net/sunrpc/sched.c

   1 /*
   2  * linux/net/sunrpc/sched.c
   3  *
   4  * Scheduling for synchronous and asynchronous RPC requests.
   5  *
   6  * Copyright (C) 1996 Olaf Kirch, <okir@monad.swb.de>
   7  *
   8  * TCP NFS related read + write fixes
   9  * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
  10  */
  11
  12 #include <linux/module.h>
  13
  14 #include <linux/sched.h>
  15 #include <linux/interrupt.h>
  16 #include <linux/slab.h>
  17 #include <linux/mempool.h>
  18 #include <linux/smp.h>
  19 #include <linux/smp_lock.h>
  20 #include <linux/spinlock.h>
  21 #include <linux/mutex.h>
  22
  23 #include <linux/sunrpc/clnt.h>
  24
  25 #ifdef RPC_DEBUG
  26 #define RPCDBG_FACILITY         RPCDBG_SCHED
  27 #define RPC_TASK_MAGIC_ID       0xf00baa
  28 static int                      rpc_task_id;
  29 #endif
  30
  31 /*
  32  * RPC slabs and memory pools
  33  */
  34 #define RPC_BUFFER_MAXSIZE      (2048)
  35 #define RPC_BUFFER_POOLSIZE     (8)
  36 #define RPC_TASK_POOLSIZE       (8)
  37 static struct kmem_cache        *rpc_task_slabp __read_mostly;
  38 static struct kmem_cache        *rpc_buffer_slabp __read_mostly;
  39 static mempool_t        *rpc_task_mempool __read_mostly;
  40 static mempool_t        *rpc_buffer_mempool __read_mostly;
  41
  42 static void                     __rpc_default_timer(struct rpc_task *task);
  43 static void                     rpciod_killall(void);
  44 static void                     rpc_async_schedule(struct work_struct *);
  45 static void                      rpc_release_task(struct rpc_task *task);
  46
  47 /*
  48  * RPC tasks sit here while waiting for conditions to improve.
  49  */
  50 static RPC_WAITQ(delay_queue, "delayq");
  51
  52 /*
  53  * All RPC tasks are linked into this list
  54  */
  55 static LIST_HEAD(all_tasks);
  56
  57 /*
  58  * rpciod-related stuff
  59  */
  60 static DEFINE_MUTEX(rpciod_mutex);
  61 static unsigned int             rpciod_users;
  62 struct workqueue_struct *rpciod_workqueue;
  63
  64 /*
  65  * Spinlock for other critical sections of code.
  66  */
  67 static DEFINE_SPINLOCK(rpc_sched_lock);
  68
  69 /*
  70  * Disable the timer for a given RPC task. Should be called with
  71  * queue->lock and bh_disabled in order to avoid races within
  72  * rpc_run_timer().
  73  */
  74 static inline void
  75 __rpc_disable_timer(struct rpc_task *task)
  76 {
  77         dprintk("RPC: %5u disabling timer\n", task->tk_pid);
  78         task->tk_timeout_fn = NULL;
  79         task->tk_timeout = 0;
  80 }
  81
  82 /*
  83  * Run a timeout function.
  84  * We use the callback in order to allow __rpc_wake_up_task()
  85  * and friends to disable the timer synchronously on SMP systems
  86  * without calling del_timer_sync(). The latter could cause a
  87  * deadlock if called while we're holding spinlocks...
  88  */
  89 static void rpc_run_timer(struct rpc_task *task)
  90 {
  91         void (*callback)(struct rpc_task *);
  92
  93         callback = task->tk_timeout_fn;
  94         task->tk_timeout_fn = NULL;
  95         if (callback && RPC_IS_QUEUED(task)) {
  96                 dprintk("RPC: %5u running timer\n", task->tk_pid);
  97                 callback(task);
  98         }
  99         smp_mb__before_clear_bit();
 100         clear_bit(RPC_TASK_HAS_TIMER, &task->tk_runstate);
 101         smp_mb__after_clear_bit();
 102 }
 103
 104 /*
 105  * Set up a timer for the current task.
 106  */
 107 static inline void
 108 __rpc_add_timer(struct rpc_task *task, rpc_action timer)
 109 {
 110         if (!task->tk_timeout)
 111                 return;
 112
 113         dprintk("RPC: %5u setting alarm for %lu ms\n",
 114                         task->tk_pid, task->tk_timeout * 1000 / HZ);
 115
 116         if (timer)
 117                 task->tk_timeout_fn = timer;
 118         else
 119                 task->tk_timeout_fn = __rpc_default_timer;
 120         set_bit(RPC_TASK_HAS_TIMER, &task->tk_runstate);
 121         mod_timer(&task->tk_timer, jiffies + task->tk_timeout);
 122 }
 123
 124 /*
 125  * Delete any timer for the current task. Because we use del_timer_sync(),
 126  * this function should never be called while holding queue->lock.
 127  */
 128 static void
 129 rpc_delete_timer(struct rpc_task *task)
 130 {
 131         if (RPC_IS_QUEUED(task))
 132                 return;
 133         if (test_and_clear_bit(RPC_TASK_HAS_TIMER, &task->tk_runstate)) {
 134                 del_singleshot_timer_sync(&task->tk_timer);
 135                 dprintk("RPC: %5u deleting timer\n", task->tk_pid);
 136         }
 137 }
 138
 139 /*
 140  * Add new request to a priority queue.
 141  */
 142 static void __rpc_add_wait_queue_priority(struct rpc_wait_queue *queue, struct rpc_task *task)
 143 {
 144         struct list_head *q;
 145         struct rpc_task *t;
 146
 147         INIT_LIST_HEAD(&task->u.tk_wait.links);
 148         q = &queue->tasks[task->tk_priority];
 149         if (unlikely(task->tk_priority > queue->maxpriority))
 150                 q = &queue->tasks[queue->maxpriority];
 151         list_for_each_entry(t, q, u.tk_wait.list) {
 152                 if (t->tk_cookie == task->tk_cookie) {
 153                         list_add_tail(&task->u.tk_wait.list, &t->u.tk_wait.links);
 154                         return;
 155                 }
 156         }
 157         list_add_tail(&task->u.tk_wait.list, q);
 158 }
 159
 160 /*
 161  * Add new request to wait queue.
 162  *
 163  * Swapper tasks always get inserted at the head of the queue.
 164  * This should avoid many nasty memory deadlocks and hopefully
 165  * improve overall performance.
 166  * Everyone else gets appended to the queue to ensure proper FIFO behavior.
 167  */
 168 static void __rpc_add_wait_queue(struct rpc_wait_queue *queue, struct rpc_task *task)
 169 {
 170         BUG_ON (RPC_IS_QUEUED(task));
 171
 172         if (RPC_IS_PRIORITY(queue))
 173                 __rpc_add_wait_queue_priority(queue, task);
 174         else if (RPC_IS_SWAPPER(task))
 175                 list_add(&task->u.tk_wait.list, &queue->tasks[0]);
 176         else
 177                 list_add_tail(&task->u.tk_wait.list, &queue->tasks[0]);
 178         task->u.tk_wait.rpc_waitq = queue;
 179         queue->qlen++;
 180         rpc_set_queued(task);
 181
 182         dprintk("RPC: %5u added to queue %p \"%s\"\n",
 183                         task->tk_pid, queue, rpc_qname(queue));
 184 }
 185
 186 /*
 187  * Remove request from a priority queue.
 188  */
 189 static void __rpc_remove_wait_queue_priority(struct rpc_task *task)
 190 {
 191         struct rpc_task *t;
 192
 193         if (!list_empty(&task->u.tk_wait.links)) {
 194                 t = list_entry(task->u.tk_wait.links.next, struct rpc_task, u.tk_wait.list);
 195                 list_move(&t->u.tk_wait.list, &task->u.tk_wait.list);
 196                 list_splice_init(&task->u.tk_wait.links, &t->u.tk_wait.links);
 197         }
 198         list_del(&task->u.tk_wait.list);
 199 }
 200
 201 /*
 202  * Remove request from queue.
 203  * Note: must be called with spin lock held.
 204  */
 205 static void __rpc_remove_wait_queue(struct rpc_task *task)
 206 {
 207         struct rpc_wait_queue *queue;
 208         queue = task->u.tk_wait.rpc_waitq;
 209
 210         if (RPC_IS_PRIORITY(queue))
 211                 __rpc_remove_wait_queue_priority(task);
 212         else
 213                 list_del(&task->u.tk_wait.list);
 214         queue->qlen--;
 215         dprintk("RPC: %5u removed from queue %p \"%s\"\n",
 216                         task->tk_pid, queue, rpc_qname(queue));
 217 }
 218
 219 static inline void rpc_set_waitqueue_priority(struct rpc_wait_queue *queue, int priority)
 220 {
 221         queue->priority = priority;
 222         queue->count = 1 << (priority * 2);
 223 }
 224
 225 static inline void rpc_set_waitqueue_cookie(struct rpc_wait_queue *queue, unsigned long cookie)
 226 {
 227         queue->cookie = cookie;
 228         queue->nr = RPC_BATCH_COUNT;
 229 }
 230
 231 static inline void rpc_reset_waitqueue_priority(struct rpc_wait_queue *queue)
 232 {
 233         rpc_set_waitqueue_priority(queue, queue->maxpriority);
 234         rpc_set_waitqueue_cookie(queue, 0);
 235 }
 236
 237 static void __rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname, int maxprio)
 238 {
 239         int i;
 240
 241         spin_lock_init(&queue->lock);
 242         for (i = 0; i < ARRAY_SIZE(queue->tasks); i++)
 243                 INIT_LIST_HEAD(&queue->tasks[i]);
 244         queue->maxpriority = maxprio;
 245         rpc_reset_waitqueue_priority(queue);
 246 #ifdef RPC_DEBUG
 247         queue->name = qname;
 248 #endif
 249 }
 250
 251 void rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname)
 252 {
 253         __rpc_init_priority_wait_queue(queue, qname, RPC_PRIORITY_HIGH);
 254 }
 255
 256 void rpc_init_wait_queue(struct rpc_wait_queue *queue, const char *qname)
 257 {
 258         __rpc_init_priority_wait_queue(queue, qname, 0);
 259 }
 260 EXPORT_SYMBOL(rpc_init_wait_queue);
 261
 262 static int rpc_wait_bit_interruptible(void *word)
 263 {
 264         if (signal_pending(current))
 265                 return -ERESTARTSYS;
 266         schedule();
 267         return 0;
 268 }
 269
 270 static void rpc_set_active(struct rpc_task *task)
 271 {
 272         if (test_and_set_bit(RPC_TASK_ACTIVE, &task->tk_runstate) != 0)
 273                 return;
 274         spin_lock(&rpc_sched_lock);
 275 #ifdef RPC_DEBUG
 276         task->tk_magic = RPC_TASK_MAGIC_ID;
 277         task->tk_pid = rpc_task_id++;
 278 #endif
 279         /* Add to global list of all tasks */
 280         list_add_tail(&task->tk_task, &all_tasks);
 281         spin_unlock(&rpc_sched_lock);
 282 }
 283
 284 /*
 285  * Mark an RPC call as having completed by clearing the 'active' bit
 286  */
 287 static void rpc_mark_complete_task(struct rpc_task *task)
 288 {
 289         smp_mb__before_clear_bit();
 290         clear_bit(RPC_TASK_ACTIVE, &task->tk_runstate);
 291         smp_mb__after_clear_bit();
 292         wake_up_bit(&task->tk_runstate, RPC_TASK_ACTIVE);
 293 }
 294
 295 /*
 296  * Allow callers to wait for completion of an RPC call
 297  */
 298 int __rpc_wait_for_completion_task(struct rpc_task *task, int (*action)(void *))
 299 {
 300         if (action == NULL)
 301                 action = rpc_wait_bit_interruptible;
 302         return wait_on_bit(&task->tk_runstate, RPC_TASK_ACTIVE,
 303                         action, TASK_INTERRUPTIBLE);
 304 }
 305 EXPORT_SYMBOL(__rpc_wait_for_completion_task);
 306
 307 /*
 308  * Make an RPC task runnable.
 309  *
 310  * Note: If the task is ASYNC, this must be called with
 311  * the spinlock held to protect the wait queue operation.
 312  */
 313 static void rpc_make_runnable(struct rpc_task *task)
 314 {
 315         BUG_ON(task->tk_timeout_fn);
 316         rpc_clear_queued(task);
 317         if (rpc_test_and_set_running(task))
 318                 return;
 319         /* We might have raced */
 320         if (RPC_IS_QUEUED(task)) {
 321                 rpc_clear_running(task);
 322                 return;
 323         }
 324         if (RPC_IS_ASYNC(task)) {
 325                 int status;
 326
 327                 INIT_WORK(&task->u.tk_work, rpc_async_schedule);
 328                 status = queue_work(task->tk_workqueue, &task->u.tk_work);
 329                 if (status < 0) {
 330                         printk(KERN_WARNING "RPC: failed to add task to queue: error: %d!\n", status);
 331                         task->tk_status = status;
 332                         return;
 333                 }
 334         } else
 335                 wake_up_bit(&task->tk_runstate, RPC_TASK_QUEUED);
 336 }
 337
 338 /*
 339  * Prepare for sleeping on a wait queue.
 340  * By always appending tasks to the list we ensure FIFO behavior.
 341  * NB: An RPC task will only receive interrupt-driven events as long
 342  * as it's on a wait queue.
 343  */
 344 static void __rpc_sleep_on(struct rpc_wait_queue *q, struct rpc_task *task,
 345                         rpc_action action, rpc_action timer)
 346 {
 347         dprintk("RPC: %5u sleep_on(queue \"%s\" time %lu)\n",
 348                         task->tk_pid, rpc_qname(q), jiffies);
 349
 350         if (!RPC_IS_ASYNC(task) && !RPC_IS_ACTIVATED(task)) {
 351                 printk(KERN_ERR "RPC: Inactive synchronous task put to sleep!\n");
 352                 return;
 353         }
 354
 355         __rpc_add_wait_queue(q, task);
 356
 357         BUG_ON(task->tk_callback != NULL);
 358         task->tk_callback = action;
 359         __rpc_add_timer(task, timer);
 360 }
 361
 362 void rpc_sleep_on(struct rpc_wait_queue *q, struct rpc_task *task,
 363                                 rpc_action action, rpc_action timer)
 364 {
 365         /* Mark the task as being activated if so needed */
 366         rpc_set_active(task);
 367
 368         /*
 369          * Protect the queue operations.
 370          */
 371         spin_lock_bh(&q->lock);
 372         __rpc_sleep_on(q, task, action, timer);
 373         spin_unlock_bh(&q->lock);
 374 }
 375
 376 /**
 377  * __rpc_do_wake_up_task - wake up a single rpc_task
 378  * @task: task to be woken up
 379  *
 380  * Caller must hold queue->lock, and have cleared the task queued flag.
 381  */
 382 static void __rpc_do_wake_up_task(struct rpc_task *task)
 383 {
 384         dprintk("RPC: %5u __rpc_wake_up_task (now %lu)\n",
 385                         task->tk_pid, jiffies);
 386
 387 #ifdef RPC_DEBUG
 388         BUG_ON(task->tk_magic != RPC_TASK_MAGIC_ID);
 389 #endif
 390         /* Has the task been executed yet? If not, we cannot wake it up! */
 391         if (!RPC_IS_ACTIVATED(task)) {
 392                 printk(KERN_ERR "RPC: Inactive task (%p) being woken up!\n", task);
 393                 return;
 394         }
 395
 396         __rpc_disable_timer(task);
 397         __rpc_remove_wait_queue(task);
 398
 399         rpc_make_runnable(task);
 400
 401         dprintk("RPC:       __rpc_wake_up_task done\n");
 402 }
 403
 404 /*
 405  * Wake up the specified task
 406  */
 407 static void __rpc_wake_up_task(struct rpc_task *task)
 408 {
 409         if (rpc_start_wakeup(task)) {
 410                 if (RPC_IS_QUEUED(task))
 411                         __rpc_do_wake_up_task(task);
 412                 rpc_finish_wakeup(task);
 413         }
 414 }
 415
 416 /*
 417  * Default timeout handler if none specified by user
 418  */
 419 static void
 420 __rpc_default_timer(struct rpc_task *task)
 421 {
 422         dprintk("RPC: %5u timeout (default timer)\n", task->tk_pid);
 423         task->tk_status = -ETIMEDOUT;
 424         rpc_wake_up_task(task);
 425 }
 426
 427 /*
 428  * Wake up the specified task
 429  */
 430 void rpc_wake_up_task(struct rpc_task *task)
 431 {
 432         rcu_read_lock_bh();
 433         if (rpc_start_wakeup(task)) {
 434                 if (RPC_IS_QUEUED(task)) {
 435                         struct rpc_wait_queue *queue = task->u.tk_wait.rpc_waitq;
 436
 437                         /* Note: we're already in a bh-safe context */
 438                         spin_lock(&queue->lock);
 439                         __rpc_do_wake_up_task(task);
 440                         spin_unlock(&queue->lock);
 441                 }
 442                 rpc_finish_wakeup(task);
 443         }
 444         rcu_read_unlock_bh();
 445 }
 446
 447 /*
 448  * Wake up the next task on a priority queue.
 449  */
 450 static struct rpc_task * __rpc_wake_up_next_priority(struct rpc_wait_queue *queue)
 451 {
 452         struct list_head *q;
 453         struct rpc_task *task;
 454
 455         /*
 456          * Service a batch of tasks from a single cookie.
 457          */
 458         q = &queue->tasks[queue->priority];
 459         if (!list_empty(q)) {
 460                 task = list_entry(q->next, struct rpc_task, u.tk_wait.list);
 461                 if (queue->cookie == task->tk_cookie) {
 462                         if (--queue->nr)
 463                                 goto out;
 464                         list_move_tail(&task->u.tk_wait.list, q);
 465                 }
 466                 /*
 467                  * Check if we need to switch queues.
 468                  */
 469                 if (--queue->count)
 470                         goto new_cookie;
 471         }
 472
 473         /*
 474          * Service the next queue.
 475          */
 476         do {
 477                 if (q == &queue->tasks[0])
 478                         q = &queue->tasks[queue->maxpriority];
 479                 else
 480                         q = q - 1;
 481                 if (!list_empty(q)) {
 482                         task = list_entry(q->next, struct rpc_task, u.tk_wait.list);
 483                         goto new_queue;
 484                 }
 485         } while (q != &queue->tasks[queue->priority]);
 486
 487         rpc_reset_waitqueue_priority(queue);
 488         return NULL;
 489
 490 new_queue:
 491         rpc_set_waitqueue_priority(queue, (unsigned int)(q - &queue->tasks[0]));
 492 new_cookie:
 493         rpc_set_waitqueue_cookie(queue, task->tk_cookie);
 494 out:
 495         __rpc_wake_up_task(task);
 496         return task;
 497 }
 498
 499 /*
 500  * Wake up the next task on the wait queue.
 501  */
 502 struct rpc_task * rpc_wake_up_next(struct rpc_wait_queue *queue)
 503 {
 504         struct rpc_task *task = NULL;
 505
 506         dprintk("RPC:       wake_up_next(%p \"%s\")\n",
 507                         queue, rpc_qname(queue));
 508         rcu_read_lock_bh();
 509         spin_lock(&queue->lock);
 510         if (RPC_IS_PRIORITY(queue))
 511                 task = __rpc_wake_up_next_priority(queue);
 512         else {
 513                 task_for_first(task, &queue->tasks[0])
 514                         __rpc_wake_up_task(task);
 515         }
 516         spin_unlock(&queue->lock);
 517         rcu_read_unlock_bh();
 518
 519         return task;
 520 }
 521
 522 /**
 523  * rpc_wake_up - wake up all rpc_tasks
 524  * @queue: rpc_wait_queue on which the tasks are sleeping
 525  *
 526  * Grabs queue->lock
 527  */
 528 void rpc_wake_up(struct rpc_wait_queue *queue)
 529 {
 530         struct rpc_task *task, *next;
 531         struct list_head *head;
 532
 533         rcu_read_lock_bh();
 534         spin_lock(&queue->lock);
 535         head = &queue->tasks[queue->maxpriority];
 536         for (;;) {
 537                 list_for_each_entry_safe(task, next, head, u.tk_wait.list)
 538                         __rpc_wake_up_task(task);
 539                 if (head == &queue->tasks[0])
 540                         break;
 541                 head--;
 542         }
 543         spin_unlock(&queue->lock);
 544         rcu_read_unlock_bh();
 545 }
 546
 547 /**
 548  * rpc_wake_up_status - wake up all rpc_tasks and set their status value.
 549  * @queue: rpc_wait_queue on which the tasks are sleeping
 550  * @status: status value to set
 551  *
 552  * Grabs queue->lock
 553  */
 554 void rpc_wake_up_status(struct rpc_wait_queue *queue, int status)
 555 {
 556         struct rpc_task *task, *next;
 557         struct list_head *head;
 558
 559         rcu_read_lock_bh();
 560         spin_lock(&queue->lock);
 561         head = &queue->tasks[queue->maxpriority];
 562         for (;;) {
 563                 list_for_each_entry_safe(task, next, head, u.tk_wait.list) {
 564                         task->tk_status = status;
 565                         __rpc_wake_up_task(task);
 566                 }
 567                 if (head == &queue->tasks[0])
 568                         break;
 569                 head--;
 570         }
 571         spin_unlock(&queue->lock);
 572         rcu_read_unlock_bh();
 573 }
 574
 575 static void __rpc_atrun(struct rpc_task *task)
 576 {
 577         rpc_wake_up_task(task);
 578 }
 579
 580 /*
 581  * Run a task at a later time
 582  */
 583 void rpc_delay(struct rpc_task *task, unsigned long delay)
 584 {
 585         task->tk_timeout = delay;
 586         rpc_sleep_on(&delay_queue, task, NULL, __rpc_atrun);
 587 }
 588
 589 /*
 590  * Helper to call task->tk_ops->rpc_call_prepare
 591  */
 592 static void rpc_prepare_task(struct rpc_task *task)
 593 {
 594         lock_kernel();
 595         task->tk_ops->rpc_call_prepare(task, task->tk_calldata);
 596         unlock_kernel();
 597 }
 598
 599 /*
 600  * Helper that calls task->tk_ops->rpc_call_done if it exists
 601  */
 602 void rpc_exit_task(struct rpc_task *task)
 603 {
 604         task->tk_action = NULL;
 605         if (task->tk_ops->rpc_call_done != NULL) {
 606                 lock_kernel();
 607                 task->tk_ops->rpc_call_done(task, task->tk_calldata);
 608                 unlock_kernel();
 609                 if (task->tk_action != NULL) {
 610                         WARN_ON(RPC_ASSASSINATED(task));
 611                         /* Always release the RPC slot and buffer memory */
 612                         xprt_release(task);
 613                 }
 614         }
 615 }
 616 EXPORT_SYMBOL(rpc_exit_task);
 617
 618 void rpc_release_calldata(const struct rpc_call_ops *ops, void *calldata)
 619 {
 620         if (ops->rpc_release != NULL) {
 621                 lock_kernel();
 622                 ops->rpc_release(calldata);
 623                 unlock_kernel();
 624         }
 625 }
 626
 627 /*
 628  * This is the RPC `scheduler' (or rather, the finite state machine).
 629  */
 630 static void __rpc_execute(struct rpc_task *task)
 631 {
 632         int             status = 0;
 633
 634         dprintk("RPC: %5u __rpc_execute flags=0x%x\n",
 635                         task->tk_pid, task->tk_flags);
 636
 637         BUG_ON(RPC_IS_QUEUED(task));
 638
 639         for (;;) {
 640                 /*
 641                  * Garbage collection of pending timers...
 642                  */
 643                 rpc_delete_timer(task);
 644
 645                 /*
 646                  * Execute any pending callback.
 647                  */
 648                 if (RPC_DO_CALLBACK(task)) {
 649                         /* Define a callback save pointer */
 650                         void (*save_callback)(struct rpc_task *);
 651
 652                         /*
 653                          * If a callback exists, save it, reset it,
 654                          * call it.
 655                          * The save is needed to stop from resetting
 656                          * another callback set within the callback handler
 657                          * - Dave
 658                          */
 659                         save_callback=task->tk_callback;
 660                         task->tk_callback=NULL;
 661                         save_callback(task);
 662                 }
 663
 664                 /*
 665                  * Perform the next FSM step.
 666                  * tk_action may be NULL when the task has been killed
 667                  * by someone else.
 668                  */
 669                 if (!RPC_IS_QUEUED(task)) {
 670                         if (task->tk_action == NULL)
 671                                 break;
 672                         task->tk_action(task);
 673                 }
 674
 675                 /*
 676                  * Lockless check for whether task is sleeping or not.
 677                  */
 678                 if (!RPC_IS_QUEUED(task))
 679                         continue;
 680                 rpc_clear_running(task);
 681                 if (RPC_IS_ASYNC(task)) {
 682                         /* Careful! we may have raced... */
 683                         if (RPC_IS_QUEUED(task))
 684                                 return;
 685                         if (rpc_test_and_set_running(task))
 686                                 return;
 687                         continue;
 688                 }
 689
 690                 /* sync task: sleep here */
 691                 dprintk("RPC: %5u sync task going to sleep\n", task->tk_pid);
 692                 /* Note: Caller should be using rpc_clnt_sigmask() */
 693                 status = out_of_line_wait_on_bit(&task->tk_runstate,
 694                                 RPC_TASK_QUEUED, rpc_wait_bit_interruptible,
 695                                 TASK_INTERRUPTIBLE);
 696                 if (status == -ERESTARTSYS) {
 697                         /*
 698                          * When a sync task receives a signal, it exits with
 699                          * -ERESTARTSYS. In order to catch any callbacks that
 700                          * clean up after sleeping on some queue, we don't
 701                          * break the loop here, but go around once more.
 702                          */
 703                         dprintk("RPC: %5u got signal\n", task->tk_pid);
 704                         task->tk_flags |= RPC_TASK_KILLED;
 705                         rpc_exit(task, -ERESTARTSYS);
 706                         rpc_wake_up_task(task);
 707                 }
 708                 rpc_set_running(task);
 709                 dprintk("RPC: %5u sync task resuming\n", task->tk_pid);
 710         }
 711
 712         dprintk("RPC: %5u return %d, status %d\n", task->tk_pid, status,
 713                         task->tk_status);
 714         /* Release all resources associated with the task */
 715         rpc_release_task(task);
 716 }
 717
 718 /*
 719  * User-visible entry point to the scheduler.
 720  *
 721  * This may be called recursively if e.g. an async NFS task updates
 722  * the attributes and finds that dirty pages must be flushed.
 723  * NOTE: Upon exit of this function the task is guaranteed to be
 724  *       released. In particular note that tk_release() will have
 725  *       been called, so your task memory may have been freed.
 726  */
 727 void rpc_execute(struct rpc_task *task)
 728 {
 729         rpc_set_active(task);
 730         rpc_set_running(task);
 731         __rpc_execute(task);
 732 }
 733
 734 static void rpc_async_schedule(struct work_struct *work)
 735 {
 736         __rpc_execute(container_of(work, struct rpc_task, u.tk_work));
 737 }
 738
 739 struct rpc_buffer {
 740         size_t  len;
 741         char    data[];
 742 };
 743
 744 /**
 745  * rpc_malloc - allocate an RPC buffer
 746  * @task: RPC task that will use this buffer
 747  * @size: requested byte size
 748  *
 749  * To prevent rpciod from hanging, this allocator never sleeps,
 750  * returning NULL if the request cannot be serviced immediately.
 751  * The caller can arrange to sleep in a way that is safe for rpciod.
 752  *
 753  * Most requests are 'small' (under 2KiB) and can be serviced from a
 754  * mempool, ensuring that NFS reads and writes can always proceed,
 755  * and that there is good locality of reference for these buffers.
 756  *
 757  * In order to avoid memory starvation triggering more writebacks of
 758  * NFS requests, we avoid using GFP_KERNEL.
 759  */
 760 void *rpc_malloc(struct rpc_task *task, size_t size)
 761 {
 762         struct rpc_buffer *buf;
 763         gfp_t gfp = RPC_IS_SWAPPER(task) ? GFP_ATOMIC : GFP_NOWAIT;
 764
 765         size += sizeof(struct rpc_buffer);
 766         if (size <= RPC_BUFFER_MAXSIZE)
 767                 buf = mempool_alloc(rpc_buffer_mempool, gfp);
 768         else
 769                 buf = kmalloc(size, gfp);
 770
 771         if (!buf)
 772                 return NULL;
 773
 774         buf->len = size;
 775         dprintk("RPC: %5u allocated buffer of size %zu at %p\n",
 776                         task->tk_pid, size, buf);
 777         return &buf->data;
 778 }
 779
 780 /**
 781  * rpc_free - free buffer allocated via rpc_malloc
 782  * @buffer: buffer to free
 783  *
 784  */
 785 void rpc_free(void *buffer)
 786 {
 787         size_t size;
 788         struct rpc_buffer *buf;
 789
 790         if (!buffer)
 791                 return;
 792
 793         buf = container_of(buffer, struct rpc_buffer, data);
 794         size = buf->len;
 795
 796         dprintk("RPC:       freeing buffer of size %zu at %p\n",
 797                         size, buf);
 798
 799         if (size <= RPC_BUFFER_MAXSIZE)
 800                 mempool_free(buf, rpc_buffer_mempool);
 801         else
 802                 kfree(buf);
 803 }
 804
 805 /*
 806  * Creation and deletion of RPC task structures
 807  */
 808 void rpc_init_task(struct rpc_task *task, struct rpc_clnt *clnt, int flags, const struct rpc_call_ops *tk_ops, void *calldata)
 809 {
 810         memset(task, 0, sizeof(*task));
 811         init_timer(&task->tk_timer);
 812         task->tk_timer.data     = (unsigned long) task;
 813         task->tk_timer.function = (void (*)(unsigned long)) rpc_run_timer;
 814         atomic_set(&task->tk_count, 1);
 815         task->tk_client = clnt;
 816         task->tk_flags  = flags;
 817         task->tk_ops = tk_ops;
 818         if (tk_ops->rpc_call_prepare != NULL)
 819                 task->tk_action = rpc_prepare_task;
 820         task->tk_calldata = calldata;
 821
 822         /* Initialize retry counters */
 823         task->tk_garb_retry = 2;
 824         task->tk_cred_retry = 2;
 825
 826         task->tk_priority = RPC_PRIORITY_NORMAL;
 827         task->tk_cookie = (unsigned long)current;
 828
 829         /* Initialize workqueue for async tasks */
 830         task->tk_workqueue = rpciod_workqueue;
 831
 832         if (clnt) {
 833                 atomic_inc(&clnt->cl_users);
 834                 if (clnt->cl_softrtry)
 835                         task->tk_flags |= RPC_TASK_SOFT;
 836                 if (!clnt->cl_intr)
 837                         task->tk_flags |= RPC_TASK_NOINTR;
 838         }
 839
 840         BUG_ON(task->tk_ops == NULL);
 841
 842         /* starting timestamp */
 843         task->tk_start = jiffies;
 844
 845         dprintk("RPC:       new task initialized, procpid %u\n",
 846                                 current->pid);
 847 }
 848
 849 static struct rpc_task *
 850 rpc_alloc_task(void)
 851 {
 852         return (struct rpc_task *)mempool_alloc(rpc_task_mempool, GFP_NOFS);
 853 }
 854
 855 static void rpc_free_task(struct rcu_head *rcu)
 856 {
 857         struct rpc_task *task = container_of(rcu, struct rpc_task, u.tk_rcu);
 858         dprintk("RPC: %5u freeing task\n", task->tk_pid);
 859         mempool_free(task, rpc_task_mempool);
 860 }
 861
 862 /*
 863  * Create a new task for the specified client.  We have to
 864  * clean up after an allocation failure, as the client may
 865  * have specified "oneshot".
 866  */
 867 struct rpc_task *rpc_new_task(struct rpc_clnt *clnt, int flags, const struct rpc_call_ops *tk_ops, void *calldata)
 868 {
 869         struct rpc_task *task;
 870
 871         task = rpc_alloc_task();
 872         if (!task)
 873                 goto cleanup;
 874
 875         rpc_init_task(task, clnt, flags, tk_ops, calldata);
 876
 877         dprintk("RPC:       allocated task %p\n", task);
 878         task->tk_flags |= RPC_TASK_DYNAMIC;
 879 out:
 880         return task;
 881
 882 cleanup:
 883         /* Check whether to release the client */
 884         if (clnt) {
 885                 printk("rpc_new_task: failed, users=%d, oneshot=%d\n",
 886                         atomic_read(&clnt->cl_users), clnt->cl_oneshot);
 887                 atomic_inc(&clnt->cl_users); /* pretend we were used ... */
 888                 rpc_release_client(clnt);
 889         }
 890         goto out;
 891 }
 892
 893
 894 void rpc_put_task(struct rpc_task *task)
 895 {
 896         const struct rpc_call_ops *tk_ops = task->tk_ops;
 897         void *calldata = task->tk_calldata;
 898
 899         if (!atomic_dec_and_test(&task->tk_count))
 900                 return;
 901         /* Release resources */
 902         if (task->tk_rqstp)
 903                 xprt_release(task);
 904         if (task->tk_msg.rpc_cred)
 905                 rpcauth_unbindcred(task);
 906         if (task->tk_client) {
 907                 rpc_release_client(task->tk_client);
 908                 task->tk_client = NULL;
 909         }
 910         if (task->tk_flags & RPC_TASK_DYNAMIC)
 911                 call_rcu_bh(&task->u.tk_rcu, rpc_free_task);
 912         rpc_release_calldata(tk_ops, calldata);
 913 }
 914 EXPORT_SYMBOL(rpc_put_task);
 915
 916 static void rpc_release_task(struct rpc_task *task)
 917 {
 918 #ifdef RPC_DEBUG
 919         BUG_ON(task->tk_magic != RPC_TASK_MAGIC_ID);
 920 #endif
 921         dprintk("RPC: %5u release task\n", task->tk_pid);
 922
 923         /* Remove from global task list */
 924         spin_lock(&rpc_sched_lock);
 925         list_del(&task->tk_task);
 926         spin_unlock(&rpc_sched_lock);
 927
 928         BUG_ON (RPC_IS_QUEUED(task));
 929
 930         /* Synchronously delete any running timer */
 931         rpc_delete_timer(task);
 932
 933 #ifdef RPC_DEBUG
 934         task->tk_magic = 0;
 935 #endif
 936         /* Wake up anyone who is waiting for task completion */
 937         rpc_mark_complete_task(task);
 938
 939         rpc_put_task(task);
 940 }
 941
 942 /**
 943  * rpc_run_task - Allocate a new RPC task, then run rpc_execute against it
 944  * @clnt: pointer to RPC client
 945  * @flags: RPC flags
 946  * @ops: RPC call ops
 947  * @data: user call data
 948  */
 949 struct rpc_task *rpc_run_task(struct rpc_clnt *clnt, int flags,
 950                                         const struct rpc_call_ops *ops,
 951                                         void *data)
 952 {
 953         struct rpc_task *task;
 954         task = rpc_new_task(clnt, flags, ops, data);
 955         if (task == NULL) {
 956                 rpc_release_calldata(ops, data);
 957                 return ERR_PTR(-ENOMEM);
 958         }
 959         atomic_inc(&task->tk_count);
 960         rpc_execute(task);
 961         return task;
 962 }
 963 EXPORT_SYMBOL(rpc_run_task);
 964
 965 /*
 966  * Kill all tasks for the given client.
 967  * XXX: kill their descendants as well?
 968  */
 969 void rpc_killall_tasks(struct rpc_clnt *clnt)
 970 {
 971         struct rpc_task *rovr;
 972         struct list_head *le;
 973
 974         dprintk("RPC:       killing all tasks for client %p\n", clnt);
 975
 976         /*
 977          * Spin lock all_tasks to prevent changes...
 978          */
 979         spin_lock(&rpc_sched_lock);
 980         alltask_for_each(rovr, le, &all_tasks) {
 981                 if (! RPC_IS_ACTIVATED(rovr))
 982                         continue;
 983                 if (!clnt || rovr->tk_client == clnt) {
 984                         rovr->tk_flags |= RPC_TASK_KILLED;
 985                         rpc_exit(rovr, -EIO);
 986                         rpc_wake_up_task(rovr);
 987                 }
 988         }
 989         spin_unlock(&rpc_sched_lock);
 990 }
 991
 992 static void rpciod_killall(void)
 993 {
 994         unsigned long flags;
 995
 996         while (!list_empty(&all_tasks)) {
 997                 clear_thread_flag(TIF_SIGPENDING);
 998                 rpc_killall_tasks(NULL);
 999                 flush_workqueue(rpciod_workqueue);
1000                 if (!list_empty(&all_tasks)) {
1001                         dprintk("RPC:       rpciod_killall: waiting for tasks "
1002                                         "to exit\n");
1003                         yield();
1004                 }
1005         }
1006
1007         spin_lock_irqsave(&current->sighand->siglock, flags);
1008         recalc_sigpending();
1009         spin_unlock_irqrestore(&current->sighand->siglock, flags);
1010 }
1011
1012 /*
1013  * Start up the rpciod process if it's not already running.
1014  */
1015 int
1016 rpciod_up(void)
1017 {
1018         struct workqueue_struct *wq;
1019         int error = 0;
1020
1021         mutex_lock(&rpciod_mutex);
1022         dprintk("RPC:       rpciod_up: users %u\n", rpciod_users);
1023         rpciod_users++;
1024         if (rpciod_workqueue)
1025                 goto out;
1026         /*
1027          * If there's no pid, we should be the first user.
1028          */
1029         if (rpciod_users > 1)
1030                 printk(KERN_WARNING "rpciod_up: no workqueue, %u users??\n", rpciod_users);
1031         /*
1032          * Create the rpciod thread and wait for it to start.
1033          */
1034         error = -ENOMEM;
1035         wq = create_workqueue("rpciod");
1036         if (wq == NULL) {
1037                 printk(KERN_WARNING "rpciod_up: create workqueue failed, error=%d\n", error);
1038                 rpciod_users--;
1039                 goto out;
1040         }
1041         rpciod_workqueue = wq;
1042         error = 0;
1043 out:
1044         mutex_unlock(&rpciod_mutex);
1045         return error;
1046 }
1047
1048 void
1049 rpciod_down(void)
1050 {
1051         mutex_lock(&rpciod_mutex);
1052         dprintk("RPC:       rpciod_down sema %u\n", rpciod_users);
1053         if (rpciod_users) {
1054                 if (--rpciod_users)
1055                         goto out;
1056         } else
1057                 printk(KERN_WARNING "rpciod_down: no users??\n");
1058
1059         if (!rpciod_workqueue) {
1060                 dprintk("RPC:       rpciod_down: Nothing to do!\n");
1061                 goto out;
1062         }
1063         rpciod_killall();
1064
1065         destroy_workqueue(rpciod_workqueue);
1066         rpciod_workqueue = NULL;
1067  out:
1068         mutex_unlock(&rpciod_mutex);
1069 }
1070
1071 #ifdef RPC_DEBUG
1072 void rpc_show_tasks(void)
1073 {
1074         struct list_head *le;
1075         struct rpc_task *t;
1076
1077         spin_lock(&rpc_sched_lock);
1078         if (list_empty(&all_tasks)) {
1079                 spin_unlock(&rpc_sched_lock);
1080                 return;
1081         }
1082         printk("-pid- proc flgs status -client- -prog- --rqstp- -timeout "
1083                 "-rpcwait -action- ---ops--\n");
1084         alltask_for_each(t, le, &all_tasks) {
1085                 const char *rpc_waitq = "none";
1086
1087                 if (RPC_IS_QUEUED(t))
1088                         rpc_waitq = rpc_qname(t->u.tk_wait.rpc_waitq);
1089
1090                 printk("%5u %04d %04x %6d %8p %6d %8p %8ld %8s %8p %8p\n",
1091                         t->tk_pid,
1092                         (t->tk_msg.rpc_proc ? t->tk_msg.rpc_proc->p_proc : -1),
1093                         t->tk_flags, t->tk_status,
1094                         t->tk_client,
1095                         (t->tk_client ? t->tk_client->cl_prog : 0),
1096                         t->tk_rqstp, t->tk_timeout,
1097                         rpc_waitq,
1098                         t->tk_action, t->tk_ops);
1099         }
1100         spin_unlock(&rpc_sched_lock);
1101 }
1102 #endif
1103
1104 void
1105 rpc_destroy_mempool(void)
1106 {
1107         if (rpc_buffer_mempool)
1108                 mempool_destroy(rpc_buffer_mempool);
1109         if (rpc_task_mempool)
1110                 mempool_destroy(rpc_task_mempool);
1111         if (rpc_task_slabp)
1112                 kmem_cache_destroy(rpc_task_slabp);
1113         if (rpc_buffer_slabp)
1114                 kmem_cache_destroy(rpc_buffer_slabp);
1115 }
1116
1117 int
1118 rpc_init_mempool(void)
1119 {
1120         rpc_task_slabp = kmem_cache_create("rpc_tasks",
1121                                              sizeof(struct rpc_task),
1122                                              0, SLAB_HWCACHE_ALIGN,
1123                                              NULL, NULL);
1124         if (!rpc_task_slabp)
1125                 goto err_nomem;
1126         rpc_buffer_slabp = kmem_cache_create("rpc_buffers",
1127                                              RPC_BUFFER_MAXSIZE,
1128                                              0, SLAB_HWCACHE_ALIGN,
1129                                              NULL, NULL);
1130         if (!rpc_buffer_slabp)
1131                 goto err_nomem;
1132         rpc_task_mempool = mempool_create_slab_pool(RPC_TASK_POOLSIZE,
1133                                                     rpc_task_slabp);
1134         if (!rpc_task_mempool)
1135                 goto err_nomem;
1136         rpc_buffer_mempool = mempool_create_slab_pool(RPC_BUFFER_POOLSIZE,
1137                                                       rpc_buffer_slabp);
1138         if (!rpc_buffer_mempool)
1139                 goto err_nomem;
1140         return 0;
1141 err_nomem:
1142         rpc_destroy_mempool();
1143         return -ENOMEM;
1144 }