11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * fs/fs-writeback.c 31da177e4SLinus Torvalds * 41da177e4SLinus Torvalds * Copyright (C) 2002, Linus Torvalds. 51da177e4SLinus Torvalds * 61da177e4SLinus Torvalds * Contains all the functions related to writing back and waiting 71da177e4SLinus Torvalds * upon dirty inodes against superblocks, and writing back dirty 81da177e4SLinus Torvalds * pages against inodes. ie: data writeback. Writeout of the 91da177e4SLinus Torvalds * inode itself is not handled here. 101da177e4SLinus Torvalds * 11e1f8e874SFrancois Cami * 10Apr2002 Andrew Morton 121da177e4SLinus Torvalds * Split out of fs/inode.c 131da177e4SLinus Torvalds * Additions for address_space-based writeback 141da177e4SLinus Torvalds */ 151da177e4SLinus Torvalds 161da177e4SLinus Torvalds #include <linux/kernel.h> 17f5ff8422SJens Axboe #include <linux/module.h> 181da177e4SLinus Torvalds #include <linux/spinlock.h> 195a0e3ad6STejun Heo #include <linux/slab.h> 201da177e4SLinus Torvalds #include <linux/sched.h> 211da177e4SLinus Torvalds #include <linux/fs.h> 221da177e4SLinus Torvalds #include <linux/mm.h> 2303ba3782SJens Axboe #include <linux/kthread.h> 2403ba3782SJens Axboe #include <linux/freezer.h> 251da177e4SLinus Torvalds #include <linux/writeback.h> 261da177e4SLinus Torvalds #include <linux/blkdev.h> 271da177e4SLinus Torvalds #include <linux/backing-dev.h> 281da177e4SLinus Torvalds #include <linux/buffer_head.h> 2907f3f05cSDavid Howells #include "internal.h" 301da177e4SLinus Torvalds 3166f3b8e2SJens Axboe #define inode_to_bdi(inode) ((inode)->i_mapping->backing_dev_info) 32f11b00f3SAdrian Bunk 3303ba3782SJens Axboe /* 34d0bceac7SJens Axboe * We don't actually have pdflush, but this one is exported though /proc... 35d0bceac7SJens Axboe */ 36d0bceac7SJens Axboe int nr_pdflush_threads; 37d0bceac7SJens Axboe 38d0bceac7SJens Axboe /* 39c4a77a6cSJens Axboe * Passed into wb_writeback(), essentially a subset of writeback_control 40c4a77a6cSJens Axboe */ 41c4a77a6cSJens Axboe struct wb_writeback_args { 42c4a77a6cSJens Axboe long nr_pages; 43c4a77a6cSJens Axboe struct super_block *sb; 44c4a77a6cSJens Axboe enum writeback_sync_modes sync_mode; 45d3ddec76SWu Fengguang int for_kupdate:1; 46d3ddec76SWu Fengguang int range_cyclic:1; 47d3ddec76SWu Fengguang int for_background:1; 48e913fc82SJens Axboe int sb_pinned:1; 49c4a77a6cSJens Axboe }; 50c4a77a6cSJens Axboe 51c4a77a6cSJens Axboe /* 5203ba3782SJens Axboe * Work items for the bdi_writeback threads 53f11b00f3SAdrian Bunk */ 5403ba3782SJens Axboe struct bdi_work { 558010c3b6SJens Axboe struct list_head list; /* pending work list */ 568010c3b6SJens Axboe struct rcu_head rcu_head; /* for RCU free/clear of work */ 5703ba3782SJens Axboe 588010c3b6SJens Axboe unsigned long seen; /* threads that have seen this work */ 598010c3b6SJens Axboe atomic_t pending; /* number of threads still to do work */ 6003ba3782SJens Axboe 618010c3b6SJens Axboe struct wb_writeback_args args; /* writeback arguments */ 6203ba3782SJens Axboe 638010c3b6SJens Axboe unsigned long state; /* flag bits, see WS_* */ 6403ba3782SJens Axboe }; 6503ba3782SJens Axboe 6603ba3782SJens Axboe enum { 6703ba3782SJens Axboe WS_USED_B = 0, 6803ba3782SJens Axboe WS_ONSTACK_B, 6903ba3782SJens Axboe }; 7003ba3782SJens Axboe 7103ba3782SJens Axboe #define WS_USED (1 << WS_USED_B) 7203ba3782SJens Axboe #define WS_ONSTACK (1 << WS_ONSTACK_B) 7303ba3782SJens Axboe 7403ba3782SJens Axboe static inline bool bdi_work_on_stack(struct bdi_work *work) 75f11b00f3SAdrian Bunk { 7603ba3782SJens Axboe return test_bit(WS_ONSTACK_B, &work->state); 7703ba3782SJens Axboe } 7803ba3782SJens Axboe 7903ba3782SJens Axboe static inline void bdi_work_init(struct bdi_work *work, 80b6e51316SJens Axboe struct wb_writeback_args *args) 8103ba3782SJens Axboe { 8203ba3782SJens Axboe INIT_RCU_HEAD(&work->rcu_head); 83b6e51316SJens Axboe work->args = *args; 8403ba3782SJens Axboe work->state = WS_USED; 8503ba3782SJens Axboe } 8603ba3782SJens Axboe 87f11b00f3SAdrian Bunk /** 88f11b00f3SAdrian Bunk * writeback_in_progress - determine whether there is writeback in progress 89f11b00f3SAdrian Bunk * @bdi: the device's backing_dev_info structure. 90f11b00f3SAdrian Bunk * 9103ba3782SJens Axboe * Determine whether there is writeback waiting to be handled against a 9203ba3782SJens Axboe * backing device. 93f11b00f3SAdrian Bunk */ 94f11b00f3SAdrian Bunk int writeback_in_progress(struct backing_dev_info *bdi) 95f11b00f3SAdrian Bunk { 9603ba3782SJens Axboe return !list_empty(&bdi->work_list); 97f11b00f3SAdrian Bunk } 98f11b00f3SAdrian Bunk 9903ba3782SJens Axboe static void bdi_work_clear(struct bdi_work *work) 100f11b00f3SAdrian Bunk { 10103ba3782SJens Axboe clear_bit(WS_USED_B, &work->state); 10203ba3782SJens Axboe smp_mb__after_clear_bit(); 1031ef7d9aaSNick Piggin /* 1041ef7d9aaSNick Piggin * work can have disappeared at this point. bit waitq functions 1051ef7d9aaSNick Piggin * should be able to tolerate this, provided bdi_sched_wait does 1061ef7d9aaSNick Piggin * not dereference it's pointer argument. 1071ef7d9aaSNick Piggin */ 10803ba3782SJens Axboe wake_up_bit(&work->state, WS_USED_B); 109f11b00f3SAdrian Bunk } 110f11b00f3SAdrian Bunk 11103ba3782SJens Axboe static void bdi_work_free(struct rcu_head *head) 1124195f73dSNick Piggin { 11303ba3782SJens Axboe struct bdi_work *work = container_of(head, struct bdi_work, rcu_head); 1144195f73dSNick Piggin 11503ba3782SJens Axboe if (!bdi_work_on_stack(work)) 11603ba3782SJens Axboe kfree(work); 11703ba3782SJens Axboe else 11803ba3782SJens Axboe bdi_work_clear(work); 1194195f73dSNick Piggin } 1204195f73dSNick Piggin 12103ba3782SJens Axboe static void wb_work_complete(struct bdi_work *work) 1221da177e4SLinus Torvalds { 123c4a77a6cSJens Axboe const enum writeback_sync_modes sync_mode = work->args.sync_mode; 12477b9d059SNick Piggin int onstack = bdi_work_on_stack(work); 1251da177e4SLinus Torvalds 1261da177e4SLinus Torvalds /* 12703ba3782SJens Axboe * For allocated work, we can clear the done/seen bit right here. 12803ba3782SJens Axboe * For on-stack work, we need to postpone both the clear and free 12903ba3782SJens Axboe * to after the RCU grace period, since the stack could be invalidated 13003ba3782SJens Axboe * as soon as bdi_work_clear() has done the wakeup. 1311da177e4SLinus Torvalds */ 13277b9d059SNick Piggin if (!onstack) 13303ba3782SJens Axboe bdi_work_clear(work); 13477b9d059SNick Piggin if (sync_mode == WB_SYNC_NONE || onstack) 13503ba3782SJens Axboe call_rcu(&work->rcu_head, bdi_work_free); 1361da177e4SLinus Torvalds } 1371da177e4SLinus Torvalds 13803ba3782SJens Axboe static void wb_clear_pending(struct bdi_writeback *wb, struct bdi_work *work) 13903ba3782SJens Axboe { 1401da177e4SLinus Torvalds /* 14103ba3782SJens Axboe * The caller has retrieved the work arguments from this work, 14203ba3782SJens Axboe * drop our reference. If this is the last ref, delete and free it 14303ba3782SJens Axboe */ 14403ba3782SJens Axboe if (atomic_dec_and_test(&work->pending)) { 14503ba3782SJens Axboe struct backing_dev_info *bdi = wb->bdi; 14603ba3782SJens Axboe 14703ba3782SJens Axboe spin_lock(&bdi->wb_lock); 14803ba3782SJens Axboe list_del_rcu(&work->list); 14903ba3782SJens Axboe spin_unlock(&bdi->wb_lock); 15003ba3782SJens Axboe 15103ba3782SJens Axboe wb_work_complete(work); 15203ba3782SJens Axboe } 15303ba3782SJens Axboe } 15403ba3782SJens Axboe 15503ba3782SJens Axboe static void bdi_queue_work(struct backing_dev_info *bdi, struct bdi_work *work) 15603ba3782SJens Axboe { 15703ba3782SJens Axboe work->seen = bdi->wb_mask; 15803ba3782SJens Axboe BUG_ON(!work->seen); 15903ba3782SJens Axboe atomic_set(&work->pending, bdi->wb_cnt); 16003ba3782SJens Axboe BUG_ON(!bdi->wb_cnt); 16103ba3782SJens Axboe 16203ba3782SJens Axboe /* 163deed62edSNick Piggin * list_add_tail_rcu() contains the necessary barriers to 164deed62edSNick Piggin * make sure the above stores are seen before the item is 165deed62edSNick Piggin * noticed on the list 1661da177e4SLinus Torvalds */ 16703ba3782SJens Axboe spin_lock(&bdi->wb_lock); 16803ba3782SJens Axboe list_add_tail_rcu(&work->list, &bdi->work_list); 16903ba3782SJens Axboe spin_unlock(&bdi->wb_lock); 1701da177e4SLinus Torvalds 1711da177e4SLinus Torvalds /* 17203ba3782SJens Axboe * If the default thread isn't there, make sure we add it. When 17303ba3782SJens Axboe * it gets created and wakes up, we'll run this work. 1741da177e4SLinus Torvalds */ 17503ba3782SJens Axboe if (unlikely(list_empty_careful(&bdi->wb_list))) 17603ba3782SJens Axboe wake_up_process(default_backing_dev_info.wb.task); 17703ba3782SJens Axboe else { 17803ba3782SJens Axboe struct bdi_writeback *wb = &bdi->wb; 1791da177e4SLinus Torvalds 1801ef7d9aaSNick Piggin if (wb->task) 18103ba3782SJens Axboe wake_up_process(wb->task); 1821da177e4SLinus Torvalds } 18303ba3782SJens Axboe } 1841da177e4SLinus Torvalds 1851da177e4SLinus Torvalds /* 18603ba3782SJens Axboe * Used for on-stack allocated work items. The caller needs to wait until 18703ba3782SJens Axboe * the wb threads have acked the work before it's safe to continue. 1881da177e4SLinus Torvalds */ 18903ba3782SJens Axboe static void bdi_wait_on_work_clear(struct bdi_work *work) 1901da177e4SLinus Torvalds { 19103ba3782SJens Axboe wait_on_bit(&work->state, WS_USED_B, bdi_sched_wait, 19203ba3782SJens Axboe TASK_UNINTERRUPTIBLE); 19303ba3782SJens Axboe } 19403ba3782SJens Axboe 195f11fcae8SJens Axboe static void bdi_alloc_queue_work(struct backing_dev_info *bdi, 1967c8a3554SJens Axboe struct wb_writeback_args *args, 1977c8a3554SJens Axboe int wait) 19803ba3782SJens Axboe { 19903ba3782SJens Axboe struct bdi_work *work; 20003ba3782SJens Axboe 201bcddc3f0SJens Axboe /* 202bcddc3f0SJens Axboe * This is WB_SYNC_NONE writeback, so if allocation fails just 203bcddc3f0SJens Axboe * wakeup the thread for old dirty data writeback 204bcddc3f0SJens Axboe */ 20503ba3782SJens Axboe work = kmalloc(sizeof(*work), GFP_ATOMIC); 206bcddc3f0SJens Axboe if (work) { 207b6e51316SJens Axboe bdi_work_init(work, args); 208f11fcae8SJens Axboe bdi_queue_work(bdi, work); 2097c8a3554SJens Axboe if (wait) 2107c8a3554SJens Axboe bdi_wait_on_work_clear(work); 211bcddc3f0SJens Axboe } else { 212bcddc3f0SJens Axboe struct bdi_writeback *wb = &bdi->wb; 213bcddc3f0SJens Axboe 214bcddc3f0SJens Axboe if (wb->task) 215bcddc3f0SJens Axboe wake_up_process(wb->task); 216bcddc3f0SJens Axboe } 21703ba3782SJens Axboe } 21803ba3782SJens Axboe 219b6e51316SJens Axboe /** 220b6e51316SJens Axboe * bdi_sync_writeback - start and wait for writeback 221b6e51316SJens Axboe * @bdi: the backing device to write from 222b6e51316SJens Axboe * @sb: write inodes from this super_block 223b6e51316SJens Axboe * 224b6e51316SJens Axboe * Description: 225b6e51316SJens Axboe * This does WB_SYNC_ALL data integrity writeback and waits for the 226b6e51316SJens Axboe * IO to complete. Callers must hold the sb s_umount semaphore for 227b6e51316SJens Axboe * reading, to avoid having the super disappear before we are done. 22803ba3782SJens Axboe */ 229b6e51316SJens Axboe static void bdi_sync_writeback(struct backing_dev_info *bdi, 230b6e51316SJens Axboe struct super_block *sb) 231b6e51316SJens Axboe { 232b6e51316SJens Axboe struct wb_writeback_args args = { 233b6e51316SJens Axboe .sb = sb, 234b6e51316SJens Axboe .sync_mode = WB_SYNC_ALL, 235b6e51316SJens Axboe .nr_pages = LONG_MAX, 236b6e51316SJens Axboe .range_cyclic = 0, 237e913fc82SJens Axboe /* 238e913fc82SJens Axboe * Setting sb_pinned is not necessary for WB_SYNC_ALL, but 239e913fc82SJens Axboe * lets make it explicitly clear. 240e913fc82SJens Axboe */ 241e913fc82SJens Axboe .sb_pinned = 1, 242b6e51316SJens Axboe }; 243f0fad8a5SChristoph Hellwig struct bdi_work work; 244f0fad8a5SChristoph Hellwig 245b6e51316SJens Axboe bdi_work_init(&work, &args); 246f0fad8a5SChristoph Hellwig work.state |= WS_ONSTACK; 247f0fad8a5SChristoph Hellwig 248b6e51316SJens Axboe bdi_queue_work(bdi, &work); 249f0fad8a5SChristoph Hellwig bdi_wait_on_work_clear(&work); 25003ba3782SJens Axboe } 251b6e51316SJens Axboe 252b6e51316SJens Axboe /** 253b6e51316SJens Axboe * bdi_start_writeback - start writeback 254b6e51316SJens Axboe * @bdi: the backing device to write from 2554b6764faSJaswinder Singh Rajput * @sb: write inodes from this super_block 256b6e51316SJens Axboe * @nr_pages: the number of pages to write 257e913fc82SJens Axboe * @sb_locked: caller already holds sb umount sem. 258b6e51316SJens Axboe * 259b6e51316SJens Axboe * Description: 260b6e51316SJens Axboe * This does WB_SYNC_NONE opportunistic writeback. The IO is only 261b6e51316SJens Axboe * started when this function returns, we make no guarentees on 262e913fc82SJens Axboe * completion. Caller specifies whether sb umount sem is held already or not. 263b6e51316SJens Axboe * 264b6e51316SJens Axboe */ 265a72bfd4dSJens Axboe void bdi_start_writeback(struct backing_dev_info *bdi, struct super_block *sb, 266e913fc82SJens Axboe long nr_pages, int sb_locked) 267b6e51316SJens Axboe { 268b6e51316SJens Axboe struct wb_writeback_args args = { 269a72bfd4dSJens Axboe .sb = sb, 270b6e51316SJens Axboe .sync_mode = WB_SYNC_NONE, 271b6e51316SJens Axboe .nr_pages = nr_pages, 272b6e51316SJens Axboe .range_cyclic = 1, 273e913fc82SJens Axboe .sb_pinned = sb_locked, 274b6e51316SJens Axboe }; 275b6e51316SJens Axboe 276d3ddec76SWu Fengguang /* 277d3ddec76SWu Fengguang * We treat @nr_pages=0 as the special case to do background writeback, 278d3ddec76SWu Fengguang * ie. to sync pages until the background dirty threshold is reached. 279d3ddec76SWu Fengguang */ 280d3ddec76SWu Fengguang if (!nr_pages) { 281d3ddec76SWu Fengguang args.nr_pages = LONG_MAX; 282d3ddec76SWu Fengguang args.for_background = 1; 283d3ddec76SWu Fengguang } 284d3ddec76SWu Fengguang 2857c8a3554SJens Axboe bdi_alloc_queue_work(bdi, &args, sb_locked); 2861da177e4SLinus Torvalds } 2871da177e4SLinus Torvalds 2881da177e4SLinus Torvalds /* 2896610a0bcSAndrew Morton * Redirty an inode: set its when-it-was dirtied timestamp and move it to the 2906610a0bcSAndrew Morton * furthest end of its superblock's dirty-inode list. 2916610a0bcSAndrew Morton * 2926610a0bcSAndrew Morton * Before stamping the inode's ->dirtied_when, we check to see whether it is 29366f3b8e2SJens Axboe * already the most-recently-dirtied inode on the b_dirty list. If that is 2946610a0bcSAndrew Morton * the case then the inode must have been redirtied while it was being written 2956610a0bcSAndrew Morton * out and we don't reset its dirtied_when. 2966610a0bcSAndrew Morton */ 2976610a0bcSAndrew Morton static void redirty_tail(struct inode *inode) 2986610a0bcSAndrew Morton { 29903ba3782SJens Axboe struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; 3006610a0bcSAndrew Morton 30103ba3782SJens Axboe if (!list_empty(&wb->b_dirty)) { 30266f3b8e2SJens Axboe struct inode *tail; 3036610a0bcSAndrew Morton 30403ba3782SJens Axboe tail = list_entry(wb->b_dirty.next, struct inode, i_list); 30566f3b8e2SJens Axboe if (time_before(inode->dirtied_when, tail->dirtied_when)) 3066610a0bcSAndrew Morton inode->dirtied_when = jiffies; 3076610a0bcSAndrew Morton } 30803ba3782SJens Axboe list_move(&inode->i_list, &wb->b_dirty); 3096610a0bcSAndrew Morton } 3106610a0bcSAndrew Morton 3116610a0bcSAndrew Morton /* 31266f3b8e2SJens Axboe * requeue inode for re-scanning after bdi->b_io list is exhausted. 313c986d1e2SAndrew Morton */ 3140e0f4fc2SKen Chen static void requeue_io(struct inode *inode) 315c986d1e2SAndrew Morton { 31603ba3782SJens Axboe struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; 31703ba3782SJens Axboe 31803ba3782SJens Axboe list_move(&inode->i_list, &wb->b_more_io); 319c986d1e2SAndrew Morton } 320c986d1e2SAndrew Morton 3211c0eeaf5SJoern Engel static void inode_sync_complete(struct inode *inode) 3221c0eeaf5SJoern Engel { 3231c0eeaf5SJoern Engel /* 3241c0eeaf5SJoern Engel * Prevent speculative execution through spin_unlock(&inode_lock); 3251c0eeaf5SJoern Engel */ 3261c0eeaf5SJoern Engel smp_mb(); 3271c0eeaf5SJoern Engel wake_up_bit(&inode->i_state, __I_SYNC); 3281c0eeaf5SJoern Engel } 3291c0eeaf5SJoern Engel 330d2caa3c5SJeff Layton static bool inode_dirtied_after(struct inode *inode, unsigned long t) 331d2caa3c5SJeff Layton { 332d2caa3c5SJeff Layton bool ret = time_after(inode->dirtied_when, t); 333d2caa3c5SJeff Layton #ifndef CONFIG_64BIT 334d2caa3c5SJeff Layton /* 335d2caa3c5SJeff Layton * For inodes being constantly redirtied, dirtied_when can get stuck. 336d2caa3c5SJeff Layton * It _appears_ to be in the future, but is actually in distant past. 337d2caa3c5SJeff Layton * This test is necessary to prevent such wrapped-around relative times 3385b0830cbSJens Axboe * from permanently stopping the whole bdi writeback. 339d2caa3c5SJeff Layton */ 340d2caa3c5SJeff Layton ret = ret && time_before_eq(inode->dirtied_when, jiffies); 341d2caa3c5SJeff Layton #endif 342d2caa3c5SJeff Layton return ret; 343d2caa3c5SJeff Layton } 344d2caa3c5SJeff Layton 345c986d1e2SAndrew Morton /* 3462c136579SFengguang Wu * Move expired dirty inodes from @delaying_queue to @dispatch_queue. 3472c136579SFengguang Wu */ 3482c136579SFengguang Wu static void move_expired_inodes(struct list_head *delaying_queue, 3492c136579SFengguang Wu struct list_head *dispatch_queue, 3502c136579SFengguang Wu unsigned long *older_than_this) 3512c136579SFengguang Wu { 3525c03449dSShaohua Li LIST_HEAD(tmp); 3535c03449dSShaohua Li struct list_head *pos, *node; 354cf137307SJens Axboe struct super_block *sb = NULL; 3555c03449dSShaohua Li struct inode *inode; 356cf137307SJens Axboe int do_sb_sort = 0; 3575c03449dSShaohua Li 3582c136579SFengguang Wu while (!list_empty(delaying_queue)) { 3595c03449dSShaohua Li inode = list_entry(delaying_queue->prev, struct inode, i_list); 3602c136579SFengguang Wu if (older_than_this && 361d2caa3c5SJeff Layton inode_dirtied_after(inode, *older_than_this)) 3622c136579SFengguang Wu break; 363cf137307SJens Axboe if (sb && sb != inode->i_sb) 364cf137307SJens Axboe do_sb_sort = 1; 365cf137307SJens Axboe sb = inode->i_sb; 3665c03449dSShaohua Li list_move(&inode->i_list, &tmp); 3675c03449dSShaohua Li } 3685c03449dSShaohua Li 369cf137307SJens Axboe /* just one sb in list, splice to dispatch_queue and we're done */ 370cf137307SJens Axboe if (!do_sb_sort) { 371cf137307SJens Axboe list_splice(&tmp, dispatch_queue); 372cf137307SJens Axboe return; 373cf137307SJens Axboe } 374cf137307SJens Axboe 3755c03449dSShaohua Li /* Move inodes from one superblock together */ 3765c03449dSShaohua Li while (!list_empty(&tmp)) { 3775c03449dSShaohua Li inode = list_entry(tmp.prev, struct inode, i_list); 3785c03449dSShaohua Li sb = inode->i_sb; 3795c03449dSShaohua Li list_for_each_prev_safe(pos, node, &tmp) { 3805c03449dSShaohua Li inode = list_entry(pos, struct inode, i_list); 3815c03449dSShaohua Li if (inode->i_sb == sb) 3822c136579SFengguang Wu list_move(&inode->i_list, dispatch_queue); 3832c136579SFengguang Wu } 3842c136579SFengguang Wu } 3855c03449dSShaohua Li } 3862c136579SFengguang Wu 3872c136579SFengguang Wu /* 3882c136579SFengguang Wu * Queue all expired dirty inodes for io, eldest first. 3892c136579SFengguang Wu */ 39003ba3782SJens Axboe static void queue_io(struct bdi_writeback *wb, unsigned long *older_than_this) 3912c136579SFengguang Wu { 39203ba3782SJens Axboe list_splice_init(&wb->b_more_io, wb->b_io.prev); 39303ba3782SJens Axboe move_expired_inodes(&wb->b_dirty, &wb->b_io, older_than_this); 39466f3b8e2SJens Axboe } 39566f3b8e2SJens Axboe 396a9185b41SChristoph Hellwig static int write_inode(struct inode *inode, struct writeback_control *wbc) 39766f3b8e2SJens Axboe { 39803ba3782SJens Axboe if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) 399a9185b41SChristoph Hellwig return inode->i_sb->s_op->write_inode(inode, wbc); 40003ba3782SJens Axboe return 0; 40166f3b8e2SJens Axboe } 40208d8e974SFengguang Wu 4032c136579SFengguang Wu /* 40401c03194SChristoph Hellwig * Wait for writeback on an inode to complete. 40501c03194SChristoph Hellwig */ 40601c03194SChristoph Hellwig static void inode_wait_for_writeback(struct inode *inode) 40701c03194SChristoph Hellwig { 40801c03194SChristoph Hellwig DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); 40901c03194SChristoph Hellwig wait_queue_head_t *wqh; 41001c03194SChristoph Hellwig 41101c03194SChristoph Hellwig wqh = bit_waitqueue(&inode->i_state, __I_SYNC); 41201c03194SChristoph Hellwig do { 41301c03194SChristoph Hellwig spin_unlock(&inode_lock); 41401c03194SChristoph Hellwig __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE); 41501c03194SChristoph Hellwig spin_lock(&inode_lock); 41601c03194SChristoph Hellwig } while (inode->i_state & I_SYNC); 41701c03194SChristoph Hellwig } 41801c03194SChristoph Hellwig 41901c03194SChristoph Hellwig /* 42001c03194SChristoph Hellwig * Write out an inode's dirty pages. Called under inode_lock. Either the 42101c03194SChristoph Hellwig * caller has ref on the inode (either via __iget or via syscall against an fd) 42201c03194SChristoph Hellwig * or the inode has I_WILL_FREE set (via generic_forget_inode) 42301c03194SChristoph Hellwig * 4241da177e4SLinus Torvalds * If `wait' is set, wait on the writeout. 4251da177e4SLinus Torvalds * 4261da177e4SLinus Torvalds * The whole writeout design is quite complex and fragile. We want to avoid 4271da177e4SLinus Torvalds * starvation of particular inodes when others are being redirtied, prevent 4281da177e4SLinus Torvalds * livelocks, etc. 4291da177e4SLinus Torvalds * 4301da177e4SLinus Torvalds * Called under inode_lock. 4311da177e4SLinus Torvalds */ 4321da177e4SLinus Torvalds static int 43301c03194SChristoph Hellwig writeback_single_inode(struct inode *inode, struct writeback_control *wbc) 4341da177e4SLinus Torvalds { 4351da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 43601c03194SChristoph Hellwig unsigned dirty; 4371da177e4SLinus Torvalds int ret; 4381da177e4SLinus Torvalds 43901c03194SChristoph Hellwig if (!atomic_read(&inode->i_count)) 44001c03194SChristoph Hellwig WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); 44101c03194SChristoph Hellwig else 44201c03194SChristoph Hellwig WARN_ON(inode->i_state & I_WILL_FREE); 44301c03194SChristoph Hellwig 44401c03194SChristoph Hellwig if (inode->i_state & I_SYNC) { 44501c03194SChristoph Hellwig /* 44601c03194SChristoph Hellwig * If this inode is locked for writeback and we are not doing 44766f3b8e2SJens Axboe * writeback-for-data-integrity, move it to b_more_io so that 44801c03194SChristoph Hellwig * writeback can proceed with the other inodes on s_io. 44901c03194SChristoph Hellwig * 45001c03194SChristoph Hellwig * We'll have another go at writing back this inode when we 45166f3b8e2SJens Axboe * completed a full scan of b_io. 45201c03194SChristoph Hellwig */ 453a9185b41SChristoph Hellwig if (wbc->sync_mode != WB_SYNC_ALL) { 45401c03194SChristoph Hellwig requeue_io(inode); 45501c03194SChristoph Hellwig return 0; 45601c03194SChristoph Hellwig } 45701c03194SChristoph Hellwig 45801c03194SChristoph Hellwig /* 45901c03194SChristoph Hellwig * It's a data-integrity sync. We must wait. 46001c03194SChristoph Hellwig */ 46101c03194SChristoph Hellwig inode_wait_for_writeback(inode); 46201c03194SChristoph Hellwig } 46301c03194SChristoph Hellwig 4641c0eeaf5SJoern Engel BUG_ON(inode->i_state & I_SYNC); 4651da177e4SLinus Torvalds 4665547e8aaSDmitry Monakhov /* Set I_SYNC, reset I_DIRTY_PAGES */ 4671c0eeaf5SJoern Engel inode->i_state |= I_SYNC; 4685547e8aaSDmitry Monakhov inode->i_state &= ~I_DIRTY_PAGES; 4691da177e4SLinus Torvalds spin_unlock(&inode_lock); 4701da177e4SLinus Torvalds 4711da177e4SLinus Torvalds ret = do_writepages(mapping, wbc); 4721da177e4SLinus Torvalds 47326821ed4SChristoph Hellwig /* 47426821ed4SChristoph Hellwig * Make sure to wait on the data before writing out the metadata. 47526821ed4SChristoph Hellwig * This is important for filesystems that modify metadata on data 47626821ed4SChristoph Hellwig * I/O completion. 47726821ed4SChristoph Hellwig */ 478a9185b41SChristoph Hellwig if (wbc->sync_mode == WB_SYNC_ALL) { 47926821ed4SChristoph Hellwig int err = filemap_fdatawait(mapping); 4801da177e4SLinus Torvalds if (ret == 0) 4811da177e4SLinus Torvalds ret = err; 4821da177e4SLinus Torvalds } 4831da177e4SLinus Torvalds 4845547e8aaSDmitry Monakhov /* 4855547e8aaSDmitry Monakhov * Some filesystems may redirty the inode during the writeback 4865547e8aaSDmitry Monakhov * due to delalloc, clear dirty metadata flags right before 4875547e8aaSDmitry Monakhov * write_inode() 4885547e8aaSDmitry Monakhov */ 4895547e8aaSDmitry Monakhov spin_lock(&inode_lock); 4905547e8aaSDmitry Monakhov dirty = inode->i_state & I_DIRTY; 4915547e8aaSDmitry Monakhov inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC); 4925547e8aaSDmitry Monakhov spin_unlock(&inode_lock); 49326821ed4SChristoph Hellwig /* Don't write the inode if only I_DIRTY_PAGES was set */ 49426821ed4SChristoph Hellwig if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { 495a9185b41SChristoph Hellwig int err = write_inode(inode, wbc); 4961da177e4SLinus Torvalds if (ret == 0) 4971da177e4SLinus Torvalds ret = err; 4981da177e4SLinus Torvalds } 4991da177e4SLinus Torvalds 5001da177e4SLinus Torvalds spin_lock(&inode_lock); 5011c0eeaf5SJoern Engel inode->i_state &= ~I_SYNC; 50284a89245SWu Fengguang if (!(inode->i_state & (I_FREEING | I_CLEAR))) { 503b3af9468SWu Fengguang if ((inode->i_state & I_DIRTY_PAGES) && wbc->for_kupdate) { 504ae1b7f7dSWu Fengguang /* 505b3af9468SWu Fengguang * More pages get dirtied by a fast dirtier. 506b3af9468SWu Fengguang */ 507b3af9468SWu Fengguang goto select_queue; 508b3af9468SWu Fengguang } else if (inode->i_state & I_DIRTY) { 509b3af9468SWu Fengguang /* 510b3af9468SWu Fengguang * At least XFS will redirty the inode during the 511b3af9468SWu Fengguang * writeback (delalloc) and on io completion (isize). 512ae1b7f7dSWu Fengguang */ 513ae1b7f7dSWu Fengguang redirty_tail(inode); 514ae1b7f7dSWu Fengguang } else if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { 5151da177e4SLinus Torvalds /* 5161da177e4SLinus Torvalds * We didn't write back all the pages. nfs_writepages() 5171da177e4SLinus Torvalds * sometimes bales out without doing anything. Redirty 51866f3b8e2SJens Axboe * the inode; Move it from b_io onto b_more_io/b_dirty. 5191b43ef91SAndrew Morton */ 5201b43ef91SAndrew Morton /* 5211b43ef91SAndrew Morton * akpm: if the caller was the kupdate function we put 52266f3b8e2SJens Axboe * this inode at the head of b_dirty so it gets first 5231b43ef91SAndrew Morton * consideration. Otherwise, move it to the tail, for 5241b43ef91SAndrew Morton * the reasons described there. I'm not really sure 5251b43ef91SAndrew Morton * how much sense this makes. Presumably I had a good 5261b43ef91SAndrew Morton * reasons for doing it this way, and I'd rather not 5271b43ef91SAndrew Morton * muck with it at present. 5281da177e4SLinus Torvalds */ 5291da177e4SLinus Torvalds if (wbc->for_kupdate) { 5301da177e4SLinus Torvalds /* 5312c136579SFengguang Wu * For the kupdate function we move the inode 53266f3b8e2SJens Axboe * to b_more_io so it will get more writeout as 5332c136579SFengguang Wu * soon as the queue becomes uncongested. 5341da177e4SLinus Torvalds */ 5351da177e4SLinus Torvalds inode->i_state |= I_DIRTY_PAGES; 536b3af9468SWu Fengguang select_queue: 5378bc3be27SFengguang Wu if (wbc->nr_to_write <= 0) { 5388bc3be27SFengguang Wu /* 5398bc3be27SFengguang Wu * slice used up: queue for next turn 5408bc3be27SFengguang Wu */ 5410e0f4fc2SKen Chen requeue_io(inode); 5421da177e4SLinus Torvalds } else { 5431da177e4SLinus Torvalds /* 5448bc3be27SFengguang Wu * somehow blocked: retry later 5458bc3be27SFengguang Wu */ 5468bc3be27SFengguang Wu redirty_tail(inode); 5478bc3be27SFengguang Wu } 5488bc3be27SFengguang Wu } else { 5498bc3be27SFengguang Wu /* 5501da177e4SLinus Torvalds * Otherwise fully redirty the inode so that 5511da177e4SLinus Torvalds * other inodes on this superblock will get some 5521da177e4SLinus Torvalds * writeout. Otherwise heavy writing to one 5531da177e4SLinus Torvalds * file would indefinitely suspend writeout of 5541da177e4SLinus Torvalds * all the other files. 5551da177e4SLinus Torvalds */ 5561da177e4SLinus Torvalds inode->i_state |= I_DIRTY_PAGES; 5571b43ef91SAndrew Morton redirty_tail(inode); 5581da177e4SLinus Torvalds } 5591da177e4SLinus Torvalds } else if (atomic_read(&inode->i_count)) { 5601da177e4SLinus Torvalds /* 5611da177e4SLinus Torvalds * The inode is clean, inuse 5621da177e4SLinus Torvalds */ 5631da177e4SLinus Torvalds list_move(&inode->i_list, &inode_in_use); 5641da177e4SLinus Torvalds } else { 5651da177e4SLinus Torvalds /* 5661da177e4SLinus Torvalds * The inode is clean, unused 5671da177e4SLinus Torvalds */ 5681da177e4SLinus Torvalds list_move(&inode->i_list, &inode_unused); 5691da177e4SLinus Torvalds } 5701da177e4SLinus Torvalds } 5711c0eeaf5SJoern Engel inode_sync_complete(inode); 5721da177e4SLinus Torvalds return ret; 5731da177e4SLinus Torvalds } 5741da177e4SLinus Torvalds 575f11c9c5cSEdward Shishkin static void unpin_sb_for_writeback(struct super_block *sb) 5769ecc2738SJens Axboe { 5779ecc2738SJens Axboe up_read(&sb->s_umount); 5789ecc2738SJens Axboe put_super(sb); 5799ecc2738SJens Axboe } 580f11c9c5cSEdward Shishkin 581f11c9c5cSEdward Shishkin enum sb_pin_state { 582f11c9c5cSEdward Shishkin SB_PINNED, 583f11c9c5cSEdward Shishkin SB_NOT_PINNED, 584f11c9c5cSEdward Shishkin SB_PIN_FAILED 585f11c9c5cSEdward Shishkin }; 5869ecc2738SJens Axboe 58703ba3782SJens Axboe /* 58803ba3782SJens Axboe * For WB_SYNC_NONE writeback, the caller does not have the sb pinned 58903ba3782SJens Axboe * before calling writeback. So make sure that we do pin it, so it doesn't 59003ba3782SJens Axboe * go away while we are writing inodes from it. 59103ba3782SJens Axboe */ 592f11c9c5cSEdward Shishkin static enum sb_pin_state pin_sb_for_writeback(struct writeback_control *wbc, 593f11c9c5cSEdward Shishkin struct super_block *sb) 5941da177e4SLinus Torvalds { 5959ecc2738SJens Axboe /* 59603ba3782SJens Axboe * Caller must already hold the ref for this 59703ba3782SJens Axboe */ 598e913fc82SJens Axboe if (wbc->sync_mode == WB_SYNC_ALL || wbc->sb_pinned) { 59903ba3782SJens Axboe WARN_ON(!rwsem_is_locked(&sb->s_umount)); 600f11c9c5cSEdward Shishkin return SB_NOT_PINNED; 60103ba3782SJens Axboe } 60203ba3782SJens Axboe spin_lock(&sb_lock); 60303ba3782SJens Axboe sb->s_count++; 60403ba3782SJens Axboe if (down_read_trylock(&sb->s_umount)) { 60503ba3782SJens Axboe if (sb->s_root) { 60603ba3782SJens Axboe spin_unlock(&sb_lock); 607f11c9c5cSEdward Shishkin return SB_PINNED; 60803ba3782SJens Axboe } 60903ba3782SJens Axboe /* 61003ba3782SJens Axboe * umounted, drop rwsem again and fall through to failure 61103ba3782SJens Axboe */ 61203ba3782SJens Axboe up_read(&sb->s_umount); 61303ba3782SJens Axboe } 61403ba3782SJens Axboe sb->s_count--; 61503ba3782SJens Axboe spin_unlock(&sb_lock); 616f11c9c5cSEdward Shishkin return SB_PIN_FAILED; 61703ba3782SJens Axboe } 61803ba3782SJens Axboe 619f11c9c5cSEdward Shishkin /* 620f11c9c5cSEdward Shishkin * Write a portion of b_io inodes which belong to @sb. 621f11c9c5cSEdward Shishkin * If @wbc->sb != NULL, then find and write all such 622f11c9c5cSEdward Shishkin * inodes. Otherwise write only ones which go sequentially 623f11c9c5cSEdward Shishkin * in reverse order. 624f11c9c5cSEdward Shishkin * Return 1, if the caller writeback routine should be 625f11c9c5cSEdward Shishkin * interrupted. Otherwise return 0. 626f11c9c5cSEdward Shishkin */ 627f11c9c5cSEdward Shishkin static int writeback_sb_inodes(struct super_block *sb, 628f11c9c5cSEdward Shishkin struct bdi_writeback *wb, 62903ba3782SJens Axboe struct writeback_control *wbc) 63003ba3782SJens Axboe { 63103ba3782SJens Axboe while (!list_empty(&wb->b_io)) { 632f11c9c5cSEdward Shishkin long pages_skipped; 63303ba3782SJens Axboe struct inode *inode = list_entry(wb->b_io.prev, 6341da177e4SLinus Torvalds struct inode, i_list); 635f11c9c5cSEdward Shishkin if (wbc->sb && sb != inode->i_sb) { 636f11c9c5cSEdward Shishkin /* super block given and doesn't 637f11c9c5cSEdward Shishkin match, skip this inode */ 63866f3b8e2SJens Axboe redirty_tail(inode); 63966f3b8e2SJens Axboe continue; 64066f3b8e2SJens Axboe } 641f11c9c5cSEdward Shishkin if (sb != inode->i_sb) 642f11c9c5cSEdward Shishkin /* finish with this superblock */ 643f11c9c5cSEdward Shishkin return 0; 64484a89245SWu Fengguang if (inode->i_state & (I_NEW | I_WILL_FREE)) { 6457ef0d737SNick Piggin requeue_io(inode); 6467ef0d737SNick Piggin continue; 6477ef0d737SNick Piggin } 648d2caa3c5SJeff Layton /* 649d2caa3c5SJeff Layton * Was this inode dirtied after sync_sb_inodes was called? 650d2caa3c5SJeff Layton * This keeps sync from extra jobs and livelock. 651d2caa3c5SJeff Layton */ 652f11c9c5cSEdward Shishkin if (inode_dirtied_after(inode, wbc->wb_start)) 653f11c9c5cSEdward Shishkin return 1; 6541da177e4SLinus Torvalds 65584a89245SWu Fengguang BUG_ON(inode->i_state & (I_FREEING | I_CLEAR)); 6561da177e4SLinus Torvalds __iget(inode); 6571da177e4SLinus Torvalds pages_skipped = wbc->pages_skipped; 65801c03194SChristoph Hellwig writeback_single_inode(inode, wbc); 6591da177e4SLinus Torvalds if (wbc->pages_skipped != pages_skipped) { 6601da177e4SLinus Torvalds /* 6611da177e4SLinus Torvalds * writeback is not making progress due to locked 6621da177e4SLinus Torvalds * buffers. Skip this inode for now. 6631da177e4SLinus Torvalds */ 664f57b9b7bSAndrew Morton redirty_tail(inode); 6651da177e4SLinus Torvalds } 6661da177e4SLinus Torvalds spin_unlock(&inode_lock); 6671da177e4SLinus Torvalds iput(inode); 6684ffc8444SOGAWA Hirofumi cond_resched(); 6691da177e4SLinus Torvalds spin_lock(&inode_lock); 6708bc3be27SFengguang Wu if (wbc->nr_to_write <= 0) { 6718bc3be27SFengguang Wu wbc->more_io = 1; 672f11c9c5cSEdward Shishkin return 1; 6731da177e4SLinus Torvalds } 67403ba3782SJens Axboe if (!list_empty(&wb->b_more_io)) 6758bc3be27SFengguang Wu wbc->more_io = 1; 6768bc3be27SFengguang Wu } 677f11c9c5cSEdward Shishkin /* b_io is empty */ 678f11c9c5cSEdward Shishkin return 1; 679f11c9c5cSEdward Shishkin } 68038f21977SNick Piggin 681f11c9c5cSEdward Shishkin static void writeback_inodes_wb(struct bdi_writeback *wb, 682f11c9c5cSEdward Shishkin struct writeback_control *wbc) 683f11c9c5cSEdward Shishkin { 684f11c9c5cSEdward Shishkin int ret = 0; 6859ecc2738SJens Axboe 686f11c9c5cSEdward Shishkin wbc->wb_start = jiffies; /* livelock avoidance */ 687f11c9c5cSEdward Shishkin spin_lock(&inode_lock); 688f11c9c5cSEdward Shishkin if (!wbc->for_kupdate || list_empty(&wb->b_io)) 689f11c9c5cSEdward Shishkin queue_io(wb, wbc->older_than_this); 690f11c9c5cSEdward Shishkin 691f11c9c5cSEdward Shishkin while (!list_empty(&wb->b_io)) { 692f11c9c5cSEdward Shishkin struct inode *inode = list_entry(wb->b_io.prev, 693f11c9c5cSEdward Shishkin struct inode, i_list); 694f11c9c5cSEdward Shishkin struct super_block *sb = inode->i_sb; 695f11c9c5cSEdward Shishkin enum sb_pin_state state; 696f11c9c5cSEdward Shishkin 697f11c9c5cSEdward Shishkin if (wbc->sb && sb != wbc->sb) { 698f11c9c5cSEdward Shishkin /* super block given and doesn't 699f11c9c5cSEdward Shishkin match, skip this inode */ 700f11c9c5cSEdward Shishkin redirty_tail(inode); 701f11c9c5cSEdward Shishkin continue; 702f11c9c5cSEdward Shishkin } 703f11c9c5cSEdward Shishkin state = pin_sb_for_writeback(wbc, sb); 704f11c9c5cSEdward Shishkin 705f11c9c5cSEdward Shishkin if (state == SB_PIN_FAILED) { 706f11c9c5cSEdward Shishkin requeue_io(inode); 707f11c9c5cSEdward Shishkin continue; 708f11c9c5cSEdward Shishkin } 709f11c9c5cSEdward Shishkin ret = writeback_sb_inodes(sb, wb, wbc); 710f11c9c5cSEdward Shishkin 711f11c9c5cSEdward Shishkin if (state == SB_PINNED) 712f11c9c5cSEdward Shishkin unpin_sb_for_writeback(sb); 713f11c9c5cSEdward Shishkin if (ret) 714f11c9c5cSEdward Shishkin break; 715f11c9c5cSEdward Shishkin } 71666f3b8e2SJens Axboe spin_unlock(&inode_lock); 71766f3b8e2SJens Axboe /* Leave any unwritten inodes on b_io */ 71866f3b8e2SJens Axboe } 71966f3b8e2SJens Axboe 72003ba3782SJens Axboe void writeback_inodes_wbc(struct writeback_control *wbc) 72103ba3782SJens Axboe { 72203ba3782SJens Axboe struct backing_dev_info *bdi = wbc->bdi; 72303ba3782SJens Axboe 72403ba3782SJens Axboe writeback_inodes_wb(&bdi->wb, wbc); 72503ba3782SJens Axboe } 72603ba3782SJens Axboe 72703ba3782SJens Axboe /* 72803ba3782SJens Axboe * The maximum number of pages to writeout in a single bdi flush/kupdate 72903ba3782SJens Axboe * operation. We do this so we don't hold I_SYNC against an inode for 73003ba3782SJens Axboe * enormous amounts of time, which would block a userspace task which has 73103ba3782SJens Axboe * been forced to throttle against that inode. Also, the code reevaluates 73203ba3782SJens Axboe * the dirty each time it has written this many pages. 73303ba3782SJens Axboe */ 73403ba3782SJens Axboe #define MAX_WRITEBACK_PAGES 1024 73503ba3782SJens Axboe 73603ba3782SJens Axboe static inline bool over_bground_thresh(void) 73703ba3782SJens Axboe { 73803ba3782SJens Axboe unsigned long background_thresh, dirty_thresh; 73903ba3782SJens Axboe 74003ba3782SJens Axboe get_dirty_limits(&background_thresh, &dirty_thresh, NULL, NULL); 74103ba3782SJens Axboe 74203ba3782SJens Axboe return (global_page_state(NR_FILE_DIRTY) + 74303ba3782SJens Axboe global_page_state(NR_UNSTABLE_NFS) >= background_thresh); 74403ba3782SJens Axboe } 74503ba3782SJens Axboe 74603ba3782SJens Axboe /* 74703ba3782SJens Axboe * Explicit flushing or periodic writeback of "old" data. 74803ba3782SJens Axboe * 74903ba3782SJens Axboe * Define "old": the first time one of an inode's pages is dirtied, we mark the 75003ba3782SJens Axboe * dirtying-time in the inode's address_space. So this periodic writeback code 75103ba3782SJens Axboe * just walks the superblock inode list, writing back any inodes which are 75203ba3782SJens Axboe * older than a specific point in time. 75303ba3782SJens Axboe * 75403ba3782SJens Axboe * Try to run once per dirty_writeback_interval. But if a writeback event 75503ba3782SJens Axboe * takes longer than a dirty_writeback_interval interval, then leave a 75603ba3782SJens Axboe * one-second gap. 75703ba3782SJens Axboe * 75803ba3782SJens Axboe * older_than_this takes precedence over nr_to_write. So we'll only write back 75903ba3782SJens Axboe * all dirty pages if they are all attached to "old" mappings. 76003ba3782SJens Axboe */ 761c4a77a6cSJens Axboe static long wb_writeback(struct bdi_writeback *wb, 762c4a77a6cSJens Axboe struct wb_writeback_args *args) 76303ba3782SJens Axboe { 76403ba3782SJens Axboe struct writeback_control wbc = { 76503ba3782SJens Axboe .bdi = wb->bdi, 766c4a77a6cSJens Axboe .sb = args->sb, 767c4a77a6cSJens Axboe .sync_mode = args->sync_mode, 76803ba3782SJens Axboe .older_than_this = NULL, 769c4a77a6cSJens Axboe .for_kupdate = args->for_kupdate, 770b17621feSWu Fengguang .for_background = args->for_background, 771c4a77a6cSJens Axboe .range_cyclic = args->range_cyclic, 772e913fc82SJens Axboe .sb_pinned = args->sb_pinned, 77303ba3782SJens Axboe }; 77403ba3782SJens Axboe unsigned long oldest_jif; 77503ba3782SJens Axboe long wrote = 0; 776a5989bdcSJan Kara struct inode *inode; 77703ba3782SJens Axboe 77803ba3782SJens Axboe if (wbc.for_kupdate) { 77903ba3782SJens Axboe wbc.older_than_this = &oldest_jif; 78003ba3782SJens Axboe oldest_jif = jiffies - 78103ba3782SJens Axboe msecs_to_jiffies(dirty_expire_interval * 10); 78203ba3782SJens Axboe } 783c4a77a6cSJens Axboe if (!wbc.range_cyclic) { 784c4a77a6cSJens Axboe wbc.range_start = 0; 785c4a77a6cSJens Axboe wbc.range_end = LLONG_MAX; 786c4a77a6cSJens Axboe } 78703ba3782SJens Axboe 78803ba3782SJens Axboe for (;;) { 78903ba3782SJens Axboe /* 790d3ddec76SWu Fengguang * Stop writeback when nr_pages has been consumed 79103ba3782SJens Axboe */ 792d3ddec76SWu Fengguang if (args->nr_pages <= 0) 79303ba3782SJens Axboe break; 79403ba3782SJens Axboe 79503ba3782SJens Axboe /* 796d3ddec76SWu Fengguang * For background writeout, stop when we are below the 797d3ddec76SWu Fengguang * background dirty threshold 79803ba3782SJens Axboe */ 799d3ddec76SWu Fengguang if (args->for_background && !over_bground_thresh()) 80003ba3782SJens Axboe break; 80103ba3782SJens Axboe 80203ba3782SJens Axboe wbc.more_io = 0; 80303ba3782SJens Axboe wbc.nr_to_write = MAX_WRITEBACK_PAGES; 80403ba3782SJens Axboe wbc.pages_skipped = 0; 80503ba3782SJens Axboe writeback_inodes_wb(wb, &wbc); 806c4a77a6cSJens Axboe args->nr_pages -= MAX_WRITEBACK_PAGES - wbc.nr_to_write; 80703ba3782SJens Axboe wrote += MAX_WRITEBACK_PAGES - wbc.nr_to_write; 80803ba3782SJens Axboe 80903ba3782SJens Axboe /* 81071fd05a8SJens Axboe * If we consumed everything, see if we have more 81103ba3782SJens Axboe */ 81271fd05a8SJens Axboe if (wbc.nr_to_write <= 0) 81371fd05a8SJens Axboe continue; 81471fd05a8SJens Axboe /* 81571fd05a8SJens Axboe * Didn't write everything and we don't have more IO, bail 81671fd05a8SJens Axboe */ 81771fd05a8SJens Axboe if (!wbc.more_io) 81871fd05a8SJens Axboe break; 81971fd05a8SJens Axboe /* 82071fd05a8SJens Axboe * Did we write something? Try for more 82171fd05a8SJens Axboe */ 822a5989bdcSJan Kara if (wbc.nr_to_write < MAX_WRITEBACK_PAGES) 82303ba3782SJens Axboe continue; 824a5989bdcSJan Kara /* 825a5989bdcSJan Kara * Nothing written. Wait for some inode to 826a5989bdcSJan Kara * become available for writeback. Otherwise 827a5989bdcSJan Kara * we'll just busyloop. 828a5989bdcSJan Kara */ 829a5989bdcSJan Kara spin_lock(&inode_lock); 830a5989bdcSJan Kara if (!list_empty(&wb->b_more_io)) { 83171fd05a8SJens Axboe inode = list_entry(wb->b_more_io.prev, 832a5989bdcSJan Kara struct inode, i_list); 833a5989bdcSJan Kara inode_wait_for_writeback(inode); 834a5989bdcSJan Kara } 835a5989bdcSJan Kara spin_unlock(&inode_lock); 83603ba3782SJens Axboe } 83703ba3782SJens Axboe 83803ba3782SJens Axboe return wrote; 83903ba3782SJens Axboe } 84003ba3782SJens Axboe 84103ba3782SJens Axboe /* 84203ba3782SJens Axboe * Return the next bdi_work struct that hasn't been processed by this 8438010c3b6SJens Axboe * wb thread yet. ->seen is initially set for each thread that exists 8448010c3b6SJens Axboe * for this device, when a thread first notices a piece of work it 8458010c3b6SJens Axboe * clears its bit. Depending on writeback type, the thread will notify 8468010c3b6SJens Axboe * completion on either receiving the work (WB_SYNC_NONE) or after 8478010c3b6SJens Axboe * it is done (WB_SYNC_ALL). 84803ba3782SJens Axboe */ 84903ba3782SJens Axboe static struct bdi_work *get_next_work_item(struct backing_dev_info *bdi, 85003ba3782SJens Axboe struct bdi_writeback *wb) 85103ba3782SJens Axboe { 85203ba3782SJens Axboe struct bdi_work *work, *ret = NULL; 85303ba3782SJens Axboe 85403ba3782SJens Axboe rcu_read_lock(); 85503ba3782SJens Axboe 85603ba3782SJens Axboe list_for_each_entry_rcu(work, &bdi->work_list, list) { 85777fad5e6SNick Piggin if (!test_bit(wb->nr, &work->seen)) 85803ba3782SJens Axboe continue; 85977fad5e6SNick Piggin clear_bit(wb->nr, &work->seen); 86003ba3782SJens Axboe 86103ba3782SJens Axboe ret = work; 86203ba3782SJens Axboe break; 86303ba3782SJens Axboe } 86403ba3782SJens Axboe 86503ba3782SJens Axboe rcu_read_unlock(); 86603ba3782SJens Axboe return ret; 86703ba3782SJens Axboe } 86803ba3782SJens Axboe 86903ba3782SJens Axboe static long wb_check_old_data_flush(struct bdi_writeback *wb) 87003ba3782SJens Axboe { 87103ba3782SJens Axboe unsigned long expired; 87203ba3782SJens Axboe long nr_pages; 87303ba3782SJens Axboe 87469b62d01SJens Axboe /* 87569b62d01SJens Axboe * When set to zero, disable periodic writeback 87669b62d01SJens Axboe */ 87769b62d01SJens Axboe if (!dirty_writeback_interval) 87869b62d01SJens Axboe return 0; 87969b62d01SJens Axboe 88003ba3782SJens Axboe expired = wb->last_old_flush + 88103ba3782SJens Axboe msecs_to_jiffies(dirty_writeback_interval * 10); 88203ba3782SJens Axboe if (time_before(jiffies, expired)) 88303ba3782SJens Axboe return 0; 88403ba3782SJens Axboe 88503ba3782SJens Axboe wb->last_old_flush = jiffies; 88603ba3782SJens Axboe nr_pages = global_page_state(NR_FILE_DIRTY) + 88703ba3782SJens Axboe global_page_state(NR_UNSTABLE_NFS) + 88803ba3782SJens Axboe (inodes_stat.nr_inodes - inodes_stat.nr_unused); 88903ba3782SJens Axboe 890c4a77a6cSJens Axboe if (nr_pages) { 891c4a77a6cSJens Axboe struct wb_writeback_args args = { 892c4a77a6cSJens Axboe .nr_pages = nr_pages, 893c4a77a6cSJens Axboe .sync_mode = WB_SYNC_NONE, 894c4a77a6cSJens Axboe .for_kupdate = 1, 895c4a77a6cSJens Axboe .range_cyclic = 1, 896c4a77a6cSJens Axboe }; 897c4a77a6cSJens Axboe 898c4a77a6cSJens Axboe return wb_writeback(wb, &args); 899c4a77a6cSJens Axboe } 90003ba3782SJens Axboe 90103ba3782SJens Axboe return 0; 90203ba3782SJens Axboe } 90303ba3782SJens Axboe 90403ba3782SJens Axboe /* 90503ba3782SJens Axboe * Retrieve work items and do the writeback they describe 90603ba3782SJens Axboe */ 90703ba3782SJens Axboe long wb_do_writeback(struct bdi_writeback *wb, int force_wait) 90803ba3782SJens Axboe { 90903ba3782SJens Axboe struct backing_dev_info *bdi = wb->bdi; 91003ba3782SJens Axboe struct bdi_work *work; 911c4a77a6cSJens Axboe long wrote = 0; 91203ba3782SJens Axboe 91303ba3782SJens Axboe while ((work = get_next_work_item(bdi, wb)) != NULL) { 914c4a77a6cSJens Axboe struct wb_writeback_args args = work->args; 9157c8a3554SJens Axboe int post_clear; 91603ba3782SJens Axboe 91703ba3782SJens Axboe /* 91803ba3782SJens Axboe * Override sync mode, in case we must wait for completion 91903ba3782SJens Axboe */ 92003ba3782SJens Axboe if (force_wait) 921c4a77a6cSJens Axboe work->args.sync_mode = args.sync_mode = WB_SYNC_ALL; 92203ba3782SJens Axboe 9237c8a3554SJens Axboe post_clear = WB_SYNC_ALL || args.sb_pinned; 9247c8a3554SJens Axboe 92503ba3782SJens Axboe /* 92603ba3782SJens Axboe * If this isn't a data integrity operation, just notify 92703ba3782SJens Axboe * that we have seen this work and we are now starting it. 92803ba3782SJens Axboe */ 9297c8a3554SJens Axboe if (!post_clear) 93003ba3782SJens Axboe wb_clear_pending(wb, work); 93103ba3782SJens Axboe 932c4a77a6cSJens Axboe wrote += wb_writeback(wb, &args); 93303ba3782SJens Axboe 93403ba3782SJens Axboe /* 93503ba3782SJens Axboe * This is a data integrity writeback, so only do the 93603ba3782SJens Axboe * notification when we have completed the work. 93703ba3782SJens Axboe */ 9387c8a3554SJens Axboe if (post_clear) 93903ba3782SJens Axboe wb_clear_pending(wb, work); 94003ba3782SJens Axboe } 94103ba3782SJens Axboe 94203ba3782SJens Axboe /* 94303ba3782SJens Axboe * Check for periodic writeback, kupdated() style 94403ba3782SJens Axboe */ 94503ba3782SJens Axboe wrote += wb_check_old_data_flush(wb); 94603ba3782SJens Axboe 94703ba3782SJens Axboe return wrote; 94803ba3782SJens Axboe } 94903ba3782SJens Axboe 95003ba3782SJens Axboe /* 95103ba3782SJens Axboe * Handle writeback of dirty data for the device backed by this bdi. Also 95203ba3782SJens Axboe * wakes up periodically and does kupdated style flushing. 95303ba3782SJens Axboe */ 95403ba3782SJens Axboe int bdi_writeback_task(struct bdi_writeback *wb) 95503ba3782SJens Axboe { 95603ba3782SJens Axboe unsigned long last_active = jiffies; 95703ba3782SJens Axboe unsigned long wait_jiffies = -1UL; 95803ba3782SJens Axboe long pages_written; 95903ba3782SJens Axboe 96003ba3782SJens Axboe while (!kthread_should_stop()) { 96103ba3782SJens Axboe pages_written = wb_do_writeback(wb, 0); 96203ba3782SJens Axboe 96303ba3782SJens Axboe if (pages_written) 96403ba3782SJens Axboe last_active = jiffies; 96503ba3782SJens Axboe else if (wait_jiffies != -1UL) { 96603ba3782SJens Axboe unsigned long max_idle; 96703ba3782SJens Axboe 96803ba3782SJens Axboe /* 96903ba3782SJens Axboe * Longest period of inactivity that we tolerate. If we 97003ba3782SJens Axboe * see dirty data again later, the task will get 97103ba3782SJens Axboe * recreated automatically. 97203ba3782SJens Axboe */ 97303ba3782SJens Axboe max_idle = max(5UL * 60 * HZ, wait_jiffies); 97403ba3782SJens Axboe if (time_after(jiffies, max_idle + last_active)) 97503ba3782SJens Axboe break; 97603ba3782SJens Axboe } 97703ba3782SJens Axboe 97869b62d01SJens Axboe if (dirty_writeback_interval) { 97903ba3782SJens Axboe wait_jiffies = msecs_to_jiffies(dirty_writeback_interval * 10); 98049db0414SJens Axboe schedule_timeout_interruptible(wait_jiffies); 981f9eadbbdSJens Axboe } else { 982f9eadbbdSJens Axboe set_current_state(TASK_INTERRUPTIBLE); 983f9eadbbdSJens Axboe if (list_empty_careful(&wb->bdi->work_list) && 984f9eadbbdSJens Axboe !kthread_should_stop()) 98569b62d01SJens Axboe schedule(); 986f9eadbbdSJens Axboe __set_current_state(TASK_RUNNING); 987f9eadbbdSJens Axboe } 98869b62d01SJens Axboe 98903ba3782SJens Axboe try_to_freeze(); 99003ba3782SJens Axboe } 99103ba3782SJens Axboe 99203ba3782SJens Axboe return 0; 99303ba3782SJens Axboe } 99403ba3782SJens Axboe 99503ba3782SJens Axboe /* 996b6e51316SJens Axboe * Schedule writeback for all backing devices. This does WB_SYNC_NONE 997b6e51316SJens Axboe * writeback, for integrity writeback see bdi_sync_writeback(). 99803ba3782SJens Axboe */ 999b6e51316SJens Axboe static void bdi_writeback_all(struct super_block *sb, long nr_pages) 100003ba3782SJens Axboe { 1001b6e51316SJens Axboe struct wb_writeback_args args = { 1002b6e51316SJens Axboe .sb = sb, 1003b6e51316SJens Axboe .nr_pages = nr_pages, 1004b6e51316SJens Axboe .sync_mode = WB_SYNC_NONE, 1005b6e51316SJens Axboe }; 100603ba3782SJens Axboe struct backing_dev_info *bdi; 100703ba3782SJens Axboe 1008cfc4ba53SJens Axboe rcu_read_lock(); 100903ba3782SJens Axboe 1010cfc4ba53SJens Axboe list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { 101103ba3782SJens Axboe if (!bdi_has_dirty_io(bdi)) 101203ba3782SJens Axboe continue; 101303ba3782SJens Axboe 10147c8a3554SJens Axboe bdi_alloc_queue_work(bdi, &args, 0); 101503ba3782SJens Axboe } 101603ba3782SJens Axboe 1017cfc4ba53SJens Axboe rcu_read_unlock(); 101803ba3782SJens Axboe } 101903ba3782SJens Axboe 102003ba3782SJens Axboe /* 102103ba3782SJens Axboe * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back 102203ba3782SJens Axboe * the whole world. 102303ba3782SJens Axboe */ 102403ba3782SJens Axboe void wakeup_flusher_threads(long nr_pages) 102503ba3782SJens Axboe { 102603ba3782SJens Axboe if (nr_pages == 0) 102703ba3782SJens Axboe nr_pages = global_page_state(NR_FILE_DIRTY) + 102803ba3782SJens Axboe global_page_state(NR_UNSTABLE_NFS); 1029b6e51316SJens Axboe bdi_writeback_all(NULL, nr_pages); 103003ba3782SJens Axboe } 103103ba3782SJens Axboe 103203ba3782SJens Axboe static noinline void block_dump___mark_inode_dirty(struct inode *inode) 103303ba3782SJens Axboe { 103403ba3782SJens Axboe if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { 103503ba3782SJens Axboe struct dentry *dentry; 103603ba3782SJens Axboe const char *name = "?"; 103703ba3782SJens Axboe 103803ba3782SJens Axboe dentry = d_find_alias(inode); 103903ba3782SJens Axboe if (dentry) { 104003ba3782SJens Axboe spin_lock(&dentry->d_lock); 104103ba3782SJens Axboe name = (const char *) dentry->d_name.name; 104203ba3782SJens Axboe } 104303ba3782SJens Axboe printk(KERN_DEBUG 104403ba3782SJens Axboe "%s(%d): dirtied inode %lu (%s) on %s\n", 104503ba3782SJens Axboe current->comm, task_pid_nr(current), inode->i_ino, 104603ba3782SJens Axboe name, inode->i_sb->s_id); 104703ba3782SJens Axboe if (dentry) { 104803ba3782SJens Axboe spin_unlock(&dentry->d_lock); 104903ba3782SJens Axboe dput(dentry); 105003ba3782SJens Axboe } 105103ba3782SJens Axboe } 105203ba3782SJens Axboe } 105303ba3782SJens Axboe 105403ba3782SJens Axboe /** 105503ba3782SJens Axboe * __mark_inode_dirty - internal function 105603ba3782SJens Axboe * @inode: inode to mark 105703ba3782SJens Axboe * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) 105803ba3782SJens Axboe * Mark an inode as dirty. Callers should use mark_inode_dirty or 105903ba3782SJens Axboe * mark_inode_dirty_sync. 106003ba3782SJens Axboe * 106103ba3782SJens Axboe * Put the inode on the super block's dirty list. 106203ba3782SJens Axboe * 106303ba3782SJens Axboe * CAREFUL! We mark it dirty unconditionally, but move it onto the 106403ba3782SJens Axboe * dirty list only if it is hashed or if it refers to a blockdev. 106503ba3782SJens Axboe * If it was not hashed, it will never be added to the dirty list 106603ba3782SJens Axboe * even if it is later hashed, as it will have been marked dirty already. 106703ba3782SJens Axboe * 106803ba3782SJens Axboe * In short, make sure you hash any inodes _before_ you start marking 106903ba3782SJens Axboe * them dirty. 107003ba3782SJens Axboe * 107103ba3782SJens Axboe * This function *must* be atomic for the I_DIRTY_PAGES case - 107203ba3782SJens Axboe * set_page_dirty() is called under spinlock in several places. 107303ba3782SJens Axboe * 107403ba3782SJens Axboe * Note that for blockdevs, inode->dirtied_when represents the dirtying time of 107503ba3782SJens Axboe * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of 107603ba3782SJens Axboe * the kernel-internal blockdev inode represents the dirtying time of the 107703ba3782SJens Axboe * blockdev's pages. This is why for I_DIRTY_PAGES we always use 107803ba3782SJens Axboe * page->mapping->host, so the page-dirtying time is recorded in the internal 107903ba3782SJens Axboe * blockdev inode. 108003ba3782SJens Axboe */ 108103ba3782SJens Axboe void __mark_inode_dirty(struct inode *inode, int flags) 108203ba3782SJens Axboe { 108303ba3782SJens Axboe struct super_block *sb = inode->i_sb; 108403ba3782SJens Axboe 108503ba3782SJens Axboe /* 108603ba3782SJens Axboe * Don't do this for I_DIRTY_PAGES - that doesn't actually 108703ba3782SJens Axboe * dirty the inode itself 108803ba3782SJens Axboe */ 108903ba3782SJens Axboe if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { 109003ba3782SJens Axboe if (sb->s_op->dirty_inode) 109103ba3782SJens Axboe sb->s_op->dirty_inode(inode); 109203ba3782SJens Axboe } 109303ba3782SJens Axboe 109403ba3782SJens Axboe /* 109503ba3782SJens Axboe * make sure that changes are seen by all cpus before we test i_state 109603ba3782SJens Axboe * -- mikulas 109703ba3782SJens Axboe */ 109803ba3782SJens Axboe smp_mb(); 109903ba3782SJens Axboe 110003ba3782SJens Axboe /* avoid the locking if we can */ 110103ba3782SJens Axboe if ((inode->i_state & flags) == flags) 110203ba3782SJens Axboe return; 110303ba3782SJens Axboe 110403ba3782SJens Axboe if (unlikely(block_dump)) 110503ba3782SJens Axboe block_dump___mark_inode_dirty(inode); 110603ba3782SJens Axboe 110703ba3782SJens Axboe spin_lock(&inode_lock); 110803ba3782SJens Axboe if ((inode->i_state & flags) != flags) { 110903ba3782SJens Axboe const int was_dirty = inode->i_state & I_DIRTY; 111003ba3782SJens Axboe 111103ba3782SJens Axboe inode->i_state |= flags; 111203ba3782SJens Axboe 111303ba3782SJens Axboe /* 111403ba3782SJens Axboe * If the inode is being synced, just update its dirty state. 111503ba3782SJens Axboe * The unlocker will place the inode on the appropriate 111603ba3782SJens Axboe * superblock list, based upon its state. 111703ba3782SJens Axboe */ 111803ba3782SJens Axboe if (inode->i_state & I_SYNC) 111903ba3782SJens Axboe goto out; 112003ba3782SJens Axboe 112103ba3782SJens Axboe /* 112203ba3782SJens Axboe * Only add valid (hashed) inodes to the superblock's 112303ba3782SJens Axboe * dirty list. Add blockdev inodes as well. 112403ba3782SJens Axboe */ 112503ba3782SJens Axboe if (!S_ISBLK(inode->i_mode)) { 112603ba3782SJens Axboe if (hlist_unhashed(&inode->i_hash)) 112703ba3782SJens Axboe goto out; 112803ba3782SJens Axboe } 112903ba3782SJens Axboe if (inode->i_state & (I_FREEING|I_CLEAR)) 113003ba3782SJens Axboe goto out; 113103ba3782SJens Axboe 113203ba3782SJens Axboe /* 113303ba3782SJens Axboe * If the inode was already on b_dirty/b_io/b_more_io, don't 113403ba3782SJens Axboe * reposition it (that would break b_dirty time-ordering). 113503ba3782SJens Axboe */ 113603ba3782SJens Axboe if (!was_dirty) { 113703ba3782SJens Axboe struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; 1138500b067cSJens Axboe struct backing_dev_info *bdi = wb->bdi; 1139500b067cSJens Axboe 1140500b067cSJens Axboe if (bdi_cap_writeback_dirty(bdi) && 1141500b067cSJens Axboe !test_bit(BDI_registered, &bdi->state)) { 1142500b067cSJens Axboe WARN_ON(1); 1143500b067cSJens Axboe printk(KERN_ERR "bdi-%s not registered\n", 1144500b067cSJens Axboe bdi->name); 1145500b067cSJens Axboe } 114603ba3782SJens Axboe 114703ba3782SJens Axboe inode->dirtied_when = jiffies; 114803ba3782SJens Axboe list_move(&inode->i_list, &wb->b_dirty); 114903ba3782SJens Axboe } 115003ba3782SJens Axboe } 115103ba3782SJens Axboe out: 115203ba3782SJens Axboe spin_unlock(&inode_lock); 115303ba3782SJens Axboe } 115403ba3782SJens Axboe EXPORT_SYMBOL(__mark_inode_dirty); 115503ba3782SJens Axboe 115666f3b8e2SJens Axboe /* 115766f3b8e2SJens Axboe * Write out a superblock's list of dirty inodes. A wait will be performed 115866f3b8e2SJens Axboe * upon no inodes, all inodes or the final one, depending upon sync_mode. 115966f3b8e2SJens Axboe * 116066f3b8e2SJens Axboe * If older_than_this is non-NULL, then only write out inodes which 116166f3b8e2SJens Axboe * had their first dirtying at a time earlier than *older_than_this. 116266f3b8e2SJens Axboe * 116366f3b8e2SJens Axboe * If `bdi' is non-zero then we're being asked to writeback a specific queue. 116466f3b8e2SJens Axboe * This function assumes that the blockdev superblock's inodes are backed by 116566f3b8e2SJens Axboe * a variety of queues, so all inodes are searched. For other superblocks, 116666f3b8e2SJens Axboe * assume that all inodes are backed by the same queue. 116766f3b8e2SJens Axboe * 116866f3b8e2SJens Axboe * The inodes to be written are parked on bdi->b_io. They are moved back onto 116966f3b8e2SJens Axboe * bdi->b_dirty as they are selected for writing. This way, none can be missed 117066f3b8e2SJens Axboe * on the writer throttling path, and we get decent balancing between many 117166f3b8e2SJens Axboe * throttled threads: we don't want them all piling up on inode_sync_wait. 117266f3b8e2SJens Axboe */ 1173b6e51316SJens Axboe static void wait_sb_inodes(struct super_block *sb) 117466f3b8e2SJens Axboe { 117538f21977SNick Piggin struct inode *inode, *old_inode = NULL; 117638f21977SNick Piggin 117703ba3782SJens Axboe /* 117803ba3782SJens Axboe * We need to be protected against the filesystem going from 117903ba3782SJens Axboe * r/o to r/w or vice versa. 118003ba3782SJens Axboe */ 1181b6e51316SJens Axboe WARN_ON(!rwsem_is_locked(&sb->s_umount)); 118203ba3782SJens Axboe 118366f3b8e2SJens Axboe spin_lock(&inode_lock); 118466f3b8e2SJens Axboe 118538f21977SNick Piggin /* 118638f21977SNick Piggin * Data integrity sync. Must wait for all pages under writeback, 118738f21977SNick Piggin * because there may have been pages dirtied before our sync 118838f21977SNick Piggin * call, but which had writeout started before we write it out. 118938f21977SNick Piggin * In which case, the inode may not be on the dirty list, but 119038f21977SNick Piggin * we still have to wait for that writeout. 119138f21977SNick Piggin */ 1192b6e51316SJens Axboe list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { 119338f21977SNick Piggin struct address_space *mapping; 119438f21977SNick Piggin 119503ba3782SJens Axboe if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE|I_NEW)) 119638f21977SNick Piggin continue; 119738f21977SNick Piggin mapping = inode->i_mapping; 119838f21977SNick Piggin if (mapping->nrpages == 0) 119938f21977SNick Piggin continue; 120038f21977SNick Piggin __iget(inode); 1201ae8547b0SHans Reiser spin_unlock(&inode_lock); 120238f21977SNick Piggin /* 120338f21977SNick Piggin * We hold a reference to 'inode' so it couldn't have 120438f21977SNick Piggin * been removed from s_inodes list while we dropped the 120538f21977SNick Piggin * inode_lock. We cannot iput the inode now as we can 120638f21977SNick Piggin * be holding the last reference and we cannot iput it 120738f21977SNick Piggin * under inode_lock. So we keep the reference and iput 120838f21977SNick Piggin * it later. 120938f21977SNick Piggin */ 121038f21977SNick Piggin iput(old_inode); 121138f21977SNick Piggin old_inode = inode; 121238f21977SNick Piggin 121338f21977SNick Piggin filemap_fdatawait(mapping); 121438f21977SNick Piggin 121538f21977SNick Piggin cond_resched(); 121638f21977SNick Piggin 121738f21977SNick Piggin spin_lock(&inode_lock); 121838f21977SNick Piggin } 121938f21977SNick Piggin spin_unlock(&inode_lock); 122038f21977SNick Piggin iput(old_inode); 122166f3b8e2SJens Axboe } 12221da177e4SLinus Torvalds 1223e913fc82SJens Axboe static void __writeback_inodes_sb(struct super_block *sb, int sb_locked) 1224e913fc82SJens Axboe { 1225e913fc82SJens Axboe unsigned long nr_dirty = global_page_state(NR_FILE_DIRTY); 1226e913fc82SJens Axboe unsigned long nr_unstable = global_page_state(NR_UNSTABLE_NFS); 1227e913fc82SJens Axboe long nr_to_write; 1228e913fc82SJens Axboe 1229e913fc82SJens Axboe nr_to_write = nr_dirty + nr_unstable + 1230e913fc82SJens Axboe (inodes_stat.nr_inodes - inodes_stat.nr_unused); 1231e913fc82SJens Axboe 1232e913fc82SJens Axboe bdi_start_writeback(sb->s_bdi, sb, nr_to_write, sb_locked); 1233e913fc82SJens Axboe } 1234e913fc82SJens Axboe 1235d8a8559cSJens Axboe /** 1236d8a8559cSJens Axboe * writeback_inodes_sb - writeback dirty inodes from given super_block 1237d8a8559cSJens Axboe * @sb: the superblock 12381da177e4SLinus Torvalds * 1239d8a8559cSJens Axboe * Start writeback on some inodes on this super_block. No guarantees are made 1240d8a8559cSJens Axboe * on how many (if any) will be written, and this function does not wait 1241d8a8559cSJens Axboe * for IO completion of submitted IO. The number of pages submitted is 1242d8a8559cSJens Axboe * returned. 12431da177e4SLinus Torvalds */ 1244b6e51316SJens Axboe void writeback_inodes_sb(struct super_block *sb) 12451da177e4SLinus Torvalds { 1246e913fc82SJens Axboe __writeback_inodes_sb(sb, 0); 12471da177e4SLinus Torvalds } 1248d8a8559cSJens Axboe EXPORT_SYMBOL(writeback_inodes_sb); 1249d8a8559cSJens Axboe 1250d8a8559cSJens Axboe /** 1251e913fc82SJens Axboe * writeback_inodes_sb_locked - writeback dirty inodes from given super_block 1252e913fc82SJens Axboe * @sb: the superblock 1253e913fc82SJens Axboe * 1254e913fc82SJens Axboe * Like writeback_inodes_sb(), except the caller already holds the 1255e913fc82SJens Axboe * sb umount sem. 1256e913fc82SJens Axboe */ 1257e913fc82SJens Axboe void writeback_inodes_sb_locked(struct super_block *sb) 1258e913fc82SJens Axboe { 1259e913fc82SJens Axboe __writeback_inodes_sb(sb, 1); 1260e913fc82SJens Axboe } 1261e913fc82SJens Axboe 1262e913fc82SJens Axboe /** 126317bd55d0SEric Sandeen * writeback_inodes_sb_if_idle - start writeback if none underway 126417bd55d0SEric Sandeen * @sb: the superblock 126517bd55d0SEric Sandeen * 126617bd55d0SEric Sandeen * Invoke writeback_inodes_sb if no writeback is currently underway. 126717bd55d0SEric Sandeen * Returns 1 if writeback was started, 0 if not. 126817bd55d0SEric Sandeen */ 126917bd55d0SEric Sandeen int writeback_inodes_sb_if_idle(struct super_block *sb) 127017bd55d0SEric Sandeen { 127117bd55d0SEric Sandeen if (!writeback_in_progress(sb->s_bdi)) { 127217bd55d0SEric Sandeen writeback_inodes_sb(sb); 127317bd55d0SEric Sandeen return 1; 127417bd55d0SEric Sandeen } else 127517bd55d0SEric Sandeen return 0; 127617bd55d0SEric Sandeen } 127717bd55d0SEric Sandeen EXPORT_SYMBOL(writeback_inodes_sb_if_idle); 127817bd55d0SEric Sandeen 127917bd55d0SEric Sandeen /** 1280d8a8559cSJens Axboe * sync_inodes_sb - sync sb inode pages 1281d8a8559cSJens Axboe * @sb: the superblock 1282d8a8559cSJens Axboe * 1283d8a8559cSJens Axboe * This function writes and waits on any dirty inode belonging to this 1284d8a8559cSJens Axboe * super_block. The number of pages synced is returned. 1285d8a8559cSJens Axboe */ 1286b6e51316SJens Axboe void sync_inodes_sb(struct super_block *sb) 1287d8a8559cSJens Axboe { 1288b6e51316SJens Axboe bdi_sync_writeback(sb->s_bdi, sb); 1289b6e51316SJens Axboe wait_sb_inodes(sb); 1290d8a8559cSJens Axboe } 1291d8a8559cSJens Axboe EXPORT_SYMBOL(sync_inodes_sb); 12921da177e4SLinus Torvalds 12931da177e4SLinus Torvalds /** 12941da177e4SLinus Torvalds * write_inode_now - write an inode to disk 12951da177e4SLinus Torvalds * @inode: inode to write to disk 12961da177e4SLinus Torvalds * @sync: whether the write should be synchronous or not 12971da177e4SLinus Torvalds * 12987f04c26dSAndrea Arcangeli * This function commits an inode to disk immediately if it is dirty. This is 12997f04c26dSAndrea Arcangeli * primarily needed by knfsd. 13007f04c26dSAndrea Arcangeli * 13017f04c26dSAndrea Arcangeli * The caller must either have a ref on the inode or must have set I_WILL_FREE. 13021da177e4SLinus Torvalds */ 13031da177e4SLinus Torvalds int write_inode_now(struct inode *inode, int sync) 13041da177e4SLinus Torvalds { 13051da177e4SLinus Torvalds int ret; 13061da177e4SLinus Torvalds struct writeback_control wbc = { 13071da177e4SLinus Torvalds .nr_to_write = LONG_MAX, 130818914b18SMike Galbraith .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, 1309111ebb6eSOGAWA Hirofumi .range_start = 0, 1310111ebb6eSOGAWA Hirofumi .range_end = LLONG_MAX, 13111da177e4SLinus Torvalds }; 13121da177e4SLinus Torvalds 13131da177e4SLinus Torvalds if (!mapping_cap_writeback_dirty(inode->i_mapping)) 131449364ce2SAndrew Morton wbc.nr_to_write = 0; 13151da177e4SLinus Torvalds 13161da177e4SLinus Torvalds might_sleep(); 13171da177e4SLinus Torvalds spin_lock(&inode_lock); 131801c03194SChristoph Hellwig ret = writeback_single_inode(inode, &wbc); 13191da177e4SLinus Torvalds spin_unlock(&inode_lock); 13201da177e4SLinus Torvalds if (sync) 13211c0eeaf5SJoern Engel inode_sync_wait(inode); 13221da177e4SLinus Torvalds return ret; 13231da177e4SLinus Torvalds } 13241da177e4SLinus Torvalds EXPORT_SYMBOL(write_inode_now); 13251da177e4SLinus Torvalds 13261da177e4SLinus Torvalds /** 13271da177e4SLinus Torvalds * sync_inode - write an inode and its pages to disk. 13281da177e4SLinus Torvalds * @inode: the inode to sync 13291da177e4SLinus Torvalds * @wbc: controls the writeback mode 13301da177e4SLinus Torvalds * 13311da177e4SLinus Torvalds * sync_inode() will write an inode and its pages to disk. It will also 13321da177e4SLinus Torvalds * correctly update the inode on its superblock's dirty inode lists and will 13331da177e4SLinus Torvalds * update inode->i_state. 13341da177e4SLinus Torvalds * 13351da177e4SLinus Torvalds * The caller must have a ref on the inode. 13361da177e4SLinus Torvalds */ 13371da177e4SLinus Torvalds int sync_inode(struct inode *inode, struct writeback_control *wbc) 13381da177e4SLinus Torvalds { 13391da177e4SLinus Torvalds int ret; 13401da177e4SLinus Torvalds 13411da177e4SLinus Torvalds spin_lock(&inode_lock); 134201c03194SChristoph Hellwig ret = writeback_single_inode(inode, wbc); 13431da177e4SLinus Torvalds spin_unlock(&inode_lock); 13441da177e4SLinus Torvalds return ret; 13451da177e4SLinus Torvalds } 13461da177e4SLinus Torvalds EXPORT_SYMBOL(sync_inode); 1347