kernel/2.6.29.6-aldebaran-rt/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_wait_queue *queue, 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         if (list_empty(&queue->timer_list.list))
  69                 del_timer(&queue->timer_list.timer);
  70 }
  71
  72 static void
  73 rpc_set_queue_timer(struct rpc_wait_queue *queue, unsigned long expires)
  74 {
  75         queue->timer_list.expires = expires;
  76         mod_timer(&queue->timer_list.timer, expires);
  77 }
  78
  79 /*
  80  * Set up a timer for the current task.
  81  */
  82 static void
  83 __rpc_add_timer(struct rpc_wait_queue *queue, struct rpc_task *task)
  84 {
  85         if (!task->tk_timeout)
  86                 return;
  87
  88         dprintk("RPC: %5u setting alarm for %lu ms\n",
  89                         task->tk_pid, task->tk_timeout * 1000 / HZ);
  90
  91         task->u.tk_wait.expires = jiffies + task->tk_timeout;
  92         if (list_empty(&queue->timer_list.list) || time_before(task->u.tk_wait.expires, queue->timer_list.expires))
  93                 rpc_set_queue_timer(queue, task->u.tk_wait.expires);
  94         list_add(&task->u.tk_wait.timer_list, &queue->timer_list.list);
  95 }
  96
  97 /*
  98  * Add new request to a priority queue.
  99  */
 100 static void __rpc_add_wait_queue_priority(struct rpc_wait_queue *queue, struct rpc_task *task)
 101 {
 102         struct list_head *q;
 103         struct rpc_task *t;
 104
 105         INIT_LIST_HEAD(&task->u.tk_wait.links);
 106         q = &queue->tasks[task->tk_priority];
 107         if (unlikely(task->tk_priority > queue->maxpriority))
 108                 q = &queue->tasks[queue->maxpriority];
 109         list_for_each_entry(t, q, u.tk_wait.list) {
 110                 if (t->tk_owner == task->tk_owner) {
 111                         list_add_tail(&task->u.tk_wait.list, &t->u.tk_wait.links);
 112                         return;
 113                 }
 114         }
 115         list_add_tail(&task->u.tk_wait.list, q);
 116 }
 117
 118 /*
 119  * Add new request to wait queue.
 120  *
 121  * Swapper tasks always get inserted at the head of the queue.
 122  * This should avoid many nasty memory deadlocks and hopefully
 123  * improve overall performance.
 124  * Everyone else gets appended to the queue to ensure proper FIFO behavior.
 125  */
 126 static void __rpc_add_wait_queue(struct rpc_wait_queue *queue, struct rpc_task *task)
 127 {
 128         BUG_ON (RPC_IS_QUEUED(task));
 129
 130         if (RPC_IS_PRIORITY(queue))
 131                 __rpc_add_wait_queue_priority(queue, task);
 132         else if (RPC_IS_SWAPPER(task))
 133                 list_add(&task->u.tk_wait.list, &queue->tasks[0]);
 134         else
 135                 list_add_tail(&task->u.tk_wait.list, &queue->tasks[0]);
 136         task->tk_waitqueue = queue;
 137         queue->qlen++;
 138         rpc_set_queued(task);
 139
 140         dprintk("RPC: %5u added to queue %p \"%s\"\n",
 141                         task->tk_pid, queue, rpc_qname(queue));
 142 }
 143
 144 /*
 145  * Remove request from a priority queue.
 146  */
 147 static void __rpc_remove_wait_queue_priority(struct rpc_task *task)
 148 {
 149         struct rpc_task *t;
 150
 151         if (!list_empty(&task->u.tk_wait.links)) {
 152                 t = list_entry(task->u.tk_wait.links.next, struct rpc_task, u.tk_wait.list);
 153                 list_move(&t->u.tk_wait.list, &task->u.tk_wait.list);
 154                 list_splice_init(&task->u.tk_wait.links, &t->u.tk_wait.links);
 155         }
 156 }
 157
 158 /*
 159  * Remove request from queue.
 160  * Note: must be called with spin lock held.
 161  */
 162 static void __rpc_remove_wait_queue(struct rpc_wait_queue *queue, struct rpc_task *task)
 163 {
 164         __rpc_disable_timer(queue, task);
 165         if (RPC_IS_PRIORITY(queue))
 166                 __rpc_remove_wait_queue_priority(task);
 167         list_del(&task->u.tk_wait.list);
 168         queue->qlen--;
 169         dprintk("RPC: %5u removed from queue %p \"%s\"\n",
 170                         task->tk_pid, queue, rpc_qname(queue));
 171 }
 172
 173 static inline void rpc_set_waitqueue_priority(struct rpc_wait_queue *queue, int priority)
 174 {
 175         queue->priority = priority;
 176         queue->count = 1 << (priority * 2);
 177 }
 178
 179 static inline void rpc_set_waitqueue_owner(struct rpc_wait_queue *queue, pid_t pid)
 180 {
 181         queue->owner = pid;
 182         queue->nr = RPC_BATCH_COUNT;
 183 }
 184
 185 static inline void rpc_reset_waitqueue_priority(struct rpc_wait_queue *queue)
 186 {
 187         rpc_set_waitqueue_priority(queue, queue->maxpriority);
 188         rpc_set_waitqueue_owner(queue, 0);
 189 }
 190
 191 static void __rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname, unsigned char nr_queues)
 192 {
 193         int i;
 194
 195         spin_lock_init(&queue->lock);
 196         for (i = 0; i < ARRAY_SIZE(queue->tasks); i++)
 197                 INIT_LIST_HEAD(&queue->tasks[i]);
 198         queue->maxpriority = nr_queues - 1;
 199         rpc_reset_waitqueue_priority(queue);
 200         queue->qlen = 0;
 201         setup_timer(&queue->timer_list.timer, __rpc_queue_timer_fn, (unsigned long)queue);
 202         INIT_LIST_HEAD(&queue->timer_list.list);
 203 #ifdef RPC_DEBUG
 204         queue->name = qname;
 205 #endif
 206 }
 207
 208 void rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname)
 209 {
 210         __rpc_init_priority_wait_queue(queue, qname, RPC_NR_PRIORITY);
 211 }
 212
 213 void rpc_init_wait_queue(struct rpc_wait_queue *queue, const char *qname)
 214 {
 215         __rpc_init_priority_wait_queue(queue, qname, 1);
 216 }
 217 EXPORT_SYMBOL_GPL(rpc_init_wait_queue);
 218
 219 void rpc_destroy_wait_queue(struct rpc_wait_queue *queue)
 220 {
 221         del_timer_sync(&queue->timer_list.timer);
 222 }
 223 EXPORT_SYMBOL_GPL(rpc_destroy_wait_queue);
 224
 225 static int rpc_wait_bit_killable(void *word)
 226 {
 227         if (fatal_signal_pending(current))
 228                 return -ERESTARTSYS;
 229         schedule();
 230         return 0;
 231 }
 232
 233 #ifdef RPC_DEBUG
 234 static void rpc_task_set_debuginfo(struct rpc_task *task)
 235 {
 236         static atomic_t rpc_pid;
 237
 238         task->tk_magic = RPC_TASK_MAGIC_ID;
 239         task->tk_pid = atomic_inc_return(&rpc_pid);
 240 }
 241 #else
 242 static inline void rpc_task_set_debuginfo(struct rpc_task *task)
 243 {
 244 }
 245 #endif
 246
 247 static void rpc_set_active(struct rpc_task *task)
 248 {
 249         struct rpc_clnt *clnt;
 250         if (test_and_set_bit(RPC_TASK_ACTIVE, &task->tk_runstate) != 0)
 251                 return;
 252         rpc_task_set_debuginfo(task);
 253         /* Add to global list of all tasks */
 254         clnt = task->tk_client;
 255         if (clnt != NULL) {
 256                 spin_lock(&clnt->cl_lock);
 257                 list_add_tail(&task->tk_task, &clnt->cl_tasks);
 258                 spin_unlock(&clnt->cl_lock);
 259         }
 260 }
 261
 262 /*
 263  * Mark an RPC call as having completed by clearing the 'active' bit
 264  */
 265 static void rpc_mark_complete_task(struct rpc_task *task)
 266 {
 267         smp_mb__before_clear_bit();
 268         clear_bit(RPC_TASK_ACTIVE, &task->tk_runstate);
 269         smp_mb__after_clear_bit();
 270         wake_up_bit(&task->tk_runstate, RPC_TASK_ACTIVE);
 271 }
 272
 273 /*
 274  * Allow callers to wait for completion of an RPC call
 275  */
 276 int __rpc_wait_for_completion_task(struct rpc_task *task, int (*action)(void *))
 277 {
 278         if (action == NULL)
 279                 action = rpc_wait_bit_killable;
 280         return wait_on_bit(&task->tk_runstate, RPC_TASK_ACTIVE,
 281                         action, TASK_KILLABLE);
 282 }
 283 EXPORT_SYMBOL_GPL(__rpc_wait_for_completion_task);
 284
 285 /*
 286  * Make an RPC task runnable.
 287  *
 288  * Note: If the task is ASYNC, this must be called with
 289  * the spinlock held to protect the wait queue operation.
 290  */
 291 static void rpc_make_runnable(struct rpc_task *task)
 292 {
 293         rpc_clear_queued(task);
 294         if (rpc_test_and_set_running(task))
 295                 return;
 296         if (RPC_IS_ASYNC(task)) {
 297                 int status;
 298
 299                 INIT_WORK(&task->u.tk_work, rpc_async_schedule);
 300                 status = queue_work(rpciod_workqueue, &task->u.tk_work);
 301                 if (status < 0) {
 302                         printk(KERN_WARNING "RPC: failed to add task to queue: error: %d!\n", status);
 303                         task->tk_status = status;
 304                         return;
 305                 }
 306         } else
 307                 wake_up_bit(&task->tk_runstate, RPC_TASK_QUEUED);
 308 }
 309
 310 /*
 311  * Prepare for sleeping on a wait queue.
 312  * By always appending tasks to the list we ensure FIFO behavior.
 313  * NB: An RPC task will only receive interrupt-driven events as long
 314  * as it's on a wait queue.
 315  */
 316 static void __rpc_sleep_on(struct rpc_wait_queue *q, struct rpc_task *task,
 317                         rpc_action action)
 318 {
 319         dprintk("RPC: %5u sleep_on(queue \"%s\" time %lu)\n",
 320                         task->tk_pid, rpc_qname(q), jiffies);
 321
 322         if (!RPC_IS_ASYNC(task) && !RPC_IS_ACTIVATED(task)) {
 323                 printk(KERN_ERR "RPC: Inactive synchronous task put to sleep!\n");
 324                 return;
 325         }
 326
 327         __rpc_add_wait_queue(q, task);
 328
 329         BUG_ON(task->tk_callback != NULL);
 330         task->tk_callback = action;
 331         __rpc_add_timer(q, task);
 332 }
 333
 334 void rpc_sleep_on(struct rpc_wait_queue *q, struct rpc_task *task,
 335                                 rpc_action action)
 336 {
 337         /* Mark the task as being activated if so needed */
 338         rpc_set_active(task);
 339
 340         /*
 341          * Protect the queue operations.
 342          */
 343         spin_lock_bh(&q->lock);
 344         __rpc_sleep_on(q, task, action);
 345         spin_unlock_bh(&q->lock);
 346 }
 347 EXPORT_SYMBOL_GPL(rpc_sleep_on);
 348
 349 /**
 350  * __rpc_do_wake_up_task - wake up a single rpc_task
 351  * @queue: wait queue
 352  * @task: task to be woken up
 353  *
 354  * Caller must hold queue->lock, and have cleared the task queued flag.
 355  */
 356 static void __rpc_do_wake_up_task(struct rpc_wait_queue *queue, struct rpc_task *task)
 357 {
 358         dprintk("RPC: %5u __rpc_wake_up_task (now %lu)\n",
 359                         task->tk_pid, jiffies);
 360
 361 #ifdef RPC_DEBUG
 362         BUG_ON(task->tk_magic != RPC_TASK_MAGIC_ID);
 363 #endif
 364         /* Has the task been executed yet? If not, we cannot wake it up! */
 365         if (!RPC_IS_ACTIVATED(task)) {
 366                 printk(KERN_ERR "RPC: Inactive task (%p) being woken up!\n", task);
 367                 return;
 368         }
 369
 370         __rpc_remove_wait_queue(queue, task);
 371
 372         rpc_make_runnable(task);
 373
 374         dprintk("RPC:       __rpc_wake_up_task done\n");
 375 }
 376
 377 /*
 378  * Wake up a queued task while the queue lock is being held
 379  */
 380 static void rpc_wake_up_task_queue_locked(struct rpc_wait_queue *queue, struct rpc_task *task)
 381 {
 382         if (RPC_IS_QUEUED(task) && task->tk_waitqueue == queue)
 383                 __rpc_do_wake_up_task(queue, task);
 384 }
 385
 386 /*
 387  * Wake up a task on a specific queue
 388  */
 389 void rpc_wake_up_queued_task(struct rpc_wait_queue *queue, struct rpc_task *task)
 390 {
 391         spin_lock_bh(&queue->lock);
 392         rpc_wake_up_task_queue_locked(queue, task);
 393         spin_unlock_bh(&queue->lock);
 394 }
 395 EXPORT_SYMBOL_GPL(rpc_wake_up_queued_task);
 396
 397 /*
 398  * Wake up the specified task
 399  */
 400 static void rpc_wake_up_task(struct rpc_task *task)
 401 {
 402         rpc_wake_up_queued_task(task->tk_waitqueue, task);
 403 }
 404
 405 /*
 406  * Wake up the next task on a priority queue.
 407  */
 408 static struct rpc_task * __rpc_wake_up_next_priority(struct rpc_wait_queue *queue)
 409 {
 410         struct list_head *q;
 411         struct rpc_task *task;
 412
 413         /*
 414          * Service a batch of tasks from a single owner.
 415          */
 416         q = &queue->tasks[queue->priority];
 417         if (!list_empty(q)) {
 418                 task = list_entry(q->next, struct rpc_task, u.tk_wait.list);
 419                 if (queue->owner == task->tk_owner) {
 420                         if (--queue->nr)
 421                                 goto out;
 422                         list_move_tail(&task->u.tk_wait.list, q);
 423                 }
 424                 /*
 425                  * Check if we need to switch queues.
 426                  */
 427                 if (--queue->count)
 428                         goto new_owner;
 429         }
 430
 431         /*
 432          * Service the next queue.
 433          */
 434         do {
 435                 if (q == &queue->tasks[0])
 436                         q = &queue->tasks[queue->maxpriority];
 437                 else
 438                         q = q - 1;
 439                 if (!list_empty(q)) {
 440                         task = list_entry(q->next, struct rpc_task, u.tk_wait.list);
 441                         goto new_queue;
 442                 }
 443         } while (q != &queue->tasks[queue->priority]);
 444
 445         rpc_reset_waitqueue_priority(queue);
 446         return NULL;
 447
 448 new_queue:
 449         rpc_set_waitqueue_priority(queue, (unsigned int)(q - &queue->tasks[0]));
 450 new_owner:
 451         rpc_set_waitqueue_owner(queue, task->tk_owner);
 452 out:
 453         rpc_wake_up_task_queue_locked(queue, task);
 454         return task;
 455 }
 456
 457 /*
 458  * Wake up the next task on the wait queue.
 459  */
 460 struct rpc_task * rpc_wake_up_next(struct rpc_wait_queue *queue)
 461 {
 462         struct rpc_task *task = NULL;
 463
 464         dprintk("RPC:       wake_up_next(%p \"%s\")\n",
 465                         queue, rpc_qname(queue));
 466         spin_lock_bh(&queue->lock);
 467         if (RPC_IS_PRIORITY(queue))
 468                 task = __rpc_wake_up_next_priority(queue);
 469         else {
 470                 task_for_first(task, &queue->tasks[0])
 471                         rpc_wake_up_task_queue_locked(queue, task);
 472         }
 473         spin_unlock_bh(&queue->lock);
 474
 475         return task;
 476 }
 477 EXPORT_SYMBOL_GPL(rpc_wake_up_next);
 478
 479 /**
 480  * rpc_wake_up - wake up all rpc_tasks
 481  * @queue: rpc_wait_queue on which the tasks are sleeping
 482  *
 483  * Grabs queue->lock
 484  */
 485 void rpc_wake_up(struct rpc_wait_queue *queue)
 486 {
 487         struct rpc_task *task, *next;
 488         struct list_head *head;
 489
 490         spin_lock_bh(&queue->lock);
 491         head = &queue->tasks[queue->maxpriority];
 492         for (;;) {
 493                 list_for_each_entry_safe(task, next, head, u.tk_wait.list)
 494                         rpc_wake_up_task_queue_locked(queue, task);
 495                 if (head == &queue->tasks[0])
 496                         break;
 497                 head--;
 498         }
 499         spin_unlock_bh(&queue->lock);
 500 }
 501 EXPORT_SYMBOL_GPL(rpc_wake_up);
 502
 503 /**
 504  * rpc_wake_up_status - wake up all rpc_tasks and set their status value.
 505  * @queue: rpc_wait_queue on which the tasks are sleeping
 506  * @status: status value to set
 507  *
 508  * Grabs queue->lock
 509  */
 510 void rpc_wake_up_status(struct rpc_wait_queue *queue, int status)
 511 {
 512         struct rpc_task *task, *next;
 513         struct list_head *head;
 514
 515         spin_lock_bh(&queue->lock);
 516         head = &queue->tasks[queue->maxpriority];
 517         for (;;) {
 518                 list_for_each_entry_safe(task, next, head, u.tk_wait.list) {
 519                         task->tk_status = status;
 520                         rpc_wake_up_task_queue_locked(queue, task);
 521                 }
 522                 if (head == &queue->tasks[0])
 523                         break;
 524                 head--;
 525         }
 526         spin_unlock_bh(&queue->lock);
 527 }
 528 EXPORT_SYMBOL_GPL(rpc_wake_up_status);
 529
 530 static void __rpc_queue_timer_fn(unsigned long ptr)
 531 {
 532         struct rpc_wait_queue *queue = (struct rpc_wait_queue *)ptr;
 533         struct rpc_task *task, *n;
 534         unsigned long expires, now, timeo;
 535
 536         spin_lock(&queue->lock);
 537         expires = now = jiffies;
 538         list_for_each_entry_safe(task, n, &queue->timer_list.list, u.tk_wait.timer_list) {
 539                 timeo = task->u.tk_wait.expires;
 540                 if (time_after_eq(now, timeo)) {
 541                         dprintk("RPC: %5u timeout\n", task->tk_pid);
 542                         task->tk_status = -ETIMEDOUT;
 543                         rpc_wake_up_task_queue_locked(queue, task);
 544                         continue;
 545                 }
 546                 if (expires == now || time_after(expires, timeo))
 547                         expires = timeo;
 548         }
 549         if (!list_empty(&queue->timer_list.list))
 550                 rpc_set_queue_timer(queue, expires);
 551         spin_unlock(&queue->lock);
 552 }
 553
 554 static void __rpc_atrun(struct rpc_task *task)
 555 {
 556         task->tk_status = 0;
 557 }
 558
 559 /*
 560  * Run a task at a later time
 561  */
 562 void rpc_delay(struct rpc_task *task, unsigned long delay)
 563 {
 564         task->tk_timeout = delay;
 565         rpc_sleep_on(&delay_queue, task, __rpc_atrun);
 566 }
 567 EXPORT_SYMBOL_GPL(rpc_delay);
 568
 569 /*
 570  * Helper to call task->tk_ops->rpc_call_prepare
 571  */
 572 static void rpc_prepare_task(struct rpc_task *task)
 573 {
 574         task->tk_ops->rpc_call_prepare(task, task->tk_calldata);
 575 }
 576
 577 /*
 578  * Helper that calls task->tk_ops->rpc_call_done if it exists
 579  */
 580 void rpc_exit_task(struct rpc_task *task)
 581 {
 582         task->tk_action = NULL;
 583         if (task->tk_ops->rpc_call_done != NULL) {
 584                 task->tk_ops->rpc_call_done(task, task->tk_calldata);
 585                 if (task->tk_action != NULL) {
 586                         WARN_ON(RPC_ASSASSINATED(task));
 587                         /* Always release the RPC slot and buffer memory */
 588                         xprt_release(task);
 589                 }
 590         }
 591 }
 592 EXPORT_SYMBOL_GPL(rpc_exit_task);
 593
 594 void rpc_release_calldata(const struct rpc_call_ops *ops, void *calldata)
 595 {
 596         if (ops->rpc_release != NULL)
 597                 ops->rpc_release(calldata);
 598 }
 599
 600 /*
 601  * This is the RPC `scheduler' (or rather, the finite state machine).
 602  */
 603 static void __rpc_execute(struct rpc_task *task)
 604 {
 605         struct rpc_wait_queue *queue;
 606         int task_is_async = RPC_IS_ASYNC(task);
 607         int status = 0;
 608
 609         dprintk("RPC: %5u __rpc_execute flags=0x%x\n",
 610                         task->tk_pid, task->tk_flags);
 611
 612         BUG_ON(RPC_IS_QUEUED(task));
 613
 614         for (;;) {
 615
 616                 /*
 617                  * Execute any pending callback.
 618                  */
 619                 if (task->tk_callback) {
 620                         void (*save_callback)(struct rpc_task *);
 621
 622                         /*
 623                          * We set tk_callback to NULL before calling it,
 624                          * in case it sets the tk_callback field itself:
 625                          */
 626                         save_callback = task->tk_callback;
 627                         task->tk_callback = NULL;
 628                         save_callback(task);
 629                 }
 630
 631                 /*
 632                  * Perform the next FSM step.
 633                  * tk_action may be NULL when the task has been killed
 634                  * by someone else.
 635                  */
 636                 if (!RPC_IS_QUEUED(task)) {
 637                         if (task->tk_action == NULL)
 638                                 break;
 639                         task->tk_action(task);
 640                 }
 641
 642                 /*
 643                  * Lockless check for whether task is sleeping or not.
 644                  */
 645                 if (!RPC_IS_QUEUED(task))
 646                         continue;
 647                 /*
 648                  * The queue->lock protects against races with
 649                  * rpc_make_runnable().
 650                  *
 651                  * Note that once we clear RPC_TASK_RUNNING on an asynchronous
 652                  * rpc_task, rpc_make_runnable() can assign it to a
 653                  * different workqueue. We therefore cannot assume that the
 654                  * rpc_task pointer may still be dereferenced.
 655                  */
 656                 queue = task->tk_waitqueue;
 657                 spin_lock_bh(&queue->lock);
 658                 if (!RPC_IS_QUEUED(task)) {
 659                         spin_unlock_bh(&queue->lock);
 660                         continue;
 661                 }
 662                 rpc_clear_running(task);
 663                 spin_unlock_bh(&queue->lock);
 664                 if (task_is_async)
 665                         return;
 666
 667                 /* sync task: sleep here */
 668                 dprintk("RPC: %5u sync task going to sleep\n", task->tk_pid);
 669                 status = out_of_line_wait_on_bit(&task->tk_runstate,
 670                                 RPC_TASK_QUEUED, rpc_wait_bit_killable,
 671                                 TASK_KILLABLE);
 672                 if (status == -ERESTARTSYS) {
 673                         /*
 674                          * When a sync task receives a signal, it exits with
 675                          * -ERESTARTSYS. In order to catch any callbacks that
 676                          * clean up after sleeping on some queue, we don't
 677                          * break the loop here, but go around once more.
 678                          */
 679                         dprintk("RPC: %5u got signal\n", task->tk_pid);
 680                         task->tk_flags |= RPC_TASK_KILLED;
 681                         rpc_exit(task, -ERESTARTSYS);
 682                         rpc_wake_up_task(task);
 683                 }
 684                 rpc_set_running(task);
 685                 dprintk("RPC: %5u sync task resuming\n", task->tk_pid);
 686         }
 687
 688         dprintk("RPC: %5u return %d, status %d\n", task->tk_pid, status,
 689                         task->tk_status);
 690         /* Release all resources associated with the task */
 691         rpc_release_task(task);
 692 }
 693
 694 /*
 695  * User-visible entry point to the scheduler.
 696  *
 697  * This may be called recursively if e.g. an async NFS task updates
 698  * the attributes and finds that dirty pages must be flushed.
 699  * NOTE: Upon exit of this function the task is guaranteed to be
 700  *       released. In particular note that tk_release() will have
 701  *       been called, so your task memory may have been freed.
 702  */
 703 void rpc_execute(struct rpc_task *task)
 704 {
 705         rpc_set_active(task);
 706         rpc_set_running(task);
 707         __rpc_execute(task);
 708 }
 709
 710 static void rpc_async_schedule(struct work_struct *work)
 711 {
 712         __rpc_execute(container_of(work, struct rpc_task, u.tk_work));
 713 }
 714
 715 struct rpc_buffer {
 716         size_t  len;
 717         char    data[];
 718 };
 719
 720 /**
 721  * rpc_malloc - allocate an RPC buffer
 722  * @task: RPC task that will use this buffer
 723  * @size: requested byte size
 724  *
 725  * To prevent rpciod from hanging, this allocator never sleeps,
 726  * returning NULL if the request cannot be serviced immediately.
 727  * The caller can arrange to sleep in a way that is safe for rpciod.
 728  *
 729  * Most requests are 'small' (under 2KiB) and can be serviced from a
 730  * mempool, ensuring that NFS reads and writes can always proceed,
 731  * and that there is good locality of reference for these buffers.
 732  *
 733  * In order to avoid memory starvation triggering more writebacks of
 734  * NFS requests, we avoid using GFP_KERNEL.
 735  */
 736 void *rpc_malloc(struct rpc_task *task, size_t size)
 737 {
 738         struct rpc_buffer *buf;
 739         gfp_t gfp = RPC_IS_SWAPPER(task) ? GFP_ATOMIC : GFP_NOWAIT;
 740
 741         size += sizeof(struct rpc_buffer);
 742         if (size <= RPC_BUFFER_MAXSIZE)
 743                 buf = mempool_alloc(rpc_buffer_mempool, gfp);
 744         else
 745                 buf = kmalloc(size, gfp);
 746
 747         if (!buf)
 748                 return NULL;
 749
 750         buf->len = size;
 751         dprintk("RPC: %5u allocated buffer of size %zu at %p\n",
 752                         task->tk_pid, size, buf);
 753         return &buf->data;
 754 }
 755 EXPORT_SYMBOL_GPL(rpc_malloc);
 756
 757 /**
 758  * rpc_free - free buffer allocated via rpc_malloc
 759  * @buffer: buffer to free
 760  *
 761  */
 762 void rpc_free(void *buffer)
 763 {
 764         size_t size;
 765         struct rpc_buffer *buf;
 766
 767         if (!buffer)
 768                 return;
 769
 770         buf = container_of(buffer, struct rpc_buffer, data);
 771         size = buf->len;
 772
 773         dprintk("RPC:       freeing buffer of size %zu at %p\n",
 774                         size, buf);
 775
 776         if (size <= RPC_BUFFER_MAXSIZE)
 777                 mempool_free(buf, rpc_buffer_mempool);
 778         else
 779                 kfree(buf);
 780 }
 781 EXPORT_SYMBOL_GPL(rpc_free);
 782
 783 /*
 784  * Creation and deletion of RPC task structures
 785  */
 786 static void rpc_init_task(struct rpc_task *task, const struct rpc_task_setup *task_setup_data)
 787 {
 788         memset(task, 0, sizeof(*task));
 789         atomic_set(&task->tk_count, 1);
 790         task->tk_flags  = task_setup_data->flags;
 791         task->tk_ops = task_setup_data->callback_ops;
 792         task->tk_calldata = task_setup_data->callback_data;
 793         INIT_LIST_HEAD(&task->tk_task);
 794
 795         /* Initialize retry counters */
 796         task->tk_garb_retry = 2;
 797         task->tk_cred_retry = 2;
 798
 799         task->tk_priority = task_setup_data->priority - RPC_PRIORITY_LOW;
 800         task->tk_owner = current->tgid;
 801
 802         /* Initialize workqueue for async tasks */
 803         task->tk_workqueue = task_setup_data->workqueue;
 804
 805         task->tk_client = task_setup_data->rpc_client;
 806         if (task->tk_client != NULL) {
 807                 kref_get(&task->tk_client->cl_kref);
 808                 if (task->tk_client->cl_softrtry)
 809                         task->tk_flags |= RPC_TASK_SOFT;
 810         }
 811
 812         if (task->tk_ops->rpc_call_prepare != NULL)
 813                 task->tk_action = rpc_prepare_task;
 814
 815         if (task_setup_data->rpc_message != NULL) {
 816                 task->tk_msg.rpc_proc = task_setup_data->rpc_message->rpc_proc;
 817                 task->tk_msg.rpc_argp = task_setup_data->rpc_message->rpc_argp;
 818                 task->tk_msg.rpc_resp = task_setup_data->rpc_message->rpc_resp;
 819                 /* Bind the user cred */
 820                 rpcauth_bindcred(task, task_setup_data->rpc_message->rpc_cred, task_setup_data->flags);
 821                 if (task->tk_action == NULL)
 822                         rpc_call_start(task);
 823         }
 824
 825         /* starting timestamp */
 826         task->tk_start = jiffies;
 827
 828         dprintk("RPC:       new task initialized, procpid %u\n",
 829                                 task_pid_nr(current));
 830 }
 831
 832 static struct rpc_task *
 833 rpc_alloc_task(void)
 834 {
 835         return (struct rpc_task *)mempool_alloc(rpc_task_mempool, GFP_NOFS);
 836 }
 837
 838 /*
 839  * Create a new task for the specified client.
 840  */
 841 struct rpc_task *rpc_new_task(const struct rpc_task_setup *setup_data)
 842 {
 843         struct rpc_task *task = setup_data->task;
 844         unsigned short flags = 0;
 845
 846         if (task == NULL) {
 847                 task = rpc_alloc_task();
 848                 if (task == NULL)
 849                         goto out;
 850                 flags = RPC_TASK_DYNAMIC;
 851         }
 852
 853         rpc_init_task(task, setup_data);
 854
 855         task->tk_flags |= flags;
 856         dprintk("RPC:       allocated task %p\n", task);
 857 out:
 858         return task;
 859 }
 860
 861 static void rpc_free_task(struct rpc_task *task)
 862 {
 863         const struct rpc_call_ops *tk_ops = task->tk_ops;
 864         void *calldata = task->tk_calldata;
 865
 866         if (task->tk_flags & RPC_TASK_DYNAMIC) {
 867                 dprintk("RPC: %5u freeing task\n", task->tk_pid);
 868                 mempool_free(task, rpc_task_mempool);
 869         }
 870         rpc_release_calldata(tk_ops, calldata);
 871 }
 872
 873 static void rpc_async_release(struct work_struct *work)
 874 {
 875         rpc_free_task(container_of(work, struct rpc_task, u.tk_work));
 876 }
 877
 878 void rpc_put_task(struct rpc_task *task)
 879 {
 880         if (!atomic_dec_and_test(&task->tk_count))
 881                 return;
 882         /* Release resources */
 883         if (task->tk_rqstp)
 884                 xprt_release(task);
 885         if (task->tk_msg.rpc_cred)
 886                 rpcauth_unbindcred(task);
 887         if (task->tk_client) {
 888                 rpc_release_client(task->tk_client);
 889                 task->tk_client = NULL;
 890         }
 891         if (task->tk_workqueue != NULL) {
 892                 INIT_WORK(&task->u.tk_work, rpc_async_release);
 893                 queue_work(task->tk_workqueue, &task->u.tk_work);
 894         } else
 895                 rpc_free_task(task);
 896 }
 897 EXPORT_SYMBOL_GPL(rpc_put_task);
 898
 899 static void rpc_release_task(struct rpc_task *task)
 900 {
 901 #ifdef RPC_DEBUG
 902         BUG_ON(task->tk_magic != RPC_TASK_MAGIC_ID);
 903 #endif
 904         dprintk("RPC: %5u release task\n", task->tk_pid);
 905
 906         if (!list_empty(&task->tk_task)) {
 907                 struct rpc_clnt *clnt = task->tk_client;
 908                 /* Remove from client task list */
 909                 spin_lock(&clnt->cl_lock);
 910                 list_del(&task->tk_task);
 911                 spin_unlock(&clnt->cl_lock);
 912         }
 913         BUG_ON (RPC_IS_QUEUED(task));
 914
 915 #ifdef RPC_DEBUG
 916         task->tk_magic = 0;
 917 #endif
 918         /* Wake up anyone who is waiting for task completion */
 919         rpc_mark_complete_task(task);
 920
 921         rpc_put_task(task);
 922 }
 923
 924 /*
 925  * Kill all tasks for the given client.
 926  * XXX: kill their descendants as well?
 927  */
 928 void rpc_killall_tasks(struct rpc_clnt *clnt)
 929 {
 930         struct rpc_task *rovr;
 931
 932
 933         if (list_empty(&clnt->cl_tasks))
 934                 return;
 935         dprintk("RPC:       killing all tasks for client %p\n", clnt);
 936         /*
 937          * Spin lock all_tasks to prevent changes...
 938          */
 939         spin_lock(&clnt->cl_lock);
 940         list_for_each_entry(rovr, &clnt->cl_tasks, tk_task) {
 941                 if (! RPC_IS_ACTIVATED(rovr))
 942                         continue;
 943                 if (!(rovr->tk_flags & RPC_TASK_KILLED)) {
 944                         rovr->tk_flags |= RPC_TASK_KILLED;
 945                         rpc_exit(rovr, -EIO);
 946                         rpc_wake_up_task(rovr);
 947                 }
 948         }
 949         spin_unlock(&clnt->cl_lock);
 950 }
 951 EXPORT_SYMBOL_GPL(rpc_killall_tasks);
 952
 953 int rpciod_up(void)
 954 {
 955         return try_module_get(THIS_MODULE) ? 0 : -EINVAL;
 956 }
 957
 958 void rpciod_down(void)
 959 {
 960         module_put(THIS_MODULE);
 961 }
 962
 963 /*
 964  * Start up the rpciod workqueue.
 965  */
 966 static int rpciod_start(void)
 967 {
 968         struct workqueue_struct *wq;
 969
 970         /*
 971          * Create the rpciod thread and wait for it to start.
 972          */
 973         dprintk("RPC:       creating workqueue rpciod\n");
 974         wq = create_workqueue("rpciod");
 975         rpciod_workqueue = wq;
 976         return rpciod_workqueue != NULL;
 977 }
 978
 979 static void rpciod_stop(void)
 980 {
 981         struct workqueue_struct *wq = NULL;
 982
 983         if (rpciod_workqueue == NULL)
 984                 return;
 985         dprintk("RPC:       destroying workqueue rpciod\n");
 986
 987         wq = rpciod_workqueue;
 988         rpciod_workqueue = NULL;
 989         destroy_workqueue(wq);
 990 }
 991
 992 void
 993 rpc_destroy_mempool(void)
 994 {
 995         rpciod_stop();
 996         if (rpc_buffer_mempool)
 997                 mempool_destroy(rpc_buffer_mempool);
 998         if (rpc_task_mempool)
 999                 mempool_destroy(rpc_task_mempool);
1000         if (rpc_task_slabp)
1001                 kmem_cache_destroy(rpc_task_slabp);
1002         if (rpc_buffer_slabp)
1003                 kmem_cache_destroy(rpc_buffer_slabp);
1004         rpc_destroy_wait_queue(&delay_queue);
1005 }
1006
1007 int
1008 rpc_init_mempool(void)
1009 {
1010         /*
1011          * The following is not strictly a mempool initialisation,
1012          * but there is no harm in doing it here
1013          */
1014         rpc_init_wait_queue(&delay_queue, "delayq");
1015         if (!rpciod_start())
1016                 goto err_nomem;
1017
1018         rpc_task_slabp = kmem_cache_create("rpc_tasks",
1019                                              sizeof(struct rpc_task),
1020                                              0, SLAB_HWCACHE_ALIGN,
1021                                              NULL);
1022         if (!rpc_task_slabp)
1023                 goto err_nomem;
1024         rpc_buffer_slabp = kmem_cache_create("rpc_buffers",
1025                                              RPC_BUFFER_MAXSIZE,
1026                                              0, SLAB_HWCACHE_ALIGN,
1027                                              NULL);
1028         if (!rpc_buffer_slabp)
1029                 goto err_nomem;
1030         rpc_task_mempool = mempool_create_slab_pool(RPC_TASK_POOLSIZE,
1031                                                     rpc_task_slabp);
1032         if (!rpc_task_mempool)
1033                 goto err_nomem;
1034         rpc_buffer_mempool = mempool_create_slab_pool(RPC_BUFFER_POOLSIZE,
1035                                                       rpc_buffer_slabp);
1036         if (!rpc_buffer_mempool)
1037                 goto err_nomem;
1038         return 0;
1039 err_nomem:
1040         rpc_destroy_mempool();
1041         return -ENOMEM;
1042 }