err.no Git - linux-2.6/blob - 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 #endif
  29
  30 /*
  31  * RPC slabs and memory pools
  32  */
  33 #define RPC_BUFFER_MAXSIZE      (2048)
  34 #define RPC_BUFFER_POOLSIZE     (8)
  35 #define RPC_TASK_POOLSIZE       (8)
  36 static struct kmem_cache        *rpc_task_slabp __read_mostly;
  37 static struct kmem_cache        *rpc_buffer_slabp __read_mostly;
  38 static mempool_t        *rpc_task_mempool __read_mostly;
  39 static mempool_t        *rpc_buffer_mempool __read_mostly;
  40
  41 static void                     rpc_async_schedule(struct work_struct *);
  42 static void                      rpc_release_task(struct rpc_task *task);
  43 static void __rpc_queue_timer_fn(unsigned long ptr);
  44
  45 /*
  46  * RPC tasks sit here while waiting for conditions to improve.
  47  */
  48 static struct rpc_wait_queue delay_queue;
  49
  50 /*
  51  * rpciod-related stuff
  52  */
  53 struct workqueue_struct *rpciod_workqueue;
  54
  55 /*
  56  * Disable the timer for a given RPC task. Should be called with
  57  * queue->lock and bh_disabled in order to avoid races within
  58  * rpc_run_timer().
  59  */
  60 static void
  61 __rpc_disable_timer(struct rpc_task *task)
  62 {
  63         if (task->tk_timeout == 0)
  64                 return;
  65         dprintk("RPC: %5u disabling timer\n", task->tk_pid);
  66         task->tk_timeout = 0;
  67         list_del(&task->u.tk_wait.timer_list);
  68 }
  69
  70 static void
  71 rpc_set_queue_timer(struct rpc_wait_queue *queue, unsigned long expires)
  72 {
  73         queue->timer_list.expires = expires;
  74         mod_timer(&queue->timer_list.timer, expires);
  75 }
  76
  77 /*
  78  * Set up a timer for the current task.
  79  */
  80 static void
  81 __rpc_add_timer(struct rpc_task *task)
  82 {
  83         if (!task->tk_timeout)
  84                 return;
  85
  86         dprintk("RPC: %5u setting alarm for %lu ms\n",
  87                         task->tk_pid, task->tk_timeout * 1000 / HZ);
  88
  89         set_bit(RPC_TASK_HAS_TIMER, &task->tk_runstate);
  90         mod_timer(&task->tk_timer, jiffies + task->tk_timeout);
  91 }
  92
  93 /*
  94  * Delete any timer for the current task. Because we use del_timer_sync(),
  95  * this function should never be called while holding queue->lock.
  96  */
  97 static void
  98 rpc_delete_timer(struct rpc_task *task)
  99 {
 100         if (RPC_IS_QUEUED(task))
 101                 return;
 102         if (test_and_clear_bit(RPC_TASK_HAS_TIMER, &task->tk_runstate)) {
 103                 del_singleshot_timer_sync(&task->tk_timer);
 104                 dprintk("RPC: %5u deleting timer\n", task->tk_pid);
 105         }
 106 }
 107
 108 /*
 109  * Add new request to a priority queue.
 110  */
 111 static void __rpc_add_wait_queue_priority(struct rpc_wait_queue *queue, struct rpc_task *task)
 112 {
 113         struct list_head *q;
 114         struct rpc_task *t;
 115
 116         INIT_LIST_HEAD(&task->u.tk_wait.links);
 117         q = &queue->tasks[task->tk_priority];
 118         if (unlikely(task->tk_priority > queue->maxpriority))
 119                 q = &queue->tasks[queue->maxpriority];
 120         list_for_each_entry(t, q, u.tk_wait.list) {
 121                 if (t->tk_owner == task->tk_owner) {
 122                         list_add_tail(&task->u.tk_wait.list, &t->u.tk_wait.links);
 123                         return;
 124                 }
 125         }
 126         list_add_tail(&task->u.tk_wait.list, q);
 127 }
 128
 129 /*
 130  * Add new request to wait queue.
 131  *
 132  * Swapper tasks always get inserted at the head of the queue.
 133  * This should avoid many nasty memory deadlocks and hopefully
 134  * improve overall performance.
 135  * Everyone else gets appended to the queue to ensure proper FIFO behavior.
 136  */
 137 static void __rpc_add_wait_queue(struct rpc_wait_queue *queue, struct rpc_task *task)
 138 {
 139         BUG_ON (RPC_IS_QUEUED(task));
 140
 141         if (RPC_IS_PRIORITY(queue))
 142                 __rpc_add_wait_queue_priority(queue, task);
 143         else if (RPC_IS_SWAPPER(task))
 144                 list_add(&task->u.tk_wait.list, &queue->tasks[0]);
 145         else
 146                 list_add_tail(&task->u.tk_wait.list, &queue->tasks[0]);
 147         task->tk_waitqueue = queue;
 148         queue->qlen++;
 149         rpc_set_queued(task);
 150
 151         dprintk("RPC: %5u added to queue %p \"%s\"\n",
 152                         task->tk_pid, queue, rpc_qname(queue));
 153 }
 154
 155 /*
 156  * Remove request from a priority queue.
 157  */
 158 static void __rpc_remove_wait_queue_priority(struct rpc_task *task)
 159 {
 160         struct rpc_task *t;
 161
 162         if (!list_empty(&task->u.tk_wait.links)) {
 163                 t = list_entry(task->u.tk_wait.links.next, struct rpc_task, u.tk_wait.list);
 164                 list_move(&t->u.tk_wait.list, &task->u.tk_wait.list);
 165                 list_splice_init(&task->u.tk_wait.links, &t->u.tk_wait.links);
 166         }
 167 }
 168
 169 /*
 170  * Remove request from queue.
 171  * Note: must be called with spin lock held.
 172  */
 173 static void __rpc_remove_wait_queue(struct rpc_wait_queue *queue, struct rpc_task *task)
 174 {
 175         __rpc_disable_timer(task);
 176         if (RPC_IS_PRIORITY(queue))
 177                 __rpc_remove_wait_queue_priority(task);
 178         list_del(&task->u.tk_wait.list);
 179         queue->qlen--;
 180         dprintk("RPC: %5u removed from queue %p \"%s\"\n",
 181                         task->tk_pid, queue, rpc_qname(queue));
 182 }
 183
 184 static inline void rpc_set_waitqueue_priority(struct rpc_wait_queue *queue, int priority)
 185 {
 186         queue->priority = priority;
 187         queue->count = 1 << (priority * 2);
 188 }
 189
 190 static inline void rpc_set_waitqueue_owner(struct rpc_wait_queue *queue, pid_t pid)
 191 {
 192         queue->owner = pid;
 193         queue->nr = RPC_BATCH_COUNT;
 194 }
 195
 196 static inline void rpc_reset_waitqueue_priority(struct rpc_wait_queue *queue)
 197 {
 198         rpc_set_waitqueue_priority(queue, queue->maxpriority);
 199         rpc_set_waitqueue_owner(queue, 0);
 200 }
 201
 202 static void __rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname, unsigned char nr_queues)
 203 {
 204         int i;
 205
 206         spin_lock_init(&queue->lock);
 207         for (i = 0; i < ARRAY_SIZE(queue->tasks); i++)
 208                 INIT_LIST_HEAD(&queue->tasks[i]);
 209         queue->maxpriority = nr_queues - 1;
 210         rpc_reset_waitqueue_priority(queue);
 211         queue->qlen = 0;
 212         setup_timer(&queue->timer_list.timer, __rpc_queue_timer_fn, (unsigned long)queue);
 213         INIT_LIST_HEAD(&queue->timer_list.list);
 214 #ifdef RPC_DEBUG
 215         queue->name = qname;
 216 #endif
 217 }
 218
 219 void rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname)
 220 {
 221         __rpc_init_priority_wait_queue(queue, qname, RPC_NR_PRIORITY);
 222 }
 223
 224 void rpc_init_wait_queue(struct rpc_wait_queue *queue, const char *qname)
 225 {
 226         __rpc_init_priority_wait_queue(queue, qname, 1);
 227 }
 228 EXPORT_SYMBOL_GPL(rpc_init_wait_queue);
 229
 230 void rpc_destroy_wait_queue(struct rpc_wait_queue *queue)
 231 {
 232         del_timer_sync(&queue->timer_list.timer);
 233 }
 234 EXPORT_SYMBOL_GPL(rpc_destroy_wait_queue);
 235
 236 static int rpc_wait_bit_killable(void *word)
 237 {
 238         if (fatal_signal_pending(current))
 239                 return -ERESTARTSYS;
 240         schedule();
 241         return 0;
 242 }
 243
 244 #ifdef RPC_DEBUG
 245 static void rpc_task_set_debuginfo(struct rpc_task *task)
 246 {
 247         static atomic_t rpc_pid;
 248
 249         task->tk_magic = RPC_TASK_MAGIC_ID;
 250         task->tk_pid = atomic_inc_return(&rpc_pid);
 251 }
 252 #else
 253 static inline void rpc_task_set_debuginfo(struct rpc_task *task)
 254 {
 255 }
 256 #endif
 257
 258 static void rpc_set_active(struct rpc_task *task)
 259 {
 260         struct rpc_clnt *clnt;
 261         if (test_and_set_bit(RPC_TASK_ACTIVE, &task->tk_runstate) != 0)
 262                 return;
 263         rpc_task_set_debuginfo(task);
 264         /* Add to global list of all tasks */
 265         clnt = task->tk_client;
 266         if (clnt != NULL) {
 267                 spin_lock(&clnt->cl_lock);
 268                 list_add_tail(&task->tk_task, &clnt->cl_tasks);
 269                 spin_unlock(&clnt->cl_lock);
 270         }
 271 }
 272
 273 /*
 274  * Mark an RPC call as having completed by clearing the 'active' bit
 275  */
 276 static void rpc_mark_complete_task(struct rpc_task *task)
 277 {
 278         smp_mb__before_clear_bit();
 279         clear_bit(RPC_TASK_ACTIVE, &task->tk_runstate);
 280         smp_mb__after_clear_bit();
 281         wake_up_bit(&task->tk_runstate, RPC_TASK_ACTIVE);
 282 }
 283
 284 /*
 285  * Allow callers to wait for completion of an RPC call
 286  */
 287 int __rpc_wait_for_completion_task(struct rpc_task *task, int (*action)(void *))
 288 {
 289         if (action == NULL)
 290                 action = rpc_wait_bit_killable;
 291         return wait_on_bit(&task->tk_runstate, RPC_TASK_ACTIVE,
 292                         action, TASK_KILLABLE);
 293 }
 294 EXPORT_SYMBOL_GPL(__rpc_wait_for_completion_task);
 295
 296 /*
 297  * Make an RPC task runnable.
 298  *
 299  * Note: If the task is ASYNC, this must be called with
 300  * the spinlock held to protect the wait queue operation.
 301  */
 302 static void rpc_make_runnable(struct rpc_task *task)
 303 {
 304         rpc_clear_queued(task);
 305         if (rpc_test_and_set_running(task))
 306                 return;
 307         /* We might have raced */
 308         if (RPC_IS_QUEUED(task)) {
 309                 rpc_clear_running(task);
 310                 return;
 311         }
 312         if (RPC_IS_ASYNC(task)) {
 313                 int status;
 314
 315                 INIT_WORK(&task->u.tk_work, rpc_async_schedule);
 316                 status = queue_work(rpciod_workqueue, &task->u.tk_work);
 317                 if (status < 0) {
 318                         printk(KERN_WARNING "RPC: failed to add task to queue: error: %d!\n", status);
 319                         task->tk_status = status;
 320                         return;
 321                 }
 322         } else
 323                 wake_up_bit(&task->tk_runstate, RPC_TASK_QUEUED);
 324 }
 325
 326 /*
 327  * Prepare for sleeping on a wait queue.
 328  * By always appending tasks to the list we ensure FIFO behavior.
 329  * NB: An RPC task will only receive interrupt-driven events as long
 330  * as it's on a wait queue.
 331  */
 332 static void __rpc_sleep_on(struct rpc_wait_queue *q, struct rpc_task *task,
 333                         rpc_action action)
 334 {
 335         dprintk("RPC: %5u sleep_on(queue \"%s\" time %lu)\n",
 336                         task->tk_pid, rpc_qname(q), jiffies);
 337
 338         if (!RPC_IS_ASYNC(task) && !RPC_IS_ACTIVATED(task)) {
 339                 printk(KERN_ERR "RPC: Inactive synchronous task put to sleep!\n");
 340                 return;
 341         }
 342
 343         __rpc_add_wait_queue(q, task);
 344
 345         BUG_ON(task->tk_callback != NULL);
 346         task->tk_callback = action;
 347         __rpc_add_timer(task);
 348 }
 349
 350 void rpc_sleep_on(struct rpc_wait_queue *q, struct rpc_task *task,
 351                                 rpc_action action)
 352 {
 353         /* Mark the task as being activated if so needed */
 354         rpc_set_active(task);
 355
 356         /*
 357          * Protect the queue operations.
 358          */
 359         spin_lock_bh(&q->lock);
 360         __rpc_sleep_on(q, task, action);
 361         spin_unlock_bh(&q->lock);
 362 }
 363 EXPORT_SYMBOL_GPL(rpc_sleep_on);
 364
 365 /**
 366  * __rpc_do_wake_up_task - wake up a single rpc_task
 367  * @queue: wait queue
 368  * @task: task to be woken up
 369  *
 370  * Caller must hold queue->lock, and have cleared the task queued flag.
 371  */
 372 static void __rpc_do_wake_up_task(struct rpc_wait_queue *queue, struct rpc_task *task)
 373 {
 374         dprintk("RPC: %5u __rpc_wake_up_task (now %lu)\n",
 375                         task->tk_pid, jiffies);
 376
 377 #ifdef RPC_DEBUG
 378         BUG_ON(task->tk_magic != RPC_TASK_MAGIC_ID);
 379 #endif
 380         /* Has the task been executed yet? If not, we cannot wake it up! */
 381         if (!RPC_IS_ACTIVATED(task)) {
 382                 printk(KERN_ERR "RPC: Inactive task (%p) being woken up!\n", task);
 383                 return;
 384         }
 385
 386         __rpc_remove_wait_queue(queue, task);
 387
 388         rpc_make_runnable(task);
 389
 390         dprintk("RPC:       __rpc_wake_up_task done\n");
 391 }
 392
 393 /*
 394  * Wake up a queued task while the queue lock is being held
 395  */
 396 static void rpc_wake_up_task_queue_locked(struct rpc_wait_queue *queue, struct rpc_task *task)
 397 {
 398         if (!RPC_IS_QUEUED(task) || task->tk_waitqueue != queue)
 399                 return;
 400         if (rpc_start_wakeup(task)) {
 401                         __rpc_do_wake_up_task(queue, task);
 402                 rpc_finish_wakeup(task);
 403         }
 404 }
 405
 406 /*
 407  * Wake up a task on a specific queue
 408  */
 409 void rpc_wake_up_queued_task(struct rpc_wait_queue *queue, struct rpc_task *task)
 410 {
 411         rcu_read_lock_bh();
 412         spin_lock(&queue->lock);
 413         rpc_wake_up_task_queue_locked(queue, task);
 414         spin_unlock(&queue->lock);
 415         rcu_read_unlock_bh();
 416 }
 417 EXPORT_SYMBOL_GPL(rpc_wake_up_queued_task);
 418
 419 /*
 420  * Wake up the specified task
 421  */
 422 static void rpc_wake_up_task(struct rpc_task *task)
 423 {
 424         rpc_wake_up_queued_task(task->tk_waitqueue, task);
 425 }
 426
 427 /*
 428  * Wake up the next task on a priority queue.
 429  */
 430 static struct rpc_task * __rpc_wake_up_next_priority(struct rpc_wait_queue *queue)
 431 {
 432         struct list_head *q;
 433         struct rpc_task *task;
 434
 435         /*
 436          * Service a batch of tasks from a single owner.
 437          */
 438         q = &queue->tasks[queue->priority];
 439         if (!list_empty(q)) {
 440                 task = list_entry(q->next, struct rpc_task, u.tk_wait.list);
 441                 if (queue->owner == task->tk_owner) {
 442                         if (--queue->nr)
 443                                 goto out;
 444                         list_move_tail(&task->u.tk_wait.list, q);
 445                 }
 446                 /*
 447                  * Check if we need to switch queues.
 448                  */
 449                 if (--queue->count)
 450                         goto new_owner;
 451         }
 452
 453         /*
 454          * Service the next queue.
 455          */
 456         do {
 457                 if (q == &queue->tasks[0])
 458                         q = &queue->tasks[queue->maxpriority];
 459                 else
 460                         q = q - 1;
 461                 if (!list_empty(q)) {
 462                         task = list_entry(q->next, struct rpc_task, u.tk_wait.list);
 463                         goto new_queue;
 464                 }
 465         } while (q != &queue->tasks[queue->priority]);
 466
 467         rpc_reset_waitqueue_priority(queue);
 468         return NULL;
 469
 470 new_queue:
 471         rpc_set_waitqueue_priority(queue, (unsigned int)(q - &queue->tasks[0]));
 472 new_owner:
 473         rpc_set_waitqueue_owner(queue, task->tk_owner);
 474 out:
 475         rpc_wake_up_task_queue_locked(queue, task);
 476         return task;
 477 }
 478
 479 /*
 480  * Wake up the next task on the wait queue.
 481  */
 482 struct rpc_task * rpc_wake_up_next(struct rpc_wait_queue *queue)
 483 {
 484         struct rpc_task *task = NULL;
 485
 486         dprintk("RPC:       wake_up_next(%p \"%s\")\n",
 487                         queue, rpc_qname(queue));
 488         rcu_read_lock_bh();
 489         spin_lock(&queue->lock);
 490         if (RPC_IS_PRIORITY(queue))
 491                 task = __rpc_wake_up_next_priority(queue);
 492         else {
 493                 task_for_first(task, &queue->tasks[0])
 494                         rpc_wake_up_task_queue_locked(queue, task);
 495         }
 496         spin_unlock(&queue->lock);
 497         rcu_read_unlock_bh();
 498
 499         return task;
 500 }
 501 EXPORT_SYMBOL_GPL(rpc_wake_up_next);
 502
 503 /**
 504  * rpc_wake_up - wake up all rpc_tasks
 505  * @queue: rpc_wait_queue on which the tasks are sleeping
 506  *
 507  * Grabs queue->lock
 508  */
 509 void rpc_wake_up(struct rpc_wait_queue *queue)
 510 {
 511         struct rpc_task *task, *next;
 512         struct list_head *head;
 513
 514         rcu_read_lock_bh();
 515         spin_lock(&queue->lock);
 516         head = &queue->tasks[queue->maxpriority];
 517         for (;;) {
 518                 list_for_each_entry_safe(task, next, head, u.tk_wait.list)
 519                         rpc_wake_up_task_queue_locked(queue, task);
 520                 if (head == &queue->tasks[0])
 521                         break;
 522                 head--;
 523         }
 524         spin_unlock(&queue->lock);
 525         rcu_read_unlock_bh();
 526 }
 527 EXPORT_SYMBOL_GPL(rpc_wake_up);
 528
 529 /**
 530  * rpc_wake_up_status - wake up all rpc_tasks and set their status value.
 531  * @queue: rpc_wait_queue on which the tasks are sleeping
 532  * @status: status value to set
 533  *
 534  * Grabs queue->lock
 535  */
 536 void rpc_wake_up_status(struct rpc_wait_queue *queue, int status)
 537 {
 538         struct rpc_task *task, *next;
 539         struct list_head *head;
 540
 541         rcu_read_lock_bh();
 542         spin_lock(&queue->lock);
 543         head = &queue->tasks[queue->maxpriority];
 544         for (;;) {
 545                 list_for_each_entry_safe(task, next, head, u.tk_wait.list) {
 546                         task->tk_status = status;
 547                         rpc_wake_up_task_queue_locked(queue, task);
 548                 }
 549                 if (head == &queue->tasks[0])
 550                         break;
 551                 head--;
 552         }
 553         spin_unlock(&queue->lock);
 554         rcu_read_unlock_bh();
 555 }
 556 EXPORT_SYMBOL_GPL(rpc_wake_up_status);
 557
 558 /*
 559  * Run a timeout function.
 560  */
 561 static void rpc_run_timer(unsigned long ptr)
 562 {
 563         struct rpc_task *task = (struct rpc_task *)ptr;
 564         struct rpc_wait_queue *queue = task->tk_waitqueue;
 565
 566         spin_lock(&queue->lock);
 567         if (RPC_IS_QUEUED(task) && task->tk_waitqueue == queue) {
 568                 dprintk("RPC: %5u timeout\n", task->tk_pid);
 569                 task->tk_status = -ETIMEDOUT;
 570                 rpc_wake_up_task_queue_locked(queue, task);
 571         }
 572         spin_unlock(&queue->lock);
 573         smp_mb__before_clear_bit();
 574         clear_bit(RPC_TASK_HAS_TIMER, &task->tk_runstate);
 575         smp_mb__after_clear_bit();
 576 }
 577
 578 static void __rpc_queue_timer_fn(unsigned long ptr)
 579 {
 580         struct rpc_wait_queue *queue = (struct rpc_wait_queue *)ptr;
 581         struct rpc_task *task, *n;
 582         unsigned long expires, now, timeo;
 583
 584         spin_lock(&queue->lock);
 585         expires = now = jiffies;
 586         list_for_each_entry_safe(task, n, &queue->timer_list.list, u.tk_wait.timer_list) {
 587                 timeo = task->u.tk_wait.expires;
 588                 if (time_after_eq(now, timeo)) {
 589                         list_del_init(&task->u.tk_wait.timer_list);
 590                         dprintk("RPC: %5u timeout\n", task->tk_pid);
 591                         task->tk_status = -ETIMEDOUT;
 592                         rpc_wake_up_task_queue_locked(queue, task);
 593                         continue;
 594                 }
 595                 if (expires == now || time_after(expires, timeo))
 596                         expires = timeo;
 597         }
 598         if (!list_empty(&queue->timer_list.list))
 599                 rpc_set_queue_timer(queue, expires);
 600         spin_unlock(&queue->lock);
 601 }
 602
 603 static void __rpc_atrun(struct rpc_task *task)
 604 {
 605         task->tk_status = 0;
 606 }
 607
 608 /*
 609  * Run a task at a later time
 610  */
 611 void rpc_delay(struct rpc_task *task, unsigned long delay)
 612 {
 613         task->tk_timeout = delay;
 614         rpc_sleep_on(&delay_queue, task, __rpc_atrun);
 615 }
 616 EXPORT_SYMBOL_GPL(rpc_delay);
 617
 618 /*
 619  * Helper to call task->tk_ops->rpc_call_prepare
 620  */
 621 static void rpc_prepare_task(struct rpc_task *task)
 622 {
 623         lock_kernel();
 624         task->tk_ops->rpc_call_prepare(task, task->tk_calldata);
 625         unlock_kernel();
 626 }
 627
 628 /*
 629  * Helper that calls task->tk_ops->rpc_call_done if it exists
 630  */
 631 void rpc_exit_task(struct rpc_task *task)
 632 {
 633         task->tk_action = NULL;
 634         if (task->tk_ops->rpc_call_done != NULL) {
 635                 lock_kernel();
 636                 task->tk_ops->rpc_call_done(task, task->tk_calldata);
 637                 unlock_kernel();
 638                 if (task->tk_action != NULL) {
 639                         WARN_ON(RPC_ASSASSINATED(task));
 640                         /* Always release the RPC slot and buffer memory */
 641                         xprt_release(task);
 642                 }
 643         }
 644 }
 645 EXPORT_SYMBOL_GPL(rpc_exit_task);
 646
 647 void rpc_release_calldata(const struct rpc_call_ops *ops, void *calldata)
 648 {
 649         if (ops->rpc_release != NULL) {
 650                 lock_kernel();
 651                 ops->rpc_release(calldata);
 652                 unlock_kernel();
 653         }
 654 }
 655
 656 /*
 657  * This is the RPC `scheduler' (or rather, the finite state machine).
 658  */
 659 static void __rpc_execute(struct rpc_task *task)
 660 {
 661         int             status = 0;
 662
 663         dprintk("RPC: %5u __rpc_execute flags=0x%x\n",
 664                         task->tk_pid, task->tk_flags);
 665
 666         BUG_ON(RPC_IS_QUEUED(task));
 667
 668         for (;;) {
 669                 /*
 670                  * Garbage collection of pending timers...
 671                  */
 672                 rpc_delete_timer(task);
 673
 674                 /*
 675                  * Execute any pending callback.
 676                  */
 677                 if (RPC_DO_CALLBACK(task)) {
 678                         /* Define a callback save pointer */
 679                         void (*save_callback)(struct rpc_task *);
 680
 681                         /*
 682                          * If a callback exists, save it, reset it,
 683                          * call it.
 684                          * The save is needed to stop from resetting
 685                          * another callback set within the callback handler
 686                          * - Dave
 687                          */
 688                         save_callback=task->tk_callback;
 689                         task->tk_callback=NULL;
 690                         save_callback(task);
 691                 }
 692
 693                 /*
 694                  * Perform the next FSM step.
 695                  * tk_action may be NULL when the task has been killed
 696                  * by someone else.
 697                  */
 698                 if (!RPC_IS_QUEUED(task)) {
 699                         if (task->tk_action == NULL)
 700                                 break;
 701                         task->tk_action(task);
 702                 }
 703
 704                 /*
 705                  * Lockless check for whether task is sleeping or not.
 706                  */
 707                 if (!RPC_IS_QUEUED(task))
 708                         continue;
 709                 rpc_clear_running(task);
 710                 if (RPC_IS_ASYNC(task)) {
 711                         /* Careful! we may have raced... */
 712                         if (RPC_IS_QUEUED(task))
 713                                 return;
 714                         if (rpc_test_and_set_running(task))
 715                                 return;
 716                         continue;
 717                 }
 718
 719                 /* sync task: sleep here */
 720                 dprintk("RPC: %5u sync task going to sleep\n", task->tk_pid);
 721                 status = out_of_line_wait_on_bit(&task->tk_runstate,
 722                                 RPC_TASK_QUEUED, rpc_wait_bit_killable,
 723                                 TASK_KILLABLE);
 724                 if (status == -ERESTARTSYS) {
 725                         /*
 726                          * When a sync task receives a signal, it exits with
 727                          * -ERESTARTSYS. In order to catch any callbacks that
 728                          * clean up after sleeping on some queue, we don't
 729                          * break the loop here, but go around once more.
 730                          */
 731                         dprintk("RPC: %5u got signal\n", task->tk_pid);
 732                         task->tk_flags |= RPC_TASK_KILLED;
 733                         rpc_exit(task, -ERESTARTSYS);
 734                         rpc_wake_up_task(task);
 735                 }
 736                 rpc_set_running(task);
 737                 dprintk("RPC: %5u sync task resuming\n", task->tk_pid);
 738         }
 739
 740         dprintk("RPC: %5u return %d, status %d\n", task->tk_pid, status,
 741                         task->tk_status);
 742         /* Release all resources associated with the task */
 743         rpc_release_task(task);
 744 }
 745
 746 /*
 747  * User-visible entry point to the scheduler.
 748  *
 749  * This may be called recursively if e.g. an async NFS task updates
 750  * the attributes and finds that dirty pages must be flushed.
 751  * NOTE: Upon exit of this function the task is guaranteed to be
 752  *       released. In particular note that tk_release() will have
 753  *       been called, so your task memory may have been freed.
 754  */
 755 void rpc_execute(struct rpc_task *task)
 756 {
 757         rpc_set_active(task);
 758         rpc_set_running(task);
 759         __rpc_execute(task);
 760 }
 761
 762 static void rpc_async_schedule(struct work_struct *work)
 763 {
 764         __rpc_execute(container_of(work, struct rpc_task, u.tk_work));
 765 }
 766
 767 struct rpc_buffer {
 768         size_t  len;
 769         char    data[];
 770 };
 771
 772 /**
 773  * rpc_malloc - allocate an RPC buffer
 774  * @task: RPC task that will use this buffer
 775  * @size: requested byte size
 776  *
 777  * To prevent rpciod from hanging, this allocator never sleeps,
 778  * returning NULL if the request cannot be serviced immediately.
 779  * The caller can arrange to sleep in a way that is safe for rpciod.
 780  *
 781  * Most requests are 'small' (under 2KiB) and can be serviced from a
 782  * mempool, ensuring that NFS reads and writes can always proceed,
 783  * and that there is good locality of reference for these buffers.
 784  *
 785  * In order to avoid memory starvation triggering more writebacks of
 786  * NFS requests, we avoid using GFP_KERNEL.
 787  */
 788 void *rpc_malloc(struct rpc_task *task, size_t size)
 789 {
 790         struct rpc_buffer *buf;
 791         gfp_t gfp = RPC_IS_SWAPPER(task) ? GFP_ATOMIC : GFP_NOWAIT;
 792
 793         size += sizeof(struct rpc_buffer);
 794         if (size <= RPC_BUFFER_MAXSIZE)
 795                 buf = mempool_alloc(rpc_buffer_mempool, gfp);
 796         else
 797                 buf = kmalloc(size, gfp);
 798
 799         if (!buf)
 800                 return NULL;
 801
 802         buf->len = size;
 803         dprintk("RPC: %5u allocated buffer of size %zu at %p\n",
 804                         task->tk_pid, size, buf);
 805         return &buf->data;
 806 }
 807 EXPORT_SYMBOL_GPL(rpc_malloc);
 808
 809 /**
 810  * rpc_free - free buffer allocated via rpc_malloc
 811  * @buffer: buffer to free
 812  *
 813  */
 814 void rpc_free(void *buffer)
 815 {
 816         size_t size;
 817         struct rpc_buffer *buf;
 818
 819         if (!buffer)
 820                 return;
 821
 822         buf = container_of(buffer, struct rpc_buffer, data);
 823         size = buf->len;
 824
 825         dprintk("RPC:       freeing buffer of size %zu at %p\n",
 826                         size, buf);
 827
 828         if (size <= RPC_BUFFER_MAXSIZE)
 829                 mempool_free(buf, rpc_buffer_mempool);
 830         else
 831                 kfree(buf);
 832 }
 833 EXPORT_SYMBOL_GPL(rpc_free);
 834
 835 /*
 836  * Creation and deletion of RPC task structures
 837  */
 838 static void rpc_init_task(struct rpc_task *task, const struct rpc_task_setup *task_setup_data)
 839 {
 840         memset(task, 0, sizeof(*task));
 841         setup_timer(&task->tk_timer, rpc_run_timer, (unsigned long)task);
 842         atomic_set(&task->tk_count, 1);
 843         task->tk_flags  = task_setup_data->flags;
 844         task->tk_ops = task_setup_data->callback_ops;
 845         task->tk_calldata = task_setup_data->callback_data;
 846         INIT_LIST_HEAD(&task->tk_task);
 847
 848         /* Initialize retry counters */
 849         task->tk_garb_retry = 2;
 850         task->tk_cred_retry = 2;
 851
 852         task->tk_priority = task_setup_data->priority - RPC_PRIORITY_LOW;
 853         task->tk_owner = current->tgid;
 854
 855         /* Initialize workqueue for async tasks */
 856         task->tk_workqueue = task_setup_data->workqueue;
 857
 858         task->tk_client = task_setup_data->rpc_client;
 859         if (task->tk_client != NULL) {
 860                 kref_get(&task->tk_client->cl_kref);
 861                 if (task->tk_client->cl_softrtry)
 862                         task->tk_flags |= RPC_TASK_SOFT;
 863         }
 864
 865         if (task->tk_ops->rpc_call_prepare != NULL)
 866                 task->tk_action = rpc_prepare_task;
 867
 868         if (task_setup_data->rpc_message != NULL) {
 869                 memcpy(&task->tk_msg, task_setup_data->rpc_message, sizeof(task->tk_msg));
 870                 /* Bind the user cred */
 871                 if (task->tk_msg.rpc_cred != NULL)
 872                         rpcauth_holdcred(task);
 873                 else
 874                         rpcauth_bindcred(task);
 875                 if (task->tk_action == NULL)
 876                         rpc_call_start(task);
 877         }
 878
 879         /* starting timestamp */
 880         task->tk_start = jiffies;
 881
 882         dprintk("RPC:       new task initialized, procpid %u\n",
 883                                 task_pid_nr(current));
 884 }
 885
 886 static struct rpc_task *
 887 rpc_alloc_task(void)
 888 {
 889         return (struct rpc_task *)mempool_alloc(rpc_task_mempool, GFP_NOFS);
 890 }
 891
 892 static void rpc_free_task_rcu(struct rcu_head *rcu)
 893 {
 894         struct rpc_task *task = container_of(rcu, struct rpc_task, u.tk_rcu);
 895         dprintk("RPC: %5u freeing task\n", task->tk_pid);
 896         mempool_free(task, rpc_task_mempool);
 897 }
 898
 899 /*
 900  * Create a new task for the specified client.
 901  */
 902 struct rpc_task *rpc_new_task(const struct rpc_task_setup *setup_data)
 903 {
 904         struct rpc_task *task = setup_data->task;
 905         unsigned short flags = 0;
 906
 907         if (task == NULL) {
 908                 task = rpc_alloc_task();
 909                 if (task == NULL)
 910                         goto out;
 911                 flags = RPC_TASK_DYNAMIC;
 912         }
 913
 914         rpc_init_task(task, setup_data);
 915
 916         task->tk_flags |= flags;
 917         dprintk("RPC:       allocated task %p\n", task);
 918 out:
 919         return task;
 920 }
 921
 922 static void rpc_free_task(struct rpc_task *task)
 923 {
 924         const struct rpc_call_ops *tk_ops = task->tk_ops;
 925         void *calldata = task->tk_calldata;
 926
 927         if (task->tk_flags & RPC_TASK_DYNAMIC)
 928                 call_rcu_bh(&task->u.tk_rcu, rpc_free_task_rcu);
 929         rpc_release_calldata(tk_ops, calldata);
 930 }
 931
 932 static void rpc_async_release(struct work_struct *work)
 933 {
 934         rpc_free_task(container_of(work, struct rpc_task, u.tk_work));
 935 }
 936
 937 void rpc_put_task(struct rpc_task *task)
 938 {
 939         if (!atomic_dec_and_test(&task->tk_count))
 940                 return;
 941         /* Release resources */
 942         if (task->tk_rqstp)
 943                 xprt_release(task);
 944         if (task->tk_msg.rpc_cred)
 945                 rpcauth_unbindcred(task);
 946         if (task->tk_client) {
 947                 rpc_release_client(task->tk_client);
 948                 task->tk_client = NULL;
 949         }
 950         if (task->tk_workqueue != NULL) {
 951                 INIT_WORK(&task->u.tk_work, rpc_async_release);
 952                 queue_work(task->tk_workqueue, &task->u.tk_work);
 953         } else
 954                 rpc_free_task(task);
 955 }
 956 EXPORT_SYMBOL_GPL(rpc_put_task);
 957
 958 static void rpc_release_task(struct rpc_task *task)
 959 {
 960 #ifdef RPC_DEBUG
 961         BUG_ON(task->tk_magic != RPC_TASK_MAGIC_ID);
 962 #endif
 963         dprintk("RPC: %5u release task\n", task->tk_pid);
 964
 965         if (!list_empty(&task->tk_task)) {
 966                 struct rpc_clnt *clnt = task->tk_client;
 967                 /* Remove from client task list */
 968                 spin_lock(&clnt->cl_lock);
 969                 list_del(&task->tk_task);
 970                 spin_unlock(&clnt->cl_lock);
 971         }
 972         BUG_ON (RPC_IS_QUEUED(task));
 973
 974         /* Synchronously delete any running timer */
 975         rpc_delete_timer(task);
 976
 977 #ifdef RPC_DEBUG
 978         task->tk_magic = 0;
 979 #endif
 980         /* Wake up anyone who is waiting for task completion */
 981         rpc_mark_complete_task(task);
 982
 983         rpc_put_task(task);
 984 }
 985
 986 /*
 987  * Kill all tasks for the given client.
 988  * XXX: kill their descendants as well?
 989  */
 990 void rpc_killall_tasks(struct rpc_clnt *clnt)
 991 {
 992         struct rpc_task *rovr;
 993
 994
 995         if (list_empty(&clnt->cl_tasks))
 996                 return;
 997         dprintk("RPC:       killing all tasks for client %p\n", clnt);
 998         /*
 999          * Spin lock all_tasks to prevent changes...
1000          */
1001         spin_lock(&clnt->cl_lock);
1002         list_for_each_entry(rovr, &clnt->cl_tasks, tk_task) {
1003                 if (! RPC_IS_ACTIVATED(rovr))
1004                         continue;
1005                 if (!(rovr->tk_flags & RPC_TASK_KILLED)) {
1006                         rovr->tk_flags |= RPC_TASK_KILLED;
1007                         rpc_exit(rovr, -EIO);
1008                         rpc_wake_up_task(rovr);
1009                 }
1010         }
1011         spin_unlock(&clnt->cl_lock);
1012 }
1013 EXPORT_SYMBOL_GPL(rpc_killall_tasks);
1014
1015 int rpciod_up(void)
1016 {
1017         return try_module_get(THIS_MODULE) ? 0 : -EINVAL;
1018 }
1019
1020 void rpciod_down(void)
1021 {
1022         module_put(THIS_MODULE);
1023 }
1024
1025 /*
1026  * Start up the rpciod workqueue.
1027  */
1028 static int rpciod_start(void)
1029 {
1030         struct workqueue_struct *wq;
1031
1032         /*
1033          * Create the rpciod thread and wait for it to start.
1034          */
1035         dprintk("RPC:       creating workqueue rpciod\n");
1036         wq = create_workqueue("rpciod");
1037         rpciod_workqueue = wq;
1038         return rpciod_workqueue != NULL;
1039 }
1040
1041 static void rpciod_stop(void)
1042 {
1043         struct workqueue_struct *wq = NULL;
1044
1045         if (rpciod_workqueue == NULL)
1046                 return;
1047         dprintk("RPC:       destroying workqueue rpciod\n");
1048
1049         wq = rpciod_workqueue;
1050         rpciod_workqueue = NULL;
1051         destroy_workqueue(wq);
1052 }
1053
1054 void
1055 rpc_destroy_mempool(void)
1056 {
1057         rpciod_stop();
1058         if (rpc_buffer_mempool)
1059                 mempool_destroy(rpc_buffer_mempool);
1060         if (rpc_task_mempool)
1061                 mempool_destroy(rpc_task_mempool);
1062         if (rpc_task_slabp)
1063                 kmem_cache_destroy(rpc_task_slabp);
1064         if (rpc_buffer_slabp)
1065                 kmem_cache_destroy(rpc_buffer_slabp);
1066         rpc_destroy_wait_queue(&delay_queue);
1067 }
1068
1069 int
1070 rpc_init_mempool(void)
1071 {
1072         /*
1073          * The following is not strictly a mempool initialisation,
1074          * but there is no harm in doing it here
1075          */
1076         rpc_init_wait_queue(&delay_queue, "delayq");
1077         if (!rpciod_start())
1078                 goto err_nomem;
1079
1080         rpc_task_slabp = kmem_cache_create("rpc_tasks",
1081                                              sizeof(struct rpc_task),
1082                                              0, SLAB_HWCACHE_ALIGN,
1083                                              NULL);
1084         if (!rpc_task_slabp)
1085                 goto err_nomem;
1086         rpc_buffer_slabp = kmem_cache_create("rpc_buffers",
1087                                              RPC_BUFFER_MAXSIZE,
1088                                              0, SLAB_HWCACHE_ALIGN,
1089                                              NULL);
1090         if (!rpc_buffer_slabp)
1091                 goto err_nomem;
1092         rpc_task_mempool = mempool_create_slab_pool(RPC_TASK_POOLSIZE,
1093                                                     rpc_task_slabp);
1094         if (!rpc_task_mempool)
1095                 goto err_nomem;
1096         rpc_buffer_mempool = mempool_create_slab_pool(RPC_BUFFER_POOLSIZE,
1097                                                       rpc_buffer_slabp);
1098         if (!rpc_buffer_mempool)
1099                 goto err_nomem;
1100         return 0;
1101 err_nomem:
1102         rpc_destroy_mempool();
1103         return -ENOMEM;
1104 }