Revert "ipv6: Fix typo in net/ipv6/Kconfig"
[safe/jmp/linux-2.6] / net / sunrpc / sched.c
index f9fd66b..4c66912 100644 (file)
@@ -4,7 +4,7 @@
  * Scheduling for synchronous and asynchronous RPC requests.
  *
  * Copyright (C) 1996 Olaf Kirch, <okir@monad.swb.de>
- * 
+ *
  * TCP NFS related read + write fixes
  * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
  */
@@ -25,7 +25,6 @@
 #ifdef RPC_DEBUG
 #define RPCDBG_FACILITY                RPCDBG_SCHED
 #define RPC_TASK_MAGIC_ID      0xf00baa
-static int                     rpc_task_id;
 #endif
 
 /*
@@ -34,38 +33,26 @@ static int                  rpc_task_id;
 #define RPC_BUFFER_MAXSIZE     (2048)
 #define RPC_BUFFER_POOLSIZE    (8)
 #define RPC_TASK_POOLSIZE      (8)
-static kmem_cache_t    *rpc_task_slabp __read_mostly;
-static kmem_cache_t    *rpc_buffer_slabp __read_mostly;
+static struct kmem_cache       *rpc_task_slabp __read_mostly;
+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(void *);
+static void                    rpc_async_schedule(struct work_struct *);
+static void                     rpc_release_task(struct rpc_task *task);
 
 /*
  * RPC tasks sit here while waiting for conditions to improve.
  */
-static RPC_WAITQ(delay_queue, "delayq");
-
-/*
- * All RPC tasks are linked into this list
- */
-static LIST_HEAD(all_tasks);
+static struct rpc_wait_queue delay_queue;
 
 /*
  * rpciod-related stuff
  */
-static DEFINE_MUTEX(rpciod_mutex);
-static unsigned int            rpciod_users;
 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().
@@ -73,7 +60,7 @@ static DEFINE_SPINLOCK(rpc_sched_lock);
 static inline void
 __rpc_disable_timer(struct rpc_task *task)
 {
-       dprintk("RPC: %4d disabling timer\n", task->tk_pid);
+       dprintk("RPC: %5u disabling timer\n", task->tk_pid);
        task->tk_timeout_fn = NULL;
        task->tk_timeout = 0;
 }
@@ -92,7 +79,7 @@ static void rpc_run_timer(struct rpc_task *task)
        callback = task->tk_timeout_fn;
        task->tk_timeout_fn = NULL;
        if (callback && RPC_IS_QUEUED(task)) {
-               dprintk("RPC: %4d running timer\n", task->tk_pid);
+               dprintk("RPC: %5u running timer\n", task->tk_pid);
                callback(task);
        }
        smp_mb__before_clear_bit();
@@ -109,7 +96,7 @@ __rpc_add_timer(struct rpc_task *task, rpc_action timer)
        if (!task->tk_timeout)
                return;
 
-       dprintk("RPC: %4d setting alarm for %lu ms\n",
+       dprintk("RPC: %5u setting alarm for %lu ms\n",
                        task->tk_pid, task->tk_timeout * 1000 / HZ);
 
        if (timer)
@@ -131,7 +118,7 @@ rpc_delete_timer(struct rpc_task *task)
                return;
        if (test_and_clear_bit(RPC_TASK_HAS_TIMER, &task->tk_runstate)) {
                del_singleshot_timer_sync(&task->tk_timer);
-               dprintk("RPC: %4d deleting timer\n", task->tk_pid);
+               dprintk("RPC: %5u deleting timer\n", task->tk_pid);
        }
 }
 
@@ -148,7 +135,7 @@ 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;
                }
@@ -178,8 +165,8 @@ static void __rpc_add_wait_queue(struct rpc_wait_queue *queue, struct rpc_task *
        queue->qlen++;
        rpc_set_queued(task);
 
-       dprintk("RPC: %4d added to queue %p \"%s\"\n",
-                               task->tk_pid, queue, rpc_qname(queue));
+       dprintk("RPC: %5u added to queue %p \"%s\"\n",
+                       task->tk_pid, queue, rpc_qname(queue));
 }
 
 /*
@@ -211,8 +198,8 @@ static void __rpc_remove_wait_queue(struct rpc_task *task)
        else
                list_del(&task->u.tk_wait.list);
        queue->qlen--;
-       dprintk("RPC: %4d removed from queue %p \"%s\"\n",
-                               task->tk_pid, queue, rpc_qname(queue));
+       dprintk("RPC: %5u removed from queue %p \"%s\"\n",
+                       task->tk_pid, queue, rpc_qname(queue));
 }
 
 static inline void rpc_set_waitqueue_priority(struct rpc_wait_queue *queue, int priority)
@@ -221,26 +208,26 @@ 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);
 #ifdef RPC_DEBUG
        queue->name = qname;
@@ -249,35 +236,50 @@ 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);
 }
 
 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)
+static int rpc_wait_bit_killable(void *word)
 {
-       if (signal_pending(current))
+       if (fatal_signal_pending(current))
                return -ERESTARTSYS;
        schedule();
        return 0;
 }
 
+#ifdef RPC_DEBUG
+static void rpc_task_set_debuginfo(struct rpc_task *task)
+{
+       static atomic_t rpc_pid;
+
+       task->tk_magic = RPC_TASK_MAGIC_ID;
+       task->tk_pid = atomic_inc_return(&rpc_pid);
+}
+#else
+static inline void rpc_task_set_debuginfo(struct rpc_task *task)
+{
+}
+#endif
+
 static void rpc_set_active(struct rpc_task *task)
 {
+       struct rpc_clnt *clnt;
        if (test_and_set_bit(RPC_TASK_ACTIVE, &task->tk_runstate) != 0)
                return;
-       spin_lock(&rpc_sched_lock);
-#ifdef RPC_DEBUG
-       task->tk_magic = RPC_TASK_MAGIC_ID;
-       task->tk_pid = rpc_task_id++;
-#endif
+       rpc_task_set_debuginfo(task);
        /* Add to global list of all tasks */
-       list_add_tail(&task->tk_task, &all_tasks);
-       spin_unlock(&rpc_sched_lock);
+       clnt = task->tk_client;
+       if (clnt != NULL) {
+               spin_lock(&clnt->cl_lock);
+               list_add_tail(&task->tk_task, &clnt->cl_tasks);
+               spin_unlock(&clnt->cl_lock);
+       }
 }
 
 /*
@@ -297,16 +299,16 @@ 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.
  *
- * Note: If the task is ASYNC, this must be called with 
+ * Note: If the task is ASYNC, this must be called with
  * the spinlock held to protect the wait queue operation.
  */
 static void rpc_make_runnable(struct rpc_task *task)
@@ -323,7 +325,7 @@ static void rpc_make_runnable(struct rpc_task *task)
        if (RPC_IS_ASYNC(task)) {
                int status;
 
-               INIT_WORK(&task->u.tk_work, rpc_async_schedule, (void *)task);
+               INIT_WORK(&task->u.tk_work, rpc_async_schedule);
                status = queue_work(task->tk_workqueue, &task->u.tk_work);
                if (status < 0) {
                        printk(KERN_WARNING "RPC: failed to add task to queue: error: %d!\n", status);
@@ -343,8 +345,8 @@ static void rpc_make_runnable(struct rpc_task *task)
 static void __rpc_sleep_on(struct rpc_wait_queue *q, struct rpc_task *task,
                        rpc_action action, rpc_action timer)
 {
-       dprintk("RPC: %4d sleep_on(queue \"%s\" time %ld)\n", task->tk_pid,
-                               rpc_qname(q), jiffies);
+       dprintk("RPC: %5u sleep_on(queue \"%s\" time %lu)\n",
+                       task->tk_pid, rpc_qname(q), jiffies);
 
        if (!RPC_IS_ASYNC(task) && !RPC_IS_ACTIVATED(task)) {
                printk(KERN_ERR "RPC: Inactive synchronous task put to sleep!\n");
@@ -371,6 +373,7 @@ void rpc_sleep_on(struct rpc_wait_queue *q, struct rpc_task *task,
        __rpc_sleep_on(q, task, action, timer);
        spin_unlock_bh(&q->lock);
 }
+EXPORT_SYMBOL_GPL(rpc_sleep_on);
 
 /**
  * __rpc_do_wake_up_task - wake up a single rpc_task
@@ -380,7 +383,8 @@ void rpc_sleep_on(struct rpc_wait_queue *q, struct rpc_task *task,
  */
 static void __rpc_do_wake_up_task(struct rpc_task *task)
 {
-       dprintk("RPC: %4d __rpc_wake_up_task (now %ld)\n", task->tk_pid, jiffies);
+       dprintk("RPC: %5u __rpc_wake_up_task (now %lu)\n",
+                       task->tk_pid, jiffies);
 
 #ifdef RPC_DEBUG
        BUG_ON(task->tk_magic != RPC_TASK_MAGIC_ID);
@@ -396,7 +400,7 @@ static void __rpc_do_wake_up_task(struct rpc_task *task)
 
        rpc_make_runnable(task);
 
-       dprintk("RPC:      __rpc_wake_up_task done\n");
+       dprintk("RPC:       __rpc_wake_up_task done\n");
 }
 
 /*
@@ -417,7 +421,7 @@ static void __rpc_wake_up_task(struct rpc_task *task)
 static void
 __rpc_default_timer(struct rpc_task *task)
 {
-       dprintk("RPC: %d timeout (default timer)\n", task->tk_pid);
+       dprintk("RPC: %5u timeout (default timer)\n", task->tk_pid);
        task->tk_status = -ETIMEDOUT;
        rpc_wake_up_task(task);
 }
@@ -441,6 +445,7 @@ void rpc_wake_up_task(struct rpc_task *task)
        }
        rcu_read_unlock_bh();
 }
+EXPORT_SYMBOL_GPL(rpc_wake_up_task);
 
 /*
  * Wake up the next task on a priority queue.
@@ -451,12 +456,12 @@ 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);
@@ -465,7 +470,7 @@ 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;
        }
 
        /*
@@ -487,8 +492,8 @@ 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);
        return task;
@@ -501,7 +506,8 @@ struct rpc_task * rpc_wake_up_next(struct rpc_wait_queue *queue)
 {
        struct rpc_task *task = NULL;
 
-       dprintk("RPC:      wake_up_next(%p \"%s\")\n", queue, rpc_qname(queue));
+       dprintk("RPC:       wake_up_next(%p \"%s\")\n",
+                       queue, rpc_qname(queue));
        rcu_read_lock_bh();
        spin_lock(&queue->lock);
        if (RPC_IS_PRIORITY(queue))
@@ -515,6 +521,7 @@ struct rpc_task * rpc_wake_up_next(struct rpc_wait_queue *queue)
 
        return task;
 }
+EXPORT_SYMBOL_GPL(rpc_wake_up_next);
 
 /**
  * rpc_wake_up - wake up all rpc_tasks
@@ -540,6 +547,7 @@ void rpc_wake_up(struct rpc_wait_queue *queue)
        spin_unlock(&queue->lock);
        rcu_read_unlock_bh();
 }
+EXPORT_SYMBOL_GPL(rpc_wake_up);
 
 /**
  * rpc_wake_up_status - wake up all rpc_tasks and set their status value.
@@ -568,6 +576,7 @@ void rpc_wake_up_status(struct rpc_wait_queue *queue, int status)
        spin_unlock(&queue->lock);
        rcu_read_unlock_bh();
 }
+EXPORT_SYMBOL_GPL(rpc_wake_up_status);
 
 static void __rpc_atrun(struct rpc_task *task)
 {
@@ -582,6 +591,7 @@ void rpc_delay(struct rpc_task *task, unsigned long delay)
        task->tk_timeout = delay;
        rpc_sleep_on(&delay_queue, task, NULL, __rpc_atrun);
 }
+EXPORT_SYMBOL_GPL(rpc_delay);
 
 /*
  * Helper to call task->tk_ops->rpc_call_prepare
@@ -610,7 +620,7 @@ 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)
 {
@@ -624,12 +634,12 @@ void rpc_release_calldata(const struct rpc_call_ops *ops, void *calldata)
 /*
  * This is the RPC `scheduler' (or rather, the finite state machine).
  */
-static int __rpc_execute(struct rpc_task *task)
+static void __rpc_execute(struct rpc_task *task)
 {
        int             status = 0;
 
-       dprintk("RPC: %4d rpc_execute flgs %x\n",
-                               task->tk_pid, task->tk_flags);
+       dprintk("RPC: %5u __rpc_execute flags=0x%x\n",
+                       task->tk_pid, task->tk_flags);
 
        BUG_ON(RPC_IS_QUEUED(task));
 
@@ -645,8 +655,8 @@ static int __rpc_execute(struct rpc_task *task)
                if (RPC_DO_CALLBACK(task)) {
                        /* Define a callback save pointer */
                        void (*save_callback)(struct rpc_task *);
-       
-                       /* 
+
+                       /*
                         * If a callback exists, save it, reset it,
                         * call it.
                         * The save is needed to stop from resetting
@@ -678,18 +688,17 @@ static int __rpc_execute(struct rpc_task *task)
                if (RPC_IS_ASYNC(task)) {
                        /* Careful! we may have raced... */
                        if (RPC_IS_QUEUED(task))
-                               return 0;
+                               return;
                        if (rpc_test_and_set_running(task))
-                               return 0;
+                               return;
                        continue;
                }
 
                /* sync task: sleep here */
-               dprintk("RPC: %4d sync task going to sleep\n", task->tk_pid);
-               /* Note: Caller should be using rpc_clnt_sigmask() */
+               dprintk("RPC: %5u sync task going to sleep\n", task->tk_pid);
                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
@@ -697,19 +706,19 @@ static int __rpc_execute(struct rpc_task *task)
                         * clean up after sleeping on some queue, we don't
                         * break the loop here, but go around once more.
                         */
-                       dprintk("RPC: %4d got signal\n", task->tk_pid);
+                       dprintk("RPC: %5u got signal\n", task->tk_pid);
                        task->tk_flags |= RPC_TASK_KILLED;
                        rpc_exit(task, -ERESTARTSYS);
                        rpc_wake_up_task(task);
                }
                rpc_set_running(task);
-               dprintk("RPC: %4d sync task resuming\n", task->tk_pid);
+               dprintk("RPC: %5u sync task resuming\n", task->tk_pid);
        }
 
-       dprintk("RPC: %4d, return %d, status %d\n", task->tk_pid, status, task->tk_status);
+       dprintk("RPC: %5u return %d, status %d\n", task->tk_pid, status,
+                       task->tk_status);
        /* Release all resources associated with the task */
        rpc_release_task(task);
-       return status;
 }
 
 /*
@@ -721,112 +730,136 @@ static int __rpc_execute(struct rpc_task *task)
  *      released. In particular note that tk_release() will have
  *      been called, so your task memory may have been freed.
  */
-int
-rpc_execute(struct rpc_task *task)
+void rpc_execute(struct rpc_task *task)
 {
        rpc_set_active(task);
        rpc_set_running(task);
-       return __rpc_execute(task);
+       __rpc_execute(task);
 }
 
-static void rpc_async_schedule(void *arg)
+static void rpc_async_schedule(struct work_struct *work)
 {
-       __rpc_execute((struct rpc_task *)arg);
+       __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
  * @size: requested byte size
  *
- * We try to ensure that some NFS reads and writes can always proceed
- * by using a mempool when allocating 'small' buffers.
+ * To prevent rpciod from hanging, this allocator never sleeps,
+ * returning NULL if the request cannot be serviced immediately.
+ * The caller can arrange to sleep in a way that is safe for rpciod.
+ *
+ * Most requests are 'small' (under 2KiB) and can be serviced from a
+ * mempool, ensuring that NFS reads and writes can always proceed,
+ * and that there is good locality of reference for these buffers.
+ *
  * In order to avoid memory starvation triggering more writebacks of
- * NFS requests, we use GFP_NOFS rather than GFP_KERNEL.
+ * NFS requests, we avoid using GFP_KERNEL.
  */
-void * rpc_malloc(struct rpc_task *task, size_t size)
+void *rpc_malloc(struct rpc_task *task, size_t size)
 {
-       struct rpc_rqst *req = task->tk_rqstp;
-       gfp_t   gfp;
+       struct rpc_buffer *buf;
+       gfp_t gfp = RPC_IS_SWAPPER(task) ? GFP_ATOMIC : GFP_NOWAIT;
 
-       if (task->tk_flags & RPC_TASK_SWAPPER)
-               gfp = GFP_ATOMIC;
+       size += sizeof(struct rpc_buffer);
+       if (size <= RPC_BUFFER_MAXSIZE)
+               buf = mempool_alloc(rpc_buffer_mempool, gfp);
        else
-               gfp = GFP_NOFS;
-
-       if (size > RPC_BUFFER_MAXSIZE) {
-               req->rq_buffer = kmalloc(size, gfp);
-               if (req->rq_buffer)
-                       req->rq_bufsize = size;
-       } else {
-               req->rq_buffer = mempool_alloc(rpc_buffer_mempool, gfp);
-               if (req->rq_buffer)
-                       req->rq_bufsize = RPC_BUFFER_MAXSIZE;
-       }
-       return req->rq_buffer;
+               buf = kmalloc(size, gfp);
+
+       if (!buf)
+               return NULL;
+
+       buf->len = size;
+       dprintk("RPC: %5u allocated buffer of size %zu at %p\n",
+                       task->tk_pid, size, buf);
+       return &buf->data;
 }
+EXPORT_SYMBOL_GPL(rpc_malloc);
 
 /**
  * rpc_free - free buffer allocated via rpc_malloc
- * @task: RPC task with a buffer to be freed
+ * @buffer: buffer to free
  *
  */
-void rpc_free(struct rpc_task *task)
+void rpc_free(void *buffer)
 {
-       struct rpc_rqst *req = task->tk_rqstp;
+       size_t size;
+       struct rpc_buffer *buf;
 
-       if (req->rq_buffer) {
-               if (req->rq_bufsize == RPC_BUFFER_MAXSIZE)
-                       mempool_free(req->rq_buffer, rpc_buffer_mempool);
-               else
-                       kfree(req->rq_buffer);
-               req->rq_buffer = NULL;
-               req->rq_bufsize = 0;
-       }
+       if (!buffer)
+               return;
+
+       buf = container_of(buffer, struct rpc_buffer, data);
+       size = buf->len;
+
+       dprintk("RPC:       freeing buffer of size %zu at %p\n",
+                       size, buf);
+
+       if (size <= RPC_BUFFER_MAXSIZE)
+               mempool_free(buf, rpc_buffer_mempool);
+       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;
+       setup_timer(&task->tk_timer, (void (*)(unsigned long))rpc_run_timer,
+                       (unsigned long)task);
        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;
 
-       if (clnt) {
-               atomic_inc(&clnt->cl_users);
-               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) {
+               memcpy(&task->tk_msg, task_setup_data->rpc_message, sizeof(task->tk_msg));
+               /* Bind the user cred */
+               if (task->tk_msg.rpc_cred != NULL)
+                       rpcauth_holdcred(task);
+               else
+                       rpcauth_bindcred(task);
+               if (task->tk_action == NULL)
+                       rpc_call_start(task);
+       }
 
        /* starting timestamp */
        task->tk_start = jiffies;
 
-       dprintk("RPC: %4d new task procpid %d\n", task->tk_pid,
-                               current->pid);
+       dprintk("RPC:       new task initialized, procpid %u\n",
+                               task_pid_nr(current));
 }
 
 static struct rpc_task *
@@ -838,39 +871,31 @@ rpc_alloc_task(void)
 static void rpc_free_task(struct rcu_head *rcu)
 {
        struct rpc_task *task = container_of(rcu, struct rpc_task, u.tk_rcu);
-       dprintk("RPC: %4d freeing task\n", task->tk_pid);
+       dprintk("RPC: %5u freeing task\n", task->tk_pid);
        mempool_free(task, rpc_task_mempool);
 }
 
 /*
- * Create a new task for the specified client.  We have to
- * clean up after an allocation failure, as the client may
- * have specified "oneshot".
+ * 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 cleanup;
+       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);
 
-       dprintk("RPC: %4d allocated task\n", task->tk_pid);
-       task->tk_flags |= RPC_TASK_DYNAMIC;
+       task->tk_flags |= flags;
+       dprintk("RPC:       allocated task %p\n", task);
 out:
        return task;
-
-cleanup:
-       /* Check whether to release the client */
-       if (clnt) {
-               printk("rpc_new_task: failed, users=%d, oneshot=%d\n",
-                       atomic_read(&clnt->cl_users), clnt->cl_oneshot);
-               atomic_inc(&clnt->cl_users); /* pretend we were used ... */
-               rpc_release_client(clnt);
-       }
-       goto out;
 }
 
 
@@ -894,20 +919,22 @@ void rpc_put_task(struct rpc_task *task)
                call_rcu_bh(&task->u.tk_rcu, rpc_free_task);
        rpc_release_calldata(tk_ops, calldata);
 }
-EXPORT_SYMBOL(rpc_put_task);
+EXPORT_SYMBOL_GPL(rpc_put_task);
 
-void rpc_release_task(struct rpc_task *task)
+static void rpc_release_task(struct rpc_task *task)
 {
 #ifdef RPC_DEBUG
        BUG_ON(task->tk_magic != RPC_TASK_MAGIC_ID);
 #endif
-       dprintk("RPC: %4d release task\n", task->tk_pid);
-
-       /* Remove from global task list */
-       spin_lock(&rpc_sched_lock);
-       list_del(&task->tk_task);
-       spin_unlock(&rpc_sched_lock);
-
+       dprintk("RPC: %5u release task\n", task->tk_pid);
+
+       if (!list_empty(&task->tk_task)) {
+               struct rpc_clnt *clnt = task->tk_client;
+               /* Remove from client task list */
+               spin_lock(&clnt->cl_lock);
+               list_del(&task->tk_task);
+               spin_unlock(&clnt->cl_lock);
+       }
        BUG_ON (RPC_IS_QUEUED(task));
 
        /* Synchronously delete any running timer */
@@ -922,29 +949,6 @@ 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?
@@ -952,142 +956,71 @@ EXPORT_SYMBOL(rpc_run_task);
 void rpc_killall_tasks(struct rpc_clnt *clnt)
 {
        struct rpc_task *rovr;
-       struct list_head *le;
 
-       dprintk("RPC:      killing all tasks for client %p\n", clnt);
 
+       if (list_empty(&clnt->cl_tasks))
+               return;
+       dprintk("RPC:       killing all tasks for client %p\n", clnt);
        /*
         * Spin lock all_tasks to prevent changes...
         */
-       spin_lock(&rpc_sched_lock);
-       alltask_for_each(rovr, le, &all_tasks) {
+       spin_lock(&clnt->cl_lock);
+       list_for_each_entry(rovr, &clnt->cl_tasks, tk_task) {
                if (! RPC_IS_ACTIVATED(rovr))
                        continue;
-               if (!clnt || rovr->tk_client == clnt) {
+               if (!(rovr->tk_flags & RPC_TASK_KILLED)) {
                        rovr->tk_flags |= RPC_TASK_KILLED;
                        rpc_exit(rovr, -EIO);
                        rpc_wake_up_task(rovr);
                }
        }
-       spin_unlock(&rpc_sched_lock);
+       spin_unlock(&clnt->cl_lock);
 }
+EXPORT_SYMBOL_GPL(rpc_killall_tasks);
 
-static DECLARE_MUTEX_LOCKED(rpciod_running);
-
-static void rpciod_killall(void)
+int rpciod_up(void)
 {
-       unsigned long flags;
-
-       while (!list_empty(&all_tasks)) {
-               clear_thread_flag(TIF_SIGPENDING);
-               rpc_killall_tasks(NULL);
-               flush_workqueue(rpciod_workqueue);
-               if (!list_empty(&all_tasks)) {
-                       dprintk("rpciod_killall: waiting for tasks to exit\n");
-                       yield();
-               }
-       }
+       return try_module_get(THIS_MODULE) ? 0 : -EINVAL;
+}
 
-       spin_lock_irqsave(&current->sighand->siglock, flags);
-       recalc_sigpending();
-       spin_unlock_irqrestore(&current->sighand->siglock, flags);
+void rpciod_down(void)
+{
+       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;
 
-       mutex_lock(&rpciod_mutex);
-       dprintk("rpciod_up: users %d\n", rpciod_users);
-       rpciod_users++;
-       if (rpciod_workqueue)
-               goto out;
-       /*
-        * If there's no pid, we should be the first user.
-        */
-       if (rpciod_users > 1)
-               printk(KERN_WARNING "rpciod_up: no workqueue, %d users??\n", rpciod_users);
        /*
         * Create the rpciod thread and wait for it to start.
         */
-       error = -ENOMEM;
+       dprintk("RPC:       creating workqueue rpciod\n");
        wq = create_workqueue("rpciod");
-       if (wq == NULL) {
-               printk(KERN_WARNING "rpciod_up: create workqueue failed, error=%d\n", error);
-               rpciod_users--;
-               goto out;
-       }
        rpciod_workqueue = wq;
-       error = 0;
-out:
-       mutex_unlock(&rpciod_mutex);
-       return error;
+       return rpciod_workqueue != NULL;
 }
 
-void
-rpciod_down(void)
+static void rpciod_stop(void)
 {
-       mutex_lock(&rpciod_mutex);
-       dprintk("rpciod_down sema %d\n", rpciod_users);
-       if (rpciod_users) {
-               if (--rpciod_users)
-                       goto out;
-       } else
-               printk(KERN_WARNING "rpciod_down: no users??\n");
+       struct workqueue_struct *wq = NULL;
 
-       if (!rpciod_workqueue) {
-               dprintk("rpciod_down: Nothing to do!\n");
-               goto out;
-       }
-       rpciod_killall();
+       if (rpciod_workqueue == NULL)
+               return;
+       dprintk("RPC:       destroying workqueue rpciod\n");
 
-       destroy_workqueue(rpciod_workqueue);
+       wq = rpciod_workqueue;
        rpciod_workqueue = NULL;
- out:
-       mutex_unlock(&rpciod_mutex);
+       destroy_workqueue(wq);
 }
 
-#ifdef RPC_DEBUG
-void rpc_show_tasks(void)
-{
-       struct list_head *le;
-       struct rpc_task *t;
-
-       spin_lock(&rpc_sched_lock);
-       if (list_empty(&all_tasks)) {
-               spin_unlock(&rpc_sched_lock);
-               return;
-       }
-       printk("-pid- proc flgs status -client- -prog- --rqstp- -timeout "
-               "-rpcwait -action- ---ops--\n");
-       alltask_for_each(t, le, &all_tasks) {
-               const char *rpc_waitq = "none";
-
-               if (RPC_IS_QUEUED(t))
-                       rpc_waitq = rpc_qname(t->u.tk_wait.rpc_waitq);
-
-               printk("%05d %04d %04x %06d %8p %6d %8p %08ld %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(&rpc_sched_lock);
-}
-#endif
-
 void
 rpc_destroy_mempool(void)
 {
+       rpciod_stop();
        if (rpc_buffer_mempool)
                mempool_destroy(rpc_buffer_mempool);
        if (rpc_task_mempool)
@@ -1104,13 +1037,13 @@ rpc_init_mempool(void)
        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,
@@ -1121,6 +1054,13 @@ rpc_init_mempool(void)
                                                      rpc_buffer_slabp);
        if (!rpc_buffer_mempool)
                goto err_nomem;
+       if (!rpciod_start())
+               goto err_nomem;
+       /*
+        * The following is not strictly a mempool initialisation,
+        * but there is no harm in doing it here
+        */
+       rpc_init_wait_queue(&delay_queue, "delayq");
        return 0;
 err_nomem:
        rpc_destroy_mempool();