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 EXPORT_SYMBOL_GPL(rpc_init_priority_wait_queue);
 214
 215 void rpc_init_wait_queue(struct rpc_wait_queue *queue, const char *qname)
 216 {
 217         __rpc_init_priority_wait_queue(queue, qname, 1);
 218 }
 219 EXPORT_SYMBOL_GPL(rpc_init_wait_queue);
 220
 221 void rpc_destroy_wait_queue(struct rpc_wait_queue *queue)
 222 {
 223         del_timer_sync(&queue->timer_list.timer);
 224 }
 225 EXPORT_SYMBOL_GPL(rpc_destroy_wait_queue);
 226
 227 static int rpc_wait_bit_killable(void *word)
 228 {
 229         if (fatal_signal_pending(current))
 230                 return -ERESTARTSYS;
 231         schedule();
 232         return 0;
 233 }
 234
 235 #ifdef RPC_DEBUG
 236 static void rpc_task_set_debuginfo(struct rpc_task *task)
 237 {
 238         static atomic_t rpc_pid;
 239
 240         task->tk_magic = RPC_TASK_MAGIC_ID;
 241         task->tk_pid = atomic_inc_return(&rpc_pid);
 242 }
 243 #else
 244 static inline void rpc_task_set_debuginfo(struct rpc_task *task)
 245 {
 246 }
 247 #endif
 248
 249 static void rpc_set_active(struct rpc_task *task)
 250 {
 251         struct rpc_clnt *clnt;
 252         if (test_and_set_bit(RPC_TASK_ACTIVE, &task->tk_runstate) != 0)
 253                 return;
 254         rpc_task_set_debuginfo(task);
 255         /* Add to global list of all tasks */
 256         clnt = task->tk_client;
 257         if (clnt != NULL) {
 258                 spin_lock(&clnt->cl_lock);
 259                 list_add_tail(&task->tk_task, &clnt->cl_tasks);
 260                 spin_unlock(&clnt->cl_lock);
 261         }
 262 }
 263
 264 /*
 265  * Mark an RPC call as having completed by clearing the 'active' bit
 266  */
 267 static void rpc_mark_complete_task(struct rpc_task *task)
 268 {
 269         smp_mb__before_clear_bit();
 270         clear_bit(RPC_TASK_ACTIVE, &task->tk_runstate);
 271         smp_mb__after_clear_bit();
 272         wake_up_bit(&task->tk_runstate, RPC_TASK_ACTIVE);
 273 }
 274
 275 /*
 276  * Allow callers to wait for completion of an RPC call
 277  */
 278 int __rpc_wait_for_completion_task(struct rpc_task *task, int (*action)(void *))
 279 {
 280         if (action == NULL)
 281                 action = rpc_wait_bit_killable;
 282         return wait_on_bit(&task->tk_runstate, RPC_TASK_ACTIVE,
 283                         action, TASK_KILLABLE);
 284 }
 285 EXPORT_SYMBOL_GPL(__rpc_wait_for_completion_task);
 286
 287 /*
 288  * Make an RPC task runnable.
 289  *
 290  * Note: If the task is ASYNC, this must be called with
 291  * the spinlock held to protect the wait queue operation.
 292  */
 293 static void rpc_make_runnable(struct rpc_task *task)
 294 {
 295         rpc_clear_queued(task);
 296         if (rpc_test_and_set_running(task))
 297                 return;
 298         if (RPC_IS_ASYNC(task)) {
 299                 int status;
 300
 301                 INIT_WORK(&task->u.tk_work, rpc_async_schedule);
 302                 status = queue_work(rpciod_workqueue, &task->u.tk_work);
 303                 if (status < 0) {
 304                         printk(KERN_WARNING "RPC: failed to add task to queue: error: %d!\n", status);
 305                         task->tk_status = status;
 306                         return;
 307                 }
 308         } else
 309                 wake_up_bit(&task->tk_runstate, RPC_TASK_QUEUED);
 310 }
 311
 312 /*
 313  * Prepare for sleeping on a wait queue.
 314  * By always appending tasks to the list we ensure FIFO behavior.
 315  * NB: An RPC task will only receive interrupt-driven events as long
 316  * as it's on a wait queue.
 317  */
 318 static void __rpc_sleep_on(struct rpc_wait_queue *q, struct rpc_task *task,
 319                         rpc_action action)
 320 {
 321         dprintk("RPC: %5u sleep_on(queue \"%s\" time %lu)\n",
 322                         task->tk_pid, rpc_qname(q), jiffies);
 323
 324         if (!RPC_IS_ASYNC(task) && !RPC_IS_ACTIVATED(task)) {
 325                 printk(KERN_ERR "RPC: Inactive synchronous task put to sleep!\n");
 326                 return;
 327         }
 328
 329         __rpc_add_wait_queue(q, task);
 330
 331         BUG_ON(task->tk_callback != NULL);
 332         task->tk_callback = action;
 333         __rpc_add_timer(q, task);
 334 }
 335
 336 void rpc_sleep_on(struct rpc_wait_queue *q, struct rpc_task *task,
 337                                 rpc_action action)
 338 {
 339         /* Mark the task as being activated if so needed */
 340         rpc_set_active(task);
 341
 342         /*
 343          * Protect the queue operations.
 344          */
 345         spin_lock_bh(&q->lock);
 346         __rpc_sleep_on(q, task, action);
 347         spin_unlock_bh(&q->lock);
 348 }
 349 EXPORT_SYMBOL_GPL(rpc_sleep_on);
 350
 351 /**
 352  * __rpc_do_wake_up_task - wake up a single rpc_task
 353  * @queue: wait queue
 354  * @task: task to be woken up
 355  *
 356  * Caller must hold queue->lock, and have cleared the task queued flag.
 357  */
 358 static void __rpc_do_wake_up_task(struct rpc_wait_queue *queue, struct rpc_task *task)
 359 {
 360         dprintk("RPC: %5u __rpc_wake_up_task (now %lu)\n",
 361                         task->tk_pid, jiffies);
 362
 363 #ifdef RPC_DEBUG
 364         BUG_ON(task->tk_magic != RPC_TASK_MAGIC_ID);
 365 #endif
 366         /* Has the task been executed yet? If not, we cannot wake it up! */
 367         if (!RPC_IS_ACTIVATED(task)) {
 368                 printk(KERN_ERR "RPC: Inactive task (%p) being woken up!\n", task);
 369                 return;
 370         }
 371
 372         __rpc_remove_wait_queue(queue, task);
 373
 374         rpc_make_runnable(task);
 375
 376         dprintk("RPC:       __rpc_wake_up_task done\n");
 377 }
 378
 379 /*
 380  * Wake up a queued task while the queue lock is being held
 381  */
 382 static void rpc_wake_up_task_queue_locked(struct rpc_wait_queue *queue, struct rpc_task *task)
 383 {
 384         if (RPC_IS_QUEUED(task) && task->tk_waitqueue == queue)
 385                 __rpc_do_wake_up_task(queue, task);
 386 }
 387
 388 /*
 389  * Tests whether rpc queue is empty
 390  */
 391 int rpc_queue_empty(struct rpc_wait_queue *queue)
 392 {
 393         int res;
 394
 395         spin_lock_bh(&queue->lock);
 396         res = queue->qlen;
 397         spin_unlock_bh(&queue->lock);
 398         return (res == 0);
 399 }
 400 EXPORT_SYMBOL_GPL(rpc_queue_empty);
 401
 402 /*
 403  * Wake up a task on a specific queue
 404  */
 405 void rpc_wake_up_queued_task(struct rpc_wait_queue *queue, struct rpc_task *task)
 406 {
 407         spin_lock_bh(&queue->lock);
 408         rpc_wake_up_task_queue_locked(queue, task);
 409         spin_unlock_bh(&queue->lock);
 410 }
 411 EXPORT_SYMBOL_GPL(rpc_wake_up_queued_task);
 412
 413 /*
 414  * Wake up the specified task
 415  */
 416 static void rpc_wake_up_task(struct rpc_task *task)
 417 {
 418         rpc_wake_up_queued_task(task->tk_waitqueue, task);
 419 }
 420
 421 /*
 422  * Wake up the next task on a priority queue.
 423  */
 424 static struct rpc_task * __rpc_wake_up_next_priority(struct rpc_wait_queue *queue)
 425 {
 426         struct list_head *q;
 427         struct rpc_task *task;
 428
 429         /*
 430          * Service a batch of tasks from a single owner.
 431          */
 432         q = &queue->tasks[queue->priority];
 433         if (!list_empty(q)) {
 434                 task = list_entry(q->next, struct rpc_task, u.tk_wait.list);
 435                 if (queue->owner == task->tk_owner) {
 436                         if (--queue->nr)
 437                                 goto out;
 438                         list_move_tail(&task->u.tk_wait.list, q);
 439                 }
 440                 /*
 441                  * Check if we need to switch queues.
 442                  */
 443                 if (--queue->count)
 444                         goto new_owner;
 445         }
 446
 447         /*
 448          * Service the next queue.
 449          */
 450         do {
 451                 if (q == &queue->tasks[0])
 452                         q = &queue->tasks[queue->maxpriority];
 453                 else
 454                         q = q - 1;
 455                 if (!list_empty(q)) {
 456                         task = list_entry(q->next, struct rpc_task, u.tk_wait.list);
 457                         goto new_queue;
 458                 }
 459         } while (q != &queue->tasks[queue->priority]);
 460
 461         rpc_reset_waitqueue_priority(queue);
 462         return NULL;
 463
 464 new_queue:
 465         rpc_set_waitqueue_priority(queue, (unsigned int)(q - &queue->tasks[0]));
 466 new_owner:
 467         rpc_set_waitqueue_owner(queue, task->tk_owner);
 468 out:
 469         rpc_wake_up_task_queue_locked(queue, task);
 470         return task;
 471 }
 472
 473 /*
 474  * Wake up the next task on the wait queue.
 475  */
 476 struct rpc_task * rpc_wake_up_next(struct rpc_wait_queue *queue)
 477 {
 478         struct rpc_task *task = NULL;
 479
 480         dprintk("RPC:       wake_up_next(%p \"%s\")\n",
 481                         queue, rpc_qname(queue));
 482         spin_lock_bh(&queue->lock);
 483         if (RPC_IS_PRIORITY(queue))
 484                 task = __rpc_wake_up_next_priority(queue);
 485         else {
 486                 task_for_first(task, &queue->tasks[0])
 487                         rpc_wake_up_task_queue_locked(queue, task);
 488         }
 489         spin_unlock_bh(&queue->lock);
 490
 491         return task;
 492 }
 493 EXPORT_SYMBOL_GPL(rpc_wake_up_next);
 494
 495 /**
 496  * rpc_wake_up - wake up all rpc_tasks
 497  * @queue: rpc_wait_queue on which the tasks are sleeping
 498  *
 499  * Grabs queue->lock
 500  */
 501 void rpc_wake_up(struct rpc_wait_queue *queue)
 502 {
 503         struct rpc_task *task, *next;
 504         struct list_head *head;
 505
 506         spin_lock_bh(&queue->lock);
 507         head = &queue->tasks[queue->maxpriority];
 508         for (;;) {
 509                 list_for_each_entry_safe(task, next, head, u.tk_wait.list)
 510                         rpc_wake_up_task_queue_locked(queue, task);
 511                 if (head == &queue->tasks[0])
 512                         break;
 513                 head--;
 514         }
 515         spin_unlock_bh(&queue->lock);
 516 }
 517 EXPORT_SYMBOL_GPL(rpc_wake_up);
 518
 519 /**
 520  * rpc_wake_up_status - wake up all rpc_tasks and set their status value.
 521  * @queue: rpc_wait_queue on which the tasks are sleeping
 522  * @status: status value to set
 523  *
 524  * Grabs queue->lock
 525  */
 526 void rpc_wake_up_status(struct rpc_wait_queue *queue, int status)
 527 {
 528         struct rpc_task *task, *next;
 529         struct list_head *head;
 530
 531         spin_lock_bh(&queue->lock);
 532         head = &queue->tasks[queue->maxpriority];
 533         for (;;) {
 534                 list_for_each_entry_safe(task, next, head, u.tk_wait.list) {
 535                         task->tk_status = status;
 536                         rpc_wake_up_task_queue_locked(queue, task);
 537                 }
 538                 if (head == &queue->tasks[0])
 539                         break;
 540                 head--;
 541         }
 542         spin_unlock_bh(&queue->lock);
 543 }
 544 EXPORT_SYMBOL_GPL(rpc_wake_up_status);
 545
 546 static void __rpc_queue_timer_fn(unsigned long ptr)
 547 {
 548         struct rpc_wait_queue *queue = (struct rpc_wait_queue *)ptr;
 549         struct rpc_task *task, *n;
 550         unsigned long expires, now, timeo;
 551
 552         spin_lock(&queue->lock);
 553         expires = now = jiffies;
 554         list_for_each_entry_safe(task, n, &queue->timer_list.list, u.tk_wait.timer_list) {
 555                 timeo = task->u.tk_wait.expires;
 556                 if (time_after_eq(now, timeo)) {
 557                         dprintk("RPC: %5u timeout\n", task->tk_pid);
 558                         task->tk_status = -ETIMEDOUT;
 559                         rpc_wake_up_task_queue_locked(queue, task);
 560                         continue;
 561                 }
 562                 if (expires == now || time_after(expires, timeo))
 563                         expires = timeo;
 564         }
 565         if (!list_empty(&queue->timer_list.list))
 566                 rpc_set_queue_timer(queue, expires);
 567         spin_unlock(&queue->lock);
 568 }
 569
 570 static void __rpc_atrun(struct rpc_task *task)
 571 {
 572         task->tk_status = 0;
 573 }
 574
 575 /*
 576  * Run a task at a later time
 577  */
 578 void rpc_delay(struct rpc_task *task, unsigned long delay)
 579 {
 580         task->tk_timeout = delay;
 581         rpc_sleep_on(&delay_queue, task, __rpc_atrun);
 582 }
 583 EXPORT_SYMBOL_GPL(rpc_delay);
 584
 585 /*
 586  * Helper to call task->tk_ops->rpc_call_prepare
 587  */
 588 void rpc_prepare_task(struct rpc_task *task)
 589 {
 590         task->tk_ops->rpc_call_prepare(task, task->tk_calldata);
 591 }
 592
 593 /*
 594  * Helper that calls task->tk_ops->rpc_call_done if it exists
 595  */
 596 void rpc_exit_task(struct rpc_task *task)
 597 {
 598         task->tk_action = NULL;
 599         if (task->tk_ops->rpc_call_done != NULL) {
 600                 task->tk_ops->rpc_call_done(task, task->tk_calldata);
 601                 if (task->tk_action != NULL) {
 602                         WARN_ON(RPC_ASSASSINATED(task));
 603                         /* Always release the RPC slot and buffer memory */
 604                         xprt_release(task);
 605                 }
 606         }
 607 }
 608 EXPORT_SYMBOL_GPL(rpc_exit_task);
 609
 610 void rpc_release_calldata(const struct rpc_call_ops *ops, void *calldata)
 611 {
 612         if (ops->rpc_release != NULL)
 613                 ops->rpc_release(calldata);
 614 }
 615
 616 /*
 617  * This is the RPC `scheduler' (or rather, the finite state machine).
 618  */
 619 static void __rpc_execute(struct rpc_task *task)
 620 {
 621         struct rpc_wait_queue *queue;
 622         int task_is_async = RPC_IS_ASYNC(task);
 623         int status = 0;
 624
 625         dprintk("RPC: %5u __rpc_execute flags=0x%x\n",
 626                         task->tk_pid, task->tk_flags);
 627
 628         BUG_ON(RPC_IS_QUEUED(task));
 629
 630         for (;;) {
 631
 632                 /*
 633                  * Execute any pending callback.
 634                  */
 635                 if (task->tk_callback) {
 636                         void (*save_callback)(struct rpc_task *);
 637
 638                         /*
 639                          * We set tk_callback to NULL before calling it,
 640                          * in case it sets the tk_callback field itself:
 641                          */
 642                         save_callback = task->tk_callback;
 643                         task->tk_callback = NULL;
 644                         save_callback(task);
 645                 }
 646
 647                 /*
 648                  * Perform the next FSM step.
 649                  * tk_action may be NULL when the task has been killed
 650                  * by someone else.
 651                  */
 652                 if (!RPC_IS_QUEUED(task)) {
 653                         if (task->tk_action == NULL)
 654                                 break;
 655                         task->tk_action(task);
 656                 }
 657
 658                 /*
 659                  * Lockless check for whether task is sleeping or not.
 660                  */
 661                 if (!RPC_IS_QUEUED(task))
 662                         continue;
 663                 /*
 664                  * The queue->lock protects against races with
 665                  * rpc_make_runnable().
 666                  *
 667                  * Note that once we clear RPC_TASK_RUNNING on an asynchronous
 668                  * rpc_task, rpc_make_runnable() can assign it to a
 669                  * different workqueue. We therefore cannot assume that the
 670                  * rpc_task pointer may still be dereferenced.
 671                  */
 672                 queue = task->tk_waitqueue;
 673                 spin_lock_bh(&queue->lock);
 674                 if (!RPC_IS_QUEUED(task)) {
 675                         spin_unlock_bh(&queue->lock);
 676                         continue;
 677                 }
 678                 rpc_clear_running(task);
 679                 spin_unlock_bh(&queue->lock);
 680                 if (task_is_async)
 681                         return;
 682
 683                 /* sync task: sleep here */
 684                 dprintk("RPC: %5u sync task going to sleep\n", task->tk_pid);
 685                 status = out_of_line_wait_on_bit(&task->tk_runstate,
 686                                 RPC_TASK_QUEUED, rpc_wait_bit_killable,
 687                                 TASK_KILLABLE);
 688                 if (status == -ERESTARTSYS) {
 689                         /*
 690                          * When a sync task receives a signal, it exits with
 691                          * -ERESTARTSYS. In order to catch any callbacks that
 692                          * clean up after sleeping on some queue, we don't
 693                          * break the loop here, but go around once more.
 694                          */
 695                         dprintk("RPC: %5u got signal\n", task->tk_pid);
 696                         task->tk_flags |= RPC_TASK_KILLED;
 697                         rpc_exit(task, -ERESTARTSYS);
 698                         rpc_wake_up_task(task);
 699                 }
 700                 rpc_set_running(task);
 701                 dprintk("RPC: %5u sync task resuming\n", task->tk_pid);
 702         }
 703
 704         dprintk("RPC: %5u return %d, status %d\n", task->tk_pid, status,
 705                         task->tk_status);
 706         /* Release all resources associated with the task */
 707         rpc_release_task(task);
 708 }
 709
 710 /*
 711  * User-visible entry point to the scheduler.
 712  *
 713  * This may be called recursively if e.g. an async NFS task updates
 714  * the attributes and finds that dirty pages must be flushed.
 715  * NOTE: Upon exit of this function the task is guaranteed to be
 716  *       released. In particular note that tk_release() will have
 717  *       been called, so your task memory may have been freed.
 718  */
 719 void rpc_execute(struct rpc_task *task)
 720 {
 721         rpc_set_active(task);
 722         rpc_set_running(task);
 723         __rpc_execute(task);
 724 }
 725
 726 static void rpc_async_schedule(struct work_struct *work)
 727 {
 728         __rpc_execute(container_of(work, struct rpc_task, u.tk_work));
 729 }
 730
 731 /**
 732  * rpc_malloc - allocate an RPC buffer
 733  * @task: RPC task that will use this buffer
 734  * @size: requested byte size
 735  *
 736  * To prevent rpciod from hanging, this allocator never sleeps,
 737  * returning NULL if the request cannot be serviced immediately.
 738  * The caller can arrange to sleep in a way that is safe for rpciod.
 739  *
 740  * Most requests are 'small' (under 2KiB) and can be serviced from a
 741  * mempool, ensuring that NFS reads and writes can always proceed,
 742  * and that there is good locality of reference for these buffers.
 743  *
 744  * In order to avoid memory starvation triggering more writebacks of
 745  * NFS requests, we avoid using GFP_KERNEL.
 746  */
 747 void *rpc_malloc(struct rpc_task *task, size_t size)
 748 {
 749         struct rpc_buffer *buf;
 750         gfp_t gfp = RPC_IS_SWAPPER(task) ? GFP_ATOMIC : GFP_NOWAIT;
 751
 752         size += sizeof(struct rpc_buffer);
 753         if (size <= RPC_BUFFER_MAXSIZE)
 754                 buf = mempool_alloc(rpc_buffer_mempool, gfp);
 755         else
 756                 buf = kmalloc(size, gfp);
 757
 758         if (!buf)
 759                 return NULL;
 760
 761         buf->len = size;
 762         dprintk("RPC: %5u allocated buffer of size %zu at %p\n",
 763                         task->tk_pid, size, buf);
 764         return &buf->data;
 765 }
 766 EXPORT_SYMBOL_GPL(rpc_malloc);
 767
 768 /**
 769  * rpc_free - free buffer allocated via rpc_malloc
 770  * @buffer: buffer to free
 771  *
 772  */
 773 void rpc_free(void *buffer)
 774 {
 775         size_t size;
 776         struct rpc_buffer *buf;
 777
 778         if (!buffer)
 779                 return;
 780
 781         buf = container_of(buffer, struct rpc_buffer, data);
 782         size = buf->len;
 783
 784         dprintk("RPC:       freeing buffer of size %zu at %p\n",
 785                         size, buf);
 786
 787         if (size <= RPC_BUFFER_MAXSIZE)
 788                 mempool_free(buf, rpc_buffer_mempool);
 789         else
 790                 kfree(buf);
 791 }
 792 EXPORT_SYMBOL_GPL(rpc_free);
 793
 794 /*
 795  * Creation and deletion of RPC task structures
 796  */
 797 static void rpc_init_task(struct rpc_task *task, const struct rpc_task_setup *task_setup_data)
 798 {
 799         memset(task, 0, sizeof(*task));
 800         atomic_set(&task->tk_count, 1);
 801         task->tk_flags  = task_setup_data->flags;
 802         task->tk_ops = task_setup_data->callback_ops;
 803         task->tk_calldata = task_setup_data->callback_data;
 804         INIT_LIST_HEAD(&task->tk_task);
 805
 806         /* Initialize retry counters */
 807         task->tk_garb_retry = 2;
 808         task->tk_cred_retry = 2;
 809
 810         task->tk_priority = task_setup_data->priority - RPC_PRIORITY_LOW;
 811         task->tk_owner = current->tgid;
 812
 813         /* Initialize workqueue for async tasks */
 814         task->tk_workqueue = task_setup_data->workqueue;
 815
 816         task->tk_client = task_setup_data->rpc_client;
 817         if (task->tk_client != NULL) {
 818                 kref_get(&task->tk_client->cl_kref);
 819                 if (task->tk_client->cl_softrtry)
 820                         task->tk_flags |= RPC_TASK_SOFT;
 821         }
 822
 823         if (task->tk_ops->rpc_call_prepare != NULL)
 824                 task->tk_action = rpc_prepare_task;
 825
 826         if (task_setup_data->rpc_message != NULL) {
 827                 task->tk_msg.rpc_proc = task_setup_data->rpc_message->rpc_proc;
 828                 task->tk_msg.rpc_argp = task_setup_data->rpc_message->rpc_argp;
 829                 task->tk_msg.rpc_resp = task_setup_data->rpc_message->rpc_resp;
 830                 /* Bind the user cred */
 831                 rpcauth_bindcred(task, task_setup_data->rpc_message->rpc_cred, task_setup_data->flags);
 832                 if (task->tk_action == NULL)
 833                         rpc_call_start(task);
 834         }
 835
 836         /* starting timestamp */
 837         task->tk_start = ktime_get();
 838
 839         dprintk("RPC:       new task initialized, procpid %u\n",
 840                                 task_pid_nr(current));
 841 }
 842
 843 static struct rpc_task *
 844 rpc_alloc_task(void)
 845 {
 846         return (struct rpc_task *)mempool_alloc(rpc_task_mempool, GFP_NOFS);
 847 }
 848
 849 /*
 850  * Create a new task for the specified client.
 851  */
 852 struct rpc_task *rpc_new_task(const struct rpc_task_setup *setup_data)
 853 {
 854         struct rpc_task *task = setup_data->task;
 855         unsigned short flags = 0;
 856
 857         if (task == NULL) {
 858                 task = rpc_alloc_task();
 859                 if (task == NULL) {
 860                         rpc_release_calldata(setup_data->callback_ops,
 861                                         setup_data->callback_data);
 862                         return ERR_PTR(-ENOMEM);
 863                 }
 864                 flags = RPC_TASK_DYNAMIC;
 865         }
 866
 867         rpc_init_task(task, setup_data);
 868         if (task->tk_status < 0) {
 869                 int err = task->tk_status;
 870                 rpc_put_task(task);
 871                 return ERR_PTR(err);
 872         }
 873
 874         task->tk_flags |= flags;
 875         dprintk("RPC:       allocated task %p\n", task);
 876         return task;
 877 }
 878
 879 static void rpc_free_task(struct rpc_task *task)
 880 {
 881         const struct rpc_call_ops *tk_ops = task->tk_ops;
 882         void *calldata = task->tk_calldata;
 883
 884         if (task->tk_flags & RPC_TASK_DYNAMIC) {
 885                 dprintk("RPC: %5u freeing task\n", task->tk_pid);
 886                 mempool_free(task, rpc_task_mempool);
 887         }
 888         rpc_release_calldata(tk_ops, calldata);
 889 }
 890
 891 static void rpc_async_release(struct work_struct *work)
 892 {
 893         rpc_free_task(container_of(work, struct rpc_task, u.tk_work));
 894 }
 895
 896 void rpc_put_task(struct rpc_task *task)
 897 {
 898         if (!atomic_dec_and_test(&task->tk_count))
 899                 return;
 900         /* Release resources */
 901         if (task->tk_rqstp)
 902                 xprt_release(task);
 903         if (task->tk_msg.rpc_cred)
 904                 rpcauth_unbindcred(task);
 905         if (task->tk_client) {
 906                 rpc_release_client(task->tk_client);
 907                 task->tk_client = NULL;
 908         }
 909         if (task->tk_workqueue != NULL) {
 910                 INIT_WORK(&task->u.tk_work, rpc_async_release);
 911                 queue_work(task->tk_workqueue, &task->u.tk_work);
 912         } else
 913                 rpc_free_task(task);
 914 }
 915 EXPORT_SYMBOL_GPL(rpc_put_task);
 916
 917 static void rpc_release_task(struct rpc_task *task)
 918 {
 919 #ifdef RPC_DEBUG
 920         BUG_ON(task->tk_magic != RPC_TASK_MAGIC_ID);
 921 #endif
 922         dprintk("RPC: %5u release task\n", task->tk_pid);
 923
 924         if (!list_empty(&task->tk_task)) {
 925                 struct rpc_clnt *clnt = task->tk_client;
 926                 /* Remove from client task list */
 927                 spin_lock(&clnt->cl_lock);
 928                 list_del(&task->tk_task);
 929                 spin_unlock(&clnt->cl_lock);
 930         }
 931         BUG_ON (RPC_IS_QUEUED(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  * Kill all tasks for the given client.
 944  * XXX: kill their descendants as well?
 945  */
 946 void rpc_killall_tasks(struct rpc_clnt *clnt)
 947 {
 948         struct rpc_task *rovr;
 949
 950
 951         if (list_empty(&clnt->cl_tasks))
 952                 return;
 953         dprintk("RPC:       killing all tasks for client %p\n", clnt);
 954         /*
 955          * Spin lock all_tasks to prevent changes...
 956          */
 957         spin_lock(&clnt->cl_lock);
 958         list_for_each_entry(rovr, &clnt->cl_tasks, tk_task) {
 959                 if (! RPC_IS_ACTIVATED(rovr))
 960                         continue;
 961                 if (!(rovr->tk_flags & RPC_TASK_KILLED)) {
 962                         rovr->tk_flags |= RPC_TASK_KILLED;
 963                         rpc_exit(rovr, -EIO);
 964                         rpc_wake_up_task(rovr);
 965                 }
 966         }
 967         spin_unlock(&clnt->cl_lock);
 968 }
 969 EXPORT_SYMBOL_GPL(rpc_killall_tasks);
 970
 971 int rpciod_up(void)
 972 {
 973         return try_module_get(THIS_MODULE) ? 0 : -EINVAL;
 974 }
 975
 976 void rpciod_down(void)
 977 {
 978         module_put(THIS_MODULE);
 979 }
 980
 981 /*
 982  * Start up the rpciod workqueue.
 983  */
 984 static int rpciod_start(void)
 985 {
 986         struct workqueue_struct *wq;
 987
 988         /*
 989          * Create the rpciod thread and wait for it to start.
 990          */
 991         dprintk("RPC:       creating workqueue rpciod\n");
 992         wq = create_workqueue("rpciod");
 993         rpciod_workqueue = wq;
 994         return rpciod_workqueue != NULL;
 995 }
 996
 997 static void rpciod_stop(void)
 998 {
 999         struct workqueue_struct *wq = NULL;
1000
1001         if (rpciod_workqueue == NULL)
1002                 return;
1003         dprintk("RPC:       destroying workqueue rpciod\n");
1004
1005         wq = rpciod_workqueue;
1006         rpciod_workqueue = NULL;
1007         destroy_workqueue(wq);
1008 }
1009
1010 void
1011 rpc_destroy_mempool(void)
1012 {
1013         rpciod_stop();
1014         if (rpc_buffer_mempool)
1015                 mempool_destroy(rpc_buffer_mempool);
1016         if (rpc_task_mempool)
1017                 mempool_destroy(rpc_task_mempool);
1018         if (rpc_task_slabp)
1019                 kmem_cache_destroy(rpc_task_slabp);
1020         if (rpc_buffer_slabp)
1021                 kmem_cache_destroy(rpc_buffer_slabp);
1022         rpc_destroy_wait_queue(&delay_queue);
1023 }
1024
1025 int
1026 rpc_init_mempool(void)
1027 {
1028         /*
1029          * The following is not strictly a mempool initialisation,
1030          * but there is no harm in doing it here
1031          */
1032         rpc_init_wait_queue(&delay_queue, "delayq");
1033         if (!rpciod_start())
1034                 goto err_nomem;
1035
1036         rpc_task_slabp = kmem_cache_create("rpc_tasks",
1037                                              sizeof(struct rpc_task),
1038                                              0, SLAB_HWCACHE_ALIGN,
1039                                              NULL);
1040         if (!rpc_task_slabp)
1041                 goto err_nomem;
1042         rpc_buffer_slabp = kmem_cache_create("rpc_buffers",
1043                                              RPC_BUFFER_MAXSIZE,
1044                                              0, SLAB_HWCACHE_ALIGN,
1045                                              NULL);
1046         if (!rpc_buffer_slabp)
1047                 goto err_nomem;
1048         rpc_task_mempool = mempool_create_slab_pool(RPC_TASK_POOLSIZE,
1049                                                     rpc_task_slabp);
1050         if (!rpc_task_mempool)
1051                 goto err_nomem;
1052         rpc_buffer_mempool = mempool_create_slab_pool(RPC_BUFFER_POOLSIZE,
1053                                                       rpc_buffer_slabp);
1054         if (!rpc_buffer_mempool)
1055                 goto err_nomem;
1056         return 0;
1057 err_nomem:
1058         rpc_destroy_mempool();
1059         return -ENOMEM;
1060 }