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 = zone_getspecific(rfs4_zone_key, curzone);
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 * Initilized when the module is loaded and used by NFSv4 state tables.
1159 * These kmem_cache free pools are used globally, the NFSv4 state
1160 * tables which make use of these kmem_cache free pools are per zone.
1161 *
1162 * initialize the global kmem_cache free pools which will be used by
1163 * the NFSv4 state tables.
1164 */
1165 /* CSTYLED */
1166 rfs4_client_mem_cache = nfs4_init_mem_cache("Client_entry_cache", 2, sizeof (rfs4_client_t), 0);
1167 /* CSTYLED */
1168 rfs4_clntIP_mem_cache = nfs4_init_mem_cache("ClntIP_entry_cache", 1, sizeof (rfs4_clntip_t), 1);
1169 /* CSTYLED */
1170 rfs4_openown_mem_cache = nfs4_init_mem_cache("OpenOwner_entry_cache", 1, sizeof (rfs4_openowner_t), 2);
1171 /* CSTYLED */
1172 rfs4_openstID_mem_cache = nfs4_init_mem_cache("OpenStateID_entry_cache", 3, sizeof (rfs4_state_t), 3);
1173 /* CSTYLED */
1174 rfs4_lockstID_mem_cache = nfs4_init_mem_cache("LockStateID_entry_cache", 3, sizeof (rfs4_lo_state_t), 4);
1175 /* CSTYLED */
1176 rfs4_lockown_mem_cache = nfs4_init_mem_cache("Lockowner_entry_cache", 2, sizeof (rfs4_lockowner_t), 5);
1177 /* CSTYLED */
1178 rfs4_file_mem_cache = nfs4_init_mem_cache("File_entry_cache", 1, sizeof (rfs4_file_t), 6);
1179 /* CSTYLED */
1180 rfs4_delegstID_mem_cache = nfs4_init_mem_cache("DelegStateID_entry_cache", 2, sizeof (rfs4_deleg_state_t), 7);
1181
1182 rfs4_client_clrst = rfs4_clear_client_state;
1183 }
1184
1185
1186 /*
1187 * Used at server shutdown to cleanup all of the NFSv4 server's structures
1188 * and other state.
1189 */
1190 void
1191 rfs4_state_g_fini()
1192 {
1193 int i;
1194 /*
1195 * Cleanup the CPR callback.
1196 */
1197 if (cpr_id)
1198 (void) callb_delete(cpr_id);
1199
1200 rfs4_client_clrst = NULL;
1201
1202 /* free the NFSv4 state databases */
1203 for (i = 0; i < RFS4_DB_MEM_CACHE_NUM; i++) {
1204 kmem_cache_destroy(rfs4_db_mem_cache_table[i].r_db_mem_cache);
1205 rfs4_db_mem_cache_table[i].r_db_mem_cache = NULL;
1206 }
1207
1208 rfs4_client_mem_cache = NULL;
1209 rfs4_clntIP_mem_cache = NULL;
1210 rfs4_openown_mem_cache = NULL;
1211 rfs4_openstID_mem_cache = NULL;
1212 rfs4_lockstID_mem_cache = NULL;
1213 rfs4_lockown_mem_cache = NULL;
1214 rfs4_file_mem_cache = NULL;
1215 rfs4_delegstID_mem_cache = NULL;
1216
1217 /* DSS: distributed stable storage */
1218 nvlist_free(rfs4_dss_oldpaths);
1219 nvlist_free(rfs4_dss_paths);
1220 rfs4_dss_paths = rfs4_dss_oldpaths = NULL;
1221 }
1222
1223 /*
1224 * Used to initialize the per zone NFSv4 server's state
1225 */
1226 void
1227 rfs4_state_zone_init(nfs4_srv_t *nsrv4)
1228 {
1229 time_t start_time;
1230 int start_grace;
1231 char *dss_path = NFS4_DSS_VAR_DIR;
1232
1233 /* DSS: distributed stable storage: initialise served paths list */
1234 nsrv4->dss_pathlist = NULL;
1235
1236 /*
1237 * Set the boot time. If the server
1238 * has been restarted quickly and has had the opportunity to
1239 * service clients, then the start_time needs to be bumped
1240 * regardless. A small window but it exists...
1241 */
1242 start_time = gethrestime_sec();
1243 if (nsrv4->rfs4_start_time < start_time)
1244 nsrv4->rfs4_start_time = start_time;
1245 else
1246 nsrv4->rfs4_start_time++;
1247
1248 /*
1249 * Create the first server instance, or a new one if the server has
1250 * been restarted; see above comments on rfs4_start_time. Don't
1251 * start its grace period; that will be done later, to maximise the
1252 * clients' recovery window.
1253 */
1254 start_grace = 0;
1255 if (curzone == global_zone && rfs4_dss_numnewpaths > 0) {
1256 int i;
1257 char **dss_allpaths = NULL;
1258 dss_allpaths = kmem_alloc(sizeof (char *) * (rfs4_dss_numnewpaths + 1), KM_SLEEP);
1259 /*
1260 * Add the default path into the list of paths for saving
1261 * state informantion.
1262 */
1263 dss_allpaths[0] = dss_path;
1264 for ( i = 0; i < rfs4_dss_numnewpaths; i++) {
1265 dss_allpaths[i + 1] = rfs4_dss_newpaths[i];
1266 }
1267 rfs4_servinst_create(nsrv4, start_grace, (rfs4_dss_numnewpaths + 1), dss_allpaths);
1268 kmem_free(dss_allpaths, (sizeof (char *) * (rfs4_dss_numnewpaths + 1)));
1269 } else {
1270 rfs4_servinst_create(nsrv4, start_grace, 1, &dss_path);
1271 }
1272
1273 /* reset the "first NFSv4 request" status */
1274 nsrv4->seen_first_compound = 0;
1275
1276 mutex_enter(&nsrv4->state_lock);
1277
1278 /*
1279 * If the server state database has already been initialized,
1280 * skip it
1281 */
1282 if (nsrv4->nfs4_server_state != NULL) {
1283 mutex_exit(&nsrv4->state_lock);
1284 return;
1285 }
1286
1287 rw_init(&nsrv4->rfs4_findclient_lock, NULL, RW_DEFAULT, NULL);
1288
1289 /* set the various cache timers for table creation */
1290 if (nsrv4->rfs4_client_cache_time == 0)
1291 nsrv4->rfs4_client_cache_time = CLIENT_CACHE_TIME;
1292 if (nsrv4->rfs4_openowner_cache_time == 0)
1293 nsrv4->rfs4_openowner_cache_time = OPENOWNER_CACHE_TIME;
1294 if (nsrv4->rfs4_state_cache_time == 0)
1295 nsrv4->rfs4_state_cache_time = STATE_CACHE_TIME;
1296 if (nsrv4->rfs4_lo_state_cache_time == 0)
1297 nsrv4->rfs4_lo_state_cache_time = LO_STATE_CACHE_TIME;
1298 if (nsrv4->rfs4_lockowner_cache_time == 0)
1299 nsrv4->rfs4_lockowner_cache_time = LOCKOWNER_CACHE_TIME;
1300 if (nsrv4->rfs4_file_cache_time == 0)
1301 nsrv4->rfs4_file_cache_time = FILE_CACHE_TIME;
1302 if (nsrv4->rfs4_deleg_state_cache_time == 0)
1303 nsrv4->rfs4_deleg_state_cache_time = DELEG_STATE_CACHE_TIME;
1304
1305 /* Create the overall database to hold all server state */
1306 nsrv4->nfs4_server_state = rfs4_database_create(rfs4_database_debug);
1307
1308 /* Now create the individual tables */
1309 nsrv4->rfs4_client_cache_time *= rfs4_lease_time;
1310 nsrv4->rfs4_client_tab = rfs4_table_create(nsrv4->nfs4_server_state,
1311 "Client",
1312 nsrv4->rfs4_client_cache_time,
1313 2,
1314 rfs4_client_create,
1315 rfs4_client_destroy,
1316 rfs4_client_expiry,
1317 sizeof (rfs4_client_t),
1318 TABSIZE,
1319 MAXTABSZ/8, 100);
1320 nsrv4->rfs4_nfsclnt_idx = rfs4_index_create(nsrv4->rfs4_client_tab,
1321 "nfs_client_id4", nfsclnt_hash,
1322 nfsclnt_compare, nfsclnt_mkkey,
1323 TRUE);
1324 nsrv4->rfs4_clientid_idx = rfs4_index_create(nsrv4->rfs4_client_tab,
1325 "client_id", clientid_hash,
1326 clientid_compare, clientid_mkkey,
1327 FALSE);
1328
1329 nsrv4->rfs4_clntip_cache_time = 86400 * 365; /* about a year */
1330 nsrv4->rfs4_clntip_tab = rfs4_table_create(nsrv4->nfs4_server_state,
1331 "ClntIP",
1332 nsrv4->rfs4_clntip_cache_time,
1333 1,
1334 rfs4_clntip_create,
1335 rfs4_clntip_destroy,
1336 rfs4_clntip_expiry,
1337 sizeof (rfs4_clntip_t),
1338 TABSIZE,
1339 MAXTABSZ, 100);
1340 nsrv4->rfs4_clntip_idx = rfs4_index_create(nsrv4->rfs4_clntip_tab,
1341 "client_ip", clntip_hash,
1342 clntip_compare, clntip_mkkey,
1343 TRUE);
1344
1345 nsrv4->rfs4_openowner_cache_time *= rfs4_lease_time;
1346 nsrv4->rfs4_openowner_tab = rfs4_table_create(nsrv4->nfs4_server_state,
1347 "OpenOwner",
1348 nsrv4->rfs4_openowner_cache_time,
1349 1,
1350 rfs4_openowner_create,
1351 rfs4_openowner_destroy,
1352 rfs4_openowner_expiry,
1353 sizeof (rfs4_openowner_t),
1354 TABSIZE,
1355 MAXTABSZ, 100);
1356 nsrv4->rfs4_openowner_idx = rfs4_index_create(nsrv4->rfs4_openowner_tab,
1357 "open_owner4", openowner_hash,
1358 openowner_compare,
1359 openowner_mkkey, TRUE);
1360
1361 nsrv4->rfs4_state_cache_time *= rfs4_lease_time;
1362 nsrv4->rfs4_state_tab = rfs4_table_create(nsrv4->nfs4_server_state,
1363 "OpenStateID",
1364 nsrv4->rfs4_state_cache_time,
1365 3,
1366 rfs4_state_create,
1367 rfs4_state_destroy,
1368 rfs4_state_expiry,
1369 sizeof (rfs4_state_t),
1370 TABSIZE,
1371 MAXTABSZ, 100);
1372
1373 /* CSTYLED */
1374 nsrv4->rfs4_state_owner_file_idx = rfs4_index_create(nsrv4->rfs4_state_tab,
1375 "Openowner-File",
1376 state_owner_file_hash,
1377 state_owner_file_compare,
1378 state_owner_file_mkkey, TRUE);
1379
1380 nsrv4->rfs4_state_idx = rfs4_index_create(nsrv4->rfs4_state_tab,
1381 "State-id", state_hash,
1382 state_compare, state_mkkey, FALSE);
1383
1384 nsrv4->rfs4_state_file_idx = rfs4_index_create(nsrv4->rfs4_state_tab,
1385 "File", state_file_hash,
1386 state_file_compare, state_file_mkkey,
1387 FALSE);
1388
1389 nsrv4->rfs4_lo_state_cache_time *= rfs4_lease_time;
1390 nsrv4->rfs4_lo_state_tab = rfs4_table_create(nsrv4->nfs4_server_state,
1391 "LockStateID",
1392 nsrv4->rfs4_lo_state_cache_time,
1393 2,
1394 rfs4_lo_state_create,
1395 rfs4_lo_state_destroy,
1396 rfs4_lo_state_expiry,
1397 sizeof (rfs4_lo_state_t),
1398 TABSIZE,
1399 MAXTABSZ, 100);
1400
1401 /* CSTYLED */
1402 nsrv4->rfs4_lo_state_owner_idx = rfs4_index_create(nsrv4->rfs4_lo_state_tab,
1403 "lockownerxstate",
1404 lo_state_lo_hash,
1405 lo_state_lo_compare,
1406 lo_state_lo_mkkey, TRUE);
1407
1408 nsrv4->rfs4_lo_state_idx = rfs4_index_create(nsrv4->rfs4_lo_state_tab,
1409 "State-id",
1410 lo_state_hash, lo_state_compare,
1411 lo_state_mkkey, FALSE);
1412
1413 nsrv4->rfs4_lockowner_cache_time *= rfs4_lease_time;
1414
1415 nsrv4->rfs4_lockowner_tab = rfs4_table_create(nsrv4->nfs4_server_state,
1416 "Lockowner",
1417 nsrv4->rfs4_lockowner_cache_time,
1418 2,
1419 rfs4_lockowner_create,
1420 rfs4_lockowner_destroy,
1421 rfs4_lockowner_expiry,
1422 sizeof (rfs4_lockowner_t),
1423 TABSIZE,
1424 MAXTABSZ, 100);
1425
1426 nsrv4->rfs4_lockowner_idx = rfs4_index_create(nsrv4->rfs4_lockowner_tab,
1427 "lock_owner4", lockowner_hash,
1428 lockowner_compare,
1429 lockowner_mkkey, TRUE);
1430
1431 /* CSTYLED */
1432 nsrv4->rfs4_lockowner_pid_idx = rfs4_index_create(nsrv4->rfs4_lockowner_tab,
1433 "pid", pid_hash,
1434 pid_compare, pid_mkkey,
1435 FALSE);
1436
1437 nsrv4->rfs4_file_cache_time *= rfs4_lease_time;
1438 nsrv4->rfs4_file_tab = rfs4_table_create(nsrv4->nfs4_server_state,
1439 "File",
1440 nsrv4->rfs4_file_cache_time,
1441 1,
1442 rfs4_file_create,
1443 rfs4_file_destroy,
1444 NULL,
1445 sizeof (rfs4_file_t),
1446 TABSIZE,
1447 MAXTABSZ, -1);
1448
1449 nsrv4->rfs4_file_idx = rfs4_index_create(nsrv4->rfs4_file_tab,
1450 "Filehandle", file_hash,
1451 file_compare, file_mkkey, TRUE);
1452
1453 nsrv4->rfs4_deleg_state_cache_time *= rfs4_lease_time;
1454 /* CSTYLED */
1455 nsrv4->rfs4_deleg_state_tab = rfs4_table_create(nsrv4->nfs4_server_state,
1456 "DelegStateID",
1457 nsrv4->rfs4_deleg_state_cache_time,
1458 2,
1459 rfs4_deleg_state_create,
1460 rfs4_deleg_state_destroy,
1461 rfs4_deleg_state_expiry,
1462 sizeof (rfs4_deleg_state_t),
1463 TABSIZE,
1464 MAXTABSZ, 100);
1465 nsrv4->rfs4_deleg_idx = rfs4_index_create(nsrv4->rfs4_deleg_state_tab,
1466 "DelegByFileClient",
1467 deleg_hash,
1468 deleg_compare,
1469 deleg_mkkey, TRUE);
1470
1471 /* CSTYLED */
1472 nsrv4->rfs4_deleg_state_idx = rfs4_index_create(nsrv4->rfs4_deleg_state_tab,
1473 "DelegState",
1474 deleg_state_hash,
1475 deleg_state_compare,
1476 deleg_state_mkkey, FALSE);
1477
1478 mutex_exit(&nsrv4->state_lock);
1479
1480 /*
1481 * Init the stable storage.
1482 */
1483 rfs4_ss_init(nsrv4);
1484 }
1485
1486 /*
1487 * Used at server shutdown to cleanup all of NFSv4 server's zone structures
1488 * and state.
1489 */
1490 void
1491 rfs4_state_zone_fini()
1492 {
1493 rfs4_database_t *dbp;
1494 nfs4_srv_t *nsrv4;
1495 nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
1496
1497 rfs4_set_deleg_policy(nsrv4, SRV_NEVER_DELEGATE);
1498
1499 mutex_enter(&nsrv4->state_lock);
1500
1501 if (nsrv4->nfs4_server_state == NULL) {
1502 mutex_exit(&nsrv4->state_lock);
1503 return;
1504 }
1505
1506 /* destroy server instances and current instance ptr */
1507 rfs4_servinst_destroy_all(nsrv4);
1508
1509 /* reset the "first NFSv4 request" status */
1510 nsrv4->seen_first_compound = 0;
1511
1512 dbp = nsrv4->nfs4_server_state;
1513 nsrv4->nfs4_server_state = NULL;
1514
1515 rw_destroy(&nsrv4->rfs4_findclient_lock);
1516
1517 /* First stop all of the reaper threads in the database */
1518 rfs4_database_shutdown(dbp);
1519 /*
1520 * XXX workaround
1521 * Skip destrying the state database yet just in case there
1522 * are unfinished operations depending on it.
1523 */
1524 /* Now destroy/release the database tables */
1525 /* rfs4_database_destroy(dbp); */
1526
1527 /* Reset the cache timers for next time */
1528 nsrv4->rfs4_client_cache_time = 0;
1529 nsrv4->rfs4_openowner_cache_time = 0;
1530 nsrv4->rfs4_state_cache_time = 0;
1531 nsrv4->rfs4_lo_state_cache_time = 0;
1532 nsrv4->rfs4_lockowner_cache_time = 0;
1533 nsrv4->rfs4_file_cache_time = 0;
1534 nsrv4->rfs4_deleg_state_cache_time = 0;
1535
1536 mutex_exit(&nsrv4->state_lock);
1537
1538 /* clean up any dangling stable storage structures */
1539 rfs4_ss_fini(nsrv4);
1540 }
1541
1542 typedef union {
1543 struct {
1544 uint32_t start_time;
1545 uint32_t c_id;
1546 } impl_id;
1547 clientid4 id4;
1548 } cid;
1549
1550 static int foreign_stateid(stateid_t *id);
1551 static int foreign_clientid(cid *cidp);
1552 static void embed_nodeid(cid *cidp);
1553
1554 typedef union {
1555 struct {
1556 uint32_t c_id;
1557 uint32_t gen_num;
1558 } cv_impl;
1559 verifier4 confirm_verf;
1560 } scid_confirm_verf;
1561
1562 static uint32_t
1563 clientid_hash(void *key)
1564 {
1565 cid *idp = key;
1566
1567 return (idp->impl_id.c_id);
1568 }
1569
1570 static bool_t
1571 clientid_compare(rfs4_entry_t entry, void *key)
1572 {
1573 rfs4_client_t *cp = (rfs4_client_t *)entry;
1574 clientid4 *idp = key;
1575
1576 return (*idp == cp->rc_clientid);
1577 }
1578
1579 static void *
1580 clientid_mkkey(rfs4_entry_t entry)
1581 {
1582 rfs4_client_t *cp = (rfs4_client_t *)entry;
1583
1584 return (&cp->rc_clientid);
1585 }
1586
1587 static uint32_t
1588 nfsclnt_hash(void *key)
1589 {
1590 nfs_client_id4 *client = key;
1591 int i;
1592 uint32_t hash = 0;
1593
1594 for (i = 0; i < client->id_len; i++) {
1595 hash <<= 1;
1596 hash += (uint_t)client->id_val[i];
1597 }
1598 return (hash);
1599 }
1600
1601
1602 static bool_t
1603 nfsclnt_compare(rfs4_entry_t entry, void *key)
1604 {
1605 rfs4_client_t *cp = (rfs4_client_t *)entry;
1606 nfs_client_id4 *nfs_client = key;
1607
1608 if (cp->rc_nfs_client.id_len != nfs_client->id_len)
1609 return (FALSE);
1610
1611 return (bcmp(cp->rc_nfs_client.id_val, nfs_client->id_val,
1612 nfs_client->id_len) == 0);
1613 }
1614
1615 static void *
1616 nfsclnt_mkkey(rfs4_entry_t entry)
1617 {
1618 rfs4_client_t *cp = (rfs4_client_t *)entry;
1619
1620 return (&cp->rc_nfs_client);
1621 }
1622
1623 static bool_t
1624 rfs4_client_expiry(rfs4_entry_t u_entry)
1625 {
1626 rfs4_client_t *cp = (rfs4_client_t *)u_entry;
1627 bool_t cp_expired;
1628
1629 if (rfs4_dbe_is_invalid(cp->rc_dbe)) {
1630 cp->rc_ss_remove = 1;
1631 return (TRUE);
1632 }
1633 /*
1634 * If the sysadmin has used clear_locks for this
1635 * entry then forced_expire will be set and we
1636 * want this entry to be reaped. Or the entry
1637 * has exceeded its lease period.
1638 */
1639 cp_expired = (cp->rc_forced_expire ||
1640 (gethrestime_sec() - cp->rc_last_access
1641 > rfs4_lease_time));
1642
1643 if (!cp->rc_ss_remove && cp_expired)
1644 cp->rc_ss_remove = 1;
1645 return (cp_expired);
1646 }
1647
1648 /*
1649 * Remove the leaf file from all distributed stable storage paths.
1650 */
1651 static void
1652 rfs4_dss_remove_cpleaf(rfs4_client_t *cp)
1653 {
1654 nfs4_srv_t *nsrv4;
1655 rfs4_servinst_t *sip;
1656 char *leaf = cp->rc_ss_pn->leaf;
1657
1658 /*
1659 * since the state files are written to all DSS
1660 * paths we must remove this leaf file instance
1661 * from all server instances.
1662 */
1663
1664 nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
1665 mutex_enter(&nsrv4->servinst_lock);
1666 for (sip = nsrv4->nfs4_cur_servinst; sip != NULL; sip = sip->prev) {
1667 /* remove the leaf file associated with this server instance */
1668 rfs4_dss_remove_leaf(sip, NFS4_DSS_STATE_LEAF, leaf);
1669 }
1670 mutex_exit(&nsrv4->servinst_lock);
1671 }
1672
1673 static void
1674 rfs4_dss_remove_leaf(rfs4_servinst_t *sip, char *dir_leaf, char *leaf)
1675 {
1676 int i, npaths = sip->dss_npaths;
1677
1678 for (i = 0; i < npaths; i++) {
1679 rfs4_dss_path_t *dss_path = sip->dss_paths[i];
1680 char *path, *dir;
1681 size_t pathlen;
1682
1683 /* the HA-NFSv4 path might have been failed-over away from us */
1684 if (dss_path == NULL)
1685 continue;
1686
1687 dir = dss_path->path;
1688
1689 /* allow 3 extra bytes for two '/' & a NUL */
1690 pathlen = strlen(dir) + strlen(dir_leaf) + strlen(leaf) + 3;
1691 path = kmem_alloc(pathlen, KM_SLEEP);
1692 (void) sprintf(path, "%s/%s/%s", dir, dir_leaf, leaf);
1693
1694 (void) vn_remove(path, UIO_SYSSPACE, RMFILE);
1695
1696 kmem_free(path, pathlen);
1697 }
1698 }
1699
1700 static void
1701 rfs4_client_destroy(rfs4_entry_t u_entry)
1702 {
1703 rfs4_client_t *cp = (rfs4_client_t *)u_entry;
1704
1705 mutex_destroy(cp->rc_cbinfo.cb_lock);
1706 cv_destroy(cp->rc_cbinfo.cb_cv);
1707 cv_destroy(cp->rc_cbinfo.cb_cv_nullcaller);
1708 list_destroy(&cp->rc_openownerlist);
1709
1710 /* free callback info */
1711 rfs4_cbinfo_free(&cp->rc_cbinfo);
1712
1713 if (cp->rc_cp_confirmed)
1714 rfs4_client_rele(cp->rc_cp_confirmed);
1715
1716 if (cp->rc_ss_pn) {
1717 /* check if the stable storage files need to be removed */
1718 if (cp->rc_ss_remove)
1719 rfs4_dss_remove_cpleaf(cp);
1720 rfs4_ss_pnfree(cp->rc_ss_pn);
1721 }
1722
1723 /* Free the client supplied client id */
1724 kmem_free(cp->rc_nfs_client.id_val, cp->rc_nfs_client.id_len);
1725
1726 if (cp->rc_sysidt != LM_NOSYSID)
1727 lm_free_sysidt(cp->rc_sysidt);
1728 }
1729
1730 static bool_t
1731 rfs4_client_create(rfs4_entry_t u_entry, void *arg)
1732 {
1733 rfs4_client_t *cp = (rfs4_client_t *)u_entry;
1734 nfs_client_id4 *client = (nfs_client_id4 *)arg;
1735 struct sockaddr *ca;
1736 cid *cidp;
1737 scid_confirm_verf *scvp;
1738 nfs4_srv_t *nsrv4;
1739
1740 nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
1741
1742 /* Get a clientid to give to the client */
1743 cidp = (cid *)&cp->rc_clientid;
1744 cidp->impl_id.start_time = nsrv4->rfs4_start_time;
1745 cidp->impl_id.c_id = (uint32_t)rfs4_dbe_getid(cp->rc_dbe);
1746
1747 /* If we are booted as a cluster node, embed our nodeid */
1748 if (cluster_bootflags & CLUSTER_BOOTED)
1749 embed_nodeid(cidp);
1750
1751 /* Allocate and copy client's client id value */
1752 cp->rc_nfs_client.id_val = kmem_alloc(client->id_len, KM_SLEEP);
1753 cp->rc_nfs_client.id_len = client->id_len;
1754 bcopy(client->id_val, cp->rc_nfs_client.id_val, client->id_len);
1755 cp->rc_nfs_client.verifier = client->verifier;
1756
1757 /* Copy client's IP address */
1758 ca = client->cl_addr;
1759 if (ca->sa_family == AF_INET)
1760 bcopy(ca, &cp->rc_addr, sizeof (struct sockaddr_in));
1761 else if (ca->sa_family == AF_INET6)
1762 bcopy(ca, &cp->rc_addr, sizeof (struct sockaddr_in6));
1763 cp->rc_nfs_client.cl_addr = (struct sockaddr *)&cp->rc_addr;
1764
1765 /* Init the value for the SETCLIENTID_CONFIRM verifier */
1766 scvp = (scid_confirm_verf *)&cp->rc_confirm_verf;
1767 scvp->cv_impl.c_id = cidp->impl_id.c_id;
1768 scvp->cv_impl.gen_num = 0;
1769
1770 /* An F_UNLKSYS has been done for this client */
1771 cp->rc_unlksys_completed = FALSE;
1772
1773 /* We need the client to ack us */
1774 cp->rc_need_confirm = TRUE;
1775 cp->rc_cp_confirmed = NULL;
1776
1777 /* TRUE all the time until the callback path actually fails */
1778 cp->rc_cbinfo.cb_notified_of_cb_path_down = TRUE;
1779
1780 /* Initialize the access time to now */
1781 cp->rc_last_access = gethrestime_sec();
1782
1783 cp->rc_cr_set = NULL;
1784
1785 cp->rc_sysidt = LM_NOSYSID;
1786
1787 list_create(&cp->rc_openownerlist, sizeof (rfs4_openowner_t),
1788 offsetof(rfs4_openowner_t, ro_node));
1789
1790 /* set up the callback control structure */
1791 cp->rc_cbinfo.cb_state = CB_UNINIT;
1792 mutex_init(cp->rc_cbinfo.cb_lock, NULL, MUTEX_DEFAULT, NULL);
1793 cv_init(cp->rc_cbinfo.cb_cv, NULL, CV_DEFAULT, NULL);
1794 cv_init(cp->rc_cbinfo.cb_cv_nullcaller, NULL, CV_DEFAULT, NULL);
1795
1796 /*
1797 * Associate the client_t with the current server instance.
1798 * The hold is solely to satisfy the calling requirement of
1799 * rfs4_servinst_assign(). In this case it's not strictly necessary.
1800 */
1801 rfs4_dbe_hold(cp->rc_dbe);
1802 rfs4_servinst_assign(nsrv4, cp, nsrv4->nfs4_cur_servinst);
1803 rfs4_dbe_rele(cp->rc_dbe);
1804
1805 return (TRUE);
1806 }
1807
1808 /*
1809 * Caller wants to generate/update the setclientid_confirm verifier
1810 * associated with a client. This is done during the SETCLIENTID
1811 * processing.
1812 */
1813 void
1814 rfs4_client_scv_next(rfs4_client_t *cp)
1815 {
1816 scid_confirm_verf *scvp;
1817
1818 /* Init the value for the SETCLIENTID_CONFIRM verifier */
1819 scvp = (scid_confirm_verf *)&cp->rc_confirm_verf;
1820 scvp->cv_impl.gen_num++;
1821 }
1822
1823 void
1824 rfs4_client_rele(rfs4_client_t *cp)
1825 {
1826 rfs4_dbe_rele(cp->rc_dbe);
1827 }
1828
1829 rfs4_client_t *
1830 rfs4_findclient(nfs_client_id4 *client, bool_t *create, rfs4_client_t *oldcp)
1831 {
1832 rfs4_client_t *cp;
1833 nfs4_srv_t *nsrv4;
1834 nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
1835
1836
1837 if (oldcp) {
1838 rw_enter(&nsrv4->rfs4_findclient_lock, RW_WRITER);
1839 rfs4_dbe_hide(oldcp->rc_dbe);
1840 } else {
1841 rw_enter(&nsrv4->rfs4_findclient_lock, RW_READER);
1842 }
1843
1844 cp = (rfs4_client_t *)rfs4_dbsearch(nsrv4->rfs4_nfsclnt_idx, client,
1845 create, (void *)client, RFS4_DBS_VALID);
1846
1847 if (oldcp)
1848 rfs4_dbe_unhide(oldcp->rc_dbe);
1849
1850 rw_exit(&nsrv4->rfs4_findclient_lock);
1851
1852 return (cp);
1853 }
1854
1855 rfs4_client_t *
1856 rfs4_findclient_by_id(clientid4 clientid, bool_t find_unconfirmed)
1857 {
1858 rfs4_client_t *cp;
1859 bool_t create = FALSE;
1860 cid *cidp = (cid *)&clientid;
1861 nfs4_srv_t *nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
1862
1863 /* If we're a cluster and the nodeid isn't right, short-circuit */
1864 if (cluster_bootflags & CLUSTER_BOOTED && foreign_clientid(cidp))
1865 return (NULL);
1866
1867 rw_enter(&nsrv4->rfs4_findclient_lock, RW_READER);
1868
1869 cp = (rfs4_client_t *)rfs4_dbsearch(nsrv4->rfs4_clientid_idx, &clientid,
1870 &create, NULL, RFS4_DBS_VALID);
1871
1872 rw_exit(&nsrv4->rfs4_findclient_lock);
1873
1874 if (cp && cp->rc_need_confirm && find_unconfirmed == FALSE) {
1875 rfs4_client_rele(cp);
1876 return (NULL);
1877 } else {
1878 return (cp);
1879 }
1880 }
1881
1882 static uint32_t
1883 clntip_hash(void *key)
1884 {
1885 struct sockaddr *addr = key;
1886 int i, len = 0;
1887 uint32_t hash = 0;
1888 char *ptr;
1889
1890 if (addr->sa_family == AF_INET) {
1891 struct sockaddr_in *a = (struct sockaddr_in *)addr;
1892 len = sizeof (struct in_addr);
1893 ptr = (char *)&a->sin_addr;
1894 } else if (addr->sa_family == AF_INET6) {
1895 struct sockaddr_in6 *a = (struct sockaddr_in6 *)addr;
1896 len = sizeof (struct in6_addr);
1897 ptr = (char *)&a->sin6_addr;
1898 } else
1899 return (0);
1900
1901 for (i = 0; i < len; i++) {
1902 hash <<= 1;
1903 hash += (uint_t)ptr[i];
1904 }
1905 return (hash);
1906 }
1907
1908 static bool_t
1909 clntip_compare(rfs4_entry_t entry, void *key)
1910 {
1911 rfs4_clntip_t *cp = (rfs4_clntip_t *)entry;
1912 struct sockaddr *addr = key;
1913 int len = 0;
1914 char *p1, *p2;
1915
1916 if (addr->sa_family == AF_INET) {
1917 struct sockaddr_in *a1 = (struct sockaddr_in *)&cp->ri_addr;
1918 struct sockaddr_in *a2 = (struct sockaddr_in *)addr;
1919 len = sizeof (struct in_addr);
1920 p1 = (char *)&a1->sin_addr;
1921 p2 = (char *)&a2->sin_addr;
1922 } else if (addr->sa_family == AF_INET6) {
1923 struct sockaddr_in6 *a1 = (struct sockaddr_in6 *)&cp->ri_addr;
1924 struct sockaddr_in6 *a2 = (struct sockaddr_in6 *)addr;
1925 len = sizeof (struct in6_addr);
1926 p1 = (char *)&a1->sin6_addr;
1927 p2 = (char *)&a2->sin6_addr;
1928 } else
1929 return (0);
1930
1931 return (bcmp(p1, p2, len) == 0);
1932 }
1933
1934 static void *
1935 clntip_mkkey(rfs4_entry_t entry)
1936 {
1937 rfs4_clntip_t *cp = (rfs4_clntip_t *)entry;
1938
1939 return (&cp->ri_addr);
1940 }
1941
1942 static bool_t
1943 rfs4_clntip_expiry(rfs4_entry_t u_entry)
1944 {
1945 rfs4_clntip_t *cp = (rfs4_clntip_t *)u_entry;
1946
1947 if (rfs4_dbe_is_invalid(cp->ri_dbe))
1948 return (TRUE);
1949 return (FALSE);
1950 }
1951
1952 /* ARGSUSED */
1953 static void
1954 rfs4_clntip_destroy(rfs4_entry_t u_entry)
1955 {
1956 }
1957
1958 static bool_t
1959 rfs4_clntip_create(rfs4_entry_t u_entry, void *arg)
1960 {
1961 rfs4_clntip_t *cp = (rfs4_clntip_t *)u_entry;
1962 struct sockaddr *ca = (struct sockaddr *)arg;
1963
1964 /* Copy client's IP address */
1965 if (ca->sa_family == AF_INET)
1966 bcopy(ca, &cp->ri_addr, sizeof (struct sockaddr_in));
1967 else if (ca->sa_family == AF_INET6)
1968 bcopy(ca, &cp->ri_addr, sizeof (struct sockaddr_in6));
1969 else
1970 return (FALSE);
1971 cp->ri_no_referrals = 1;
1972
1973 return (TRUE);
1974 }
1975
1976 rfs4_clntip_t *
1977 rfs4_find_clntip(struct sockaddr *addr, bool_t *create)
1978 {
1979 rfs4_clntip_t *cp;
1980 nfs4_srv_t *nsrv4;
1981
1982 nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
1983
1984 rw_enter(&nsrv4->rfs4_findclient_lock, RW_READER);
1985
1986 cp = (rfs4_clntip_t *)rfs4_dbsearch(nsrv4->rfs4_clntip_idx, addr,
1987 create, addr, RFS4_DBS_VALID);
1988
1989 rw_exit(&nsrv4->rfs4_findclient_lock);
1990
1991 return (cp);
1992 }
1993
1994 void
1995 rfs4_invalidate_clntip(struct sockaddr *addr)
1996 {
1997 rfs4_clntip_t *cp;
1998 bool_t create = FALSE;
1999 nfs4_srv_t *nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
2000
2001 rw_enter(&nsrv4->rfs4_findclient_lock, RW_READER);
2002
2003 cp = (rfs4_clntip_t *)rfs4_dbsearch(nsrv4->rfs4_clntip_idx, addr,
2004 &create, NULL, RFS4_DBS_VALID);
2005 if (cp == NULL) {
2006 rw_exit(&nsrv4->rfs4_findclient_lock);
2007 return;
2008 }
2009 rfs4_dbe_invalidate(cp->ri_dbe);
2010 rfs4_dbe_rele(cp->ri_dbe);
2011
2012 rw_exit(&nsrv4->rfs4_findclient_lock);
2013 }
2014
2015 bool_t
2016 rfs4_lease_expired(rfs4_client_t *cp)
2017 {
2018 bool_t rc;
2019
2020 rfs4_dbe_lock(cp->rc_dbe);
2021
2022 /*
2023 * If the admin has executed clear_locks for this
2024 * client id, force expire will be set, so no need
2025 * to calculate anything because it's "outa here".
2026 */
2027 if (cp->rc_forced_expire) {
2028 rc = TRUE;
2029 } else {
2030 rc = (gethrestime_sec() - cp->rc_last_access > rfs4_lease_time);
2031 }
2032
2033 /*
2034 * If the lease has expired we will also want
2035 * to remove any stable storage state data. So
2036 * mark the client id accordingly.
2037 */
2038 if (!cp->rc_ss_remove)
2039 cp->rc_ss_remove = (rc == TRUE);
2040
2041 rfs4_dbe_unlock(cp->rc_dbe);
2042
2043 return (rc);
2044 }
2045
2046 void
2047 rfs4_update_lease(rfs4_client_t *cp)
2048 {
2049 rfs4_dbe_lock(cp->rc_dbe);
2050 if (!cp->rc_forced_expire)
2051 cp->rc_last_access = gethrestime_sec();
2052 rfs4_dbe_unlock(cp->rc_dbe);
2053 }
2054
2055
2056 static bool_t
2057 EQOPENOWNER(open_owner4 *a, open_owner4 *b)
2058 {
2059 bool_t rc;
2060
2061 if (a->clientid != b->clientid)
2062 return (FALSE);
2063
2064 if (a->owner_len != b->owner_len)
2065 return (FALSE);
2066
2067 rc = (bcmp(a->owner_val, b->owner_val, a->owner_len) == 0);
2068
2069 return (rc);
2070 }
2071
2072 static uint_t
2073 openowner_hash(void *key)
2074 {
2075 int i;
2076 open_owner4 *openowner = key;
2077 uint_t hash = 0;
2078
2079 for (i = 0; i < openowner->owner_len; i++) {
2080 hash <<= 4;
2081 hash += (uint_t)openowner->owner_val[i];
2082 }
2083 hash += (uint_t)openowner->clientid;
2084 hash |= (openowner->clientid >> 32);
2085
2086 return (hash);
2087 }
2088
2089 static bool_t
2090 openowner_compare(rfs4_entry_t u_entry, void *key)
2091 {
2092 rfs4_openowner_t *oo = (rfs4_openowner_t *)u_entry;
2093 open_owner4 *arg = key;
2094
2095 return (EQOPENOWNER(&oo->ro_owner, arg));
2096 }
2097
2098 void *
2099 openowner_mkkey(rfs4_entry_t u_entry)
2100 {
2101 rfs4_openowner_t *oo = (rfs4_openowner_t *)u_entry;
2102
2103 return (&oo->ro_owner);
2104 }
2105
2106 /* ARGSUSED */
2107 static bool_t
2108 rfs4_openowner_expiry(rfs4_entry_t u_entry)
2109 {
2110 /* openstateid held us and did all needed delay */
2111 return (TRUE);
2112 }
2113
2114 static void
2115 rfs4_openowner_destroy(rfs4_entry_t u_entry)
2116 {
2117 rfs4_openowner_t *oo = (rfs4_openowner_t *)u_entry;
2118
2119 /* Remove open owner from client's lists of open owners */
2120 rfs4_dbe_lock(oo->ro_client->rc_dbe);
2121 list_remove(&oo->ro_client->rc_openownerlist, oo);
2122 rfs4_dbe_unlock(oo->ro_client->rc_dbe);
2123
2124 /* One less reference to the client */
2125 rfs4_client_rele(oo->ro_client);
2126 oo->ro_client = NULL;
2127
2128 /* Free the last reply for this lock owner */
2129 rfs4_free_reply(&oo->ro_reply);
2130
2131 if (oo->ro_reply_fh.nfs_fh4_val) {
2132 kmem_free(oo->ro_reply_fh.nfs_fh4_val,
2133 oo->ro_reply_fh.nfs_fh4_len);
2134 oo->ro_reply_fh.nfs_fh4_val = NULL;
2135 oo->ro_reply_fh.nfs_fh4_len = 0;
2136 }
2137
2138 rfs4_sw_destroy(&oo->ro_sw);
2139 list_destroy(&oo->ro_statelist);
2140
2141 /* Free the lock owner id */
2142 kmem_free(oo->ro_owner.owner_val, oo->ro_owner.owner_len);
2143 }
2144
2145 void
2146 rfs4_openowner_rele(rfs4_openowner_t *oo)
2147 {
2148 rfs4_dbe_rele(oo->ro_dbe);
2149 }
2150
2151 static bool_t
2152 rfs4_openowner_create(rfs4_entry_t u_entry, void *arg)
2153 {
2154 rfs4_openowner_t *oo = (rfs4_openowner_t *)u_entry;
2155 rfs4_openowner_t *argp = (rfs4_openowner_t *)arg;
2156 open_owner4 *openowner = &argp->ro_owner;
2157 seqid4 seqid = argp->ro_open_seqid;
2158 rfs4_client_t *cp;
2159 bool_t create = FALSE;
2160 nfs4_srv_t *nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
2161
2162 rw_enter(&nsrv4->rfs4_findclient_lock, RW_READER);
2163
2164 cp = (rfs4_client_t *)rfs4_dbsearch(nsrv4->rfs4_clientid_idx,
2165 &openowner->clientid,
2166 &create, NULL, RFS4_DBS_VALID);
2167
2168 rw_exit(&nsrv4->rfs4_findclient_lock);
2169
2170 if (cp == NULL)
2171 return (FALSE);
2172
2173 oo->ro_reply_fh.nfs_fh4_len = 0;
2174 oo->ro_reply_fh.nfs_fh4_val = NULL;
2175
2176 oo->ro_owner.clientid = openowner->clientid;
2177 oo->ro_owner.owner_val =
2178 kmem_alloc(openowner->owner_len, KM_SLEEP);
2179
2180 bcopy(openowner->owner_val,
2181 oo->ro_owner.owner_val, openowner->owner_len);
2182
2183 oo->ro_owner.owner_len = openowner->owner_len;
2184
2185 oo->ro_need_confirm = TRUE;
2186
2187 rfs4_sw_init(&oo->ro_sw);
2188
2189 oo->ro_open_seqid = seqid;
2190 bzero(&oo->ro_reply, sizeof (nfs_resop4));
2191 oo->ro_client = cp;
2192 oo->ro_cr_set = NULL;
2193
2194 list_create(&oo->ro_statelist, sizeof (rfs4_state_t),
2195 offsetof(rfs4_state_t, rs_node));
2196
2197 /* Insert openowner into client's open owner list */
2198 rfs4_dbe_lock(cp->rc_dbe);
2199 list_insert_tail(&cp->rc_openownerlist, oo);
2200 rfs4_dbe_unlock(cp->rc_dbe);
2201
2202 return (TRUE);
2203 }
2204
2205 rfs4_openowner_t *
2206 rfs4_findopenowner(open_owner4 *openowner, bool_t *create, seqid4 seqid)
2207 {
2208 rfs4_openowner_t *oo;
2209 rfs4_openowner_t arg;
2210 nfs4_srv_t *nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
2211
2212 arg.ro_owner = *openowner;
2213 arg.ro_open_seqid = seqid;
2214 /* CSTYLED */
2215 oo = (rfs4_openowner_t *)rfs4_dbsearch(nsrv4->rfs4_openowner_idx, openowner,
2216 create, &arg, RFS4_DBS_VALID);
2217
2218 return (oo);
2219 }
2220
2221 void
2222 rfs4_update_open_sequence(rfs4_openowner_t *oo)
2223 {
2224
2225 rfs4_dbe_lock(oo->ro_dbe);
2226
2227 oo->ro_open_seqid++;
2228
2229 rfs4_dbe_unlock(oo->ro_dbe);
2230 }
2231
2232 void
2233 rfs4_update_open_resp(rfs4_openowner_t *oo, nfs_resop4 *resp, nfs_fh4 *fh)
2234 {
2235
2236 rfs4_dbe_lock(oo->ro_dbe);
2237
2238 rfs4_free_reply(&oo->ro_reply);
2239
2240 rfs4_copy_reply(&oo->ro_reply, resp);
2241
2242 /* Save the filehandle if provided and free if not used */
2243 if (resp->nfs_resop4_u.opopen.status == NFS4_OK &&
2244 fh && fh->nfs_fh4_len) {
2245 if (oo->ro_reply_fh.nfs_fh4_val == NULL)
2246 oo->ro_reply_fh.nfs_fh4_val =
2247 kmem_alloc(fh->nfs_fh4_len, KM_SLEEP);
2248 nfs_fh4_copy(fh, &oo->ro_reply_fh);
2249 } else {
2250 if (oo->ro_reply_fh.nfs_fh4_val) {
2251 kmem_free(oo->ro_reply_fh.nfs_fh4_val,
2252 oo->ro_reply_fh.nfs_fh4_len);
2253 oo->ro_reply_fh.nfs_fh4_val = NULL;
2254 oo->ro_reply_fh.nfs_fh4_len = 0;
2255 }
2256 }
2257
2258 rfs4_dbe_unlock(oo->ro_dbe);
2259 }
2260
2261 static bool_t
2262 lockowner_compare(rfs4_entry_t u_entry, void *key)
2263 {
2264 rfs4_lockowner_t *lo = (rfs4_lockowner_t *)u_entry;
2265 lock_owner4 *b = (lock_owner4 *)key;
2266
2267 if (lo->rl_owner.clientid != b->clientid)
2268 return (FALSE);
2269
2270 if (lo->rl_owner.owner_len != b->owner_len)
2271 return (FALSE);
2272
2273 return (bcmp(lo->rl_owner.owner_val, b->owner_val,
2274 lo->rl_owner.owner_len) == 0);
2275 }
2276
2277 void *
2278 lockowner_mkkey(rfs4_entry_t u_entry)
2279 {
2280 rfs4_lockowner_t *lo = (rfs4_lockowner_t *)u_entry;
2281
2282 return (&lo->rl_owner);
2283 }
2284
2285 static uint32_t
2286 lockowner_hash(void *key)
2287 {
2288 int i;
2289 lock_owner4 *lockowner = key;
2290 uint_t hash = 0;
2291
2292 for (i = 0; i < lockowner->owner_len; i++) {
2293 hash <<= 4;
2294 hash += (uint_t)lockowner->owner_val[i];
2295 }
2296 hash += (uint_t)lockowner->clientid;
2297 hash |= (lockowner->clientid >> 32);
2298
2299 return (hash);
2300 }
2301
2302 static uint32_t
2303 pid_hash(void *key)
2304 {
2305 return ((uint32_t)(uintptr_t)key);
2306 }
2307
2308 static void *
2309 pid_mkkey(rfs4_entry_t u_entry)
2310 {
2311 rfs4_lockowner_t *lo = (rfs4_lockowner_t *)u_entry;
2312
2313 return ((void *)(uintptr_t)lo->rl_pid);
2314 }
2315
2316 static bool_t
2317 pid_compare(rfs4_entry_t u_entry, void *key)
2318 {
2319 rfs4_lockowner_t *lo = (rfs4_lockowner_t *)u_entry;
2320
2321 return (lo->rl_pid == (pid_t)(uintptr_t)key);
2322 }
2323
2324 static void
2325 rfs4_lockowner_destroy(rfs4_entry_t u_entry)
2326 {
2327 rfs4_lockowner_t *lo = (rfs4_lockowner_t *)u_entry;
2328
2329 /* Free the lock owner id */
2330 kmem_free(lo->rl_owner.owner_val, lo->rl_owner.owner_len);
2331 rfs4_client_rele(lo->rl_client);
2332 }
2333
2334 void
2335 rfs4_lockowner_rele(rfs4_lockowner_t *lo)
2336 {
2337 rfs4_dbe_rele(lo->rl_dbe);
2338 }
2339
2340 /* ARGSUSED */
2341 static bool_t
2342 rfs4_lockowner_expiry(rfs4_entry_t u_entry)
2343 {
2344 /*
2345 * Since expiry is called with no other references on
2346 * this struct, go ahead and have it removed.
2347 */
2348 return (TRUE);
2349 }
2350
2351 static bool_t
2352 rfs4_lockowner_create(rfs4_entry_t u_entry, void *arg)
2353 {
2354 rfs4_lockowner_t *lo = (rfs4_lockowner_t *)u_entry;
2355 lock_owner4 *lockowner = (lock_owner4 *)arg;
2356 rfs4_client_t *cp;
2357 bool_t create = FALSE;
2358 nfs4_srv_t *nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
2359
2360 rw_enter(&nsrv4->rfs4_findclient_lock, RW_READER);
2361
2362 cp = (rfs4_client_t *)rfs4_dbsearch(nsrv4->rfs4_clientid_idx,
2363 &lockowner->clientid,
2364 &create, NULL, RFS4_DBS_VALID);
2365
2366 rw_exit(&nsrv4->rfs4_findclient_lock);
2367
2368 if (cp == NULL)
2369 return (FALSE);
2370
2371 /* Reference client */
2372 lo->rl_client = cp;
2373 lo->rl_owner.clientid = lockowner->clientid;
2374 lo->rl_owner.owner_val = kmem_alloc(lockowner->owner_len, KM_SLEEP);
2375 bcopy(lockowner->owner_val, lo->rl_owner.owner_val,
2376 lockowner->owner_len);
2377 lo->rl_owner.owner_len = lockowner->owner_len;
2378 lo->rl_pid = rfs4_dbe_getid(lo->rl_dbe);
2379
2380 return (TRUE);
2381 }
2382
2383 rfs4_lockowner_t *
2384 rfs4_findlockowner(lock_owner4 *lockowner, bool_t *create)
2385 {
2386 rfs4_lockowner_t *lo;
2387 nfs4_srv_t *nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
2388
2389 /* CSTYLED */
2390 lo = (rfs4_lockowner_t *)rfs4_dbsearch(nsrv4->rfs4_lockowner_idx, lockowner,
2391 create, lockowner, RFS4_DBS_VALID);
2392
2393 return (lo);
2394 }
2395
2396 rfs4_lockowner_t *
2397 rfs4_findlockowner_by_pid(pid_t pid)
2398 {
2399 rfs4_lockowner_t *lo;
2400 bool_t create = FALSE;
2401 nfs4_srv_t *nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
2402
2403 lo = (rfs4_lockowner_t *)rfs4_dbsearch(nsrv4->rfs4_lockowner_pid_idx,
2404 (void *)(uintptr_t)pid, &create, NULL, RFS4_DBS_VALID);
2405
2406 return (lo);
2407 }
2408
2409
2410 static uint32_t
2411 file_hash(void *key)
2412 {
2413 return (ADDRHASH(key));
2414 }
2415
2416 static void *
2417 file_mkkey(rfs4_entry_t u_entry)
2418 {
2419 rfs4_file_t *fp = (rfs4_file_t *)u_entry;
2420
2421 return (fp->rf_vp);
2422 }
2423
2424 static bool_t
2425 file_compare(rfs4_entry_t u_entry, void *key)
2426 {
2427 rfs4_file_t *fp = (rfs4_file_t *)u_entry;
2428
2429 return (fp->rf_vp == (vnode_t *)key);
2430 }
2431
2432 static void
2433 rfs4_file_destroy(rfs4_entry_t u_entry)
2434 {
2435 rfs4_file_t *fp = (rfs4_file_t *)u_entry;
2436
2437 list_destroy(&fp->rf_delegstatelist);
2438
2439 if (fp->rf_filehandle.nfs_fh4_val)
2440 kmem_free(fp->rf_filehandle.nfs_fh4_val,
2441 fp->rf_filehandle.nfs_fh4_len);
2442 cv_destroy(fp->rf_dinfo.rd_recall_cv);
2443 if (fp->rf_vp) {
2444 vnode_t *vp = fp->rf_vp;
2445
2446 mutex_enter(&vp->v_vsd_lock);
2447 (void) vsd_set(vp, nfs4_srv_vkey, NULL);
2448 mutex_exit(&vp->v_vsd_lock);
2449 VN_RELE(vp);
2450 fp->rf_vp = NULL;
2451 }
2452 rw_destroy(&fp->rf_file_rwlock);
2453 }
2454
2455 /*
2456 * Used to unlock the underlying dbe struct only
2457 */
2458 void
2459 rfs4_file_rele(rfs4_file_t *fp)
2460 {
2461 rfs4_dbe_rele(fp->rf_dbe);
2462 }
2463
2464 typedef struct {
2465 vnode_t *vp;
2466 nfs_fh4 *fh;
2467 } rfs4_fcreate_arg;
2468
2469 static bool_t
2470 rfs4_file_create(rfs4_entry_t u_entry, void *arg)
2471 {
2472 rfs4_file_t *fp = (rfs4_file_t *)u_entry;
2473 rfs4_fcreate_arg *ap = (rfs4_fcreate_arg *)arg;
2474 vnode_t *vp = ap->vp;
2475 nfs_fh4 *fh = ap->fh;
2476
2477 VN_HOLD(vp);
2478
2479 fp->rf_filehandle.nfs_fh4_len = 0;
2480 fp->rf_filehandle.nfs_fh4_val = NULL;
2481 ASSERT(fh && fh->nfs_fh4_len);
2482 if (fh && fh->nfs_fh4_len) {
2483 fp->rf_filehandle.nfs_fh4_val =
2484 kmem_alloc(fh->nfs_fh4_len, KM_SLEEP);
2485 nfs_fh4_copy(fh, &fp->rf_filehandle);
2486 }
2487 fp->rf_vp = vp;
2488
2489 list_create(&fp->rf_delegstatelist, sizeof (rfs4_deleg_state_t),
2490 offsetof(rfs4_deleg_state_t, rds_node));
2491
2492 fp->rf_share_deny = fp->rf_share_access = fp->rf_access_read = 0;
2493 fp->rf_access_write = fp->rf_deny_read = fp->rf_deny_write = 0;
2494
2495 mutex_init(fp->rf_dinfo.rd_recall_lock, NULL, MUTEX_DEFAULT, NULL);
2496 cv_init(fp->rf_dinfo.rd_recall_cv, NULL, CV_DEFAULT, NULL);
2497
2498 fp->rf_dinfo.rd_dtype = OPEN_DELEGATE_NONE;
2499
2500 rw_init(&fp->rf_file_rwlock, NULL, RW_DEFAULT, NULL);
2501
2502 mutex_enter(&vp->v_vsd_lock);
2503 VERIFY(vsd_set(vp, nfs4_srv_vkey, (void *)fp) == 0);
2504 mutex_exit(&vp->v_vsd_lock);
2505
2506 return (TRUE);
2507 }
2508
2509 rfs4_file_t *
2510 rfs4_findfile(vnode_t *vp, nfs_fh4 *fh, bool_t *create)
2511 {
2512 rfs4_file_t *fp;
2513 rfs4_fcreate_arg arg;
2514 nfs4_srv_t *nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
2515
2516 arg.vp = vp;
2517 arg.fh = fh;
2518
2519 if (*create == TRUE)
2520 /* CSTYLED */
2521 fp = (rfs4_file_t *)rfs4_dbsearch(nsrv4->rfs4_file_idx, vp, create,
2522 &arg, RFS4_DBS_VALID);
2523 else {
2524 mutex_enter(&vp->v_vsd_lock);
2525 fp = (rfs4_file_t *)vsd_get(vp, nfs4_srv_vkey);
2526 if (fp) {
2527 rfs4_dbe_lock(fp->rf_dbe);
2528 if (rfs4_dbe_is_invalid(fp->rf_dbe) ||
2529 (rfs4_dbe_refcnt(fp->rf_dbe) == 0)) {
2530 rfs4_dbe_unlock(fp->rf_dbe);
2531 fp = NULL;
2532 } else {
2533 rfs4_dbe_hold(fp->rf_dbe);
2534 rfs4_dbe_unlock(fp->rf_dbe);
2535 }
2536 }
2537 mutex_exit(&vp->v_vsd_lock);
2538 }
2539 return (fp);
2540 }
2541
2542 /*
2543 * Find a file in the db and once it is located, take the rw lock.
2544 * Need to check the vnode pointer and if it does not exist (it was
2545 * removed between the db location and check) redo the find. This
2546 * assumes that a file struct that has a NULL vnode pointer is marked
2547 * at 'invalid' and will not be found in the db the second time
2548 * around.
2549 */
2550 rfs4_file_t *
2551 rfs4_findfile_withlock(vnode_t *vp, nfs_fh4 *fh, bool_t *create)
2552 {
2553 rfs4_file_t *fp;
2554 rfs4_fcreate_arg arg;
2555 bool_t screate = *create;
2556 nfs4_srv_t *nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
2557
2558 if (screate == FALSE) {
2559 mutex_enter(&vp->v_vsd_lock);
2560 fp = (rfs4_file_t *)vsd_get(vp, nfs4_srv_vkey);
2561 if (fp) {
2562 rfs4_dbe_lock(fp->rf_dbe);
2563 if (rfs4_dbe_is_invalid(fp->rf_dbe) ||
2564 (rfs4_dbe_refcnt(fp->rf_dbe) == 0)) {
2565 rfs4_dbe_unlock(fp->rf_dbe);
2566 mutex_exit(&vp->v_vsd_lock);
2567 fp = NULL;
2568 } else {
2569 rfs4_dbe_hold(fp->rf_dbe);
2570 rfs4_dbe_unlock(fp->rf_dbe);
2571 mutex_exit(&vp->v_vsd_lock);
2572 rw_enter(&fp->rf_file_rwlock, RW_WRITER);
2573 if (fp->rf_vp == NULL) {
2574 rw_exit(&fp->rf_file_rwlock);
2575 rfs4_file_rele(fp);
2576 fp = NULL;
2577 }
2578 }
2579 } else {
2580 mutex_exit(&vp->v_vsd_lock);
2581 }
2582 } else {
2583 retry:
2584 arg.vp = vp;
2585 arg.fh = fh;
2586
2587 fp = (rfs4_file_t *)rfs4_dbsearch(nsrv4->rfs4_file_idx, vp,
2588 create, &arg, RFS4_DBS_VALID);
2589 if (fp != NULL) {
2590 rw_enter(&fp->rf_file_rwlock, RW_WRITER);
2591 if (fp->rf_vp == NULL) {
2592 rw_exit(&fp->rf_file_rwlock);
2593 rfs4_file_rele(fp);
2594 *create = screate;
2595 goto retry;
2596 }
2597 }
2598 }
2599
2600 return (fp);
2601 }
2602
2603 static uint32_t
2604 lo_state_hash(void *key)
2605 {
2606 stateid_t *id = key;
2607
2608 return (id->bits.ident+id->bits.pid);
2609 }
2610
2611 static bool_t
2612 lo_state_compare(rfs4_entry_t u_entry, void *key)
2613 {
2614 rfs4_lo_state_t *lsp = (rfs4_lo_state_t *)u_entry;
2615 stateid_t *id = key;
2616 bool_t rc;
2617
2618 rc = (lsp->rls_lockid.bits.boottime == id->bits.boottime &&
2619 lsp->rls_lockid.bits.type == id->bits.type &&
2620 lsp->rls_lockid.bits.ident == id->bits.ident &&
2621 lsp->rls_lockid.bits.pid == id->bits.pid);
2622
2623 return (rc);
2624 }
2625
2626 static void *
2627 lo_state_mkkey(rfs4_entry_t u_entry)
2628 {
2629 rfs4_lo_state_t *lsp = (rfs4_lo_state_t *)u_entry;
2630
2631 return (&lsp->rls_lockid);
2632 }
2633
2634 static bool_t
2635 rfs4_lo_state_expiry(rfs4_entry_t u_entry)
2636 {
2637 rfs4_lo_state_t *lsp = (rfs4_lo_state_t *)u_entry;
2638
2639 if (rfs4_dbe_is_invalid(lsp->rls_dbe))
2640 return (TRUE);
2641 if (lsp->rls_state->rs_closed)
2642 return (TRUE);
2643 return ((gethrestime_sec() -
2644 lsp->rls_state->rs_owner->ro_client->rc_last_access
2645 > rfs4_lease_time));
2646 }
2647
2648 static void
2649 rfs4_lo_state_destroy(rfs4_entry_t u_entry)
2650 {
2651 rfs4_lo_state_t *lsp = (rfs4_lo_state_t *)u_entry;
2652
2653 rfs4_dbe_lock(lsp->rls_state->rs_dbe);
2654 list_remove(&lsp->rls_state->rs_lostatelist, lsp);
2655 rfs4_dbe_unlock(lsp->rls_state->rs_dbe);
2656
2657 rfs4_sw_destroy(&lsp->rls_sw);
2658
2659 /* Make sure to release the file locks */
2660 if (lsp->rls_locks_cleaned == FALSE) {
2661 lsp->rls_locks_cleaned = TRUE;
2662 if (lsp->rls_locker->rl_client->rc_sysidt != LM_NOSYSID) {
2663 /* Is the PxFS kernel module loaded? */
2664 if (lm_remove_file_locks != NULL) {
2665 int new_sysid;
2666
2667 /* Encode the cluster nodeid in new sysid */
2668 new_sysid =
2669 lsp->rls_locker->rl_client->rc_sysidt;
2670 lm_set_nlmid_flk(&new_sysid);
2671
2672 /*
2673 * This PxFS routine removes file locks for a
2674 * client over all nodes of a cluster.
2675 */
2676 DTRACE_PROBE1(nfss_i_clust_rm_lck,
2677 int, new_sysid);
2678 (*lm_remove_file_locks)(new_sysid);
2679 } else {
2680 (void) cleanlocks(
2681 lsp->rls_state->rs_finfo->rf_vp,
2682 lsp->rls_locker->rl_pid,
2683 lsp->rls_locker->rl_client->rc_sysidt);
2684 }
2685 }
2686 }
2687
2688 /* Free the last reply for this state */
2689 rfs4_free_reply(&lsp->rls_reply);
2690
2691 rfs4_lockowner_rele(lsp->rls_locker);
2692 lsp->rls_locker = NULL;
2693
2694 rfs4_state_rele_nounlock(lsp->rls_state);
2695 lsp->rls_state = NULL;
2696 }
2697
2698 static bool_t
2699 rfs4_lo_state_create(rfs4_entry_t u_entry, void *arg)
2700 {
2701 rfs4_lo_state_t *lsp = (rfs4_lo_state_t *)u_entry;
2702 rfs4_lo_state_t *argp = (rfs4_lo_state_t *)arg;
2703 rfs4_lockowner_t *lo = argp->rls_locker;
2704 rfs4_state_t *sp = argp->rls_state;
2705
2706 lsp->rls_state = sp;
2707
2708 lsp->rls_lockid = sp->rs_stateid;
2709 lsp->rls_lockid.bits.type = LOCKID;
2710 lsp->rls_lockid.bits.chgseq = 0;
2711 lsp->rls_lockid.bits.pid = lo->rl_pid;
2712
2713 lsp->rls_locks_cleaned = FALSE;
2714 lsp->rls_lock_completed = FALSE;
2715
2716 rfs4_sw_init(&lsp->rls_sw);
2717
2718 /* Attached the supplied lock owner */
2719 rfs4_dbe_hold(lo->rl_dbe);
2720 lsp->rls_locker = lo;
2721
2722 rfs4_dbe_lock(sp->rs_dbe);
2723 list_insert_tail(&sp->rs_lostatelist, lsp);
2724 rfs4_dbe_hold(sp->rs_dbe);
2725 rfs4_dbe_unlock(sp->rs_dbe);
2726
2727 return (TRUE);
2728 }
2729
2730 void
2731 rfs4_lo_state_rele(rfs4_lo_state_t *lsp, bool_t unlock_fp)
2732 {
2733 if (unlock_fp == TRUE)
2734 rw_exit(&lsp->rls_state->rs_finfo->rf_file_rwlock);
2735 rfs4_dbe_rele(lsp->rls_dbe);
2736 }
2737
2738 static rfs4_lo_state_t *
2739 rfs4_findlo_state(stateid_t *id, bool_t lock_fp)
2740 {
2741 rfs4_lo_state_t *lsp;
2742 bool_t create = FALSE;
2743 nfs4_srv_t *nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
2744
2745 lsp = (rfs4_lo_state_t *)rfs4_dbsearch(nsrv4->rfs4_lo_state_idx, id,
2746 &create, NULL, RFS4_DBS_VALID);
2747 if (lock_fp == TRUE && lsp != NULL)
2748 rw_enter(&lsp->rls_state->rs_finfo->rf_file_rwlock, RW_READER);
2749
2750 return (lsp);
2751 }
2752
2753
2754 static uint32_t
2755 lo_state_lo_hash(void *key)
2756 {
2757 rfs4_lo_state_t *lsp = key;
2758
2759 return (ADDRHASH(lsp->rls_locker) ^ ADDRHASH(lsp->rls_state));
2760 }
2761
2762 static bool_t
2763 lo_state_lo_compare(rfs4_entry_t u_entry, void *key)
2764 {
2765 rfs4_lo_state_t *lsp = (rfs4_lo_state_t *)u_entry;
2766 rfs4_lo_state_t *keyp = key;
2767
2768 return (keyp->rls_locker == lsp->rls_locker &&
2769 keyp->rls_state == lsp->rls_state);
2770 }
2771
2772 static void *
2773 lo_state_lo_mkkey(rfs4_entry_t u_entry)
2774 {
2775 return (u_entry);
2776 }
2777
2778 rfs4_lo_state_t *
2779 rfs4_findlo_state_by_owner(rfs4_lockowner_t *lo, rfs4_state_t *sp,
2780 bool_t *create)
2781 {
2782 rfs4_lo_state_t *lsp;
2783 rfs4_lo_state_t arg;
2784 nfs4_srv_t *nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
2785
2786 arg.rls_locker = lo;
2787 arg.rls_state = sp;
2788
2789 lsp = (rfs4_lo_state_t *)rfs4_dbsearch(nsrv4->rfs4_lo_state_owner_idx,
2790 &arg, create, &arg, RFS4_DBS_VALID);
2791
2792 return (lsp);
2793 }
2794
2795 static stateid_t
2796 get_stateid(id_t eid)
2797 {
2798 stateid_t id;
2799 nfs4_srv_t *nsrv4;
2800
2801 nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
2802
2803 id.bits.boottime = nsrv4->rfs4_start_time;
2804 id.bits.ident = eid;
2805 id.bits.chgseq = 0;
2806 id.bits.type = 0;
2807 id.bits.pid = 0;
2808
2809 /*
2810 * If we are booted as a cluster node, embed our nodeid.
2811 * We've already done sanity checks in rfs4_client_create() so no
2812 * need to repeat them here.
2813 */
2814 id.bits.clnodeid = (cluster_bootflags & CLUSTER_BOOTED) ?
2815 clconf_get_nodeid() : 0;
2816
2817 return (id);
2818 }
2819
2820 /*
2821 * For use only when booted as a cluster node.
2822 * Returns TRUE if the embedded nodeid indicates that this stateid was
2823 * generated on another node.
2824 */
2825 static int
2826 foreign_stateid(stateid_t *id)
2827 {
2828 ASSERT(cluster_bootflags & CLUSTER_BOOTED);
2829 return (id->bits.clnodeid != (uint32_t)clconf_get_nodeid());
2830 }
2831
2832 /*
2833 * For use only when booted as a cluster node.
2834 * Returns TRUE if the embedded nodeid indicates that this clientid was
2835 * generated on another node.
2836 */
2837 static int
2838 foreign_clientid(cid *cidp)
2839 {
2840 ASSERT(cluster_bootflags & CLUSTER_BOOTED);
2841 return (cidp->impl_id.c_id >> CLUSTER_NODEID_SHIFT !=
2842 (uint32_t)clconf_get_nodeid());
2843 }
2844
2845 /*
2846 * For use only when booted as a cluster node.
2847 * Embed our cluster nodeid into the clientid.
2848 */
2849 static void
2850 embed_nodeid(cid *cidp)
2851 {
2852 int clnodeid;
2853 /*
2854 * Currently, our state tables are small enough that their
2855 * ids will leave enough bits free for the nodeid. If the
2856 * tables become larger, we mustn't overwrite the id.
2857 * Equally, we only have room for so many bits of nodeid, so
2858 * must check that too.
2859 */
2860 ASSERT(cluster_bootflags & CLUSTER_BOOTED);
2861 ASSERT(cidp->impl_id.c_id >> CLUSTER_NODEID_SHIFT == 0);
2862 clnodeid = clconf_get_nodeid();
2863 ASSERT(clnodeid <= CLUSTER_MAX_NODEID);
2864 ASSERT(clnodeid != NODEID_UNKNOWN);
2865 cidp->impl_id.c_id |= (clnodeid << CLUSTER_NODEID_SHIFT);
2866 }
2867
2868 static uint32_t
2869 state_hash(void *key)
2870 {
2871 stateid_t *ip = (stateid_t *)key;
2872
2873 return (ip->bits.ident);
2874 }
2875
2876 static bool_t
2877 state_compare(rfs4_entry_t u_entry, void *key)
2878 {
2879 rfs4_state_t *sp = (rfs4_state_t *)u_entry;
2880 stateid_t *id = (stateid_t *)key;
2881 bool_t rc;
2882
2883 rc = (sp->rs_stateid.bits.boottime == id->bits.boottime &&
2884 sp->rs_stateid.bits.ident == id->bits.ident);
2885
2886 return (rc);
2887 }
2888
2889 static void *
2890 state_mkkey(rfs4_entry_t u_entry)
2891 {
2892 rfs4_state_t *sp = (rfs4_state_t *)u_entry;
2893
2894 return (&sp->rs_stateid);
2895 }
2896
2897 static void
2898 rfs4_state_destroy(rfs4_entry_t u_entry)
2899 {
2900 rfs4_state_t *sp = (rfs4_state_t *)u_entry;
2901
2902 /* remove from openowner list */
2903 rfs4_dbe_lock(sp->rs_owner->ro_dbe);
2904 list_remove(&sp->rs_owner->ro_statelist, sp);
2905 rfs4_dbe_unlock(sp->rs_owner->ro_dbe);
2906
2907 list_destroy(&sp->rs_lostatelist);
2908
2909 /* release any share locks for this stateid if it's still open */
2910 if (!sp->rs_closed) {
2911 rfs4_dbe_lock(sp->rs_dbe);
2912 (void) rfs4_unshare(sp);
2913 rfs4_dbe_unlock(sp->rs_dbe);
2914 }
2915
2916 /* Were done with the file */
2917 rfs4_file_rele(sp->rs_finfo);
2918 sp->rs_finfo = NULL;
2919
2920 /* And now with the openowner */
2921 rfs4_openowner_rele(sp->rs_owner);
2922 sp->rs_owner = NULL;
2923 }
2924
2925 static void
2926 rfs4_state_rele_nounlock(rfs4_state_t *sp)
2927 {
2928 rfs4_dbe_rele(sp->rs_dbe);
2929 }
2930
2931 void
2932 rfs4_state_rele(rfs4_state_t *sp)
2933 {
2934 rw_exit(&sp->rs_finfo->rf_file_rwlock);
2935 rfs4_dbe_rele(sp->rs_dbe);
2936 }
2937
2938 static uint32_t
2939 deleg_hash(void *key)
2940 {
2941 rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)key;
2942
2943 return (ADDRHASH(dsp->rds_client) ^ ADDRHASH(dsp->rds_finfo));
2944 }
2945
2946 static bool_t
2947 deleg_compare(rfs4_entry_t u_entry, void *key)
2948 {
2949 rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
2950 rfs4_deleg_state_t *kdsp = (rfs4_deleg_state_t *)key;
2951
2952 return (dsp->rds_client == kdsp->rds_client &&
2953 dsp->rds_finfo == kdsp->rds_finfo);
2954 }
2955
2956 static void *
2957 deleg_mkkey(rfs4_entry_t u_entry)
2958 {
2959 return (u_entry);
2960 }
2961
2962 static uint32_t
2963 deleg_state_hash(void *key)
2964 {
2965 stateid_t *ip = (stateid_t *)key;
2966
2967 return (ip->bits.ident);
2968 }
2969
2970 static bool_t
2971 deleg_state_compare(rfs4_entry_t u_entry, void *key)
2972 {
2973 rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
2974 stateid_t *id = (stateid_t *)key;
2975 bool_t rc;
2976
2977 if (id->bits.type != DELEGID)
2978 return (FALSE);
2979
2980 rc = (dsp->rds_delegid.bits.boottime == id->bits.boottime &&
2981 dsp->rds_delegid.bits.ident == id->bits.ident);
2982
2983 return (rc);
2984 }
2985
2986 static void *
2987 deleg_state_mkkey(rfs4_entry_t u_entry)
2988 {
2989 rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
2990
2991 return (&dsp->rds_delegid);
2992 }
2993
2994 static bool_t
2995 rfs4_deleg_state_expiry(rfs4_entry_t u_entry)
2996 {
2997 rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
2998
2999 if (rfs4_dbe_is_invalid(dsp->rds_dbe))
3000 return (TRUE);
3001
3002 if (dsp->rds_dtype == OPEN_DELEGATE_NONE)
3003 return (TRUE);
3004
3005 if ((gethrestime_sec() - dsp->rds_client->rc_last_access
3006 > rfs4_lease_time)) {
3007 rfs4_dbe_invalidate(dsp->rds_dbe);
3008 return (TRUE);
3009 }
3010
3011 return (FALSE);
3012 }
3013
3014 static bool_t
3015 rfs4_deleg_state_create(rfs4_entry_t u_entry, void *argp)
3016 {
3017 rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
3018 rfs4_file_t *fp = ((rfs4_deleg_state_t *)argp)->rds_finfo;
3019 rfs4_client_t *cp = ((rfs4_deleg_state_t *)argp)->rds_client;
3020
3021 rfs4_dbe_hold(fp->rf_dbe);
3022 rfs4_dbe_hold(cp->rc_dbe);
3023
3024 dsp->rds_delegid = get_stateid(rfs4_dbe_getid(dsp->rds_dbe));
3025 dsp->rds_delegid.bits.type = DELEGID;
3026 dsp->rds_finfo = fp;
3027 dsp->rds_client = cp;
3028 dsp->rds_dtype = OPEN_DELEGATE_NONE;
3029
3030 dsp->rds_time_granted = gethrestime_sec(); /* observability */
3031 dsp->rds_time_revoked = 0;
3032
3033 list_link_init(&dsp->rds_node);
3034
3035 return (TRUE);
3036 }
3037
3038 static void
3039 rfs4_deleg_state_destroy(rfs4_entry_t u_entry)
3040 {
3041 rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
3042
3043 /* return delegation if necessary */
3044 rfs4_return_deleg(dsp, FALSE);
3045
3046 /* Were done with the file */
3047 rfs4_file_rele(dsp->rds_finfo);
3048 dsp->rds_finfo = NULL;
3049
3050 /* And now with the openowner */
3051 rfs4_client_rele(dsp->rds_client);
3052 dsp->rds_client = NULL;
3053 }
3054
3055 rfs4_deleg_state_t *
3056 rfs4_finddeleg(rfs4_state_t *sp, bool_t *create)
3057 {
3058 rfs4_deleg_state_t ds, *dsp;
3059 nfs4_srv_t *nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
3060
3061 ds.rds_client = sp->rs_owner->ro_client;
3062 ds.rds_finfo = sp->rs_finfo;
3063
3064 dsp = (rfs4_deleg_state_t *)rfs4_dbsearch(nsrv4->rfs4_deleg_idx, &ds,
3065 create, &ds, RFS4_DBS_VALID);
3066
3067 return (dsp);
3068 }
3069
3070 rfs4_deleg_state_t *
3071 rfs4_finddelegstate(stateid_t *id)
3072 {
3073 rfs4_deleg_state_t *dsp;
3074 bool_t create = FALSE;
3075 nfs4_srv_t *nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
3076
3077 dsp = (rfs4_deleg_state_t *)rfs4_dbsearch(nsrv4->rfs4_deleg_state_idx,
3078 id, &create, NULL, RFS4_DBS_VALID);
3079
3080 return (dsp);
3081 }
3082
3083 void
3084 rfs4_deleg_state_rele(rfs4_deleg_state_t *dsp)
3085 {
3086 rfs4_dbe_rele(dsp->rds_dbe);
3087 }
3088
3089 void
3090 rfs4_update_lock_sequence(rfs4_lo_state_t *lsp)
3091 {
3092
3093 rfs4_dbe_lock(lsp->rls_dbe);
3094
3095 /*
3096 * If we are skipping sequence id checking, this means that
3097 * this is the first lock request and therefore the sequence
3098 * id does not need to be updated. This only happens on the
3099 * first lock request for a lockowner
3100 */
3101 if (!lsp->rls_skip_seqid_check)
3102 lsp->rls_seqid++;
3103
3104 rfs4_dbe_unlock(lsp->rls_dbe);
3105 }
3106
3107 void
3108 rfs4_update_lock_resp(rfs4_lo_state_t *lsp, nfs_resop4 *resp)
3109 {
3110
3111 rfs4_dbe_lock(lsp->rls_dbe);
3112
3113 rfs4_free_reply(&lsp->rls_reply);
3114
3115 rfs4_copy_reply(&lsp->rls_reply, resp);
3116
3117 rfs4_dbe_unlock(lsp->rls_dbe);
3118 }
3119
3120 void
3121 rfs4_free_opens(rfs4_openowner_t *oo, bool_t invalidate,
3122 bool_t close_of_client)
3123 {
3124 rfs4_state_t *sp;
3125
3126 rfs4_dbe_lock(oo->ro_dbe);
3127
3128 for (sp = list_head(&oo->ro_statelist); sp != NULL;
3129 sp = list_next(&oo->ro_statelist, sp)) {
3130 rfs4_state_close(sp, FALSE, close_of_client, CRED());
3131 if (invalidate == TRUE)
3132 rfs4_dbe_invalidate(sp->rs_dbe);
3133 }
3134
3135 rfs4_dbe_invalidate(oo->ro_dbe);
3136 rfs4_dbe_unlock(oo->ro_dbe);
3137 }
3138
3139 static uint32_t
3140 state_owner_file_hash(void *key)
3141 {
3142 rfs4_state_t *sp = key;
3143
3144 return (ADDRHASH(sp->rs_owner) ^ ADDRHASH(sp->rs_finfo));
3145 }
3146
3147 static bool_t
3148 state_owner_file_compare(rfs4_entry_t u_entry, void *key)
3149 {
3150 rfs4_state_t *sp = (rfs4_state_t *)u_entry;
3151 rfs4_state_t *arg = key;
3152
3153 if (sp->rs_closed == TRUE)
3154 return (FALSE);
3155
3156 return (arg->rs_owner == sp->rs_owner && arg->rs_finfo == sp->rs_finfo);
3157 }
3158
3159 static void *
3160 state_owner_file_mkkey(rfs4_entry_t u_entry)
3161 {
3162 return (u_entry);
3163 }
3164
3165 static uint32_t
3166 state_file_hash(void *key)
3167 {
3168 return (ADDRHASH(key));
3169 }
3170
3171 static bool_t
3172 state_file_compare(rfs4_entry_t u_entry, void *key)
3173 {
3174 rfs4_state_t *sp = (rfs4_state_t *)u_entry;
3175 rfs4_file_t *fp = key;
3176
3177 if (sp->rs_closed == TRUE)
3178 return (FALSE);
3179
3180 return (fp == sp->rs_finfo);
3181 }
3182
3183 static void *
3184 state_file_mkkey(rfs4_entry_t u_entry)
3185 {
3186 rfs4_state_t *sp = (rfs4_state_t *)u_entry;
3187
3188 return (sp->rs_finfo);
3189 }
3190
3191 rfs4_state_t *
3192 rfs4_findstate_by_owner_file(rfs4_openowner_t *oo, rfs4_file_t *fp,
3193 bool_t *create)
3194 {
3195 rfs4_state_t *sp;
3196 rfs4_state_t key;
3197 nfs4_srv_t *nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
3198
3199 key.rs_owner = oo;
3200 key.rs_finfo = fp;
3201
3202 sp = (rfs4_state_t *)rfs4_dbsearch(nsrv4->rfs4_state_owner_file_idx,
3203 &key, create, &key, RFS4_DBS_VALID);
3204
3205 return (sp);
3206 }
3207
3208 /* This returns ANY state struct that refers to this file */
3209 static rfs4_state_t *
3210 rfs4_findstate_by_file(rfs4_file_t *fp)
3211 {
3212 bool_t create = FALSE;
3213 nfs4_srv_t *nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
3214
3215 return ((rfs4_state_t *)rfs4_dbsearch(nsrv4->rfs4_state_file_idx, fp,
3216 &create, fp, RFS4_DBS_VALID));
3217 }
3218
3219 static bool_t
3220 rfs4_state_expiry(rfs4_entry_t u_entry)
3221 {
3222 rfs4_state_t *sp = (rfs4_state_t *)u_entry;
3223
3224 if (rfs4_dbe_is_invalid(sp->rs_dbe))
3225 return (TRUE);
3226
3227 if (sp->rs_closed == TRUE &&
3228 ((gethrestime_sec() - rfs4_dbe_get_timerele(sp->rs_dbe))
3229 > rfs4_lease_time))
3230 return (TRUE);
3231
3232 return ((gethrestime_sec() - sp->rs_owner->ro_client->rc_last_access
3233 > rfs4_lease_time));
3234 }
3235
3236 static bool_t
3237 rfs4_state_create(rfs4_entry_t u_entry, void *argp)
3238 {
3239 rfs4_state_t *sp = (rfs4_state_t *)u_entry;
3240 rfs4_file_t *fp = ((rfs4_state_t *)argp)->rs_finfo;
3241 rfs4_openowner_t *oo = ((rfs4_state_t *)argp)->rs_owner;
3242
3243 rfs4_dbe_hold(fp->rf_dbe);
3244 rfs4_dbe_hold(oo->ro_dbe);
3245 sp->rs_stateid = get_stateid(rfs4_dbe_getid(sp->rs_dbe));
3246 sp->rs_stateid.bits.type = OPENID;
3247 sp->rs_owner = oo;
3248 sp->rs_finfo = fp;
3249
3250 list_create(&sp->rs_lostatelist, sizeof (rfs4_lo_state_t),
3251 offsetof(rfs4_lo_state_t, rls_node));
3252
3253 /* Insert state on per open owner's list */
3254 rfs4_dbe_lock(oo->ro_dbe);
3255 list_insert_tail(&oo->ro_statelist, sp);
3256 rfs4_dbe_unlock(oo->ro_dbe);
3257
3258 return (TRUE);
3259 }
3260
3261 static rfs4_state_t *
3262 rfs4_findstate(stateid_t *id, rfs4_dbsearch_type_t find_invalid, bool_t lock_fp)
3263 {
3264 rfs4_state_t *sp;
3265 bool_t create = FALSE;
3266 nfs4_srv_t *nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
3267
3268 sp = (rfs4_state_t *)rfs4_dbsearch(nsrv4->rfs4_state_idx, id,
3269 &create, NULL, find_invalid);
3270 if (lock_fp == TRUE && sp != NULL)
3271 rw_enter(&sp->rs_finfo->rf_file_rwlock, RW_READER);
3272
3273 return (sp);
3274 }
3275
3276 void
3277 rfs4_state_close(rfs4_state_t *sp, bool_t lock_held, bool_t close_of_client,
3278 cred_t *cr)
3279 {
3280 /* Remove the associated lo_state owners */
3281 if (!lock_held)
3282 rfs4_dbe_lock(sp->rs_dbe);
3283
3284 /*
3285 * If refcnt == 0, the dbe is about to be destroyed.
3286 * lock state will be released by the reaper thread.
3287 */
3288
3289 if (rfs4_dbe_refcnt(sp->rs_dbe) > 0) {
3290 if (sp->rs_closed == FALSE) {
3291 rfs4_release_share_lock_state(sp, cr, close_of_client);
3292 sp->rs_closed = TRUE;
3293 }
3294 }
3295
3296 if (!lock_held)
3297 rfs4_dbe_unlock(sp->rs_dbe);
3298 }
3299
3300 /*
3301 * Remove all state associated with the given client.
3302 */
3303 void
3304 rfs4_client_state_remove(rfs4_client_t *cp)
3305 {
3306 rfs4_openowner_t *oo;
3307
3308 rfs4_dbe_lock(cp->rc_dbe);
3309
3310 for (oo = list_head(&cp->rc_openownerlist); oo != NULL;
3311 oo = list_next(&cp->rc_openownerlist, oo)) {
3312 rfs4_free_opens(oo, TRUE, TRUE);
3313 }
3314
3315 rfs4_dbe_unlock(cp->rc_dbe);
3316 }
3317
3318 void
3319 rfs4_client_close(rfs4_client_t *cp)
3320 {
3321 /* Mark client as going away. */
3322 rfs4_dbe_lock(cp->rc_dbe);
3323 rfs4_dbe_invalidate(cp->rc_dbe);
3324 rfs4_dbe_unlock(cp->rc_dbe);
3325
3326 rfs4_client_state_remove(cp);
3327
3328 /* Release the client */
3329 rfs4_client_rele(cp);
3330 }
3331
3332 nfsstat4
3333 rfs4_check_clientid(clientid4 *cp, int setclid_confirm)
3334 {
3335 cid *cidp = (cid *) cp;
3336 nfs4_srv_t *nsrv4;
3337
3338 nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
3339
3340 /*
3341 * If we are booted as a cluster node, check the embedded nodeid.
3342 * If it indicates that this clientid was generated on another node,
3343 * inform the client accordingly.
3344 */
3345 if (cluster_bootflags & CLUSTER_BOOTED && foreign_clientid(cidp))
3346 return (NFS4ERR_STALE_CLIENTID);
3347
3348 /*
3349 * If the server start time matches the time provided
3350 * by the client (via the clientid) and this is NOT a
3351 * setclientid_confirm then return EXPIRED.
3352 */
3353 if (!setclid_confirm &&
3354 cidp->impl_id.start_time == nsrv4->rfs4_start_time)
3355 return (NFS4ERR_EXPIRED);
3356
3357 return (NFS4ERR_STALE_CLIENTID);
3358 }
3359
3360 /*
3361 * This is used when a stateid has not been found amongst the
3362 * current server's state. Check the stateid to see if it
3363 * was from this server instantiation or not.
3364 */
3365 static nfsstat4
3366 what_stateid_error(stateid_t *id, stateid_type_t type)
3367 {
3368 nfs4_srv_t *nsrv4;
3369
3370 nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
3371
3372 /* If we are booted as a cluster node, was stateid locally generated? */
3373 if ((cluster_bootflags & CLUSTER_BOOTED) && foreign_stateid(id))
3374 return (NFS4ERR_STALE_STATEID);
3375
3376 /* If types don't match then no use checking further */
3377 if (type != id->bits.type)
3378 return (NFS4ERR_BAD_STATEID);
3379
3380 /* From a different server instantiation, return STALE */
3381 if (id->bits.boottime != nsrv4->rfs4_start_time)
3382 return (NFS4ERR_STALE_STATEID);
3383
3384 /*
3385 * From this server but the state is most likely beyond lease
3386 * timeout: return NFS4ERR_EXPIRED. However, there is the
3387 * case of a delegation stateid. For delegations, there is a
3388 * case where the state can be removed without the client's
3389 * knowledge/consent: revocation. In the case of delegation
3390 * revocation, the delegation state will be removed and will
3391 * not be found. If the client does something like a
3392 * DELEGRETURN or even a READ/WRITE with a delegatoin stateid
3393 * that has been revoked, the server should return BAD_STATEID
3394 * instead of the more common EXPIRED error.
3395 */
3396 if (id->bits.boottime == nsrv4->rfs4_start_time) {
3397 if (type == DELEGID)
3398 return (NFS4ERR_BAD_STATEID);
3399 else
3400 return (NFS4ERR_EXPIRED);
3401 }
3402
3403 return (NFS4ERR_BAD_STATEID);
3404 }
3405
3406 /*
3407 * Used later on to find the various state structs. When called from
3408 * rfs4_check_stateid()->rfs4_get_all_state(), no file struct lock is
3409 * taken (it is not needed) and helps on the read/write path with
3410 * respect to performance.
3411 */
3412 static nfsstat4
3413 rfs4_get_state_lockit(stateid4 *stateid, rfs4_state_t **spp,
3414 rfs4_dbsearch_type_t find_invalid, bool_t lock_fp)
3415 {
3416 stateid_t *id = (stateid_t *)stateid;
3417 rfs4_state_t *sp;
3418
3419 *spp = NULL;
3420
3421 /* If we are booted as a cluster node, was stateid locally generated? */
3422 if ((cluster_bootflags & CLUSTER_BOOTED) && foreign_stateid(id))
3423 return (NFS4ERR_STALE_STATEID);
3424
3425 sp = rfs4_findstate(id, find_invalid, lock_fp);
3426 if (sp == NULL) {
3427 return (what_stateid_error(id, OPENID));
3428 }
3429
3430 if (rfs4_lease_expired(sp->rs_owner->ro_client)) {
3431 if (lock_fp == TRUE)
3432 rfs4_state_rele(sp);
3433 else
3434 rfs4_state_rele_nounlock(sp);
3435 return (NFS4ERR_EXPIRED);
3436 }
3437
3438 *spp = sp;
3439
3440 return (NFS4_OK);
3441 }
3442
3443 nfsstat4
3444 rfs4_get_state(stateid4 *stateid, rfs4_state_t **spp,
3445 rfs4_dbsearch_type_t find_invalid)
3446 {
3447 return (rfs4_get_state_lockit(stateid, spp, find_invalid, TRUE));
3448 }
3449
3450 int
3451 rfs4_check_stateid_seqid(rfs4_state_t *sp, stateid4 *stateid)
3452 {
3453 stateid_t *id = (stateid_t *)stateid;
3454
3455 if (rfs4_lease_expired(sp->rs_owner->ro_client))
3456 return (NFS4_CHECK_STATEID_EXPIRED);
3457
3458 /* Stateid is some time in the future - that's bad */
3459 if (sp->rs_stateid.bits.chgseq < id->bits.chgseq)
3460 return (NFS4_CHECK_STATEID_BAD);
3461
3462 if (sp->rs_stateid.bits.chgseq == id->bits.chgseq + 1)
3463 return (NFS4_CHECK_STATEID_REPLAY);
3464
3465 /* Stateid is some time in the past - that's old */
3466 if (sp->rs_stateid.bits.chgseq > id->bits.chgseq)
3467 return (NFS4_CHECK_STATEID_OLD);
3468
3469 /* Caller needs to know about confirmation before closure */
3470 if (sp->rs_owner->ro_need_confirm)
3471 return (NFS4_CHECK_STATEID_UNCONFIRMED);
3472
3473 if (sp->rs_closed == TRUE)
3474 return (NFS4_CHECK_STATEID_CLOSED);
3475
3476 return (NFS4_CHECK_STATEID_OKAY);
3477 }
3478
3479 int
3480 rfs4_check_lo_stateid_seqid(rfs4_lo_state_t *lsp, stateid4 *stateid)
3481 {
3482 stateid_t *id = (stateid_t *)stateid;
3483
3484 if (rfs4_lease_expired(lsp->rls_state->rs_owner->ro_client))
3485 return (NFS4_CHECK_STATEID_EXPIRED);
3486
3487 /* Stateid is some time in the future - that's bad */
3488 if (lsp->rls_lockid.bits.chgseq < id->bits.chgseq)
3489 return (NFS4_CHECK_STATEID_BAD);
3490
3491 if (lsp->rls_lockid.bits.chgseq == id->bits.chgseq + 1)
3492 return (NFS4_CHECK_STATEID_REPLAY);
3493
3494 /* Stateid is some time in the past - that's old */
3495 if (lsp->rls_lockid.bits.chgseq > id->bits.chgseq)
3496 return (NFS4_CHECK_STATEID_OLD);
3497
3498 if (lsp->rls_state->rs_closed == TRUE)
3499 return (NFS4_CHECK_STATEID_CLOSED);
3500
3501 return (NFS4_CHECK_STATEID_OKAY);
3502 }
3503
3504 nfsstat4
3505 rfs4_get_deleg_state(stateid4 *stateid, rfs4_deleg_state_t **dspp)
3506 {
3507 stateid_t *id = (stateid_t *)stateid;
3508 rfs4_deleg_state_t *dsp;
3509
3510 *dspp = NULL;
3511
3512 /* If we are booted as a cluster node, was stateid locally generated? */
3513 if ((cluster_bootflags & CLUSTER_BOOTED) && foreign_stateid(id))
3514 return (NFS4ERR_STALE_STATEID);
3515
3516 dsp = rfs4_finddelegstate(id);
3517 if (dsp == NULL) {
3518 return (what_stateid_error(id, DELEGID));
3519 }
3520
3521 if (rfs4_lease_expired(dsp->rds_client)) {
3522 rfs4_deleg_state_rele(dsp);
3523 return (NFS4ERR_EXPIRED);
3524 }
3525
3526 *dspp = dsp;
3527
3528 return (NFS4_OK);
3529 }
3530
3531 nfsstat4
3532 rfs4_get_lo_state(stateid4 *stateid, rfs4_lo_state_t **lspp, bool_t lock_fp)
3533 {
3534 stateid_t *id = (stateid_t *)stateid;
3535 rfs4_lo_state_t *lsp;
3536
3537 *lspp = NULL;
3538
3539 /* If we are booted as a cluster node, was stateid locally generated? */
3540 if ((cluster_bootflags & CLUSTER_BOOTED) && foreign_stateid(id))
3541 return (NFS4ERR_STALE_STATEID);
3542
3543 lsp = rfs4_findlo_state(id, lock_fp);
3544 if (lsp == NULL) {
3545 return (what_stateid_error(id, LOCKID));
3546 }
3547
3548 if (rfs4_lease_expired(lsp->rls_state->rs_owner->ro_client)) {
3549 rfs4_lo_state_rele(lsp, lock_fp);
3550 return (NFS4ERR_EXPIRED);
3551 }
3552
3553 *lspp = lsp;
3554
3555 return (NFS4_OK);
3556 }
3557
3558 static nfsstat4
3559 rfs4_get_all_state(stateid4 *sid, rfs4_state_t **spp,
3560 rfs4_deleg_state_t **dspp, rfs4_lo_state_t **lspp)
3561 {
3562 rfs4_state_t *sp = NULL;
3563 rfs4_deleg_state_t *dsp = NULL;
3564 rfs4_lo_state_t *lsp = NULL;
3565 stateid_t *id;
3566 nfsstat4 status;
3567
3568 *spp = NULL; *dspp = NULL; *lspp = NULL;
3569
3570 id = (stateid_t *)sid;
3571 switch (id->bits.type) {
3572 case OPENID:
3573 status = rfs4_get_state_lockit(sid, &sp, FALSE, FALSE);
3574 break;
3575 case DELEGID:
3576 status = rfs4_get_deleg_state(sid, &dsp);
3577 break;
3578 case LOCKID:
3579 status = rfs4_get_lo_state(sid, &lsp, FALSE);
3580 if (status == NFS4_OK) {
3581 sp = lsp->rls_state;
3582 rfs4_dbe_hold(sp->rs_dbe);
3583 }
3584 break;
3585 default:
3586 status = NFS4ERR_BAD_STATEID;
3587 }
3588
3589 if (status == NFS4_OK) {
3590 *spp = sp;
3591 *dspp = dsp;
3592 *lspp = lsp;
3593 }
3594
3595 return (status);
3596 }
3597
3598 /*
3599 * Given the I/O mode (FREAD or FWRITE), this checks whether the
3600 * rfs4_state_t struct has access to do this operation and if so
3601 * return NFS4_OK; otherwise the proper NFSv4 error is returned.
3602 */
3603 nfsstat4
3604 rfs4_state_has_access(rfs4_state_t *sp, int mode, vnode_t *vp)
3605 {
3606 nfsstat4 stat = NFS4_OK;
3607 rfs4_file_t *fp;
3608 bool_t create = FALSE;
3609
3610 rfs4_dbe_lock(sp->rs_dbe);
3611 if (mode == FWRITE) {
3612 if (!(sp->rs_share_access & OPEN4_SHARE_ACCESS_WRITE)) {
3613 stat = NFS4ERR_OPENMODE;
3614 }
3615 } else if (mode == FREAD) {
3616 if (!(sp->rs_share_access & OPEN4_SHARE_ACCESS_READ)) {
3617 /*
3618 * If we have OPENed the file with DENYing access
3619 * to both READ and WRITE then no one else could
3620 * have OPENed the file, hence no conflicting READ
3621 * deny. This check is merely an optimization.
3622 */
3623 if (sp->rs_share_deny == OPEN4_SHARE_DENY_BOTH)
3624 goto out;
3625
3626 /* Check against file struct's DENY mode */
3627 fp = rfs4_findfile(vp, NULL, &create);
3628 if (fp != NULL) {
3629 int deny_read = 0;
3630 rfs4_dbe_lock(fp->rf_dbe);
3631 /*
3632 * Check if any other open owner has the file
3633 * OPENed with deny READ.
3634 */
3635 if (sp->rs_share_deny & OPEN4_SHARE_DENY_READ)
3636 deny_read = 1;
3637 ASSERT(fp->rf_deny_read >= deny_read);
3638 if (fp->rf_deny_read > deny_read)
3639 stat = NFS4ERR_OPENMODE;
3640 rfs4_dbe_unlock(fp->rf_dbe);
3641 rfs4_file_rele(fp);
3642 }
3643 }
3644 } else {
3645 /* Illegal I/O mode */
3646 stat = NFS4ERR_INVAL;
3647 }
3648 out:
3649 rfs4_dbe_unlock(sp->rs_dbe);
3650 return (stat);
3651 }
3652
3653 /*
3654 * Given the I/O mode (FREAD or FWRITE), the vnode, the stateid and whether
3655 * the file is being truncated, return NFS4_OK if allowed or appropriate
3656 * V4 error if not. Note NFS4ERR_DELAY will be returned and a recall on
3657 * the associated file will be done if the I/O is not consistent with any
3658 * delegation in effect on the file. Should be holding VOP_RWLOCK, either
3659 * as reader or writer as appropriate. rfs4_op_open will acquire the
3660 * VOP_RWLOCK as writer when setting up delegation. If the stateid is bad
3661 * this routine will return NFS4ERR_BAD_STATEID. In addition, through the
3662 * deleg parameter, we will return whether a write delegation is held by
3663 * the client associated with this stateid.
3664 * If the server instance associated with the relevant client is in its
3665 * grace period, return NFS4ERR_GRACE.
3666 */
3667
3668 nfsstat4
3669 rfs4_check_stateid(int mode, vnode_t *vp,
3670 stateid4 *stateid, bool_t trunc, bool_t *deleg,
3671 bool_t do_access, caller_context_t *ct)
3672 {
3673 rfs4_file_t *fp;
3674 bool_t create = FALSE;
3675 rfs4_state_t *sp;
3676 rfs4_deleg_state_t *dsp;
3677 rfs4_lo_state_t *lsp;
3678 stateid_t *id = (stateid_t *)stateid;
3679 nfsstat4 stat = NFS4_OK;
3680
3681 if (ct != NULL) {
3682 ct->cc_sysid = 0;
3683 ct->cc_pid = 0;
3684 ct->cc_caller_id = nfs4_srv_caller_id;
3685 ct->cc_flags = CC_DONTBLOCK;
3686 }
3687
3688 if (ISSPECIAL(stateid)) {
3689 fp = rfs4_findfile(vp, NULL, &create);
3690 if (fp == NULL)
3691 return (NFS4_OK);
3692 if (fp->rf_dinfo.rd_dtype == OPEN_DELEGATE_NONE) {
3693 rfs4_file_rele(fp);
3694 return (NFS4_OK);
3695 }
3696 if (mode == FWRITE ||
3697 fp->rf_dinfo.rd_dtype == OPEN_DELEGATE_WRITE) {
3698 rfs4_recall_deleg(fp, trunc, NULL);
3699 rfs4_file_rele(fp);
3700 return (NFS4ERR_DELAY);
3701 }
3702 rfs4_file_rele(fp);
3703 return (NFS4_OK);
3704 } else {
3705 stat = rfs4_get_all_state(stateid, &sp, &dsp, &lsp);
3706 if (stat != NFS4_OK)
3707 return (stat);
3708 if (lsp != NULL) {
3709 /* Is associated server instance in its grace period? */
3710 if (rfs4_clnt_in_grace(lsp->rls_locker->rl_client)) {
3711 rfs4_lo_state_rele(lsp, FALSE);
3712 if (sp != NULL)
3713 rfs4_state_rele_nounlock(sp);
3714 return (NFS4ERR_GRACE);
3715 }
3716 if (id->bits.type == LOCKID) {
3717 /* Seqid in the future? - that's bad */
3718 if (lsp->rls_lockid.bits.chgseq <
3719 id->bits.chgseq) {
3720 rfs4_lo_state_rele(lsp, FALSE);
3721 if (sp != NULL)
3722 rfs4_state_rele_nounlock(sp);
3723 return (NFS4ERR_BAD_STATEID);
3724 }
3725 /* Seqid in the past? - that's old */
3726 if (lsp->rls_lockid.bits.chgseq >
3727 id->bits.chgseq) {
3728 rfs4_lo_state_rele(lsp, FALSE);
3729 if (sp != NULL)
3730 rfs4_state_rele_nounlock(sp);
3731 return (NFS4ERR_OLD_STATEID);
3732 }
3733 /* Ensure specified filehandle matches */
3734 if (lsp->rls_state->rs_finfo->rf_vp != vp) {
3735 rfs4_lo_state_rele(lsp, FALSE);
3736 if (sp != NULL)
3737 rfs4_state_rele_nounlock(sp);
3738 return (NFS4ERR_BAD_STATEID);
3739 }
3740 }
3741 if (ct != NULL) {
3742 ct->cc_sysid =
3743 lsp->rls_locker->rl_client->rc_sysidt;
3744 ct->cc_pid = lsp->rls_locker->rl_pid;
3745 }
3746 rfs4_lo_state_rele(lsp, FALSE);
3747 }
3748
3749 /* Stateid provided was an "open" stateid */
3750 if (sp != NULL) {
3751 /* Is associated server instance in its grace period? */
3752 if (rfs4_clnt_in_grace(sp->rs_owner->ro_client)) {
3753 rfs4_state_rele_nounlock(sp);
3754 return (NFS4ERR_GRACE);
3755 }
3756 if (id->bits.type == OPENID) {
3757 /* Seqid in the future? - that's bad */
3758 if (sp->rs_stateid.bits.chgseq <
3759 id->bits.chgseq) {
3760 rfs4_state_rele_nounlock(sp);
3761 return (NFS4ERR_BAD_STATEID);
3762 }
3763 /* Seqid in the past - that's old */
3764 if (sp->rs_stateid.bits.chgseq >
3765 id->bits.chgseq) {
3766 rfs4_state_rele_nounlock(sp);
3767 return (NFS4ERR_OLD_STATEID);
3768 }
3769 }
3770 /* Ensure specified filehandle matches */
3771 if (sp->rs_finfo->rf_vp != vp) {
3772 rfs4_state_rele_nounlock(sp);
3773 return (NFS4ERR_BAD_STATEID);
3774 }
3775
3776 if (sp->rs_owner->ro_need_confirm) {
3777 rfs4_state_rele_nounlock(sp);
3778 return (NFS4ERR_BAD_STATEID);
3779 }
3780
3781 if (sp->rs_closed == TRUE) {
3782 rfs4_state_rele_nounlock(sp);
3783 return (NFS4ERR_OLD_STATEID);
3784 }
3785
3786 if (do_access)
3787 stat = rfs4_state_has_access(sp, mode, vp);
3788 else
3789 stat = NFS4_OK;
3790
3791 /*
3792 * Return whether this state has write
3793 * delegation if desired
3794 */
3795 if (deleg && (sp->rs_finfo->rf_dinfo.rd_dtype ==
3796 OPEN_DELEGATE_WRITE))
3797 *deleg = TRUE;
3798
3799 /*
3800 * We got a valid stateid, so we update the
3801 * lease on the client. Ideally we would like
3802 * to do this after the calling op succeeds,
3803 * but for now this will be good
3804 * enough. Callers of this routine are
3805 * currently insulated from the state stuff.
3806 */
3807 rfs4_update_lease(sp->rs_owner->ro_client);
3808
3809 /*
3810 * If a delegation is present on this file and
3811 * this is a WRITE, then update the lastwrite
3812 * time to indicate that activity is present.
3813 */
3814 if (sp->rs_finfo->rf_dinfo.rd_dtype ==
3815 OPEN_DELEGATE_WRITE &&
3816 mode == FWRITE) {
3817 sp->rs_finfo->rf_dinfo.rd_time_lastwrite =
3818 gethrestime_sec();
3819 }
3820
3821 rfs4_state_rele_nounlock(sp);
3822
3823 return (stat);
3824 }
3825
3826 if (dsp != NULL) {
3827 /* Is associated server instance in its grace period? */
3828 if (rfs4_clnt_in_grace(dsp->rds_client)) {
3829 rfs4_deleg_state_rele(dsp);
3830 return (NFS4ERR_GRACE);
3831 }
3832 if (dsp->rds_delegid.bits.chgseq != id->bits.chgseq) {
3833 rfs4_deleg_state_rele(dsp);
3834 return (NFS4ERR_BAD_STATEID);
3835 }
3836
3837 /* Ensure specified filehandle matches */
3838 if (dsp->rds_finfo->rf_vp != vp) {
3839 rfs4_deleg_state_rele(dsp);
3840 return (NFS4ERR_BAD_STATEID);
3841 }
3842 /*
3843 * Return whether this state has write
3844 * delegation if desired
3845 */
3846 if (deleg && (dsp->rds_finfo->rf_dinfo.rd_dtype ==
3847 OPEN_DELEGATE_WRITE))
3848 *deleg = TRUE;
3849
3850 rfs4_update_lease(dsp->rds_client);
3851
3852 /*
3853 * If a delegation is present on this file and
3854 * this is a WRITE, then update the lastwrite
3855 * time to indicate that activity is present.
3856 */
3857 if (dsp->rds_finfo->rf_dinfo.rd_dtype ==
3858 OPEN_DELEGATE_WRITE && mode == FWRITE) {
3859 dsp->rds_finfo->rf_dinfo.rd_time_lastwrite =
3860 gethrestime_sec();
3861 }
3862
3863 /*
3864 * XXX - what happens if this is a WRITE and the
3865 * delegation type of for READ.
3866 */
3867 rfs4_deleg_state_rele(dsp);
3868
3869 return (stat);
3870 }
3871 /*
3872 * If we got this far, something bad happened
3873 */
3874 return (NFS4ERR_BAD_STATEID);
3875 }
3876 }
3877
3878
3879 /*
3880 * This is a special function in that for the file struct provided the
3881 * server wants to remove/close all current state associated with the
3882 * file. The prime use of this would be with OP_REMOVE to force the
3883 * release of state and particularly of file locks.
3884 *
3885 * There is an assumption that there is no delegations outstanding on
3886 * this file at this point. The caller should have waited for those
3887 * to be returned or revoked.
3888 */
3889 void
3890 rfs4_close_all_state(rfs4_file_t *fp)
3891 {
3892 rfs4_state_t *sp;
3893
3894 rfs4_dbe_lock(fp->rf_dbe);
3895
3896 #ifdef DEBUG
3897 /* only applies when server is handing out delegations */
3898 if (nfs4_get_deleg_policy() != SRV_NEVER_DELEGATE)
3899 ASSERT(fp->rf_dinfo.rd_hold_grant > 0);
3900 #endif
3901
3902 /* No delegations for this file */
3903 ASSERT(list_is_empty(&fp->rf_delegstatelist));
3904
3905 /* Make sure that it can not be found */
3906 rfs4_dbe_invalidate(fp->rf_dbe);
3907
3908 if (fp->rf_vp == NULL) {
3909 rfs4_dbe_unlock(fp->rf_dbe);
3910 return;
3911 }
3912 rfs4_dbe_unlock(fp->rf_dbe);
3913
3914 /*
3915 * Hold as writer to prevent other server threads from
3916 * processing requests related to the file while all state is
3917 * being removed.
3918 */
3919 rw_enter(&fp->rf_file_rwlock, RW_WRITER);
3920
3921 /* Remove ALL state from the file */
3922 while (sp = rfs4_findstate_by_file(fp)) {
3923 rfs4_state_close(sp, FALSE, FALSE, CRED());
3924 rfs4_state_rele_nounlock(sp);
3925 }
3926
3927 /*
3928 * This is only safe since there are no further references to
3929 * the file.
3930 */
3931 rfs4_dbe_lock(fp->rf_dbe);
3932 if (fp->rf_vp) {
3933 vnode_t *vp = fp->rf_vp;
3934
3935 mutex_enter(&vp->v_vsd_lock);
3936 (void) vsd_set(vp, nfs4_srv_vkey, NULL);
3937 mutex_exit(&vp->v_vsd_lock);
3938 VN_RELE(vp);
3939 fp->rf_vp = NULL;
3940 }
3941 rfs4_dbe_unlock(fp->rf_dbe);
3942
3943 /* Finally let other references to proceed */
3944 rw_exit(&fp->rf_file_rwlock);
3945 }
3946
3947 /*
3948 * This function is used as a target for the rfs4_dbe_walk() call
3949 * below. The purpose of this function is to see if the
3950 * lockowner_state refers to a file that resides within the exportinfo
3951 * export. If so, then remove the lock_owner state (file locks and
3952 * share "locks") for this object since the intent is the server is
3953 * unexporting the specified directory. Be sure to invalidate the
3954 * object after the state has been released
3955 */
3956 static void
3957 rfs4_lo_state_walk_callout(rfs4_entry_t u_entry, void *e)
3958 {
3959 rfs4_lo_state_t *lsp = (rfs4_lo_state_t *)u_entry;
3960 struct exportinfo *exi = (struct exportinfo *)e;
3961 nfs_fh4_fmt_t fhfmt4, *exi_fhp, *finfo_fhp;
3962 fhandle_t *efhp;
3963
3964 efhp = (fhandle_t *)&exi->exi_fh;
3965 exi_fhp = (nfs_fh4_fmt_t *)&fhfmt4;
3966
3967 FH_TO_FMT4(efhp, exi_fhp);
3968
3969 finfo_fhp = (nfs_fh4_fmt_t *)lsp->rls_state->rs_finfo->
3970 rf_filehandle.nfs_fh4_val;
3971
3972 if (EQFSID(&finfo_fhp->fh4_fsid, &exi_fhp->fh4_fsid) &&
3973 bcmp(&finfo_fhp->fh4_xdata, &exi_fhp->fh4_xdata,
3974 exi_fhp->fh4_xlen) == 0) {
3975 rfs4_state_close(lsp->rls_state, FALSE, FALSE, CRED());
3976 rfs4_dbe_invalidate(lsp->rls_dbe);
3977 rfs4_dbe_invalidate(lsp->rls_state->rs_dbe);
3978 }
3979 }
3980
3981 /*
3982 * This function is used as a target for the rfs4_dbe_walk() call
3983 * below. The purpose of this function is to see if the state refers
3984 * to a file that resides within the exportinfo export. If so, then
3985 * remove the open state for this object since the intent is the
3986 * server is unexporting the specified directory. The main result for
3987 * this type of entry is to invalidate it such it will not be found in
3988 * the future.
3989 */
3990 static void
3991 rfs4_state_walk_callout(rfs4_entry_t u_entry, void *e)
3992 {
3993 rfs4_state_t *sp = (rfs4_state_t *)u_entry;
3994 struct exportinfo *exi = (struct exportinfo *)e;
3995 nfs_fh4_fmt_t fhfmt4, *exi_fhp, *finfo_fhp;
3996 fhandle_t *efhp;
3997
3998 efhp = (fhandle_t *)&exi->exi_fh;
3999 exi_fhp = (nfs_fh4_fmt_t *)&fhfmt4;
4000
4001 FH_TO_FMT4(efhp, exi_fhp);
4002
4003 finfo_fhp =
4004 (nfs_fh4_fmt_t *)sp->rs_finfo->rf_filehandle.nfs_fh4_val;
4005
4006 if (EQFSID(&finfo_fhp->fh4_fsid, &exi_fhp->fh4_fsid) &&
4007 bcmp(&finfo_fhp->fh4_xdata, &exi_fhp->fh4_xdata,
4008 exi_fhp->fh4_xlen) == 0) {
4009 rfs4_state_close(sp, TRUE, FALSE, CRED());
4010 rfs4_dbe_invalidate(sp->rs_dbe);
4011 }
4012 }
4013
4014 /*
4015 * This function is used as a target for the rfs4_dbe_walk() call
4016 * below. The purpose of this function is to see if the state refers
4017 * to a file that resides within the exportinfo export. If so, then
4018 * remove the deleg state for this object since the intent is the
4019 * server is unexporting the specified directory. The main result for
4020 * this type of entry is to invalidate it such it will not be found in
4021 * the future.
4022 */
4023 static void
4024 rfs4_deleg_state_walk_callout(rfs4_entry_t u_entry, void *e)
4025 {
4026 rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
4027 struct exportinfo *exi = (struct exportinfo *)e;
4028 nfs_fh4_fmt_t fhfmt4, *exi_fhp, *finfo_fhp;
4029 fhandle_t *efhp;
4030
4031 efhp = (fhandle_t *)&exi->exi_fh;
4032 exi_fhp = (nfs_fh4_fmt_t *)&fhfmt4;
4033
4034 FH_TO_FMT4(efhp, exi_fhp);
4035
4036 finfo_fhp =
4037 (nfs_fh4_fmt_t *)dsp->rds_finfo->rf_filehandle.nfs_fh4_val;
4038
4039 if (EQFSID(&finfo_fhp->fh4_fsid, &exi_fhp->fh4_fsid) &&
4040 bcmp(&finfo_fhp->fh4_xdata, &exi_fhp->fh4_xdata,
4041 exi_fhp->fh4_xlen) == 0) {
4042 rfs4_dbe_invalidate(dsp->rds_dbe);
4043 }
4044 }
4045
4046 /*
4047 * This function is used as a target for the rfs4_dbe_walk() call
4048 * below. The purpose of this function is to see if the state refers
4049 * to a file that resides within the exportinfo export. If so, then
4050 * release vnode hold for this object since the intent is the server
4051 * is unexporting the specified directory. Invalidation will prevent
4052 * this struct from being found in the future.
4053 */
4054 static void
4055 rfs4_file_walk_callout(rfs4_entry_t u_entry, void *e)
4056 {
4057 rfs4_file_t *fp = (rfs4_file_t *)u_entry;
4058 struct exportinfo *exi = (struct exportinfo *)e;
4059 nfs_fh4_fmt_t fhfmt4, *exi_fhp, *finfo_fhp;
4060 fhandle_t *efhp;
4061
4062 efhp = (fhandle_t *)&exi->exi_fh;
4063 exi_fhp = (nfs_fh4_fmt_t *)&fhfmt4;
4064
4065 FH_TO_FMT4(efhp, exi_fhp);
4066
4067 finfo_fhp = (nfs_fh4_fmt_t *)fp->rf_filehandle.nfs_fh4_val;
4068
4069 if (EQFSID(&finfo_fhp->fh4_fsid, &exi_fhp->fh4_fsid) &&
4070 bcmp(&finfo_fhp->fh4_xdata, &exi_fhp->fh4_xdata,
4071 exi_fhp->fh4_xlen) == 0) {
4072 if (fp->rf_vp) {
4073 vnode_t *vp = fp->rf_vp;
4074
4075 /*
4076 * don't leak monitors and remove the reference
4077 * put on the vnode when the delegation was granted.
4078 */
4079 if (fp->rf_dinfo.rd_dtype == OPEN_DELEGATE_READ) {
4080 (void) fem_uninstall(vp, deleg_rdops,
4081 (void *)fp);
4082 vn_open_downgrade(vp, FREAD);
4083 } else if (fp->rf_dinfo.rd_dtype ==
4084 OPEN_DELEGATE_WRITE) {
4085 (void) fem_uninstall(vp, deleg_wrops,
4086 (void *)fp);
4087 vn_open_downgrade(vp, FREAD|FWRITE);
4088 }
4089 mutex_enter(&vp->v_vsd_lock);
4090 (void) vsd_set(vp, nfs4_srv_vkey, NULL);
4091 mutex_exit(&vp->v_vsd_lock);
4092 VN_RELE(vp);
4093 fp->rf_vp = NULL;
4094 }
4095 rfs4_dbe_invalidate(fp->rf_dbe);
4096 }
4097 }
4098
4099 /*
4100 * Given a directory that is being unexported, cleanup/release all
4101 * state in the server that refers to objects residing underneath this
4102 * particular export. The ordering of the release is important.
4103 * Lock_owner, then state and then file.
4104 */
4105 void
4106 rfs4_clean_state_exi(struct exportinfo *exi)
4107 {
4108 nfs4_srv_t *nsrv4;
4109
4110 /* curzone mightn't be exi_zone, so use exi_zone instead. */
4111 ASSERT(exi->exi_zone == curzone || curzone == global_zone);
4112 nsrv4 = zone_getspecific(rfs4_zone_key, exi->exi_zone);
4113 if (nsrv4 == NULL) /* NOTE: NFSv4 cleanup MAY have already happened. */
4114 return;
4115 mutex_enter(&nsrv4->state_lock);
4116
4117 if (nsrv4->nfs4_server_state == NULL) {
4118 mutex_exit(&nsrv4->state_lock);
4119 return;
4120 }
4121
4122 /* CSTYLED */
4123 rfs4_dbe_walk(nsrv4->rfs4_lo_state_tab, rfs4_lo_state_walk_callout, exi);
4124 rfs4_dbe_walk(nsrv4->rfs4_state_tab, rfs4_state_walk_callout, exi);
4125 /* CSTYLED */
4126 rfs4_dbe_walk(nsrv4->rfs4_deleg_state_tab, rfs4_deleg_state_walk_callout, exi);
4127 rfs4_dbe_walk(nsrv4->rfs4_file_tab, rfs4_file_walk_callout, exi);
4128
4129 mutex_exit(&nsrv4->state_lock);
4130 }