linux/fs/xfs/xfs_trans_ail.c
<<
>>
Prefs
   1/*
   2 * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
   3 * Copyright (c) 2008 Dave Chinner
   4 * All Rights Reserved.
   5 *
   6 * This program is free software; you can redistribute it and/or
   7 * modify it under the terms of the GNU General Public License as
   8 * published by the Free Software Foundation.
   9 *
  10 * This program is distributed in the hope that it would be useful,
  11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13 * GNU General Public License for more details.
  14 *
  15 * You should have received a copy of the GNU General Public License
  16 * along with this program; if not, write the Free Software Foundation,
  17 * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  18 */
  19#include "xfs.h"
  20#include "xfs_fs.h"
  21#include "xfs_format.h"
  22#include "xfs_log_format.h"
  23#include "xfs_trans_resv.h"
  24#include "xfs_mount.h"
  25#include "xfs_trans.h"
  26#include "xfs_trans_priv.h"
  27#include "xfs_trace.h"
  28#include "xfs_error.h"
  29#include "xfs_log.h"
  30
  31#ifdef DEBUG
  32/*
  33 * Check that the list is sorted as it should be.
  34 */
  35STATIC void
  36xfs_ail_check(
  37        struct xfs_ail  *ailp,
  38        xfs_log_item_t  *lip)
  39{
  40        xfs_log_item_t  *prev_lip;
  41
  42        if (list_empty(&ailp->xa_ail))
  43                return;
  44
  45        /*
  46         * Check the next and previous entries are valid.
  47         */
  48        ASSERT((lip->li_flags & XFS_LI_IN_AIL) != 0);
  49        prev_lip = list_entry(lip->li_ail.prev, xfs_log_item_t, li_ail);
  50        if (&prev_lip->li_ail != &ailp->xa_ail)
  51                ASSERT(XFS_LSN_CMP(prev_lip->li_lsn, lip->li_lsn) <= 0);
  52
  53        prev_lip = list_entry(lip->li_ail.next, xfs_log_item_t, li_ail);
  54        if (&prev_lip->li_ail != &ailp->xa_ail)
  55                ASSERT(XFS_LSN_CMP(prev_lip->li_lsn, lip->li_lsn) >= 0);
  56
  57
  58}
  59#else /* !DEBUG */
  60#define xfs_ail_check(a,l)
  61#endif /* DEBUG */
  62
  63/*
  64 * Return a pointer to the last item in the AIL.  If the AIL is empty, then
  65 * return NULL.
  66 */
  67static xfs_log_item_t *
  68xfs_ail_max(
  69        struct xfs_ail  *ailp)
  70{
  71        if (list_empty(&ailp->xa_ail))
  72                return NULL;
  73
  74        return list_entry(ailp->xa_ail.prev, xfs_log_item_t, li_ail);
  75}
  76
  77/*
  78 * Return a pointer to the item which follows the given item in the AIL.  If
  79 * the given item is the last item in the list, then return NULL.
  80 */
  81static xfs_log_item_t *
  82xfs_ail_next(
  83        struct xfs_ail  *ailp,
  84        xfs_log_item_t  *lip)
  85{
  86        if (lip->li_ail.next == &ailp->xa_ail)
  87                return NULL;
  88
  89        return list_first_entry(&lip->li_ail, xfs_log_item_t, li_ail);
  90}
  91
  92/*
  93 * This is called by the log manager code to determine the LSN of the tail of
  94 * the log.  This is exactly the LSN of the first item in the AIL.  If the AIL
  95 * is empty, then this function returns 0.
  96 *
  97 * We need the AIL lock in order to get a coherent read of the lsn of the last
  98 * item in the AIL.
  99 */
 100xfs_lsn_t
 101xfs_ail_min_lsn(
 102        struct xfs_ail  *ailp)
 103{
 104        xfs_lsn_t       lsn = 0;
 105        xfs_log_item_t  *lip;
 106
 107        spin_lock(&ailp->xa_lock);
 108        lip = xfs_ail_min(ailp);
 109        if (lip)
 110                lsn = lip->li_lsn;
 111        spin_unlock(&ailp->xa_lock);
 112
 113        return lsn;
 114}
 115
 116/*
 117 * Return the maximum lsn held in the AIL, or zero if the AIL is empty.
 118 */
 119static xfs_lsn_t
 120xfs_ail_max_lsn(
 121        struct xfs_ail  *ailp)
 122{
 123        xfs_lsn_t       lsn = 0;
 124        xfs_log_item_t  *lip;
 125
 126        spin_lock(&ailp->xa_lock);
 127        lip = xfs_ail_max(ailp);
 128        if (lip)
 129                lsn = lip->li_lsn;
 130        spin_unlock(&ailp->xa_lock);
 131
 132        return lsn;
 133}
 134
 135/*
 136 * The cursor keeps track of where our current traversal is up to by tracking
 137 * the next item in the list for us. However, for this to be safe, removing an
 138 * object from the AIL needs to invalidate any cursor that points to it. hence
 139 * the traversal cursor needs to be linked to the struct xfs_ail so that
 140 * deletion can search all the active cursors for invalidation.
 141 */
 142STATIC void
 143xfs_trans_ail_cursor_init(
 144        struct xfs_ail          *ailp,
 145        struct xfs_ail_cursor   *cur)
 146{
 147        cur->item = NULL;
 148        list_add_tail(&cur->list, &ailp->xa_cursors);
 149}
 150
 151/*
 152 * Get the next item in the traversal and advance the cursor.  If the cursor
 153 * was invalidated (indicated by a lip of 1), restart the traversal.
 154 */
 155struct xfs_log_item *
 156xfs_trans_ail_cursor_next(
 157        struct xfs_ail          *ailp,
 158        struct xfs_ail_cursor   *cur)
 159{
 160        struct xfs_log_item     *lip = cur->item;
 161
 162        if ((uintptr_t)lip & 1)
 163                lip = xfs_ail_min(ailp);
 164        if (lip)
 165                cur->item = xfs_ail_next(ailp, lip);
 166        return lip;
 167}
 168
 169/*
 170 * When the traversal is complete, we need to remove the cursor from the list
 171 * of traversing cursors.
 172 */
 173void
 174xfs_trans_ail_cursor_done(
 175        struct xfs_ail_cursor   *cur)
 176{
 177        cur->item = NULL;
 178        list_del_init(&cur->list);
 179}
 180
 181/*
 182 * Invalidate any cursor that is pointing to this item. This is called when an
 183 * item is removed from the AIL. Any cursor pointing to this object is now
 184 * invalid and the traversal needs to be terminated so it doesn't reference a
 185 * freed object. We set the low bit of the cursor item pointer so we can
 186 * distinguish between an invalidation and the end of the list when getting the
 187 * next item from the cursor.
 188 */
 189STATIC void
 190xfs_trans_ail_cursor_clear(
 191        struct xfs_ail          *ailp,
 192        struct xfs_log_item     *lip)
 193{
 194        struct xfs_ail_cursor   *cur;
 195
 196        list_for_each_entry(cur, &ailp->xa_cursors, list) {
 197                if (cur->item == lip)
 198                        cur->item = (struct xfs_log_item *)
 199                                        ((uintptr_t)cur->item | 1);
 200        }
 201}
 202
 203/*
 204 * Find the first item in the AIL with the given @lsn by searching in ascending
 205 * LSN order and initialise the cursor to point to the next item for a
 206 * ascending traversal.  Pass a @lsn of zero to initialise the cursor to the
 207 * first item in the AIL. Returns NULL if the list is empty.
 208 */
 209xfs_log_item_t *
 210xfs_trans_ail_cursor_first(
 211        struct xfs_ail          *ailp,
 212        struct xfs_ail_cursor   *cur,
 213        xfs_lsn_t               lsn)
 214{
 215        xfs_log_item_t          *lip;
 216
 217        xfs_trans_ail_cursor_init(ailp, cur);
 218
 219        if (lsn == 0) {
 220                lip = xfs_ail_min(ailp);
 221                goto out;
 222        }
 223
 224        list_for_each_entry(lip, &ailp->xa_ail, li_ail) {
 225                if (XFS_LSN_CMP(lip->li_lsn, lsn) >= 0)
 226                        goto out;
 227        }
 228        return NULL;
 229
 230out:
 231        if (lip)
 232                cur->item = xfs_ail_next(ailp, lip);
 233        return lip;
 234}
 235
 236static struct xfs_log_item *
 237__xfs_trans_ail_cursor_last(
 238        struct xfs_ail          *ailp,
 239        xfs_lsn_t               lsn)
 240{
 241        xfs_log_item_t          *lip;
 242
 243        list_for_each_entry_reverse(lip, &ailp->xa_ail, li_ail) {
 244                if (XFS_LSN_CMP(lip->li_lsn, lsn) <= 0)
 245                        return lip;
 246        }
 247        return NULL;
 248}
 249
 250/*
 251 * Find the last item in the AIL with the given @lsn by searching in descending
 252 * LSN order and initialise the cursor to point to that item.  If there is no
 253 * item with the value of @lsn, then it sets the cursor to the last item with an
 254 * LSN lower than @lsn.  Returns NULL if the list is empty.
 255 */
 256struct xfs_log_item *
 257xfs_trans_ail_cursor_last(
 258        struct xfs_ail          *ailp,
 259        struct xfs_ail_cursor   *cur,
 260        xfs_lsn_t               lsn)
 261{
 262        xfs_trans_ail_cursor_init(ailp, cur);
 263        cur->item = __xfs_trans_ail_cursor_last(ailp, lsn);
 264        return cur->item;
 265}
 266
 267/*
 268 * Splice the log item list into the AIL at the given LSN. We splice to the
 269 * tail of the given LSN to maintain insert order for push traversals. The
 270 * cursor is optional, allowing repeated updates to the same LSN to avoid
 271 * repeated traversals.  This should not be called with an empty list.
 272 */
 273static void
 274xfs_ail_splice(
 275        struct xfs_ail          *ailp,
 276        struct xfs_ail_cursor   *cur,
 277        struct list_head        *list,
 278        xfs_lsn_t               lsn)
 279{
 280        struct xfs_log_item     *lip;
 281
 282        ASSERT(!list_empty(list));
 283
 284        /*
 285         * Use the cursor to determine the insertion point if one is
 286         * provided.  If not, or if the one we got is not valid,
 287         * find the place in the AIL where the items belong.
 288         */
 289        lip = cur ? cur->item : NULL;
 290        if (!lip || (uintptr_t)lip & 1)
 291                lip = __xfs_trans_ail_cursor_last(ailp, lsn);
 292
 293        /*
 294         * If a cursor is provided, we know we're processing the AIL
 295         * in lsn order, and future items to be spliced in will
 296         * follow the last one being inserted now.  Update the
 297         * cursor to point to that last item, now while we have a
 298         * reliable pointer to it.
 299         */
 300        if (cur)
 301                cur->item = list_entry(list->prev, struct xfs_log_item, li_ail);
 302
 303        /*
 304         * Finally perform the splice.  Unless the AIL was empty,
 305         * lip points to the item in the AIL _after_ which the new
 306         * items should go.  If lip is null the AIL was empty, so
 307         * the new items go at the head of the AIL.
 308         */
 309        if (lip)
 310                list_splice(list, &lip->li_ail);
 311        else
 312                list_splice(list, &ailp->xa_ail);
 313}
 314
 315/*
 316 * Delete the given item from the AIL.  Return a pointer to the item.
 317 */
 318static void
 319xfs_ail_delete(
 320        struct xfs_ail  *ailp,
 321        xfs_log_item_t  *lip)
 322{
 323        xfs_ail_check(ailp, lip);
 324        list_del(&lip->li_ail);
 325        xfs_trans_ail_cursor_clear(ailp, lip);
 326}
 327
 328static inline uint
 329xfsaild_push_item(
 330        struct xfs_ail          *ailp,
 331        struct xfs_log_item     *lip)
 332{
 333        /*
 334         * If log item pinning is enabled, skip the push and track the item as
 335         * pinned. This can help induce head-behind-tail conditions.
 336         */
 337        if (XFS_TEST_ERROR(false, ailp->xa_mount, XFS_ERRTAG_LOG_ITEM_PIN))
 338                return XFS_ITEM_PINNED;
 339
 340        return lip->li_ops->iop_push(lip, &ailp->xa_buf_list);
 341}
 342
 343static long
 344xfsaild_push(
 345        struct xfs_ail          *ailp)
 346{
 347        xfs_mount_t             *mp = ailp->xa_mount;
 348        struct xfs_ail_cursor   cur;
 349        xfs_log_item_t          *lip;
 350        xfs_lsn_t               lsn;
 351        xfs_lsn_t               target;
 352        long                    tout;
 353        int                     stuck = 0;
 354        int                     flushing = 0;
 355        int                     count = 0;
 356
 357        /*
 358         * If we encountered pinned items or did not finish writing out all
 359         * buffers the last time we ran, force the log first and wait for it
 360         * before pushing again.
 361         */
 362        if (ailp->xa_log_flush && ailp->xa_last_pushed_lsn == 0 &&
 363            (!list_empty_careful(&ailp->xa_buf_list) ||
 364             xfs_ail_min_lsn(ailp))) {
 365                ailp->xa_log_flush = 0;
 366
 367                XFS_STATS_INC(mp, xs_push_ail_flush);
 368                xfs_log_force(mp, XFS_LOG_SYNC);
 369        }
 370
 371        spin_lock(&ailp->xa_lock);
 372
 373        /* barrier matches the xa_target update in xfs_ail_push() */
 374        smp_rmb();
 375        target = ailp->xa_target;
 376        ailp->xa_target_prev = target;
 377
 378        lip = xfs_trans_ail_cursor_first(ailp, &cur, ailp->xa_last_pushed_lsn);
 379        if (!lip) {
 380                /*
 381                 * If the AIL is empty or our push has reached the end we are
 382                 * done now.
 383                 */
 384                xfs_trans_ail_cursor_done(&cur);
 385                spin_unlock(&ailp->xa_lock);
 386                goto out_done;
 387        }
 388
 389        XFS_STATS_INC(mp, xs_push_ail);
 390
 391        lsn = lip->li_lsn;
 392        while ((XFS_LSN_CMP(lip->li_lsn, target) <= 0)) {
 393                int     lock_result;
 394
 395                /*
 396                 * Note that iop_push may unlock and reacquire the AIL lock.  We
 397                 * rely on the AIL cursor implementation to be able to deal with
 398                 * the dropped lock.
 399                 */
 400                lock_result = xfsaild_push_item(ailp, lip);
 401                switch (lock_result) {
 402                case XFS_ITEM_SUCCESS:
 403                        XFS_STATS_INC(mp, xs_push_ail_success);
 404                        trace_xfs_ail_push(lip);
 405
 406                        ailp->xa_last_pushed_lsn = lsn;
 407                        break;
 408
 409                case XFS_ITEM_FLUSHING:
 410                        /*
 411                         * The item or its backing buffer is already beeing
 412                         * flushed.  The typical reason for that is that an
 413                         * inode buffer is locked because we already pushed the
 414                         * updates to it as part of inode clustering.
 415                         *
 416                         * We do not want to to stop flushing just because lots
 417                         * of items are already beeing flushed, but we need to
 418                         * re-try the flushing relatively soon if most of the
 419                         * AIL is beeing flushed.
 420                         */
 421                        XFS_STATS_INC(mp, xs_push_ail_flushing);
 422                        trace_xfs_ail_flushing(lip);
 423
 424                        flushing++;
 425                        ailp->xa_last_pushed_lsn = lsn;
 426                        break;
 427
 428                case XFS_ITEM_PINNED:
 429                        XFS_STATS_INC(mp, xs_push_ail_pinned);
 430                        trace_xfs_ail_pinned(lip);
 431
 432                        stuck++;
 433                        ailp->xa_log_flush++;
 434                        break;
 435                case XFS_ITEM_LOCKED:
 436                        XFS_STATS_INC(mp, xs_push_ail_locked);
 437                        trace_xfs_ail_locked(lip);
 438
 439                        stuck++;
 440                        break;
 441                default:
 442                        ASSERT(0);
 443                        break;
 444                }
 445
 446                count++;
 447
 448                /*
 449                 * Are there too many items we can't do anything with?
 450                 *
 451                 * If we we are skipping too many items because we can't flush
 452                 * them or they are already being flushed, we back off and
 453                 * given them time to complete whatever operation is being
 454                 * done. i.e. remove pressure from the AIL while we can't make
 455                 * progress so traversals don't slow down further inserts and
 456                 * removals to/from the AIL.
 457                 *
 458                 * The value of 100 is an arbitrary magic number based on
 459                 * observation.
 460                 */
 461                if (stuck > 100)
 462                        break;
 463
 464                lip = xfs_trans_ail_cursor_next(ailp, &cur);
 465                if (lip == NULL)
 466                        break;
 467                lsn = lip->li_lsn;
 468        }
 469        xfs_trans_ail_cursor_done(&cur);
 470        spin_unlock(&ailp->xa_lock);
 471
 472        if (xfs_buf_delwri_submit_nowait(&ailp->xa_buf_list))
 473                ailp->xa_log_flush++;
 474
 475        if (!count || XFS_LSN_CMP(lsn, target) >= 0) {
 476out_done:
 477                /*
 478                 * We reached the target or the AIL is empty, so wait a bit
 479                 * longer for I/O to complete and remove pushed items from the
 480                 * AIL before we start the next scan from the start of the AIL.
 481                 */
 482                tout = 50;
 483                ailp->xa_last_pushed_lsn = 0;
 484        } else if (((stuck + flushing) * 100) / count > 90) {
 485                /*
 486                 * Either there is a lot of contention on the AIL or we are
 487                 * stuck due to operations in progress. "Stuck" in this case
 488                 * is defined as >90% of the items we tried to push were stuck.
 489                 *
 490                 * Backoff a bit more to allow some I/O to complete before
 491                 * restarting from the start of the AIL. This prevents us from
 492                 * spinning on the same items, and if they are pinned will all
 493                 * the restart to issue a log force to unpin the stuck items.
 494                 */
 495                tout = 20;
 496                ailp->xa_last_pushed_lsn = 0;
 497        } else {
 498                /*
 499                 * Assume we have more work to do in a short while.
 500                 */
 501                tout = 10;
 502        }
 503
 504        return tout;
 505}
 506
 507static int
 508xfsaild(
 509        void            *data)
 510{
 511        struct xfs_ail  *ailp = data;
 512        long            tout = 0;       /* milliseconds */
 513
 514        current->flags |= PF_MEMALLOC;
 515        set_freezable();
 516
 517        while (!kthread_should_stop()) {
 518                if (tout && tout <= 20)
 519                        __set_current_state(TASK_KILLABLE);
 520                else
 521                        __set_current_state(TASK_INTERRUPTIBLE);
 522
 523                spin_lock(&ailp->xa_lock);
 524
 525                /*
 526                 * Idle if the AIL is empty and we are not racing with a target
 527                 * update. We check the AIL after we set the task to a sleep
 528                 * state to guarantee that we either catch an xa_target update
 529                 * or that a wake_up resets the state to TASK_RUNNING.
 530                 * Otherwise, we run the risk of sleeping indefinitely.
 531                 *
 532                 * The barrier matches the xa_target update in xfs_ail_push().
 533                 */
 534                smp_rmb();
 535                if (!xfs_ail_min(ailp) &&
 536                    ailp->xa_target == ailp->xa_target_prev) {
 537                        spin_unlock(&ailp->xa_lock);
 538                        freezable_schedule();
 539                        tout = 0;
 540                        continue;
 541                }
 542                spin_unlock(&ailp->xa_lock);
 543
 544                if (tout)
 545                        freezable_schedule_timeout(msecs_to_jiffies(tout));
 546
 547                __set_current_state(TASK_RUNNING);
 548
 549                try_to_freeze();
 550
 551                tout = xfsaild_push(ailp);
 552        }
 553
 554        return 0;
 555}
 556
 557/*
 558 * This routine is called to move the tail of the AIL forward.  It does this by
 559 * trying to flush items in the AIL whose lsns are below the given
 560 * threshold_lsn.
 561 *
 562 * The push is run asynchronously in a workqueue, which means the caller needs
 563 * to handle waiting on the async flush for space to become available.
 564 * We don't want to interrupt any push that is in progress, hence we only queue
 565 * work if we set the pushing bit approriately.
 566 *
 567 * We do this unlocked - we only need to know whether there is anything in the
 568 * AIL at the time we are called. We don't need to access the contents of
 569 * any of the objects, so the lock is not needed.
 570 */
 571void
 572xfs_ail_push(
 573        struct xfs_ail  *ailp,
 574        xfs_lsn_t       threshold_lsn)
 575{
 576        xfs_log_item_t  *lip;
 577
 578        lip = xfs_ail_min(ailp);
 579        if (!lip || XFS_FORCED_SHUTDOWN(ailp->xa_mount) ||
 580            XFS_LSN_CMP(threshold_lsn, ailp->xa_target) <= 0)
 581                return;
 582
 583        /*
 584         * Ensure that the new target is noticed in push code before it clears
 585         * the XFS_AIL_PUSHING_BIT.
 586         */
 587        smp_wmb();
 588        xfs_trans_ail_copy_lsn(ailp, &ailp->xa_target, &threshold_lsn);
 589        smp_wmb();
 590
 591        wake_up_process(ailp->xa_task);
 592}
 593
 594/*
 595 * Push out all items in the AIL immediately
 596 */
 597void
 598xfs_ail_push_all(
 599        struct xfs_ail  *ailp)
 600{
 601        xfs_lsn_t       threshold_lsn = xfs_ail_max_lsn(ailp);
 602
 603        if (threshold_lsn)
 604                xfs_ail_push(ailp, threshold_lsn);
 605}
 606
 607/*
 608 * Push out all items in the AIL immediately and wait until the AIL is empty.
 609 */
 610void
 611xfs_ail_push_all_sync(
 612        struct xfs_ail  *ailp)
 613{
 614        struct xfs_log_item     *lip;
 615        DEFINE_WAIT(wait);
 616
 617        spin_lock(&ailp->xa_lock);
 618        while ((lip = xfs_ail_max(ailp)) != NULL) {
 619                prepare_to_wait(&ailp->xa_empty, &wait, TASK_UNINTERRUPTIBLE);
 620                ailp->xa_target = lip->li_lsn;
 621                wake_up_process(ailp->xa_task);
 622                spin_unlock(&ailp->xa_lock);
 623                schedule();
 624                spin_lock(&ailp->xa_lock);
 625        }
 626        spin_unlock(&ailp->xa_lock);
 627
 628        finish_wait(&ailp->xa_empty, &wait);
 629}
 630
 631/*
 632 * xfs_trans_ail_update - bulk AIL insertion operation.
 633 *
 634 * @xfs_trans_ail_update takes an array of log items that all need to be
 635 * positioned at the same LSN in the AIL. If an item is not in the AIL, it will
 636 * be added.  Otherwise, it will be repositioned  by removing it and re-adding
 637 * it to the AIL. If we move the first item in the AIL, update the log tail to
 638 * match the new minimum LSN in the AIL.
 639 *
 640 * This function takes the AIL lock once to execute the update operations on
 641 * all the items in the array, and as such should not be called with the AIL
 642 * lock held. As a result, once we have the AIL lock, we need to check each log
 643 * item LSN to confirm it needs to be moved forward in the AIL.
 644 *
 645 * To optimise the insert operation, we delete all the items from the AIL in
 646 * the first pass, moving them into a temporary list, then splice the temporary
 647 * list into the correct position in the AIL. This avoids needing to do an
 648 * insert operation on every item.
 649 *
 650 * This function must be called with the AIL lock held.  The lock is dropped
 651 * before returning.
 652 */
 653void
 654xfs_trans_ail_update_bulk(
 655        struct xfs_ail          *ailp,
 656        struct xfs_ail_cursor   *cur,
 657        struct xfs_log_item     **log_items,
 658        int                     nr_items,
 659        xfs_lsn_t               lsn) __releases(ailp->xa_lock)
 660{
 661        xfs_log_item_t          *mlip;
 662        int                     mlip_changed = 0;
 663        int                     i;
 664        LIST_HEAD(tmp);
 665
 666        ASSERT(nr_items > 0);           /* Not required, but true. */
 667        mlip = xfs_ail_min(ailp);
 668
 669        for (i = 0; i < nr_items; i++) {
 670                struct xfs_log_item *lip = log_items[i];
 671                if (lip->li_flags & XFS_LI_IN_AIL) {
 672                        /* check if we really need to move the item */
 673                        if (XFS_LSN_CMP(lsn, lip->li_lsn) <= 0)
 674                                continue;
 675
 676                        trace_xfs_ail_move(lip, lip->li_lsn, lsn);
 677                        xfs_ail_delete(ailp, lip);
 678                        if (mlip == lip)
 679                                mlip_changed = 1;
 680                } else {
 681                        lip->li_flags |= XFS_LI_IN_AIL;
 682                        trace_xfs_ail_insert(lip, 0, lsn);
 683                }
 684                lip->li_lsn = lsn;
 685                list_add(&lip->li_ail, &tmp);
 686        }
 687
 688        if (!list_empty(&tmp))
 689                xfs_ail_splice(ailp, cur, &tmp, lsn);
 690
 691        if (mlip_changed) {
 692                if (!XFS_FORCED_SHUTDOWN(ailp->xa_mount))
 693                        xlog_assign_tail_lsn_locked(ailp->xa_mount);
 694                spin_unlock(&ailp->xa_lock);
 695
 696                xfs_log_space_wake(ailp->xa_mount);
 697        } else {
 698                spin_unlock(&ailp->xa_lock);
 699        }
 700}
 701
 702bool
 703xfs_ail_delete_one(
 704        struct xfs_ail          *ailp,
 705        struct xfs_log_item     *lip)
 706{
 707        struct xfs_log_item     *mlip = xfs_ail_min(ailp);
 708
 709        trace_xfs_ail_delete(lip, mlip->li_lsn, lip->li_lsn);
 710        xfs_ail_delete(ailp, lip);
 711        xfs_clear_li_failed(lip);
 712        lip->li_flags &= ~XFS_LI_IN_AIL;
 713        lip->li_lsn = 0;
 714
 715        return mlip == lip;
 716}
 717
 718/**
 719 * Remove a log items from the AIL
 720 *
 721 * @xfs_trans_ail_delete_bulk takes an array of log items that all need to
 722 * removed from the AIL. The caller is already holding the AIL lock, and done
 723 * all the checks necessary to ensure the items passed in via @log_items are
 724 * ready for deletion. This includes checking that the items are in the AIL.
 725 *
 726 * For each log item to be removed, unlink it  from the AIL, clear the IN_AIL
 727 * flag from the item and reset the item's lsn to 0. If we remove the first
 728 * item in the AIL, update the log tail to match the new minimum LSN in the
 729 * AIL.
 730 *
 731 * This function will not drop the AIL lock until all items are removed from
 732 * the AIL to minimise the amount of lock traffic on the AIL. This does not
 733 * greatly increase the AIL hold time, but does significantly reduce the amount
 734 * of traffic on the lock, especially during IO completion.
 735 *
 736 * This function must be called with the AIL lock held.  The lock is dropped
 737 * before returning.
 738 */
 739void
 740xfs_trans_ail_delete(
 741        struct xfs_ail          *ailp,
 742        struct xfs_log_item     *lip,
 743        int                     shutdown_type) __releases(ailp->xa_lock)
 744{
 745        struct xfs_mount        *mp = ailp->xa_mount;
 746        bool                    mlip_changed;
 747
 748        if (!(lip->li_flags & XFS_LI_IN_AIL)) {
 749                spin_unlock(&ailp->xa_lock);
 750                if (!XFS_FORCED_SHUTDOWN(mp)) {
 751                        xfs_alert_tag(mp, XFS_PTAG_AILDELETE,
 752        "%s: attempting to delete a log item that is not in the AIL",
 753                                        __func__);
 754                        xfs_force_shutdown(mp, shutdown_type);
 755                }
 756                return;
 757        }
 758
 759        mlip_changed = xfs_ail_delete_one(ailp, lip);
 760        if (mlip_changed) {
 761                if (!XFS_FORCED_SHUTDOWN(mp))
 762                        xlog_assign_tail_lsn_locked(mp);
 763                if (list_empty(&ailp->xa_ail))
 764                        wake_up_all(&ailp->xa_empty);
 765        }
 766
 767        spin_unlock(&ailp->xa_lock);
 768        if (mlip_changed)
 769                xfs_log_space_wake(ailp->xa_mount);
 770}
 771
 772int
 773xfs_trans_ail_init(
 774        xfs_mount_t     *mp)
 775{
 776        struct xfs_ail  *ailp;
 777
 778        ailp = kmem_zalloc(sizeof(struct xfs_ail), KM_MAYFAIL);
 779        if (!ailp)
 780                return -ENOMEM;
 781
 782        ailp->xa_mount = mp;
 783        INIT_LIST_HEAD(&ailp->xa_ail);
 784        INIT_LIST_HEAD(&ailp->xa_cursors);
 785        spin_lock_init(&ailp->xa_lock);
 786        INIT_LIST_HEAD(&ailp->xa_buf_list);
 787        init_waitqueue_head(&ailp->xa_empty);
 788
 789        ailp->xa_task = kthread_run(xfsaild, ailp, "xfsaild/%s",
 790                        ailp->xa_mount->m_fsname);
 791        if (IS_ERR(ailp->xa_task))
 792                goto out_free_ailp;
 793
 794        mp->m_ail = ailp;
 795        return 0;
 796
 797out_free_ailp:
 798        kmem_free(ailp);
 799        return -ENOMEM;
 800}
 801
 802void
 803xfs_trans_ail_destroy(
 804        xfs_mount_t     *mp)
 805{
 806        struct xfs_ail  *ailp = mp->m_ail;
 807
 808        kthread_stop(ailp->xa_task);
 809        kmem_free(ailp);
 810}
 811