SUNRPC: Change svc_create_xprt() to take a @family argument
[safe/jmp/linux-2.6] / fs / lockd / svc.c
index 2c3d5ac..390c559 100644 (file)
@@ -25,6 +25,8 @@
 #include <linux/smp.h>
 #include <linux/smp_lock.h>
 #include <linux/mutex.h>
+#include <linux/kthread.h>
+#include <linux/freezer.h>
 
 #include <linux/sunrpc/types.h>
 #include <linux/sunrpc/stats.h>
@@ -33,7 +35,6 @@
 #include <linux/sunrpc/svcsock.h>
 #include <net/ip.h>
 #include <linux/lockd/lockd.h>
-#include <linux/lockd/sm_inter.h>
 #include <linux/nfs.h>
 
 #define NLMDBG_FACILITY                NLMDBG_SVC
 static struct svc_program      nlmsvc_program;
 
 struct nlmsvc_binding *                nlmsvc_ops;
-EXPORT_SYMBOL(nlmsvc_ops);
+EXPORT_SYMBOL_GPL(nlmsvc_ops);
 
 static DEFINE_MUTEX(nlmsvc_mutex);
 static unsigned int            nlmsvc_users;
-static pid_t                   nlmsvc_pid;
-static struct svc_serv         *nlmsvc_serv;
-int                            nlmsvc_grace_period;
+static struct task_struct      *nlmsvc_task;
+static struct svc_rqst         *nlmsvc_rqst;
 unsigned long                  nlmsvc_timeout;
 
-static DECLARE_COMPLETION(lockd_start_done);
-static DECLARE_WAIT_QUEUE_HEAD(lockd_exit);
+/*
+ * If the kernel has IPv6 support available, always listen for
+ * both AF_INET and AF_INET6 requests.
+ */
+#if (defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)) && \
+       defined(CONFIG_SUNRPC_REGISTER_V4)
+static const sa_family_t       nlmsvc_family = AF_INET6;
+#else  /* (CONFIG_IPV6 || CONFIG_IPV6_MODULE) && CONFIG_SUNRPC_REGISTER_V4 */
+static const sa_family_t       nlmsvc_family = AF_INET;
+#endif /* (CONFIG_IPV6 || CONFIG_IPV6_MODULE) && CONFIG_SUNRPC_REGISTER_V4 */
 
 /*
  * These can be set at insmod time (useful for NFS as root filesystem),
@@ -62,7 +70,9 @@ static DECLARE_WAIT_QUEUE_HEAD(lockd_exit);
 static unsigned long           nlm_grace_period;
 static unsigned long           nlm_timeout = LOCKD_DFLT_TIMEO;
 static int                     nlm_udpport, nlm_tcpport;
-int                            nsm_use_hostnames = 0;
+
+/* RLIM_NOFILE defaults to 1024. That seems like a reasonable default here. */
+static unsigned int            nlm_max_connections = 1024;
 
 /*
  * Constants needed for the sysctl interface.
@@ -73,202 +83,188 @@ static const unsigned long        nlm_timeout_min = 3;
 static const unsigned long     nlm_timeout_max = 20;
 static const int               nlm_port_min = 0, nlm_port_max = 65535;
 
+#ifdef CONFIG_SYSCTL
 static struct ctl_table_header * nlm_sysctl_table;
+#endif
 
-static unsigned long set_grace_period(void)
+static unsigned long get_lockd_grace_period(void)
 {
-       unsigned long grace_period;
-
        /* Note: nlm_timeout should always be nonzero */
        if (nlm_grace_period)
-               grace_period = ((nlm_grace_period + nlm_timeout - 1)
-                               / nlm_timeout) * nlm_timeout * HZ;
+               return roundup(nlm_grace_period, nlm_timeout) * HZ;
        else
-               grace_period = nlm_timeout * 5 * HZ;
-       nlmsvc_grace_period = 1;
-       return grace_period + jiffies;
+               return nlm_timeout * 5 * HZ;
+}
+
+static struct lock_manager lockd_manager = {
+};
+
+static void grace_ender(struct work_struct *not_used)
+{
+       locks_end_grace(&lockd_manager);
 }
 
-static inline void clear_grace_period(void)
+static DECLARE_DELAYED_WORK(grace_period_end, grace_ender);
+
+static void set_grace_period(void)
 {
-       nlmsvc_grace_period = 0;
+       unsigned long grace_period = get_lockd_grace_period();
+
+       locks_start_grace(&lockd_manager);
+       cancel_delayed_work_sync(&grace_period_end);
+       schedule_delayed_work(&grace_period_end, grace_period);
 }
 
 /*
  * This is the lockd kernel thread
  */
-static void
-lockd(struct svc_rqst *rqstp)
+static int
+lockd(void *vrqstp)
 {
-       int             err = 0;
-       unsigned long grace_period_expire;
-
-       /* Lock module and set up kernel thread */
-       /* lockd_up is waiting for us to startup, so will
-        * be holding a reference to this module, so it
-        * is safe to just claim another reference
-        */
-       __module_get(THIS_MODULE);
-       lock_kernel();
-
-       /*
-        * Let our maker know we're running.
-        */
-       nlmsvc_pid = current->pid;
-       nlmsvc_serv = rqstp->rq_server;
-       complete(&lockd_start_done);
+       int             err = 0, preverr = 0;
+       struct svc_rqst *rqstp = vrqstp;
 
-       daemonize("lockd");
+       /* try_to_freeze() is called from svc_recv() */
+       set_freezable();
 
-       /* Process request with signals blocked, but allow SIGKILL.  */
+       /* Allow SIGKILL to tell lockd to drop all of its locks */
        allow_signal(SIGKILL);
 
-       /* kick rpciod */
-       rpciod_up();
-
        dprintk("NFS locking service started (ver " LOCKD_VERSION ").\n");
 
+       /*
+        * FIXME: it would be nice if lockd didn't spend its entire life
+        * running under the BKL. At the very least, it would be good to
+        * have someone clarify what it's intended to protect here. I've
+        * seen some handwavy posts about posix locking needing to be
+        * done under the BKL, but it's far from clear.
+        */
+       lock_kernel();
+
        if (!nlm_timeout)
                nlm_timeout = LOCKD_DFLT_TIMEO;
        nlmsvc_timeout = nlm_timeout * HZ;
 
-       grace_period_expire = set_grace_period();
+       set_grace_period();
 
        /*
         * The main request loop. We don't terminate until the last
-        * NFS mount or NFS daemon has gone away, and we've been sent a
-        * signal, or else another process has taken over our job.
+        * NFS mount or NFS daemon has gone away.
         */
-       while ((nlmsvc_users || !signalled()) && nlmsvc_pid == current->pid) {
+       while (!kthread_should_stop()) {
                long timeout = MAX_SCHEDULE_TIMEOUT;
+               RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
+
+               /* update sv_maxconn if it has changed */
+               rqstp->rq_server->sv_maxconn = nlm_max_connections;
 
                if (signalled()) {
                        flush_signals(current);
                        if (nlmsvc_ops) {
                                nlmsvc_invalidate_all();
-                               grace_period_expire = set_grace_period();
+                               set_grace_period();
                        }
+                       continue;
                }
 
-               /*
-                * Retry any blocked locks that have been notified by
-                * the VFS. Don't do this during grace period.
-                * (Theoretically, there shouldn't even be blocked locks
-                * during grace period).
-                */
-               if (!nlmsvc_grace_period) {
-                       timeout = nlmsvc_retry_blocked();
-               } else if (time_before(grace_period_expire, jiffies))
-                       clear_grace_period();
+               timeout = nlmsvc_retry_blocked();
 
                /*
                 * Find a socket with data available and call its
                 * recvfrom routine.
                 */
                err = svc_recv(rqstp, timeout);
-               if (err == -EAGAIN || err == -EINTR)
+               if (err == -EAGAIN || err == -EINTR) {
+                       preverr = err;
                        continue;
+               }
                if (err < 0) {
-                       printk(KERN_WARNING
-                              "lockd: terminating on error %d\n",
-                              -err);
-                       break;
+                       if (err != preverr) {
+                               printk(KERN_WARNING "%s: unexpected error "
+                                       "from svc_recv (%d)\n", __func__, err);
+                               preverr = err;
+                       }
+                       schedule_timeout_interruptible(HZ);
+                       continue;
                }
+               preverr = err;
 
-               dprintk("lockd: request from %08x\n",
-                       (unsigned)ntohl(rqstp->rq_addr.sin_addr.s_addr));
+               dprintk("lockd: request from %s\n",
+                               svc_print_addr(rqstp, buf, sizeof(buf)));
 
                svc_process(rqstp);
-
        }
-
        flush_signals(current);
-
-       /*
-        * Check whether there's a new lockd process before
-        * shutting down the hosts and clearing the slot.
-        */
-       if (!nlmsvc_pid || current->pid == nlmsvc_pid) {
-               if (nlmsvc_ops)
-                       nlmsvc_invalidate_all();
-               nlm_shutdown_hosts();
-               nlmsvc_pid = 0;
-               nlmsvc_serv = NULL;
-       } else
-               printk(KERN_DEBUG
-                       "lockd: new process, skipping host shutdown\n");
-       wake_up(&lockd_exit);
-
-       /* Exit the RPC thread */
-       svc_exit_thread(rqstp);
-
-       /* release rpciod */
-       rpciod_down();
-
-       /* Release module */
+       cancel_delayed_work_sync(&grace_period_end);
+       locks_end_grace(&lockd_manager);
+       if (nlmsvc_ops)
+               nlmsvc_invalidate_all();
+       nlm_shutdown_hosts();
        unlock_kernel();
-       module_put_and_exit(0);
+       return 0;
 }
 
-
-static int find_socket(struct svc_serv *serv, int proto)
+static int create_lockd_listener(struct svc_serv *serv, char *name,
+                                unsigned short port)
 {
-       struct svc_sock *svsk;
-       int found = 0;
-       list_for_each_entry(svsk, &serv->sv_permsocks, sk_list)
-               if (svsk->sk_sk->sk_protocol == proto) {
-                       found = 1;
-                       break;
-               }
-       return found;
+       struct svc_xprt *xprt;
+
+       xprt = svc_find_xprt(serv, name, 0, 0);
+       if (xprt == NULL)
+               return svc_create_xprt(serv, name, nlmsvc_family,
+                                       port, SVC_SOCK_DEFAULTS);
+
+       svc_xprt_put(xprt);
+       return 0;
 }
 
 /*
- * Make any sockets that are needed but not present.
- * If nlm_udpport or nlm_tcpport were set as module
- * options, make those sockets unconditionally
+ * Ensure there are active UDP and TCP listeners for lockd.
+ *
+ * Even if we have only TCP NFS mounts and/or TCP NFSDs, some
+ * local services (such as rpc.statd) still require UDP, and
+ * some NFS servers do not yet support NLM over TCP.
+ *
+ * Returns zero if all listeners are available; otherwise a
+ * negative errno value is returned.
  */
-static int make_socks(struct svc_serv *serv, int proto)
+static int make_socks(struct svc_serv *serv)
 {
        static int warned;
-       int err = 0;
-
-       if (proto == IPPROTO_UDP || nlm_udpport)
-               if (!find_socket(serv, IPPROTO_UDP))
-                       err = svc_makesock(serv, IPPROTO_UDP, nlm_udpport,
-                                               SVC_SOCK_DEFAULTS);
-       if (err >= 0 && (proto == IPPROTO_TCP || nlm_tcpport))
-               if (!find_socket(serv, IPPROTO_TCP))
-                       err = svc_makesock(serv, IPPROTO_TCP, nlm_tcpport,
-                                               SVC_SOCK_DEFAULTS);
-
-       if (err >= 0) {
-               warned = 0;
-               err = 0;
-       } else if (warned++ == 0)
+       int err;
+
+       err = create_lockd_listener(serv, "udp", nlm_udpport);
+       if (err < 0)
+               goto out_err;
+
+       err = create_lockd_listener(serv, "tcp", nlm_tcpport);
+       if (err < 0)
+               goto out_err;
+
+       warned = 0;
+       return 0;
+
+out_err:
+       if (warned++ == 0)
                printk(KERN_WARNING
-                      "lockd_up: makesock failed, error=%d\n", err);
+                       "lockd_up: makesock failed, error=%d\n", err);
        return err;
 }
 
 /*
  * Bring up the lockd process if it's not already up.
  */
-int
-lockd_up(int proto) /* Maybe add a 'family' option when IPv6 is supported ?? */
+int lockd_up(void)
 {
-       struct svc_serv *       serv;
-       int                     error = 0;
+       struct svc_serv *serv;
+       int             error = 0;
 
        mutex_lock(&nlmsvc_mutex);
        /*
         * Check whether we're already up and running.
         */
-       if (nlmsvc_pid) {
-               if (proto)
-                       error = make_socks(nlmsvc_serv, proto);
+       if (nlmsvc_rqst)
                goto out;
-       }
 
        /*
         * Sanity check: if there's no pid,
@@ -279,25 +275,42 @@ lockd_up(int proto) /* Maybe add a 'family' option when IPv6 is supported ?? */
                        "lockd_up: no pid, %d users??\n", nlmsvc_users);
 
        error = -ENOMEM;
-       serv = svc_create(&nlmsvc_program, LOCKD_BUFSIZE, NULL);
+       serv = svc_create(&nlmsvc_program, LOCKD_BUFSIZE, nlmsvc_family, NULL);
        if (!serv) {
                printk(KERN_WARNING "lockd_up: create service failed\n");
                goto out;
        }
 
-       if ((error = make_socks(serv, proto)) < 0)
+       error = make_socks(serv);
+       if (error < 0)
                goto destroy_and_out;
 
        /*
         * Create the kernel thread and wait for it to start.
         */
-       error = svc_create_thread(lockd, serv);
-       if (error) {
+       nlmsvc_rqst = svc_prepare_thread(serv, &serv->sv_pools[0]);
+       if (IS_ERR(nlmsvc_rqst)) {
+               error = PTR_ERR(nlmsvc_rqst);
+               nlmsvc_rqst = NULL;
                printk(KERN_WARNING
-                       "lockd_up: create thread failed, error=%d\n", error);
+                       "lockd_up: svc_rqst allocation failed, error=%d\n",
+                       error);
+               goto destroy_and_out;
+       }
+
+       svc_sock_update_bufs(serv);
+       serv->sv_maxconn = nlm_max_connections;
+
+       nlmsvc_task = kthread_run(lockd, nlmsvc_rqst, serv->sv_name);
+       if (IS_ERR(nlmsvc_task)) {
+               error = PTR_ERR(nlmsvc_task);
+               svc_exit_thread(nlmsvc_rqst);
+               nlmsvc_task = NULL;
+               nlmsvc_rqst = NULL;
+               printk(KERN_WARNING
+                       "lockd_up: kthread_run failed, error=%d\n", error);
                goto destroy_and_out;
        }
-       wait_for_completion(&lockd_start_done);
 
        /*
         * Note: svc_serv structures have an initial use count of 1,
@@ -311,7 +324,7 @@ out:
        mutex_unlock(&nlmsvc_mutex);
        return error;
 }
-EXPORT_SYMBOL(lockd_up);
+EXPORT_SYMBOL_GPL(lockd_up);
 
 /*
  * Decrement the user count and bring down lockd if we're the last.
@@ -319,41 +332,30 @@ EXPORT_SYMBOL(lockd_up);
 void
 lockd_down(void)
 {
-       static int warned;
-
        mutex_lock(&nlmsvc_mutex);
        if (nlmsvc_users) {
                if (--nlmsvc_users)
                        goto out;
-       } else
-               printk(KERN_WARNING "lockd_down: no users! pid=%d\n", nlmsvc_pid);
-
-       if (!nlmsvc_pid) {
-               if (warned++ == 0)
-                       printk(KERN_WARNING "lockd_down: no lockd running.\n"); 
-               goto out;
+       } else {
+               printk(KERN_ERR "lockd_down: no users! task=%p\n",
+                       nlmsvc_task);
+               BUG();
        }
-       warned = 0;
 
-       kill_proc(nlmsvc_pid, SIGKILL, 1);
-       /*
-        * Wait for the lockd process to exit, but since we're holding
-        * the lockd semaphore, we can't wait around forever ...
-        */
-       clear_thread_flag(TIF_SIGPENDING);
-       interruptible_sleep_on_timeout(&lockd_exit, HZ);
-       if (nlmsvc_pid) {
-               printk(KERN_WARNING 
-                       "lockd_down: lockd failed to exit, clearing pid\n");
-               nlmsvc_pid = 0;
+       if (!nlmsvc_task) {
+               printk(KERN_ERR "lockd_down: no lockd running.\n");
+               BUG();
        }
-       spin_lock_irq(&current->sighand->siglock);
-       recalc_sigpending();
-       spin_unlock_irq(&current->sighand->siglock);
+       kthread_stop(nlmsvc_task);
+       svc_exit_thread(nlmsvc_rqst);
+       nlmsvc_task = NULL;
+       nlmsvc_rqst = NULL;
 out:
        mutex_unlock(&nlmsvc_mutex);
 }
-EXPORT_SYMBOL(lockd_down);
+EXPORT_SYMBOL_GPL(lockd_down);
+
+#ifdef CONFIG_SYSCTL
 
 /*
  * Sysctl parameters (same as module parameters, different interface).
@@ -439,8 +441,10 @@ static ctl_table nlm_sysctl_root[] = {
        { .ctl_name = 0 }
 };
 
+#endif /* CONFIG_SYSCTL */
+
 /*
- * Module (and driverfs) parameters.
+ * Module (and sysfs) parameters.
  */
 
 #define param_set_min_max(name, type, which_strtol, min, max)          \
@@ -505,6 +509,7 @@ module_param_call(nlm_udpport, param_set_port, param_get_int,
 module_param_call(nlm_tcpport, param_set_port, param_get_int,
                  &nlm_tcpport, 0644);
 module_param(nsm_use_hostnames, bool, 0644);
+module_param(nlm_max_connections, uint, 0644);
 
 /*
  * Initialising and terminating the module.
@@ -512,15 +517,21 @@ module_param(nsm_use_hostnames, bool, 0644);
 
 static int __init init_nlm(void)
 {
-       nlm_sysctl_table = register_sysctl_table(nlm_sysctl_root, 0);
+#ifdef CONFIG_SYSCTL
+       nlm_sysctl_table = register_sysctl_table(nlm_sysctl_root);
        return nlm_sysctl_table ? 0 : -ENOMEM;
+#else
+       return 0;
+#endif
 }
 
 static void __exit exit_nlm(void)
 {
        /* FIXME: delete all NLM clients */
        nlm_shutdown_hosts();
+#ifdef CONFIG_SYSCTL
        unregister_sysctl_table(nlm_sysctl_table);
+#endif
 }
 
 module_init(init_nlm);