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