fedora core 6 1.2949 + vserver 2.2.0
[linux-2.6.git] / fs / lockd / host.c
index 82f7a0b..3d4610c 100644 (file)
@@ -16,6 +16,7 @@
 #include <linux/sunrpc/svc.h>
 #include <linux/lockd/lockd.h>
 #include <linux/lockd/sm_inter.h>
+#include <linux/mutex.h>
 
 
 #define NLMDBG_FACILITY                NLMDBG_HOSTCACHE
 #define NLM_HOST_REBIND                (60 * HZ)
 #define NLM_HOST_EXPIRE                ((nrhosts > NLM_HOST_MAX)? 300 * HZ : 120 * HZ)
 #define NLM_HOST_COLLECT       ((nrhosts > NLM_HOST_MAX)? 120 * HZ :  60 * HZ)
-#define NLM_HOST_ADDR(sv)      (&(sv)->s_nlmclnt->cl_xprt->addr)
 
-static struct nlm_host *       nlm_hosts[NLM_HOST_NRHASH];
+static struct hlist_head       nlm_hosts[NLM_HOST_NRHASH];
 static unsigned long           next_gc;
 static int                     nrhosts;
-static DECLARE_MUTEX(nlm_host_sema);
+static DEFINE_MUTEX(nlm_host_mutex);
 
 
 static void                    nlm_gc_hosts(void);
-
-/*
- * Find an NLM server handle in the cache. If there is none, create it.
- */
-struct nlm_host *
-nlmclnt_lookup_host(struct sockaddr_in *sin, int proto, int version)
-{
-       return nlm_lookup_host(0, sin, proto, version);
-}
-
-/*
- * Find an NLM client handle in the cache. If there is none, create it.
- */
-struct nlm_host *
-nlmsvc_lookup_host(struct svc_rqst *rqstp)
-{
-       return nlm_lookup_host(1, &rqstp->rq_addr,
-                              rqstp->rq_prot, rqstp->rq_vers);
-}
+static struct nsm_handle *     __nsm_find(const struct sockaddr_in *,
+                                       const char *, int, int);
+static struct nsm_handle *     nsm_find(const struct sockaddr_in *sin,
+                                        const char *hostname,
+                                        int hostname_len);
 
 /*
  * Common host lookup routine for server & client
  */
-struct nlm_host *
-nlm_lookup_host(int server, struct sockaddr_in *sin,
-                                       int proto, int version)
+static struct nlm_host *
+nlm_lookup_host(int server, const struct sockaddr_in *sin,
+                                       int proto, int version,
+                                       const char *hostname,
+                                       int hostname_len)
 {
-       struct nlm_host *host, **hp;
-       u32             addr;
+       struct hlist_head *chain;
+       struct hlist_node *pos;
+       struct nlm_host *host;
+       struct nsm_handle *nsm = NULL;
        int             hash;
 
-       dprintk("lockd: nlm_lookup_host(%08x, p=%d, v=%d)\n",
-                       (unsigned)(sin? ntohl(sin->sin_addr.s_addr) : 0), proto, version);
+       dprintk("lockd: nlm_lookup_host(%u.%u.%u.%u, p=%d, v=%d, my role=%s, name=%.*s)\n",
+                       NIPQUAD(sin->sin_addr.s_addr), proto, version,
+                       server? "server" : "client",
+                       hostname_len,
+                       hostname? hostname : "<none>");
+
 
        hash = NLM_ADDRHASH(sin->sin_addr.s_addr);
 
        /* Lock hash table */
-       down(&nlm_host_sema);
+       mutex_lock(&nlm_host_mutex);
 
        if (time_after_eq(jiffies, next_gc))
                nlm_gc_hosts();
 
-       for (hp = &nlm_hosts[hash]; (host = *hp) != 0; hp = &host->h_next) {
+       /* We may keep several nlm_host objects for a peer, because each
+        * nlm_host is identified by
+        * (address, protocol, version, server/client)
+        * We could probably simplify this a little by putting all those
+        * different NLM rpc_clients into one single nlm_host object.
+        * This would allow us to have one nlm_host per address.
+        */
+       chain = &nlm_hosts[hash];
+       hlist_for_each_entry(host, pos, chain, h_hash) {
+               if (!nlm_cmp_addr(&host->h_addr, sin))
+                       continue;
+
+               /* See if we have an NSM handle for this client */
+               if (!nsm)
+                       nsm = host->h_nsmhandle;
+
                if (host->h_proto != proto)
                        continue;
                if (host->h_version != version)
@@ -84,78 +93,110 @@ nlm_lookup_host(int server, struct sockaddr_in *sin,
                if (host->h_server != server)
                        continue;
 
-               if (nlm_cmp_addr(&host->h_addr, sin)) {
-                       if (hp != nlm_hosts + hash) {
-                               *hp = host->h_next;
-                               host->h_next = nlm_hosts[hash];
-                               nlm_hosts[hash] = host;
-                       }
-                       nlm_get_host(host);
-                       up(&nlm_host_sema);
-                       return host;
-               }
-       }
+               /* Move to head of hash chain. */
+               hlist_del(&host->h_hash);
+               hlist_add_head(&host->h_hash, chain);
 
-       /* Ooops, no host found, create it */
-       dprintk("lockd: creating host entry\n");
+               nlm_get_host(host);
+               goto out;
+       }
+       if (nsm)
+               atomic_inc(&nsm->sm_count);
 
-       if (!(host = (struct nlm_host *) kmalloc(sizeof(*host), GFP_KERNEL)))
-               goto nohost;
-       memset(host, 0, sizeof(*host));
+       host = NULL;
 
-       addr = sin->sin_addr.s_addr;
-       sprintf(host->h_name, "%u.%u.%u.%u", NIPQUAD(addr));
+       /* Sadly, the host isn't in our hash table yet. See if
+        * we have an NSM handle for it. If not, create one.
+        */
+       if (!nsm && !(nsm = nsm_find(sin, hostname, hostname_len)))
+               goto out;
 
+       host = kzalloc(sizeof(*host), GFP_KERNEL);
+       if (!host) {
+               nsm_release(nsm);
+               goto out;
+       }
+       host->h_name       = nsm->sm_name;
        host->h_addr       = *sin;
        host->h_addr.sin_port = 0;      /* ouch! */
        host->h_version    = version;
        host->h_proto      = proto;
        host->h_rpcclnt    = NULL;
-       init_MUTEX(&host->h_sema);
+       mutex_init(&host->h_mutex);
        host->h_nextrebind = jiffies + NLM_HOST_REBIND;
        host->h_expires    = jiffies + NLM_HOST_EXPIRE;
        atomic_set(&host->h_count, 1);
        init_waitqueue_head(&host->h_gracewait);
+       init_rwsem(&host->h_rwsem);
        host->h_state      = 0;                 /* pseudo NSM state */
        host->h_nsmstate   = 0;                 /* real NSM state */
+       host->h_nsmhandle  = nsm;
        host->h_server     = server;
-       host->h_next       = nlm_hosts[hash];
-       nlm_hosts[hash]    = host;
+       hlist_add_head(&host->h_hash, chain);
        INIT_LIST_HEAD(&host->h_lockowners);
        spin_lock_init(&host->h_lock);
+       INIT_LIST_HEAD(&host->h_granted);
+       INIT_LIST_HEAD(&host->h_reclaim);
 
        if (++nrhosts > NLM_HOST_MAX)
                next_gc = 0;
 
-nohost:
-       up(&nlm_host_sema);
+out:
+       mutex_unlock(&nlm_host_mutex);
        return host;
 }
 
-struct nlm_host *
-nlm_find_client(void)
+/*
+ * Destroy a host
+ */
+static void
+nlm_destroy_host(struct nlm_host *host)
 {
-       /* find a nlm_host for a client for which h_killed == 0.
-        * and return it
+       struct rpc_clnt *clnt;
+
+       BUG_ON(!list_empty(&host->h_lockowners));
+       BUG_ON(atomic_read(&host->h_count));
+
+       /*
+        * Release NSM handle and unmonitor host.
         */
-       int hash;
-       down(&nlm_host_sema);
-       for (hash = 0 ; hash < NLM_HOST_NRHASH; hash++) {
-               struct nlm_host *host, **hp;
-               for (hp = &nlm_hosts[hash]; (host = *hp) != 0; hp = &host->h_next) {
-                       if (host->h_server &&
-                           host->h_killed == 0) {
-                               nlm_get_host(host);
-                               up(&nlm_host_sema);
-                               return host;
-                       }
+       nsm_unmonitor(host);
+
+       if ((clnt = host->h_rpcclnt) != NULL) {
+               if (atomic_read(&clnt->cl_users)) {
+                       printk(KERN_WARNING
+                               "lockd: active RPC handle\n");
+                       clnt->cl_dead = 1;
+               } else {
+                       rpc_destroy_client(host->h_rpcclnt);
                }
        }
-       up(&nlm_host_sema);
-       return NULL;
+       kfree(host);
+}
+
+/*
+ * Find an NLM server handle in the cache. If there is none, create it.
+ */
+struct nlm_host *
+nlmclnt_lookup_host(const struct sockaddr_in *sin, int proto, int version,
+                       const char *hostname, int hostname_len)
+{
+       return nlm_lookup_host(0, sin, proto, version,
+                              hostname, hostname_len);
+}
+
+/*
+ * Find an NLM client handle in the cache. If there is none, create it.
+ */
+struct nlm_host *
+nlmsvc_lookup_host(struct svc_rqst *rqstp,
+                       const char *hostname, int hostname_len)
+{
+       return nlm_lookup_host(1, &rqstp->rq_addr,
+                              rqstp->rq_prot, rqstp->rq_vers,
+                              hostname, hostname_len);
 }
 
-                               
 /*
  * Create the NLM RPC client for an NLM peer
  */
@@ -163,19 +204,17 @@ struct rpc_clnt *
 nlm_bind_host(struct nlm_host *host)
 {
        struct rpc_clnt *clnt;
-       struct rpc_xprt *xprt;
 
        dprintk("lockd: nlm_bind_host(%08x)\n",
                        (unsigned)ntohl(host->h_addr.sin_addr.s_addr));
 
        /* Lock host handle */
-       down(&host->h_sema);
+       mutex_lock(&host->h_mutex);
 
        /* If we've already created an RPC client, check whether
         * RPC rebind is required
         */
        if ((clnt = host->h_rpcclnt) != NULL) {
-               xprt = clnt->cl_xprt;
                if (time_after_eq(jiffies, host->h_nextrebind)) {
                        rpc_force_rebind(clnt);
                        host->h_nextrebind = jiffies + NLM_HOST_REBIND;
@@ -183,30 +222,37 @@ nlm_bind_host(struct nlm_host *host)
                                        host->h_nextrebind - jiffies);
                }
        } else {
-               xprt = xprt_create_proto(host->h_proto, &host->h_addr, NULL);
-               if (IS_ERR(xprt))
-                       goto forgetit;
-
-               xprt_set_timeout(&xprt->timeout, 5, nlmsvc_timeout);
-               xprt->resvport = 1;     /* NLM requires a reserved port */
-
-               /* Existing NLM servers accept AUTH_UNIX only */
-               clnt = rpc_create_client(xprt, host->h_name, &nlm_program,
-                                       host->h_version, RPC_AUTH_UNIX);
-               if (IS_ERR(clnt))
-                       goto forgetit;
-               clnt->cl_autobind = 1;  /* turn on pmap queries */
-
-               host->h_rpcclnt = clnt;
+               unsigned long increment = nlmsvc_timeout * HZ;
+               struct rpc_timeout timeparms = {
+                       .to_initval     = increment,
+                       .to_increment   = increment,
+                       .to_maxval      = increment * 6UL,
+                       .to_retries     = 5U,
+               };
+               struct rpc_create_args args = {
+                       .protocol       = host->h_proto,
+                       .address        = (struct sockaddr *)&host->h_addr,
+                       .addrsize       = sizeof(host->h_addr),
+                       .timeout        = &timeparms,
+                       .servername     = host->h_name,
+                       .program        = &nlm_program,
+                       .version        = host->h_version,
+                       .authflavor     = RPC_AUTH_UNIX,
+                       .flags          = (RPC_CLNT_CREATE_HARDRTRY |
+                                          RPC_CLNT_CREATE_AUTOBIND),
+               };
+
+               clnt = rpc_create(&args);
+               if (!IS_ERR(clnt))
+                       host->h_rpcclnt = clnt;
+               else {
+                       printk("lockd: couldn't create RPC handle for %s\n", host->h_name);
+                       clnt = NULL;
+               }
        }
 
-       up(&host->h_sema);
+       mutex_unlock(&host->h_mutex);
        return clnt;
-
-forgetit:
-       printk("lockd: couldn't create RPC handle for %s\n", host->h_name);
-       up(&host->h_sema);
-       return NULL;
 }
 
 /*
@@ -242,11 +288,74 @@ void nlm_release_host(struct nlm_host *host)
 {
        if (host != NULL) {
                dprintk("lockd: release host %s\n", host->h_name);
-               atomic_dec(&host->h_count);
                BUG_ON(atomic_read(&host->h_count) < 0);
+               if (atomic_dec_and_test(&host->h_count)) {
+                       BUG_ON(!list_empty(&host->h_lockowners));
+                       BUG_ON(!list_empty(&host->h_granted));
+                       BUG_ON(!list_empty(&host->h_reclaim));
+               }
        }
 }
 
+/*
+ * We were notified that the host indicated by address &sin
+ * has rebooted.
+ * Release all resources held by that peer.
+ */
+void nlm_host_rebooted(const struct sockaddr_in *sin,
+                               const char *hostname, int hostname_len,
+                               u32 new_state)
+{
+       struct hlist_head *chain;
+       struct hlist_node *pos;
+       struct nsm_handle *nsm;
+       struct nlm_host *host;
+
+       dprintk("lockd: nlm_host_rebooted(%s, %u.%u.%u.%u)\n",
+                       hostname, NIPQUAD(sin->sin_addr));
+
+       /* Find the NSM handle for this peer */
+       if (!(nsm = __nsm_find(sin, hostname, hostname_len, 0)))
+               return;
+
+       /* When reclaiming locks on this peer, make sure that
+        * we set up a new notification */
+       nsm->sm_monitored = 0;
+
+       /* Mark all hosts tied to this NSM state as having rebooted.
+        * We run the loop repeatedly, because we drop the host table
+        * lock for this.
+        * To avoid processing a host several times, we match the nsmstate.
+        */
+again: mutex_lock(&nlm_host_mutex);
+       for (chain = nlm_hosts; chain < nlm_hosts + NLM_HOST_NRHASH; ++chain) {
+               hlist_for_each_entry(host, pos, chain, h_hash) {
+                       if (host->h_nsmhandle == nsm
+                        && host->h_nsmstate != new_state) {
+                               host->h_nsmstate = new_state;
+                               host->h_state++;
+
+                               nlm_get_host(host);
+                               mutex_unlock(&nlm_host_mutex);
+
+                               if (host->h_server) {
+                                       /* We're server for this guy, just ditch
+                                        * all the locks he held. */
+                                       nlmsvc_free_host_resources(host);
+                               } else {
+                                       /* He's the server, initiate lock recovery. */
+                                       nlmclnt_recovery(host);
+                               }
+
+                               nlm_release_host(host);
+                               goto again;
+                       }
+               }
+       }
+
+       mutex_unlock(&nlm_host_mutex);
+}
+
 /*
  * Shut down the hosts module.
  * Note that this routine is called only at server shutdown time.
@@ -254,29 +363,30 @@ void nlm_release_host(struct nlm_host *host)
 void
 nlm_shutdown_hosts(void)
 {
+       struct hlist_head *chain;
+       struct hlist_node *pos;
        struct nlm_host *host;
-       int             i;
 
        dprintk("lockd: shutting down host module\n");
-       down(&nlm_host_sema);
+       mutex_lock(&nlm_host_mutex);
 
        /* First, make all hosts eligible for gc */
        dprintk("lockd: nuking all hosts...\n");
-       for (i = 0; i < NLM_HOST_NRHASH; i++) {
-               for (host = nlm_hosts[i]; host; host = host->h_next)
+       for (chain = nlm_hosts; chain < nlm_hosts + NLM_HOST_NRHASH; ++chain) {
+               hlist_for_each_entry(host, pos, chain, h_hash)
                        host->h_expires = jiffies - 1;
        }
 
        /* Then, perform a garbage collection pass */
        nlm_gc_hosts();
-       up(&nlm_host_sema);
+       mutex_unlock(&nlm_host_mutex);
 
        /* complain if any hosts are left */
        if (nrhosts) {
                printk(KERN_WARNING "lockd: couldn't shutdown host module!\n");
                dprintk("lockd: %d hosts left:\n", nrhosts);
-               for (i = 0; i < NLM_HOST_NRHASH; i++) {
-                       for (host = nlm_hosts[i]; host; host = host->h_next) {
+               for (chain = nlm_hosts; chain < nlm_hosts + NLM_HOST_NRHASH; ++chain) {
+                       hlist_for_each_entry(host, pos, chain, h_hash) {
                                dprintk("       %s (cnt %d use %d exp %ld)\n",
                                        host->h_name, atomic_read(&host->h_count),
                                        host->h_inuse, host->h_expires);
@@ -293,46 +403,32 @@ nlm_shutdown_hosts(void)
 static void
 nlm_gc_hosts(void)
 {
-       struct nlm_host **q, *host;
-       struct rpc_clnt *clnt;
-       int             i;
+       struct hlist_head *chain;
+       struct hlist_node *pos, *next;
+       struct nlm_host *host;
 
        dprintk("lockd: host garbage collection\n");
-       for (i = 0; i < NLM_HOST_NRHASH; i++) {
-               for (host = nlm_hosts[i]; host; host = host->h_next)
+       for (chain = nlm_hosts; chain < nlm_hosts + NLM_HOST_NRHASH; ++chain) {
+               hlist_for_each_entry(host, pos, chain, h_hash)
                        host->h_inuse = 0;
        }
 
        /* Mark all hosts that hold locks, blocks or shares */
        nlmsvc_mark_resources();
 
-       for (i = 0; i < NLM_HOST_NRHASH; i++) {
-               q = &nlm_hosts[i];
-               while ((host = *q) != NULL) {
+       for (chain = nlm_hosts; chain < nlm_hosts + NLM_HOST_NRHASH; ++chain) {
+               hlist_for_each_entry_safe(host, pos, next, chain, h_hash) {
                        if (atomic_read(&host->h_count) || host->h_inuse
                         || time_before(jiffies, host->h_expires)) {
                                dprintk("nlm_gc_hosts skipping %s (cnt %d use %d exp %ld)\n",
                                        host->h_name, atomic_read(&host->h_count),
                                        host->h_inuse, host->h_expires);
-                               q = &host->h_next;
                                continue;
                        }
                        dprintk("lockd: delete host %s\n", host->h_name);
-                       *q = host->h_next;
-                       /* Don't unmonitor hosts that have been invalidated */
-                       if (host->h_monitored && !host->h_killed)
-                               nsm_unmonitor(host);
-                       if ((clnt = host->h_rpcclnt) != NULL) {
-                               if (atomic_read(&clnt->cl_users)) {
-                                       printk(KERN_WARNING
-                                               "lockd: active RPC handle\n");
-                                       clnt->cl_dead = 1;
-                               } else {
-                                       rpc_destroy_client(host->h_rpcclnt);
-                               }
-                       }
-                       BUG_ON(!list_empty(&host->h_lockowners));
-                       kfree(host);
+                       hlist_del_init(&host->h_hash);
+
+                       nlm_destroy_host(host);
                        nrhosts--;
                }
        }
@@ -340,3 +436,88 @@ nlm_gc_hosts(void)
        next_gc = jiffies + NLM_HOST_COLLECT;
 }
 
+
+/*
+ * Manage NSM handles
+ */
+static LIST_HEAD(nsm_handles);
+static DEFINE_MUTEX(nsm_mutex);
+
+static struct nsm_handle *
+__nsm_find(const struct sockaddr_in *sin,
+               const char *hostname, int hostname_len,
+               int create)
+{
+       struct nsm_handle *nsm = NULL;
+       struct list_head *pos;
+
+       if (!sin)
+               return NULL;
+
+       if (hostname && memchr(hostname, '/', hostname_len) != NULL) {
+               if (printk_ratelimit()) {
+                       printk(KERN_WARNING "Invalid hostname \"%.*s\" "
+                                           "in NFS lock request\n",
+                               hostname_len, hostname);
+               }
+               return NULL;
+       }
+
+       mutex_lock(&nsm_mutex);
+       list_for_each(pos, &nsm_handles) {
+               nsm = list_entry(pos, struct nsm_handle, sm_link);
+
+               if (hostname && nsm_use_hostnames) {
+                       if (strlen(nsm->sm_name) != hostname_len
+                        || memcmp(nsm->sm_name, hostname, hostname_len))
+                               continue;
+               } else if (!nlm_cmp_addr(&nsm->sm_addr, sin))
+                       continue;
+               atomic_inc(&nsm->sm_count);
+               goto out;
+       }
+
+       if (!create) {
+               nsm = NULL;
+               goto out;
+       }
+
+       nsm = kzalloc(sizeof(*nsm) + hostname_len + 1, GFP_KERNEL);
+       if (nsm != NULL) {
+               nsm->sm_addr = *sin;
+               nsm->sm_name = (char *) (nsm + 1);
+               memcpy(nsm->sm_name, hostname, hostname_len);
+               nsm->sm_name[hostname_len] = '\0';
+               atomic_set(&nsm->sm_count, 1);
+
+               list_add(&nsm->sm_link, &nsm_handles);
+       }
+
+out:
+       mutex_unlock(&nsm_mutex);
+       return nsm;
+}
+
+static struct nsm_handle *
+nsm_find(const struct sockaddr_in *sin, const char *hostname, int hostname_len)
+{
+       return __nsm_find(sin, hostname, hostname_len, 1);
+}
+
+/*
+ * Release an NSM handle
+ */
+void
+nsm_release(struct nsm_handle *nsm)
+{
+       if (!nsm)
+               return;
+       if (atomic_dec_and_test(&nsm->sm_count)) {
+               mutex_lock(&nsm_mutex);
+               if (atomic_read(&nsm->sm_count) == 0) {
+                       list_del(&nsm->sm_link);
+                       kfree(nsm);
+               }
+               mutex_unlock(&nsm_mutex);
+       }
+}