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 8283ba7b07SChristoph Hellwig static void bdi_queue_work(struct backing_dev_info *bdi, 8383ba7b07SChristoph Hellwig struct wb_writeback_work *work) 844195f73dSNick Piggin { 85455b2864SDave Chinner trace_writeback_queue(bdi, work); 86455b2864SDave Chinner 876467716aSArtem Bityutskiy spin_lock_bh(&bdi->wb_lock); 8883ba7b07SChristoph Hellwig list_add_tail(&work->list, &bdi->work_list); 89fff5b85aSArtem Bityutskiy if (bdi->wb.task) { 90fff5b85aSArtem Bityutskiy wake_up_process(bdi->wb.task); 91fff5b85aSArtem Bityutskiy } else { 921da177e4SLinus Torvalds /* 93fff5b85aSArtem Bityutskiy * The bdi thread isn't there, wake up the forker thread which 94fff5b85aSArtem Bityutskiy * will create and run it. 951da177e4SLinus Torvalds */ 96455b2864SDave Chinner trace_writeback_nothread(bdi, work); 9703ba3782SJens Axboe wake_up_process(default_backing_dev_info.wb.task); 981da177e4SLinus Torvalds } 996467716aSArtem Bityutskiy spin_unlock_bh(&bdi->wb_lock); 10003ba3782SJens Axboe } 1011da177e4SLinus Torvalds 10283ba7b07SChristoph Hellwig static void 10383ba7b07SChristoph Hellwig __bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, 10483ba7b07SChristoph Hellwig bool range_cyclic, bool for_background) 1051da177e4SLinus Torvalds { 10683ba7b07SChristoph Hellwig struct wb_writeback_work *work; 10703ba3782SJens Axboe 108bcddc3f0SJens Axboe /* 109bcddc3f0SJens Axboe * This is WB_SYNC_NONE writeback, so if allocation fails just 110bcddc3f0SJens Axboe * wakeup the thread for old dirty data writeback 111bcddc3f0SJens Axboe */ 11283ba7b07SChristoph Hellwig work = kzalloc(sizeof(*work), GFP_ATOMIC); 11383ba7b07SChristoph Hellwig if (!work) { 114455b2864SDave Chinner if (bdi->wb.task) { 115455b2864SDave Chinner trace_writeback_nowork(bdi); 11683ba7b07SChristoph Hellwig wake_up_process(bdi->wb.task); 117455b2864SDave Chinner } 11883ba7b07SChristoph Hellwig return; 11983ba7b07SChristoph Hellwig } 12083ba7b07SChristoph Hellwig 12183ba7b07SChristoph Hellwig work->sync_mode = WB_SYNC_NONE; 12283ba7b07SChristoph Hellwig work->nr_pages = nr_pages; 12383ba7b07SChristoph Hellwig work->range_cyclic = range_cyclic; 12483ba7b07SChristoph Hellwig work->for_background = for_background; 12583ba7b07SChristoph Hellwig 126f11fcae8SJens Axboe bdi_queue_work(bdi, work); 12703ba3782SJens Axboe } 128b6e51316SJens Axboe 129b6e51316SJens Axboe /** 130b6e51316SJens Axboe * bdi_start_writeback - start writeback 131b6e51316SJens Axboe * @bdi: the backing device to write from 132b6e51316SJens Axboe * @nr_pages: the number of pages to write 133b6e51316SJens Axboe * 134b6e51316SJens Axboe * Description: 135b6e51316SJens Axboe * This does WB_SYNC_NONE opportunistic writeback. The IO is only 136b6e51316SJens Axboe * started when this function returns, we make no guarentees on 1370e3c9a22SJens Axboe * completion. Caller need not hold sb s_umount semaphore. 138b6e51316SJens Axboe * 139b6e51316SJens Axboe */ 140c5444198SChristoph Hellwig void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages) 141b6e51316SJens Axboe { 14283ba7b07SChristoph Hellwig __bdi_start_writeback(bdi, nr_pages, true, false); 143d3ddec76SWu Fengguang } 144d3ddec76SWu Fengguang 145c5444198SChristoph Hellwig /** 146c5444198SChristoph Hellwig * bdi_start_background_writeback - start background writeback 147c5444198SChristoph Hellwig * @bdi: the backing device to write from 148c5444198SChristoph Hellwig * 149c5444198SChristoph Hellwig * Description: 150c5444198SChristoph Hellwig * This does WB_SYNC_NONE background writeback. The IO is only 151c5444198SChristoph Hellwig * started when this function returns, we make no guarentees on 152c5444198SChristoph Hellwig * completion. Caller need not hold sb s_umount semaphore. 153c5444198SChristoph Hellwig */ 154c5444198SChristoph Hellwig void bdi_start_background_writeback(struct backing_dev_info *bdi) 155c5444198SChristoph Hellwig { 15683ba7b07SChristoph Hellwig __bdi_start_writeback(bdi, LONG_MAX, true, true); 1571da177e4SLinus Torvalds } 1581da177e4SLinus Torvalds 1591da177e4SLinus Torvalds /* 1606610a0bcSAndrew Morton * Redirty an inode: set its when-it-was dirtied timestamp and move it to the 1616610a0bcSAndrew Morton * furthest end of its superblock's dirty-inode list. 1626610a0bcSAndrew Morton * 1636610a0bcSAndrew Morton * Before stamping the inode's ->dirtied_when, we check to see whether it is 16466f3b8e2SJens Axboe * already the most-recently-dirtied inode on the b_dirty list. If that is 1656610a0bcSAndrew Morton * the case then the inode must have been redirtied while it was being written 1666610a0bcSAndrew Morton * out and we don't reset its dirtied_when. 1676610a0bcSAndrew Morton */ 1686610a0bcSAndrew Morton static void redirty_tail(struct inode *inode) 1696610a0bcSAndrew Morton { 17003ba3782SJens Axboe struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; 1716610a0bcSAndrew Morton 17203ba3782SJens Axboe if (!list_empty(&wb->b_dirty)) { 17366f3b8e2SJens Axboe struct inode *tail; 1746610a0bcSAndrew Morton 17503ba3782SJens Axboe tail = list_entry(wb->b_dirty.next, struct inode, i_list); 17666f3b8e2SJens Axboe if (time_before(inode->dirtied_when, tail->dirtied_when)) 1776610a0bcSAndrew Morton inode->dirtied_when = jiffies; 1786610a0bcSAndrew Morton } 17903ba3782SJens Axboe list_move(&inode->i_list, &wb->b_dirty); 1806610a0bcSAndrew Morton } 1816610a0bcSAndrew Morton 1826610a0bcSAndrew Morton /* 18366f3b8e2SJens Axboe * requeue inode for re-scanning after bdi->b_io list is exhausted. 184c986d1e2SAndrew Morton */ 1850e0f4fc2SKen Chen static void requeue_io(struct inode *inode) 186c986d1e2SAndrew Morton { 18703ba3782SJens Axboe struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; 18803ba3782SJens Axboe 18903ba3782SJens Axboe list_move(&inode->i_list, &wb->b_more_io); 190c986d1e2SAndrew Morton } 191c986d1e2SAndrew Morton 1921c0eeaf5SJoern Engel static void inode_sync_complete(struct inode *inode) 1931c0eeaf5SJoern Engel { 1941c0eeaf5SJoern Engel /* 1951c0eeaf5SJoern Engel * Prevent speculative execution through spin_unlock(&inode_lock); 1961c0eeaf5SJoern Engel */ 1971c0eeaf5SJoern Engel smp_mb(); 1981c0eeaf5SJoern Engel wake_up_bit(&inode->i_state, __I_SYNC); 1991c0eeaf5SJoern Engel } 2001c0eeaf5SJoern Engel 201d2caa3c5SJeff Layton static bool inode_dirtied_after(struct inode *inode, unsigned long t) 202d2caa3c5SJeff Layton { 203d2caa3c5SJeff Layton bool ret = time_after(inode->dirtied_when, t); 204d2caa3c5SJeff Layton #ifndef CONFIG_64BIT 205d2caa3c5SJeff Layton /* 206d2caa3c5SJeff Layton * For inodes being constantly redirtied, dirtied_when can get stuck. 207d2caa3c5SJeff Layton * It _appears_ to be in the future, but is actually in distant past. 208d2caa3c5SJeff Layton * This test is necessary to prevent such wrapped-around relative times 2095b0830cbSJens Axboe * from permanently stopping the whole bdi writeback. 210d2caa3c5SJeff Layton */ 211d2caa3c5SJeff Layton ret = ret && time_before_eq(inode->dirtied_when, jiffies); 212d2caa3c5SJeff Layton #endif 213d2caa3c5SJeff Layton return ret; 214d2caa3c5SJeff Layton } 215d2caa3c5SJeff Layton 216c986d1e2SAndrew Morton /* 2172c136579SFengguang Wu * Move expired dirty inodes from @delaying_queue to @dispatch_queue. 2182c136579SFengguang Wu */ 2192c136579SFengguang Wu static void move_expired_inodes(struct list_head *delaying_queue, 2202c136579SFengguang Wu struct list_head *dispatch_queue, 2212c136579SFengguang Wu unsigned long *older_than_this) 2222c136579SFengguang Wu { 2235c03449dSShaohua Li LIST_HEAD(tmp); 2245c03449dSShaohua Li struct list_head *pos, *node; 225cf137307SJens Axboe struct super_block *sb = NULL; 2265c03449dSShaohua Li struct inode *inode; 227cf137307SJens Axboe int do_sb_sort = 0; 2285c03449dSShaohua Li 2292c136579SFengguang Wu while (!list_empty(delaying_queue)) { 2305c03449dSShaohua Li inode = list_entry(delaying_queue->prev, struct inode, i_list); 2312c136579SFengguang Wu if (older_than_this && 232d2caa3c5SJeff Layton inode_dirtied_after(inode, *older_than_this)) 2332c136579SFengguang Wu break; 234cf137307SJens Axboe if (sb && sb != inode->i_sb) 235cf137307SJens Axboe do_sb_sort = 1; 236cf137307SJens Axboe sb = inode->i_sb; 2375c03449dSShaohua Li list_move(&inode->i_list, &tmp); 2385c03449dSShaohua Li } 2395c03449dSShaohua Li 240cf137307SJens Axboe /* just one sb in list, splice to dispatch_queue and we're done */ 241cf137307SJens Axboe if (!do_sb_sort) { 242cf137307SJens Axboe list_splice(&tmp, dispatch_queue); 243cf137307SJens Axboe return; 244cf137307SJens Axboe } 245cf137307SJens Axboe 2465c03449dSShaohua Li /* Move inodes from one superblock together */ 2475c03449dSShaohua Li while (!list_empty(&tmp)) { 2485c03449dSShaohua Li inode = list_entry(tmp.prev, struct inode, i_list); 2495c03449dSShaohua Li sb = inode->i_sb; 2505c03449dSShaohua Li list_for_each_prev_safe(pos, node, &tmp) { 2515c03449dSShaohua Li inode = list_entry(pos, struct inode, i_list); 2525c03449dSShaohua Li if (inode->i_sb == sb) 2532c136579SFengguang Wu list_move(&inode->i_list, dispatch_queue); 2542c136579SFengguang Wu } 2552c136579SFengguang Wu } 2565c03449dSShaohua Li } 2572c136579SFengguang Wu 2582c136579SFengguang Wu /* 2592c136579SFengguang Wu * Queue all expired dirty inodes for io, eldest first. 2604ea879b9SWu Fengguang * Before 2614ea879b9SWu Fengguang * newly dirtied b_dirty b_io b_more_io 2624ea879b9SWu Fengguang * =============> gf edc BA 2634ea879b9SWu Fengguang * After 2644ea879b9SWu Fengguang * newly dirtied b_dirty b_io b_more_io 2654ea879b9SWu Fengguang * =============> g fBAedc 2664ea879b9SWu Fengguang * | 2674ea879b9SWu Fengguang * +--> dequeue for IO 2682c136579SFengguang Wu */ 26903ba3782SJens Axboe static void queue_io(struct bdi_writeback *wb, unsigned long *older_than_this) 2702c136579SFengguang Wu { 2714ea879b9SWu Fengguang list_splice_init(&wb->b_more_io, &wb->b_io); 27203ba3782SJens Axboe move_expired_inodes(&wb->b_dirty, &wb->b_io, older_than_this); 27366f3b8e2SJens Axboe } 27466f3b8e2SJens Axboe 275a9185b41SChristoph Hellwig static int write_inode(struct inode *inode, struct writeback_control *wbc) 27666f3b8e2SJens Axboe { 27703ba3782SJens Axboe if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) 278a9185b41SChristoph Hellwig return inode->i_sb->s_op->write_inode(inode, wbc); 27903ba3782SJens Axboe return 0; 28066f3b8e2SJens Axboe } 28108d8e974SFengguang Wu 2822c136579SFengguang Wu /* 28301c03194SChristoph Hellwig * Wait for writeback on an inode to complete. 28401c03194SChristoph Hellwig */ 28501c03194SChristoph Hellwig static void inode_wait_for_writeback(struct inode *inode) 28601c03194SChristoph Hellwig { 28701c03194SChristoph Hellwig DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); 28801c03194SChristoph Hellwig wait_queue_head_t *wqh; 28901c03194SChristoph Hellwig 29001c03194SChristoph Hellwig wqh = bit_waitqueue(&inode->i_state, __I_SYNC); 29158a9d3d8SRichard Kennedy while (inode->i_state & I_SYNC) { 29201c03194SChristoph Hellwig spin_unlock(&inode_lock); 29301c03194SChristoph Hellwig __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE); 29401c03194SChristoph Hellwig spin_lock(&inode_lock); 29558a9d3d8SRichard Kennedy } 29601c03194SChristoph Hellwig } 29701c03194SChristoph Hellwig 29801c03194SChristoph Hellwig /* 29901c03194SChristoph Hellwig * Write out an inode's dirty pages. Called under inode_lock. Either the 30001c03194SChristoph Hellwig * caller has ref on the inode (either via __iget or via syscall against an fd) 30101c03194SChristoph Hellwig * or the inode has I_WILL_FREE set (via generic_forget_inode) 30201c03194SChristoph Hellwig * 3031da177e4SLinus Torvalds * If `wait' is set, wait on the writeout. 3041da177e4SLinus Torvalds * 3051da177e4SLinus Torvalds * The whole writeout design is quite complex and fragile. We want to avoid 3061da177e4SLinus Torvalds * starvation of particular inodes when others are being redirtied, prevent 3071da177e4SLinus Torvalds * livelocks, etc. 3081da177e4SLinus Torvalds * 3091da177e4SLinus Torvalds * Called under inode_lock. 3101da177e4SLinus Torvalds */ 3111da177e4SLinus Torvalds static int 31201c03194SChristoph Hellwig writeback_single_inode(struct inode *inode, struct writeback_control *wbc) 3131da177e4SLinus Torvalds { 3141da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 31501c03194SChristoph Hellwig unsigned dirty; 3161da177e4SLinus Torvalds int ret; 3171da177e4SLinus Torvalds 31801c03194SChristoph Hellwig if (!atomic_read(&inode->i_count)) 31901c03194SChristoph Hellwig WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); 32001c03194SChristoph Hellwig else 32101c03194SChristoph Hellwig WARN_ON(inode->i_state & I_WILL_FREE); 32201c03194SChristoph Hellwig 32301c03194SChristoph Hellwig if (inode->i_state & I_SYNC) { 32401c03194SChristoph Hellwig /* 32501c03194SChristoph Hellwig * If this inode is locked for writeback and we are not doing 32666f3b8e2SJens Axboe * writeback-for-data-integrity, move it to b_more_io so that 32701c03194SChristoph Hellwig * writeback can proceed with the other inodes on s_io. 32801c03194SChristoph Hellwig * 32901c03194SChristoph Hellwig * We'll have another go at writing back this inode when we 33066f3b8e2SJens Axboe * completed a full scan of b_io. 33101c03194SChristoph Hellwig */ 332a9185b41SChristoph Hellwig if (wbc->sync_mode != WB_SYNC_ALL) { 33301c03194SChristoph Hellwig requeue_io(inode); 33401c03194SChristoph Hellwig return 0; 33501c03194SChristoph Hellwig } 33601c03194SChristoph Hellwig 33701c03194SChristoph Hellwig /* 33801c03194SChristoph Hellwig * It's a data-integrity sync. We must wait. 33901c03194SChristoph Hellwig */ 34001c03194SChristoph Hellwig inode_wait_for_writeback(inode); 34101c03194SChristoph Hellwig } 34201c03194SChristoph Hellwig 3431c0eeaf5SJoern Engel BUG_ON(inode->i_state & I_SYNC); 3441da177e4SLinus Torvalds 3455547e8aaSDmitry Monakhov /* Set I_SYNC, reset I_DIRTY_PAGES */ 3461c0eeaf5SJoern Engel inode->i_state |= I_SYNC; 3475547e8aaSDmitry Monakhov inode->i_state &= ~I_DIRTY_PAGES; 3481da177e4SLinus Torvalds spin_unlock(&inode_lock); 3491da177e4SLinus Torvalds 3501da177e4SLinus Torvalds ret = do_writepages(mapping, wbc); 3511da177e4SLinus Torvalds 35226821ed4SChristoph Hellwig /* 35326821ed4SChristoph Hellwig * Make sure to wait on the data before writing out the metadata. 35426821ed4SChristoph Hellwig * This is important for filesystems that modify metadata on data 35526821ed4SChristoph Hellwig * I/O completion. 35626821ed4SChristoph Hellwig */ 357a9185b41SChristoph Hellwig if (wbc->sync_mode == WB_SYNC_ALL) { 35826821ed4SChristoph Hellwig int err = filemap_fdatawait(mapping); 3591da177e4SLinus Torvalds if (ret == 0) 3601da177e4SLinus Torvalds ret = err; 3611da177e4SLinus Torvalds } 3621da177e4SLinus Torvalds 3635547e8aaSDmitry Monakhov /* 3645547e8aaSDmitry Monakhov * Some filesystems may redirty the inode during the writeback 3655547e8aaSDmitry Monakhov * due to delalloc, clear dirty metadata flags right before 3665547e8aaSDmitry Monakhov * write_inode() 3675547e8aaSDmitry Monakhov */ 3685547e8aaSDmitry Monakhov spin_lock(&inode_lock); 3695547e8aaSDmitry Monakhov dirty = inode->i_state & I_DIRTY; 3705547e8aaSDmitry Monakhov inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC); 3715547e8aaSDmitry Monakhov spin_unlock(&inode_lock); 37226821ed4SChristoph Hellwig /* Don't write the inode if only I_DIRTY_PAGES was set */ 37326821ed4SChristoph Hellwig if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { 374a9185b41SChristoph Hellwig int err = write_inode(inode, wbc); 3751da177e4SLinus Torvalds if (ret == 0) 3761da177e4SLinus Torvalds ret = err; 3771da177e4SLinus Torvalds } 3781da177e4SLinus Torvalds 3791da177e4SLinus Torvalds spin_lock(&inode_lock); 3801c0eeaf5SJoern Engel inode->i_state &= ~I_SYNC; 381a4ffdde6SAl Viro if (!(inode->i_state & I_FREEING)) { 38223539afcSWu Fengguang if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { 3831da177e4SLinus Torvalds /* 3841da177e4SLinus Torvalds * We didn't write back all the pages. nfs_writepages() 385a50aeb40SWu Fengguang * sometimes bales out without doing anything. 3861da177e4SLinus Torvalds */ 3871da177e4SLinus Torvalds inode->i_state |= I_DIRTY_PAGES; 3888bc3be27SFengguang Wu if (wbc->nr_to_write <= 0) { 3898bc3be27SFengguang Wu /* 3908bc3be27SFengguang Wu * slice used up: queue for next turn 3918bc3be27SFengguang Wu */ 3920e0f4fc2SKen Chen requeue_io(inode); 3931da177e4SLinus Torvalds } else { 3941da177e4SLinus Torvalds /* 395a50aeb40SWu Fengguang * Writeback blocked by something other than 396a50aeb40SWu Fengguang * congestion. Delay the inode for some time to 397a50aeb40SWu Fengguang * avoid spinning on the CPU (100% iowait) 398a50aeb40SWu Fengguang * retrying writeback of the dirty page/inode 399a50aeb40SWu Fengguang * that cannot be performed immediately. 4008bc3be27SFengguang Wu */ 4018bc3be27SFengguang Wu redirty_tail(inode); 4028bc3be27SFengguang Wu } 40323539afcSWu Fengguang } else if (inode->i_state & I_DIRTY) { 40423539afcSWu Fengguang /* 40523539afcSWu Fengguang * Filesystems can dirty the inode during writeback 40623539afcSWu Fengguang * operations, such as delayed allocation during 40723539afcSWu Fengguang * submission or metadata updates after data IO 40823539afcSWu Fengguang * completion. 40923539afcSWu Fengguang */ 41023539afcSWu Fengguang redirty_tail(inode); 4111da177e4SLinus Torvalds } else if (atomic_read(&inode->i_count)) { 4121da177e4SLinus Torvalds /* 4131da177e4SLinus Torvalds * The inode is clean, inuse 4141da177e4SLinus Torvalds */ 4151da177e4SLinus Torvalds list_move(&inode->i_list, &inode_in_use); 4161da177e4SLinus Torvalds } else { 4171da177e4SLinus Torvalds /* 4181da177e4SLinus Torvalds * The inode is clean, unused 4191da177e4SLinus Torvalds */ 4201da177e4SLinus Torvalds list_move(&inode->i_list, &inode_unused); 4211da177e4SLinus Torvalds } 4221da177e4SLinus Torvalds } 4231c0eeaf5SJoern Engel inode_sync_complete(inode); 4241da177e4SLinus Torvalds return ret; 4251da177e4SLinus Torvalds } 4261da177e4SLinus Torvalds 42703ba3782SJens Axboe /* 428d19de7edSChristoph Hellwig * For background writeback the caller does not have the sb pinned 42903ba3782SJens Axboe * before calling writeback. So make sure that we do pin it, so it doesn't 43003ba3782SJens Axboe * go away while we are writing inodes from it. 43103ba3782SJens Axboe */ 432d19de7edSChristoph Hellwig static bool pin_sb_for_writeback(struct super_block *sb) 4331da177e4SLinus Torvalds { 43403ba3782SJens Axboe spin_lock(&sb_lock); 43529cb4859SChristoph Hellwig if (list_empty(&sb->s_instances)) { 43603ba3782SJens Axboe spin_unlock(&sb_lock); 43729cb4859SChristoph Hellwig return false; 43803ba3782SJens Axboe } 43929cb4859SChristoph Hellwig 44029cb4859SChristoph Hellwig sb->s_count++; 44129cb4859SChristoph Hellwig spin_unlock(&sb_lock); 44229cb4859SChristoph Hellwig 44329cb4859SChristoph Hellwig if (down_read_trylock(&sb->s_umount)) { 44429cb4859SChristoph Hellwig if (sb->s_root) 44529cb4859SChristoph Hellwig return true; 44603ba3782SJens Axboe up_read(&sb->s_umount); 44703ba3782SJens Axboe } 44829cb4859SChristoph Hellwig 44929cb4859SChristoph Hellwig put_super(sb); 450d19de7edSChristoph Hellwig return false; 45103ba3782SJens Axboe } 45203ba3782SJens Axboe 453f11c9c5cSEdward Shishkin /* 454f11c9c5cSEdward Shishkin * Write a portion of b_io inodes which belong to @sb. 455edadfb10SChristoph Hellwig * 456edadfb10SChristoph Hellwig * If @only_this_sb is true, then find and write all such 457f11c9c5cSEdward Shishkin * inodes. Otherwise write only ones which go sequentially 458f11c9c5cSEdward Shishkin * in reverse order. 459edadfb10SChristoph Hellwig * 460f11c9c5cSEdward Shishkin * Return 1, if the caller writeback routine should be 461f11c9c5cSEdward Shishkin * interrupted. Otherwise return 0. 462f11c9c5cSEdward Shishkin */ 463edadfb10SChristoph Hellwig static int writeback_sb_inodes(struct super_block *sb, struct bdi_writeback *wb, 464edadfb10SChristoph Hellwig struct writeback_control *wbc, bool only_this_sb) 46503ba3782SJens Axboe { 46603ba3782SJens Axboe while (!list_empty(&wb->b_io)) { 467f11c9c5cSEdward Shishkin long pages_skipped; 46803ba3782SJens Axboe struct inode *inode = list_entry(wb->b_io.prev, 4691da177e4SLinus Torvalds struct inode, i_list); 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 49084a89245SWu Fengguang if (inode->i_state & (I_NEW | I_WILL_FREE)) { 4917ef0d737SNick Piggin requeue_io(inode); 4927ef0d737SNick Piggin continue; 4937ef0d737SNick Piggin } 494d2caa3c5SJeff Layton /* 495d2caa3c5SJeff Layton * Was this inode dirtied after sync_sb_inodes was called? 496d2caa3c5SJeff Layton * This keeps sync from extra jobs and livelock. 497d2caa3c5SJeff Layton */ 498f11c9c5cSEdward Shishkin if (inode_dirtied_after(inode, wbc->wb_start)) 499f11c9c5cSEdward Shishkin return 1; 5001da177e4SLinus Torvalds 501a4ffdde6SAl Viro BUG_ON(inode->i_state & I_FREEING); 5021da177e4SLinus Torvalds __iget(inode); 5031da177e4SLinus Torvalds pages_skipped = wbc->pages_skipped; 50401c03194SChristoph Hellwig writeback_single_inode(inode, wbc); 5051da177e4SLinus Torvalds if (wbc->pages_skipped != pages_skipped) { 5061da177e4SLinus Torvalds /* 5071da177e4SLinus Torvalds * writeback is not making progress due to locked 5081da177e4SLinus Torvalds * buffers. Skip this inode for now. 5091da177e4SLinus Torvalds */ 510f57b9b7bSAndrew Morton redirty_tail(inode); 5111da177e4SLinus Torvalds } 5121da177e4SLinus Torvalds spin_unlock(&inode_lock); 5131da177e4SLinus Torvalds iput(inode); 5144ffc8444SOGAWA Hirofumi cond_resched(); 5151da177e4SLinus Torvalds spin_lock(&inode_lock); 5168bc3be27SFengguang Wu if (wbc->nr_to_write <= 0) { 5178bc3be27SFengguang Wu wbc->more_io = 1; 518f11c9c5cSEdward Shishkin return 1; 5191da177e4SLinus Torvalds } 52003ba3782SJens Axboe if (!list_empty(&wb->b_more_io)) 5218bc3be27SFengguang Wu wbc->more_io = 1; 5228bc3be27SFengguang Wu } 523f11c9c5cSEdward Shishkin /* b_io is empty */ 524f11c9c5cSEdward Shishkin return 1; 525f11c9c5cSEdward Shishkin } 52638f21977SNick Piggin 5279c3a8ee8SChristoph Hellwig void writeback_inodes_wb(struct bdi_writeback *wb, 528f11c9c5cSEdward Shishkin struct writeback_control *wbc) 529f11c9c5cSEdward Shishkin { 530f11c9c5cSEdward Shishkin int ret = 0; 5319ecc2738SJens Axboe 5327624ee72SJan Kara if (!wbc->wb_start) 533f11c9c5cSEdward Shishkin wbc->wb_start = jiffies; /* livelock avoidance */ 534f11c9c5cSEdward Shishkin spin_lock(&inode_lock); 535f11c9c5cSEdward Shishkin if (!wbc->for_kupdate || list_empty(&wb->b_io)) 536f11c9c5cSEdward Shishkin queue_io(wb, wbc->older_than_this); 537f11c9c5cSEdward Shishkin 538f11c9c5cSEdward Shishkin while (!list_empty(&wb->b_io)) { 539f11c9c5cSEdward Shishkin struct inode *inode = list_entry(wb->b_io.prev, 540f11c9c5cSEdward Shishkin struct inode, i_list); 541f11c9c5cSEdward Shishkin struct super_block *sb = inode->i_sb; 542f11c9c5cSEdward Shishkin 543334132aeSChristoph Hellwig if (!pin_sb_for_writeback(sb)) { 544334132aeSChristoph Hellwig requeue_io(inode); 545d19de7edSChristoph Hellwig continue; 546334132aeSChristoph Hellwig } 547edadfb10SChristoph Hellwig ret = writeback_sb_inodes(sb, wb, wbc, false); 548d19de7edSChristoph Hellwig drop_super(sb); 549f11c9c5cSEdward Shishkin 550f11c9c5cSEdward Shishkin if (ret) 551f11c9c5cSEdward Shishkin break; 552f11c9c5cSEdward Shishkin } 55366f3b8e2SJens Axboe spin_unlock(&inode_lock); 55466f3b8e2SJens Axboe /* Leave any unwritten inodes on b_io */ 55566f3b8e2SJens Axboe } 55666f3b8e2SJens Axboe 557edadfb10SChristoph Hellwig static void __writeback_inodes_sb(struct super_block *sb, 558edadfb10SChristoph Hellwig struct bdi_writeback *wb, struct writeback_control *wbc) 559edadfb10SChristoph Hellwig { 560edadfb10SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 561edadfb10SChristoph Hellwig 562edadfb10SChristoph Hellwig spin_lock(&inode_lock); 563edadfb10SChristoph Hellwig if (!wbc->for_kupdate || list_empty(&wb->b_io)) 564edadfb10SChristoph Hellwig queue_io(wb, wbc->older_than_this); 565edadfb10SChristoph Hellwig writeback_sb_inodes(sb, wb, wbc, true); 566edadfb10SChristoph Hellwig spin_unlock(&inode_lock); 567edadfb10SChristoph Hellwig } 568edadfb10SChristoph Hellwig 56903ba3782SJens Axboe /* 57003ba3782SJens Axboe * The maximum number of pages to writeout in a single bdi flush/kupdate 57103ba3782SJens Axboe * operation. We do this so we don't hold I_SYNC against an inode for 57203ba3782SJens Axboe * enormous amounts of time, which would block a userspace task which has 57303ba3782SJens Axboe * been forced to throttle against that inode. Also, the code reevaluates 57403ba3782SJens Axboe * the dirty each time it has written this many pages. 57503ba3782SJens Axboe */ 57603ba3782SJens Axboe #define MAX_WRITEBACK_PAGES 1024 57703ba3782SJens Axboe 57803ba3782SJens Axboe static inline bool over_bground_thresh(void) 57903ba3782SJens Axboe { 58003ba3782SJens Axboe unsigned long background_thresh, dirty_thresh; 58103ba3782SJens Axboe 58216c4042fSWu Fengguang global_dirty_limits(&background_thresh, &dirty_thresh); 58303ba3782SJens Axboe 58403ba3782SJens Axboe return (global_page_state(NR_FILE_DIRTY) + 58503ba3782SJens Axboe global_page_state(NR_UNSTABLE_NFS) >= background_thresh); 58603ba3782SJens Axboe } 58703ba3782SJens Axboe 58803ba3782SJens Axboe /* 58903ba3782SJens Axboe * Explicit flushing or periodic writeback of "old" data. 59003ba3782SJens Axboe * 59103ba3782SJens Axboe * Define "old": the first time one of an inode's pages is dirtied, we mark the 59203ba3782SJens Axboe * dirtying-time in the inode's address_space. So this periodic writeback code 59303ba3782SJens Axboe * just walks the superblock inode list, writing back any inodes which are 59403ba3782SJens Axboe * older than a specific point in time. 59503ba3782SJens Axboe * 59603ba3782SJens Axboe * Try to run once per dirty_writeback_interval. But if a writeback event 59703ba3782SJens Axboe * takes longer than a dirty_writeback_interval interval, then leave a 59803ba3782SJens Axboe * one-second gap. 59903ba3782SJens Axboe * 60003ba3782SJens Axboe * older_than_this takes precedence over nr_to_write. So we'll only write back 60103ba3782SJens Axboe * all dirty pages if they are all attached to "old" mappings. 60203ba3782SJens Axboe */ 603c4a77a6cSJens Axboe static long wb_writeback(struct bdi_writeback *wb, 60483ba7b07SChristoph Hellwig struct wb_writeback_work *work) 60503ba3782SJens Axboe { 60603ba3782SJens Axboe struct writeback_control wbc = { 60783ba7b07SChristoph Hellwig .sync_mode = work->sync_mode, 60803ba3782SJens Axboe .older_than_this = NULL, 60983ba7b07SChristoph Hellwig .for_kupdate = work->for_kupdate, 61083ba7b07SChristoph Hellwig .for_background = work->for_background, 61183ba7b07SChristoph Hellwig .range_cyclic = work->range_cyclic, 61203ba3782SJens Axboe }; 61303ba3782SJens Axboe unsigned long oldest_jif; 61403ba3782SJens Axboe long wrote = 0; 615a5989bdcSJan Kara struct inode *inode; 61603ba3782SJens Axboe 61703ba3782SJens Axboe if (wbc.for_kupdate) { 61803ba3782SJens Axboe wbc.older_than_this = &oldest_jif; 61903ba3782SJens Axboe oldest_jif = jiffies - 62003ba3782SJens Axboe msecs_to_jiffies(dirty_expire_interval * 10); 62103ba3782SJens Axboe } 622c4a77a6cSJens Axboe if (!wbc.range_cyclic) { 623c4a77a6cSJens Axboe wbc.range_start = 0; 624c4a77a6cSJens Axboe wbc.range_end = LLONG_MAX; 625c4a77a6cSJens Axboe } 62603ba3782SJens Axboe 6277624ee72SJan Kara wbc.wb_start = jiffies; /* livelock avoidance */ 62803ba3782SJens Axboe for (;;) { 62903ba3782SJens Axboe /* 630d3ddec76SWu Fengguang * Stop writeback when nr_pages has been consumed 63103ba3782SJens Axboe */ 63283ba7b07SChristoph Hellwig if (work->nr_pages <= 0) 63303ba3782SJens Axboe break; 63403ba3782SJens Axboe 63503ba3782SJens Axboe /* 636d3ddec76SWu Fengguang * For background writeout, stop when we are below the 637d3ddec76SWu Fengguang * background dirty threshold 63803ba3782SJens Axboe */ 63983ba7b07SChristoph Hellwig if (work->for_background && !over_bground_thresh()) 64003ba3782SJens Axboe break; 64103ba3782SJens Axboe 64203ba3782SJens Axboe wbc.more_io = 0; 64303ba3782SJens Axboe wbc.nr_to_write = MAX_WRITEBACK_PAGES; 64403ba3782SJens Axboe wbc.pages_skipped = 0; 645028c2dd1SDave Chinner 646028c2dd1SDave Chinner trace_wbc_writeback_start(&wbc, wb->bdi); 64783ba7b07SChristoph Hellwig if (work->sb) 64883ba7b07SChristoph Hellwig __writeback_inodes_sb(work->sb, wb, &wbc); 649edadfb10SChristoph Hellwig else 65003ba3782SJens Axboe writeback_inodes_wb(wb, &wbc); 651028c2dd1SDave Chinner trace_wbc_writeback_written(&wbc, wb->bdi); 652028c2dd1SDave Chinner 65383ba7b07SChristoph Hellwig work->nr_pages -= MAX_WRITEBACK_PAGES - wbc.nr_to_write; 65403ba3782SJens Axboe wrote += MAX_WRITEBACK_PAGES - wbc.nr_to_write; 65503ba3782SJens Axboe 65603ba3782SJens Axboe /* 65771fd05a8SJens Axboe * If we consumed everything, see if we have more 65803ba3782SJens Axboe */ 65971fd05a8SJens Axboe if (wbc.nr_to_write <= 0) 66071fd05a8SJens Axboe continue; 66171fd05a8SJens Axboe /* 66271fd05a8SJens Axboe * Didn't write everything and we don't have more IO, bail 66371fd05a8SJens Axboe */ 66471fd05a8SJens Axboe if (!wbc.more_io) 66571fd05a8SJens Axboe break; 66671fd05a8SJens Axboe /* 66771fd05a8SJens Axboe * Did we write something? Try for more 66871fd05a8SJens Axboe */ 669a5989bdcSJan Kara if (wbc.nr_to_write < MAX_WRITEBACK_PAGES) 67003ba3782SJens Axboe continue; 671a5989bdcSJan Kara /* 672a5989bdcSJan Kara * Nothing written. Wait for some inode to 673a5989bdcSJan Kara * become available for writeback. Otherwise 674a5989bdcSJan Kara * we'll just busyloop. 675a5989bdcSJan Kara */ 676a5989bdcSJan Kara spin_lock(&inode_lock); 677a5989bdcSJan Kara if (!list_empty(&wb->b_more_io)) { 67871fd05a8SJens Axboe inode = list_entry(wb->b_more_io.prev, 679a5989bdcSJan Kara struct inode, i_list); 680028c2dd1SDave Chinner trace_wbc_writeback_wait(&wbc, wb->bdi); 681a5989bdcSJan Kara inode_wait_for_writeback(inode); 682a5989bdcSJan Kara } 683a5989bdcSJan Kara spin_unlock(&inode_lock); 68403ba3782SJens Axboe } 68503ba3782SJens Axboe 68603ba3782SJens Axboe return wrote; 68703ba3782SJens Axboe } 68803ba3782SJens Axboe 68903ba3782SJens Axboe /* 69083ba7b07SChristoph Hellwig * Return the next wb_writeback_work struct that hasn't been processed yet. 69103ba3782SJens Axboe */ 69283ba7b07SChristoph Hellwig static struct wb_writeback_work * 69308852b6dSMinchan Kim get_next_work_item(struct backing_dev_info *bdi) 69403ba3782SJens Axboe { 69583ba7b07SChristoph Hellwig struct wb_writeback_work *work = NULL; 69603ba3782SJens Axboe 6976467716aSArtem Bityutskiy spin_lock_bh(&bdi->wb_lock); 69883ba7b07SChristoph Hellwig if (!list_empty(&bdi->work_list)) { 69983ba7b07SChristoph Hellwig work = list_entry(bdi->work_list.next, 70083ba7b07SChristoph Hellwig struct wb_writeback_work, list); 70183ba7b07SChristoph Hellwig list_del_init(&work->list); 70203ba3782SJens Axboe } 7036467716aSArtem Bityutskiy spin_unlock_bh(&bdi->wb_lock); 70483ba7b07SChristoph Hellwig return work; 70503ba3782SJens Axboe } 70603ba3782SJens Axboe 70703ba3782SJens Axboe static long wb_check_old_data_flush(struct bdi_writeback *wb) 70803ba3782SJens Axboe { 70903ba3782SJens Axboe unsigned long expired; 71003ba3782SJens Axboe long nr_pages; 71103ba3782SJens Axboe 71269b62d01SJens Axboe /* 71369b62d01SJens Axboe * When set to zero, disable periodic writeback 71469b62d01SJens Axboe */ 71569b62d01SJens Axboe if (!dirty_writeback_interval) 71669b62d01SJens Axboe return 0; 71769b62d01SJens Axboe 71803ba3782SJens Axboe expired = wb->last_old_flush + 71903ba3782SJens Axboe msecs_to_jiffies(dirty_writeback_interval * 10); 72003ba3782SJens Axboe if (time_before(jiffies, expired)) 72103ba3782SJens Axboe return 0; 72203ba3782SJens Axboe 72303ba3782SJens Axboe wb->last_old_flush = jiffies; 72403ba3782SJens Axboe nr_pages = global_page_state(NR_FILE_DIRTY) + 72503ba3782SJens Axboe global_page_state(NR_UNSTABLE_NFS) + 726cffbc8aaSDave Chinner get_nr_dirty_inodes(); 72703ba3782SJens Axboe 728c4a77a6cSJens Axboe if (nr_pages) { 72983ba7b07SChristoph Hellwig struct wb_writeback_work work = { 730c4a77a6cSJens Axboe .nr_pages = nr_pages, 731c4a77a6cSJens Axboe .sync_mode = WB_SYNC_NONE, 732c4a77a6cSJens Axboe .for_kupdate = 1, 733c4a77a6cSJens Axboe .range_cyclic = 1, 734c4a77a6cSJens Axboe }; 735c4a77a6cSJens Axboe 73683ba7b07SChristoph Hellwig return wb_writeback(wb, &work); 737c4a77a6cSJens Axboe } 73803ba3782SJens Axboe 73903ba3782SJens Axboe return 0; 74003ba3782SJens Axboe } 74103ba3782SJens Axboe 74203ba3782SJens Axboe /* 74303ba3782SJens Axboe * Retrieve work items and do the writeback they describe 74403ba3782SJens Axboe */ 74503ba3782SJens Axboe long wb_do_writeback(struct bdi_writeback *wb, int force_wait) 74603ba3782SJens Axboe { 74703ba3782SJens Axboe struct backing_dev_info *bdi = wb->bdi; 74883ba7b07SChristoph Hellwig struct wb_writeback_work *work; 749c4a77a6cSJens Axboe long wrote = 0; 75003ba3782SJens Axboe 75181d73a32SJan Kara set_bit(BDI_writeback_running, &wb->bdi->state); 75208852b6dSMinchan Kim while ((work = get_next_work_item(bdi)) != NULL) { 75303ba3782SJens Axboe /* 75403ba3782SJens Axboe * Override sync mode, in case we must wait for completion 75583ba7b07SChristoph Hellwig * because this thread is exiting now. 75603ba3782SJens Axboe */ 75703ba3782SJens Axboe if (force_wait) 75883ba7b07SChristoph Hellwig work->sync_mode = WB_SYNC_ALL; 75983ba7b07SChristoph Hellwig 760455b2864SDave Chinner trace_writeback_exec(bdi, work); 761455b2864SDave Chinner 76283ba7b07SChristoph Hellwig wrote += wb_writeback(wb, work); 76303ba3782SJens Axboe 76403ba3782SJens Axboe /* 76583ba7b07SChristoph Hellwig * Notify the caller of completion if this is a synchronous 76683ba7b07SChristoph Hellwig * work item, otherwise just free it. 76703ba3782SJens Axboe */ 76883ba7b07SChristoph Hellwig if (work->done) 76983ba7b07SChristoph Hellwig complete(work->done); 77083ba7b07SChristoph Hellwig else 77183ba7b07SChristoph Hellwig kfree(work); 77203ba3782SJens Axboe } 77303ba3782SJens Axboe 77403ba3782SJens Axboe /* 77503ba3782SJens Axboe * Check for periodic writeback, kupdated() style 77603ba3782SJens Axboe */ 77703ba3782SJens Axboe wrote += wb_check_old_data_flush(wb); 77881d73a32SJan Kara clear_bit(BDI_writeback_running, &wb->bdi->state); 77903ba3782SJens Axboe 78003ba3782SJens Axboe return wrote; 78103ba3782SJens Axboe } 78203ba3782SJens Axboe 78303ba3782SJens Axboe /* 78403ba3782SJens Axboe * Handle writeback of dirty data for the device backed by this bdi. Also 78503ba3782SJens Axboe * wakes up periodically and does kupdated style flushing. 78603ba3782SJens Axboe */ 78708243900SChristoph Hellwig int bdi_writeback_thread(void *data) 78803ba3782SJens Axboe { 78908243900SChristoph Hellwig struct bdi_writeback *wb = data; 79008243900SChristoph Hellwig struct backing_dev_info *bdi = wb->bdi; 79103ba3782SJens Axboe long pages_written; 79203ba3782SJens Axboe 79308243900SChristoph Hellwig current->flags |= PF_FLUSHER | PF_SWAPWRITE; 79408243900SChristoph Hellwig set_freezable(); 795ecd58403SArtem Bityutskiy wb->last_active = jiffies; 79603ba3782SJens Axboe 79703ba3782SJens Axboe /* 79808243900SChristoph Hellwig * Our parent may run at a different priority, just set us to normal 79903ba3782SJens Axboe */ 80008243900SChristoph Hellwig set_user_nice(current, 0); 80108243900SChristoph Hellwig 802455b2864SDave Chinner trace_writeback_thread_start(bdi); 803455b2864SDave Chinner 80403ba3782SJens Axboe while (!kthread_should_stop()) { 8056467716aSArtem Bityutskiy /* 8066467716aSArtem Bityutskiy * Remove own delayed wake-up timer, since we are already awake 8076467716aSArtem Bityutskiy * and we'll take care of the preriodic write-back. 8086467716aSArtem Bityutskiy */ 8096467716aSArtem Bityutskiy del_timer(&wb->wakeup_timer); 8106467716aSArtem Bityutskiy 81103ba3782SJens Axboe pages_written = wb_do_writeback(wb, 0); 81203ba3782SJens Axboe 813455b2864SDave Chinner trace_writeback_pages_written(pages_written); 814455b2864SDave Chinner 81503ba3782SJens Axboe if (pages_written) 816ecd58403SArtem Bityutskiy wb->last_active = jiffies; 81703ba3782SJens Axboe 818297252c8SArtem Bityutskiy set_current_state(TASK_INTERRUPTIBLE); 819b76b4014SJ. Bruce Fields if (!list_empty(&bdi->work_list) || kthread_should_stop()) { 820297252c8SArtem Bityutskiy __set_current_state(TASK_RUNNING); 821297252c8SArtem Bityutskiy continue; 82203ba3782SJens Axboe } 82303ba3782SJens Axboe 824253c34e9SArtem Bityutskiy if (wb_has_dirty_io(wb) && dirty_writeback_interval) 825fff5b85aSArtem Bityutskiy schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10)); 826253c34e9SArtem Bityutskiy else { 827253c34e9SArtem Bityutskiy /* 828253c34e9SArtem Bityutskiy * We have nothing to do, so can go sleep without any 829253c34e9SArtem Bityutskiy * timeout and save power. When a work is queued or 830253c34e9SArtem Bityutskiy * something is made dirty - we will be woken up. 831253c34e9SArtem Bityutskiy */ 83269b62d01SJens Axboe schedule(); 833f9eadbbdSJens Axboe } 83469b62d01SJens Axboe 83503ba3782SJens Axboe try_to_freeze(); 83603ba3782SJens Axboe } 83703ba3782SJens Axboe 838fff5b85aSArtem Bityutskiy /* Flush any work that raced with us exiting */ 83908243900SChristoph Hellwig if (!list_empty(&bdi->work_list)) 84008243900SChristoph Hellwig wb_do_writeback(wb, 1); 841455b2864SDave Chinner 842455b2864SDave Chinner trace_writeback_thread_stop(bdi); 84303ba3782SJens Axboe return 0; 84403ba3782SJens Axboe } 84503ba3782SJens Axboe 84608243900SChristoph Hellwig 84703ba3782SJens Axboe /* 84803ba3782SJens Axboe * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back 84903ba3782SJens Axboe * the whole world. 85003ba3782SJens Axboe */ 85103ba3782SJens Axboe void wakeup_flusher_threads(long nr_pages) 85203ba3782SJens Axboe { 853b8c2f347SChristoph Hellwig struct backing_dev_info *bdi; 854b8c2f347SChristoph Hellwig 85583ba7b07SChristoph Hellwig if (!nr_pages) { 85683ba7b07SChristoph Hellwig nr_pages = global_page_state(NR_FILE_DIRTY) + 85703ba3782SJens Axboe global_page_state(NR_UNSTABLE_NFS); 858b8c2f347SChristoph Hellwig } 859b8c2f347SChristoph Hellwig 860b8c2f347SChristoph Hellwig rcu_read_lock(); 861b8c2f347SChristoph Hellwig list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { 862b8c2f347SChristoph Hellwig if (!bdi_has_dirty_io(bdi)) 863b8c2f347SChristoph Hellwig continue; 86483ba7b07SChristoph Hellwig __bdi_start_writeback(bdi, nr_pages, false, false); 865b8c2f347SChristoph Hellwig } 866b8c2f347SChristoph Hellwig rcu_read_unlock(); 86703ba3782SJens Axboe } 86803ba3782SJens Axboe 86903ba3782SJens Axboe static noinline void block_dump___mark_inode_dirty(struct inode *inode) 87003ba3782SJens Axboe { 87103ba3782SJens Axboe if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { 87203ba3782SJens Axboe struct dentry *dentry; 87303ba3782SJens Axboe const char *name = "?"; 87403ba3782SJens Axboe 87503ba3782SJens Axboe dentry = d_find_alias(inode); 87603ba3782SJens Axboe if (dentry) { 87703ba3782SJens Axboe spin_lock(&dentry->d_lock); 87803ba3782SJens Axboe name = (const char *) dentry->d_name.name; 87903ba3782SJens Axboe } 88003ba3782SJens Axboe printk(KERN_DEBUG 88103ba3782SJens Axboe "%s(%d): dirtied inode %lu (%s) on %s\n", 88203ba3782SJens Axboe current->comm, task_pid_nr(current), inode->i_ino, 88303ba3782SJens Axboe name, inode->i_sb->s_id); 88403ba3782SJens Axboe if (dentry) { 88503ba3782SJens Axboe spin_unlock(&dentry->d_lock); 88603ba3782SJens Axboe dput(dentry); 88703ba3782SJens Axboe } 88803ba3782SJens Axboe } 88903ba3782SJens Axboe } 89003ba3782SJens Axboe 89103ba3782SJens Axboe /** 89203ba3782SJens Axboe * __mark_inode_dirty - internal function 89303ba3782SJens Axboe * @inode: inode to mark 89403ba3782SJens Axboe * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) 89503ba3782SJens Axboe * Mark an inode as dirty. Callers should use mark_inode_dirty or 89603ba3782SJens Axboe * mark_inode_dirty_sync. 89703ba3782SJens Axboe * 89803ba3782SJens Axboe * Put the inode on the super block's dirty list. 89903ba3782SJens Axboe * 90003ba3782SJens Axboe * CAREFUL! We mark it dirty unconditionally, but move it onto the 90103ba3782SJens Axboe * dirty list only if it is hashed or if it refers to a blockdev. 90203ba3782SJens Axboe * If it was not hashed, it will never be added to the dirty list 90303ba3782SJens Axboe * even if it is later hashed, as it will have been marked dirty already. 90403ba3782SJens Axboe * 90503ba3782SJens Axboe * In short, make sure you hash any inodes _before_ you start marking 90603ba3782SJens Axboe * them dirty. 90703ba3782SJens Axboe * 90803ba3782SJens Axboe * This function *must* be atomic for the I_DIRTY_PAGES case - 90903ba3782SJens Axboe * set_page_dirty() is called under spinlock in several places. 91003ba3782SJens Axboe * 91103ba3782SJens Axboe * Note that for blockdevs, inode->dirtied_when represents the dirtying time of 91203ba3782SJens Axboe * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of 91303ba3782SJens Axboe * the kernel-internal blockdev inode represents the dirtying time of the 91403ba3782SJens Axboe * blockdev's pages. This is why for I_DIRTY_PAGES we always use 91503ba3782SJens Axboe * page->mapping->host, so the page-dirtying time is recorded in the internal 91603ba3782SJens Axboe * blockdev inode. 91703ba3782SJens Axboe */ 91803ba3782SJens Axboe void __mark_inode_dirty(struct inode *inode, int flags) 91903ba3782SJens Axboe { 92003ba3782SJens Axboe struct super_block *sb = inode->i_sb; 921253c34e9SArtem Bityutskiy struct backing_dev_info *bdi = NULL; 922253c34e9SArtem Bityutskiy bool wakeup_bdi = false; 92303ba3782SJens Axboe 92403ba3782SJens Axboe /* 92503ba3782SJens Axboe * Don't do this for I_DIRTY_PAGES - that doesn't actually 92603ba3782SJens Axboe * dirty the inode itself 92703ba3782SJens Axboe */ 92803ba3782SJens Axboe if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { 92903ba3782SJens Axboe if (sb->s_op->dirty_inode) 93003ba3782SJens Axboe sb->s_op->dirty_inode(inode); 93103ba3782SJens Axboe } 93203ba3782SJens Axboe 93303ba3782SJens Axboe /* 93403ba3782SJens Axboe * make sure that changes are seen by all cpus before we test i_state 93503ba3782SJens Axboe * -- mikulas 93603ba3782SJens Axboe */ 93703ba3782SJens Axboe smp_mb(); 93803ba3782SJens Axboe 93903ba3782SJens Axboe /* avoid the locking if we can */ 94003ba3782SJens Axboe if ((inode->i_state & flags) == flags) 94103ba3782SJens Axboe return; 94203ba3782SJens Axboe 94303ba3782SJens Axboe if (unlikely(block_dump)) 94403ba3782SJens Axboe block_dump___mark_inode_dirty(inode); 94503ba3782SJens Axboe 94603ba3782SJens Axboe spin_lock(&inode_lock); 94703ba3782SJens Axboe if ((inode->i_state & flags) != flags) { 94803ba3782SJens Axboe const int was_dirty = inode->i_state & I_DIRTY; 94903ba3782SJens Axboe 95003ba3782SJens Axboe inode->i_state |= flags; 95103ba3782SJens Axboe 95203ba3782SJens Axboe /* 95303ba3782SJens Axboe * If the inode is being synced, just update its dirty state. 95403ba3782SJens Axboe * The unlocker will place the inode on the appropriate 95503ba3782SJens Axboe * superblock list, based upon its state. 95603ba3782SJens Axboe */ 95703ba3782SJens Axboe if (inode->i_state & I_SYNC) 95803ba3782SJens Axboe goto out; 95903ba3782SJens Axboe 96003ba3782SJens Axboe /* 96103ba3782SJens Axboe * Only add valid (hashed) inodes to the superblock's 96203ba3782SJens Axboe * dirty list. Add blockdev inodes as well. 96303ba3782SJens Axboe */ 96403ba3782SJens Axboe if (!S_ISBLK(inode->i_mode)) { 9651d3382cbSAl Viro if (inode_unhashed(inode)) 96603ba3782SJens Axboe goto out; 96703ba3782SJens Axboe } 968a4ffdde6SAl Viro if (inode->i_state & I_FREEING) 96903ba3782SJens Axboe goto out; 97003ba3782SJens Axboe 97103ba3782SJens Axboe /* 97203ba3782SJens Axboe * If the inode was already on b_dirty/b_io/b_more_io, don't 97303ba3782SJens Axboe * reposition it (that would break b_dirty time-ordering). 97403ba3782SJens Axboe */ 97503ba3782SJens Axboe if (!was_dirty) { 976253c34e9SArtem Bityutskiy bdi = inode_to_bdi(inode); 977500b067cSJens Axboe 978253c34e9SArtem Bityutskiy if (bdi_cap_writeback_dirty(bdi)) { 979253c34e9SArtem Bityutskiy WARN(!test_bit(BDI_registered, &bdi->state), 980253c34e9SArtem Bityutskiy "bdi-%s not registered\n", bdi->name); 981253c34e9SArtem Bityutskiy 982253c34e9SArtem Bityutskiy /* 983253c34e9SArtem Bityutskiy * If this is the first dirty inode for this 984253c34e9SArtem Bityutskiy * bdi, we have to wake-up the corresponding 985253c34e9SArtem Bityutskiy * bdi thread to make sure background 986253c34e9SArtem Bityutskiy * write-back happens later. 987253c34e9SArtem Bityutskiy */ 988253c34e9SArtem Bityutskiy if (!wb_has_dirty_io(&bdi->wb)) 989253c34e9SArtem Bityutskiy wakeup_bdi = true; 990500b067cSJens Axboe } 99103ba3782SJens Axboe 99203ba3782SJens Axboe inode->dirtied_when = jiffies; 993253c34e9SArtem Bityutskiy list_move(&inode->i_list, &bdi->wb.b_dirty); 99403ba3782SJens Axboe } 99503ba3782SJens Axboe } 99603ba3782SJens Axboe out: 99703ba3782SJens Axboe spin_unlock(&inode_lock); 998253c34e9SArtem Bityutskiy 999253c34e9SArtem Bityutskiy if (wakeup_bdi) 10006467716aSArtem Bityutskiy bdi_wakeup_thread_delayed(bdi); 100103ba3782SJens Axboe } 100203ba3782SJens Axboe EXPORT_SYMBOL(__mark_inode_dirty); 100303ba3782SJens Axboe 100466f3b8e2SJens Axboe /* 100566f3b8e2SJens Axboe * Write out a superblock's list of dirty inodes. A wait will be performed 100666f3b8e2SJens Axboe * upon no inodes, all inodes or the final one, depending upon sync_mode. 100766f3b8e2SJens Axboe * 100866f3b8e2SJens Axboe * If older_than_this is non-NULL, then only write out inodes which 100966f3b8e2SJens Axboe * had their first dirtying at a time earlier than *older_than_this. 101066f3b8e2SJens Axboe * 101166f3b8e2SJens Axboe * If `bdi' is non-zero then we're being asked to writeback a specific queue. 101266f3b8e2SJens Axboe * This function assumes that the blockdev superblock's inodes are backed by 101366f3b8e2SJens Axboe * a variety of queues, so all inodes are searched. For other superblocks, 101466f3b8e2SJens Axboe * assume that all inodes are backed by the same queue. 101566f3b8e2SJens Axboe * 101666f3b8e2SJens Axboe * The inodes to be written are parked on bdi->b_io. They are moved back onto 101766f3b8e2SJens Axboe * bdi->b_dirty as they are selected for writing. This way, none can be missed 101866f3b8e2SJens Axboe * on the writer throttling path, and we get decent balancing between many 101966f3b8e2SJens Axboe * throttled threads: we don't want them all piling up on inode_sync_wait. 102066f3b8e2SJens Axboe */ 1021b6e51316SJens Axboe static void wait_sb_inodes(struct super_block *sb) 102266f3b8e2SJens Axboe { 102338f21977SNick Piggin struct inode *inode, *old_inode = NULL; 102438f21977SNick Piggin 102503ba3782SJens Axboe /* 102603ba3782SJens Axboe * We need to be protected against the filesystem going from 102703ba3782SJens Axboe * r/o to r/w or vice versa. 102803ba3782SJens Axboe */ 1029b6e51316SJens Axboe WARN_ON(!rwsem_is_locked(&sb->s_umount)); 103003ba3782SJens Axboe 103166f3b8e2SJens Axboe spin_lock(&inode_lock); 103266f3b8e2SJens Axboe 103338f21977SNick Piggin /* 103438f21977SNick Piggin * Data integrity sync. Must wait for all pages under writeback, 103538f21977SNick Piggin * because there may have been pages dirtied before our sync 103638f21977SNick Piggin * call, but which had writeout started before we write it out. 103738f21977SNick Piggin * In which case, the inode may not be on the dirty list, but 103838f21977SNick Piggin * we still have to wait for that writeout. 103938f21977SNick Piggin */ 1040b6e51316SJens Axboe list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { 104138f21977SNick Piggin struct address_space *mapping; 104238f21977SNick Piggin 1043a4ffdde6SAl Viro if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) 104438f21977SNick Piggin continue; 104538f21977SNick Piggin mapping = inode->i_mapping; 104638f21977SNick Piggin if (mapping->nrpages == 0) 104738f21977SNick Piggin continue; 104838f21977SNick Piggin __iget(inode); 1049ae8547b0SHans Reiser spin_unlock(&inode_lock); 105038f21977SNick Piggin /* 105138f21977SNick Piggin * We hold a reference to 'inode' so it couldn't have 105238f21977SNick Piggin * been removed from s_inodes list while we dropped the 105338f21977SNick Piggin * inode_lock. We cannot iput the inode now as we can 105438f21977SNick Piggin * be holding the last reference and we cannot iput it 105538f21977SNick Piggin * under inode_lock. So we keep the reference and iput 105638f21977SNick Piggin * it later. 105738f21977SNick Piggin */ 105838f21977SNick Piggin iput(old_inode); 105938f21977SNick Piggin old_inode = inode; 106038f21977SNick Piggin 106138f21977SNick Piggin filemap_fdatawait(mapping); 106238f21977SNick Piggin 106338f21977SNick Piggin cond_resched(); 106438f21977SNick Piggin 106538f21977SNick Piggin spin_lock(&inode_lock); 106638f21977SNick Piggin } 106738f21977SNick Piggin spin_unlock(&inode_lock); 106838f21977SNick Piggin iput(old_inode); 106966f3b8e2SJens Axboe } 10701da177e4SLinus Torvalds 1071d8a8559cSJens Axboe /** 1072d8a8559cSJens Axboe * writeback_inodes_sb - writeback dirty inodes from given super_block 1073d8a8559cSJens Axboe * @sb: the superblock 10741da177e4SLinus Torvalds * 1075d8a8559cSJens Axboe * Start writeback on some inodes on this super_block. No guarantees are made 1076d8a8559cSJens Axboe * on how many (if any) will be written, and this function does not wait 1077d8a8559cSJens Axboe * for IO completion of submitted IO. The number of pages submitted is 1078d8a8559cSJens Axboe * returned. 10791da177e4SLinus Torvalds */ 1080b6e51316SJens Axboe void writeback_inodes_sb(struct super_block *sb) 10811da177e4SLinus Torvalds { 10820e3c9a22SJens Axboe unsigned long nr_dirty = global_page_state(NR_FILE_DIRTY); 10830e3c9a22SJens Axboe unsigned long nr_unstable = global_page_state(NR_UNSTABLE_NFS); 108483ba7b07SChristoph Hellwig DECLARE_COMPLETION_ONSTACK(done); 108583ba7b07SChristoph Hellwig struct wb_writeback_work work = { 10863c4d7165SChristoph Hellwig .sb = sb, 10873c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_NONE, 108883ba7b07SChristoph Hellwig .done = &done, 10893c4d7165SChristoph Hellwig }; 10900e3c9a22SJens Axboe 1091cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 1092cf37e972SChristoph Hellwig 1093cffbc8aaSDave Chinner work.nr_pages = nr_dirty + nr_unstable + get_nr_dirty_inodes(); 10940e3c9a22SJens Axboe 109583ba7b07SChristoph Hellwig bdi_queue_work(sb->s_bdi, &work); 109683ba7b07SChristoph Hellwig wait_for_completion(&done); 10971da177e4SLinus Torvalds } 1098d8a8559cSJens Axboe EXPORT_SYMBOL(writeback_inodes_sb); 1099d8a8559cSJens Axboe 1100d8a8559cSJens Axboe /** 110117bd55d0SEric Sandeen * writeback_inodes_sb_if_idle - start writeback if none underway 110217bd55d0SEric Sandeen * @sb: the superblock 110317bd55d0SEric Sandeen * 110417bd55d0SEric Sandeen * Invoke writeback_inodes_sb if no writeback is currently underway. 110517bd55d0SEric Sandeen * Returns 1 if writeback was started, 0 if not. 110617bd55d0SEric Sandeen */ 110717bd55d0SEric Sandeen int writeback_inodes_sb_if_idle(struct super_block *sb) 110817bd55d0SEric Sandeen { 110917bd55d0SEric Sandeen if (!writeback_in_progress(sb->s_bdi)) { 1110cf37e972SChristoph Hellwig down_read(&sb->s_umount); 111117bd55d0SEric Sandeen writeback_inodes_sb(sb); 1112cf37e972SChristoph Hellwig up_read(&sb->s_umount); 111317bd55d0SEric Sandeen return 1; 111417bd55d0SEric Sandeen } else 111517bd55d0SEric Sandeen return 0; 111617bd55d0SEric Sandeen } 111717bd55d0SEric Sandeen EXPORT_SYMBOL(writeback_inodes_sb_if_idle); 111817bd55d0SEric Sandeen 111917bd55d0SEric Sandeen /** 1120d8a8559cSJens Axboe * sync_inodes_sb - sync sb inode pages 1121d8a8559cSJens Axboe * @sb: the superblock 1122d8a8559cSJens Axboe * 1123d8a8559cSJens Axboe * This function writes and waits on any dirty inode belonging to this 1124d8a8559cSJens Axboe * super_block. The number of pages synced is returned. 1125d8a8559cSJens Axboe */ 1126b6e51316SJens Axboe void sync_inodes_sb(struct super_block *sb) 1127d8a8559cSJens Axboe { 112883ba7b07SChristoph Hellwig DECLARE_COMPLETION_ONSTACK(done); 112983ba7b07SChristoph Hellwig struct wb_writeback_work work = { 11303c4d7165SChristoph Hellwig .sb = sb, 11313c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_ALL, 11323c4d7165SChristoph Hellwig .nr_pages = LONG_MAX, 11333c4d7165SChristoph Hellwig .range_cyclic = 0, 113483ba7b07SChristoph Hellwig .done = &done, 11353c4d7165SChristoph Hellwig }; 11363c4d7165SChristoph Hellwig 1137cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 1138cf37e972SChristoph Hellwig 113983ba7b07SChristoph Hellwig bdi_queue_work(sb->s_bdi, &work); 114083ba7b07SChristoph Hellwig wait_for_completion(&done); 114183ba7b07SChristoph Hellwig 1142b6e51316SJens Axboe wait_sb_inodes(sb); 1143d8a8559cSJens Axboe } 1144d8a8559cSJens Axboe EXPORT_SYMBOL(sync_inodes_sb); 11451da177e4SLinus Torvalds 11461da177e4SLinus Torvalds /** 11471da177e4SLinus Torvalds * write_inode_now - write an inode to disk 11481da177e4SLinus Torvalds * @inode: inode to write to disk 11491da177e4SLinus Torvalds * @sync: whether the write should be synchronous or not 11501da177e4SLinus Torvalds * 11517f04c26dSAndrea Arcangeli * This function commits an inode to disk immediately if it is dirty. This is 11527f04c26dSAndrea Arcangeli * primarily needed by knfsd. 11537f04c26dSAndrea Arcangeli * 11547f04c26dSAndrea Arcangeli * The caller must either have a ref on the inode or must have set I_WILL_FREE. 11551da177e4SLinus Torvalds */ 11561da177e4SLinus Torvalds int write_inode_now(struct inode *inode, int sync) 11571da177e4SLinus Torvalds { 11581da177e4SLinus Torvalds int ret; 11591da177e4SLinus Torvalds struct writeback_control wbc = { 11601da177e4SLinus Torvalds .nr_to_write = LONG_MAX, 116118914b18SMike Galbraith .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, 1162111ebb6eSOGAWA Hirofumi .range_start = 0, 1163111ebb6eSOGAWA Hirofumi .range_end = LLONG_MAX, 11641da177e4SLinus Torvalds }; 11651da177e4SLinus Torvalds 11661da177e4SLinus Torvalds if (!mapping_cap_writeback_dirty(inode->i_mapping)) 116749364ce2SAndrew Morton wbc.nr_to_write = 0; 11681da177e4SLinus Torvalds 11691da177e4SLinus Torvalds might_sleep(); 11701da177e4SLinus Torvalds spin_lock(&inode_lock); 117101c03194SChristoph Hellwig ret = writeback_single_inode(inode, &wbc); 11721da177e4SLinus Torvalds spin_unlock(&inode_lock); 11731da177e4SLinus Torvalds if (sync) 11741c0eeaf5SJoern Engel inode_sync_wait(inode); 11751da177e4SLinus Torvalds return ret; 11761da177e4SLinus Torvalds } 11771da177e4SLinus Torvalds EXPORT_SYMBOL(write_inode_now); 11781da177e4SLinus Torvalds 11791da177e4SLinus Torvalds /** 11801da177e4SLinus Torvalds * sync_inode - write an inode and its pages to disk. 11811da177e4SLinus Torvalds * @inode: the inode to sync 11821da177e4SLinus Torvalds * @wbc: controls the writeback mode 11831da177e4SLinus Torvalds * 11841da177e4SLinus Torvalds * sync_inode() will write an inode and its pages to disk. It will also 11851da177e4SLinus Torvalds * correctly update the inode on its superblock's dirty inode lists and will 11861da177e4SLinus Torvalds * update inode->i_state. 11871da177e4SLinus Torvalds * 11881da177e4SLinus Torvalds * The caller must have a ref on the inode. 11891da177e4SLinus Torvalds */ 11901da177e4SLinus Torvalds int sync_inode(struct inode *inode, struct writeback_control *wbc) 11911da177e4SLinus Torvalds { 11921da177e4SLinus Torvalds int ret; 11931da177e4SLinus Torvalds 11941da177e4SLinus Torvalds spin_lock(&inode_lock); 119501c03194SChristoph Hellwig ret = writeback_single_inode(inode, wbc); 11961da177e4SLinus Torvalds spin_unlock(&inode_lock); 11971da177e4SLinus Torvalds return ret; 11981da177e4SLinus Torvalds } 11991da177e4SLinus Torvalds EXPORT_SYMBOL(sync_inode); 1200c3765016SChristoph Hellwig 1201c3765016SChristoph Hellwig /** 1202c3765016SChristoph Hellwig * sync_inode - write an inode to disk 1203c3765016SChristoph Hellwig * @inode: the inode to sync 1204c3765016SChristoph Hellwig * @wait: wait for I/O to complete. 1205c3765016SChristoph Hellwig * 1206c3765016SChristoph Hellwig * Write an inode to disk and adjust it's dirty state after completion. 1207c3765016SChristoph Hellwig * 1208c3765016SChristoph Hellwig * Note: only writes the actual inode, no associated data or other metadata. 1209c3765016SChristoph Hellwig */ 1210c3765016SChristoph Hellwig int sync_inode_metadata(struct inode *inode, int wait) 1211c3765016SChristoph Hellwig { 1212c3765016SChristoph Hellwig struct writeback_control wbc = { 1213c3765016SChristoph Hellwig .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE, 1214c3765016SChristoph Hellwig .nr_to_write = 0, /* metadata-only */ 1215c3765016SChristoph Hellwig }; 1216c3765016SChristoph Hellwig 1217c3765016SChristoph Hellwig return sync_inode(inode, &wbc); 1218c3765016SChristoph Hellwig } 1219c3765016SChristoph Hellwig EXPORT_SYMBOL(sync_inode_metadata); 1220