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 69692ebd17SJan Kara static inline struct backing_dev_info *inode_to_bdi(struct inode *inode) 70692ebd17SJan Kara { 71692ebd17SJan Kara struct super_block *sb = inode->i_sb; 72495a276eSChristoph Hellwig #ifdef CONFIG_BLOCK 73a8855990SJan Kara if (sb_is_blkdev_sb(sb)) 74495a276eSChristoph Hellwig return blk_get_backing_dev_info(I_BDEV(inode)); 75495a276eSChristoph Hellwig #endif 76692ebd17SJan Kara return sb->s_bdi; 77692ebd17SJan Kara } 78692ebd17SJan Kara 797ccf19a8SNick Piggin static inline struct inode *wb_inode(struct list_head *head) 807ccf19a8SNick Piggin { 817ccf19a8SNick Piggin return list_entry(head, struct inode, i_wb_list); 827ccf19a8SNick Piggin } 837ccf19a8SNick Piggin 8415eb77a0SWu Fengguang /* 8515eb77a0SWu Fengguang * Include the creation of the trace points after defining the 8615eb77a0SWu Fengguang * wb_writeback_work structure and inline functions so that the definition 8715eb77a0SWu Fengguang * remains local to this file. 8815eb77a0SWu Fengguang */ 8915eb77a0SWu Fengguang #define CREATE_TRACE_POINTS 9015eb77a0SWu Fengguang #include <trace/events/writeback.h> 9115eb77a0SWu Fengguang 92774016b2SSteven Whitehouse EXPORT_TRACEPOINT_SYMBOL_GPL(wbc_writepage); 93774016b2SSteven Whitehouse 945acda9d1SJan Kara static void bdi_wakeup_thread(struct backing_dev_info *bdi) 955acda9d1SJan Kara { 965acda9d1SJan Kara spin_lock_bh(&bdi->wb_lock); 975acda9d1SJan Kara if (test_bit(BDI_registered, &bdi->state)) 985acda9d1SJan Kara mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0); 995acda9d1SJan Kara spin_unlock_bh(&bdi->wb_lock); 1005acda9d1SJan Kara } 1015acda9d1SJan 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); 1085acda9d1SJan Kara if (!test_bit(BDI_registered, &bdi->state)) { 1095acda9d1SJan Kara if (work->done) 1105acda9d1SJan Kara complete(work->done); 1115acda9d1SJan Kara goto out_unlock; 1125acda9d1SJan Kara } 1136585027aSJan Kara list_add_tail(&work->list, &bdi->work_list); 114839a8e86STejun Heo mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0); 1155acda9d1SJan Kara out_unlock: 1165acda9d1SJan Kara spin_unlock_bh(&bdi->wb_lock); 11703ba3782SJens Axboe } 1181da177e4SLinus Torvalds 11983ba7b07SChristoph Hellwig static void 12083ba7b07SChristoph Hellwig __bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, 1210e175a18SCurt Wohlgemuth bool range_cyclic, enum wb_reason reason) 1221da177e4SLinus Torvalds { 12383ba7b07SChristoph Hellwig struct wb_writeback_work *work; 12403ba3782SJens Axboe 125bcddc3f0SJens Axboe /* 126bcddc3f0SJens Axboe * This is WB_SYNC_NONE writeback, so if allocation fails just 127bcddc3f0SJens Axboe * wakeup the thread for old dirty data writeback 128bcddc3f0SJens Axboe */ 12983ba7b07SChristoph Hellwig work = kzalloc(sizeof(*work), GFP_ATOMIC); 13083ba7b07SChristoph Hellwig if (!work) { 131455b2864SDave Chinner trace_writeback_nowork(bdi); 1325acda9d1SJan Kara bdi_wakeup_thread(bdi); 13383ba7b07SChristoph Hellwig return; 13483ba7b07SChristoph Hellwig } 13583ba7b07SChristoph Hellwig 13683ba7b07SChristoph Hellwig work->sync_mode = WB_SYNC_NONE; 13783ba7b07SChristoph Hellwig work->nr_pages = nr_pages; 13883ba7b07SChristoph Hellwig work->range_cyclic = range_cyclic; 1390e175a18SCurt Wohlgemuth work->reason = reason; 14083ba7b07SChristoph Hellwig 141f11fcae8SJens Axboe bdi_queue_work(bdi, work); 14203ba3782SJens Axboe } 143b6e51316SJens Axboe 144b6e51316SJens Axboe /** 145b6e51316SJens Axboe * bdi_start_writeback - start writeback 146b6e51316SJens Axboe * @bdi: the backing device to write from 147b6e51316SJens Axboe * @nr_pages: the number of pages to write 148786228abSMarcos Paulo de Souza * @reason: reason why some writeback work was initiated 149b6e51316SJens Axboe * 150b6e51316SJens Axboe * Description: 151b6e51316SJens Axboe * This does WB_SYNC_NONE opportunistic writeback. The IO is only 15225985edcSLucas De Marchi * started when this function returns, we make no guarantees on 1530e3c9a22SJens Axboe * completion. Caller need not hold sb s_umount semaphore. 154b6e51316SJens Axboe * 155b6e51316SJens Axboe */ 1560e175a18SCurt Wohlgemuth void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, 1570e175a18SCurt Wohlgemuth enum wb_reason reason) 158b6e51316SJens Axboe { 1590e175a18SCurt Wohlgemuth __bdi_start_writeback(bdi, nr_pages, true, reason); 160d3ddec76SWu Fengguang } 161d3ddec76SWu Fengguang 162c5444198SChristoph Hellwig /** 163c5444198SChristoph Hellwig * bdi_start_background_writeback - start background writeback 164c5444198SChristoph Hellwig * @bdi: the backing device to write from 165c5444198SChristoph Hellwig * 166c5444198SChristoph Hellwig * Description: 1676585027aSJan Kara * This makes sure WB_SYNC_NONE background writeback happens. When 1686585027aSJan Kara * this function returns, it is only guaranteed that for given BDI 1696585027aSJan Kara * some IO is happening if we are over background dirty threshold. 1706585027aSJan Kara * Caller need not hold sb s_umount semaphore. 171c5444198SChristoph Hellwig */ 172c5444198SChristoph Hellwig void bdi_start_background_writeback(struct backing_dev_info *bdi) 173c5444198SChristoph Hellwig { 1746585027aSJan Kara /* 1756585027aSJan Kara * We just wake up the flusher thread. It will perform background 1766585027aSJan Kara * writeback as soon as there is no other work to do. 1776585027aSJan Kara */ 17871927e84SWu Fengguang trace_writeback_wake_background(bdi); 1795acda9d1SJan Kara bdi_wakeup_thread(bdi); 1801da177e4SLinus Torvalds } 1811da177e4SLinus Torvalds 1821da177e4SLinus Torvalds /* 183a66979abSDave Chinner * Remove the inode from the writeback list it is on. 184a66979abSDave Chinner */ 185a66979abSDave Chinner void inode_wb_list_del(struct inode *inode) 186a66979abSDave Chinner { 187f758eeabSChristoph Hellwig struct backing_dev_info *bdi = inode_to_bdi(inode); 188a66979abSDave Chinner 189f758eeabSChristoph Hellwig spin_lock(&bdi->wb.list_lock); 190f758eeabSChristoph Hellwig list_del_init(&inode->i_wb_list); 191f758eeabSChristoph Hellwig spin_unlock(&bdi->wb.list_lock); 192f758eeabSChristoph Hellwig } 193a66979abSDave Chinner 194a66979abSDave Chinner /* 1956610a0bcSAndrew Morton * Redirty an inode: set its when-it-was dirtied timestamp and move it to the 1966610a0bcSAndrew Morton * furthest end of its superblock's dirty-inode list. 1976610a0bcSAndrew Morton * 1986610a0bcSAndrew Morton * Before stamping the inode's ->dirtied_when, we check to see whether it is 19966f3b8e2SJens Axboe * already the most-recently-dirtied inode on the b_dirty list. If that is 2006610a0bcSAndrew Morton * the case then the inode must have been redirtied while it was being written 2016610a0bcSAndrew Morton * out and we don't reset its dirtied_when. 2026610a0bcSAndrew Morton */ 203f758eeabSChristoph Hellwig static void redirty_tail(struct inode *inode, struct bdi_writeback *wb) 2046610a0bcSAndrew Morton { 205f758eeabSChristoph Hellwig assert_spin_locked(&wb->list_lock); 20603ba3782SJens Axboe if (!list_empty(&wb->b_dirty)) { 20766f3b8e2SJens Axboe struct inode *tail; 2086610a0bcSAndrew Morton 2097ccf19a8SNick Piggin tail = wb_inode(wb->b_dirty.next); 21066f3b8e2SJens Axboe if (time_before(inode->dirtied_when, tail->dirtied_when)) 2116610a0bcSAndrew Morton inode->dirtied_when = jiffies; 2126610a0bcSAndrew Morton } 2137ccf19a8SNick Piggin list_move(&inode->i_wb_list, &wb->b_dirty); 2146610a0bcSAndrew Morton } 2156610a0bcSAndrew Morton 2166610a0bcSAndrew Morton /* 21766f3b8e2SJens Axboe * requeue inode for re-scanning after bdi->b_io list is exhausted. 218c986d1e2SAndrew Morton */ 219f758eeabSChristoph Hellwig static void requeue_io(struct inode *inode, struct bdi_writeback *wb) 220c986d1e2SAndrew Morton { 221f758eeabSChristoph Hellwig assert_spin_locked(&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 { 227365b94aeSJan Kara inode->i_state &= ~I_SYNC; 2284eff96ddSJan Kara /* If inode is clean an unused, put it into LRU now... */ 2294eff96ddSJan Kara inode_add_lru(inode); 230365b94aeSJan Kara /* Waiters must see I_SYNC cleared before being woken up */ 2311c0eeaf5SJoern Engel smp_mb(); 2321c0eeaf5SJoern Engel wake_up_bit(&inode->i_state, __I_SYNC); 2331c0eeaf5SJoern Engel } 2341c0eeaf5SJoern Engel 235d2caa3c5SJeff Layton static bool inode_dirtied_after(struct inode *inode, unsigned long t) 236d2caa3c5SJeff Layton { 237d2caa3c5SJeff Layton bool ret = time_after(inode->dirtied_when, t); 238d2caa3c5SJeff Layton #ifndef CONFIG_64BIT 239d2caa3c5SJeff Layton /* 240d2caa3c5SJeff Layton * For inodes being constantly redirtied, dirtied_when can get stuck. 241d2caa3c5SJeff Layton * It _appears_ to be in the future, but is actually in distant past. 242d2caa3c5SJeff Layton * This test is necessary to prevent such wrapped-around relative times 2435b0830cbSJens Axboe * from permanently stopping the whole bdi writeback. 244d2caa3c5SJeff Layton */ 245d2caa3c5SJeff Layton ret = ret && time_before_eq(inode->dirtied_when, jiffies); 246d2caa3c5SJeff Layton #endif 247d2caa3c5SJeff Layton return ret; 248d2caa3c5SJeff Layton } 249d2caa3c5SJeff Layton 250c986d1e2SAndrew Morton /* 2510e2f2b23SWang Sheng-Hui * Move expired (dirtied before work->older_than_this) dirty inodes from 252697e6fedSJan Kara * @delaying_queue to @dispatch_queue. 2532c136579SFengguang Wu */ 254e84d0a4fSWu Fengguang static int move_expired_inodes(struct list_head *delaying_queue, 2552c136579SFengguang Wu struct list_head *dispatch_queue, 256ad4e38ddSCurt Wohlgemuth struct wb_writeback_work *work) 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; 263e84d0a4fSWu Fengguang int moved = 0; 2645c03449dSShaohua Li 2652c136579SFengguang Wu while (!list_empty(delaying_queue)) { 2667ccf19a8SNick Piggin inode = wb_inode(delaying_queue->prev); 2670dc83bd3SJan Kara if (work->older_than_this && 2680dc83bd3SJan Kara inode_dirtied_after(inode, *work->older_than_this)) 2692c136579SFengguang Wu break; 270a8855990SJan Kara list_move(&inode->i_wb_list, &tmp); 271a8855990SJan Kara moved++; 272a8855990SJan Kara if (sb_is_blkdev_sb(inode->i_sb)) 273a8855990SJan Kara continue; 274cf137307SJens Axboe if (sb && sb != inode->i_sb) 275cf137307SJens Axboe do_sb_sort = 1; 276cf137307SJens Axboe sb = inode->i_sb; 2775c03449dSShaohua Li } 2785c03449dSShaohua Li 279cf137307SJens Axboe /* just one sb in list, splice to dispatch_queue and we're done */ 280cf137307SJens Axboe if (!do_sb_sort) { 281cf137307SJens Axboe list_splice(&tmp, dispatch_queue); 282e84d0a4fSWu Fengguang goto out; 283cf137307SJens Axboe } 284cf137307SJens Axboe 2855c03449dSShaohua Li /* Move inodes from one superblock together */ 2865c03449dSShaohua Li while (!list_empty(&tmp)) { 2877ccf19a8SNick Piggin sb = wb_inode(tmp.prev)->i_sb; 2885c03449dSShaohua Li list_for_each_prev_safe(pos, node, &tmp) { 2897ccf19a8SNick Piggin inode = wb_inode(pos); 2905c03449dSShaohua Li if (inode->i_sb == sb) 2917ccf19a8SNick Piggin list_move(&inode->i_wb_list, dispatch_queue); 2922c136579SFengguang Wu } 2932c136579SFengguang Wu } 294e84d0a4fSWu Fengguang out: 295e84d0a4fSWu Fengguang return moved; 2965c03449dSShaohua Li } 2972c136579SFengguang Wu 2982c136579SFengguang Wu /* 2992c136579SFengguang Wu * Queue all expired dirty inodes for io, eldest first. 3004ea879b9SWu Fengguang * Before 3014ea879b9SWu Fengguang * newly dirtied b_dirty b_io b_more_io 3024ea879b9SWu Fengguang * =============> gf edc BA 3034ea879b9SWu Fengguang * After 3044ea879b9SWu Fengguang * newly dirtied b_dirty b_io b_more_io 3054ea879b9SWu Fengguang * =============> g fBAedc 3064ea879b9SWu Fengguang * | 3074ea879b9SWu Fengguang * +--> dequeue for IO 3082c136579SFengguang Wu */ 309ad4e38ddSCurt Wohlgemuth static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work) 3102c136579SFengguang Wu { 311e84d0a4fSWu Fengguang int moved; 312f758eeabSChristoph Hellwig assert_spin_locked(&wb->list_lock); 3134ea879b9SWu Fengguang list_splice_init(&wb->b_more_io, &wb->b_io); 314ad4e38ddSCurt Wohlgemuth moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, work); 315ad4e38ddSCurt Wohlgemuth trace_writeback_queue_io(wb, work, moved); 31666f3b8e2SJens Axboe } 31766f3b8e2SJens Axboe 318a9185b41SChristoph Hellwig static int write_inode(struct inode *inode, struct writeback_control *wbc) 31966f3b8e2SJens Axboe { 3209fb0a7daSTejun Heo int ret; 3219fb0a7daSTejun Heo 3229fb0a7daSTejun Heo if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) { 3239fb0a7daSTejun Heo trace_writeback_write_inode_start(inode, wbc); 3249fb0a7daSTejun Heo ret = inode->i_sb->s_op->write_inode(inode, wbc); 3259fb0a7daSTejun Heo trace_writeback_write_inode(inode, wbc); 3269fb0a7daSTejun Heo return ret; 3279fb0a7daSTejun Heo } 32803ba3782SJens Axboe return 0; 32966f3b8e2SJens Axboe } 33008d8e974SFengguang Wu 3312c136579SFengguang Wu /* 332169ebd90SJan Kara * Wait for writeback on an inode to complete. Called with i_lock held. 333169ebd90SJan Kara * Caller must make sure inode cannot go away when we drop i_lock. 33401c03194SChristoph Hellwig */ 335169ebd90SJan Kara static void __inode_wait_for_writeback(struct inode *inode) 336169ebd90SJan Kara __releases(inode->i_lock) 337169ebd90SJan Kara __acquires(inode->i_lock) 33801c03194SChristoph Hellwig { 33901c03194SChristoph Hellwig DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); 34001c03194SChristoph Hellwig wait_queue_head_t *wqh; 34101c03194SChristoph Hellwig 34201c03194SChristoph Hellwig wqh = bit_waitqueue(&inode->i_state, __I_SYNC); 34358a9d3d8SRichard Kennedy while (inode->i_state & I_SYNC) { 344250df6edSDave Chinner spin_unlock(&inode->i_lock); 34574316201SNeilBrown __wait_on_bit(wqh, &wq, bit_wait, 34674316201SNeilBrown TASK_UNINTERRUPTIBLE); 347250df6edSDave Chinner spin_lock(&inode->i_lock); 34858a9d3d8SRichard Kennedy } 34901c03194SChristoph Hellwig } 35001c03194SChristoph Hellwig 35101c03194SChristoph Hellwig /* 352169ebd90SJan Kara * Wait for writeback on an inode to complete. Caller must have inode pinned. 353169ebd90SJan Kara */ 354169ebd90SJan Kara void inode_wait_for_writeback(struct inode *inode) 355169ebd90SJan Kara { 356169ebd90SJan Kara spin_lock(&inode->i_lock); 357169ebd90SJan Kara __inode_wait_for_writeback(inode); 358169ebd90SJan Kara spin_unlock(&inode->i_lock); 359169ebd90SJan Kara } 360169ebd90SJan Kara 361169ebd90SJan Kara /* 362169ebd90SJan Kara * Sleep until I_SYNC is cleared. This function must be called with i_lock 363169ebd90SJan Kara * held and drops it. It is aimed for callers not holding any inode reference 364169ebd90SJan Kara * so once i_lock is dropped, inode can go away. 365169ebd90SJan Kara */ 366169ebd90SJan Kara static void inode_sleep_on_writeback(struct inode *inode) 367169ebd90SJan Kara __releases(inode->i_lock) 368169ebd90SJan Kara { 369169ebd90SJan Kara DEFINE_WAIT(wait); 370169ebd90SJan Kara wait_queue_head_t *wqh = bit_waitqueue(&inode->i_state, __I_SYNC); 371169ebd90SJan Kara int sleep; 372169ebd90SJan Kara 373169ebd90SJan Kara prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); 374169ebd90SJan Kara sleep = inode->i_state & I_SYNC; 375169ebd90SJan Kara spin_unlock(&inode->i_lock); 376169ebd90SJan Kara if (sleep) 377169ebd90SJan Kara schedule(); 378169ebd90SJan Kara finish_wait(wqh, &wait); 379169ebd90SJan Kara } 380169ebd90SJan Kara 381169ebd90SJan Kara /* 382ccb26b5aSJan Kara * Find proper writeback list for the inode depending on its current state and 383ccb26b5aSJan Kara * possibly also change of its state while we were doing writeback. Here we 384ccb26b5aSJan Kara * handle things such as livelock prevention or fairness of writeback among 385ccb26b5aSJan Kara * inodes. This function can be called only by flusher thread - noone else 386ccb26b5aSJan Kara * processes all inodes in writeback lists and requeueing inodes behind flusher 387ccb26b5aSJan Kara * thread's back can have unexpected consequences. 388ccb26b5aSJan Kara */ 389ccb26b5aSJan Kara static void requeue_inode(struct inode *inode, struct bdi_writeback *wb, 390ccb26b5aSJan Kara struct writeback_control *wbc) 391ccb26b5aSJan Kara { 392ccb26b5aSJan Kara if (inode->i_state & I_FREEING) 393ccb26b5aSJan Kara return; 394ccb26b5aSJan Kara 395ccb26b5aSJan Kara /* 396ccb26b5aSJan Kara * Sync livelock prevention. Each inode is tagged and synced in one 397ccb26b5aSJan Kara * shot. If still dirty, it will be redirty_tail()'ed below. Update 398ccb26b5aSJan Kara * the dirty time to prevent enqueue and sync it again. 399ccb26b5aSJan Kara */ 400ccb26b5aSJan Kara if ((inode->i_state & I_DIRTY) && 401ccb26b5aSJan Kara (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)) 402ccb26b5aSJan Kara inode->dirtied_when = jiffies; 403ccb26b5aSJan Kara 4044f8ad655SJan Kara if (wbc->pages_skipped) { 4054f8ad655SJan Kara /* 4064f8ad655SJan Kara * writeback is not making progress due to locked 4074f8ad655SJan Kara * buffers. Skip this inode for now. 4084f8ad655SJan Kara */ 4094f8ad655SJan Kara redirty_tail(inode, wb); 4104f8ad655SJan Kara return; 4114f8ad655SJan Kara } 4124f8ad655SJan Kara 413ccb26b5aSJan Kara if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) { 414ccb26b5aSJan Kara /* 415ccb26b5aSJan Kara * We didn't write back all the pages. nfs_writepages() 416ccb26b5aSJan Kara * sometimes bales out without doing anything. 417ccb26b5aSJan Kara */ 418ccb26b5aSJan Kara if (wbc->nr_to_write <= 0) { 419ccb26b5aSJan Kara /* Slice used up. Queue for next turn. */ 420ccb26b5aSJan Kara requeue_io(inode, wb); 421ccb26b5aSJan Kara } else { 422ccb26b5aSJan Kara /* 423ccb26b5aSJan Kara * Writeback blocked by something other than 424ccb26b5aSJan Kara * congestion. Delay the inode for some time to 425ccb26b5aSJan Kara * avoid spinning on the CPU (100% iowait) 426ccb26b5aSJan Kara * retrying writeback of the dirty page/inode 427ccb26b5aSJan Kara * that cannot be performed immediately. 428ccb26b5aSJan Kara */ 429ccb26b5aSJan Kara redirty_tail(inode, wb); 430ccb26b5aSJan Kara } 431ccb26b5aSJan Kara } else if (inode->i_state & I_DIRTY) { 432ccb26b5aSJan Kara /* 433ccb26b5aSJan Kara * Filesystems can dirty the inode during writeback operations, 434ccb26b5aSJan Kara * such as delayed allocation during submission or metadata 435ccb26b5aSJan Kara * updates after data IO completion. 436ccb26b5aSJan Kara */ 437ccb26b5aSJan Kara redirty_tail(inode, wb); 438ccb26b5aSJan Kara } else { 439ccb26b5aSJan Kara /* The inode is clean. Remove from writeback lists. */ 440ccb26b5aSJan Kara list_del_init(&inode->i_wb_list); 441ccb26b5aSJan Kara } 442ccb26b5aSJan Kara } 443ccb26b5aSJan Kara 444ccb26b5aSJan Kara /* 4454f8ad655SJan Kara * Write out an inode and its dirty pages. Do not update the writeback list 4464f8ad655SJan Kara * linkage. That is left to the caller. The caller is also responsible for 4474f8ad655SJan Kara * setting I_SYNC flag and calling inode_sync_complete() to clear it. 4481da177e4SLinus Torvalds */ 4491da177e4SLinus Torvalds static int 450cd8ed2a4SYan Hong __writeback_single_inode(struct inode *inode, struct writeback_control *wbc) 4511da177e4SLinus Torvalds { 4521da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 453251d6a47SWu Fengguang long nr_to_write = wbc->nr_to_write; 45401c03194SChristoph Hellwig unsigned dirty; 4551da177e4SLinus Torvalds int ret; 4561da177e4SLinus Torvalds 4574f8ad655SJan Kara WARN_ON(!(inode->i_state & I_SYNC)); 4581da177e4SLinus Torvalds 4599fb0a7daSTejun Heo trace_writeback_single_inode_start(inode, wbc, nr_to_write); 4609fb0a7daSTejun Heo 4611da177e4SLinus Torvalds ret = do_writepages(mapping, wbc); 4621da177e4SLinus Torvalds 46326821ed4SChristoph Hellwig /* 46426821ed4SChristoph Hellwig * Make sure to wait on the data before writing out the metadata. 46526821ed4SChristoph Hellwig * This is important for filesystems that modify metadata on data 4667747bd4bSDave Chinner * I/O completion. We don't do it for sync(2) writeback because it has a 4677747bd4bSDave Chinner * separate, external IO completion path and ->sync_fs for guaranteeing 4687747bd4bSDave Chinner * inode metadata is written back correctly. 46926821ed4SChristoph Hellwig */ 4707747bd4bSDave Chinner if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) { 47126821ed4SChristoph Hellwig int err = filemap_fdatawait(mapping); 4721da177e4SLinus Torvalds if (ret == 0) 4731da177e4SLinus Torvalds ret = err; 4741da177e4SLinus Torvalds } 4751da177e4SLinus Torvalds 4765547e8aaSDmitry Monakhov /* 4775547e8aaSDmitry Monakhov * Some filesystems may redirty the inode during the writeback 4785547e8aaSDmitry Monakhov * due to delalloc, clear dirty metadata flags right before 4795547e8aaSDmitry Monakhov * write_inode() 4805547e8aaSDmitry Monakhov */ 481250df6edSDave Chinner spin_lock(&inode->i_lock); 4829c6ac78eSTejun Heo 4835547e8aaSDmitry Monakhov dirty = inode->i_state & I_DIRTY; 4849c6ac78eSTejun Heo inode->i_state &= ~I_DIRTY; 4859c6ac78eSTejun Heo 4869c6ac78eSTejun Heo /* 4879c6ac78eSTejun Heo * Paired with smp_mb() in __mark_inode_dirty(). This allows 4889c6ac78eSTejun Heo * __mark_inode_dirty() to test i_state without grabbing i_lock - 4899c6ac78eSTejun Heo * either they see the I_DIRTY bits cleared or we see the dirtied 4909c6ac78eSTejun Heo * inode. 4919c6ac78eSTejun Heo * 4929c6ac78eSTejun Heo * I_DIRTY_PAGES is always cleared together above even if @mapping 4939c6ac78eSTejun Heo * still has dirty pages. The flag is reinstated after smp_mb() if 4949c6ac78eSTejun Heo * necessary. This guarantees that either __mark_inode_dirty() 4959c6ac78eSTejun Heo * sees clear I_DIRTY_PAGES or we see PAGECACHE_TAG_DIRTY. 4969c6ac78eSTejun Heo */ 4979c6ac78eSTejun Heo smp_mb(); 4989c6ac78eSTejun Heo 4999c6ac78eSTejun Heo if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) 5009c6ac78eSTejun Heo inode->i_state |= I_DIRTY_PAGES; 5019c6ac78eSTejun Heo 502250df6edSDave Chinner spin_unlock(&inode->i_lock); 5039c6ac78eSTejun Heo 50426821ed4SChristoph Hellwig /* Don't write the inode if only I_DIRTY_PAGES was set */ 50526821ed4SChristoph Hellwig if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { 506a9185b41SChristoph Hellwig int err = write_inode(inode, wbc); 5071da177e4SLinus Torvalds if (ret == 0) 5081da177e4SLinus Torvalds ret = err; 5091da177e4SLinus Torvalds } 5104f8ad655SJan Kara trace_writeback_single_inode(inode, wbc, nr_to_write); 5114f8ad655SJan Kara return ret; 5124f8ad655SJan Kara } 5134f8ad655SJan Kara 5144f8ad655SJan Kara /* 5154f8ad655SJan Kara * Write out an inode's dirty pages. Either the caller has an active reference 5164f8ad655SJan Kara * on the inode or the inode has I_WILL_FREE set. 5174f8ad655SJan Kara * 5184f8ad655SJan Kara * This function is designed to be called for writing back one inode which 5194f8ad655SJan Kara * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode() 5204f8ad655SJan Kara * and does more profound writeback list handling in writeback_sb_inodes(). 5214f8ad655SJan Kara */ 5224f8ad655SJan Kara static int 5234f8ad655SJan Kara writeback_single_inode(struct inode *inode, struct bdi_writeback *wb, 5244f8ad655SJan Kara struct writeback_control *wbc) 5254f8ad655SJan Kara { 5264f8ad655SJan Kara int ret = 0; 5274f8ad655SJan Kara 5284f8ad655SJan Kara spin_lock(&inode->i_lock); 5294f8ad655SJan Kara if (!atomic_read(&inode->i_count)) 5304f8ad655SJan Kara WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); 5314f8ad655SJan Kara else 5324f8ad655SJan Kara WARN_ON(inode->i_state & I_WILL_FREE); 5334f8ad655SJan Kara 5344f8ad655SJan Kara if (inode->i_state & I_SYNC) { 5354f8ad655SJan Kara if (wbc->sync_mode != WB_SYNC_ALL) 5364f8ad655SJan Kara goto out; 5374f8ad655SJan Kara /* 538169ebd90SJan Kara * It's a data-integrity sync. We must wait. Since callers hold 539169ebd90SJan Kara * inode reference or inode has I_WILL_FREE set, it cannot go 540169ebd90SJan Kara * away under us. 5414f8ad655SJan Kara */ 542169ebd90SJan Kara __inode_wait_for_writeback(inode); 5434f8ad655SJan Kara } 5444f8ad655SJan Kara WARN_ON(inode->i_state & I_SYNC); 5454f8ad655SJan Kara /* 546f9b0e058SJan Kara * Skip inode if it is clean and we have no outstanding writeback in 547f9b0e058SJan Kara * WB_SYNC_ALL mode. We don't want to mess with writeback lists in this 548f9b0e058SJan Kara * function since flusher thread may be doing for example sync in 549f9b0e058SJan Kara * parallel and if we move the inode, it could get skipped. So here we 550f9b0e058SJan Kara * make sure inode is on some writeback list and leave it there unless 551f9b0e058SJan Kara * we have completely cleaned the inode. 5524f8ad655SJan Kara */ 553f9b0e058SJan Kara if (!(inode->i_state & I_DIRTY) && 554f9b0e058SJan Kara (wbc->sync_mode != WB_SYNC_ALL || 555f9b0e058SJan Kara !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK))) 5564f8ad655SJan Kara goto out; 5574f8ad655SJan Kara inode->i_state |= I_SYNC; 5584f8ad655SJan Kara spin_unlock(&inode->i_lock); 5594f8ad655SJan Kara 560cd8ed2a4SYan Hong ret = __writeback_single_inode(inode, wbc); 5611da177e4SLinus Torvalds 562f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 563250df6edSDave Chinner spin_lock(&inode->i_lock); 5644f8ad655SJan Kara /* 5654f8ad655SJan Kara * If inode is clean, remove it from writeback lists. Otherwise don't 5664f8ad655SJan Kara * touch it. See comment above for explanation. 5674f8ad655SJan Kara */ 5684f8ad655SJan Kara if (!(inode->i_state & I_DIRTY)) 5694f8ad655SJan Kara list_del_init(&inode->i_wb_list); 5704f8ad655SJan Kara spin_unlock(&wb->list_lock); 5711c0eeaf5SJoern Engel inode_sync_complete(inode); 5724f8ad655SJan Kara out: 5734f8ad655SJan Kara spin_unlock(&inode->i_lock); 5741da177e4SLinus Torvalds return ret; 5751da177e4SLinus Torvalds } 5761da177e4SLinus Torvalds 5771a12d8bdSWu Fengguang static long writeback_chunk_size(struct backing_dev_info *bdi, 5781a12d8bdSWu Fengguang struct wb_writeback_work *work) 579d46db3d5SWu Fengguang { 580d46db3d5SWu Fengguang long pages; 581d46db3d5SWu Fengguang 582d46db3d5SWu Fengguang /* 583d46db3d5SWu Fengguang * WB_SYNC_ALL mode does livelock avoidance by syncing dirty 584d46db3d5SWu Fengguang * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX 585d46db3d5SWu Fengguang * here avoids calling into writeback_inodes_wb() more than once. 586d46db3d5SWu Fengguang * 587d46db3d5SWu Fengguang * The intended call sequence for WB_SYNC_ALL writeback is: 588d46db3d5SWu Fengguang * 589d46db3d5SWu Fengguang * wb_writeback() 590d46db3d5SWu Fengguang * writeback_sb_inodes() <== called only once 591d46db3d5SWu Fengguang * write_cache_pages() <== called once for each inode 592d46db3d5SWu Fengguang * (quickly) tag currently dirty pages 593d46db3d5SWu Fengguang * (maybe slowly) sync all tagged pages 594d46db3d5SWu Fengguang */ 595d46db3d5SWu Fengguang if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages) 596d46db3d5SWu Fengguang pages = LONG_MAX; 5971a12d8bdSWu Fengguang else { 5981a12d8bdSWu Fengguang pages = min(bdi->avg_write_bandwidth / 2, 5991a12d8bdSWu Fengguang global_dirty_limit / DIRTY_SCOPE); 6001a12d8bdSWu Fengguang pages = min(pages, work->nr_pages); 6011a12d8bdSWu Fengguang pages = round_down(pages + MIN_WRITEBACK_PAGES, 6021a12d8bdSWu Fengguang MIN_WRITEBACK_PAGES); 6031a12d8bdSWu Fengguang } 604d46db3d5SWu Fengguang 605d46db3d5SWu Fengguang return pages; 606d46db3d5SWu Fengguang } 607d46db3d5SWu Fengguang 60803ba3782SJens Axboe /* 609f11c9c5cSEdward Shishkin * Write a portion of b_io inodes which belong to @sb. 610edadfb10SChristoph Hellwig * 611d46db3d5SWu Fengguang * Return the number of pages and/or inodes written. 612f11c9c5cSEdward Shishkin */ 613d46db3d5SWu Fengguang static long writeback_sb_inodes(struct super_block *sb, 614d46db3d5SWu Fengguang struct bdi_writeback *wb, 615d46db3d5SWu Fengguang struct wb_writeback_work *work) 61603ba3782SJens Axboe { 617d46db3d5SWu Fengguang struct writeback_control wbc = { 618d46db3d5SWu Fengguang .sync_mode = work->sync_mode, 619d46db3d5SWu Fengguang .tagged_writepages = work->tagged_writepages, 620d46db3d5SWu Fengguang .for_kupdate = work->for_kupdate, 621d46db3d5SWu Fengguang .for_background = work->for_background, 6227747bd4bSDave Chinner .for_sync = work->for_sync, 623d46db3d5SWu Fengguang .range_cyclic = work->range_cyclic, 624d46db3d5SWu Fengguang .range_start = 0, 625d46db3d5SWu Fengguang .range_end = LLONG_MAX, 626d46db3d5SWu Fengguang }; 627d46db3d5SWu Fengguang unsigned long start_time = jiffies; 628d46db3d5SWu Fengguang long write_chunk; 629d46db3d5SWu Fengguang long wrote = 0; /* count both pages and inodes */ 630d46db3d5SWu Fengguang 63103ba3782SJens Axboe while (!list_empty(&wb->b_io)) { 6327ccf19a8SNick Piggin struct inode *inode = wb_inode(wb->b_io.prev); 633edadfb10SChristoph Hellwig 634edadfb10SChristoph Hellwig if (inode->i_sb != sb) { 635d46db3d5SWu Fengguang if (work->sb) { 636edadfb10SChristoph Hellwig /* 637edadfb10SChristoph Hellwig * We only want to write back data for this 638edadfb10SChristoph Hellwig * superblock, move all inodes not belonging 639edadfb10SChristoph Hellwig * to it back onto the dirty list. 640edadfb10SChristoph Hellwig */ 641f758eeabSChristoph Hellwig redirty_tail(inode, wb); 64266f3b8e2SJens Axboe continue; 64366f3b8e2SJens Axboe } 644edadfb10SChristoph Hellwig 645edadfb10SChristoph Hellwig /* 646edadfb10SChristoph Hellwig * The inode belongs to a different superblock. 647edadfb10SChristoph Hellwig * Bounce back to the caller to unpin this and 648edadfb10SChristoph Hellwig * pin the next superblock. 649edadfb10SChristoph Hellwig */ 650d46db3d5SWu Fengguang break; 651edadfb10SChristoph Hellwig } 652edadfb10SChristoph Hellwig 6539843b76aSChristoph Hellwig /* 654331cbdeeSWanpeng Li * Don't bother with new inodes or inodes being freed, first 655331cbdeeSWanpeng Li * kind does not need periodic writeout yet, and for the latter 6569843b76aSChristoph Hellwig * kind writeout is handled by the freer. 6579843b76aSChristoph Hellwig */ 658250df6edSDave Chinner spin_lock(&inode->i_lock); 6599843b76aSChristoph Hellwig if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { 660250df6edSDave Chinner spin_unlock(&inode->i_lock); 661fcc5c222SWu Fengguang redirty_tail(inode, wb); 6627ef0d737SNick Piggin continue; 6637ef0d737SNick Piggin } 664cc1676d9SJan Kara if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) { 665cc1676d9SJan Kara /* 666cc1676d9SJan Kara * If this inode is locked for writeback and we are not 667cc1676d9SJan Kara * doing writeback-for-data-integrity, move it to 668cc1676d9SJan Kara * b_more_io so that writeback can proceed with the 669cc1676d9SJan Kara * other inodes on s_io. 670cc1676d9SJan Kara * 671cc1676d9SJan Kara * We'll have another go at writing back this inode 672cc1676d9SJan Kara * when we completed a full scan of b_io. 673cc1676d9SJan Kara */ 674cc1676d9SJan Kara spin_unlock(&inode->i_lock); 675cc1676d9SJan Kara requeue_io(inode, wb); 676cc1676d9SJan Kara trace_writeback_sb_inodes_requeue(inode); 677cc1676d9SJan Kara continue; 678cc1676d9SJan Kara } 679f0d07b7fSJan Kara spin_unlock(&wb->list_lock); 680f0d07b7fSJan Kara 6814f8ad655SJan Kara /* 6824f8ad655SJan Kara * We already requeued the inode if it had I_SYNC set and we 6834f8ad655SJan Kara * are doing WB_SYNC_NONE writeback. So this catches only the 6844f8ad655SJan Kara * WB_SYNC_ALL case. 6854f8ad655SJan Kara */ 686169ebd90SJan Kara if (inode->i_state & I_SYNC) { 687169ebd90SJan Kara /* Wait for I_SYNC. This function drops i_lock... */ 688169ebd90SJan Kara inode_sleep_on_writeback(inode); 689169ebd90SJan Kara /* Inode may be gone, start again */ 690ead188f9SJan Kara spin_lock(&wb->list_lock); 691169ebd90SJan Kara continue; 692169ebd90SJan Kara } 6934f8ad655SJan Kara inode->i_state |= I_SYNC; 6944f8ad655SJan Kara spin_unlock(&inode->i_lock); 695169ebd90SJan Kara 6961a12d8bdSWu Fengguang write_chunk = writeback_chunk_size(wb->bdi, work); 697d46db3d5SWu Fengguang wbc.nr_to_write = write_chunk; 698d46db3d5SWu Fengguang wbc.pages_skipped = 0; 699250df6edSDave Chinner 700169ebd90SJan Kara /* 701169ebd90SJan Kara * We use I_SYNC to pin the inode in memory. While it is set 702169ebd90SJan Kara * evict_inode() will wait so the inode cannot be freed. 703169ebd90SJan Kara */ 704cd8ed2a4SYan Hong __writeback_single_inode(inode, &wbc); 705d46db3d5SWu Fengguang 706d46db3d5SWu Fengguang work->nr_pages -= write_chunk - wbc.nr_to_write; 707d46db3d5SWu Fengguang wrote += write_chunk - wbc.nr_to_write; 7084f8ad655SJan Kara spin_lock(&wb->list_lock); 7094f8ad655SJan Kara spin_lock(&inode->i_lock); 710d46db3d5SWu Fengguang if (!(inode->i_state & I_DIRTY)) 711d46db3d5SWu Fengguang wrote++; 7124f8ad655SJan Kara requeue_inode(inode, wb, &wbc); 7134f8ad655SJan Kara inode_sync_complete(inode); 7140f1b1fd8SDave Chinner spin_unlock(&inode->i_lock); 715169ebd90SJan Kara cond_resched_lock(&wb->list_lock); 716d46db3d5SWu Fengguang /* 717d46db3d5SWu Fengguang * bail out to wb_writeback() often enough to check 718d46db3d5SWu Fengguang * background threshold and other termination conditions. 719d46db3d5SWu Fengguang */ 720d46db3d5SWu Fengguang if (wrote) { 721d46db3d5SWu Fengguang if (time_is_before_jiffies(start_time + HZ / 10UL)) 722d46db3d5SWu Fengguang break; 723d46db3d5SWu Fengguang if (work->nr_pages <= 0) 724d46db3d5SWu Fengguang break; 7251da177e4SLinus Torvalds } 7268bc3be27SFengguang Wu } 727d46db3d5SWu Fengguang return wrote; 728f11c9c5cSEdward Shishkin } 72938f21977SNick Piggin 730d46db3d5SWu Fengguang static long __writeback_inodes_wb(struct bdi_writeback *wb, 731d46db3d5SWu Fengguang struct wb_writeback_work *work) 732f11c9c5cSEdward Shishkin { 733d46db3d5SWu Fengguang unsigned long start_time = jiffies; 734d46db3d5SWu Fengguang long wrote = 0; 735f11c9c5cSEdward Shishkin 736f11c9c5cSEdward Shishkin while (!list_empty(&wb->b_io)) { 7377ccf19a8SNick Piggin struct inode *inode = wb_inode(wb->b_io.prev); 738f11c9c5cSEdward Shishkin struct super_block *sb = inode->i_sb; 739f11c9c5cSEdward Shishkin 74012ad3ab6SDave Chinner if (!grab_super_passive(sb)) { 7410e995816SWu Fengguang /* 7420e995816SWu Fengguang * grab_super_passive() may fail consistently due to 7430e995816SWu Fengguang * s_umount being grabbed by someone else. Don't use 7440e995816SWu Fengguang * requeue_io() to avoid busy retrying the inode/sb. 7450e995816SWu Fengguang */ 7460e995816SWu Fengguang redirty_tail(inode, wb); 747d19de7edSChristoph Hellwig continue; 748334132aeSChristoph Hellwig } 749d46db3d5SWu Fengguang wrote += writeback_sb_inodes(sb, wb, work); 750d19de7edSChristoph Hellwig drop_super(sb); 751f11c9c5cSEdward Shishkin 752d46db3d5SWu Fengguang /* refer to the same tests at the end of writeback_sb_inodes */ 753d46db3d5SWu Fengguang if (wrote) { 754d46db3d5SWu Fengguang if (time_is_before_jiffies(start_time + HZ / 10UL)) 755d46db3d5SWu Fengguang break; 756d46db3d5SWu Fengguang if (work->nr_pages <= 0) 757f11c9c5cSEdward Shishkin break; 758f11c9c5cSEdward Shishkin } 759d46db3d5SWu Fengguang } 76066f3b8e2SJens Axboe /* Leave any unwritten inodes on b_io */ 761d46db3d5SWu Fengguang return wrote; 76266f3b8e2SJens Axboe } 76366f3b8e2SJens Axboe 7647d9f073bSWanpeng Li static long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages, 7650e175a18SCurt Wohlgemuth enum wb_reason reason) 766edadfb10SChristoph Hellwig { 767d46db3d5SWu Fengguang struct wb_writeback_work work = { 768d46db3d5SWu Fengguang .nr_pages = nr_pages, 769d46db3d5SWu Fengguang .sync_mode = WB_SYNC_NONE, 770d46db3d5SWu Fengguang .range_cyclic = 1, 7710e175a18SCurt Wohlgemuth .reason = reason, 772d46db3d5SWu Fengguang }; 773edadfb10SChristoph Hellwig 774f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 775424b351fSWu Fengguang if (list_empty(&wb->b_io)) 776ad4e38ddSCurt Wohlgemuth queue_io(wb, &work); 777d46db3d5SWu Fengguang __writeback_inodes_wb(wb, &work); 778f758eeabSChristoph Hellwig spin_unlock(&wb->list_lock); 779edadfb10SChristoph Hellwig 780d46db3d5SWu Fengguang return nr_pages - work.nr_pages; 78166f3b8e2SJens Axboe } 78266f3b8e2SJens Axboe 783b00949aaSWu Fengguang static bool over_bground_thresh(struct backing_dev_info *bdi) 78403ba3782SJens Axboe { 78503ba3782SJens Axboe unsigned long background_thresh, dirty_thresh; 78603ba3782SJens Axboe 78716c4042fSWu Fengguang global_dirty_limits(&background_thresh, &dirty_thresh); 78803ba3782SJens Axboe 789b00949aaSWu Fengguang if (global_page_state(NR_FILE_DIRTY) + 790b00949aaSWu Fengguang global_page_state(NR_UNSTABLE_NFS) > background_thresh) 791b00949aaSWu Fengguang return true; 792b00949aaSWu Fengguang 793b00949aaSWu Fengguang if (bdi_stat(bdi, BDI_RECLAIMABLE) > 794b00949aaSWu Fengguang bdi_dirty_limit(bdi, background_thresh)) 795b00949aaSWu Fengguang return true; 796b00949aaSWu Fengguang 797b00949aaSWu Fengguang return false; 79803ba3782SJens Axboe } 79903ba3782SJens Axboe 80003ba3782SJens Axboe /* 801e98be2d5SWu Fengguang * Called under wb->list_lock. If there are multiple wb per bdi, 802e98be2d5SWu Fengguang * only the flusher working on the first wb should do it. 803e98be2d5SWu Fengguang */ 804e98be2d5SWu Fengguang static void wb_update_bandwidth(struct bdi_writeback *wb, 805e98be2d5SWu Fengguang unsigned long start_time) 806e98be2d5SWu Fengguang { 807af6a3113SWu Fengguang __bdi_update_bandwidth(wb->bdi, 0, 0, 0, 0, 0, start_time); 808e98be2d5SWu Fengguang } 809e98be2d5SWu Fengguang 810e98be2d5SWu Fengguang /* 81103ba3782SJens Axboe * Explicit flushing or periodic writeback of "old" data. 81203ba3782SJens Axboe * 81303ba3782SJens Axboe * Define "old": the first time one of an inode's pages is dirtied, we mark the 81403ba3782SJens Axboe * dirtying-time in the inode's address_space. So this periodic writeback code 81503ba3782SJens Axboe * just walks the superblock inode list, writing back any inodes which are 81603ba3782SJens Axboe * older than a specific point in time. 81703ba3782SJens Axboe * 81803ba3782SJens Axboe * Try to run once per dirty_writeback_interval. But if a writeback event 81903ba3782SJens Axboe * takes longer than a dirty_writeback_interval interval, then leave a 82003ba3782SJens Axboe * one-second gap. 82103ba3782SJens Axboe * 82203ba3782SJens Axboe * older_than_this takes precedence over nr_to_write. So we'll only write back 82303ba3782SJens Axboe * all dirty pages if they are all attached to "old" mappings. 82403ba3782SJens Axboe */ 825c4a77a6cSJens Axboe static long wb_writeback(struct bdi_writeback *wb, 82683ba7b07SChristoph Hellwig struct wb_writeback_work *work) 82703ba3782SJens Axboe { 828e98be2d5SWu Fengguang unsigned long wb_start = jiffies; 829d46db3d5SWu Fengguang long nr_pages = work->nr_pages; 8300dc83bd3SJan Kara unsigned long oldest_jif; 831a5989bdcSJan Kara struct inode *inode; 832d46db3d5SWu Fengguang long progress; 83303ba3782SJens Axboe 8340dc83bd3SJan Kara oldest_jif = jiffies; 8350dc83bd3SJan Kara work->older_than_this = &oldest_jif; 83603ba3782SJens Axboe 837e8dfc305SWu Fengguang spin_lock(&wb->list_lock); 83803ba3782SJens Axboe for (;;) { 83903ba3782SJens Axboe /* 840d3ddec76SWu Fengguang * Stop writeback when nr_pages has been consumed 84103ba3782SJens Axboe */ 84283ba7b07SChristoph Hellwig if (work->nr_pages <= 0) 84303ba3782SJens Axboe break; 84403ba3782SJens Axboe 84503ba3782SJens Axboe /* 846aa373cf5SJan Kara * Background writeout and kupdate-style writeback may 847aa373cf5SJan Kara * run forever. Stop them if there is other work to do 848aa373cf5SJan Kara * so that e.g. sync can proceed. They'll be restarted 849aa373cf5SJan Kara * after the other works are all done. 850aa373cf5SJan Kara */ 851aa373cf5SJan Kara if ((work->for_background || work->for_kupdate) && 852aa373cf5SJan Kara !list_empty(&wb->bdi->work_list)) 853aa373cf5SJan Kara break; 854aa373cf5SJan Kara 855aa373cf5SJan Kara /* 856d3ddec76SWu Fengguang * For background writeout, stop when we are below the 857d3ddec76SWu Fengguang * background dirty threshold 85803ba3782SJens Axboe */ 859b00949aaSWu Fengguang if (work->for_background && !over_bground_thresh(wb->bdi)) 86003ba3782SJens Axboe break; 86103ba3782SJens Axboe 8621bc36b64SJan Kara /* 8631bc36b64SJan Kara * Kupdate and background works are special and we want to 8641bc36b64SJan Kara * include all inodes that need writing. Livelock avoidance is 8651bc36b64SJan Kara * handled by these works yielding to any other work so we are 8661bc36b64SJan Kara * safe. 8671bc36b64SJan Kara */ 868ba9aa839SWu Fengguang if (work->for_kupdate) { 8690dc83bd3SJan Kara oldest_jif = jiffies - 870ba9aa839SWu Fengguang msecs_to_jiffies(dirty_expire_interval * 10); 8711bc36b64SJan Kara } else if (work->for_background) 8720dc83bd3SJan Kara oldest_jif = jiffies; 873028c2dd1SDave Chinner 874d46db3d5SWu Fengguang trace_writeback_start(wb->bdi, work); 875e8dfc305SWu Fengguang if (list_empty(&wb->b_io)) 876ad4e38ddSCurt Wohlgemuth queue_io(wb, work); 87783ba7b07SChristoph Hellwig if (work->sb) 878d46db3d5SWu Fengguang progress = writeback_sb_inodes(work->sb, wb, work); 879edadfb10SChristoph Hellwig else 880d46db3d5SWu Fengguang progress = __writeback_inodes_wb(wb, work); 881d46db3d5SWu Fengguang trace_writeback_written(wb->bdi, work); 882028c2dd1SDave Chinner 883e98be2d5SWu Fengguang wb_update_bandwidth(wb, wb_start); 88403ba3782SJens Axboe 88503ba3782SJens Axboe /* 88671fd05a8SJens Axboe * Did we write something? Try for more 887e6fb6da2SWu Fengguang * 888e6fb6da2SWu Fengguang * Dirty inodes are moved to b_io for writeback in batches. 889e6fb6da2SWu Fengguang * The completion of the current batch does not necessarily 890e6fb6da2SWu Fengguang * mean the overall work is done. So we keep looping as long 891e6fb6da2SWu Fengguang * as made some progress on cleaning pages or inodes. 89271fd05a8SJens Axboe */ 893d46db3d5SWu Fengguang if (progress) 89403ba3782SJens Axboe continue; 895a5989bdcSJan Kara /* 896e6fb6da2SWu Fengguang * No more inodes for IO, bail 897a5989bdcSJan Kara */ 898b7a2441fSWu Fengguang if (list_empty(&wb->b_more_io)) 89903ba3782SJens Axboe break; 90003ba3782SJens Axboe /* 9018010c3b6SJens Axboe * Nothing written. Wait for some inode to 9028010c3b6SJens Axboe * become available for writeback. Otherwise 9038010c3b6SJens Axboe * we'll just busyloop. 90403ba3782SJens Axboe */ 90503ba3782SJens Axboe if (!list_empty(&wb->b_more_io)) { 906d46db3d5SWu Fengguang trace_writeback_wait(wb->bdi, work); 90703ba3782SJens Axboe inode = wb_inode(wb->b_more_io.prev); 908250df6edSDave Chinner spin_lock(&inode->i_lock); 909f0d07b7fSJan Kara spin_unlock(&wb->list_lock); 910169ebd90SJan Kara /* This function drops i_lock... */ 911169ebd90SJan Kara inode_sleep_on_writeback(inode); 912f0d07b7fSJan Kara spin_lock(&wb->list_lock); 91303ba3782SJens Axboe } 91403ba3782SJens Axboe } 915e8dfc305SWu Fengguang spin_unlock(&wb->list_lock); 91603ba3782SJens Axboe 917d46db3d5SWu Fengguang return nr_pages - work->nr_pages; 91803ba3782SJens Axboe } 91903ba3782SJens Axboe 92003ba3782SJens Axboe /* 92183ba7b07SChristoph Hellwig * Return the next wb_writeback_work struct that hasn't been processed yet. 92203ba3782SJens Axboe */ 92383ba7b07SChristoph Hellwig static struct wb_writeback_work * 92408852b6dSMinchan Kim get_next_work_item(struct backing_dev_info *bdi) 92503ba3782SJens Axboe { 92683ba7b07SChristoph Hellwig struct wb_writeback_work *work = NULL; 92703ba3782SJens Axboe 9286467716aSArtem Bityutskiy spin_lock_bh(&bdi->wb_lock); 92983ba7b07SChristoph Hellwig if (!list_empty(&bdi->work_list)) { 93083ba7b07SChristoph Hellwig work = list_entry(bdi->work_list.next, 93183ba7b07SChristoph Hellwig struct wb_writeback_work, list); 93283ba7b07SChristoph Hellwig list_del_init(&work->list); 93303ba3782SJens Axboe } 9346467716aSArtem Bityutskiy spin_unlock_bh(&bdi->wb_lock); 93583ba7b07SChristoph Hellwig return work; 93603ba3782SJens Axboe } 93703ba3782SJens Axboe 938cdf01dd5SLinus Torvalds /* 939cdf01dd5SLinus Torvalds * Add in the number of potentially dirty inodes, because each inode 940cdf01dd5SLinus Torvalds * write can dirty pagecache in the underlying blockdev. 941cdf01dd5SLinus Torvalds */ 942cdf01dd5SLinus Torvalds static unsigned long get_nr_dirty_pages(void) 943cdf01dd5SLinus Torvalds { 944cdf01dd5SLinus Torvalds return global_page_state(NR_FILE_DIRTY) + 945cdf01dd5SLinus Torvalds global_page_state(NR_UNSTABLE_NFS) + 946cdf01dd5SLinus Torvalds get_nr_dirty_inodes(); 947cdf01dd5SLinus Torvalds } 948cdf01dd5SLinus Torvalds 9496585027aSJan Kara static long wb_check_background_flush(struct bdi_writeback *wb) 9506585027aSJan Kara { 951b00949aaSWu Fengguang if (over_bground_thresh(wb->bdi)) { 9526585027aSJan Kara 9536585027aSJan Kara struct wb_writeback_work work = { 9546585027aSJan Kara .nr_pages = LONG_MAX, 9556585027aSJan Kara .sync_mode = WB_SYNC_NONE, 9566585027aSJan Kara .for_background = 1, 9576585027aSJan Kara .range_cyclic = 1, 9580e175a18SCurt Wohlgemuth .reason = WB_REASON_BACKGROUND, 9596585027aSJan Kara }; 9606585027aSJan Kara 9616585027aSJan Kara return wb_writeback(wb, &work); 9626585027aSJan Kara } 9636585027aSJan Kara 9646585027aSJan Kara return 0; 9656585027aSJan Kara } 9666585027aSJan Kara 96703ba3782SJens Axboe static long wb_check_old_data_flush(struct bdi_writeback *wb) 96803ba3782SJens Axboe { 96903ba3782SJens Axboe unsigned long expired; 97003ba3782SJens Axboe long nr_pages; 97103ba3782SJens Axboe 97269b62d01SJens Axboe /* 97369b62d01SJens Axboe * When set to zero, disable periodic writeback 97469b62d01SJens Axboe */ 97569b62d01SJens Axboe if (!dirty_writeback_interval) 97669b62d01SJens Axboe return 0; 97769b62d01SJens Axboe 97803ba3782SJens Axboe expired = wb->last_old_flush + 97903ba3782SJens Axboe msecs_to_jiffies(dirty_writeback_interval * 10); 98003ba3782SJens Axboe if (time_before(jiffies, expired)) 98103ba3782SJens Axboe return 0; 98203ba3782SJens Axboe 98303ba3782SJens Axboe wb->last_old_flush = jiffies; 984cdf01dd5SLinus Torvalds nr_pages = get_nr_dirty_pages(); 98503ba3782SJens Axboe 986c4a77a6cSJens Axboe if (nr_pages) { 98783ba7b07SChristoph Hellwig struct wb_writeback_work work = { 988c4a77a6cSJens Axboe .nr_pages = nr_pages, 989c4a77a6cSJens Axboe .sync_mode = WB_SYNC_NONE, 990c4a77a6cSJens Axboe .for_kupdate = 1, 991c4a77a6cSJens Axboe .range_cyclic = 1, 9920e175a18SCurt Wohlgemuth .reason = WB_REASON_PERIODIC, 993c4a77a6cSJens Axboe }; 994c4a77a6cSJens Axboe 99583ba7b07SChristoph Hellwig return wb_writeback(wb, &work); 996c4a77a6cSJens Axboe } 99703ba3782SJens Axboe 99803ba3782SJens Axboe return 0; 99903ba3782SJens Axboe } 100003ba3782SJens Axboe 100103ba3782SJens Axboe /* 100203ba3782SJens Axboe * Retrieve work items and do the writeback they describe 100303ba3782SJens Axboe */ 100425d130baSWanpeng Li static long wb_do_writeback(struct bdi_writeback *wb) 100503ba3782SJens Axboe { 100603ba3782SJens Axboe struct backing_dev_info *bdi = wb->bdi; 100783ba7b07SChristoph Hellwig struct wb_writeback_work *work; 1008c4a77a6cSJens Axboe long wrote = 0; 100903ba3782SJens Axboe 101081d73a32SJan Kara set_bit(BDI_writeback_running, &wb->bdi->state); 101108852b6dSMinchan Kim while ((work = get_next_work_item(bdi)) != NULL) { 101283ba7b07SChristoph Hellwig 1013455b2864SDave Chinner trace_writeback_exec(bdi, work); 1014455b2864SDave Chinner 101583ba7b07SChristoph Hellwig wrote += wb_writeback(wb, work); 101603ba3782SJens Axboe 101703ba3782SJens Axboe /* 101883ba7b07SChristoph Hellwig * Notify the caller of completion if this is a synchronous 101983ba7b07SChristoph Hellwig * work item, otherwise just free it. 102003ba3782SJens Axboe */ 102183ba7b07SChristoph Hellwig if (work->done) 102283ba7b07SChristoph Hellwig complete(work->done); 102383ba7b07SChristoph Hellwig else 102483ba7b07SChristoph Hellwig kfree(work); 102503ba3782SJens Axboe } 102603ba3782SJens Axboe 102703ba3782SJens Axboe /* 102803ba3782SJens Axboe * Check for periodic writeback, kupdated() style 102903ba3782SJens Axboe */ 103003ba3782SJens Axboe wrote += wb_check_old_data_flush(wb); 10316585027aSJan Kara wrote += wb_check_background_flush(wb); 103281d73a32SJan Kara clear_bit(BDI_writeback_running, &wb->bdi->state); 103303ba3782SJens Axboe 103403ba3782SJens Axboe return wrote; 103503ba3782SJens Axboe } 103603ba3782SJens Axboe 103703ba3782SJens Axboe /* 103803ba3782SJens Axboe * Handle writeback of dirty data for the device backed by this bdi. Also 1039839a8e86STejun Heo * reschedules periodically and does kupdated style flushing. 104003ba3782SJens Axboe */ 1041839a8e86STejun Heo void bdi_writeback_workfn(struct work_struct *work) 104203ba3782SJens Axboe { 1043839a8e86STejun Heo struct bdi_writeback *wb = container_of(to_delayed_work(work), 1044839a8e86STejun Heo struct bdi_writeback, dwork); 104508243900SChristoph Hellwig struct backing_dev_info *bdi = wb->bdi; 104603ba3782SJens Axboe long pages_written; 104703ba3782SJens Axboe 1048ef3b1019STejun Heo set_worker_desc("flush-%s", dev_name(bdi->dev)); 1049766f9164SPeter Zijlstra current->flags |= PF_SWAPWRITE; 105003ba3782SJens Axboe 1051839a8e86STejun Heo if (likely(!current_is_workqueue_rescuer() || 10525acda9d1SJan Kara !test_bit(BDI_registered, &bdi->state))) { 105303ba3782SJens Axboe /* 1054839a8e86STejun Heo * The normal path. Keep writing back @bdi until its 1055839a8e86STejun Heo * work_list is empty. Note that this path is also taken 1056839a8e86STejun Heo * if @bdi is shutting down even when we're running off the 1057839a8e86STejun Heo * rescuer as work_list needs to be drained. 105803ba3782SJens Axboe */ 1059839a8e86STejun Heo do { 106025d130baSWanpeng Li pages_written = wb_do_writeback(wb); 1061455b2864SDave Chinner trace_writeback_pages_written(pages_written); 1062839a8e86STejun Heo } while (!list_empty(&bdi->work_list)); 1063839a8e86STejun Heo } else { 1064253c34e9SArtem Bityutskiy /* 1065839a8e86STejun Heo * bdi_wq can't get enough workers and we're running off 1066839a8e86STejun Heo * the emergency worker. Don't hog it. Hopefully, 1024 is 1067839a8e86STejun Heo * enough for efficient IO. 1068253c34e9SArtem Bityutskiy */ 1069839a8e86STejun Heo pages_written = writeback_inodes_wb(&bdi->wb, 1024, 1070839a8e86STejun Heo WB_REASON_FORKER_THREAD); 1071839a8e86STejun Heo trace_writeback_pages_written(pages_written); 107203ba3782SJens Axboe } 107303ba3782SJens Axboe 10746ca738d6SDerek Basehore if (!list_empty(&bdi->work_list)) 10756ca738d6SDerek Basehore mod_delayed_work(bdi_wq, &wb->dwork, 0); 10766ca738d6SDerek Basehore else if (wb_has_dirty_io(wb) && dirty_writeback_interval) 10776ca738d6SDerek Basehore bdi_wakeup_thread_delayed(bdi); 1078455b2864SDave Chinner 1079839a8e86STejun Heo current->flags &= ~PF_SWAPWRITE; 108003ba3782SJens Axboe } 108103ba3782SJens Axboe 108203ba3782SJens Axboe /* 108303ba3782SJens Axboe * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back 108403ba3782SJens Axboe * the whole world. 108503ba3782SJens Axboe */ 10860e175a18SCurt Wohlgemuth void wakeup_flusher_threads(long nr_pages, enum wb_reason reason) 108703ba3782SJens Axboe { 1088b8c2f347SChristoph Hellwig struct backing_dev_info *bdi; 1089b8c2f347SChristoph Hellwig 109047df3ddeSJan Kara if (!nr_pages) 109147df3ddeSJan Kara nr_pages = get_nr_dirty_pages(); 1092b8c2f347SChristoph Hellwig 1093b8c2f347SChristoph Hellwig rcu_read_lock(); 1094b8c2f347SChristoph Hellwig list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { 1095b8c2f347SChristoph Hellwig if (!bdi_has_dirty_io(bdi)) 1096b8c2f347SChristoph Hellwig continue; 10970e175a18SCurt Wohlgemuth __bdi_start_writeback(bdi, nr_pages, false, reason); 1098b8c2f347SChristoph Hellwig } 1099b8c2f347SChristoph Hellwig rcu_read_unlock(); 110003ba3782SJens Axboe } 110103ba3782SJens Axboe 110203ba3782SJens Axboe static noinline void block_dump___mark_inode_dirty(struct inode *inode) 110303ba3782SJens Axboe { 110403ba3782SJens Axboe if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { 110503ba3782SJens Axboe struct dentry *dentry; 110603ba3782SJens Axboe const char *name = "?"; 110703ba3782SJens Axboe 110803ba3782SJens Axboe dentry = d_find_alias(inode); 110903ba3782SJens Axboe if (dentry) { 111003ba3782SJens Axboe spin_lock(&dentry->d_lock); 111103ba3782SJens Axboe name = (const char *) dentry->d_name.name; 111203ba3782SJens Axboe } 111303ba3782SJens Axboe printk(KERN_DEBUG 111403ba3782SJens Axboe "%s(%d): dirtied inode %lu (%s) on %s\n", 111503ba3782SJens Axboe current->comm, task_pid_nr(current), inode->i_ino, 111603ba3782SJens Axboe name, inode->i_sb->s_id); 111703ba3782SJens Axboe if (dentry) { 111803ba3782SJens Axboe spin_unlock(&dentry->d_lock); 111903ba3782SJens Axboe dput(dentry); 112003ba3782SJens Axboe } 112103ba3782SJens Axboe } 112203ba3782SJens Axboe } 112303ba3782SJens Axboe 112403ba3782SJens Axboe /** 112503ba3782SJens Axboe * __mark_inode_dirty - internal function 112603ba3782SJens Axboe * @inode: inode to mark 112703ba3782SJens Axboe * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) 112803ba3782SJens Axboe * Mark an inode as dirty. Callers should use mark_inode_dirty or 112903ba3782SJens Axboe * mark_inode_dirty_sync. 113003ba3782SJens Axboe * 113103ba3782SJens Axboe * Put the inode on the super block's dirty list. 113203ba3782SJens Axboe * 113303ba3782SJens Axboe * CAREFUL! We mark it dirty unconditionally, but move it onto the 113403ba3782SJens Axboe * dirty list only if it is hashed or if it refers to a blockdev. 113503ba3782SJens Axboe * If it was not hashed, it will never be added to the dirty list 113603ba3782SJens Axboe * even if it is later hashed, as it will have been marked dirty already. 113703ba3782SJens Axboe * 113803ba3782SJens Axboe * In short, make sure you hash any inodes _before_ you start marking 113903ba3782SJens Axboe * them dirty. 114003ba3782SJens Axboe * 114103ba3782SJens Axboe * Note that for blockdevs, inode->dirtied_when represents the dirtying time of 114203ba3782SJens Axboe * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of 114303ba3782SJens Axboe * the kernel-internal blockdev inode represents the dirtying time of the 114403ba3782SJens Axboe * blockdev's pages. This is why for I_DIRTY_PAGES we always use 114503ba3782SJens Axboe * page->mapping->host, so the page-dirtying time is recorded in the internal 114603ba3782SJens Axboe * blockdev inode. 114703ba3782SJens Axboe */ 114803ba3782SJens Axboe void __mark_inode_dirty(struct inode *inode, int flags) 114903ba3782SJens Axboe { 115003ba3782SJens Axboe struct super_block *sb = inode->i_sb; 1151253c34e9SArtem Bityutskiy struct backing_dev_info *bdi = NULL; 115203ba3782SJens Axboe 115303ba3782SJens Axboe /* 115403ba3782SJens Axboe * Don't do this for I_DIRTY_PAGES - that doesn't actually 115503ba3782SJens Axboe * dirty the inode itself 115603ba3782SJens Axboe */ 115703ba3782SJens Axboe if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { 11589fb0a7daSTejun Heo trace_writeback_dirty_inode_start(inode, flags); 11599fb0a7daSTejun Heo 116003ba3782SJens Axboe if (sb->s_op->dirty_inode) 1161aa385729SChristoph Hellwig sb->s_op->dirty_inode(inode, flags); 11629fb0a7daSTejun Heo 11639fb0a7daSTejun Heo trace_writeback_dirty_inode(inode, flags); 116403ba3782SJens Axboe } 116503ba3782SJens Axboe 116603ba3782SJens Axboe /* 11679c6ac78eSTejun Heo * Paired with smp_mb() in __writeback_single_inode() for the 11689c6ac78eSTejun Heo * following lockless i_state test. See there for details. 116903ba3782SJens Axboe */ 117003ba3782SJens Axboe smp_mb(); 117103ba3782SJens Axboe 117203ba3782SJens Axboe if ((inode->i_state & flags) == flags) 117303ba3782SJens Axboe return; 117403ba3782SJens Axboe 117503ba3782SJens Axboe if (unlikely(block_dump)) 117603ba3782SJens Axboe block_dump___mark_inode_dirty(inode); 117703ba3782SJens Axboe 1178250df6edSDave Chinner spin_lock(&inode->i_lock); 117903ba3782SJens Axboe if ((inode->i_state & flags) != flags) { 118003ba3782SJens Axboe const int was_dirty = inode->i_state & I_DIRTY; 118103ba3782SJens Axboe 118203ba3782SJens Axboe inode->i_state |= flags; 118303ba3782SJens Axboe 118403ba3782SJens Axboe /* 118503ba3782SJens Axboe * If the inode is being synced, just update its dirty state. 118603ba3782SJens Axboe * The unlocker will place the inode on the appropriate 118703ba3782SJens Axboe * superblock list, based upon its state. 118803ba3782SJens Axboe */ 118903ba3782SJens Axboe if (inode->i_state & I_SYNC) 1190250df6edSDave Chinner goto out_unlock_inode; 119103ba3782SJens Axboe 119203ba3782SJens Axboe /* 119303ba3782SJens Axboe * Only add valid (hashed) inodes to the superblock's 119403ba3782SJens Axboe * dirty list. Add blockdev inodes as well. 119503ba3782SJens Axboe */ 119603ba3782SJens Axboe if (!S_ISBLK(inode->i_mode)) { 11971d3382cbSAl Viro if (inode_unhashed(inode)) 1198250df6edSDave Chinner goto out_unlock_inode; 119903ba3782SJens Axboe } 1200a4ffdde6SAl Viro if (inode->i_state & I_FREEING) 1201250df6edSDave Chinner goto out_unlock_inode; 120203ba3782SJens Axboe 120303ba3782SJens Axboe /* 120403ba3782SJens Axboe * If the inode was already on b_dirty/b_io/b_more_io, don't 120503ba3782SJens Axboe * reposition it (that would break b_dirty time-ordering). 120603ba3782SJens Axboe */ 120703ba3782SJens Axboe if (!was_dirty) { 1208a66979abSDave Chinner bool wakeup_bdi = false; 1209253c34e9SArtem Bityutskiy bdi = inode_to_bdi(inode); 1210500b067cSJens Axboe 1211146d7009SJunxiao Bi spin_unlock(&inode->i_lock); 1212146d7009SJunxiao Bi spin_lock(&bdi->wb.list_lock); 1213253c34e9SArtem Bityutskiy if (bdi_cap_writeback_dirty(bdi)) { 1214253c34e9SArtem Bityutskiy WARN(!test_bit(BDI_registered, &bdi->state), 1215253c34e9SArtem Bityutskiy "bdi-%s not registered\n", bdi->name); 1216253c34e9SArtem Bityutskiy 1217253c34e9SArtem Bityutskiy /* 1218253c34e9SArtem Bityutskiy * If this is the first dirty inode for this 1219253c34e9SArtem Bityutskiy * bdi, we have to wake-up the corresponding 1220253c34e9SArtem Bityutskiy * bdi thread to make sure background 1221253c34e9SArtem Bityutskiy * write-back happens later. 1222253c34e9SArtem Bityutskiy */ 1223253c34e9SArtem Bityutskiy if (!wb_has_dirty_io(&bdi->wb)) 1224253c34e9SArtem Bityutskiy wakeup_bdi = true; 1225500b067cSJens Axboe } 122603ba3782SJens Axboe 122703ba3782SJens Axboe inode->dirtied_when = jiffies; 12287ccf19a8SNick Piggin list_move(&inode->i_wb_list, &bdi->wb.b_dirty); 1229f758eeabSChristoph Hellwig spin_unlock(&bdi->wb.list_lock); 1230253c34e9SArtem Bityutskiy 1231253c34e9SArtem Bityutskiy if (wakeup_bdi) 12326467716aSArtem Bityutskiy bdi_wakeup_thread_delayed(bdi); 1233a66979abSDave Chinner return; 1234a66979abSDave Chinner } 1235a66979abSDave Chinner } 1236a66979abSDave Chinner out_unlock_inode: 1237a66979abSDave Chinner spin_unlock(&inode->i_lock); 1238a66979abSDave Chinner 123903ba3782SJens Axboe } 124003ba3782SJens Axboe EXPORT_SYMBOL(__mark_inode_dirty); 124103ba3782SJens Axboe 1242b6e51316SJens Axboe static void wait_sb_inodes(struct super_block *sb) 124366f3b8e2SJens Axboe { 124438f21977SNick Piggin struct inode *inode, *old_inode = NULL; 124538f21977SNick Piggin 124603ba3782SJens Axboe /* 124703ba3782SJens Axboe * We need to be protected against the filesystem going from 124803ba3782SJens Axboe * r/o to r/w or vice versa. 124903ba3782SJens Axboe */ 1250b6e51316SJens Axboe WARN_ON(!rwsem_is_locked(&sb->s_umount)); 125103ba3782SJens Axboe 125255fa6091SDave Chinner spin_lock(&inode_sb_list_lock); 125366f3b8e2SJens Axboe 125438f21977SNick Piggin /* 125538f21977SNick Piggin * Data integrity sync. Must wait for all pages under writeback, 125638f21977SNick Piggin * because there may have been pages dirtied before our sync 125738f21977SNick Piggin * call, but which had writeout started before we write it out. 125838f21977SNick Piggin * In which case, the inode may not be on the dirty list, but 125938f21977SNick Piggin * we still have to wait for that writeout. 126038f21977SNick Piggin */ 1261b6e51316SJens Axboe list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { 1262250df6edSDave Chinner struct address_space *mapping = inode->i_mapping; 126338f21977SNick Piggin 1264250df6edSDave Chinner spin_lock(&inode->i_lock); 1265250df6edSDave Chinner if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) || 1266250df6edSDave Chinner (mapping->nrpages == 0)) { 1267250df6edSDave Chinner spin_unlock(&inode->i_lock); 126838f21977SNick Piggin continue; 1269250df6edSDave Chinner } 127038f21977SNick Piggin __iget(inode); 1271250df6edSDave Chinner spin_unlock(&inode->i_lock); 127255fa6091SDave Chinner spin_unlock(&inode_sb_list_lock); 127355fa6091SDave Chinner 127438f21977SNick Piggin /* 127555fa6091SDave Chinner * We hold a reference to 'inode' so it couldn't have been 127655fa6091SDave Chinner * removed from s_inodes list while we dropped the 127755fa6091SDave Chinner * inode_sb_list_lock. We cannot iput the inode now as we can 127855fa6091SDave Chinner * be holding the last reference and we cannot iput it under 127955fa6091SDave Chinner * inode_sb_list_lock. So we keep the reference and iput it 128055fa6091SDave Chinner * later. 128138f21977SNick Piggin */ 128238f21977SNick Piggin iput(old_inode); 128338f21977SNick Piggin old_inode = inode; 128438f21977SNick Piggin 128538f21977SNick Piggin filemap_fdatawait(mapping); 128638f21977SNick Piggin 128738f21977SNick Piggin cond_resched(); 128838f21977SNick Piggin 128955fa6091SDave Chinner spin_lock(&inode_sb_list_lock); 129038f21977SNick Piggin } 129155fa6091SDave Chinner spin_unlock(&inode_sb_list_lock); 129238f21977SNick Piggin iput(old_inode); 129366f3b8e2SJens Axboe } 12941da177e4SLinus Torvalds 1295d8a8559cSJens Axboe /** 12963259f8beSChris Mason * writeback_inodes_sb_nr - writeback dirty inodes from given super_block 1297d8a8559cSJens Axboe * @sb: the superblock 12983259f8beSChris Mason * @nr: the number of pages to write 1299786228abSMarcos Paulo de Souza * @reason: reason why some writeback work initiated 13001da177e4SLinus Torvalds * 1301d8a8559cSJens Axboe * Start writeback on some inodes on this super_block. No guarantees are made 1302d8a8559cSJens Axboe * on how many (if any) will be written, and this function does not wait 13033259f8beSChris Mason * for IO completion of submitted IO. 13041da177e4SLinus Torvalds */ 13050e175a18SCurt Wohlgemuth void writeback_inodes_sb_nr(struct super_block *sb, 13060e175a18SCurt Wohlgemuth unsigned long nr, 13070e175a18SCurt Wohlgemuth enum wb_reason reason) 13081da177e4SLinus Torvalds { 130983ba7b07SChristoph Hellwig DECLARE_COMPLETION_ONSTACK(done); 131083ba7b07SChristoph Hellwig struct wb_writeback_work work = { 13113c4d7165SChristoph Hellwig .sb = sb, 13123c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_NONE, 13136e6938b6SWu Fengguang .tagged_writepages = 1, 131483ba7b07SChristoph Hellwig .done = &done, 13153259f8beSChris Mason .nr_pages = nr, 13160e175a18SCurt Wohlgemuth .reason = reason, 13173c4d7165SChristoph Hellwig }; 13180e3c9a22SJens Axboe 13196eedc701SJan Kara if (sb->s_bdi == &noop_backing_dev_info) 13206eedc701SJan Kara return; 1321cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 132283ba7b07SChristoph Hellwig bdi_queue_work(sb->s_bdi, &work); 132383ba7b07SChristoph Hellwig wait_for_completion(&done); 13241da177e4SLinus Torvalds } 13253259f8beSChris Mason EXPORT_SYMBOL(writeback_inodes_sb_nr); 13263259f8beSChris Mason 13273259f8beSChris Mason /** 13283259f8beSChris Mason * writeback_inodes_sb - writeback dirty inodes from given super_block 13293259f8beSChris Mason * @sb: the superblock 1330786228abSMarcos Paulo de Souza * @reason: reason why some writeback work was initiated 13313259f8beSChris Mason * 13323259f8beSChris Mason * Start writeback on some inodes on this super_block. No guarantees are made 13333259f8beSChris Mason * on how many (if any) will be written, and this function does not wait 13343259f8beSChris Mason * for IO completion of submitted IO. 13353259f8beSChris Mason */ 13360e175a18SCurt Wohlgemuth void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) 13373259f8beSChris Mason { 13380e175a18SCurt Wohlgemuth return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); 13393259f8beSChris Mason } 1340d8a8559cSJens Axboe EXPORT_SYMBOL(writeback_inodes_sb); 1341d8a8559cSJens Axboe 1342d8a8559cSJens Axboe /** 134310ee27a0SMiao Xie * try_to_writeback_inodes_sb_nr - try to start writeback if none underway 13443259f8beSChris Mason * @sb: the superblock 13453259f8beSChris Mason * @nr: the number of pages to write 134610ee27a0SMiao Xie * @reason: the reason of writeback 13473259f8beSChris Mason * 134810ee27a0SMiao Xie * Invoke writeback_inodes_sb_nr if no writeback is currently underway. 13493259f8beSChris Mason * Returns 1 if writeback was started, 0 if not. 13503259f8beSChris Mason */ 135110ee27a0SMiao Xie int try_to_writeback_inodes_sb_nr(struct super_block *sb, 13520e175a18SCurt Wohlgemuth unsigned long nr, 13530e175a18SCurt Wohlgemuth enum wb_reason reason) 13543259f8beSChris Mason { 135510ee27a0SMiao Xie if (writeback_in_progress(sb->s_bdi)) 135610ee27a0SMiao Xie return 1; 135710ee27a0SMiao Xie 135810ee27a0SMiao Xie if (!down_read_trylock(&sb->s_umount)) 135910ee27a0SMiao Xie return 0; 136010ee27a0SMiao Xie 13610e175a18SCurt Wohlgemuth writeback_inodes_sb_nr(sb, nr, reason); 13623259f8beSChris Mason up_read(&sb->s_umount); 13633259f8beSChris Mason return 1; 13643259f8beSChris Mason } 136510ee27a0SMiao Xie EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr); 136610ee27a0SMiao Xie 136710ee27a0SMiao Xie /** 136810ee27a0SMiao Xie * try_to_writeback_inodes_sb - try to start writeback if none underway 136910ee27a0SMiao Xie * @sb: the superblock 137010ee27a0SMiao Xie * @reason: reason why some writeback work was initiated 137110ee27a0SMiao Xie * 137210ee27a0SMiao Xie * Implement by try_to_writeback_inodes_sb_nr() 137310ee27a0SMiao Xie * Returns 1 if writeback was started, 0 if not. 137410ee27a0SMiao Xie */ 137510ee27a0SMiao Xie int try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) 137610ee27a0SMiao Xie { 137710ee27a0SMiao Xie return try_to_writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); 137810ee27a0SMiao Xie } 137910ee27a0SMiao Xie EXPORT_SYMBOL(try_to_writeback_inodes_sb); 13803259f8beSChris Mason 13813259f8beSChris Mason /** 1382d8a8559cSJens Axboe * sync_inodes_sb - sync sb inode pages 1383d8a8559cSJens Axboe * @sb: the superblock 1384d8a8559cSJens Axboe * 1385d8a8559cSJens Axboe * This function writes and waits on any dirty inode belonging to this 13860dc83bd3SJan Kara * super_block. 1387d8a8559cSJens Axboe */ 13880dc83bd3SJan Kara void sync_inodes_sb(struct super_block *sb) 1389d8a8559cSJens Axboe { 139083ba7b07SChristoph Hellwig DECLARE_COMPLETION_ONSTACK(done); 139183ba7b07SChristoph Hellwig struct wb_writeback_work work = { 13923c4d7165SChristoph Hellwig .sb = sb, 13933c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_ALL, 13943c4d7165SChristoph Hellwig .nr_pages = LONG_MAX, 13953c4d7165SChristoph Hellwig .range_cyclic = 0, 139683ba7b07SChristoph Hellwig .done = &done, 13970e175a18SCurt Wohlgemuth .reason = WB_REASON_SYNC, 13987747bd4bSDave Chinner .for_sync = 1, 13993c4d7165SChristoph Hellwig }; 14003c4d7165SChristoph Hellwig 14016eedc701SJan Kara /* Nothing to do? */ 14026eedc701SJan Kara if (sb->s_bdi == &noop_backing_dev_info) 14036eedc701SJan Kara return; 1404cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 1405cf37e972SChristoph Hellwig 140683ba7b07SChristoph Hellwig bdi_queue_work(sb->s_bdi, &work); 140783ba7b07SChristoph Hellwig wait_for_completion(&done); 140883ba7b07SChristoph Hellwig 1409b6e51316SJens Axboe wait_sb_inodes(sb); 1410d8a8559cSJens Axboe } 1411d8a8559cSJens Axboe EXPORT_SYMBOL(sync_inodes_sb); 14121da177e4SLinus Torvalds 14131da177e4SLinus Torvalds /** 14141da177e4SLinus Torvalds * write_inode_now - write an inode to disk 14151da177e4SLinus Torvalds * @inode: inode to write to disk 14161da177e4SLinus Torvalds * @sync: whether the write should be synchronous or not 14171da177e4SLinus Torvalds * 14187f04c26dSAndrea Arcangeli * This function commits an inode to disk immediately if it is dirty. This is 14197f04c26dSAndrea Arcangeli * primarily needed by knfsd. 14207f04c26dSAndrea Arcangeli * 14217f04c26dSAndrea Arcangeli * The caller must either have a ref on the inode or must have set I_WILL_FREE. 14221da177e4SLinus Torvalds */ 14231da177e4SLinus Torvalds int write_inode_now(struct inode *inode, int sync) 14241da177e4SLinus Torvalds { 1425f758eeabSChristoph Hellwig struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; 14261da177e4SLinus Torvalds struct writeback_control wbc = { 14271da177e4SLinus Torvalds .nr_to_write = LONG_MAX, 142818914b18SMike Galbraith .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, 1429111ebb6eSOGAWA Hirofumi .range_start = 0, 1430111ebb6eSOGAWA Hirofumi .range_end = LLONG_MAX, 14311da177e4SLinus Torvalds }; 14321da177e4SLinus Torvalds 14331da177e4SLinus Torvalds if (!mapping_cap_writeback_dirty(inode->i_mapping)) 143449364ce2SAndrew Morton wbc.nr_to_write = 0; 14351da177e4SLinus Torvalds 14361da177e4SLinus Torvalds might_sleep(); 14374f8ad655SJan Kara return writeback_single_inode(inode, wb, &wbc); 14381da177e4SLinus Torvalds } 14391da177e4SLinus Torvalds EXPORT_SYMBOL(write_inode_now); 14401da177e4SLinus Torvalds 14411da177e4SLinus Torvalds /** 14421da177e4SLinus Torvalds * sync_inode - write an inode and its pages to disk. 14431da177e4SLinus Torvalds * @inode: the inode to sync 14441da177e4SLinus Torvalds * @wbc: controls the writeback mode 14451da177e4SLinus Torvalds * 14461da177e4SLinus Torvalds * sync_inode() will write an inode and its pages to disk. It will also 14471da177e4SLinus Torvalds * correctly update the inode on its superblock's dirty inode lists and will 14481da177e4SLinus Torvalds * update inode->i_state. 14491da177e4SLinus Torvalds * 14501da177e4SLinus Torvalds * The caller must have a ref on the inode. 14511da177e4SLinus Torvalds */ 14521da177e4SLinus Torvalds int sync_inode(struct inode *inode, struct writeback_control *wbc) 14531da177e4SLinus Torvalds { 14544f8ad655SJan Kara return writeback_single_inode(inode, &inode_to_bdi(inode)->wb, wbc); 14551da177e4SLinus Torvalds } 14561da177e4SLinus Torvalds EXPORT_SYMBOL(sync_inode); 1457c3765016SChristoph Hellwig 1458c3765016SChristoph Hellwig /** 1459c691b9d9SAndrew Morton * sync_inode_metadata - write an inode to disk 1460c3765016SChristoph Hellwig * @inode: the inode to sync 1461c3765016SChristoph Hellwig * @wait: wait for I/O to complete. 1462c3765016SChristoph Hellwig * 1463c691b9d9SAndrew Morton * Write an inode to disk and adjust its dirty state after completion. 1464c3765016SChristoph Hellwig * 1465c3765016SChristoph Hellwig * Note: only writes the actual inode, no associated data or other metadata. 1466c3765016SChristoph Hellwig */ 1467c3765016SChristoph Hellwig int sync_inode_metadata(struct inode *inode, int wait) 1468c3765016SChristoph Hellwig { 1469c3765016SChristoph Hellwig struct writeback_control wbc = { 1470c3765016SChristoph Hellwig .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE, 1471c3765016SChristoph Hellwig .nr_to_write = 0, /* metadata-only */ 1472c3765016SChristoph Hellwig }; 1473c3765016SChristoph Hellwig 1474c3765016SChristoph Hellwig return sync_inode(inode, &wbc); 1475c3765016SChristoph Hellwig } 1476c3765016SChristoph Hellwig EXPORT_SYMBOL(sync_inode_metadata); 1477