Print this page
Revert "NEX-20260 NFS hung in transitional state when RSF marks it maintenance"
This reverts commit 9bf6e5f740709f470ba350df64cd9f2c93f3f0a7.
| Split |
Close |
| Expand all |
| Collapse all |
--- old/usr/src/uts/common/fs/nfs/nfs4_state.c
+++ new/usr/src/uts/common/fs/nfs/nfs4_state.c
1 1 /*
2 2 * CDDL HEADER START
3 3 *
4 4 * The contents of this file are subject to the terms of the
5 5 * Common Development and Distribution License (the "License").
6 6 * You may not use this file except in compliance with the License.
7 7 *
8 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 9 * or http://www.opensolaris.org/os/licensing.
10 10 * See the License for the specific language governing permissions
11 11 * and limitations under the License.
12 12 *
13 13 * When distributing Covered Code, include this CDDL HEADER in each
14 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 15 * If applicable, add the following below this CDDL HEADER, with the
16 16 * fields enclosed by brackets "[]" replaced with your own identifying
17 17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 18 *
19 19 * CDDL HEADER END
20 20 */
21 21
22 22 /*
23 23 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
24 24 */
25 25
26 26 /*
27 27 * Copyright 2018 Nexenta Systems, Inc.
28 28 * Copyright 2019 Nexenta by DDN, Inc.
29 29 */
30 30
31 31 #include <sys/systm.h>
32 32 #include <sys/kmem.h>
33 33 #include <sys/cmn_err.h>
34 34 #include <sys/atomic.h>
35 35 #include <sys/clconf.h>
36 36 #include <sys/cladm.h>
37 37 #include <sys/flock.h>
38 38 #include <nfs/export.h>
39 39 #include <nfs/nfs.h>
40 40 #include <nfs/nfs4.h>
41 41 #include <nfs/nfssys.h>
42 42 #include <nfs/lm.h>
43 43 #include <sys/pathname.h>
44 44 #include <sys/sdt.h>
45 45 #include <sys/nvpair.h>
46 46
47 47 extern u_longlong_t nfs4_srv_caller_id;
48 48
49 49 extern uint_t nfs4_srv_vkey;
50 50
51 51 stateid4 special0 = {
52 52 0,
53 53 { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 }
54 54 };
55 55
56 56 stateid4 special1 = {
57 57 0xffffffff,
58 58 {
59 59 (char)0xff, (char)0xff, (char)0xff, (char)0xff,
60 60 (char)0xff, (char)0xff, (char)0xff, (char)0xff,
61 61 (char)0xff, (char)0xff, (char)0xff, (char)0xff
62 62 }
63 63 };
64 64
65 65
66 66 #define ISSPECIAL(id) (stateid4_cmp(id, &special0) || \
67 67 stateid4_cmp(id, &special1))
68 68
69 69 /* For embedding the cluster nodeid into our clientid */
70 70 #define CLUSTER_NODEID_SHIFT 24
71 71 #define CLUSTER_MAX_NODEID 255
72 72
73 73 #ifdef DEBUG
74 74 int rfs4_debug;
75 75 #endif
76 76
77 77 static uint32_t rfs4_database_debug = 0x00;
78 78
79 79 /* CSTYLED */
80 80 static void rfs4_ss_clid_write(nfs4_srv_t *nsrv4, rfs4_client_t *cp, char *leaf);
81 81 static void rfs4_ss_clid_write_one(rfs4_client_t *cp, char *dir, char *leaf);
82 82 static void rfs4_dss_clear_oldstate(rfs4_servinst_t *sip);
83 83 static void rfs4_ss_chkclid_sip(rfs4_client_t *cp, rfs4_servinst_t *sip);
84 84
85 85 /*
86 86 * Couple of simple init/destroy functions for a general waiter
87 87 */
88 88 void
89 89 rfs4_sw_init(rfs4_state_wait_t *swp)
90 90 {
91 91 mutex_init(swp->sw_cv_lock, NULL, MUTEX_DEFAULT, NULL);
92 92 cv_init(swp->sw_cv, NULL, CV_DEFAULT, NULL);
93 93 swp->sw_active = FALSE;
94 94 swp->sw_wait_count = 0;
95 95 }
96 96
97 97 void
98 98 rfs4_sw_destroy(rfs4_state_wait_t *swp)
99 99 {
100 100 mutex_destroy(swp->sw_cv_lock);
101 101 cv_destroy(swp->sw_cv);
102 102 }
103 103
104 104 void
105 105 rfs4_sw_enter(rfs4_state_wait_t *swp)
106 106 {
107 107 mutex_enter(swp->sw_cv_lock);
108 108 while (swp->sw_active) {
109 109 swp->sw_wait_count++;
110 110 cv_wait(swp->sw_cv, swp->sw_cv_lock);
111 111 swp->sw_wait_count--;
112 112 }
113 113 ASSERT(swp->sw_active == FALSE);
114 114 swp->sw_active = TRUE;
115 115 mutex_exit(swp->sw_cv_lock);
116 116 }
117 117
118 118 void
119 119 rfs4_sw_exit(rfs4_state_wait_t *swp)
120 120 {
121 121 mutex_enter(swp->sw_cv_lock);
122 122 ASSERT(swp->sw_active == TRUE);
123 123 swp->sw_active = FALSE;
124 124 if (swp->sw_wait_count != 0)
125 125 cv_broadcast(swp->sw_cv);
126 126 mutex_exit(swp->sw_cv_lock);
127 127 }
128 128
129 129 static void
130 130 deep_lock_copy(LOCK4res *dres, LOCK4res *sres)
131 131 {
132 132 lock_owner4 *slo = &sres->LOCK4res_u.denied.owner;
133 133 lock_owner4 *dlo = &dres->LOCK4res_u.denied.owner;
134 134
135 135 if (sres->status == NFS4ERR_DENIED) {
136 136 dlo->owner_val = kmem_alloc(slo->owner_len, KM_SLEEP);
137 137 bcopy(slo->owner_val, dlo->owner_val, slo->owner_len);
138 138 }
139 139 }
140 140
141 141 /*
142 142 * CPR callback id -- not related to v4 callbacks
143 143 */
144 144 static callb_id_t cpr_id = 0;
145 145
146 146 static void
147 147 deep_lock_free(LOCK4res *res)
148 148 {
149 149 lock_owner4 *lo = &res->LOCK4res_u.denied.owner;
150 150
151 151 if (res->status == NFS4ERR_DENIED)
152 152 kmem_free(lo->owner_val, lo->owner_len);
153 153 }
154 154
155 155 static void
156 156 deep_open_copy(OPEN4res *dres, OPEN4res *sres)
157 157 {
158 158 nfsace4 *sacep, *dacep;
159 159
160 160 if (sres->status != NFS4_OK) {
161 161 return;
162 162 }
163 163
164 164 dres->attrset = sres->attrset;
165 165
166 166 switch (sres->delegation.delegation_type) {
167 167 case OPEN_DELEGATE_NONE:
168 168 return;
169 169 case OPEN_DELEGATE_READ:
170 170 sacep = &sres->delegation.open_delegation4_u.read.permissions;
171 171 dacep = &dres->delegation.open_delegation4_u.read.permissions;
172 172 break;
173 173 case OPEN_DELEGATE_WRITE:
174 174 sacep = &sres->delegation.open_delegation4_u.write.permissions;
175 175 dacep = &dres->delegation.open_delegation4_u.write.permissions;
176 176 break;
177 177 }
178 178 dacep->who.utf8string_val =
179 179 kmem_alloc(sacep->who.utf8string_len, KM_SLEEP);
180 180 bcopy(sacep->who.utf8string_val, dacep->who.utf8string_val,
181 181 sacep->who.utf8string_len);
182 182 }
183 183
184 184 static void
185 185 deep_open_free(OPEN4res *res)
186 186 {
187 187 nfsace4 *acep;
188 188 if (res->status != NFS4_OK)
189 189 return;
190 190
191 191 switch (res->delegation.delegation_type) {
192 192 case OPEN_DELEGATE_NONE:
193 193 return;
194 194 case OPEN_DELEGATE_READ:
195 195 acep = &res->delegation.open_delegation4_u.read.permissions;
196 196 break;
197 197 case OPEN_DELEGATE_WRITE:
198 198 acep = &res->delegation.open_delegation4_u.write.permissions;
199 199 break;
200 200 }
201 201
202 202 if (acep->who.utf8string_val) {
203 203 kmem_free(acep->who.utf8string_val, acep->who.utf8string_len);
204 204 acep->who.utf8string_val = NULL;
205 205 }
206 206 }
207 207
208 208 void
209 209 rfs4_free_reply(nfs_resop4 *rp)
210 210 {
211 211 switch (rp->resop) {
212 212 case OP_LOCK:
213 213 deep_lock_free(&rp->nfs_resop4_u.oplock);
214 214 break;
215 215 case OP_OPEN:
216 216 deep_open_free(&rp->nfs_resop4_u.opopen);
217 217 default:
218 218 break;
219 219 }
220 220 }
221 221
222 222 void
223 223 rfs4_copy_reply(nfs_resop4 *dst, nfs_resop4 *src)
224 224 {
225 225 *dst = *src;
226 226
227 227 /* Handle responses that need deep copy */
228 228 switch (src->resop) {
229 229 case OP_LOCK:
230 230 deep_lock_copy(&dst->nfs_resop4_u.oplock,
231 231 &src->nfs_resop4_u.oplock);
232 232 break;
233 233 case OP_OPEN:
234 234 deep_open_copy(&dst->nfs_resop4_u.opopen,
235 235 &src->nfs_resop4_u.opopen);
236 236 break;
237 237 default:
238 238 break;
239 239 };
240 240 }
241 241
242 242 /*
243 243 * This is the implementation of the underlying state engine. The
244 244 * public interface to this engine is described by
245 245 * nfs4_state.h. Callers to the engine should hold no state engine
246 246 * locks when they call in to it. If the protocol needs to lock data
247 247 * structures it should do so after acquiring all references to them
248 248 * first and then follow the following lock order:
249 249 *
250 250 * client > openowner > state > lo_state > lockowner > file.
251 251 *
252 252 * Internally we only allow a thread to hold one hash bucket lock at a
253 253 * time and the lock is higher in the lock order (must be acquired
254 254 * first) than the data structure that is on that hash list.
255 255 *
256 256 * If a new reference was acquired by the caller, that reference needs
257 257 * to be released after releasing all acquired locks with the
258 258 * corresponding rfs4_*_rele routine.
259 259 */
260 260
261 261 /*
262 262 * This code is some what prototypical for now. Its purpose currently is to
263 263 * implement the interfaces sufficiently to finish the higher protocol
264 264 * elements. This will be replaced by a dynamically resizeable tables
265 265 * backed by kmem_cache allocator. However synchronization is handled
266 266 * correctly (I hope) and will not change by much. The mutexes for
267 267 * the hash buckets that can be used to create new instances of data
268 268 * structures might be good candidates to evolve into reader writer
269 269 * locks. If it has to do a creation, it would be holding the
270 270 * mutex across a kmem_alloc with KM_SLEEP specified.
271 271 */
272 272
273 273 #ifdef DEBUG
274 274 #define TABSIZE 17
275 275 #else
276 276 #define TABSIZE 2047
277 277 #endif
278 278
279 279 #define ADDRHASH(key) ((unsigned long)(key) >> 3)
280 280
281 281 #define MAXTABSZ 1024*1024
282 282
283 283 /* The values below are rfs4_lease_time units */
284 284
285 285 #ifdef DEBUG
286 286 #define CLIENT_CACHE_TIME 1
287 287 #define OPENOWNER_CACHE_TIME 1
288 288 #define STATE_CACHE_TIME 1
289 289 #define LO_STATE_CACHE_TIME 1
290 290 #define LOCKOWNER_CACHE_TIME 1
291 291 #define FILE_CACHE_TIME 3
292 292 #define DELEG_STATE_CACHE_TIME 1
293 293 #else
294 294 #define CLIENT_CACHE_TIME 10
295 295 #define OPENOWNER_CACHE_TIME 5
296 296 #define STATE_CACHE_TIME 1
297 297 #define LO_STATE_CACHE_TIME 1
298 298 #define LOCKOWNER_CACHE_TIME 3
299 299 #define FILE_CACHE_TIME 40
300 300 #define DELEG_STATE_CACHE_TIME 1
301 301 #endif
302 302
303 303 /*
304 304 * NFSv4 server state databases
305 305 *
306 306 * Initilized when the module is loaded and used by NFSv4 state tables.
307 307 * These kmem_cache databases are global, the tables that make use of these
308 308 * are per zone.
309 309 */
310 310 kmem_cache_t *rfs4_client_mem_cache;
311 311 kmem_cache_t *rfs4_clntIP_mem_cache;
312 312 kmem_cache_t *rfs4_openown_mem_cache;
313 313 kmem_cache_t *rfs4_openstID_mem_cache;
314 314 kmem_cache_t *rfs4_lockstID_mem_cache;
315 315 kmem_cache_t *rfs4_lockown_mem_cache;
316 316 kmem_cache_t *rfs4_file_mem_cache;
317 317 kmem_cache_t *rfs4_delegstID_mem_cache;
318 318
319 319 /*
320 320 * NFSv4 state table functions
321 321 */
322 322 static bool_t rfs4_client_create(rfs4_entry_t, void *);
323 323 static void rfs4_dss_remove_cpleaf(rfs4_client_t *);
324 324 static void rfs4_dss_remove_leaf(rfs4_servinst_t *, char *, char *);
325 325 static void rfs4_client_destroy(rfs4_entry_t);
326 326 static bool_t rfs4_client_expiry(rfs4_entry_t);
327 327 static uint32_t clientid_hash(void *);
328 328 static bool_t clientid_compare(rfs4_entry_t, void *);
329 329 static void *clientid_mkkey(rfs4_entry_t);
330 330 static uint32_t nfsclnt_hash(void *);
331 331 static bool_t nfsclnt_compare(rfs4_entry_t, void *);
332 332 static void *nfsclnt_mkkey(rfs4_entry_t);
333 333 static bool_t rfs4_clntip_expiry(rfs4_entry_t);
334 334 static void rfs4_clntip_destroy(rfs4_entry_t);
335 335 static bool_t rfs4_clntip_create(rfs4_entry_t, void *);
336 336 static uint32_t clntip_hash(void *);
337 337 static bool_t clntip_compare(rfs4_entry_t, void *);
338 338 static void *clntip_mkkey(rfs4_entry_t);
339 339 static bool_t rfs4_openowner_create(rfs4_entry_t, void *);
340 340 static void rfs4_openowner_destroy(rfs4_entry_t);
341 341 static bool_t rfs4_openowner_expiry(rfs4_entry_t);
342 342 static uint32_t openowner_hash(void *);
343 343 static bool_t openowner_compare(rfs4_entry_t, void *);
344 344 static void *openowner_mkkey(rfs4_entry_t);
345 345 static bool_t rfs4_state_create(rfs4_entry_t, void *);
346 346 static void rfs4_state_destroy(rfs4_entry_t);
347 347 static bool_t rfs4_state_expiry(rfs4_entry_t);
348 348 static uint32_t state_hash(void *);
349 349 static bool_t state_compare(rfs4_entry_t, void *);
350 350 static void *state_mkkey(rfs4_entry_t);
351 351 static uint32_t state_owner_file_hash(void *);
352 352 static bool_t state_owner_file_compare(rfs4_entry_t, void *);
353 353 static void *state_owner_file_mkkey(rfs4_entry_t);
354 354 static uint32_t state_file_hash(void *);
355 355 static bool_t state_file_compare(rfs4_entry_t, void *);
356 356 static void *state_file_mkkey(rfs4_entry_t);
357 357 static bool_t rfs4_lo_state_create(rfs4_entry_t, void *);
358 358 static void rfs4_lo_state_destroy(rfs4_entry_t);
359 359 static bool_t rfs4_lo_state_expiry(rfs4_entry_t);
360 360 static uint32_t lo_state_hash(void *);
361 361 static bool_t lo_state_compare(rfs4_entry_t, void *);
362 362 static void *lo_state_mkkey(rfs4_entry_t);
363 363 static uint32_t lo_state_lo_hash(void *);
364 364 static bool_t lo_state_lo_compare(rfs4_entry_t, void *);
365 365 static void *lo_state_lo_mkkey(rfs4_entry_t);
366 366 static bool_t rfs4_lockowner_create(rfs4_entry_t, void *);
367 367 static void rfs4_lockowner_destroy(rfs4_entry_t);
368 368 static bool_t rfs4_lockowner_expiry(rfs4_entry_t);
369 369 static uint32_t lockowner_hash(void *);
370 370 static bool_t lockowner_compare(rfs4_entry_t, void *);
371 371 static void *lockowner_mkkey(rfs4_entry_t);
372 372 static uint32_t pid_hash(void *);
373 373 static bool_t pid_compare(rfs4_entry_t, void *);
374 374 static void *pid_mkkey(rfs4_entry_t);
375 375 static bool_t rfs4_file_create(rfs4_entry_t, void *);
376 376 static void rfs4_file_destroy(rfs4_entry_t);
377 377 static uint32_t file_hash(void *);
378 378 static bool_t file_compare(rfs4_entry_t, void *);
379 379 static void *file_mkkey(rfs4_entry_t);
380 380 static bool_t rfs4_deleg_state_create(rfs4_entry_t, void *);
381 381 static void rfs4_deleg_state_destroy(rfs4_entry_t);
382 382 static bool_t rfs4_deleg_state_expiry(rfs4_entry_t);
383 383 static uint32_t deleg_hash(void *);
384 384 static bool_t deleg_compare(rfs4_entry_t, void *);
385 385 static void *deleg_mkkey(rfs4_entry_t);
386 386 static uint32_t deleg_state_hash(void *);
387 387 static bool_t deleg_state_compare(rfs4_entry_t, void *);
388 388 static void *deleg_state_mkkey(rfs4_entry_t);
389 389
390 390 static void rfs4_state_rele_nounlock(rfs4_state_t *);
391 391
392 392 static int rfs4_ss_enabled = 0;
393 393
394 394 extern void (*rfs4_client_clrst)(struct nfs4clrst_args *);
395 395
396 396 void
397 397 rfs4_ss_pnfree(rfs4_ss_pn_t *ss_pn)
398 398 {
399 399 kmem_free(ss_pn, sizeof (rfs4_ss_pn_t));
400 400 }
401 401
402 402 static rfs4_ss_pn_t *
403 403 rfs4_ss_pnalloc(char *dir, char *leaf)
404 404 {
405 405 rfs4_ss_pn_t *ss_pn;
406 406 int dir_len, leaf_len;
407 407
408 408 /*
409 409 * validate we have a resonable path
410 410 * (account for the '/' and trailing null)
411 411 */
412 412 if ((dir_len = strlen(dir)) > MAXPATHLEN ||
413 413 (leaf_len = strlen(leaf)) > MAXNAMELEN ||
414 414 (dir_len + leaf_len + 2) > MAXPATHLEN) {
415 415 return (NULL);
416 416 }
417 417
418 418 ss_pn = kmem_alloc(sizeof (rfs4_ss_pn_t), KM_SLEEP);
419 419
420 420 (void) snprintf(ss_pn->pn, MAXPATHLEN, "%s/%s", dir, leaf);
421 421 /* Handy pointer to just the leaf name */
422 422 ss_pn->leaf = ss_pn->pn + dir_len + 1;
423 423 return (ss_pn);
424 424 }
425 425
426 426
427 427 /*
428 428 * Move the "leaf" filename from "sdir" directory
429 429 * to the "ddir" directory. Return the pathname of
430 430 * the destination unless the rename fails in which
431 431 * case we need to return the source pathname.
432 432 */
433 433 static rfs4_ss_pn_t *
434 434 rfs4_ss_movestate(char *sdir, char *ddir, char *leaf)
435 435 {
436 436 rfs4_ss_pn_t *src, *dst;
437 437
438 438 if ((src = rfs4_ss_pnalloc(sdir, leaf)) == NULL)
439 439 return (NULL);
440 440
441 441 if ((dst = rfs4_ss_pnalloc(ddir, leaf)) == NULL) {
442 442 rfs4_ss_pnfree(src);
443 443 return (NULL);
444 444 }
445 445
446 446 /*
447 447 * If the rename fails we shall return the src
448 448 * pathname and free the dst. Otherwise we need
449 449 * to free the src and return the dst pathanme.
450 450 */
451 451 if (vn_rename(src->pn, dst->pn, UIO_SYSSPACE)) {
452 452 rfs4_ss_pnfree(dst);
453 453 return (src);
454 454 }
455 455 rfs4_ss_pnfree(src);
456 456 return (dst);
457 457 }
458 458
459 459
460 460 static rfs4_oldstate_t *
461 461 rfs4_ss_getstate(vnode_t *dvp, rfs4_ss_pn_t *ss_pn)
462 462 {
463 463 struct uio uio;
464 464 struct iovec iov[3];
465 465
466 466 rfs4_oldstate_t *cl_ss = NULL;
467 467 vnode_t *vp;
468 468 vattr_t va;
469 469 uint_t id_len;
470 470 int err, kill_file, file_vers;
471 471
472 472 if (ss_pn == NULL)
473 473 return (NULL);
474 474
475 475 /*
476 476 * open the state file.
477 477 */
478 478 if (vn_open(ss_pn->pn, UIO_SYSSPACE, FREAD, 0, &vp, 0, 0) != 0) {
479 479 return (NULL);
480 480 }
481 481
482 482 if (vp->v_type != VREG) {
483 483 (void) VOP_CLOSE(vp, FREAD, 1, (offset_t)0, CRED(), NULL);
484 484 VN_RELE(vp);
485 485 return (NULL);
486 486 }
487 487
488 488 err = VOP_ACCESS(vp, VREAD, 0, CRED(), NULL);
489 489 if (err) {
490 490 /*
491 491 * We don't have read access? better get the heck out.
492 492 */
493 493 (void) VOP_CLOSE(vp, FREAD, 1, (offset_t)0, CRED(), NULL);
494 494 VN_RELE(vp);
495 495 return (NULL);
496 496 }
497 497
498 498 (void) VOP_RWLOCK(vp, V_WRITELOCK_FALSE, NULL);
499 499 /*
500 500 * get the file size to do some basic validation
501 501 */
502 502 va.va_mask = AT_SIZE;
503 503 err = VOP_GETATTR(vp, &va, 0, CRED(), NULL);
504 504
505 505 kill_file = (va.va_size == 0 || va.va_size <
506 506 (NFS4_VERIFIER_SIZE + sizeof (uint_t)+1));
507 507
508 508 if (err || kill_file) {
509 509 VOP_RWUNLOCK(vp, V_WRITELOCK_FALSE, NULL);
510 510 (void) VOP_CLOSE(vp, FREAD, 1, (offset_t)0, CRED(), NULL);
511 511 VN_RELE(vp);
512 512 if (kill_file) {
513 513 (void) VOP_REMOVE(dvp, ss_pn->leaf, CRED(), NULL, 0);
514 514 }
515 515 return (NULL);
516 516 }
517 517
518 518 cl_ss = kmem_alloc(sizeof (rfs4_oldstate_t), KM_SLEEP);
519 519
520 520 /*
521 521 * build iovecs to read in the file_version, verifier and id_len
522 522 */
523 523 iov[0].iov_base = (caddr_t)&file_vers;
524 524 iov[0].iov_len = sizeof (int);
525 525 iov[1].iov_base = (caddr_t)&cl_ss->cl_id4.verifier;
526 526 iov[1].iov_len = NFS4_VERIFIER_SIZE;
527 527 iov[2].iov_base = (caddr_t)&id_len;
528 528 iov[2].iov_len = sizeof (uint_t);
529 529
530 530 uio.uio_iov = iov;
531 531 uio.uio_iovcnt = 3;
532 532 uio.uio_segflg = UIO_SYSSPACE;
533 533 uio.uio_loffset = 0;
534 534 uio.uio_resid = sizeof (int) + NFS4_VERIFIER_SIZE + sizeof (uint_t);
535 535
536 536 if (err = VOP_READ(vp, &uio, FREAD, CRED(), NULL)) {
537 537 VOP_RWUNLOCK(vp, V_WRITELOCK_FALSE, NULL);
538 538 (void) VOP_CLOSE(vp, FREAD, 1, (offset_t)0, CRED(), NULL);
539 539 VN_RELE(vp);
540 540 kmem_free(cl_ss, sizeof (rfs4_oldstate_t));
541 541 return (NULL);
542 542 }
543 543
544 544 /*
545 545 * if the file_version doesn't match or if the
546 546 * id_len is zero or the combination of the verifier,
547 547 * id_len and id_val is bigger than the file we have
548 548 * a problem. If so ditch the file.
549 549 */
550 550 kill_file = (file_vers != NFS4_SS_VERSION || id_len == 0 ||
551 551 (id_len + NFS4_VERIFIER_SIZE + sizeof (uint_t)) > va.va_size);
552 552
553 553 if (err || kill_file) {
554 554 VOP_RWUNLOCK(vp, V_WRITELOCK_FALSE, NULL);
555 555 (void) VOP_CLOSE(vp, FREAD, 1, (offset_t)0, CRED(), NULL);
556 556 VN_RELE(vp);
557 557 kmem_free(cl_ss, sizeof (rfs4_oldstate_t));
558 558 if (kill_file) {
559 559 (void) VOP_REMOVE(dvp, ss_pn->leaf, CRED(), NULL, 0);
560 560 }
561 561 return (NULL);
562 562 }
563 563
564 564 /*
565 565 * now get the client id value
566 566 */
567 567 cl_ss->cl_id4.id_val = kmem_alloc(id_len, KM_SLEEP);
568 568 iov[0].iov_base = cl_ss->cl_id4.id_val;
569 569 iov[0].iov_len = id_len;
570 570
571 571 uio.uio_iov = iov;
572 572 uio.uio_iovcnt = 1;
573 573 uio.uio_segflg = UIO_SYSSPACE;
574 574 uio.uio_resid = cl_ss->cl_id4.id_len = id_len;
575 575
576 576 if (err = VOP_READ(vp, &uio, FREAD, CRED(), NULL)) {
577 577 VOP_RWUNLOCK(vp, V_WRITELOCK_FALSE, NULL);
578 578 (void) VOP_CLOSE(vp, FREAD, 1, (offset_t)0, CRED(), NULL);
579 579 VN_RELE(vp);
580 580 kmem_free(cl_ss->cl_id4.id_val, id_len);
581 581 kmem_free(cl_ss, sizeof (rfs4_oldstate_t));
582 582 return (NULL);
583 583 }
584 584
585 585 VOP_RWUNLOCK(vp, V_WRITELOCK_FALSE, NULL);
586 586 (void) VOP_CLOSE(vp, FREAD, 1, (offset_t)0, CRED(), NULL);
587 587 VN_RELE(vp);
588 588 return (cl_ss);
589 589 }
590 590
591 591 #ifdef nextdp
592 592 #undef nextdp
593 593 #endif
594 594 #define nextdp(dp) ((struct dirent64 *)((char *)(dp) + (dp)->d_reclen))
595 595
596 596 /*
597 597 * Add entries from statedir to supplied oldstate list.
598 598 * Optionally, move all entries from statedir -> destdir.
599 599 */
600 600 void
601 601 rfs4_ss_oldstate(rfs4_oldstate_t *oldstate, char *statedir, char *destdir)
602 602 {
603 603 rfs4_ss_pn_t *ss_pn;
604 604 rfs4_oldstate_t *cl_ss = NULL;
605 605 char *dirt = NULL;
606 606 int err, dir_eof = 0, size = 0;
607 607 vnode_t *dvp;
608 608 struct iovec iov;
609 609 struct uio uio;
610 610 struct dirent64 *dep;
611 611 offset_t dirchunk_offset = 0;
612 612
613 613 /*
614 614 * open the state directory
615 615 */
616 616 if (vn_open(statedir, UIO_SYSSPACE, FREAD, 0, &dvp, 0, 0))
617 617 return;
618 618
619 619 if (dvp->v_type != VDIR || VOP_ACCESS(dvp, VREAD, 0, CRED(), NULL))
620 620 goto out;
621 621
622 622 dirt = kmem_alloc(RFS4_SS_DIRSIZE, KM_SLEEP);
623 623
624 624 /*
625 625 * Get and process the directory entries
626 626 */
627 627 while (!dir_eof) {
628 628 (void) VOP_RWLOCK(dvp, V_WRITELOCK_FALSE, NULL);
629 629 iov.iov_base = dirt;
630 630 iov.iov_len = RFS4_SS_DIRSIZE;
631 631 uio.uio_iov = &iov;
632 632 uio.uio_iovcnt = 1;
633 633 uio.uio_segflg = UIO_SYSSPACE;
634 634 uio.uio_loffset = dirchunk_offset;
635 635 uio.uio_resid = RFS4_SS_DIRSIZE;
636 636
637 637 err = VOP_READDIR(dvp, &uio, CRED(), &dir_eof, NULL, 0);
638 638 VOP_RWUNLOCK(dvp, V_WRITELOCK_FALSE, NULL);
639 639 if (err)
640 640 goto out;
641 641
642 642 size = RFS4_SS_DIRSIZE - uio.uio_resid;
643 643
644 644 /*
645 645 * Process all the directory entries in this
646 646 * readdir chunk
647 647 */
648 648 for (dep = (struct dirent64 *)dirt; size > 0;
649 649 dep = nextdp(dep)) {
650 650
651 651 size -= dep->d_reclen;
652 652 dirchunk_offset = dep->d_off;
653 653
654 654 /*
655 655 * Skip '.' and '..'
656 656 */
657 657 if (NFS_IS_DOTNAME(dep->d_name))
658 658 continue;
659 659
660 660 ss_pn = rfs4_ss_pnalloc(statedir, dep->d_name);
661 661 if (ss_pn == NULL)
662 662 continue;
663 663
664 664 if (cl_ss = rfs4_ss_getstate(dvp, ss_pn)) {
665 665 if (destdir != NULL) {
666 666 rfs4_ss_pnfree(ss_pn);
667 667 cl_ss->ss_pn = rfs4_ss_movestate(
668 668 statedir, destdir, dep->d_name);
669 669 } else {
670 670 cl_ss->ss_pn = ss_pn;
671 671 }
672 672 insque(cl_ss, oldstate);
673 673 } else {
674 674 rfs4_ss_pnfree(ss_pn);
675 675 }
676 676 }
677 677 }
678 678
679 679 out:
680 680 (void) VOP_CLOSE(dvp, FREAD, 1, (offset_t)0, CRED(), NULL);
681 681 VN_RELE(dvp);
682 682 if (dirt)
683 683 kmem_free((caddr_t)dirt, RFS4_SS_DIRSIZE);
684 684 }
685 685
686 686 static void
687 687 rfs4_ss_init(nfs4_srv_t *nsrv4)
688 688 {
689 689 int npaths = 1;
690 690 char *default_dss_path = NFS4_DSS_VAR_DIR;
691 691
692 692 /* read the default stable storage state */
693 693 rfs4_dss_readstate(nsrv4, npaths, &default_dss_path);
694 694
695 695 rfs4_ss_enabled = 1;
696 696 }
697 697
698 698 static void
699 699 rfs4_ss_fini(nfs4_srv_t *nsrv4)
700 700 {
701 701 rfs4_servinst_t *sip;
702 702
703 703 mutex_enter(&nsrv4->servinst_lock);
704 704 sip = nsrv4->nfs4_cur_servinst;
705 705 while (sip != NULL) {
706 706 rfs4_dss_clear_oldstate(sip);
707 707 sip = sip->next;
708 708 }
709 709 mutex_exit(&nsrv4->servinst_lock);
710 710 }
711 711
712 712 /*
713 713 * Remove all oldstate files referenced by this servinst.
714 714 */
715 715 static void
716 716 rfs4_dss_clear_oldstate(rfs4_servinst_t *sip)
717 717 {
718 718 rfs4_oldstate_t *os_head, *osp;
719 719
720 720 rw_enter(&sip->oldstate_lock, RW_WRITER);
721 721 os_head = sip->oldstate;
722 722
723 723 if (os_head == NULL) {
724 724 rw_exit(&sip->oldstate_lock);
725 725 return;
726 726 }
727 727
728 728 /* skip dummy entry */
729 729 osp = os_head->next;
730 730 while (osp != os_head) {
731 731 char *leaf = osp->ss_pn->leaf;
732 732 rfs4_oldstate_t *os_next;
733 733
734 734 rfs4_dss_remove_leaf(sip, NFS4_DSS_OLDSTATE_LEAF, leaf);
735 735
736 736 if (osp->cl_id4.id_val)
737 737 kmem_free(osp->cl_id4.id_val, osp->cl_id4.id_len);
738 738 rfs4_ss_pnfree(osp->ss_pn);
739 739
740 740 os_next = osp->next;
741 741 remque(osp);
742 742 kmem_free(osp, sizeof (rfs4_oldstate_t));
743 743 osp = os_next;
744 744 }
745 745
746 746 rw_exit(&sip->oldstate_lock);
747 747 }
748 748
749 749 /*
750 750 * Form the state and oldstate paths, and read in the stable storage files.
751 751 */
752 752 void
753 753 rfs4_dss_readstate(nfs4_srv_t *nsrv4, int npaths, char **paths)
754 754 {
755 755 int i;
756 756 char *state, *oldstate;
757 757
758 758 state = kmem_alloc(MAXPATHLEN, KM_SLEEP);
759 759 oldstate = kmem_alloc(MAXPATHLEN, KM_SLEEP);
760 760
761 761 for (i = 0; i < npaths; i++) {
762 762 char *path = paths[i];
763 763
764 764 (void) sprintf(state, "%s/%s", path, NFS4_DSS_STATE_LEAF);
765 765 (void) sprintf(oldstate, "%s/%s", path, NFS4_DSS_OLDSTATE_LEAF);
766 766
767 767 /*
768 768 * Populate the current server instance's oldstate list.
769 769 *
770 770 * 1. Read stable storage data from old state directory,
771 771 * leaving its contents alone.
772 772 *
773 773 * 2. Read stable storage data from state directory,
774 774 * and move the latter's contents to old state
775 775 * directory.
776 776 */
777 777 /* CSTYLED */
778 778 rfs4_ss_oldstate(nsrv4->nfs4_cur_servinst->oldstate, oldstate, NULL);
779 779 /* CSTYLED */
780 780 rfs4_ss_oldstate(nsrv4->nfs4_cur_servinst->oldstate, state, oldstate);
781 781 }
782 782
783 783 kmem_free(state, MAXPATHLEN);
784 784 kmem_free(oldstate, MAXPATHLEN);
785 785 }
786 786
787 787
788 788 /*
789 789 * Check if we are still in grace and if the client can be
790 790 * granted permission to perform reclaims.
791 791 */
792 792 void
793 793 rfs4_ss_chkclid(nfs4_srv_t *nsrv4, rfs4_client_t *cp)
794 794 {
795 795 rfs4_servinst_t *sip;
796 796
797 797 /*
798 798 * It should be sufficient to check the oldstate data for just
799 799 * this client's instance. However, since our per-instance
800 800 * client grouping is solely temporal, HA-NFSv4 RG failover
801 801 * might result in clients of the same RG being partitioned into
802 802 * separate instances.
803 803 *
804 804 * Until the client grouping is improved, we must check the
805 805 * oldstate data for all instances with an active grace period.
806 806 *
807 807 * This also serves as the mechanism to remove stale oldstate data.
808 808 * The first time we check an instance after its grace period has
809 809 * expired, the oldstate data should be cleared.
810 810 *
811 811 * Start at the current instance, and walk the list backwards
812 812 * to the first.
813 813 */
814 814 mutex_enter(&nsrv4->servinst_lock);
815 815 for (sip = nsrv4->nfs4_cur_servinst; sip != NULL; sip = sip->prev) {
816 816 rfs4_ss_chkclid_sip(cp, sip);
817 817
818 818 /* if the above check found this client, we're done */
819 819 if (cp->rc_can_reclaim)
820 820 break;
821 821 }
822 822 mutex_exit(&nsrv4->servinst_lock);
823 823 }
824 824
825 825 static void
826 826 rfs4_ss_chkclid_sip(rfs4_client_t *cp, rfs4_servinst_t *sip)
827 827 {
828 828 rfs4_oldstate_t *osp, *os_head;
829 829
830 830 /* short circuit everything if this server instance has no oldstate */
831 831 rw_enter(&sip->oldstate_lock, RW_READER);
832 832 os_head = sip->oldstate;
833 833 rw_exit(&sip->oldstate_lock);
834 834 if (os_head == NULL)
835 835 return;
836 836
837 837 /*
838 838 * If this server instance is no longer in a grace period then
839 839 * the client won't be able to reclaim. No further need for this
840 840 * instance's oldstate data, so it can be cleared.
841 841 */
842 842 if (!rfs4_servinst_in_grace(sip))
843 843 return;
844 844
845 845 /* this instance is still in grace; search for the clientid */
846 846
847 847 rw_enter(&sip->oldstate_lock, RW_READER);
848 848
849 849 os_head = sip->oldstate;
850 850 /* skip dummy entry */
851 851 osp = os_head->next;
852 852 while (osp != os_head) {
853 853 if (osp->cl_id4.id_len == cp->rc_nfs_client.id_len) {
854 854 if (bcmp(osp->cl_id4.id_val, cp->rc_nfs_client.id_val,
855 855 osp->cl_id4.id_len) == 0) {
856 856 cp->rc_can_reclaim = 1;
857 857 break;
858 858 }
859 859 }
860 860 osp = osp->next;
861 861 }
862 862
863 863 rw_exit(&sip->oldstate_lock);
864 864 }
865 865
866 866 /*
867 867 * Place client information into stable storage: 1/3.
868 868 * First, generate the leaf filename, from the client's IP address and
869 869 * the server-generated short-hand clientid.
870 870 */
871 871 void
872 872 rfs4_ss_clid(nfs4_srv_t *nsrv4, rfs4_client_t *cp)
873 873 {
874 874 const char *kinet_ntop6(uchar_t *, char *, size_t);
875 875 char leaf[MAXNAMELEN], buf[INET6_ADDRSTRLEN];
876 876 struct sockaddr *ca;
877 877 uchar_t *b;
878 878
879 879 if (rfs4_ss_enabled == 0) {
880 880 return;
881 881 }
882 882
883 883 buf[0] = 0;
884 884
885 885 ca = (struct sockaddr *)&cp->rc_addr;
886 886
887 887 /*
888 888 * Convert the caller's IP address to a dotted string
889 889 */
890 890 if (ca->sa_family == AF_INET) {
891 891 b = (uchar_t *)&((struct sockaddr_in *)ca)->sin_addr;
892 892 (void) sprintf(buf, "%03d.%03d.%03d.%03d", b[0] & 0xFF,
893 893 b[1] & 0xFF, b[2] & 0xFF, b[3] & 0xFF);
894 894 } else if (ca->sa_family == AF_INET6) {
895 895 struct sockaddr_in6 *sin6;
896 896
897 897 sin6 = (struct sockaddr_in6 *)ca;
898 898 (void) kinet_ntop6((uchar_t *)&sin6->sin6_addr,
899 899 buf, INET6_ADDRSTRLEN);
900 900 }
901 901
902 902 (void) snprintf(leaf, MAXNAMELEN, "%s-%llx", buf,
903 903 (longlong_t)cp->rc_clientid);
904 904 rfs4_ss_clid_write(nsrv4, cp, leaf);
905 905 }
906 906
907 907 /*
908 908 * Place client information into stable storage: 2/3.
909 909 * DSS: distributed stable storage: the file may need to be written to
910 910 * multiple directories.
911 911 */
912 912 static void
913 913 rfs4_ss_clid_write(nfs4_srv_t *nsrv4, rfs4_client_t *cp, char *leaf)
914 914 {
915 915 rfs4_servinst_t *sip;
916 916
917 917 /*
918 918 * It should be sufficient to write the leaf file to (all) DSS paths
919 919 * associated with just this client's instance. However, since our
920 920 * per-instance client grouping is solely temporal, HA-NFSv4 RG
921 921 * failover might result in us losing DSS data.
922 922 *
923 923 * Until the client grouping is improved, we must write the DSS data
924 924 * to all instances' paths. Start at the current instance, and
925 925 * walk the list backwards to the first.
926 926 */
927 927 mutex_enter(&nsrv4->servinst_lock);
928 928 for (sip = nsrv4->nfs4_cur_servinst; sip != NULL; sip = sip->prev) {
929 929 int i, npaths = sip->dss_npaths;
930 930
931 931 /* write the leaf file to all DSS paths */
932 932 for (i = 0; i < npaths; i++) {
933 933 rfs4_dss_path_t *dss_path = sip->dss_paths[i];
934 934
935 935 /* HA-NFSv4 path might have been failed-away from us */
936 936 if (dss_path == NULL)
937 937 continue;
938 938
939 939 rfs4_ss_clid_write_one(cp, dss_path->path, leaf);
940 940 }
941 941 }
942 942 mutex_exit(&nsrv4->servinst_lock);
943 943 }
944 944
945 945 /*
946 946 * Place client information into stable storage: 3/3.
947 947 * Write the stable storage data to the requested file.
948 948 */
949 949 static void
950 950 rfs4_ss_clid_write_one(rfs4_client_t *cp, char *dss_path, char *leaf)
951 951 {
952 952 int ioflag;
953 953 int file_vers = NFS4_SS_VERSION;
954 954 size_t dirlen;
955 955 struct uio uio;
956 956 struct iovec iov[4];
957 957 char *dir;
958 958 rfs4_ss_pn_t *ss_pn;
959 959 vnode_t *vp;
960 960 nfs_client_id4 *cl_id4 = &(cp->rc_nfs_client);
961 961
962 962 /* allow 2 extra bytes for '/' & NUL */
963 963 dirlen = strlen(dss_path) + strlen(NFS4_DSS_STATE_LEAF) + 2;
964 964 dir = kmem_alloc(dirlen, KM_SLEEP);
965 965 (void) sprintf(dir, "%s/%s", dss_path, NFS4_DSS_STATE_LEAF);
966 966
967 967 ss_pn = rfs4_ss_pnalloc(dir, leaf);
968 968 /* rfs4_ss_pnalloc takes its own copy */
969 969 kmem_free(dir, dirlen);
970 970 if (ss_pn == NULL)
971 971 return;
972 972
973 973 if (vn_open(ss_pn->pn, UIO_SYSSPACE, FCREAT|FWRITE, 0600, &vp,
974 974 CRCREAT, 0)) {
975 975 rfs4_ss_pnfree(ss_pn);
976 976 return;
977 977 }
978 978
979 979 /*
980 980 * We need to record leaf - i.e. the filename - so that we know
981 981 * what to remove, in the future. However, the dir part of cp->ss_pn
982 982 * should never be referenced directly, since it's potentially only
983 983 * one of several paths with this leaf in it.
984 984 */
985 985 if (cp->rc_ss_pn != NULL) {
986 986 if (strcmp(cp->rc_ss_pn->leaf, leaf) == 0) {
987 987 /* we've already recorded *this* leaf */
988 988 rfs4_ss_pnfree(ss_pn);
989 989 } else {
990 990 /* replace with this leaf */
991 991 rfs4_ss_pnfree(cp->rc_ss_pn);
992 992 cp->rc_ss_pn = ss_pn;
993 993 }
994 994 } else {
995 995 cp->rc_ss_pn = ss_pn;
996 996 }
997 997
998 998 /*
999 999 * Build a scatter list that points to the nfs_client_id4
1000 1000 */
1001 1001 iov[0].iov_base = (caddr_t)&file_vers;
1002 1002 iov[0].iov_len = sizeof (int);
1003 1003 iov[1].iov_base = (caddr_t)&(cl_id4->verifier);
1004 1004 iov[1].iov_len = NFS4_VERIFIER_SIZE;
1005 1005 iov[2].iov_base = (caddr_t)&(cl_id4->id_len);
1006 1006 iov[2].iov_len = sizeof (uint_t);
1007 1007 iov[3].iov_base = (caddr_t)cl_id4->id_val;
1008 1008 iov[3].iov_len = cl_id4->id_len;
1009 1009
1010 1010 uio.uio_iov = iov;
1011 1011 uio.uio_iovcnt = 4;
1012 1012 uio.uio_loffset = 0;
1013 1013 uio.uio_segflg = UIO_SYSSPACE;
1014 1014 uio.uio_llimit = (rlim64_t)MAXOFFSET_T;
1015 1015 uio.uio_resid = cl_id4->id_len + sizeof (int) +
1016 1016 NFS4_VERIFIER_SIZE + sizeof (uint_t);
1017 1017
1018 1018 ioflag = uio.uio_fmode = (FWRITE|FSYNC);
1019 1019 uio.uio_extflg = UIO_COPY_DEFAULT;
1020 1020
1021 1021 (void) VOP_RWLOCK(vp, V_WRITELOCK_TRUE, NULL);
1022 1022 /* write the full client id to the file. */
1023 1023 (void) VOP_WRITE(vp, &uio, ioflag, CRED(), NULL);
1024 1024 VOP_RWUNLOCK(vp, V_WRITELOCK_TRUE, NULL);
1025 1025
1026 1026 (void) VOP_CLOSE(vp, FWRITE, 1, (offset_t)0, CRED(), NULL);
1027 1027 VN_RELE(vp);
1028 1028 }
1029 1029
1030 1030 /*
1031 1031 * DSS: distributed stable storage.
1032 1032 * Unpack the list of paths passed by nfsd.
1033 1033 * Use nvlist_alloc(9F) to manage the data.
1034 1034 * The caller is responsible for allocating and freeing the buffer.
1035 1035 */
1036 1036 int
1037 1037 rfs4_dss_setpaths(char *buf, size_t buflen)
1038 1038 {
1039 1039 int error;
1040 1040
1041 1041 /*
1042 1042 * If this is a "warm start", i.e. we previously had DSS paths,
1043 1043 * preserve the old paths.
1044 1044 */
1045 1045 if (rfs4_dss_paths != NULL) {
1046 1046 /*
1047 1047 * Before we lose the ptr, destroy the nvlist and pathnames
1048 1048 * array from the warm start before this one.
1049 1049 */
1050 1050 nvlist_free(rfs4_dss_oldpaths);
1051 1051 rfs4_dss_oldpaths = rfs4_dss_paths;
1052 1052 }
1053 1053
1054 1054 /* unpack the buffer into a searchable nvlist */
1055 1055 error = nvlist_unpack(buf, buflen, &rfs4_dss_paths, KM_SLEEP);
1056 1056 if (error)
1057 1057 return (error);
1058 1058
1059 1059 /*
1060 1060 * Search the nvlist for the pathnames nvpair (which is the only nvpair
1061 1061 * in the list, and record its location.
1062 1062 */
1063 1063 error = nvlist_lookup_string_array(rfs4_dss_paths, NFS4_DSS_NVPAIR_NAME,
1064 1064 &rfs4_dss_newpaths, &rfs4_dss_numnewpaths);
1065 1065 return (error);
1066 1066 }
1067 1067
1068 1068 /*
1069 1069 * Ultimately the nfssys() call NFS4_CLR_STATE endsup here
1070 1070 * to find and mark the client for forced expire.
1071 1071 */
1072 1072 static void
1073 1073 rfs4_client_scrub(rfs4_entry_t ent, void *arg)
1074 1074 {
1075 1075 rfs4_client_t *cp = (rfs4_client_t *)ent;
1076 1076 struct nfs4clrst_args *clr = arg;
1077 1077 struct sockaddr_in6 *ent_sin6;
1078 1078 struct in6_addr clr_in6;
1079 1079 struct sockaddr_in *ent_sin;
1080 1080 struct in_addr clr_in;
1081 1081
1082 1082 if (clr->addr_type != cp->rc_addr.ss_family) {
1083 1083 return;
1084 1084 }
1085 1085
1086 1086 switch (clr->addr_type) {
1087 1087
1088 1088 case AF_INET6:
1089 1089 /* copyin the address from user space */
1090 1090 if (copyin(clr->ap, &clr_in6, sizeof (clr_in6))) {
1091 1091 break;
1092 1092 }
1093 1093
1094 1094 ent_sin6 = (struct sockaddr_in6 *)&cp->rc_addr;
1095 1095
1096 1096 /*
1097 1097 * now compare, and if equivalent mark entry
1098 1098 * for forced expiration
1099 1099 */
1100 1100 if (IN6_ARE_ADDR_EQUAL(&ent_sin6->sin6_addr, &clr_in6)) {
1101 1101 cp->rc_forced_expire = 1;
1102 1102 }
1103 1103 break;
1104 1104
1105 1105 case AF_INET:
1106 1106 /* copyin the address from user space */
1107 1107 if (copyin(clr->ap, &clr_in, sizeof (clr_in))) {
1108 1108 break;
1109 1109 }
1110 1110
1111 1111 ent_sin = (struct sockaddr_in *)&cp->rc_addr;
1112 1112
1113 1113 /*
1114 1114 * now compare, and if equivalent mark entry
1115 1115 * for forced expiration
1116 1116 */
1117 1117 if (ent_sin->sin_addr.s_addr == clr_in.s_addr) {
1118 1118 cp->rc_forced_expire = 1;
1119 1119 }
1120 1120 break;
1121 1121
1122 1122 default:
1123 1123 /* force this assert to fail */
1124 1124 ASSERT(clr->addr_type != clr->addr_type);
1125 1125 }
1126 1126 }
1127 1127
1128 1128 /*
1129 1129 * This is called from nfssys() in order to clear server state
1130 1130 * for the specified client IP Address.
1131 1131 */
1132 1132 void
1133 1133 rfs4_clear_client_state(struct nfs4clrst_args *clr)
1134 1134 {
1135 1135 nfs4_srv_t *nsrv4;
1136 1136 nsrv4 = nfs4_get_srv();
1137 1137 (void) rfs4_dbe_walk(nsrv4->rfs4_client_tab, rfs4_client_scrub, clr);
1138 1138 }
1139 1139
1140 1140 /*
1141 1141 * Used to initialize the NFSv4 server's state or database. All of
1142 1142 * the tables are created and timers are set.
1143 1143 */
1144 1144 void
1145 1145 rfs4_state_g_init()
1146 1146 {
1147 1147 extern boolean_t rfs4_cpr_callb(void *, int);
1148 1148 /*
1149 1149 * Add a CPR callback so that we can update client
1150 1150 * access times to extend the lease after a suspend
1151 1151 * and resume (using the same class as rpcmod/connmgr)
1152 1152 */
1153 1153 cpr_id = callb_add(rfs4_cpr_callb, 0, CB_CL_CPR_RPC, "rfs4");
1154 1154
1155 1155 /*
1156 1156 * NFSv4 server state databases
1157 1157 *
1158 1158 * Initialized when the module is loaded and used by NFSv4 state
1159 1159 * tables. These kmem_cache free pools are used globally, the NFSv4
1160 1160 * state tables which make use of these kmem_cache free pools are per
1161 1161 * zone.
1162 1162 *
1163 1163 * initialize the global kmem_cache free pools which will be used by
1164 1164 * the NFSv4 state tables.
1165 1165 */
1166 1166 /* CSTYLED */
1167 1167 rfs4_client_mem_cache = nfs4_init_mem_cache("Client_entry_cache", 2, sizeof (rfs4_client_t), 0);
1168 1168 /* CSTYLED */
1169 1169 rfs4_clntIP_mem_cache = nfs4_init_mem_cache("ClntIP_entry_cache", 1, sizeof (rfs4_clntip_t), 1);
1170 1170 /* CSTYLED */
1171 1171 rfs4_openown_mem_cache = nfs4_init_mem_cache("OpenOwner_entry_cache", 1, sizeof (rfs4_openowner_t), 2);
1172 1172 /* CSTYLED */
1173 1173 rfs4_openstID_mem_cache = nfs4_init_mem_cache("OpenStateID_entry_cache", 3, sizeof (rfs4_state_t), 3);
1174 1174 /* CSTYLED */
1175 1175 rfs4_lockstID_mem_cache = nfs4_init_mem_cache("LockStateID_entry_cache", 3, sizeof (rfs4_lo_state_t), 4);
1176 1176 /* CSTYLED */
1177 1177 rfs4_lockown_mem_cache = nfs4_init_mem_cache("Lockowner_entry_cache", 2, sizeof (rfs4_lockowner_t), 5);
1178 1178 /* CSTYLED */
1179 1179 rfs4_file_mem_cache = nfs4_init_mem_cache("File_entry_cache", 1, sizeof (rfs4_file_t), 6);
1180 1180 /* CSTYLED */
1181 1181 rfs4_delegstID_mem_cache = nfs4_init_mem_cache("DelegStateID_entry_cache", 2, sizeof (rfs4_deleg_state_t), 7);
1182 1182
1183 1183 rfs4_client_clrst = rfs4_clear_client_state;
1184 1184 }
1185 1185
1186 1186
1187 1187 /*
1188 1188 * Used at server shutdown to cleanup all of the NFSv4 server's structures
1189 1189 * and other state.
1190 1190 */
1191 1191 void
1192 1192 rfs4_state_g_fini()
1193 1193 {
1194 1194 int i;
1195 1195 /*
1196 1196 * Cleanup the CPR callback.
1197 1197 */
1198 1198 if (cpr_id)
1199 1199 (void) callb_delete(cpr_id);
1200 1200
1201 1201 rfs4_client_clrst = NULL;
1202 1202
1203 1203 /* free the NFSv4 state databases */
1204 1204 for (i = 0; i < RFS4_DB_MEM_CACHE_NUM; i++) {
1205 1205 kmem_cache_destroy(rfs4_db_mem_cache_table[i].r_db_mem_cache);
1206 1206 rfs4_db_mem_cache_table[i].r_db_mem_cache = NULL;
1207 1207 }
1208 1208
1209 1209 rfs4_client_mem_cache = NULL;
1210 1210 rfs4_clntIP_mem_cache = NULL;
1211 1211 rfs4_openown_mem_cache = NULL;
1212 1212 rfs4_openstID_mem_cache = NULL;
1213 1213 rfs4_lockstID_mem_cache = NULL;
1214 1214 rfs4_lockown_mem_cache = NULL;
1215 1215 rfs4_file_mem_cache = NULL;
1216 1216 rfs4_delegstID_mem_cache = NULL;
1217 1217
1218 1218 /* DSS: distributed stable storage */
1219 1219 nvlist_free(rfs4_dss_oldpaths);
1220 1220 nvlist_free(rfs4_dss_paths);
1221 1221 rfs4_dss_paths = rfs4_dss_oldpaths = NULL;
1222 1222 }
1223 1223
1224 1224 /*
1225 1225 * Used to initialize the per zone NFSv4 server's state
1226 1226 */
1227 1227 void
1228 1228 rfs4_state_zone_init(nfs4_srv_t *nsrv4)
1229 1229 {
1230 1230 time_t start_time;
1231 1231 int start_grace;
1232 1232 char *dss_path = NFS4_DSS_VAR_DIR;
1233 1233
1234 1234 /* DSS: distributed stable storage: initialise served paths list */
1235 1235 nsrv4->dss_pathlist = NULL;
1236 1236
1237 1237 /*
1238 1238 * Set the boot time. If the server
1239 1239 * has been restarted quickly and has had the opportunity to
1240 1240 * service clients, then the start_time needs to be bumped
1241 1241 * regardless. A small window but it exists...
1242 1242 */
1243 1243 start_time = gethrestime_sec();
1244 1244 if (nsrv4->rfs4_start_time < start_time)
1245 1245 nsrv4->rfs4_start_time = start_time;
1246 1246 else
1247 1247 nsrv4->rfs4_start_time++;
1248 1248
1249 1249 /*
1250 1250 * Create the first server instance, or a new one if the server has
1251 1251 * been restarted; see above comments on rfs4_start_time. Don't
1252 1252 * start its grace period; that will be done later, to maximise the
1253 1253 * clients' recovery window.
1254 1254 */
1255 1255 start_grace = 0;
1256 1256 if (curzone == global_zone && rfs4_dss_numnewpaths > 0) {
1257 1257 int i;
1258 1258 char **dss_allpaths = NULL;
1259 1259 dss_allpaths = kmem_alloc(sizeof (char *) * (rfs4_dss_numnewpaths + 1), KM_SLEEP);
1260 1260 /*
1261 1261 * Add the default path into the list of paths for saving
1262 1262 * state informantion.
1263 1263 */
1264 1264 dss_allpaths[0] = dss_path;
1265 1265 for ( i = 0; i < rfs4_dss_numnewpaths; i++) {
1266 1266 dss_allpaths[i + 1] = rfs4_dss_newpaths[i];
1267 1267 }
1268 1268 rfs4_servinst_create(nsrv4, start_grace, (rfs4_dss_numnewpaths + 1), dss_allpaths);
1269 1269 kmem_free(dss_allpaths, (sizeof (char *) * (rfs4_dss_numnewpaths + 1)));
1270 1270 } else {
1271 1271 rfs4_servinst_create(nsrv4, start_grace, 1, &dss_path);
1272 1272 }
1273 1273
1274 1274 /* reset the "first NFSv4 request" status */
1275 1275 nsrv4->seen_first_compound = 0;
1276 1276
1277 1277 mutex_enter(&nsrv4->state_lock);
1278 1278
1279 1279 /*
1280 1280 * If the server state database has already been initialized,
1281 1281 * skip it
1282 1282 */
1283 1283 if (nsrv4->nfs4_server_state != NULL) {
1284 1284 mutex_exit(&nsrv4->state_lock);
1285 1285 return;
1286 1286 }
1287 1287
1288 1288 rw_init(&nsrv4->rfs4_findclient_lock, NULL, RW_DEFAULT, NULL);
1289 1289
1290 1290 /* set the various cache timers for table creation */
1291 1291 if (nsrv4->rfs4_client_cache_time == 0)
1292 1292 nsrv4->rfs4_client_cache_time = CLIENT_CACHE_TIME;
1293 1293 if (nsrv4->rfs4_openowner_cache_time == 0)
1294 1294 nsrv4->rfs4_openowner_cache_time = OPENOWNER_CACHE_TIME;
1295 1295 if (nsrv4->rfs4_state_cache_time == 0)
1296 1296 nsrv4->rfs4_state_cache_time = STATE_CACHE_TIME;
1297 1297 if (nsrv4->rfs4_lo_state_cache_time == 0)
1298 1298 nsrv4->rfs4_lo_state_cache_time = LO_STATE_CACHE_TIME;
1299 1299 if (nsrv4->rfs4_lockowner_cache_time == 0)
1300 1300 nsrv4->rfs4_lockowner_cache_time = LOCKOWNER_CACHE_TIME;
1301 1301 if (nsrv4->rfs4_file_cache_time == 0)
1302 1302 nsrv4->rfs4_file_cache_time = FILE_CACHE_TIME;
1303 1303 if (nsrv4->rfs4_deleg_state_cache_time == 0)
1304 1304 nsrv4->rfs4_deleg_state_cache_time = DELEG_STATE_CACHE_TIME;
1305 1305
1306 1306 /* Create the overall database to hold all server state */
1307 1307 nsrv4->nfs4_server_state = rfs4_database_create(rfs4_database_debug);
1308 1308
1309 1309 /* Now create the individual tables */
1310 1310 nsrv4->rfs4_client_cache_time *= rfs4_lease_time;
1311 1311 nsrv4->rfs4_client_tab = rfs4_table_create(nsrv4->nfs4_server_state,
1312 1312 "Client",
1313 1313 nsrv4->rfs4_client_cache_time,
1314 1314 2,
1315 1315 rfs4_client_create,
1316 1316 rfs4_client_destroy,
1317 1317 rfs4_client_expiry,
1318 1318 sizeof (rfs4_client_t),
1319 1319 TABSIZE,
1320 1320 MAXTABSZ/8, 100);
1321 1321 nsrv4->rfs4_nfsclnt_idx = rfs4_index_create(nsrv4->rfs4_client_tab,
1322 1322 "nfs_client_id4", nfsclnt_hash,
1323 1323 nfsclnt_compare, nfsclnt_mkkey,
1324 1324 TRUE);
1325 1325 nsrv4->rfs4_clientid_idx = rfs4_index_create(nsrv4->rfs4_client_tab,
1326 1326 "client_id", clientid_hash,
1327 1327 clientid_compare, clientid_mkkey,
1328 1328 FALSE);
1329 1329
1330 1330 nsrv4->rfs4_clntip_cache_time = 86400 * 365; /* about a year */
1331 1331 nsrv4->rfs4_clntip_tab = rfs4_table_create(nsrv4->nfs4_server_state,
1332 1332 "ClntIP",
1333 1333 nsrv4->rfs4_clntip_cache_time,
1334 1334 1,
1335 1335 rfs4_clntip_create,
1336 1336 rfs4_clntip_destroy,
1337 1337 rfs4_clntip_expiry,
1338 1338 sizeof (rfs4_clntip_t),
1339 1339 TABSIZE,
1340 1340 MAXTABSZ, 100);
1341 1341 nsrv4->rfs4_clntip_idx = rfs4_index_create(nsrv4->rfs4_clntip_tab,
1342 1342 "client_ip", clntip_hash,
1343 1343 clntip_compare, clntip_mkkey,
1344 1344 TRUE);
1345 1345
1346 1346 nsrv4->rfs4_openowner_cache_time *= rfs4_lease_time;
1347 1347 nsrv4->rfs4_openowner_tab = rfs4_table_create(nsrv4->nfs4_server_state,
1348 1348 "OpenOwner",
1349 1349 nsrv4->rfs4_openowner_cache_time,
1350 1350 1,
1351 1351 rfs4_openowner_create,
1352 1352 rfs4_openowner_destroy,
1353 1353 rfs4_openowner_expiry,
1354 1354 sizeof (rfs4_openowner_t),
1355 1355 TABSIZE,
1356 1356 MAXTABSZ, 100);
1357 1357 nsrv4->rfs4_openowner_idx = rfs4_index_create(nsrv4->rfs4_openowner_tab,
1358 1358 "open_owner4", openowner_hash,
1359 1359 openowner_compare,
1360 1360 openowner_mkkey, TRUE);
1361 1361
1362 1362 nsrv4->rfs4_state_cache_time *= rfs4_lease_time;
1363 1363 nsrv4->rfs4_state_tab = rfs4_table_create(nsrv4->nfs4_server_state,
1364 1364 "OpenStateID",
1365 1365 nsrv4->rfs4_state_cache_time,
1366 1366 3,
1367 1367 rfs4_state_create,
1368 1368 rfs4_state_destroy,
1369 1369 rfs4_state_expiry,
1370 1370 sizeof (rfs4_state_t),
1371 1371 TABSIZE,
1372 1372 MAXTABSZ, 100);
1373 1373
1374 1374 /* CSTYLED */
1375 1375 nsrv4->rfs4_state_owner_file_idx = rfs4_index_create(nsrv4->rfs4_state_tab,
1376 1376 "Openowner-File",
1377 1377 state_owner_file_hash,
1378 1378 state_owner_file_compare,
1379 1379 state_owner_file_mkkey, TRUE);
1380 1380
1381 1381 nsrv4->rfs4_state_idx = rfs4_index_create(nsrv4->rfs4_state_tab,
1382 1382 "State-id", state_hash,
1383 1383 state_compare, state_mkkey, FALSE);
1384 1384
1385 1385 nsrv4->rfs4_state_file_idx = rfs4_index_create(nsrv4->rfs4_state_tab,
1386 1386 "File", state_file_hash,
1387 1387 state_file_compare, state_file_mkkey,
1388 1388 FALSE);
1389 1389
1390 1390 nsrv4->rfs4_lo_state_cache_time *= rfs4_lease_time;
1391 1391 nsrv4->rfs4_lo_state_tab = rfs4_table_create(nsrv4->nfs4_server_state,
1392 1392 "LockStateID",
1393 1393 nsrv4->rfs4_lo_state_cache_time,
1394 1394 2,
1395 1395 rfs4_lo_state_create,
1396 1396 rfs4_lo_state_destroy,
1397 1397 rfs4_lo_state_expiry,
1398 1398 sizeof (rfs4_lo_state_t),
1399 1399 TABSIZE,
1400 1400 MAXTABSZ, 100);
1401 1401
1402 1402 /* CSTYLED */
1403 1403 nsrv4->rfs4_lo_state_owner_idx = rfs4_index_create(nsrv4->rfs4_lo_state_tab,
1404 1404 "lockownerxstate",
1405 1405 lo_state_lo_hash,
1406 1406 lo_state_lo_compare,
1407 1407 lo_state_lo_mkkey, TRUE);
1408 1408
1409 1409 nsrv4->rfs4_lo_state_idx = rfs4_index_create(nsrv4->rfs4_lo_state_tab,
1410 1410 "State-id",
1411 1411 lo_state_hash, lo_state_compare,
1412 1412 lo_state_mkkey, FALSE);
1413 1413
1414 1414 nsrv4->rfs4_lockowner_cache_time *= rfs4_lease_time;
1415 1415
1416 1416 nsrv4->rfs4_lockowner_tab = rfs4_table_create(nsrv4->nfs4_server_state,
1417 1417 "Lockowner",
1418 1418 nsrv4->rfs4_lockowner_cache_time,
1419 1419 2,
1420 1420 rfs4_lockowner_create,
1421 1421 rfs4_lockowner_destroy,
1422 1422 rfs4_lockowner_expiry,
1423 1423 sizeof (rfs4_lockowner_t),
1424 1424 TABSIZE,
1425 1425 MAXTABSZ, 100);
1426 1426
1427 1427 nsrv4->rfs4_lockowner_idx = rfs4_index_create(nsrv4->rfs4_lockowner_tab,
1428 1428 "lock_owner4", lockowner_hash,
1429 1429 lockowner_compare,
1430 1430 lockowner_mkkey, TRUE);
1431 1431
1432 1432 /* CSTYLED */
1433 1433 nsrv4->rfs4_lockowner_pid_idx = rfs4_index_create(nsrv4->rfs4_lockowner_tab,
1434 1434 "pid", pid_hash,
1435 1435 pid_compare, pid_mkkey,
1436 1436 FALSE);
1437 1437
1438 1438 nsrv4->rfs4_file_cache_time *= rfs4_lease_time;
1439 1439 nsrv4->rfs4_file_tab = rfs4_table_create(nsrv4->nfs4_server_state,
1440 1440 "File",
1441 1441 nsrv4->rfs4_file_cache_time,
1442 1442 1,
1443 1443 rfs4_file_create,
1444 1444 rfs4_file_destroy,
1445 1445 NULL,
1446 1446 sizeof (rfs4_file_t),
1447 1447 TABSIZE,
1448 1448 MAXTABSZ, -1);
1449 1449
1450 1450 nsrv4->rfs4_file_idx = rfs4_index_create(nsrv4->rfs4_file_tab,
1451 1451 "Filehandle", file_hash,
1452 1452 file_compare, file_mkkey, TRUE);
1453 1453
1454 1454 nsrv4->rfs4_deleg_state_cache_time *= rfs4_lease_time;
1455 1455 /* CSTYLED */
1456 1456 nsrv4->rfs4_deleg_state_tab = rfs4_table_create(nsrv4->nfs4_server_state,
1457 1457 "DelegStateID",
1458 1458 nsrv4->rfs4_deleg_state_cache_time,
1459 1459 2,
1460 1460 rfs4_deleg_state_create,
1461 1461 rfs4_deleg_state_destroy,
1462 1462 rfs4_deleg_state_expiry,
1463 1463 sizeof (rfs4_deleg_state_t),
1464 1464 TABSIZE,
1465 1465 MAXTABSZ, 100);
1466 1466 nsrv4->rfs4_deleg_idx = rfs4_index_create(nsrv4->rfs4_deleg_state_tab,
1467 1467 "DelegByFileClient",
1468 1468 deleg_hash,
1469 1469 deleg_compare,
1470 1470 deleg_mkkey, TRUE);
1471 1471
1472 1472 /* CSTYLED */
1473 1473 nsrv4->rfs4_deleg_state_idx = rfs4_index_create(nsrv4->rfs4_deleg_state_tab,
1474 1474 "DelegState",
1475 1475 deleg_state_hash,
1476 1476 deleg_state_compare,
1477 1477 deleg_state_mkkey, FALSE);
1478 1478
1479 1479 mutex_exit(&nsrv4->state_lock);
1480 1480
1481 1481 /*
1482 1482 * Init the stable storage.
1483 1483 */
1484 1484 rfs4_ss_init(nsrv4);
1485 1485 }
1486 1486
1487 1487 /*
1488 1488 * Used at server shutdown to cleanup all of NFSv4 server's zone structures
1489 1489 * and state.
1490 1490 */
1491 1491 void
1492 1492 rfs4_state_zone_fini()
1493 1493 {
1494 1494 rfs4_database_t *dbp;
1495 1495 nfs4_srv_t *nsrv4;
1496 1496 nsrv4 = nfs4_get_srv();
1497 1497
1498 1498 rfs4_set_deleg_policy(nsrv4, SRV_NEVER_DELEGATE);
1499 1499
1500 1500 /*
1501 1501 * Clean up any dangling stable storage structures BEFORE calling
1502 1502 * rfs4_servinst_destroy_all() so there are no dangling structures
1503 1503 * (i.e. the srvinsts are all cleared of danglers BEFORE they get
1504 1504 * freed).
1505 1505 */
1506 1506 rfs4_ss_fini(nsrv4);
1507 1507
1508 1508 mutex_enter(&nsrv4->state_lock);
1509 1509
1510 1510 if (nsrv4->nfs4_server_state == NULL) {
1511 1511 mutex_exit(&nsrv4->state_lock);
1512 1512 return;
1513 1513 }
1514 1514
1515 1515 /* destroy server instances and current instance ptr */
1516 1516 rfs4_servinst_destroy_all(nsrv4);
1517 1517
|
↓ open down ↓ |
1517 lines elided |
↑ open up ↑ |
1518 1518 /* reset the "first NFSv4 request" status */
1519 1519 nsrv4->seen_first_compound = 0;
1520 1520
1521 1521 dbp = nsrv4->nfs4_server_state;
1522 1522 nsrv4->nfs4_server_state = NULL;
1523 1523
1524 1524 rw_destroy(&nsrv4->rfs4_findclient_lock);
1525 1525
1526 1526 /* First stop all of the reaper threads in the database */
1527 1527 rfs4_database_shutdown(dbp);
1528 +
1528 1529 /*
1529 1530 * WARNING: There may be consumers of the rfs4 database still
1530 1531 * active as we destroy these. IF that's the case, consider putting
1531 1532 * some of their _zone_fini()-like functions into the zsd key as
1532 1533 * ~~SHUTDOWN~~ functions instead of ~~DESTROY~~ functions. We can
1533 1534 * maintain some ordering guarantees better that way.
1534 1535 */
1535 1536 /* Now destroy/release the database tables */
1536 1537 rfs4_database_destroy(dbp);
1537 1538
1538 1539 /* Reset the cache timers for next time */
1539 1540 nsrv4->rfs4_client_cache_time = 0;
1540 1541 nsrv4->rfs4_openowner_cache_time = 0;
1541 1542 nsrv4->rfs4_state_cache_time = 0;
1542 1543 nsrv4->rfs4_lo_state_cache_time = 0;
1543 1544 nsrv4->rfs4_lockowner_cache_time = 0;
1544 1545 nsrv4->rfs4_file_cache_time = 0;
1545 1546 nsrv4->rfs4_deleg_state_cache_time = 0;
1546 1547
1547 1548 mutex_exit(&nsrv4->state_lock);
1548 1549 }
1549 1550
1550 1551 typedef union {
1551 1552 struct {
1552 1553 uint32_t start_time;
1553 1554 uint32_t c_id;
1554 1555 } impl_id;
1555 1556 clientid4 id4;
1556 1557 } cid;
1557 1558
1558 1559 static int foreign_stateid(stateid_t *id);
1559 1560 static int foreign_clientid(cid *cidp);
1560 1561 static void embed_nodeid(cid *cidp);
1561 1562
1562 1563 typedef union {
1563 1564 struct {
1564 1565 uint32_t c_id;
1565 1566 uint32_t gen_num;
1566 1567 } cv_impl;
1567 1568 verifier4 confirm_verf;
1568 1569 } scid_confirm_verf;
1569 1570
1570 1571 static uint32_t
1571 1572 clientid_hash(void *key)
1572 1573 {
1573 1574 cid *idp = key;
1574 1575
1575 1576 return (idp->impl_id.c_id);
1576 1577 }
1577 1578
1578 1579 static bool_t
1579 1580 clientid_compare(rfs4_entry_t entry, void *key)
1580 1581 {
1581 1582 rfs4_client_t *cp = (rfs4_client_t *)entry;
1582 1583 clientid4 *idp = key;
1583 1584
1584 1585 return (*idp == cp->rc_clientid);
1585 1586 }
1586 1587
1587 1588 static void *
1588 1589 clientid_mkkey(rfs4_entry_t entry)
1589 1590 {
1590 1591 rfs4_client_t *cp = (rfs4_client_t *)entry;
1591 1592
1592 1593 return (&cp->rc_clientid);
1593 1594 }
1594 1595
1595 1596 static uint32_t
1596 1597 nfsclnt_hash(void *key)
1597 1598 {
1598 1599 nfs_client_id4 *client = key;
1599 1600 int i;
1600 1601 uint32_t hash = 0;
1601 1602
1602 1603 for (i = 0; i < client->id_len; i++) {
1603 1604 hash <<= 1;
1604 1605 hash += (uint_t)client->id_val[i];
1605 1606 }
1606 1607 return (hash);
1607 1608 }
1608 1609
1609 1610
1610 1611 static bool_t
1611 1612 nfsclnt_compare(rfs4_entry_t entry, void *key)
1612 1613 {
1613 1614 rfs4_client_t *cp = (rfs4_client_t *)entry;
1614 1615 nfs_client_id4 *nfs_client = key;
1615 1616
1616 1617 if (cp->rc_nfs_client.id_len != nfs_client->id_len)
1617 1618 return (FALSE);
1618 1619
1619 1620 return (bcmp(cp->rc_nfs_client.id_val, nfs_client->id_val,
1620 1621 nfs_client->id_len) == 0);
1621 1622 }
1622 1623
1623 1624 static void *
1624 1625 nfsclnt_mkkey(rfs4_entry_t entry)
1625 1626 {
1626 1627 rfs4_client_t *cp = (rfs4_client_t *)entry;
1627 1628
1628 1629 return (&cp->rc_nfs_client);
1629 1630 }
1630 1631
1631 1632 static bool_t
1632 1633 rfs4_client_expiry(rfs4_entry_t u_entry)
1633 1634 {
1634 1635 rfs4_client_t *cp = (rfs4_client_t *)u_entry;
1635 1636 bool_t cp_expired;
1636 1637
1637 1638 if (rfs4_dbe_is_invalid(cp->rc_dbe)) {
1638 1639 cp->rc_ss_remove = 1;
1639 1640 return (TRUE);
1640 1641 }
1641 1642 /*
1642 1643 * If the sysadmin has used clear_locks for this
1643 1644 * entry then forced_expire will be set and we
1644 1645 * want this entry to be reaped. Or the entry
1645 1646 * has exceeded its lease period.
1646 1647 */
1647 1648 cp_expired = (cp->rc_forced_expire ||
1648 1649 (gethrestime_sec() - cp->rc_last_access
1649 1650 > rfs4_lease_time));
1650 1651
1651 1652 if (!cp->rc_ss_remove && cp_expired)
1652 1653 cp->rc_ss_remove = 1;
1653 1654 return (cp_expired);
1654 1655 }
1655 1656
1656 1657 /*
1657 1658 * Remove the leaf file from all distributed stable storage paths.
1658 1659 */
1659 1660 static void
1660 1661 rfs4_dss_remove_cpleaf(rfs4_client_t *cp)
1661 1662 {
1662 1663 nfs4_srv_t *nsrv4;
1663 1664 rfs4_servinst_t *sip;
1664 1665 char *leaf = cp->rc_ss_pn->leaf;
1665 1666
1666 1667 /*
1667 1668 * since the state files are written to all DSS
1668 1669 * paths we must remove this leaf file instance
1669 1670 * from all server instances.
1670 1671 */
1671 1672
1672 1673 nsrv4 = nfs4_get_srv();
1673 1674 mutex_enter(&nsrv4->servinst_lock);
1674 1675 for (sip = nsrv4->nfs4_cur_servinst; sip != NULL; sip = sip->prev) {
1675 1676 /* remove the leaf file associated with this server instance */
1676 1677 rfs4_dss_remove_leaf(sip, NFS4_DSS_STATE_LEAF, leaf);
1677 1678 }
1678 1679 mutex_exit(&nsrv4->servinst_lock);
1679 1680 }
1680 1681
1681 1682 static void
1682 1683 rfs4_dss_remove_leaf(rfs4_servinst_t *sip, char *dir_leaf, char *leaf)
1683 1684 {
1684 1685 int i, npaths = sip->dss_npaths;
1685 1686
1686 1687 for (i = 0; i < npaths; i++) {
1687 1688 rfs4_dss_path_t *dss_path = sip->dss_paths[i];
1688 1689 char *path, *dir;
1689 1690 size_t pathlen;
1690 1691
1691 1692 /* the HA-NFSv4 path might have been failed-over away from us */
1692 1693 if (dss_path == NULL)
1693 1694 continue;
1694 1695
1695 1696 dir = dss_path->path;
1696 1697
1697 1698 /* allow 3 extra bytes for two '/' & a NUL */
1698 1699 pathlen = strlen(dir) + strlen(dir_leaf) + strlen(leaf) + 3;
1699 1700 path = kmem_alloc(pathlen, KM_SLEEP);
1700 1701 (void) sprintf(path, "%s/%s/%s", dir, dir_leaf, leaf);
1701 1702
1702 1703 (void) vn_remove(path, UIO_SYSSPACE, RMFILE);
1703 1704
1704 1705 kmem_free(path, pathlen);
1705 1706 }
1706 1707 }
1707 1708
1708 1709 static void
1709 1710 rfs4_client_destroy(rfs4_entry_t u_entry)
1710 1711 {
1711 1712 rfs4_client_t *cp = (rfs4_client_t *)u_entry;
1712 1713
1713 1714 mutex_destroy(cp->rc_cbinfo.cb_lock);
1714 1715 cv_destroy(cp->rc_cbinfo.cb_cv);
1715 1716 cv_destroy(cp->rc_cbinfo.cb_cv_nullcaller);
1716 1717 list_destroy(&cp->rc_openownerlist);
1717 1718
1718 1719 /* free callback info */
1719 1720 rfs4_cbinfo_free(&cp->rc_cbinfo);
1720 1721
1721 1722 if (cp->rc_cp_confirmed)
1722 1723 rfs4_client_rele(cp->rc_cp_confirmed);
1723 1724
1724 1725 if (cp->rc_ss_pn) {
1725 1726 /* check if the stable storage files need to be removed */
1726 1727 if (cp->rc_ss_remove)
1727 1728 rfs4_dss_remove_cpleaf(cp);
1728 1729 rfs4_ss_pnfree(cp->rc_ss_pn);
1729 1730 }
1730 1731
1731 1732 /* Free the client supplied client id */
1732 1733 kmem_free(cp->rc_nfs_client.id_val, cp->rc_nfs_client.id_len);
1733 1734
1734 1735 if (cp->rc_sysidt != LM_NOSYSID)
1735 1736 lm_free_sysidt(cp->rc_sysidt);
1736 1737 }
1737 1738
1738 1739 static bool_t
1739 1740 rfs4_client_create(rfs4_entry_t u_entry, void *arg)
1740 1741 {
1741 1742 rfs4_client_t *cp = (rfs4_client_t *)u_entry;
1742 1743 nfs_client_id4 *client = (nfs_client_id4 *)arg;
1743 1744 struct sockaddr *ca;
1744 1745 cid *cidp;
1745 1746 scid_confirm_verf *scvp;
1746 1747 nfs4_srv_t *nsrv4;
1747 1748
1748 1749 nsrv4 = nfs4_get_srv();
1749 1750
1750 1751 /* Get a clientid to give to the client */
1751 1752 cidp = (cid *)&cp->rc_clientid;
1752 1753 cidp->impl_id.start_time = nsrv4->rfs4_start_time;
1753 1754 cidp->impl_id.c_id = (uint32_t)rfs4_dbe_getid(cp->rc_dbe);
1754 1755
1755 1756 /* If we are booted as a cluster node, embed our nodeid */
1756 1757 if (cluster_bootflags & CLUSTER_BOOTED)
1757 1758 embed_nodeid(cidp);
1758 1759
1759 1760 /* Allocate and copy client's client id value */
1760 1761 cp->rc_nfs_client.id_val = kmem_alloc(client->id_len, KM_SLEEP);
1761 1762 cp->rc_nfs_client.id_len = client->id_len;
1762 1763 bcopy(client->id_val, cp->rc_nfs_client.id_val, client->id_len);
1763 1764 cp->rc_nfs_client.verifier = client->verifier;
1764 1765
1765 1766 /* Copy client's IP address */
1766 1767 ca = client->cl_addr;
1767 1768 if (ca->sa_family == AF_INET)
1768 1769 bcopy(ca, &cp->rc_addr, sizeof (struct sockaddr_in));
1769 1770 else if (ca->sa_family == AF_INET6)
1770 1771 bcopy(ca, &cp->rc_addr, sizeof (struct sockaddr_in6));
1771 1772 cp->rc_nfs_client.cl_addr = (struct sockaddr *)&cp->rc_addr;
1772 1773
1773 1774 /* Init the value for the SETCLIENTID_CONFIRM verifier */
1774 1775 scvp = (scid_confirm_verf *)&cp->rc_confirm_verf;
1775 1776 scvp->cv_impl.c_id = cidp->impl_id.c_id;
1776 1777 scvp->cv_impl.gen_num = 0;
1777 1778
1778 1779 /* An F_UNLKSYS has been done for this client */
1779 1780 cp->rc_unlksys_completed = FALSE;
1780 1781
1781 1782 /* We need the client to ack us */
1782 1783 cp->rc_need_confirm = TRUE;
1783 1784 cp->rc_cp_confirmed = NULL;
1784 1785
1785 1786 /* TRUE all the time until the callback path actually fails */
1786 1787 cp->rc_cbinfo.cb_notified_of_cb_path_down = TRUE;
1787 1788
1788 1789 /* Initialize the access time to now */
1789 1790 cp->rc_last_access = gethrestime_sec();
1790 1791
1791 1792 cp->rc_cr_set = NULL;
1792 1793
1793 1794 cp->rc_sysidt = LM_NOSYSID;
1794 1795
1795 1796 list_create(&cp->rc_openownerlist, sizeof (rfs4_openowner_t),
1796 1797 offsetof(rfs4_openowner_t, ro_node));
1797 1798
1798 1799 /* set up the callback control structure */
1799 1800 cp->rc_cbinfo.cb_state = CB_UNINIT;
1800 1801 mutex_init(cp->rc_cbinfo.cb_lock, NULL, MUTEX_DEFAULT, NULL);
1801 1802 cv_init(cp->rc_cbinfo.cb_cv, NULL, CV_DEFAULT, NULL);
1802 1803 cv_init(cp->rc_cbinfo.cb_cv_nullcaller, NULL, CV_DEFAULT, NULL);
1803 1804
1804 1805 /*
1805 1806 * Associate the client_t with the current server instance.
1806 1807 * The hold is solely to satisfy the calling requirement of
1807 1808 * rfs4_servinst_assign(). In this case it's not strictly necessary.
1808 1809 */
1809 1810 rfs4_dbe_hold(cp->rc_dbe);
1810 1811 rfs4_servinst_assign(nsrv4, cp, nsrv4->nfs4_cur_servinst);
1811 1812 rfs4_dbe_rele(cp->rc_dbe);
1812 1813
1813 1814 return (TRUE);
1814 1815 }
1815 1816
1816 1817 /*
1817 1818 * Caller wants to generate/update the setclientid_confirm verifier
1818 1819 * associated with a client. This is done during the SETCLIENTID
1819 1820 * processing.
1820 1821 */
1821 1822 void
1822 1823 rfs4_client_scv_next(rfs4_client_t *cp)
1823 1824 {
1824 1825 scid_confirm_verf *scvp;
1825 1826
1826 1827 /* Init the value for the SETCLIENTID_CONFIRM verifier */
1827 1828 scvp = (scid_confirm_verf *)&cp->rc_confirm_verf;
1828 1829 scvp->cv_impl.gen_num++;
1829 1830 }
1830 1831
1831 1832 void
1832 1833 rfs4_client_rele(rfs4_client_t *cp)
1833 1834 {
1834 1835 rfs4_dbe_rele(cp->rc_dbe);
1835 1836 }
1836 1837
1837 1838 rfs4_client_t *
1838 1839 rfs4_findclient(nfs_client_id4 *client, bool_t *create, rfs4_client_t *oldcp)
1839 1840 {
1840 1841 rfs4_client_t *cp;
1841 1842 nfs4_srv_t *nsrv4;
1842 1843 nsrv4 = nfs4_get_srv();
1843 1844
1844 1845
1845 1846 if (oldcp) {
1846 1847 rw_enter(&nsrv4->rfs4_findclient_lock, RW_WRITER);
1847 1848 rfs4_dbe_hide(oldcp->rc_dbe);
1848 1849 } else {
1849 1850 rw_enter(&nsrv4->rfs4_findclient_lock, RW_READER);
1850 1851 }
1851 1852
1852 1853 cp = (rfs4_client_t *)rfs4_dbsearch(nsrv4->rfs4_nfsclnt_idx, client,
1853 1854 create, (void *)client, RFS4_DBS_VALID);
1854 1855
1855 1856 if (oldcp)
1856 1857 rfs4_dbe_unhide(oldcp->rc_dbe);
1857 1858
1858 1859 rw_exit(&nsrv4->rfs4_findclient_lock);
1859 1860
1860 1861 return (cp);
1861 1862 }
1862 1863
1863 1864 rfs4_client_t *
1864 1865 rfs4_findclient_by_id(clientid4 clientid, bool_t find_unconfirmed)
1865 1866 {
1866 1867 rfs4_client_t *cp;
1867 1868 bool_t create = FALSE;
1868 1869 cid *cidp = (cid *)&clientid;
1869 1870 nfs4_srv_t *nsrv4 = nfs4_get_srv();
1870 1871
1871 1872 /* If we're a cluster and the nodeid isn't right, short-circuit */
1872 1873 if (cluster_bootflags & CLUSTER_BOOTED && foreign_clientid(cidp))
1873 1874 return (NULL);
1874 1875
1875 1876 rw_enter(&nsrv4->rfs4_findclient_lock, RW_READER);
1876 1877
1877 1878 cp = (rfs4_client_t *)rfs4_dbsearch(nsrv4->rfs4_clientid_idx, &clientid,
1878 1879 &create, NULL, RFS4_DBS_VALID);
1879 1880
1880 1881 rw_exit(&nsrv4->rfs4_findclient_lock);
1881 1882
1882 1883 if (cp && cp->rc_need_confirm && find_unconfirmed == FALSE) {
1883 1884 rfs4_client_rele(cp);
1884 1885 return (NULL);
1885 1886 } else {
1886 1887 return (cp);
1887 1888 }
1888 1889 }
1889 1890
1890 1891 static uint32_t
1891 1892 clntip_hash(void *key)
1892 1893 {
1893 1894 struct sockaddr *addr = key;
1894 1895 int i, len = 0;
1895 1896 uint32_t hash = 0;
1896 1897 char *ptr;
1897 1898
1898 1899 if (addr->sa_family == AF_INET) {
1899 1900 struct sockaddr_in *a = (struct sockaddr_in *)addr;
1900 1901 len = sizeof (struct in_addr);
1901 1902 ptr = (char *)&a->sin_addr;
1902 1903 } else if (addr->sa_family == AF_INET6) {
1903 1904 struct sockaddr_in6 *a = (struct sockaddr_in6 *)addr;
1904 1905 len = sizeof (struct in6_addr);
1905 1906 ptr = (char *)&a->sin6_addr;
1906 1907 } else
1907 1908 return (0);
1908 1909
1909 1910 for (i = 0; i < len; i++) {
1910 1911 hash <<= 1;
1911 1912 hash += (uint_t)ptr[i];
1912 1913 }
1913 1914 return (hash);
1914 1915 }
1915 1916
1916 1917 static bool_t
1917 1918 clntip_compare(rfs4_entry_t entry, void *key)
1918 1919 {
1919 1920 rfs4_clntip_t *cp = (rfs4_clntip_t *)entry;
1920 1921 struct sockaddr *addr = key;
1921 1922 int len = 0;
1922 1923 char *p1, *p2;
1923 1924
1924 1925 if (addr->sa_family == AF_INET) {
1925 1926 struct sockaddr_in *a1 = (struct sockaddr_in *)&cp->ri_addr;
1926 1927 struct sockaddr_in *a2 = (struct sockaddr_in *)addr;
1927 1928 len = sizeof (struct in_addr);
1928 1929 p1 = (char *)&a1->sin_addr;
1929 1930 p2 = (char *)&a2->sin_addr;
1930 1931 } else if (addr->sa_family == AF_INET6) {
1931 1932 struct sockaddr_in6 *a1 = (struct sockaddr_in6 *)&cp->ri_addr;
1932 1933 struct sockaddr_in6 *a2 = (struct sockaddr_in6 *)addr;
1933 1934 len = sizeof (struct in6_addr);
1934 1935 p1 = (char *)&a1->sin6_addr;
1935 1936 p2 = (char *)&a2->sin6_addr;
1936 1937 } else
1937 1938 return (0);
1938 1939
1939 1940 return (bcmp(p1, p2, len) == 0);
1940 1941 }
1941 1942
1942 1943 static void *
1943 1944 clntip_mkkey(rfs4_entry_t entry)
1944 1945 {
1945 1946 rfs4_clntip_t *cp = (rfs4_clntip_t *)entry;
1946 1947
1947 1948 return (&cp->ri_addr);
1948 1949 }
1949 1950
1950 1951 static bool_t
1951 1952 rfs4_clntip_expiry(rfs4_entry_t u_entry)
1952 1953 {
1953 1954 rfs4_clntip_t *cp = (rfs4_clntip_t *)u_entry;
1954 1955
1955 1956 if (rfs4_dbe_is_invalid(cp->ri_dbe))
1956 1957 return (TRUE);
1957 1958 return (FALSE);
1958 1959 }
1959 1960
1960 1961 /* ARGSUSED */
1961 1962 static void
1962 1963 rfs4_clntip_destroy(rfs4_entry_t u_entry)
1963 1964 {
1964 1965 }
1965 1966
1966 1967 static bool_t
1967 1968 rfs4_clntip_create(rfs4_entry_t u_entry, void *arg)
1968 1969 {
1969 1970 rfs4_clntip_t *cp = (rfs4_clntip_t *)u_entry;
1970 1971 struct sockaddr *ca = (struct sockaddr *)arg;
1971 1972
1972 1973 /* Copy client's IP address */
1973 1974 if (ca->sa_family == AF_INET)
1974 1975 bcopy(ca, &cp->ri_addr, sizeof (struct sockaddr_in));
1975 1976 else if (ca->sa_family == AF_INET6)
1976 1977 bcopy(ca, &cp->ri_addr, sizeof (struct sockaddr_in6));
1977 1978 else
1978 1979 return (FALSE);
1979 1980 cp->ri_no_referrals = 1;
1980 1981
1981 1982 return (TRUE);
1982 1983 }
1983 1984
1984 1985 rfs4_clntip_t *
1985 1986 rfs4_find_clntip(struct sockaddr *addr, bool_t *create)
1986 1987 {
1987 1988 rfs4_clntip_t *cp;
1988 1989 nfs4_srv_t *nsrv4;
1989 1990
1990 1991 nsrv4 = nfs4_get_srv();
1991 1992
1992 1993 rw_enter(&nsrv4->rfs4_findclient_lock, RW_READER);
1993 1994
1994 1995 cp = (rfs4_clntip_t *)rfs4_dbsearch(nsrv4->rfs4_clntip_idx, addr,
1995 1996 create, addr, RFS4_DBS_VALID);
1996 1997
1997 1998 rw_exit(&nsrv4->rfs4_findclient_lock);
1998 1999
1999 2000 return (cp);
2000 2001 }
2001 2002
2002 2003 void
2003 2004 rfs4_invalidate_clntip(struct sockaddr *addr)
2004 2005 {
2005 2006 rfs4_clntip_t *cp;
2006 2007 bool_t create = FALSE;
2007 2008 nfs4_srv_t *nsrv4 = nfs4_get_srv();
2008 2009
2009 2010 rw_enter(&nsrv4->rfs4_findclient_lock, RW_READER);
2010 2011
2011 2012 cp = (rfs4_clntip_t *)rfs4_dbsearch(nsrv4->rfs4_clntip_idx, addr,
2012 2013 &create, NULL, RFS4_DBS_VALID);
2013 2014 if (cp == NULL) {
2014 2015 rw_exit(&nsrv4->rfs4_findclient_lock);
2015 2016 return;
2016 2017 }
2017 2018 rfs4_dbe_invalidate(cp->ri_dbe);
2018 2019 rfs4_dbe_rele(cp->ri_dbe);
2019 2020
2020 2021 rw_exit(&nsrv4->rfs4_findclient_lock);
2021 2022 }
2022 2023
2023 2024 bool_t
2024 2025 rfs4_lease_expired(rfs4_client_t *cp)
2025 2026 {
2026 2027 bool_t rc;
2027 2028
2028 2029 rfs4_dbe_lock(cp->rc_dbe);
2029 2030
2030 2031 /*
2031 2032 * If the admin has executed clear_locks for this
2032 2033 * client id, force expire will be set, so no need
2033 2034 * to calculate anything because it's "outa here".
2034 2035 */
2035 2036 if (cp->rc_forced_expire) {
2036 2037 rc = TRUE;
2037 2038 } else {
2038 2039 rc = (gethrestime_sec() - cp->rc_last_access > rfs4_lease_time);
2039 2040 }
2040 2041
2041 2042 /*
2042 2043 * If the lease has expired we will also want
2043 2044 * to remove any stable storage state data. So
2044 2045 * mark the client id accordingly.
2045 2046 */
2046 2047 if (!cp->rc_ss_remove)
2047 2048 cp->rc_ss_remove = (rc == TRUE);
2048 2049
2049 2050 rfs4_dbe_unlock(cp->rc_dbe);
2050 2051
2051 2052 return (rc);
2052 2053 }
2053 2054
2054 2055 void
2055 2056 rfs4_update_lease(rfs4_client_t *cp)
2056 2057 {
2057 2058 rfs4_dbe_lock(cp->rc_dbe);
2058 2059 if (!cp->rc_forced_expire)
2059 2060 cp->rc_last_access = gethrestime_sec();
2060 2061 rfs4_dbe_unlock(cp->rc_dbe);
2061 2062 }
2062 2063
2063 2064
2064 2065 static bool_t
2065 2066 EQOPENOWNER(open_owner4 *a, open_owner4 *b)
2066 2067 {
2067 2068 bool_t rc;
2068 2069
2069 2070 if (a->clientid != b->clientid)
2070 2071 return (FALSE);
2071 2072
2072 2073 if (a->owner_len != b->owner_len)
2073 2074 return (FALSE);
2074 2075
2075 2076 rc = (bcmp(a->owner_val, b->owner_val, a->owner_len) == 0);
2076 2077
2077 2078 return (rc);
2078 2079 }
2079 2080
2080 2081 static uint_t
2081 2082 openowner_hash(void *key)
2082 2083 {
2083 2084 int i;
2084 2085 open_owner4 *openowner = key;
2085 2086 uint_t hash = 0;
2086 2087
2087 2088 for (i = 0; i < openowner->owner_len; i++) {
2088 2089 hash <<= 4;
2089 2090 hash += (uint_t)openowner->owner_val[i];
2090 2091 }
2091 2092 hash += (uint_t)openowner->clientid;
2092 2093 hash |= (openowner->clientid >> 32);
2093 2094
2094 2095 return (hash);
2095 2096 }
2096 2097
2097 2098 static bool_t
2098 2099 openowner_compare(rfs4_entry_t u_entry, void *key)
2099 2100 {
2100 2101 rfs4_openowner_t *oo = (rfs4_openowner_t *)u_entry;
2101 2102 open_owner4 *arg = key;
2102 2103
2103 2104 return (EQOPENOWNER(&oo->ro_owner, arg));
2104 2105 }
2105 2106
2106 2107 void *
2107 2108 openowner_mkkey(rfs4_entry_t u_entry)
2108 2109 {
2109 2110 rfs4_openowner_t *oo = (rfs4_openowner_t *)u_entry;
2110 2111
2111 2112 return (&oo->ro_owner);
2112 2113 }
2113 2114
2114 2115 /* ARGSUSED */
2115 2116 static bool_t
2116 2117 rfs4_openowner_expiry(rfs4_entry_t u_entry)
2117 2118 {
2118 2119 /* openstateid held us and did all needed delay */
2119 2120 return (TRUE);
2120 2121 }
2121 2122
2122 2123 static void
2123 2124 rfs4_openowner_destroy(rfs4_entry_t u_entry)
2124 2125 {
2125 2126 rfs4_openowner_t *oo = (rfs4_openowner_t *)u_entry;
2126 2127
2127 2128 /* Remove open owner from client's lists of open owners */
2128 2129 rfs4_dbe_lock(oo->ro_client->rc_dbe);
2129 2130 list_remove(&oo->ro_client->rc_openownerlist, oo);
2130 2131 rfs4_dbe_unlock(oo->ro_client->rc_dbe);
2131 2132
2132 2133 /* One less reference to the client */
2133 2134 rfs4_client_rele(oo->ro_client);
2134 2135 oo->ro_client = NULL;
2135 2136
2136 2137 /* Free the last reply for this lock owner */
2137 2138 rfs4_free_reply(&oo->ro_reply);
2138 2139
2139 2140 if (oo->ro_reply_fh.nfs_fh4_val) {
2140 2141 kmem_free(oo->ro_reply_fh.nfs_fh4_val,
2141 2142 oo->ro_reply_fh.nfs_fh4_len);
2142 2143 oo->ro_reply_fh.nfs_fh4_val = NULL;
2143 2144 oo->ro_reply_fh.nfs_fh4_len = 0;
2144 2145 }
2145 2146
2146 2147 rfs4_sw_destroy(&oo->ro_sw);
2147 2148 list_destroy(&oo->ro_statelist);
2148 2149
2149 2150 /* Free the lock owner id */
2150 2151 kmem_free(oo->ro_owner.owner_val, oo->ro_owner.owner_len);
2151 2152 }
2152 2153
2153 2154 void
2154 2155 rfs4_openowner_rele(rfs4_openowner_t *oo)
2155 2156 {
2156 2157 rfs4_dbe_rele(oo->ro_dbe);
2157 2158 }
2158 2159
2159 2160 static bool_t
2160 2161 rfs4_openowner_create(rfs4_entry_t u_entry, void *arg)
2161 2162 {
2162 2163 rfs4_openowner_t *oo = (rfs4_openowner_t *)u_entry;
2163 2164 rfs4_openowner_t *argp = (rfs4_openowner_t *)arg;
2164 2165 open_owner4 *openowner = &argp->ro_owner;
2165 2166 seqid4 seqid = argp->ro_open_seqid;
2166 2167 rfs4_client_t *cp;
2167 2168 bool_t create = FALSE;
2168 2169 nfs4_srv_t *nsrv4 = nfs4_get_srv();
2169 2170
2170 2171 rw_enter(&nsrv4->rfs4_findclient_lock, RW_READER);
2171 2172
2172 2173 cp = (rfs4_client_t *)rfs4_dbsearch(nsrv4->rfs4_clientid_idx,
2173 2174 &openowner->clientid,
2174 2175 &create, NULL, RFS4_DBS_VALID);
2175 2176
2176 2177 rw_exit(&nsrv4->rfs4_findclient_lock);
2177 2178
2178 2179 if (cp == NULL)
2179 2180 return (FALSE);
2180 2181
2181 2182 oo->ro_reply_fh.nfs_fh4_len = 0;
2182 2183 oo->ro_reply_fh.nfs_fh4_val = NULL;
2183 2184
2184 2185 oo->ro_owner.clientid = openowner->clientid;
2185 2186 oo->ro_owner.owner_val =
2186 2187 kmem_alloc(openowner->owner_len, KM_SLEEP);
2187 2188
2188 2189 bcopy(openowner->owner_val,
2189 2190 oo->ro_owner.owner_val, openowner->owner_len);
2190 2191
2191 2192 oo->ro_owner.owner_len = openowner->owner_len;
2192 2193
2193 2194 oo->ro_need_confirm = TRUE;
2194 2195
2195 2196 rfs4_sw_init(&oo->ro_sw);
2196 2197
2197 2198 oo->ro_open_seqid = seqid;
2198 2199 bzero(&oo->ro_reply, sizeof (nfs_resop4));
2199 2200 oo->ro_client = cp;
2200 2201 oo->ro_cr_set = NULL;
2201 2202
2202 2203 list_create(&oo->ro_statelist, sizeof (rfs4_state_t),
2203 2204 offsetof(rfs4_state_t, rs_node));
2204 2205
2205 2206 /* Insert openowner into client's open owner list */
2206 2207 rfs4_dbe_lock(cp->rc_dbe);
2207 2208 list_insert_tail(&cp->rc_openownerlist, oo);
2208 2209 rfs4_dbe_unlock(cp->rc_dbe);
2209 2210
2210 2211 return (TRUE);
2211 2212 }
2212 2213
2213 2214 rfs4_openowner_t *
2214 2215 rfs4_findopenowner(open_owner4 *openowner, bool_t *create, seqid4 seqid)
2215 2216 {
2216 2217 rfs4_openowner_t *oo;
2217 2218 rfs4_openowner_t arg;
2218 2219 nfs4_srv_t *nsrv4 = nfs4_get_srv();
2219 2220
2220 2221 arg.ro_owner = *openowner;
2221 2222 arg.ro_open_seqid = seqid;
2222 2223 /* CSTYLED */
2223 2224 oo = (rfs4_openowner_t *)rfs4_dbsearch(nsrv4->rfs4_openowner_idx, openowner,
2224 2225 create, &arg, RFS4_DBS_VALID);
2225 2226
2226 2227 return (oo);
2227 2228 }
2228 2229
2229 2230 void
2230 2231 rfs4_update_open_sequence(rfs4_openowner_t *oo)
2231 2232 {
2232 2233
2233 2234 rfs4_dbe_lock(oo->ro_dbe);
2234 2235
2235 2236 oo->ro_open_seqid++;
2236 2237
2237 2238 rfs4_dbe_unlock(oo->ro_dbe);
2238 2239 }
2239 2240
2240 2241 void
2241 2242 rfs4_update_open_resp(rfs4_openowner_t *oo, nfs_resop4 *resp, nfs_fh4 *fh)
2242 2243 {
2243 2244
2244 2245 rfs4_dbe_lock(oo->ro_dbe);
2245 2246
2246 2247 rfs4_free_reply(&oo->ro_reply);
2247 2248
2248 2249 rfs4_copy_reply(&oo->ro_reply, resp);
2249 2250
2250 2251 /* Save the filehandle if provided and free if not used */
2251 2252 if (resp->nfs_resop4_u.opopen.status == NFS4_OK &&
2252 2253 fh && fh->nfs_fh4_len) {
2253 2254 if (oo->ro_reply_fh.nfs_fh4_val == NULL)
2254 2255 oo->ro_reply_fh.nfs_fh4_val =
2255 2256 kmem_alloc(fh->nfs_fh4_len, KM_SLEEP);
2256 2257 nfs_fh4_copy(fh, &oo->ro_reply_fh);
2257 2258 } else {
2258 2259 if (oo->ro_reply_fh.nfs_fh4_val) {
2259 2260 kmem_free(oo->ro_reply_fh.nfs_fh4_val,
2260 2261 oo->ro_reply_fh.nfs_fh4_len);
2261 2262 oo->ro_reply_fh.nfs_fh4_val = NULL;
2262 2263 oo->ro_reply_fh.nfs_fh4_len = 0;
2263 2264 }
2264 2265 }
2265 2266
2266 2267 rfs4_dbe_unlock(oo->ro_dbe);
2267 2268 }
2268 2269
2269 2270 static bool_t
2270 2271 lockowner_compare(rfs4_entry_t u_entry, void *key)
2271 2272 {
2272 2273 rfs4_lockowner_t *lo = (rfs4_lockowner_t *)u_entry;
2273 2274 lock_owner4 *b = (lock_owner4 *)key;
2274 2275
2275 2276 if (lo->rl_owner.clientid != b->clientid)
2276 2277 return (FALSE);
2277 2278
2278 2279 if (lo->rl_owner.owner_len != b->owner_len)
2279 2280 return (FALSE);
2280 2281
2281 2282 return (bcmp(lo->rl_owner.owner_val, b->owner_val,
2282 2283 lo->rl_owner.owner_len) == 0);
2283 2284 }
2284 2285
2285 2286 void *
2286 2287 lockowner_mkkey(rfs4_entry_t u_entry)
2287 2288 {
2288 2289 rfs4_lockowner_t *lo = (rfs4_lockowner_t *)u_entry;
2289 2290
2290 2291 return (&lo->rl_owner);
2291 2292 }
2292 2293
2293 2294 static uint32_t
2294 2295 lockowner_hash(void *key)
2295 2296 {
2296 2297 int i;
2297 2298 lock_owner4 *lockowner = key;
2298 2299 uint_t hash = 0;
2299 2300
2300 2301 for (i = 0; i < lockowner->owner_len; i++) {
2301 2302 hash <<= 4;
2302 2303 hash += (uint_t)lockowner->owner_val[i];
2303 2304 }
2304 2305 hash += (uint_t)lockowner->clientid;
2305 2306 hash |= (lockowner->clientid >> 32);
2306 2307
2307 2308 return (hash);
2308 2309 }
2309 2310
2310 2311 static uint32_t
2311 2312 pid_hash(void *key)
2312 2313 {
2313 2314 return ((uint32_t)(uintptr_t)key);
2314 2315 }
2315 2316
2316 2317 static void *
2317 2318 pid_mkkey(rfs4_entry_t u_entry)
2318 2319 {
2319 2320 rfs4_lockowner_t *lo = (rfs4_lockowner_t *)u_entry;
2320 2321
2321 2322 return ((void *)(uintptr_t)lo->rl_pid);
2322 2323 }
2323 2324
2324 2325 static bool_t
2325 2326 pid_compare(rfs4_entry_t u_entry, void *key)
2326 2327 {
2327 2328 rfs4_lockowner_t *lo = (rfs4_lockowner_t *)u_entry;
2328 2329
2329 2330 return (lo->rl_pid == (pid_t)(uintptr_t)key);
2330 2331 }
2331 2332
2332 2333 static void
2333 2334 rfs4_lockowner_destroy(rfs4_entry_t u_entry)
2334 2335 {
2335 2336 rfs4_lockowner_t *lo = (rfs4_lockowner_t *)u_entry;
2336 2337
2337 2338 /* Free the lock owner id */
2338 2339 kmem_free(lo->rl_owner.owner_val, lo->rl_owner.owner_len);
2339 2340 rfs4_client_rele(lo->rl_client);
2340 2341 }
2341 2342
2342 2343 void
2343 2344 rfs4_lockowner_rele(rfs4_lockowner_t *lo)
2344 2345 {
2345 2346 rfs4_dbe_rele(lo->rl_dbe);
2346 2347 }
2347 2348
2348 2349 /* ARGSUSED */
2349 2350 static bool_t
2350 2351 rfs4_lockowner_expiry(rfs4_entry_t u_entry)
2351 2352 {
2352 2353 /*
2353 2354 * Since expiry is called with no other references on
2354 2355 * this struct, go ahead and have it removed.
2355 2356 */
2356 2357 return (TRUE);
2357 2358 }
2358 2359
2359 2360 static bool_t
2360 2361 rfs4_lockowner_create(rfs4_entry_t u_entry, void *arg)
2361 2362 {
2362 2363 rfs4_lockowner_t *lo = (rfs4_lockowner_t *)u_entry;
2363 2364 lock_owner4 *lockowner = (lock_owner4 *)arg;
2364 2365 rfs4_client_t *cp;
2365 2366 bool_t create = FALSE;
2366 2367 nfs4_srv_t *nsrv4 = nfs4_get_srv();
2367 2368
2368 2369 rw_enter(&nsrv4->rfs4_findclient_lock, RW_READER);
2369 2370
2370 2371 cp = (rfs4_client_t *)rfs4_dbsearch(nsrv4->rfs4_clientid_idx,
2371 2372 &lockowner->clientid,
2372 2373 &create, NULL, RFS4_DBS_VALID);
2373 2374
2374 2375 rw_exit(&nsrv4->rfs4_findclient_lock);
2375 2376
2376 2377 if (cp == NULL)
2377 2378 return (FALSE);
2378 2379
2379 2380 /* Reference client */
2380 2381 lo->rl_client = cp;
2381 2382 lo->rl_owner.clientid = lockowner->clientid;
2382 2383 lo->rl_owner.owner_val = kmem_alloc(lockowner->owner_len, KM_SLEEP);
2383 2384 bcopy(lockowner->owner_val, lo->rl_owner.owner_val,
2384 2385 lockowner->owner_len);
2385 2386 lo->rl_owner.owner_len = lockowner->owner_len;
2386 2387 lo->rl_pid = rfs4_dbe_getid(lo->rl_dbe);
2387 2388
2388 2389 return (TRUE);
2389 2390 }
2390 2391
2391 2392 rfs4_lockowner_t *
2392 2393 rfs4_findlockowner(lock_owner4 *lockowner, bool_t *create)
2393 2394 {
2394 2395 rfs4_lockowner_t *lo;
2395 2396 nfs4_srv_t *nsrv4 = nfs4_get_srv();
2396 2397
2397 2398 /* CSTYLED */
2398 2399 lo = (rfs4_lockowner_t *)rfs4_dbsearch(nsrv4->rfs4_lockowner_idx, lockowner,
2399 2400 create, lockowner, RFS4_DBS_VALID);
2400 2401
2401 2402 return (lo);
2402 2403 }
2403 2404
2404 2405 rfs4_lockowner_t *
2405 2406 rfs4_findlockowner_by_pid(pid_t pid)
2406 2407 {
2407 2408 rfs4_lockowner_t *lo;
2408 2409 bool_t create = FALSE;
2409 2410 nfs4_srv_t *nsrv4 = nfs4_get_srv();
2410 2411
2411 2412 lo = (rfs4_lockowner_t *)rfs4_dbsearch(nsrv4->rfs4_lockowner_pid_idx,
2412 2413 (void *)(uintptr_t)pid, &create, NULL, RFS4_DBS_VALID);
2413 2414
2414 2415 return (lo);
2415 2416 }
2416 2417
2417 2418
2418 2419 static uint32_t
2419 2420 file_hash(void *key)
2420 2421 {
2421 2422 return (ADDRHASH(key));
2422 2423 }
2423 2424
2424 2425 static void *
2425 2426 file_mkkey(rfs4_entry_t u_entry)
2426 2427 {
2427 2428 rfs4_file_t *fp = (rfs4_file_t *)u_entry;
2428 2429
2429 2430 return (fp->rf_vp);
2430 2431 }
2431 2432
2432 2433 static bool_t
2433 2434 file_compare(rfs4_entry_t u_entry, void *key)
2434 2435 {
2435 2436 rfs4_file_t *fp = (rfs4_file_t *)u_entry;
2436 2437
2437 2438 return (fp->rf_vp == (vnode_t *)key);
2438 2439 }
2439 2440
2440 2441 static void
2441 2442 rfs4_file_destroy(rfs4_entry_t u_entry)
2442 2443 {
2443 2444 rfs4_file_t *fp = (rfs4_file_t *)u_entry;
2444 2445
2445 2446 list_destroy(&fp->rf_delegstatelist);
2446 2447
2447 2448 if (fp->rf_filehandle.nfs_fh4_val)
2448 2449 kmem_free(fp->rf_filehandle.nfs_fh4_val,
2449 2450 fp->rf_filehandle.nfs_fh4_len);
2450 2451 cv_destroy(fp->rf_dinfo.rd_recall_cv);
2451 2452 if (fp->rf_vp) {
2452 2453 vnode_t *vp = fp->rf_vp;
2453 2454
2454 2455 mutex_enter(&vp->v_vsd_lock);
2455 2456 (void) vsd_set(vp, nfs4_srv_vkey, NULL);
2456 2457 mutex_exit(&vp->v_vsd_lock);
2457 2458 VN_RELE(vp);
2458 2459 fp->rf_vp = NULL;
2459 2460 }
2460 2461 rw_destroy(&fp->rf_file_rwlock);
2461 2462 }
2462 2463
2463 2464 /*
2464 2465 * Used to unlock the underlying dbe struct only
2465 2466 */
2466 2467 void
2467 2468 rfs4_file_rele(rfs4_file_t *fp)
2468 2469 {
2469 2470 rfs4_dbe_rele(fp->rf_dbe);
2470 2471 }
2471 2472
2472 2473 typedef struct {
2473 2474 vnode_t *vp;
2474 2475 nfs_fh4 *fh;
2475 2476 } rfs4_fcreate_arg;
2476 2477
2477 2478 static bool_t
2478 2479 rfs4_file_create(rfs4_entry_t u_entry, void *arg)
2479 2480 {
2480 2481 rfs4_file_t *fp = (rfs4_file_t *)u_entry;
2481 2482 rfs4_fcreate_arg *ap = (rfs4_fcreate_arg *)arg;
2482 2483 vnode_t *vp = ap->vp;
2483 2484 nfs_fh4 *fh = ap->fh;
2484 2485
2485 2486 VN_HOLD(vp);
2486 2487
2487 2488 fp->rf_filehandle.nfs_fh4_len = 0;
2488 2489 fp->rf_filehandle.nfs_fh4_val = NULL;
2489 2490 ASSERT(fh && fh->nfs_fh4_len);
2490 2491 if (fh && fh->nfs_fh4_len) {
2491 2492 fp->rf_filehandle.nfs_fh4_val =
2492 2493 kmem_alloc(fh->nfs_fh4_len, KM_SLEEP);
2493 2494 nfs_fh4_copy(fh, &fp->rf_filehandle);
2494 2495 }
2495 2496 fp->rf_vp = vp;
2496 2497
2497 2498 list_create(&fp->rf_delegstatelist, sizeof (rfs4_deleg_state_t),
2498 2499 offsetof(rfs4_deleg_state_t, rds_node));
2499 2500
2500 2501 fp->rf_share_deny = fp->rf_share_access = fp->rf_access_read = 0;
2501 2502 fp->rf_access_write = fp->rf_deny_read = fp->rf_deny_write = 0;
2502 2503
2503 2504 mutex_init(fp->rf_dinfo.rd_recall_lock, NULL, MUTEX_DEFAULT, NULL);
2504 2505 cv_init(fp->rf_dinfo.rd_recall_cv, NULL, CV_DEFAULT, NULL);
2505 2506
2506 2507 fp->rf_dinfo.rd_dtype = OPEN_DELEGATE_NONE;
2507 2508
2508 2509 rw_init(&fp->rf_file_rwlock, NULL, RW_DEFAULT, NULL);
2509 2510
2510 2511 mutex_enter(&vp->v_vsd_lock);
2511 2512 VERIFY(vsd_set(vp, nfs4_srv_vkey, (void *)fp) == 0);
2512 2513 mutex_exit(&vp->v_vsd_lock);
2513 2514
2514 2515 return (TRUE);
2515 2516 }
2516 2517
2517 2518 rfs4_file_t *
2518 2519 rfs4_findfile(vnode_t *vp, nfs_fh4 *fh, bool_t *create)
2519 2520 {
2520 2521 rfs4_file_t *fp;
2521 2522 rfs4_fcreate_arg arg;
2522 2523 nfs4_srv_t *nsrv4 = nfs4_get_srv();
2523 2524
2524 2525 arg.vp = vp;
2525 2526 arg.fh = fh;
2526 2527
2527 2528 if (*create == TRUE)
2528 2529 /* CSTYLED */
2529 2530 fp = (rfs4_file_t *)rfs4_dbsearch(nsrv4->rfs4_file_idx, vp, create,
2530 2531 &arg, RFS4_DBS_VALID);
2531 2532 else {
2532 2533 mutex_enter(&vp->v_vsd_lock);
2533 2534 fp = (rfs4_file_t *)vsd_get(vp, nfs4_srv_vkey);
2534 2535 if (fp) {
2535 2536 rfs4_dbe_lock(fp->rf_dbe);
2536 2537 if (rfs4_dbe_is_invalid(fp->rf_dbe) ||
2537 2538 (rfs4_dbe_refcnt(fp->rf_dbe) == 0)) {
2538 2539 rfs4_dbe_unlock(fp->rf_dbe);
2539 2540 fp = NULL;
2540 2541 } else {
2541 2542 rfs4_dbe_hold(fp->rf_dbe);
2542 2543 rfs4_dbe_unlock(fp->rf_dbe);
2543 2544 }
2544 2545 }
2545 2546 mutex_exit(&vp->v_vsd_lock);
2546 2547 }
2547 2548 return (fp);
2548 2549 }
2549 2550
2550 2551 /*
2551 2552 * Find a file in the db and once it is located, take the rw lock.
2552 2553 * Need to check the vnode pointer and if it does not exist (it was
2553 2554 * removed between the db location and check) redo the find. This
2554 2555 * assumes that a file struct that has a NULL vnode pointer is marked
2555 2556 * at 'invalid' and will not be found in the db the second time
2556 2557 * around.
2557 2558 */
2558 2559 rfs4_file_t *
2559 2560 rfs4_findfile_withlock(vnode_t *vp, nfs_fh4 *fh, bool_t *create)
2560 2561 {
2561 2562 rfs4_file_t *fp;
2562 2563 rfs4_fcreate_arg arg;
2563 2564 bool_t screate = *create;
2564 2565 nfs4_srv_t *nsrv4 = nfs4_get_srv();
2565 2566
2566 2567 if (screate == FALSE) {
2567 2568 mutex_enter(&vp->v_vsd_lock);
2568 2569 fp = (rfs4_file_t *)vsd_get(vp, nfs4_srv_vkey);
2569 2570 if (fp) {
2570 2571 rfs4_dbe_lock(fp->rf_dbe);
2571 2572 if (rfs4_dbe_is_invalid(fp->rf_dbe) ||
2572 2573 (rfs4_dbe_refcnt(fp->rf_dbe) == 0)) {
2573 2574 rfs4_dbe_unlock(fp->rf_dbe);
2574 2575 mutex_exit(&vp->v_vsd_lock);
2575 2576 fp = NULL;
2576 2577 } else {
2577 2578 rfs4_dbe_hold(fp->rf_dbe);
2578 2579 rfs4_dbe_unlock(fp->rf_dbe);
2579 2580 mutex_exit(&vp->v_vsd_lock);
2580 2581 rw_enter(&fp->rf_file_rwlock, RW_WRITER);
2581 2582 if (fp->rf_vp == NULL) {
2582 2583 rw_exit(&fp->rf_file_rwlock);
2583 2584 rfs4_file_rele(fp);
2584 2585 fp = NULL;
2585 2586 }
2586 2587 }
2587 2588 } else {
2588 2589 mutex_exit(&vp->v_vsd_lock);
2589 2590 }
2590 2591 } else {
2591 2592 retry:
2592 2593 arg.vp = vp;
2593 2594 arg.fh = fh;
2594 2595
2595 2596 fp = (rfs4_file_t *)rfs4_dbsearch(nsrv4->rfs4_file_idx, vp,
2596 2597 create, &arg, RFS4_DBS_VALID);
2597 2598 if (fp != NULL) {
2598 2599 rw_enter(&fp->rf_file_rwlock, RW_WRITER);
2599 2600 if (fp->rf_vp == NULL) {
2600 2601 rw_exit(&fp->rf_file_rwlock);
2601 2602 rfs4_file_rele(fp);
2602 2603 *create = screate;
2603 2604 goto retry;
2604 2605 }
2605 2606 }
2606 2607 }
2607 2608
2608 2609 return (fp);
2609 2610 }
2610 2611
2611 2612 static uint32_t
2612 2613 lo_state_hash(void *key)
2613 2614 {
2614 2615 stateid_t *id = key;
2615 2616
2616 2617 return (id->bits.ident+id->bits.pid);
2617 2618 }
2618 2619
2619 2620 static bool_t
2620 2621 lo_state_compare(rfs4_entry_t u_entry, void *key)
2621 2622 {
2622 2623 rfs4_lo_state_t *lsp = (rfs4_lo_state_t *)u_entry;
2623 2624 stateid_t *id = key;
2624 2625 bool_t rc;
2625 2626
2626 2627 rc = (lsp->rls_lockid.bits.boottime == id->bits.boottime &&
2627 2628 lsp->rls_lockid.bits.type == id->bits.type &&
2628 2629 lsp->rls_lockid.bits.ident == id->bits.ident &&
2629 2630 lsp->rls_lockid.bits.pid == id->bits.pid);
2630 2631
2631 2632 return (rc);
2632 2633 }
2633 2634
2634 2635 static void *
2635 2636 lo_state_mkkey(rfs4_entry_t u_entry)
2636 2637 {
2637 2638 rfs4_lo_state_t *lsp = (rfs4_lo_state_t *)u_entry;
2638 2639
2639 2640 return (&lsp->rls_lockid);
2640 2641 }
2641 2642
2642 2643 static bool_t
2643 2644 rfs4_lo_state_expiry(rfs4_entry_t u_entry)
2644 2645 {
2645 2646 rfs4_lo_state_t *lsp = (rfs4_lo_state_t *)u_entry;
2646 2647
2647 2648 if (rfs4_dbe_is_invalid(lsp->rls_dbe))
2648 2649 return (TRUE);
2649 2650 if (lsp->rls_state->rs_closed)
2650 2651 return (TRUE);
2651 2652 return ((gethrestime_sec() -
2652 2653 lsp->rls_state->rs_owner->ro_client->rc_last_access
2653 2654 > rfs4_lease_time));
2654 2655 }
2655 2656
2656 2657 static void
2657 2658 rfs4_lo_state_destroy(rfs4_entry_t u_entry)
2658 2659 {
2659 2660 rfs4_lo_state_t *lsp = (rfs4_lo_state_t *)u_entry;
2660 2661
2661 2662 rfs4_dbe_lock(lsp->rls_state->rs_dbe);
2662 2663 list_remove(&lsp->rls_state->rs_lostatelist, lsp);
2663 2664 rfs4_dbe_unlock(lsp->rls_state->rs_dbe);
2664 2665
2665 2666 rfs4_sw_destroy(&lsp->rls_sw);
2666 2667
2667 2668 /* Make sure to release the file locks */
2668 2669 if (lsp->rls_locks_cleaned == FALSE) {
2669 2670 lsp->rls_locks_cleaned = TRUE;
2670 2671 if (lsp->rls_locker->rl_client->rc_sysidt != LM_NOSYSID) {
2671 2672 /* Is the PxFS kernel module loaded? */
2672 2673 if (lm_remove_file_locks != NULL) {
2673 2674 int new_sysid;
2674 2675
2675 2676 /* Encode the cluster nodeid in new sysid */
2676 2677 new_sysid =
2677 2678 lsp->rls_locker->rl_client->rc_sysidt;
2678 2679 lm_set_nlmid_flk(&new_sysid);
2679 2680
2680 2681 /*
2681 2682 * This PxFS routine removes file locks for a
2682 2683 * client over all nodes of a cluster.
2683 2684 */
2684 2685 DTRACE_PROBE1(nfss_i_clust_rm_lck,
2685 2686 int, new_sysid);
2686 2687 (*lm_remove_file_locks)(new_sysid);
2687 2688 } else {
2688 2689 (void) cleanlocks(
2689 2690 lsp->rls_state->rs_finfo->rf_vp,
2690 2691 lsp->rls_locker->rl_pid,
2691 2692 lsp->rls_locker->rl_client->rc_sysidt);
2692 2693 }
2693 2694 }
2694 2695 }
2695 2696
2696 2697 /* Free the last reply for this state */
2697 2698 rfs4_free_reply(&lsp->rls_reply);
2698 2699
2699 2700 rfs4_lockowner_rele(lsp->rls_locker);
2700 2701 lsp->rls_locker = NULL;
2701 2702
2702 2703 rfs4_state_rele_nounlock(lsp->rls_state);
2703 2704 lsp->rls_state = NULL;
2704 2705 }
2705 2706
2706 2707 static bool_t
2707 2708 rfs4_lo_state_create(rfs4_entry_t u_entry, void *arg)
2708 2709 {
2709 2710 rfs4_lo_state_t *lsp = (rfs4_lo_state_t *)u_entry;
2710 2711 rfs4_lo_state_t *argp = (rfs4_lo_state_t *)arg;
2711 2712 rfs4_lockowner_t *lo = argp->rls_locker;
2712 2713 rfs4_state_t *sp = argp->rls_state;
2713 2714
2714 2715 lsp->rls_state = sp;
2715 2716
2716 2717 lsp->rls_lockid = sp->rs_stateid;
2717 2718 lsp->rls_lockid.bits.type = LOCKID;
2718 2719 lsp->rls_lockid.bits.chgseq = 0;
2719 2720 lsp->rls_lockid.bits.pid = lo->rl_pid;
2720 2721
2721 2722 lsp->rls_locks_cleaned = FALSE;
2722 2723 lsp->rls_lock_completed = FALSE;
2723 2724
2724 2725 rfs4_sw_init(&lsp->rls_sw);
2725 2726
2726 2727 /* Attached the supplied lock owner */
2727 2728 rfs4_dbe_hold(lo->rl_dbe);
2728 2729 lsp->rls_locker = lo;
2729 2730
2730 2731 rfs4_dbe_lock(sp->rs_dbe);
2731 2732 list_insert_tail(&sp->rs_lostatelist, lsp);
2732 2733 rfs4_dbe_hold(sp->rs_dbe);
2733 2734 rfs4_dbe_unlock(sp->rs_dbe);
2734 2735
2735 2736 return (TRUE);
2736 2737 }
2737 2738
2738 2739 void
2739 2740 rfs4_lo_state_rele(rfs4_lo_state_t *lsp, bool_t unlock_fp)
2740 2741 {
2741 2742 if (unlock_fp == TRUE)
2742 2743 rw_exit(&lsp->rls_state->rs_finfo->rf_file_rwlock);
2743 2744 rfs4_dbe_rele(lsp->rls_dbe);
2744 2745 }
2745 2746
2746 2747 static rfs4_lo_state_t *
2747 2748 rfs4_findlo_state(stateid_t *id, bool_t lock_fp)
2748 2749 {
2749 2750 rfs4_lo_state_t *lsp;
2750 2751 bool_t create = FALSE;
2751 2752 nfs4_srv_t *nsrv4 = nfs4_get_srv();
2752 2753
2753 2754 lsp = (rfs4_lo_state_t *)rfs4_dbsearch(nsrv4->rfs4_lo_state_idx, id,
2754 2755 &create, NULL, RFS4_DBS_VALID);
2755 2756 if (lock_fp == TRUE && lsp != NULL)
2756 2757 rw_enter(&lsp->rls_state->rs_finfo->rf_file_rwlock, RW_READER);
2757 2758
2758 2759 return (lsp);
2759 2760 }
2760 2761
2761 2762
2762 2763 static uint32_t
2763 2764 lo_state_lo_hash(void *key)
2764 2765 {
2765 2766 rfs4_lo_state_t *lsp = key;
2766 2767
2767 2768 return (ADDRHASH(lsp->rls_locker) ^ ADDRHASH(lsp->rls_state));
2768 2769 }
2769 2770
2770 2771 static bool_t
2771 2772 lo_state_lo_compare(rfs4_entry_t u_entry, void *key)
2772 2773 {
2773 2774 rfs4_lo_state_t *lsp = (rfs4_lo_state_t *)u_entry;
2774 2775 rfs4_lo_state_t *keyp = key;
2775 2776
2776 2777 return (keyp->rls_locker == lsp->rls_locker &&
2777 2778 keyp->rls_state == lsp->rls_state);
2778 2779 }
2779 2780
2780 2781 static void *
2781 2782 lo_state_lo_mkkey(rfs4_entry_t u_entry)
2782 2783 {
2783 2784 return (u_entry);
2784 2785 }
2785 2786
2786 2787 rfs4_lo_state_t *
2787 2788 rfs4_findlo_state_by_owner(rfs4_lockowner_t *lo, rfs4_state_t *sp,
2788 2789 bool_t *create)
2789 2790 {
2790 2791 rfs4_lo_state_t *lsp;
2791 2792 rfs4_lo_state_t arg;
2792 2793 nfs4_srv_t *nsrv4 = nfs4_get_srv();
2793 2794
2794 2795 arg.rls_locker = lo;
2795 2796 arg.rls_state = sp;
2796 2797
2797 2798 lsp = (rfs4_lo_state_t *)rfs4_dbsearch(nsrv4->rfs4_lo_state_owner_idx,
2798 2799 &arg, create, &arg, RFS4_DBS_VALID);
2799 2800
2800 2801 return (lsp);
2801 2802 }
2802 2803
2803 2804 static stateid_t
2804 2805 get_stateid(id_t eid)
2805 2806 {
2806 2807 stateid_t id;
2807 2808 nfs4_srv_t *nsrv4;
2808 2809
2809 2810 nsrv4 = nfs4_get_srv();
2810 2811
2811 2812 id.bits.boottime = nsrv4->rfs4_start_time;
2812 2813 id.bits.ident = eid;
2813 2814 id.bits.chgseq = 0;
2814 2815 id.bits.type = 0;
2815 2816 id.bits.pid = 0;
2816 2817
2817 2818 /*
2818 2819 * If we are booted as a cluster node, embed our nodeid.
2819 2820 * We've already done sanity checks in rfs4_client_create() so no
2820 2821 * need to repeat them here.
2821 2822 */
2822 2823 id.bits.clnodeid = (cluster_bootflags & CLUSTER_BOOTED) ?
2823 2824 clconf_get_nodeid() : 0;
2824 2825
2825 2826 return (id);
2826 2827 }
2827 2828
2828 2829 /*
2829 2830 * For use only when booted as a cluster node.
2830 2831 * Returns TRUE if the embedded nodeid indicates that this stateid was
2831 2832 * generated on another node.
2832 2833 */
2833 2834 static int
2834 2835 foreign_stateid(stateid_t *id)
2835 2836 {
2836 2837 ASSERT(cluster_bootflags & CLUSTER_BOOTED);
2837 2838 return (id->bits.clnodeid != (uint32_t)clconf_get_nodeid());
2838 2839 }
2839 2840
2840 2841 /*
2841 2842 * For use only when booted as a cluster node.
2842 2843 * Returns TRUE if the embedded nodeid indicates that this clientid was
2843 2844 * generated on another node.
2844 2845 */
2845 2846 static int
2846 2847 foreign_clientid(cid *cidp)
2847 2848 {
2848 2849 ASSERT(cluster_bootflags & CLUSTER_BOOTED);
2849 2850 return (cidp->impl_id.c_id >> CLUSTER_NODEID_SHIFT !=
2850 2851 (uint32_t)clconf_get_nodeid());
2851 2852 }
2852 2853
2853 2854 /*
2854 2855 * For use only when booted as a cluster node.
2855 2856 * Embed our cluster nodeid into the clientid.
2856 2857 */
2857 2858 static void
2858 2859 embed_nodeid(cid *cidp)
2859 2860 {
2860 2861 int clnodeid;
2861 2862 /*
2862 2863 * Currently, our state tables are small enough that their
2863 2864 * ids will leave enough bits free for the nodeid. If the
2864 2865 * tables become larger, we mustn't overwrite the id.
2865 2866 * Equally, we only have room for so many bits of nodeid, so
2866 2867 * must check that too.
2867 2868 */
2868 2869 ASSERT(cluster_bootflags & CLUSTER_BOOTED);
2869 2870 ASSERT(cidp->impl_id.c_id >> CLUSTER_NODEID_SHIFT == 0);
2870 2871 clnodeid = clconf_get_nodeid();
2871 2872 ASSERT(clnodeid <= CLUSTER_MAX_NODEID);
2872 2873 ASSERT(clnodeid != NODEID_UNKNOWN);
2873 2874 cidp->impl_id.c_id |= (clnodeid << CLUSTER_NODEID_SHIFT);
2874 2875 }
2875 2876
2876 2877 static uint32_t
2877 2878 state_hash(void *key)
2878 2879 {
2879 2880 stateid_t *ip = (stateid_t *)key;
2880 2881
2881 2882 return (ip->bits.ident);
2882 2883 }
2883 2884
2884 2885 static bool_t
2885 2886 state_compare(rfs4_entry_t u_entry, void *key)
2886 2887 {
2887 2888 rfs4_state_t *sp = (rfs4_state_t *)u_entry;
2888 2889 stateid_t *id = (stateid_t *)key;
2889 2890 bool_t rc;
2890 2891
2891 2892 rc = (sp->rs_stateid.bits.boottime == id->bits.boottime &&
2892 2893 sp->rs_stateid.bits.ident == id->bits.ident);
2893 2894
2894 2895 return (rc);
2895 2896 }
2896 2897
2897 2898 static void *
2898 2899 state_mkkey(rfs4_entry_t u_entry)
2899 2900 {
2900 2901 rfs4_state_t *sp = (rfs4_state_t *)u_entry;
2901 2902
2902 2903 return (&sp->rs_stateid);
2903 2904 }
2904 2905
2905 2906 static void
2906 2907 rfs4_state_destroy(rfs4_entry_t u_entry)
2907 2908 {
2908 2909 rfs4_state_t *sp = (rfs4_state_t *)u_entry;
2909 2910
2910 2911 /* remove from openowner list */
2911 2912 rfs4_dbe_lock(sp->rs_owner->ro_dbe);
2912 2913 list_remove(&sp->rs_owner->ro_statelist, sp);
2913 2914 rfs4_dbe_unlock(sp->rs_owner->ro_dbe);
2914 2915
2915 2916 list_destroy(&sp->rs_lostatelist);
2916 2917
2917 2918 /* release any share locks for this stateid if it's still open */
2918 2919 if (!sp->rs_closed) {
2919 2920 rfs4_dbe_lock(sp->rs_dbe);
2920 2921 (void) rfs4_unshare(sp);
2921 2922 rfs4_dbe_unlock(sp->rs_dbe);
2922 2923 }
2923 2924
2924 2925 /* Were done with the file */
2925 2926 rfs4_file_rele(sp->rs_finfo);
2926 2927 sp->rs_finfo = NULL;
2927 2928
2928 2929 /* And now with the openowner */
2929 2930 rfs4_openowner_rele(sp->rs_owner);
2930 2931 sp->rs_owner = NULL;
2931 2932 }
2932 2933
2933 2934 static void
2934 2935 rfs4_state_rele_nounlock(rfs4_state_t *sp)
2935 2936 {
2936 2937 rfs4_dbe_rele(sp->rs_dbe);
2937 2938 }
2938 2939
2939 2940 void
2940 2941 rfs4_state_rele(rfs4_state_t *sp)
2941 2942 {
2942 2943 rw_exit(&sp->rs_finfo->rf_file_rwlock);
2943 2944 rfs4_dbe_rele(sp->rs_dbe);
2944 2945 }
2945 2946
2946 2947 static uint32_t
2947 2948 deleg_hash(void *key)
2948 2949 {
2949 2950 rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)key;
2950 2951
2951 2952 return (ADDRHASH(dsp->rds_client) ^ ADDRHASH(dsp->rds_finfo));
2952 2953 }
2953 2954
2954 2955 static bool_t
2955 2956 deleg_compare(rfs4_entry_t u_entry, void *key)
2956 2957 {
2957 2958 rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
2958 2959 rfs4_deleg_state_t *kdsp = (rfs4_deleg_state_t *)key;
2959 2960
2960 2961 return (dsp->rds_client == kdsp->rds_client &&
2961 2962 dsp->rds_finfo == kdsp->rds_finfo);
2962 2963 }
2963 2964
2964 2965 static void *
2965 2966 deleg_mkkey(rfs4_entry_t u_entry)
2966 2967 {
2967 2968 return (u_entry);
2968 2969 }
2969 2970
2970 2971 static uint32_t
2971 2972 deleg_state_hash(void *key)
2972 2973 {
2973 2974 stateid_t *ip = (stateid_t *)key;
2974 2975
2975 2976 return (ip->bits.ident);
2976 2977 }
2977 2978
2978 2979 static bool_t
2979 2980 deleg_state_compare(rfs4_entry_t u_entry, void *key)
2980 2981 {
2981 2982 rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
2982 2983 stateid_t *id = (stateid_t *)key;
2983 2984 bool_t rc;
2984 2985
2985 2986 if (id->bits.type != DELEGID)
2986 2987 return (FALSE);
2987 2988
2988 2989 rc = (dsp->rds_delegid.bits.boottime == id->bits.boottime &&
2989 2990 dsp->rds_delegid.bits.ident == id->bits.ident);
2990 2991
2991 2992 return (rc);
2992 2993 }
2993 2994
2994 2995 static void *
2995 2996 deleg_state_mkkey(rfs4_entry_t u_entry)
2996 2997 {
2997 2998 rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
2998 2999
2999 3000 return (&dsp->rds_delegid);
3000 3001 }
3001 3002
3002 3003 static bool_t
3003 3004 rfs4_deleg_state_expiry(rfs4_entry_t u_entry)
3004 3005 {
3005 3006 rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
3006 3007
3007 3008 if (rfs4_dbe_is_invalid(dsp->rds_dbe))
3008 3009 return (TRUE);
3009 3010
3010 3011 if (dsp->rds_dtype == OPEN_DELEGATE_NONE)
3011 3012 return (TRUE);
3012 3013
3013 3014 if ((gethrestime_sec() - dsp->rds_client->rc_last_access
3014 3015 > rfs4_lease_time)) {
3015 3016 rfs4_dbe_invalidate(dsp->rds_dbe);
3016 3017 return (TRUE);
3017 3018 }
3018 3019
3019 3020 return (FALSE);
3020 3021 }
3021 3022
3022 3023 static bool_t
3023 3024 rfs4_deleg_state_create(rfs4_entry_t u_entry, void *argp)
3024 3025 {
3025 3026 rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
3026 3027 rfs4_file_t *fp = ((rfs4_deleg_state_t *)argp)->rds_finfo;
3027 3028 rfs4_client_t *cp = ((rfs4_deleg_state_t *)argp)->rds_client;
3028 3029
3029 3030 rfs4_dbe_hold(fp->rf_dbe);
3030 3031 rfs4_dbe_hold(cp->rc_dbe);
3031 3032
3032 3033 dsp->rds_delegid = get_stateid(rfs4_dbe_getid(dsp->rds_dbe));
3033 3034 dsp->rds_delegid.bits.type = DELEGID;
3034 3035 dsp->rds_finfo = fp;
3035 3036 dsp->rds_client = cp;
3036 3037 dsp->rds_dtype = OPEN_DELEGATE_NONE;
3037 3038
3038 3039 dsp->rds_time_granted = gethrestime_sec(); /* observability */
3039 3040 dsp->rds_time_revoked = 0;
3040 3041
3041 3042 list_link_init(&dsp->rds_node);
3042 3043
3043 3044 return (TRUE);
3044 3045 }
3045 3046
3046 3047 static void
3047 3048 rfs4_deleg_state_destroy(rfs4_entry_t u_entry)
3048 3049 {
3049 3050 rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
3050 3051
3051 3052 /* return delegation if necessary */
3052 3053 rfs4_return_deleg(dsp, FALSE);
3053 3054
3054 3055 /* Were done with the file */
3055 3056 rfs4_file_rele(dsp->rds_finfo);
3056 3057 dsp->rds_finfo = NULL;
3057 3058
3058 3059 /* And now with the openowner */
3059 3060 rfs4_client_rele(dsp->rds_client);
3060 3061 dsp->rds_client = NULL;
3061 3062 }
3062 3063
3063 3064 rfs4_deleg_state_t *
3064 3065 rfs4_finddeleg(rfs4_state_t *sp, bool_t *create)
3065 3066 {
3066 3067 rfs4_deleg_state_t ds, *dsp;
3067 3068 nfs4_srv_t *nsrv4 = nfs4_get_srv();
3068 3069
3069 3070 ds.rds_client = sp->rs_owner->ro_client;
3070 3071 ds.rds_finfo = sp->rs_finfo;
3071 3072
3072 3073 dsp = (rfs4_deleg_state_t *)rfs4_dbsearch(nsrv4->rfs4_deleg_idx, &ds,
3073 3074 create, &ds, RFS4_DBS_VALID);
3074 3075
3075 3076 return (dsp);
3076 3077 }
3077 3078
3078 3079 rfs4_deleg_state_t *
3079 3080 rfs4_finddelegstate(stateid_t *id)
3080 3081 {
3081 3082 rfs4_deleg_state_t *dsp;
3082 3083 bool_t create = FALSE;
3083 3084 nfs4_srv_t *nsrv4 = nfs4_get_srv();
3084 3085
3085 3086 dsp = (rfs4_deleg_state_t *)rfs4_dbsearch(nsrv4->rfs4_deleg_state_idx,
3086 3087 id, &create, NULL, RFS4_DBS_VALID);
3087 3088
3088 3089 return (dsp);
3089 3090 }
3090 3091
3091 3092 void
3092 3093 rfs4_deleg_state_rele(rfs4_deleg_state_t *dsp)
3093 3094 {
3094 3095 rfs4_dbe_rele(dsp->rds_dbe);
3095 3096 }
3096 3097
3097 3098 void
3098 3099 rfs4_update_lock_sequence(rfs4_lo_state_t *lsp)
3099 3100 {
3100 3101
3101 3102 rfs4_dbe_lock(lsp->rls_dbe);
3102 3103
3103 3104 /*
3104 3105 * If we are skipping sequence id checking, this means that
3105 3106 * this is the first lock request and therefore the sequence
3106 3107 * id does not need to be updated. This only happens on the
3107 3108 * first lock request for a lockowner
3108 3109 */
3109 3110 if (!lsp->rls_skip_seqid_check)
3110 3111 lsp->rls_seqid++;
3111 3112
3112 3113 rfs4_dbe_unlock(lsp->rls_dbe);
3113 3114 }
3114 3115
3115 3116 void
3116 3117 rfs4_update_lock_resp(rfs4_lo_state_t *lsp, nfs_resop4 *resp)
3117 3118 {
3118 3119
3119 3120 rfs4_dbe_lock(lsp->rls_dbe);
3120 3121
3121 3122 rfs4_free_reply(&lsp->rls_reply);
3122 3123
3123 3124 rfs4_copy_reply(&lsp->rls_reply, resp);
3124 3125
3125 3126 rfs4_dbe_unlock(lsp->rls_dbe);
3126 3127 }
3127 3128
3128 3129 void
3129 3130 rfs4_free_opens(rfs4_openowner_t *oo, bool_t invalidate,
3130 3131 bool_t close_of_client)
3131 3132 {
3132 3133 rfs4_state_t *sp;
3133 3134
3134 3135 rfs4_dbe_lock(oo->ro_dbe);
3135 3136
3136 3137 for (sp = list_head(&oo->ro_statelist); sp != NULL;
3137 3138 sp = list_next(&oo->ro_statelist, sp)) {
3138 3139 rfs4_state_close(sp, FALSE, close_of_client, CRED());
3139 3140 if (invalidate == TRUE)
3140 3141 rfs4_dbe_invalidate(sp->rs_dbe);
3141 3142 }
3142 3143
3143 3144 rfs4_dbe_invalidate(oo->ro_dbe);
3144 3145 rfs4_dbe_unlock(oo->ro_dbe);
3145 3146 }
3146 3147
3147 3148 static uint32_t
3148 3149 state_owner_file_hash(void *key)
3149 3150 {
3150 3151 rfs4_state_t *sp = key;
3151 3152
3152 3153 return (ADDRHASH(sp->rs_owner) ^ ADDRHASH(sp->rs_finfo));
3153 3154 }
3154 3155
3155 3156 static bool_t
3156 3157 state_owner_file_compare(rfs4_entry_t u_entry, void *key)
3157 3158 {
3158 3159 rfs4_state_t *sp = (rfs4_state_t *)u_entry;
3159 3160 rfs4_state_t *arg = key;
3160 3161
3161 3162 if (sp->rs_closed == TRUE)
3162 3163 return (FALSE);
3163 3164
3164 3165 return (arg->rs_owner == sp->rs_owner && arg->rs_finfo == sp->rs_finfo);
3165 3166 }
3166 3167
3167 3168 static void *
3168 3169 state_owner_file_mkkey(rfs4_entry_t u_entry)
3169 3170 {
3170 3171 return (u_entry);
3171 3172 }
3172 3173
3173 3174 static uint32_t
3174 3175 state_file_hash(void *key)
3175 3176 {
3176 3177 return (ADDRHASH(key));
3177 3178 }
3178 3179
3179 3180 static bool_t
3180 3181 state_file_compare(rfs4_entry_t u_entry, void *key)
3181 3182 {
3182 3183 rfs4_state_t *sp = (rfs4_state_t *)u_entry;
3183 3184 rfs4_file_t *fp = key;
3184 3185
3185 3186 if (sp->rs_closed == TRUE)
3186 3187 return (FALSE);
3187 3188
3188 3189 return (fp == sp->rs_finfo);
3189 3190 }
3190 3191
3191 3192 static void *
3192 3193 state_file_mkkey(rfs4_entry_t u_entry)
3193 3194 {
3194 3195 rfs4_state_t *sp = (rfs4_state_t *)u_entry;
3195 3196
3196 3197 return (sp->rs_finfo);
3197 3198 }
3198 3199
3199 3200 rfs4_state_t *
3200 3201 rfs4_findstate_by_owner_file(rfs4_openowner_t *oo, rfs4_file_t *fp,
3201 3202 bool_t *create)
3202 3203 {
3203 3204 rfs4_state_t *sp;
3204 3205 rfs4_state_t key;
3205 3206 nfs4_srv_t *nsrv4 = nfs4_get_srv();
3206 3207
3207 3208 key.rs_owner = oo;
3208 3209 key.rs_finfo = fp;
3209 3210
3210 3211 sp = (rfs4_state_t *)rfs4_dbsearch(nsrv4->rfs4_state_owner_file_idx,
3211 3212 &key, create, &key, RFS4_DBS_VALID);
3212 3213
3213 3214 return (sp);
3214 3215 }
3215 3216
3216 3217 /* This returns ANY state struct that refers to this file */
3217 3218 static rfs4_state_t *
3218 3219 rfs4_findstate_by_file(rfs4_file_t *fp)
3219 3220 {
3220 3221 bool_t create = FALSE;
3221 3222 nfs4_srv_t *nsrv4 = nfs4_get_srv();
3222 3223
3223 3224 return ((rfs4_state_t *)rfs4_dbsearch(nsrv4->rfs4_state_file_idx, fp,
3224 3225 &create, fp, RFS4_DBS_VALID));
3225 3226 }
3226 3227
3227 3228 static bool_t
3228 3229 rfs4_state_expiry(rfs4_entry_t u_entry)
3229 3230 {
3230 3231 rfs4_state_t *sp = (rfs4_state_t *)u_entry;
3231 3232
3232 3233 if (rfs4_dbe_is_invalid(sp->rs_dbe))
3233 3234 return (TRUE);
3234 3235
3235 3236 if (sp->rs_closed == TRUE &&
3236 3237 ((gethrestime_sec() - rfs4_dbe_get_timerele(sp->rs_dbe))
3237 3238 > rfs4_lease_time))
3238 3239 return (TRUE);
3239 3240
3240 3241 return ((gethrestime_sec() - sp->rs_owner->ro_client->rc_last_access
3241 3242 > rfs4_lease_time));
3242 3243 }
3243 3244
3244 3245 static bool_t
3245 3246 rfs4_state_create(rfs4_entry_t u_entry, void *argp)
3246 3247 {
3247 3248 rfs4_state_t *sp = (rfs4_state_t *)u_entry;
3248 3249 rfs4_file_t *fp = ((rfs4_state_t *)argp)->rs_finfo;
3249 3250 rfs4_openowner_t *oo = ((rfs4_state_t *)argp)->rs_owner;
3250 3251
3251 3252 rfs4_dbe_hold(fp->rf_dbe);
3252 3253 rfs4_dbe_hold(oo->ro_dbe);
3253 3254 sp->rs_stateid = get_stateid(rfs4_dbe_getid(sp->rs_dbe));
3254 3255 sp->rs_stateid.bits.type = OPENID;
3255 3256 sp->rs_owner = oo;
3256 3257 sp->rs_finfo = fp;
3257 3258
3258 3259 list_create(&sp->rs_lostatelist, sizeof (rfs4_lo_state_t),
3259 3260 offsetof(rfs4_lo_state_t, rls_node));
3260 3261
3261 3262 /* Insert state on per open owner's list */
3262 3263 rfs4_dbe_lock(oo->ro_dbe);
3263 3264 list_insert_tail(&oo->ro_statelist, sp);
3264 3265 rfs4_dbe_unlock(oo->ro_dbe);
3265 3266
3266 3267 return (TRUE);
3267 3268 }
3268 3269
3269 3270 static rfs4_state_t *
3270 3271 rfs4_findstate(stateid_t *id, rfs4_dbsearch_type_t find_invalid, bool_t lock_fp)
3271 3272 {
3272 3273 rfs4_state_t *sp;
3273 3274 bool_t create = FALSE;
3274 3275 nfs4_srv_t *nsrv4 = nfs4_get_srv();
3275 3276
3276 3277 sp = (rfs4_state_t *)rfs4_dbsearch(nsrv4->rfs4_state_idx, id,
3277 3278 &create, NULL, find_invalid);
3278 3279 if (lock_fp == TRUE && sp != NULL)
3279 3280 rw_enter(&sp->rs_finfo->rf_file_rwlock, RW_READER);
3280 3281
3281 3282 return (sp);
3282 3283 }
3283 3284
3284 3285 void
3285 3286 rfs4_state_close(rfs4_state_t *sp, bool_t lock_held, bool_t close_of_client,
3286 3287 cred_t *cr)
3287 3288 {
3288 3289 /* Remove the associated lo_state owners */
3289 3290 if (!lock_held)
3290 3291 rfs4_dbe_lock(sp->rs_dbe);
3291 3292
3292 3293 /*
3293 3294 * If refcnt == 0, the dbe is about to be destroyed.
3294 3295 * lock state will be released by the reaper thread.
3295 3296 */
3296 3297
3297 3298 if (rfs4_dbe_refcnt(sp->rs_dbe) > 0) {
3298 3299 if (sp->rs_closed == FALSE) {
3299 3300 rfs4_release_share_lock_state(sp, cr, close_of_client);
3300 3301 sp->rs_closed = TRUE;
3301 3302 }
3302 3303 }
3303 3304
3304 3305 if (!lock_held)
3305 3306 rfs4_dbe_unlock(sp->rs_dbe);
3306 3307 }
3307 3308
3308 3309 /*
3309 3310 * Remove all state associated with the given client.
3310 3311 */
3311 3312 void
3312 3313 rfs4_client_state_remove(rfs4_client_t *cp)
3313 3314 {
3314 3315 rfs4_openowner_t *oo;
3315 3316
3316 3317 rfs4_dbe_lock(cp->rc_dbe);
3317 3318
3318 3319 for (oo = list_head(&cp->rc_openownerlist); oo != NULL;
3319 3320 oo = list_next(&cp->rc_openownerlist, oo)) {
3320 3321 rfs4_free_opens(oo, TRUE, TRUE);
3321 3322 }
3322 3323
3323 3324 rfs4_dbe_unlock(cp->rc_dbe);
3324 3325 }
3325 3326
3326 3327 void
3327 3328 rfs4_client_close(rfs4_client_t *cp)
3328 3329 {
3329 3330 /* Mark client as going away. */
3330 3331 rfs4_dbe_lock(cp->rc_dbe);
3331 3332 rfs4_dbe_invalidate(cp->rc_dbe);
3332 3333 rfs4_dbe_unlock(cp->rc_dbe);
3333 3334
3334 3335 rfs4_client_state_remove(cp);
3335 3336
3336 3337 /* Release the client */
3337 3338 rfs4_client_rele(cp);
3338 3339 }
3339 3340
3340 3341 nfsstat4
3341 3342 rfs4_check_clientid(clientid4 *cp, int setclid_confirm)
3342 3343 {
3343 3344 cid *cidp = (cid *) cp;
3344 3345 nfs4_srv_t *nsrv4;
3345 3346
3346 3347 nsrv4 = nfs4_get_srv();
3347 3348
3348 3349 /*
3349 3350 * If we are booted as a cluster node, check the embedded nodeid.
3350 3351 * If it indicates that this clientid was generated on another node,
3351 3352 * inform the client accordingly.
3352 3353 */
3353 3354 if (cluster_bootflags & CLUSTER_BOOTED && foreign_clientid(cidp))
3354 3355 return (NFS4ERR_STALE_CLIENTID);
3355 3356
3356 3357 /*
3357 3358 * If the server start time matches the time provided
3358 3359 * by the client (via the clientid) and this is NOT a
3359 3360 * setclientid_confirm then return EXPIRED.
3360 3361 */
3361 3362 if (!setclid_confirm &&
3362 3363 cidp->impl_id.start_time == nsrv4->rfs4_start_time)
3363 3364 return (NFS4ERR_EXPIRED);
3364 3365
3365 3366 return (NFS4ERR_STALE_CLIENTID);
3366 3367 }
3367 3368
3368 3369 /*
3369 3370 * This is used when a stateid has not been found amongst the
3370 3371 * current server's state. Check the stateid to see if it
3371 3372 * was from this server instantiation or not.
3372 3373 */
3373 3374 static nfsstat4
3374 3375 what_stateid_error(stateid_t *id, stateid_type_t type)
3375 3376 {
3376 3377 nfs4_srv_t *nsrv4;
3377 3378
3378 3379 nsrv4 = nfs4_get_srv();
3379 3380
3380 3381 /* If we are booted as a cluster node, was stateid locally generated? */
3381 3382 if ((cluster_bootflags & CLUSTER_BOOTED) && foreign_stateid(id))
3382 3383 return (NFS4ERR_STALE_STATEID);
3383 3384
3384 3385 /* If types don't match then no use checking further */
3385 3386 if (type != id->bits.type)
3386 3387 return (NFS4ERR_BAD_STATEID);
3387 3388
3388 3389 /* From a different server instantiation, return STALE */
3389 3390 if (id->bits.boottime != nsrv4->rfs4_start_time)
3390 3391 return (NFS4ERR_STALE_STATEID);
3391 3392
3392 3393 /*
3393 3394 * From this server but the state is most likely beyond lease
3394 3395 * timeout: return NFS4ERR_EXPIRED. However, there is the
3395 3396 * case of a delegation stateid. For delegations, there is a
3396 3397 * case where the state can be removed without the client's
3397 3398 * knowledge/consent: revocation. In the case of delegation
3398 3399 * revocation, the delegation state will be removed and will
3399 3400 * not be found. If the client does something like a
3400 3401 * DELEGRETURN or even a READ/WRITE with a delegatoin stateid
3401 3402 * that has been revoked, the server should return BAD_STATEID
3402 3403 * instead of the more common EXPIRED error.
3403 3404 */
3404 3405 if (id->bits.boottime == nsrv4->rfs4_start_time) {
3405 3406 if (type == DELEGID)
3406 3407 return (NFS4ERR_BAD_STATEID);
3407 3408 else
3408 3409 return (NFS4ERR_EXPIRED);
3409 3410 }
3410 3411
3411 3412 return (NFS4ERR_BAD_STATEID);
3412 3413 }
3413 3414
3414 3415 /*
3415 3416 * Used later on to find the various state structs. When called from
3416 3417 * rfs4_check_stateid()->rfs4_get_all_state(), no file struct lock is
3417 3418 * taken (it is not needed) and helps on the read/write path with
3418 3419 * respect to performance.
3419 3420 */
3420 3421 static nfsstat4
3421 3422 rfs4_get_state_lockit(stateid4 *stateid, rfs4_state_t **spp,
3422 3423 rfs4_dbsearch_type_t find_invalid, bool_t lock_fp)
3423 3424 {
3424 3425 stateid_t *id = (stateid_t *)stateid;
3425 3426 rfs4_state_t *sp;
3426 3427
3427 3428 *spp = NULL;
3428 3429
3429 3430 /* If we are booted as a cluster node, was stateid locally generated? */
3430 3431 if ((cluster_bootflags & CLUSTER_BOOTED) && foreign_stateid(id))
3431 3432 return (NFS4ERR_STALE_STATEID);
3432 3433
3433 3434 sp = rfs4_findstate(id, find_invalid, lock_fp);
3434 3435 if (sp == NULL) {
3435 3436 return (what_stateid_error(id, OPENID));
3436 3437 }
3437 3438
3438 3439 if (rfs4_lease_expired(sp->rs_owner->ro_client)) {
3439 3440 if (lock_fp == TRUE)
3440 3441 rfs4_state_rele(sp);
3441 3442 else
3442 3443 rfs4_state_rele_nounlock(sp);
3443 3444 return (NFS4ERR_EXPIRED);
3444 3445 }
3445 3446
3446 3447 *spp = sp;
3447 3448
3448 3449 return (NFS4_OK);
3449 3450 }
3450 3451
3451 3452 nfsstat4
3452 3453 rfs4_get_state(stateid4 *stateid, rfs4_state_t **spp,
3453 3454 rfs4_dbsearch_type_t find_invalid)
3454 3455 {
3455 3456 return (rfs4_get_state_lockit(stateid, spp, find_invalid, TRUE));
3456 3457 }
3457 3458
3458 3459 int
3459 3460 rfs4_check_stateid_seqid(rfs4_state_t *sp, stateid4 *stateid)
3460 3461 {
3461 3462 stateid_t *id = (stateid_t *)stateid;
3462 3463
3463 3464 if (rfs4_lease_expired(sp->rs_owner->ro_client))
3464 3465 return (NFS4_CHECK_STATEID_EXPIRED);
3465 3466
3466 3467 /* Stateid is some time in the future - that's bad */
3467 3468 if (sp->rs_stateid.bits.chgseq < id->bits.chgseq)
3468 3469 return (NFS4_CHECK_STATEID_BAD);
3469 3470
3470 3471 if (sp->rs_stateid.bits.chgseq == id->bits.chgseq + 1)
3471 3472 return (NFS4_CHECK_STATEID_REPLAY);
3472 3473
3473 3474 /* Stateid is some time in the past - that's old */
3474 3475 if (sp->rs_stateid.bits.chgseq > id->bits.chgseq)
3475 3476 return (NFS4_CHECK_STATEID_OLD);
3476 3477
3477 3478 /* Caller needs to know about confirmation before closure */
3478 3479 if (sp->rs_owner->ro_need_confirm)
3479 3480 return (NFS4_CHECK_STATEID_UNCONFIRMED);
3480 3481
3481 3482 if (sp->rs_closed == TRUE)
3482 3483 return (NFS4_CHECK_STATEID_CLOSED);
3483 3484
3484 3485 return (NFS4_CHECK_STATEID_OKAY);
3485 3486 }
3486 3487
3487 3488 int
3488 3489 rfs4_check_lo_stateid_seqid(rfs4_lo_state_t *lsp, stateid4 *stateid)
3489 3490 {
3490 3491 stateid_t *id = (stateid_t *)stateid;
3491 3492
3492 3493 if (rfs4_lease_expired(lsp->rls_state->rs_owner->ro_client))
3493 3494 return (NFS4_CHECK_STATEID_EXPIRED);
3494 3495
3495 3496 /* Stateid is some time in the future - that's bad */
3496 3497 if (lsp->rls_lockid.bits.chgseq < id->bits.chgseq)
3497 3498 return (NFS4_CHECK_STATEID_BAD);
3498 3499
3499 3500 if (lsp->rls_lockid.bits.chgseq == id->bits.chgseq + 1)
3500 3501 return (NFS4_CHECK_STATEID_REPLAY);
3501 3502
3502 3503 /* Stateid is some time in the past - that's old */
3503 3504 if (lsp->rls_lockid.bits.chgseq > id->bits.chgseq)
3504 3505 return (NFS4_CHECK_STATEID_OLD);
3505 3506
3506 3507 if (lsp->rls_state->rs_closed == TRUE)
3507 3508 return (NFS4_CHECK_STATEID_CLOSED);
3508 3509
3509 3510 return (NFS4_CHECK_STATEID_OKAY);
3510 3511 }
3511 3512
3512 3513 nfsstat4
3513 3514 rfs4_get_deleg_state(stateid4 *stateid, rfs4_deleg_state_t **dspp)
3514 3515 {
3515 3516 stateid_t *id = (stateid_t *)stateid;
3516 3517 rfs4_deleg_state_t *dsp;
3517 3518
3518 3519 *dspp = NULL;
3519 3520
3520 3521 /* If we are booted as a cluster node, was stateid locally generated? */
3521 3522 if ((cluster_bootflags & CLUSTER_BOOTED) && foreign_stateid(id))
3522 3523 return (NFS4ERR_STALE_STATEID);
3523 3524
3524 3525 dsp = rfs4_finddelegstate(id);
3525 3526 if (dsp == NULL) {
3526 3527 return (what_stateid_error(id, DELEGID));
3527 3528 }
3528 3529
3529 3530 if (rfs4_lease_expired(dsp->rds_client)) {
3530 3531 rfs4_deleg_state_rele(dsp);
3531 3532 return (NFS4ERR_EXPIRED);
3532 3533 }
3533 3534
3534 3535 *dspp = dsp;
3535 3536
3536 3537 return (NFS4_OK);
3537 3538 }
3538 3539
3539 3540 nfsstat4
3540 3541 rfs4_get_lo_state(stateid4 *stateid, rfs4_lo_state_t **lspp, bool_t lock_fp)
3541 3542 {
3542 3543 stateid_t *id = (stateid_t *)stateid;
3543 3544 rfs4_lo_state_t *lsp;
3544 3545
3545 3546 *lspp = NULL;
3546 3547
3547 3548 /* If we are booted as a cluster node, was stateid locally generated? */
3548 3549 if ((cluster_bootflags & CLUSTER_BOOTED) && foreign_stateid(id))
3549 3550 return (NFS4ERR_STALE_STATEID);
3550 3551
3551 3552 lsp = rfs4_findlo_state(id, lock_fp);
3552 3553 if (lsp == NULL) {
3553 3554 return (what_stateid_error(id, LOCKID));
3554 3555 }
3555 3556
3556 3557 if (rfs4_lease_expired(lsp->rls_state->rs_owner->ro_client)) {
3557 3558 rfs4_lo_state_rele(lsp, lock_fp);
3558 3559 return (NFS4ERR_EXPIRED);
3559 3560 }
3560 3561
3561 3562 *lspp = lsp;
3562 3563
3563 3564 return (NFS4_OK);
3564 3565 }
3565 3566
3566 3567 static nfsstat4
3567 3568 rfs4_get_all_state(stateid4 *sid, rfs4_state_t **spp,
3568 3569 rfs4_deleg_state_t **dspp, rfs4_lo_state_t **lspp)
3569 3570 {
3570 3571 rfs4_state_t *sp = NULL;
3571 3572 rfs4_deleg_state_t *dsp = NULL;
3572 3573 rfs4_lo_state_t *lsp = NULL;
3573 3574 stateid_t *id;
3574 3575 nfsstat4 status;
3575 3576
3576 3577 *spp = NULL; *dspp = NULL; *lspp = NULL;
3577 3578
3578 3579 id = (stateid_t *)sid;
3579 3580 switch (id->bits.type) {
3580 3581 case OPENID:
3581 3582 status = rfs4_get_state_lockit(sid, &sp, FALSE, FALSE);
3582 3583 break;
3583 3584 case DELEGID:
3584 3585 status = rfs4_get_deleg_state(sid, &dsp);
3585 3586 break;
3586 3587 case LOCKID:
3587 3588 status = rfs4_get_lo_state(sid, &lsp, FALSE);
3588 3589 if (status == NFS4_OK) {
3589 3590 sp = lsp->rls_state;
3590 3591 rfs4_dbe_hold(sp->rs_dbe);
3591 3592 }
3592 3593 break;
3593 3594 default:
3594 3595 status = NFS4ERR_BAD_STATEID;
3595 3596 }
3596 3597
3597 3598 if (status == NFS4_OK) {
3598 3599 *spp = sp;
3599 3600 *dspp = dsp;
3600 3601 *lspp = lsp;
3601 3602 }
3602 3603
3603 3604 return (status);
3604 3605 }
3605 3606
3606 3607 /*
3607 3608 * Given the I/O mode (FREAD or FWRITE), this checks whether the
3608 3609 * rfs4_state_t struct has access to do this operation and if so
3609 3610 * return NFS4_OK; otherwise the proper NFSv4 error is returned.
3610 3611 */
3611 3612 nfsstat4
3612 3613 rfs4_state_has_access(rfs4_state_t *sp, int mode, vnode_t *vp)
3613 3614 {
3614 3615 nfsstat4 stat = NFS4_OK;
3615 3616 rfs4_file_t *fp;
3616 3617 bool_t create = FALSE;
3617 3618
3618 3619 rfs4_dbe_lock(sp->rs_dbe);
3619 3620 if (mode == FWRITE) {
3620 3621 if (!(sp->rs_share_access & OPEN4_SHARE_ACCESS_WRITE)) {
3621 3622 stat = NFS4ERR_OPENMODE;
3622 3623 }
3623 3624 } else if (mode == FREAD) {
3624 3625 if (!(sp->rs_share_access & OPEN4_SHARE_ACCESS_READ)) {
3625 3626 /*
3626 3627 * If we have OPENed the file with DENYing access
3627 3628 * to both READ and WRITE then no one else could
3628 3629 * have OPENed the file, hence no conflicting READ
3629 3630 * deny. This check is merely an optimization.
3630 3631 */
3631 3632 if (sp->rs_share_deny == OPEN4_SHARE_DENY_BOTH)
3632 3633 goto out;
3633 3634
3634 3635 /* Check against file struct's DENY mode */
3635 3636 fp = rfs4_findfile(vp, NULL, &create);
3636 3637 if (fp != NULL) {
3637 3638 int deny_read = 0;
3638 3639 rfs4_dbe_lock(fp->rf_dbe);
3639 3640 /*
3640 3641 * Check if any other open owner has the file
3641 3642 * OPENed with deny READ.
3642 3643 */
3643 3644 if (sp->rs_share_deny & OPEN4_SHARE_DENY_READ)
3644 3645 deny_read = 1;
3645 3646 ASSERT(fp->rf_deny_read >= deny_read);
3646 3647 if (fp->rf_deny_read > deny_read)
3647 3648 stat = NFS4ERR_OPENMODE;
3648 3649 rfs4_dbe_unlock(fp->rf_dbe);
3649 3650 rfs4_file_rele(fp);
3650 3651 }
3651 3652 }
3652 3653 } else {
3653 3654 /* Illegal I/O mode */
3654 3655 stat = NFS4ERR_INVAL;
3655 3656 }
3656 3657 out:
3657 3658 rfs4_dbe_unlock(sp->rs_dbe);
3658 3659 return (stat);
3659 3660 }
3660 3661
3661 3662 /*
3662 3663 * Given the I/O mode (FREAD or FWRITE), the vnode, the stateid and whether
3663 3664 * the file is being truncated, return NFS4_OK if allowed or appropriate
3664 3665 * V4 error if not. Note NFS4ERR_DELAY will be returned and a recall on
3665 3666 * the associated file will be done if the I/O is not consistent with any
3666 3667 * delegation in effect on the file. Should be holding VOP_RWLOCK, either
3667 3668 * as reader or writer as appropriate. rfs4_op_open will acquire the
3668 3669 * VOP_RWLOCK as writer when setting up delegation. If the stateid is bad
3669 3670 * this routine will return NFS4ERR_BAD_STATEID. In addition, through the
3670 3671 * deleg parameter, we will return whether a write delegation is held by
3671 3672 * the client associated with this stateid.
3672 3673 * If the server instance associated with the relevant client is in its
3673 3674 * grace period, return NFS4ERR_GRACE.
3674 3675 */
3675 3676
3676 3677 nfsstat4
3677 3678 rfs4_check_stateid(int mode, vnode_t *vp,
3678 3679 stateid4 *stateid, bool_t trunc, bool_t *deleg,
3679 3680 bool_t do_access, caller_context_t *ct)
3680 3681 {
3681 3682 rfs4_file_t *fp;
3682 3683 bool_t create = FALSE;
3683 3684 rfs4_state_t *sp;
3684 3685 rfs4_deleg_state_t *dsp;
3685 3686 rfs4_lo_state_t *lsp;
3686 3687 stateid_t *id = (stateid_t *)stateid;
3687 3688 nfsstat4 stat = NFS4_OK;
3688 3689
3689 3690 if (ct != NULL) {
3690 3691 ct->cc_sysid = 0;
3691 3692 ct->cc_pid = 0;
3692 3693 ct->cc_caller_id = nfs4_srv_caller_id;
3693 3694 ct->cc_flags = CC_DONTBLOCK;
3694 3695 }
3695 3696
3696 3697 if (ISSPECIAL(stateid)) {
3697 3698 fp = rfs4_findfile(vp, NULL, &create);
3698 3699 if (fp == NULL)
3699 3700 return (NFS4_OK);
3700 3701 if (fp->rf_dinfo.rd_dtype == OPEN_DELEGATE_NONE) {
3701 3702 rfs4_file_rele(fp);
3702 3703 return (NFS4_OK);
3703 3704 }
3704 3705 if (mode == FWRITE ||
3705 3706 fp->rf_dinfo.rd_dtype == OPEN_DELEGATE_WRITE) {
3706 3707 rfs4_recall_deleg(fp, trunc, NULL);
3707 3708 rfs4_file_rele(fp);
3708 3709 return (NFS4ERR_DELAY);
3709 3710 }
3710 3711 rfs4_file_rele(fp);
3711 3712 return (NFS4_OK);
3712 3713 } else {
3713 3714 stat = rfs4_get_all_state(stateid, &sp, &dsp, &lsp);
3714 3715 if (stat != NFS4_OK)
3715 3716 return (stat);
3716 3717 if (lsp != NULL) {
3717 3718 /* Is associated server instance in its grace period? */
3718 3719 if (rfs4_clnt_in_grace(lsp->rls_locker->rl_client)) {
3719 3720 rfs4_lo_state_rele(lsp, FALSE);
3720 3721 if (sp != NULL)
3721 3722 rfs4_state_rele_nounlock(sp);
3722 3723 return (NFS4ERR_GRACE);
3723 3724 }
3724 3725 if (id->bits.type == LOCKID) {
3725 3726 /* Seqid in the future? - that's bad */
3726 3727 if (lsp->rls_lockid.bits.chgseq <
3727 3728 id->bits.chgseq) {
3728 3729 rfs4_lo_state_rele(lsp, FALSE);
3729 3730 if (sp != NULL)
3730 3731 rfs4_state_rele_nounlock(sp);
3731 3732 return (NFS4ERR_BAD_STATEID);
3732 3733 }
3733 3734 /* Seqid in the past? - that's old */
3734 3735 if (lsp->rls_lockid.bits.chgseq >
3735 3736 id->bits.chgseq) {
3736 3737 rfs4_lo_state_rele(lsp, FALSE);
3737 3738 if (sp != NULL)
3738 3739 rfs4_state_rele_nounlock(sp);
3739 3740 return (NFS4ERR_OLD_STATEID);
3740 3741 }
3741 3742 /* Ensure specified filehandle matches */
3742 3743 if (lsp->rls_state->rs_finfo->rf_vp != vp) {
3743 3744 rfs4_lo_state_rele(lsp, FALSE);
3744 3745 if (sp != NULL)
3745 3746 rfs4_state_rele_nounlock(sp);
3746 3747 return (NFS4ERR_BAD_STATEID);
3747 3748 }
3748 3749 }
3749 3750 if (ct != NULL) {
3750 3751 ct->cc_sysid =
3751 3752 lsp->rls_locker->rl_client->rc_sysidt;
3752 3753 ct->cc_pid = lsp->rls_locker->rl_pid;
3753 3754 }
3754 3755 rfs4_lo_state_rele(lsp, FALSE);
3755 3756 }
3756 3757
3757 3758 /* Stateid provided was an "open" stateid */
3758 3759 if (sp != NULL) {
3759 3760 /* Is associated server instance in its grace period? */
3760 3761 if (rfs4_clnt_in_grace(sp->rs_owner->ro_client)) {
3761 3762 rfs4_state_rele_nounlock(sp);
3762 3763 return (NFS4ERR_GRACE);
3763 3764 }
3764 3765 if (id->bits.type == OPENID) {
3765 3766 /* Seqid in the future? - that's bad */
3766 3767 if (sp->rs_stateid.bits.chgseq <
3767 3768 id->bits.chgseq) {
3768 3769 rfs4_state_rele_nounlock(sp);
3769 3770 return (NFS4ERR_BAD_STATEID);
3770 3771 }
3771 3772 /* Seqid in the past - that's old */
3772 3773 if (sp->rs_stateid.bits.chgseq >
3773 3774 id->bits.chgseq) {
3774 3775 rfs4_state_rele_nounlock(sp);
3775 3776 return (NFS4ERR_OLD_STATEID);
3776 3777 }
3777 3778 }
3778 3779 /* Ensure specified filehandle matches */
3779 3780 if (sp->rs_finfo->rf_vp != vp) {
3780 3781 rfs4_state_rele_nounlock(sp);
3781 3782 return (NFS4ERR_BAD_STATEID);
3782 3783 }
3783 3784
3784 3785 if (sp->rs_owner->ro_need_confirm) {
3785 3786 rfs4_state_rele_nounlock(sp);
3786 3787 return (NFS4ERR_BAD_STATEID);
3787 3788 }
3788 3789
3789 3790 if (sp->rs_closed == TRUE) {
3790 3791 rfs4_state_rele_nounlock(sp);
3791 3792 return (NFS4ERR_OLD_STATEID);
3792 3793 }
3793 3794
3794 3795 if (do_access)
3795 3796 stat = rfs4_state_has_access(sp, mode, vp);
3796 3797 else
3797 3798 stat = NFS4_OK;
3798 3799
3799 3800 /*
3800 3801 * Return whether this state has write
3801 3802 * delegation if desired
3802 3803 */
3803 3804 if (deleg && (sp->rs_finfo->rf_dinfo.rd_dtype ==
3804 3805 OPEN_DELEGATE_WRITE))
3805 3806 *deleg = TRUE;
3806 3807
3807 3808 /*
3808 3809 * We got a valid stateid, so we update the
3809 3810 * lease on the client. Ideally we would like
3810 3811 * to do this after the calling op succeeds,
3811 3812 * but for now this will be good
3812 3813 * enough. Callers of this routine are
3813 3814 * currently insulated from the state stuff.
3814 3815 */
3815 3816 rfs4_update_lease(sp->rs_owner->ro_client);
3816 3817
3817 3818 /*
3818 3819 * If a delegation is present on this file and
3819 3820 * this is a WRITE, then update the lastwrite
3820 3821 * time to indicate that activity is present.
3821 3822 */
3822 3823 if (sp->rs_finfo->rf_dinfo.rd_dtype ==
3823 3824 OPEN_DELEGATE_WRITE &&
3824 3825 mode == FWRITE) {
3825 3826 sp->rs_finfo->rf_dinfo.rd_time_lastwrite =
3826 3827 gethrestime_sec();
3827 3828 }
3828 3829
3829 3830 rfs4_state_rele_nounlock(sp);
3830 3831
3831 3832 return (stat);
3832 3833 }
3833 3834
3834 3835 if (dsp != NULL) {
3835 3836 /* Is associated server instance in its grace period? */
3836 3837 if (rfs4_clnt_in_grace(dsp->rds_client)) {
3837 3838 rfs4_deleg_state_rele(dsp);
3838 3839 return (NFS4ERR_GRACE);
3839 3840 }
3840 3841 if (dsp->rds_delegid.bits.chgseq != id->bits.chgseq) {
3841 3842 rfs4_deleg_state_rele(dsp);
3842 3843 return (NFS4ERR_BAD_STATEID);
3843 3844 }
3844 3845
3845 3846 /* Ensure specified filehandle matches */
3846 3847 if (dsp->rds_finfo->rf_vp != vp) {
3847 3848 rfs4_deleg_state_rele(dsp);
3848 3849 return (NFS4ERR_BAD_STATEID);
3849 3850 }
3850 3851 /*
3851 3852 * Return whether this state has write
3852 3853 * delegation if desired
3853 3854 */
3854 3855 if (deleg && (dsp->rds_finfo->rf_dinfo.rd_dtype ==
3855 3856 OPEN_DELEGATE_WRITE))
3856 3857 *deleg = TRUE;
3857 3858
3858 3859 rfs4_update_lease(dsp->rds_client);
3859 3860
3860 3861 /*
3861 3862 * If a delegation is present on this file and
3862 3863 * this is a WRITE, then update the lastwrite
3863 3864 * time to indicate that activity is present.
3864 3865 */
3865 3866 if (dsp->rds_finfo->rf_dinfo.rd_dtype ==
3866 3867 OPEN_DELEGATE_WRITE && mode == FWRITE) {
3867 3868 dsp->rds_finfo->rf_dinfo.rd_time_lastwrite =
3868 3869 gethrestime_sec();
3869 3870 }
3870 3871
3871 3872 /*
3872 3873 * XXX - what happens if this is a WRITE and the
3873 3874 * delegation type of for READ.
3874 3875 */
3875 3876 rfs4_deleg_state_rele(dsp);
3876 3877
3877 3878 return (stat);
3878 3879 }
3879 3880 /*
3880 3881 * If we got this far, something bad happened
3881 3882 */
3882 3883 return (NFS4ERR_BAD_STATEID);
3883 3884 }
3884 3885 }
3885 3886
3886 3887
3887 3888 /*
3888 3889 * This is a special function in that for the file struct provided the
3889 3890 * server wants to remove/close all current state associated with the
3890 3891 * file. The prime use of this would be with OP_REMOVE to force the
3891 3892 * release of state and particularly of file locks.
3892 3893 *
3893 3894 * There is an assumption that there is no delegations outstanding on
3894 3895 * this file at this point. The caller should have waited for those
3895 3896 * to be returned or revoked.
3896 3897 */
3897 3898 void
3898 3899 rfs4_close_all_state(rfs4_file_t *fp)
3899 3900 {
3900 3901 rfs4_state_t *sp;
3901 3902
3902 3903 rfs4_dbe_lock(fp->rf_dbe);
3903 3904
3904 3905 #ifdef DEBUG
3905 3906 /* only applies when server is handing out delegations */
3906 3907 if (nfs4_get_deleg_policy() != SRV_NEVER_DELEGATE)
3907 3908 ASSERT(fp->rf_dinfo.rd_hold_grant > 0);
3908 3909 #endif
3909 3910
3910 3911 /* No delegations for this file */
3911 3912 ASSERT(list_is_empty(&fp->rf_delegstatelist));
3912 3913
3913 3914 /* Make sure that it can not be found */
3914 3915 rfs4_dbe_invalidate(fp->rf_dbe);
3915 3916
3916 3917 if (fp->rf_vp == NULL) {
3917 3918 rfs4_dbe_unlock(fp->rf_dbe);
3918 3919 return;
3919 3920 }
3920 3921 rfs4_dbe_unlock(fp->rf_dbe);
3921 3922
3922 3923 /*
3923 3924 * Hold as writer to prevent other server threads from
3924 3925 * processing requests related to the file while all state is
3925 3926 * being removed.
3926 3927 */
3927 3928 rw_enter(&fp->rf_file_rwlock, RW_WRITER);
3928 3929
3929 3930 /* Remove ALL state from the file */
3930 3931 while (sp = rfs4_findstate_by_file(fp)) {
3931 3932 rfs4_state_close(sp, FALSE, FALSE, CRED());
3932 3933 rfs4_state_rele_nounlock(sp);
3933 3934 }
3934 3935
3935 3936 /*
3936 3937 * This is only safe since there are no further references to
3937 3938 * the file.
3938 3939 */
3939 3940 rfs4_dbe_lock(fp->rf_dbe);
3940 3941 if (fp->rf_vp) {
3941 3942 vnode_t *vp = fp->rf_vp;
3942 3943
3943 3944 mutex_enter(&vp->v_vsd_lock);
3944 3945 (void) vsd_set(vp, nfs4_srv_vkey, NULL);
3945 3946 mutex_exit(&vp->v_vsd_lock);
3946 3947 VN_RELE(vp);
3947 3948 fp->rf_vp = NULL;
3948 3949 }
3949 3950 rfs4_dbe_unlock(fp->rf_dbe);
3950 3951
3951 3952 /* Finally let other references to proceed */
3952 3953 rw_exit(&fp->rf_file_rwlock);
3953 3954 }
3954 3955
3955 3956 /*
3956 3957 * This function is used as a target for the rfs4_dbe_walk() call
3957 3958 * below. The purpose of this function is to see if the
3958 3959 * lockowner_state refers to a file that resides within the exportinfo
3959 3960 * export. If so, then remove the lock_owner state (file locks and
3960 3961 * share "locks") for this object since the intent is the server is
3961 3962 * unexporting the specified directory. Be sure to invalidate the
3962 3963 * object after the state has been released
3963 3964 */
3964 3965 static void
3965 3966 rfs4_lo_state_walk_callout(rfs4_entry_t u_entry, void *e)
3966 3967 {
3967 3968 rfs4_lo_state_t *lsp = (rfs4_lo_state_t *)u_entry;
3968 3969 struct exportinfo *exi = (struct exportinfo *)e;
3969 3970 nfs_fh4_fmt_t fhfmt4, *exi_fhp, *finfo_fhp;
3970 3971 fhandle_t *efhp;
3971 3972
3972 3973 efhp = (fhandle_t *)&exi->exi_fh;
3973 3974 exi_fhp = (nfs_fh4_fmt_t *)&fhfmt4;
3974 3975
3975 3976 FH_TO_FMT4(efhp, exi_fhp);
3976 3977
3977 3978 finfo_fhp = (nfs_fh4_fmt_t *)lsp->rls_state->rs_finfo->
3978 3979 rf_filehandle.nfs_fh4_val;
3979 3980
3980 3981 if (EQFSID(&finfo_fhp->fh4_fsid, &exi_fhp->fh4_fsid) &&
3981 3982 bcmp(&finfo_fhp->fh4_xdata, &exi_fhp->fh4_xdata,
3982 3983 exi_fhp->fh4_xlen) == 0) {
3983 3984 rfs4_state_close(lsp->rls_state, FALSE, FALSE, CRED());
3984 3985 rfs4_dbe_invalidate(lsp->rls_dbe);
3985 3986 rfs4_dbe_invalidate(lsp->rls_state->rs_dbe);
3986 3987 }
3987 3988 }
3988 3989
3989 3990 /*
3990 3991 * This function is used as a target for the rfs4_dbe_walk() call
3991 3992 * below. The purpose of this function is to see if the state refers
3992 3993 * to a file that resides within the exportinfo export. If so, then
3993 3994 * remove the open state for this object since the intent is the
3994 3995 * server is unexporting the specified directory. The main result for
3995 3996 * this type of entry is to invalidate it such it will not be found in
3996 3997 * the future.
3997 3998 */
3998 3999 static void
3999 4000 rfs4_state_walk_callout(rfs4_entry_t u_entry, void *e)
4000 4001 {
4001 4002 rfs4_state_t *sp = (rfs4_state_t *)u_entry;
4002 4003 struct exportinfo *exi = (struct exportinfo *)e;
4003 4004 nfs_fh4_fmt_t fhfmt4, *exi_fhp, *finfo_fhp;
4004 4005 fhandle_t *efhp;
4005 4006
4006 4007 efhp = (fhandle_t *)&exi->exi_fh;
4007 4008 exi_fhp = (nfs_fh4_fmt_t *)&fhfmt4;
4008 4009
4009 4010 FH_TO_FMT4(efhp, exi_fhp);
4010 4011
4011 4012 finfo_fhp =
4012 4013 (nfs_fh4_fmt_t *)sp->rs_finfo->rf_filehandle.nfs_fh4_val;
4013 4014
4014 4015 if (EQFSID(&finfo_fhp->fh4_fsid, &exi_fhp->fh4_fsid) &&
4015 4016 bcmp(&finfo_fhp->fh4_xdata, &exi_fhp->fh4_xdata,
4016 4017 exi_fhp->fh4_xlen) == 0) {
4017 4018 rfs4_state_close(sp, TRUE, FALSE, CRED());
4018 4019 rfs4_dbe_invalidate(sp->rs_dbe);
4019 4020 }
4020 4021 }
4021 4022
4022 4023 /*
4023 4024 * This function is used as a target for the rfs4_dbe_walk() call
4024 4025 * below. The purpose of this function is to see if the state refers
4025 4026 * to a file that resides within the exportinfo export. If so, then
4026 4027 * remove the deleg state for this object since the intent is the
4027 4028 * server is unexporting the specified directory. The main result for
4028 4029 * this type of entry is to invalidate it such it will not be found in
4029 4030 * the future.
4030 4031 */
4031 4032 static void
4032 4033 rfs4_deleg_state_walk_callout(rfs4_entry_t u_entry, void *e)
4033 4034 {
4034 4035 rfs4_deleg_state_t *dsp = (rfs4_deleg_state_t *)u_entry;
4035 4036 struct exportinfo *exi = (struct exportinfo *)e;
4036 4037 nfs_fh4_fmt_t fhfmt4, *exi_fhp, *finfo_fhp;
4037 4038 fhandle_t *efhp;
4038 4039
4039 4040 efhp = (fhandle_t *)&exi->exi_fh;
4040 4041 exi_fhp = (nfs_fh4_fmt_t *)&fhfmt4;
4041 4042
4042 4043 FH_TO_FMT4(efhp, exi_fhp);
4043 4044
4044 4045 finfo_fhp =
4045 4046 (nfs_fh4_fmt_t *)dsp->rds_finfo->rf_filehandle.nfs_fh4_val;
4046 4047
4047 4048 if (EQFSID(&finfo_fhp->fh4_fsid, &exi_fhp->fh4_fsid) &&
4048 4049 bcmp(&finfo_fhp->fh4_xdata, &exi_fhp->fh4_xdata,
4049 4050 exi_fhp->fh4_xlen) == 0) {
4050 4051 rfs4_dbe_invalidate(dsp->rds_dbe);
4051 4052 }
4052 4053 }
4053 4054
4054 4055 /*
4055 4056 * This function is used as a target for the rfs4_dbe_walk() call
4056 4057 * below. The purpose of this function is to see if the state refers
4057 4058 * to a file that resides within the exportinfo export. If so, then
4058 4059 * release vnode hold for this object since the intent is the server
4059 4060 * is unexporting the specified directory. Invalidation will prevent
4060 4061 * this struct from being found in the future.
4061 4062 */
4062 4063 static void
4063 4064 rfs4_file_walk_callout(rfs4_entry_t u_entry, void *e)
4064 4065 {
4065 4066 rfs4_file_t *fp = (rfs4_file_t *)u_entry;
4066 4067 struct exportinfo *exi = (struct exportinfo *)e;
4067 4068 nfs_fh4_fmt_t fhfmt4, *exi_fhp, *finfo_fhp;
4068 4069 fhandle_t *efhp;
4069 4070
4070 4071 efhp = (fhandle_t *)&exi->exi_fh;
4071 4072 exi_fhp = (nfs_fh4_fmt_t *)&fhfmt4;
4072 4073
4073 4074 FH_TO_FMT4(efhp, exi_fhp);
4074 4075
4075 4076 finfo_fhp = (nfs_fh4_fmt_t *)fp->rf_filehandle.nfs_fh4_val;
4076 4077
4077 4078 if (EQFSID(&finfo_fhp->fh4_fsid, &exi_fhp->fh4_fsid) &&
4078 4079 bcmp(&finfo_fhp->fh4_xdata, &exi_fhp->fh4_xdata,
4079 4080 exi_fhp->fh4_xlen) == 0) {
4080 4081 if (fp->rf_vp) {
4081 4082 vnode_t *vp = fp->rf_vp;
4082 4083
4083 4084 /*
4084 4085 * don't leak monitors and remove the reference
4085 4086 * put on the vnode when the delegation was granted.
4086 4087 */
4087 4088 if (fp->rf_dinfo.rd_dtype == OPEN_DELEGATE_READ) {
4088 4089 (void) fem_uninstall(vp, deleg_rdops,
4089 4090 (void *)fp);
4090 4091 vn_open_downgrade(vp, FREAD);
4091 4092 } else if (fp->rf_dinfo.rd_dtype ==
4092 4093 OPEN_DELEGATE_WRITE) {
4093 4094 (void) fem_uninstall(vp, deleg_wrops,
4094 4095 (void *)fp);
4095 4096 vn_open_downgrade(vp, FREAD|FWRITE);
4096 4097 }
4097 4098 mutex_enter(&vp->v_vsd_lock);
4098 4099 (void) vsd_set(vp, nfs4_srv_vkey, NULL);
4099 4100 mutex_exit(&vp->v_vsd_lock);
4100 4101 VN_RELE(vp);
4101 4102 fp->rf_vp = NULL;
4102 4103 }
4103 4104 rfs4_dbe_invalidate(fp->rf_dbe);
4104 4105 }
4105 4106 }
4106 4107
4107 4108 /*
4108 4109 * Given a directory that is being unexported, cleanup/release all
4109 4110 * state in the server that refers to objects residing underneath this
4110 4111 * particular export. The ordering of the release is important.
4111 4112 * Lock_owner, then state and then file.
4112 4113 *
4113 4114 * NFS zones note: nfs_export.c:unexport() calls this from a
4114 4115 * thread in the global zone for NGZ data structures, so we
4115 4116 * CANNOT use zone_getspecific anywhere in this code path.
4116 4117 */
4117 4118 void
4118 4119 rfs4_clean_state_exi(nfs_export_t *ne, struct exportinfo *exi)
4119 4120 {
4120 4121 nfs_globals_t *ng;
4121 4122 nfs4_srv_t *nsrv4;
4122 4123
4123 4124 ng = ne->ne_globals;
4124 4125 ASSERT(ng->nfs_zoneid == exi->exi_zoneid);
4125 4126 nsrv4 = ng->nfs4_srv;
4126 4127
4127 4128 mutex_enter(&nsrv4->state_lock);
4128 4129
4129 4130 if (nsrv4->nfs4_server_state == NULL) {
4130 4131 mutex_exit(&nsrv4->state_lock);
4131 4132 return;
4132 4133 }
4133 4134
4134 4135 rfs4_dbe_walk(nsrv4->rfs4_lo_state_tab,
4135 4136 rfs4_lo_state_walk_callout, exi);
4136 4137 rfs4_dbe_walk(nsrv4->rfs4_state_tab, rfs4_state_walk_callout, exi);
4137 4138 rfs4_dbe_walk(nsrv4->rfs4_deleg_state_tab,
4138 4139 rfs4_deleg_state_walk_callout, exi);
4139 4140 rfs4_dbe_walk(nsrv4->rfs4_file_tab, rfs4_file_walk_callout, exi);
4140 4141
4141 4142 mutex_exit(&nsrv4->state_lock);
4142 4143 }
|
↓ open down ↓ |
2605 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX