linux/fs/xfs/xfs_icache.c
<<
>>
Prefs
   1/*
   2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
   3 * All Rights Reserved.
   4 *
   5 * This program is free software; you can redistribute it and/or
   6 * modify it under the terms of the GNU General Public License as
   7 * published by the Free Software Foundation.
   8 *
   9 * This program is distributed in the hope that it would be useful,
  10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12 * GNU General Public License for more details.
  13 *
  14 * You should have received a copy of the GNU General Public License
  15 * along with this program; if not, write the Free Software Foundation,
  16 * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17 */
  18#include "xfs.h"
  19#include "xfs_fs.h"
  20#include "xfs_types.h"
  21#include "xfs_log.h"
  22#include "xfs_log_priv.h"
  23#include "xfs_inum.h"
  24#include "xfs_trans.h"
  25#include "xfs_trans_priv.h"
  26#include "xfs_sb.h"
  27#include "xfs_ag.h"
  28#include "xfs_mount.h"
  29#include "xfs_bmap_btree.h"
  30#include "xfs_inode.h"
  31#include "xfs_dinode.h"
  32#include "xfs_error.h"
  33#include "xfs_filestream.h"
  34#include "xfs_vnodeops.h"
  35#include "xfs_inode_item.h"
  36#include "xfs_quota.h"
  37#include "xfs_trace.h"
  38#include "xfs_fsops.h"
  39#include "xfs_icache.h"
  40
  41#include <linux/kthread.h>
  42#include <linux/freezer.h>
  43
  44STATIC void __xfs_inode_clear_reclaim_tag(struct xfs_mount *mp,
  45                                struct xfs_perag *pag, struct xfs_inode *ip);
  46
  47/*
  48 * Allocate and initialise an xfs_inode.
  49 */
  50STATIC struct xfs_inode *
  51xfs_inode_alloc(
  52        struct xfs_mount        *mp,
  53        xfs_ino_t               ino)
  54{
  55        struct xfs_inode        *ip;
  56
  57        /*
  58         * if this didn't occur in transactions, we could use
  59         * KM_MAYFAIL and return NULL here on ENOMEM. Set the
  60         * code up to do this anyway.
  61         */
  62        ip = kmem_zone_alloc(xfs_inode_zone, KM_SLEEP);
  63        if (!ip)
  64                return NULL;
  65        if (inode_init_always(mp->m_super, VFS_I(ip))) {
  66                kmem_zone_free(xfs_inode_zone, ip);
  67                return NULL;
  68        }
  69
  70        ASSERT(atomic_read(&ip->i_pincount) == 0);
  71        ASSERT(!spin_is_locked(&ip->i_flags_lock));
  72        ASSERT(!xfs_isiflocked(ip));
  73        ASSERT(ip->i_ino == 0);
  74
  75        mrlock_init(&ip->i_iolock, MRLOCK_BARRIER, "xfsio", ip->i_ino);
  76
  77        /* initialise the xfs inode */
  78        ip->i_ino = ino;
  79        ip->i_mount = mp;
  80        memset(&ip->i_imap, 0, sizeof(struct xfs_imap));
  81        ip->i_afp = NULL;
  82        memset(&ip->i_df, 0, sizeof(xfs_ifork_t));
  83        ip->i_flags = 0;
  84        ip->i_delayed_blks = 0;
  85        memset(&ip->i_d, 0, sizeof(xfs_icdinode_t));
  86
  87        return ip;
  88}
  89
  90STATIC void
  91xfs_inode_free_callback(
  92        struct rcu_head         *head)
  93{
  94        struct inode            *inode = container_of(head, struct inode, i_rcu);
  95        struct xfs_inode        *ip = XFS_I(inode);
  96
  97        kmem_zone_free(xfs_inode_zone, ip);
  98}
  99
 100STATIC void
 101xfs_inode_free(
 102        struct xfs_inode        *ip)
 103{
 104        switch (ip->i_d.di_mode & S_IFMT) {
 105        case S_IFREG:
 106        case S_IFDIR:
 107        case S_IFLNK:
 108                xfs_idestroy_fork(ip, XFS_DATA_FORK);
 109                break;
 110        }
 111
 112        if (ip->i_afp)
 113                xfs_idestroy_fork(ip, XFS_ATTR_FORK);
 114
 115        if (ip->i_itemp) {
 116                ASSERT(!(ip->i_itemp->ili_item.li_flags & XFS_LI_IN_AIL));
 117                xfs_inode_item_destroy(ip);
 118                ip->i_itemp = NULL;
 119        }
 120
 121        /* asserts to verify all state is correct here */
 122        ASSERT(atomic_read(&ip->i_pincount) == 0);
 123        ASSERT(!spin_is_locked(&ip->i_flags_lock));
 124        ASSERT(!xfs_isiflocked(ip));
 125
 126        /*
 127         * Because we use RCU freeing we need to ensure the inode always
 128         * appears to be reclaimed with an invalid inode number when in the
 129         * free state. The ip->i_flags_lock provides the barrier against lookup
 130         * races.
 131         */
 132        spin_lock(&ip->i_flags_lock);
 133        ip->i_flags = XFS_IRECLAIM;
 134        ip->i_ino = 0;
 135        spin_unlock(&ip->i_flags_lock);
 136
 137        call_rcu(&VFS_I(ip)->i_rcu, xfs_inode_free_callback);
 138}
 139
 140/*
 141 * Check the validity of the inode we just found it the cache
 142 */
 143static int
 144xfs_iget_cache_hit(
 145        struct xfs_perag        *pag,
 146        struct xfs_inode        *ip,
 147        xfs_ino_t               ino,
 148        int                     flags,
 149        int                     lock_flags) __releases(RCU)
 150{
 151        struct inode            *inode = VFS_I(ip);
 152        struct xfs_mount        *mp = ip->i_mount;
 153        int                     error;
 154
 155        /*
 156         * check for re-use of an inode within an RCU grace period due to the
 157         * radix tree nodes not being updated yet. We monitor for this by
 158         * setting the inode number to zero before freeing the inode structure.
 159         * If the inode has been reallocated and set up, then the inode number
 160         * will not match, so check for that, too.
 161         */
 162        spin_lock(&ip->i_flags_lock);
 163        if (ip->i_ino != ino) {
 164                trace_xfs_iget_skip(ip);
 165                XFS_STATS_INC(xs_ig_frecycle);
 166                error = EAGAIN;
 167                goto out_error;
 168        }
 169
 170
 171        /*
 172         * If we are racing with another cache hit that is currently
 173         * instantiating this inode or currently recycling it out of
 174         * reclaimabe state, wait for the initialisation to complete
 175         * before continuing.
 176         *
 177         * XXX(hch): eventually we should do something equivalent to
 178         *           wait_on_inode to wait for these flags to be cleared
 179         *           instead of polling for it.
 180         */
 181        if (ip->i_flags & (XFS_INEW|XFS_IRECLAIM)) {
 182                trace_xfs_iget_skip(ip);
 183                XFS_STATS_INC(xs_ig_frecycle);
 184                error = EAGAIN;
 185                goto out_error;
 186        }
 187
 188        /*
 189         * If lookup is racing with unlink return an error immediately.
 190         */
 191        if (ip->i_d.di_mode == 0 && !(flags & XFS_IGET_CREATE)) {
 192                error = ENOENT;
 193                goto out_error;
 194        }
 195
 196        /*
 197         * If IRECLAIMABLE is set, we've torn down the VFS inode already.
 198         * Need to carefully get it back into useable state.
 199         */
 200        if (ip->i_flags & XFS_IRECLAIMABLE) {
 201                trace_xfs_iget_reclaim(ip);
 202
 203                /*
 204                 * We need to set XFS_IRECLAIM to prevent xfs_reclaim_inode
 205                 * from stomping over us while we recycle the inode.  We can't
 206                 * clear the radix tree reclaimable tag yet as it requires
 207                 * pag_ici_lock to be held exclusive.
 208                 */
 209                ip->i_flags |= XFS_IRECLAIM;
 210
 211                spin_unlock(&ip->i_flags_lock);
 212                rcu_read_unlock();
 213
 214                error = -inode_init_always(mp->m_super, inode);
 215                if (error) {
 216                        /*
 217                         * Re-initializing the inode failed, and we are in deep
 218                         * trouble.  Try to re-add it to the reclaim list.
 219                         */
 220                        rcu_read_lock();
 221                        spin_lock(&ip->i_flags_lock);
 222
 223                        ip->i_flags &= ~(XFS_INEW | XFS_IRECLAIM);
 224                        ASSERT(ip->i_flags & XFS_IRECLAIMABLE);
 225                        trace_xfs_iget_reclaim_fail(ip);
 226                        goto out_error;
 227                }
 228
 229                spin_lock(&pag->pag_ici_lock);
 230                spin_lock(&ip->i_flags_lock);
 231
 232                /*
 233                 * Clear the per-lifetime state in the inode as we are now
 234                 * effectively a new inode and need to return to the initial
 235                 * state before reuse occurs.
 236                 */
 237                ip->i_flags &= ~XFS_IRECLAIM_RESET_FLAGS;
 238                ip->i_flags |= XFS_INEW;
 239                __xfs_inode_clear_reclaim_tag(mp, pag, ip);
 240                inode->i_state = I_NEW;
 241
 242                ASSERT(!rwsem_is_locked(&ip->i_iolock.mr_lock));
 243                mrlock_init(&ip->i_iolock, MRLOCK_BARRIER, "xfsio", ip->i_ino);
 244
 245                spin_unlock(&ip->i_flags_lock);
 246                spin_unlock(&pag->pag_ici_lock);
 247        } else {
 248                /* If the VFS inode is being torn down, pause and try again. */
 249                if (!igrab(inode)) {
 250                        trace_xfs_iget_skip(ip);
 251                        error = EAGAIN;
 252                        goto out_error;
 253                }
 254
 255                /* We've got a live one. */
 256                spin_unlock(&ip->i_flags_lock);
 257                rcu_read_unlock();
 258                trace_xfs_iget_hit(ip);
 259        }
 260
 261        if (lock_flags != 0)
 262                xfs_ilock(ip, lock_flags);
 263
 264        xfs_iflags_clear(ip, XFS_ISTALE | XFS_IDONTCACHE);
 265        XFS_STATS_INC(xs_ig_found);
 266
 267        return 0;
 268
 269out_error:
 270        spin_unlock(&ip->i_flags_lock);
 271        rcu_read_unlock();
 272        return error;
 273}
 274
 275
 276static int
 277xfs_iget_cache_miss(
 278        struct xfs_mount        *mp,
 279        struct xfs_perag        *pag,
 280        xfs_trans_t             *tp,
 281        xfs_ino_t               ino,
 282        struct xfs_inode        **ipp,
 283        int                     flags,
 284        int                     lock_flags)
 285{
 286        struct xfs_inode        *ip;
 287        int                     error;
 288        xfs_agino_t             agino = XFS_INO_TO_AGINO(mp, ino);
 289        int                     iflags;
 290
 291        ip = xfs_inode_alloc(mp, ino);
 292        if (!ip)
 293                return ENOMEM;
 294
 295        error = xfs_iread(mp, tp, ip, flags);
 296        if (error)
 297                goto out_destroy;
 298
 299        trace_xfs_iget_miss(ip);
 300
 301        if ((ip->i_d.di_mode == 0) && !(flags & XFS_IGET_CREATE)) {
 302                error = ENOENT;
 303                goto out_destroy;
 304        }
 305
 306        /*
 307         * Preload the radix tree so we can insert safely under the
 308         * write spinlock. Note that we cannot sleep inside the preload
 309         * region. Since we can be called from transaction context, don't
 310         * recurse into the file system.
 311         */
 312        if (radix_tree_preload(GFP_NOFS)) {
 313                error = EAGAIN;
 314                goto out_destroy;
 315        }
 316
 317        /*
 318         * Because the inode hasn't been added to the radix-tree yet it can't
 319         * be found by another thread, so we can do the non-sleeping lock here.
 320         */
 321        if (lock_flags) {
 322                if (!xfs_ilock_nowait(ip, lock_flags))
 323                        BUG();
 324        }
 325
 326        /*
 327         * These values must be set before inserting the inode into the radix
 328         * tree as the moment it is inserted a concurrent lookup (allowed by the
 329         * RCU locking mechanism) can find it and that lookup must see that this
 330         * is an inode currently under construction (i.e. that XFS_INEW is set).
 331         * The ip->i_flags_lock that protects the XFS_INEW flag forms the
 332         * memory barrier that ensures this detection works correctly at lookup
 333         * time.
 334         */
 335        iflags = XFS_INEW;
 336        if (flags & XFS_IGET_DONTCACHE)
 337                iflags |= XFS_IDONTCACHE;
 338        ip->i_udquot = NULL;
 339        ip->i_gdquot = NULL;
 340        ip->i_pdquot = NULL;
 341        xfs_iflags_set(ip, iflags);
 342
 343        /* insert the new inode */
 344        spin_lock(&pag->pag_ici_lock);
 345        error = radix_tree_insert(&pag->pag_ici_root, agino, ip);
 346        if (unlikely(error)) {
 347                WARN_ON(error != -EEXIST);
 348                XFS_STATS_INC(xs_ig_dup);
 349                error = EAGAIN;
 350                goto out_preload_end;
 351        }
 352        spin_unlock(&pag->pag_ici_lock);
 353        radix_tree_preload_end();
 354
 355        *ipp = ip;
 356        return 0;
 357
 358out_preload_end:
 359        spin_unlock(&pag->pag_ici_lock);
 360        radix_tree_preload_end();
 361        if (lock_flags)
 362                xfs_iunlock(ip, lock_flags);
 363out_destroy:
 364        __destroy_inode(VFS_I(ip));
 365        xfs_inode_free(ip);
 366        return error;
 367}
 368
 369/*
 370 * Look up an inode by number in the given file system.
 371 * The inode is looked up in the cache held in each AG.
 372 * If the inode is found in the cache, initialise the vfs inode
 373 * if necessary.
 374 *
 375 * If it is not in core, read it in from the file system's device,
 376 * add it to the cache and initialise the vfs inode.
 377 *
 378 * The inode is locked according to the value of the lock_flags parameter.
 379 * This flag parameter indicates how and if the inode's IO lock and inode lock
 380 * should be taken.
 381 *
 382 * mp -- the mount point structure for the current file system.  It points
 383 *       to the inode hash table.
 384 * tp -- a pointer to the current transaction if there is one.  This is
 385 *       simply passed through to the xfs_iread() call.
 386 * ino -- the number of the inode desired.  This is the unique identifier
 387 *        within the file system for the inode being requested.
 388 * lock_flags -- flags indicating how to lock the inode.  See the comment
 389 *               for xfs_ilock() for a list of valid values.
 390 */
 391int
 392xfs_iget(
 393        xfs_mount_t     *mp,
 394        xfs_trans_t     *tp,
 395        xfs_ino_t       ino,
 396        uint            flags,
 397        uint            lock_flags,
 398        xfs_inode_t     **ipp)
 399{
 400        xfs_inode_t     *ip;
 401        int             error;
 402        xfs_perag_t     *pag;
 403        xfs_agino_t     agino;
 404
 405        /*
 406         * xfs_reclaim_inode() uses the ILOCK to ensure an inode
 407         * doesn't get freed while it's being referenced during a
 408         * radix tree traversal here.  It assumes this function
 409         * aqcuires only the ILOCK (and therefore it has no need to
 410         * involve the IOLOCK in this synchronization).
 411         */
 412        ASSERT((lock_flags & (XFS_IOLOCK_EXCL | XFS_IOLOCK_SHARED)) == 0);
 413
 414        /* reject inode numbers outside existing AGs */
 415        if (!ino || XFS_INO_TO_AGNO(mp, ino) >= mp->m_sb.sb_agcount)
 416                return EINVAL;
 417
 418        /* get the perag structure and ensure that it's inode capable */
 419        pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ino));
 420        agino = XFS_INO_TO_AGINO(mp, ino);
 421
 422again:
 423        error = 0;
 424        rcu_read_lock();
 425        ip = radix_tree_lookup(&pag->pag_ici_root, agino);
 426
 427        if (ip) {
 428                error = xfs_iget_cache_hit(pag, ip, ino, flags, lock_flags);
 429                if (error)
 430                        goto out_error_or_again;
 431        } else {
 432                rcu_read_unlock();
 433                XFS_STATS_INC(xs_ig_missed);
 434
 435                error = xfs_iget_cache_miss(mp, pag, tp, ino, &ip,
 436                                                        flags, lock_flags);
 437                if (error)
 438                        goto out_error_or_again;
 439        }
 440        xfs_perag_put(pag);
 441
 442        *ipp = ip;
 443
 444        /*
 445         * If we have a real type for an on-disk inode, we can set ops(&unlock)
 446         * now.  If it's a new inode being created, xfs_ialloc will handle it.
 447         */
 448        if (xfs_iflags_test(ip, XFS_INEW) && ip->i_d.di_mode != 0)
 449                xfs_setup_inode(ip);
 450        return 0;
 451
 452out_error_or_again:
 453        if (error == EAGAIN) {
 454                delay(1);
 455                goto again;
 456        }
 457        xfs_perag_put(pag);
 458        return error;
 459}
 460
 461/*
 462 * The inode lookup is done in batches to keep the amount of lock traffic and
 463 * radix tree lookups to a minimum. The batch size is a trade off between
 464 * lookup reduction and stack usage. This is in the reclaim path, so we can't
 465 * be too greedy.
 466 */
 467#define XFS_LOOKUP_BATCH        32
 468
 469STATIC int
 470xfs_inode_ag_walk_grab(
 471        struct xfs_inode        *ip)
 472{
 473        struct inode            *inode = VFS_I(ip);
 474
 475        ASSERT(rcu_read_lock_held());
 476
 477        /*
 478         * check for stale RCU freed inode
 479         *
 480         * If the inode has been reallocated, it doesn't matter if it's not in
 481         * the AG we are walking - we are walking for writeback, so if it
 482         * passes all the "valid inode" checks and is dirty, then we'll write
 483         * it back anyway.  If it has been reallocated and still being
 484         * initialised, the XFS_INEW check below will catch it.
 485         */
 486        spin_lock(&ip->i_flags_lock);
 487        if (!ip->i_ino)
 488                goto out_unlock_noent;
 489
 490        /* avoid new or reclaimable inodes. Leave for reclaim code to flush */
 491        if (__xfs_iflags_test(ip, XFS_INEW | XFS_IRECLAIMABLE | XFS_IRECLAIM))
 492                goto out_unlock_noent;
 493        spin_unlock(&ip->i_flags_lock);
 494
 495        /* nothing to sync during shutdown */
 496        if (XFS_FORCED_SHUTDOWN(ip->i_mount))
 497                return EFSCORRUPTED;
 498
 499        /* If we can't grab the inode, it must on it's way to reclaim. */
 500        if (!igrab(inode))
 501                return ENOENT;
 502
 503        if (is_bad_inode(inode)) {
 504                IRELE(ip);
 505                return ENOENT;
 506        }
 507
 508        /* inode is valid */
 509        return 0;
 510
 511out_unlock_noent:
 512        spin_unlock(&ip->i_flags_lock);
 513        return ENOENT;
 514}
 515
 516STATIC int
 517xfs_inode_ag_walk(
 518        struct xfs_mount        *mp,
 519        struct xfs_perag        *pag,
 520        int                     (*execute)(struct xfs_inode *ip,
 521                                           struct xfs_perag *pag, int flags,
 522                                           void *args),
 523        int                     flags,
 524        void                    *args,
 525        int                     tag)
 526{
 527        uint32_t                first_index;
 528        int                     last_error = 0;
 529        int                     skipped;
 530        int                     done;
 531        int                     nr_found;
 532
 533restart:
 534        done = 0;
 535        skipped = 0;
 536        first_index = 0;
 537        nr_found = 0;
 538        do {
 539                struct xfs_inode *batch[XFS_LOOKUP_BATCH];
 540                int             error = 0;
 541                int             i;
 542
 543                rcu_read_lock();
 544
 545                if (tag == -1)
 546                        nr_found = radix_tree_gang_lookup(&pag->pag_ici_root,
 547                                        (void **)batch, first_index,
 548                                        XFS_LOOKUP_BATCH);
 549                else
 550                        nr_found = radix_tree_gang_lookup_tag(
 551                                        &pag->pag_ici_root,
 552                                        (void **) batch, first_index,
 553                                        XFS_LOOKUP_BATCH, tag);
 554
 555                if (!nr_found) {
 556                        rcu_read_unlock();
 557                        break;
 558                }
 559
 560                /*
 561                 * Grab the inodes before we drop the lock. if we found
 562                 * nothing, nr == 0 and the loop will be skipped.
 563                 */
 564                for (i = 0; i < nr_found; i++) {
 565                        struct xfs_inode *ip = batch[i];
 566
 567                        if (done || xfs_inode_ag_walk_grab(ip))
 568                                batch[i] = NULL;
 569
 570                        /*
 571                         * Update the index for the next lookup. Catch
 572                         * overflows into the next AG range which can occur if
 573                         * we have inodes in the last block of the AG and we
 574                         * are currently pointing to the last inode.
 575                         *
 576                         * Because we may see inodes that are from the wrong AG
 577                         * due to RCU freeing and reallocation, only update the
 578                         * index if it lies in this AG. It was a race that lead
 579                         * us to see this inode, so another lookup from the
 580                         * same index will not find it again.
 581                         */
 582                        if (XFS_INO_TO_AGNO(mp, ip->i_ino) != pag->pag_agno)
 583                                continue;
 584                        first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1);
 585                        if (first_index < XFS_INO_TO_AGINO(mp, ip->i_ino))
 586                                done = 1;
 587                }
 588
 589                /* unlock now we've grabbed the inodes. */
 590                rcu_read_unlock();
 591
 592                for (i = 0; i < nr_found; i++) {
 593                        if (!batch[i])
 594                                continue;
 595                        error = execute(batch[i], pag, flags, args);
 596                        IRELE(batch[i]);
 597                        if (error == EAGAIN) {
 598                                skipped++;
 599                                continue;
 600                        }
 601                        if (error && last_error != EFSCORRUPTED)
 602                                last_error = error;
 603                }
 604
 605                /* bail out if the filesystem is corrupted.  */
 606                if (error == EFSCORRUPTED)
 607                        break;
 608
 609                cond_resched();
 610
 611        } while (nr_found && !done);
 612
 613        if (skipped) {
 614                delay(1);
 615                goto restart;
 616        }
 617        return last_error;
 618}
 619
 620/*
 621 * Background scanning to trim post-EOF preallocated space. This is queued
 622 * based on the 'background_prealloc_discard_period' tunable (5m by default).
 623 */
 624STATIC void
 625xfs_queue_eofblocks(
 626        struct xfs_mount *mp)
 627{
 628        rcu_read_lock();
 629        if (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_EOFBLOCKS_TAG))
 630                queue_delayed_work(mp->m_eofblocks_workqueue,
 631                                   &mp->m_eofblocks_work,
 632                                   msecs_to_jiffies(xfs_eofb_secs * 1000));
 633        rcu_read_unlock();
 634}
 635
 636void
 637xfs_eofblocks_worker(
 638        struct work_struct *work)
 639{
 640        struct xfs_mount *mp = container_of(to_delayed_work(work),
 641                                struct xfs_mount, m_eofblocks_work);
 642        xfs_icache_free_eofblocks(mp, NULL);
 643        xfs_queue_eofblocks(mp);
 644}
 645
 646int
 647xfs_inode_ag_iterator(
 648        struct xfs_mount        *mp,
 649        int                     (*execute)(struct xfs_inode *ip,
 650                                           struct xfs_perag *pag, int flags,
 651                                           void *args),
 652        int                     flags,
 653        void                    *args)
 654{
 655        struct xfs_perag        *pag;
 656        int                     error = 0;
 657        int                     last_error = 0;
 658        xfs_agnumber_t          ag;
 659
 660        ag = 0;
 661        while ((pag = xfs_perag_get(mp, ag))) {
 662                ag = pag->pag_agno + 1;
 663                error = xfs_inode_ag_walk(mp, pag, execute, flags, args, -1);
 664                xfs_perag_put(pag);
 665                if (error) {
 666                        last_error = error;
 667                        if (error == EFSCORRUPTED)
 668                                break;
 669                }
 670        }
 671        return XFS_ERROR(last_error);
 672}
 673
 674int
 675xfs_inode_ag_iterator_tag(
 676        struct xfs_mount        *mp,
 677        int                     (*execute)(struct xfs_inode *ip,
 678                                           struct xfs_perag *pag, int flags,
 679                                           void *args),
 680        int                     flags,
 681        void                    *args,
 682        int                     tag)
 683{
 684        struct xfs_perag        *pag;
 685        int                     error = 0;
 686        int                     last_error = 0;
 687        xfs_agnumber_t          ag;
 688
 689        ag = 0;
 690        while ((pag = xfs_perag_get_tag(mp, ag, tag))) {
 691                ag = pag->pag_agno + 1;
 692                error = xfs_inode_ag_walk(mp, pag, execute, flags, args, tag);
 693                xfs_perag_put(pag);
 694                if (error) {
 695                        last_error = error;
 696                        if (error == EFSCORRUPTED)
 697                                break;
 698                }
 699        }
 700        return XFS_ERROR(last_error);
 701}
 702
 703/*
 704 * Queue a new inode reclaim pass if there are reclaimable inodes and there
 705 * isn't a reclaim pass already in progress. By default it runs every 5s based
 706 * on the xfs periodic sync default of 30s. Perhaps this should have it's own
 707 * tunable, but that can be done if this method proves to be ineffective or too
 708 * aggressive.
 709 */
 710static void
 711xfs_reclaim_work_queue(
 712        struct xfs_mount        *mp)
 713{
 714
 715        rcu_read_lock();
 716        if (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_RECLAIM_TAG)) {
 717                queue_delayed_work(mp->m_reclaim_workqueue, &mp->m_reclaim_work,
 718                        msecs_to_jiffies(xfs_syncd_centisecs / 6 * 10));
 719        }
 720        rcu_read_unlock();
 721}
 722
 723/*
 724 * This is a fast pass over the inode cache to try to get reclaim moving on as
 725 * many inodes as possible in a short period of time. It kicks itself every few
 726 * seconds, as well as being kicked by the inode cache shrinker when memory
 727 * goes low. It scans as quickly as possible avoiding locked inodes or those
 728 * already being flushed, and once done schedules a future pass.
 729 */
 730void
 731xfs_reclaim_worker(
 732        struct work_struct *work)
 733{
 734        struct xfs_mount *mp = container_of(to_delayed_work(work),
 735                                        struct xfs_mount, m_reclaim_work);
 736
 737        xfs_reclaim_inodes(mp, SYNC_TRYLOCK);
 738        xfs_reclaim_work_queue(mp);
 739}
 740
 741static void
 742__xfs_inode_set_reclaim_tag(
 743        struct xfs_perag        *pag,
 744        struct xfs_inode        *ip)
 745{
 746        radix_tree_tag_set(&pag->pag_ici_root,
 747                           XFS_INO_TO_AGINO(ip->i_mount, ip->i_ino),
 748                           XFS_ICI_RECLAIM_TAG);
 749
 750        if (!pag->pag_ici_reclaimable) {
 751                /* propagate the reclaim tag up into the perag radix tree */
 752                spin_lock(&ip->i_mount->m_perag_lock);
 753                radix_tree_tag_set(&ip->i_mount->m_perag_tree,
 754                                XFS_INO_TO_AGNO(ip->i_mount, ip->i_ino),
 755                                XFS_ICI_RECLAIM_TAG);
 756                spin_unlock(&ip->i_mount->m_perag_lock);
 757
 758                /* schedule periodic background inode reclaim */
 759                xfs_reclaim_work_queue(ip->i_mount);
 760
 761                trace_xfs_perag_set_reclaim(ip->i_mount, pag->pag_agno,
 762                                                        -1, _RET_IP_);
 763        }
 764        pag->pag_ici_reclaimable++;
 765}
 766
 767/*
 768 * We set the inode flag atomically with the radix tree tag.
 769 * Once we get tag lookups on the radix tree, this inode flag
 770 * can go away.
 771 */
 772void
 773xfs_inode_set_reclaim_tag(
 774        xfs_inode_t     *ip)
 775{
 776        struct xfs_mount *mp = ip->i_mount;
 777        struct xfs_perag *pag;
 778
 779        pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
 780        spin_lock(&pag->pag_ici_lock);
 781        spin_lock(&ip->i_flags_lock);
 782        __xfs_inode_set_reclaim_tag(pag, ip);
 783        __xfs_iflags_set(ip, XFS_IRECLAIMABLE);
 784        spin_unlock(&ip->i_flags_lock);
 785        spin_unlock(&pag->pag_ici_lock);
 786        xfs_perag_put(pag);
 787}
 788
 789STATIC void
 790__xfs_inode_clear_reclaim(
 791        xfs_perag_t     *pag,
 792        xfs_inode_t     *ip)
 793{
 794        pag->pag_ici_reclaimable--;
 795        if (!pag->pag_ici_reclaimable) {
 796                /* clear the reclaim tag from the perag radix tree */
 797                spin_lock(&ip->i_mount->m_perag_lock);
 798                radix_tree_tag_clear(&ip->i_mount->m_perag_tree,
 799                                XFS_INO_TO_AGNO(ip->i_mount, ip->i_ino),
 800                                XFS_ICI_RECLAIM_TAG);
 801                spin_unlock(&ip->i_mount->m_perag_lock);
 802                trace_xfs_perag_clear_reclaim(ip->i_mount, pag->pag_agno,
 803                                                        -1, _RET_IP_);
 804        }
 805}
 806
 807STATIC void
 808__xfs_inode_clear_reclaim_tag(
 809        xfs_mount_t     *mp,
 810        xfs_perag_t     *pag,
 811        xfs_inode_t     *ip)
 812{
 813        radix_tree_tag_clear(&pag->pag_ici_root,
 814                        XFS_INO_TO_AGINO(mp, ip->i_ino), XFS_ICI_RECLAIM_TAG);
 815        __xfs_inode_clear_reclaim(pag, ip);
 816}
 817
 818/*
 819 * Grab the inode for reclaim exclusively.
 820 * Return 0 if we grabbed it, non-zero otherwise.
 821 */
 822STATIC int
 823xfs_reclaim_inode_grab(
 824        struct xfs_inode        *ip,
 825        int                     flags)
 826{
 827        ASSERT(rcu_read_lock_held());
 828
 829        /* quick check for stale RCU freed inode */
 830        if (!ip->i_ino)
 831                return 1;
 832
 833        /*
 834         * If we are asked for non-blocking operation, do unlocked checks to
 835         * see if the inode already is being flushed or in reclaim to avoid
 836         * lock traffic.
 837         */
 838        if ((flags & SYNC_TRYLOCK) &&
 839            __xfs_iflags_test(ip, XFS_IFLOCK | XFS_IRECLAIM))
 840                return 1;
 841
 842        /*
 843         * The radix tree lock here protects a thread in xfs_iget from racing
 844         * with us starting reclaim on the inode.  Once we have the
 845         * XFS_IRECLAIM flag set it will not touch us.
 846         *
 847         * Due to RCU lookup, we may find inodes that have been freed and only
 848         * have XFS_IRECLAIM set.  Indeed, we may see reallocated inodes that
 849         * aren't candidates for reclaim at all, so we must check the
 850         * XFS_IRECLAIMABLE is set first before proceeding to reclaim.
 851         */
 852        spin_lock(&ip->i_flags_lock);
 853        if (!__xfs_iflags_test(ip, XFS_IRECLAIMABLE) ||
 854            __xfs_iflags_test(ip, XFS_IRECLAIM)) {
 855                /* not a reclaim candidate. */
 856                spin_unlock(&ip->i_flags_lock);
 857                return 1;
 858        }
 859        __xfs_iflags_set(ip, XFS_IRECLAIM);
 860        spin_unlock(&ip->i_flags_lock);
 861        return 0;
 862}
 863
 864/*
 865 * Inodes in different states need to be treated differently. The following
 866 * table lists the inode states and the reclaim actions necessary:
 867 *
 868 *      inode state          iflush ret         required action
 869 *      ---------------      ----------         ---------------
 870 *      bad                     -               reclaim
 871 *      shutdown                EIO             unpin and reclaim
 872 *      clean, unpinned         0               reclaim
 873 *      stale, unpinned         0               reclaim
 874 *      clean, pinned(*)        0               requeue
 875 *      stale, pinned           EAGAIN          requeue
 876 *      dirty, async            -               requeue
 877 *      dirty, sync             0               reclaim
 878 *
 879 * (*) dgc: I don't think the clean, pinned state is possible but it gets
 880 * handled anyway given the order of checks implemented.
 881 *
 882 * Also, because we get the flush lock first, we know that any inode that has
 883 * been flushed delwri has had the flush completed by the time we check that
 884 * the inode is clean.
 885 *
 886 * Note that because the inode is flushed delayed write by AIL pushing, the
 887 * flush lock may already be held here and waiting on it can result in very
 888 * long latencies.  Hence for sync reclaims, where we wait on the flush lock,
 889 * the caller should push the AIL first before trying to reclaim inodes to
 890 * minimise the amount of time spent waiting.  For background relaim, we only
 891 * bother to reclaim clean inodes anyway.
 892 *
 893 * Hence the order of actions after gaining the locks should be:
 894 *      bad             => reclaim
 895 *      shutdown        => unpin and reclaim
 896 *      pinned, async   => requeue
 897 *      pinned, sync    => unpin
 898 *      stale           => reclaim
 899 *      clean           => reclaim
 900 *      dirty, async    => requeue
 901 *      dirty, sync     => flush, wait and reclaim
 902 */
 903STATIC int
 904xfs_reclaim_inode(
 905        struct xfs_inode        *ip,
 906        struct xfs_perag        *pag,
 907        int                     sync_mode)
 908{
 909        struct xfs_buf          *bp = NULL;
 910        int                     error;
 911
 912restart:
 913        error = 0;
 914        xfs_ilock(ip, XFS_ILOCK_EXCL);
 915        if (!xfs_iflock_nowait(ip)) {
 916                if (!(sync_mode & SYNC_WAIT))
 917                        goto out;
 918                xfs_iflock(ip);
 919        }
 920
 921        if (is_bad_inode(VFS_I(ip)))
 922                goto reclaim;
 923        if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
 924                xfs_iunpin_wait(ip);
 925                xfs_iflush_abort(ip, false);
 926                goto reclaim;
 927        }
 928        if (xfs_ipincount(ip)) {
 929                if (!(sync_mode & SYNC_WAIT))
 930                        goto out_ifunlock;
 931                xfs_iunpin_wait(ip);
 932        }
 933        if (xfs_iflags_test(ip, XFS_ISTALE))
 934                goto reclaim;
 935        if (xfs_inode_clean(ip))
 936                goto reclaim;
 937
 938        /*
 939         * Never flush out dirty data during non-blocking reclaim, as it would
 940         * just contend with AIL pushing trying to do the same job.
 941         */
 942        if (!(sync_mode & SYNC_WAIT))
 943                goto out_ifunlock;
 944
 945        /*
 946         * Now we have an inode that needs flushing.
 947         *
 948         * Note that xfs_iflush will never block on the inode buffer lock, as
 949         * xfs_ifree_cluster() can lock the inode buffer before it locks the
 950         * ip->i_lock, and we are doing the exact opposite here.  As a result,
 951         * doing a blocking xfs_imap_to_bp() to get the cluster buffer would
 952         * result in an ABBA deadlock with xfs_ifree_cluster().
 953         *
 954         * As xfs_ifree_cluser() must gather all inodes that are active in the
 955         * cache to mark them stale, if we hit this case we don't actually want
 956         * to do IO here - we want the inode marked stale so we can simply
 957         * reclaim it.  Hence if we get an EAGAIN error here,  just unlock the
 958         * inode, back off and try again.  Hopefully the next pass through will
 959         * see the stale flag set on the inode.
 960         */
 961        error = xfs_iflush(ip, &bp);
 962        if (error == EAGAIN) {
 963                xfs_iunlock(ip, XFS_ILOCK_EXCL);
 964                /* backoff longer than in xfs_ifree_cluster */
 965                delay(2);
 966                goto restart;
 967        }
 968
 969        if (!error) {
 970                error = xfs_bwrite(bp);
 971                xfs_buf_relse(bp);
 972        }
 973
 974        xfs_iflock(ip);
 975reclaim:
 976        xfs_ifunlock(ip);
 977        xfs_iunlock(ip, XFS_ILOCK_EXCL);
 978
 979        XFS_STATS_INC(xs_ig_reclaims);
 980        /*
 981         * Remove the inode from the per-AG radix tree.
 982         *
 983         * Because radix_tree_delete won't complain even if the item was never
 984         * added to the tree assert that it's been there before to catch
 985         * problems with the inode life time early on.
 986         */
 987        spin_lock(&pag->pag_ici_lock);
 988        if (!radix_tree_delete(&pag->pag_ici_root,
 989                                XFS_INO_TO_AGINO(ip->i_mount, ip->i_ino)))
 990                ASSERT(0);
 991        __xfs_inode_clear_reclaim(pag, ip);
 992        spin_unlock(&pag->pag_ici_lock);
 993
 994        /*
 995         * Here we do an (almost) spurious inode lock in order to coordinate
 996         * with inode cache radix tree lookups.  This is because the lookup
 997         * can reference the inodes in the cache without taking references.
 998         *
 999         * We make that OK here by ensuring that we wait until the inode is
1000         * unlocked after the lookup before we go ahead and free it.
1001         */
1002        xfs_ilock(ip, XFS_ILOCK_EXCL);
1003        xfs_qm_dqdetach(ip);
1004        xfs_iunlock(ip, XFS_ILOCK_EXCL);
1005
1006        xfs_inode_free(ip);
1007        return error;
1008
1009out_ifunlock:
1010        xfs_ifunlock(ip);
1011out:
1012        xfs_iflags_clear(ip, XFS_IRECLAIM);
1013        xfs_iunlock(ip, XFS_ILOCK_EXCL);
1014        /*
1015         * We could return EAGAIN here to make reclaim rescan the inode tree in
1016         * a short while. However, this just burns CPU time scanning the tree
1017         * waiting for IO to complete and the reclaim work never goes back to
1018         * the idle state. Instead, return 0 to let the next scheduled
1019         * background reclaim attempt to reclaim the inode again.
1020         */
1021        return 0;
1022}
1023
1024/*
1025 * Walk the AGs and reclaim the inodes in them. Even if the filesystem is
1026 * corrupted, we still want to try to reclaim all the inodes. If we don't,
1027 * then a shut down during filesystem unmount reclaim walk leak all the
1028 * unreclaimed inodes.
1029 */
1030STATIC int
1031xfs_reclaim_inodes_ag(
1032        struct xfs_mount        *mp,
1033        int                     flags,
1034        int                     *nr_to_scan)
1035{
1036        struct xfs_perag        *pag;
1037        int                     error = 0;
1038        int                     last_error = 0;
1039        xfs_agnumber_t          ag;
1040        int                     trylock = flags & SYNC_TRYLOCK;
1041        int                     skipped;
1042
1043restart:
1044        ag = 0;
1045        skipped = 0;
1046        while ((pag = xfs_perag_get_tag(mp, ag, XFS_ICI_RECLAIM_TAG))) {
1047                unsigned long   first_index = 0;
1048                int             done = 0;
1049                int             nr_found = 0;
1050
1051                ag = pag->pag_agno + 1;
1052
1053                if (trylock) {
1054                        if (!mutex_trylock(&pag->pag_ici_reclaim_lock)) {
1055                                skipped++;
1056                                xfs_perag_put(pag);
1057                                continue;
1058                        }
1059                        first_index = pag->pag_ici_reclaim_cursor;
1060                } else
1061                        mutex_lock(&pag->pag_ici_reclaim_lock);
1062
1063                do {
1064                        struct xfs_inode *batch[XFS_LOOKUP_BATCH];
1065                        int     i;
1066
1067                        rcu_read_lock();
1068                        nr_found = radix_tree_gang_lookup_tag(
1069                                        &pag->pag_ici_root,
1070                                        (void **)batch, first_index,
1071                                        XFS_LOOKUP_BATCH,
1072                                        XFS_ICI_RECLAIM_TAG);
1073                        if (!nr_found) {
1074                                done = 1;
1075                                rcu_read_unlock();
1076                                break;
1077                        }
1078
1079                        /*
1080                         * Grab the inodes before we drop the lock. if we found
1081                         * nothing, nr == 0 and the loop will be skipped.
1082                         */
1083                        for (i = 0; i < nr_found; i++) {
1084                                struct xfs_inode *ip = batch[i];
1085
1086                                if (done || xfs_reclaim_inode_grab(ip, flags))
1087                                        batch[i] = NULL;
1088
1089                                /*
1090                                 * Update the index for the next lookup. Catch
1091                                 * overflows into the next AG range which can
1092                                 * occur if we have inodes in the last block of
1093                                 * the AG and we are currently pointing to the
1094                                 * last inode.
1095                                 *
1096                                 * Because we may see inodes that are from the
1097                                 * wrong AG due to RCU freeing and
1098                                 * reallocation, only update the index if it
1099                                 * lies in this AG. It was a race that lead us
1100                                 * to see this inode, so another lookup from
1101                                 * the same index will not find it again.
1102                                 */
1103                                if (XFS_INO_TO_AGNO(mp, ip->i_ino) !=
1104                                                                pag->pag_agno)
1105                                        continue;
1106                                first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1);
1107                                if (first_index < XFS_INO_TO_AGINO(mp, ip->i_ino))
1108                                        done = 1;
1109                        }
1110
1111                        /* unlock now we've grabbed the inodes. */
1112                        rcu_read_unlock();
1113
1114                        for (i = 0; i < nr_found; i++) {
1115                                if (!batch[i])
1116                                        continue;
1117                                error = xfs_reclaim_inode(batch[i], pag, flags);
1118                                if (error && last_error != EFSCORRUPTED)
1119                                        last_error = error;
1120                        }
1121
1122                        *nr_to_scan -= XFS_LOOKUP_BATCH;
1123
1124                        cond_resched();
1125
1126                } while (nr_found && !done && *nr_to_scan > 0);
1127
1128                if (trylock && !done)
1129                        pag->pag_ici_reclaim_cursor = first_index;
1130                else
1131                        pag->pag_ici_reclaim_cursor = 0;
1132                mutex_unlock(&pag->pag_ici_reclaim_lock);
1133                xfs_perag_put(pag);
1134        }
1135
1136        /*
1137         * if we skipped any AG, and we still have scan count remaining, do
1138         * another pass this time using blocking reclaim semantics (i.e
1139         * waiting on the reclaim locks and ignoring the reclaim cursors). This
1140         * ensure that when we get more reclaimers than AGs we block rather
1141         * than spin trying to execute reclaim.
1142         */
1143        if (skipped && (flags & SYNC_WAIT) && *nr_to_scan > 0) {
1144                trylock = 0;
1145                goto restart;
1146        }
1147        return XFS_ERROR(last_error);
1148}
1149
1150int
1151xfs_reclaim_inodes(
1152        xfs_mount_t     *mp,
1153        int             mode)
1154{
1155        int             nr_to_scan = INT_MAX;
1156
1157        return xfs_reclaim_inodes_ag(mp, mode, &nr_to_scan);
1158}
1159
1160/*
1161 * Scan a certain number of inodes for reclaim.
1162 *
1163 * When called we make sure that there is a background (fast) inode reclaim in
1164 * progress, while we will throttle the speed of reclaim via doing synchronous
1165 * reclaim of inodes. That means if we come across dirty inodes, we wait for
1166 * them to be cleaned, which we hope will not be very long due to the
1167 * background walker having already kicked the IO off on those dirty inodes.
1168 */
1169void
1170xfs_reclaim_inodes_nr(
1171        struct xfs_mount        *mp,
1172        int                     nr_to_scan)
1173{
1174        /* kick background reclaimer and push the AIL */
1175        xfs_reclaim_work_queue(mp);
1176        xfs_ail_push_all(mp->m_ail);
1177
1178        xfs_reclaim_inodes_ag(mp, SYNC_TRYLOCK | SYNC_WAIT, &nr_to_scan);
1179}
1180
1181/*
1182 * Return the number of reclaimable inodes in the filesystem for
1183 * the shrinker to determine how much to reclaim.
1184 */
1185int
1186xfs_reclaim_inodes_count(
1187        struct xfs_mount        *mp)
1188{
1189        struct xfs_perag        *pag;
1190        xfs_agnumber_t          ag = 0;
1191        int                     reclaimable = 0;
1192
1193        while ((pag = xfs_perag_get_tag(mp, ag, XFS_ICI_RECLAIM_TAG))) {
1194                ag = pag->pag_agno + 1;
1195                reclaimable += pag->pag_ici_reclaimable;
1196                xfs_perag_put(pag);
1197        }
1198        return reclaimable;
1199}
1200
1201STATIC int
1202xfs_inode_match_id(
1203        struct xfs_inode        *ip,
1204        struct xfs_eofblocks    *eofb)
1205{
1206        if (eofb->eof_flags & XFS_EOF_FLAGS_UID &&
1207            ip->i_d.di_uid != eofb->eof_uid)
1208                return 0;
1209
1210        if (eofb->eof_flags & XFS_EOF_FLAGS_GID &&
1211            ip->i_d.di_gid != eofb->eof_gid)
1212                return 0;
1213
1214        if (eofb->eof_flags & XFS_EOF_FLAGS_PRID &&
1215            xfs_get_projid(ip) != eofb->eof_prid)
1216                return 0;
1217
1218        return 1;
1219}
1220
1221STATIC int
1222xfs_inode_free_eofblocks(
1223        struct xfs_inode        *ip,
1224        struct xfs_perag        *pag,
1225        int                     flags,
1226        void                    *args)
1227{
1228        int ret;
1229        struct xfs_eofblocks *eofb = args;
1230
1231        if (!xfs_can_free_eofblocks(ip, false)) {
1232                /* inode could be preallocated or append-only */
1233                trace_xfs_inode_free_eofblocks_invalid(ip);
1234                xfs_inode_clear_eofblocks_tag(ip);
1235                return 0;
1236        }
1237
1238        /*
1239         * If the mapping is dirty the operation can block and wait for some
1240         * time. Unless we are waiting, skip it.
1241         */
1242        if (!(flags & SYNC_WAIT) &&
1243            mapping_tagged(VFS_I(ip)->i_mapping, PAGECACHE_TAG_DIRTY))
1244                return 0;
1245
1246        if (eofb) {
1247                if (!xfs_inode_match_id(ip, eofb))
1248                        return 0;
1249
1250                /* skip the inode if the file size is too small */
1251                if (eofb->eof_flags & XFS_EOF_FLAGS_MINFILESIZE &&
1252                    XFS_ISIZE(ip) < eofb->eof_min_file_size)
1253                        return 0;
1254        }
1255
1256        ret = xfs_free_eofblocks(ip->i_mount, ip, true);
1257
1258        /* don't revisit the inode if we're not waiting */
1259        if (ret == EAGAIN && !(flags & SYNC_WAIT))
1260                ret = 0;
1261
1262        return ret;
1263}
1264
1265int
1266xfs_icache_free_eofblocks(
1267        struct xfs_mount        *mp,
1268        struct xfs_eofblocks    *eofb)
1269{
1270        int flags = SYNC_TRYLOCK;
1271
1272        if (eofb && (eofb->eof_flags & XFS_EOF_FLAGS_SYNC))
1273                flags = SYNC_WAIT;
1274
1275        return xfs_inode_ag_iterator_tag(mp, xfs_inode_free_eofblocks, flags,
1276                                         eofb, XFS_ICI_EOFBLOCKS_TAG);
1277}
1278
1279void
1280xfs_inode_set_eofblocks_tag(
1281        xfs_inode_t     *ip)
1282{
1283        struct xfs_mount *mp = ip->i_mount;
1284        struct xfs_perag *pag;
1285        int tagged;
1286
1287        pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
1288        spin_lock(&pag->pag_ici_lock);
1289        trace_xfs_inode_set_eofblocks_tag(ip);
1290
1291        tagged = radix_tree_tagged(&pag->pag_ici_root,
1292                                   XFS_ICI_EOFBLOCKS_TAG);
1293        radix_tree_tag_set(&pag->pag_ici_root,
1294                           XFS_INO_TO_AGINO(ip->i_mount, ip->i_ino),
1295                           XFS_ICI_EOFBLOCKS_TAG);
1296        if (!tagged) {
1297                /* propagate the eofblocks tag up into the perag radix tree */
1298                spin_lock(&ip->i_mount->m_perag_lock);
1299                radix_tree_tag_set(&ip->i_mount->m_perag_tree,
1300                                   XFS_INO_TO_AGNO(ip->i_mount, ip->i_ino),
1301                                   XFS_ICI_EOFBLOCKS_TAG);
1302                spin_unlock(&ip->i_mount->m_perag_lock);
1303
1304                /* kick off background trimming */
1305                xfs_queue_eofblocks(ip->i_mount);
1306
1307                trace_xfs_perag_set_eofblocks(ip->i_mount, pag->pag_agno,
1308                                              -1, _RET_IP_);
1309        }
1310
1311        spin_unlock(&pag->pag_ici_lock);
1312        xfs_perag_put(pag);
1313}
1314
1315void
1316xfs_inode_clear_eofblocks_tag(
1317        xfs_inode_t     *ip)
1318{
1319        struct xfs_mount *mp = ip->i_mount;
1320        struct xfs_perag *pag;
1321
1322        pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
1323        spin_lock(&pag->pag_ici_lock);
1324        trace_xfs_inode_clear_eofblocks_tag(ip);
1325
1326        radix_tree_tag_clear(&pag->pag_ici_root,
1327                             XFS_INO_TO_AGINO(ip->i_mount, ip->i_ino),
1328                             XFS_ICI_EOFBLOCKS_TAG);
1329        if (!radix_tree_tagged(&pag->pag_ici_root, XFS_ICI_EOFBLOCKS_TAG)) {
1330                /* clear the eofblocks tag from the perag radix tree */
1331                spin_lock(&ip->i_mount->m_perag_lock);
1332                radix_tree_tag_clear(&ip->i_mount->m_perag_tree,
1333                                     XFS_INO_TO_AGNO(ip->i_mount, ip->i_ino),
1334                                     XFS_ICI_EOFBLOCKS_TAG);
1335                spin_unlock(&ip->i_mount->m_perag_lock);
1336                trace_xfs_perag_clear_eofblocks(ip->i_mount, pag->pag_agno,
1337                                               -1, _RET_IP_);
1338        }
1339
1340        spin_unlock(&pag->pag_ici_lock);
1341        xfs_perag_put(pag);
1342}
1343
1344
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.