fs/btrfs/async-thread.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/kthread.h>
  20 #include <linux/slab.h>
  21 #include <linux/list.h>
  22 #include <linux/spinlock.h>
  23 #include <linux/freezer.h>
  24 #include "async-thread.h"
  25
  26 #define WORK_QUEUED_BIT 0
  27 #define WORK_DONE_BIT 1
  28 #define WORK_ORDER_DONE_BIT 2
  29 #define WORK_HIGH_PRIO_BIT 3
  30
  31 /*
  32  * container for the kthread task pointer and the list of pending work
  33  * One of these is allocated per thread.
  34  */
  35 struct btrfs_worker_thread {
  36         /* pool we belong to */
  37         struct btrfs_workers *workers;
  38
  39         /* list of struct btrfs_work that are waiting for service */
  40         struct list_head pending;
  41         struct list_head prio_pending;
  42
  43         /* list of worker threads from struct btrfs_workers */
  44         struct list_head worker_list;
  45
  46         /* kthread */
  47         struct task_struct *task;
  48
  49         /* number of things on the pending list */
  50         atomic_t num_pending;
  51
  52         /* reference counter for this struct */
  53         atomic_t refs;
  54
  55         unsigned long sequence;
  56
  57         /* protects the pending list. */
  58         spinlock_t lock;
  59
  60         /* set to non-zero when this thread is already awake and kicking */
  61         int working;
  62
  63         /* are we currently idle */
  64         int idle;
  65 };
  66
  67 static int __btrfs_start_workers(struct btrfs_workers *workers);
  68
  69 /*
  70  * btrfs_start_workers uses kthread_run, which can block waiting for memory
  71  * for a very long time.  It will actually throttle on page writeback,
  72  * and so it may not make progress until after our btrfs worker threads
  73  * process all of the pending work structs in their queue
  74  *
  75  * This means we can't use btrfs_start_workers from inside a btrfs worker
  76  * thread that is used as part of cleaning dirty memory, which pretty much
  77  * involves all of the worker threads.
  78  *
  79  * Instead we have a helper queue who never has more than one thread
  80  * where we scheduler thread start operations.  This worker_start struct
  81  * is used to contain the work and hold a pointer to the queue that needs
  82  * another worker.
  83  */
  84 struct worker_start {
  85         struct btrfs_work work;
  86         struct btrfs_workers *queue;
  87 };
  88
  89 static void start_new_worker_func(struct btrfs_work *work)
  90 {
  91         struct worker_start *start;
  92         start = container_of(work, struct worker_start, work);
  93         __btrfs_start_workers(start->queue);
  94         kfree(start);
  95 }
  96
  97 /*
  98  * helper function to move a thread onto the idle list after it
  99  * has finished some requests.
 100  */
 101 static void check_idle_worker(struct btrfs_worker_thread *worker)
 102 {
 103         if (!worker->idle && atomic_read(&worker->num_pending) <
 104             worker->workers->idle_thresh / 2) {
 105                 unsigned long flags;
 106                 spin_lock_irqsave(&worker->workers->lock, flags);
 107                 worker->idle = 1;
 108
 109                 /* the list may be empty if the worker is just starting */
 110                 if (!list_empty(&worker->worker_list)) {
 111                         list_move(&worker->worker_list,
 112                                  &worker->workers->idle_list);
 113                 }
 114                 spin_unlock_irqrestore(&worker->workers->lock, flags);
 115         }
 116 }
 117
 118 /*
 119  * helper function to move a thread off the idle list after new
 120  * pending work is added.
 121  */
 122 static void check_busy_worker(struct btrfs_worker_thread *worker)
 123 {
 124         if (worker->idle && atomic_read(&worker->num_pending) >=
 125             worker->workers->idle_thresh) {
 126                 unsigned long flags;
 127                 spin_lock_irqsave(&worker->workers->lock, flags);
 128                 worker->idle = 0;
 129
 130                 if (!list_empty(&worker->worker_list)) {
 131                         list_move_tail(&worker->worker_list,
 132                                       &worker->workers->worker_list);
 133                 }
 134                 spin_unlock_irqrestore(&worker->workers->lock, flags);
 135         }
 136 }
 137
 138 static void check_pending_worker_creates(struct btrfs_worker_thread *worker)
 139 {
 140         struct btrfs_workers *workers = worker->workers;
 141         struct worker_start *start;
 142         unsigned long flags;
 143
 144         rmb();
 145         if (!workers->atomic_start_pending)
 146                 return;
 147
 148         start = kzalloc(sizeof(*start), GFP_NOFS);
 149         if (!start)
 150                 return;
 151
 152         start->work.func = start_new_worker_func;
 153         start->queue = workers;
 154
 155         spin_lock_irqsave(&workers->lock, flags);
 156         if (!workers->atomic_start_pending)
 157                 goto out;
 158
 159         workers->atomic_start_pending = 0;
 160         if (workers->num_workers + workers->num_workers_starting >=
 161             workers->max_workers)
 162                 goto out;
 163
 164         workers->num_workers_starting += 1;
 165         spin_unlock_irqrestore(&workers->lock, flags);
 166         btrfs_queue_worker(workers->atomic_worker_start, &start->work);
 167         return;
 168
 169 out:
 170         kfree(start);
 171         spin_unlock_irqrestore(&workers->lock, flags);
 172 }
 173
 174 static noinline void run_ordered_completions(struct btrfs_workers *workers,
 175                                             struct btrfs_work *work)
 176 {
 177         if (!workers->ordered)
 178                 return;
 179
 180         set_bit(WORK_DONE_BIT, &work->flags);
 181
 182         spin_lock(&workers->order_lock);
 183
 184         while (1) {
 185                 if (!list_empty(&workers->prio_order_list)) {
 186                         work = list_entry(workers->prio_order_list.next,
 187                                           struct btrfs_work, order_list);
 188                 } else if (!list_empty(&workers->order_list)) {
 189                         work = list_entry(workers->order_list.next,
 190                                           struct btrfs_work, order_list);
 191                 } else {
 192                         break;
 193                 }
 194                 if (!test_bit(WORK_DONE_BIT, &work->flags))
 195                         break;
 196
 197                 /* we are going to call the ordered done function, but
 198                  * we leave the work item on the list as a barrier so
 199                  * that later work items that are done don't have their
 200                  * functions called before this one returns
 201                  */
 202                 if (test_and_set_bit(WORK_ORDER_DONE_BIT, &work->flags))
 203                         break;
 204
 205                 spin_unlock(&workers->order_lock);
 206
 207                 work->ordered_func(work);
 208
 209                 /* now take the lock again and call the freeing code */
 210                 spin_lock(&workers->order_lock);
 211                 list_del(&work->order_list);
 212                 work->ordered_free(work);
 213         }
 214
 215         spin_unlock(&workers->order_lock);
 216 }
 217
 218 static void put_worker(struct btrfs_worker_thread *worker)
 219 {
 220         if (atomic_dec_and_test(&worker->refs))
 221                 kfree(worker);
 222 }
 223
 224 static int try_worker_shutdown(struct btrfs_worker_thread *worker)
 225 {
 226         int freeit = 0;
 227
 228         spin_lock_irq(&worker->lock);
 229         spin_lock(&worker->workers->lock);
 230         if (worker->workers->num_workers > 1 &&
 231             worker->idle &&
 232             !worker->working &&
 233             !list_empty(&worker->worker_list) &&
 234             list_empty(&worker->prio_pending) &&
 235             list_empty(&worker->pending) &&
 236             atomic_read(&worker->num_pending) == 0) {
 237                 freeit = 1;
 238                 list_del_init(&worker->worker_list);
 239                 worker->workers->num_workers--;
 240         }
 241         spin_unlock(&worker->workers->lock);
 242         spin_unlock_irq(&worker->lock);
 243
 244         if (freeit)
 245                 put_worker(worker);
 246         return freeit;
 247 }
 248
 249 static struct btrfs_work *get_next_work(struct btrfs_worker_thread *worker,
 250                                         struct list_head *prio_head,
 251                                         struct list_head *head)
 252 {
 253         struct btrfs_work *work = NULL;
 254         struct list_head *cur = NULL;
 255
 256         if(!list_empty(prio_head))
 257                 cur = prio_head->next;
 258
 259         smp_mb();
 260         if (!list_empty(&worker->prio_pending))
 261                 goto refill;
 262
 263         if (!list_empty(head))
 264                 cur = head->next;
 265
 266         if (cur)
 267                 goto out;
 268
 269 refill:
 270         spin_lock_irq(&worker->lock);
 271         list_splice_tail_init(&worker->prio_pending, prio_head);
 272         list_splice_tail_init(&worker->pending, head);
 273
 274         if (!list_empty(prio_head))
 275                 cur = prio_head->next;
 276         else if (!list_empty(head))
 277                 cur = head->next;
 278         spin_unlock_irq(&worker->lock);
 279
 280         if (!cur)
 281                 goto out_fail;
 282
 283 out:
 284         work = list_entry(cur, struct btrfs_work, list);
 285
 286 out_fail:
 287         return work;
 288 }
 289
 290 /*
 291  * main loop for servicing work items
 292  */
 293 static int worker_loop(void *arg)
 294 {
 295         struct btrfs_worker_thread *worker = arg;
 296         struct list_head head;
 297         struct list_head prio_head;
 298         struct btrfs_work *work;
 299
 300         INIT_LIST_HEAD(&head);
 301         INIT_LIST_HEAD(&prio_head);
 302
 303         do {
 304 again:
 305                 while (1) {
 306
 307
 308                         work = get_next_work(worker, &prio_head, &head);
 309                         if (!work)
 310                                 break;
 311
 312                         list_del(&work->list);
 313                         clear_bit(WORK_QUEUED_BIT, &work->flags);
 314
 315                         work->worker = worker;
 316
 317                         work->func(work);
 318
 319                         atomic_dec(&worker->num_pending);
 320                         /*
 321                          * unless this is an ordered work queue,
 322                          * 'work' was probably freed by func above.
 323                          */
 324                         run_ordered_completions(worker->workers, work);
 325
 326                         check_pending_worker_creates(worker);
 327                         cond_resched();
 328                 }
 329
 330                 spin_lock_irq(&worker->lock);
 331                 check_idle_worker(worker);
 332
 333                 if (freezing(current)) {
 334                         worker->working = 0;
 335                         spin_unlock_irq(&worker->lock);
 336                         try_to_freeze();
 337                 } else {
 338                         spin_unlock_irq(&worker->lock);
 339                         if (!kthread_should_stop()) {
 340                                 cpu_relax();
 341                                 /*
 342                                  * we've dropped the lock, did someone else
 343                                  * jump_in?
 344                                  */
 345                                 smp_mb();
 346                                 if (!list_empty(&worker->pending) ||
 347                                     !list_empty(&worker->prio_pending))
 348                                         continue;
 349
 350                                 /*
 351                                  * this short schedule allows more work to
 352                                  * come in without the queue functions
 353                                  * needing to go through wake_up_process()
 354                                  *
 355                                  * worker->working is still 1, so nobody
 356                                  * is going to try and wake us up
 357                                  */
 358                                 schedule_timeout(1);
 359                                 smp_mb();
 360                                 if (!list_empty(&worker->pending) ||
 361                                     !list_empty(&worker->prio_pending))
 362                                         continue;
 363
 364                                 if (kthread_should_stop())
 365                                         break;
 366
 367                                 /* still no more work?, sleep for real */
 368                                 spin_lock_irq(&worker->lock);
 369                                 set_current_state(TASK_INTERRUPTIBLE);
 370                                 if (!list_empty(&worker->pending) ||
 371                                     !list_empty(&worker->prio_pending)) {
 372                                         spin_unlock_irq(&worker->lock);
 373                                         set_current_state(TASK_RUNNING);
 374                                         goto again;
 375                                 }
 376
 377                                 /*
 378                                  * this makes sure we get a wakeup when someone
 379                                  * adds something new to the queue
 380                                  */
 381                                 worker->working = 0;
 382                                 spin_unlock_irq(&worker->lock);
 383
 384                                 if (!kthread_should_stop()) {
 385                                         schedule_timeout(HZ * 120);
 386                                         if (!worker->working &&
 387                                             try_worker_shutdown(worker)) {
 388                                                 return 0;
 389                                         }
 390                                 }
 391                         }
 392                         __set_current_state(TASK_RUNNING);
 393                 }
 394         } while (!kthread_should_stop());
 395         return 0;
 396 }
 397
 398 /*
 399  * this will wait for all the worker threads to shutdown
 400  */
 401 void btrfs_stop_workers(struct btrfs_workers *workers)
 402 {
 403         struct list_head *cur;
 404         struct btrfs_worker_thread *worker;
 405         int can_stop;
 406
 407         spin_lock_irq(&workers->lock);
 408         list_splice_init(&workers->idle_list, &workers->worker_list);
 409         while (!list_empty(&workers->worker_list)) {
 410                 cur = workers->worker_list.next;
 411                 worker = list_entry(cur, struct btrfs_worker_thread,
 412                                     worker_list);
 413
 414                 atomic_inc(&worker->refs);
 415                 workers->num_workers -= 1;
 416                 if (!list_empty(&worker->worker_list)) {
 417                         list_del_init(&worker->worker_list);
 418                         put_worker(worker);
 419                         can_stop = 1;
 420                 } else
 421                         can_stop = 0;
 422                 spin_unlock_irq(&workers->lock);
 423                 if (can_stop)
 424                         kthread_stop(worker->task);
 425                 spin_lock_irq(&workers->lock);
 426                 put_worker(worker);
 427         }
 428         spin_unlock_irq(&workers->lock);
 429 }
 430
 431 /*
 432  * simple init on struct btrfs_workers
 433  */
 434 void btrfs_init_workers(struct btrfs_workers *workers, char *name, int max,
 435                         struct btrfs_workers *async_helper)
 436 {
 437         workers->num_workers = 0;
 438         workers->num_workers_starting = 0;
 439         INIT_LIST_HEAD(&workers->worker_list);
 440         INIT_LIST_HEAD(&workers->idle_list);
 441         INIT_LIST_HEAD(&workers->order_list);
 442         INIT_LIST_HEAD(&workers->prio_order_list);
 443         spin_lock_init(&workers->lock);
 444         spin_lock_init(&workers->order_lock);
 445         workers->max_workers = max;
 446         workers->idle_thresh = 32;
 447         workers->name = name;
 448         workers->ordered = 0;
 449         workers->atomic_start_pending = 0;
 450         workers->atomic_worker_start = async_helper;
 451 }
 452
 453 /*
 454  * starts new worker threads.  This does not enforce the max worker
 455  * count in case you need to temporarily go past it.
 456  */
 457 static int __btrfs_start_workers(struct btrfs_workers *workers)
 458 {
 459         struct btrfs_worker_thread *worker;
 460         int ret = 0;
 461
 462         worker = kzalloc(sizeof(*worker), GFP_NOFS);
 463         if (!worker) {
 464                 ret = -ENOMEM;
 465                 goto fail;
 466         }
 467
 468         INIT_LIST_HEAD(&worker->pending);
 469         INIT_LIST_HEAD(&worker->prio_pending);
 470         INIT_LIST_HEAD(&worker->worker_list);
 471         spin_lock_init(&worker->lock);
 472
 473         atomic_set(&worker->num_pending, 0);
 474         atomic_set(&worker->refs, 1);
 475         worker->workers = workers;
 476         worker->task = kthread_run(worker_loop, worker,
 477                                    "btrfs-%s-%d", workers->name,
 478                                    workers->num_workers + 1);
 479         if (IS_ERR(worker->task)) {
 480                 ret = PTR_ERR(worker->task);
 481                 kfree(worker);
 482                 goto fail;
 483         }
 484         spin_lock_irq(&workers->lock);
 485         list_add_tail(&worker->worker_list, &workers->idle_list);
 486         worker->idle = 1;
 487         workers->num_workers++;
 488         workers->num_workers_starting--;
 489         WARN_ON(workers->num_workers_starting < 0);
 490         spin_unlock_irq(&workers->lock);
 491
 492         return 0;
 493 fail:
 494         spin_lock_irq(&workers->lock);
 495         workers->num_workers_starting--;
 496         spin_unlock_irq(&workers->lock);
 497         return ret;
 498 }
 499
 500 int btrfs_start_workers(struct btrfs_workers *workers)
 501 {
 502         spin_lock_irq(&workers->lock);
 503         workers->num_workers_starting++;
 504         spin_unlock_irq(&workers->lock);
 505         return __btrfs_start_workers(workers);
 506 }
 507
 508 /*
 509  * run through the list and find a worker thread that doesn't have a lot
 510  * to do right now.  This can return null if we aren't yet at the thread
 511  * count limit and all of the threads are busy.
 512  */
 513 static struct btrfs_worker_thread *next_worker(struct btrfs_workers *workers)
 514 {
 515         struct btrfs_worker_thread *worker;
 516         struct list_head *next;
 517         int enforce_min;
 518
 519         enforce_min = (workers->num_workers + workers->num_workers_starting) <
 520                 workers->max_workers;
 521
 522         /*
 523          * if we find an idle thread, don't move it to the end of the
 524          * idle list.  This improves the chance that the next submission
 525          * will reuse the same thread, and maybe catch it while it is still
 526          * working
 527          */
 528         if (!list_empty(&workers->idle_list)) {
 529                 next = workers->idle_list.next;
 530                 worker = list_entry(next, struct btrfs_worker_thread,
 531                                     worker_list);
 532                 return worker;
 533         }
 534         if (enforce_min || list_empty(&workers->worker_list))
 535                 return NULL;
 536
 537         /*
 538          * if we pick a busy task, move the task to the end of the list.
 539          * hopefully this will keep things somewhat evenly balanced.
 540          * Do the move in batches based on the sequence number.  This groups
 541          * requests submitted at roughly the same time onto the same worker.
 542          */
 543         next = workers->worker_list.next;
 544         worker = list_entry(next, struct btrfs_worker_thread, worker_list);
 545         worker->sequence++;
 546
 547         if (worker->sequence % workers->idle_thresh == 0)
 548                 list_move_tail(next, &workers->worker_list);
 549         return worker;
 550 }
 551
 552 /*
 553  * selects a worker thread to take the next job.  This will either find
 554  * an idle worker, start a new worker up to the max count, or just return
 555  * one of the existing busy workers.
 556  */
 557 static struct btrfs_worker_thread *find_worker(struct btrfs_workers *workers)
 558 {
 559         struct btrfs_worker_thread *worker;
 560         unsigned long flags;
 561         struct list_head *fallback;
 562         int ret;
 563
 564         spin_lock_irqsave(&workers->lock, flags);
 565 again:
 566         worker = next_worker(workers);
 567
 568         if (!worker) {
 569                 if (workers->num_workers + workers->num_workers_starting >=
 570                     workers->max_workers) {
 571                         goto fallback;
 572                 } else if (workers->atomic_worker_start) {
 573                         workers->atomic_start_pending = 1;
 574                         goto fallback;
 575                 } else {
 576                         workers->num_workers_starting++;
 577                         spin_unlock_irqrestore(&workers->lock, flags);
 578                         /* we're below the limit, start another worker */
 579                         ret = __btrfs_start_workers(workers);
 580                         spin_lock_irqsave(&workers->lock, flags);
 581                         if (ret)
 582                                 goto fallback;
 583                         goto again;
 584                 }
 585         }
 586         goto found;
 587
 588 fallback:
 589         fallback = NULL;
 590         /*
 591          * we have failed to find any workers, just
 592          * return the first one we can find.
 593          */
 594         if (!list_empty(&workers->worker_list))
 595                 fallback = workers->worker_list.next;
 596         if (!list_empty(&workers->idle_list))
 597                 fallback = workers->idle_list.next;
 598         BUG_ON(!fallback);
 599         worker = list_entry(fallback,
 600                   struct btrfs_worker_thread, worker_list);
 601 found:
 602         /*
 603          * this makes sure the worker doesn't exit before it is placed
 604          * onto a busy/idle list
 605          */
 606         atomic_inc(&worker->num_pending);
 607         spin_unlock_irqrestore(&workers->lock, flags);
 608         return worker;
 609 }
 610
 611 /*
 612  * btrfs_requeue_work just puts the work item back on the tail of the list
 613  * it was taken from.  It is intended for use with long running work functions
 614  * that make some progress and want to give the cpu up for others.
 615  */
 616 void btrfs_requeue_work(struct btrfs_work *work)
 617 {
 618         struct btrfs_worker_thread *worker = work->worker;
 619         unsigned long flags;
 620         int wake = 0;
 621
 622         if (test_and_set_bit(WORK_QUEUED_BIT, &work->flags))
 623                 return;
 624
 625         spin_lock_irqsave(&worker->lock, flags);
 626         if (test_bit(WORK_HIGH_PRIO_BIT, &work->flags))
 627                 list_add_tail(&work->list, &worker->prio_pending);
 628         else
 629                 list_add_tail(&work->list, &worker->pending);
 630         atomic_inc(&worker->num_pending);
 631
 632         /* by definition we're busy, take ourselves off the idle
 633          * list
 634          */
 635         if (worker->idle) {
 636                 spin_lock(&worker->workers->lock);
 637                 worker->idle = 0;
 638                 list_move_tail(&worker->worker_list,
 639                               &worker->workers->worker_list);
 640                 spin_unlock(&worker->workers->lock);
 641         }
 642         if (!worker->working) {
 643                 wake = 1;
 644                 worker->working = 1;
 645         }
 646
 647         if (wake)
 648                 wake_up_process(worker->task);
 649         spin_unlock_irqrestore(&worker->lock, flags);
 650 }
 651
 652 void btrfs_set_work_high_prio(struct btrfs_work *work)
 653 {
 654         set_bit(WORK_HIGH_PRIO_BIT, &work->flags);
 655 }
 656
 657 /*
 658  * places a struct btrfs_work into the pending queue of one of the kthreads
 659  */
 660 void btrfs_queue_worker(struct btrfs_workers *workers, struct btrfs_work *work)
 661 {
 662         struct btrfs_worker_thread *worker;
 663         unsigned long flags;
 664         int wake = 0;
 665
 666         /* don't requeue something already on a list */
 667         if (test_and_set_bit(WORK_QUEUED_BIT, &work->flags))
 668                 return;
 669
 670         worker = find_worker(workers);
 671         if (workers->ordered) {
 672                 /*
 673                  * you're not allowed to do ordered queues from an
 674                  * interrupt handler
 675                  */
 676                 spin_lock(&workers->order_lock);
 677                 if (test_bit(WORK_HIGH_PRIO_BIT, &work->flags)) {
 678                         list_add_tail(&work->order_list,
 679                                       &workers->prio_order_list);
 680                 } else {
 681                         list_add_tail(&work->order_list, &workers->order_list);
 682                 }
 683                 spin_unlock(&workers->order_lock);
 684         } else {
 685                 INIT_LIST_HEAD(&work->order_list);
 686         }
 687
 688         spin_lock_irqsave(&worker->lock, flags);
 689
 690         if (test_bit(WORK_HIGH_PRIO_BIT, &work->flags))
 691                 list_add_tail(&work->list, &worker->prio_pending);
 692         else
 693                 list_add_tail(&work->list, &worker->pending);
 694         check_busy_worker(worker);
 695
 696         /*
 697          * avoid calling into wake_up_process if this thread has already
 698          * been kicked
 699          */
 700         if (!worker->working)
 701                 wake = 1;
 702         worker->working = 1;
 703
 704         if (wake)
 705                 wake_up_process(worker->task);
 706         spin_unlock_irqrestore(&worker->lock, flags);
 707 }