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 876585027aSJan Kara /* Wakeup flusher thread or forker thread to fork it. Requires bdi->wb_lock. */ 886585027aSJan Kara static void bdi_wakeup_flusher(struct backing_dev_info *bdi) 894195f73dSNick Piggin { 90fff5b85aSArtem Bityutskiy if (bdi->wb.task) { 91fff5b85aSArtem Bityutskiy wake_up_process(bdi->wb.task); 92fff5b85aSArtem Bityutskiy } else { 931da177e4SLinus Torvalds /* 94fff5b85aSArtem Bityutskiy * The bdi thread isn't there, wake up the forker thread which 95fff5b85aSArtem Bityutskiy * will create and run it. 961da177e4SLinus Torvalds */ 9703ba3782SJens Axboe wake_up_process(default_backing_dev_info.wb.task); 981da177e4SLinus Torvalds } 996585027aSJan Kara } 1006585027aSJan Kara 1016585027aSJan Kara static void bdi_queue_work(struct backing_dev_info *bdi, 1026585027aSJan Kara struct wb_writeback_work *work) 1036585027aSJan Kara { 1046585027aSJan Kara trace_writeback_queue(bdi, work); 1056585027aSJan Kara 1066585027aSJan Kara spin_lock_bh(&bdi->wb_lock); 1076585027aSJan Kara list_add_tail(&work->list, &bdi->work_list); 1086585027aSJan Kara if (!bdi->wb.task) 1096585027aSJan Kara trace_writeback_nothread(bdi, work); 1106585027aSJan Kara bdi_wakeup_flusher(bdi); 1116467716aSArtem Bityutskiy spin_unlock_bh(&bdi->wb_lock); 11203ba3782SJens Axboe } 1131da177e4SLinus Torvalds 11483ba7b07SChristoph Hellwig static void 11583ba7b07SChristoph Hellwig __bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, 1166585027aSJan Kara bool range_cyclic) 1171da177e4SLinus Torvalds { 11883ba7b07SChristoph Hellwig struct wb_writeback_work *work; 11903ba3782SJens Axboe 120bcddc3f0SJens Axboe /* 121bcddc3f0SJens Axboe * This is WB_SYNC_NONE writeback, so if allocation fails just 122bcddc3f0SJens Axboe * wakeup the thread for old dirty data writeback 123bcddc3f0SJens Axboe */ 12483ba7b07SChristoph Hellwig work = kzalloc(sizeof(*work), GFP_ATOMIC); 12583ba7b07SChristoph Hellwig if (!work) { 126455b2864SDave Chinner if (bdi->wb.task) { 127455b2864SDave Chinner trace_writeback_nowork(bdi); 12883ba7b07SChristoph Hellwig wake_up_process(bdi->wb.task); 129455b2864SDave Chinner } 13083ba7b07SChristoph Hellwig return; 13183ba7b07SChristoph Hellwig } 13283ba7b07SChristoph Hellwig 13383ba7b07SChristoph Hellwig work->sync_mode = WB_SYNC_NONE; 13483ba7b07SChristoph Hellwig work->nr_pages = nr_pages; 13583ba7b07SChristoph Hellwig work->range_cyclic = range_cyclic; 13683ba7b07SChristoph Hellwig 137f11fcae8SJens Axboe bdi_queue_work(bdi, work); 13803ba3782SJens Axboe } 139b6e51316SJens Axboe 140b6e51316SJens Axboe /** 141b6e51316SJens Axboe * bdi_start_writeback - start writeback 142b6e51316SJens Axboe * @bdi: the backing device to write from 143b6e51316SJens Axboe * @nr_pages: the number of pages to write 144b6e51316SJens Axboe * 145b6e51316SJens Axboe * Description: 146b6e51316SJens Axboe * This does WB_SYNC_NONE opportunistic writeback. The IO is only 147b6e51316SJens Axboe * started when this function returns, we make no guarentees on 1480e3c9a22SJens Axboe * completion. Caller need not hold sb s_umount semaphore. 149b6e51316SJens Axboe * 150b6e51316SJens Axboe */ 151c5444198SChristoph Hellwig void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages) 152b6e51316SJens Axboe { 1536585027aSJan Kara __bdi_start_writeback(bdi, nr_pages, true); 154d3ddec76SWu Fengguang } 155d3ddec76SWu Fengguang 156c5444198SChristoph Hellwig /** 157c5444198SChristoph Hellwig * bdi_start_background_writeback - start background writeback 158c5444198SChristoph Hellwig * @bdi: the backing device to write from 159c5444198SChristoph Hellwig * 160c5444198SChristoph Hellwig * Description: 1616585027aSJan Kara * This makes sure WB_SYNC_NONE background writeback happens. When 1626585027aSJan Kara * this function returns, it is only guaranteed that for given BDI 1636585027aSJan Kara * some IO is happening if we are over background dirty threshold. 1646585027aSJan Kara * Caller need not hold sb s_umount semaphore. 165c5444198SChristoph Hellwig */ 166c5444198SChristoph Hellwig void bdi_start_background_writeback(struct backing_dev_info *bdi) 167c5444198SChristoph Hellwig { 1686585027aSJan Kara /* 1696585027aSJan Kara * We just wake up the flusher thread. It will perform background 1706585027aSJan Kara * writeback as soon as there is no other work to do. 1716585027aSJan Kara */ 17271927e84SWu Fengguang trace_writeback_wake_background(bdi); 1736585027aSJan Kara spin_lock_bh(&bdi->wb_lock); 1746585027aSJan Kara bdi_wakeup_flusher(bdi); 1756585027aSJan Kara spin_unlock_bh(&bdi->wb_lock); 1761da177e4SLinus Torvalds } 1771da177e4SLinus Torvalds 1781da177e4SLinus Torvalds /* 1796610a0bcSAndrew Morton * Redirty an inode: set its when-it-was dirtied timestamp and move it to the 1806610a0bcSAndrew Morton * furthest end of its superblock's dirty-inode list. 1816610a0bcSAndrew Morton * 1826610a0bcSAndrew Morton * Before stamping the inode's ->dirtied_when, we check to see whether it is 18366f3b8e2SJens Axboe * already the most-recently-dirtied inode on the b_dirty list. If that is 1846610a0bcSAndrew Morton * the case then the inode must have been redirtied while it was being written 1856610a0bcSAndrew Morton * out and we don't reset its dirtied_when. 1866610a0bcSAndrew Morton */ 1876610a0bcSAndrew Morton static void redirty_tail(struct inode *inode) 1886610a0bcSAndrew Morton { 18903ba3782SJens Axboe struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; 1906610a0bcSAndrew Morton 19103ba3782SJens Axboe if (!list_empty(&wb->b_dirty)) { 19266f3b8e2SJens Axboe struct inode *tail; 1936610a0bcSAndrew Morton 1947ccf19a8SNick Piggin tail = wb_inode(wb->b_dirty.next); 19566f3b8e2SJens Axboe if (time_before(inode->dirtied_when, tail->dirtied_when)) 1966610a0bcSAndrew Morton inode->dirtied_when = jiffies; 1976610a0bcSAndrew Morton } 1987ccf19a8SNick Piggin list_move(&inode->i_wb_list, &wb->b_dirty); 1996610a0bcSAndrew Morton } 2006610a0bcSAndrew Morton 2016610a0bcSAndrew Morton /* 20266f3b8e2SJens Axboe * requeue inode for re-scanning after bdi->b_io list is exhausted. 203c986d1e2SAndrew Morton */ 2040e0f4fc2SKen Chen static void requeue_io(struct inode *inode) 205c986d1e2SAndrew Morton { 20603ba3782SJens Axboe struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; 20703ba3782SJens Axboe 2087ccf19a8SNick Piggin list_move(&inode->i_wb_list, &wb->b_more_io); 209c986d1e2SAndrew Morton } 210c986d1e2SAndrew Morton 2111c0eeaf5SJoern Engel static void inode_sync_complete(struct inode *inode) 2121c0eeaf5SJoern Engel { 2131c0eeaf5SJoern Engel /* 2141c0eeaf5SJoern Engel * Prevent speculative execution through spin_unlock(&inode_lock); 2151c0eeaf5SJoern Engel */ 2161c0eeaf5SJoern Engel smp_mb(); 2171c0eeaf5SJoern Engel wake_up_bit(&inode->i_state, __I_SYNC); 2181c0eeaf5SJoern Engel } 2191c0eeaf5SJoern Engel 220d2caa3c5SJeff Layton static bool inode_dirtied_after(struct inode *inode, unsigned long t) 221d2caa3c5SJeff Layton { 222d2caa3c5SJeff Layton bool ret = time_after(inode->dirtied_when, t); 223d2caa3c5SJeff Layton #ifndef CONFIG_64BIT 224d2caa3c5SJeff Layton /* 225d2caa3c5SJeff Layton * For inodes being constantly redirtied, dirtied_when can get stuck. 226d2caa3c5SJeff Layton * It _appears_ to be in the future, but is actually in distant past. 227d2caa3c5SJeff Layton * This test is necessary to prevent such wrapped-around relative times 2285b0830cbSJens Axboe * from permanently stopping the whole bdi writeback. 229d2caa3c5SJeff Layton */ 230d2caa3c5SJeff Layton ret = ret && time_before_eq(inode->dirtied_when, jiffies); 231d2caa3c5SJeff Layton #endif 232d2caa3c5SJeff Layton return ret; 233d2caa3c5SJeff Layton } 234d2caa3c5SJeff Layton 235c986d1e2SAndrew Morton /* 2362c136579SFengguang Wu * Move expired dirty inodes from @delaying_queue to @dispatch_queue. 2372c136579SFengguang Wu */ 2382c136579SFengguang Wu static void move_expired_inodes(struct list_head *delaying_queue, 2392c136579SFengguang Wu struct list_head *dispatch_queue, 2402c136579SFengguang Wu unsigned long *older_than_this) 2412c136579SFengguang Wu { 2425c03449dSShaohua Li LIST_HEAD(tmp); 2435c03449dSShaohua Li struct list_head *pos, *node; 244cf137307SJens Axboe struct super_block *sb = NULL; 2455c03449dSShaohua Li struct inode *inode; 246cf137307SJens Axboe int do_sb_sort = 0; 2475c03449dSShaohua Li 2482c136579SFengguang Wu while (!list_empty(delaying_queue)) { 2497ccf19a8SNick Piggin inode = wb_inode(delaying_queue->prev); 2502c136579SFengguang Wu if (older_than_this && 251d2caa3c5SJeff Layton inode_dirtied_after(inode, *older_than_this)) 2522c136579SFengguang Wu break; 253cf137307SJens Axboe if (sb && sb != inode->i_sb) 254cf137307SJens Axboe do_sb_sort = 1; 255cf137307SJens Axboe sb = inode->i_sb; 2567ccf19a8SNick Piggin list_move(&inode->i_wb_list, &tmp); 2575c03449dSShaohua Li } 2585c03449dSShaohua Li 259cf137307SJens Axboe /* just one sb in list, splice to dispatch_queue and we're done */ 260cf137307SJens Axboe if (!do_sb_sort) { 261cf137307SJens Axboe list_splice(&tmp, dispatch_queue); 262cf137307SJens Axboe return; 263cf137307SJens Axboe } 264cf137307SJens Axboe 2655c03449dSShaohua Li /* Move inodes from one superblock together */ 2665c03449dSShaohua Li while (!list_empty(&tmp)) { 2677ccf19a8SNick Piggin sb = wb_inode(tmp.prev)->i_sb; 2685c03449dSShaohua Li list_for_each_prev_safe(pos, node, &tmp) { 2697ccf19a8SNick Piggin inode = wb_inode(pos); 2705c03449dSShaohua Li if (inode->i_sb == sb) 2717ccf19a8SNick Piggin list_move(&inode->i_wb_list, dispatch_queue); 2722c136579SFengguang Wu } 2732c136579SFengguang Wu } 2745c03449dSShaohua Li } 2752c136579SFengguang Wu 2762c136579SFengguang Wu /* 2772c136579SFengguang Wu * Queue all expired dirty inodes for io, eldest first. 2784ea879b9SWu Fengguang * Before 2794ea879b9SWu Fengguang * newly dirtied b_dirty b_io b_more_io 2804ea879b9SWu Fengguang * =============> gf edc BA 2814ea879b9SWu Fengguang * After 2824ea879b9SWu Fengguang * newly dirtied b_dirty b_io b_more_io 2834ea879b9SWu Fengguang * =============> g fBAedc 2844ea879b9SWu Fengguang * | 2854ea879b9SWu Fengguang * +--> dequeue for IO 2862c136579SFengguang Wu */ 28703ba3782SJens Axboe static void queue_io(struct bdi_writeback *wb, unsigned long *older_than_this) 2882c136579SFengguang Wu { 2894ea879b9SWu Fengguang list_splice_init(&wb->b_more_io, &wb->b_io); 29003ba3782SJens Axboe move_expired_inodes(&wb->b_dirty, &wb->b_io, older_than_this); 29166f3b8e2SJens Axboe } 29266f3b8e2SJens Axboe 293a9185b41SChristoph Hellwig static int write_inode(struct inode *inode, struct writeback_control *wbc) 29466f3b8e2SJens Axboe { 29503ba3782SJens Axboe if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) 296a9185b41SChristoph Hellwig return inode->i_sb->s_op->write_inode(inode, wbc); 29703ba3782SJens Axboe return 0; 29866f3b8e2SJens Axboe } 29908d8e974SFengguang Wu 3002c136579SFengguang Wu /* 30101c03194SChristoph Hellwig * Wait for writeback on an inode to complete. 30201c03194SChristoph Hellwig */ 30301c03194SChristoph Hellwig static void inode_wait_for_writeback(struct inode *inode) 30401c03194SChristoph Hellwig { 30501c03194SChristoph Hellwig DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); 30601c03194SChristoph Hellwig wait_queue_head_t *wqh; 30701c03194SChristoph Hellwig 30801c03194SChristoph Hellwig wqh = bit_waitqueue(&inode->i_state, __I_SYNC); 30958a9d3d8SRichard Kennedy while (inode->i_state & I_SYNC) { 31001c03194SChristoph Hellwig spin_unlock(&inode_lock); 31101c03194SChristoph Hellwig __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE); 31201c03194SChristoph Hellwig spin_lock(&inode_lock); 31358a9d3d8SRichard Kennedy } 31401c03194SChristoph Hellwig } 31501c03194SChristoph Hellwig 31601c03194SChristoph Hellwig /* 31701c03194SChristoph Hellwig * Write out an inode's dirty pages. Called under inode_lock. Either the 31801c03194SChristoph Hellwig * caller has ref on the inode (either via __iget or via syscall against an fd) 31901c03194SChristoph Hellwig * or the inode has I_WILL_FREE set (via generic_forget_inode) 32001c03194SChristoph Hellwig * 3211da177e4SLinus Torvalds * If `wait' is set, wait on the writeout. 3221da177e4SLinus Torvalds * 3231da177e4SLinus Torvalds * The whole writeout design is quite complex and fragile. We want to avoid 3241da177e4SLinus Torvalds * starvation of particular inodes when others are being redirtied, prevent 3251da177e4SLinus Torvalds * livelocks, etc. 3261da177e4SLinus Torvalds * 3271da177e4SLinus Torvalds * Called under inode_lock. 3281da177e4SLinus Torvalds */ 3291da177e4SLinus Torvalds static int 33001c03194SChristoph Hellwig writeback_single_inode(struct inode *inode, struct writeback_control *wbc) 3311da177e4SLinus Torvalds { 3321da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 33301c03194SChristoph Hellwig unsigned dirty; 3341da177e4SLinus Torvalds int ret; 3351da177e4SLinus Torvalds 33601c03194SChristoph Hellwig if (!atomic_read(&inode->i_count)) 33701c03194SChristoph Hellwig WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); 33801c03194SChristoph Hellwig else 33901c03194SChristoph Hellwig WARN_ON(inode->i_state & I_WILL_FREE); 34001c03194SChristoph Hellwig 34101c03194SChristoph Hellwig if (inode->i_state & I_SYNC) { 34201c03194SChristoph Hellwig /* 34301c03194SChristoph Hellwig * If this inode is locked for writeback and we are not doing 34466f3b8e2SJens Axboe * writeback-for-data-integrity, move it to b_more_io so that 34501c03194SChristoph Hellwig * writeback can proceed with the other inodes on s_io. 34601c03194SChristoph Hellwig * 34701c03194SChristoph Hellwig * We'll have another go at writing back this inode when we 34866f3b8e2SJens Axboe * completed a full scan of b_io. 34901c03194SChristoph Hellwig */ 350a9185b41SChristoph Hellwig if (wbc->sync_mode != WB_SYNC_ALL) { 35101c03194SChristoph Hellwig requeue_io(inode); 35201c03194SChristoph Hellwig return 0; 35301c03194SChristoph Hellwig } 35401c03194SChristoph Hellwig 35501c03194SChristoph Hellwig /* 35601c03194SChristoph Hellwig * It's a data-integrity sync. We must wait. 35701c03194SChristoph Hellwig */ 35801c03194SChristoph Hellwig inode_wait_for_writeback(inode); 35901c03194SChristoph Hellwig } 36001c03194SChristoph Hellwig 3611c0eeaf5SJoern Engel BUG_ON(inode->i_state & I_SYNC); 3621da177e4SLinus Torvalds 3635547e8aaSDmitry Monakhov /* Set I_SYNC, reset I_DIRTY_PAGES */ 3641c0eeaf5SJoern Engel inode->i_state |= I_SYNC; 3655547e8aaSDmitry Monakhov inode->i_state &= ~I_DIRTY_PAGES; 3661da177e4SLinus Torvalds spin_unlock(&inode_lock); 3671da177e4SLinus Torvalds 3681da177e4SLinus Torvalds ret = do_writepages(mapping, wbc); 3691da177e4SLinus Torvalds 37026821ed4SChristoph Hellwig /* 37126821ed4SChristoph Hellwig * Make sure to wait on the data before writing out the metadata. 37226821ed4SChristoph Hellwig * This is important for filesystems that modify metadata on data 37326821ed4SChristoph Hellwig * I/O completion. 37426821ed4SChristoph Hellwig */ 375a9185b41SChristoph Hellwig if (wbc->sync_mode == WB_SYNC_ALL) { 37626821ed4SChristoph Hellwig int err = filemap_fdatawait(mapping); 3771da177e4SLinus Torvalds if (ret == 0) 3781da177e4SLinus Torvalds ret = err; 3791da177e4SLinus Torvalds } 3801da177e4SLinus Torvalds 3815547e8aaSDmitry Monakhov /* 3825547e8aaSDmitry Monakhov * Some filesystems may redirty the inode during the writeback 3835547e8aaSDmitry Monakhov * due to delalloc, clear dirty metadata flags right before 3845547e8aaSDmitry Monakhov * write_inode() 3855547e8aaSDmitry Monakhov */ 3865547e8aaSDmitry Monakhov spin_lock(&inode_lock); 3875547e8aaSDmitry Monakhov dirty = inode->i_state & I_DIRTY; 3885547e8aaSDmitry Monakhov inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC); 3895547e8aaSDmitry Monakhov spin_unlock(&inode_lock); 39026821ed4SChristoph Hellwig /* Don't write the inode if only I_DIRTY_PAGES was set */ 39126821ed4SChristoph Hellwig if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { 392a9185b41SChristoph Hellwig int err = write_inode(inode, wbc); 3931da177e4SLinus Torvalds if (ret == 0) 3941da177e4SLinus Torvalds ret = err; 3951da177e4SLinus Torvalds } 3961da177e4SLinus Torvalds 3971da177e4SLinus Torvalds spin_lock(&inode_lock); 3981c0eeaf5SJoern Engel inode->i_state &= ~I_SYNC; 399a4ffdde6SAl Viro if (!(inode->i_state & I_FREEING)) { 40023539afcSWu Fengguang if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { 4011da177e4SLinus Torvalds /* 4021da177e4SLinus Torvalds * We didn't write back all the pages. nfs_writepages() 403a50aeb40SWu Fengguang * sometimes bales out without doing anything. 4041da177e4SLinus Torvalds */ 4051da177e4SLinus Torvalds inode->i_state |= I_DIRTY_PAGES; 4068bc3be27SFengguang Wu if (wbc->nr_to_write <= 0) { 4078bc3be27SFengguang Wu /* 4088bc3be27SFengguang Wu * slice used up: queue for next turn 4098bc3be27SFengguang Wu */ 4100e0f4fc2SKen Chen requeue_io(inode); 4111da177e4SLinus Torvalds } else { 4121da177e4SLinus Torvalds /* 413a50aeb40SWu Fengguang * Writeback blocked by something other than 414a50aeb40SWu Fengguang * congestion. Delay the inode for some time to 415a50aeb40SWu Fengguang * avoid spinning on the CPU (100% iowait) 416a50aeb40SWu Fengguang * retrying writeback of the dirty page/inode 417a50aeb40SWu Fengguang * that cannot be performed immediately. 4188bc3be27SFengguang Wu */ 4198bc3be27SFengguang Wu redirty_tail(inode); 4208bc3be27SFengguang Wu } 42123539afcSWu Fengguang } else if (inode->i_state & I_DIRTY) { 42223539afcSWu Fengguang /* 42323539afcSWu Fengguang * Filesystems can dirty the inode during writeback 42423539afcSWu Fengguang * operations, such as delayed allocation during 42523539afcSWu Fengguang * submission or metadata updates after data IO 42623539afcSWu Fengguang * completion. 42723539afcSWu Fengguang */ 42823539afcSWu Fengguang redirty_tail(inode); 4291da177e4SLinus Torvalds } else { 4301da177e4SLinus Torvalds /* 4319e38d86fSNick Piggin * The inode is clean. At this point we either have 4329e38d86fSNick Piggin * a reference to the inode or it's on it's way out. 4339e38d86fSNick Piggin * No need to add it back to the LRU. 4341da177e4SLinus Torvalds */ 4357ccf19a8SNick Piggin list_del_init(&inode->i_wb_list); 4361da177e4SLinus Torvalds } 4371da177e4SLinus Torvalds } 4381c0eeaf5SJoern Engel inode_sync_complete(inode); 4391da177e4SLinus Torvalds return ret; 4401da177e4SLinus Torvalds } 4411da177e4SLinus Torvalds 44203ba3782SJens Axboe /* 443d19de7edSChristoph Hellwig * For background writeback the caller does not have the sb pinned 44403ba3782SJens Axboe * before calling writeback. So make sure that we do pin it, so it doesn't 44503ba3782SJens Axboe * go away while we are writing inodes from it. 44603ba3782SJens Axboe */ 447d19de7edSChristoph Hellwig static bool pin_sb_for_writeback(struct super_block *sb) 4481da177e4SLinus Torvalds { 44903ba3782SJens Axboe spin_lock(&sb_lock); 45029cb4859SChristoph Hellwig if (list_empty(&sb->s_instances)) { 45103ba3782SJens Axboe spin_unlock(&sb_lock); 45229cb4859SChristoph Hellwig return false; 45303ba3782SJens Axboe } 45429cb4859SChristoph Hellwig 45529cb4859SChristoph Hellwig sb->s_count++; 45629cb4859SChristoph Hellwig spin_unlock(&sb_lock); 45729cb4859SChristoph Hellwig 45829cb4859SChristoph Hellwig if (down_read_trylock(&sb->s_umount)) { 45929cb4859SChristoph Hellwig if (sb->s_root) 46029cb4859SChristoph Hellwig return true; 46103ba3782SJens Axboe up_read(&sb->s_umount); 46203ba3782SJens Axboe } 46329cb4859SChristoph Hellwig 46429cb4859SChristoph Hellwig put_super(sb); 465d19de7edSChristoph Hellwig return false; 46603ba3782SJens Axboe } 46703ba3782SJens Axboe 468f11c9c5cSEdward Shishkin /* 469f11c9c5cSEdward Shishkin * Write a portion of b_io inodes which belong to @sb. 470edadfb10SChristoph Hellwig * 471edadfb10SChristoph Hellwig * If @only_this_sb is true, then find and write all such 472f11c9c5cSEdward Shishkin * inodes. Otherwise write only ones which go sequentially 473f11c9c5cSEdward Shishkin * in reverse order. 474edadfb10SChristoph Hellwig * 475f11c9c5cSEdward Shishkin * Return 1, if the caller writeback routine should be 476f11c9c5cSEdward Shishkin * interrupted. Otherwise return 0. 477f11c9c5cSEdward Shishkin */ 478edadfb10SChristoph Hellwig static int writeback_sb_inodes(struct super_block *sb, struct bdi_writeback *wb, 479edadfb10SChristoph Hellwig struct writeback_control *wbc, bool only_this_sb) 48003ba3782SJens Axboe { 48103ba3782SJens Axboe while (!list_empty(&wb->b_io)) { 482f11c9c5cSEdward Shishkin long pages_skipped; 4837ccf19a8SNick Piggin struct inode *inode = wb_inode(wb->b_io.prev); 484edadfb10SChristoph Hellwig 485edadfb10SChristoph Hellwig if (inode->i_sb != sb) { 486edadfb10SChristoph Hellwig if (only_this_sb) { 487edadfb10SChristoph Hellwig /* 488edadfb10SChristoph Hellwig * We only want to write back data for this 489edadfb10SChristoph Hellwig * superblock, move all inodes not belonging 490edadfb10SChristoph Hellwig * to it back onto the dirty list. 491edadfb10SChristoph Hellwig */ 49266f3b8e2SJens Axboe redirty_tail(inode); 49366f3b8e2SJens Axboe continue; 49466f3b8e2SJens Axboe } 495edadfb10SChristoph Hellwig 496edadfb10SChristoph Hellwig /* 497edadfb10SChristoph Hellwig * The inode belongs to a different superblock. 498edadfb10SChristoph Hellwig * Bounce back to the caller to unpin this and 499edadfb10SChristoph Hellwig * pin the next superblock. 500edadfb10SChristoph Hellwig */ 501f11c9c5cSEdward Shishkin return 0; 502edadfb10SChristoph Hellwig } 503edadfb10SChristoph Hellwig 5049843b76aSChristoph Hellwig /* 5059843b76aSChristoph Hellwig * Don't bother with new inodes or inodes beeing freed, first 5069843b76aSChristoph Hellwig * kind does not need peridic writeout yet, and for the latter 5079843b76aSChristoph Hellwig * kind writeout is handled by the freer. 5089843b76aSChristoph Hellwig */ 5099843b76aSChristoph Hellwig if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { 5107ef0d737SNick Piggin requeue_io(inode); 5117ef0d737SNick Piggin continue; 5127ef0d737SNick Piggin } 5139843b76aSChristoph Hellwig 514d2caa3c5SJeff Layton /* 515d2caa3c5SJeff Layton * Was this inode dirtied after sync_sb_inodes was called? 516d2caa3c5SJeff Layton * This keeps sync from extra jobs and livelock. 517d2caa3c5SJeff Layton */ 518f11c9c5cSEdward Shishkin if (inode_dirtied_after(inode, wbc->wb_start)) 519f11c9c5cSEdward Shishkin return 1; 5201da177e4SLinus Torvalds 5211da177e4SLinus Torvalds __iget(inode); 5221da177e4SLinus Torvalds pages_skipped = wbc->pages_skipped; 52301c03194SChristoph Hellwig writeback_single_inode(inode, wbc); 5241da177e4SLinus Torvalds if (wbc->pages_skipped != pages_skipped) { 5251da177e4SLinus Torvalds /* 5261da177e4SLinus Torvalds * writeback is not making progress due to locked 5271da177e4SLinus Torvalds * buffers. Skip this inode for now. 5281da177e4SLinus Torvalds */ 529f57b9b7bSAndrew Morton redirty_tail(inode); 5301da177e4SLinus Torvalds } 5311da177e4SLinus Torvalds spin_unlock(&inode_lock); 5321da177e4SLinus Torvalds iput(inode); 5334ffc8444SOGAWA Hirofumi cond_resched(); 5341da177e4SLinus Torvalds spin_lock(&inode_lock); 5358bc3be27SFengguang Wu if (wbc->nr_to_write <= 0) { 5368bc3be27SFengguang Wu wbc->more_io = 1; 537f11c9c5cSEdward Shishkin return 1; 5381da177e4SLinus Torvalds } 53903ba3782SJens Axboe if (!list_empty(&wb->b_more_io)) 5408bc3be27SFengguang Wu wbc->more_io = 1; 5418bc3be27SFengguang Wu } 542f11c9c5cSEdward Shishkin /* b_io is empty */ 543f11c9c5cSEdward Shishkin return 1; 544f11c9c5cSEdward Shishkin } 54538f21977SNick Piggin 5469c3a8ee8SChristoph Hellwig void writeback_inodes_wb(struct bdi_writeback *wb, 547f11c9c5cSEdward Shishkin struct writeback_control *wbc) 548f11c9c5cSEdward Shishkin { 549f11c9c5cSEdward Shishkin int ret = 0; 5509ecc2738SJens Axboe 5517624ee72SJan Kara if (!wbc->wb_start) 552f11c9c5cSEdward Shishkin wbc->wb_start = jiffies; /* livelock avoidance */ 553f11c9c5cSEdward Shishkin spin_lock(&inode_lock); 554f11c9c5cSEdward Shishkin if (!wbc->for_kupdate || list_empty(&wb->b_io)) 555f11c9c5cSEdward Shishkin queue_io(wb, wbc->older_than_this); 556f11c9c5cSEdward Shishkin 557f11c9c5cSEdward Shishkin while (!list_empty(&wb->b_io)) { 5587ccf19a8SNick Piggin struct inode *inode = wb_inode(wb->b_io.prev); 559f11c9c5cSEdward Shishkin struct super_block *sb = inode->i_sb; 560f11c9c5cSEdward Shishkin 561334132aeSChristoph Hellwig if (!pin_sb_for_writeback(sb)) { 562334132aeSChristoph Hellwig requeue_io(inode); 563d19de7edSChristoph Hellwig continue; 564334132aeSChristoph Hellwig } 565edadfb10SChristoph Hellwig ret = writeback_sb_inodes(sb, wb, wbc, false); 566d19de7edSChristoph Hellwig drop_super(sb); 567f11c9c5cSEdward Shishkin 568f11c9c5cSEdward Shishkin if (ret) 569f11c9c5cSEdward Shishkin break; 570f11c9c5cSEdward Shishkin } 57166f3b8e2SJens Axboe spin_unlock(&inode_lock); 57266f3b8e2SJens Axboe /* Leave any unwritten inodes on b_io */ 57366f3b8e2SJens Axboe } 57466f3b8e2SJens Axboe 575edadfb10SChristoph Hellwig static void __writeback_inodes_sb(struct super_block *sb, 576edadfb10SChristoph Hellwig struct bdi_writeback *wb, struct writeback_control *wbc) 577edadfb10SChristoph Hellwig { 578edadfb10SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 579edadfb10SChristoph Hellwig 580edadfb10SChristoph Hellwig spin_lock(&inode_lock); 581edadfb10SChristoph Hellwig if (!wbc->for_kupdate || list_empty(&wb->b_io)) 582edadfb10SChristoph Hellwig queue_io(wb, wbc->older_than_this); 583edadfb10SChristoph Hellwig writeback_sb_inodes(sb, wb, wbc, true); 584edadfb10SChristoph Hellwig spin_unlock(&inode_lock); 585edadfb10SChristoph Hellwig } 586edadfb10SChristoph Hellwig 58703ba3782SJens Axboe /* 58803ba3782SJens Axboe * The maximum number of pages to writeout in a single bdi flush/kupdate 58903ba3782SJens Axboe * operation. We do this so we don't hold I_SYNC against an inode for 59003ba3782SJens Axboe * enormous amounts of time, which would block a userspace task which has 59103ba3782SJens Axboe * been forced to throttle against that inode. Also, the code reevaluates 59203ba3782SJens Axboe * the dirty each time it has written this many pages. 59303ba3782SJens Axboe */ 59403ba3782SJens Axboe #define MAX_WRITEBACK_PAGES 1024 59503ba3782SJens Axboe 59603ba3782SJens Axboe static inline bool over_bground_thresh(void) 59703ba3782SJens Axboe { 59803ba3782SJens Axboe unsigned long background_thresh, dirty_thresh; 59903ba3782SJens Axboe 60016c4042fSWu Fengguang global_dirty_limits(&background_thresh, &dirty_thresh); 60103ba3782SJens Axboe 60203ba3782SJens Axboe return (global_page_state(NR_FILE_DIRTY) + 6034cbec4c8SWu Fengguang global_page_state(NR_UNSTABLE_NFS) > background_thresh); 60403ba3782SJens Axboe } 60503ba3782SJens Axboe 60603ba3782SJens Axboe /* 60703ba3782SJens Axboe * Explicit flushing or periodic writeback of "old" data. 60803ba3782SJens Axboe * 60903ba3782SJens Axboe * Define "old": the first time one of an inode's pages is dirtied, we mark the 61003ba3782SJens Axboe * dirtying-time in the inode's address_space. So this periodic writeback code 61103ba3782SJens Axboe * just walks the superblock inode list, writing back any inodes which are 61203ba3782SJens Axboe * older than a specific point in time. 61303ba3782SJens Axboe * 61403ba3782SJens Axboe * Try to run once per dirty_writeback_interval. But if a writeback event 61503ba3782SJens Axboe * takes longer than a dirty_writeback_interval interval, then leave a 61603ba3782SJens Axboe * one-second gap. 61703ba3782SJens Axboe * 61803ba3782SJens Axboe * older_than_this takes precedence over nr_to_write. So we'll only write back 61903ba3782SJens Axboe * all dirty pages if they are all attached to "old" mappings. 62003ba3782SJens Axboe */ 621c4a77a6cSJens Axboe static long wb_writeback(struct bdi_writeback *wb, 62283ba7b07SChristoph Hellwig struct wb_writeback_work *work) 62303ba3782SJens Axboe { 62403ba3782SJens Axboe struct writeback_control wbc = { 62583ba7b07SChristoph Hellwig .sync_mode = work->sync_mode, 62603ba3782SJens Axboe .older_than_this = NULL, 62783ba7b07SChristoph Hellwig .for_kupdate = work->for_kupdate, 62883ba7b07SChristoph Hellwig .for_background = work->for_background, 62983ba7b07SChristoph Hellwig .range_cyclic = work->range_cyclic, 63003ba3782SJens Axboe }; 63103ba3782SJens Axboe unsigned long oldest_jif; 63203ba3782SJens Axboe long wrote = 0; 633b9543dacSJan Kara long write_chunk; 634a5989bdcSJan Kara struct inode *inode; 63503ba3782SJens Axboe 63603ba3782SJens Axboe if (wbc.for_kupdate) { 63703ba3782SJens Axboe wbc.older_than_this = &oldest_jif; 63803ba3782SJens Axboe oldest_jif = jiffies - 63903ba3782SJens Axboe msecs_to_jiffies(dirty_expire_interval * 10); 64003ba3782SJens Axboe } 641c4a77a6cSJens Axboe if (!wbc.range_cyclic) { 642c4a77a6cSJens Axboe wbc.range_start = 0; 643c4a77a6cSJens Axboe wbc.range_end = LLONG_MAX; 644c4a77a6cSJens Axboe } 64503ba3782SJens Axboe 646b9543dacSJan Kara /* 647b9543dacSJan Kara * WB_SYNC_ALL mode does livelock avoidance by syncing dirty 648b9543dacSJan Kara * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX 649b9543dacSJan Kara * here avoids calling into writeback_inodes_wb() more than once. 650b9543dacSJan Kara * 651b9543dacSJan Kara * The intended call sequence for WB_SYNC_ALL writeback is: 652b9543dacSJan Kara * 653b9543dacSJan Kara * wb_writeback() 654b9543dacSJan Kara * __writeback_inodes_sb() <== called only once 655b9543dacSJan Kara * write_cache_pages() <== called once for each inode 656b9543dacSJan Kara * (quickly) tag currently dirty pages 657b9543dacSJan Kara * (maybe slowly) sync all tagged pages 658b9543dacSJan Kara */ 659b9543dacSJan Kara if (wbc.sync_mode == WB_SYNC_NONE) 660b9543dacSJan Kara write_chunk = MAX_WRITEBACK_PAGES; 661b9543dacSJan Kara else 662b9543dacSJan Kara write_chunk = LONG_MAX; 663b9543dacSJan Kara 6647624ee72SJan Kara wbc.wb_start = jiffies; /* livelock avoidance */ 66503ba3782SJens Axboe for (;;) { 66603ba3782SJens Axboe /* 667d3ddec76SWu Fengguang * Stop writeback when nr_pages has been consumed 66803ba3782SJens Axboe */ 66983ba7b07SChristoph Hellwig if (work->nr_pages <= 0) 67003ba3782SJens Axboe break; 67103ba3782SJens Axboe 67203ba3782SJens Axboe /* 673aa373cf5SJan Kara * Background writeout and kupdate-style writeback may 674aa373cf5SJan Kara * run forever. Stop them if there is other work to do 675aa373cf5SJan Kara * so that e.g. sync can proceed. They'll be restarted 676aa373cf5SJan Kara * after the other works are all done. 677aa373cf5SJan Kara */ 678aa373cf5SJan Kara if ((work->for_background || work->for_kupdate) && 679aa373cf5SJan Kara !list_empty(&wb->bdi->work_list)) 680aa373cf5SJan Kara break; 681aa373cf5SJan Kara 682aa373cf5SJan Kara /* 683d3ddec76SWu Fengguang * For background writeout, stop when we are below the 684d3ddec76SWu Fengguang * background dirty threshold 68503ba3782SJens Axboe */ 68683ba7b07SChristoph Hellwig if (work->for_background && !over_bground_thresh()) 68703ba3782SJens Axboe break; 68803ba3782SJens Axboe 68903ba3782SJens Axboe wbc.more_io = 0; 690b9543dacSJan Kara wbc.nr_to_write = write_chunk; 69103ba3782SJens Axboe wbc.pages_skipped = 0; 692028c2dd1SDave Chinner 693028c2dd1SDave Chinner trace_wbc_writeback_start(&wbc, wb->bdi); 69483ba7b07SChristoph Hellwig if (work->sb) 69583ba7b07SChristoph Hellwig __writeback_inodes_sb(work->sb, wb, &wbc); 696edadfb10SChristoph Hellwig else 69703ba3782SJens Axboe writeback_inodes_wb(wb, &wbc); 698028c2dd1SDave Chinner trace_wbc_writeback_written(&wbc, wb->bdi); 699028c2dd1SDave Chinner 700b9543dacSJan Kara work->nr_pages -= write_chunk - wbc.nr_to_write; 701b9543dacSJan Kara wrote += write_chunk - wbc.nr_to_write; 70203ba3782SJens Axboe 70303ba3782SJens Axboe /* 70471fd05a8SJens Axboe * If we consumed everything, see if we have more 70503ba3782SJens Axboe */ 70671fd05a8SJens Axboe if (wbc.nr_to_write <= 0) 70771fd05a8SJens Axboe continue; 70871fd05a8SJens Axboe /* 70971fd05a8SJens Axboe * Didn't write everything and we don't have more IO, bail 71071fd05a8SJens Axboe */ 71171fd05a8SJens Axboe if (!wbc.more_io) 71271fd05a8SJens Axboe break; 71371fd05a8SJens Axboe /* 71471fd05a8SJens Axboe * Did we write something? Try for more 71571fd05a8SJens Axboe */ 716b9543dacSJan Kara if (wbc.nr_to_write < write_chunk) 71703ba3782SJens Axboe continue; 718a5989bdcSJan Kara /* 719a5989bdcSJan Kara * Nothing written. Wait for some inode to 720a5989bdcSJan Kara * become available for writeback. Otherwise 721a5989bdcSJan Kara * we'll just busyloop. 722a5989bdcSJan Kara */ 723a5989bdcSJan Kara spin_lock(&inode_lock); 724a5989bdcSJan Kara if (!list_empty(&wb->b_more_io)) { 7257ccf19a8SNick Piggin inode = wb_inode(wb->b_more_io.prev); 726028c2dd1SDave Chinner trace_wbc_writeback_wait(&wbc, wb->bdi); 727a5989bdcSJan Kara inode_wait_for_writeback(inode); 728a5989bdcSJan Kara } 729a5989bdcSJan Kara spin_unlock(&inode_lock); 73003ba3782SJens Axboe } 73103ba3782SJens Axboe 73203ba3782SJens Axboe return wrote; 73303ba3782SJens Axboe } 73403ba3782SJens Axboe 73503ba3782SJens Axboe /* 73683ba7b07SChristoph Hellwig * Return the next wb_writeback_work struct that hasn't been processed yet. 73703ba3782SJens Axboe */ 73883ba7b07SChristoph Hellwig static struct wb_writeback_work * 73908852b6dSMinchan Kim get_next_work_item(struct backing_dev_info *bdi) 74003ba3782SJens Axboe { 74183ba7b07SChristoph Hellwig struct wb_writeback_work *work = NULL; 74203ba3782SJens Axboe 7436467716aSArtem Bityutskiy spin_lock_bh(&bdi->wb_lock); 74483ba7b07SChristoph Hellwig if (!list_empty(&bdi->work_list)) { 74583ba7b07SChristoph Hellwig work = list_entry(bdi->work_list.next, 74683ba7b07SChristoph Hellwig struct wb_writeback_work, list); 74783ba7b07SChristoph Hellwig list_del_init(&work->list); 74803ba3782SJens Axboe } 7496467716aSArtem Bityutskiy spin_unlock_bh(&bdi->wb_lock); 75083ba7b07SChristoph Hellwig return work; 75103ba3782SJens Axboe } 75203ba3782SJens Axboe 753cdf01dd5SLinus Torvalds /* 754cdf01dd5SLinus Torvalds * Add in the number of potentially dirty inodes, because each inode 755cdf01dd5SLinus Torvalds * write can dirty pagecache in the underlying blockdev. 756cdf01dd5SLinus Torvalds */ 757cdf01dd5SLinus Torvalds static unsigned long get_nr_dirty_pages(void) 758cdf01dd5SLinus Torvalds { 759cdf01dd5SLinus Torvalds return global_page_state(NR_FILE_DIRTY) + 760cdf01dd5SLinus Torvalds global_page_state(NR_UNSTABLE_NFS) + 761cdf01dd5SLinus Torvalds get_nr_dirty_inodes(); 762cdf01dd5SLinus Torvalds } 763cdf01dd5SLinus Torvalds 7646585027aSJan Kara static long wb_check_background_flush(struct bdi_writeback *wb) 7656585027aSJan Kara { 7666585027aSJan Kara if (over_bground_thresh()) { 7676585027aSJan Kara 7686585027aSJan Kara struct wb_writeback_work work = { 7696585027aSJan Kara .nr_pages = LONG_MAX, 7706585027aSJan Kara .sync_mode = WB_SYNC_NONE, 7716585027aSJan Kara .for_background = 1, 7726585027aSJan Kara .range_cyclic = 1, 7736585027aSJan Kara }; 7746585027aSJan Kara 7756585027aSJan Kara return wb_writeback(wb, &work); 7766585027aSJan Kara } 7776585027aSJan Kara 7786585027aSJan Kara return 0; 7796585027aSJan Kara } 7806585027aSJan Kara 78103ba3782SJens Axboe static long wb_check_old_data_flush(struct bdi_writeback *wb) 78203ba3782SJens Axboe { 78303ba3782SJens Axboe unsigned long expired; 78403ba3782SJens Axboe long nr_pages; 78503ba3782SJens Axboe 78669b62d01SJens Axboe /* 78769b62d01SJens Axboe * When set to zero, disable periodic writeback 78869b62d01SJens Axboe */ 78969b62d01SJens Axboe if (!dirty_writeback_interval) 79069b62d01SJens Axboe return 0; 79169b62d01SJens Axboe 79203ba3782SJens Axboe expired = wb->last_old_flush + 79303ba3782SJens Axboe msecs_to_jiffies(dirty_writeback_interval * 10); 79403ba3782SJens Axboe if (time_before(jiffies, expired)) 79503ba3782SJens Axboe return 0; 79603ba3782SJens Axboe 79703ba3782SJens Axboe wb->last_old_flush = jiffies; 798cdf01dd5SLinus Torvalds nr_pages = get_nr_dirty_pages(); 79903ba3782SJens Axboe 800c4a77a6cSJens Axboe if (nr_pages) { 80183ba7b07SChristoph Hellwig struct wb_writeback_work work = { 802c4a77a6cSJens Axboe .nr_pages = nr_pages, 803c4a77a6cSJens Axboe .sync_mode = WB_SYNC_NONE, 804c4a77a6cSJens Axboe .for_kupdate = 1, 805c4a77a6cSJens Axboe .range_cyclic = 1, 806c4a77a6cSJens Axboe }; 807c4a77a6cSJens Axboe 80883ba7b07SChristoph Hellwig return wb_writeback(wb, &work); 809c4a77a6cSJens Axboe } 81003ba3782SJens Axboe 81103ba3782SJens Axboe return 0; 81203ba3782SJens Axboe } 81303ba3782SJens Axboe 81403ba3782SJens Axboe /* 81503ba3782SJens Axboe * Retrieve work items and do the writeback they describe 81603ba3782SJens Axboe */ 81703ba3782SJens Axboe long wb_do_writeback(struct bdi_writeback *wb, int force_wait) 81803ba3782SJens Axboe { 81903ba3782SJens Axboe struct backing_dev_info *bdi = wb->bdi; 82083ba7b07SChristoph Hellwig struct wb_writeback_work *work; 821c4a77a6cSJens Axboe long wrote = 0; 82203ba3782SJens Axboe 82381d73a32SJan Kara set_bit(BDI_writeback_running, &wb->bdi->state); 82408852b6dSMinchan Kim while ((work = get_next_work_item(bdi)) != NULL) { 82503ba3782SJens Axboe /* 82603ba3782SJens Axboe * Override sync mode, in case we must wait for completion 82783ba7b07SChristoph Hellwig * because this thread is exiting now. 82803ba3782SJens Axboe */ 82903ba3782SJens Axboe if (force_wait) 83083ba7b07SChristoph Hellwig work->sync_mode = WB_SYNC_ALL; 83183ba7b07SChristoph Hellwig 832455b2864SDave Chinner trace_writeback_exec(bdi, work); 833455b2864SDave Chinner 83483ba7b07SChristoph Hellwig wrote += wb_writeback(wb, work); 83503ba3782SJens Axboe 83603ba3782SJens Axboe /* 83783ba7b07SChristoph Hellwig * Notify the caller of completion if this is a synchronous 83883ba7b07SChristoph Hellwig * work item, otherwise just free it. 83903ba3782SJens Axboe */ 84083ba7b07SChristoph Hellwig if (work->done) 84183ba7b07SChristoph Hellwig complete(work->done); 84283ba7b07SChristoph Hellwig else 84383ba7b07SChristoph Hellwig kfree(work); 84403ba3782SJens Axboe } 84503ba3782SJens Axboe 84603ba3782SJens Axboe /* 84703ba3782SJens Axboe * Check for periodic writeback, kupdated() style 84803ba3782SJens Axboe */ 84903ba3782SJens Axboe wrote += wb_check_old_data_flush(wb); 8506585027aSJan Kara wrote += wb_check_background_flush(wb); 85181d73a32SJan Kara clear_bit(BDI_writeback_running, &wb->bdi->state); 85203ba3782SJens Axboe 85303ba3782SJens Axboe return wrote; 85403ba3782SJens Axboe } 85503ba3782SJens Axboe 85603ba3782SJens Axboe /* 85703ba3782SJens Axboe * Handle writeback of dirty data for the device backed by this bdi. Also 85803ba3782SJens Axboe * wakes up periodically and does kupdated style flushing. 85903ba3782SJens Axboe */ 86008243900SChristoph Hellwig int bdi_writeback_thread(void *data) 86103ba3782SJens Axboe { 86208243900SChristoph Hellwig struct bdi_writeback *wb = data; 86308243900SChristoph Hellwig struct backing_dev_info *bdi = wb->bdi; 86403ba3782SJens Axboe long pages_written; 86503ba3782SJens Axboe 866766f9164SPeter Zijlstra current->flags |= PF_SWAPWRITE; 86708243900SChristoph Hellwig set_freezable(); 868ecd58403SArtem Bityutskiy wb->last_active = jiffies; 86903ba3782SJens Axboe 87003ba3782SJens Axboe /* 87108243900SChristoph Hellwig * Our parent may run at a different priority, just set us to normal 87203ba3782SJens Axboe */ 87308243900SChristoph Hellwig set_user_nice(current, 0); 87408243900SChristoph Hellwig 875455b2864SDave Chinner trace_writeback_thread_start(bdi); 876455b2864SDave Chinner 87703ba3782SJens Axboe while (!kthread_should_stop()) { 8786467716aSArtem Bityutskiy /* 8796467716aSArtem Bityutskiy * Remove own delayed wake-up timer, since we are already awake 8806467716aSArtem Bityutskiy * and we'll take care of the preriodic write-back. 8816467716aSArtem Bityutskiy */ 8826467716aSArtem Bityutskiy del_timer(&wb->wakeup_timer); 8836467716aSArtem Bityutskiy 88403ba3782SJens Axboe pages_written = wb_do_writeback(wb, 0); 88503ba3782SJens Axboe 886455b2864SDave Chinner trace_writeback_pages_written(pages_written); 887455b2864SDave Chinner 88803ba3782SJens Axboe if (pages_written) 889ecd58403SArtem Bityutskiy wb->last_active = jiffies; 89003ba3782SJens Axboe 891297252c8SArtem Bityutskiy set_current_state(TASK_INTERRUPTIBLE); 892b76b4014SJ. Bruce Fields if (!list_empty(&bdi->work_list) || kthread_should_stop()) { 893297252c8SArtem Bityutskiy __set_current_state(TASK_RUNNING); 894297252c8SArtem Bityutskiy continue; 89503ba3782SJens Axboe } 89603ba3782SJens Axboe 897253c34e9SArtem Bityutskiy if (wb_has_dirty_io(wb) && dirty_writeback_interval) 898fff5b85aSArtem Bityutskiy schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10)); 899253c34e9SArtem Bityutskiy else { 900253c34e9SArtem Bityutskiy /* 901253c34e9SArtem Bityutskiy * We have nothing to do, so can go sleep without any 902253c34e9SArtem Bityutskiy * timeout and save power. When a work is queued or 903253c34e9SArtem Bityutskiy * something is made dirty - we will be woken up. 904253c34e9SArtem Bityutskiy */ 90569b62d01SJens Axboe schedule(); 906f9eadbbdSJens Axboe } 90769b62d01SJens Axboe 90803ba3782SJens Axboe try_to_freeze(); 90903ba3782SJens Axboe } 91003ba3782SJens Axboe 911fff5b85aSArtem Bityutskiy /* Flush any work that raced with us exiting */ 91208243900SChristoph Hellwig if (!list_empty(&bdi->work_list)) 91308243900SChristoph Hellwig wb_do_writeback(wb, 1); 914455b2864SDave Chinner 915455b2864SDave Chinner trace_writeback_thread_stop(bdi); 91603ba3782SJens Axboe return 0; 91703ba3782SJens Axboe } 91803ba3782SJens Axboe 91908243900SChristoph Hellwig 92003ba3782SJens Axboe /* 92103ba3782SJens Axboe * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back 92203ba3782SJens Axboe * the whole world. 92303ba3782SJens Axboe */ 92403ba3782SJens Axboe void wakeup_flusher_threads(long nr_pages) 92503ba3782SJens Axboe { 926b8c2f347SChristoph Hellwig struct backing_dev_info *bdi; 927b8c2f347SChristoph Hellwig 92883ba7b07SChristoph Hellwig if (!nr_pages) { 92983ba7b07SChristoph Hellwig nr_pages = global_page_state(NR_FILE_DIRTY) + 93003ba3782SJens Axboe global_page_state(NR_UNSTABLE_NFS); 931b8c2f347SChristoph Hellwig } 932b8c2f347SChristoph Hellwig 933b8c2f347SChristoph Hellwig rcu_read_lock(); 934b8c2f347SChristoph Hellwig list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { 935b8c2f347SChristoph Hellwig if (!bdi_has_dirty_io(bdi)) 936b8c2f347SChristoph Hellwig continue; 9376585027aSJan Kara __bdi_start_writeback(bdi, nr_pages, false); 938b8c2f347SChristoph Hellwig } 939b8c2f347SChristoph Hellwig rcu_read_unlock(); 94003ba3782SJens Axboe } 94103ba3782SJens Axboe 94203ba3782SJens Axboe static noinline void block_dump___mark_inode_dirty(struct inode *inode) 94303ba3782SJens Axboe { 94403ba3782SJens Axboe if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { 94503ba3782SJens Axboe struct dentry *dentry; 94603ba3782SJens Axboe const char *name = "?"; 94703ba3782SJens Axboe 94803ba3782SJens Axboe dentry = d_find_alias(inode); 94903ba3782SJens Axboe if (dentry) { 95003ba3782SJens Axboe spin_lock(&dentry->d_lock); 95103ba3782SJens Axboe name = (const char *) dentry->d_name.name; 95203ba3782SJens Axboe } 95303ba3782SJens Axboe printk(KERN_DEBUG 95403ba3782SJens Axboe "%s(%d): dirtied inode %lu (%s) on %s\n", 95503ba3782SJens Axboe current->comm, task_pid_nr(current), inode->i_ino, 95603ba3782SJens Axboe name, inode->i_sb->s_id); 95703ba3782SJens Axboe if (dentry) { 95803ba3782SJens Axboe spin_unlock(&dentry->d_lock); 95903ba3782SJens Axboe dput(dentry); 96003ba3782SJens Axboe } 96103ba3782SJens Axboe } 96203ba3782SJens Axboe } 96303ba3782SJens Axboe 96403ba3782SJens Axboe /** 96503ba3782SJens Axboe * __mark_inode_dirty - internal function 96603ba3782SJens Axboe * @inode: inode to mark 96703ba3782SJens Axboe * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) 96803ba3782SJens Axboe * Mark an inode as dirty. Callers should use mark_inode_dirty or 96903ba3782SJens Axboe * mark_inode_dirty_sync. 97003ba3782SJens Axboe * 97103ba3782SJens Axboe * Put the inode on the super block's dirty list. 97203ba3782SJens Axboe * 97303ba3782SJens Axboe * CAREFUL! We mark it dirty unconditionally, but move it onto the 97403ba3782SJens Axboe * dirty list only if it is hashed or if it refers to a blockdev. 97503ba3782SJens Axboe * If it was not hashed, it will never be added to the dirty list 97603ba3782SJens Axboe * even if it is later hashed, as it will have been marked dirty already. 97703ba3782SJens Axboe * 97803ba3782SJens Axboe * In short, make sure you hash any inodes _before_ you start marking 97903ba3782SJens Axboe * them dirty. 98003ba3782SJens Axboe * 98103ba3782SJens Axboe * This function *must* be atomic for the I_DIRTY_PAGES case - 98203ba3782SJens Axboe * set_page_dirty() is called under spinlock in several places. 98303ba3782SJens Axboe * 98403ba3782SJens Axboe * Note that for blockdevs, inode->dirtied_when represents the dirtying time of 98503ba3782SJens Axboe * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of 98603ba3782SJens Axboe * the kernel-internal blockdev inode represents the dirtying time of the 98703ba3782SJens Axboe * blockdev's pages. This is why for I_DIRTY_PAGES we always use 98803ba3782SJens Axboe * page->mapping->host, so the page-dirtying time is recorded in the internal 98903ba3782SJens Axboe * blockdev inode. 99003ba3782SJens Axboe */ 99103ba3782SJens Axboe void __mark_inode_dirty(struct inode *inode, int flags) 99203ba3782SJens Axboe { 99303ba3782SJens Axboe struct super_block *sb = inode->i_sb; 994253c34e9SArtem Bityutskiy struct backing_dev_info *bdi = NULL; 995253c34e9SArtem Bityutskiy bool wakeup_bdi = false; 99603ba3782SJens Axboe 99703ba3782SJens Axboe /* 99803ba3782SJens Axboe * Don't do this for I_DIRTY_PAGES - that doesn't actually 99903ba3782SJens Axboe * dirty the inode itself 100003ba3782SJens Axboe */ 100103ba3782SJens Axboe if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { 100203ba3782SJens Axboe if (sb->s_op->dirty_inode) 100303ba3782SJens Axboe sb->s_op->dirty_inode(inode); 100403ba3782SJens Axboe } 100503ba3782SJens Axboe 100603ba3782SJens Axboe /* 100703ba3782SJens Axboe * make sure that changes are seen by all cpus before we test i_state 100803ba3782SJens Axboe * -- mikulas 100903ba3782SJens Axboe */ 101003ba3782SJens Axboe smp_mb(); 101103ba3782SJens Axboe 101203ba3782SJens Axboe /* avoid the locking if we can */ 101303ba3782SJens Axboe if ((inode->i_state & flags) == flags) 101403ba3782SJens Axboe return; 101503ba3782SJens Axboe 101603ba3782SJens Axboe if (unlikely(block_dump)) 101703ba3782SJens Axboe block_dump___mark_inode_dirty(inode); 101803ba3782SJens Axboe 101903ba3782SJens Axboe spin_lock(&inode_lock); 102003ba3782SJens Axboe if ((inode->i_state & flags) != flags) { 102103ba3782SJens Axboe const int was_dirty = inode->i_state & I_DIRTY; 102203ba3782SJens Axboe 102303ba3782SJens Axboe inode->i_state |= flags; 102403ba3782SJens Axboe 102503ba3782SJens Axboe /* 102603ba3782SJens Axboe * If the inode is being synced, just update its dirty state. 102703ba3782SJens Axboe * The unlocker will place the inode on the appropriate 102803ba3782SJens Axboe * superblock list, based upon its state. 102903ba3782SJens Axboe */ 103003ba3782SJens Axboe if (inode->i_state & I_SYNC) 103103ba3782SJens Axboe goto out; 103203ba3782SJens Axboe 103303ba3782SJens Axboe /* 103403ba3782SJens Axboe * Only add valid (hashed) inodes to the superblock's 103503ba3782SJens Axboe * dirty list. Add blockdev inodes as well. 103603ba3782SJens Axboe */ 103703ba3782SJens Axboe if (!S_ISBLK(inode->i_mode)) { 10381d3382cbSAl Viro if (inode_unhashed(inode)) 103903ba3782SJens Axboe goto out; 104003ba3782SJens Axboe } 1041a4ffdde6SAl Viro if (inode->i_state & I_FREEING) 104203ba3782SJens Axboe goto out; 104303ba3782SJens Axboe 104403ba3782SJens Axboe /* 104503ba3782SJens Axboe * If the inode was already on b_dirty/b_io/b_more_io, don't 104603ba3782SJens Axboe * reposition it (that would break b_dirty time-ordering). 104703ba3782SJens Axboe */ 104803ba3782SJens Axboe if (!was_dirty) { 1049253c34e9SArtem Bityutskiy bdi = inode_to_bdi(inode); 1050500b067cSJens Axboe 1051253c34e9SArtem Bityutskiy if (bdi_cap_writeback_dirty(bdi)) { 1052253c34e9SArtem Bityutskiy WARN(!test_bit(BDI_registered, &bdi->state), 1053253c34e9SArtem Bityutskiy "bdi-%s not registered\n", bdi->name); 1054253c34e9SArtem Bityutskiy 1055253c34e9SArtem Bityutskiy /* 1056253c34e9SArtem Bityutskiy * If this is the first dirty inode for this 1057253c34e9SArtem Bityutskiy * bdi, we have to wake-up the corresponding 1058253c34e9SArtem Bityutskiy * bdi thread to make sure background 1059253c34e9SArtem Bityutskiy * write-back happens later. 1060253c34e9SArtem Bityutskiy */ 1061253c34e9SArtem Bityutskiy if (!wb_has_dirty_io(&bdi->wb)) 1062253c34e9SArtem Bityutskiy wakeup_bdi = true; 1063500b067cSJens Axboe } 106403ba3782SJens Axboe 106503ba3782SJens Axboe inode->dirtied_when = jiffies; 10667ccf19a8SNick Piggin list_move(&inode->i_wb_list, &bdi->wb.b_dirty); 106703ba3782SJens Axboe } 106803ba3782SJens Axboe } 106903ba3782SJens Axboe out: 107003ba3782SJens Axboe spin_unlock(&inode_lock); 1071253c34e9SArtem Bityutskiy 1072253c34e9SArtem Bityutskiy if (wakeup_bdi) 10736467716aSArtem Bityutskiy bdi_wakeup_thread_delayed(bdi); 107403ba3782SJens Axboe } 107503ba3782SJens Axboe EXPORT_SYMBOL(__mark_inode_dirty); 107603ba3782SJens Axboe 107766f3b8e2SJens Axboe /* 107866f3b8e2SJens Axboe * Write out a superblock's list of dirty inodes. A wait will be performed 107966f3b8e2SJens Axboe * upon no inodes, all inodes or the final one, depending upon sync_mode. 108066f3b8e2SJens Axboe * 108166f3b8e2SJens Axboe * If older_than_this is non-NULL, then only write out inodes which 108266f3b8e2SJens Axboe * had their first dirtying at a time earlier than *older_than_this. 108366f3b8e2SJens Axboe * 108466f3b8e2SJens Axboe * If `bdi' is non-zero then we're being asked to writeback a specific queue. 108566f3b8e2SJens Axboe * This function assumes that the blockdev superblock's inodes are backed by 108666f3b8e2SJens Axboe * a variety of queues, so all inodes are searched. For other superblocks, 108766f3b8e2SJens Axboe * assume that all inodes are backed by the same queue. 108866f3b8e2SJens Axboe * 108966f3b8e2SJens Axboe * The inodes to be written are parked on bdi->b_io. They are moved back onto 109066f3b8e2SJens Axboe * bdi->b_dirty as they are selected for writing. This way, none can be missed 109166f3b8e2SJens Axboe * on the writer throttling path, and we get decent balancing between many 109266f3b8e2SJens Axboe * throttled threads: we don't want them all piling up on inode_sync_wait. 109366f3b8e2SJens Axboe */ 1094b6e51316SJens Axboe static void wait_sb_inodes(struct super_block *sb) 109566f3b8e2SJens Axboe { 109638f21977SNick Piggin struct inode *inode, *old_inode = NULL; 109738f21977SNick Piggin 109803ba3782SJens Axboe /* 109903ba3782SJens Axboe * We need to be protected against the filesystem going from 110003ba3782SJens Axboe * r/o to r/w or vice versa. 110103ba3782SJens Axboe */ 1102b6e51316SJens Axboe WARN_ON(!rwsem_is_locked(&sb->s_umount)); 110303ba3782SJens Axboe 110466f3b8e2SJens Axboe spin_lock(&inode_lock); 110566f3b8e2SJens Axboe 110638f21977SNick Piggin /* 110738f21977SNick Piggin * Data integrity sync. Must wait for all pages under writeback, 110838f21977SNick Piggin * because there may have been pages dirtied before our sync 110938f21977SNick Piggin * call, but which had writeout started before we write it out. 111038f21977SNick Piggin * In which case, the inode may not be on the dirty list, but 111138f21977SNick Piggin * we still have to wait for that writeout. 111238f21977SNick Piggin */ 1113b6e51316SJens Axboe list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { 111438f21977SNick Piggin struct address_space *mapping; 111538f21977SNick Piggin 1116a4ffdde6SAl Viro if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) 111738f21977SNick Piggin continue; 111838f21977SNick Piggin mapping = inode->i_mapping; 111938f21977SNick Piggin if (mapping->nrpages == 0) 112038f21977SNick Piggin continue; 112138f21977SNick Piggin __iget(inode); 1122ae8547b0SHans Reiser spin_unlock(&inode_lock); 112338f21977SNick Piggin /* 112438f21977SNick Piggin * We hold a reference to 'inode' so it couldn't have 112538f21977SNick Piggin * been removed from s_inodes list while we dropped the 112638f21977SNick Piggin * inode_lock. We cannot iput the inode now as we can 112738f21977SNick Piggin * be holding the last reference and we cannot iput it 112838f21977SNick Piggin * under inode_lock. So we keep the reference and iput 112938f21977SNick Piggin * it later. 113038f21977SNick Piggin */ 113138f21977SNick Piggin iput(old_inode); 113238f21977SNick Piggin old_inode = inode; 113338f21977SNick Piggin 113438f21977SNick Piggin filemap_fdatawait(mapping); 113538f21977SNick Piggin 113638f21977SNick Piggin cond_resched(); 113738f21977SNick Piggin 113838f21977SNick Piggin spin_lock(&inode_lock); 113938f21977SNick Piggin } 114038f21977SNick Piggin spin_unlock(&inode_lock); 114138f21977SNick Piggin iput(old_inode); 114266f3b8e2SJens Axboe } 11431da177e4SLinus Torvalds 1144d8a8559cSJens Axboe /** 11453259f8beSChris Mason * writeback_inodes_sb_nr - writeback dirty inodes from given super_block 1146d8a8559cSJens Axboe * @sb: the superblock 11473259f8beSChris Mason * @nr: the number of pages to write 11481da177e4SLinus Torvalds * 1149d8a8559cSJens Axboe * Start writeback on some inodes on this super_block. No guarantees are made 1150d8a8559cSJens Axboe * on how many (if any) will be written, and this function does not wait 11513259f8beSChris Mason * for IO completion of submitted IO. 11521da177e4SLinus Torvalds */ 11533259f8beSChris Mason void writeback_inodes_sb_nr(struct super_block *sb, unsigned long nr) 11541da177e4SLinus Torvalds { 115583ba7b07SChristoph Hellwig DECLARE_COMPLETION_ONSTACK(done); 115683ba7b07SChristoph Hellwig struct wb_writeback_work work = { 11573c4d7165SChristoph Hellwig .sb = sb, 11583c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_NONE, 115983ba7b07SChristoph Hellwig .done = &done, 11603259f8beSChris Mason .nr_pages = nr, 11613c4d7165SChristoph Hellwig }; 11620e3c9a22SJens Axboe 1163cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 116483ba7b07SChristoph Hellwig bdi_queue_work(sb->s_bdi, &work); 116583ba7b07SChristoph Hellwig wait_for_completion(&done); 11661da177e4SLinus Torvalds } 11673259f8beSChris Mason EXPORT_SYMBOL(writeback_inodes_sb_nr); 11683259f8beSChris Mason 11693259f8beSChris Mason /** 11703259f8beSChris Mason * writeback_inodes_sb - writeback dirty inodes from given super_block 11713259f8beSChris Mason * @sb: the superblock 11723259f8beSChris Mason * 11733259f8beSChris Mason * Start writeback on some inodes on this super_block. No guarantees are made 11743259f8beSChris Mason * on how many (if any) will be written, and this function does not wait 11753259f8beSChris Mason * for IO completion of submitted IO. 11763259f8beSChris Mason */ 11773259f8beSChris Mason void writeback_inodes_sb(struct super_block *sb) 11783259f8beSChris Mason { 1179925d169fSLinus Torvalds return writeback_inodes_sb_nr(sb, get_nr_dirty_pages()); 11803259f8beSChris Mason } 1181d8a8559cSJens Axboe EXPORT_SYMBOL(writeback_inodes_sb); 1182d8a8559cSJens Axboe 1183d8a8559cSJens Axboe /** 118417bd55d0SEric Sandeen * writeback_inodes_sb_if_idle - start writeback if none underway 118517bd55d0SEric Sandeen * @sb: the superblock 118617bd55d0SEric Sandeen * 118717bd55d0SEric Sandeen * Invoke writeback_inodes_sb if no writeback is currently underway. 118817bd55d0SEric Sandeen * Returns 1 if writeback was started, 0 if not. 118917bd55d0SEric Sandeen */ 119017bd55d0SEric Sandeen int writeback_inodes_sb_if_idle(struct super_block *sb) 119117bd55d0SEric Sandeen { 119217bd55d0SEric Sandeen if (!writeback_in_progress(sb->s_bdi)) { 1193cf37e972SChristoph Hellwig down_read(&sb->s_umount); 119417bd55d0SEric Sandeen writeback_inodes_sb(sb); 1195cf37e972SChristoph Hellwig up_read(&sb->s_umount); 119617bd55d0SEric Sandeen return 1; 119717bd55d0SEric Sandeen } else 119817bd55d0SEric Sandeen return 0; 119917bd55d0SEric Sandeen } 120017bd55d0SEric Sandeen EXPORT_SYMBOL(writeback_inodes_sb_if_idle); 120117bd55d0SEric Sandeen 120217bd55d0SEric Sandeen /** 12033259f8beSChris Mason * writeback_inodes_sb_if_idle - start writeback if none underway 12043259f8beSChris Mason * @sb: the superblock 12053259f8beSChris Mason * @nr: the number of pages to write 12063259f8beSChris Mason * 12073259f8beSChris Mason * Invoke writeback_inodes_sb if no writeback is currently underway. 12083259f8beSChris Mason * Returns 1 if writeback was started, 0 if not. 12093259f8beSChris Mason */ 12103259f8beSChris Mason int writeback_inodes_sb_nr_if_idle(struct super_block *sb, 12113259f8beSChris Mason unsigned long nr) 12123259f8beSChris Mason { 12133259f8beSChris Mason if (!writeback_in_progress(sb->s_bdi)) { 12143259f8beSChris Mason down_read(&sb->s_umount); 12153259f8beSChris Mason writeback_inodes_sb_nr(sb, nr); 12163259f8beSChris Mason up_read(&sb->s_umount); 12173259f8beSChris Mason return 1; 12183259f8beSChris Mason } else 12193259f8beSChris Mason return 0; 12203259f8beSChris Mason } 12213259f8beSChris Mason EXPORT_SYMBOL(writeback_inodes_sb_nr_if_idle); 12223259f8beSChris Mason 12233259f8beSChris Mason /** 1224d8a8559cSJens Axboe * sync_inodes_sb - sync sb inode pages 1225d8a8559cSJens Axboe * @sb: the superblock 1226d8a8559cSJens Axboe * 1227d8a8559cSJens Axboe * This function writes and waits on any dirty inode belonging to this 1228d8a8559cSJens Axboe * super_block. The number of pages synced is returned. 1229d8a8559cSJens Axboe */ 1230b6e51316SJens Axboe void sync_inodes_sb(struct super_block *sb) 1231d8a8559cSJens Axboe { 123283ba7b07SChristoph Hellwig DECLARE_COMPLETION_ONSTACK(done); 123383ba7b07SChristoph Hellwig struct wb_writeback_work work = { 12343c4d7165SChristoph Hellwig .sb = sb, 12353c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_ALL, 12363c4d7165SChristoph Hellwig .nr_pages = LONG_MAX, 12373c4d7165SChristoph Hellwig .range_cyclic = 0, 123883ba7b07SChristoph Hellwig .done = &done, 12393c4d7165SChristoph Hellwig }; 12403c4d7165SChristoph Hellwig 1241cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 1242cf37e972SChristoph Hellwig 124383ba7b07SChristoph Hellwig bdi_queue_work(sb->s_bdi, &work); 124483ba7b07SChristoph Hellwig wait_for_completion(&done); 124583ba7b07SChristoph Hellwig 1246b6e51316SJens Axboe wait_sb_inodes(sb); 1247d8a8559cSJens Axboe } 1248d8a8559cSJens Axboe EXPORT_SYMBOL(sync_inodes_sb); 12491da177e4SLinus Torvalds 12501da177e4SLinus Torvalds /** 12511da177e4SLinus Torvalds * write_inode_now - write an inode to disk 12521da177e4SLinus Torvalds * @inode: inode to write to disk 12531da177e4SLinus Torvalds * @sync: whether the write should be synchronous or not 12541da177e4SLinus Torvalds * 12557f04c26dSAndrea Arcangeli * This function commits an inode to disk immediately if it is dirty. This is 12567f04c26dSAndrea Arcangeli * primarily needed by knfsd. 12577f04c26dSAndrea Arcangeli * 12587f04c26dSAndrea Arcangeli * The caller must either have a ref on the inode or must have set I_WILL_FREE. 12591da177e4SLinus Torvalds */ 12601da177e4SLinus Torvalds int write_inode_now(struct inode *inode, int sync) 12611da177e4SLinus Torvalds { 12621da177e4SLinus Torvalds int ret; 12631da177e4SLinus Torvalds struct writeback_control wbc = { 12641da177e4SLinus Torvalds .nr_to_write = LONG_MAX, 126518914b18SMike Galbraith .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, 1266111ebb6eSOGAWA Hirofumi .range_start = 0, 1267111ebb6eSOGAWA Hirofumi .range_end = LLONG_MAX, 12681da177e4SLinus Torvalds }; 12691da177e4SLinus Torvalds 12701da177e4SLinus Torvalds if (!mapping_cap_writeback_dirty(inode->i_mapping)) 127149364ce2SAndrew Morton wbc.nr_to_write = 0; 12721da177e4SLinus Torvalds 12731da177e4SLinus Torvalds might_sleep(); 12741da177e4SLinus Torvalds spin_lock(&inode_lock); 127501c03194SChristoph Hellwig ret = writeback_single_inode(inode, &wbc); 12761da177e4SLinus Torvalds spin_unlock(&inode_lock); 12771da177e4SLinus Torvalds if (sync) 12781c0eeaf5SJoern Engel inode_sync_wait(inode); 12791da177e4SLinus Torvalds return ret; 12801da177e4SLinus Torvalds } 12811da177e4SLinus Torvalds EXPORT_SYMBOL(write_inode_now); 12821da177e4SLinus Torvalds 12831da177e4SLinus Torvalds /** 12841da177e4SLinus Torvalds * sync_inode - write an inode and its pages to disk. 12851da177e4SLinus Torvalds * @inode: the inode to sync 12861da177e4SLinus Torvalds * @wbc: controls the writeback mode 12871da177e4SLinus Torvalds * 12881da177e4SLinus Torvalds * sync_inode() will write an inode and its pages to disk. It will also 12891da177e4SLinus Torvalds * correctly update the inode on its superblock's dirty inode lists and will 12901da177e4SLinus Torvalds * update inode->i_state. 12911da177e4SLinus Torvalds * 12921da177e4SLinus Torvalds * The caller must have a ref on the inode. 12931da177e4SLinus Torvalds */ 12941da177e4SLinus Torvalds int sync_inode(struct inode *inode, struct writeback_control *wbc) 12951da177e4SLinus Torvalds { 12961da177e4SLinus Torvalds int ret; 12971da177e4SLinus Torvalds 12981da177e4SLinus Torvalds spin_lock(&inode_lock); 129901c03194SChristoph Hellwig ret = writeback_single_inode(inode, wbc); 13001da177e4SLinus Torvalds spin_unlock(&inode_lock); 13011da177e4SLinus Torvalds return ret; 13021da177e4SLinus Torvalds } 13031da177e4SLinus Torvalds EXPORT_SYMBOL(sync_inode); 1304c3765016SChristoph Hellwig 1305c3765016SChristoph Hellwig /** 1306c691b9d9SAndrew Morton * sync_inode_metadata - write an inode to disk 1307c3765016SChristoph Hellwig * @inode: the inode to sync 1308c3765016SChristoph Hellwig * @wait: wait for I/O to complete. 1309c3765016SChristoph Hellwig * 1310c691b9d9SAndrew Morton * Write an inode to disk and adjust its dirty state after completion. 1311c3765016SChristoph Hellwig * 1312c3765016SChristoph Hellwig * Note: only writes the actual inode, no associated data or other metadata. 1313c3765016SChristoph Hellwig */ 1314c3765016SChristoph Hellwig int sync_inode_metadata(struct inode *inode, int wait) 1315c3765016SChristoph Hellwig { 1316c3765016SChristoph Hellwig struct writeback_control wbc = { 1317c3765016SChristoph Hellwig .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE, 1318c3765016SChristoph Hellwig .nr_to_write = 0, /* metadata-only */ 1319c3765016SChristoph Hellwig }; 1320c3765016SChristoph Hellwig 1321c3765016SChristoph Hellwig return sync_inode(inode, &wbc); 1322c3765016SChristoph Hellwig } 1323c3765016SChristoph Hellwig EXPORT_SYMBOL(sync_inode_metadata); 1324