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> 2503ba3782SJens Axboe #include <linux/freezer.h> 261da177e4SLinus Torvalds #include <linux/writeback.h> 271da177e4SLinus Torvalds #include <linux/blkdev.h> 281da177e4SLinus Torvalds #include <linux/backing-dev.h> 29455b2864SDave Chinner #include <linux/tracepoint.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; 43d46db3d5SWu Fengguang 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; 490e175a18SCurt Wohlgemuth enum wb_reason reason; /* why was writeback initiated? */ 50c4a77a6cSJens Axboe 518010c3b6SJens Axboe struct list_head list; /* pending work list */ 5283ba7b07SChristoph Hellwig struct completion *done; /* set if the caller waits */ 5303ba3782SJens Axboe }; 5403ba3782SJens Axboe 55455b2864SDave Chinner /* 56455b2864SDave Chinner * We don't actually have pdflush, but this one is exported though /proc... 57455b2864SDave Chinner */ 58455b2864SDave Chinner int nr_pdflush_threads; 59455b2864SDave Chinner 60f11b00f3SAdrian Bunk /** 61f11b00f3SAdrian Bunk * writeback_in_progress - determine whether there is writeback in progress 62f11b00f3SAdrian Bunk * @bdi: the device's backing_dev_info structure. 63f11b00f3SAdrian Bunk * 6403ba3782SJens Axboe * Determine whether there is writeback waiting to be handled against a 6503ba3782SJens Axboe * backing device. 66f11b00f3SAdrian Bunk */ 67f11b00f3SAdrian Bunk int writeback_in_progress(struct backing_dev_info *bdi) 68f11b00f3SAdrian Bunk { 6981d73a32SJan Kara return test_bit(BDI_writeback_running, &bdi->state); 70f11b00f3SAdrian Bunk } 71f11b00f3SAdrian Bunk 72692ebd17SJan Kara static inline struct backing_dev_info *inode_to_bdi(struct inode *inode) 73692ebd17SJan Kara { 74692ebd17SJan Kara struct super_block *sb = inode->i_sb; 75692ebd17SJan Kara 76aaead25bSChristoph Hellwig if (strcmp(sb->s_type->name, "bdev") == 0) 77aaead25bSChristoph Hellwig return inode->i_mapping->backing_dev_info; 78aaead25bSChristoph Hellwig 79692ebd17SJan Kara return sb->s_bdi; 80692ebd17SJan Kara } 81692ebd17SJan Kara 827ccf19a8SNick Piggin static inline struct inode *wb_inode(struct list_head *head) 837ccf19a8SNick Piggin { 847ccf19a8SNick Piggin return list_entry(head, struct inode, i_wb_list); 857ccf19a8SNick Piggin } 867ccf19a8SNick Piggin 8715eb77a0SWu Fengguang /* 8815eb77a0SWu Fengguang * Include the creation of the trace points after defining the 8915eb77a0SWu Fengguang * wb_writeback_work structure and inline functions so that the definition 9015eb77a0SWu Fengguang * remains local to this file. 9115eb77a0SWu Fengguang */ 9215eb77a0SWu Fengguang #define CREATE_TRACE_POINTS 9315eb77a0SWu Fengguang #include <trace/events/writeback.h> 9415eb77a0SWu Fengguang 956585027aSJan Kara /* Wakeup flusher thread or forker thread to fork it. Requires bdi->wb_lock. */ 966585027aSJan Kara static void bdi_wakeup_flusher(struct backing_dev_info *bdi) 974195f73dSNick Piggin { 98fff5b85aSArtem Bityutskiy if (bdi->wb.task) { 99fff5b85aSArtem Bityutskiy wake_up_process(bdi->wb.task); 100fff5b85aSArtem Bityutskiy } else { 1011da177e4SLinus Torvalds /* 102fff5b85aSArtem Bityutskiy * The bdi thread isn't there, wake up the forker thread which 103fff5b85aSArtem Bityutskiy * will create and run it. 1041da177e4SLinus Torvalds */ 10503ba3782SJens Axboe wake_up_process(default_backing_dev_info.wb.task); 1061da177e4SLinus Torvalds } 1076585027aSJan Kara } 1086585027aSJan Kara 1096585027aSJan Kara static void bdi_queue_work(struct backing_dev_info *bdi, 1106585027aSJan Kara struct wb_writeback_work *work) 1116585027aSJan Kara { 1126585027aSJan Kara trace_writeback_queue(bdi, work); 1136585027aSJan Kara 1146585027aSJan Kara spin_lock_bh(&bdi->wb_lock); 1156585027aSJan Kara list_add_tail(&work->list, &bdi->work_list); 1166585027aSJan Kara if (!bdi->wb.task) 1176585027aSJan Kara trace_writeback_nothread(bdi, work); 1186585027aSJan Kara bdi_wakeup_flusher(bdi); 1196467716aSArtem Bityutskiy spin_unlock_bh(&bdi->wb_lock); 12003ba3782SJens Axboe } 1211da177e4SLinus Torvalds 12283ba7b07SChristoph Hellwig static void 12383ba7b07SChristoph Hellwig __bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, 1240e175a18SCurt Wohlgemuth bool range_cyclic, enum wb_reason reason) 1251da177e4SLinus Torvalds { 12683ba7b07SChristoph Hellwig struct wb_writeback_work *work; 12703ba3782SJens Axboe 128bcddc3f0SJens Axboe /* 129bcddc3f0SJens Axboe * This is WB_SYNC_NONE writeback, so if allocation fails just 130bcddc3f0SJens Axboe * wakeup the thread for old dirty data writeback 131bcddc3f0SJens Axboe */ 13283ba7b07SChristoph Hellwig work = kzalloc(sizeof(*work), GFP_ATOMIC); 13383ba7b07SChristoph Hellwig if (!work) { 134455b2864SDave Chinner if (bdi->wb.task) { 135455b2864SDave Chinner trace_writeback_nowork(bdi); 13683ba7b07SChristoph Hellwig wake_up_process(bdi->wb.task); 137455b2864SDave Chinner } 13883ba7b07SChristoph Hellwig return; 13983ba7b07SChristoph Hellwig } 14083ba7b07SChristoph Hellwig 14183ba7b07SChristoph Hellwig work->sync_mode = WB_SYNC_NONE; 14283ba7b07SChristoph Hellwig work->nr_pages = nr_pages; 14383ba7b07SChristoph Hellwig work->range_cyclic = range_cyclic; 1440e175a18SCurt Wohlgemuth work->reason = reason; 14583ba7b07SChristoph Hellwig 146f11fcae8SJens Axboe bdi_queue_work(bdi, work); 14703ba3782SJens Axboe } 148b6e51316SJens Axboe 149b6e51316SJens Axboe /** 150b6e51316SJens Axboe * bdi_start_writeback - start writeback 151b6e51316SJens Axboe * @bdi: the backing device to write from 152b6e51316SJens Axboe * @nr_pages: the number of pages to write 153786228abSMarcos Paulo de Souza * @reason: reason why some writeback work was initiated 154b6e51316SJens Axboe * 155b6e51316SJens Axboe * Description: 156b6e51316SJens Axboe * This does WB_SYNC_NONE opportunistic writeback. The IO is only 15725985edcSLucas De Marchi * started when this function returns, we make no guarantees on 1580e3c9a22SJens Axboe * completion. Caller need not hold sb s_umount semaphore. 159b6e51316SJens Axboe * 160b6e51316SJens Axboe */ 1610e175a18SCurt Wohlgemuth void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, 1620e175a18SCurt Wohlgemuth enum wb_reason reason) 163b6e51316SJens Axboe { 1640e175a18SCurt Wohlgemuth __bdi_start_writeback(bdi, nr_pages, true, reason); 165d3ddec76SWu Fengguang } 166d3ddec76SWu Fengguang 167c5444198SChristoph Hellwig /** 168c5444198SChristoph Hellwig * bdi_start_background_writeback - start background writeback 169c5444198SChristoph Hellwig * @bdi: the backing device to write from 170c5444198SChristoph Hellwig * 171c5444198SChristoph Hellwig * Description: 1726585027aSJan Kara * This makes sure WB_SYNC_NONE background writeback happens. When 1736585027aSJan Kara * this function returns, it is only guaranteed that for given BDI 1746585027aSJan Kara * some IO is happening if we are over background dirty threshold. 1756585027aSJan Kara * Caller need not hold sb s_umount semaphore. 176c5444198SChristoph Hellwig */ 177c5444198SChristoph Hellwig void bdi_start_background_writeback(struct backing_dev_info *bdi) 178c5444198SChristoph Hellwig { 1796585027aSJan Kara /* 1806585027aSJan Kara * We just wake up the flusher thread. It will perform background 1816585027aSJan Kara * writeback as soon as there is no other work to do. 1826585027aSJan Kara */ 18371927e84SWu Fengguang trace_writeback_wake_background(bdi); 1846585027aSJan Kara spin_lock_bh(&bdi->wb_lock); 1856585027aSJan Kara bdi_wakeup_flusher(bdi); 1866585027aSJan Kara spin_unlock_bh(&bdi->wb_lock); 1871da177e4SLinus Torvalds } 1881da177e4SLinus Torvalds 1891da177e4SLinus Torvalds /* 190a66979abSDave Chinner * Remove the inode from the writeback list it is on. 191a66979abSDave Chinner */ 192a66979abSDave Chinner void inode_wb_list_del(struct inode *inode) 193a66979abSDave Chinner { 194f758eeabSChristoph Hellwig struct backing_dev_info *bdi = inode_to_bdi(inode); 195a66979abSDave Chinner 196f758eeabSChristoph Hellwig spin_lock(&bdi->wb.list_lock); 197f758eeabSChristoph Hellwig list_del_init(&inode->i_wb_list); 198f758eeabSChristoph Hellwig spin_unlock(&bdi->wb.list_lock); 199f758eeabSChristoph Hellwig } 200a66979abSDave Chinner 201a66979abSDave Chinner /* 2026610a0bcSAndrew Morton * Redirty an inode: set its when-it-was dirtied timestamp and move it to the 2036610a0bcSAndrew Morton * furthest end of its superblock's dirty-inode list. 2046610a0bcSAndrew Morton * 2056610a0bcSAndrew Morton * Before stamping the inode's ->dirtied_when, we check to see whether it is 20666f3b8e2SJens Axboe * already the most-recently-dirtied inode on the b_dirty list. If that is 2076610a0bcSAndrew Morton * the case then the inode must have been redirtied while it was being written 2086610a0bcSAndrew Morton * out and we don't reset its dirtied_when. 2096610a0bcSAndrew Morton */ 210f758eeabSChristoph Hellwig static void redirty_tail(struct inode *inode, struct bdi_writeback *wb) 2116610a0bcSAndrew Morton { 212f758eeabSChristoph Hellwig assert_spin_locked(&wb->list_lock); 21303ba3782SJens Axboe if (!list_empty(&wb->b_dirty)) { 21466f3b8e2SJens Axboe struct inode *tail; 2156610a0bcSAndrew Morton 2167ccf19a8SNick Piggin tail = wb_inode(wb->b_dirty.next); 21766f3b8e2SJens Axboe if (time_before(inode->dirtied_when, tail->dirtied_when)) 2186610a0bcSAndrew Morton inode->dirtied_when = jiffies; 2196610a0bcSAndrew Morton } 2207ccf19a8SNick Piggin list_move(&inode->i_wb_list, &wb->b_dirty); 2216610a0bcSAndrew Morton } 2226610a0bcSAndrew Morton 2236610a0bcSAndrew Morton /* 22466f3b8e2SJens Axboe * requeue inode for re-scanning after bdi->b_io list is exhausted. 225c986d1e2SAndrew Morton */ 226f758eeabSChristoph Hellwig static void requeue_io(struct inode *inode, struct bdi_writeback *wb) 227c986d1e2SAndrew Morton { 228f758eeabSChristoph Hellwig assert_spin_locked(&wb->list_lock); 2297ccf19a8SNick Piggin list_move(&inode->i_wb_list, &wb->b_more_io); 230c986d1e2SAndrew Morton } 231c986d1e2SAndrew Morton 2321c0eeaf5SJoern Engel static void inode_sync_complete(struct inode *inode) 2331c0eeaf5SJoern Engel { 234365b94aeSJan Kara inode->i_state &= ~I_SYNC; 235365b94aeSJan Kara /* Waiters must see I_SYNC cleared before being woken up */ 2361c0eeaf5SJoern Engel smp_mb(); 2371c0eeaf5SJoern Engel wake_up_bit(&inode->i_state, __I_SYNC); 2381c0eeaf5SJoern Engel } 2391c0eeaf5SJoern Engel 240d2caa3c5SJeff Layton static bool inode_dirtied_after(struct inode *inode, unsigned long t) 241d2caa3c5SJeff Layton { 242d2caa3c5SJeff Layton bool ret = time_after(inode->dirtied_when, t); 243d2caa3c5SJeff Layton #ifndef CONFIG_64BIT 244d2caa3c5SJeff Layton /* 245d2caa3c5SJeff Layton * For inodes being constantly redirtied, dirtied_when can get stuck. 246d2caa3c5SJeff Layton * It _appears_ to be in the future, but is actually in distant past. 247d2caa3c5SJeff Layton * This test is necessary to prevent such wrapped-around relative times 2485b0830cbSJens Axboe * from permanently stopping the whole bdi writeback. 249d2caa3c5SJeff Layton */ 250d2caa3c5SJeff Layton ret = ret && time_before_eq(inode->dirtied_when, jiffies); 251d2caa3c5SJeff Layton #endif 252d2caa3c5SJeff Layton return ret; 253d2caa3c5SJeff Layton } 254d2caa3c5SJeff Layton 255c986d1e2SAndrew Morton /* 256697e6fedSJan Kara * Move expired (dirtied after work->older_than_this) dirty inodes from 257697e6fedSJan Kara * @delaying_queue to @dispatch_queue. 2582c136579SFengguang Wu */ 259e84d0a4fSWu Fengguang static int move_expired_inodes(struct list_head *delaying_queue, 2602c136579SFengguang Wu struct list_head *dispatch_queue, 261ad4e38ddSCurt Wohlgemuth struct wb_writeback_work *work) 2622c136579SFengguang Wu { 2635c03449dSShaohua Li LIST_HEAD(tmp); 2645c03449dSShaohua Li struct list_head *pos, *node; 265cf137307SJens Axboe struct super_block *sb = NULL; 2665c03449dSShaohua Li struct inode *inode; 267cf137307SJens Axboe int do_sb_sort = 0; 268e84d0a4fSWu Fengguang int moved = 0; 2695c03449dSShaohua Li 2702c136579SFengguang Wu while (!list_empty(delaying_queue)) { 2717ccf19a8SNick Piggin inode = wb_inode(delaying_queue->prev); 272ad4e38ddSCurt Wohlgemuth if (work->older_than_this && 273ad4e38ddSCurt Wohlgemuth inode_dirtied_after(inode, *work->older_than_this)) 2742c136579SFengguang Wu break; 275cf137307SJens Axboe if (sb && sb != inode->i_sb) 276cf137307SJens Axboe do_sb_sort = 1; 277cf137307SJens Axboe sb = inode->i_sb; 2787ccf19a8SNick Piggin list_move(&inode->i_wb_list, &tmp); 279e84d0a4fSWu Fengguang moved++; 2805c03449dSShaohua Li } 2815c03449dSShaohua Li 282cf137307SJens Axboe /* just one sb in list, splice to dispatch_queue and we're done */ 283cf137307SJens Axboe if (!do_sb_sort) { 284cf137307SJens Axboe list_splice(&tmp, dispatch_queue); 285e84d0a4fSWu Fengguang goto out; 286cf137307SJens Axboe } 287cf137307SJens Axboe 2885c03449dSShaohua Li /* Move inodes from one superblock together */ 2895c03449dSShaohua Li while (!list_empty(&tmp)) { 2907ccf19a8SNick Piggin sb = wb_inode(tmp.prev)->i_sb; 2915c03449dSShaohua Li list_for_each_prev_safe(pos, node, &tmp) { 2927ccf19a8SNick Piggin inode = wb_inode(pos); 2935c03449dSShaohua Li if (inode->i_sb == sb) 2947ccf19a8SNick Piggin list_move(&inode->i_wb_list, dispatch_queue); 2952c136579SFengguang Wu } 2962c136579SFengguang Wu } 297e84d0a4fSWu Fengguang out: 298e84d0a4fSWu Fengguang return moved; 2995c03449dSShaohua Li } 3002c136579SFengguang Wu 3012c136579SFengguang Wu /* 3022c136579SFengguang Wu * Queue all expired dirty inodes for io, eldest first. 3034ea879b9SWu Fengguang * Before 3044ea879b9SWu Fengguang * newly dirtied b_dirty b_io b_more_io 3054ea879b9SWu Fengguang * =============> gf edc BA 3064ea879b9SWu Fengguang * After 3074ea879b9SWu Fengguang * newly dirtied b_dirty b_io b_more_io 3084ea879b9SWu Fengguang * =============> g fBAedc 3094ea879b9SWu Fengguang * | 3104ea879b9SWu Fengguang * +--> dequeue for IO 3112c136579SFengguang Wu */ 312ad4e38ddSCurt Wohlgemuth static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work) 3132c136579SFengguang Wu { 314e84d0a4fSWu Fengguang int moved; 315f758eeabSChristoph Hellwig assert_spin_locked(&wb->list_lock); 3164ea879b9SWu Fengguang list_splice_init(&wb->b_more_io, &wb->b_io); 317ad4e38ddSCurt Wohlgemuth moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, work); 318ad4e38ddSCurt Wohlgemuth trace_writeback_queue_io(wb, work, moved); 31966f3b8e2SJens Axboe } 32066f3b8e2SJens Axboe 321a9185b41SChristoph Hellwig static int write_inode(struct inode *inode, struct writeback_control *wbc) 32266f3b8e2SJens Axboe { 32303ba3782SJens Axboe if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) 324a9185b41SChristoph Hellwig return inode->i_sb->s_op->write_inode(inode, wbc); 32503ba3782SJens Axboe return 0; 32666f3b8e2SJens Axboe } 32708d8e974SFengguang Wu 3282c136579SFengguang Wu /* 329169ebd90SJan Kara * Wait for writeback on an inode to complete. Called with i_lock held. 330169ebd90SJan Kara * Caller must make sure inode cannot go away when we drop i_lock. 33101c03194SChristoph Hellwig */ 332169ebd90SJan Kara static void __inode_wait_for_writeback(struct inode *inode) 333169ebd90SJan Kara __releases(inode->i_lock) 334169ebd90SJan Kara __acquires(inode->i_lock) 33501c03194SChristoph Hellwig { 33601c03194SChristoph Hellwig DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); 33701c03194SChristoph Hellwig wait_queue_head_t *wqh; 33801c03194SChristoph Hellwig 33901c03194SChristoph Hellwig wqh = bit_waitqueue(&inode->i_state, __I_SYNC); 34058a9d3d8SRichard Kennedy while (inode->i_state & I_SYNC) { 341250df6edSDave Chinner spin_unlock(&inode->i_lock); 34201c03194SChristoph Hellwig __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE); 343250df6edSDave Chinner spin_lock(&inode->i_lock); 34458a9d3d8SRichard Kennedy } 34501c03194SChristoph Hellwig } 34601c03194SChristoph Hellwig 34701c03194SChristoph Hellwig /* 348169ebd90SJan Kara * Wait for writeback on an inode to complete. Caller must have inode pinned. 349169ebd90SJan Kara */ 350169ebd90SJan Kara void inode_wait_for_writeback(struct inode *inode) 351169ebd90SJan Kara { 352169ebd90SJan Kara spin_lock(&inode->i_lock); 353169ebd90SJan Kara __inode_wait_for_writeback(inode); 354169ebd90SJan Kara spin_unlock(&inode->i_lock); 355169ebd90SJan Kara } 356169ebd90SJan Kara 357169ebd90SJan Kara /* 358169ebd90SJan Kara * Sleep until I_SYNC is cleared. This function must be called with i_lock 359169ebd90SJan Kara * held and drops it. It is aimed for callers not holding any inode reference 360169ebd90SJan Kara * so once i_lock is dropped, inode can go away. 361169ebd90SJan Kara */ 362169ebd90SJan Kara static void inode_sleep_on_writeback(struct inode *inode) 363169ebd90SJan Kara __releases(inode->i_lock) 364169ebd90SJan Kara { 365169ebd90SJan Kara DEFINE_WAIT(wait); 366169ebd90SJan Kara wait_queue_head_t *wqh = bit_waitqueue(&inode->i_state, __I_SYNC); 367169ebd90SJan Kara int sleep; 368169ebd90SJan Kara 369169ebd90SJan Kara prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); 370169ebd90SJan Kara sleep = inode->i_state & I_SYNC; 371169ebd90SJan Kara spin_unlock(&inode->i_lock); 372169ebd90SJan Kara if (sleep) 373169ebd90SJan Kara schedule(); 374169ebd90SJan Kara finish_wait(wqh, &wait); 375169ebd90SJan Kara } 376169ebd90SJan Kara 377169ebd90SJan Kara /* 378ccb26b5aSJan Kara * Find proper writeback list for the inode depending on its current state and 379ccb26b5aSJan Kara * possibly also change of its state while we were doing writeback. Here we 380ccb26b5aSJan Kara * handle things such as livelock prevention or fairness of writeback among 381ccb26b5aSJan Kara * inodes. This function can be called only by flusher thread - noone else 382ccb26b5aSJan Kara * processes all inodes in writeback lists and requeueing inodes behind flusher 383ccb26b5aSJan Kara * thread's back can have unexpected consequences. 384ccb26b5aSJan Kara */ 385ccb26b5aSJan Kara static void requeue_inode(struct inode *inode, struct bdi_writeback *wb, 386ccb26b5aSJan Kara struct writeback_control *wbc) 387ccb26b5aSJan Kara { 388ccb26b5aSJan Kara if (inode->i_state & I_FREEING) 389ccb26b5aSJan Kara return; 390ccb26b5aSJan Kara 391ccb26b5aSJan Kara /* 392ccb26b5aSJan Kara * Sync livelock prevention. Each inode is tagged and synced in one 393ccb26b5aSJan Kara * shot. If still dirty, it will be redirty_tail()'ed below. Update 394ccb26b5aSJan Kara * the dirty time to prevent enqueue and sync it again. 395ccb26b5aSJan Kara */ 396ccb26b5aSJan Kara if ((inode->i_state & I_DIRTY) && 397ccb26b5aSJan Kara (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)) 398ccb26b5aSJan Kara inode->dirtied_when = jiffies; 399ccb26b5aSJan Kara 4004f8ad655SJan Kara if (wbc->pages_skipped) { 4014f8ad655SJan Kara /* 4024f8ad655SJan Kara * writeback is not making progress due to locked 4034f8ad655SJan Kara * buffers. Skip this inode for now. 4044f8ad655SJan Kara */ 4054f8ad655SJan Kara redirty_tail(inode, wb); 4064f8ad655SJan Kara return; 4074f8ad655SJan Kara } 4084f8ad655SJan Kara 409ccb26b5aSJan Kara if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) { 410ccb26b5aSJan Kara /* 411ccb26b5aSJan Kara * We didn't write back all the pages. nfs_writepages() 412ccb26b5aSJan Kara * sometimes bales out without doing anything. 413ccb26b5aSJan Kara */ 414ccb26b5aSJan Kara if (wbc->nr_to_write <= 0) { 415ccb26b5aSJan Kara /* Slice used up. Queue for next turn. */ 416ccb26b5aSJan Kara requeue_io(inode, wb); 417ccb26b5aSJan Kara } else { 418ccb26b5aSJan Kara /* 419ccb26b5aSJan Kara * Writeback blocked by something other than 420ccb26b5aSJan Kara * congestion. Delay the inode for some time to 421ccb26b5aSJan Kara * avoid spinning on the CPU (100% iowait) 422ccb26b5aSJan Kara * retrying writeback of the dirty page/inode 423ccb26b5aSJan Kara * that cannot be performed immediately. 424ccb26b5aSJan Kara */ 425ccb26b5aSJan Kara redirty_tail(inode, wb); 426ccb26b5aSJan Kara } 427ccb26b5aSJan Kara } else if (inode->i_state & I_DIRTY) { 428ccb26b5aSJan Kara /* 429ccb26b5aSJan Kara * Filesystems can dirty the inode during writeback operations, 430ccb26b5aSJan Kara * such as delayed allocation during submission or metadata 431ccb26b5aSJan Kara * updates after data IO completion. 432ccb26b5aSJan Kara */ 433ccb26b5aSJan Kara redirty_tail(inode, wb); 434ccb26b5aSJan Kara } else { 435ccb26b5aSJan Kara /* The inode is clean. Remove from writeback lists. */ 436ccb26b5aSJan Kara list_del_init(&inode->i_wb_list); 437ccb26b5aSJan Kara } 438ccb26b5aSJan Kara } 439ccb26b5aSJan Kara 440ccb26b5aSJan Kara /* 4414f8ad655SJan Kara * Write out an inode and its dirty pages. Do not update the writeback list 4424f8ad655SJan Kara * linkage. That is left to the caller. The caller is also responsible for 4434f8ad655SJan Kara * setting I_SYNC flag and calling inode_sync_complete() to clear it. 4441da177e4SLinus Torvalds */ 4451da177e4SLinus Torvalds static int 4464f8ad655SJan Kara __writeback_single_inode(struct inode *inode, struct bdi_writeback *wb, 447f758eeabSChristoph Hellwig struct writeback_control *wbc) 4481da177e4SLinus Torvalds { 4491da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 450251d6a47SWu Fengguang long nr_to_write = wbc->nr_to_write; 45101c03194SChristoph Hellwig unsigned dirty; 4521da177e4SLinus Torvalds int ret; 4531da177e4SLinus Torvalds 4544f8ad655SJan Kara WARN_ON(!(inode->i_state & I_SYNC)); 4551da177e4SLinus Torvalds 4561da177e4SLinus Torvalds ret = do_writepages(mapping, wbc); 4571da177e4SLinus Torvalds 45826821ed4SChristoph Hellwig /* 45926821ed4SChristoph Hellwig * Make sure to wait on the data before writing out the metadata. 46026821ed4SChristoph Hellwig * This is important for filesystems that modify metadata on data 46126821ed4SChristoph Hellwig * I/O completion. 46226821ed4SChristoph Hellwig */ 463a9185b41SChristoph Hellwig if (wbc->sync_mode == WB_SYNC_ALL) { 46426821ed4SChristoph Hellwig int err = filemap_fdatawait(mapping); 4651da177e4SLinus Torvalds if (ret == 0) 4661da177e4SLinus Torvalds ret = err; 4671da177e4SLinus Torvalds } 4681da177e4SLinus Torvalds 4695547e8aaSDmitry Monakhov /* 4705547e8aaSDmitry Monakhov * Some filesystems may redirty the inode during the writeback 4715547e8aaSDmitry Monakhov * due to delalloc, clear dirty metadata flags right before 4725547e8aaSDmitry Monakhov * write_inode() 4735547e8aaSDmitry Monakhov */ 474250df6edSDave Chinner spin_lock(&inode->i_lock); 4756290be1cSJan Kara /* Clear I_DIRTY_PAGES if we've written out all dirty pages */ 4766290be1cSJan Kara if (!mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) 4776290be1cSJan Kara inode->i_state &= ~I_DIRTY_PAGES; 4785547e8aaSDmitry Monakhov dirty = inode->i_state & I_DIRTY; 4795547e8aaSDmitry Monakhov inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC); 480250df6edSDave Chinner spin_unlock(&inode->i_lock); 48126821ed4SChristoph Hellwig /* Don't write the inode if only I_DIRTY_PAGES was set */ 48226821ed4SChristoph Hellwig if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { 483a9185b41SChristoph Hellwig int err = write_inode(inode, wbc); 4841da177e4SLinus Torvalds if (ret == 0) 4851da177e4SLinus Torvalds ret = err; 4861da177e4SLinus Torvalds } 4874f8ad655SJan Kara trace_writeback_single_inode(inode, wbc, nr_to_write); 4884f8ad655SJan Kara return ret; 4894f8ad655SJan Kara } 4904f8ad655SJan Kara 4914f8ad655SJan Kara /* 4924f8ad655SJan Kara * Write out an inode's dirty pages. Either the caller has an active reference 4934f8ad655SJan Kara * on the inode or the inode has I_WILL_FREE set. 4944f8ad655SJan Kara * 4954f8ad655SJan Kara * This function is designed to be called for writing back one inode which 4964f8ad655SJan Kara * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode() 4974f8ad655SJan Kara * and does more profound writeback list handling in writeback_sb_inodes(). 4984f8ad655SJan Kara */ 4994f8ad655SJan Kara static int 5004f8ad655SJan Kara writeback_single_inode(struct inode *inode, struct bdi_writeback *wb, 5014f8ad655SJan Kara struct writeback_control *wbc) 5024f8ad655SJan Kara { 5034f8ad655SJan Kara int ret = 0; 5044f8ad655SJan Kara 5054f8ad655SJan Kara spin_lock(&inode->i_lock); 5064f8ad655SJan Kara if (!atomic_read(&inode->i_count)) 5074f8ad655SJan Kara WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); 5084f8ad655SJan Kara else 5094f8ad655SJan Kara WARN_ON(inode->i_state & I_WILL_FREE); 5104f8ad655SJan Kara 5114f8ad655SJan Kara if (inode->i_state & I_SYNC) { 5124f8ad655SJan Kara if (wbc->sync_mode != WB_SYNC_ALL) 5134f8ad655SJan Kara goto out; 5144f8ad655SJan Kara /* 515169ebd90SJan Kara * It's a data-integrity sync. We must wait. Since callers hold 516169ebd90SJan Kara * inode reference or inode has I_WILL_FREE set, it cannot go 517169ebd90SJan Kara * away under us. 5184f8ad655SJan Kara */ 519169ebd90SJan Kara __inode_wait_for_writeback(inode); 5204f8ad655SJan Kara } 5214f8ad655SJan Kara WARN_ON(inode->i_state & I_SYNC); 5224f8ad655SJan Kara /* 5234f8ad655SJan Kara * Skip inode if it is clean. We don't want to mess with writeback 5244f8ad655SJan Kara * lists in this function since flusher thread may be doing for example 5254f8ad655SJan Kara * sync in parallel and if we move the inode, it could get skipped. So 5264f8ad655SJan Kara * here we make sure inode is on some writeback list and leave it there 5274f8ad655SJan Kara * unless we have completely cleaned the inode. 5284f8ad655SJan Kara */ 5294f8ad655SJan Kara if (!(inode->i_state & I_DIRTY)) 5304f8ad655SJan Kara goto out; 5314f8ad655SJan Kara inode->i_state |= I_SYNC; 5324f8ad655SJan Kara spin_unlock(&inode->i_lock); 5334f8ad655SJan Kara 5344f8ad655SJan Kara ret = __writeback_single_inode(inode, wb, wbc); 5351da177e4SLinus Torvalds 536f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 537250df6edSDave Chinner spin_lock(&inode->i_lock); 5384f8ad655SJan Kara /* 5394f8ad655SJan Kara * If inode is clean, remove it from writeback lists. Otherwise don't 5404f8ad655SJan Kara * touch it. See comment above for explanation. 5414f8ad655SJan Kara */ 5424f8ad655SJan Kara if (!(inode->i_state & I_DIRTY)) 5434f8ad655SJan Kara list_del_init(&inode->i_wb_list); 5444f8ad655SJan Kara spin_unlock(&wb->list_lock); 5451c0eeaf5SJoern Engel inode_sync_complete(inode); 5464f8ad655SJan Kara out: 5474f8ad655SJan Kara spin_unlock(&inode->i_lock); 5481da177e4SLinus Torvalds return ret; 5491da177e4SLinus Torvalds } 5501da177e4SLinus Torvalds 5511a12d8bdSWu Fengguang static long writeback_chunk_size(struct backing_dev_info *bdi, 5521a12d8bdSWu Fengguang struct wb_writeback_work *work) 553d46db3d5SWu Fengguang { 554d46db3d5SWu Fengguang long pages; 555d46db3d5SWu Fengguang 556d46db3d5SWu Fengguang /* 557d46db3d5SWu Fengguang * WB_SYNC_ALL mode does livelock avoidance by syncing dirty 558d46db3d5SWu Fengguang * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX 559d46db3d5SWu Fengguang * here avoids calling into writeback_inodes_wb() more than once. 560d46db3d5SWu Fengguang * 561d46db3d5SWu Fengguang * The intended call sequence for WB_SYNC_ALL writeback is: 562d46db3d5SWu Fengguang * 563d46db3d5SWu Fengguang * wb_writeback() 564d46db3d5SWu Fengguang * writeback_sb_inodes() <== called only once 565d46db3d5SWu Fengguang * write_cache_pages() <== called once for each inode 566d46db3d5SWu Fengguang * (quickly) tag currently dirty pages 567d46db3d5SWu Fengguang * (maybe slowly) sync all tagged pages 568d46db3d5SWu Fengguang */ 569d46db3d5SWu Fengguang if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages) 570d46db3d5SWu Fengguang pages = LONG_MAX; 5711a12d8bdSWu Fengguang else { 5721a12d8bdSWu Fengguang pages = min(bdi->avg_write_bandwidth / 2, 5731a12d8bdSWu Fengguang global_dirty_limit / DIRTY_SCOPE); 5741a12d8bdSWu Fengguang pages = min(pages, work->nr_pages); 5751a12d8bdSWu Fengguang pages = round_down(pages + MIN_WRITEBACK_PAGES, 5761a12d8bdSWu Fengguang MIN_WRITEBACK_PAGES); 5771a12d8bdSWu Fengguang } 578d46db3d5SWu Fengguang 579d46db3d5SWu Fengguang return pages; 580d46db3d5SWu Fengguang } 581d46db3d5SWu Fengguang 58203ba3782SJens Axboe /* 583f11c9c5cSEdward Shishkin * Write a portion of b_io inodes which belong to @sb. 584edadfb10SChristoph Hellwig * 585edadfb10SChristoph Hellwig * If @only_this_sb is true, then find and write all such 586f11c9c5cSEdward Shishkin * inodes. Otherwise write only ones which go sequentially 587f11c9c5cSEdward Shishkin * in reverse order. 588edadfb10SChristoph Hellwig * 589d46db3d5SWu Fengguang * Return the number of pages and/or inodes written. 590f11c9c5cSEdward Shishkin */ 591d46db3d5SWu Fengguang static long writeback_sb_inodes(struct super_block *sb, 592d46db3d5SWu Fengguang struct bdi_writeback *wb, 593d46db3d5SWu Fengguang struct wb_writeback_work *work) 59403ba3782SJens Axboe { 595d46db3d5SWu Fengguang struct writeback_control wbc = { 596d46db3d5SWu Fengguang .sync_mode = work->sync_mode, 597d46db3d5SWu Fengguang .tagged_writepages = work->tagged_writepages, 598d46db3d5SWu Fengguang .for_kupdate = work->for_kupdate, 599d46db3d5SWu Fengguang .for_background = work->for_background, 600d46db3d5SWu Fengguang .range_cyclic = work->range_cyclic, 601d46db3d5SWu Fengguang .range_start = 0, 602d46db3d5SWu Fengguang .range_end = LLONG_MAX, 603d46db3d5SWu Fengguang }; 604d46db3d5SWu Fengguang unsigned long start_time = jiffies; 605d46db3d5SWu Fengguang long write_chunk; 606d46db3d5SWu Fengguang long wrote = 0; /* count both pages and inodes */ 607d46db3d5SWu Fengguang 60803ba3782SJens Axboe while (!list_empty(&wb->b_io)) { 6097ccf19a8SNick Piggin struct inode *inode = wb_inode(wb->b_io.prev); 610edadfb10SChristoph Hellwig 611edadfb10SChristoph Hellwig if (inode->i_sb != sb) { 612d46db3d5SWu Fengguang if (work->sb) { 613edadfb10SChristoph Hellwig /* 614edadfb10SChristoph Hellwig * We only want to write back data for this 615edadfb10SChristoph Hellwig * superblock, move all inodes not belonging 616edadfb10SChristoph Hellwig * to it back onto the dirty list. 617edadfb10SChristoph Hellwig */ 618f758eeabSChristoph Hellwig redirty_tail(inode, wb); 61966f3b8e2SJens Axboe continue; 62066f3b8e2SJens Axboe } 621edadfb10SChristoph Hellwig 622edadfb10SChristoph Hellwig /* 623edadfb10SChristoph Hellwig * The inode belongs to a different superblock. 624edadfb10SChristoph Hellwig * Bounce back to the caller to unpin this and 625edadfb10SChristoph Hellwig * pin the next superblock. 626edadfb10SChristoph Hellwig */ 627d46db3d5SWu Fengguang break; 628edadfb10SChristoph Hellwig } 629edadfb10SChristoph Hellwig 6309843b76aSChristoph Hellwig /* 6319843b76aSChristoph Hellwig * Don't bother with new inodes or inodes beeing freed, first 6329843b76aSChristoph Hellwig * kind does not need peridic writeout yet, and for the latter 6339843b76aSChristoph Hellwig * kind writeout is handled by the freer. 6349843b76aSChristoph Hellwig */ 635250df6edSDave Chinner spin_lock(&inode->i_lock); 6369843b76aSChristoph Hellwig if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { 637250df6edSDave Chinner spin_unlock(&inode->i_lock); 638fcc5c222SWu Fengguang redirty_tail(inode, wb); 6397ef0d737SNick Piggin continue; 6407ef0d737SNick Piggin } 641cc1676d9SJan Kara if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) { 642cc1676d9SJan Kara /* 643cc1676d9SJan Kara * If this inode is locked for writeback and we are not 644cc1676d9SJan Kara * doing writeback-for-data-integrity, move it to 645cc1676d9SJan Kara * b_more_io so that writeback can proceed with the 646cc1676d9SJan Kara * other inodes on s_io. 647cc1676d9SJan Kara * 648cc1676d9SJan Kara * We'll have another go at writing back this inode 649cc1676d9SJan Kara * when we completed a full scan of b_io. 650cc1676d9SJan Kara */ 651cc1676d9SJan Kara spin_unlock(&inode->i_lock); 652cc1676d9SJan Kara requeue_io(inode, wb); 653cc1676d9SJan Kara trace_writeback_sb_inodes_requeue(inode); 654cc1676d9SJan Kara continue; 655cc1676d9SJan Kara } 656f0d07b7fSJan Kara spin_unlock(&wb->list_lock); 657f0d07b7fSJan Kara 6584f8ad655SJan Kara /* 6594f8ad655SJan Kara * We already requeued the inode if it had I_SYNC set and we 6604f8ad655SJan Kara * are doing WB_SYNC_NONE writeback. So this catches only the 6614f8ad655SJan Kara * WB_SYNC_ALL case. 6624f8ad655SJan Kara */ 663169ebd90SJan Kara if (inode->i_state & I_SYNC) { 664169ebd90SJan Kara /* Wait for I_SYNC. This function drops i_lock... */ 665169ebd90SJan Kara inode_sleep_on_writeback(inode); 666169ebd90SJan Kara /* Inode may be gone, start again */ 667ead188f9SJan Kara spin_lock(&wb->list_lock); 668169ebd90SJan Kara continue; 669169ebd90SJan Kara } 6704f8ad655SJan Kara inode->i_state |= I_SYNC; 6714f8ad655SJan Kara spin_unlock(&inode->i_lock); 672169ebd90SJan Kara 6731a12d8bdSWu Fengguang write_chunk = writeback_chunk_size(wb->bdi, work); 674d46db3d5SWu Fengguang wbc.nr_to_write = write_chunk; 675d46db3d5SWu Fengguang wbc.pages_skipped = 0; 676250df6edSDave Chinner 677169ebd90SJan Kara /* 678169ebd90SJan Kara * We use I_SYNC to pin the inode in memory. While it is set 679169ebd90SJan Kara * evict_inode() will wait so the inode cannot be freed. 680169ebd90SJan Kara */ 6814f8ad655SJan Kara __writeback_single_inode(inode, wb, &wbc); 682d46db3d5SWu Fengguang 683d46db3d5SWu Fengguang work->nr_pages -= write_chunk - wbc.nr_to_write; 684d46db3d5SWu Fengguang wrote += write_chunk - wbc.nr_to_write; 6854f8ad655SJan Kara spin_lock(&wb->list_lock); 6864f8ad655SJan Kara spin_lock(&inode->i_lock); 687d46db3d5SWu Fengguang if (!(inode->i_state & I_DIRTY)) 688d46db3d5SWu Fengguang wrote++; 6894f8ad655SJan Kara requeue_inode(inode, wb, &wbc); 6904f8ad655SJan Kara inode_sync_complete(inode); 6910f1b1fd8SDave Chinner spin_unlock(&inode->i_lock); 692169ebd90SJan Kara cond_resched_lock(&wb->list_lock); 693d46db3d5SWu Fengguang /* 694d46db3d5SWu Fengguang * bail out to wb_writeback() often enough to check 695d46db3d5SWu Fengguang * background threshold and other termination conditions. 696d46db3d5SWu Fengguang */ 697d46db3d5SWu Fengguang if (wrote) { 698d46db3d5SWu Fengguang if (time_is_before_jiffies(start_time + HZ / 10UL)) 699d46db3d5SWu Fengguang break; 700d46db3d5SWu Fengguang if (work->nr_pages <= 0) 701d46db3d5SWu Fengguang break; 7021da177e4SLinus Torvalds } 7038bc3be27SFengguang Wu } 704d46db3d5SWu Fengguang return wrote; 705f11c9c5cSEdward Shishkin } 70638f21977SNick Piggin 707d46db3d5SWu Fengguang static long __writeback_inodes_wb(struct bdi_writeback *wb, 708d46db3d5SWu Fengguang struct wb_writeback_work *work) 709f11c9c5cSEdward Shishkin { 710d46db3d5SWu Fengguang unsigned long start_time = jiffies; 711d46db3d5SWu Fengguang long wrote = 0; 712f11c9c5cSEdward Shishkin 713f11c9c5cSEdward Shishkin while (!list_empty(&wb->b_io)) { 7147ccf19a8SNick Piggin struct inode *inode = wb_inode(wb->b_io.prev); 715f11c9c5cSEdward Shishkin struct super_block *sb = inode->i_sb; 716f11c9c5cSEdward Shishkin 71712ad3ab6SDave Chinner if (!grab_super_passive(sb)) { 7180e995816SWu Fengguang /* 7190e995816SWu Fengguang * grab_super_passive() may fail consistently due to 7200e995816SWu Fengguang * s_umount being grabbed by someone else. Don't use 7210e995816SWu Fengguang * requeue_io() to avoid busy retrying the inode/sb. 7220e995816SWu Fengguang */ 7230e995816SWu Fengguang redirty_tail(inode, wb); 724d19de7edSChristoph Hellwig continue; 725334132aeSChristoph Hellwig } 726d46db3d5SWu Fengguang wrote += writeback_sb_inodes(sb, wb, work); 727d19de7edSChristoph Hellwig drop_super(sb); 728f11c9c5cSEdward Shishkin 729d46db3d5SWu Fengguang /* refer to the same tests at the end of writeback_sb_inodes */ 730d46db3d5SWu Fengguang if (wrote) { 731d46db3d5SWu Fengguang if (time_is_before_jiffies(start_time + HZ / 10UL)) 732d46db3d5SWu Fengguang break; 733d46db3d5SWu Fengguang if (work->nr_pages <= 0) 734f11c9c5cSEdward Shishkin break; 735f11c9c5cSEdward Shishkin } 736d46db3d5SWu Fengguang } 73766f3b8e2SJens Axboe /* Leave any unwritten inodes on b_io */ 738d46db3d5SWu Fengguang return wrote; 73966f3b8e2SJens Axboe } 74066f3b8e2SJens Axboe 7410e175a18SCurt Wohlgemuth long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages, 7420e175a18SCurt Wohlgemuth enum wb_reason reason) 743edadfb10SChristoph Hellwig { 744d46db3d5SWu Fengguang struct wb_writeback_work work = { 745d46db3d5SWu Fengguang .nr_pages = nr_pages, 746d46db3d5SWu Fengguang .sync_mode = WB_SYNC_NONE, 747d46db3d5SWu Fengguang .range_cyclic = 1, 7480e175a18SCurt Wohlgemuth .reason = reason, 749d46db3d5SWu Fengguang }; 750edadfb10SChristoph Hellwig 751f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 752424b351fSWu Fengguang if (list_empty(&wb->b_io)) 753ad4e38ddSCurt Wohlgemuth queue_io(wb, &work); 754d46db3d5SWu Fengguang __writeback_inodes_wb(wb, &work); 755f758eeabSChristoph Hellwig spin_unlock(&wb->list_lock); 756edadfb10SChristoph Hellwig 757d46db3d5SWu Fengguang return nr_pages - work.nr_pages; 75866f3b8e2SJens Axboe } 75966f3b8e2SJens Axboe 760b00949aaSWu Fengguang static bool over_bground_thresh(struct backing_dev_info *bdi) 76103ba3782SJens Axboe { 76203ba3782SJens Axboe unsigned long background_thresh, dirty_thresh; 76303ba3782SJens Axboe 76416c4042fSWu Fengguang global_dirty_limits(&background_thresh, &dirty_thresh); 76503ba3782SJens Axboe 766b00949aaSWu Fengguang if (global_page_state(NR_FILE_DIRTY) + 767b00949aaSWu Fengguang global_page_state(NR_UNSTABLE_NFS) > background_thresh) 768b00949aaSWu Fengguang return true; 769b00949aaSWu Fengguang 770b00949aaSWu Fengguang if (bdi_stat(bdi, BDI_RECLAIMABLE) > 771b00949aaSWu Fengguang bdi_dirty_limit(bdi, background_thresh)) 772b00949aaSWu Fengguang return true; 773b00949aaSWu Fengguang 774b00949aaSWu Fengguang return false; 77503ba3782SJens Axboe } 77603ba3782SJens Axboe 77703ba3782SJens Axboe /* 778e98be2d5SWu Fengguang * Called under wb->list_lock. If there are multiple wb per bdi, 779e98be2d5SWu Fengguang * only the flusher working on the first wb should do it. 780e98be2d5SWu Fengguang */ 781e98be2d5SWu Fengguang static void wb_update_bandwidth(struct bdi_writeback *wb, 782e98be2d5SWu Fengguang unsigned long start_time) 783e98be2d5SWu Fengguang { 784af6a3113SWu Fengguang __bdi_update_bandwidth(wb->bdi, 0, 0, 0, 0, 0, start_time); 785e98be2d5SWu Fengguang } 786e98be2d5SWu Fengguang 787e98be2d5SWu Fengguang /* 78803ba3782SJens Axboe * Explicit flushing or periodic writeback of "old" data. 78903ba3782SJens Axboe * 79003ba3782SJens Axboe * Define "old": the first time one of an inode's pages is dirtied, we mark the 79103ba3782SJens Axboe * dirtying-time in the inode's address_space. So this periodic writeback code 79203ba3782SJens Axboe * just walks the superblock inode list, writing back any inodes which are 79303ba3782SJens Axboe * older than a specific point in time. 79403ba3782SJens Axboe * 79503ba3782SJens Axboe * Try to run once per dirty_writeback_interval. But if a writeback event 79603ba3782SJens Axboe * takes longer than a dirty_writeback_interval interval, then leave a 79703ba3782SJens Axboe * one-second gap. 79803ba3782SJens Axboe * 79903ba3782SJens Axboe * older_than_this takes precedence over nr_to_write. So we'll only write back 80003ba3782SJens Axboe * all dirty pages if they are all attached to "old" mappings. 80103ba3782SJens Axboe */ 802c4a77a6cSJens Axboe static long wb_writeback(struct bdi_writeback *wb, 80383ba7b07SChristoph Hellwig struct wb_writeback_work *work) 80403ba3782SJens Axboe { 805e98be2d5SWu Fengguang unsigned long wb_start = jiffies; 806d46db3d5SWu Fengguang long nr_pages = work->nr_pages; 80703ba3782SJens Axboe unsigned long oldest_jif; 808a5989bdcSJan Kara struct inode *inode; 809d46db3d5SWu Fengguang long progress; 81003ba3782SJens Axboe 811e185dda8SWu Fengguang oldest_jif = jiffies; 812d46db3d5SWu Fengguang work->older_than_this = &oldest_jif; 81303ba3782SJens Axboe 814e8dfc305SWu Fengguang spin_lock(&wb->list_lock); 81503ba3782SJens Axboe for (;;) { 81603ba3782SJens Axboe /* 817d3ddec76SWu Fengguang * Stop writeback when nr_pages has been consumed 81803ba3782SJens Axboe */ 81983ba7b07SChristoph Hellwig if (work->nr_pages <= 0) 82003ba3782SJens Axboe break; 82103ba3782SJens Axboe 82203ba3782SJens Axboe /* 823aa373cf5SJan Kara * Background writeout and kupdate-style writeback may 824aa373cf5SJan Kara * run forever. Stop them if there is other work to do 825aa373cf5SJan Kara * so that e.g. sync can proceed. They'll be restarted 826aa373cf5SJan Kara * after the other works are all done. 827aa373cf5SJan Kara */ 828aa373cf5SJan Kara if ((work->for_background || work->for_kupdate) && 829aa373cf5SJan Kara !list_empty(&wb->bdi->work_list)) 830aa373cf5SJan Kara break; 831aa373cf5SJan Kara 832aa373cf5SJan Kara /* 833d3ddec76SWu Fengguang * For background writeout, stop when we are below the 834d3ddec76SWu Fengguang * background dirty threshold 83503ba3782SJens Axboe */ 836b00949aaSWu Fengguang if (work->for_background && !over_bground_thresh(wb->bdi)) 83703ba3782SJens Axboe break; 83803ba3782SJens Axboe 8391bc36b64SJan Kara /* 8401bc36b64SJan Kara * Kupdate and background works are special and we want to 8411bc36b64SJan Kara * include all inodes that need writing. Livelock avoidance is 8421bc36b64SJan Kara * handled by these works yielding to any other work so we are 8431bc36b64SJan Kara * safe. 8441bc36b64SJan Kara */ 845ba9aa839SWu Fengguang if (work->for_kupdate) { 846ba9aa839SWu Fengguang oldest_jif = jiffies - 847ba9aa839SWu Fengguang msecs_to_jiffies(dirty_expire_interval * 10); 8481bc36b64SJan Kara } else if (work->for_background) 8491bc36b64SJan Kara oldest_jif = jiffies; 850028c2dd1SDave Chinner 851d46db3d5SWu Fengguang trace_writeback_start(wb->bdi, work); 852e8dfc305SWu Fengguang if (list_empty(&wb->b_io)) 853ad4e38ddSCurt Wohlgemuth queue_io(wb, work); 85483ba7b07SChristoph Hellwig if (work->sb) 855d46db3d5SWu Fengguang progress = writeback_sb_inodes(work->sb, wb, work); 856edadfb10SChristoph Hellwig else 857d46db3d5SWu Fengguang progress = __writeback_inodes_wb(wb, work); 858d46db3d5SWu Fengguang trace_writeback_written(wb->bdi, work); 859028c2dd1SDave Chinner 860e98be2d5SWu Fengguang wb_update_bandwidth(wb, wb_start); 86103ba3782SJens Axboe 86203ba3782SJens Axboe /* 86371fd05a8SJens Axboe * Did we write something? Try for more 864e6fb6da2SWu Fengguang * 865e6fb6da2SWu Fengguang * Dirty inodes are moved to b_io for writeback in batches. 866e6fb6da2SWu Fengguang * The completion of the current batch does not necessarily 867e6fb6da2SWu Fengguang * mean the overall work is done. So we keep looping as long 868e6fb6da2SWu Fengguang * as made some progress on cleaning pages or inodes. 86971fd05a8SJens Axboe */ 870d46db3d5SWu Fengguang if (progress) 87103ba3782SJens Axboe continue; 872a5989bdcSJan Kara /* 873e6fb6da2SWu Fengguang * No more inodes for IO, bail 874a5989bdcSJan Kara */ 875b7a2441fSWu Fengguang if (list_empty(&wb->b_more_io)) 87603ba3782SJens Axboe break; 87703ba3782SJens Axboe /* 8788010c3b6SJens Axboe * Nothing written. Wait for some inode to 8798010c3b6SJens Axboe * become available for writeback. Otherwise 8808010c3b6SJens Axboe * we'll just busyloop. 88103ba3782SJens Axboe */ 88203ba3782SJens Axboe if (!list_empty(&wb->b_more_io)) { 883d46db3d5SWu Fengguang trace_writeback_wait(wb->bdi, work); 88403ba3782SJens Axboe inode = wb_inode(wb->b_more_io.prev); 885250df6edSDave Chinner spin_lock(&inode->i_lock); 886f0d07b7fSJan Kara spin_unlock(&wb->list_lock); 887169ebd90SJan Kara /* This function drops i_lock... */ 888169ebd90SJan Kara inode_sleep_on_writeback(inode); 889f0d07b7fSJan Kara spin_lock(&wb->list_lock); 89003ba3782SJens Axboe } 89103ba3782SJens Axboe } 892e8dfc305SWu Fengguang spin_unlock(&wb->list_lock); 89303ba3782SJens Axboe 894d46db3d5SWu Fengguang return nr_pages - work->nr_pages; 89503ba3782SJens Axboe } 89603ba3782SJens Axboe 89703ba3782SJens Axboe /* 89883ba7b07SChristoph Hellwig * Return the next wb_writeback_work struct that hasn't been processed yet. 89903ba3782SJens Axboe */ 90083ba7b07SChristoph Hellwig static struct wb_writeback_work * 90108852b6dSMinchan Kim get_next_work_item(struct backing_dev_info *bdi) 90203ba3782SJens Axboe { 90383ba7b07SChristoph Hellwig struct wb_writeback_work *work = NULL; 90403ba3782SJens Axboe 9056467716aSArtem Bityutskiy spin_lock_bh(&bdi->wb_lock); 90683ba7b07SChristoph Hellwig if (!list_empty(&bdi->work_list)) { 90783ba7b07SChristoph Hellwig work = list_entry(bdi->work_list.next, 90883ba7b07SChristoph Hellwig struct wb_writeback_work, list); 90983ba7b07SChristoph Hellwig list_del_init(&work->list); 91003ba3782SJens Axboe } 9116467716aSArtem Bityutskiy spin_unlock_bh(&bdi->wb_lock); 91283ba7b07SChristoph Hellwig return work; 91303ba3782SJens Axboe } 91403ba3782SJens Axboe 915cdf01dd5SLinus Torvalds /* 916cdf01dd5SLinus Torvalds * Add in the number of potentially dirty inodes, because each inode 917cdf01dd5SLinus Torvalds * write can dirty pagecache in the underlying blockdev. 918cdf01dd5SLinus Torvalds */ 919cdf01dd5SLinus Torvalds static unsigned long get_nr_dirty_pages(void) 920cdf01dd5SLinus Torvalds { 921cdf01dd5SLinus Torvalds return global_page_state(NR_FILE_DIRTY) + 922cdf01dd5SLinus Torvalds global_page_state(NR_UNSTABLE_NFS) + 923cdf01dd5SLinus Torvalds get_nr_dirty_inodes(); 924cdf01dd5SLinus Torvalds } 925cdf01dd5SLinus Torvalds 9266585027aSJan Kara static long wb_check_background_flush(struct bdi_writeback *wb) 9276585027aSJan Kara { 928b00949aaSWu Fengguang if (over_bground_thresh(wb->bdi)) { 9296585027aSJan Kara 9306585027aSJan Kara struct wb_writeback_work work = { 9316585027aSJan Kara .nr_pages = LONG_MAX, 9326585027aSJan Kara .sync_mode = WB_SYNC_NONE, 9336585027aSJan Kara .for_background = 1, 9346585027aSJan Kara .range_cyclic = 1, 9350e175a18SCurt Wohlgemuth .reason = WB_REASON_BACKGROUND, 9366585027aSJan Kara }; 9376585027aSJan Kara 9386585027aSJan Kara return wb_writeback(wb, &work); 9396585027aSJan Kara } 9406585027aSJan Kara 9416585027aSJan Kara return 0; 9426585027aSJan Kara } 9436585027aSJan Kara 94403ba3782SJens Axboe static long wb_check_old_data_flush(struct bdi_writeback *wb) 94503ba3782SJens Axboe { 94603ba3782SJens Axboe unsigned long expired; 94703ba3782SJens Axboe long nr_pages; 94803ba3782SJens Axboe 94969b62d01SJens Axboe /* 95069b62d01SJens Axboe * When set to zero, disable periodic writeback 95169b62d01SJens Axboe */ 95269b62d01SJens Axboe if (!dirty_writeback_interval) 95369b62d01SJens Axboe return 0; 95469b62d01SJens Axboe 95503ba3782SJens Axboe expired = wb->last_old_flush + 95603ba3782SJens Axboe msecs_to_jiffies(dirty_writeback_interval * 10); 95703ba3782SJens Axboe if (time_before(jiffies, expired)) 95803ba3782SJens Axboe return 0; 95903ba3782SJens Axboe 96003ba3782SJens Axboe wb->last_old_flush = jiffies; 961cdf01dd5SLinus Torvalds nr_pages = get_nr_dirty_pages(); 96203ba3782SJens Axboe 963c4a77a6cSJens Axboe if (nr_pages) { 96483ba7b07SChristoph Hellwig struct wb_writeback_work work = { 965c4a77a6cSJens Axboe .nr_pages = nr_pages, 966c4a77a6cSJens Axboe .sync_mode = WB_SYNC_NONE, 967c4a77a6cSJens Axboe .for_kupdate = 1, 968c4a77a6cSJens Axboe .range_cyclic = 1, 9690e175a18SCurt Wohlgemuth .reason = WB_REASON_PERIODIC, 970c4a77a6cSJens Axboe }; 971c4a77a6cSJens Axboe 97283ba7b07SChristoph Hellwig return wb_writeback(wb, &work); 973c4a77a6cSJens Axboe } 97403ba3782SJens Axboe 97503ba3782SJens Axboe return 0; 97603ba3782SJens Axboe } 97703ba3782SJens Axboe 97803ba3782SJens Axboe /* 97903ba3782SJens Axboe * Retrieve work items and do the writeback they describe 98003ba3782SJens Axboe */ 98103ba3782SJens Axboe long wb_do_writeback(struct bdi_writeback *wb, int force_wait) 98203ba3782SJens Axboe { 98303ba3782SJens Axboe struct backing_dev_info *bdi = wb->bdi; 98483ba7b07SChristoph Hellwig struct wb_writeback_work *work; 985c4a77a6cSJens Axboe long wrote = 0; 98603ba3782SJens Axboe 98781d73a32SJan Kara set_bit(BDI_writeback_running, &wb->bdi->state); 98808852b6dSMinchan Kim while ((work = get_next_work_item(bdi)) != NULL) { 98903ba3782SJens Axboe /* 99003ba3782SJens Axboe * Override sync mode, in case we must wait for completion 99183ba7b07SChristoph Hellwig * because this thread is exiting now. 99203ba3782SJens Axboe */ 99303ba3782SJens Axboe if (force_wait) 99483ba7b07SChristoph Hellwig work->sync_mode = WB_SYNC_ALL; 99583ba7b07SChristoph Hellwig 996455b2864SDave Chinner trace_writeback_exec(bdi, work); 997455b2864SDave Chinner 99883ba7b07SChristoph Hellwig wrote += wb_writeback(wb, work); 99903ba3782SJens Axboe 100003ba3782SJens Axboe /* 100183ba7b07SChristoph Hellwig * Notify the caller of completion if this is a synchronous 100283ba7b07SChristoph Hellwig * work item, otherwise just free it. 100303ba3782SJens Axboe */ 100483ba7b07SChristoph Hellwig if (work->done) 100583ba7b07SChristoph Hellwig complete(work->done); 100683ba7b07SChristoph Hellwig else 100783ba7b07SChristoph Hellwig kfree(work); 100803ba3782SJens Axboe } 100903ba3782SJens Axboe 101003ba3782SJens Axboe /* 101103ba3782SJens Axboe * Check for periodic writeback, kupdated() style 101203ba3782SJens Axboe */ 101303ba3782SJens Axboe wrote += wb_check_old_data_flush(wb); 10146585027aSJan Kara wrote += wb_check_background_flush(wb); 101581d73a32SJan Kara clear_bit(BDI_writeback_running, &wb->bdi->state); 101603ba3782SJens Axboe 101703ba3782SJens Axboe return wrote; 101803ba3782SJens Axboe } 101903ba3782SJens Axboe 102003ba3782SJens Axboe /* 102103ba3782SJens Axboe * Handle writeback of dirty data for the device backed by this bdi. Also 102203ba3782SJens Axboe * wakes up periodically and does kupdated style flushing. 102303ba3782SJens Axboe */ 102408243900SChristoph Hellwig int bdi_writeback_thread(void *data) 102503ba3782SJens Axboe { 102608243900SChristoph Hellwig struct bdi_writeback *wb = data; 102708243900SChristoph Hellwig struct backing_dev_info *bdi = wb->bdi; 102803ba3782SJens Axboe long pages_written; 102903ba3782SJens Axboe 1030766f9164SPeter Zijlstra current->flags |= PF_SWAPWRITE; 103108243900SChristoph Hellwig set_freezable(); 1032ecd58403SArtem Bityutskiy wb->last_active = jiffies; 103303ba3782SJens Axboe 103403ba3782SJens Axboe /* 103508243900SChristoph Hellwig * Our parent may run at a different priority, just set us to normal 103603ba3782SJens Axboe */ 103708243900SChristoph Hellwig set_user_nice(current, 0); 103808243900SChristoph Hellwig 1039455b2864SDave Chinner trace_writeback_thread_start(bdi); 1040455b2864SDave Chinner 10418a32c441STejun Heo while (!kthread_freezable_should_stop(NULL)) { 10426467716aSArtem Bityutskiy /* 10436467716aSArtem Bityutskiy * Remove own delayed wake-up timer, since we are already awake 10446467716aSArtem Bityutskiy * and we'll take care of the preriodic write-back. 10456467716aSArtem Bityutskiy */ 10466467716aSArtem Bityutskiy del_timer(&wb->wakeup_timer); 10476467716aSArtem Bityutskiy 104803ba3782SJens Axboe pages_written = wb_do_writeback(wb, 0); 104903ba3782SJens Axboe 1050455b2864SDave Chinner trace_writeback_pages_written(pages_written); 1051455b2864SDave Chinner 105203ba3782SJens Axboe if (pages_written) 1053ecd58403SArtem Bityutskiy wb->last_active = jiffies; 105403ba3782SJens Axboe 1055297252c8SArtem Bityutskiy set_current_state(TASK_INTERRUPTIBLE); 1056b76b4014SJ. Bruce Fields if (!list_empty(&bdi->work_list) || kthread_should_stop()) { 1057297252c8SArtem Bityutskiy __set_current_state(TASK_RUNNING); 1058297252c8SArtem Bityutskiy continue; 105903ba3782SJens Axboe } 106003ba3782SJens Axboe 1061253c34e9SArtem Bityutskiy if (wb_has_dirty_io(wb) && dirty_writeback_interval) 1062fff5b85aSArtem Bityutskiy schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10)); 1063253c34e9SArtem Bityutskiy else { 1064253c34e9SArtem Bityutskiy /* 1065253c34e9SArtem Bityutskiy * We have nothing to do, so can go sleep without any 1066253c34e9SArtem Bityutskiy * timeout and save power. When a work is queued or 1067253c34e9SArtem Bityutskiy * something is made dirty - we will be woken up. 1068253c34e9SArtem Bityutskiy */ 106969b62d01SJens Axboe schedule(); 1070f9eadbbdSJens Axboe } 107103ba3782SJens Axboe } 107203ba3782SJens Axboe 1073fff5b85aSArtem Bityutskiy /* Flush any work that raced with us exiting */ 107408243900SChristoph Hellwig if (!list_empty(&bdi->work_list)) 107508243900SChristoph Hellwig wb_do_writeback(wb, 1); 1076455b2864SDave Chinner 1077455b2864SDave Chinner trace_writeback_thread_stop(bdi); 107803ba3782SJens Axboe return 0; 107903ba3782SJens Axboe } 108003ba3782SJens Axboe 108108243900SChristoph Hellwig 108203ba3782SJens Axboe /* 108303ba3782SJens Axboe * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back 108403ba3782SJens Axboe * the whole world. 108503ba3782SJens Axboe */ 10860e175a18SCurt Wohlgemuth void wakeup_flusher_threads(long nr_pages, enum wb_reason reason) 108703ba3782SJens Axboe { 1088b8c2f347SChristoph Hellwig struct backing_dev_info *bdi; 1089b8c2f347SChristoph Hellwig 109083ba7b07SChristoph Hellwig if (!nr_pages) { 109183ba7b07SChristoph Hellwig nr_pages = global_page_state(NR_FILE_DIRTY) + 109203ba3782SJens Axboe global_page_state(NR_UNSTABLE_NFS); 1093b8c2f347SChristoph Hellwig } 1094b8c2f347SChristoph Hellwig 1095b8c2f347SChristoph Hellwig rcu_read_lock(); 1096b8c2f347SChristoph Hellwig list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { 1097b8c2f347SChristoph Hellwig if (!bdi_has_dirty_io(bdi)) 1098b8c2f347SChristoph Hellwig continue; 10990e175a18SCurt Wohlgemuth __bdi_start_writeback(bdi, nr_pages, false, reason); 1100b8c2f347SChristoph Hellwig } 1101b8c2f347SChristoph Hellwig rcu_read_unlock(); 110203ba3782SJens Axboe } 110303ba3782SJens Axboe 110403ba3782SJens Axboe static noinline void block_dump___mark_inode_dirty(struct inode *inode) 110503ba3782SJens Axboe { 110603ba3782SJens Axboe if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { 110703ba3782SJens Axboe struct dentry *dentry; 110803ba3782SJens Axboe const char *name = "?"; 110903ba3782SJens Axboe 111003ba3782SJens Axboe dentry = d_find_alias(inode); 111103ba3782SJens Axboe if (dentry) { 111203ba3782SJens Axboe spin_lock(&dentry->d_lock); 111303ba3782SJens Axboe name = (const char *) dentry->d_name.name; 111403ba3782SJens Axboe } 111503ba3782SJens Axboe printk(KERN_DEBUG 111603ba3782SJens Axboe "%s(%d): dirtied inode %lu (%s) on %s\n", 111703ba3782SJens Axboe current->comm, task_pid_nr(current), inode->i_ino, 111803ba3782SJens Axboe name, inode->i_sb->s_id); 111903ba3782SJens Axboe if (dentry) { 112003ba3782SJens Axboe spin_unlock(&dentry->d_lock); 112103ba3782SJens Axboe dput(dentry); 112203ba3782SJens Axboe } 112303ba3782SJens Axboe } 112403ba3782SJens Axboe } 112503ba3782SJens Axboe 112603ba3782SJens Axboe /** 112703ba3782SJens Axboe * __mark_inode_dirty - internal function 112803ba3782SJens Axboe * @inode: inode to mark 112903ba3782SJens Axboe * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) 113003ba3782SJens Axboe * Mark an inode as dirty. Callers should use mark_inode_dirty or 113103ba3782SJens Axboe * mark_inode_dirty_sync. 113203ba3782SJens Axboe * 113303ba3782SJens Axboe * Put the inode on the super block's dirty list. 113403ba3782SJens Axboe * 113503ba3782SJens Axboe * CAREFUL! We mark it dirty unconditionally, but move it onto the 113603ba3782SJens Axboe * dirty list only if it is hashed or if it refers to a blockdev. 113703ba3782SJens Axboe * If it was not hashed, it will never be added to the dirty list 113803ba3782SJens Axboe * even if it is later hashed, as it will have been marked dirty already. 113903ba3782SJens Axboe * 114003ba3782SJens Axboe * In short, make sure you hash any inodes _before_ you start marking 114103ba3782SJens Axboe * them dirty. 114203ba3782SJens Axboe * 114303ba3782SJens Axboe * Note that for blockdevs, inode->dirtied_when represents the dirtying time of 114403ba3782SJens Axboe * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of 114503ba3782SJens Axboe * the kernel-internal blockdev inode represents the dirtying time of the 114603ba3782SJens Axboe * blockdev's pages. This is why for I_DIRTY_PAGES we always use 114703ba3782SJens Axboe * page->mapping->host, so the page-dirtying time is recorded in the internal 114803ba3782SJens Axboe * blockdev inode. 114903ba3782SJens Axboe */ 115003ba3782SJens Axboe void __mark_inode_dirty(struct inode *inode, int flags) 115103ba3782SJens Axboe { 115203ba3782SJens Axboe struct super_block *sb = inode->i_sb; 1153253c34e9SArtem Bityutskiy struct backing_dev_info *bdi = NULL; 115403ba3782SJens Axboe 115503ba3782SJens Axboe /* 115603ba3782SJens Axboe * Don't do this for I_DIRTY_PAGES - that doesn't actually 115703ba3782SJens Axboe * dirty the inode itself 115803ba3782SJens Axboe */ 115903ba3782SJens Axboe if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { 116003ba3782SJens Axboe if (sb->s_op->dirty_inode) 1161aa385729SChristoph Hellwig sb->s_op->dirty_inode(inode, flags); 116203ba3782SJens Axboe } 116303ba3782SJens Axboe 116403ba3782SJens Axboe /* 116503ba3782SJens Axboe * make sure that changes are seen by all cpus before we test i_state 116603ba3782SJens Axboe * -- mikulas 116703ba3782SJens Axboe */ 116803ba3782SJens Axboe smp_mb(); 116903ba3782SJens Axboe 117003ba3782SJens Axboe /* avoid the locking if we can */ 117103ba3782SJens Axboe if ((inode->i_state & flags) == flags) 117203ba3782SJens Axboe return; 117303ba3782SJens Axboe 117403ba3782SJens Axboe if (unlikely(block_dump)) 117503ba3782SJens Axboe block_dump___mark_inode_dirty(inode); 117603ba3782SJens Axboe 1177250df6edSDave Chinner spin_lock(&inode->i_lock); 117803ba3782SJens Axboe if ((inode->i_state & flags) != flags) { 117903ba3782SJens Axboe const int was_dirty = inode->i_state & I_DIRTY; 118003ba3782SJens Axboe 118103ba3782SJens Axboe inode->i_state |= flags; 118203ba3782SJens Axboe 118303ba3782SJens Axboe /* 118403ba3782SJens Axboe * If the inode is being synced, just update its dirty state. 118503ba3782SJens Axboe * The unlocker will place the inode on the appropriate 118603ba3782SJens Axboe * superblock list, based upon its state. 118703ba3782SJens Axboe */ 118803ba3782SJens Axboe if (inode->i_state & I_SYNC) 1189250df6edSDave Chinner goto out_unlock_inode; 119003ba3782SJens Axboe 119103ba3782SJens Axboe /* 119203ba3782SJens Axboe * Only add valid (hashed) inodes to the superblock's 119303ba3782SJens Axboe * dirty list. Add blockdev inodes as well. 119403ba3782SJens Axboe */ 119503ba3782SJens Axboe if (!S_ISBLK(inode->i_mode)) { 11961d3382cbSAl Viro if (inode_unhashed(inode)) 1197250df6edSDave Chinner goto out_unlock_inode; 119803ba3782SJens Axboe } 1199a4ffdde6SAl Viro if (inode->i_state & I_FREEING) 1200250df6edSDave Chinner goto out_unlock_inode; 120103ba3782SJens Axboe 120203ba3782SJens Axboe /* 120303ba3782SJens Axboe * If the inode was already on b_dirty/b_io/b_more_io, don't 120403ba3782SJens Axboe * reposition it (that would break b_dirty time-ordering). 120503ba3782SJens Axboe */ 120603ba3782SJens Axboe if (!was_dirty) { 1207a66979abSDave Chinner bool wakeup_bdi = false; 1208253c34e9SArtem Bityutskiy bdi = inode_to_bdi(inode); 1209500b067cSJens Axboe 1210253c34e9SArtem Bityutskiy if (bdi_cap_writeback_dirty(bdi)) { 1211253c34e9SArtem Bityutskiy WARN(!test_bit(BDI_registered, &bdi->state), 1212253c34e9SArtem Bityutskiy "bdi-%s not registered\n", bdi->name); 1213253c34e9SArtem Bityutskiy 1214253c34e9SArtem Bityutskiy /* 1215253c34e9SArtem Bityutskiy * If this is the first dirty inode for this 1216253c34e9SArtem Bityutskiy * bdi, we have to wake-up the corresponding 1217253c34e9SArtem Bityutskiy * bdi thread to make sure background 1218253c34e9SArtem Bityutskiy * write-back happens later. 1219253c34e9SArtem Bityutskiy */ 1220253c34e9SArtem Bityutskiy if (!wb_has_dirty_io(&bdi->wb)) 1221253c34e9SArtem Bityutskiy wakeup_bdi = true; 1222500b067cSJens Axboe } 122303ba3782SJens Axboe 1224a66979abSDave Chinner spin_unlock(&inode->i_lock); 1225f758eeabSChristoph Hellwig spin_lock(&bdi->wb.list_lock); 122603ba3782SJens Axboe inode->dirtied_when = jiffies; 12277ccf19a8SNick Piggin list_move(&inode->i_wb_list, &bdi->wb.b_dirty); 1228f758eeabSChristoph Hellwig spin_unlock(&bdi->wb.list_lock); 1229253c34e9SArtem Bityutskiy 1230253c34e9SArtem Bityutskiy if (wakeup_bdi) 12316467716aSArtem Bityutskiy bdi_wakeup_thread_delayed(bdi); 1232a66979abSDave Chinner return; 1233a66979abSDave Chinner } 1234a66979abSDave Chinner } 1235a66979abSDave Chinner out_unlock_inode: 1236a66979abSDave Chinner spin_unlock(&inode->i_lock); 1237a66979abSDave Chinner 123803ba3782SJens Axboe } 123903ba3782SJens Axboe EXPORT_SYMBOL(__mark_inode_dirty); 124003ba3782SJens Axboe 1241b6e51316SJens Axboe static void wait_sb_inodes(struct super_block *sb) 124266f3b8e2SJens Axboe { 124338f21977SNick Piggin struct inode *inode, *old_inode = NULL; 124438f21977SNick Piggin 124503ba3782SJens Axboe /* 124603ba3782SJens Axboe * We need to be protected against the filesystem going from 124703ba3782SJens Axboe * r/o to r/w or vice versa. 124803ba3782SJens Axboe */ 1249b6e51316SJens Axboe WARN_ON(!rwsem_is_locked(&sb->s_umount)); 125003ba3782SJens Axboe 125155fa6091SDave Chinner spin_lock(&inode_sb_list_lock); 125266f3b8e2SJens Axboe 125338f21977SNick Piggin /* 125438f21977SNick Piggin * Data integrity sync. Must wait for all pages under writeback, 125538f21977SNick Piggin * because there may have been pages dirtied before our sync 125638f21977SNick Piggin * call, but which had writeout started before we write it out. 125738f21977SNick Piggin * In which case, the inode may not be on the dirty list, but 125838f21977SNick Piggin * we still have to wait for that writeout. 125938f21977SNick Piggin */ 1260b6e51316SJens Axboe list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { 1261250df6edSDave Chinner struct address_space *mapping = inode->i_mapping; 126238f21977SNick Piggin 1263250df6edSDave Chinner spin_lock(&inode->i_lock); 1264250df6edSDave Chinner if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) || 1265250df6edSDave Chinner (mapping->nrpages == 0)) { 1266250df6edSDave Chinner spin_unlock(&inode->i_lock); 126738f21977SNick Piggin continue; 1268250df6edSDave Chinner } 126938f21977SNick Piggin __iget(inode); 1270250df6edSDave Chinner spin_unlock(&inode->i_lock); 127155fa6091SDave Chinner spin_unlock(&inode_sb_list_lock); 127255fa6091SDave Chinner 127338f21977SNick Piggin /* 127455fa6091SDave Chinner * We hold a reference to 'inode' so it couldn't have been 127555fa6091SDave Chinner * removed from s_inodes list while we dropped the 127655fa6091SDave Chinner * inode_sb_list_lock. We cannot iput the inode now as we can 127755fa6091SDave Chinner * be holding the last reference and we cannot iput it under 127855fa6091SDave Chinner * inode_sb_list_lock. So we keep the reference and iput it 127955fa6091SDave Chinner * later. 128038f21977SNick Piggin */ 128138f21977SNick Piggin iput(old_inode); 128238f21977SNick Piggin old_inode = inode; 128338f21977SNick Piggin 128438f21977SNick Piggin filemap_fdatawait(mapping); 128538f21977SNick Piggin 128638f21977SNick Piggin cond_resched(); 128738f21977SNick Piggin 128855fa6091SDave Chinner spin_lock(&inode_sb_list_lock); 128938f21977SNick Piggin } 129055fa6091SDave Chinner spin_unlock(&inode_sb_list_lock); 129138f21977SNick Piggin iput(old_inode); 129266f3b8e2SJens Axboe } 12931da177e4SLinus Torvalds 1294d8a8559cSJens Axboe /** 12953259f8beSChris Mason * writeback_inodes_sb_nr - writeback dirty inodes from given super_block 1296d8a8559cSJens Axboe * @sb: the superblock 12973259f8beSChris Mason * @nr: the number of pages to write 1298786228abSMarcos Paulo de Souza * @reason: reason why some writeback work initiated 12991da177e4SLinus Torvalds * 1300d8a8559cSJens Axboe * Start writeback on some inodes on this super_block. No guarantees are made 1301d8a8559cSJens Axboe * on how many (if any) will be written, and this function does not wait 13023259f8beSChris Mason * for IO completion of submitted IO. 13031da177e4SLinus Torvalds */ 13040e175a18SCurt Wohlgemuth void writeback_inodes_sb_nr(struct super_block *sb, 13050e175a18SCurt Wohlgemuth unsigned long nr, 13060e175a18SCurt Wohlgemuth enum wb_reason reason) 13071da177e4SLinus Torvalds { 130883ba7b07SChristoph Hellwig DECLARE_COMPLETION_ONSTACK(done); 130983ba7b07SChristoph Hellwig struct wb_writeback_work work = { 13103c4d7165SChristoph Hellwig .sb = sb, 13113c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_NONE, 13126e6938b6SWu Fengguang .tagged_writepages = 1, 131383ba7b07SChristoph Hellwig .done = &done, 13143259f8beSChris Mason .nr_pages = nr, 13150e175a18SCurt Wohlgemuth .reason = reason, 13163c4d7165SChristoph Hellwig }; 13170e3c9a22SJens Axboe 13186eedc701SJan Kara if (sb->s_bdi == &noop_backing_dev_info) 13196eedc701SJan Kara return; 1320cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 132183ba7b07SChristoph Hellwig bdi_queue_work(sb->s_bdi, &work); 132283ba7b07SChristoph Hellwig wait_for_completion(&done); 13231da177e4SLinus Torvalds } 13243259f8beSChris Mason EXPORT_SYMBOL(writeback_inodes_sb_nr); 13253259f8beSChris Mason 13263259f8beSChris Mason /** 13273259f8beSChris Mason * writeback_inodes_sb - writeback dirty inodes from given super_block 13283259f8beSChris Mason * @sb: the superblock 1329786228abSMarcos Paulo de Souza * @reason: reason why some writeback work was initiated 13303259f8beSChris Mason * 13313259f8beSChris Mason * Start writeback on some inodes on this super_block. No guarantees are made 13323259f8beSChris Mason * on how many (if any) will be written, and this function does not wait 13333259f8beSChris Mason * for IO completion of submitted IO. 13343259f8beSChris Mason */ 13350e175a18SCurt Wohlgemuth void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) 13363259f8beSChris Mason { 13370e175a18SCurt Wohlgemuth return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); 13383259f8beSChris Mason } 1339d8a8559cSJens Axboe EXPORT_SYMBOL(writeback_inodes_sb); 1340d8a8559cSJens Axboe 1341d8a8559cSJens Axboe /** 134217bd55d0SEric Sandeen * writeback_inodes_sb_if_idle - start writeback if none underway 134317bd55d0SEric Sandeen * @sb: the superblock 1344786228abSMarcos Paulo de Souza * @reason: reason why some writeback work was initiated 134517bd55d0SEric Sandeen * 134617bd55d0SEric Sandeen * Invoke writeback_inodes_sb if no writeback is currently underway. 134717bd55d0SEric Sandeen * Returns 1 if writeback was started, 0 if not. 134817bd55d0SEric Sandeen */ 13490e175a18SCurt Wohlgemuth int writeback_inodes_sb_if_idle(struct super_block *sb, enum wb_reason reason) 135017bd55d0SEric Sandeen { 135117bd55d0SEric Sandeen if (!writeback_in_progress(sb->s_bdi)) { 1352cf37e972SChristoph Hellwig down_read(&sb->s_umount); 13530e175a18SCurt Wohlgemuth writeback_inodes_sb(sb, reason); 1354cf37e972SChristoph Hellwig up_read(&sb->s_umount); 135517bd55d0SEric Sandeen return 1; 135617bd55d0SEric Sandeen } else 135717bd55d0SEric Sandeen return 0; 135817bd55d0SEric Sandeen } 135917bd55d0SEric Sandeen EXPORT_SYMBOL(writeback_inodes_sb_if_idle); 136017bd55d0SEric Sandeen 136117bd55d0SEric Sandeen /** 1362c097b2caSFengguang Wu * writeback_inodes_sb_nr_if_idle - start writeback if none underway 13633259f8beSChris Mason * @sb: the superblock 13643259f8beSChris Mason * @nr: the number of pages to write 1365786228abSMarcos Paulo de Souza * @reason: reason why some writeback work was initiated 13663259f8beSChris Mason * 13673259f8beSChris Mason * Invoke writeback_inodes_sb if no writeback is currently underway. 13683259f8beSChris Mason * Returns 1 if writeback was started, 0 if not. 13693259f8beSChris Mason */ 13703259f8beSChris Mason int writeback_inodes_sb_nr_if_idle(struct super_block *sb, 13710e175a18SCurt Wohlgemuth unsigned long nr, 13720e175a18SCurt Wohlgemuth enum wb_reason reason) 13733259f8beSChris Mason { 13743259f8beSChris Mason if (!writeback_in_progress(sb->s_bdi)) { 13753259f8beSChris Mason down_read(&sb->s_umount); 13760e175a18SCurt Wohlgemuth writeback_inodes_sb_nr(sb, nr, reason); 13773259f8beSChris Mason up_read(&sb->s_umount); 13783259f8beSChris Mason return 1; 13793259f8beSChris Mason } else 13803259f8beSChris Mason return 0; 13813259f8beSChris Mason } 13823259f8beSChris Mason EXPORT_SYMBOL(writeback_inodes_sb_nr_if_idle); 13833259f8beSChris Mason 13843259f8beSChris Mason /** 1385d8a8559cSJens Axboe * sync_inodes_sb - sync sb inode pages 1386d8a8559cSJens Axboe * @sb: the superblock 1387d8a8559cSJens Axboe * 1388d8a8559cSJens Axboe * This function writes and waits on any dirty inode belonging to this 1389cb9ef8d5SStefan Hajnoczi * super_block. 1390d8a8559cSJens Axboe */ 1391b6e51316SJens Axboe void sync_inodes_sb(struct super_block *sb) 1392d8a8559cSJens Axboe { 139383ba7b07SChristoph Hellwig DECLARE_COMPLETION_ONSTACK(done); 139483ba7b07SChristoph Hellwig struct wb_writeback_work work = { 13953c4d7165SChristoph Hellwig .sb = sb, 13963c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_ALL, 13973c4d7165SChristoph Hellwig .nr_pages = LONG_MAX, 13983c4d7165SChristoph Hellwig .range_cyclic = 0, 139983ba7b07SChristoph Hellwig .done = &done, 14000e175a18SCurt Wohlgemuth .reason = WB_REASON_SYNC, 14013c4d7165SChristoph Hellwig }; 14023c4d7165SChristoph Hellwig 14036eedc701SJan Kara /* Nothing to do? */ 14046eedc701SJan Kara if (sb->s_bdi == &noop_backing_dev_info) 14056eedc701SJan Kara return; 1406cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 1407cf37e972SChristoph Hellwig 140883ba7b07SChristoph Hellwig bdi_queue_work(sb->s_bdi, &work); 140983ba7b07SChristoph Hellwig wait_for_completion(&done); 141083ba7b07SChristoph Hellwig 1411b6e51316SJens Axboe wait_sb_inodes(sb); 1412d8a8559cSJens Axboe } 1413d8a8559cSJens Axboe EXPORT_SYMBOL(sync_inodes_sb); 14141da177e4SLinus Torvalds 14151da177e4SLinus Torvalds /** 14161da177e4SLinus Torvalds * write_inode_now - write an inode to disk 14171da177e4SLinus Torvalds * @inode: inode to write to disk 14181da177e4SLinus Torvalds * @sync: whether the write should be synchronous or not 14191da177e4SLinus Torvalds * 14207f04c26dSAndrea Arcangeli * This function commits an inode to disk immediately if it is dirty. This is 14217f04c26dSAndrea Arcangeli * primarily needed by knfsd. 14227f04c26dSAndrea Arcangeli * 14237f04c26dSAndrea Arcangeli * The caller must either have a ref on the inode or must have set I_WILL_FREE. 14241da177e4SLinus Torvalds */ 14251da177e4SLinus Torvalds int write_inode_now(struct inode *inode, int sync) 14261da177e4SLinus Torvalds { 1427f758eeabSChristoph Hellwig struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; 14281da177e4SLinus Torvalds struct writeback_control wbc = { 14291da177e4SLinus Torvalds .nr_to_write = LONG_MAX, 143018914b18SMike Galbraith .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, 1431111ebb6eSOGAWA Hirofumi .range_start = 0, 1432111ebb6eSOGAWA Hirofumi .range_end = LLONG_MAX, 14331da177e4SLinus Torvalds }; 14341da177e4SLinus Torvalds 14351da177e4SLinus Torvalds if (!mapping_cap_writeback_dirty(inode->i_mapping)) 143649364ce2SAndrew Morton wbc.nr_to_write = 0; 14371da177e4SLinus Torvalds 14381da177e4SLinus Torvalds might_sleep(); 14394f8ad655SJan Kara return writeback_single_inode(inode, wb, &wbc); 14401da177e4SLinus Torvalds } 14411da177e4SLinus Torvalds EXPORT_SYMBOL(write_inode_now); 14421da177e4SLinus Torvalds 14431da177e4SLinus Torvalds /** 14441da177e4SLinus Torvalds * sync_inode - write an inode and its pages to disk. 14451da177e4SLinus Torvalds * @inode: the inode to sync 14461da177e4SLinus Torvalds * @wbc: controls the writeback mode 14471da177e4SLinus Torvalds * 14481da177e4SLinus Torvalds * sync_inode() will write an inode and its pages to disk. It will also 14491da177e4SLinus Torvalds * correctly update the inode on its superblock's dirty inode lists and will 14501da177e4SLinus Torvalds * update inode->i_state. 14511da177e4SLinus Torvalds * 14521da177e4SLinus Torvalds * The caller must have a ref on the inode. 14531da177e4SLinus Torvalds */ 14541da177e4SLinus Torvalds int sync_inode(struct inode *inode, struct writeback_control *wbc) 14551da177e4SLinus Torvalds { 14564f8ad655SJan Kara return writeback_single_inode(inode, &inode_to_bdi(inode)->wb, wbc); 14571da177e4SLinus Torvalds } 14581da177e4SLinus Torvalds EXPORT_SYMBOL(sync_inode); 1459c3765016SChristoph Hellwig 1460c3765016SChristoph Hellwig /** 1461c691b9d9SAndrew Morton * sync_inode_metadata - write an inode to disk 1462c3765016SChristoph Hellwig * @inode: the inode to sync 1463c3765016SChristoph Hellwig * @wait: wait for I/O to complete. 1464c3765016SChristoph Hellwig * 1465c691b9d9SAndrew Morton * Write an inode to disk and adjust its dirty state after completion. 1466c3765016SChristoph Hellwig * 1467c3765016SChristoph Hellwig * Note: only writes the actual inode, no associated data or other metadata. 1468c3765016SChristoph Hellwig */ 1469c3765016SChristoph Hellwig int sync_inode_metadata(struct inode *inode, int wait) 1470c3765016SChristoph Hellwig { 1471c3765016SChristoph Hellwig struct writeback_control wbc = { 1472c3765016SChristoph Hellwig .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE, 1473c3765016SChristoph Hellwig .nr_to_write = 0, /* metadata-only */ 1474c3765016SChristoph Hellwig }; 1475c3765016SChristoph Hellwig 1476c3765016SChristoph Hellwig return sync_inode(inode, &wbc); 1477c3765016SChristoph Hellwig } 1478c3765016SChristoph Hellwig EXPORT_SYMBOL(sync_inode_metadata); 1479