Print this page
NEX-15279 support NFS server in zone
NEX-15520 online NFS shares cause zoneadm halt to hang in nfs_export_zone_fini
Portions contributed by: Dan Kruchinin dan.kruchinin@nexenta.com
Portions contributed by: Stepan Zastupov stepan.zastupov@gmail.com
Reviewed by: Joyce McIntosh <joyce.mcintosh@nexenta.com>
Reviewed by: Rob Gittins <rob.gittins@nexenta.com>
Reviewed by: Gordon Ross <gordon.ross@nexenta.com>
NEX-9275 Got "bad mutex" panic when run IO to nfs share from clients
Reviewed by: Yuri Pankov <yuri.pankov@nexenta.com>
Reviewed by: Sanjay Nadkarni <sanjay.nadkarni@nexenta.com>
Reviewed by: Rob Gittins <rob.gittins@nexenta.com>
NEX-5667 nfssrv_stats_flags does not work for aggregated kstats
NEX-4472 nfsauth_retrieve() flood caused by NFS clients with personal identity problems
Reviewed by: Yuri Pankov <yuri.pankov@nexenta.com>
NEX-2345 nfsauth_cache_get() could spend a lot of time walking exi_cache
Reviewed by: Gordon Ross <gordon.ross@nexenta.com>
NEX-2972 bad free at checkauth+0x1a2()
Reviewed by: Jan Kryl <jan.kryl@nexenta.com>
Reviewed by: Josef 'Jeff' Sipek <josef.sipek@nexenta.com>
NEX-2949 Panic due to bad mutex, from auth_cache being previously freed
NEX-1974 Support for more than 16 groups with AUTH_SYS
Reviewed by: Gordon Ross <gordon.ross@nexenta.com>
Reviewed by: Josef 'Jeff' Sipek <josef.sipek@nexenta.com>
NEX-1128 NFS server: Generic uid and gid remapping for AUTH_SYS
Reviewed by: Jan Kryl <jan.kryl@nexenta.com>
OS-152 NFS extremely slow in nfsauth
Reviewed by: Jan Kryl <jan.kryl@nexenta.com>
Reviewed by: Ilya Usvyatsky <ilya.usvyatsky@nexenta.com>
Reviewed by: Josef Sipek <josef.sipek@nexenta.com>


   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  */
  27 
  28 #include <sys/param.h>
  29 #include <sys/errno.h>
  30 #include <sys/vfs.h>
  31 #include <sys/vnode.h>
  32 #include <sys/cred.h>
  33 #include <sys/cmn_err.h>
  34 #include <sys/systm.h>
  35 #include <sys/kmem.h>
  36 #include <sys/pathname.h>
  37 #include <sys/utsname.h>
  38 #include <sys/debug.h>
  39 #include <sys/door.h>
  40 #include <sys/sdt.h>
  41 #include <sys/thread.h>
  42 #include <sys/avl.h>
  43 
  44 #include <rpc/types.h>
  45 #include <rpc/auth.h>
  46 #include <rpc/clnt.h>
  47 
  48 #include <nfs/nfs.h>
  49 #include <nfs/export.h>
  50 #include <nfs/nfs_clnt.h>
  51 #include <nfs/auth.h>
  52 
  53 static struct kmem_cache *exi_cache_handle;
  54 static void exi_cache_reclaim(void *);
  55 static void exi_cache_trim(struct exportinfo *exi);



  56 
  57 extern pri_t minclsyspri;
  58 

  59 volatile uint_t nfsauth_cache_hit;
  60 volatile uint_t nfsauth_cache_miss;
  61 volatile uint_t nfsauth_cache_refresh;
  62 volatile uint_t nfsauth_cache_reclaim;
  63 volatile uint_t exi_cache_auth_reclaim_failed;
  64 volatile uint_t exi_cache_clnt_reclaim_failed;
  65 
  66 /*
  67  * The lifetime of an auth cache entry:
  68  * ------------------------------------
  69  *
  70  * An auth cache entry is created with both the auth_time
  71  * and auth_freshness times set to the current time.
  72  *
  73  * Upon every client access which results in a hit, the
  74  * auth_time will be updated.
  75  *
  76  * If a client access determines that the auth_freshness
  77  * indicates that the entry is STALE, then it will be
  78  * refreshed. Note that this will explicitly reset


 102 
 103 /*
 104  * While we could encapuslate the exi_list inside the
 105  * exi structure, we can't do that for the auth_list.
 106  * So, to keep things looking clean, we keep them both
 107  * in these external lists.
 108  */
 109 typedef struct refreshq_exi_node {
 110         struct exportinfo       *ren_exi;
 111         list_t                  ren_authlist;
 112         list_node_t             ren_node;
 113 } refreshq_exi_node_t;
 114 
 115 typedef struct refreshq_auth_node {
 116         struct auth_cache       *ran_auth;
 117         char                    *ran_netid;
 118         list_node_t             ran_node;
 119 } refreshq_auth_node_t;
 120 
 121 /*
 122  * Used to manipulate things on the refreshq_queue.
 123  * Note that the refresh thread will effectively
 124  * pop a node off of the queue, at which point it
 125  * will no longer need to hold the mutex.
 126  */
 127 static kmutex_t refreshq_lock;
 128 static list_t refreshq_queue;
 129 static kcondvar_t refreshq_cv;
 130 
 131 /*
 132  * If there is ever a problem with loading the
 133  * module, then nfsauth_fini() needs to be called
 134  * to remove state. In that event, since the
 135  * refreshq thread has been started, they need to
 136  * 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 } nfsauth_refreshq_thread_state_t;
 143 
 144 nfsauth_refreshq_thread_state_t
 145 refreshq_thread_state = REFRESHQ_THREAD_HALTED;

 146 
























 147 static void nfsauth_free_node(struct auth_cache *);
 148 static void nfsauth_refresh_thread(void);
 149 
 150 static int nfsauth_cache_compar(const void *, const void *);
 151 
 152 /*
 153  * mountd is a server-side only daemon. This will need to be
 154  * revisited if the NFS server is ever made zones-aware.
 155  */
 156 kmutex_t        mountd_lock;
 157 door_handle_t   mountd_dh;
 158 
 159 void
 160 mountd_args(uint_t did)
 161 {
 162         mutex_enter(&mountd_lock);
 163         if (mountd_dh != NULL)
 164                 door_ki_rele(mountd_dh);
 165         mountd_dh = door_ki_lookup(did);
 166         mutex_exit(&mountd_lock);



 167 }
 168 
 169 void
 170 nfsauth_init(void)
 171 {
 172         /*
 173          * mountd can be restarted by smf(5). We need to make sure
 174          * the updated door handle will safely make it to mountd_dh
 175          */
 176         mutex_init(&mountd_lock, NULL, MUTEX_DEFAULT, NULL);
 177 
 178         mutex_init(&refreshq_lock, NULL, MUTEX_DEFAULT, NULL);
 179         list_create(&refreshq_queue, sizeof (refreshq_exi_node_t),
 180             offsetof(refreshq_exi_node_t, ren_node));
 181 
 182         cv_init(&refreshq_cv, NULL, CV_DEFAULT, NULL);
 183 
 184         /*
 185          * Allocate nfsauth cache handle
 186          */
 187         exi_cache_handle = kmem_cache_create("exi_cache_handle",
 188             sizeof (struct auth_cache), 0, NULL, NULL,
 189             exi_cache_reclaim, NULL, NULL, 0);
 190 
 191         refreshq_thread_state = REFRESHQ_THREAD_RUNNING;
 192         (void) zthread_create(NULL, 0, nfsauth_refresh_thread,
 193             NULL, 0, minclsyspri);
 194 }
 195 
 196 /*
 197  * Finalization routine for nfsauth. It is important to call this routine
 198  * before destroying the exported_lock.
 199  */
 200 void
 201 nfsauth_fini(void)
 202 {
 203         refreshq_exi_node_t     *ren;

 204 
 205         /*
 206          * Prevent the nfsauth_refresh_thread from getting new
 207          * work.
 208          */
 209         mutex_enter(&refreshq_lock);
 210         if (refreshq_thread_state != REFRESHQ_THREAD_HALTED) {
 211                 refreshq_thread_state = REFRESHQ_THREAD_FINI_REQ;
 212                 cv_broadcast(&refreshq_cv);
 213 


 214                 /*
 215                  * Also, wait for nfsauth_refresh_thread() to exit.

 216                  */
 217                 while (refreshq_thread_state != REFRESHQ_THREAD_HALTED) {
 218                         cv_wait(&refreshq_cv, &refreshq_lock);
























 219                 }
 220         }
 221         mutex_exit(&refreshq_lock);
 222 
 223         /*
 224          * Walk the exi_list and in turn, walk the auth_lists and free all
 225          * lists.  In addition, free INVALID auth_cache entries.
 226          */
 227         while ((ren = list_remove_head(&refreshq_queue))) {
 228                 refreshq_auth_node_t *ran;
 229 
 230                 while ((ran = list_remove_head(&ren->ren_authlist)) != NULL) {
 231                         struct auth_cache *p = ran->ran_auth;
 232                         if (p->auth_state == NFS_AUTH_INVALID)
 233                                 nfsauth_free_node(p);
 234                         strfree(ran->ran_netid);
 235                         kmem_free(ran, sizeof (refreshq_auth_node_t));
 236                 }
 237 
 238                 list_destroy(&ren->ren_authlist);
 239                 exi_rele(ren->ren_exi);
 240                 kmem_free(ren, sizeof (refreshq_exi_node_t));
 241         }
 242         list_destroy(&refreshq_queue);
 243 
 244         cv_destroy(&refreshq_cv);
 245         mutex_destroy(&refreshq_lock);



 246 
 247         mutex_destroy(&mountd_lock);
 248 
 249         /*
 250          * Deallocate nfsauth cache handle
 251          */
 252         kmem_cache_destroy(exi_cache_handle);
 253 }
 254 
 255 /*
 256  * Convert the address in a netbuf to
 257  * a hash index for the auth_cache table.
 258  */
 259 static int
 260 hash(struct netbuf *a)
 261 {
 262         int i, h = 0;
 263 
 264         for (i = 0; i < a->len; i++)
 265                 h ^= a->buf[i];
 266 
 267         return (h & (AUTH_TABLESIZE - 1));
 268 }
 269 
 270 /*
 271  * Mask out the components of an
 272  * address that do not identify


 324 static void
 325 sys_log(const char *msg)
 326 {
 327         static time_t   tstamp = 0;
 328         time_t          now;
 329 
 330         /*
 331          * msg is shown (at most) once per minute
 332          */
 333         now = gethrestime_sec();
 334         if ((tstamp + 60) < now) {
 335                 tstamp = now;
 336                 cmn_err(CE_WARN, msg);
 337         }
 338 }
 339 
 340 /*
 341  * Callup to the mountd to get access information in the kernel.
 342  */
 343 static bool_t
 344 nfsauth_retrieve(struct exportinfo *exi, char *req_netid, int flavor,
 345     struct netbuf *addr, int *access, cred_t *clnt_cred, uid_t *srv_uid,
 346     gid_t *srv_gid, uint_t *srv_gids_cnt, gid_t **srv_gids)

 347 {
 348         varg_t                    varg = {0};
 349         nfsauth_res_t             res = {0};
 350         XDR                       xdrs;
 351         size_t                    absz;
 352         caddr_t                   abuf;
 353         int                       last = 0;
 354         door_arg_t                da;
 355         door_info_t               di;
 356         door_handle_t             dh;
 357         uint_t                    ntries = 0;
 358 
 359         /*
 360          * No entry in the cache for this client/flavor
 361          * so we need to call the nfsauth service in the
 362          * mount daemon.
 363          */
 364 
 365         varg.vers = V_PROTO;
 366         varg.arg_u.arg.cmd = NFSAUTH_ACCESS;


 399         XDR_DESTROY(&xdrs);
 400 
 401         /*
 402          * Prepare the door arguments
 403          *
 404          * We don't know the size of the message the daemon
 405          * will pass back to us.  By setting rbuf to NULL,
 406          * we force the door code to allocate a buf of the
 407          * appropriate size.  We must set rsize > 0, however,
 408          * else the door code acts as if no response was
 409          * expected and doesn't pass the data to us.
 410          */
 411         da.data_ptr = (char *)abuf;
 412         da.data_size = absz;
 413         da.desc_ptr = NULL;
 414         da.desc_num = 0;
 415         da.rbuf = NULL;
 416         da.rsize = 1;
 417 
 418 retry:
 419         mutex_enter(&mountd_lock);
 420         dh = mountd_dh;
 421         if (dh != NULL)
 422                 door_ki_hold(dh);
 423         mutex_exit(&mountd_lock);
 424 
 425         if (dh == NULL) {
 426                 /*
 427                  * The rendezvous point has not been established yet!
 428                  * This could mean that either mountd(1m) has not yet
 429                  * been started or that _this_ routine nuked the door
 430                  * handle after receiving an EINTR for a REVOKED door.
 431                  *
 432                  * Returning NFSAUTH_DROP will cause the NFS client
 433                  * to retransmit the request, so let's try to be more
 434                  * rescillient and attempt for ntries before we bail.
 435                  */
 436                 if (++ntries % NFSAUTH_DR_TRYCNT) {
 437                         delay(hz);
 438                         goto retry;
 439                 }
 440 
 441                 kmem_free(abuf, absz);
 442 
 443                 sys_log("nfsauth: mountd has not established door");


 473                 /*
 474                  * Server out of resources; back off for a bit
 475                  */
 476                 door_ki_rele(dh);
 477                 delay(hz);
 478                 goto retry;
 479                 /* NOTREACHED */
 480 
 481         case EINTR:
 482                 if (!door_ki_info(dh, &di)) {
 483                         door_ki_rele(dh);
 484 
 485                         if (di.di_attributes & DOOR_REVOKED) {
 486                                 /*
 487                                  * The server barfed and revoked
 488                                  * the (existing) door on us; we
 489                                  * want to wait to give smf(5) a
 490                                  * chance to restart mountd(1m)
 491                                  * and establish a new door handle.
 492                                  */
 493                                 mutex_enter(&mountd_lock);
 494                                 if (dh == mountd_dh) {
 495                                         door_ki_rele(mountd_dh);
 496                                         mountd_dh = NULL;
 497                                 }
 498                                 mutex_exit(&mountd_lock);
 499                                 delay(hz);
 500                                 goto retry;
 501                         }
 502                         /*
 503                          * If the door was _not_ revoked on us,
 504                          * then more than likely we took an INTR,
 505                          * so we need to fail the operation.
 506                          */
 507                         goto fail;
 508                 }
 509                 /*
 510                  * The only failure that can occur from getting
 511                  * the door info is EINVAL, so we let the code
 512                  * below handle it.
 513                  */
 514                 /* FALLTHROUGH */
 515 
 516         case EBADF:
 517         case EINVAL:
 518         default:


 570 
 571                 case NFSAUTH_DR_EFAIL:
 572                 case NFSAUTH_DR_DECERR:
 573                 case NFSAUTH_DR_BADCMD:
 574                 default:
 575                         xdr_free(xdr_nfsauth_res, (char *)&res);
 576 fail:
 577                         *access = NFSAUTH_DENIED;
 578                         kmem_free(abuf, absz);
 579                         return (FALSE);
 580                         /* NOTREACHED */
 581         }
 582 
 583         xdr_free(xdr_nfsauth_res, (char *)&res);
 584         kmem_free(abuf, absz);
 585 
 586         return (TRUE);
 587 }
 588 
 589 static void
 590 nfsauth_refresh_thread(void)
 591 {
 592         refreshq_exi_node_t     *ren;
 593         refreshq_auth_node_t    *ran;
 594 
 595         struct exportinfo       *exi;
 596 
 597         int                     access;
 598         bool_t                  retrieval;
 599 
 600         callb_cpr_t             cprinfo;
 601 
 602         CALLB_CPR_INIT(&cprinfo, &refreshq_lock, callb_generic_cpr,
 603             "nfsauth_refresh");
 604 
 605         for (;;) {
 606                 mutex_enter(&refreshq_lock);
 607                 if (refreshq_thread_state != REFRESHQ_THREAD_RUNNING) {
 608                         /* Keep the hold on the lock! */
 609                         break;
 610                 }
 611 
 612                 ren = list_remove_head(&refreshq_queue);
 613                 if (ren == NULL) {
 614                         CALLB_CPR_SAFE_BEGIN(&cprinfo);
 615                         cv_wait(&refreshq_cv, &refreshq_lock);
 616                         CALLB_CPR_SAFE_END(&cprinfo, &refreshq_lock);
 617                         mutex_exit(&refreshq_lock);
 618                         continue;
 619                 }
 620                 mutex_exit(&refreshq_lock);
 621 
 622                 exi = ren->ren_exi;
 623                 ASSERT(exi != NULL);
 624 
 625                 /*
 626                  * Since the ren was removed from the refreshq_queue above,
 627                  * this is the only thread aware about the ren existence, so we
 628                  * have the exclusive ownership of it and we do not need to
 629                  * protect it by any lock.
 630                  */
 631                 while ((ran = list_remove_head(&ren->ren_authlist))) {
 632                         uid_t uid;
 633                         gid_t gid;
 634                         uint_t ngids;
 635                         gid_t *gids;
 636                         struct auth_cache *p = ran->ran_auth;
 637                         char *netid = ran->ran_netid;
 638 
 639                         ASSERT(p != NULL);
 640                         ASSERT(netid != NULL);


 647                          * Once the entry goes INVALID, it can not change
 648                          * state.
 649                          *
 650                          * No need to refresh entries also in a case we are
 651                          * just shutting down.
 652                          *
 653                          * In general, there is no need to hold the
 654                          * refreshq_lock to test the refreshq_thread_state.  We
 655                          * do hold it at other places because there is some
 656                          * related thread synchronization (or some other tasks)
 657                          * close to the refreshq_thread_state check.
 658                          *
 659                          * The check for the refreshq_thread_state value here
 660                          * is purely advisory to allow the faster
 661                          * nfsauth_refresh_thread() shutdown.  In a case we
 662                          * will miss such advisory, nothing catastrophic
 663                          * happens: we will just spin longer here before the
 664                          * shutdown.
 665                          */
 666                         if (p->auth_state == NFS_AUTH_INVALID ||
 667                             refreshq_thread_state != REFRESHQ_THREAD_RUNNING) {

 668                                 mutex_exit(&p->auth_lock);
 669 
 670                                 if (p->auth_state == NFS_AUTH_INVALID)
 671                                         nfsauth_free_node(p);
 672 
 673                                 strfree(netid);
 674 
 675                                 continue;
 676                         }
 677 
 678                         /*
 679                          * Make sure the state is valid.  Note that once we
 680                          * change the state to NFS_AUTH_REFRESHING, no other
 681                          * thread will be able to work on this entry.
 682                          */
 683                         ASSERT(p->auth_state == NFS_AUTH_STALE);
 684 
 685                         p->auth_state = NFS_AUTH_REFRESHING;
 686                         mutex_exit(&p->auth_lock);
 687 
 688                         DTRACE_PROBE2(nfsauth__debug__cache__refresh,
 689                             struct exportinfo *, exi,
 690                             struct auth_cache *, p);
 691 
 692                         /*
 693                          * The first caching of the access rights
 694                          * is done with the netid pulled out of the
 695                          * request from the client. All subsequent
 696                          * users of the cache may or may not have
 697                          * the same netid. It doesn't matter. So
 698                          * when we refresh, we simply use the netid
 699                          * of the request which triggered the
 700                          * refresh attempt.
 701                          */
 702                         retrieval = nfsauth_retrieve(exi, netid,
 703                             p->auth_flavor, &p->auth_clnt->authc_addr, &access,
 704                             p->auth_clnt_cred, &uid, &gid, &ngids, &gids);
 705 
 706                         /*
 707                          * This can only be set in one other place
 708                          * and the state has to be NFS_AUTH_FRESH.
 709                          */
 710                         strfree(netid);
 711 
 712                         mutex_enter(&p->auth_lock);
 713                         if (p->auth_state == NFS_AUTH_INVALID) {
 714                                 mutex_exit(&p->auth_lock);
 715                                 nfsauth_free_node(p);
 716                                 if (retrieval == TRUE)
 717                                         kmem_free(gids, ngids * sizeof (gid_t));
 718                         } else {
 719                                 /*
 720                                  * If we got an error, do not reset the
 721                                  * time. This will cause the next access
 722                                  * check for the client to reschedule this


 725                                 if (retrieval == TRUE) {
 726                                         p->auth_access = access;
 727 
 728                                         p->auth_srv_uid = uid;
 729                                         p->auth_srv_gid = gid;
 730                                         kmem_free(p->auth_srv_gids,
 731                                             p->auth_srv_ngids * sizeof (gid_t));
 732                                         p->auth_srv_ngids = ngids;
 733                                         p->auth_srv_gids = gids;
 734 
 735                                         p->auth_freshness = gethrestime_sec();
 736                                 }
 737                                 p->auth_state = NFS_AUTH_FRESH;
 738 
 739                                 cv_broadcast(&p->auth_cv);
 740                                 mutex_exit(&p->auth_lock);
 741                         }
 742                 }
 743 
 744                 list_destroy(&ren->ren_authlist);
 745                 exi_rele(ren->ren_exi);
 746                 kmem_free(ren, sizeof (refreshq_exi_node_t));
 747         }
 748 
 749         refreshq_thread_state = REFRESHQ_THREAD_HALTED;
 750         cv_broadcast(&refreshq_cv);
 751         CALLB_CPR_EXIT(&cprinfo);

 752         zthread_exit();
 753 }
 754 
 755 int
 756 nfsauth_cache_clnt_compar(const void *v1, const void *v2)
 757 {
 758         int c;
 759 
 760         const struct auth_cache_clnt *a1 = (const struct auth_cache_clnt *)v1;
 761         const struct auth_cache_clnt *a2 = (const struct auth_cache_clnt *)v2;
 762 
 763         if (a1->authc_addr.len < a2->authc_addr.len)
 764                 return (-1);
 765         if (a1->authc_addr.len > a2->authc_addr.len)
 766                 return (1);
 767 
 768         c = memcmp(a1->authc_addr.buf, a2->authc_addr.buf, a1->authc_addr.len);
 769         if (c < 0)
 770                 return (-1);
 771         if (c > 0)


 803                 return (1);
 804 
 805         c = memcmp(crgetgroups(a1->auth_clnt_cred),
 806             crgetgroups(a2->auth_clnt_cred), crgetngroups(a1->auth_clnt_cred));
 807         if (c < 0)
 808                 return (-1);
 809         if (c > 0)
 810                 return (1);
 811 
 812         return (0);
 813 }
 814 
 815 /*
 816  * Get the access information from the cache or callup to the mountd
 817  * to get and cache the access information in the kernel.
 818  */
 819 static int
 820 nfsauth_cache_get(struct exportinfo *exi, struct svc_req *req, int flavor,
 821     cred_t *cr, uid_t *uid, gid_t *gid, uint_t *ngids, gid_t **gids)
 822 {

 823         struct netbuf           *taddrmask;
 824         struct netbuf           addr;   /* temporary copy of client's address */
 825         const struct netbuf     *claddr;
 826         avl_tree_t              *tree;
 827         struct auth_cache       ac;     /* used as a template for avl_find() */
 828         struct auth_cache_clnt  *c;
 829         struct auth_cache_clnt  acc;    /* used as a template for avl_find() */
 830         struct auth_cache       *p = NULL;
 831         int                     access;
 832 
 833         uid_t                   tmpuid;
 834         gid_t                   tmpgid;
 835         uint_t                  tmpngids;
 836         gid_t                   *tmpgids;
 837 
 838         avl_index_t             where;  /* used for avl_find()/avl_insert() */
 839 
 840         ASSERT(cr != NULL);
 841 


 842         /*
 843          * Now check whether this client already
 844          * has an entry for this flavor in the cache
 845          * for this export.
 846          * Get the caller's address, mask off the
 847          * parts of the address that do not identify
 848          * the host (port number, etc), and then hash
 849          * it to find the chain of cache entries.
 850          */
 851 
 852         claddr = svc_getrpccaller(req->rq_xprt);
 853         addr = *claddr;
 854         addr.buf = kmem_alloc(addr.maxlen, KM_SLEEP);
 855         bcopy(claddr->buf, addr.buf, claddr->len);
 856 
 857         SVC_GETADDRMASK(req->rq_xprt, SVC_TATTR_ADDRMASK, (void **)&taddrmask);
 858         ASSERT(taddrmask != NULL);
 859         addrmask(&addr, taddrmask);
 860 
 861         ac.auth_flavor = flavor;


 981 
 982         /*
 983          * If the cache entry is not valid yet, we need to retrieve the
 984          * info ourselves.
 985          */
 986         if (p->auth_state == NFS_AUTH_NEW) {
 987                 bool_t res;
 988                 /*
 989                  * NFS_AUTH_NEW is the default output auth_state value in a
 990                  * case we failed somewhere below.
 991                  */
 992                 auth_state_t state = NFS_AUTH_NEW;
 993 
 994                 p->auth_state = NFS_AUTH_WAITING;
 995                 mutex_exit(&p->auth_lock);
 996                 kmem_free(addr.buf, addr.maxlen);
 997                 addr = p->auth_clnt->authc_addr;
 998 
 999                 atomic_inc_uint(&nfsauth_cache_miss);
1000 
1001                 res = nfsauth_retrieve(exi, svc_getnetid(req->rq_xprt), flavor,
1002                     &addr, &access, cr, &tmpuid, &tmpgid, &tmpngids, &tmpgids);

1003 
1004                 p->auth_access = access;
1005                 p->auth_time = p->auth_freshness = gethrestime_sec();
1006 
1007                 if (res == TRUE) {
1008                         if (uid != NULL)
1009                                 *uid = tmpuid;
1010                         if (gid != NULL)
1011                                 *gid = tmpgid;
1012                         if (ngids != NULL && gids != NULL) {
1013                                 *ngids = tmpngids;
1014                                 *gids = tmpgids;
1015 
1016                                 /*
1017                                  * We need a copy of gids for the
1018                                  * auth_cache entry
1019                                  */
1020                                 tmpgids = kmem_alloc(tmpngids * sizeof (gid_t),
1021                                     KM_NOSLEEP | KM_NORMALPRI);
1022                                 if (tmpgids != NULL)


1063 
1064                 if ((refresh > NFSAUTH_CACHE_REFRESH) &&
1065                     p->auth_state == NFS_AUTH_FRESH) {
1066                         refreshq_auth_node_t *ran;
1067                         uint_t nacr;
1068 
1069                         p->auth_state = NFS_AUTH_STALE;
1070                         mutex_exit(&p->auth_lock);
1071 
1072                         nacr = atomic_inc_uint_nv(&nfsauth_cache_refresh);
1073                         DTRACE_PROBE3(nfsauth__debug__cache__stale,
1074                             struct exportinfo *, exi,
1075                             struct auth_cache *, p,
1076                             uint_t, nacr);
1077 
1078                         ran = kmem_alloc(sizeof (refreshq_auth_node_t),
1079                             KM_SLEEP);
1080                         ran->ran_auth = p;
1081                         ran->ran_netid = strdup(svc_getnetid(req->rq_xprt));
1082 
1083                         mutex_enter(&refreshq_lock);











1084                         /*
1085                          * We should not add a work queue
1086                          * item if the thread is not
1087                          * accepting them.
1088                          */
1089                         if (refreshq_thread_state == REFRESHQ_THREAD_RUNNING) {

1090                                 refreshq_exi_node_t *ren;
1091 
1092                                 /*
1093                                  * Is there an existing exi_list?
1094                                  */
1095                                 for (ren = list_head(&refreshq_queue);
1096                                     ren != NULL;
1097                                     ren = list_next(&refreshq_queue, ren)) {

1098                                         if (ren->ren_exi == exi) {
1099                                                 list_insert_tail(
1100                                                     &ren->ren_authlist, ran);
1101                                                 break;
1102                                         }
1103                                 }
1104 
1105                                 if (ren == NULL) {
1106                                         ren = kmem_alloc(
1107                                             sizeof (refreshq_exi_node_t),
1108                                             KM_SLEEP);
1109 
1110                                         exi_hold(exi);
1111                                         ren->ren_exi = exi;
1112 
1113                                         list_create(&ren->ren_authlist,
1114                                             sizeof (refreshq_auth_node_t),
1115                                             offsetof(refreshq_auth_node_t,
1116                                             ran_node));
1117 
1118                                         list_insert_tail(&ren->ren_authlist,
1119                                             ran);
1120                                         list_insert_tail(&refreshq_queue, ren);

1121                                 }
1122 
1123                                 cv_broadcast(&refreshq_cv);
1124                         } else {
1125                                 strfree(ran->ran_netid);
1126                                 kmem_free(ran, sizeof (refreshq_auth_node_t));
1127                         }
1128 
1129                         mutex_exit(&refreshq_lock);
1130                 } else {
1131                         mutex_exit(&p->auth_lock);
1132                 }
1133 
1134                 nach = atomic_inc_uint_nv(&nfsauth_cache_hit);
1135                 DTRACE_PROBE2(nfsauth__debug__cache__hit,
1136                     uint_t, nach,
1137                     time_t, refresh);
1138 
1139                 kmem_free(addr.buf, addr.maxlen);
1140         }
1141 
1142         return (access);
1143 
1144 retrieve:
1145         crfree(ac.auth_clnt_cred);
1146 
1147         /*
1148          * Retrieve the required data without caching.
1149          */
1150 
1151         ASSERT(p == NULL);
1152 
1153         atomic_inc_uint(&nfsauth_cache_miss);
1154 
1155         if (nfsauth_retrieve(exi, svc_getnetid(req->rq_xprt), flavor, &addr,
1156             &access, cr, &tmpuid, &tmpgid, &tmpngids, &tmpgids)) {
1157                 if (uid != NULL)
1158                         *uid = tmpuid;
1159                 if (gid != NULL)
1160                         *gid = tmpgid;
1161                 if (ngids != NULL && gids != NULL) {
1162                         *ngids = tmpngids;
1163                         *gids = tmpgids;
1164                 } else {
1165                         kmem_free(tmpgids, tmpngids * sizeof (gid_t));
1166                 }
1167         }
1168 
1169         kmem_free(addr.buf, addr.maxlen);
1170 
1171         return (access);
1172 }
1173 
1174 /*
1175  * Check if the requesting client has access to the filesystem with
1176  * a given nfs flavor number which is an explicitly shared flavor.


1394                 void *cookie = NULL;
1395                 struct auth_cache_clnt *node;
1396 
1397                 while ((node = avl_destroy_nodes(tree, &cookie)) != NULL)
1398                         nfsauth_free_clnt_node(node);
1399         }
1400 }
1401 
1402 /*
1403  * Called by the kernel memory allocator when
1404  * memory is low. Free unused cache entries.
1405  * If that's not enough, the VM system will
1406  * call again for some more.
1407  */
1408 /*ARGSUSED*/
1409 void
1410 exi_cache_reclaim(void *cdrarg)
1411 {
1412         int i;
1413         struct exportinfo *exi;

1414 
1415         rw_enter(&exported_lock, RW_READER);
1416 
1417         for (i = 0; i < EXPTABLESIZE; i++) {
1418                 for (exi = exptable[i]; exi; exi = exi->fid_hash.next) {
1419                         exi_cache_trim(exi);
1420                 }
1421         }
1422 
1423         rw_exit(&exported_lock);
1424 
1425         atomic_inc_uint(&nfsauth_cache_reclaim);
1426 }
1427 
1428 void
1429 exi_cache_trim(struct exportinfo *exi)
1430 {
1431         struct auth_cache_clnt *c;
1432         struct auth_cache_clnt *nextc;
1433         struct auth_cache *p;
1434         struct auth_cache *next;
1435         int i;
1436         time_t stale_time;
1437         avl_tree_t *tree;
1438 
1439         for (i = 0; i < AUTH_TABLESIZE; i++) {
1440                 tree = exi->exi_cache[i];
1441                 stale_time = gethrestime_sec() - NFSAUTH_CACHE_TRIM;
1442                 rw_enter(&exi->exi_cache_lock, RW_READER);
1443 




   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  */
  25 
  26 /*
  27  * Copyright 2018 Nexenta Systems, Inc.
  28  * Copyright (c) 2015 by Delphix. All rights reserved.
  29  */
  30 
  31 #include <sys/param.h>
  32 #include <sys/errno.h>
  33 #include <sys/vfs.h>
  34 #include <sys/vnode.h>
  35 #include <sys/cred.h>
  36 #include <sys/cmn_err.h>
  37 #include <sys/systm.h>
  38 #include <sys/kmem.h>
  39 #include <sys/pathname.h>
  40 #include <sys/utsname.h>
  41 #include <sys/debug.h>
  42 #include <sys/door.h>
  43 #include <sys/sdt.h>
  44 #include <sys/thread.h>
  45 #include <sys/avl.h>
  46 
  47 #include <rpc/types.h>
  48 #include <rpc/auth.h>
  49 #include <rpc/clnt.h>
  50 
  51 #include <nfs/nfs.h>
  52 #include <nfs/export.h>
  53 #include <nfs/nfs_clnt.h>
  54 #include <nfs/auth.h>
  55 
  56 static struct kmem_cache *exi_cache_handle;
  57 static void exi_cache_reclaim(void *);
  58 static void exi_cache_trim(struct exportinfo *exi);
  59 static void *nfsauth_zone_init(zoneid_t);
  60 static void nfsauth_zone_shutdown(zoneid_t zoneid, void *data);
  61 static void nfsauth_zone_fini(zoneid_t, void *);
  62 
  63 extern pri_t minclsyspri;
  64 
  65 /* NFS auth cache statistics */
  66 volatile uint_t nfsauth_cache_hit;
  67 volatile uint_t nfsauth_cache_miss;
  68 volatile uint_t nfsauth_cache_refresh;
  69 volatile uint_t nfsauth_cache_reclaim;
  70 volatile uint_t exi_cache_auth_reclaim_failed;
  71 volatile uint_t exi_cache_clnt_reclaim_failed;
  72 
  73 /*
  74  * The lifetime of an auth cache entry:
  75  * ------------------------------------
  76  *
  77  * An auth cache entry is created with both the auth_time
  78  * and auth_freshness times set to the current time.
  79  *
  80  * Upon every client access which results in a hit, the
  81  * auth_time will be updated.
  82  *
  83  * If a client access determines that the auth_freshness
  84  * indicates that the entry is STALE, then it will be
  85  * refreshed. Note that this will explicitly reset


 109 
 110 /*
 111  * While we could encapuslate the exi_list inside the
 112  * exi structure, we can't do that for the auth_list.
 113  * So, to keep things looking clean, we keep them both
 114  * in these external lists.
 115  */
 116 typedef struct refreshq_exi_node {
 117         struct exportinfo       *ren_exi;
 118         list_t                  ren_authlist;
 119         list_node_t             ren_node;
 120 } refreshq_exi_node_t;
 121 
 122 typedef struct refreshq_auth_node {
 123         struct auth_cache       *ran_auth;
 124         char                    *ran_netid;
 125         list_node_t             ran_node;
 126 } refreshq_auth_node_t;
 127 
 128 /*
 129  * Used to manipulate things on the refreshq_queue.  Note that the refresh
 130  * thread will effectively pop a node off of the queue, at which point it

 131  * will no longer need to hold the mutex.
 132  */
 133 static kmutex_t refreshq_lock;
 134 static list_t refreshq_queue;
 135 static kcondvar_t refreshq_cv;
 136 
 137 /*
 138  * If there is ever a problem with loading the module, then nfsauth_fini()
 139  * needs to be called to remove state.  In that event, since the refreshq
 140  * thread has been started, they need to work together to get rid of state.


 141  */
 142 typedef enum nfsauth_refreshq_thread_state {
 143         REFRESHQ_THREAD_RUNNING,
 144         REFRESHQ_THREAD_FINI_REQ,
 145         REFRESHQ_THREAD_HALTED,
 146         REFRESHQ_THREAD_NEED_CREATE
 147 } nfsauth_refreshq_thread_state_t;
 148 
 149 typedef struct nfsauth_globals {
 150         kmutex_t        mountd_lock;
 151         door_handle_t   mountd_dh;
 152 
 153         /*
 154          * Used to manipulate things on the refreshq_queue.  Note that the
 155          * refresh thread will effectively pop a node off of the queue,
 156          * at which point it will no longer need to hold the mutex.
 157          */
 158         kmutex_t        refreshq_lock;
 159         list_t          refreshq_queue;
 160         kcondvar_t      refreshq_cv;
 161 
 162         /*
 163          * A list_t would be overkill.  These are auth_cache entries which are
 164          * no longer linked to an exi.  It should be the case that all of their
 165          * states are NFS_AUTH_INVALID, i.e., the only way to be put on this
 166          * list is iff their state indicated that they had been placed on the
 167          * refreshq_queue.
 168          *
 169          * Note that while there is no link from the exi or back to the exi,
 170          * the exi can not go away until these entries are harvested.
 171          */
 172         struct auth_cache               *refreshq_dead_entries;
 173         nfsauth_refreshq_thread_state_t refreshq_thread_state;
 174 
 175 } nfsauth_globals_t;
 176 
 177 static void nfsauth_free_node(struct auth_cache *);
 178 static void nfsauth_refresh_thread(nfsauth_globals_t *);
 179 
 180 static int nfsauth_cache_compar(const void *, const void *);
 181 
 182 static zone_key_t       nfsauth_zone_key;





 183 
 184 void
 185 mountd_args(uint_t did)
 186 {
 187         nfsauth_globals_t *nag;
 188 
 189         nag = zone_getspecific(nfsauth_zone_key, curzone);
 190         mutex_enter(&nag->mountd_lock);
 191         if (nag->mountd_dh != NULL)
 192                 door_ki_rele(nag->mountd_dh);
 193         nag->mountd_dh = door_ki_lookup(did);
 194         mutex_exit(&nag->mountd_lock);
 195 }
 196 
 197 void
 198 nfsauth_init(void)
 199 {
 200         zone_key_create(&nfsauth_zone_key, nfsauth_zone_init,
 201             nfsauth_zone_shutdown, nfsauth_zone_fini);



 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 /*ARGSUSED*/
 215 static void *
 216 nfsauth_zone_init(zoneid_t zoneid)
 217 {
 218         nfsauth_globals_t *nag;



 219 
 220         nag = kmem_zalloc(sizeof (*nag), KM_SLEEP);
 221 
 222         /*
 223          * mountd can be restarted by smf(5).  We need to make sure
 224          * the updated door handle will safely make it to mountd_dh.
 225          */
 226         mutex_init(&nag->mountd_lock, NULL, MUTEX_DEFAULT, NULL);
 227         mutex_init(&nag->refreshq_lock, NULL, MUTEX_DEFAULT, NULL);
 228         list_create(&nag->refreshq_queue, sizeof (refreshq_exi_node_t),
 229             offsetof(refreshq_exi_node_t, ren_node));
 230         cv_init(&nag->refreshq_cv, NULL, CV_DEFAULT, NULL);
 231         nag->refreshq_thread_state = REFRESHQ_THREAD_NEED_CREATE;
 232 
 233         return (nag);
 234 }
 235 
 236 /*ARGSUSED*/
 237 static void
 238 nfsauth_zone_shutdown(zoneid_t zoneid, void *data)
 239 {
 240         refreshq_exi_node_t     *ren;
 241         nfsauth_globals_t       *nag = data;
 242 
 243         /* Prevent the nfsauth_refresh_thread from getting new work */
 244         mutex_enter(&nag->refreshq_lock);
 245         if (nag->refreshq_thread_state == REFRESHQ_THREAD_RUNNING) {
 246                 nag->refreshq_thread_state = REFRESHQ_THREAD_FINI_REQ;
 247                 cv_broadcast(&nag->refreshq_cv);
 248 
 249                 /* Wait for nfsauth_refresh_thread() to exit */
 250                 while (nag->refreshq_thread_state != REFRESHQ_THREAD_HALTED)
 251                         cv_wait(&nag->refreshq_cv, &nag->refreshq_lock);
 252         }
 253         mutex_exit(&nag->refreshq_lock);

 254 
 255         /*
 256          * Walk the exi_list and in turn, walk the auth_lists and free all
 257          * lists.  In addition, free INVALID auth_cache entries.
 258          */
 259         while ((ren = list_remove_head(&nag->refreshq_queue))) {
 260                 refreshq_auth_node_t *ran;
 261 
 262                 while ((ran = list_remove_head(&ren->ren_authlist)) != NULL) {
 263                         struct auth_cache *p = ran->ran_auth;
 264                         if (p->auth_state == NFS_AUTH_INVALID)
 265                                 nfsauth_free_node(p);
 266                         strfree(ran->ran_netid);
 267                         kmem_free(ran, sizeof (*ran));
 268                 }
 269 
 270                 list_destroy(&ren->ren_authlist);
 271                 exi_rele(&ren->ren_exi);
 272                 kmem_free(ren, sizeof (*ren));
 273         }
 274 }
 275 
 276 /*ARGSUSED*/
 277 static void
 278 nfsauth_zone_fini(zoneid_t zoneid, void *data)
 279 {
 280         nfsauth_globals_t *nag = data;
 281 
 282         list_destroy(&nag->refreshq_queue);
 283         cv_destroy(&nag->refreshq_cv);
 284         mutex_destroy(&nag->refreshq_lock);
 285         mutex_destroy(&nag->mountd_lock);
 286         kmem_free(nag, sizeof (*nag));

 287 }
 288 
 289 /*
 290  * Convert the address in a netbuf to
 291  * a hash index for the auth_cache table.
 292  */
 293 static int
 294 hash(struct netbuf *a)
 295 {
 296         int i, h = 0;
 297 
 298         for (i = 0; i < a->len; i++)
 299                 h ^= a->buf[i];
 300 
 301         return (h & (AUTH_TABLESIZE - 1));
 302 }
 303 
 304 /*
 305  * Mask out the components of an
 306  * address that do not identify


 358 static void
 359 sys_log(const char *msg)
 360 {
 361         static time_t   tstamp = 0;
 362         time_t          now;
 363 
 364         /*
 365          * msg is shown (at most) once per minute
 366          */
 367         now = gethrestime_sec();
 368         if ((tstamp + 60) < now) {
 369                 tstamp = now;
 370                 cmn_err(CE_WARN, msg);
 371         }
 372 }
 373 
 374 /*
 375  * Callup to the mountd to get access information in the kernel.
 376  */
 377 static bool_t
 378 nfsauth_retrieve(nfsauth_globals_t *nag, struct exportinfo *exi,
 379     char *req_netid, int flavor, struct netbuf *addr, int *access,
 380     cred_t *clnt_cred, uid_t *srv_uid, gid_t *srv_gid, uint_t *srv_gids_cnt,
 381     gid_t **srv_gids)
 382 {
 383         varg_t                    varg = {0};
 384         nfsauth_res_t             res = {0};
 385         XDR                       xdrs;
 386         size_t                    absz;
 387         caddr_t                   abuf;
 388         int                       last = 0;
 389         door_arg_t                da;
 390         door_info_t               di;
 391         door_handle_t             dh;
 392         uint_t                    ntries = 0;
 393 
 394         /*
 395          * No entry in the cache for this client/flavor
 396          * so we need to call the nfsauth service in the
 397          * mount daemon.
 398          */
 399 
 400         varg.vers = V_PROTO;
 401         varg.arg_u.arg.cmd = NFSAUTH_ACCESS;


 434         XDR_DESTROY(&xdrs);
 435 
 436         /*
 437          * Prepare the door arguments
 438          *
 439          * We don't know the size of the message the daemon
 440          * will pass back to us.  By setting rbuf to NULL,
 441          * we force the door code to allocate a buf of the
 442          * appropriate size.  We must set rsize > 0, however,
 443          * else the door code acts as if no response was
 444          * expected and doesn't pass the data to us.
 445          */
 446         da.data_ptr = (char *)abuf;
 447         da.data_size = absz;
 448         da.desc_ptr = NULL;
 449         da.desc_num = 0;
 450         da.rbuf = NULL;
 451         da.rsize = 1;
 452 
 453 retry:
 454         mutex_enter(&nag->mountd_lock);
 455         dh = nag->mountd_dh;
 456         if (dh != NULL)
 457                 door_ki_hold(dh);
 458         mutex_exit(&nag->mountd_lock);
 459 
 460         if (dh == NULL) {
 461                 /*
 462                  * The rendezvous point has not been established yet!
 463                  * This could mean that either mountd(1m) has not yet
 464                  * been started or that _this_ routine nuked the door
 465                  * handle after receiving an EINTR for a REVOKED door.
 466                  *
 467                  * Returning NFSAUTH_DROP will cause the NFS client
 468                  * to retransmit the request, so let's try to be more
 469                  * rescillient and attempt for ntries before we bail.
 470                  */
 471                 if (++ntries % NFSAUTH_DR_TRYCNT) {
 472                         delay(hz);
 473                         goto retry;
 474                 }
 475 
 476                 kmem_free(abuf, absz);
 477 
 478                 sys_log("nfsauth: mountd has not established door");


 508                 /*
 509                  * Server out of resources; back off for a bit
 510                  */
 511                 door_ki_rele(dh);
 512                 delay(hz);
 513                 goto retry;
 514                 /* NOTREACHED */
 515 
 516         case EINTR:
 517                 if (!door_ki_info(dh, &di)) {
 518                         door_ki_rele(dh);
 519 
 520                         if (di.di_attributes & DOOR_REVOKED) {
 521                                 /*
 522                                  * The server barfed and revoked
 523                                  * the (existing) door on us; we
 524                                  * want to wait to give smf(5) a
 525                                  * chance to restart mountd(1m)
 526                                  * and establish a new door handle.
 527                                  */
 528                                 mutex_enter(&nag->mountd_lock);
 529                                 if (dh == nag->mountd_dh) {
 530                                         door_ki_rele(nag->mountd_dh);
 531                                         nag->mountd_dh = NULL;
 532                                 }
 533                                 mutex_exit(&nag->mountd_lock);
 534                                 delay(hz);
 535                                 goto retry;
 536                         }
 537                         /*
 538                          * If the door was _not_ revoked on us,
 539                          * then more than likely we took an INTR,
 540                          * so we need to fail the operation.
 541                          */
 542                         goto fail;
 543                 }
 544                 /*
 545                  * The only failure that can occur from getting
 546                  * the door info is EINVAL, so we let the code
 547                  * below handle it.
 548                  */
 549                 /* FALLTHROUGH */
 550 
 551         case EBADF:
 552         case EINVAL:
 553         default:


 605 
 606                 case NFSAUTH_DR_EFAIL:
 607                 case NFSAUTH_DR_DECERR:
 608                 case NFSAUTH_DR_BADCMD:
 609                 default:
 610                         xdr_free(xdr_nfsauth_res, (char *)&res);
 611 fail:
 612                         *access = NFSAUTH_DENIED;
 613                         kmem_free(abuf, absz);
 614                         return (FALSE);
 615                         /* NOTREACHED */
 616         }
 617 
 618         xdr_free(xdr_nfsauth_res, (char *)&res);
 619         kmem_free(abuf, absz);
 620 
 621         return (TRUE);
 622 }
 623 
 624 static void
 625 nfsauth_refresh_thread(nfsauth_globals_t *nag)
 626 {
 627         refreshq_exi_node_t     *ren;
 628         refreshq_auth_node_t    *ran;
 629 
 630         struct exportinfo       *exi;
 631 
 632         int                     access;
 633         bool_t                  retrieval;
 634 
 635         callb_cpr_t             cprinfo;
 636 
 637         CALLB_CPR_INIT(&cprinfo, &nag->refreshq_lock, callb_generic_cpr,
 638             "nfsauth_refresh");
 639 
 640         for (;;) {
 641                 mutex_enter(&nag->refreshq_lock);
 642                 if (nag->refreshq_thread_state != REFRESHQ_THREAD_RUNNING) {
 643                         /* Keep the hold on the lock! */
 644                         break;
 645                 }
 646 
 647                 ren = list_remove_head(&nag->refreshq_queue);
 648                 if (ren == NULL) {
 649                         CALLB_CPR_SAFE_BEGIN(&cprinfo);
 650                         cv_wait(&nag->refreshq_cv, &nag->refreshq_lock);
 651                         CALLB_CPR_SAFE_END(&cprinfo, &nag->refreshq_lock);
 652                         mutex_exit(&nag->refreshq_lock);
 653                         continue;
 654                 }
 655                 mutex_exit(&nag->refreshq_lock);
 656 
 657                 exi = ren->ren_exi;
 658                 ASSERT(exi != NULL);
 659 
 660                 /*
 661                  * Since the ren was removed from the refreshq_queue above,
 662                  * this is the only thread aware about the ren existence, so we
 663                  * have the exclusive ownership of it and we do not need to
 664                  * protect it by any lock.
 665                  */
 666                 while ((ran = list_remove_head(&ren->ren_authlist))) {
 667                         uid_t uid;
 668                         gid_t gid;
 669                         uint_t ngids;
 670                         gid_t *gids;
 671                         struct auth_cache *p = ran->ran_auth;
 672                         char *netid = ran->ran_netid;
 673 
 674                         ASSERT(p != NULL);
 675                         ASSERT(netid != NULL);


 682                          * Once the entry goes INVALID, it can not change
 683                          * state.
 684                          *
 685                          * No need to refresh entries also in a case we are
 686                          * just shutting down.
 687                          *
 688                          * In general, there is no need to hold the
 689                          * refreshq_lock to test the refreshq_thread_state.  We
 690                          * do hold it at other places because there is some
 691                          * related thread synchronization (or some other tasks)
 692                          * close to the refreshq_thread_state check.
 693                          *
 694                          * The check for the refreshq_thread_state value here
 695                          * is purely advisory to allow the faster
 696                          * nfsauth_refresh_thread() shutdown.  In a case we
 697                          * will miss such advisory, nothing catastrophic
 698                          * happens: we will just spin longer here before the
 699                          * shutdown.
 700                          */
 701                         if (p->auth_state == NFS_AUTH_INVALID ||
 702                             nag->refreshq_thread_state !=
 703                             REFRESHQ_THREAD_RUNNING) {
 704                                 mutex_exit(&p->auth_lock);
 705 
 706                                 if (p->auth_state == NFS_AUTH_INVALID)
 707                                         nfsauth_free_node(p);
 708 
 709                                 strfree(netid);
 710 
 711                                 continue;
 712                         }
 713 
 714                         /*
 715                          * Make sure the state is valid.  Note that once we
 716                          * change the state to NFS_AUTH_REFRESHING, no other
 717                          * thread will be able to work on this entry.
 718                          */
 719                         ASSERT(p->auth_state == NFS_AUTH_STALE);
 720 
 721                         p->auth_state = NFS_AUTH_REFRESHING;
 722                         mutex_exit(&p->auth_lock);
 723 
 724                         DTRACE_PROBE2(nfsauth__debug__cache__refresh,
 725                             struct exportinfo *, exi,
 726                             struct auth_cache *, p);
 727 
 728                         /*
 729                          * The first caching of the access rights
 730                          * is done with the netid pulled out of the
 731                          * request from the client. All subsequent
 732                          * users of the cache may or may not have
 733                          * the same netid. It doesn't matter. So
 734                          * when we refresh, we simply use the netid
 735                          * of the request which triggered the
 736                          * refresh attempt.
 737                          */
 738                         retrieval = nfsauth_retrieve(nag, exi, netid,
 739                             p->auth_flavor, &p->auth_clnt->authc_addr, &access,
 740                             p->auth_clnt_cred, &uid, &gid, &ngids, &gids);
 741 
 742                         /*
 743                          * This can only be set in one other place
 744                          * and the state has to be NFS_AUTH_FRESH.
 745                          */
 746                         strfree(netid);
 747 
 748                         mutex_enter(&p->auth_lock);
 749                         if (p->auth_state == NFS_AUTH_INVALID) {
 750                                 mutex_exit(&p->auth_lock);
 751                                 nfsauth_free_node(p);
 752                                 if (retrieval == TRUE)
 753                                         kmem_free(gids, ngids * sizeof (gid_t));
 754                         } else {
 755                                 /*
 756                                  * If we got an error, do not reset the
 757                                  * time. This will cause the next access
 758                                  * check for the client to reschedule this


 761                                 if (retrieval == TRUE) {
 762                                         p->auth_access = access;
 763 
 764                                         p->auth_srv_uid = uid;
 765                                         p->auth_srv_gid = gid;
 766                                         kmem_free(p->auth_srv_gids,
 767                                             p->auth_srv_ngids * sizeof (gid_t));
 768                                         p->auth_srv_ngids = ngids;
 769                                         p->auth_srv_gids = gids;
 770 
 771                                         p->auth_freshness = gethrestime_sec();
 772                                 }
 773                                 p->auth_state = NFS_AUTH_FRESH;
 774 
 775                                 cv_broadcast(&p->auth_cv);
 776                                 mutex_exit(&p->auth_lock);
 777                         }
 778                 }
 779 
 780                 list_destroy(&ren->ren_authlist);
 781                 exi_rele(&ren->ren_exi);
 782                 kmem_free(ren, sizeof (refreshq_exi_node_t));
 783         }
 784 
 785         nag->refreshq_thread_state = REFRESHQ_THREAD_HALTED;
 786         cv_broadcast(&nag->refreshq_cv);
 787         CALLB_CPR_EXIT(&cprinfo);
 788         DTRACE_PROBE(nfsauth__nfsauth__refresh__thread__exit);
 789         zthread_exit();
 790 }
 791 
 792 int
 793 nfsauth_cache_clnt_compar(const void *v1, const void *v2)
 794 {
 795         int c;
 796 
 797         const struct auth_cache_clnt *a1 = (const struct auth_cache_clnt *)v1;
 798         const struct auth_cache_clnt *a2 = (const struct auth_cache_clnt *)v2;
 799 
 800         if (a1->authc_addr.len < a2->authc_addr.len)
 801                 return (-1);
 802         if (a1->authc_addr.len > a2->authc_addr.len)
 803                 return (1);
 804 
 805         c = memcmp(a1->authc_addr.buf, a2->authc_addr.buf, a1->authc_addr.len);
 806         if (c < 0)
 807                 return (-1);
 808         if (c > 0)


 840                 return (1);
 841 
 842         c = memcmp(crgetgroups(a1->auth_clnt_cred),
 843             crgetgroups(a2->auth_clnt_cred), crgetngroups(a1->auth_clnt_cred));
 844         if (c < 0)
 845                 return (-1);
 846         if (c > 0)
 847                 return (1);
 848 
 849         return (0);
 850 }
 851 
 852 /*
 853  * Get the access information from the cache or callup to the mountd
 854  * to get and cache the access information in the kernel.
 855  */
 856 static int
 857 nfsauth_cache_get(struct exportinfo *exi, struct svc_req *req, int flavor,
 858     cred_t *cr, uid_t *uid, gid_t *gid, uint_t *ngids, gid_t **gids)
 859 {
 860         nfsauth_globals_t       *nag;
 861         struct netbuf           *taddrmask;
 862         struct netbuf           addr;   /* temporary copy of client's address */
 863         const struct netbuf     *claddr;
 864         avl_tree_t              *tree;
 865         struct auth_cache       ac;     /* used as a template for avl_find() */
 866         struct auth_cache_clnt  *c;
 867         struct auth_cache_clnt  acc;    /* used as a template for avl_find() */
 868         struct auth_cache       *p = NULL;
 869         int                     access;
 870 
 871         uid_t                   tmpuid;
 872         gid_t                   tmpgid;
 873         uint_t                  tmpngids;
 874         gid_t                   *tmpgids;
 875 
 876         avl_index_t             where;  /* used for avl_find()/avl_insert() */
 877 
 878         ASSERT(cr != NULL);
 879 
 880         nag = zone_getspecific(nfsauth_zone_key, curzone);
 881 
 882         /*
 883          * Now check whether this client already
 884          * has an entry for this flavor in the cache
 885          * for this export.
 886          * Get the caller's address, mask off the
 887          * parts of the address that do not identify
 888          * the host (port number, etc), and then hash
 889          * it to find the chain of cache entries.
 890          */
 891 
 892         claddr = svc_getrpccaller(req->rq_xprt);
 893         addr = *claddr;
 894         addr.buf = kmem_alloc(addr.maxlen, KM_SLEEP);
 895         bcopy(claddr->buf, addr.buf, claddr->len);
 896 
 897         SVC_GETADDRMASK(req->rq_xprt, SVC_TATTR_ADDRMASK, (void **)&taddrmask);
 898         ASSERT(taddrmask != NULL);
 899         addrmask(&addr, taddrmask);
 900 
 901         ac.auth_flavor = flavor;


1021 
1022         /*
1023          * If the cache entry is not valid yet, we need to retrieve the
1024          * info ourselves.
1025          */
1026         if (p->auth_state == NFS_AUTH_NEW) {
1027                 bool_t res;
1028                 /*
1029                  * NFS_AUTH_NEW is the default output auth_state value in a
1030                  * case we failed somewhere below.
1031                  */
1032                 auth_state_t state = NFS_AUTH_NEW;
1033 
1034                 p->auth_state = NFS_AUTH_WAITING;
1035                 mutex_exit(&p->auth_lock);
1036                 kmem_free(addr.buf, addr.maxlen);
1037                 addr = p->auth_clnt->authc_addr;
1038 
1039                 atomic_inc_uint(&nfsauth_cache_miss);
1040 
1041                 res = nfsauth_retrieve(nag, exi, svc_getnetid(req->rq_xprt),
1042                     flavor, &addr, &access, cr, &tmpuid, &tmpgid, &tmpngids,
1043                     &tmpgids);
1044 
1045                 p->auth_access = access;
1046                 p->auth_time = p->auth_freshness = gethrestime_sec();
1047 
1048                 if (res == TRUE) {
1049                         if (uid != NULL)
1050                                 *uid = tmpuid;
1051                         if (gid != NULL)
1052                                 *gid = tmpgid;
1053                         if (ngids != NULL && gids != NULL) {
1054                                 *ngids = tmpngids;
1055                                 *gids = tmpgids;
1056 
1057                                 /*
1058                                  * We need a copy of gids for the
1059                                  * auth_cache entry
1060                                  */
1061                                 tmpgids = kmem_alloc(tmpngids * sizeof (gid_t),
1062                                     KM_NOSLEEP | KM_NORMALPRI);
1063                                 if (tmpgids != NULL)


1104 
1105                 if ((refresh > NFSAUTH_CACHE_REFRESH) &&
1106                     p->auth_state == NFS_AUTH_FRESH) {
1107                         refreshq_auth_node_t *ran;
1108                         uint_t nacr;
1109 
1110                         p->auth_state = NFS_AUTH_STALE;
1111                         mutex_exit(&p->auth_lock);
1112 
1113                         nacr = atomic_inc_uint_nv(&nfsauth_cache_refresh);
1114                         DTRACE_PROBE3(nfsauth__debug__cache__stale,
1115                             struct exportinfo *, exi,
1116                             struct auth_cache *, p,
1117                             uint_t, nacr);
1118 
1119                         ran = kmem_alloc(sizeof (refreshq_auth_node_t),
1120                             KM_SLEEP);
1121                         ran->ran_auth = p;
1122                         ran->ran_netid = strdup(svc_getnetid(req->rq_xprt));
1123 
1124                         mutex_enter(&nag->refreshq_lock);
1125 
1126                         if (nag->refreshq_thread_state ==
1127                             REFRESHQ_THREAD_NEED_CREATE) {
1128                                 /* Launch nfsauth refresh thread */
1129                                 nag->refreshq_thread_state =
1130                                     REFRESHQ_THREAD_RUNNING;
1131                                 (void) zthread_create(NULL, 0,
1132                                     nfsauth_refresh_thread, nag, 0,
1133                                     minclsyspri);
1134                         }
1135 
1136                         /*
1137                          * We should not add a work queue item if the thread
1138                          * is not accepting them.

1139                          */
1140                         if (nag->refreshq_thread_state ==
1141                             REFRESHQ_THREAD_RUNNING) {
1142                                 refreshq_exi_node_t *ren;
1143 
1144                                 /*
1145                                  * Is there an existing exi_list?
1146                                  */
1147                                 for (ren = list_head(&nag->refreshq_queue);
1148                                     ren != NULL;
1149                                     ren = list_next(&nag->refreshq_queue,
1150                                     ren)) {
1151                                         if (ren->ren_exi == exi) {
1152                                                 list_insert_tail(
1153                                                     &ren->ren_authlist, ran);
1154                                                 break;
1155                                         }
1156                                 }
1157 
1158                                 if (ren == NULL) {
1159                                         ren = kmem_alloc(
1160                                             sizeof (refreshq_exi_node_t),
1161                                             KM_SLEEP);
1162 
1163                                         exi_hold(exi);
1164                                         ren->ren_exi = exi;
1165 
1166                                         list_create(&ren->ren_authlist,
1167                                             sizeof (refreshq_auth_node_t),
1168                                             offsetof(refreshq_auth_node_t,
1169                                             ran_node));
1170 
1171                                         list_insert_tail(&ren->ren_authlist,
1172                                             ran);
1173                                         list_insert_tail(&nag->refreshq_queue,
1174                                             ren);
1175                                 }
1176 
1177                                 cv_broadcast(&nag->refreshq_cv);
1178                         } else {
1179                                 strfree(ran->ran_netid);
1180                                 kmem_free(ran, sizeof (refreshq_auth_node_t));
1181                         }
1182 
1183                         mutex_exit(&nag->refreshq_lock);
1184                 } else {
1185                         mutex_exit(&p->auth_lock);
1186                 }
1187 
1188                 nach = atomic_inc_uint_nv(&nfsauth_cache_hit);
1189                 DTRACE_PROBE2(nfsauth__debug__cache__hit,
1190                     uint_t, nach,
1191                     time_t, refresh);
1192 
1193                 kmem_free(addr.buf, addr.maxlen);
1194         }
1195 
1196         return (access);
1197 
1198 retrieve:
1199         crfree(ac.auth_clnt_cred);
1200 
1201         /*
1202          * Retrieve the required data without caching.
1203          */
1204 
1205         ASSERT(p == NULL);
1206 
1207         atomic_inc_uint(&nfsauth_cache_miss);
1208 
1209         if (nfsauth_retrieve(nag, exi, svc_getnetid(req->rq_xprt), flavor,
1210             &addr, &access, cr, &tmpuid, &tmpgid, &tmpngids, &tmpgids)) {
1211                 if (uid != NULL)
1212                         *uid = tmpuid;
1213                 if (gid != NULL)
1214                         *gid = tmpgid;
1215                 if (ngids != NULL && gids != NULL) {
1216                         *ngids = tmpngids;
1217                         *gids = tmpgids;
1218                 } else {
1219                         kmem_free(tmpgids, tmpngids * sizeof (gid_t));
1220                 }
1221         }
1222 
1223         kmem_free(addr.buf, addr.maxlen);
1224 
1225         return (access);
1226 }
1227 
1228 /*
1229  * Check if the requesting client has access to the filesystem with
1230  * a given nfs flavor number which is an explicitly shared flavor.


1448                 void *cookie = NULL;
1449                 struct auth_cache_clnt *node;
1450 
1451                 while ((node = avl_destroy_nodes(tree, &cookie)) != NULL)
1452                         nfsauth_free_clnt_node(node);
1453         }
1454 }
1455 
1456 /*
1457  * Called by the kernel memory allocator when
1458  * memory is low. Free unused cache entries.
1459  * If that's not enough, the VM system will
1460  * call again for some more.
1461  */
1462 /*ARGSUSED*/
1463 void
1464 exi_cache_reclaim(void *cdrarg)
1465 {
1466         int i;
1467         struct exportinfo *exi;
1468         nfs_export_t *ne = nfs_get_export();
1469 
1470         rw_enter(&ne->exported_lock, RW_READER);
1471 
1472         for (i = 0; i < EXPTABLESIZE; i++) {
1473                 for (exi = ne->exptable[i]; exi; exi = exi->fid_hash.next)
1474                         exi_cache_trim(exi);
1475         }

1476 
1477         rw_exit(&ne->exported_lock);
1478 
1479         atomic_inc_uint(&nfsauth_cache_reclaim);
1480 }
1481 
1482 void
1483 exi_cache_trim(struct exportinfo *exi)
1484 {
1485         struct auth_cache_clnt *c;
1486         struct auth_cache_clnt *nextc;
1487         struct auth_cache *p;
1488         struct auth_cache *next;
1489         int i;
1490         time_t stale_time;
1491         avl_tree_t *tree;
1492 
1493         for (i = 0; i < AUTH_TABLESIZE; i++) {
1494                 tree = exi->exi_cache[i];
1495                 stale_time = gethrestime_sec() - NFSAUTH_CACHE_TRIM;
1496                 rw_enter(&exi->exi_cache_lock, RW_READER);
1497