1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21
22 /*
23 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
24 */
25
26 /*
27 * Copyright 2018 Nexenta Systems, Inc.
28 * Copyright 2019 Nexenta by DDN, Inc.
29 */
30
31 #include <sys/systm.h>
32 #include <sys/kmem.h>
33 #include <sys/cmn_err.h>
34 #include <sys/atomic.h>
35 #include <sys/clconf.h>
36 #include <sys/cladm.h>
37 #include <sys/flock.h>
38 #include <nfs/export.h>
39 #include <nfs/nfs.h>
40 #include <nfs/nfs4.h>
41 #include <nfs/nfssys.h>
42 #include <nfs/lm.h>
43 #include <sys/pathname.h>
44 #include <sys/sdt.h>
45 #include <sys/nvpair.h>
46
47 extern u_longlong_t nfs4_srv_caller_id;
48
49 extern uint_t nfs4_srv_vkey;
50
51 stateid4 special0 = {
52 0,
53 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 }
54 };
55
56 stateid4 special1 = {
57 0xffffffff,
58 {
59 (char)0xff, (char)0xff, (char)0xff, (char)0xff,
60 (char)0xff, (char)0xff, (char)0xff, (char)0xff,
61 (char)0xff, (char)0xff, (char)0xff, (char)0xff
62 }
63 };
64
65
66 #define ISSPECIAL(id) (stateid4_cmp(id, &special0) || \
67 stateid4_cmp(id, &special1))
68
69 /* For embedding the cluster nodeid into our clientid */
70 #define CLUSTER_NODEID_SHIFT 24
71 #define CLUSTER_MAX_NODEID 255
72
73 #ifdef DEBUG
74 int rfs4_debug;
75 #endif
76
77 static uint32_t rfs4_database_debug = 0x00;
78
79 /* CSTYLED */
80 static void rfs4_ss_clid_write(nfs4_srv_t *nsrv4, rfs4_client_t *cp, char *leaf);
81 static void rfs4_ss_clid_write_one(rfs4_client_t *cp, char *dir, char *leaf);
82 static void rfs4_dss_clear_oldstate(rfs4_servinst_t *sip);
83 static void rfs4_ss_chkclid_sip(rfs4_client_t *cp, rfs4_servinst_t *sip);
84
85 /*
86 * Couple of simple init/destroy functions for a general waiter
87 */
88 void
89 rfs4_sw_init(rfs4_state_wait_t *swp)
90 {
91 mutex_init(swp->sw_cv_lock, NULL, MUTEX_DEFAULT, NULL);
92 cv_init(swp->sw_cv, NULL, CV_DEFAULT, NULL);
93 swp->sw_active = FALSE;
94 swp->sw_wait_count = 0;
95 }
96
97 void
98 rfs4_sw_destroy(rfs4_state_wait_t *swp)
99 {
100 mutex_destroy(swp->sw_cv_lock);
101 cv_destroy(swp->sw_cv);
102 }
103
104 void
105 rfs4_sw_enter(rfs4_state_wait_t *swp)
106 {
107 mutex_enter(swp->sw_cv_lock);
108 while (swp->sw_active) {
109 swp->sw_wait_count++;
110 cv_wait(swp->sw_cv, swp->sw_cv_lock);
111 swp->sw_wait_count--;
112 }
113 ASSERT(swp->sw_active == FALSE);
114 swp->sw_active = TRUE;
115 mutex_exit(swp->sw_cv_lock);
116 }
117
118 void
119 rfs4_sw_exit(rfs4_state_wait_t *swp)
120 {
121 mutex_enter(swp->sw_cv_lock);
122 ASSERT(swp->sw_active == TRUE);
123 swp->sw_active = FALSE;
124 if (swp->sw_wait_count != 0)
125 cv_broadcast(swp->sw_cv);
126 mutex_exit(swp->sw_cv_lock);
127 }
128
129 static void
130 deep_lock_copy(LOCK4res *dres, LOCK4res *sres)
131 {
132 lock_owner4 *slo = &sres->LOCK4res_u.denied.owner;
133 lock_owner4 *dlo = &dres->LOCK4res_u.denied.owner;
134
135 if (sres->status == NFS4ERR_DENIED) {
136 dlo->owner_val = kmem_alloc(slo->owner_len, KM_SLEEP);
137 bcopy(slo->owner_val, dlo->owner_val, slo->owner_len);
138 }
139 }
140
141 /*
142 * CPR callback id -- not related to v4 callbacks
143 */
144 static callb_id_t cpr_id = 0;
145
146 static void
147 deep_lock_free(LOCK4res *res)
148 {
149 lock_owner4 *lo = &res->LOCK4res_u.denied.owner;
150
151 if (res->status == NFS4ERR_DENIED)
152 kmem_free(lo->owner_val, lo->owner_len);
153 }
154
155 static void
156 deep_open_copy(OPEN4res *dres, OPEN4res *sres)
157 {
158 nfsace4 *sacep, *dacep;
159
160 if (sres->status != NFS4_OK) {
161 return;
162 }
163
164 dres->attrset = sres->attrset;
165
166 switch (sres->delegation.delegation_type) {
167 case OPEN_DELEGATE_NONE:
168 return;
169 case OPEN_DELEGATE_READ:
170 sacep = &sres->delegation.open_delegation4_u.read.permissions;
171 dacep = &dres->delegation.open_delegation4_u.read.permissions;
172 break;
173 case OPEN_DELEGATE_WRITE:
174 sacep = &sres->delegation.open_delegation4_u.write.permissions;
175 dacep = &dres->delegation.open_delegation4_u.write.permissions;
176 break;
177 }
178 dacep->who.utf8string_val =
179 kmem_alloc(sacep->who.utf8string_len, KM_SLEEP);
180 bcopy(sacep->who.utf8string_val, dacep->who.utf8string_val,
181 sacep->who.utf8string_len);
182 }
183
184 static void
185 deep_open_free(OPEN4res *res)
186 {
187 nfsace4 *acep;
188 if (res->status != NFS4_OK)
189 return;
190
191 switch (res->delegation.delegation_type) {
192 case OPEN_DELEGATE_NONE:
193 return;
194 case OPEN_DELEGATE_READ:
195 acep = &res->delegation.open_delegation4_u.read.permissions;
196 break;
197 case OPEN_DELEGATE_WRITE:
198 acep = &res->delegation.open_delegation4_u.write.permissions;
199 break;
200 }
201
202 if (acep->who.utf8string_val) {
203 kmem_free(acep->who.utf8string_val, acep->who.utf8string_len);
204 acep->who.utf8string_val = NULL;
205 }
206 }
207
208 void
209 rfs4_free_reply(nfs_resop4 *rp)
210 {
211 switch (rp->resop) {
212 case OP_LOCK:
213 deep_lock_free(&rp->nfs_resop4_u.oplock);
214 break;
215 case OP_OPEN:
216 deep_open_free(&rp->nfs_resop4_u.opopen);
217 default:
218 break;
219 }
220 }
221
222 void
223 rfs4_copy_reply(nfs_resop4 *dst, nfs_resop4 *src)
224 {
225 *dst = *src;
226
227 /* Handle responses that need deep copy */
228 switch (src->resop) {
229 case OP_LOCK:
230 deep_lock_copy(&dst->nfs_resop4_u.oplock,
231 &src->nfs_resop4_u.oplock);
232 break;
233 case OP_OPEN:
234 deep_open_copy(&dst->nfs_resop4_u.opopen,
235 &src->nfs_resop4_u.opopen);
236 break;
237 default:
238 break;
239 };
240 }
241
242 /*
243 * This is the implementation of the underlying state engine. The
244 * public interface to this engine is described by
245 * nfs4_state.h. Callers to the engine should hold no state engine
246 * locks when they call in to it. If the protocol needs to lock data
247 * structures it should do so after acquiring all references to them
248 * first and then follow the following lock order:
249 *
250 * client > openowner > state > lo_state > lockowner > file.
251 *
252 * Internally we only allow a thread to hold one hash bucket lock at a
253 * time and the lock is higher in the lock order (must be acquired
254 * first) than the data structure that is on that hash list.
255 *
256 * If a new reference was acquired by the caller, that reference needs
257 * to be released after releasing all acquired locks with the
258 * corresponding rfs4_*_rele routine.
259 */
260
261 /*
262 * This code is some what prototypical for now. Its purpose currently is to
263 * implement the interfaces sufficiently to finish the higher protocol
264 * elements. This will be replaced by a dynamically resizeable tables
265 * backed by kmem_cache allocator. However synchronization is handled
266 * correctly (I hope) and will not change by much. The mutexes for
267 * the hash buckets that can be used to create new instances of data
268 * structures might be good candidates to evolve into reader writer
269 * locks. If it has to do a creation, it would be holding the
270 * mutex across a kmem_alloc with KM_SLEEP specified.
271 */
272
273 #ifdef DEBUG
274 #define TABSIZE 17
275 #else
276 #define TABSIZE 2047
277 #endif
278
279 #define ADDRHASH(key) ((unsigned long)(key) >> 3)
280
281 #define MAXTABSZ 1024*1024
282
283 /* The values below are rfs4_lease_time units */
284
285 #ifdef DEBUG
286 #define CLIENT_CACHE_TIME 1
287 #define OPENOWNER_CACHE_TIME 1
288 #define STATE_CACHE_TIME 1
289 #define LO_STATE_CACHE_TIME 1
290 #define LOCKOWNER_CACHE_TIME 1
291 #define FILE_CACHE_TIME 3
292 #define DELEG_STATE_CACHE_TIME 1
293 #else
294 #define CLIENT_CACHE_TIME 10
295 #define OPENOWNER_CACHE_TIME 5
296 #define STATE_CACHE_TIME 1
297 #define LO_STATE_CACHE_TIME 1
298 #define LOCKOWNER_CACHE_TIME 3
299 #define FILE_CACHE_TIME 40
300 #define DELEG_STATE_CACHE_TIME 1
301 #endif
302
303 /*
304 * NFSv4 server state databases
305 *
306 * Initilized when the module is loaded and used by NFSv4 state tables.
307 * These kmem_cache databases are global, the tables that make use of these
308 * are per zone.
309 */
310 kmem_cache_t *rfs4_client_mem_cache;
311 kmem_cache_t *rfs4_clntIP_mem_cache;
312 kmem_cache_t *rfs4_openown_mem_cache;
313 kmem_cache_t *rfs4_openstID_mem_cache;
314 kmem_cache_t *rfs4_lockstID_mem_cache;
315 kmem_cache_t *rfs4_lockown_mem_cache;
316 kmem_cache_t *rfs4_file_mem_cache;
317 kmem_cache_t *rfs4_delegstID_mem_cache;
318
319 /*
320 * NFSv4 state table functions
321 */
322 static bool_t rfs4_client_create(rfs4_entry_t, void *);
323 static void rfs4_dss_remove_cpleaf(rfs4_client_t *);
324 static void rfs4_dss_remove_leaf(rfs4_servinst_t *, char *, char *);
325 static void rfs4_client_destroy(rfs4_entry_t);
326 static bool_t rfs4_client_expiry(rfs4_entry_t);
327 static uint32_t clientid_hash(void *);
328 static bool_t clientid_compare(rfs4_entry_t, void *);
329 static void *clientid_mkkey(rfs4_entry_t);
330 static uint32_t nfsclnt_hash(void *);
331 static bool_t nfsclnt_compare(rfs4_entry_t, void *);
332 static void *nfsclnt_mkkey(rfs4_entry_t);
333 static bool_t rfs4_clntip_expiry(rfs4_entry_t);
334 static void rfs4_clntip_destroy(rfs4_entry_t);
335 static bool_t rfs4_clntip_create(rfs4_entry_t, void *);
336 static uint32_t clntip_hash(void *);
337 static bool_t clntip_compare(rfs4_entry_t, void *);
338 static void *clntip_mkkey(rfs4_entry_t);
339 static bool_t rfs4_openowner_create(rfs4_entry_t, void *);
340 static void rfs4_openowner_destroy(rfs4_entry_t);
341 static bool_t rfs4_openowner_expiry(rfs4_entry_t);
342 static uint32_t openowner_hash(void *);
343 static bool_t openowner_compare(rfs4_entry_t, void *);
344 static void *openowner_mkkey(rfs4_entry_t);
345 static bool_t rfs4_state_create(rfs4_entry_t, void *);
346 static void rfs4_state_destroy(rfs4_entry_t);
347 static bool_t rfs4_state_expiry(rfs4_entry_t);
348 static uint32_t state_hash(void *);
349 static bool_t state_compare(rfs4_entry_t, void *);
350 static void *state_mkkey(rfs4_entry_t);
351 static uint32_t state_owner_file_hash(void *);
352 static bool_t state_owner_file_compare(rfs4_entry_t, void *);
353 static void *state_owner_file_mkkey(rfs4_entry_t);
354 static uint32_t state_file_hash(void *);
355 static bool_t state_file_compare(rfs4_entry_t, void *);
356 static void *state_file_mkkey(rfs4_entry_t);
357 static bool_t rfs4_lo_state_create(rfs4_entry_t, void *);
358 static void rfs4_lo_state_destroy(rfs4_entry_t);
359 static bool_t rfs4_lo_state_expiry(rfs4_entry_t);
360 static uint32_t lo_state_hash(void *);
361 static bool_t lo_state_compare(rfs4_entry_t, void *);
362 static void *lo_state_mkkey(rfs4_entry_t);
363 static uint32_t lo_state_lo_hash(void *);
364 static bool_t lo_state_lo_compare(rfs4_entry_t, void *);
365 static void *lo_state_lo_mkkey(rfs4_entry_t);
366 static bool_t rfs4_lockowner_create(rfs4_entry_t, void *);
367 static void rfs4_lockowner_destroy(rfs4_entry_t);
368 static bool_t rfs4_lockowner_expiry(rfs4_entry_t);
369 static uint32_t lockowner_hash(void *);
370 static bool_t lockowner_compare(rfs4_entry_t, void *);
371 static void *lockowner_mkkey(rfs4_entry_t);
372 static uint32_t pid_hash(void *);
373 static bool_t pid_compare(rfs4_entry_t, void *);
374 static void *pid_mkkey(rfs4_entry_t);
375 static bool_t rfs4_file_create(rfs4_entry_t, void *);
376 static void rfs4_file_destroy(rfs4_entry_t);
377 static uint32_t file_hash(void *);
378 static bool_t file_compare(rfs4_entry_t, void *);
379 static void *file_mkkey(rfs4_entry_t);
380 static bool_t rfs4_deleg_state_create(rfs4_entry_t, void *);
381 static void rfs4_deleg_state_destroy(rfs4_entry_t);
382 static bool_t rfs4_deleg_state_expiry(rfs4_entry_t);
383 static uint32_t deleg_hash(void *);
384 static bool_t deleg_compare(rfs4_entry_t, void *);
385 static void *deleg_mkkey(rfs4_entry_t);
386 static uint32_t deleg_state_hash(void *);
387 static bool_t deleg_state_compare(rfs4_entry_t, void *);
388 static void *deleg_state_mkkey(rfs4_entry_t);
389
390 static void rfs4_state_rele_nounlock(rfs4_state_t *);
391
392 static int rfs4_ss_enabled = 0;
393
394 extern void (*rfs4_client_clrst)(struct nfs4clrst_args *);
395
396 void
397 rfs4_ss_pnfree(rfs4_ss_pn_t *ss_pn)
398 {
399 kmem_free(ss_pn, sizeof (rfs4_ss_pn_t));
400 }
401
402 static rfs4_ss_pn_t *
403 rfs4_ss_pnalloc(char *dir, char *leaf)
404 {
405 rfs4_ss_pn_t *ss_pn;
406 int dir_len, leaf_len;
407
408 /*
409 * validate we have a resonable path
410 * (account for the '/' and trailing null)
411 */
412 if ((dir_len = strlen(dir)) > MAXPATHLEN ||
413 (leaf_len = strlen(leaf)) > MAXNAMELEN ||
414 (dir_len + leaf_len + 2) > MAXPATHLEN) {
415 return (NULL);
416 }
417
418 ss_pn = kmem_alloc(sizeof (rfs4_ss_pn_t), KM_SLEEP);
419
420 (void) snprintf(ss_pn->pn, MAXPATHLEN, "%s/%s", dir, leaf);
421 /* Handy pointer to just the leaf name */
422 ss_pn->leaf = ss_pn->pn + dir_len + 1;
423 return (ss_pn);
424 }
425
426
427 /*
428 * Move the "leaf" filename from "sdir" directory
429 * to the "ddir" directory. Return the pathname of
430 * the destination unless the rename fails in which
431 * case we need to return the source pathname.
432 */
433 static rfs4_ss_pn_t *
434 rfs4_ss_movestate(char *sdir, char *ddir, char *leaf)
435 {
436 rfs4_ss_pn_t *src, *dst;
437
438 if ((src = rfs4_ss_pnalloc(sdir, leaf)) == NULL)
439 return (NULL);
440
441 if ((dst = rfs4_ss_pnalloc(ddir, leaf)) == NULL) {
442 rfs4_ss_pnfree(src);
443 return (NULL);
444 }
445
446 /*
447 * If the rename fails we shall return the src
448 * pathname and free the dst. Otherwise we need
449 * to free the src and return the dst pathanme.
450 */
451 if (vn_rename(src->pn, dst->pn, UIO_SYSSPACE)) {
452 rfs4_ss_pnfree(dst);
453 return (src);
454 }
455 rfs4_ss_pnfree(src);
456 return (dst);
457 }
458
459
460 static rfs4_oldstate_t *
461 rfs4_ss_getstate(vnode_t *dvp, rfs4_ss_pn_t *ss_pn)
462 {
463 struct uio uio;
464 struct iovec iov[3];
465
466 rfs4_oldstate_t *cl_ss = NULL;
467 vnode_t *vp;
468 vattr_t va;
469 uint_t id_len;
470 int err, kill_file, file_vers;
471
472 if (ss_pn == NULL)
473 return (NULL);
474
475 /*
476 * open the state file.
477 */
478 if (vn_open(ss_pn->pn, UIO_SYSSPACE, FREAD, 0, &vp, 0, 0) != 0) {
479 return (NULL);
480 }
481
482 if (vp->v_type != VREG) {
483 (void) VOP_CLOSE(vp, FREAD, 1, (offset_t)0, CRED(), NULL);
484 VN_RELE(vp);
485 return (NULL);
486 }
487
488 err = VOP_ACCESS(vp, VREAD, 0, CRED(), NULL);
489 if (err) {
490 /*
491 * We don't have read access? better get the heck out.
492 */
493 (void) VOP_CLOSE(vp, FREAD, 1, (offset_t)0, CRED(), NULL);
494 VN_RELE(vp);
495 return (NULL);
496 }
497
498 (void) VOP_RWLOCK(vp, V_WRITELOCK_FALSE, NULL);
499 /*
500 * get the file size to do some basic validation
501 */
502 va.va_mask = AT_SIZE;
503 err = VOP_GETATTR(vp, &va, 0, CRED(), NULL);
504
505 kill_file = (va.va_size == 0 || va.va_size <
506 (NFS4_VERIFIER_SIZE + sizeof (uint_t)+1));
507
508 if (err || kill_file) {
509 VOP_RWUNLOCK(vp, V_WRITELOCK_FALSE, NULL);
510 (void) VOP_CLOSE(vp, FREAD, 1, (offset_t)0, CRED(), NULL);
511 VN_RELE(vp);
512 if (kill_file) {
513 (void) VOP_REMOVE(dvp, ss_pn->leaf, CRED(), NULL, 0);
514 }
515 return (NULL);
516 }
517
518 cl_ss = kmem_alloc(sizeof (rfs4_oldstate_t), KM_SLEEP);
519
520 /*
521 * build iovecs to read in the file_version, verifier and id_len
522 */
523 iov[0].iov_base = (caddr_t)&file_vers;
524 iov[0].iov_len = sizeof (int);
525 iov[1].iov_base = (caddr_t)&cl_ss->cl_id4.verifier;
526 iov[1].iov_len = NFS4_VERIFIER_SIZE;
527 iov[2].iov_base = (caddr_t)&id_len;
528 iov[2].iov_len = sizeof (uint_t);
529
530 uio.uio_iov = iov;
531 uio.uio_iovcnt = 3;
532 uio.uio_segflg = UIO_SYSSPACE;
533 uio.uio_loffset = 0;
534 uio.uio_resid = sizeof (int) + NFS4_VERIFIER_SIZE + sizeof (uint_t);
535
536 if (err = VOP_READ(vp, &uio, FREAD, CRED(), NULL)) {
537 VOP_RWUNLOCK(vp, V_WRITELOCK_FALSE, NULL);
538 (void) VOP_CLOSE(vp, FREAD, 1, (offset_t)0, CRED(), NULL);
539 VN_RELE(vp);
540 kmem_free(cl_ss, sizeof (rfs4_oldstate_t));
541 return (NULL);
542 }
543
544 /*
545 * if the file_version doesn't match or if the
546 * id_len is zero or the combination of the verifier,
547 * id_len and id_val is bigger than the file we have
548 * a problem. If so ditch the file.
549 */
550 kill_file = (file_vers != NFS4_SS_VERSION || id_len == 0 ||
551 (id_len + NFS4_VERIFIER_SIZE + sizeof (uint_t)) > va.va_size);
552
553 if (err || kill_file) {
554 VOP_RWUNLOCK(vp, V_WRITELOCK_FALSE, NULL);
555 (void) VOP_CLOSE(vp, FREAD, 1, (offset_t)0, CRED(), NULL);
556 VN_RELE(vp);
557 kmem_free(cl_ss, sizeof (rfs4_oldstate_t));
558 if (kill_file) {
559 (void) VOP_REMOVE(dvp, ss_pn->leaf, CRED(), NULL, 0);
560 }
561 return (NULL);
562 }
563
564 /*
565 * now get the client id value
566 */
567 cl_ss->cl_id4.id_val = kmem_alloc(id_len, KM_SLEEP);
568 iov[0].iov_base = cl_ss->cl_id4.id_val;
569 iov[0].iov_len = id_len;
570
571 uio.uio_iov = iov;
572 uio.uio_iovcnt = 1;
573 uio.uio_segflg = UIO_SYSSPACE;
574 uio.uio_resid = cl_ss->cl_id4.id_len = id_len;
575
576 if (err = VOP_READ(vp, &uio, FREAD, CRED(), NULL)) {
577 VOP_RWUNLOCK(vp, V_WRITELOCK_FALSE, NULL);
578 (void) VOP_CLOSE(vp, FREAD, 1, (offset_t)0, CRED(), NULL);
579 VN_RELE(vp);
580 kmem_free(cl_ss->cl_id4.id_val, id_len);
581 kmem_free(cl_ss, sizeof (rfs4_oldstate_t));
582 return (NULL);
583 }
584
585 VOP_RWUNLOCK(vp, V_WRITELOCK_FALSE, NULL);
586 (void) VOP_CLOSE(vp, FREAD, 1, (offset_t)0, CRED(), NULL);
587 VN_RELE(vp);
588 return (cl_ss);
589 }
590
591 #ifdef nextdp
592 #undef nextdp
593 #endif
594 #define nextdp(dp) ((struct dirent64 *)((char *)(dp) + (dp)->d_reclen))
595
596 /*
597 * Add entries from statedir to supplied oldstate list.
598 * Optionally, move all entries from statedir -> destdir.
599 */
600 void
601 rfs4_ss_oldstate(rfs4_oldstate_t *oldstate, char *statedir, char *destdir)
602 {
603 rfs4_ss_pn_t *ss_pn;
604 rfs4_oldstate_t *cl_ss = NULL;
605 char *dirt = NULL;
606 int err, dir_eof = 0, size = 0;
607 vnode_t *dvp;
608 struct iovec iov;
609 struct uio uio;
610 struct dirent64 *dep;
611 offset_t dirchunk_offset = 0;
612
613 /*
614 * open the state directory
615 */
616 if (vn_open(statedir, UIO_SYSSPACE, FREAD, 0, &dvp, 0, 0))
617 return;
618
619 if (dvp->v_type != VDIR || VOP_ACCESS(dvp, VREAD, 0, CRED(), NULL))
620 goto out;
621
622 dirt = kmem_alloc(RFS4_SS_DIRSIZE, KM_SLEEP);
623
624 /*
625 * Get and process the directory entries
626 */
627 while (!dir_eof) {
628 (void) VOP_RWLOCK(dvp, V_WRITELOCK_FALSE, NULL);
629 iov.iov_base = dirt;
630 iov.iov_len = RFS4_SS_DIRSIZE;
631 uio.uio_iov = &iov;
632 uio.uio_iovcnt = 1;
633 uio.uio_segflg = UIO_SYSSPACE;
634 uio.uio_loffset = dirchunk_offset;
635 uio.uio_resid = RFS4_SS_DIRSIZE;
636
637 err = VOP_READDIR(dvp, &uio, CRED(), &dir_eof, NULL, 0);
638 VOP_RWUNLOCK(dvp, V_WRITELOCK_FALSE, NULL);
639 if (err)
640 goto out;
641
642 size = RFS4_SS_DIRSIZE - uio.uio_resid;
643
644 /*
645 * Process all the directory entries in this
646 * readdir chunk
647 */
648 for (dep = (struct dirent64 *)dirt; size > 0;
649 dep = nextdp(dep)) {
650
651 size -= dep->d_reclen;
652 dirchunk_offset = dep->d_off;
653
654 /*
655 * Skip '.' and '..'
656 */
657 if (NFS_IS_DOTNAME(dep->d_name))
658 continue;
659
660 ss_pn = rfs4_ss_pnalloc(statedir, dep->d_name);
661 if (ss_pn == NULL)
662 continue;
663
664 if (cl_ss = rfs4_ss_getstate(dvp, ss_pn)) {
665 if (destdir != NULL) {
666 rfs4_ss_pnfree(ss_pn);
667 cl_ss->ss_pn = rfs4_ss_movestate(
668 statedir, destdir, dep->d_name);
669 } else {
670 cl_ss->ss_pn = ss_pn;
671 }
672 insque(cl_ss, oldstate);
673 } else {
674 rfs4_ss_pnfree(ss_pn);
675 }
676 }
677 }
678
679 out:
680 (void) VOP_CLOSE(dvp, FREAD, 1, (offset_t)0, CRED(), NULL);
681 VN_RELE(dvp);
682 if (dirt)
683 kmem_free((caddr_t)dirt, RFS4_SS_DIRSIZE);
684 }
685
686 static void
687 rfs4_ss_init(nfs4_srv_t *nsrv4)
688 {
689 int npaths = 1;
690 char *default_dss_path = NFS4_DSS_VAR_DIR;
691
692 /* read the default stable storage state */
693 rfs4_dss_readstate(nsrv4, npaths, &default_dss_path);
694
695 rfs4_ss_enabled = 1;
696 }
697
698 static void
699 rfs4_ss_fini(nfs4_srv_t *nsrv4)
700 {
701 rfs4_servinst_t *sip;
702
703 mutex_enter(&nsrv4->servinst_lock);
704 sip = nsrv4->nfs4_cur_servinst;
705 while (sip != NULL) {
706 rfs4_dss_clear_oldstate(sip);
707 sip = sip->next;
708 }
709 mutex_exit(&nsrv4->servinst_lock);
710 }
711
712 /*
713 * Remove all oldstate files referenced by this servinst.
714 */
715 static void
716 rfs4_dss_clear_oldstate(rfs4_servinst_t *sip)
717 {
718 rfs4_oldstate_t *os_head, *osp;
719
720 rw_enter(&sip->oldstate_lock, RW_WRITER);
721 os_head = sip->oldstate;
722
723 if (os_head == NULL) {
724 rw_exit(&sip->oldstate_lock);
725 return;
726 }
727
728 /* skip dummy entry */
729 osp = os_head->next;
730 while (osp != os_head) {
731 char *leaf = osp->ss_pn->leaf;
732 rfs4_oldstate_t *os_next;
733
734 rfs4_dss_remove_leaf(sip, NFS4_DSS_OLDSTATE_LEAF, leaf);
735
736 if (osp->cl_id4.id_val)
737 kmem_free(osp->cl_id4.id_val, osp->cl_id4.id_len);
738 rfs4_ss_pnfree(osp->ss_pn);
739
740 os_next = osp->next;
741 remque(osp);
742 kmem_free(osp, sizeof (rfs4_oldstate_t));
743 osp = os_next;
744 }
745
746 rw_exit(&sip->oldstate_lock);
747 }
748
749 /*
750 * Form the state and oldstate paths, and read in the stable storage files.
751 */
752 void
753 rfs4_dss_readstate(nfs4_srv_t *nsrv4, int npaths, char **paths)
754 {
755 int i;
756 char *state, *oldstate;
757
758 state = kmem_alloc(MAXPATHLEN, KM_SLEEP);
759 oldstate = kmem_alloc(MAXPATHLEN, KM_SLEEP);
760
761 for (i = 0; i < npaths; i++) {
762 char *path = paths[i];
763
764 (void) sprintf(state, "%s/%s", path, NFS4_DSS_STATE_LEAF);
765 (void) sprintf(oldstate, "%s/%s", path, NFS4_DSS_OLDSTATE_LEAF);
766
767 /*
768 * Populate the current server instance's oldstate list.
769 *
770 * 1. Read stable storage data from old state directory,
771 * leaving its contents alone.
772 *
773 * 2. Read stable storage data from state directory,
774 * and move the latter's contents to old state
775 * directory.
776 */
777 /* CSTYLED */
778 rfs4_ss_oldstate(nsrv4->nfs4_cur_servinst->oldstate, oldstate, NULL);
779 /* CSTYLED */
780 rfs4_ss_oldstate(nsrv4->nfs4_cur_servinst->oldstate, state, oldstate);
781 }
782
783 kmem_free(state, MAXPATHLEN);
784 kmem_free(oldstate, MAXPATHLEN);
785 }
786
787
788 /*
789 * Check if we are still in grace and if the client can be
790 * granted permission to perform reclaims.
791 */
792 void
793 rfs4_ss_chkclid(nfs4_srv_t *nsrv4, rfs4_client_t *cp)
794 {
795 rfs4_servinst_t *sip;
796
797 /*
798 * It should be sufficient to check the oldstate data for just
799 * this client's instance. However, since our per-instance
800 * client grouping is solely temporal, HA-NFSv4 RG failover
801 * might result in clients of the same RG being partitioned into
802 * separate instances.
803 *
804 * Until the client grouping is improved, we must check the
805 * oldstate data for all instances with an active grace period.
806 *
807 * This also serves as the mechanism to remove stale oldstate data.
808 * The first time we check an instance after its grace period has
809 * expired, the oldstate data should be cleared.
810 *
811 * Start at the current instance, and walk the list backwards
812 * to the first.
813 */
814 mutex_enter(&nsrv4->servinst_lock);
815 for (sip = nsrv4->nfs4_cur_servinst; sip != NULL; sip = sip->prev) {
816 rfs4_ss_chkclid_sip(cp, sip);
817
818 /* if the above check found this client, we're done */
819 if (cp->rc_can_reclaim)
820 break;
821 }
822 mutex_exit(&nsrv4->servinst_lock);
823 }
824
825 static void
826 rfs4_ss_chkclid_sip(rfs4_client_t *cp, rfs4_servinst_t *sip)
827 {
828 rfs4_oldstate_t *osp, *os_head;
829
830 /* short circuit everything if this server instance has no oldstate */
831 rw_enter(&sip->oldstate_lock, RW_READER);
832 os_head = sip->oldstate;
833 rw_exit(&sip->oldstate_lock);
834 if (os_head == NULL)
835 return;
836
837 /*
838 * If this server instance is no longer in a grace period then
839 * the client won't be able to reclaim. No further need for this
840 * instance's oldstate data, so it can be cleared.
841 */
842 if (!rfs4_servinst_in_grace(sip))
843 return;
844
845 /* this instance is still in grace; search for the clientid */
846
847 rw_enter(&sip->oldstate_lock, RW_READER);
848
849 os_head = sip->oldstate;
850 /* skip dummy entry */
851 osp = os_head->next;
852 while (osp != os_head) {
853 if (osp->cl_id4.id_len == cp->rc_nfs_client.id_len) {
854 if (bcmp(osp->cl_id4.id_val, cp->rc_nfs_client.id_val,
855 osp->cl_id4.id_len) == 0) {
856 cp->rc_can_reclaim = 1;
857 break;
858 }
859 }
860 osp = osp->next;
861 }
862
863 rw_exit(&sip->oldstate_lock);
864 }
865
866 /*
867 * Place client information into stable storage: 1/3.
868 * First, generate the leaf filename, from the client's IP address and
869 * the server-generated short-hand clientid.
870 */
871 void
872 rfs4_ss_clid(nfs4_srv_t *nsrv4, rfs4_client_t *cp)
873 {
874 const char *kinet_ntop6(uchar_t *, char *, size_t);
875 char leaf[MAXNAMELEN], buf[INET6_ADDRSTRLEN];
876 struct sockaddr *ca;
877 uchar_t *b;
878
879 if (rfs4_ss_enabled == 0) {
880 return;
881 }
882
883 buf[0] = 0;
884
885 ca = (struct sockaddr *)&cp->rc_addr;
886
887 /*
888 * Convert the caller's IP address to a dotted string
889 */
890 if (ca->sa_family == AF_INET) {
891 b = (uchar_t *)&((struct sockaddr_in *)ca)->sin_addr;
892 (void) sprintf(buf, "%03d.%03d.%03d.%03d", b[0] & 0xFF,
893 b[1] & 0xFF, b[2] & 0xFF, b[3] & 0xFF);
894 } else if (ca->sa_family == AF_INET6) {
895 struct sockaddr_in6 *sin6;
896
897 sin6 = (struct sockaddr_in6 *)ca;
898 (void) kinet_ntop6((uchar_t *)&sin6->sin6_addr,
899 buf, INET6_ADDRSTRLEN);
900 }
901
902 (void) snprintf(leaf, MAXNAMELEN, "%s-%llx", buf,
903 (longlong_t)cp->rc_clientid);
904 rfs4_ss_clid_write(nsrv4, cp, leaf);
905 }
906
907 /*
908 * Place client information into stable storage: 2/3.
909 * DSS: distributed stable storage: the file may need to be written to
910 * multiple directories.
911 */
912 static void
913 rfs4_ss_clid_write(nfs4_srv_t *nsrv4, rfs4_client_t *cp, char *leaf)
914 {
915 rfs4_servinst_t *sip;
916
917 /*
918 * It should be sufficient to write the leaf file to (all) DSS paths
919 * associated with just this client's instance. However, since our
920 * per-instance client grouping is solely temporal, HA-NFSv4 RG
921 * failover might result in us losing DSS data.
922 *
923 * Until the client grouping is improved, we must write the DSS data
924 * to all instances' paths. Start at the current instance, and
925 * walk the list backwards to the first.
926 */
927 mutex_enter(&nsrv4->servinst_lock);
928 for (sip = nsrv4->nfs4_cur_servinst; sip != NULL; sip = sip->prev) {
929 int i, npaths = sip->dss_npaths;
930
931 /* write the leaf file to all DSS paths */
932 for (i = 0; i < npaths; i++) {
933 rfs4_dss_path_t *dss_path = sip->dss_paths[i];
934
935 /* HA-NFSv4 path might have been failed-away from us */
936 if (dss_path == NULL)
937 continue;
938
939 rfs4_ss_clid_write_one(cp, dss_path->path, leaf);
940 }
941 }
942 mutex_exit(&nsrv4->servinst_lock);
943 }
944
945 /*
946 * Place client information into stable storage: 3/3.
947 * Write the stable storage data to the requested file.
948 */
949 static void
950 rfs4_ss_clid_write_one(rfs4_client_t *cp, char *dss_path, char *leaf)
951 {
952 int ioflag;
953 int file_vers = NFS4_SS_VERSION;
954 size_t dirlen;
955 struct uio uio;
956 struct iovec iov[4];
957 char *dir;
958 rfs4_ss_pn_t *ss_pn;
959 vnode_t *vp;
960 nfs_client_id4 *cl_id4 = &(cp->rc_nfs_client);
961
962 /* allow 2 extra bytes for '/' & NUL */
963 dirlen = strlen(dss_path) + strlen(NFS4_DSS_STATE_LEAF) + 2;
964 dir = kmem_alloc(dirlen, KM_SLEEP);
965 (void) sprintf(dir, "%s/%s", dss_path, NFS4_DSS_STATE_LEAF);
966
967 ss_pn = rfs4_ss_pnalloc(dir, leaf);
968 /* rfs4_ss_pnalloc takes its own copy */
969 kmem_free(dir, dirlen);
970 if (ss_pn == NULL)
971 return;
972
973 if (vn_open(ss_pn->pn, UIO_SYSSPACE, FCREAT|FWRITE, 0600, &vp,
974 CRCREAT, 0)) {
975 rfs4_ss_pnfree(ss_pn);
976 return;
977 }
978
979 /*
980 * We need to record leaf - i.e. the filename - so that we know
981 * what to remove, in the future. However, the dir part of cp->ss_pn
982 * should never be referenced directly, since it's potentially only
983 * one of several paths with this leaf in it.
984 */
985 if (cp->rc_ss_pn != NULL) {
986 if (strcmp(cp->rc_ss_pn->leaf, leaf) == 0) {
987 /* we've already recorded *this* leaf */
988 rfs4_ss_pnfree(ss_pn);
989 } else {
990 /* replace with this leaf */
991 rfs4_ss_pnfree(cp->rc_ss_pn);
992 cp->rc_ss_pn = ss_pn;
993 }
994 } else {
995 cp->rc_ss_pn = ss_pn;
996 }
997
998 /*
999 * Build a scatter list that points to the nfs_client_id4
1000 */
1001 iov[0].iov_base = (caddr_t)&file_vers;
1002 iov[0].iov_len = sizeof (int);
1003 iov[1].iov_base = (caddr_t)&(cl_id4->verifier);
1004 iov[1].iov_len = NFS4_VERIFIER_SIZE;
1005 iov[2].iov_base = (caddr_t)&(cl_id4->id_len);
1006 iov[2].iov_len = sizeof (uint_t);
1007 iov[3].iov_base = (caddr_t)cl_id4->id_val;
1008 iov[3].iov_len = cl_id4->id_len;
1009
1010 uio.uio_iov = iov;
1011 uio.uio_iovcnt = 4;
1012 uio.uio_loffset = 0;
1013 uio.uio_segflg = UIO_SYSSPACE;
1014 uio.uio_llimit = (rlim64_t)MAXOFFSET_T;
1015 uio.uio_resid = cl_id4->id_len + sizeof (int) +
1016 NFS4_VERIFIER_SIZE + sizeof (uint_t);
1017
1018 ioflag = uio.uio_fmode = (FWRITE|FSYNC);
1019 uio.uio_extflg = UIO_COPY_DEFAULT;
1020
1021 (void) VOP_RWLOCK(vp, V_WRITELOCK_TRUE, NULL);
1022 /* write the full client id to the file. */
1023 (void) VOP_WRITE(vp, &uio, ioflag, CRED(), NULL);
1024 VOP_RWUNLOCK(vp, V_WRITELOCK_TRUE, NULL);
1025
1026 (void) VOP_CLOSE(vp, FWRITE, 1, (offset_t)0, CRED(), NULL);
1027 VN_RELE(vp);
1028 }
1029
1030 /*
1031 * DSS: distributed stable storage.
1032 * Unpack the list of paths passed by nfsd.
1033 * Use nvlist_alloc(9F) to manage the data.
1034 * The caller is responsible for allocating and freeing the buffer.
1035 */
1036 int
1037 rfs4_dss_setpaths(char *buf, size_t buflen)
1038 {
1039 int error;
1040
1041 /*
1042 * If this is a "warm start", i.e. we previously had DSS paths,
1043 * preserve the old paths.
1044 */
1045 if (rfs4_dss_paths != NULL) {
1046 /*
1047 * Before we lose the ptr, destroy the nvlist and pathnames
1048 * array from the warm start before this one.
1049 */
1050 nvlist_free(rfs4_dss_oldpaths);
1051 rfs4_dss_oldpaths = rfs4_dss_paths;
1052 }
1053
1054 /* unpack the buffer into a searchable nvlist */
1055 error = nvlist_unpack(buf, buflen, &rfs4_dss_paths, KM_SLEEP);
1056 if (error)
1057 return (error);
1058
1059 /*
1060 * Search the nvlist for the pathnames nvpair (which is the only nvpair
1061 * in the list, and record its location.
1062 */
1063 error = nvlist_lookup_string_array(rfs4_dss_paths, NFS4_DSS_NVPAIR_NAME,
1064 &rfs4_dss_newpaths, &rfs4_dss_numnewpaths);
1065 return (error);
1066 }
1067
1068 /*
1069 * Ultimately the nfssys() call NFS4_CLR_STATE endsup here
1070 * to find and mark the client for forced expire.
1071 */
1072 static void
1073 rfs4_client_scrub(rfs4_entry_t ent, void *arg)
1074 {
1075 rfs4_client_t *cp = (rfs4_client_t *)ent;
1076 struct nfs4clrst_args *clr = arg;
1077 struct sockaddr_in6 *ent_sin6;
1078 struct in6_addr clr_in6;
1079 struct sockaddr_in *ent_sin;
1080 struct in_addr clr_in;
1081
1082 if (clr->addr_type != cp->rc_addr.ss_family) {
1083 return;
1084 }
1085
1086 switch (clr->addr_type) {
1087
1088 case AF_INET6:
1089 /* copyin the address from user space */
1090 if (copyin(clr->ap, &clr_in6, sizeof (clr_in6))) {
1091 break;
1092 }
1093
1094 ent_sin6 = (struct sockaddr_in6 *)&cp->rc_addr;
1095
1096 /*
1097 * now compare, and if equivalent mark entry
1098 * for forced expiration
1099 */
1100 if (IN6_ARE_ADDR_EQUAL(&ent_sin6->sin6_addr, &clr_in6)) {
1101 cp->rc_forced_expire = 1;
1102 }
1103 break;
1104
1105 case AF_INET:
1106 /* copyin the address from user space */
1107 if (copyin(clr->ap, &clr_in, sizeof (clr_in))) {
1108 break;
1109 }
1110
1111 ent_sin = (struct sockaddr_in *)&cp->rc_addr;
1112
1113 /*
1114 * now compare, and if equivalent mark entry
1115 * for forced expiration
1116 */
1117 if (ent_sin->sin_addr.s_addr == clr_in.s_addr) {
1118 cp->rc_forced_expire = 1;
1119 }
1120 break;
1121
1122 default:
1123 /* force this assert to fail */
1124 ASSERT(clr->addr_type != clr->addr_type);
1125 }
1126 }
1127
1128 /*
1129 * This is called from nfssys() in order to clear server state
1130 * for the specified client IP Address.
1131 */
1132 void
1133 rfs4_clear_client_state(struct nfs4clrst_args *clr)
1134 {
1135 nfs4_srv_t *nsrv4;
1136 nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
1137 (void) rfs4_dbe_walk(nsrv4->rfs4_client_tab, rfs4_client_scrub, clr);
1138 }
1139
1140 /*
1141 * Used to initialize the NFSv4 server's state or database. All of
1142 * the tables are created and timers are set.
1143 */
1144 void
1145 rfs4_state_g_init()
1146 {
1147 extern boolean_t rfs4_cpr_callb(void *, int);
1148 /*
1149 * Add a CPR callback so that we can update client
1150 * access times to extend the lease after a suspend
1151 * and resume (using the same class as rpcmod/connmgr)
1152 */
1153 cpr_id = callb_add(rfs4_cpr_callb, 0, CB_CL_CPR_RPC, "rfs4");
1154
1155 /*
1156 * NFSv4 server state databases
1157 *
1158 * Initilized when the module is loaded and used by NFSv4 state tables.
1159 * These kmem_cache free pools are used globally, the NFSv4 state
1160 * tables which make use of these kmem_cache free pools are per zone.
1161 *
1162 * initialize the global kmem_cache free pools which will be used by
1163 * the NFSv4 state tables.
1164 */
1165 /* CSTYLED */
1166 rfs4_client_mem_cache = nfs4_init_mem_cache("Client_entry_cache", 2, sizeof (rfs4_client_t), 0);
1167 /* CSTYLED */
1168 rfs4_clntIP_mem_cache = nfs4_init_mem_cache("ClntIP_entry_cache", 1, sizeof (rfs4_clntip_t), 1);
1169 /* CSTYLED */
1170 rfs4_openown_mem_cache = nfs4_init_mem_cache("OpenOwner_entry_cache", 1, sizeof (rfs4_openowner_t), 2);
1171 /* CSTYLED */
1172 rfs4_openstID_mem_cache = nfs4_init_mem_cache("OpenStateID_entry_cache", 3, sizeof (rfs4_state_t), 3);
1173 /* CSTYLED */
1174 rfs4_lockstID_mem_cache = nfs4_init_mem_cache("LockStateID_entry_cache", 3, sizeof (rfs4_lo_state_t), 4);
1175 /* CSTYLED */
1176 rfs4_lockown_mem_cache = nfs4_init_mem_cache("Lockowner_entry_cache", 2, sizeof (rfs4_lockowner_t), 5);
1177 /* CSTYLED */
1178 rfs4_file_mem_cache = nfs4_init_mem_cache("File_entry_cache", 1, sizeof (rfs4_file_t), 6);
1179 /* CSTYLED */
1180 rfs4_delegstID_mem_cache = nfs4_init_mem_cache("DelegStateID_entry_cache", 2, sizeof (rfs4_deleg_state_t), 7);
1181
1182 rfs4_client_clrst = rfs4_clear_client_state;
1183 }
1184
1185
1186 /*
1187 * Used at server shutdown to cleanup all of the NFSv4 server's structures
1188 * and other state.
1189 */
1190 void
1191 rfs4_state_g_fini()
1192 {
1193 int i;
1194 /*
1195 * Cleanup the CPR callback.
1196 */
1197 if (cpr_id)
1198 (void) callb_delete(cpr_id);
1199
1200 rfs4_client_clrst = NULL;
1201
1202 /* free the NFSv4 state databases */
1203 for (i = 0; i < RFS4_DB_MEM_CACHE_NUM; i++) {
1204 kmem_cache_destroy(rfs4_db_mem_cache_table[i].r_db_mem_cache);
1205 rfs4_db_mem_cache_table[i].r_db_mem_cache = NULL;
1206 }
1207
1208 rfs4_client_mem_cache = NULL;
1209 rfs4_clntIP_mem_cache = NULL;
1210 rfs4_openown_mem_cache = NULL;
1211 rfs4_openstID_mem_cache = NULL;
1212 rfs4_lockstID_mem_cache = NULL;
1213 rfs4_lockown_mem_cache = NULL;
1214 rfs4_file_mem_cache = NULL;
1215 rfs4_delegstID_mem_cache = NULL;
1216
1217 /* DSS: distributed stable storage */
1218 nvlist_free(rfs4_dss_oldpaths);
1219 nvlist_free(rfs4_dss_paths);
1220 rfs4_dss_paths = rfs4_dss_oldpaths = NULL;
1221 }
1222
1223 /*
1224 * Used to initialize the per zone NFSv4 server's state
1225 */
1226 void
1227 rfs4_state_zone_init(nfs4_srv_t *nsrv4)
1228 {
1229 time_t start_time;
1230 int start_grace;
1231 char *dss_path = NFS4_DSS_VAR_DIR;
1232
1233 /* DSS: distributed stable storage: initialise served paths list */
1234 nsrv4->dss_pathlist = NULL;
1235
1236 /*
1237 * Set the boot time. If the server
1238 * has been restarted quickly and has had the opportunity to
1239 * service clients, then the start_time needs to be bumped
1240 * regardless. A small window but it exists...
1241 */
1242 start_time = gethrestime_sec();
1243 if (nsrv4->rfs4_start_time < start_time)
1244 nsrv4->rfs4_start_time = start_time;
1245 else
1246 nsrv4->rfs4_start_time++;
1247
1248 /*
1249 * Create the first server instance, or a new one if the server has
1250 * been restarted; see above comments on rfs4_start_time. Don't
1251 * start its grace period; that will be done later, to maximise the
1252 * clients' recovery window.
1253 */
1254 start_grace = 0;
1255 if (curzone == global_zone && rfs4_dss_numnewpaths > 0) {
1256 int i;
1257 char **dss_allpaths = NULL;
1258 dss_allpaths = kmem_alloc(sizeof (char *) * (rfs4_dss_numnewpaths + 1), KM_SLEEP);
1259 /*
1260 * Add the default path into the list of paths for saving
1261 * state informantion.
1262 */
1263 dss_allpaths[0] = dss_path;
1264 for ( i = 0; i < rfs4_dss_numnewpaths; i++) {
1265 dss_allpaths[i + 1] = rfs4_dss_newpaths[i];
1266 }
1267 rfs4_servinst_create(nsrv4, start_grace, (rfs4_dss_numnewpaths + 1), dss_allpaths);
1268 kmem_free(dss_allpaths, (sizeof (char *) * (rfs4_dss_numnewpaths + 1)));
1269 } else {
1270 rfs4_servinst_create(nsrv4, start_grace, 1, &dss_path);
1271 }
1272
1273 /* reset the "first NFSv4 request" status */
1274 nsrv4->seen_first_compound = 0;
1275
1276 mutex_enter(&nsrv4->state_lock);
1277
1278 /*
1279 * If the server state database has already been initialized,
1280 * skip it
1281 */
1282 if (nsrv4->nfs4_server_state != NULL) {
1283 mutex_exit(&nsrv4->state_lock);
1284 return;
1285 }
1286
1287 rw_init(&nsrv4->rfs4_findclient_lock, NULL, RW_DEFAULT, NULL);
1288
1289 /* set the various cache timers for table creation */
1290 if (nsrv4->rfs4_client_cache_time == 0)
1291 nsrv4->rfs4_client_cache_time = CLIENT_CACHE_TIME;
1292 if (nsrv4->rfs4_openowner_cache_time == 0)
1293 nsrv4->rfs4_openowner_cache_time = OPENOWNER_CACHE_TIME;
1294 if (nsrv4->rfs4_state_cache_time == 0)
1295 nsrv4->rfs4_state_cache_time = STATE_CACHE_TIME;
1296 if (nsrv4->rfs4_lo_state_cache_time == 0)
1297 nsrv4->rfs4_lo_state_cache_time = LO_STATE_CACHE_TIME;
1298 if (nsrv4->rfs4_lockowner_cache_time == 0)
1299 nsrv4->rfs4_lockowner_cache_time = LOCKOWNER_CACHE_TIME;
1300 if (nsrv4->rfs4_file_cache_time == 0)
1301 nsrv4->rfs4_file_cache_time = FILE_CACHE_TIME;
1302 if (nsrv4->rfs4_deleg_state_cache_time == 0)
1303 nsrv4->rfs4_deleg_state_cache_time = DELEG_STATE_CACHE_TIME;
1304
1305 /* Create the overall database to hold all server state */
1306 nsrv4->nfs4_server_state = rfs4_database_create(rfs4_database_debug);
1307
1308 /* Now create the individual tables */
1309 nsrv4->rfs4_client_cache_time *= rfs4_lease_time;
1310 nsrv4->rfs4_client_tab = rfs4_table_create(nsrv4->nfs4_server_state,
1311 "Client",
1312 nsrv4->rfs4_client_cache_time,
1313 2,
1314 rfs4_client_create,
1315 rfs4_client_destroy,
1316 rfs4_client_expiry,
1317 sizeof (rfs4_client_t),
1318 TABSIZE,
1319 MAXTABSZ/8, 100);
1320 nsrv4->rfs4_nfsclnt_idx = rfs4_index_create(nsrv4->rfs4_client_tab,
1321 "nfs_client_id4", nfsclnt_hash,
1322 nfsclnt_compare, nfsclnt_mkkey,
1323 TRUE);
1324 nsrv4->rfs4_clientid_idx = rfs4_index_create(nsrv4->rfs4_client_tab,
1325 "client_id", clientid_hash,
1326 clientid_compare, clientid_mkkey,
1327 FALSE);
1328
1329 nsrv4->rfs4_clntip_cache_time = 86400 * 365; /* about a year */
1330 nsrv4->rfs4_clntip_tab = rfs4_table_create(nsrv4->nfs4_server_state,
1331 "ClntIP",
1332 nsrv4->rfs4_clntip_cache_time,
1333 1,
1334 rfs4_clntip_create,
1335 rfs4_clntip_destroy,
1336 rfs4_clntip_expiry,
1337 sizeof (rfs4_clntip_t),
1338 TABSIZE,
1339 MAXTABSZ, 100);
1340 nsrv4->rfs4_clntip_idx = rfs4_index_create(nsrv4->rfs4_clntip_tab,
1341 "client_ip", clntip_hash,
1342 clntip_compare, clntip_mkkey,
1343 TRUE);
1344
1345 nsrv4->rfs4_openowner_cache_time *= rfs4_lease_time;
1346 nsrv4->rfs4_openowner_tab = rfs4_table_create(nsrv4->nfs4_server_state,
1347 "OpenOwner",
1348 nsrv4->rfs4_openowner_cache_time,
1349 1,
1350 rfs4_openowner_create,
1351 rfs4_openowner_destroy,
1352 rfs4_openowner_expiry,
1353 sizeof (rfs4_openowner_t),
1354 TABSIZE,
1355 MAXTABSZ, 100);
1356 nsrv4->rfs4_openowner_idx = rfs4_index_create(nsrv4->rfs4_openowner_tab,
1357 "open_owner4", openowner_hash,
1358 openowner_compare,
1359 openowner_mkkey, TRUE);
1360
1361 nsrv4->rfs4_state_cache_time *= rfs4_lease_time;
1362 nsrv4->rfs4_state_tab = rfs4_table_create(nsrv4->nfs4_server_state,
1363 "OpenStateID",
1364 nsrv4->rfs4_state_cache_time,
1365 3,
1366 rfs4_state_create,
1367 rfs4_state_destroy,
1368 rfs4_state_expiry,
1369 sizeof (rfs4_state_t),
1370 TABSIZE,
1371 MAXTABSZ, 100);
1372
1373 /* CSTYLED */
1374 nsrv4->rfs4_state_owner_file_idx = rfs4_index_create(nsrv4->rfs4_state_tab,
1375 "Openowner-File",
1376 state_owner_file_hash,
1377 state_owner_file_compare,
1378 state_owner_file_mkkey, TRUE);
1379
1380 nsrv4->rfs4_state_idx = rfs4_index_create(nsrv4->rfs4_state_tab,
1381 "State-id", state_hash,
1382 state_compare, state_mkkey, FALSE);
1383
1384 nsrv4->rfs4_state_file_idx = rfs4_index_create(nsrv4->rfs4_state_tab,
1385 "File", state_file_hash,
1386 state_file_compare, state_file_mkkey,
1387 FALSE);
1388
1389 nsrv4->rfs4_lo_state_cache_time *= rfs4_lease_time;
1390 nsrv4->rfs4_lo_state_tab = rfs4_table_create(nsrv4->nfs4_server_state,
1391 "LockStateID",
1392 nsrv4->rfs4_lo_state_cache_time,
1393 2,
1394 rfs4_lo_state_create,
1395 rfs4_lo_state_destroy,
1396 rfs4_lo_state_expiry,
1397 sizeof (rfs4_lo_state_t),
1398 TABSIZE,
1399 MAXTABSZ, 100);
1400
1401 /* CSTYLED */
1402 nsrv4->rfs4_lo_state_owner_idx = rfs4_index_create(nsrv4->rfs4_lo_state_tab,
1403 "lockownerxstate",
1404 lo_state_lo_hash,
1405 lo_state_lo_compare,
1406 lo_state_lo_mkkey, TRUE);
1407
1408 nsrv4->rfs4_lo_state_idx = rfs4_index_create(nsrv4->rfs4_lo_state_tab,
1409 "State-id",
1410 lo_state_hash, lo_state_compare,
1411 lo_state_mkkey, FALSE);
1412
1413 nsrv4->rfs4_lockowner_cache_time *= rfs4_lease_time;
1414
1415 nsrv4->rfs4_lockowner_tab = rfs4_table_create(nsrv4->nfs4_server_state,
1416 "Lockowner",
1417 nsrv4->rfs4_lockowner_cache_time,
1418 2,
1419 rfs4_lockowner_create,
1420 rfs4_lockowner_destroy,
1421 rfs4_lockowner_expiry,
1422 sizeof (rfs4_lockowner_t),
1423 TABSIZE,
1424 MAXTABSZ, 100);
1425
1426 nsrv4->rfs4_lockowner_idx = rfs4_index_create(nsrv4->rfs4_lockowner_tab,
1427 "lock_owner4", lockowner_hash,
1428 lockowner_compare,
1429 lockowner_mkkey, TRUE);
1430
1431 /* CSTYLED */
1432 nsrv4->rfs4_lockowner_pid_idx = rfs4_index_create(nsrv4->rfs4_lockowner_tab,
1433 "pid", pid_hash,
1434 pid_compare, pid_mkkey,
1435 FALSE);
1436
1437 nsrv4->rfs4_file_cache_time *= rfs4_lease_time;
1438 nsrv4->rfs4_file_tab = rfs4_table_create(nsrv4->nfs4_server_state,
1439 "File",
1440 nsrv4->rfs4_file_cache_time,
1441 1,
1442 rfs4_file_create,
1443 rfs4_file_destroy,
1444 NULL,
1445 sizeof (rfs4_file_t),
1446 TABSIZE,
1447 MAXTABSZ, -1);
1448
1449 nsrv4->rfs4_file_idx = rfs4_index_create(nsrv4->rfs4_file_tab,
1450 "Filehandle", file_hash,
1451 file_compare, file_mkkey, TRUE);
1452
1453 nsrv4->rfs4_deleg_state_cache_time *= rfs4_lease_time;
1454 /* CSTYLED */
1455 nsrv4->rfs4_deleg_state_tab = rfs4_table_create(nsrv4->nfs4_server_state,
1456 "DelegStateID",
1457 nsrv4->rfs4_deleg_state_cache_time,
1458 2,
1459 rfs4_deleg_state_create,
1460 rfs4_deleg_state_destroy,
1461 rfs4_deleg_state_expiry,
1462 sizeof (rfs4_deleg_state_t),
1463 TABSIZE,
1464 MAXTABSZ, 100);
1465 nsrv4->rfs4_deleg_idx = rfs4_index_create(nsrv4->rfs4_deleg_state_tab,
1466 "DelegByFileClient",
1467 deleg_hash,
1468 deleg_compare,
1469 deleg_mkkey, TRUE);
1470
1471 /* CSTYLED */
1472 nsrv4->rfs4_deleg_state_idx = rfs4_index_create(nsrv4->rfs4_deleg_state_tab,
1473 "DelegState",
1474 deleg_state_hash,
1475 deleg_state_compare,
1476 deleg_state_mkkey, FALSE);
1477
1478 mutex_exit(&nsrv4->state_lock);
1479
1480 /*
1481 * Init the stable storage.
1482 */
1483 rfs4_ss_init(nsrv4);
1484 }
1485
1486 /*
1487 * Used at server shutdown to cleanup all of NFSv4 server's zone structures
1488 * and state.
1489 */
1490 void
1491 rfs4_state_zone_fini()
1492 {
1493 rfs4_database_t *dbp;
1494 nfs4_srv_t *nsrv4;
1495 nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
1496
1497 rfs4_set_deleg_policy(nsrv4, SRV_NEVER_DELEGATE);
1498
1499 mutex_enter(&nsrv4->state_lock);
1500
1501 if (nsrv4->nfs4_server_state == NULL) {
1502 mutex_exit(&nsrv4->state_lock);
1503 return;
1504 }
1505
1506 /* destroy server instances and current instance ptr */
1507 rfs4_servinst_destroy_all(nsrv4);
1508
1509 /* reset the "first NFSv4 request" status */
1510 nsrv4->seen_first_compound = 0;
1511
1512 dbp = nsrv4->nfs4_server_state;
1513 nsrv4->nfs4_server_state = NULL;
1514
1515 rw_destroy(&nsrv4->rfs4_findclient_lock);
1516
1517 /* First stop all of the reaper threads in the database */
1518 rfs4_database_shutdown(dbp);
1519 /*
1520 * WARNING: There may be consumers of the rfs4 database still
1521 * active as we destroy these. IF that's the case, consider putting
1522 * some of their _zone_fini()-like functions into the zsd key as
1523 * ~~SHUTDOWN~~ functions instead of ~~DESTROY~~ functions. We can
1524 * maintain some ordering guarantees better that way.
1525 */
1526 /* Now destroy/release the database tables */
1527 rfs4_database_destroy(dbp);
1528
1529 /* Reset the cache timers for next time */
1530 nsrv4->rfs4_client_cache_time = 0;
1531 nsrv4->rfs4_openowner_cache_time = 0;
1532 nsrv4->rfs4_state_cache_time = 0;
1533 nsrv4->rfs4_lo_state_cache_time = 0;
1534 nsrv4->rfs4_lockowner_cache_time = 0;
1535 nsrv4->rfs4_file_cache_time = 0;
1536 nsrv4->rfs4_deleg_state_cache_time = 0;
1537
1538 mutex_exit(&nsrv4->state_lock);
1539
1540 /* clean up any dangling stable storage structures */
1541 rfs4_ss_fini(nsrv4);
1542 }
1543
1544 typedef union {
1545 struct {
1546 uint32_t start_time;
1547 uint32_t c_id;
1548 } impl_id;
1549 clientid4 id4;
1550 } cid;
1551
1552 static int foreign_stateid(stateid_t *id);
1553 static int foreign_clientid(cid *cidp);
1554 static void embed_nodeid(cid *cidp);
1555
1556 typedef union {
1557 struct {
1558 uint32_t c_id;
1559 uint32_t gen_num;
1560 } cv_impl;
1561 verifier4 confirm_verf;
1562 } scid_confirm_verf;
1563
1564 static uint32_t
1565 clientid_hash(void *key)
1566 {
1567 cid *idp = key;
1568
1569 return (idp->impl_id.c_id);
1570 }
1571
1572 static bool_t
1573 clientid_compare(rfs4_entry_t entry, void *key)
1574 {
1575 rfs4_client_t *cp = (rfs4_client_t *)entry;
1576 clientid4 *idp = key;
1577
1578 return (*idp == cp->rc_clientid);
1579 }
1580
1581 static void *
1582 clientid_mkkey(rfs4_entry_t entry)
1583 {
1584 rfs4_client_t *cp = (rfs4_client_t *)entry;
1585
1586 return (&cp->rc_clientid);
1587 }
1588
1589 static uint32_t
1590 nfsclnt_hash(void *key)
1591 {
1592 nfs_client_id4 *client = key;
1593 int i;
1594 uint32_t hash = 0;
1595
1596 for (i = 0; i < client->id_len; i++) {
1597 hash <<= 1;
1598 hash += (uint_t)client->id_val[i];
1599 }
1600 return (hash);
1601 }
1602
1603
1604 static bool_t
1605 nfsclnt_compare(rfs4_entry_t entry, void *key)
1606 {
1607 rfs4_client_t *cp = (rfs4_client_t *)entry;
1608 nfs_client_id4 *nfs_client = key;
1609
1610 if (cp->rc_nfs_client.id_len != nfs_client->id_len)
1611 return (FALSE);
1612
1613 return (bcmp(cp->rc_nfs_client.id_val, nfs_client->id_val,
1614 nfs_client->id_len) == 0);
1615 }
1616
1617 static void *
1618 nfsclnt_mkkey(rfs4_entry_t entry)
1619 {
1620 rfs4_client_t *cp = (rfs4_client_t *)entry;
1621
1622 return (&cp->rc_nfs_client);
1623 }
1624
1625 static bool_t
1626 rfs4_client_expiry(rfs4_entry_t u_entry)
1627 {
1628 rfs4_client_t *cp = (rfs4_client_t *)u_entry;
1629 bool_t cp_expired;
1630
1631 if (rfs4_dbe_is_invalid(cp->rc_dbe)) {
1632 cp->rc_ss_remove = 1;
1633 return (TRUE);
1634 }
1635 /*
1636 * If the sysadmin has used clear_locks for this
1637 * entry then forced_expire will be set and we
1638 * want this entry to be reaped. Or the entry
1639 * has exceeded its lease period.
1640 */
1641 cp_expired = (cp->rc_forced_expire ||
1642 (gethrestime_sec() - cp->rc_last_access
1643 > rfs4_lease_time));
1644
1645 if (!cp->rc_ss_remove && cp_expired)
1646 cp->rc_ss_remove = 1;
1647 return (cp_expired);
1648 }
1649
1650 /*
1651 * Remove the leaf file from all distributed stable storage paths.
1652 */
1653 static void
1654 rfs4_dss_remove_cpleaf(rfs4_client_t *cp)
1655 {
1656 nfs4_srv_t *nsrv4;
1657 rfs4_servinst_t *sip;
1658 char *leaf = cp->rc_ss_pn->leaf;
1659
1660 /*
1661 * since the state files are written to all DSS
1662 * paths we must remove this leaf file instance
1663 * from all server instances.
1664 */
1665
1666 nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
1667 mutex_enter(&nsrv4->servinst_lock);
1668 for (sip = nsrv4->nfs4_cur_servinst; sip != NULL; sip = sip->prev) {
1669 /* remove the leaf file associated with this server instance */
1670 rfs4_dss_remove_leaf(sip, NFS4_DSS_STATE_LEAF, leaf);
1671 }
1672 mutex_exit(&nsrv4->servinst_lock);
1673 }
1674
1675 static void
1676 rfs4_dss_remove_leaf(rfs4_servinst_t *sip, char *dir_leaf, char *leaf)
1677 {
1678 int i, npaths = sip->dss_npaths;
1679
1680 for (i = 0; i < npaths; i++) {
1681 rfs4_dss_path_t *dss_path = sip->dss_paths[i];
1682 char *path, *dir;
1683 size_t pathlen;
1684
1685 /* the HA-NFSv4 path might have been failed-over away from us */
1686 if (dss_path == NULL)
1687 continue;
1688
1689 dir = dss_path->path;
1690
1691 /* allow 3 extra bytes for two '/' & a NUL */
1692 pathlen = strlen(dir) + strlen(dir_leaf) + strlen(leaf) + 3;
1693 path = kmem_alloc(pathlen, KM_SLEEP);
1694 (void) sprintf(path, "%s/%s/%s", dir, dir_leaf, leaf);
1695
1696 (void) vn_remove(path, UIO_SYSSPACE, RMFILE);
1697
1698 kmem_free(path, pathlen);
1699 }
1700 }
1701
1702 static void
1703 rfs4_client_destroy(rfs4_entry_t u_entry)
1704 {
1705 rfs4_client_t *cp = (rfs4_client_t *)u_entry;
1706
1707 mutex_destroy(cp->rc_cbinfo.cb_lock);
1708 cv_destroy(cp->rc_cbinfo.cb_cv);
1709 cv_destroy(cp->rc_cbinfo.cb_cv_nullcaller);
1710 list_destroy(&cp->rc_openownerlist);
1711
1712 /* free callback info */
1713 rfs4_cbinfo_free(&cp->rc_cbinfo);
1714
1715 if (cp->rc_cp_confirmed)
1716 rfs4_client_rele(cp->rc_cp_confirmed);
1717
1718 if (cp->rc_ss_pn) {
1719 /* check if the stable storage files need to be removed */
1720 if (cp->rc_ss_remove)
1721 rfs4_dss_remove_cpleaf(cp);
1722 rfs4_ss_pnfree(cp->rc_ss_pn);
1723 }
1724
1725 /* Free the client supplied client id */
1726 kmem_free(cp->rc_nfs_client.id_val, cp->rc_nfs_client.id_len);
1727
1728 if (cp->rc_sysidt != LM_NOSYSID)
1729 lm_free_sysidt(cp->rc_sysidt);
1730 }
1731
1732 static bool_t
1733 rfs4_client_create(rfs4_entry_t u_entry, void *arg)
1734 {
1735 rfs4_client_t *cp = (rfs4_client_t *)u_entry;
1736 nfs_client_id4 *client = (nfs_client_id4 *)arg;
1737 struct sockaddr *ca;
1738 cid *cidp;
1739 scid_confirm_verf *scvp;
1740 nfs4_srv_t *nsrv4;
1741
1742 nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
1743
1744 /* Get a clientid to give to the client */
1745 cidp = (cid *)&cp->rc_clientid;
1746 cidp->impl_id.start_time = nsrv4->rfs4_start_time;
1747 cidp->impl_id.c_id = (uint32_t)rfs4_dbe_getid(cp->rc_dbe);
1748
1749 /* If we are booted as a cluster node, embed our nodeid */
1750 if (cluster_bootflags & CLUSTER_BOOTED)
1751 embed_nodeid(cidp);
1752
1753 /* Allocate and copy client's client id value */
1754 cp->rc_nfs_client.id_val = kmem_alloc(client->id_len, KM_SLEEP);
1755 cp->rc_nfs_client.id_len = client->id_len;
1756 bcopy(client->id_val, cp->rc_nfs_client.id_val, client->id_len);
1757 cp->rc_nfs_client.verifier = client->verifier;
1758
1759 /* Copy client's IP address */
1760 ca = client->cl_addr;
1761 if (ca->sa_family == AF_INET)
1762 bcopy(ca, &cp->rc_addr, sizeof (struct sockaddr_in));
1763 else if (ca->sa_family == AF_INET6)
1764 bcopy(ca, &cp->rc_addr, sizeof (struct sockaddr_in6));
1765 cp->rc_nfs_client.cl_addr = (struct sockaddr *)&cp->rc_addr;
1766
1767 /* Init the value for the SETCLIENTID_CONFIRM verifier */
1768 scvp = (scid_confirm_verf *)&cp->rc_confirm_verf;
1769 scvp->cv_impl.c_id = cidp->impl_id.c_id;
1770 scvp->cv_impl.gen_num = 0;
1771
1772 /* An F_UNLKSYS has been done for this client */
1773 cp->rc_unlksys_completed = FALSE;
1774
1775 /* We need the client to ack us */
1776 cp->rc_need_confirm = TRUE;
1777 cp->rc_cp_confirmed = NULL;
1778
1779 /* TRUE all the time until the callback path actually fails */
1780 cp->rc_cbinfo.cb_notified_of_cb_path_down = TRUE;
1781
1782 /* Initialize the access time to now */
1783 cp->rc_last_access = gethrestime_sec();
1784
1785 cp->rc_cr_set = NULL;
1786
1787 cp->rc_sysidt = LM_NOSYSID;
1788
1789 list_create(&cp->rc_openownerlist, sizeof (rfs4_openowner_t),
1790 offsetof(rfs4_openowner_t, ro_node));
1791
1792 /* set up the callback control structure */
1793 cp->rc_cbinfo.cb_state = CB_UNINIT;
1794 mutex_init(cp->rc_cbinfo.cb_lock, NULL, MUTEX_DEFAULT, NULL);
1795 cv_init(cp->rc_cbinfo.cb_cv, NULL, CV_DEFAULT, NULL);
1796 cv_init(cp->rc_cbinfo.cb_cv_nullcaller, NULL, CV_DEFAULT, NULL);
1797
1798 /*
1799 * Associate the client_t with the current server instance.
1800 * The hold is solely to satisfy the calling requirement of
1801 * rfs4_servinst_assign(). In this case it's not strictly necessary.
1802 */
1803 rfs4_dbe_hold(cp->rc_dbe);
1804 rfs4_servinst_assign(nsrv4, cp, nsrv4->nfs4_cur_servinst);
1805 rfs4_dbe_rele(cp->rc_dbe);
1806
1807 return (TRUE);
1808 }
1809
1810 /*
1811 * Caller wants to generate/update the setclientid_confirm verifier
1812 * associated with a client. This is done during the SETCLIENTID
1813 * processing.
1814 */
1815 void
1816 rfs4_client_scv_next(rfs4_client_t *cp)
1817 {
1818 scid_confirm_verf *scvp;
1819
1820 /* Init the value for the SETCLIENTID_CONFIRM verifier */
1821 scvp = (scid_confirm_verf *)&cp->rc_confirm_verf;
1822 scvp->cv_impl.gen_num++;
1823 }
1824
1825 void
1826 rfs4_client_rele(rfs4_client_t *cp)
1827 {
1828 rfs4_dbe_rele(cp->rc_dbe);
1829 }
1830
1831 rfs4_client_t *
1832 rfs4_findclient(nfs_client_id4 *client, bool_t *create, rfs4_client_t *oldcp)
1833 {
1834 rfs4_client_t *cp;
1835 nfs4_srv_t *nsrv4;
1836 nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
1837
1838
1839 if (oldcp) {
1840 rw_enter(&nsrv4->rfs4_findclient_lock, RW_WRITER);
1841 rfs4_dbe_hide(oldcp->rc_dbe);
1842 } else {
1843 rw_enter(&nsrv4->rfs4_findclient_lock, RW_READER);
1844 }
1845
1846 cp = (rfs4_client_t *)rfs4_dbsearch(nsrv4->rfs4_nfsclnt_idx, client,
1847 create, (void *)client, RFS4_DBS_VALID);
1848
1849 if (oldcp)
1850 rfs4_dbe_unhide(oldcp->rc_dbe);
1851
1852 rw_exit(&nsrv4->rfs4_findclient_lock);
1853
1854 return (cp);
1855 }
1856
1857 rfs4_client_t *
1858 rfs4_findclient_by_id(clientid4 clientid, bool_t find_unconfirmed)
1859 {
1860 rfs4_client_t *cp;
1861 bool_t create = FALSE;
1862 cid *cidp = (cid *)&clientid;
1863 nfs4_srv_t *nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
1864
1865 /* If we're a cluster and the nodeid isn't right, short-circuit */
1866 if (cluster_bootflags & CLUSTER_BOOTED && foreign_clientid(cidp))
1867 return (NULL);
1868
1869 rw_enter(&nsrv4->rfs4_findclient_lock, RW_READER);
1870
1871 cp = (rfs4_client_t *)rfs4_dbsearch(nsrv4->rfs4_clientid_idx, &clientid,
1872 &create, NULL, RFS4_DBS_VALID);
1873
1874 rw_exit(&nsrv4->rfs4_findclient_lock);
1875
1876 if (cp && cp->rc_need_confirm && find_unconfirmed == FALSE) {
1877 rfs4_client_rele(cp);
1878 return (NULL);
1879 } else {
1880 return (cp);
1881 }
1882 }
1883
1884 static uint32_t
1885 clntip_hash(void *key)
1886 {
1887 struct sockaddr *addr = key;
1888 int i, len = 0;
1889 uint32_t hash = 0;
1890 char *ptr;
1891
1892 if (addr->sa_family == AF_INET) {
1893 struct sockaddr_in *a = (struct sockaddr_in *)addr;
1894 len = sizeof (struct in_addr);
1895 ptr = (char *)&a->sin_addr;
1896 } else if (addr->sa_family == AF_INET6) {
1897 struct sockaddr_in6 *a = (struct sockaddr_in6 *)addr;
1898 len = sizeof (struct in6_addr);
1899 ptr = (char *)&a->sin6_addr;
1900 } else
1901 return (0);
1902
1903 for (i = 0; i < len; i++) {
1904 hash <<= 1;
1905 hash += (uint_t)ptr[i];
1906 }
1907 return (hash);
1908 }
1909
1910 static bool_t
1911 clntip_compare(rfs4_entry_t entry, void *key)
1912 {
1913 rfs4_clntip_t *cp = (rfs4_clntip_t *)entry;
1914 struct sockaddr *addr = key;
1915 int len = 0;
1916 char *p1, *p2;
1917
1918 if (addr->sa_family == AF_INET) {
1919 struct sockaddr_in *a1 = (struct sockaddr_in *)&cp->ri_addr;
1920 struct sockaddr_in *a2 = (struct sockaddr_in *)addr;
1921 len = sizeof (struct in_addr);
1922 p1 = (char *)&a1->sin_addr;
1923 p2 = (char *)&a2->sin_addr;
1924 } else if (addr->sa_family == AF_INET6) {
1925 struct sockaddr_in6 *a1 = (struct sockaddr_in6 *)&cp->ri_addr;
1926 struct sockaddr_in6 *a2 = (struct sockaddr_in6 *)addr;
1927 len = sizeof (struct in6_addr);
1928 p1 = (char *)&a1->sin6_addr;
1929 p2 = (char *)&a2->sin6_addr;
1930 } else
1931 return (0);
1932
1933 return (bcmp(p1, p2, len) == 0);
1934 }
1935
1936 static void *
1937 clntip_mkkey(rfs4_entry_t entry)
1938 {
1939 rfs4_clntip_t *cp = (rfs4_clntip_t *)entry;
1940
1941 return (&cp->ri_addr);
1942 }
1943
1944 static bool_t
1945 rfs4_clntip_expiry(rfs4_entry_t u_entry)
1946 {
1947 rfs4_clntip_t *cp = (rfs4_clntip_t *)u_entry;
1948
1949 if (rfs4_dbe_is_invalid(cp->ri_dbe))
1950 return (TRUE);
1951 return (FALSE);
1952 }
1953
1954 /* ARGSUSED */
1955 static void
1956 rfs4_clntip_destroy(rfs4_entry_t u_entry)
1957 {
1958 }
1959
1960 static bool_t
1961 rfs4_clntip_create(rfs4_entry_t u_entry, void *arg)
1962 {
1963 rfs4_clntip_t *cp = (rfs4_clntip_t *)u_entry;
1964 struct sockaddr *ca = (struct sockaddr *)arg;
1965
1966 /* Copy client's IP address */
1967 if (ca->sa_family == AF_INET)
1968 bcopy(ca, &cp->ri_addr, sizeof (struct sockaddr_in));
1969 else if (ca->sa_family == AF_INET6)
1970 bcopy(ca, &cp->ri_addr, sizeof (struct sockaddr_in6));
1971 else
1972 return (FALSE);
1973 cp->ri_no_referrals = 1;
1974
1975 return (TRUE);
1976 }
1977
1978 rfs4_clntip_t *
1979 rfs4_find_clntip(struct sockaddr *addr, bool_t *create)
1980 {
1981 rfs4_clntip_t *cp;
1982 nfs4_srv_t *nsrv4;
1983
1984 nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
1985
1986 rw_enter(&nsrv4->rfs4_findclient_lock, RW_READER);
1987
1988 cp = (rfs4_clntip_t *)rfs4_dbsearch(nsrv4->rfs4_clntip_idx, addr,
1989 create, addr, RFS4_DBS_VALID);
1990
1991 rw_exit(&nsrv4->rfs4_findclient_lock);
1992
1993 return (cp);
1994 }
1995
1996 void
1997 rfs4_invalidate_clntip(struct sockaddr *addr)
1998 {
1999 rfs4_clntip_t *cp;
2000 bool_t create = FALSE;
2001 nfs4_srv_t *nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
2002
2003 rw_enter(&nsrv4->rfs4_findclient_lock, RW_READER);
2004
2005 cp = (rfs4_clntip_t *)rfs4_dbsearch(nsrv4->rfs4_clntip_idx, addr,
2006 &create, NULL, RFS4_DBS_VALID);
2007 if (cp == NULL) {
2008 rw_exit(&nsrv4->rfs4_findclient_lock);
2009 return;
2010 }
2011 rfs4_dbe_invalidate(cp->ri_dbe);
2012 rfs4_dbe_rele(cp->ri_dbe);
2013
2014 rw_exit(&nsrv4->rfs4_findclient_lock);
2015 }
2016
2017 bool_t
2018 rfs4_lease_expired(rfs4_client_t *cp)
2019 {
2020 bool_t rc;
2021
2022 rfs4_dbe_lock(cp->rc_dbe);
2023
2024 /*
2025 * If the admin has executed clear_locks for this
2026 * client id, force expire will be set, so no need
2027 * to calculate anything because it's "outa here".
2028 */
2029 if (cp->rc_forced_expire) {
2030 rc = TRUE;
2031 } else {
2032 rc = (gethrestime_sec() - cp->rc_last_access > rfs4_lease_time);
2033 }
2034
2035 /*
2036 * If the lease has expired we will also want
2037 * to remove any stable storage state data. So
2038 * mark the client id accordingly.
2039 */
2040 if (!cp->rc_ss_remove)
2041 cp->rc_ss_remove = (rc == TRUE);
2042
2043 rfs4_dbe_unlock(cp->rc_dbe);
2044
2045 return (rc);
2046 }
2047
2048 void
2049 rfs4_update_lease(rfs4_client_t *cp)
2050 {
2051 rfs4_dbe_lock(cp->rc_dbe);
2052 if (!cp->rc_forced_expire)
2053 cp->rc_last_access = gethrestime_sec();
2054 rfs4_dbe_unlock(cp->rc_dbe);
2055 }
2056
2057
2058 static bool_t
2059 EQOPENOWNER(open_owner4 *a, open_owner4 *b)
2060 {
2061 bool_t rc;
2062
2063 if (a->clientid != b->clientid)
2064 return (FALSE);
2065
2066 if (a->owner_len != b->owner_len)
2067 return (FALSE);
2068
2069 rc = (bcmp(a->owner_val, b->owner_val, a->owner_len) == 0);
2070
2071 return (rc);
2072 }
2073
2074 static uint_t
2075 openowner_hash(void *key)
2076 {
2077 int i;
2078 open_owner4 *openowner = key;
2079 uint_t hash = 0;
2080
2081 for (i = 0; i < openowner->owner_len; i++) {
2082 hash <<= 4;
2083 hash += (uint_t)openowner->owner_val[i];
2084 }
2085 hash += (uint_t)openowner->clientid;
2086 hash |= (openowner->clientid >> 32);
2087
2088 return (hash);
2089 }
2090
2091 static bool_t
2092 openowner_compare(rfs4_entry_t u_entry, void *key)
2093 {
2094 rfs4_openowner_t *oo = (rfs4_openowner_t *)u_entry;
2095 open_owner4 *arg = key;
2096
2097 return (EQOPENOWNER(&oo->ro_owner, arg));
2098 }
2099
2100 void *
2101 openowner_mkkey(rfs4_entry_t u_entry)
2102 {
2103 rfs4_openowner_t *oo = (rfs4_openowner_t *)u_entry;
2104
2105 return (&oo->ro_owner);
2106 }
2107
2108 /* ARGSUSED */
2109 static bool_t
2110 rfs4_openowner_expiry(rfs4_entry_t u_entry)
2111 {
2112 /* openstateid held us and did all needed delay */
2113 return (TRUE);
2114 }
2115
2116 static void
2117 rfs4_openowner_destroy(rfs4_entry_t u_entry)
2118 {
2119 rfs4_openowner_t *oo = (rfs4_openowner_t *)u_entry;
2120
2121 /* Remove open owner from client's lists of open owners */
2122 rfs4_dbe_lock(oo->ro_client->rc_dbe);
2123 list_remove(&oo->ro_client->rc_openownerlist, oo);
2124 rfs4_dbe_unlock(oo->ro_client->rc_dbe);
2125
2126 /* One less reference to the client */
2127 rfs4_client_rele(oo->ro_client);
2128 oo->ro_client = NULL;
2129
2130 /* Free the last reply for this lock owner */
2131 rfs4_free_reply(&oo->ro_reply);
2132
2133 if (oo->ro_reply_fh.nfs_fh4_val) {
2134 kmem_free(oo->ro_reply_fh.nfs_fh4_val,
2135 oo->ro_reply_fh.nfs_fh4_len);
2136 oo->ro_reply_fh.nfs_fh4_val = NULL;
2137 oo->ro_reply_fh.nfs_fh4_len = 0;
2138 }
2139
2140 rfs4_sw_destroy(&oo->ro_sw);
2141 list_destroy(&oo->ro_statelist);
2142
2143 /* Free the lock owner id */
2144 kmem_free(oo->ro_owner.owner_val, oo->ro_owner.owner_len);
2145 }
2146
2147 void
2148 rfs4_openowner_rele(rfs4_openowner_t *oo)
2149 {
2150 rfs4_dbe_rele(oo->ro_dbe);
2151 }
2152
2153 static bool_t
2154 rfs4_openowner_create(rfs4_entry_t u_entry, void *arg)
2155 {
2156 rfs4_openowner_t *oo = (rfs4_openowner_t *)u_entry;
2157 rfs4_openowner_t *argp = (rfs4_openowner_t *)arg;
2158 open_owner4 *openowner = &argp->ro_owner;
2159 seqid4 seqid = argp->ro_open_seqid;
2160 rfs4_client_t *cp;
2161 bool_t create = FALSE;
2162 nfs4_srv_t *nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
2163
2164 rw_enter(&nsrv4->rfs4_findclient_lock, RW_READER);
2165
2166 cp = (rfs4_client_t *)rfs4_dbsearch(nsrv4->rfs4_clientid_idx,
2167 &openowner->clientid,
2168 &create, NULL, RFS4_DBS_VALID);
2169
2170 rw_exit(&nsrv4->rfs4_findclient_lock);
2171
2172 if (cp == NULL)
2173 return (FALSE);
2174
2175 oo->ro_reply_fh.nfs_fh4_len = 0;
2176 oo->ro_reply_fh.nfs_fh4_val = NULL;
2177
2178 oo->ro_owner.clientid = openowner->clientid;
2179 oo->ro_owner.owner_val =
2180 kmem_alloc(openowner->owner_len, KM_SLEEP);
2181
2182 bcopy(openowner->owner_val,
2183 oo->ro_owner.owner_val, openowner->owner_len);
2184
2185 oo->ro_owner.owner_len = openowner->owner_len;
2186
2187 oo->ro_need_confirm = TRUE;
2188
2189 rfs4_sw_init(&oo->ro_sw);
2190
2191 oo->ro_open_seqid = seqid;
2192 bzero(&oo->ro_reply, sizeof (nfs_resop4));
2193 oo->ro_client = cp;
2194 oo->ro_cr_set = NULL;
2195
2196 list_create(&oo->ro_statelist, sizeof (rfs4_state_t),
2197 offsetof(rfs4_state_t, rs_node));
2198
2199 /* Insert openowner into client's open owner list */
2200 rfs4_dbe_lock(cp->rc_dbe);
2201 list_insert_tail(&cp->rc_openownerlist, oo);
2202 rfs4_dbe_unlock(cp->rc_dbe);
2203
2204 return (TRUE);
2205 }
2206
2207 rfs4_openowner_t *
2208 rfs4_findopenowner(open_owner4 *openowner, bool_t *create, seqid4 seqid)
2209 {
2210 rfs4_openowner_t *oo;
2211 rfs4_openowner_t arg;
2212 nfs4_srv_t *nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
2213
2214 arg.ro_owner = *openowner;
2215 arg.ro_open_seqid = seqid;
2216 /* CSTYLED */
2217 oo = (rfs4_openowner_t *)rfs4_dbsearch(nsrv4->rfs4_openowner_idx, openowner,
2218 create, &arg, RFS4_DBS_VALID);
2219
2220 return (oo);
2221 }
2222
2223 void
2224 rfs4_update_open_sequence(rfs4_openowner_t *oo)
2225 {
2226
2227 rfs4_dbe_lock(oo->ro_dbe);
2228
2229 oo->ro_open_seqid++;
2230
2231 rfs4_dbe_unlock(oo->ro_dbe);
2232 }
2233
2234 void
2235 rfs4_update_open_resp(rfs4_openowner_t *oo, nfs_resop4 *resp, nfs_fh4 *fh)
2236 {
2237
2238 rfs4_dbe_lock(oo->ro_dbe);
2239
2240 rfs4_free_reply(&oo->ro_reply);
2241
2242 rfs4_copy_reply(&oo->ro_reply, resp);
2243
2244 /* Save the filehandle if provided and free if not used */
2245 if (resp->nfs_resop4_u.opopen.status == NFS4_OK &&
2246 fh && fh->nfs_fh4_len) {
2247 if (oo->ro_reply_fh.nfs_fh4_val == NULL)
2248 oo->ro_reply_fh.nfs_fh4_val =
2249 kmem_alloc(fh->nfs_fh4_len, KM_SLEEP);
2250 nfs_fh4_copy(fh, &oo->ro_reply_fh);
2251 } else {
2252 if (oo->ro_reply_fh.nfs_fh4_val) {
2253 kmem_free(oo->ro_reply_fh.nfs_fh4_val,
2254 oo->ro_reply_fh.nfs_fh4_len);
2255 oo->ro_reply_fh.nfs_fh4_val = NULL;
2256 oo->ro_reply_fh.nfs_fh4_len = 0;
2257 }
2258 }
2259
2260 rfs4_dbe_unlock(oo->ro_dbe);
2261 }
2262
2263 static bool_t
2264 lockowner_compare(rfs4_entry_t u_entry, void *key)
2265 {
2266 rfs4_lockowner_t *lo = (rfs4_lockowner_t *)u_entry;
2267 lock_owner4 *b = (lock_owner4 *)key;
2268
2269 if (lo->rl_owner.clientid != b->clientid)
2270 return (FALSE);
2271
2272 if (lo->rl_owner.owner_len != b->owner_len)
2273 return (FALSE);
2274
2275 return (bcmp(lo->rl_owner.owner_val, b->owner_val,
2276 lo->rl_owner.owner_len) == 0);
2277 }
2278
2279 void *
2280 lockowner_mkkey(rfs4_entry_t u_entry)
2281 {
2282 rfs4_lockowner_t *lo = (rfs4_lockowner_t *)u_entry;
2283
2284 return (&lo->rl_owner);
2285 }
2286
2287 static uint32_t
2288 lockowner_hash(void *key)
2289 {
2290 int i;
2291 lock_owner4 *lockowner = key;
2292 uint_t hash = 0;
2293
2294 for (i = 0; i < lockowner->owner_len; i++) {
2295 hash <<= 4;
2296 hash += (uint_t)lockowner->owner_val[i];
2297 }
2298 hash += (uint_t)lockowner->clientid;
2299 hash |= (lockowner->clientid >> 32);
2300
2301 return (hash);
2302 }
2303
2304 static uint32_t
2305 pid_hash(void *key)
2306 {
2307 return ((uint32_t)(uintptr_t)key);
2308 }
2309
2310 static void *
2311 pid_mkkey(rfs4_entry_t u_entry)
2312 {
2313 rfs4_lockowner_t *lo = (rfs4_lockowner_t *)u_entry;
2314
2315 return ((void *)(uintptr_t)lo->rl_pid);
2316 }
2317
2318 static bool_t
2319 pid_compare(rfs4_entry_t u_entry, void *key)
2320 {
2321 rfs4_lockowner_t *lo = (rfs4_lockowner_t *)u_entry;
2322
2323 return (lo->rl_pid == (pid_t)(uintptr_t)key);
2324 }
2325
2326 static void
2327 rfs4_lockowner_destroy(rfs4_entry_t u_entry)
2328 {
2329 rfs4_lockowner_t *lo = (rfs4_lockowner_t *)u_entry;
2330
2331 /* Free the lock owner id */
2332 kmem_free(lo->rl_owner.owner_val, lo->rl_owner.owner_len);
2333 rfs4_client_rele(lo->rl_client);
2334 }
2335
2336 void
2337 rfs4_lockowner_rele(rfs4_lockowner_t *lo)
2338 {
2339 rfs4_dbe_rele(lo->rl_dbe);
2340 }
2341
2342 /* ARGSUSED */
2343 static bool_t
2344 rfs4_lockowner_expiry(rfs4_entry_t u_entry)
2345 {
2346 /*
2347 * Since expiry is called with no other references on
2348 * this struct, go ahead and have it removed.
2349 */
2350 return (TRUE);
2351 }
2352
2353 static bool_t
2354 rfs4_lockowner_create(rfs4_entry_t u_entry, void *arg)
2355 {
2356 rfs4_lockowner_t *lo = (rfs4_lockowner_t *)u_entry;
2357 lock_owner4 *lockowner = (lock_owner4 *)arg;
2358 rfs4_client_t *cp;
2359 bool_t create = FALSE;
2360 nfs4_srv_t *nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
2361
2362 rw_enter(&nsrv4->rfs4_findclient_lock, RW_READER);
2363
2364 cp = (rfs4_client_t *)rfs4_dbsearch(nsrv4->rfs4_clientid_idx,
2365 &lockowner->clientid,
2366 &create, NULL, RFS4_DBS_VALID);
2367
2368 rw_exit(&nsrv4->rfs4_findclient_lock);
2369
2370 if (cp == NULL)
2371 return (FALSE);
2372
2373 /* Reference client */
2374 lo->rl_client = cp;
2375 lo->rl_owner.clientid = lockowner->clientid;
2376 lo->rl_owner.owner_val = kmem_alloc(lockowner->owner_len, KM_SLEEP);
2377 bcopy(lockowner->owner_val, lo->rl_owner.owner_val,
2378 lockowner->owner_len);
2379 lo->rl_owner.owner_len = lockowner->owner_len;
2380 lo->rl_pid = rfs4_dbe_getid(lo->rl_dbe);
2381
2382 return (TRUE);
2383 }
2384
2385 rfs4_lockowner_t *
2386 rfs4_findlockowner(lock_owner4 *lockowner, bool_t *create)
2387 {
2388 rfs4_lockowner_t *lo;
2389 nfs4_srv_t *nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
2390
2391 /* CSTYLED */
2392 lo = (rfs4_lockowner_t *)rfs4_dbsearch(nsrv4->rfs4_lockowner_idx, lockowner,
2393 create, lockowner, RFS4_DBS_VALID);
2394
2395 return (lo);
2396 }
2397
2398 rfs4_lockowner_t *
2399 rfs4_findlockowner_by_pid(pid_t pid)
2400 {
2401 rfs4_lockowner_t *lo;
2402 bool_t create = FALSE;
2403 nfs4_srv_t *nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
2404
2405 lo = (rfs4_lockowner_t *)rfs4_dbsearch(nsrv4->rfs4_lockowner_pid_idx,
2406 (void *)(uintptr_t)pid, &create, NULL, RFS4_DBS_VALID);
2407
2408 return (lo);
2409 }
2410
2411
2412 static uint32_t
2413 file_hash(void *key)
2414 {
2415 return (ADDRHASH(key));
2416 }
2417
2418 static void *
2419 file_mkkey(rfs4_entry_t u_entry)
2420 {
2421 rfs4_file_t *fp = (rfs4_file_t *)u_entry;
2422
2423 return (fp->rf_vp);
2424 }
2425
2426 static bool_t
2427 file_compare(rfs4_entry_t u_entry, void *key)
2428 {
2429 rfs4_file_t *fp = (rfs4_file_t *)u_entry;
2430
2431 return (fp->rf_vp == (vnode_t *)key);
2432 }
2433
2434 static void
2435 rfs4_file_destroy(rfs4_entry_t u_entry)
2436 {
2437 rfs4_file_t *fp = (rfs4_file_t *)u_entry;
2438
2439 list_destroy(&fp->rf_delegstatelist);
2440
2441 if (fp->rf_filehandle.nfs_fh4_val)
2442 kmem_free(fp->rf_filehandle.nfs_fh4_val,
2443 fp->rf_filehandle.nfs_fh4_len);
2444 cv_destroy(fp->rf_dinfo.rd_recall_cv);
2445 if (fp->rf_vp) {
2446 vnode_t *vp = fp->rf_vp;
2447
2448 mutex_enter(&vp->v_vsd_lock);
2449 (void) vsd_set(vp, nfs4_srv_vkey, NULL);
2450 mutex_exit(&vp->v_vsd_lock);
2451 VN_RELE(vp);
2452 fp->rf_vp = NULL;
2453 }
2454 rw_destroy(&fp->rf_file_rwlock);
2455 }
2456
2457 /*
2458 * Used to unlock the underlying dbe struct only
2459 */
2460 void
2461 rfs4_file_rele(rfs4_file_t *fp)
2462 {
2463 rfs4_dbe_rele(fp->rf_dbe);
2464 }
2465
2466 typedef struct {
2467 vnode_t *vp;
2468 nfs_fh4 *fh;
2469 } rfs4_fcreate_arg;
2470
2471 static bool_t
2472 rfs4_file_create(rfs4_entry_t u_entry, void *arg)
2473 {
2474 rfs4_file_t *fp = (rfs4_file_t *)u_entry;
2475 rfs4_fcreate_arg *ap = (rfs4_fcreate_arg *)arg;
2476 vnode_t *vp = ap->vp;
2477 nfs_fh4 *fh = ap->fh;
2478
2479 VN_HOLD(vp);
2480
2481 fp->rf_filehandle.nfs_fh4_len = 0;
2482 fp->rf_filehandle.nfs_fh4_val = NULL;
2483 ASSERT(fh && fh->nfs_fh4_len);
2484 if (fh && fh->nfs_fh4_len) {
2485 fp->rf_filehandle.nfs_fh4_val =
2486 kmem_alloc(fh->nfs_fh4_len, KM_SLEEP);
2487 nfs_fh4_copy(fh, &fp->rf_filehandle);
2488 }
2489 fp->rf_vp = vp;
2490
2491 list_create(&fp->rf_delegstatelist, sizeof (rfs4_deleg_state_t),
2492 offsetof(rfs4_deleg_state_t, rds_node));
2493
2494 fp->rf_share_deny = fp->rf_share_access = fp->rf_access_read = 0;
2495 fp->rf_access_write = fp->rf_deny_read = fp->rf_deny_write = 0;
2496
2497 mutex_init(fp->rf_dinfo.rd_recall_lock, NULL, MUTEX_DEFAULT, NULL);
2498 cv_init(fp->rf_dinfo.rd_recall_cv, NULL, CV_DEFAULT, NULL);
2499
2500 fp->rf_dinfo.rd_dtype = OPEN_DELEGATE_NONE;
2501
2502 rw_init(&fp->rf_file_rwlock, NULL, RW_DEFAULT, NULL);
2503
2504 mutex_enter(&vp->v_vsd_lock);
2505 VERIFY(vsd_set(vp, nfs4_srv_vkey, (void *)fp) == 0);
2506 mutex_exit(&vp->v_vsd_lock);
2507
2508 return (TRUE);
2509 }
2510
2511 rfs4_file_t *
2512 rfs4_findfile(vnode_t *vp, nfs_fh4 *fh, bool_t *create)
2513 {
2514 rfs4_file_t *fp;
2515 rfs4_fcreate_arg arg;
2516 nfs4_srv_t *nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
2517
2518 arg.vp = vp;
2519 arg.fh = fh;
2520
2521 if (*create == TRUE)
2522 /* CSTYLED */
2523 fp = (rfs4_file_t *)rfs4_dbsearch(nsrv4->rfs4_file_idx, vp, create,
2524 &arg, RFS4_DBS_VALID);
2525 else {
2526 mutex_enter(&vp->v_vsd_lock);
2527 fp = (rfs4_file_t *)vsd_get(vp, nfs4_srv_vkey);
2528 if (fp) {
2529 rfs4_dbe_lock(fp->rf_dbe);
2530 if (rfs4_dbe_is_invalid(fp->rf_dbe) ||
2531 (rfs4_dbe_refcnt(fp->rf_dbe) == 0)) {
2532 rfs4_dbe_unlock(fp->rf_dbe);
2533 fp = NULL;
2534 } else {
2535 rfs4_dbe_hold(fp->rf_dbe);
2536 rfs4_dbe_unlock(fp->rf_dbe);
2537 }
2538 }
2539 mutex_exit(&vp->v_vsd_lock);
2540 }
2541 return (fp);
2542 }
2543
2544 /*
2545 * Find a file in the db and once it is located, take the rw lock.
2546 * Need to check the vnode pointer and if it does not exist (it was
2547 * removed between the db location and check) redo the find. This
2548 * assumes that a file struct that has a NULL vnode pointer is marked
2549 * at 'invalid' and will not be found in the db the second time
2550 * around.
2551 */
2552 rfs4_file_t *
2553 rfs4_findfile_withlock(vnode_t *vp, nfs_fh4 *fh, bool_t *create)
2554 {
2555 rfs4_file_t *fp;
2556 rfs4_fcreate_arg arg;
2557 bool_t screate = *create;
2558 nfs4_srv_t *nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
2559
2560 if (screate == FALSE) {
2561 mutex_enter(&vp->v_vsd_lock);
2562 fp = (rfs4_file_t *)vsd_get(vp, nfs4_srv_vkey);
2563 if (fp) {
2564 rfs4_dbe_lock(fp->rf_dbe);
2565 if (rfs4_dbe_is_invalid(fp->rf_dbe) ||
2566 (rfs4_dbe_refcnt(fp->rf_dbe) == 0)) {
2567 rfs4_dbe_unlock(fp->rf_dbe);
2568 mutex_exit(&vp->v_vsd_lock);
2569 fp = NULL;
2570 } else {
2571 rfs4_dbe_hold(fp->rf_dbe);
2572 rfs4_dbe_unlock(fp->rf_dbe);
2573 mutex_exit(&vp->v_vsd_lock);
2574 rw_enter(&fp->rf_file_rwlock, RW_WRITER);
2575 if (fp->rf_vp == NULL) {
2576 rw_exit(&fp->rf_file_rwlock);
2577 rfs4_file_rele(fp);
2578 fp = NULL;
2579 }
2580 }
2581 } else {
2582 mutex_exit(&vp->v_vsd_lock);
2583 }
2584 } else {
2585 retry:
2586 arg.vp = vp;
2587 arg.fh = fh;
2588
2589 fp = (rfs4_file_t *)rfs4_dbsearch(nsrv4->rfs4_file_idx, vp,
2590 create, &arg, RFS4_DBS_VALID);
2591 if (fp != NULL) {
2592 rw_enter(&fp->rf_file_rwlock, RW_WRITER);
2593 if (fp->rf_vp == NULL) {
2594 rw_exit(&fp->rf_file_rwlock);
2595 rfs4_file_rele(fp);
2596 *create = screate;
2597 goto retry;
2598 }
2599 }
2600 }
2601
2602 return (fp);
2603 }
2604
2605 static uint32_t
2606 lo_state_hash(void *key)
2607 {
2608 stateid_t *id = key;
2609
2610 return (id->bits.ident+id->bits.pid);
2611 }
2612
2613 static bool_t
2614 lo_state_compare(rfs4_entry_t u_entry, void *key)
2615 {
2616 rfs4_lo_state_t *lsp = (rfs4_lo_state_t *)u_entry;
2617 stateid_t *id = key;
2618 bool_t rc;
2619
2620 rc = (lsp->rls_lockid.bits.boottime == id->bits.boottime &&
2621 lsp->rls_lockid.bits.type == id->bits.type &&
2622 lsp->rls_lockid.bits.ident == id->bits.ident &&
2623 lsp->rls_lockid.bits.pid == id->bits.pid);
2624
2625 return (rc);
2626 }
2627
2628 static void *
2629 lo_state_mkkey(rfs4_entry_t u_entry)
2630 {
2631 rfs4_lo_state_t *lsp = (rfs4_lo_state_t *)u_entry;
2632
2633 return (&lsp->rls_lockid);
2634 }
2635
2636 static bool_t
2637 rfs4_lo_state_expiry(rfs4_entry_t u_entry)
2638 {
2639 rfs4_lo_state_t *lsp = (rfs4_lo_state_t *)u_entry;
2640
2641 if (rfs4_dbe_is_invalid(lsp->rls_dbe))
2642 return (TRUE);
2643 if (lsp->rls_state->rs_closed)
2644 return (TRUE);
2645 return ((gethrestime_sec() -
2646 lsp->rls_state->rs_owner->ro_client->rc_last_access
2647 > rfs4_lease_time));
2648 }
2649
2650 static void
2651 rfs4_lo_state_destroy(rfs4_entry_t u_entry)
2652 {
2653 rfs4_lo_state_t *lsp = (rfs4_lo_state_t *)u_entry;
2654
2655 rfs4_dbe_lock(lsp->rls_state->rs_dbe);
2656 list_remove(&lsp->rls_state->rs_lostatelist, lsp);
2657 rfs4_dbe_unlock(lsp->rls_state->rs_dbe);
2658
2659 rfs4_sw_destroy(&lsp->rls_sw);
2660
2661 /* Make sure to release the file locks */
2662 if (lsp->rls_locks_cleaned == FALSE) {
2663 lsp->rls_locks_cleaned = TRUE;
2664 if (lsp->rls_locker->rl_client->rc_sysidt != LM_NOSYSID) {
2665 /* Is the PxFS kernel module loaded? */
2666 if (lm_remove_file_locks != NULL) {
2667 int new_sysid;
2668
2669 /* Encode the cluster nodeid in new sysid */
2670 new_sysid =
2671 lsp->rls_locker->rl_client->rc_sysidt;
2672 lm_set_nlmid_flk(&new_sysid);
2673
2674 /*
2675 * This PxFS routine removes file locks for a
2676 * client over all nodes of a cluster.
2677 */
2678 DTRACE_PROBE1(nfss_i_clust_rm_lck,
2679 int, new_sysid);
2680 (*lm_remove_file_locks)(new_sysid);
2681 } else {
2682 (void) cleanlocks(
2683 lsp->rls_state->rs_finfo->rf_vp,
2684 lsp->rls_locker->rl_pid,
2685 lsp->rls_locker->rl_client->rc_sysidt);
2686 }
2687 }
2688 }
2689
2690 /* Free the last reply for this state */
2691 rfs4_free_reply(&lsp->rls_reply);
2692
2693 rfs4_lockowner_rele(lsp->rls_locker);
2694 lsp->rls_locker = NULL;
2695
2696 rfs4_state_rele_nounlock(lsp->rls_state);
2697 lsp->rls_state = NULL;
2698 }
2699
2700 static bool_t
2701 rfs4_lo_state_create(rfs4_entry_t u_entry, void *arg)
2702 {
2703 rfs4_lo_state_t *lsp = (rfs4_lo_state_t *)u_entry;
2704 rfs4_lo_state_t *argp = (rfs4_lo_state_t *)arg;
2705 rfs4_lockowner_t *lo = argp->rls_locker;
2706 rfs4_state_t *sp = argp->rls_state;
2707
2708 lsp->rls_state = sp;
2709
2710 lsp->rls_lockid = sp->rs_stateid;
2711 lsp->rls_lockid.bits.type = LOCKID;
2712 lsp->rls_lockid.bits.chgseq = 0;
2713 lsp->rls_lockid.bits.pid = lo->rl_pid;
2714
2715 lsp->rls_locks_cleaned = FALSE;
2716 lsp->rls_lock_completed = FALSE;
2717
2718 rfs4_sw_init(&lsp->rls_sw);
2719
2720 /* Attached the supplied lock owner */
2721 rfs4_dbe_hold(lo->rl_dbe);
2722 lsp->rls_locker = lo;
2723
2724 rfs4_dbe_lock(sp->rs_dbe);
2725 list_insert_tail(&sp->rs_lostatelist, lsp);
2726 rfs4_dbe_hold(sp->rs_dbe);
2727 rfs4_dbe_unlock(sp->rs_dbe);
2728
2729 return (TRUE);
2730 }
2731
2732 void
2733 rfs4_lo_state_rele(rfs4_lo_state_t *lsp, bool_t unlock_fp)
2734 {
2735 if (unlock_fp == TRUE)
2736 rw_exit(&lsp->rls_state->rs_finfo->rf_file_rwlock);
2737 rfs4_dbe_rele(lsp->rls_dbe);
2738 }
2739
2740 static rfs4_lo_state_t *
2741 rfs4_findlo_state(stateid_t *id, bool_t lock_fp)
2742 {
2743 rfs4_lo_state_t *lsp;
2744 bool_t create = FALSE;
2745 nfs4_srv_t *nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
2746
2747 lsp = (rfs4_lo_state_t *)rfs4_dbsearch(nsrv4->rfs4_lo_state_idx, id,
2748 &create, NULL, RFS4_DBS_VALID);
2749 if (lock_fp == TRUE && lsp != NULL)
2750 rw_enter(&lsp->rls_state->rs_finfo->rf_file_rwlock, RW_READER);
2751
2752 return (lsp);
2753 }
2754
2755
2756 static uint32_t
2757 lo_state_lo_hash(void *key)
2758 {
2759 rfs4_lo_state_t *lsp = key;
2760
2761 return (ADDRHASH(lsp->rls_locker) ^ ADDRHASH(lsp->rls_state));
2762 }
2763
2764 static bool_t
2765 lo_state_lo_compare(rfs4_entry_t u_entry, void *key)
2766 {
2767 rfs4_lo_state_t *lsp = (rfs4_lo_state_t *)u_entry;
2768 rfs4_lo_state_t *keyp = key;
2769
2770 return (keyp->rls_locker == lsp->rls_locker &&
2771 keyp->rls_state == lsp->rls_state);
2772 }
2773
2774 static void *
2775 lo_state_lo_mkkey(rfs4_entry_t u_entry)
2776 {
2777 return (u_entry);
2778 }
2779
2780 rfs4_lo_state_t *
2781 rfs4_findlo_state_by_owner(rfs4_lockowner_t *lo, rfs4_state_t *sp,
2782 bool_t *create)
2783 {
2784 rfs4_lo_state_t *lsp;
2785 rfs4_lo_state_t arg;
2786 nfs4_srv_t *nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
2787
2788 arg.rls_locker = lo;
2789 arg.rls_state = sp;
2790
2791 lsp = (rfs4_lo_state_t *)rfs4_dbsearch(nsrv4->rfs4_lo_state_owner_idx,
2792 &arg, create, &arg, RFS4_DBS_VALID);
2793
2794 return (lsp);
2795 }
2796
2797 static stateid_t
2798 get_stateid(id_t eid)
2799 {
2800 stateid_t id;
2801 nfs4_srv_t *nsrv4;
2802
2803 nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
2804
2805 id.bits.boottime = nsrv4->rfs4_start_time;
2806 id.bits.ident = eid;
2807 id.bits.chgseq = 0;
2808 id.bits.type = 0;
2809 id.bits.pid = 0;
2810
2811 /*
2812 * If we are booted as a cluster node, embed our nodeid.
2813 * We've already done sanity checks in rfs4_client_create() so no
2814 * need to repeat them here.
2815 */
2816 id.bits.clnodeid = (cluster_bootflags & CLUSTER_BOOTED) ?
2817 clconf_get_nodeid() : 0;
2818
2819 return (id);
2820 }
2821
2822 /*
2823 * For use only when booted as a cluster node.
2824 * Returns TRUE if the embedded nodeid indicates that this stateid was
2825 * generated on another node.
2826 */
2827 static int
2828 foreign_stateid(stateid_t *id)
2829 {
2830 ASSERT(cluster_bootflags & CLUSTER_BOOTED);
2831 return (id->bits.clnodeid != (uint32_t)clconf_get_nodeid());
2832 }
2833
2834 /*
2835 * For use only when booted as a cluster node.
2836 * Returns TRUE if the embedded nodeid indicates that this clientid was
2837 * generated on another node.
2838 */
2839 static int
2840 foreign_clientid(cid *cidp)
2841 {
2842 ASSERT(cluster_bootflags & CLUSTER_BOOTED);
2843 return (cidp->impl_id.c_id >> CLUSTER_NODEID_SHIFT !=
2844 (uint32_t)clconf_get_nodeid());
2845 }
2846
2847 /*
2848 * For use only when booted as a cluster node.
2849 * Embed our cluster nodeid into the clientid.
2850 */
2851 static void
2852 embed_nodeid(cid *cidp)
2853 {
2854 int clnodeid;
2855 /*
2856 * Currently, our state tables are small enough that their
2857 * ids will leave enough bits free for the nodeid. If the
2858 * tables become larger, we mustn't overwrite the id.
2859 * Equally, we only have room for so many bits of nodeid, so
2860 * must check that too.
2861 */
2862 ASSERT(cluster_bootflags & CLUSTER_BOOTED);
2863 ASSERT(cidp->impl_id.c_id >> CLUSTER_NODEID_SHIFT == 0);
2864 clnodeid = clconf_get_nodeid();
2865 ASSERT(clnodeid <= CLUSTER_MAX_NODEID);
2866 ASSERT(clnodeid != NODEID_UNKNOWN);
2867 cidp->impl_id.c_id |= (clnodeid << CLUSTER_NODEID_SHIFT);
2868 }
2869
2870 static uint32_t
2871 state_hash(void *key)
2872 {
2873 stateid_t *ip = (stateid_t *)key;
2874
2875 return (ip->bits.ident);
2876 }
2877
2878 static bool_t
2879 state_compare(rfs4_entry_t u_entry, void *key)
2880 {
2881 rfs4_state_t *sp = (rfs4_state_t *)u_entry;
2882 stateid_t *id = (stateid_t *)key;
2883 bool_t rc;
2884
2885 rc = (sp->rs_stateid.bits.boottime == id->bits.boottime &&
2886 sp->rs_stateid.bits.ident == id->bits.ident);
2887
2888 return (rc);
2889 }
2890
2891 static void *
2892 state_mkkey(rfs4_entry_t u_entry)
2893 {
2894 rfs4_state_t *sp = (rfs4_state_t *)u_entry;
2895
2896 return (&sp->rs_stateid);
2897 }
2898
2899 static void
2900 rfs4_state_destroy(rfs4_entry_t u_entry)
2901 {
2902 rfs4_state_t *sp = (rfs4_state_t *)u_entry;
2903
2904 /* remove from openowner list */
2905 rfs4_dbe_lock(sp->rs_owner->ro_dbe);
2906 list_remove(&sp->rs_owner->ro_statelist, sp);
2907 rfs4_dbe_unlock(sp->rs_owner->ro_dbe);
2908
2909 list_destroy(&sp->rs_lostatelist);
2910
2911 /* release any share locks for this stateid if it's still open */
2912 if (!sp->rs_closed) {
2913 rfs4_dbe_lock(sp->rs_dbe);
2914 (void) rfs4_unshare(sp);
2915 rfs4_dbe_unlock(sp->rs_dbe);
2916 }
2917
2918 /* Were done with the file */
2919 rfs4_file_rele(sp->rs_finfo);
2920 sp->rs_finfo = NULL;
2921
2922 /* And now with the openowner */
2923 rfs4_openowner_rele(sp->rs_owner);
2924 sp->rs_owner = NULL;
2925 }
2926
2927 static void
2928 rfs4_state_rele_nounlock(rfs4_state_t *sp)
2929 {
2930 rfs4_dbe_rele(sp->rs_dbe);
2931 }
2932
2933 void
2934 rfs4_state_rele(rfs4_state_t *sp)
2935 {
2936 rw_exit(&sp->rs_finfo->rf_file_rwlock);
2937 rfs4_dbe_rele(sp->rs_dbe);
2938 }
2939
2940 static uint32_t
2941 deleg_hash(void *key)
2942 {
2943 rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)key;
2944
2945 return (ADDRHASH(dsp->rds_client) ^ ADDRHASH(dsp->rds_finfo));
2946 }
2947
2948 static bool_t
2949 deleg_compare(rfs4_entry_t u_entry, void *key)
2950 {
2951 rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
2952 rfs4_deleg_state_t *kdsp = (rfs4_deleg_state_t *)key;
2953
2954 return (dsp->rds_client == kdsp->rds_client &&
2955 dsp->rds_finfo == kdsp->rds_finfo);
2956 }
2957
2958 static void *
2959 deleg_mkkey(rfs4_entry_t u_entry)
2960 {
2961 return (u_entry);
2962 }
2963
2964 static uint32_t
2965 deleg_state_hash(void *key)
2966 {
2967 stateid_t *ip = (stateid_t *)key;
2968
2969 return (ip->bits.ident);
2970 }
2971
2972 static bool_t
2973 deleg_state_compare(rfs4_entry_t u_entry, void *key)
2974 {
2975 rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
2976 stateid_t *id = (stateid_t *)key;
2977 bool_t rc;
2978
2979 if (id->bits.type != DELEGID)
2980 return (FALSE);
2981
2982 rc = (dsp->rds_delegid.bits.boottime == id->bits.boottime &&
2983 dsp->rds_delegid.bits.ident == id->bits.ident);
2984
2985 return (rc);
2986 }
2987
2988 static void *
2989 deleg_state_mkkey(rfs4_entry_t u_entry)
2990 {
2991 rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
2992
2993 return (&dsp->rds_delegid);
2994 }
2995
2996 static bool_t
2997 rfs4_deleg_state_expiry(rfs4_entry_t u_entry)
2998 {
2999 rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
3000
3001 if (rfs4_dbe_is_invalid(dsp->rds_dbe))
3002 return (TRUE);
3003
3004 if (dsp->rds_dtype == OPEN_DELEGATE_NONE)
3005 return (TRUE);
3006
3007 if ((gethrestime_sec() - dsp->rds_client->rc_last_access
3008 > rfs4_lease_time)) {
3009 rfs4_dbe_invalidate(dsp->rds_dbe);
3010 return (TRUE);
3011 }
3012
3013 return (FALSE);
3014 }
3015
3016 static bool_t
3017 rfs4_deleg_state_create(rfs4_entry_t u_entry, void *argp)
3018 {
3019 rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
3020 rfs4_file_t *fp = ((rfs4_deleg_state_t *)argp)->rds_finfo;
3021 rfs4_client_t *cp = ((rfs4_deleg_state_t *)argp)->rds_client;
3022
3023 rfs4_dbe_hold(fp->rf_dbe);
3024 rfs4_dbe_hold(cp->rc_dbe);
3025
3026 dsp->rds_delegid = get_stateid(rfs4_dbe_getid(dsp->rds_dbe));
3027 dsp->rds_delegid.bits.type = DELEGID;
3028 dsp->rds_finfo = fp;
3029 dsp->rds_client = cp;
3030 dsp->rds_dtype = OPEN_DELEGATE_NONE;
3031
3032 dsp->rds_time_granted = gethrestime_sec(); /* observability */
3033 dsp->rds_time_revoked = 0;
3034
3035 list_link_init(&dsp->rds_node);
3036
3037 return (TRUE);
3038 }
3039
3040 static void
3041 rfs4_deleg_state_destroy(rfs4_entry_t u_entry)
3042 {
3043 rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
3044
3045 /* return delegation if necessary */
3046 rfs4_return_deleg(dsp, FALSE);
3047
3048 /* Were done with the file */
3049 rfs4_file_rele(dsp->rds_finfo);
3050 dsp->rds_finfo = NULL;
3051
3052 /* And now with the openowner */
3053 rfs4_client_rele(dsp->rds_client);
3054 dsp->rds_client = NULL;
3055 }
3056
3057 rfs4_deleg_state_t *
3058 rfs4_finddeleg(rfs4_state_t *sp, bool_t *create)
3059 {
3060 rfs4_deleg_state_t ds, *dsp;
3061 nfs4_srv_t *nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
3062
3063 ds.rds_client = sp->rs_owner->ro_client;
3064 ds.rds_finfo = sp->rs_finfo;
3065
3066 dsp = (rfs4_deleg_state_t *)rfs4_dbsearch(nsrv4->rfs4_deleg_idx, &ds,
3067 create, &ds, RFS4_DBS_VALID);
3068
3069 return (dsp);
3070 }
3071
3072 rfs4_deleg_state_t *
3073 rfs4_finddelegstate(stateid_t *id)
3074 {
3075 rfs4_deleg_state_t *dsp;
3076 bool_t create = FALSE;
3077 nfs4_srv_t *nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
3078
3079 dsp = (rfs4_deleg_state_t *)rfs4_dbsearch(nsrv4->rfs4_deleg_state_idx,
3080 id, &create, NULL, RFS4_DBS_VALID);
3081
3082 return (dsp);
3083 }
3084
3085 void
3086 rfs4_deleg_state_rele(rfs4_deleg_state_t *dsp)
3087 {
3088 rfs4_dbe_rele(dsp->rds_dbe);
3089 }
3090
3091 void
3092 rfs4_update_lock_sequence(rfs4_lo_state_t *lsp)
3093 {
3094
3095 rfs4_dbe_lock(lsp->rls_dbe);
3096
3097 /*
3098 * If we are skipping sequence id checking, this means that
3099 * this is the first lock request and therefore the sequence
3100 * id does not need to be updated. This only happens on the
3101 * first lock request for a lockowner
3102 */
3103 if (!lsp->rls_skip_seqid_check)
3104 lsp->rls_seqid++;
3105
3106 rfs4_dbe_unlock(lsp->rls_dbe);
3107 }
3108
3109 void
3110 rfs4_update_lock_resp(rfs4_lo_state_t *lsp, nfs_resop4 *resp)
3111 {
3112
3113 rfs4_dbe_lock(lsp->rls_dbe);
3114
3115 rfs4_free_reply(&lsp->rls_reply);
3116
3117 rfs4_copy_reply(&lsp->rls_reply, resp);
3118
3119 rfs4_dbe_unlock(lsp->rls_dbe);
3120 }
3121
3122 void
3123 rfs4_free_opens(rfs4_openowner_t *oo, bool_t invalidate,
3124 bool_t close_of_client)
3125 {
3126 rfs4_state_t *sp;
3127
3128 rfs4_dbe_lock(oo->ro_dbe);
3129
3130 for (sp = list_head(&oo->ro_statelist); sp != NULL;
3131 sp = list_next(&oo->ro_statelist, sp)) {
3132 rfs4_state_close(sp, FALSE, close_of_client, CRED());
3133 if (invalidate == TRUE)
3134 rfs4_dbe_invalidate(sp->rs_dbe);
3135 }
3136
3137 rfs4_dbe_invalidate(oo->ro_dbe);
3138 rfs4_dbe_unlock(oo->ro_dbe);
3139 }
3140
3141 static uint32_t
3142 state_owner_file_hash(void *key)
3143 {
3144 rfs4_state_t *sp = key;
3145
3146 return (ADDRHASH(sp->rs_owner) ^ ADDRHASH(sp->rs_finfo));
3147 }
3148
3149 static bool_t
3150 state_owner_file_compare(rfs4_entry_t u_entry, void *key)
3151 {
3152 rfs4_state_t *sp = (rfs4_state_t *)u_entry;
3153 rfs4_state_t *arg = key;
3154
3155 if (sp->rs_closed == TRUE)
3156 return (FALSE);
3157
3158 return (arg->rs_owner == sp->rs_owner && arg->rs_finfo == sp->rs_finfo);
3159 }
3160
3161 static void *
3162 state_owner_file_mkkey(rfs4_entry_t u_entry)
3163 {
3164 return (u_entry);
3165 }
3166
3167 static uint32_t
3168 state_file_hash(void *key)
3169 {
3170 return (ADDRHASH(key));
3171 }
3172
3173 static bool_t
3174 state_file_compare(rfs4_entry_t u_entry, void *key)
3175 {
3176 rfs4_state_t *sp = (rfs4_state_t *)u_entry;
3177 rfs4_file_t *fp = key;
3178
3179 if (sp->rs_closed == TRUE)
3180 return (FALSE);
3181
3182 return (fp == sp->rs_finfo);
3183 }
3184
3185 static void *
3186 state_file_mkkey(rfs4_entry_t u_entry)
3187 {
3188 rfs4_state_t *sp = (rfs4_state_t *)u_entry;
3189
3190 return (sp->rs_finfo);
3191 }
3192
3193 rfs4_state_t *
3194 rfs4_findstate_by_owner_file(rfs4_openowner_t *oo, rfs4_file_t *fp,
3195 bool_t *create)
3196 {
3197 rfs4_state_t *sp;
3198 rfs4_state_t key;
3199 nfs4_srv_t *nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
3200
3201 key.rs_owner = oo;
3202 key.rs_finfo = fp;
3203
3204 sp = (rfs4_state_t *)rfs4_dbsearch(nsrv4->rfs4_state_owner_file_idx,
3205 &key, create, &key, RFS4_DBS_VALID);
3206
3207 return (sp);
3208 }
3209
3210 /* This returns ANY state struct that refers to this file */
3211 static rfs4_state_t *
3212 rfs4_findstate_by_file(rfs4_file_t *fp)
3213 {
3214 bool_t create = FALSE;
3215 nfs4_srv_t *nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
3216
3217 return ((rfs4_state_t *)rfs4_dbsearch(nsrv4->rfs4_state_file_idx, fp,
3218 &create, fp, RFS4_DBS_VALID));
3219 }
3220
3221 static bool_t
3222 rfs4_state_expiry(rfs4_entry_t u_entry)
3223 {
3224 rfs4_state_t *sp = (rfs4_state_t *)u_entry;
3225
3226 if (rfs4_dbe_is_invalid(sp->rs_dbe))
3227 return (TRUE);
3228
3229 if (sp->rs_closed == TRUE &&
3230 ((gethrestime_sec() - rfs4_dbe_get_timerele(sp->rs_dbe))
3231 > rfs4_lease_time))
3232 return (TRUE);
3233
3234 return ((gethrestime_sec() - sp->rs_owner->ro_client->rc_last_access
3235 > rfs4_lease_time));
3236 }
3237
3238 static bool_t
3239 rfs4_state_create(rfs4_entry_t u_entry, void *argp)
3240 {
3241 rfs4_state_t *sp = (rfs4_state_t *)u_entry;
3242 rfs4_file_t *fp = ((rfs4_state_t *)argp)->rs_finfo;
3243 rfs4_openowner_t *oo = ((rfs4_state_t *)argp)->rs_owner;
3244
3245 rfs4_dbe_hold(fp->rf_dbe);
3246 rfs4_dbe_hold(oo->ro_dbe);
3247 sp->rs_stateid = get_stateid(rfs4_dbe_getid(sp->rs_dbe));
3248 sp->rs_stateid.bits.type = OPENID;
3249 sp->rs_owner = oo;
3250 sp->rs_finfo = fp;
3251
3252 list_create(&sp->rs_lostatelist, sizeof (rfs4_lo_state_t),
3253 offsetof(rfs4_lo_state_t, rls_node));
3254
3255 /* Insert state on per open owner's list */
3256 rfs4_dbe_lock(oo->ro_dbe);
3257 list_insert_tail(&oo->ro_statelist, sp);
3258 rfs4_dbe_unlock(oo->ro_dbe);
3259
3260 return (TRUE);
3261 }
3262
3263 static rfs4_state_t *
3264 rfs4_findstate(stateid_t *id, rfs4_dbsearch_type_t find_invalid, bool_t lock_fp)
3265 {
3266 rfs4_state_t *sp;
3267 bool_t create = FALSE;
3268 nfs4_srv_t *nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
3269
3270 sp = (rfs4_state_t *)rfs4_dbsearch(nsrv4->rfs4_state_idx, id,
3271 &create, NULL, find_invalid);
3272 if (lock_fp == TRUE && sp != NULL)
3273 rw_enter(&sp->rs_finfo->rf_file_rwlock, RW_READER);
3274
3275 return (sp);
3276 }
3277
3278 void
3279 rfs4_state_close(rfs4_state_t *sp, bool_t lock_held, bool_t close_of_client,
3280 cred_t *cr)
3281 {
3282 /* Remove the associated lo_state owners */
3283 if (!lock_held)
3284 rfs4_dbe_lock(sp->rs_dbe);
3285
3286 /*
3287 * If refcnt == 0, the dbe is about to be destroyed.
3288 * lock state will be released by the reaper thread.
3289 */
3290
3291 if (rfs4_dbe_refcnt(sp->rs_dbe) > 0) {
3292 if (sp->rs_closed == FALSE) {
3293 rfs4_release_share_lock_state(sp, cr, close_of_client);
3294 sp->rs_closed = TRUE;
3295 }
3296 }
3297
3298 if (!lock_held)
3299 rfs4_dbe_unlock(sp->rs_dbe);
3300 }
3301
3302 /*
3303 * Remove all state associated with the given client.
3304 */
3305 void
3306 rfs4_client_state_remove(rfs4_client_t *cp)
3307 {
3308 rfs4_openowner_t *oo;
3309
3310 rfs4_dbe_lock(cp->rc_dbe);
3311
3312 for (oo = list_head(&cp->rc_openownerlist); oo != NULL;
3313 oo = list_next(&cp->rc_openownerlist, oo)) {
3314 rfs4_free_opens(oo, TRUE, TRUE);
3315 }
3316
3317 rfs4_dbe_unlock(cp->rc_dbe);
3318 }
3319
3320 void
3321 rfs4_client_close(rfs4_client_t *cp)
3322 {
3323 /* Mark client as going away. */
3324 rfs4_dbe_lock(cp->rc_dbe);
3325 rfs4_dbe_invalidate(cp->rc_dbe);
3326 rfs4_dbe_unlock(cp->rc_dbe);
3327
3328 rfs4_client_state_remove(cp);
3329
3330 /* Release the client */
3331 rfs4_client_rele(cp);
3332 }
3333
3334 nfsstat4
3335 rfs4_check_clientid(clientid4 *cp, int setclid_confirm)
3336 {
3337 cid *cidp = (cid *) cp;
3338 nfs4_srv_t *nsrv4;
3339
3340 nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
3341
3342 /*
3343 * If we are booted as a cluster node, check the embedded nodeid.
3344 * If it indicates that this clientid was generated on another node,
3345 * inform the client accordingly.
3346 */
3347 if (cluster_bootflags & CLUSTER_BOOTED && foreign_clientid(cidp))
3348 return (NFS4ERR_STALE_CLIENTID);
3349
3350 /*
3351 * If the server start time matches the time provided
3352 * by the client (via the clientid) and this is NOT a
3353 * setclientid_confirm then return EXPIRED.
3354 */
3355 if (!setclid_confirm &&
3356 cidp->impl_id.start_time == nsrv4->rfs4_start_time)
3357 return (NFS4ERR_EXPIRED);
3358
3359 return (NFS4ERR_STALE_CLIENTID);
3360 }
3361
3362 /*
3363 * This is used when a stateid has not been found amongst the
3364 * current server's state. Check the stateid to see if it
3365 * was from this server instantiation or not.
3366 */
3367 static nfsstat4
3368 what_stateid_error(stateid_t *id, stateid_type_t type)
3369 {
3370 nfs4_srv_t *nsrv4;
3371
3372 nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
3373
3374 /* If we are booted as a cluster node, was stateid locally generated? */
3375 if ((cluster_bootflags & CLUSTER_BOOTED) && foreign_stateid(id))
3376 return (NFS4ERR_STALE_STATEID);
3377
3378 /* If types don't match then no use checking further */
3379 if (type != id->bits.type)
3380 return (NFS4ERR_BAD_STATEID);
3381
3382 /* From a different server instantiation, return STALE */
3383 if (id->bits.boottime != nsrv4->rfs4_start_time)
3384 return (NFS4ERR_STALE_STATEID);
3385
3386 /*
3387 * From this server but the state is most likely beyond lease
3388 * timeout: return NFS4ERR_EXPIRED. However, there is the
3389 * case of a delegation stateid. For delegations, there is a
3390 * case where the state can be removed without the client's
3391 * knowledge/consent: revocation. In the case of delegation
3392 * revocation, the delegation state will be removed and will
3393 * not be found. If the client does something like a
3394 * DELEGRETURN or even a READ/WRITE with a delegatoin stateid
3395 * that has been revoked, the server should return BAD_STATEID
3396 * instead of the more common EXPIRED error.
3397 */
3398 if (id->bits.boottime == nsrv4->rfs4_start_time) {
3399 if (type == DELEGID)
3400 return (NFS4ERR_BAD_STATEID);
3401 else
3402 return (NFS4ERR_EXPIRED);
3403 }
3404
3405 return (NFS4ERR_BAD_STATEID);
3406 }
3407
3408 /*
3409 * Used later on to find the various state structs. When called from
3410 * rfs4_check_stateid()->rfs4_get_all_state(), no file struct lock is
3411 * taken (it is not needed) and helps on the read/write path with
3412 * respect to performance.
3413 */
3414 static nfsstat4
3415 rfs4_get_state_lockit(stateid4 *stateid, rfs4_state_t **spp,
3416 rfs4_dbsearch_type_t find_invalid, bool_t lock_fp)
3417 {
3418 stateid_t *id = (stateid_t *)stateid;
3419 rfs4_state_t *sp;
3420
3421 *spp = NULL;
3422
3423 /* If we are booted as a cluster node, was stateid locally generated? */
3424 if ((cluster_bootflags & CLUSTER_BOOTED) && foreign_stateid(id))
3425 return (NFS4ERR_STALE_STATEID);
3426
3427 sp = rfs4_findstate(id, find_invalid, lock_fp);
3428 if (sp == NULL) {
3429 return (what_stateid_error(id, OPENID));
3430 }
3431
3432 if (rfs4_lease_expired(sp->rs_owner->ro_client)) {
3433 if (lock_fp == TRUE)
3434 rfs4_state_rele(sp);
3435 else
3436 rfs4_state_rele_nounlock(sp);
3437 return (NFS4ERR_EXPIRED);
3438 }
3439
3440 *spp = sp;
3441
3442 return (NFS4_OK);
3443 }
3444
3445 nfsstat4
3446 rfs4_get_state(stateid4 *stateid, rfs4_state_t **spp,
3447 rfs4_dbsearch_type_t find_invalid)
3448 {
3449 return (rfs4_get_state_lockit(stateid, spp, find_invalid, TRUE));
3450 }
3451
3452 int
3453 rfs4_check_stateid_seqid(rfs4_state_t *sp, stateid4 *stateid)
3454 {
3455 stateid_t *id = (stateid_t *)stateid;
3456
3457 if (rfs4_lease_expired(sp->rs_owner->ro_client))
3458 return (NFS4_CHECK_STATEID_EXPIRED);
3459
3460 /* Stateid is some time in the future - that's bad */
3461 if (sp->rs_stateid.bits.chgseq < id->bits.chgseq)
3462 return (NFS4_CHECK_STATEID_BAD);
3463
3464 if (sp->rs_stateid.bits.chgseq == id->bits.chgseq + 1)
3465 return (NFS4_CHECK_STATEID_REPLAY);
3466
3467 /* Stateid is some time in the past - that's old */
3468 if (sp->rs_stateid.bits.chgseq > id->bits.chgseq)
3469 return (NFS4_CHECK_STATEID_OLD);
3470
3471 /* Caller needs to know about confirmation before closure */
3472 if (sp->rs_owner->ro_need_confirm)
3473 return (NFS4_CHECK_STATEID_UNCONFIRMED);
3474
3475 if (sp->rs_closed == TRUE)
3476 return (NFS4_CHECK_STATEID_CLOSED);
3477
3478 return (NFS4_CHECK_STATEID_OKAY);
3479 }
3480
3481 int
3482 rfs4_check_lo_stateid_seqid(rfs4_lo_state_t *lsp, stateid4 *stateid)
3483 {
3484 stateid_t *id = (stateid_t *)stateid;
3485
3486 if (rfs4_lease_expired(lsp->rls_state->rs_owner->ro_client))
3487 return (NFS4_CHECK_STATEID_EXPIRED);
3488
3489 /* Stateid is some time in the future - that's bad */
3490 if (lsp->rls_lockid.bits.chgseq < id->bits.chgseq)
3491 return (NFS4_CHECK_STATEID_BAD);
3492
3493 if (lsp->rls_lockid.bits.chgseq == id->bits.chgseq + 1)
3494 return (NFS4_CHECK_STATEID_REPLAY);
3495
3496 /* Stateid is some time in the past - that's old */
3497 if (lsp->rls_lockid.bits.chgseq > id->bits.chgseq)
3498 return (NFS4_CHECK_STATEID_OLD);
3499
3500 if (lsp->rls_state->rs_closed == TRUE)
3501 return (NFS4_CHECK_STATEID_CLOSED);
3502
3503 return (NFS4_CHECK_STATEID_OKAY);
3504 }
3505
3506 nfsstat4
3507 rfs4_get_deleg_state(stateid4 *stateid, rfs4_deleg_state_t **dspp)
3508 {
3509 stateid_t *id = (stateid_t *)stateid;
3510 rfs4_deleg_state_t *dsp;
3511
3512 *dspp = NULL;
3513
3514 /* If we are booted as a cluster node, was stateid locally generated? */
3515 if ((cluster_bootflags & CLUSTER_BOOTED) && foreign_stateid(id))
3516 return (NFS4ERR_STALE_STATEID);
3517
3518 dsp = rfs4_finddelegstate(id);
3519 if (dsp == NULL) {
3520 return (what_stateid_error(id, DELEGID));
3521 }
3522
3523 if (rfs4_lease_expired(dsp->rds_client)) {
3524 rfs4_deleg_state_rele(dsp);
3525 return (NFS4ERR_EXPIRED);
3526 }
3527
3528 *dspp = dsp;
3529
3530 return (NFS4_OK);
3531 }
3532
3533 nfsstat4
3534 rfs4_get_lo_state(stateid4 *stateid, rfs4_lo_state_t **lspp, bool_t lock_fp)
3535 {
3536 stateid_t *id = (stateid_t *)stateid;
3537 rfs4_lo_state_t *lsp;
3538
3539 *lspp = NULL;
3540
3541 /* If we are booted as a cluster node, was stateid locally generated? */
3542 if ((cluster_bootflags & CLUSTER_BOOTED) && foreign_stateid(id))
3543 return (NFS4ERR_STALE_STATEID);
3544
3545 lsp = rfs4_findlo_state(id, lock_fp);
3546 if (lsp == NULL) {
3547 return (what_stateid_error(id, LOCKID));
3548 }
3549
3550 if (rfs4_lease_expired(lsp->rls_state->rs_owner->ro_client)) {
3551 rfs4_lo_state_rele(lsp, lock_fp);
3552 return (NFS4ERR_EXPIRED);
3553 }
3554
3555 *lspp = lsp;
3556
3557 return (NFS4_OK);
3558 }
3559
3560 static nfsstat4
3561 rfs4_get_all_state(stateid4 *sid, rfs4_state_t **spp,
3562 rfs4_deleg_state_t **dspp, rfs4_lo_state_t **lspp)
3563 {
3564 rfs4_state_t *sp = NULL;
3565 rfs4_deleg_state_t *dsp = NULL;
3566 rfs4_lo_state_t *lsp = NULL;
3567 stateid_t *id;
3568 nfsstat4 status;
3569
3570 *spp = NULL; *dspp = NULL; *lspp = NULL;
3571
3572 id = (stateid_t *)sid;
3573 switch (id->bits.type) {
3574 case OPENID:
3575 status = rfs4_get_state_lockit(sid, &sp, FALSE, FALSE);
3576 break;
3577 case DELEGID:
3578 status = rfs4_get_deleg_state(sid, &dsp);
3579 break;
3580 case LOCKID:
3581 status = rfs4_get_lo_state(sid, &lsp, FALSE);
3582 if (status == NFS4_OK) {
3583 sp = lsp->rls_state;
3584 rfs4_dbe_hold(sp->rs_dbe);
3585 }
3586 break;
3587 default:
3588 status = NFS4ERR_BAD_STATEID;
3589 }
3590
3591 if (status == NFS4_OK) {
3592 *spp = sp;
3593 *dspp = dsp;
3594 *lspp = lsp;
3595 }
3596
3597 return (status);
3598 }
3599
3600 /*
3601 * Given the I/O mode (FREAD or FWRITE), this checks whether the
3602 * rfs4_state_t struct has access to do this operation and if so
3603 * return NFS4_OK; otherwise the proper NFSv4 error is returned.
3604 */
3605 nfsstat4
3606 rfs4_state_has_access(rfs4_state_t *sp, int mode, vnode_t *vp)
3607 {
3608 nfsstat4 stat = NFS4_OK;
3609 rfs4_file_t *fp;
3610 bool_t create = FALSE;
3611
3612 rfs4_dbe_lock(sp->rs_dbe);
3613 if (mode == FWRITE) {
3614 if (!(sp->rs_share_access & OPEN4_SHARE_ACCESS_WRITE)) {
3615 stat = NFS4ERR_OPENMODE;
3616 }
3617 } else if (mode == FREAD) {
3618 if (!(sp->rs_share_access & OPEN4_SHARE_ACCESS_READ)) {
3619 /*
3620 * If we have OPENed the file with DENYing access
3621 * to both READ and WRITE then no one else could
3622 * have OPENed the file, hence no conflicting READ
3623 * deny. This check is merely an optimization.
3624 */
3625 if (sp->rs_share_deny == OPEN4_SHARE_DENY_BOTH)
3626 goto out;
3627
3628 /* Check against file struct's DENY mode */
3629 fp = rfs4_findfile(vp, NULL, &create);
3630 if (fp != NULL) {
3631 int deny_read = 0;
3632 rfs4_dbe_lock(fp->rf_dbe);
3633 /*
3634 * Check if any other open owner has the file
3635 * OPENed with deny READ.
3636 */
3637 if (sp->rs_share_deny & OPEN4_SHARE_DENY_READ)
3638 deny_read = 1;
3639 ASSERT(fp->rf_deny_read >= deny_read);
3640 if (fp->rf_deny_read > deny_read)
3641 stat = NFS4ERR_OPENMODE;
3642 rfs4_dbe_unlock(fp->rf_dbe);
3643 rfs4_file_rele(fp);
3644 }
3645 }
3646 } else {
3647 /* Illegal I/O mode */
3648 stat = NFS4ERR_INVAL;
3649 }
3650 out:
3651 rfs4_dbe_unlock(sp->rs_dbe);
3652 return (stat);
3653 }
3654
3655 /*
3656 * Given the I/O mode (FREAD or FWRITE), the vnode, the stateid and whether
3657 * the file is being truncated, return NFS4_OK if allowed or appropriate
3658 * V4 error if not. Note NFS4ERR_DELAY will be returned and a recall on
3659 * the associated file will be done if the I/O is not consistent with any
3660 * delegation in effect on the file. Should be holding VOP_RWLOCK, either
3661 * as reader or writer as appropriate. rfs4_op_open will acquire the
3662 * VOP_RWLOCK as writer when setting up delegation. If the stateid is bad
3663 * this routine will return NFS4ERR_BAD_STATEID. In addition, through the
3664 * deleg parameter, we will return whether a write delegation is held by
3665 * the client associated with this stateid.
3666 * If the server instance associated with the relevant client is in its
3667 * grace period, return NFS4ERR_GRACE.
3668 */
3669
3670 nfsstat4
3671 rfs4_check_stateid(int mode, vnode_t *vp,
3672 stateid4 *stateid, bool_t trunc, bool_t *deleg,
3673 bool_t do_access, caller_context_t *ct)
3674 {
3675 rfs4_file_t *fp;
3676 bool_t create = FALSE;
3677 rfs4_state_t *sp;
3678 rfs4_deleg_state_t *dsp;
3679 rfs4_lo_state_t *lsp;
3680 stateid_t *id = (stateid_t *)stateid;
3681 nfsstat4 stat = NFS4_OK;
3682
3683 if (ct != NULL) {
3684 ct->cc_sysid = 0;
3685 ct->cc_pid = 0;
3686 ct->cc_caller_id = nfs4_srv_caller_id;
3687 ct->cc_flags = CC_DONTBLOCK;
3688 }
3689
3690 if (ISSPECIAL(stateid)) {
3691 fp = rfs4_findfile(vp, NULL, &create);
3692 if (fp == NULL)
3693 return (NFS4_OK);
3694 if (fp->rf_dinfo.rd_dtype == OPEN_DELEGATE_NONE) {
3695 rfs4_file_rele(fp);
3696 return (NFS4_OK);
3697 }
3698 if (mode == FWRITE ||
3699 fp->rf_dinfo.rd_dtype == OPEN_DELEGATE_WRITE) {
3700 rfs4_recall_deleg(fp, trunc, NULL);
3701 rfs4_file_rele(fp);
3702 return (NFS4ERR_DELAY);
3703 }
3704 rfs4_file_rele(fp);
3705 return (NFS4_OK);
3706 } else {
3707 stat = rfs4_get_all_state(stateid, &sp, &dsp, &lsp);
3708 if (stat != NFS4_OK)
3709 return (stat);
3710 if (lsp != NULL) {
3711 /* Is associated server instance in its grace period? */
3712 if (rfs4_clnt_in_grace(lsp->rls_locker->rl_client)) {
3713 rfs4_lo_state_rele(lsp, FALSE);
3714 if (sp != NULL)
3715 rfs4_state_rele_nounlock(sp);
3716 return (NFS4ERR_GRACE);
3717 }
3718 if (id->bits.type == LOCKID) {
3719 /* Seqid in the future? - that's bad */
3720 if (lsp->rls_lockid.bits.chgseq <
3721 id->bits.chgseq) {
3722 rfs4_lo_state_rele(lsp, FALSE);
3723 if (sp != NULL)
3724 rfs4_state_rele_nounlock(sp);
3725 return (NFS4ERR_BAD_STATEID);
3726 }
3727 /* Seqid in the past? - that's old */
3728 if (lsp->rls_lockid.bits.chgseq >
3729 id->bits.chgseq) {
3730 rfs4_lo_state_rele(lsp, FALSE);
3731 if (sp != NULL)
3732 rfs4_state_rele_nounlock(sp);
3733 return (NFS4ERR_OLD_STATEID);
3734 }
3735 /* Ensure specified filehandle matches */
3736 if (lsp->rls_state->rs_finfo->rf_vp != vp) {
3737 rfs4_lo_state_rele(lsp, FALSE);
3738 if (sp != NULL)
3739 rfs4_state_rele_nounlock(sp);
3740 return (NFS4ERR_BAD_STATEID);
3741 }
3742 }
3743 if (ct != NULL) {
3744 ct->cc_sysid =
3745 lsp->rls_locker->rl_client->rc_sysidt;
3746 ct->cc_pid = lsp->rls_locker->rl_pid;
3747 }
3748 rfs4_lo_state_rele(lsp, FALSE);
3749 }
3750
3751 /* Stateid provided was an "open" stateid */
3752 if (sp != NULL) {
3753 /* Is associated server instance in its grace period? */
3754 if (rfs4_clnt_in_grace(sp->rs_owner->ro_client)) {
3755 rfs4_state_rele_nounlock(sp);
3756 return (NFS4ERR_GRACE);
3757 }
3758 if (id->bits.type == OPENID) {
3759 /* Seqid in the future? - that's bad */
3760 if (sp->rs_stateid.bits.chgseq <
3761 id->bits.chgseq) {
3762 rfs4_state_rele_nounlock(sp);
3763 return (NFS4ERR_BAD_STATEID);
3764 }
3765 /* Seqid in the past - that's old */
3766 if (sp->rs_stateid.bits.chgseq >
3767 id->bits.chgseq) {
3768 rfs4_state_rele_nounlock(sp);
3769 return (NFS4ERR_OLD_STATEID);
3770 }
3771 }
3772 /* Ensure specified filehandle matches */
3773 if (sp->rs_finfo->rf_vp != vp) {
3774 rfs4_state_rele_nounlock(sp);
3775 return (NFS4ERR_BAD_STATEID);
3776 }
3777
3778 if (sp->rs_owner->ro_need_confirm) {
3779 rfs4_state_rele_nounlock(sp);
3780 return (NFS4ERR_BAD_STATEID);
3781 }
3782
3783 if (sp->rs_closed == TRUE) {
3784 rfs4_state_rele_nounlock(sp);
3785 return (NFS4ERR_OLD_STATEID);
3786 }
3787
3788 if (do_access)
3789 stat = rfs4_state_has_access(sp, mode, vp);
3790 else
3791 stat = NFS4_OK;
3792
3793 /*
3794 * Return whether this state has write
3795 * delegation if desired
3796 */
3797 if (deleg && (sp->rs_finfo->rf_dinfo.rd_dtype ==
3798 OPEN_DELEGATE_WRITE))
3799 *deleg = TRUE;
3800
3801 /*
3802 * We got a valid stateid, so we update the
3803 * lease on the client. Ideally we would like
3804 * to do this after the calling op succeeds,
3805 * but for now this will be good
3806 * enough. Callers of this routine are
3807 * currently insulated from the state stuff.
3808 */
3809 rfs4_update_lease(sp->rs_owner->ro_client);
3810
3811 /*
3812 * If a delegation is present on this file and
3813 * this is a WRITE, then update the lastwrite
3814 * time to indicate that activity is present.
3815 */
3816 if (sp->rs_finfo->rf_dinfo.rd_dtype ==
3817 OPEN_DELEGATE_WRITE &&
3818 mode == FWRITE) {
3819 sp->rs_finfo->rf_dinfo.rd_time_lastwrite =
3820 gethrestime_sec();
3821 }
3822
3823 rfs4_state_rele_nounlock(sp);
3824
3825 return (stat);
3826 }
3827
3828 if (dsp != NULL) {
3829 /* Is associated server instance in its grace period? */
3830 if (rfs4_clnt_in_grace(dsp->rds_client)) {
3831 rfs4_deleg_state_rele(dsp);
3832 return (NFS4ERR_GRACE);
3833 }
3834 if (dsp->rds_delegid.bits.chgseq != id->bits.chgseq) {
3835 rfs4_deleg_state_rele(dsp);
3836 return (NFS4ERR_BAD_STATEID);
3837 }
3838
3839 /* Ensure specified filehandle matches */
3840 if (dsp->rds_finfo->rf_vp != vp) {
3841 rfs4_deleg_state_rele(dsp);
3842 return (NFS4ERR_BAD_STATEID);
3843 }
3844 /*
3845 * Return whether this state has write
3846 * delegation if desired
3847 */
3848 if (deleg && (dsp->rds_finfo->rf_dinfo.rd_dtype ==
3849 OPEN_DELEGATE_WRITE))
3850 *deleg = TRUE;
3851
3852 rfs4_update_lease(dsp->rds_client);
3853
3854 /*
3855 * If a delegation is present on this file and
3856 * this is a WRITE, then update the lastwrite
3857 * time to indicate that activity is present.
3858 */
3859 if (dsp->rds_finfo->rf_dinfo.rd_dtype ==
3860 OPEN_DELEGATE_WRITE && mode == FWRITE) {
3861 dsp->rds_finfo->rf_dinfo.rd_time_lastwrite =
3862 gethrestime_sec();
3863 }
3864
3865 /*
3866 * XXX - what happens if this is a WRITE and the
3867 * delegation type of for READ.
3868 */
3869 rfs4_deleg_state_rele(dsp);
3870
3871 return (stat);
3872 }
3873 /*
3874 * If we got this far, something bad happened
3875 */
3876 return (NFS4ERR_BAD_STATEID);
3877 }
3878 }
3879
3880
3881 /*
3882 * This is a special function in that for the file struct provided the
3883 * server wants to remove/close all current state associated with the
3884 * file. The prime use of this would be with OP_REMOVE to force the
3885 * release of state and particularly of file locks.
3886 *
3887 * There is an assumption that there is no delegations outstanding on
3888 * this file at this point. The caller should have waited for those
3889 * to be returned or revoked.
3890 */
3891 void
3892 rfs4_close_all_state(rfs4_file_t *fp)
3893 {
3894 rfs4_state_t *sp;
3895
3896 rfs4_dbe_lock(fp->rf_dbe);
3897
3898 #ifdef DEBUG
3899 /* only applies when server is handing out delegations */
3900 if (nfs4_get_deleg_policy() != SRV_NEVER_DELEGATE)
3901 ASSERT(fp->rf_dinfo.rd_hold_grant > 0);
3902 #endif
3903
3904 /* No delegations for this file */
3905 ASSERT(list_is_empty(&fp->rf_delegstatelist));
3906
3907 /* Make sure that it can not be found */
3908 rfs4_dbe_invalidate(fp->rf_dbe);
3909
3910 if (fp->rf_vp == NULL) {
3911 rfs4_dbe_unlock(fp->rf_dbe);
3912 return;
3913 }
3914 rfs4_dbe_unlock(fp->rf_dbe);
3915
3916 /*
3917 * Hold as writer to prevent other server threads from
3918 * processing requests related to the file while all state is
3919 * being removed.
3920 */
3921 rw_enter(&fp->rf_file_rwlock, RW_WRITER);
3922
3923 /* Remove ALL state from the file */
3924 while (sp = rfs4_findstate_by_file(fp)) {
3925 rfs4_state_close(sp, FALSE, FALSE, CRED());
3926 rfs4_state_rele_nounlock(sp);
3927 }
3928
3929 /*
3930 * This is only safe since there are no further references to
3931 * the file.
3932 */
3933 rfs4_dbe_lock(fp->rf_dbe);
3934 if (fp->rf_vp) {
3935 vnode_t *vp = fp->rf_vp;
3936
3937 mutex_enter(&vp->v_vsd_lock);
3938 (void) vsd_set(vp, nfs4_srv_vkey, NULL);
3939 mutex_exit(&vp->v_vsd_lock);
3940 VN_RELE(vp);
3941 fp->rf_vp = NULL;
3942 }
3943 rfs4_dbe_unlock(fp->rf_dbe);
3944
3945 /* Finally let other references to proceed */
3946 rw_exit(&fp->rf_file_rwlock);
3947 }
3948
3949 /*
3950 * This function is used as a target for the rfs4_dbe_walk() call
3951 * below. The purpose of this function is to see if the
3952 * lockowner_state refers to a file that resides within the exportinfo
3953 * export. If so, then remove the lock_owner state (file locks and
3954 * share "locks") for this object since the intent is the server is
3955 * unexporting the specified directory. Be sure to invalidate the
3956 * object after the state has been released
3957 */
3958 static void
3959 rfs4_lo_state_walk_callout(rfs4_entry_t u_entry, void *e)
3960 {
3961 rfs4_lo_state_t *lsp = (rfs4_lo_state_t *)u_entry;
3962 struct exportinfo *exi = (struct exportinfo *)e;
3963 nfs_fh4_fmt_t fhfmt4, *exi_fhp, *finfo_fhp;
3964 fhandle_t *efhp;
3965
3966 efhp = (fhandle_t *)&exi->exi_fh;
3967 exi_fhp = (nfs_fh4_fmt_t *)&fhfmt4;
3968
3969 FH_TO_FMT4(efhp, exi_fhp);
3970
3971 finfo_fhp = (nfs_fh4_fmt_t *)lsp->rls_state->rs_finfo->
3972 rf_filehandle.nfs_fh4_val;
3973
3974 if (EQFSID(&finfo_fhp->fh4_fsid, &exi_fhp->fh4_fsid) &&
3975 bcmp(&finfo_fhp->fh4_xdata, &exi_fhp->fh4_xdata,
3976 exi_fhp->fh4_xlen) == 0) {
3977 rfs4_state_close(lsp->rls_state, FALSE, FALSE, CRED());
3978 rfs4_dbe_invalidate(lsp->rls_dbe);
3979 rfs4_dbe_invalidate(lsp->rls_state->rs_dbe);
3980 }
3981 }
3982
3983 /*
3984 * This function is used as a target for the rfs4_dbe_walk() call
3985 * below. The purpose of this function is to see if the state refers
3986 * to a file that resides within the exportinfo export. If so, then
3987 * remove the open state for this object since the intent is the
3988 * server is unexporting the specified directory. The main result for
3989 * this type of entry is to invalidate it such it will not be found in
3990 * the future.
3991 */
3992 static void
3993 rfs4_state_walk_callout(rfs4_entry_t u_entry, void *e)
3994 {
3995 rfs4_state_t *sp = (rfs4_state_t *)u_entry;
3996 struct exportinfo *exi = (struct exportinfo *)e;
3997 nfs_fh4_fmt_t fhfmt4, *exi_fhp, *finfo_fhp;
3998 fhandle_t *efhp;
3999
4000 efhp = (fhandle_t *)&exi->exi_fh;
4001 exi_fhp = (nfs_fh4_fmt_t *)&fhfmt4;
4002
4003 FH_TO_FMT4(efhp, exi_fhp);
4004
4005 finfo_fhp =
4006 (nfs_fh4_fmt_t *)sp->rs_finfo->rf_filehandle.nfs_fh4_val;
4007
4008 if (EQFSID(&finfo_fhp->fh4_fsid, &exi_fhp->fh4_fsid) &&
4009 bcmp(&finfo_fhp->fh4_xdata, &exi_fhp->fh4_xdata,
4010 exi_fhp->fh4_xlen) == 0) {
4011 rfs4_state_close(sp, TRUE, FALSE, CRED());
4012 rfs4_dbe_invalidate(sp->rs_dbe);
4013 }
4014 }
4015
4016 /*
4017 * This function is used as a target for the rfs4_dbe_walk() call
4018 * below. The purpose of this function is to see if the state refers
4019 * to a file that resides within the exportinfo export. If so, then
4020 * remove the deleg state for this object since the intent is the
4021 * server is unexporting the specified directory. The main result for
4022 * this type of entry is to invalidate it such it will not be found in
4023 * the future.
4024 */
4025 static void
4026 rfs4_deleg_state_walk_callout(rfs4_entry_t u_entry, void *e)
4027 {
4028 rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
4029 struct exportinfo *exi = (struct exportinfo *)e;
4030 nfs_fh4_fmt_t fhfmt4, *exi_fhp, *finfo_fhp;
4031 fhandle_t *efhp;
4032
4033 efhp = (fhandle_t *)&exi->exi_fh;
4034 exi_fhp = (nfs_fh4_fmt_t *)&fhfmt4;
4035
4036 FH_TO_FMT4(efhp, exi_fhp);
4037
4038 finfo_fhp =
4039 (nfs_fh4_fmt_t *)dsp->rds_finfo->rf_filehandle.nfs_fh4_val;
4040
4041 if (EQFSID(&finfo_fhp->fh4_fsid, &exi_fhp->fh4_fsid) &&
4042 bcmp(&finfo_fhp->fh4_xdata, &exi_fhp->fh4_xdata,
4043 exi_fhp->fh4_xlen) == 0) {
4044 rfs4_dbe_invalidate(dsp->rds_dbe);
4045 }
4046 }
4047
4048 /*
4049 * This function is used as a target for the rfs4_dbe_walk() call
4050 * below. The purpose of this function is to see if the state refers
4051 * to a file that resides within the exportinfo export. If so, then
4052 * release vnode hold for this object since the intent is the server
4053 * is unexporting the specified directory. Invalidation will prevent
4054 * this struct from being found in the future.
4055 */
4056 static void
4057 rfs4_file_walk_callout(rfs4_entry_t u_entry, void *e)
4058 {
4059 rfs4_file_t *fp = (rfs4_file_t *)u_entry;
4060 struct exportinfo *exi = (struct exportinfo *)e;
4061 nfs_fh4_fmt_t fhfmt4, *exi_fhp, *finfo_fhp;
4062 fhandle_t *efhp;
4063
4064 efhp = (fhandle_t *)&exi->exi_fh;
4065 exi_fhp = (nfs_fh4_fmt_t *)&fhfmt4;
4066
4067 FH_TO_FMT4(efhp, exi_fhp);
4068
4069 finfo_fhp = (nfs_fh4_fmt_t *)fp->rf_filehandle.nfs_fh4_val;
4070
4071 if (EQFSID(&finfo_fhp->fh4_fsid, &exi_fhp->fh4_fsid) &&
4072 bcmp(&finfo_fhp->fh4_xdata, &exi_fhp->fh4_xdata,
4073 exi_fhp->fh4_xlen) == 0) {
4074 if (fp->rf_vp) {
4075 vnode_t *vp = fp->rf_vp;
4076
4077 /*
4078 * don't leak monitors and remove the reference
4079 * put on the vnode when the delegation was granted.
4080 */
4081 if (fp->rf_dinfo.rd_dtype == OPEN_DELEGATE_READ) {
4082 (void) fem_uninstall(vp, deleg_rdops,
4083 (void *)fp);
4084 vn_open_downgrade(vp, FREAD);
4085 } else if (fp->rf_dinfo.rd_dtype ==
4086 OPEN_DELEGATE_WRITE) {
4087 (void) fem_uninstall(vp, deleg_wrops,
4088 (void *)fp);
4089 vn_open_downgrade(vp, FREAD|FWRITE);
4090 }
4091 mutex_enter(&vp->v_vsd_lock);
4092 (void) vsd_set(vp, nfs4_srv_vkey, NULL);
4093 mutex_exit(&vp->v_vsd_lock);
4094 VN_RELE(vp);
4095 fp->rf_vp = NULL;
4096 }
4097 rfs4_dbe_invalidate(fp->rf_dbe);
4098 }
4099 }
4100
4101 /*
4102 * Given a directory that is being unexported, cleanup/release all
4103 * state in the server that refers to objects residing underneath this
4104 * particular export. The ordering of the release is important.
4105 * Lock_owner, then state and then file.
4106 */
4107 void
4108 rfs4_clean_state_exi(struct exportinfo *exi)
4109 {
4110 nfs4_srv_t *nsrv4;
4111
4112 /* curzone mightn't be exi_zone, so use exi_zone instead. */
4113 ASSERT(exi->exi_zone == curzone || curzone == global_zone);
4114 nsrv4 = zone_getspecific(rfs4_zone_key, exi->exi_zone);
4115 if (nsrv4 == NULL) /* NOTE: NFSv4 cleanup MAY have already happened. */
4116 return;
4117 mutex_enter(&nsrv4->state_lock);
4118
4119 if (nsrv4->nfs4_server_state == NULL) {
4120 mutex_exit(&nsrv4->state_lock);
4121 return;
4122 }
4123
4124 /* CSTYLED */
4125 rfs4_dbe_walk(nsrv4->rfs4_lo_state_tab, rfs4_lo_state_walk_callout, exi);
4126 rfs4_dbe_walk(nsrv4->rfs4_state_tab, rfs4_state_walk_callout, exi);
4127 /* CSTYLED */
4128 rfs4_dbe_walk(nsrv4->rfs4_deleg_state_tab, rfs4_deleg_state_walk_callout, exi);
4129 rfs4_dbe_walk(nsrv4->rfs4_file_tab, rfs4_file_walk_callout, exi);
4130
4131 mutex_exit(&nsrv4->state_lock);
4132 }