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/writeback.h>
  26#include <linux/blkdev.h>
  27#include <linux/backing-dev.h>
  28#include <linux/tracepoint.h>
  29#include "internal.h"
  30
  31/*
  32 * 4MB minimal write chunk size
  33 */
  34#define MIN_WRITEBACK_PAGES     (4096UL >> (PAGE_CACHE_SHIFT - 10))
  35
  36/*
  37 * Passed into wb_writeback(), essentially a subset of writeback_control
  38 */
  39struct wb_writeback_work {
  40        long nr_pages;
  41        struct super_block *sb;
  42        unsigned long *older_than_this;
  43        enum writeback_sync_modes sync_mode;
  44        unsigned int tagged_writepages:1;
  45        unsigned int for_kupdate:1;
  46        unsigned int range_cyclic:1;
  47        unsigned int for_background:1;
  48        unsigned int for_sync:1;        /* sync(2) WB_SYNC_ALL writeback */
  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
  91static void bdi_queue_work(struct backing_dev_info *bdi,
  92                           struct wb_writeback_work *work)
  93{
  94        trace_writeback_queue(bdi, work);
  95
  96        spin_lock_bh(&bdi->wb_lock);
  97        list_add_tail(&work->list, &bdi->work_list);
  98        spin_unlock_bh(&bdi->wb_lock);
  99
 100        mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0);
 101}
 102
 103static void
 104__bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages,
 105                      bool range_cyclic, enum wb_reason reason)
 106{
 107        struct wb_writeback_work *work;
 108
 109        /*
 110         * This is WB_SYNC_NONE writeback, so if allocation fails just
 111         * wakeup the thread for old dirty data writeback
 112         */
 113        work = kzalloc(sizeof(*work), GFP_ATOMIC);
 114        if (!work) {
 115                trace_writeback_nowork(bdi);
 116                mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0);
 117                return;
 118        }
 119
 120        work->sync_mode = WB_SYNC_NONE;
 121        work->nr_pages  = nr_pages;
 122        work->range_cyclic = range_cyclic;
 123        work->reason    = reason;
 124
 125        bdi_queue_work(bdi, work);
 126}
 127
 128/**
 129 * bdi_start_writeback - start writeback
 130 * @bdi: the backing device to write from
 131 * @nr_pages: the number of pages to write
 132 * @reason: reason why some writeback work was initiated
 133 *
 134 * Description:
 135 *   This does WB_SYNC_NONE opportunistic writeback. The IO is only
 136 *   started when this function returns, we make no guarantees on
 137 *   completion. Caller need not hold sb s_umount semaphore.
 138 *
 139 */
 140void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages,
 141                        enum wb_reason reason)
 142{
 143        __bdi_start_writeback(bdi, nr_pages, true, reason);
 144}
 145
 146/**
 147 * bdi_start_background_writeback - start background writeback
 148 * @bdi: the backing device to write from
 149 *
 150 * Description:
 151 *   This makes sure WB_SYNC_NONE background writeback happens. When
 152 *   this function returns, it is only guaranteed that for given BDI
 153 *   some IO is happening if we are over background dirty threshold.
 154 *   Caller need not hold sb s_umount semaphore.
 155 */
 156void bdi_start_background_writeback(struct backing_dev_info *bdi)
 157{
 158        /*
 159         * We just wake up the flusher thread. It will perform background
 160         * writeback as soon as there is no other work to do.
 161         */
 162        trace_writeback_wake_background(bdi);
 163        mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0);
 164}
 165
 166/*
 167 * Remove the inode from the writeback list it is on.
 168 */
 169void inode_wb_list_del(struct inode *inode)
 170{
 171        struct backing_dev_info *bdi = inode_to_bdi(inode);
 172
 173        spin_lock(&bdi->wb.list_lock);
 174        list_del_init(&inode->i_wb_list);
 175        spin_unlock(&bdi->wb.list_lock);
 176}
 177
 178/*
 179 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
 180 * furthest end of its superblock's dirty-inode list.
 181 *
 182 * Before stamping the inode's ->dirtied_when, we check to see whether it is
 183 * already the most-recently-dirtied inode on the b_dirty list.  If that is
 184 * the case then the inode must have been redirtied while it was being written
 185 * out and we don't reset its dirtied_when.
 186 */
 187static void redirty_tail(struct inode *inode, struct bdi_writeback *wb)
 188{
 189        assert_spin_locked(&wb->list_lock);
 190        if (!list_empty(&wb->b_dirty)) {
 191                struct inode *tail;
 192
 193                tail = wb_inode(wb->b_dirty.next);
 194                if (time_before(inode->dirtied_when, tail->dirtied_when))
 195                        inode->dirtied_when = jiffies;
 196        }
 197        list_move(&inode->i_wb_list, &wb->b_dirty);
 198}
 199
 200/*
 201 * requeue inode for re-scanning after bdi->b_io list is exhausted.
 202 */
 203static void requeue_io(struct inode *inode, struct bdi_writeback *wb)
 204{
 205        assert_spin_locked(&wb->list_lock);
 206        list_move(&inode->i_wb_list, &wb->b_more_io);
 207}
 208
 209static void inode_sync_complete(struct inode *inode)
 210{
 211        inode->i_state &= ~I_SYNC;
 212        /* If inode is clean an unused, put it into LRU now... */
 213        inode_add_lru(inode);
 214        /* Waiters must see I_SYNC cleared before being woken up */
 215        smp_mb();
 216        wake_up_bit(&inode->i_state, __I_SYNC);
 217}
 218
 219static bool inode_dirtied_after(struct inode *inode, unsigned long t)
 220{
 221        bool ret = time_after(inode->dirtied_when, t);
 222#ifndef CONFIG_64BIT
 223        /*
 224         * For inodes being constantly redirtied, dirtied_when can get stuck.
 225         * It _appears_ to be in the future, but is actually in distant past.
 226         * This test is necessary to prevent such wrapped-around relative times
 227         * from permanently stopping the whole bdi writeback.
 228         */
 229        ret = ret && time_before_eq(inode->dirtied_when, jiffies);
 230#endif
 231        return ret;
 232}
 233
 234/*
 235 * Move expired (dirtied before work->older_than_this) dirty inodes from
 236 * @delaying_queue to @dispatch_queue.
 237 */
 238static int move_expired_inodes(struct list_head *delaying_queue,
 239                               struct list_head *dispatch_queue,
 240                               struct wb_writeback_work *work)
 241{
 242        LIST_HEAD(tmp);
 243        struct list_head *pos, *node;
 244        struct super_block *sb = NULL;
 245        struct inode *inode;
 246        int do_sb_sort = 0;
 247        int moved = 0;
 248
 249        while (!list_empty(delaying_queue)) {
 250                inode = wb_inode(delaying_queue->prev);
 251                if (work->older_than_this &&
 252                    inode_dirtied_after(inode, *work->older_than_this))
 253                        break;
 254                if (sb && sb != inode->i_sb)
 255                        do_sb_sort = 1;
 256                sb = inode->i_sb;
 257                list_move(&inode->i_wb_list, &tmp);
 258                moved++;
 259        }
 260
 261        /* just one sb in list, splice to dispatch_queue and we're done */
 262        if (!do_sb_sort) {
 263                list_splice(&tmp, dispatch_queue);
 264                goto out;
 265        }
 266
 267        /* Move inodes from one superblock together */
 268        while (!list_empty(&tmp)) {
 269                sb = wb_inode(tmp.prev)->i_sb;
 270                list_for_each_prev_safe(pos, node, &tmp) {
 271                        inode = wb_inode(pos);
 272                        if (inode->i_sb == sb)
 273                                list_move(&inode->i_wb_list, dispatch_queue);
 274                }
 275        }
 276out:
 277        return moved;
 278}
 279
 280/*
 281 * Queue all expired dirty inodes for io, eldest first.
 282 * Before
 283 *         newly dirtied     b_dirty    b_io    b_more_io
 284 *         =============>    gf         edc     BA
 285 * After
 286 *         newly dirtied     b_dirty    b_io    b_more_io
 287 *         =============>    g          fBAedc
 288 *                                           |
 289 *                                           +--> dequeue for IO
 290 */
 291static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work)
 292{
 293        int moved;
 294        assert_spin_locked(&wb->list_lock);
 295        list_splice_init(&wb->b_more_io, &wb->b_io);
 296        moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, work);
 297        trace_writeback_queue_io(wb, work, moved);
 298}
 299
 300static int write_inode(struct inode *inode, struct writeback_control *wbc)
 301{
 302        int ret;
 303
 304        if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) {
 305                trace_writeback_write_inode_start(inode, wbc);
 306                ret = inode->i_sb->s_op->write_inode(inode, wbc);
 307                trace_writeback_write_inode(inode, wbc);
 308                return ret;
 309        }
 310        return 0;
 311}
 312
 313/*
 314 * Wait for writeback on an inode to complete. Called with i_lock held.
 315 * Caller must make sure inode cannot go away when we drop i_lock.
 316 */
 317static void __inode_wait_for_writeback(struct inode *inode)
 318        __releases(inode->i_lock)
 319        __acquires(inode->i_lock)
 320{
 321        DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC);
 322        wait_queue_head_t *wqh;
 323
 324        wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
 325        while (inode->i_state & I_SYNC) {
 326                spin_unlock(&inode->i_lock);
 327                __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE);
 328                spin_lock(&inode->i_lock);
 329        }
 330}
 331
 332/*
 333 * Wait for writeback on an inode to complete. Caller must have inode pinned.
 334 */
 335void inode_wait_for_writeback(struct inode *inode)
 336{
 337        spin_lock(&inode->i_lock);
 338        __inode_wait_for_writeback(inode);
 339        spin_unlock(&inode->i_lock);
 340}
 341
 342/*
 343 * Sleep until I_SYNC is cleared. This function must be called with i_lock
 344 * held and drops it. It is aimed for callers not holding any inode reference
 345 * so once i_lock is dropped, inode can go away.
 346 */
 347static void inode_sleep_on_writeback(struct inode *inode)
 348        __releases(inode->i_lock)
 349{
 350        DEFINE_WAIT(wait);
 351        wait_queue_head_t *wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
 352        int sleep;
 353
 354        prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
 355        sleep = inode->i_state & I_SYNC;
 356        spin_unlock(&inode->i_lock);
 357        if (sleep)
 358                schedule();
 359        finish_wait(wqh, &wait);
 360}
 361
 362/*
 363 * Find proper writeback list for the inode depending on its current state and
 364 * possibly also change of its state while we were doing writeback.  Here we
 365 * handle things such as livelock prevention or fairness of writeback among
 366 * inodes. This function can be called only by flusher thread - noone else
 367 * processes all inodes in writeback lists and requeueing inodes behind flusher
 368 * thread's back can have unexpected consequences.
 369 */
 370static void requeue_inode(struct inode *inode, struct bdi_writeback *wb,
 371                          struct writeback_control *wbc)
 372{
 373        if (inode->i_state & I_FREEING)
 374                return;
 375
 376        /*
 377         * Sync livelock prevention. Each inode is tagged and synced in one
 378         * shot. If still dirty, it will be redirty_tail()'ed below.  Update
 379         * the dirty time to prevent enqueue and sync it again.
 380         */
 381        if ((inode->i_state & I_DIRTY) &&
 382            (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages))
 383                inode->dirtied_when = jiffies;
 384
 385        if (wbc->pages_skipped) {
 386                /*
 387                 * writeback is not making progress due to locked
 388                 * buffers. Skip this inode for now.
 389                 */
 390                redirty_tail(inode, wb);
 391                return;
 392        }
 393
 394        if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) {
 395                /*
 396                 * We didn't write back all the pages.  nfs_writepages()
 397                 * sometimes bales out without doing anything.
 398                 */
 399                if (wbc->nr_to_write <= 0) {
 400                        /* Slice used up. Queue for next turn. */
 401                        requeue_io(inode, wb);
 402                } else {
 403                        /*
 404                         * Writeback blocked by something other than
 405                         * congestion. Delay the inode for some time to
 406                         * avoid spinning on the CPU (100% iowait)
 407                         * retrying writeback of the dirty page/inode
 408                         * that cannot be performed immediately.
 409                         */
 410                        redirty_tail(inode, wb);
 411                }
 412        } else if (inode->i_state & I_DIRTY) {
 413                /*
 414                 * Filesystems can dirty the inode during writeback operations,
 415                 * such as delayed allocation during submission or metadata
 416                 * updates after data IO completion.
 417                 */
 418                redirty_tail(inode, wb);
 419        } else {
 420                /* The inode is clean. Remove from writeback lists. */
 421                list_del_init(&inode->i_wb_list);
 422        }
 423}
 424
 425/*
 426 * Write out an inode and its dirty pages. Do not update the writeback list
 427 * linkage. That is left to the caller. The caller is also responsible for
 428 * setting I_SYNC flag and calling inode_sync_complete() to clear it.
 429 */
 430static int
 431__writeback_single_inode(struct inode *inode, struct writeback_control *wbc)
 432{
 433        struct address_space *mapping = inode->i_mapping;
 434        long nr_to_write = wbc->nr_to_write;
 435        unsigned dirty;
 436        int ret;
 437
 438        WARN_ON(!(inode->i_state & I_SYNC));
 439
 440        trace_writeback_single_inode_start(inode, wbc, nr_to_write);
 441
 442        ret = do_writepages(mapping, wbc);
 443
 444        /*
 445         * Make sure to wait on the data before writing out the metadata.
 446         * This is important for filesystems that modify metadata on data
 447         * I/O completion. We don't do it for sync(2) writeback because it has a
 448         * separate, external IO completion path and ->sync_fs for guaranteeing
 449         * inode metadata is written back correctly.
 450         */
 451        if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) {
 452                int err = filemap_fdatawait(mapping);
 453                if (ret == 0)
 454                        ret = err;
 455        }
 456
 457        /*
 458         * Some filesystems may redirty the inode during the writeback
 459         * due to delalloc, clear dirty metadata flags right before
 460         * write_inode()
 461         */
 462        spin_lock(&inode->i_lock);
 463        /* Clear I_DIRTY_PAGES if we've written out all dirty pages */
 464        if (!mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
 465                inode->i_state &= ~I_DIRTY_PAGES;
 466        dirty = inode->i_state & I_DIRTY;
 467        inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC);
 468        spin_unlock(&inode->i_lock);
 469        /* Don't write the inode if only I_DIRTY_PAGES was set */
 470        if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
 471                int err = write_inode(inode, wbc);
 472                if (ret == 0)
 473                        ret = err;
 474        }
 475        trace_writeback_single_inode(inode, wbc, nr_to_write);
 476        return ret;
 477}
 478
 479/*
 480 * Write out an inode's dirty pages. Either the caller has an active reference
 481 * on the inode or the inode has I_WILL_FREE set.
 482 *
 483 * This function is designed to be called for writing back one inode which
 484 * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode()
 485 * and does more profound writeback list handling in writeback_sb_inodes().
 486 */
 487static int
 488writeback_single_inode(struct inode *inode, struct bdi_writeback *wb,
 489                       struct writeback_control *wbc)
 490{
 491        int ret = 0;
 492
 493        spin_lock(&inode->i_lock);
 494        if (!atomic_read(&inode->i_count))
 495                WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING)));
 496        else
 497                WARN_ON(inode->i_state & I_WILL_FREE);
 498
 499        if (inode->i_state & I_SYNC) {
 500                if (wbc->sync_mode != WB_SYNC_ALL)
 501                        goto out;
 502                /*
 503                 * It's a data-integrity sync. We must wait. Since callers hold
 504                 * inode reference or inode has I_WILL_FREE set, it cannot go
 505                 * away under us.
 506                 */
 507                __inode_wait_for_writeback(inode);
 508        }
 509        WARN_ON(inode->i_state & I_SYNC);
 510        /*
 511         * Skip inode if it is clean. We don't want to mess with writeback
 512         * lists in this function since flusher thread may be doing for example
 513         * sync in parallel and if we move the inode, it could get skipped. So
 514         * here we make sure inode is on some writeback list and leave it there
 515         * unless we have completely cleaned the inode.
 516         */
 517        if (!(inode->i_state & I_DIRTY))
 518                goto out;
 519        inode->i_state |= I_SYNC;
 520        spin_unlock(&inode->i_lock);
 521
 522        ret = __writeback_single_inode(inode, wbc);
 523
 524        spin_lock(&wb->list_lock);
 525        spin_lock(&inode->i_lock);
 526        /*
 527         * If inode is clean, remove it from writeback lists. Otherwise don't
 528         * touch it. See comment above for explanation.
 529         */
 530        if (!(inode->i_state & I_DIRTY))
 531                list_del_init(&inode->i_wb_list);
 532        spin_unlock(&wb->list_lock);
 533        inode_sync_complete(inode);
 534out:
 535        spin_unlock(&inode->i_lock);
 536        return ret;
 537}
 538
 539static long writeback_chunk_size(struct backing_dev_info *bdi,
 540                                 struct wb_writeback_work *work)
 541{
 542        long pages;
 543
 544        /*
 545         * WB_SYNC_ALL mode does livelock avoidance by syncing dirty
 546         * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX
 547         * here avoids calling into writeback_inodes_wb() more than once.
 548         *
 549         * The intended call sequence for WB_SYNC_ALL writeback is:
 550         *
 551         *      wb_writeback()
 552         *          writeback_sb_inodes()       <== called only once
 553         *              write_cache_pages()     <== called once for each inode
 554         *                   (quickly) tag currently dirty pages
 555         *                   (maybe slowly) sync all tagged pages
 556         */
 557        if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages)
 558                pages = LONG_MAX;
 559        else {
 560                pages = min(bdi->avg_write_bandwidth / 2,
 561                            global_dirty_limit / DIRTY_SCOPE);
 562                pages = min(pages, work->nr_pages);
 563                pages = round_down(pages + MIN_WRITEBACK_PAGES,
 564                                   MIN_WRITEBACK_PAGES);
 565        }
 566
 567        return pages;
 568}
 569
 570/*
 571 * Write a portion of b_io inodes which belong to @sb.
 572 *
 573 * Return the number of pages and/or inodes written.
 574 */
 575static long writeback_sb_inodes(struct super_block *sb,
 576                                struct bdi_writeback *wb,
 577                                struct wb_writeback_work *work)
 578{
 579        struct writeback_control wbc = {
 580                .sync_mode              = work->sync_mode,
 581                .tagged_writepages      = work->tagged_writepages,
 582                .for_kupdate            = work->for_kupdate,
 583                .for_background         = work->for_background,
 584                .for_sync               = work->for_sync,
 585                .range_cyclic           = work->range_cyclic,
 586                .range_start            = 0,
 587                .range_end              = LLONG_MAX,
 588        };
 589        unsigned long start_time = jiffies;
 590        long write_chunk;
 591        long wrote = 0;  /* count both pages and inodes */
 592
 593        while (!list_empty(&wb->b_io)) {
 594                struct inode *inode = wb_inode(wb->b_io.prev);
 595
 596                if (inode->i_sb != sb) {
 597                        if (work->sb) {
 598                                /*
 599                                 * We only want to write back data for this
 600                                 * superblock, move all inodes not belonging
 601                                 * to it back onto the dirty list.
 602                                 */
 603                                redirty_tail(inode, wb);
 604                                continue;
 605                        }
 606
 607                        /*
 608                         * The inode belongs to a different superblock.
 609                         * Bounce back to the caller to unpin this and
 610                         * pin the next superblock.
 611                         */
 612                        break;
 613                }
 614
 615                /*
 616                 * Don't bother with new inodes or inodes being freed, first
 617                 * kind does not need periodic writeout yet, and for the latter
 618                 * kind writeout is handled by the freer.
 619                 */
 620                spin_lock(&inode->i_lock);
 621                if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
 622                        spin_unlock(&inode->i_lock);
 623                        redirty_tail(inode, wb);
 624                        continue;
 625                }
 626                if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) {
 627                        /*
 628                         * If this inode is locked for writeback and we are not
 629                         * doing writeback-for-data-integrity, move it to
 630                         * b_more_io so that writeback can proceed with the
 631                         * other inodes on s_io.
 632                         *
 633                         * We'll have another go at writing back this inode
 634                         * when we completed a full scan of b_io.
 635                         */
 636                        spin_unlock(&inode->i_lock);
 637                        requeue_io(inode, wb);
 638                        trace_writeback_sb_inodes_requeue(inode);
 639                        continue;
 640                }
 641                spin_unlock(&wb->list_lock);
 642
 643                /*
 644                 * We already requeued the inode if it had I_SYNC set and we
 645                 * are doing WB_SYNC_NONE writeback. So this catches only the
 646                 * WB_SYNC_ALL case.
 647                 */
 648                if (inode->i_state & I_SYNC) {
 649                        /* Wait for I_SYNC. This function drops i_lock... */
 650                        inode_sleep_on_writeback(inode);
 651                        /* Inode may be gone, start again */
 652                        spin_lock(&wb->list_lock);
 653                        continue;
 654                }
 655                inode->i_state |= I_SYNC;
 656                spin_unlock(&inode->i_lock);
 657
 658                write_chunk = writeback_chunk_size(wb->bdi, work);
 659                wbc.nr_to_write = write_chunk;
 660                wbc.pages_skipped = 0;
 661
 662                /*
 663                 * We use I_SYNC to pin the inode in memory. While it is set
 664                 * evict_inode() will wait so the inode cannot be freed.
 665                 */
 666                __writeback_single_inode(inode, &wbc);
 667
 668                work->nr_pages -= write_chunk - wbc.nr_to_write;
 669                wrote += write_chunk - wbc.nr_to_write;
 670                spin_lock(&wb->list_lock);
 671                spin_lock(&inode->i_lock);
 672                if (!(inode->i_state & I_DIRTY))
 673                        wrote++;
 674                requeue_inode(inode, wb, &wbc);
 675                inode_sync_complete(inode);
 676                spin_unlock(&inode->i_lock);
 677                cond_resched_lock(&wb->list_lock);
 678                /*
 679                 * bail out to wb_writeback() often enough to check
 680                 * background threshold and other termination conditions.
 681                 */
 682                if (wrote) {
 683                        if (time_is_before_jiffies(start_time + HZ / 10UL))
 684                                break;
 685                        if (work->nr_pages <= 0)
 686                                break;
 687                }
 688        }
 689        return wrote;
 690}
 691
 692static long __writeback_inodes_wb(struct bdi_writeback *wb,
 693                                  struct wb_writeback_work *work)
 694{
 695        unsigned long start_time = jiffies;
 696        long wrote = 0;
 697
 698        while (!list_empty(&wb->b_io)) {
 699                struct inode *inode = wb_inode(wb->b_io.prev);
 700                struct super_block *sb = inode->i_sb;
 701
 702                if (!grab_super_passive(sb)) {
 703                        /*
 704                         * grab_super_passive() may fail consistently due to
 705                         * s_umount being grabbed by someone else. Don't use
 706                         * requeue_io() to avoid busy retrying the inode/sb.
 707                         */
 708                        redirty_tail(inode, wb);
 709                        continue;
 710                }
 711                wrote += writeback_sb_inodes(sb, wb, work);
 712                drop_super(sb);
 713
 714                /* refer to the same tests at the end of writeback_sb_inodes */
 715                if (wrote) {
 716                        if (time_is_before_jiffies(start_time + HZ / 10UL))
 717                                break;
 718                        if (work->nr_pages <= 0)
 719                                break;
 720                }
 721        }
 722        /* Leave any unwritten inodes on b_io */
 723        return wrote;
 724}
 725
 726long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages,
 727                                enum wb_reason reason)
 728{
 729        struct wb_writeback_work work = {
 730                .nr_pages       = nr_pages,
 731                .sync_mode      = WB_SYNC_NONE,
 732                .range_cyclic   = 1,
 733                .reason         = reason,
 734        };
 735
 736        spin_lock(&wb->list_lock);
 737        if (list_empty(&wb->b_io))
 738                queue_io(wb, &work);
 739        __writeback_inodes_wb(wb, &work);
 740        spin_unlock(&wb->list_lock);
 741
 742        return nr_pages - work.nr_pages;
 743}
 744
 745static bool over_bground_thresh(struct backing_dev_info *bdi)
 746{
 747        unsigned long background_thresh, dirty_thresh;
 748
 749        global_dirty_limits(&background_thresh, &dirty_thresh);
 750
 751        if (global_page_state(NR_FILE_DIRTY) +
 752            global_page_state(NR_UNSTABLE_NFS) > background_thresh)
 753                return true;
 754
 755        if (bdi_stat(bdi, BDI_RECLAIMABLE) >
 756                                bdi_dirty_limit(bdi, background_thresh))
 757                return true;
 758
 759        return false;
 760}
 761
 762/*
 763 * Called under wb->list_lock. If there are multiple wb per bdi,
 764 * only the flusher working on the first wb should do it.
 765 */
 766static void wb_update_bandwidth(struct bdi_writeback *wb,
 767                                unsigned long start_time)
 768{
 769        __bdi_update_bandwidth(wb->bdi, 0, 0, 0, 0, 0, start_time);
 770}
 771
 772/*
 773 * Explicit flushing or periodic writeback of "old" data.
 774 *
 775 * Define "old": the first time one of an inode's pages is dirtied, we mark the
 776 * dirtying-time in the inode's address_space.  So this periodic writeback code
 777 * just walks the superblock inode list, writing back any inodes which are
 778 * older than a specific point in time.
 779 *
 780 * Try to run once per dirty_writeback_interval.  But if a writeback event
 781 * takes longer than a dirty_writeback_interval interval, then leave a
 782 * one-second gap.
 783 *
 784 * older_than_this takes precedence over nr_to_write.  So we'll only write back
 785 * all dirty pages if they are all attached to "old" mappings.
 786 */
 787static long wb_writeback(struct bdi_writeback *wb,
 788                         struct wb_writeback_work *work)
 789{
 790        unsigned long wb_start = jiffies;
 791        long nr_pages = work->nr_pages;
 792        unsigned long oldest_jif;
 793        struct inode *inode;
 794        long progress;
 795
 796        oldest_jif = jiffies;
 797        work->older_than_this = &oldest_jif;
 798
 799        spin_lock(&wb->list_lock);
 800        for (;;) {
 801                /*
 802                 * Stop writeback when nr_pages has been consumed
 803                 */
 804                if (work->nr_pages <= 0)
 805                        break;
 806
 807                /*
 808                 * Background writeout and kupdate-style writeback may
 809                 * run forever. Stop them if there is other work to do
 810                 * so that e.g. sync can proceed. They'll be restarted
 811                 * after the other works are all done.
 812                 */
 813                if ((work->for_background || work->for_kupdate) &&
 814                    !list_empty(&wb->bdi->work_list))
 815                        break;
 816
 817                /*
 818                 * For background writeout, stop when we are below the
 819                 * background dirty threshold
 820                 */
 821                if (work->for_background && !over_bground_thresh(wb->bdi))
 822                        break;
 823
 824                /*
 825                 * Kupdate and background works are special and we want to
 826                 * include all inodes that need writing. Livelock avoidance is
 827                 * handled by these works yielding to any other work so we are
 828                 * safe.
 829                 */
 830                if (work->for_kupdate) {
 831                        oldest_jif = jiffies -
 832                                msecs_to_jiffies(dirty_expire_interval * 10);
 833                } else if (work->for_background)
 834                        oldest_jif = jiffies;
 835
 836                trace_writeback_start(wb->bdi, work);
 837                if (list_empty(&wb->b_io))
 838                        queue_io(wb, work);
 839                if (work->sb)
 840                        progress = writeback_sb_inodes(work->sb, wb, work);
 841                else
 842                        progress = __writeback_inodes_wb(wb, work);
 843                trace_writeback_written(wb->bdi, work);
 844
 845                wb_update_bandwidth(wb, wb_start);
 846
 847                /*
 848                 * Did we write something? Try for more
 849                 *
 850                 * Dirty inodes are moved to b_io for writeback in batches.
 851                 * The completion of the current batch does not necessarily
 852                 * mean the overall work is done. So we keep looping as long
 853                 * as made some progress on cleaning pages or inodes.
 854                 */
 855                if (progress)
 856                        continue;
 857                /*
 858                 * No more inodes for IO, bail
 859                 */
 860                if (list_empty(&wb->b_more_io))
 861                        break;
 862                /*
 863                 * Nothing written. Wait for some inode to
 864                 * become available for writeback. Otherwise
 865                 * we'll just busyloop.
 866                 */
 867                if (!list_empty(&wb->b_more_io))  {
 868                        trace_writeback_wait(wb->bdi, work);
 869                        inode = wb_inode(wb->b_more_io.prev);
 870                        spin_lock(&inode->i_lock);
 871                        spin_unlock(&wb->list_lock);
 872                        /* This function drops i_lock... */
 873                        inode_sleep_on_writeback(inode);
 874                        spin_lock(&wb->list_lock);
 875                }
 876        }
 877        spin_unlock(&wb->list_lock);
 878
 879        return nr_pages - work->nr_pages;
 880}
 881
 882/*
 883 * Return the next wb_writeback_work struct that hasn't been processed yet.
 884 */
 885static struct wb_writeback_work *
 886get_next_work_item(struct backing_dev_info *bdi)
 887{
 888        struct wb_writeback_work *work = NULL;
 889
 890        spin_lock_bh(&bdi->wb_lock);
 891        if (!list_empty(&bdi->work_list)) {
 892                work = list_entry(bdi->work_list.next,
 893                                  struct wb_writeback_work, list);
 894                list_del_init(&work->list);
 895        }
 896        spin_unlock_bh(&bdi->wb_lock);
 897        return work;
 898}
 899
 900/*
 901 * Add in the number of potentially dirty inodes, because each inode
 902 * write can dirty pagecache in the underlying blockdev.
 903 */
 904static unsigned long get_nr_dirty_pages(void)
 905{
 906        return global_page_state(NR_FILE_DIRTY) +
 907                global_page_state(NR_UNSTABLE_NFS) +
 908                get_nr_dirty_inodes();
 909}
 910
 911static long wb_check_background_flush(struct bdi_writeback *wb)
 912{
 913        if (over_bground_thresh(wb->bdi)) {
 914
 915                struct wb_writeback_work work = {
 916                        .nr_pages       = LONG_MAX,
 917                        .sync_mode      = WB_SYNC_NONE,
 918                        .for_background = 1,
 919                        .range_cyclic   = 1,
 920                        .reason         = WB_REASON_BACKGROUND,
 921                };
 922
 923                return wb_writeback(wb, &work);
 924        }
 925
 926        return 0;
 927}
 928
 929static long wb_check_old_data_flush(struct bdi_writeback *wb)
 930{
 931        unsigned long expired;
 932        long nr_pages;
 933
 934        /*
 935         * When set to zero, disable periodic writeback
 936         */
 937        if (!dirty_writeback_interval)
 938                return 0;
 939
 940        expired = wb->last_old_flush +
 941                        msecs_to_jiffies(dirty_writeback_interval * 10);
 942        if (time_before(jiffies, expired))
 943                return 0;
 944
 945        wb->last_old_flush = jiffies;
 946        nr_pages = get_nr_dirty_pages();
 947
 948        if (nr_pages) {
 949                struct wb_writeback_work work = {
 950                        .nr_pages       = nr_pages,
 951                        .sync_mode      = WB_SYNC_NONE,
 952                        .for_kupdate    = 1,
 953                        .range_cyclic   = 1,
 954                        .reason         = WB_REASON_PERIODIC,
 955                };
 956
 957                return wb_writeback(wb, &work);
 958        }
 959
 960        return 0;
 961}
 962
 963/*
 964 * Retrieve work items and do the writeback they describe
 965 */
 966static long wb_do_writeback(struct bdi_writeback *wb)
 967{
 968        struct backing_dev_info *bdi = wb->bdi;
 969        struct wb_writeback_work *work;
 970        long wrote = 0;
 971
 972        set_bit(BDI_writeback_running, &wb->bdi->state);
 973        while ((work = get_next_work_item(bdi)) != NULL) {
 974
 975                trace_writeback_exec(bdi, work);
 976
 977                wrote += wb_writeback(wb, work);
 978
 979                /*
 980                 * Notify the caller of completion if this is a synchronous
 981                 * work item, otherwise just free it.
 982                 */
 983                if (work->done)
 984                        complete(work->done);
 985                else
 986                        kfree(work);
 987        }
 988
 989        /*
 990         * Check for periodic writeback, kupdated() style
 991         */
 992        wrote += wb_check_old_data_flush(wb);
 993        wrote += wb_check_background_flush(wb);
 994        clear_bit(BDI_writeback_running, &wb->bdi->state);
 995
 996        return wrote;
 997}
 998
 999/*
1000 * Handle writeback of dirty data for the device backed by this bdi. Also
1001 * reschedules periodically and does kupdated style flushing.
1002 */
1003void bdi_writeback_workfn(struct work_struct *work)
1004{
1005        struct bdi_writeback *wb = container_of(to_delayed_work(work),
1006                                                struct bdi_writeback, dwork);
1007        struct backing_dev_info *bdi = wb->bdi;
1008        long pages_written;
1009
1010        set_worker_desc("flush-%s", dev_name(bdi->dev));
1011        current->flags |= PF_SWAPWRITE;
1012
1013        if (likely(!current_is_workqueue_rescuer() ||
1014                   list_empty(&bdi->bdi_list))) {
1015                /*
1016                 * The normal path.  Keep writing back @bdi until its
1017                 * work_list is empty.  Note that this path is also taken
1018                 * if @bdi is shutting down even when we're running off the
1019                 * rescuer as work_list needs to be drained.
1020                 */
1021                do {
1022                        pages_written = wb_do_writeback(wb);
1023                        trace_writeback_pages_written(pages_written);
1024                } while (!list_empty(&bdi->work_list));
1025        } else {
1026                /*
1027                 * bdi_wq can't get enough workers and we're running off
1028                 * the emergency worker.  Don't hog it.  Hopefully, 1024 is
1029                 * enough for efficient IO.
1030                 */
1031                pages_written = writeback_inodes_wb(&bdi->wb, 1024,
1032                                                    WB_REASON_FORKER_THREAD);
1033                trace_writeback_pages_written(pages_written);
1034        }
1035
1036        if (!list_empty(&bdi->work_list) ||
1037            (wb_has_dirty_io(wb) && dirty_writeback_interval))
1038                queue_delayed_work(bdi_wq, &wb->dwork,
1039                        msecs_to_jiffies(dirty_writeback_interval * 10));
1040
1041        current->flags &= ~PF_SWAPWRITE;
1042}
1043
1044/*
1045 * Start writeback of `nr_pages' pages.  If `nr_pages' is zero, write back
1046 * the whole world.
1047 */
1048void wakeup_flusher_threads(long nr_pages, enum wb_reason reason)
1049{
1050        struct backing_dev_info *bdi;
1051
1052        if (!nr_pages) {
1053                nr_pages = global_page_state(NR_FILE_DIRTY) +
1054                                global_page_state(NR_UNSTABLE_NFS);
1055        }
1056
1057        rcu_read_lock();
1058        list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) {
1059                if (!bdi_has_dirty_io(bdi))
1060                        continue;
1061                __bdi_start_writeback(bdi, nr_pages, false, reason);
1062        }
1063        rcu_read_unlock();
1064}
1065
1066static noinline void block_dump___mark_inode_dirty(struct inode *inode)
1067{
1068        if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) {
1069                struct dentry *dentry;
1070                const char *name = "?";
1071
1072                dentry = d_find_alias(inode);
1073                if (dentry) {
1074                        spin_lock(&dentry->d_lock);
1075                        name = (const char *) dentry->d_name.name;
1076                }
1077                printk(KERN_DEBUG
1078                       "%s(%d): dirtied inode %lu (%s) on %s\n",
1079                       current->comm, task_pid_nr(current), inode->i_ino,
1080                       name, inode->i_sb->s_id);
1081                if (dentry) {
1082                        spin_unlock(&dentry->d_lock);
1083                        dput(dentry);
1084                }
1085        }
1086}
1087
1088/**
1089 *      __mark_inode_dirty -    internal function
1090 *      @inode: inode to mark
1091 *      @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
1092 *      Mark an inode as dirty. Callers should use mark_inode_dirty or
1093 *      mark_inode_dirty_sync.
1094 *
1095 * Put the inode on the super block's dirty list.
1096 *
1097 * CAREFUL! We mark it dirty unconditionally, but move it onto the
1098 * dirty list only if it is hashed or if it refers to a blockdev.
1099 * If it was not hashed, it will never be added to the dirty list
1100 * even if it is later hashed, as it will have been marked dirty already.
1101 *
1102 * In short, make sure you hash any inodes _before_ you start marking
1103 * them dirty.
1104 *
1105 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
1106 * the block-special inode (/dev/hda1) itself.  And the ->dirtied_when field of
1107 * the kernel-internal blockdev inode represents the dirtying time of the
1108 * blockdev's pages.  This is why for I_DIRTY_PAGES we always use
1109 * page->mapping->host, so the page-dirtying time is recorded in the internal
1110 * blockdev inode.
1111 */
1112void __mark_inode_dirty(struct inode *inode, int flags)
1113{
1114        struct super_block *sb = inode->i_sb;
1115        struct backing_dev_info *bdi = NULL;
1116
1117        /*
1118         * Don't do this for I_DIRTY_PAGES - that doesn't actually
1119         * dirty the inode itself
1120         */
1121        if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
1122                trace_writeback_dirty_inode_start(inode, flags);
1123
1124                if (sb->s_op->dirty_inode)
1125                        sb->s_op->dirty_inode(inode, flags);
1126
1127                trace_writeback_dirty_inode(inode, flags);
1128        }
1129
1130        /*
1131         * make sure that changes are seen by all cpus before we test i_state
1132         * -- mikulas
1133         */
1134        smp_mb();
1135
1136        /* avoid the locking if we can */
1137        if ((inode->i_state & flags) == flags)
1138                return;
1139
1140        if (unlikely(block_dump))
1141                block_dump___mark_inode_dirty(inode);
1142
1143        spin_lock(&inode->i_lock);
1144        if ((inode->i_state & flags) != flags) {
1145                const int was_dirty = inode->i_state & I_DIRTY;
1146
1147                inode->i_state |= flags;
1148
1149                /*
1150                 * If the inode is being synced, just update its dirty state.
1151                 * The unlocker will place the inode on the appropriate
1152                 * superblock list, based upon its state.
1153                 */
1154                if (inode->i_state & I_SYNC)
1155                        goto out_unlock_inode;
1156
1157                /*
1158                 * Only add valid (hashed) inodes to the superblock's
1159                 * dirty list.  Add blockdev inodes as well.
1160                 */
1161                if (!S_ISBLK(inode->i_mode)) {
1162                        if (inode_unhashed(inode))
1163                                goto out_unlock_inode;
1164                }
1165                if (inode->i_state & I_FREEING)
1166                        goto out_unlock_inode;
1167
1168                /*
1169                 * If the inode was already on b_dirty/b_io/b_more_io, don't
1170                 * reposition it (that would break b_dirty time-ordering).
1171                 */
1172                if (!was_dirty) {
1173                        bool wakeup_bdi = false;
1174                        bdi = inode_to_bdi(inode);
1175
1176                        if (bdi_cap_writeback_dirty(bdi)) {
1177                                WARN(!test_bit(BDI_registered, &bdi->state),
1178                                     "bdi-%s not registered\n", bdi->name);
1179
1180                                /*
1181                                 * If this is the first dirty inode for this
1182                                 * bdi, we have to wake-up the corresponding
1183                                 * bdi thread to make sure background
1184                                 * write-back happens later.
1185                                 */
1186                                if (!wb_has_dirty_io(&bdi->wb))
1187                                        wakeup_bdi = true;
1188                        }
1189
1190                        spin_unlock(&inode->i_lock);
1191                        spin_lock(&bdi->wb.list_lock);
1192                        inode->dirtied_when = jiffies;
1193                        list_move(&inode->i_wb_list, &bdi->wb.b_dirty);
1194                        spin_unlock(&bdi->wb.list_lock);
1195
1196                        if (wakeup_bdi)
1197                                bdi_wakeup_thread_delayed(bdi);
1198                        return;
1199                }
1200        }
1201out_unlock_inode:
1202        spin_unlock(&inode->i_lock);
1203
1204}
1205EXPORT_SYMBOL(__mark_inode_dirty);
1206
1207static void wait_sb_inodes(struct super_block *sb)
1208{
1209        struct inode *inode, *old_inode = NULL;
1210
1211        /*
1212         * We need to be protected against the filesystem going from
1213         * r/o to r/w or vice versa.
1214         */
1215        WARN_ON(!rwsem_is_locked(&sb->s_umount));
1216
1217        spin_lock(&inode_sb_list_lock);
1218
1219        /*
1220         * Data integrity sync. Must wait for all pages under writeback,
1221         * because there may have been pages dirtied before our sync
1222         * call, but which had writeout started before we write it out.
1223         * In which case, the inode may not be on the dirty list, but
1224         * we still have to wait for that writeout.
1225         */
1226        list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1227                struct address_space *mapping = inode->i_mapping;
1228
1229                spin_lock(&inode->i_lock);
1230                if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
1231                    (mapping->nrpages == 0)) {
1232                        spin_unlock(&inode->i_lock);
1233                        continue;
1234                }
1235                __iget(inode);
1236                spin_unlock(&inode->i_lock);
1237                spin_unlock(&inode_sb_list_lock);
1238
1239                /*
1240                 * We hold a reference to 'inode' so it couldn't have been
1241                 * removed from s_inodes list while we dropped the
1242                 * inode_sb_list_lock.  We cannot iput the inode now as we can
1243                 * be holding the last reference and we cannot iput it under
1244                 * inode_sb_list_lock. So we keep the reference and iput it
1245                 * later.
1246                 */
1247                iput(old_inode);
1248                old_inode = inode;
1249
1250                filemap_fdatawait(mapping);
1251
1252                cond_resched();
1253
1254                spin_lock(&inode_sb_list_lock);
1255        }
1256        spin_unlock(&inode_sb_list_lock);
1257        iput(old_inode);
1258}
1259
1260/**
1261 * writeback_inodes_sb_nr -     writeback dirty inodes from given super_block
1262 * @sb: the superblock
1263 * @nr: the number of pages to write
1264 * @reason: reason why some writeback work initiated
1265 *
1266 * Start writeback on some inodes on this super_block. No guarantees are made
1267 * on how many (if any) will be written, and this function does not wait
1268 * for IO completion of submitted IO.
1269 */
1270void writeback_inodes_sb_nr(struct super_block *sb,
1271                            unsigned long nr,
1272                            enum wb_reason reason)
1273{
1274        DECLARE_COMPLETION_ONSTACK(done);
1275        struct wb_writeback_work work = {
1276                .sb                     = sb,
1277                .sync_mode              = WB_SYNC_NONE,
1278                .tagged_writepages      = 1,
1279                .done                   = &done,
1280                .nr_pages               = nr,
1281                .reason                 = reason,
1282        };
1283
1284        if (sb->s_bdi == &noop_backing_dev_info)
1285                return;
1286        WARN_ON(!rwsem_is_locked(&sb->s_umount));
1287        bdi_queue_work(sb->s_bdi, &work);
1288        wait_for_completion(&done);
1289}
1290EXPORT_SYMBOL(writeback_inodes_sb_nr);
1291
1292/**
1293 * writeback_inodes_sb  -       writeback dirty inodes from given super_block
1294 * @sb: the superblock
1295 * @reason: reason why some writeback work was initiated
1296 *
1297 * Start writeback on some inodes on this super_block. No guarantees are made
1298 * on how many (if any) will be written, and this function does not wait
1299 * for IO completion of submitted IO.
1300 */
1301void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason)
1302{
1303        return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason);
1304}
1305EXPORT_SYMBOL(writeback_inodes_sb);
1306
1307/**
1308 * try_to_writeback_inodes_sb_nr - try to start writeback if none underway
1309 * @sb: the superblock
1310 * @nr: the number of pages to write
1311 * @reason: the reason of writeback
1312 *
1313 * Invoke writeback_inodes_sb_nr if no writeback is currently underway.
1314 * Returns 1 if writeback was started, 0 if not.
1315 */
1316int try_to_writeback_inodes_sb_nr(struct super_block *sb,
1317                                  unsigned long nr,
1318                                  enum wb_reason reason)
1319{
1320        if (writeback_in_progress(sb->s_bdi))
1321                return 1;
1322
1323        if (!down_read_trylock(&sb->s_umount))
1324                return 0;
1325
1326        writeback_inodes_sb_nr(sb, nr, reason);
1327        up_read(&sb->s_umount);
1328        return 1;
1329}
1330EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr);
1331
1332/**
1333 * try_to_writeback_inodes_sb - try to start writeback if none underway
1334 * @sb: the superblock
1335 * @reason: reason why some writeback work was initiated
1336 *
1337 * Implement by try_to_writeback_inodes_sb_nr()
1338 * Returns 1 if writeback was started, 0 if not.
1339 */
1340int try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason)
1341{
1342        return try_to_writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason);
1343}
1344EXPORT_SYMBOL(try_to_writeback_inodes_sb);
1345
1346/**
1347 * sync_inodes_sb       -       sync sb inode pages
1348 * @sb: the superblock
1349 *
1350 * This function writes and waits on any dirty inode belonging to this
1351 * super_block.
1352 */
1353void sync_inodes_sb(struct super_block *sb)
1354{
1355        DECLARE_COMPLETION_ONSTACK(done);
1356        struct wb_writeback_work work = {
1357                .sb             = sb,
1358                .sync_mode      = WB_SYNC_ALL,
1359                .nr_pages       = LONG_MAX,
1360                .range_cyclic   = 0,
1361                .done           = &done,
1362                .reason         = WB_REASON_SYNC,
1363                .for_sync       = 1,
1364        };
1365
1366        /* Nothing to do? */
1367        if (sb->s_bdi == &noop_backing_dev_info)
1368                return;
1369        WARN_ON(!rwsem_is_locked(&sb->s_umount));
1370
1371        bdi_queue_work(sb->s_bdi, &work);
1372        wait_for_completion(&done);
1373
1374        wait_sb_inodes(sb);
1375}
1376EXPORT_SYMBOL(sync_inodes_sb);
1377
1378/**
1379 * write_inode_now      -       write an inode to disk
1380 * @inode: inode to write to disk
1381 * @sync: whether the write should be synchronous or not
1382 *
1383 * This function commits an inode to disk immediately if it is dirty. This is
1384 * primarily needed by knfsd.
1385 *
1386 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
1387 */
1388int write_inode_now(struct inode *inode, int sync)
1389{
1390        struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
1391        struct writeback_control wbc = {
1392                .nr_to_write = LONG_MAX,
1393                .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE,
1394                .range_start = 0,
1395                .range_end = LLONG_MAX,
1396        };
1397
1398        if (!mapping_cap_writeback_dirty(inode->i_mapping))
1399                wbc.nr_to_write = 0;
1400
1401        might_sleep();
1402        return writeback_single_inode(inode, wb, &wbc);
1403}
1404EXPORT_SYMBOL(write_inode_now);
1405
1406/**
1407 * sync_inode - write an inode and its pages to disk.
1408 * @inode: the inode to sync
1409 * @wbc: controls the writeback mode
1410 *
1411 * sync_inode() will write an inode and its pages to disk.  It will also
1412 * correctly update the inode on its superblock's dirty inode lists and will
1413 * update inode->i_state.
1414 *
1415 * The caller must have a ref on the inode.
1416 */
1417int sync_inode(struct inode *inode, struct writeback_control *wbc)
1418{
1419        return writeback_single_inode(inode, &inode_to_bdi(inode)->wb, wbc);
1420}
1421EXPORT_SYMBOL(sync_inode);
1422
1423/**
1424 * sync_inode_metadata - write an inode to disk
1425 * @inode: the inode to sync
1426 * @wait: wait for I/O to complete.
1427 *
1428 * Write an inode to disk and adjust its dirty state after completion.
1429 *
1430 * Note: only writes the actual inode, no associated data or other metadata.
1431 */
1432int sync_inode_metadata(struct inode *inode, int wait)
1433{
1434        struct writeback_control wbc = {
1435                .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE,
1436                .nr_to_write = 0, /* metadata-only */
1437        };
1438
1439        return sync_inode(inode, &wbc);
1440}
1441EXPORT_SYMBOL(sync_inode_metadata);
1442
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.