linux/fs/fs-writeback.c
<<
>>
Prefs
   1/*
   2 * fs/fs-writeback.c
   3 *
   4 * Copyright (C) 2002, Linus Torvalds.
   5 *
   6 * Contains all the functions related to writing back and waiting
   7 * upon dirty inodes against superblocks, and writing back dirty
   8 * pages against inodes.  ie: data writeback.  Writeout of the
   9 * inode itself is not handled here.
  10 *
  11 * 10Apr2002    Andrew Morton
  12 *              Split out of fs/inode.c
  13 *              Additions for address_space-based writeback
  14 */
  15
  16#include <linux/kernel.h>
  17#include <linux/module.h>
  18#include <linux/spinlock.h>
  19#include <linux/slab.h>
  20#include <linux/sched.h>
  21#include <linux/fs.h>
  22#include <linux/mm.h>
  23#include <linux/pagemap.h>
  24#include <linux/kthread.h>
  25#include <linux/freezer.h>
  26#include <linux/writeback.h>
  27#include <linux/blkdev.h>
  28#include <linux/backing-dev.h>
  29#include <linux/tracepoint.h>
  30#include "internal.h"
  31
  32/*
  33 * 4MB minimal write chunk size
  34 */
  35#define MIN_WRITEBACK_PAGES     (4096UL >> (PAGE_CACHE_SHIFT - 10))
  36
  37/*
  38 * Passed into wb_writeback(), essentially a subset of writeback_control
  39 */
  40struct wb_writeback_work {
  41        long nr_pages;
  42        struct super_block *sb;
  43        unsigned long *older_than_this;
  44        enum writeback_sync_modes sync_mode;
  45        unsigned int tagged_writepages:1;
  46        unsigned int for_kupdate:1;
  47        unsigned int range_cyclic:1;
  48        unsigned int for_background:1;
  49        enum wb_reason reason;          /* why was writeback initiated? */
  50
  51        struct list_head list;          /* pending work list */
  52        struct completion *done;        /* set if the caller waits */
  53};
  54
  55/*
  56 * We don't actually have pdflush, but this one is exported though /proc...
  57 */
  58int nr_pdflush_threads;
  59
  60/**
  61 * writeback_in_progress - determine whether there is writeback in progress
  62 * @bdi: the device's backing_dev_info structure.
  63 *
  64 * Determine whether there is writeback waiting to be handled against a
  65 * backing device.
  66 */
  67int writeback_in_progress(struct backing_dev_info *bdi)
  68{
  69        return test_bit(BDI_writeback_running, &bdi->state);
  70}
  71
  72static inline struct backing_dev_info *inode_to_bdi(struct inode *inode)
  73{
  74        struct super_block *sb = inode->i_sb;
  75
  76        if (strcmp(sb->s_type->name, "bdev") == 0)
  77                return inode->i_mapping->backing_dev_info;
  78
  79        return sb->s_bdi;
  80}
  81
  82static inline struct inode *wb_inode(struct list_head *head)
  83{
  84        return list_entry(head, struct inode, i_wb_list);
  85}
  86
  87/*
  88 * Include the creation of the trace points after defining the
  89 * wb_writeback_work structure and inline functions so that the definition
  90 * remains local to this file.
  91 */
  92#define CREATE_TRACE_POINTS
  93#include <trace/events/writeback.h>
  94
  95/* Wakeup flusher thread or forker thread to fork it. Requires bdi->wb_lock. */
  96static void bdi_wakeup_flusher(struct backing_dev_info *bdi)
  97{
  98        if (bdi->wb.task) {
  99                wake_up_process(bdi->wb.task);
 100        } else {
 101                /*
 102                 * The bdi thread isn't there, wake up the forker thread which
 103                 * will create and run it.
 104                 */
 105                wake_up_process(default_backing_dev_info.wb.task);
 106        }
 107}
 108
 109static void bdi_queue_work(struct backing_dev_info *bdi,
 110                           struct wb_writeback_work *work)
 111{
 112        trace_writeback_queue(bdi, work);
 113
 114        spin_lock_bh(&bdi->wb_lock);
 115        list_add_tail(&work->list, &bdi->work_list);
 116        if (!bdi->wb.task)
 117                trace_writeback_nothread(bdi, work);
 118        bdi_wakeup_flusher(bdi);
 119        spin_unlock_bh(&bdi->wb_lock);
 120}
 121
 122static void
 123__bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages,
 124                      bool range_cyclic, enum wb_reason reason)
 125{
 126        struct wb_writeback_work *work;
 127
 128        /*
 129         * This is WB_SYNC_NONE writeback, so if allocation fails just
 130         * wakeup the thread for old dirty data writeback
 131         */
 132        work = kzalloc(sizeof(*work), GFP_ATOMIC);
 133        if (!work) {
 134                if (bdi->wb.task) {
 135                        trace_writeback_nowork(bdi);
 136                        wake_up_process(bdi->wb.task);
 137                }
 138                return;
 139        }
 140
 141        work->sync_mode = WB_SYNC_NONE;
 142        work->nr_pages  = nr_pages;
 143        work->range_cyclic = range_cyclic;
 144        work->reason    = reason;
 145
 146        bdi_queue_work(bdi, work);
 147}
 148
 149/**
 150 * bdi_start_writeback - start writeback
 151 * @bdi: the backing device to write from
 152 * @nr_pages: the number of pages to write
 153 * @reason: reason why some writeback work was initiated
 154 *
 155 * Description:
 156 *   This does WB_SYNC_NONE opportunistic writeback. The IO is only
 157 *   started when this function returns, we make no guarantees on
 158 *   completion. Caller need not hold sb s_umount semaphore.
 159 *
 160 */
 161void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages,
 162                        enum wb_reason reason)
 163{
 164        __bdi_start_writeback(bdi, nr_pages, true, reason);
 165}
 166
 167/**
 168 * bdi_start_background_writeback - start background writeback
 169 * @bdi: the backing device to write from
 170 *
 171 * Description:
 172 *   This makes sure WB_SYNC_NONE background writeback happens. When
 173 *   this function returns, it is only guaranteed that for given BDI
 174 *   some IO is happening if we are over background dirty threshold.
 175 *   Caller need not hold sb s_umount semaphore.
 176 */
 177void bdi_start_background_writeback(struct backing_dev_info *bdi)
 178{
 179        /*
 180         * We just wake up the flusher thread. It will perform background
 181         * writeback as soon as there is no other work to do.
 182         */
 183        trace_writeback_wake_background(bdi);
 184        spin_lock_bh(&bdi->wb_lock);
 185        bdi_wakeup_flusher(bdi);
 186        spin_unlock_bh(&bdi->wb_lock);
 187}
 188
 189/*
 190 * Remove the inode from the writeback list it is on.
 191 */
 192void inode_wb_list_del(struct inode *inode)
 193{
 194        struct backing_dev_info *bdi = inode_to_bdi(inode);
 195
 196        spin_lock(&bdi->wb.list_lock);
 197        list_del_init(&inode->i_wb_list);
 198        spin_unlock(&bdi->wb.list_lock);
 199}
 200
 201/*
 202 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
 203 * furthest end of its superblock's dirty-inode list.
 204 *
 205 * Before stamping the inode's ->dirtied_when, we check to see whether it is
 206 * already the most-recently-dirtied inode on the b_dirty list.  If that is
 207 * the case then the inode must have been redirtied while it was being written
 208 * out and we don't reset its dirtied_when.
 209 */
 210static void redirty_tail(struct inode *inode, struct bdi_writeback *wb)
 211{
 212        assert_spin_locked(&wb->list_lock);
 213        if (!list_empty(&wb->b_dirty)) {
 214                struct inode *tail;
 215
 216                tail = wb_inode(wb->b_dirty.next);
 217                if (time_before(inode->dirtied_when, tail->dirtied_when))
 218                        inode->dirtied_when = jiffies;
 219        }
 220        list_move(&inode->i_wb_list, &wb->b_dirty);
 221}
 222
 223/*
 224 * requeue inode for re-scanning after bdi->b_io list is exhausted.
 225 */
 226static void requeue_io(struct inode *inode, struct bdi_writeback *wb)
 227{
 228        assert_spin_locked(&wb->list_lock);
 229        list_move(&inode->i_wb_list, &wb->b_more_io);
 230}
 231
 232static void inode_sync_complete(struct inode *inode)
 233{
 234        /*
 235         * Prevent speculative execution through
 236         * spin_unlock(&wb->list_lock);
 237         */
 238
 239        smp_mb();
 240        wake_up_bit(&inode->i_state, __I_SYNC);
 241}
 242
 243static bool inode_dirtied_after(struct inode *inode, unsigned long t)
 244{
 245        bool ret = time_after(inode->dirtied_when, t);
 246#ifndef CONFIG_64BIT
 247        /*
 248         * For inodes being constantly redirtied, dirtied_when can get stuck.
 249         * It _appears_ to be in the future, but is actually in distant past.
 250         * This test is necessary to prevent such wrapped-around relative times
 251         * from permanently stopping the whole bdi writeback.
 252         */
 253        ret = ret && time_before_eq(inode->dirtied_when, jiffies);
 254#endif
 255        return ret;
 256}
 257
 258/*
 259 * Move expired dirty inodes from @delaying_queue to @dispatch_queue.
 260 */
 261static int move_expired_inodes(struct list_head *delaying_queue,
 262                               struct list_head *dispatch_queue,
 263                               struct wb_writeback_work *work)
 264{
 265        LIST_HEAD(tmp);
 266        struct list_head *pos, *node;
 267        struct super_block *sb = NULL;
 268        struct inode *inode;
 269        int do_sb_sort = 0;
 270        int moved = 0;
 271
 272        while (!list_empty(delaying_queue)) {
 273                inode = wb_inode(delaying_queue->prev);
 274                if (work->older_than_this &&
 275                    inode_dirtied_after(inode, *work->older_than_this))
 276                        break;
 277                if (sb && sb != inode->i_sb)
 278                        do_sb_sort = 1;
 279                sb = inode->i_sb;
 280                list_move(&inode->i_wb_list, &tmp);
 281                moved++;
 282        }
 283
 284        /* just one sb in list, splice to dispatch_queue and we're done */
 285        if (!do_sb_sort) {
 286                list_splice(&tmp, dispatch_queue);
 287                goto out;
 288        }
 289
 290        /* Move inodes from one superblock together */
 291        while (!list_empty(&tmp)) {
 292                sb = wb_inode(tmp.prev)->i_sb;
 293                list_for_each_prev_safe(pos, node, &tmp) {
 294                        inode = wb_inode(pos);
 295                        if (inode->i_sb == sb)
 296                                list_move(&inode->i_wb_list, dispatch_queue);
 297                }
 298        }
 299out:
 300        return moved;
 301}
 302
 303/*
 304 * Queue all expired dirty inodes for io, eldest first.
 305 * Before
 306 *         newly dirtied     b_dirty    b_io    b_more_io
 307 *         =============>    gf         edc     BA
 308 * After
 309 *         newly dirtied     b_dirty    b_io    b_more_io
 310 *         =============>    g          fBAedc
 311 *                                           |
 312 *                                           +--> dequeue for IO
 313 */
 314static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work)
 315{
 316        int moved;
 317        assert_spin_locked(&wb->list_lock);
 318        list_splice_init(&wb->b_more_io, &wb->b_io);
 319        moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, work);
 320        trace_writeback_queue_io(wb, work, moved);
 321}
 322
 323static int write_inode(struct inode *inode, struct writeback_control *wbc)
 324{
 325        if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode))
 326                return inode->i_sb->s_op->write_inode(inode, wbc);
 327        return 0;
 328}
 329
 330/*
 331 * Wait for writeback on an inode to complete.
 332 */
 333static void inode_wait_for_writeback(struct inode *inode,
 334                                     struct bdi_writeback *wb)
 335{
 336        DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC);
 337        wait_queue_head_t *wqh;
 338
 339        wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
 340        while (inode->i_state & I_SYNC) {
 341                spin_unlock(&inode->i_lock);
 342                spin_unlock(&wb->list_lock);
 343                __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE);
 344                spin_lock(&wb->list_lock);
 345                spin_lock(&inode->i_lock);
 346        }
 347}
 348
 349/*
 350 * Write out an inode's dirty pages.  Called under wb->list_lock and
 351 * inode->i_lock.  Either the caller has an active reference on the inode or
 352 * the inode has I_WILL_FREE set.
 353 *
 354 * If `wait' is set, wait on the writeout.
 355 *
 356 * The whole writeout design is quite complex and fragile.  We want to avoid
 357 * starvation of particular inodes when others are being redirtied, prevent
 358 * livelocks, etc.
 359 */
 360static int
 361writeback_single_inode(struct inode *inode, struct bdi_writeback *wb,
 362                       struct writeback_control *wbc)
 363{
 364        struct address_space *mapping = inode->i_mapping;
 365        long nr_to_write = wbc->nr_to_write;
 366        unsigned dirty;
 367        int ret;
 368
 369        assert_spin_locked(&wb->list_lock);
 370        assert_spin_locked(&inode->i_lock);
 371
 372        if (!atomic_read(&inode->i_count))
 373                WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING)));
 374        else
 375                WARN_ON(inode->i_state & I_WILL_FREE);
 376
 377        if (inode->i_state & I_SYNC) {
 378                /*
 379                 * If this inode is locked for writeback and we are not doing
 380                 * writeback-for-data-integrity, move it to b_more_io so that
 381                 * writeback can proceed with the other inodes on s_io.
 382                 *
 383                 * We'll have another go at writing back this inode when we
 384                 * completed a full scan of b_io.
 385                 */
 386                if (wbc->sync_mode != WB_SYNC_ALL) {
 387                        requeue_io(inode, wb);
 388                        trace_writeback_single_inode_requeue(inode, wbc,
 389                                                             nr_to_write);
 390                        return 0;
 391                }
 392
 393                /*
 394                 * It's a data-integrity sync.  We must wait.
 395                 */
 396                inode_wait_for_writeback(inode, wb);
 397        }
 398
 399        BUG_ON(inode->i_state & I_SYNC);
 400
 401        /* Set I_SYNC, reset I_DIRTY_PAGES */
 402        inode->i_state |= I_SYNC;
 403        inode->i_state &= ~I_DIRTY_PAGES;
 404        spin_unlock(&inode->i_lock);
 405        spin_unlock(&wb->list_lock);
 406
 407        ret = do_writepages(mapping, wbc);
 408
 409        /*
 410         * Make sure to wait on the data before writing out the metadata.
 411         * This is important for filesystems that modify metadata on data
 412         * I/O completion.
 413         */
 414        if (wbc->sync_mode == WB_SYNC_ALL) {
 415                int err = filemap_fdatawait(mapping);
 416                if (ret == 0)
 417                        ret = err;
 418        }
 419
 420        /*
 421         * Some filesystems may redirty the inode during the writeback
 422         * due to delalloc, clear dirty metadata flags right before
 423         * write_inode()
 424         */
 425        spin_lock(&inode->i_lock);
 426        dirty = inode->i_state & I_DIRTY;
 427        inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC);
 428        spin_unlock(&inode->i_lock);
 429        /* Don't write the inode if only I_DIRTY_PAGES was set */
 430        if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
 431                int err = write_inode(inode, wbc);
 432                if (ret == 0)
 433                        ret = err;
 434        }
 435
 436        spin_lock(&wb->list_lock);
 437        spin_lock(&inode->i_lock);
 438        inode->i_state &= ~I_SYNC;
 439        if (!(inode->i_state & I_FREEING)) {
 440                /*
 441                 * Sync livelock prevention. Each inode is tagged and synced in
 442                 * one shot. If still dirty, it will be redirty_tail()'ed below.
 443                 * Update the dirty time to prevent enqueue and sync it again.
 444                 */
 445                if ((inode->i_state & I_DIRTY) &&
 446                    (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages))
 447                        inode->dirtied_when = jiffies;
 448
 449                if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
 450                        /*
 451                         * We didn't write back all the pages.  nfs_writepages()
 452                         * sometimes bales out without doing anything.
 453                         */
 454                        inode->i_state |= I_DIRTY_PAGES;
 455                        if (wbc->nr_to_write <= 0) {
 456                                /*
 457                                 * slice used up: queue for next turn
 458                                 */
 459                                requeue_io(inode, wb);
 460                        } else {
 461                                /*
 462                                 * Writeback blocked by something other than
 463                                 * congestion. Delay the inode for some time to
 464                                 * avoid spinning on the CPU (100% iowait)
 465                                 * retrying writeback of the dirty page/inode
 466                                 * that cannot be performed immediately.
 467                                 */
 468                                redirty_tail(inode, wb);
 469                        }
 470                } else if (inode->i_state & I_DIRTY) {
 471                        /*
 472                         * Filesystems can dirty the inode during writeback
 473                         * operations, such as delayed allocation during
 474                         * submission or metadata updates after data IO
 475                         * completion.
 476                         */
 477                        redirty_tail(inode, wb);
 478                } else {
 479                        /*
 480                         * The inode is clean.  At this point we either have
 481                         * a reference to the inode or it's on it's way out.
 482                         * No need to add it back to the LRU.
 483                         */
 484                        list_del_init(&inode->i_wb_list);
 485                }
 486        }
 487        inode_sync_complete(inode);
 488        trace_writeback_single_inode(inode, wbc, nr_to_write);
 489        return ret;
 490}
 491
 492static long writeback_chunk_size(struct backing_dev_info *bdi,
 493                                 struct wb_writeback_work *work)
 494{
 495        long pages;
 496
 497        /*
 498         * WB_SYNC_ALL mode does livelock avoidance by syncing dirty
 499         * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX
 500         * here avoids calling into writeback_inodes_wb() more than once.
 501         *
 502         * The intended call sequence for WB_SYNC_ALL writeback is:
 503         *
 504         *      wb_writeback()
 505         *          writeback_sb_inodes()       <== called only once
 506         *              write_cache_pages()     <== called once for each inode
 507         *                   (quickly) tag currently dirty pages
 508         *                   (maybe slowly) sync all tagged pages
 509         */
 510        if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages)
 511                pages = LONG_MAX;
 512        else {
 513                pages = min(bdi->avg_write_bandwidth / 2,
 514                            global_dirty_limit / DIRTY_SCOPE);
 515                pages = min(pages, work->nr_pages);
 516                pages = round_down(pages + MIN_WRITEBACK_PAGES,
 517                                   MIN_WRITEBACK_PAGES);
 518        }
 519
 520        return pages;
 521}
 522
 523/*
 524 * Write a portion of b_io inodes which belong to @sb.
 525 *
 526 * If @only_this_sb is true, then find and write all such
 527 * inodes. Otherwise write only ones which go sequentially
 528 * in reverse order.
 529 *
 530 * Return the number of pages and/or inodes written.
 531 */
 532static long writeback_sb_inodes(struct super_block *sb,
 533                                struct bdi_writeback *wb,
 534                                struct wb_writeback_work *work)
 535{
 536        struct writeback_control wbc = {
 537                .sync_mode              = work->sync_mode,
 538                .tagged_writepages      = work->tagged_writepages,
 539                .for_kupdate            = work->for_kupdate,
 540                .for_background         = work->for_background,
 541                .range_cyclic           = work->range_cyclic,
 542                .range_start            = 0,
 543                .range_end              = LLONG_MAX,
 544        };
 545        unsigned long start_time = jiffies;
 546        long write_chunk;
 547        long wrote = 0;  /* count both pages and inodes */
 548
 549        while (!list_empty(&wb->b_io)) {
 550                struct inode *inode = wb_inode(wb->b_io.prev);
 551
 552                if (inode->i_sb != sb) {
 553                        if (work->sb) {
 554                                /*
 555                                 * We only want to write back data for this
 556                                 * superblock, move all inodes not belonging
 557                                 * to it back onto the dirty list.
 558                                 */
 559                                redirty_tail(inode, wb);
 560                                continue;
 561                        }
 562
 563                        /*
 564                         * The inode belongs to a different superblock.
 565                         * Bounce back to the caller to unpin this and
 566                         * pin the next superblock.
 567                         */
 568                        break;
 569                }
 570
 571                /*
 572                 * Don't bother with new inodes or inodes beeing freed, first
 573                 * kind does not need peridic writeout yet, and for the latter
 574                 * kind writeout is handled by the freer.
 575                 */
 576                spin_lock(&inode->i_lock);
 577                if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
 578                        spin_unlock(&inode->i_lock);
 579                        redirty_tail(inode, wb);
 580                        continue;
 581                }
 582                __iget(inode);
 583                write_chunk = writeback_chunk_size(wb->bdi, work);
 584                wbc.nr_to_write = write_chunk;
 585                wbc.pages_skipped = 0;
 586
 587                writeback_single_inode(inode, wb, &wbc);
 588
 589                work->nr_pages -= write_chunk - wbc.nr_to_write;
 590                wrote += write_chunk - wbc.nr_to_write;
 591                if (!(inode->i_state & I_DIRTY))
 592                        wrote++;
 593                if (wbc.pages_skipped) {
 594                        /*
 595                         * writeback is not making progress due to locked
 596                         * buffers.  Skip this inode for now.
 597                         */
 598                        redirty_tail(inode, wb);
 599                }
 600                spin_unlock(&inode->i_lock);
 601                spin_unlock(&wb->list_lock);
 602                iput(inode);
 603                cond_resched();
 604                spin_lock(&wb->list_lock);
 605                /*
 606                 * bail out to wb_writeback() often enough to check
 607                 * background threshold and other termination conditions.
 608                 */
 609                if (wrote) {
 610                        if (time_is_before_jiffies(start_time + HZ / 10UL))
 611                                break;
 612                        if (work->nr_pages <= 0)
 613                                break;
 614                }
 615        }
 616        return wrote;
 617}
 618
 619static long __writeback_inodes_wb(struct bdi_writeback *wb,
 620                                  struct wb_writeback_work *work)
 621{
 622        unsigned long start_time = jiffies;
 623        long wrote = 0;
 624
 625        while (!list_empty(&wb->b_io)) {
 626                struct inode *inode = wb_inode(wb->b_io.prev);
 627                struct super_block *sb = inode->i_sb;
 628
 629                if (!grab_super_passive(sb)) {
 630                        /*
 631                         * grab_super_passive() may fail consistently due to
 632                         * s_umount being grabbed by someone else. Don't use
 633                         * requeue_io() to avoid busy retrying the inode/sb.
 634                         */
 635                        redirty_tail(inode, wb);
 636                        continue;
 637                }
 638                wrote += writeback_sb_inodes(sb, wb, work);
 639                drop_super(sb);
 640
 641                /* refer to the same tests at the end of writeback_sb_inodes */
 642                if (wrote) {
 643                        if (time_is_before_jiffies(start_time + HZ / 10UL))
 644                                break;
 645                        if (work->nr_pages <= 0)
 646                                break;
 647                }
 648        }
 649        /* Leave any unwritten inodes on b_io */
 650        return wrote;
 651}
 652
 653long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages,
 654                                enum wb_reason reason)
 655{
 656        struct wb_writeback_work work = {
 657                .nr_pages       = nr_pages,
 658                .sync_mode      = WB_SYNC_NONE,
 659                .range_cyclic   = 1,
 660                .reason         = reason,
 661        };
 662
 663        spin_lock(&wb->list_lock);
 664        if (list_empty(&wb->b_io))
 665                queue_io(wb, &work);
 666        __writeback_inodes_wb(wb, &work);
 667        spin_unlock(&wb->list_lock);
 668
 669        return nr_pages - work.nr_pages;
 670}
 671
 672static bool over_bground_thresh(struct backing_dev_info *bdi)
 673{
 674        unsigned long background_thresh, dirty_thresh;
 675
 676        global_dirty_limits(&background_thresh, &dirty_thresh);
 677
 678        if (global_page_state(NR_FILE_DIRTY) +
 679            global_page_state(NR_UNSTABLE_NFS) > background_thresh)
 680                return true;
 681
 682        if (bdi_stat(bdi, BDI_RECLAIMABLE) >
 683                                bdi_dirty_limit(bdi, background_thresh))
 684                return true;
 685
 686        return false;
 687}
 688
 689/*
 690 * Called under wb->list_lock. If there are multiple wb per bdi,
 691 * only the flusher working on the first wb should do it.
 692 */
 693static void wb_update_bandwidth(struct bdi_writeback *wb,
 694                                unsigned long start_time)
 695{
 696        __bdi_update_bandwidth(wb->bdi, 0, 0, 0, 0, 0, start_time);
 697}
 698
 699/*
 700 * Explicit flushing or periodic writeback of "old" data.
 701 *
 702 * Define "old": the first time one of an inode's pages is dirtied, we mark the
 703 * dirtying-time in the inode's address_space.  So this periodic writeback code
 704 * just walks the superblock inode list, writing back any inodes which are
 705 * older than a specific point in time.
 706 *
 707 * Try to run once per dirty_writeback_interval.  But if a writeback event
 708 * takes longer than a dirty_writeback_interval interval, then leave a
 709 * one-second gap.
 710 *
 711 * older_than_this takes precedence over nr_to_write.  So we'll only write back
 712 * all dirty pages if they are all attached to "old" mappings.
 713 */
 714static long wb_writeback(struct bdi_writeback *wb,
 715                         struct wb_writeback_work *work)
 716{
 717        unsigned long wb_start = jiffies;
 718        long nr_pages = work->nr_pages;
 719        unsigned long oldest_jif;
 720        struct inode *inode;
 721        long progress;
 722
 723        oldest_jif = jiffies;
 724        work->older_than_this = &oldest_jif;
 725
 726        spin_lock(&wb->list_lock);
 727        for (;;) {
 728                /*
 729                 * Stop writeback when nr_pages has been consumed
 730                 */
 731                if (work->nr_pages <= 0)
 732                        break;
 733
 734                /*
 735                 * Background writeout and kupdate-style writeback may
 736                 * run forever. Stop them if there is other work to do
 737                 * so that e.g. sync can proceed. They'll be restarted
 738                 * after the other works are all done.
 739                 */
 740                if ((work->for_background || work->for_kupdate) &&
 741                    !list_empty(&wb->bdi->work_list))
 742                        break;
 743
 744                /*
 745                 * For background writeout, stop when we are below the
 746                 * background dirty threshold
 747                 */
 748                if (work->for_background && !over_bground_thresh(wb->bdi))
 749                        break;
 750
 751                /*
 752                 * Kupdate and background works are special and we want to
 753                 * include all inodes that need writing. Livelock avoidance is
 754                 * handled by these works yielding to any other work so we are
 755                 * safe.
 756                 */
 757                if (work->for_kupdate) {
 758                        oldest_jif = jiffies -
 759                                msecs_to_jiffies(dirty_expire_interval * 10);
 760                } else if (work->for_background)
 761                        oldest_jif = jiffies;
 762
 763                trace_writeback_start(wb->bdi, work);
 764                if (list_empty(&wb->b_io))
 765                        queue_io(wb, work);
 766                if (work->sb)
 767                        progress = writeback_sb_inodes(work->sb, wb, work);
 768                else
 769                        progress = __writeback_inodes_wb(wb, work);
 770                trace_writeback_written(wb->bdi, work);
 771
 772                wb_update_bandwidth(wb, wb_start);
 773
 774                /*
 775                 * Did we write something? Try for more
 776                 *
 777                 * Dirty inodes are moved to b_io for writeback in batches.
 778                 * The completion of the current batch does not necessarily
 779                 * mean the overall work is done. So we keep looping as long
 780                 * as made some progress on cleaning pages or inodes.
 781                 */
 782                if (progress)
 783                        continue;
 784                /*
 785                 * No more inodes for IO, bail
 786                 */
 787                if (list_empty(&wb->b_more_io))
 788                        break;
 789                /*
 790                 * Nothing written. Wait for some inode to
 791                 * become available for writeback. Otherwise
 792                 * we'll just busyloop.
 793                 */
 794                if (!list_empty(&wb->b_more_io))  {
 795                        trace_writeback_wait(wb->bdi, work);
 796                        inode = wb_inode(wb->b_more_io.prev);
 797                        spin_lock(&inode->i_lock);
 798                        inode_wait_for_writeback(inode, wb);
 799                        spin_unlock(&inode->i_lock);
 800                }
 801        }
 802        spin_unlock(&wb->list_lock);
 803
 804        return nr_pages - work->nr_pages;
 805}
 806
 807/*
 808 * Return the next wb_writeback_work struct that hasn't been processed yet.
 809 */
 810static struct wb_writeback_work *
 811get_next_work_item(struct backing_dev_info *bdi)
 812{
 813        struct wb_writeback_work *work = NULL;
 814
 815        spin_lock_bh(&bdi->wb_lock);
 816        if (!list_empty(&bdi->work_list)) {
 817                work = list_entry(bdi->work_list.next,
 818                                  struct wb_writeback_work, list);
 819                list_del_init(&work->list);
 820        }
 821        spin_unlock_bh(&bdi->wb_lock);
 822        return work;
 823}
 824
 825/*
 826 * Add in the number of potentially dirty inodes, because each inode
 827 * write can dirty pagecache in the underlying blockdev.
 828 */
 829static unsigned long get_nr_dirty_pages(void)
 830{
 831        return global_page_state(NR_FILE_DIRTY) +
 832                global_page_state(NR_UNSTABLE_NFS) +
 833                get_nr_dirty_inodes();
 834}
 835
 836static long wb_check_background_flush(struct bdi_writeback *wb)
 837{
 838        if (over_bground_thresh(wb->bdi)) {
 839
 840                struct wb_writeback_work work = {
 841                        .nr_pages       = LONG_MAX,
 842                        .sync_mode      = WB_SYNC_NONE,
 843                        .for_background = 1,
 844                        .range_cyclic   = 1,
 845                        .reason         = WB_REASON_BACKGROUND,
 846                };
 847
 848                return wb_writeback(wb, &work);
 849        }
 850
 851        return 0;
 852}
 853
 854static long wb_check_old_data_flush(struct bdi_writeback *wb)
 855{
 856        unsigned long expired;
 857        long nr_pages;
 858
 859        /*
 860         * When set to zero, disable periodic writeback
 861         */
 862        if (!dirty_writeback_interval)
 863                return 0;
 864
 865        expired = wb->last_old_flush +
 866                        msecs_to_jiffies(dirty_writeback_interval * 10);
 867        if (time_before(jiffies, expired))
 868                return 0;
 869
 870        wb->last_old_flush = jiffies;
 871        nr_pages = get_nr_dirty_pages();
 872
 873        if (nr_pages) {
 874                struct wb_writeback_work work = {
 875                        .nr_pages       = nr_pages,
 876                        .sync_mode      = WB_SYNC_NONE,
 877                        .for_kupdate    = 1,
 878                        .range_cyclic   = 1,
 879                        .reason         = WB_REASON_PERIODIC,
 880                };
 881
 882                return wb_writeback(wb, &work);
 883        }
 884
 885        return 0;
 886}
 887
 888/*
 889 * Retrieve work items and do the writeback they describe
 890 */
 891long wb_do_writeback(struct bdi_writeback *wb, int force_wait)
 892{
 893        struct backing_dev_info *bdi = wb->bdi;
 894        struct wb_writeback_work *work;
 895        long wrote = 0;
 896
 897        set_bit(BDI_writeback_running, &wb->bdi->state);
 898        while ((work = get_next_work_item(bdi)) != NULL) {
 899                /*
 900                 * Override sync mode, in case we must wait for completion
 901                 * because this thread is exiting now.
 902                 */
 903                if (force_wait)
 904                        work->sync_mode = WB_SYNC_ALL;
 905
 906                trace_writeback_exec(bdi, work);
 907
 908                wrote += wb_writeback(wb, work);
 909
 910                /*
 911                 * Notify the caller of completion if this is a synchronous
 912                 * work item, otherwise just free it.
 913                 */
 914                if (work->done)
 915                        complete(work->done);
 916                else
 917                        kfree(work);
 918        }
 919
 920        /*
 921         * Check for periodic writeback, kupdated() style
 922         */
 923        wrote += wb_check_old_data_flush(wb);
 924        wrote += wb_check_background_flush(wb);
 925        clear_bit(BDI_writeback_running, &wb->bdi->state);
 926
 927        return wrote;
 928}
 929
 930/*
 931 * Handle writeback of dirty data for the device backed by this bdi. Also
 932 * wakes up periodically and does kupdated style flushing.
 933 */
 934int bdi_writeback_thread(void *data)
 935{
 936        struct bdi_writeback *wb = data;
 937        struct backing_dev_info *bdi = wb->bdi;
 938        long pages_written;
 939
 940        current->flags |= PF_SWAPWRITE;
 941        set_freezable();
 942        wb->last_active = jiffies;
 943
 944        /*
 945         * Our parent may run at a different priority, just set us to normal
 946         */
 947        set_user_nice(current, 0);
 948
 949        trace_writeback_thread_start(bdi);
 950
 951        while (!kthread_freezable_should_stop(NULL)) {
 952                /*
 953                 * Remove own delayed wake-up timer, since we are already awake
 954                 * and we'll take care of the preriodic write-back.
 955                 */
 956                del_timer(&wb->wakeup_timer);
 957
 958                pages_written = wb_do_writeback(wb, 0);
 959
 960                trace_writeback_pages_written(pages_written);
 961
 962                if (pages_written)
 963                        wb->last_active = jiffies;
 964
 965                set_current_state(TASK_INTERRUPTIBLE);
 966                if (!list_empty(&bdi->work_list) || kthread_should_stop()) {
 967                        __set_current_state(TASK_RUNNING);
 968                        continue;
 969                }
 970
 971                if (wb_has_dirty_io(wb) && dirty_writeback_interval)
 972                        schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10));
 973                else {
 974                        /*
 975                         * We have nothing to do, so can go sleep without any
 976                         * timeout and save power. When a work is queued or
 977                         * something is made dirty - we will be woken up.
 978                         */
 979                        schedule();
 980                }
 981        }
 982
 983        /* Flush any work that raced with us exiting */
 984        if (!list_empty(&bdi->work_list))
 985                wb_do_writeback(wb, 1);
 986
 987        trace_writeback_thread_stop(bdi);
 988        return 0;
 989}
 990
 991
 992/*
 993 * Start writeback of `nr_pages' pages.  If `nr_pages' is zero, write back
 994 * the whole world.
 995 */
 996void wakeup_flusher_threads(long nr_pages, enum wb_reason reason)
 997{
 998        struct backing_dev_info *bdi;
 999
1000        if (!nr_pages) {
1001                nr_pages = global_page_state(NR_FILE_DIRTY) +
1002                                global_page_state(NR_UNSTABLE_NFS);
1003        }
1004
1005        rcu_read_lock();
1006        list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) {
1007                if (!bdi_has_dirty_io(bdi))
1008                        continue;
1009                __bdi_start_writeback(bdi, nr_pages, false, reason);
1010        }
1011        rcu_read_unlock();
1012}
1013
1014static noinline void block_dump___mark_inode_dirty(struct inode *inode)
1015{
1016        if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) {
1017                struct dentry *dentry;
1018                const char *name = "?";
1019
1020                dentry = d_find_alias(inode);
1021                if (dentry) {
1022                        spin_lock(&dentry->d_lock);
1023                        name = (const char *) dentry->d_name.name;
1024                }
1025                printk(KERN_DEBUG
1026                       "%s(%d): dirtied inode %lu (%s) on %s\n",
1027                       current->comm, task_pid_nr(current), inode->i_ino,
1028                       name, inode->i_sb->s_id);
1029                if (dentry) {
1030                        spin_unlock(&dentry->d_lock);
1031                        dput(dentry);
1032                }
1033        }
1034}
1035
1036/**
1037 *      __mark_inode_dirty -    internal function
1038 *      @inode: inode to mark
1039 *      @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
1040 *      Mark an inode as dirty. Callers should use mark_inode_dirty or
1041 *      mark_inode_dirty_sync.
1042 *
1043 * Put the inode on the super block's dirty list.
1044 *
1045 * CAREFUL! We mark it dirty unconditionally, but move it onto the
1046 * dirty list only if it is hashed or if it refers to a blockdev.
1047 * If it was not hashed, it will never be added to the dirty list
1048 * even if it is later hashed, as it will have been marked dirty already.
1049 *
1050 * In short, make sure you hash any inodes _before_ you start marking
1051 * them dirty.
1052 *
1053 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
1054 * the block-special inode (/dev/hda1) itself.  And the ->dirtied_when field of
1055 * the kernel-internal blockdev inode represents the dirtying time of the
1056 * blockdev's pages.  This is why for I_DIRTY_PAGES we always use
1057 * page->mapping->host, so the page-dirtying time is recorded in the internal
1058 * blockdev inode.
1059 */
1060void __mark_inode_dirty(struct inode *inode, int flags)
1061{
1062        struct super_block *sb = inode->i_sb;
1063        struct backing_dev_info *bdi = NULL;
1064
1065        /*
1066         * Don't do this for I_DIRTY_PAGES - that doesn't actually
1067         * dirty the inode itself
1068         */
1069        if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
1070                if (sb->s_op->dirty_inode)
1071                        sb->s_op->dirty_inode(inode, flags);
1072        }
1073
1074        /*
1075         * make sure that changes are seen by all cpus before we test i_state
1076         * -- mikulas
1077         */
1078        smp_mb();
1079
1080        /* avoid the locking if we can */
1081        if ((inode->i_state & flags) == flags)
1082                return;
1083
1084        if (unlikely(block_dump))
1085                block_dump___mark_inode_dirty(inode);
1086
1087        spin_lock(&inode->i_lock);
1088        if ((inode->i_state & flags) != flags) {
1089                const int was_dirty = inode->i_state & I_DIRTY;
1090
1091                inode->i_state |= flags;
1092
1093                /*
1094                 * If the inode is being synced, just update its dirty state.
1095                 * The unlocker will place the inode on the appropriate
1096                 * superblock list, based upon its state.
1097                 */
1098                if (inode->i_state & I_SYNC)
1099                        goto out_unlock_inode;
1100
1101                /*
1102                 * Only add valid (hashed) inodes to the superblock's
1103                 * dirty list.  Add blockdev inodes as well.
1104                 */
1105                if (!S_ISBLK(inode->i_mode)) {
1106                        if (inode_unhashed(inode))
1107                                goto out_unlock_inode;
1108                }
1109                if (inode->i_state & I_FREEING)
1110                        goto out_unlock_inode;
1111
1112                /*
1113                 * If the inode was already on b_dirty/b_io/b_more_io, don't
1114                 * reposition it (that would break b_dirty time-ordering).
1115                 */
1116                if (!was_dirty) {
1117                        bool wakeup_bdi = false;
1118                        bdi = inode_to_bdi(inode);
1119
1120                        if (bdi_cap_writeback_dirty(bdi)) {
1121                                WARN(!test_bit(BDI_registered, &bdi->state),
1122                                     "bdi-%s not registered\n", bdi->name);
1123
1124                                /*
1125                                 * If this is the first dirty inode for this
1126                                 * bdi, we have to wake-up the corresponding
1127                                 * bdi thread to make sure background
1128                                 * write-back happens later.
1129                                 */
1130                                if (!wb_has_dirty_io(&bdi->wb))
1131                                        wakeup_bdi = true;
1132                        }
1133
1134                        spin_unlock(&inode->i_lock);
1135                        spin_lock(&bdi->wb.list_lock);
1136                        inode->dirtied_when = jiffies;
1137                        list_move(&inode->i_wb_list, &bdi->wb.b_dirty);
1138                        spin_unlock(&bdi->wb.list_lock);
1139
1140                        if (wakeup_bdi)
1141                                bdi_wakeup_thread_delayed(bdi);
1142                        return;
1143                }
1144        }
1145out_unlock_inode:
1146        spin_unlock(&inode->i_lock);
1147
1148}
1149EXPORT_SYMBOL(__mark_inode_dirty);
1150
1151/*
1152 * Write out a superblock's list of dirty inodes.  A wait will be performed
1153 * upon no inodes, all inodes or the final one, depending upon sync_mode.
1154 *
1155 * If older_than_this is non-NULL, then only write out inodes which
1156 * had their first dirtying at a time earlier than *older_than_this.
1157 *
1158 * If `bdi' is non-zero then we're being asked to writeback a specific queue.
1159 * This function assumes that the blockdev superblock's inodes are backed by
1160 * a variety of queues, so all inodes are searched.  For other superblocks,
1161 * assume that all inodes are backed by the same queue.
1162 *
1163 * The inodes to be written are parked on bdi->b_io.  They are moved back onto
1164 * bdi->b_dirty as they are selected for writing.  This way, none can be missed
1165 * on the writer throttling path, and we get decent balancing between many
1166 * throttled threads: we don't want them all piling up on inode_sync_wait.
1167 */
1168static void wait_sb_inodes(struct super_block *sb)
1169{
1170        struct inode *inode, *old_inode = NULL;
1171
1172        /*
1173         * We need to be protected against the filesystem going from
1174         * r/o to r/w or vice versa.
1175         */
1176        WARN_ON(!rwsem_is_locked(&sb->s_umount));
1177
1178        spin_lock(&inode_sb_list_lock);
1179
1180        /*
1181         * Data integrity sync. Must wait for all pages under writeback,
1182         * because there may have been pages dirtied before our sync
1183         * call, but which had writeout started before we write it out.
1184         * In which case, the inode may not be on the dirty list, but
1185         * we still have to wait for that writeout.
1186         */
1187        list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1188                struct address_space *mapping = inode->i_mapping;
1189
1190                spin_lock(&inode->i_lock);
1191                if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
1192                    (mapping->nrpages == 0)) {
1193                        spin_unlock(&inode->i_lock);
1194                        continue;
1195                }
1196                __iget(inode);
1197                spin_unlock(&inode->i_lock);
1198                spin_unlock(&inode_sb_list_lock);
1199
1200                /*
1201                 * We hold a reference to 'inode' so it couldn't have been
1202                 * removed from s_inodes list while we dropped the
1203                 * inode_sb_list_lock.  We cannot iput the inode now as we can
1204                 * be holding the last reference and we cannot iput it under
1205                 * inode_sb_list_lock. So we keep the reference and iput it
1206                 * later.
1207                 */
1208                iput(old_inode);
1209                old_inode = inode;
1210
1211                filemap_fdatawait(mapping);
1212
1213                cond_resched();
1214
1215                spin_lock(&inode_sb_list_lock);
1216        }
1217        spin_unlock(&inode_sb_list_lock);
1218        iput(old_inode);
1219}
1220
1221/**
1222 * writeback_inodes_sb_nr -     writeback dirty inodes from given super_block
1223 * @sb: the superblock
1224 * @nr: the number of pages to write
1225 * @reason: reason why some writeback work initiated
1226 *
1227 * Start writeback on some inodes on this super_block. No guarantees are made
1228 * on how many (if any) will be written, and this function does not wait
1229 * for IO completion of submitted IO.
1230 */
1231void writeback_inodes_sb_nr(struct super_block *sb,
1232                            unsigned long nr,
1233                            enum wb_reason reason)
1234{
1235        DECLARE_COMPLETION_ONSTACK(done);
1236        struct wb_writeback_work work = {
1237                .sb                     = sb,
1238                .sync_mode              = WB_SYNC_NONE,
1239                .tagged_writepages      = 1,
1240                .done                   = &done,
1241                .nr_pages               = nr,
1242                .reason                 = reason,
1243        };
1244
1245        WARN_ON(!rwsem_is_locked(&sb->s_umount));
1246        bdi_queue_work(sb->s_bdi, &work);
1247        wait_for_completion(&done);
1248}
1249EXPORT_SYMBOL(writeback_inodes_sb_nr);
1250
1251/**
1252 * writeback_inodes_sb  -       writeback dirty inodes from given super_block
1253 * @sb: the superblock
1254 * @reason: reason why some writeback work was initiated
1255 *
1256 * Start writeback on some inodes on this super_block. No guarantees are made
1257 * on how many (if any) will be written, and this function does not wait
1258 * for IO completion of submitted IO.
1259 */
1260void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason)
1261{
1262        return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason);
1263}
1264EXPORT_SYMBOL(writeback_inodes_sb);
1265
1266/**
1267 * writeback_inodes_sb_if_idle  -       start writeback if none underway
1268 * @sb: the superblock
1269 * @reason: reason why some writeback work was initiated
1270 *
1271 * Invoke writeback_inodes_sb if no writeback is currently underway.
1272 * Returns 1 if writeback was started, 0 if not.
1273 */
1274int writeback_inodes_sb_if_idle(struct super_block *sb, enum wb_reason reason)
1275{
1276        if (!writeback_in_progress(sb->s_bdi)) {
1277                down_read(&sb->s_umount);
1278                writeback_inodes_sb(sb, reason);
1279                up_read(&sb->s_umount);
1280                return 1;
1281        } else
1282                return 0;
1283}
1284EXPORT_SYMBOL(writeback_inodes_sb_if_idle);
1285
1286/**
1287 * writeback_inodes_sb_if_idle  -       start writeback if none underway
1288 * @sb: the superblock
1289 * @nr: the number of pages to write
1290 * @reason: reason why some writeback work was initiated
1291 *
1292 * Invoke writeback_inodes_sb if no writeback is currently underway.
1293 * Returns 1 if writeback was started, 0 if not.
1294 */
1295int writeback_inodes_sb_nr_if_idle(struct super_block *sb,
1296                                   unsigned long nr,
1297                                   enum wb_reason reason)
1298{
1299        if (!writeback_in_progress(sb->s_bdi)) {
1300                down_read(&sb->s_umount);
1301                writeback_inodes_sb_nr(sb, nr, reason);
1302                up_read(&sb->s_umount);
1303                return 1;
1304        } else
1305                return 0;
1306}
1307EXPORT_SYMBOL(writeback_inodes_sb_nr_if_idle);
1308
1309/**
1310 * sync_inodes_sb       -       sync sb inode pages
1311 * @sb: the superblock
1312 *
1313 * This function writes and waits on any dirty inode belonging to this
1314 * super_block.
1315 */
1316void sync_inodes_sb(struct super_block *sb)
1317{
1318        DECLARE_COMPLETION_ONSTACK(done);
1319        struct wb_writeback_work work = {
1320                .sb             = sb,
1321                .sync_mode      = WB_SYNC_ALL,
1322                .nr_pages       = LONG_MAX,
1323                .range_cyclic   = 0,
1324                .done           = &done,
1325                .reason         = WB_REASON_SYNC,
1326        };
1327
1328        WARN_ON(!rwsem_is_locked(&sb->s_umount));
1329
1330        bdi_queue_work(sb->s_bdi, &work);
1331        wait_for_completion(&done);
1332
1333        wait_sb_inodes(sb);
1334}
1335EXPORT_SYMBOL(sync_inodes_sb);
1336
1337/**
1338 * write_inode_now      -       write an inode to disk
1339 * @inode: inode to write to disk
1340 * @sync: whether the write should be synchronous or not
1341 *
1342 * This function commits an inode to disk immediately if it is dirty. This is
1343 * primarily needed by knfsd.
1344 *
1345 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
1346 */
1347int write_inode_now(struct inode *inode, int sync)
1348{
1349        struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
1350        int ret;
1351        struct writeback_control wbc = {
1352                .nr_to_write = LONG_MAX,
1353                .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE,
1354                .range_start = 0,
1355                .range_end = LLONG_MAX,
1356        };
1357
1358        if (!mapping_cap_writeback_dirty(inode->i_mapping))
1359                wbc.nr_to_write = 0;
1360
1361        might_sleep();
1362        spin_lock(&wb->list_lock);
1363        spin_lock(&inode->i_lock);
1364        ret = writeback_single_inode(inode, wb, &wbc);
1365        spin_unlock(&inode->i_lock);
1366        spin_unlock(&wb->list_lock);
1367        if (sync)
1368                inode_sync_wait(inode);
1369        return ret;
1370}
1371EXPORT_SYMBOL(write_inode_now);
1372
1373/**
1374 * sync_inode - write an inode and its pages to disk.
1375 * @inode: the inode to sync
1376 * @wbc: controls the writeback mode
1377 *
1378 * sync_inode() will write an inode and its pages to disk.  It will also
1379 * correctly update the inode on its superblock's dirty inode lists and will
1380 * update inode->i_state.
1381 *
1382 * The caller must have a ref on the inode.
1383 */
1384int sync_inode(struct inode *inode, struct writeback_control *wbc)
1385{
1386        struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
1387        int ret;
1388
1389        spin_lock(&wb->list_lock);
1390        spin_lock(&inode->i_lock);
1391        ret = writeback_single_inode(inode, wb, wbc);
1392        spin_unlock(&inode->i_lock);
1393        spin_unlock(&wb->list_lock);
1394        return ret;
1395}
1396EXPORT_SYMBOL(sync_inode);
1397
1398/**
1399 * sync_inode_metadata - write an inode to disk
1400 * @inode: the inode to sync
1401 * @wait: wait for I/O to complete.
1402 *
1403 * Write an inode to disk and adjust its dirty state after completion.
1404 *
1405 * Note: only writes the actual inode, no associated data or other metadata.
1406 */
1407int sync_inode_metadata(struct inode *inode, int wait)
1408{
1409        struct writeback_control wbc = {
1410                .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE,
1411                .nr_to_write = 0, /* metadata-only */
1412        };
1413
1414        return sync_inode(inode, &wbc);
1415}
1416EXPORT_SYMBOL(sync_inode_metadata);
1417