The Design and Implementation of the FreeBSD Operating System, Second Edition
Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]

FreeBSD/Linux Kernel Cross Reference
sys/nlm/nlm_prot_impl.c

Version: -  FREEBSD  -  FREEBSD-13-STABLE  -  FREEBSD-13-0  -  FREEBSD-12-STABLE  -  FREEBSD-12-0  -  FREEBSD-11-STABLE  -  FREEBSD-11-0  -  FREEBSD-10-STABLE  -  FREEBSD-10-0  -  FREEBSD-9-STABLE  -  FREEBSD-9-0  -  FREEBSD-8-STABLE  -  FREEBSD-8-0  -  FREEBSD-7-STABLE  -  FREEBSD-7-0  -  FREEBSD-6-STABLE  -  FREEBSD-6-0  -  FREEBSD-5-STABLE  -  FREEBSD-5-0  -  FREEBSD-4-STABLE  -  FREEBSD-3-STABLE  -  FREEBSD22  -  l41  -  OPENBSD  -  linux-2.6  -  MK84  -  PLAN9  -  xnu-8792 
SearchContext: -  none  -  3  -  10 

    1 /*-
    2  * Copyright (c) 2008 Isilon Inc http://www.isilon.com/
    3  * Authors: Doug Rabson <dfr@rabson.org>
    4  * Developed with Red Inc: Alfred Perlstein <alfred@freebsd.org>
    5  *
    6  * Redistribution and use in source and binary forms, with or without
    7  * modification, are permitted provided that the following conditions
    8  * are met:
    9  * 1. Redistributions of source code must retain the above copyright
   10  *    notice, this list of conditions and the following disclaimer.
   11  * 2. Redistributions in binary form must reproduce the above copyright
   12  *    notice, this list of conditions and the following disclaimer in the
   13  *    documentation and/or other materials provided with the distribution.
   14  *
   15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
   16  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
   17  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
   18  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
   19  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
   20  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
   21  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
   22  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
   23  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
   24  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
   25  * SUCH DAMAGE.
   26  */
   27 
   28 #include "opt_inet6.h"
   29 
   30 #include <sys/cdefs.h>
   31 __FBSDID("$FreeBSD$");
   32 
   33 #include <sys/param.h>
   34 #include <sys/fail.h>
   35 #include <sys/fcntl.h>
   36 #include <sys/kernel.h>
   37 #include <sys/kthread.h>
   38 #include <sys/lockf.h>
   39 #include <sys/malloc.h>
   40 #include <sys/mount.h>
   41 #if __FreeBSD_version >= 700000
   42 #include <sys/priv.h>
   43 #endif
   44 #include <sys/proc.h>
   45 #include <sys/socket.h>
   46 #include <sys/socketvar.h>
   47 #include <sys/syscall.h>
   48 #include <sys/sysctl.h>
   49 #include <sys/sysent.h>
   50 #include <sys/syslog.h>
   51 #include <sys/sysproto.h>
   52 #include <sys/systm.h>
   53 #include <sys/taskqueue.h>
   54 #include <sys/unistd.h>
   55 #include <sys/vnode.h>
   56 
   57 #include <nfs/nfsproto.h>
   58 #include <nfs/nfs_lock.h>
   59 
   60 #include <nlm/nlm_prot.h>
   61 #include <nlm/sm_inter.h>
   62 #include <nlm/nlm.h>
   63 #include <rpc/rpc_com.h>
   64 #include <rpc/rpcb_prot.h>
   65 
   66 MALLOC_DEFINE(M_NLM, "NLM", "Network Lock Manager");
   67 
   68 /*
   69  * If a host is inactive (and holds no locks) for this amount of
   70  * seconds, we consider it idle and stop tracking it.
   71  */
   72 #define NLM_IDLE_TIMEOUT        30
   73 
   74 /*
   75  * We check the host list for idle every few seconds.
   76  */
   77 #define NLM_IDLE_PERIOD         5
   78 
   79 /*
   80  * We only look for GRANTED_RES messages for a little while.
   81  */
   82 #define NLM_EXPIRE_TIMEOUT      10
   83 
   84 /*
   85  * Support for sysctl vfs.nlm.sysid
   86  */
   87 static SYSCTL_NODE(_vfs, OID_AUTO, nlm, CTLFLAG_RW, NULL,
   88     "Network Lock Manager");
   89 static SYSCTL_NODE(_vfs_nlm, OID_AUTO, sysid, CTLFLAG_RW, NULL, "");
   90 
   91 /*
   92  * Syscall hooks
   93  */
   94 static int nlm_syscall_offset = SYS_nlm_syscall;
   95 static struct sysent nlm_syscall_prev_sysent;
   96 #if __FreeBSD_version < 700000
   97 static struct sysent nlm_syscall_sysent = {
   98         (sizeof(struct nlm_syscall_args) / sizeof(register_t)) | SYF_MPSAFE,
   99         (sy_call_t *) nlm_syscall
  100 };
  101 #else
  102 MAKE_SYSENT(nlm_syscall);
  103 #endif
  104 static bool_t nlm_syscall_registered = FALSE;
  105 
  106 /*
  107  * Debug level passed in from userland. We also support a sysctl hook
  108  * so that it can be changed on a live system.
  109  */
  110 static int nlm_debug_level;
  111 SYSCTL_INT(_debug, OID_AUTO, nlm_debug, CTLFLAG_RW, &nlm_debug_level, 0, "");
  112 
  113 #define NLM_DEBUG(_level, args...)                      \
  114         do {                                            \
  115                 if (nlm_debug_level >= (_level))        \
  116                         log(LOG_DEBUG, args);           \
  117         } while(0)
  118 #define NLM_ERR(args...)                        \
  119         do {                                    \
  120                 log(LOG_ERR, args);             \
  121         } while(0)
  122 
  123 /*
  124  * Grace period handling. The value of nlm_grace_threshold is the
  125  * value of time_uptime after which we are serving requests normally.
  126  */
  127 static time_t nlm_grace_threshold;
  128 
  129 /*
  130  * We check for idle hosts if time_uptime is greater than
  131  * nlm_next_idle_check,
  132  */
  133 static time_t nlm_next_idle_check;
  134 
  135 /*
  136  * A flag to indicate the server is already running.
  137  */
  138 static int nlm_is_running;
  139 
  140 /*
  141  * A socket to use for RPC - shared by all IPv4 RPC clients.
  142  */
  143 static struct socket *nlm_socket;
  144 
  145 #ifdef INET6
  146 
  147 /*
  148  * A socket to use for RPC - shared by all IPv6 RPC clients.
  149  */
  150 static struct socket *nlm_socket6;
  151 
  152 #endif
  153 
  154 /*
  155  * An RPC client handle that can be used to communicate with the local
  156  * NSM.
  157  */
  158 static CLIENT *nlm_nsm;
  159 
  160 /*
  161  * An AUTH handle for the server's creds.
  162  */
  163 static AUTH *nlm_auth;
  164 
  165 /*
  166  * A zero timeval for sending async RPC messages.
  167  */
  168 struct timeval nlm_zero_tv = { 0, 0 };
  169 
  170 /*
  171  * The local NSM state number
  172  */
  173 int nlm_nsm_state;
  174 
  175 
  176 /*
  177  * A lock to protect the host list and waiting lock list.
  178  */
  179 static struct mtx nlm_global_lock;
  180 
  181 /*
  182  * Locks:
  183  * (l)          locked by nh_lock
  184  * (s)          only accessed via server RPC which is single threaded
  185  * (g)          locked by nlm_global_lock
  186  * (c)          const until freeing
  187  * (a)          modified using atomic ops
  188  */
  189 
  190 /*
  191  * A pending client-side lock request, stored on the nlm_waiting_locks
  192  * list.
  193  */
  194 struct nlm_waiting_lock {
  195         TAILQ_ENTRY(nlm_waiting_lock) nw_link; /* (g) */
  196         bool_t          nw_waiting;            /* (g) */
  197         nlm4_lock       nw_lock;               /* (c) */
  198         union nfsfh     nw_fh;                 /* (c) */
  199         struct vnode    *nw_vp;                /* (c) */
  200 };
  201 TAILQ_HEAD(nlm_waiting_lock_list, nlm_waiting_lock);
  202 
  203 struct nlm_waiting_lock_list nlm_waiting_locks; /* (g) */
  204 
  205 /*
  206  * A pending server-side asynchronous lock request, stored on the
  207  * nh_pending list of the NLM host.
  208  */
  209 struct nlm_async_lock {
  210         TAILQ_ENTRY(nlm_async_lock) af_link; /* (l) host's list of locks */
  211         struct task     af_task;        /* (c) async callback details */
  212         void            *af_cookie;     /* (l) lock manager cancel token */
  213         struct vnode    *af_vp;         /* (l) vnode to lock */
  214         struct flock    af_fl;          /* (c) lock details */
  215         struct nlm_host *af_host;       /* (c) host which is locking */
  216         CLIENT          *af_rpc;        /* (c) rpc client to send message */
  217         nlm4_testargs   af_granted;     /* (c) notification details */
  218         time_t          af_expiretime;  /* (c) notification time */
  219 };
  220 TAILQ_HEAD(nlm_async_lock_list, nlm_async_lock);
  221 
  222 /*
  223  * NLM host.
  224  */
  225 enum nlm_host_state {
  226         NLM_UNMONITORED,
  227         NLM_MONITORED,
  228         NLM_MONITOR_FAILED,
  229         NLM_RECOVERING
  230 };
  231 
  232 struct nlm_rpc {
  233         CLIENT          *nr_client;    /* (l) RPC client handle */
  234         time_t          nr_create_time; /* (l) when client was created */
  235 };
  236 
  237 struct nlm_host {
  238         struct mtx      nh_lock;
  239         volatile u_int  nh_refs;       /* (a) reference count */
  240         TAILQ_ENTRY(nlm_host) nh_link; /* (g) global list of hosts */
  241         char            nh_caller_name[MAXNAMELEN]; /* (c) printable name of host */
  242         uint32_t        nh_sysid;        /* (c) our allocaed system ID */
  243         char            nh_sysid_string[10]; /* (c) string rep. of sysid */
  244         struct sockaddr_storage nh_addr; /* (s) remote address of host */
  245         struct nlm_rpc  nh_srvrpc;       /* (l) RPC for server replies */
  246         struct nlm_rpc  nh_clntrpc;      /* (l) RPC for client requests */
  247         rpcvers_t       nh_vers;         /* (s) NLM version of host */
  248         int             nh_state;        /* (s) last seen NSM state of host */
  249         enum nlm_host_state nh_monstate; /* (l) local NSM monitoring state */
  250         time_t          nh_idle_timeout; /* (s) Time at which host is idle */
  251         struct sysctl_ctx_list nh_sysctl; /* (c) vfs.nlm.sysid nodes */
  252         uint32_t        nh_grantcookie;  /* (l) grant cookie counter */
  253         struct nlm_async_lock_list nh_pending; /* (l) pending async locks */
  254         struct nlm_async_lock_list nh_granted; /* (l) granted locks */
  255         struct nlm_async_lock_list nh_finished; /* (l) finished async locks */
  256 };
  257 TAILQ_HEAD(nlm_host_list, nlm_host);
  258 
  259 static struct nlm_host_list nlm_hosts; /* (g) */
  260 static uint32_t nlm_next_sysid = 1;    /* (g) */
  261 
  262 static void     nlm_host_unmonitor(struct nlm_host *);
  263 
  264 struct nlm_grantcookie {
  265         uint32_t        ng_sysid;
  266         uint32_t        ng_cookie;
  267 };
  268 
  269 static inline uint32_t
  270 ng_sysid(struct netobj *src)
  271 {
  272 
  273         return ((struct nlm_grantcookie *)src->n_bytes)->ng_sysid;
  274 }
  275 
  276 static inline uint32_t
  277 ng_cookie(struct netobj *src)
  278 {
  279 
  280         return ((struct nlm_grantcookie *)src->n_bytes)->ng_cookie;
  281 }
  282 
  283 /**********************************************************************/
  284 
  285 /*
  286  * Initialise NLM globals.
  287  */
  288 static void
  289 nlm_init(void *dummy)
  290 {
  291         int error;
  292 
  293         mtx_init(&nlm_global_lock, "nlm_global_lock", NULL, MTX_DEF);
  294         TAILQ_INIT(&nlm_waiting_locks);
  295         TAILQ_INIT(&nlm_hosts);
  296 
  297         error = syscall_register(&nlm_syscall_offset, &nlm_syscall_sysent,
  298             &nlm_syscall_prev_sysent);
  299         if (error)
  300                 NLM_ERR("Can't register NLM syscall\n");
  301         else
  302                 nlm_syscall_registered = TRUE;
  303 }
  304 SYSINIT(nlm_init, SI_SUB_LOCK, SI_ORDER_FIRST, nlm_init, NULL);
  305 
  306 static void
  307 nlm_uninit(void *dummy)
  308 {
  309 
  310         if (nlm_syscall_registered)
  311                 syscall_deregister(&nlm_syscall_offset,
  312                     &nlm_syscall_prev_sysent);
  313 }
  314 SYSUNINIT(nlm_uninit, SI_SUB_LOCK, SI_ORDER_FIRST, nlm_uninit, NULL);
  315 
  316 /*
  317  * Create a netobj from an arbitrary source.
  318  */
  319 void
  320 nlm_make_netobj(struct netobj *dst, caddr_t src, size_t srcsize,
  321     struct malloc_type *type)
  322 {
  323 
  324         dst->n_len = srcsize;
  325         dst->n_bytes = malloc(srcsize, type, M_WAITOK);
  326         memcpy(dst->n_bytes, src, srcsize);
  327 }
  328 
  329 /*
  330  * Copy a struct netobj.
  331  */ 
  332 void
  333 nlm_copy_netobj(struct netobj *dst, struct netobj *src,
  334     struct malloc_type *type)
  335 {
  336 
  337         nlm_make_netobj(dst, src->n_bytes, src->n_len, type);
  338 }
  339 
  340 
  341 /*
  342  * Create an RPC client handle for the given (address,prog,vers)
  343  * triple using UDP.
  344  */
  345 static CLIENT *
  346 nlm_get_rpc(struct sockaddr *sa, rpcprog_t prog, rpcvers_t vers)
  347 {
  348         char *wchan = "nlmrcv";
  349         const char* protofmly;
  350         struct sockaddr_storage ss;
  351         struct socket *so;
  352         CLIENT *rpcb;
  353         struct timeval timo;
  354         RPCB parms;
  355         char *uaddr;
  356         enum clnt_stat stat = RPC_SUCCESS;
  357         int rpcvers = RPCBVERS4;
  358         bool_t do_tcp = FALSE;
  359         bool_t tryagain = FALSE;
  360         struct portmap mapping;
  361         u_short port = 0;
  362 
  363         /*
  364          * First we need to contact the remote RPCBIND service to find
  365          * the right port.
  366          */
  367         memcpy(&ss, sa, sa->sa_len);
  368         switch (ss.ss_family) {
  369         case AF_INET:
  370                 ((struct sockaddr_in *)&ss)->sin_port = htons(111);
  371                 protofmly = "inet";
  372                 so = nlm_socket;
  373                 break;
  374                 
  375 #ifdef INET6
  376         case AF_INET6:
  377                 ((struct sockaddr_in6 *)&ss)->sin6_port = htons(111);
  378                 protofmly = "inet6";
  379                 so = nlm_socket6;
  380                 break;
  381 #endif
  382 
  383         default:
  384                 /*
  385                  * Unsupported address family - fail.
  386                  */
  387                 return (NULL);
  388         }
  389 
  390         rpcb = clnt_dg_create(so, (struct sockaddr *)&ss,
  391             RPCBPROG, rpcvers, 0, 0);
  392         if (!rpcb)
  393                 return (NULL);
  394 
  395 try_tcp:
  396         parms.r_prog = prog;
  397         parms.r_vers = vers;
  398         if (do_tcp)
  399                 parms.r_netid = "tcp";
  400         else
  401                 parms.r_netid = "udp";
  402         parms.r_addr = "";
  403         parms.r_owner = "";
  404 
  405         /*
  406          * Use the default timeout.
  407          */
  408         timo.tv_sec = 25;
  409         timo.tv_usec = 0;
  410 again:
  411         switch (rpcvers) {
  412         case RPCBVERS4:
  413         case RPCBVERS:
  414                 /*
  415                  * Try RPCBIND 4 then 3.
  416                  */
  417                 uaddr = NULL;
  418                 stat = CLNT_CALL(rpcb, (rpcprog_t) RPCBPROC_GETADDR,
  419                     (xdrproc_t) xdr_rpcb, &parms,
  420                     (xdrproc_t) xdr_wrapstring, &uaddr, timo);
  421                 if (stat == RPC_SUCCESS) {
  422                         /*
  423                          * We have a reply from the remote RPCBIND - turn it
  424                          * into an appropriate address and make a new client
  425                          * that can talk to the remote NLM.
  426                          *
  427                          * XXX fixup IPv6 scope ID.
  428                          */
  429                         struct netbuf *a;
  430                         a = __rpc_uaddr2taddr_af(ss.ss_family, uaddr);
  431                         if (!a) {
  432                                 tryagain = TRUE;
  433                         } else {
  434                                 tryagain = FALSE;
  435                                 memcpy(&ss, a->buf, a->len);
  436                                 free(a->buf, M_RPC);
  437                                 free(a, M_RPC);
  438                                 xdr_free((xdrproc_t) xdr_wrapstring, &uaddr);
  439                         }
  440                 }
  441                 if (tryagain || stat == RPC_PROGVERSMISMATCH) {
  442                         if (rpcvers == RPCBVERS4)
  443                                 rpcvers = RPCBVERS;
  444                         else if (rpcvers == RPCBVERS)
  445                                 rpcvers = PMAPVERS;
  446                         CLNT_CONTROL(rpcb, CLSET_VERS, &rpcvers);
  447                         goto again;
  448                 }
  449                 break;
  450         case PMAPVERS:
  451                 /*
  452                  * Try portmap.
  453                  */
  454                 mapping.pm_prog = parms.r_prog;
  455                 mapping.pm_vers = parms.r_vers;
  456                 mapping.pm_prot = do_tcp ? IPPROTO_TCP : IPPROTO_UDP;
  457                 mapping.pm_port = 0;
  458 
  459                 stat = CLNT_CALL(rpcb, (rpcprog_t) PMAPPROC_GETPORT,
  460                     (xdrproc_t) xdr_portmap, &mapping,
  461                     (xdrproc_t) xdr_u_short, &port, timo);
  462 
  463                 if (stat == RPC_SUCCESS) {
  464                         switch (ss.ss_family) {
  465                         case AF_INET:
  466                                 ((struct sockaddr_in *)&ss)->sin_port =
  467                                         htons(port);
  468                                 break;
  469                 
  470 #ifdef INET6
  471                         case AF_INET6:
  472                                 ((struct sockaddr_in6 *)&ss)->sin6_port =
  473                                         htons(port);
  474                                 break;
  475 #endif
  476                         }
  477                 }
  478                 break;
  479         default:
  480                 panic("invalid rpcvers %d", rpcvers);
  481         }
  482         /*
  483          * We may have a positive response from the portmapper, but the NLM
  484          * service was not found. Make sure we received a valid port.
  485          */
  486         switch (ss.ss_family) {
  487         case AF_INET:
  488                 port = ((struct sockaddr_in *)&ss)->sin_port;
  489                 break;
  490 #ifdef INET6
  491         case AF_INET6:
  492                 port = ((struct sockaddr_in6 *)&ss)->sin6_port;
  493                 break;
  494 #endif
  495         }
  496         if (stat != RPC_SUCCESS || !port) {
  497                 /*
  498                  * If we were able to talk to rpcbind or portmap, but the udp
  499                  * variant wasn't available, ask about tcp.
  500                  *
  501                  * XXX - We could also check for a TCP portmapper, but
  502                  * if the host is running a portmapper at all, we should be able
  503                  * to hail it over UDP.
  504                  */
  505                 if (stat == RPC_SUCCESS && !do_tcp) {
  506                         do_tcp = TRUE;
  507                         goto try_tcp;
  508                 }
  509 
  510                 /* Otherwise, bad news. */
  511                 NLM_ERR("NLM: failed to contact remote rpcbind, "
  512                     "stat = %d, port = %d\n", (int) stat, port);
  513                 CLNT_DESTROY(rpcb);
  514                 return (NULL);
  515         }
  516 
  517         if (do_tcp) {
  518                 /*
  519                  * Destroy the UDP client we used to speak to rpcbind and
  520                  * recreate as a TCP client.
  521                  */
  522                 struct netconfig *nconf = NULL;
  523 
  524                 CLNT_DESTROY(rpcb);
  525 
  526                 switch (ss.ss_family) {
  527                 case AF_INET:
  528                         nconf = getnetconfigent("tcp");
  529                         break;
  530 #ifdef INET6
  531                 case AF_INET6:
  532                         nconf = getnetconfigent("tcp6");
  533                         break;
  534 #endif
  535                 }
  536 
  537                 rpcb = clnt_reconnect_create(nconf, (struct sockaddr *)&ss,
  538                     prog, vers, 0, 0);
  539                 CLNT_CONTROL(rpcb, CLSET_WAITCHAN, wchan);
  540                 rpcb->cl_auth = nlm_auth;
  541                 
  542         } else {
  543                 /*
  544                  * Re-use the client we used to speak to rpcbind.
  545                  */
  546                 CLNT_CONTROL(rpcb, CLSET_SVC_ADDR, &ss);
  547                 CLNT_CONTROL(rpcb, CLSET_PROG, &prog);
  548                 CLNT_CONTROL(rpcb, CLSET_VERS, &vers);
  549                 CLNT_CONTROL(rpcb, CLSET_WAITCHAN, wchan);
  550                 rpcb->cl_auth = nlm_auth;
  551         }
  552 
  553         return (rpcb);
  554 }
  555 
  556 /*
  557  * This async callback after when an async lock request has been
  558  * granted. We notify the host which initiated the request.
  559  */
  560 static void
  561 nlm_lock_callback(void *arg, int pending)
  562 {
  563         struct nlm_async_lock *af = (struct nlm_async_lock *) arg;
  564         struct rpc_callextra ext;
  565 
  566         NLM_DEBUG(2, "NLM: async lock %p for %s (sysid %d) granted,"
  567             " cookie %d:%d\n", af, af->af_host->nh_caller_name,
  568             af->af_host->nh_sysid, ng_sysid(&af->af_granted.cookie),
  569             ng_cookie(&af->af_granted.cookie));
  570 
  571         /*
  572          * Send the results back to the host.
  573          *
  574          * Note: there is a possible race here with nlm_host_notify
  575          * destroying the RPC client. To avoid problems, the first
  576          * thing nlm_host_notify does is to cancel pending async lock
  577          * requests.
  578          */
  579         memset(&ext, 0, sizeof(ext));
  580         ext.rc_auth = nlm_auth;
  581         if (af->af_host->nh_vers == NLM_VERS4) {
  582                 nlm4_granted_msg_4(&af->af_granted,
  583                     NULL, af->af_rpc, &ext, nlm_zero_tv);
  584         } else {
  585                 /*
  586                  * Back-convert to legacy protocol
  587                  */
  588                 nlm_testargs granted;
  589                 granted.cookie = af->af_granted.cookie;
  590                 granted.exclusive = af->af_granted.exclusive;
  591                 granted.alock.caller_name =
  592                         af->af_granted.alock.caller_name;
  593                 granted.alock.fh = af->af_granted.alock.fh;
  594                 granted.alock.oh = af->af_granted.alock.oh;
  595                 granted.alock.svid = af->af_granted.alock.svid;
  596                 granted.alock.l_offset =
  597                         af->af_granted.alock.l_offset;
  598                 granted.alock.l_len =
  599                         af->af_granted.alock.l_len;
  600 
  601                 nlm_granted_msg_1(&granted,
  602                     NULL, af->af_rpc, &ext, nlm_zero_tv);
  603         }
  604 
  605         /*
  606          * Move this entry to the nh_granted list.
  607          */
  608         af->af_expiretime = time_uptime + NLM_EXPIRE_TIMEOUT;
  609         mtx_lock(&af->af_host->nh_lock);
  610         TAILQ_REMOVE(&af->af_host->nh_pending, af, af_link);
  611         TAILQ_INSERT_TAIL(&af->af_host->nh_granted, af, af_link);
  612         mtx_unlock(&af->af_host->nh_lock);
  613 }
  614 
  615 /*
  616  * Free an async lock request. The request must have been removed from
  617  * any list.
  618  */
  619 static void
  620 nlm_free_async_lock(struct nlm_async_lock *af)
  621 {
  622         /*
  623          * Free an async lock.
  624          */
  625         if (af->af_rpc)
  626                 CLNT_RELEASE(af->af_rpc);
  627         xdr_free((xdrproc_t) xdr_nlm4_testargs, &af->af_granted);
  628         if (af->af_vp)
  629                 vrele(af->af_vp);
  630         free(af, M_NLM);
  631 }
  632 
  633 /*
  634  * Cancel our async request - this must be called with
  635  * af->nh_host->nh_lock held. This is slightly complicated by a
  636  * potential race with our own callback. If we fail to cancel the
  637  * lock, it must already have been granted - we make sure our async
  638  * task has completed by calling taskqueue_drain in this case.
  639  */
  640 static int
  641 nlm_cancel_async_lock(struct nlm_async_lock *af)
  642 {
  643         struct nlm_host *host = af->af_host;
  644         int error;
  645 
  646         mtx_assert(&host->nh_lock, MA_OWNED);
  647 
  648         mtx_unlock(&host->nh_lock);
  649 
  650         error = VOP_ADVLOCKASYNC(af->af_vp, NULL, F_CANCEL, &af->af_fl,
  651             F_REMOTE, NULL, &af->af_cookie);
  652 
  653         if (error) {
  654                 /*
  655                  * We failed to cancel - make sure our callback has
  656                  * completed before we continue.
  657                  */
  658                 taskqueue_drain(taskqueue_thread, &af->af_task);
  659         }
  660 
  661         mtx_lock(&host->nh_lock);
  662         
  663         if (!error) {
  664                 NLM_DEBUG(2, "NLM: async lock %p for %s (sysid %d) "
  665                     "cancelled\n", af, host->nh_caller_name, host->nh_sysid);
  666 
  667                 /*
  668                  * Remove from the nh_pending list and free now that
  669                  * we are safe from the callback.
  670                  */
  671                 TAILQ_REMOVE(&host->nh_pending, af, af_link);
  672                 mtx_unlock(&host->nh_lock);
  673                 nlm_free_async_lock(af);
  674                 mtx_lock(&host->nh_lock);
  675         }
  676 
  677         return (error);
  678 }
  679 
  680 static void
  681 nlm_check_expired_locks(struct nlm_host *host)
  682 {
  683         struct nlm_async_lock *af;
  684         time_t uptime = time_uptime;
  685 
  686         mtx_lock(&host->nh_lock);
  687         while ((af = TAILQ_FIRST(&host->nh_granted)) != NULL
  688             && uptime >= af->af_expiretime) {
  689                 NLM_DEBUG(2, "NLM: async lock %p for %s (sysid %d) expired,"
  690                     " cookie %d:%d\n", af, af->af_host->nh_caller_name,
  691                     af->af_host->nh_sysid, ng_sysid(&af->af_granted.cookie),
  692                     ng_cookie(&af->af_granted.cookie));
  693                 TAILQ_REMOVE(&host->nh_granted, af, af_link);
  694                 mtx_unlock(&host->nh_lock);
  695                 nlm_free_async_lock(af);
  696                 mtx_lock(&host->nh_lock);
  697         }
  698         while ((af = TAILQ_FIRST(&host->nh_finished)) != NULL) {
  699                 TAILQ_REMOVE(&host->nh_finished, af, af_link);
  700                 mtx_unlock(&host->nh_lock);
  701                 nlm_free_async_lock(af);
  702                 mtx_lock(&host->nh_lock);
  703         }
  704         mtx_unlock(&host->nh_lock);
  705 }
  706 
  707 /*
  708  * Free resources used by a host. This is called after the reference
  709  * count has reached zero so it doesn't need to worry about locks.
  710  */
  711 static void
  712 nlm_host_destroy(struct nlm_host *host)
  713 {
  714 
  715         mtx_lock(&nlm_global_lock);
  716         TAILQ_REMOVE(&nlm_hosts, host, nh_link);
  717         mtx_unlock(&nlm_global_lock);
  718 
  719         if (host->nh_srvrpc.nr_client)
  720                 CLNT_RELEASE(host->nh_srvrpc.nr_client);
  721         if (host->nh_clntrpc.nr_client)
  722                 CLNT_RELEASE(host->nh_clntrpc.nr_client);
  723         mtx_destroy(&host->nh_lock);
  724         sysctl_ctx_free(&host->nh_sysctl);
  725         free(host, M_NLM);
  726 }
  727 
  728 /*
  729  * Thread start callback for client lock recovery
  730  */
  731 static void
  732 nlm_client_recovery_start(void *arg)
  733 {
  734         struct nlm_host *host = (struct nlm_host *) arg;
  735 
  736         NLM_DEBUG(1, "NLM: client lock recovery for %s started\n",
  737             host->nh_caller_name);
  738 
  739         nlm_client_recovery(host);
  740 
  741         NLM_DEBUG(1, "NLM: client lock recovery for %s completed\n",
  742             host->nh_caller_name);
  743 
  744         host->nh_monstate = NLM_MONITORED;
  745         nlm_host_release(host);
  746 
  747         kthread_exit();
  748 }
  749 
  750 /*
  751  * This is called when we receive a host state change notification. We
  752  * unlock any active locks owned by the host. When rpc.lockd is
  753  * shutting down, this function is called with newstate set to zero
  754  * which allows us to cancel any pending async locks and clear the
  755  * locking state.
  756  */
  757 static void
  758 nlm_host_notify(struct nlm_host *host, int newstate)
  759 {
  760         struct nlm_async_lock *af;
  761 
  762         if (newstate) {
  763                 NLM_DEBUG(1, "NLM: host %s (sysid %d) rebooted, new "
  764                     "state is %d\n", host->nh_caller_name,
  765                     host->nh_sysid, newstate);
  766         }
  767 
  768         /*
  769          * Cancel any pending async locks for this host.
  770          */
  771         mtx_lock(&host->nh_lock);
  772         while ((af = TAILQ_FIRST(&host->nh_pending)) != NULL) {
  773                 /*
  774                  * nlm_cancel_async_lock will remove the entry from
  775                  * nh_pending and free it.
  776                  */
  777                 nlm_cancel_async_lock(af);
  778         }
  779         mtx_unlock(&host->nh_lock);
  780         nlm_check_expired_locks(host);
  781 
  782         /*
  783          * The host just rebooted - trash its locks.
  784          */
  785         lf_clearremotesys(host->nh_sysid);
  786         host->nh_state = newstate;
  787 
  788         /*
  789          * If we have any remote locks for this host (i.e. it
  790          * represents a remote NFS server that our local NFS client
  791          * has locks for), start a recovery thread.
  792          */
  793         if (newstate != 0
  794             && host->nh_monstate != NLM_RECOVERING
  795             && lf_countlocks(NLM_SYSID_CLIENT | host->nh_sysid) > 0) {
  796                 struct thread *td;
  797                 host->nh_monstate = NLM_RECOVERING;
  798                 refcount_acquire(&host->nh_refs);
  799                 kthread_add(nlm_client_recovery_start, host, curproc, &td, 0, 0,
  800                     "NFS lock recovery for %s", host->nh_caller_name);
  801         }
  802 }
  803 
  804 /*
  805  * Sysctl handler to count the number of locks for a sysid.
  806  */
  807 static int
  808 nlm_host_lock_count_sysctl(SYSCTL_HANDLER_ARGS)
  809 {
  810         struct nlm_host *host;
  811         int count;
  812 
  813         host = oidp->oid_arg1;
  814         count = lf_countlocks(host->nh_sysid);
  815         return sysctl_handle_int(oidp, &count, 0, req);
  816 }
  817 
  818 /*
  819  * Sysctl handler to count the number of client locks for a sysid.
  820  */
  821 static int
  822 nlm_host_client_lock_count_sysctl(SYSCTL_HANDLER_ARGS)
  823 {
  824         struct nlm_host *host;
  825         int count;
  826 
  827         host = oidp->oid_arg1;
  828         count = lf_countlocks(NLM_SYSID_CLIENT | host->nh_sysid);
  829         return sysctl_handle_int(oidp, &count, 0, req);
  830 }
  831 
  832 /*
  833  * Create a new NLM host.
  834  */
  835 static struct nlm_host *
  836 nlm_create_host(const char* caller_name)
  837 {
  838         struct nlm_host *host;
  839         struct sysctl_oid *oid;
  840 
  841         mtx_assert(&nlm_global_lock, MA_OWNED);
  842 
  843         NLM_DEBUG(1, "NLM: new host %s (sysid %d)\n",
  844             caller_name, nlm_next_sysid);
  845         host = malloc(sizeof(struct nlm_host), M_NLM, M_NOWAIT|M_ZERO);
  846         if (!host)
  847                 return (NULL);
  848         mtx_init(&host->nh_lock, "nh_lock", NULL, MTX_DEF);
  849         host->nh_refs = 1;
  850         strlcpy(host->nh_caller_name, caller_name, MAXNAMELEN);
  851         host->nh_sysid = nlm_next_sysid++;
  852         snprintf(host->nh_sysid_string, sizeof(host->nh_sysid_string),
  853                 "%d", host->nh_sysid);
  854         host->nh_vers = 0;
  855         host->nh_state = 0;
  856         host->nh_monstate = NLM_UNMONITORED;
  857         host->nh_grantcookie = 1;
  858         TAILQ_INIT(&host->nh_pending);
  859         TAILQ_INIT(&host->nh_granted);
  860         TAILQ_INIT(&host->nh_finished);
  861         TAILQ_INSERT_TAIL(&nlm_hosts, host, nh_link);
  862 
  863         mtx_unlock(&nlm_global_lock);
  864 
  865         sysctl_ctx_init(&host->nh_sysctl);
  866         oid = SYSCTL_ADD_NODE(&host->nh_sysctl,
  867             SYSCTL_STATIC_CHILDREN(_vfs_nlm_sysid),
  868             OID_AUTO, host->nh_sysid_string, CTLFLAG_RD, NULL, "");
  869         SYSCTL_ADD_STRING(&host->nh_sysctl, SYSCTL_CHILDREN(oid), OID_AUTO,
  870             "hostname", CTLFLAG_RD, host->nh_caller_name, 0, "");
  871         SYSCTL_ADD_UINT(&host->nh_sysctl, SYSCTL_CHILDREN(oid), OID_AUTO,
  872             "version", CTLFLAG_RD, &host->nh_vers, 0, "");
  873         SYSCTL_ADD_UINT(&host->nh_sysctl, SYSCTL_CHILDREN(oid), OID_AUTO,
  874             "monitored", CTLFLAG_RD, &host->nh_monstate, 0, "");
  875         SYSCTL_ADD_PROC(&host->nh_sysctl, SYSCTL_CHILDREN(oid), OID_AUTO,
  876             "lock_count", CTLTYPE_INT | CTLFLAG_RD, host, 0,
  877             nlm_host_lock_count_sysctl, "I", "");
  878         SYSCTL_ADD_PROC(&host->nh_sysctl, SYSCTL_CHILDREN(oid), OID_AUTO,
  879             "client_lock_count", CTLTYPE_INT | CTLFLAG_RD, host, 0,
  880             nlm_host_client_lock_count_sysctl, "I", "");
  881 
  882         mtx_lock(&nlm_global_lock);
  883 
  884         return (host);
  885 }
  886 
  887 /*
  888  * Acquire the next sysid for remote locks not handled by the NLM.
  889  */
  890 uint32_t
  891 nlm_acquire_next_sysid(void)
  892 {
  893         uint32_t next_sysid;
  894 
  895         mtx_lock(&nlm_global_lock);
  896         next_sysid = nlm_next_sysid++;
  897         mtx_unlock(&nlm_global_lock);
  898         return (next_sysid);
  899 }
  900 
  901 /*
  902  * Return non-zero if the address parts of the two sockaddrs are the
  903  * same.
  904  */
  905 static int
  906 nlm_compare_addr(const struct sockaddr *a, const struct sockaddr *b)
  907 {
  908         const struct sockaddr_in *a4, *b4;
  909 #ifdef INET6
  910         const struct sockaddr_in6 *a6, *b6;
  911 #endif
  912 
  913         if (a->sa_family != b->sa_family)
  914                 return (FALSE);
  915 
  916         switch (a->sa_family) {
  917         case AF_INET:
  918                 a4 = (const struct sockaddr_in *) a;
  919                 b4 = (const struct sockaddr_in *) b;
  920                 return !memcmp(&a4->sin_addr, &b4->sin_addr,
  921                     sizeof(a4->sin_addr));
  922 #ifdef INET6
  923         case AF_INET6:
  924                 a6 = (const struct sockaddr_in6 *) a;
  925                 b6 = (const struct sockaddr_in6 *) b;
  926                 return !memcmp(&a6->sin6_addr, &b6->sin6_addr,
  927                     sizeof(a6->sin6_addr));
  928 #endif
  929         }
  930 
  931         return (0);
  932 }
  933 
  934 /*
  935  * Check for idle hosts and stop monitoring them. We could also free
  936  * the host structure here, possibly after a larger timeout but that
  937  * would require some care to avoid races with
  938  * e.g. nlm_host_lock_count_sysctl.
  939  */
  940 static void
  941 nlm_check_idle(void)
  942 {
  943         struct nlm_host *host;
  944 
  945         mtx_assert(&nlm_global_lock, MA_OWNED);
  946 
  947         if (time_uptime <= nlm_next_idle_check)
  948                 return;
  949 
  950         nlm_next_idle_check = time_uptime + NLM_IDLE_PERIOD;
  951 
  952         TAILQ_FOREACH(host, &nlm_hosts, nh_link) {
  953                 if (host->nh_monstate == NLM_MONITORED
  954                     && time_uptime > host->nh_idle_timeout) {
  955                         mtx_unlock(&nlm_global_lock);
  956                         if (lf_countlocks(host->nh_sysid) > 0
  957                             || lf_countlocks(NLM_SYSID_CLIENT
  958                                 + host->nh_sysid)) {
  959                                 host->nh_idle_timeout =
  960                                         time_uptime + NLM_IDLE_TIMEOUT;
  961                                 mtx_lock(&nlm_global_lock);
  962                                 continue;
  963                         }
  964                         nlm_host_unmonitor(host);
  965                         mtx_lock(&nlm_global_lock);
  966                 } 
  967         }
  968 }
  969 
  970 /*
  971  * Search for an existing NLM host that matches the given name
  972  * (typically the caller_name element of an nlm4_lock).  If none is
  973  * found, create a new host. If 'addr' is non-NULL, record the remote
  974  * address of the host so that we can call it back for async
  975  * responses. If 'vers' is greater than zero then record the NLM
  976  * program version to use to communicate with this client.
  977  */
  978 struct nlm_host *
  979 nlm_find_host_by_name(const char *name, const struct sockaddr *addr,
  980     rpcvers_t vers)
  981 {
  982         struct nlm_host *host;
  983 
  984         mtx_lock(&nlm_global_lock);
  985 
  986         /*
  987          * The remote host is determined by caller_name.
  988          */
  989         TAILQ_FOREACH(host, &nlm_hosts, nh_link) {
  990                 if (!strcmp(host->nh_caller_name, name))
  991                         break;
  992         }
  993 
  994         if (!host) {
  995                 host = nlm_create_host(name);
  996                 if (!host) {
  997                         mtx_unlock(&nlm_global_lock);
  998                         return (NULL);
  999                 }
 1000         }
 1001         refcount_acquire(&host->nh_refs);
 1002 
 1003         host->nh_idle_timeout = time_uptime + NLM_IDLE_TIMEOUT;
 1004 
 1005         /*
 1006          * If we have an address for the host, record it so that we
 1007          * can send async replies etc.
 1008          */
 1009         if (addr) {
 1010                 
 1011                 KASSERT(addr->sa_len < sizeof(struct sockaddr_storage),
 1012                     ("Strange remote transport address length"));
 1013 
 1014                 /*
 1015                  * If we have seen an address before and we currently
 1016                  * have an RPC client handle, make sure the address is
 1017                  * the same, otherwise discard the client handle.
 1018                  */
 1019                 if (host->nh_addr.ss_len && host->nh_srvrpc.nr_client) {
 1020                         if (!nlm_compare_addr(
 1021                                     (struct sockaddr *) &host->nh_addr,
 1022                                     addr)
 1023                             || host->nh_vers != vers) {
 1024                                 CLIENT *client;
 1025                                 mtx_lock(&host->nh_lock);
 1026                                 client = host->nh_srvrpc.nr_client;
 1027                                 host->nh_srvrpc.nr_client = NULL;
 1028                                 mtx_unlock(&host->nh_lock);
 1029                                 if (client) {
 1030                                         CLNT_RELEASE(client);
 1031                                 }
 1032                         }
 1033                 }
 1034                 memcpy(&host->nh_addr, addr, addr->sa_len);
 1035                 host->nh_vers = vers;
 1036         }
 1037 
 1038         nlm_check_idle();
 1039 
 1040         mtx_unlock(&nlm_global_lock);
 1041 
 1042         return (host);
 1043 }
 1044 
 1045 /*
 1046  * Search for an existing NLM host that matches the given remote
 1047  * address. If none is found, create a new host with the requested
 1048  * address and remember 'vers' as the NLM protocol version to use for
 1049  * that host.
 1050  */
 1051 struct nlm_host *
 1052 nlm_find_host_by_addr(const struct sockaddr *addr, int vers)
 1053 {
 1054         /*
 1055          * Fake up a name using inet_ntop. This buffer is
 1056          * large enough for an IPv6 address.
 1057          */
 1058         char tmp[sizeof "ffff:ffff:ffff:ffff:ffff:ffff:255.255.255.255"];
 1059         struct nlm_host *host;
 1060 
 1061         switch (addr->sa_family) {
 1062         case AF_INET:
 1063                 inet_ntop(AF_INET,
 1064                     &((const struct sockaddr_in *) addr)->sin_addr,
 1065                     tmp, sizeof tmp);
 1066                 break;
 1067 #ifdef INET6
 1068         case AF_INET6:
 1069                 inet_ntop(AF_INET6,
 1070                     &((const struct sockaddr_in6 *) addr)->sin6_addr,
 1071                     tmp, sizeof tmp);
 1072                 break;
 1073 #endif
 1074         default:
 1075                 strlcpy(tmp, "<unknown>", sizeof(tmp));
 1076         }
 1077 
 1078 
 1079         mtx_lock(&nlm_global_lock);
 1080 
 1081         /*
 1082          * The remote host is determined by caller_name.
 1083          */
 1084         TAILQ_FOREACH(host, &nlm_hosts, nh_link) {
 1085                 if (nlm_compare_addr(addr,
 1086                         (const struct sockaddr *) &host->nh_addr))
 1087                         break;
 1088         }
 1089 
 1090         if (!host) {
 1091                 host = nlm_create_host(tmp);
 1092                 if (!host) {
 1093                         mtx_unlock(&nlm_global_lock);
 1094                         return (NULL);
 1095                 }
 1096                 memcpy(&host->nh_addr, addr, addr->sa_len);
 1097                 host->nh_vers = vers;
 1098         }
 1099         refcount_acquire(&host->nh_refs);
 1100 
 1101         host->nh_idle_timeout = time_uptime + NLM_IDLE_TIMEOUT;
 1102 
 1103         nlm_check_idle();
 1104 
 1105         mtx_unlock(&nlm_global_lock);
 1106 
 1107         return (host);
 1108 }
 1109 
 1110 /*
 1111  * Find the NLM host that matches the value of 'sysid'. If none
 1112  * exists, return NULL.
 1113  */
 1114 static struct nlm_host *
 1115 nlm_find_host_by_sysid(int sysid)
 1116 {
 1117         struct nlm_host *host;
 1118 
 1119         TAILQ_FOREACH(host, &nlm_hosts, nh_link) {
 1120                 if (host->nh_sysid == sysid) {
 1121                         refcount_acquire(&host->nh_refs);
 1122                         return (host);
 1123                 }
 1124         }
 1125 
 1126         return (NULL);
 1127 }
 1128 
 1129 void nlm_host_release(struct nlm_host *host)
 1130 {
 1131         if (refcount_release(&host->nh_refs)) {
 1132                 /*
 1133                  * Free the host
 1134                  */
 1135                 nlm_host_destroy(host);
 1136         }
 1137 }
 1138 
 1139 /*
 1140  * Unregister this NLM host with the local NSM due to idleness.
 1141  */
 1142 static void
 1143 nlm_host_unmonitor(struct nlm_host *host)
 1144 {
 1145         mon_id smmonid;
 1146         sm_stat_res smstat;
 1147         struct timeval timo;
 1148         enum clnt_stat stat;
 1149 
 1150         NLM_DEBUG(1, "NLM: unmonitoring %s (sysid %d)\n",
 1151             host->nh_caller_name, host->nh_sysid);
 1152 
 1153         /*
 1154          * We put our assigned system ID value in the priv field to
 1155          * make it simpler to find the host if we are notified of a
 1156          * host restart.
 1157          */
 1158         smmonid.mon_name = host->nh_caller_name;
 1159         smmonid.my_id.my_name = "localhost";
 1160         smmonid.my_id.my_prog = NLM_PROG;
 1161         smmonid.my_id.my_vers = NLM_SM;
 1162         smmonid.my_id.my_proc = NLM_SM_NOTIFY;
 1163 
 1164         timo.tv_sec = 25;
 1165         timo.tv_usec = 0;
 1166         stat = CLNT_CALL(nlm_nsm, SM_UNMON,
 1167             (xdrproc_t) xdr_mon, &smmonid,
 1168             (xdrproc_t) xdr_sm_stat, &smstat, timo);
 1169 
 1170         if (stat != RPC_SUCCESS) {
 1171                 NLM_ERR("Failed to contact local NSM - rpc error %d\n", stat);
 1172                 return;
 1173         }
 1174         if (smstat.res_stat == stat_fail) {
 1175                 NLM_ERR("Local NSM refuses to unmonitor %s\n",
 1176                     host->nh_caller_name);
 1177                 return;
 1178         }
 1179 
 1180         host->nh_monstate = NLM_UNMONITORED;
 1181 }
 1182 
 1183 /*
 1184  * Register this NLM host with the local NSM so that we can be
 1185  * notified if it reboots.
 1186  */
 1187 void
 1188 nlm_host_monitor(struct nlm_host *host, int state)
 1189 {
 1190         mon smmon;
 1191         sm_stat_res smstat;
 1192         struct timeval timo;
 1193         enum clnt_stat stat;
 1194 
 1195         if (state && !host->nh_state) {
 1196                 /*
 1197                  * This is the first time we have seen an NSM state
 1198                  * value for this host. We record it here to help
 1199                  * detect host reboots.
 1200                  */
 1201                 host->nh_state = state;
 1202                 NLM_DEBUG(1, "NLM: host %s (sysid %d) has NSM state %d\n",
 1203                     host->nh_caller_name, host->nh_sysid, state);
 1204         }
 1205 
 1206         mtx_lock(&host->nh_lock);
 1207         if (host->nh_monstate != NLM_UNMONITORED) {
 1208                 mtx_unlock(&host->nh_lock);
 1209                 return;
 1210         }
 1211         host->nh_monstate = NLM_MONITORED;
 1212         mtx_unlock(&host->nh_lock);
 1213 
 1214         NLM_DEBUG(1, "NLM: monitoring %s (sysid %d)\n",
 1215             host->nh_caller_name, host->nh_sysid);
 1216 
 1217         /*
 1218          * We put our assigned system ID value in the priv field to
 1219          * make it simpler to find the host if we are notified of a
 1220          * host restart.
 1221          */
 1222         smmon.mon_id.mon_name = host->nh_caller_name;
 1223         smmon.mon_id.my_id.my_name = "localhost";
 1224         smmon.mon_id.my_id.my_prog = NLM_PROG;
 1225         smmon.mon_id.my_id.my_vers = NLM_SM;
 1226         smmon.mon_id.my_id.my_proc = NLM_SM_NOTIFY;
 1227         memcpy(smmon.priv, &host->nh_sysid, sizeof(host->nh_sysid));
 1228 
 1229         timo.tv_sec = 25;
 1230         timo.tv_usec = 0;
 1231         stat = CLNT_CALL(nlm_nsm, SM_MON,
 1232             (xdrproc_t) xdr_mon, &smmon,
 1233             (xdrproc_t) xdr_sm_stat, &smstat, timo);
 1234 
 1235         if (stat != RPC_SUCCESS) {
 1236                 NLM_ERR("Failed to contact local NSM - rpc error %d\n", stat);
 1237                 return;
 1238         }
 1239         if (smstat.res_stat == stat_fail) {
 1240                 NLM_ERR("Local NSM refuses to monitor %s\n",
 1241                     host->nh_caller_name);
 1242                 mtx_lock(&host->nh_lock);
 1243                 host->nh_monstate = NLM_MONITOR_FAILED;
 1244                 mtx_unlock(&host->nh_lock);
 1245                 return;
 1246         }
 1247 
 1248         host->nh_monstate = NLM_MONITORED;
 1249 }
 1250 
 1251 /*
 1252  * Return an RPC client handle that can be used to talk to the NLM
 1253  * running on the given host.
 1254  */
 1255 CLIENT *
 1256 nlm_host_get_rpc(struct nlm_host *host, bool_t isserver)
 1257 {
 1258         struct nlm_rpc *rpc;
 1259         CLIENT *client;
 1260 
 1261         mtx_lock(&host->nh_lock);
 1262 
 1263         if (isserver)
 1264                 rpc = &host->nh_srvrpc;
 1265         else
 1266                 rpc = &host->nh_clntrpc;
 1267 
 1268         /*
 1269          * We can't hold onto RPC handles for too long - the async
 1270          * call/reply protocol used by some NLM clients makes it hard
 1271          * to tell when they change port numbers (e.g. after a
 1272          * reboot). Note that if a client reboots while it isn't
 1273          * holding any locks, it won't bother to notify us. We
 1274          * expire the RPC handles after two minutes.
 1275          */
 1276         if (rpc->nr_client && time_uptime > rpc->nr_create_time + 2*60) {
 1277                 client = rpc->nr_client;
 1278                 rpc->nr_client = NULL;
 1279                 mtx_unlock(&host->nh_lock);
 1280                 CLNT_RELEASE(client);
 1281                 mtx_lock(&host->nh_lock);
 1282         }
 1283 
 1284         if (!rpc->nr_client) {
 1285                 mtx_unlock(&host->nh_lock);
 1286                 client = nlm_get_rpc((struct sockaddr *)&host->nh_addr,
 1287                     NLM_PROG, host->nh_vers);
 1288                 mtx_lock(&host->nh_lock);
 1289 
 1290                 if (client) {
 1291                         if (rpc->nr_client) {
 1292                                 mtx_unlock(&host->nh_lock);
 1293                                 CLNT_DESTROY(client);
 1294                                 mtx_lock(&host->nh_lock);
 1295                         } else {
 1296                                 rpc->nr_client = client;
 1297                                 rpc->nr_create_time = time_uptime;
 1298                         }
 1299                 }
 1300         }
 1301 
 1302         client = rpc->nr_client;
 1303         if (client)
 1304                 CLNT_ACQUIRE(client);
 1305         mtx_unlock(&host->nh_lock);
 1306 
 1307         return (client);
 1308 
 1309 }
 1310 
 1311 int nlm_host_get_sysid(struct nlm_host *host)
 1312 {
 1313 
 1314         return (host->nh_sysid);
 1315 }
 1316 
 1317 int
 1318 nlm_host_get_state(struct nlm_host *host)
 1319 {
 1320 
 1321         return (host->nh_state);
 1322 }
 1323 
 1324 void *
 1325 nlm_register_wait_lock(struct nlm4_lock *lock, struct vnode *vp)
 1326 {
 1327         struct nlm_waiting_lock *nw;
 1328 
 1329         nw = malloc(sizeof(struct nlm_waiting_lock), M_NLM, M_WAITOK);
 1330         nw->nw_lock = *lock;
 1331         memcpy(&nw->nw_fh.fh_bytes, nw->nw_lock.fh.n_bytes,
 1332             nw->nw_lock.fh.n_len);
 1333         nw->nw_lock.fh.n_bytes = nw->nw_fh.fh_bytes;
 1334         nw->nw_waiting = TRUE;
 1335         nw->nw_vp = vp;
 1336         mtx_lock(&nlm_global_lock);
 1337         TAILQ_INSERT_TAIL(&nlm_waiting_locks, nw, nw_link);
 1338         mtx_unlock(&nlm_global_lock);
 1339 
 1340         return nw;
 1341 }
 1342 
 1343 void
 1344 nlm_deregister_wait_lock(void *handle)
 1345 {
 1346         struct nlm_waiting_lock *nw = handle;
 1347 
 1348         mtx_lock(&nlm_global_lock);
 1349         TAILQ_REMOVE(&nlm_waiting_locks, nw, nw_link);
 1350         mtx_unlock(&nlm_global_lock);
 1351         
 1352         free(nw, M_NLM);
 1353 }
 1354 
 1355 int
 1356 nlm_wait_lock(void *handle, int timo)
 1357 {
 1358         struct nlm_waiting_lock *nw = handle;
 1359         int error;
 1360 
 1361         /*
 1362          * If the granted message arrived before we got here,
 1363          * nw->nw_waiting will be FALSE - in that case, don't sleep.
 1364          */
 1365         mtx_lock(&nlm_global_lock);
 1366         error = 0;
 1367         if (nw->nw_waiting)
 1368                 error = msleep(nw, &nlm_global_lock, PCATCH, "nlmlock", timo);
 1369         TAILQ_REMOVE(&nlm_waiting_locks, nw, nw_link);
 1370         if (error) {
 1371                 /*
 1372                  * The granted message may arrive after the
 1373                  * interrupt/timeout but before we manage to lock the
 1374                  * mutex. Detect this by examining nw_lock.
 1375                  */
 1376                 if (!nw->nw_waiting)
 1377                         error = 0;
 1378         } else {
 1379                 /*
 1380                  * If nlm_cancel_wait is called, then error will be
 1381                  * zero but nw_waiting will still be TRUE. We
 1382                  * translate this into EINTR.
 1383                  */
 1384                 if (nw->nw_waiting)
 1385                         error = EINTR;
 1386         }
 1387         mtx_unlock(&nlm_global_lock);
 1388 
 1389         free(nw, M_NLM);
 1390 
 1391         return (error);
 1392 }
 1393 
 1394 void
 1395 nlm_cancel_wait(struct vnode *vp)
 1396 {
 1397         struct nlm_waiting_lock *nw;
 1398 
 1399         mtx_lock(&nlm_global_lock);
 1400         TAILQ_FOREACH(nw, &nlm_waiting_locks, nw_link) {
 1401                 if (nw->nw_vp == vp) {
 1402                         wakeup(nw);
 1403                 }
 1404         }
 1405         mtx_unlock(&nlm_global_lock);
 1406 }
 1407 
 1408 
 1409 /**********************************************************************/
 1410 
 1411 /*
 1412  * Syscall interface with userland.
 1413  */
 1414 
 1415 extern void nlm_prog_0(struct svc_req *rqstp, SVCXPRT *transp);
 1416 extern void nlm_prog_1(struct svc_req *rqstp, SVCXPRT *transp);
 1417 extern void nlm_prog_3(struct svc_req *rqstp, SVCXPRT *transp);
 1418 extern void nlm_prog_4(struct svc_req *rqstp, SVCXPRT *transp);
 1419 
 1420 static int
 1421 nlm_register_services(SVCPOOL *pool, int addr_count, char **addrs)
 1422 {
 1423         static rpcvers_t versions[] = {
 1424                 NLM_SM, NLM_VERS, NLM_VERSX, NLM_VERS4
 1425         };
 1426         static void (*dispatchers[])(struct svc_req *, SVCXPRT *) = {
 1427                 nlm_prog_0, nlm_prog_1, nlm_prog_3, nlm_prog_4
 1428         };
 1429         static const int version_count = sizeof(versions) / sizeof(versions[0]);
 1430 
 1431         SVCXPRT **xprts;
 1432         char netid[16];
 1433         char uaddr[128];
 1434         struct netconfig *nconf;
 1435         int i, j, error;
 1436 
 1437         if (!addr_count) {
 1438                 NLM_ERR("NLM: no service addresses given - can't start server");
 1439                 return (EINVAL);
 1440         }
 1441 
 1442         xprts = malloc(addr_count * sizeof(SVCXPRT *), M_NLM, M_WAITOK|M_ZERO);
 1443         for (i = 0; i < version_count; i++) {
 1444                 for (j = 0; j < addr_count; j++) {
 1445                         /*
 1446                          * Create transports for the first version and
 1447                          * then just register everything else to the
 1448                          * same transports.
 1449                          */
 1450                         if (i == 0) {
 1451                                 char *up;
 1452 
 1453                                 error = copyin(&addrs[2*j], &up,
 1454                                     sizeof(char*));
 1455                                 if (error)
 1456                                         goto out;
 1457                                 error = copyinstr(up, netid, sizeof(netid),
 1458                                     NULL);
 1459                                 if (error)
 1460                                         goto out;
 1461                                 error = copyin(&addrs[2*j+1], &up,
 1462                                     sizeof(char*));
 1463                                 if (error)
 1464                                         goto out;
 1465                                 error = copyinstr(up, uaddr, sizeof(uaddr),
 1466                                     NULL);
 1467                                 if (error)
 1468                                         goto out;
 1469                                 nconf = getnetconfigent(netid);
 1470                                 if (!nconf) {
 1471                                         NLM_ERR("Can't lookup netid %s\n",
 1472                                             netid);
 1473                                         error = EINVAL;
 1474                                         goto out;
 1475                                 }
 1476                                 xprts[j] = svc_tp_create(pool, dispatchers[i],
 1477                                     NLM_PROG, versions[i], uaddr, nconf);
 1478                                 if (!xprts[j]) {
 1479                                         NLM_ERR("NLM: unable to create "
 1480                                             "(NLM_PROG, %d).\n", versions[i]);
 1481                                         error = EINVAL;
 1482                                         goto out;
 1483                                 }
 1484                                 freenetconfigent(nconf);
 1485                         } else {
 1486                                 nconf = getnetconfigent(xprts[j]->xp_netid);
 1487                                 rpcb_unset(NLM_PROG, versions[i], nconf);
 1488                                 if (!svc_reg(xprts[j], NLM_PROG, versions[i],
 1489                                         dispatchers[i], nconf)) {
 1490                                         NLM_ERR("NLM: can't register "
 1491                                             "(NLM_PROG, %d)\n", versions[i]);
 1492                                         error = EINVAL;
 1493                                         goto out;
 1494                                 }
 1495                         }
 1496                 }
 1497         }
 1498         error = 0;
 1499 out:
 1500         for (j = 0; j < addr_count; j++) {
 1501                 if (xprts[j])
 1502                         SVC_RELEASE(xprts[j]);
 1503         }
 1504         free(xprts, M_NLM);
 1505         return (error);
 1506 }
 1507 
 1508 /*
 1509  * Main server entry point. Contacts the local NSM to get its current
 1510  * state and send SM_UNMON_ALL. Registers the NLM services and then
 1511  * services requests. Does not return until the server is interrupted
 1512  * by a signal.
 1513  */
 1514 static int
 1515 nlm_server_main(int addr_count, char **addrs)
 1516 {
 1517         struct thread *td = curthread;
 1518         int error;
 1519         SVCPOOL *pool = NULL;
 1520         struct sockopt opt;
 1521         int portlow;
 1522 #ifdef INET6
 1523         struct sockaddr_in6 sin6;
 1524 #endif
 1525         struct sockaddr_in sin;
 1526         my_id id;
 1527         sm_stat smstat;
 1528         struct timeval timo;
 1529         enum clnt_stat stat;
 1530         struct nlm_host *host, *nhost;
 1531         struct nlm_waiting_lock *nw;
 1532         vop_advlock_t *old_nfs_advlock;
 1533         vop_reclaim_t *old_nfs_reclaim;
 1534 
 1535         if (nlm_is_running != 0) {
 1536                 NLM_ERR("NLM: can't start server - "
 1537                     "it appears to be running already\n");
 1538                 return (EPERM);
 1539         }
 1540 
 1541         if (nlm_socket == NULL) {
 1542                 memset(&opt, 0, sizeof(opt));
 1543 
 1544                 error = socreate(AF_INET, &nlm_socket, SOCK_DGRAM, 0,
 1545                     td->td_ucred, td);
 1546                 if (error) {
 1547                         NLM_ERR("NLM: can't create IPv4 socket - error %d\n",
 1548                             error);
 1549                         return (error);
 1550                 }
 1551                 opt.sopt_dir = SOPT_SET;
 1552                 opt.sopt_level = IPPROTO_IP;
 1553                 opt.sopt_name = IP_PORTRANGE;
 1554                 portlow = IP_PORTRANGE_LOW;
 1555                 opt.sopt_val = &portlow;
 1556                 opt.sopt_valsize = sizeof(portlow);
 1557                 sosetopt(nlm_socket, &opt);
 1558 
 1559 #ifdef INET6
 1560                 nlm_socket6 = NULL;
 1561                 error = socreate(AF_INET6, &nlm_socket6, SOCK_DGRAM, 0,
 1562                     td->td_ucred, td);
 1563                 if (error) {
 1564                         NLM_ERR("NLM: can't create IPv6 socket - error %d\n",
 1565                             error);
 1566                         soclose(nlm_socket);
 1567                         nlm_socket = NULL;
 1568                         return (error);
 1569                 }
 1570                 opt.sopt_dir = SOPT_SET;
 1571                 opt.sopt_level = IPPROTO_IPV6;
 1572                 opt.sopt_name = IPV6_PORTRANGE;
 1573                 portlow = IPV6_PORTRANGE_LOW;
 1574                 opt.sopt_val = &portlow;
 1575                 opt.sopt_valsize = sizeof(portlow);
 1576                 sosetopt(nlm_socket6, &opt);
 1577 #endif
 1578         }
 1579 
 1580         nlm_auth = authunix_create(curthread->td_ucred);
 1581 
 1582 #ifdef INET6
 1583         memset(&sin6, 0, sizeof(sin6));
 1584         sin6.sin6_len = sizeof(sin6);
 1585         sin6.sin6_family = AF_INET6;
 1586         sin6.sin6_addr = in6addr_loopback;
 1587         nlm_nsm = nlm_get_rpc((struct sockaddr *) &sin6, SM_PROG, SM_VERS);
 1588         if (!nlm_nsm) {
 1589 #endif
 1590                 memset(&sin, 0, sizeof(sin));
 1591                 sin.sin_len = sizeof(sin);
 1592                 sin.sin_family = AF_INET;
 1593                 sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
 1594                 nlm_nsm = nlm_get_rpc((struct sockaddr *) &sin, SM_PROG,
 1595                     SM_VERS);
 1596 #ifdef INET6
 1597         }
 1598 #endif
 1599 
 1600         if (!nlm_nsm) {
 1601                 NLM_ERR("Can't start NLM - unable to contact NSM\n");
 1602                 error = EINVAL;
 1603                 goto out;
 1604         }
 1605 
 1606         pool = svcpool_create("NLM", NULL);
 1607 
 1608         error = nlm_register_services(pool, addr_count, addrs);
 1609         if (error)
 1610                 goto out;
 1611 
 1612         memset(&id, 0, sizeof(id));
 1613         id.my_name = "NFS NLM";
 1614 
 1615         timo.tv_sec = 25;
 1616         timo.tv_usec = 0;
 1617         stat = CLNT_CALL(nlm_nsm, SM_UNMON_ALL,
 1618             (xdrproc_t) xdr_my_id, &id,
 1619             (xdrproc_t) xdr_sm_stat, &smstat, timo);
 1620 
 1621         if (stat != RPC_SUCCESS) {
 1622                 struct rpc_err err;
 1623 
 1624                 CLNT_GETERR(nlm_nsm, &err);
 1625                 NLM_ERR("NLM: unexpected error contacting NSM, "
 1626                     "stat=%d, errno=%d\n", stat, err.re_errno);
 1627                 error = EINVAL;
 1628                 goto out;
 1629         }
 1630         nlm_is_running = 1;
 1631 
 1632         NLM_DEBUG(1, "NLM: local NSM state is %d\n", smstat.state);
 1633         nlm_nsm_state = smstat.state;
 1634 
 1635         old_nfs_advlock = nfs_advlock_p;
 1636         nfs_advlock_p = nlm_advlock;
 1637         old_nfs_reclaim = nfs_reclaim_p;
 1638         nfs_reclaim_p = nlm_reclaim;
 1639 
 1640         svc_run(pool);
 1641         error = 0;
 1642 
 1643         nfs_advlock_p = old_nfs_advlock;
 1644         nfs_reclaim_p = old_nfs_reclaim;
 1645 
 1646 out:
 1647         nlm_is_running = 0;
 1648         if (pool)
 1649                 svcpool_destroy(pool);
 1650 
 1651         /*
 1652          * We are finished communicating with the NSM.
 1653          */
 1654         if (nlm_nsm) {
 1655                 CLNT_RELEASE(nlm_nsm);
 1656                 nlm_nsm = NULL;
 1657         }
 1658 
 1659         /*
 1660          * Trash all the existing state so that if the server
 1661          * restarts, it gets a clean slate. This is complicated by the
 1662          * possibility that there may be other threads trying to make
 1663          * client locking requests.
 1664          *
 1665          * First we fake a client reboot notification which will
 1666          * cancel any pending async locks and purge remote lock state
 1667          * from the local lock manager. We release the reference from
 1668          * nlm_hosts to the host (which may remove it from the list
 1669          * and free it). After this phase, the only entries in the
 1670          * nlm_host list should be from other threads performing
 1671          * client lock requests.
 1672          */
 1673         mtx_lock(&nlm_global_lock);
 1674         TAILQ_FOREACH(nw, &nlm_waiting_locks, nw_link) {
 1675                 wakeup(nw);
 1676         }
 1677         TAILQ_FOREACH_SAFE(host, &nlm_hosts, nh_link, nhost) {
 1678                 mtx_unlock(&nlm_global_lock);
 1679                 nlm_host_notify(host, 0);
 1680                 nlm_host_release(host);
 1681                 mtx_lock(&nlm_global_lock);
 1682         }
 1683         mtx_unlock(&nlm_global_lock);
 1684 
 1685         AUTH_DESTROY(nlm_auth);
 1686 
 1687         return (error);
 1688 }
 1689 
 1690 int
 1691 sys_nlm_syscall(struct thread *td, struct nlm_syscall_args *uap)
 1692 {
 1693         int error;
 1694 
 1695 #if __FreeBSD_version >= 700000
 1696         error = priv_check(td, PRIV_NFS_LOCKD);
 1697 #else
 1698         error = suser(td);
 1699 #endif
 1700         if (error)
 1701                 return (error);
 1702 
 1703         nlm_debug_level = uap->debug_level;
 1704         nlm_grace_threshold = time_uptime + uap->grace_period;
 1705         nlm_next_idle_check = time_uptime + NLM_IDLE_PERIOD;
 1706 
 1707         return nlm_server_main(uap->addr_count, uap->addrs);
 1708 }
 1709 
 1710 /**********************************************************************/
 1711 
 1712 /*
 1713  * NLM implementation details, called from the RPC stubs.
 1714  */
 1715 
 1716 
 1717 void
 1718 nlm_sm_notify(struct nlm_sm_status *argp)
 1719 {
 1720         uint32_t sysid;
 1721         struct nlm_host *host;
 1722 
 1723         NLM_DEBUG(3, "nlm_sm_notify(): mon_name = %s\n", argp->mon_name);
 1724         memcpy(&sysid, &argp->priv, sizeof(sysid));
 1725         host = nlm_find_host_by_sysid(sysid);
 1726         if (host) {
 1727                 nlm_host_notify(host, argp->state);
 1728                 nlm_host_release(host);
 1729         }
 1730 }
 1731 
 1732 static void
 1733 nlm_convert_to_fhandle_t(fhandle_t *fhp, struct netobj *p)
 1734 {
 1735         memcpy(fhp, p->n_bytes, sizeof(fhandle_t));
 1736 }
 1737 
 1738 struct vfs_state {
 1739         struct mount    *vs_mp;
 1740         struct vnode    *vs_vp;
 1741         int             vs_vfslocked;
 1742         int             vs_vnlocked;
 1743 };
 1744 
 1745 static int
 1746 nlm_get_vfs_state(struct nlm_host *host, struct svc_req *rqstp,
 1747     fhandle_t *fhp, struct vfs_state *vs, accmode_t accmode)
 1748 {
 1749         int error, exflags;
 1750         struct ucred *cred = NULL, *credanon = NULL;
 1751         
 1752         memset(vs, 0, sizeof(*vs));
 1753 
 1754         vs->vs_mp = vfs_getvfs(&fhp->fh_fsid);
 1755         if (!vs->vs_mp) {
 1756                 return (ESTALE);
 1757         }
 1758         vs->vs_vfslocked = VFS_LOCK_GIANT(vs->vs_mp);
 1759 
 1760         /* accmode == 0 means don't check, since it is an unlock. */
 1761         if (accmode != 0) {
 1762                 error = VFS_CHECKEXP(vs->vs_mp,
 1763                     (struct sockaddr *)&host->nh_addr, &exflags, &credanon,
 1764                     NULL, NULL);
 1765                 if (error)
 1766                         goto out;
 1767 
 1768                 if (exflags & MNT_EXRDONLY ||
 1769                     (vs->vs_mp->mnt_flag & MNT_RDONLY)) {
 1770                         error = EROFS;
 1771                         goto out;
 1772                 }
 1773         }
 1774 
 1775         error = VFS_FHTOVP(vs->vs_mp, &fhp->fh_fid, LK_EXCLUSIVE, &vs->vs_vp);
 1776         if (error)
 1777                 goto out;
 1778         vs->vs_vnlocked = TRUE;
 1779 
 1780         if (accmode != 0) {
 1781                 if (!svc_getcred(rqstp, &cred, NULL)) {
 1782                         error = EINVAL;
 1783                         goto out;
 1784                 }
 1785                 if (cred->cr_uid == 0 || (exflags & MNT_EXPORTANON)) {
 1786                         crfree(cred);
 1787                         cred = credanon;
 1788                         credanon = NULL;
 1789                 }
 1790 
 1791                 /*
 1792                  * Check cred.
 1793                  */
 1794                 error = VOP_ACCESS(vs->vs_vp, accmode, cred, curthread);
 1795                 /*
 1796                  * If this failed and accmode != VWRITE, try again with
 1797                  * VWRITE to maintain backwards compatibility with the
 1798                  * old code that always used VWRITE.
 1799                  */
 1800                 if (error != 0 && accmode != VWRITE)
 1801                         error = VOP_ACCESS(vs->vs_vp, VWRITE, cred, curthread);
 1802                 if (error)
 1803                         goto out;
 1804         }
 1805 
 1806 #if __FreeBSD_version < 800011
 1807         VOP_UNLOCK(vs->vs_vp, 0, curthread);
 1808 #else
 1809         VOP_UNLOCK(vs->vs_vp, 0);
 1810 #endif
 1811         vs->vs_vnlocked = FALSE;
 1812 
 1813 out:
 1814         if (cred)
 1815                 crfree(cred);
 1816         if (credanon)
 1817                 crfree(credanon);
 1818 
 1819         return (error);
 1820 }
 1821 
 1822 static void
 1823 nlm_release_vfs_state(struct vfs_state *vs)
 1824 {
 1825 
 1826         if (vs->vs_vp) {
 1827                 if (vs->vs_vnlocked)
 1828                         vput(vs->vs_vp);
 1829                 else
 1830                         vrele(vs->vs_vp);
 1831         }
 1832         if (vs->vs_mp)
 1833                 vfs_rel(vs->vs_mp);
 1834         VFS_UNLOCK_GIANT(vs->vs_vfslocked);
 1835 }
 1836 
 1837 static nlm4_stats
 1838 nlm_convert_error(int error)
 1839 {
 1840 
 1841         if (error == ESTALE)
 1842                 return nlm4_stale_fh;
 1843         else if (error == EROFS)
 1844                 return nlm4_rofs;
 1845         else
 1846                 return nlm4_failed;
 1847 }
 1848 
 1849 int
 1850 nlm_do_test(nlm4_testargs *argp, nlm4_testres *result, struct svc_req *rqstp,
 1851         CLIENT **rpcp)
 1852 {
 1853         fhandle_t fh;
 1854         struct vfs_state vs;
 1855         struct nlm_host *host, *bhost;
 1856         int error, sysid;
 1857         struct flock fl;
 1858         accmode_t accmode;
 1859         
 1860         memset(result, 0, sizeof(*result));
 1861         memset(&vs, 0, sizeof(vs));
 1862 
 1863         host = nlm_find_host_by_name(argp->alock.caller_name,
 1864             svc_getrpccaller(rqstp), rqstp->rq_vers);
 1865         if (!host) {
 1866                 result->stat.stat = nlm4_denied_nolocks;
 1867                 return (ENOMEM);
 1868         }
 1869 
 1870         NLM_DEBUG(3, "nlm_do_test(): caller_name = %s (sysid = %d)\n",
 1871             host->nh_caller_name, host->nh_sysid);
 1872 
 1873         nlm_check_expired_locks(host);
 1874         sysid = host->nh_sysid;
 1875 
 1876         nlm_convert_to_fhandle_t(&fh, &argp->alock.fh);
 1877         nlm_copy_netobj(&result->cookie, &argp->cookie, M_RPC);
 1878 
 1879         if (time_uptime < nlm_grace_threshold) {
 1880                 result->stat.stat = nlm4_denied_grace_period;
 1881                 goto out;
 1882         }
 1883 
 1884         accmode = argp->exclusive ? VWRITE : VREAD;
 1885         error = nlm_get_vfs_state(host, rqstp, &fh, &vs, accmode);
 1886         if (error) {
 1887                 result->stat.stat = nlm_convert_error(error);
 1888                 goto out;
 1889         }
 1890 
 1891         fl.l_start = argp->alock.l_offset;
 1892         fl.l_len = argp->alock.l_len;
 1893         fl.l_pid = argp->alock.svid;
 1894         fl.l_sysid = sysid;
 1895         fl.l_whence = SEEK_SET;
 1896         if (argp->exclusive)
 1897                 fl.l_type = F_WRLCK;
 1898         else
 1899                 fl.l_type = F_RDLCK;
 1900         error = VOP_ADVLOCK(vs.vs_vp, NULL, F_GETLK, &fl, F_REMOTE);
 1901         if (error) {
 1902                 result->stat.stat = nlm4_failed;
 1903                 goto out;
 1904         }
 1905 
 1906         if (fl.l_type == F_UNLCK) {
 1907                 result->stat.stat = nlm4_granted;
 1908         } else {
 1909                 result->stat.stat = nlm4_denied;
 1910                 result->stat.nlm4_testrply_u.holder.exclusive =
 1911                         (fl.l_type == F_WRLCK);
 1912                 result->stat.nlm4_testrply_u.holder.svid = fl.l_pid;
 1913                 bhost = nlm_find_host_by_sysid(fl.l_sysid);
 1914                 if (bhost) {
 1915                         /*
 1916                          * We don't have any useful way of recording
 1917                          * the value of oh used in the original lock
 1918                          * request. Ideally, the test reply would have
 1919                          * a space for the owning host's name allowing
 1920                          * our caller's NLM to keep track.
 1921                          *
 1922                          * As far as I can see, Solaris uses an eight
 1923                          * byte structure for oh which contains a four
 1924                          * byte pid encoded in local byte order and
 1925                          * the first four bytes of the host
 1926                          * name. Linux uses a variable length string
 1927                          * 'pid@hostname' in ascii but doesn't even
 1928                          * return that in test replies.
 1929                          *
 1930                          * For the moment, return nothing in oh
 1931                          * (already zero'ed above).
 1932                          */
 1933                         nlm_host_release(bhost);
 1934                 }
 1935                 result->stat.nlm4_testrply_u.holder.l_offset = fl.l_start;
 1936                 result->stat.nlm4_testrply_u.holder.l_len = fl.l_len;
 1937         }
 1938 
 1939 out:
 1940         nlm_release_vfs_state(&vs);
 1941         if (rpcp)
 1942                 *rpcp = nlm_host_get_rpc(host, TRUE);
 1943         nlm_host_release(host);
 1944         return (0);
 1945 }
 1946 
 1947 int
 1948 nlm_do_lock(nlm4_lockargs *argp, nlm4_res *result, struct svc_req *rqstp,
 1949     bool_t monitor, CLIENT **rpcp)
 1950 {
 1951         fhandle_t fh;
 1952         struct vfs_state vs;
 1953         struct nlm_host *host;
 1954         int error, sysid;
 1955         struct flock fl;
 1956         accmode_t accmode;
 1957         
 1958         memset(result, 0, sizeof(*result));
 1959         memset(&vs, 0, sizeof(vs));
 1960 
 1961         host = nlm_find_host_by_name(argp->alock.caller_name,
 1962             svc_getrpccaller(rqstp), rqstp->rq_vers);
 1963         if (!host) {
 1964                 result->stat.stat = nlm4_denied_nolocks;
 1965                 return (ENOMEM);
 1966         }
 1967 
 1968         NLM_DEBUG(3, "nlm_do_lock(): caller_name = %s (sysid = %d)\n",
 1969             host->nh_caller_name, host->nh_sysid);
 1970 
 1971         if (monitor && host->nh_state && argp->state
 1972             && host->nh_state != argp->state) {
 1973                 /*
 1974                  * The host rebooted without telling us. Trash its
 1975                  * locks.
 1976                  */
 1977                 nlm_host_notify(host, argp->state);
 1978         }
 1979 
 1980         nlm_check_expired_locks(host);
 1981         sysid = host->nh_sysid;
 1982 
 1983         nlm_convert_to_fhandle_t(&fh, &argp->alock.fh);
 1984         nlm_copy_netobj(&result->cookie, &argp->cookie, M_RPC);
 1985 
 1986         if (time_uptime < nlm_grace_threshold && !argp->reclaim) {
 1987                 result->stat.stat = nlm4_denied_grace_period;
 1988                 goto out;
 1989         }
 1990 
 1991         accmode = argp->exclusive ? VWRITE : VREAD;
 1992         error = nlm_get_vfs_state(host, rqstp, &fh, &vs, accmode);
 1993         if (error) {
 1994                 result->stat.stat = nlm_convert_error(error);
 1995                 goto out;
 1996         }
 1997 
 1998         fl.l_start = argp->alock.l_offset;
 1999         fl.l_len = argp->alock.l_len;
 2000         fl.l_pid = argp->alock.svid;
 2001         fl.l_sysid = sysid;
 2002         fl.l_whence = SEEK_SET;
 2003         if (argp->exclusive)
 2004                 fl.l_type = F_WRLCK;
 2005         else
 2006                 fl.l_type = F_RDLCK;
 2007         if (argp->block) {
 2008                 struct nlm_async_lock *af;
 2009                 CLIENT *client;
 2010                 struct nlm_grantcookie cookie;
 2011 
 2012                 /*
 2013                  * First, make sure we can contact the host's NLM.
 2014                  */
 2015                 client = nlm_host_get_rpc(host, TRUE);
 2016                 if (!client) {
 2017                         result->stat.stat = nlm4_failed;
 2018                         goto out;
 2019                 }
 2020 
 2021                 /*
 2022                  * First we need to check and see if there is an
 2023                  * existing blocked lock that matches. This could be a
 2024                  * badly behaved client or an RPC re-send. If we find
 2025                  * one, just return nlm4_blocked.
 2026                  */
 2027                 mtx_lock(&host->nh_lock);
 2028                 TAILQ_FOREACH(af, &host->nh_pending, af_link) {
 2029                         if (af->af_fl.l_start == fl.l_start
 2030                             && af->af_fl.l_len == fl.l_len
 2031                             && af->af_fl.l_pid == fl.l_pid
 2032                             && af->af_fl.l_type == fl.l_type) {
 2033                                 break;
 2034                         }
 2035                 }
 2036                 if (!af) {
 2037                         cookie.ng_sysid = host->nh_sysid;
 2038                         cookie.ng_cookie = host->nh_grantcookie++;
 2039                 }
 2040                 mtx_unlock(&host->nh_lock);
 2041                 if (af) {
 2042                         CLNT_RELEASE(client);
 2043                         result->stat.stat = nlm4_blocked;
 2044                         goto out;
 2045                 }
 2046 
 2047                 af = malloc(sizeof(struct nlm_async_lock), M_NLM,
 2048                     M_WAITOK|M_ZERO);
 2049                 TASK_INIT(&af->af_task, 0, nlm_lock_callback, af);
 2050                 af->af_vp = vs.vs_vp;
 2051                 af->af_fl = fl;
 2052                 af->af_host = host;
 2053                 af->af_rpc = client;
 2054                 /*
 2055                  * We use M_RPC here so that we can xdr_free the thing
 2056                  * later.
 2057                  */
 2058                 nlm_make_netobj(&af->af_granted.cookie,
 2059                     (caddr_t)&cookie, sizeof(cookie), M_RPC);
 2060                 af->af_granted.exclusive = argp->exclusive;
 2061                 af->af_granted.alock.caller_name =
 2062                         strdup(argp->alock.caller_name, M_RPC);
 2063                 nlm_copy_netobj(&af->af_granted.alock.fh,
 2064                     &argp->alock.fh, M_RPC);
 2065                 nlm_copy_netobj(&af->af_granted.alock.oh,
 2066                     &argp->alock.oh, M_RPC);
 2067                 af->af_granted.alock.svid = argp->alock.svid;
 2068                 af->af_granted.alock.l_offset = argp->alock.l_offset;
 2069                 af->af_granted.alock.l_len = argp->alock.l_len;
 2070 
 2071                 /*
 2072                  * Put the entry on the pending list before calling
 2073                  * VOP_ADVLOCKASYNC. We do this in case the lock
 2074                  * request was blocked (returning EINPROGRESS) but
 2075                  * then granted before we manage to run again. The
 2076                  * client may receive the granted message before we
 2077                  * send our blocked reply but thats their problem.
 2078                  */
 2079                 mtx_lock(&host->nh_lock);
 2080                 TAILQ_INSERT_TAIL(&host->nh_pending, af, af_link);
 2081                 mtx_unlock(&host->nh_lock);
 2082 
 2083                 error = VOP_ADVLOCKASYNC(vs.vs_vp, NULL, F_SETLK, &fl, F_REMOTE,
 2084                     &af->af_task, &af->af_cookie);
 2085 
 2086                 /*
 2087                  * If the lock completed synchronously, just free the
 2088                  * tracking structure now.
 2089                  */
 2090                 if (error != EINPROGRESS) {
 2091                         CLNT_RELEASE(af->af_rpc);
 2092                         mtx_lock(&host->nh_lock);
 2093                         TAILQ_REMOVE(&host->nh_pending, af, af_link);
 2094                         mtx_unlock(&host->nh_lock);
 2095                         xdr_free((xdrproc_t) xdr_nlm4_testargs,
 2096                             &af->af_granted);
 2097                         free(af, M_NLM);
 2098                 } else {
 2099                         NLM_DEBUG(2, "NLM: pending async lock %p for %s "
 2100                             "(sysid %d)\n", af, host->nh_caller_name, sysid);
 2101                         /*
 2102                          * Don't vrele the vnode just yet - this must
 2103                          * wait until either the async callback
 2104                          * happens or the lock is cancelled.
 2105                          */
 2106                         vs.vs_vp = NULL;
 2107                 }
 2108         } else {
 2109                 error = VOP_ADVLOCK(vs.vs_vp, NULL, F_SETLK, &fl, F_REMOTE);
 2110         }
 2111 
 2112         if (error) {
 2113                 if (error == EINPROGRESS) {
 2114                         result->stat.stat = nlm4_blocked;
 2115                 } else if (error == EDEADLK) {
 2116                         result->stat.stat = nlm4_deadlck;
 2117                 } else if (error == EAGAIN) {
 2118                         result->stat.stat = nlm4_denied;
 2119                 } else {
 2120                         result->stat.stat = nlm4_failed;
 2121                 }
 2122         } else {
 2123                 if (monitor)
 2124                         nlm_host_monitor(host, argp->state);
 2125                 result->stat.stat = nlm4_granted;
 2126         }       
 2127 
 2128 out:
 2129         nlm_release_vfs_state(&vs);
 2130         if (rpcp)
 2131                 *rpcp = nlm_host_get_rpc(host, TRUE);
 2132         nlm_host_release(host);
 2133         return (0);
 2134 }
 2135 
 2136 int
 2137 nlm_do_cancel(nlm4_cancargs *argp, nlm4_res *result, struct svc_req *rqstp,
 2138     CLIENT **rpcp)
 2139 {
 2140         fhandle_t fh;
 2141         struct vfs_state vs;
 2142         struct nlm_host *host;
 2143         int error, sysid;
 2144         struct flock fl;
 2145         struct nlm_async_lock *af;
 2146         
 2147         memset(result, 0, sizeof(*result));
 2148         memset(&vs, 0, sizeof(vs));
 2149 
 2150         host = nlm_find_host_by_name(argp->alock.caller_name,
 2151             svc_getrpccaller(rqstp), rqstp->rq_vers);
 2152         if (!host) {
 2153                 result->stat.stat = nlm4_denied_nolocks;
 2154                 return (ENOMEM);
 2155         }
 2156 
 2157         NLM_DEBUG(3, "nlm_do_cancel(): caller_name = %s (sysid = %d)\n",
 2158             host->nh_caller_name, host->nh_sysid);
 2159 
 2160         nlm_check_expired_locks(host);
 2161         sysid = host->nh_sysid;
 2162 
 2163         nlm_convert_to_fhandle_t(&fh, &argp->alock.fh);
 2164         nlm_copy_netobj(&result->cookie, &argp->cookie, M_RPC);
 2165 
 2166         if (time_uptime < nlm_grace_threshold) {
 2167                 result->stat.stat = nlm4_denied_grace_period;
 2168                 goto out;
 2169         }
 2170 
 2171         error = nlm_get_vfs_state(host, rqstp, &fh, &vs, (accmode_t)0);
 2172         if (error) {
 2173                 result->stat.stat = nlm_convert_error(error);
 2174                 goto out;
 2175         }
 2176 
 2177         fl.l_start = argp->alock.l_offset;
 2178         fl.l_len = argp->alock.l_len;
 2179         fl.l_pid = argp->alock.svid;
 2180         fl.l_sysid = sysid;
 2181         fl.l_whence = SEEK_SET;
 2182         if (argp->exclusive)
 2183                 fl.l_type = F_WRLCK;
 2184         else
 2185                 fl.l_type = F_RDLCK;
 2186 
 2187         /*
 2188          * First we need to try and find the async lock request - if
 2189          * there isn't one, we give up and return nlm4_denied.
 2190          */
 2191         mtx_lock(&host->nh_lock);
 2192 
 2193         TAILQ_FOREACH(af, &host->nh_pending, af_link) {
 2194                 if (af->af_fl.l_start == fl.l_start
 2195                     && af->af_fl.l_len == fl.l_len
 2196                     && af->af_fl.l_pid == fl.l_pid
 2197                     && af->af_fl.l_type == fl.l_type) {
 2198                         break;
 2199                 }
 2200         }
 2201 
 2202         if (!af) {
 2203                 mtx_unlock(&host->nh_lock);
 2204                 result->stat.stat = nlm4_denied;
 2205                 goto out;
 2206         }
 2207 
 2208         error = nlm_cancel_async_lock(af);
 2209 
 2210         if (error) {
 2211                 result->stat.stat = nlm4_denied;
 2212         } else {
 2213                 result->stat.stat = nlm4_granted;
 2214         }
 2215 
 2216         mtx_unlock(&host->nh_lock);
 2217 
 2218 out:
 2219         nlm_release_vfs_state(&vs);
 2220         if (rpcp)
 2221                 *rpcp = nlm_host_get_rpc(host, TRUE);
 2222         nlm_host_release(host);
 2223         return (0);
 2224 }
 2225 
 2226 int
 2227 nlm_do_unlock(nlm4_unlockargs *argp, nlm4_res *result, struct svc_req *rqstp,
 2228     CLIENT **rpcp)
 2229 {
 2230         fhandle_t fh;
 2231         struct vfs_state vs;
 2232         struct nlm_host *host;
 2233         int error, sysid;
 2234         struct flock fl;
 2235         
 2236         memset(result, 0, sizeof(*result));
 2237         memset(&vs, 0, sizeof(vs));
 2238 
 2239         host = nlm_find_host_by_name(argp->alock.caller_name,
 2240             svc_getrpccaller(rqstp), rqstp->rq_vers);
 2241         if (!host) {
 2242                 result->stat.stat = nlm4_denied_nolocks;
 2243                 return (ENOMEM);
 2244         }
 2245 
 2246         NLM_DEBUG(3, "nlm_do_unlock(): caller_name = %s (sysid = %d)\n",
 2247             host->nh_caller_name, host->nh_sysid);
 2248 
 2249         nlm_check_expired_locks(host);
 2250         sysid = host->nh_sysid;
 2251 
 2252         nlm_convert_to_fhandle_t(&fh, &argp->alock.fh);
 2253         nlm_copy_netobj(&result->cookie, &argp->cookie, M_RPC);
 2254 
 2255         if (time_uptime < nlm_grace_threshold) {
 2256                 result->stat.stat = nlm4_denied_grace_period;
 2257                 goto out;
 2258         }
 2259 
 2260         error = nlm_get_vfs_state(host, rqstp, &fh, &vs, (accmode_t)0);
 2261         if (error) {
 2262                 result->stat.stat = nlm_convert_error(error);
 2263                 goto out;
 2264         }
 2265 
 2266         fl.l_start = argp->alock.l_offset;
 2267         fl.l_len = argp->alock.l_len;
 2268         fl.l_pid = argp->alock.svid;
 2269         fl.l_sysid = sysid;
 2270         fl.l_whence = SEEK_SET;
 2271         fl.l_type = F_UNLCK;
 2272         error = VOP_ADVLOCK(vs.vs_vp, NULL, F_UNLCK, &fl, F_REMOTE);
 2273 
 2274         /*
 2275          * Ignore the error - there is no result code for failure,
 2276          * only for grace period.
 2277          */
 2278         result->stat.stat = nlm4_granted;
 2279 
 2280 out:
 2281         nlm_release_vfs_state(&vs);
 2282         if (rpcp)
 2283                 *rpcp = nlm_host_get_rpc(host, TRUE);
 2284         nlm_host_release(host);
 2285         return (0);
 2286 }
 2287 
 2288 int
 2289 nlm_do_granted(nlm4_testargs *argp, nlm4_res *result, struct svc_req *rqstp,
 2290 
 2291     CLIENT **rpcp)
 2292 {
 2293         struct nlm_host *host;
 2294         struct nlm_waiting_lock *nw;
 2295         
 2296         memset(result, 0, sizeof(*result));
 2297 
 2298         host = nlm_find_host_by_addr(svc_getrpccaller(rqstp), rqstp->rq_vers);
 2299         if (!host) {
 2300                 result->stat.stat = nlm4_denied_nolocks;
 2301                 return (ENOMEM);
 2302         }
 2303 
 2304         nlm_copy_netobj(&result->cookie, &argp->cookie, M_RPC);
 2305         result->stat.stat = nlm4_denied;
 2306         KFAIL_POINT_CODE(DEBUG_FP, nlm_deny_grant, goto out);
 2307 
 2308         mtx_lock(&nlm_global_lock);
 2309         TAILQ_FOREACH(nw, &nlm_waiting_locks, nw_link) {
 2310                 if (!nw->nw_waiting)
 2311                         continue;
 2312                 if (argp->alock.svid == nw->nw_lock.svid
 2313                     && argp->alock.l_offset == nw->nw_lock.l_offset
 2314                     && argp->alock.l_len == nw->nw_lock.l_len
 2315                     && argp->alock.fh.n_len == nw->nw_lock.fh.n_len
 2316                     && !memcmp(argp->alock.fh.n_bytes, nw->nw_lock.fh.n_bytes,
 2317                         nw->nw_lock.fh.n_len)) {
 2318                         nw->nw_waiting = FALSE;
 2319                         wakeup(nw);
 2320                         result->stat.stat = nlm4_granted;
 2321                         break;
 2322                 }
 2323         }
 2324         mtx_unlock(&nlm_global_lock);
 2325 
 2326 out:
 2327         if (rpcp)
 2328                 *rpcp = nlm_host_get_rpc(host, TRUE);
 2329         nlm_host_release(host);
 2330         return (0);
 2331 }
 2332 
 2333 void
 2334 nlm_do_granted_res(nlm4_res *argp, struct svc_req *rqstp)
 2335 {
 2336         struct nlm_host *host = NULL;
 2337         struct nlm_async_lock *af = NULL;
 2338         int error;
 2339 
 2340         if (argp->cookie.n_len != sizeof(struct nlm_grantcookie)) {
 2341                 NLM_DEBUG(1, "NLM: bogus grant cookie");
 2342                 goto out;
 2343         }
 2344 
 2345         host = nlm_find_host_by_sysid(ng_sysid(&argp->cookie));
 2346         if (!host) {
 2347                 NLM_DEBUG(1, "NLM: Unknown host rejected our grant");
 2348                 goto out;
 2349         }
 2350 
 2351         mtx_lock(&host->nh_lock);
 2352         TAILQ_FOREACH(af, &host->nh_granted, af_link)
 2353             if (ng_cookie(&argp->cookie) ==
 2354                 ng_cookie(&af->af_granted.cookie))
 2355                     break;
 2356         if (af)
 2357                 TAILQ_REMOVE(&host->nh_granted, af, af_link);
 2358         mtx_unlock(&host->nh_lock);
 2359 
 2360         if (!af) {
 2361                 NLM_DEBUG(1, "NLM: host %s (sysid %d) replied to our grant "
 2362                     "with unrecognized cookie %d:%d", host->nh_caller_name,
 2363                     host->nh_sysid, ng_sysid(&argp->cookie),
 2364                     ng_cookie(&argp->cookie));
 2365                 goto out;
 2366         }
 2367 
 2368         if (argp->stat.stat != nlm4_granted) {
 2369                 af->af_fl.l_type = F_UNLCK;
 2370                 error = VOP_ADVLOCK(af->af_vp, NULL, F_UNLCK, &af->af_fl, F_REMOTE);
 2371                 if (error) {
 2372                         NLM_DEBUG(1, "NLM: host %s (sysid %d) rejected our grant "
 2373                             "and we failed to unlock (%d)", host->nh_caller_name,
 2374                             host->nh_sysid, error);
 2375                         goto out;
 2376                 }
 2377 
 2378                 NLM_DEBUG(5, "NLM: async lock %p rejected by host %s (sysid %d)",
 2379                     af, host->nh_caller_name, host->nh_sysid);
 2380         } else {
 2381                 NLM_DEBUG(5, "NLM: async lock %p accepted by host %s (sysid %d)",
 2382                     af, host->nh_caller_name, host->nh_sysid);
 2383         }
 2384 
 2385  out:
 2386         if (af)
 2387                 nlm_free_async_lock(af);
 2388         if (host)
 2389                 nlm_host_release(host);
 2390 }
 2391 
 2392 void
 2393 nlm_do_free_all(nlm4_notify *argp)
 2394 {
 2395         struct nlm_host *host, *thost;
 2396 
 2397         TAILQ_FOREACH_SAFE(host, &nlm_hosts, nh_link, thost) {
 2398                 if (!strcmp(host->nh_caller_name, argp->name))
 2399                         nlm_host_notify(host, argp->state);
 2400         }
 2401 }
 2402 
 2403 /*
 2404  * Kernel module glue
 2405  */
 2406 static int
 2407 nfslockd_modevent(module_t mod, int type, void *data)
 2408 {
 2409 
 2410         switch (type) {
 2411         case MOD_LOAD:
 2412                 return (0);
 2413         case MOD_UNLOAD:
 2414                 /* The NLM module cannot be safely unloaded. */
 2415                 /* FALLTHROUGH */
 2416         default:
 2417                 return (EOPNOTSUPP);
 2418         }
 2419 }
 2420 static moduledata_t nfslockd_mod = {
 2421         "nfslockd",
 2422         nfslockd_modevent,
 2423         NULL,
 2424 };
 2425 DECLARE_MODULE(nfslockd, nfslockd_mod, SI_SUB_VFS, SI_ORDER_ANY);
 2426 
 2427 /* So that loader and kldload(2) can find us, wherever we are.. */
 2428 MODULE_DEPEND(nfslockd, krpc, 1, 1, 1);
 2429 MODULE_DEPEND(nfslockd, nfslock, 1, 1, 1);
 2430 MODULE_VERSION(nfslockd, 1);

Cache object: e79a19a24888b5ef663a8dcf4eba6e89


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]


This page is part of the FreeBSD/Linux Linux Kernel Cross-Reference, and was automatically generated using a modified version of the LXR engine.