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; 396e6938b6SWu Fengguang unsigned int tagged_writepages:1; 4052957fe1SH Hartley Sweeten unsigned int for_kupdate:1; 4152957fe1SH Hartley Sweeten unsigned int range_cyclic:1; 4252957fe1SH Hartley Sweeten unsigned int for_background:1; 43c4a77a6cSJens Axboe 448010c3b6SJens Axboe struct list_head list; /* pending work list */ 4583ba7b07SChristoph Hellwig struct completion *done; /* set if the caller waits */ 4603ba3782SJens Axboe }; 4703ba3782SJens Axboe 48455b2864SDave Chinner /* 49455b2864SDave Chinner * Include the creation of the trace points after defining the 50455b2864SDave Chinner * wb_writeback_work structure so that the definition remains local to this 51455b2864SDave Chinner * file. 52455b2864SDave Chinner */ 53455b2864SDave Chinner #define CREATE_TRACE_POINTS 54455b2864SDave Chinner #include <trace/events/writeback.h> 55455b2864SDave Chinner 56455b2864SDave Chinner /* 57455b2864SDave Chinner * We don't actually have pdflush, but this one is exported though /proc... 58455b2864SDave Chinner */ 59455b2864SDave Chinner int nr_pdflush_threads; 60455b2864SDave Chinner 61f11b00f3SAdrian Bunk /** 62f11b00f3SAdrian Bunk * writeback_in_progress - determine whether there is writeback in progress 63f11b00f3SAdrian Bunk * @bdi: the device's backing_dev_info structure. 64f11b00f3SAdrian Bunk * 6503ba3782SJens Axboe * Determine whether there is writeback waiting to be handled against a 6603ba3782SJens Axboe * backing device. 67f11b00f3SAdrian Bunk */ 68f11b00f3SAdrian Bunk int writeback_in_progress(struct backing_dev_info *bdi) 69f11b00f3SAdrian Bunk { 7081d73a32SJan Kara return test_bit(BDI_writeback_running, &bdi->state); 71f11b00f3SAdrian Bunk } 72f11b00f3SAdrian Bunk 73692ebd17SJan Kara static inline struct backing_dev_info *inode_to_bdi(struct inode *inode) 74692ebd17SJan Kara { 75692ebd17SJan Kara struct super_block *sb = inode->i_sb; 76692ebd17SJan Kara 77aaead25bSChristoph Hellwig if (strcmp(sb->s_type->name, "bdev") == 0) 78aaead25bSChristoph Hellwig return inode->i_mapping->backing_dev_info; 79aaead25bSChristoph Hellwig 80692ebd17SJan Kara return sb->s_bdi; 81692ebd17SJan Kara } 82692ebd17SJan Kara 837ccf19a8SNick Piggin static inline struct inode *wb_inode(struct list_head *head) 847ccf19a8SNick Piggin { 857ccf19a8SNick Piggin return list_entry(head, struct inode, i_wb_list); 867ccf19a8SNick Piggin } 877ccf19a8SNick Piggin 886585027aSJan Kara /* Wakeup flusher thread or forker thread to fork it. Requires bdi->wb_lock. */ 896585027aSJan Kara static void bdi_wakeup_flusher(struct backing_dev_info *bdi) 904195f73dSNick Piggin { 91fff5b85aSArtem Bityutskiy if (bdi->wb.task) { 92fff5b85aSArtem Bityutskiy wake_up_process(bdi->wb.task); 93fff5b85aSArtem Bityutskiy } else { 941da177e4SLinus Torvalds /* 95fff5b85aSArtem Bityutskiy * The bdi thread isn't there, wake up the forker thread which 96fff5b85aSArtem Bityutskiy * will create and run it. 971da177e4SLinus Torvalds */ 9803ba3782SJens Axboe wake_up_process(default_backing_dev_info.wb.task); 991da177e4SLinus Torvalds } 1006585027aSJan Kara } 1016585027aSJan Kara 1026585027aSJan Kara static void bdi_queue_work(struct backing_dev_info *bdi, 1036585027aSJan Kara struct wb_writeback_work *work) 1046585027aSJan Kara { 1056585027aSJan Kara trace_writeback_queue(bdi, work); 1066585027aSJan Kara 1076585027aSJan Kara spin_lock_bh(&bdi->wb_lock); 1086585027aSJan Kara list_add_tail(&work->list, &bdi->work_list); 1096585027aSJan Kara if (!bdi->wb.task) 1106585027aSJan Kara trace_writeback_nothread(bdi, work); 1116585027aSJan Kara bdi_wakeup_flusher(bdi); 1126467716aSArtem Bityutskiy spin_unlock_bh(&bdi->wb_lock); 11303ba3782SJens Axboe } 1141da177e4SLinus Torvalds 11583ba7b07SChristoph Hellwig static void 11683ba7b07SChristoph Hellwig __bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, 1176585027aSJan Kara bool range_cyclic) 1181da177e4SLinus Torvalds { 11983ba7b07SChristoph Hellwig struct wb_writeback_work *work; 12003ba3782SJens Axboe 121bcddc3f0SJens Axboe /* 122bcddc3f0SJens Axboe * This is WB_SYNC_NONE writeback, so if allocation fails just 123bcddc3f0SJens Axboe * wakeup the thread for old dirty data writeback 124bcddc3f0SJens Axboe */ 12583ba7b07SChristoph Hellwig work = kzalloc(sizeof(*work), GFP_ATOMIC); 12683ba7b07SChristoph Hellwig if (!work) { 127455b2864SDave Chinner if (bdi->wb.task) { 128455b2864SDave Chinner trace_writeback_nowork(bdi); 12983ba7b07SChristoph Hellwig wake_up_process(bdi->wb.task); 130455b2864SDave Chinner } 13183ba7b07SChristoph Hellwig return; 13283ba7b07SChristoph Hellwig } 13383ba7b07SChristoph Hellwig 13483ba7b07SChristoph Hellwig work->sync_mode = WB_SYNC_NONE; 13583ba7b07SChristoph Hellwig work->nr_pages = nr_pages; 13683ba7b07SChristoph Hellwig work->range_cyclic = range_cyclic; 13783ba7b07SChristoph Hellwig 138f11fcae8SJens Axboe bdi_queue_work(bdi, work); 13903ba3782SJens Axboe } 140b6e51316SJens Axboe 141b6e51316SJens Axboe /** 142b6e51316SJens Axboe * bdi_start_writeback - start writeback 143b6e51316SJens Axboe * @bdi: the backing device to write from 144b6e51316SJens Axboe * @nr_pages: the number of pages to write 145b6e51316SJens Axboe * 146b6e51316SJens Axboe * Description: 147b6e51316SJens Axboe * This does WB_SYNC_NONE opportunistic writeback. The IO is only 14825985edcSLucas De Marchi * started when this function returns, we make no guarantees on 1490e3c9a22SJens Axboe * completion. Caller need not hold sb s_umount semaphore. 150b6e51316SJens Axboe * 151b6e51316SJens Axboe */ 152c5444198SChristoph Hellwig void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages) 153b6e51316SJens Axboe { 1546585027aSJan Kara __bdi_start_writeback(bdi, nr_pages, true); 155d3ddec76SWu Fengguang } 156d3ddec76SWu Fengguang 157c5444198SChristoph Hellwig /** 158c5444198SChristoph Hellwig * bdi_start_background_writeback - start background writeback 159c5444198SChristoph Hellwig * @bdi: the backing device to write from 160c5444198SChristoph Hellwig * 161c5444198SChristoph Hellwig * Description: 1626585027aSJan Kara * This makes sure WB_SYNC_NONE background writeback happens. When 1636585027aSJan Kara * this function returns, it is only guaranteed that for given BDI 1646585027aSJan Kara * some IO is happening if we are over background dirty threshold. 1656585027aSJan Kara * Caller need not hold sb s_umount semaphore. 166c5444198SChristoph Hellwig */ 167c5444198SChristoph Hellwig void bdi_start_background_writeback(struct backing_dev_info *bdi) 168c5444198SChristoph Hellwig { 1696585027aSJan Kara /* 1706585027aSJan Kara * We just wake up the flusher thread. It will perform background 1716585027aSJan Kara * writeback as soon as there is no other work to do. 1726585027aSJan Kara */ 17371927e84SWu Fengguang trace_writeback_wake_background(bdi); 1746585027aSJan Kara spin_lock_bh(&bdi->wb_lock); 1756585027aSJan Kara bdi_wakeup_flusher(bdi); 1766585027aSJan Kara spin_unlock_bh(&bdi->wb_lock); 1771da177e4SLinus Torvalds } 1781da177e4SLinus Torvalds 1791da177e4SLinus Torvalds /* 180a66979abSDave Chinner * Remove the inode from the writeback list it is on. 181a66979abSDave Chinner */ 182a66979abSDave Chinner void inode_wb_list_del(struct inode *inode) 183a66979abSDave Chinner { 184a66979abSDave Chinner spin_lock(&inode_wb_list_lock); 185a66979abSDave Chinner list_del_init(&inode->i_wb_list); 186a66979abSDave Chinner spin_unlock(&inode_wb_list_lock); 187a66979abSDave Chinner } 188a66979abSDave Chinner 189a66979abSDave Chinner 190a66979abSDave Chinner /* 1916610a0bcSAndrew Morton * Redirty an inode: set its when-it-was dirtied timestamp and move it to the 1926610a0bcSAndrew Morton * furthest end of its superblock's dirty-inode list. 1936610a0bcSAndrew Morton * 1946610a0bcSAndrew Morton * Before stamping the inode's ->dirtied_when, we check to see whether it is 19566f3b8e2SJens Axboe * already the most-recently-dirtied inode on the b_dirty list. If that is 1966610a0bcSAndrew Morton * the case then the inode must have been redirtied while it was being written 1976610a0bcSAndrew Morton * out and we don't reset its dirtied_when. 1986610a0bcSAndrew Morton */ 1996610a0bcSAndrew Morton static void redirty_tail(struct inode *inode) 2006610a0bcSAndrew Morton { 20103ba3782SJens Axboe struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; 2026610a0bcSAndrew Morton 203a66979abSDave Chinner assert_spin_locked(&inode_wb_list_lock); 20403ba3782SJens Axboe if (!list_empty(&wb->b_dirty)) { 20566f3b8e2SJens Axboe struct inode *tail; 2066610a0bcSAndrew Morton 2077ccf19a8SNick Piggin tail = wb_inode(wb->b_dirty.next); 20866f3b8e2SJens Axboe if (time_before(inode->dirtied_when, tail->dirtied_when)) 2096610a0bcSAndrew Morton inode->dirtied_when = jiffies; 2106610a0bcSAndrew Morton } 2117ccf19a8SNick Piggin list_move(&inode->i_wb_list, &wb->b_dirty); 2126610a0bcSAndrew Morton } 2136610a0bcSAndrew Morton 2146610a0bcSAndrew Morton /* 21566f3b8e2SJens Axboe * requeue inode for re-scanning after bdi->b_io list is exhausted. 216c986d1e2SAndrew Morton */ 2170e0f4fc2SKen Chen static void requeue_io(struct inode *inode) 218c986d1e2SAndrew Morton { 21903ba3782SJens Axboe struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; 22003ba3782SJens Axboe 221a66979abSDave Chinner assert_spin_locked(&inode_wb_list_lock); 2227ccf19a8SNick Piggin list_move(&inode->i_wb_list, &wb->b_more_io); 223c986d1e2SAndrew Morton } 224c986d1e2SAndrew Morton 2251c0eeaf5SJoern Engel static void inode_sync_complete(struct inode *inode) 2261c0eeaf5SJoern Engel { 2271c0eeaf5SJoern Engel /* 228a66979abSDave Chinner * Prevent speculative execution through 229a66979abSDave Chinner * spin_unlock(&inode_wb_list_lock); 2301c0eeaf5SJoern Engel */ 231a66979abSDave Chinner 2321c0eeaf5SJoern Engel smp_mb(); 2331c0eeaf5SJoern Engel wake_up_bit(&inode->i_state, __I_SYNC); 2341c0eeaf5SJoern Engel } 2351c0eeaf5SJoern Engel 236d2caa3c5SJeff Layton static bool inode_dirtied_after(struct inode *inode, unsigned long t) 237d2caa3c5SJeff Layton { 238d2caa3c5SJeff Layton bool ret = time_after(inode->dirtied_when, t); 239d2caa3c5SJeff Layton #ifndef CONFIG_64BIT 240d2caa3c5SJeff Layton /* 241d2caa3c5SJeff Layton * For inodes being constantly redirtied, dirtied_when can get stuck. 242d2caa3c5SJeff Layton * It _appears_ to be in the future, but is actually in distant past. 243d2caa3c5SJeff Layton * This test is necessary to prevent such wrapped-around relative times 2445b0830cbSJens Axboe * from permanently stopping the whole bdi writeback. 245d2caa3c5SJeff Layton */ 246d2caa3c5SJeff Layton ret = ret && time_before_eq(inode->dirtied_when, jiffies); 247d2caa3c5SJeff Layton #endif 248d2caa3c5SJeff Layton return ret; 249d2caa3c5SJeff Layton } 250d2caa3c5SJeff Layton 251c986d1e2SAndrew Morton /* 2522c136579SFengguang Wu * Move expired dirty inodes from @delaying_queue to @dispatch_queue. 2532c136579SFengguang Wu */ 2542c136579SFengguang Wu static void move_expired_inodes(struct list_head *delaying_queue, 2552c136579SFengguang Wu struct list_head *dispatch_queue, 2562c136579SFengguang Wu unsigned long *older_than_this) 2572c136579SFengguang Wu { 2585c03449dSShaohua Li LIST_HEAD(tmp); 2595c03449dSShaohua Li struct list_head *pos, *node; 260cf137307SJens Axboe struct super_block *sb = NULL; 2615c03449dSShaohua Li struct inode *inode; 262cf137307SJens Axboe int do_sb_sort = 0; 2635c03449dSShaohua Li 2642c136579SFengguang Wu while (!list_empty(delaying_queue)) { 2657ccf19a8SNick Piggin inode = wb_inode(delaying_queue->prev); 2662c136579SFengguang Wu if (older_than_this && 267d2caa3c5SJeff Layton inode_dirtied_after(inode, *older_than_this)) 2682c136579SFengguang Wu break; 269cf137307SJens Axboe if (sb && sb != inode->i_sb) 270cf137307SJens Axboe do_sb_sort = 1; 271cf137307SJens Axboe sb = inode->i_sb; 2727ccf19a8SNick Piggin list_move(&inode->i_wb_list, &tmp); 2735c03449dSShaohua Li } 2745c03449dSShaohua Li 275cf137307SJens Axboe /* just one sb in list, splice to dispatch_queue and we're done */ 276cf137307SJens Axboe if (!do_sb_sort) { 277cf137307SJens Axboe list_splice(&tmp, dispatch_queue); 278cf137307SJens Axboe return; 279cf137307SJens Axboe } 280cf137307SJens Axboe 2815c03449dSShaohua Li /* Move inodes from one superblock together */ 2825c03449dSShaohua Li while (!list_empty(&tmp)) { 2837ccf19a8SNick Piggin sb = wb_inode(tmp.prev)->i_sb; 2845c03449dSShaohua Li list_for_each_prev_safe(pos, node, &tmp) { 2857ccf19a8SNick Piggin inode = wb_inode(pos); 2865c03449dSShaohua Li if (inode->i_sb == sb) 2877ccf19a8SNick Piggin list_move(&inode->i_wb_list, dispatch_queue); 2882c136579SFengguang Wu } 2892c136579SFengguang Wu } 2905c03449dSShaohua Li } 2912c136579SFengguang Wu 2922c136579SFengguang Wu /* 2932c136579SFengguang Wu * Queue all expired dirty inodes for io, eldest first. 2944ea879b9SWu Fengguang * Before 2954ea879b9SWu Fengguang * newly dirtied b_dirty b_io b_more_io 2964ea879b9SWu Fengguang * =============> gf edc BA 2974ea879b9SWu Fengguang * After 2984ea879b9SWu Fengguang * newly dirtied b_dirty b_io b_more_io 2994ea879b9SWu Fengguang * =============> g fBAedc 3004ea879b9SWu Fengguang * | 3014ea879b9SWu Fengguang * +--> dequeue for IO 3022c136579SFengguang Wu */ 30303ba3782SJens Axboe static void queue_io(struct bdi_writeback *wb, unsigned long *older_than_this) 3042c136579SFengguang Wu { 305a66979abSDave Chinner assert_spin_locked(&inode_wb_list_lock); 3064ea879b9SWu Fengguang list_splice_init(&wb->b_more_io, &wb->b_io); 30703ba3782SJens Axboe move_expired_inodes(&wb->b_dirty, &wb->b_io, older_than_this); 30866f3b8e2SJens Axboe } 30966f3b8e2SJens Axboe 310a9185b41SChristoph Hellwig static int write_inode(struct inode *inode, struct writeback_control *wbc) 31166f3b8e2SJens Axboe { 31203ba3782SJens Axboe if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) 313a9185b41SChristoph Hellwig return inode->i_sb->s_op->write_inode(inode, wbc); 31403ba3782SJens Axboe return 0; 31566f3b8e2SJens Axboe } 31608d8e974SFengguang Wu 3172c136579SFengguang Wu /* 31801c03194SChristoph Hellwig * Wait for writeback on an inode to complete. 31901c03194SChristoph Hellwig */ 32001c03194SChristoph Hellwig static void inode_wait_for_writeback(struct inode *inode) 32101c03194SChristoph Hellwig { 32201c03194SChristoph Hellwig DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); 32301c03194SChristoph Hellwig wait_queue_head_t *wqh; 32401c03194SChristoph Hellwig 32501c03194SChristoph Hellwig wqh = bit_waitqueue(&inode->i_state, __I_SYNC); 32658a9d3d8SRichard Kennedy while (inode->i_state & I_SYNC) { 327250df6edSDave Chinner spin_unlock(&inode->i_lock); 328a66979abSDave Chinner spin_unlock(&inode_wb_list_lock); 32901c03194SChristoph Hellwig __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE); 330a66979abSDave Chinner spin_lock(&inode_wb_list_lock); 331250df6edSDave Chinner spin_lock(&inode->i_lock); 33258a9d3d8SRichard Kennedy } 33301c03194SChristoph Hellwig } 33401c03194SChristoph Hellwig 33501c03194SChristoph Hellwig /* 3360f1b1fd8SDave Chinner * Write out an inode's dirty pages. Called under inode_wb_list_lock and 3370f1b1fd8SDave Chinner * inode->i_lock. Either the caller has an active reference on the inode or 3380f1b1fd8SDave Chinner * the inode has I_WILL_FREE set. 33901c03194SChristoph Hellwig * 3401da177e4SLinus Torvalds * If `wait' is set, wait on the writeout. 3411da177e4SLinus Torvalds * 3421da177e4SLinus Torvalds * The whole writeout design is quite complex and fragile. We want to avoid 3431da177e4SLinus Torvalds * starvation of particular inodes when others are being redirtied, prevent 3441da177e4SLinus Torvalds * livelocks, etc. 3451da177e4SLinus Torvalds */ 3461da177e4SLinus Torvalds static int 34701c03194SChristoph Hellwig writeback_single_inode(struct inode *inode, struct writeback_control *wbc) 3481da177e4SLinus Torvalds { 3491da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 35001c03194SChristoph Hellwig unsigned dirty; 3511da177e4SLinus Torvalds int ret; 3521da177e4SLinus Torvalds 3530f1b1fd8SDave Chinner assert_spin_locked(&inode_wb_list_lock); 3540f1b1fd8SDave Chinner assert_spin_locked(&inode->i_lock); 3550f1b1fd8SDave Chinner 35601c03194SChristoph Hellwig if (!atomic_read(&inode->i_count)) 35701c03194SChristoph Hellwig WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); 35801c03194SChristoph Hellwig else 35901c03194SChristoph Hellwig WARN_ON(inode->i_state & I_WILL_FREE); 36001c03194SChristoph Hellwig 36101c03194SChristoph Hellwig if (inode->i_state & I_SYNC) { 36201c03194SChristoph Hellwig /* 36301c03194SChristoph Hellwig * If this inode is locked for writeback and we are not doing 36466f3b8e2SJens Axboe * writeback-for-data-integrity, move it to b_more_io so that 36501c03194SChristoph Hellwig * writeback can proceed with the other inodes on s_io. 36601c03194SChristoph Hellwig * 36701c03194SChristoph Hellwig * We'll have another go at writing back this inode when we 36866f3b8e2SJens Axboe * completed a full scan of b_io. 36901c03194SChristoph Hellwig */ 370a9185b41SChristoph Hellwig if (wbc->sync_mode != WB_SYNC_ALL) { 37101c03194SChristoph Hellwig requeue_io(inode); 37201c03194SChristoph Hellwig return 0; 37301c03194SChristoph Hellwig } 37401c03194SChristoph Hellwig 37501c03194SChristoph Hellwig /* 37601c03194SChristoph Hellwig * It's a data-integrity sync. We must wait. 37701c03194SChristoph Hellwig */ 37801c03194SChristoph Hellwig inode_wait_for_writeback(inode); 37901c03194SChristoph Hellwig } 38001c03194SChristoph Hellwig 3811c0eeaf5SJoern Engel BUG_ON(inode->i_state & I_SYNC); 3821da177e4SLinus Torvalds 3835547e8aaSDmitry Monakhov /* Set I_SYNC, reset I_DIRTY_PAGES */ 3841c0eeaf5SJoern Engel inode->i_state |= I_SYNC; 3855547e8aaSDmitry Monakhov inode->i_state &= ~I_DIRTY_PAGES; 386250df6edSDave Chinner spin_unlock(&inode->i_lock); 387a66979abSDave Chinner spin_unlock(&inode_wb_list_lock); 3881da177e4SLinus Torvalds 3891da177e4SLinus Torvalds ret = do_writepages(mapping, wbc); 3901da177e4SLinus Torvalds 39126821ed4SChristoph Hellwig /* 39226821ed4SChristoph Hellwig * Make sure to wait on the data before writing out the metadata. 39326821ed4SChristoph Hellwig * This is important for filesystems that modify metadata on data 39426821ed4SChristoph Hellwig * I/O completion. 39526821ed4SChristoph Hellwig */ 396a9185b41SChristoph Hellwig if (wbc->sync_mode == WB_SYNC_ALL) { 39726821ed4SChristoph Hellwig int err = filemap_fdatawait(mapping); 3981da177e4SLinus Torvalds if (ret == 0) 3991da177e4SLinus Torvalds ret = err; 4001da177e4SLinus Torvalds } 4011da177e4SLinus Torvalds 4025547e8aaSDmitry Monakhov /* 4035547e8aaSDmitry Monakhov * Some filesystems may redirty the inode during the writeback 4045547e8aaSDmitry Monakhov * due to delalloc, clear dirty metadata flags right before 4055547e8aaSDmitry Monakhov * write_inode() 4065547e8aaSDmitry Monakhov */ 407250df6edSDave Chinner spin_lock(&inode->i_lock); 4085547e8aaSDmitry Monakhov dirty = inode->i_state & I_DIRTY; 4095547e8aaSDmitry Monakhov inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC); 410250df6edSDave Chinner spin_unlock(&inode->i_lock); 41126821ed4SChristoph Hellwig /* Don't write the inode if only I_DIRTY_PAGES was set */ 41226821ed4SChristoph Hellwig if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { 413a9185b41SChristoph Hellwig int err = write_inode(inode, wbc); 4141da177e4SLinus Torvalds if (ret == 0) 4151da177e4SLinus Torvalds ret = err; 4161da177e4SLinus Torvalds } 4171da177e4SLinus Torvalds 418a66979abSDave Chinner spin_lock(&inode_wb_list_lock); 419250df6edSDave Chinner spin_lock(&inode->i_lock); 4201c0eeaf5SJoern Engel inode->i_state &= ~I_SYNC; 421a4ffdde6SAl Viro if (!(inode->i_state & I_FREEING)) { 42294c3dcbbSWu Fengguang /* 42394c3dcbbSWu Fengguang * Sync livelock prevention. Each inode is tagged and synced in 42494c3dcbbSWu Fengguang * one shot. If still dirty, it will be redirty_tail()'ed below. 42594c3dcbbSWu Fengguang * Update the dirty time to prevent enqueue and sync it again. 42694c3dcbbSWu Fengguang */ 42794c3dcbbSWu Fengguang if ((inode->i_state & I_DIRTY) && 42894c3dcbbSWu Fengguang (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)) 42994c3dcbbSWu Fengguang inode->dirtied_when = jiffies; 43094c3dcbbSWu Fengguang 43123539afcSWu Fengguang if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { 4321da177e4SLinus Torvalds /* 4331da177e4SLinus Torvalds * We didn't write back all the pages. nfs_writepages() 434a50aeb40SWu Fengguang * sometimes bales out without doing anything. 4351da177e4SLinus Torvalds */ 4361da177e4SLinus Torvalds inode->i_state |= I_DIRTY_PAGES; 4378bc3be27SFengguang Wu if (wbc->nr_to_write <= 0) { 4388bc3be27SFengguang Wu /* 4398bc3be27SFengguang Wu * slice used up: queue for next turn 4408bc3be27SFengguang Wu */ 4410e0f4fc2SKen Chen requeue_io(inode); 4421da177e4SLinus Torvalds } else { 4431da177e4SLinus Torvalds /* 444a50aeb40SWu Fengguang * Writeback blocked by something other than 445a50aeb40SWu Fengguang * congestion. Delay the inode for some time to 446a50aeb40SWu Fengguang * avoid spinning on the CPU (100% iowait) 447a50aeb40SWu Fengguang * retrying writeback of the dirty page/inode 448a50aeb40SWu Fengguang * that cannot be performed immediately. 4498bc3be27SFengguang Wu */ 4508bc3be27SFengguang Wu redirty_tail(inode); 4518bc3be27SFengguang Wu } 45223539afcSWu Fengguang } else if (inode->i_state & I_DIRTY) { 45323539afcSWu Fengguang /* 45423539afcSWu Fengguang * Filesystems can dirty the inode during writeback 45523539afcSWu Fengguang * operations, such as delayed allocation during 45623539afcSWu Fengguang * submission or metadata updates after data IO 45723539afcSWu Fengguang * completion. 45823539afcSWu Fengguang */ 45923539afcSWu Fengguang redirty_tail(inode); 4601da177e4SLinus Torvalds } else { 4611da177e4SLinus Torvalds /* 4629e38d86fSNick Piggin * The inode is clean. At this point we either have 4639e38d86fSNick Piggin * a reference to the inode or it's on it's way out. 4649e38d86fSNick Piggin * No need to add it back to the LRU. 4651da177e4SLinus Torvalds */ 4667ccf19a8SNick Piggin list_del_init(&inode->i_wb_list); 467cb9bd115SWu Fengguang wbc->inodes_written++; 4681da177e4SLinus Torvalds } 4691da177e4SLinus Torvalds } 4701c0eeaf5SJoern Engel inode_sync_complete(inode); 4711da177e4SLinus Torvalds return ret; 4721da177e4SLinus Torvalds } 4731da177e4SLinus Torvalds 47403ba3782SJens Axboe /* 475d19de7edSChristoph Hellwig * For background writeback the caller does not have the sb pinned 47603ba3782SJens Axboe * before calling writeback. So make sure that we do pin it, so it doesn't 47703ba3782SJens Axboe * go away while we are writing inodes from it. 47803ba3782SJens Axboe */ 479d19de7edSChristoph Hellwig static bool pin_sb_for_writeback(struct super_block *sb) 4801da177e4SLinus Torvalds { 48103ba3782SJens Axboe spin_lock(&sb_lock); 48229cb4859SChristoph Hellwig if (list_empty(&sb->s_instances)) { 48303ba3782SJens Axboe spin_unlock(&sb_lock); 48429cb4859SChristoph Hellwig return false; 48503ba3782SJens Axboe } 48629cb4859SChristoph Hellwig 48729cb4859SChristoph Hellwig sb->s_count++; 48829cb4859SChristoph Hellwig spin_unlock(&sb_lock); 48929cb4859SChristoph Hellwig 49029cb4859SChristoph Hellwig if (down_read_trylock(&sb->s_umount)) { 49129cb4859SChristoph Hellwig if (sb->s_root) 49229cb4859SChristoph Hellwig return true; 49303ba3782SJens Axboe up_read(&sb->s_umount); 49403ba3782SJens Axboe } 49529cb4859SChristoph Hellwig 49629cb4859SChristoph Hellwig put_super(sb); 497d19de7edSChristoph Hellwig return false; 49803ba3782SJens Axboe } 49903ba3782SJens Axboe 500f11c9c5cSEdward Shishkin /* 501f11c9c5cSEdward Shishkin * Write a portion of b_io inodes which belong to @sb. 502edadfb10SChristoph Hellwig * 503edadfb10SChristoph Hellwig * If @only_this_sb is true, then find and write all such 504f11c9c5cSEdward Shishkin * inodes. Otherwise write only ones which go sequentially 505f11c9c5cSEdward Shishkin * in reverse order. 506edadfb10SChristoph Hellwig * 507f11c9c5cSEdward Shishkin * Return 1, if the caller writeback routine should be 508f11c9c5cSEdward Shishkin * interrupted. Otherwise return 0. 509f11c9c5cSEdward Shishkin */ 510edadfb10SChristoph Hellwig static int writeback_sb_inodes(struct super_block *sb, struct bdi_writeback *wb, 511edadfb10SChristoph Hellwig struct writeback_control *wbc, bool only_this_sb) 51203ba3782SJens Axboe { 51303ba3782SJens Axboe while (!list_empty(&wb->b_io)) { 514f11c9c5cSEdward Shishkin long pages_skipped; 5157ccf19a8SNick Piggin struct inode *inode = wb_inode(wb->b_io.prev); 516edadfb10SChristoph Hellwig 517edadfb10SChristoph Hellwig if (inode->i_sb != sb) { 518edadfb10SChristoph Hellwig if (only_this_sb) { 519edadfb10SChristoph Hellwig /* 520edadfb10SChristoph Hellwig * We only want to write back data for this 521edadfb10SChristoph Hellwig * superblock, move all inodes not belonging 522edadfb10SChristoph Hellwig * to it back onto the dirty list. 523edadfb10SChristoph Hellwig */ 52466f3b8e2SJens Axboe redirty_tail(inode); 52566f3b8e2SJens Axboe continue; 52666f3b8e2SJens Axboe } 527edadfb10SChristoph Hellwig 528edadfb10SChristoph Hellwig /* 529edadfb10SChristoph Hellwig * The inode belongs to a different superblock. 530edadfb10SChristoph Hellwig * Bounce back to the caller to unpin this and 531edadfb10SChristoph Hellwig * pin the next superblock. 532edadfb10SChristoph Hellwig */ 533f11c9c5cSEdward Shishkin return 0; 534edadfb10SChristoph Hellwig } 535edadfb10SChristoph Hellwig 5369843b76aSChristoph Hellwig /* 5379843b76aSChristoph Hellwig * Don't bother with new inodes or inodes beeing freed, first 5389843b76aSChristoph Hellwig * kind does not need peridic writeout yet, and for the latter 5399843b76aSChristoph Hellwig * kind writeout is handled by the freer. 5409843b76aSChristoph Hellwig */ 541250df6edSDave Chinner spin_lock(&inode->i_lock); 5429843b76aSChristoph Hellwig if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { 543250df6edSDave Chinner spin_unlock(&inode->i_lock); 5447ef0d737SNick Piggin requeue_io(inode); 5457ef0d737SNick Piggin continue; 5467ef0d737SNick Piggin } 5479843b76aSChristoph Hellwig 548d2caa3c5SJeff Layton /* 549d2caa3c5SJeff Layton * Was this inode dirtied after sync_sb_inodes was called? 550d2caa3c5SJeff Layton * This keeps sync from extra jobs and livelock. 551d2caa3c5SJeff Layton */ 552250df6edSDave Chinner if (inode_dirtied_after(inode, wbc->wb_start)) { 553250df6edSDave Chinner spin_unlock(&inode->i_lock); 554f11c9c5cSEdward Shishkin return 1; 555250df6edSDave Chinner } 5561da177e4SLinus Torvalds 5571da177e4SLinus Torvalds __iget(inode); 558250df6edSDave Chinner 5591da177e4SLinus Torvalds pages_skipped = wbc->pages_skipped; 56001c03194SChristoph Hellwig writeback_single_inode(inode, wbc); 5611da177e4SLinus Torvalds if (wbc->pages_skipped != pages_skipped) { 5621da177e4SLinus Torvalds /* 5631da177e4SLinus Torvalds * writeback is not making progress due to locked 5641da177e4SLinus Torvalds * buffers. Skip this inode for now. 5651da177e4SLinus Torvalds */ 566f57b9b7bSAndrew Morton redirty_tail(inode); 5671da177e4SLinus Torvalds } 5680f1b1fd8SDave Chinner spin_unlock(&inode->i_lock); 569a66979abSDave Chinner spin_unlock(&inode_wb_list_lock); 5701da177e4SLinus Torvalds iput(inode); 5714ffc8444SOGAWA Hirofumi cond_resched(); 572a66979abSDave Chinner spin_lock(&inode_wb_list_lock); 5738bc3be27SFengguang Wu if (wbc->nr_to_write <= 0) { 5748bc3be27SFengguang Wu wbc->more_io = 1; 575f11c9c5cSEdward Shishkin return 1; 5761da177e4SLinus Torvalds } 57703ba3782SJens Axboe if (!list_empty(&wb->b_more_io)) 5788bc3be27SFengguang Wu wbc->more_io = 1; 5798bc3be27SFengguang Wu } 580f11c9c5cSEdward Shishkin /* b_io is empty */ 581f11c9c5cSEdward Shishkin return 1; 582f11c9c5cSEdward Shishkin } 58338f21977SNick Piggin 5849c3a8ee8SChristoph Hellwig void writeback_inodes_wb(struct bdi_writeback *wb, 585f11c9c5cSEdward Shishkin struct writeback_control *wbc) 586f11c9c5cSEdward Shishkin { 587f11c9c5cSEdward Shishkin int ret = 0; 5889ecc2738SJens Axboe 5897624ee72SJan Kara if (!wbc->wb_start) 590f11c9c5cSEdward Shishkin wbc->wb_start = jiffies; /* livelock avoidance */ 591a66979abSDave Chinner spin_lock(&inode_wb_list_lock); 592f11c9c5cSEdward Shishkin if (!wbc->for_kupdate || list_empty(&wb->b_io)) 593f11c9c5cSEdward Shishkin queue_io(wb, wbc->older_than_this); 594f11c9c5cSEdward Shishkin 595f11c9c5cSEdward Shishkin while (!list_empty(&wb->b_io)) { 5967ccf19a8SNick Piggin struct inode *inode = wb_inode(wb->b_io.prev); 597f11c9c5cSEdward Shishkin struct super_block *sb = inode->i_sb; 598f11c9c5cSEdward Shishkin 599334132aeSChristoph Hellwig if (!pin_sb_for_writeback(sb)) { 600334132aeSChristoph Hellwig requeue_io(inode); 601d19de7edSChristoph Hellwig continue; 602334132aeSChristoph Hellwig } 603edadfb10SChristoph Hellwig ret = writeback_sb_inodes(sb, wb, wbc, false); 604d19de7edSChristoph Hellwig drop_super(sb); 605f11c9c5cSEdward Shishkin 606f11c9c5cSEdward Shishkin if (ret) 607f11c9c5cSEdward Shishkin break; 608f11c9c5cSEdward Shishkin } 609a66979abSDave Chinner spin_unlock(&inode_wb_list_lock); 61066f3b8e2SJens Axboe /* Leave any unwritten inodes on b_io */ 61166f3b8e2SJens Axboe } 61266f3b8e2SJens Axboe 613edadfb10SChristoph Hellwig static void __writeback_inodes_sb(struct super_block *sb, 614edadfb10SChristoph Hellwig struct bdi_writeback *wb, struct writeback_control *wbc) 615edadfb10SChristoph Hellwig { 616edadfb10SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 617edadfb10SChristoph Hellwig 618a66979abSDave Chinner spin_lock(&inode_wb_list_lock); 619edadfb10SChristoph Hellwig if (!wbc->for_kupdate || list_empty(&wb->b_io)) 620edadfb10SChristoph Hellwig queue_io(wb, wbc->older_than_this); 621edadfb10SChristoph Hellwig writeback_sb_inodes(sb, wb, wbc, true); 622a66979abSDave Chinner spin_unlock(&inode_wb_list_lock); 623edadfb10SChristoph Hellwig } 624edadfb10SChristoph Hellwig 62503ba3782SJens Axboe /* 62603ba3782SJens Axboe * The maximum number of pages to writeout in a single bdi flush/kupdate 62703ba3782SJens Axboe * operation. We do this so we don't hold I_SYNC against an inode for 62803ba3782SJens Axboe * enormous amounts of time, which would block a userspace task which has 62903ba3782SJens Axboe * been forced to throttle against that inode. Also, the code reevaluates 63003ba3782SJens Axboe * the dirty each time it has written this many pages. 63103ba3782SJens Axboe */ 63203ba3782SJens Axboe #define MAX_WRITEBACK_PAGES 1024 63303ba3782SJens Axboe 63403ba3782SJens Axboe static inline bool over_bground_thresh(void) 63503ba3782SJens Axboe { 63603ba3782SJens Axboe unsigned long background_thresh, dirty_thresh; 63703ba3782SJens Axboe 63816c4042fSWu Fengguang global_dirty_limits(&background_thresh, &dirty_thresh); 63903ba3782SJens Axboe 64003ba3782SJens Axboe return (global_page_state(NR_FILE_DIRTY) + 6414cbec4c8SWu Fengguang global_page_state(NR_UNSTABLE_NFS) > background_thresh); 64203ba3782SJens Axboe } 64303ba3782SJens Axboe 64403ba3782SJens Axboe /* 64503ba3782SJens Axboe * Explicit flushing or periodic writeback of "old" data. 64603ba3782SJens Axboe * 64703ba3782SJens Axboe * Define "old": the first time one of an inode's pages is dirtied, we mark the 64803ba3782SJens Axboe * dirtying-time in the inode's address_space. So this periodic writeback code 64903ba3782SJens Axboe * just walks the superblock inode list, writing back any inodes which are 65003ba3782SJens Axboe * older than a specific point in time. 65103ba3782SJens Axboe * 65203ba3782SJens Axboe * Try to run once per dirty_writeback_interval. But if a writeback event 65303ba3782SJens Axboe * takes longer than a dirty_writeback_interval interval, then leave a 65403ba3782SJens Axboe * one-second gap. 65503ba3782SJens Axboe * 65603ba3782SJens Axboe * older_than_this takes precedence over nr_to_write. So we'll only write back 65703ba3782SJens Axboe * all dirty pages if they are all attached to "old" mappings. 65803ba3782SJens Axboe */ 659c4a77a6cSJens Axboe static long wb_writeback(struct bdi_writeback *wb, 66083ba7b07SChristoph Hellwig struct wb_writeback_work *work) 66103ba3782SJens Axboe { 66203ba3782SJens Axboe struct writeback_control wbc = { 66383ba7b07SChristoph Hellwig .sync_mode = work->sync_mode, 6646e6938b6SWu Fengguang .tagged_writepages = work->tagged_writepages, 66503ba3782SJens Axboe .older_than_this = NULL, 66683ba7b07SChristoph Hellwig .for_kupdate = work->for_kupdate, 66783ba7b07SChristoph Hellwig .for_background = work->for_background, 66883ba7b07SChristoph Hellwig .range_cyclic = work->range_cyclic, 66903ba3782SJens Axboe }; 67003ba3782SJens Axboe unsigned long oldest_jif; 67103ba3782SJens Axboe long wrote = 0; 6726e6938b6SWu Fengguang long write_chunk = MAX_WRITEBACK_PAGES; 673a5989bdcSJan Kara struct inode *inode; 67403ba3782SJens Axboe 675c4a77a6cSJens Axboe if (!wbc.range_cyclic) { 676c4a77a6cSJens Axboe wbc.range_start = 0; 677c4a77a6cSJens Axboe wbc.range_end = LLONG_MAX; 678c4a77a6cSJens Axboe } 67903ba3782SJens Axboe 680b9543dacSJan Kara /* 681b9543dacSJan Kara * WB_SYNC_ALL mode does livelock avoidance by syncing dirty 682b9543dacSJan Kara * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX 683b9543dacSJan Kara * here avoids calling into writeback_inodes_wb() more than once. 684b9543dacSJan Kara * 685b9543dacSJan Kara * The intended call sequence for WB_SYNC_ALL writeback is: 686b9543dacSJan Kara * 687b9543dacSJan Kara * wb_writeback() 688b9543dacSJan Kara * __writeback_inodes_sb() <== called only once 689b9543dacSJan Kara * write_cache_pages() <== called once for each inode 690b9543dacSJan Kara * (quickly) tag currently dirty pages 691b9543dacSJan Kara * (maybe slowly) sync all tagged pages 692b9543dacSJan Kara */ 6936e6938b6SWu Fengguang if (wbc.sync_mode == WB_SYNC_ALL || wbc.tagged_writepages) 694b9543dacSJan Kara write_chunk = LONG_MAX; 695b9543dacSJan Kara 6967624ee72SJan Kara wbc.wb_start = jiffies; /* livelock avoidance */ 69703ba3782SJens Axboe for (;;) { 69803ba3782SJens Axboe /* 699d3ddec76SWu Fengguang * Stop writeback when nr_pages has been consumed 70003ba3782SJens Axboe */ 70183ba7b07SChristoph Hellwig if (work->nr_pages <= 0) 70203ba3782SJens Axboe break; 70303ba3782SJens Axboe 70403ba3782SJens Axboe /* 705aa373cf5SJan Kara * Background writeout and kupdate-style writeback may 706aa373cf5SJan Kara * run forever. Stop them if there is other work to do 707aa373cf5SJan Kara * so that e.g. sync can proceed. They'll be restarted 708aa373cf5SJan Kara * after the other works are all done. 709aa373cf5SJan Kara */ 710aa373cf5SJan Kara if ((work->for_background || work->for_kupdate) && 711aa373cf5SJan Kara !list_empty(&wb->bdi->work_list)) 712aa373cf5SJan Kara break; 713aa373cf5SJan Kara 714aa373cf5SJan Kara /* 715d3ddec76SWu Fengguang * For background writeout, stop when we are below the 716d3ddec76SWu Fengguang * background dirty threshold 71703ba3782SJens Axboe */ 71883ba7b07SChristoph Hellwig if (work->for_background && !over_bground_thresh()) 71903ba3782SJens Axboe break; 72003ba3782SJens Axboe 721ba9aa839SWu Fengguang if (work->for_kupdate) { 722ba9aa839SWu Fengguang oldest_jif = jiffies - 723ba9aa839SWu Fengguang msecs_to_jiffies(dirty_expire_interval * 10); 724ba9aa839SWu Fengguang wbc.older_than_this = &oldest_jif; 725ba9aa839SWu Fengguang } 726ba9aa839SWu Fengguang 72703ba3782SJens Axboe wbc.more_io = 0; 728b9543dacSJan Kara wbc.nr_to_write = write_chunk; 72903ba3782SJens Axboe wbc.pages_skipped = 0; 730cb9bd115SWu Fengguang wbc.inodes_written = 0; 731028c2dd1SDave Chinner 732028c2dd1SDave Chinner trace_wbc_writeback_start(&wbc, wb->bdi); 73383ba7b07SChristoph Hellwig if (work->sb) 73483ba7b07SChristoph Hellwig __writeback_inodes_sb(work->sb, wb, &wbc); 735edadfb10SChristoph Hellwig else 73603ba3782SJens Axboe writeback_inodes_wb(wb, &wbc); 737028c2dd1SDave Chinner trace_wbc_writeback_written(&wbc, wb->bdi); 738028c2dd1SDave Chinner 739b9543dacSJan Kara work->nr_pages -= write_chunk - wbc.nr_to_write; 740b9543dacSJan Kara wrote += write_chunk - wbc.nr_to_write; 74103ba3782SJens Axboe 74203ba3782SJens Axboe /* 743e6fb6da2SWu Fengguang * Did we write something? Try for more 744e6fb6da2SWu Fengguang * 745e6fb6da2SWu Fengguang * Dirty inodes are moved to b_io for writeback in batches. 746e6fb6da2SWu Fengguang * The completion of the current batch does not necessarily 747e6fb6da2SWu Fengguang * mean the overall work is done. So we keep looping as long 748e6fb6da2SWu Fengguang * as made some progress on cleaning pages or inodes. 74903ba3782SJens Axboe */ 750e6fb6da2SWu Fengguang if (wbc.nr_to_write < write_chunk) 75171fd05a8SJens Axboe continue; 752cb9bd115SWu Fengguang if (wbc.inodes_written) 753cb9bd115SWu Fengguang continue; 75471fd05a8SJens Axboe /* 755e6fb6da2SWu Fengguang * No more inodes for IO, bail 75671fd05a8SJens Axboe */ 75771fd05a8SJens Axboe if (!wbc.more_io) 75871fd05a8SJens Axboe break; 75971fd05a8SJens Axboe /* 760a5989bdcSJan Kara * Nothing written. Wait for some inode to 761a5989bdcSJan Kara * become available for writeback. Otherwise 762a5989bdcSJan Kara * we'll just busyloop. 763a5989bdcSJan Kara */ 764a66979abSDave Chinner spin_lock(&inode_wb_list_lock); 765a5989bdcSJan Kara if (!list_empty(&wb->b_more_io)) { 7667ccf19a8SNick Piggin inode = wb_inode(wb->b_more_io.prev); 767028c2dd1SDave Chinner trace_wbc_writeback_wait(&wbc, wb->bdi); 768250df6edSDave Chinner spin_lock(&inode->i_lock); 769a5989bdcSJan Kara inode_wait_for_writeback(inode); 770250df6edSDave Chinner spin_unlock(&inode->i_lock); 771a5989bdcSJan Kara } 772a66979abSDave Chinner spin_unlock(&inode_wb_list_lock); 77303ba3782SJens Axboe } 77403ba3782SJens Axboe 77503ba3782SJens Axboe return wrote; 77603ba3782SJens Axboe } 77703ba3782SJens Axboe 77803ba3782SJens Axboe /* 77983ba7b07SChristoph Hellwig * Return the next wb_writeback_work struct that hasn't been processed yet. 78003ba3782SJens Axboe */ 78183ba7b07SChristoph Hellwig static struct wb_writeback_work * 78208852b6dSMinchan Kim get_next_work_item(struct backing_dev_info *bdi) 78303ba3782SJens Axboe { 78483ba7b07SChristoph Hellwig struct wb_writeback_work *work = NULL; 78503ba3782SJens Axboe 7866467716aSArtem Bityutskiy spin_lock_bh(&bdi->wb_lock); 78783ba7b07SChristoph Hellwig if (!list_empty(&bdi->work_list)) { 78883ba7b07SChristoph Hellwig work = list_entry(bdi->work_list.next, 78983ba7b07SChristoph Hellwig struct wb_writeback_work, list); 79083ba7b07SChristoph Hellwig list_del_init(&work->list); 79103ba3782SJens Axboe } 7926467716aSArtem Bityutskiy spin_unlock_bh(&bdi->wb_lock); 79383ba7b07SChristoph Hellwig return work; 79403ba3782SJens Axboe } 79503ba3782SJens Axboe 796cdf01dd5SLinus Torvalds /* 797cdf01dd5SLinus Torvalds * Add in the number of potentially dirty inodes, because each inode 798cdf01dd5SLinus Torvalds * write can dirty pagecache in the underlying blockdev. 799cdf01dd5SLinus Torvalds */ 800cdf01dd5SLinus Torvalds static unsigned long get_nr_dirty_pages(void) 801cdf01dd5SLinus Torvalds { 802cdf01dd5SLinus Torvalds return global_page_state(NR_FILE_DIRTY) + 803cdf01dd5SLinus Torvalds global_page_state(NR_UNSTABLE_NFS) + 804cdf01dd5SLinus Torvalds get_nr_dirty_inodes(); 805cdf01dd5SLinus Torvalds } 806cdf01dd5SLinus Torvalds 8076585027aSJan Kara static long wb_check_background_flush(struct bdi_writeback *wb) 8086585027aSJan Kara { 8096585027aSJan Kara if (over_bground_thresh()) { 8106585027aSJan Kara 8116585027aSJan Kara struct wb_writeback_work work = { 8126585027aSJan Kara .nr_pages = LONG_MAX, 8136585027aSJan Kara .sync_mode = WB_SYNC_NONE, 8146585027aSJan Kara .for_background = 1, 8156585027aSJan Kara .range_cyclic = 1, 8166585027aSJan Kara }; 8176585027aSJan Kara 8186585027aSJan Kara return wb_writeback(wb, &work); 8196585027aSJan Kara } 8206585027aSJan Kara 8216585027aSJan Kara return 0; 8226585027aSJan Kara } 8236585027aSJan Kara 82403ba3782SJens Axboe static long wb_check_old_data_flush(struct bdi_writeback *wb) 82503ba3782SJens Axboe { 82603ba3782SJens Axboe unsigned long expired; 82703ba3782SJens Axboe long nr_pages; 82803ba3782SJens Axboe 82969b62d01SJens Axboe /* 83069b62d01SJens Axboe * When set to zero, disable periodic writeback 83169b62d01SJens Axboe */ 83269b62d01SJens Axboe if (!dirty_writeback_interval) 83369b62d01SJens Axboe return 0; 83469b62d01SJens Axboe 83503ba3782SJens Axboe expired = wb->last_old_flush + 83603ba3782SJens Axboe msecs_to_jiffies(dirty_writeback_interval * 10); 83703ba3782SJens Axboe if (time_before(jiffies, expired)) 83803ba3782SJens Axboe return 0; 83903ba3782SJens Axboe 84003ba3782SJens Axboe wb->last_old_flush = jiffies; 841cdf01dd5SLinus Torvalds nr_pages = get_nr_dirty_pages(); 84203ba3782SJens Axboe 843c4a77a6cSJens Axboe if (nr_pages) { 84483ba7b07SChristoph Hellwig struct wb_writeback_work work = { 845c4a77a6cSJens Axboe .nr_pages = nr_pages, 846c4a77a6cSJens Axboe .sync_mode = WB_SYNC_NONE, 847c4a77a6cSJens Axboe .for_kupdate = 1, 848c4a77a6cSJens Axboe .range_cyclic = 1, 849c4a77a6cSJens Axboe }; 850c4a77a6cSJens Axboe 85183ba7b07SChristoph Hellwig return wb_writeback(wb, &work); 852c4a77a6cSJens Axboe } 85303ba3782SJens Axboe 85403ba3782SJens Axboe return 0; 85503ba3782SJens Axboe } 85603ba3782SJens Axboe 85703ba3782SJens Axboe /* 85803ba3782SJens Axboe * Retrieve work items and do the writeback they describe 85903ba3782SJens Axboe */ 86003ba3782SJens Axboe long wb_do_writeback(struct bdi_writeback *wb, int force_wait) 86103ba3782SJens Axboe { 86203ba3782SJens Axboe struct backing_dev_info *bdi = wb->bdi; 86383ba7b07SChristoph Hellwig struct wb_writeback_work *work; 864c4a77a6cSJens Axboe long wrote = 0; 86503ba3782SJens Axboe 86681d73a32SJan Kara set_bit(BDI_writeback_running, &wb->bdi->state); 86708852b6dSMinchan Kim while ((work = get_next_work_item(bdi)) != NULL) { 86803ba3782SJens Axboe /* 86903ba3782SJens Axboe * Override sync mode, in case we must wait for completion 87083ba7b07SChristoph Hellwig * because this thread is exiting now. 87103ba3782SJens Axboe */ 87203ba3782SJens Axboe if (force_wait) 87383ba7b07SChristoph Hellwig work->sync_mode = WB_SYNC_ALL; 87483ba7b07SChristoph Hellwig 875455b2864SDave Chinner trace_writeback_exec(bdi, work); 876455b2864SDave Chinner 87783ba7b07SChristoph Hellwig wrote += wb_writeback(wb, work); 87803ba3782SJens Axboe 87903ba3782SJens Axboe /* 88083ba7b07SChristoph Hellwig * Notify the caller of completion if this is a synchronous 88183ba7b07SChristoph Hellwig * work item, otherwise just free it. 88203ba3782SJens Axboe */ 88383ba7b07SChristoph Hellwig if (work->done) 88483ba7b07SChristoph Hellwig complete(work->done); 88583ba7b07SChristoph Hellwig else 88683ba7b07SChristoph Hellwig kfree(work); 88703ba3782SJens Axboe } 88803ba3782SJens Axboe 88903ba3782SJens Axboe /* 89003ba3782SJens Axboe * Check for periodic writeback, kupdated() style 89103ba3782SJens Axboe */ 89203ba3782SJens Axboe wrote += wb_check_old_data_flush(wb); 8936585027aSJan Kara wrote += wb_check_background_flush(wb); 89481d73a32SJan Kara clear_bit(BDI_writeback_running, &wb->bdi->state); 89503ba3782SJens Axboe 89603ba3782SJens Axboe return wrote; 89703ba3782SJens Axboe } 89803ba3782SJens Axboe 89903ba3782SJens Axboe /* 90003ba3782SJens Axboe * Handle writeback of dirty data for the device backed by this bdi. Also 90103ba3782SJens Axboe * wakes up periodically and does kupdated style flushing. 90203ba3782SJens Axboe */ 90308243900SChristoph Hellwig int bdi_writeback_thread(void *data) 90403ba3782SJens Axboe { 90508243900SChristoph Hellwig struct bdi_writeback *wb = data; 90608243900SChristoph Hellwig struct backing_dev_info *bdi = wb->bdi; 90703ba3782SJens Axboe long pages_written; 90803ba3782SJens Axboe 909766f9164SPeter Zijlstra current->flags |= PF_SWAPWRITE; 91008243900SChristoph Hellwig set_freezable(); 911ecd58403SArtem Bityutskiy wb->last_active = jiffies; 91203ba3782SJens Axboe 91303ba3782SJens Axboe /* 91408243900SChristoph Hellwig * Our parent may run at a different priority, just set us to normal 91503ba3782SJens Axboe */ 91608243900SChristoph Hellwig set_user_nice(current, 0); 91708243900SChristoph Hellwig 918455b2864SDave Chinner trace_writeback_thread_start(bdi); 919455b2864SDave Chinner 92003ba3782SJens Axboe while (!kthread_should_stop()) { 9216467716aSArtem Bityutskiy /* 9226467716aSArtem Bityutskiy * Remove own delayed wake-up timer, since we are already awake 9236467716aSArtem Bityutskiy * and we'll take care of the preriodic write-back. 9246467716aSArtem Bityutskiy */ 9256467716aSArtem Bityutskiy del_timer(&wb->wakeup_timer); 9266467716aSArtem Bityutskiy 92703ba3782SJens Axboe pages_written = wb_do_writeback(wb, 0); 92803ba3782SJens Axboe 929455b2864SDave Chinner trace_writeback_pages_written(pages_written); 930455b2864SDave Chinner 93103ba3782SJens Axboe if (pages_written) 932ecd58403SArtem Bityutskiy wb->last_active = jiffies; 93303ba3782SJens Axboe 934297252c8SArtem Bityutskiy set_current_state(TASK_INTERRUPTIBLE); 935b76b4014SJ. Bruce Fields if (!list_empty(&bdi->work_list) || kthread_should_stop()) { 936297252c8SArtem Bityutskiy __set_current_state(TASK_RUNNING); 937297252c8SArtem Bityutskiy continue; 93803ba3782SJens Axboe } 93903ba3782SJens Axboe 940253c34e9SArtem Bityutskiy if (wb_has_dirty_io(wb) && dirty_writeback_interval) 941fff5b85aSArtem Bityutskiy schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10)); 942253c34e9SArtem Bityutskiy else { 943253c34e9SArtem Bityutskiy /* 944253c34e9SArtem Bityutskiy * We have nothing to do, so can go sleep without any 945253c34e9SArtem Bityutskiy * timeout and save power. When a work is queued or 946253c34e9SArtem Bityutskiy * something is made dirty - we will be woken up. 947253c34e9SArtem Bityutskiy */ 94869b62d01SJens Axboe schedule(); 949f9eadbbdSJens Axboe } 95069b62d01SJens Axboe 95103ba3782SJens Axboe try_to_freeze(); 95203ba3782SJens Axboe } 95303ba3782SJens Axboe 954fff5b85aSArtem Bityutskiy /* Flush any work that raced with us exiting */ 95508243900SChristoph Hellwig if (!list_empty(&bdi->work_list)) 95608243900SChristoph Hellwig wb_do_writeback(wb, 1); 957455b2864SDave Chinner 958455b2864SDave Chinner trace_writeback_thread_stop(bdi); 95903ba3782SJens Axboe return 0; 96003ba3782SJens Axboe } 96103ba3782SJens Axboe 96208243900SChristoph Hellwig 96303ba3782SJens Axboe /* 96403ba3782SJens Axboe * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back 96503ba3782SJens Axboe * the whole world. 96603ba3782SJens Axboe */ 96703ba3782SJens Axboe void wakeup_flusher_threads(long nr_pages) 96803ba3782SJens Axboe { 969b8c2f347SChristoph Hellwig struct backing_dev_info *bdi; 970b8c2f347SChristoph Hellwig 97183ba7b07SChristoph Hellwig if (!nr_pages) { 97283ba7b07SChristoph Hellwig nr_pages = global_page_state(NR_FILE_DIRTY) + 97303ba3782SJens Axboe global_page_state(NR_UNSTABLE_NFS); 974b8c2f347SChristoph Hellwig } 975b8c2f347SChristoph Hellwig 976b8c2f347SChristoph Hellwig rcu_read_lock(); 977b8c2f347SChristoph Hellwig list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { 978b8c2f347SChristoph Hellwig if (!bdi_has_dirty_io(bdi)) 979b8c2f347SChristoph Hellwig continue; 9806585027aSJan Kara __bdi_start_writeback(bdi, nr_pages, false); 981b8c2f347SChristoph Hellwig } 982b8c2f347SChristoph Hellwig rcu_read_unlock(); 98303ba3782SJens Axboe } 98403ba3782SJens Axboe 98503ba3782SJens Axboe static noinline void block_dump___mark_inode_dirty(struct inode *inode) 98603ba3782SJens Axboe { 98703ba3782SJens Axboe if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { 98803ba3782SJens Axboe struct dentry *dentry; 98903ba3782SJens Axboe const char *name = "?"; 99003ba3782SJens Axboe 99103ba3782SJens Axboe dentry = d_find_alias(inode); 99203ba3782SJens Axboe if (dentry) { 99303ba3782SJens Axboe spin_lock(&dentry->d_lock); 99403ba3782SJens Axboe name = (const char *) dentry->d_name.name; 99503ba3782SJens Axboe } 99603ba3782SJens Axboe printk(KERN_DEBUG 99703ba3782SJens Axboe "%s(%d): dirtied inode %lu (%s) on %s\n", 99803ba3782SJens Axboe current->comm, task_pid_nr(current), inode->i_ino, 99903ba3782SJens Axboe name, inode->i_sb->s_id); 100003ba3782SJens Axboe if (dentry) { 100103ba3782SJens Axboe spin_unlock(&dentry->d_lock); 100203ba3782SJens Axboe dput(dentry); 100303ba3782SJens Axboe } 100403ba3782SJens Axboe } 100503ba3782SJens Axboe } 100603ba3782SJens Axboe 100703ba3782SJens Axboe /** 100803ba3782SJens Axboe * __mark_inode_dirty - internal function 100903ba3782SJens Axboe * @inode: inode to mark 101003ba3782SJens Axboe * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) 101103ba3782SJens Axboe * Mark an inode as dirty. Callers should use mark_inode_dirty or 101203ba3782SJens Axboe * mark_inode_dirty_sync. 101303ba3782SJens Axboe * 101403ba3782SJens Axboe * Put the inode on the super block's dirty list. 101503ba3782SJens Axboe * 101603ba3782SJens Axboe * CAREFUL! We mark it dirty unconditionally, but move it onto the 101703ba3782SJens Axboe * dirty list only if it is hashed or if it refers to a blockdev. 101803ba3782SJens Axboe * If it was not hashed, it will never be added to the dirty list 101903ba3782SJens Axboe * even if it is later hashed, as it will have been marked dirty already. 102003ba3782SJens Axboe * 102103ba3782SJens Axboe * In short, make sure you hash any inodes _before_ you start marking 102203ba3782SJens Axboe * them dirty. 102303ba3782SJens Axboe * 102403ba3782SJens Axboe * Note that for blockdevs, inode->dirtied_when represents the dirtying time of 102503ba3782SJens Axboe * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of 102603ba3782SJens Axboe * the kernel-internal blockdev inode represents the dirtying time of the 102703ba3782SJens Axboe * blockdev's pages. This is why for I_DIRTY_PAGES we always use 102803ba3782SJens Axboe * page->mapping->host, so the page-dirtying time is recorded in the internal 102903ba3782SJens Axboe * blockdev inode. 103003ba3782SJens Axboe */ 103103ba3782SJens Axboe void __mark_inode_dirty(struct inode *inode, int flags) 103203ba3782SJens Axboe { 103303ba3782SJens Axboe struct super_block *sb = inode->i_sb; 1034253c34e9SArtem Bityutskiy struct backing_dev_info *bdi = NULL; 103503ba3782SJens Axboe 103603ba3782SJens Axboe /* 103703ba3782SJens Axboe * Don't do this for I_DIRTY_PAGES - that doesn't actually 103803ba3782SJens Axboe * dirty the inode itself 103903ba3782SJens Axboe */ 104003ba3782SJens Axboe if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { 104103ba3782SJens Axboe if (sb->s_op->dirty_inode) 1042aa385729SChristoph Hellwig sb->s_op->dirty_inode(inode, flags); 104303ba3782SJens Axboe } 104403ba3782SJens Axboe 104503ba3782SJens Axboe /* 104603ba3782SJens Axboe * make sure that changes are seen by all cpus before we test i_state 104703ba3782SJens Axboe * -- mikulas 104803ba3782SJens Axboe */ 104903ba3782SJens Axboe smp_mb(); 105003ba3782SJens Axboe 105103ba3782SJens Axboe /* avoid the locking if we can */ 105203ba3782SJens Axboe if ((inode->i_state & flags) == flags) 105303ba3782SJens Axboe return; 105403ba3782SJens Axboe 105503ba3782SJens Axboe if (unlikely(block_dump)) 105603ba3782SJens Axboe block_dump___mark_inode_dirty(inode); 105703ba3782SJens Axboe 1058250df6edSDave Chinner spin_lock(&inode->i_lock); 105903ba3782SJens Axboe if ((inode->i_state & flags) != flags) { 106003ba3782SJens Axboe const int was_dirty = inode->i_state & I_DIRTY; 106103ba3782SJens Axboe 106203ba3782SJens Axboe inode->i_state |= flags; 106303ba3782SJens Axboe 106403ba3782SJens Axboe /* 106503ba3782SJens Axboe * If the inode is being synced, just update its dirty state. 106603ba3782SJens Axboe * The unlocker will place the inode on the appropriate 106703ba3782SJens Axboe * superblock list, based upon its state. 106803ba3782SJens Axboe */ 106903ba3782SJens Axboe if (inode->i_state & I_SYNC) 1070250df6edSDave Chinner goto out_unlock_inode; 107103ba3782SJens Axboe 107203ba3782SJens Axboe /* 107303ba3782SJens Axboe * Only add valid (hashed) inodes to the superblock's 107403ba3782SJens Axboe * dirty list. Add blockdev inodes as well. 107503ba3782SJens Axboe */ 107603ba3782SJens Axboe if (!S_ISBLK(inode->i_mode)) { 10771d3382cbSAl Viro if (inode_unhashed(inode)) 1078250df6edSDave Chinner goto out_unlock_inode; 107903ba3782SJens Axboe } 1080a4ffdde6SAl Viro if (inode->i_state & I_FREEING) 1081250df6edSDave Chinner goto out_unlock_inode; 108203ba3782SJens Axboe 108303ba3782SJens Axboe /* 108403ba3782SJens Axboe * If the inode was already on b_dirty/b_io/b_more_io, don't 108503ba3782SJens Axboe * reposition it (that would break b_dirty time-ordering). 108603ba3782SJens Axboe */ 108703ba3782SJens Axboe if (!was_dirty) { 1088a66979abSDave Chinner bool wakeup_bdi = false; 1089253c34e9SArtem Bityutskiy bdi = inode_to_bdi(inode); 1090500b067cSJens Axboe 1091253c34e9SArtem Bityutskiy if (bdi_cap_writeback_dirty(bdi)) { 1092253c34e9SArtem Bityutskiy WARN(!test_bit(BDI_registered, &bdi->state), 1093253c34e9SArtem Bityutskiy "bdi-%s not registered\n", bdi->name); 1094253c34e9SArtem Bityutskiy 1095253c34e9SArtem Bityutskiy /* 1096253c34e9SArtem Bityutskiy * If this is the first dirty inode for this 1097253c34e9SArtem Bityutskiy * bdi, we have to wake-up the corresponding 1098253c34e9SArtem Bityutskiy * bdi thread to make sure background 1099253c34e9SArtem Bityutskiy * write-back happens later. 1100253c34e9SArtem Bityutskiy */ 1101253c34e9SArtem Bityutskiy if (!wb_has_dirty_io(&bdi->wb)) 1102253c34e9SArtem Bityutskiy wakeup_bdi = true; 1103500b067cSJens Axboe } 110403ba3782SJens Axboe 1105a66979abSDave Chinner spin_unlock(&inode->i_lock); 1106a66979abSDave Chinner spin_lock(&inode_wb_list_lock); 110703ba3782SJens Axboe inode->dirtied_when = jiffies; 11087ccf19a8SNick Piggin list_move(&inode->i_wb_list, &bdi->wb.b_dirty); 1109a66979abSDave Chinner spin_unlock(&inode_wb_list_lock); 1110253c34e9SArtem Bityutskiy 1111253c34e9SArtem Bityutskiy if (wakeup_bdi) 11126467716aSArtem Bityutskiy bdi_wakeup_thread_delayed(bdi); 1113a66979abSDave Chinner return; 1114a66979abSDave Chinner } 1115a66979abSDave Chinner } 1116a66979abSDave Chinner out_unlock_inode: 1117a66979abSDave Chinner spin_unlock(&inode->i_lock); 1118a66979abSDave Chinner 111903ba3782SJens Axboe } 112003ba3782SJens Axboe EXPORT_SYMBOL(__mark_inode_dirty); 112103ba3782SJens Axboe 112266f3b8e2SJens Axboe /* 112366f3b8e2SJens Axboe * Write out a superblock's list of dirty inodes. A wait will be performed 112466f3b8e2SJens Axboe * upon no inodes, all inodes or the final one, depending upon sync_mode. 112566f3b8e2SJens Axboe * 112666f3b8e2SJens Axboe * If older_than_this is non-NULL, then only write out inodes which 112766f3b8e2SJens Axboe * had their first dirtying at a time earlier than *older_than_this. 112866f3b8e2SJens Axboe * 112966f3b8e2SJens Axboe * If `bdi' is non-zero then we're being asked to writeback a specific queue. 113066f3b8e2SJens Axboe * This function assumes that the blockdev superblock's inodes are backed by 113166f3b8e2SJens Axboe * a variety of queues, so all inodes are searched. For other superblocks, 113266f3b8e2SJens Axboe * assume that all inodes are backed by the same queue. 113366f3b8e2SJens Axboe * 113466f3b8e2SJens Axboe * The inodes to be written are parked on bdi->b_io. They are moved back onto 113566f3b8e2SJens Axboe * bdi->b_dirty as they are selected for writing. This way, none can be missed 113666f3b8e2SJens Axboe * on the writer throttling path, and we get decent balancing between many 113766f3b8e2SJens Axboe * throttled threads: we don't want them all piling up on inode_sync_wait. 113866f3b8e2SJens Axboe */ 1139b6e51316SJens Axboe static void wait_sb_inodes(struct super_block *sb) 114066f3b8e2SJens Axboe { 114138f21977SNick Piggin struct inode *inode, *old_inode = NULL; 114238f21977SNick Piggin 114303ba3782SJens Axboe /* 114403ba3782SJens Axboe * We need to be protected against the filesystem going from 114503ba3782SJens Axboe * r/o to r/w or vice versa. 114603ba3782SJens Axboe */ 1147b6e51316SJens Axboe WARN_ON(!rwsem_is_locked(&sb->s_umount)); 114803ba3782SJens Axboe 114955fa6091SDave Chinner spin_lock(&inode_sb_list_lock); 115066f3b8e2SJens Axboe 115138f21977SNick Piggin /* 115238f21977SNick Piggin * Data integrity sync. Must wait for all pages under writeback, 115338f21977SNick Piggin * because there may have been pages dirtied before our sync 115438f21977SNick Piggin * call, but which had writeout started before we write it out. 115538f21977SNick Piggin * In which case, the inode may not be on the dirty list, but 115638f21977SNick Piggin * we still have to wait for that writeout. 115738f21977SNick Piggin */ 1158b6e51316SJens Axboe list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { 1159250df6edSDave Chinner struct address_space *mapping = inode->i_mapping; 116038f21977SNick Piggin 1161250df6edSDave Chinner spin_lock(&inode->i_lock); 1162250df6edSDave Chinner if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) || 1163250df6edSDave Chinner (mapping->nrpages == 0)) { 1164250df6edSDave Chinner spin_unlock(&inode->i_lock); 116538f21977SNick Piggin continue; 1166250df6edSDave Chinner } 116738f21977SNick Piggin __iget(inode); 1168250df6edSDave Chinner spin_unlock(&inode->i_lock); 116955fa6091SDave Chinner spin_unlock(&inode_sb_list_lock); 117055fa6091SDave Chinner 117138f21977SNick Piggin /* 117255fa6091SDave Chinner * We hold a reference to 'inode' so it couldn't have been 117355fa6091SDave Chinner * removed from s_inodes list while we dropped the 117455fa6091SDave Chinner * inode_sb_list_lock. We cannot iput the inode now as we can 117555fa6091SDave Chinner * be holding the last reference and we cannot iput it under 117655fa6091SDave Chinner * inode_sb_list_lock. So we keep the reference and iput it 117755fa6091SDave Chinner * later. 117838f21977SNick Piggin */ 117938f21977SNick Piggin iput(old_inode); 118038f21977SNick Piggin old_inode = inode; 118138f21977SNick Piggin 118238f21977SNick Piggin filemap_fdatawait(mapping); 118338f21977SNick Piggin 118438f21977SNick Piggin cond_resched(); 118538f21977SNick Piggin 118655fa6091SDave Chinner spin_lock(&inode_sb_list_lock); 118738f21977SNick Piggin } 118855fa6091SDave Chinner spin_unlock(&inode_sb_list_lock); 118938f21977SNick Piggin iput(old_inode); 119066f3b8e2SJens Axboe } 11911da177e4SLinus Torvalds 1192d8a8559cSJens Axboe /** 11933259f8beSChris Mason * writeback_inodes_sb_nr - writeback dirty inodes from given super_block 1194d8a8559cSJens Axboe * @sb: the superblock 11953259f8beSChris Mason * @nr: the number of pages to write 11961da177e4SLinus Torvalds * 1197d8a8559cSJens Axboe * Start writeback on some inodes on this super_block. No guarantees are made 1198d8a8559cSJens Axboe * on how many (if any) will be written, and this function does not wait 11993259f8beSChris Mason * for IO completion of submitted IO. 12001da177e4SLinus Torvalds */ 12013259f8beSChris Mason void writeback_inodes_sb_nr(struct super_block *sb, unsigned long nr) 12021da177e4SLinus Torvalds { 120383ba7b07SChristoph Hellwig DECLARE_COMPLETION_ONSTACK(done); 120483ba7b07SChristoph Hellwig struct wb_writeback_work work = { 12053c4d7165SChristoph Hellwig .sb = sb, 12063c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_NONE, 12076e6938b6SWu Fengguang .tagged_writepages = 1, 120883ba7b07SChristoph Hellwig .done = &done, 12093259f8beSChris Mason .nr_pages = nr, 12103c4d7165SChristoph Hellwig }; 12110e3c9a22SJens Axboe 1212cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 121383ba7b07SChristoph Hellwig bdi_queue_work(sb->s_bdi, &work); 121483ba7b07SChristoph Hellwig wait_for_completion(&done); 12151da177e4SLinus Torvalds } 12163259f8beSChris Mason EXPORT_SYMBOL(writeback_inodes_sb_nr); 12173259f8beSChris Mason 12183259f8beSChris Mason /** 12193259f8beSChris Mason * writeback_inodes_sb - writeback dirty inodes from given super_block 12203259f8beSChris Mason * @sb: the superblock 12213259f8beSChris Mason * 12223259f8beSChris Mason * Start writeback on some inodes on this super_block. No guarantees are made 12233259f8beSChris Mason * on how many (if any) will be written, and this function does not wait 12243259f8beSChris Mason * for IO completion of submitted IO. 12253259f8beSChris Mason */ 12263259f8beSChris Mason void writeback_inodes_sb(struct super_block *sb) 12273259f8beSChris Mason { 1228925d169fSLinus Torvalds return writeback_inodes_sb_nr(sb, get_nr_dirty_pages()); 12293259f8beSChris Mason } 1230d8a8559cSJens Axboe EXPORT_SYMBOL(writeback_inodes_sb); 1231d8a8559cSJens Axboe 1232d8a8559cSJens Axboe /** 123317bd55d0SEric Sandeen * writeback_inodes_sb_if_idle - start writeback if none underway 123417bd55d0SEric Sandeen * @sb: the superblock 123517bd55d0SEric Sandeen * 123617bd55d0SEric Sandeen * Invoke writeback_inodes_sb if no writeback is currently underway. 123717bd55d0SEric Sandeen * Returns 1 if writeback was started, 0 if not. 123817bd55d0SEric Sandeen */ 123917bd55d0SEric Sandeen int writeback_inodes_sb_if_idle(struct super_block *sb) 124017bd55d0SEric Sandeen { 124117bd55d0SEric Sandeen if (!writeback_in_progress(sb->s_bdi)) { 1242cf37e972SChristoph Hellwig down_read(&sb->s_umount); 124317bd55d0SEric Sandeen writeback_inodes_sb(sb); 1244cf37e972SChristoph Hellwig up_read(&sb->s_umount); 124517bd55d0SEric Sandeen return 1; 124617bd55d0SEric Sandeen } else 124717bd55d0SEric Sandeen return 0; 124817bd55d0SEric Sandeen } 124917bd55d0SEric Sandeen EXPORT_SYMBOL(writeback_inodes_sb_if_idle); 125017bd55d0SEric Sandeen 125117bd55d0SEric Sandeen /** 12523259f8beSChris Mason * writeback_inodes_sb_if_idle - start writeback if none underway 12533259f8beSChris Mason * @sb: the superblock 12543259f8beSChris Mason * @nr: the number of pages to write 12553259f8beSChris Mason * 12563259f8beSChris Mason * Invoke writeback_inodes_sb if no writeback is currently underway. 12573259f8beSChris Mason * Returns 1 if writeback was started, 0 if not. 12583259f8beSChris Mason */ 12593259f8beSChris Mason int writeback_inodes_sb_nr_if_idle(struct super_block *sb, 12603259f8beSChris Mason unsigned long nr) 12613259f8beSChris Mason { 12623259f8beSChris Mason if (!writeback_in_progress(sb->s_bdi)) { 12633259f8beSChris Mason down_read(&sb->s_umount); 12643259f8beSChris Mason writeback_inodes_sb_nr(sb, nr); 12653259f8beSChris Mason up_read(&sb->s_umount); 12663259f8beSChris Mason return 1; 12673259f8beSChris Mason } else 12683259f8beSChris Mason return 0; 12693259f8beSChris Mason } 12703259f8beSChris Mason EXPORT_SYMBOL(writeback_inodes_sb_nr_if_idle); 12713259f8beSChris Mason 12723259f8beSChris Mason /** 1273d8a8559cSJens Axboe * sync_inodes_sb - sync sb inode pages 1274d8a8559cSJens Axboe * @sb: the superblock 1275d8a8559cSJens Axboe * 1276d8a8559cSJens Axboe * This function writes and waits on any dirty inode belonging to this 1277cb9ef8d5SStefan Hajnoczi * super_block. 1278d8a8559cSJens Axboe */ 1279b6e51316SJens Axboe void sync_inodes_sb(struct super_block *sb) 1280d8a8559cSJens Axboe { 128183ba7b07SChristoph Hellwig DECLARE_COMPLETION_ONSTACK(done); 128283ba7b07SChristoph Hellwig struct wb_writeback_work work = { 12833c4d7165SChristoph Hellwig .sb = sb, 12843c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_ALL, 12853c4d7165SChristoph Hellwig .nr_pages = LONG_MAX, 12863c4d7165SChristoph Hellwig .range_cyclic = 0, 128783ba7b07SChristoph Hellwig .done = &done, 12883c4d7165SChristoph Hellwig }; 12893c4d7165SChristoph Hellwig 1290cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 1291cf37e972SChristoph Hellwig 129283ba7b07SChristoph Hellwig bdi_queue_work(sb->s_bdi, &work); 129383ba7b07SChristoph Hellwig wait_for_completion(&done); 129483ba7b07SChristoph Hellwig 1295b6e51316SJens Axboe wait_sb_inodes(sb); 1296d8a8559cSJens Axboe } 1297d8a8559cSJens Axboe EXPORT_SYMBOL(sync_inodes_sb); 12981da177e4SLinus Torvalds 12991da177e4SLinus Torvalds /** 13001da177e4SLinus Torvalds * write_inode_now - write an inode to disk 13011da177e4SLinus Torvalds * @inode: inode to write to disk 13021da177e4SLinus Torvalds * @sync: whether the write should be synchronous or not 13031da177e4SLinus Torvalds * 13047f04c26dSAndrea Arcangeli * This function commits an inode to disk immediately if it is dirty. This is 13057f04c26dSAndrea Arcangeli * primarily needed by knfsd. 13067f04c26dSAndrea Arcangeli * 13077f04c26dSAndrea Arcangeli * The caller must either have a ref on the inode or must have set I_WILL_FREE. 13081da177e4SLinus Torvalds */ 13091da177e4SLinus Torvalds int write_inode_now(struct inode *inode, int sync) 13101da177e4SLinus Torvalds { 13111da177e4SLinus Torvalds int ret; 13121da177e4SLinus Torvalds struct writeback_control wbc = { 13131da177e4SLinus Torvalds .nr_to_write = LONG_MAX, 131418914b18SMike Galbraith .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, 1315111ebb6eSOGAWA Hirofumi .range_start = 0, 1316111ebb6eSOGAWA Hirofumi .range_end = LLONG_MAX, 13171da177e4SLinus Torvalds }; 13181da177e4SLinus Torvalds 13191da177e4SLinus Torvalds if (!mapping_cap_writeback_dirty(inode->i_mapping)) 132049364ce2SAndrew Morton wbc.nr_to_write = 0; 13211da177e4SLinus Torvalds 13221da177e4SLinus Torvalds might_sleep(); 1323a66979abSDave Chinner spin_lock(&inode_wb_list_lock); 13240f1b1fd8SDave Chinner spin_lock(&inode->i_lock); 132501c03194SChristoph Hellwig ret = writeback_single_inode(inode, &wbc); 13260f1b1fd8SDave Chinner spin_unlock(&inode->i_lock); 1327a66979abSDave Chinner spin_unlock(&inode_wb_list_lock); 13281da177e4SLinus Torvalds if (sync) 13291c0eeaf5SJoern Engel inode_sync_wait(inode); 13301da177e4SLinus Torvalds return ret; 13311da177e4SLinus Torvalds } 13321da177e4SLinus Torvalds EXPORT_SYMBOL(write_inode_now); 13331da177e4SLinus Torvalds 13341da177e4SLinus Torvalds /** 13351da177e4SLinus Torvalds * sync_inode - write an inode and its pages to disk. 13361da177e4SLinus Torvalds * @inode: the inode to sync 13371da177e4SLinus Torvalds * @wbc: controls the writeback mode 13381da177e4SLinus Torvalds * 13391da177e4SLinus Torvalds * sync_inode() will write an inode and its pages to disk. It will also 13401da177e4SLinus Torvalds * correctly update the inode on its superblock's dirty inode lists and will 13411da177e4SLinus Torvalds * update inode->i_state. 13421da177e4SLinus Torvalds * 13431da177e4SLinus Torvalds * The caller must have a ref on the inode. 13441da177e4SLinus Torvalds */ 13451da177e4SLinus Torvalds int sync_inode(struct inode *inode, struct writeback_control *wbc) 13461da177e4SLinus Torvalds { 13471da177e4SLinus Torvalds int ret; 13481da177e4SLinus Torvalds 1349a66979abSDave Chinner spin_lock(&inode_wb_list_lock); 13500f1b1fd8SDave Chinner spin_lock(&inode->i_lock); 135101c03194SChristoph Hellwig ret = writeback_single_inode(inode, wbc); 13520f1b1fd8SDave Chinner spin_unlock(&inode->i_lock); 1353a66979abSDave Chinner spin_unlock(&inode_wb_list_lock); 13541da177e4SLinus Torvalds return ret; 13551da177e4SLinus Torvalds } 13561da177e4SLinus Torvalds EXPORT_SYMBOL(sync_inode); 1357c3765016SChristoph Hellwig 1358c3765016SChristoph Hellwig /** 1359c691b9d9SAndrew Morton * sync_inode_metadata - write an inode to disk 1360c3765016SChristoph Hellwig * @inode: the inode to sync 1361c3765016SChristoph Hellwig * @wait: wait for I/O to complete. 1362c3765016SChristoph Hellwig * 1363c691b9d9SAndrew Morton * Write an inode to disk and adjust its dirty state after completion. 1364c3765016SChristoph Hellwig * 1365c3765016SChristoph Hellwig * Note: only writes the actual inode, no associated data or other metadata. 1366c3765016SChristoph Hellwig */ 1367c3765016SChristoph Hellwig int sync_inode_metadata(struct inode *inode, int wait) 1368c3765016SChristoph Hellwig { 1369c3765016SChristoph Hellwig struct writeback_control wbc = { 1370c3765016SChristoph Hellwig .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE, 1371c3765016SChristoph Hellwig .nr_to_write = 0, /* metadata-only */ 1372c3765016SChristoph Hellwig }; 1373c3765016SChristoph Hellwig 1374c3765016SChristoph Hellwig return sync_inode(inode, &wbc); 1375c3765016SChristoph Hellwig } 1376c3765016SChristoph Hellwig EXPORT_SYMBOL(sync_inode_metadata); 1377