netfilter: ip6table_raw: fix table priority
[safe/jmp/linux-2.6] / net / sunrpc / sched.c
index f6eed4d..aae6907 100644 (file)
 #include <linux/slab.h>
 #include <linux/mempool.h>
 #include <linux/smp.h>
-#include <linux/smp_lock.h>
 #include <linux/spinlock.h>
 #include <linux/mutex.h>
 
 #include <linux/sunrpc/clnt.h>
 
+#include "sunrpc.h"
+
 #ifdef RPC_DEBUG
 #define RPCDBG_FACILITY                RPCDBG_SCHED
 #define RPC_TASK_MAGIC_ID      0xf00baa
@@ -38,74 +39,49 @@ static struct kmem_cache    *rpc_buffer_slabp __read_mostly;
 static mempool_t       *rpc_task_mempool __read_mostly;
 static mempool_t       *rpc_buffer_mempool __read_mostly;
 
-static void                    __rpc_default_timer(struct rpc_task *task);
-static void                    rpciod_killall(void);
 static void                    rpc_async_schedule(struct work_struct *);
 static void                     rpc_release_task(struct rpc_task *task);
+static void __rpc_queue_timer_fn(unsigned long ptr);
 
 /*
  * RPC tasks sit here while waiting for conditions to improve.
  */
-static RPC_WAITQ(delay_queue, "delayq");
-
-/*
- * All RPC clients are linked into this list
- */
-static LIST_HEAD(all_clients);
-static DECLARE_WAIT_QUEUE_HEAD(client_kill_wait);
+static struct rpc_wait_queue delay_queue;
 
 /*
  * rpciod-related stuff
  */
-static DEFINE_MUTEX(rpciod_mutex);
-static atomic_t rpciod_users = ATOMIC_INIT(0);
 struct workqueue_struct *rpciod_workqueue;
 
 /*
- * Spinlock for other critical sections of code.
- */
-static DEFINE_SPINLOCK(rpc_sched_lock);
-
-/*
  * Disable the timer for a given RPC task. Should be called with
  * queue->lock and bh_disabled in order to avoid races within
  * rpc_run_timer().
  */
-static inline void
-__rpc_disable_timer(struct rpc_task *task)
+static void
+__rpc_disable_timer(struct rpc_wait_queue *queue, struct rpc_task *task)
 {
+       if (task->tk_timeout == 0)
+               return;
        dprintk("RPC: %5u disabling timer\n", task->tk_pid);
-       task->tk_timeout_fn = NULL;
        task->tk_timeout = 0;
+       list_del(&task->u.tk_wait.timer_list);
+       if (list_empty(&queue->timer_list.list))
+               del_timer(&queue->timer_list.timer);
 }
 
-/*
- * Run a timeout function.
- * We use the callback in order to allow __rpc_wake_up_task()
- * and friends to disable the timer synchronously on SMP systems
- * without calling del_timer_sync(). The latter could cause a
- * deadlock if called while we're holding spinlocks...
- */
-static void rpc_run_timer(struct rpc_task *task)
+static void
+rpc_set_queue_timer(struct rpc_wait_queue *queue, unsigned long expires)
 {
-       void (*callback)(struct rpc_task *);
-
-       callback = task->tk_timeout_fn;
-       task->tk_timeout_fn = NULL;
-       if (callback && RPC_IS_QUEUED(task)) {
-               dprintk("RPC: %5u running timer\n", task->tk_pid);
-               callback(task);
-       }
-       smp_mb__before_clear_bit();
-       clear_bit(RPC_TASK_HAS_TIMER, &task->tk_runstate);
-       smp_mb__after_clear_bit();
+       queue->timer_list.expires = expires;
+       mod_timer(&queue->timer_list.timer, expires);
 }
 
 /*
  * Set up a timer for the current task.
  */
-static inline void
-__rpc_add_timer(struct rpc_task *task, rpc_action timer)
+static void
+__rpc_add_timer(struct rpc_wait_queue *queue, struct rpc_task *task)
 {
        if (!task->tk_timeout)
                return;
@@ -113,27 +89,10 @@ __rpc_add_timer(struct rpc_task *task, rpc_action timer)
        dprintk("RPC: %5u setting alarm for %lu ms\n",
                        task->tk_pid, task->tk_timeout * 1000 / HZ);
 
-       if (timer)
-               task->tk_timeout_fn = timer;
-       else
-               task->tk_timeout_fn = __rpc_default_timer;
-       set_bit(RPC_TASK_HAS_TIMER, &task->tk_runstate);
-       mod_timer(&task->tk_timer, jiffies + task->tk_timeout);
-}
-
-/*
- * Delete any timer for the current task. Because we use del_timer_sync(),
- * this function should never be called while holding queue->lock.
- */
-static void
-rpc_delete_timer(struct rpc_task *task)
-{
-       if (RPC_IS_QUEUED(task))
-               return;
-       if (test_and_clear_bit(RPC_TASK_HAS_TIMER, &task->tk_runstate)) {
-               del_singleshot_timer_sync(&task->tk_timer);
-               dprintk("RPC: %5u deleting timer\n", task->tk_pid);
-       }
+       task->u.tk_wait.expires = jiffies + task->tk_timeout;
+       if (list_empty(&queue->timer_list.list) || time_before(task->u.tk_wait.expires, queue->timer_list.expires))
+               rpc_set_queue_timer(queue, task->u.tk_wait.expires);
+       list_add(&task->u.tk_wait.timer_list, &queue->timer_list.list);
 }
 
 /*
@@ -149,7 +108,7 @@ static void __rpc_add_wait_queue_priority(struct rpc_wait_queue *queue, struct r
        if (unlikely(task->tk_priority > queue->maxpriority))
                q = &queue->tasks[queue->maxpriority];
        list_for_each_entry(t, q, u.tk_wait.list) {
-               if (t->tk_cookie == task->tk_cookie) {
+               if (t->tk_owner == task->tk_owner) {
                        list_add_tail(&task->u.tk_wait.list, &t->u.tk_wait.links);
                        return;
                }
@@ -175,7 +134,7 @@ static void __rpc_add_wait_queue(struct rpc_wait_queue *queue, struct rpc_task *
                list_add(&task->u.tk_wait.list, &queue->tasks[0]);
        else
                list_add_tail(&task->u.tk_wait.list, &queue->tasks[0]);
-       task->u.tk_wait.rpc_waitq = queue;
+       task->tk_waitqueue = queue;
        queue->qlen++;
        rpc_set_queued(task);
 
@@ -195,22 +154,18 @@ static void __rpc_remove_wait_queue_priority(struct rpc_task *task)
                list_move(&t->u.tk_wait.list, &task->u.tk_wait.list);
                list_splice_init(&task->u.tk_wait.links, &t->u.tk_wait.links);
        }
-       list_del(&task->u.tk_wait.list);
 }
 
 /*
  * Remove request from queue.
  * Note: must be called with spin lock held.
  */
-static void __rpc_remove_wait_queue(struct rpc_task *task)
+static void __rpc_remove_wait_queue(struct rpc_wait_queue *queue, struct rpc_task *task)
 {
-       struct rpc_wait_queue *queue;
-       queue = task->u.tk_wait.rpc_waitq;
-
+       __rpc_disable_timer(queue, task);
        if (RPC_IS_PRIORITY(queue))
                __rpc_remove_wait_queue_priority(task);
-       else
-               list_del(&task->u.tk_wait.list);
+       list_del(&task->u.tk_wait.list);
        queue->qlen--;
        dprintk("RPC: %5u removed from queue %p \"%s\"\n",
                        task->tk_pid, queue, rpc_qname(queue));
@@ -222,27 +177,30 @@ static inline void rpc_set_waitqueue_priority(struct rpc_wait_queue *queue, int
        queue->count = 1 << (priority * 2);
 }
 
-static inline void rpc_set_waitqueue_cookie(struct rpc_wait_queue *queue, unsigned long cookie)
+static inline void rpc_set_waitqueue_owner(struct rpc_wait_queue *queue, pid_t pid)
 {
-       queue->cookie = cookie;
+       queue->owner = pid;
        queue->nr = RPC_BATCH_COUNT;
 }
 
 static inline void rpc_reset_waitqueue_priority(struct rpc_wait_queue *queue)
 {
        rpc_set_waitqueue_priority(queue, queue->maxpriority);
-       rpc_set_waitqueue_cookie(queue, 0);
+       rpc_set_waitqueue_owner(queue, 0);
 }
 
-static void __rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname, int maxprio)
+static void __rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname, unsigned char nr_queues)
 {
        int i;
 
        spin_lock_init(&queue->lock);
        for (i = 0; i < ARRAY_SIZE(queue->tasks); i++)
                INIT_LIST_HEAD(&queue->tasks[i]);
-       queue->maxpriority = maxprio;
+       queue->maxpriority = nr_queues - 1;
        rpc_reset_waitqueue_priority(queue);
+       queue->qlen = 0;
+       setup_timer(&queue->timer_list.timer, __rpc_queue_timer_fn, (unsigned long)queue);
+       INIT_LIST_HEAD(&queue->timer_list.list);
 #ifdef RPC_DEBUG
        queue->name = qname;
 #endif
@@ -250,18 +208,25 @@ static void __rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const c
 
 void rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname)
 {
-       __rpc_init_priority_wait_queue(queue, qname, RPC_PRIORITY_HIGH);
+       __rpc_init_priority_wait_queue(queue, qname, RPC_NR_PRIORITY);
 }
+EXPORT_SYMBOL_GPL(rpc_init_priority_wait_queue);
 
 void rpc_init_wait_queue(struct rpc_wait_queue *queue, const char *qname)
 {
-       __rpc_init_priority_wait_queue(queue, qname, 0);
+       __rpc_init_priority_wait_queue(queue, qname, 1);
 }
-EXPORT_SYMBOL(rpc_init_wait_queue);
+EXPORT_SYMBOL_GPL(rpc_init_wait_queue);
 
-static int rpc_wait_bit_interruptible(void *word)
+void rpc_destroy_wait_queue(struct rpc_wait_queue *queue)
 {
-       if (signal_pending(current))
+       del_timer_sync(&queue->timer_list.timer);
+}
+EXPORT_SYMBOL_GPL(rpc_destroy_wait_queue);
+
+static int rpc_wait_bit_killable(void *word)
+{
+       if (fatal_signal_pending(current))
                return -ERESTARTSYS;
        schedule();
        return 0;
@@ -313,11 +278,11 @@ static void rpc_mark_complete_task(struct rpc_task *task)
 int __rpc_wait_for_completion_task(struct rpc_task *task, int (*action)(void *))
 {
        if (action == NULL)
-               action = rpc_wait_bit_interruptible;
+               action = rpc_wait_bit_killable;
        return wait_on_bit(&task->tk_runstate, RPC_TASK_ACTIVE,
-                       action, TASK_INTERRUPTIBLE);
+                       action, TASK_KILLABLE);
 }
-EXPORT_SYMBOL(__rpc_wait_for_completion_task);
+EXPORT_SYMBOL_GPL(__rpc_wait_for_completion_task);
 
 /*
  * Make an RPC task runnable.
@@ -327,20 +292,14 @@ EXPORT_SYMBOL(__rpc_wait_for_completion_task);
  */
 static void rpc_make_runnable(struct rpc_task *task)
 {
-       BUG_ON(task->tk_timeout_fn);
        rpc_clear_queued(task);
        if (rpc_test_and_set_running(task))
                return;
-       /* We might have raced */
-       if (RPC_IS_QUEUED(task)) {
-               rpc_clear_running(task);
-               return;
-       }
        if (RPC_IS_ASYNC(task)) {
                int status;
 
                INIT_WORK(&task->u.tk_work, rpc_async_schedule);
-               status = queue_work(task->tk_workqueue, &task->u.tk_work);
+               status = queue_work(rpciod_workqueue, &task->u.tk_work);
                if (status < 0) {
                        printk(KERN_WARNING "RPC: failed to add task to queue: error: %d!\n", status);
                        task->tk_status = status;
@@ -357,7 +316,7 @@ static void rpc_make_runnable(struct rpc_task *task)
  * as it's on a wait queue.
  */
 static void __rpc_sleep_on(struct rpc_wait_queue *q, struct rpc_task *task,
-                       rpc_action action, rpc_action timer)
+                       rpc_action action)
 {
        dprintk("RPC: %5u sleep_on(queue \"%s\" time %lu)\n",
                        task->tk_pid, rpc_qname(q), jiffies);
@@ -371,11 +330,11 @@ static void __rpc_sleep_on(struct rpc_wait_queue *q, struct rpc_task *task,
 
        BUG_ON(task->tk_callback != NULL);
        task->tk_callback = action;
-       __rpc_add_timer(task, timer);
+       __rpc_add_timer(q, task);
 }
 
 void rpc_sleep_on(struct rpc_wait_queue *q, struct rpc_task *task,
-                               rpc_action action, rpc_action timer)
+                               rpc_action action)
 {
        /* Mark the task as being activated if so needed */
        rpc_set_active(task);
@@ -384,17 +343,19 @@ void rpc_sleep_on(struct rpc_wait_queue *q, struct rpc_task *task,
         * Protect the queue operations.
         */
        spin_lock_bh(&q->lock);
-       __rpc_sleep_on(q, task, action, timer);
+       __rpc_sleep_on(q, task, action);
        spin_unlock_bh(&q->lock);
 }
+EXPORT_SYMBOL_GPL(rpc_sleep_on);
 
 /**
  * __rpc_do_wake_up_task - wake up a single rpc_task
+ * @queue: wait queue
  * @task: task to be woken up
  *
  * Caller must hold queue->lock, and have cleared the task queued flag.
  */
-static void __rpc_do_wake_up_task(struct rpc_task *task)
+static void __rpc_do_wake_up_task(struct rpc_wait_queue *queue, struct rpc_task *task)
 {
        dprintk("RPC: %5u __rpc_wake_up_task (now %lu)\n",
                        task->tk_pid, jiffies);
@@ -408,8 +369,7 @@ static void __rpc_do_wake_up_task(struct rpc_task *task)
                return;
        }
 
-       __rpc_disable_timer(task);
-       __rpc_remove_wait_queue(task);
+       __rpc_remove_wait_queue(queue, task);
 
        rpc_make_runnable(task);
 
@@ -417,46 +377,45 @@ static void __rpc_do_wake_up_task(struct rpc_task *task)
 }
 
 /*
- * Wake up the specified task
+ * Wake up a queued task while the queue lock is being held
  */
-static void __rpc_wake_up_task(struct rpc_task *task)
+static void rpc_wake_up_task_queue_locked(struct rpc_wait_queue *queue, struct rpc_task *task)
 {
-       if (rpc_start_wakeup(task)) {
-               if (RPC_IS_QUEUED(task))
-                       __rpc_do_wake_up_task(task);
-               rpc_finish_wakeup(task);
-       }
+       if (RPC_IS_QUEUED(task) && task->tk_waitqueue == queue)
+               __rpc_do_wake_up_task(queue, task);
 }
 
 /*
- * Default timeout handler if none specified by user
+ * Tests whether rpc queue is empty
  */
-static void
-__rpc_default_timer(struct rpc_task *task)
+int rpc_queue_empty(struct rpc_wait_queue *queue)
+{
+       int res;
+
+       spin_lock_bh(&queue->lock);
+       res = queue->qlen;
+       spin_unlock_bh(&queue->lock);
+       return (res == 0);
+}
+EXPORT_SYMBOL_GPL(rpc_queue_empty);
+
+/*
+ * Wake up a task on a specific queue
+ */
+void rpc_wake_up_queued_task(struct rpc_wait_queue *queue, struct rpc_task *task)
 {
-       dprintk("RPC: %5u timeout (default timer)\n", task->tk_pid);
-       task->tk_status = -ETIMEDOUT;
-       rpc_wake_up_task(task);
+       spin_lock_bh(&queue->lock);
+       rpc_wake_up_task_queue_locked(queue, task);
+       spin_unlock_bh(&queue->lock);
 }
+EXPORT_SYMBOL_GPL(rpc_wake_up_queued_task);
 
 /*
  * Wake up the specified task
  */
-void rpc_wake_up_task(struct rpc_task *task)
+static void rpc_wake_up_task(struct rpc_task *task)
 {
-       rcu_read_lock_bh();
-       if (rpc_start_wakeup(task)) {
-               if (RPC_IS_QUEUED(task)) {
-                       struct rpc_wait_queue *queue = task->u.tk_wait.rpc_waitq;
-
-                       /* Note: we're already in a bh-safe context */
-                       spin_lock(&queue->lock);
-                       __rpc_do_wake_up_task(task);
-                       spin_unlock(&queue->lock);
-               }
-               rpc_finish_wakeup(task);
-       }
-       rcu_read_unlock_bh();
+       rpc_wake_up_queued_task(task->tk_waitqueue, task);
 }
 
 /*
@@ -468,12 +427,12 @@ static struct rpc_task * __rpc_wake_up_next_priority(struct rpc_wait_queue *queu
        struct rpc_task *task;
 
        /*
-        * Service a batch of tasks from a single cookie.
+        * Service a batch of tasks from a single owner.
         */
        q = &queue->tasks[queue->priority];
        if (!list_empty(q)) {
                task = list_entry(q->next, struct rpc_task, u.tk_wait.list);
-               if (queue->cookie == task->tk_cookie) {
+               if (queue->owner == task->tk_owner) {
                        if (--queue->nr)
                                goto out;
                        list_move_tail(&task->u.tk_wait.list, q);
@@ -482,7 +441,7 @@ static struct rpc_task * __rpc_wake_up_next_priority(struct rpc_wait_queue *queu
                 * Check if we need to switch queues.
                 */
                if (--queue->count)
-                       goto new_cookie;
+                       goto new_owner;
        }
 
        /*
@@ -504,10 +463,10 @@ static struct rpc_task * __rpc_wake_up_next_priority(struct rpc_wait_queue *queu
 
 new_queue:
        rpc_set_waitqueue_priority(queue, (unsigned int)(q - &queue->tasks[0]));
-new_cookie:
-       rpc_set_waitqueue_cookie(queue, task->tk_cookie);
+new_owner:
+       rpc_set_waitqueue_owner(queue, task->tk_owner);
 out:
-       __rpc_wake_up_task(task);
+       rpc_wake_up_task_queue_locked(queue, task);
        return task;
 }
 
@@ -520,19 +479,18 @@ struct rpc_task * rpc_wake_up_next(struct rpc_wait_queue *queue)
 
        dprintk("RPC:       wake_up_next(%p \"%s\")\n",
                        queue, rpc_qname(queue));
-       rcu_read_lock_bh();
-       spin_lock(&queue->lock);
+       spin_lock_bh(&queue->lock);
        if (RPC_IS_PRIORITY(queue))
                task = __rpc_wake_up_next_priority(queue);
        else {
                task_for_first(task, &queue->tasks[0])
-                       __rpc_wake_up_task(task);
+                       rpc_wake_up_task_queue_locked(queue, task);
        }
-       spin_unlock(&queue->lock);
-       rcu_read_unlock_bh();
+       spin_unlock_bh(&queue->lock);
 
        return task;
 }
+EXPORT_SYMBOL_GPL(rpc_wake_up_next);
 
 /**
  * rpc_wake_up - wake up all rpc_tasks
@@ -545,19 +503,18 @@ void rpc_wake_up(struct rpc_wait_queue *queue)
        struct rpc_task *task, *next;
        struct list_head *head;
 
-       rcu_read_lock_bh();
-       spin_lock(&queue->lock);
+       spin_lock_bh(&queue->lock);
        head = &queue->tasks[queue->maxpriority];
        for (;;) {
                list_for_each_entry_safe(task, next, head, u.tk_wait.list)
-                       __rpc_wake_up_task(task);
+                       rpc_wake_up_task_queue_locked(queue, task);
                if (head == &queue->tasks[0])
                        break;
                head--;
        }
-       spin_unlock(&queue->lock);
-       rcu_read_unlock_bh();
+       spin_unlock_bh(&queue->lock);
 }
+EXPORT_SYMBOL_GPL(rpc_wake_up);
 
 /**
  * rpc_wake_up_status - wake up all rpc_tasks and set their status value.
@@ -571,25 +528,48 @@ void rpc_wake_up_status(struct rpc_wait_queue *queue, int status)
        struct rpc_task *task, *next;
        struct list_head *head;
 
-       rcu_read_lock_bh();
-       spin_lock(&queue->lock);
+       spin_lock_bh(&queue->lock);
        head = &queue->tasks[queue->maxpriority];
        for (;;) {
                list_for_each_entry_safe(task, next, head, u.tk_wait.list) {
                        task->tk_status = status;
-                       __rpc_wake_up_task(task);
+                       rpc_wake_up_task_queue_locked(queue, task);
                }
                if (head == &queue->tasks[0])
                        break;
                head--;
        }
+       spin_unlock_bh(&queue->lock);
+}
+EXPORT_SYMBOL_GPL(rpc_wake_up_status);
+
+static void __rpc_queue_timer_fn(unsigned long ptr)
+{
+       struct rpc_wait_queue *queue = (struct rpc_wait_queue *)ptr;
+       struct rpc_task *task, *n;
+       unsigned long expires, now, timeo;
+
+       spin_lock(&queue->lock);
+       expires = now = jiffies;
+       list_for_each_entry_safe(task, n, &queue->timer_list.list, u.tk_wait.timer_list) {
+               timeo = task->u.tk_wait.expires;
+               if (time_after_eq(now, timeo)) {
+                       dprintk("RPC: %5u timeout\n", task->tk_pid);
+                       task->tk_status = -ETIMEDOUT;
+                       rpc_wake_up_task_queue_locked(queue, task);
+                       continue;
+               }
+               if (expires == now || time_after(expires, timeo))
+                       expires = timeo;
+       }
+       if (!list_empty(&queue->timer_list.list))
+               rpc_set_queue_timer(queue, expires);
        spin_unlock(&queue->lock);
-       rcu_read_unlock_bh();
 }
 
 static void __rpc_atrun(struct rpc_task *task)
 {
-       rpc_wake_up_task(task);
+       task->tk_status = 0;
 }
 
 /*
@@ -598,17 +578,16 @@ static void __rpc_atrun(struct rpc_task *task)
 void rpc_delay(struct rpc_task *task, unsigned long delay)
 {
        task->tk_timeout = delay;
-       rpc_sleep_on(&delay_queue, task, NULL, __rpc_atrun);
+       rpc_sleep_on(&delay_queue, task, __rpc_atrun);
 }
+EXPORT_SYMBOL_GPL(rpc_delay);
 
 /*
  * Helper to call task->tk_ops->rpc_call_prepare
  */
-static void rpc_prepare_task(struct rpc_task *task)
+void rpc_prepare_task(struct rpc_task *task)
 {
-       lock_kernel();
        task->tk_ops->rpc_call_prepare(task, task->tk_calldata);
-       unlock_kernel();
 }
 
 /*
@@ -618,9 +597,7 @@ void rpc_exit_task(struct rpc_task *task)
 {
        task->tk_action = NULL;
        if (task->tk_ops->rpc_call_done != NULL) {
-               lock_kernel();
                task->tk_ops->rpc_call_done(task, task->tk_calldata);
-               unlock_kernel();
                if (task->tk_action != NULL) {
                        WARN_ON(RPC_ASSASSINATED(task));
                        /* Always release the RPC slot and buffer memory */
@@ -628,15 +605,12 @@ void rpc_exit_task(struct rpc_task *task)
                }
        }
 }
-EXPORT_SYMBOL(rpc_exit_task);
+EXPORT_SYMBOL_GPL(rpc_exit_task);
 
 void rpc_release_calldata(const struct rpc_call_ops *ops, void *calldata)
 {
-       if (ops->rpc_release != NULL) {
-               lock_kernel();
+       if (ops->rpc_release != NULL)
                ops->rpc_release(calldata);
-               unlock_kernel();
-       }
 }
 
 /*
@@ -644,7 +618,9 @@ void rpc_release_calldata(const struct rpc_call_ops *ops, void *calldata)
  */
 static void __rpc_execute(struct rpc_task *task)
 {
-       int             status = 0;
+       struct rpc_wait_queue *queue;
+       int task_is_async = RPC_IS_ASYNC(task);
+       int status = 0;
 
        dprintk("RPC: %5u __rpc_execute flags=0x%x\n",
                        task->tk_pid, task->tk_flags);
@@ -652,27 +628,19 @@ static void __rpc_execute(struct rpc_task *task)
        BUG_ON(RPC_IS_QUEUED(task));
 
        for (;;) {
-               /*
-                * Garbage collection of pending timers...
-                */
-               rpc_delete_timer(task);
 
                /*
                 * Execute any pending callback.
                 */
-               if (RPC_DO_CALLBACK(task)) {
-                       /* Define a callback save pointer */
+               if (task->tk_callback) {
                        void (*save_callback)(struct rpc_task *);
 
                        /*
-                        * If a callback exists, save it, reset it,
-                        * call it.
-                        * The save is needed to stop from resetting
-                        * another callback set within the callback handler
-                        * - Dave
+                        * We set tk_callback to NULL before calling it,
+                        * in case it sets the tk_callback field itself:
                         */
-                       save_callback=task->tk_callback;
-                       task->tk_callback=NULL;
+                       save_callback = task->tk_callback;
+                       task->tk_callback = NULL;
                        save_callback(task);
                }
 
@@ -692,22 +660,31 @@ static void __rpc_execute(struct rpc_task *task)
                 */
                if (!RPC_IS_QUEUED(task))
                        continue;
-               rpc_clear_running(task);
-               if (RPC_IS_ASYNC(task)) {
-                       /* Careful! we may have raced... */
-                       if (RPC_IS_QUEUED(task))
-                               return;
-                       if (rpc_test_and_set_running(task))
-                               return;
+               /*
+                * The queue->lock protects against races with
+                * rpc_make_runnable().
+                *
+                * Note that once we clear RPC_TASK_RUNNING on an asynchronous
+                * rpc_task, rpc_make_runnable() can assign it to a
+                * different workqueue. We therefore cannot assume that the
+                * rpc_task pointer may still be dereferenced.
+                */
+               queue = task->tk_waitqueue;
+               spin_lock_bh(&queue->lock);
+               if (!RPC_IS_QUEUED(task)) {
+                       spin_unlock_bh(&queue->lock);
                        continue;
                }
+               rpc_clear_running(task);
+               spin_unlock_bh(&queue->lock);
+               if (task_is_async)
+                       return;
 
                /* sync task: sleep here */
                dprintk("RPC: %5u sync task going to sleep\n", task->tk_pid);
-               /* Note: Caller should be using rpc_clnt_sigmask() */
                status = out_of_line_wait_on_bit(&task->tk_runstate,
-                               RPC_TASK_QUEUED, rpc_wait_bit_interruptible,
-                               TASK_INTERRUPTIBLE);
+                               RPC_TASK_QUEUED, rpc_wait_bit_killable,
+                               TASK_KILLABLE);
                if (status == -ERESTARTSYS) {
                        /*
                         * When a sync task receives a signal, it exits with
@@ -751,11 +728,6 @@ static void rpc_async_schedule(struct work_struct *work)
        __rpc_execute(container_of(work, struct rpc_task, u.tk_work));
 }
 
-struct rpc_buffer {
-       size_t  len;
-       char    data[];
-};
-
 /**
  * rpc_malloc - allocate an RPC buffer
  * @task: RPC task that will use this buffer
@@ -791,6 +763,7 @@ void *rpc_malloc(struct rpc_task *task, size_t size)
                        task->tk_pid, size, buf);
        return &buf->data;
 }
+EXPORT_SYMBOL_GPL(rpc_malloc);
 
 /**
  * rpc_free - free buffer allocated via rpc_malloc
@@ -816,50 +789,55 @@ void rpc_free(void *buffer)
        else
                kfree(buf);
 }
+EXPORT_SYMBOL_GPL(rpc_free);
 
 /*
  * Creation and deletion of RPC task structures
  */
-void rpc_init_task(struct rpc_task *task, struct rpc_clnt *clnt, int flags, const struct rpc_call_ops *tk_ops, void *calldata)
+static void rpc_init_task(struct rpc_task *task, const struct rpc_task_setup *task_setup_data)
 {
        memset(task, 0, sizeof(*task));
-       init_timer(&task->tk_timer);
-       task->tk_timer.data     = (unsigned long) task;
-       task->tk_timer.function = (void (*)(unsigned long)) rpc_run_timer;
        atomic_set(&task->tk_count, 1);
-       task->tk_client = clnt;
-       task->tk_flags  = flags;
-       task->tk_ops = tk_ops;
-       if (tk_ops->rpc_call_prepare != NULL)
-               task->tk_action = rpc_prepare_task;
-       task->tk_calldata = calldata;
+       task->tk_flags  = task_setup_data->flags;
+       task->tk_ops = task_setup_data->callback_ops;
+       task->tk_calldata = task_setup_data->callback_data;
        INIT_LIST_HEAD(&task->tk_task);
 
        /* Initialize retry counters */
        task->tk_garb_retry = 2;
        task->tk_cred_retry = 2;
 
-       task->tk_priority = RPC_PRIORITY_NORMAL;
-       task->tk_cookie = (unsigned long)current;
+       task->tk_priority = task_setup_data->priority - RPC_PRIORITY_LOW;
+       task->tk_owner = current->tgid;
 
        /* Initialize workqueue for async tasks */
-       task->tk_workqueue = rpciod_workqueue;
+       task->tk_workqueue = task_setup_data->workqueue;
 
-       if (clnt) {
-               kref_get(&clnt->cl_kref);
-               if (clnt->cl_softrtry)
+       task->tk_client = task_setup_data->rpc_client;
+       if (task->tk_client != NULL) {
+               kref_get(&task->tk_client->cl_kref);
+               if (task->tk_client->cl_softrtry)
                        task->tk_flags |= RPC_TASK_SOFT;
-               if (!clnt->cl_intr)
-                       task->tk_flags |= RPC_TASK_NOINTR;
        }
 
-       BUG_ON(task->tk_ops == NULL);
+       if (task->tk_ops->rpc_call_prepare != NULL)
+               task->tk_action = rpc_prepare_task;
+
+       if (task_setup_data->rpc_message != NULL) {
+               task->tk_msg.rpc_proc = task_setup_data->rpc_message->rpc_proc;
+               task->tk_msg.rpc_argp = task_setup_data->rpc_message->rpc_argp;
+               task->tk_msg.rpc_resp = task_setup_data->rpc_message->rpc_resp;
+               /* Bind the user cred */
+               rpcauth_bindcred(task, task_setup_data->rpc_message->rpc_cred, task_setup_data->flags);
+               if (task->tk_action == NULL)
+                       rpc_call_start(task);
+       }
 
        /* starting timestamp */
        task->tk_start = jiffies;
 
        dprintk("RPC:       new task initialized, procpid %u\n",
-                               current->pid);
+                               task_pid_nr(current));
 }
 
 static struct rpc_task *
@@ -868,38 +846,48 @@ rpc_alloc_task(void)
        return (struct rpc_task *)mempool_alloc(rpc_task_mempool, GFP_NOFS);
 }
 
-static void rpc_free_task(struct rcu_head *rcu)
-{
-       struct rpc_task *task = container_of(rcu, struct rpc_task, u.tk_rcu);
-       dprintk("RPC: %5u freeing task\n", task->tk_pid);
-       mempool_free(task, rpc_task_mempool);
-}
-
 /*
  * Create a new task for the specified client.
  */
-struct rpc_task *rpc_new_task(struct rpc_clnt *clnt, int flags, const struct rpc_call_ops *tk_ops, void *calldata)
+struct rpc_task *rpc_new_task(const struct rpc_task_setup *setup_data)
 {
-       struct rpc_task *task;
+       struct rpc_task *task = setup_data->task;
+       unsigned short flags = 0;
 
-       task = rpc_alloc_task();
-       if (!task)
-               goto out;
+       if (task == NULL) {
+               task = rpc_alloc_task();
+               if (task == NULL)
+                       goto out;
+               flags = RPC_TASK_DYNAMIC;
+       }
 
-       rpc_init_task(task, clnt, flags, tk_ops, calldata);
+       rpc_init_task(task, setup_data);
 
+       task->tk_flags |= flags;
        dprintk("RPC:       allocated task %p\n", task);
-       task->tk_flags |= RPC_TASK_DYNAMIC;
 out:
        return task;
 }
 
-
-void rpc_put_task(struct rpc_task *task)
+static void rpc_free_task(struct rpc_task *task)
 {
        const struct rpc_call_ops *tk_ops = task->tk_ops;
        void *calldata = task->tk_calldata;
 
+       if (task->tk_flags & RPC_TASK_DYNAMIC) {
+               dprintk("RPC: %5u freeing task\n", task->tk_pid);
+               mempool_free(task, rpc_task_mempool);
+       }
+       rpc_release_calldata(tk_ops, calldata);
+}
+
+static void rpc_async_release(struct work_struct *work)
+{
+       rpc_free_task(container_of(work, struct rpc_task, u.tk_work));
+}
+
+void rpc_put_task(struct rpc_task *task)
+{
        if (!atomic_dec_and_test(&task->tk_count))
                return;
        /* Release resources */
@@ -911,11 +899,13 @@ void rpc_put_task(struct rpc_task *task)
                rpc_release_client(task->tk_client);
                task->tk_client = NULL;
        }
-       if (task->tk_flags & RPC_TASK_DYNAMIC)
-               call_rcu_bh(&task->u.tk_rcu, rpc_free_task);
-       rpc_release_calldata(tk_ops, calldata);
+       if (task->tk_workqueue != NULL) {
+               INIT_WORK(&task->u.tk_work, rpc_async_release);
+               queue_work(task->tk_workqueue, &task->u.tk_work);
+       } else
+               rpc_free_task(task);
 }
-EXPORT_SYMBOL(rpc_put_task);
+EXPORT_SYMBOL_GPL(rpc_put_task);
 
 static void rpc_release_task(struct rpc_task *task)
 {
@@ -933,9 +923,6 @@ static void rpc_release_task(struct rpc_task *task)
        }
        BUG_ON (RPC_IS_QUEUED(task));
 
-       /* Synchronously delete any running timer */
-       rpc_delete_timer(task);
-
 #ifdef RPC_DEBUG
        task->tk_magic = 0;
 #endif
@@ -945,29 +932,6 @@ static void rpc_release_task(struct rpc_task *task)
        rpc_put_task(task);
 }
 
-/**
- * rpc_run_task - Allocate a new RPC task, then run rpc_execute against it
- * @clnt: pointer to RPC client
- * @flags: RPC flags
- * @ops: RPC call ops
- * @data: user call data
- */
-struct rpc_task *rpc_run_task(struct rpc_clnt *clnt, int flags,
-                                       const struct rpc_call_ops *ops,
-                                       void *data)
-{
-       struct rpc_task *task;
-       task = rpc_new_task(clnt, flags, ops, data);
-       if (task == NULL) {
-               rpc_release_calldata(ops, data);
-               return ERR_PTR(-ENOMEM);
-       }
-       atomic_inc(&task->tk_count);
-       rpc_execute(task);
-       return task;
-}
-EXPORT_SYMBOL(rpc_run_task);
-
 /*
  * Kill all tasks for the given client.
  * XXX: kill their descendants as well?
@@ -995,142 +959,51 @@ void rpc_killall_tasks(struct rpc_clnt *clnt)
        }
        spin_unlock(&clnt->cl_lock);
 }
+EXPORT_SYMBOL_GPL(rpc_killall_tasks);
 
-static void rpciod_killall(void)
-{
-       struct rpc_clnt *clnt;
-       unsigned long flags;
-
-       for(;;) {
-               clear_thread_flag(TIF_SIGPENDING);
-
-               spin_lock(&rpc_sched_lock);
-               list_for_each_entry(clnt, &all_clients, cl_clients)
-                       rpc_killall_tasks(clnt);
-               spin_unlock(&rpc_sched_lock);
-               flush_workqueue(rpciod_workqueue);
-               if (!list_empty(&all_clients))
-                       break;
-               dprintk("RPC:       rpciod_killall: waiting for tasks "
-                                       "to exit\n");
-               wait_event_timeout(client_kill_wait,
-                               list_empty(&all_clients), 1*HZ);
-       }
-
-       spin_lock_irqsave(&current->sighand->siglock, flags);
-       recalc_sigpending();
-       spin_unlock_irqrestore(&current->sighand->siglock, flags);
-}
-
-void rpc_register_client(struct rpc_clnt *clnt)
+int rpciod_up(void)
 {
-       spin_lock(&rpc_sched_lock);
-       list_add(&clnt->cl_clients, &all_clients);
-       spin_unlock(&rpc_sched_lock);
+       return try_module_get(THIS_MODULE) ? 0 : -EINVAL;
 }
 
-void rpc_unregister_client(struct rpc_clnt *clnt)
+void rpciod_down(void)
 {
-       spin_lock(&rpc_sched_lock);
-       list_del(&clnt->cl_clients);
-       if (list_empty(&all_clients))
-               wake_up(&client_kill_wait);
-       spin_unlock(&rpc_sched_lock);
+       module_put(THIS_MODULE);
 }
 
 /*
- * Start up the rpciod process if it's not already running.
+ * Start up the rpciod workqueue.
  */
-int
-rpciod_up(void)
+static int rpciod_start(void)
 {
        struct workqueue_struct *wq;
-       int error = 0;
 
-       if (atomic_inc_not_zero(&rpciod_users))
-               return 0;
-
-       mutex_lock(&rpciod_mutex);
-
-       /* Guard against races with rpciod_down() */
-       if (rpciod_workqueue != NULL)
-               goto out_ok;
        /*
         * Create the rpciod thread and wait for it to start.
         */
        dprintk("RPC:       creating workqueue rpciod\n");
-       error = -ENOMEM;
        wq = create_workqueue("rpciod");
-       if (wq == NULL)
-               goto out;
-
        rpciod_workqueue = wq;
-       error = 0;
-out_ok:
-       atomic_inc(&rpciod_users);
-out:
-       mutex_unlock(&rpciod_mutex);
-       return error;
+       return rpciod_workqueue != NULL;
 }
 
-void
-rpciod_down(void)
+static void rpciod_stop(void)
 {
-       if (!atomic_dec_and_test(&rpciod_users))
-               return;
+       struct workqueue_struct *wq = NULL;
 
-       mutex_lock(&rpciod_mutex);
+       if (rpciod_workqueue == NULL)
+               return;
        dprintk("RPC:       destroying workqueue rpciod\n");
 
-       if (atomic_read(&rpciod_users) == 0 && rpciod_workqueue != NULL) {
-               rpciod_killall();
-               destroy_workqueue(rpciod_workqueue);
-               rpciod_workqueue = NULL;
-       }
-       mutex_unlock(&rpciod_mutex);
+       wq = rpciod_workqueue;
+       rpciod_workqueue = NULL;
+       destroy_workqueue(wq);
 }
 
-#ifdef RPC_DEBUG
-void rpc_show_tasks(void)
-{
-       struct rpc_clnt *clnt;
-       struct rpc_task *t;
-
-       spin_lock(&rpc_sched_lock);
-       if (list_empty(&all_clients))
-               goto out;
-       printk("-pid- proc flgs status -client- -prog- --rqstp- -timeout "
-               "-rpcwait -action- ---ops--\n");
-       list_for_each_entry(clnt, &all_clients, cl_clients) {
-               if (list_empty(&clnt->cl_tasks))
-                       continue;
-               spin_lock(&clnt->cl_lock);
-               list_for_each_entry(t, &clnt->cl_tasks, tk_task) {
-                       const char *rpc_waitq = "none";
-
-                       if (RPC_IS_QUEUED(t))
-                               rpc_waitq = rpc_qname(t->u.tk_wait.rpc_waitq);
-
-                       printk("%5u %04d %04x %6d %8p %6d %8p %8ld %8s %8p %8p\n",
-                               t->tk_pid,
-                               (t->tk_msg.rpc_proc ? t->tk_msg.rpc_proc->p_proc : -1),
-                               t->tk_flags, t->tk_status,
-                               t->tk_client,
-                               (t->tk_client ? t->tk_client->cl_prog : 0),
-                               t->tk_rqstp, t->tk_timeout,
-                               rpc_waitq,
-                               t->tk_action, t->tk_ops);
-               }
-               spin_unlock(&clnt->cl_lock);
-       }
-out:
-       spin_unlock(&rpc_sched_lock);
-}
-#endif
-
 void
 rpc_destroy_mempool(void)
 {
+       rpciod_stop();
        if (rpc_buffer_mempool)
                mempool_destroy(rpc_buffer_mempool);
        if (rpc_task_mempool)
@@ -1139,21 +1012,30 @@ rpc_destroy_mempool(void)
                kmem_cache_destroy(rpc_task_slabp);
        if (rpc_buffer_slabp)
                kmem_cache_destroy(rpc_buffer_slabp);
+       rpc_destroy_wait_queue(&delay_queue);
 }
 
 int
 rpc_init_mempool(void)
 {
+       /*
+        * The following is not strictly a mempool initialisation,
+        * but there is no harm in doing it here
+        */
+       rpc_init_wait_queue(&delay_queue, "delayq");
+       if (!rpciod_start())
+               goto err_nomem;
+
        rpc_task_slabp = kmem_cache_create("rpc_tasks",
                                             sizeof(struct rpc_task),
                                             0, SLAB_HWCACHE_ALIGN,
-                                            NULL, NULL);
+                                            NULL);
        if (!rpc_task_slabp)
                goto err_nomem;
        rpc_buffer_slabp = kmem_cache_create("rpc_buffers",
                                             RPC_BUFFER_MAXSIZE,
                                             0, SLAB_HWCACHE_ALIGN,
-                                            NULL, NULL);
+                                            NULL);
        if (!rpc_buffer_slabp)
                goto err_nomem;
        rpc_task_mempool = mempool_create_slab_pool(RPC_TASK_POOLSIZE,