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/export.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 * writeback_in_progress - determine whether there is writeback in progress
  57 * @bdi: the device's backing_dev_info structure.
  58 *
  59 * Determine whether there is writeback waiting to be handled against a
  60 * backing device.
  61 */
  62int writeback_in_progress(struct backing_dev_info *bdi)
  63{
  64        return test_bit(BDI_writeback_running, &bdi->state);
  65}
  66EXPORT_SYMBOL(writeback_in_progress);
  67
  68static inline struct backing_dev_info *inode_to_bdi(struct inode *inode)
  69{
  70        struct super_block *sb = inode->i_sb;
  71
  72        if (strcmp(sb->s_type->name, "bdev") == 0)
  73                return inode->i_mapping->backing_dev_info;
  74
  75        return sb->s_bdi;
  76}
  77
  78static inline struct inode *wb_inode(struct list_head *head)
  79{
  80        return list_entry(head, struct inode, i_wb_list);
  81}
  82
  83/*
  84 * Include the creation of the trace points after defining the
  85 * wb_writeback_work structure and inline functions so that the definition
  86 * remains local to this file.
  87 */
  88#define CREATE_TRACE_POINTS
  89#include <trace/events/writeback.h>
  90
  91/* Wakeup flusher thread or forker thread to fork it. Requires bdi->wb_lock. */
  92static void bdi_wakeup_flusher(struct backing_dev_info *bdi)
  93{
  94        if (bdi->wb.task) {
  95                wake_up_process(bdi->wb.task);
  96        } else {
  97                /*
  98                 * The bdi thread isn't there, wake up the forker thread which
  99                 * will create and run it.
 100                 */
 101                wake_up_process(default_backing_dev_info.wb.task);
 102        }
 103}
 104
 105static void bdi_queue_work(struct backing_dev_info *bdi,
 106                           struct wb_writeback_work *work)
 107{
 108        trace_writeback_queue(bdi, work);
 109
 110        spin_lock_bh(&bdi->wb_lock);
 111        list_add_tail(&work->list, &bdi->work_list);
 112        if (!bdi->wb.task)
 113                trace_writeback_nothread(bdi, work);
 114        bdi_wakeup_flusher(bdi);
 115        spin_unlock_bh(&bdi->wb_lock);
 116}
 117
 118static void
 119__bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages,
 120                      bool range_cyclic, enum wb_reason reason)
 121{
 122        struct wb_writeback_work *work;
 123
 124        /*
 125         * This is WB_SYNC_NONE writeback, so if allocation fails just
 126         * wakeup the thread for old dirty data writeback
 127         */
 128        work = kzalloc(sizeof(*work), GFP_ATOMIC);
 129        if (!work) {
 130                if (bdi->wb.task) {
 131                        trace_writeback_nowork(bdi);
 132                        wake_up_process(bdi->wb.task);
 133                }
 134                return;
 135        }
 136
 137        work->sync_mode = WB_SYNC_NONE;
 138        work->nr_pages  = nr_pages;
 139        work->range_cyclic = range_cyclic;
 140        work->reason    = reason;
 141
 142        bdi_queue_work(bdi, work);
 143}
 144
 145/**
 146 * bdi_start_writeback - start writeback
 147 * @bdi: the backing device to write from
 148 * @nr_pages: the number of pages to write
 149 * @reason: reason why some writeback work was initiated
 150 *
 151 * Description:
 152 *   This does WB_SYNC_NONE opportunistic writeback. The IO is only
 153 *   started when this function returns, we make no guarantees on
 154 *   completion. Caller need not hold sb s_umount semaphore.
 155 *
 156 */
 157void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages,
 158                        enum wb_reason reason)
 159{
 160        __bdi_start_writeback(bdi, nr_pages, true, reason);
 161}
 162
 163/**
 164 * bdi_start_background_writeback - start background writeback
 165 * @bdi: the backing device to write from
 166 *
 167 * Description:
 168 *   This makes sure WB_SYNC_NONE background writeback happens. When
 169 *   this function returns, it is only guaranteed that for given BDI
 170 *   some IO is happening if we are over background dirty threshold.
 171 *   Caller need not hold sb s_umount semaphore.
 172 */
 173void bdi_start_background_writeback(struct backing_dev_info *bdi)
 174{
 175        /*
 176         * We just wake up the flusher thread. It will perform background
 177         * writeback as soon as there is no other work to do.
 178         */
 179        trace_writeback_wake_background(bdi);
 180        spin_lock_bh(&bdi->wb_lock);
 181        bdi_wakeup_flusher(bdi);
 182        spin_unlock_bh(&bdi->wb_lock);
 183}
 184
 185/*
 186 * Remove the inode from the writeback list it is on.
 187 */
 188void inode_wb_list_del(struct inode *inode)
 189{
 190        struct backing_dev_info *bdi = inode_to_bdi(inode);
 191
 192        spin_lock(&bdi->wb.list_lock);
 193        list_del_init(&inode->i_wb_list);
 194        spin_unlock(&bdi->wb.list_lock);
 195}
 196
 197/*
 198 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
 199 * furthest end of its superblock's dirty-inode list.
 200 *
 201 * Before stamping the inode's ->dirtied_when, we check to see whether it is
 202 * already the most-recently-dirtied inode on the b_dirty list.  If that is
 203 * the case then the inode must have been redirtied while it was being written
 204 * out and we don't reset its dirtied_when.
 205 */
 206static void redirty_tail(struct inode *inode, struct bdi_writeback *wb)
 207{
 208        assert_spin_locked(&wb->list_lock);
 209        if (!list_empty(&wb->b_dirty)) {
 210                struct inode *tail;
 211
 212                tail = wb_inode(wb->b_dirty.next);
 213                if (time_before(inode->dirtied_when, tail->dirtied_when))
 214                        inode->dirtied_when = jiffies;
 215        }
 216        list_move(&inode->i_wb_list, &wb->b_dirty);
 217}
 218
 219/*
 220 * requeue inode for re-scanning after bdi->b_io list is exhausted.
 221 */
 222static void requeue_io(struct inode *inode, struct bdi_writeback *wb)
 223{
 224        assert_spin_locked(&wb->list_lock);
 225        list_move(&inode->i_wb_list, &wb->b_more_io);
 226}
 227
 228static void inode_sync_complete(struct inode *inode)
 229{
 230        inode->i_state &= ~I_SYNC;
 231        /* If inode is clean an unused, put it into LRU now... */
 232        inode_add_lru(inode);
 233        /* Waiters must see I_SYNC cleared before being woken up */
 234        smp_mb();
 235        wake_up_bit(&inode->i_state, __I_SYNC);
 236}
 237
 238static bool inode_dirtied_after(struct inode *inode, unsigned long t)
 239{
 240        bool ret = time_after(inode->dirtied_when, t);
 241#ifndef CONFIG_64BIT
 242        /*
 243         * For inodes being constantly redirtied, dirtied_when can get stuck.
 244         * It _appears_ to be in the future, but is actually in distant past.
 245         * This test is necessary to prevent such wrapped-around relative times
 246         * from permanently stopping the whole bdi writeback.
 247         */
 248        ret = ret && time_before_eq(inode->dirtied_when, jiffies);
 249#endif
 250        return ret;
 251}
 252
 253/*
 254 * Move expired (dirtied after work->older_than_this) dirty inodes from
 255 * @delaying_queue to @dispatch_queue.
 256 */
 257static int move_expired_inodes(struct list_head *delaying_queue,
 258                               struct list_head *dispatch_queue,
 259                               struct wb_writeback_work *work)
 260{
 261        LIST_HEAD(tmp);
 262        struct list_head *pos, *node;
 263        struct super_block *sb = NULL;
 264        struct inode *inode;
 265        int do_sb_sort = 0;
 266        int moved = 0;
 267
 268        while (!list_empty(delaying_queue)) {
 269                inode = wb_inode(delaying_queue->prev);
 270                if (work->older_than_this &&
 271                    inode_dirtied_after(inode, *work->older_than_this))
 272                        break;
 273                if (sb && sb != inode->i_sb)
 274                        do_sb_sort = 1;
 275                sb = inode->i_sb;
 276                list_move(&inode->i_wb_list, &tmp);
 277                moved++;
 278        }
 279
 280        /* just one sb in list, splice to dispatch_queue and we're done */
 281        if (!do_sb_sort) {
 282                list_splice(&tmp, dispatch_queue);
 283                goto out;
 284        }
 285
 286        /* Move inodes from one superblock together */
 287        while (!list_empty(&tmp)) {
 288                sb = wb_inode(tmp.prev)->i_sb;
 289                list_for_each_prev_safe(pos, node, &tmp) {
 290                        inode = wb_inode(pos);
 291                        if (inode->i_sb == sb)
 292                                list_move(&inode->i_wb_list, dispatch_queue);
 293                }
 294        }
 295out:
 296        return moved;
 297}
 298
 299/*
 300 * Queue all expired dirty inodes for io, eldest first.
 301 * Before
 302 *         newly dirtied     b_dirty    b_io    b_more_io
 303 *         =============>    gf         edc     BA
 304 * After
 305 *         newly dirtied     b_dirty    b_io    b_more_io
 306 *         =============>    g          fBAedc
 307 *                                           |
 308 *                                           +--> dequeue for IO
 309 */
 310static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work)
 311{
 312        int moved;
 313        assert_spin_locked(&wb->list_lock);
 314        list_splice_init(&wb->b_more_io, &wb->b_io);
 315        moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, work);
 316        trace_writeback_queue_io(wb, work, moved);
 317}
 318
 319static int write_inode(struct inode *inode, struct writeback_control *wbc)
 320{
 321        if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode))
 322                return inode->i_sb->s_op->write_inode(inode, wbc);
 323        return 0;
 324}
 325
 326/*
 327 * Wait for writeback on an inode to complete. Called with i_lock held.
 328 * Caller must make sure inode cannot go away when we drop i_lock.
 329 */
 330static void __inode_wait_for_writeback(struct inode *inode)
 331        __releases(inode->i_lock)
 332        __acquires(inode->i_lock)
 333{
 334        DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC);
 335        wait_queue_head_t *wqh;
 336
 337        wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
 338        while (inode->i_state & I_SYNC) {
 339                spin_unlock(&inode->i_lock);
 340                __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE);
 341                spin_lock(&inode->i_lock);
 342        }
 343}
 344
 345/*
 346 * Wait for writeback on an inode to complete. Caller must have inode pinned.
 347 */
 348void inode_wait_for_writeback(struct inode *inode)
 349{
 350        spin_lock(&inode->i_lock);
 351        __inode_wait_for_writeback(inode);
 352        spin_unlock(&inode->i_lock);
 353}
 354
 355/*
 356 * Sleep until I_SYNC is cleared. This function must be called with i_lock
 357 * held and drops it. It is aimed for callers not holding any inode reference
 358 * so once i_lock is dropped, inode can go away.
 359 */
 360static void inode_sleep_on_writeback(struct inode *inode)
 361        __releases(inode->i_lock)
 362{
 363        DEFINE_WAIT(wait);
 364        wait_queue_head_t *wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
 365        int sleep;
 366
 367        prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
 368        sleep = inode->i_state & I_SYNC;
 369        spin_unlock(&inode->i_lock);
 370        if (sleep)
 371                schedule();
 372        finish_wait(wqh, &wait);
 373}
 374
 375/*
 376 * Find proper writeback list for the inode depending on its current state and
 377 * possibly also change of its state while we were doing writeback.  Here we
 378 * handle things such as livelock prevention or fairness of writeback among
 379 * inodes. This function can be called only by flusher thread - noone else
 380 * processes all inodes in writeback lists and requeueing inodes behind flusher
 381 * thread's back can have unexpected consequences.
 382 */
 383static void requeue_inode(struct inode *inode, struct bdi_writeback *wb,
 384                          struct writeback_control *wbc)
 385{
 386        if (inode->i_state & I_FREEING)
 387                return;
 388
 389        /*
 390         * Sync livelock prevention. Each inode is tagged and synced in one
 391         * shot. If still dirty, it will be redirty_tail()'ed below.  Update
 392         * the dirty time to prevent enqueue and sync it again.
 393         */
 394        if ((inode->i_state & I_DIRTY) &&
 395            (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages))
 396                inode->dirtied_when = jiffies;
 397
 398        if (wbc->pages_skipped) {
 399                /*
 400                 * writeback is not making progress due to locked
 401                 * buffers. Skip this inode for now.
 402                 */
 403                redirty_tail(inode, wb);
 404                return;
 405        }
 406
 407        if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) {
 408                /*
 409                 * We didn't write back all the pages.  nfs_writepages()
 410                 * sometimes bales out without doing anything.
 411                 */
 412                if (wbc->nr_to_write <= 0) {
 413                        /* Slice used up. Queue for next turn. */
 414                        requeue_io(inode, wb);
 415                } else {
 416                        /*
 417                         * Writeback blocked by something other than
 418                         * congestion. Delay the inode for some time to
 419                         * avoid spinning on the CPU (100% iowait)
 420                         * retrying writeback of the dirty page/inode
 421                         * that cannot be performed immediately.
 422                         */
 423                        redirty_tail(inode, wb);
 424                }
 425        } else if (inode->i_state & I_DIRTY) {
 426                /*
 427                 * Filesystems can dirty the inode during writeback operations,
 428                 * such as delayed allocation during submission or metadata
 429                 * updates after data IO completion.
 430                 */
 431                redirty_tail(inode, wb);
 432        } else {
 433                /* The inode is clean. Remove from writeback lists. */
 434                list_del_init(&inode->i_wb_list);
 435        }
 436}
 437
 438/*
 439 * Write out an inode and its dirty pages. Do not update the writeback list
 440 * linkage. That is left to the caller. The caller is also responsible for
 441 * setting I_SYNC flag and calling inode_sync_complete() to clear it.
 442 */
 443static int
 444__writeback_single_inode(struct inode *inode, struct bdi_writeback *wb,
 445                         struct writeback_control *wbc)
 446{
 447        struct address_space *mapping = inode->i_mapping;
 448        long nr_to_write = wbc->nr_to_write;
 449        unsigned dirty;
 450        int ret;
 451
 452        WARN_ON(!(inode->i_state & I_SYNC));
 453
 454        ret = do_writepages(mapping, wbc);
 455
 456        /*
 457         * Make sure to wait on the data before writing out the metadata.
 458         * This is important for filesystems that modify metadata on data
 459         * I/O completion.
 460         */
 461        if (wbc->sync_mode == WB_SYNC_ALL) {
 462                int err = filemap_fdatawait(mapping);
 463                if (ret == 0)
 464                        ret = err;
 465        }
 466
 467        /*
 468         * Some filesystems may redirty the inode during the writeback
 469         * due to delalloc, clear dirty metadata flags right before
 470         * write_inode()
 471         */
 472        spin_lock(&inode->i_lock);
 473        /* Clear I_DIRTY_PAGES if we've written out all dirty pages */
 474        if (!mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
 475                inode->i_state &= ~I_DIRTY_PAGES;
 476        dirty = inode->i_state & I_DIRTY;
 477        inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC);
 478        spin_unlock(&inode->i_lock);
 479        /* Don't write the inode if only I_DIRTY_PAGES was set */
 480        if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
 481                int err = write_inode(inode, wbc);
 482                if (ret == 0)
 483                        ret = err;
 484        }
 485        trace_writeback_single_inode(inode, wbc, nr_to_write);
 486        return ret;
 487}
 488
 489/*
 490 * Write out an inode's dirty pages. Either the caller has an active reference
 491 * on the inode or the inode has I_WILL_FREE set.
 492 *
 493 * This function is designed to be called for writing back one inode which
 494 * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode()
 495 * and does more profound writeback list handling in writeback_sb_inodes().
 496 */
 497static int
 498writeback_single_inode(struct inode *inode, struct bdi_writeback *wb,
 499                       struct writeback_control *wbc)
 500{
 501        int ret = 0;
 502
 503        spin_lock(&inode->i_lock);
 504        if (!atomic_read(&inode->i_count))
 505                WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING)));
 506        else
 507                WARN_ON(inode->i_state & I_WILL_FREE);
 508
 509        if (inode->i_state & I_SYNC) {
 510                if (wbc->sync_mode != WB_SYNC_ALL)
 511                        goto out;
 512                /*
 513                 * It's a data-integrity sync. We must wait. Since callers hold
 514                 * inode reference or inode has I_WILL_FREE set, it cannot go
 515                 * away under us.
 516                 */
 517                __inode_wait_for_writeback(inode);
 518        }
 519        WARN_ON(inode->i_state & I_SYNC);
 520        /*
 521         * Skip inode if it is clean. We don't want to mess with writeback
 522         * lists in this function since flusher thread may be doing for example
 523         * sync in parallel and if we move the inode, it could get skipped. So
 524         * here we make sure inode is on some writeback list and leave it there
 525         * unless we have completely cleaned the inode.
 526         */
 527        if (!(inode->i_state & I_DIRTY))
 528                goto out;
 529        inode->i_state |= I_SYNC;
 530        spin_unlock(&inode->i_lock);
 531
 532        ret = __writeback_single_inode(inode, wb, wbc);
 533
 534        spin_lock(&wb->list_lock);
 535        spin_lock(&inode->i_lock);
 536        /*
 537         * If inode is clean, remove it from writeback lists. Otherwise don't
 538         * touch it. See comment above for explanation.
 539         */
 540        if (!(inode->i_state & I_DIRTY))
 541                list_del_init(&inode->i_wb_list);
 542        spin_unlock(&wb->list_lock);
 543        inode_sync_complete(inode);
 544out:
 545        spin_unlock(&inode->i_lock);
 546        return ret;
 547}
 548
 549static long writeback_chunk_size(struct backing_dev_info *bdi,
 550                                 struct wb_writeback_work *work)
 551{
 552        long pages;
 553
 554        /*
 555         * WB_SYNC_ALL mode does livelock avoidance by syncing dirty
 556         * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX
 557         * here avoids calling into writeback_inodes_wb() more than once.
 558         *
 559         * The intended call sequence for WB_SYNC_ALL writeback is:
 560         *
 561         *      wb_writeback()
 562         *          writeback_sb_inodes()       <== called only once
 563         *              write_cache_pages()     <== called once for each inode
 564         *                   (quickly) tag currently dirty pages
 565         *                   (maybe slowly) sync all tagged pages
 566         */
 567        if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages)
 568                pages = LONG_MAX;
 569        else {
 570                pages = min(bdi->avg_write_bandwidth / 2,
 571                            global_dirty_limit / DIRTY_SCOPE);
 572                pages = min(pages, work->nr_pages);
 573                pages = round_down(pages + MIN_WRITEBACK_PAGES,
 574                                   MIN_WRITEBACK_PAGES);
 575        }
 576
 577        return pages;
 578}
 579
 580/*
 581 * Write a portion of b_io inodes which belong to @sb.
 582 *
 583 * If @only_this_sb is true, then find and write all such
 584 * inodes. Otherwise write only ones which go sequentially
 585 * in reverse order.
 586 *
 587 * Return the number of pages and/or inodes written.
 588 */
 589static long writeback_sb_inodes(struct super_block *sb,
 590                                struct bdi_writeback *wb,
 591                                struct wb_writeback_work *work)
 592{
 593        struct writeback_control wbc = {
 594                .sync_mode              = work->sync_mode,
 595                .tagged_writepages      = work->tagged_writepages,
 596                .for_kupdate            = work->for_kupdate,
 597                .for_background         = work->for_background,
 598                .range_cyclic           = work->range_cyclic,
 599                .range_start            = 0,
 600                .range_end              = LLONG_MAX,
 601        };
 602        unsigned long start_time = jiffies;
 603        long write_chunk;
 604        long wrote = 0;  /* count both pages and inodes */
 605
 606        while (!list_empty(&wb->b_io)) {
 607                struct inode *inode = wb_inode(wb->b_io.prev);
 608
 609                if (inode->i_sb != sb) {
 610                        if (work->sb) {
 611                                /*
 612                                 * We only want to write back data for this
 613                                 * superblock, move all inodes not belonging
 614                                 * to it back onto the dirty list.
 615                                 */
 616                                redirty_tail(inode, wb);
 617                                continue;
 618                        }
 619
 620                        /*
 621                         * The inode belongs to a different superblock.
 622                         * Bounce back to the caller to unpin this and
 623                         * pin the next superblock.
 624                         */
 625                        break;
 626                }
 627
 628                /*
 629                 * Don't bother with new inodes or inodes being freed, first
 630                 * kind does not need periodic writeout yet, and for the latter
 631                 * kind writeout is handled by the freer.
 632                 */
 633                spin_lock(&inode->i_lock);
 634                if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
 635                        spin_unlock(&inode->i_lock);
 636                        redirty_tail(inode, wb);
 637                        continue;
 638                }
 639                if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) {
 640                        /*
 641                         * If this inode is locked for writeback and we are not
 642                         * doing writeback-for-data-integrity, move it to
 643                         * b_more_io so that writeback can proceed with the
 644                         * other inodes on s_io.
 645                         *
 646                         * We'll have another go at writing back this inode
 647                         * when we completed a full scan of b_io.
 648                         */
 649                        spin_unlock(&inode->i_lock);
 650                        requeue_io(inode, wb);
 651                        trace_writeback_sb_inodes_requeue(inode);
 652                        continue;
 653                }
 654                spin_unlock(&wb->list_lock);
 655
 656                /*
 657                 * We already requeued the inode if it had I_SYNC set and we
 658                 * are doing WB_SYNC_NONE writeback. So this catches only the
 659                 * WB_SYNC_ALL case.
 660                 */
 661                if (inode->i_state & I_SYNC) {
 662                        /* Wait for I_SYNC. This function drops i_lock... */
 663                        inode_sleep_on_writeback(inode);
 664                        /* Inode may be gone, start again */
 665                        spin_lock(&wb->list_lock);
 666                        continue;
 667                }
 668                inode->i_state |= I_SYNC;
 669                spin_unlock(&inode->i_lock);
 670
 671                write_chunk = writeback_chunk_size(wb->bdi, work);
 672                wbc.nr_to_write = write_chunk;
 673                wbc.pages_skipped = 0;
 674
 675                /*
 676                 * We use I_SYNC to pin the inode in memory. While it is set
 677                 * evict_inode() will wait so the inode cannot be freed.
 678                 */
 679                __writeback_single_inode(inode, wb, &wbc);
 680
 681                work->nr_pages -= write_chunk - wbc.nr_to_write;
 682                wrote += write_chunk - wbc.nr_to_write;
 683                spin_lock(&wb->list_lock);
 684                spin_lock(&inode->i_lock);
 685                if (!(inode->i_state & I_DIRTY))
 686                        wrote++;
 687                requeue_inode(inode, wb, &wbc);
 688                inode_sync_complete(inode);
 689                spin_unlock(&inode->i_lock);
 690                cond_resched_lock(&wb->list_lock);
 691                /*
 692                 * bail out to wb_writeback() often enough to check
 693                 * background threshold and other termination conditions.
 694                 */
 695                if (wrote) {
 696                        if (time_is_before_jiffies(start_time + HZ / 10UL))
 697                                break;
 698                        if (work->nr_pages <= 0)
 699                                break;
 700                }
 701        }
 702        return wrote;
 703}
 704
 705static long __writeback_inodes_wb(struct bdi_writeback *wb,
 706                                  struct wb_writeback_work *work)
 707{
 708        unsigned long start_time = jiffies;
 709        long wrote = 0;
 710
 711        while (!list_empty(&wb->b_io)) {
 712                struct inode *inode = wb_inode(wb->b_io.prev);
 713                struct super_block *sb = inode->i_sb;
 714
 715                if (!grab_super_passive(sb)) {
 716                        /*
 717                         * grab_super_passive() may fail consistently due to
 718                         * s_umount being grabbed by someone else. Don't use
 719                         * requeue_io() to avoid busy retrying the inode/sb.
 720                         */
 721                        redirty_tail(inode, wb);
 722                        continue;
 723                }
 724                wrote += writeback_sb_inodes(sb, wb, work);
 725                drop_super(sb);
 726
 727                /* refer to the same tests at the end of writeback_sb_inodes */
 728                if (wrote) {
 729                        if (time_is_before_jiffies(start_time + HZ / 10UL))
 730                                break;
 731                        if (work->nr_pages <= 0)
 732                                break;
 733                }
 734        }
 735        /* Leave any unwritten inodes on b_io */
 736        return wrote;
 737}
 738
 739long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages,
 740                                enum wb_reason reason)
 741{
 742        struct wb_writeback_work work = {
 743                .nr_pages       = nr_pages,
 744                .sync_mode      = WB_SYNC_NONE,
 745                .range_cyclic   = 1,
 746                .reason         = reason,
 747        };
 748
 749        spin_lock(&wb->list_lock);
 750        if (list_empty(&wb->b_io))
 751                queue_io(wb, &work);
 752        __writeback_inodes_wb(wb, &work);
 753        spin_unlock(&wb->list_lock);
 754
 755        return nr_pages - work.nr_pages;
 756}
 757
 758static bool over_bground_thresh(struct backing_dev_info *bdi)
 759{
 760        unsigned long background_thresh, dirty_thresh;
 761
 762        global_dirty_limits(&background_thresh, &dirty_thresh);
 763
 764        if (global_page_state(NR_FILE_DIRTY) +
 765            global_page_state(NR_UNSTABLE_NFS) > background_thresh)
 766                return true;
 767
 768        if (bdi_stat(bdi, BDI_RECLAIMABLE) >
 769                                bdi_dirty_limit(bdi, background_thresh))
 770                return true;
 771
 772        return false;
 773}
 774
 775/*
 776 * Called under wb->list_lock. If there are multiple wb per bdi,
 777 * only the flusher working on the first wb should do it.
 778 */
 779static void wb_update_bandwidth(struct bdi_writeback *wb,
 780                                unsigned long start_time)
 781{
 782        __bdi_update_bandwidth(wb->bdi, 0, 0, 0, 0, 0, start_time);
 783}
 784
 785/*
 786 * Explicit flushing or periodic writeback of "old" data.
 787 *
 788 * Define "old": the first time one of an inode's pages is dirtied, we mark the
 789 * dirtying-time in the inode's address_space.  So this periodic writeback code
 790 * just walks the superblock inode list, writing back any inodes which are
 791 * older than a specific point in time.
 792 *
 793 * Try to run once per dirty_writeback_interval.  But if a writeback event
 794 * takes longer than a dirty_writeback_interval interval, then leave a
 795 * one-second gap.
 796 *
 797 * older_than_this takes precedence over nr_to_write.  So we'll only write back
 798 * all dirty pages if they are all attached to "old" mappings.
 799 */
 800static long wb_writeback(struct bdi_writeback *wb,
 801                         struct wb_writeback_work *work)
 802{
 803        unsigned long wb_start = jiffies;
 804        long nr_pages = work->nr_pages;
 805        unsigned long oldest_jif;
 806        struct inode *inode;
 807        long progress;
 808
 809        oldest_jif = jiffies;
 810        work->older_than_this = &oldest_jif;
 811
 812        spin_lock(&wb->list_lock);
 813        for (;;) {
 814                /*
 815                 * Stop writeback when nr_pages has been consumed
 816                 */
 817                if (work->nr_pages <= 0)
 818                        break;
 819
 820                /*
 821                 * Background writeout and kupdate-style writeback may
 822                 * run forever. Stop them if there is other work to do
 823                 * so that e.g. sync can proceed. They'll be restarted
 824                 * after the other works are all done.
 825                 */
 826                if ((work->for_background || work->for_kupdate) &&
 827                    !list_empty(&wb->bdi->work_list))
 828                        break;
 829
 830                /*
 831                 * For background writeout, stop when we are below the
 832                 * background dirty threshold
 833                 */
 834                if (work->for_background && !over_bground_thresh(wb->bdi))
 835                        break;
 836
 837                /*
 838                 * Kupdate and background works are special and we want to
 839                 * include all inodes that need writing. Livelock avoidance is
 840                 * handled by these works yielding to any other work so we are
 841                 * safe.
 842                 */
 843                if (work->for_kupdate) {
 844                        oldest_jif = jiffies -
 845                                msecs_to_jiffies(dirty_expire_interval * 10);
 846                } else if (work->for_background)
 847                        oldest_jif = jiffies;
 848
 849                trace_writeback_start(wb->bdi, work);
 850                if (list_empty(&wb->b_io))
 851                        queue_io(wb, work);
 852                if (work->sb)
 853                        progress = writeback_sb_inodes(work->sb, wb, work);
 854                else
 855                        progress = __writeback_inodes_wb(wb, work);
 856                trace_writeback_written(wb->bdi, work);
 857
 858                wb_update_bandwidth(wb, wb_start);
 859
 860                /*
 861                 * Did we write something? Try for more
 862                 *
 863                 * Dirty inodes are moved to b_io for writeback in batches.
 864                 * The completion of the current batch does not necessarily
 865                 * mean the overall work is done. So we keep looping as long
 866                 * as made some progress on cleaning pages or inodes.
 867                 */
 868                if (progress)
 869                        continue;
 870                /*
 871                 * No more inodes for IO, bail
 872                 */
 873                if (list_empty(&wb->b_more_io))
 874                        break;
 875                /*
 876                 * Nothing written. Wait for some inode to
 877                 * become available for writeback. Otherwise
 878                 * we'll just busyloop.
 879                 */
 880                if (!list_empty(&wb->b_more_io))  {
 881                        trace_writeback_wait(wb->bdi, work);
 882                        inode = wb_inode(wb->b_more_io.prev);
 883                        spin_lock(&inode->i_lock);
 884                        spin_unlock(&wb->list_lock);
 885                        /* This function drops i_lock... */
 886                        inode_sleep_on_writeback(inode);
 887                        spin_lock(&wb->list_lock);
 888                }
 889        }
 890        spin_unlock(&wb->list_lock);
 891
 892        return nr_pages - work->nr_pages;
 893}
 894
 895/*
 896 * Return the next wb_writeback_work struct that hasn't been processed yet.
 897 */
 898static struct wb_writeback_work *
 899get_next_work_item(struct backing_dev_info *bdi)
 900{
 901        struct wb_writeback_work *work = NULL;
 902
 903        spin_lock_bh(&bdi->wb_lock);
 904        if (!list_empty(&bdi->work_list)) {
 905                work = list_entry(bdi->work_list.next,
 906                                  struct wb_writeback_work, list);
 907                list_del_init(&work->list);
 908        }
 909        spin_unlock_bh(&bdi->wb_lock);
 910        return work;
 911}
 912
 913/*
 914 * Add in the number of potentially dirty inodes, because each inode
 915 * write can dirty pagecache in the underlying blockdev.
 916 */
 917static unsigned long get_nr_dirty_pages(void)
 918{
 919        return global_page_state(NR_FILE_DIRTY) +
 920                global_page_state(NR_UNSTABLE_NFS) +
 921                get_nr_dirty_inodes();
 922}
 923
 924static long wb_check_background_flush(struct bdi_writeback *wb)
 925{
 926        if (over_bground_thresh(wb->bdi)) {
 927
 928                struct wb_writeback_work work = {
 929                        .nr_pages       = LONG_MAX,
 930                        .sync_mode      = WB_SYNC_NONE,
 931                        .for_background = 1,
 932                        .range_cyclic   = 1,
 933                        .reason         = WB_REASON_BACKGROUND,
 934                };
 935
 936                return wb_writeback(wb, &work);
 937        }
 938
 939        return 0;
 940}
 941
 942static long wb_check_old_data_flush(struct bdi_writeback *wb)
 943{
 944        unsigned long expired;
 945        long nr_pages;
 946
 947        /*
 948         * When set to zero, disable periodic writeback
 949         */
 950        if (!dirty_writeback_interval)
 951                return 0;
 952
 953        expired = wb->last_old_flush +
 954                        msecs_to_jiffies(dirty_writeback_interval * 10);
 955        if (time_before(jiffies, expired))
 956                return 0;
 957
 958        wb->last_old_flush = jiffies;
 959        nr_pages = get_nr_dirty_pages();
 960
 961        if (nr_pages) {
 962                struct wb_writeback_work work = {
 963                        .nr_pages       = nr_pages,
 964                        .sync_mode      = WB_SYNC_NONE,
 965                        .for_kupdate    = 1,
 966                        .range_cyclic   = 1,
 967                        .reason         = WB_REASON_PERIODIC,
 968                };
 969
 970                return wb_writeback(wb, &work);
 971        }
 972
 973        return 0;
 974}
 975
 976/*
 977 * Retrieve work items and do the writeback they describe
 978 */
 979long wb_do_writeback(struct bdi_writeback *wb, int force_wait)
 980{
 981        struct backing_dev_info *bdi = wb->bdi;
 982        struct wb_writeback_work *work;
 983        long wrote = 0;
 984
 985        set_bit(BDI_writeback_running, &wb->bdi->state);
 986        while ((work = get_next_work_item(bdi)) != NULL) {
 987                /*
 988                 * Override sync mode, in case we must wait for completion
 989                 * because this thread is exiting now.
 990                 */
 991                if (force_wait)
 992                        work->sync_mode = WB_SYNC_ALL;
 993
 994                trace_writeback_exec(bdi, work);
 995
 996                wrote += wb_writeback(wb, work);
 997
 998                /*
 999                 * Notify the caller of completion if this is a synchronous
1000                 * work item, otherwise just free it.
1001                 */
1002                if (work->done)
1003                        complete(work->done);
1004                else
1005                        kfree(work);
1006        }
1007
1008        /*
1009         * Check for periodic writeback, kupdated() style
1010         */
1011        wrote += wb_check_old_data_flush(wb);
1012        wrote += wb_check_background_flush(wb);
1013        clear_bit(BDI_writeback_running, &wb->bdi->state);
1014
1015        return wrote;
1016}
1017
1018/*
1019 * Handle writeback of dirty data for the device backed by this bdi. Also
1020 * wakes up periodically and does kupdated style flushing.
1021 */
1022int bdi_writeback_thread(void *data)
1023{
1024        struct bdi_writeback *wb = data;
1025        struct backing_dev_info *bdi = wb->bdi;
1026        long pages_written;
1027
1028        current->flags |= PF_SWAPWRITE;
1029        set_freezable();
1030        wb->last_active = jiffies;
1031
1032        /*
1033         * Our parent may run at a different priority, just set us to normal
1034         */
1035        set_user_nice(current, 0);
1036
1037        trace_writeback_thread_start(bdi);
1038
1039        while (!kthread_freezable_should_stop(NULL)) {
1040                /*
1041                 * Remove own delayed wake-up timer, since we are already awake
1042                 * and we'll take care of the preriodic write-back.
1043                 */
1044                del_timer(&wb->wakeup_timer);
1045
1046                pages_written = wb_do_writeback(wb, 0);
1047
1048                trace_writeback_pages_written(pages_written);
1049
1050                if (pages_written)
1051                        wb->last_active = jiffies;
1052
1053                set_current_state(TASK_INTERRUPTIBLE);
1054                if (!list_empty(&bdi->work_list) || kthread_should_stop()) {
1055                        __set_current_state(TASK_RUNNING);
1056                        continue;
1057                }
1058
1059                if (wb_has_dirty_io(wb) && dirty_writeback_interval)
1060                        schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10));
1061                else {
1062                        /*
1063                         * We have nothing to do, so can go sleep without any
1064                         * timeout and save power. When a work is queued or
1065                         * something is made dirty - we will be woken up.
1066                         */
1067                        schedule();
1068                }
1069        }
1070
1071        /* Flush any work that raced with us exiting */
1072        if (!list_empty(&bdi->work_list))
1073                wb_do_writeback(wb, 1);
1074
1075        trace_writeback_thread_stop(bdi);
1076        return 0;
1077}
1078
1079
1080/*
1081 * Start writeback of `nr_pages' pages.  If `nr_pages' is zero, write back
1082 * the whole world.
1083 */
1084void wakeup_flusher_threads(long nr_pages, enum wb_reason reason)
1085{
1086        struct backing_dev_info *bdi;
1087
1088        if (!nr_pages) {
1089                nr_pages = global_page_state(NR_FILE_DIRTY) +
1090                                global_page_state(NR_UNSTABLE_NFS);
1091        }
1092
1093        rcu_read_lock();
1094        list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) {
1095                if (!bdi_has_dirty_io(bdi))
1096                        continue;
1097                __bdi_start_writeback(bdi, nr_pages, false, reason);
1098        }
1099        rcu_read_unlock();
1100}
1101
1102static noinline void block_dump___mark_inode_dirty(struct inode *inode)
1103{
1104        if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) {
1105                struct dentry *dentry;
1106                const char *name = "?";
1107
1108                dentry = d_find_alias(inode);
1109                if (dentry) {
1110                        spin_lock(&dentry->d_lock);
1111                        name = (const char *) dentry->d_name.name;
1112                }
1113                printk(KERN_DEBUG
1114                       "%s(%d): dirtied inode %lu (%s) on %s\n",
1115                       current->comm, task_pid_nr(current), inode->i_ino,
1116                       name, inode->i_sb->s_id);
1117                if (dentry) {
1118                        spin_unlock(&dentry->d_lock);
1119                        dput(dentry);
1120                }
1121        }
1122}
1123
1124/**
1125 *      __mark_inode_dirty -    internal function
1126 *      @inode: inode to mark
1127 *      @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
1128 *      Mark an inode as dirty. Callers should use mark_inode_dirty or
1129 *      mark_inode_dirty_sync.
1130 *
1131 * Put the inode on the super block's dirty list.
1132 *
1133 * CAREFUL! We mark it dirty unconditionally, but move it onto the
1134 * dirty list only if it is hashed or if it refers to a blockdev.
1135 * If it was not hashed, it will never be added to the dirty list
1136 * even if it is later hashed, as it will have been marked dirty already.
1137 *
1138 * In short, make sure you hash any inodes _before_ you start marking
1139 * them dirty.
1140 *
1141 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
1142 * the block-special inode (/dev/hda1) itself.  And the ->dirtied_when field of
1143 * the kernel-internal blockdev inode represents the dirtying time of the
1144 * blockdev's pages.  This is why for I_DIRTY_PAGES we always use
1145 * page->mapping->host, so the page-dirtying time is recorded in the internal
1146 * blockdev inode.
1147 */
1148void __mark_inode_dirty(struct inode *inode, int flags)
1149{
1150        struct super_block *sb = inode->i_sb;
1151        struct backing_dev_info *bdi = NULL;
1152
1153        /*
1154         * Don't do this for I_DIRTY_PAGES - that doesn't actually
1155         * dirty the inode itself
1156         */
1157        if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
1158                if (sb->s_op->dirty_inode)
1159                        sb->s_op->dirty_inode(inode, flags);
1160        }
1161
1162        /*
1163         * make sure that changes are seen by all cpus before we test i_state
1164         * -- mikulas
1165         */
1166        smp_mb();
1167
1168        /* avoid the locking if we can */
1169        if ((inode->i_state & flags) == flags)
1170                return;
1171
1172        if (unlikely(block_dump))
1173                block_dump___mark_inode_dirty(inode);
1174
1175        spin_lock(&inode->i_lock);
1176        if ((inode->i_state & flags) != flags) {
1177                const int was_dirty = inode->i_state & I_DIRTY;
1178
1179                inode->i_state |= flags;
1180
1181                /*
1182                 * If the inode is being synced, just update its dirty state.
1183                 * The unlocker will place the inode on the appropriate
1184                 * superblock list, based upon its state.
1185                 */
1186                if (inode->i_state & I_SYNC)
1187                        goto out_unlock_inode;
1188
1189                /*
1190                 * Only add valid (hashed) inodes to the superblock's
1191                 * dirty list.  Add blockdev inodes as well.
1192                 */
1193                if (!S_ISBLK(inode->i_mode)) {
1194                        if (inode_unhashed(inode))
1195                                goto out_unlock_inode;
1196                }
1197                if (inode->i_state & I_FREEING)
1198                        goto out_unlock_inode;
1199
1200                /*
1201                 * If the inode was already on b_dirty/b_io/b_more_io, don't
1202                 * reposition it (that would break b_dirty time-ordering).
1203                 */
1204                if (!was_dirty) {
1205                        bool wakeup_bdi = false;
1206                        bdi = inode_to_bdi(inode);
1207
1208                        if (bdi_cap_writeback_dirty(bdi)) {
1209                                WARN(!test_bit(BDI_registered, &bdi->state),
1210                                     "bdi-%s not registered\n", bdi->name);
1211
1212                                /*
1213                                 * If this is the first dirty inode for this
1214                                 * bdi, we have to wake-up the corresponding
1215                                 * bdi thread to make sure background
1216                                 * write-back happens later.
1217                                 */
1218                                if (!wb_has_dirty_io(&bdi->wb))
1219                                        wakeup_bdi = true;
1220                        }
1221
1222                        spin_unlock(&inode->i_lock);
1223                        spin_lock(&bdi->wb.list_lock);
1224                        inode->dirtied_when = jiffies;
1225                        list_move(&inode->i_wb_list, &bdi->wb.b_dirty);
1226                        spin_unlock(&bdi->wb.list_lock);
1227
1228                        if (wakeup_bdi)
1229                                bdi_wakeup_thread_delayed(bdi);
1230                        return;
1231                }
1232        }
1233out_unlock_inode:
1234        spin_unlock(&inode->i_lock);
1235
1236}
1237EXPORT_SYMBOL(__mark_inode_dirty);
1238
1239static void wait_sb_inodes(struct super_block *sb)
1240{
1241        struct inode *inode, *old_inode = NULL;
1242
1243        /*
1244         * We need to be protected against the filesystem going from
1245         * r/o to r/w or vice versa.
1246         */
1247        WARN_ON(!rwsem_is_locked(&sb->s_umount));
1248
1249        spin_lock(&inode_sb_list_lock);
1250
1251        /*
1252         * Data integrity sync. Must wait for all pages under writeback,
1253         * because there may have been pages dirtied before our sync
1254         * call, but which had writeout started before we write it out.
1255         * In which case, the inode may not be on the dirty list, but
1256         * we still have to wait for that writeout.
1257         */
1258        list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1259                struct address_space *mapping = inode->i_mapping;
1260
1261                spin_lock(&inode->i_lock);
1262                if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
1263                    (mapping->nrpages == 0)) {
1264                        spin_unlock(&inode->i_lock);
1265                        continue;
1266                }
1267                __iget(inode);
1268                spin_unlock(&inode->i_lock);
1269                spin_unlock(&inode_sb_list_lock);
1270
1271                /*
1272                 * We hold a reference to 'inode' so it couldn't have been
1273                 * removed from s_inodes list while we dropped the
1274                 * inode_sb_list_lock.  We cannot iput the inode now as we can
1275                 * be holding the last reference and we cannot iput it under
1276                 * inode_sb_list_lock. So we keep the reference and iput it
1277                 * later.
1278                 */
1279                iput(old_inode);
1280                old_inode = inode;
1281
1282                filemap_fdatawait(mapping);
1283
1284                cond_resched();
1285
1286                spin_lock(&inode_sb_list_lock);
1287        }
1288        spin_unlock(&inode_sb_list_lock);
1289        iput(old_inode);
1290}
1291
1292/**
1293 * writeback_inodes_sb_nr -     writeback dirty inodes from given super_block
1294 * @sb: the superblock
1295 * @nr: the number of pages to write
1296 * @reason: reason why some writeback work initiated
1297 *
1298 * Start writeback on some inodes on this super_block. No guarantees are made
1299 * on how many (if any) will be written, and this function does not wait
1300 * for IO completion of submitted IO.
1301 */
1302void writeback_inodes_sb_nr(struct super_block *sb,
1303                            unsigned long nr,
1304                            enum wb_reason reason)
1305{
1306        DECLARE_COMPLETION_ONSTACK(done);
1307        struct wb_writeback_work work = {
1308                .sb                     = sb,
1309                .sync_mode              = WB_SYNC_NONE,
1310                .tagged_writepages      = 1,
1311                .done                   = &done,
1312                .nr_pages               = nr,
1313                .reason                 = reason,
1314        };
1315
1316        if (sb->s_bdi == &noop_backing_dev_info)
1317                return;
1318        WARN_ON(!rwsem_is_locked(&sb->s_umount));
1319        bdi_queue_work(sb->s_bdi, &work);
1320        wait_for_completion(&done);
1321}
1322EXPORT_SYMBOL(writeback_inodes_sb_nr);
1323
1324/**
1325 * writeback_inodes_sb  -       writeback dirty inodes from given super_block
1326 * @sb: the superblock
1327 * @reason: reason why some writeback work was initiated
1328 *
1329 * Start writeback on some inodes on this super_block. No guarantees are made
1330 * on how many (if any) will be written, and this function does not wait
1331 * for IO completion of submitted IO.
1332 */
1333void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason)
1334{
1335        return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason);
1336}
1337EXPORT_SYMBOL(writeback_inodes_sb);
1338
1339/**
1340 * writeback_inodes_sb_if_idle  -       start writeback if none underway
1341 * @sb: the superblock
1342 * @reason: reason why some writeback work was initiated
1343 *
1344 * Invoke writeback_inodes_sb if no writeback is currently underway.
1345 * Returns 1 if writeback was started, 0 if not.
1346 */
1347int writeback_inodes_sb_if_idle(struct super_block *sb, enum wb_reason reason)
1348{
1349        if (!writeback_in_progress(sb->s_bdi)) {
1350                down_read(&sb->s_umount);
1351                writeback_inodes_sb(sb, reason);
1352                up_read(&sb->s_umount);
1353                return 1;
1354        } else
1355                return 0;
1356}
1357EXPORT_SYMBOL(writeback_inodes_sb_if_idle);
1358
1359/**
1360 * writeback_inodes_sb_nr_if_idle       -       start writeback if none underway
1361 * @sb: the superblock
1362 * @nr: the number of pages to write
1363 * @reason: reason why some writeback work was initiated
1364 *
1365 * Invoke writeback_inodes_sb if no writeback is currently underway.
1366 * Returns 1 if writeback was started, 0 if not.
1367 */
1368int writeback_inodes_sb_nr_if_idle(struct super_block *sb,
1369                                   unsigned long nr,
1370                                   enum wb_reason reason)
1371{
1372        if (!writeback_in_progress(sb->s_bdi)) {
1373                down_read(&sb->s_umount);
1374                writeback_inodes_sb_nr(sb, nr, reason);
1375                up_read(&sb->s_umount);
1376                return 1;
1377        } else
1378                return 0;
1379}
1380EXPORT_SYMBOL(writeback_inodes_sb_nr_if_idle);
1381
1382/**
1383 * sync_inodes_sb       -       sync sb inode pages
1384 * @sb: the superblock
1385 *
1386 * This function writes and waits on any dirty inode belonging to this
1387 * super_block.
1388 */
1389void sync_inodes_sb(struct super_block *sb)
1390{
1391        DECLARE_COMPLETION_ONSTACK(done);
1392        struct wb_writeback_work work = {
1393                .sb             = sb,
1394                .sync_mode      = WB_SYNC_ALL,
1395                .nr_pages       = LONG_MAX,
1396                .range_cyclic   = 0,
1397                .done           = &done,
1398                .reason         = WB_REASON_SYNC,
1399        };
1400
1401        /* Nothing to do? */
1402        if (sb->s_bdi == &noop_backing_dev_info)
1403                return;
1404        WARN_ON(!rwsem_is_locked(&sb->s_umount));
1405
1406        bdi_queue_work(sb->s_bdi, &work);
1407        wait_for_completion(&done);
1408
1409        wait_sb_inodes(sb);
1410}
1411EXPORT_SYMBOL(sync_inodes_sb);
1412
1413/**
1414 * write_inode_now      -       write an inode to disk
1415 * @inode: inode to write to disk
1416 * @sync: whether the write should be synchronous or not
1417 *
1418 * This function commits an inode to disk immediately if it is dirty. This is
1419 * primarily needed by knfsd.
1420 *
1421 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
1422 */
1423int write_inode_now(struct inode *inode, int sync)
1424{
1425        struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
1426        struct writeback_control wbc = {
1427                .nr_to_write = LONG_MAX,
1428                .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE,
1429                .range_start = 0,
1430                .range_end = LLONG_MAX,
1431        };
1432
1433        if (!mapping_cap_writeback_dirty(inode->i_mapping))
1434                wbc.nr_to_write = 0;
1435
1436        might_sleep();
1437        return writeback_single_inode(inode, wb, &wbc);
1438}
1439EXPORT_SYMBOL(write_inode_now);
1440
1441/**
1442 * sync_inode - write an inode and its pages to disk.
1443 * @inode: the inode to sync
1444 * @wbc: controls the writeback mode
1445 *
1446 * sync_inode() will write an inode and its pages to disk.  It will also
1447 * correctly update the inode on its superblock's dirty inode lists and will
1448 * update inode->i_state.
1449 *
1450 * The caller must have a ref on the inode.
1451 */
1452int sync_inode(struct inode *inode, struct writeback_control *wbc)
1453{
1454        return writeback_single_inode(inode, &inode_to_bdi(inode)->wb, wbc);
1455}
1456EXPORT_SYMBOL(sync_inode);
1457
1458/**
1459 * sync_inode_metadata - write an inode to disk
1460 * @inode: the inode to sync
1461 * @wait: wait for I/O to complete.
1462 *
1463 * Write an inode to disk and adjust its dirty state after completion.
1464 *
1465 * Note: only writes the actual inode, no associated data or other metadata.
1466 */
1467int sync_inode_metadata(struct inode *inode, int wait)
1468{
1469        struct writeback_control wbc = {
1470                .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE,
1471                .nr_to_write = 0, /* metadata-only */
1472        };
1473
1474        return sync_inode(inode, &wbc);
1475}
1476EXPORT_SYMBOL(sync_inode_metadata);
1477
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.