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> 17630d9c47SPaul Gortmaker #include <linux/export.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> 23bc31b86aSWu Fengguang #include <linux/pagemap.h> 2403ba3782SJens Axboe #include <linux/kthread.h> 251da177e4SLinus Torvalds #include <linux/writeback.h> 261da177e4SLinus Torvalds #include <linux/blkdev.h> 271da177e4SLinus Torvalds #include <linux/backing-dev.h> 28455b2864SDave Chinner #include <linux/tracepoint.h> 29719ea2fbSAl Viro #include <linux/device.h> 3007f3f05cSDavid Howells #include "internal.h" 311da177e4SLinus Torvalds 32d0bceac7SJens Axboe /* 33bc31b86aSWu Fengguang * 4MB minimal write chunk size 34bc31b86aSWu Fengguang */ 35bc31b86aSWu Fengguang #define MIN_WRITEBACK_PAGES (4096UL >> (PAGE_CACHE_SHIFT - 10)) 36bc31b86aSWu Fengguang 37bc31b86aSWu Fengguang /* 38c4a77a6cSJens Axboe * Passed into wb_writeback(), essentially a subset of writeback_control 39c4a77a6cSJens Axboe */ 4083ba7b07SChristoph Hellwig struct wb_writeback_work { 41c4a77a6cSJens Axboe long nr_pages; 42c4a77a6cSJens Axboe struct super_block *sb; 430dc83bd3SJan Kara unsigned long *older_than_this; 44c4a77a6cSJens Axboe enum writeback_sync_modes sync_mode; 456e6938b6SWu Fengguang unsigned int tagged_writepages:1; 4652957fe1SH Hartley Sweeten unsigned int for_kupdate:1; 4752957fe1SH Hartley Sweeten unsigned int range_cyclic:1; 4852957fe1SH Hartley Sweeten unsigned int for_background:1; 497747bd4bSDave Chinner unsigned int for_sync:1; /* sync(2) WB_SYNC_ALL writeback */ 500e175a18SCurt Wohlgemuth enum wb_reason reason; /* why was writeback initiated? */ 51c4a77a6cSJens Axboe 528010c3b6SJens Axboe struct list_head list; /* pending work list */ 5383ba7b07SChristoph Hellwig struct completion *done; /* set if the caller waits */ 5403ba3782SJens Axboe }; 5503ba3782SJens Axboe 56f11b00f3SAdrian Bunk /** 57f11b00f3SAdrian Bunk * writeback_in_progress - determine whether there is writeback in progress 58f11b00f3SAdrian Bunk * @bdi: the device's backing_dev_info structure. 59f11b00f3SAdrian Bunk * 6003ba3782SJens Axboe * Determine whether there is writeback waiting to be handled against a 6103ba3782SJens Axboe * backing device. 62f11b00f3SAdrian Bunk */ 63f11b00f3SAdrian Bunk int writeback_in_progress(struct backing_dev_info *bdi) 64f11b00f3SAdrian Bunk { 6581d73a32SJan Kara return test_bit(BDI_writeback_running, &bdi->state); 66f11b00f3SAdrian Bunk } 6700d4e736STheodore Ts'o EXPORT_SYMBOL(writeback_in_progress); 68f11b00f3SAdrian Bunk 69de1414a6SChristoph Hellwig struct backing_dev_info *inode_to_bdi(struct inode *inode) 70692ebd17SJan Kara { 71b520252aSJens Axboe struct super_block *sb; 72b520252aSJens Axboe 73b520252aSJens Axboe if (!inode) 74b520252aSJens Axboe return &noop_backing_dev_info; 75b520252aSJens Axboe 76b520252aSJens Axboe sb = inode->i_sb; 77495a276eSChristoph Hellwig #ifdef CONFIG_BLOCK 78a8855990SJan Kara if (sb_is_blkdev_sb(sb)) 79495a276eSChristoph Hellwig return blk_get_backing_dev_info(I_BDEV(inode)); 80495a276eSChristoph Hellwig #endif 81692ebd17SJan Kara return sb->s_bdi; 82692ebd17SJan Kara } 83de1414a6SChristoph Hellwig EXPORT_SYMBOL_GPL(inode_to_bdi); 84692ebd17SJan Kara 857ccf19a8SNick Piggin static inline struct inode *wb_inode(struct list_head *head) 867ccf19a8SNick Piggin { 877ccf19a8SNick Piggin return list_entry(head, struct inode, i_wb_list); 887ccf19a8SNick Piggin } 897ccf19a8SNick Piggin 9015eb77a0SWu Fengguang /* 9115eb77a0SWu Fengguang * Include the creation of the trace points after defining the 9215eb77a0SWu Fengguang * wb_writeback_work structure and inline functions so that the definition 9315eb77a0SWu Fengguang * remains local to this file. 9415eb77a0SWu Fengguang */ 9515eb77a0SWu Fengguang #define CREATE_TRACE_POINTS 9615eb77a0SWu Fengguang #include <trace/events/writeback.h> 9715eb77a0SWu Fengguang 98774016b2SSteven Whitehouse EXPORT_TRACEPOINT_SYMBOL_GPL(wbc_writepage); 99774016b2SSteven Whitehouse 1005acda9d1SJan Kara static void bdi_wakeup_thread(struct backing_dev_info *bdi) 1015acda9d1SJan Kara { 1025acda9d1SJan Kara spin_lock_bh(&bdi->wb_lock); 1035acda9d1SJan Kara if (test_bit(BDI_registered, &bdi->state)) 1045acda9d1SJan Kara mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0); 1055acda9d1SJan Kara spin_unlock_bh(&bdi->wb_lock); 1065acda9d1SJan Kara } 1075acda9d1SJan Kara 1086585027aSJan Kara static void bdi_queue_work(struct backing_dev_info *bdi, 1096585027aSJan Kara struct wb_writeback_work *work) 1106585027aSJan Kara { 1116585027aSJan Kara trace_writeback_queue(bdi, work); 1126585027aSJan Kara 1136585027aSJan Kara spin_lock_bh(&bdi->wb_lock); 1145acda9d1SJan Kara if (!test_bit(BDI_registered, &bdi->state)) { 1155acda9d1SJan Kara if (work->done) 1165acda9d1SJan Kara complete(work->done); 1175acda9d1SJan Kara goto out_unlock; 1185acda9d1SJan Kara } 1196585027aSJan Kara list_add_tail(&work->list, &bdi->work_list); 120839a8e86STejun Heo mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0); 1215acda9d1SJan Kara out_unlock: 1225acda9d1SJan Kara spin_unlock_bh(&bdi->wb_lock); 12303ba3782SJens Axboe } 1241da177e4SLinus Torvalds 12583ba7b07SChristoph Hellwig static void 12683ba7b07SChristoph Hellwig __bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, 1270e175a18SCurt Wohlgemuth bool range_cyclic, enum wb_reason reason) 1281da177e4SLinus Torvalds { 12983ba7b07SChristoph Hellwig struct wb_writeback_work *work; 13003ba3782SJens Axboe 131bcddc3f0SJens Axboe /* 132bcddc3f0SJens Axboe * This is WB_SYNC_NONE writeback, so if allocation fails just 133bcddc3f0SJens Axboe * wakeup the thread for old dirty data writeback 134bcddc3f0SJens Axboe */ 13583ba7b07SChristoph Hellwig work = kzalloc(sizeof(*work), GFP_ATOMIC); 13683ba7b07SChristoph Hellwig if (!work) { 137455b2864SDave Chinner trace_writeback_nowork(bdi); 1385acda9d1SJan Kara bdi_wakeup_thread(bdi); 13983ba7b07SChristoph Hellwig return; 14083ba7b07SChristoph Hellwig } 14183ba7b07SChristoph Hellwig 14283ba7b07SChristoph Hellwig work->sync_mode = WB_SYNC_NONE; 14383ba7b07SChristoph Hellwig work->nr_pages = nr_pages; 14483ba7b07SChristoph Hellwig work->range_cyclic = range_cyclic; 1450e175a18SCurt Wohlgemuth work->reason = reason; 14683ba7b07SChristoph Hellwig 147f11fcae8SJens Axboe bdi_queue_work(bdi, work); 14803ba3782SJens Axboe } 149b6e51316SJens Axboe 150b6e51316SJens Axboe /** 151b6e51316SJens Axboe * bdi_start_writeback - start writeback 152b6e51316SJens Axboe * @bdi: the backing device to write from 153b6e51316SJens Axboe * @nr_pages: the number of pages to write 154786228abSMarcos Paulo de Souza * @reason: reason why some writeback work was initiated 155b6e51316SJens Axboe * 156b6e51316SJens Axboe * Description: 157b6e51316SJens Axboe * This does WB_SYNC_NONE opportunistic writeback. The IO is only 15825985edcSLucas De Marchi * started when this function returns, we make no guarantees on 1590e3c9a22SJens Axboe * completion. Caller need not hold sb s_umount semaphore. 160b6e51316SJens Axboe * 161b6e51316SJens Axboe */ 1620e175a18SCurt Wohlgemuth void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, 1630e175a18SCurt Wohlgemuth enum wb_reason reason) 164b6e51316SJens Axboe { 1650e175a18SCurt Wohlgemuth __bdi_start_writeback(bdi, nr_pages, true, reason); 166d3ddec76SWu Fengguang } 167d3ddec76SWu Fengguang 168c5444198SChristoph Hellwig /** 169c5444198SChristoph Hellwig * bdi_start_background_writeback - start background writeback 170c5444198SChristoph Hellwig * @bdi: the backing device to write from 171c5444198SChristoph Hellwig * 172c5444198SChristoph Hellwig * Description: 1736585027aSJan Kara * This makes sure WB_SYNC_NONE background writeback happens. When 1746585027aSJan Kara * this function returns, it is only guaranteed that for given BDI 1756585027aSJan Kara * some IO is happening if we are over background dirty threshold. 1766585027aSJan Kara * Caller need not hold sb s_umount semaphore. 177c5444198SChristoph Hellwig */ 178c5444198SChristoph Hellwig void bdi_start_background_writeback(struct backing_dev_info *bdi) 179c5444198SChristoph Hellwig { 1806585027aSJan Kara /* 1816585027aSJan Kara * We just wake up the flusher thread. It will perform background 1826585027aSJan Kara * writeback as soon as there is no other work to do. 1836585027aSJan Kara */ 18471927e84SWu Fengguang trace_writeback_wake_background(bdi); 1855acda9d1SJan Kara bdi_wakeup_thread(bdi); 1861da177e4SLinus Torvalds } 1871da177e4SLinus Torvalds 1881da177e4SLinus Torvalds /* 189a66979abSDave Chinner * Remove the inode from the writeback list it is on. 190a66979abSDave Chinner */ 191a66979abSDave Chinner void inode_wb_list_del(struct inode *inode) 192a66979abSDave Chinner { 193f758eeabSChristoph Hellwig struct backing_dev_info *bdi = inode_to_bdi(inode); 194a66979abSDave Chinner 195f758eeabSChristoph Hellwig spin_lock(&bdi->wb.list_lock); 196f758eeabSChristoph Hellwig list_del_init(&inode->i_wb_list); 197f758eeabSChristoph Hellwig spin_unlock(&bdi->wb.list_lock); 198f758eeabSChristoph Hellwig } 199a66979abSDave Chinner 200a66979abSDave Chinner /* 2016610a0bcSAndrew Morton * Redirty an inode: set its when-it-was dirtied timestamp and move it to the 2026610a0bcSAndrew Morton * furthest end of its superblock's dirty-inode list. 2036610a0bcSAndrew Morton * 2046610a0bcSAndrew Morton * Before stamping the inode's ->dirtied_when, we check to see whether it is 20566f3b8e2SJens Axboe * already the most-recently-dirtied inode on the b_dirty list. If that is 2066610a0bcSAndrew Morton * the case then the inode must have been redirtied while it was being written 2076610a0bcSAndrew Morton * out and we don't reset its dirtied_when. 2086610a0bcSAndrew Morton */ 209f758eeabSChristoph Hellwig static void redirty_tail(struct inode *inode, struct bdi_writeback *wb) 2106610a0bcSAndrew Morton { 211f758eeabSChristoph Hellwig assert_spin_locked(&wb->list_lock); 21203ba3782SJens Axboe if (!list_empty(&wb->b_dirty)) { 21366f3b8e2SJens Axboe struct inode *tail; 2146610a0bcSAndrew Morton 2157ccf19a8SNick Piggin tail = wb_inode(wb->b_dirty.next); 21666f3b8e2SJens Axboe if (time_before(inode->dirtied_when, tail->dirtied_when)) 2176610a0bcSAndrew Morton inode->dirtied_when = jiffies; 2186610a0bcSAndrew Morton } 2197ccf19a8SNick Piggin list_move(&inode->i_wb_list, &wb->b_dirty); 2206610a0bcSAndrew Morton } 2216610a0bcSAndrew Morton 2226610a0bcSAndrew Morton /* 22366f3b8e2SJens Axboe * requeue inode for re-scanning after bdi->b_io list is exhausted. 224c986d1e2SAndrew Morton */ 225f758eeabSChristoph Hellwig static void requeue_io(struct inode *inode, struct bdi_writeback *wb) 226c986d1e2SAndrew Morton { 227f758eeabSChristoph Hellwig assert_spin_locked(&wb->list_lock); 2287ccf19a8SNick Piggin list_move(&inode->i_wb_list, &wb->b_more_io); 229c986d1e2SAndrew Morton } 230c986d1e2SAndrew Morton 2311c0eeaf5SJoern Engel static void inode_sync_complete(struct inode *inode) 2321c0eeaf5SJoern Engel { 233365b94aeSJan Kara inode->i_state &= ~I_SYNC; 2344eff96ddSJan Kara /* If inode is clean an unused, put it into LRU now... */ 2354eff96ddSJan Kara inode_add_lru(inode); 236365b94aeSJan Kara /* Waiters must see I_SYNC cleared before being woken up */ 2371c0eeaf5SJoern Engel smp_mb(); 2381c0eeaf5SJoern Engel wake_up_bit(&inode->i_state, __I_SYNC); 2391c0eeaf5SJoern Engel } 2401c0eeaf5SJoern Engel 241d2caa3c5SJeff Layton static bool inode_dirtied_after(struct inode *inode, unsigned long t) 242d2caa3c5SJeff Layton { 243d2caa3c5SJeff Layton bool ret = time_after(inode->dirtied_when, t); 244d2caa3c5SJeff Layton #ifndef CONFIG_64BIT 245d2caa3c5SJeff Layton /* 246d2caa3c5SJeff Layton * For inodes being constantly redirtied, dirtied_when can get stuck. 247d2caa3c5SJeff Layton * It _appears_ to be in the future, but is actually in distant past. 248d2caa3c5SJeff Layton * This test is necessary to prevent such wrapped-around relative times 2495b0830cbSJens Axboe * from permanently stopping the whole bdi writeback. 250d2caa3c5SJeff Layton */ 251d2caa3c5SJeff Layton ret = ret && time_before_eq(inode->dirtied_when, jiffies); 252d2caa3c5SJeff Layton #endif 253d2caa3c5SJeff Layton return ret; 254d2caa3c5SJeff Layton } 255d2caa3c5SJeff Layton 256c986d1e2SAndrew Morton /* 2570e2f2b23SWang Sheng-Hui * Move expired (dirtied before work->older_than_this) dirty inodes from 258697e6fedSJan Kara * @delaying_queue to @dispatch_queue. 2592c136579SFengguang Wu */ 260e84d0a4fSWu Fengguang static int move_expired_inodes(struct list_head *delaying_queue, 2612c136579SFengguang Wu struct list_head *dispatch_queue, 262ad4e38ddSCurt Wohlgemuth struct wb_writeback_work *work) 2632c136579SFengguang Wu { 2645c03449dSShaohua Li LIST_HEAD(tmp); 2655c03449dSShaohua Li struct list_head *pos, *node; 266cf137307SJens Axboe struct super_block *sb = NULL; 2675c03449dSShaohua Li struct inode *inode; 268cf137307SJens Axboe int do_sb_sort = 0; 269e84d0a4fSWu Fengguang int moved = 0; 2705c03449dSShaohua Li 2712c136579SFengguang Wu while (!list_empty(delaying_queue)) { 2727ccf19a8SNick Piggin inode = wb_inode(delaying_queue->prev); 2730dc83bd3SJan Kara if (work->older_than_this && 2740dc83bd3SJan Kara inode_dirtied_after(inode, *work->older_than_this)) 2752c136579SFengguang Wu break; 276a8855990SJan Kara list_move(&inode->i_wb_list, &tmp); 277a8855990SJan Kara moved++; 278a8855990SJan Kara if (sb_is_blkdev_sb(inode->i_sb)) 279a8855990SJan Kara continue; 280cf137307SJens Axboe if (sb && sb != inode->i_sb) 281cf137307SJens Axboe do_sb_sort = 1; 282cf137307SJens Axboe sb = inode->i_sb; 2835c03449dSShaohua Li } 2845c03449dSShaohua Li 285cf137307SJens Axboe /* just one sb in list, splice to dispatch_queue and we're done */ 286cf137307SJens Axboe if (!do_sb_sort) { 287cf137307SJens Axboe list_splice(&tmp, dispatch_queue); 288e84d0a4fSWu Fengguang goto out; 289cf137307SJens Axboe } 290cf137307SJens Axboe 2915c03449dSShaohua Li /* Move inodes from one superblock together */ 2925c03449dSShaohua Li while (!list_empty(&tmp)) { 2937ccf19a8SNick Piggin sb = wb_inode(tmp.prev)->i_sb; 2945c03449dSShaohua Li list_for_each_prev_safe(pos, node, &tmp) { 2957ccf19a8SNick Piggin inode = wb_inode(pos); 2965c03449dSShaohua Li if (inode->i_sb == sb) 2977ccf19a8SNick Piggin list_move(&inode->i_wb_list, dispatch_queue); 2982c136579SFengguang Wu } 2992c136579SFengguang Wu } 300e84d0a4fSWu Fengguang out: 301e84d0a4fSWu Fengguang return moved; 3025c03449dSShaohua Li } 3032c136579SFengguang Wu 3042c136579SFengguang Wu /* 3052c136579SFengguang Wu * Queue all expired dirty inodes for io, eldest first. 3064ea879b9SWu Fengguang * Before 3074ea879b9SWu Fengguang * newly dirtied b_dirty b_io b_more_io 3084ea879b9SWu Fengguang * =============> gf edc BA 3094ea879b9SWu Fengguang * After 3104ea879b9SWu Fengguang * newly dirtied b_dirty b_io b_more_io 3114ea879b9SWu Fengguang * =============> g fBAedc 3124ea879b9SWu Fengguang * | 3134ea879b9SWu Fengguang * +--> dequeue for IO 3142c136579SFengguang Wu */ 315ad4e38ddSCurt Wohlgemuth static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work) 3162c136579SFengguang Wu { 317e84d0a4fSWu Fengguang int moved; 318f758eeabSChristoph Hellwig assert_spin_locked(&wb->list_lock); 3194ea879b9SWu Fengguang list_splice_init(&wb->b_more_io, &wb->b_io); 320ad4e38ddSCurt Wohlgemuth moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, work); 321ad4e38ddSCurt Wohlgemuth trace_writeback_queue_io(wb, work, moved); 32266f3b8e2SJens Axboe } 32366f3b8e2SJens Axboe 324a9185b41SChristoph Hellwig static int write_inode(struct inode *inode, struct writeback_control *wbc) 32566f3b8e2SJens Axboe { 3269fb0a7daSTejun Heo int ret; 3279fb0a7daSTejun Heo 3289fb0a7daSTejun Heo if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) { 3299fb0a7daSTejun Heo trace_writeback_write_inode_start(inode, wbc); 3309fb0a7daSTejun Heo ret = inode->i_sb->s_op->write_inode(inode, wbc); 3319fb0a7daSTejun Heo trace_writeback_write_inode(inode, wbc); 3329fb0a7daSTejun Heo return ret; 3339fb0a7daSTejun Heo } 33403ba3782SJens Axboe return 0; 33566f3b8e2SJens Axboe } 33608d8e974SFengguang Wu 3372c136579SFengguang Wu /* 338169ebd90SJan Kara * Wait for writeback on an inode to complete. Called with i_lock held. 339169ebd90SJan Kara * Caller must make sure inode cannot go away when we drop i_lock. 34001c03194SChristoph Hellwig */ 341169ebd90SJan Kara static void __inode_wait_for_writeback(struct inode *inode) 342169ebd90SJan Kara __releases(inode->i_lock) 343169ebd90SJan Kara __acquires(inode->i_lock) 34401c03194SChristoph Hellwig { 34501c03194SChristoph Hellwig DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); 34601c03194SChristoph Hellwig wait_queue_head_t *wqh; 34701c03194SChristoph Hellwig 34801c03194SChristoph Hellwig wqh = bit_waitqueue(&inode->i_state, __I_SYNC); 34958a9d3d8SRichard Kennedy while (inode->i_state & I_SYNC) { 350250df6edSDave Chinner spin_unlock(&inode->i_lock); 35174316201SNeilBrown __wait_on_bit(wqh, &wq, bit_wait, 35274316201SNeilBrown TASK_UNINTERRUPTIBLE); 353250df6edSDave Chinner spin_lock(&inode->i_lock); 35458a9d3d8SRichard Kennedy } 35501c03194SChristoph Hellwig } 35601c03194SChristoph Hellwig 35701c03194SChristoph Hellwig /* 358169ebd90SJan Kara * Wait for writeback on an inode to complete. Caller must have inode pinned. 359169ebd90SJan Kara */ 360169ebd90SJan Kara void inode_wait_for_writeback(struct inode *inode) 361169ebd90SJan Kara { 362169ebd90SJan Kara spin_lock(&inode->i_lock); 363169ebd90SJan Kara __inode_wait_for_writeback(inode); 364169ebd90SJan Kara spin_unlock(&inode->i_lock); 365169ebd90SJan Kara } 366169ebd90SJan Kara 367169ebd90SJan Kara /* 368169ebd90SJan Kara * Sleep until I_SYNC is cleared. This function must be called with i_lock 369169ebd90SJan Kara * held and drops it. It is aimed for callers not holding any inode reference 370169ebd90SJan Kara * so once i_lock is dropped, inode can go away. 371169ebd90SJan Kara */ 372169ebd90SJan Kara static void inode_sleep_on_writeback(struct inode *inode) 373169ebd90SJan Kara __releases(inode->i_lock) 374169ebd90SJan Kara { 375169ebd90SJan Kara DEFINE_WAIT(wait); 376169ebd90SJan Kara wait_queue_head_t *wqh = bit_waitqueue(&inode->i_state, __I_SYNC); 377169ebd90SJan Kara int sleep; 378169ebd90SJan Kara 379169ebd90SJan Kara prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); 380169ebd90SJan Kara sleep = inode->i_state & I_SYNC; 381169ebd90SJan Kara spin_unlock(&inode->i_lock); 382169ebd90SJan Kara if (sleep) 383169ebd90SJan Kara schedule(); 384169ebd90SJan Kara finish_wait(wqh, &wait); 385169ebd90SJan Kara } 386169ebd90SJan Kara 387169ebd90SJan Kara /* 388ccb26b5aSJan Kara * Find proper writeback list for the inode depending on its current state and 389ccb26b5aSJan Kara * possibly also change of its state while we were doing writeback. Here we 390ccb26b5aSJan Kara * handle things such as livelock prevention or fairness of writeback among 391ccb26b5aSJan Kara * inodes. This function can be called only by flusher thread - noone else 392ccb26b5aSJan Kara * processes all inodes in writeback lists and requeueing inodes behind flusher 393ccb26b5aSJan Kara * thread's back can have unexpected consequences. 394ccb26b5aSJan Kara */ 395ccb26b5aSJan Kara static void requeue_inode(struct inode *inode, struct bdi_writeback *wb, 396ccb26b5aSJan Kara struct writeback_control *wbc) 397ccb26b5aSJan Kara { 398ccb26b5aSJan Kara if (inode->i_state & I_FREEING) 399ccb26b5aSJan Kara return; 400ccb26b5aSJan Kara 401ccb26b5aSJan Kara /* 402ccb26b5aSJan Kara * Sync livelock prevention. Each inode is tagged and synced in one 403ccb26b5aSJan Kara * shot. If still dirty, it will be redirty_tail()'ed below. Update 404ccb26b5aSJan Kara * the dirty time to prevent enqueue and sync it again. 405ccb26b5aSJan Kara */ 406ccb26b5aSJan Kara if ((inode->i_state & I_DIRTY) && 407ccb26b5aSJan Kara (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)) 408ccb26b5aSJan Kara inode->dirtied_when = jiffies; 409ccb26b5aSJan Kara 4104f8ad655SJan Kara if (wbc->pages_skipped) { 4114f8ad655SJan Kara /* 4124f8ad655SJan Kara * writeback is not making progress due to locked 4134f8ad655SJan Kara * buffers. Skip this inode for now. 4144f8ad655SJan Kara */ 4154f8ad655SJan Kara redirty_tail(inode, wb); 4164f8ad655SJan Kara return; 4174f8ad655SJan Kara } 4184f8ad655SJan Kara 419ccb26b5aSJan Kara if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) { 420ccb26b5aSJan Kara /* 421ccb26b5aSJan Kara * We didn't write back all the pages. nfs_writepages() 422ccb26b5aSJan Kara * sometimes bales out without doing anything. 423ccb26b5aSJan Kara */ 424ccb26b5aSJan Kara if (wbc->nr_to_write <= 0) { 425ccb26b5aSJan Kara /* Slice used up. Queue for next turn. */ 426ccb26b5aSJan Kara requeue_io(inode, wb); 427ccb26b5aSJan Kara } else { 428ccb26b5aSJan Kara /* 429ccb26b5aSJan Kara * Writeback blocked by something other than 430ccb26b5aSJan Kara * congestion. Delay the inode for some time to 431ccb26b5aSJan Kara * avoid spinning on the CPU (100% iowait) 432ccb26b5aSJan Kara * retrying writeback of the dirty page/inode 433ccb26b5aSJan Kara * that cannot be performed immediately. 434ccb26b5aSJan Kara */ 435ccb26b5aSJan Kara redirty_tail(inode, wb); 436ccb26b5aSJan Kara } 437ccb26b5aSJan Kara } else if (inode->i_state & I_DIRTY) { 438ccb26b5aSJan Kara /* 439ccb26b5aSJan Kara * Filesystems can dirty the inode during writeback operations, 440ccb26b5aSJan Kara * such as delayed allocation during submission or metadata 441ccb26b5aSJan Kara * updates after data IO completion. 442ccb26b5aSJan Kara */ 443ccb26b5aSJan Kara redirty_tail(inode, wb); 444ccb26b5aSJan Kara } else { 445ccb26b5aSJan Kara /* The inode is clean. Remove from writeback lists. */ 446ccb26b5aSJan Kara list_del_init(&inode->i_wb_list); 447ccb26b5aSJan Kara } 448ccb26b5aSJan Kara } 449ccb26b5aSJan Kara 450ccb26b5aSJan Kara /* 4514f8ad655SJan Kara * Write out an inode and its dirty pages. Do not update the writeback list 4524f8ad655SJan Kara * linkage. That is left to the caller. The caller is also responsible for 4534f8ad655SJan Kara * setting I_SYNC flag and calling inode_sync_complete() to clear it. 4541da177e4SLinus Torvalds */ 4551da177e4SLinus Torvalds static int 456cd8ed2a4SYan Hong __writeback_single_inode(struct inode *inode, struct writeback_control *wbc) 4571da177e4SLinus Torvalds { 4581da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 459251d6a47SWu Fengguang long nr_to_write = wbc->nr_to_write; 46001c03194SChristoph Hellwig unsigned dirty; 4611da177e4SLinus Torvalds int ret; 4621da177e4SLinus Torvalds 4634f8ad655SJan Kara WARN_ON(!(inode->i_state & I_SYNC)); 4641da177e4SLinus Torvalds 4659fb0a7daSTejun Heo trace_writeback_single_inode_start(inode, wbc, nr_to_write); 4669fb0a7daSTejun Heo 4671da177e4SLinus Torvalds ret = do_writepages(mapping, wbc); 4681da177e4SLinus Torvalds 46926821ed4SChristoph Hellwig /* 47026821ed4SChristoph Hellwig * Make sure to wait on the data before writing out the metadata. 47126821ed4SChristoph Hellwig * This is important for filesystems that modify metadata on data 4727747bd4bSDave Chinner * I/O completion. We don't do it for sync(2) writeback because it has a 4737747bd4bSDave Chinner * separate, external IO completion path and ->sync_fs for guaranteeing 4747747bd4bSDave Chinner * inode metadata is written back correctly. 47526821ed4SChristoph Hellwig */ 4767747bd4bSDave Chinner if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) { 47726821ed4SChristoph Hellwig int err = filemap_fdatawait(mapping); 4781da177e4SLinus Torvalds if (ret == 0) 4791da177e4SLinus Torvalds ret = err; 4801da177e4SLinus Torvalds } 4811da177e4SLinus Torvalds 4825547e8aaSDmitry Monakhov /* 4835547e8aaSDmitry Monakhov * Some filesystems may redirty the inode during the writeback 4845547e8aaSDmitry Monakhov * due to delalloc, clear dirty metadata flags right before 4855547e8aaSDmitry Monakhov * write_inode() 4865547e8aaSDmitry Monakhov */ 487250df6edSDave Chinner spin_lock(&inode->i_lock); 4889c6ac78eSTejun Heo 4895547e8aaSDmitry Monakhov dirty = inode->i_state & I_DIRTY; 4909c6ac78eSTejun Heo inode->i_state &= ~I_DIRTY; 4919c6ac78eSTejun Heo 4929c6ac78eSTejun Heo /* 4939c6ac78eSTejun Heo * Paired with smp_mb() in __mark_inode_dirty(). This allows 4949c6ac78eSTejun Heo * __mark_inode_dirty() to test i_state without grabbing i_lock - 4959c6ac78eSTejun Heo * either they see the I_DIRTY bits cleared or we see the dirtied 4969c6ac78eSTejun Heo * inode. 4979c6ac78eSTejun Heo * 4989c6ac78eSTejun Heo * I_DIRTY_PAGES is always cleared together above even if @mapping 4999c6ac78eSTejun Heo * still has dirty pages. The flag is reinstated after smp_mb() if 5009c6ac78eSTejun Heo * necessary. This guarantees that either __mark_inode_dirty() 5019c6ac78eSTejun Heo * sees clear I_DIRTY_PAGES or we see PAGECACHE_TAG_DIRTY. 5029c6ac78eSTejun Heo */ 5039c6ac78eSTejun Heo smp_mb(); 5049c6ac78eSTejun Heo 5059c6ac78eSTejun Heo if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) 5069c6ac78eSTejun Heo inode->i_state |= I_DIRTY_PAGES; 5079c6ac78eSTejun Heo 508250df6edSDave Chinner spin_unlock(&inode->i_lock); 5099c6ac78eSTejun Heo 51026821ed4SChristoph Hellwig /* Don't write the inode if only I_DIRTY_PAGES was set */ 51126821ed4SChristoph Hellwig if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { 512a9185b41SChristoph Hellwig int err = write_inode(inode, wbc); 5131da177e4SLinus Torvalds if (ret == 0) 5141da177e4SLinus Torvalds ret = err; 5151da177e4SLinus Torvalds } 5164f8ad655SJan Kara trace_writeback_single_inode(inode, wbc, nr_to_write); 5174f8ad655SJan Kara return ret; 5184f8ad655SJan Kara } 5194f8ad655SJan Kara 5204f8ad655SJan Kara /* 5214f8ad655SJan Kara * Write out an inode's dirty pages. Either the caller has an active reference 5224f8ad655SJan Kara * on the inode or the inode has I_WILL_FREE set. 5234f8ad655SJan Kara * 5244f8ad655SJan Kara * This function is designed to be called for writing back one inode which 5254f8ad655SJan Kara * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode() 5264f8ad655SJan Kara * and does more profound writeback list handling in writeback_sb_inodes(). 5274f8ad655SJan Kara */ 5284f8ad655SJan Kara static int 5294f8ad655SJan Kara writeback_single_inode(struct inode *inode, struct bdi_writeback *wb, 5304f8ad655SJan Kara struct writeback_control *wbc) 5314f8ad655SJan Kara { 5324f8ad655SJan Kara int ret = 0; 5334f8ad655SJan Kara 5344f8ad655SJan Kara spin_lock(&inode->i_lock); 5354f8ad655SJan Kara if (!atomic_read(&inode->i_count)) 5364f8ad655SJan Kara WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); 5374f8ad655SJan Kara else 5384f8ad655SJan Kara WARN_ON(inode->i_state & I_WILL_FREE); 5394f8ad655SJan Kara 5404f8ad655SJan Kara if (inode->i_state & I_SYNC) { 5414f8ad655SJan Kara if (wbc->sync_mode != WB_SYNC_ALL) 5424f8ad655SJan Kara goto out; 5434f8ad655SJan Kara /* 544169ebd90SJan Kara * It's a data-integrity sync. We must wait. Since callers hold 545169ebd90SJan Kara * inode reference or inode has I_WILL_FREE set, it cannot go 546169ebd90SJan Kara * away under us. 5474f8ad655SJan Kara */ 548169ebd90SJan Kara __inode_wait_for_writeback(inode); 5494f8ad655SJan Kara } 5504f8ad655SJan Kara WARN_ON(inode->i_state & I_SYNC); 5514f8ad655SJan Kara /* 552f9b0e058SJan Kara * Skip inode if it is clean and we have no outstanding writeback in 553f9b0e058SJan Kara * WB_SYNC_ALL mode. We don't want to mess with writeback lists in this 554f9b0e058SJan Kara * function since flusher thread may be doing for example sync in 555f9b0e058SJan Kara * parallel and if we move the inode, it could get skipped. So here we 556f9b0e058SJan Kara * make sure inode is on some writeback list and leave it there unless 557f9b0e058SJan Kara * we have completely cleaned the inode. 5584f8ad655SJan Kara */ 559f9b0e058SJan Kara if (!(inode->i_state & I_DIRTY) && 560f9b0e058SJan Kara (wbc->sync_mode != WB_SYNC_ALL || 561f9b0e058SJan Kara !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK))) 5624f8ad655SJan Kara goto out; 5634f8ad655SJan Kara inode->i_state |= I_SYNC; 5644f8ad655SJan Kara spin_unlock(&inode->i_lock); 5654f8ad655SJan Kara 566cd8ed2a4SYan Hong ret = __writeback_single_inode(inode, wbc); 5671da177e4SLinus Torvalds 568f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 569250df6edSDave Chinner spin_lock(&inode->i_lock); 5704f8ad655SJan Kara /* 5714f8ad655SJan Kara * If inode is clean, remove it from writeback lists. Otherwise don't 5724f8ad655SJan Kara * touch it. See comment above for explanation. 5734f8ad655SJan Kara */ 5744f8ad655SJan Kara if (!(inode->i_state & I_DIRTY)) 5754f8ad655SJan Kara list_del_init(&inode->i_wb_list); 5764f8ad655SJan Kara spin_unlock(&wb->list_lock); 5771c0eeaf5SJoern Engel inode_sync_complete(inode); 5784f8ad655SJan Kara out: 5794f8ad655SJan Kara spin_unlock(&inode->i_lock); 5801da177e4SLinus Torvalds return ret; 5811da177e4SLinus Torvalds } 5821da177e4SLinus Torvalds 5831a12d8bdSWu Fengguang static long writeback_chunk_size(struct backing_dev_info *bdi, 5841a12d8bdSWu Fengguang struct wb_writeback_work *work) 585d46db3d5SWu Fengguang { 586d46db3d5SWu Fengguang long pages; 587d46db3d5SWu Fengguang 588d46db3d5SWu Fengguang /* 589d46db3d5SWu Fengguang * WB_SYNC_ALL mode does livelock avoidance by syncing dirty 590d46db3d5SWu Fengguang * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX 591d46db3d5SWu Fengguang * here avoids calling into writeback_inodes_wb() more than once. 592d46db3d5SWu Fengguang * 593d46db3d5SWu Fengguang * The intended call sequence for WB_SYNC_ALL writeback is: 594d46db3d5SWu Fengguang * 595d46db3d5SWu Fengguang * wb_writeback() 596d46db3d5SWu Fengguang * writeback_sb_inodes() <== called only once 597d46db3d5SWu Fengguang * write_cache_pages() <== called once for each inode 598d46db3d5SWu Fengguang * (quickly) tag currently dirty pages 599d46db3d5SWu Fengguang * (maybe slowly) sync all tagged pages 600d46db3d5SWu Fengguang */ 601d46db3d5SWu Fengguang if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages) 602d46db3d5SWu Fengguang pages = LONG_MAX; 6031a12d8bdSWu Fengguang else { 6041a12d8bdSWu Fengguang pages = min(bdi->avg_write_bandwidth / 2, 6051a12d8bdSWu Fengguang global_dirty_limit / DIRTY_SCOPE); 6061a12d8bdSWu Fengguang pages = min(pages, work->nr_pages); 6071a12d8bdSWu Fengguang pages = round_down(pages + MIN_WRITEBACK_PAGES, 6081a12d8bdSWu Fengguang MIN_WRITEBACK_PAGES); 6091a12d8bdSWu Fengguang } 610d46db3d5SWu Fengguang 611d46db3d5SWu Fengguang return pages; 612d46db3d5SWu Fengguang } 613d46db3d5SWu Fengguang 61403ba3782SJens Axboe /* 615f11c9c5cSEdward Shishkin * Write a portion of b_io inodes which belong to @sb. 616edadfb10SChristoph Hellwig * 617d46db3d5SWu Fengguang * Return the number of pages and/or inodes written. 618f11c9c5cSEdward Shishkin */ 619d46db3d5SWu Fengguang static long writeback_sb_inodes(struct super_block *sb, 620d46db3d5SWu Fengguang struct bdi_writeback *wb, 621d46db3d5SWu Fengguang struct wb_writeback_work *work) 62203ba3782SJens Axboe { 623d46db3d5SWu Fengguang struct writeback_control wbc = { 624d46db3d5SWu Fengguang .sync_mode = work->sync_mode, 625d46db3d5SWu Fengguang .tagged_writepages = work->tagged_writepages, 626d46db3d5SWu Fengguang .for_kupdate = work->for_kupdate, 627d46db3d5SWu Fengguang .for_background = work->for_background, 6287747bd4bSDave Chinner .for_sync = work->for_sync, 629d46db3d5SWu Fengguang .range_cyclic = work->range_cyclic, 630d46db3d5SWu Fengguang .range_start = 0, 631d46db3d5SWu Fengguang .range_end = LLONG_MAX, 632d46db3d5SWu Fengguang }; 633d46db3d5SWu Fengguang unsigned long start_time = jiffies; 634d46db3d5SWu Fengguang long write_chunk; 635d46db3d5SWu Fengguang long wrote = 0; /* count both pages and inodes */ 636d46db3d5SWu Fengguang 63703ba3782SJens Axboe while (!list_empty(&wb->b_io)) { 6387ccf19a8SNick Piggin struct inode *inode = wb_inode(wb->b_io.prev); 639edadfb10SChristoph Hellwig 640edadfb10SChristoph Hellwig if (inode->i_sb != sb) { 641d46db3d5SWu Fengguang if (work->sb) { 642edadfb10SChristoph Hellwig /* 643edadfb10SChristoph Hellwig * We only want to write back data for this 644edadfb10SChristoph Hellwig * superblock, move all inodes not belonging 645edadfb10SChristoph Hellwig * to it back onto the dirty list. 646edadfb10SChristoph Hellwig */ 647f758eeabSChristoph Hellwig redirty_tail(inode, wb); 64866f3b8e2SJens Axboe continue; 64966f3b8e2SJens Axboe } 650edadfb10SChristoph Hellwig 651edadfb10SChristoph Hellwig /* 652edadfb10SChristoph Hellwig * The inode belongs to a different superblock. 653edadfb10SChristoph Hellwig * Bounce back to the caller to unpin this and 654edadfb10SChristoph Hellwig * pin the next superblock. 655edadfb10SChristoph Hellwig */ 656d46db3d5SWu Fengguang break; 657edadfb10SChristoph Hellwig } 658edadfb10SChristoph Hellwig 6599843b76aSChristoph Hellwig /* 660331cbdeeSWanpeng Li * Don't bother with new inodes or inodes being freed, first 661331cbdeeSWanpeng Li * kind does not need periodic writeout yet, and for the latter 6629843b76aSChristoph Hellwig * kind writeout is handled by the freer. 6639843b76aSChristoph Hellwig */ 664250df6edSDave Chinner spin_lock(&inode->i_lock); 6659843b76aSChristoph Hellwig if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { 666250df6edSDave Chinner spin_unlock(&inode->i_lock); 667fcc5c222SWu Fengguang redirty_tail(inode, wb); 6687ef0d737SNick Piggin continue; 6697ef0d737SNick Piggin } 670cc1676d9SJan Kara if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) { 671cc1676d9SJan Kara /* 672cc1676d9SJan Kara * If this inode is locked for writeback and we are not 673cc1676d9SJan Kara * doing writeback-for-data-integrity, move it to 674cc1676d9SJan Kara * b_more_io so that writeback can proceed with the 675cc1676d9SJan Kara * other inodes on s_io. 676cc1676d9SJan Kara * 677cc1676d9SJan Kara * We'll have another go at writing back this inode 678cc1676d9SJan Kara * when we completed a full scan of b_io. 679cc1676d9SJan Kara */ 680cc1676d9SJan Kara spin_unlock(&inode->i_lock); 681cc1676d9SJan Kara requeue_io(inode, wb); 682cc1676d9SJan Kara trace_writeback_sb_inodes_requeue(inode); 683cc1676d9SJan Kara continue; 684cc1676d9SJan Kara } 685f0d07b7fSJan Kara spin_unlock(&wb->list_lock); 686f0d07b7fSJan Kara 6874f8ad655SJan Kara /* 6884f8ad655SJan Kara * We already requeued the inode if it had I_SYNC set and we 6894f8ad655SJan Kara * are doing WB_SYNC_NONE writeback. So this catches only the 6904f8ad655SJan Kara * WB_SYNC_ALL case. 6914f8ad655SJan Kara */ 692169ebd90SJan Kara if (inode->i_state & I_SYNC) { 693169ebd90SJan Kara /* Wait for I_SYNC. This function drops i_lock... */ 694169ebd90SJan Kara inode_sleep_on_writeback(inode); 695169ebd90SJan Kara /* Inode may be gone, start again */ 696ead188f9SJan Kara spin_lock(&wb->list_lock); 697169ebd90SJan Kara continue; 698169ebd90SJan Kara } 6994f8ad655SJan Kara inode->i_state |= I_SYNC; 7004f8ad655SJan Kara spin_unlock(&inode->i_lock); 701169ebd90SJan Kara 7021a12d8bdSWu Fengguang write_chunk = writeback_chunk_size(wb->bdi, work); 703d46db3d5SWu Fengguang wbc.nr_to_write = write_chunk; 704d46db3d5SWu Fengguang wbc.pages_skipped = 0; 705250df6edSDave Chinner 706169ebd90SJan Kara /* 707169ebd90SJan Kara * We use I_SYNC to pin the inode in memory. While it is set 708169ebd90SJan Kara * evict_inode() will wait so the inode cannot be freed. 709169ebd90SJan Kara */ 710cd8ed2a4SYan Hong __writeback_single_inode(inode, &wbc); 711d46db3d5SWu Fengguang 712d46db3d5SWu Fengguang work->nr_pages -= write_chunk - wbc.nr_to_write; 713d46db3d5SWu Fengguang wrote += write_chunk - wbc.nr_to_write; 7144f8ad655SJan Kara spin_lock(&wb->list_lock); 7154f8ad655SJan Kara spin_lock(&inode->i_lock); 716d46db3d5SWu Fengguang if (!(inode->i_state & I_DIRTY)) 717d46db3d5SWu Fengguang wrote++; 7184f8ad655SJan Kara requeue_inode(inode, wb, &wbc); 7194f8ad655SJan Kara inode_sync_complete(inode); 7200f1b1fd8SDave Chinner spin_unlock(&inode->i_lock); 721169ebd90SJan Kara cond_resched_lock(&wb->list_lock); 722d46db3d5SWu Fengguang /* 723d46db3d5SWu Fengguang * bail out to wb_writeback() often enough to check 724d46db3d5SWu Fengguang * background threshold and other termination conditions. 725d46db3d5SWu Fengguang */ 726d46db3d5SWu Fengguang if (wrote) { 727d46db3d5SWu Fengguang if (time_is_before_jiffies(start_time + HZ / 10UL)) 728d46db3d5SWu Fengguang break; 729d46db3d5SWu Fengguang if (work->nr_pages <= 0) 730d46db3d5SWu Fengguang break; 7311da177e4SLinus Torvalds } 7328bc3be27SFengguang Wu } 733d46db3d5SWu Fengguang return wrote; 734f11c9c5cSEdward Shishkin } 73538f21977SNick Piggin 736d46db3d5SWu Fengguang static long __writeback_inodes_wb(struct bdi_writeback *wb, 737d46db3d5SWu Fengguang struct wb_writeback_work *work) 738f11c9c5cSEdward Shishkin { 739d46db3d5SWu Fengguang unsigned long start_time = jiffies; 740d46db3d5SWu Fengguang long wrote = 0; 741f11c9c5cSEdward Shishkin 742f11c9c5cSEdward Shishkin while (!list_empty(&wb->b_io)) { 7437ccf19a8SNick Piggin struct inode *inode = wb_inode(wb->b_io.prev); 744f11c9c5cSEdward Shishkin struct super_block *sb = inode->i_sb; 745f11c9c5cSEdward Shishkin 74612ad3ab6SDave Chinner if (!grab_super_passive(sb)) { 7470e995816SWu Fengguang /* 7480e995816SWu Fengguang * grab_super_passive() may fail consistently due to 7490e995816SWu Fengguang * s_umount being grabbed by someone else. Don't use 7500e995816SWu Fengguang * requeue_io() to avoid busy retrying the inode/sb. 7510e995816SWu Fengguang */ 7520e995816SWu Fengguang redirty_tail(inode, wb); 753d19de7edSChristoph Hellwig continue; 754334132aeSChristoph Hellwig } 755d46db3d5SWu Fengguang wrote += writeback_sb_inodes(sb, wb, work); 756d19de7edSChristoph Hellwig drop_super(sb); 757f11c9c5cSEdward Shishkin 758d46db3d5SWu Fengguang /* refer to the same tests at the end of writeback_sb_inodes */ 759d46db3d5SWu Fengguang if (wrote) { 760d46db3d5SWu Fengguang if (time_is_before_jiffies(start_time + HZ / 10UL)) 761d46db3d5SWu Fengguang break; 762d46db3d5SWu Fengguang if (work->nr_pages <= 0) 763f11c9c5cSEdward Shishkin break; 764f11c9c5cSEdward Shishkin } 765d46db3d5SWu Fengguang } 76666f3b8e2SJens Axboe /* Leave any unwritten inodes on b_io */ 767d46db3d5SWu Fengguang return wrote; 76866f3b8e2SJens Axboe } 76966f3b8e2SJens Axboe 7707d9f073bSWanpeng Li static long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages, 7710e175a18SCurt Wohlgemuth enum wb_reason reason) 772edadfb10SChristoph Hellwig { 773d46db3d5SWu Fengguang struct wb_writeback_work work = { 774d46db3d5SWu Fengguang .nr_pages = nr_pages, 775d46db3d5SWu Fengguang .sync_mode = WB_SYNC_NONE, 776d46db3d5SWu Fengguang .range_cyclic = 1, 7770e175a18SCurt Wohlgemuth .reason = reason, 778d46db3d5SWu Fengguang }; 779edadfb10SChristoph Hellwig 780f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 781424b351fSWu Fengguang if (list_empty(&wb->b_io)) 782ad4e38ddSCurt Wohlgemuth queue_io(wb, &work); 783d46db3d5SWu Fengguang __writeback_inodes_wb(wb, &work); 784f758eeabSChristoph Hellwig spin_unlock(&wb->list_lock); 785edadfb10SChristoph Hellwig 786d46db3d5SWu Fengguang return nr_pages - work.nr_pages; 78766f3b8e2SJens Axboe } 78866f3b8e2SJens Axboe 789b00949aaSWu Fengguang static bool over_bground_thresh(struct backing_dev_info *bdi) 79003ba3782SJens Axboe { 79103ba3782SJens Axboe unsigned long background_thresh, dirty_thresh; 79203ba3782SJens Axboe 79316c4042fSWu Fengguang global_dirty_limits(&background_thresh, &dirty_thresh); 79403ba3782SJens Axboe 795b00949aaSWu Fengguang if (global_page_state(NR_FILE_DIRTY) + 796b00949aaSWu Fengguang global_page_state(NR_UNSTABLE_NFS) > background_thresh) 797b00949aaSWu Fengguang return true; 798b00949aaSWu Fengguang 799b00949aaSWu Fengguang if (bdi_stat(bdi, BDI_RECLAIMABLE) > 800b00949aaSWu Fengguang bdi_dirty_limit(bdi, background_thresh)) 801b00949aaSWu Fengguang return true; 802b00949aaSWu Fengguang 803b00949aaSWu Fengguang return false; 80403ba3782SJens Axboe } 80503ba3782SJens Axboe 80603ba3782SJens Axboe /* 807e98be2d5SWu Fengguang * Called under wb->list_lock. If there are multiple wb per bdi, 808e98be2d5SWu Fengguang * only the flusher working on the first wb should do it. 809e98be2d5SWu Fengguang */ 810e98be2d5SWu Fengguang static void wb_update_bandwidth(struct bdi_writeback *wb, 811e98be2d5SWu Fengguang unsigned long start_time) 812e98be2d5SWu Fengguang { 813af6a3113SWu Fengguang __bdi_update_bandwidth(wb->bdi, 0, 0, 0, 0, 0, start_time); 814e98be2d5SWu Fengguang } 815e98be2d5SWu Fengguang 816e98be2d5SWu Fengguang /* 81703ba3782SJens Axboe * Explicit flushing or periodic writeback of "old" data. 81803ba3782SJens Axboe * 81903ba3782SJens Axboe * Define "old": the first time one of an inode's pages is dirtied, we mark the 82003ba3782SJens Axboe * dirtying-time in the inode's address_space. So this periodic writeback code 82103ba3782SJens Axboe * just walks the superblock inode list, writing back any inodes which are 82203ba3782SJens Axboe * older than a specific point in time. 82303ba3782SJens Axboe * 82403ba3782SJens Axboe * Try to run once per dirty_writeback_interval. But if a writeback event 82503ba3782SJens Axboe * takes longer than a dirty_writeback_interval interval, then leave a 82603ba3782SJens Axboe * one-second gap. 82703ba3782SJens Axboe * 82803ba3782SJens Axboe * older_than_this takes precedence over nr_to_write. So we'll only write back 82903ba3782SJens Axboe * all dirty pages if they are all attached to "old" mappings. 83003ba3782SJens Axboe */ 831c4a77a6cSJens Axboe static long wb_writeback(struct bdi_writeback *wb, 83283ba7b07SChristoph Hellwig struct wb_writeback_work *work) 83303ba3782SJens Axboe { 834e98be2d5SWu Fengguang unsigned long wb_start = jiffies; 835d46db3d5SWu Fengguang long nr_pages = work->nr_pages; 8360dc83bd3SJan Kara unsigned long oldest_jif; 837a5989bdcSJan Kara struct inode *inode; 838d46db3d5SWu Fengguang long progress; 83903ba3782SJens Axboe 8400dc83bd3SJan Kara oldest_jif = jiffies; 8410dc83bd3SJan Kara work->older_than_this = &oldest_jif; 84203ba3782SJens Axboe 843e8dfc305SWu Fengguang spin_lock(&wb->list_lock); 84403ba3782SJens Axboe for (;;) { 84503ba3782SJens Axboe /* 846d3ddec76SWu Fengguang * Stop writeback when nr_pages has been consumed 84703ba3782SJens Axboe */ 84883ba7b07SChristoph Hellwig if (work->nr_pages <= 0) 84903ba3782SJens Axboe break; 85003ba3782SJens Axboe 85103ba3782SJens Axboe /* 852aa373cf5SJan Kara * Background writeout and kupdate-style writeback may 853aa373cf5SJan Kara * run forever. Stop them if there is other work to do 854aa373cf5SJan Kara * so that e.g. sync can proceed. They'll be restarted 855aa373cf5SJan Kara * after the other works are all done. 856aa373cf5SJan Kara */ 857aa373cf5SJan Kara if ((work->for_background || work->for_kupdate) && 858aa373cf5SJan Kara !list_empty(&wb->bdi->work_list)) 859aa373cf5SJan Kara break; 860aa373cf5SJan Kara 861aa373cf5SJan Kara /* 862d3ddec76SWu Fengguang * For background writeout, stop when we are below the 863d3ddec76SWu Fengguang * background dirty threshold 86403ba3782SJens Axboe */ 865b00949aaSWu Fengguang if (work->for_background && !over_bground_thresh(wb->bdi)) 86603ba3782SJens Axboe break; 86703ba3782SJens Axboe 8681bc36b64SJan Kara /* 8691bc36b64SJan Kara * Kupdate and background works are special and we want to 8701bc36b64SJan Kara * include all inodes that need writing. Livelock avoidance is 8711bc36b64SJan Kara * handled by these works yielding to any other work so we are 8721bc36b64SJan Kara * safe. 8731bc36b64SJan Kara */ 874ba9aa839SWu Fengguang if (work->for_kupdate) { 8750dc83bd3SJan Kara oldest_jif = jiffies - 876ba9aa839SWu Fengguang msecs_to_jiffies(dirty_expire_interval * 10); 8771bc36b64SJan Kara } else if (work->for_background) 8780dc83bd3SJan Kara oldest_jif = jiffies; 879028c2dd1SDave Chinner 880d46db3d5SWu Fengguang trace_writeback_start(wb->bdi, work); 881e8dfc305SWu Fengguang if (list_empty(&wb->b_io)) 882ad4e38ddSCurt Wohlgemuth queue_io(wb, work); 88383ba7b07SChristoph Hellwig if (work->sb) 884d46db3d5SWu Fengguang progress = writeback_sb_inodes(work->sb, wb, work); 885edadfb10SChristoph Hellwig else 886d46db3d5SWu Fengguang progress = __writeback_inodes_wb(wb, work); 887d46db3d5SWu Fengguang trace_writeback_written(wb->bdi, work); 888028c2dd1SDave Chinner 889e98be2d5SWu Fengguang wb_update_bandwidth(wb, wb_start); 89003ba3782SJens Axboe 89103ba3782SJens Axboe /* 89271fd05a8SJens Axboe * Did we write something? Try for more 893e6fb6da2SWu Fengguang * 894e6fb6da2SWu Fengguang * Dirty inodes are moved to b_io for writeback in batches. 895e6fb6da2SWu Fengguang * The completion of the current batch does not necessarily 896e6fb6da2SWu Fengguang * mean the overall work is done. So we keep looping as long 897e6fb6da2SWu Fengguang * as made some progress on cleaning pages or inodes. 89871fd05a8SJens Axboe */ 899d46db3d5SWu Fengguang if (progress) 90003ba3782SJens Axboe continue; 901a5989bdcSJan Kara /* 902e6fb6da2SWu Fengguang * No more inodes for IO, bail 903a5989bdcSJan Kara */ 904b7a2441fSWu Fengguang if (list_empty(&wb->b_more_io)) 90503ba3782SJens Axboe break; 90603ba3782SJens Axboe /* 9078010c3b6SJens Axboe * Nothing written. Wait for some inode to 9088010c3b6SJens Axboe * become available for writeback. Otherwise 9098010c3b6SJens Axboe * we'll just busyloop. 91003ba3782SJens Axboe */ 91103ba3782SJens Axboe if (!list_empty(&wb->b_more_io)) { 912d46db3d5SWu Fengguang trace_writeback_wait(wb->bdi, work); 91303ba3782SJens Axboe inode = wb_inode(wb->b_more_io.prev); 914250df6edSDave Chinner spin_lock(&inode->i_lock); 915f0d07b7fSJan Kara spin_unlock(&wb->list_lock); 916169ebd90SJan Kara /* This function drops i_lock... */ 917169ebd90SJan Kara inode_sleep_on_writeback(inode); 918f0d07b7fSJan Kara spin_lock(&wb->list_lock); 91903ba3782SJens Axboe } 92003ba3782SJens Axboe } 921e8dfc305SWu Fengguang spin_unlock(&wb->list_lock); 92203ba3782SJens Axboe 923d46db3d5SWu Fengguang return nr_pages - work->nr_pages; 92403ba3782SJens Axboe } 92503ba3782SJens Axboe 92603ba3782SJens Axboe /* 92783ba7b07SChristoph Hellwig * Return the next wb_writeback_work struct that hasn't been processed yet. 92803ba3782SJens Axboe */ 92983ba7b07SChristoph Hellwig static struct wb_writeback_work * 93008852b6dSMinchan Kim get_next_work_item(struct backing_dev_info *bdi) 93103ba3782SJens Axboe { 93283ba7b07SChristoph Hellwig struct wb_writeback_work *work = NULL; 93303ba3782SJens Axboe 9346467716aSArtem Bityutskiy spin_lock_bh(&bdi->wb_lock); 93583ba7b07SChristoph Hellwig if (!list_empty(&bdi->work_list)) { 93683ba7b07SChristoph Hellwig work = list_entry(bdi->work_list.next, 93783ba7b07SChristoph Hellwig struct wb_writeback_work, list); 93883ba7b07SChristoph Hellwig list_del_init(&work->list); 93903ba3782SJens Axboe } 9406467716aSArtem Bityutskiy spin_unlock_bh(&bdi->wb_lock); 94183ba7b07SChristoph Hellwig return work; 94203ba3782SJens Axboe } 94303ba3782SJens Axboe 944cdf01dd5SLinus Torvalds /* 945cdf01dd5SLinus Torvalds * Add in the number of potentially dirty inodes, because each inode 946cdf01dd5SLinus Torvalds * write can dirty pagecache in the underlying blockdev. 947cdf01dd5SLinus Torvalds */ 948cdf01dd5SLinus Torvalds static unsigned long get_nr_dirty_pages(void) 949cdf01dd5SLinus Torvalds { 950cdf01dd5SLinus Torvalds return global_page_state(NR_FILE_DIRTY) + 951cdf01dd5SLinus Torvalds global_page_state(NR_UNSTABLE_NFS) + 952cdf01dd5SLinus Torvalds get_nr_dirty_inodes(); 953cdf01dd5SLinus Torvalds } 954cdf01dd5SLinus Torvalds 9556585027aSJan Kara static long wb_check_background_flush(struct bdi_writeback *wb) 9566585027aSJan Kara { 957b00949aaSWu Fengguang if (over_bground_thresh(wb->bdi)) { 9586585027aSJan Kara 9596585027aSJan Kara struct wb_writeback_work work = { 9606585027aSJan Kara .nr_pages = LONG_MAX, 9616585027aSJan Kara .sync_mode = WB_SYNC_NONE, 9626585027aSJan Kara .for_background = 1, 9636585027aSJan Kara .range_cyclic = 1, 9640e175a18SCurt Wohlgemuth .reason = WB_REASON_BACKGROUND, 9656585027aSJan Kara }; 9666585027aSJan Kara 9676585027aSJan Kara return wb_writeback(wb, &work); 9686585027aSJan Kara } 9696585027aSJan Kara 9706585027aSJan Kara return 0; 9716585027aSJan Kara } 9726585027aSJan Kara 97303ba3782SJens Axboe static long wb_check_old_data_flush(struct bdi_writeback *wb) 97403ba3782SJens Axboe { 97503ba3782SJens Axboe unsigned long expired; 97603ba3782SJens Axboe long nr_pages; 97703ba3782SJens Axboe 97869b62d01SJens Axboe /* 97969b62d01SJens Axboe * When set to zero, disable periodic writeback 98069b62d01SJens Axboe */ 98169b62d01SJens Axboe if (!dirty_writeback_interval) 98269b62d01SJens Axboe return 0; 98369b62d01SJens Axboe 98403ba3782SJens Axboe expired = wb->last_old_flush + 98503ba3782SJens Axboe msecs_to_jiffies(dirty_writeback_interval * 10); 98603ba3782SJens Axboe if (time_before(jiffies, expired)) 98703ba3782SJens Axboe return 0; 98803ba3782SJens Axboe 98903ba3782SJens Axboe wb->last_old_flush = jiffies; 990cdf01dd5SLinus Torvalds nr_pages = get_nr_dirty_pages(); 99103ba3782SJens Axboe 992c4a77a6cSJens Axboe if (nr_pages) { 99383ba7b07SChristoph Hellwig struct wb_writeback_work work = { 994c4a77a6cSJens Axboe .nr_pages = nr_pages, 995c4a77a6cSJens Axboe .sync_mode = WB_SYNC_NONE, 996c4a77a6cSJens Axboe .for_kupdate = 1, 997c4a77a6cSJens Axboe .range_cyclic = 1, 9980e175a18SCurt Wohlgemuth .reason = WB_REASON_PERIODIC, 999c4a77a6cSJens Axboe }; 1000c4a77a6cSJens Axboe 100183ba7b07SChristoph Hellwig return wb_writeback(wb, &work); 1002c4a77a6cSJens Axboe } 100303ba3782SJens Axboe 100403ba3782SJens Axboe return 0; 100503ba3782SJens Axboe } 100603ba3782SJens Axboe 100703ba3782SJens Axboe /* 100803ba3782SJens Axboe * Retrieve work items and do the writeback they describe 100903ba3782SJens Axboe */ 101025d130baSWanpeng Li static long wb_do_writeback(struct bdi_writeback *wb) 101103ba3782SJens Axboe { 101203ba3782SJens Axboe struct backing_dev_info *bdi = wb->bdi; 101383ba7b07SChristoph Hellwig struct wb_writeback_work *work; 1014c4a77a6cSJens Axboe long wrote = 0; 101503ba3782SJens Axboe 101681d73a32SJan Kara set_bit(BDI_writeback_running, &wb->bdi->state); 101708852b6dSMinchan Kim while ((work = get_next_work_item(bdi)) != NULL) { 101883ba7b07SChristoph Hellwig 1019455b2864SDave Chinner trace_writeback_exec(bdi, work); 1020455b2864SDave Chinner 102183ba7b07SChristoph Hellwig wrote += wb_writeback(wb, work); 102203ba3782SJens Axboe 102303ba3782SJens Axboe /* 102483ba7b07SChristoph Hellwig * Notify the caller of completion if this is a synchronous 102583ba7b07SChristoph Hellwig * work item, otherwise just free it. 102603ba3782SJens Axboe */ 102783ba7b07SChristoph Hellwig if (work->done) 102883ba7b07SChristoph Hellwig complete(work->done); 102983ba7b07SChristoph Hellwig else 103083ba7b07SChristoph Hellwig kfree(work); 103103ba3782SJens Axboe } 103203ba3782SJens Axboe 103303ba3782SJens Axboe /* 103403ba3782SJens Axboe * Check for periodic writeback, kupdated() style 103503ba3782SJens Axboe */ 103603ba3782SJens Axboe wrote += wb_check_old_data_flush(wb); 10376585027aSJan Kara wrote += wb_check_background_flush(wb); 103881d73a32SJan Kara clear_bit(BDI_writeback_running, &wb->bdi->state); 103903ba3782SJens Axboe 104003ba3782SJens Axboe return wrote; 104103ba3782SJens Axboe } 104203ba3782SJens Axboe 104303ba3782SJens Axboe /* 104403ba3782SJens Axboe * Handle writeback of dirty data for the device backed by this bdi. Also 1045839a8e86STejun Heo * reschedules periodically and does kupdated style flushing. 104603ba3782SJens Axboe */ 1047839a8e86STejun Heo void bdi_writeback_workfn(struct work_struct *work) 104803ba3782SJens Axboe { 1049839a8e86STejun Heo struct bdi_writeback *wb = container_of(to_delayed_work(work), 1050839a8e86STejun Heo struct bdi_writeback, dwork); 105108243900SChristoph Hellwig struct backing_dev_info *bdi = wb->bdi; 105203ba3782SJens Axboe long pages_written; 105303ba3782SJens Axboe 1054ef3b1019STejun Heo set_worker_desc("flush-%s", dev_name(bdi->dev)); 1055766f9164SPeter Zijlstra current->flags |= PF_SWAPWRITE; 105603ba3782SJens Axboe 1057839a8e86STejun Heo if (likely(!current_is_workqueue_rescuer() || 10585acda9d1SJan Kara !test_bit(BDI_registered, &bdi->state))) { 105903ba3782SJens Axboe /* 1060839a8e86STejun Heo * The normal path. Keep writing back @bdi until its 1061839a8e86STejun Heo * work_list is empty. Note that this path is also taken 1062839a8e86STejun Heo * if @bdi is shutting down even when we're running off the 1063839a8e86STejun Heo * rescuer as work_list needs to be drained. 106403ba3782SJens Axboe */ 1065839a8e86STejun Heo do { 106625d130baSWanpeng Li pages_written = wb_do_writeback(wb); 1067455b2864SDave Chinner trace_writeback_pages_written(pages_written); 1068839a8e86STejun Heo } while (!list_empty(&bdi->work_list)); 1069839a8e86STejun Heo } else { 1070253c34e9SArtem Bityutskiy /* 1071839a8e86STejun Heo * bdi_wq can't get enough workers and we're running off 1072839a8e86STejun Heo * the emergency worker. Don't hog it. Hopefully, 1024 is 1073839a8e86STejun Heo * enough for efficient IO. 1074253c34e9SArtem Bityutskiy */ 1075839a8e86STejun Heo pages_written = writeback_inodes_wb(&bdi->wb, 1024, 1076839a8e86STejun Heo WB_REASON_FORKER_THREAD); 1077839a8e86STejun Heo trace_writeback_pages_written(pages_written); 107803ba3782SJens Axboe } 107903ba3782SJens Axboe 10806ca738d6SDerek Basehore if (!list_empty(&bdi->work_list)) 10816ca738d6SDerek Basehore mod_delayed_work(bdi_wq, &wb->dwork, 0); 10826ca738d6SDerek Basehore else if (wb_has_dirty_io(wb) && dirty_writeback_interval) 10836ca738d6SDerek Basehore bdi_wakeup_thread_delayed(bdi); 1084455b2864SDave Chinner 1085839a8e86STejun Heo current->flags &= ~PF_SWAPWRITE; 108603ba3782SJens Axboe } 108703ba3782SJens Axboe 108803ba3782SJens Axboe /* 108903ba3782SJens Axboe * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back 109003ba3782SJens Axboe * the whole world. 109103ba3782SJens Axboe */ 10920e175a18SCurt Wohlgemuth void wakeup_flusher_threads(long nr_pages, enum wb_reason reason) 109303ba3782SJens Axboe { 1094b8c2f347SChristoph Hellwig struct backing_dev_info *bdi; 1095b8c2f347SChristoph Hellwig 109647df3ddeSJan Kara if (!nr_pages) 109747df3ddeSJan Kara nr_pages = get_nr_dirty_pages(); 1098b8c2f347SChristoph Hellwig 1099b8c2f347SChristoph Hellwig rcu_read_lock(); 1100b8c2f347SChristoph Hellwig list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { 1101b8c2f347SChristoph Hellwig if (!bdi_has_dirty_io(bdi)) 1102b8c2f347SChristoph Hellwig continue; 11030e175a18SCurt Wohlgemuth __bdi_start_writeback(bdi, nr_pages, false, reason); 1104b8c2f347SChristoph Hellwig } 1105b8c2f347SChristoph Hellwig rcu_read_unlock(); 110603ba3782SJens Axboe } 110703ba3782SJens Axboe 110803ba3782SJens Axboe static noinline void block_dump___mark_inode_dirty(struct inode *inode) 110903ba3782SJens Axboe { 111003ba3782SJens Axboe if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { 111103ba3782SJens Axboe struct dentry *dentry; 111203ba3782SJens Axboe const char *name = "?"; 111303ba3782SJens Axboe 111403ba3782SJens Axboe dentry = d_find_alias(inode); 111503ba3782SJens Axboe if (dentry) { 111603ba3782SJens Axboe spin_lock(&dentry->d_lock); 111703ba3782SJens Axboe name = (const char *) dentry->d_name.name; 111803ba3782SJens Axboe } 111903ba3782SJens Axboe printk(KERN_DEBUG 112003ba3782SJens Axboe "%s(%d): dirtied inode %lu (%s) on %s\n", 112103ba3782SJens Axboe current->comm, task_pid_nr(current), inode->i_ino, 112203ba3782SJens Axboe name, inode->i_sb->s_id); 112303ba3782SJens Axboe if (dentry) { 112403ba3782SJens Axboe spin_unlock(&dentry->d_lock); 112503ba3782SJens Axboe dput(dentry); 112603ba3782SJens Axboe } 112703ba3782SJens Axboe } 112803ba3782SJens Axboe } 112903ba3782SJens Axboe 113003ba3782SJens Axboe /** 113103ba3782SJens Axboe * __mark_inode_dirty - internal function 113203ba3782SJens Axboe * @inode: inode to mark 113303ba3782SJens Axboe * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) 113403ba3782SJens Axboe * Mark an inode as dirty. Callers should use mark_inode_dirty or 113503ba3782SJens Axboe * mark_inode_dirty_sync. 113603ba3782SJens Axboe * 113703ba3782SJens Axboe * Put the inode on the super block's dirty list. 113803ba3782SJens Axboe * 113903ba3782SJens Axboe * CAREFUL! We mark it dirty unconditionally, but move it onto the 114003ba3782SJens Axboe * dirty list only if it is hashed or if it refers to a blockdev. 114103ba3782SJens Axboe * If it was not hashed, it will never be added to the dirty list 114203ba3782SJens Axboe * even if it is later hashed, as it will have been marked dirty already. 114303ba3782SJens Axboe * 114403ba3782SJens Axboe * In short, make sure you hash any inodes _before_ you start marking 114503ba3782SJens Axboe * them dirty. 114603ba3782SJens Axboe * 114703ba3782SJens Axboe * Note that for blockdevs, inode->dirtied_when represents the dirtying time of 114803ba3782SJens Axboe * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of 114903ba3782SJens Axboe * the kernel-internal blockdev inode represents the dirtying time of the 115003ba3782SJens Axboe * blockdev's pages. This is why for I_DIRTY_PAGES we always use 115103ba3782SJens Axboe * page->mapping->host, so the page-dirtying time is recorded in the internal 115203ba3782SJens Axboe * blockdev inode. 115303ba3782SJens Axboe */ 115403ba3782SJens Axboe void __mark_inode_dirty(struct inode *inode, int flags) 115503ba3782SJens Axboe { 115603ba3782SJens Axboe struct super_block *sb = inode->i_sb; 1157253c34e9SArtem Bityutskiy struct backing_dev_info *bdi = NULL; 115803ba3782SJens Axboe 115903ba3782SJens Axboe /* 116003ba3782SJens Axboe * Don't do this for I_DIRTY_PAGES - that doesn't actually 116103ba3782SJens Axboe * dirty the inode itself 116203ba3782SJens Axboe */ 116303ba3782SJens Axboe if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { 11649fb0a7daSTejun Heo trace_writeback_dirty_inode_start(inode, flags); 11659fb0a7daSTejun Heo 116603ba3782SJens Axboe if (sb->s_op->dirty_inode) 1167aa385729SChristoph Hellwig sb->s_op->dirty_inode(inode, flags); 11689fb0a7daSTejun Heo 11699fb0a7daSTejun Heo trace_writeback_dirty_inode(inode, flags); 117003ba3782SJens Axboe } 117103ba3782SJens Axboe 117203ba3782SJens Axboe /* 11739c6ac78eSTejun Heo * Paired with smp_mb() in __writeback_single_inode() for the 11749c6ac78eSTejun Heo * following lockless i_state test. See there for details. 117503ba3782SJens Axboe */ 117603ba3782SJens Axboe smp_mb(); 117703ba3782SJens Axboe 117803ba3782SJens Axboe if ((inode->i_state & flags) == flags) 117903ba3782SJens Axboe return; 118003ba3782SJens Axboe 118103ba3782SJens Axboe if (unlikely(block_dump)) 118203ba3782SJens Axboe block_dump___mark_inode_dirty(inode); 118303ba3782SJens Axboe 1184250df6edSDave Chinner spin_lock(&inode->i_lock); 118503ba3782SJens Axboe if ((inode->i_state & flags) != flags) { 118603ba3782SJens Axboe const int was_dirty = inode->i_state & I_DIRTY; 118703ba3782SJens Axboe 118803ba3782SJens Axboe inode->i_state |= flags; 118903ba3782SJens Axboe 119003ba3782SJens Axboe /* 119103ba3782SJens Axboe * If the inode is being synced, just update its dirty state. 119203ba3782SJens Axboe * The unlocker will place the inode on the appropriate 119303ba3782SJens Axboe * superblock list, based upon its state. 119403ba3782SJens Axboe */ 119503ba3782SJens Axboe if (inode->i_state & I_SYNC) 1196250df6edSDave Chinner goto out_unlock_inode; 119703ba3782SJens Axboe 119803ba3782SJens Axboe /* 119903ba3782SJens Axboe * Only add valid (hashed) inodes to the superblock's 120003ba3782SJens Axboe * dirty list. Add blockdev inodes as well. 120103ba3782SJens Axboe */ 120203ba3782SJens Axboe if (!S_ISBLK(inode->i_mode)) { 12031d3382cbSAl Viro if (inode_unhashed(inode)) 1204250df6edSDave Chinner goto out_unlock_inode; 120503ba3782SJens Axboe } 1206a4ffdde6SAl Viro if (inode->i_state & I_FREEING) 1207250df6edSDave Chinner goto out_unlock_inode; 120803ba3782SJens Axboe 120903ba3782SJens Axboe /* 121003ba3782SJens Axboe * If the inode was already on b_dirty/b_io/b_more_io, don't 121103ba3782SJens Axboe * reposition it (that would break b_dirty time-ordering). 121203ba3782SJens Axboe */ 121303ba3782SJens Axboe if (!was_dirty) { 1214a66979abSDave Chinner bool wakeup_bdi = false; 1215253c34e9SArtem Bityutskiy bdi = inode_to_bdi(inode); 1216500b067cSJens Axboe 1217146d7009SJunxiao Bi spin_unlock(&inode->i_lock); 1218146d7009SJunxiao Bi spin_lock(&bdi->wb.list_lock); 1219253c34e9SArtem Bityutskiy if (bdi_cap_writeback_dirty(bdi)) { 1220253c34e9SArtem Bityutskiy WARN(!test_bit(BDI_registered, &bdi->state), 1221253c34e9SArtem Bityutskiy "bdi-%s not registered\n", bdi->name); 1222253c34e9SArtem Bityutskiy 1223253c34e9SArtem Bityutskiy /* 1224253c34e9SArtem Bityutskiy * If this is the first dirty inode for this 1225253c34e9SArtem Bityutskiy * bdi, we have to wake-up the corresponding 1226253c34e9SArtem Bityutskiy * bdi thread to make sure background 1227253c34e9SArtem Bityutskiy * write-back happens later. 1228253c34e9SArtem Bityutskiy */ 1229253c34e9SArtem Bityutskiy if (!wb_has_dirty_io(&bdi->wb)) 1230253c34e9SArtem Bityutskiy wakeup_bdi = true; 1231500b067cSJens Axboe } 123203ba3782SJens Axboe 123303ba3782SJens Axboe inode->dirtied_when = jiffies; 12347ccf19a8SNick Piggin list_move(&inode->i_wb_list, &bdi->wb.b_dirty); 1235f758eeabSChristoph Hellwig spin_unlock(&bdi->wb.list_lock); 1236253c34e9SArtem Bityutskiy 1237253c34e9SArtem Bityutskiy if (wakeup_bdi) 12386467716aSArtem Bityutskiy bdi_wakeup_thread_delayed(bdi); 1239a66979abSDave Chinner return; 1240a66979abSDave Chinner } 1241a66979abSDave Chinner } 1242a66979abSDave Chinner out_unlock_inode: 1243a66979abSDave Chinner spin_unlock(&inode->i_lock); 1244a66979abSDave Chinner 124503ba3782SJens Axboe } 124603ba3782SJens Axboe EXPORT_SYMBOL(__mark_inode_dirty); 124703ba3782SJens Axboe 1248b6e51316SJens Axboe static void wait_sb_inodes(struct super_block *sb) 124966f3b8e2SJens Axboe { 125038f21977SNick Piggin struct inode *inode, *old_inode = NULL; 125138f21977SNick Piggin 125203ba3782SJens Axboe /* 125303ba3782SJens Axboe * We need to be protected against the filesystem going from 125403ba3782SJens Axboe * r/o to r/w or vice versa. 125503ba3782SJens Axboe */ 1256b6e51316SJens Axboe WARN_ON(!rwsem_is_locked(&sb->s_umount)); 125703ba3782SJens Axboe 125855fa6091SDave Chinner spin_lock(&inode_sb_list_lock); 125966f3b8e2SJens Axboe 126038f21977SNick Piggin /* 126138f21977SNick Piggin * Data integrity sync. Must wait for all pages under writeback, 126238f21977SNick Piggin * because there may have been pages dirtied before our sync 126338f21977SNick Piggin * call, but which had writeout started before we write it out. 126438f21977SNick Piggin * In which case, the inode may not be on the dirty list, but 126538f21977SNick Piggin * we still have to wait for that writeout. 126638f21977SNick Piggin */ 1267b6e51316SJens Axboe list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { 1268250df6edSDave Chinner struct address_space *mapping = inode->i_mapping; 126938f21977SNick Piggin 1270250df6edSDave Chinner spin_lock(&inode->i_lock); 1271250df6edSDave Chinner if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) || 1272250df6edSDave Chinner (mapping->nrpages == 0)) { 1273250df6edSDave Chinner spin_unlock(&inode->i_lock); 127438f21977SNick Piggin continue; 1275250df6edSDave Chinner } 127638f21977SNick Piggin __iget(inode); 1277250df6edSDave Chinner spin_unlock(&inode->i_lock); 127855fa6091SDave Chinner spin_unlock(&inode_sb_list_lock); 127955fa6091SDave Chinner 128038f21977SNick Piggin /* 128155fa6091SDave Chinner * We hold a reference to 'inode' so it couldn't have been 128255fa6091SDave Chinner * removed from s_inodes list while we dropped the 128355fa6091SDave Chinner * inode_sb_list_lock. We cannot iput the inode now as we can 128455fa6091SDave Chinner * be holding the last reference and we cannot iput it under 128555fa6091SDave Chinner * inode_sb_list_lock. So we keep the reference and iput it 128655fa6091SDave Chinner * later. 128738f21977SNick Piggin */ 128838f21977SNick Piggin iput(old_inode); 128938f21977SNick Piggin old_inode = inode; 129038f21977SNick Piggin 129138f21977SNick Piggin filemap_fdatawait(mapping); 129238f21977SNick Piggin 129338f21977SNick Piggin cond_resched(); 129438f21977SNick Piggin 129555fa6091SDave Chinner spin_lock(&inode_sb_list_lock); 129638f21977SNick Piggin } 129755fa6091SDave Chinner spin_unlock(&inode_sb_list_lock); 129838f21977SNick Piggin iput(old_inode); 129966f3b8e2SJens Axboe } 13001da177e4SLinus Torvalds 1301d8a8559cSJens Axboe /** 13023259f8beSChris Mason * writeback_inodes_sb_nr - writeback dirty inodes from given super_block 1303d8a8559cSJens Axboe * @sb: the superblock 13043259f8beSChris Mason * @nr: the number of pages to write 1305786228abSMarcos Paulo de Souza * @reason: reason why some writeback work initiated 13061da177e4SLinus Torvalds * 1307d8a8559cSJens Axboe * Start writeback on some inodes on this super_block. No guarantees are made 1308d8a8559cSJens Axboe * on how many (if any) will be written, and this function does not wait 13093259f8beSChris Mason * for IO completion of submitted IO. 13101da177e4SLinus Torvalds */ 13110e175a18SCurt Wohlgemuth void writeback_inodes_sb_nr(struct super_block *sb, 13120e175a18SCurt Wohlgemuth unsigned long nr, 13130e175a18SCurt Wohlgemuth enum wb_reason reason) 13141da177e4SLinus Torvalds { 131583ba7b07SChristoph Hellwig DECLARE_COMPLETION_ONSTACK(done); 131683ba7b07SChristoph Hellwig struct wb_writeback_work work = { 13173c4d7165SChristoph Hellwig .sb = sb, 13183c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_NONE, 13196e6938b6SWu Fengguang .tagged_writepages = 1, 132083ba7b07SChristoph Hellwig .done = &done, 13213259f8beSChris Mason .nr_pages = nr, 13220e175a18SCurt Wohlgemuth .reason = reason, 13233c4d7165SChristoph Hellwig }; 13240e3c9a22SJens Axboe 13256eedc701SJan Kara if (sb->s_bdi == &noop_backing_dev_info) 13266eedc701SJan Kara return; 1327cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 132883ba7b07SChristoph Hellwig bdi_queue_work(sb->s_bdi, &work); 132983ba7b07SChristoph Hellwig wait_for_completion(&done); 13301da177e4SLinus Torvalds } 13313259f8beSChris Mason EXPORT_SYMBOL(writeback_inodes_sb_nr); 13323259f8beSChris Mason 13333259f8beSChris Mason /** 13343259f8beSChris Mason * writeback_inodes_sb - writeback dirty inodes from given super_block 13353259f8beSChris Mason * @sb: the superblock 1336786228abSMarcos Paulo de Souza * @reason: reason why some writeback work was initiated 13373259f8beSChris Mason * 13383259f8beSChris Mason * Start writeback on some inodes on this super_block. No guarantees are made 13393259f8beSChris Mason * on how many (if any) will be written, and this function does not wait 13403259f8beSChris Mason * for IO completion of submitted IO. 13413259f8beSChris Mason */ 13420e175a18SCurt Wohlgemuth void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) 13433259f8beSChris Mason { 13440e175a18SCurt Wohlgemuth return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); 13453259f8beSChris Mason } 1346d8a8559cSJens Axboe EXPORT_SYMBOL(writeback_inodes_sb); 1347d8a8559cSJens Axboe 1348d8a8559cSJens Axboe /** 134910ee27a0SMiao Xie * try_to_writeback_inodes_sb_nr - try to start writeback if none underway 13503259f8beSChris Mason * @sb: the superblock 13513259f8beSChris Mason * @nr: the number of pages to write 135210ee27a0SMiao Xie * @reason: the reason of writeback 13533259f8beSChris Mason * 135410ee27a0SMiao Xie * Invoke writeback_inodes_sb_nr if no writeback is currently underway. 13553259f8beSChris Mason * Returns 1 if writeback was started, 0 if not. 13563259f8beSChris Mason */ 135710ee27a0SMiao Xie int try_to_writeback_inodes_sb_nr(struct super_block *sb, 13580e175a18SCurt Wohlgemuth unsigned long nr, 13590e175a18SCurt Wohlgemuth enum wb_reason reason) 13603259f8beSChris Mason { 136110ee27a0SMiao Xie if (writeback_in_progress(sb->s_bdi)) 136210ee27a0SMiao Xie return 1; 136310ee27a0SMiao Xie 136410ee27a0SMiao Xie if (!down_read_trylock(&sb->s_umount)) 136510ee27a0SMiao Xie return 0; 136610ee27a0SMiao Xie 13670e175a18SCurt Wohlgemuth writeback_inodes_sb_nr(sb, nr, reason); 13683259f8beSChris Mason up_read(&sb->s_umount); 13693259f8beSChris Mason return 1; 13703259f8beSChris Mason } 137110ee27a0SMiao Xie EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr); 137210ee27a0SMiao Xie 137310ee27a0SMiao Xie /** 137410ee27a0SMiao Xie * try_to_writeback_inodes_sb - try to start writeback if none underway 137510ee27a0SMiao Xie * @sb: the superblock 137610ee27a0SMiao Xie * @reason: reason why some writeback work was initiated 137710ee27a0SMiao Xie * 137810ee27a0SMiao Xie * Implement by try_to_writeback_inodes_sb_nr() 137910ee27a0SMiao Xie * Returns 1 if writeback was started, 0 if not. 138010ee27a0SMiao Xie */ 138110ee27a0SMiao Xie int try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) 138210ee27a0SMiao Xie { 138310ee27a0SMiao Xie return try_to_writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); 138410ee27a0SMiao Xie } 138510ee27a0SMiao Xie EXPORT_SYMBOL(try_to_writeback_inodes_sb); 13863259f8beSChris Mason 13873259f8beSChris Mason /** 1388d8a8559cSJens Axboe * sync_inodes_sb - sync sb inode pages 1389d8a8559cSJens Axboe * @sb: the superblock 1390d8a8559cSJens Axboe * 1391d8a8559cSJens Axboe * This function writes and waits on any dirty inode belonging to this 13920dc83bd3SJan Kara * super_block. 1393d8a8559cSJens Axboe */ 13940dc83bd3SJan Kara void sync_inodes_sb(struct super_block *sb) 1395d8a8559cSJens Axboe { 139683ba7b07SChristoph Hellwig DECLARE_COMPLETION_ONSTACK(done); 139783ba7b07SChristoph Hellwig struct wb_writeback_work work = { 13983c4d7165SChristoph Hellwig .sb = sb, 13993c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_ALL, 14003c4d7165SChristoph Hellwig .nr_pages = LONG_MAX, 14013c4d7165SChristoph Hellwig .range_cyclic = 0, 140283ba7b07SChristoph Hellwig .done = &done, 14030e175a18SCurt Wohlgemuth .reason = WB_REASON_SYNC, 14047747bd4bSDave Chinner .for_sync = 1, 14053c4d7165SChristoph Hellwig }; 14063c4d7165SChristoph Hellwig 14076eedc701SJan Kara /* Nothing to do? */ 14086eedc701SJan Kara if (sb->s_bdi == &noop_backing_dev_info) 14096eedc701SJan Kara return; 1410cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 1411cf37e972SChristoph Hellwig 141283ba7b07SChristoph Hellwig bdi_queue_work(sb->s_bdi, &work); 141383ba7b07SChristoph Hellwig wait_for_completion(&done); 141483ba7b07SChristoph Hellwig 1415b6e51316SJens Axboe wait_sb_inodes(sb); 1416d8a8559cSJens Axboe } 1417d8a8559cSJens Axboe EXPORT_SYMBOL(sync_inodes_sb); 14181da177e4SLinus Torvalds 14191da177e4SLinus Torvalds /** 14201da177e4SLinus Torvalds * write_inode_now - write an inode to disk 14211da177e4SLinus Torvalds * @inode: inode to write to disk 14221da177e4SLinus Torvalds * @sync: whether the write should be synchronous or not 14231da177e4SLinus Torvalds * 14247f04c26dSAndrea Arcangeli * This function commits an inode to disk immediately if it is dirty. This is 14257f04c26dSAndrea Arcangeli * primarily needed by knfsd. 14267f04c26dSAndrea Arcangeli * 14277f04c26dSAndrea Arcangeli * The caller must either have a ref on the inode or must have set I_WILL_FREE. 14281da177e4SLinus Torvalds */ 14291da177e4SLinus Torvalds int write_inode_now(struct inode *inode, int sync) 14301da177e4SLinus Torvalds { 1431f758eeabSChristoph Hellwig struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; 14321da177e4SLinus Torvalds struct writeback_control wbc = { 14331da177e4SLinus Torvalds .nr_to_write = LONG_MAX, 143418914b18SMike Galbraith .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, 1435111ebb6eSOGAWA Hirofumi .range_start = 0, 1436111ebb6eSOGAWA Hirofumi .range_end = LLONG_MAX, 14371da177e4SLinus Torvalds }; 14381da177e4SLinus Torvalds 14391da177e4SLinus Torvalds if (!mapping_cap_writeback_dirty(inode->i_mapping)) 144049364ce2SAndrew Morton wbc.nr_to_write = 0; 14411da177e4SLinus Torvalds 14421da177e4SLinus Torvalds might_sleep(); 14434f8ad655SJan Kara return writeback_single_inode(inode, wb, &wbc); 14441da177e4SLinus Torvalds } 14451da177e4SLinus Torvalds EXPORT_SYMBOL(write_inode_now); 14461da177e4SLinus Torvalds 14471da177e4SLinus Torvalds /** 14481da177e4SLinus Torvalds * sync_inode - write an inode and its pages to disk. 14491da177e4SLinus Torvalds * @inode: the inode to sync 14501da177e4SLinus Torvalds * @wbc: controls the writeback mode 14511da177e4SLinus Torvalds * 14521da177e4SLinus Torvalds * sync_inode() will write an inode and its pages to disk. It will also 14531da177e4SLinus Torvalds * correctly update the inode on its superblock's dirty inode lists and will 14541da177e4SLinus Torvalds * update inode->i_state. 14551da177e4SLinus Torvalds * 14561da177e4SLinus Torvalds * The caller must have a ref on the inode. 14571da177e4SLinus Torvalds */ 14581da177e4SLinus Torvalds int sync_inode(struct inode *inode, struct writeback_control *wbc) 14591da177e4SLinus Torvalds { 14604f8ad655SJan Kara return writeback_single_inode(inode, &inode_to_bdi(inode)->wb, wbc); 14611da177e4SLinus Torvalds } 14621da177e4SLinus Torvalds EXPORT_SYMBOL(sync_inode); 1463c3765016SChristoph Hellwig 1464c3765016SChristoph Hellwig /** 1465c691b9d9SAndrew Morton * sync_inode_metadata - write an inode to disk 1466c3765016SChristoph Hellwig * @inode: the inode to sync 1467c3765016SChristoph Hellwig * @wait: wait for I/O to complete. 1468c3765016SChristoph Hellwig * 1469c691b9d9SAndrew Morton * Write an inode to disk and adjust its dirty state after completion. 1470c3765016SChristoph Hellwig * 1471c3765016SChristoph Hellwig * Note: only writes the actual inode, no associated data or other metadata. 1472c3765016SChristoph Hellwig */ 1473c3765016SChristoph Hellwig int sync_inode_metadata(struct inode *inode, int wait) 1474c3765016SChristoph Hellwig { 1475c3765016SChristoph Hellwig struct writeback_control wbc = { 1476c3765016SChristoph Hellwig .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE, 1477c3765016SChristoph Hellwig .nr_to_write = 0, /* metadata-only */ 1478c3765016SChristoph Hellwig }; 1479c3765016SChristoph Hellwig 1480c3765016SChristoph Hellwig return sync_inode(inode, &wbc); 1481c3765016SChristoph Hellwig } 1482c3765016SChristoph Hellwig EXPORT_SYMBOL(sync_inode_metadata); 1483