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