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 /*
23 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
24 */
25
26 /*
27 * Copyright 2018 Nexenta Systems, Inc.
28 * Copyright 2019 Nexenta by DDN, Inc.
29 */
30
31 #include <sys/systm.h>
32 #include <sys/kmem.h>
33 #include <sys/cmn_err.h>
34 #include <sys/atomic.h>
35 #include <sys/clconf.h>
36 #include <sys/cladm.h>
37 #include <sys/flock.h>
38 #include <nfs/export.h>
39 #include <nfs/nfs.h>
40 #include <nfs/nfs4.h>
41 #include <nfs/nfssys.h>
42 #include <nfs/lm.h>
43 #include <sys/pathname.h>
44 #include <sys/sdt.h>
45 #include <sys/nvpair.h>
46
47 extern u_longlong_t nfs4_srv_caller_id;
48
49 extern uint_t nfs4_srv_vkey;
50
51 stateid4 special0 = {
52 0,
53 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 }
54 };
55
56 stateid4 special1 = {
57 0xffffffff,
58 {
59 (char)0xff, (char)0xff, (char)0xff, (char)0xff,
60 (char)0xff, (char)0xff, (char)0xff, (char)0xff,
61 (char)0xff, (char)0xff, (char)0xff, (char)0xff
62 }
63 };
64
65
66 #define ISSPECIAL(id) (stateid4_cmp(id, &special0) || \
67 stateid4_cmp(id, &special1))
68
69 /* For embedding the cluster nodeid into our clientid */
70 #define CLUSTER_NODEID_SHIFT 24
71 #define CLUSTER_MAX_NODEID 255
72
73 #ifdef DEBUG
74 int rfs4_debug;
75 #endif
76
77 static uint32_t rfs4_database_debug = 0x00;
78
79 /* CSTYLED */
80 static void rfs4_ss_clid_write(nfs4_srv_t *nsrv4, rfs4_client_t *cp, char *leaf);
81 static void rfs4_ss_clid_write_one(rfs4_client_t *cp, char *dir, char *leaf);
82 static void rfs4_dss_clear_oldstate(rfs4_servinst_t *sip);
83 static void rfs4_ss_chkclid_sip(rfs4_client_t *cp, rfs4_servinst_t *sip);
84
85 /*
86 * Couple of simple init/destroy functions for a general waiter
87 */
88 void
89 rfs4_sw_init(rfs4_state_wait_t *swp)
90 {
91 mutex_init(swp->sw_cv_lock, NULL, MUTEX_DEFAULT, NULL);
92 cv_init(swp->sw_cv, NULL, CV_DEFAULT, NULL);
93 swp->sw_active = FALSE;
94 swp->sw_wait_count = 0;
95 }
96
97 void
98 rfs4_sw_destroy(rfs4_state_wait_t *swp)
99 {
100 mutex_destroy(swp->sw_cv_lock);
101 cv_destroy(swp->sw_cv);
102 }
103
104 void
105 rfs4_sw_enter(rfs4_state_wait_t *swp)
106 {
107 mutex_enter(swp->sw_cv_lock);
108 while (swp->sw_active) {
109 swp->sw_wait_count++;
110 cv_wait(swp->sw_cv, swp->sw_cv_lock);
111 swp->sw_wait_count--;
112 }
113 ASSERT(swp->sw_active == FALSE);
114 swp->sw_active = TRUE;
115 mutex_exit(swp->sw_cv_lock);
116 }
117
118 void
119 rfs4_sw_exit(rfs4_state_wait_t *swp)
120 {
121 mutex_enter(swp->sw_cv_lock);
122 ASSERT(swp->sw_active == TRUE);
123 swp->sw_active = FALSE;
124 if (swp->sw_wait_count != 0)
125 cv_broadcast(swp->sw_cv);
126 mutex_exit(swp->sw_cv_lock);
127 }
128
129 static void
130 deep_lock_copy(LOCK4res *dres, LOCK4res *sres)
131 {
132 lock_owner4 *slo = &sres->LOCK4res_u.denied.owner;
133 lock_owner4 *dlo = &dres->LOCK4res_u.denied.owner;
134
135 if (sres->status == NFS4ERR_DENIED) {
136 dlo->owner_val = kmem_alloc(slo->owner_len, KM_SLEEP);
137 bcopy(slo->owner_val, dlo->owner_val, slo->owner_len);
138 }
139 }
140
141 /*
142 * CPR callback id -- not related to v4 callbacks
143 */
144 static callb_id_t cpr_id = 0;
145
146 static void
147 deep_lock_free(LOCK4res *res)
148 {
149 lock_owner4 *lo = &res->LOCK4res_u.denied.owner;
150
151 if (res->status == NFS4ERR_DENIED)
152 kmem_free(lo->owner_val, lo->owner_len);
153 }
154
155 static void
156 deep_open_copy(OPEN4res *dres, OPEN4res *sres)
157 {
158 nfsace4 *sacep, *dacep;
159
160 if (sres->status != NFS4_OK) {
161 return;
162 }
163
164 dres->attrset = sres->attrset;
165
166 switch (sres->delegation.delegation_type) {
167 case OPEN_DELEGATE_NONE:
168 return;
169 case OPEN_DELEGATE_READ:
170 sacep = &sres->delegation.open_delegation4_u.read.permissions;
171 dacep = &dres->delegation.open_delegation4_u.read.permissions;
172 break;
173 case OPEN_DELEGATE_WRITE:
174 sacep = &sres->delegation.open_delegation4_u.write.permissions;
175 dacep = &dres->delegation.open_delegation4_u.write.permissions;
176 break;
177 }
178 dacep->who.utf8string_val =
179 kmem_alloc(sacep->who.utf8string_len, KM_SLEEP);
180 bcopy(sacep->who.utf8string_val, dacep->who.utf8string_val,
181 sacep->who.utf8string_len);
182 }
183
184 static void
185 deep_open_free(OPEN4res *res)
186 {
187 nfsace4 *acep;
188 if (res->status != NFS4_OK)
189 return;
190
191 switch (res->delegation.delegation_type) {
192 case OPEN_DELEGATE_NONE:
193 return;
194 case OPEN_DELEGATE_READ:
195 acep = &res->delegation.open_delegation4_u.read.permissions;
196 break;
197 case OPEN_DELEGATE_WRITE:
198 acep = &res->delegation.open_delegation4_u.write.permissions;
199 break;
200 }
201
202 if (acep->who.utf8string_val) {
203 kmem_free(acep->who.utf8string_val, acep->who.utf8string_len);
204 acep->who.utf8string_val = NULL;
205 }
206 }
207
208 void
209 rfs4_free_reply(nfs_resop4 *rp)
210 {
211 switch (rp->resop) {
212 case OP_LOCK:
213 deep_lock_free(&rp->nfs_resop4_u.oplock);
214 break;
215 case OP_OPEN:
216 deep_open_free(&rp->nfs_resop4_u.opopen);
217 default:
218 break;
219 }
220 }
221
222 void
223 rfs4_copy_reply(nfs_resop4 *dst, nfs_resop4 *src)
224 {
225 *dst = *src;
226
227 /* Handle responses that need deep copy */
228 switch (src->resop) {
229 case OP_LOCK:
230 deep_lock_copy(&dst->nfs_resop4_u.oplock,
231 &src->nfs_resop4_u.oplock);
232 break;
233 case OP_OPEN:
234 deep_open_copy(&dst->nfs_resop4_u.opopen,
235 &src->nfs_resop4_u.opopen);
236 break;
237 default:
238 break;
239 };
240 }
241
242 /*
243 * This is the implementation of the underlying state engine. The
244 * public interface to this engine is described by
245 * nfs4_state.h. Callers to the engine should hold no state engine
246 * locks when they call in to it. If the protocol needs to lock data
247 * structures it should do so after acquiring all references to them
248 * first and then follow the following lock order:
249 *
250 * client > openowner > state > lo_state > lockowner > file.
251 *
252 * Internally we only allow a thread to hold one hash bucket lock at a
253 * time and the lock is higher in the lock order (must be acquired
254 * first) than the data structure that is on that hash list.
255 *
256 * If a new reference was acquired by the caller, that reference needs
257 * to be released after releasing all acquired locks with the
258 * corresponding rfs4_*_rele routine.
259 */
260
261 /*
262 * This code is some what prototypical for now. Its purpose currently is to
263 * implement the interfaces sufficiently to finish the higher protocol
264 * elements. This will be replaced by a dynamically resizeable tables
265 * backed by kmem_cache allocator. However synchronization is handled
266 * correctly (I hope) and will not change by much. The mutexes for
267 * the hash buckets that can be used to create new instances of data
268 * structures might be good candidates to evolve into reader writer
269 * locks. If it has to do a creation, it would be holding the
270 * mutex across a kmem_alloc with KM_SLEEP specified.
271 */
272
273 #ifdef DEBUG
274 #define TABSIZE 17
275 #else
276 #define TABSIZE 2047
277 #endif
278
279 #define ADDRHASH(key) ((unsigned long)(key) >> 3)
280
281 #define MAXTABSZ 1024*1024
282
283 /* The values below are rfs4_lease_time units */
284
285 #ifdef DEBUG
286 #define CLIENT_CACHE_TIME 1
287 #define OPENOWNER_CACHE_TIME 1
288 #define STATE_CACHE_TIME 1
289 #define LO_STATE_CACHE_TIME 1
290 #define LOCKOWNER_CACHE_TIME 1
291 #define FILE_CACHE_TIME 3
292 #define DELEG_STATE_CACHE_TIME 1
293 #else
294 #define CLIENT_CACHE_TIME 10
295 #define OPENOWNER_CACHE_TIME 5
296 #define STATE_CACHE_TIME 1
297 #define LO_STATE_CACHE_TIME 1
298 #define LOCKOWNER_CACHE_TIME 3
299 #define FILE_CACHE_TIME 40
300 #define DELEG_STATE_CACHE_TIME 1
301 #endif
302
303 /*
304 * NFSv4 server state databases
305 *
306 * Initilized when the module is loaded and used by NFSv4 state tables.
307 * These kmem_cache databases are global, the tables that make use of these
308 * are per zone.
309 */
310 kmem_cache_t *rfs4_client_mem_cache;
311 kmem_cache_t *rfs4_clntIP_mem_cache;
312 kmem_cache_t *rfs4_openown_mem_cache;
313 kmem_cache_t *rfs4_openstID_mem_cache;
314 kmem_cache_t *rfs4_lockstID_mem_cache;
315 kmem_cache_t *rfs4_lockown_mem_cache;
316 kmem_cache_t *rfs4_file_mem_cache;
317 kmem_cache_t *rfs4_delegstID_mem_cache;
318
319 /*
320 * NFSv4 state table functions
321 */
322 static bool_t rfs4_client_create(rfs4_entry_t, void *);
323 static void rfs4_dss_remove_cpleaf(rfs4_client_t *);
324 static void rfs4_dss_remove_leaf(rfs4_servinst_t *, char *, char *);
325 static void rfs4_client_destroy(rfs4_entry_t);
326 static bool_t rfs4_client_expiry(rfs4_entry_t);
327 static uint32_t clientid_hash(void *);
328 static bool_t clientid_compare(rfs4_entry_t, void *);
329 static void *clientid_mkkey(rfs4_entry_t);
330 static uint32_t nfsclnt_hash(void *);
331 static bool_t nfsclnt_compare(rfs4_entry_t, void *);
332 static void *nfsclnt_mkkey(rfs4_entry_t);
333 static bool_t rfs4_clntip_expiry(rfs4_entry_t);
334 static void rfs4_clntip_destroy(rfs4_entry_t);
335 static bool_t rfs4_clntip_create(rfs4_entry_t, void *);
336 static uint32_t clntip_hash(void *);
337 static bool_t clntip_compare(rfs4_entry_t, void *);
338 static void *clntip_mkkey(rfs4_entry_t);
339 static bool_t rfs4_openowner_create(rfs4_entry_t, void *);
340 static void rfs4_openowner_destroy(rfs4_entry_t);
341 static bool_t rfs4_openowner_expiry(rfs4_entry_t);
342 static uint32_t openowner_hash(void *);
343 static bool_t openowner_compare(rfs4_entry_t, void *);
344 static void *openowner_mkkey(rfs4_entry_t);
345 static bool_t rfs4_state_create(rfs4_entry_t, void *);
346 static void rfs4_state_destroy(rfs4_entry_t);
347 static bool_t rfs4_state_expiry(rfs4_entry_t);
348 static uint32_t state_hash(void *);
349 static bool_t state_compare(rfs4_entry_t, void *);
350 static void *state_mkkey(rfs4_entry_t);
351 static uint32_t state_owner_file_hash(void *);
352 static bool_t state_owner_file_compare(rfs4_entry_t, void *);
353 static void *state_owner_file_mkkey(rfs4_entry_t);
354 static uint32_t state_file_hash(void *);
355 static bool_t state_file_compare(rfs4_entry_t, void *);
356 static void *state_file_mkkey(rfs4_entry_t);
357 static bool_t rfs4_lo_state_create(rfs4_entry_t, void *);
358 static void rfs4_lo_state_destroy(rfs4_entry_t);
359 static bool_t rfs4_lo_state_expiry(rfs4_entry_t);
360 static uint32_t lo_state_hash(void *);
361 static bool_t lo_state_compare(rfs4_entry_t, void *);
362 static void *lo_state_mkkey(rfs4_entry_t);
363 static uint32_t lo_state_lo_hash(void *);
364 static bool_t lo_state_lo_compare(rfs4_entry_t, void *);
365 static void *lo_state_lo_mkkey(rfs4_entry_t);
366 static bool_t rfs4_lockowner_create(rfs4_entry_t, void *);
367 static void rfs4_lockowner_destroy(rfs4_entry_t);
368 static bool_t rfs4_lockowner_expiry(rfs4_entry_t);
369 static uint32_t lockowner_hash(void *);
370 static bool_t lockowner_compare(rfs4_entry_t, void *);
371 static void *lockowner_mkkey(rfs4_entry_t);
372 static uint32_t pid_hash(void *);
373 static bool_t pid_compare(rfs4_entry_t, void *);
374 static void *pid_mkkey(rfs4_entry_t);
375 static bool_t rfs4_file_create(rfs4_entry_t, void *);
376 static void rfs4_file_destroy(rfs4_entry_t);
377 static uint32_t file_hash(void *);
378 static bool_t file_compare(rfs4_entry_t, void *);
379 static void *file_mkkey(rfs4_entry_t);
380 static bool_t rfs4_deleg_state_create(rfs4_entry_t, void *);
381 static void rfs4_deleg_state_destroy(rfs4_entry_t);
382 static bool_t rfs4_deleg_state_expiry(rfs4_entry_t);
383 static uint32_t deleg_hash(void *);
384 static bool_t deleg_compare(rfs4_entry_t, void *);
385 static void *deleg_mkkey(rfs4_entry_t);
386 static uint32_t deleg_state_hash(void *);
387 static bool_t deleg_state_compare(rfs4_entry_t, void *);
388 static void *deleg_state_mkkey(rfs4_entry_t);
389
390 static void rfs4_state_rele_nounlock(rfs4_state_t *);
391
392 static int rfs4_ss_enabled = 0;
393
394 extern void (*rfs4_client_clrst)(struct nfs4clrst_args *);
395
396 void
397 rfs4_ss_pnfree(rfs4_ss_pn_t *ss_pn)
398 {
399 kmem_free(ss_pn, sizeof (rfs4_ss_pn_t));
400 }
401
402 static rfs4_ss_pn_t *
403 rfs4_ss_pnalloc(char *dir, char *leaf)
404 {
405 rfs4_ss_pn_t *ss_pn;
406 int dir_len, leaf_len;
407
408 /*
409 * validate we have a resonable path
410 * (account for the '/' and trailing null)
411 */
412 if ((dir_len = strlen(dir)) > MAXPATHLEN ||
413 (leaf_len = strlen(leaf)) > MAXNAMELEN ||
414 (dir_len + leaf_len + 2) > MAXPATHLEN) {
415 return (NULL);
416 }
417
418 ss_pn = kmem_alloc(sizeof (rfs4_ss_pn_t), KM_SLEEP);
419
420 (void) snprintf(ss_pn->pn, MAXPATHLEN, "%s/%s", dir, leaf);
421 /* Handy pointer to just the leaf name */
422 ss_pn->leaf = ss_pn->pn + dir_len + 1;
423 return (ss_pn);
424 }
425
426
427 /*
428 * Move the "leaf" filename from "sdir" directory
429 * to the "ddir" directory. Return the pathname of
430 * the destination unless the rename fails in which
431 * case we need to return the source pathname.
432 */
433 static rfs4_ss_pn_t *
434 rfs4_ss_movestate(char *sdir, char *ddir, char *leaf)
435 {
436 rfs4_ss_pn_t *src, *dst;
437
438 if ((src = rfs4_ss_pnalloc(sdir, leaf)) == NULL)
439 return (NULL);
440
441 if ((dst = rfs4_ss_pnalloc(ddir, leaf)) == NULL) {
442 rfs4_ss_pnfree(src);
443 return (NULL);
444 }
445
446 /*
447 * If the rename fails we shall return the src
448 * pathname and free the dst. Otherwise we need
449 * to free the src and return the dst pathanme.
450 */
451 if (vn_rename(src->pn, dst->pn, UIO_SYSSPACE)) {
452 rfs4_ss_pnfree(dst);
453 return (src);
454 }
455 rfs4_ss_pnfree(src);
456 return (dst);
457 }
458
459
460 static rfs4_oldstate_t *
461 rfs4_ss_getstate(vnode_t *dvp, rfs4_ss_pn_t *ss_pn)
462 {
463 struct uio uio;
464 struct iovec iov[3];
465
466 rfs4_oldstate_t *cl_ss = NULL;
467 vnode_t *vp;
468 vattr_t va;
469 uint_t id_len;
470 int err, kill_file, file_vers;
471
472 if (ss_pn == NULL)
473 return (NULL);
474
475 /*
476 * open the state file.
477 */
478 if (vn_open(ss_pn->pn, UIO_SYSSPACE, FREAD, 0, &vp, 0, 0) != 0) {
479 return (NULL);
480 }
481
482 if (vp->v_type != VREG) {
483 (void) VOP_CLOSE(vp, FREAD, 1, (offset_t)0, CRED(), NULL);
484 VN_RELE(vp);
485 return (NULL);
486 }
487
488 err = VOP_ACCESS(vp, VREAD, 0, CRED(), NULL);
489 if (err) {
490 /*
491 * We don't have read access? better get the heck out.
492 */
493 (void) VOP_CLOSE(vp, FREAD, 1, (offset_t)0, CRED(), NULL);
494 VN_RELE(vp);
495 return (NULL);
496 }
497
498 (void) VOP_RWLOCK(vp, V_WRITELOCK_FALSE, NULL);
499 /*
500 * get the file size to do some basic validation
501 */
502 va.va_mask = AT_SIZE;
503 err = VOP_GETATTR(vp, &va, 0, CRED(), NULL);
504
505 kill_file = (va.va_size == 0 || va.va_size <
506 (NFS4_VERIFIER_SIZE + sizeof (uint_t)+1));
507
508 if (err || kill_file) {
509 VOP_RWUNLOCK(vp, V_WRITELOCK_FALSE, NULL);
510 (void) VOP_CLOSE(vp, FREAD, 1, (offset_t)0, CRED(), NULL);
511 VN_RELE(vp);
512 if (kill_file) {
513 (void) VOP_REMOVE(dvp, ss_pn->leaf, CRED(), NULL, 0);
514 }
515 return (NULL);
516 }
517
518 cl_ss = kmem_alloc(sizeof (rfs4_oldstate_t), KM_SLEEP);
519
520 /*
521 * build iovecs to read in the file_version, verifier and id_len
522 */
523 iov[0].iov_base = (caddr_t)&file_vers;
524 iov[0].iov_len = sizeof (int);
525 iov[1].iov_base = (caddr_t)&cl_ss->cl_id4.verifier;
526 iov[1].iov_len = NFS4_VERIFIER_SIZE;
527 iov[2].iov_base = (caddr_t)&id_len;
528 iov[2].iov_len = sizeof (uint_t);
529
530 uio.uio_iov = iov;
531 uio.uio_iovcnt = 3;
532 uio.uio_segflg = UIO_SYSSPACE;
533 uio.uio_loffset = 0;
534 uio.uio_resid = sizeof (int) + NFS4_VERIFIER_SIZE + sizeof (uint_t);
535
536 if (err = VOP_READ(vp, &uio, FREAD, CRED(), NULL)) {
537 VOP_RWUNLOCK(vp, V_WRITELOCK_FALSE, NULL);
538 (void) VOP_CLOSE(vp, FREAD, 1, (offset_t)0, CRED(), NULL);
539 VN_RELE(vp);
540 kmem_free(cl_ss, sizeof (rfs4_oldstate_t));
541 return (NULL);
542 }
543
544 /*
545 * if the file_version doesn't match or if the
546 * id_len is zero or the combination of the verifier,
547 * id_len and id_val is bigger than the file we have
548 * a problem. If so ditch the file.
549 */
550 kill_file = (file_vers != NFS4_SS_VERSION || id_len == 0 ||
551 (id_len + NFS4_VERIFIER_SIZE + sizeof (uint_t)) > va.va_size);
552
553 if (err || kill_file) {
554 VOP_RWUNLOCK(vp, V_WRITELOCK_FALSE, NULL);
555 (void) VOP_CLOSE(vp, FREAD, 1, (offset_t)0, CRED(), NULL);
556 VN_RELE(vp);
557 kmem_free(cl_ss, sizeof (rfs4_oldstate_t));
558 if (kill_file) {
559 (void) VOP_REMOVE(dvp, ss_pn->leaf, CRED(), NULL, 0);
560 }
561 return (NULL);
562 }
563
564 /*
565 * now get the client id value
566 */
567 cl_ss->cl_id4.id_val = kmem_alloc(id_len, KM_SLEEP);
568 iov[0].iov_base = cl_ss->cl_id4.id_val;
569 iov[0].iov_len = id_len;
570
571 uio.uio_iov = iov;
572 uio.uio_iovcnt = 1;
573 uio.uio_segflg = UIO_SYSSPACE;
574 uio.uio_resid = cl_ss->cl_id4.id_len = id_len;
575
576 if (err = VOP_READ(vp, &uio, FREAD, CRED(), NULL)) {
577 VOP_RWUNLOCK(vp, V_WRITELOCK_FALSE, NULL);
578 (void) VOP_CLOSE(vp, FREAD, 1, (offset_t)0, CRED(), NULL);
579 VN_RELE(vp);
580 kmem_free(cl_ss->cl_id4.id_val, id_len);
581 kmem_free(cl_ss, sizeof (rfs4_oldstate_t));
582 return (NULL);
583 }
584
585 VOP_RWUNLOCK(vp, V_WRITELOCK_FALSE, NULL);
586 (void) VOP_CLOSE(vp, FREAD, 1, (offset_t)0, CRED(), NULL);
587 VN_RELE(vp);
588 return (cl_ss);
589 }
590
591 #ifdef nextdp
592 #undef nextdp
593 #endif
594 #define nextdp(dp) ((struct dirent64 *)((char *)(dp) + (dp)->d_reclen))
595
596 /*
597 * Add entries from statedir to supplied oldstate list.
598 * Optionally, move all entries from statedir -> destdir.
599 */
600 void
601 rfs4_ss_oldstate(rfs4_oldstate_t *oldstate, char *statedir, char *destdir)
602 {
603 rfs4_ss_pn_t *ss_pn;
604 rfs4_oldstate_t *cl_ss = NULL;
605 char *dirt = NULL;
606 int err, dir_eof = 0, size = 0;
607 vnode_t *dvp;
608 struct iovec iov;
609 struct uio uio;
610 struct dirent64 *dep;
611 offset_t dirchunk_offset = 0;
612
613 /*
614 * open the state directory
615 */
616 if (vn_open(statedir, UIO_SYSSPACE, FREAD, 0, &dvp, 0, 0))
617 return;
618
619 if (dvp->v_type != VDIR || VOP_ACCESS(dvp, VREAD, 0, CRED(), NULL))
620 goto out;
621
622 dirt = kmem_alloc(RFS4_SS_DIRSIZE, KM_SLEEP);
623
624 /*
625 * Get and process the directory entries
626 */
627 while (!dir_eof) {
628 (void) VOP_RWLOCK(dvp, V_WRITELOCK_FALSE, NULL);
629 iov.iov_base = dirt;
630 iov.iov_len = RFS4_SS_DIRSIZE;
631 uio.uio_iov = &iov;
632 uio.uio_iovcnt = 1;
633 uio.uio_segflg = UIO_SYSSPACE;
634 uio.uio_loffset = dirchunk_offset;
635 uio.uio_resid = RFS4_SS_DIRSIZE;
636
637 err = VOP_READDIR(dvp, &uio, CRED(), &dir_eof, NULL, 0);
638 VOP_RWUNLOCK(dvp, V_WRITELOCK_FALSE, NULL);
639 if (err)
640 goto out;
641
642 size = RFS4_SS_DIRSIZE - uio.uio_resid;
643
644 /*
645 * Process all the directory entries in this
646 * readdir chunk
647 */
648 for (dep = (struct dirent64 *)dirt; size > 0;
649 dep = nextdp(dep)) {
650
651 size -= dep->d_reclen;
652 dirchunk_offset = dep->d_off;
653
654 /*
655 * Skip '.' and '..'
656 */
657 if (NFS_IS_DOTNAME(dep->d_name))
658 continue;
659
660 ss_pn = rfs4_ss_pnalloc(statedir, dep->d_name);
661 if (ss_pn == NULL)
662 continue;
663
664 if (cl_ss = rfs4_ss_getstate(dvp, ss_pn)) {
665 if (destdir != NULL) {
666 rfs4_ss_pnfree(ss_pn);
667 cl_ss->ss_pn = rfs4_ss_movestate(
668 statedir, destdir, dep->d_name);
669 } else {
670 cl_ss->ss_pn = ss_pn;
671 }
672 insque(cl_ss, oldstate);
673 } else {
674 rfs4_ss_pnfree(ss_pn);
675 }
676 }
677 }
678
679 out:
680 (void) VOP_CLOSE(dvp, FREAD, 1, (offset_t)0, CRED(), NULL);
681 VN_RELE(dvp);
682 if (dirt)
683 kmem_free((caddr_t)dirt, RFS4_SS_DIRSIZE);
684 }
685
686 static void
687 rfs4_ss_init(nfs4_srv_t *nsrv4)
688 {
689 int npaths = 1;
690 char *default_dss_path = NFS4_DSS_VAR_DIR;
691
692 /* read the default stable storage state */
693 rfs4_dss_readstate(nsrv4, npaths, &default_dss_path);
694
695 rfs4_ss_enabled = 1;
696 }
697
698 static void
699 rfs4_ss_fini(nfs4_srv_t *nsrv4)
700 {
701 rfs4_servinst_t *sip;
702
703 mutex_enter(&nsrv4->servinst_lock);
704 sip = nsrv4->nfs4_cur_servinst;
705 while (sip != NULL) {
706 rfs4_dss_clear_oldstate(sip);
707 sip = sip->next;
708 }
709 mutex_exit(&nsrv4->servinst_lock);
710 }
711
712 /*
713 * Remove all oldstate files referenced by this servinst.
714 */
715 static void
716 rfs4_dss_clear_oldstate(rfs4_servinst_t *sip)
717 {
718 rfs4_oldstate_t *os_head, *osp;
719
720 rw_enter(&sip->oldstate_lock, RW_WRITER);
721 os_head = sip->oldstate;
722
723 if (os_head == NULL) {
724 rw_exit(&sip->oldstate_lock);
725 return;
726 }
727
728 /* skip dummy entry */
729 osp = os_head->next;
730 while (osp != os_head) {
731 char *leaf = osp->ss_pn->leaf;
732 rfs4_oldstate_t *os_next;
733
734 rfs4_dss_remove_leaf(sip, NFS4_DSS_OLDSTATE_LEAF, leaf);
735
736 if (osp->cl_id4.id_val)
737 kmem_free(osp->cl_id4.id_val, osp->cl_id4.id_len);
738 rfs4_ss_pnfree(osp->ss_pn);
739
740 os_next = osp->next;
741 remque(osp);
742 kmem_free(osp, sizeof (rfs4_oldstate_t));
743 osp = os_next;
744 }
745
746 rw_exit(&sip->oldstate_lock);
747 }
748
749 /*
750 * Form the state and oldstate paths, and read in the stable storage files.
751 */
752 void
753 rfs4_dss_readstate(nfs4_srv_t *nsrv4, int npaths, char **paths)
754 {
755 int i;
756 char *state, *oldstate;
757
758 state = kmem_alloc(MAXPATHLEN, KM_SLEEP);
759 oldstate = kmem_alloc(MAXPATHLEN, KM_SLEEP);
760
761 for (i = 0; i < npaths; i++) {
762 char *path = paths[i];
763
764 (void) sprintf(state, "%s/%s", path, NFS4_DSS_STATE_LEAF);
765 (void) sprintf(oldstate, "%s/%s", path, NFS4_DSS_OLDSTATE_LEAF);
766
767 /*
768 * Populate the current server instance's oldstate list.
769 *
770 * 1. Read stable storage data from old state directory,
771 * leaving its contents alone.
772 *
773 * 2. Read stable storage data from state directory,
774 * and move the latter's contents to old state
775 * directory.
776 */
777 /* CSTYLED */
778 rfs4_ss_oldstate(nsrv4->nfs4_cur_servinst->oldstate, oldstate, NULL);
779 /* CSTYLED */
780 rfs4_ss_oldstate(nsrv4->nfs4_cur_servinst->oldstate, state, oldstate);
781 }
782
783 kmem_free(state, MAXPATHLEN);
784 kmem_free(oldstate, MAXPATHLEN);
785 }
786
787
788 /*
789 * Check if we are still in grace and if the client can be
790 * granted permission to perform reclaims.
791 */
792 void
793 rfs4_ss_chkclid(nfs4_srv_t *nsrv4, rfs4_client_t *cp)
794 {
795 rfs4_servinst_t *sip;
796
797 /*
798 * It should be sufficient to check the oldstate data for just
799 * this client's instance. However, since our per-instance
800 * client grouping is solely temporal, HA-NFSv4 RG failover
801 * might result in clients of the same RG being partitioned into
802 * separate instances.
803 *
804 * Until the client grouping is improved, we must check the
805 * oldstate data for all instances with an active grace period.
806 *
807 * This also serves as the mechanism to remove stale oldstate data.
808 * The first time we check an instance after its grace period has
809 * expired, the oldstate data should be cleared.
810 *
811 * Start at the current instance, and walk the list backwards
812 * to the first.
813 */
814 mutex_enter(&nsrv4->servinst_lock);
815 for (sip = nsrv4->nfs4_cur_servinst; sip != NULL; sip = sip->prev) {
816 rfs4_ss_chkclid_sip(cp, sip);
817
818 /* if the above check found this client, we're done */
819 if (cp->rc_can_reclaim)
820 break;
821 }
822 mutex_exit(&nsrv4->servinst_lock);
823 }
824
825 static void
826 rfs4_ss_chkclid_sip(rfs4_client_t *cp, rfs4_servinst_t *sip)
827 {
828 rfs4_oldstate_t *osp, *os_head;
829
830 /* short circuit everything if this server instance has no oldstate */
831 rw_enter(&sip->oldstate_lock, RW_READER);
832 os_head = sip->oldstate;
833 rw_exit(&sip->oldstate_lock);
834 if (os_head == NULL)
835 return;
836
837 /*
838 * If this server instance is no longer in a grace period then
839 * the client won't be able to reclaim. No further need for this
840 * instance's oldstate data, so it can be cleared.
841 */
842 if (!rfs4_servinst_in_grace(sip))
843 return;
844
845 /* this instance is still in grace; search for the clientid */
846
847 rw_enter(&sip->oldstate_lock, RW_READER);
848
849 os_head = sip->oldstate;
850 /* skip dummy entry */
851 osp = os_head->next;
852 while (osp != os_head) {
853 if (osp->cl_id4.id_len == cp->rc_nfs_client.id_len) {
854 if (bcmp(osp->cl_id4.id_val, cp->rc_nfs_client.id_val,
855 osp->cl_id4.id_len) == 0) {
856 cp->rc_can_reclaim = 1;
857 break;
858 }
859 }
860 osp = osp->next;
861 }
862
863 rw_exit(&sip->oldstate_lock);
864 }
865
866 /*
867 * Place client information into stable storage: 1/3.
868 * First, generate the leaf filename, from the client's IP address and
869 * the server-generated short-hand clientid.
870 */
871 void
872 rfs4_ss_clid(nfs4_srv_t *nsrv4, rfs4_client_t *cp)
873 {
874 const char *kinet_ntop6(uchar_t *, char *, size_t);
875 char leaf[MAXNAMELEN], buf[INET6_ADDRSTRLEN];
876 struct sockaddr *ca;
877 uchar_t *b;
878
879 if (rfs4_ss_enabled == 0) {
880 return;
881 }
882
883 buf[0] = 0;
884
885 ca = (struct sockaddr *)&cp->rc_addr;
886
887 /*
888 * Convert the caller's IP address to a dotted string
889 */
890 if (ca->sa_family == AF_INET) {
891 b = (uchar_t *)&((struct sockaddr_in *)ca)->sin_addr;
892 (void) sprintf(buf, "%03d.%03d.%03d.%03d", b[0] & 0xFF,
893 b[1] & 0xFF, b[2] & 0xFF, b[3] & 0xFF);
894 } else if (ca->sa_family == AF_INET6) {
895 struct sockaddr_in6 *sin6;
896
897 sin6 = (struct sockaddr_in6 *)ca;
898 (void) kinet_ntop6((uchar_t *)&sin6->sin6_addr,
899 buf, INET6_ADDRSTRLEN);
900 }
901
902 (void) snprintf(leaf, MAXNAMELEN, "%s-%llx", buf,
903 (longlong_t)cp->rc_clientid);
904 rfs4_ss_clid_write(nsrv4, cp, leaf);
905 }
906
907 /*
908 * Place client information into stable storage: 2/3.
909 * DSS: distributed stable storage: the file may need to be written to
910 * multiple directories.
911 */
912 static void
913 rfs4_ss_clid_write(nfs4_srv_t *nsrv4, rfs4_client_t *cp, char *leaf)
914 {
915 rfs4_servinst_t *sip;
916
917 /*
918 * It should be sufficient to write the leaf file to (all) DSS paths
919 * associated with just this client's instance. However, since our
920 * per-instance client grouping is solely temporal, HA-NFSv4 RG
921 * failover might result in us losing DSS data.
922 *
923 * Until the client grouping is improved, we must write the DSS data
924 * to all instances' paths. Start at the current instance, and
925 * walk the list backwards to the first.
926 */
927 mutex_enter(&nsrv4->servinst_lock);
928 for (sip = nsrv4->nfs4_cur_servinst; sip != NULL; sip = sip->prev) {
929 int i, npaths = sip->dss_npaths;
930
931 /* write the leaf file to all DSS paths */
932 for (i = 0; i < npaths; i++) {
933 rfs4_dss_path_t *dss_path = sip->dss_paths[i];
934
935 /* HA-NFSv4 path might have been failed-away from us */
936 if (dss_path == NULL)
937 continue;
938
939 rfs4_ss_clid_write_one(cp, dss_path->path, leaf);
940 }
941 }
942 mutex_exit(&nsrv4->servinst_lock);
943 }
944
945 /*
946 * Place client information into stable storage: 3/3.
947 * Write the stable storage data to the requested file.
948 */
949 static void
950 rfs4_ss_clid_write_one(rfs4_client_t *cp, char *dss_path, char *leaf)
951 {
952 int ioflag;
953 int file_vers = NFS4_SS_VERSION;
954 size_t dirlen;
955 struct uio uio;
956 struct iovec iov[4];
957 char *dir;
958 rfs4_ss_pn_t *ss_pn;
959 vnode_t *vp;
960 nfs_client_id4 *cl_id4 = &(cp->rc_nfs_client);
961
962 /* allow 2 extra bytes for '/' & NUL */
963 dirlen = strlen(dss_path) + strlen(NFS4_DSS_STATE_LEAF) + 2;
964 dir = kmem_alloc(dirlen, KM_SLEEP);
965 (void) sprintf(dir, "%s/%s", dss_path, NFS4_DSS_STATE_LEAF);
966
967 ss_pn = rfs4_ss_pnalloc(dir, leaf);
968 /* rfs4_ss_pnalloc takes its own copy */
969 kmem_free(dir, dirlen);
970 if (ss_pn == NULL)
971 return;
972
973 if (vn_open(ss_pn->pn, UIO_SYSSPACE, FCREAT|FWRITE, 0600, &vp,
974 CRCREAT, 0)) {
975 rfs4_ss_pnfree(ss_pn);
976 return;
977 }
978
979 /*
980 * We need to record leaf - i.e. the filename - so that we know
981 * what to remove, in the future. However, the dir part of cp->ss_pn
982 * should never be referenced directly, since it's potentially only
983 * one of several paths with this leaf in it.
984 */
985 if (cp->rc_ss_pn != NULL) {
986 if (strcmp(cp->rc_ss_pn->leaf, leaf) == 0) {
987 /* we've already recorded *this* leaf */
988 rfs4_ss_pnfree(ss_pn);
989 } else {
990 /* replace with this leaf */
991 rfs4_ss_pnfree(cp->rc_ss_pn);
992 cp->rc_ss_pn = ss_pn;
993 }
994 } else {
995 cp->rc_ss_pn = ss_pn;
996 }
997
998 /*
999 * Build a scatter list that points to the nfs_client_id4
1000 */
1001 iov[0].iov_base = (caddr_t)&file_vers;
1002 iov[0].iov_len = sizeof (int);
1003 iov[1].iov_base = (caddr_t)&(cl_id4->verifier);
1004 iov[1].iov_len = NFS4_VERIFIER_SIZE;
1005 iov[2].iov_base = (caddr_t)&(cl_id4->id_len);
1006 iov[2].iov_len = sizeof (uint_t);
1007 iov[3].iov_base = (caddr_t)cl_id4->id_val;
1008 iov[3].iov_len = cl_id4->id_len;
1009
1010 uio.uio_iov = iov;
1011 uio.uio_iovcnt = 4;
1012 uio.uio_loffset = 0;
1013 uio.uio_segflg = UIO_SYSSPACE;
1014 uio.uio_llimit = (rlim64_t)MAXOFFSET_T;
1015 uio.uio_resid = cl_id4->id_len + sizeof (int) +
1016 NFS4_VERIFIER_SIZE + sizeof (uint_t);
1017
1018 ioflag = uio.uio_fmode = (FWRITE|FSYNC);
1019 uio.uio_extflg = UIO_COPY_DEFAULT;
1020
1021 (void) VOP_RWLOCK(vp, V_WRITELOCK_TRUE, NULL);
1022 /* write the full client id to the file. */
1023 (void) VOP_WRITE(vp, &uio, ioflag, CRED(), NULL);
1024 VOP_RWUNLOCK(vp, V_WRITELOCK_TRUE, NULL);
1025
1026 (void) VOP_CLOSE(vp, FWRITE, 1, (offset_t)0, CRED(), NULL);
1027 VN_RELE(vp);
1028 }
1029
1030 /*
1031 * DSS: distributed stable storage.
1032 * Unpack the list of paths passed by nfsd.
1033 * Use nvlist_alloc(9F) to manage the data.
1034 * The caller is responsible for allocating and freeing the buffer.
1035 */
1036 int
1037 rfs4_dss_setpaths(char *buf, size_t buflen)
1038 {
1039 int error;
1040
1041 /*
1042 * If this is a "warm start", i.e. we previously had DSS paths,
1043 * preserve the old paths.
1044 */
1045 if (rfs4_dss_paths != NULL) {
1046 /*
1047 * Before we lose the ptr, destroy the nvlist and pathnames
1048 * array from the warm start before this one.
1049 */
1050 nvlist_free(rfs4_dss_oldpaths);
1051 rfs4_dss_oldpaths = rfs4_dss_paths;
1052 }
1053
1054 /* unpack the buffer into a searchable nvlist */
1055 error = nvlist_unpack(buf, buflen, &rfs4_dss_paths, KM_SLEEP);
1056 if (error)
1057 return (error);
1058
1059 /*
1060 * Search the nvlist for the pathnames nvpair (which is the only nvpair
1061 * in the list, and record its location.
1062 */
1063 error = nvlist_lookup_string_array(rfs4_dss_paths, NFS4_DSS_NVPAIR_NAME,
1064 &rfs4_dss_newpaths, &rfs4_dss_numnewpaths);
1065 return (error);
1066 }
1067
1068 /*
1069 * Ultimately the nfssys() call NFS4_CLR_STATE endsup here
1070 * to find and mark the client for forced expire.
1071 */
1072 static void
1073 rfs4_client_scrub(rfs4_entry_t ent, void *arg)
1074 {
1075 rfs4_client_t *cp = (rfs4_client_t *)ent;
1076 struct nfs4clrst_args *clr = arg;
1077 struct sockaddr_in6 *ent_sin6;
1078 struct in6_addr clr_in6;
1079 struct sockaddr_in *ent_sin;
1080 struct in_addr clr_in;
1081
1082 if (clr->addr_type != cp->rc_addr.ss_family) {
1083 return;
1084 }
1085
1086 switch (clr->addr_type) {
1087
1088 case AF_INET6:
1089 /* copyin the address from user space */
1090 if (copyin(clr->ap, &clr_in6, sizeof (clr_in6))) {
1091 break;
1092 }
1093
1094 ent_sin6 = (struct sockaddr_in6 *)&cp->rc_addr;
1095
1096 /*
1097 * now compare, and if equivalent mark entry
1098 * for forced expiration
1099 */
1100 if (IN6_ARE_ADDR_EQUAL(&ent_sin6->sin6_addr, &clr_in6)) {
1101 cp->rc_forced_expire = 1;
1102 }
1103 break;
1104
1105 case AF_INET:
1106 /* copyin the address from user space */
1107 if (copyin(clr->ap, &clr_in, sizeof (clr_in))) {
1108 break;
1109 }
1110
1111 ent_sin = (struct sockaddr_in *)&cp->rc_addr;
1112
1113 /*
1114 * now compare, and if equivalent mark entry
1115 * for forced expiration
1116 */
1117 if (ent_sin->sin_addr.s_addr == clr_in.s_addr) {
1118 cp->rc_forced_expire = 1;
1119 }
1120 break;
1121
1122 default:
1123 /* force this assert to fail */
1124 ASSERT(clr->addr_type != clr->addr_type);
1125 }
1126 }
1127
1128 /*
1129 * This is called from nfssys() in order to clear server state
1130 * for the specified client IP Address.
1131 */
1132 void
1133 rfs4_clear_client_state(struct nfs4clrst_args *clr)
1134 {
1135 nfs4_srv_t *nsrv4;
1136 nsrv4 = nfs4_get_srv();
1137 (void) rfs4_dbe_walk(nsrv4->rfs4_client_tab, rfs4_client_scrub, clr);
1138 }
1139
1140 /*
1141 * Used to initialize the NFSv4 server's state or database. All of
1142 * the tables are created and timers are set.
1143 */
1144 void
1145 rfs4_state_g_init()
1146 {
1147 extern boolean_t rfs4_cpr_callb(void *, int);
1148 /*
1149 * Add a CPR callback so that we can update client
1150 * access times to extend the lease after a suspend
1151 * and resume (using the same class as rpcmod/connmgr)
1152 */
1153 cpr_id = callb_add(rfs4_cpr_callb, 0, CB_CL_CPR_RPC, "rfs4");
1154
1155 /*
1156 * NFSv4 server state databases
1157 *
1158 * Initialized when the module is loaded and used by NFSv4 state
1159 * tables. These kmem_cache free pools are used globally, the NFSv4
1160 * state tables which make use of these kmem_cache free pools are per
1161 * zone.
1162 *
1163 * initialize the global kmem_cache free pools which will be used by
1164 * the NFSv4 state tables.
1165 */
1166 /* CSTYLED */
1167 rfs4_client_mem_cache = nfs4_init_mem_cache("Client_entry_cache", 2, sizeof (rfs4_client_t), 0);
1168 /* CSTYLED */
1169 rfs4_clntIP_mem_cache = nfs4_init_mem_cache("ClntIP_entry_cache", 1, sizeof (rfs4_clntip_t), 1);
1170 /* CSTYLED */
1171 rfs4_openown_mem_cache = nfs4_init_mem_cache("OpenOwner_entry_cache", 1, sizeof (rfs4_openowner_t), 2);
1172 /* CSTYLED */
1173 rfs4_openstID_mem_cache = nfs4_init_mem_cache("OpenStateID_entry_cache", 3, sizeof (rfs4_state_t), 3);
1174 /* CSTYLED */
1175 rfs4_lockstID_mem_cache = nfs4_init_mem_cache("LockStateID_entry_cache", 3, sizeof (rfs4_lo_state_t), 4);
1176 /* CSTYLED */
1177 rfs4_lockown_mem_cache = nfs4_init_mem_cache("Lockowner_entry_cache", 2, sizeof (rfs4_lockowner_t), 5);
1178 /* CSTYLED */
1179 rfs4_file_mem_cache = nfs4_init_mem_cache("File_entry_cache", 1, sizeof (rfs4_file_t), 6);
1180 /* CSTYLED */
1181 rfs4_delegstID_mem_cache = nfs4_init_mem_cache("DelegStateID_entry_cache", 2, sizeof (rfs4_deleg_state_t), 7);
1182
1183 rfs4_client_clrst = rfs4_clear_client_state;
1184 }
1185
1186
1187 /*
1188 * Used at server shutdown to cleanup all of the NFSv4 server's structures
1189 * and other state.
1190 */
1191 void
1192 rfs4_state_g_fini()
1193 {
1194 int i;
1195 /*
1196 * Cleanup the CPR callback.
1197 */
1198 if (cpr_id)
1199 (void) callb_delete(cpr_id);
1200
1201 rfs4_client_clrst = NULL;
1202
1203 /* free the NFSv4 state databases */
1204 for (i = 0; i < RFS4_DB_MEM_CACHE_NUM; i++) {
1205 kmem_cache_destroy(rfs4_db_mem_cache_table[i].r_db_mem_cache);
1206 rfs4_db_mem_cache_table[i].r_db_mem_cache = NULL;
1207 }
1208
1209 rfs4_client_mem_cache = NULL;
1210 rfs4_clntIP_mem_cache = NULL;
1211 rfs4_openown_mem_cache = NULL;
1212 rfs4_openstID_mem_cache = NULL;
1213 rfs4_lockstID_mem_cache = NULL;
1214 rfs4_lockown_mem_cache = NULL;
1215 rfs4_file_mem_cache = NULL;
1216 rfs4_delegstID_mem_cache = NULL;
1217
1218 /* DSS: distributed stable storage */
1219 nvlist_free(rfs4_dss_oldpaths);
1220 nvlist_free(rfs4_dss_paths);
1221 rfs4_dss_paths = rfs4_dss_oldpaths = NULL;
1222 }
1223
1224 /*
1225 * Used to initialize the per zone NFSv4 server's state
1226 */
1227 void
1228 rfs4_state_zone_init(nfs4_srv_t *nsrv4)
1229 {
1230 time_t start_time;
1231 int start_grace;
1232 char *dss_path = NFS4_DSS_VAR_DIR;
1233
1234 /* DSS: distributed stable storage: initialise served paths list */
1235 nsrv4->dss_pathlist = NULL;
1236
1237 /*
1238 * Set the boot time. If the server
1239 * has been restarted quickly and has had the opportunity to
1240 * service clients, then the start_time needs to be bumped
1241 * regardless. A small window but it exists...
1242 */
1243 start_time = gethrestime_sec();
1244 if (nsrv4->rfs4_start_time < start_time)
1245 nsrv4->rfs4_start_time = start_time;
1246 else
1247 nsrv4->rfs4_start_time++;
1248
1249 /*
1250 * Create the first server instance, or a new one if the server has
1251 * been restarted; see above comments on rfs4_start_time. Don't
1252 * start its grace period; that will be done later, to maximise the
1253 * clients' recovery window.
1254 */
1255 start_grace = 0;
1256 if (curzone == global_zone && rfs4_dss_numnewpaths > 0) {
1257 int i;
1258 char **dss_allpaths = NULL;
1259 dss_allpaths = kmem_alloc(sizeof (char *) *
1260 (rfs4_dss_numnewpaths + 1), KM_SLEEP);
1261 /*
1262 * Add the default path into the list of paths for saving
1263 * state informantion.
1264 */
1265 dss_allpaths[0] = dss_path;
1266 for (i = 0; i < rfs4_dss_numnewpaths; i++) {
1267 dss_allpaths[i + 1] = rfs4_dss_newpaths[i];
1268 }
1269 rfs4_servinst_create(nsrv4, start_grace,
1270 (rfs4_dss_numnewpaths + 1), dss_allpaths);
1271 kmem_free(dss_allpaths,
1272 (sizeof (char *) * (rfs4_dss_numnewpaths + 1)));
1273 } else {
1274 rfs4_servinst_create(nsrv4, start_grace, 1, &dss_path);
1275 }
1276
1277 /* reset the "first NFSv4 request" status */
1278 nsrv4->seen_first_compound = 0;
1279
1280 mutex_enter(&nsrv4->state_lock);
1281
1282 /*
1283 * If the server state database has already been initialized,
1284 * skip it
1285 */
1286 if (nsrv4->nfs4_server_state != NULL) {
1287 mutex_exit(&nsrv4->state_lock);
1288 return;
1289 }
1290
1291 rw_init(&nsrv4->rfs4_findclient_lock, NULL, RW_DEFAULT, NULL);
1292
1293 /* set the various cache timers for table creation */
1294 if (nsrv4->rfs4_client_cache_time == 0)
1295 nsrv4->rfs4_client_cache_time = CLIENT_CACHE_TIME;
1296 if (nsrv4->rfs4_openowner_cache_time == 0)
1297 nsrv4->rfs4_openowner_cache_time = OPENOWNER_CACHE_TIME;
1298 if (nsrv4->rfs4_state_cache_time == 0)
1299 nsrv4->rfs4_state_cache_time = STATE_CACHE_TIME;
1300 if (nsrv4->rfs4_lo_state_cache_time == 0)
1301 nsrv4->rfs4_lo_state_cache_time = LO_STATE_CACHE_TIME;
1302 if (nsrv4->rfs4_lockowner_cache_time == 0)
1303 nsrv4->rfs4_lockowner_cache_time = LOCKOWNER_CACHE_TIME;
1304 if (nsrv4->rfs4_file_cache_time == 0)
1305 nsrv4->rfs4_file_cache_time = FILE_CACHE_TIME;
1306 if (nsrv4->rfs4_deleg_state_cache_time == 0)
1307 nsrv4->rfs4_deleg_state_cache_time = DELEG_STATE_CACHE_TIME;
1308
1309 /* Create the overall database to hold all server state */
1310 nsrv4->nfs4_server_state = rfs4_database_create(rfs4_database_debug);
1311
1312 /* Now create the individual tables */
1313 nsrv4->rfs4_client_cache_time *= rfs4_lease_time;
1314 nsrv4->rfs4_client_tab = rfs4_table_create(nsrv4->nfs4_server_state,
1315 "Client",
1316 nsrv4->rfs4_client_cache_time,
1317 2,
1318 rfs4_client_create,
1319 rfs4_client_destroy,
1320 rfs4_client_expiry,
1321 sizeof (rfs4_client_t),
1322 TABSIZE,
1323 MAXTABSZ/8, 100);
1324 nsrv4->rfs4_nfsclnt_idx = rfs4_index_create(nsrv4->rfs4_client_tab,
1325 "nfs_client_id4", nfsclnt_hash,
1326 nfsclnt_compare, nfsclnt_mkkey,
1327 TRUE);
1328 nsrv4->rfs4_clientid_idx = rfs4_index_create(nsrv4->rfs4_client_tab,
1329 "client_id", clientid_hash,
1330 clientid_compare, clientid_mkkey,
1331 FALSE);
1332
1333 nsrv4->rfs4_clntip_cache_time = 86400 * 365; /* about a year */
1334 nsrv4->rfs4_clntip_tab = rfs4_table_create(nsrv4->nfs4_server_state,
1335 "ClntIP",
1336 nsrv4->rfs4_clntip_cache_time,
1337 1,
1338 rfs4_clntip_create,
1339 rfs4_clntip_destroy,
1340 rfs4_clntip_expiry,
1341 sizeof (rfs4_clntip_t),
1342 TABSIZE,
1343 MAXTABSZ, 100);
1344 nsrv4->rfs4_clntip_idx = rfs4_index_create(nsrv4->rfs4_clntip_tab,
1345 "client_ip", clntip_hash,
1346 clntip_compare, clntip_mkkey,
1347 TRUE);
1348
1349 nsrv4->rfs4_openowner_cache_time *= rfs4_lease_time;
1350 nsrv4->rfs4_openowner_tab = rfs4_table_create(nsrv4->nfs4_server_state,
1351 "OpenOwner",
1352 nsrv4->rfs4_openowner_cache_time,
1353 1,
1354 rfs4_openowner_create,
1355 rfs4_openowner_destroy,
1356 rfs4_openowner_expiry,
1357 sizeof (rfs4_openowner_t),
1358 TABSIZE,
1359 MAXTABSZ, 100);
1360 nsrv4->rfs4_openowner_idx = rfs4_index_create(nsrv4->rfs4_openowner_tab,
1361 "open_owner4", openowner_hash,
1362 openowner_compare,
1363 openowner_mkkey, TRUE);
1364
1365 nsrv4->rfs4_state_cache_time *= rfs4_lease_time;
1366 nsrv4->rfs4_state_tab = rfs4_table_create(nsrv4->nfs4_server_state,
1367 "OpenStateID",
1368 nsrv4->rfs4_state_cache_time,
1369 3,
1370 rfs4_state_create,
1371 rfs4_state_destroy,
1372 rfs4_state_expiry,
1373 sizeof (rfs4_state_t),
1374 TABSIZE,
1375 MAXTABSZ, 100);
1376
1377 /* CSTYLED */
1378 nsrv4->rfs4_state_owner_file_idx = rfs4_index_create(nsrv4->rfs4_state_tab,
1379 "Openowner-File",
1380 state_owner_file_hash,
1381 state_owner_file_compare,
1382 state_owner_file_mkkey, TRUE);
1383
1384 nsrv4->rfs4_state_idx = rfs4_index_create(nsrv4->rfs4_state_tab,
1385 "State-id", state_hash,
1386 state_compare, state_mkkey, FALSE);
1387
1388 nsrv4->rfs4_state_file_idx = rfs4_index_create(nsrv4->rfs4_state_tab,
1389 "File", state_file_hash,
1390 state_file_compare, state_file_mkkey,
1391 FALSE);
1392
1393 nsrv4->rfs4_lo_state_cache_time *= rfs4_lease_time;
1394 nsrv4->rfs4_lo_state_tab = rfs4_table_create(nsrv4->nfs4_server_state,
1395 "LockStateID",
1396 nsrv4->rfs4_lo_state_cache_time,
1397 2,
1398 rfs4_lo_state_create,
1399 rfs4_lo_state_destroy,
1400 rfs4_lo_state_expiry,
1401 sizeof (rfs4_lo_state_t),
1402 TABSIZE,
1403 MAXTABSZ, 100);
1404
1405 /* CSTYLED */
1406 nsrv4->rfs4_lo_state_owner_idx = rfs4_index_create(nsrv4->rfs4_lo_state_tab,
1407 "lockownerxstate",
1408 lo_state_lo_hash,
1409 lo_state_lo_compare,
1410 lo_state_lo_mkkey, TRUE);
1411
1412 nsrv4->rfs4_lo_state_idx = rfs4_index_create(nsrv4->rfs4_lo_state_tab,
1413 "State-id",
1414 lo_state_hash, lo_state_compare,
1415 lo_state_mkkey, FALSE);
1416
1417 nsrv4->rfs4_lockowner_cache_time *= rfs4_lease_time;
1418
1419 nsrv4->rfs4_lockowner_tab = rfs4_table_create(nsrv4->nfs4_server_state,
1420 "Lockowner",
1421 nsrv4->rfs4_lockowner_cache_time,
1422 2,
1423 rfs4_lockowner_create,
1424 rfs4_lockowner_destroy,
1425 rfs4_lockowner_expiry,
1426 sizeof (rfs4_lockowner_t),
1427 TABSIZE,
1428 MAXTABSZ, 100);
1429
1430 nsrv4->rfs4_lockowner_idx = rfs4_index_create(nsrv4->rfs4_lockowner_tab,
1431 "lock_owner4", lockowner_hash,
1432 lockowner_compare,
1433 lockowner_mkkey, TRUE);
1434
1435 /* CSTYLED */
1436 nsrv4->rfs4_lockowner_pid_idx = rfs4_index_create(nsrv4->rfs4_lockowner_tab,
1437 "pid", pid_hash,
1438 pid_compare, pid_mkkey,
1439 FALSE);
1440
1441 nsrv4->rfs4_file_cache_time *= rfs4_lease_time;
1442 nsrv4->rfs4_file_tab = rfs4_table_create(nsrv4->nfs4_server_state,
1443 "File",
1444 nsrv4->rfs4_file_cache_time,
1445 1,
1446 rfs4_file_create,
1447 rfs4_file_destroy,
1448 NULL,
1449 sizeof (rfs4_file_t),
1450 TABSIZE,
1451 MAXTABSZ, -1);
1452
1453 nsrv4->rfs4_file_idx = rfs4_index_create(nsrv4->rfs4_file_tab,
1454 "Filehandle", file_hash,
1455 file_compare, file_mkkey, TRUE);
1456
1457 nsrv4->rfs4_deleg_state_cache_time *= rfs4_lease_time;
1458 /* CSTYLED */
1459 nsrv4->rfs4_deleg_state_tab = rfs4_table_create(nsrv4->nfs4_server_state,
1460 "DelegStateID",
1461 nsrv4->rfs4_deleg_state_cache_time,
1462 2,
1463 rfs4_deleg_state_create,
1464 rfs4_deleg_state_destroy,
1465 rfs4_deleg_state_expiry,
1466 sizeof (rfs4_deleg_state_t),
1467 TABSIZE,
1468 MAXTABSZ, 100);
1469 nsrv4->rfs4_deleg_idx = rfs4_index_create(nsrv4->rfs4_deleg_state_tab,
1470 "DelegByFileClient",
1471 deleg_hash,
1472 deleg_compare,
1473 deleg_mkkey, TRUE);
1474
1475 /* CSTYLED */
1476 nsrv4->rfs4_deleg_state_idx = rfs4_index_create(nsrv4->rfs4_deleg_state_tab,
1477 "DelegState",
1478 deleg_state_hash,
1479 deleg_state_compare,
1480 deleg_state_mkkey, FALSE);
1481
1482 mutex_exit(&nsrv4->state_lock);
1483
1484 /*
1485 * Init the stable storage.
1486 */
1487 rfs4_ss_init(nsrv4);
1488 }
1489
1490 /*
1491 * Used at server shutdown to cleanup all of NFSv4 server's zone structures
1492 * and state.
1493 */
1494 void
1495 rfs4_state_zone_fini()
1496 {
1497 rfs4_database_t *dbp;
1498 nfs4_srv_t *nsrv4;
1499 nsrv4 = nfs4_get_srv();
1500
1501 rfs4_set_deleg_policy(nsrv4, SRV_NEVER_DELEGATE);
1502
1503 /*
1504 * Clean up any dangling stable storage structures BEFORE calling
1505 * rfs4_servinst_destroy_all() so there are no dangling structures
1506 * (i.e. the srvinsts are all cleared of danglers BEFORE they get
1507 * freed).
1508 */
1509 rfs4_ss_fini(nsrv4);
1510
1511 mutex_enter(&nsrv4->state_lock);
1512
1513 if (nsrv4->nfs4_server_state == NULL) {
1514 mutex_exit(&nsrv4->state_lock);
1515 return;
1516 }
1517
1518 /* destroy server instances and current instance ptr */
1519 rfs4_servinst_destroy_all(nsrv4);
1520
1521 /* reset the "first NFSv4 request" status */
1522 nsrv4->seen_first_compound = 0;
1523
1524 dbp = nsrv4->nfs4_server_state;
1525 nsrv4->nfs4_server_state = NULL;
1526
1527 rw_destroy(&nsrv4->rfs4_findclient_lock);
1528
1529 /* First stop all of the reaper threads in the database */
1530 rfs4_database_shutdown(dbp);
1531
1532 /*
1533 * WARNING: There may be consumers of the rfs4 database still
1534 * active as we destroy these. IF that's the case, consider putting
1535 * some of their _zone_fini()-like functions into the zsd key as
1536 * ~~SHUTDOWN~~ functions instead of ~~DESTROY~~ functions. We can
1537 * maintain some ordering guarantees better that way.
1538 */
1539 /* Now destroy/release the database tables */
1540 rfs4_database_destroy(dbp);
1541
1542 /* Reset the cache timers for next time */
1543 nsrv4->rfs4_client_cache_time = 0;
1544 nsrv4->rfs4_openowner_cache_time = 0;
1545 nsrv4->rfs4_state_cache_time = 0;
1546 nsrv4->rfs4_lo_state_cache_time = 0;
1547 nsrv4->rfs4_lockowner_cache_time = 0;
1548 nsrv4->rfs4_file_cache_time = 0;
1549 nsrv4->rfs4_deleg_state_cache_time = 0;
1550
1551 mutex_exit(&nsrv4->state_lock);
1552 }
1553
1554 typedef union {
1555 struct {
1556 uint32_t start_time;
1557 uint32_t c_id;
1558 } impl_id;
1559 clientid4 id4;
1560 } cid;
1561
1562 static int foreign_stateid(stateid_t *id);
1563 static int foreign_clientid(cid *cidp);
1564 static void embed_nodeid(cid *cidp);
1565
1566 typedef union {
1567 struct {
1568 uint32_t c_id;
1569 uint32_t gen_num;
1570 } cv_impl;
1571 verifier4 confirm_verf;
1572 } scid_confirm_verf;
1573
1574 static uint32_t
1575 clientid_hash(void *key)
1576 {
1577 cid *idp = key;
1578
1579 return (idp->impl_id.c_id);
1580 }
1581
1582 static bool_t
1583 clientid_compare(rfs4_entry_t entry, void *key)
1584 {
1585 rfs4_client_t *cp = (rfs4_client_t *)entry;
1586 clientid4 *idp = key;
1587
1588 return (*idp == cp->rc_clientid);
1589 }
1590
1591 static void *
1592 clientid_mkkey(rfs4_entry_t entry)
1593 {
1594 rfs4_client_t *cp = (rfs4_client_t *)entry;
1595
1596 return (&cp->rc_clientid);
1597 }
1598
1599 static uint32_t
1600 nfsclnt_hash(void *key)
1601 {
1602 nfs_client_id4 *client = key;
1603 int i;
1604 uint32_t hash = 0;
1605
1606 for (i = 0; i < client->id_len; i++) {
1607 hash <<= 1;
1608 hash += (uint_t)client->id_val[i];
1609 }
1610 return (hash);
1611 }
1612
1613
1614 static bool_t
1615 nfsclnt_compare(rfs4_entry_t entry, void *key)
1616 {
1617 rfs4_client_t *cp = (rfs4_client_t *)entry;
1618 nfs_client_id4 *nfs_client = key;
1619
1620 if (cp->rc_nfs_client.id_len != nfs_client->id_len)
1621 return (FALSE);
1622
1623 return (bcmp(cp->rc_nfs_client.id_val, nfs_client->id_val,
1624 nfs_client->id_len) == 0);
1625 }
1626
1627 static void *
1628 nfsclnt_mkkey(rfs4_entry_t entry)
1629 {
1630 rfs4_client_t *cp = (rfs4_client_t *)entry;
1631
1632 return (&cp->rc_nfs_client);
1633 }
1634
1635 static bool_t
1636 rfs4_client_expiry(rfs4_entry_t u_entry)
1637 {
1638 rfs4_client_t *cp = (rfs4_client_t *)u_entry;
1639 bool_t cp_expired;
1640
1641 if (rfs4_dbe_is_invalid(cp->rc_dbe)) {
1642 cp->rc_ss_remove = 1;
1643 return (TRUE);
1644 }
1645 /*
1646 * If the sysadmin has used clear_locks for this
1647 * entry then forced_expire will be set and we
1648 * want this entry to be reaped. Or the entry
1649 * has exceeded its lease period.
1650 */
1651 cp_expired = (cp->rc_forced_expire ||
1652 (gethrestime_sec() - cp->rc_last_access
1653 > rfs4_lease_time));
1654
1655 if (!cp->rc_ss_remove && cp_expired)
1656 cp->rc_ss_remove = 1;
1657 return (cp_expired);
1658 }
1659
1660 /*
1661 * Remove the leaf file from all distributed stable storage paths.
1662 */
1663 static void
1664 rfs4_dss_remove_cpleaf(rfs4_client_t *cp)
1665 {
1666 nfs4_srv_t *nsrv4;
1667 rfs4_servinst_t *sip;
1668 char *leaf = cp->rc_ss_pn->leaf;
1669
1670 /*
1671 * since the state files are written to all DSS
1672 * paths we must remove this leaf file instance
1673 * from all server instances.
1674 */
1675
1676 nsrv4 = nfs4_get_srv();
1677 mutex_enter(&nsrv4->servinst_lock);
1678 for (sip = nsrv4->nfs4_cur_servinst; sip != NULL; sip = sip->prev) {
1679 /* remove the leaf file associated with this server instance */
1680 rfs4_dss_remove_leaf(sip, NFS4_DSS_STATE_LEAF, leaf);
1681 }
1682 mutex_exit(&nsrv4->servinst_lock);
1683 }
1684
1685 static void
1686 rfs4_dss_remove_leaf(rfs4_servinst_t *sip, char *dir_leaf, char *leaf)
1687 {
1688 int i, npaths = sip->dss_npaths;
1689
1690 for (i = 0; i < npaths; i++) {
1691 rfs4_dss_path_t *dss_path = sip->dss_paths[i];
1692 char *path, *dir;
1693 size_t pathlen;
1694
1695 /* the HA-NFSv4 path might have been failed-over away from us */
1696 if (dss_path == NULL)
1697 continue;
1698
1699 dir = dss_path->path;
1700
1701 /* allow 3 extra bytes for two '/' & a NUL */
1702 pathlen = strlen(dir) + strlen(dir_leaf) + strlen(leaf) + 3;
1703 path = kmem_alloc(pathlen, KM_SLEEP);
1704 (void) sprintf(path, "%s/%s/%s", dir, dir_leaf, leaf);
1705
1706 (void) vn_remove(path, UIO_SYSSPACE, RMFILE);
1707
1708 kmem_free(path, pathlen);
1709 }
1710 }
1711
1712 static void
1713 rfs4_client_destroy(rfs4_entry_t u_entry)
1714 {
1715 rfs4_client_t *cp = (rfs4_client_t *)u_entry;
1716
1717 mutex_destroy(cp->rc_cbinfo.cb_lock);
1718 cv_destroy(cp->rc_cbinfo.cb_cv);
1719 cv_destroy(cp->rc_cbinfo.cb_cv_nullcaller);
1720 list_destroy(&cp->rc_openownerlist);
1721
1722 /* free callback info */
1723 rfs4_cbinfo_free(&cp->rc_cbinfo);
1724
1725 if (cp->rc_cp_confirmed)
1726 rfs4_client_rele(cp->rc_cp_confirmed);
1727
1728 if (cp->rc_ss_pn) {
1729 /* check if the stable storage files need to be removed */
1730 if (cp->rc_ss_remove)
1731 rfs4_dss_remove_cpleaf(cp);
1732 rfs4_ss_pnfree(cp->rc_ss_pn);
1733 }
1734
1735 /* Free the client supplied client id */
1736 kmem_free(cp->rc_nfs_client.id_val, cp->rc_nfs_client.id_len);
1737
1738 if (cp->rc_sysidt != LM_NOSYSID)
1739 lm_free_sysidt(cp->rc_sysidt);
1740 }
1741
1742 static bool_t
1743 rfs4_client_create(rfs4_entry_t u_entry, void *arg)
1744 {
1745 rfs4_client_t *cp = (rfs4_client_t *)u_entry;
1746 nfs_client_id4 *client = (nfs_client_id4 *)arg;
1747 struct sockaddr *ca;
1748 cid *cidp;
1749 scid_confirm_verf *scvp;
1750 nfs4_srv_t *nsrv4;
1751
1752 nsrv4 = nfs4_get_srv();
1753
1754 /* Get a clientid to give to the client */
1755 cidp = (cid *)&cp->rc_clientid;
1756 cidp->impl_id.start_time = nsrv4->rfs4_start_time;
1757 cidp->impl_id.c_id = (uint32_t)rfs4_dbe_getid(cp->rc_dbe);
1758
1759 /* If we are booted as a cluster node, embed our nodeid */
1760 if (cluster_bootflags & CLUSTER_BOOTED)
1761 embed_nodeid(cidp);
1762
1763 /* Allocate and copy client's client id value */
1764 cp->rc_nfs_client.id_val = kmem_alloc(client->id_len, KM_SLEEP);
1765 cp->rc_nfs_client.id_len = client->id_len;
1766 bcopy(client->id_val, cp->rc_nfs_client.id_val, client->id_len);
1767 cp->rc_nfs_client.verifier = client->verifier;
1768
1769 /* Copy client's IP address */
1770 ca = client->cl_addr;
1771 if (ca->sa_family == AF_INET)
1772 bcopy(ca, &cp->rc_addr, sizeof (struct sockaddr_in));
1773 else if (ca->sa_family == AF_INET6)
1774 bcopy(ca, &cp->rc_addr, sizeof (struct sockaddr_in6));
1775 cp->rc_nfs_client.cl_addr = (struct sockaddr *)&cp->rc_addr;
1776
1777 /* Init the value for the SETCLIENTID_CONFIRM verifier */
1778 scvp = (scid_confirm_verf *)&cp->rc_confirm_verf;
1779 scvp->cv_impl.c_id = cidp->impl_id.c_id;
1780 scvp->cv_impl.gen_num = 0;
1781
1782 /* An F_UNLKSYS has been done for this client */
1783 cp->rc_unlksys_completed = FALSE;
1784
1785 /* We need the client to ack us */
1786 cp->rc_need_confirm = TRUE;
1787 cp->rc_cp_confirmed = NULL;
1788
1789 /* TRUE all the time until the callback path actually fails */
1790 cp->rc_cbinfo.cb_notified_of_cb_path_down = TRUE;
1791
1792 /* Initialize the access time to now */
1793 cp->rc_last_access = gethrestime_sec();
1794
1795 cp->rc_cr_set = NULL;
1796
1797 cp->rc_sysidt = LM_NOSYSID;
1798
1799 list_create(&cp->rc_openownerlist, sizeof (rfs4_openowner_t),
1800 offsetof(rfs4_openowner_t, ro_node));
1801
1802 /* set up the callback control structure */
1803 cp->rc_cbinfo.cb_state = CB_UNINIT;
1804 mutex_init(cp->rc_cbinfo.cb_lock, NULL, MUTEX_DEFAULT, NULL);
1805 cv_init(cp->rc_cbinfo.cb_cv, NULL, CV_DEFAULT, NULL);
1806 cv_init(cp->rc_cbinfo.cb_cv_nullcaller, NULL, CV_DEFAULT, NULL);
1807
1808 /*
1809 * Associate the client_t with the current server instance.
1810 * The hold is solely to satisfy the calling requirement of
1811 * rfs4_servinst_assign(). In this case it's not strictly necessary.
1812 */
1813 rfs4_dbe_hold(cp->rc_dbe);
1814 rfs4_servinst_assign(nsrv4, cp, nsrv4->nfs4_cur_servinst);
1815 rfs4_dbe_rele(cp->rc_dbe);
1816
1817 return (TRUE);
1818 }
1819
1820 /*
1821 * Caller wants to generate/update the setclientid_confirm verifier
1822 * associated with a client. This is done during the SETCLIENTID
1823 * processing.
1824 */
1825 void
1826 rfs4_client_scv_next(rfs4_client_t *cp)
1827 {
1828 scid_confirm_verf *scvp;
1829
1830 /* Init the value for the SETCLIENTID_CONFIRM verifier */
1831 scvp = (scid_confirm_verf *)&cp->rc_confirm_verf;
1832 scvp->cv_impl.gen_num++;
1833 }
1834
1835 void
1836 rfs4_client_rele(rfs4_client_t *cp)
1837 {
1838 rfs4_dbe_rele(cp->rc_dbe);
1839 }
1840
1841 rfs4_client_t *
1842 rfs4_findclient(nfs_client_id4 *client, bool_t *create, rfs4_client_t *oldcp)
1843 {
1844 rfs4_client_t *cp;
1845 nfs4_srv_t *nsrv4;
1846 nsrv4 = nfs4_get_srv();
1847
1848
1849 if (oldcp) {
1850 rw_enter(&nsrv4->rfs4_findclient_lock, RW_WRITER);
1851 rfs4_dbe_hide(oldcp->rc_dbe);
1852 } else {
1853 rw_enter(&nsrv4->rfs4_findclient_lock, RW_READER);
1854 }
1855
1856 cp = (rfs4_client_t *)rfs4_dbsearch(nsrv4->rfs4_nfsclnt_idx, client,
1857 create, (void *)client, RFS4_DBS_VALID);
1858
1859 if (oldcp)
1860 rfs4_dbe_unhide(oldcp->rc_dbe);
1861
1862 rw_exit(&nsrv4->rfs4_findclient_lock);
1863
1864 return (cp);
1865 }
1866
1867 rfs4_client_t *
1868 rfs4_findclient_by_id(clientid4 clientid, bool_t find_unconfirmed)
1869 {
1870 rfs4_client_t *cp;
1871 bool_t create = FALSE;
1872 cid *cidp = (cid *)&clientid;
1873 nfs4_srv_t *nsrv4 = nfs4_get_srv();
1874
1875 /* If we're a cluster and the nodeid isn't right, short-circuit */
1876 if (cluster_bootflags & CLUSTER_BOOTED && foreign_clientid(cidp))
1877 return (NULL);
1878
1879 rw_enter(&nsrv4->rfs4_findclient_lock, RW_READER);
1880
1881 cp = (rfs4_client_t *)rfs4_dbsearch(nsrv4->rfs4_clientid_idx, &clientid,
1882 &create, NULL, RFS4_DBS_VALID);
1883
1884 rw_exit(&nsrv4->rfs4_findclient_lock);
1885
1886 if (cp && cp->rc_need_confirm && find_unconfirmed == FALSE) {
1887 rfs4_client_rele(cp);
1888 return (NULL);
1889 } else {
1890 return (cp);
1891 }
1892 }
1893
1894 static uint32_t
1895 clntip_hash(void *key)
1896 {
1897 struct sockaddr *addr = key;
1898 int i, len = 0;
1899 uint32_t hash = 0;
1900 char *ptr;
1901
1902 if (addr->sa_family == AF_INET) {
1903 struct sockaddr_in *a = (struct sockaddr_in *)addr;
1904 len = sizeof (struct in_addr);
1905 ptr = (char *)&a->sin_addr;
1906 } else if (addr->sa_family == AF_INET6) {
1907 struct sockaddr_in6 *a = (struct sockaddr_in6 *)addr;
1908 len = sizeof (struct in6_addr);
1909 ptr = (char *)&a->sin6_addr;
1910 } else
1911 return (0);
1912
1913 for (i = 0; i < len; i++) {
1914 hash <<= 1;
1915 hash += (uint_t)ptr[i];
1916 }
1917 return (hash);
1918 }
1919
1920 static bool_t
1921 clntip_compare(rfs4_entry_t entry, void *key)
1922 {
1923 rfs4_clntip_t *cp = (rfs4_clntip_t *)entry;
1924 struct sockaddr *addr = key;
1925 int len = 0;
1926 char *p1, *p2;
1927
1928 if (addr->sa_family == AF_INET) {
1929 struct sockaddr_in *a1 = (struct sockaddr_in *)&cp->ri_addr;
1930 struct sockaddr_in *a2 = (struct sockaddr_in *)addr;
1931 len = sizeof (struct in_addr);
1932 p1 = (char *)&a1->sin_addr;
1933 p2 = (char *)&a2->sin_addr;
1934 } else if (addr->sa_family == AF_INET6) {
1935 struct sockaddr_in6 *a1 = (struct sockaddr_in6 *)&cp->ri_addr;
1936 struct sockaddr_in6 *a2 = (struct sockaddr_in6 *)addr;
1937 len = sizeof (struct in6_addr);
1938 p1 = (char *)&a1->sin6_addr;
1939 p2 = (char *)&a2->sin6_addr;
1940 } else
1941 return (0);
1942
1943 return (bcmp(p1, p2, len) == 0);
1944 }
1945
1946 static void *
1947 clntip_mkkey(rfs4_entry_t entry)
1948 {
1949 rfs4_clntip_t *cp = (rfs4_clntip_t *)entry;
1950
1951 return (&cp->ri_addr);
1952 }
1953
1954 static bool_t
1955 rfs4_clntip_expiry(rfs4_entry_t u_entry)
1956 {
1957 rfs4_clntip_t *cp = (rfs4_clntip_t *)u_entry;
1958
1959 if (rfs4_dbe_is_invalid(cp->ri_dbe))
1960 return (TRUE);
1961 return (FALSE);
1962 }
1963
1964 /* ARGSUSED */
1965 static void
1966 rfs4_clntip_destroy(rfs4_entry_t u_entry)
1967 {
1968 }
1969
1970 static bool_t
1971 rfs4_clntip_create(rfs4_entry_t u_entry, void *arg)
1972 {
1973 rfs4_clntip_t *cp = (rfs4_clntip_t *)u_entry;
1974 struct sockaddr *ca = (struct sockaddr *)arg;
1975
1976 /* Copy client's IP address */
1977 if (ca->sa_family == AF_INET)
1978 bcopy(ca, &cp->ri_addr, sizeof (struct sockaddr_in));
1979 else if (ca->sa_family == AF_INET6)
1980 bcopy(ca, &cp->ri_addr, sizeof (struct sockaddr_in6));
1981 else
1982 return (FALSE);
1983 cp->ri_no_referrals = 1;
1984
1985 return (TRUE);
1986 }
1987
1988 rfs4_clntip_t *
1989 rfs4_find_clntip(struct sockaddr *addr, bool_t *create)
1990 {
1991 rfs4_clntip_t *cp;
1992 nfs4_srv_t *nsrv4;
1993
1994 nsrv4 = nfs4_get_srv();
1995
1996 rw_enter(&nsrv4->rfs4_findclient_lock, RW_READER);
1997
1998 cp = (rfs4_clntip_t *)rfs4_dbsearch(nsrv4->rfs4_clntip_idx, addr,
1999 create, addr, RFS4_DBS_VALID);
2000
2001 rw_exit(&nsrv4->rfs4_findclient_lock);
2002
2003 return (cp);
2004 }
2005
2006 void
2007 rfs4_invalidate_clntip(struct sockaddr *addr)
2008 {
2009 rfs4_clntip_t *cp;
2010 bool_t create = FALSE;
2011 nfs4_srv_t *nsrv4 = nfs4_get_srv();
2012
2013 rw_enter(&nsrv4->rfs4_findclient_lock, RW_READER);
2014
2015 cp = (rfs4_clntip_t *)rfs4_dbsearch(nsrv4->rfs4_clntip_idx, addr,
2016 &create, NULL, RFS4_DBS_VALID);
2017 if (cp == NULL) {
2018 rw_exit(&nsrv4->rfs4_findclient_lock);
2019 return;
2020 }
2021 rfs4_dbe_invalidate(cp->ri_dbe);
2022 rfs4_dbe_rele(cp->ri_dbe);
2023
2024 rw_exit(&nsrv4->rfs4_findclient_lock);
2025 }
2026
2027 bool_t
2028 rfs4_lease_expired(rfs4_client_t *cp)
2029 {
2030 bool_t rc;
2031
2032 rfs4_dbe_lock(cp->rc_dbe);
2033
2034 /*
2035 * If the admin has executed clear_locks for this
2036 * client id, force expire will be set, so no need
2037 * to calculate anything because it's "outa here".
2038 */
2039 if (cp->rc_forced_expire) {
2040 rc = TRUE;
2041 } else {
2042 rc = (gethrestime_sec() - cp->rc_last_access > rfs4_lease_time);
2043 }
2044
2045 /*
2046 * If the lease has expired we will also want
2047 * to remove any stable storage state data. So
2048 * mark the client id accordingly.
2049 */
2050 if (!cp->rc_ss_remove)
2051 cp->rc_ss_remove = (rc == TRUE);
2052
2053 rfs4_dbe_unlock(cp->rc_dbe);
2054
2055 return (rc);
2056 }
2057
2058 void
2059 rfs4_update_lease(rfs4_client_t *cp)
2060 {
2061 rfs4_dbe_lock(cp->rc_dbe);
2062 if (!cp->rc_forced_expire)
2063 cp->rc_last_access = gethrestime_sec();
2064 rfs4_dbe_unlock(cp->rc_dbe);
2065 }
2066
2067
2068 static bool_t
2069 EQOPENOWNER(open_owner4 *a, open_owner4 *b)
2070 {
2071 bool_t rc;
2072
2073 if (a->clientid != b->clientid)
2074 return (FALSE);
2075
2076 if (a->owner_len != b->owner_len)
2077 return (FALSE);
2078
2079 rc = (bcmp(a->owner_val, b->owner_val, a->owner_len) == 0);
2080
2081 return (rc);
2082 }
2083
2084 static uint_t
2085 openowner_hash(void *key)
2086 {
2087 int i;
2088 open_owner4 *openowner = key;
2089 uint_t hash = 0;
2090
2091 for (i = 0; i < openowner->owner_len; i++) {
2092 hash <<= 4;
2093 hash += (uint_t)openowner->owner_val[i];
2094 }
2095 hash += (uint_t)openowner->clientid;
2096 hash |= (openowner->clientid >> 32);
2097
2098 return (hash);
2099 }
2100
2101 static bool_t
2102 openowner_compare(rfs4_entry_t u_entry, void *key)
2103 {
2104 rfs4_openowner_t *oo = (rfs4_openowner_t *)u_entry;
2105 open_owner4 *arg = key;
2106
2107 return (EQOPENOWNER(&oo->ro_owner, arg));
2108 }
2109
2110 void *
2111 openowner_mkkey(rfs4_entry_t u_entry)
2112 {
2113 rfs4_openowner_t *oo = (rfs4_openowner_t *)u_entry;
2114
2115 return (&oo->ro_owner);
2116 }
2117
2118 /* ARGSUSED */
2119 static bool_t
2120 rfs4_openowner_expiry(rfs4_entry_t u_entry)
2121 {
2122 /* openstateid held us and did all needed delay */
2123 return (TRUE);
2124 }
2125
2126 static void
2127 rfs4_openowner_destroy(rfs4_entry_t u_entry)
2128 {
2129 rfs4_openowner_t *oo = (rfs4_openowner_t *)u_entry;
2130
2131 /* Remove open owner from client's lists of open owners */
2132 rfs4_dbe_lock(oo->ro_client->rc_dbe);
2133 list_remove(&oo->ro_client->rc_openownerlist, oo);
2134 rfs4_dbe_unlock(oo->ro_client->rc_dbe);
2135
2136 /* One less reference to the client */
2137 rfs4_client_rele(oo->ro_client);
2138 oo->ro_client = NULL;
2139
2140 /* Free the last reply for this lock owner */
2141 rfs4_free_reply(&oo->ro_reply);
2142
2143 if (oo->ro_reply_fh.nfs_fh4_val) {
2144 kmem_free(oo->ro_reply_fh.nfs_fh4_val,
2145 oo->ro_reply_fh.nfs_fh4_len);
2146 oo->ro_reply_fh.nfs_fh4_val = NULL;
2147 oo->ro_reply_fh.nfs_fh4_len = 0;
2148 }
2149
2150 rfs4_sw_destroy(&oo->ro_sw);
2151 list_destroy(&oo->ro_statelist);
2152
2153 /* Free the lock owner id */
2154 kmem_free(oo->ro_owner.owner_val, oo->ro_owner.owner_len);
2155 }
2156
2157 void
2158 rfs4_openowner_rele(rfs4_openowner_t *oo)
2159 {
2160 rfs4_dbe_rele(oo->ro_dbe);
2161 }
2162
2163 static bool_t
2164 rfs4_openowner_create(rfs4_entry_t u_entry, void *arg)
2165 {
2166 rfs4_openowner_t *oo = (rfs4_openowner_t *)u_entry;
2167 rfs4_openowner_t *argp = (rfs4_openowner_t *)arg;
2168 open_owner4 *openowner = &argp->ro_owner;
2169 seqid4 seqid = argp->ro_open_seqid;
2170 rfs4_client_t *cp;
2171 bool_t create = FALSE;
2172 nfs4_srv_t *nsrv4 = nfs4_get_srv();
2173
2174 rw_enter(&nsrv4->rfs4_findclient_lock, RW_READER);
2175
2176 cp = (rfs4_client_t *)rfs4_dbsearch(nsrv4->rfs4_clientid_idx,
2177 &openowner->clientid,
2178 &create, NULL, RFS4_DBS_VALID);
2179
2180 rw_exit(&nsrv4->rfs4_findclient_lock);
2181
2182 if (cp == NULL)
2183 return (FALSE);
2184
2185 oo->ro_reply_fh.nfs_fh4_len = 0;
2186 oo->ro_reply_fh.nfs_fh4_val = NULL;
2187
2188 oo->ro_owner.clientid = openowner->clientid;
2189 oo->ro_owner.owner_val =
2190 kmem_alloc(openowner->owner_len, KM_SLEEP);
2191
2192 bcopy(openowner->owner_val,
2193 oo->ro_owner.owner_val, openowner->owner_len);
2194
2195 oo->ro_owner.owner_len = openowner->owner_len;
2196
2197 oo->ro_need_confirm = TRUE;
2198
2199 rfs4_sw_init(&oo->ro_sw);
2200
2201 oo->ro_open_seqid = seqid;
2202 bzero(&oo->ro_reply, sizeof (nfs_resop4));
2203 oo->ro_client = cp;
2204 oo->ro_cr_set = NULL;
2205
2206 list_create(&oo->ro_statelist, sizeof (rfs4_state_t),
2207 offsetof(rfs4_state_t, rs_node));
2208
2209 /* Insert openowner into client's open owner list */
2210 rfs4_dbe_lock(cp->rc_dbe);
2211 list_insert_tail(&cp->rc_openownerlist, oo);
2212 rfs4_dbe_unlock(cp->rc_dbe);
2213
2214 return (TRUE);
2215 }
2216
2217 rfs4_openowner_t *
2218 rfs4_findopenowner(open_owner4 *openowner, bool_t *create, seqid4 seqid)
2219 {
2220 rfs4_openowner_t *oo;
2221 rfs4_openowner_t arg;
2222 nfs4_srv_t *nsrv4 = nfs4_get_srv();
2223
2224 arg.ro_owner = *openowner;
2225 arg.ro_open_seqid = seqid;
2226 /* CSTYLED */
2227 oo = (rfs4_openowner_t *)rfs4_dbsearch(nsrv4->rfs4_openowner_idx, openowner,
2228 create, &arg, RFS4_DBS_VALID);
2229
2230 return (oo);
2231 }
2232
2233 void
2234 rfs4_update_open_sequence(rfs4_openowner_t *oo)
2235 {
2236
2237 rfs4_dbe_lock(oo->ro_dbe);
2238
2239 oo->ro_open_seqid++;
2240
2241 rfs4_dbe_unlock(oo->ro_dbe);
2242 }
2243
2244 void
2245 rfs4_update_open_resp(rfs4_openowner_t *oo, nfs_resop4 *resp, nfs_fh4 *fh)
2246 {
2247
2248 rfs4_dbe_lock(oo->ro_dbe);
2249
2250 rfs4_free_reply(&oo->ro_reply);
2251
2252 rfs4_copy_reply(&oo->ro_reply, resp);
2253
2254 /* Save the filehandle if provided and free if not used */
2255 if (resp->nfs_resop4_u.opopen.status == NFS4_OK &&
2256 fh && fh->nfs_fh4_len) {
2257 if (oo->ro_reply_fh.nfs_fh4_val == NULL)
2258 oo->ro_reply_fh.nfs_fh4_val =
2259 kmem_alloc(fh->nfs_fh4_len, KM_SLEEP);
2260 nfs_fh4_copy(fh, &oo->ro_reply_fh);
2261 } else {
2262 if (oo->ro_reply_fh.nfs_fh4_val) {
2263 kmem_free(oo->ro_reply_fh.nfs_fh4_val,
2264 oo->ro_reply_fh.nfs_fh4_len);
2265 oo->ro_reply_fh.nfs_fh4_val = NULL;
2266 oo->ro_reply_fh.nfs_fh4_len = 0;
2267 }
2268 }
2269
2270 rfs4_dbe_unlock(oo->ro_dbe);
2271 }
2272
2273 static bool_t
2274 lockowner_compare(rfs4_entry_t u_entry, void *key)
2275 {
2276 rfs4_lockowner_t *lo = (rfs4_lockowner_t *)u_entry;
2277 lock_owner4 *b = (lock_owner4 *)key;
2278
2279 if (lo->rl_owner.clientid != b->clientid)
2280 return (FALSE);
2281
2282 if (lo->rl_owner.owner_len != b->owner_len)
2283 return (FALSE);
2284
2285 return (bcmp(lo->rl_owner.owner_val, b->owner_val,
2286 lo->rl_owner.owner_len) == 0);
2287 }
2288
2289 void *
2290 lockowner_mkkey(rfs4_entry_t u_entry)
2291 {
2292 rfs4_lockowner_t *lo = (rfs4_lockowner_t *)u_entry;
2293
2294 return (&lo->rl_owner);
2295 }
2296
2297 static uint32_t
2298 lockowner_hash(void *key)
2299 {
2300 int i;
2301 lock_owner4 *lockowner = key;
2302 uint_t hash = 0;
2303
2304 for (i = 0; i < lockowner->owner_len; i++) {
2305 hash <<= 4;
2306 hash += (uint_t)lockowner->owner_val[i];
2307 }
2308 hash += (uint_t)lockowner->clientid;
2309 hash |= (lockowner->clientid >> 32);
2310
2311 return (hash);
2312 }
2313
2314 static uint32_t
2315 pid_hash(void *key)
2316 {
2317 return ((uint32_t)(uintptr_t)key);
2318 }
2319
2320 static void *
2321 pid_mkkey(rfs4_entry_t u_entry)
2322 {
2323 rfs4_lockowner_t *lo = (rfs4_lockowner_t *)u_entry;
2324
2325 return ((void *)(uintptr_t)lo->rl_pid);
2326 }
2327
2328 static bool_t
2329 pid_compare(rfs4_entry_t u_entry, void *key)
2330 {
2331 rfs4_lockowner_t *lo = (rfs4_lockowner_t *)u_entry;
2332
2333 return (lo->rl_pid == (pid_t)(uintptr_t)key);
2334 }
2335
2336 static void
2337 rfs4_lockowner_destroy(rfs4_entry_t u_entry)
2338 {
2339 rfs4_lockowner_t *lo = (rfs4_lockowner_t *)u_entry;
2340
2341 /* Free the lock owner id */
2342 kmem_free(lo->rl_owner.owner_val, lo->rl_owner.owner_len);
2343 rfs4_client_rele(lo->rl_client);
2344 }
2345
2346 void
2347 rfs4_lockowner_rele(rfs4_lockowner_t *lo)
2348 {
2349 rfs4_dbe_rele(lo->rl_dbe);
2350 }
2351
2352 /* ARGSUSED */
2353 static bool_t
2354 rfs4_lockowner_expiry(rfs4_entry_t u_entry)
2355 {
2356 /*
2357 * Since expiry is called with no other references on
2358 * this struct, go ahead and have it removed.
2359 */
2360 return (TRUE);
2361 }
2362
2363 static bool_t
2364 rfs4_lockowner_create(rfs4_entry_t u_entry, void *arg)
2365 {
2366 rfs4_lockowner_t *lo = (rfs4_lockowner_t *)u_entry;
2367 lock_owner4 *lockowner = (lock_owner4 *)arg;
2368 rfs4_client_t *cp;
2369 bool_t create = FALSE;
2370 nfs4_srv_t *nsrv4 = nfs4_get_srv();
2371
2372 rw_enter(&nsrv4->rfs4_findclient_lock, RW_READER);
2373
2374 cp = (rfs4_client_t *)rfs4_dbsearch(nsrv4->rfs4_clientid_idx,
2375 &lockowner->clientid,
2376 &create, NULL, RFS4_DBS_VALID);
2377
2378 rw_exit(&nsrv4->rfs4_findclient_lock);
2379
2380 if (cp == NULL)
2381 return (FALSE);
2382
2383 /* Reference client */
2384 lo->rl_client = cp;
2385 lo->rl_owner.clientid = lockowner->clientid;
2386 lo->rl_owner.owner_val = kmem_alloc(lockowner->owner_len, KM_SLEEP);
2387 bcopy(lockowner->owner_val, lo->rl_owner.owner_val,
2388 lockowner->owner_len);
2389 lo->rl_owner.owner_len = lockowner->owner_len;
2390 lo->rl_pid = rfs4_dbe_getid(lo->rl_dbe);
2391
2392 return (TRUE);
2393 }
2394
2395 rfs4_lockowner_t *
2396 rfs4_findlockowner(lock_owner4 *lockowner, bool_t *create)
2397 {
2398 rfs4_lockowner_t *lo;
2399 nfs4_srv_t *nsrv4 = nfs4_get_srv();
2400
2401 /* CSTYLED */
2402 lo = (rfs4_lockowner_t *)rfs4_dbsearch(nsrv4->rfs4_lockowner_idx, lockowner,
2403 create, lockowner, RFS4_DBS_VALID);
2404
2405 return (lo);
2406 }
2407
2408 rfs4_lockowner_t *
2409 rfs4_findlockowner_by_pid(pid_t pid)
2410 {
2411 rfs4_lockowner_t *lo;
2412 bool_t create = FALSE;
2413 nfs4_srv_t *nsrv4 = nfs4_get_srv();
2414
2415 lo = (rfs4_lockowner_t *)rfs4_dbsearch(nsrv4->rfs4_lockowner_pid_idx,
2416 (void *)(uintptr_t)pid, &create, NULL, RFS4_DBS_VALID);
2417
2418 return (lo);
2419 }
2420
2421
2422 static uint32_t
2423 file_hash(void *key)
2424 {
2425 return (ADDRHASH(key));
2426 }
2427
2428 static void *
2429 file_mkkey(rfs4_entry_t u_entry)
2430 {
2431 rfs4_file_t *fp = (rfs4_file_t *)u_entry;
2432
2433 return (fp->rf_vp);
2434 }
2435
2436 static bool_t
2437 file_compare(rfs4_entry_t u_entry, void *key)
2438 {
2439 rfs4_file_t *fp = (rfs4_file_t *)u_entry;
2440
2441 return (fp->rf_vp == (vnode_t *)key);
2442 }
2443
2444 static void
2445 rfs4_file_destroy(rfs4_entry_t u_entry)
2446 {
2447 rfs4_file_t *fp = (rfs4_file_t *)u_entry;
2448
2449 list_destroy(&fp->rf_delegstatelist);
2450
2451 if (fp->rf_filehandle.nfs_fh4_val)
2452 kmem_free(fp->rf_filehandle.nfs_fh4_val,
2453 fp->rf_filehandle.nfs_fh4_len);
2454 cv_destroy(fp->rf_dinfo.rd_recall_cv);
2455 if (fp->rf_vp) {
2456 vnode_t *vp = fp->rf_vp;
2457
2458 mutex_enter(&vp->v_vsd_lock);
2459 (void) vsd_set(vp, nfs4_srv_vkey, NULL);
2460 mutex_exit(&vp->v_vsd_lock);
2461 VN_RELE(vp);
2462 fp->rf_vp = NULL;
2463 }
2464 rw_destroy(&fp->rf_file_rwlock);
2465 }
2466
2467 /*
2468 * Used to unlock the underlying dbe struct only
2469 */
2470 void
2471 rfs4_file_rele(rfs4_file_t *fp)
2472 {
2473 rfs4_dbe_rele(fp->rf_dbe);
2474 }
2475
2476 typedef struct {
2477 vnode_t *vp;
2478 nfs_fh4 *fh;
2479 } rfs4_fcreate_arg;
2480
2481 static bool_t
2482 rfs4_file_create(rfs4_entry_t u_entry, void *arg)
2483 {
2484 rfs4_file_t *fp = (rfs4_file_t *)u_entry;
2485 rfs4_fcreate_arg *ap = (rfs4_fcreate_arg *)arg;
2486 vnode_t *vp = ap->vp;
2487 nfs_fh4 *fh = ap->fh;
2488
2489 VN_HOLD(vp);
2490
2491 fp->rf_filehandle.nfs_fh4_len = 0;
2492 fp->rf_filehandle.nfs_fh4_val = NULL;
2493 ASSERT(fh && fh->nfs_fh4_len);
2494 if (fh && fh->nfs_fh4_len) {
2495 fp->rf_filehandle.nfs_fh4_val =
2496 kmem_alloc(fh->nfs_fh4_len, KM_SLEEP);
2497 nfs_fh4_copy(fh, &fp->rf_filehandle);
2498 }
2499 fp->rf_vp = vp;
2500
2501 list_create(&fp->rf_delegstatelist, sizeof (rfs4_deleg_state_t),
2502 offsetof(rfs4_deleg_state_t, rds_node));
2503
2504 fp->rf_share_deny = fp->rf_share_access = fp->rf_access_read = 0;
2505 fp->rf_access_write = fp->rf_deny_read = fp->rf_deny_write = 0;
2506
2507 mutex_init(fp->rf_dinfo.rd_recall_lock, NULL, MUTEX_DEFAULT, NULL);
2508 cv_init(fp->rf_dinfo.rd_recall_cv, NULL, CV_DEFAULT, NULL);
2509
2510 fp->rf_dinfo.rd_dtype = OPEN_DELEGATE_NONE;
2511
2512 rw_init(&fp->rf_file_rwlock, NULL, RW_DEFAULT, NULL);
2513
2514 mutex_enter(&vp->v_vsd_lock);
2515 VERIFY(vsd_set(vp, nfs4_srv_vkey, (void *)fp) == 0);
2516 mutex_exit(&vp->v_vsd_lock);
2517
2518 return (TRUE);
2519 }
2520
2521 rfs4_file_t *
2522 rfs4_findfile(vnode_t *vp, nfs_fh4 *fh, bool_t *create)
2523 {
2524 rfs4_file_t *fp;
2525 rfs4_fcreate_arg arg;
2526 nfs4_srv_t *nsrv4 = nfs4_get_srv();
2527
2528 arg.vp = vp;
2529 arg.fh = fh;
2530
2531 if (*create == TRUE)
2532 /* CSTYLED */
2533 fp = (rfs4_file_t *)rfs4_dbsearch(nsrv4->rfs4_file_idx, vp, create,
2534 &arg, RFS4_DBS_VALID);
2535 else {
2536 mutex_enter(&vp->v_vsd_lock);
2537 fp = (rfs4_file_t *)vsd_get(vp, nfs4_srv_vkey);
2538 if (fp) {
2539 rfs4_dbe_lock(fp->rf_dbe);
2540 if (rfs4_dbe_is_invalid(fp->rf_dbe) ||
2541 (rfs4_dbe_refcnt(fp->rf_dbe) == 0)) {
2542 rfs4_dbe_unlock(fp->rf_dbe);
2543 fp = NULL;
2544 } else {
2545 rfs4_dbe_hold(fp->rf_dbe);
2546 rfs4_dbe_unlock(fp->rf_dbe);
2547 }
2548 }
2549 mutex_exit(&vp->v_vsd_lock);
2550 }
2551 return (fp);
2552 }
2553
2554 /*
2555 * Find a file in the db and once it is located, take the rw lock.
2556 * Need to check the vnode pointer and if it does not exist (it was
2557 * removed between the db location and check) redo the find. This
2558 * assumes that a file struct that has a NULL vnode pointer is marked
2559 * at 'invalid' and will not be found in the db the second time
2560 * around.
2561 */
2562 rfs4_file_t *
2563 rfs4_findfile_withlock(vnode_t *vp, nfs_fh4 *fh, bool_t *create)
2564 {
2565 rfs4_file_t *fp;
2566 rfs4_fcreate_arg arg;
2567 bool_t screate = *create;
2568 nfs4_srv_t *nsrv4 = nfs4_get_srv();
2569
2570 if (screate == FALSE) {
2571 mutex_enter(&vp->v_vsd_lock);
2572 fp = (rfs4_file_t *)vsd_get(vp, nfs4_srv_vkey);
2573 if (fp) {
2574 rfs4_dbe_lock(fp->rf_dbe);
2575 if (rfs4_dbe_is_invalid(fp->rf_dbe) ||
2576 (rfs4_dbe_refcnt(fp->rf_dbe) == 0)) {
2577 rfs4_dbe_unlock(fp->rf_dbe);
2578 mutex_exit(&vp->v_vsd_lock);
2579 fp = NULL;
2580 } else {
2581 rfs4_dbe_hold(fp->rf_dbe);
2582 rfs4_dbe_unlock(fp->rf_dbe);
2583 mutex_exit(&vp->v_vsd_lock);
2584 rw_enter(&fp->rf_file_rwlock, RW_WRITER);
2585 if (fp->rf_vp == NULL) {
2586 rw_exit(&fp->rf_file_rwlock);
2587 rfs4_file_rele(fp);
2588 fp = NULL;
2589 }
2590 }
2591 } else {
2592 mutex_exit(&vp->v_vsd_lock);
2593 }
2594 } else {
2595 retry:
2596 arg.vp = vp;
2597 arg.fh = fh;
2598
2599 fp = (rfs4_file_t *)rfs4_dbsearch(nsrv4->rfs4_file_idx, vp,
2600 create, &arg, RFS4_DBS_VALID);
2601 if (fp != NULL) {
2602 rw_enter(&fp->rf_file_rwlock, RW_WRITER);
2603 if (fp->rf_vp == NULL) {
2604 rw_exit(&fp->rf_file_rwlock);
2605 rfs4_file_rele(fp);
2606 *create = screate;
2607 goto retry;
2608 }
2609 }
2610 }
2611
2612 return (fp);
2613 }
2614
2615 static uint32_t
2616 lo_state_hash(void *key)
2617 {
2618 stateid_t *id = key;
2619
2620 return (id->bits.ident+id->bits.pid);
2621 }
2622
2623 static bool_t
2624 lo_state_compare(rfs4_entry_t u_entry, void *key)
2625 {
2626 rfs4_lo_state_t *lsp = (rfs4_lo_state_t *)u_entry;
2627 stateid_t *id = key;
2628 bool_t rc;
2629
2630 rc = (lsp->rls_lockid.bits.boottime == id->bits.boottime &&
2631 lsp->rls_lockid.bits.type == id->bits.type &&
2632 lsp->rls_lockid.bits.ident == id->bits.ident &&
2633 lsp->rls_lockid.bits.pid == id->bits.pid);
2634
2635 return (rc);
2636 }
2637
2638 static void *
2639 lo_state_mkkey(rfs4_entry_t u_entry)
2640 {
2641 rfs4_lo_state_t *lsp = (rfs4_lo_state_t *)u_entry;
2642
2643 return (&lsp->rls_lockid);
2644 }
2645
2646 static bool_t
2647 rfs4_lo_state_expiry(rfs4_entry_t u_entry)
2648 {
2649 rfs4_lo_state_t *lsp = (rfs4_lo_state_t *)u_entry;
2650
2651 if (rfs4_dbe_is_invalid(lsp->rls_dbe))
2652 return (TRUE);
2653 if (lsp->rls_state->rs_closed)
2654 return (TRUE);
2655 return ((gethrestime_sec() -
2656 lsp->rls_state->rs_owner->ro_client->rc_last_access
2657 > rfs4_lease_time));
2658 }
2659
2660 static void
2661 rfs4_lo_state_destroy(rfs4_entry_t u_entry)
2662 {
2663 rfs4_lo_state_t *lsp = (rfs4_lo_state_t *)u_entry;
2664
2665 rfs4_dbe_lock(lsp->rls_state->rs_dbe);
2666 list_remove(&lsp->rls_state->rs_lostatelist, lsp);
2667 rfs4_dbe_unlock(lsp->rls_state->rs_dbe);
2668
2669 rfs4_sw_destroy(&lsp->rls_sw);
2670
2671 /* Make sure to release the file locks */
2672 if (lsp->rls_locks_cleaned == FALSE) {
2673 lsp->rls_locks_cleaned = TRUE;
2674 if (lsp->rls_locker->rl_client->rc_sysidt != LM_NOSYSID) {
2675 /* Is the PxFS kernel module loaded? */
2676 if (lm_remove_file_locks != NULL) {
2677 int new_sysid;
2678
2679 /* Encode the cluster nodeid in new sysid */
2680 new_sysid =
2681 lsp->rls_locker->rl_client->rc_sysidt;
2682 lm_set_nlmid_flk(&new_sysid);
2683
2684 /*
2685 * This PxFS routine removes file locks for a
2686 * client over all nodes of a cluster.
2687 */
2688 DTRACE_PROBE1(nfss_i_clust_rm_lck,
2689 int, new_sysid);
2690 (*lm_remove_file_locks)(new_sysid);
2691 } else {
2692 (void) cleanlocks(
2693 lsp->rls_state->rs_finfo->rf_vp,
2694 lsp->rls_locker->rl_pid,
2695 lsp->rls_locker->rl_client->rc_sysidt);
2696 }
2697 }
2698 }
2699
2700 /* Free the last reply for this state */
2701 rfs4_free_reply(&lsp->rls_reply);
2702
2703 rfs4_lockowner_rele(lsp->rls_locker);
2704 lsp->rls_locker = NULL;
2705
2706 rfs4_state_rele_nounlock(lsp->rls_state);
2707 lsp->rls_state = NULL;
2708 }
2709
2710 static bool_t
2711 rfs4_lo_state_create(rfs4_entry_t u_entry, void *arg)
2712 {
2713 rfs4_lo_state_t *lsp = (rfs4_lo_state_t *)u_entry;
2714 rfs4_lo_state_t *argp = (rfs4_lo_state_t *)arg;
2715 rfs4_lockowner_t *lo = argp->rls_locker;
2716 rfs4_state_t *sp = argp->rls_state;
2717
2718 lsp->rls_state = sp;
2719
2720 lsp->rls_lockid = sp->rs_stateid;
2721 lsp->rls_lockid.bits.type = LOCKID;
2722 lsp->rls_lockid.bits.chgseq = 0;
2723 lsp->rls_lockid.bits.pid = lo->rl_pid;
2724
2725 lsp->rls_locks_cleaned = FALSE;
2726 lsp->rls_lock_completed = FALSE;
2727
2728 rfs4_sw_init(&lsp->rls_sw);
2729
2730 /* Attached the supplied lock owner */
2731 rfs4_dbe_hold(lo->rl_dbe);
2732 lsp->rls_locker = lo;
2733
2734 rfs4_dbe_lock(sp->rs_dbe);
2735 list_insert_tail(&sp->rs_lostatelist, lsp);
2736 rfs4_dbe_hold(sp->rs_dbe);
2737 rfs4_dbe_unlock(sp->rs_dbe);
2738
2739 return (TRUE);
2740 }
2741
2742 void
2743 rfs4_lo_state_rele(rfs4_lo_state_t *lsp, bool_t unlock_fp)
2744 {
2745 if (unlock_fp == TRUE)
2746 rw_exit(&lsp->rls_state->rs_finfo->rf_file_rwlock);
2747 rfs4_dbe_rele(lsp->rls_dbe);
2748 }
2749
2750 static rfs4_lo_state_t *
2751 rfs4_findlo_state(stateid_t *id, bool_t lock_fp)
2752 {
2753 rfs4_lo_state_t *lsp;
2754 bool_t create = FALSE;
2755 nfs4_srv_t *nsrv4 = nfs4_get_srv();
2756
2757 lsp = (rfs4_lo_state_t *)rfs4_dbsearch(nsrv4->rfs4_lo_state_idx, id,
2758 &create, NULL, RFS4_DBS_VALID);
2759 if (lock_fp == TRUE && lsp != NULL)
2760 rw_enter(&lsp->rls_state->rs_finfo->rf_file_rwlock, RW_READER);
2761
2762 return (lsp);
2763 }
2764
2765
2766 static uint32_t
2767 lo_state_lo_hash(void *key)
2768 {
2769 rfs4_lo_state_t *lsp = key;
2770
2771 return (ADDRHASH(lsp->rls_locker) ^ ADDRHASH(lsp->rls_state));
2772 }
2773
2774 static bool_t
2775 lo_state_lo_compare(rfs4_entry_t u_entry, void *key)
2776 {
2777 rfs4_lo_state_t *lsp = (rfs4_lo_state_t *)u_entry;
2778 rfs4_lo_state_t *keyp = key;
2779
2780 return (keyp->rls_locker == lsp->rls_locker &&
2781 keyp->rls_state == lsp->rls_state);
2782 }
2783
2784 static void *
2785 lo_state_lo_mkkey(rfs4_entry_t u_entry)
2786 {
2787 return (u_entry);
2788 }
2789
2790 rfs4_lo_state_t *
2791 rfs4_findlo_state_by_owner(rfs4_lockowner_t *lo, rfs4_state_t *sp,
2792 bool_t *create)
2793 {
2794 rfs4_lo_state_t *lsp;
2795 rfs4_lo_state_t arg;
2796 nfs4_srv_t *nsrv4 = nfs4_get_srv();
2797
2798 arg.rls_locker = lo;
2799 arg.rls_state = sp;
2800
2801 lsp = (rfs4_lo_state_t *)rfs4_dbsearch(nsrv4->rfs4_lo_state_owner_idx,
2802 &arg, create, &arg, RFS4_DBS_VALID);
2803
2804 return (lsp);
2805 }
2806
2807 static stateid_t
2808 get_stateid(id_t eid)
2809 {
2810 stateid_t id;
2811 nfs4_srv_t *nsrv4;
2812
2813 nsrv4 = nfs4_get_srv();
2814
2815 id.bits.boottime = nsrv4->rfs4_start_time;
2816 id.bits.ident = eid;
2817 id.bits.chgseq = 0;
2818 id.bits.type = 0;
2819 id.bits.pid = 0;
2820
2821 /*
2822 * If we are booted as a cluster node, embed our nodeid.
2823 * We've already done sanity checks in rfs4_client_create() so no
2824 * need to repeat them here.
2825 */
2826 id.bits.clnodeid = (cluster_bootflags & CLUSTER_BOOTED) ?
2827 clconf_get_nodeid() : 0;
2828
2829 return (id);
2830 }
2831
2832 /*
2833 * For use only when booted as a cluster node.
2834 * Returns TRUE if the embedded nodeid indicates that this stateid was
2835 * generated on another node.
2836 */
2837 static int
2838 foreign_stateid(stateid_t *id)
2839 {
2840 ASSERT(cluster_bootflags & CLUSTER_BOOTED);
2841 return (id->bits.clnodeid != (uint32_t)clconf_get_nodeid());
2842 }
2843
2844 /*
2845 * For use only when booted as a cluster node.
2846 * Returns TRUE if the embedded nodeid indicates that this clientid was
2847 * generated on another node.
2848 */
2849 static int
2850 foreign_clientid(cid *cidp)
2851 {
2852 ASSERT(cluster_bootflags & CLUSTER_BOOTED);
2853 return (cidp->impl_id.c_id >> CLUSTER_NODEID_SHIFT !=
2854 (uint32_t)clconf_get_nodeid());
2855 }
2856
2857 /*
2858 * For use only when booted as a cluster node.
2859 * Embed our cluster nodeid into the clientid.
2860 */
2861 static void
2862 embed_nodeid(cid *cidp)
2863 {
2864 int clnodeid;
2865 /*
2866 * Currently, our state tables are small enough that their
2867 * ids will leave enough bits free for the nodeid. If the
2868 * tables become larger, we mustn't overwrite the id.
2869 * Equally, we only have room for so many bits of nodeid, so
2870 * must check that too.
2871 */
2872 ASSERT(cluster_bootflags & CLUSTER_BOOTED);
2873 ASSERT(cidp->impl_id.c_id >> CLUSTER_NODEID_SHIFT == 0);
2874 clnodeid = clconf_get_nodeid();
2875 ASSERT(clnodeid <= CLUSTER_MAX_NODEID);
2876 ASSERT(clnodeid != NODEID_UNKNOWN);
2877 cidp->impl_id.c_id |= (clnodeid << CLUSTER_NODEID_SHIFT);
2878 }
2879
2880 static uint32_t
2881 state_hash(void *key)
2882 {
2883 stateid_t *ip = (stateid_t *)key;
2884
2885 return (ip->bits.ident);
2886 }
2887
2888 static bool_t
2889 state_compare(rfs4_entry_t u_entry, void *key)
2890 {
2891 rfs4_state_t *sp = (rfs4_state_t *)u_entry;
2892 stateid_t *id = (stateid_t *)key;
2893 bool_t rc;
2894
2895 rc = (sp->rs_stateid.bits.boottime == id->bits.boottime &&
2896 sp->rs_stateid.bits.ident == id->bits.ident);
2897
2898 return (rc);
2899 }
2900
2901 static void *
2902 state_mkkey(rfs4_entry_t u_entry)
2903 {
2904 rfs4_state_t *sp = (rfs4_state_t *)u_entry;
2905
2906 return (&sp->rs_stateid);
2907 }
2908
2909 static void
2910 rfs4_state_destroy(rfs4_entry_t u_entry)
2911 {
2912 rfs4_state_t *sp = (rfs4_state_t *)u_entry;
2913
2914 /* remove from openowner list */
2915 rfs4_dbe_lock(sp->rs_owner->ro_dbe);
2916 list_remove(&sp->rs_owner->ro_statelist, sp);
2917 rfs4_dbe_unlock(sp->rs_owner->ro_dbe);
2918
2919 list_destroy(&sp->rs_lostatelist);
2920
2921 /* release any share locks for this stateid if it's still open */
2922 if (!sp->rs_closed) {
2923 rfs4_dbe_lock(sp->rs_dbe);
2924 (void) rfs4_unshare(sp);
2925 rfs4_dbe_unlock(sp->rs_dbe);
2926 }
2927
2928 /* Were done with the file */
2929 rfs4_file_rele(sp->rs_finfo);
2930 sp->rs_finfo = NULL;
2931
2932 /* And now with the openowner */
2933 rfs4_openowner_rele(sp->rs_owner);
2934 sp->rs_owner = NULL;
2935 }
2936
2937 static void
2938 rfs4_state_rele_nounlock(rfs4_state_t *sp)
2939 {
2940 rfs4_dbe_rele(sp->rs_dbe);
2941 }
2942
2943 void
2944 rfs4_state_rele(rfs4_state_t *sp)
2945 {
2946 rw_exit(&sp->rs_finfo->rf_file_rwlock);
2947 rfs4_dbe_rele(sp->rs_dbe);
2948 }
2949
2950 static uint32_t
2951 deleg_hash(void *key)
2952 {
2953 rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)key;
2954
2955 return (ADDRHASH(dsp->rds_client) ^ ADDRHASH(dsp->rds_finfo));
2956 }
2957
2958 static bool_t
2959 deleg_compare(rfs4_entry_t u_entry, void *key)
2960 {
2961 rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
2962 rfs4_deleg_state_t *kdsp = (rfs4_deleg_state_t *)key;
2963
2964 return (dsp->rds_client == kdsp->rds_client &&
2965 dsp->rds_finfo == kdsp->rds_finfo);
2966 }
2967
2968 static void *
2969 deleg_mkkey(rfs4_entry_t u_entry)
2970 {
2971 return (u_entry);
2972 }
2973
2974 static uint32_t
2975 deleg_state_hash(void *key)
2976 {
2977 stateid_t *ip = (stateid_t *)key;
2978
2979 return (ip->bits.ident);
2980 }
2981
2982 static bool_t
2983 deleg_state_compare(rfs4_entry_t u_entry, void *key)
2984 {
2985 rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
2986 stateid_t *id = (stateid_t *)key;
2987 bool_t rc;
2988
2989 if (id->bits.type != DELEGID)
2990 return (FALSE);
2991
2992 rc = (dsp->rds_delegid.bits.boottime == id->bits.boottime &&
2993 dsp->rds_delegid.bits.ident == id->bits.ident);
2994
2995 return (rc);
2996 }
2997
2998 static void *
2999 deleg_state_mkkey(rfs4_entry_t u_entry)
3000 {
3001 rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
3002
3003 return (&dsp->rds_delegid);
3004 }
3005
3006 static bool_t
3007 rfs4_deleg_state_expiry(rfs4_entry_t u_entry)
3008 {
3009 rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
3010
3011 if (rfs4_dbe_is_invalid(dsp->rds_dbe))
3012 return (TRUE);
3013
3014 if (dsp->rds_dtype == OPEN_DELEGATE_NONE)
3015 return (TRUE);
3016
3017 if ((gethrestime_sec() - dsp->rds_client->rc_last_access
3018 > rfs4_lease_time)) {
3019 rfs4_dbe_invalidate(dsp->rds_dbe);
3020 return (TRUE);
3021 }
3022
3023 return (FALSE);
3024 }
3025
3026 static bool_t
3027 rfs4_deleg_state_create(rfs4_entry_t u_entry, void *argp)
3028 {
3029 rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
3030 rfs4_file_t *fp = ((rfs4_deleg_state_t *)argp)->rds_finfo;
3031 rfs4_client_t *cp = ((rfs4_deleg_state_t *)argp)->rds_client;
3032
3033 rfs4_dbe_hold(fp->rf_dbe);
3034 rfs4_dbe_hold(cp->rc_dbe);
3035
3036 dsp->rds_delegid = get_stateid(rfs4_dbe_getid(dsp->rds_dbe));
3037 dsp->rds_delegid.bits.type = DELEGID;
3038 dsp->rds_finfo = fp;
3039 dsp->rds_client = cp;
3040 dsp->rds_dtype = OPEN_DELEGATE_NONE;
3041
3042 dsp->rds_time_granted = gethrestime_sec(); /* observability */
3043 dsp->rds_time_revoked = 0;
3044
3045 list_link_init(&dsp->rds_node);
3046
3047 return (TRUE);
3048 }
3049
3050 static void
3051 rfs4_deleg_state_destroy(rfs4_entry_t u_entry)
3052 {
3053 rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
3054
3055 /* return delegation if necessary */
3056 rfs4_return_deleg(dsp, FALSE);
3057
3058 /* Were done with the file */
3059 rfs4_file_rele(dsp->rds_finfo);
3060 dsp->rds_finfo = NULL;
3061
3062 /* And now with the openowner */
3063 rfs4_client_rele(dsp->rds_client);
3064 dsp->rds_client = NULL;
3065 }
3066
3067 rfs4_deleg_state_t *
3068 rfs4_finddeleg(rfs4_state_t *sp, bool_t *create)
3069 {
3070 rfs4_deleg_state_t ds, *dsp;
3071 nfs4_srv_t *nsrv4 = nfs4_get_srv();
3072
3073 ds.rds_client = sp->rs_owner->ro_client;
3074 ds.rds_finfo = sp->rs_finfo;
3075
3076 dsp = (rfs4_deleg_state_t *)rfs4_dbsearch(nsrv4->rfs4_deleg_idx, &ds,
3077 create, &ds, RFS4_DBS_VALID);
3078
3079 return (dsp);
3080 }
3081
3082 rfs4_deleg_state_t *
3083 rfs4_finddelegstate(stateid_t *id)
3084 {
3085 rfs4_deleg_state_t *dsp;
3086 bool_t create = FALSE;
3087 nfs4_srv_t *nsrv4 = nfs4_get_srv();
3088
3089 dsp = (rfs4_deleg_state_t *)rfs4_dbsearch(nsrv4->rfs4_deleg_state_idx,
3090 id, &create, NULL, RFS4_DBS_VALID);
3091
3092 return (dsp);
3093 }
3094
3095 void
3096 rfs4_deleg_state_rele(rfs4_deleg_state_t *dsp)
3097 {
3098 rfs4_dbe_rele(dsp->rds_dbe);
3099 }
3100
3101 void
3102 rfs4_update_lock_sequence(rfs4_lo_state_t *lsp)
3103 {
3104
3105 rfs4_dbe_lock(lsp->rls_dbe);
3106
3107 /*
3108 * If we are skipping sequence id checking, this means that
3109 * this is the first lock request and therefore the sequence
3110 * id does not need to be updated. This only happens on the
3111 * first lock request for a lockowner
3112 */
3113 if (!lsp->rls_skip_seqid_check)
3114 lsp->rls_seqid++;
3115
3116 rfs4_dbe_unlock(lsp->rls_dbe);
3117 }
3118
3119 void
3120 rfs4_update_lock_resp(rfs4_lo_state_t *lsp, nfs_resop4 *resp)
3121 {
3122
3123 rfs4_dbe_lock(lsp->rls_dbe);
3124
3125 rfs4_free_reply(&lsp->rls_reply);
3126
3127 rfs4_copy_reply(&lsp->rls_reply, resp);
3128
3129 rfs4_dbe_unlock(lsp->rls_dbe);
3130 }
3131
3132 void
3133 rfs4_free_opens(rfs4_openowner_t *oo, bool_t invalidate,
3134 bool_t close_of_client)
3135 {
3136 rfs4_state_t *sp;
3137
3138 rfs4_dbe_lock(oo->ro_dbe);
3139
3140 for (sp = list_head(&oo->ro_statelist); sp != NULL;
3141 sp = list_next(&oo->ro_statelist, sp)) {
3142 rfs4_state_close(sp, FALSE, close_of_client, CRED());
3143 if (invalidate == TRUE)
3144 rfs4_dbe_invalidate(sp->rs_dbe);
3145 }
3146
3147 rfs4_dbe_invalidate(oo->ro_dbe);
3148 rfs4_dbe_unlock(oo->ro_dbe);
3149 }
3150
3151 static uint32_t
3152 state_owner_file_hash(void *key)
3153 {
3154 rfs4_state_t *sp = key;
3155
3156 return (ADDRHASH(sp->rs_owner) ^ ADDRHASH(sp->rs_finfo));
3157 }
3158
3159 static bool_t
3160 state_owner_file_compare(rfs4_entry_t u_entry, void *key)
3161 {
3162 rfs4_state_t *sp = (rfs4_state_t *)u_entry;
3163 rfs4_state_t *arg = key;
3164
3165 if (sp->rs_closed == TRUE)
3166 return (FALSE);
3167
3168 return (arg->rs_owner == sp->rs_owner && arg->rs_finfo == sp->rs_finfo);
3169 }
3170
3171 static void *
3172 state_owner_file_mkkey(rfs4_entry_t u_entry)
3173 {
3174 return (u_entry);
3175 }
3176
3177 static uint32_t
3178 state_file_hash(void *key)
3179 {
3180 return (ADDRHASH(key));
3181 }
3182
3183 static bool_t
3184 state_file_compare(rfs4_entry_t u_entry, void *key)
3185 {
3186 rfs4_state_t *sp = (rfs4_state_t *)u_entry;
3187 rfs4_file_t *fp = key;
3188
3189 if (sp->rs_closed == TRUE)
3190 return (FALSE);
3191
3192 return (fp == sp->rs_finfo);
3193 }
3194
3195 static void *
3196 state_file_mkkey(rfs4_entry_t u_entry)
3197 {
3198 rfs4_state_t *sp = (rfs4_state_t *)u_entry;
3199
3200 return (sp->rs_finfo);
3201 }
3202
3203 rfs4_state_t *
3204 rfs4_findstate_by_owner_file(rfs4_openowner_t *oo, rfs4_file_t *fp,
3205 bool_t *create)
3206 {
3207 rfs4_state_t *sp;
3208 rfs4_state_t key;
3209 nfs4_srv_t *nsrv4 = nfs4_get_srv();
3210
3211 key.rs_owner = oo;
3212 key.rs_finfo = fp;
3213
3214 sp = (rfs4_state_t *)rfs4_dbsearch(nsrv4->rfs4_state_owner_file_idx,
3215 &key, create, &key, RFS4_DBS_VALID);
3216
3217 return (sp);
3218 }
3219
3220 /* This returns ANY state struct that refers to this file */
3221 static rfs4_state_t *
3222 rfs4_findstate_by_file(rfs4_file_t *fp)
3223 {
3224 bool_t create = FALSE;
3225 nfs4_srv_t *nsrv4 = nfs4_get_srv();
3226
3227 return ((rfs4_state_t *)rfs4_dbsearch(nsrv4->rfs4_state_file_idx, fp,
3228 &create, fp, RFS4_DBS_VALID));
3229 }
3230
3231 static bool_t
3232 rfs4_state_expiry(rfs4_entry_t u_entry)
3233 {
3234 rfs4_state_t *sp = (rfs4_state_t *)u_entry;
3235
3236 if (rfs4_dbe_is_invalid(sp->rs_dbe))
3237 return (TRUE);
3238
3239 if (sp->rs_closed == TRUE &&
3240 ((gethrestime_sec() - rfs4_dbe_get_timerele(sp->rs_dbe))
3241 > rfs4_lease_time))
3242 return (TRUE);
3243
3244 return ((gethrestime_sec() - sp->rs_owner->ro_client->rc_last_access
3245 > rfs4_lease_time));
3246 }
3247
3248 static bool_t
3249 rfs4_state_create(rfs4_entry_t u_entry, void *argp)
3250 {
3251 rfs4_state_t *sp = (rfs4_state_t *)u_entry;
3252 rfs4_file_t *fp = ((rfs4_state_t *)argp)->rs_finfo;
3253 rfs4_openowner_t *oo = ((rfs4_state_t *)argp)->rs_owner;
3254
3255 rfs4_dbe_hold(fp->rf_dbe);
3256 rfs4_dbe_hold(oo->ro_dbe);
3257 sp->rs_stateid = get_stateid(rfs4_dbe_getid(sp->rs_dbe));
3258 sp->rs_stateid.bits.type = OPENID;
3259 sp->rs_owner = oo;
3260 sp->rs_finfo = fp;
3261
3262 list_create(&sp->rs_lostatelist, sizeof (rfs4_lo_state_t),
3263 offsetof(rfs4_lo_state_t, rls_node));
3264
3265 /* Insert state on per open owner's list */
3266 rfs4_dbe_lock(oo->ro_dbe);
3267 list_insert_tail(&oo->ro_statelist, sp);
3268 rfs4_dbe_unlock(oo->ro_dbe);
3269
3270 return (TRUE);
3271 }
3272
3273 static rfs4_state_t *
3274 rfs4_findstate(stateid_t *id, rfs4_dbsearch_type_t find_invalid, bool_t lock_fp)
3275 {
3276 rfs4_state_t *sp;
3277 bool_t create = FALSE;
3278 nfs4_srv_t *nsrv4 = nfs4_get_srv();
3279
3280 sp = (rfs4_state_t *)rfs4_dbsearch(nsrv4->rfs4_state_idx, id,
3281 &create, NULL, find_invalid);
3282 if (lock_fp == TRUE && sp != NULL)
3283 rw_enter(&sp->rs_finfo->rf_file_rwlock, RW_READER);
3284
3285 return (sp);
3286 }
3287
3288 void
3289 rfs4_state_close(rfs4_state_t *sp, bool_t lock_held, bool_t close_of_client,
3290 cred_t *cr)
3291 {
3292 /* Remove the associated lo_state owners */
3293 if (!lock_held)
3294 rfs4_dbe_lock(sp->rs_dbe);
3295
3296 /*
3297 * If refcnt == 0, the dbe is about to be destroyed.
3298 * lock state will be released by the reaper thread.
3299 */
3300
3301 if (rfs4_dbe_refcnt(sp->rs_dbe) > 0) {
3302 if (sp->rs_closed == FALSE) {
3303 rfs4_release_share_lock_state(sp, cr, close_of_client);
3304 sp->rs_closed = TRUE;
3305 }
3306 }
3307
3308 if (!lock_held)
3309 rfs4_dbe_unlock(sp->rs_dbe);
3310 }
3311
3312 /*
3313 * Remove all state associated with the given client.
3314 */
3315 void
3316 rfs4_client_state_remove(rfs4_client_t *cp)
3317 {
3318 rfs4_openowner_t *oo;
3319
3320 rfs4_dbe_lock(cp->rc_dbe);
3321
3322 for (oo = list_head(&cp->rc_openownerlist); oo != NULL;
3323 oo = list_next(&cp->rc_openownerlist, oo)) {
3324 rfs4_free_opens(oo, TRUE, TRUE);
3325 }
3326
3327 rfs4_dbe_unlock(cp->rc_dbe);
3328 }
3329
3330 void
3331 rfs4_client_close(rfs4_client_t *cp)
3332 {
3333 /* Mark client as going away. */
3334 rfs4_dbe_lock(cp->rc_dbe);
3335 rfs4_dbe_invalidate(cp->rc_dbe);
3336 rfs4_dbe_unlock(cp->rc_dbe);
3337
3338 rfs4_client_state_remove(cp);
3339
3340 /* Release the client */
3341 rfs4_client_rele(cp);
3342 }
3343
3344 nfsstat4
3345 rfs4_check_clientid(clientid4 *cp, int setclid_confirm)
3346 {
3347 cid *cidp = (cid *) cp;
3348 nfs4_srv_t *nsrv4;
3349
3350 nsrv4 = nfs4_get_srv();
3351
3352 /*
3353 * If we are booted as a cluster node, check the embedded nodeid.
3354 * If it indicates that this clientid was generated on another node,
3355 * inform the client accordingly.
3356 */
3357 if (cluster_bootflags & CLUSTER_BOOTED && foreign_clientid(cidp))
3358 return (NFS4ERR_STALE_CLIENTID);
3359
3360 /*
3361 * If the server start time matches the time provided
3362 * by the client (via the clientid) and this is NOT a
3363 * setclientid_confirm then return EXPIRED.
3364 */
3365 if (!setclid_confirm &&
3366 cidp->impl_id.start_time == nsrv4->rfs4_start_time)
3367 return (NFS4ERR_EXPIRED);
3368
3369 return (NFS4ERR_STALE_CLIENTID);
3370 }
3371
3372 /*
3373 * This is used when a stateid has not been found amongst the
3374 * current server's state. Check the stateid to see if it
3375 * was from this server instantiation or not.
3376 */
3377 static nfsstat4
3378 what_stateid_error(stateid_t *id, stateid_type_t type)
3379 {
3380 nfs4_srv_t *nsrv4;
3381
3382 nsrv4 = nfs4_get_srv();
3383
3384 /* If we are booted as a cluster node, was stateid locally generated? */
3385 if ((cluster_bootflags & CLUSTER_BOOTED) && foreign_stateid(id))
3386 return (NFS4ERR_STALE_STATEID);
3387
3388 /* If types don't match then no use checking further */
3389 if (type != id->bits.type)
3390 return (NFS4ERR_BAD_STATEID);
3391
3392 /* From a different server instantiation, return STALE */
3393 if (id->bits.boottime != nsrv4->rfs4_start_time)
3394 return (NFS4ERR_STALE_STATEID);
3395
3396 /*
3397 * From this server but the state is most likely beyond lease
3398 * timeout: return NFS4ERR_EXPIRED. However, there is the
3399 * case of a delegation stateid. For delegations, there is a
3400 * case where the state can be removed without the client's
3401 * knowledge/consent: revocation. In the case of delegation
3402 * revocation, the delegation state will be removed and will
3403 * not be found. If the client does something like a
3404 * DELEGRETURN or even a READ/WRITE with a delegatoin stateid
3405 * that has been revoked, the server should return BAD_STATEID
3406 * instead of the more common EXPIRED error.
3407 */
3408 if (id->bits.boottime == nsrv4->rfs4_start_time) {
3409 if (type == DELEGID)
3410 return (NFS4ERR_BAD_STATEID);
3411 else
3412 return (NFS4ERR_EXPIRED);
3413 }
3414
3415 return (NFS4ERR_BAD_STATEID);
3416 }
3417
3418 /*
3419 * Used later on to find the various state structs. When called from
3420 * rfs4_check_stateid()->rfs4_get_all_state(), no file struct lock is
3421 * taken (it is not needed) and helps on the read/write path with
3422 * respect to performance.
3423 */
3424 static nfsstat4
3425 rfs4_get_state_lockit(stateid4 *stateid, rfs4_state_t **spp,
3426 rfs4_dbsearch_type_t find_invalid, bool_t lock_fp)
3427 {
3428 stateid_t *id = (stateid_t *)stateid;
3429 rfs4_state_t *sp;
3430
3431 *spp = NULL;
3432
3433 /* If we are booted as a cluster node, was stateid locally generated? */
3434 if ((cluster_bootflags & CLUSTER_BOOTED) && foreign_stateid(id))
3435 return (NFS4ERR_STALE_STATEID);
3436
3437 sp = rfs4_findstate(id, find_invalid, lock_fp);
3438 if (sp == NULL) {
3439 return (what_stateid_error(id, OPENID));
3440 }
3441
3442 if (rfs4_lease_expired(sp->rs_owner->ro_client)) {
3443 if (lock_fp == TRUE)
3444 rfs4_state_rele(sp);
3445 else
3446 rfs4_state_rele_nounlock(sp);
3447 return (NFS4ERR_EXPIRED);
3448 }
3449
3450 *spp = sp;
3451
3452 return (NFS4_OK);
3453 }
3454
3455 nfsstat4
3456 rfs4_get_state(stateid4 *stateid, rfs4_state_t **spp,
3457 rfs4_dbsearch_type_t find_invalid)
3458 {
3459 return (rfs4_get_state_lockit(stateid, spp, find_invalid, TRUE));
3460 }
3461
3462 int
3463 rfs4_check_stateid_seqid(rfs4_state_t *sp, stateid4 *stateid)
3464 {
3465 stateid_t *id = (stateid_t *)stateid;
3466
3467 if (rfs4_lease_expired(sp->rs_owner->ro_client))
3468 return (NFS4_CHECK_STATEID_EXPIRED);
3469
3470 /* Stateid is some time in the future - that's bad */
3471 if (sp->rs_stateid.bits.chgseq < id->bits.chgseq)
3472 return (NFS4_CHECK_STATEID_BAD);
3473
3474 if (sp->rs_stateid.bits.chgseq == id->bits.chgseq + 1)
3475 return (NFS4_CHECK_STATEID_REPLAY);
3476
3477 /* Stateid is some time in the past - that's old */
3478 if (sp->rs_stateid.bits.chgseq > id->bits.chgseq)
3479 return (NFS4_CHECK_STATEID_OLD);
3480
3481 /* Caller needs to know about confirmation before closure */
3482 if (sp->rs_owner->ro_need_confirm)
3483 return (NFS4_CHECK_STATEID_UNCONFIRMED);
3484
3485 if (sp->rs_closed == TRUE)
3486 return (NFS4_CHECK_STATEID_CLOSED);
3487
3488 return (NFS4_CHECK_STATEID_OKAY);
3489 }
3490
3491 int
3492 rfs4_check_lo_stateid_seqid(rfs4_lo_state_t *lsp, stateid4 *stateid)
3493 {
3494 stateid_t *id = (stateid_t *)stateid;
3495
3496 if (rfs4_lease_expired(lsp->rls_state->rs_owner->ro_client))
3497 return (NFS4_CHECK_STATEID_EXPIRED);
3498
3499 /* Stateid is some time in the future - that's bad */
3500 if (lsp->rls_lockid.bits.chgseq < id->bits.chgseq)
3501 return (NFS4_CHECK_STATEID_BAD);
3502
3503 if (lsp->rls_lockid.bits.chgseq == id->bits.chgseq + 1)
3504 return (NFS4_CHECK_STATEID_REPLAY);
3505
3506 /* Stateid is some time in the past - that's old */
3507 if (lsp->rls_lockid.bits.chgseq > id->bits.chgseq)
3508 return (NFS4_CHECK_STATEID_OLD);
3509
3510 if (lsp->rls_state->rs_closed == TRUE)
3511 return (NFS4_CHECK_STATEID_CLOSED);
3512
3513 return (NFS4_CHECK_STATEID_OKAY);
3514 }
3515
3516 nfsstat4
3517 rfs4_get_deleg_state(stateid4 *stateid, rfs4_deleg_state_t **dspp)
3518 {
3519 stateid_t *id = (stateid_t *)stateid;
3520 rfs4_deleg_state_t *dsp;
3521
3522 *dspp = NULL;
3523
3524 /* If we are booted as a cluster node, was stateid locally generated? */
3525 if ((cluster_bootflags & CLUSTER_BOOTED) && foreign_stateid(id))
3526 return (NFS4ERR_STALE_STATEID);
3527
3528 dsp = rfs4_finddelegstate(id);
3529 if (dsp == NULL) {
3530 return (what_stateid_error(id, DELEGID));
3531 }
3532
3533 if (rfs4_lease_expired(dsp->rds_client)) {
3534 rfs4_deleg_state_rele(dsp);
3535 return (NFS4ERR_EXPIRED);
3536 }
3537
3538 *dspp = dsp;
3539
3540 return (NFS4_OK);
3541 }
3542
3543 nfsstat4
3544 rfs4_get_lo_state(stateid4 *stateid, rfs4_lo_state_t **lspp, bool_t lock_fp)
3545 {
3546 stateid_t *id = (stateid_t *)stateid;
3547 rfs4_lo_state_t *lsp;
3548
3549 *lspp = NULL;
3550
3551 /* If we are booted as a cluster node, was stateid locally generated? */
3552 if ((cluster_bootflags & CLUSTER_BOOTED) && foreign_stateid(id))
3553 return (NFS4ERR_STALE_STATEID);
3554
3555 lsp = rfs4_findlo_state(id, lock_fp);
3556 if (lsp == NULL) {
3557 return (what_stateid_error(id, LOCKID));
3558 }
3559
3560 if (rfs4_lease_expired(lsp->rls_state->rs_owner->ro_client)) {
3561 rfs4_lo_state_rele(lsp, lock_fp);
3562 return (NFS4ERR_EXPIRED);
3563 }
3564
3565 *lspp = lsp;
3566
3567 return (NFS4_OK);
3568 }
3569
3570 static nfsstat4
3571 rfs4_get_all_state(stateid4 *sid, rfs4_state_t **spp,
3572 rfs4_deleg_state_t **dspp, rfs4_lo_state_t **lspp)
3573 {
3574 rfs4_state_t *sp = NULL;
3575 rfs4_deleg_state_t *dsp = NULL;
3576 rfs4_lo_state_t *lsp = NULL;
3577 stateid_t *id;
3578 nfsstat4 status;
3579
3580 *spp = NULL; *dspp = NULL; *lspp = NULL;
3581
3582 id = (stateid_t *)sid;
3583 switch (id->bits.type) {
3584 case OPENID:
3585 status = rfs4_get_state_lockit(sid, &sp, FALSE, FALSE);
3586 break;
3587 case DELEGID:
3588 status = rfs4_get_deleg_state(sid, &dsp);
3589 break;
3590 case LOCKID:
3591 status = rfs4_get_lo_state(sid, &lsp, FALSE);
3592 if (status == NFS4_OK) {
3593 sp = lsp->rls_state;
3594 rfs4_dbe_hold(sp->rs_dbe);
3595 }
3596 break;
3597 default:
3598 status = NFS4ERR_BAD_STATEID;
3599 }
3600
3601 if (status == NFS4_OK) {
3602 *spp = sp;
3603 *dspp = dsp;
3604 *lspp = lsp;
3605 }
3606
3607 return (status);
3608 }
3609
3610 /*
3611 * Given the I/O mode (FREAD or FWRITE), this checks whether the
3612 * rfs4_state_t struct has access to do this operation and if so
3613 * return NFS4_OK; otherwise the proper NFSv4 error is returned.
3614 */
3615 nfsstat4
3616 rfs4_state_has_access(rfs4_state_t *sp, int mode, vnode_t *vp)
3617 {
3618 nfsstat4 stat = NFS4_OK;
3619 rfs4_file_t *fp;
3620 bool_t create = FALSE;
3621
3622 rfs4_dbe_lock(sp->rs_dbe);
3623 if (mode == FWRITE) {
3624 if (!(sp->rs_share_access & OPEN4_SHARE_ACCESS_WRITE)) {
3625 stat = NFS4ERR_OPENMODE;
3626 }
3627 } else if (mode == FREAD) {
3628 if (!(sp->rs_share_access & OPEN4_SHARE_ACCESS_READ)) {
3629 /*
3630 * If we have OPENed the file with DENYing access
3631 * to both READ and WRITE then no one else could
3632 * have OPENed the file, hence no conflicting READ
3633 * deny. This check is merely an optimization.
3634 */
3635 if (sp->rs_share_deny == OPEN4_SHARE_DENY_BOTH)
3636 goto out;
3637
3638 /* Check against file struct's DENY mode */
3639 fp = rfs4_findfile(vp, NULL, &create);
3640 if (fp != NULL) {
3641 int deny_read = 0;
3642 rfs4_dbe_lock(fp->rf_dbe);
3643 /*
3644 * Check if any other open owner has the file
3645 * OPENed with deny READ.
3646 */
3647 if (sp->rs_share_deny & OPEN4_SHARE_DENY_READ)
3648 deny_read = 1;
3649 ASSERT(fp->rf_deny_read >= deny_read);
3650 if (fp->rf_deny_read > deny_read)
3651 stat = NFS4ERR_OPENMODE;
3652 rfs4_dbe_unlock(fp->rf_dbe);
3653 rfs4_file_rele(fp);
3654 }
3655 }
3656 } else {
3657 /* Illegal I/O mode */
3658 stat = NFS4ERR_INVAL;
3659 }
3660 out:
3661 rfs4_dbe_unlock(sp->rs_dbe);
3662 return (stat);
3663 }
3664
3665 /*
3666 * Given the I/O mode (FREAD or FWRITE), the vnode, the stateid and whether
3667 * the file is being truncated, return NFS4_OK if allowed or appropriate
3668 * V4 error if not. Note NFS4ERR_DELAY will be returned and a recall on
3669 * the associated file will be done if the I/O is not consistent with any
3670 * delegation in effect on the file. Should be holding VOP_RWLOCK, either
3671 * as reader or writer as appropriate. rfs4_op_open will acquire the
3672 * VOP_RWLOCK as writer when setting up delegation. If the stateid is bad
3673 * this routine will return NFS4ERR_BAD_STATEID. In addition, through the
3674 * deleg parameter, we will return whether a write delegation is held by
3675 * the client associated with this stateid.
3676 * If the server instance associated with the relevant client is in its
3677 * grace period, return NFS4ERR_GRACE.
3678 */
3679
3680 nfsstat4
3681 rfs4_check_stateid(int mode, vnode_t *vp,
3682 stateid4 *stateid, bool_t trunc, bool_t *deleg,
3683 bool_t do_access, caller_context_t *ct)
3684 {
3685 rfs4_file_t *fp;
3686 bool_t create = FALSE;
3687 rfs4_state_t *sp;
3688 rfs4_deleg_state_t *dsp;
3689 rfs4_lo_state_t *lsp;
3690 stateid_t *id = (stateid_t *)stateid;
3691 nfsstat4 stat = NFS4_OK;
3692
3693 if (ct != NULL) {
3694 ct->cc_sysid = 0;
3695 ct->cc_pid = 0;
3696 ct->cc_caller_id = nfs4_srv_caller_id;
3697 ct->cc_flags = CC_DONTBLOCK;
3698 }
3699
3700 if (ISSPECIAL(stateid)) {
3701 fp = rfs4_findfile(vp, NULL, &create);
3702 if (fp == NULL)
3703 return (NFS4_OK);
3704 if (fp->rf_dinfo.rd_dtype == OPEN_DELEGATE_NONE) {
3705 rfs4_file_rele(fp);
3706 return (NFS4_OK);
3707 }
3708 if (mode == FWRITE ||
3709 fp->rf_dinfo.rd_dtype == OPEN_DELEGATE_WRITE) {
3710 rfs4_recall_deleg(fp, trunc, NULL);
3711 rfs4_file_rele(fp);
3712 return (NFS4ERR_DELAY);
3713 }
3714 rfs4_file_rele(fp);
3715 return (NFS4_OK);
3716 } else {
3717 stat = rfs4_get_all_state(stateid, &sp, &dsp, &lsp);
3718 if (stat != NFS4_OK)
3719 return (stat);
3720 if (lsp != NULL) {
3721 /* Is associated server instance in its grace period? */
3722 if (rfs4_clnt_in_grace(lsp->rls_locker->rl_client)) {
3723 rfs4_lo_state_rele(lsp, FALSE);
3724 if (sp != NULL)
3725 rfs4_state_rele_nounlock(sp);
3726 return (NFS4ERR_GRACE);
3727 }
3728 if (id->bits.type == LOCKID) {
3729 /* Seqid in the future? - that's bad */
3730 if (lsp->rls_lockid.bits.chgseq <
3731 id->bits.chgseq) {
3732 rfs4_lo_state_rele(lsp, FALSE);
3733 if (sp != NULL)
3734 rfs4_state_rele_nounlock(sp);
3735 return (NFS4ERR_BAD_STATEID);
3736 }
3737 /* Seqid in the past? - that's old */
3738 if (lsp->rls_lockid.bits.chgseq >
3739 id->bits.chgseq) {
3740 rfs4_lo_state_rele(lsp, FALSE);
3741 if (sp != NULL)
3742 rfs4_state_rele_nounlock(sp);
3743 return (NFS4ERR_OLD_STATEID);
3744 }
3745 /* Ensure specified filehandle matches */
3746 if (lsp->rls_state->rs_finfo->rf_vp != vp) {
3747 rfs4_lo_state_rele(lsp, FALSE);
3748 if (sp != NULL)
3749 rfs4_state_rele_nounlock(sp);
3750 return (NFS4ERR_BAD_STATEID);
3751 }
3752 }
3753 if (ct != NULL) {
3754 ct->cc_sysid =
3755 lsp->rls_locker->rl_client->rc_sysidt;
3756 ct->cc_pid = lsp->rls_locker->rl_pid;
3757 }
3758 rfs4_lo_state_rele(lsp, FALSE);
3759 }
3760
3761 /* Stateid provided was an "open" stateid */
3762 if (sp != NULL) {
3763 /* Is associated server instance in its grace period? */
3764 if (rfs4_clnt_in_grace(sp->rs_owner->ro_client)) {
3765 rfs4_state_rele_nounlock(sp);
3766 return (NFS4ERR_GRACE);
3767 }
3768 if (id->bits.type == OPENID) {
3769 /* Seqid in the future? - that's bad */
3770 if (sp->rs_stateid.bits.chgseq <
3771 id->bits.chgseq) {
3772 rfs4_state_rele_nounlock(sp);
3773 return (NFS4ERR_BAD_STATEID);
3774 }
3775 /* Seqid in the past - that's old */
3776 if (sp->rs_stateid.bits.chgseq >
3777 id->bits.chgseq) {
3778 rfs4_state_rele_nounlock(sp);
3779 return (NFS4ERR_OLD_STATEID);
3780 }
3781 }
3782 /* Ensure specified filehandle matches */
3783 if (sp->rs_finfo->rf_vp != vp) {
3784 rfs4_state_rele_nounlock(sp);
3785 return (NFS4ERR_BAD_STATEID);
3786 }
3787
3788 if (sp->rs_owner->ro_need_confirm) {
3789 rfs4_state_rele_nounlock(sp);
3790 return (NFS4ERR_BAD_STATEID);
3791 }
3792
3793 if (sp->rs_closed == TRUE) {
3794 rfs4_state_rele_nounlock(sp);
3795 return (NFS4ERR_OLD_STATEID);
3796 }
3797
3798 if (do_access)
3799 stat = rfs4_state_has_access(sp, mode, vp);
3800 else
3801 stat = NFS4_OK;
3802
3803 /*
3804 * Return whether this state has write
3805 * delegation if desired
3806 */
3807 if (deleg && (sp->rs_finfo->rf_dinfo.rd_dtype ==
3808 OPEN_DELEGATE_WRITE))
3809 *deleg = TRUE;
3810
3811 /*
3812 * We got a valid stateid, so we update the
3813 * lease on the client. Ideally we would like
3814 * to do this after the calling op succeeds,
3815 * but for now this will be good
3816 * enough. Callers of this routine are
3817 * currently insulated from the state stuff.
3818 */
3819 rfs4_update_lease(sp->rs_owner->ro_client);
3820
3821 /*
3822 * If a delegation is present on this file and
3823 * this is a WRITE, then update the lastwrite
3824 * time to indicate that activity is present.
3825 */
3826 if (sp->rs_finfo->rf_dinfo.rd_dtype ==
3827 OPEN_DELEGATE_WRITE &&
3828 mode == FWRITE) {
3829 sp->rs_finfo->rf_dinfo.rd_time_lastwrite =
3830 gethrestime_sec();
3831 }
3832
3833 rfs4_state_rele_nounlock(sp);
3834
3835 return (stat);
3836 }
3837
3838 if (dsp != NULL) {
3839 /* Is associated server instance in its grace period? */
3840 if (rfs4_clnt_in_grace(dsp->rds_client)) {
3841 rfs4_deleg_state_rele(dsp);
3842 return (NFS4ERR_GRACE);
3843 }
3844 if (dsp->rds_delegid.bits.chgseq != id->bits.chgseq) {
3845 rfs4_deleg_state_rele(dsp);
3846 return (NFS4ERR_BAD_STATEID);
3847 }
3848
3849 /* Ensure specified filehandle matches */
3850 if (dsp->rds_finfo->rf_vp != vp) {
3851 rfs4_deleg_state_rele(dsp);
3852 return (NFS4ERR_BAD_STATEID);
3853 }
3854 /*
3855 * Return whether this state has write
3856 * delegation if desired
3857 */
3858 if (deleg && (dsp->rds_finfo->rf_dinfo.rd_dtype ==
3859 OPEN_DELEGATE_WRITE))
3860 *deleg = TRUE;
3861
3862 rfs4_update_lease(dsp->rds_client);
3863
3864 /*
3865 * If a delegation is present on this file and
3866 * this is a WRITE, then update the lastwrite
3867 * time to indicate that activity is present.
3868 */
3869 if (dsp->rds_finfo->rf_dinfo.rd_dtype ==
3870 OPEN_DELEGATE_WRITE && mode == FWRITE) {
3871 dsp->rds_finfo->rf_dinfo.rd_time_lastwrite =
3872 gethrestime_sec();
3873 }
3874
3875 /*
3876 * XXX - what happens if this is a WRITE and the
3877 * delegation type of for READ.
3878 */
3879 rfs4_deleg_state_rele(dsp);
3880
3881 return (stat);
3882 }
3883 /*
3884 * If we got this far, something bad happened
3885 */
3886 return (NFS4ERR_BAD_STATEID);
3887 }
3888 }
3889
3890
3891 /*
3892 * This is a special function in that for the file struct provided the
3893 * server wants to remove/close all current state associated with the
3894 * file. The prime use of this would be with OP_REMOVE to force the
3895 * release of state and particularly of file locks.
3896 *
3897 * There is an assumption that there is no delegations outstanding on
3898 * this file at this point. The caller should have waited for those
3899 * to be returned or revoked.
3900 */
3901 void
3902 rfs4_close_all_state(rfs4_file_t *fp)
3903 {
3904 rfs4_state_t *sp;
3905
3906 rfs4_dbe_lock(fp->rf_dbe);
3907
3908 #ifdef DEBUG
3909 /* only applies when server is handing out delegations */
3910 if (nfs4_get_deleg_policy() != SRV_NEVER_DELEGATE)
3911 ASSERT(fp->rf_dinfo.rd_hold_grant > 0);
3912 #endif
3913
3914 /* No delegations for this file */
3915 ASSERT(list_is_empty(&fp->rf_delegstatelist));
3916
3917 /* Make sure that it can not be found */
3918 rfs4_dbe_invalidate(fp->rf_dbe);
3919
3920 if (fp->rf_vp == NULL) {
3921 rfs4_dbe_unlock(fp->rf_dbe);
3922 return;
3923 }
3924 rfs4_dbe_unlock(fp->rf_dbe);
3925
3926 /*
3927 * Hold as writer to prevent other server threads from
3928 * processing requests related to the file while all state is
3929 * being removed.
3930 */
3931 rw_enter(&fp->rf_file_rwlock, RW_WRITER);
3932
3933 /* Remove ALL state from the file */
3934 while (sp = rfs4_findstate_by_file(fp)) {
3935 rfs4_state_close(sp, FALSE, FALSE, CRED());
3936 rfs4_state_rele_nounlock(sp);
3937 }
3938
3939 /*
3940 * This is only safe since there are no further references to
3941 * the file.
3942 */
3943 rfs4_dbe_lock(fp->rf_dbe);
3944 if (fp->rf_vp) {
3945 vnode_t *vp = fp->rf_vp;
3946
3947 mutex_enter(&vp->v_vsd_lock);
3948 (void) vsd_set(vp, nfs4_srv_vkey, NULL);
3949 mutex_exit(&vp->v_vsd_lock);
3950 VN_RELE(vp);
3951 fp->rf_vp = NULL;
3952 }
3953 rfs4_dbe_unlock(fp->rf_dbe);
3954
3955 /* Finally let other references to proceed */
3956 rw_exit(&fp->rf_file_rwlock);
3957 }
3958
3959 /*
3960 * This function is used as a target for the rfs4_dbe_walk() call
3961 * below. The purpose of this function is to see if the
3962 * lockowner_state refers to a file that resides within the exportinfo
3963 * export. If so, then remove the lock_owner state (file locks and
3964 * share "locks") for this object since the intent is the server is
3965 * unexporting the specified directory. Be sure to invalidate the
3966 * object after the state has been released
3967 */
3968 static void
3969 rfs4_lo_state_walk_callout(rfs4_entry_t u_entry, void *e)
3970 {
3971 rfs4_lo_state_t *lsp = (rfs4_lo_state_t *)u_entry;
3972 struct exportinfo *exi = (struct exportinfo *)e;
3973 nfs_fh4_fmt_t fhfmt4, *exi_fhp, *finfo_fhp;
3974 fhandle_t *efhp;
3975
3976 efhp = (fhandle_t *)&exi->exi_fh;
3977 exi_fhp = (nfs_fh4_fmt_t *)&fhfmt4;
3978
3979 FH_TO_FMT4(efhp, exi_fhp);
3980
3981 finfo_fhp = (nfs_fh4_fmt_t *)lsp->rls_state->rs_finfo->
3982 rf_filehandle.nfs_fh4_val;
3983
3984 if (EQFSID(&finfo_fhp->fh4_fsid, &exi_fhp->fh4_fsid) &&
3985 bcmp(&finfo_fhp->fh4_xdata, &exi_fhp->fh4_xdata,
3986 exi_fhp->fh4_xlen) == 0) {
3987 rfs4_state_close(lsp->rls_state, FALSE, FALSE, CRED());
3988 rfs4_dbe_invalidate(lsp->rls_dbe);
3989 rfs4_dbe_invalidate(lsp->rls_state->rs_dbe);
3990 }
3991 }
3992
3993 /*
3994 * This function is used as a target for the rfs4_dbe_walk() call
3995 * below. The purpose of this function is to see if the state refers
3996 * to a file that resides within the exportinfo export. If so, then
3997 * remove the open state for this object since the intent is the
3998 * server is unexporting the specified directory. The main result for
3999 * this type of entry is to invalidate it such it will not be found in
4000 * the future.
4001 */
4002 static void
4003 rfs4_state_walk_callout(rfs4_entry_t u_entry, void *e)
4004 {
4005 rfs4_state_t *sp = (rfs4_state_t *)u_entry;
4006 struct exportinfo *exi = (struct exportinfo *)e;
4007 nfs_fh4_fmt_t fhfmt4, *exi_fhp, *finfo_fhp;
4008 fhandle_t *efhp;
4009
4010 efhp = (fhandle_t *)&exi->exi_fh;
4011 exi_fhp = (nfs_fh4_fmt_t *)&fhfmt4;
4012
4013 FH_TO_FMT4(efhp, exi_fhp);
4014
4015 finfo_fhp =
4016 (nfs_fh4_fmt_t *)sp->rs_finfo->rf_filehandle.nfs_fh4_val;
4017
4018 if (EQFSID(&finfo_fhp->fh4_fsid, &exi_fhp->fh4_fsid) &&
4019 bcmp(&finfo_fhp->fh4_xdata, &exi_fhp->fh4_xdata,
4020 exi_fhp->fh4_xlen) == 0) {
4021 rfs4_state_close(sp, TRUE, FALSE, CRED());
4022 rfs4_dbe_invalidate(sp->rs_dbe);
4023 }
4024 }
4025
4026 /*
4027 * This function is used as a target for the rfs4_dbe_walk() call
4028 * below. The purpose of this function is to see if the state refers
4029 * to a file that resides within the exportinfo export. If so, then
4030 * remove the deleg state for this object since the intent is the
4031 * server is unexporting the specified directory. The main result for
4032 * this type of entry is to invalidate it such it will not be found in
4033 * the future.
4034 */
4035 static void
4036 rfs4_deleg_state_walk_callout(rfs4_entry_t u_entry, void *e)
4037 {
4038 rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
4039 struct exportinfo *exi = (struct exportinfo *)e;
4040 nfs_fh4_fmt_t fhfmt4, *exi_fhp, *finfo_fhp;
4041 fhandle_t *efhp;
4042
4043 efhp = (fhandle_t *)&exi->exi_fh;
4044 exi_fhp = (nfs_fh4_fmt_t *)&fhfmt4;
4045
4046 FH_TO_FMT4(efhp, exi_fhp);
4047
4048 finfo_fhp =
4049 (nfs_fh4_fmt_t *)dsp->rds_finfo->rf_filehandle.nfs_fh4_val;
4050
4051 if (EQFSID(&finfo_fhp->fh4_fsid, &exi_fhp->fh4_fsid) &&
4052 bcmp(&finfo_fhp->fh4_xdata, &exi_fhp->fh4_xdata,
4053 exi_fhp->fh4_xlen) == 0) {
4054 rfs4_dbe_invalidate(dsp->rds_dbe);
4055 }
4056 }
4057
4058 /*
4059 * This function is used as a target for the rfs4_dbe_walk() call
4060 * below. The purpose of this function is to see if the state refers
4061 * to a file that resides within the exportinfo export. If so, then
4062 * release vnode hold for this object since the intent is the server
4063 * is unexporting the specified directory. Invalidation will prevent
4064 * this struct from being found in the future.
4065 */
4066 static void
4067 rfs4_file_walk_callout(rfs4_entry_t u_entry, void *e)
4068 {
4069 rfs4_file_t *fp = (rfs4_file_t *)u_entry;
4070 struct exportinfo *exi = (struct exportinfo *)e;
4071 nfs_fh4_fmt_t fhfmt4, *exi_fhp, *finfo_fhp;
4072 fhandle_t *efhp;
4073
4074 efhp = (fhandle_t *)&exi->exi_fh;
4075 exi_fhp = (nfs_fh4_fmt_t *)&fhfmt4;
4076
4077 FH_TO_FMT4(efhp, exi_fhp);
4078
4079 finfo_fhp = (nfs_fh4_fmt_t *)fp->rf_filehandle.nfs_fh4_val;
4080
4081 if (EQFSID(&finfo_fhp->fh4_fsid, &exi_fhp->fh4_fsid) &&
4082 bcmp(&finfo_fhp->fh4_xdata, &exi_fhp->fh4_xdata,
4083 exi_fhp->fh4_xlen) == 0) {
4084 if (fp->rf_vp) {
4085 vnode_t *vp = fp->rf_vp;
4086
4087 /*
4088 * don't leak monitors and remove the reference
4089 * put on the vnode when the delegation was granted.
4090 */
4091 if (fp->rf_dinfo.rd_dtype == OPEN_DELEGATE_READ) {
4092 (void) fem_uninstall(vp, deleg_rdops,
4093 (void *)fp);
4094 vn_open_downgrade(vp, FREAD);
4095 } else if (fp->rf_dinfo.rd_dtype ==
4096 OPEN_DELEGATE_WRITE) {
4097 (void) fem_uninstall(vp, deleg_wrops,
4098 (void *)fp);
4099 vn_open_downgrade(vp, FREAD|FWRITE);
4100 }
4101 mutex_enter(&vp->v_vsd_lock);
4102 (void) vsd_set(vp, nfs4_srv_vkey, NULL);
4103 mutex_exit(&vp->v_vsd_lock);
4104 VN_RELE(vp);
4105 fp->rf_vp = NULL;
4106 }
4107 rfs4_dbe_invalidate(fp->rf_dbe);
4108 }
4109 }
4110
4111 /*
4112 * Given a directory that is being unexported, cleanup/release all
4113 * state in the server that refers to objects residing underneath this
4114 * particular export. The ordering of the release is important.
4115 * Lock_owner, then state and then file.
4116 *
4117 * NFS zones note: nfs_export.c:unexport() calls this from a
4118 * thread in the global zone for NGZ data structures, so we
4119 * CANNOT use zone_getspecific anywhere in this code path.
4120 */
4121 void
4122 rfs4_clean_state_exi(nfs_export_t *ne, struct exportinfo *exi)
4123 {
4124 nfs_globals_t *ng;
4125 nfs4_srv_t *nsrv4;
4126
4127 ng = ne->ne_globals;
4128 ASSERT(ng->nfs_zoneid == exi->exi_zoneid);
4129 nsrv4 = ng->nfs4_srv;
4130
4131 mutex_enter(&nsrv4->state_lock);
4132
4133 if (nsrv4->nfs4_server_state == NULL) {
4134 mutex_exit(&nsrv4->state_lock);
4135 return;
4136 }
4137
4138 rfs4_dbe_walk(nsrv4->rfs4_lo_state_tab,
4139 rfs4_lo_state_walk_callout, exi);
4140 rfs4_dbe_walk(nsrv4->rfs4_state_tab, rfs4_state_walk_callout, exi);
4141 rfs4_dbe_walk(nsrv4->rfs4_deleg_state_tab,
4142 rfs4_deleg_state_walk_callout, exi);
4143 rfs4_dbe_walk(nsrv4->rfs4_file_tab, rfs4_file_walk_callout, exi);
4144
4145 mutex_exit(&nsrv4->state_lock);
4146 }