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 2016 Nexenta Systems, Inc. All rights reserved.
24 * Copyright (c) 1995, 2010, Oracle and/or its affiliates. All rights reserved.
25 * Copyright (c) 2015 by Delphix. All rights reserved.
26 * Copyright (c) 2015 Joyent, Inc. All rights reserved.
27 */
28
29 #include <sys/param.h>
30 #include <sys/errno.h>
31 #include <sys/vfs.h>
32 #include <sys/vnode.h>
33 #include <sys/cred.h>
34 #include <sys/cmn_err.h>
35 #include <sys/systm.h>
36 #include <sys/kmem.h>
37 #include <sys/pathname.h>
38 #include <sys/utsname.h>
39 #include <sys/debug.h>
40 #include <sys/door.h>
41 #include <sys/sdt.h>
42 #include <sys/thread.h>
43 #include <sys/avl.h>
44
45 #include <rpc/types.h>
46 #include <rpc/auth.h>
47 #include <rpc/clnt.h>
48
49 #include <nfs/nfs.h>
50 #include <nfs/export.h>
51 #include <nfs/nfs_clnt.h>
52 #include <nfs/auth.h>
53
54 static struct kmem_cache *exi_cache_handle;
55 static void exi_cache_reclaim(void *);
56 static void exi_cache_trim(struct exportinfo *exi);
57
58 extern pri_t minclsyspri;
59
60 volatile uint_t nfsauth_cache_hit;
61 volatile uint_t nfsauth_cache_miss;
62 volatile uint_t nfsauth_cache_refresh;
63 volatile uint_t nfsauth_cache_reclaim;
64 volatile uint_t exi_cache_auth_reclaim_failed;
65 volatile uint_t exi_cache_clnt_reclaim_failed;
66
67 /*
68 * The lifetime of an auth cache entry:
69 * ------------------------------------
70 *
71 * An auth cache entry is created with both the auth_time
72 * and auth_freshness times set to the current time.
73 *
74 * Upon every client access which results in a hit, the
75 * auth_time will be updated.
76 *
77 * If a client access determines that the auth_freshness
78 * indicates that the entry is STALE, then it will be
79 * refreshed. Note that this will explicitly reset
103
104 /*
105 * While we could encapuslate the exi_list inside the
106 * exi structure, we can't do that for the auth_list.
107 * So, to keep things looking clean, we keep them both
108 * in these external lists.
109 */
110 typedef struct refreshq_exi_node {
111 struct exportinfo *ren_exi;
112 list_t ren_authlist;
113 list_node_t ren_node;
114 } refreshq_exi_node_t;
115
116 typedef struct refreshq_auth_node {
117 struct auth_cache *ran_auth;
118 char *ran_netid;
119 list_node_t ran_node;
120 } refreshq_auth_node_t;
121
122 /*
123 * Used to manipulate things on the refreshq_queue.
124 * Note that the refresh thread will effectively
125 * pop a node off of the queue, at which point it
126 * will no longer need to hold the mutex.
127 */
128 static kmutex_t refreshq_lock;
129 static list_t refreshq_queue;
130 static kcondvar_t refreshq_cv;
131
132 /*
133 * If there is ever a problem with loading the
134 * module, then nfsauth_fini() needs to be called
135 * to remove state. In that event, since the
136 * refreshq thread has been started, they need to
137 * work together to get rid of state.
138 */
139 typedef enum nfsauth_refreshq_thread_state {
140 REFRESHQ_THREAD_RUNNING,
141 REFRESHQ_THREAD_FINI_REQ,
142 REFRESHQ_THREAD_HALTED
143 } nfsauth_refreshq_thread_state_t;
144
145 nfsauth_refreshq_thread_state_t
146 refreshq_thread_state = REFRESHQ_THREAD_HALTED;
147
148 static void nfsauth_free_node(struct auth_cache *);
149 static void nfsauth_refresh_thread(void);
150
151 static int nfsauth_cache_compar(const void *, const void *);
152
153 /*
154 * mountd is a server-side only daemon. This will need to be
155 * revisited if the NFS server is ever made zones-aware.
156 */
157 kmutex_t mountd_lock;
158 door_handle_t mountd_dh;
159
160 void
161 mountd_args(uint_t did)
162 {
163 mutex_enter(&mountd_lock);
164 if (mountd_dh != NULL)
165 door_ki_rele(mountd_dh);
166 mountd_dh = door_ki_lookup(did);
167 mutex_exit(&mountd_lock);
168 }
169
170 void
171 nfsauth_init(void)
172 {
173 /*
174 * mountd can be restarted by smf(5). We need to make sure
175 * the updated door handle will safely make it to mountd_dh
176 */
177 mutex_init(&mountd_lock, NULL, MUTEX_DEFAULT, NULL);
178
179 mutex_init(&refreshq_lock, NULL, MUTEX_DEFAULT, NULL);
180 list_create(&refreshq_queue, sizeof (refreshq_exi_node_t),
181 offsetof(refreshq_exi_node_t, ren_node));
182
183 cv_init(&refreshq_cv, NULL, CV_DEFAULT, NULL);
184
185 /*
186 * Allocate nfsauth cache handle
187 */
188 exi_cache_handle = kmem_cache_create("exi_cache_handle",
189 sizeof (struct auth_cache), 0, NULL, NULL,
190 exi_cache_reclaim, NULL, NULL, 0);
191
192 refreshq_thread_state = REFRESHQ_THREAD_RUNNING;
193 (void) zthread_create(NULL, 0, nfsauth_refresh_thread,
194 NULL, 0, minclsyspri);
195 }
196
197 /*
198 * Finalization routine for nfsauth. It is important to call this routine
199 * before destroying the exported_lock.
200 */
201 void
202 nfsauth_fini(void)
203 {
204 refreshq_exi_node_t *ren;
205
206 /*
207 * Prevent the nfsauth_refresh_thread from getting new
208 * work.
209 */
210 mutex_enter(&refreshq_lock);
211 if (refreshq_thread_state != REFRESHQ_THREAD_HALTED) {
212 refreshq_thread_state = REFRESHQ_THREAD_FINI_REQ;
213 cv_broadcast(&refreshq_cv);
214
215 /*
216 * Also, wait for nfsauth_refresh_thread() to exit.
217 */
218 while (refreshq_thread_state != REFRESHQ_THREAD_HALTED) {
219 cv_wait(&refreshq_cv, &refreshq_lock);
220 }
221 }
222 mutex_exit(&refreshq_lock);
223
224 /*
225 * Walk the exi_list and in turn, walk the auth_lists and free all
226 * lists. In addition, free INVALID auth_cache entries.
227 */
228 while ((ren = list_remove_head(&refreshq_queue))) {
229 refreshq_auth_node_t *ran;
230
231 while ((ran = list_remove_head(&ren->ren_authlist)) != NULL) {
232 struct auth_cache *p = ran->ran_auth;
233 if (p->auth_state == NFS_AUTH_INVALID)
234 nfsauth_free_node(p);
235 strfree(ran->ran_netid);
236 kmem_free(ran, sizeof (refreshq_auth_node_t));
237 }
238
239 list_destroy(&ren->ren_authlist);
240 exi_rele(ren->ren_exi);
241 kmem_free(ren, sizeof (refreshq_exi_node_t));
242 }
243 list_destroy(&refreshq_queue);
244
245 cv_destroy(&refreshq_cv);
246 mutex_destroy(&refreshq_lock);
247
248 mutex_destroy(&mountd_lock);
249
250 /*
251 * Deallocate nfsauth cache handle
252 */
253 kmem_cache_destroy(exi_cache_handle);
254 }
255
256 /*
257 * Convert the address in a netbuf to
258 * a hash index for the auth_cache table.
259 */
260 static int
261 hash(struct netbuf *a)
262 {
263 int i, h = 0;
264
265 for (i = 0; i < a->len; i++)
266 h ^= a->buf[i];
267
268 return (h & (AUTH_TABLESIZE - 1));
269 }
270
271 /*
272 * Mask out the components of an
273 * address that do not identify
325 static void
326 sys_log(const char *msg)
327 {
328 static time_t tstamp = 0;
329 time_t now;
330
331 /*
332 * msg is shown (at most) once per minute
333 */
334 now = gethrestime_sec();
335 if ((tstamp + 60) < now) {
336 tstamp = now;
337 cmn_err(CE_WARN, msg);
338 }
339 }
340
341 /*
342 * Callup to the mountd to get access information in the kernel.
343 */
344 static bool_t
345 nfsauth_retrieve(struct exportinfo *exi, char *req_netid, int flavor,
346 struct netbuf *addr, int *access, cred_t *clnt_cred, uid_t *srv_uid,
347 gid_t *srv_gid, uint_t *srv_gids_cnt, gid_t **srv_gids)
348 {
349 varg_t varg = {0};
350 nfsauth_res_t res = {0};
351 XDR xdrs;
352 size_t absz;
353 caddr_t abuf;
354 int last = 0;
355 door_arg_t da;
356 door_info_t di;
357 door_handle_t dh;
358 uint_t ntries = 0;
359
360 /*
361 * No entry in the cache for this client/flavor
362 * so we need to call the nfsauth service in the
363 * mount daemon.
364 */
365
366 varg.vers = V_PROTO;
367 varg.arg_u.arg.cmd = NFSAUTH_ACCESS;
400 XDR_DESTROY(&xdrs);
401
402 /*
403 * Prepare the door arguments
404 *
405 * We don't know the size of the message the daemon
406 * will pass back to us. By setting rbuf to NULL,
407 * we force the door code to allocate a buf of the
408 * appropriate size. We must set rsize > 0, however,
409 * else the door code acts as if no response was
410 * expected and doesn't pass the data to us.
411 */
412 da.data_ptr = (char *)abuf;
413 da.data_size = absz;
414 da.desc_ptr = NULL;
415 da.desc_num = 0;
416 da.rbuf = NULL;
417 da.rsize = 1;
418
419 retry:
420 mutex_enter(&mountd_lock);
421 dh = mountd_dh;
422 if (dh != NULL)
423 door_ki_hold(dh);
424 mutex_exit(&mountd_lock);
425
426 if (dh == NULL) {
427 /*
428 * The rendezvous point has not been established yet!
429 * This could mean that either mountd(1m) has not yet
430 * been started or that _this_ routine nuked the door
431 * handle after receiving an EINTR for a REVOKED door.
432 *
433 * Returning NFSAUTH_DROP will cause the NFS client
434 * to retransmit the request, so let's try to be more
435 * rescillient and attempt for ntries before we bail.
436 */
437 if (++ntries % NFSAUTH_DR_TRYCNT) {
438 delay(hz);
439 goto retry;
440 }
441
442 kmem_free(abuf, absz);
443
444 sys_log("nfsauth: mountd has not established door");
474 /*
475 * Server out of resources; back off for a bit
476 */
477 door_ki_rele(dh);
478 delay(hz);
479 goto retry;
480 /* NOTREACHED */
481
482 case EINTR:
483 if (!door_ki_info(dh, &di)) {
484 door_ki_rele(dh);
485
486 if (di.di_attributes & DOOR_REVOKED) {
487 /*
488 * The server barfed and revoked
489 * the (existing) door on us; we
490 * want to wait to give smf(5) a
491 * chance to restart mountd(1m)
492 * and establish a new door handle.
493 */
494 mutex_enter(&mountd_lock);
495 if (dh == mountd_dh) {
496 door_ki_rele(mountd_dh);
497 mountd_dh = NULL;
498 }
499 mutex_exit(&mountd_lock);
500 delay(hz);
501 goto retry;
502 }
503 /*
504 * If the door was _not_ revoked on us,
505 * then more than likely we took an INTR,
506 * so we need to fail the operation.
507 */
508 goto fail;
509 }
510 /*
511 * The only failure that can occur from getting
512 * the door info is EINVAL, so we let the code
513 * below handle it.
514 */
515 /* FALLTHROUGH */
516
517 case EBADF:
518 case EINVAL:
519 default:
576
577 case NFSAUTH_DR_EFAIL:
578 case NFSAUTH_DR_DECERR:
579 case NFSAUTH_DR_BADCMD:
580 default:
581 xdr_free(xdr_nfsauth_res, (char *)&res);
582 fail:
583 *access = NFSAUTH_DENIED;
584 kmem_free(abuf, absz);
585 return (FALSE);
586 /* NOTREACHED */
587 }
588
589 xdr_free(xdr_nfsauth_res, (char *)&res);
590 kmem_free(abuf, absz);
591
592 return (TRUE);
593 }
594
595 static void
596 nfsauth_refresh_thread(void)
597 {
598 refreshq_exi_node_t *ren;
599 refreshq_auth_node_t *ran;
600
601 struct exportinfo *exi;
602
603 int access;
604 bool_t retrieval;
605
606 callb_cpr_t cprinfo;
607
608 CALLB_CPR_INIT(&cprinfo, &refreshq_lock, callb_generic_cpr,
609 "nfsauth_refresh");
610
611 for (;;) {
612 mutex_enter(&refreshq_lock);
613 if (refreshq_thread_state != REFRESHQ_THREAD_RUNNING) {
614 /* Keep the hold on the lock! */
615 break;
616 }
617
618 ren = list_remove_head(&refreshq_queue);
619 if (ren == NULL) {
620 CALLB_CPR_SAFE_BEGIN(&cprinfo);
621 cv_wait(&refreshq_cv, &refreshq_lock);
622 CALLB_CPR_SAFE_END(&cprinfo, &refreshq_lock);
623 mutex_exit(&refreshq_lock);
624 continue;
625 }
626 mutex_exit(&refreshq_lock);
627
628 exi = ren->ren_exi;
629 ASSERT(exi != NULL);
630
631 /*
632 * Since the ren was removed from the refreshq_queue above,
633 * this is the only thread aware about the ren existence, so we
634 * have the exclusive ownership of it and we do not need to
635 * protect it by any lock.
636 */
637 while ((ran = list_remove_head(&ren->ren_authlist))) {
638 uid_t uid;
639 gid_t gid;
640 uint_t ngids;
641 gid_t *gids;
642 struct auth_cache *p = ran->ran_auth;
643 char *netid = ran->ran_netid;
644
645 ASSERT(p != NULL);
646 ASSERT(netid != NULL);
653 * Once the entry goes INVALID, it can not change
654 * state.
655 *
656 * No need to refresh entries also in a case we are
657 * just shutting down.
658 *
659 * In general, there is no need to hold the
660 * refreshq_lock to test the refreshq_thread_state. We
661 * do hold it at other places because there is some
662 * related thread synchronization (or some other tasks)
663 * close to the refreshq_thread_state check.
664 *
665 * The check for the refreshq_thread_state value here
666 * is purely advisory to allow the faster
667 * nfsauth_refresh_thread() shutdown. In a case we
668 * will miss such advisory, nothing catastrophic
669 * happens: we will just spin longer here before the
670 * shutdown.
671 */
672 if (p->auth_state == NFS_AUTH_INVALID ||
673 refreshq_thread_state != REFRESHQ_THREAD_RUNNING) {
674 mutex_exit(&p->auth_lock);
675
676 if (p->auth_state == NFS_AUTH_INVALID)
677 nfsauth_free_node(p);
678
679 strfree(netid);
680
681 continue;
682 }
683
684 /*
685 * Make sure the state is valid. Note that once we
686 * change the state to NFS_AUTH_REFRESHING, no other
687 * thread will be able to work on this entry.
688 */
689 ASSERT(p->auth_state == NFS_AUTH_STALE);
690
691 p->auth_state = NFS_AUTH_REFRESHING;
692 mutex_exit(&p->auth_lock);
693
694 DTRACE_PROBE2(nfsauth__debug__cache__refresh,
695 struct exportinfo *, exi,
696 struct auth_cache *, p);
697
698 /*
699 * The first caching of the access rights
700 * is done with the netid pulled out of the
701 * request from the client. All subsequent
702 * users of the cache may or may not have
703 * the same netid. It doesn't matter. So
704 * when we refresh, we simply use the netid
705 * of the request which triggered the
706 * refresh attempt.
707 */
708 retrieval = nfsauth_retrieve(exi, netid,
709 p->auth_flavor, &p->auth_clnt->authc_addr, &access,
710 p->auth_clnt_cred, &uid, &gid, &ngids, &gids);
711
712 /*
713 * This can only be set in one other place
714 * and the state has to be NFS_AUTH_FRESH.
715 */
716 strfree(netid);
717
718 mutex_enter(&p->auth_lock);
719 if (p->auth_state == NFS_AUTH_INVALID) {
720 mutex_exit(&p->auth_lock);
721 nfsauth_free_node(p);
722 if (retrieval == TRUE)
723 kmem_free(gids, ngids * sizeof (gid_t));
724 } else {
725 /*
726 * If we got an error, do not reset the
727 * time. This will cause the next access
728 * check for the client to reschedule this
735 p->auth_srv_gid = gid;
736 kmem_free(p->auth_srv_gids,
737 p->auth_srv_ngids * sizeof (gid_t));
738 p->auth_srv_ngids = ngids;
739 p->auth_srv_gids = gids;
740
741 p->auth_freshness = gethrestime_sec();
742 }
743 p->auth_state = NFS_AUTH_FRESH;
744
745 cv_broadcast(&p->auth_cv);
746 mutex_exit(&p->auth_lock);
747 }
748 }
749
750 list_destroy(&ren->ren_authlist);
751 exi_rele(ren->ren_exi);
752 kmem_free(ren, sizeof (refreshq_exi_node_t));
753 }
754
755 refreshq_thread_state = REFRESHQ_THREAD_HALTED;
756 cv_broadcast(&refreshq_cv);
757 CALLB_CPR_EXIT(&cprinfo);
758 zthread_exit();
759 }
760
761 int
762 nfsauth_cache_clnt_compar(const void *v1, const void *v2)
763 {
764 int c;
765
766 const struct auth_cache_clnt *a1 = (const struct auth_cache_clnt *)v1;
767 const struct auth_cache_clnt *a2 = (const struct auth_cache_clnt *)v2;
768
769 if (a1->authc_addr.len < a2->authc_addr.len)
770 return (-1);
771 if (a1->authc_addr.len > a2->authc_addr.len)
772 return (1);
773
774 c = memcmp(a1->authc_addr.buf, a2->authc_addr.buf, a1->authc_addr.len);
775 if (c < 0)
776 return (-1);
777 if (c > 0)
809 return (1);
810
811 c = memcmp(crgetgroups(a1->auth_clnt_cred),
812 crgetgroups(a2->auth_clnt_cred), crgetngroups(a1->auth_clnt_cred));
813 if (c < 0)
814 return (-1);
815 if (c > 0)
816 return (1);
817
818 return (0);
819 }
820
821 /*
822 * Get the access information from the cache or callup to the mountd
823 * to get and cache the access information in the kernel.
824 */
825 static int
826 nfsauth_cache_get(struct exportinfo *exi, struct svc_req *req, int flavor,
827 cred_t *cr, uid_t *uid, gid_t *gid, uint_t *ngids, gid_t **gids)
828 {
829 struct netbuf *taddrmask;
830 struct netbuf addr; /* temporary copy of client's address */
831 const struct netbuf *claddr;
832 avl_tree_t *tree;
833 struct auth_cache ac; /* used as a template for avl_find() */
834 struct auth_cache_clnt *c;
835 struct auth_cache_clnt acc; /* used as a template for avl_find() */
836 struct auth_cache *p = NULL;
837 int access;
838
839 uid_t tmpuid;
840 gid_t tmpgid;
841 uint_t tmpngids;
842 gid_t *tmpgids;
843
844 avl_index_t where; /* used for avl_find()/avl_insert() */
845
846 ASSERT(cr != NULL);
847
848 /*
849 * Now check whether this client already
850 * has an entry for this flavor in the cache
851 * for this export.
852 * Get the caller's address, mask off the
853 * parts of the address that do not identify
854 * the host (port number, etc), and then hash
855 * it to find the chain of cache entries.
856 */
857
858 claddr = svc_getrpccaller(req->rq_xprt);
859 addr = *claddr;
860 addr.buf = kmem_alloc(addr.maxlen, KM_SLEEP);
861 bcopy(claddr->buf, addr.buf, claddr->len);
862
863 SVC_GETADDRMASK(req->rq_xprt, SVC_TATTR_ADDRMASK, (void **)&taddrmask);
864 ASSERT(taddrmask != NULL);
865 addrmask(&addr, taddrmask);
866
867 ac.auth_flavor = flavor;
868 ac.auth_clnt_cred = crdup(cr);
869
870 acc.authc_addr = addr;
871
872 tree = exi->exi_cache[hash(&addr)];
873
874 rw_enter(&exi->exi_cache_lock, RW_READER);
875 c = (struct auth_cache_clnt *)avl_find(tree, &acc, NULL);
876
877 if (c == NULL) {
878 struct auth_cache_clnt *nc;
879
880 rw_exit(&exi->exi_cache_lock);
881
987
988 /*
989 * If the cache entry is not valid yet, we need to retrieve the
990 * info ourselves.
991 */
992 if (p->auth_state == NFS_AUTH_NEW) {
993 bool_t res;
994 /*
995 * NFS_AUTH_NEW is the default output auth_state value in a
996 * case we failed somewhere below.
997 */
998 auth_state_t state = NFS_AUTH_NEW;
999
1000 p->auth_state = NFS_AUTH_WAITING;
1001 mutex_exit(&p->auth_lock);
1002 kmem_free(addr.buf, addr.maxlen);
1003 addr = p->auth_clnt->authc_addr;
1004
1005 atomic_inc_uint(&nfsauth_cache_miss);
1006
1007 res = nfsauth_retrieve(exi, svc_getnetid(req->rq_xprt), flavor,
1008 &addr, &access, cr, &tmpuid, &tmpgid, &tmpngids, &tmpgids);
1009
1010 p->auth_access = access;
1011 p->auth_time = p->auth_freshness = gethrestime_sec();
1012
1013 if (res == TRUE) {
1014 if (uid != NULL)
1015 *uid = tmpuid;
1016 if (gid != NULL)
1017 *gid = tmpgid;
1018 if (ngids != NULL && gids != NULL) {
1019 *ngids = tmpngids;
1020 *gids = tmpgids;
1021
1022 /*
1023 * We need a copy of gids for the
1024 * auth_cache entry
1025 */
1026 tmpgids = kmem_alloc(tmpngids * sizeof (gid_t),
1027 KM_NOSLEEP | KM_NORMALPRI);
1028 if (tmpgids != NULL)
1073
1074 if ((refresh > NFSAUTH_CACHE_REFRESH) &&
1075 p->auth_state == NFS_AUTH_FRESH) {
1076 refreshq_auth_node_t *ran;
1077 uint_t nacr;
1078
1079 p->auth_state = NFS_AUTH_STALE;
1080 mutex_exit(&p->auth_lock);
1081
1082 nacr = atomic_inc_uint_nv(&nfsauth_cache_refresh);
1083 DTRACE_PROBE3(nfsauth__debug__cache__stale,
1084 struct exportinfo *, exi,
1085 struct auth_cache *, p,
1086 uint_t, nacr);
1087
1088 ran = kmem_alloc(sizeof (refreshq_auth_node_t),
1089 KM_SLEEP);
1090 ran->ran_auth = p;
1091 ran->ran_netid = strdup(svc_getnetid(req->rq_xprt));
1092
1093 mutex_enter(&refreshq_lock);
1094 /*
1095 * We should not add a work queue
1096 * item if the thread is not
1097 * accepting them.
1098 */
1099 if (refreshq_thread_state == REFRESHQ_THREAD_RUNNING) {
1100 refreshq_exi_node_t *ren;
1101
1102 /*
1103 * Is there an existing exi_list?
1104 */
1105 for (ren = list_head(&refreshq_queue);
1106 ren != NULL;
1107 ren = list_next(&refreshq_queue, ren)) {
1108 if (ren->ren_exi == exi) {
1109 list_insert_tail(
1110 &ren->ren_authlist, ran);
1111 break;
1112 }
1113 }
1114
1115 if (ren == NULL) {
1116 ren = kmem_alloc(
1117 sizeof (refreshq_exi_node_t),
1118 KM_SLEEP);
1119
1120 exi_hold(exi);
1121 ren->ren_exi = exi;
1122
1123 list_create(&ren->ren_authlist,
1124 sizeof (refreshq_auth_node_t),
1125 offsetof(refreshq_auth_node_t,
1126 ran_node));
1127
1128 list_insert_tail(&ren->ren_authlist,
1129 ran);
1130 list_insert_tail(&refreshq_queue, ren);
1131 }
1132
1133 cv_broadcast(&refreshq_cv);
1134 } else {
1135 strfree(ran->ran_netid);
1136 kmem_free(ran, sizeof (refreshq_auth_node_t));
1137 }
1138
1139 mutex_exit(&refreshq_lock);
1140 } else {
1141 mutex_exit(&p->auth_lock);
1142 }
1143
1144 nach = atomic_inc_uint_nv(&nfsauth_cache_hit);
1145 DTRACE_PROBE2(nfsauth__debug__cache__hit,
1146 uint_t, nach,
1147 time_t, refresh);
1148
1149 kmem_free(addr.buf, addr.maxlen);
1150 }
1151
1152 return (access);
1153
1154 retrieve:
1155 crfree(ac.auth_clnt_cred);
1156
1157 /*
1158 * Retrieve the required data without caching.
1159 */
1160
1161 ASSERT(p == NULL);
1162
1163 atomic_inc_uint(&nfsauth_cache_miss);
1164
1165 if (nfsauth_retrieve(exi, svc_getnetid(req->rq_xprt), flavor, &addr,
1166 &access, cr, &tmpuid, &tmpgid, &tmpngids, &tmpgids)) {
1167 if (uid != NULL)
1168 *uid = tmpuid;
1169 if (gid != NULL)
1170 *gid = tmpgid;
1171 if (ngids != NULL && gids != NULL) {
1172 *ngids = tmpngids;
1173 *gids = tmpgids;
1174 } else {
1175 kmem_free(tmpgids, tmpngids * sizeof (gid_t));
1176 }
1177 }
1178
1179 kmem_free(addr.buf, addr.maxlen);
1180
1181 return (access);
1182 }
1183
1184 /*
1185 * Check if the requesting client has access to the filesystem with
1186 * a given nfs flavor number which is an explicitly shared flavor.
1393 nfsauth_cache_free(struct exportinfo *exi)
1394 {
1395 int i;
1396
1397 /*
1398 * The only way we got here was with an exi_rele, which means that no
1399 * auth cache entry is being refreshed.
1400 */
1401
1402 for (i = 0; i < AUTH_TABLESIZE; i++) {
1403 avl_tree_t *tree = exi->exi_cache[i];
1404 void *cookie = NULL;
1405 struct auth_cache_clnt *node;
1406
1407 while ((node = avl_destroy_nodes(tree, &cookie)) != NULL)
1408 nfsauth_free_clnt_node(node);
1409 }
1410 }
1411
1412 /*
1413 * Called by the kernel memory allocator when
1414 * memory is low. Free unused cache entries.
1415 * If that's not enough, the VM system will
1416 * call again for some more.
1417 */
1418 /*ARGSUSED*/
1419 void
1420 exi_cache_reclaim(void *cdrarg)
1421 {
1422 int i;
1423 struct exportinfo *exi;
1424
1425 rw_enter(&exported_lock, RW_READER);
1426
1427 for (i = 0; i < EXPTABLESIZE; i++) {
1428 for (exi = exptable[i]; exi; exi = exi->fid_hash.next) {
1429 exi_cache_trim(exi);
1430 }
1431 }
1432
1433 rw_exit(&exported_lock);
1434
1435 atomic_inc_uint(&nfsauth_cache_reclaim);
1436 }
1437
1438 void
1439 exi_cache_trim(struct exportinfo *exi)
1440 {
1441 struct auth_cache_clnt *c;
1442 struct auth_cache_clnt *nextc;
1443 struct auth_cache *p;
1444 struct auth_cache *next;
1445 int i;
1446 time_t stale_time;
1447 avl_tree_t *tree;
1448
1449 for (i = 0; i < AUTH_TABLESIZE; i++) {
1450 tree = exi->exi_cache[i];
1451 stale_time = gethrestime_sec() - NFSAUTH_CACHE_TRIM;
1452 rw_enter(&exi->exi_cache_lock, RW_READER);
1453
1454 /*
1455 * Free entries that have not been
1456 * used for NFSAUTH_CACHE_TRIM seconds.
1457 */
1458 for (c = avl_first(tree); c != NULL; c = AVL_NEXT(tree, c)) {
|
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21
22 /*
23 * Copyright (c) 1995, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2015 by Delphix. All rights reserved.
25 * Copyright (c) 2015 Joyent, Inc. All rights reserved.
26 * Copyright 2018 Nexenta Systems, Inc. All rights reserved.
27 */
28
29 #include <sys/param.h>
30 #include <sys/errno.h>
31 #include <sys/vfs.h>
32 #include <sys/vnode.h>
33 #include <sys/cred.h>
34 #include <sys/cmn_err.h>
35 #include <sys/systm.h>
36 #include <sys/kmem.h>
37 #include <sys/pathname.h>
38 #include <sys/utsname.h>
39 #include <sys/debug.h>
40 #include <sys/door.h>
41 #include <sys/sdt.h>
42 #include <sys/thread.h>
43 #include <sys/avl.h>
44
45 #include <rpc/types.h>
46 #include <rpc/auth.h>
47 #include <rpc/clnt.h>
48
49 #include <nfs/nfs.h>
50 #include <nfs/export.h>
51 #include <nfs/nfs_clnt.h>
52 #include <nfs/auth.h>
53
54 static struct kmem_cache *exi_cache_handle;
55 static void exi_cache_reclaim(void *);
56 static void exi_cache_reclaim_zone(nfs_globals_t *);
57 static void exi_cache_trim(struct exportinfo *exi);
58
59 extern pri_t minclsyspri;
60
61 /* NFS auth cache statistics */
62 volatile uint_t nfsauth_cache_hit;
63 volatile uint_t nfsauth_cache_miss;
64 volatile uint_t nfsauth_cache_refresh;
65 volatile uint_t nfsauth_cache_reclaim;
66 volatile uint_t exi_cache_auth_reclaim_failed;
67 volatile uint_t exi_cache_clnt_reclaim_failed;
68
69 /*
70 * The lifetime of an auth cache entry:
71 * ------------------------------------
72 *
73 * An auth cache entry is created with both the auth_time
74 * and auth_freshness times set to the current time.
75 *
76 * Upon every client access which results in a hit, the
77 * auth_time will be updated.
78 *
79 * If a client access determines that the auth_freshness
80 * indicates that the entry is STALE, then it will be
81 * refreshed. Note that this will explicitly reset
105
106 /*
107 * While we could encapuslate the exi_list inside the
108 * exi structure, we can't do that for the auth_list.
109 * So, to keep things looking clean, we keep them both
110 * in these external lists.
111 */
112 typedef struct refreshq_exi_node {
113 struct exportinfo *ren_exi;
114 list_t ren_authlist;
115 list_node_t ren_node;
116 } refreshq_exi_node_t;
117
118 typedef struct refreshq_auth_node {
119 struct auth_cache *ran_auth;
120 char *ran_netid;
121 list_node_t ran_node;
122 } refreshq_auth_node_t;
123
124 /*
125 * Used to manipulate things on the refreshq_queue. Note that the refresh
126 * thread will effectively pop a node off of the queue, at which point it
127 * will no longer need to hold the mutex.
128 */
129 static kmutex_t refreshq_lock;
130 static list_t refreshq_queue;
131 static kcondvar_t refreshq_cv;
132
133 /*
134 * If there is ever a problem with loading the module, then nfsauth_fini()
135 * needs to be called to remove state. In that event, since the refreshq
136 * thread has been started, they need to work together to get rid of state.
137 */
138 typedef enum nfsauth_refreshq_thread_state {
139 REFRESHQ_THREAD_RUNNING,
140 REFRESHQ_THREAD_FINI_REQ,
141 REFRESHQ_THREAD_HALTED,
142 REFRESHQ_THREAD_NEED_CREATE
143 } nfsauth_refreshq_thread_state_t;
144
145 typedef struct nfsauth_globals {
146 kmutex_t mountd_lock;
147 door_handle_t mountd_dh;
148
149 /*
150 * Used to manipulate things on the refreshq_queue. Note that the
151 * refresh thread will effectively pop a node off of the queue,
152 * at which point it will no longer need to hold the mutex.
153 */
154 kmutex_t refreshq_lock;
155 list_t refreshq_queue;
156 kcondvar_t refreshq_cv;
157
158 /*
159 * A list_t would be overkill. These are auth_cache entries which are
160 * no longer linked to an exi. It should be the case that all of their
161 * states are NFS_AUTH_INVALID, i.e., the only way to be put on this
162 * list is iff their state indicated that they had been placed on the
163 * refreshq_queue.
164 *
165 * Note that while there is no link from the exi or back to the exi,
166 * the exi can not go away until these entries are harvested.
167 */
168 struct auth_cache *refreshq_dead_entries;
169 nfsauth_refreshq_thread_state_t refreshq_thread_state;
170
171 } nfsauth_globals_t;
172
173 static void nfsauth_free_node(struct auth_cache *);
174 static void nfsauth_refresh_thread(nfsauth_globals_t *);
175
176 static int nfsauth_cache_compar(const void *, const void *);
177
178 static nfsauth_globals_t *
179 nfsauth_get_zg(void)
180 {
181 nfs_globals_t *ng = nfs_srv_getzg();
182 nfsauth_globals_t *nag = ng->nfs_auth;
183 ASSERT(nag != NULL);
184 return (nag);
185 }
186
187 void
188 mountd_args(uint_t did)
189 {
190 nfsauth_globals_t *nag;
191
192 nag = nfsauth_get_zg();
193 mutex_enter(&nag->mountd_lock);
194 if (nag->mountd_dh != NULL)
195 door_ki_rele(nag->mountd_dh);
196 nag->mountd_dh = door_ki_lookup(did);
197 mutex_exit(&nag->mountd_lock);
198 }
199
200 void
201 nfsauth_init(void)
202 {
203 exi_cache_handle = kmem_cache_create("exi_cache_handle",
204 sizeof (struct auth_cache), 0, NULL, NULL,
205 exi_cache_reclaim, NULL, NULL, 0);
206 }
207
208 void
209 nfsauth_fini(void)
210 {
211 kmem_cache_destroy(exi_cache_handle);
212 }
213
214 void
215 nfsauth_zone_init(nfs_globals_t *ng)
216 {
217 nfsauth_globals_t *nag;
218
219 nag = kmem_zalloc(sizeof (*nag), KM_SLEEP);
220
221 /*
222 * mountd can be restarted by smf(5). We need to make sure
223 * the updated door handle will safely make it to mountd_dh.
224 */
225 mutex_init(&nag->mountd_lock, NULL, MUTEX_DEFAULT, NULL);
226 mutex_init(&nag->refreshq_lock, NULL, MUTEX_DEFAULT, NULL);
227 list_create(&nag->refreshq_queue, sizeof (refreshq_exi_node_t),
228 offsetof(refreshq_exi_node_t, ren_node));
229 cv_init(&nag->refreshq_cv, NULL, CV_DEFAULT, NULL);
230 nag->refreshq_thread_state = REFRESHQ_THREAD_NEED_CREATE;
231
232 ng->nfs_auth = nag;
233 }
234
235 void
236 nfsauth_zone_shutdown(nfs_globals_t *ng)
237 {
238 refreshq_exi_node_t *ren;
239 nfsauth_globals_t *nag = ng->nfs_auth;
240
241 /* Prevent the nfsauth_refresh_thread from getting new work */
242 mutex_enter(&nag->refreshq_lock);
243 if (nag->refreshq_thread_state == REFRESHQ_THREAD_RUNNING) {
244 nag->refreshq_thread_state = REFRESHQ_THREAD_FINI_REQ;
245 cv_broadcast(&nag->refreshq_cv);
246
247 /* Wait for nfsauth_refresh_thread() to exit */
248 while (nag->refreshq_thread_state != REFRESHQ_THREAD_HALTED)
249 cv_wait(&nag->refreshq_cv, &nag->refreshq_lock);
250 }
251 mutex_exit(&nag->refreshq_lock);
252
253 /*
254 * Walk the exi_list and in turn, walk the auth_lists and free all
255 * lists. In addition, free INVALID auth_cache entries.
256 */
257 while ((ren = list_remove_head(&nag->refreshq_queue))) {
258 refreshq_auth_node_t *ran;
259
260 while ((ran = list_remove_head(&ren->ren_authlist)) != NULL) {
261 struct auth_cache *p = ran->ran_auth;
262 if (p->auth_state == NFS_AUTH_INVALID)
263 nfsauth_free_node(p);
264 strfree(ran->ran_netid);
265 kmem_free(ran, sizeof (*ran));
266 }
267
268 list_destroy(&ren->ren_authlist);
269 exi_rele(ren->ren_exi);
270 kmem_free(ren, sizeof (*ren));
271 }
272 }
273
274 void
275 nfsauth_zone_fini(nfs_globals_t *ng)
276 {
277 nfsauth_globals_t *nag = ng->nfs_auth;
278
279 ng->nfs_auth = NULL;
280
281 list_destroy(&nag->refreshq_queue);
282 cv_destroy(&nag->refreshq_cv);
283 mutex_destroy(&nag->refreshq_lock);
284 mutex_destroy(&nag->mountd_lock);
285 /* Extra cleanup. */
286 if (nag->mountd_dh != NULL)
287 door_ki_rele(nag->mountd_dh);
288 kmem_free(nag, sizeof (*nag));
289 }
290
291 /*
292 * Convert the address in a netbuf to
293 * a hash index for the auth_cache table.
294 */
295 static int
296 hash(struct netbuf *a)
297 {
298 int i, h = 0;
299
300 for (i = 0; i < a->len; i++)
301 h ^= a->buf[i];
302
303 return (h & (AUTH_TABLESIZE - 1));
304 }
305
306 /*
307 * Mask out the components of an
308 * address that do not identify
360 static void
361 sys_log(const char *msg)
362 {
363 static time_t tstamp = 0;
364 time_t now;
365
366 /*
367 * msg is shown (at most) once per minute
368 */
369 now = gethrestime_sec();
370 if ((tstamp + 60) < now) {
371 tstamp = now;
372 cmn_err(CE_WARN, msg);
373 }
374 }
375
376 /*
377 * Callup to the mountd to get access information in the kernel.
378 */
379 static bool_t
380 nfsauth_retrieve(nfsauth_globals_t *nag, struct exportinfo *exi,
381 char *req_netid, int flavor, struct netbuf *addr, int *access,
382 cred_t *clnt_cred, uid_t *srv_uid, gid_t *srv_gid, uint_t *srv_gids_cnt,
383 gid_t **srv_gids)
384 {
385 varg_t varg = {0};
386 nfsauth_res_t res = {0};
387 XDR xdrs;
388 size_t absz;
389 caddr_t abuf;
390 int last = 0;
391 door_arg_t da;
392 door_info_t di;
393 door_handle_t dh;
394 uint_t ntries = 0;
395
396 /*
397 * No entry in the cache for this client/flavor
398 * so we need to call the nfsauth service in the
399 * mount daemon.
400 */
401
402 varg.vers = V_PROTO;
403 varg.arg_u.arg.cmd = NFSAUTH_ACCESS;
436 XDR_DESTROY(&xdrs);
437
438 /*
439 * Prepare the door arguments
440 *
441 * We don't know the size of the message the daemon
442 * will pass back to us. By setting rbuf to NULL,
443 * we force the door code to allocate a buf of the
444 * appropriate size. We must set rsize > 0, however,
445 * else the door code acts as if no response was
446 * expected and doesn't pass the data to us.
447 */
448 da.data_ptr = (char *)abuf;
449 da.data_size = absz;
450 da.desc_ptr = NULL;
451 da.desc_num = 0;
452 da.rbuf = NULL;
453 da.rsize = 1;
454
455 retry:
456 mutex_enter(&nag->mountd_lock);
457 dh = nag->mountd_dh;
458 if (dh != NULL)
459 door_ki_hold(dh);
460 mutex_exit(&nag->mountd_lock);
461
462 if (dh == NULL) {
463 /*
464 * The rendezvous point has not been established yet!
465 * This could mean that either mountd(1m) has not yet
466 * been started or that _this_ routine nuked the door
467 * handle after receiving an EINTR for a REVOKED door.
468 *
469 * Returning NFSAUTH_DROP will cause the NFS client
470 * to retransmit the request, so let's try to be more
471 * rescillient and attempt for ntries before we bail.
472 */
473 if (++ntries % NFSAUTH_DR_TRYCNT) {
474 delay(hz);
475 goto retry;
476 }
477
478 kmem_free(abuf, absz);
479
480 sys_log("nfsauth: mountd has not established door");
510 /*
511 * Server out of resources; back off for a bit
512 */
513 door_ki_rele(dh);
514 delay(hz);
515 goto retry;
516 /* NOTREACHED */
517
518 case EINTR:
519 if (!door_ki_info(dh, &di)) {
520 door_ki_rele(dh);
521
522 if (di.di_attributes & DOOR_REVOKED) {
523 /*
524 * The server barfed and revoked
525 * the (existing) door on us; we
526 * want to wait to give smf(5) a
527 * chance to restart mountd(1m)
528 * and establish a new door handle.
529 */
530 mutex_enter(&nag->mountd_lock);
531 if (dh == nag->mountd_dh) {
532 door_ki_rele(nag->mountd_dh);
533 nag->mountd_dh = NULL;
534 }
535 mutex_exit(&nag->mountd_lock);
536 delay(hz);
537 goto retry;
538 }
539 /*
540 * If the door was _not_ revoked on us,
541 * then more than likely we took an INTR,
542 * so we need to fail the operation.
543 */
544 goto fail;
545 }
546 /*
547 * The only failure that can occur from getting
548 * the door info is EINVAL, so we let the code
549 * below handle it.
550 */
551 /* FALLTHROUGH */
552
553 case EBADF:
554 case EINVAL:
555 default:
612
613 case NFSAUTH_DR_EFAIL:
614 case NFSAUTH_DR_DECERR:
615 case NFSAUTH_DR_BADCMD:
616 default:
617 xdr_free(xdr_nfsauth_res, (char *)&res);
618 fail:
619 *access = NFSAUTH_DENIED;
620 kmem_free(abuf, absz);
621 return (FALSE);
622 /* NOTREACHED */
623 }
624
625 xdr_free(xdr_nfsauth_res, (char *)&res);
626 kmem_free(abuf, absz);
627
628 return (TRUE);
629 }
630
631 static void
632 nfsauth_refresh_thread(nfsauth_globals_t *nag)
633 {
634 refreshq_exi_node_t *ren;
635 refreshq_auth_node_t *ran;
636
637 struct exportinfo *exi;
638
639 int access;
640 bool_t retrieval;
641
642 callb_cpr_t cprinfo;
643
644 CALLB_CPR_INIT(&cprinfo, &nag->refreshq_lock, callb_generic_cpr,
645 "nfsauth_refresh");
646
647 for (;;) {
648 mutex_enter(&nag->refreshq_lock);
649 if (nag->refreshq_thread_state != REFRESHQ_THREAD_RUNNING) {
650 /* Keep the hold on the lock! */
651 break;
652 }
653
654 ren = list_remove_head(&nag->refreshq_queue);
655 if (ren == NULL) {
656 CALLB_CPR_SAFE_BEGIN(&cprinfo);
657 cv_wait(&nag->refreshq_cv, &nag->refreshq_lock);
658 CALLB_CPR_SAFE_END(&cprinfo, &nag->refreshq_lock);
659 mutex_exit(&nag->refreshq_lock);
660 continue;
661 }
662 mutex_exit(&nag->refreshq_lock);
663
664 exi = ren->ren_exi;
665 ASSERT(exi != NULL);
666
667 /*
668 * Since the ren was removed from the refreshq_queue above,
669 * this is the only thread aware about the ren existence, so we
670 * have the exclusive ownership of it and we do not need to
671 * protect it by any lock.
672 */
673 while ((ran = list_remove_head(&ren->ren_authlist))) {
674 uid_t uid;
675 gid_t gid;
676 uint_t ngids;
677 gid_t *gids;
678 struct auth_cache *p = ran->ran_auth;
679 char *netid = ran->ran_netid;
680
681 ASSERT(p != NULL);
682 ASSERT(netid != NULL);
689 * Once the entry goes INVALID, it can not change
690 * state.
691 *
692 * No need to refresh entries also in a case we are
693 * just shutting down.
694 *
695 * In general, there is no need to hold the
696 * refreshq_lock to test the refreshq_thread_state. We
697 * do hold it at other places because there is some
698 * related thread synchronization (or some other tasks)
699 * close to the refreshq_thread_state check.
700 *
701 * The check for the refreshq_thread_state value here
702 * is purely advisory to allow the faster
703 * nfsauth_refresh_thread() shutdown. In a case we
704 * will miss such advisory, nothing catastrophic
705 * happens: we will just spin longer here before the
706 * shutdown.
707 */
708 if (p->auth_state == NFS_AUTH_INVALID ||
709 nag->refreshq_thread_state !=
710 REFRESHQ_THREAD_RUNNING) {
711 mutex_exit(&p->auth_lock);
712
713 if (p->auth_state == NFS_AUTH_INVALID)
714 nfsauth_free_node(p);
715
716 strfree(netid);
717
718 continue;
719 }
720
721 /*
722 * Make sure the state is valid. Note that once we
723 * change the state to NFS_AUTH_REFRESHING, no other
724 * thread will be able to work on this entry.
725 */
726 ASSERT(p->auth_state == NFS_AUTH_STALE);
727
728 p->auth_state = NFS_AUTH_REFRESHING;
729 mutex_exit(&p->auth_lock);
730
731 DTRACE_PROBE2(nfsauth__debug__cache__refresh,
732 struct exportinfo *, exi,
733 struct auth_cache *, p);
734
735 /*
736 * The first caching of the access rights
737 * is done with the netid pulled out of the
738 * request from the client. All subsequent
739 * users of the cache may or may not have
740 * the same netid. It doesn't matter. So
741 * when we refresh, we simply use the netid
742 * of the request which triggered the
743 * refresh attempt.
744 */
745 retrieval = nfsauth_retrieve(nag, exi, netid,
746 p->auth_flavor, &p->auth_clnt->authc_addr, &access,
747 p->auth_clnt_cred, &uid, &gid, &ngids, &gids);
748
749 /*
750 * This can only be set in one other place
751 * and the state has to be NFS_AUTH_FRESH.
752 */
753 strfree(netid);
754
755 mutex_enter(&p->auth_lock);
756 if (p->auth_state == NFS_AUTH_INVALID) {
757 mutex_exit(&p->auth_lock);
758 nfsauth_free_node(p);
759 if (retrieval == TRUE)
760 kmem_free(gids, ngids * sizeof (gid_t));
761 } else {
762 /*
763 * If we got an error, do not reset the
764 * time. This will cause the next access
765 * check for the client to reschedule this
772 p->auth_srv_gid = gid;
773 kmem_free(p->auth_srv_gids,
774 p->auth_srv_ngids * sizeof (gid_t));
775 p->auth_srv_ngids = ngids;
776 p->auth_srv_gids = gids;
777
778 p->auth_freshness = gethrestime_sec();
779 }
780 p->auth_state = NFS_AUTH_FRESH;
781
782 cv_broadcast(&p->auth_cv);
783 mutex_exit(&p->auth_lock);
784 }
785 }
786
787 list_destroy(&ren->ren_authlist);
788 exi_rele(ren->ren_exi);
789 kmem_free(ren, sizeof (refreshq_exi_node_t));
790 }
791
792 nag->refreshq_thread_state = REFRESHQ_THREAD_HALTED;
793 cv_broadcast(&nag->refreshq_cv);
794 CALLB_CPR_EXIT(&cprinfo);
795 DTRACE_PROBE(nfsauth__nfsauth__refresh__thread__exit);
796 zthread_exit();
797 }
798
799 int
800 nfsauth_cache_clnt_compar(const void *v1, const void *v2)
801 {
802 int c;
803
804 const struct auth_cache_clnt *a1 = (const struct auth_cache_clnt *)v1;
805 const struct auth_cache_clnt *a2 = (const struct auth_cache_clnt *)v2;
806
807 if (a1->authc_addr.len < a2->authc_addr.len)
808 return (-1);
809 if (a1->authc_addr.len > a2->authc_addr.len)
810 return (1);
811
812 c = memcmp(a1->authc_addr.buf, a2->authc_addr.buf, a1->authc_addr.len);
813 if (c < 0)
814 return (-1);
815 if (c > 0)
847 return (1);
848
849 c = memcmp(crgetgroups(a1->auth_clnt_cred),
850 crgetgroups(a2->auth_clnt_cred), crgetngroups(a1->auth_clnt_cred));
851 if (c < 0)
852 return (-1);
853 if (c > 0)
854 return (1);
855
856 return (0);
857 }
858
859 /*
860 * Get the access information from the cache or callup to the mountd
861 * to get and cache the access information in the kernel.
862 */
863 static int
864 nfsauth_cache_get(struct exportinfo *exi, struct svc_req *req, int flavor,
865 cred_t *cr, uid_t *uid, gid_t *gid, uint_t *ngids, gid_t **gids)
866 {
867 nfsauth_globals_t *nag;
868 struct netbuf *taddrmask;
869 struct netbuf addr; /* temporary copy of client's address */
870 const struct netbuf *claddr;
871 avl_tree_t *tree;
872 struct auth_cache ac; /* used as a template for avl_find() */
873 struct auth_cache_clnt *c;
874 struct auth_cache_clnt acc; /* used as a template for avl_find() */
875 struct auth_cache *p = NULL;
876 int access;
877
878 uid_t tmpuid;
879 gid_t tmpgid;
880 uint_t tmpngids;
881 gid_t *tmpgids;
882
883 avl_index_t where; /* used for avl_find()/avl_insert() */
884
885 ASSERT(cr != NULL);
886
887 ASSERT3P(curzone->zone_id, ==, exi->exi_zoneid);
888 nag = nfsauth_get_zg();
889
890 /*
891 * Now check whether this client already
892 * has an entry for this flavor in the cache
893 * for this export.
894 * Get the caller's address, mask off the
895 * parts of the address that do not identify
896 * the host (port number, etc), and then hash
897 * it to find the chain of cache entries.
898 */
899
900 claddr = svc_getrpccaller(req->rq_xprt);
901 addr = *claddr;
902 if (claddr->len != 0) {
903 addr.buf = kmem_alloc(addr.maxlen, KM_SLEEP);
904 bcopy(claddr->buf, addr.buf, claddr->len);
905 } else {
906 addr.buf = NULL;
907 }
908
909 SVC_GETADDRMASK(req->rq_xprt, SVC_TATTR_ADDRMASK, (void **)&taddrmask);
910 ASSERT(taddrmask != NULL);
911 addrmask(&addr, taddrmask);
912
913 ac.auth_flavor = flavor;
914 ac.auth_clnt_cred = crdup(cr);
915
916 acc.authc_addr = addr;
917
918 tree = exi->exi_cache[hash(&addr)];
919
920 rw_enter(&exi->exi_cache_lock, RW_READER);
921 c = (struct auth_cache_clnt *)avl_find(tree, &acc, NULL);
922
923 if (c == NULL) {
924 struct auth_cache_clnt *nc;
925
926 rw_exit(&exi->exi_cache_lock);
927
1033
1034 /*
1035 * If the cache entry is not valid yet, we need to retrieve the
1036 * info ourselves.
1037 */
1038 if (p->auth_state == NFS_AUTH_NEW) {
1039 bool_t res;
1040 /*
1041 * NFS_AUTH_NEW is the default output auth_state value in a
1042 * case we failed somewhere below.
1043 */
1044 auth_state_t state = NFS_AUTH_NEW;
1045
1046 p->auth_state = NFS_AUTH_WAITING;
1047 mutex_exit(&p->auth_lock);
1048 kmem_free(addr.buf, addr.maxlen);
1049 addr = p->auth_clnt->authc_addr;
1050
1051 atomic_inc_uint(&nfsauth_cache_miss);
1052
1053 res = nfsauth_retrieve(nag, exi, svc_getnetid(req->rq_xprt),
1054 flavor, &addr, &access, cr, &tmpuid, &tmpgid, &tmpngids,
1055 &tmpgids);
1056
1057 p->auth_access = access;
1058 p->auth_time = p->auth_freshness = gethrestime_sec();
1059
1060 if (res == TRUE) {
1061 if (uid != NULL)
1062 *uid = tmpuid;
1063 if (gid != NULL)
1064 *gid = tmpgid;
1065 if (ngids != NULL && gids != NULL) {
1066 *ngids = tmpngids;
1067 *gids = tmpgids;
1068
1069 /*
1070 * We need a copy of gids for the
1071 * auth_cache entry
1072 */
1073 tmpgids = kmem_alloc(tmpngids * sizeof (gid_t),
1074 KM_NOSLEEP | KM_NORMALPRI);
1075 if (tmpgids != NULL)
1120
1121 if ((refresh > NFSAUTH_CACHE_REFRESH) &&
1122 p->auth_state == NFS_AUTH_FRESH) {
1123 refreshq_auth_node_t *ran;
1124 uint_t nacr;
1125
1126 p->auth_state = NFS_AUTH_STALE;
1127 mutex_exit(&p->auth_lock);
1128
1129 nacr = atomic_inc_uint_nv(&nfsauth_cache_refresh);
1130 DTRACE_PROBE3(nfsauth__debug__cache__stale,
1131 struct exportinfo *, exi,
1132 struct auth_cache *, p,
1133 uint_t, nacr);
1134
1135 ran = kmem_alloc(sizeof (refreshq_auth_node_t),
1136 KM_SLEEP);
1137 ran->ran_auth = p;
1138 ran->ran_netid = strdup(svc_getnetid(req->rq_xprt));
1139
1140 mutex_enter(&nag->refreshq_lock);
1141
1142 if (nag->refreshq_thread_state ==
1143 REFRESHQ_THREAD_NEED_CREATE) {
1144 /* Launch nfsauth refresh thread */
1145 nag->refreshq_thread_state =
1146 REFRESHQ_THREAD_RUNNING;
1147 (void) zthread_create(NULL, 0,
1148 nfsauth_refresh_thread, nag, 0,
1149 minclsyspri);
1150 }
1151
1152 /*
1153 * We should not add a work queue item if the thread
1154 * is not accepting them.
1155 */
1156 if (nag->refreshq_thread_state ==
1157 REFRESHQ_THREAD_RUNNING) {
1158 refreshq_exi_node_t *ren;
1159
1160 /*
1161 * Is there an existing exi_list?
1162 */
1163 for (ren = list_head(&nag->refreshq_queue);
1164 ren != NULL;
1165 ren = list_next(&nag->refreshq_queue,
1166 ren)) {
1167 if (ren->ren_exi == exi) {
1168 list_insert_tail(
1169 &ren->ren_authlist, ran);
1170 break;
1171 }
1172 }
1173
1174 if (ren == NULL) {
1175 ren = kmem_alloc(
1176 sizeof (refreshq_exi_node_t),
1177 KM_SLEEP);
1178
1179 exi_hold(exi);
1180 ren->ren_exi = exi;
1181
1182 list_create(&ren->ren_authlist,
1183 sizeof (refreshq_auth_node_t),
1184 offsetof(refreshq_auth_node_t,
1185 ran_node));
1186
1187 list_insert_tail(&ren->ren_authlist,
1188 ran);
1189 list_insert_tail(&nag->refreshq_queue,
1190 ren);
1191 }
1192
1193 cv_broadcast(&nag->refreshq_cv);
1194 } else {
1195 strfree(ran->ran_netid);
1196 kmem_free(ran, sizeof (refreshq_auth_node_t));
1197 }
1198
1199 mutex_exit(&nag->refreshq_lock);
1200 } else {
1201 mutex_exit(&p->auth_lock);
1202 }
1203
1204 nach = atomic_inc_uint_nv(&nfsauth_cache_hit);
1205 DTRACE_PROBE2(nfsauth__debug__cache__hit,
1206 uint_t, nach,
1207 time_t, refresh);
1208
1209 kmem_free(addr.buf, addr.maxlen);
1210 }
1211
1212 return (access);
1213
1214 retrieve:
1215 crfree(ac.auth_clnt_cred);
1216
1217 /*
1218 * Retrieve the required data without caching.
1219 */
1220
1221 ASSERT(p == NULL);
1222
1223 atomic_inc_uint(&nfsauth_cache_miss);
1224
1225 if (nfsauth_retrieve(nag, exi, svc_getnetid(req->rq_xprt), flavor,
1226 &addr, &access, cr, &tmpuid, &tmpgid, &tmpngids, &tmpgids)) {
1227 if (uid != NULL)
1228 *uid = tmpuid;
1229 if (gid != NULL)
1230 *gid = tmpgid;
1231 if (ngids != NULL && gids != NULL) {
1232 *ngids = tmpngids;
1233 *gids = tmpgids;
1234 } else {
1235 kmem_free(tmpgids, tmpngids * sizeof (gid_t));
1236 }
1237 }
1238
1239 kmem_free(addr.buf, addr.maxlen);
1240
1241 return (access);
1242 }
1243
1244 /*
1245 * Check if the requesting client has access to the filesystem with
1246 * a given nfs flavor number which is an explicitly shared flavor.
1453 nfsauth_cache_free(struct exportinfo *exi)
1454 {
1455 int i;
1456
1457 /*
1458 * The only way we got here was with an exi_rele, which means that no
1459 * auth cache entry is being refreshed.
1460 */
1461
1462 for (i = 0; i < AUTH_TABLESIZE; i++) {
1463 avl_tree_t *tree = exi->exi_cache[i];
1464 void *cookie = NULL;
1465 struct auth_cache_clnt *node;
1466
1467 while ((node = avl_destroy_nodes(tree, &cookie)) != NULL)
1468 nfsauth_free_clnt_node(node);
1469 }
1470 }
1471
1472 /*
1473 * Called by the kernel memory allocator when memory is low.
1474 * Free unused cache entries. If that's not enough, the VM system
1475 * will call again for some more.
1476 *
1477 * This needs to operate on all zones, so we take a reader lock
1478 * on the list of zones and walk the list. This is OK here
1479 * becuase exi_cache_trim doesn't block or cause new objects
1480 * to be allocated (basically just frees lots of stuff).
1481 * Use care if nfssrv_globals_rwl is taken as reader in any
1482 * other cases because it will block nfs_server_zone_init
1483 * and nfs_server_zone_fini, which enter as writer.
1484 */
1485 /*ARGSUSED*/
1486 void
1487 exi_cache_reclaim(void *cdrarg)
1488 {
1489 nfs_globals_t *ng;
1490
1491 rw_enter(&nfssrv_globals_rwl, RW_READER);
1492
1493 ng = list_head(&nfssrv_globals_list);
1494 while (ng != NULL) {
1495 exi_cache_reclaim_zone(ng);
1496 ng = list_next(&nfssrv_globals_list, ng);
1497 }
1498
1499 rw_exit(&nfssrv_globals_rwl);
1500 }
1501
1502 static void
1503 exi_cache_reclaim_zone(nfs_globals_t *ng)
1504 {
1505 int i;
1506 struct exportinfo *exi;
1507 nfs_export_t *ne = ng->nfs_export;
1508
1509 rw_enter(&ne->exported_lock, RW_READER);
1510
1511 for (i = 0; i < EXPTABLESIZE; i++) {
1512 for (exi = ne->exptable[i]; exi; exi = exi->fid_hash.next)
1513 exi_cache_trim(exi);
1514 }
1515
1516 rw_exit(&ne->exported_lock);
1517
1518 atomic_inc_uint(&nfsauth_cache_reclaim);
1519 }
1520
1521 static void
1522 exi_cache_trim(struct exportinfo *exi)
1523 {
1524 struct auth_cache_clnt *c;
1525 struct auth_cache_clnt *nextc;
1526 struct auth_cache *p;
1527 struct auth_cache *next;
1528 int i;
1529 time_t stale_time;
1530 avl_tree_t *tree;
1531
1532 for (i = 0; i < AUTH_TABLESIZE; i++) {
1533 tree = exi->exi_cache[i];
1534 stale_time = gethrestime_sec() - NFSAUTH_CACHE_TRIM;
1535 rw_enter(&exi->exi_cache_lock, RW_READER);
1536
1537 /*
1538 * Free entries that have not been
1539 * used for NFSAUTH_CACHE_TRIM seconds.
1540 */
1541 for (c = avl_first(tree); c != NULL; c = AVL_NEXT(tree, c)) {
|