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; 44c4a77a6cSJens Axboe int for_kupdate; 45c4a77a6cSJens Axboe int range_cyclic; 46c4a77a6cSJens Axboe }; 47c4a77a6cSJens Axboe 48c4a77a6cSJens Axboe /* 4903ba3782SJens Axboe * Work items for the bdi_writeback threads 50f11b00f3SAdrian Bunk */ 5103ba3782SJens Axboe struct bdi_work { 528010c3b6SJens Axboe struct list_head list; /* pending work list */ 538010c3b6SJens Axboe struct rcu_head rcu_head; /* for RCU free/clear of work */ 5403ba3782SJens Axboe 558010c3b6SJens Axboe unsigned long seen; /* threads that have seen this work */ 568010c3b6SJens Axboe atomic_t pending; /* number of threads still to do work */ 5703ba3782SJens Axboe 588010c3b6SJens Axboe struct wb_writeback_args args; /* writeback arguments */ 5903ba3782SJens Axboe 608010c3b6SJens Axboe unsigned long state; /* flag bits, see WS_* */ 6103ba3782SJens Axboe }; 6203ba3782SJens Axboe 6303ba3782SJens Axboe enum { 6403ba3782SJens Axboe WS_USED_B = 0, 6503ba3782SJens Axboe WS_ONSTACK_B, 6603ba3782SJens Axboe }; 6703ba3782SJens Axboe 6803ba3782SJens Axboe #define WS_USED (1 << WS_USED_B) 6903ba3782SJens Axboe #define WS_ONSTACK (1 << WS_ONSTACK_B) 7003ba3782SJens Axboe 7103ba3782SJens Axboe static inline bool bdi_work_on_stack(struct bdi_work *work) 72f11b00f3SAdrian Bunk { 7303ba3782SJens Axboe return test_bit(WS_ONSTACK_B, &work->state); 7403ba3782SJens Axboe } 7503ba3782SJens Axboe 7603ba3782SJens Axboe static inline void bdi_work_init(struct bdi_work *work, 77b6e51316SJens Axboe struct wb_writeback_args *args) 7803ba3782SJens Axboe { 7903ba3782SJens Axboe INIT_RCU_HEAD(&work->rcu_head); 80b6e51316SJens Axboe work->args = *args; 8103ba3782SJens Axboe work->state = WS_USED; 8203ba3782SJens Axboe } 8303ba3782SJens Axboe 84f11b00f3SAdrian Bunk /** 85f11b00f3SAdrian Bunk * writeback_in_progress - determine whether there is writeback in progress 86f11b00f3SAdrian Bunk * @bdi: the device's backing_dev_info structure. 87f11b00f3SAdrian Bunk * 8803ba3782SJens Axboe * Determine whether there is writeback waiting to be handled against a 8903ba3782SJens Axboe * backing device. 90f11b00f3SAdrian Bunk */ 91f11b00f3SAdrian Bunk int writeback_in_progress(struct backing_dev_info *bdi) 92f11b00f3SAdrian Bunk { 9303ba3782SJens Axboe return !list_empty(&bdi->work_list); 94f11b00f3SAdrian Bunk } 95f11b00f3SAdrian Bunk 9603ba3782SJens Axboe static void bdi_work_clear(struct bdi_work *work) 97f11b00f3SAdrian Bunk { 9803ba3782SJens Axboe clear_bit(WS_USED_B, &work->state); 9903ba3782SJens Axboe smp_mb__after_clear_bit(); 1001ef7d9aaSNick Piggin /* 1011ef7d9aaSNick Piggin * work can have disappeared at this point. bit waitq functions 1021ef7d9aaSNick Piggin * should be able to tolerate this, provided bdi_sched_wait does 1031ef7d9aaSNick Piggin * not dereference it's pointer argument. 1041ef7d9aaSNick Piggin */ 10503ba3782SJens Axboe wake_up_bit(&work->state, WS_USED_B); 106f11b00f3SAdrian Bunk } 107f11b00f3SAdrian Bunk 10803ba3782SJens Axboe static void bdi_work_free(struct rcu_head *head) 1094195f73dSNick Piggin { 11003ba3782SJens Axboe struct bdi_work *work = container_of(head, struct bdi_work, rcu_head); 1114195f73dSNick Piggin 11203ba3782SJens Axboe if (!bdi_work_on_stack(work)) 11303ba3782SJens Axboe kfree(work); 11403ba3782SJens Axboe else 11503ba3782SJens Axboe bdi_work_clear(work); 1164195f73dSNick Piggin } 1174195f73dSNick Piggin 11803ba3782SJens Axboe static void wb_work_complete(struct bdi_work *work) 1191da177e4SLinus Torvalds { 120c4a77a6cSJens Axboe const enum writeback_sync_modes sync_mode = work->args.sync_mode; 12177b9d059SNick Piggin int onstack = bdi_work_on_stack(work); 1221da177e4SLinus Torvalds 1231da177e4SLinus Torvalds /* 12403ba3782SJens Axboe * For allocated work, we can clear the done/seen bit right here. 12503ba3782SJens Axboe * For on-stack work, we need to postpone both the clear and free 12603ba3782SJens Axboe * to after the RCU grace period, since the stack could be invalidated 12703ba3782SJens Axboe * as soon as bdi_work_clear() has done the wakeup. 1281da177e4SLinus Torvalds */ 12977b9d059SNick Piggin if (!onstack) 13003ba3782SJens Axboe bdi_work_clear(work); 13177b9d059SNick Piggin if (sync_mode == WB_SYNC_NONE || onstack) 13203ba3782SJens Axboe call_rcu(&work->rcu_head, bdi_work_free); 1331da177e4SLinus Torvalds } 1341da177e4SLinus Torvalds 13503ba3782SJens Axboe static void wb_clear_pending(struct bdi_writeback *wb, struct bdi_work *work) 13603ba3782SJens Axboe { 1371da177e4SLinus Torvalds /* 13803ba3782SJens Axboe * The caller has retrieved the work arguments from this work, 13903ba3782SJens Axboe * drop our reference. If this is the last ref, delete and free it 14003ba3782SJens Axboe */ 14103ba3782SJens Axboe if (atomic_dec_and_test(&work->pending)) { 14203ba3782SJens Axboe struct backing_dev_info *bdi = wb->bdi; 14303ba3782SJens Axboe 14403ba3782SJens Axboe spin_lock(&bdi->wb_lock); 14503ba3782SJens Axboe list_del_rcu(&work->list); 14603ba3782SJens Axboe spin_unlock(&bdi->wb_lock); 14703ba3782SJens Axboe 14803ba3782SJens Axboe wb_work_complete(work); 14903ba3782SJens Axboe } 15003ba3782SJens Axboe } 15103ba3782SJens Axboe 15203ba3782SJens Axboe static void bdi_queue_work(struct backing_dev_info *bdi, struct bdi_work *work) 15303ba3782SJens Axboe { 15403ba3782SJens Axboe work->seen = bdi->wb_mask; 15503ba3782SJens Axboe BUG_ON(!work->seen); 15603ba3782SJens Axboe atomic_set(&work->pending, bdi->wb_cnt); 15703ba3782SJens Axboe BUG_ON(!bdi->wb_cnt); 15803ba3782SJens Axboe 15903ba3782SJens Axboe /* 160deed62edSNick Piggin * list_add_tail_rcu() contains the necessary barriers to 161deed62edSNick Piggin * make sure the above stores are seen before the item is 162deed62edSNick Piggin * noticed on the list 1631da177e4SLinus Torvalds */ 16403ba3782SJens Axboe spin_lock(&bdi->wb_lock); 16503ba3782SJens Axboe list_add_tail_rcu(&work->list, &bdi->work_list); 16603ba3782SJens Axboe spin_unlock(&bdi->wb_lock); 1671da177e4SLinus Torvalds 1681da177e4SLinus Torvalds /* 16903ba3782SJens Axboe * If the default thread isn't there, make sure we add it. When 17003ba3782SJens Axboe * it gets created and wakes up, we'll run this work. 1711da177e4SLinus Torvalds */ 17203ba3782SJens Axboe if (unlikely(list_empty_careful(&bdi->wb_list))) 17303ba3782SJens Axboe wake_up_process(default_backing_dev_info.wb.task); 17403ba3782SJens Axboe else { 17503ba3782SJens Axboe struct bdi_writeback *wb = &bdi->wb; 1761da177e4SLinus Torvalds 1771ef7d9aaSNick Piggin if (wb->task) 17803ba3782SJens Axboe wake_up_process(wb->task); 1791da177e4SLinus Torvalds } 18003ba3782SJens Axboe } 1811da177e4SLinus Torvalds 1821da177e4SLinus Torvalds /* 18303ba3782SJens Axboe * Used for on-stack allocated work items. The caller needs to wait until 18403ba3782SJens Axboe * the wb threads have acked the work before it's safe to continue. 1851da177e4SLinus Torvalds */ 18603ba3782SJens Axboe static void bdi_wait_on_work_clear(struct bdi_work *work) 1871da177e4SLinus Torvalds { 18803ba3782SJens Axboe wait_on_bit(&work->state, WS_USED_B, bdi_sched_wait, 18903ba3782SJens Axboe TASK_UNINTERRUPTIBLE); 19003ba3782SJens Axboe } 19103ba3782SJens Axboe 192f11fcae8SJens Axboe static void bdi_alloc_queue_work(struct backing_dev_info *bdi, 193b6e51316SJens Axboe struct wb_writeback_args *args) 19403ba3782SJens Axboe { 19503ba3782SJens Axboe struct bdi_work *work; 19603ba3782SJens Axboe 197bcddc3f0SJens Axboe /* 198bcddc3f0SJens Axboe * This is WB_SYNC_NONE writeback, so if allocation fails just 199bcddc3f0SJens Axboe * wakeup the thread for old dirty data writeback 200bcddc3f0SJens Axboe */ 20103ba3782SJens Axboe work = kmalloc(sizeof(*work), GFP_ATOMIC); 202bcddc3f0SJens Axboe if (work) { 203b6e51316SJens Axboe bdi_work_init(work, args); 204f11fcae8SJens Axboe bdi_queue_work(bdi, work); 205bcddc3f0SJens Axboe } else { 206bcddc3f0SJens Axboe struct bdi_writeback *wb = &bdi->wb; 207bcddc3f0SJens Axboe 208bcddc3f0SJens Axboe if (wb->task) 209bcddc3f0SJens Axboe wake_up_process(wb->task); 210bcddc3f0SJens Axboe } 21103ba3782SJens Axboe } 21203ba3782SJens Axboe 213b6e51316SJens Axboe /** 214b6e51316SJens Axboe * bdi_sync_writeback - start and wait for writeback 215b6e51316SJens Axboe * @bdi: the backing device to write from 216b6e51316SJens Axboe * @sb: write inodes from this super_block 217b6e51316SJens Axboe * 218b6e51316SJens Axboe * Description: 219b6e51316SJens Axboe * This does WB_SYNC_ALL data integrity writeback and waits for the 220b6e51316SJens Axboe * IO to complete. Callers must hold the sb s_umount semaphore for 221b6e51316SJens Axboe * reading, to avoid having the super disappear before we are done. 22203ba3782SJens Axboe */ 223b6e51316SJens Axboe static void bdi_sync_writeback(struct backing_dev_info *bdi, 224b6e51316SJens Axboe struct super_block *sb) 225b6e51316SJens Axboe { 226b6e51316SJens Axboe struct wb_writeback_args args = { 227b6e51316SJens Axboe .sb = sb, 228b6e51316SJens Axboe .sync_mode = WB_SYNC_ALL, 229b6e51316SJens Axboe .nr_pages = LONG_MAX, 230b6e51316SJens Axboe .range_cyclic = 0, 231b6e51316SJens Axboe }; 232f0fad8a5SChristoph Hellwig struct bdi_work work; 233f0fad8a5SChristoph Hellwig 234b6e51316SJens Axboe bdi_work_init(&work, &args); 235f0fad8a5SChristoph Hellwig work.state |= WS_ONSTACK; 236f0fad8a5SChristoph Hellwig 237b6e51316SJens Axboe bdi_queue_work(bdi, &work); 238f0fad8a5SChristoph Hellwig bdi_wait_on_work_clear(&work); 23903ba3782SJens Axboe } 240b6e51316SJens Axboe 241b6e51316SJens Axboe /** 242b6e51316SJens Axboe * bdi_start_writeback - start writeback 243b6e51316SJens Axboe * @bdi: the backing device to write from 244b6e51316SJens Axboe * @nr_pages: the number of pages to write 245b6e51316SJens Axboe * 246b6e51316SJens Axboe * Description: 247b6e51316SJens Axboe * This does WB_SYNC_NONE opportunistic writeback. The IO is only 248b6e51316SJens Axboe * started when this function returns, we make no guarentees on 249b6e51316SJens Axboe * completion. Caller need not hold sb s_umount semaphore. 250b6e51316SJens Axboe * 251b6e51316SJens Axboe */ 252b6e51316SJens Axboe void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages) 253b6e51316SJens Axboe { 254b6e51316SJens Axboe struct wb_writeback_args args = { 255b6e51316SJens Axboe .sync_mode = WB_SYNC_NONE, 256b6e51316SJens Axboe .nr_pages = nr_pages, 257b6e51316SJens Axboe .range_cyclic = 1, 258b6e51316SJens Axboe }; 259b6e51316SJens Axboe 260b6e51316SJens Axboe bdi_alloc_queue_work(bdi, &args); 2611da177e4SLinus Torvalds } 2621da177e4SLinus Torvalds 2631da177e4SLinus Torvalds /* 2646610a0bcSAndrew Morton * Redirty an inode: set its when-it-was dirtied timestamp and move it to the 2656610a0bcSAndrew Morton * furthest end of its superblock's dirty-inode list. 2666610a0bcSAndrew Morton * 2676610a0bcSAndrew Morton * Before stamping the inode's ->dirtied_when, we check to see whether it is 26866f3b8e2SJens Axboe * already the most-recently-dirtied inode on the b_dirty list. If that is 2696610a0bcSAndrew Morton * the case then the inode must have been redirtied while it was being written 2706610a0bcSAndrew Morton * out and we don't reset its dirtied_when. 2716610a0bcSAndrew Morton */ 2726610a0bcSAndrew Morton static void redirty_tail(struct inode *inode) 2736610a0bcSAndrew Morton { 27403ba3782SJens Axboe struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; 2756610a0bcSAndrew Morton 27603ba3782SJens Axboe if (!list_empty(&wb->b_dirty)) { 27766f3b8e2SJens Axboe struct inode *tail; 2786610a0bcSAndrew Morton 27903ba3782SJens Axboe tail = list_entry(wb->b_dirty.next, struct inode, i_list); 28066f3b8e2SJens Axboe if (time_before(inode->dirtied_when, tail->dirtied_when)) 2816610a0bcSAndrew Morton inode->dirtied_when = jiffies; 2826610a0bcSAndrew Morton } 28303ba3782SJens Axboe list_move(&inode->i_list, &wb->b_dirty); 2846610a0bcSAndrew Morton } 2856610a0bcSAndrew Morton 2866610a0bcSAndrew Morton /* 28766f3b8e2SJens Axboe * requeue inode for re-scanning after bdi->b_io list is exhausted. 288c986d1e2SAndrew Morton */ 2890e0f4fc2SKen Chen static void requeue_io(struct inode *inode) 290c986d1e2SAndrew Morton { 29103ba3782SJens Axboe struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; 29203ba3782SJens Axboe 29303ba3782SJens Axboe list_move(&inode->i_list, &wb->b_more_io); 294c986d1e2SAndrew Morton } 295c986d1e2SAndrew Morton 2961c0eeaf5SJoern Engel static void inode_sync_complete(struct inode *inode) 2971c0eeaf5SJoern Engel { 2981c0eeaf5SJoern Engel /* 2991c0eeaf5SJoern Engel * Prevent speculative execution through spin_unlock(&inode_lock); 3001c0eeaf5SJoern Engel */ 3011c0eeaf5SJoern Engel smp_mb(); 3021c0eeaf5SJoern Engel wake_up_bit(&inode->i_state, __I_SYNC); 3031c0eeaf5SJoern Engel } 3041c0eeaf5SJoern Engel 305d2caa3c5SJeff Layton static bool inode_dirtied_after(struct inode *inode, unsigned long t) 306d2caa3c5SJeff Layton { 307d2caa3c5SJeff Layton bool ret = time_after(inode->dirtied_when, t); 308d2caa3c5SJeff Layton #ifndef CONFIG_64BIT 309d2caa3c5SJeff Layton /* 310d2caa3c5SJeff Layton * For inodes being constantly redirtied, dirtied_when can get stuck. 311d2caa3c5SJeff Layton * It _appears_ to be in the future, but is actually in distant past. 312d2caa3c5SJeff Layton * This test is necessary to prevent such wrapped-around relative times 313d2caa3c5SJeff Layton * from permanently stopping the whole pdflush writeback. 314d2caa3c5SJeff Layton */ 315d2caa3c5SJeff Layton ret = ret && time_before_eq(inode->dirtied_when, jiffies); 316d2caa3c5SJeff Layton #endif 317d2caa3c5SJeff Layton return ret; 318d2caa3c5SJeff Layton } 319d2caa3c5SJeff Layton 320c986d1e2SAndrew Morton /* 3212c136579SFengguang Wu * Move expired dirty inodes from @delaying_queue to @dispatch_queue. 3222c136579SFengguang Wu */ 3232c136579SFengguang Wu static void move_expired_inodes(struct list_head *delaying_queue, 3242c136579SFengguang Wu struct list_head *dispatch_queue, 3252c136579SFengguang Wu unsigned long *older_than_this) 3262c136579SFengguang Wu { 3272c136579SFengguang Wu while (!list_empty(delaying_queue)) { 3282c136579SFengguang Wu struct inode *inode = list_entry(delaying_queue->prev, 3292c136579SFengguang Wu struct inode, i_list); 3302c136579SFengguang Wu if (older_than_this && 331d2caa3c5SJeff Layton inode_dirtied_after(inode, *older_than_this)) 3322c136579SFengguang Wu break; 3332c136579SFengguang Wu list_move(&inode->i_list, dispatch_queue); 3342c136579SFengguang Wu } 3352c136579SFengguang Wu } 3362c136579SFengguang Wu 3372c136579SFengguang Wu /* 3382c136579SFengguang Wu * Queue all expired dirty inodes for io, eldest first. 3392c136579SFengguang Wu */ 34003ba3782SJens Axboe static void queue_io(struct bdi_writeback *wb, unsigned long *older_than_this) 3412c136579SFengguang Wu { 34203ba3782SJens Axboe list_splice_init(&wb->b_more_io, wb->b_io.prev); 34303ba3782SJens Axboe move_expired_inodes(&wb->b_dirty, &wb->b_io, older_than_this); 34466f3b8e2SJens Axboe } 34566f3b8e2SJens Axboe 34603ba3782SJens Axboe static int write_inode(struct inode *inode, int sync) 34766f3b8e2SJens Axboe { 34803ba3782SJens Axboe if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) 34903ba3782SJens Axboe return inode->i_sb->s_op->write_inode(inode, sync); 35003ba3782SJens Axboe return 0; 35166f3b8e2SJens Axboe } 35208d8e974SFengguang Wu 3532c136579SFengguang Wu /* 35401c03194SChristoph Hellwig * Wait for writeback on an inode to complete. 35501c03194SChristoph Hellwig */ 35601c03194SChristoph Hellwig static void inode_wait_for_writeback(struct inode *inode) 35701c03194SChristoph Hellwig { 35801c03194SChristoph Hellwig DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); 35901c03194SChristoph Hellwig wait_queue_head_t *wqh; 36001c03194SChristoph Hellwig 36101c03194SChristoph Hellwig wqh = bit_waitqueue(&inode->i_state, __I_SYNC); 36201c03194SChristoph Hellwig do { 36301c03194SChristoph Hellwig spin_unlock(&inode_lock); 36401c03194SChristoph Hellwig __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE); 36501c03194SChristoph Hellwig spin_lock(&inode_lock); 36601c03194SChristoph Hellwig } while (inode->i_state & I_SYNC); 36701c03194SChristoph Hellwig } 36801c03194SChristoph Hellwig 36901c03194SChristoph Hellwig /* 37001c03194SChristoph Hellwig * Write out an inode's dirty pages. Called under inode_lock. Either the 37101c03194SChristoph Hellwig * caller has ref on the inode (either via __iget or via syscall against an fd) 37201c03194SChristoph Hellwig * or the inode has I_WILL_FREE set (via generic_forget_inode) 37301c03194SChristoph Hellwig * 3741da177e4SLinus Torvalds * If `wait' is set, wait on the writeout. 3751da177e4SLinus Torvalds * 3761da177e4SLinus Torvalds * The whole writeout design is quite complex and fragile. We want to avoid 3771da177e4SLinus Torvalds * starvation of particular inodes when others are being redirtied, prevent 3781da177e4SLinus Torvalds * livelocks, etc. 3791da177e4SLinus Torvalds * 3801da177e4SLinus Torvalds * Called under inode_lock. 3811da177e4SLinus Torvalds */ 3821da177e4SLinus Torvalds static int 38301c03194SChristoph Hellwig writeback_single_inode(struct inode *inode, struct writeback_control *wbc) 3841da177e4SLinus Torvalds { 3851da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 3861da177e4SLinus Torvalds int wait = wbc->sync_mode == WB_SYNC_ALL; 38701c03194SChristoph Hellwig unsigned dirty; 3881da177e4SLinus Torvalds int ret; 3891da177e4SLinus Torvalds 39001c03194SChristoph Hellwig if (!atomic_read(&inode->i_count)) 39101c03194SChristoph Hellwig WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); 39201c03194SChristoph Hellwig else 39301c03194SChristoph Hellwig WARN_ON(inode->i_state & I_WILL_FREE); 39401c03194SChristoph Hellwig 39501c03194SChristoph Hellwig if (inode->i_state & I_SYNC) { 39601c03194SChristoph Hellwig /* 39701c03194SChristoph Hellwig * If this inode is locked for writeback and we are not doing 39866f3b8e2SJens Axboe * writeback-for-data-integrity, move it to b_more_io so that 39901c03194SChristoph Hellwig * writeback can proceed with the other inodes on s_io. 40001c03194SChristoph Hellwig * 40101c03194SChristoph Hellwig * We'll have another go at writing back this inode when we 40266f3b8e2SJens Axboe * completed a full scan of b_io. 40301c03194SChristoph Hellwig */ 40401c03194SChristoph Hellwig if (!wait) { 40501c03194SChristoph Hellwig requeue_io(inode); 40601c03194SChristoph Hellwig return 0; 40701c03194SChristoph Hellwig } 40801c03194SChristoph Hellwig 40901c03194SChristoph Hellwig /* 41001c03194SChristoph Hellwig * It's a data-integrity sync. We must wait. 41101c03194SChristoph Hellwig */ 41201c03194SChristoph Hellwig inode_wait_for_writeback(inode); 41301c03194SChristoph Hellwig } 41401c03194SChristoph Hellwig 4151c0eeaf5SJoern Engel BUG_ON(inode->i_state & I_SYNC); 4161da177e4SLinus Torvalds 4171c0eeaf5SJoern Engel /* Set I_SYNC, reset I_DIRTY */ 4181da177e4SLinus Torvalds dirty = inode->i_state & I_DIRTY; 4191c0eeaf5SJoern Engel inode->i_state |= I_SYNC; 4201da177e4SLinus Torvalds inode->i_state &= ~I_DIRTY; 4211da177e4SLinus Torvalds 4221da177e4SLinus Torvalds spin_unlock(&inode_lock); 4231da177e4SLinus Torvalds 4241da177e4SLinus Torvalds ret = do_writepages(mapping, wbc); 4251da177e4SLinus Torvalds 4261da177e4SLinus Torvalds /* Don't write the inode if only I_DIRTY_PAGES was set */ 4271da177e4SLinus Torvalds if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { 4281da177e4SLinus Torvalds int err = write_inode(inode, wait); 4291da177e4SLinus Torvalds if (ret == 0) 4301da177e4SLinus Torvalds ret = err; 4311da177e4SLinus Torvalds } 4321da177e4SLinus Torvalds 4331da177e4SLinus Torvalds if (wait) { 4341da177e4SLinus Torvalds int err = filemap_fdatawait(mapping); 4351da177e4SLinus Torvalds if (ret == 0) 4361da177e4SLinus Torvalds ret = err; 4371da177e4SLinus Torvalds } 4381da177e4SLinus Torvalds 4391da177e4SLinus Torvalds spin_lock(&inode_lock); 4401c0eeaf5SJoern Engel inode->i_state &= ~I_SYNC; 44184a89245SWu Fengguang if (!(inode->i_state & (I_FREEING | I_CLEAR))) { 4421da177e4SLinus Torvalds if (!(inode->i_state & I_DIRTY) && 4431da177e4SLinus Torvalds mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { 4441da177e4SLinus Torvalds /* 4451da177e4SLinus Torvalds * We didn't write back all the pages. nfs_writepages() 4461da177e4SLinus Torvalds * sometimes bales out without doing anything. Redirty 44766f3b8e2SJens Axboe * the inode; Move it from b_io onto b_more_io/b_dirty. 4481b43ef91SAndrew Morton */ 4491b43ef91SAndrew Morton /* 4501b43ef91SAndrew Morton * akpm: if the caller was the kupdate function we put 45166f3b8e2SJens Axboe * this inode at the head of b_dirty so it gets first 4521b43ef91SAndrew Morton * consideration. Otherwise, move it to the tail, for 4531b43ef91SAndrew Morton * the reasons described there. I'm not really sure 4541b43ef91SAndrew Morton * how much sense this makes. Presumably I had a good 4551b43ef91SAndrew Morton * reasons for doing it this way, and I'd rather not 4561b43ef91SAndrew Morton * muck with it at present. 4571da177e4SLinus Torvalds */ 4581da177e4SLinus Torvalds if (wbc->for_kupdate) { 4591da177e4SLinus Torvalds /* 4602c136579SFengguang Wu * For the kupdate function we move the inode 46166f3b8e2SJens Axboe * to b_more_io so it will get more writeout as 4622c136579SFengguang Wu * soon as the queue becomes uncongested. 4631da177e4SLinus Torvalds */ 4641da177e4SLinus Torvalds inode->i_state |= I_DIRTY_PAGES; 4658bc3be27SFengguang Wu if (wbc->nr_to_write <= 0) { 4668bc3be27SFengguang Wu /* 4678bc3be27SFengguang Wu * slice used up: queue for next turn 4688bc3be27SFengguang Wu */ 4690e0f4fc2SKen Chen requeue_io(inode); 4701da177e4SLinus Torvalds } else { 4711da177e4SLinus Torvalds /* 4728bc3be27SFengguang Wu * somehow blocked: retry later 4738bc3be27SFengguang Wu */ 4748bc3be27SFengguang Wu redirty_tail(inode); 4758bc3be27SFengguang Wu } 4768bc3be27SFengguang Wu } else { 4778bc3be27SFengguang Wu /* 4781da177e4SLinus Torvalds * Otherwise fully redirty the inode so that 4791da177e4SLinus Torvalds * other inodes on this superblock will get some 4801da177e4SLinus Torvalds * writeout. Otherwise heavy writing to one 4811da177e4SLinus Torvalds * file would indefinitely suspend writeout of 4821da177e4SLinus Torvalds * all the other files. 4831da177e4SLinus Torvalds */ 4841da177e4SLinus Torvalds inode->i_state |= I_DIRTY_PAGES; 4851b43ef91SAndrew Morton redirty_tail(inode); 4861da177e4SLinus Torvalds } 4871da177e4SLinus Torvalds } else if (inode->i_state & I_DIRTY) { 4881da177e4SLinus Torvalds /* 4891da177e4SLinus Torvalds * Someone redirtied the inode while were writing back 4901da177e4SLinus Torvalds * the pages. 4911da177e4SLinus Torvalds */ 4926610a0bcSAndrew Morton redirty_tail(inode); 4931da177e4SLinus Torvalds } else if (atomic_read(&inode->i_count)) { 4941da177e4SLinus Torvalds /* 4951da177e4SLinus Torvalds * The inode is clean, inuse 4961da177e4SLinus Torvalds */ 4971da177e4SLinus Torvalds list_move(&inode->i_list, &inode_in_use); 4981da177e4SLinus Torvalds } else { 4991da177e4SLinus Torvalds /* 5001da177e4SLinus Torvalds * The inode is clean, unused 5011da177e4SLinus Torvalds */ 5021da177e4SLinus Torvalds list_move(&inode->i_list, &inode_unused); 5031da177e4SLinus Torvalds } 5041da177e4SLinus Torvalds } 5051c0eeaf5SJoern Engel inode_sync_complete(inode); 5061da177e4SLinus Torvalds return ret; 5071da177e4SLinus Torvalds } 5081da177e4SLinus Torvalds 50903ba3782SJens Axboe /* 51003ba3782SJens Axboe * For WB_SYNC_NONE writeback, the caller does not have the sb pinned 51103ba3782SJens Axboe * before calling writeback. So make sure that we do pin it, so it doesn't 51203ba3782SJens Axboe * go away while we are writing inodes from it. 51303ba3782SJens Axboe * 51403ba3782SJens Axboe * Returns 0 if the super was successfully pinned (or pinning wasn't needed), 51503ba3782SJens Axboe * 1 if we failed. 51603ba3782SJens Axboe */ 51703ba3782SJens Axboe static int pin_sb_for_writeback(struct writeback_control *wbc, 51803ba3782SJens Axboe struct inode *inode) 5191da177e4SLinus Torvalds { 52003ba3782SJens Axboe struct super_block *sb = inode->i_sb; 52103ba3782SJens Axboe 52203ba3782SJens Axboe /* 52303ba3782SJens Axboe * Caller must already hold the ref for this 52403ba3782SJens Axboe */ 52503ba3782SJens Axboe if (wbc->sync_mode == WB_SYNC_ALL) { 52603ba3782SJens Axboe WARN_ON(!rwsem_is_locked(&sb->s_umount)); 52703ba3782SJens Axboe return 0; 52803ba3782SJens Axboe } 52903ba3782SJens Axboe 53003ba3782SJens Axboe spin_lock(&sb_lock); 53103ba3782SJens Axboe sb->s_count++; 53203ba3782SJens Axboe if (down_read_trylock(&sb->s_umount)) { 53303ba3782SJens Axboe if (sb->s_root) { 53403ba3782SJens Axboe spin_unlock(&sb_lock); 53503ba3782SJens Axboe return 0; 53603ba3782SJens Axboe } 53703ba3782SJens Axboe /* 53803ba3782SJens Axboe * umounted, drop rwsem again and fall through to failure 53903ba3782SJens Axboe */ 54003ba3782SJens Axboe up_read(&sb->s_umount); 54103ba3782SJens Axboe } 54203ba3782SJens Axboe 54303ba3782SJens Axboe sb->s_count--; 54403ba3782SJens Axboe spin_unlock(&sb_lock); 54503ba3782SJens Axboe return 1; 54603ba3782SJens Axboe } 54703ba3782SJens Axboe 54803ba3782SJens Axboe static void unpin_sb_for_writeback(struct writeback_control *wbc, 54903ba3782SJens Axboe struct inode *inode) 55003ba3782SJens Axboe { 55103ba3782SJens Axboe struct super_block *sb = inode->i_sb; 55203ba3782SJens Axboe 55303ba3782SJens Axboe if (wbc->sync_mode == WB_SYNC_ALL) 55403ba3782SJens Axboe return; 55503ba3782SJens Axboe 55603ba3782SJens Axboe up_read(&sb->s_umount); 55703ba3782SJens Axboe put_super(sb); 55803ba3782SJens Axboe } 55903ba3782SJens Axboe 56003ba3782SJens Axboe static void writeback_inodes_wb(struct bdi_writeback *wb, 56103ba3782SJens Axboe struct writeback_control *wbc) 56203ba3782SJens Axboe { 56303ba3782SJens Axboe struct super_block *sb = wbc->sb; 56466f3b8e2SJens Axboe const int is_blkdev_sb = sb_is_blkdev_sb(sb); 5651da177e4SLinus Torvalds const unsigned long start = jiffies; /* livelock avoidance */ 5661da177e4SLinus Torvalds 567ae8547b0SHans Reiser spin_lock(&inode_lock); 5681da177e4SLinus Torvalds 56903ba3782SJens Axboe if (!wbc->for_kupdate || list_empty(&wb->b_io)) 57003ba3782SJens Axboe queue_io(wb, wbc->older_than_this); 57166f3b8e2SJens Axboe 57203ba3782SJens Axboe while (!list_empty(&wb->b_io)) { 57303ba3782SJens Axboe struct inode *inode = list_entry(wb->b_io.prev, 5741da177e4SLinus Torvalds struct inode, i_list); 5751da177e4SLinus Torvalds long pages_skipped; 5761da177e4SLinus Torvalds 57766f3b8e2SJens Axboe /* 57866f3b8e2SJens Axboe * super block given and doesn't match, skip this inode 57966f3b8e2SJens Axboe */ 58066f3b8e2SJens Axboe if (sb && sb != inode->i_sb) { 58166f3b8e2SJens Axboe redirty_tail(inode); 58266f3b8e2SJens Axboe continue; 58366f3b8e2SJens Axboe } 58466f3b8e2SJens Axboe 58503ba3782SJens Axboe if (!bdi_cap_writeback_dirty(wb->bdi)) { 5869852a0e7SAndrew Morton redirty_tail(inode); 58766f3b8e2SJens Axboe if (is_blkdev_sb) { 5881da177e4SLinus Torvalds /* 5891da177e4SLinus Torvalds * Dirty memory-backed blockdev: the ramdisk 5901da177e4SLinus Torvalds * driver does this. Skip just this inode 5911da177e4SLinus Torvalds */ 5921da177e4SLinus Torvalds continue; 5931da177e4SLinus Torvalds } 5941da177e4SLinus Torvalds /* 5951da177e4SLinus Torvalds * Dirty memory-backed inode against a filesystem other 5961da177e4SLinus Torvalds * than the kernel-internal bdev filesystem. Skip the 5971da177e4SLinus Torvalds * entire superblock. 5981da177e4SLinus Torvalds */ 5991da177e4SLinus Torvalds break; 6001da177e4SLinus Torvalds } 6011da177e4SLinus Torvalds 60284a89245SWu Fengguang if (inode->i_state & (I_NEW | I_WILL_FREE)) { 6037ef0d737SNick Piggin requeue_io(inode); 6047ef0d737SNick Piggin continue; 6057ef0d737SNick Piggin } 6067ef0d737SNick Piggin 60703ba3782SJens Axboe if (wbc->nonblocking && bdi_write_congested(wb->bdi)) { 6081da177e4SLinus Torvalds wbc->encountered_congestion = 1; 60966f3b8e2SJens Axboe if (!is_blkdev_sb) 6101da177e4SLinus Torvalds break; /* Skip a congested fs */ 6110e0f4fc2SKen Chen requeue_io(inode); 6121da177e4SLinus Torvalds continue; /* Skip a congested blockdev */ 6131da177e4SLinus Torvalds } 6141da177e4SLinus Torvalds 615d2caa3c5SJeff Layton /* 616d2caa3c5SJeff Layton * Was this inode dirtied after sync_sb_inodes was called? 617d2caa3c5SJeff Layton * This keeps sync from extra jobs and livelock. 618d2caa3c5SJeff Layton */ 619d2caa3c5SJeff Layton if (inode_dirtied_after(inode, start)) 6201da177e4SLinus Torvalds break; 6211da177e4SLinus Torvalds 62203ba3782SJens Axboe if (pin_sb_for_writeback(wbc, inode)) { 62303ba3782SJens Axboe requeue_io(inode); 62403ba3782SJens Axboe continue; 62503ba3782SJens Axboe } 6261da177e4SLinus Torvalds 62784a89245SWu Fengguang BUG_ON(inode->i_state & (I_FREEING | I_CLEAR)); 6281da177e4SLinus Torvalds __iget(inode); 6291da177e4SLinus Torvalds pages_skipped = wbc->pages_skipped; 63001c03194SChristoph Hellwig writeback_single_inode(inode, wbc); 63103ba3782SJens Axboe unpin_sb_for_writeback(wbc, inode); 6321da177e4SLinus Torvalds if (wbc->pages_skipped != pages_skipped) { 6331da177e4SLinus Torvalds /* 6341da177e4SLinus Torvalds * writeback is not making progress due to locked 6351da177e4SLinus Torvalds * buffers. Skip this inode for now. 6361da177e4SLinus Torvalds */ 637f57b9b7bSAndrew Morton redirty_tail(inode); 6381da177e4SLinus Torvalds } 6391da177e4SLinus Torvalds spin_unlock(&inode_lock); 6401da177e4SLinus Torvalds iput(inode); 6414ffc8444SOGAWA Hirofumi cond_resched(); 6421da177e4SLinus Torvalds spin_lock(&inode_lock); 6438bc3be27SFengguang Wu if (wbc->nr_to_write <= 0) { 6448bc3be27SFengguang Wu wbc->more_io = 1; 6451da177e4SLinus Torvalds break; 6461da177e4SLinus Torvalds } 64703ba3782SJens Axboe if (!list_empty(&wb->b_more_io)) 6488bc3be27SFengguang Wu wbc->more_io = 1; 6498bc3be27SFengguang Wu } 65038f21977SNick Piggin 65166f3b8e2SJens Axboe spin_unlock(&inode_lock); 65266f3b8e2SJens Axboe /* Leave any unwritten inodes on b_io */ 65366f3b8e2SJens Axboe } 65466f3b8e2SJens Axboe 65503ba3782SJens Axboe void writeback_inodes_wbc(struct writeback_control *wbc) 65603ba3782SJens Axboe { 65703ba3782SJens Axboe struct backing_dev_info *bdi = wbc->bdi; 65803ba3782SJens Axboe 65903ba3782SJens Axboe writeback_inodes_wb(&bdi->wb, wbc); 66003ba3782SJens Axboe } 66103ba3782SJens Axboe 66203ba3782SJens Axboe /* 66303ba3782SJens Axboe * The maximum number of pages to writeout in a single bdi flush/kupdate 66403ba3782SJens Axboe * operation. We do this so we don't hold I_SYNC against an inode for 66503ba3782SJens Axboe * enormous amounts of time, which would block a userspace task which has 66603ba3782SJens Axboe * been forced to throttle against that inode. Also, the code reevaluates 66703ba3782SJens Axboe * the dirty each time it has written this many pages. 66803ba3782SJens Axboe */ 66903ba3782SJens Axboe #define MAX_WRITEBACK_PAGES 1024 67003ba3782SJens Axboe 67103ba3782SJens Axboe static inline bool over_bground_thresh(void) 67203ba3782SJens Axboe { 67303ba3782SJens Axboe unsigned long background_thresh, dirty_thresh; 67403ba3782SJens Axboe 67503ba3782SJens Axboe get_dirty_limits(&background_thresh, &dirty_thresh, NULL, NULL); 67603ba3782SJens Axboe 67703ba3782SJens Axboe return (global_page_state(NR_FILE_DIRTY) + 67803ba3782SJens Axboe global_page_state(NR_UNSTABLE_NFS) >= background_thresh); 67903ba3782SJens Axboe } 68003ba3782SJens Axboe 68103ba3782SJens Axboe /* 68203ba3782SJens Axboe * Explicit flushing or periodic writeback of "old" data. 68303ba3782SJens Axboe * 68403ba3782SJens Axboe * Define "old": the first time one of an inode's pages is dirtied, we mark the 68503ba3782SJens Axboe * dirtying-time in the inode's address_space. So this periodic writeback code 68603ba3782SJens Axboe * just walks the superblock inode list, writing back any inodes which are 68703ba3782SJens Axboe * older than a specific point in time. 68803ba3782SJens Axboe * 68903ba3782SJens Axboe * Try to run once per dirty_writeback_interval. But if a writeback event 69003ba3782SJens Axboe * takes longer than a dirty_writeback_interval interval, then leave a 69103ba3782SJens Axboe * one-second gap. 69203ba3782SJens Axboe * 69303ba3782SJens Axboe * older_than_this takes precedence over nr_to_write. So we'll only write back 69403ba3782SJens Axboe * all dirty pages if they are all attached to "old" mappings. 69503ba3782SJens Axboe */ 696c4a77a6cSJens Axboe static long wb_writeback(struct bdi_writeback *wb, 697c4a77a6cSJens Axboe struct wb_writeback_args *args) 69803ba3782SJens Axboe { 69903ba3782SJens Axboe struct writeback_control wbc = { 70003ba3782SJens Axboe .bdi = wb->bdi, 701c4a77a6cSJens Axboe .sb = args->sb, 702c4a77a6cSJens Axboe .sync_mode = args->sync_mode, 70303ba3782SJens Axboe .older_than_this = NULL, 704c4a77a6cSJens Axboe .for_kupdate = args->for_kupdate, 705c4a77a6cSJens Axboe .range_cyclic = args->range_cyclic, 70603ba3782SJens Axboe }; 70703ba3782SJens Axboe unsigned long oldest_jif; 70803ba3782SJens Axboe long wrote = 0; 70903ba3782SJens Axboe 71003ba3782SJens Axboe if (wbc.for_kupdate) { 71103ba3782SJens Axboe wbc.older_than_this = &oldest_jif; 71203ba3782SJens Axboe oldest_jif = jiffies - 71303ba3782SJens Axboe msecs_to_jiffies(dirty_expire_interval * 10); 71403ba3782SJens Axboe } 715c4a77a6cSJens Axboe if (!wbc.range_cyclic) { 716c4a77a6cSJens Axboe wbc.range_start = 0; 717c4a77a6cSJens Axboe wbc.range_end = LLONG_MAX; 718c4a77a6cSJens Axboe } 71903ba3782SJens Axboe 72003ba3782SJens Axboe for (;;) { 72103ba3782SJens Axboe /* 72203ba3782SJens Axboe * Don't flush anything for non-integrity writeback where 72303ba3782SJens Axboe * no nr_pages was given 72403ba3782SJens Axboe */ 725c4a77a6cSJens Axboe if (!args->for_kupdate && args->nr_pages <= 0 && 726c4a77a6cSJens Axboe args->sync_mode == WB_SYNC_NONE) 72703ba3782SJens Axboe break; 72803ba3782SJens Axboe 72903ba3782SJens Axboe /* 73003ba3782SJens Axboe * If no specific pages were given and this is just a 73103ba3782SJens Axboe * periodic background writeout and we are below the 73203ba3782SJens Axboe * background dirty threshold, don't do anything 73303ba3782SJens Axboe */ 734c4a77a6cSJens Axboe if (args->for_kupdate && args->nr_pages <= 0 && 735c4a77a6cSJens Axboe !over_bground_thresh()) 73603ba3782SJens Axboe break; 73703ba3782SJens Axboe 73803ba3782SJens Axboe wbc.more_io = 0; 73903ba3782SJens Axboe wbc.encountered_congestion = 0; 74003ba3782SJens Axboe wbc.nr_to_write = MAX_WRITEBACK_PAGES; 74103ba3782SJens Axboe wbc.pages_skipped = 0; 74203ba3782SJens Axboe writeback_inodes_wb(wb, &wbc); 743c4a77a6cSJens Axboe args->nr_pages -= MAX_WRITEBACK_PAGES - wbc.nr_to_write; 74403ba3782SJens Axboe wrote += MAX_WRITEBACK_PAGES - wbc.nr_to_write; 74503ba3782SJens Axboe 74603ba3782SJens Axboe /* 74703ba3782SJens Axboe * If we ran out of stuff to write, bail unless more_io got set 74803ba3782SJens Axboe */ 74903ba3782SJens Axboe if (wbc.nr_to_write > 0 || wbc.pages_skipped > 0) { 75003ba3782SJens Axboe if (wbc.more_io && !wbc.for_kupdate) 75103ba3782SJens Axboe continue; 75203ba3782SJens Axboe break; 75303ba3782SJens Axboe } 75403ba3782SJens Axboe } 75503ba3782SJens Axboe 75603ba3782SJens Axboe return wrote; 75703ba3782SJens Axboe } 75803ba3782SJens Axboe 75903ba3782SJens Axboe /* 76003ba3782SJens Axboe * Return the next bdi_work struct that hasn't been processed by this 7618010c3b6SJens Axboe * wb thread yet. ->seen is initially set for each thread that exists 7628010c3b6SJens Axboe * for this device, when a thread first notices a piece of work it 7638010c3b6SJens Axboe * clears its bit. Depending on writeback type, the thread will notify 7648010c3b6SJens Axboe * completion on either receiving the work (WB_SYNC_NONE) or after 7658010c3b6SJens Axboe * it is done (WB_SYNC_ALL). 76603ba3782SJens Axboe */ 76703ba3782SJens Axboe static struct bdi_work *get_next_work_item(struct backing_dev_info *bdi, 76803ba3782SJens Axboe struct bdi_writeback *wb) 76903ba3782SJens Axboe { 77003ba3782SJens Axboe struct bdi_work *work, *ret = NULL; 77103ba3782SJens Axboe 77203ba3782SJens Axboe rcu_read_lock(); 77303ba3782SJens Axboe 77403ba3782SJens Axboe list_for_each_entry_rcu(work, &bdi->work_list, list) { 77577fad5e6SNick Piggin if (!test_bit(wb->nr, &work->seen)) 77603ba3782SJens Axboe continue; 77777fad5e6SNick Piggin clear_bit(wb->nr, &work->seen); 77803ba3782SJens Axboe 77903ba3782SJens Axboe ret = work; 78003ba3782SJens Axboe break; 78103ba3782SJens Axboe } 78203ba3782SJens Axboe 78303ba3782SJens Axboe rcu_read_unlock(); 78403ba3782SJens Axboe return ret; 78503ba3782SJens Axboe } 78603ba3782SJens Axboe 78703ba3782SJens Axboe static long wb_check_old_data_flush(struct bdi_writeback *wb) 78803ba3782SJens Axboe { 78903ba3782SJens Axboe unsigned long expired; 79003ba3782SJens Axboe long nr_pages; 79103ba3782SJens Axboe 79203ba3782SJens Axboe expired = wb->last_old_flush + 79303ba3782SJens Axboe msecs_to_jiffies(dirty_writeback_interval * 10); 79403ba3782SJens Axboe if (time_before(jiffies, expired)) 79503ba3782SJens Axboe return 0; 79603ba3782SJens Axboe 79703ba3782SJens Axboe wb->last_old_flush = jiffies; 79803ba3782SJens Axboe nr_pages = global_page_state(NR_FILE_DIRTY) + 79903ba3782SJens Axboe global_page_state(NR_UNSTABLE_NFS) + 80003ba3782SJens Axboe (inodes_stat.nr_inodes - inodes_stat.nr_unused); 80103ba3782SJens Axboe 802c4a77a6cSJens Axboe if (nr_pages) { 803c4a77a6cSJens Axboe struct wb_writeback_args args = { 804c4a77a6cSJens Axboe .nr_pages = nr_pages, 805c4a77a6cSJens Axboe .sync_mode = WB_SYNC_NONE, 806c4a77a6cSJens Axboe .for_kupdate = 1, 807c4a77a6cSJens Axboe .range_cyclic = 1, 808c4a77a6cSJens Axboe }; 809c4a77a6cSJens Axboe 810c4a77a6cSJens Axboe return wb_writeback(wb, &args); 811c4a77a6cSJens Axboe } 81203ba3782SJens Axboe 81303ba3782SJens Axboe return 0; 81403ba3782SJens Axboe } 81503ba3782SJens Axboe 81603ba3782SJens Axboe /* 81703ba3782SJens Axboe * Retrieve work items and do the writeback they describe 81803ba3782SJens Axboe */ 81903ba3782SJens Axboe long wb_do_writeback(struct bdi_writeback *wb, int force_wait) 82003ba3782SJens Axboe { 82103ba3782SJens Axboe struct backing_dev_info *bdi = wb->bdi; 82203ba3782SJens Axboe struct bdi_work *work; 823c4a77a6cSJens Axboe long wrote = 0; 82403ba3782SJens Axboe 82503ba3782SJens Axboe while ((work = get_next_work_item(bdi, wb)) != NULL) { 826c4a77a6cSJens Axboe struct wb_writeback_args args = work->args; 82703ba3782SJens Axboe 82803ba3782SJens Axboe /* 82903ba3782SJens Axboe * Override sync mode, in case we must wait for completion 83003ba3782SJens Axboe */ 83103ba3782SJens Axboe if (force_wait) 832c4a77a6cSJens Axboe work->args.sync_mode = args.sync_mode = WB_SYNC_ALL; 83303ba3782SJens Axboe 83403ba3782SJens Axboe /* 83503ba3782SJens Axboe * If this isn't a data integrity operation, just notify 83603ba3782SJens Axboe * that we have seen this work and we are now starting it. 83703ba3782SJens Axboe */ 838c4a77a6cSJens Axboe if (args.sync_mode == WB_SYNC_NONE) 83903ba3782SJens Axboe wb_clear_pending(wb, work); 84003ba3782SJens Axboe 841c4a77a6cSJens Axboe wrote += wb_writeback(wb, &args); 84203ba3782SJens Axboe 84303ba3782SJens Axboe /* 84403ba3782SJens Axboe * This is a data integrity writeback, so only do the 84503ba3782SJens Axboe * notification when we have completed the work. 84603ba3782SJens Axboe */ 847c4a77a6cSJens Axboe if (args.sync_mode == WB_SYNC_ALL) 84803ba3782SJens Axboe wb_clear_pending(wb, work); 84903ba3782SJens Axboe } 85003ba3782SJens Axboe 85103ba3782SJens Axboe /* 85203ba3782SJens Axboe * Check for periodic writeback, kupdated() style 85303ba3782SJens Axboe */ 85403ba3782SJens Axboe wrote += wb_check_old_data_flush(wb); 85503ba3782SJens Axboe 85603ba3782SJens Axboe return wrote; 85703ba3782SJens Axboe } 85803ba3782SJens Axboe 85903ba3782SJens Axboe /* 86003ba3782SJens Axboe * Handle writeback of dirty data for the device backed by this bdi. Also 86103ba3782SJens Axboe * wakes up periodically and does kupdated style flushing. 86203ba3782SJens Axboe */ 86303ba3782SJens Axboe int bdi_writeback_task(struct bdi_writeback *wb) 86403ba3782SJens Axboe { 86503ba3782SJens Axboe unsigned long last_active = jiffies; 86603ba3782SJens Axboe unsigned long wait_jiffies = -1UL; 86703ba3782SJens Axboe long pages_written; 86803ba3782SJens Axboe 86903ba3782SJens Axboe while (!kthread_should_stop()) { 87003ba3782SJens Axboe pages_written = wb_do_writeback(wb, 0); 87103ba3782SJens Axboe 87203ba3782SJens Axboe if (pages_written) 87303ba3782SJens Axboe last_active = jiffies; 87403ba3782SJens Axboe else if (wait_jiffies != -1UL) { 87503ba3782SJens Axboe unsigned long max_idle; 87603ba3782SJens Axboe 87703ba3782SJens Axboe /* 87803ba3782SJens Axboe * Longest period of inactivity that we tolerate. If we 87903ba3782SJens Axboe * see dirty data again later, the task will get 88003ba3782SJens Axboe * recreated automatically. 88103ba3782SJens Axboe */ 88203ba3782SJens Axboe max_idle = max(5UL * 60 * HZ, wait_jiffies); 88303ba3782SJens Axboe if (time_after(jiffies, max_idle + last_active)) 88403ba3782SJens Axboe break; 88503ba3782SJens Axboe } 88603ba3782SJens Axboe 88703ba3782SJens Axboe wait_jiffies = msecs_to_jiffies(dirty_writeback_interval * 10); 88849db0414SJens Axboe schedule_timeout_interruptible(wait_jiffies); 88903ba3782SJens Axboe try_to_freeze(); 89003ba3782SJens Axboe } 89103ba3782SJens Axboe 89203ba3782SJens Axboe return 0; 89303ba3782SJens Axboe } 89403ba3782SJens Axboe 89503ba3782SJens Axboe /* 896b6e51316SJens Axboe * Schedule writeback for all backing devices. This does WB_SYNC_NONE 897b6e51316SJens Axboe * writeback, for integrity writeback see bdi_sync_writeback(). 89803ba3782SJens Axboe */ 899b6e51316SJens Axboe static void bdi_writeback_all(struct super_block *sb, long nr_pages) 90003ba3782SJens Axboe { 901b6e51316SJens Axboe struct wb_writeback_args args = { 902b6e51316SJens Axboe .sb = sb, 903b6e51316SJens Axboe .nr_pages = nr_pages, 904b6e51316SJens Axboe .sync_mode = WB_SYNC_NONE, 905b6e51316SJens Axboe }; 90603ba3782SJens Axboe struct backing_dev_info *bdi; 90703ba3782SJens Axboe 908cfc4ba53SJens Axboe rcu_read_lock(); 90903ba3782SJens Axboe 910cfc4ba53SJens Axboe list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { 91103ba3782SJens Axboe if (!bdi_has_dirty_io(bdi)) 91203ba3782SJens Axboe continue; 91303ba3782SJens Axboe 914b6e51316SJens Axboe bdi_alloc_queue_work(bdi, &args); 91503ba3782SJens Axboe } 91603ba3782SJens Axboe 917cfc4ba53SJens Axboe rcu_read_unlock(); 91803ba3782SJens Axboe } 91903ba3782SJens Axboe 92003ba3782SJens Axboe /* 92103ba3782SJens Axboe * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back 92203ba3782SJens Axboe * the whole world. 92303ba3782SJens Axboe */ 92403ba3782SJens Axboe void wakeup_flusher_threads(long nr_pages) 92503ba3782SJens Axboe { 92603ba3782SJens Axboe if (nr_pages == 0) 92703ba3782SJens Axboe nr_pages = global_page_state(NR_FILE_DIRTY) + 92803ba3782SJens Axboe global_page_state(NR_UNSTABLE_NFS); 929b6e51316SJens Axboe bdi_writeback_all(NULL, nr_pages); 93003ba3782SJens Axboe } 93103ba3782SJens Axboe 93203ba3782SJens Axboe static noinline void block_dump___mark_inode_dirty(struct inode *inode) 93303ba3782SJens Axboe { 93403ba3782SJens Axboe if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { 93503ba3782SJens Axboe struct dentry *dentry; 93603ba3782SJens Axboe const char *name = "?"; 93703ba3782SJens Axboe 93803ba3782SJens Axboe dentry = d_find_alias(inode); 93903ba3782SJens Axboe if (dentry) { 94003ba3782SJens Axboe spin_lock(&dentry->d_lock); 94103ba3782SJens Axboe name = (const char *) dentry->d_name.name; 94203ba3782SJens Axboe } 94303ba3782SJens Axboe printk(KERN_DEBUG 94403ba3782SJens Axboe "%s(%d): dirtied inode %lu (%s) on %s\n", 94503ba3782SJens Axboe current->comm, task_pid_nr(current), inode->i_ino, 94603ba3782SJens Axboe name, inode->i_sb->s_id); 94703ba3782SJens Axboe if (dentry) { 94803ba3782SJens Axboe spin_unlock(&dentry->d_lock); 94903ba3782SJens Axboe dput(dentry); 95003ba3782SJens Axboe } 95103ba3782SJens Axboe } 95203ba3782SJens Axboe } 95303ba3782SJens Axboe 95403ba3782SJens Axboe /** 95503ba3782SJens Axboe * __mark_inode_dirty - internal function 95603ba3782SJens Axboe * @inode: inode to mark 95703ba3782SJens Axboe * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) 95803ba3782SJens Axboe * Mark an inode as dirty. Callers should use mark_inode_dirty or 95903ba3782SJens Axboe * mark_inode_dirty_sync. 96003ba3782SJens Axboe * 96103ba3782SJens Axboe * Put the inode on the super block's dirty list. 96203ba3782SJens Axboe * 96303ba3782SJens Axboe * CAREFUL! We mark it dirty unconditionally, but move it onto the 96403ba3782SJens Axboe * dirty list only if it is hashed or if it refers to a blockdev. 96503ba3782SJens Axboe * If it was not hashed, it will never be added to the dirty list 96603ba3782SJens Axboe * even if it is later hashed, as it will have been marked dirty already. 96703ba3782SJens Axboe * 96803ba3782SJens Axboe * In short, make sure you hash any inodes _before_ you start marking 96903ba3782SJens Axboe * them dirty. 97003ba3782SJens Axboe * 97103ba3782SJens Axboe * This function *must* be atomic for the I_DIRTY_PAGES case - 97203ba3782SJens Axboe * set_page_dirty() is called under spinlock in several places. 97303ba3782SJens Axboe * 97403ba3782SJens Axboe * Note that for blockdevs, inode->dirtied_when represents the dirtying time of 97503ba3782SJens Axboe * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of 97603ba3782SJens Axboe * the kernel-internal blockdev inode represents the dirtying time of the 97703ba3782SJens Axboe * blockdev's pages. This is why for I_DIRTY_PAGES we always use 97803ba3782SJens Axboe * page->mapping->host, so the page-dirtying time is recorded in the internal 97903ba3782SJens Axboe * blockdev inode. 98003ba3782SJens Axboe */ 98103ba3782SJens Axboe void __mark_inode_dirty(struct inode *inode, int flags) 98203ba3782SJens Axboe { 98303ba3782SJens Axboe struct super_block *sb = inode->i_sb; 98403ba3782SJens Axboe 98503ba3782SJens Axboe /* 98603ba3782SJens Axboe * Don't do this for I_DIRTY_PAGES - that doesn't actually 98703ba3782SJens Axboe * dirty the inode itself 98803ba3782SJens Axboe */ 98903ba3782SJens Axboe if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { 99003ba3782SJens Axboe if (sb->s_op->dirty_inode) 99103ba3782SJens Axboe sb->s_op->dirty_inode(inode); 99203ba3782SJens Axboe } 99303ba3782SJens Axboe 99403ba3782SJens Axboe /* 99503ba3782SJens Axboe * make sure that changes are seen by all cpus before we test i_state 99603ba3782SJens Axboe * -- mikulas 99703ba3782SJens Axboe */ 99803ba3782SJens Axboe smp_mb(); 99903ba3782SJens Axboe 100003ba3782SJens Axboe /* avoid the locking if we can */ 100103ba3782SJens Axboe if ((inode->i_state & flags) == flags) 100203ba3782SJens Axboe return; 100303ba3782SJens Axboe 100403ba3782SJens Axboe if (unlikely(block_dump)) 100503ba3782SJens Axboe block_dump___mark_inode_dirty(inode); 100603ba3782SJens Axboe 100703ba3782SJens Axboe spin_lock(&inode_lock); 100803ba3782SJens Axboe if ((inode->i_state & flags) != flags) { 100903ba3782SJens Axboe const int was_dirty = inode->i_state & I_DIRTY; 101003ba3782SJens Axboe 101103ba3782SJens Axboe inode->i_state |= flags; 101203ba3782SJens Axboe 101303ba3782SJens Axboe /* 101403ba3782SJens Axboe * If the inode is being synced, just update its dirty state. 101503ba3782SJens Axboe * The unlocker will place the inode on the appropriate 101603ba3782SJens Axboe * superblock list, based upon its state. 101703ba3782SJens Axboe */ 101803ba3782SJens Axboe if (inode->i_state & I_SYNC) 101903ba3782SJens Axboe goto out; 102003ba3782SJens Axboe 102103ba3782SJens Axboe /* 102203ba3782SJens Axboe * Only add valid (hashed) inodes to the superblock's 102303ba3782SJens Axboe * dirty list. Add blockdev inodes as well. 102403ba3782SJens Axboe */ 102503ba3782SJens Axboe if (!S_ISBLK(inode->i_mode)) { 102603ba3782SJens Axboe if (hlist_unhashed(&inode->i_hash)) 102703ba3782SJens Axboe goto out; 102803ba3782SJens Axboe } 102903ba3782SJens Axboe if (inode->i_state & (I_FREEING|I_CLEAR)) 103003ba3782SJens Axboe goto out; 103103ba3782SJens Axboe 103203ba3782SJens Axboe /* 103303ba3782SJens Axboe * If the inode was already on b_dirty/b_io/b_more_io, don't 103403ba3782SJens Axboe * reposition it (that would break b_dirty time-ordering). 103503ba3782SJens Axboe */ 103603ba3782SJens Axboe if (!was_dirty) { 103703ba3782SJens Axboe struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; 1038500b067cSJens Axboe struct backing_dev_info *bdi = wb->bdi; 1039500b067cSJens Axboe 1040500b067cSJens Axboe if (bdi_cap_writeback_dirty(bdi) && 1041500b067cSJens Axboe !test_bit(BDI_registered, &bdi->state)) { 1042500b067cSJens Axboe WARN_ON(1); 1043500b067cSJens Axboe printk(KERN_ERR "bdi-%s not registered\n", 1044500b067cSJens Axboe bdi->name); 1045500b067cSJens Axboe } 104603ba3782SJens Axboe 104703ba3782SJens Axboe inode->dirtied_when = jiffies; 104803ba3782SJens Axboe list_move(&inode->i_list, &wb->b_dirty); 104903ba3782SJens Axboe } 105003ba3782SJens Axboe } 105103ba3782SJens Axboe out: 105203ba3782SJens Axboe spin_unlock(&inode_lock); 105303ba3782SJens Axboe } 105403ba3782SJens Axboe EXPORT_SYMBOL(__mark_inode_dirty); 105503ba3782SJens Axboe 105666f3b8e2SJens Axboe /* 105766f3b8e2SJens Axboe * Write out a superblock's list of dirty inodes. A wait will be performed 105866f3b8e2SJens Axboe * upon no inodes, all inodes or the final one, depending upon sync_mode. 105966f3b8e2SJens Axboe * 106066f3b8e2SJens Axboe * If older_than_this is non-NULL, then only write out inodes which 106166f3b8e2SJens Axboe * had their first dirtying at a time earlier than *older_than_this. 106266f3b8e2SJens Axboe * 106366f3b8e2SJens Axboe * If we're a pdlfush thread, then implement pdflush collision avoidance 106466f3b8e2SJens Axboe * against the entire list. 106566f3b8e2SJens Axboe * 106666f3b8e2SJens Axboe * If `bdi' is non-zero then we're being asked to writeback a specific queue. 106766f3b8e2SJens Axboe * This function assumes that the blockdev superblock's inodes are backed by 106866f3b8e2SJens Axboe * a variety of queues, so all inodes are searched. For other superblocks, 106966f3b8e2SJens Axboe * assume that all inodes are backed by the same queue. 107066f3b8e2SJens Axboe * 107166f3b8e2SJens Axboe * The inodes to be written are parked on bdi->b_io. They are moved back onto 107266f3b8e2SJens Axboe * bdi->b_dirty as they are selected for writing. This way, none can be missed 107366f3b8e2SJens Axboe * on the writer throttling path, and we get decent balancing between many 107466f3b8e2SJens Axboe * throttled threads: we don't want them all piling up on inode_sync_wait. 107566f3b8e2SJens Axboe */ 1076b6e51316SJens Axboe static void wait_sb_inodes(struct super_block *sb) 107766f3b8e2SJens Axboe { 107838f21977SNick Piggin struct inode *inode, *old_inode = NULL; 107938f21977SNick Piggin 108003ba3782SJens Axboe /* 108103ba3782SJens Axboe * We need to be protected against the filesystem going from 108203ba3782SJens Axboe * r/o to r/w or vice versa. 108303ba3782SJens Axboe */ 1084b6e51316SJens Axboe WARN_ON(!rwsem_is_locked(&sb->s_umount)); 108503ba3782SJens Axboe 108666f3b8e2SJens Axboe spin_lock(&inode_lock); 108766f3b8e2SJens Axboe 108838f21977SNick Piggin /* 108938f21977SNick Piggin * Data integrity sync. Must wait for all pages under writeback, 109038f21977SNick Piggin * because there may have been pages dirtied before our sync 109138f21977SNick Piggin * call, but which had writeout started before we write it out. 109238f21977SNick Piggin * In which case, the inode may not be on the dirty list, but 109338f21977SNick Piggin * we still have to wait for that writeout. 109438f21977SNick Piggin */ 1095b6e51316SJens Axboe list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { 109638f21977SNick Piggin struct address_space *mapping; 109738f21977SNick Piggin 109803ba3782SJens Axboe if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE|I_NEW)) 109938f21977SNick Piggin continue; 110038f21977SNick Piggin mapping = inode->i_mapping; 110138f21977SNick Piggin if (mapping->nrpages == 0) 110238f21977SNick Piggin continue; 110338f21977SNick Piggin __iget(inode); 1104ae8547b0SHans Reiser spin_unlock(&inode_lock); 110538f21977SNick Piggin /* 110638f21977SNick Piggin * We hold a reference to 'inode' so it couldn't have 110738f21977SNick Piggin * been removed from s_inodes list while we dropped the 110838f21977SNick Piggin * inode_lock. We cannot iput the inode now as we can 110938f21977SNick Piggin * be holding the last reference and we cannot iput it 111038f21977SNick Piggin * under inode_lock. So we keep the reference and iput 111138f21977SNick Piggin * it later. 111238f21977SNick Piggin */ 111338f21977SNick Piggin iput(old_inode); 111438f21977SNick Piggin old_inode = inode; 111538f21977SNick Piggin 111638f21977SNick Piggin filemap_fdatawait(mapping); 111738f21977SNick Piggin 111838f21977SNick Piggin cond_resched(); 111938f21977SNick Piggin 112038f21977SNick Piggin spin_lock(&inode_lock); 112138f21977SNick Piggin } 112238f21977SNick Piggin spin_unlock(&inode_lock); 112338f21977SNick Piggin iput(old_inode); 112466f3b8e2SJens Axboe } 11251da177e4SLinus Torvalds 1126d8a8559cSJens Axboe /** 1127d8a8559cSJens Axboe * writeback_inodes_sb - writeback dirty inodes from given super_block 1128d8a8559cSJens Axboe * @sb: the superblock 11291da177e4SLinus Torvalds * 1130d8a8559cSJens Axboe * Start writeback on some inodes on this super_block. No guarantees are made 1131d8a8559cSJens Axboe * on how many (if any) will be written, and this function does not wait 1132d8a8559cSJens Axboe * for IO completion of submitted IO. The number of pages submitted is 1133d8a8559cSJens Axboe * returned. 11341da177e4SLinus Torvalds */ 1135b6e51316SJens Axboe void writeback_inodes_sb(struct super_block *sb) 11361da177e4SLinus Torvalds { 1137b1e7a8fdSChristoph Lameter unsigned long nr_dirty = global_page_state(NR_FILE_DIRTY); 1138fd39fc85SChristoph Lameter unsigned long nr_unstable = global_page_state(NR_UNSTABLE_NFS); 1139d8a8559cSJens Axboe long nr_to_write; 11401da177e4SLinus Torvalds 1141d8a8559cSJens Axboe nr_to_write = nr_dirty + nr_unstable + 114238f21977SNick Piggin (inodes_stat.nr_inodes - inodes_stat.nr_unused); 114338f21977SNick Piggin 1144b6e51316SJens Axboe bdi_writeback_all(sb, nr_to_write); 11451da177e4SLinus Torvalds } 1146d8a8559cSJens Axboe EXPORT_SYMBOL(writeback_inodes_sb); 1147d8a8559cSJens Axboe 1148d8a8559cSJens Axboe /** 1149d8a8559cSJens Axboe * sync_inodes_sb - sync sb inode pages 1150d8a8559cSJens Axboe * @sb: the superblock 1151d8a8559cSJens Axboe * 1152d8a8559cSJens Axboe * This function writes and waits on any dirty inode belonging to this 1153d8a8559cSJens Axboe * super_block. The number of pages synced is returned. 1154d8a8559cSJens Axboe */ 1155b6e51316SJens Axboe void sync_inodes_sb(struct super_block *sb) 1156d8a8559cSJens Axboe { 1157b6e51316SJens Axboe bdi_sync_writeback(sb->s_bdi, sb); 1158b6e51316SJens Axboe wait_sb_inodes(sb); 1159d8a8559cSJens Axboe } 1160d8a8559cSJens Axboe EXPORT_SYMBOL(sync_inodes_sb); 11611da177e4SLinus Torvalds 11621da177e4SLinus Torvalds /** 11631da177e4SLinus Torvalds * write_inode_now - write an inode to disk 11641da177e4SLinus Torvalds * @inode: inode to write to disk 11651da177e4SLinus Torvalds * @sync: whether the write should be synchronous or not 11661da177e4SLinus Torvalds * 11677f04c26dSAndrea Arcangeli * This function commits an inode to disk immediately if it is dirty. This is 11687f04c26dSAndrea Arcangeli * primarily needed by knfsd. 11697f04c26dSAndrea Arcangeli * 11707f04c26dSAndrea Arcangeli * The caller must either have a ref on the inode or must have set I_WILL_FREE. 11711da177e4SLinus Torvalds */ 11721da177e4SLinus Torvalds int write_inode_now(struct inode *inode, int sync) 11731da177e4SLinus Torvalds { 11741da177e4SLinus Torvalds int ret; 11751da177e4SLinus Torvalds struct writeback_control wbc = { 11761da177e4SLinus Torvalds .nr_to_write = LONG_MAX, 117718914b18SMike Galbraith .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, 1178111ebb6eSOGAWA Hirofumi .range_start = 0, 1179111ebb6eSOGAWA Hirofumi .range_end = LLONG_MAX, 11801da177e4SLinus Torvalds }; 11811da177e4SLinus Torvalds 11821da177e4SLinus Torvalds if (!mapping_cap_writeback_dirty(inode->i_mapping)) 118349364ce2SAndrew Morton wbc.nr_to_write = 0; 11841da177e4SLinus Torvalds 11851da177e4SLinus Torvalds might_sleep(); 11861da177e4SLinus Torvalds spin_lock(&inode_lock); 118701c03194SChristoph Hellwig ret = writeback_single_inode(inode, &wbc); 11881da177e4SLinus Torvalds spin_unlock(&inode_lock); 11891da177e4SLinus Torvalds if (sync) 11901c0eeaf5SJoern Engel inode_sync_wait(inode); 11911da177e4SLinus Torvalds return ret; 11921da177e4SLinus Torvalds } 11931da177e4SLinus Torvalds EXPORT_SYMBOL(write_inode_now); 11941da177e4SLinus Torvalds 11951da177e4SLinus Torvalds /** 11961da177e4SLinus Torvalds * sync_inode - write an inode and its pages to disk. 11971da177e4SLinus Torvalds * @inode: the inode to sync 11981da177e4SLinus Torvalds * @wbc: controls the writeback mode 11991da177e4SLinus Torvalds * 12001da177e4SLinus Torvalds * sync_inode() will write an inode and its pages to disk. It will also 12011da177e4SLinus Torvalds * correctly update the inode on its superblock's dirty inode lists and will 12021da177e4SLinus Torvalds * update inode->i_state. 12031da177e4SLinus Torvalds * 12041da177e4SLinus Torvalds * The caller must have a ref on the inode. 12051da177e4SLinus Torvalds */ 12061da177e4SLinus Torvalds int sync_inode(struct inode *inode, struct writeback_control *wbc) 12071da177e4SLinus Torvalds { 12081da177e4SLinus Torvalds int ret; 12091da177e4SLinus Torvalds 12101da177e4SLinus Torvalds spin_lock(&inode_lock); 121101c03194SChristoph Hellwig ret = writeback_single_inode(inode, wbc); 12121da177e4SLinus Torvalds spin_unlock(&inode_lock); 12131da177e4SLinus Torvalds return ret; 12141da177e4SLinus Torvalds } 12151da177e4SLinus Torvalds EXPORT_SYMBOL(sync_inode); 1216