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 926585027aSJan Kara static void bdi_queue_work(struct backing_dev_info *bdi, 936585027aSJan Kara struct wb_writeback_work *work) 946585027aSJan Kara { 956585027aSJan Kara trace_writeback_queue(bdi, work); 966585027aSJan Kara 976585027aSJan Kara spin_lock_bh(&bdi->wb_lock); 986585027aSJan Kara list_add_tail(&work->list, &bdi->work_list); 996467716aSArtem Bityutskiy spin_unlock_bh(&bdi->wb_lock); 100839a8e86STejun Heo 101839a8e86STejun Heo mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0); 10203ba3782SJens Axboe } 1031da177e4SLinus Torvalds 10483ba7b07SChristoph Hellwig static void 10583ba7b07SChristoph Hellwig __bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, 1060e175a18SCurt Wohlgemuth bool range_cyclic, enum wb_reason reason) 1071da177e4SLinus Torvalds { 10883ba7b07SChristoph Hellwig struct wb_writeback_work *work; 10903ba3782SJens Axboe 110bcddc3f0SJens Axboe /* 111bcddc3f0SJens Axboe * This is WB_SYNC_NONE writeback, so if allocation fails just 112bcddc3f0SJens Axboe * wakeup the thread for old dirty data writeback 113bcddc3f0SJens Axboe */ 11483ba7b07SChristoph Hellwig work = kzalloc(sizeof(*work), GFP_ATOMIC); 11583ba7b07SChristoph Hellwig if (!work) { 116455b2864SDave Chinner trace_writeback_nowork(bdi); 117839a8e86STejun Heo mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0); 11883ba7b07SChristoph Hellwig return; 11983ba7b07SChristoph Hellwig } 12083ba7b07SChristoph Hellwig 12183ba7b07SChristoph Hellwig work->sync_mode = WB_SYNC_NONE; 12283ba7b07SChristoph Hellwig work->nr_pages = nr_pages; 12383ba7b07SChristoph Hellwig work->range_cyclic = range_cyclic; 1240e175a18SCurt Wohlgemuth work->reason = reason; 12583ba7b07SChristoph Hellwig 126f11fcae8SJens Axboe bdi_queue_work(bdi, work); 12703ba3782SJens Axboe } 128b6e51316SJens Axboe 129b6e51316SJens Axboe /** 130b6e51316SJens Axboe * bdi_start_writeback - start writeback 131b6e51316SJens Axboe * @bdi: the backing device to write from 132b6e51316SJens Axboe * @nr_pages: the number of pages to write 133786228abSMarcos Paulo de Souza * @reason: reason why some writeback work was initiated 134b6e51316SJens Axboe * 135b6e51316SJens Axboe * Description: 136b6e51316SJens Axboe * This does WB_SYNC_NONE opportunistic writeback. The IO is only 13725985edcSLucas De Marchi * started when this function returns, we make no guarantees on 1380e3c9a22SJens Axboe * completion. Caller need not hold sb s_umount semaphore. 139b6e51316SJens Axboe * 140b6e51316SJens Axboe */ 1410e175a18SCurt Wohlgemuth void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, 1420e175a18SCurt Wohlgemuth enum wb_reason reason) 143b6e51316SJens Axboe { 1440e175a18SCurt Wohlgemuth __bdi_start_writeback(bdi, nr_pages, true, reason); 145d3ddec76SWu Fengguang } 146d3ddec76SWu Fengguang 147c5444198SChristoph Hellwig /** 148c5444198SChristoph Hellwig * bdi_start_background_writeback - start background writeback 149c5444198SChristoph Hellwig * @bdi: the backing device to write from 150c5444198SChristoph Hellwig * 151c5444198SChristoph Hellwig * Description: 1526585027aSJan Kara * This makes sure WB_SYNC_NONE background writeback happens. When 1536585027aSJan Kara * this function returns, it is only guaranteed that for given BDI 1546585027aSJan Kara * some IO is happening if we are over background dirty threshold. 1556585027aSJan Kara * Caller need not hold sb s_umount semaphore. 156c5444198SChristoph Hellwig */ 157c5444198SChristoph Hellwig void bdi_start_background_writeback(struct backing_dev_info *bdi) 158c5444198SChristoph Hellwig { 1596585027aSJan Kara /* 1606585027aSJan Kara * We just wake up the flusher thread. It will perform background 1616585027aSJan Kara * writeback as soon as there is no other work to do. 1626585027aSJan Kara */ 16371927e84SWu Fengguang trace_writeback_wake_background(bdi); 164839a8e86STejun Heo mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0); 1651da177e4SLinus Torvalds } 1661da177e4SLinus Torvalds 1671da177e4SLinus Torvalds /* 168a66979abSDave Chinner * Remove the inode from the writeback list it is on. 169a66979abSDave Chinner */ 170a66979abSDave Chinner void inode_wb_list_del(struct inode *inode) 171a66979abSDave Chinner { 172f758eeabSChristoph Hellwig struct backing_dev_info *bdi = inode_to_bdi(inode); 173a66979abSDave Chinner 174f758eeabSChristoph Hellwig spin_lock(&bdi->wb.list_lock); 175f758eeabSChristoph Hellwig list_del_init(&inode->i_wb_list); 176f758eeabSChristoph Hellwig spin_unlock(&bdi->wb.list_lock); 177f758eeabSChristoph Hellwig } 178a66979abSDave Chinner 179a66979abSDave Chinner /* 1806610a0bcSAndrew Morton * Redirty an inode: set its when-it-was dirtied timestamp and move it to the 1816610a0bcSAndrew Morton * furthest end of its superblock's dirty-inode list. 1826610a0bcSAndrew Morton * 1836610a0bcSAndrew Morton * Before stamping the inode's ->dirtied_when, we check to see whether it is 18466f3b8e2SJens Axboe * already the most-recently-dirtied inode on the b_dirty list. If that is 1856610a0bcSAndrew Morton * the case then the inode must have been redirtied while it was being written 1866610a0bcSAndrew Morton * out and we don't reset its dirtied_when. 1876610a0bcSAndrew Morton */ 188f758eeabSChristoph Hellwig static void redirty_tail(struct inode *inode, struct bdi_writeback *wb) 1896610a0bcSAndrew Morton { 190f758eeabSChristoph Hellwig assert_spin_locked(&wb->list_lock); 19103ba3782SJens Axboe if (!list_empty(&wb->b_dirty)) { 19266f3b8e2SJens Axboe struct inode *tail; 1936610a0bcSAndrew Morton 1947ccf19a8SNick Piggin tail = wb_inode(wb->b_dirty.next); 19566f3b8e2SJens Axboe if (time_before(inode->dirtied_when, tail->dirtied_when)) 1966610a0bcSAndrew Morton inode->dirtied_when = jiffies; 1976610a0bcSAndrew Morton } 1987ccf19a8SNick Piggin list_move(&inode->i_wb_list, &wb->b_dirty); 1996610a0bcSAndrew Morton } 2006610a0bcSAndrew Morton 2016610a0bcSAndrew Morton /* 20266f3b8e2SJens Axboe * requeue inode for re-scanning after bdi->b_io list is exhausted. 203c986d1e2SAndrew Morton */ 204f758eeabSChristoph Hellwig static void requeue_io(struct inode *inode, struct bdi_writeback *wb) 205c986d1e2SAndrew Morton { 206f758eeabSChristoph Hellwig assert_spin_locked(&wb->list_lock); 2077ccf19a8SNick Piggin list_move(&inode->i_wb_list, &wb->b_more_io); 208c986d1e2SAndrew Morton } 209c986d1e2SAndrew Morton 2101c0eeaf5SJoern Engel static void inode_sync_complete(struct inode *inode) 2111c0eeaf5SJoern Engel { 212365b94aeSJan Kara inode->i_state &= ~I_SYNC; 2134eff96ddSJan Kara /* If inode is clean an unused, put it into LRU now... */ 2144eff96ddSJan Kara inode_add_lru(inode); 215365b94aeSJan Kara /* Waiters must see I_SYNC cleared before being woken up */ 2161c0eeaf5SJoern Engel smp_mb(); 2171c0eeaf5SJoern Engel wake_up_bit(&inode->i_state, __I_SYNC); 2181c0eeaf5SJoern Engel } 2191c0eeaf5SJoern Engel 220d2caa3c5SJeff Layton static bool inode_dirtied_after(struct inode *inode, unsigned long t) 221d2caa3c5SJeff Layton { 222d2caa3c5SJeff Layton bool ret = time_after(inode->dirtied_when, t); 223d2caa3c5SJeff Layton #ifndef CONFIG_64BIT 224d2caa3c5SJeff Layton /* 225d2caa3c5SJeff Layton * For inodes being constantly redirtied, dirtied_when can get stuck. 226d2caa3c5SJeff Layton * It _appears_ to be in the future, but is actually in distant past. 227d2caa3c5SJeff Layton * This test is necessary to prevent such wrapped-around relative times 2285b0830cbSJens Axboe * from permanently stopping the whole bdi writeback. 229d2caa3c5SJeff Layton */ 230d2caa3c5SJeff Layton ret = ret && time_before_eq(inode->dirtied_when, jiffies); 231d2caa3c5SJeff Layton #endif 232d2caa3c5SJeff Layton return ret; 233d2caa3c5SJeff Layton } 234d2caa3c5SJeff Layton 235c986d1e2SAndrew Morton /* 2360e2f2b23SWang Sheng-Hui * Move expired (dirtied before work->older_than_this) dirty inodes from 237697e6fedSJan Kara * @delaying_queue to @dispatch_queue. 2382c136579SFengguang Wu */ 239e84d0a4fSWu Fengguang static int move_expired_inodes(struct list_head *delaying_queue, 2402c136579SFengguang Wu struct list_head *dispatch_queue, 241ad4e38ddSCurt Wohlgemuth struct wb_writeback_work *work) 2422c136579SFengguang Wu { 2435c03449dSShaohua Li LIST_HEAD(tmp); 2445c03449dSShaohua Li struct list_head *pos, *node; 245cf137307SJens Axboe struct super_block *sb = NULL; 2465c03449dSShaohua Li struct inode *inode; 247cf137307SJens Axboe int do_sb_sort = 0; 248e84d0a4fSWu Fengguang int moved = 0; 2495c03449dSShaohua Li 2502c136579SFengguang Wu while (!list_empty(delaying_queue)) { 2517ccf19a8SNick Piggin inode = wb_inode(delaying_queue->prev); 2520dc83bd3SJan Kara if (work->older_than_this && 2530dc83bd3SJan Kara inode_dirtied_after(inode, *work->older_than_this)) 2542c136579SFengguang Wu break; 255a8855990SJan Kara list_move(&inode->i_wb_list, &tmp); 256a8855990SJan Kara moved++; 257a8855990SJan Kara if (sb_is_blkdev_sb(inode->i_sb)) 258a8855990SJan Kara continue; 259cf137307SJens Axboe if (sb && sb != inode->i_sb) 260cf137307SJens Axboe do_sb_sort = 1; 261cf137307SJens Axboe sb = inode->i_sb; 2625c03449dSShaohua Li } 2635c03449dSShaohua Li 264cf137307SJens Axboe /* just one sb in list, splice to dispatch_queue and we're done */ 265cf137307SJens Axboe if (!do_sb_sort) { 266cf137307SJens Axboe list_splice(&tmp, dispatch_queue); 267e84d0a4fSWu Fengguang goto out; 268cf137307SJens Axboe } 269cf137307SJens Axboe 2705c03449dSShaohua Li /* Move inodes from one superblock together */ 2715c03449dSShaohua Li while (!list_empty(&tmp)) { 2727ccf19a8SNick Piggin sb = wb_inode(tmp.prev)->i_sb; 2735c03449dSShaohua Li list_for_each_prev_safe(pos, node, &tmp) { 2747ccf19a8SNick Piggin inode = wb_inode(pos); 2755c03449dSShaohua Li if (inode->i_sb == sb) 2767ccf19a8SNick Piggin list_move(&inode->i_wb_list, dispatch_queue); 2772c136579SFengguang Wu } 2782c136579SFengguang Wu } 279e84d0a4fSWu Fengguang out: 280e84d0a4fSWu Fengguang return moved; 2815c03449dSShaohua Li } 2822c136579SFengguang Wu 2832c136579SFengguang Wu /* 2842c136579SFengguang Wu * Queue all expired dirty inodes for io, eldest first. 2854ea879b9SWu Fengguang * Before 2864ea879b9SWu Fengguang * newly dirtied b_dirty b_io b_more_io 2874ea879b9SWu Fengguang * =============> gf edc BA 2884ea879b9SWu Fengguang * After 2894ea879b9SWu Fengguang * newly dirtied b_dirty b_io b_more_io 2904ea879b9SWu Fengguang * =============> g fBAedc 2914ea879b9SWu Fengguang * | 2924ea879b9SWu Fengguang * +--> dequeue for IO 2932c136579SFengguang Wu */ 294ad4e38ddSCurt Wohlgemuth static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work) 2952c136579SFengguang Wu { 296e84d0a4fSWu Fengguang int moved; 297f758eeabSChristoph Hellwig assert_spin_locked(&wb->list_lock); 2984ea879b9SWu Fengguang list_splice_init(&wb->b_more_io, &wb->b_io); 299ad4e38ddSCurt Wohlgemuth moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, work); 300ad4e38ddSCurt Wohlgemuth trace_writeback_queue_io(wb, work, moved); 30166f3b8e2SJens Axboe } 30266f3b8e2SJens Axboe 303a9185b41SChristoph Hellwig static int write_inode(struct inode *inode, struct writeback_control *wbc) 30466f3b8e2SJens Axboe { 3059fb0a7daSTejun Heo int ret; 3069fb0a7daSTejun Heo 3079fb0a7daSTejun Heo if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) { 3089fb0a7daSTejun Heo trace_writeback_write_inode_start(inode, wbc); 3099fb0a7daSTejun Heo ret = inode->i_sb->s_op->write_inode(inode, wbc); 3109fb0a7daSTejun Heo trace_writeback_write_inode(inode, wbc); 3119fb0a7daSTejun Heo return ret; 3129fb0a7daSTejun Heo } 31303ba3782SJens Axboe return 0; 31466f3b8e2SJens Axboe } 31508d8e974SFengguang Wu 3162c136579SFengguang Wu /* 317169ebd90SJan Kara * Wait for writeback on an inode to complete. Called with i_lock held. 318169ebd90SJan Kara * Caller must make sure inode cannot go away when we drop i_lock. 31901c03194SChristoph Hellwig */ 320169ebd90SJan Kara static void __inode_wait_for_writeback(struct inode *inode) 321169ebd90SJan Kara __releases(inode->i_lock) 322169ebd90SJan Kara __acquires(inode->i_lock) 32301c03194SChristoph Hellwig { 32401c03194SChristoph Hellwig DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); 32501c03194SChristoph Hellwig wait_queue_head_t *wqh; 32601c03194SChristoph Hellwig 32701c03194SChristoph Hellwig wqh = bit_waitqueue(&inode->i_state, __I_SYNC); 32858a9d3d8SRichard Kennedy while (inode->i_state & I_SYNC) { 329250df6edSDave Chinner spin_unlock(&inode->i_lock); 33001c03194SChristoph Hellwig __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE); 331250df6edSDave Chinner spin_lock(&inode->i_lock); 33258a9d3d8SRichard Kennedy } 33301c03194SChristoph Hellwig } 33401c03194SChristoph Hellwig 33501c03194SChristoph Hellwig /* 336169ebd90SJan Kara * Wait for writeback on an inode to complete. Caller must have inode pinned. 337169ebd90SJan Kara */ 338169ebd90SJan Kara void inode_wait_for_writeback(struct inode *inode) 339169ebd90SJan Kara { 340169ebd90SJan Kara spin_lock(&inode->i_lock); 341169ebd90SJan Kara __inode_wait_for_writeback(inode); 342169ebd90SJan Kara spin_unlock(&inode->i_lock); 343169ebd90SJan Kara } 344169ebd90SJan Kara 345169ebd90SJan Kara /* 346169ebd90SJan Kara * Sleep until I_SYNC is cleared. This function must be called with i_lock 347169ebd90SJan Kara * held and drops it. It is aimed for callers not holding any inode reference 348169ebd90SJan Kara * so once i_lock is dropped, inode can go away. 349169ebd90SJan Kara */ 350169ebd90SJan Kara static void inode_sleep_on_writeback(struct inode *inode) 351169ebd90SJan Kara __releases(inode->i_lock) 352169ebd90SJan Kara { 353169ebd90SJan Kara DEFINE_WAIT(wait); 354169ebd90SJan Kara wait_queue_head_t *wqh = bit_waitqueue(&inode->i_state, __I_SYNC); 355169ebd90SJan Kara int sleep; 356169ebd90SJan Kara 357169ebd90SJan Kara prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); 358169ebd90SJan Kara sleep = inode->i_state & I_SYNC; 359169ebd90SJan Kara spin_unlock(&inode->i_lock); 360169ebd90SJan Kara if (sleep) 361169ebd90SJan Kara schedule(); 362169ebd90SJan Kara finish_wait(wqh, &wait); 363169ebd90SJan Kara } 364169ebd90SJan Kara 365169ebd90SJan Kara /* 366ccb26b5aSJan Kara * Find proper writeback list for the inode depending on its current state and 367ccb26b5aSJan Kara * possibly also change of its state while we were doing writeback. Here we 368ccb26b5aSJan Kara * handle things such as livelock prevention or fairness of writeback among 369ccb26b5aSJan Kara * inodes. This function can be called only by flusher thread - noone else 370ccb26b5aSJan Kara * processes all inodes in writeback lists and requeueing inodes behind flusher 371ccb26b5aSJan Kara * thread's back can have unexpected consequences. 372ccb26b5aSJan Kara */ 373ccb26b5aSJan Kara static void requeue_inode(struct inode *inode, struct bdi_writeback *wb, 374ccb26b5aSJan Kara struct writeback_control *wbc) 375ccb26b5aSJan Kara { 376ccb26b5aSJan Kara if (inode->i_state & I_FREEING) 377ccb26b5aSJan Kara return; 378ccb26b5aSJan Kara 379ccb26b5aSJan Kara /* 380ccb26b5aSJan Kara * Sync livelock prevention. Each inode is tagged and synced in one 381ccb26b5aSJan Kara * shot. If still dirty, it will be redirty_tail()'ed below. Update 382ccb26b5aSJan Kara * the dirty time to prevent enqueue and sync it again. 383ccb26b5aSJan Kara */ 384ccb26b5aSJan Kara if ((inode->i_state & I_DIRTY) && 385ccb26b5aSJan Kara (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)) 386ccb26b5aSJan Kara inode->dirtied_when = jiffies; 387ccb26b5aSJan Kara 3884f8ad655SJan Kara if (wbc->pages_skipped) { 3894f8ad655SJan Kara /* 3904f8ad655SJan Kara * writeback is not making progress due to locked 3914f8ad655SJan Kara * buffers. Skip this inode for now. 3924f8ad655SJan Kara */ 3934f8ad655SJan Kara redirty_tail(inode, wb); 3944f8ad655SJan Kara return; 3954f8ad655SJan Kara } 3964f8ad655SJan Kara 397ccb26b5aSJan Kara if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) { 398ccb26b5aSJan Kara /* 399ccb26b5aSJan Kara * We didn't write back all the pages. nfs_writepages() 400ccb26b5aSJan Kara * sometimes bales out without doing anything. 401ccb26b5aSJan Kara */ 402ccb26b5aSJan Kara if (wbc->nr_to_write <= 0) { 403ccb26b5aSJan Kara /* Slice used up. Queue for next turn. */ 404ccb26b5aSJan Kara requeue_io(inode, wb); 405ccb26b5aSJan Kara } else { 406ccb26b5aSJan Kara /* 407ccb26b5aSJan Kara * Writeback blocked by something other than 408ccb26b5aSJan Kara * congestion. Delay the inode for some time to 409ccb26b5aSJan Kara * avoid spinning on the CPU (100% iowait) 410ccb26b5aSJan Kara * retrying writeback of the dirty page/inode 411ccb26b5aSJan Kara * that cannot be performed immediately. 412ccb26b5aSJan Kara */ 413ccb26b5aSJan Kara redirty_tail(inode, wb); 414ccb26b5aSJan Kara } 415ccb26b5aSJan Kara } else if (inode->i_state & I_DIRTY) { 416ccb26b5aSJan Kara /* 417ccb26b5aSJan Kara * Filesystems can dirty the inode during writeback operations, 418ccb26b5aSJan Kara * such as delayed allocation during submission or metadata 419ccb26b5aSJan Kara * updates after data IO completion. 420ccb26b5aSJan Kara */ 421ccb26b5aSJan Kara redirty_tail(inode, wb); 422ccb26b5aSJan Kara } else { 423ccb26b5aSJan Kara /* The inode is clean. Remove from writeback lists. */ 424ccb26b5aSJan Kara list_del_init(&inode->i_wb_list); 425ccb26b5aSJan Kara } 426ccb26b5aSJan Kara } 427ccb26b5aSJan Kara 428ccb26b5aSJan Kara /* 4294f8ad655SJan Kara * Write out an inode and its dirty pages. Do not update the writeback list 4304f8ad655SJan Kara * linkage. That is left to the caller. The caller is also responsible for 4314f8ad655SJan Kara * setting I_SYNC flag and calling inode_sync_complete() to clear it. 4321da177e4SLinus Torvalds */ 4331da177e4SLinus Torvalds static int 434cd8ed2a4SYan Hong __writeback_single_inode(struct inode *inode, struct writeback_control *wbc) 4351da177e4SLinus Torvalds { 4361da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 437251d6a47SWu Fengguang long nr_to_write = wbc->nr_to_write; 43801c03194SChristoph Hellwig unsigned dirty; 4391da177e4SLinus Torvalds int ret; 4401da177e4SLinus Torvalds 4414f8ad655SJan Kara WARN_ON(!(inode->i_state & I_SYNC)); 4421da177e4SLinus Torvalds 4439fb0a7daSTejun Heo trace_writeback_single_inode_start(inode, wbc, nr_to_write); 4449fb0a7daSTejun Heo 4451da177e4SLinus Torvalds ret = do_writepages(mapping, wbc); 4461da177e4SLinus Torvalds 44726821ed4SChristoph Hellwig /* 44826821ed4SChristoph Hellwig * Make sure to wait on the data before writing out the metadata. 44926821ed4SChristoph Hellwig * This is important for filesystems that modify metadata on data 4507747bd4bSDave Chinner * I/O completion. We don't do it for sync(2) writeback because it has a 4517747bd4bSDave Chinner * separate, external IO completion path and ->sync_fs for guaranteeing 4527747bd4bSDave Chinner * inode metadata is written back correctly. 45326821ed4SChristoph Hellwig */ 4547747bd4bSDave Chinner if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) { 45526821ed4SChristoph Hellwig int err = filemap_fdatawait(mapping); 4561da177e4SLinus Torvalds if (ret == 0) 4571da177e4SLinus Torvalds ret = err; 4581da177e4SLinus Torvalds } 4591da177e4SLinus Torvalds 4605547e8aaSDmitry Monakhov /* 4615547e8aaSDmitry Monakhov * Some filesystems may redirty the inode during the writeback 4625547e8aaSDmitry Monakhov * due to delalloc, clear dirty metadata flags right before 4635547e8aaSDmitry Monakhov * write_inode() 4645547e8aaSDmitry Monakhov */ 465250df6edSDave Chinner spin_lock(&inode->i_lock); 4666290be1cSJan Kara /* Clear I_DIRTY_PAGES if we've written out all dirty pages */ 4676290be1cSJan Kara if (!mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) 4686290be1cSJan Kara inode->i_state &= ~I_DIRTY_PAGES; 4695547e8aaSDmitry Monakhov dirty = inode->i_state & I_DIRTY; 4705547e8aaSDmitry Monakhov inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC); 471250df6edSDave Chinner spin_unlock(&inode->i_lock); 47226821ed4SChristoph Hellwig /* Don't write the inode if only I_DIRTY_PAGES was set */ 47326821ed4SChristoph Hellwig if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { 474a9185b41SChristoph Hellwig int err = write_inode(inode, wbc); 4751da177e4SLinus Torvalds if (ret == 0) 4761da177e4SLinus Torvalds ret = err; 4771da177e4SLinus Torvalds } 4784f8ad655SJan Kara trace_writeback_single_inode(inode, wbc, nr_to_write); 4794f8ad655SJan Kara return ret; 4804f8ad655SJan Kara } 4814f8ad655SJan Kara 4824f8ad655SJan Kara /* 4834f8ad655SJan Kara * Write out an inode's dirty pages. Either the caller has an active reference 4844f8ad655SJan Kara * on the inode or the inode has I_WILL_FREE set. 4854f8ad655SJan Kara * 4864f8ad655SJan Kara * This function is designed to be called for writing back one inode which 4874f8ad655SJan Kara * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode() 4884f8ad655SJan Kara * and does more profound writeback list handling in writeback_sb_inodes(). 4894f8ad655SJan Kara */ 4904f8ad655SJan Kara static int 4914f8ad655SJan Kara writeback_single_inode(struct inode *inode, struct bdi_writeback *wb, 4924f8ad655SJan Kara struct writeback_control *wbc) 4934f8ad655SJan Kara { 4944f8ad655SJan Kara int ret = 0; 4954f8ad655SJan Kara 4964f8ad655SJan Kara spin_lock(&inode->i_lock); 4974f8ad655SJan Kara if (!atomic_read(&inode->i_count)) 4984f8ad655SJan Kara WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); 4994f8ad655SJan Kara else 5004f8ad655SJan Kara WARN_ON(inode->i_state & I_WILL_FREE); 5014f8ad655SJan Kara 5024f8ad655SJan Kara if (inode->i_state & I_SYNC) { 5034f8ad655SJan Kara if (wbc->sync_mode != WB_SYNC_ALL) 5044f8ad655SJan Kara goto out; 5054f8ad655SJan Kara /* 506169ebd90SJan Kara * It's a data-integrity sync. We must wait. Since callers hold 507169ebd90SJan Kara * inode reference or inode has I_WILL_FREE set, it cannot go 508169ebd90SJan Kara * away under us. 5094f8ad655SJan Kara */ 510169ebd90SJan Kara __inode_wait_for_writeback(inode); 5114f8ad655SJan Kara } 5124f8ad655SJan Kara WARN_ON(inode->i_state & I_SYNC); 5134f8ad655SJan Kara /* 514f9b0e058SJan Kara * Skip inode if it is clean and we have no outstanding writeback in 515f9b0e058SJan Kara * WB_SYNC_ALL mode. We don't want to mess with writeback lists in this 516f9b0e058SJan Kara * function since flusher thread may be doing for example sync in 517f9b0e058SJan Kara * parallel and if we move the inode, it could get skipped. So here we 518f9b0e058SJan Kara * make sure inode is on some writeback list and leave it there unless 519f9b0e058SJan Kara * we have completely cleaned the inode. 5204f8ad655SJan Kara */ 521f9b0e058SJan Kara if (!(inode->i_state & I_DIRTY) && 522f9b0e058SJan Kara (wbc->sync_mode != WB_SYNC_ALL || 523f9b0e058SJan Kara !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK))) 5244f8ad655SJan Kara goto out; 5254f8ad655SJan Kara inode->i_state |= I_SYNC; 5264f8ad655SJan Kara spin_unlock(&inode->i_lock); 5274f8ad655SJan Kara 528cd8ed2a4SYan Hong ret = __writeback_single_inode(inode, wbc); 5291da177e4SLinus Torvalds 530f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 531250df6edSDave Chinner spin_lock(&inode->i_lock); 5324f8ad655SJan Kara /* 5334f8ad655SJan Kara * If inode is clean, remove it from writeback lists. Otherwise don't 5344f8ad655SJan Kara * touch it. See comment above for explanation. 5354f8ad655SJan Kara */ 5364f8ad655SJan Kara if (!(inode->i_state & I_DIRTY)) 5374f8ad655SJan Kara list_del_init(&inode->i_wb_list); 5384f8ad655SJan Kara spin_unlock(&wb->list_lock); 5391c0eeaf5SJoern Engel inode_sync_complete(inode); 5404f8ad655SJan Kara out: 5414f8ad655SJan Kara spin_unlock(&inode->i_lock); 5421da177e4SLinus Torvalds return ret; 5431da177e4SLinus Torvalds } 5441da177e4SLinus Torvalds 5451a12d8bdSWu Fengguang static long writeback_chunk_size(struct backing_dev_info *bdi, 5461a12d8bdSWu Fengguang struct wb_writeback_work *work) 547d46db3d5SWu Fengguang { 548d46db3d5SWu Fengguang long pages; 549d46db3d5SWu Fengguang 550d46db3d5SWu Fengguang /* 551d46db3d5SWu Fengguang * WB_SYNC_ALL mode does livelock avoidance by syncing dirty 552d46db3d5SWu Fengguang * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX 553d46db3d5SWu Fengguang * here avoids calling into writeback_inodes_wb() more than once. 554d46db3d5SWu Fengguang * 555d46db3d5SWu Fengguang * The intended call sequence for WB_SYNC_ALL writeback is: 556d46db3d5SWu Fengguang * 557d46db3d5SWu Fengguang * wb_writeback() 558d46db3d5SWu Fengguang * writeback_sb_inodes() <== called only once 559d46db3d5SWu Fengguang * write_cache_pages() <== called once for each inode 560d46db3d5SWu Fengguang * (quickly) tag currently dirty pages 561d46db3d5SWu Fengguang * (maybe slowly) sync all tagged pages 562d46db3d5SWu Fengguang */ 563d46db3d5SWu Fengguang if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages) 564d46db3d5SWu Fengguang pages = LONG_MAX; 5651a12d8bdSWu Fengguang else { 5661a12d8bdSWu Fengguang pages = min(bdi->avg_write_bandwidth / 2, 5671a12d8bdSWu Fengguang global_dirty_limit / DIRTY_SCOPE); 5681a12d8bdSWu Fengguang pages = min(pages, work->nr_pages); 5691a12d8bdSWu Fengguang pages = round_down(pages + MIN_WRITEBACK_PAGES, 5701a12d8bdSWu Fengguang MIN_WRITEBACK_PAGES); 5711a12d8bdSWu Fengguang } 572d46db3d5SWu Fengguang 573d46db3d5SWu Fengguang return pages; 574d46db3d5SWu Fengguang } 575d46db3d5SWu Fengguang 57603ba3782SJens Axboe /* 577f11c9c5cSEdward Shishkin * Write a portion of b_io inodes which belong to @sb. 578edadfb10SChristoph Hellwig * 579d46db3d5SWu Fengguang * Return the number of pages and/or inodes written. 580f11c9c5cSEdward Shishkin */ 581d46db3d5SWu Fengguang static long writeback_sb_inodes(struct super_block *sb, 582d46db3d5SWu Fengguang struct bdi_writeback *wb, 583d46db3d5SWu Fengguang struct wb_writeback_work *work) 58403ba3782SJens Axboe { 585d46db3d5SWu Fengguang struct writeback_control wbc = { 586d46db3d5SWu Fengguang .sync_mode = work->sync_mode, 587d46db3d5SWu Fengguang .tagged_writepages = work->tagged_writepages, 588d46db3d5SWu Fengguang .for_kupdate = work->for_kupdate, 589d46db3d5SWu Fengguang .for_background = work->for_background, 5907747bd4bSDave Chinner .for_sync = work->for_sync, 591d46db3d5SWu Fengguang .range_cyclic = work->range_cyclic, 592d46db3d5SWu Fengguang .range_start = 0, 593d46db3d5SWu Fengguang .range_end = LLONG_MAX, 594d46db3d5SWu Fengguang }; 595d46db3d5SWu Fengguang unsigned long start_time = jiffies; 596d46db3d5SWu Fengguang long write_chunk; 597d46db3d5SWu Fengguang long wrote = 0; /* count both pages and inodes */ 598d46db3d5SWu Fengguang 59903ba3782SJens Axboe while (!list_empty(&wb->b_io)) { 6007ccf19a8SNick Piggin struct inode *inode = wb_inode(wb->b_io.prev); 601edadfb10SChristoph Hellwig 602edadfb10SChristoph Hellwig if (inode->i_sb != sb) { 603d46db3d5SWu Fengguang if (work->sb) { 604edadfb10SChristoph Hellwig /* 605edadfb10SChristoph Hellwig * We only want to write back data for this 606edadfb10SChristoph Hellwig * superblock, move all inodes not belonging 607edadfb10SChristoph Hellwig * to it back onto the dirty list. 608edadfb10SChristoph Hellwig */ 609f758eeabSChristoph Hellwig redirty_tail(inode, wb); 61066f3b8e2SJens Axboe continue; 61166f3b8e2SJens Axboe } 612edadfb10SChristoph Hellwig 613edadfb10SChristoph Hellwig /* 614edadfb10SChristoph Hellwig * The inode belongs to a different superblock. 615edadfb10SChristoph Hellwig * Bounce back to the caller to unpin this and 616edadfb10SChristoph Hellwig * pin the next superblock. 617edadfb10SChristoph Hellwig */ 618d46db3d5SWu Fengguang break; 619edadfb10SChristoph Hellwig } 620edadfb10SChristoph Hellwig 6219843b76aSChristoph Hellwig /* 622331cbdeeSWanpeng Li * Don't bother with new inodes or inodes being freed, first 623331cbdeeSWanpeng Li * kind does not need periodic writeout yet, and for the latter 6249843b76aSChristoph Hellwig * kind writeout is handled by the freer. 6259843b76aSChristoph Hellwig */ 626250df6edSDave Chinner spin_lock(&inode->i_lock); 6279843b76aSChristoph Hellwig if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { 628250df6edSDave Chinner spin_unlock(&inode->i_lock); 629fcc5c222SWu Fengguang redirty_tail(inode, wb); 6307ef0d737SNick Piggin continue; 6317ef0d737SNick Piggin } 632cc1676d9SJan Kara if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) { 633cc1676d9SJan Kara /* 634cc1676d9SJan Kara * If this inode is locked for writeback and we are not 635cc1676d9SJan Kara * doing writeback-for-data-integrity, move it to 636cc1676d9SJan Kara * b_more_io so that writeback can proceed with the 637cc1676d9SJan Kara * other inodes on s_io. 638cc1676d9SJan Kara * 639cc1676d9SJan Kara * We'll have another go at writing back this inode 640cc1676d9SJan Kara * when we completed a full scan of b_io. 641cc1676d9SJan Kara */ 642cc1676d9SJan Kara spin_unlock(&inode->i_lock); 643cc1676d9SJan Kara requeue_io(inode, wb); 644cc1676d9SJan Kara trace_writeback_sb_inodes_requeue(inode); 645cc1676d9SJan Kara continue; 646cc1676d9SJan Kara } 647f0d07b7fSJan Kara spin_unlock(&wb->list_lock); 648f0d07b7fSJan Kara 6494f8ad655SJan Kara /* 6504f8ad655SJan Kara * We already requeued the inode if it had I_SYNC set and we 6514f8ad655SJan Kara * are doing WB_SYNC_NONE writeback. So this catches only the 6524f8ad655SJan Kara * WB_SYNC_ALL case. 6534f8ad655SJan Kara */ 654169ebd90SJan Kara if (inode->i_state & I_SYNC) { 655169ebd90SJan Kara /* Wait for I_SYNC. This function drops i_lock... */ 656169ebd90SJan Kara inode_sleep_on_writeback(inode); 657169ebd90SJan Kara /* Inode may be gone, start again */ 658ead188f9SJan Kara spin_lock(&wb->list_lock); 659169ebd90SJan Kara continue; 660169ebd90SJan Kara } 6614f8ad655SJan Kara inode->i_state |= I_SYNC; 6624f8ad655SJan Kara spin_unlock(&inode->i_lock); 663169ebd90SJan Kara 6641a12d8bdSWu Fengguang write_chunk = writeback_chunk_size(wb->bdi, work); 665d46db3d5SWu Fengguang wbc.nr_to_write = write_chunk; 666d46db3d5SWu Fengguang wbc.pages_skipped = 0; 667250df6edSDave Chinner 668169ebd90SJan Kara /* 669169ebd90SJan Kara * We use I_SYNC to pin the inode in memory. While it is set 670169ebd90SJan Kara * evict_inode() will wait so the inode cannot be freed. 671169ebd90SJan Kara */ 672cd8ed2a4SYan Hong __writeback_single_inode(inode, &wbc); 673d46db3d5SWu Fengguang 674d46db3d5SWu Fengguang work->nr_pages -= write_chunk - wbc.nr_to_write; 675d46db3d5SWu Fengguang wrote += write_chunk - wbc.nr_to_write; 6764f8ad655SJan Kara spin_lock(&wb->list_lock); 6774f8ad655SJan Kara spin_lock(&inode->i_lock); 678d46db3d5SWu Fengguang if (!(inode->i_state & I_DIRTY)) 679d46db3d5SWu Fengguang wrote++; 6804f8ad655SJan Kara requeue_inode(inode, wb, &wbc); 6814f8ad655SJan Kara inode_sync_complete(inode); 6820f1b1fd8SDave Chinner spin_unlock(&inode->i_lock); 683169ebd90SJan Kara cond_resched_lock(&wb->list_lock); 684d46db3d5SWu Fengguang /* 685d46db3d5SWu Fengguang * bail out to wb_writeback() often enough to check 686d46db3d5SWu Fengguang * background threshold and other termination conditions. 687d46db3d5SWu Fengguang */ 688d46db3d5SWu Fengguang if (wrote) { 689d46db3d5SWu Fengguang if (time_is_before_jiffies(start_time + HZ / 10UL)) 690d46db3d5SWu Fengguang break; 691d46db3d5SWu Fengguang if (work->nr_pages <= 0) 692d46db3d5SWu Fengguang break; 6931da177e4SLinus Torvalds } 6948bc3be27SFengguang Wu } 695d46db3d5SWu Fengguang return wrote; 696f11c9c5cSEdward Shishkin } 69738f21977SNick Piggin 698d46db3d5SWu Fengguang static long __writeback_inodes_wb(struct bdi_writeback *wb, 699d46db3d5SWu Fengguang struct wb_writeback_work *work) 700f11c9c5cSEdward Shishkin { 701d46db3d5SWu Fengguang unsigned long start_time = jiffies; 702d46db3d5SWu Fengguang long wrote = 0; 703f11c9c5cSEdward Shishkin 704f11c9c5cSEdward Shishkin while (!list_empty(&wb->b_io)) { 7057ccf19a8SNick Piggin struct inode *inode = wb_inode(wb->b_io.prev); 706f11c9c5cSEdward Shishkin struct super_block *sb = inode->i_sb; 707f11c9c5cSEdward Shishkin 70812ad3ab6SDave Chinner if (!grab_super_passive(sb)) { 7090e995816SWu Fengguang /* 7100e995816SWu Fengguang * grab_super_passive() may fail consistently due to 7110e995816SWu Fengguang * s_umount being grabbed by someone else. Don't use 7120e995816SWu Fengguang * requeue_io() to avoid busy retrying the inode/sb. 7130e995816SWu Fengguang */ 7140e995816SWu Fengguang redirty_tail(inode, wb); 715d19de7edSChristoph Hellwig continue; 716334132aeSChristoph Hellwig } 717d46db3d5SWu Fengguang wrote += writeback_sb_inodes(sb, wb, work); 718d19de7edSChristoph Hellwig drop_super(sb); 719f11c9c5cSEdward Shishkin 720d46db3d5SWu Fengguang /* refer to the same tests at the end of writeback_sb_inodes */ 721d46db3d5SWu Fengguang if (wrote) { 722d46db3d5SWu Fengguang if (time_is_before_jiffies(start_time + HZ / 10UL)) 723d46db3d5SWu Fengguang break; 724d46db3d5SWu Fengguang if (work->nr_pages <= 0) 725f11c9c5cSEdward Shishkin break; 726f11c9c5cSEdward Shishkin } 727d46db3d5SWu Fengguang } 72866f3b8e2SJens Axboe /* Leave any unwritten inodes on b_io */ 729d46db3d5SWu Fengguang return wrote; 73066f3b8e2SJens Axboe } 73166f3b8e2SJens Axboe 7327d9f073bSWanpeng Li static long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages, 7330e175a18SCurt Wohlgemuth enum wb_reason reason) 734edadfb10SChristoph Hellwig { 735d46db3d5SWu Fengguang struct wb_writeback_work work = { 736d46db3d5SWu Fengguang .nr_pages = nr_pages, 737d46db3d5SWu Fengguang .sync_mode = WB_SYNC_NONE, 738d46db3d5SWu Fengguang .range_cyclic = 1, 7390e175a18SCurt Wohlgemuth .reason = reason, 740d46db3d5SWu Fengguang }; 741edadfb10SChristoph Hellwig 742f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 743424b351fSWu Fengguang if (list_empty(&wb->b_io)) 744ad4e38ddSCurt Wohlgemuth queue_io(wb, &work); 745d46db3d5SWu Fengguang __writeback_inodes_wb(wb, &work); 746f758eeabSChristoph Hellwig spin_unlock(&wb->list_lock); 747edadfb10SChristoph Hellwig 748d46db3d5SWu Fengguang return nr_pages - work.nr_pages; 74966f3b8e2SJens Axboe } 75066f3b8e2SJens Axboe 751b00949aaSWu Fengguang static bool over_bground_thresh(struct backing_dev_info *bdi) 75203ba3782SJens Axboe { 75303ba3782SJens Axboe unsigned long background_thresh, dirty_thresh; 75403ba3782SJens Axboe 75516c4042fSWu Fengguang global_dirty_limits(&background_thresh, &dirty_thresh); 75603ba3782SJens Axboe 757b00949aaSWu Fengguang if (global_page_state(NR_FILE_DIRTY) + 758b00949aaSWu Fengguang global_page_state(NR_UNSTABLE_NFS) > background_thresh) 759b00949aaSWu Fengguang return true; 760b00949aaSWu Fengguang 761b00949aaSWu Fengguang if (bdi_stat(bdi, BDI_RECLAIMABLE) > 762b00949aaSWu Fengguang bdi_dirty_limit(bdi, background_thresh)) 763b00949aaSWu Fengguang return true; 764b00949aaSWu Fengguang 765b00949aaSWu Fengguang return false; 76603ba3782SJens Axboe } 76703ba3782SJens Axboe 76803ba3782SJens Axboe /* 769e98be2d5SWu Fengguang * Called under wb->list_lock. If there are multiple wb per bdi, 770e98be2d5SWu Fengguang * only the flusher working on the first wb should do it. 771e98be2d5SWu Fengguang */ 772e98be2d5SWu Fengguang static void wb_update_bandwidth(struct bdi_writeback *wb, 773e98be2d5SWu Fengguang unsigned long start_time) 774e98be2d5SWu Fengguang { 775af6a3113SWu Fengguang __bdi_update_bandwidth(wb->bdi, 0, 0, 0, 0, 0, start_time); 776e98be2d5SWu Fengguang } 777e98be2d5SWu Fengguang 778e98be2d5SWu Fengguang /* 77903ba3782SJens Axboe * Explicit flushing or periodic writeback of "old" data. 78003ba3782SJens Axboe * 78103ba3782SJens Axboe * Define "old": the first time one of an inode's pages is dirtied, we mark the 78203ba3782SJens Axboe * dirtying-time in the inode's address_space. So this periodic writeback code 78303ba3782SJens Axboe * just walks the superblock inode list, writing back any inodes which are 78403ba3782SJens Axboe * older than a specific point in time. 78503ba3782SJens Axboe * 78603ba3782SJens Axboe * Try to run once per dirty_writeback_interval. But if a writeback event 78703ba3782SJens Axboe * takes longer than a dirty_writeback_interval interval, then leave a 78803ba3782SJens Axboe * one-second gap. 78903ba3782SJens Axboe * 79003ba3782SJens Axboe * older_than_this takes precedence over nr_to_write. So we'll only write back 79103ba3782SJens Axboe * all dirty pages if they are all attached to "old" mappings. 79203ba3782SJens Axboe */ 793c4a77a6cSJens Axboe static long wb_writeback(struct bdi_writeback *wb, 79483ba7b07SChristoph Hellwig struct wb_writeback_work *work) 79503ba3782SJens Axboe { 796e98be2d5SWu Fengguang unsigned long wb_start = jiffies; 797d46db3d5SWu Fengguang long nr_pages = work->nr_pages; 7980dc83bd3SJan Kara unsigned long oldest_jif; 799a5989bdcSJan Kara struct inode *inode; 800d46db3d5SWu Fengguang long progress; 80103ba3782SJens Axboe 8020dc83bd3SJan Kara oldest_jif = jiffies; 8030dc83bd3SJan Kara work->older_than_this = &oldest_jif; 80403ba3782SJens Axboe 805e8dfc305SWu Fengguang spin_lock(&wb->list_lock); 80603ba3782SJens Axboe for (;;) { 80703ba3782SJens Axboe /* 808d3ddec76SWu Fengguang * Stop writeback when nr_pages has been consumed 80903ba3782SJens Axboe */ 81083ba7b07SChristoph Hellwig if (work->nr_pages <= 0) 81103ba3782SJens Axboe break; 81203ba3782SJens Axboe 81303ba3782SJens Axboe /* 814aa373cf5SJan Kara * Background writeout and kupdate-style writeback may 815aa373cf5SJan Kara * run forever. Stop them if there is other work to do 816aa373cf5SJan Kara * so that e.g. sync can proceed. They'll be restarted 817aa373cf5SJan Kara * after the other works are all done. 818aa373cf5SJan Kara */ 819aa373cf5SJan Kara if ((work->for_background || work->for_kupdate) && 820aa373cf5SJan Kara !list_empty(&wb->bdi->work_list)) 821aa373cf5SJan Kara break; 822aa373cf5SJan Kara 823aa373cf5SJan Kara /* 824d3ddec76SWu Fengguang * For background writeout, stop when we are below the 825d3ddec76SWu Fengguang * background dirty threshold 82603ba3782SJens Axboe */ 827b00949aaSWu Fengguang if (work->for_background && !over_bground_thresh(wb->bdi)) 82803ba3782SJens Axboe break; 82903ba3782SJens Axboe 8301bc36b64SJan Kara /* 8311bc36b64SJan Kara * Kupdate and background works are special and we want to 8321bc36b64SJan Kara * include all inodes that need writing. Livelock avoidance is 8331bc36b64SJan Kara * handled by these works yielding to any other work so we are 8341bc36b64SJan Kara * safe. 8351bc36b64SJan Kara */ 836ba9aa839SWu Fengguang if (work->for_kupdate) { 8370dc83bd3SJan Kara oldest_jif = jiffies - 838ba9aa839SWu Fengguang msecs_to_jiffies(dirty_expire_interval * 10); 8391bc36b64SJan Kara } else if (work->for_background) 8400dc83bd3SJan Kara oldest_jif = jiffies; 841028c2dd1SDave Chinner 842d46db3d5SWu Fengguang trace_writeback_start(wb->bdi, work); 843e8dfc305SWu Fengguang if (list_empty(&wb->b_io)) 844ad4e38ddSCurt Wohlgemuth queue_io(wb, work); 84583ba7b07SChristoph Hellwig if (work->sb) 846d46db3d5SWu Fengguang progress = writeback_sb_inodes(work->sb, wb, work); 847edadfb10SChristoph Hellwig else 848d46db3d5SWu Fengguang progress = __writeback_inodes_wb(wb, work); 849d46db3d5SWu Fengguang trace_writeback_written(wb->bdi, work); 850028c2dd1SDave Chinner 851e98be2d5SWu Fengguang wb_update_bandwidth(wb, wb_start); 85203ba3782SJens Axboe 85303ba3782SJens Axboe /* 85471fd05a8SJens Axboe * Did we write something? Try for more 855e6fb6da2SWu Fengguang * 856e6fb6da2SWu Fengguang * Dirty inodes are moved to b_io for writeback in batches. 857e6fb6da2SWu Fengguang * The completion of the current batch does not necessarily 858e6fb6da2SWu Fengguang * mean the overall work is done. So we keep looping as long 859e6fb6da2SWu Fengguang * as made some progress on cleaning pages or inodes. 86071fd05a8SJens Axboe */ 861d46db3d5SWu Fengguang if (progress) 86203ba3782SJens Axboe continue; 863a5989bdcSJan Kara /* 864e6fb6da2SWu Fengguang * No more inodes for IO, bail 865a5989bdcSJan Kara */ 866b7a2441fSWu Fengguang if (list_empty(&wb->b_more_io)) 86703ba3782SJens Axboe break; 86803ba3782SJens Axboe /* 8698010c3b6SJens Axboe * Nothing written. Wait for some inode to 8708010c3b6SJens Axboe * become available for writeback. Otherwise 8718010c3b6SJens Axboe * we'll just busyloop. 87203ba3782SJens Axboe */ 87303ba3782SJens Axboe if (!list_empty(&wb->b_more_io)) { 874d46db3d5SWu Fengguang trace_writeback_wait(wb->bdi, work); 87503ba3782SJens Axboe inode = wb_inode(wb->b_more_io.prev); 876250df6edSDave Chinner spin_lock(&inode->i_lock); 877f0d07b7fSJan Kara spin_unlock(&wb->list_lock); 878169ebd90SJan Kara /* This function drops i_lock... */ 879169ebd90SJan Kara inode_sleep_on_writeback(inode); 880f0d07b7fSJan Kara spin_lock(&wb->list_lock); 88103ba3782SJens Axboe } 88203ba3782SJens Axboe } 883e8dfc305SWu Fengguang spin_unlock(&wb->list_lock); 88403ba3782SJens Axboe 885d46db3d5SWu Fengguang return nr_pages - work->nr_pages; 88603ba3782SJens Axboe } 88703ba3782SJens Axboe 88803ba3782SJens Axboe /* 88983ba7b07SChristoph Hellwig * Return the next wb_writeback_work struct that hasn't been processed yet. 89003ba3782SJens Axboe */ 89183ba7b07SChristoph Hellwig static struct wb_writeback_work * 89208852b6dSMinchan Kim get_next_work_item(struct backing_dev_info *bdi) 89303ba3782SJens Axboe { 89483ba7b07SChristoph Hellwig struct wb_writeback_work *work = NULL; 89503ba3782SJens Axboe 8966467716aSArtem Bityutskiy spin_lock_bh(&bdi->wb_lock); 89783ba7b07SChristoph Hellwig if (!list_empty(&bdi->work_list)) { 89883ba7b07SChristoph Hellwig work = list_entry(bdi->work_list.next, 89983ba7b07SChristoph Hellwig struct wb_writeback_work, list); 90083ba7b07SChristoph Hellwig list_del_init(&work->list); 90103ba3782SJens Axboe } 9026467716aSArtem Bityutskiy spin_unlock_bh(&bdi->wb_lock); 90383ba7b07SChristoph Hellwig return work; 90403ba3782SJens Axboe } 90503ba3782SJens Axboe 906cdf01dd5SLinus Torvalds /* 907cdf01dd5SLinus Torvalds * Add in the number of potentially dirty inodes, because each inode 908cdf01dd5SLinus Torvalds * write can dirty pagecache in the underlying blockdev. 909cdf01dd5SLinus Torvalds */ 910cdf01dd5SLinus Torvalds static unsigned long get_nr_dirty_pages(void) 911cdf01dd5SLinus Torvalds { 912cdf01dd5SLinus Torvalds return global_page_state(NR_FILE_DIRTY) + 913cdf01dd5SLinus Torvalds global_page_state(NR_UNSTABLE_NFS) + 914cdf01dd5SLinus Torvalds get_nr_dirty_inodes(); 915cdf01dd5SLinus Torvalds } 916cdf01dd5SLinus Torvalds 9176585027aSJan Kara static long wb_check_background_flush(struct bdi_writeback *wb) 9186585027aSJan Kara { 919b00949aaSWu Fengguang if (over_bground_thresh(wb->bdi)) { 9206585027aSJan Kara 9216585027aSJan Kara struct wb_writeback_work work = { 9226585027aSJan Kara .nr_pages = LONG_MAX, 9236585027aSJan Kara .sync_mode = WB_SYNC_NONE, 9246585027aSJan Kara .for_background = 1, 9256585027aSJan Kara .range_cyclic = 1, 9260e175a18SCurt Wohlgemuth .reason = WB_REASON_BACKGROUND, 9276585027aSJan Kara }; 9286585027aSJan Kara 9296585027aSJan Kara return wb_writeback(wb, &work); 9306585027aSJan Kara } 9316585027aSJan Kara 9326585027aSJan Kara return 0; 9336585027aSJan Kara } 9346585027aSJan Kara 93503ba3782SJens Axboe static long wb_check_old_data_flush(struct bdi_writeback *wb) 93603ba3782SJens Axboe { 93703ba3782SJens Axboe unsigned long expired; 93803ba3782SJens Axboe long nr_pages; 93903ba3782SJens Axboe 94069b62d01SJens Axboe /* 94169b62d01SJens Axboe * When set to zero, disable periodic writeback 94269b62d01SJens Axboe */ 94369b62d01SJens Axboe if (!dirty_writeback_interval) 94469b62d01SJens Axboe return 0; 94569b62d01SJens Axboe 94603ba3782SJens Axboe expired = wb->last_old_flush + 94703ba3782SJens Axboe msecs_to_jiffies(dirty_writeback_interval * 10); 94803ba3782SJens Axboe if (time_before(jiffies, expired)) 94903ba3782SJens Axboe return 0; 95003ba3782SJens Axboe 95103ba3782SJens Axboe wb->last_old_flush = jiffies; 952cdf01dd5SLinus Torvalds nr_pages = get_nr_dirty_pages(); 95303ba3782SJens Axboe 954c4a77a6cSJens Axboe if (nr_pages) { 95583ba7b07SChristoph Hellwig struct wb_writeback_work work = { 956c4a77a6cSJens Axboe .nr_pages = nr_pages, 957c4a77a6cSJens Axboe .sync_mode = WB_SYNC_NONE, 958c4a77a6cSJens Axboe .for_kupdate = 1, 959c4a77a6cSJens Axboe .range_cyclic = 1, 9600e175a18SCurt Wohlgemuth .reason = WB_REASON_PERIODIC, 961c4a77a6cSJens Axboe }; 962c4a77a6cSJens Axboe 96383ba7b07SChristoph Hellwig return wb_writeback(wb, &work); 964c4a77a6cSJens Axboe } 96503ba3782SJens Axboe 96603ba3782SJens Axboe return 0; 96703ba3782SJens Axboe } 96803ba3782SJens Axboe 96903ba3782SJens Axboe /* 97003ba3782SJens Axboe * Retrieve work items and do the writeback they describe 97103ba3782SJens Axboe */ 97225d130baSWanpeng Li static long wb_do_writeback(struct bdi_writeback *wb) 97303ba3782SJens Axboe { 97403ba3782SJens Axboe struct backing_dev_info *bdi = wb->bdi; 97583ba7b07SChristoph Hellwig struct wb_writeback_work *work; 976c4a77a6cSJens Axboe long wrote = 0; 97703ba3782SJens Axboe 97881d73a32SJan Kara set_bit(BDI_writeback_running, &wb->bdi->state); 97908852b6dSMinchan Kim while ((work = get_next_work_item(bdi)) != NULL) { 98083ba7b07SChristoph Hellwig 981455b2864SDave Chinner trace_writeback_exec(bdi, work); 982455b2864SDave Chinner 98383ba7b07SChristoph Hellwig wrote += wb_writeback(wb, work); 98403ba3782SJens Axboe 98503ba3782SJens Axboe /* 98683ba7b07SChristoph Hellwig * Notify the caller of completion if this is a synchronous 98783ba7b07SChristoph Hellwig * work item, otherwise just free it. 98803ba3782SJens Axboe */ 98983ba7b07SChristoph Hellwig if (work->done) 99083ba7b07SChristoph Hellwig complete(work->done); 99183ba7b07SChristoph Hellwig else 99283ba7b07SChristoph Hellwig kfree(work); 99303ba3782SJens Axboe } 99403ba3782SJens Axboe 99503ba3782SJens Axboe /* 99603ba3782SJens Axboe * Check for periodic writeback, kupdated() style 99703ba3782SJens Axboe */ 99803ba3782SJens Axboe wrote += wb_check_old_data_flush(wb); 9996585027aSJan Kara wrote += wb_check_background_flush(wb); 100081d73a32SJan Kara clear_bit(BDI_writeback_running, &wb->bdi->state); 100103ba3782SJens Axboe 100203ba3782SJens Axboe return wrote; 100303ba3782SJens Axboe } 100403ba3782SJens Axboe 100503ba3782SJens Axboe /* 100603ba3782SJens Axboe * Handle writeback of dirty data for the device backed by this bdi. Also 1007839a8e86STejun Heo * reschedules periodically and does kupdated style flushing. 100803ba3782SJens Axboe */ 1009839a8e86STejun Heo void bdi_writeback_workfn(struct work_struct *work) 101003ba3782SJens Axboe { 1011839a8e86STejun Heo struct bdi_writeback *wb = container_of(to_delayed_work(work), 1012839a8e86STejun Heo struct bdi_writeback, dwork); 101308243900SChristoph Hellwig struct backing_dev_info *bdi = wb->bdi; 101403ba3782SJens Axboe long pages_written; 101503ba3782SJens Axboe 1016ef3b1019STejun Heo set_worker_desc("flush-%s", dev_name(bdi->dev)); 1017766f9164SPeter Zijlstra current->flags |= PF_SWAPWRITE; 101803ba3782SJens Axboe 1019839a8e86STejun Heo if (likely(!current_is_workqueue_rescuer() || 1020839a8e86STejun Heo list_empty(&bdi->bdi_list))) { 102103ba3782SJens Axboe /* 1022839a8e86STejun Heo * The normal path. Keep writing back @bdi until its 1023839a8e86STejun Heo * work_list is empty. Note that this path is also taken 1024839a8e86STejun Heo * if @bdi is shutting down even when we're running off the 1025839a8e86STejun Heo * rescuer as work_list needs to be drained. 102603ba3782SJens Axboe */ 1027839a8e86STejun Heo do { 102825d130baSWanpeng Li pages_written = wb_do_writeback(wb); 1029455b2864SDave Chinner trace_writeback_pages_written(pages_written); 1030839a8e86STejun Heo } while (!list_empty(&bdi->work_list)); 1031839a8e86STejun Heo } else { 1032253c34e9SArtem Bityutskiy /* 1033839a8e86STejun Heo * bdi_wq can't get enough workers and we're running off 1034839a8e86STejun Heo * the emergency worker. Don't hog it. Hopefully, 1024 is 1035839a8e86STejun Heo * enough for efficient IO. 1036253c34e9SArtem Bityutskiy */ 1037839a8e86STejun Heo pages_written = writeback_inodes_wb(&bdi->wb, 1024, 1038839a8e86STejun Heo WB_REASON_FORKER_THREAD); 1039839a8e86STejun Heo trace_writeback_pages_written(pages_written); 104003ba3782SJens Axboe } 104103ba3782SJens Axboe 10426ca738d6SDerek Basehore if (!list_empty(&bdi->work_list)) 10436ca738d6SDerek Basehore mod_delayed_work(bdi_wq, &wb->dwork, 0); 10446ca738d6SDerek Basehore else if (wb_has_dirty_io(wb) && dirty_writeback_interval) 10456ca738d6SDerek Basehore bdi_wakeup_thread_delayed(bdi); 1046455b2864SDave Chinner 1047839a8e86STejun Heo current->flags &= ~PF_SWAPWRITE; 104803ba3782SJens Axboe } 104903ba3782SJens Axboe 105003ba3782SJens Axboe /* 105103ba3782SJens Axboe * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back 105203ba3782SJens Axboe * the whole world. 105303ba3782SJens Axboe */ 10540e175a18SCurt Wohlgemuth void wakeup_flusher_threads(long nr_pages, enum wb_reason reason) 105503ba3782SJens Axboe { 1056b8c2f347SChristoph Hellwig struct backing_dev_info *bdi; 1057b8c2f347SChristoph Hellwig 105847df3ddeSJan Kara if (!nr_pages) 105947df3ddeSJan Kara nr_pages = get_nr_dirty_pages(); 1060b8c2f347SChristoph Hellwig 1061b8c2f347SChristoph Hellwig rcu_read_lock(); 1062b8c2f347SChristoph Hellwig list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { 1063b8c2f347SChristoph Hellwig if (!bdi_has_dirty_io(bdi)) 1064b8c2f347SChristoph Hellwig continue; 10650e175a18SCurt Wohlgemuth __bdi_start_writeback(bdi, nr_pages, false, reason); 1066b8c2f347SChristoph Hellwig } 1067b8c2f347SChristoph Hellwig rcu_read_unlock(); 106803ba3782SJens Axboe } 106903ba3782SJens Axboe 107003ba3782SJens Axboe static noinline void block_dump___mark_inode_dirty(struct inode *inode) 107103ba3782SJens Axboe { 107203ba3782SJens Axboe if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { 107303ba3782SJens Axboe struct dentry *dentry; 107403ba3782SJens Axboe const char *name = "?"; 107503ba3782SJens Axboe 107603ba3782SJens Axboe dentry = d_find_alias(inode); 107703ba3782SJens Axboe if (dentry) { 107803ba3782SJens Axboe spin_lock(&dentry->d_lock); 107903ba3782SJens Axboe name = (const char *) dentry->d_name.name; 108003ba3782SJens Axboe } 108103ba3782SJens Axboe printk(KERN_DEBUG 108203ba3782SJens Axboe "%s(%d): dirtied inode %lu (%s) on %s\n", 108303ba3782SJens Axboe current->comm, task_pid_nr(current), inode->i_ino, 108403ba3782SJens Axboe name, inode->i_sb->s_id); 108503ba3782SJens Axboe if (dentry) { 108603ba3782SJens Axboe spin_unlock(&dentry->d_lock); 108703ba3782SJens Axboe dput(dentry); 108803ba3782SJens Axboe } 108903ba3782SJens Axboe } 109003ba3782SJens Axboe } 109103ba3782SJens Axboe 109203ba3782SJens Axboe /** 109303ba3782SJens Axboe * __mark_inode_dirty - internal function 109403ba3782SJens Axboe * @inode: inode to mark 109503ba3782SJens Axboe * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) 109603ba3782SJens Axboe * Mark an inode as dirty. Callers should use mark_inode_dirty or 109703ba3782SJens Axboe * mark_inode_dirty_sync. 109803ba3782SJens Axboe * 109903ba3782SJens Axboe * Put the inode on the super block's dirty list. 110003ba3782SJens Axboe * 110103ba3782SJens Axboe * CAREFUL! We mark it dirty unconditionally, but move it onto the 110203ba3782SJens Axboe * dirty list only if it is hashed or if it refers to a blockdev. 110303ba3782SJens Axboe * If it was not hashed, it will never be added to the dirty list 110403ba3782SJens Axboe * even if it is later hashed, as it will have been marked dirty already. 110503ba3782SJens Axboe * 110603ba3782SJens Axboe * In short, make sure you hash any inodes _before_ you start marking 110703ba3782SJens Axboe * them dirty. 110803ba3782SJens Axboe * 110903ba3782SJens Axboe * Note that for blockdevs, inode->dirtied_when represents the dirtying time of 111003ba3782SJens Axboe * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of 111103ba3782SJens Axboe * the kernel-internal blockdev inode represents the dirtying time of the 111203ba3782SJens Axboe * blockdev's pages. This is why for I_DIRTY_PAGES we always use 111303ba3782SJens Axboe * page->mapping->host, so the page-dirtying time is recorded in the internal 111403ba3782SJens Axboe * blockdev inode. 111503ba3782SJens Axboe */ 111603ba3782SJens Axboe void __mark_inode_dirty(struct inode *inode, int flags) 111703ba3782SJens Axboe { 111803ba3782SJens Axboe struct super_block *sb = inode->i_sb; 1119253c34e9SArtem Bityutskiy struct backing_dev_info *bdi = NULL; 112003ba3782SJens Axboe 112103ba3782SJens Axboe /* 112203ba3782SJens Axboe * Don't do this for I_DIRTY_PAGES - that doesn't actually 112303ba3782SJens Axboe * dirty the inode itself 112403ba3782SJens Axboe */ 112503ba3782SJens Axboe if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { 11269fb0a7daSTejun Heo trace_writeback_dirty_inode_start(inode, flags); 11279fb0a7daSTejun Heo 112803ba3782SJens Axboe if (sb->s_op->dirty_inode) 1129aa385729SChristoph Hellwig sb->s_op->dirty_inode(inode, flags); 11309fb0a7daSTejun Heo 11319fb0a7daSTejun Heo trace_writeback_dirty_inode(inode, flags); 113203ba3782SJens Axboe } 113303ba3782SJens Axboe 113403ba3782SJens Axboe /* 113503ba3782SJens Axboe * make sure that changes are seen by all cpus before we test i_state 113603ba3782SJens Axboe * -- mikulas 113703ba3782SJens Axboe */ 113803ba3782SJens Axboe smp_mb(); 113903ba3782SJens Axboe 114003ba3782SJens Axboe /* avoid the locking if we can */ 114103ba3782SJens Axboe if ((inode->i_state & flags) == flags) 114203ba3782SJens Axboe return; 114303ba3782SJens Axboe 114403ba3782SJens Axboe if (unlikely(block_dump)) 114503ba3782SJens Axboe block_dump___mark_inode_dirty(inode); 114603ba3782SJens Axboe 1147250df6edSDave Chinner spin_lock(&inode->i_lock); 114803ba3782SJens Axboe if ((inode->i_state & flags) != flags) { 114903ba3782SJens Axboe const int was_dirty = inode->i_state & I_DIRTY; 115003ba3782SJens Axboe 115103ba3782SJens Axboe inode->i_state |= flags; 115203ba3782SJens Axboe 115303ba3782SJens Axboe /* 115403ba3782SJens Axboe * If the inode is being synced, just update its dirty state. 115503ba3782SJens Axboe * The unlocker will place the inode on the appropriate 115603ba3782SJens Axboe * superblock list, based upon its state. 115703ba3782SJens Axboe */ 115803ba3782SJens Axboe if (inode->i_state & I_SYNC) 1159250df6edSDave Chinner goto out_unlock_inode; 116003ba3782SJens Axboe 116103ba3782SJens Axboe /* 116203ba3782SJens Axboe * Only add valid (hashed) inodes to the superblock's 116303ba3782SJens Axboe * dirty list. Add blockdev inodes as well. 116403ba3782SJens Axboe */ 116503ba3782SJens Axboe if (!S_ISBLK(inode->i_mode)) { 11661d3382cbSAl Viro if (inode_unhashed(inode)) 1167250df6edSDave Chinner goto out_unlock_inode; 116803ba3782SJens Axboe } 1169a4ffdde6SAl Viro if (inode->i_state & I_FREEING) 1170250df6edSDave Chinner goto out_unlock_inode; 117103ba3782SJens Axboe 117203ba3782SJens Axboe /* 117303ba3782SJens Axboe * If the inode was already on b_dirty/b_io/b_more_io, don't 117403ba3782SJens Axboe * reposition it (that would break b_dirty time-ordering). 117503ba3782SJens Axboe */ 117603ba3782SJens Axboe if (!was_dirty) { 1177a66979abSDave Chinner bool wakeup_bdi = false; 1178253c34e9SArtem Bityutskiy bdi = inode_to_bdi(inode); 1179500b067cSJens Axboe 1180146d7009SJunxiao Bi spin_unlock(&inode->i_lock); 1181146d7009SJunxiao Bi spin_lock(&bdi->wb.list_lock); 1182253c34e9SArtem Bityutskiy if (bdi_cap_writeback_dirty(bdi)) { 1183253c34e9SArtem Bityutskiy WARN(!test_bit(BDI_registered, &bdi->state), 1184253c34e9SArtem Bityutskiy "bdi-%s not registered\n", bdi->name); 1185253c34e9SArtem Bityutskiy 1186253c34e9SArtem Bityutskiy /* 1187253c34e9SArtem Bityutskiy * If this is the first dirty inode for this 1188253c34e9SArtem Bityutskiy * bdi, we have to wake-up the corresponding 1189253c34e9SArtem Bityutskiy * bdi thread to make sure background 1190253c34e9SArtem Bityutskiy * write-back happens later. 1191253c34e9SArtem Bityutskiy */ 1192253c34e9SArtem Bityutskiy if (!wb_has_dirty_io(&bdi->wb)) 1193253c34e9SArtem Bityutskiy wakeup_bdi = true; 1194500b067cSJens Axboe } 119503ba3782SJens Axboe 119603ba3782SJens Axboe inode->dirtied_when = jiffies; 11977ccf19a8SNick Piggin list_move(&inode->i_wb_list, &bdi->wb.b_dirty); 1198f758eeabSChristoph Hellwig spin_unlock(&bdi->wb.list_lock); 1199253c34e9SArtem Bityutskiy 1200253c34e9SArtem Bityutskiy if (wakeup_bdi) 12016467716aSArtem Bityutskiy bdi_wakeup_thread_delayed(bdi); 1202a66979abSDave Chinner return; 1203a66979abSDave Chinner } 1204a66979abSDave Chinner } 1205a66979abSDave Chinner out_unlock_inode: 1206a66979abSDave Chinner spin_unlock(&inode->i_lock); 1207a66979abSDave Chinner 120803ba3782SJens Axboe } 120903ba3782SJens Axboe EXPORT_SYMBOL(__mark_inode_dirty); 121003ba3782SJens Axboe 1211b6e51316SJens Axboe static void wait_sb_inodes(struct super_block *sb) 121266f3b8e2SJens Axboe { 121338f21977SNick Piggin struct inode *inode, *old_inode = NULL; 121438f21977SNick Piggin 121503ba3782SJens Axboe /* 121603ba3782SJens Axboe * We need to be protected against the filesystem going from 121703ba3782SJens Axboe * r/o to r/w or vice versa. 121803ba3782SJens Axboe */ 1219b6e51316SJens Axboe WARN_ON(!rwsem_is_locked(&sb->s_umount)); 122003ba3782SJens Axboe 122155fa6091SDave Chinner spin_lock(&inode_sb_list_lock); 122266f3b8e2SJens Axboe 122338f21977SNick Piggin /* 122438f21977SNick Piggin * Data integrity sync. Must wait for all pages under writeback, 122538f21977SNick Piggin * because there may have been pages dirtied before our sync 122638f21977SNick Piggin * call, but which had writeout started before we write it out. 122738f21977SNick Piggin * In which case, the inode may not be on the dirty list, but 122838f21977SNick Piggin * we still have to wait for that writeout. 122938f21977SNick Piggin */ 1230b6e51316SJens Axboe list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { 1231250df6edSDave Chinner struct address_space *mapping = inode->i_mapping; 123238f21977SNick Piggin 1233250df6edSDave Chinner spin_lock(&inode->i_lock); 1234250df6edSDave Chinner if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) || 1235250df6edSDave Chinner (mapping->nrpages == 0)) { 1236250df6edSDave Chinner spin_unlock(&inode->i_lock); 123738f21977SNick Piggin continue; 1238250df6edSDave Chinner } 123938f21977SNick Piggin __iget(inode); 1240250df6edSDave Chinner spin_unlock(&inode->i_lock); 124155fa6091SDave Chinner spin_unlock(&inode_sb_list_lock); 124255fa6091SDave Chinner 124338f21977SNick Piggin /* 124455fa6091SDave Chinner * We hold a reference to 'inode' so it couldn't have been 124555fa6091SDave Chinner * removed from s_inodes list while we dropped the 124655fa6091SDave Chinner * inode_sb_list_lock. We cannot iput the inode now as we can 124755fa6091SDave Chinner * be holding the last reference and we cannot iput it under 124855fa6091SDave Chinner * inode_sb_list_lock. So we keep the reference and iput it 124955fa6091SDave Chinner * later. 125038f21977SNick Piggin */ 125138f21977SNick Piggin iput(old_inode); 125238f21977SNick Piggin old_inode = inode; 125338f21977SNick Piggin 125438f21977SNick Piggin filemap_fdatawait(mapping); 125538f21977SNick Piggin 125638f21977SNick Piggin cond_resched(); 125738f21977SNick Piggin 125855fa6091SDave Chinner spin_lock(&inode_sb_list_lock); 125938f21977SNick Piggin } 126055fa6091SDave Chinner spin_unlock(&inode_sb_list_lock); 126138f21977SNick Piggin iput(old_inode); 126266f3b8e2SJens Axboe } 12631da177e4SLinus Torvalds 1264d8a8559cSJens Axboe /** 12653259f8beSChris Mason * writeback_inodes_sb_nr - writeback dirty inodes from given super_block 1266d8a8559cSJens Axboe * @sb: the superblock 12673259f8beSChris Mason * @nr: the number of pages to write 1268786228abSMarcos Paulo de Souza * @reason: reason why some writeback work initiated 12691da177e4SLinus Torvalds * 1270d8a8559cSJens Axboe * Start writeback on some inodes on this super_block. No guarantees are made 1271d8a8559cSJens Axboe * on how many (if any) will be written, and this function does not wait 12723259f8beSChris Mason * for IO completion of submitted IO. 12731da177e4SLinus Torvalds */ 12740e175a18SCurt Wohlgemuth void writeback_inodes_sb_nr(struct super_block *sb, 12750e175a18SCurt Wohlgemuth unsigned long nr, 12760e175a18SCurt Wohlgemuth enum wb_reason reason) 12771da177e4SLinus Torvalds { 127883ba7b07SChristoph Hellwig DECLARE_COMPLETION_ONSTACK(done); 127983ba7b07SChristoph Hellwig struct wb_writeback_work work = { 12803c4d7165SChristoph Hellwig .sb = sb, 12813c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_NONE, 12826e6938b6SWu Fengguang .tagged_writepages = 1, 128383ba7b07SChristoph Hellwig .done = &done, 12843259f8beSChris Mason .nr_pages = nr, 12850e175a18SCurt Wohlgemuth .reason = reason, 12863c4d7165SChristoph Hellwig }; 12870e3c9a22SJens Axboe 12886eedc701SJan Kara if (sb->s_bdi == &noop_backing_dev_info) 12896eedc701SJan Kara return; 1290cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 129183ba7b07SChristoph Hellwig bdi_queue_work(sb->s_bdi, &work); 129283ba7b07SChristoph Hellwig wait_for_completion(&done); 12931da177e4SLinus Torvalds } 12943259f8beSChris Mason EXPORT_SYMBOL(writeback_inodes_sb_nr); 12953259f8beSChris Mason 12963259f8beSChris Mason /** 12973259f8beSChris Mason * writeback_inodes_sb - writeback dirty inodes from given super_block 12983259f8beSChris Mason * @sb: the superblock 1299786228abSMarcos Paulo de Souza * @reason: reason why some writeback work was initiated 13003259f8beSChris Mason * 13013259f8beSChris Mason * Start writeback on some inodes on this super_block. No guarantees are made 13023259f8beSChris Mason * on how many (if any) will be written, and this function does not wait 13033259f8beSChris Mason * for IO completion of submitted IO. 13043259f8beSChris Mason */ 13050e175a18SCurt Wohlgemuth void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) 13063259f8beSChris Mason { 13070e175a18SCurt Wohlgemuth return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); 13083259f8beSChris Mason } 1309d8a8559cSJens Axboe EXPORT_SYMBOL(writeback_inodes_sb); 1310d8a8559cSJens Axboe 1311d8a8559cSJens Axboe /** 131210ee27a0SMiao Xie * try_to_writeback_inodes_sb_nr - try to start writeback if none underway 13133259f8beSChris Mason * @sb: the superblock 13143259f8beSChris Mason * @nr: the number of pages to write 131510ee27a0SMiao Xie * @reason: the reason of writeback 13163259f8beSChris Mason * 131710ee27a0SMiao Xie * Invoke writeback_inodes_sb_nr if no writeback is currently underway. 13183259f8beSChris Mason * Returns 1 if writeback was started, 0 if not. 13193259f8beSChris Mason */ 132010ee27a0SMiao Xie int try_to_writeback_inodes_sb_nr(struct super_block *sb, 13210e175a18SCurt Wohlgemuth unsigned long nr, 13220e175a18SCurt Wohlgemuth enum wb_reason reason) 13233259f8beSChris Mason { 132410ee27a0SMiao Xie if (writeback_in_progress(sb->s_bdi)) 132510ee27a0SMiao Xie return 1; 132610ee27a0SMiao Xie 132710ee27a0SMiao Xie if (!down_read_trylock(&sb->s_umount)) 132810ee27a0SMiao Xie return 0; 132910ee27a0SMiao Xie 13300e175a18SCurt Wohlgemuth writeback_inodes_sb_nr(sb, nr, reason); 13313259f8beSChris Mason up_read(&sb->s_umount); 13323259f8beSChris Mason return 1; 13333259f8beSChris Mason } 133410ee27a0SMiao Xie EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr); 133510ee27a0SMiao Xie 133610ee27a0SMiao Xie /** 133710ee27a0SMiao Xie * try_to_writeback_inodes_sb - try to start writeback if none underway 133810ee27a0SMiao Xie * @sb: the superblock 133910ee27a0SMiao Xie * @reason: reason why some writeback work was initiated 134010ee27a0SMiao Xie * 134110ee27a0SMiao Xie * Implement by try_to_writeback_inodes_sb_nr() 134210ee27a0SMiao Xie * Returns 1 if writeback was started, 0 if not. 134310ee27a0SMiao Xie */ 134410ee27a0SMiao Xie int try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) 134510ee27a0SMiao Xie { 134610ee27a0SMiao Xie return try_to_writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); 134710ee27a0SMiao Xie } 134810ee27a0SMiao Xie EXPORT_SYMBOL(try_to_writeback_inodes_sb); 13493259f8beSChris Mason 13503259f8beSChris Mason /** 1351d8a8559cSJens Axboe * sync_inodes_sb - sync sb inode pages 1352d8a8559cSJens Axboe * @sb: the superblock 1353d8a8559cSJens Axboe * 1354d8a8559cSJens Axboe * This function writes and waits on any dirty inode belonging to this 13550dc83bd3SJan Kara * super_block. 1356d8a8559cSJens Axboe */ 13570dc83bd3SJan Kara void sync_inodes_sb(struct super_block *sb) 1358d8a8559cSJens Axboe { 135983ba7b07SChristoph Hellwig DECLARE_COMPLETION_ONSTACK(done); 136083ba7b07SChristoph Hellwig struct wb_writeback_work work = { 13613c4d7165SChristoph Hellwig .sb = sb, 13623c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_ALL, 13633c4d7165SChristoph Hellwig .nr_pages = LONG_MAX, 13643c4d7165SChristoph Hellwig .range_cyclic = 0, 136583ba7b07SChristoph Hellwig .done = &done, 13660e175a18SCurt Wohlgemuth .reason = WB_REASON_SYNC, 13677747bd4bSDave Chinner .for_sync = 1, 13683c4d7165SChristoph Hellwig }; 13693c4d7165SChristoph Hellwig 13706eedc701SJan Kara /* Nothing to do? */ 13716eedc701SJan Kara if (sb->s_bdi == &noop_backing_dev_info) 13726eedc701SJan Kara return; 1373cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 1374cf37e972SChristoph Hellwig 137583ba7b07SChristoph Hellwig bdi_queue_work(sb->s_bdi, &work); 137683ba7b07SChristoph Hellwig wait_for_completion(&done); 137783ba7b07SChristoph Hellwig 1378b6e51316SJens Axboe wait_sb_inodes(sb); 1379d8a8559cSJens Axboe } 1380d8a8559cSJens Axboe EXPORT_SYMBOL(sync_inodes_sb); 13811da177e4SLinus Torvalds 13821da177e4SLinus Torvalds /** 13831da177e4SLinus Torvalds * write_inode_now - write an inode to disk 13841da177e4SLinus Torvalds * @inode: inode to write to disk 13851da177e4SLinus Torvalds * @sync: whether the write should be synchronous or not 13861da177e4SLinus Torvalds * 13877f04c26dSAndrea Arcangeli * This function commits an inode to disk immediately if it is dirty. This is 13887f04c26dSAndrea Arcangeli * primarily needed by knfsd. 13897f04c26dSAndrea Arcangeli * 13907f04c26dSAndrea Arcangeli * The caller must either have a ref on the inode or must have set I_WILL_FREE. 13911da177e4SLinus Torvalds */ 13921da177e4SLinus Torvalds int write_inode_now(struct inode *inode, int sync) 13931da177e4SLinus Torvalds { 1394f758eeabSChristoph Hellwig struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; 13951da177e4SLinus Torvalds struct writeback_control wbc = { 13961da177e4SLinus Torvalds .nr_to_write = LONG_MAX, 139718914b18SMike Galbraith .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, 1398111ebb6eSOGAWA Hirofumi .range_start = 0, 1399111ebb6eSOGAWA Hirofumi .range_end = LLONG_MAX, 14001da177e4SLinus Torvalds }; 14011da177e4SLinus Torvalds 14021da177e4SLinus Torvalds if (!mapping_cap_writeback_dirty(inode->i_mapping)) 140349364ce2SAndrew Morton wbc.nr_to_write = 0; 14041da177e4SLinus Torvalds 14051da177e4SLinus Torvalds might_sleep(); 14064f8ad655SJan Kara return writeback_single_inode(inode, wb, &wbc); 14071da177e4SLinus Torvalds } 14081da177e4SLinus Torvalds EXPORT_SYMBOL(write_inode_now); 14091da177e4SLinus Torvalds 14101da177e4SLinus Torvalds /** 14111da177e4SLinus Torvalds * sync_inode - write an inode and its pages to disk. 14121da177e4SLinus Torvalds * @inode: the inode to sync 14131da177e4SLinus Torvalds * @wbc: controls the writeback mode 14141da177e4SLinus Torvalds * 14151da177e4SLinus Torvalds * sync_inode() will write an inode and its pages to disk. It will also 14161da177e4SLinus Torvalds * correctly update the inode on its superblock's dirty inode lists and will 14171da177e4SLinus Torvalds * update inode->i_state. 14181da177e4SLinus Torvalds * 14191da177e4SLinus Torvalds * The caller must have a ref on the inode. 14201da177e4SLinus Torvalds */ 14211da177e4SLinus Torvalds int sync_inode(struct inode *inode, struct writeback_control *wbc) 14221da177e4SLinus Torvalds { 14234f8ad655SJan Kara return writeback_single_inode(inode, &inode_to_bdi(inode)->wb, wbc); 14241da177e4SLinus Torvalds } 14251da177e4SLinus Torvalds EXPORT_SYMBOL(sync_inode); 1426c3765016SChristoph Hellwig 1427c3765016SChristoph Hellwig /** 1428c691b9d9SAndrew Morton * sync_inode_metadata - write an inode to disk 1429c3765016SChristoph Hellwig * @inode: the inode to sync 1430c3765016SChristoph Hellwig * @wait: wait for I/O to complete. 1431c3765016SChristoph Hellwig * 1432c691b9d9SAndrew Morton * Write an inode to disk and adjust its dirty state after completion. 1433c3765016SChristoph Hellwig * 1434c3765016SChristoph Hellwig * Note: only writes the actual inode, no associated data or other metadata. 1435c3765016SChristoph Hellwig */ 1436c3765016SChristoph Hellwig int sync_inode_metadata(struct inode *inode, int wait) 1437c3765016SChristoph Hellwig { 1438c3765016SChristoph Hellwig struct writeback_control wbc = { 1439c3765016SChristoph Hellwig .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE, 1440c3765016SChristoph Hellwig .nr_to_write = 0, /* metadata-only */ 1441c3765016SChristoph Hellwig }; 1442c3765016SChristoph Hellwig 1443c3765016SChristoph Hellwig return sync_inode(inode, &wbc); 1444c3765016SChristoph Hellwig } 1445c3765016SChristoph Hellwig EXPORT_SYMBOL(sync_inode_metadata); 1446