11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * fs/fs-writeback.c 31da177e4SLinus Torvalds * 41da177e4SLinus Torvalds * Copyright (C) 2002, Linus Torvalds. 51da177e4SLinus Torvalds * 61da177e4SLinus Torvalds * Contains all the functions related to writing back and waiting 71da177e4SLinus Torvalds * upon dirty inodes against superblocks, and writing back dirty 81da177e4SLinus Torvalds * pages against inodes. ie: data writeback. Writeout of the 91da177e4SLinus Torvalds * inode itself is not handled here. 101da177e4SLinus Torvalds * 11e1f8e874SFrancois Cami * 10Apr2002 Andrew Morton 121da177e4SLinus Torvalds * Split out of fs/inode.c 131da177e4SLinus Torvalds * Additions for address_space-based writeback 141da177e4SLinus Torvalds */ 151da177e4SLinus Torvalds 161da177e4SLinus Torvalds #include <linux/kernel.h> 17f5ff8422SJens Axboe #include <linux/module.h> 181da177e4SLinus Torvalds #include <linux/spinlock.h> 195a0e3ad6STejun Heo #include <linux/slab.h> 201da177e4SLinus Torvalds #include <linux/sched.h> 211da177e4SLinus Torvalds #include <linux/fs.h> 221da177e4SLinus Torvalds #include <linux/mm.h> 2303ba3782SJens Axboe #include <linux/kthread.h> 2403ba3782SJens Axboe #include <linux/freezer.h> 251da177e4SLinus Torvalds #include <linux/writeback.h> 261da177e4SLinus Torvalds #include <linux/blkdev.h> 271da177e4SLinus Torvalds #include <linux/backing-dev.h> 281da177e4SLinus Torvalds #include <linux/buffer_head.h> 29455b2864SDave Chinner #include <linux/tracepoint.h> 3007f3f05cSDavid Howells #include "internal.h" 311da177e4SLinus Torvalds 32d0bceac7SJens Axboe /* 33c4a77a6cSJens Axboe * Passed into wb_writeback(), essentially a subset of writeback_control 34c4a77a6cSJens Axboe */ 3583ba7b07SChristoph Hellwig struct wb_writeback_work { 36c4a77a6cSJens Axboe long nr_pages; 37c4a77a6cSJens Axboe struct super_block *sb; 38c4a77a6cSJens Axboe enum writeback_sync_modes sync_mode; 3952957fe1SH Hartley Sweeten unsigned int for_kupdate:1; 4052957fe1SH Hartley Sweeten unsigned int range_cyclic:1; 4152957fe1SH Hartley Sweeten unsigned int for_background:1; 42c4a77a6cSJens Axboe 438010c3b6SJens Axboe struct list_head list; /* pending work list */ 4483ba7b07SChristoph Hellwig struct completion *done; /* set if the caller waits */ 4503ba3782SJens Axboe }; 4603ba3782SJens Axboe 47455b2864SDave Chinner /* 48455b2864SDave Chinner * Include the creation of the trace points after defining the 49455b2864SDave Chinner * wb_writeback_work structure so that the definition remains local to this 50455b2864SDave Chinner * file. 51455b2864SDave Chinner */ 52455b2864SDave Chinner #define CREATE_TRACE_POINTS 53455b2864SDave Chinner #include <trace/events/writeback.h> 54455b2864SDave Chinner 55455b2864SDave Chinner /* 56455b2864SDave Chinner * We don't actually have pdflush, but this one is exported though /proc... 57455b2864SDave Chinner */ 58455b2864SDave Chinner int nr_pdflush_threads; 59455b2864SDave Chinner 60f11b00f3SAdrian Bunk /** 61f11b00f3SAdrian Bunk * writeback_in_progress - determine whether there is writeback in progress 62f11b00f3SAdrian Bunk * @bdi: the device's backing_dev_info structure. 63f11b00f3SAdrian Bunk * 6403ba3782SJens Axboe * Determine whether there is writeback waiting to be handled against a 6503ba3782SJens Axboe * backing device. 66f11b00f3SAdrian Bunk */ 67f11b00f3SAdrian Bunk int writeback_in_progress(struct backing_dev_info *bdi) 68f11b00f3SAdrian Bunk { 6981d73a32SJan Kara return test_bit(BDI_writeback_running, &bdi->state); 70f11b00f3SAdrian Bunk } 71f11b00f3SAdrian Bunk 72692ebd17SJan Kara static inline struct backing_dev_info *inode_to_bdi(struct inode *inode) 73692ebd17SJan Kara { 74692ebd17SJan Kara struct super_block *sb = inode->i_sb; 75692ebd17SJan Kara 76aaead25bSChristoph Hellwig if (strcmp(sb->s_type->name, "bdev") == 0) 77aaead25bSChristoph Hellwig return inode->i_mapping->backing_dev_info; 78aaead25bSChristoph Hellwig 79692ebd17SJan Kara return sb->s_bdi; 80692ebd17SJan Kara } 81692ebd17SJan Kara 827ccf19a8SNick Piggin static inline struct inode *wb_inode(struct list_head *head) 837ccf19a8SNick Piggin { 847ccf19a8SNick Piggin return list_entry(head, struct inode, i_wb_list); 857ccf19a8SNick Piggin } 867ccf19a8SNick Piggin 8783ba7b07SChristoph Hellwig static void bdi_queue_work(struct backing_dev_info *bdi, 8883ba7b07SChristoph Hellwig struct wb_writeback_work *work) 894195f73dSNick Piggin { 90455b2864SDave Chinner trace_writeback_queue(bdi, work); 91455b2864SDave Chinner 926467716aSArtem Bityutskiy spin_lock_bh(&bdi->wb_lock); 9383ba7b07SChristoph Hellwig list_add_tail(&work->list, &bdi->work_list); 94fff5b85aSArtem Bityutskiy if (bdi->wb.task) { 95fff5b85aSArtem Bityutskiy wake_up_process(bdi->wb.task); 96fff5b85aSArtem Bityutskiy } else { 971da177e4SLinus Torvalds /* 98fff5b85aSArtem Bityutskiy * The bdi thread isn't there, wake up the forker thread which 99fff5b85aSArtem Bityutskiy * will create and run it. 1001da177e4SLinus Torvalds */ 101455b2864SDave Chinner trace_writeback_nothread(bdi, work); 10203ba3782SJens Axboe wake_up_process(default_backing_dev_info.wb.task); 1031da177e4SLinus Torvalds } 1046467716aSArtem Bityutskiy spin_unlock_bh(&bdi->wb_lock); 10503ba3782SJens Axboe } 1061da177e4SLinus Torvalds 10783ba7b07SChristoph Hellwig static void 10883ba7b07SChristoph Hellwig __bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, 10983ba7b07SChristoph Hellwig bool range_cyclic, bool for_background) 1101da177e4SLinus Torvalds { 11183ba7b07SChristoph Hellwig struct wb_writeback_work *work; 11203ba3782SJens Axboe 113bcddc3f0SJens Axboe /* 114bcddc3f0SJens Axboe * This is WB_SYNC_NONE writeback, so if allocation fails just 115bcddc3f0SJens Axboe * wakeup the thread for old dirty data writeback 116bcddc3f0SJens Axboe */ 11783ba7b07SChristoph Hellwig work = kzalloc(sizeof(*work), GFP_ATOMIC); 11883ba7b07SChristoph Hellwig if (!work) { 119455b2864SDave Chinner if (bdi->wb.task) { 120455b2864SDave Chinner trace_writeback_nowork(bdi); 12183ba7b07SChristoph Hellwig wake_up_process(bdi->wb.task); 122455b2864SDave Chinner } 12383ba7b07SChristoph Hellwig return; 12483ba7b07SChristoph Hellwig } 12583ba7b07SChristoph Hellwig 12683ba7b07SChristoph Hellwig work->sync_mode = WB_SYNC_NONE; 12783ba7b07SChristoph Hellwig work->nr_pages = nr_pages; 12883ba7b07SChristoph Hellwig work->range_cyclic = range_cyclic; 12983ba7b07SChristoph Hellwig work->for_background = for_background; 13083ba7b07SChristoph Hellwig 131f11fcae8SJens Axboe bdi_queue_work(bdi, work); 13203ba3782SJens Axboe } 133b6e51316SJens Axboe 134b6e51316SJens Axboe /** 135b6e51316SJens Axboe * bdi_start_writeback - start writeback 136b6e51316SJens Axboe * @bdi: the backing device to write from 137b6e51316SJens Axboe * @nr_pages: the number of pages to write 138b6e51316SJens Axboe * 139b6e51316SJens Axboe * Description: 140b6e51316SJens Axboe * This does WB_SYNC_NONE opportunistic writeback. The IO is only 141b6e51316SJens Axboe * started when this function returns, we make no guarentees on 1420e3c9a22SJens Axboe * completion. Caller need not hold sb s_umount semaphore. 143b6e51316SJens Axboe * 144b6e51316SJens Axboe */ 145c5444198SChristoph Hellwig void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages) 146b6e51316SJens Axboe { 14783ba7b07SChristoph Hellwig __bdi_start_writeback(bdi, nr_pages, true, false); 148d3ddec76SWu Fengguang } 149d3ddec76SWu Fengguang 150c5444198SChristoph Hellwig /** 151c5444198SChristoph Hellwig * bdi_start_background_writeback - start background writeback 152c5444198SChristoph Hellwig * @bdi: the backing device to write from 153c5444198SChristoph Hellwig * 154c5444198SChristoph Hellwig * Description: 155c5444198SChristoph Hellwig * This does WB_SYNC_NONE background writeback. The IO is only 156c5444198SChristoph Hellwig * started when this function returns, we make no guarentees on 157c5444198SChristoph Hellwig * completion. Caller need not hold sb s_umount semaphore. 158c5444198SChristoph Hellwig */ 159c5444198SChristoph Hellwig void bdi_start_background_writeback(struct backing_dev_info *bdi) 160c5444198SChristoph Hellwig { 16183ba7b07SChristoph Hellwig __bdi_start_writeback(bdi, LONG_MAX, true, true); 1621da177e4SLinus Torvalds } 1631da177e4SLinus Torvalds 1641da177e4SLinus Torvalds /* 1656610a0bcSAndrew Morton * Redirty an inode: set its when-it-was dirtied timestamp and move it to the 1666610a0bcSAndrew Morton * furthest end of its superblock's dirty-inode list. 1676610a0bcSAndrew Morton * 1686610a0bcSAndrew Morton * Before stamping the inode's ->dirtied_when, we check to see whether it is 16966f3b8e2SJens Axboe * already the most-recently-dirtied inode on the b_dirty list. If that is 1706610a0bcSAndrew Morton * the case then the inode must have been redirtied while it was being written 1716610a0bcSAndrew Morton * out and we don't reset its dirtied_when. 1726610a0bcSAndrew Morton */ 1736610a0bcSAndrew Morton static void redirty_tail(struct inode *inode) 1746610a0bcSAndrew Morton { 17503ba3782SJens Axboe struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; 1766610a0bcSAndrew Morton 17703ba3782SJens Axboe if (!list_empty(&wb->b_dirty)) { 17866f3b8e2SJens Axboe struct inode *tail; 1796610a0bcSAndrew Morton 1807ccf19a8SNick Piggin tail = wb_inode(wb->b_dirty.next); 18166f3b8e2SJens Axboe if (time_before(inode->dirtied_when, tail->dirtied_when)) 1826610a0bcSAndrew Morton inode->dirtied_when = jiffies; 1836610a0bcSAndrew Morton } 1847ccf19a8SNick Piggin list_move(&inode->i_wb_list, &wb->b_dirty); 1856610a0bcSAndrew Morton } 1866610a0bcSAndrew Morton 1876610a0bcSAndrew Morton /* 18866f3b8e2SJens Axboe * requeue inode for re-scanning after bdi->b_io list is exhausted. 189c986d1e2SAndrew Morton */ 1900e0f4fc2SKen Chen static void requeue_io(struct inode *inode) 191c986d1e2SAndrew Morton { 19203ba3782SJens Axboe struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; 19303ba3782SJens Axboe 1947ccf19a8SNick Piggin list_move(&inode->i_wb_list, &wb->b_more_io); 195c986d1e2SAndrew Morton } 196c986d1e2SAndrew Morton 1971c0eeaf5SJoern Engel static void inode_sync_complete(struct inode *inode) 1981c0eeaf5SJoern Engel { 1991c0eeaf5SJoern Engel /* 2001c0eeaf5SJoern Engel * Prevent speculative execution through spin_unlock(&inode_lock); 2011c0eeaf5SJoern Engel */ 2021c0eeaf5SJoern Engel smp_mb(); 2031c0eeaf5SJoern Engel wake_up_bit(&inode->i_state, __I_SYNC); 2041c0eeaf5SJoern Engel } 2051c0eeaf5SJoern Engel 206d2caa3c5SJeff Layton static bool inode_dirtied_after(struct inode *inode, unsigned long t) 207d2caa3c5SJeff Layton { 208d2caa3c5SJeff Layton bool ret = time_after(inode->dirtied_when, t); 209d2caa3c5SJeff Layton #ifndef CONFIG_64BIT 210d2caa3c5SJeff Layton /* 211d2caa3c5SJeff Layton * For inodes being constantly redirtied, dirtied_when can get stuck. 212d2caa3c5SJeff Layton * It _appears_ to be in the future, but is actually in distant past. 213d2caa3c5SJeff Layton * This test is necessary to prevent such wrapped-around relative times 2145b0830cbSJens Axboe * from permanently stopping the whole bdi writeback. 215d2caa3c5SJeff Layton */ 216d2caa3c5SJeff Layton ret = ret && time_before_eq(inode->dirtied_when, jiffies); 217d2caa3c5SJeff Layton #endif 218d2caa3c5SJeff Layton return ret; 219d2caa3c5SJeff Layton } 220d2caa3c5SJeff Layton 221c986d1e2SAndrew Morton /* 2222c136579SFengguang Wu * Move expired dirty inodes from @delaying_queue to @dispatch_queue. 2232c136579SFengguang Wu */ 2242c136579SFengguang Wu static void move_expired_inodes(struct list_head *delaying_queue, 2252c136579SFengguang Wu struct list_head *dispatch_queue, 2262c136579SFengguang Wu unsigned long *older_than_this) 2272c136579SFengguang Wu { 2285c03449dSShaohua Li LIST_HEAD(tmp); 2295c03449dSShaohua Li struct list_head *pos, *node; 230cf137307SJens Axboe struct super_block *sb = NULL; 2315c03449dSShaohua Li struct inode *inode; 232cf137307SJens Axboe int do_sb_sort = 0; 2335c03449dSShaohua Li 2342c136579SFengguang Wu while (!list_empty(delaying_queue)) { 2357ccf19a8SNick Piggin inode = wb_inode(delaying_queue->prev); 2362c136579SFengguang Wu if (older_than_this && 237d2caa3c5SJeff Layton inode_dirtied_after(inode, *older_than_this)) 2382c136579SFengguang Wu break; 239cf137307SJens Axboe if (sb && sb != inode->i_sb) 240cf137307SJens Axboe do_sb_sort = 1; 241cf137307SJens Axboe sb = inode->i_sb; 2427ccf19a8SNick Piggin list_move(&inode->i_wb_list, &tmp); 2435c03449dSShaohua Li } 2445c03449dSShaohua Li 245cf137307SJens Axboe /* just one sb in list, splice to dispatch_queue and we're done */ 246cf137307SJens Axboe if (!do_sb_sort) { 247cf137307SJens Axboe list_splice(&tmp, dispatch_queue); 248cf137307SJens Axboe return; 249cf137307SJens Axboe } 250cf137307SJens Axboe 2515c03449dSShaohua Li /* Move inodes from one superblock together */ 2525c03449dSShaohua Li while (!list_empty(&tmp)) { 2537ccf19a8SNick Piggin sb = wb_inode(tmp.prev)->i_sb; 2545c03449dSShaohua Li list_for_each_prev_safe(pos, node, &tmp) { 2557ccf19a8SNick Piggin inode = wb_inode(pos); 2565c03449dSShaohua Li if (inode->i_sb == sb) 2577ccf19a8SNick Piggin list_move(&inode->i_wb_list, dispatch_queue); 2582c136579SFengguang Wu } 2592c136579SFengguang Wu } 2605c03449dSShaohua Li } 2612c136579SFengguang Wu 2622c136579SFengguang Wu /* 2632c136579SFengguang Wu * Queue all expired dirty inodes for io, eldest first. 2644ea879b9SWu Fengguang * Before 2654ea879b9SWu Fengguang * newly dirtied b_dirty b_io b_more_io 2664ea879b9SWu Fengguang * =============> gf edc BA 2674ea879b9SWu Fengguang * After 2684ea879b9SWu Fengguang * newly dirtied b_dirty b_io b_more_io 2694ea879b9SWu Fengguang * =============> g fBAedc 2704ea879b9SWu Fengguang * | 2714ea879b9SWu Fengguang * +--> dequeue for IO 2722c136579SFengguang Wu */ 27303ba3782SJens Axboe static void queue_io(struct bdi_writeback *wb, unsigned long *older_than_this) 2742c136579SFengguang Wu { 2754ea879b9SWu Fengguang list_splice_init(&wb->b_more_io, &wb->b_io); 27603ba3782SJens Axboe move_expired_inodes(&wb->b_dirty, &wb->b_io, older_than_this); 27766f3b8e2SJens Axboe } 27866f3b8e2SJens Axboe 279a9185b41SChristoph Hellwig static int write_inode(struct inode *inode, struct writeback_control *wbc) 28066f3b8e2SJens Axboe { 28103ba3782SJens Axboe if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) 282a9185b41SChristoph Hellwig return inode->i_sb->s_op->write_inode(inode, wbc); 28303ba3782SJens Axboe return 0; 28466f3b8e2SJens Axboe } 28508d8e974SFengguang Wu 2862c136579SFengguang Wu /* 28701c03194SChristoph Hellwig * Wait for writeback on an inode to complete. 28801c03194SChristoph Hellwig */ 28901c03194SChristoph Hellwig static void inode_wait_for_writeback(struct inode *inode) 29001c03194SChristoph Hellwig { 29101c03194SChristoph Hellwig DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); 29201c03194SChristoph Hellwig wait_queue_head_t *wqh; 29301c03194SChristoph Hellwig 29401c03194SChristoph Hellwig wqh = bit_waitqueue(&inode->i_state, __I_SYNC); 29558a9d3d8SRichard Kennedy while (inode->i_state & I_SYNC) { 29601c03194SChristoph Hellwig spin_unlock(&inode_lock); 29701c03194SChristoph Hellwig __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE); 29801c03194SChristoph Hellwig spin_lock(&inode_lock); 29958a9d3d8SRichard Kennedy } 30001c03194SChristoph Hellwig } 30101c03194SChristoph Hellwig 30201c03194SChristoph Hellwig /* 30301c03194SChristoph Hellwig * Write out an inode's dirty pages. Called under inode_lock. Either the 30401c03194SChristoph Hellwig * caller has ref on the inode (either via __iget or via syscall against an fd) 30501c03194SChristoph Hellwig * or the inode has I_WILL_FREE set (via generic_forget_inode) 30601c03194SChristoph Hellwig * 3071da177e4SLinus Torvalds * If `wait' is set, wait on the writeout. 3081da177e4SLinus Torvalds * 3091da177e4SLinus Torvalds * The whole writeout design is quite complex and fragile. We want to avoid 3101da177e4SLinus Torvalds * starvation of particular inodes when others are being redirtied, prevent 3111da177e4SLinus Torvalds * livelocks, etc. 3121da177e4SLinus Torvalds * 3131da177e4SLinus Torvalds * Called under inode_lock. 3141da177e4SLinus Torvalds */ 3151da177e4SLinus Torvalds static int 31601c03194SChristoph Hellwig writeback_single_inode(struct inode *inode, struct writeback_control *wbc) 3171da177e4SLinus Torvalds { 3181da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 31901c03194SChristoph Hellwig unsigned dirty; 3201da177e4SLinus Torvalds int ret; 3211da177e4SLinus Torvalds 32201c03194SChristoph Hellwig if (!atomic_read(&inode->i_count)) 32301c03194SChristoph Hellwig WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); 32401c03194SChristoph Hellwig else 32501c03194SChristoph Hellwig WARN_ON(inode->i_state & I_WILL_FREE); 32601c03194SChristoph Hellwig 32701c03194SChristoph Hellwig if (inode->i_state & I_SYNC) { 32801c03194SChristoph Hellwig /* 32901c03194SChristoph Hellwig * If this inode is locked for writeback and we are not doing 33066f3b8e2SJens Axboe * writeback-for-data-integrity, move it to b_more_io so that 33101c03194SChristoph Hellwig * writeback can proceed with the other inodes on s_io. 33201c03194SChristoph Hellwig * 33301c03194SChristoph Hellwig * We'll have another go at writing back this inode when we 33466f3b8e2SJens Axboe * completed a full scan of b_io. 33501c03194SChristoph Hellwig */ 336a9185b41SChristoph Hellwig if (wbc->sync_mode != WB_SYNC_ALL) { 33701c03194SChristoph Hellwig requeue_io(inode); 33801c03194SChristoph Hellwig return 0; 33901c03194SChristoph Hellwig } 34001c03194SChristoph Hellwig 34101c03194SChristoph Hellwig /* 34201c03194SChristoph Hellwig * It's a data-integrity sync. We must wait. 34301c03194SChristoph Hellwig */ 34401c03194SChristoph Hellwig inode_wait_for_writeback(inode); 34501c03194SChristoph Hellwig } 34601c03194SChristoph Hellwig 3471c0eeaf5SJoern Engel BUG_ON(inode->i_state & I_SYNC); 3481da177e4SLinus Torvalds 3495547e8aaSDmitry Monakhov /* Set I_SYNC, reset I_DIRTY_PAGES */ 3501c0eeaf5SJoern Engel inode->i_state |= I_SYNC; 3515547e8aaSDmitry Monakhov inode->i_state &= ~I_DIRTY_PAGES; 3521da177e4SLinus Torvalds spin_unlock(&inode_lock); 3531da177e4SLinus Torvalds 3541da177e4SLinus Torvalds ret = do_writepages(mapping, wbc); 3551da177e4SLinus Torvalds 35626821ed4SChristoph Hellwig /* 35726821ed4SChristoph Hellwig * Make sure to wait on the data before writing out the metadata. 35826821ed4SChristoph Hellwig * This is important for filesystems that modify metadata on data 35926821ed4SChristoph Hellwig * I/O completion. 36026821ed4SChristoph Hellwig */ 361a9185b41SChristoph Hellwig if (wbc->sync_mode == WB_SYNC_ALL) { 36226821ed4SChristoph Hellwig int err = filemap_fdatawait(mapping); 3631da177e4SLinus Torvalds if (ret == 0) 3641da177e4SLinus Torvalds ret = err; 3651da177e4SLinus Torvalds } 3661da177e4SLinus Torvalds 3675547e8aaSDmitry Monakhov /* 3685547e8aaSDmitry Monakhov * Some filesystems may redirty the inode during the writeback 3695547e8aaSDmitry Monakhov * due to delalloc, clear dirty metadata flags right before 3705547e8aaSDmitry Monakhov * write_inode() 3715547e8aaSDmitry Monakhov */ 3725547e8aaSDmitry Monakhov spin_lock(&inode_lock); 3735547e8aaSDmitry Monakhov dirty = inode->i_state & I_DIRTY; 3745547e8aaSDmitry Monakhov inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC); 3755547e8aaSDmitry Monakhov spin_unlock(&inode_lock); 37626821ed4SChristoph Hellwig /* Don't write the inode if only I_DIRTY_PAGES was set */ 37726821ed4SChristoph Hellwig if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { 378a9185b41SChristoph Hellwig int err = write_inode(inode, wbc); 3791da177e4SLinus Torvalds if (ret == 0) 3801da177e4SLinus Torvalds ret = err; 3811da177e4SLinus Torvalds } 3821da177e4SLinus Torvalds 3831da177e4SLinus Torvalds spin_lock(&inode_lock); 3841c0eeaf5SJoern Engel inode->i_state &= ~I_SYNC; 385a4ffdde6SAl Viro if (!(inode->i_state & I_FREEING)) { 38623539afcSWu Fengguang if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { 3871da177e4SLinus Torvalds /* 3881da177e4SLinus Torvalds * We didn't write back all the pages. nfs_writepages() 389a50aeb40SWu Fengguang * sometimes bales out without doing anything. 3901da177e4SLinus Torvalds */ 3911da177e4SLinus Torvalds inode->i_state |= I_DIRTY_PAGES; 3928bc3be27SFengguang Wu if (wbc->nr_to_write <= 0) { 3938bc3be27SFengguang Wu /* 3948bc3be27SFengguang Wu * slice used up: queue for next turn 3958bc3be27SFengguang Wu */ 3960e0f4fc2SKen Chen requeue_io(inode); 3971da177e4SLinus Torvalds } else { 3981da177e4SLinus Torvalds /* 399a50aeb40SWu Fengguang * Writeback blocked by something other than 400a50aeb40SWu Fengguang * congestion. Delay the inode for some time to 401a50aeb40SWu Fengguang * avoid spinning on the CPU (100% iowait) 402a50aeb40SWu Fengguang * retrying writeback of the dirty page/inode 403a50aeb40SWu Fengguang * that cannot be performed immediately. 4048bc3be27SFengguang Wu */ 4058bc3be27SFengguang Wu redirty_tail(inode); 4068bc3be27SFengguang Wu } 40723539afcSWu Fengguang } else if (inode->i_state & I_DIRTY) { 40823539afcSWu Fengguang /* 40923539afcSWu Fengguang * Filesystems can dirty the inode during writeback 41023539afcSWu Fengguang * operations, such as delayed allocation during 41123539afcSWu Fengguang * submission or metadata updates after data IO 41223539afcSWu Fengguang * completion. 41323539afcSWu Fengguang */ 41423539afcSWu Fengguang redirty_tail(inode); 4151da177e4SLinus Torvalds } else { 4161da177e4SLinus Torvalds /* 4179e38d86fSNick Piggin * The inode is clean. At this point we either have 4189e38d86fSNick Piggin * a reference to the inode or it's on it's way out. 4199e38d86fSNick Piggin * No need to add it back to the LRU. 4201da177e4SLinus Torvalds */ 4217ccf19a8SNick Piggin list_del_init(&inode->i_wb_list); 4221da177e4SLinus Torvalds } 4231da177e4SLinus Torvalds } 4241c0eeaf5SJoern Engel inode_sync_complete(inode); 4251da177e4SLinus Torvalds return ret; 4261da177e4SLinus Torvalds } 4271da177e4SLinus Torvalds 42803ba3782SJens Axboe /* 429d19de7edSChristoph Hellwig * For background writeback the caller does not have the sb pinned 43003ba3782SJens Axboe * before calling writeback. So make sure that we do pin it, so it doesn't 43103ba3782SJens Axboe * go away while we are writing inodes from it. 43203ba3782SJens Axboe */ 433d19de7edSChristoph Hellwig static bool pin_sb_for_writeback(struct super_block *sb) 4341da177e4SLinus Torvalds { 43503ba3782SJens Axboe spin_lock(&sb_lock); 43629cb4859SChristoph Hellwig if (list_empty(&sb->s_instances)) { 43703ba3782SJens Axboe spin_unlock(&sb_lock); 43829cb4859SChristoph Hellwig return false; 43903ba3782SJens Axboe } 44029cb4859SChristoph Hellwig 44129cb4859SChristoph Hellwig sb->s_count++; 44229cb4859SChristoph Hellwig spin_unlock(&sb_lock); 44329cb4859SChristoph Hellwig 44429cb4859SChristoph Hellwig if (down_read_trylock(&sb->s_umount)) { 44529cb4859SChristoph Hellwig if (sb->s_root) 44629cb4859SChristoph Hellwig return true; 44703ba3782SJens Axboe up_read(&sb->s_umount); 44803ba3782SJens Axboe } 44929cb4859SChristoph Hellwig 45029cb4859SChristoph Hellwig put_super(sb); 451d19de7edSChristoph Hellwig return false; 45203ba3782SJens Axboe } 45303ba3782SJens Axboe 454f11c9c5cSEdward Shishkin /* 455f11c9c5cSEdward Shishkin * Write a portion of b_io inodes which belong to @sb. 456edadfb10SChristoph Hellwig * 457edadfb10SChristoph Hellwig * If @only_this_sb is true, then find and write all such 458f11c9c5cSEdward Shishkin * inodes. Otherwise write only ones which go sequentially 459f11c9c5cSEdward Shishkin * in reverse order. 460edadfb10SChristoph Hellwig * 461f11c9c5cSEdward Shishkin * Return 1, if the caller writeback routine should be 462f11c9c5cSEdward Shishkin * interrupted. Otherwise return 0. 463f11c9c5cSEdward Shishkin */ 464edadfb10SChristoph Hellwig static int writeback_sb_inodes(struct super_block *sb, struct bdi_writeback *wb, 465edadfb10SChristoph Hellwig struct writeback_control *wbc, bool only_this_sb) 46603ba3782SJens Axboe { 46703ba3782SJens Axboe while (!list_empty(&wb->b_io)) { 468f11c9c5cSEdward Shishkin long pages_skipped; 4697ccf19a8SNick Piggin struct inode *inode = wb_inode(wb->b_io.prev); 470edadfb10SChristoph Hellwig 471edadfb10SChristoph Hellwig if (inode->i_sb != sb) { 472edadfb10SChristoph Hellwig if (only_this_sb) { 473edadfb10SChristoph Hellwig /* 474edadfb10SChristoph Hellwig * We only want to write back data for this 475edadfb10SChristoph Hellwig * superblock, move all inodes not belonging 476edadfb10SChristoph Hellwig * to it back onto the dirty list. 477edadfb10SChristoph Hellwig */ 47866f3b8e2SJens Axboe redirty_tail(inode); 47966f3b8e2SJens Axboe continue; 48066f3b8e2SJens Axboe } 481edadfb10SChristoph Hellwig 482edadfb10SChristoph Hellwig /* 483edadfb10SChristoph Hellwig * The inode belongs to a different superblock. 484edadfb10SChristoph Hellwig * Bounce back to the caller to unpin this and 485edadfb10SChristoph Hellwig * pin the next superblock. 486edadfb10SChristoph Hellwig */ 487f11c9c5cSEdward Shishkin return 0; 488edadfb10SChristoph Hellwig } 489edadfb10SChristoph Hellwig 4909843b76aSChristoph Hellwig /* 4919843b76aSChristoph Hellwig * Don't bother with new inodes or inodes beeing freed, first 4929843b76aSChristoph Hellwig * kind does not need peridic writeout yet, and for the latter 4939843b76aSChristoph Hellwig * kind writeout is handled by the freer. 4949843b76aSChristoph Hellwig */ 4959843b76aSChristoph Hellwig if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { 4967ef0d737SNick Piggin requeue_io(inode); 4977ef0d737SNick Piggin continue; 4987ef0d737SNick Piggin } 4999843b76aSChristoph Hellwig 500d2caa3c5SJeff Layton /* 501d2caa3c5SJeff Layton * Was this inode dirtied after sync_sb_inodes was called? 502d2caa3c5SJeff Layton * This keeps sync from extra jobs and livelock. 503d2caa3c5SJeff Layton */ 504f11c9c5cSEdward Shishkin if (inode_dirtied_after(inode, wbc->wb_start)) 505f11c9c5cSEdward Shishkin return 1; 5061da177e4SLinus Torvalds 5071da177e4SLinus Torvalds __iget(inode); 5081da177e4SLinus Torvalds pages_skipped = wbc->pages_skipped; 50901c03194SChristoph Hellwig writeback_single_inode(inode, wbc); 5101da177e4SLinus Torvalds if (wbc->pages_skipped != pages_skipped) { 5111da177e4SLinus Torvalds /* 5121da177e4SLinus Torvalds * writeback is not making progress due to locked 5131da177e4SLinus Torvalds * buffers. Skip this inode for now. 5141da177e4SLinus Torvalds */ 515f57b9b7bSAndrew Morton redirty_tail(inode); 5161da177e4SLinus Torvalds } 5171da177e4SLinus Torvalds spin_unlock(&inode_lock); 5181da177e4SLinus Torvalds iput(inode); 5194ffc8444SOGAWA Hirofumi cond_resched(); 5201da177e4SLinus Torvalds spin_lock(&inode_lock); 5218bc3be27SFengguang Wu if (wbc->nr_to_write <= 0) { 5228bc3be27SFengguang Wu wbc->more_io = 1; 523f11c9c5cSEdward Shishkin return 1; 5241da177e4SLinus Torvalds } 52503ba3782SJens Axboe if (!list_empty(&wb->b_more_io)) 5268bc3be27SFengguang Wu wbc->more_io = 1; 5278bc3be27SFengguang Wu } 528f11c9c5cSEdward Shishkin /* b_io is empty */ 529f11c9c5cSEdward Shishkin return 1; 530f11c9c5cSEdward Shishkin } 53138f21977SNick Piggin 5329c3a8ee8SChristoph Hellwig void writeback_inodes_wb(struct bdi_writeback *wb, 533f11c9c5cSEdward Shishkin struct writeback_control *wbc) 534f11c9c5cSEdward Shishkin { 535f11c9c5cSEdward Shishkin int ret = 0; 5369ecc2738SJens Axboe 5377624ee72SJan Kara if (!wbc->wb_start) 538f11c9c5cSEdward Shishkin wbc->wb_start = jiffies; /* livelock avoidance */ 539f11c9c5cSEdward Shishkin spin_lock(&inode_lock); 540f11c9c5cSEdward Shishkin if (!wbc->for_kupdate || list_empty(&wb->b_io)) 541f11c9c5cSEdward Shishkin queue_io(wb, wbc->older_than_this); 542f11c9c5cSEdward Shishkin 543f11c9c5cSEdward Shishkin while (!list_empty(&wb->b_io)) { 5447ccf19a8SNick Piggin struct inode *inode = wb_inode(wb->b_io.prev); 545f11c9c5cSEdward Shishkin struct super_block *sb = inode->i_sb; 546f11c9c5cSEdward Shishkin 547334132aeSChristoph Hellwig if (!pin_sb_for_writeback(sb)) { 548334132aeSChristoph Hellwig requeue_io(inode); 549d19de7edSChristoph Hellwig continue; 550334132aeSChristoph Hellwig } 551edadfb10SChristoph Hellwig ret = writeback_sb_inodes(sb, wb, wbc, false); 552d19de7edSChristoph Hellwig drop_super(sb); 553f11c9c5cSEdward Shishkin 554f11c9c5cSEdward Shishkin if (ret) 555f11c9c5cSEdward Shishkin break; 556f11c9c5cSEdward Shishkin } 55766f3b8e2SJens Axboe spin_unlock(&inode_lock); 55866f3b8e2SJens Axboe /* Leave any unwritten inodes on b_io */ 55966f3b8e2SJens Axboe } 56066f3b8e2SJens Axboe 561edadfb10SChristoph Hellwig static void __writeback_inodes_sb(struct super_block *sb, 562edadfb10SChristoph Hellwig struct bdi_writeback *wb, struct writeback_control *wbc) 563edadfb10SChristoph Hellwig { 564edadfb10SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 565edadfb10SChristoph Hellwig 566edadfb10SChristoph Hellwig spin_lock(&inode_lock); 567edadfb10SChristoph Hellwig if (!wbc->for_kupdate || list_empty(&wb->b_io)) 568edadfb10SChristoph Hellwig queue_io(wb, wbc->older_than_this); 569edadfb10SChristoph Hellwig writeback_sb_inodes(sb, wb, wbc, true); 570edadfb10SChristoph Hellwig spin_unlock(&inode_lock); 571edadfb10SChristoph Hellwig } 572edadfb10SChristoph Hellwig 57303ba3782SJens Axboe /* 57403ba3782SJens Axboe * The maximum number of pages to writeout in a single bdi flush/kupdate 57503ba3782SJens Axboe * operation. We do this so we don't hold I_SYNC against an inode for 57603ba3782SJens Axboe * enormous amounts of time, which would block a userspace task which has 57703ba3782SJens Axboe * been forced to throttle against that inode. Also, the code reevaluates 57803ba3782SJens Axboe * the dirty each time it has written this many pages. 57903ba3782SJens Axboe */ 58003ba3782SJens Axboe #define MAX_WRITEBACK_PAGES 1024 58103ba3782SJens Axboe 58203ba3782SJens Axboe static inline bool over_bground_thresh(void) 58303ba3782SJens Axboe { 58403ba3782SJens Axboe unsigned long background_thresh, dirty_thresh; 58503ba3782SJens Axboe 58616c4042fSWu Fengguang global_dirty_limits(&background_thresh, &dirty_thresh); 58703ba3782SJens Axboe 58803ba3782SJens Axboe return (global_page_state(NR_FILE_DIRTY) + 58903ba3782SJens Axboe global_page_state(NR_UNSTABLE_NFS) >= background_thresh); 59003ba3782SJens Axboe } 59103ba3782SJens Axboe 59203ba3782SJens Axboe /* 59303ba3782SJens Axboe * Explicit flushing or periodic writeback of "old" data. 59403ba3782SJens Axboe * 59503ba3782SJens Axboe * Define "old": the first time one of an inode's pages is dirtied, we mark the 59603ba3782SJens Axboe * dirtying-time in the inode's address_space. So this periodic writeback code 59703ba3782SJens Axboe * just walks the superblock inode list, writing back any inodes which are 59803ba3782SJens Axboe * older than a specific point in time. 59903ba3782SJens Axboe * 60003ba3782SJens Axboe * Try to run once per dirty_writeback_interval. But if a writeback event 60103ba3782SJens Axboe * takes longer than a dirty_writeback_interval interval, then leave a 60203ba3782SJens Axboe * one-second gap. 60303ba3782SJens Axboe * 60403ba3782SJens Axboe * older_than_this takes precedence over nr_to_write. So we'll only write back 60503ba3782SJens Axboe * all dirty pages if they are all attached to "old" mappings. 60603ba3782SJens Axboe */ 607c4a77a6cSJens Axboe static long wb_writeback(struct bdi_writeback *wb, 60883ba7b07SChristoph Hellwig struct wb_writeback_work *work) 60903ba3782SJens Axboe { 61003ba3782SJens Axboe struct writeback_control wbc = { 61183ba7b07SChristoph Hellwig .sync_mode = work->sync_mode, 61203ba3782SJens Axboe .older_than_this = NULL, 61383ba7b07SChristoph Hellwig .for_kupdate = work->for_kupdate, 61483ba7b07SChristoph Hellwig .for_background = work->for_background, 61583ba7b07SChristoph Hellwig .range_cyclic = work->range_cyclic, 61603ba3782SJens Axboe }; 61703ba3782SJens Axboe unsigned long oldest_jif; 61803ba3782SJens Axboe long wrote = 0; 619a5989bdcSJan Kara struct inode *inode; 62003ba3782SJens Axboe 62103ba3782SJens Axboe if (wbc.for_kupdate) { 62203ba3782SJens Axboe wbc.older_than_this = &oldest_jif; 62303ba3782SJens Axboe oldest_jif = jiffies - 62403ba3782SJens Axboe msecs_to_jiffies(dirty_expire_interval * 10); 62503ba3782SJens Axboe } 626c4a77a6cSJens Axboe if (!wbc.range_cyclic) { 627c4a77a6cSJens Axboe wbc.range_start = 0; 628c4a77a6cSJens Axboe wbc.range_end = LLONG_MAX; 629c4a77a6cSJens Axboe } 63003ba3782SJens Axboe 6317624ee72SJan Kara wbc.wb_start = jiffies; /* livelock avoidance */ 63203ba3782SJens Axboe for (;;) { 63303ba3782SJens Axboe /* 634d3ddec76SWu Fengguang * Stop writeback when nr_pages has been consumed 63503ba3782SJens Axboe */ 63683ba7b07SChristoph Hellwig if (work->nr_pages <= 0) 63703ba3782SJens Axboe break; 63803ba3782SJens Axboe 63903ba3782SJens Axboe /* 640d3ddec76SWu Fengguang * For background writeout, stop when we are below the 641d3ddec76SWu Fengguang * background dirty threshold 64203ba3782SJens Axboe */ 64383ba7b07SChristoph Hellwig if (work->for_background && !over_bground_thresh()) 64403ba3782SJens Axboe break; 64503ba3782SJens Axboe 64603ba3782SJens Axboe wbc.more_io = 0; 64703ba3782SJens Axboe wbc.nr_to_write = MAX_WRITEBACK_PAGES; 64803ba3782SJens Axboe wbc.pages_skipped = 0; 649028c2dd1SDave Chinner 650028c2dd1SDave Chinner trace_wbc_writeback_start(&wbc, wb->bdi); 65183ba7b07SChristoph Hellwig if (work->sb) 65283ba7b07SChristoph Hellwig __writeback_inodes_sb(work->sb, wb, &wbc); 653edadfb10SChristoph Hellwig else 65403ba3782SJens Axboe writeback_inodes_wb(wb, &wbc); 655028c2dd1SDave Chinner trace_wbc_writeback_written(&wbc, wb->bdi); 656028c2dd1SDave Chinner 65783ba7b07SChristoph Hellwig work->nr_pages -= MAX_WRITEBACK_PAGES - wbc.nr_to_write; 65803ba3782SJens Axboe wrote += MAX_WRITEBACK_PAGES - wbc.nr_to_write; 65903ba3782SJens Axboe 66003ba3782SJens Axboe /* 66171fd05a8SJens Axboe * If we consumed everything, see if we have more 66203ba3782SJens Axboe */ 66371fd05a8SJens Axboe if (wbc.nr_to_write <= 0) 66471fd05a8SJens Axboe continue; 66571fd05a8SJens Axboe /* 66671fd05a8SJens Axboe * Didn't write everything and we don't have more IO, bail 66771fd05a8SJens Axboe */ 66871fd05a8SJens Axboe if (!wbc.more_io) 66971fd05a8SJens Axboe break; 67071fd05a8SJens Axboe /* 67171fd05a8SJens Axboe * Did we write something? Try for more 67271fd05a8SJens Axboe */ 673a5989bdcSJan Kara if (wbc.nr_to_write < MAX_WRITEBACK_PAGES) 67403ba3782SJens Axboe continue; 675a5989bdcSJan Kara /* 676a5989bdcSJan Kara * Nothing written. Wait for some inode to 677a5989bdcSJan Kara * become available for writeback. Otherwise 678a5989bdcSJan Kara * we'll just busyloop. 679a5989bdcSJan Kara */ 680a5989bdcSJan Kara spin_lock(&inode_lock); 681a5989bdcSJan Kara if (!list_empty(&wb->b_more_io)) { 6827ccf19a8SNick Piggin inode = wb_inode(wb->b_more_io.prev); 683028c2dd1SDave Chinner trace_wbc_writeback_wait(&wbc, wb->bdi); 684a5989bdcSJan Kara inode_wait_for_writeback(inode); 685a5989bdcSJan Kara } 686a5989bdcSJan Kara spin_unlock(&inode_lock); 68703ba3782SJens Axboe } 68803ba3782SJens Axboe 68903ba3782SJens Axboe return wrote; 69003ba3782SJens Axboe } 69103ba3782SJens Axboe 69203ba3782SJens Axboe /* 69383ba7b07SChristoph Hellwig * Return the next wb_writeback_work struct that hasn't been processed yet. 69403ba3782SJens Axboe */ 69583ba7b07SChristoph Hellwig static struct wb_writeback_work * 69608852b6dSMinchan Kim get_next_work_item(struct backing_dev_info *bdi) 69703ba3782SJens Axboe { 69883ba7b07SChristoph Hellwig struct wb_writeback_work *work = NULL; 69903ba3782SJens Axboe 7006467716aSArtem Bityutskiy spin_lock_bh(&bdi->wb_lock); 70183ba7b07SChristoph Hellwig if (!list_empty(&bdi->work_list)) { 70283ba7b07SChristoph Hellwig work = list_entry(bdi->work_list.next, 70383ba7b07SChristoph Hellwig struct wb_writeback_work, list); 70483ba7b07SChristoph Hellwig list_del_init(&work->list); 70503ba3782SJens Axboe } 7066467716aSArtem Bityutskiy spin_unlock_bh(&bdi->wb_lock); 70783ba7b07SChristoph Hellwig return work; 70803ba3782SJens Axboe } 70903ba3782SJens Axboe 71003ba3782SJens Axboe static long wb_check_old_data_flush(struct bdi_writeback *wb) 71103ba3782SJens Axboe { 71203ba3782SJens Axboe unsigned long expired; 71303ba3782SJens Axboe long nr_pages; 71403ba3782SJens Axboe 71569b62d01SJens Axboe /* 71669b62d01SJens Axboe * When set to zero, disable periodic writeback 71769b62d01SJens Axboe */ 71869b62d01SJens Axboe if (!dirty_writeback_interval) 71969b62d01SJens Axboe return 0; 72069b62d01SJens Axboe 72103ba3782SJens Axboe expired = wb->last_old_flush + 72203ba3782SJens Axboe msecs_to_jiffies(dirty_writeback_interval * 10); 72303ba3782SJens Axboe if (time_before(jiffies, expired)) 72403ba3782SJens Axboe return 0; 72503ba3782SJens Axboe 72603ba3782SJens Axboe wb->last_old_flush = jiffies; 72703ba3782SJens Axboe nr_pages = global_page_state(NR_FILE_DIRTY) + 72803ba3782SJens Axboe global_page_state(NR_UNSTABLE_NFS) + 729cffbc8aaSDave Chinner get_nr_dirty_inodes(); 73003ba3782SJens Axboe 731c4a77a6cSJens Axboe if (nr_pages) { 73283ba7b07SChristoph Hellwig struct wb_writeback_work work = { 733c4a77a6cSJens Axboe .nr_pages = nr_pages, 734c4a77a6cSJens Axboe .sync_mode = WB_SYNC_NONE, 735c4a77a6cSJens Axboe .for_kupdate = 1, 736c4a77a6cSJens Axboe .range_cyclic = 1, 737c4a77a6cSJens Axboe }; 738c4a77a6cSJens Axboe 73983ba7b07SChristoph Hellwig return wb_writeback(wb, &work); 740c4a77a6cSJens Axboe } 74103ba3782SJens Axboe 74203ba3782SJens Axboe return 0; 74303ba3782SJens Axboe } 74403ba3782SJens Axboe 74503ba3782SJens Axboe /* 74603ba3782SJens Axboe * Retrieve work items and do the writeback they describe 74703ba3782SJens Axboe */ 74803ba3782SJens Axboe long wb_do_writeback(struct bdi_writeback *wb, int force_wait) 74903ba3782SJens Axboe { 75003ba3782SJens Axboe struct backing_dev_info *bdi = wb->bdi; 75183ba7b07SChristoph Hellwig struct wb_writeback_work *work; 752c4a77a6cSJens Axboe long wrote = 0; 75303ba3782SJens Axboe 75481d73a32SJan Kara set_bit(BDI_writeback_running, &wb->bdi->state); 75508852b6dSMinchan Kim while ((work = get_next_work_item(bdi)) != NULL) { 75603ba3782SJens Axboe /* 75703ba3782SJens Axboe * Override sync mode, in case we must wait for completion 75883ba7b07SChristoph Hellwig * because this thread is exiting now. 75903ba3782SJens Axboe */ 76003ba3782SJens Axboe if (force_wait) 76183ba7b07SChristoph Hellwig work->sync_mode = WB_SYNC_ALL; 76283ba7b07SChristoph Hellwig 763455b2864SDave Chinner trace_writeback_exec(bdi, work); 764455b2864SDave Chinner 76583ba7b07SChristoph Hellwig wrote += wb_writeback(wb, work); 76603ba3782SJens Axboe 76703ba3782SJens Axboe /* 76883ba7b07SChristoph Hellwig * Notify the caller of completion if this is a synchronous 76983ba7b07SChristoph Hellwig * work item, otherwise just free it. 77003ba3782SJens Axboe */ 77183ba7b07SChristoph Hellwig if (work->done) 77283ba7b07SChristoph Hellwig complete(work->done); 77383ba7b07SChristoph Hellwig else 77483ba7b07SChristoph Hellwig kfree(work); 77503ba3782SJens Axboe } 77603ba3782SJens Axboe 77703ba3782SJens Axboe /* 77803ba3782SJens Axboe * Check for periodic writeback, kupdated() style 77903ba3782SJens Axboe */ 78003ba3782SJens Axboe wrote += wb_check_old_data_flush(wb); 78181d73a32SJan Kara clear_bit(BDI_writeback_running, &wb->bdi->state); 78203ba3782SJens Axboe 78303ba3782SJens Axboe return wrote; 78403ba3782SJens Axboe } 78503ba3782SJens Axboe 78603ba3782SJens Axboe /* 78703ba3782SJens Axboe * Handle writeback of dirty data for the device backed by this bdi. Also 78803ba3782SJens Axboe * wakes up periodically and does kupdated style flushing. 78903ba3782SJens Axboe */ 79008243900SChristoph Hellwig int bdi_writeback_thread(void *data) 79103ba3782SJens Axboe { 79208243900SChristoph Hellwig struct bdi_writeback *wb = data; 79308243900SChristoph Hellwig struct backing_dev_info *bdi = wb->bdi; 79403ba3782SJens Axboe long pages_written; 79503ba3782SJens Axboe 79608243900SChristoph Hellwig current->flags |= PF_FLUSHER | PF_SWAPWRITE; 79708243900SChristoph Hellwig set_freezable(); 798ecd58403SArtem Bityutskiy wb->last_active = jiffies; 79903ba3782SJens Axboe 80003ba3782SJens Axboe /* 80108243900SChristoph Hellwig * Our parent may run at a different priority, just set us to normal 80203ba3782SJens Axboe */ 80308243900SChristoph Hellwig set_user_nice(current, 0); 80408243900SChristoph Hellwig 805455b2864SDave Chinner trace_writeback_thread_start(bdi); 806455b2864SDave Chinner 80703ba3782SJens Axboe while (!kthread_should_stop()) { 8086467716aSArtem Bityutskiy /* 8096467716aSArtem Bityutskiy * Remove own delayed wake-up timer, since we are already awake 8106467716aSArtem Bityutskiy * and we'll take care of the preriodic write-back. 8116467716aSArtem Bityutskiy */ 8126467716aSArtem Bityutskiy del_timer(&wb->wakeup_timer); 8136467716aSArtem Bityutskiy 81403ba3782SJens Axboe pages_written = wb_do_writeback(wb, 0); 81503ba3782SJens Axboe 816455b2864SDave Chinner trace_writeback_pages_written(pages_written); 817455b2864SDave Chinner 81803ba3782SJens Axboe if (pages_written) 819ecd58403SArtem Bityutskiy wb->last_active = jiffies; 82003ba3782SJens Axboe 821297252c8SArtem Bityutskiy set_current_state(TASK_INTERRUPTIBLE); 822b76b4014SJ. Bruce Fields if (!list_empty(&bdi->work_list) || kthread_should_stop()) { 823297252c8SArtem Bityutskiy __set_current_state(TASK_RUNNING); 824297252c8SArtem Bityutskiy continue; 82503ba3782SJens Axboe } 82603ba3782SJens Axboe 827253c34e9SArtem Bityutskiy if (wb_has_dirty_io(wb) && dirty_writeback_interval) 828fff5b85aSArtem Bityutskiy schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10)); 829253c34e9SArtem Bityutskiy else { 830253c34e9SArtem Bityutskiy /* 831253c34e9SArtem Bityutskiy * We have nothing to do, so can go sleep without any 832253c34e9SArtem Bityutskiy * timeout and save power. When a work is queued or 833253c34e9SArtem Bityutskiy * something is made dirty - we will be woken up. 834253c34e9SArtem Bityutskiy */ 83569b62d01SJens Axboe schedule(); 836f9eadbbdSJens Axboe } 83769b62d01SJens Axboe 83803ba3782SJens Axboe try_to_freeze(); 83903ba3782SJens Axboe } 84003ba3782SJens Axboe 841fff5b85aSArtem Bityutskiy /* Flush any work that raced with us exiting */ 84208243900SChristoph Hellwig if (!list_empty(&bdi->work_list)) 84308243900SChristoph Hellwig wb_do_writeback(wb, 1); 844455b2864SDave Chinner 845455b2864SDave Chinner trace_writeback_thread_stop(bdi); 84603ba3782SJens Axboe return 0; 84703ba3782SJens Axboe } 84803ba3782SJens Axboe 84908243900SChristoph Hellwig 85003ba3782SJens Axboe /* 85103ba3782SJens Axboe * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back 85203ba3782SJens Axboe * the whole world. 85303ba3782SJens Axboe */ 85403ba3782SJens Axboe void wakeup_flusher_threads(long nr_pages) 85503ba3782SJens Axboe { 856b8c2f347SChristoph Hellwig struct backing_dev_info *bdi; 857b8c2f347SChristoph Hellwig 85883ba7b07SChristoph Hellwig if (!nr_pages) { 85983ba7b07SChristoph Hellwig nr_pages = global_page_state(NR_FILE_DIRTY) + 86003ba3782SJens Axboe global_page_state(NR_UNSTABLE_NFS); 861b8c2f347SChristoph Hellwig } 862b8c2f347SChristoph Hellwig 863b8c2f347SChristoph Hellwig rcu_read_lock(); 864b8c2f347SChristoph Hellwig list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { 865b8c2f347SChristoph Hellwig if (!bdi_has_dirty_io(bdi)) 866b8c2f347SChristoph Hellwig continue; 86783ba7b07SChristoph Hellwig __bdi_start_writeback(bdi, nr_pages, false, false); 868b8c2f347SChristoph Hellwig } 869b8c2f347SChristoph Hellwig rcu_read_unlock(); 87003ba3782SJens Axboe } 87103ba3782SJens Axboe 87203ba3782SJens Axboe static noinline void block_dump___mark_inode_dirty(struct inode *inode) 87303ba3782SJens Axboe { 87403ba3782SJens Axboe if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { 87503ba3782SJens Axboe struct dentry *dentry; 87603ba3782SJens Axboe const char *name = "?"; 87703ba3782SJens Axboe 87803ba3782SJens Axboe dentry = d_find_alias(inode); 87903ba3782SJens Axboe if (dentry) { 88003ba3782SJens Axboe spin_lock(&dentry->d_lock); 88103ba3782SJens Axboe name = (const char *) dentry->d_name.name; 88203ba3782SJens Axboe } 88303ba3782SJens Axboe printk(KERN_DEBUG 88403ba3782SJens Axboe "%s(%d): dirtied inode %lu (%s) on %s\n", 88503ba3782SJens Axboe current->comm, task_pid_nr(current), inode->i_ino, 88603ba3782SJens Axboe name, inode->i_sb->s_id); 88703ba3782SJens Axboe if (dentry) { 88803ba3782SJens Axboe spin_unlock(&dentry->d_lock); 88903ba3782SJens Axboe dput(dentry); 89003ba3782SJens Axboe } 89103ba3782SJens Axboe } 89203ba3782SJens Axboe } 89303ba3782SJens Axboe 89403ba3782SJens Axboe /** 89503ba3782SJens Axboe * __mark_inode_dirty - internal function 89603ba3782SJens Axboe * @inode: inode to mark 89703ba3782SJens Axboe * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) 89803ba3782SJens Axboe * Mark an inode as dirty. Callers should use mark_inode_dirty or 89903ba3782SJens Axboe * mark_inode_dirty_sync. 90003ba3782SJens Axboe * 90103ba3782SJens Axboe * Put the inode on the super block's dirty list. 90203ba3782SJens Axboe * 90303ba3782SJens Axboe * CAREFUL! We mark it dirty unconditionally, but move it onto the 90403ba3782SJens Axboe * dirty list only if it is hashed or if it refers to a blockdev. 90503ba3782SJens Axboe * If it was not hashed, it will never be added to the dirty list 90603ba3782SJens Axboe * even if it is later hashed, as it will have been marked dirty already. 90703ba3782SJens Axboe * 90803ba3782SJens Axboe * In short, make sure you hash any inodes _before_ you start marking 90903ba3782SJens Axboe * them dirty. 91003ba3782SJens Axboe * 91103ba3782SJens Axboe * This function *must* be atomic for the I_DIRTY_PAGES case - 91203ba3782SJens Axboe * set_page_dirty() is called under spinlock in several places. 91303ba3782SJens Axboe * 91403ba3782SJens Axboe * Note that for blockdevs, inode->dirtied_when represents the dirtying time of 91503ba3782SJens Axboe * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of 91603ba3782SJens Axboe * the kernel-internal blockdev inode represents the dirtying time of the 91703ba3782SJens Axboe * blockdev's pages. This is why for I_DIRTY_PAGES we always use 91803ba3782SJens Axboe * page->mapping->host, so the page-dirtying time is recorded in the internal 91903ba3782SJens Axboe * blockdev inode. 92003ba3782SJens Axboe */ 92103ba3782SJens Axboe void __mark_inode_dirty(struct inode *inode, int flags) 92203ba3782SJens Axboe { 92303ba3782SJens Axboe struct super_block *sb = inode->i_sb; 924253c34e9SArtem Bityutskiy struct backing_dev_info *bdi = NULL; 925253c34e9SArtem Bityutskiy bool wakeup_bdi = false; 92603ba3782SJens Axboe 92703ba3782SJens Axboe /* 92803ba3782SJens Axboe * Don't do this for I_DIRTY_PAGES - that doesn't actually 92903ba3782SJens Axboe * dirty the inode itself 93003ba3782SJens Axboe */ 93103ba3782SJens Axboe if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { 93203ba3782SJens Axboe if (sb->s_op->dirty_inode) 93303ba3782SJens Axboe sb->s_op->dirty_inode(inode); 93403ba3782SJens Axboe } 93503ba3782SJens Axboe 93603ba3782SJens Axboe /* 93703ba3782SJens Axboe * make sure that changes are seen by all cpus before we test i_state 93803ba3782SJens Axboe * -- mikulas 93903ba3782SJens Axboe */ 94003ba3782SJens Axboe smp_mb(); 94103ba3782SJens Axboe 94203ba3782SJens Axboe /* avoid the locking if we can */ 94303ba3782SJens Axboe if ((inode->i_state & flags) == flags) 94403ba3782SJens Axboe return; 94503ba3782SJens Axboe 94603ba3782SJens Axboe if (unlikely(block_dump)) 94703ba3782SJens Axboe block_dump___mark_inode_dirty(inode); 94803ba3782SJens Axboe 94903ba3782SJens Axboe spin_lock(&inode_lock); 95003ba3782SJens Axboe if ((inode->i_state & flags) != flags) { 95103ba3782SJens Axboe const int was_dirty = inode->i_state & I_DIRTY; 95203ba3782SJens Axboe 95303ba3782SJens Axboe inode->i_state |= flags; 95403ba3782SJens Axboe 95503ba3782SJens Axboe /* 95603ba3782SJens Axboe * If the inode is being synced, just update its dirty state. 95703ba3782SJens Axboe * The unlocker will place the inode on the appropriate 95803ba3782SJens Axboe * superblock list, based upon its state. 95903ba3782SJens Axboe */ 96003ba3782SJens Axboe if (inode->i_state & I_SYNC) 96103ba3782SJens Axboe goto out; 96203ba3782SJens Axboe 96303ba3782SJens Axboe /* 96403ba3782SJens Axboe * Only add valid (hashed) inodes to the superblock's 96503ba3782SJens Axboe * dirty list. Add blockdev inodes as well. 96603ba3782SJens Axboe */ 96703ba3782SJens Axboe if (!S_ISBLK(inode->i_mode)) { 9681d3382cbSAl Viro if (inode_unhashed(inode)) 96903ba3782SJens Axboe goto out; 97003ba3782SJens Axboe } 971a4ffdde6SAl Viro if (inode->i_state & I_FREEING) 97203ba3782SJens Axboe goto out; 97303ba3782SJens Axboe 97403ba3782SJens Axboe /* 97503ba3782SJens Axboe * If the inode was already on b_dirty/b_io/b_more_io, don't 97603ba3782SJens Axboe * reposition it (that would break b_dirty time-ordering). 97703ba3782SJens Axboe */ 97803ba3782SJens Axboe if (!was_dirty) { 979253c34e9SArtem Bityutskiy bdi = inode_to_bdi(inode); 980500b067cSJens Axboe 981253c34e9SArtem Bityutskiy if (bdi_cap_writeback_dirty(bdi)) { 982253c34e9SArtem Bityutskiy WARN(!test_bit(BDI_registered, &bdi->state), 983253c34e9SArtem Bityutskiy "bdi-%s not registered\n", bdi->name); 984253c34e9SArtem Bityutskiy 985253c34e9SArtem Bityutskiy /* 986253c34e9SArtem Bityutskiy * If this is the first dirty inode for this 987253c34e9SArtem Bityutskiy * bdi, we have to wake-up the corresponding 988253c34e9SArtem Bityutskiy * bdi thread to make sure background 989253c34e9SArtem Bityutskiy * write-back happens later. 990253c34e9SArtem Bityutskiy */ 991253c34e9SArtem Bityutskiy if (!wb_has_dirty_io(&bdi->wb)) 992253c34e9SArtem Bityutskiy wakeup_bdi = true; 993500b067cSJens Axboe } 99403ba3782SJens Axboe 99503ba3782SJens Axboe inode->dirtied_when = jiffies; 9967ccf19a8SNick Piggin list_move(&inode->i_wb_list, &bdi->wb.b_dirty); 99703ba3782SJens Axboe } 99803ba3782SJens Axboe } 99903ba3782SJens Axboe out: 100003ba3782SJens Axboe spin_unlock(&inode_lock); 1001253c34e9SArtem Bityutskiy 1002253c34e9SArtem Bityutskiy if (wakeup_bdi) 10036467716aSArtem Bityutskiy bdi_wakeup_thread_delayed(bdi); 100403ba3782SJens Axboe } 100503ba3782SJens Axboe EXPORT_SYMBOL(__mark_inode_dirty); 100603ba3782SJens Axboe 100766f3b8e2SJens Axboe /* 100866f3b8e2SJens Axboe * Write out a superblock's list of dirty inodes. A wait will be performed 100966f3b8e2SJens Axboe * upon no inodes, all inodes or the final one, depending upon sync_mode. 101066f3b8e2SJens Axboe * 101166f3b8e2SJens Axboe * If older_than_this is non-NULL, then only write out inodes which 101266f3b8e2SJens Axboe * had their first dirtying at a time earlier than *older_than_this. 101366f3b8e2SJens Axboe * 101466f3b8e2SJens Axboe * If `bdi' is non-zero then we're being asked to writeback a specific queue. 101566f3b8e2SJens Axboe * This function assumes that the blockdev superblock's inodes are backed by 101666f3b8e2SJens Axboe * a variety of queues, so all inodes are searched. For other superblocks, 101766f3b8e2SJens Axboe * assume that all inodes are backed by the same queue. 101866f3b8e2SJens Axboe * 101966f3b8e2SJens Axboe * The inodes to be written are parked on bdi->b_io. They are moved back onto 102066f3b8e2SJens Axboe * bdi->b_dirty as they are selected for writing. This way, none can be missed 102166f3b8e2SJens Axboe * on the writer throttling path, and we get decent balancing between many 102266f3b8e2SJens Axboe * throttled threads: we don't want them all piling up on inode_sync_wait. 102366f3b8e2SJens Axboe */ 1024b6e51316SJens Axboe static void wait_sb_inodes(struct super_block *sb) 102566f3b8e2SJens Axboe { 102638f21977SNick Piggin struct inode *inode, *old_inode = NULL; 102738f21977SNick Piggin 102803ba3782SJens Axboe /* 102903ba3782SJens Axboe * We need to be protected against the filesystem going from 103003ba3782SJens Axboe * r/o to r/w or vice versa. 103103ba3782SJens Axboe */ 1032b6e51316SJens Axboe WARN_ON(!rwsem_is_locked(&sb->s_umount)); 103303ba3782SJens Axboe 103466f3b8e2SJens Axboe spin_lock(&inode_lock); 103566f3b8e2SJens Axboe 103638f21977SNick Piggin /* 103738f21977SNick Piggin * Data integrity sync. Must wait for all pages under writeback, 103838f21977SNick Piggin * because there may have been pages dirtied before our sync 103938f21977SNick Piggin * call, but which had writeout started before we write it out. 104038f21977SNick Piggin * In which case, the inode may not be on the dirty list, but 104138f21977SNick Piggin * we still have to wait for that writeout. 104238f21977SNick Piggin */ 1043b6e51316SJens Axboe list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { 104438f21977SNick Piggin struct address_space *mapping; 104538f21977SNick Piggin 1046a4ffdde6SAl Viro if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) 104738f21977SNick Piggin continue; 104838f21977SNick Piggin mapping = inode->i_mapping; 104938f21977SNick Piggin if (mapping->nrpages == 0) 105038f21977SNick Piggin continue; 105138f21977SNick Piggin __iget(inode); 1052ae8547b0SHans Reiser spin_unlock(&inode_lock); 105338f21977SNick Piggin /* 105438f21977SNick Piggin * We hold a reference to 'inode' so it couldn't have 105538f21977SNick Piggin * been removed from s_inodes list while we dropped the 105638f21977SNick Piggin * inode_lock. We cannot iput the inode now as we can 105738f21977SNick Piggin * be holding the last reference and we cannot iput it 105838f21977SNick Piggin * under inode_lock. So we keep the reference and iput 105938f21977SNick Piggin * it later. 106038f21977SNick Piggin */ 106138f21977SNick Piggin iput(old_inode); 106238f21977SNick Piggin old_inode = inode; 106338f21977SNick Piggin 106438f21977SNick Piggin filemap_fdatawait(mapping); 106538f21977SNick Piggin 106638f21977SNick Piggin cond_resched(); 106738f21977SNick Piggin 106838f21977SNick Piggin spin_lock(&inode_lock); 106938f21977SNick Piggin } 107038f21977SNick Piggin spin_unlock(&inode_lock); 107138f21977SNick Piggin iput(old_inode); 107266f3b8e2SJens Axboe } 10731da177e4SLinus Torvalds 1074d8a8559cSJens Axboe /** 1075d8a8559cSJens Axboe * writeback_inodes_sb - writeback dirty inodes from given super_block 1076d8a8559cSJens Axboe * @sb: the superblock 10771da177e4SLinus Torvalds * 1078d8a8559cSJens Axboe * Start writeback on some inodes on this super_block. No guarantees are made 1079d8a8559cSJens Axboe * on how many (if any) will be written, and this function does not wait 1080d8a8559cSJens Axboe * for IO completion of submitted IO. The number of pages submitted is 1081d8a8559cSJens Axboe * returned. 10821da177e4SLinus Torvalds */ 1083b6e51316SJens Axboe void writeback_inodes_sb(struct super_block *sb) 10841da177e4SLinus Torvalds { 10850e3c9a22SJens Axboe unsigned long nr_dirty = global_page_state(NR_FILE_DIRTY); 10860e3c9a22SJens Axboe unsigned long nr_unstable = global_page_state(NR_UNSTABLE_NFS); 108783ba7b07SChristoph Hellwig DECLARE_COMPLETION_ONSTACK(done); 108883ba7b07SChristoph Hellwig struct wb_writeback_work work = { 10893c4d7165SChristoph Hellwig .sb = sb, 10903c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_NONE, 109183ba7b07SChristoph Hellwig .done = &done, 10923c4d7165SChristoph Hellwig }; 10930e3c9a22SJens Axboe 1094cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 1095cf37e972SChristoph Hellwig 1096cffbc8aaSDave Chinner work.nr_pages = nr_dirty + nr_unstable + get_nr_dirty_inodes(); 10970e3c9a22SJens Axboe 109883ba7b07SChristoph Hellwig bdi_queue_work(sb->s_bdi, &work); 109983ba7b07SChristoph Hellwig wait_for_completion(&done); 11001da177e4SLinus Torvalds } 1101d8a8559cSJens Axboe EXPORT_SYMBOL(writeback_inodes_sb); 1102d8a8559cSJens Axboe 1103d8a8559cSJens Axboe /** 110417bd55d0SEric Sandeen * writeback_inodes_sb_if_idle - start writeback if none underway 110517bd55d0SEric Sandeen * @sb: the superblock 110617bd55d0SEric Sandeen * 110717bd55d0SEric Sandeen * Invoke writeback_inodes_sb if no writeback is currently underway. 110817bd55d0SEric Sandeen * Returns 1 if writeback was started, 0 if not. 110917bd55d0SEric Sandeen */ 111017bd55d0SEric Sandeen int writeback_inodes_sb_if_idle(struct super_block *sb) 111117bd55d0SEric Sandeen { 111217bd55d0SEric Sandeen if (!writeback_in_progress(sb->s_bdi)) { 1113cf37e972SChristoph Hellwig down_read(&sb->s_umount); 111417bd55d0SEric Sandeen writeback_inodes_sb(sb); 1115cf37e972SChristoph Hellwig up_read(&sb->s_umount); 111617bd55d0SEric Sandeen return 1; 111717bd55d0SEric Sandeen } else 111817bd55d0SEric Sandeen return 0; 111917bd55d0SEric Sandeen } 112017bd55d0SEric Sandeen EXPORT_SYMBOL(writeback_inodes_sb_if_idle); 112117bd55d0SEric Sandeen 112217bd55d0SEric Sandeen /** 1123d8a8559cSJens Axboe * sync_inodes_sb - sync sb inode pages 1124d8a8559cSJens Axboe * @sb: the superblock 1125d8a8559cSJens Axboe * 1126d8a8559cSJens Axboe * This function writes and waits on any dirty inode belonging to this 1127d8a8559cSJens Axboe * super_block. The number of pages synced is returned. 1128d8a8559cSJens Axboe */ 1129b6e51316SJens Axboe void sync_inodes_sb(struct super_block *sb) 1130d8a8559cSJens Axboe { 113183ba7b07SChristoph Hellwig DECLARE_COMPLETION_ONSTACK(done); 113283ba7b07SChristoph Hellwig struct wb_writeback_work work = { 11333c4d7165SChristoph Hellwig .sb = sb, 11343c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_ALL, 11353c4d7165SChristoph Hellwig .nr_pages = LONG_MAX, 11363c4d7165SChristoph Hellwig .range_cyclic = 0, 113783ba7b07SChristoph Hellwig .done = &done, 11383c4d7165SChristoph Hellwig }; 11393c4d7165SChristoph Hellwig 1140cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 1141cf37e972SChristoph Hellwig 114283ba7b07SChristoph Hellwig bdi_queue_work(sb->s_bdi, &work); 114383ba7b07SChristoph Hellwig wait_for_completion(&done); 114483ba7b07SChristoph Hellwig 1145b6e51316SJens Axboe wait_sb_inodes(sb); 1146d8a8559cSJens Axboe } 1147d8a8559cSJens Axboe EXPORT_SYMBOL(sync_inodes_sb); 11481da177e4SLinus Torvalds 11491da177e4SLinus Torvalds /** 11501da177e4SLinus Torvalds * write_inode_now - write an inode to disk 11511da177e4SLinus Torvalds * @inode: inode to write to disk 11521da177e4SLinus Torvalds * @sync: whether the write should be synchronous or not 11531da177e4SLinus Torvalds * 11547f04c26dSAndrea Arcangeli * This function commits an inode to disk immediately if it is dirty. This is 11557f04c26dSAndrea Arcangeli * primarily needed by knfsd. 11567f04c26dSAndrea Arcangeli * 11577f04c26dSAndrea Arcangeli * The caller must either have a ref on the inode or must have set I_WILL_FREE. 11581da177e4SLinus Torvalds */ 11591da177e4SLinus Torvalds int write_inode_now(struct inode *inode, int sync) 11601da177e4SLinus Torvalds { 11611da177e4SLinus Torvalds int ret; 11621da177e4SLinus Torvalds struct writeback_control wbc = { 11631da177e4SLinus Torvalds .nr_to_write = LONG_MAX, 116418914b18SMike Galbraith .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, 1165111ebb6eSOGAWA Hirofumi .range_start = 0, 1166111ebb6eSOGAWA Hirofumi .range_end = LLONG_MAX, 11671da177e4SLinus Torvalds }; 11681da177e4SLinus Torvalds 11691da177e4SLinus Torvalds if (!mapping_cap_writeback_dirty(inode->i_mapping)) 117049364ce2SAndrew Morton wbc.nr_to_write = 0; 11711da177e4SLinus Torvalds 11721da177e4SLinus Torvalds might_sleep(); 11731da177e4SLinus Torvalds spin_lock(&inode_lock); 117401c03194SChristoph Hellwig ret = writeback_single_inode(inode, &wbc); 11751da177e4SLinus Torvalds spin_unlock(&inode_lock); 11761da177e4SLinus Torvalds if (sync) 11771c0eeaf5SJoern Engel inode_sync_wait(inode); 11781da177e4SLinus Torvalds return ret; 11791da177e4SLinus Torvalds } 11801da177e4SLinus Torvalds EXPORT_SYMBOL(write_inode_now); 11811da177e4SLinus Torvalds 11821da177e4SLinus Torvalds /** 11831da177e4SLinus Torvalds * sync_inode - write an inode and its pages to disk. 11841da177e4SLinus Torvalds * @inode: the inode to sync 11851da177e4SLinus Torvalds * @wbc: controls the writeback mode 11861da177e4SLinus Torvalds * 11871da177e4SLinus Torvalds * sync_inode() will write an inode and its pages to disk. It will also 11881da177e4SLinus Torvalds * correctly update the inode on its superblock's dirty inode lists and will 11891da177e4SLinus Torvalds * update inode->i_state. 11901da177e4SLinus Torvalds * 11911da177e4SLinus Torvalds * The caller must have a ref on the inode. 11921da177e4SLinus Torvalds */ 11931da177e4SLinus Torvalds int sync_inode(struct inode *inode, struct writeback_control *wbc) 11941da177e4SLinus Torvalds { 11951da177e4SLinus Torvalds int ret; 11961da177e4SLinus Torvalds 11971da177e4SLinus Torvalds spin_lock(&inode_lock); 119801c03194SChristoph Hellwig ret = writeback_single_inode(inode, wbc); 11991da177e4SLinus Torvalds spin_unlock(&inode_lock); 12001da177e4SLinus Torvalds return ret; 12011da177e4SLinus Torvalds } 12021da177e4SLinus Torvalds EXPORT_SYMBOL(sync_inode); 1203c3765016SChristoph Hellwig 1204c3765016SChristoph Hellwig /** 1205c3765016SChristoph Hellwig * sync_inode - write an inode to disk 1206c3765016SChristoph Hellwig * @inode: the inode to sync 1207c3765016SChristoph Hellwig * @wait: wait for I/O to complete. 1208c3765016SChristoph Hellwig * 1209c3765016SChristoph Hellwig * Write an inode to disk and adjust it's dirty state after completion. 1210c3765016SChristoph Hellwig * 1211c3765016SChristoph Hellwig * Note: only writes the actual inode, no associated data or other metadata. 1212c3765016SChristoph Hellwig */ 1213c3765016SChristoph Hellwig int sync_inode_metadata(struct inode *inode, int wait) 1214c3765016SChristoph Hellwig { 1215c3765016SChristoph Hellwig struct writeback_control wbc = { 1216c3765016SChristoph Hellwig .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE, 1217c3765016SChristoph Hellwig .nr_to_write = 0, /* metadata-only */ 1218c3765016SChristoph Hellwig }; 1219c3765016SChristoph Hellwig 1220c3765016SChristoph Hellwig return sync_inode(inode, &wbc); 1221c3765016SChristoph Hellwig } 1222c3765016SChristoph Hellwig EXPORT_SYMBOL(sync_inode_metadata); 1223