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