1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21
22 /*
23 * Copyright 2009 Sun Microsystems, Inc. All rights reserved.
24 * Use is subject to license terms.
25 */
26
27 /*
28 * Copyright 2018 Nexenta Systems, Inc.
29 */
30
31 #include <sys/systm.h>
32 #include <rpc/auth.h>
33 #include <rpc/clnt.h>
34 #include <nfs/nfs4_kprot.h>
35 #include <nfs/nfs4.h>
36 #include <nfs/lm.h>
37 #include <sys/cmn_err.h>
38 #include <sys/disp.h>
39 #include <sys/sdt.h>
40
41 #include <sys/pathname.h>
42
43 #include <sys/strsubr.h>
44 #include <sys/ddi.h>
45
46 #include <sys/vnode.h>
47 #include <sys/sdt.h>
48 #include <inet/common.h>
49 #include <inet/ip.h>
50 #include <inet/ip6.h>
51
52 #define MAX_READ_DELEGATIONS 5
53
54 static int rfs4_deleg_wlp = 5;
55 static int rfs4_deleg_disabled;
56 static int rfs4_max_setup_cb_tries = 5;
57
58 #ifdef DEBUG
59
60 static int rfs4_test_cbgetattr_fail = 0;
61 int rfs4_cb_null;
62 int rfs4_cb_debug;
63 int rfs4_deleg_debug;
64
65 #endif
66
67 static void rfs4_recall_file(rfs4_file_t *,
68 void (*recall)(rfs4_deleg_state_t *, bool_t),
69 bool_t, rfs4_client_t *);
70 static void rfs4_revoke_file(rfs4_file_t *);
71 static void rfs4_cb_chflush(rfs4_cbinfo_t *);
72 static CLIENT *rfs4_cb_getch(rfs4_cbinfo_t *);
73 static void rfs4_cb_freech(rfs4_cbinfo_t *, CLIENT *, bool_t);
74 static rfs4_deleg_state_t *rfs4_deleg_state(rfs4_state_t *,
75 open_delegation_type4, int *);
76
77 /*
78 * Convert a universal address to an transport specific
79 * address using inet_pton.
80 */
81 static int
82 uaddr2sockaddr(int af, char *ua, void *ap, in_port_t *pp)
83 {
84 int dots = 0, i, j, len, k;
85 unsigned char c;
86 in_port_t port = 0;
87
88 len = strlen(ua);
89
90 for (i = len-1; i >= 0; i--) {
91
92 if (ua[i] == '.')
93 dots++;
94
95 if (dots == 2) {
96
97 ua[i] = '\0';
98 /*
99 * We use k to remember were to stick '.' back, since
100 * ua was kmem_allocateded from the pool len+1.
101 */
102 k = i;
103 if (inet_pton(af, ua, ap) == 1) {
104
105 c = 0;
106
107 for (j = i+1; j < len; j++) {
108 if (ua[j] == '.') {
109 port = c << 8;
110 c = 0;
111 } else if (ua[j] >= '0' &&
112 ua[j] <= '9') {
113 c *= 10;
114 c += ua[j] - '0';
115 } else {
116 ua[k] = '.';
117 return (EINVAL);
118 }
119 }
120 port += c;
121
122 *pp = htons(port);
123
124 ua[k] = '.';
125 return (0);
126 } else {
127 ua[k] = '.';
128 return (EINVAL);
129 }
130 }
131 }
132
133 return (EINVAL);
134 }
135
136 /*
137 * Update the delegation policy with the
138 * value of "new_policy"
139 */
140 void
141 rfs4_set_deleg_policy(nfs4_srv_t *nsrv4, srv_deleg_policy_t new_policy)
142 {
143 rw_enter(&nsrv4->deleg_policy_lock, RW_WRITER);
144 nsrv4->nfs4_deleg_policy = new_policy;
145 rw_exit(&nsrv4->deleg_policy_lock);
146 }
147
148 void
149 rfs4_hold_deleg_policy(nfs4_srv_t *nsrv4)
150 {
151 rw_enter(&nsrv4->deleg_policy_lock, RW_READER);
152 }
153
154 void
155 rfs4_rele_deleg_policy(nfs4_srv_t *nsrv4)
156 {
157 rw_exit(&nsrv4->deleg_policy_lock);
158 }
159
160 srv_deleg_policy_t
161 nfs4_get_deleg_policy()
162 {
163 nfs4_srv_t *nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
164 return (nsrv4->nfs4_deleg_policy);
165 }
166
167
168 /*
169 * This free function is to be used when the client struct is being
170 * released and nothing at all is needed of the callback info any
171 * longer.
172 */
173 void
174 rfs4_cbinfo_free(rfs4_cbinfo_t *cbp)
175 {
176 char *addr = cbp->cb_callback.cb_location.r_addr;
177 char *netid = cbp->cb_callback.cb_location.r_netid;
178
179 /* Free old address if any */
180
181 if (addr)
182 kmem_free(addr, strlen(addr) + 1);
183 if (netid)
184 kmem_free(netid, strlen(netid) + 1);
185
186 addr = cbp->cb_newer.cb_callback.cb_location.r_addr;
187 netid = cbp->cb_newer.cb_callback.cb_location.r_netid;
188
189 if (addr)
190 kmem_free(addr, strlen(addr) + 1);
191 if (netid)
192 kmem_free(netid, strlen(netid) + 1);
193
194 if (cbp->cb_chc_free) {
195 rfs4_cb_chflush(cbp);
196 }
197 }
198
199 /*
200 * The server uses this to check the callback path supplied by the
201 * client. The callback connection is marked "in progress" while this
202 * work is going on and then eventually marked either OK or FAILED.
203 * This work can be done as part of a separate thread and at the end
204 * of this the thread will exit or it may be done such that the caller
205 * will continue with other work.
206 */
207 static void
208 rfs4_do_cb_null(rfs4_client_t *cp)
209 {
210 struct timeval tv;
211 CLIENT *ch;
212 rfs4_cbstate_t newstate;
213 rfs4_cbinfo_t *cbp = &cp->rc_cbinfo;
214
215 mutex_enter(cbp->cb_lock);
216 /* If another thread is doing CB_NULL RPC then return */
217 if (cbp->cb_nullcaller == TRUE) {
218 mutex_exit(cbp->cb_lock);
219 rfs4_client_rele(cp);
220 zthread_exit();
221 }
222
223 /* Mark the cbinfo as having a thread in the NULL callback */
224 cbp->cb_nullcaller = TRUE;
225
226 /*
227 * Are there other threads still using the cbinfo client
228 * handles? If so, this thread must wait before going and
229 * mucking aroiund with the callback information
230 */
231 while (cbp->cb_refcnt != 0)
232 cv_wait(cbp->cb_cv_nullcaller, cbp->cb_lock);
233
234 /*
235 * This thread itself may find that new callback info has
236 * arrived and is set up to handle this case and redrive the
237 * call to the client's callback server.
238 */
239 retry:
240 if (cbp->cb_newer.cb_new == TRUE &&
241 cbp->cb_newer.cb_confirmed == TRUE) {
242 char *addr = cbp->cb_callback.cb_location.r_addr;
243 char *netid = cbp->cb_callback.cb_location.r_netid;
244
245 /*
246 * Free the old stuff if it exists; may be the first
247 * time through this path
248 */
249 if (addr)
250 kmem_free(addr, strlen(addr) + 1);
251 if (netid)
252 kmem_free(netid, strlen(netid) + 1);
253
254 /* Move over the addr/netid */
255 cbp->cb_callback.cb_location.r_addr =
256 cbp->cb_newer.cb_callback.cb_location.r_addr;
257 cbp->cb_newer.cb_callback.cb_location.r_addr = NULL;
258 cbp->cb_callback.cb_location.r_netid =
259 cbp->cb_newer.cb_callback.cb_location.r_netid;
260 cbp->cb_newer.cb_callback.cb_location.r_netid = NULL;
261
262 /* Get the program number */
263 cbp->cb_callback.cb_program =
264 cbp->cb_newer.cb_callback.cb_program;
265 cbp->cb_newer.cb_callback.cb_program = 0;
266
267 /* Don't forget the protocol's "cb_ident" field */
268 cbp->cb_ident = cbp->cb_newer.cb_ident;
269 cbp->cb_newer.cb_ident = 0;
270
271 /* no longer new */
272 cbp->cb_newer.cb_new = FALSE;
273 cbp->cb_newer.cb_confirmed = FALSE;
274
275 /* get rid of the old client handles that may exist */
276 rfs4_cb_chflush(cbp);
277
278 cbp->cb_state = CB_NONE;
279 cbp->cb_timefailed = 0; /* reset the clock */
280 cbp->cb_notified_of_cb_path_down = TRUE;
281 }
282
283 if (cbp->cb_state != CB_NONE) {
284 cv_broadcast(cbp->cb_cv); /* let the others know */
285 cbp->cb_nullcaller = FALSE;
286 mutex_exit(cbp->cb_lock);
287 rfs4_client_rele(cp);
288 zthread_exit();
289 }
290
291 /* mark rfs4_client_t as CALLBACK NULL in progress */
292 cbp->cb_state = CB_INPROG;
293 mutex_exit(cbp->cb_lock);
294
295 /* get/generate a client handle */
296 if ((ch = rfs4_cb_getch(cbp)) == NULL) {
297 mutex_enter(cbp->cb_lock);
298 cbp->cb_state = CB_BAD;
299 cbp->cb_timefailed = gethrestime_sec(); /* observability */
300 goto retry;
301 }
302
303
304 tv.tv_sec = 30;
305 tv.tv_usec = 0;
306 if (clnt_call(ch, CB_NULL, xdr_void, NULL, xdr_void, NULL, tv) != 0) {
307 newstate = CB_BAD;
308 } else {
309 newstate = CB_OK;
310 #ifdef DEBUG
311 rfs4_cb_null++;
312 #endif
313 }
314
315 /* Check to see if the client has specified new callback info */
316 mutex_enter(cbp->cb_lock);
317 rfs4_cb_freech(cbp, ch, TRUE);
318 if (cbp->cb_newer.cb_new == TRUE &&
319 cbp->cb_newer.cb_confirmed == TRUE) {
320 goto retry; /* give the CB_NULL another chance */
321 }
322
323 cbp->cb_state = newstate;
324 if (cbp->cb_state == CB_BAD)
325 cbp->cb_timefailed = gethrestime_sec(); /* observability */
326
327 cv_broadcast(cbp->cb_cv); /* start up the other threads */
328 cbp->cb_nullcaller = FALSE;
329 mutex_exit(cbp->cb_lock);
330 rfs4_client_rele(cp);
331 zthread_exit();
332 }
333
334 /*
335 * Given a client struct, inspect the callback info to see if the
336 * callback path is up and available.
337 *
338 * If new callback path is available and no one has set it up then
339 * try to set it up. If setup is not successful after 5 tries (5 secs)
340 * then gives up and returns NULL.
341 *
342 * If callback path is being initialized, then wait for the CB_NULL RPC
343 * call to occur.
344 */
345 static rfs4_cbinfo_t *
346 rfs4_cbinfo_hold(rfs4_client_t *cp)
347 {
348 rfs4_cbinfo_t *cbp = &cp->rc_cbinfo;
349 int retries = 0;
350
351 mutex_enter(cbp->cb_lock);
352
353 while (cbp->cb_newer.cb_new == TRUE && cbp->cb_nullcaller == FALSE) {
354 /*
355 * Looks like a new callback path may be available and
356 * noone has set it up.
357 */
358 mutex_exit(cbp->cb_lock);
359 rfs4_dbe_hold(cp->rc_dbe);
360 rfs4_do_cb_null(cp); /* caller will release client hold */
361
362 mutex_enter(cbp->cb_lock);
363 /*
364 * If callback path is no longer new, or it's being setup
365 * then stop and wait for it to be done.
366 */
367 if (cbp->cb_newer.cb_new == FALSE || cbp->cb_nullcaller == TRUE)
368 break;
369 mutex_exit(cbp->cb_lock);
370
371 if (++retries >= rfs4_max_setup_cb_tries)
372 return (NULL);
373 delay(hz);
374 mutex_enter(cbp->cb_lock);
375 }
376
377 /* Is there a thread working on doing the CB_NULL RPC? */
378 if (cbp->cb_nullcaller == TRUE)
379 cv_wait(cbp->cb_cv, cbp->cb_lock); /* if so, wait on it */
380
381 /* If the callback path is not okay (up and running), just quit */
382 if (cbp->cb_state != CB_OK) {
383 mutex_exit(cbp->cb_lock);
384 return (NULL);
385 }
386
387 /* Let someone know we are using the current callback info */
388 cbp->cb_refcnt++;
389 mutex_exit(cbp->cb_lock);
390 return (cbp);
391 }
392
393 /*
394 * The caller is done with the callback info. It may be that the
395 * caller's RPC failed and the NFSv4 client has actually provided new
396 * callback information. If so, let the caller know so they can
397 * advantage of this and maybe retry the RPC that originally failed.
398 */
399 static int
400 rfs4_cbinfo_rele(rfs4_cbinfo_t *cbp, rfs4_cbstate_t newstate)
401 {
402 int cb_new = FALSE;
403
404 mutex_enter(cbp->cb_lock);
405
406 /* The caller gets a chance to mark the callback info as bad */
407 if (newstate != CB_NOCHANGE)
408 cbp->cb_state = newstate;
409 if (newstate == CB_FAILED) {
410 cbp->cb_timefailed = gethrestime_sec(); /* observability */
411 cbp->cb_notified_of_cb_path_down = FALSE;
412 }
413
414 cbp->cb_refcnt--; /* no longer using the information */
415
416 /*
417 * A thread may be waiting on this one to finish and if so,
418 * let it know that it is okay to do the CB_NULL to the
419 * client's callback server.
420 */
421 if (cbp->cb_refcnt == 0 && cbp->cb_nullcaller)
422 cv_broadcast(cbp->cb_cv_nullcaller);
423
424 /*
425 * If this is the last thread to use the callback info and
426 * there is new callback information to try and no thread is
427 * there ready to do the CB_NULL, then return true to teh
428 * caller so they can do the CB_NULL
429 */
430 if (cbp->cb_refcnt == 0 &&
431 cbp->cb_nullcaller == FALSE &&
432 cbp->cb_newer.cb_new == TRUE &&
433 cbp->cb_newer.cb_confirmed == TRUE)
434 cb_new = TRUE;
435
436 mutex_exit(cbp->cb_lock);
437
438 return (cb_new);
439 }
440
441 /*
442 * Given the information in the callback info struct, create a client
443 * handle that can be used by the server for its callback path.
444 */
445 static CLIENT *
446 rfs4_cbch_init(rfs4_cbinfo_t *cbp)
447 {
448 struct knetconfig knc;
449 vnode_t *vp;
450 struct sockaddr_in addr4;
451 struct sockaddr_in6 addr6;
452 void *addr, *taddr;
453 in_port_t *pp;
454 int af;
455 char *devnam;
456 struct netbuf nb;
457 int size;
458 CLIENT *ch = NULL;
459 int useresvport = 0;
460
461 mutex_enter(cbp->cb_lock);
462
463 if (cbp->cb_callback.cb_location.r_netid == NULL ||
464 cbp->cb_callback.cb_location.r_addr == NULL) {
465 goto cb_init_out;
466 }
467
468 if (strcmp(cbp->cb_callback.cb_location.r_netid, "tcp") == 0) {
469 knc.knc_semantics = NC_TPI_COTS;
470 knc.knc_protofmly = "inet";
471 knc.knc_proto = "tcp";
472 devnam = "/dev/tcp";
473 af = AF_INET;
474 } else if (strcmp(cbp->cb_callback.cb_location.r_netid, "udp")
475 == 0) {
476 knc.knc_semantics = NC_TPI_CLTS;
477 knc.knc_protofmly = "inet";
478 knc.knc_proto = "udp";
479 devnam = "/dev/udp";
480 af = AF_INET;
481 } else if (strcmp(cbp->cb_callback.cb_location.r_netid, "tcp6")
482 == 0) {
483 knc.knc_semantics = NC_TPI_COTS;
484 knc.knc_protofmly = "inet6";
485 knc.knc_proto = "tcp";
486 devnam = "/dev/tcp6";
487 af = AF_INET6;
488 } else if (strcmp(cbp->cb_callback.cb_location.r_netid, "udp6")
489 == 0) {
490 knc.knc_semantics = NC_TPI_CLTS;
491 knc.knc_protofmly = "inet6";
492 knc.knc_proto = "udp";
493 devnam = "/dev/udp6";
494 af = AF_INET6;
495 } else {
496 goto cb_init_out;
497 }
498
499 if (lookupname(devnam, UIO_SYSSPACE, FOLLOW, NULLVPP, &vp) != 0) {
500
501 goto cb_init_out;
502 }
503
504 if (vp->v_type != VCHR) {
505 VN_RELE(vp);
506 goto cb_init_out;
507 }
508
509 knc.knc_rdev = vp->v_rdev;
510
511 VN_RELE(vp);
512
513 if (af == AF_INET) {
514 size = sizeof (addr4);
515 bzero(&addr4, size);
516 addr4.sin_family = (sa_family_t)af;
517 addr = &addr4.sin_addr;
518 pp = &addr4.sin_port;
519 taddr = &addr4;
520 } else /* AF_INET6 */ {
521 size = sizeof (addr6);
522 bzero(&addr6, size);
523 addr6.sin6_family = (sa_family_t)af;
524 addr = &addr6.sin6_addr;
525 pp = &addr6.sin6_port;
526 taddr = &addr6;
527 }
528
529 if (uaddr2sockaddr(af,
530 cbp->cb_callback.cb_location.r_addr, addr, pp)) {
531
532 goto cb_init_out;
533 }
534
535
536 nb.maxlen = nb.len = size;
537 nb.buf = (char *)taddr;
538
539 if (clnt_tli_kcreate(&knc, &nb, cbp->cb_callback.cb_program,
540 NFS_CB, 0, 0, curthread->t_cred, &ch)) {
541
542 ch = NULL;
543 }
544
545 /* turn off reserved port usage */
546 (void) CLNT_CONTROL(ch, CLSET_BINDRESVPORT, (char *)&useresvport);
547
548 cb_init_out:
549 mutex_exit(cbp->cb_lock);
550 return (ch);
551 }
552
553 /*
554 * Iterate over the client handle cache and
555 * destroy it.
556 */
557 static void
558 rfs4_cb_chflush(rfs4_cbinfo_t *cbp)
559 {
560 CLIENT *ch;
561
562 while (cbp->cb_chc_free) {
563 cbp->cb_chc_free--;
564 ch = cbp->cb_chc[cbp->cb_chc_free];
565 cbp->cb_chc[cbp->cb_chc_free] = NULL;
566 if (ch) {
567 if (ch->cl_auth)
568 auth_destroy(ch->cl_auth);
569 clnt_destroy(ch);
570 }
571 }
572 }
573
574 /*
575 * Return a client handle, either from a the small
576 * rfs4_client_t cache or one that we just created.
577 */
578 static CLIENT *
579 rfs4_cb_getch(rfs4_cbinfo_t *cbp)
580 {
581 CLIENT *cbch = NULL;
582 uint32_t zilch = 0;
583
584 mutex_enter(cbp->cb_lock);
585
586 if (cbp->cb_chc_free) {
587 cbp->cb_chc_free--;
588 cbch = cbp->cb_chc[ cbp->cb_chc_free ];
589 mutex_exit(cbp->cb_lock);
590 (void) CLNT_CONTROL(cbch, CLSET_XID, (char *)&zilch);
591 return (cbch);
592 }
593
594 mutex_exit(cbp->cb_lock);
595
596 /* none free so make it now */
597 cbch = rfs4_cbch_init(cbp);
598
599 return (cbch);
600 }
601
602 /*
603 * Return the client handle to the small cache or
604 * destroy it.
605 */
606 static void
607 rfs4_cb_freech(rfs4_cbinfo_t *cbp, CLIENT *ch, bool_t lockheld)
608 {
609 if (lockheld == FALSE)
610 mutex_enter(cbp->cb_lock);
611
612 if (cbp->cb_chc_free < RFS4_CBCH_MAX) {
613 cbp->cb_chc[ cbp->cb_chc_free++ ] = ch;
614 if (lockheld == FALSE)
615 mutex_exit(cbp->cb_lock);
616 return;
617 }
618 if (lockheld == FALSE)
619 mutex_exit(cbp->cb_lock);
620
621 /*
622 * cache maxed out of free entries, obliterate
623 * this client handle, destroy it, throw it away.
624 */
625 if (ch->cl_auth)
626 auth_destroy(ch->cl_auth);
627 clnt_destroy(ch);
628 }
629
630 /*
631 * With the supplied callback information - initialize the client
632 * callback data. If there is a callback in progress, save the
633 * callback info so that a thread can pick it up in the future.
634 */
635 void
636 rfs4_client_setcb(rfs4_client_t *cp, cb_client4 *cb, uint32_t cb_ident)
637 {
638 char *addr = NULL;
639 char *netid = NULL;
640 rfs4_cbinfo_t *cbp = &cp->rc_cbinfo;
641 size_t len;
642
643 /* Set the call back for the client */
644 if (cb->cb_location.r_addr && cb->cb_location.r_addr[0] != '\0' &&
645 cb->cb_location.r_netid && cb->cb_location.r_netid[0] != '\0') {
646 len = strlen(cb->cb_location.r_addr) + 1;
647 addr = kmem_alloc(len, KM_SLEEP);
648 bcopy(cb->cb_location.r_addr, addr, len);
649 len = strlen(cb->cb_location.r_netid) + 1;
650 netid = kmem_alloc(len, KM_SLEEP);
651 bcopy(cb->cb_location.r_netid, netid, len);
652 }
653 /* ready to save the new information but first free old, if exists */
654 mutex_enter(cbp->cb_lock);
655
656 cbp->cb_newer.cb_callback.cb_program = cb->cb_program;
657
658 if (cbp->cb_newer.cb_callback.cb_location.r_addr != NULL)
659 kmem_free(cbp->cb_newer.cb_callback.cb_location.r_addr,
660 strlen(cbp->cb_newer.cb_callback.cb_location.r_addr) + 1);
661 cbp->cb_newer.cb_callback.cb_location.r_addr = addr;
662
663 if (cbp->cb_newer.cb_callback.cb_location.r_netid != NULL)
664 kmem_free(cbp->cb_newer.cb_callback.cb_location.r_netid,
665 strlen(cbp->cb_newer.cb_callback.cb_location.r_netid) + 1);
666 cbp->cb_newer.cb_callback.cb_location.r_netid = netid;
667
668 cbp->cb_newer.cb_ident = cb_ident;
669
670 if (addr && *addr && netid && *netid) {
671 cbp->cb_newer.cb_new = TRUE;
672 cbp->cb_newer.cb_confirmed = FALSE;
673 } else {
674 cbp->cb_newer.cb_new = FALSE;
675 cbp->cb_newer.cb_confirmed = FALSE;
676 }
677
678 mutex_exit(cbp->cb_lock);
679 }
680
681 /*
682 * The server uses this when processing SETCLIENTID_CONFIRM. Callback
683 * information may have been provided on SETCLIENTID and this call
684 * marks that information as confirmed and then starts a thread to
685 * test the callback path.
686 */
687 void
688 rfs4_deleg_cb_check(rfs4_client_t *cp)
689 {
690 if (cp->rc_cbinfo.cb_newer.cb_new == FALSE)
691 return;
692
693 cp->rc_cbinfo.cb_newer.cb_confirmed = TRUE;
694
695 rfs4_dbe_hold(cp->rc_dbe); /* hold the client struct for thread */
696
697 (void) zthread_create(NULL, 0, rfs4_do_cb_null, cp, 0,
698 minclsyspri);
699 }
700
701 static void
702 rfs4args_cb_recall_free(nfs_cb_argop4 *argop)
703 {
704 CB_RECALL4args *rec_argp;
705
706 rec_argp = &argop->nfs_cb_argop4_u.opcbrecall;
707 if (rec_argp->fh.nfs_fh4_val)
708 kmem_free(rec_argp->fh.nfs_fh4_val, rec_argp->fh.nfs_fh4_len);
709 }
710
711 /* ARGSUSED */
712 static void
713 rfs4args_cb_getattr_free(nfs_cb_argop4 *argop)
714 {
715 CB_GETATTR4args *argp;
716
717 argp = &argop->nfs_cb_argop4_u.opcbgetattr;
718 if (argp->fh.nfs_fh4_val)
719 kmem_free(argp->fh.nfs_fh4_val, argp->fh.nfs_fh4_len);
720 }
721
722 static void
723 rfs4freeargres(CB_COMPOUND4args *args, CB_COMPOUND4res *resp)
724 {
725 int i, arglen;
726 nfs_cb_argop4 *argop;
727
728 /*
729 * First free any special args alloc'd for specific ops.
730 */
731 arglen = args->array_len;
732 argop = args->array;
733 for (i = 0; i < arglen; i++, argop++) {
734
735 switch (argop->argop) {
736 case OP_CB_RECALL:
737 rfs4args_cb_recall_free(argop);
738 break;
739
740 case OP_CB_GETATTR:
741 rfs4args_cb_getattr_free(argop);
742 break;
743
744 default:
745 return;
746 }
747 }
748
749 if (args->tag.utf8string_len > 0)
750 UTF8STRING_FREE(args->tag)
751
752 kmem_free(args->array, arglen * sizeof (nfs_cb_argop4));
753 if (resp)
754 xdr_free(xdr_CB_COMPOUND4res, (caddr_t)resp);
755 }
756
757 /*
758 * General callback routine for the server to the client.
759 */
760 static enum clnt_stat
761 rfs4_do_callback(rfs4_client_t *cp, CB_COMPOUND4args *args,
762 CB_COMPOUND4res *res, struct timeval timeout)
763 {
764 rfs4_cbinfo_t *cbp;
765 CLIENT *ch;
766 /* start with this in case cb_getch() fails */
767 enum clnt_stat stat = RPC_FAILED;
768
769 res->tag.utf8string_val = NULL;
770 res->array = NULL;
771
772 retry:
773 cbp = rfs4_cbinfo_hold(cp);
774 if (cbp == NULL)
775 return (stat);
776
777 /* get a client handle */
778 if ((ch = rfs4_cb_getch(cbp)) != NULL) {
779 /*
780 * reset the cb_ident since it may have changed in
781 * rfs4_cbinfo_hold()
782 */
783 args->callback_ident = cbp->cb_ident;
784
785 stat = clnt_call(ch, CB_COMPOUND, xdr_CB_COMPOUND4args_srv,
786 (caddr_t)args, xdr_CB_COMPOUND4res,
787 (caddr_t)res, timeout);
788
789 /* free client handle */
790 rfs4_cb_freech(cbp, ch, FALSE);
791 }
792
793 /*
794 * If the rele says that there may be new callback info then
795 * retry this sequence and it may succeed as a result of the
796 * new callback path
797 */
798 if (rfs4_cbinfo_rele(cbp,
799 (stat == RPC_SUCCESS ? CB_NOCHANGE : CB_FAILED)) == TRUE)
800 goto retry;
801
802 return (stat);
803 }
804
805 /*
806 * Used by the NFSv4 server to get attributes for a file while
807 * handling the case where a file has been write delegated. For the
808 * time being, VOP_GETATTR() is called and CB_GETATTR processing is
809 * not undertaken. This call site is maintained in case the server is
810 * updated in the future to handle write delegation space guarantees.
811 */
812 nfsstat4
813 rfs4_vop_getattr(vnode_t *vp, vattr_t *vap, int flag, cred_t *cr)
814 {
815
816 int error;
817
818 error = VOP_GETATTR(vp, vap, flag, cr, NULL);
819 return (puterrno4(error));
820 }
821
822 /*
823 * This is used everywhere in the v2/v3 server to allow the
824 * integration of all NFS versions and the support of delegation. For
825 * now, just call the VOP_GETATTR(). If the NFSv4 server is enhanced
826 * in the future to provide space guarantees for write delegations
827 * then this call site should be expanded to interact with the client.
828 */
829 int
830 rfs4_delegated_getattr(vnode_t *vp, vattr_t *vap, int flag, cred_t *cr)
831 {
832 return (VOP_GETATTR(vp, vap, flag, cr, NULL));
833 }
834
835 /*
836 * Place the actual cb_recall otw call to client.
837 */
838 static void
839 rfs4_do_cb_recall(rfs4_deleg_state_t *dsp, bool_t trunc)
840 {
841 CB_COMPOUND4args cb4_args;
842 CB_COMPOUND4res cb4_res;
843 CB_RECALL4args *rec_argp;
844 CB_RECALL4res *rec_resp;
845 nfs_cb_argop4 *argop;
846 int numops;
847 int argoplist_size;
848 struct timeval timeout;
849 nfs_fh4 *fhp;
850 enum clnt_stat call_stat;
851
852 /*
853 * set up the compound args
854 */
855 numops = 1; /* CB_RECALL only */
856
857 argoplist_size = numops * sizeof (nfs_cb_argop4);
858 argop = kmem_zalloc(argoplist_size, KM_SLEEP);
859 argop->argop = OP_CB_RECALL;
860 rec_argp = &argop->nfs_cb_argop4_u.opcbrecall;
861
862 (void) str_to_utf8("cb_recall", &cb4_args.tag);
863 cb4_args.minorversion = CB4_MINORVERSION;
864 /* cb4_args.callback_ident is set in rfs4_do_callback() */
865 cb4_args.array_len = numops;
866 cb4_args.array = argop;
867
868 /*
869 * fill in the args struct
870 */
871 bcopy(&dsp->rds_delegid.stateid, &rec_argp->stateid, sizeof (stateid4));
872 rec_argp->truncate = trunc;
873
874 fhp = &dsp->rds_finfo->rf_filehandle;
875 rec_argp->fh.nfs_fh4_val = kmem_alloc(sizeof (char) *
876 fhp->nfs_fh4_len, KM_SLEEP);
877 nfs_fh4_copy(fhp, &rec_argp->fh);
878
879 /* Keep track of when we did this for observability */
880 dsp->rds_time_recalled = gethrestime_sec();
881
882 /*
883 * Set up the timeout for the callback and make the actual call.
884 * Timeout will be 80% of the lease period for this server.
885 */
886 timeout.tv_sec = (rfs4_lease_time * 80) / 100;
887 timeout.tv_usec = 0;
888
889 DTRACE_NFSV4_3(cb__recall__start, rfs4_client_t *, dsp->rds_client,
890 rfs4_deleg_state_t *, dsp, CB_RECALL4args *, rec_argp);
891
892 call_stat = rfs4_do_callback(dsp->rds_client, &cb4_args, &cb4_res,
893 timeout);
894
895 rec_resp = (cb4_res.array_len == 0) ? NULL :
896 &cb4_res.array[0].nfs_cb_resop4_u.opcbrecall;
897
898 DTRACE_NFSV4_3(cb__recall__done, rfs4_client_t *, dsp->rds_client,
899 rfs4_deleg_state_t *, dsp, CB_RECALL4res *, rec_resp);
900
901 if (call_stat != RPC_SUCCESS || cb4_res.status != NFS4_OK) {
902 rfs4_return_deleg(dsp, TRUE);
903 }
904
905 rfs4freeargres(&cb4_args, &cb4_res);
906 }
907
908 struct recall_arg {
909 rfs4_deleg_state_t *dsp;
910 void (*recall)(rfs4_deleg_state_t *, bool_t trunc);
911 bool_t trunc;
912 };
913
914 static void
915 do_recall(struct recall_arg *arg)
916 {
917 rfs4_deleg_state_t *dsp = arg->dsp;
918 rfs4_file_t *fp = dsp->rds_finfo;
919 callb_cpr_t cpr_info;
920 kmutex_t cpr_lock;
921
922 mutex_init(&cpr_lock, NULL, MUTEX_DEFAULT, NULL);
923 CALLB_CPR_INIT(&cpr_info, &cpr_lock, callb_generic_cpr, "nfsv4Recall");
924
925 /*
926 * It is possible that before this thread starts
927 * the client has send us a return_delegation, and
928 * if that is the case we do not need to send the
929 * recall callback.
930 */
931 if (dsp->rds_dtype != OPEN_DELEGATE_NONE) {
932 DTRACE_PROBE3(nfss__i__recall,
933 struct recall_arg *, arg,
934 struct rfs4_deleg_state_t *, dsp,
935 struct rfs4_file_t *, fp);
936
937 if (arg->recall)
938 (void) (*arg->recall)(dsp, arg->trunc);
939 }
940
941 mutex_enter(fp->rf_dinfo.rd_recall_lock);
942 /*
943 * Recall count may go negative if the parent thread that is
944 * creating the individual callback threads does not modify
945 * the recall_count field before the callback thread actually
946 * gets a response from the CB_RECALL
947 */
948 fp->rf_dinfo.rd_recall_count--;
949 if (fp->rf_dinfo.rd_recall_count == 0)
950 cv_signal(fp->rf_dinfo.rd_recall_cv);
951 mutex_exit(fp->rf_dinfo.rd_recall_lock);
952
953 mutex_enter(&cpr_lock);
954 CALLB_CPR_EXIT(&cpr_info);
955 mutex_destroy(&cpr_lock);
956
957 rfs4_deleg_state_rele(dsp); /* release the hold for this thread */
958 kmem_free(arg, sizeof (struct recall_arg));
959 zthread_exit();
960 }
961
962 struct master_recall_args {
963 rfs4_file_t *fp;
964 void (*recall)(rfs4_deleg_state_t *, bool_t);
965 bool_t trunc;
966 };
967
968 static void
969 do_recall_file(struct master_recall_args *map)
970 {
971 rfs4_file_t *fp = map->fp;
972 rfs4_deleg_state_t *dsp;
973 struct recall_arg *arg;
974 callb_cpr_t cpr_info;
975 kmutex_t cpr_lock;
976 int32_t recall_count;
977
978 rfs4_dbe_lock(fp->rf_dbe);
979
980 /* Recall already in progress ? */
981 mutex_enter(fp->rf_dinfo.rd_recall_lock);
982 if (fp->rf_dinfo.rd_recall_count != 0) {
983 mutex_exit(fp->rf_dinfo.rd_recall_lock);
984 rfs4_dbe_rele_nolock(fp->rf_dbe);
985 rfs4_dbe_unlock(fp->rf_dbe);
986 kmem_free(map, sizeof (struct master_recall_args));
987 zthread_exit();
988 }
989
990 mutex_exit(fp->rf_dinfo.rd_recall_lock);
991
992 mutex_init(&cpr_lock, NULL, MUTEX_DEFAULT, NULL);
993 CALLB_CPR_INIT(&cpr_info, &cpr_lock, callb_generic_cpr, "v4RecallFile");
994
995 recall_count = 0;
996 for (dsp = list_head(&fp->rf_delegstatelist); dsp != NULL;
997 dsp = list_next(&fp->rf_delegstatelist, dsp)) {
998
999 rfs4_dbe_lock(dsp->rds_dbe);
1000 /*
1001 * if this delegation state
1002 * is being reaped skip it
1003 */
1004 if (rfs4_dbe_is_invalid(dsp->rds_dbe)) {
1005 rfs4_dbe_unlock(dsp->rds_dbe);
1006 continue;
1007 }
1008
1009 /* hold for receiving thread */
1010 rfs4_dbe_hold(dsp->rds_dbe);
1011 rfs4_dbe_unlock(dsp->rds_dbe);
1012
1013 arg = kmem_alloc(sizeof (struct recall_arg), KM_SLEEP);
1014 arg->recall = map->recall;
1015 arg->trunc = map->trunc;
1016 arg->dsp = dsp;
1017
1018 recall_count++;
1019
1020 (void) zthread_create(NULL, 0, do_recall, arg, 0,
1021 minclsyspri);
1022 }
1023
1024 rfs4_dbe_unlock(fp->rf_dbe);
1025
1026 mutex_enter(fp->rf_dinfo.rd_recall_lock);
1027 /*
1028 * Recall count may go negative if the parent thread that is
1029 * creating the individual callback threads does not modify
1030 * the recall_count field before the callback thread actually
1031 * gets a response from the CB_RECALL
1032 */
1033 fp->rf_dinfo.rd_recall_count += recall_count;
1034 while (fp->rf_dinfo.rd_recall_count)
1035 cv_wait(fp->rf_dinfo.rd_recall_cv, fp->rf_dinfo.rd_recall_lock);
1036
1037 mutex_exit(fp->rf_dinfo.rd_recall_lock);
1038
1039 DTRACE_PROBE1(nfss__i__recall_done, rfs4_file_t *, fp);
1040 rfs4_file_rele(fp);
1041 kmem_free(map, sizeof (struct master_recall_args));
1042 mutex_enter(&cpr_lock);
1043 CALLB_CPR_EXIT(&cpr_info);
1044 mutex_destroy(&cpr_lock);
1045 zthread_exit();
1046 }
1047
1048 static void
1049 rfs4_recall_file(rfs4_file_t *fp,
1050 void (*recall)(rfs4_deleg_state_t *, bool_t trunc),
1051 bool_t trunc, rfs4_client_t *cp)
1052 {
1053 struct master_recall_args *args;
1054
1055 rfs4_dbe_lock(fp->rf_dbe);
1056 if (fp->rf_dinfo.rd_dtype == OPEN_DELEGATE_NONE) {
1057 rfs4_dbe_unlock(fp->rf_dbe);
1058 return;
1059 }
1060 rfs4_dbe_hold(fp->rf_dbe); /* hold for new thread */
1061
1062 /*
1063 * Mark the time we started the recall processing.
1064 * If it has been previously recalled, do not reset the
1065 * timer since this is used for the revocation decision.
1066 */
1067 if (fp->rf_dinfo.rd_time_recalled == 0)
1068 fp->rf_dinfo.rd_time_recalled = gethrestime_sec();
1069 fp->rf_dinfo.rd_ever_recalled = TRUE; /* used for policy decision */
1070 /* Client causing recall not always available */
1071 if (cp)
1072 fp->rf_dinfo.rd_conflicted_client = cp->rc_clientid;
1073
1074 rfs4_dbe_unlock(fp->rf_dbe);
1075
1076 args = kmem_alloc(sizeof (struct master_recall_args), KM_SLEEP);
1077 args->fp = fp;
1078 args->recall = recall;
1079 args->trunc = trunc;
1080
1081 (void) zthread_create(NULL, 0, do_recall_file, args, 0,
1082 minclsyspri);
1083 }
1084
1085 void
1086 rfs4_recall_deleg(rfs4_file_t *fp, bool_t trunc, rfs4_client_t *cp)
1087 {
1088 time_t elapsed1, elapsed2;
1089
1090 if (fp->rf_dinfo.rd_time_recalled != 0) {
1091 elapsed1 = gethrestime_sec() - fp->rf_dinfo.rd_time_recalled;
1092 elapsed2 = gethrestime_sec() - fp->rf_dinfo.rd_time_lastwrite;
1093 /* First check to see if a revocation should occur */
1094 if (elapsed1 > rfs4_lease_time &&
1095 elapsed2 > rfs4_lease_time) {
1096 rfs4_revoke_file(fp);
1097 return;
1098 }
1099 /*
1100 * Next check to see if a recall should be done again
1101 * so quickly.
1102 */
1103 if (elapsed1 <= ((rfs4_lease_time * 20) / 100))
1104 return;
1105 }
1106 rfs4_recall_file(fp, rfs4_do_cb_recall, trunc, cp);
1107 }
1108
1109 /*
1110 * rfs4_check_recall is called from rfs4_do_open to determine if the current
1111 * open conflicts with the delegation.
1112 * Return true if we need recall otherwise false.
1113 * Assumes entry locks for sp and sp->rs_finfo are held.
1114 */
1115 bool_t
1116 rfs4_check_recall(rfs4_state_t *sp, uint32_t access)
1117 {
1118 open_delegation_type4 dtype = sp->rs_finfo->rf_dinfo.rd_dtype;
1119
1120 switch (dtype) {
1121 case OPEN_DELEGATE_NONE:
1122 /* Not currently delegated so there is nothing to do */
1123 return (FALSE);
1124 case OPEN_DELEGATE_READ:
1125 /*
1126 * If the access is only asking for READ then there is
1127 * no conflict and nothing to do. If it is asking
1128 * for write, then there will be conflict and the read
1129 * delegation should be recalled.
1130 */
1131 if (access == OPEN4_SHARE_ACCESS_READ)
1132 return (FALSE);
1133 else
1134 return (TRUE);
1135 case OPEN_DELEGATE_WRITE:
1136 /* Check to see if this client has the delegation */
1137 return (rfs4_is_deleg(sp));
1138 }
1139
1140 return (FALSE);
1141 }
1142
1143 /*
1144 * Return the "best" allowable delegation available given the current
1145 * delegation type and the desired access and deny modes on the file.
1146 * At the point that this routine is called we know that the access and
1147 * deny modes are consistent with the file modes.
1148 */
1149 static open_delegation_type4
1150 rfs4_check_delegation(rfs4_state_t *sp, rfs4_file_t *fp)
1151 {
1152 open_delegation_type4 dtype = fp->rf_dinfo.rd_dtype;
1153 uint32_t access = sp->rs_share_access;
1154 uint32_t deny = sp->rs_share_deny;
1155 int readcnt = 0;
1156 int writecnt = 0;
1157
1158 switch (dtype) {
1159 case OPEN_DELEGATE_NONE:
1160 /*
1161 * Determine if more than just this OPEN have the file
1162 * open and if so, no delegation may be provided to
1163 * the client.
1164 */
1165 if (access & OPEN4_SHARE_ACCESS_WRITE)
1166 writecnt++;
1167 if (access & OPEN4_SHARE_ACCESS_READ)
1168 readcnt++;
1169
1170 if (fp->rf_access_read > readcnt ||
1171 fp->rf_access_write > writecnt)
1172 return (OPEN_DELEGATE_NONE);
1173
1174 /*
1175 * If the client is going to write, or if the client
1176 * has exclusive access, return a write delegation.
1177 */
1178 if ((access & OPEN4_SHARE_ACCESS_WRITE) ||
1179 (deny & (OPEN4_SHARE_DENY_READ | OPEN4_SHARE_DENY_WRITE)))
1180 return (OPEN_DELEGATE_WRITE);
1181 /*
1182 * If we don't want to write or we've haven't denied read
1183 * access to others, return a read delegation.
1184 */
1185 if ((access & ~OPEN4_SHARE_ACCESS_WRITE) ||
1186 (deny & ~OPEN4_SHARE_DENY_READ))
1187 return (OPEN_DELEGATE_READ);
1188
1189 /* Shouldn't get here */
1190 return (OPEN_DELEGATE_NONE);
1191
1192 case OPEN_DELEGATE_READ:
1193 /*
1194 * If the file is delegated for read but we wan't to
1195 * write or deny others to read then we can't delegate
1196 * the file. We shouldn't get here since the delegation should
1197 * have been recalled already.
1198 */
1199 if ((access & OPEN4_SHARE_ACCESS_WRITE) ||
1200 (deny & OPEN4_SHARE_DENY_READ))
1201 return (OPEN_DELEGATE_NONE);
1202 return (OPEN_DELEGATE_READ);
1203
1204 case OPEN_DELEGATE_WRITE:
1205 return (OPEN_DELEGATE_WRITE);
1206 }
1207
1208 /* Shouldn't get here */
1209 return (OPEN_DELEGATE_NONE);
1210 }
1211
1212 /*
1213 * Given the desired delegation type and the "history" of the file
1214 * determine the actual delegation type to return.
1215 */
1216 static open_delegation_type4
1217 rfs4_delegation_policy(nfs4_srv_t *nsrv4, open_delegation_type4 dtype,
1218 rfs4_dinfo_t *dinfo, clientid4 cid)
1219 {
1220 time_t elapsed;
1221
1222 if (nsrv4->nfs4_deleg_policy != SRV_NORMAL_DELEGATE)
1223 return (OPEN_DELEGATE_NONE);
1224
1225 /*
1226 * Has this file/delegation ever been recalled? If not then
1227 * no further checks for a delegation race need to be done.
1228 * However if a recall has occurred, then check to see if a
1229 * client has caused its own delegation recall to occur. If
1230 * not, then has a delegation for this file been returned
1231 * recently? If so, then do not assign a new delegation to
1232 * avoid a "delegation race" between the original client and
1233 * the new/conflicting client.
1234 */
1235 if (dinfo->rd_ever_recalled == TRUE) {
1236 if (dinfo->rd_conflicted_client != cid) {
1237 elapsed = gethrestime_sec() - dinfo->rd_time_returned;
1238 if (elapsed < rfs4_lease_time)
1239 return (OPEN_DELEGATE_NONE);
1240 }
1241 }
1242
1243 /* Limit the number of read grants */
1244 if (dtype == OPEN_DELEGATE_READ &&
1245 dinfo->rd_rdgrants > MAX_READ_DELEGATIONS)
1246 return (OPEN_DELEGATE_NONE);
1247
1248 /*
1249 * Should consider limiting total number of read/write
1250 * delegations the server will permit.
1251 */
1252
1253 return (dtype);
1254 }
1255
1256 /*
1257 * Try and grant a delegation for an open give the state. The routine
1258 * returns the delegation type granted. This could be OPEN_DELEGATE_NONE.
1259 *
1260 * The state and associate file entry must be locked
1261 */
1262 rfs4_deleg_state_t *
1263 rfs4_grant_delegation(delegreq_t dreq, rfs4_state_t *sp, int *recall)
1264 {
1265 nfs4_srv_t *nsrv4;
1266 rfs4_file_t *fp = sp->rs_finfo;
1267 open_delegation_type4 dtype;
1268 int no_delegation;
1269
1270 ASSERT(rfs4_dbe_islocked(sp->rs_dbe));
1271 ASSERT(rfs4_dbe_islocked(fp->rf_dbe));
1272
1273 nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
1274
1275 /* Is the server even providing delegations? */
1276 if (nsrv4->nfs4_deleg_policy == SRV_NEVER_DELEGATE || dreq == DELEG_NONE)
1277 return (NULL);
1278
1279 /* Check to see if delegations have been temporarily disabled */
1280 mutex_enter(&nsrv4->deleg_lock);
1281 no_delegation = rfs4_deleg_disabled;
1282 mutex_exit(&nsrv4->deleg_lock);
1283
1284 if (no_delegation)
1285 return (NULL);
1286
1287 /* Don't grant a delegation if a deletion is impending. */
1288 if (fp->rf_dinfo.rd_hold_grant > 0) {
1289 return (NULL);
1290 }
1291
1292 /*
1293 * Don't grant a delegation if there are any lock manager
1294 * (NFSv2/v3) locks for the file. This is a bit of a hack (e.g.,
1295 * if there are only read locks we should be able to grant a
1296 * read-only delegation), but it's good enough for now.
1297 *
1298 * MT safety: the lock manager checks for conflicting delegations
1299 * before processing a lock request. That check will block until
1300 * we are done here. So if the lock manager acquires a lock after
1301 * we decide to grant the delegation, the delegation will get
1302 * immediately recalled (if there's a conflict), so we're safe.
1303 */
1304 if (lm_vp_active(fp->rf_vp)) {
1305 return (NULL);
1306 }
1307
1308 /*
1309 * Based on the type of delegation request passed in, take the
1310 * appropriate action (DELEG_NONE is handled above)
1311 */
1312 switch (dreq) {
1313
1314 case DELEG_READ:
1315 case DELEG_WRITE:
1316 /*
1317 * The server "must" grant the delegation in this case.
1318 * Client is using open previous
1319 */
1320 dtype = (open_delegation_type4)dreq;
1321 *recall = 1;
1322 break;
1323 case DELEG_ANY:
1324 /*
1325 * If a valid callback path does not exist, no delegation may
1326 * be granted.
1327 */
1328 if (sp->rs_owner->ro_client->rc_cbinfo.cb_state != CB_OK)
1329 return (NULL);
1330
1331 /*
1332 * If the original operation which caused time_rm_delayed
1333 * to be set hasn't been retried and completed for one
1334 * full lease period, clear it and allow delegations to
1335 * get granted again.
1336 */
1337 if (fp->rf_dinfo.rd_time_rm_delayed > 0 &&
1338 gethrestime_sec() >
1339 fp->rf_dinfo.rd_time_rm_delayed + rfs4_lease_time)
1340 fp->rf_dinfo.rd_time_rm_delayed = 0;
1341
1342 /*
1343 * If we are waiting for a delegation to be returned then
1344 * don't delegate this file. We do this for correctness as
1345 * well as if the file is being recalled we would likely
1346 * recall this file again.
1347 */
1348
1349 if (fp->rf_dinfo.rd_time_recalled != 0 ||
1350 fp->rf_dinfo.rd_time_rm_delayed != 0)
1351 return (NULL);
1352
1353 /* Get the "best" delegation candidate */
1354 dtype = rfs4_check_delegation(sp, fp);
1355
1356 if (dtype == OPEN_DELEGATE_NONE)
1357 return (NULL);
1358
1359 /*
1360 * Based on policy and the history of the file get the
1361 * actual delegation.
1362 */
1363 dtype = rfs4_delegation_policy(nsrv4, dtype, &fp->rf_dinfo,
1364 sp->rs_owner->ro_client->rc_clientid);
1365
1366 if (dtype == OPEN_DELEGATE_NONE)
1367 return (NULL);
1368 break;
1369 default:
1370 return (NULL);
1371 }
1372
1373 /* set the delegation for the state */
1374 return (rfs4_deleg_state(sp, dtype, recall));
1375 }
1376
1377 void
1378 rfs4_set_deleg_response(rfs4_deleg_state_t *dsp, open_delegation4 *dp,
1379 nfsace4 *ace, int recall)
1380 {
1381 open_write_delegation4 *wp;
1382 open_read_delegation4 *rp;
1383 nfs_space_limit4 *spl;
1384 nfsace4 nace;
1385
1386 /*
1387 * We need to allocate a new copy of the who string.
1388 * this string will be freed by the rfs4_op_open dis_resfree
1389 * routine. We need to do this allocation since replays will
1390 * be allocated and rfs4_compound can't tell the difference from
1391 * a replay and an inital open. N.B. if an ace is passed in, it
1392 * the caller's responsibility to free it.
1393 */
1394
1395 if (ace == NULL) {
1396 /*
1397 * Default is to deny all access, the client will have
1398 * to contact the server. XXX Do we want to actually
1399 * set a deny for every one, or do we simply want to
1400 * construct an entity that will match no one?
1401 */
1402 nace.type = ACE4_ACCESS_DENIED_ACE_TYPE;
1403 nace.flag = 0;
1404 nace.access_mask = ACE4_VALID_MASK_BITS;
1405 (void) str_to_utf8(ACE4_WHO_EVERYONE, &nace.who);
1406 } else {
1407 nace.type = ace->type;
1408 nace.flag = ace->flag;
1409 nace.access_mask = ace->access_mask;
1410 (void) utf8_copy(&ace->who, &nace.who);
1411 }
1412
1413 dp->delegation_type = dsp->rds_dtype;
1414
1415 switch (dsp->rds_dtype) {
1416 case OPEN_DELEGATE_NONE:
1417 break;
1418 case OPEN_DELEGATE_READ:
1419 rp = &dp->open_delegation4_u.read;
1420 rp->stateid = dsp->rds_delegid.stateid;
1421 rp->recall = (bool_t)recall;
1422 rp->permissions = nace;
1423 break;
1424 case OPEN_DELEGATE_WRITE:
1425 wp = &dp->open_delegation4_u.write;
1426 wp->stateid = dsp->rds_delegid.stateid;
1427 wp->recall = (bool_t)recall;
1428 spl = &wp->space_limit;
1429 spl->limitby = NFS_LIMIT_SIZE;
1430 spl->nfs_space_limit4_u.filesize = 0;
1431 wp->permissions = nace;
1432 break;
1433 }
1434 }
1435
1436 /*
1437 * Check if the file is delegated via the provided file struct.
1438 * Return TRUE if it is delegated. This is intended for use by
1439 * the v4 server. The v2/v3 server code should use rfs4_check_delegated().
1440 *
1441 * Note that if the file is found to have a delegation, it is
1442 * recalled, unless the clientid of the caller matches the clientid of the
1443 * delegation. If the caller has specified, there is a slight delay
1444 * inserted in the hopes that the delegation will be returned quickly.
1445 */
1446 bool_t
1447 rfs4_check_delegated_byfp(int mode, rfs4_file_t *fp,
1448 bool_t trunc, bool_t do_delay, bool_t is_rm, clientid4 *cp)
1449 {
1450 rfs4_deleg_state_t *dsp;
1451
1452 nfs4_srv_t *nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
1453
1454 /* Is delegation enabled? */
1455 if (nsrv4->nfs4_deleg_policy == SRV_NEVER_DELEGATE)
1456 return (FALSE);
1457
1458 /* do we have a delegation on this file? */
1459 rfs4_dbe_lock(fp->rf_dbe);
1460 if (fp->rf_dinfo.rd_dtype == OPEN_DELEGATE_NONE) {
1461 if (is_rm)
1462 fp->rf_dinfo.rd_hold_grant++;
1463 rfs4_dbe_unlock(fp->rf_dbe);
1464 return (FALSE);
1465 }
1466 /*
1467 * do we have a write delegation on this file or are we
1468 * requesting write access to a file with any type of existing
1469 * delegation?
1470 */
1471 if (mode == FWRITE || fp->rf_dinfo.rd_dtype == OPEN_DELEGATE_WRITE) {
1472 if (cp != NULL) {
1473 dsp = list_head(&fp->rf_delegstatelist);
1474 if (dsp == NULL) {
1475 rfs4_dbe_unlock(fp->rf_dbe);
1476 return (FALSE);
1477 }
1478 /*
1479 * Does the requestor already own the delegation?
1480 */
1481 if (dsp->rds_client->rc_clientid == *(cp)) {
1482 rfs4_dbe_unlock(fp->rf_dbe);
1483 return (FALSE);
1484 }
1485 }
1486
1487 rfs4_dbe_unlock(fp->rf_dbe);
1488 rfs4_recall_deleg(fp, trunc, NULL);
1489
1490 if (!do_delay) {
1491 rfs4_dbe_lock(fp->rf_dbe);
1492 fp->rf_dinfo.rd_time_rm_delayed = gethrestime_sec();
1493 rfs4_dbe_unlock(fp->rf_dbe);
1494 return (TRUE);
1495 }
1496
1497 delay(NFS4_DELEGATION_CONFLICT_DELAY);
1498
1499 rfs4_dbe_lock(fp->rf_dbe);
1500 if (fp->rf_dinfo.rd_dtype != OPEN_DELEGATE_NONE) {
1501 fp->rf_dinfo.rd_time_rm_delayed = gethrestime_sec();
1502 rfs4_dbe_unlock(fp->rf_dbe);
1503 return (TRUE);
1504 }
1505 }
1506 if (is_rm)
1507 fp->rf_dinfo.rd_hold_grant++;
1508 rfs4_dbe_unlock(fp->rf_dbe);
1509 return (FALSE);
1510 }
1511
1512 /*
1513 * Check if the file is delegated in the case of a v2 or v3 access.
1514 * Return TRUE if it is delegated which in turn means that v2 should
1515 * drop the request and in the case of v3 JUKEBOX should be returned.
1516 */
1517 bool_t
1518 rfs4_check_delegated(int mode, vnode_t *vp, bool_t trunc)
1519 {
1520 nfs4_srv_t *nsrv4;
1521 rfs4_file_t *fp;
1522 bool_t create = FALSE;
1523 bool_t rc = FALSE;
1524
1525 nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
1526 rfs4_hold_deleg_policy(nsrv4);
1527
1528 /* Is delegation enabled? */
1529 if (nsrv4->nfs4_deleg_policy != SRV_NEVER_DELEGATE) {
1530 fp = rfs4_findfile(vp, NULL, &create);
1531 if (fp != NULL) {
1532 if (rfs4_check_delegated_byfp(mode, fp, trunc,
1533 TRUE, FALSE, NULL)) {
1534 rc = TRUE;
1535 }
1536 rfs4_file_rele(fp);
1537 }
1538 }
1539 rfs4_rele_deleg_policy(nsrv4);
1540 return (rc);
1541 }
1542
1543 /*
1544 * Release a hold on the hold_grant counter which
1545 * prevents delegation from being granted while a remove
1546 * or a rename is in progress.
1547 */
1548 void
1549 rfs4_clear_dont_grant(rfs4_file_t *fp)
1550 {
1551 nfs4_srv_t *nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
1552
1553 if (nsrv4->nfs4_deleg_policy == SRV_NEVER_DELEGATE)
1554 return;
1555 rfs4_dbe_lock(fp->rf_dbe);
1556 ASSERT(fp->rf_dinfo.rd_hold_grant > 0);
1557 fp->rf_dinfo.rd_hold_grant--;
1558 fp->rf_dinfo.rd_time_rm_delayed = 0;
1559 rfs4_dbe_unlock(fp->rf_dbe);
1560 }
1561
1562 /*
1563 * State support for delegation.
1564 * Set the state delegation type for this state;
1565 * This routine is called from open via rfs4_grant_delegation and the entry
1566 * locks on sp and sp->rs_finfo are assumed.
1567 */
1568 static rfs4_deleg_state_t *
1569 rfs4_deleg_state(rfs4_state_t *sp, open_delegation_type4 dtype, int *recall)
1570 {
1571 rfs4_file_t *fp = sp->rs_finfo;
1572 bool_t create = TRUE;
1573 rfs4_deleg_state_t *dsp;
1574 vnode_t *vp;
1575 int open_prev = *recall;
1576 int ret;
1577 int fflags = 0;
1578
1579 ASSERT(rfs4_dbe_islocked(sp->rs_dbe));
1580 ASSERT(rfs4_dbe_islocked(fp->rf_dbe));
1581
1582 /* Shouldn't happen */
1583 if (fp->rf_dinfo.rd_recall_count != 0 ||
1584 (fp->rf_dinfo.rd_dtype == OPEN_DELEGATE_READ &&
1585 dtype != OPEN_DELEGATE_READ)) {
1586 return (NULL);
1587 }
1588
1589 /* Unlock to avoid deadlock */
1590 rfs4_dbe_unlock(fp->rf_dbe);
1591 rfs4_dbe_unlock(sp->rs_dbe);
1592
1593 dsp = rfs4_finddeleg(sp, &create);
1594
1595 rfs4_dbe_lock(sp->rs_dbe);
1596 rfs4_dbe_lock(fp->rf_dbe);
1597
1598 if (dsp == NULL)
1599 return (NULL);
1600
1601 /*
1602 * It is possible that since we dropped the lock
1603 * in order to call finddeleg, the rfs4_file_t
1604 * was marked such that we should not grant a
1605 * delegation, if so bail out.
1606 */
1607 if (fp->rf_dinfo.rd_hold_grant > 0) {
1608 rfs4_deleg_state_rele(dsp);
1609 return (NULL);
1610 }
1611
1612 if (create == FALSE) {
1613 if (sp->rs_owner->ro_client == dsp->rds_client &&
1614 dsp->rds_dtype == dtype) {
1615 return (dsp);
1616 } else {
1617 rfs4_deleg_state_rele(dsp);
1618 return (NULL);
1619 }
1620 }
1621
1622 /*
1623 * Check that this file has not been delegated to another
1624 * client
1625 */
1626 if (fp->rf_dinfo.rd_recall_count != 0 ||
1627 fp->rf_dinfo.rd_dtype == OPEN_DELEGATE_WRITE ||
1628 (fp->rf_dinfo.rd_dtype == OPEN_DELEGATE_READ &&
1629 dtype != OPEN_DELEGATE_READ)) {
1630 rfs4_deleg_state_rele(dsp);
1631 return (NULL);
1632 }
1633
1634 vp = fp->rf_vp;
1635 /* vnevent_support returns 0 if file system supports vnevents */
1636 if (vnevent_support(vp, NULL)) {
1637 rfs4_deleg_state_rele(dsp);
1638 return (NULL);
1639 }
1640
1641 /* Calculate the fflags for this OPEN. */
1642 if (sp->rs_share_access & OPEN4_SHARE_ACCESS_READ)
1643 fflags |= FREAD;
1644 if (sp->rs_share_access & OPEN4_SHARE_ACCESS_WRITE)
1645 fflags |= FWRITE;
1646
1647 *recall = 0;
1648 /*
1649 * Before granting a delegation we need to know if anyone else has
1650 * opened the file in a conflicting mode. However, first we need to
1651 * know how we opened the file to check the counts properly.
1652 */
1653 if (dtype == OPEN_DELEGATE_READ) {
1654 if (((fflags & FWRITE) && vn_has_other_opens(vp, V_WRITE)) ||
1655 (((fflags & FWRITE) == 0) && vn_is_opened(vp, V_WRITE)) ||
1656 vn_is_mapped(vp, V_WRITE)) {
1657 if (open_prev) {
1658 *recall = 1;
1659 } else {
1660 rfs4_deleg_state_rele(dsp);
1661 return (NULL);
1662 }
1663 }
1664 ret = fem_install(vp, deleg_rdops, (void *)fp, OPUNIQ,
1665 rfs4_mon_hold, rfs4_mon_rele);
1666 if (((fflags & FWRITE) && vn_has_other_opens(vp, V_WRITE)) ||
1667 (((fflags & FWRITE) == 0) && vn_is_opened(vp, V_WRITE)) ||
1668 vn_is_mapped(vp, V_WRITE)) {
1669 if (open_prev) {
1670 *recall = 1;
1671 } else {
1672 (void) fem_uninstall(vp, deleg_rdops,
1673 (void *)fp);
1674 rfs4_deleg_state_rele(dsp);
1675 return (NULL);
1676 }
1677 }
1678 /*
1679 * Because a client can hold onto a delegation after the
1680 * file has been closed, we need to keep track of the
1681 * access to this file. Otherwise the CIFS server would
1682 * not know about the client accessing the file and could
1683 * inappropriately grant an OPLOCK.
1684 * fem_install() returns EBUSY when asked to install a
1685 * OPUNIQ monitor more than once. Therefore, check the
1686 * return code because we only want this done once.
1687 */
1688 if (ret == 0)
1689 vn_open_upgrade(vp, FREAD);
1690 } else { /* WRITE */
1691 if (((fflags & FWRITE) && vn_has_other_opens(vp, V_WRITE)) ||
1692 (((fflags & FWRITE) == 0) && vn_is_opened(vp, V_WRITE)) ||
1693 ((fflags & FREAD) && vn_has_other_opens(vp, V_READ)) ||
1694 (((fflags & FREAD) == 0) && vn_is_opened(vp, V_READ)) ||
1695 vn_is_mapped(vp, V_RDORWR)) {
1696 if (open_prev) {
1697 *recall = 1;
1698 } else {
1699 rfs4_deleg_state_rele(dsp);
1700 return (NULL);
1701 }
1702 }
1703 ret = fem_install(vp, deleg_wrops, (void *)fp, OPUNIQ,
1704 rfs4_mon_hold, rfs4_mon_rele);
1705 if (((fflags & FWRITE) && vn_has_other_opens(vp, V_WRITE)) ||
1706 (((fflags & FWRITE) == 0) && vn_is_opened(vp, V_WRITE)) ||
1707 ((fflags & FREAD) && vn_has_other_opens(vp, V_READ)) ||
1708 (((fflags & FREAD) == 0) && vn_is_opened(vp, V_READ)) ||
1709 vn_is_mapped(vp, V_RDORWR)) {
1710 if (open_prev) {
1711 *recall = 1;
1712 } else {
1713 (void) fem_uninstall(vp, deleg_wrops,
1714 (void *)fp);
1715 rfs4_deleg_state_rele(dsp);
1716 return (NULL);
1717 }
1718 }
1719 /*
1720 * Because a client can hold onto a delegation after the
1721 * file has been closed, we need to keep track of the
1722 * access to this file. Otherwise the CIFS server would
1723 * not know about the client accessing the file and could
1724 * inappropriately grant an OPLOCK.
1725 * fem_install() returns EBUSY when asked to install a
1726 * OPUNIQ monitor more than once. Therefore, check the
1727 * return code because we only want this done once.
1728 */
1729 if (ret == 0)
1730 vn_open_upgrade(vp, FREAD|FWRITE);
1731 }
1732 /* Place on delegation list for file */
1733 ASSERT(!list_link_active(&dsp->rds_node));
1734 list_insert_tail(&fp->rf_delegstatelist, dsp);
1735
1736 dsp->rds_dtype = fp->rf_dinfo.rd_dtype = dtype;
1737
1738 /* Update delegation stats for this file */
1739 fp->rf_dinfo.rd_time_lastgrant = gethrestime_sec();
1740
1741 /* reset since this is a new delegation */
1742 fp->rf_dinfo.rd_conflicted_client = 0;
1743 fp->rf_dinfo.rd_ever_recalled = FALSE;
1744
1745 if (dtype == OPEN_DELEGATE_READ)
1746 fp->rf_dinfo.rd_rdgrants++;
1747 else
1748 fp->rf_dinfo.rd_wrgrants++;
1749
1750 return (dsp);
1751 }
1752
1753 /*
1754 * State routine for the server when a delegation is returned.
1755 */
1756 void
1757 rfs4_return_deleg(rfs4_deleg_state_t *dsp, bool_t revoked)
1758 {
1759 rfs4_file_t *fp = dsp->rds_finfo;
1760 open_delegation_type4 dtypewas;
1761
1762 rfs4_dbe_lock(fp->rf_dbe);
1763
1764 /* nothing to do if no longer on list */
1765 if (!list_link_active(&dsp->rds_node)) {
1766 rfs4_dbe_unlock(fp->rf_dbe);
1767 return;
1768 }
1769
1770 /* Remove state from recall list */
1771 list_remove(&fp->rf_delegstatelist, dsp);
1772
1773 if (list_is_empty(&fp->rf_delegstatelist)) {
1774 dtypewas = fp->rf_dinfo.rd_dtype;
1775 fp->rf_dinfo.rd_dtype = OPEN_DELEGATE_NONE;
1776 rfs4_dbe_cv_broadcast(fp->rf_dbe);
1777
1778 /* if file system was unshared, the vp will be NULL */
1779 if (fp->rf_vp != NULL) {
1780 /*
1781 * Once a delegation is no longer held by any client,
1782 * the monitor is uninstalled. At this point, the
1783 * client must send OPEN otw, so we don't need the
1784 * reference on the vnode anymore. The open
1785 * downgrade removes the reference put on earlier.
1786 */
1787 if (dtypewas == OPEN_DELEGATE_READ) {
1788 (void) fem_uninstall(fp->rf_vp, deleg_rdops,
1789 (void *)fp);
1790 vn_open_downgrade(fp->rf_vp, FREAD);
1791 } else if (dtypewas == OPEN_DELEGATE_WRITE) {
1792 (void) fem_uninstall(fp->rf_vp, deleg_wrops,
1793 (void *)fp);
1794 vn_open_downgrade(fp->rf_vp, FREAD|FWRITE);
1795 }
1796 }
1797 }
1798
1799 switch (dsp->rds_dtype) {
1800 case OPEN_DELEGATE_READ:
1801 fp->rf_dinfo.rd_rdgrants--;
1802 break;
1803 case OPEN_DELEGATE_WRITE:
1804 fp->rf_dinfo.rd_wrgrants--;
1805 break;
1806 default:
1807 break;
1808 }
1809
1810 /* used in the policy decision */
1811 fp->rf_dinfo.rd_time_returned = gethrestime_sec();
1812
1813 /*
1814 * reset the time_recalled field so future delegations are not
1815 * accidentally revoked
1816 */
1817 if ((fp->rf_dinfo.rd_rdgrants + fp->rf_dinfo.rd_wrgrants) == 0)
1818 fp->rf_dinfo.rd_time_recalled = 0;
1819
1820 rfs4_dbe_unlock(fp->rf_dbe);
1821
1822 rfs4_dbe_lock(dsp->rds_dbe);
1823
1824 dsp->rds_dtype = OPEN_DELEGATE_NONE;
1825
1826 if (revoked == TRUE)
1827 dsp->rds_time_revoked = gethrestime_sec();
1828
1829 rfs4_dbe_invalidate(dsp->rds_dbe);
1830
1831 rfs4_dbe_unlock(dsp->rds_dbe);
1832
1833 if (revoked == TRUE) {
1834 rfs4_dbe_lock(dsp->rds_client->rc_dbe);
1835 dsp->rds_client->rc_deleg_revoked++; /* observability */
1836 rfs4_dbe_unlock(dsp->rds_client->rc_dbe);
1837 }
1838 }
1839
1840 static void
1841 rfs4_revoke_file(rfs4_file_t *fp)
1842 {
1843 rfs4_deleg_state_t *dsp;
1844
1845 /*
1846 * The lock for rfs4_file_t must be held when traversing the
1847 * delegation list but that lock needs to be released to call
1848 * rfs4_return_deleg()
1849 */
1850 rfs4_dbe_lock(fp->rf_dbe);
1851 while (dsp = list_head(&fp->rf_delegstatelist)) {
1852 rfs4_dbe_hold(dsp->rds_dbe);
1853 rfs4_dbe_unlock(fp->rf_dbe);
1854 rfs4_return_deleg(dsp, TRUE);
1855 rfs4_deleg_state_rele(dsp);
1856 rfs4_dbe_lock(fp->rf_dbe);
1857 }
1858 rfs4_dbe_unlock(fp->rf_dbe);
1859 }
1860
1861 /*
1862 * A delegation is assumed to be present on the file associated with
1863 * "sp". Check to see if the delegation matches is associated with
1864 * the same client as referenced by "sp". If it is not, TRUE is
1865 * returned. If the delegation DOES match the client (or no
1866 * delegation is present), return FALSE.
1867 * Assume the state entry and file entry are locked.
1868 */
1869 bool_t
1870 rfs4_is_deleg(rfs4_state_t *sp)
1871 {
1872 rfs4_deleg_state_t *dsp;
1873 rfs4_file_t *fp = sp->rs_finfo;
1874 rfs4_client_t *cp = sp->rs_owner->ro_client;
1875
1876 ASSERT(rfs4_dbe_islocked(fp->rf_dbe));
1877 for (dsp = list_head(&fp->rf_delegstatelist); dsp != NULL;
1878 dsp = list_next(&fp->rf_delegstatelist, dsp)) {
1879 if (cp != dsp->rds_client) {
1880 return (TRUE);
1881 }
1882 }
1883 return (FALSE);
1884 }
1885
1886 void
1887 rfs4_disable_delegation(void)
1888 {
1889 nfs4_srv_t *nsrv4;
1890
1891 nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
1892 mutex_enter(&nsrv4->deleg_lock);
1893 rfs4_deleg_disabled++;
1894 mutex_exit(&nsrv4->deleg_lock);
1895 }
1896
1897 void
1898 rfs4_enable_delegation(void)
1899 {
1900 nfs4_srv_t *nsrv4;
1901
1902 nsrv4 = zone_getspecific(rfs4_zone_key, curzone);
1903 mutex_enter(&nsrv4->deleg_lock);
1904 ASSERT(rfs4_deleg_disabled > 0);
1905 rfs4_deleg_disabled--;
1906 mutex_exit(&nsrv4->deleg_lock);
1907 }
1908
1909 void
1910 rfs4_mon_hold(void *arg)
1911 {
1912 rfs4_file_t *fp = arg;
1913
1914 rfs4_dbe_hold(fp->rf_dbe);
1915 }
1916
1917 void
1918 rfs4_mon_rele(void *arg)
1919 {
1920 rfs4_file_t *fp = arg;
1921
1922 rfs4_dbe_rele_nolock(fp->rf_dbe);
1923 }