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; 72692ebd17SJan Kara 73a8855990SJan Kara if (sb_is_blkdev_sb(sb)) 74aaead25bSChristoph Hellwig return inode->i_mapping->backing_dev_info; 75aaead25bSChristoph Hellwig 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); 4826290be1cSJan Kara /* Clear I_DIRTY_PAGES if we've written out all dirty pages */ 4836290be1cSJan Kara if (!mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) 4846290be1cSJan Kara inode->i_state &= ~I_DIRTY_PAGES; 4855547e8aaSDmitry Monakhov dirty = inode->i_state & I_DIRTY; 4865547e8aaSDmitry Monakhov inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC); 487250df6edSDave Chinner spin_unlock(&inode->i_lock); 48826821ed4SChristoph Hellwig /* Don't write the inode if only I_DIRTY_PAGES was set */ 48926821ed4SChristoph Hellwig if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { 490a9185b41SChristoph Hellwig int err = write_inode(inode, wbc); 4911da177e4SLinus Torvalds if (ret == 0) 4921da177e4SLinus Torvalds ret = err; 4931da177e4SLinus Torvalds } 4944f8ad655SJan Kara trace_writeback_single_inode(inode, wbc, nr_to_write); 4954f8ad655SJan Kara return ret; 4964f8ad655SJan Kara } 4974f8ad655SJan Kara 4984f8ad655SJan Kara /* 4994f8ad655SJan Kara * Write out an inode's dirty pages. Either the caller has an active reference 5004f8ad655SJan Kara * on the inode or the inode has I_WILL_FREE set. 5014f8ad655SJan Kara * 5024f8ad655SJan Kara * This function is designed to be called for writing back one inode which 5034f8ad655SJan Kara * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode() 5044f8ad655SJan Kara * and does more profound writeback list handling in writeback_sb_inodes(). 5054f8ad655SJan Kara */ 5064f8ad655SJan Kara static int 5074f8ad655SJan Kara writeback_single_inode(struct inode *inode, struct bdi_writeback *wb, 5084f8ad655SJan Kara struct writeback_control *wbc) 5094f8ad655SJan Kara { 5104f8ad655SJan Kara int ret = 0; 5114f8ad655SJan Kara 5124f8ad655SJan Kara spin_lock(&inode->i_lock); 5134f8ad655SJan Kara if (!atomic_read(&inode->i_count)) 5144f8ad655SJan Kara WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); 5154f8ad655SJan Kara else 5164f8ad655SJan Kara WARN_ON(inode->i_state & I_WILL_FREE); 5174f8ad655SJan Kara 5184f8ad655SJan Kara if (inode->i_state & I_SYNC) { 5194f8ad655SJan Kara if (wbc->sync_mode != WB_SYNC_ALL) 5204f8ad655SJan Kara goto out; 5214f8ad655SJan Kara /* 522169ebd90SJan Kara * It's a data-integrity sync. We must wait. Since callers hold 523169ebd90SJan Kara * inode reference or inode has I_WILL_FREE set, it cannot go 524169ebd90SJan Kara * away under us. 5254f8ad655SJan Kara */ 526169ebd90SJan Kara __inode_wait_for_writeback(inode); 5274f8ad655SJan Kara } 5284f8ad655SJan Kara WARN_ON(inode->i_state & I_SYNC); 5294f8ad655SJan Kara /* 530f9b0e058SJan Kara * Skip inode if it is clean and we have no outstanding writeback in 531f9b0e058SJan Kara * WB_SYNC_ALL mode. We don't want to mess with writeback lists in this 532f9b0e058SJan Kara * function since flusher thread may be doing for example sync in 533f9b0e058SJan Kara * parallel and if we move the inode, it could get skipped. So here we 534f9b0e058SJan Kara * make sure inode is on some writeback list and leave it there unless 535f9b0e058SJan Kara * we have completely cleaned the inode. 5364f8ad655SJan Kara */ 537f9b0e058SJan Kara if (!(inode->i_state & I_DIRTY) && 538f9b0e058SJan Kara (wbc->sync_mode != WB_SYNC_ALL || 539f9b0e058SJan Kara !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK))) 5404f8ad655SJan Kara goto out; 5414f8ad655SJan Kara inode->i_state |= I_SYNC; 5424f8ad655SJan Kara spin_unlock(&inode->i_lock); 5434f8ad655SJan Kara 544cd8ed2a4SYan Hong ret = __writeback_single_inode(inode, wbc); 5451da177e4SLinus Torvalds 546f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 547250df6edSDave Chinner spin_lock(&inode->i_lock); 5484f8ad655SJan Kara /* 5494f8ad655SJan Kara * If inode is clean, remove it from writeback lists. Otherwise don't 5504f8ad655SJan Kara * touch it. See comment above for explanation. 5514f8ad655SJan Kara */ 5524f8ad655SJan Kara if (!(inode->i_state & I_DIRTY)) 5534f8ad655SJan Kara list_del_init(&inode->i_wb_list); 5544f8ad655SJan Kara spin_unlock(&wb->list_lock); 5551c0eeaf5SJoern Engel inode_sync_complete(inode); 5564f8ad655SJan Kara out: 5574f8ad655SJan Kara spin_unlock(&inode->i_lock); 5581da177e4SLinus Torvalds return ret; 5591da177e4SLinus Torvalds } 5601da177e4SLinus Torvalds 5611a12d8bdSWu Fengguang static long writeback_chunk_size(struct backing_dev_info *bdi, 5621a12d8bdSWu Fengguang struct wb_writeback_work *work) 563d46db3d5SWu Fengguang { 564d46db3d5SWu Fengguang long pages; 565d46db3d5SWu Fengguang 566d46db3d5SWu Fengguang /* 567d46db3d5SWu Fengguang * WB_SYNC_ALL mode does livelock avoidance by syncing dirty 568d46db3d5SWu Fengguang * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX 569d46db3d5SWu Fengguang * here avoids calling into writeback_inodes_wb() more than once. 570d46db3d5SWu Fengguang * 571d46db3d5SWu Fengguang * The intended call sequence for WB_SYNC_ALL writeback is: 572d46db3d5SWu Fengguang * 573d46db3d5SWu Fengguang * wb_writeback() 574d46db3d5SWu Fengguang * writeback_sb_inodes() <== called only once 575d46db3d5SWu Fengguang * write_cache_pages() <== called once for each inode 576d46db3d5SWu Fengguang * (quickly) tag currently dirty pages 577d46db3d5SWu Fengguang * (maybe slowly) sync all tagged pages 578d46db3d5SWu Fengguang */ 579d46db3d5SWu Fengguang if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages) 580d46db3d5SWu Fengguang pages = LONG_MAX; 5811a12d8bdSWu Fengguang else { 5821a12d8bdSWu Fengguang pages = min(bdi->avg_write_bandwidth / 2, 5831a12d8bdSWu Fengguang global_dirty_limit / DIRTY_SCOPE); 5841a12d8bdSWu Fengguang pages = min(pages, work->nr_pages); 5851a12d8bdSWu Fengguang pages = round_down(pages + MIN_WRITEBACK_PAGES, 5861a12d8bdSWu Fengguang MIN_WRITEBACK_PAGES); 5871a12d8bdSWu Fengguang } 588d46db3d5SWu Fengguang 589d46db3d5SWu Fengguang return pages; 590d46db3d5SWu Fengguang } 591d46db3d5SWu Fengguang 59203ba3782SJens Axboe /* 593f11c9c5cSEdward Shishkin * Write a portion of b_io inodes which belong to @sb. 594edadfb10SChristoph Hellwig * 595d46db3d5SWu Fengguang * Return the number of pages and/or inodes written. 596f11c9c5cSEdward Shishkin */ 597d46db3d5SWu Fengguang static long writeback_sb_inodes(struct super_block *sb, 598d46db3d5SWu Fengguang struct bdi_writeback *wb, 599d46db3d5SWu Fengguang struct wb_writeback_work *work) 60003ba3782SJens Axboe { 601d46db3d5SWu Fengguang struct writeback_control wbc = { 602d46db3d5SWu Fengguang .sync_mode = work->sync_mode, 603d46db3d5SWu Fengguang .tagged_writepages = work->tagged_writepages, 604d46db3d5SWu Fengguang .for_kupdate = work->for_kupdate, 605d46db3d5SWu Fengguang .for_background = work->for_background, 6067747bd4bSDave Chinner .for_sync = work->for_sync, 607d46db3d5SWu Fengguang .range_cyclic = work->range_cyclic, 608d46db3d5SWu Fengguang .range_start = 0, 609d46db3d5SWu Fengguang .range_end = LLONG_MAX, 610d46db3d5SWu Fengguang }; 611d46db3d5SWu Fengguang unsigned long start_time = jiffies; 612d46db3d5SWu Fengguang long write_chunk; 613d46db3d5SWu Fengguang long wrote = 0; /* count both pages and inodes */ 614d46db3d5SWu Fengguang 61503ba3782SJens Axboe while (!list_empty(&wb->b_io)) { 6167ccf19a8SNick Piggin struct inode *inode = wb_inode(wb->b_io.prev); 617edadfb10SChristoph Hellwig 618edadfb10SChristoph Hellwig if (inode->i_sb != sb) { 619d46db3d5SWu Fengguang if (work->sb) { 620edadfb10SChristoph Hellwig /* 621edadfb10SChristoph Hellwig * We only want to write back data for this 622edadfb10SChristoph Hellwig * superblock, move all inodes not belonging 623edadfb10SChristoph Hellwig * to it back onto the dirty list. 624edadfb10SChristoph Hellwig */ 625f758eeabSChristoph Hellwig redirty_tail(inode, wb); 62666f3b8e2SJens Axboe continue; 62766f3b8e2SJens Axboe } 628edadfb10SChristoph Hellwig 629edadfb10SChristoph Hellwig /* 630edadfb10SChristoph Hellwig * The inode belongs to a different superblock. 631edadfb10SChristoph Hellwig * Bounce back to the caller to unpin this and 632edadfb10SChristoph Hellwig * pin the next superblock. 633edadfb10SChristoph Hellwig */ 634d46db3d5SWu Fengguang break; 635edadfb10SChristoph Hellwig } 636edadfb10SChristoph Hellwig 6379843b76aSChristoph Hellwig /* 638331cbdeeSWanpeng Li * Don't bother with new inodes or inodes being freed, first 639331cbdeeSWanpeng Li * kind does not need periodic writeout yet, and for the latter 6409843b76aSChristoph Hellwig * kind writeout is handled by the freer. 6419843b76aSChristoph Hellwig */ 642250df6edSDave Chinner spin_lock(&inode->i_lock); 6439843b76aSChristoph Hellwig if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { 644250df6edSDave Chinner spin_unlock(&inode->i_lock); 645fcc5c222SWu Fengguang redirty_tail(inode, wb); 6467ef0d737SNick Piggin continue; 6477ef0d737SNick Piggin } 648cc1676d9SJan Kara if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) { 649cc1676d9SJan Kara /* 650cc1676d9SJan Kara * If this inode is locked for writeback and we are not 651cc1676d9SJan Kara * doing writeback-for-data-integrity, move it to 652cc1676d9SJan Kara * b_more_io so that writeback can proceed with the 653cc1676d9SJan Kara * other inodes on s_io. 654cc1676d9SJan Kara * 655cc1676d9SJan Kara * We'll have another go at writing back this inode 656cc1676d9SJan Kara * when we completed a full scan of b_io. 657cc1676d9SJan Kara */ 658cc1676d9SJan Kara spin_unlock(&inode->i_lock); 659cc1676d9SJan Kara requeue_io(inode, wb); 660cc1676d9SJan Kara trace_writeback_sb_inodes_requeue(inode); 661cc1676d9SJan Kara continue; 662cc1676d9SJan Kara } 663f0d07b7fSJan Kara spin_unlock(&wb->list_lock); 664f0d07b7fSJan Kara 6654f8ad655SJan Kara /* 6664f8ad655SJan Kara * We already requeued the inode if it had I_SYNC set and we 6674f8ad655SJan Kara * are doing WB_SYNC_NONE writeback. So this catches only the 6684f8ad655SJan Kara * WB_SYNC_ALL case. 6694f8ad655SJan Kara */ 670169ebd90SJan Kara if (inode->i_state & I_SYNC) { 671169ebd90SJan Kara /* Wait for I_SYNC. This function drops i_lock... */ 672169ebd90SJan Kara inode_sleep_on_writeback(inode); 673169ebd90SJan Kara /* Inode may be gone, start again */ 674ead188f9SJan Kara spin_lock(&wb->list_lock); 675169ebd90SJan Kara continue; 676169ebd90SJan Kara } 6774f8ad655SJan Kara inode->i_state |= I_SYNC; 6784f8ad655SJan Kara spin_unlock(&inode->i_lock); 679169ebd90SJan Kara 6801a12d8bdSWu Fengguang write_chunk = writeback_chunk_size(wb->bdi, work); 681d46db3d5SWu Fengguang wbc.nr_to_write = write_chunk; 682d46db3d5SWu Fengguang wbc.pages_skipped = 0; 683250df6edSDave Chinner 684169ebd90SJan Kara /* 685169ebd90SJan Kara * We use I_SYNC to pin the inode in memory. While it is set 686169ebd90SJan Kara * evict_inode() will wait so the inode cannot be freed. 687169ebd90SJan Kara */ 688cd8ed2a4SYan Hong __writeback_single_inode(inode, &wbc); 689d46db3d5SWu Fengguang 690d46db3d5SWu Fengguang work->nr_pages -= write_chunk - wbc.nr_to_write; 691d46db3d5SWu Fengguang wrote += write_chunk - wbc.nr_to_write; 6924f8ad655SJan Kara spin_lock(&wb->list_lock); 6934f8ad655SJan Kara spin_lock(&inode->i_lock); 694d46db3d5SWu Fengguang if (!(inode->i_state & I_DIRTY)) 695d46db3d5SWu Fengguang wrote++; 6964f8ad655SJan Kara requeue_inode(inode, wb, &wbc); 6974f8ad655SJan Kara inode_sync_complete(inode); 6980f1b1fd8SDave Chinner spin_unlock(&inode->i_lock); 699169ebd90SJan Kara cond_resched_lock(&wb->list_lock); 700d46db3d5SWu Fengguang /* 701d46db3d5SWu Fengguang * bail out to wb_writeback() often enough to check 702d46db3d5SWu Fengguang * background threshold and other termination conditions. 703d46db3d5SWu Fengguang */ 704d46db3d5SWu Fengguang if (wrote) { 705d46db3d5SWu Fengguang if (time_is_before_jiffies(start_time + HZ / 10UL)) 706d46db3d5SWu Fengguang break; 707d46db3d5SWu Fengguang if (work->nr_pages <= 0) 708d46db3d5SWu Fengguang break; 7091da177e4SLinus Torvalds } 7108bc3be27SFengguang Wu } 711d46db3d5SWu Fengguang return wrote; 712f11c9c5cSEdward Shishkin } 71338f21977SNick Piggin 714d46db3d5SWu Fengguang static long __writeback_inodes_wb(struct bdi_writeback *wb, 715d46db3d5SWu Fengguang struct wb_writeback_work *work) 716f11c9c5cSEdward Shishkin { 717d46db3d5SWu Fengguang unsigned long start_time = jiffies; 718d46db3d5SWu Fengguang long wrote = 0; 719f11c9c5cSEdward Shishkin 720f11c9c5cSEdward Shishkin while (!list_empty(&wb->b_io)) { 7217ccf19a8SNick Piggin struct inode *inode = wb_inode(wb->b_io.prev); 722f11c9c5cSEdward Shishkin struct super_block *sb = inode->i_sb; 723f11c9c5cSEdward Shishkin 72412ad3ab6SDave Chinner if (!grab_super_passive(sb)) { 7250e995816SWu Fengguang /* 7260e995816SWu Fengguang * grab_super_passive() may fail consistently due to 7270e995816SWu Fengguang * s_umount being grabbed by someone else. Don't use 7280e995816SWu Fengguang * requeue_io() to avoid busy retrying the inode/sb. 7290e995816SWu Fengguang */ 7300e995816SWu Fengguang redirty_tail(inode, wb); 731d19de7edSChristoph Hellwig continue; 732334132aeSChristoph Hellwig } 733d46db3d5SWu Fengguang wrote += writeback_sb_inodes(sb, wb, work); 734d19de7edSChristoph Hellwig drop_super(sb); 735f11c9c5cSEdward Shishkin 736d46db3d5SWu Fengguang /* refer to the same tests at the end of writeback_sb_inodes */ 737d46db3d5SWu Fengguang if (wrote) { 738d46db3d5SWu Fengguang if (time_is_before_jiffies(start_time + HZ / 10UL)) 739d46db3d5SWu Fengguang break; 740d46db3d5SWu Fengguang if (work->nr_pages <= 0) 741f11c9c5cSEdward Shishkin break; 742f11c9c5cSEdward Shishkin } 743d46db3d5SWu Fengguang } 74466f3b8e2SJens Axboe /* Leave any unwritten inodes on b_io */ 745d46db3d5SWu Fengguang return wrote; 74666f3b8e2SJens Axboe } 74766f3b8e2SJens Axboe 7487d9f073bSWanpeng Li static long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages, 7490e175a18SCurt Wohlgemuth enum wb_reason reason) 750edadfb10SChristoph Hellwig { 751d46db3d5SWu Fengguang struct wb_writeback_work work = { 752d46db3d5SWu Fengguang .nr_pages = nr_pages, 753d46db3d5SWu Fengguang .sync_mode = WB_SYNC_NONE, 754d46db3d5SWu Fengguang .range_cyclic = 1, 7550e175a18SCurt Wohlgemuth .reason = reason, 756d46db3d5SWu Fengguang }; 757edadfb10SChristoph Hellwig 758f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 759424b351fSWu Fengguang if (list_empty(&wb->b_io)) 760ad4e38ddSCurt Wohlgemuth queue_io(wb, &work); 761d46db3d5SWu Fengguang __writeback_inodes_wb(wb, &work); 762f758eeabSChristoph Hellwig spin_unlock(&wb->list_lock); 763edadfb10SChristoph Hellwig 764d46db3d5SWu Fengguang return nr_pages - work.nr_pages; 76566f3b8e2SJens Axboe } 76666f3b8e2SJens Axboe 767b00949aaSWu Fengguang static bool over_bground_thresh(struct backing_dev_info *bdi) 76803ba3782SJens Axboe { 76903ba3782SJens Axboe unsigned long background_thresh, dirty_thresh; 77003ba3782SJens Axboe 77116c4042fSWu Fengguang global_dirty_limits(&background_thresh, &dirty_thresh); 77203ba3782SJens Axboe 773b00949aaSWu Fengguang if (global_page_state(NR_FILE_DIRTY) + 774b00949aaSWu Fengguang global_page_state(NR_UNSTABLE_NFS) > background_thresh) 775b00949aaSWu Fengguang return true; 776b00949aaSWu Fengguang 777b00949aaSWu Fengguang if (bdi_stat(bdi, BDI_RECLAIMABLE) > 778b00949aaSWu Fengguang bdi_dirty_limit(bdi, background_thresh)) 779b00949aaSWu Fengguang return true; 780b00949aaSWu Fengguang 781b00949aaSWu Fengguang return false; 78203ba3782SJens Axboe } 78303ba3782SJens Axboe 78403ba3782SJens Axboe /* 785e98be2d5SWu Fengguang * Called under wb->list_lock. If there are multiple wb per bdi, 786e98be2d5SWu Fengguang * only the flusher working on the first wb should do it. 787e98be2d5SWu Fengguang */ 788e98be2d5SWu Fengguang static void wb_update_bandwidth(struct bdi_writeback *wb, 789e98be2d5SWu Fengguang unsigned long start_time) 790e98be2d5SWu Fengguang { 791af6a3113SWu Fengguang __bdi_update_bandwidth(wb->bdi, 0, 0, 0, 0, 0, start_time); 792e98be2d5SWu Fengguang } 793e98be2d5SWu Fengguang 794e98be2d5SWu Fengguang /* 79503ba3782SJens Axboe * Explicit flushing or periodic writeback of "old" data. 79603ba3782SJens Axboe * 79703ba3782SJens Axboe * Define "old": the first time one of an inode's pages is dirtied, we mark the 79803ba3782SJens Axboe * dirtying-time in the inode's address_space. So this periodic writeback code 79903ba3782SJens Axboe * just walks the superblock inode list, writing back any inodes which are 80003ba3782SJens Axboe * older than a specific point in time. 80103ba3782SJens Axboe * 80203ba3782SJens Axboe * Try to run once per dirty_writeback_interval. But if a writeback event 80303ba3782SJens Axboe * takes longer than a dirty_writeback_interval interval, then leave a 80403ba3782SJens Axboe * one-second gap. 80503ba3782SJens Axboe * 80603ba3782SJens Axboe * older_than_this takes precedence over nr_to_write. So we'll only write back 80703ba3782SJens Axboe * all dirty pages if they are all attached to "old" mappings. 80803ba3782SJens Axboe */ 809c4a77a6cSJens Axboe static long wb_writeback(struct bdi_writeback *wb, 81083ba7b07SChristoph Hellwig struct wb_writeback_work *work) 81103ba3782SJens Axboe { 812e98be2d5SWu Fengguang unsigned long wb_start = jiffies; 813d46db3d5SWu Fengguang long nr_pages = work->nr_pages; 8140dc83bd3SJan Kara unsigned long oldest_jif; 815a5989bdcSJan Kara struct inode *inode; 816d46db3d5SWu Fengguang long progress; 81703ba3782SJens Axboe 8180dc83bd3SJan Kara oldest_jif = jiffies; 8190dc83bd3SJan Kara work->older_than_this = &oldest_jif; 82003ba3782SJens Axboe 821e8dfc305SWu Fengguang spin_lock(&wb->list_lock); 82203ba3782SJens Axboe for (;;) { 82303ba3782SJens Axboe /* 824d3ddec76SWu Fengguang * Stop writeback when nr_pages has been consumed 82503ba3782SJens Axboe */ 82683ba7b07SChristoph Hellwig if (work->nr_pages <= 0) 82703ba3782SJens Axboe break; 82803ba3782SJens Axboe 82903ba3782SJens Axboe /* 830aa373cf5SJan Kara * Background writeout and kupdate-style writeback may 831aa373cf5SJan Kara * run forever. Stop them if there is other work to do 832aa373cf5SJan Kara * so that e.g. sync can proceed. They'll be restarted 833aa373cf5SJan Kara * after the other works are all done. 834aa373cf5SJan Kara */ 835aa373cf5SJan Kara if ((work->for_background || work->for_kupdate) && 836aa373cf5SJan Kara !list_empty(&wb->bdi->work_list)) 837aa373cf5SJan Kara break; 838aa373cf5SJan Kara 839aa373cf5SJan Kara /* 840d3ddec76SWu Fengguang * For background writeout, stop when we are below the 841d3ddec76SWu Fengguang * background dirty threshold 84203ba3782SJens Axboe */ 843b00949aaSWu Fengguang if (work->for_background && !over_bground_thresh(wb->bdi)) 84403ba3782SJens Axboe break; 84503ba3782SJens Axboe 8461bc36b64SJan Kara /* 8471bc36b64SJan Kara * Kupdate and background works are special and we want to 8481bc36b64SJan Kara * include all inodes that need writing. Livelock avoidance is 8491bc36b64SJan Kara * handled by these works yielding to any other work so we are 8501bc36b64SJan Kara * safe. 8511bc36b64SJan Kara */ 852ba9aa839SWu Fengguang if (work->for_kupdate) { 8530dc83bd3SJan Kara oldest_jif = jiffies - 854ba9aa839SWu Fengguang msecs_to_jiffies(dirty_expire_interval * 10); 8551bc36b64SJan Kara } else if (work->for_background) 8560dc83bd3SJan Kara oldest_jif = jiffies; 857028c2dd1SDave Chinner 858d46db3d5SWu Fengguang trace_writeback_start(wb->bdi, work); 859e8dfc305SWu Fengguang if (list_empty(&wb->b_io)) 860ad4e38ddSCurt Wohlgemuth queue_io(wb, work); 86183ba7b07SChristoph Hellwig if (work->sb) 862d46db3d5SWu Fengguang progress = writeback_sb_inodes(work->sb, wb, work); 863edadfb10SChristoph Hellwig else 864d46db3d5SWu Fengguang progress = __writeback_inodes_wb(wb, work); 865d46db3d5SWu Fengguang trace_writeback_written(wb->bdi, work); 866028c2dd1SDave Chinner 867e98be2d5SWu Fengguang wb_update_bandwidth(wb, wb_start); 86803ba3782SJens Axboe 86903ba3782SJens Axboe /* 87071fd05a8SJens Axboe * Did we write something? Try for more 871e6fb6da2SWu Fengguang * 872e6fb6da2SWu Fengguang * Dirty inodes are moved to b_io for writeback in batches. 873e6fb6da2SWu Fengguang * The completion of the current batch does not necessarily 874e6fb6da2SWu Fengguang * mean the overall work is done. So we keep looping as long 875e6fb6da2SWu Fengguang * as made some progress on cleaning pages or inodes. 87671fd05a8SJens Axboe */ 877d46db3d5SWu Fengguang if (progress) 87803ba3782SJens Axboe continue; 879a5989bdcSJan Kara /* 880e6fb6da2SWu Fengguang * No more inodes for IO, bail 881a5989bdcSJan Kara */ 882b7a2441fSWu Fengguang if (list_empty(&wb->b_more_io)) 88303ba3782SJens Axboe break; 88403ba3782SJens Axboe /* 8858010c3b6SJens Axboe * Nothing written. Wait for some inode to 8868010c3b6SJens Axboe * become available for writeback. Otherwise 8878010c3b6SJens Axboe * we'll just busyloop. 88803ba3782SJens Axboe */ 88903ba3782SJens Axboe if (!list_empty(&wb->b_more_io)) { 890d46db3d5SWu Fengguang trace_writeback_wait(wb->bdi, work); 89103ba3782SJens Axboe inode = wb_inode(wb->b_more_io.prev); 892250df6edSDave Chinner spin_lock(&inode->i_lock); 893f0d07b7fSJan Kara spin_unlock(&wb->list_lock); 894169ebd90SJan Kara /* This function drops i_lock... */ 895169ebd90SJan Kara inode_sleep_on_writeback(inode); 896f0d07b7fSJan Kara spin_lock(&wb->list_lock); 89703ba3782SJens Axboe } 89803ba3782SJens Axboe } 899e8dfc305SWu Fengguang spin_unlock(&wb->list_lock); 90003ba3782SJens Axboe 901d46db3d5SWu Fengguang return nr_pages - work->nr_pages; 90203ba3782SJens Axboe } 90303ba3782SJens Axboe 90403ba3782SJens Axboe /* 90583ba7b07SChristoph Hellwig * Return the next wb_writeback_work struct that hasn't been processed yet. 90603ba3782SJens Axboe */ 90783ba7b07SChristoph Hellwig static struct wb_writeback_work * 90808852b6dSMinchan Kim get_next_work_item(struct backing_dev_info *bdi) 90903ba3782SJens Axboe { 91083ba7b07SChristoph Hellwig struct wb_writeback_work *work = NULL; 91103ba3782SJens Axboe 9126467716aSArtem Bityutskiy spin_lock_bh(&bdi->wb_lock); 91383ba7b07SChristoph Hellwig if (!list_empty(&bdi->work_list)) { 91483ba7b07SChristoph Hellwig work = list_entry(bdi->work_list.next, 91583ba7b07SChristoph Hellwig struct wb_writeback_work, list); 91683ba7b07SChristoph Hellwig list_del_init(&work->list); 91703ba3782SJens Axboe } 9186467716aSArtem Bityutskiy spin_unlock_bh(&bdi->wb_lock); 91983ba7b07SChristoph Hellwig return work; 92003ba3782SJens Axboe } 92103ba3782SJens Axboe 922cdf01dd5SLinus Torvalds /* 923cdf01dd5SLinus Torvalds * Add in the number of potentially dirty inodes, because each inode 924cdf01dd5SLinus Torvalds * write can dirty pagecache in the underlying blockdev. 925cdf01dd5SLinus Torvalds */ 926cdf01dd5SLinus Torvalds static unsigned long get_nr_dirty_pages(void) 927cdf01dd5SLinus Torvalds { 928cdf01dd5SLinus Torvalds return global_page_state(NR_FILE_DIRTY) + 929cdf01dd5SLinus Torvalds global_page_state(NR_UNSTABLE_NFS) + 930cdf01dd5SLinus Torvalds get_nr_dirty_inodes(); 931cdf01dd5SLinus Torvalds } 932cdf01dd5SLinus Torvalds 9336585027aSJan Kara static long wb_check_background_flush(struct bdi_writeback *wb) 9346585027aSJan Kara { 935b00949aaSWu Fengguang if (over_bground_thresh(wb->bdi)) { 9366585027aSJan Kara 9376585027aSJan Kara struct wb_writeback_work work = { 9386585027aSJan Kara .nr_pages = LONG_MAX, 9396585027aSJan Kara .sync_mode = WB_SYNC_NONE, 9406585027aSJan Kara .for_background = 1, 9416585027aSJan Kara .range_cyclic = 1, 9420e175a18SCurt Wohlgemuth .reason = WB_REASON_BACKGROUND, 9436585027aSJan Kara }; 9446585027aSJan Kara 9456585027aSJan Kara return wb_writeback(wb, &work); 9466585027aSJan Kara } 9476585027aSJan Kara 9486585027aSJan Kara return 0; 9496585027aSJan Kara } 9506585027aSJan Kara 95103ba3782SJens Axboe static long wb_check_old_data_flush(struct bdi_writeback *wb) 95203ba3782SJens Axboe { 95303ba3782SJens Axboe unsigned long expired; 95403ba3782SJens Axboe long nr_pages; 95503ba3782SJens Axboe 95669b62d01SJens Axboe /* 95769b62d01SJens Axboe * When set to zero, disable periodic writeback 95869b62d01SJens Axboe */ 95969b62d01SJens Axboe if (!dirty_writeback_interval) 96069b62d01SJens Axboe return 0; 96169b62d01SJens Axboe 96203ba3782SJens Axboe expired = wb->last_old_flush + 96303ba3782SJens Axboe msecs_to_jiffies(dirty_writeback_interval * 10); 96403ba3782SJens Axboe if (time_before(jiffies, expired)) 96503ba3782SJens Axboe return 0; 96603ba3782SJens Axboe 96703ba3782SJens Axboe wb->last_old_flush = jiffies; 968cdf01dd5SLinus Torvalds nr_pages = get_nr_dirty_pages(); 96903ba3782SJens Axboe 970c4a77a6cSJens Axboe if (nr_pages) { 97183ba7b07SChristoph Hellwig struct wb_writeback_work work = { 972c4a77a6cSJens Axboe .nr_pages = nr_pages, 973c4a77a6cSJens Axboe .sync_mode = WB_SYNC_NONE, 974c4a77a6cSJens Axboe .for_kupdate = 1, 975c4a77a6cSJens Axboe .range_cyclic = 1, 9760e175a18SCurt Wohlgemuth .reason = WB_REASON_PERIODIC, 977c4a77a6cSJens Axboe }; 978c4a77a6cSJens Axboe 97983ba7b07SChristoph Hellwig return wb_writeback(wb, &work); 980c4a77a6cSJens Axboe } 98103ba3782SJens Axboe 98203ba3782SJens Axboe return 0; 98303ba3782SJens Axboe } 98403ba3782SJens Axboe 98503ba3782SJens Axboe /* 98603ba3782SJens Axboe * Retrieve work items and do the writeback they describe 98703ba3782SJens Axboe */ 98825d130baSWanpeng Li static long wb_do_writeback(struct bdi_writeback *wb) 98903ba3782SJens Axboe { 99003ba3782SJens Axboe struct backing_dev_info *bdi = wb->bdi; 99183ba7b07SChristoph Hellwig struct wb_writeback_work *work; 992c4a77a6cSJens Axboe long wrote = 0; 99303ba3782SJens Axboe 99481d73a32SJan Kara set_bit(BDI_writeback_running, &wb->bdi->state); 99508852b6dSMinchan Kim while ((work = get_next_work_item(bdi)) != NULL) { 99683ba7b07SChristoph Hellwig 997455b2864SDave Chinner trace_writeback_exec(bdi, work); 998455b2864SDave Chinner 99983ba7b07SChristoph Hellwig wrote += wb_writeback(wb, work); 100003ba3782SJens Axboe 100103ba3782SJens Axboe /* 100283ba7b07SChristoph Hellwig * Notify the caller of completion if this is a synchronous 100383ba7b07SChristoph Hellwig * work item, otherwise just free it. 100403ba3782SJens Axboe */ 100583ba7b07SChristoph Hellwig if (work->done) 100683ba7b07SChristoph Hellwig complete(work->done); 100783ba7b07SChristoph Hellwig else 100883ba7b07SChristoph Hellwig kfree(work); 100903ba3782SJens Axboe } 101003ba3782SJens Axboe 101103ba3782SJens Axboe /* 101203ba3782SJens Axboe * Check for periodic writeback, kupdated() style 101303ba3782SJens Axboe */ 101403ba3782SJens Axboe wrote += wb_check_old_data_flush(wb); 10156585027aSJan Kara wrote += wb_check_background_flush(wb); 101681d73a32SJan Kara clear_bit(BDI_writeback_running, &wb->bdi->state); 101703ba3782SJens Axboe 101803ba3782SJens Axboe return wrote; 101903ba3782SJens Axboe } 102003ba3782SJens Axboe 102103ba3782SJens Axboe /* 102203ba3782SJens Axboe * Handle writeback of dirty data for the device backed by this bdi. Also 1023839a8e86STejun Heo * reschedules periodically and does kupdated style flushing. 102403ba3782SJens Axboe */ 1025839a8e86STejun Heo void bdi_writeback_workfn(struct work_struct *work) 102603ba3782SJens Axboe { 1027839a8e86STejun Heo struct bdi_writeback *wb = container_of(to_delayed_work(work), 1028839a8e86STejun Heo struct bdi_writeback, dwork); 102908243900SChristoph Hellwig struct backing_dev_info *bdi = wb->bdi; 103003ba3782SJens Axboe long pages_written; 103103ba3782SJens Axboe 1032ef3b1019STejun Heo set_worker_desc("flush-%s", dev_name(bdi->dev)); 1033766f9164SPeter Zijlstra current->flags |= PF_SWAPWRITE; 103403ba3782SJens Axboe 1035839a8e86STejun Heo if (likely(!current_is_workqueue_rescuer() || 10365acda9d1SJan Kara !test_bit(BDI_registered, &bdi->state))) { 103703ba3782SJens Axboe /* 1038839a8e86STejun Heo * The normal path. Keep writing back @bdi until its 1039839a8e86STejun Heo * work_list is empty. Note that this path is also taken 1040839a8e86STejun Heo * if @bdi is shutting down even when we're running off the 1041839a8e86STejun Heo * rescuer as work_list needs to be drained. 104203ba3782SJens Axboe */ 1043839a8e86STejun Heo do { 104425d130baSWanpeng Li pages_written = wb_do_writeback(wb); 1045455b2864SDave Chinner trace_writeback_pages_written(pages_written); 1046839a8e86STejun Heo } while (!list_empty(&bdi->work_list)); 1047839a8e86STejun Heo } else { 1048253c34e9SArtem Bityutskiy /* 1049839a8e86STejun Heo * bdi_wq can't get enough workers and we're running off 1050839a8e86STejun Heo * the emergency worker. Don't hog it. Hopefully, 1024 is 1051839a8e86STejun Heo * enough for efficient IO. 1052253c34e9SArtem Bityutskiy */ 1053839a8e86STejun Heo pages_written = writeback_inodes_wb(&bdi->wb, 1024, 1054839a8e86STejun Heo WB_REASON_FORKER_THREAD); 1055839a8e86STejun Heo trace_writeback_pages_written(pages_written); 105603ba3782SJens Axboe } 105703ba3782SJens Axboe 10586ca738d6SDerek Basehore if (!list_empty(&bdi->work_list)) 10596ca738d6SDerek Basehore mod_delayed_work(bdi_wq, &wb->dwork, 0); 10606ca738d6SDerek Basehore else if (wb_has_dirty_io(wb) && dirty_writeback_interval) 10616ca738d6SDerek Basehore bdi_wakeup_thread_delayed(bdi); 1062455b2864SDave Chinner 1063839a8e86STejun Heo current->flags &= ~PF_SWAPWRITE; 106403ba3782SJens Axboe } 106503ba3782SJens Axboe 106603ba3782SJens Axboe /* 106703ba3782SJens Axboe * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back 106803ba3782SJens Axboe * the whole world. 106903ba3782SJens Axboe */ 10700e175a18SCurt Wohlgemuth void wakeup_flusher_threads(long nr_pages, enum wb_reason reason) 107103ba3782SJens Axboe { 1072b8c2f347SChristoph Hellwig struct backing_dev_info *bdi; 1073b8c2f347SChristoph Hellwig 107447df3ddeSJan Kara if (!nr_pages) 107547df3ddeSJan Kara nr_pages = get_nr_dirty_pages(); 1076b8c2f347SChristoph Hellwig 1077b8c2f347SChristoph Hellwig rcu_read_lock(); 1078b8c2f347SChristoph Hellwig list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { 1079b8c2f347SChristoph Hellwig if (!bdi_has_dirty_io(bdi)) 1080b8c2f347SChristoph Hellwig continue; 10810e175a18SCurt Wohlgemuth __bdi_start_writeback(bdi, nr_pages, false, reason); 1082b8c2f347SChristoph Hellwig } 1083b8c2f347SChristoph Hellwig rcu_read_unlock(); 108403ba3782SJens Axboe } 108503ba3782SJens Axboe 108603ba3782SJens Axboe static noinline void block_dump___mark_inode_dirty(struct inode *inode) 108703ba3782SJens Axboe { 108803ba3782SJens Axboe if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { 108903ba3782SJens Axboe struct dentry *dentry; 109003ba3782SJens Axboe const char *name = "?"; 109103ba3782SJens Axboe 109203ba3782SJens Axboe dentry = d_find_alias(inode); 109303ba3782SJens Axboe if (dentry) { 109403ba3782SJens Axboe spin_lock(&dentry->d_lock); 109503ba3782SJens Axboe name = (const char *) dentry->d_name.name; 109603ba3782SJens Axboe } 109703ba3782SJens Axboe printk(KERN_DEBUG 109803ba3782SJens Axboe "%s(%d): dirtied inode %lu (%s) on %s\n", 109903ba3782SJens Axboe current->comm, task_pid_nr(current), inode->i_ino, 110003ba3782SJens Axboe name, inode->i_sb->s_id); 110103ba3782SJens Axboe if (dentry) { 110203ba3782SJens Axboe spin_unlock(&dentry->d_lock); 110303ba3782SJens Axboe dput(dentry); 110403ba3782SJens Axboe } 110503ba3782SJens Axboe } 110603ba3782SJens Axboe } 110703ba3782SJens Axboe 110803ba3782SJens Axboe /** 110903ba3782SJens Axboe * __mark_inode_dirty - internal function 111003ba3782SJens Axboe * @inode: inode to mark 111103ba3782SJens Axboe * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) 111203ba3782SJens Axboe * Mark an inode as dirty. Callers should use mark_inode_dirty or 111303ba3782SJens Axboe * mark_inode_dirty_sync. 111403ba3782SJens Axboe * 111503ba3782SJens Axboe * Put the inode on the super block's dirty list. 111603ba3782SJens Axboe * 111703ba3782SJens Axboe * CAREFUL! We mark it dirty unconditionally, but move it onto the 111803ba3782SJens Axboe * dirty list only if it is hashed or if it refers to a blockdev. 111903ba3782SJens Axboe * If it was not hashed, it will never be added to the dirty list 112003ba3782SJens Axboe * even if it is later hashed, as it will have been marked dirty already. 112103ba3782SJens Axboe * 112203ba3782SJens Axboe * In short, make sure you hash any inodes _before_ you start marking 112303ba3782SJens Axboe * them dirty. 112403ba3782SJens Axboe * 112503ba3782SJens Axboe * Note that for blockdevs, inode->dirtied_when represents the dirtying time of 112603ba3782SJens Axboe * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of 112703ba3782SJens Axboe * the kernel-internal blockdev inode represents the dirtying time of the 112803ba3782SJens Axboe * blockdev's pages. This is why for I_DIRTY_PAGES we always use 112903ba3782SJens Axboe * page->mapping->host, so the page-dirtying time is recorded in the internal 113003ba3782SJens Axboe * blockdev inode. 113103ba3782SJens Axboe */ 113203ba3782SJens Axboe void __mark_inode_dirty(struct inode *inode, int flags) 113303ba3782SJens Axboe { 113403ba3782SJens Axboe struct super_block *sb = inode->i_sb; 1135253c34e9SArtem Bityutskiy struct backing_dev_info *bdi = NULL; 113603ba3782SJens Axboe 113703ba3782SJens Axboe /* 113803ba3782SJens Axboe * Don't do this for I_DIRTY_PAGES - that doesn't actually 113903ba3782SJens Axboe * dirty the inode itself 114003ba3782SJens Axboe */ 114103ba3782SJens Axboe if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { 11429fb0a7daSTejun Heo trace_writeback_dirty_inode_start(inode, flags); 11439fb0a7daSTejun Heo 114403ba3782SJens Axboe if (sb->s_op->dirty_inode) 1145aa385729SChristoph Hellwig sb->s_op->dirty_inode(inode, flags); 11469fb0a7daSTejun Heo 11479fb0a7daSTejun Heo trace_writeback_dirty_inode(inode, flags); 114803ba3782SJens Axboe } 114903ba3782SJens Axboe 115003ba3782SJens Axboe /* 115103ba3782SJens Axboe * make sure that changes are seen by all cpus before we test i_state 115203ba3782SJens Axboe * -- mikulas 115303ba3782SJens Axboe */ 115403ba3782SJens Axboe smp_mb(); 115503ba3782SJens Axboe 115603ba3782SJens Axboe /* avoid the locking if we can */ 115703ba3782SJens Axboe if ((inode->i_state & flags) == flags) 115803ba3782SJens Axboe return; 115903ba3782SJens Axboe 116003ba3782SJens Axboe if (unlikely(block_dump)) 116103ba3782SJens Axboe block_dump___mark_inode_dirty(inode); 116203ba3782SJens Axboe 1163250df6edSDave Chinner spin_lock(&inode->i_lock); 116403ba3782SJens Axboe if ((inode->i_state & flags) != flags) { 116503ba3782SJens Axboe const int was_dirty = inode->i_state & I_DIRTY; 116603ba3782SJens Axboe 116703ba3782SJens Axboe inode->i_state |= flags; 116803ba3782SJens Axboe 116903ba3782SJens Axboe /* 117003ba3782SJens Axboe * If the inode is being synced, just update its dirty state. 117103ba3782SJens Axboe * The unlocker will place the inode on the appropriate 117203ba3782SJens Axboe * superblock list, based upon its state. 117303ba3782SJens Axboe */ 117403ba3782SJens Axboe if (inode->i_state & I_SYNC) 1175250df6edSDave Chinner goto out_unlock_inode; 117603ba3782SJens Axboe 117703ba3782SJens Axboe /* 117803ba3782SJens Axboe * Only add valid (hashed) inodes to the superblock's 117903ba3782SJens Axboe * dirty list. Add blockdev inodes as well. 118003ba3782SJens Axboe */ 118103ba3782SJens Axboe if (!S_ISBLK(inode->i_mode)) { 11821d3382cbSAl Viro if (inode_unhashed(inode)) 1183250df6edSDave Chinner goto out_unlock_inode; 118403ba3782SJens Axboe } 1185a4ffdde6SAl Viro if (inode->i_state & I_FREEING) 1186250df6edSDave Chinner goto out_unlock_inode; 118703ba3782SJens Axboe 118803ba3782SJens Axboe /* 118903ba3782SJens Axboe * If the inode was already on b_dirty/b_io/b_more_io, don't 119003ba3782SJens Axboe * reposition it (that would break b_dirty time-ordering). 119103ba3782SJens Axboe */ 119203ba3782SJens Axboe if (!was_dirty) { 1193a66979abSDave Chinner bool wakeup_bdi = false; 1194253c34e9SArtem Bityutskiy bdi = inode_to_bdi(inode); 1195500b067cSJens Axboe 1196146d7009SJunxiao Bi spin_unlock(&inode->i_lock); 1197146d7009SJunxiao Bi spin_lock(&bdi->wb.list_lock); 1198253c34e9SArtem Bityutskiy if (bdi_cap_writeback_dirty(bdi)) { 1199253c34e9SArtem Bityutskiy WARN(!test_bit(BDI_registered, &bdi->state), 1200253c34e9SArtem Bityutskiy "bdi-%s not registered\n", bdi->name); 1201253c34e9SArtem Bityutskiy 1202253c34e9SArtem Bityutskiy /* 1203253c34e9SArtem Bityutskiy * If this is the first dirty inode for this 1204253c34e9SArtem Bityutskiy * bdi, we have to wake-up the corresponding 1205253c34e9SArtem Bityutskiy * bdi thread to make sure background 1206253c34e9SArtem Bityutskiy * write-back happens later. 1207253c34e9SArtem Bityutskiy */ 1208253c34e9SArtem Bityutskiy if (!wb_has_dirty_io(&bdi->wb)) 1209253c34e9SArtem Bityutskiy wakeup_bdi = true; 1210500b067cSJens Axboe } 121103ba3782SJens Axboe 121203ba3782SJens Axboe inode->dirtied_when = jiffies; 12137ccf19a8SNick Piggin list_move(&inode->i_wb_list, &bdi->wb.b_dirty); 1214f758eeabSChristoph Hellwig spin_unlock(&bdi->wb.list_lock); 1215253c34e9SArtem Bityutskiy 1216253c34e9SArtem Bityutskiy if (wakeup_bdi) 12176467716aSArtem Bityutskiy bdi_wakeup_thread_delayed(bdi); 1218a66979abSDave Chinner return; 1219a66979abSDave Chinner } 1220a66979abSDave Chinner } 1221a66979abSDave Chinner out_unlock_inode: 1222a66979abSDave Chinner spin_unlock(&inode->i_lock); 1223a66979abSDave Chinner 122403ba3782SJens Axboe } 122503ba3782SJens Axboe EXPORT_SYMBOL(__mark_inode_dirty); 122603ba3782SJens Axboe 1227b6e51316SJens Axboe static void wait_sb_inodes(struct super_block *sb) 122866f3b8e2SJens Axboe { 122938f21977SNick Piggin struct inode *inode, *old_inode = NULL; 123038f21977SNick Piggin 123103ba3782SJens Axboe /* 123203ba3782SJens Axboe * We need to be protected against the filesystem going from 123303ba3782SJens Axboe * r/o to r/w or vice versa. 123403ba3782SJens Axboe */ 1235b6e51316SJens Axboe WARN_ON(!rwsem_is_locked(&sb->s_umount)); 123603ba3782SJens Axboe 123755fa6091SDave Chinner spin_lock(&inode_sb_list_lock); 123866f3b8e2SJens Axboe 123938f21977SNick Piggin /* 124038f21977SNick Piggin * Data integrity sync. Must wait for all pages under writeback, 124138f21977SNick Piggin * because there may have been pages dirtied before our sync 124238f21977SNick Piggin * call, but which had writeout started before we write it out. 124338f21977SNick Piggin * In which case, the inode may not be on the dirty list, but 124438f21977SNick Piggin * we still have to wait for that writeout. 124538f21977SNick Piggin */ 1246b6e51316SJens Axboe list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { 1247250df6edSDave Chinner struct address_space *mapping = inode->i_mapping; 124838f21977SNick Piggin 1249250df6edSDave Chinner spin_lock(&inode->i_lock); 1250250df6edSDave Chinner if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) || 1251250df6edSDave Chinner (mapping->nrpages == 0)) { 1252250df6edSDave Chinner spin_unlock(&inode->i_lock); 125338f21977SNick Piggin continue; 1254250df6edSDave Chinner } 125538f21977SNick Piggin __iget(inode); 1256250df6edSDave Chinner spin_unlock(&inode->i_lock); 125755fa6091SDave Chinner spin_unlock(&inode_sb_list_lock); 125855fa6091SDave Chinner 125938f21977SNick Piggin /* 126055fa6091SDave Chinner * We hold a reference to 'inode' so it couldn't have been 126155fa6091SDave Chinner * removed from s_inodes list while we dropped the 126255fa6091SDave Chinner * inode_sb_list_lock. We cannot iput the inode now as we can 126355fa6091SDave Chinner * be holding the last reference and we cannot iput it under 126455fa6091SDave Chinner * inode_sb_list_lock. So we keep the reference and iput it 126555fa6091SDave Chinner * later. 126638f21977SNick Piggin */ 126738f21977SNick Piggin iput(old_inode); 126838f21977SNick Piggin old_inode = inode; 126938f21977SNick Piggin 127038f21977SNick Piggin filemap_fdatawait(mapping); 127138f21977SNick Piggin 127238f21977SNick Piggin cond_resched(); 127338f21977SNick Piggin 127455fa6091SDave Chinner spin_lock(&inode_sb_list_lock); 127538f21977SNick Piggin } 127655fa6091SDave Chinner spin_unlock(&inode_sb_list_lock); 127738f21977SNick Piggin iput(old_inode); 127866f3b8e2SJens Axboe } 12791da177e4SLinus Torvalds 1280d8a8559cSJens Axboe /** 12813259f8beSChris Mason * writeback_inodes_sb_nr - writeback dirty inodes from given super_block 1282d8a8559cSJens Axboe * @sb: the superblock 12833259f8beSChris Mason * @nr: the number of pages to write 1284786228abSMarcos Paulo de Souza * @reason: reason why some writeback work initiated 12851da177e4SLinus Torvalds * 1286d8a8559cSJens Axboe * Start writeback on some inodes on this super_block. No guarantees are made 1287d8a8559cSJens Axboe * on how many (if any) will be written, and this function does not wait 12883259f8beSChris Mason * for IO completion of submitted IO. 12891da177e4SLinus Torvalds */ 12900e175a18SCurt Wohlgemuth void writeback_inodes_sb_nr(struct super_block *sb, 12910e175a18SCurt Wohlgemuth unsigned long nr, 12920e175a18SCurt Wohlgemuth enum wb_reason reason) 12931da177e4SLinus Torvalds { 129483ba7b07SChristoph Hellwig DECLARE_COMPLETION_ONSTACK(done); 129583ba7b07SChristoph Hellwig struct wb_writeback_work work = { 12963c4d7165SChristoph Hellwig .sb = sb, 12973c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_NONE, 12986e6938b6SWu Fengguang .tagged_writepages = 1, 129983ba7b07SChristoph Hellwig .done = &done, 13003259f8beSChris Mason .nr_pages = nr, 13010e175a18SCurt Wohlgemuth .reason = reason, 13023c4d7165SChristoph Hellwig }; 13030e3c9a22SJens Axboe 13046eedc701SJan Kara if (sb->s_bdi == &noop_backing_dev_info) 13056eedc701SJan Kara return; 1306cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 130783ba7b07SChristoph Hellwig bdi_queue_work(sb->s_bdi, &work); 130883ba7b07SChristoph Hellwig wait_for_completion(&done); 13091da177e4SLinus Torvalds } 13103259f8beSChris Mason EXPORT_SYMBOL(writeback_inodes_sb_nr); 13113259f8beSChris Mason 13123259f8beSChris Mason /** 13133259f8beSChris Mason * writeback_inodes_sb - writeback dirty inodes from given super_block 13143259f8beSChris Mason * @sb: the superblock 1315786228abSMarcos Paulo de Souza * @reason: reason why some writeback work was initiated 13163259f8beSChris Mason * 13173259f8beSChris Mason * Start writeback on some inodes on this super_block. No guarantees are made 13183259f8beSChris Mason * on how many (if any) will be written, and this function does not wait 13193259f8beSChris Mason * for IO completion of submitted IO. 13203259f8beSChris Mason */ 13210e175a18SCurt Wohlgemuth void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) 13223259f8beSChris Mason { 13230e175a18SCurt Wohlgemuth return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); 13243259f8beSChris Mason } 1325d8a8559cSJens Axboe EXPORT_SYMBOL(writeback_inodes_sb); 1326d8a8559cSJens Axboe 1327d8a8559cSJens Axboe /** 132810ee27a0SMiao Xie * try_to_writeback_inodes_sb_nr - try to start writeback if none underway 13293259f8beSChris Mason * @sb: the superblock 13303259f8beSChris Mason * @nr: the number of pages to write 133110ee27a0SMiao Xie * @reason: the reason of writeback 13323259f8beSChris Mason * 133310ee27a0SMiao Xie * Invoke writeback_inodes_sb_nr if no writeback is currently underway. 13343259f8beSChris Mason * Returns 1 if writeback was started, 0 if not. 13353259f8beSChris Mason */ 133610ee27a0SMiao Xie int try_to_writeback_inodes_sb_nr(struct super_block *sb, 13370e175a18SCurt Wohlgemuth unsigned long nr, 13380e175a18SCurt Wohlgemuth enum wb_reason reason) 13393259f8beSChris Mason { 134010ee27a0SMiao Xie if (writeback_in_progress(sb->s_bdi)) 134110ee27a0SMiao Xie return 1; 134210ee27a0SMiao Xie 134310ee27a0SMiao Xie if (!down_read_trylock(&sb->s_umount)) 134410ee27a0SMiao Xie return 0; 134510ee27a0SMiao Xie 13460e175a18SCurt Wohlgemuth writeback_inodes_sb_nr(sb, nr, reason); 13473259f8beSChris Mason up_read(&sb->s_umount); 13483259f8beSChris Mason return 1; 13493259f8beSChris Mason } 135010ee27a0SMiao Xie EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr); 135110ee27a0SMiao Xie 135210ee27a0SMiao Xie /** 135310ee27a0SMiao Xie * try_to_writeback_inodes_sb - try to start writeback if none underway 135410ee27a0SMiao Xie * @sb: the superblock 135510ee27a0SMiao Xie * @reason: reason why some writeback work was initiated 135610ee27a0SMiao Xie * 135710ee27a0SMiao Xie * Implement by try_to_writeback_inodes_sb_nr() 135810ee27a0SMiao Xie * Returns 1 if writeback was started, 0 if not. 135910ee27a0SMiao Xie */ 136010ee27a0SMiao Xie int try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) 136110ee27a0SMiao Xie { 136210ee27a0SMiao Xie return try_to_writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); 136310ee27a0SMiao Xie } 136410ee27a0SMiao Xie EXPORT_SYMBOL(try_to_writeback_inodes_sb); 13653259f8beSChris Mason 13663259f8beSChris Mason /** 1367d8a8559cSJens Axboe * sync_inodes_sb - sync sb inode pages 1368d8a8559cSJens Axboe * @sb: the superblock 1369d8a8559cSJens Axboe * 1370d8a8559cSJens Axboe * This function writes and waits on any dirty inode belonging to this 13710dc83bd3SJan Kara * super_block. 1372d8a8559cSJens Axboe */ 13730dc83bd3SJan Kara void sync_inodes_sb(struct super_block *sb) 1374d8a8559cSJens Axboe { 137583ba7b07SChristoph Hellwig DECLARE_COMPLETION_ONSTACK(done); 137683ba7b07SChristoph Hellwig struct wb_writeback_work work = { 13773c4d7165SChristoph Hellwig .sb = sb, 13783c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_ALL, 13793c4d7165SChristoph Hellwig .nr_pages = LONG_MAX, 13803c4d7165SChristoph Hellwig .range_cyclic = 0, 138183ba7b07SChristoph Hellwig .done = &done, 13820e175a18SCurt Wohlgemuth .reason = WB_REASON_SYNC, 13837747bd4bSDave Chinner .for_sync = 1, 13843c4d7165SChristoph Hellwig }; 13853c4d7165SChristoph Hellwig 13866eedc701SJan Kara /* Nothing to do? */ 13876eedc701SJan Kara if (sb->s_bdi == &noop_backing_dev_info) 13886eedc701SJan Kara return; 1389cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 1390cf37e972SChristoph Hellwig 139183ba7b07SChristoph Hellwig bdi_queue_work(sb->s_bdi, &work); 139283ba7b07SChristoph Hellwig wait_for_completion(&done); 139383ba7b07SChristoph Hellwig 1394b6e51316SJens Axboe wait_sb_inodes(sb); 1395d8a8559cSJens Axboe } 1396d8a8559cSJens Axboe EXPORT_SYMBOL(sync_inodes_sb); 13971da177e4SLinus Torvalds 13981da177e4SLinus Torvalds /** 13991da177e4SLinus Torvalds * write_inode_now - write an inode to disk 14001da177e4SLinus Torvalds * @inode: inode to write to disk 14011da177e4SLinus Torvalds * @sync: whether the write should be synchronous or not 14021da177e4SLinus Torvalds * 14037f04c26dSAndrea Arcangeli * This function commits an inode to disk immediately if it is dirty. This is 14047f04c26dSAndrea Arcangeli * primarily needed by knfsd. 14057f04c26dSAndrea Arcangeli * 14067f04c26dSAndrea Arcangeli * The caller must either have a ref on the inode or must have set I_WILL_FREE. 14071da177e4SLinus Torvalds */ 14081da177e4SLinus Torvalds int write_inode_now(struct inode *inode, int sync) 14091da177e4SLinus Torvalds { 1410f758eeabSChristoph Hellwig struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; 14111da177e4SLinus Torvalds struct writeback_control wbc = { 14121da177e4SLinus Torvalds .nr_to_write = LONG_MAX, 141318914b18SMike Galbraith .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, 1414111ebb6eSOGAWA Hirofumi .range_start = 0, 1415111ebb6eSOGAWA Hirofumi .range_end = LLONG_MAX, 14161da177e4SLinus Torvalds }; 14171da177e4SLinus Torvalds 14181da177e4SLinus Torvalds if (!mapping_cap_writeback_dirty(inode->i_mapping)) 141949364ce2SAndrew Morton wbc.nr_to_write = 0; 14201da177e4SLinus Torvalds 14211da177e4SLinus Torvalds might_sleep(); 14224f8ad655SJan Kara return writeback_single_inode(inode, wb, &wbc); 14231da177e4SLinus Torvalds } 14241da177e4SLinus Torvalds EXPORT_SYMBOL(write_inode_now); 14251da177e4SLinus Torvalds 14261da177e4SLinus Torvalds /** 14271da177e4SLinus Torvalds * sync_inode - write an inode and its pages to disk. 14281da177e4SLinus Torvalds * @inode: the inode to sync 14291da177e4SLinus Torvalds * @wbc: controls the writeback mode 14301da177e4SLinus Torvalds * 14311da177e4SLinus Torvalds * sync_inode() will write an inode and its pages to disk. It will also 14321da177e4SLinus Torvalds * correctly update the inode on its superblock's dirty inode lists and will 14331da177e4SLinus Torvalds * update inode->i_state. 14341da177e4SLinus Torvalds * 14351da177e4SLinus Torvalds * The caller must have a ref on the inode. 14361da177e4SLinus Torvalds */ 14371da177e4SLinus Torvalds int sync_inode(struct inode *inode, struct writeback_control *wbc) 14381da177e4SLinus Torvalds { 14394f8ad655SJan Kara return writeback_single_inode(inode, &inode_to_bdi(inode)->wb, wbc); 14401da177e4SLinus Torvalds } 14411da177e4SLinus Torvalds EXPORT_SYMBOL(sync_inode); 1442c3765016SChristoph Hellwig 1443c3765016SChristoph Hellwig /** 1444c691b9d9SAndrew Morton * sync_inode_metadata - write an inode to disk 1445c3765016SChristoph Hellwig * @inode: the inode to sync 1446c3765016SChristoph Hellwig * @wait: wait for I/O to complete. 1447c3765016SChristoph Hellwig * 1448c691b9d9SAndrew Morton * Write an inode to disk and adjust its dirty state after completion. 1449c3765016SChristoph Hellwig * 1450c3765016SChristoph Hellwig * Note: only writes the actual inode, no associated data or other metadata. 1451c3765016SChristoph Hellwig */ 1452c3765016SChristoph Hellwig int sync_inode_metadata(struct inode *inode, int wait) 1453c3765016SChristoph Hellwig { 1454c3765016SChristoph Hellwig struct writeback_control wbc = { 1455c3765016SChristoph Hellwig .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE, 1456c3765016SChristoph Hellwig .nr_to_write = 0, /* metadata-only */ 1457c3765016SChristoph Hellwig }; 1458c3765016SChristoph Hellwig 1459c3765016SChristoph Hellwig return sync_inode(inode, &wbc); 1460c3765016SChristoph Hellwig } 1461c3765016SChristoph Hellwig EXPORT_SYMBOL(sync_inode_metadata); 1462