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 { 5203ba3782SJens Axboe struct list_head list; 5303ba3782SJens Axboe struct rcu_head rcu_head; 5403ba3782SJens Axboe 5503ba3782SJens Axboe unsigned long seen; 5603ba3782SJens Axboe atomic_t pending; 5703ba3782SJens Axboe 58c4a77a6cSJens Axboe struct wb_writeback_args args; 5903ba3782SJens Axboe 6003ba3782SJens Axboe unsigned long state; 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, 7703ba3782SJens Axboe struct writeback_control *wbc) 7803ba3782SJens Axboe { 7903ba3782SJens Axboe INIT_RCU_HEAD(&work->rcu_head); 80c4a77a6cSJens Axboe work->args.sb = wbc->sb; 81c4a77a6cSJens Axboe work->args.nr_pages = wbc->nr_to_write; 82c4a77a6cSJens Axboe work->args.sync_mode = wbc->sync_mode; 83c4a77a6cSJens Axboe work->args.range_cyclic = wbc->range_cyclic; 84c4a77a6cSJens Axboe work->args.for_kupdate = 0; 8503ba3782SJens Axboe work->state = WS_USED; 8603ba3782SJens Axboe } 8703ba3782SJens Axboe 88f11b00f3SAdrian Bunk /** 89f11b00f3SAdrian Bunk * writeback_in_progress - determine whether there is writeback in progress 90f11b00f3SAdrian Bunk * @bdi: the device's backing_dev_info structure. 91f11b00f3SAdrian Bunk * 9203ba3782SJens Axboe * Determine whether there is writeback waiting to be handled against a 9303ba3782SJens Axboe * backing device. 94f11b00f3SAdrian Bunk */ 95f11b00f3SAdrian Bunk int writeback_in_progress(struct backing_dev_info *bdi) 96f11b00f3SAdrian Bunk { 9703ba3782SJens Axboe return !list_empty(&bdi->work_list); 98f11b00f3SAdrian Bunk } 99f11b00f3SAdrian Bunk 10003ba3782SJens Axboe static void bdi_work_clear(struct bdi_work *work) 101f11b00f3SAdrian Bunk { 10203ba3782SJens Axboe clear_bit(WS_USED_B, &work->state); 10303ba3782SJens Axboe smp_mb__after_clear_bit(); 10403ba3782SJens Axboe wake_up_bit(&work->state, WS_USED_B); 105f11b00f3SAdrian Bunk } 106f11b00f3SAdrian Bunk 10703ba3782SJens Axboe static void bdi_work_free(struct rcu_head *head) 1084195f73dSNick Piggin { 10903ba3782SJens Axboe struct bdi_work *work = container_of(head, struct bdi_work, rcu_head); 1104195f73dSNick Piggin 11103ba3782SJens Axboe if (!bdi_work_on_stack(work)) 11203ba3782SJens Axboe kfree(work); 11303ba3782SJens Axboe else 11403ba3782SJens Axboe bdi_work_clear(work); 1154195f73dSNick Piggin } 1164195f73dSNick Piggin 11703ba3782SJens Axboe static void wb_work_complete(struct bdi_work *work) 1181da177e4SLinus Torvalds { 119c4a77a6cSJens Axboe const enum writeback_sync_modes sync_mode = work->args.sync_mode; 1201da177e4SLinus Torvalds 1211da177e4SLinus Torvalds /* 12203ba3782SJens Axboe * For allocated work, we can clear the done/seen bit right here. 12303ba3782SJens Axboe * For on-stack work, we need to postpone both the clear and free 12403ba3782SJens Axboe * to after the RCU grace period, since the stack could be invalidated 12503ba3782SJens Axboe * as soon as bdi_work_clear() has done the wakeup. 1261da177e4SLinus Torvalds */ 12703ba3782SJens Axboe if (!bdi_work_on_stack(work)) 12803ba3782SJens Axboe bdi_work_clear(work); 12903ba3782SJens Axboe if (sync_mode == WB_SYNC_NONE || bdi_work_on_stack(work)) 13003ba3782SJens Axboe call_rcu(&work->rcu_head, bdi_work_free); 1311da177e4SLinus Torvalds } 1321da177e4SLinus Torvalds 13303ba3782SJens Axboe static void wb_clear_pending(struct bdi_writeback *wb, struct bdi_work *work) 13403ba3782SJens Axboe { 1351da177e4SLinus Torvalds /* 13603ba3782SJens Axboe * The caller has retrieved the work arguments from this work, 13703ba3782SJens Axboe * drop our reference. If this is the last ref, delete and free it 13803ba3782SJens Axboe */ 13903ba3782SJens Axboe if (atomic_dec_and_test(&work->pending)) { 14003ba3782SJens Axboe struct backing_dev_info *bdi = wb->bdi; 14103ba3782SJens Axboe 14203ba3782SJens Axboe spin_lock(&bdi->wb_lock); 14303ba3782SJens Axboe list_del_rcu(&work->list); 14403ba3782SJens Axboe spin_unlock(&bdi->wb_lock); 14503ba3782SJens Axboe 14603ba3782SJens Axboe wb_work_complete(work); 14703ba3782SJens Axboe } 14803ba3782SJens Axboe } 14903ba3782SJens Axboe 15003ba3782SJens Axboe static void bdi_queue_work(struct backing_dev_info *bdi, struct bdi_work *work) 15103ba3782SJens Axboe { 15203ba3782SJens Axboe work->seen = bdi->wb_mask; 15303ba3782SJens Axboe BUG_ON(!work->seen); 15403ba3782SJens Axboe atomic_set(&work->pending, bdi->wb_cnt); 15503ba3782SJens Axboe BUG_ON(!bdi->wb_cnt); 15603ba3782SJens Axboe 15703ba3782SJens Axboe /* 15803ba3782SJens Axboe * Make sure stores are seen before it appears on the list 1591da177e4SLinus Torvalds */ 1601da177e4SLinus Torvalds smp_mb(); 1611da177e4SLinus Torvalds 16203ba3782SJens Axboe spin_lock(&bdi->wb_lock); 16303ba3782SJens Axboe list_add_tail_rcu(&work->list, &bdi->work_list); 16403ba3782SJens Axboe spin_unlock(&bdi->wb_lock); 1651da177e4SLinus Torvalds 1661da177e4SLinus Torvalds /* 16703ba3782SJens Axboe * If the default thread isn't there, make sure we add it. When 16803ba3782SJens Axboe * it gets created and wakes up, we'll run this work. 1691da177e4SLinus Torvalds */ 17003ba3782SJens Axboe if (unlikely(list_empty_careful(&bdi->wb_list))) 17103ba3782SJens Axboe wake_up_process(default_backing_dev_info.wb.task); 17203ba3782SJens Axboe else { 17303ba3782SJens Axboe struct bdi_writeback *wb = &bdi->wb; 1741da177e4SLinus Torvalds 1751da177e4SLinus Torvalds /* 176bcddc3f0SJens Axboe * End work now if this wb has no dirty IO pending. Otherwise 177bcddc3f0SJens Axboe * wakeup the handling thread 1781da177e4SLinus Torvalds */ 179bcddc3f0SJens Axboe if (!wb_has_dirty_io(wb)) 18003ba3782SJens Axboe wb_clear_pending(wb, work); 181bcddc3f0SJens Axboe else if (wb->task) 18203ba3782SJens Axboe wake_up_process(wb->task); 1831da177e4SLinus Torvalds } 18403ba3782SJens Axboe } 1851da177e4SLinus Torvalds 1861da177e4SLinus Torvalds /* 18703ba3782SJens Axboe * Used for on-stack allocated work items. The caller needs to wait until 18803ba3782SJens Axboe * the wb threads have acked the work before it's safe to continue. 1891da177e4SLinus Torvalds */ 19003ba3782SJens Axboe static void bdi_wait_on_work_clear(struct bdi_work *work) 1911da177e4SLinus Torvalds { 19203ba3782SJens Axboe wait_on_bit(&work->state, WS_USED_B, bdi_sched_wait, 19303ba3782SJens Axboe TASK_UNINTERRUPTIBLE); 19403ba3782SJens Axboe } 19503ba3782SJens Axboe 196f11fcae8SJens Axboe static void bdi_alloc_queue_work(struct backing_dev_info *bdi, 197f11fcae8SJens Axboe struct writeback_control *wbc) 19803ba3782SJens Axboe { 19903ba3782SJens Axboe struct bdi_work *work; 20003ba3782SJens Axboe 201bcddc3f0SJens Axboe /* 202bcddc3f0SJens Axboe * This is WB_SYNC_NONE writeback, so if allocation fails just 203bcddc3f0SJens Axboe * wakeup the thread for old dirty data writeback 204bcddc3f0SJens Axboe */ 20503ba3782SJens Axboe work = kmalloc(sizeof(*work), GFP_ATOMIC); 206bcddc3f0SJens Axboe if (work) { 20703ba3782SJens Axboe bdi_work_init(work, wbc); 208f11fcae8SJens Axboe bdi_queue_work(bdi, work); 209bcddc3f0SJens Axboe } else { 210bcddc3f0SJens Axboe struct bdi_writeback *wb = &bdi->wb; 211bcddc3f0SJens Axboe 212bcddc3f0SJens Axboe if (wb->task) 213bcddc3f0SJens Axboe wake_up_process(wb->task); 214bcddc3f0SJens Axboe } 21503ba3782SJens Axboe } 21603ba3782SJens Axboe 21703ba3782SJens Axboe void bdi_start_writeback(struct writeback_control *wbc) 21803ba3782SJens Axboe { 21903ba3782SJens Axboe /* 220f0fad8a5SChristoph Hellwig * WB_SYNC_NONE is opportunistic writeback. If this allocation fails, 221f0fad8a5SChristoph Hellwig * bdi_queue_work() will wake up the thread and flush old data. This 222f0fad8a5SChristoph Hellwig * should ensure some amount of progress in freeing memory. 22303ba3782SJens Axboe */ 224f11fcae8SJens Axboe if (wbc->sync_mode != WB_SYNC_ALL) 225f11fcae8SJens Axboe bdi_alloc_queue_work(wbc->bdi, wbc); 226f11fcae8SJens Axboe else { 227f0fad8a5SChristoph Hellwig struct bdi_work work; 228f0fad8a5SChristoph Hellwig 229f0fad8a5SChristoph Hellwig bdi_work_init(&work, wbc); 230f0fad8a5SChristoph Hellwig work.state |= WS_ONSTACK; 231f0fad8a5SChristoph Hellwig 232f0fad8a5SChristoph Hellwig bdi_queue_work(wbc->bdi, &work); 233f0fad8a5SChristoph Hellwig bdi_wait_on_work_clear(&work); 23403ba3782SJens Axboe } 2351da177e4SLinus Torvalds } 2361da177e4SLinus Torvalds 2371da177e4SLinus Torvalds /* 2386610a0bcSAndrew Morton * Redirty an inode: set its when-it-was dirtied timestamp and move it to the 2396610a0bcSAndrew Morton * furthest end of its superblock's dirty-inode list. 2406610a0bcSAndrew Morton * 2416610a0bcSAndrew Morton * Before stamping the inode's ->dirtied_when, we check to see whether it is 24266f3b8e2SJens Axboe * already the most-recently-dirtied inode on the b_dirty list. If that is 2436610a0bcSAndrew Morton * the case then the inode must have been redirtied while it was being written 2446610a0bcSAndrew Morton * out and we don't reset its dirtied_when. 2456610a0bcSAndrew Morton */ 2466610a0bcSAndrew Morton static void redirty_tail(struct inode *inode) 2476610a0bcSAndrew Morton { 24803ba3782SJens Axboe struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; 2496610a0bcSAndrew Morton 25003ba3782SJens Axboe if (!list_empty(&wb->b_dirty)) { 25166f3b8e2SJens Axboe struct inode *tail; 2526610a0bcSAndrew Morton 25303ba3782SJens Axboe tail = list_entry(wb->b_dirty.next, struct inode, i_list); 25466f3b8e2SJens Axboe if (time_before(inode->dirtied_when, tail->dirtied_when)) 2556610a0bcSAndrew Morton inode->dirtied_when = jiffies; 2566610a0bcSAndrew Morton } 25703ba3782SJens Axboe list_move(&inode->i_list, &wb->b_dirty); 2586610a0bcSAndrew Morton } 2596610a0bcSAndrew Morton 2606610a0bcSAndrew Morton /* 26166f3b8e2SJens Axboe * requeue inode for re-scanning after bdi->b_io list is exhausted. 262c986d1e2SAndrew Morton */ 2630e0f4fc2SKen Chen static void requeue_io(struct inode *inode) 264c986d1e2SAndrew Morton { 26503ba3782SJens Axboe struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; 26603ba3782SJens Axboe 26703ba3782SJens Axboe list_move(&inode->i_list, &wb->b_more_io); 268c986d1e2SAndrew Morton } 269c986d1e2SAndrew Morton 2701c0eeaf5SJoern Engel static void inode_sync_complete(struct inode *inode) 2711c0eeaf5SJoern Engel { 2721c0eeaf5SJoern Engel /* 2731c0eeaf5SJoern Engel * Prevent speculative execution through spin_unlock(&inode_lock); 2741c0eeaf5SJoern Engel */ 2751c0eeaf5SJoern Engel smp_mb(); 2761c0eeaf5SJoern Engel wake_up_bit(&inode->i_state, __I_SYNC); 2771c0eeaf5SJoern Engel } 2781c0eeaf5SJoern Engel 279d2caa3c5SJeff Layton static bool inode_dirtied_after(struct inode *inode, unsigned long t) 280d2caa3c5SJeff Layton { 281d2caa3c5SJeff Layton bool ret = time_after(inode->dirtied_when, t); 282d2caa3c5SJeff Layton #ifndef CONFIG_64BIT 283d2caa3c5SJeff Layton /* 284d2caa3c5SJeff Layton * For inodes being constantly redirtied, dirtied_when can get stuck. 285d2caa3c5SJeff Layton * It _appears_ to be in the future, but is actually in distant past. 286d2caa3c5SJeff Layton * This test is necessary to prevent such wrapped-around relative times 287d2caa3c5SJeff Layton * from permanently stopping the whole pdflush writeback. 288d2caa3c5SJeff Layton */ 289d2caa3c5SJeff Layton ret = ret && time_before_eq(inode->dirtied_when, jiffies); 290d2caa3c5SJeff Layton #endif 291d2caa3c5SJeff Layton return ret; 292d2caa3c5SJeff Layton } 293d2caa3c5SJeff Layton 294c986d1e2SAndrew Morton /* 2952c136579SFengguang Wu * Move expired dirty inodes from @delaying_queue to @dispatch_queue. 2962c136579SFengguang Wu */ 2972c136579SFengguang Wu static void move_expired_inodes(struct list_head *delaying_queue, 2982c136579SFengguang Wu struct list_head *dispatch_queue, 2992c136579SFengguang Wu unsigned long *older_than_this) 3002c136579SFengguang Wu { 3012c136579SFengguang Wu while (!list_empty(delaying_queue)) { 3022c136579SFengguang Wu struct inode *inode = list_entry(delaying_queue->prev, 3032c136579SFengguang Wu struct inode, i_list); 3042c136579SFengguang Wu if (older_than_this && 305d2caa3c5SJeff Layton inode_dirtied_after(inode, *older_than_this)) 3062c136579SFengguang Wu break; 3072c136579SFengguang Wu list_move(&inode->i_list, dispatch_queue); 3082c136579SFengguang Wu } 3092c136579SFengguang Wu } 3102c136579SFengguang Wu 3112c136579SFengguang Wu /* 3122c136579SFengguang Wu * Queue all expired dirty inodes for io, eldest first. 3132c136579SFengguang Wu */ 31403ba3782SJens Axboe static void queue_io(struct bdi_writeback *wb, unsigned long *older_than_this) 3152c136579SFengguang Wu { 31603ba3782SJens Axboe list_splice_init(&wb->b_more_io, wb->b_io.prev); 31703ba3782SJens Axboe move_expired_inodes(&wb->b_dirty, &wb->b_io, older_than_this); 31866f3b8e2SJens Axboe } 31966f3b8e2SJens Axboe 32003ba3782SJens Axboe static int write_inode(struct inode *inode, int sync) 32166f3b8e2SJens Axboe { 32203ba3782SJens Axboe if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) 32303ba3782SJens Axboe return inode->i_sb->s_op->write_inode(inode, sync); 32403ba3782SJens Axboe return 0; 32566f3b8e2SJens Axboe } 32608d8e974SFengguang Wu 3272c136579SFengguang Wu /* 32801c03194SChristoph Hellwig * Wait for writeback on an inode to complete. 32901c03194SChristoph Hellwig */ 33001c03194SChristoph Hellwig static void inode_wait_for_writeback(struct inode *inode) 33101c03194SChristoph Hellwig { 33201c03194SChristoph Hellwig DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); 33301c03194SChristoph Hellwig wait_queue_head_t *wqh; 33401c03194SChristoph Hellwig 33501c03194SChristoph Hellwig wqh = bit_waitqueue(&inode->i_state, __I_SYNC); 33601c03194SChristoph Hellwig do { 33701c03194SChristoph Hellwig spin_unlock(&inode_lock); 33801c03194SChristoph Hellwig __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE); 33901c03194SChristoph Hellwig spin_lock(&inode_lock); 34001c03194SChristoph Hellwig } while (inode->i_state & I_SYNC); 34101c03194SChristoph Hellwig } 34201c03194SChristoph Hellwig 34301c03194SChristoph Hellwig /* 34401c03194SChristoph Hellwig * Write out an inode's dirty pages. Called under inode_lock. Either the 34501c03194SChristoph Hellwig * caller has ref on the inode (either via __iget or via syscall against an fd) 34601c03194SChristoph Hellwig * or the inode has I_WILL_FREE set (via generic_forget_inode) 34701c03194SChristoph Hellwig * 3481da177e4SLinus Torvalds * If `wait' is set, wait on the writeout. 3491da177e4SLinus Torvalds * 3501da177e4SLinus Torvalds * The whole writeout design is quite complex and fragile. We want to avoid 3511da177e4SLinus Torvalds * starvation of particular inodes when others are being redirtied, prevent 3521da177e4SLinus Torvalds * livelocks, etc. 3531da177e4SLinus Torvalds * 3541da177e4SLinus Torvalds * Called under inode_lock. 3551da177e4SLinus Torvalds */ 3561da177e4SLinus Torvalds static int 35701c03194SChristoph Hellwig writeback_single_inode(struct inode *inode, struct writeback_control *wbc) 3581da177e4SLinus Torvalds { 3591da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 3601da177e4SLinus Torvalds int wait = wbc->sync_mode == WB_SYNC_ALL; 36101c03194SChristoph Hellwig unsigned dirty; 3621da177e4SLinus Torvalds int ret; 3631da177e4SLinus Torvalds 36401c03194SChristoph Hellwig if (!atomic_read(&inode->i_count)) 36501c03194SChristoph Hellwig WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); 36601c03194SChristoph Hellwig else 36701c03194SChristoph Hellwig WARN_ON(inode->i_state & I_WILL_FREE); 36801c03194SChristoph Hellwig 36901c03194SChristoph Hellwig if (inode->i_state & I_SYNC) { 37001c03194SChristoph Hellwig /* 37101c03194SChristoph Hellwig * If this inode is locked for writeback and we are not doing 37266f3b8e2SJens Axboe * writeback-for-data-integrity, move it to b_more_io so that 37301c03194SChristoph Hellwig * writeback can proceed with the other inodes on s_io. 37401c03194SChristoph Hellwig * 37501c03194SChristoph Hellwig * We'll have another go at writing back this inode when we 37666f3b8e2SJens Axboe * completed a full scan of b_io. 37701c03194SChristoph Hellwig */ 37801c03194SChristoph Hellwig if (!wait) { 37901c03194SChristoph Hellwig requeue_io(inode); 38001c03194SChristoph Hellwig return 0; 38101c03194SChristoph Hellwig } 38201c03194SChristoph Hellwig 38301c03194SChristoph Hellwig /* 38401c03194SChristoph Hellwig * It's a data-integrity sync. We must wait. 38501c03194SChristoph Hellwig */ 38601c03194SChristoph Hellwig inode_wait_for_writeback(inode); 38701c03194SChristoph Hellwig } 38801c03194SChristoph Hellwig 3891c0eeaf5SJoern Engel BUG_ON(inode->i_state & I_SYNC); 3901da177e4SLinus Torvalds 3911c0eeaf5SJoern Engel /* Set I_SYNC, reset I_DIRTY */ 3921da177e4SLinus Torvalds dirty = inode->i_state & I_DIRTY; 3931c0eeaf5SJoern Engel inode->i_state |= I_SYNC; 3941da177e4SLinus Torvalds inode->i_state &= ~I_DIRTY; 3951da177e4SLinus Torvalds 3961da177e4SLinus Torvalds spin_unlock(&inode_lock); 3971da177e4SLinus Torvalds 3981da177e4SLinus Torvalds ret = do_writepages(mapping, wbc); 3991da177e4SLinus Torvalds 4001da177e4SLinus Torvalds /* Don't write the inode if only I_DIRTY_PAGES was set */ 4011da177e4SLinus Torvalds if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { 4021da177e4SLinus Torvalds int err = write_inode(inode, wait); 4031da177e4SLinus Torvalds if (ret == 0) 4041da177e4SLinus Torvalds ret = err; 4051da177e4SLinus Torvalds } 4061da177e4SLinus Torvalds 4071da177e4SLinus Torvalds if (wait) { 4081da177e4SLinus Torvalds int err = filemap_fdatawait(mapping); 4091da177e4SLinus Torvalds if (ret == 0) 4101da177e4SLinus Torvalds ret = err; 4111da177e4SLinus Torvalds } 4121da177e4SLinus Torvalds 4131da177e4SLinus Torvalds spin_lock(&inode_lock); 4141c0eeaf5SJoern Engel inode->i_state &= ~I_SYNC; 41584a89245SWu Fengguang if (!(inode->i_state & (I_FREEING | I_CLEAR))) { 4161da177e4SLinus Torvalds if (!(inode->i_state & I_DIRTY) && 4171da177e4SLinus Torvalds mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { 4181da177e4SLinus Torvalds /* 4191da177e4SLinus Torvalds * We didn't write back all the pages. nfs_writepages() 4201da177e4SLinus Torvalds * sometimes bales out without doing anything. Redirty 42166f3b8e2SJens Axboe * the inode; Move it from b_io onto b_more_io/b_dirty. 4221b43ef91SAndrew Morton */ 4231b43ef91SAndrew Morton /* 4241b43ef91SAndrew Morton * akpm: if the caller was the kupdate function we put 42566f3b8e2SJens Axboe * this inode at the head of b_dirty so it gets first 4261b43ef91SAndrew Morton * consideration. Otherwise, move it to the tail, for 4271b43ef91SAndrew Morton * the reasons described there. I'm not really sure 4281b43ef91SAndrew Morton * how much sense this makes. Presumably I had a good 4291b43ef91SAndrew Morton * reasons for doing it this way, and I'd rather not 4301b43ef91SAndrew Morton * muck with it at present. 4311da177e4SLinus Torvalds */ 4321da177e4SLinus Torvalds if (wbc->for_kupdate) { 4331da177e4SLinus Torvalds /* 4342c136579SFengguang Wu * For the kupdate function we move the inode 43566f3b8e2SJens Axboe * to b_more_io so it will get more writeout as 4362c136579SFengguang Wu * soon as the queue becomes uncongested. 4371da177e4SLinus Torvalds */ 4381da177e4SLinus Torvalds inode->i_state |= I_DIRTY_PAGES; 4398bc3be27SFengguang Wu if (wbc->nr_to_write <= 0) { 4408bc3be27SFengguang Wu /* 4418bc3be27SFengguang Wu * slice used up: queue for next turn 4428bc3be27SFengguang Wu */ 4430e0f4fc2SKen Chen requeue_io(inode); 4441da177e4SLinus Torvalds } else { 4451da177e4SLinus Torvalds /* 4468bc3be27SFengguang Wu * somehow blocked: retry later 4478bc3be27SFengguang Wu */ 4488bc3be27SFengguang Wu redirty_tail(inode); 4498bc3be27SFengguang Wu } 4508bc3be27SFengguang Wu } else { 4518bc3be27SFengguang Wu /* 4521da177e4SLinus Torvalds * Otherwise fully redirty the inode so that 4531da177e4SLinus Torvalds * other inodes on this superblock will get some 4541da177e4SLinus Torvalds * writeout. Otherwise heavy writing to one 4551da177e4SLinus Torvalds * file would indefinitely suspend writeout of 4561da177e4SLinus Torvalds * all the other files. 4571da177e4SLinus Torvalds */ 4581da177e4SLinus Torvalds inode->i_state |= I_DIRTY_PAGES; 4591b43ef91SAndrew Morton redirty_tail(inode); 4601da177e4SLinus Torvalds } 4611da177e4SLinus Torvalds } else if (inode->i_state & I_DIRTY) { 4621da177e4SLinus Torvalds /* 4631da177e4SLinus Torvalds * Someone redirtied the inode while were writing back 4641da177e4SLinus Torvalds * the pages. 4651da177e4SLinus Torvalds */ 4666610a0bcSAndrew Morton redirty_tail(inode); 4671da177e4SLinus Torvalds } else if (atomic_read(&inode->i_count)) { 4681da177e4SLinus Torvalds /* 4691da177e4SLinus Torvalds * The inode is clean, inuse 4701da177e4SLinus Torvalds */ 4711da177e4SLinus Torvalds list_move(&inode->i_list, &inode_in_use); 4721da177e4SLinus Torvalds } else { 4731da177e4SLinus Torvalds /* 4741da177e4SLinus Torvalds * The inode is clean, unused 4751da177e4SLinus Torvalds */ 4761da177e4SLinus Torvalds list_move(&inode->i_list, &inode_unused); 4771da177e4SLinus Torvalds } 4781da177e4SLinus Torvalds } 4791c0eeaf5SJoern Engel inode_sync_complete(inode); 4801da177e4SLinus Torvalds return ret; 4811da177e4SLinus Torvalds } 4821da177e4SLinus Torvalds 48303ba3782SJens Axboe /* 48403ba3782SJens Axboe * For WB_SYNC_NONE writeback, the caller does not have the sb pinned 48503ba3782SJens Axboe * before calling writeback. So make sure that we do pin it, so it doesn't 48603ba3782SJens Axboe * go away while we are writing inodes from it. 48703ba3782SJens Axboe * 48803ba3782SJens Axboe * Returns 0 if the super was successfully pinned (or pinning wasn't needed), 48903ba3782SJens Axboe * 1 if we failed. 49003ba3782SJens Axboe */ 49103ba3782SJens Axboe static int pin_sb_for_writeback(struct writeback_control *wbc, 49203ba3782SJens Axboe struct inode *inode) 4931da177e4SLinus Torvalds { 49403ba3782SJens Axboe struct super_block *sb = inode->i_sb; 49503ba3782SJens Axboe 49603ba3782SJens Axboe /* 49703ba3782SJens Axboe * Caller must already hold the ref for this 49803ba3782SJens Axboe */ 49903ba3782SJens Axboe if (wbc->sync_mode == WB_SYNC_ALL) { 50003ba3782SJens Axboe WARN_ON(!rwsem_is_locked(&sb->s_umount)); 50103ba3782SJens Axboe return 0; 50203ba3782SJens Axboe } 50303ba3782SJens Axboe 50403ba3782SJens Axboe spin_lock(&sb_lock); 50503ba3782SJens Axboe sb->s_count++; 50603ba3782SJens Axboe if (down_read_trylock(&sb->s_umount)) { 50703ba3782SJens Axboe if (sb->s_root) { 50803ba3782SJens Axboe spin_unlock(&sb_lock); 50903ba3782SJens Axboe return 0; 51003ba3782SJens Axboe } 51103ba3782SJens Axboe /* 51203ba3782SJens Axboe * umounted, drop rwsem again and fall through to failure 51303ba3782SJens Axboe */ 51403ba3782SJens Axboe up_read(&sb->s_umount); 51503ba3782SJens Axboe } 51603ba3782SJens Axboe 51703ba3782SJens Axboe sb->s_count--; 51803ba3782SJens Axboe spin_unlock(&sb_lock); 51903ba3782SJens Axboe return 1; 52003ba3782SJens Axboe } 52103ba3782SJens Axboe 52203ba3782SJens Axboe static void unpin_sb_for_writeback(struct writeback_control *wbc, 52303ba3782SJens Axboe struct inode *inode) 52403ba3782SJens Axboe { 52503ba3782SJens Axboe struct super_block *sb = inode->i_sb; 52603ba3782SJens Axboe 52703ba3782SJens Axboe if (wbc->sync_mode == WB_SYNC_ALL) 52803ba3782SJens Axboe return; 52903ba3782SJens Axboe 53003ba3782SJens Axboe up_read(&sb->s_umount); 53103ba3782SJens Axboe put_super(sb); 53203ba3782SJens Axboe } 53303ba3782SJens Axboe 53403ba3782SJens Axboe static void writeback_inodes_wb(struct bdi_writeback *wb, 53503ba3782SJens Axboe struct writeback_control *wbc) 53603ba3782SJens Axboe { 53703ba3782SJens Axboe struct super_block *sb = wbc->sb; 53866f3b8e2SJens Axboe const int is_blkdev_sb = sb_is_blkdev_sb(sb); 5391da177e4SLinus Torvalds const unsigned long start = jiffies; /* livelock avoidance */ 5401da177e4SLinus Torvalds 541ae8547b0SHans Reiser spin_lock(&inode_lock); 5421da177e4SLinus Torvalds 54303ba3782SJens Axboe if (!wbc->for_kupdate || list_empty(&wb->b_io)) 54403ba3782SJens Axboe queue_io(wb, wbc->older_than_this); 54566f3b8e2SJens Axboe 54603ba3782SJens Axboe while (!list_empty(&wb->b_io)) { 54703ba3782SJens Axboe struct inode *inode = list_entry(wb->b_io.prev, 5481da177e4SLinus Torvalds struct inode, i_list); 5491da177e4SLinus Torvalds long pages_skipped; 5501da177e4SLinus Torvalds 55166f3b8e2SJens Axboe /* 55266f3b8e2SJens Axboe * super block given and doesn't match, skip this inode 55366f3b8e2SJens Axboe */ 55466f3b8e2SJens Axboe if (sb && sb != inode->i_sb) { 55566f3b8e2SJens Axboe redirty_tail(inode); 55666f3b8e2SJens Axboe continue; 55766f3b8e2SJens Axboe } 55866f3b8e2SJens Axboe 55903ba3782SJens Axboe if (!bdi_cap_writeback_dirty(wb->bdi)) { 5609852a0e7SAndrew Morton redirty_tail(inode); 56166f3b8e2SJens Axboe if (is_blkdev_sb) { 5621da177e4SLinus Torvalds /* 5631da177e4SLinus Torvalds * Dirty memory-backed blockdev: the ramdisk 5641da177e4SLinus Torvalds * driver does this. Skip just this inode 5651da177e4SLinus Torvalds */ 5661da177e4SLinus Torvalds continue; 5671da177e4SLinus Torvalds } 5681da177e4SLinus Torvalds /* 5691da177e4SLinus Torvalds * Dirty memory-backed inode against a filesystem other 5701da177e4SLinus Torvalds * than the kernel-internal bdev filesystem. Skip the 5711da177e4SLinus Torvalds * entire superblock. 5721da177e4SLinus Torvalds */ 5731da177e4SLinus Torvalds break; 5741da177e4SLinus Torvalds } 5751da177e4SLinus Torvalds 57684a89245SWu Fengguang if (inode->i_state & (I_NEW | I_WILL_FREE)) { 5777ef0d737SNick Piggin requeue_io(inode); 5787ef0d737SNick Piggin continue; 5797ef0d737SNick Piggin } 5807ef0d737SNick Piggin 58103ba3782SJens Axboe if (wbc->nonblocking && bdi_write_congested(wb->bdi)) { 5821da177e4SLinus Torvalds wbc->encountered_congestion = 1; 58366f3b8e2SJens Axboe if (!is_blkdev_sb) 5841da177e4SLinus Torvalds break; /* Skip a congested fs */ 5850e0f4fc2SKen Chen requeue_io(inode); 5861da177e4SLinus Torvalds continue; /* Skip a congested blockdev */ 5871da177e4SLinus Torvalds } 5881da177e4SLinus Torvalds 589d2caa3c5SJeff Layton /* 590d2caa3c5SJeff Layton * Was this inode dirtied after sync_sb_inodes was called? 591d2caa3c5SJeff Layton * This keeps sync from extra jobs and livelock. 592d2caa3c5SJeff Layton */ 593d2caa3c5SJeff Layton if (inode_dirtied_after(inode, start)) 5941da177e4SLinus Torvalds break; 5951da177e4SLinus Torvalds 59603ba3782SJens Axboe if (pin_sb_for_writeback(wbc, inode)) { 59703ba3782SJens Axboe requeue_io(inode); 59803ba3782SJens Axboe continue; 59903ba3782SJens Axboe } 6001da177e4SLinus Torvalds 60184a89245SWu Fengguang BUG_ON(inode->i_state & (I_FREEING | I_CLEAR)); 6021da177e4SLinus Torvalds __iget(inode); 6031da177e4SLinus Torvalds pages_skipped = wbc->pages_skipped; 60401c03194SChristoph Hellwig writeback_single_inode(inode, wbc); 60503ba3782SJens Axboe unpin_sb_for_writeback(wbc, inode); 6061da177e4SLinus Torvalds if (wbc->pages_skipped != pages_skipped) { 6071da177e4SLinus Torvalds /* 6081da177e4SLinus Torvalds * writeback is not making progress due to locked 6091da177e4SLinus Torvalds * buffers. Skip this inode for now. 6101da177e4SLinus Torvalds */ 611f57b9b7bSAndrew Morton redirty_tail(inode); 6121da177e4SLinus Torvalds } 6131da177e4SLinus Torvalds spin_unlock(&inode_lock); 6141da177e4SLinus Torvalds iput(inode); 6154ffc8444SOGAWA Hirofumi cond_resched(); 6161da177e4SLinus Torvalds spin_lock(&inode_lock); 6178bc3be27SFengguang Wu if (wbc->nr_to_write <= 0) { 6188bc3be27SFengguang Wu wbc->more_io = 1; 6191da177e4SLinus Torvalds break; 6201da177e4SLinus Torvalds } 62103ba3782SJens Axboe if (!list_empty(&wb->b_more_io)) 6228bc3be27SFengguang Wu wbc->more_io = 1; 6238bc3be27SFengguang Wu } 62438f21977SNick Piggin 62566f3b8e2SJens Axboe spin_unlock(&inode_lock); 62666f3b8e2SJens Axboe /* Leave any unwritten inodes on b_io */ 62766f3b8e2SJens Axboe } 62866f3b8e2SJens Axboe 62903ba3782SJens Axboe void writeback_inodes_wbc(struct writeback_control *wbc) 63003ba3782SJens Axboe { 63103ba3782SJens Axboe struct backing_dev_info *bdi = wbc->bdi; 63203ba3782SJens Axboe 63303ba3782SJens Axboe writeback_inodes_wb(&bdi->wb, wbc); 63403ba3782SJens Axboe } 63503ba3782SJens Axboe 63603ba3782SJens Axboe /* 63703ba3782SJens Axboe * The maximum number of pages to writeout in a single bdi flush/kupdate 63803ba3782SJens Axboe * operation. We do this so we don't hold I_SYNC against an inode for 63903ba3782SJens Axboe * enormous amounts of time, which would block a userspace task which has 64003ba3782SJens Axboe * been forced to throttle against that inode. Also, the code reevaluates 64103ba3782SJens Axboe * the dirty each time it has written this many pages. 64203ba3782SJens Axboe */ 64303ba3782SJens Axboe #define MAX_WRITEBACK_PAGES 1024 64403ba3782SJens Axboe 64503ba3782SJens Axboe static inline bool over_bground_thresh(void) 64603ba3782SJens Axboe { 64703ba3782SJens Axboe unsigned long background_thresh, dirty_thresh; 64803ba3782SJens Axboe 64903ba3782SJens Axboe get_dirty_limits(&background_thresh, &dirty_thresh, NULL, NULL); 65003ba3782SJens Axboe 65103ba3782SJens Axboe return (global_page_state(NR_FILE_DIRTY) + 65203ba3782SJens Axboe global_page_state(NR_UNSTABLE_NFS) >= background_thresh); 65303ba3782SJens Axboe } 65403ba3782SJens Axboe 65503ba3782SJens Axboe /* 65603ba3782SJens Axboe * Explicit flushing or periodic writeback of "old" data. 65703ba3782SJens Axboe * 65803ba3782SJens Axboe * Define "old": the first time one of an inode's pages is dirtied, we mark the 65903ba3782SJens Axboe * dirtying-time in the inode's address_space. So this periodic writeback code 66003ba3782SJens Axboe * just walks the superblock inode list, writing back any inodes which are 66103ba3782SJens Axboe * older than a specific point in time. 66203ba3782SJens Axboe * 66303ba3782SJens Axboe * Try to run once per dirty_writeback_interval. But if a writeback event 66403ba3782SJens Axboe * takes longer than a dirty_writeback_interval interval, then leave a 66503ba3782SJens Axboe * one-second gap. 66603ba3782SJens Axboe * 66703ba3782SJens Axboe * older_than_this takes precedence over nr_to_write. So we'll only write back 66803ba3782SJens Axboe * all dirty pages if they are all attached to "old" mappings. 66903ba3782SJens Axboe */ 670c4a77a6cSJens Axboe static long wb_writeback(struct bdi_writeback *wb, 671c4a77a6cSJens Axboe struct wb_writeback_args *args) 67203ba3782SJens Axboe { 67303ba3782SJens Axboe struct writeback_control wbc = { 67403ba3782SJens Axboe .bdi = wb->bdi, 675c4a77a6cSJens Axboe .sb = args->sb, 676c4a77a6cSJens Axboe .sync_mode = args->sync_mode, 67703ba3782SJens Axboe .older_than_this = NULL, 678c4a77a6cSJens Axboe .for_kupdate = args->for_kupdate, 679c4a77a6cSJens Axboe .range_cyclic = args->range_cyclic, 68003ba3782SJens Axboe }; 68103ba3782SJens Axboe unsigned long oldest_jif; 68203ba3782SJens Axboe long wrote = 0; 68303ba3782SJens Axboe 68403ba3782SJens Axboe if (wbc.for_kupdate) { 68503ba3782SJens Axboe wbc.older_than_this = &oldest_jif; 68603ba3782SJens Axboe oldest_jif = jiffies - 68703ba3782SJens Axboe msecs_to_jiffies(dirty_expire_interval * 10); 68803ba3782SJens Axboe } 689c4a77a6cSJens Axboe if (!wbc.range_cyclic) { 690c4a77a6cSJens Axboe wbc.range_start = 0; 691c4a77a6cSJens Axboe wbc.range_end = LLONG_MAX; 692c4a77a6cSJens Axboe } 69303ba3782SJens Axboe 69403ba3782SJens Axboe for (;;) { 69503ba3782SJens Axboe /* 69603ba3782SJens Axboe * Don't flush anything for non-integrity writeback where 69703ba3782SJens Axboe * no nr_pages was given 69803ba3782SJens Axboe */ 699c4a77a6cSJens Axboe if (!args->for_kupdate && args->nr_pages <= 0 && 700c4a77a6cSJens Axboe args->sync_mode == WB_SYNC_NONE) 70103ba3782SJens Axboe break; 70203ba3782SJens Axboe 70303ba3782SJens Axboe /* 70403ba3782SJens Axboe * If no specific pages were given and this is just a 70503ba3782SJens Axboe * periodic background writeout and we are below the 70603ba3782SJens Axboe * background dirty threshold, don't do anything 70703ba3782SJens Axboe */ 708c4a77a6cSJens Axboe if (args->for_kupdate && args->nr_pages <= 0 && 709c4a77a6cSJens Axboe !over_bground_thresh()) 71003ba3782SJens Axboe break; 71103ba3782SJens Axboe 71203ba3782SJens Axboe wbc.more_io = 0; 71303ba3782SJens Axboe wbc.encountered_congestion = 0; 71403ba3782SJens Axboe wbc.nr_to_write = MAX_WRITEBACK_PAGES; 71503ba3782SJens Axboe wbc.pages_skipped = 0; 71603ba3782SJens Axboe writeback_inodes_wb(wb, &wbc); 717c4a77a6cSJens Axboe args->nr_pages -= MAX_WRITEBACK_PAGES - wbc.nr_to_write; 71803ba3782SJens Axboe wrote += MAX_WRITEBACK_PAGES - wbc.nr_to_write; 71903ba3782SJens Axboe 72003ba3782SJens Axboe /* 72103ba3782SJens Axboe * If we ran out of stuff to write, bail unless more_io got set 72203ba3782SJens Axboe */ 72303ba3782SJens Axboe if (wbc.nr_to_write > 0 || wbc.pages_skipped > 0) { 72403ba3782SJens Axboe if (wbc.more_io && !wbc.for_kupdate) 72503ba3782SJens Axboe continue; 72603ba3782SJens Axboe break; 72703ba3782SJens Axboe } 72803ba3782SJens Axboe } 72903ba3782SJens Axboe 73003ba3782SJens Axboe return wrote; 73103ba3782SJens Axboe } 73203ba3782SJens Axboe 73303ba3782SJens Axboe /* 73403ba3782SJens Axboe * Return the next bdi_work struct that hasn't been processed by this 73503ba3782SJens Axboe * wb thread yet 73603ba3782SJens Axboe */ 73703ba3782SJens Axboe static struct bdi_work *get_next_work_item(struct backing_dev_info *bdi, 73803ba3782SJens Axboe struct bdi_writeback *wb) 73903ba3782SJens Axboe { 74003ba3782SJens Axboe struct bdi_work *work, *ret = NULL; 74103ba3782SJens Axboe 74203ba3782SJens Axboe rcu_read_lock(); 74303ba3782SJens Axboe 74403ba3782SJens Axboe list_for_each_entry_rcu(work, &bdi->work_list, list) { 74503ba3782SJens Axboe if (!test_and_clear_bit(wb->nr, &work->seen)) 74603ba3782SJens Axboe continue; 74703ba3782SJens Axboe 74803ba3782SJens Axboe ret = work; 74903ba3782SJens Axboe break; 75003ba3782SJens Axboe } 75103ba3782SJens Axboe 75203ba3782SJens Axboe rcu_read_unlock(); 75303ba3782SJens Axboe return ret; 75403ba3782SJens Axboe } 75503ba3782SJens Axboe 75603ba3782SJens Axboe static long wb_check_old_data_flush(struct bdi_writeback *wb) 75703ba3782SJens Axboe { 75803ba3782SJens Axboe unsigned long expired; 75903ba3782SJens Axboe long nr_pages; 76003ba3782SJens Axboe 76103ba3782SJens Axboe expired = wb->last_old_flush + 76203ba3782SJens Axboe msecs_to_jiffies(dirty_writeback_interval * 10); 76303ba3782SJens Axboe if (time_before(jiffies, expired)) 76403ba3782SJens Axboe return 0; 76503ba3782SJens Axboe 76603ba3782SJens Axboe wb->last_old_flush = jiffies; 76703ba3782SJens Axboe nr_pages = global_page_state(NR_FILE_DIRTY) + 76803ba3782SJens Axboe global_page_state(NR_UNSTABLE_NFS) + 76903ba3782SJens Axboe (inodes_stat.nr_inodes - inodes_stat.nr_unused); 77003ba3782SJens Axboe 771c4a77a6cSJens Axboe if (nr_pages) { 772c4a77a6cSJens Axboe struct wb_writeback_args args = { 773c4a77a6cSJens Axboe .nr_pages = nr_pages, 774c4a77a6cSJens Axboe .sync_mode = WB_SYNC_NONE, 775c4a77a6cSJens Axboe .for_kupdate = 1, 776c4a77a6cSJens Axboe .range_cyclic = 1, 777c4a77a6cSJens Axboe }; 778c4a77a6cSJens Axboe 779c4a77a6cSJens Axboe return wb_writeback(wb, &args); 780c4a77a6cSJens Axboe } 78103ba3782SJens Axboe 78203ba3782SJens Axboe return 0; 78303ba3782SJens Axboe } 78403ba3782SJens Axboe 78503ba3782SJens Axboe /* 78603ba3782SJens Axboe * Retrieve work items and do the writeback they describe 78703ba3782SJens Axboe */ 78803ba3782SJens Axboe long wb_do_writeback(struct bdi_writeback *wb, int force_wait) 78903ba3782SJens Axboe { 79003ba3782SJens Axboe struct backing_dev_info *bdi = wb->bdi; 79103ba3782SJens Axboe struct bdi_work *work; 792c4a77a6cSJens Axboe long wrote = 0; 79303ba3782SJens Axboe 79403ba3782SJens Axboe while ((work = get_next_work_item(bdi, wb)) != NULL) { 795c4a77a6cSJens Axboe struct wb_writeback_args args = work->args; 79603ba3782SJens Axboe 79703ba3782SJens Axboe /* 79803ba3782SJens Axboe * Override sync mode, in case we must wait for completion 79903ba3782SJens Axboe */ 80003ba3782SJens Axboe if (force_wait) 801c4a77a6cSJens Axboe work->args.sync_mode = args.sync_mode = WB_SYNC_ALL; 80203ba3782SJens Axboe 80303ba3782SJens Axboe /* 80403ba3782SJens Axboe * If this isn't a data integrity operation, just notify 80503ba3782SJens Axboe * that we have seen this work and we are now starting it. 80603ba3782SJens Axboe */ 807c4a77a6cSJens Axboe if (args.sync_mode == WB_SYNC_NONE) 80803ba3782SJens Axboe wb_clear_pending(wb, work); 80903ba3782SJens Axboe 810c4a77a6cSJens Axboe wrote += wb_writeback(wb, &args); 81103ba3782SJens Axboe 81203ba3782SJens Axboe /* 81303ba3782SJens Axboe * This is a data integrity writeback, so only do the 81403ba3782SJens Axboe * notification when we have completed the work. 81503ba3782SJens Axboe */ 816c4a77a6cSJens Axboe if (args.sync_mode == WB_SYNC_ALL) 81703ba3782SJens Axboe wb_clear_pending(wb, work); 81803ba3782SJens Axboe } 81903ba3782SJens Axboe 82003ba3782SJens Axboe /* 82103ba3782SJens Axboe * Check for periodic writeback, kupdated() style 82203ba3782SJens Axboe */ 82303ba3782SJens Axboe wrote += wb_check_old_data_flush(wb); 82403ba3782SJens Axboe 82503ba3782SJens Axboe return wrote; 82603ba3782SJens Axboe } 82703ba3782SJens Axboe 82803ba3782SJens Axboe /* 82903ba3782SJens Axboe * Handle writeback of dirty data for the device backed by this bdi. Also 83003ba3782SJens Axboe * wakes up periodically and does kupdated style flushing. 83103ba3782SJens Axboe */ 83203ba3782SJens Axboe int bdi_writeback_task(struct bdi_writeback *wb) 83303ba3782SJens Axboe { 83403ba3782SJens Axboe unsigned long last_active = jiffies; 83503ba3782SJens Axboe unsigned long wait_jiffies = -1UL; 83603ba3782SJens Axboe long pages_written; 83703ba3782SJens Axboe 83803ba3782SJens Axboe while (!kthread_should_stop()) { 83903ba3782SJens Axboe pages_written = wb_do_writeback(wb, 0); 84003ba3782SJens Axboe 84103ba3782SJens Axboe if (pages_written) 84203ba3782SJens Axboe last_active = jiffies; 84303ba3782SJens Axboe else if (wait_jiffies != -1UL) { 84403ba3782SJens Axboe unsigned long max_idle; 84503ba3782SJens Axboe 84603ba3782SJens Axboe /* 84703ba3782SJens Axboe * Longest period of inactivity that we tolerate. If we 84803ba3782SJens Axboe * see dirty data again later, the task will get 84903ba3782SJens Axboe * recreated automatically. 85003ba3782SJens Axboe */ 85103ba3782SJens Axboe max_idle = max(5UL * 60 * HZ, wait_jiffies); 85203ba3782SJens Axboe if (time_after(jiffies, max_idle + last_active)) 85303ba3782SJens Axboe break; 85403ba3782SJens Axboe } 85503ba3782SJens Axboe 85603ba3782SJens Axboe wait_jiffies = msecs_to_jiffies(dirty_writeback_interval * 10); 85703ba3782SJens Axboe set_current_state(TASK_INTERRUPTIBLE); 85803ba3782SJens Axboe schedule_timeout(wait_jiffies); 85903ba3782SJens Axboe try_to_freeze(); 86003ba3782SJens Axboe } 86103ba3782SJens Axboe 86203ba3782SJens Axboe return 0; 86303ba3782SJens Axboe } 86403ba3782SJens Axboe 86503ba3782SJens Axboe /* 866f11fcae8SJens Axboe * Schedule writeback for all backing devices. Can only be used for 867f11fcae8SJens Axboe * WB_SYNC_NONE writeback, WB_SYNC_ALL should use bdi_start_writeback() 868f11fcae8SJens Axboe * and pass in the superblock. 86903ba3782SJens Axboe */ 87003ba3782SJens Axboe static void bdi_writeback_all(struct writeback_control *wbc) 87103ba3782SJens Axboe { 87203ba3782SJens Axboe struct backing_dev_info *bdi; 87303ba3782SJens Axboe 874f11fcae8SJens Axboe WARN_ON(wbc->sync_mode == WB_SYNC_ALL); 875f11fcae8SJens Axboe 876cfc4ba53SJens Axboe rcu_read_lock(); 87703ba3782SJens Axboe 878cfc4ba53SJens Axboe list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { 87903ba3782SJens Axboe if (!bdi_has_dirty_io(bdi)) 88003ba3782SJens Axboe continue; 88103ba3782SJens Axboe 882f11fcae8SJens Axboe bdi_alloc_queue_work(bdi, wbc); 88303ba3782SJens Axboe } 88403ba3782SJens Axboe 885cfc4ba53SJens Axboe rcu_read_unlock(); 88603ba3782SJens Axboe } 88703ba3782SJens Axboe 88803ba3782SJens Axboe /* 88903ba3782SJens Axboe * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back 89003ba3782SJens Axboe * the whole world. 89103ba3782SJens Axboe */ 89203ba3782SJens Axboe void wakeup_flusher_threads(long nr_pages) 89303ba3782SJens Axboe { 89403ba3782SJens Axboe struct writeback_control wbc = { 89503ba3782SJens Axboe .sync_mode = WB_SYNC_NONE, 89603ba3782SJens Axboe .older_than_this = NULL, 89703ba3782SJens Axboe .range_cyclic = 1, 89803ba3782SJens Axboe }; 89903ba3782SJens Axboe 90003ba3782SJens Axboe if (nr_pages == 0) 90103ba3782SJens Axboe nr_pages = global_page_state(NR_FILE_DIRTY) + 90203ba3782SJens Axboe global_page_state(NR_UNSTABLE_NFS); 90303ba3782SJens Axboe wbc.nr_to_write = nr_pages; 90403ba3782SJens Axboe bdi_writeback_all(&wbc); 90503ba3782SJens Axboe } 90603ba3782SJens Axboe 90703ba3782SJens Axboe static noinline void block_dump___mark_inode_dirty(struct inode *inode) 90803ba3782SJens Axboe { 90903ba3782SJens Axboe if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { 91003ba3782SJens Axboe struct dentry *dentry; 91103ba3782SJens Axboe const char *name = "?"; 91203ba3782SJens Axboe 91303ba3782SJens Axboe dentry = d_find_alias(inode); 91403ba3782SJens Axboe if (dentry) { 91503ba3782SJens Axboe spin_lock(&dentry->d_lock); 91603ba3782SJens Axboe name = (const char *) dentry->d_name.name; 91703ba3782SJens Axboe } 91803ba3782SJens Axboe printk(KERN_DEBUG 91903ba3782SJens Axboe "%s(%d): dirtied inode %lu (%s) on %s\n", 92003ba3782SJens Axboe current->comm, task_pid_nr(current), inode->i_ino, 92103ba3782SJens Axboe name, inode->i_sb->s_id); 92203ba3782SJens Axboe if (dentry) { 92303ba3782SJens Axboe spin_unlock(&dentry->d_lock); 92403ba3782SJens Axboe dput(dentry); 92503ba3782SJens Axboe } 92603ba3782SJens Axboe } 92703ba3782SJens Axboe } 92803ba3782SJens Axboe 92903ba3782SJens Axboe /** 93003ba3782SJens Axboe * __mark_inode_dirty - internal function 93103ba3782SJens Axboe * @inode: inode to mark 93203ba3782SJens Axboe * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) 93303ba3782SJens Axboe * Mark an inode as dirty. Callers should use mark_inode_dirty or 93403ba3782SJens Axboe * mark_inode_dirty_sync. 93503ba3782SJens Axboe * 93603ba3782SJens Axboe * Put the inode on the super block's dirty list. 93703ba3782SJens Axboe * 93803ba3782SJens Axboe * CAREFUL! We mark it dirty unconditionally, but move it onto the 93903ba3782SJens Axboe * dirty list only if it is hashed or if it refers to a blockdev. 94003ba3782SJens Axboe * If it was not hashed, it will never be added to the dirty list 94103ba3782SJens Axboe * even if it is later hashed, as it will have been marked dirty already. 94203ba3782SJens Axboe * 94303ba3782SJens Axboe * In short, make sure you hash any inodes _before_ you start marking 94403ba3782SJens Axboe * them dirty. 94503ba3782SJens Axboe * 94603ba3782SJens Axboe * This function *must* be atomic for the I_DIRTY_PAGES case - 94703ba3782SJens Axboe * set_page_dirty() is called under spinlock in several places. 94803ba3782SJens Axboe * 94903ba3782SJens Axboe * Note that for blockdevs, inode->dirtied_when represents the dirtying time of 95003ba3782SJens Axboe * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of 95103ba3782SJens Axboe * the kernel-internal blockdev inode represents the dirtying time of the 95203ba3782SJens Axboe * blockdev's pages. This is why for I_DIRTY_PAGES we always use 95303ba3782SJens Axboe * page->mapping->host, so the page-dirtying time is recorded in the internal 95403ba3782SJens Axboe * blockdev inode. 95503ba3782SJens Axboe */ 95603ba3782SJens Axboe void __mark_inode_dirty(struct inode *inode, int flags) 95703ba3782SJens Axboe { 95803ba3782SJens Axboe struct super_block *sb = inode->i_sb; 95903ba3782SJens Axboe 96003ba3782SJens Axboe /* 96103ba3782SJens Axboe * Don't do this for I_DIRTY_PAGES - that doesn't actually 96203ba3782SJens Axboe * dirty the inode itself 96303ba3782SJens Axboe */ 96403ba3782SJens Axboe if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { 96503ba3782SJens Axboe if (sb->s_op->dirty_inode) 96603ba3782SJens Axboe sb->s_op->dirty_inode(inode); 96703ba3782SJens Axboe } 96803ba3782SJens Axboe 96903ba3782SJens Axboe /* 97003ba3782SJens Axboe * make sure that changes are seen by all cpus before we test i_state 97103ba3782SJens Axboe * -- mikulas 97203ba3782SJens Axboe */ 97303ba3782SJens Axboe smp_mb(); 97403ba3782SJens Axboe 97503ba3782SJens Axboe /* avoid the locking if we can */ 97603ba3782SJens Axboe if ((inode->i_state & flags) == flags) 97703ba3782SJens Axboe return; 97803ba3782SJens Axboe 97903ba3782SJens Axboe if (unlikely(block_dump)) 98003ba3782SJens Axboe block_dump___mark_inode_dirty(inode); 98103ba3782SJens Axboe 98203ba3782SJens Axboe spin_lock(&inode_lock); 98303ba3782SJens Axboe if ((inode->i_state & flags) != flags) { 98403ba3782SJens Axboe const int was_dirty = inode->i_state & I_DIRTY; 98503ba3782SJens Axboe 98603ba3782SJens Axboe inode->i_state |= flags; 98703ba3782SJens Axboe 98803ba3782SJens Axboe /* 98903ba3782SJens Axboe * If the inode is being synced, just update its dirty state. 99003ba3782SJens Axboe * The unlocker will place the inode on the appropriate 99103ba3782SJens Axboe * superblock list, based upon its state. 99203ba3782SJens Axboe */ 99303ba3782SJens Axboe if (inode->i_state & I_SYNC) 99403ba3782SJens Axboe goto out; 99503ba3782SJens Axboe 99603ba3782SJens Axboe /* 99703ba3782SJens Axboe * Only add valid (hashed) inodes to the superblock's 99803ba3782SJens Axboe * dirty list. Add blockdev inodes as well. 99903ba3782SJens Axboe */ 100003ba3782SJens Axboe if (!S_ISBLK(inode->i_mode)) { 100103ba3782SJens Axboe if (hlist_unhashed(&inode->i_hash)) 100203ba3782SJens Axboe goto out; 100303ba3782SJens Axboe } 100403ba3782SJens Axboe if (inode->i_state & (I_FREEING|I_CLEAR)) 100503ba3782SJens Axboe goto out; 100603ba3782SJens Axboe 100703ba3782SJens Axboe /* 100803ba3782SJens Axboe * If the inode was already on b_dirty/b_io/b_more_io, don't 100903ba3782SJens Axboe * reposition it (that would break b_dirty time-ordering). 101003ba3782SJens Axboe */ 101103ba3782SJens Axboe if (!was_dirty) { 101203ba3782SJens Axboe struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; 1013500b067cSJens Axboe struct backing_dev_info *bdi = wb->bdi; 1014500b067cSJens Axboe 1015500b067cSJens Axboe if (bdi_cap_writeback_dirty(bdi) && 1016500b067cSJens Axboe !test_bit(BDI_registered, &bdi->state)) { 1017500b067cSJens Axboe WARN_ON(1); 1018500b067cSJens Axboe printk(KERN_ERR "bdi-%s not registered\n", 1019500b067cSJens Axboe bdi->name); 1020500b067cSJens Axboe } 102103ba3782SJens Axboe 102203ba3782SJens Axboe inode->dirtied_when = jiffies; 102303ba3782SJens Axboe list_move(&inode->i_list, &wb->b_dirty); 102403ba3782SJens Axboe } 102503ba3782SJens Axboe } 102603ba3782SJens Axboe out: 102703ba3782SJens Axboe spin_unlock(&inode_lock); 102803ba3782SJens Axboe } 102903ba3782SJens Axboe EXPORT_SYMBOL(__mark_inode_dirty); 103003ba3782SJens Axboe 103166f3b8e2SJens Axboe /* 103266f3b8e2SJens Axboe * Write out a superblock's list of dirty inodes. A wait will be performed 103366f3b8e2SJens Axboe * upon no inodes, all inodes or the final one, depending upon sync_mode. 103466f3b8e2SJens Axboe * 103566f3b8e2SJens Axboe * If older_than_this is non-NULL, then only write out inodes which 103666f3b8e2SJens Axboe * had their first dirtying at a time earlier than *older_than_this. 103766f3b8e2SJens Axboe * 103866f3b8e2SJens Axboe * If we're a pdlfush thread, then implement pdflush collision avoidance 103966f3b8e2SJens Axboe * against the entire list. 104066f3b8e2SJens Axboe * 104166f3b8e2SJens Axboe * If `bdi' is non-zero then we're being asked to writeback a specific queue. 104266f3b8e2SJens Axboe * This function assumes that the blockdev superblock's inodes are backed by 104366f3b8e2SJens Axboe * a variety of queues, so all inodes are searched. For other superblocks, 104466f3b8e2SJens Axboe * assume that all inodes are backed by the same queue. 104566f3b8e2SJens Axboe * 104666f3b8e2SJens Axboe * The inodes to be written are parked on bdi->b_io. They are moved back onto 104766f3b8e2SJens Axboe * bdi->b_dirty as they are selected for writing. This way, none can be missed 104866f3b8e2SJens Axboe * on the writer throttling path, and we get decent balancing between many 104966f3b8e2SJens Axboe * throttled threads: we don't want them all piling up on inode_sync_wait. 105066f3b8e2SJens Axboe */ 105103ba3782SJens Axboe static void wait_sb_inodes(struct writeback_control *wbc) 105266f3b8e2SJens Axboe { 105338f21977SNick Piggin struct inode *inode, *old_inode = NULL; 105438f21977SNick Piggin 105503ba3782SJens Axboe /* 105603ba3782SJens Axboe * We need to be protected against the filesystem going from 105703ba3782SJens Axboe * r/o to r/w or vice versa. 105803ba3782SJens Axboe */ 105903ba3782SJens Axboe WARN_ON(!rwsem_is_locked(&wbc->sb->s_umount)); 106003ba3782SJens Axboe 106166f3b8e2SJens Axboe spin_lock(&inode_lock); 106266f3b8e2SJens Axboe 106338f21977SNick Piggin /* 106438f21977SNick Piggin * Data integrity sync. Must wait for all pages under writeback, 106538f21977SNick Piggin * because there may have been pages dirtied before our sync 106638f21977SNick Piggin * call, but which had writeout started before we write it out. 106738f21977SNick Piggin * In which case, the inode may not be on the dirty list, but 106838f21977SNick Piggin * we still have to wait for that writeout. 106938f21977SNick Piggin */ 107003ba3782SJens Axboe list_for_each_entry(inode, &wbc->sb->s_inodes, i_sb_list) { 107138f21977SNick Piggin struct address_space *mapping; 107238f21977SNick Piggin 107303ba3782SJens Axboe if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE|I_NEW)) 107438f21977SNick Piggin continue; 107538f21977SNick Piggin mapping = inode->i_mapping; 107638f21977SNick Piggin if (mapping->nrpages == 0) 107738f21977SNick Piggin continue; 107838f21977SNick Piggin __iget(inode); 1079ae8547b0SHans Reiser spin_unlock(&inode_lock); 108038f21977SNick Piggin /* 108138f21977SNick Piggin * We hold a reference to 'inode' so it couldn't have 108238f21977SNick Piggin * been removed from s_inodes list while we dropped the 108338f21977SNick Piggin * inode_lock. We cannot iput the inode now as we can 108438f21977SNick Piggin * be holding the last reference and we cannot iput it 108538f21977SNick Piggin * under inode_lock. So we keep the reference and iput 108638f21977SNick Piggin * it later. 108738f21977SNick Piggin */ 108838f21977SNick Piggin iput(old_inode); 108938f21977SNick Piggin old_inode = inode; 109038f21977SNick Piggin 109138f21977SNick Piggin filemap_fdatawait(mapping); 109238f21977SNick Piggin 109338f21977SNick Piggin cond_resched(); 109438f21977SNick Piggin 109538f21977SNick Piggin spin_lock(&inode_lock); 109638f21977SNick Piggin } 109738f21977SNick Piggin spin_unlock(&inode_lock); 109838f21977SNick Piggin iput(old_inode); 109966f3b8e2SJens Axboe } 11001da177e4SLinus Torvalds 1101d8a8559cSJens Axboe /** 1102d8a8559cSJens Axboe * writeback_inodes_sb - writeback dirty inodes from given super_block 1103d8a8559cSJens Axboe * @sb: the superblock 11041da177e4SLinus Torvalds * 1105d8a8559cSJens Axboe * Start writeback on some inodes on this super_block. No guarantees are made 1106d8a8559cSJens Axboe * on how many (if any) will be written, and this function does not wait 1107d8a8559cSJens Axboe * for IO completion of submitted IO. The number of pages submitted is 1108d8a8559cSJens Axboe * returned. 11091da177e4SLinus Torvalds */ 1110d8a8559cSJens Axboe long writeback_inodes_sb(struct super_block *sb) 11111da177e4SLinus Torvalds { 11121da177e4SLinus Torvalds struct writeback_control wbc = { 111303ba3782SJens Axboe .sb = sb, 1114d8a8559cSJens Axboe .sync_mode = WB_SYNC_NONE, 1115111ebb6eSOGAWA Hirofumi .range_start = 0, 1116111ebb6eSOGAWA Hirofumi .range_end = LLONG_MAX, 11171da177e4SLinus Torvalds }; 1118b1e7a8fdSChristoph Lameter unsigned long nr_dirty = global_page_state(NR_FILE_DIRTY); 1119fd39fc85SChristoph Lameter unsigned long nr_unstable = global_page_state(NR_UNSTABLE_NFS); 1120d8a8559cSJens Axboe long nr_to_write; 11211da177e4SLinus Torvalds 1122d8a8559cSJens Axboe nr_to_write = nr_dirty + nr_unstable + 112338f21977SNick Piggin (inodes_stat.nr_inodes - inodes_stat.nr_unused); 112438f21977SNick Piggin 1125d8a8559cSJens Axboe wbc.nr_to_write = nr_to_write; 112603ba3782SJens Axboe bdi_writeback_all(&wbc); 1127d8a8559cSJens Axboe return nr_to_write - wbc.nr_to_write; 11281da177e4SLinus Torvalds } 1129d8a8559cSJens Axboe EXPORT_SYMBOL(writeback_inodes_sb); 1130d8a8559cSJens Axboe 1131d8a8559cSJens Axboe /** 1132d8a8559cSJens Axboe * sync_inodes_sb - sync sb inode pages 1133d8a8559cSJens Axboe * @sb: the superblock 1134d8a8559cSJens Axboe * 1135d8a8559cSJens Axboe * This function writes and waits on any dirty inode belonging to this 1136d8a8559cSJens Axboe * super_block. The number of pages synced is returned. 1137d8a8559cSJens Axboe */ 1138d8a8559cSJens Axboe long sync_inodes_sb(struct super_block *sb) 1139d8a8559cSJens Axboe { 1140d8a8559cSJens Axboe struct writeback_control wbc = { 114103ba3782SJens Axboe .sb = sb, 1142f11fcae8SJens Axboe .bdi = sb->s_bdi, 1143d8a8559cSJens Axboe .sync_mode = WB_SYNC_ALL, 1144d8a8559cSJens Axboe .range_start = 0, 1145d8a8559cSJens Axboe .range_end = LLONG_MAX, 1146d8a8559cSJens Axboe }; 1147d8a8559cSJens Axboe long nr_to_write = LONG_MAX; /* doesn't actually matter */ 1148d8a8559cSJens Axboe 1149d8a8559cSJens Axboe wbc.nr_to_write = nr_to_write; 1150f11fcae8SJens Axboe bdi_start_writeback(&wbc); 115103ba3782SJens Axboe wait_sb_inodes(&wbc); 1152d8a8559cSJens Axboe return nr_to_write - wbc.nr_to_write; 1153d8a8559cSJens Axboe } 1154d8a8559cSJens Axboe EXPORT_SYMBOL(sync_inodes_sb); 11551da177e4SLinus Torvalds 11561da177e4SLinus Torvalds /** 11571da177e4SLinus Torvalds * write_inode_now - write an inode to disk 11581da177e4SLinus Torvalds * @inode: inode to write to disk 11591da177e4SLinus Torvalds * @sync: whether the write should be synchronous or not 11601da177e4SLinus Torvalds * 11617f04c26dSAndrea Arcangeli * This function commits an inode to disk immediately if it is dirty. This is 11627f04c26dSAndrea Arcangeli * primarily needed by knfsd. 11637f04c26dSAndrea Arcangeli * 11647f04c26dSAndrea Arcangeli * The caller must either have a ref on the inode or must have set I_WILL_FREE. 11651da177e4SLinus Torvalds */ 11661da177e4SLinus Torvalds int write_inode_now(struct inode *inode, int sync) 11671da177e4SLinus Torvalds { 11681da177e4SLinus Torvalds int ret; 11691da177e4SLinus Torvalds struct writeback_control wbc = { 11701da177e4SLinus Torvalds .nr_to_write = LONG_MAX, 117118914b18SMike Galbraith .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, 1172111ebb6eSOGAWA Hirofumi .range_start = 0, 1173111ebb6eSOGAWA Hirofumi .range_end = LLONG_MAX, 11741da177e4SLinus Torvalds }; 11751da177e4SLinus Torvalds 11761da177e4SLinus Torvalds if (!mapping_cap_writeback_dirty(inode->i_mapping)) 117749364ce2SAndrew Morton wbc.nr_to_write = 0; 11781da177e4SLinus Torvalds 11791da177e4SLinus Torvalds might_sleep(); 11801da177e4SLinus Torvalds spin_lock(&inode_lock); 118101c03194SChristoph Hellwig ret = writeback_single_inode(inode, &wbc); 11821da177e4SLinus Torvalds spin_unlock(&inode_lock); 11831da177e4SLinus Torvalds if (sync) 11841c0eeaf5SJoern Engel inode_sync_wait(inode); 11851da177e4SLinus Torvalds return ret; 11861da177e4SLinus Torvalds } 11871da177e4SLinus Torvalds EXPORT_SYMBOL(write_inode_now); 11881da177e4SLinus Torvalds 11891da177e4SLinus Torvalds /** 11901da177e4SLinus Torvalds * sync_inode - write an inode and its pages to disk. 11911da177e4SLinus Torvalds * @inode: the inode to sync 11921da177e4SLinus Torvalds * @wbc: controls the writeback mode 11931da177e4SLinus Torvalds * 11941da177e4SLinus Torvalds * sync_inode() will write an inode and its pages to disk. It will also 11951da177e4SLinus Torvalds * correctly update the inode on its superblock's dirty inode lists and will 11961da177e4SLinus Torvalds * update inode->i_state. 11971da177e4SLinus Torvalds * 11981da177e4SLinus Torvalds * The caller must have a ref on the inode. 11991da177e4SLinus Torvalds */ 12001da177e4SLinus Torvalds int sync_inode(struct inode *inode, struct writeback_control *wbc) 12011da177e4SLinus Torvalds { 12021da177e4SLinus Torvalds int ret; 12031da177e4SLinus Torvalds 12041da177e4SLinus Torvalds spin_lock(&inode_lock); 120501c03194SChristoph Hellwig ret = writeback_single_inode(inode, wbc); 12061da177e4SLinus Torvalds spin_unlock(&inode_lock); 12071da177e4SLinus Torvalds return ret; 12081da177e4SLinus Torvalds } 12091da177e4SLinus Torvalds EXPORT_SYMBOL(sync_inode); 1210