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) { 310250df6edSDave Chinner spin_unlock(&inode->i_lock); 31101c03194SChristoph Hellwig spin_unlock(&inode_lock); 31201c03194SChristoph Hellwig __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE); 31301c03194SChristoph Hellwig spin_lock(&inode_lock); 314250df6edSDave Chinner spin_lock(&inode->i_lock); 31558a9d3d8SRichard Kennedy } 31601c03194SChristoph Hellwig } 31701c03194SChristoph Hellwig 31801c03194SChristoph Hellwig /* 31901c03194SChristoph Hellwig * Write out an inode's dirty pages. Called under inode_lock. Either the 32001c03194SChristoph Hellwig * caller has ref on the inode (either via __iget or via syscall against an fd) 32101c03194SChristoph Hellwig * or the inode has I_WILL_FREE set (via generic_forget_inode) 32201c03194SChristoph Hellwig * 3231da177e4SLinus Torvalds * If `wait' is set, wait on the writeout. 3241da177e4SLinus Torvalds * 3251da177e4SLinus Torvalds * The whole writeout design is quite complex and fragile. We want to avoid 3261da177e4SLinus Torvalds * starvation of particular inodes when others are being redirtied, prevent 3271da177e4SLinus Torvalds * livelocks, etc. 3281da177e4SLinus Torvalds * 3291da177e4SLinus Torvalds * Called under inode_lock. 3301da177e4SLinus Torvalds */ 3311da177e4SLinus Torvalds static int 33201c03194SChristoph Hellwig writeback_single_inode(struct inode *inode, struct writeback_control *wbc) 3331da177e4SLinus Torvalds { 3341da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 33501c03194SChristoph Hellwig unsigned dirty; 3361da177e4SLinus Torvalds int ret; 3371da177e4SLinus Torvalds 338250df6edSDave Chinner spin_lock(&inode->i_lock); 33901c03194SChristoph Hellwig if (!atomic_read(&inode->i_count)) 34001c03194SChristoph Hellwig WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); 34101c03194SChristoph Hellwig else 34201c03194SChristoph Hellwig WARN_ON(inode->i_state & I_WILL_FREE); 34301c03194SChristoph Hellwig 34401c03194SChristoph Hellwig if (inode->i_state & I_SYNC) { 34501c03194SChristoph Hellwig /* 34601c03194SChristoph Hellwig * If this inode is locked for writeback and we are not doing 34766f3b8e2SJens Axboe * writeback-for-data-integrity, move it to b_more_io so that 34801c03194SChristoph Hellwig * writeback can proceed with the other inodes on s_io. 34901c03194SChristoph Hellwig * 35001c03194SChristoph Hellwig * We'll have another go at writing back this inode when we 35166f3b8e2SJens Axboe * completed a full scan of b_io. 35201c03194SChristoph Hellwig */ 353a9185b41SChristoph Hellwig if (wbc->sync_mode != WB_SYNC_ALL) { 354250df6edSDave Chinner spin_unlock(&inode->i_lock); 35501c03194SChristoph Hellwig requeue_io(inode); 35601c03194SChristoph Hellwig return 0; 35701c03194SChristoph Hellwig } 35801c03194SChristoph Hellwig 35901c03194SChristoph Hellwig /* 36001c03194SChristoph Hellwig * It's a data-integrity sync. We must wait. 36101c03194SChristoph Hellwig */ 36201c03194SChristoph Hellwig inode_wait_for_writeback(inode); 36301c03194SChristoph Hellwig } 36401c03194SChristoph Hellwig 3651c0eeaf5SJoern Engel BUG_ON(inode->i_state & I_SYNC); 3661da177e4SLinus Torvalds 3675547e8aaSDmitry Monakhov /* Set I_SYNC, reset I_DIRTY_PAGES */ 3681c0eeaf5SJoern Engel inode->i_state |= I_SYNC; 3695547e8aaSDmitry Monakhov inode->i_state &= ~I_DIRTY_PAGES; 370250df6edSDave Chinner spin_unlock(&inode->i_lock); 3711da177e4SLinus Torvalds spin_unlock(&inode_lock); 3721da177e4SLinus Torvalds 3731da177e4SLinus Torvalds ret = do_writepages(mapping, wbc); 3741da177e4SLinus Torvalds 37526821ed4SChristoph Hellwig /* 37626821ed4SChristoph Hellwig * Make sure to wait on the data before writing out the metadata. 37726821ed4SChristoph Hellwig * This is important for filesystems that modify metadata on data 37826821ed4SChristoph Hellwig * I/O completion. 37926821ed4SChristoph Hellwig */ 380a9185b41SChristoph Hellwig if (wbc->sync_mode == WB_SYNC_ALL) { 38126821ed4SChristoph Hellwig int err = filemap_fdatawait(mapping); 3821da177e4SLinus Torvalds if (ret == 0) 3831da177e4SLinus Torvalds ret = err; 3841da177e4SLinus Torvalds } 3851da177e4SLinus Torvalds 3865547e8aaSDmitry Monakhov /* 3875547e8aaSDmitry Monakhov * Some filesystems may redirty the inode during the writeback 3885547e8aaSDmitry Monakhov * due to delalloc, clear dirty metadata flags right before 3895547e8aaSDmitry Monakhov * write_inode() 3905547e8aaSDmitry Monakhov */ 3915547e8aaSDmitry Monakhov spin_lock(&inode_lock); 392250df6edSDave Chinner spin_lock(&inode->i_lock); 3935547e8aaSDmitry Monakhov dirty = inode->i_state & I_DIRTY; 3945547e8aaSDmitry Monakhov inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC); 395250df6edSDave Chinner spin_unlock(&inode->i_lock); 3965547e8aaSDmitry Monakhov spin_unlock(&inode_lock); 39726821ed4SChristoph Hellwig /* Don't write the inode if only I_DIRTY_PAGES was set */ 39826821ed4SChristoph Hellwig if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { 399a9185b41SChristoph Hellwig int err = write_inode(inode, wbc); 4001da177e4SLinus Torvalds if (ret == 0) 4011da177e4SLinus Torvalds ret = err; 4021da177e4SLinus Torvalds } 4031da177e4SLinus Torvalds 4041da177e4SLinus Torvalds spin_lock(&inode_lock); 405250df6edSDave Chinner spin_lock(&inode->i_lock); 4061c0eeaf5SJoern Engel inode->i_state &= ~I_SYNC; 407a4ffdde6SAl Viro if (!(inode->i_state & I_FREEING)) { 40823539afcSWu Fengguang if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { 4091da177e4SLinus Torvalds /* 4101da177e4SLinus Torvalds * We didn't write back all the pages. nfs_writepages() 411a50aeb40SWu Fengguang * sometimes bales out without doing anything. 4121da177e4SLinus Torvalds */ 4131da177e4SLinus Torvalds inode->i_state |= I_DIRTY_PAGES; 4148bc3be27SFengguang Wu if (wbc->nr_to_write <= 0) { 4158bc3be27SFengguang Wu /* 4168bc3be27SFengguang Wu * slice used up: queue for next turn 4178bc3be27SFengguang Wu */ 4180e0f4fc2SKen Chen requeue_io(inode); 4191da177e4SLinus Torvalds } else { 4201da177e4SLinus Torvalds /* 421a50aeb40SWu Fengguang * Writeback blocked by something other than 422a50aeb40SWu Fengguang * congestion. Delay the inode for some time to 423a50aeb40SWu Fengguang * avoid spinning on the CPU (100% iowait) 424a50aeb40SWu Fengguang * retrying writeback of the dirty page/inode 425a50aeb40SWu Fengguang * that cannot be performed immediately. 4268bc3be27SFengguang Wu */ 4278bc3be27SFengguang Wu redirty_tail(inode); 4288bc3be27SFengguang Wu } 42923539afcSWu Fengguang } else if (inode->i_state & I_DIRTY) { 43023539afcSWu Fengguang /* 43123539afcSWu Fengguang * Filesystems can dirty the inode during writeback 43223539afcSWu Fengguang * operations, such as delayed allocation during 43323539afcSWu Fengguang * submission or metadata updates after data IO 43423539afcSWu Fengguang * completion. 43523539afcSWu Fengguang */ 43623539afcSWu Fengguang redirty_tail(inode); 4371da177e4SLinus Torvalds } else { 4381da177e4SLinus Torvalds /* 4399e38d86fSNick Piggin * The inode is clean. At this point we either have 4409e38d86fSNick Piggin * a reference to the inode or it's on it's way out. 4419e38d86fSNick Piggin * No need to add it back to the LRU. 4421da177e4SLinus Torvalds */ 4437ccf19a8SNick Piggin list_del_init(&inode->i_wb_list); 4441da177e4SLinus Torvalds } 4451da177e4SLinus Torvalds } 4461c0eeaf5SJoern Engel inode_sync_complete(inode); 447250df6edSDave Chinner spin_unlock(&inode->i_lock); 4481da177e4SLinus Torvalds return ret; 4491da177e4SLinus Torvalds } 4501da177e4SLinus Torvalds 45103ba3782SJens Axboe /* 452d19de7edSChristoph Hellwig * For background writeback the caller does not have the sb pinned 45303ba3782SJens Axboe * before calling writeback. So make sure that we do pin it, so it doesn't 45403ba3782SJens Axboe * go away while we are writing inodes from it. 45503ba3782SJens Axboe */ 456d19de7edSChristoph Hellwig static bool pin_sb_for_writeback(struct super_block *sb) 4571da177e4SLinus Torvalds { 45803ba3782SJens Axboe spin_lock(&sb_lock); 45929cb4859SChristoph Hellwig if (list_empty(&sb->s_instances)) { 46003ba3782SJens Axboe spin_unlock(&sb_lock); 46129cb4859SChristoph Hellwig return false; 46203ba3782SJens Axboe } 46329cb4859SChristoph Hellwig 46429cb4859SChristoph Hellwig sb->s_count++; 46529cb4859SChristoph Hellwig spin_unlock(&sb_lock); 46629cb4859SChristoph Hellwig 46729cb4859SChristoph Hellwig if (down_read_trylock(&sb->s_umount)) { 46829cb4859SChristoph Hellwig if (sb->s_root) 46929cb4859SChristoph Hellwig return true; 47003ba3782SJens Axboe up_read(&sb->s_umount); 47103ba3782SJens Axboe } 47229cb4859SChristoph Hellwig 47329cb4859SChristoph Hellwig put_super(sb); 474d19de7edSChristoph Hellwig return false; 47503ba3782SJens Axboe } 47603ba3782SJens Axboe 477f11c9c5cSEdward Shishkin /* 478f11c9c5cSEdward Shishkin * Write a portion of b_io inodes which belong to @sb. 479edadfb10SChristoph Hellwig * 480edadfb10SChristoph Hellwig * If @only_this_sb is true, then find and write all such 481f11c9c5cSEdward Shishkin * inodes. Otherwise write only ones which go sequentially 482f11c9c5cSEdward Shishkin * in reverse order. 483edadfb10SChristoph Hellwig * 484f11c9c5cSEdward Shishkin * Return 1, if the caller writeback routine should be 485f11c9c5cSEdward Shishkin * interrupted. Otherwise return 0. 486f11c9c5cSEdward Shishkin */ 487edadfb10SChristoph Hellwig static int writeback_sb_inodes(struct super_block *sb, struct bdi_writeback *wb, 488edadfb10SChristoph Hellwig struct writeback_control *wbc, bool only_this_sb) 48903ba3782SJens Axboe { 49003ba3782SJens Axboe while (!list_empty(&wb->b_io)) { 491f11c9c5cSEdward Shishkin long pages_skipped; 4927ccf19a8SNick Piggin struct inode *inode = wb_inode(wb->b_io.prev); 493edadfb10SChristoph Hellwig 494edadfb10SChristoph Hellwig if (inode->i_sb != sb) { 495edadfb10SChristoph Hellwig if (only_this_sb) { 496edadfb10SChristoph Hellwig /* 497edadfb10SChristoph Hellwig * We only want to write back data for this 498edadfb10SChristoph Hellwig * superblock, move all inodes not belonging 499edadfb10SChristoph Hellwig * to it back onto the dirty list. 500edadfb10SChristoph Hellwig */ 50166f3b8e2SJens Axboe redirty_tail(inode); 50266f3b8e2SJens Axboe continue; 50366f3b8e2SJens Axboe } 504edadfb10SChristoph Hellwig 505edadfb10SChristoph Hellwig /* 506edadfb10SChristoph Hellwig * The inode belongs to a different superblock. 507edadfb10SChristoph Hellwig * Bounce back to the caller to unpin this and 508edadfb10SChristoph Hellwig * pin the next superblock. 509edadfb10SChristoph Hellwig */ 510f11c9c5cSEdward Shishkin return 0; 511edadfb10SChristoph Hellwig } 512edadfb10SChristoph Hellwig 5139843b76aSChristoph Hellwig /* 5149843b76aSChristoph Hellwig * Don't bother with new inodes or inodes beeing freed, first 5159843b76aSChristoph Hellwig * kind does not need peridic writeout yet, and for the latter 5169843b76aSChristoph Hellwig * kind writeout is handled by the freer. 5179843b76aSChristoph Hellwig */ 518250df6edSDave Chinner spin_lock(&inode->i_lock); 5199843b76aSChristoph Hellwig if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { 520250df6edSDave Chinner spin_unlock(&inode->i_lock); 5217ef0d737SNick Piggin requeue_io(inode); 5227ef0d737SNick Piggin continue; 5237ef0d737SNick Piggin } 5249843b76aSChristoph Hellwig 525d2caa3c5SJeff Layton /* 526d2caa3c5SJeff Layton * Was this inode dirtied after sync_sb_inodes was called? 527d2caa3c5SJeff Layton * This keeps sync from extra jobs and livelock. 528d2caa3c5SJeff Layton */ 529250df6edSDave Chinner if (inode_dirtied_after(inode, wbc->wb_start)) { 530250df6edSDave Chinner spin_unlock(&inode->i_lock); 531f11c9c5cSEdward Shishkin return 1; 532250df6edSDave Chinner } 5331da177e4SLinus Torvalds 5341da177e4SLinus Torvalds __iget(inode); 535250df6edSDave Chinner spin_unlock(&inode->i_lock); 536250df6edSDave Chinner 5371da177e4SLinus Torvalds pages_skipped = wbc->pages_skipped; 53801c03194SChristoph Hellwig writeback_single_inode(inode, wbc); 5391da177e4SLinus Torvalds if (wbc->pages_skipped != pages_skipped) { 5401da177e4SLinus Torvalds /* 5411da177e4SLinus Torvalds * writeback is not making progress due to locked 5421da177e4SLinus Torvalds * buffers. Skip this inode for now. 5431da177e4SLinus Torvalds */ 544f57b9b7bSAndrew Morton redirty_tail(inode); 5451da177e4SLinus Torvalds } 5461da177e4SLinus Torvalds spin_unlock(&inode_lock); 5471da177e4SLinus Torvalds iput(inode); 5484ffc8444SOGAWA Hirofumi cond_resched(); 5491da177e4SLinus Torvalds spin_lock(&inode_lock); 5508bc3be27SFengguang Wu if (wbc->nr_to_write <= 0) { 5518bc3be27SFengguang Wu wbc->more_io = 1; 552f11c9c5cSEdward Shishkin return 1; 5531da177e4SLinus Torvalds } 55403ba3782SJens Axboe if (!list_empty(&wb->b_more_io)) 5558bc3be27SFengguang Wu wbc->more_io = 1; 5568bc3be27SFengguang Wu } 557f11c9c5cSEdward Shishkin /* b_io is empty */ 558f11c9c5cSEdward Shishkin return 1; 559f11c9c5cSEdward Shishkin } 56038f21977SNick Piggin 5619c3a8ee8SChristoph Hellwig void writeback_inodes_wb(struct bdi_writeback *wb, 562f11c9c5cSEdward Shishkin struct writeback_control *wbc) 563f11c9c5cSEdward Shishkin { 564f11c9c5cSEdward Shishkin int ret = 0; 5659ecc2738SJens Axboe 5667624ee72SJan Kara if (!wbc->wb_start) 567f11c9c5cSEdward Shishkin wbc->wb_start = jiffies; /* livelock avoidance */ 568f11c9c5cSEdward Shishkin spin_lock(&inode_lock); 569f11c9c5cSEdward Shishkin if (!wbc->for_kupdate || list_empty(&wb->b_io)) 570f11c9c5cSEdward Shishkin queue_io(wb, wbc->older_than_this); 571f11c9c5cSEdward Shishkin 572f11c9c5cSEdward Shishkin while (!list_empty(&wb->b_io)) { 5737ccf19a8SNick Piggin struct inode *inode = wb_inode(wb->b_io.prev); 574f11c9c5cSEdward Shishkin struct super_block *sb = inode->i_sb; 575f11c9c5cSEdward Shishkin 576334132aeSChristoph Hellwig if (!pin_sb_for_writeback(sb)) { 577334132aeSChristoph Hellwig requeue_io(inode); 578d19de7edSChristoph Hellwig continue; 579334132aeSChristoph Hellwig } 580edadfb10SChristoph Hellwig ret = writeback_sb_inodes(sb, wb, wbc, false); 581d19de7edSChristoph Hellwig drop_super(sb); 582f11c9c5cSEdward Shishkin 583f11c9c5cSEdward Shishkin if (ret) 584f11c9c5cSEdward Shishkin break; 585f11c9c5cSEdward Shishkin } 58666f3b8e2SJens Axboe spin_unlock(&inode_lock); 58766f3b8e2SJens Axboe /* Leave any unwritten inodes on b_io */ 58866f3b8e2SJens Axboe } 58966f3b8e2SJens Axboe 590edadfb10SChristoph Hellwig static void __writeback_inodes_sb(struct super_block *sb, 591edadfb10SChristoph Hellwig struct bdi_writeback *wb, struct writeback_control *wbc) 592edadfb10SChristoph Hellwig { 593edadfb10SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 594edadfb10SChristoph Hellwig 595edadfb10SChristoph Hellwig spin_lock(&inode_lock); 596edadfb10SChristoph Hellwig if (!wbc->for_kupdate || list_empty(&wb->b_io)) 597edadfb10SChristoph Hellwig queue_io(wb, wbc->older_than_this); 598edadfb10SChristoph Hellwig writeback_sb_inodes(sb, wb, wbc, true); 599edadfb10SChristoph Hellwig spin_unlock(&inode_lock); 600edadfb10SChristoph Hellwig } 601edadfb10SChristoph Hellwig 60203ba3782SJens Axboe /* 60303ba3782SJens Axboe * The maximum number of pages to writeout in a single bdi flush/kupdate 60403ba3782SJens Axboe * operation. We do this so we don't hold I_SYNC against an inode for 60503ba3782SJens Axboe * enormous amounts of time, which would block a userspace task which has 60603ba3782SJens Axboe * been forced to throttle against that inode. Also, the code reevaluates 60703ba3782SJens Axboe * the dirty each time it has written this many pages. 60803ba3782SJens Axboe */ 60903ba3782SJens Axboe #define MAX_WRITEBACK_PAGES 1024 61003ba3782SJens Axboe 61103ba3782SJens Axboe static inline bool over_bground_thresh(void) 61203ba3782SJens Axboe { 61303ba3782SJens Axboe unsigned long background_thresh, dirty_thresh; 61403ba3782SJens Axboe 61516c4042fSWu Fengguang global_dirty_limits(&background_thresh, &dirty_thresh); 61603ba3782SJens Axboe 61703ba3782SJens Axboe return (global_page_state(NR_FILE_DIRTY) + 6184cbec4c8SWu Fengguang global_page_state(NR_UNSTABLE_NFS) > background_thresh); 61903ba3782SJens Axboe } 62003ba3782SJens Axboe 62103ba3782SJens Axboe /* 62203ba3782SJens Axboe * Explicit flushing or periodic writeback of "old" data. 62303ba3782SJens Axboe * 62403ba3782SJens Axboe * Define "old": the first time one of an inode's pages is dirtied, we mark the 62503ba3782SJens Axboe * dirtying-time in the inode's address_space. So this periodic writeback code 62603ba3782SJens Axboe * just walks the superblock inode list, writing back any inodes which are 62703ba3782SJens Axboe * older than a specific point in time. 62803ba3782SJens Axboe * 62903ba3782SJens Axboe * Try to run once per dirty_writeback_interval. But if a writeback event 63003ba3782SJens Axboe * takes longer than a dirty_writeback_interval interval, then leave a 63103ba3782SJens Axboe * one-second gap. 63203ba3782SJens Axboe * 63303ba3782SJens Axboe * older_than_this takes precedence over nr_to_write. So we'll only write back 63403ba3782SJens Axboe * all dirty pages if they are all attached to "old" mappings. 63503ba3782SJens Axboe */ 636c4a77a6cSJens Axboe static long wb_writeback(struct bdi_writeback *wb, 63783ba7b07SChristoph Hellwig struct wb_writeback_work *work) 63803ba3782SJens Axboe { 63903ba3782SJens Axboe struct writeback_control wbc = { 64083ba7b07SChristoph Hellwig .sync_mode = work->sync_mode, 64103ba3782SJens Axboe .older_than_this = NULL, 64283ba7b07SChristoph Hellwig .for_kupdate = work->for_kupdate, 64383ba7b07SChristoph Hellwig .for_background = work->for_background, 64483ba7b07SChristoph Hellwig .range_cyclic = work->range_cyclic, 64503ba3782SJens Axboe }; 64603ba3782SJens Axboe unsigned long oldest_jif; 64703ba3782SJens Axboe long wrote = 0; 648b9543dacSJan Kara long write_chunk; 649a5989bdcSJan Kara struct inode *inode; 65003ba3782SJens Axboe 65103ba3782SJens Axboe if (wbc.for_kupdate) { 65203ba3782SJens Axboe wbc.older_than_this = &oldest_jif; 65303ba3782SJens Axboe oldest_jif = jiffies - 65403ba3782SJens Axboe msecs_to_jiffies(dirty_expire_interval * 10); 65503ba3782SJens Axboe } 656c4a77a6cSJens Axboe if (!wbc.range_cyclic) { 657c4a77a6cSJens Axboe wbc.range_start = 0; 658c4a77a6cSJens Axboe wbc.range_end = LLONG_MAX; 659c4a77a6cSJens Axboe } 66003ba3782SJens Axboe 661b9543dacSJan Kara /* 662b9543dacSJan Kara * WB_SYNC_ALL mode does livelock avoidance by syncing dirty 663b9543dacSJan Kara * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX 664b9543dacSJan Kara * here avoids calling into writeback_inodes_wb() more than once. 665b9543dacSJan Kara * 666b9543dacSJan Kara * The intended call sequence for WB_SYNC_ALL writeback is: 667b9543dacSJan Kara * 668b9543dacSJan Kara * wb_writeback() 669b9543dacSJan Kara * __writeback_inodes_sb() <== called only once 670b9543dacSJan Kara * write_cache_pages() <== called once for each inode 671b9543dacSJan Kara * (quickly) tag currently dirty pages 672b9543dacSJan Kara * (maybe slowly) sync all tagged pages 673b9543dacSJan Kara */ 674b9543dacSJan Kara if (wbc.sync_mode == WB_SYNC_NONE) 675b9543dacSJan Kara write_chunk = MAX_WRITEBACK_PAGES; 676b9543dacSJan Kara else 677b9543dacSJan Kara write_chunk = LONG_MAX; 678b9543dacSJan Kara 6797624ee72SJan Kara wbc.wb_start = jiffies; /* livelock avoidance */ 68003ba3782SJens Axboe for (;;) { 68103ba3782SJens Axboe /* 682d3ddec76SWu Fengguang * Stop writeback when nr_pages has been consumed 68303ba3782SJens Axboe */ 68483ba7b07SChristoph Hellwig if (work->nr_pages <= 0) 68503ba3782SJens Axboe break; 68603ba3782SJens Axboe 68703ba3782SJens Axboe /* 688aa373cf5SJan Kara * Background writeout and kupdate-style writeback may 689aa373cf5SJan Kara * run forever. Stop them if there is other work to do 690aa373cf5SJan Kara * so that e.g. sync can proceed. They'll be restarted 691aa373cf5SJan Kara * after the other works are all done. 692aa373cf5SJan Kara */ 693aa373cf5SJan Kara if ((work->for_background || work->for_kupdate) && 694aa373cf5SJan Kara !list_empty(&wb->bdi->work_list)) 695aa373cf5SJan Kara break; 696aa373cf5SJan Kara 697aa373cf5SJan Kara /* 698d3ddec76SWu Fengguang * For background writeout, stop when we are below the 699d3ddec76SWu Fengguang * background dirty threshold 70003ba3782SJens Axboe */ 70183ba7b07SChristoph Hellwig if (work->for_background && !over_bground_thresh()) 70203ba3782SJens Axboe break; 70303ba3782SJens Axboe 70403ba3782SJens Axboe wbc.more_io = 0; 705b9543dacSJan Kara wbc.nr_to_write = write_chunk; 70603ba3782SJens Axboe wbc.pages_skipped = 0; 707028c2dd1SDave Chinner 708028c2dd1SDave Chinner trace_wbc_writeback_start(&wbc, wb->bdi); 70983ba7b07SChristoph Hellwig if (work->sb) 71083ba7b07SChristoph Hellwig __writeback_inodes_sb(work->sb, wb, &wbc); 711edadfb10SChristoph Hellwig else 71203ba3782SJens Axboe writeback_inodes_wb(wb, &wbc); 713028c2dd1SDave Chinner trace_wbc_writeback_written(&wbc, wb->bdi); 714028c2dd1SDave Chinner 715b9543dacSJan Kara work->nr_pages -= write_chunk - wbc.nr_to_write; 716b9543dacSJan Kara wrote += write_chunk - wbc.nr_to_write; 71703ba3782SJens Axboe 71803ba3782SJens Axboe /* 71971fd05a8SJens Axboe * If we consumed everything, see if we have more 72003ba3782SJens Axboe */ 72171fd05a8SJens Axboe if (wbc.nr_to_write <= 0) 72271fd05a8SJens Axboe continue; 72371fd05a8SJens Axboe /* 72471fd05a8SJens Axboe * Didn't write everything and we don't have more IO, bail 72571fd05a8SJens Axboe */ 72671fd05a8SJens Axboe if (!wbc.more_io) 72771fd05a8SJens Axboe break; 72871fd05a8SJens Axboe /* 72971fd05a8SJens Axboe * Did we write something? Try for more 73071fd05a8SJens Axboe */ 731b9543dacSJan Kara if (wbc.nr_to_write < write_chunk) 73203ba3782SJens Axboe continue; 733a5989bdcSJan Kara /* 734a5989bdcSJan Kara * Nothing written. Wait for some inode to 735a5989bdcSJan Kara * become available for writeback. Otherwise 736a5989bdcSJan Kara * we'll just busyloop. 737a5989bdcSJan Kara */ 738a5989bdcSJan Kara spin_lock(&inode_lock); 739a5989bdcSJan Kara if (!list_empty(&wb->b_more_io)) { 7407ccf19a8SNick Piggin inode = wb_inode(wb->b_more_io.prev); 741028c2dd1SDave Chinner trace_wbc_writeback_wait(&wbc, wb->bdi); 742250df6edSDave Chinner spin_lock(&inode->i_lock); 743a5989bdcSJan Kara inode_wait_for_writeback(inode); 744250df6edSDave Chinner spin_unlock(&inode->i_lock); 745a5989bdcSJan Kara } 746a5989bdcSJan Kara spin_unlock(&inode_lock); 74703ba3782SJens Axboe } 74803ba3782SJens Axboe 74903ba3782SJens Axboe return wrote; 75003ba3782SJens Axboe } 75103ba3782SJens Axboe 75203ba3782SJens Axboe /* 75383ba7b07SChristoph Hellwig * Return the next wb_writeback_work struct that hasn't been processed yet. 75403ba3782SJens Axboe */ 75583ba7b07SChristoph Hellwig static struct wb_writeback_work * 75608852b6dSMinchan Kim get_next_work_item(struct backing_dev_info *bdi) 75703ba3782SJens Axboe { 75883ba7b07SChristoph Hellwig struct wb_writeback_work *work = NULL; 75903ba3782SJens Axboe 7606467716aSArtem Bityutskiy spin_lock_bh(&bdi->wb_lock); 76183ba7b07SChristoph Hellwig if (!list_empty(&bdi->work_list)) { 76283ba7b07SChristoph Hellwig work = list_entry(bdi->work_list.next, 76383ba7b07SChristoph Hellwig struct wb_writeback_work, list); 76483ba7b07SChristoph Hellwig list_del_init(&work->list); 76503ba3782SJens Axboe } 7666467716aSArtem Bityutskiy spin_unlock_bh(&bdi->wb_lock); 76783ba7b07SChristoph Hellwig return work; 76803ba3782SJens Axboe } 76903ba3782SJens Axboe 770cdf01dd5SLinus Torvalds /* 771cdf01dd5SLinus Torvalds * Add in the number of potentially dirty inodes, because each inode 772cdf01dd5SLinus Torvalds * write can dirty pagecache in the underlying blockdev. 773cdf01dd5SLinus Torvalds */ 774cdf01dd5SLinus Torvalds static unsigned long get_nr_dirty_pages(void) 775cdf01dd5SLinus Torvalds { 776cdf01dd5SLinus Torvalds return global_page_state(NR_FILE_DIRTY) + 777cdf01dd5SLinus Torvalds global_page_state(NR_UNSTABLE_NFS) + 778cdf01dd5SLinus Torvalds get_nr_dirty_inodes(); 779cdf01dd5SLinus Torvalds } 780cdf01dd5SLinus Torvalds 7816585027aSJan Kara static long wb_check_background_flush(struct bdi_writeback *wb) 7826585027aSJan Kara { 7836585027aSJan Kara if (over_bground_thresh()) { 7846585027aSJan Kara 7856585027aSJan Kara struct wb_writeback_work work = { 7866585027aSJan Kara .nr_pages = LONG_MAX, 7876585027aSJan Kara .sync_mode = WB_SYNC_NONE, 7886585027aSJan Kara .for_background = 1, 7896585027aSJan Kara .range_cyclic = 1, 7906585027aSJan Kara }; 7916585027aSJan Kara 7926585027aSJan Kara return wb_writeback(wb, &work); 7936585027aSJan Kara } 7946585027aSJan Kara 7956585027aSJan Kara return 0; 7966585027aSJan Kara } 7976585027aSJan Kara 79803ba3782SJens Axboe static long wb_check_old_data_flush(struct bdi_writeback *wb) 79903ba3782SJens Axboe { 80003ba3782SJens Axboe unsigned long expired; 80103ba3782SJens Axboe long nr_pages; 80203ba3782SJens Axboe 80369b62d01SJens Axboe /* 80469b62d01SJens Axboe * When set to zero, disable periodic writeback 80569b62d01SJens Axboe */ 80669b62d01SJens Axboe if (!dirty_writeback_interval) 80769b62d01SJens Axboe return 0; 80869b62d01SJens Axboe 80903ba3782SJens Axboe expired = wb->last_old_flush + 81003ba3782SJens Axboe msecs_to_jiffies(dirty_writeback_interval * 10); 81103ba3782SJens Axboe if (time_before(jiffies, expired)) 81203ba3782SJens Axboe return 0; 81303ba3782SJens Axboe 81403ba3782SJens Axboe wb->last_old_flush = jiffies; 815cdf01dd5SLinus Torvalds nr_pages = get_nr_dirty_pages(); 81603ba3782SJens Axboe 817c4a77a6cSJens Axboe if (nr_pages) { 81883ba7b07SChristoph Hellwig struct wb_writeback_work work = { 819c4a77a6cSJens Axboe .nr_pages = nr_pages, 820c4a77a6cSJens Axboe .sync_mode = WB_SYNC_NONE, 821c4a77a6cSJens Axboe .for_kupdate = 1, 822c4a77a6cSJens Axboe .range_cyclic = 1, 823c4a77a6cSJens Axboe }; 824c4a77a6cSJens Axboe 82583ba7b07SChristoph Hellwig return wb_writeback(wb, &work); 826c4a77a6cSJens Axboe } 82703ba3782SJens Axboe 82803ba3782SJens Axboe return 0; 82903ba3782SJens Axboe } 83003ba3782SJens Axboe 83103ba3782SJens Axboe /* 83203ba3782SJens Axboe * Retrieve work items and do the writeback they describe 83303ba3782SJens Axboe */ 83403ba3782SJens Axboe long wb_do_writeback(struct bdi_writeback *wb, int force_wait) 83503ba3782SJens Axboe { 83603ba3782SJens Axboe struct backing_dev_info *bdi = wb->bdi; 83783ba7b07SChristoph Hellwig struct wb_writeback_work *work; 838c4a77a6cSJens Axboe long wrote = 0; 83903ba3782SJens Axboe 84081d73a32SJan Kara set_bit(BDI_writeback_running, &wb->bdi->state); 84108852b6dSMinchan Kim while ((work = get_next_work_item(bdi)) != NULL) { 84203ba3782SJens Axboe /* 84303ba3782SJens Axboe * Override sync mode, in case we must wait for completion 84483ba7b07SChristoph Hellwig * because this thread is exiting now. 84503ba3782SJens Axboe */ 84603ba3782SJens Axboe if (force_wait) 84783ba7b07SChristoph Hellwig work->sync_mode = WB_SYNC_ALL; 84883ba7b07SChristoph Hellwig 849455b2864SDave Chinner trace_writeback_exec(bdi, work); 850455b2864SDave Chinner 85183ba7b07SChristoph Hellwig wrote += wb_writeback(wb, work); 85203ba3782SJens Axboe 85303ba3782SJens Axboe /* 85483ba7b07SChristoph Hellwig * Notify the caller of completion if this is a synchronous 85583ba7b07SChristoph Hellwig * work item, otherwise just free it. 85603ba3782SJens Axboe */ 85783ba7b07SChristoph Hellwig if (work->done) 85883ba7b07SChristoph Hellwig complete(work->done); 85983ba7b07SChristoph Hellwig else 86083ba7b07SChristoph Hellwig kfree(work); 86103ba3782SJens Axboe } 86203ba3782SJens Axboe 86303ba3782SJens Axboe /* 86403ba3782SJens Axboe * Check for periodic writeback, kupdated() style 86503ba3782SJens Axboe */ 86603ba3782SJens Axboe wrote += wb_check_old_data_flush(wb); 8676585027aSJan Kara wrote += wb_check_background_flush(wb); 86881d73a32SJan Kara clear_bit(BDI_writeback_running, &wb->bdi->state); 86903ba3782SJens Axboe 87003ba3782SJens Axboe return wrote; 87103ba3782SJens Axboe } 87203ba3782SJens Axboe 87303ba3782SJens Axboe /* 87403ba3782SJens Axboe * Handle writeback of dirty data for the device backed by this bdi. Also 87503ba3782SJens Axboe * wakes up periodically and does kupdated style flushing. 87603ba3782SJens Axboe */ 87708243900SChristoph Hellwig int bdi_writeback_thread(void *data) 87803ba3782SJens Axboe { 87908243900SChristoph Hellwig struct bdi_writeback *wb = data; 88008243900SChristoph Hellwig struct backing_dev_info *bdi = wb->bdi; 88103ba3782SJens Axboe long pages_written; 88203ba3782SJens Axboe 883766f9164SPeter Zijlstra current->flags |= PF_SWAPWRITE; 88408243900SChristoph Hellwig set_freezable(); 885ecd58403SArtem Bityutskiy wb->last_active = jiffies; 88603ba3782SJens Axboe 88703ba3782SJens Axboe /* 88808243900SChristoph Hellwig * Our parent may run at a different priority, just set us to normal 88903ba3782SJens Axboe */ 89008243900SChristoph Hellwig set_user_nice(current, 0); 89108243900SChristoph Hellwig 892455b2864SDave Chinner trace_writeback_thread_start(bdi); 893455b2864SDave Chinner 89403ba3782SJens Axboe while (!kthread_should_stop()) { 8956467716aSArtem Bityutskiy /* 8966467716aSArtem Bityutskiy * Remove own delayed wake-up timer, since we are already awake 8976467716aSArtem Bityutskiy * and we'll take care of the preriodic write-back. 8986467716aSArtem Bityutskiy */ 8996467716aSArtem Bityutskiy del_timer(&wb->wakeup_timer); 9006467716aSArtem Bityutskiy 90103ba3782SJens Axboe pages_written = wb_do_writeback(wb, 0); 90203ba3782SJens Axboe 903455b2864SDave Chinner trace_writeback_pages_written(pages_written); 904455b2864SDave Chinner 90503ba3782SJens Axboe if (pages_written) 906ecd58403SArtem Bityutskiy wb->last_active = jiffies; 90703ba3782SJens Axboe 908297252c8SArtem Bityutskiy set_current_state(TASK_INTERRUPTIBLE); 909b76b4014SJ. Bruce Fields if (!list_empty(&bdi->work_list) || kthread_should_stop()) { 910297252c8SArtem Bityutskiy __set_current_state(TASK_RUNNING); 911297252c8SArtem Bityutskiy continue; 91203ba3782SJens Axboe } 91303ba3782SJens Axboe 914253c34e9SArtem Bityutskiy if (wb_has_dirty_io(wb) && dirty_writeback_interval) 915fff5b85aSArtem Bityutskiy schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10)); 916253c34e9SArtem Bityutskiy else { 917253c34e9SArtem Bityutskiy /* 918253c34e9SArtem Bityutskiy * We have nothing to do, so can go sleep without any 919253c34e9SArtem Bityutskiy * timeout and save power. When a work is queued or 920253c34e9SArtem Bityutskiy * something is made dirty - we will be woken up. 921253c34e9SArtem Bityutskiy */ 92269b62d01SJens Axboe schedule(); 923f9eadbbdSJens Axboe } 92469b62d01SJens Axboe 92503ba3782SJens Axboe try_to_freeze(); 92603ba3782SJens Axboe } 92703ba3782SJens Axboe 928fff5b85aSArtem Bityutskiy /* Flush any work that raced with us exiting */ 92908243900SChristoph Hellwig if (!list_empty(&bdi->work_list)) 93008243900SChristoph Hellwig wb_do_writeback(wb, 1); 931455b2864SDave Chinner 932455b2864SDave Chinner trace_writeback_thread_stop(bdi); 93303ba3782SJens Axboe return 0; 93403ba3782SJens Axboe } 93503ba3782SJens Axboe 93608243900SChristoph Hellwig 93703ba3782SJens Axboe /* 93803ba3782SJens Axboe * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back 93903ba3782SJens Axboe * the whole world. 94003ba3782SJens Axboe */ 94103ba3782SJens Axboe void wakeup_flusher_threads(long nr_pages) 94203ba3782SJens Axboe { 943b8c2f347SChristoph Hellwig struct backing_dev_info *bdi; 944b8c2f347SChristoph Hellwig 94583ba7b07SChristoph Hellwig if (!nr_pages) { 94683ba7b07SChristoph Hellwig nr_pages = global_page_state(NR_FILE_DIRTY) + 94703ba3782SJens Axboe global_page_state(NR_UNSTABLE_NFS); 948b8c2f347SChristoph Hellwig } 949b8c2f347SChristoph Hellwig 950b8c2f347SChristoph Hellwig rcu_read_lock(); 951b8c2f347SChristoph Hellwig list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { 952b8c2f347SChristoph Hellwig if (!bdi_has_dirty_io(bdi)) 953b8c2f347SChristoph Hellwig continue; 9546585027aSJan Kara __bdi_start_writeback(bdi, nr_pages, false); 955b8c2f347SChristoph Hellwig } 956b8c2f347SChristoph Hellwig rcu_read_unlock(); 95703ba3782SJens Axboe } 95803ba3782SJens Axboe 95903ba3782SJens Axboe static noinline void block_dump___mark_inode_dirty(struct inode *inode) 96003ba3782SJens Axboe { 96103ba3782SJens Axboe if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { 96203ba3782SJens Axboe struct dentry *dentry; 96303ba3782SJens Axboe const char *name = "?"; 96403ba3782SJens Axboe 96503ba3782SJens Axboe dentry = d_find_alias(inode); 96603ba3782SJens Axboe if (dentry) { 96703ba3782SJens Axboe spin_lock(&dentry->d_lock); 96803ba3782SJens Axboe name = (const char *) dentry->d_name.name; 96903ba3782SJens Axboe } 97003ba3782SJens Axboe printk(KERN_DEBUG 97103ba3782SJens Axboe "%s(%d): dirtied inode %lu (%s) on %s\n", 97203ba3782SJens Axboe current->comm, task_pid_nr(current), inode->i_ino, 97303ba3782SJens Axboe name, inode->i_sb->s_id); 97403ba3782SJens Axboe if (dentry) { 97503ba3782SJens Axboe spin_unlock(&dentry->d_lock); 97603ba3782SJens Axboe dput(dentry); 97703ba3782SJens Axboe } 97803ba3782SJens Axboe } 97903ba3782SJens Axboe } 98003ba3782SJens Axboe 98103ba3782SJens Axboe /** 98203ba3782SJens Axboe * __mark_inode_dirty - internal function 98303ba3782SJens Axboe * @inode: inode to mark 98403ba3782SJens Axboe * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) 98503ba3782SJens Axboe * Mark an inode as dirty. Callers should use mark_inode_dirty or 98603ba3782SJens Axboe * mark_inode_dirty_sync. 98703ba3782SJens Axboe * 98803ba3782SJens Axboe * Put the inode on the super block's dirty list. 98903ba3782SJens Axboe * 99003ba3782SJens Axboe * CAREFUL! We mark it dirty unconditionally, but move it onto the 99103ba3782SJens Axboe * dirty list only if it is hashed or if it refers to a blockdev. 99203ba3782SJens Axboe * If it was not hashed, it will never be added to the dirty list 99303ba3782SJens Axboe * even if it is later hashed, as it will have been marked dirty already. 99403ba3782SJens Axboe * 99503ba3782SJens Axboe * In short, make sure you hash any inodes _before_ you start marking 99603ba3782SJens Axboe * them dirty. 99703ba3782SJens Axboe * 99803ba3782SJens Axboe * This function *must* be atomic for the I_DIRTY_PAGES case - 99903ba3782SJens Axboe * set_page_dirty() is called under spinlock in several places. 100003ba3782SJens Axboe * 100103ba3782SJens Axboe * Note that for blockdevs, inode->dirtied_when represents the dirtying time of 100203ba3782SJens Axboe * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of 100303ba3782SJens Axboe * the kernel-internal blockdev inode represents the dirtying time of the 100403ba3782SJens Axboe * blockdev's pages. This is why for I_DIRTY_PAGES we always use 100503ba3782SJens Axboe * page->mapping->host, so the page-dirtying time is recorded in the internal 100603ba3782SJens Axboe * blockdev inode. 100703ba3782SJens Axboe */ 100803ba3782SJens Axboe void __mark_inode_dirty(struct inode *inode, int flags) 100903ba3782SJens Axboe { 101003ba3782SJens Axboe struct super_block *sb = inode->i_sb; 1011253c34e9SArtem Bityutskiy struct backing_dev_info *bdi = NULL; 1012253c34e9SArtem Bityutskiy bool wakeup_bdi = false; 101303ba3782SJens Axboe 101403ba3782SJens Axboe /* 101503ba3782SJens Axboe * Don't do this for I_DIRTY_PAGES - that doesn't actually 101603ba3782SJens Axboe * dirty the inode itself 101703ba3782SJens Axboe */ 101803ba3782SJens Axboe if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { 101903ba3782SJens Axboe if (sb->s_op->dirty_inode) 102003ba3782SJens Axboe sb->s_op->dirty_inode(inode); 102103ba3782SJens Axboe } 102203ba3782SJens Axboe 102303ba3782SJens Axboe /* 102403ba3782SJens Axboe * make sure that changes are seen by all cpus before we test i_state 102503ba3782SJens Axboe * -- mikulas 102603ba3782SJens Axboe */ 102703ba3782SJens Axboe smp_mb(); 102803ba3782SJens Axboe 102903ba3782SJens Axboe /* avoid the locking if we can */ 103003ba3782SJens Axboe if ((inode->i_state & flags) == flags) 103103ba3782SJens Axboe return; 103203ba3782SJens Axboe 103303ba3782SJens Axboe if (unlikely(block_dump)) 103403ba3782SJens Axboe block_dump___mark_inode_dirty(inode); 103503ba3782SJens Axboe 103603ba3782SJens Axboe spin_lock(&inode_lock); 1037250df6edSDave Chinner spin_lock(&inode->i_lock); 103803ba3782SJens Axboe if ((inode->i_state & flags) != flags) { 103903ba3782SJens Axboe const int was_dirty = inode->i_state & I_DIRTY; 104003ba3782SJens Axboe 104103ba3782SJens Axboe inode->i_state |= flags; 104203ba3782SJens Axboe 104303ba3782SJens Axboe /* 104403ba3782SJens Axboe * If the inode is being synced, just update its dirty state. 104503ba3782SJens Axboe * The unlocker will place the inode on the appropriate 104603ba3782SJens Axboe * superblock list, based upon its state. 104703ba3782SJens Axboe */ 104803ba3782SJens Axboe if (inode->i_state & I_SYNC) 1049250df6edSDave Chinner goto out_unlock_inode; 105003ba3782SJens Axboe 105103ba3782SJens Axboe /* 105203ba3782SJens Axboe * Only add valid (hashed) inodes to the superblock's 105303ba3782SJens Axboe * dirty list. Add blockdev inodes as well. 105403ba3782SJens Axboe */ 105503ba3782SJens Axboe if (!S_ISBLK(inode->i_mode)) { 10561d3382cbSAl Viro if (inode_unhashed(inode)) 1057250df6edSDave Chinner goto out_unlock_inode; 105803ba3782SJens Axboe } 1059a4ffdde6SAl Viro if (inode->i_state & I_FREEING) 1060250df6edSDave Chinner goto out_unlock_inode; 106103ba3782SJens Axboe 1062250df6edSDave Chinner spin_unlock(&inode->i_lock); 106303ba3782SJens Axboe /* 106403ba3782SJens Axboe * If the inode was already on b_dirty/b_io/b_more_io, don't 106503ba3782SJens Axboe * reposition it (that would break b_dirty time-ordering). 106603ba3782SJens Axboe */ 106703ba3782SJens Axboe if (!was_dirty) { 1068253c34e9SArtem Bityutskiy bdi = inode_to_bdi(inode); 1069500b067cSJens Axboe 1070253c34e9SArtem Bityutskiy if (bdi_cap_writeback_dirty(bdi)) { 1071253c34e9SArtem Bityutskiy WARN(!test_bit(BDI_registered, &bdi->state), 1072253c34e9SArtem Bityutskiy "bdi-%s not registered\n", bdi->name); 1073253c34e9SArtem Bityutskiy 1074253c34e9SArtem Bityutskiy /* 1075253c34e9SArtem Bityutskiy * If this is the first dirty inode for this 1076253c34e9SArtem Bityutskiy * bdi, we have to wake-up the corresponding 1077253c34e9SArtem Bityutskiy * bdi thread to make sure background 1078253c34e9SArtem Bityutskiy * write-back happens later. 1079253c34e9SArtem Bityutskiy */ 1080253c34e9SArtem Bityutskiy if (!wb_has_dirty_io(&bdi->wb)) 1081253c34e9SArtem Bityutskiy wakeup_bdi = true; 1082500b067cSJens Axboe } 108303ba3782SJens Axboe 108403ba3782SJens Axboe inode->dirtied_when = jiffies; 10857ccf19a8SNick Piggin list_move(&inode->i_wb_list, &bdi->wb.b_dirty); 108603ba3782SJens Axboe } 1087250df6edSDave Chinner goto out; 108803ba3782SJens Axboe } 1089250df6edSDave Chinner out_unlock_inode: 1090250df6edSDave Chinner spin_unlock(&inode->i_lock); 109103ba3782SJens Axboe out: 109203ba3782SJens Axboe spin_unlock(&inode_lock); 1093253c34e9SArtem Bityutskiy 1094253c34e9SArtem Bityutskiy if (wakeup_bdi) 10956467716aSArtem Bityutskiy bdi_wakeup_thread_delayed(bdi); 109603ba3782SJens Axboe } 109703ba3782SJens Axboe EXPORT_SYMBOL(__mark_inode_dirty); 109803ba3782SJens Axboe 109966f3b8e2SJens Axboe /* 110066f3b8e2SJens Axboe * Write out a superblock's list of dirty inodes. A wait will be performed 110166f3b8e2SJens Axboe * upon no inodes, all inodes or the final one, depending upon sync_mode. 110266f3b8e2SJens Axboe * 110366f3b8e2SJens Axboe * If older_than_this is non-NULL, then only write out inodes which 110466f3b8e2SJens Axboe * had their first dirtying at a time earlier than *older_than_this. 110566f3b8e2SJens Axboe * 110666f3b8e2SJens Axboe * If `bdi' is non-zero then we're being asked to writeback a specific queue. 110766f3b8e2SJens Axboe * This function assumes that the blockdev superblock's inodes are backed by 110866f3b8e2SJens Axboe * a variety of queues, so all inodes are searched. For other superblocks, 110966f3b8e2SJens Axboe * assume that all inodes are backed by the same queue. 111066f3b8e2SJens Axboe * 111166f3b8e2SJens Axboe * The inodes to be written are parked on bdi->b_io. They are moved back onto 111266f3b8e2SJens Axboe * bdi->b_dirty as they are selected for writing. This way, none can be missed 111366f3b8e2SJens Axboe * on the writer throttling path, and we get decent balancing between many 111466f3b8e2SJens Axboe * throttled threads: we don't want them all piling up on inode_sync_wait. 111566f3b8e2SJens Axboe */ 1116b6e51316SJens Axboe static void wait_sb_inodes(struct super_block *sb) 111766f3b8e2SJens Axboe { 111838f21977SNick Piggin struct inode *inode, *old_inode = NULL; 111938f21977SNick Piggin 112003ba3782SJens Axboe /* 112103ba3782SJens Axboe * We need to be protected against the filesystem going from 112203ba3782SJens Axboe * r/o to r/w or vice versa. 112303ba3782SJens Axboe */ 1124b6e51316SJens Axboe WARN_ON(!rwsem_is_locked(&sb->s_umount)); 112503ba3782SJens Axboe 112655fa6091SDave Chinner spin_lock(&inode_sb_list_lock); 112766f3b8e2SJens Axboe 112838f21977SNick Piggin /* 112938f21977SNick Piggin * Data integrity sync. Must wait for all pages under writeback, 113038f21977SNick Piggin * because there may have been pages dirtied before our sync 113138f21977SNick Piggin * call, but which had writeout started before we write it out. 113238f21977SNick Piggin * In which case, the inode may not be on the dirty list, but 113338f21977SNick Piggin * we still have to wait for that writeout. 113438f21977SNick Piggin */ 1135b6e51316SJens Axboe list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { 1136250df6edSDave Chinner struct address_space *mapping = inode->i_mapping; 113738f21977SNick Piggin 1138250df6edSDave Chinner spin_lock(&inode->i_lock); 1139250df6edSDave Chinner if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) || 1140250df6edSDave Chinner (mapping->nrpages == 0)) { 1141250df6edSDave Chinner spin_unlock(&inode->i_lock); 114238f21977SNick Piggin continue; 1143250df6edSDave Chinner } 114438f21977SNick Piggin __iget(inode); 1145250df6edSDave Chinner spin_unlock(&inode->i_lock); 114655fa6091SDave Chinner spin_unlock(&inode_sb_list_lock); 114755fa6091SDave Chinner 114838f21977SNick Piggin /* 114955fa6091SDave Chinner * We hold a reference to 'inode' so it couldn't have been 115055fa6091SDave Chinner * removed from s_inodes list while we dropped the 115155fa6091SDave Chinner * inode_sb_list_lock. We cannot iput the inode now as we can 115255fa6091SDave Chinner * be holding the last reference and we cannot iput it under 115355fa6091SDave Chinner * inode_sb_list_lock. So we keep the reference and iput it 115455fa6091SDave Chinner * later. 115538f21977SNick Piggin */ 115638f21977SNick Piggin iput(old_inode); 115738f21977SNick Piggin old_inode = inode; 115838f21977SNick Piggin 115938f21977SNick Piggin filemap_fdatawait(mapping); 116038f21977SNick Piggin 116138f21977SNick Piggin cond_resched(); 116238f21977SNick Piggin 116355fa6091SDave Chinner spin_lock(&inode_sb_list_lock); 116438f21977SNick Piggin } 116555fa6091SDave Chinner spin_unlock(&inode_sb_list_lock); 116638f21977SNick Piggin iput(old_inode); 116766f3b8e2SJens Axboe } 11681da177e4SLinus Torvalds 1169d8a8559cSJens Axboe /** 11703259f8beSChris Mason * writeback_inodes_sb_nr - writeback dirty inodes from given super_block 1171d8a8559cSJens Axboe * @sb: the superblock 11723259f8beSChris Mason * @nr: the number of pages to write 11731da177e4SLinus Torvalds * 1174d8a8559cSJens Axboe * Start writeback on some inodes on this super_block. No guarantees are made 1175d8a8559cSJens Axboe * on how many (if any) will be written, and this function does not wait 11763259f8beSChris Mason * for IO completion of submitted IO. 11771da177e4SLinus Torvalds */ 11783259f8beSChris Mason void writeback_inodes_sb_nr(struct super_block *sb, unsigned long nr) 11791da177e4SLinus Torvalds { 118083ba7b07SChristoph Hellwig DECLARE_COMPLETION_ONSTACK(done); 118183ba7b07SChristoph Hellwig struct wb_writeback_work work = { 11823c4d7165SChristoph Hellwig .sb = sb, 11833c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_NONE, 118483ba7b07SChristoph Hellwig .done = &done, 11853259f8beSChris Mason .nr_pages = nr, 11863c4d7165SChristoph Hellwig }; 11870e3c9a22SJens Axboe 1188cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 118983ba7b07SChristoph Hellwig bdi_queue_work(sb->s_bdi, &work); 119083ba7b07SChristoph Hellwig wait_for_completion(&done); 11911da177e4SLinus Torvalds } 11923259f8beSChris Mason EXPORT_SYMBOL(writeback_inodes_sb_nr); 11933259f8beSChris Mason 11943259f8beSChris Mason /** 11953259f8beSChris Mason * writeback_inodes_sb - writeback dirty inodes from given super_block 11963259f8beSChris Mason * @sb: the superblock 11973259f8beSChris Mason * 11983259f8beSChris Mason * Start writeback on some inodes on this super_block. No guarantees are made 11993259f8beSChris Mason * on how many (if any) will be written, and this function does not wait 12003259f8beSChris Mason * for IO completion of submitted IO. 12013259f8beSChris Mason */ 12023259f8beSChris Mason void writeback_inodes_sb(struct super_block *sb) 12033259f8beSChris Mason { 1204925d169fSLinus Torvalds return writeback_inodes_sb_nr(sb, get_nr_dirty_pages()); 12053259f8beSChris Mason } 1206d8a8559cSJens Axboe EXPORT_SYMBOL(writeback_inodes_sb); 1207d8a8559cSJens Axboe 1208d8a8559cSJens Axboe /** 120917bd55d0SEric Sandeen * writeback_inodes_sb_if_idle - start writeback if none underway 121017bd55d0SEric Sandeen * @sb: the superblock 121117bd55d0SEric Sandeen * 121217bd55d0SEric Sandeen * Invoke writeback_inodes_sb if no writeback is currently underway. 121317bd55d0SEric Sandeen * Returns 1 if writeback was started, 0 if not. 121417bd55d0SEric Sandeen */ 121517bd55d0SEric Sandeen int writeback_inodes_sb_if_idle(struct super_block *sb) 121617bd55d0SEric Sandeen { 121717bd55d0SEric Sandeen if (!writeback_in_progress(sb->s_bdi)) { 1218cf37e972SChristoph Hellwig down_read(&sb->s_umount); 121917bd55d0SEric Sandeen writeback_inodes_sb(sb); 1220cf37e972SChristoph Hellwig up_read(&sb->s_umount); 122117bd55d0SEric Sandeen return 1; 122217bd55d0SEric Sandeen } else 122317bd55d0SEric Sandeen return 0; 122417bd55d0SEric Sandeen } 122517bd55d0SEric Sandeen EXPORT_SYMBOL(writeback_inodes_sb_if_idle); 122617bd55d0SEric Sandeen 122717bd55d0SEric Sandeen /** 12283259f8beSChris Mason * writeback_inodes_sb_if_idle - start writeback if none underway 12293259f8beSChris Mason * @sb: the superblock 12303259f8beSChris Mason * @nr: the number of pages to write 12313259f8beSChris Mason * 12323259f8beSChris Mason * Invoke writeback_inodes_sb if no writeback is currently underway. 12333259f8beSChris Mason * Returns 1 if writeback was started, 0 if not. 12343259f8beSChris Mason */ 12353259f8beSChris Mason int writeback_inodes_sb_nr_if_idle(struct super_block *sb, 12363259f8beSChris Mason unsigned long nr) 12373259f8beSChris Mason { 12383259f8beSChris Mason if (!writeback_in_progress(sb->s_bdi)) { 12393259f8beSChris Mason down_read(&sb->s_umount); 12403259f8beSChris Mason writeback_inodes_sb_nr(sb, nr); 12413259f8beSChris Mason up_read(&sb->s_umount); 12423259f8beSChris Mason return 1; 12433259f8beSChris Mason } else 12443259f8beSChris Mason return 0; 12453259f8beSChris Mason } 12463259f8beSChris Mason EXPORT_SYMBOL(writeback_inodes_sb_nr_if_idle); 12473259f8beSChris Mason 12483259f8beSChris Mason /** 1249d8a8559cSJens Axboe * sync_inodes_sb - sync sb inode pages 1250d8a8559cSJens Axboe * @sb: the superblock 1251d8a8559cSJens Axboe * 1252d8a8559cSJens Axboe * This function writes and waits on any dirty inode belonging to this 1253cb9ef8d5SStefan Hajnoczi * super_block. 1254d8a8559cSJens Axboe */ 1255b6e51316SJens Axboe void sync_inodes_sb(struct super_block *sb) 1256d8a8559cSJens Axboe { 125783ba7b07SChristoph Hellwig DECLARE_COMPLETION_ONSTACK(done); 125883ba7b07SChristoph Hellwig struct wb_writeback_work work = { 12593c4d7165SChristoph Hellwig .sb = sb, 12603c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_ALL, 12613c4d7165SChristoph Hellwig .nr_pages = LONG_MAX, 12623c4d7165SChristoph Hellwig .range_cyclic = 0, 126383ba7b07SChristoph Hellwig .done = &done, 12643c4d7165SChristoph Hellwig }; 12653c4d7165SChristoph Hellwig 1266cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 1267cf37e972SChristoph Hellwig 126883ba7b07SChristoph Hellwig bdi_queue_work(sb->s_bdi, &work); 126983ba7b07SChristoph Hellwig wait_for_completion(&done); 127083ba7b07SChristoph Hellwig 1271b6e51316SJens Axboe wait_sb_inodes(sb); 1272d8a8559cSJens Axboe } 1273d8a8559cSJens Axboe EXPORT_SYMBOL(sync_inodes_sb); 12741da177e4SLinus Torvalds 12751da177e4SLinus Torvalds /** 12761da177e4SLinus Torvalds * write_inode_now - write an inode to disk 12771da177e4SLinus Torvalds * @inode: inode to write to disk 12781da177e4SLinus Torvalds * @sync: whether the write should be synchronous or not 12791da177e4SLinus Torvalds * 12807f04c26dSAndrea Arcangeli * This function commits an inode to disk immediately if it is dirty. This is 12817f04c26dSAndrea Arcangeli * primarily needed by knfsd. 12827f04c26dSAndrea Arcangeli * 12837f04c26dSAndrea Arcangeli * The caller must either have a ref on the inode or must have set I_WILL_FREE. 12841da177e4SLinus Torvalds */ 12851da177e4SLinus Torvalds int write_inode_now(struct inode *inode, int sync) 12861da177e4SLinus Torvalds { 12871da177e4SLinus Torvalds int ret; 12881da177e4SLinus Torvalds struct writeback_control wbc = { 12891da177e4SLinus Torvalds .nr_to_write = LONG_MAX, 129018914b18SMike Galbraith .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, 1291111ebb6eSOGAWA Hirofumi .range_start = 0, 1292111ebb6eSOGAWA Hirofumi .range_end = LLONG_MAX, 12931da177e4SLinus Torvalds }; 12941da177e4SLinus Torvalds 12951da177e4SLinus Torvalds if (!mapping_cap_writeback_dirty(inode->i_mapping)) 129649364ce2SAndrew Morton wbc.nr_to_write = 0; 12971da177e4SLinus Torvalds 12981da177e4SLinus Torvalds might_sleep(); 12991da177e4SLinus Torvalds spin_lock(&inode_lock); 130001c03194SChristoph Hellwig ret = writeback_single_inode(inode, &wbc); 13011da177e4SLinus Torvalds spin_unlock(&inode_lock); 13021da177e4SLinus Torvalds if (sync) 13031c0eeaf5SJoern Engel inode_sync_wait(inode); 13041da177e4SLinus Torvalds return ret; 13051da177e4SLinus Torvalds } 13061da177e4SLinus Torvalds EXPORT_SYMBOL(write_inode_now); 13071da177e4SLinus Torvalds 13081da177e4SLinus Torvalds /** 13091da177e4SLinus Torvalds * sync_inode - write an inode and its pages to disk. 13101da177e4SLinus Torvalds * @inode: the inode to sync 13111da177e4SLinus Torvalds * @wbc: controls the writeback mode 13121da177e4SLinus Torvalds * 13131da177e4SLinus Torvalds * sync_inode() will write an inode and its pages to disk. It will also 13141da177e4SLinus Torvalds * correctly update the inode on its superblock's dirty inode lists and will 13151da177e4SLinus Torvalds * update inode->i_state. 13161da177e4SLinus Torvalds * 13171da177e4SLinus Torvalds * The caller must have a ref on the inode. 13181da177e4SLinus Torvalds */ 13191da177e4SLinus Torvalds int sync_inode(struct inode *inode, struct writeback_control *wbc) 13201da177e4SLinus Torvalds { 13211da177e4SLinus Torvalds int ret; 13221da177e4SLinus Torvalds 13231da177e4SLinus Torvalds spin_lock(&inode_lock); 132401c03194SChristoph Hellwig ret = writeback_single_inode(inode, wbc); 13251da177e4SLinus Torvalds spin_unlock(&inode_lock); 13261da177e4SLinus Torvalds return ret; 13271da177e4SLinus Torvalds } 13281da177e4SLinus Torvalds EXPORT_SYMBOL(sync_inode); 1329c3765016SChristoph Hellwig 1330c3765016SChristoph Hellwig /** 1331c691b9d9SAndrew Morton * sync_inode_metadata - write an inode to disk 1332c3765016SChristoph Hellwig * @inode: the inode to sync 1333c3765016SChristoph Hellwig * @wait: wait for I/O to complete. 1334c3765016SChristoph Hellwig * 1335c691b9d9SAndrew Morton * Write an inode to disk and adjust its dirty state after completion. 1336c3765016SChristoph Hellwig * 1337c3765016SChristoph Hellwig * Note: only writes the actual inode, no associated data or other metadata. 1338c3765016SChristoph Hellwig */ 1339c3765016SChristoph Hellwig int sync_inode_metadata(struct inode *inode, int wait) 1340c3765016SChristoph Hellwig { 1341c3765016SChristoph Hellwig struct writeback_control wbc = { 1342c3765016SChristoph Hellwig .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE, 1343c3765016SChristoph Hellwig .nr_to_write = 0, /* metadata-only */ 1344c3765016SChristoph Hellwig }; 1345c3765016SChristoph Hellwig 1346c3765016SChristoph Hellwig return sync_inode(inode, &wbc); 1347c3765016SChristoph Hellwig } 1348c3765016SChristoph Hellwig EXPORT_SYMBOL(sync_inode_metadata); 1349