_fcreate_arg arg;
2467 	bool_t screate = *create;
2468 
2469 	if (screate == FALSE) {
2470 		mutex_enter(&vp->v_vsd_lock);
2471 		fp = (rfs4_file_t *)vsd_get(vp, nfs4_srv_vkey);
2472 		if (fp) {
2473 			rfs4_dbe_lock(fp->rf_dbe);
2474 			if (rfs4_dbe_is_invalid(fp->rf_dbe) ||
2475 			    (rfs4_dbe_refcnt(fp->rf_dbe) == 0)) {
2476 				rfs4_dbe_unlock(fp->rf_dbe);
2477 				mutex_exit(&vp->v_vsd_lock);
2478 				fp = NULL;
2479 			} else {
2480 				rfs4_dbe_hold(fp->rf_dbe);
2481 				rfs4_dbe_unlock(fp->rf_dbe);
2482 				mutex_exit(&vp->v_vsd_lock);
2483 				rw_enter(&fp->rf_file_rwlock, RW_WRITER);
2484 				if (fp->rf_vp == NULL) {
2485 					rw_exit(&fp->rf_file_rwlock);
2486 					rfs4_file_rele(fp);
2487 					fp = NULL;
2488 				}
2489 			}
2490 		} else {
2491 			mutex_exit(&vp->v_vsd_lock);
2492 		}
2493 	} else {
2494 retry:
2495 		arg.vp = vp;
2496 		arg.fh = fh;
2497 
2498 		fp = (rfs4_file_t *)rfs4_dbsearch(rfs4_file_idx, vp, create,
2499 		    &arg, RFS4_DBS_VALID);
2500 		if (fp != NULL) {
2501 			rw_enter(&fp->rf_file_rwlock, RW_WRITER);
2502 			if (fp->rf_vp == NULL) {
2503 				rw_exit(&fp->rf_file_rwlock);
2504 				rfs4_file_rele(fp);
2505 				*create = screate;
2506 				goto retry;
2507 			}
2508 		}
2509 	}
2510 
2511 	return (fp);
2512 }
2513 
2514 static uint32_t
2515 lo_state_hash(void *key)
2516 {
2517 	stateid_t *id = key;
2518 
2519 	return (id->bits.ident+id->bits.pid);
2520 }
2521 
2522 static bool_t
2523 lo_state_compare(rfs4_entry_t u_entry, void *key)
2524 {
2525 	rfs4_lo_state_t *lsp = (rfs4_lo_state_t *)u_entry;
2526 	stateid_t *id = key;
2527 	bool_t rc;
2528 
2529 	rc = (lsp->rls_lockid.bits.boottime == id->bits.boottime &&
2530 	    lsp->rls_lockid.bits.type == id->bits.type &&
2531 	    lsp->rls_lockid.bits.ident == id->bits.ident &&
2532 	    lsp->rls_lockid.bits.pid == id->bits.pid);
2533 
2534 	return (rc);
2535 }
2536 
2537 static void *
2538 lo_state_mkkey(rfs4_entry_t u_entry)
2539 {
2540 	rfs4_lo_state_t *lsp = (rfs4_lo_state_t *)u_entry;
2541 
2542 	return (&lsp->rls_lockid);
2543 }
2544 
2545 static bool_t
2546 rfs4_lo_state_expiry(rfs4_entry_t u_entry)
2547 {
2548 	rfs4_lo_state_t *lsp = (rfs4_lo_state_t *)u_entry;
2549 
2550 	if (rfs4_dbe_is_invalid(lsp->rls_dbe))
2551 		return (TRUE);
2552 	if (lsp->rls_state->rs_closed)
2553 		return (TRUE);
2554 	return ((gethrestime_sec() -
2555 	    lsp->rls_state->rs_owner->ro_client->rc_last_access
2556 	    > rfs4_lease_time));
2557 }
2558 
2559 static void
2560 rfs4_lo_state_destroy(rfs4_entry_t u_entry)
2561 {
2562 	rfs4_lo_state_t *lsp = (rfs4_lo_state_t *)u_entry;
2563 
2564 	rfs4_dbe_lock(lsp->rls_state->rs_dbe);
2565 	list_remove(&lsp->rls_state->rs_lostatelist, lsp);
2566 	rfs4_dbe_unlock(lsp->rls_state->rs_dbe);
2567 
2568 	rfs4_sw_destroy(&lsp->rls_sw);
2569 
2570 	/* Make sure to release the file locks */
2571 	if (lsp->rls_locks_cleaned == FALSE) {
2572 		lsp->rls_locks_cleaned = TRUE;
2573 		if (lsp->rls_locker->rl_client->rc_sysidt != LM_NOSYSID) {
2574 			/* Is the PxFS kernel module loaded? */
2575 			if (lm_remove_file_locks != NULL) {
2576 				int new_sysid;
2577 
2578 				/* Encode the cluster nodeid in new sysid */
2579 				new_sysid =
2580 				    lsp->rls_locker->rl_client->rc_sysidt;
2581 				lm_set_nlmid_flk(&new_sysid);
2582 
2583 				/*
2584 				 * This PxFS routine removes file locks for a
2585 				 * client over all nodes of a cluster.
2586 				 */
2587 				DTRACE_PROBE1(nfss_i_clust_rm_lck,
2588 				    int, new_sysid);
2589 				(*lm_remove_file_locks)(new_sysid);
2590 			} else {
2591 				(void) cleanlocks(
2592 				    lsp->rls_state->rs_finfo->rf_vp,
2593 				    lsp->rls_locker->rl_pid,
2594 				    lsp->rls_locker->rl_client->rc_sysidt);
2595 			}
2596 		}
2597 	}
2598 
2599 	/* Free the last reply for this state */
2600 	rfs4_free_reply(&lsp->rls_reply);
2601 
2602 	rfs4_lockowner_rele(lsp->rls_locker);
2603 	lsp->rls_locker = NULL;
2604 
2605 	rfs4_state_rele_nounlock(lsp->rls_state);
2606 	lsp->rls_state = NULL;
2607 }
2608 
2609 static bool_t
2610 rfs4_lo_state_create(rfs4_entry_t u_entry, void *arg)
2611 {
2612 	rfs4_lo_state_t *lsp = (rfs4_lo_state_t *)u_entry;
2613 	rfs4_lo_state_t *argp = (rfs4_lo_state_t *)arg;
2614 	rfs4_lockowner_t *lo = argp->rls_locker;
2615 	rfs4_state_t *sp = argp->rls_state;
2616 
2617 	lsp->rls_state = sp;
2618 
2619 	lsp->rls_lockid = sp->rs_stateid;
2620 	lsp->rls_lockid.bits.type = LOCKID;
2621 	lsp->rls_lockid.bits.chgseq = 0;
2622 	lsp->rls_lockid.bits.pid = lo->rl_pid;
2623 
2624 	lsp->rls_locks_cleaned = FALSE;
2625 	lsp->rls_lock_completed = FALSE;
2626 
2627 	rfs4_sw_init(&lsp->rls_sw);
2628 
2629 	/* Attached the supplied lock owner */
2630 	rfs4_dbe_hold(lo->rl_dbe);
2631 	lsp->rls_locker = lo;
2632 
2633 	rfs4_dbe_lock(sp->rs_dbe);
2634 	list_insert_tail(&sp->rs_lostatelist, lsp);
2635 	rfs4_dbe_hold(sp->rs_dbe);
2636 	rfs4_dbe_unlock(sp->rs_dbe);
2637 
2638 	return (TRUE);
2639 }
2640 
2641 void
2642 rfs4_lo_state_rele(rfs4_lo_state_t *lsp, bool_t unlock_fp)
2643 {
2644 	if (unlock_fp == TRUE)
2645 		rw_exit(&lsp->rls_state->rs_finfo->rf_file_rwlock);
2646 	rfs4_dbe_rele(lsp->rls_dbe);
2647 }
2648 
2649 static rfs4_lo_state_t *
2650 rfs4_findlo_state(stateid_t *id, bool_t lock_fp)
2651 {
2652 	rfs4_lo_state_t *lsp;
2653 	bool_t create = FALSE;
2654 
2655 	lsp = (rfs4_lo_state_t *)rfs4_dbsearch(rfs4_lo_state_idx, id,
2656 	    &create, NULL, RFS4_DBS_VALID);
2657 	if (lock_fp == TRUE && lsp != NULL)
2658 		rw_enter(&lsp->rls_state->rs_finfo->rf_file_rwlock, RW_READER);
2659 
2660 	return (lsp);
2661 }
2662 
2663 
2664 static uint32_t
2665 lo_state_lo_hash(void *key)
2666 {
2667 	rfs4_lo_state_t *lsp = key;
2668 
2669 	return (ADDRHASH(lsp->rls_locker) ^ ADDRHASH(lsp->rls_state));
2670 }
2671 
2672 static bool_t
2673 lo_state_lo_compare(rfs4_entry_t u_entry, void *key)
2674 {
2675 	rfs4_lo_state_t *lsp = (rfs4_lo_state_t *)u_entry;
2676 	rfs4_lo_state_t *keyp = key;
2677 
2678 	return (keyp->rls_locker == lsp->rls_locker &&
2679 	    keyp->rls_state == lsp->rls_state);
2680 }
2681 
2682 static void *
2683 lo_state_lo_mkkey(rfs4_entry_t u_entry)
2684 {
2685 	return (u_entry);
2686 }
2687 
2688 rfs4_lo_state_t *
2689 rfs4_findlo_state_by_owner(rfs4_lockowner_t *lo, rfs4_state_t *sp,
2690     bool_t *create)
2691 {
2692 	rfs4_lo_state_t *lsp;
2693 	rfs4_lo_state_t arg;
2694 
2695 	arg.rls_locker = lo;
2696 	arg.rls_state = sp;
2697 
2698 	lsp = (rfs4_lo_state_t *)rfs4_dbsearch(rfs4_lo_state_owner_idx, &arg,
2699 	    create, &arg, RFS4_DBS_VALID);
2700 
2701 	return (lsp);
2702 }
2703 
2704 static stateid_t
2705 get_stateid(id_t eid)
2706 {
2707 	stateid_t id;
2708 
2709 	id.bits.boottime = rfs4_start_time;
2710 	id.bits.ident = eid;
2711 	id.bits.chgseq = 0;
2712 	id.bits.type = 0;
2713 	id.bits.pid = 0;
2714 
2715 	/*
2716 	 * If we are booted as a cluster node, embed our nodeid.
2717 	 * We've already done sanity checks in rfs4_client_create() so no
2718 	 * need to repeat them here.
2719 	 */
2720 	id.bits.clnodeid = (cluster_bootflags & CLUSTER_BOOTED) ?
2721 	    clconf_get_nodeid() : 0;
2722 
2723 	return (id);
2724 }
2725 
2726 /*
2727  * For use only when booted as a cluster node.
2728  * Returns TRUE if the embedded nodeid indicates that this stateid was
2729  * generated on another node.
2730  */
2731 static int
2732 foreign_stateid(stateid_t *id)
2733 {
2734 	ASSERT(cluster_bootflags & CLUSTER_BOOTED);
2735 	return (id->bits.clnodeid != (uint32_t)clconf_get_nodeid());
2736 }
2737 
2738 /*
2739  * For use only when booted as a cluster node.
2740  * Returns TRUE if the embedded nodeid indicates that this clientid was
2741  * generated on another node.
2742  */
2743 static int
2744 foreign_clientid(cid *cidp)
2745 {
2746 	ASSERT(cluster_bootflags & CLUSTER_BOOTED);
2747 	return (cidp->impl_id.c_id >> CLUSTER_NODEID_SHIFT !=
2748 	    (uint32_t)clconf_get_nodeid());
2749 }
2750 
2751 /*
2752  * For use only when booted as a cluster node.
2753  * Embed our cluster nodeid into the clientid.
2754  */
2755 static void
2756 embed_nodeid(cid *cidp)
2757 {
2758 	int clnodeid;
2759 	/*
2760 	 * Currently, our state tables are small enough that their
2761 	 * ids will leave enough bits free for the nodeid. If the
2762 	 * tables become larger, we mustn't overwrite the id.
2763 	 * Equally, we only have room for so many bits of nodeid, so
2764 	 * must check that too.
2765 	 */
2766 	ASSERT(cluster_bootflags & CLUSTER_BOOTED);
2767 	ASSERT(cidp->impl_id.c_id >> CLUSTER_NODEID_SHIFT == 0);
2768 	clnodeid = clconf_get_nodeid();
2769 	ASSERT(clnodeid <= CLUSTER_MAX_NODEID);
2770 	ASSERT(clnodeid != NODEID_UNKNOWN);
2771 	cidp->impl_id.c_id |= (clnodeid << CLUSTER_NODEID_SHIFT);
2772 }
2773 
2774 static uint32_t
2775 state_hash(void *key)
2776 {
2777 	stateid_t *ip = (stateid_t *)key;
2778 
2779 	return (ip->bits.ident);
2780 }
2781 
2782 static bool_t
2783 state_compare(rfs4_entry_t u_entry, void *key)
2784 {
2785 	rfs4_state_t *sp = (rfs4_state_t *)u_entry;
2786 	stateid_t *id = (stateid_t *)key;
2787 	bool_t rc;
2788 
2789 	rc = (sp->rs_stateid.bits.boottime == id->bits.boottime &&
2790 	    sp->rs_stateid.bits.ident == id->bits.ident);
2791 
2792 	return (rc);
2793 }
2794 
2795 static void *
2796 state_mkkey(rfs4_entry_t u_entry)
2797 {
2798 	rfs4_state_t *sp = (rfs4_state_t *)u_entry;
2799 
2800 	return (&sp->rs_stateid);
2801 }
2802 
2803 static void
2804 rfs4_state_destroy(rfs4_entry_t u_entry)
2805 {
2806 	rfs4_state_t *sp = (rfs4_state_t *)u_entry;
2807 
2808 	/* remove from openowner list */
2809 	rfs4_dbe_lock(sp->rs_owner->ro_dbe);
2810 	list_remove(&sp->rs_owner->ro_statelist, sp);
2811 	rfs4_dbe_unlock(sp->rs_owner->ro_dbe);
2812 
2813 	list_destroy(&sp->rs_lostatelist);
2814 
2815 	/* release any share locks for this stateid if it's still open */
2816 	if (!sp->rs_closed) {
2817 		rfs4_dbe_lock(sp->rs_dbe);
2818 		(void) rfs4_unshare(sp);
2819 		rfs4_dbe_unlock(sp->rs_dbe);
2820 	}
2821 
2822 	/* Were done with the file */
2823 	rfs4_file_rele(sp->rs_finfo);
2824 	sp->rs_finfo = NULL;
2825 
2826 	/* And now with the openowner */
2827 	rfs4_openowner_rele(sp->rs_owner);
2828 	sp->rs_owner = NULL;
2829 }
2830 
2831 static void
2832 rfs4_state_rele_nounlock(rfs4_state_t *sp)
2833 {
2834 	rfs4_dbe_rele(sp->rs_dbe);
2835 }
2836 
2837 void
2838 rfs4_state_rele(rfs4_state_t *sp)
2839 {
2840 	rw_exit(&sp->rs_finfo->rf_file_rwlock);
2841 	rfs4_dbe_rele(sp->rs_dbe);
2842 }
2843 
2844 static uint32_t
2845 deleg_hash(void *key)
2846 {
2847 	rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)key;
2848 
2849 	return (ADDRHASH(dsp->rds_client) ^ ADDRHASH(dsp->rds_finfo));
2850 }
2851 
2852 static bool_t
2853 deleg_compare(rfs4_entry_t u_entry, void *key)
2854 {
2855 	rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
2856 	rfs4_deleg_state_t *kdsp = (rfs4_deleg_state_t *)key;
2857 
2858 	return (dsp->rds_client == kdsp->rds_client &&
2859 	    dsp->rds_finfo == kdsp->rds_finfo);
2860 }
2861 
2862 static void *
2863 deleg_mkkey(rfs4_entry_t u_entry)
2864 {
2865 	return (u_entry);
2866 }
2867 
2868 static uint32_t
2869 deleg_state_hash(void *key)
2870 {
2871 	stateid_t *ip = (stateid_t *)key;
2872 
2873 	return (ip->bits.ident);
2874 }
2875 
2876 static bool_t
2877 deleg_state_compare(rfs4_entry_t u_entry, void *key)
2878 {
2879 	rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
2880 	stateid_t *id = (stateid_t *)key;
2881 	bool_t rc;
2882 
2883 	if (id->bits.type != DELEGID)
2884 		return (FALSE);
2885 
2886 	rc = (dsp->rds_delegid.bits.boottime == id->bits.boottime &&
2887 	    dsp->rds_delegid.bits.ident == id->bits.ident);
2888 
2889 	return (rc);
2890 }
2891 
2892 static void *
2893 deleg_state_mkkey(rfs4_entry_t u_entry)
2894 {
2895 	rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
2896 
2897 	return (&dsp->rds_delegid);
2898 }
2899 
2900 static bool_t
2901 rfs4_deleg_state_expiry(rfs4_entry_t u_entry)
2902 {
2903 	rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
2904 
2905 	if (rfs4_dbe_is_invalid(dsp->rds_dbe))
2906 		return (TRUE);
2907 
2908 	if (dsp->rds_dtype == OPEN_DELEGATE_NONE)
2909 		return (TRUE);
2910 
2911 	if ((gethrestime_sec() - dsp->rds_client->rc_last_access
2912 	    > rfs4_lease_time)) {
2913 		rfs4_dbe_invalidate(dsp->rds_dbe);
2914 		return (TRUE);
2915 	}
2916 
2917 	return (FALSE);
2918 }
2919 
2920 static bool_t
2921 rfs4_deleg_state_create(rfs4_entry_t u_entry, void *argp)
2922 {
2923 	rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
2924 	rfs4_file_t *fp = ((rfs4_deleg_state_t *)argp)->rds_finfo;
2925 	rfs4_client_t *cp = ((rfs4_deleg_state_t *)argp)->rds_client;
2926 
2927 	rfs4_dbe_hold(fp->rf_dbe);
2928 	rfs4_dbe_hold(cp->rc_dbe);
2929 
2930 	dsp->rds_delegid = get_stateid(rfs4_dbe_getid(dsp->rds_dbe));
2931 	dsp->rds_delegid.bits.type = DELEGID;
2932 	dsp->rds_finfo = fp;
2933 	dsp->rds_client = cp;
2934 	dsp->rds_dtype = OPEN_DELEGATE_NONE;
2935 
2936 	dsp->rds_time_granted = gethrestime_sec();	/* observability */
2937 	dsp->rds_time_revoked = 0;
2938 
2939 	list_link_init(&dsp->rds_node);
2940 
2941 	return (TRUE);
2942 }
2943 
2944 static void
2945 rfs4_deleg_state_destroy(rfs4_entry_t u_entry)
2946 {
2947 	rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
2948 
2949 	/* return delegation if necessary */
2950 	rfs4_return_deleg(dsp, FALSE);
2951 
2952 	/* Were done with the file */
2953 	rfs4_file_rele(dsp->rds_finfo);
2954 	dsp->rds_finfo = NULL;
2955 
2956 	/* And now with the openowner */
2957 	rfs4_client_rele(dsp->rds_client);
2958 	dsp->rds_client = NULL;
2959 }
2960 
2961 rfs4_deleg_state_t *
2962 rfs4_finddeleg(rfs4_state_t *sp, bool_t *create)
2963 {
2964 	rfs4_deleg_state_t ds, *dsp;
2965 
2966 	ds.rds_client = sp->rs_owner->ro_client;
2967 	ds.rds_finfo = sp->rs_finfo;
2968 
2969 	dsp = (rfs4_deleg_state_t *)rfs4_dbsearch(rfs4_deleg_idx, &ds,
2970 	    create, &ds, RFS4_DBS_VALID);
2971 
2972 	return (dsp);
2973 }
2974 
2975 rfs4_deleg_state_t *
2976 rfs4_finddelegstate(stateid_t *id)
2977 {
2978 	rfs4_deleg_state_t *dsp;
2979 	bool_t create = FALSE;
2980 
2981 	dsp = (rfs4_deleg_state_t *)rfs4_dbsearch(rfs4_deleg_state_idx, id,
2982 	    &create, NULL, RFS4_DBS_VALID);
2983 
2984 	return (dsp);
2985 }
2986 
2987 void
2988 rfs4_deleg_state_rele(rfs4_deleg_state_t *dsp)
2989 {
2990 	rfs4_dbe_rele(dsp->rds_dbe);
2991 }
2992 
2993 void
2994 rfs4_update_lock_sequence(rfs4_lo_state_t *lsp)
2995 {
2996 
2997 	rfs4_dbe_lock(lsp->rls_dbe);
2998 
2999 	/*
3000 	 * If we are skipping sequence id checking, this means that
3001 	 * this is the first lock request and therefore the sequence
3002 	 * id does not need to be updated.  This only happens on the
3003 	 * first lock request for a lockowner
3004 	 */
3005 	if (!lsp->rls_skip_seqid_check)
3006 		lsp->rls_seqid++;
3007 
3008 	rfs4_dbe_unlock(lsp->rls_dbe);
3009 }
3010 
3011 void
3012 rfs4_update_lock_resp(rfs4_lo_state_t *lsp, nfs_resop4 *resp)
3013 {
3014 
3015 	rfs4_dbe_lock(lsp->rls_dbe);
3016 
3017 	rfs4_free_reply(&lsp->rls_reply);
3018 
3019 	rfs4_copy_reply(&lsp->rls_reply, resp);
3020 
3021 	rfs4_dbe_unlock(lsp->rls_dbe);
3022 }
3023 
3024 void
3025 rfs4_free_opens(rfs4_openowner_t *oo, bool_t invalidate,
3026     bool_t close_of_client)
3027 {
3028 	rfs4_state_t *sp;
3029 
3030 	rfs4_dbe_lock(oo->ro_dbe);
3031 
3032 	for (sp = list_head(&oo->ro_statelist); sp != NULL;
3033 	    sp = list_next(&oo->ro_statelist, sp)) {
3034 		rfs4_state_close(sp, FALSE, close_of_client, CRED());
3035 		if (invalidate == TRUE)
3036 			rfs4_dbe_invalidate(sp->rs_dbe);
3037 	}
3038 
3039 	rfs4_dbe_invalidate(oo->ro_dbe);
3040 	rfs4_dbe_unlock(oo->ro_dbe);
3041 }
3042 
3043 static uint32_t
3044 state_owner_file_hash(void *key)
3045 {
3046 	rfs4_state_t *sp = key;
3047 
3048 	return (ADDRHASH(sp->rs_owner) ^ ADDRHASH(sp->rs_finfo));
3049 }
3050 
3051 static bool_t
3052 state_owner_file_compare(rfs4_entry_t u_entry, void *key)
3053 {
3054 	rfs4_state_t *sp = (rfs4_state_t *)u_entry;
3055 	rfs4_state_t *arg = key;
3056 
3057 	if (sp->rs_closed == TRUE)
3058 		return (FALSE);
3059 
3060 	return (arg->rs_owner == sp->rs_owner && arg->rs_finfo == sp->rs_finfo);
3061 }
3062 
3063 static void *
3064 state_owner_file_mkkey(rfs4_entry_t u_entry)
3065 {
3066 	return (u_entry);
3067 }
3068 
3069 static uint32_t
3070 state_file_hash(void *key)
3071 {
3072 	return (ADDRHASH(key));
3073 }
3074 
3075 static bool_t
3076 state_file_compare(rfs4_entry_t u_entry, void *key)
3077 {
3078 	rfs4_state_t *sp = (rfs4_state_t *)u_entry;
3079 	rfs4_file_t *fp = key;
3080 
3081 	if (sp->rs_closed == TRUE)
3082 		return (FALSE);
3083 
3084 	return (fp == sp->rs_finfo);
3085 }
3086 
3087 static void *
3088 state_file_mkkey(rfs4_entry_t u_entry)
3089 {
3090 	rfs4_state_t *sp = (rfs4_state_t *)u_entry;
3091 
3092 	return (sp->rs_finfo);
3093 }
3094 
3095 rfs4_state_t *
3096 rfs4_findstate_by_owner_file(rfs4_openowner_t *oo, rfs4_file_t *fp,
3097 	bool_t *create)
3098 {
3099 	rfs4_state_t *sp;
3100 	rfs4_state_t key;
3101 
3102 	key.rs_owner = oo;
3103 	key.rs_finfo = fp;
3104 
3105 	sp = (rfs4_state_t *)rfs4_dbsearch(rfs4_state_owner_file_idx, &key,
3106 	    create, &key, RFS4_DBS_VALID);
3107 
3108 	return (sp);
3109 }
3110 
3111 /* This returns ANY state struct that refers to this file */
3112 static rfs4_state_t *
3113 rfs4_findstate_by_file(rfs4_file_t *fp)
3114 {
3115 	bool_t create = FALSE;
3116 
3117 	return ((rfs4_state_t *)rfs4_dbsearch(rfs4_state_file_idx, fp,
3118 	    &create, fp, RFS4_DBS_VALID));
3119 }
3120 
3121 static bool_t
3122 rfs4_state_expiry(rfs4_entry_t u_entry)
3123 {
3124 	rfs4_state_t *sp = (rfs4_state_t *)u_entry;
3125 
3126 	if (rfs4_dbe_is_invalid(sp->rs_dbe))
3127 		return (TRUE);
3128 
3129 	if (sp->rs_closed == TRUE &&
3130 	    ((gethrestime_sec() - rfs4_dbe_get_timerele(sp->rs_dbe))
3131 	    > rfs4_lease_time))
3132 		return (TRUE);
3133 
3134 	return ((gethrestime_sec() - sp->rs_owner->ro_client->rc_last_access
3135 	    > rfs4_lease_time));
3136 }
3137 
3138 static bool_t
3139 rfs4_state_create(rfs4_entry_t u_entry, void *argp)
3140 {
3141 	rfs4_state_t *sp = (rfs4_state_t *)u_entry;
3142 	rfs4_file_t *fp = ((rfs4_state_t *)argp)->rs_finfo;
3143 	rfs4_openowner_t *oo = ((rfs4_state_t *)argp)->rs_owner;
3144 
3145 	rfs4_dbe_hold(fp->rf_dbe);
3146 	rfs4_dbe_hold(oo->ro_dbe);
3147 	sp->rs_stateid = get_stateid(rfs4_dbe_getid(sp->rs_dbe));
3148 	sp->rs_stateid.bits.type = OPENID;
3149 	sp->rs_owner = oo;
3150 	sp->rs_finfo = fp;
3151 
3152 	list_create(&sp->rs_lostatelist, sizeof (rfs4_lo_state_t),
3153 	    offsetof(rfs4_lo_state_t, rls_node));
3154 
3155 	/* Insert state on per open owner's list */
3156 	rfs4_dbe_lock(oo->ro_dbe);
3157 	list_insert_tail(&oo->ro_statelist, sp);
3158 	rfs4_dbe_unlock(oo->ro_dbe);
3159 
3160 	return (TRUE);
3161 }
3162 
3163 static rfs4_state_t *
3164 rfs4_findstate(stateid_t *id, rfs4_dbsearch_type_t find_invalid, bool_t lock_fp)
3165 {
3166 	rfs4_state_t *sp;
3167 	bool_t create = FALSE;
3168 
3169 	sp = (rfs4_state_t *)rfs4_dbsearch(rfs4_state_idx, id,
3170 	    &create, NULL, find_invalid);
3171 	if (lock_fp == TRUE && sp != NULL)
3172 		rw_enter(&sp->rs_finfo->rf_file_rwlock, RW_READER);
3173 
3174 	return (sp);
3175 }
3176 
3177 void
3178 rfs4_state_close(rfs4_state_t *sp, bool_t lock_held, bool_t close_of_client,
3179     cred_t *cr)
3180 {
3181 	/* Remove the associated lo_state owners */
3182 	if (!lock_held)
3183 		rfs4_dbe_lock(sp->rs_dbe);
3184 
3185 	/*
3186 	 * If refcnt == 0, the dbe is about to be destroyed.
3187 	 * lock state will be released by the reaper thread.
3188 	 */
3189 
3190 	if (rfs4_dbe_refcnt(sp->rs_dbe) > 0) {
3191 		if (sp->rs_closed == FALSE) {
3192 			rfs4_release_share_lock_state(sp, cr, close_of_client);
3193 			sp->rs_closed = TRUE;
3194 		}
3195 	}
3196 
3197 	if (!lock_held)
3198 		rfs4_dbe_unlock(sp->rs_dbe);
3199 }
3200 
3201 /*
3202  * Remove all state associated with the given client.
3203  */
3204 void
3205 rfs4_client_state_remove(rfs4_client_t *cp)
3206 {
3207 	rfs4_openowner_t *oo;
3208 
3209 	rfs4_dbe_lock(cp->rc_dbe);
3210 
3211 	for (oo = list_head(&cp->rc_openownerlist); oo != NULL;
3212 	    oo = list_next(&cp->rc_openownerlist, oo)) {
3213 		rfs4_free_opens(oo, TRUE, TRUE);
3214 	}
3215 
3216 	rfs4_dbe_unlock(cp->rc_dbe);
3217 }
3218 
3219 void
3220 rfs4_client_close(rfs4_client_t *cp)
3221 {
3222 	/* Mark client as going away. */
3223 	rfs4_dbe_lock(cp->rc_dbe);
3224 	rfs4_dbe_invalidate(cp->rc_dbe);
3225 	rfs4_dbe_unlock(cp->rc_dbe);
3226 
3227 	rfs4_client_state_remove(cp);
3228 
3229 	/* Release the client */
3230 	rfs4_client_rele(cp);
3231 }
3232 
3233 nfsstat4
3234 rfs4_check_clientid(clientid4 *cp, int setclid_confirm)
3235 {
3236 	cid *cidp = (cid *) cp;
3237 
3238 	/*
3239 	 * If we are booted as a cluster node, check the embedded nodeid.
3240 	 * If it indicates that this clientid was generated on another node,
3241 	 * inform the client accordingly.
3242 	 */
3243 	if (cluster_bootflags & CLUSTER_BOOTED && foreign_clientid(cidp))
3244 		return (NFS4ERR_STALE_CLIENTID);
3245 
3246 	/*
3247 	 * If the server start time matches the time provided
3248 	 * by the client (via the clientid) and this is NOT a
3249 	 * setclientid_confirm then return EXPIRED.
3250 	 */
3251 	if (!setclid_confirm && cidp->impl_id.start_time == rfs4_start_time)
3252 		return (NFS4ERR_EXPIRED);
3253 
3254 	return (NFS4ERR_STALE_CLIENTID);
3255 }
3256 
3257 /*
3258  * This is used when a stateid has not been found amongst the
3259  * current server's state.  Check the stateid to see if it
3260  * was from this server instantiation or not.
3261  */
3262 static nfsstat4
3263 what_stateid_error(stateid_t *id, stateid_type_t type)
3264 {
3265 	/* If we are booted as a cluster node, was stateid locally generated? */
3266 	if ((cluster_bootflags & CLUSTER_BOOTED) && foreign_stateid(id))
3267 		return (NFS4ERR_STALE_STATEID);
3268 
3269 	/* If types don't match then no use checking further */
3270 	if (type != id->bits.type)
3271 		return (NFS4ERR_BAD_STATEID);
3272 
3273 	/* From a previous server instantiation, return STALE */
3274 	if (id->bits.boottime < rfs4_start_time)
3275 		return (NFS4ERR_STALE_STATEID);
3276 
3277 	/*
3278 	 * From this server but the state is most likely beyond lease
3279 	 * timeout: return NFS4ERR_EXPIRED.  However, there is the
3280 	 * case of a delegation stateid.  For delegations, there is a
3281 	 * case where the state can be removed without the client's
3282 	 * knowledge/consent: revocation.  In the case of delegation
3283 	 * revocation, the delegation state will be removed and will
3284 	 * not be found.  If the client does something like a
3285 	 * DELEGRETURN or even a READ/WRITE with a delegatoin stateid
3286 	 * that has been revoked, the server should return BAD_STATEID
3287 	 * instead of the more common EXPIRED error.
3288 	 */
3289 	if (id->bits.boottime == rfs4_start_time) {
3290 		if (type == DELEGID)
3291 			return (NFS4ERR_BAD_STATEID);
3292 		else
3293 			return (NFS4ERR_EXPIRED);
3294 	}
3295 
3296 	return (NFS4ERR_BAD_STATEID);
3297 }
3298 
3299 /*
3300  * Used later on to find the various state structs.  When called from
3301  * rfs4_check_stateid()->rfs4_get_all_state(), no file struct lock is
3302  * taken (it is not needed) and helps on the read/write path with
3303  * respect to performance.
3304  */
3305 static nfsstat4
3306 rfs4_get_state_lockit(stateid4 *stateid, rfs4_state_t **spp,
3307     rfs4_dbsearch_type_t find_invalid, bool_t lock_fp)
3308 {
3309 	stateid_t *id = (stateid_t *)stateid;
3310 	rfs4_state_t *sp;
3311 
3312 	*spp = NULL;
3313 
3314 	/* If we are booted as a cluster node, was stateid locally generated? */
3315 	if ((cluster_bootflags & CLUSTER_BOOTED) && foreign_stateid(id))
3316 		return (NFS4ERR_STALE_STATEID);
3317 
3318 	sp = rfs4_findstate(id, find_invalid, lock_fp);
3319 	if (sp == NULL) {
3320 		return (what_stateid_error(id, OPENID));
3321 	}
3322 
3323 	if (rfs4_lease_expired(sp->rs_owner->ro_client)) {
3324 		if (lock_fp == TRUE)
3325 			rfs4_state_rele(sp);
3326 		else
3327 			rfs4_state_rele_nounlock(sp);
3328 		return (NFS4ERR_EXPIRED);
3329 	}
3330 
3331 	*spp = sp;
3332 
3333 	return (NFS4_OK);
3334 }
3335 
3336 nfsstat4
3337 rfs4_get_state(stateid4 *stateid, rfs4_state_t **spp,
3338     rfs4_dbsearch_type_t find_invalid)
3339 {
3340 	return (rfs4_get_state_lockit(stateid, spp, find_invalid, TRUE));
3341 }
3342 
3343 int
3344 rfs4_check_stateid_seqid(rfs4_state_t *sp, stateid4 *stateid)
3345 {
3346 	stateid_t *id = (stateid_t *)stateid;
3347 
3348 	if (rfs4_lease_expired(sp->rs_owner->ro_client))
3349 		return (NFS4_CHECK_STATEID_EXPIRED);
3350 
3351 	/* Stateid is some time in the future - that's bad */
3352 	if (sp->rs_stateid.bits.chgseq < id->bits.chgseq)
3353 		return (NFS4_CHECK_STATEID_BAD);
3354 
3355 	if (sp->rs_stateid.bits.chgseq == id->bits.chgseq + 1)
3356 		return (NFS4_CHECK_STATEID_REPLAY);
3357 
3358 	/* Stateid is some time in the past - that's old */
3359 	if (sp->rs_stateid.bits.chgseq > id->bits.chgseq)
3360 		return (NFS4_CHECK_STATEID_OLD);
3361 
3362 	/* Caller needs to know about confirmation before closure */
3363 	if (sp->rs_owner->ro_need_confirm)
3364 		return (NFS4_CHECK_STATEID_UNCONFIRMED);
3365 
3366 	if (sp->rs_closed == TRUE)
3367 		return (NFS4_CHECK_STATEID_CLOSED);
3368 
3369 	return (NFS4_CHECK_STATEID_OKAY);
3370 }
3371 
3372 int
3373 rfs4_check_lo_stateid_seqid(rfs4_lo_state_t *lsp, stateid4 *stateid)
3374 {
3375 	stateid_t *id = (stateid_t *)stateid;
3376 
3377 	if (rfs4_lease_expired(lsp->rls_state->rs_owner->ro_client))
3378 		return (NFS4_CHECK_STATEID_EXPIRED);
3379 
3380 	/* Stateid is some time in the future - that's bad */
3381 	if (lsp->rls_lockid.bits.chgseq < id->bits.chgseq)
3382 		return (NFS4_CHECK_STATEID_BAD);
3383 
3384 	if (lsp->rls_lockid.bits.chgseq == id->bits.chgseq + 1)
3385 		return (NFS4_CHECK_STATEID_REPLAY);
3386 
3387 	/* Stateid is some time in the past - that's old */
3388 	if (lsp->rls_lockid.bits.chgseq > id->bits.chgseq)
3389 		return (NFS4_CHECK_STATEID_OLD);
3390 
3391 	if (lsp->rls_state->rs_closed == TRUE)
3392 		return (NFS4_CHECK_STATEID_CLOSED);
3393 
3394 	return (NFS4_CHECK_STATEID_OKAY);
3395 }
3396 
3397 nfsstat4
3398 rfs4_get_deleg_state(stateid4 *stateid, rfs4_deleg_state_t **dspp)
3399 {
3400 	stateid_t *id = (stateid_t *)stateid;
3401 	rfs4_deleg_state_t *dsp;
3402 
3403 	*dspp = NULL;
3404 
3405 	/* If we are booted as a cluster node, was stateid locally generated? */
3406 	if ((cluster_bootflags & CLUSTER_BOOTED) && foreign_stateid(id))
3407 		return (NFS4ERR_STALE_STATEID);
3408 
3409 	dsp = rfs4_finddelegstate(id);
3410 	if (dsp == NULL) {
3411 		return (what_stateid_error(id, DELEGID));
3412 	}
3413 
3414 	if (rfs4_lease_expired(dsp->rds_client)) {
3415 		rfs4_deleg_state_rele(dsp);
3416 		return (NFS4ERR_EXPIRED);
3417 	}
3418 
3419 	*dspp = dsp;
3420 
3421 	return (NFS4_OK);
3422 }
3423 
3424 nfsstat4
3425 rfs4_get_lo_state(stateid4 *stateid, rfs4_lo_state_t **lspp, bool_t lock_fp)
3426 {
3427 	stateid_t *id = (stateid_t *)stateid;
3428 	rfs4_lo_state_t *lsp;
3429 
3430 	*lspp = NULL;
3431 
3432 	/* If we are booted as a cluster node, was stateid locally generated? */
3433 	if ((cluster_bootflags & CLUSTER_BOOTED) && foreign_stateid(id))
3434 		return (NFS4ERR_STALE_STATEID);
3435 
3436 	lsp = rfs4_findlo_state(id, lock_fp);
3437 	if (lsp == NULL) {
3438 		return (what_stateid_error(id, LOCKID));
3439 	}
3440 
3441 	if (rfs4_lease_expired(lsp->rls_state->rs_owner->ro_client)) {
3442 		rfs4_lo_state_rele(lsp, lock_fp);
3443 		return (NFS4ERR_EXPIRED);
3444 	}
3445 
3446 	*lspp = lsp;
3447 
3448 	return (NFS4_OK);
3449 }
3450 
3451 static nfsstat4
3452 rfs4_get_all_state(stateid4 *sid, rfs4_state_t **spp,
3453     rfs4_deleg_state_t **dspp, rfs4_lo_state_t **lspp)
3454 {
3455 	rfs4_state_t *sp = NULL;
3456 	rfs4_deleg_state_t *dsp = NULL;
3457 	rfs4_lo_state_t *lsp = NULL;
3458 	stateid_t *id;
3459 	nfsstat4 status;
3460 
3461 	*spp = NULL; *dspp = NULL; *lspp = NULL;
3462 
3463 	id = (stateid_t *)sid;
3464 	switch (id->bits.type) {
3465 	case OPENID:
3466 		status = rfs4_get_state_lockit(sid, &sp, FALSE, FALSE);
3467 		break;
3468 	case DELEGID:
3469 		status = rfs4_get_deleg_state(sid, &dsp);
3470 		break;
3471 	case LOCKID:
3472 		status = rfs4_get_lo_state(sid, &lsp, FALSE);
3473 		if (status == NFS4_OK) {
3474 			sp = lsp->rls_state;
3475 			rfs4_dbe_hold(sp->rs_dbe);
3476 		}
3477 		break;
3478 	default:
3479 		status = NFS4ERR_BAD_STATEID;
3480 	}
3481 
3482 	if (status == NFS4_OK) {
3483 		*spp = sp;
3484 		*dspp = dsp;
3485 		*lspp = lsp;
3486 	}
3487 
3488 	return (status);
3489 }
3490 
3491 /*
3492  * Given the I/O mode (FREAD or FWRITE), this checks whether the
3493  * rfs4_state_t struct has access to do this operation and if so
3494  * return NFS4_OK; otherwise the proper NFSv4 error is returned.
3495  */
3496 nfsstat4
3497 rfs4_state_has_access(rfs4_state_t *sp, int mode, vnode_t *vp)
3498 {
3499 	nfsstat4 stat = NFS4_OK;
3500 	rfs4_file_t *fp;
3501 	bool_t create = FALSE;
3502 
3503 	rfs4_dbe_lock(sp->rs_dbe);
3504 	if (mode == FWRITE) {
3505 		if (!(sp->rs_share_access & OPEN4_SHARE_ACCESS_WRITE)) {
3506 			stat = NFS4ERR_OPENMODE;
3507 		}
3508 	} else if (mode == FREAD) {
3509 		if (!(sp->rs_share_access & OPEN4_SHARE_ACCESS_READ)) {
3510 			/*
3511 			 * If we have OPENed the file with DENYing access
3512 			 * to both READ and WRITE then no one else could
3513 			 * have OPENed the file, hence no conflicting READ
3514 			 * deny.  This check is merely an optimization.
3515 			 */
3516 			if (sp->rs_share_deny == OPEN4_SHARE_DENY_BOTH)
3517 				goto out;
3518 
3519 			/* Check against file struct's DENY mode */
3520 			fp = rfs4_findfile(vp, NULL, &create);
3521 			if (fp != NULL) {
3522 				int deny_read = 0;
3523 				rfs4_dbe_lock(fp->rf_dbe);
3524 				/*
3525 				 * Check if any other open owner has the file
3526 				 * OPENed with deny READ.
3527 				 */
3528 				if (sp->rs_share_deny & OPEN4_SHARE_DENY_READ)
3529 					deny_read = 1;
3530 				ASSERT(fp->rf_deny_read >= deny_read);
3531 				if (fp->rf_deny_read > deny_read)
3532 					stat = NFS4ERR_OPENMODE;
3533 				rfs4_dbe_unlock(fp->rf_dbe);
3534 				rfs4_file_rele(fp);
3535 			}
3536 		}
3537 	} else {
3538 		/* Illegal I/O mode */
3539 		stat = NFS4ERR_INVAL;
3540 	}
3541 out:
3542 	rfs4_dbe_unlock(sp->rs_dbe);
3543 	return (stat);
3544 }
3545 
3546 /*
3547  * Given the I/O mode (FREAD or FWRITE), the vnode, the stateid and whether
3548  * the file is being truncated, return NFS4_OK if allowed or appropriate
3549  * V4 error if not. Note NFS4ERR_DELAY will be returned and a recall on
3550  * the associated file will be done if the I/O is not consistent with any
3551  * delegation in effect on the file. Should be holding VOP_RWLOCK, either
3552  * as reader or writer as appropriate. rfs4_op_open will acquire the
3553  * VOP_RWLOCK as writer when setting up delegation. If the stateid is bad
3554  * this routine will return NFS4ERR_BAD_STATEID. In addition, through the
3555  * deleg parameter, we will return whether a write delegation is held by
3556  * the client associated with this stateid.
3557  * If the server instance associated with the relevant client is in its
3558  * grace period, return NFS4ERR_GRACE.
3559  */
3560 
3561 nfsstat4
3562 rfs4_check_stateid(int mode, vnode_t *vp,
3563     stateid4 *stateid, bool_t trunc, bool_t *deleg,
3564     bool_t do_access, caller_context_t *ct)
3565 {
3566 	rfs4_file_t *fp;
3567 	bool_t create = FALSE;
3568 	rfs4_state_t *sp;
3569 	rfs4_deleg_state_t *dsp;
3570 	rfs4_lo_state_t *lsp;
3571 	stateid_t *id = (stateid_t *)stateid;
3572 	nfsstat4 stat = NFS4_OK;
3573 
3574 	if (ct != NULL) {
3575 		ct->cc_sysid = 0;
3576 		ct->cc_pid = 0;
3577 		ct->cc_caller_id = nfs4_srv_caller_id;
3578 		ct->cc_flags = CC_DONTBLOCK;
3579 	}
3580 
3581 	if (ISSPECIAL(stateid)) {
3582 		fp = rfs4_findfile(vp, NULL, &create);
3583 		if (fp == NULL)
3584 			return (NFS4_OK);
3585 		if (fp->rf_dinfo.rd_dtype == OPEN_DELEGATE_NONE) {
3586 			rfs4_file_rele(fp);
3587 			return (NFS4_OK);
3588 		}
3589 		if (mode == FWRITE ||
3590 		    fp->rf_dinfo.rd_dtype == OPEN_DELEGATE_WRITE) {
3591 			rfs4_recall_deleg(fp, trunc, NULL);
3592 			rfs4_file_rele(fp);
3593 			return (NFS4ERR_DELAY);
3594 		}
3595 		rfs4_file_rele(fp);
3596 		return (NFS4_OK);
3597 	} else {
3598 		stat = rfs4_get_all_state(stateid, &sp, &dsp, &lsp);
3599 		if (stat != NFS4_OK)
3600 			return (stat);
3601 		if (lsp != NULL) {
3602 			/* Is associated server instance in its grace period? */
3603 			if (rfs4_clnt_in_grace(lsp->rls_locker->rl_client)) {
3604 				rfs4_lo_state_rele(lsp, FALSE);
3605 				if (sp != NULL)
3606 					rfs4_state_rele_nounlock(sp);
3607 				return (NFS4ERR_GRACE);
3608 			}
3609 			if (id->bits.type == LOCKID) {
3610 				/* Seqid in the future? - that's bad */
3611 				if (lsp->rls_lockid.bits.chgseq <
3612 				    id->bits.chgseq) {
3613 					rfs4_lo_state_rele(lsp, FALSE);
3614 					if (sp != NULL)
3615 						rfs4_state_rele_nounlock(sp);
3616 					return (NFS4ERR_BAD_STATEID);
3617 				}
3618 				/* Seqid in the past? - that's old */
3619 				if (lsp->rls_lockid.bits.chgseq >
3620 				    id->bits.chgseq) {
3621 					rfs4_lo_state_rele(lsp, FALSE);
3622 					if (sp != NULL)
3623 						rfs4_state_rele_nounlock(sp);
3624 					return (NFS4ERR_OLD_STATEID);
3625 				}
3626 				/* Ensure specified filehandle matches */
3627 				if (lsp->rls_state->rs_finfo->rf_vp != vp) {
3628 					rfs4_lo_state_rele(lsp, FALSE);
3629 					if (sp != NULL)
3630 						rfs4_state_rele_nounlock(sp);
3631 					return (NFS4ERR_BAD_STATEID);
3632 				}
3633 			}
3634 			if (ct != NULL) {
3635 				ct->cc_sysid =
3636 				    lsp->rls_locker->rl_client->rc_sysidt;
3637 				ct->cc_pid = lsp->rls_locker->rl_pid;
3638 			}
3639 			rfs4_lo_state_rele(lsp, FALSE);
3640 		}
3641 
3642 		/* Stateid provided was an "open" stateid */
3643 		if (sp != NULL) {
3644 			/* Is associated server instance in its grace period? */
3645 			if (rfs4_clnt_in_grace(sp->rs_owner->ro_client)) {
3646 				rfs4_state_rele_nounlock(sp);
3647 				return (NFS4ERR_GRACE);
3648 			}
3649 			if (id->bits.type == OPENID) {
3650 				/* Seqid in the future? - that's bad */
3651 				if (sp->rs_stateid.bits.chgseq <
3652 				    id->bits.chgseq) {
3653 					rfs4_state_rele_nounlock(sp);
3654 					return (NFS4ERR_BAD_STATEID);
3655 				}
3656 				/* Seqid in the past - that's old */
3657 				if (sp->rs_stateid.bits.chgseq >
3658 				    id->bits.chgseq) {
3659 					rfs4_state_rele_nounlock(sp);
3660 					return (NFS4ERR_OLD_STATEID);
3661 				}
3662 			}
3663 			/* Ensure specified filehandle matches */
3664 			if (sp->rs_finfo->rf_vp != vp) {
3665 				rfs4_state_rele_nounlock(sp);
3666 				return (NFS4ERR_BAD_STATEID);
3667 			}
3668 
3669 			if (sp->rs_owner->ro_need_confirm) {
3670 				rfs4_state_rele_nounlock(sp);
3671 				return (NFS4ERR_BAD_STATEID);
3672 			}
3673 
3674 			if (sp->rs_closed == TRUE) {
3675 				rfs4_state_rele_nounlock(sp);
3676 				return (NFS4ERR_OLD_STATEID);
3677 			}
3678 
3679 			if (do_access)
3680 				stat = rfs4_state_has_access(sp, mode, vp);
3681 			else
3682 				stat = NFS4_OK;
3683 
3684 			/*
3685 			 * Return whether this state has write
3686 			 * delegation if desired
3687 			 */
3688 			if (deleg && (sp->rs_finfo->rf_dinfo.rd_dtype ==
3689 			    OPEN_DELEGATE_WRITE))
3690 				*deleg = TRUE;
3691 
3692 			/*
3693 			 * We got a valid stateid, so we update the
3694 			 * lease on the client. Ideally we would like
3695 			 * to do this after the calling op succeeds,
3696 			 * but for now this will be good
3697 			 * enough. Callers of this routine are
3698 			 * currently insulated from the state stuff.
3699 			 */
3700 			rfs4_update_lease(sp->rs_owner->ro_client);
3701 
3702 			/*
3703 			 * If a delegation is present on this file and
3704 			 * this is a WRITE, then update the lastwrite
3705 			 * time to indicate that activity is present.
3706 			 */
3707 			if (sp->rs_finfo->rf_dinfo.rd_dtype ==
3708 			    OPEN_DELEGATE_WRITE &&
3709 			    mode == FWRITE) {
3710 				sp->rs_finfo->rf_dinfo.rd_time_lastwrite =
3711 				    gethrestime_sec();
3712 			}
3713 
3714 			rfs4_state_rele_nounlock(sp);
3715 
3716 			return (stat);
3717 		}
3718 
3719 		if (dsp != NULL) {
3720 			/* Is associated server instance in its grace period? */
3721 			if (rfs4_clnt_in_grace(dsp->rds_client)) {
3722 				rfs4_deleg_state_rele(dsp);
3723 				return (NFS4ERR_GRACE);
3724 			}
3725 			if (dsp->rds_delegid.bits.chgseq != id->bits.chgseq) {
3726 				rfs4_deleg_state_rele(dsp);
3727 				return (NFS4ERR_BAD_STATEID);
3728 			}
3729 
3730 			/* Ensure specified filehandle matches */
3731 			if (dsp->rds_finfo->rf_vp != vp) {
3732 				rfs4_deleg_state_rele(dsp);
3733 				return (NFS4ERR_BAD_STATEID);
3734 			}
3735 			/*
3736 			 * Return whether this state has write
3737 			 * delegation if desired
3738 			 */
3739 			if (deleg && (dsp->rds_finfo->rf_dinfo.rd_dtype ==
3740 			    OPEN_DELEGATE_WRITE))
3741 				*deleg = TRUE;
3742 
3743 			rfs4_update_lease(dsp->rds_client);
3744 
3745 			/*
3746 			 * If a delegation is present on this file and
3747 			 * this is a WRITE, then update the lastwrite
3748 			 * time to indicate that activity is present.
3749 			 */
3750 			if (dsp->rds_finfo->rf_dinfo.rd_dtype ==
3751 			    OPEN_DELEGATE_WRITE && mode == FWRITE) {
3752 				dsp->rds_finfo->rf_dinfo.rd_time_lastwrite =
3753 				    gethrestime_sec();
3754 			}
3755 
3756 			/*
3757 			 * XXX - what happens if this is a WRITE and the
3758 			 * delegation type of for READ.
3759 			 */
3760 			rfs4_deleg_state_rele(dsp);
3761 
3762 			return (stat);
3763 		}
3764 		/*
3765 		 * If we got this far, something bad happened
3766 		 */
3767 		return (NFS4ERR_BAD_STATEID);
3768 	}
3769 }
3770 
3771 
3772 /*
3773  * This is a special function in that for the file struct provided the
3774  * server wants to remove/close all current state associated with the
3775  * file.  The prime use of this would be with OP_REMOVE to force the
3776  * release of state and particularly of file locks.
3777  *
3778  * There is an assumption that there is no delegations outstanding on
3779  * this file at this point.  The caller should have waited for those
3780  * to be returned or revoked.
3781  */
3782 void
3783 rfs4_close_all_state(rfs4_file_t *fp)
3784 {
3785 	rfs4_state_t *sp;
3786 
3787 	rfs4_dbe_lock(fp->rf_dbe);
3788 
3789 #ifdef DEBUG
3790 	/* only applies when server is handing out delegations */
3791 	if (rfs4_deleg_policy != SRV_NEVER_DELEGATE)
3792 		ASSERT(fp->rf_dinfo.rd_hold_grant > 0);
3793 #endif
3794 
3795 	/* No delegations for this file */
3796 	ASSERT(list_is_empty(&fp->rf_delegstatelist));
3797 
3798 	/* Make sure that it can not be found */
3799 	rfs4_dbe_invalidate(fp->rf_dbe);
3800 
3801 	if (fp->rf_vp == NULL) {
3802 		rfs4_dbe_unlock(fp->rf_dbe);
3803 		return;
3804 	}
3805 	rfs4_dbe_unlock(fp->rf_dbe);
3806 
3807 	/*
3808 	 * Hold as writer to prevent other server threads from
3809 	 * processing requests related to the file while all state is
3810 	 * being removed.
3811 	 */
3812 	rw_enter(&fp->rf_file_rwlock, RW_WRITER);
3813 
3814 	/* Remove ALL state from the file */
3815 	while (sp = rfs4_findstate_by_file(fp)) {
3816 		rfs4_state_close(sp, FALSE, FALSE, CRED());
3817 		rfs4_state_rele_nounlock(sp);
3818 	}
3819 
3820 	/*
3821 	 * This is only safe since there are no further references to
3822 	 * the file.
3823 	 */
3824 	rfs4_dbe_lock(fp->rf_dbe);
3825 	if (fp->rf_vp) {
3826 		vnode_t *vp = fp->rf_vp;
3827 
3828 		mutex_enter(&vp->v_vsd_lock);
3829 		(void) vsd_set(vp, nfs4_srv_vkey, NULL);
3830 		mutex_exit(&vp->v_vsd_lock);
3831 		VN_RELE(vp);
3832 		fp->rf_vp = NULL;
3833 	}
3834 	rfs4_dbe_unlock(fp->rf_dbe);
3835 
3836 	/* Finally let other references to proceed */
3837 	rw_exit(&fp->rf_file_rwlock);
3838 }
3839 
3840 /*
3841  * This function is used as a target for the rfs4_dbe_walk() call
3842  * below.  The purpose of this function is to see if the
3843  * lockowner_state refers to a file that resides within the exportinfo
3844  * export.  If so, then remove the lock_owner state (file locks and
3845  * share "locks") for this object since the intent is the server is
3846  * unexporting the specified directory.  Be sure to invalidate the
3847  * object after the state has been released
3848  */
3849 static void
3850 rfs4_lo_state_walk_callout(rfs4_entry_t u_entry, void *e)
3851 {
3852 	rfs4_lo_state_t *lsp = (rfs4_lo_state_t *)u_entry;
3853 	struct exportinfo *exi = (struct exportinfo *)e;
3854 	nfs_fh4_fmt_t   fhfmt4, *exi_fhp, *finfo_fhp;
3855 	fhandle_t *efhp;
3856 
3857 	efhp = (fhandle_t *)&exi->exi_fh;
3858 	exi_fhp = (nfs_fh4_fmt_t *)&fhfmt4;
3859 
3860 	FH_TO_FMT4(efhp, exi_fhp);
3861 
3862 	finfo_fhp = (nfs_fh4_fmt_t *)lsp->rls_state->rs_finfo->
3863 	    rf_filehandle.nfs_fh4_val;
3864 
3865 	if (EQFSID(&finfo_fhp->fh4_fsid, &exi_fhp->fh4_fsid) &&
3866 	    bcmp(&finfo_fhp->fh4_xdata, &exi_fhp->fh4_xdata,
3867 	    exi_fhp->fh4_xlen) == 0) {
3868 		rfs4_state_close(lsp->rls_state, FALSE, FALSE, CRED());
3869 		rfs4_dbe_invalidate(lsp->rls_dbe);
3870 		rfs4_dbe_invalidate(lsp->rls_state->rs_dbe);
3871 	}
3872 }
3873 
3874 /*
3875  * This function is used as a target for the rfs4_dbe_walk() call
3876  * below.  The purpose of this function is to see if the state refers
3877  * to a file that resides within the exportinfo export.  If so, then
3878  * remove the open state for this object since the intent is the
3879  * server is unexporting the specified directory.  The main result for
3880  * this type of entry is to invalidate it such it will not be found in
3881  * the future.
3882  */
3883 static void
3884 rfs4_state_walk_callout(rfs4_entry_t u_entry, void *e)
3885 {
3886 	rfs4_state_t *sp = (rfs4_state_t *)u_entry;
3887 	struct exportinfo *exi = (struct exportinfo *)e;
3888 	nfs_fh4_fmt_t   fhfmt4, *exi_fhp, *finfo_fhp;
3889 	fhandle_t *efhp;
3890 
3891 	efhp = (fhandle_t *)&exi->exi_fh;
3892 	exi_fhp = (nfs_fh4_fmt_t *)&fhfmt4;
3893 
3894 	FH_TO_FMT4(efhp, exi_fhp);
3895 
3896 	finfo_fhp =
3897 	    (nfs_fh4_fmt_t *)sp->rs_finfo->rf_filehandle.nfs_fh4_val;
3898 
3899 	if (EQFSID(&finfo_fhp->fh4_fsid, &exi_fhp->fh4_fsid) &&
3900 	    bcmp(&finfo_fhp->fh4_xdata, &exi_fhp->fh4_xdata,
3901 	    exi_fhp->fh4_xlen) == 0) {
3902 		rfs4_state_close(sp, TRUE, FALSE, CRED());
3903 		rfs4_dbe_invalidate(sp->rs_dbe);
3904 	}
3905 }
3906 
3907 /*
3908  * This function is used as a target for the rfs4_dbe_walk() call
3909  * below.  The purpose of this function is to see if the state refers
3910  * to a file that resides within the exportinfo export.  If so, then
3911  * remove the deleg state for this object since the intent is the
3912  * server is unexporting the specified directory.  The main result for
3913  * this type of entry is to invalidate it such it will not be found in
3914  * the future.
3915  */
3916 static void
3917 rfs4_deleg_state_walk_callout(rfs4_entry_t u_entry, void *e)
3918 {
3919 	rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
3920 	struct exportinfo *exi = (struct exportinfo *)e;
3921 	nfs_fh4_fmt_t   fhfmt4, *exi_fhp, *finfo_fhp;
3922 	fhandle_t *efhp;
3923 
3924 	efhp = (fhandle_t *)&exi->exi_fh;
3925 	exi_fhp = (nfs_fh4_fmt_t *)&fhfmt4;
3926 
3927 	FH_TO_FMT4(efhp, exi_fhp);
3928 
3929 	finfo_fhp =
3930 	    (nfs_fh4_fmt_t *)dsp->rds_finfo->rf_filehandle.nfs_fh4_val;
3931 
3932 	if (EQFSID(&finfo_fhp->fh4_fsid, &exi_fhp->fh4_fsid) &&
3933 	    bcmp(&finfo_fhp->fh4_xdata, &exi_fhp->fh4_xdata,
3934 	    exi_fhp->fh4_xlen) == 0) {
3935 		rfs4_dbe_invalidate(dsp->rds_dbe);
3936 	}
3937 }
3938 
3939 /*
3940  * This function is used as a target for the rfs4_dbe_walk() call
3941  * below.  The purpose of this function is to see if the state refers
3942  * to a file that resides within the exportinfo export.  If so, then
3943  * release vnode hold for this object since the intent is the server
3944  * is unexporting the specified directory.  Invalidation will prevent
3945  * this struct from being found in the future.
3946  */
3947 static void
3948 rfs4_file_walk_callout(rfs4_entry_t u_entry, void *e)
3949 {
3950 	rfs4_file_t *fp = (rfs4_file_t *)u_entry;
3951 	struct exportinfo *exi = (struct exportinfo *)e;
3952 	nfs_fh4_fmt_t   fhfmt4, *exi_fhp, *finfo_fhp;
3953 	fhandle_t *efhp;
3954 
3955 	efhp = (fhandle_t *)&exi->exi_fh;
3956 	exi_fhp = (nfs_fh4_fmt_t *)&fhfmt4;
3957 
3958 	FH_TO_FMT4(efhp, exi_fhp);
3959 
3960 	finfo_fhp = (nfs_fh4_fmt_t *)fp->rf_filehandle.nfs_fh4_val;
3961 
3962 	if (EQFSID(&finfo_fhp->fh4_fsid, &exi_fhp->fh4_fsid) &&
3963 	    bcmp(&finfo_fhp->fh4_xdata, &exi_fhp->fh4_xdata,
3964 	    exi_fhp->fh4_xlen) == 0) {
3965 		if (fp->rf_vp) {
3966 			vnode_t *vp = fp->rf_vp;
3967 
3968 			/*
3969 			 * don't leak monitors and remove the reference
3970 			 * put on the vnode when the delegation was granted.
3971 			 */
3972 			if (fp->rf_dinfo.rd_dtype == OPEN_DELEGATE_READ) {
3973 				(void) fem_uninstall(vp, deleg_rdops,
3974 				    (void *)fp);
3975 				vn_open_downgrade(vp, FREAD);
3976 			} else if (fp->rf_dinfo.rd_dtype ==
3977 			    OPEN_DELEGATE_WRITE) {
3978 				(void) fem_uninstall(vp, deleg_wrops,
3979 				    (void *)fp);
3980 				vn_open_downgrade(vp, FREAD|FWRITE);
3981 			}
3982 			mutex_enter(&vp->v_vsd_lock);
3983 			(void) vsd_set(vp, nfs4_srv_vkey, NULL);
3984 			mutex_exit(&vp->v_vsd_lock);
3985 			VN_RELE(vp);
3986 			fp->rf_vp = NULL;
3987 		}
3988 		rfs4_dbe_invalidate(fp->rf_dbe);
3989 	}
3990 }
3991 
3992 /*
3993  * Given a directory that is being unexported, cleanup/release all
3994  * state in the server that refers to objects residing underneath this
3995  * particular export.  The ordering of the release is important.
3996  * Lock_owner, then state and then file.
3997  */
3998 void
3999 rfs4_clean_state_exi(struct exportinfo *exi)
4000 {
4001 	mutex_enter(&rfs4_state_lock);
4002 
4003 	if (rfs4_server_state == NULL) {
4004 		mutex_exit(&rfs4_state_lock);
4005 		return;
4006 	}
4007 
4008 	rfs4_dbe_walk(rfs4_lo_state_tab, rfs4_lo_state_walk_callout, exi);
4009 	rfs4_dbe_walk(rfs4_state_tab, rfs4_state_walk_callout, exi);
4010 	rfs4_dbe_walk(rfs4_deleg_state_tab, rfs4_deleg_state_walk_callout, exi);
4011 	rfs4_dbe_walk(rfs4_file_tab, rfs4_file_walk_callout, exi);
4012 
4013 	mutex_exit(&rfs4_state_lock);
4014 }
4015