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, <okir@monad.swb.de> 8 * 9 * TCP NFS related read + write fixes 10 * (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie> 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 } 1364