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 /*
1500 * Clean up any dangling stable storage structures BEFORE calling
1501 * rfs4_servinst_destroy_all() so there are no dangling structures
1502 * (i.e. the srvinsts are all cleared of danglers BEFORE they get
1503 * freed).
1504 */
1505 rfs4_ss_fini(nsrv4);
1506
1507 mutex_enter(&nsrv4->state_lock);
1508
1509 if (nsrv4->nfs4_server_state == NULL) {
1510 mutex_exit(&nsrv4->state_lock);
1511 return;
1512 }
1513
1514 /* destroy server instances and current instance ptr */
1515 rfs4_servinst_destroy_all(nsrv4);
1516
1517 /* reset the "first NFSv4 request" status */
1518 nsrv4->seen_first_compound = 0;
1519
1520 dbp = nsrv4->nfs4_server_state;
1521 nsrv4->nfs4_server_state = NULL;
1522
1523 rw_destroy(&nsrv4->rfs4_findclient_lock);
1524
1525 /* First stop all of the reaper threads in the database */
1526 rfs4_database_shutdown(dbp);
1527 /*
1528 * WARNING: There may be consumers of the rfs4 database still
1529 * active as we destroy these. IF that's the case, consider putting
1530 * some of their _zone_fini()-like functions into the zsd key as
1531 * ~~SHUTDOWN~~ functions instead of ~~DESTROY~~ functions. We can
1532 * maintain some ordering guarantees better that way.
1533 */
1534 /* Now destroy/release the database tables */
1535 rfs4_database_destroy(dbp);
1536
1537 /* Reset the cache timers for next time */
1538 nsrv4->rfs4_client_cache_time = 0;
1539 nsrv4->rfs4_openowner_cache_time = 0;
1540 nsrv4->rfs4_state_cache_time = 0;
1541 nsrv4->rfs4_lo_state_cache_time = 0;
1542 nsrv4->rfs4_lockowner_cache_time = 0;
1543 nsrv4->rfs4_file_cache_time = 0;
1544 nsrv4->rfs4_deleg_state_cache_time = 0;
1545
1546 mutex_exit(&nsrv4->state_lock);
1547 }
1548
1549 typedef union {
1550 struct {
1551 uint32_t start_time;
1552 uint32_t c_id;
1553 } impl_id;
1554 clientid4 id4;
1555 } cid;
1556
1557 static int foreign_stateid(stateid_t *id);
1558 static int foreign_clientid(cid *cidp);
1559 static void embed_nodeid(cid *cidp);
1560
1561 typedef union {
1562 struct {
1563 uint32_t c_id;
1564 uint32_t gen_num;
1565 } cv_impl;
1566 verifier4 confirm_verf;
1567 } scid_confirm_verf;
1568
1569 static uint32_t
1570 clientid_hash(void *key)
1571 {
1572 cid *idp = key;
1573
1574 return (idp->impl_id.c_id);
1575 }
1576
1577 static bool_t
1578 clientid_compare(rfs4_entry_t entry, void *key)
1579 {
1580 rfs4_client_t *cp = (rfs4_client_t *)entry;
1581 clientid4 *idp = key;
1582
1583 return (*idp == cp->rc_clientid);
1584 }
1585
1586 static void *
1587 clientid_mkkey(rfs4_entry_t entry)
1588 {
1589 rfs4_client_t *cp = (rfs4_client_t *)entry;
1590
1591 return (&cp->rc_clientid);
1592 }
1593
1594 static uint32_t
1595 nfsclnt_hash(void *key)
1596 {
1597 nfs_client_id4 *client = key;
1598 int i;
1599 uint32_t hash = 0;
1600
1601 for (i = 0; i < client->id_len; i++) {
1602 hash <<= 1;
1603 hash += (uint_t)client->id_val[i];
1604 }
1605 return (hash);
1606 }
1607
1608
1609 static bool_t
1610 nfsclnt_compare(rfs4_entry_t entry, void *key)
1611 {
1612 rfs4_client_t *cp = (rfs4_client_t *)entry;
1613 nfs_client_id4 *nfs_client = key;
1614
1615 if (cp->rc_nfs_client.id_len != nfs_client->id_len)
1616 return (FALSE);
1617
1618 return (bcmp(cp->rc_nfs_client.id_val, nfs_client->id_val,
1619 nfs_client->id_len) == 0);
1620 }
1621
1622 static void *
1623 nfsclnt_mkkey(rfs4_entry_t entry)
1624 {
1625 rfs4_client_t *cp = (rfs4_client_t *)entry;
1626
1627 return (&cp->rc_nfs_client);
1628 }
1629
1630 static bool_t
1631 rfs4_client_expiry(rfs4_entry_t u_entry)
1632 {
1633 rfs4_client_t *cp = (rfs4_client_t *)u_entry;
1634 bool_t cp_expired;
1635
1636 if (rfs4_dbe_is_invalid(cp->rc_dbe)) {
1637 cp->rc_ss_remove = 1;
1638 return (TRUE);
1639 }
1640 /*
1641 * If the sysadmin has used clear_locks for this
1642 * entry then forced_expire will be set and we
1643 * want this entry to be reaped. Or the entry
1644 * has exceeded its lease period.
1645 */
1646 cp_expired = (cp->rc_forced_expire ||
1647 (gethrestime_sec() - cp->rc_last_access
1648 > rfs4_lease_time));
1649
1650 if (!cp->rc_ss_remove && cp_expired)
1651 cp->rc_ss_remove = 1;
1652 return (cp_expired);
1653 }
1654
1655 /*
1656 * Remove the leaf file from all distributed stable storage paths.
1657 */
1658 static void
1659 rfs4_dss_remove_cpleaf(rfs4_client_t *cp)
1660 {
1661 nfs4_srv_t *nsrv4;
1662 rfs4_servinst_t *sip;
1663 char *leaf = cp->rc_ss_pn->leaf;
1664
1665 /*
1666 * since the state files are written to all DSS
1667 * paths we must remove this leaf file instance
1668 * from all server instances.
1669 */
1670
1671 nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
1672 mutex_enter(&nsrv4->servinst_lock);
1673 for (sip = nsrv4->nfs4_cur_servinst; sip != NULL; sip = sip->prev) {
1674 /* remove the leaf file associated with this server instance */
1675 rfs4_dss_remove_leaf(sip, NFS4_DSS_STATE_LEAF, leaf);
1676 }
1677 mutex_exit(&nsrv4->servinst_lock);
1678 }
1679
1680 static void
1681 rfs4_dss_remove_leaf(rfs4_servinst_t *sip, char *dir_leaf, char *leaf)
1682 {
1683 int i, npaths = sip->dss_npaths;
1684
1685 for (i = 0; i < npaths; i++) {
1686 rfs4_dss_path_t *dss_path = sip->dss_paths[i];
1687 char *path, *dir;
1688 size_t pathlen;
1689
1690 /* the HA-NFSv4 path might have been failed-over away from us */
1691 if (dss_path == NULL)
1692 continue;
1693
1694 dir = dss_path->path;
1695
1696 /* allow 3 extra bytes for two '/' & a NUL */
1697 pathlen = strlen(dir) + strlen(dir_leaf) + strlen(leaf) + 3;
1698 path = kmem_alloc(pathlen, KM_SLEEP);
1699 (void) sprintf(path, "%s/%s/%s", dir, dir_leaf, leaf);
1700
1701 (void) vn_remove(path, UIO_SYSSPACE, RMFILE);
1702
1703 kmem_free(path, pathlen);
1704 }
1705 }
1706
1707 static void
1708 rfs4_client_destroy(rfs4_entry_t u_entry)
1709 {
1710 rfs4_client_t *cp = (rfs4_client_t *)u_entry;
1711
1712 mutex_destroy(cp->rc_cbinfo.cb_lock);
1713 cv_destroy(cp->rc_cbinfo.cb_cv);
1714 cv_destroy(cp->rc_cbinfo.cb_cv_nullcaller);
1715 list_destroy(&cp->rc_openownerlist);
1716
1717 /* free callback info */
1718 rfs4_cbinfo_free(&cp->rc_cbinfo);
1719
1720 if (cp->rc_cp_confirmed)
1721 rfs4_client_rele(cp->rc_cp_confirmed);
1722
1723 if (cp->rc_ss_pn) {
1724 /* check if the stable storage files need to be removed */
1725 if (cp->rc_ss_remove)
1726 rfs4_dss_remove_cpleaf(cp);
1727 rfs4_ss_pnfree(cp->rc_ss_pn);
1728 }
1729
1730 /* Free the client supplied client id */
1731 kmem_free(cp->rc_nfs_client.id_val, cp->rc_nfs_client.id_len);
1732
1733 if (cp->rc_sysidt != LM_NOSYSID)
1734 lm_free_sysidt(cp->rc_sysidt);
1735 }
1736
1737 static bool_t
1738 rfs4_client_create(rfs4_entry_t u_entry, void *arg)
1739 {
1740 rfs4_client_t *cp = (rfs4_client_t *)u_entry;
1741 nfs_client_id4 *client = (nfs_client_id4 *)arg;
1742 struct sockaddr *ca;
1743 cid *cidp;
1744 scid_confirm_verf *scvp;
1745 nfs4_srv_t *nsrv4;
1746
1747 nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
1748
1749 /* Get a clientid to give to the client */
1750 cidp = (cid *)&cp->rc_clientid;
1751 cidp->impl_id.start_time = nsrv4->rfs4_start_time;
1752 cidp->impl_id.c_id = (uint32_t)rfs4_dbe_getid(cp->rc_dbe);
1753
1754 /* If we are booted as a cluster node, embed our nodeid */
1755 if (cluster_bootflags & CLUSTER_BOOTED)
1756 embed_nodeid(cidp);
1757
1758 /* Allocate and copy client's client id value */
1759 cp->rc_nfs_client.id_val = kmem_alloc(client->id_len, KM_SLEEP);
1760 cp->rc_nfs_client.id_len = client->id_len;
1761 bcopy(client->id_val, cp->rc_nfs_client.id_val, client->id_len);
1762 cp->rc_nfs_client.verifier = client->verifier;
1763
1764 /* Copy client's IP address */
1765 ca = client->cl_addr;
1766 if (ca->sa_family == AF_INET)
1767 bcopy(ca, &cp->rc_addr, sizeof (struct sockaddr_in));
1768 else if (ca->sa_family == AF_INET6)
1769 bcopy(ca, &cp->rc_addr, sizeof (struct sockaddr_in6));
1770 cp->rc_nfs_client.cl_addr = (struct sockaddr *)&cp->rc_addr;
1771
1772 /* Init the value for the SETCLIENTID_CONFIRM verifier */
1773 scvp = (scid_confirm_verf *)&cp->rc_confirm_verf;
1774 scvp->cv_impl.c_id = cidp->impl_id.c_id;
1775 scvp->cv_impl.gen_num = 0;
1776
1777 /* An F_UNLKSYS has been done for this client */
1778 cp->rc_unlksys_completed = FALSE;
1779
1780 /* We need the client to ack us */
1781 cp->rc_need_confirm = TRUE;
1782 cp->rc_cp_confirmed = NULL;
1783
1784 /* TRUE all the time until the callback path actually fails */
1785 cp->rc_cbinfo.cb_notified_of_cb_path_down = TRUE;
1786
1787 /* Initialize the access time to now */
1788 cp->rc_last_access = gethrestime_sec();
1789
1790 cp->rc_cr_set = NULL;
1791
1792 cp->rc_sysidt = LM_NOSYSID;
1793
1794 list_create(&cp->rc_openownerlist, sizeof (rfs4_openowner_t),
1795 offsetof(rfs4_openowner_t, ro_node));
1796
1797 /* set up the callback control structure */
1798 cp->rc_cbinfo.cb_state = CB_UNINIT;
1799 mutex_init(cp->rc_cbinfo.cb_lock, NULL, MUTEX_DEFAULT, NULL);
1800 cv_init(cp->rc_cbinfo.cb_cv, NULL, CV_DEFAULT, NULL);
1801 cv_init(cp->rc_cbinfo.cb_cv_nullcaller, NULL, CV_DEFAULT, NULL);
1802
1803 /*
1804 * Associate the client_t with the current server instance.
1805 * The hold is solely to satisfy the calling requirement of
1806 * rfs4_servinst_assign(). In this case it's not strictly necessary.
1807 */
1808 rfs4_dbe_hold(cp->rc_dbe);
1809 rfs4_servinst_assign(nsrv4, cp, nsrv4->nfs4_cur_servinst);
1810 rfs4_dbe_rele(cp->rc_dbe);
1811
1812 return (TRUE);
1813 }
1814
1815 /*
1816 * Caller wants to generate/update the setclientid_confirm verifier
1817 * associated with a client. This is done during the SETCLIENTID
1818 * processing.
1819 */
1820 void
1821 rfs4_client_scv_next(rfs4_client_t *cp)
1822 {
1823 scid_confirm_verf *scvp;
1824
1825 /* Init the value for the SETCLIENTID_CONFIRM verifier */
1826 scvp = (scid_confirm_verf *)&cp->rc_confirm_verf;
1827 scvp->cv_impl.gen_num++;
1828 }
1829
1830 void
1831 rfs4_client_rele(rfs4_client_t *cp)
1832 {
1833 rfs4_dbe_rele(cp->rc_dbe);
1834 }
1835
1836 rfs4_client_t *
1837 rfs4_findclient(nfs_client_id4 *client, bool_t *create, rfs4_client_t *oldcp)
1838 {
1839 rfs4_client_t *cp;
1840 nfs4_srv_t *nsrv4;
1841 nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
1842
1843
1844 if (oldcp) {
1845 rw_enter(&nsrv4->rfs4_findclient_lock, RW_WRITER);
1846 rfs4_dbe_hide(oldcp->rc_dbe);
1847 } else {
1848 rw_enter(&nsrv4->rfs4_findclient_lock, RW_READER);
1849 }
1850
1851 cp = (rfs4_client_t *)rfs4_dbsearch(nsrv4->rfs4_nfsclnt_idx, client,
1852 create, (void *)client, RFS4_DBS_VALID);
1853
1854 if (oldcp)
1855 rfs4_dbe_unhide(oldcp->rc_dbe);
1856
1857 rw_exit(&nsrv4->rfs4_findclient_lock);
1858
1859 return (cp);
1860 }
1861
1862 rfs4_client_t *
1863 rfs4_findclient_by_id(clientid4 clientid, bool_t find_unconfirmed)
1864 {
1865 rfs4_client_t *cp;
1866 bool_t create = FALSE;
1867 cid *cidp = (cid *)&clientid;
1868 nfs4_srv_t *nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
1869
1870 /* If we're a cluster and the nodeid isn't right, short-circuit */
1871 if (cluster_bootflags & CLUSTER_BOOTED && foreign_clientid(cidp))
1872 return (NULL);
1873
1874 rw_enter(&nsrv4->rfs4_findclient_lock, RW_READER);
1875
1876 cp = (rfs4_client_t *)rfs4_dbsearch(nsrv4->rfs4_clientid_idx, &clientid,
1877 &create, NULL, RFS4_DBS_VALID);
1878
1879 rw_exit(&nsrv4->rfs4_findclient_lock);
1880
1881 if (cp && cp->rc_need_confirm && find_unconfirmed == FALSE) {
1882 rfs4_client_rele(cp);
1883 return (NULL);
1884 } else {
1885 return (cp);
1886 }
1887 }
1888
1889 static uint32_t
1890 clntip_hash(void *key)
1891 {
1892 struct sockaddr *addr = key;
1893 int i, len = 0;
1894 uint32_t hash = 0;
1895 char *ptr;
1896
1897 if (addr->sa_family == AF_INET) {
1898 struct sockaddr_in *a = (struct sockaddr_in *)addr;
1899 len = sizeof (struct in_addr);
1900 ptr = (char *)&a->sin_addr;
1901 } else if (addr->sa_family == AF_INET6) {
1902 struct sockaddr_in6 *a = (struct sockaddr_in6 *)addr;
1903 len = sizeof (struct in6_addr);
1904 ptr = (char *)&a->sin6_addr;
1905 } else
1906 return (0);
1907
1908 for (i = 0; i < len; i++) {
1909 hash <<= 1;
1910 hash += (uint_t)ptr[i];
1911 }
1912 return (hash);
1913 }
1914
1915 static bool_t
1916 clntip_compare(rfs4_entry_t entry, void *key)
1917 {
1918 rfs4_clntip_t *cp = (rfs4_clntip_t *)entry;
1919 struct sockaddr *addr = key;
1920 int len = 0;
1921 char *p1, *p2;
1922
1923 if (addr->sa_family == AF_INET) {
1924 struct sockaddr_in *a1 = (struct sockaddr_in *)&cp->ri_addr;
1925 struct sockaddr_in *a2 = (struct sockaddr_in *)addr;
1926 len = sizeof (struct in_addr);
1927 p1 = (char *)&a1->sin_addr;
1928 p2 = (char *)&a2->sin_addr;
1929 } else if (addr->sa_family == AF_INET6) {
1930 struct sockaddr_in6 *a1 = (struct sockaddr_in6 *)&cp->ri_addr;
1931 struct sockaddr_in6 *a2 = (struct sockaddr_in6 *)addr;
1932 len = sizeof (struct in6_addr);
1933 p1 = (char *)&a1->sin6_addr;
1934 p2 = (char *)&a2->sin6_addr;
1935 } else
1936 return (0);
1937
1938 return (bcmp(p1, p2, len) == 0);
1939 }
1940
1941 static void *
1942 clntip_mkkey(rfs4_entry_t entry)
1943 {
1944 rfs4_clntip_t *cp = (rfs4_clntip_t *)entry;
1945
1946 return (&cp->ri_addr);
1947 }
1948
1949 static bool_t
1950 rfs4_clntip_expiry(rfs4_entry_t u_entry)
1951 {
1952 rfs4_clntip_t *cp = (rfs4_clntip_t *)u_entry;
1953
1954 if (rfs4_dbe_is_invalid(cp->ri_dbe))
1955 return (TRUE);
1956 return (FALSE);
1957 }
1958
1959 /* ARGSUSED */
1960 static void
1961 rfs4_clntip_destroy(rfs4_entry_t u_entry)
1962 {
1963 }
1964
1965 static bool_t
1966 rfs4_clntip_create(rfs4_entry_t u_entry, void *arg)
1967 {
1968 rfs4_clntip_t *cp = (rfs4_clntip_t *)u_entry;
1969 struct sockaddr *ca = (struct sockaddr *)arg;
1970
1971 /* Copy client's IP address */
1972 if (ca->sa_family == AF_INET)
1973 bcopy(ca, &cp->ri_addr, sizeof (struct sockaddr_in));
1974 else if (ca->sa_family == AF_INET6)
1975 bcopy(ca, &cp->ri_addr, sizeof (struct sockaddr_in6));
1976 else
1977 return (FALSE);
1978 cp->ri_no_referrals = 1;
1979
1980 return (TRUE);
1981 }
1982
1983 rfs4_clntip_t *
1984 rfs4_find_clntip(struct sockaddr *addr, bool_t *create)
1985 {
1986 rfs4_clntip_t *cp;
1987 nfs4_srv_t *nsrv4;
1988
1989 nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
1990
1991 rw_enter(&nsrv4->rfs4_findclient_lock, RW_READER);
1992
1993 cp = (rfs4_clntip_t *)rfs4_dbsearch(nsrv4->rfs4_clntip_idx, addr,
1994 create, addr, RFS4_DBS_VALID);
1995
1996 rw_exit(&nsrv4->rfs4_findclient_lock);
1997
1998 return (cp);
1999 }
2000
2001 void
2002 rfs4_invalidate_clntip(struct sockaddr *addr)
2003 {
2004 rfs4_clntip_t *cp;
2005 bool_t create = FALSE;
2006 nfs4_srv_t *nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
2007
2008 rw_enter(&nsrv4->rfs4_findclient_lock, RW_READER);
2009
2010 cp = (rfs4_clntip_t *)rfs4_dbsearch(nsrv4->rfs4_clntip_idx, addr,
2011 &create, NULL, RFS4_DBS_VALID);
2012 if (cp == NULL) {
2013 rw_exit(&nsrv4->rfs4_findclient_lock);
2014 return;
2015 }
2016 rfs4_dbe_invalidate(cp->ri_dbe);
2017 rfs4_dbe_rele(cp->ri_dbe);
2018
2019 rw_exit(&nsrv4->rfs4_findclient_lock);
2020 }
2021
2022 bool_t
2023 rfs4_lease_expired(rfs4_client_t *cp)
2024 {
2025 bool_t rc;
2026
2027 rfs4_dbe_lock(cp->rc_dbe);
2028
2029 /*
2030 * If the admin has executed clear_locks for this
2031 * client id, force expire will be set, so no need
2032 * to calculate anything because it's "outa here".
2033 */
2034 if (cp->rc_forced_expire) {
2035 rc = TRUE;
2036 } else {
2037 rc = (gethrestime_sec() - cp->rc_last_access > rfs4_lease_time);
2038 }
2039
2040 /*
2041 * If the lease has expired we will also want
2042 * to remove any stable storage state data. So
2043 * mark the client id accordingly.
2044 */
2045 if (!cp->rc_ss_remove)
2046 cp->rc_ss_remove = (rc == TRUE);
2047
2048 rfs4_dbe_unlock(cp->rc_dbe);
2049
2050 return (rc);
2051 }
2052
2053 void
2054 rfs4_update_lease(rfs4_client_t *cp)
2055 {
2056 rfs4_dbe_lock(cp->rc_dbe);
2057 if (!cp->rc_forced_expire)
2058 cp->rc_last_access = gethrestime_sec();
2059 rfs4_dbe_unlock(cp->rc_dbe);
2060 }
2061
2062
2063 static bool_t
2064 EQOPENOWNER(open_owner4 *a, open_owner4 *b)
2065 {
2066 bool_t rc;
2067
2068 if (a->clientid != b->clientid)
2069 return (FALSE);
2070
2071 if (a->owner_len != b->owner_len)
2072 return (FALSE);
2073
2074 rc = (bcmp(a->owner_val, b->owner_val, a->owner_len) == 0);
2075
2076 return (rc);
2077 }
2078
2079 static uint_t
2080 openowner_hash(void *key)
2081 {
2082 int i;
2083 open_owner4 *openowner = key;
2084 uint_t hash = 0;
2085
2086 for (i = 0; i < openowner->owner_len; i++) {
2087 hash <<= 4;
2088 hash += (uint_t)openowner->owner_val[i];
2089 }
2090 hash += (uint_t)openowner->clientid;
2091 hash |= (openowner->clientid >> 32);
2092
2093 return (hash);
2094 }
2095
2096 static bool_t
2097 openowner_compare(rfs4_entry_t u_entry, void *key)
2098 {
2099 rfs4_openowner_t *oo = (rfs4_openowner_t *)u_entry;
2100 open_owner4 *arg = key;
2101
2102 return (EQOPENOWNER(&oo->ro_owner, arg));
2103 }
2104
2105 void *
2106 openowner_mkkey(rfs4_entry_t u_entry)
2107 {
2108 rfs4_openowner_t *oo = (rfs4_openowner_t *)u_entry;
2109
2110 return (&oo->ro_owner);
2111 }
2112
2113 /* ARGSUSED */
2114 static bool_t
2115 rfs4_openowner_expiry(rfs4_entry_t u_entry)
2116 {
2117 /* openstateid held us and did all needed delay */
2118 return (TRUE);
2119 }
2120
2121 static void
2122 rfs4_openowner_destroy(rfs4_entry_t u_entry)
2123 {
2124 rfs4_openowner_t *oo = (rfs4_openowner_t *)u_entry;
2125
2126 /* Remove open owner from client's lists of open owners */
2127 rfs4_dbe_lock(oo->ro_client->rc_dbe);
2128 list_remove(&oo->ro_client->rc_openownerlist, oo);
2129 rfs4_dbe_unlock(oo->ro_client->rc_dbe);
2130
2131 /* One less reference to the client */
2132 rfs4_client_rele(oo->ro_client);
2133 oo->ro_client = NULL;
2134
2135 /* Free the last reply for this lock owner */
2136 rfs4_free_reply(&oo->ro_reply);
2137
2138 if (oo->ro_reply_fh.nfs_fh4_val) {
2139 kmem_free(oo->ro_reply_fh.nfs_fh4_val,
2140 oo->ro_reply_fh.nfs_fh4_len);
2141 oo->ro_reply_fh.nfs_fh4_val = NULL;
2142 oo->ro_reply_fh.nfs_fh4_len = 0;
2143 }
2144
2145 rfs4_sw_destroy(&oo->ro_sw);
2146 list_destroy(&oo->ro_statelist);
2147
2148 /* Free the lock owner id */
2149 kmem_free(oo->ro_owner.owner_val, oo->ro_owner.owner_len);
2150 }
2151
2152 void
2153 rfs4_openowner_rele(rfs4_openowner_t *oo)
2154 {
2155 rfs4_dbe_rele(oo->ro_dbe);
2156 }
2157
2158 static bool_t
2159 rfs4_openowner_create(rfs4_entry_t u_entry, void *arg)
2160 {
2161 rfs4_openowner_t *oo = (rfs4_openowner_t *)u_entry;
2162 rfs4_openowner_t *argp = (rfs4_openowner_t *)arg;
2163 open_owner4 *openowner = &argp->ro_owner;
2164 seqid4 seqid = argp->ro_open_seqid;
2165 rfs4_client_t *cp;
2166 bool_t create = FALSE;
2167 nfs4_srv_t *nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
2168
2169 rw_enter(&nsrv4->rfs4_findclient_lock, RW_READER);
2170
2171 cp = (rfs4_client_t *)rfs4_dbsearch(nsrv4->rfs4_clientid_idx,
2172 &openowner->clientid,
2173 &create, NULL, RFS4_DBS_VALID);
2174
2175 rw_exit(&nsrv4->rfs4_findclient_lock);
2176
2177 if (cp == NULL)
2178 return (FALSE);
2179
2180 oo->ro_reply_fh.nfs_fh4_len = 0;
2181 oo->ro_reply_fh.nfs_fh4_val = NULL;
2182
2183 oo->ro_owner.clientid = openowner->clientid;
2184 oo->ro_owner.owner_val =
2185 kmem_alloc(openowner->owner_len, KM_SLEEP);
2186
2187 bcopy(openowner->owner_val,
2188 oo->ro_owner.owner_val, openowner->owner_len);
2189
2190 oo->ro_owner.owner_len = openowner->owner_len;
2191
2192 oo->ro_need_confirm = TRUE;
2193
2194 rfs4_sw_init(&oo->ro_sw);
2195
2196 oo->ro_open_seqid = seqid;
2197 bzero(&oo->ro_reply, sizeof (nfs_resop4));
2198 oo->ro_client = cp;
2199 oo->ro_cr_set = NULL;
2200
2201 list_create(&oo->ro_statelist, sizeof (rfs4_state_t),
2202 offsetof(rfs4_state_t, rs_node));
2203
2204 /* Insert openowner into client's open owner list */
2205 rfs4_dbe_lock(cp->rc_dbe);
2206 list_insert_tail(&cp->rc_openownerlist, oo);
2207 rfs4_dbe_unlock(cp->rc_dbe);
2208
2209 return (TRUE);
2210 }
2211
2212 rfs4_openowner_t *
2213 rfs4_findopenowner(open_owner4 *openowner, bool_t *create, seqid4 seqid)
2214 {
2215 rfs4_openowner_t *oo;
2216 rfs4_openowner_t arg;
2217 nfs4_srv_t *nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
2218
2219 arg.ro_owner = *openowner;
2220 arg.ro_open_seqid = seqid;
2221 /* CSTYLED */
2222 oo = (rfs4_openowner_t *)rfs4_dbsearch(nsrv4->rfs4_openowner_idx, openowner,
2223 create, &arg, RFS4_DBS_VALID);
2224
2225 return (oo);
2226 }
2227
2228 void
2229 rfs4_update_open_sequence(rfs4_openowner_t *oo)
2230 {
2231
2232 rfs4_dbe_lock(oo->ro_dbe);
2233
2234 oo->ro_open_seqid++;
2235
2236 rfs4_dbe_unlock(oo->ro_dbe);
2237 }
2238
2239 void
2240 rfs4_update_open_resp(rfs4_openowner_t *oo, nfs_resop4 *resp, nfs_fh4 *fh)
2241 {
2242
2243 rfs4_dbe_lock(oo->ro_dbe);
2244
2245 rfs4_free_reply(&oo->ro_reply);
2246
2247 rfs4_copy_reply(&oo->ro_reply, resp);
2248
2249 /* Save the filehandle if provided and free if not used */
2250 if (resp->nfs_resop4_u.opopen.status == NFS4_OK &&
2251 fh && fh->nfs_fh4_len) {
2252 if (oo->ro_reply_fh.nfs_fh4_val == NULL)
2253 oo->ro_reply_fh.nfs_fh4_val =
2254 kmem_alloc(fh->nfs_fh4_len, KM_SLEEP);
2255 nfs_fh4_copy(fh, &oo->ro_reply_fh);
2256 } else {
2257 if (oo->ro_reply_fh.nfs_fh4_val) {
2258 kmem_free(oo->ro_reply_fh.nfs_fh4_val,
2259 oo->ro_reply_fh.nfs_fh4_len);
2260 oo->ro_reply_fh.nfs_fh4_val = NULL;
2261 oo->ro_reply_fh.nfs_fh4_len = 0;
2262 }
2263 }
2264
2265 rfs4_dbe_unlock(oo->ro_dbe);
2266 }
2267
2268 static bool_t
2269 lockowner_compare(rfs4_entry_t u_entry, void *key)
2270 {
2271 rfs4_lockowner_t *lo = (rfs4_lockowner_t *)u_entry;
2272 lock_owner4 *b = (lock_owner4 *)key;
2273
2274 if (lo->rl_owner.clientid != b->clientid)
2275 return (FALSE);
2276
2277 if (lo->rl_owner.owner_len != b->owner_len)
2278 return (FALSE);
2279
2280 return (bcmp(lo->rl_owner.owner_val, b->owner_val,
2281 lo->rl_owner.owner_len) == 0);
2282 }
2283
2284 void *
2285 lockowner_mkkey(rfs4_entry_t u_entry)
2286 {
2287 rfs4_lockowner_t *lo = (rfs4_lockowner_t *)u_entry;
2288
2289 return (&lo->rl_owner);
2290 }
2291
2292 static uint32_t
2293 lockowner_hash(void *key)
2294 {
2295 int i;
2296 lock_owner4 *lockowner = key;
2297 uint_t hash = 0;
2298
2299 for (i = 0; i < lockowner->owner_len; i++) {
2300 hash <<= 4;
2301 hash += (uint_t)lockowner->owner_val[i];
2302 }
2303 hash += (uint_t)lockowner->clientid;
2304 hash |= (lockowner->clientid >> 32);
2305
2306 return (hash);
2307 }
2308
2309 static uint32_t
2310 pid_hash(void *key)
2311 {
2312 return ((uint32_t)(uintptr_t)key);
2313 }
2314
2315 static void *
2316 pid_mkkey(rfs4_entry_t u_entry)
2317 {
2318 rfs4_lockowner_t *lo = (rfs4_lockowner_t *)u_entry;
2319
2320 return ((void *)(uintptr_t)lo->rl_pid);
2321 }
2322
2323 static bool_t
2324 pid_compare(rfs4_entry_t u_entry, void *key)
2325 {
2326 rfs4_lockowner_t *lo = (rfs4_lockowner_t *)u_entry;
2327
2328 return (lo->rl_pid == (pid_t)(uintptr_t)key);
2329 }
2330
2331 static void
2332 rfs4_lockowner_destroy(rfs4_entry_t u_entry)
2333 {
2334 rfs4_lockowner_t *lo = (rfs4_lockowner_t *)u_entry;
2335
2336 /* Free the lock owner id */
2337 kmem_free(lo->rl_owner.owner_val, lo->rl_owner.owner_len);
2338 rfs4_client_rele(lo->rl_client);
2339 }
2340
2341 void
2342 rfs4_lockowner_rele(rfs4_lockowner_t *lo)
2343 {
2344 rfs4_dbe_rele(lo->rl_dbe);
2345 }
2346
2347 /* ARGSUSED */
2348 static bool_t
2349 rfs4_lockowner_expiry(rfs4_entry_t u_entry)
2350 {
2351 /*
2352 * Since expiry is called with no other references on
2353 * this struct, go ahead and have it removed.
2354 */
2355 return (TRUE);
2356 }
2357
2358 static bool_t
2359 rfs4_lockowner_create(rfs4_entry_t u_entry, void *arg)
2360 {
2361 rfs4_lockowner_t *lo = (rfs4_lockowner_t *)u_entry;
2362 lock_owner4 *lockowner = (lock_owner4 *)arg;
2363 rfs4_client_t *cp;
2364 bool_t create = FALSE;
2365 nfs4_srv_t *nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
2366
2367 rw_enter(&nsrv4->rfs4_findclient_lock, RW_READER);
2368
2369 cp = (rfs4_client_t *)rfs4_dbsearch(nsrv4->rfs4_clientid_idx,
2370 &lockowner->clientid,
2371 &create, NULL, RFS4_DBS_VALID);
2372
2373 rw_exit(&nsrv4->rfs4_findclient_lock);
2374
2375 if (cp == NULL)
2376 return (FALSE);
2377
2378 /* Reference client */
2379 lo->rl_client = cp;
2380 lo->rl_owner.clientid = lockowner->clientid;
2381 lo->rl_owner.owner_val = kmem_alloc(lockowner->owner_len, KM_SLEEP);
2382 bcopy(lockowner->owner_val, lo->rl_owner.owner_val,
2383 lockowner->owner_len);
2384 lo->rl_owner.owner_len = lockowner->owner_len;
2385 lo->rl_pid = rfs4_dbe_getid(lo->rl_dbe);
2386
2387 return (TRUE);
2388 }
2389
2390 rfs4_lockowner_t *
2391 rfs4_findlockowner(lock_owner4 *lockowner, bool_t *create)
2392 {
2393 rfs4_lockowner_t *lo;
2394 nfs4_srv_t *nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
2395
2396 /* CSTYLED */
2397 lo = (rfs4_lockowner_t *)rfs4_dbsearch(nsrv4->rfs4_lockowner_idx, lockowner,
2398 create, lockowner, RFS4_DBS_VALID);
2399
2400 return (lo);
2401 }
2402
2403 rfs4_lockowner_t *
2404 rfs4_findlockowner_by_pid(pid_t pid)
2405 {
2406 rfs4_lockowner_t *lo;
2407 bool_t create = FALSE;
2408 nfs4_srv_t *nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
2409
2410 lo = (rfs4_lockowner_t *)rfs4_dbsearch(nsrv4->rfs4_lockowner_pid_idx,
2411 (void *)(uintptr_t)pid, &create, NULL, RFS4_DBS_VALID);
2412
2413 return (lo);
2414 }
2415
2416
2417 static uint32_t
2418 file_hash(void *key)
2419 {
2420 return (ADDRHASH(key));
2421 }
2422
2423 static void *
2424 file_mkkey(rfs4_entry_t u_entry)
2425 {
2426 rfs4_file_t *fp = (rfs4_file_t *)u_entry;
2427
2428 return (fp->rf_vp);
2429 }
2430
2431 static bool_t
2432 file_compare(rfs4_entry_t u_entry, void *key)
2433 {
2434 rfs4_file_t *fp = (rfs4_file_t *)u_entry;
2435
2436 return (fp->rf_vp == (vnode_t *)key);
2437 }
2438
2439 static void
2440 rfs4_file_destroy(rfs4_entry_t u_entry)
2441 {
2442 rfs4_file_t *fp = (rfs4_file_t *)u_entry;
2443
2444 list_destroy(&fp->rf_delegstatelist);
2445
2446 if (fp->rf_filehandle.nfs_fh4_val)
2447 kmem_free(fp->rf_filehandle.nfs_fh4_val,
2448 fp->rf_filehandle.nfs_fh4_len);
2449 cv_destroy(fp->rf_dinfo.rd_recall_cv);
2450 if (fp->rf_vp) {
2451 vnode_t *vp = fp->rf_vp;
2452
2453 mutex_enter(&vp->v_vsd_lock);
2454 (void) vsd_set(vp, nfs4_srv_vkey, NULL);
2455 mutex_exit(&vp->v_vsd_lock);
2456 VN_RELE(vp);
2457 fp->rf_vp = NULL;
2458 }
2459 rw_destroy(&fp->rf_file_rwlock);
2460 }
2461
2462 /*
2463 * Used to unlock the underlying dbe struct only
2464 */
2465 void
2466 rfs4_file_rele(rfs4_file_t *fp)
2467 {
2468 rfs4_dbe_rele(fp->rf_dbe);
2469 }
2470
2471 typedef struct {
2472 vnode_t *vp;
2473 nfs_fh4 *fh;
2474 } rfs4_fcreate_arg;
2475
2476 static bool_t
2477 rfs4_file_create(rfs4_entry_t u_entry, void *arg)
2478 {
2479 rfs4_file_t *fp = (rfs4_file_t *)u_entry;
2480 rfs4_fcreate_arg *ap = (rfs4_fcreate_arg *)arg;
2481 vnode_t *vp = ap->vp;
2482 nfs_fh4 *fh = ap->fh;
2483
2484 VN_HOLD(vp);
2485
2486 fp->rf_filehandle.nfs_fh4_len = 0;
2487 fp->rf_filehandle.nfs_fh4_val = NULL;
2488 ASSERT(fh && fh->nfs_fh4_len);
2489 if (fh && fh->nfs_fh4_len) {
2490 fp->rf_filehandle.nfs_fh4_val =
2491 kmem_alloc(fh->nfs_fh4_len, KM_SLEEP);
2492 nfs_fh4_copy(fh, &fp->rf_filehandle);
2493 }
2494 fp->rf_vp = vp;
2495
2496 list_create(&fp->rf_delegstatelist, sizeof (rfs4_deleg_state_t),
2497 offsetof(rfs4_deleg_state_t, rds_node));
2498
2499 fp->rf_share_deny = fp->rf_share_access = fp->rf_access_read = 0;
2500 fp->rf_access_write = fp->rf_deny_read = fp->rf_deny_write = 0;
2501
2502 mutex_init(fp->rf_dinfo.rd_recall_lock, NULL, MUTEX_DEFAULT, NULL);
2503 cv_init(fp->rf_dinfo.rd_recall_cv, NULL, CV_DEFAULT, NULL);
2504
2505 fp->rf_dinfo.rd_dtype = OPEN_DELEGATE_NONE;
2506
2507 rw_init(&fp->rf_file_rwlock, NULL, RW_DEFAULT, NULL);
2508
2509 mutex_enter(&vp->v_vsd_lock);
2510 VERIFY(vsd_set(vp, nfs4_srv_vkey, (void *)fp) == 0);
2511 mutex_exit(&vp->v_vsd_lock);
2512
2513 return (TRUE);
2514 }
2515
2516 rfs4_file_t *
2517 rfs4_findfile(vnode_t *vp, nfs_fh4 *fh, bool_t *create)
2518 {
2519 rfs4_file_t *fp;
2520 rfs4_fcreate_arg arg;
2521 nfs4_srv_t *nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
2522
2523 arg.vp = vp;
2524 arg.fh = fh;
2525
2526 if (*create == TRUE)
2527 /* CSTYLED */
2528 fp = (rfs4_file_t *)rfs4_dbsearch(nsrv4->rfs4_file_idx, vp, create,
2529 &arg, RFS4_DBS_VALID);
2530 else {
2531 mutex_enter(&vp->v_vsd_lock);
2532 fp = (rfs4_file_t *)vsd_get(vp, nfs4_srv_vkey);
2533 if (fp) {
2534 rfs4_dbe_lock(fp->rf_dbe);
2535 if (rfs4_dbe_is_invalid(fp->rf_dbe) ||
2536 (rfs4_dbe_refcnt(fp->rf_dbe) == 0)) {
2537 rfs4_dbe_unlock(fp->rf_dbe);
2538 fp = NULL;
2539 } else {
2540 rfs4_dbe_hold(fp->rf_dbe);
2541 rfs4_dbe_unlock(fp->rf_dbe);
2542 }
2543 }
2544 mutex_exit(&vp->v_vsd_lock);
2545 }
2546 return (fp);
2547 }
2548
2549 /*
2550 * Find a file in the db and once it is located, take the rw lock.
2551 * Need to check the vnode pointer and if it does not exist (it was
2552 * removed between the db location and check) redo the find. This
2553 * assumes that a file struct that has a NULL vnode pointer is marked
2554 * at 'invalid' and will not be found in the db the second time
2555 * around.
2556 */
2557 rfs4_file_t *
2558 rfs4_findfile_withlock(vnode_t *vp, nfs_fh4 *fh, bool_t *create)
2559 {
2560 rfs4_file_t *fp;
2561 rfs4_fcreate_arg arg;
2562 bool_t screate = *create;
2563 nfs4_srv_t *nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
2564
2565 if (screate == FALSE) {
2566 mutex_enter(&vp->v_vsd_lock);
2567 fp = (rfs4_file_t *)vsd_get(vp, nfs4_srv_vkey);
2568 if (fp) {
2569 rfs4_dbe_lock(fp->rf_dbe);
2570 if (rfs4_dbe_is_invalid(fp->rf_dbe) ||
2571 (rfs4_dbe_refcnt(fp->rf_dbe) == 0)) {
2572 rfs4_dbe_unlock(fp->rf_dbe);
2573 mutex_exit(&vp->v_vsd_lock);
2574 fp = NULL;
2575 } else {
2576 rfs4_dbe_hold(fp->rf_dbe);
2577 rfs4_dbe_unlock(fp->rf_dbe);
2578 mutex_exit(&vp->v_vsd_lock);
2579 rw_enter(&fp->rf_file_rwlock, RW_WRITER);
2580 if (fp->rf_vp == NULL) {
2581 rw_exit(&fp->rf_file_rwlock);
2582 rfs4_file_rele(fp);
2583 fp = NULL;
2584 }
2585 }
2586 } else {
2587 mutex_exit(&vp->v_vsd_lock);
2588 }
2589 } else {
2590 retry:
2591 arg.vp = vp;
2592 arg.fh = fh;
2593
2594 fp = (rfs4_file_t *)rfs4_dbsearch(nsrv4->rfs4_file_idx, vp,
2595 create, &arg, RFS4_DBS_VALID);
2596 if (fp != NULL) {
2597 rw_enter(&fp->rf_file_rwlock, RW_WRITER);
2598 if (fp->rf_vp == NULL) {
2599 rw_exit(&fp->rf_file_rwlock);
2600 rfs4_file_rele(fp);
2601 *create = screate;
2602 goto retry;
2603 }
2604 }
2605 }
2606
2607 return (fp);
2608 }
2609
2610 static uint32_t
2611 lo_state_hash(void *key)
2612 {
2613 stateid_t *id = key;
2614
2615 return (id->bits.ident+id->bits.pid);
2616 }
2617
2618 static bool_t
2619 lo_state_compare(rfs4_entry_t u_entry, void *key)
2620 {
2621 rfs4_lo_state_t *lsp = (rfs4_lo_state_t *)u_entry;
2622 stateid_t *id = key;
2623 bool_t rc;
2624
2625 rc = (lsp->rls_lockid.bits.boottime == id->bits.boottime &&
2626 lsp->rls_lockid.bits.type == id->bits.type &&
2627 lsp->rls_lockid.bits.ident == id->bits.ident &&
2628 lsp->rls_lockid.bits.pid == id->bits.pid);
2629
2630 return (rc);
2631 }
2632
2633 static void *
2634 lo_state_mkkey(rfs4_entry_t u_entry)
2635 {
2636 rfs4_lo_state_t *lsp = (rfs4_lo_state_t *)u_entry;
2637
2638 return (&lsp->rls_lockid);
2639 }
2640
2641 static bool_t
2642 rfs4_lo_state_expiry(rfs4_entry_t u_entry)
2643 {
2644 rfs4_lo_state_t *lsp = (rfs4_lo_state_t *)u_entry;
2645
2646 if (rfs4_dbe_is_invalid(lsp->rls_dbe))
2647 return (TRUE);
2648 if (lsp->rls_state->rs_closed)
2649 return (TRUE);
2650 return ((gethrestime_sec() -
2651 lsp->rls_state->rs_owner->ro_client->rc_last_access
2652 > rfs4_lease_time));
2653 }
2654
2655 static void
2656 rfs4_lo_state_destroy(rfs4_entry_t u_entry)
2657 {
2658 rfs4_lo_state_t *lsp = (rfs4_lo_state_t *)u_entry;
2659
2660 rfs4_dbe_lock(lsp->rls_state->rs_dbe);
2661 list_remove(&lsp->rls_state->rs_lostatelist, lsp);
2662 rfs4_dbe_unlock(lsp->rls_state->rs_dbe);
2663
2664 rfs4_sw_destroy(&lsp->rls_sw);
2665
2666 /* Make sure to release the file locks */
2667 if (lsp->rls_locks_cleaned == FALSE) {
2668 lsp->rls_locks_cleaned = TRUE;
2669 if (lsp->rls_locker->rl_client->rc_sysidt != LM_NOSYSID) {
2670 /* Is the PxFS kernel module loaded? */
2671 if (lm_remove_file_locks != NULL) {
2672 int new_sysid;
2673
2674 /* Encode the cluster nodeid in new sysid */
2675 new_sysid =
2676 lsp->rls_locker->rl_client->rc_sysidt;
2677 lm_set_nlmid_flk(&new_sysid);
2678
2679 /*
2680 * This PxFS routine removes file locks for a
2681 * client over all nodes of a cluster.
2682 */
2683 DTRACE_PROBE1(nfss_i_clust_rm_lck,
2684 int, new_sysid);
2685 (*lm_remove_file_locks)(new_sysid);
2686 } else {
2687 (void) cleanlocks(
2688 lsp->rls_state->rs_finfo->rf_vp,
2689 lsp->rls_locker->rl_pid,
2690 lsp->rls_locker->rl_client->rc_sysidt);
2691 }
2692 }
2693 }
2694
2695 /* Free the last reply for this state */
2696 rfs4_free_reply(&lsp->rls_reply);
2697
2698 rfs4_lockowner_rele(lsp->rls_locker);
2699 lsp->rls_locker = NULL;
2700
2701 rfs4_state_rele_nounlock(lsp->rls_state);
2702 lsp->rls_state = NULL;
2703 }
2704
2705 static bool_t
2706 rfs4_lo_state_create(rfs4_entry_t u_entry, void *arg)
2707 {
2708 rfs4_lo_state_t *lsp = (rfs4_lo_state_t *)u_entry;
2709 rfs4_lo_state_t *argp = (rfs4_lo_state_t *)arg;
2710 rfs4_lockowner_t *lo = argp->rls_locker;
2711 rfs4_state_t *sp = argp->rls_state;
2712
2713 lsp->rls_state = sp;
2714
2715 lsp->rls_lockid = sp->rs_stateid;
2716 lsp->rls_lockid.bits.type = LOCKID;
2717 lsp->rls_lockid.bits.chgseq = 0;
2718 lsp->rls_lockid.bits.pid = lo->rl_pid;
2719
2720 lsp->rls_locks_cleaned = FALSE;
2721 lsp->rls_lock_completed = FALSE;
2722
2723 rfs4_sw_init(&lsp->rls_sw);
2724
2725 /* Attached the supplied lock owner */
2726 rfs4_dbe_hold(lo->rl_dbe);
2727 lsp->rls_locker = lo;
2728
2729 rfs4_dbe_lock(sp->rs_dbe);
2730 list_insert_tail(&sp->rs_lostatelist, lsp);
2731 rfs4_dbe_hold(sp->rs_dbe);
2732 rfs4_dbe_unlock(sp->rs_dbe);
2733
2734 return (TRUE);
2735 }
2736
2737 void
2738 rfs4_lo_state_rele(rfs4_lo_state_t *lsp, bool_t unlock_fp)
2739 {
2740 if (unlock_fp == TRUE)
2741 rw_exit(&lsp->rls_state->rs_finfo->rf_file_rwlock);
2742 rfs4_dbe_rele(lsp->rls_dbe);
2743 }
2744
2745 static rfs4_lo_state_t *
2746 rfs4_findlo_state(stateid_t *id, bool_t lock_fp)
2747 {
2748 rfs4_lo_state_t *lsp;
2749 bool_t create = FALSE;
2750 nfs4_srv_t *nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
2751
2752 lsp = (rfs4_lo_state_t *)rfs4_dbsearch(nsrv4->rfs4_lo_state_idx, id,
2753 &create, NULL, RFS4_DBS_VALID);
2754 if (lock_fp == TRUE && lsp != NULL)
2755 rw_enter(&lsp->rls_state->rs_finfo->rf_file_rwlock, RW_READER);
2756
2757 return (lsp);
2758 }
2759
2760
2761 static uint32_t
2762 lo_state_lo_hash(void *key)
2763 {
2764 rfs4_lo_state_t *lsp = key;
2765
2766 return (ADDRHASH(lsp->rls_locker) ^ ADDRHASH(lsp->rls_state));
2767 }
2768
2769 static bool_t
2770 lo_state_lo_compare(rfs4_entry_t u_entry, void *key)
2771 {
2772 rfs4_lo_state_t *lsp = (rfs4_lo_state_t *)u_entry;
2773 rfs4_lo_state_t *keyp = key;
2774
2775 return (keyp->rls_locker == lsp->rls_locker &&
2776 keyp->rls_state == lsp->rls_state);
2777 }
2778
2779 static void *
2780 lo_state_lo_mkkey(rfs4_entry_t u_entry)
2781 {
2782 return (u_entry);
2783 }
2784
2785 rfs4_lo_state_t *
2786 rfs4_findlo_state_by_owner(rfs4_lockowner_t *lo, rfs4_state_t *sp,
2787 bool_t *create)
2788 {
2789 rfs4_lo_state_t *lsp;
2790 rfs4_lo_state_t arg;
2791 nfs4_srv_t *nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
2792
2793 arg.rls_locker = lo;
2794 arg.rls_state = sp;
2795
2796 lsp = (rfs4_lo_state_t *)rfs4_dbsearch(nsrv4->rfs4_lo_state_owner_idx,
2797 &arg, create, &arg, RFS4_DBS_VALID);
2798
2799 return (lsp);
2800 }
2801
2802 static stateid_t
2803 get_stateid(id_t eid)
2804 {
2805 stateid_t id;
2806 nfs4_srv_t *nsrv4;
2807
2808 nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
2809
2810 id.bits.boottime = nsrv4->rfs4_start_time;
2811 id.bits.ident = eid;
2812 id.bits.chgseq = 0;
2813 id.bits.type = 0;
2814 id.bits.pid = 0;
2815
2816 /*
2817 * If we are booted as a cluster node, embed our nodeid.
2818 * We've already done sanity checks in rfs4_client_create() so no
2819 * need to repeat them here.
2820 */
2821 id.bits.clnodeid = (cluster_bootflags & CLUSTER_BOOTED) ?
2822 clconf_get_nodeid() : 0;
2823
2824 return (id);
2825 }
2826
2827 /*
2828 * For use only when booted as a cluster node.
2829 * Returns TRUE if the embedded nodeid indicates that this stateid was
2830 * generated on another node.
2831 */
2832 static int
2833 foreign_stateid(stateid_t *id)
2834 {
2835 ASSERT(cluster_bootflags & CLUSTER_BOOTED);
2836 return (id->bits.clnodeid != (uint32_t)clconf_get_nodeid());
2837 }
2838
2839 /*
2840 * For use only when booted as a cluster node.
2841 * Returns TRUE if the embedded nodeid indicates that this clientid was
2842 * generated on another node.
2843 */
2844 static int
2845 foreign_clientid(cid *cidp)
2846 {
2847 ASSERT(cluster_bootflags & CLUSTER_BOOTED);
2848 return (cidp->impl_id.c_id >> CLUSTER_NODEID_SHIFT !=
2849 (uint32_t)clconf_get_nodeid());
2850 }
2851
2852 /*
2853 * For use only when booted as a cluster node.
2854 * Embed our cluster nodeid into the clientid.
2855 */
2856 static void
2857 embed_nodeid(cid *cidp)
2858 {
2859 int clnodeid;
2860 /*
2861 * Currently, our state tables are small enough that their
2862 * ids will leave enough bits free for the nodeid. If the
2863 * tables become larger, we mustn't overwrite the id.
2864 * Equally, we only have room for so many bits of nodeid, so
2865 * must check that too.
2866 */
2867 ASSERT(cluster_bootflags & CLUSTER_BOOTED);
2868 ASSERT(cidp->impl_id.c_id >> CLUSTER_NODEID_SHIFT == 0);
2869 clnodeid = clconf_get_nodeid();
2870 ASSERT(clnodeid <= CLUSTER_MAX_NODEID);
2871 ASSERT(clnodeid != NODEID_UNKNOWN);
2872 cidp->impl_id.c_id |= (clnodeid << CLUSTER_NODEID_SHIFT);
2873 }
2874
2875 static uint32_t
2876 state_hash(void *key)
2877 {
2878 stateid_t *ip = (stateid_t *)key;
2879
2880 return (ip->bits.ident);
2881 }
2882
2883 static bool_t
2884 state_compare(rfs4_entry_t u_entry, void *key)
2885 {
2886 rfs4_state_t *sp = (rfs4_state_t *)u_entry;
2887 stateid_t *id = (stateid_t *)key;
2888 bool_t rc;
2889
2890 rc = (sp->rs_stateid.bits.boottime == id->bits.boottime &&
2891 sp->rs_stateid.bits.ident == id->bits.ident);
2892
2893 return (rc);
2894 }
2895
2896 static void *
2897 state_mkkey(rfs4_entry_t u_entry)
2898 {
2899 rfs4_state_t *sp = (rfs4_state_t *)u_entry;
2900
2901 return (&sp->rs_stateid);
2902 }
2903
2904 static void
2905 rfs4_state_destroy(rfs4_entry_t u_entry)
2906 {
2907 rfs4_state_t *sp = (rfs4_state_t *)u_entry;
2908
2909 /* remove from openowner list */
2910 rfs4_dbe_lock(sp->rs_owner->ro_dbe);
2911 list_remove(&sp->rs_owner->ro_statelist, sp);
2912 rfs4_dbe_unlock(sp->rs_owner->ro_dbe);
2913
2914 list_destroy(&sp->rs_lostatelist);
2915
2916 /* release any share locks for this stateid if it's still open */
2917 if (!sp->rs_closed) {
2918 rfs4_dbe_lock(sp->rs_dbe);
2919 (void) rfs4_unshare(sp);
2920 rfs4_dbe_unlock(sp->rs_dbe);
2921 }
2922
2923 /* Were done with the file */
2924 rfs4_file_rele(sp->rs_finfo);
2925 sp->rs_finfo = NULL;
2926
2927 /* And now with the openowner */
2928 rfs4_openowner_rele(sp->rs_owner);
2929 sp->rs_owner = NULL;
2930 }
2931
2932 static void
2933 rfs4_state_rele_nounlock(rfs4_state_t *sp)
2934 {
2935 rfs4_dbe_rele(sp->rs_dbe);
2936 }
2937
2938 void
2939 rfs4_state_rele(rfs4_state_t *sp)
2940 {
2941 rw_exit(&sp->rs_finfo->rf_file_rwlock);
2942 rfs4_dbe_rele(sp->rs_dbe);
2943 }
2944
2945 static uint32_t
2946 deleg_hash(void *key)
2947 {
2948 rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)key;
2949
2950 return (ADDRHASH(dsp->rds_client) ^ ADDRHASH(dsp->rds_finfo));
2951 }
2952
2953 static bool_t
2954 deleg_compare(rfs4_entry_t u_entry, void *key)
2955 {
2956 rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
2957 rfs4_deleg_state_t *kdsp = (rfs4_deleg_state_t *)key;
2958
2959 return (dsp->rds_client == kdsp->rds_client &&
2960 dsp->rds_finfo == kdsp->rds_finfo);
2961 }
2962
2963 static void *
2964 deleg_mkkey(rfs4_entry_t u_entry)
2965 {
2966 return (u_entry);
2967 }
2968
2969 static uint32_t
2970 deleg_state_hash(void *key)
2971 {
2972 stateid_t *ip = (stateid_t *)key;
2973
2974 return (ip->bits.ident);
2975 }
2976
2977 static bool_t
2978 deleg_state_compare(rfs4_entry_t u_entry, void *key)
2979 {
2980 rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
2981 stateid_t *id = (stateid_t *)key;
2982 bool_t rc;
2983
2984 if (id->bits.type != DELEGID)
2985 return (FALSE);
2986
2987 rc = (dsp->rds_delegid.bits.boottime == id->bits.boottime &&
2988 dsp->rds_delegid.bits.ident == id->bits.ident);
2989
2990 return (rc);
2991 }
2992
2993 static void *
2994 deleg_state_mkkey(rfs4_entry_t u_entry)
2995 {
2996 rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
2997
2998 return (&dsp->rds_delegid);
2999 }
3000
3001 static bool_t
3002 rfs4_deleg_state_expiry(rfs4_entry_t u_entry)
3003 {
3004 rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
3005
3006 if (rfs4_dbe_is_invalid(dsp->rds_dbe))
3007 return (TRUE);
3008
3009 if (dsp->rds_dtype == OPEN_DELEGATE_NONE)
3010 return (TRUE);
3011
3012 if ((gethrestime_sec() - dsp->rds_client->rc_last_access
3013 > rfs4_lease_time)) {
3014 rfs4_dbe_invalidate(dsp->rds_dbe);
3015 return (TRUE);
3016 }
3017
3018 return (FALSE);
3019 }
3020
3021 static bool_t
3022 rfs4_deleg_state_create(rfs4_entry_t u_entry, void *argp)
3023 {
3024 rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
3025 rfs4_file_t *fp = ((rfs4_deleg_state_t *)argp)->rds_finfo;
3026 rfs4_client_t *cp = ((rfs4_deleg_state_t *)argp)->rds_client;
3027
3028 rfs4_dbe_hold(fp->rf_dbe);
3029 rfs4_dbe_hold(cp->rc_dbe);
3030
3031 dsp->rds_delegid = get_stateid(rfs4_dbe_getid(dsp->rds_dbe));
3032 dsp->rds_delegid.bits.type = DELEGID;
3033 dsp->rds_finfo = fp;
3034 dsp->rds_client = cp;
3035 dsp->rds_dtype = OPEN_DELEGATE_NONE;
3036
3037 dsp->rds_time_granted = gethrestime_sec(); /* observability */
3038 dsp->rds_time_revoked = 0;
3039
3040 list_link_init(&dsp->rds_node);
3041
3042 return (TRUE);
3043 }
3044
3045 static void
3046 rfs4_deleg_state_destroy(rfs4_entry_t u_entry)
3047 {
3048 rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
3049
3050 /* return delegation if necessary */
3051 rfs4_return_deleg(dsp, FALSE);
3052
3053 /* Were done with the file */
3054 rfs4_file_rele(dsp->rds_finfo);
3055 dsp->rds_finfo = NULL;
3056
3057 /* And now with the openowner */
3058 rfs4_client_rele(dsp->rds_client);
3059 dsp->rds_client = NULL;
3060 }
3061
3062 rfs4_deleg_state_t *
3063 rfs4_finddeleg(rfs4_state_t *sp, bool_t *create)
3064 {
3065 rfs4_deleg_state_t ds, *dsp;
3066 nfs4_srv_t *nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
3067
3068 ds.rds_client = sp->rs_owner->ro_client;
3069 ds.rds_finfo = sp->rs_finfo;
3070
3071 dsp = (rfs4_deleg_state_t *)rfs4_dbsearch(nsrv4->rfs4_deleg_idx, &ds,
3072 create, &ds, RFS4_DBS_VALID);
3073
3074 return (dsp);
3075 }
3076
3077 rfs4_deleg_state_t *
3078 rfs4_finddelegstate(stateid_t *id)
3079 {
3080 rfs4_deleg_state_t *dsp;
3081 bool_t create = FALSE;
3082 nfs4_srv_t *nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
3083
3084 dsp = (rfs4_deleg_state_t *)rfs4_dbsearch(nsrv4->rfs4_deleg_state_idx,
3085 id, &create, NULL, RFS4_DBS_VALID);
3086
3087 return (dsp);
3088 }
3089
3090 void
3091 rfs4_deleg_state_rele(rfs4_deleg_state_t *dsp)
3092 {
3093 rfs4_dbe_rele(dsp->rds_dbe);
3094 }
3095
3096 void
3097 rfs4_update_lock_sequence(rfs4_lo_state_t *lsp)
3098 {
3099
3100 rfs4_dbe_lock(lsp->rls_dbe);
3101
3102 /*
3103 * If we are skipping sequence id checking, this means that
3104 * this is the first lock request and therefore the sequence
3105 * id does not need to be updated. This only happens on the
3106 * first lock request for a lockowner
3107 */
3108 if (!lsp->rls_skip_seqid_check)
3109 lsp->rls_seqid++;
3110
3111 rfs4_dbe_unlock(lsp->rls_dbe);
3112 }
3113
3114 void
3115 rfs4_update_lock_resp(rfs4_lo_state_t *lsp, nfs_resop4 *resp)
3116 {
3117
3118 rfs4_dbe_lock(lsp->rls_dbe);
3119
3120 rfs4_free_reply(&lsp->rls_reply);
3121
3122 rfs4_copy_reply(&lsp->rls_reply, resp);
3123
3124 rfs4_dbe_unlock(lsp->rls_dbe);
3125 }
3126
3127 void
3128 rfs4_free_opens(rfs4_openowner_t *oo, bool_t invalidate,
3129 bool_t close_of_client)
3130 {
3131 rfs4_state_t *sp;
3132
3133 rfs4_dbe_lock(oo->ro_dbe);
3134
3135 for (sp = list_head(&oo->ro_statelist); sp != NULL;
3136 sp = list_next(&oo->ro_statelist, sp)) {
3137 rfs4_state_close(sp, FALSE, close_of_client, CRED());
3138 if (invalidate == TRUE)
3139 rfs4_dbe_invalidate(sp->rs_dbe);
3140 }
3141
3142 rfs4_dbe_invalidate(oo->ro_dbe);
3143 rfs4_dbe_unlock(oo->ro_dbe);
3144 }
3145
3146 static uint32_t
3147 state_owner_file_hash(void *key)
3148 {
3149 rfs4_state_t *sp = key;
3150
3151 return (ADDRHASH(sp->rs_owner) ^ ADDRHASH(sp->rs_finfo));
3152 }
3153
3154 static bool_t
3155 state_owner_file_compare(rfs4_entry_t u_entry, void *key)
3156 {
3157 rfs4_state_t *sp = (rfs4_state_t *)u_entry;
3158 rfs4_state_t *arg = key;
3159
3160 if (sp->rs_closed == TRUE)
3161 return (FALSE);
3162
3163 return (arg->rs_owner == sp->rs_owner && arg->rs_finfo == sp->rs_finfo);
3164 }
3165
3166 static void *
3167 state_owner_file_mkkey(rfs4_entry_t u_entry)
3168 {
3169 return (u_entry);
3170 }
3171
3172 static uint32_t
3173 state_file_hash(void *key)
3174 {
3175 return (ADDRHASH(key));
3176 }
3177
3178 static bool_t
3179 state_file_compare(rfs4_entry_t u_entry, void *key)
3180 {
3181 rfs4_state_t *sp = (rfs4_state_t *)u_entry;
3182 rfs4_file_t *fp = key;
3183
3184 if (sp->rs_closed == TRUE)
3185 return (FALSE);
3186
3187 return (fp == sp->rs_finfo);
3188 }
3189
3190 static void *
3191 state_file_mkkey(rfs4_entry_t u_entry)
3192 {
3193 rfs4_state_t *sp = (rfs4_state_t *)u_entry;
3194
3195 return (sp->rs_finfo);
3196 }
3197
3198 rfs4_state_t *
3199 rfs4_findstate_by_owner_file(rfs4_openowner_t *oo, rfs4_file_t *fp,
3200 bool_t *create)
3201 {
3202 rfs4_state_t *sp;
3203 rfs4_state_t key;
3204 nfs4_srv_t *nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
3205
3206 key.rs_owner = oo;
3207 key.rs_finfo = fp;
3208
3209 sp = (rfs4_state_t *)rfs4_dbsearch(nsrv4->rfs4_state_owner_file_idx,
3210 &key, create, &key, RFS4_DBS_VALID);
3211
3212 return (sp);
3213 }
3214
3215 /* This returns ANY state struct that refers to this file */
3216 static rfs4_state_t *
3217 rfs4_findstate_by_file(rfs4_file_t *fp)
3218 {
3219 bool_t create = FALSE;
3220 nfs4_srv_t *nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
3221
3222 return ((rfs4_state_t *)rfs4_dbsearch(nsrv4->rfs4_state_file_idx, fp,
3223 &create, fp, RFS4_DBS_VALID));
3224 }
3225
3226 static bool_t
3227 rfs4_state_expiry(rfs4_entry_t u_entry)
3228 {
3229 rfs4_state_t *sp = (rfs4_state_t *)u_entry;
3230
3231 if (rfs4_dbe_is_invalid(sp->rs_dbe))
3232 return (TRUE);
3233
3234 if (sp->rs_closed == TRUE &&
3235 ((gethrestime_sec() - rfs4_dbe_get_timerele(sp->rs_dbe))
3236 > rfs4_lease_time))
3237 return (TRUE);
3238
3239 return ((gethrestime_sec() - sp->rs_owner->ro_client->rc_last_access
3240 > rfs4_lease_time));
3241 }
3242
3243 static bool_t
3244 rfs4_state_create(rfs4_entry_t u_entry, void *argp)
3245 {
3246 rfs4_state_t *sp = (rfs4_state_t *)u_entry;
3247 rfs4_file_t *fp = ((rfs4_state_t *)argp)->rs_finfo;
3248 rfs4_openowner_t *oo = ((rfs4_state_t *)argp)->rs_owner;
3249
3250 rfs4_dbe_hold(fp->rf_dbe);
3251 rfs4_dbe_hold(oo->ro_dbe);
3252 sp->rs_stateid = get_stateid(rfs4_dbe_getid(sp->rs_dbe));
3253 sp->rs_stateid.bits.type = OPENID;
3254 sp->rs_owner = oo;
3255 sp->rs_finfo = fp;
3256
3257 list_create(&sp->rs_lostatelist, sizeof (rfs4_lo_state_t),
3258 offsetof(rfs4_lo_state_t, rls_node));
3259
3260 /* Insert state on per open owner's list */
3261 rfs4_dbe_lock(oo->ro_dbe);
3262 list_insert_tail(&oo->ro_statelist, sp);
3263 rfs4_dbe_unlock(oo->ro_dbe);
3264
3265 return (TRUE);
3266 }
3267
3268 static rfs4_state_t *
3269 rfs4_findstate(stateid_t *id, rfs4_dbsearch_type_t find_invalid, bool_t lock_fp)
3270 {
3271 rfs4_state_t *sp;
3272 bool_t create = FALSE;
3273 nfs4_srv_t *nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
3274
3275 sp = (rfs4_state_t *)rfs4_dbsearch(nsrv4->rfs4_state_idx, id,
3276 &create, NULL, find_invalid);
3277 if (lock_fp == TRUE && sp != NULL)
3278 rw_enter(&sp->rs_finfo->rf_file_rwlock, RW_READER);
3279
3280 return (sp);
3281 }
3282
3283 void
3284 rfs4_state_close(rfs4_state_t *sp, bool_t lock_held, bool_t close_of_client,
3285 cred_t *cr)
3286 {
3287 /* Remove the associated lo_state owners */
3288 if (!lock_held)
3289 rfs4_dbe_lock(sp->rs_dbe);
3290
3291 /*
3292 * If refcnt == 0, the dbe is about to be destroyed.
3293 * lock state will be released by the reaper thread.
3294 */
3295
3296 if (rfs4_dbe_refcnt(sp->rs_dbe) > 0) {
3297 if (sp->rs_closed == FALSE) {
3298 rfs4_release_share_lock_state(sp, cr, close_of_client);
3299 sp->rs_closed = TRUE;
3300 }
3301 }
3302
3303 if (!lock_held)
3304 rfs4_dbe_unlock(sp->rs_dbe);
3305 }
3306
3307 /*
3308 * Remove all state associated with the given client.
3309 */
3310 void
3311 rfs4_client_state_remove(rfs4_client_t *cp)
3312 {
3313 rfs4_openowner_t *oo;
3314
3315 rfs4_dbe_lock(cp->rc_dbe);
3316
3317 for (oo = list_head(&cp->rc_openownerlist); oo != NULL;
3318 oo = list_next(&cp->rc_openownerlist, oo)) {
3319 rfs4_free_opens(oo, TRUE, TRUE);
3320 }
3321
3322 rfs4_dbe_unlock(cp->rc_dbe);
3323 }
3324
3325 void
3326 rfs4_client_close(rfs4_client_t *cp)
3327 {
3328 /* Mark client as going away. */
3329 rfs4_dbe_lock(cp->rc_dbe);
3330 rfs4_dbe_invalidate(cp->rc_dbe);
3331 rfs4_dbe_unlock(cp->rc_dbe);
3332
3333 rfs4_client_state_remove(cp);
3334
3335 /* Release the client */
3336 rfs4_client_rele(cp);
3337 }
3338
3339 nfsstat4
3340 rfs4_check_clientid(clientid4 *cp, int setclid_confirm)
3341 {
3342 cid *cidp = (cid *) cp;
3343 nfs4_srv_t *nsrv4;
3344
3345 nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
3346
3347 /*
3348 * If we are booted as a cluster node, check the embedded nodeid.
3349 * If it indicates that this clientid was generated on another node,
3350 * inform the client accordingly.
3351 */
3352 if (cluster_bootflags & CLUSTER_BOOTED && foreign_clientid(cidp))
3353 return (NFS4ERR_STALE_CLIENTID);
3354
3355 /*
3356 * If the server start time matches the time provided
3357 * by the client (via the clientid) and this is NOT a
3358 * setclientid_confirm then return EXPIRED.
3359 */
3360 if (!setclid_confirm &&
3361 cidp->impl_id.start_time == nsrv4->rfs4_start_time)
3362 return (NFS4ERR_EXPIRED);
3363
3364 return (NFS4ERR_STALE_CLIENTID);
3365 }
3366
3367 /*
3368 * This is used when a stateid has not been found amongst the
3369 * current server's state. Check the stateid to see if it
3370 * was from this server instantiation or not.
3371 */
3372 static nfsstat4
3373 what_stateid_error(stateid_t *id, stateid_type_t type)
3374 {
3375 nfs4_srv_t *nsrv4;
3376
3377 nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
3378
3379 /* If we are booted as a cluster node, was stateid locally generated? */
3380 if ((cluster_bootflags & CLUSTER_BOOTED) && foreign_stateid(id))
3381 return (NFS4ERR_STALE_STATEID);
3382
3383 /* If types don't match then no use checking further */
3384 if (type != id->bits.type)
3385 return (NFS4ERR_BAD_STATEID);
3386
3387 /* From a different server instantiation, return STALE */
3388 if (id->bits.boottime != nsrv4->rfs4_start_time)
3389 return (NFS4ERR_STALE_STATEID);
3390
3391 /*
3392 * From this server but the state is most likely beyond lease
3393 * timeout: return NFS4ERR_EXPIRED. However, there is the
3394 * case of a delegation stateid. For delegations, there is a
3395 * case where the state can be removed without the client's
3396 * knowledge/consent: revocation. In the case of delegation
3397 * revocation, the delegation state will be removed and will
3398 * not be found. If the client does something like a
3399 * DELEGRETURN or even a READ/WRITE with a delegatoin stateid
3400 * that has been revoked, the server should return BAD_STATEID
3401 * instead of the more common EXPIRED error.
3402 */
3403 if (id->bits.boottime == nsrv4->rfs4_start_time) {
3404 if (type == DELEGID)
3405 return (NFS4ERR_BAD_STATEID);
3406 else
3407 return (NFS4ERR_EXPIRED);
3408 }
3409
3410 return (NFS4ERR_BAD_STATEID);
3411 }
3412
3413 /*
3414 * Used later on to find the various state structs. When called from
3415 * rfs4_check_stateid()->rfs4_get_all_state(), no file struct lock is
3416 * taken (it is not needed) and helps on the read/write path with
3417 * respect to performance.
3418 */
3419 static nfsstat4
3420 rfs4_get_state_lockit(stateid4 *stateid, rfs4_state_t **spp,
3421 rfs4_dbsearch_type_t find_invalid, bool_t lock_fp)
3422 {
3423 stateid_t *id = (stateid_t *)stateid;
3424 rfs4_state_t *sp;
3425
3426 *spp = NULL;
3427
3428 /* If we are booted as a cluster node, was stateid locally generated? */
3429 if ((cluster_bootflags & CLUSTER_BOOTED) && foreign_stateid(id))
3430 return (NFS4ERR_STALE_STATEID);
3431
3432 sp = rfs4_findstate(id, find_invalid, lock_fp);
3433 if (sp == NULL) {
3434 return (what_stateid_error(id, OPENID));
3435 }
3436
3437 if (rfs4_lease_expired(sp->rs_owner->ro_client)) {
3438 if (lock_fp == TRUE)
3439 rfs4_state_rele(sp);
3440 else
3441 rfs4_state_rele_nounlock(sp);
3442 return (NFS4ERR_EXPIRED);
3443 }
3444
3445 *spp = sp;
3446
3447 return (NFS4_OK);
3448 }
3449
3450 nfsstat4
3451 rfs4_get_state(stateid4 *stateid, rfs4_state_t **spp,
3452 rfs4_dbsearch_type_t find_invalid)
3453 {
3454 return (rfs4_get_state_lockit(stateid, spp, find_invalid, TRUE));
3455 }
3456
3457 int
3458 rfs4_check_stateid_seqid(rfs4_state_t *sp, stateid4 *stateid)
3459 {
3460 stateid_t *id = (stateid_t *)stateid;
3461
3462 if (rfs4_lease_expired(sp->rs_owner->ro_client))
3463 return (NFS4_CHECK_STATEID_EXPIRED);
3464
3465 /* Stateid is some time in the future - that's bad */
3466 if (sp->rs_stateid.bits.chgseq < id->bits.chgseq)
3467 return (NFS4_CHECK_STATEID_BAD);
3468
3469 if (sp->rs_stateid.bits.chgseq == id->bits.chgseq + 1)
3470 return (NFS4_CHECK_STATEID_REPLAY);
3471
3472 /* Stateid is some time in the past - that's old */
3473 if (sp->rs_stateid.bits.chgseq > id->bits.chgseq)
3474 return (NFS4_CHECK_STATEID_OLD);
3475
3476 /* Caller needs to know about confirmation before closure */
3477 if (sp->rs_owner->ro_need_confirm)
3478 return (NFS4_CHECK_STATEID_UNCONFIRMED);
3479
3480 if (sp->rs_closed == TRUE)
3481 return (NFS4_CHECK_STATEID_CLOSED);
3482
3483 return (NFS4_CHECK_STATEID_OKAY);
3484 }
3485
3486 int
3487 rfs4_check_lo_stateid_seqid(rfs4_lo_state_t *lsp, stateid4 *stateid)
3488 {
3489 stateid_t *id = (stateid_t *)stateid;
3490
3491 if (rfs4_lease_expired(lsp->rls_state->rs_owner->ro_client))
3492 return (NFS4_CHECK_STATEID_EXPIRED);
3493
3494 /* Stateid is some time in the future - that's bad */
3495 if (lsp->rls_lockid.bits.chgseq < id->bits.chgseq)
3496 return (NFS4_CHECK_STATEID_BAD);
3497
3498 if (lsp->rls_lockid.bits.chgseq == id->bits.chgseq + 1)
3499 return (NFS4_CHECK_STATEID_REPLAY);
3500
3501 /* Stateid is some time in the past - that's old */
3502 if (lsp->rls_lockid.bits.chgseq > id->bits.chgseq)
3503 return (NFS4_CHECK_STATEID_OLD);
3504
3505 if (lsp->rls_state->rs_closed == TRUE)
3506 return (NFS4_CHECK_STATEID_CLOSED);
3507
3508 return (NFS4_CHECK_STATEID_OKAY);
3509 }
3510
3511 nfsstat4
3512 rfs4_get_deleg_state(stateid4 *stateid, rfs4_deleg_state_t **dspp)
3513 {
3514 stateid_t *id = (stateid_t *)stateid;
3515 rfs4_deleg_state_t *dsp;
3516
3517 *dspp = NULL;
3518
3519 /* If we are booted as a cluster node, was stateid locally generated? */
3520 if ((cluster_bootflags & CLUSTER_BOOTED) && foreign_stateid(id))
3521 return (NFS4ERR_STALE_STATEID);
3522
3523 dsp = rfs4_finddelegstate(id);
3524 if (dsp == NULL) {
3525 return (what_stateid_error(id, DELEGID));
3526 }
3527
3528 if (rfs4_lease_expired(dsp->rds_client)) {
3529 rfs4_deleg_state_rele(dsp);
3530 return (NFS4ERR_EXPIRED);
3531 }
3532
3533 *dspp = dsp;
3534
3535 return (NFS4_OK);
3536 }
3537
3538 nfsstat4
3539 rfs4_get_lo_state(stateid4 *stateid, rfs4_lo_state_t **lspp, bool_t lock_fp)
3540 {
3541 stateid_t *id = (stateid_t *)stateid;
3542 rfs4_lo_state_t *lsp;
3543
3544 *lspp = NULL;
3545
3546 /* If we are booted as a cluster node, was stateid locally generated? */
3547 if ((cluster_bootflags & CLUSTER_BOOTED) && foreign_stateid(id))
3548 return (NFS4ERR_STALE_STATEID);
3549
3550 lsp = rfs4_findlo_state(id, lock_fp);
3551 if (lsp == NULL) {
3552 return (what_stateid_error(id, LOCKID));
3553 }
3554
3555 if (rfs4_lease_expired(lsp->rls_state->rs_owner->ro_client)) {
3556 rfs4_lo_state_rele(lsp, lock_fp);
3557 return (NFS4ERR_EXPIRED);
3558 }
3559
3560 *lspp = lsp;
3561
3562 return (NFS4_OK);
3563 }
3564
3565 static nfsstat4
3566 rfs4_get_all_state(stateid4 *sid, rfs4_state_t **spp,
3567 rfs4_deleg_state_t **dspp, rfs4_lo_state_t **lspp)
3568 {
3569 rfs4_state_t *sp = NULL;
3570 rfs4_deleg_state_t *dsp = NULL;
3571 rfs4_lo_state_t *lsp = NULL;
3572 stateid_t *id;
3573 nfsstat4 status;
3574
3575 *spp = NULL; *dspp = NULL; *lspp = NULL;
3576
3577 id = (stateid_t *)sid;
3578 switch (id->bits.type) {
3579 case OPENID:
3580 status = rfs4_get_state_lockit(sid, &sp, FALSE, FALSE);
3581 break;
3582 case DELEGID:
3583 status = rfs4_get_deleg_state(sid, &dsp);
3584 break;
3585 case LOCKID:
3586 status = rfs4_get_lo_state(sid, &lsp, FALSE);
3587 if (status == NFS4_OK) {
3588 sp = lsp->rls_state;
3589 rfs4_dbe_hold(sp->rs_dbe);
3590 }
3591 break;
3592 default:
3593 status = NFS4ERR_BAD_STATEID;
3594 }
3595
3596 if (status == NFS4_OK) {
3597 *spp = sp;
3598 *dspp = dsp;
3599 *lspp = lsp;
3600 }
3601
3602 return (status);
3603 }
3604
3605 /*
3606 * Given the I/O mode (FREAD or FWRITE), this checks whether the
3607 * rfs4_state_t struct has access to do this operation and if so
3608 * return NFS4_OK; otherwise the proper NFSv4 error is returned.
3609 */
3610 nfsstat4
3611 rfs4_state_has_access(rfs4_state_t *sp, int mode, vnode_t *vp)
3612 {
3613 nfsstat4 stat = NFS4_OK;
3614 rfs4_file_t *fp;
3615 bool_t create = FALSE;
3616
3617 rfs4_dbe_lock(sp->rs_dbe);
3618 if (mode == FWRITE) {
3619 if (!(sp->rs_share_access & OPEN4_SHARE_ACCESS_WRITE)) {
3620 stat = NFS4ERR_OPENMODE;
3621 }
3622 } else if (mode == FREAD) {
3623 if (!(sp->rs_share_access & OPEN4_SHARE_ACCESS_READ)) {
3624 /*
3625 * If we have OPENed the file with DENYing access
3626 * to both READ and WRITE then no one else could
3627 * have OPENed the file, hence no conflicting READ
3628 * deny. This check is merely an optimization.
3629 */
3630 if (sp->rs_share_deny == OPEN4_SHARE_DENY_BOTH)
3631 goto out;
3632
3633 /* Check against file struct's DENY mode */
3634 fp = rfs4_findfile(vp, NULL, &create);
3635 if (fp != NULL) {
3636 int deny_read = 0;
3637 rfs4_dbe_lock(fp->rf_dbe);
3638 /*
3639 * Check if any other open owner has the file
3640 * OPENed with deny READ.
3641 */
3642 if (sp->rs_share_deny & OPEN4_SHARE_DENY_READ)
3643 deny_read = 1;
3644 ASSERT(fp->rf_deny_read >= deny_read);
3645 if (fp->rf_deny_read > deny_read)
3646 stat = NFS4ERR_OPENMODE;
3647 rfs4_dbe_unlock(fp->rf_dbe);
3648 rfs4_file_rele(fp);
3649 }
3650 }
3651 } else {
3652 /* Illegal I/O mode */
3653 stat = NFS4ERR_INVAL;
3654 }
3655 out:
3656 rfs4_dbe_unlock(sp->rs_dbe);
3657 return (stat);
3658 }
3659
3660 /*
3661 * Given the I/O mode (FREAD or FWRITE), the vnode, the stateid and whether
3662 * the file is being truncated, return NFS4_OK if allowed or appropriate
3663 * V4 error if not. Note NFS4ERR_DELAY will be returned and a recall on
3664 * the associated file will be done if the I/O is not consistent with any
3665 * delegation in effect on the file. Should be holding VOP_RWLOCK, either
3666 * as reader or writer as appropriate. rfs4_op_open will acquire the
3667 * VOP_RWLOCK as writer when setting up delegation. If the stateid is bad
3668 * this routine will return NFS4ERR_BAD_STATEID. In addition, through the
3669 * deleg parameter, we will return whether a write delegation is held by
3670 * the client associated with this stateid.
3671 * If the server instance associated with the relevant client is in its
3672 * grace period, return NFS4ERR_GRACE.
3673 */
3674
3675 nfsstat4
3676 rfs4_check_stateid(int mode, vnode_t *vp,
3677 stateid4 *stateid, bool_t trunc, bool_t *deleg,
3678 bool_t do_access, caller_context_t *ct)
3679 {
3680 rfs4_file_t *fp;
3681 bool_t create = FALSE;
3682 rfs4_state_t *sp;
3683 rfs4_deleg_state_t *dsp;
3684 rfs4_lo_state_t *lsp;
3685 stateid_t *id = (stateid_t *)stateid;
3686 nfsstat4 stat = NFS4_OK;
3687
3688 if (ct != NULL) {
3689 ct->cc_sysid = 0;
3690 ct->cc_pid = 0;
3691 ct->cc_caller_id = nfs4_srv_caller_id;
3692 ct->cc_flags = CC_DONTBLOCK;
3693 }
3694
3695 if (ISSPECIAL(stateid)) {
3696 fp = rfs4_findfile(vp, NULL, &create);
3697 if (fp == NULL)
3698 return (NFS4_OK);
3699 if (fp->rf_dinfo.rd_dtype == OPEN_DELEGATE_NONE) {
3700 rfs4_file_rele(fp);
3701 return (NFS4_OK);
3702 }
3703 if (mode == FWRITE ||
3704 fp->rf_dinfo.rd_dtype == OPEN_DELEGATE_WRITE) {
3705 rfs4_recall_deleg(fp, trunc, NULL);
3706 rfs4_file_rele(fp);
3707 return (NFS4ERR_DELAY);
3708 }
3709 rfs4_file_rele(fp);
3710 return (NFS4_OK);
3711 } else {
3712 stat = rfs4_get_all_state(stateid, &sp, &dsp, &lsp);
3713 if (stat != NFS4_OK)
3714 return (stat);
3715 if (lsp != NULL) {
3716 /* Is associated server instance in its grace period? */
3717 if (rfs4_clnt_in_grace(lsp->rls_locker->rl_client)) {
3718 rfs4_lo_state_rele(lsp, FALSE);
3719 if (sp != NULL)
3720 rfs4_state_rele_nounlock(sp);
3721 return (NFS4ERR_GRACE);
3722 }
3723 if (id->bits.type == LOCKID) {
3724 /* Seqid in the future? - that's bad */
3725 if (lsp->rls_lockid.bits.chgseq <
3726 id->bits.chgseq) {
3727 rfs4_lo_state_rele(lsp, FALSE);
3728 if (sp != NULL)
3729 rfs4_state_rele_nounlock(sp);
3730 return (NFS4ERR_BAD_STATEID);
3731 }
3732 /* Seqid in the past? - that's old */
3733 if (lsp->rls_lockid.bits.chgseq >
3734 id->bits.chgseq) {
3735 rfs4_lo_state_rele(lsp, FALSE);
3736 if (sp != NULL)
3737 rfs4_state_rele_nounlock(sp);
3738 return (NFS4ERR_OLD_STATEID);
3739 }
3740 /* Ensure specified filehandle matches */
3741 if (lsp->rls_state->rs_finfo->rf_vp != vp) {
3742 rfs4_lo_state_rele(lsp, FALSE);
3743 if (sp != NULL)
3744 rfs4_state_rele_nounlock(sp);
3745 return (NFS4ERR_BAD_STATEID);
3746 }
3747 }
3748 if (ct != NULL) {
3749 ct->cc_sysid =
3750 lsp->rls_locker->rl_client->rc_sysidt;
3751 ct->cc_pid = lsp->rls_locker->rl_pid;
3752 }
3753 rfs4_lo_state_rele(lsp, FALSE);
3754 }
3755
3756 /* Stateid provided was an "open" stateid */
3757 if (sp != NULL) {
3758 /* Is associated server instance in its grace period? */
3759 if (rfs4_clnt_in_grace(sp->rs_owner->ro_client)) {
3760 rfs4_state_rele_nounlock(sp);
3761 return (NFS4ERR_GRACE);
3762 }
3763 if (id->bits.type == OPENID) {
3764 /* Seqid in the future? - that's bad */
3765 if (sp->rs_stateid.bits.chgseq <
3766 id->bits.chgseq) {
3767 rfs4_state_rele_nounlock(sp);
3768 return (NFS4ERR_BAD_STATEID);
3769 }
3770 /* Seqid in the past - that's old */
3771 if (sp->rs_stateid.bits.chgseq >
3772 id->bits.chgseq) {
3773 rfs4_state_rele_nounlock(sp);
3774 return (NFS4ERR_OLD_STATEID);
3775 }
3776 }
3777 /* Ensure specified filehandle matches */
3778 if (sp->rs_finfo->rf_vp != vp) {
3779 rfs4_state_rele_nounlock(sp);
3780 return (NFS4ERR_BAD_STATEID);
3781 }
3782
3783 if (sp->rs_owner->ro_need_confirm) {
3784 rfs4_state_rele_nounlock(sp);
3785 return (NFS4ERR_BAD_STATEID);
3786 }
3787
3788 if (sp->rs_closed == TRUE) {
3789 rfs4_state_rele_nounlock(sp);
3790 return (NFS4ERR_OLD_STATEID);
3791 }
3792
3793 if (do_access)
3794 stat = rfs4_state_has_access(sp, mode, vp);
3795 else
3796 stat = NFS4_OK;
3797
3798 /*
3799 * Return whether this state has write
3800 * delegation if desired
3801 */
3802 if (deleg && (sp->rs_finfo->rf_dinfo.rd_dtype ==
3803 OPEN_DELEGATE_WRITE))
3804 *deleg = TRUE;
3805
3806 /*
3807 * We got a valid stateid, so we update the
3808 * lease on the client. Ideally we would like
3809 * to do this after the calling op succeeds,
3810 * but for now this will be good
3811 * enough. Callers of this routine are
3812 * currently insulated from the state stuff.
3813 */
3814 rfs4_update_lease(sp->rs_owner->ro_client);
3815
3816 /*
3817 * If a delegation is present on this file and
3818 * this is a WRITE, then update the lastwrite
3819 * time to indicate that activity is present.
3820 */
3821 if (sp->rs_finfo->rf_dinfo.rd_dtype ==
3822 OPEN_DELEGATE_WRITE &&
3823 mode == FWRITE) {
3824 sp->rs_finfo->rf_dinfo.rd_time_lastwrite =
3825 gethrestime_sec();
3826 }
3827
3828 rfs4_state_rele_nounlock(sp);
3829
3830 return (stat);
3831 }
3832
3833 if (dsp != NULL) {
3834 /* Is associated server instance in its grace period? */
3835 if (rfs4_clnt_in_grace(dsp->rds_client)) {
3836 rfs4_deleg_state_rele(dsp);
3837 return (NFS4ERR_GRACE);
3838 }
3839 if (dsp->rds_delegid.bits.chgseq != id->bits.chgseq) {
3840 rfs4_deleg_state_rele(dsp);
3841 return (NFS4ERR_BAD_STATEID);
3842 }
3843
3844 /* Ensure specified filehandle matches */
3845 if (dsp->rds_finfo->rf_vp != vp) {
3846 rfs4_deleg_state_rele(dsp);
3847 return (NFS4ERR_BAD_STATEID);
3848 }
3849 /*
3850 * Return whether this state has write
3851 * delegation if desired
3852 */
3853 if (deleg && (dsp->rds_finfo->rf_dinfo.rd_dtype ==
3854 OPEN_DELEGATE_WRITE))
3855 *deleg = TRUE;
3856
3857 rfs4_update_lease(dsp->rds_client);
3858
3859 /*
3860 * If a delegation is present on this file and
3861 * this is a WRITE, then update the lastwrite
3862 * time to indicate that activity is present.
3863 */
3864 if (dsp->rds_finfo->rf_dinfo.rd_dtype ==
3865 OPEN_DELEGATE_WRITE && mode == FWRITE) {
3866 dsp->rds_finfo->rf_dinfo.rd_time_lastwrite =
3867 gethrestime_sec();
3868 }
3869
3870 /*
3871 * XXX - what happens if this is a WRITE and the
3872 * delegation type of for READ.
3873 */
3874 rfs4_deleg_state_rele(dsp);
3875
3876 return (stat);
3877 }
3878 /*
3879 * If we got this far, something bad happened
3880 */
3881 return (NFS4ERR_BAD_STATEID);
3882 }
3883 }
3884
3885
3886 /*
3887 * This is a special function in that for the file struct provided the
3888 * server wants to remove/close all current state associated with the
3889 * file. The prime use of this would be with OP_REMOVE to force the
3890 * release of state and particularly of file locks.
3891 *
3892 * There is an assumption that there is no delegations outstanding on
3893 * this file at this point. The caller should have waited for those
3894 * to be returned or revoked.
3895 */
3896 void
3897 rfs4_close_all_state(rfs4_file_t *fp)
3898 {
3899 rfs4_state_t *sp;
3900
3901 rfs4_dbe_lock(fp->rf_dbe);
3902
3903 #ifdef DEBUG
3904 /* only applies when server is handing out delegations */
3905 if (nfs4_get_deleg_policy() != SRV_NEVER_DELEGATE)
3906 ASSERT(fp->rf_dinfo.rd_hold_grant > 0);
3907 #endif
3908
3909 /* No delegations for this file */
3910 ASSERT(list_is_empty(&fp->rf_delegstatelist));
3911
3912 /* Make sure that it can not be found */
3913 rfs4_dbe_invalidate(fp->rf_dbe);
3914
3915 if (fp->rf_vp == NULL) {
3916 rfs4_dbe_unlock(fp->rf_dbe);
3917 return;
3918 }
3919 rfs4_dbe_unlock(fp->rf_dbe);
3920
3921 /*
3922 * Hold as writer to prevent other server threads from
3923 * processing requests related to the file while all state is
3924 * being removed.
3925 */
3926 rw_enter(&fp->rf_file_rwlock, RW_WRITER);
3927
3928 /* Remove ALL state from the file */
3929 while (sp = rfs4_findstate_by_file(fp)) {
3930 rfs4_state_close(sp, FALSE, FALSE, CRED());
3931 rfs4_state_rele_nounlock(sp);
3932 }
3933
3934 /*
3935 * This is only safe since there are no further references to
3936 * the file.
3937 */
3938 rfs4_dbe_lock(fp->rf_dbe);
3939 if (fp->rf_vp) {
3940 vnode_t *vp = fp->rf_vp;
3941
3942 mutex_enter(&vp->v_vsd_lock);
3943 (void) vsd_set(vp, nfs4_srv_vkey, NULL);
3944 mutex_exit(&vp->v_vsd_lock);
3945 VN_RELE(vp);
3946 fp->rf_vp = NULL;
3947 }
3948 rfs4_dbe_unlock(fp->rf_dbe);
3949
3950 /* Finally let other references to proceed */
3951 rw_exit(&fp->rf_file_rwlock);
3952 }
3953
3954 /*
3955 * This function is used as a target for the rfs4_dbe_walk() call
3956 * below. The purpose of this function is to see if the
3957 * lockowner_state refers to a file that resides within the exportinfo
3958 * export. If so, then remove the lock_owner state (file locks and
3959 * share "locks") for this object since the intent is the server is
3960 * unexporting the specified directory. Be sure to invalidate the
3961 * object after the state has been released
3962 */
3963 static void
3964 rfs4_lo_state_walk_callout(rfs4_entry_t u_entry, void *e)
3965 {
3966 rfs4_lo_state_t *lsp = (rfs4_lo_state_t *)u_entry;
3967 struct exportinfo *exi = (struct exportinfo *)e;
3968 nfs_fh4_fmt_t fhfmt4, *exi_fhp, *finfo_fhp;
3969 fhandle_t *efhp;
3970
3971 efhp = (fhandle_t *)&exi->exi_fh;
3972 exi_fhp = (nfs_fh4_fmt_t *)&fhfmt4;
3973
3974 FH_TO_FMT4(efhp, exi_fhp);
3975
3976 finfo_fhp = (nfs_fh4_fmt_t *)lsp->rls_state->rs_finfo->
3977 rf_filehandle.nfs_fh4_val;
3978
3979 if (EQFSID(&finfo_fhp->fh4_fsid, &exi_fhp->fh4_fsid) &&
3980 bcmp(&finfo_fhp->fh4_xdata, &exi_fhp->fh4_xdata,
3981 exi_fhp->fh4_xlen) == 0) {
3982 rfs4_state_close(lsp->rls_state, FALSE, FALSE, CRED());
3983 rfs4_dbe_invalidate(lsp->rls_dbe);
3984 rfs4_dbe_invalidate(lsp->rls_state->rs_dbe);
3985 }
3986 }
3987
3988 /*
3989 * This function is used as a target for the rfs4_dbe_walk() call
3990 * below. The purpose of this function is to see if the state refers
3991 * to a file that resides within the exportinfo export. If so, then
3992 * remove the open state for this object since the intent is the
3993 * server is unexporting the specified directory. The main result for
3994 * this type of entry is to invalidate it such it will not be found in
3995 * the future.
3996 */
3997 static void
3998 rfs4_state_walk_callout(rfs4_entry_t u_entry, void *e)
3999 {
4000 rfs4_state_t *sp = (rfs4_state_t *)u_entry;
4001 struct exportinfo *exi = (struct exportinfo *)e;
4002 nfs_fh4_fmt_t fhfmt4, *exi_fhp, *finfo_fhp;
4003 fhandle_t *efhp;
4004
4005 efhp = (fhandle_t *)&exi->exi_fh;
4006 exi_fhp = (nfs_fh4_fmt_t *)&fhfmt4;
4007
4008 FH_TO_FMT4(efhp, exi_fhp);
4009
4010 finfo_fhp =
4011 (nfs_fh4_fmt_t *)sp->rs_finfo->rf_filehandle.nfs_fh4_val;
4012
4013 if (EQFSID(&finfo_fhp->fh4_fsid, &exi_fhp->fh4_fsid) &&
4014 bcmp(&finfo_fhp->fh4_xdata, &exi_fhp->fh4_xdata,
4015 exi_fhp->fh4_xlen) == 0) {
4016 rfs4_state_close(sp, TRUE, FALSE, CRED());
4017 rfs4_dbe_invalidate(sp->rs_dbe);
4018 }
4019 }
4020
4021 /*
4022 * This function is used as a target for the rfs4_dbe_walk() call
4023 * below. The purpose of this function is to see if the state refers
4024 * to a file that resides within the exportinfo export. If so, then
4025 * remove the deleg state for this object since the intent is the
4026 * server is unexporting the specified directory. The main result for
4027 * this type of entry is to invalidate it such it will not be found in
4028 * the future.
4029 */
4030 static void
4031 rfs4_deleg_state_walk_callout(rfs4_entry_t u_entry, void *e)
4032 {
4033 rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
4034 struct exportinfo *exi = (struct exportinfo *)e;
4035 nfs_fh4_fmt_t fhfmt4, *exi_fhp, *finfo_fhp;
4036 fhandle_t *efhp;
4037
4038 efhp = (fhandle_t *)&exi->exi_fh;
4039 exi_fhp = (nfs_fh4_fmt_t *)&fhfmt4;
4040
4041 FH_TO_FMT4(efhp, exi_fhp);
4042
4043 finfo_fhp =
4044 (nfs_fh4_fmt_t *)dsp->rds_finfo->rf_filehandle.nfs_fh4_val;
4045
4046 if (EQFSID(&finfo_fhp->fh4_fsid, &exi_fhp->fh4_fsid) &&
4047 bcmp(&finfo_fhp->fh4_xdata, &exi_fhp->fh4_xdata,
4048 exi_fhp->fh4_xlen) == 0) {
4049 rfs4_dbe_invalidate(dsp->rds_dbe);
4050 }
4051 }
4052
4053 /*
4054 * This function is used as a target for the rfs4_dbe_walk() call
4055 * below. The purpose of this function is to see if the state refers
4056 * to a file that resides within the exportinfo export. If so, then
4057 * release vnode hold for this object since the intent is the server
4058 * is unexporting the specified directory. Invalidation will prevent
4059 * this struct from being found in the future.
4060 */
4061 static void
4062 rfs4_file_walk_callout(rfs4_entry_t u_entry, void *e)
4063 {
4064 rfs4_file_t *fp = (rfs4_file_t *)u_entry;
4065 struct exportinfo *exi = (struct exportinfo *)e;
4066 nfs_fh4_fmt_t fhfmt4, *exi_fhp, *finfo_fhp;
4067 fhandle_t *efhp;
4068
4069 efhp = (fhandle_t *)&exi->exi_fh;
4070 exi_fhp = (nfs_fh4_fmt_t *)&fhfmt4;
4071
4072 FH_TO_FMT4(efhp, exi_fhp);
4073
4074 finfo_fhp = (nfs_fh4_fmt_t *)fp->rf_filehandle.nfs_fh4_val;
4075
4076 if (EQFSID(&finfo_fhp->fh4_fsid, &exi_fhp->fh4_fsid) &&
4077 bcmp(&finfo_fhp->fh4_xdata, &exi_fhp->fh4_xdata,
4078 exi_fhp->fh4_xlen) == 0) {
4079 if (fp->rf_vp) {
4080 vnode_t *vp = fp->rf_vp;
4081
4082 /*
4083 * don't leak monitors and remove the reference
4084 * put on the vnode when the delegation was granted.
4085 */
4086 if (fp->rf_dinfo.rd_dtype == OPEN_DELEGATE_READ) {
4087 (void) fem_uninstall(vp, deleg_rdops,
4088 (void *)fp);
4089 vn_open_downgrade(vp, FREAD);
4090 } else if (fp->rf_dinfo.rd_dtype ==
4091 OPEN_DELEGATE_WRITE) {
4092 (void) fem_uninstall(vp, deleg_wrops,
4093 (void *)fp);
4094 vn_open_downgrade(vp, FREAD|FWRITE);
4095 }
4096 mutex_enter(&vp->v_vsd_lock);
4097 (void) vsd_set(vp, nfs4_srv_vkey, NULL);
4098 mutex_exit(&vp->v_vsd_lock);
4099 VN_RELE(vp);
4100 fp->rf_vp = NULL;
4101 }
4102 rfs4_dbe_invalidate(fp->rf_dbe);
4103 }
4104 }
4105
4106 /*
4107 * Given a directory that is being unexported, cleanup/release all
4108 * state in the server that refers to objects residing underneath this
4109 * particular export. The ordering of the release is important.
4110 * Lock_owner, then state and then file.
4111 */
4112 void
4113 rfs4_clean_state_exi(struct exportinfo *exi)
4114 {
4115 nfs4_srv_t *nsrv4;
4116
4117 /* curzone mightn't be exi_zone, so use exi_zone instead. */
4118 ASSERT(exi->exi_zone == curzone || curzone == global_zone);
4119 nsrv4 = zone_getspecific(rfs4_zone_key, exi->exi_zone);
4120 if (nsrv4 == NULL) /* NOTE: NFSv4 cleanup MAY have already happened. */
4121 return;
4122 mutex_enter(&nsrv4->state_lock);
4123
4124 if (nsrv4->nfs4_server_state == NULL) {
4125 mutex_exit(&nsrv4->state_lock);
4126 return;
4127 }
4128
4129 /* CSTYLED */
4130 rfs4_dbe_walk(nsrv4->rfs4_lo_state_tab, rfs4_lo_state_walk_callout, exi);
4131 rfs4_dbe_walk(nsrv4->rfs4_state_tab, rfs4_state_walk_callout, exi);
4132 /* CSTYLED */
4133 rfs4_dbe_walk(nsrv4->rfs4_deleg_state_tab, rfs4_deleg_state_walk_callout, exi);
4134 rfs4_dbe_walk(nsrv4->rfs4_file_tab, rfs4_file_walk_callout, exi);
4135
4136 mutex_exit(&nsrv4->state_lock);
4137 }