net/sunrpc/sched.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * linux/net/sunrpc/sched.c
   4  *
   5  * Scheduling for synchronous and asynchronous RPC requests.
   6  *
   7  * Copyright (C) 1996 Olaf Kirch, <[email protected]>
   8  *
   9  * TCP NFS related read + write fixes
  10  * (C) 1999 Dave Airlie, University of Limerick, Ireland <[email protected]>
  11  */
  12
  13 #include <linux/module.h>
  14
  15 #include <linux/sched.h>
  16 #include <linux/interrupt.h>
  17 #include <linux/slab.h>
  18 #include <linux/mempool.h>
  19 #include <linux/smp.h>
  20 #include <linux/spinlock.h>
  21 #include <linux/mutex.h>
  22 #include <linux/freezer.h>
  23 #include <linux/sched/mm.h>
  24
  25 #include <linux/sunrpc/clnt.h>
  26 #include <linux/sunrpc/metrics.h>
  27
  28 #include "sunrpc.h"
  29
  30 #define CREATE_TRACE_POINTS
  31 #include <trace/events/sunrpc.h>
  32
  33 /*
  34  * RPC slabs and memory pools
  35  */
  36 #define RPC_BUFFER_MAXSIZE      (2048)
  37 #define RPC_BUFFER_POOLSIZE     (8)
  38 #define RPC_TASK_POOLSIZE       (8)
  39 static struct kmem_cache        *rpc_task_slabp __read_mostly;
  40 static struct kmem_cache        *rpc_buffer_slabp __read_mostly;
  41 static mempool_t        *rpc_task_mempool __read_mostly;
  42 static mempool_t        *rpc_buffer_mempool __read_mostly;
  43
  44 static void                     rpc_async_schedule(struct work_struct *);
  45 static void                      rpc_release_task(struct rpc_task *task);
  46 static void __rpc_queue_timer_fn(struct work_struct *);
  47
  48 /*
  49  * RPC tasks sit here while waiting for conditions to improve.
  50  */
  51 static struct rpc_wait_queue delay_queue;
  52
  53 /*
  54  * rpciod-related stuff
  55  */
  56 struct workqueue_struct *rpciod_workqueue __read_mostly;
  57 struct workqueue_struct *xprtiod_workqueue __read_mostly;
  58 EXPORT_SYMBOL_GPL(xprtiod_workqueue);
  59
  60 gfp_t rpc_task_gfp_mask(void)
  61 {
  62         if (current->flags & PF_WQ_WORKER)
  63                 return GFP_KERNEL | __GFP_NORETRY | __GFP_NOWARN;
  64         return GFP_KERNEL;
  65 }
  66 EXPORT_SYMBOL_GPL(rpc_task_gfp_mask);
  67
  68 bool rpc_task_set_rpc_status(struct rpc_task *task, int rpc_status)
  69 {
  70         if (cmpxchg(&task->tk_rpc_status, 0, rpc_status) == 0)
  71                 return true;
  72         return false;
  73 }
  74
  75 unsigned long
  76 rpc_task_timeout(const struct rpc_task *task)
  77 {
  78         unsigned long timeout = READ_ONCE(task->tk_timeout);
  79
  80         if (timeout != 0) {
  81                 unsigned long now = jiffies;
  82                 if (time_before(now, timeout))
  83                         return timeout - now;
  84         }
  85         return 0;
  86 }
  87 EXPORT_SYMBOL_GPL(rpc_task_timeout);
  88
  89 /*
  90  * Disable the timer for a given RPC task. Should be called with
  91  * queue->lock and bh_disabled in order to avoid races within
  92  * rpc_run_timer().
  93  */
  94 static void
  95 __rpc_disable_timer(struct rpc_wait_queue *queue, struct rpc_task *task)
  96 {
  97         if (list_empty(&task->u.tk_wait.timer_list))
  98                 return;
  99         task->tk_timeout = 0;
 100         list_del(&task->u.tk_wait.timer_list);
 101         if (list_empty(&queue->timer_list.list))
 102                 cancel_delayed_work(&queue->timer_list.dwork);
 103 }
 104
 105 static void
 106 rpc_set_queue_timer(struct rpc_wait_queue *queue, unsigned long expires)
 107 {
 108         unsigned long now = jiffies;
 109         queue->timer_list.expires = expires;
 110         if (time_before_eq(expires, now))
 111                 expires = 0;
 112         else
 113                 expires -= now;
 114         mod_delayed_work(rpciod_workqueue, &queue->timer_list.dwork, expires);
 115 }
 116
 117 /*
 118  * Set up a timer for the current task.
 119  */
 120 static void
 121 __rpc_add_timer(struct rpc_wait_queue *queue, struct rpc_task *task,
 122                 unsigned long timeout)
 123 {
 124         task->tk_timeout = timeout;
 125         if (list_empty(&queue->timer_list.list) || time_before(timeout, queue->timer_list.expires))
 126                 rpc_set_queue_timer(queue, timeout);
 127         list_add(&task->u.tk_wait.timer_list, &queue->timer_list.list);
 128 }
 129
 130 static void rpc_set_waitqueue_priority(struct rpc_wait_queue *queue, int priority)
 131 {
 132         if (queue->priority != priority) {
 133                 queue->priority = priority;
 134                 queue->nr = 1U << priority;
 135         }
 136 }
 137
 138 static void rpc_reset_waitqueue_priority(struct rpc_wait_queue *queue)
 139 {
 140         rpc_set_waitqueue_priority(queue, queue->maxpriority);
 141 }
 142
 143 /*
 144  * Add a request to a queue list
 145  */
 146 static void
 147 __rpc_list_enqueue_task(struct list_head *q, struct rpc_task *task)
 148 {
 149         struct rpc_task *t;
 150
 151         list_for_each_entry(t, q, u.tk_wait.list) {
 152                 if (t->tk_owner == task->tk_owner) {
 153                         list_add_tail(&task->u.tk_wait.links,
 154                                         &t->u.tk_wait.links);
 155                         /* Cache the queue head in task->u.tk_wait.list */
 156                         task->u.tk_wait.list.next = q;
 157                         task->u.tk_wait.list.prev = NULL;
 158                         return;
 159                 }
 160         }
 161         INIT_LIST_HEAD(&task->u.tk_wait.links);
 162         list_add_tail(&task->u.tk_wait.list, q);
 163 }
 164
 165 /*
 166  * Remove request from a queue list
 167  */
 168 static void
 169 __rpc_list_dequeue_task(struct rpc_task *task)
 170 {
 171         struct list_head *q;
 172         struct rpc_task *t;
 173
 174         if (task->u.tk_wait.list.prev == NULL) {
 175                 list_del(&task->u.tk_wait.links);
 176                 return;
 177         }
 178         if (!list_empty(&task->u.tk_wait.links)) {
 179                 t = list_first_entry(&task->u.tk_wait.links,
 180                                 struct rpc_task,
 181                                 u.tk_wait.links);
 182                 /* Assume __rpc_list_enqueue_task() cached the queue head */
 183                 q = t->u.tk_wait.list.next;
 184                 list_add_tail(&t->u.tk_wait.list, q);
 185                 list_del(&task->u.tk_wait.links);
 186         }
 187         list_del(&task->u.tk_wait.list);
 188 }
 189
 190 /*
 191  * Add new request to a priority queue.
 192  */
 193 static void __rpc_add_wait_queue_priority(struct rpc_wait_queue *queue,
 194                 struct rpc_task *task,
 195                 unsigned char queue_priority)
 196 {
 197         if (unlikely(queue_priority > queue->maxpriority))
 198                 queue_priority = queue->maxpriority;
 199         __rpc_list_enqueue_task(&queue->tasks[queue_priority], task);
 200 }
 201
 202 /*
 203  * Add new request to wait queue.
 204  */
 205 static void __rpc_add_wait_queue(struct rpc_wait_queue *queue,
 206                 struct rpc_task *task,
 207                 unsigned char queue_priority)
 208 {
 209         INIT_LIST_HEAD(&task->u.tk_wait.timer_list);
 210         if (RPC_IS_PRIORITY(queue))
 211                 __rpc_add_wait_queue_priority(queue, task, queue_priority);
 212         else
 213                 list_add_tail(&task->u.tk_wait.list, &queue->tasks[0]);
 214         task->tk_waitqueue = queue;
 215         queue->qlen++;
 216         /* barrier matches the read in rpc_wake_up_task_queue_locked() */
 217         smp_wmb();
 218         rpc_set_queued(task);
 219 }
 220
 221 /*
 222  * Remove request from a priority queue.
 223  */
 224 static void __rpc_remove_wait_queue_priority(struct rpc_task *task)
 225 {
 226         __rpc_list_dequeue_task(task);
 227 }
 228
 229 /*
 230  * Remove request from queue.
 231  * Note: must be called with spin lock held.
 232  */
 233 static void __rpc_remove_wait_queue(struct rpc_wait_queue *queue, struct rpc_task *task)
 234 {
 235         __rpc_disable_timer(queue, task);
 236         if (RPC_IS_PRIORITY(queue))
 237                 __rpc_remove_wait_queue_priority(task);
 238         else
 239                 list_del(&task->u.tk_wait.list);
 240         queue->qlen--;
 241 }
 242
 243 static void __rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname, unsigned char nr_queues)
 244 {
 245         int i;
 246
 247         spin_lock_init(&queue->lock);
 248         for (i = 0; i < ARRAY_SIZE(queue->tasks); i++)
 249                 INIT_LIST_HEAD(&queue->tasks[i]);
 250         queue->maxpriority = nr_queues - 1;
 251         rpc_reset_waitqueue_priority(queue);
 252         queue->qlen = 0;
 253         queue->timer_list.expires = 0;
 254         INIT_DELAYED_WORK(&queue->timer_list.dwork, __rpc_queue_timer_fn);
 255         INIT_LIST_HEAD(&queue->timer_list.list);
 256         rpc_assign_waitqueue_name(queue, qname);
 257 }
 258
 259 void rpc_init_priority_wait_queue(struct rpc_wait_queue *queue, const char *qname)
 260 {
 261         __rpc_init_priority_wait_queue(queue, qname, RPC_NR_PRIORITY);
 262 }
 263 EXPORT_SYMBOL_GPL(rpc_init_priority_wait_queue);
 264
 265 void rpc_init_wait_queue(struct rpc_wait_queue *queue, const char *qname)
 266 {
 267         __rpc_init_priority_wait_queue(queue, qname, 1);
 268 }
 269 EXPORT_SYMBOL_GPL(rpc_init_wait_queue);
 270
 271 void rpc_destroy_wait_queue(struct rpc_wait_queue *queue)
 272 {
 273         cancel_delayed_work_sync(&queue->timer_list.dwork);
 274 }
 275 EXPORT_SYMBOL_GPL(rpc_destroy_wait_queue);
 276
 277 static int rpc_wait_bit_killable(struct wait_bit_key *key, int mode)
 278 {
 279         schedule();
 280         if (signal_pending_state(mode, current))
 281                 return -ERESTARTSYS;
 282         return 0;
 283 }
 284
 285 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG) || IS_ENABLED(CONFIG_TRACEPOINTS)
 286 static void rpc_task_set_debuginfo(struct rpc_task *task)
 287 {
 288         struct rpc_clnt *clnt = task->tk_client;
 289
 290         /* Might be a task carrying a reverse-direction operation */
 291         if (!clnt) {
 292                 static atomic_t rpc_pid;
 293
 294                 task->tk_pid = atomic_inc_return(&rpc_pid);
 295                 return;
 296         }
 297
 298         task->tk_pid = atomic_inc_return(&clnt->cl_pid);
 299 }
 300 #else
 301 static inline void rpc_task_set_debuginfo(struct rpc_task *task)
 302 {
 303 }
 304 #endif
 305
 306 static void rpc_set_active(struct rpc_task *task)
 307 {
 308         rpc_task_set_debuginfo(task);
 309         set_bit(RPC_TASK_ACTIVE, &task->tk_runstate);
 310         trace_rpc_task_begin(task, NULL);
 311 }
 312
 313 /*
 314  * Mark an RPC call as having completed by clearing the 'active' bit
 315  * and then waking up all tasks that were sleeping.
 316  */
 317 static int rpc_complete_task(struct rpc_task *task)
 318 {
 319         void *m = &task->tk_runstate;
 320         wait_queue_head_t *wq = bit_waitqueue(m, RPC_TASK_ACTIVE);
 321         struct wait_bit_key k = __WAIT_BIT_KEY_INITIALIZER(m, RPC_TASK_ACTIVE);
 322         unsigned long flags;
 323         int ret;
 324
 325         trace_rpc_task_complete(task, NULL);
 326
 327         spin_lock_irqsave(&wq->lock, flags);
 328         clear_bit(RPC_TASK_ACTIVE, &task->tk_runstate);
 329         ret = atomic_dec_and_test(&task->tk_count);
 330         if (waitqueue_active(wq))
 331                 __wake_up_locked_key(wq, TASK_NORMAL, &k);
 332         spin_unlock_irqrestore(&wq->lock, flags);
 333         return ret;
 334 }
 335
 336 /*
 337  * Allow callers to wait for completion of an RPC call
 338  *
 339  * Note the use of out_of_line_wait_on_bit() rather than wait_on_bit()
 340  * to enforce taking of the wq->lock and hence avoid races with
 341  * rpc_complete_task().
 342  */
 343 int rpc_wait_for_completion_task(struct rpc_task *task)
 344 {
 345         return out_of_line_wait_on_bit(&task->tk_runstate, RPC_TASK_ACTIVE,
 346                         rpc_wait_bit_killable, TASK_KILLABLE|TASK_FREEZABLE_UNSAFE);
 347 }
 348 EXPORT_SYMBOL_GPL(rpc_wait_for_completion_task);
 349
 350 /*
 351  * Make an RPC task runnable.
 352  *
 353  * Note: If the task is ASYNC, and is being made runnable after sitting on an
 354  * rpc_wait_queue, this must be called with the queue spinlock held to protect
 355  * the wait queue operation.
 356  * Note the ordering of rpc_test_and_set_running() and rpc_clear_queued(),
 357  * which is needed to ensure that __rpc_execute() doesn't loop (due to the
 358  * lockless RPC_IS_QUEUED() test) before we've had a chance to test
 359  * the RPC_TASK_RUNNING flag.
 360  */
 361 static void rpc_make_runnable(struct workqueue_struct *wq,
 362                 struct rpc_task *task)
 363 {
 364         bool need_wakeup = !rpc_test_and_set_running(task);
 365
 366         rpc_clear_queued(task);
 367         if (!need_wakeup)
 368                 return;
 369         if (RPC_IS_ASYNC(task)) {
 370                 INIT_WORK(&task->u.tk_work, rpc_async_schedule);
 371                 queue_work(wq, &task->u.tk_work);
 372         } else {
 373                 smp_mb__after_atomic();
 374                 wake_up_bit(&task->tk_runstate, RPC_TASK_QUEUED);
 375         }
 376 }
 377
 378 /*
 379  * Prepare for sleeping on a wait queue.
 380  * By always appending tasks to the list we ensure FIFO behavior.
 381  * NB: An RPC task will only receive interrupt-driven events as long
 382  * as it's on a wait queue.
 383  */
 384 static void __rpc_do_sleep_on_priority(struct rpc_wait_queue *q,
 385                 struct rpc_task *task,
 386                 unsigned char queue_priority)
 387 {
 388         trace_rpc_task_sleep(task, q);
 389
 390         __rpc_add_wait_queue(q, task, queue_priority);
 391 }
 392
 393 static void __rpc_sleep_on_priority(struct rpc_wait_queue *q,
 394                 struct rpc_task *task,
 395                 unsigned char queue_priority)
 396 {
 397         if (WARN_ON_ONCE(RPC_IS_QUEUED(task)))
 398                 return;
 399         __rpc_do_sleep_on_priority(q, task, queue_priority);
 400 }
 401
 402 static void __rpc_sleep_on_priority_timeout(struct rpc_wait_queue *q,
 403                 struct rpc_task *task, unsigned long timeout,
 404                 unsigned char queue_priority)
 405 {
 406         if (WARN_ON_ONCE(RPC_IS_QUEUED(task)))
 407                 return;
 408         if (time_is_after_jiffies(timeout)) {
 409                 __rpc_do_sleep_on_priority(q, task, queue_priority);
 410                 __rpc_add_timer(q, task, timeout);
 411         } else
 412                 task->tk_status = -ETIMEDOUT;
 413 }
 414
 415 static void rpc_set_tk_callback(struct rpc_task *task, rpc_action action)
 416 {
 417         if (action && !WARN_ON_ONCE(task->tk_callback != NULL))
 418                 task->tk_callback = action;
 419 }
 420
 421 static bool rpc_sleep_check_activated(struct rpc_task *task)
 422 {
 423         /* We shouldn't ever put an inactive task to sleep */
 424         if (WARN_ON_ONCE(!RPC_IS_ACTIVATED(task))) {
 425                 task->tk_status = -EIO;
 426                 rpc_put_task_async(task);
 427                 return false;
 428         }
 429         return true;
 430 }
 431
 432 void rpc_sleep_on_timeout(struct rpc_wait_queue *q, struct rpc_task *task,
 433                                 rpc_action action, unsigned long timeout)
 434 {
 435         if (!rpc_sleep_check_activated(task))
 436                 return;
 437
 438         rpc_set_tk_callback(task, action);
 439
 440         /*
 441          * Protect the queue operations.
 442          */
 443         spin_lock(&q->lock);
 444         __rpc_sleep_on_priority_timeout(q, task, timeout, task->tk_priority);
 445         spin_unlock(&q->lock);
 446 }
 447 EXPORT_SYMBOL_GPL(rpc_sleep_on_timeout);
 448
 449 void rpc_sleep_on(struct rpc_wait_queue *q, struct rpc_task *task,
 450                                 rpc_action action)
 451 {
 452         if (!rpc_sleep_check_activated(task))
 453                 return;
 454
 455         rpc_set_tk_callback(task, action);
 456
 457         WARN_ON_ONCE(task->tk_timeout != 0);
 458         /*
 459          * Protect the queue operations.
 460          */
 461         spin_lock(&q->lock);
 462         __rpc_sleep_on_priority(q, task, task->tk_priority);
 463         spin_unlock(&q->lock);
 464 }
 465 EXPORT_SYMBOL_GPL(rpc_sleep_on);
 466
 467 void rpc_sleep_on_priority_timeout(struct rpc_wait_queue *q,
 468                 struct rpc_task *task, unsigned long timeout, int priority)
 469 {
 470         if (!rpc_sleep_check_activated(task))
 471                 return;
 472
 473         priority -= RPC_PRIORITY_LOW;
 474         /*
 475          * Protect the queue operations.
 476          */
 477         spin_lock(&q->lock);
 478         __rpc_sleep_on_priority_timeout(q, task, timeout, priority);
 479         spin_unlock(&q->lock);
 480 }
 481 EXPORT_SYMBOL_GPL(rpc_sleep_on_priority_timeout);
 482
 483 void rpc_sleep_on_priority(struct rpc_wait_queue *q, struct rpc_task *task,
 484                 int priority)
 485 {
 486         if (!rpc_sleep_check_activated(task))
 487                 return;
 488
 489         WARN_ON_ONCE(task->tk_timeout != 0);
 490         priority -= RPC_PRIORITY_LOW;
 491         /*
 492          * Protect the queue operations.
 493          */
 494         spin_lock(&q->lock);
 495         __rpc_sleep_on_priority(q, task, priority);
 496         spin_unlock(&q->lock);
 497 }
 498 EXPORT_SYMBOL_GPL(rpc_sleep_on_priority);
 499
 500 /**
 501  * __rpc_do_wake_up_task_on_wq - wake up a single rpc_task
 502  * @wq: workqueue on which to run task
 503  * @queue: wait queue
 504  * @task: task to be woken up
 505  *
 506  * Caller must hold queue->lock, and have cleared the task queued flag.
 507  */
 508 static void __rpc_do_wake_up_task_on_wq(struct workqueue_struct *wq,
 509                 struct rpc_wait_queue *queue,
 510                 struct rpc_task *task)
 511 {
 512         /* Has the task been executed yet? If not, we cannot wake it up! */
 513         if (!RPC_IS_ACTIVATED(task)) {
 514                 printk(KERN_ERR "RPC: Inactive task (%p) being woken up!\n", task);
 515                 return;
 516         }
 517
 518         trace_rpc_task_wakeup(task, queue);
 519
 520         __rpc_remove_wait_queue(queue, task);
 521
 522         rpc_make_runnable(wq, task);
 523 }
 524
 525 /*
 526  * Wake up a queued task while the queue lock is being held
 527  */
 528 static struct rpc_task *
 529 rpc_wake_up_task_on_wq_queue_action_locked(struct workqueue_struct *wq,
 530                 struct rpc_wait_queue *queue, struct rpc_task *task,
 531                 bool (*action)(struct rpc_task *, void *), void *data)
 532 {
 533         if (RPC_IS_QUEUED(task)) {
 534                 smp_rmb();
 535                 if (task->tk_waitqueue == queue) {
 536                         if (action == NULL || action(task, data)) {
 537                                 __rpc_do_wake_up_task_on_wq(wq, queue, task);
 538                                 return task;
 539                         }
 540                 }
 541         }
 542         return NULL;
 543 }
 544
 545 /*
 546  * Wake up a queued task while the queue lock is being held
 547  */
 548 static void rpc_wake_up_task_queue_locked(struct rpc_wait_queue *queue,
 549                                           struct rpc_task *task)
 550 {
 551         rpc_wake_up_task_on_wq_queue_action_locked(rpciod_workqueue, queue,
 552                                                    task, NULL, NULL);
 553 }
 554
 555 /*
 556  * Wake up a task on a specific queue
 557  */
 558 void rpc_wake_up_queued_task(struct rpc_wait_queue *queue, struct rpc_task *task)
 559 {
 560         if (!RPC_IS_QUEUED(task))
 561                 return;
 562         spin_lock(&queue->lock);
 563         rpc_wake_up_task_queue_locked(queue, task);
 564         spin_unlock(&queue->lock);
 565 }
 566 EXPORT_SYMBOL_GPL(rpc_wake_up_queued_task);
 567
 568 static bool rpc_task_action_set_status(struct rpc_task *task, void *status)
 569 {
 570         task->tk_status = *(int *)status;
 571         return true;
 572 }
 573
 574 static void
 575 rpc_wake_up_task_queue_set_status_locked(struct rpc_wait_queue *queue,
 576                 struct rpc_task *task, int status)
 577 {
 578         rpc_wake_up_task_on_wq_queue_action_locked(rpciod_workqueue, queue,
 579                         task, rpc_task_action_set_status, &status);
 580 }
 581
 582 /**
 583  * rpc_wake_up_queued_task_set_status - wake up a task and set task->tk_status
 584  * @queue: pointer to rpc_wait_queue
 585  * @task: pointer to rpc_task
 586  * @status: integer error value
 587  *
 588  * If @task is queued on @queue, then it is woken up, and @task->tk_status is
 589  * set to the value of @status.
 590  */
 591 void
 592 rpc_wake_up_queued_task_set_status(struct rpc_wait_queue *queue,
 593                 struct rpc_task *task, int status)
 594 {
 595         if (!RPC_IS_QUEUED(task))
 596                 return;
 597         spin_lock(&queue->lock);
 598         rpc_wake_up_task_queue_set_status_locked(queue, task, status);
 599         spin_unlock(&queue->lock);
 600 }
 601
 602 /*
 603  * Wake up the next task on a priority queue.
 604  */
 605 static struct rpc_task *__rpc_find_next_queued_priority(struct rpc_wait_queue *queue)
 606 {
 607         struct list_head *q;
 608         struct rpc_task *task;
 609
 610         /*
 611          * Service the privileged queue.
 612          */
 613         q = &queue->tasks[RPC_NR_PRIORITY - 1];
 614         if (queue->maxpriority > RPC_PRIORITY_PRIVILEGED && !list_empty(q)) {
 615                 task = list_first_entry(q, struct rpc_task, u.tk_wait.list);
 616                 goto out;
 617         }
 618
 619         /*
 620          * Service a batch of tasks from a single owner.
 621          */
 622         q = &queue->tasks[queue->priority];
 623         if (!list_empty(q) && queue->nr) {
 624                 queue->nr--;
 625                 task = list_first_entry(q, struct rpc_task, u.tk_wait.list);
 626                 goto out;
 627         }
 628
 629         /*
 630          * Service the next queue.
 631          */
 632         do {
 633                 if (q == &queue->tasks[0])
 634                         q = &queue->tasks[queue->maxpriority];
 635                 else
 636                         q = q - 1;
 637                 if (!list_empty(q)) {
 638                         task = list_first_entry(q, struct rpc_task, u.tk_wait.list);
 639                         goto new_queue;
 640                 }
 641         } while (q != &queue->tasks[queue->priority]);
 642
 643         rpc_reset_waitqueue_priority(queue);
 644         return NULL;
 645
 646 new_queue:
 647         rpc_set_waitqueue_priority(queue, (unsigned int)(q - &queue->tasks[0]));
 648 out:
 649         return task;
 650 }
 651
 652 static struct rpc_task *__rpc_find_next_queued(struct rpc_wait_queue *queue)
 653 {
 654         if (RPC_IS_PRIORITY(queue))
 655                 return __rpc_find_next_queued_priority(queue);
 656         if (!list_empty(&queue->tasks[0]))
 657                 return list_first_entry(&queue->tasks[0], struct rpc_task, u.tk_wait.list);
 658         return NULL;
 659 }
 660
 661 /*
 662  * Wake up the first task on the wait queue.
 663  */
 664 struct rpc_task *rpc_wake_up_first_on_wq(struct workqueue_struct *wq,
 665                 struct rpc_wait_queue *queue,
 666                 bool (*func)(struct rpc_task *, void *), void *data)
 667 {
 668         struct rpc_task *task = NULL;
 669
 670         spin_lock(&queue->lock);
 671         task = __rpc_find_next_queued(queue);
 672         if (task != NULL)
 673                 task = rpc_wake_up_task_on_wq_queue_action_locked(wq, queue,
 674                                 task, func, data);
 675         spin_unlock(&queue->lock);
 676
 677         return task;
 678 }
 679
 680 /*
 681  * Wake up the first task on the wait queue.
 682  */
 683 struct rpc_task *rpc_wake_up_first(struct rpc_wait_queue *queue,
 684                 bool (*func)(struct rpc_task *, void *), void *data)
 685 {
 686         return rpc_wake_up_first_on_wq(rpciod_workqueue, queue, func, data);
 687 }
 688 EXPORT_SYMBOL_GPL(rpc_wake_up_first);
 689
 690 static bool rpc_wake_up_next_func(struct rpc_task *task, void *data)
 691 {
 692         return true;
 693 }
 694
 695 /*
 696  * Wake up the next task on the wait queue.
 697 */
 698 struct rpc_task *rpc_wake_up_next(struct rpc_wait_queue *queue)
 699 {
 700         return rpc_wake_up_first(queue, rpc_wake_up_next_func, NULL);
 701 }
 702 EXPORT_SYMBOL_GPL(rpc_wake_up_next);
 703
 704 /**
 705  * rpc_wake_up_locked - wake up all rpc_tasks
 706  * @queue: rpc_wait_queue on which the tasks are sleeping
 707  *
 708  */
 709 static void rpc_wake_up_locked(struct rpc_wait_queue *queue)
 710 {
 711         struct rpc_task *task;
 712
 713         for (;;) {
 714                 task = __rpc_find_next_queued(queue);
 715                 if (task == NULL)
 716                         break;
 717                 rpc_wake_up_task_queue_locked(queue, task);
 718         }
 719 }
 720
 721 /**
 722  * rpc_wake_up - wake up all rpc_tasks
 723  * @queue: rpc_wait_queue on which the tasks are sleeping
 724  *
 725  * Grabs queue->lock
 726  */
 727 void rpc_wake_up(struct rpc_wait_queue *queue)
 728 {
 729         spin_lock(&queue->lock);
 730         rpc_wake_up_locked(queue);
 731         spin_unlock(&queue->lock);
 732 }
 733 EXPORT_SYMBOL_GPL(rpc_wake_up);
 734
 735 /**
 736  * rpc_wake_up_status_locked - wake up all rpc_tasks and set their status value.
 737  * @queue: rpc_wait_queue on which the tasks are sleeping
 738  * @status: status value to set
 739  */
 740 static void rpc_wake_up_status_locked(struct rpc_wait_queue *queue, int status)
 741 {
 742         struct rpc_task *task;
 743
 744         for (;;) {
 745                 task = __rpc_find_next_queued(queue);
 746                 if (task == NULL)
 747                         break;
 748                 rpc_wake_up_task_queue_set_status_locked(queue, task, status);
 749         }
 750 }
 751
 752 /**
 753  * rpc_wake_up_status - wake up all rpc_tasks and set their status value.
 754  * @queue: rpc_wait_queue on which the tasks are sleeping
 755  * @status: status value to set
 756  *
 757  * Grabs queue->lock
 758  */
 759 void rpc_wake_up_status(struct rpc_wait_queue *queue, int status)
 760 {
 761         spin_lock(&queue->lock);
 762         rpc_wake_up_status_locked(queue, status);
 763         spin_unlock(&queue->lock);
 764 }
 765 EXPORT_SYMBOL_GPL(rpc_wake_up_status);
 766
 767 static void __rpc_queue_timer_fn(struct work_struct *work)
 768 {
 769         struct rpc_wait_queue *queue = container_of(work,
 770                         struct rpc_wait_queue,
 771                         timer_list.dwork.work);
 772         struct rpc_task *task, *n;
 773         unsigned long expires, now, timeo;
 774
 775         spin_lock(&queue->lock);
 776         expires = now = jiffies;
 777         list_for_each_entry_safe(task, n, &queue->timer_list.list, u.tk_wait.timer_list) {
 778                 timeo = task->tk_timeout;
 779                 if (time_after_eq(now, timeo)) {
 780                         trace_rpc_task_timeout(task, task->tk_action);
 781                         task->tk_status = -ETIMEDOUT;
 782                         rpc_wake_up_task_queue_locked(queue, task);
 783                         continue;
 784                 }
 785                 if (expires == now || time_after(expires, timeo))
 786                         expires = timeo;
 787         }
 788         if (!list_empty(&queue->timer_list.list))
 789                 rpc_set_queue_timer(queue, expires);
 790         spin_unlock(&queue->lock);
 791 }
 792
 793 static void __rpc_atrun(struct rpc_task *task)
 794 {
 795         if (task->tk_status == -ETIMEDOUT)
 796                 task->tk_status = 0;
 797 }
 798
 799 /*
 800  * Run a task at a later time
 801  */
 802 void rpc_delay(struct rpc_task *task, unsigned long delay)
 803 {
 804         rpc_sleep_on_timeout(&delay_queue, task, __rpc_atrun, jiffies + delay);
 805 }
 806 EXPORT_SYMBOL_GPL(rpc_delay);
 807
 808 /*
 809  * Helper to call task->tk_ops->rpc_call_prepare
 810  */
 811 void rpc_prepare_task(struct rpc_task *task)
 812 {
 813         task->tk_ops->rpc_call_prepare(task, task->tk_calldata);
 814 }
 815
 816 static void
 817 rpc_init_task_statistics(struct rpc_task *task)
 818 {
 819         /* Initialize retry counters */
 820         task->tk_garb_retry = 2;
 821         task->tk_cred_retry = 2;
 822
 823         /* starting timestamp */
 824         task->tk_start = ktime_get();
 825 }
 826
 827 static void
 828 rpc_reset_task_statistics(struct rpc_task *task)
 829 {
 830         task->tk_timeouts = 0;
 831         task->tk_flags &= ~(RPC_CALL_MAJORSEEN|RPC_TASK_SENT);
 832         rpc_init_task_statistics(task);
 833 }
 834
 835 /*
 836  * Helper that calls task->tk_ops->rpc_call_done if it exists
 837  */
 838 void rpc_exit_task(struct rpc_task *task)
 839 {
 840         trace_rpc_task_end(task, task->tk_action);
 841         task->tk_action = NULL;
 842         if (task->tk_ops->rpc_count_stats)
 843                 task->tk_ops->rpc_count_stats(task, task->tk_calldata);
 844         else if (task->tk_client)
 845                 rpc_count_iostats(task, task->tk_client->cl_metrics);
 846         if (task->tk_ops->rpc_call_done != NULL) {
 847                 trace_rpc_task_call_done(task, task->tk_ops->rpc_call_done);
 848                 task->tk_ops->rpc_call_done(task, task->tk_calldata);
 849                 if (task->tk_action != NULL) {
 850                         /* Always release the RPC slot and buffer memory */
 851                         xprt_release(task);
 852                         rpc_reset_task_statistics(task);
 853                 }
 854         }
 855 }
 856
 857 void rpc_signal_task(struct rpc_task *task)
 858 {
 859         struct rpc_wait_queue *queue;
 860
 861         if (!RPC_IS_ACTIVATED(task))
 862                 return;
 863
 864         if (!rpc_task_set_rpc_status(task, -ERESTARTSYS))
 865                 return;
 866         trace_rpc_task_signalled(task, task->tk_action);
 867         set_bit(RPC_TASK_SIGNALLED, &task->tk_runstate);
 868         smp_mb__after_atomic();
 869         queue = READ_ONCE(task->tk_waitqueue);
 870         if (queue)
 871                 rpc_wake_up_queued_task(queue, task);
 872 }
 873
 874 void rpc_task_try_cancel(struct rpc_task *task, int error)
 875 {
 876         struct rpc_wait_queue *queue;
 877
 878         if (!rpc_task_set_rpc_status(task, error))
 879                 return;
 880         queue = READ_ONCE(task->tk_waitqueue);
 881         if (queue)
 882                 rpc_wake_up_queued_task(queue, task);
 883 }
 884
 885 void rpc_exit(struct rpc_task *task, int status)
 886 {
 887         task->tk_status = status;
 888         task->tk_action = rpc_exit_task;
 889         rpc_wake_up_queued_task(task->tk_waitqueue, task);
 890 }
 891 EXPORT_SYMBOL_GPL(rpc_exit);
 892
 893 void rpc_release_calldata(const struct rpc_call_ops *ops, void *calldata)
 894 {
 895         if (ops->rpc_release != NULL)
 896                 ops->rpc_release(calldata);
 897 }
 898
 899 static bool xprt_needs_memalloc(struct rpc_xprt *xprt, struct rpc_task *tk)
 900 {
 901         if (!xprt)
 902                 return false;
 903         if (!atomic_read(&xprt->swapper))
 904                 return false;
 905         return test_bit(XPRT_LOCKED, &xprt->state) && xprt->snd_task == tk;
 906 }
 907
 908 /*
 909  * This is the RPC `scheduler' (or rather, the finite state machine).
 910  */
 911 static void __rpc_execute(struct rpc_task *task)
 912 {
 913         struct rpc_wait_queue *queue;
 914         int task_is_async = RPC_IS_ASYNC(task);
 915         int status = 0;
 916         unsigned long pflags = current->flags;
 917
 918         WARN_ON_ONCE(RPC_IS_QUEUED(task));
 919         if (RPC_IS_QUEUED(task))
 920                 return;
 921
 922         for (;;) {
 923                 void (*do_action)(struct rpc_task *);
 924
 925                 /*
 926                  * Perform the next FSM step or a pending callback.
 927                  *
 928                  * tk_action may be NULL if the task has been killed.
 929                  */
 930                 do_action = task->tk_action;
 931                 /* Tasks with an RPC error status should exit */
 932                 if (do_action && do_action != rpc_exit_task &&
 933                     (status = READ_ONCE(task->tk_rpc_status)) != 0) {
 934                         task->tk_status = status;
 935                         do_action = rpc_exit_task;
 936                 }
 937                 /* Callbacks override all actions */
 938                 if (task->tk_callback) {
 939                         do_action = task->tk_callback;
 940                         task->tk_callback = NULL;
 941                 }
 942                 if (!do_action)
 943                         break;
 944                 if (RPC_IS_SWAPPER(task) ||
 945                     xprt_needs_memalloc(task->tk_xprt, task))
 946                         current->flags |= PF_MEMALLOC;
 947
 948                 trace_rpc_task_run_action(task, do_action);
 949                 do_action(task);
 950
 951                 /*
 952                  * Lockless check for whether task is sleeping or not.
 953                  */
 954                 if (!RPC_IS_QUEUED(task)) {
 955                         cond_resched();
 956                         continue;
 957                 }
 958
 959                 /*
 960                  * The queue->lock protects against races with
 961                  * rpc_make_runnable().
 962                  *
 963                  * Note that once we clear RPC_TASK_RUNNING on an asynchronous
 964                  * rpc_task, rpc_make_runnable() can assign it to a
 965                  * different workqueue. We therefore cannot assume that the
 966                  * rpc_task pointer may still be dereferenced.
 967                  */
 968                 queue = task->tk_waitqueue;
 969                 spin_lock(&queue->lock);
 970                 if (!RPC_IS_QUEUED(task)) {
 971                         spin_unlock(&queue->lock);
 972                         continue;
 973                 }
 974                 /* Wake up any task that has an exit status */
 975                 if (READ_ONCE(task->tk_rpc_status) != 0) {
 976                         rpc_wake_up_task_queue_locked(queue, task);
 977                         spin_unlock(&queue->lock);
 978                         continue;
 979                 }
 980                 rpc_clear_running(task);
 981                 spin_unlock(&queue->lock);
 982                 if (task_is_async)
 983                         goto out;
 984
 985                 /* sync task: sleep here */
 986                 trace_rpc_task_sync_sleep(task, task->tk_action);
 987                 status = out_of_line_wait_on_bit(&task->tk_runstate,
 988                                 RPC_TASK_QUEUED, rpc_wait_bit_killable,
 989                                 TASK_KILLABLE|TASK_FREEZABLE);
 990                 if (status < 0) {
 991                         /*
 992                          * When a sync task receives a signal, it exits with
 993                          * -ERESTARTSYS. In order to catch any callbacks that
 994                          * clean up after sleeping on some queue, we don't
 995                          * break the loop here, but go around once more.
 996                          */
 997                         rpc_signal_task(task);
 998                 }
 999                 trace_rpc_task_sync_wake(task, task->tk_action);
1000         }
1001
1002         /* Release all resources associated with the task */
1003         rpc_release_task(task);
1004 out:
1005         current_restore_flags(pflags, PF_MEMALLOC);
1006 }
1007
1008 /*
1009  * User-visible entry point to the scheduler.
1010  *
1011  * This may be called recursively if e.g. an async NFS task updates
1012  * the attributes and finds that dirty pages must be flushed.
1013  * NOTE: Upon exit of this function the task is guaranteed to be
1014  *       released. In particular note that tk_release() will have
1015  *       been called, so your task memory may have been freed.
1016  */
1017 void rpc_execute(struct rpc_task *task)
1018 {
1019         bool is_async = RPC_IS_ASYNC(task);
1020
1021         rpc_set_active(task);
1022         rpc_make_runnable(rpciod_workqueue, task);
1023         if (!is_async) {
1024                 unsigned int pflags = memalloc_nofs_save();
1025                 __rpc_execute(task);
1026                 memalloc_nofs_restore(pflags);
1027         }
1028 }
1029
1030 static void rpc_async_schedule(struct work_struct *work)
1031 {
1032         unsigned int pflags = memalloc_nofs_save();
1033
1034         __rpc_execute(container_of(work, struct rpc_task, u.tk_work));
1035         memalloc_nofs_restore(pflags);
1036 }
1037
1038 /**
1039  * rpc_malloc - allocate RPC buffer resources
1040  * @task: RPC task
1041  *
1042  * A single memory region is allocated, which is split between the
1043  * RPC call and RPC reply that this task is being used for. When
1044  * this RPC is retired, the memory is released by calling rpc_free.
1045  *
1046  * To prevent rpciod from hanging, this allocator never sleeps,
1047  * returning -ENOMEM and suppressing warning if the request cannot
1048  * be serviced immediately. The caller can arrange to sleep in a
1049  * way that is safe for rpciod.
1050  *
1051  * Most requests are 'small' (under 2KiB) and can be serviced from a
1052  * mempool, ensuring that NFS reads and writes can always proceed,
1053  * and that there is good locality of reference for these buffers.
1054  */
1055 int rpc_malloc(struct rpc_task *task)
1056 {
1057         struct rpc_rqst *rqst = task->tk_rqstp;
1058         size_t size = rqst->rq_callsize + rqst->rq_rcvsize;
1059         struct rpc_buffer *buf;
1060         gfp_t gfp = rpc_task_gfp_mask();
1061
1062         size += sizeof(struct rpc_buffer);
1063         if (size <= RPC_BUFFER_MAXSIZE) {
1064                 buf = kmem_cache_alloc(rpc_buffer_slabp, gfp);
1065                 /* Reach for the mempool if dynamic allocation fails */
1066                 if (!buf && RPC_IS_ASYNC(task))
1067                         buf = mempool_alloc(rpc_buffer_mempool, GFP_NOWAIT);
1068         } else
1069                 buf = kmalloc(size, gfp);
1070
1071         if (!buf)
1072                 return -ENOMEM;
1073
1074         buf->len = size;
1075         rqst->rq_buffer = buf->data;
1076         rqst->rq_rbuffer = (char *)rqst->rq_buffer + rqst->rq_callsize;
1077         return 0;
1078 }
1079 EXPORT_SYMBOL_GPL(rpc_malloc);
1080
1081 /**
1082  * rpc_free - free RPC buffer resources allocated via rpc_malloc
1083  * @task: RPC task
1084  *
1085  */
1086 void rpc_free(struct rpc_task *task)
1087 {
1088         void *buffer = task->tk_rqstp->rq_buffer;
1089         size_t size;
1090         struct rpc_buffer *buf;
1091
1092         buf = container_of(buffer, struct rpc_buffer, data);
1093         size = buf->len;
1094
1095         if (size <= RPC_BUFFER_MAXSIZE)
1096                 mempool_free(buf, rpc_buffer_mempool);
1097         else
1098                 kfree(buf);
1099 }
1100 EXPORT_SYMBOL_GPL(rpc_free);
1101
1102 /*
1103  * Creation and deletion of RPC task structures
1104  */
1105 static void rpc_init_task(struct rpc_task *task, const struct rpc_task_setup *task_setup_data)
1106 {
1107         memset(task, 0, sizeof(*task));
1108         atomic_set(&task->tk_count, 1);
1109         task->tk_flags  = task_setup_data->flags;
1110         task->tk_ops = task_setup_data->callback_ops;
1111         task->tk_calldata = task_setup_data->callback_data;
1112         INIT_LIST_HEAD(&task->tk_task);
1113
1114         task->tk_priority = task_setup_data->priority - RPC_PRIORITY_LOW;
1115         task->tk_owner = current->tgid;
1116
1117         /* Initialize workqueue for async tasks */
1118         task->tk_workqueue = task_setup_data->workqueue;
1119
1120         task->tk_xprt = rpc_task_get_xprt(task_setup_data->rpc_client,
1121                         xprt_get(task_setup_data->rpc_xprt));
1122
1123         task->tk_op_cred = get_rpccred(task_setup_data->rpc_op_cred);
1124
1125         if (task->tk_ops->rpc_call_prepare != NULL)
1126                 task->tk_action = rpc_prepare_task;
1127
1128         rpc_init_task_statistics(task);
1129 }
1130
1131 static struct rpc_task *rpc_alloc_task(void)
1132 {
1133         struct rpc_task *task;
1134
1135         task = kmem_cache_alloc(rpc_task_slabp, rpc_task_gfp_mask());
1136         if (task)
1137                 return task;
1138         return mempool_alloc(rpc_task_mempool, GFP_NOWAIT);
1139 }
1140
1141 /*
1142  * Create a new task for the specified client.
1143  */
1144 struct rpc_task *rpc_new_task(const struct rpc_task_setup *setup_data)
1145 {
1146         struct rpc_task *task = setup_data->task;
1147         unsigned short flags = 0;
1148
1149         if (task == NULL) {
1150                 task = rpc_alloc_task();
1151                 if (task == NULL) {
1152                         rpc_release_calldata(setup_data->callback_ops,
1153                                              setup_data->callback_data);
1154                         return ERR_PTR(-ENOMEM);
1155                 }
1156                 flags = RPC_TASK_DYNAMIC;
1157         }
1158
1159         rpc_init_task(task, setup_data);
1160         task->tk_flags |= flags;
1161         return task;
1162 }
1163
1164 /*
1165  * rpc_free_task - release rpc task and perform cleanups
1166  *
1167  * Note that we free up the rpc_task _after_ rpc_release_calldata()
1168  * in order to work around a workqueue dependency issue.
1169  *
1170  * Tejun Heo states:
1171  * "Workqueue currently considers two work items to be the same if they're
1172  * on the same address and won't execute them concurrently - ie. it
1173  * makes a work item which is queued again while being executed wait
1174  * for the previous execution to complete.
1175  *
1176  * If a work function frees the work item, and then waits for an event
1177  * which should be performed by another work item and *that* work item
1178  * recycles the freed work item, it can create a false dependency loop.
1179  * There really is no reliable way to detect this short of verifying
1180  * every memory free."
1181  *
1182  */
1183 static void rpc_free_task(struct rpc_task *task)
1184 {
1185         unsigned short tk_flags = task->tk_flags;
1186
1187         put_rpccred(task->tk_op_cred);
1188         rpc_release_calldata(task->tk_ops, task->tk_calldata);
1189
1190         if (tk_flags & RPC_TASK_DYNAMIC)
1191                 mempool_free(task, rpc_task_mempool);
1192 }
1193
1194 static void rpc_async_release(struct work_struct *work)
1195 {
1196         unsigned int pflags = memalloc_nofs_save();
1197
1198         rpc_free_task(container_of(work, struct rpc_task, u.tk_work));
1199         memalloc_nofs_restore(pflags);
1200 }
1201
1202 static void rpc_release_resources_task(struct rpc_task *task)
1203 {
1204         xprt_release(task);
1205         if (task->tk_msg.rpc_cred) {
1206                 if (!(task->tk_flags & RPC_TASK_CRED_NOREF))
1207                         put_cred(task->tk_msg.rpc_cred);
1208                 task->tk_msg.rpc_cred = NULL;
1209         }
1210         rpc_task_release_client(task);
1211 }
1212
1213 static void rpc_final_put_task(struct rpc_task *task,
1214                 struct workqueue_struct *q)
1215 {
1216         if (q != NULL) {
1217                 INIT_WORK(&task->u.tk_work, rpc_async_release);
1218                 queue_work(q, &task->u.tk_work);
1219         } else
1220                 rpc_free_task(task);
1221 }
1222
1223 static void rpc_do_put_task(struct rpc_task *task, struct workqueue_struct *q)
1224 {
1225         if (atomic_dec_and_test(&task->tk_count)) {
1226                 rpc_release_resources_task(task);
1227                 rpc_final_put_task(task, q);
1228         }
1229 }
1230
1231 void rpc_put_task(struct rpc_task *task)
1232 {
1233         rpc_do_put_task(task, NULL);
1234 }
1235 EXPORT_SYMBOL_GPL(rpc_put_task);
1236
1237 void rpc_put_task_async(struct rpc_task *task)
1238 {
1239         rpc_do_put_task(task, task->tk_workqueue);
1240 }
1241 EXPORT_SYMBOL_GPL(rpc_put_task_async);
1242
1243 static void rpc_release_task(struct rpc_task *task)
1244 {
1245         WARN_ON_ONCE(RPC_IS_QUEUED(task));
1246
1247         rpc_release_resources_task(task);
1248
1249         /*
1250          * Note: at this point we have been removed from rpc_clnt->cl_tasks,
1251          * so it should be safe to use task->tk_count as a test for whether
1252          * or not any other processes still hold references to our rpc_task.
1253          */
1254         if (atomic_read(&task->tk_count) != 1 + !RPC_IS_ASYNC(task)) {
1255                 /* Wake up anyone who may be waiting for task completion */
1256                 if (!rpc_complete_task(task))
1257                         return;
1258         } else {
1259                 if (!atomic_dec_and_test(&task->tk_count))
1260                         return;
1261         }
1262         rpc_final_put_task(task, task->tk_workqueue);
1263 }
1264
1265 int rpciod_up(void)
1266 {
1267         return try_module_get(THIS_MODULE) ? 0 : -EINVAL;
1268 }
1269
1270 void rpciod_down(void)
1271 {
1272         module_put(THIS_MODULE);
1273 }
1274
1275 /*
1276  * Start up the rpciod workqueue.
1277  */
1278 static int rpciod_start(void)
1279 {
1280         struct workqueue_struct *wq;
1281
1282         /*
1283          * Create the rpciod thread and wait for it to start.
1284          */
1285         wq = alloc_workqueue("rpciod", WQ_MEM_RECLAIM | WQ_UNBOUND, 0);
1286         if (!wq)
1287                 goto out_failed;
1288         rpciod_workqueue = wq;
1289         wq = alloc_workqueue("xprtiod", WQ_UNBOUND | WQ_MEM_RECLAIM, 0);
1290         if (!wq)
1291                 goto free_rpciod;
1292         xprtiod_workqueue = wq;
1293         return 1;
1294 free_rpciod:
1295         wq = rpciod_workqueue;
1296         rpciod_workqueue = NULL;
1297         destroy_workqueue(wq);
1298 out_failed:
1299         return 0;
1300 }
1301
1302 static void rpciod_stop(void)
1303 {
1304         struct workqueue_struct *wq = NULL;
1305
1306         if (rpciod_workqueue == NULL)
1307                 return;
1308
1309         wq = rpciod_workqueue;
1310         rpciod_workqueue = NULL;
1311         destroy_workqueue(wq);
1312         wq = xprtiod_workqueue;
1313         xprtiod_workqueue = NULL;
1314         destroy_workqueue(wq);
1315 }
1316
1317 void
1318 rpc_destroy_mempool(void)
1319 {
1320         rpciod_stop();
1321         mempool_destroy(rpc_buffer_mempool);
1322         mempool_destroy(rpc_task_mempool);
1323         kmem_cache_destroy(rpc_task_slabp);
1324         kmem_cache_destroy(rpc_buffer_slabp);
1325         rpc_destroy_wait_queue(&delay_queue);
1326 }
1327
1328 int
1329 rpc_init_mempool(void)
1330 {
1331         /*
1332          * The following is not strictly a mempool initialisation,
1333          * but there is no harm in doing it here
1334          */
1335         rpc_init_wait_queue(&delay_queue, "delayq");
1336         if (!rpciod_start())
1337                 goto err_nomem;
1338
1339         rpc_task_slabp = kmem_cache_create("rpc_tasks",
1340                                              sizeof(struct rpc_task),
1341                                              0, SLAB_HWCACHE_ALIGN,
1342                                              NULL);
1343         if (!rpc_task_slabp)
1344                 goto err_nomem;
1345         rpc_buffer_slabp = kmem_cache_create("rpc_buffers",
1346                                              RPC_BUFFER_MAXSIZE,
1347                                              0, SLAB_HWCACHE_ALIGN,
1348                                              NULL);
1349         if (!rpc_buffer_slabp)
1350                 goto err_nomem;
1351         rpc_task_mempool = mempool_create_slab_pool(RPC_TASK_POOLSIZE,
1352                                                     rpc_task_slabp);
1353         if (!rpc_task_mempool)
1354                 goto err_nomem;
1355         rpc_buffer_mempool = mempool_create_slab_pool(RPC_BUFFER_POOLSIZE,
1356                                                       rpc_buffer_slabp);
1357         if (!rpc_buffer_mempool)
1358                 goto err_nomem;
1359         return 0;
1360 err_nomem:
1361         rpc_destroy_mempool();
1362         return -ENOMEM;
1363 }