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> 191da177e4SLinus Torvalds #include <linux/sched.h> 201da177e4SLinus Torvalds #include <linux/fs.h> 211da177e4SLinus Torvalds #include <linux/mm.h> 2203ba3782SJens Axboe #include <linux/kthread.h> 2303ba3782SJens Axboe #include <linux/freezer.h> 241da177e4SLinus Torvalds #include <linux/writeback.h> 251da177e4SLinus Torvalds #include <linux/blkdev.h> 261da177e4SLinus Torvalds #include <linux/backing-dev.h> 271da177e4SLinus Torvalds #include <linux/buffer_head.h> 2807f3f05cSDavid Howells #include "internal.h" 291da177e4SLinus Torvalds 3066f3b8e2SJens Axboe #define inode_to_bdi(inode) ((inode)->i_mapping->backing_dev_info) 31f11b00f3SAdrian Bunk 3203ba3782SJens Axboe /* 33d0bceac7SJens Axboe * We don't actually have pdflush, but this one is exported though /proc... 34d0bceac7SJens Axboe */ 35d0bceac7SJens Axboe int nr_pdflush_threads; 36d0bceac7SJens Axboe 37d0bceac7SJens Axboe /* 38c4a77a6cSJens Axboe * Passed into wb_writeback(), essentially a subset of writeback_control 39c4a77a6cSJens Axboe */ 40c4a77a6cSJens Axboe struct wb_writeback_args { 41c4a77a6cSJens Axboe long nr_pages; 42c4a77a6cSJens Axboe struct super_block *sb; 43c4a77a6cSJens Axboe enum writeback_sync_modes sync_mode; 44d3ddec76SWu Fengguang int for_kupdate:1; 45d3ddec76SWu Fengguang int range_cyclic:1; 46d3ddec76SWu Fengguang int for_background:1; 47c4a77a6cSJens Axboe }; 48c4a77a6cSJens Axboe 49c4a77a6cSJens Axboe /* 5003ba3782SJens Axboe * Work items for the bdi_writeback threads 51f11b00f3SAdrian Bunk */ 5203ba3782SJens Axboe struct bdi_work { 538010c3b6SJens Axboe struct list_head list; /* pending work list */ 548010c3b6SJens Axboe struct rcu_head rcu_head; /* for RCU free/clear of work */ 5503ba3782SJens Axboe 568010c3b6SJens Axboe unsigned long seen; /* threads that have seen this work */ 578010c3b6SJens Axboe atomic_t pending; /* number of threads still to do work */ 5803ba3782SJens Axboe 598010c3b6SJens Axboe struct wb_writeback_args args; /* writeback arguments */ 6003ba3782SJens Axboe 618010c3b6SJens Axboe unsigned long state; /* flag bits, see WS_* */ 6203ba3782SJens Axboe }; 6303ba3782SJens Axboe 6403ba3782SJens Axboe enum { 6503ba3782SJens Axboe WS_USED_B = 0, 6603ba3782SJens Axboe WS_ONSTACK_B, 6703ba3782SJens Axboe }; 6803ba3782SJens Axboe 6903ba3782SJens Axboe #define WS_USED (1 << WS_USED_B) 7003ba3782SJens Axboe #define WS_ONSTACK (1 << WS_ONSTACK_B) 7103ba3782SJens Axboe 7203ba3782SJens Axboe static inline bool bdi_work_on_stack(struct bdi_work *work) 73f11b00f3SAdrian Bunk { 7403ba3782SJens Axboe return test_bit(WS_ONSTACK_B, &work->state); 7503ba3782SJens Axboe } 7603ba3782SJens Axboe 7703ba3782SJens Axboe static inline void bdi_work_init(struct bdi_work *work, 78b6e51316SJens Axboe struct wb_writeback_args *args) 7903ba3782SJens Axboe { 8003ba3782SJens Axboe INIT_RCU_HEAD(&work->rcu_head); 81b6e51316SJens Axboe work->args = *args; 8203ba3782SJens Axboe work->state = WS_USED; 8303ba3782SJens Axboe } 8403ba3782SJens Axboe 85f11b00f3SAdrian Bunk /** 86f11b00f3SAdrian Bunk * writeback_in_progress - determine whether there is writeback in progress 87f11b00f3SAdrian Bunk * @bdi: the device's backing_dev_info structure. 88f11b00f3SAdrian Bunk * 8903ba3782SJens Axboe * Determine whether there is writeback waiting to be handled against a 9003ba3782SJens Axboe * backing device. 91f11b00f3SAdrian Bunk */ 92f11b00f3SAdrian Bunk int writeback_in_progress(struct backing_dev_info *bdi) 93f11b00f3SAdrian Bunk { 9403ba3782SJens Axboe return !list_empty(&bdi->work_list); 95f11b00f3SAdrian Bunk } 96f11b00f3SAdrian Bunk 9703ba3782SJens Axboe static void bdi_work_clear(struct bdi_work *work) 98f11b00f3SAdrian Bunk { 9903ba3782SJens Axboe clear_bit(WS_USED_B, &work->state); 10003ba3782SJens Axboe smp_mb__after_clear_bit(); 1011ef7d9aaSNick Piggin /* 1021ef7d9aaSNick Piggin * work can have disappeared at this point. bit waitq functions 1031ef7d9aaSNick Piggin * should be able to tolerate this, provided bdi_sched_wait does 1041ef7d9aaSNick Piggin * not dereference it's pointer argument. 1051ef7d9aaSNick Piggin */ 10603ba3782SJens Axboe wake_up_bit(&work->state, WS_USED_B); 107f11b00f3SAdrian Bunk } 108f11b00f3SAdrian Bunk 10903ba3782SJens Axboe static void bdi_work_free(struct rcu_head *head) 1104195f73dSNick Piggin { 11103ba3782SJens Axboe struct bdi_work *work = container_of(head, struct bdi_work, rcu_head); 1124195f73dSNick Piggin 11303ba3782SJens Axboe if (!bdi_work_on_stack(work)) 11403ba3782SJens Axboe kfree(work); 11503ba3782SJens Axboe else 11603ba3782SJens Axboe bdi_work_clear(work); 1174195f73dSNick Piggin } 1184195f73dSNick Piggin 11903ba3782SJens Axboe static void wb_work_complete(struct bdi_work *work) 1201da177e4SLinus Torvalds { 121c4a77a6cSJens Axboe const enum writeback_sync_modes sync_mode = work->args.sync_mode; 12277b9d059SNick Piggin int onstack = bdi_work_on_stack(work); 1231da177e4SLinus Torvalds 1241da177e4SLinus Torvalds /* 12503ba3782SJens Axboe * For allocated work, we can clear the done/seen bit right here. 12603ba3782SJens Axboe * For on-stack work, we need to postpone both the clear and free 12703ba3782SJens Axboe * to after the RCU grace period, since the stack could be invalidated 12803ba3782SJens Axboe * as soon as bdi_work_clear() has done the wakeup. 1291da177e4SLinus Torvalds */ 13077b9d059SNick Piggin if (!onstack) 13103ba3782SJens Axboe bdi_work_clear(work); 13277b9d059SNick Piggin if (sync_mode == WB_SYNC_NONE || onstack) 13303ba3782SJens Axboe call_rcu(&work->rcu_head, bdi_work_free); 1341da177e4SLinus Torvalds } 1351da177e4SLinus Torvalds 13603ba3782SJens Axboe static void wb_clear_pending(struct bdi_writeback *wb, struct bdi_work *work) 13703ba3782SJens Axboe { 1381da177e4SLinus Torvalds /* 13903ba3782SJens Axboe * The caller has retrieved the work arguments from this work, 14003ba3782SJens Axboe * drop our reference. If this is the last ref, delete and free it 14103ba3782SJens Axboe */ 14203ba3782SJens Axboe if (atomic_dec_and_test(&work->pending)) { 14303ba3782SJens Axboe struct backing_dev_info *bdi = wb->bdi; 14403ba3782SJens Axboe 14503ba3782SJens Axboe spin_lock(&bdi->wb_lock); 14603ba3782SJens Axboe list_del_rcu(&work->list); 14703ba3782SJens Axboe spin_unlock(&bdi->wb_lock); 14803ba3782SJens Axboe 14903ba3782SJens Axboe wb_work_complete(work); 15003ba3782SJens Axboe } 15103ba3782SJens Axboe } 15203ba3782SJens Axboe 15303ba3782SJens Axboe static void bdi_queue_work(struct backing_dev_info *bdi, struct bdi_work *work) 15403ba3782SJens Axboe { 15503ba3782SJens Axboe work->seen = bdi->wb_mask; 15603ba3782SJens Axboe BUG_ON(!work->seen); 15703ba3782SJens Axboe atomic_set(&work->pending, bdi->wb_cnt); 15803ba3782SJens Axboe BUG_ON(!bdi->wb_cnt); 15903ba3782SJens Axboe 16003ba3782SJens Axboe /* 161deed62edSNick Piggin * list_add_tail_rcu() contains the necessary barriers to 162deed62edSNick Piggin * make sure the above stores are seen before the item is 163deed62edSNick Piggin * noticed on the list 1641da177e4SLinus Torvalds */ 16503ba3782SJens Axboe spin_lock(&bdi->wb_lock); 16603ba3782SJens Axboe list_add_tail_rcu(&work->list, &bdi->work_list); 16703ba3782SJens Axboe spin_unlock(&bdi->wb_lock); 1681da177e4SLinus Torvalds 1691da177e4SLinus Torvalds /* 17003ba3782SJens Axboe * If the default thread isn't there, make sure we add it. When 17103ba3782SJens Axboe * it gets created and wakes up, we'll run this work. 1721da177e4SLinus Torvalds */ 17303ba3782SJens Axboe if (unlikely(list_empty_careful(&bdi->wb_list))) 17403ba3782SJens Axboe wake_up_process(default_backing_dev_info.wb.task); 17503ba3782SJens Axboe else { 17603ba3782SJens Axboe struct bdi_writeback *wb = &bdi->wb; 1771da177e4SLinus Torvalds 1781ef7d9aaSNick Piggin if (wb->task) 17903ba3782SJens Axboe wake_up_process(wb->task); 1801da177e4SLinus Torvalds } 18103ba3782SJens Axboe } 1821da177e4SLinus Torvalds 1831da177e4SLinus Torvalds /* 18403ba3782SJens Axboe * Used for on-stack allocated work items. The caller needs to wait until 18503ba3782SJens Axboe * the wb threads have acked the work before it's safe to continue. 1861da177e4SLinus Torvalds */ 18703ba3782SJens Axboe static void bdi_wait_on_work_clear(struct bdi_work *work) 1881da177e4SLinus Torvalds { 18903ba3782SJens Axboe wait_on_bit(&work->state, WS_USED_B, bdi_sched_wait, 19003ba3782SJens Axboe TASK_UNINTERRUPTIBLE); 19103ba3782SJens Axboe } 19203ba3782SJens Axboe 193f11fcae8SJens Axboe static void bdi_alloc_queue_work(struct backing_dev_info *bdi, 194b6e51316SJens Axboe struct wb_writeback_args *args) 19503ba3782SJens Axboe { 19603ba3782SJens Axboe struct bdi_work *work; 19703ba3782SJens Axboe 198bcddc3f0SJens Axboe /* 199bcddc3f0SJens Axboe * This is WB_SYNC_NONE writeback, so if allocation fails just 200bcddc3f0SJens Axboe * wakeup the thread for old dirty data writeback 201bcddc3f0SJens Axboe */ 20203ba3782SJens Axboe work = kmalloc(sizeof(*work), GFP_ATOMIC); 203bcddc3f0SJens Axboe if (work) { 204b6e51316SJens Axboe bdi_work_init(work, args); 205f11fcae8SJens Axboe bdi_queue_work(bdi, work); 206bcddc3f0SJens Axboe } else { 207bcddc3f0SJens Axboe struct bdi_writeback *wb = &bdi->wb; 208bcddc3f0SJens Axboe 209bcddc3f0SJens Axboe if (wb->task) 210bcddc3f0SJens Axboe wake_up_process(wb->task); 211bcddc3f0SJens Axboe } 21203ba3782SJens Axboe } 21303ba3782SJens Axboe 214b6e51316SJens Axboe /** 215b6e51316SJens Axboe * bdi_sync_writeback - start and wait for writeback 216b6e51316SJens Axboe * @bdi: the backing device to write from 217b6e51316SJens Axboe * @sb: write inodes from this super_block 218b6e51316SJens Axboe * 219b6e51316SJens Axboe * Description: 220b6e51316SJens Axboe * This does WB_SYNC_ALL data integrity writeback and waits for the 221b6e51316SJens Axboe * IO to complete. Callers must hold the sb s_umount semaphore for 222b6e51316SJens Axboe * reading, to avoid having the super disappear before we are done. 22303ba3782SJens Axboe */ 224b6e51316SJens Axboe static void bdi_sync_writeback(struct backing_dev_info *bdi, 225b6e51316SJens Axboe struct super_block *sb) 226b6e51316SJens Axboe { 227b6e51316SJens Axboe struct wb_writeback_args args = { 228b6e51316SJens Axboe .sb = sb, 229b6e51316SJens Axboe .sync_mode = WB_SYNC_ALL, 230b6e51316SJens Axboe .nr_pages = LONG_MAX, 231b6e51316SJens Axboe .range_cyclic = 0, 232b6e51316SJens Axboe }; 233f0fad8a5SChristoph Hellwig struct bdi_work work; 234f0fad8a5SChristoph Hellwig 235b6e51316SJens Axboe bdi_work_init(&work, &args); 236f0fad8a5SChristoph Hellwig work.state |= WS_ONSTACK; 237f0fad8a5SChristoph Hellwig 238b6e51316SJens Axboe bdi_queue_work(bdi, &work); 239f0fad8a5SChristoph Hellwig bdi_wait_on_work_clear(&work); 24003ba3782SJens Axboe } 241b6e51316SJens Axboe 242b6e51316SJens Axboe /** 243b6e51316SJens Axboe * bdi_start_writeback - start writeback 244b6e51316SJens Axboe * @bdi: the backing device to write from 245b6e51316SJens Axboe * @nr_pages: the number of pages to write 246b6e51316SJens Axboe * 247b6e51316SJens Axboe * Description: 248b6e51316SJens Axboe * This does WB_SYNC_NONE opportunistic writeback. The IO is only 249b6e51316SJens Axboe * started when this function returns, we make no guarentees on 250b6e51316SJens Axboe * completion. Caller need not hold sb s_umount semaphore. 251b6e51316SJens Axboe * 252b6e51316SJens Axboe */ 253b6e51316SJens Axboe void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages) 254b6e51316SJens Axboe { 255b6e51316SJens Axboe struct wb_writeback_args args = { 256b6e51316SJens Axboe .sync_mode = WB_SYNC_NONE, 257b6e51316SJens Axboe .nr_pages = nr_pages, 258b6e51316SJens Axboe .range_cyclic = 1, 259b6e51316SJens Axboe }; 260b6e51316SJens Axboe 261d3ddec76SWu Fengguang /* 262d3ddec76SWu Fengguang * We treat @nr_pages=0 as the special case to do background writeback, 263d3ddec76SWu Fengguang * ie. to sync pages until the background dirty threshold is reached. 264d3ddec76SWu Fengguang */ 265d3ddec76SWu Fengguang if (!nr_pages) { 266d3ddec76SWu Fengguang args.nr_pages = LONG_MAX; 267d3ddec76SWu Fengguang args.for_background = 1; 268d3ddec76SWu Fengguang } 269d3ddec76SWu Fengguang 270b6e51316SJens Axboe bdi_alloc_queue_work(bdi, &args); 2711da177e4SLinus Torvalds } 2721da177e4SLinus Torvalds 2731da177e4SLinus Torvalds /* 2746610a0bcSAndrew Morton * Redirty an inode: set its when-it-was dirtied timestamp and move it to the 2756610a0bcSAndrew Morton * furthest end of its superblock's dirty-inode list. 2766610a0bcSAndrew Morton * 2776610a0bcSAndrew Morton * Before stamping the inode's ->dirtied_when, we check to see whether it is 27866f3b8e2SJens Axboe * already the most-recently-dirtied inode on the b_dirty list. If that is 2796610a0bcSAndrew Morton * the case then the inode must have been redirtied while it was being written 2806610a0bcSAndrew Morton * out and we don't reset its dirtied_when. 2816610a0bcSAndrew Morton */ 2826610a0bcSAndrew Morton static void redirty_tail(struct inode *inode) 2836610a0bcSAndrew Morton { 28403ba3782SJens Axboe struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; 2856610a0bcSAndrew Morton 28603ba3782SJens Axboe if (!list_empty(&wb->b_dirty)) { 28766f3b8e2SJens Axboe struct inode *tail; 2886610a0bcSAndrew Morton 28903ba3782SJens Axboe tail = list_entry(wb->b_dirty.next, struct inode, i_list); 29066f3b8e2SJens Axboe if (time_before(inode->dirtied_when, tail->dirtied_when)) 2916610a0bcSAndrew Morton inode->dirtied_when = jiffies; 2926610a0bcSAndrew Morton } 29303ba3782SJens Axboe list_move(&inode->i_list, &wb->b_dirty); 2946610a0bcSAndrew Morton } 2956610a0bcSAndrew Morton 2966610a0bcSAndrew Morton /* 29766f3b8e2SJens Axboe * requeue inode for re-scanning after bdi->b_io list is exhausted. 298c986d1e2SAndrew Morton */ 2990e0f4fc2SKen Chen static void requeue_io(struct inode *inode) 300c986d1e2SAndrew Morton { 30103ba3782SJens Axboe struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; 30203ba3782SJens Axboe 30303ba3782SJens Axboe list_move(&inode->i_list, &wb->b_more_io); 304c986d1e2SAndrew Morton } 305c986d1e2SAndrew Morton 3061c0eeaf5SJoern Engel static void inode_sync_complete(struct inode *inode) 3071c0eeaf5SJoern Engel { 3081c0eeaf5SJoern Engel /* 3091c0eeaf5SJoern Engel * Prevent speculative execution through spin_unlock(&inode_lock); 3101c0eeaf5SJoern Engel */ 3111c0eeaf5SJoern Engel smp_mb(); 3121c0eeaf5SJoern Engel wake_up_bit(&inode->i_state, __I_SYNC); 3131c0eeaf5SJoern Engel } 3141c0eeaf5SJoern Engel 315d2caa3c5SJeff Layton static bool inode_dirtied_after(struct inode *inode, unsigned long t) 316d2caa3c5SJeff Layton { 317d2caa3c5SJeff Layton bool ret = time_after(inode->dirtied_when, t); 318d2caa3c5SJeff Layton #ifndef CONFIG_64BIT 319d2caa3c5SJeff Layton /* 320d2caa3c5SJeff Layton * For inodes being constantly redirtied, dirtied_when can get stuck. 321d2caa3c5SJeff Layton * It _appears_ to be in the future, but is actually in distant past. 322d2caa3c5SJeff Layton * This test is necessary to prevent such wrapped-around relative times 3235b0830cbSJens Axboe * from permanently stopping the whole bdi writeback. 324d2caa3c5SJeff Layton */ 325d2caa3c5SJeff Layton ret = ret && time_before_eq(inode->dirtied_when, jiffies); 326d2caa3c5SJeff Layton #endif 327d2caa3c5SJeff Layton return ret; 328d2caa3c5SJeff Layton } 329d2caa3c5SJeff Layton 330c986d1e2SAndrew Morton /* 3312c136579SFengguang Wu * Move expired dirty inodes from @delaying_queue to @dispatch_queue. 3322c136579SFengguang Wu */ 3332c136579SFengguang Wu static void move_expired_inodes(struct list_head *delaying_queue, 3342c136579SFengguang Wu struct list_head *dispatch_queue, 3352c136579SFengguang Wu unsigned long *older_than_this) 3362c136579SFengguang Wu { 3375c03449dSShaohua Li LIST_HEAD(tmp); 3385c03449dSShaohua Li struct list_head *pos, *node; 3395c03449dSShaohua Li struct super_block *sb; 3405c03449dSShaohua Li struct inode *inode; 3415c03449dSShaohua Li 3422c136579SFengguang Wu while (!list_empty(delaying_queue)) { 3435c03449dSShaohua Li inode = list_entry(delaying_queue->prev, struct inode, i_list); 3442c136579SFengguang Wu if (older_than_this && 345d2caa3c5SJeff Layton inode_dirtied_after(inode, *older_than_this)) 3462c136579SFengguang Wu break; 3475c03449dSShaohua Li list_move(&inode->i_list, &tmp); 3485c03449dSShaohua Li } 3495c03449dSShaohua Li 3505c03449dSShaohua Li /* Move inodes from one superblock together */ 3515c03449dSShaohua Li while (!list_empty(&tmp)) { 3525c03449dSShaohua Li inode = list_entry(tmp.prev, struct inode, i_list); 3535c03449dSShaohua Li sb = inode->i_sb; 3545c03449dSShaohua Li list_for_each_prev_safe(pos, node, &tmp) { 3555c03449dSShaohua Li inode = list_entry(pos, struct inode, i_list); 3565c03449dSShaohua Li if (inode->i_sb == sb) 3572c136579SFengguang Wu list_move(&inode->i_list, dispatch_queue); 3582c136579SFengguang Wu } 3592c136579SFengguang Wu } 3605c03449dSShaohua Li } 3612c136579SFengguang Wu 3622c136579SFengguang Wu /* 3632c136579SFengguang Wu * Queue all expired dirty inodes for io, eldest first. 3642c136579SFengguang Wu */ 36503ba3782SJens Axboe static void queue_io(struct bdi_writeback *wb, unsigned long *older_than_this) 3662c136579SFengguang Wu { 36703ba3782SJens Axboe list_splice_init(&wb->b_more_io, wb->b_io.prev); 36803ba3782SJens Axboe move_expired_inodes(&wb->b_dirty, &wb->b_io, older_than_this); 36966f3b8e2SJens Axboe } 37066f3b8e2SJens Axboe 37103ba3782SJens Axboe static int write_inode(struct inode *inode, int sync) 37266f3b8e2SJens Axboe { 37303ba3782SJens Axboe if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) 37403ba3782SJens Axboe return inode->i_sb->s_op->write_inode(inode, sync); 37503ba3782SJens Axboe return 0; 37666f3b8e2SJens Axboe } 37708d8e974SFengguang Wu 3782c136579SFengguang Wu /* 37901c03194SChristoph Hellwig * Wait for writeback on an inode to complete. 38001c03194SChristoph Hellwig */ 38101c03194SChristoph Hellwig static void inode_wait_for_writeback(struct inode *inode) 38201c03194SChristoph Hellwig { 38301c03194SChristoph Hellwig DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); 38401c03194SChristoph Hellwig wait_queue_head_t *wqh; 38501c03194SChristoph Hellwig 38601c03194SChristoph Hellwig wqh = bit_waitqueue(&inode->i_state, __I_SYNC); 38701c03194SChristoph Hellwig do { 38801c03194SChristoph Hellwig spin_unlock(&inode_lock); 38901c03194SChristoph Hellwig __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE); 39001c03194SChristoph Hellwig spin_lock(&inode_lock); 39101c03194SChristoph Hellwig } while (inode->i_state & I_SYNC); 39201c03194SChristoph Hellwig } 39301c03194SChristoph Hellwig 39401c03194SChristoph Hellwig /* 39501c03194SChristoph Hellwig * Write out an inode's dirty pages. Called under inode_lock. Either the 39601c03194SChristoph Hellwig * caller has ref on the inode (either via __iget or via syscall against an fd) 39701c03194SChristoph Hellwig * or the inode has I_WILL_FREE set (via generic_forget_inode) 39801c03194SChristoph Hellwig * 3991da177e4SLinus Torvalds * If `wait' is set, wait on the writeout. 4001da177e4SLinus Torvalds * 4011da177e4SLinus Torvalds * The whole writeout design is quite complex and fragile. We want to avoid 4021da177e4SLinus Torvalds * starvation of particular inodes when others are being redirtied, prevent 4031da177e4SLinus Torvalds * livelocks, etc. 4041da177e4SLinus Torvalds * 4051da177e4SLinus Torvalds * Called under inode_lock. 4061da177e4SLinus Torvalds */ 4071da177e4SLinus Torvalds static int 40801c03194SChristoph Hellwig writeback_single_inode(struct inode *inode, struct writeback_control *wbc) 4091da177e4SLinus Torvalds { 4101da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 4111da177e4SLinus Torvalds int wait = wbc->sync_mode == WB_SYNC_ALL; 41201c03194SChristoph Hellwig unsigned dirty; 4131da177e4SLinus Torvalds int ret; 4141da177e4SLinus Torvalds 41501c03194SChristoph Hellwig if (!atomic_read(&inode->i_count)) 41601c03194SChristoph Hellwig WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); 41701c03194SChristoph Hellwig else 41801c03194SChristoph Hellwig WARN_ON(inode->i_state & I_WILL_FREE); 41901c03194SChristoph Hellwig 42001c03194SChristoph Hellwig if (inode->i_state & I_SYNC) { 42101c03194SChristoph Hellwig /* 42201c03194SChristoph Hellwig * If this inode is locked for writeback and we are not doing 42366f3b8e2SJens Axboe * writeback-for-data-integrity, move it to b_more_io so that 42401c03194SChristoph Hellwig * writeback can proceed with the other inodes on s_io. 42501c03194SChristoph Hellwig * 42601c03194SChristoph Hellwig * We'll have another go at writing back this inode when we 42766f3b8e2SJens Axboe * completed a full scan of b_io. 42801c03194SChristoph Hellwig */ 42901c03194SChristoph Hellwig if (!wait) { 43001c03194SChristoph Hellwig requeue_io(inode); 43101c03194SChristoph Hellwig return 0; 43201c03194SChristoph Hellwig } 43301c03194SChristoph Hellwig 43401c03194SChristoph Hellwig /* 43501c03194SChristoph Hellwig * It's a data-integrity sync. We must wait. 43601c03194SChristoph Hellwig */ 43701c03194SChristoph Hellwig inode_wait_for_writeback(inode); 43801c03194SChristoph Hellwig } 43901c03194SChristoph Hellwig 4401c0eeaf5SJoern Engel BUG_ON(inode->i_state & I_SYNC); 4411da177e4SLinus Torvalds 4421c0eeaf5SJoern Engel /* Set I_SYNC, reset I_DIRTY */ 4431da177e4SLinus Torvalds dirty = inode->i_state & I_DIRTY; 4441c0eeaf5SJoern Engel inode->i_state |= I_SYNC; 4451da177e4SLinus Torvalds inode->i_state &= ~I_DIRTY; 4461da177e4SLinus Torvalds 4471da177e4SLinus Torvalds spin_unlock(&inode_lock); 4481da177e4SLinus Torvalds 4491da177e4SLinus Torvalds ret = do_writepages(mapping, wbc); 4501da177e4SLinus Torvalds 4511da177e4SLinus Torvalds /* Don't write the inode if only I_DIRTY_PAGES was set */ 4521da177e4SLinus Torvalds if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { 4531da177e4SLinus Torvalds int err = write_inode(inode, wait); 4541da177e4SLinus Torvalds if (ret == 0) 4551da177e4SLinus Torvalds ret = err; 4561da177e4SLinus Torvalds } 4571da177e4SLinus Torvalds 4581da177e4SLinus Torvalds if (wait) { 4591da177e4SLinus Torvalds int err = filemap_fdatawait(mapping); 4601da177e4SLinus Torvalds if (ret == 0) 4611da177e4SLinus Torvalds ret = err; 4621da177e4SLinus Torvalds } 4631da177e4SLinus Torvalds 4641da177e4SLinus Torvalds spin_lock(&inode_lock); 4651c0eeaf5SJoern Engel inode->i_state &= ~I_SYNC; 46684a89245SWu Fengguang if (!(inode->i_state & (I_FREEING | I_CLEAR))) { 467ae1b7f7dSWu Fengguang if (inode->i_state & I_DIRTY) { 468ae1b7f7dSWu Fengguang /* 469ae1b7f7dSWu Fengguang * Someone redirtied the inode while were writing back 470ae1b7f7dSWu Fengguang * the pages. 471ae1b7f7dSWu Fengguang */ 472ae1b7f7dSWu Fengguang redirty_tail(inode); 473ae1b7f7dSWu Fengguang } else if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { 4741da177e4SLinus Torvalds /* 4751da177e4SLinus Torvalds * We didn't write back all the pages. nfs_writepages() 4761da177e4SLinus Torvalds * sometimes bales out without doing anything. Redirty 47766f3b8e2SJens Axboe * the inode; Move it from b_io onto b_more_io/b_dirty. 4781b43ef91SAndrew Morton */ 4791b43ef91SAndrew Morton /* 4801b43ef91SAndrew Morton * akpm: if the caller was the kupdate function we put 48166f3b8e2SJens Axboe * this inode at the head of b_dirty so it gets first 4821b43ef91SAndrew Morton * consideration. Otherwise, move it to the tail, for 4831b43ef91SAndrew Morton * the reasons described there. I'm not really sure 4841b43ef91SAndrew Morton * how much sense this makes. Presumably I had a good 4851b43ef91SAndrew Morton * reasons for doing it this way, and I'd rather not 4861b43ef91SAndrew Morton * muck with it at present. 4871da177e4SLinus Torvalds */ 4881da177e4SLinus Torvalds if (wbc->for_kupdate) { 4891da177e4SLinus Torvalds /* 4902c136579SFengguang Wu * For the kupdate function we move the inode 49166f3b8e2SJens Axboe * to b_more_io so it will get more writeout as 4922c136579SFengguang Wu * soon as the queue becomes uncongested. 4931da177e4SLinus Torvalds */ 4941da177e4SLinus Torvalds inode->i_state |= I_DIRTY_PAGES; 4958bc3be27SFengguang Wu if (wbc->nr_to_write <= 0) { 4968bc3be27SFengguang Wu /* 4978bc3be27SFengguang Wu * slice used up: queue for next turn 4988bc3be27SFengguang Wu */ 4990e0f4fc2SKen Chen requeue_io(inode); 5001da177e4SLinus Torvalds } else { 5011da177e4SLinus Torvalds /* 5028bc3be27SFengguang Wu * somehow blocked: retry later 5038bc3be27SFengguang Wu */ 5048bc3be27SFengguang Wu redirty_tail(inode); 5058bc3be27SFengguang Wu } 5068bc3be27SFengguang Wu } else { 5078bc3be27SFengguang Wu /* 5081da177e4SLinus Torvalds * Otherwise fully redirty the inode so that 5091da177e4SLinus Torvalds * other inodes on this superblock will get some 5101da177e4SLinus Torvalds * writeout. Otherwise heavy writing to one 5111da177e4SLinus Torvalds * file would indefinitely suspend writeout of 5121da177e4SLinus Torvalds * all the other files. 5131da177e4SLinus Torvalds */ 5141da177e4SLinus Torvalds inode->i_state |= I_DIRTY_PAGES; 5151b43ef91SAndrew Morton redirty_tail(inode); 5161da177e4SLinus Torvalds } 5171da177e4SLinus Torvalds } else if (atomic_read(&inode->i_count)) { 5181da177e4SLinus Torvalds /* 5191da177e4SLinus Torvalds * The inode is clean, inuse 5201da177e4SLinus Torvalds */ 5211da177e4SLinus Torvalds list_move(&inode->i_list, &inode_in_use); 5221da177e4SLinus Torvalds } else { 5231da177e4SLinus Torvalds /* 5241da177e4SLinus Torvalds * The inode is clean, unused 5251da177e4SLinus Torvalds */ 5261da177e4SLinus Torvalds list_move(&inode->i_list, &inode_unused); 5271da177e4SLinus Torvalds } 5281da177e4SLinus Torvalds } 5291c0eeaf5SJoern Engel inode_sync_complete(inode); 5301da177e4SLinus Torvalds return ret; 5311da177e4SLinus Torvalds } 5321da177e4SLinus Torvalds 53303ba3782SJens Axboe /* 53403ba3782SJens Axboe * For WB_SYNC_NONE writeback, the caller does not have the sb pinned 53503ba3782SJens Axboe * before calling writeback. So make sure that we do pin it, so it doesn't 53603ba3782SJens Axboe * go away while we are writing inodes from it. 53703ba3782SJens Axboe * 53803ba3782SJens Axboe * Returns 0 if the super was successfully pinned (or pinning wasn't needed), 53903ba3782SJens Axboe * 1 if we failed. 54003ba3782SJens Axboe */ 54103ba3782SJens Axboe static int pin_sb_for_writeback(struct writeback_control *wbc, 54203ba3782SJens Axboe struct inode *inode) 5431da177e4SLinus Torvalds { 54403ba3782SJens Axboe struct super_block *sb = inode->i_sb; 54503ba3782SJens Axboe 54603ba3782SJens Axboe /* 54703ba3782SJens Axboe * Caller must already hold the ref for this 54803ba3782SJens Axboe */ 54903ba3782SJens Axboe if (wbc->sync_mode == WB_SYNC_ALL) { 55003ba3782SJens Axboe WARN_ON(!rwsem_is_locked(&sb->s_umount)); 55103ba3782SJens Axboe return 0; 55203ba3782SJens Axboe } 55303ba3782SJens Axboe 55403ba3782SJens Axboe spin_lock(&sb_lock); 55503ba3782SJens Axboe sb->s_count++; 55603ba3782SJens Axboe if (down_read_trylock(&sb->s_umount)) { 55703ba3782SJens Axboe if (sb->s_root) { 55803ba3782SJens Axboe spin_unlock(&sb_lock); 55903ba3782SJens Axboe return 0; 56003ba3782SJens Axboe } 56103ba3782SJens Axboe /* 56203ba3782SJens Axboe * umounted, drop rwsem again and fall through to failure 56303ba3782SJens Axboe */ 56403ba3782SJens Axboe up_read(&sb->s_umount); 56503ba3782SJens Axboe } 56603ba3782SJens Axboe 56703ba3782SJens Axboe sb->s_count--; 56803ba3782SJens Axboe spin_unlock(&sb_lock); 56903ba3782SJens Axboe return 1; 57003ba3782SJens Axboe } 57103ba3782SJens Axboe 57203ba3782SJens Axboe static void unpin_sb_for_writeback(struct writeback_control *wbc, 57303ba3782SJens Axboe struct inode *inode) 57403ba3782SJens Axboe { 57503ba3782SJens Axboe struct super_block *sb = inode->i_sb; 57603ba3782SJens Axboe 57703ba3782SJens Axboe if (wbc->sync_mode == WB_SYNC_ALL) 57803ba3782SJens Axboe return; 57903ba3782SJens Axboe 58003ba3782SJens Axboe up_read(&sb->s_umount); 58103ba3782SJens Axboe put_super(sb); 58203ba3782SJens Axboe } 58303ba3782SJens Axboe 58403ba3782SJens Axboe static void writeback_inodes_wb(struct bdi_writeback *wb, 58503ba3782SJens Axboe struct writeback_control *wbc) 58603ba3782SJens Axboe { 58703ba3782SJens Axboe struct super_block *sb = wbc->sb; 58866f3b8e2SJens Axboe const int is_blkdev_sb = sb_is_blkdev_sb(sb); 5891da177e4SLinus Torvalds const unsigned long start = jiffies; /* livelock avoidance */ 5901da177e4SLinus Torvalds 591ae8547b0SHans Reiser spin_lock(&inode_lock); 5921da177e4SLinus Torvalds 59303ba3782SJens Axboe if (!wbc->for_kupdate || list_empty(&wb->b_io)) 59403ba3782SJens Axboe queue_io(wb, wbc->older_than_this); 59566f3b8e2SJens Axboe 59603ba3782SJens Axboe while (!list_empty(&wb->b_io)) { 59703ba3782SJens Axboe struct inode *inode = list_entry(wb->b_io.prev, 5981da177e4SLinus Torvalds struct inode, i_list); 5991da177e4SLinus Torvalds long pages_skipped; 6001da177e4SLinus Torvalds 60166f3b8e2SJens Axboe /* 60266f3b8e2SJens Axboe * super block given and doesn't match, skip this inode 60366f3b8e2SJens Axboe */ 60466f3b8e2SJens Axboe if (sb && sb != inode->i_sb) { 60566f3b8e2SJens Axboe redirty_tail(inode); 60666f3b8e2SJens Axboe continue; 60766f3b8e2SJens Axboe } 60866f3b8e2SJens Axboe 60903ba3782SJens Axboe if (!bdi_cap_writeback_dirty(wb->bdi)) { 6109852a0e7SAndrew Morton redirty_tail(inode); 61166f3b8e2SJens Axboe if (is_blkdev_sb) { 6121da177e4SLinus Torvalds /* 6131da177e4SLinus Torvalds * Dirty memory-backed blockdev: the ramdisk 6141da177e4SLinus Torvalds * driver does this. Skip just this inode 6151da177e4SLinus Torvalds */ 6161da177e4SLinus Torvalds continue; 6171da177e4SLinus Torvalds } 6181da177e4SLinus Torvalds /* 6191da177e4SLinus Torvalds * Dirty memory-backed inode against a filesystem other 6201da177e4SLinus Torvalds * than the kernel-internal bdev filesystem. Skip the 6211da177e4SLinus Torvalds * entire superblock. 6221da177e4SLinus Torvalds */ 6231da177e4SLinus Torvalds break; 6241da177e4SLinus Torvalds } 6251da177e4SLinus Torvalds 62684a89245SWu Fengguang if (inode->i_state & (I_NEW | I_WILL_FREE)) { 6277ef0d737SNick Piggin requeue_io(inode); 6287ef0d737SNick Piggin continue; 6297ef0d737SNick Piggin } 6307ef0d737SNick Piggin 63103ba3782SJens Axboe if (wbc->nonblocking && bdi_write_congested(wb->bdi)) { 6321da177e4SLinus Torvalds wbc->encountered_congestion = 1; 63366f3b8e2SJens Axboe if (!is_blkdev_sb) 6341da177e4SLinus Torvalds break; /* Skip a congested fs */ 6350e0f4fc2SKen Chen requeue_io(inode); 6361da177e4SLinus Torvalds continue; /* Skip a congested blockdev */ 6371da177e4SLinus Torvalds } 6381da177e4SLinus Torvalds 639d2caa3c5SJeff Layton /* 640d2caa3c5SJeff Layton * Was this inode dirtied after sync_sb_inodes was called? 641d2caa3c5SJeff Layton * This keeps sync from extra jobs and livelock. 642d2caa3c5SJeff Layton */ 643d2caa3c5SJeff Layton if (inode_dirtied_after(inode, start)) 6441da177e4SLinus Torvalds break; 6451da177e4SLinus Torvalds 64603ba3782SJens Axboe if (pin_sb_for_writeback(wbc, inode)) { 64703ba3782SJens Axboe requeue_io(inode); 64803ba3782SJens Axboe continue; 64903ba3782SJens Axboe } 6501da177e4SLinus Torvalds 65184a89245SWu Fengguang BUG_ON(inode->i_state & (I_FREEING | I_CLEAR)); 6521da177e4SLinus Torvalds __iget(inode); 6531da177e4SLinus Torvalds pages_skipped = wbc->pages_skipped; 65401c03194SChristoph Hellwig writeback_single_inode(inode, wbc); 65503ba3782SJens Axboe unpin_sb_for_writeback(wbc, inode); 6561da177e4SLinus Torvalds if (wbc->pages_skipped != pages_skipped) { 6571da177e4SLinus Torvalds /* 6581da177e4SLinus Torvalds * writeback is not making progress due to locked 6591da177e4SLinus Torvalds * buffers. Skip this inode for now. 6601da177e4SLinus Torvalds */ 661f57b9b7bSAndrew Morton redirty_tail(inode); 6621da177e4SLinus Torvalds } 6631da177e4SLinus Torvalds spin_unlock(&inode_lock); 6641da177e4SLinus Torvalds iput(inode); 6654ffc8444SOGAWA Hirofumi cond_resched(); 6661da177e4SLinus Torvalds spin_lock(&inode_lock); 6678bc3be27SFengguang Wu if (wbc->nr_to_write <= 0) { 6688bc3be27SFengguang Wu wbc->more_io = 1; 6691da177e4SLinus Torvalds break; 6701da177e4SLinus Torvalds } 67103ba3782SJens Axboe if (!list_empty(&wb->b_more_io)) 6728bc3be27SFengguang Wu wbc->more_io = 1; 6738bc3be27SFengguang Wu } 67438f21977SNick Piggin 67566f3b8e2SJens Axboe spin_unlock(&inode_lock); 67666f3b8e2SJens Axboe /* Leave any unwritten inodes on b_io */ 67766f3b8e2SJens Axboe } 67866f3b8e2SJens Axboe 67903ba3782SJens Axboe void writeback_inodes_wbc(struct writeback_control *wbc) 68003ba3782SJens Axboe { 68103ba3782SJens Axboe struct backing_dev_info *bdi = wbc->bdi; 68203ba3782SJens Axboe 68303ba3782SJens Axboe writeback_inodes_wb(&bdi->wb, wbc); 68403ba3782SJens Axboe } 68503ba3782SJens Axboe 68603ba3782SJens Axboe /* 68703ba3782SJens Axboe * The maximum number of pages to writeout in a single bdi flush/kupdate 68803ba3782SJens Axboe * operation. We do this so we don't hold I_SYNC against an inode for 68903ba3782SJens Axboe * enormous amounts of time, which would block a userspace task which has 69003ba3782SJens Axboe * been forced to throttle against that inode. Also, the code reevaluates 69103ba3782SJens Axboe * the dirty each time it has written this many pages. 69203ba3782SJens Axboe */ 69303ba3782SJens Axboe #define MAX_WRITEBACK_PAGES 1024 69403ba3782SJens Axboe 69503ba3782SJens Axboe static inline bool over_bground_thresh(void) 69603ba3782SJens Axboe { 69703ba3782SJens Axboe unsigned long background_thresh, dirty_thresh; 69803ba3782SJens Axboe 69903ba3782SJens Axboe get_dirty_limits(&background_thresh, &dirty_thresh, NULL, NULL); 70003ba3782SJens Axboe 70103ba3782SJens Axboe return (global_page_state(NR_FILE_DIRTY) + 70203ba3782SJens Axboe global_page_state(NR_UNSTABLE_NFS) >= background_thresh); 70303ba3782SJens Axboe } 70403ba3782SJens Axboe 70503ba3782SJens Axboe /* 70603ba3782SJens Axboe * Explicit flushing or periodic writeback of "old" data. 70703ba3782SJens Axboe * 70803ba3782SJens Axboe * Define "old": the first time one of an inode's pages is dirtied, we mark the 70903ba3782SJens Axboe * dirtying-time in the inode's address_space. So this periodic writeback code 71003ba3782SJens Axboe * just walks the superblock inode list, writing back any inodes which are 71103ba3782SJens Axboe * older than a specific point in time. 71203ba3782SJens Axboe * 71303ba3782SJens Axboe * Try to run once per dirty_writeback_interval. But if a writeback event 71403ba3782SJens Axboe * takes longer than a dirty_writeback_interval interval, then leave a 71503ba3782SJens Axboe * one-second gap. 71603ba3782SJens Axboe * 71703ba3782SJens Axboe * older_than_this takes precedence over nr_to_write. So we'll only write back 71803ba3782SJens Axboe * all dirty pages if they are all attached to "old" mappings. 71903ba3782SJens Axboe */ 720c4a77a6cSJens Axboe static long wb_writeback(struct bdi_writeback *wb, 721c4a77a6cSJens Axboe struct wb_writeback_args *args) 72203ba3782SJens Axboe { 72303ba3782SJens Axboe struct writeback_control wbc = { 72403ba3782SJens Axboe .bdi = wb->bdi, 725c4a77a6cSJens Axboe .sb = args->sb, 726c4a77a6cSJens Axboe .sync_mode = args->sync_mode, 72703ba3782SJens Axboe .older_than_this = NULL, 728c4a77a6cSJens Axboe .for_kupdate = args->for_kupdate, 729c4a77a6cSJens Axboe .range_cyclic = args->range_cyclic, 73003ba3782SJens Axboe }; 73103ba3782SJens Axboe unsigned long oldest_jif; 73203ba3782SJens Axboe long wrote = 0; 733a5989bdcSJan Kara struct inode *inode; 73403ba3782SJens Axboe 73503ba3782SJens Axboe if (wbc.for_kupdate) { 73603ba3782SJens Axboe wbc.older_than_this = &oldest_jif; 73703ba3782SJens Axboe oldest_jif = jiffies - 73803ba3782SJens Axboe msecs_to_jiffies(dirty_expire_interval * 10); 73903ba3782SJens Axboe } 740c4a77a6cSJens Axboe if (!wbc.range_cyclic) { 741c4a77a6cSJens Axboe wbc.range_start = 0; 742c4a77a6cSJens Axboe wbc.range_end = LLONG_MAX; 743c4a77a6cSJens Axboe } 74403ba3782SJens Axboe 74503ba3782SJens Axboe for (;;) { 74603ba3782SJens Axboe /* 747d3ddec76SWu Fengguang * Stop writeback when nr_pages has been consumed 74803ba3782SJens Axboe */ 749d3ddec76SWu Fengguang if (args->nr_pages <= 0) 75003ba3782SJens Axboe break; 75103ba3782SJens Axboe 75203ba3782SJens Axboe /* 753d3ddec76SWu Fengguang * For background writeout, stop when we are below the 754d3ddec76SWu Fengguang * background dirty threshold 75503ba3782SJens Axboe */ 756d3ddec76SWu Fengguang if (args->for_background && !over_bground_thresh()) 75703ba3782SJens Axboe break; 75803ba3782SJens Axboe 75903ba3782SJens Axboe wbc.more_io = 0; 76003ba3782SJens Axboe wbc.encountered_congestion = 0; 76103ba3782SJens Axboe wbc.nr_to_write = MAX_WRITEBACK_PAGES; 76203ba3782SJens Axboe wbc.pages_skipped = 0; 76303ba3782SJens Axboe writeback_inodes_wb(wb, &wbc); 764c4a77a6cSJens Axboe args->nr_pages -= MAX_WRITEBACK_PAGES - wbc.nr_to_write; 76503ba3782SJens Axboe wrote += MAX_WRITEBACK_PAGES - wbc.nr_to_write; 76603ba3782SJens Axboe 76703ba3782SJens Axboe /* 76871fd05a8SJens Axboe * If we consumed everything, see if we have more 76903ba3782SJens Axboe */ 77071fd05a8SJens Axboe if (wbc.nr_to_write <= 0) 77171fd05a8SJens Axboe continue; 77271fd05a8SJens Axboe /* 77371fd05a8SJens Axboe * Didn't write everything and we don't have more IO, bail 77471fd05a8SJens Axboe */ 77571fd05a8SJens Axboe if (!wbc.more_io) 77671fd05a8SJens Axboe break; 77771fd05a8SJens Axboe /* 77871fd05a8SJens Axboe * Did we write something? Try for more 77971fd05a8SJens Axboe */ 780a5989bdcSJan Kara if (wbc.nr_to_write < MAX_WRITEBACK_PAGES) 78103ba3782SJens Axboe continue; 782a5989bdcSJan Kara /* 783a5989bdcSJan Kara * Nothing written. Wait for some inode to 784a5989bdcSJan Kara * become available for writeback. Otherwise 785a5989bdcSJan Kara * we'll just busyloop. 786a5989bdcSJan Kara */ 787a5989bdcSJan Kara spin_lock(&inode_lock); 788a5989bdcSJan Kara if (!list_empty(&wb->b_more_io)) { 78971fd05a8SJens Axboe inode = list_entry(wb->b_more_io.prev, 790a5989bdcSJan Kara struct inode, i_list); 791a5989bdcSJan Kara inode_wait_for_writeback(inode); 792a5989bdcSJan Kara } 793a5989bdcSJan Kara spin_unlock(&inode_lock); 79403ba3782SJens Axboe } 79503ba3782SJens Axboe 79603ba3782SJens Axboe return wrote; 79703ba3782SJens Axboe } 79803ba3782SJens Axboe 79903ba3782SJens Axboe /* 80003ba3782SJens Axboe * Return the next bdi_work struct that hasn't been processed by this 8018010c3b6SJens Axboe * wb thread yet. ->seen is initially set for each thread that exists 8028010c3b6SJens Axboe * for this device, when a thread first notices a piece of work it 8038010c3b6SJens Axboe * clears its bit. Depending on writeback type, the thread will notify 8048010c3b6SJens Axboe * completion on either receiving the work (WB_SYNC_NONE) or after 8058010c3b6SJens Axboe * it is done (WB_SYNC_ALL). 80603ba3782SJens Axboe */ 80703ba3782SJens Axboe static struct bdi_work *get_next_work_item(struct backing_dev_info *bdi, 80803ba3782SJens Axboe struct bdi_writeback *wb) 80903ba3782SJens Axboe { 81003ba3782SJens Axboe struct bdi_work *work, *ret = NULL; 81103ba3782SJens Axboe 81203ba3782SJens Axboe rcu_read_lock(); 81303ba3782SJens Axboe 81403ba3782SJens Axboe list_for_each_entry_rcu(work, &bdi->work_list, list) { 81577fad5e6SNick Piggin if (!test_bit(wb->nr, &work->seen)) 81603ba3782SJens Axboe continue; 81777fad5e6SNick Piggin clear_bit(wb->nr, &work->seen); 81803ba3782SJens Axboe 81903ba3782SJens Axboe ret = work; 82003ba3782SJens Axboe break; 82103ba3782SJens Axboe } 82203ba3782SJens Axboe 82303ba3782SJens Axboe rcu_read_unlock(); 82403ba3782SJens Axboe return ret; 82503ba3782SJens Axboe } 82603ba3782SJens Axboe 82703ba3782SJens Axboe static long wb_check_old_data_flush(struct bdi_writeback *wb) 82803ba3782SJens Axboe { 82903ba3782SJens Axboe unsigned long expired; 83003ba3782SJens Axboe long nr_pages; 83103ba3782SJens Axboe 83203ba3782SJens Axboe expired = wb->last_old_flush + 83303ba3782SJens Axboe msecs_to_jiffies(dirty_writeback_interval * 10); 83403ba3782SJens Axboe if (time_before(jiffies, expired)) 83503ba3782SJens Axboe return 0; 83603ba3782SJens Axboe 83703ba3782SJens Axboe wb->last_old_flush = jiffies; 83803ba3782SJens Axboe nr_pages = global_page_state(NR_FILE_DIRTY) + 83903ba3782SJens Axboe global_page_state(NR_UNSTABLE_NFS) + 84003ba3782SJens Axboe (inodes_stat.nr_inodes - inodes_stat.nr_unused); 84103ba3782SJens Axboe 842c4a77a6cSJens Axboe if (nr_pages) { 843c4a77a6cSJens Axboe struct wb_writeback_args args = { 844c4a77a6cSJens Axboe .nr_pages = nr_pages, 845c4a77a6cSJens Axboe .sync_mode = WB_SYNC_NONE, 846c4a77a6cSJens Axboe .for_kupdate = 1, 847c4a77a6cSJens Axboe .range_cyclic = 1, 848c4a77a6cSJens Axboe }; 849c4a77a6cSJens Axboe 850c4a77a6cSJens Axboe return wb_writeback(wb, &args); 851c4a77a6cSJens Axboe } 85203ba3782SJens Axboe 85303ba3782SJens Axboe return 0; 85403ba3782SJens Axboe } 85503ba3782SJens Axboe 85603ba3782SJens Axboe /* 85703ba3782SJens Axboe * Retrieve work items and do the writeback they describe 85803ba3782SJens Axboe */ 85903ba3782SJens Axboe long wb_do_writeback(struct bdi_writeback *wb, int force_wait) 86003ba3782SJens Axboe { 86103ba3782SJens Axboe struct backing_dev_info *bdi = wb->bdi; 86203ba3782SJens Axboe struct bdi_work *work; 863c4a77a6cSJens Axboe long wrote = 0; 86403ba3782SJens Axboe 86503ba3782SJens Axboe while ((work = get_next_work_item(bdi, wb)) != NULL) { 866c4a77a6cSJens Axboe struct wb_writeback_args args = work->args; 86703ba3782SJens Axboe 86803ba3782SJens Axboe /* 86903ba3782SJens Axboe * Override sync mode, in case we must wait for completion 87003ba3782SJens Axboe */ 87103ba3782SJens Axboe if (force_wait) 872c4a77a6cSJens Axboe work->args.sync_mode = args.sync_mode = WB_SYNC_ALL; 87303ba3782SJens Axboe 87403ba3782SJens Axboe /* 87503ba3782SJens Axboe * If this isn't a data integrity operation, just notify 87603ba3782SJens Axboe * that we have seen this work and we are now starting it. 87703ba3782SJens Axboe */ 878c4a77a6cSJens Axboe if (args.sync_mode == WB_SYNC_NONE) 87903ba3782SJens Axboe wb_clear_pending(wb, work); 88003ba3782SJens Axboe 881c4a77a6cSJens Axboe wrote += wb_writeback(wb, &args); 88203ba3782SJens Axboe 88303ba3782SJens Axboe /* 88403ba3782SJens Axboe * This is a data integrity writeback, so only do the 88503ba3782SJens Axboe * notification when we have completed the work. 88603ba3782SJens Axboe */ 887c4a77a6cSJens Axboe if (args.sync_mode == WB_SYNC_ALL) 88803ba3782SJens Axboe wb_clear_pending(wb, work); 88903ba3782SJens Axboe } 89003ba3782SJens Axboe 89103ba3782SJens Axboe /* 89203ba3782SJens Axboe * Check for periodic writeback, kupdated() style 89303ba3782SJens Axboe */ 89403ba3782SJens Axboe wrote += wb_check_old_data_flush(wb); 89503ba3782SJens Axboe 89603ba3782SJens Axboe return wrote; 89703ba3782SJens Axboe } 89803ba3782SJens Axboe 89903ba3782SJens Axboe /* 90003ba3782SJens Axboe * Handle writeback of dirty data for the device backed by this bdi. Also 90103ba3782SJens Axboe * wakes up periodically and does kupdated style flushing. 90203ba3782SJens Axboe */ 90303ba3782SJens Axboe int bdi_writeback_task(struct bdi_writeback *wb) 90403ba3782SJens Axboe { 90503ba3782SJens Axboe unsigned long last_active = jiffies; 90603ba3782SJens Axboe unsigned long wait_jiffies = -1UL; 90703ba3782SJens Axboe long pages_written; 90803ba3782SJens Axboe 90903ba3782SJens Axboe while (!kthread_should_stop()) { 91003ba3782SJens Axboe pages_written = wb_do_writeback(wb, 0); 91103ba3782SJens Axboe 91203ba3782SJens Axboe if (pages_written) 91303ba3782SJens Axboe last_active = jiffies; 91403ba3782SJens Axboe else if (wait_jiffies != -1UL) { 91503ba3782SJens Axboe unsigned long max_idle; 91603ba3782SJens Axboe 91703ba3782SJens Axboe /* 91803ba3782SJens Axboe * Longest period of inactivity that we tolerate. If we 91903ba3782SJens Axboe * see dirty data again later, the task will get 92003ba3782SJens Axboe * recreated automatically. 92103ba3782SJens Axboe */ 92203ba3782SJens Axboe max_idle = max(5UL * 60 * HZ, wait_jiffies); 92303ba3782SJens Axboe if (time_after(jiffies, max_idle + last_active)) 92403ba3782SJens Axboe break; 92503ba3782SJens Axboe } 92603ba3782SJens Axboe 92703ba3782SJens Axboe wait_jiffies = msecs_to_jiffies(dirty_writeback_interval * 10); 92849db0414SJens Axboe schedule_timeout_interruptible(wait_jiffies); 92903ba3782SJens Axboe try_to_freeze(); 93003ba3782SJens Axboe } 93103ba3782SJens Axboe 93203ba3782SJens Axboe return 0; 93303ba3782SJens Axboe } 93403ba3782SJens Axboe 93503ba3782SJens Axboe /* 936b6e51316SJens Axboe * Schedule writeback for all backing devices. This does WB_SYNC_NONE 937b6e51316SJens Axboe * writeback, for integrity writeback see bdi_sync_writeback(). 93803ba3782SJens Axboe */ 939b6e51316SJens Axboe static void bdi_writeback_all(struct super_block *sb, long nr_pages) 94003ba3782SJens Axboe { 941b6e51316SJens Axboe struct wb_writeback_args args = { 942b6e51316SJens Axboe .sb = sb, 943b6e51316SJens Axboe .nr_pages = nr_pages, 944b6e51316SJens Axboe .sync_mode = WB_SYNC_NONE, 945b6e51316SJens Axboe }; 94603ba3782SJens Axboe struct backing_dev_info *bdi; 94703ba3782SJens Axboe 948cfc4ba53SJens Axboe rcu_read_lock(); 94903ba3782SJens Axboe 950cfc4ba53SJens Axboe list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { 95103ba3782SJens Axboe if (!bdi_has_dirty_io(bdi)) 95203ba3782SJens Axboe continue; 95303ba3782SJens Axboe 954b6e51316SJens Axboe bdi_alloc_queue_work(bdi, &args); 95503ba3782SJens Axboe } 95603ba3782SJens Axboe 957cfc4ba53SJens Axboe rcu_read_unlock(); 95803ba3782SJens Axboe } 95903ba3782SJens Axboe 96003ba3782SJens Axboe /* 96103ba3782SJens Axboe * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back 96203ba3782SJens Axboe * the whole world. 96303ba3782SJens Axboe */ 96403ba3782SJens Axboe void wakeup_flusher_threads(long nr_pages) 96503ba3782SJens Axboe { 96603ba3782SJens Axboe if (nr_pages == 0) 96703ba3782SJens Axboe nr_pages = global_page_state(NR_FILE_DIRTY) + 96803ba3782SJens Axboe global_page_state(NR_UNSTABLE_NFS); 969b6e51316SJens Axboe bdi_writeback_all(NULL, nr_pages); 97003ba3782SJens Axboe } 97103ba3782SJens Axboe 97203ba3782SJens Axboe static noinline void block_dump___mark_inode_dirty(struct inode *inode) 97303ba3782SJens Axboe { 97403ba3782SJens Axboe if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { 97503ba3782SJens Axboe struct dentry *dentry; 97603ba3782SJens Axboe const char *name = "?"; 97703ba3782SJens Axboe 97803ba3782SJens Axboe dentry = d_find_alias(inode); 97903ba3782SJens Axboe if (dentry) { 98003ba3782SJens Axboe spin_lock(&dentry->d_lock); 98103ba3782SJens Axboe name = (const char *) dentry->d_name.name; 98203ba3782SJens Axboe } 98303ba3782SJens Axboe printk(KERN_DEBUG 98403ba3782SJens Axboe "%s(%d): dirtied inode %lu (%s) on %s\n", 98503ba3782SJens Axboe current->comm, task_pid_nr(current), inode->i_ino, 98603ba3782SJens Axboe name, inode->i_sb->s_id); 98703ba3782SJens Axboe if (dentry) { 98803ba3782SJens Axboe spin_unlock(&dentry->d_lock); 98903ba3782SJens Axboe dput(dentry); 99003ba3782SJens Axboe } 99103ba3782SJens Axboe } 99203ba3782SJens Axboe } 99303ba3782SJens Axboe 99403ba3782SJens Axboe /** 99503ba3782SJens Axboe * __mark_inode_dirty - internal function 99603ba3782SJens Axboe * @inode: inode to mark 99703ba3782SJens Axboe * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) 99803ba3782SJens Axboe * Mark an inode as dirty. Callers should use mark_inode_dirty or 99903ba3782SJens Axboe * mark_inode_dirty_sync. 100003ba3782SJens Axboe * 100103ba3782SJens Axboe * Put the inode on the super block's dirty list. 100203ba3782SJens Axboe * 100303ba3782SJens Axboe * CAREFUL! We mark it dirty unconditionally, but move it onto the 100403ba3782SJens Axboe * dirty list only if it is hashed or if it refers to a blockdev. 100503ba3782SJens Axboe * If it was not hashed, it will never be added to the dirty list 100603ba3782SJens Axboe * even if it is later hashed, as it will have been marked dirty already. 100703ba3782SJens Axboe * 100803ba3782SJens Axboe * In short, make sure you hash any inodes _before_ you start marking 100903ba3782SJens Axboe * them dirty. 101003ba3782SJens Axboe * 101103ba3782SJens Axboe * This function *must* be atomic for the I_DIRTY_PAGES case - 101203ba3782SJens Axboe * set_page_dirty() is called under spinlock in several places. 101303ba3782SJens Axboe * 101403ba3782SJens Axboe * Note that for blockdevs, inode->dirtied_when represents the dirtying time of 101503ba3782SJens Axboe * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of 101603ba3782SJens Axboe * the kernel-internal blockdev inode represents the dirtying time of the 101703ba3782SJens Axboe * blockdev's pages. This is why for I_DIRTY_PAGES we always use 101803ba3782SJens Axboe * page->mapping->host, so the page-dirtying time is recorded in the internal 101903ba3782SJens Axboe * blockdev inode. 102003ba3782SJens Axboe */ 102103ba3782SJens Axboe void __mark_inode_dirty(struct inode *inode, int flags) 102203ba3782SJens Axboe { 102303ba3782SJens Axboe struct super_block *sb = inode->i_sb; 102403ba3782SJens Axboe 102503ba3782SJens Axboe /* 102603ba3782SJens Axboe * Don't do this for I_DIRTY_PAGES - that doesn't actually 102703ba3782SJens Axboe * dirty the inode itself 102803ba3782SJens Axboe */ 102903ba3782SJens Axboe if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { 103003ba3782SJens Axboe if (sb->s_op->dirty_inode) 103103ba3782SJens Axboe sb->s_op->dirty_inode(inode); 103203ba3782SJens Axboe } 103303ba3782SJens Axboe 103403ba3782SJens Axboe /* 103503ba3782SJens Axboe * make sure that changes are seen by all cpus before we test i_state 103603ba3782SJens Axboe * -- mikulas 103703ba3782SJens Axboe */ 103803ba3782SJens Axboe smp_mb(); 103903ba3782SJens Axboe 104003ba3782SJens Axboe /* avoid the locking if we can */ 104103ba3782SJens Axboe if ((inode->i_state & flags) == flags) 104203ba3782SJens Axboe return; 104303ba3782SJens Axboe 104403ba3782SJens Axboe if (unlikely(block_dump)) 104503ba3782SJens Axboe block_dump___mark_inode_dirty(inode); 104603ba3782SJens Axboe 104703ba3782SJens Axboe spin_lock(&inode_lock); 104803ba3782SJens Axboe if ((inode->i_state & flags) != flags) { 104903ba3782SJens Axboe const int was_dirty = inode->i_state & I_DIRTY; 105003ba3782SJens Axboe 105103ba3782SJens Axboe inode->i_state |= flags; 105203ba3782SJens Axboe 105303ba3782SJens Axboe /* 105403ba3782SJens Axboe * If the inode is being synced, just update its dirty state. 105503ba3782SJens Axboe * The unlocker will place the inode on the appropriate 105603ba3782SJens Axboe * superblock list, based upon its state. 105703ba3782SJens Axboe */ 105803ba3782SJens Axboe if (inode->i_state & I_SYNC) 105903ba3782SJens Axboe goto out; 106003ba3782SJens Axboe 106103ba3782SJens Axboe /* 106203ba3782SJens Axboe * Only add valid (hashed) inodes to the superblock's 106303ba3782SJens Axboe * dirty list. Add blockdev inodes as well. 106403ba3782SJens Axboe */ 106503ba3782SJens Axboe if (!S_ISBLK(inode->i_mode)) { 106603ba3782SJens Axboe if (hlist_unhashed(&inode->i_hash)) 106703ba3782SJens Axboe goto out; 106803ba3782SJens Axboe } 106903ba3782SJens Axboe if (inode->i_state & (I_FREEING|I_CLEAR)) 107003ba3782SJens Axboe goto out; 107103ba3782SJens Axboe 107203ba3782SJens Axboe /* 107303ba3782SJens Axboe * If the inode was already on b_dirty/b_io/b_more_io, don't 107403ba3782SJens Axboe * reposition it (that would break b_dirty time-ordering). 107503ba3782SJens Axboe */ 107603ba3782SJens Axboe if (!was_dirty) { 107703ba3782SJens Axboe struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; 1078500b067cSJens Axboe struct backing_dev_info *bdi = wb->bdi; 1079500b067cSJens Axboe 1080500b067cSJens Axboe if (bdi_cap_writeback_dirty(bdi) && 1081500b067cSJens Axboe !test_bit(BDI_registered, &bdi->state)) { 1082500b067cSJens Axboe WARN_ON(1); 1083500b067cSJens Axboe printk(KERN_ERR "bdi-%s not registered\n", 1084500b067cSJens Axboe bdi->name); 1085500b067cSJens Axboe } 108603ba3782SJens Axboe 108703ba3782SJens Axboe inode->dirtied_when = jiffies; 108803ba3782SJens Axboe list_move(&inode->i_list, &wb->b_dirty); 108903ba3782SJens Axboe } 109003ba3782SJens Axboe } 109103ba3782SJens Axboe out: 109203ba3782SJens Axboe spin_unlock(&inode_lock); 109303ba3782SJens Axboe } 109403ba3782SJens Axboe EXPORT_SYMBOL(__mark_inode_dirty); 109503ba3782SJens Axboe 109666f3b8e2SJens Axboe /* 109766f3b8e2SJens Axboe * Write out a superblock's list of dirty inodes. A wait will be performed 109866f3b8e2SJens Axboe * upon no inodes, all inodes or the final one, depending upon sync_mode. 109966f3b8e2SJens Axboe * 110066f3b8e2SJens Axboe * If older_than_this is non-NULL, then only write out inodes which 110166f3b8e2SJens Axboe * had their first dirtying at a time earlier than *older_than_this. 110266f3b8e2SJens Axboe * 110366f3b8e2SJens Axboe * If `bdi' is non-zero then we're being asked to writeback a specific queue. 110466f3b8e2SJens Axboe * This function assumes that the blockdev superblock's inodes are backed by 110566f3b8e2SJens Axboe * a variety of queues, so all inodes are searched. For other superblocks, 110666f3b8e2SJens Axboe * assume that all inodes are backed by the same queue. 110766f3b8e2SJens Axboe * 110866f3b8e2SJens Axboe * The inodes to be written are parked on bdi->b_io. They are moved back onto 110966f3b8e2SJens Axboe * bdi->b_dirty as they are selected for writing. This way, none can be missed 111066f3b8e2SJens Axboe * on the writer throttling path, and we get decent balancing between many 111166f3b8e2SJens Axboe * throttled threads: we don't want them all piling up on inode_sync_wait. 111266f3b8e2SJens Axboe */ 1113b6e51316SJens Axboe static void wait_sb_inodes(struct super_block *sb) 111466f3b8e2SJens Axboe { 111538f21977SNick Piggin struct inode *inode, *old_inode = NULL; 111638f21977SNick Piggin 111703ba3782SJens Axboe /* 111803ba3782SJens Axboe * We need to be protected against the filesystem going from 111903ba3782SJens Axboe * r/o to r/w or vice versa. 112003ba3782SJens Axboe */ 1121b6e51316SJens Axboe WARN_ON(!rwsem_is_locked(&sb->s_umount)); 112203ba3782SJens Axboe 112366f3b8e2SJens Axboe spin_lock(&inode_lock); 112466f3b8e2SJens Axboe 112538f21977SNick Piggin /* 112638f21977SNick Piggin * Data integrity sync. Must wait for all pages under writeback, 112738f21977SNick Piggin * because there may have been pages dirtied before our sync 112838f21977SNick Piggin * call, but which had writeout started before we write it out. 112938f21977SNick Piggin * In which case, the inode may not be on the dirty list, but 113038f21977SNick Piggin * we still have to wait for that writeout. 113138f21977SNick Piggin */ 1132b6e51316SJens Axboe list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { 113338f21977SNick Piggin struct address_space *mapping; 113438f21977SNick Piggin 113503ba3782SJens Axboe if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE|I_NEW)) 113638f21977SNick Piggin continue; 113738f21977SNick Piggin mapping = inode->i_mapping; 113838f21977SNick Piggin if (mapping->nrpages == 0) 113938f21977SNick Piggin continue; 114038f21977SNick Piggin __iget(inode); 1141ae8547b0SHans Reiser spin_unlock(&inode_lock); 114238f21977SNick Piggin /* 114338f21977SNick Piggin * We hold a reference to 'inode' so it couldn't have 114438f21977SNick Piggin * been removed from s_inodes list while we dropped the 114538f21977SNick Piggin * inode_lock. We cannot iput the inode now as we can 114638f21977SNick Piggin * be holding the last reference and we cannot iput it 114738f21977SNick Piggin * under inode_lock. So we keep the reference and iput 114838f21977SNick Piggin * it later. 114938f21977SNick Piggin */ 115038f21977SNick Piggin iput(old_inode); 115138f21977SNick Piggin old_inode = inode; 115238f21977SNick Piggin 115338f21977SNick Piggin filemap_fdatawait(mapping); 115438f21977SNick Piggin 115538f21977SNick Piggin cond_resched(); 115638f21977SNick Piggin 115738f21977SNick Piggin spin_lock(&inode_lock); 115838f21977SNick Piggin } 115938f21977SNick Piggin spin_unlock(&inode_lock); 116038f21977SNick Piggin iput(old_inode); 116166f3b8e2SJens Axboe } 11621da177e4SLinus Torvalds 1163d8a8559cSJens Axboe /** 1164d8a8559cSJens Axboe * writeback_inodes_sb - writeback dirty inodes from given super_block 1165d8a8559cSJens Axboe * @sb: the superblock 11661da177e4SLinus Torvalds * 1167d8a8559cSJens Axboe * Start writeback on some inodes on this super_block. No guarantees are made 1168d8a8559cSJens Axboe * on how many (if any) will be written, and this function does not wait 1169d8a8559cSJens Axboe * for IO completion of submitted IO. The number of pages submitted is 1170d8a8559cSJens Axboe * returned. 11711da177e4SLinus Torvalds */ 1172b6e51316SJens Axboe void writeback_inodes_sb(struct super_block *sb) 11731da177e4SLinus Torvalds { 1174b1e7a8fdSChristoph Lameter unsigned long nr_dirty = global_page_state(NR_FILE_DIRTY); 1175fd39fc85SChristoph Lameter unsigned long nr_unstable = global_page_state(NR_UNSTABLE_NFS); 1176d8a8559cSJens Axboe long nr_to_write; 11771da177e4SLinus Torvalds 1178d8a8559cSJens Axboe nr_to_write = nr_dirty + nr_unstable + 117938f21977SNick Piggin (inodes_stat.nr_inodes - inodes_stat.nr_unused); 118038f21977SNick Piggin 1181b6e51316SJens Axboe bdi_writeback_all(sb, nr_to_write); 11821da177e4SLinus Torvalds } 1183d8a8559cSJens Axboe EXPORT_SYMBOL(writeback_inodes_sb); 1184d8a8559cSJens Axboe 1185d8a8559cSJens Axboe /** 1186d8a8559cSJens Axboe * sync_inodes_sb - sync sb inode pages 1187d8a8559cSJens Axboe * @sb: the superblock 1188d8a8559cSJens Axboe * 1189d8a8559cSJens Axboe * This function writes and waits on any dirty inode belonging to this 1190d8a8559cSJens Axboe * super_block. The number of pages synced is returned. 1191d8a8559cSJens Axboe */ 1192b6e51316SJens Axboe void sync_inodes_sb(struct super_block *sb) 1193d8a8559cSJens Axboe { 1194b6e51316SJens Axboe bdi_sync_writeback(sb->s_bdi, sb); 1195b6e51316SJens Axboe wait_sb_inodes(sb); 1196d8a8559cSJens Axboe } 1197d8a8559cSJens Axboe EXPORT_SYMBOL(sync_inodes_sb); 11981da177e4SLinus Torvalds 11991da177e4SLinus Torvalds /** 12001da177e4SLinus Torvalds * write_inode_now - write an inode to disk 12011da177e4SLinus Torvalds * @inode: inode to write to disk 12021da177e4SLinus Torvalds * @sync: whether the write should be synchronous or not 12031da177e4SLinus Torvalds * 12047f04c26dSAndrea Arcangeli * This function commits an inode to disk immediately if it is dirty. This is 12057f04c26dSAndrea Arcangeli * primarily needed by knfsd. 12067f04c26dSAndrea Arcangeli * 12077f04c26dSAndrea Arcangeli * The caller must either have a ref on the inode or must have set I_WILL_FREE. 12081da177e4SLinus Torvalds */ 12091da177e4SLinus Torvalds int write_inode_now(struct inode *inode, int sync) 12101da177e4SLinus Torvalds { 12111da177e4SLinus Torvalds int ret; 12121da177e4SLinus Torvalds struct writeback_control wbc = { 12131da177e4SLinus Torvalds .nr_to_write = LONG_MAX, 121418914b18SMike Galbraith .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, 1215111ebb6eSOGAWA Hirofumi .range_start = 0, 1216111ebb6eSOGAWA Hirofumi .range_end = LLONG_MAX, 12171da177e4SLinus Torvalds }; 12181da177e4SLinus Torvalds 12191da177e4SLinus Torvalds if (!mapping_cap_writeback_dirty(inode->i_mapping)) 122049364ce2SAndrew Morton wbc.nr_to_write = 0; 12211da177e4SLinus Torvalds 12221da177e4SLinus Torvalds might_sleep(); 12231da177e4SLinus Torvalds spin_lock(&inode_lock); 122401c03194SChristoph Hellwig ret = writeback_single_inode(inode, &wbc); 12251da177e4SLinus Torvalds spin_unlock(&inode_lock); 12261da177e4SLinus Torvalds if (sync) 12271c0eeaf5SJoern Engel inode_sync_wait(inode); 12281da177e4SLinus Torvalds return ret; 12291da177e4SLinus Torvalds } 12301da177e4SLinus Torvalds EXPORT_SYMBOL(write_inode_now); 12311da177e4SLinus Torvalds 12321da177e4SLinus Torvalds /** 12331da177e4SLinus Torvalds * sync_inode - write an inode and its pages to disk. 12341da177e4SLinus Torvalds * @inode: the inode to sync 12351da177e4SLinus Torvalds * @wbc: controls the writeback mode 12361da177e4SLinus Torvalds * 12371da177e4SLinus Torvalds * sync_inode() will write an inode and its pages to disk. It will also 12381da177e4SLinus Torvalds * correctly update the inode on its superblock's dirty inode lists and will 12391da177e4SLinus Torvalds * update inode->i_state. 12401da177e4SLinus Torvalds * 12411da177e4SLinus Torvalds * The caller must have a ref on the inode. 12421da177e4SLinus Torvalds */ 12431da177e4SLinus Torvalds int sync_inode(struct inode *inode, struct writeback_control *wbc) 12441da177e4SLinus Torvalds { 12451da177e4SLinus Torvalds int ret; 12461da177e4SLinus Torvalds 12471da177e4SLinus Torvalds spin_lock(&inode_lock); 124801c03194SChristoph Hellwig ret = writeback_single_inode(inode, wbc); 12491da177e4SLinus Torvalds spin_unlock(&inode_lock); 12501da177e4SLinus Torvalds return ret; 12511da177e4SLinus Torvalds } 12521da177e4SLinus Torvalds EXPORT_SYMBOL(sync_inode); 1253