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 55f11b00f3SAdrian Bunk /** 56f11b00f3SAdrian Bunk * writeback_in_progress - determine whether there is writeback in progress 57f11b00f3SAdrian Bunk * @bdi: the device's backing_dev_info structure. 58f11b00f3SAdrian Bunk * 5903ba3782SJens Axboe * Determine whether there is writeback waiting to be handled against a 6003ba3782SJens Axboe * backing device. 61f11b00f3SAdrian Bunk */ 62f11b00f3SAdrian Bunk int writeback_in_progress(struct backing_dev_info *bdi) 63f11b00f3SAdrian Bunk { 6481d73a32SJan Kara return test_bit(BDI_writeback_running, &bdi->state); 65f11b00f3SAdrian Bunk } 66f11b00f3SAdrian Bunk 67692ebd17SJan Kara static inline struct backing_dev_info *inode_to_bdi(struct inode *inode) 68692ebd17SJan Kara { 69692ebd17SJan Kara struct super_block *sb = inode->i_sb; 70692ebd17SJan Kara 71aaead25bSChristoph Hellwig if (strcmp(sb->s_type->name, "bdev") == 0) 72aaead25bSChristoph Hellwig return inode->i_mapping->backing_dev_info; 73aaead25bSChristoph Hellwig 74692ebd17SJan Kara return sb->s_bdi; 75692ebd17SJan Kara } 76692ebd17SJan Kara 777ccf19a8SNick Piggin static inline struct inode *wb_inode(struct list_head *head) 787ccf19a8SNick Piggin { 797ccf19a8SNick Piggin return list_entry(head, struct inode, i_wb_list); 807ccf19a8SNick Piggin } 817ccf19a8SNick Piggin 8215eb77a0SWu Fengguang /* 8315eb77a0SWu Fengguang * Include the creation of the trace points after defining the 8415eb77a0SWu Fengguang * wb_writeback_work structure and inline functions so that the definition 8515eb77a0SWu Fengguang * remains local to this file. 8615eb77a0SWu Fengguang */ 8715eb77a0SWu Fengguang #define CREATE_TRACE_POINTS 8815eb77a0SWu Fengguang #include <trace/events/writeback.h> 8915eb77a0SWu Fengguang 906585027aSJan Kara /* Wakeup flusher thread or forker thread to fork it. Requires bdi->wb_lock. */ 916585027aSJan Kara static void bdi_wakeup_flusher(struct backing_dev_info *bdi) 924195f73dSNick Piggin { 93fff5b85aSArtem Bityutskiy if (bdi->wb.task) { 94fff5b85aSArtem Bityutskiy wake_up_process(bdi->wb.task); 95fff5b85aSArtem Bityutskiy } else { 961da177e4SLinus Torvalds /* 97fff5b85aSArtem Bityutskiy * The bdi thread isn't there, wake up the forker thread which 98fff5b85aSArtem Bityutskiy * will create and run it. 991da177e4SLinus Torvalds */ 10003ba3782SJens Axboe wake_up_process(default_backing_dev_info.wb.task); 1011da177e4SLinus Torvalds } 1026585027aSJan Kara } 1036585027aSJan Kara 1046585027aSJan Kara static void bdi_queue_work(struct backing_dev_info *bdi, 1056585027aSJan Kara struct wb_writeback_work *work) 1066585027aSJan Kara { 1076585027aSJan Kara trace_writeback_queue(bdi, work); 1086585027aSJan Kara 1096585027aSJan Kara spin_lock_bh(&bdi->wb_lock); 1106585027aSJan Kara list_add_tail(&work->list, &bdi->work_list); 1116585027aSJan Kara if (!bdi->wb.task) 1126585027aSJan Kara trace_writeback_nothread(bdi, work); 1136585027aSJan Kara bdi_wakeup_flusher(bdi); 1146467716aSArtem Bityutskiy spin_unlock_bh(&bdi->wb_lock); 11503ba3782SJens Axboe } 1161da177e4SLinus Torvalds 11783ba7b07SChristoph Hellwig static void 11883ba7b07SChristoph Hellwig __bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, 1190e175a18SCurt Wohlgemuth bool range_cyclic, enum wb_reason reason) 1201da177e4SLinus Torvalds { 12183ba7b07SChristoph Hellwig struct wb_writeback_work *work; 12203ba3782SJens Axboe 123bcddc3f0SJens Axboe /* 124bcddc3f0SJens Axboe * This is WB_SYNC_NONE writeback, so if allocation fails just 125bcddc3f0SJens Axboe * wakeup the thread for old dirty data writeback 126bcddc3f0SJens Axboe */ 12783ba7b07SChristoph Hellwig work = kzalloc(sizeof(*work), GFP_ATOMIC); 12883ba7b07SChristoph Hellwig if (!work) { 129455b2864SDave Chinner if (bdi->wb.task) { 130455b2864SDave Chinner trace_writeback_nowork(bdi); 13183ba7b07SChristoph Hellwig wake_up_process(bdi->wb.task); 132455b2864SDave Chinner } 13383ba7b07SChristoph Hellwig return; 13483ba7b07SChristoph Hellwig } 13583ba7b07SChristoph Hellwig 13683ba7b07SChristoph Hellwig work->sync_mode = WB_SYNC_NONE; 13783ba7b07SChristoph Hellwig work->nr_pages = nr_pages; 13883ba7b07SChristoph Hellwig work->range_cyclic = range_cyclic; 1390e175a18SCurt Wohlgemuth work->reason = reason; 14083ba7b07SChristoph Hellwig 141f11fcae8SJens Axboe bdi_queue_work(bdi, work); 14203ba3782SJens Axboe } 143b6e51316SJens Axboe 144b6e51316SJens Axboe /** 145b6e51316SJens Axboe * bdi_start_writeback - start writeback 146b6e51316SJens Axboe * @bdi: the backing device to write from 147b6e51316SJens Axboe * @nr_pages: the number of pages to write 148786228abSMarcos Paulo de Souza * @reason: reason why some writeback work was initiated 149b6e51316SJens Axboe * 150b6e51316SJens Axboe * Description: 151b6e51316SJens Axboe * This does WB_SYNC_NONE opportunistic writeback. The IO is only 15225985edcSLucas De Marchi * started when this function returns, we make no guarantees on 1530e3c9a22SJens Axboe * completion. Caller need not hold sb s_umount semaphore. 154b6e51316SJens Axboe * 155b6e51316SJens Axboe */ 1560e175a18SCurt Wohlgemuth void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, 1570e175a18SCurt Wohlgemuth enum wb_reason reason) 158b6e51316SJens Axboe { 1590e175a18SCurt Wohlgemuth __bdi_start_writeback(bdi, nr_pages, true, reason); 160d3ddec76SWu Fengguang } 161d3ddec76SWu Fengguang 162c5444198SChristoph Hellwig /** 163c5444198SChristoph Hellwig * bdi_start_background_writeback - start background writeback 164c5444198SChristoph Hellwig * @bdi: the backing device to write from 165c5444198SChristoph Hellwig * 166c5444198SChristoph Hellwig * Description: 1676585027aSJan Kara * This makes sure WB_SYNC_NONE background writeback happens. When 1686585027aSJan Kara * this function returns, it is only guaranteed that for given BDI 1696585027aSJan Kara * some IO is happening if we are over background dirty threshold. 1706585027aSJan Kara * Caller need not hold sb s_umount semaphore. 171c5444198SChristoph Hellwig */ 172c5444198SChristoph Hellwig void bdi_start_background_writeback(struct backing_dev_info *bdi) 173c5444198SChristoph Hellwig { 1746585027aSJan Kara /* 1756585027aSJan Kara * We just wake up the flusher thread. It will perform background 1766585027aSJan Kara * writeback as soon as there is no other work to do. 1776585027aSJan Kara */ 17871927e84SWu Fengguang trace_writeback_wake_background(bdi); 1796585027aSJan Kara spin_lock_bh(&bdi->wb_lock); 1806585027aSJan Kara bdi_wakeup_flusher(bdi); 1816585027aSJan Kara spin_unlock_bh(&bdi->wb_lock); 1821da177e4SLinus Torvalds } 1831da177e4SLinus Torvalds 1841da177e4SLinus Torvalds /* 185a66979abSDave Chinner * Remove the inode from the writeback list it is on. 186a66979abSDave Chinner */ 187a66979abSDave Chinner void inode_wb_list_del(struct inode *inode) 188a66979abSDave Chinner { 189f758eeabSChristoph Hellwig struct backing_dev_info *bdi = inode_to_bdi(inode); 190a66979abSDave Chinner 191f758eeabSChristoph Hellwig spin_lock(&bdi->wb.list_lock); 192f758eeabSChristoph Hellwig list_del_init(&inode->i_wb_list); 193f758eeabSChristoph Hellwig spin_unlock(&bdi->wb.list_lock); 194f758eeabSChristoph Hellwig } 195a66979abSDave Chinner 196a66979abSDave Chinner /* 1976610a0bcSAndrew Morton * Redirty an inode: set its when-it-was dirtied timestamp and move it to the 1986610a0bcSAndrew Morton * furthest end of its superblock's dirty-inode list. 1996610a0bcSAndrew Morton * 2006610a0bcSAndrew Morton * Before stamping the inode's ->dirtied_when, we check to see whether it is 20166f3b8e2SJens Axboe * already the most-recently-dirtied inode on the b_dirty list. If that is 2026610a0bcSAndrew Morton * the case then the inode must have been redirtied while it was being written 2036610a0bcSAndrew Morton * out and we don't reset its dirtied_when. 2046610a0bcSAndrew Morton */ 205f758eeabSChristoph Hellwig static void redirty_tail(struct inode *inode, struct bdi_writeback *wb) 2066610a0bcSAndrew Morton { 207f758eeabSChristoph Hellwig assert_spin_locked(&wb->list_lock); 20803ba3782SJens Axboe if (!list_empty(&wb->b_dirty)) { 20966f3b8e2SJens Axboe struct inode *tail; 2106610a0bcSAndrew Morton 2117ccf19a8SNick Piggin tail = wb_inode(wb->b_dirty.next); 21266f3b8e2SJens Axboe if (time_before(inode->dirtied_when, tail->dirtied_when)) 2136610a0bcSAndrew Morton inode->dirtied_when = jiffies; 2146610a0bcSAndrew Morton } 2157ccf19a8SNick Piggin list_move(&inode->i_wb_list, &wb->b_dirty); 2166610a0bcSAndrew Morton } 2176610a0bcSAndrew Morton 2186610a0bcSAndrew Morton /* 21966f3b8e2SJens Axboe * requeue inode for re-scanning after bdi->b_io list is exhausted. 220c986d1e2SAndrew Morton */ 221f758eeabSChristoph Hellwig static void requeue_io(struct inode *inode, struct bdi_writeback *wb) 222c986d1e2SAndrew Morton { 223f758eeabSChristoph Hellwig assert_spin_locked(&wb->list_lock); 2247ccf19a8SNick Piggin list_move(&inode->i_wb_list, &wb->b_more_io); 225c986d1e2SAndrew Morton } 226c986d1e2SAndrew Morton 2271c0eeaf5SJoern Engel static void inode_sync_complete(struct inode *inode) 2281c0eeaf5SJoern Engel { 229365b94aeSJan Kara inode->i_state &= ~I_SYNC; 230365b94aeSJan Kara /* Waiters must see I_SYNC cleared before being woken up */ 2311c0eeaf5SJoern Engel smp_mb(); 2321c0eeaf5SJoern Engel wake_up_bit(&inode->i_state, __I_SYNC); 2331c0eeaf5SJoern Engel } 2341c0eeaf5SJoern Engel 235d2caa3c5SJeff Layton static bool inode_dirtied_after(struct inode *inode, unsigned long t) 236d2caa3c5SJeff Layton { 237d2caa3c5SJeff Layton bool ret = time_after(inode->dirtied_when, t); 238d2caa3c5SJeff Layton #ifndef CONFIG_64BIT 239d2caa3c5SJeff Layton /* 240d2caa3c5SJeff Layton * For inodes being constantly redirtied, dirtied_when can get stuck. 241d2caa3c5SJeff Layton * It _appears_ to be in the future, but is actually in distant past. 242d2caa3c5SJeff Layton * This test is necessary to prevent such wrapped-around relative times 2435b0830cbSJens Axboe * from permanently stopping the whole bdi writeback. 244d2caa3c5SJeff Layton */ 245d2caa3c5SJeff Layton ret = ret && time_before_eq(inode->dirtied_when, jiffies); 246d2caa3c5SJeff Layton #endif 247d2caa3c5SJeff Layton return ret; 248d2caa3c5SJeff Layton } 249d2caa3c5SJeff Layton 250c986d1e2SAndrew Morton /* 2510e2f2b23SWang Sheng-Hui * Move expired (dirtied before work->older_than_this) dirty inodes from 252697e6fedSJan Kara * @delaying_queue to @dispatch_queue. 2532c136579SFengguang Wu */ 254e84d0a4fSWu Fengguang static int move_expired_inodes(struct list_head *delaying_queue, 2552c136579SFengguang Wu struct list_head *dispatch_queue, 256ad4e38ddSCurt Wohlgemuth struct wb_writeback_work *work) 2572c136579SFengguang Wu { 2585c03449dSShaohua Li LIST_HEAD(tmp); 2595c03449dSShaohua Li struct list_head *pos, *node; 260cf137307SJens Axboe struct super_block *sb = NULL; 2615c03449dSShaohua Li struct inode *inode; 262cf137307SJens Axboe int do_sb_sort = 0; 263e84d0a4fSWu Fengguang int moved = 0; 2645c03449dSShaohua Li 2652c136579SFengguang Wu while (!list_empty(delaying_queue)) { 2667ccf19a8SNick Piggin inode = wb_inode(delaying_queue->prev); 267ad4e38ddSCurt Wohlgemuth if (work->older_than_this && 268ad4e38ddSCurt Wohlgemuth inode_dirtied_after(inode, *work->older_than_this)) 2692c136579SFengguang Wu break; 270cf137307SJens Axboe if (sb && sb != inode->i_sb) 271cf137307SJens Axboe do_sb_sort = 1; 272cf137307SJens Axboe sb = inode->i_sb; 2737ccf19a8SNick Piggin list_move(&inode->i_wb_list, &tmp); 274e84d0a4fSWu Fengguang moved++; 2755c03449dSShaohua Li } 2765c03449dSShaohua Li 277cf137307SJens Axboe /* just one sb in list, splice to dispatch_queue and we're done */ 278cf137307SJens Axboe if (!do_sb_sort) { 279cf137307SJens Axboe list_splice(&tmp, dispatch_queue); 280e84d0a4fSWu Fengguang goto out; 281cf137307SJens Axboe } 282cf137307SJens Axboe 2835c03449dSShaohua Li /* Move inodes from one superblock together */ 2845c03449dSShaohua Li while (!list_empty(&tmp)) { 2857ccf19a8SNick Piggin sb = wb_inode(tmp.prev)->i_sb; 2865c03449dSShaohua Li list_for_each_prev_safe(pos, node, &tmp) { 2877ccf19a8SNick Piggin inode = wb_inode(pos); 2885c03449dSShaohua Li if (inode->i_sb == sb) 2897ccf19a8SNick Piggin list_move(&inode->i_wb_list, dispatch_queue); 2902c136579SFengguang Wu } 2912c136579SFengguang Wu } 292e84d0a4fSWu Fengguang out: 293e84d0a4fSWu Fengguang return moved; 2945c03449dSShaohua Li } 2952c136579SFengguang Wu 2962c136579SFengguang Wu /* 2972c136579SFengguang Wu * Queue all expired dirty inodes for io, eldest first. 2984ea879b9SWu Fengguang * Before 2994ea879b9SWu Fengguang * newly dirtied b_dirty b_io b_more_io 3004ea879b9SWu Fengguang * =============> gf edc BA 3014ea879b9SWu Fengguang * After 3024ea879b9SWu Fengguang * newly dirtied b_dirty b_io b_more_io 3034ea879b9SWu Fengguang * =============> g fBAedc 3044ea879b9SWu Fengguang * | 3054ea879b9SWu Fengguang * +--> dequeue for IO 3062c136579SFengguang Wu */ 307ad4e38ddSCurt Wohlgemuth static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work) 3082c136579SFengguang Wu { 309e84d0a4fSWu Fengguang int moved; 310f758eeabSChristoph Hellwig assert_spin_locked(&wb->list_lock); 3114ea879b9SWu Fengguang list_splice_init(&wb->b_more_io, &wb->b_io); 312ad4e38ddSCurt Wohlgemuth moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, work); 313ad4e38ddSCurt Wohlgemuth trace_writeback_queue_io(wb, work, moved); 31466f3b8e2SJens Axboe } 31566f3b8e2SJens Axboe 316a9185b41SChristoph Hellwig static int write_inode(struct inode *inode, struct writeback_control *wbc) 31766f3b8e2SJens Axboe { 31803ba3782SJens Axboe if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) 319a9185b41SChristoph Hellwig return inode->i_sb->s_op->write_inode(inode, wbc); 32003ba3782SJens Axboe return 0; 32166f3b8e2SJens Axboe } 32208d8e974SFengguang Wu 3232c136579SFengguang Wu /* 324169ebd90SJan Kara * Wait for writeback on an inode to complete. Called with i_lock held. 325169ebd90SJan Kara * Caller must make sure inode cannot go away when we drop i_lock. 32601c03194SChristoph Hellwig */ 327169ebd90SJan Kara static void __inode_wait_for_writeback(struct inode *inode) 328169ebd90SJan Kara __releases(inode->i_lock) 329169ebd90SJan Kara __acquires(inode->i_lock) 33001c03194SChristoph Hellwig { 33101c03194SChristoph Hellwig DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); 33201c03194SChristoph Hellwig wait_queue_head_t *wqh; 33301c03194SChristoph Hellwig 33401c03194SChristoph Hellwig wqh = bit_waitqueue(&inode->i_state, __I_SYNC); 33558a9d3d8SRichard Kennedy while (inode->i_state & I_SYNC) { 336250df6edSDave Chinner spin_unlock(&inode->i_lock); 33701c03194SChristoph Hellwig __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE); 338250df6edSDave Chinner spin_lock(&inode->i_lock); 33958a9d3d8SRichard Kennedy } 34001c03194SChristoph Hellwig } 34101c03194SChristoph Hellwig 34201c03194SChristoph Hellwig /* 343169ebd90SJan Kara * Wait for writeback on an inode to complete. Caller must have inode pinned. 344169ebd90SJan Kara */ 345169ebd90SJan Kara void inode_wait_for_writeback(struct inode *inode) 346169ebd90SJan Kara { 347169ebd90SJan Kara spin_lock(&inode->i_lock); 348169ebd90SJan Kara __inode_wait_for_writeback(inode); 349169ebd90SJan Kara spin_unlock(&inode->i_lock); 350169ebd90SJan Kara } 351169ebd90SJan Kara 352169ebd90SJan Kara /* 353169ebd90SJan Kara * Sleep until I_SYNC is cleared. This function must be called with i_lock 354169ebd90SJan Kara * held and drops it. It is aimed for callers not holding any inode reference 355169ebd90SJan Kara * so once i_lock is dropped, inode can go away. 356169ebd90SJan Kara */ 357169ebd90SJan Kara static void inode_sleep_on_writeback(struct inode *inode) 358169ebd90SJan Kara __releases(inode->i_lock) 359169ebd90SJan Kara { 360169ebd90SJan Kara DEFINE_WAIT(wait); 361169ebd90SJan Kara wait_queue_head_t *wqh = bit_waitqueue(&inode->i_state, __I_SYNC); 362169ebd90SJan Kara int sleep; 363169ebd90SJan Kara 364169ebd90SJan Kara prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); 365169ebd90SJan Kara sleep = inode->i_state & I_SYNC; 366169ebd90SJan Kara spin_unlock(&inode->i_lock); 367169ebd90SJan Kara if (sleep) 368169ebd90SJan Kara schedule(); 369169ebd90SJan Kara finish_wait(wqh, &wait); 370169ebd90SJan Kara } 371169ebd90SJan Kara 372169ebd90SJan Kara /* 373ccb26b5aSJan Kara * Find proper writeback list for the inode depending on its current state and 374ccb26b5aSJan Kara * possibly also change of its state while we were doing writeback. Here we 375ccb26b5aSJan Kara * handle things such as livelock prevention or fairness of writeback among 376ccb26b5aSJan Kara * inodes. This function can be called only by flusher thread - noone else 377ccb26b5aSJan Kara * processes all inodes in writeback lists and requeueing inodes behind flusher 378ccb26b5aSJan Kara * thread's back can have unexpected consequences. 379ccb26b5aSJan Kara */ 380ccb26b5aSJan Kara static void requeue_inode(struct inode *inode, struct bdi_writeback *wb, 381ccb26b5aSJan Kara struct writeback_control *wbc) 382ccb26b5aSJan Kara { 383ccb26b5aSJan Kara if (inode->i_state & I_FREEING) 384ccb26b5aSJan Kara return; 385ccb26b5aSJan Kara 386ccb26b5aSJan Kara /* 387ccb26b5aSJan Kara * Sync livelock prevention. Each inode is tagged and synced in one 388ccb26b5aSJan Kara * shot. If still dirty, it will be redirty_tail()'ed below. Update 389ccb26b5aSJan Kara * the dirty time to prevent enqueue and sync it again. 390ccb26b5aSJan Kara */ 391ccb26b5aSJan Kara if ((inode->i_state & I_DIRTY) && 392ccb26b5aSJan Kara (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)) 393ccb26b5aSJan Kara inode->dirtied_when = jiffies; 394ccb26b5aSJan Kara 3954f8ad655SJan Kara if (wbc->pages_skipped) { 3964f8ad655SJan Kara /* 3974f8ad655SJan Kara * writeback is not making progress due to locked 3984f8ad655SJan Kara * buffers. Skip this inode for now. 3994f8ad655SJan Kara */ 4004f8ad655SJan Kara redirty_tail(inode, wb); 4014f8ad655SJan Kara return; 4024f8ad655SJan Kara } 4034f8ad655SJan Kara 404ccb26b5aSJan Kara if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) { 405ccb26b5aSJan Kara /* 406ccb26b5aSJan Kara * We didn't write back all the pages. nfs_writepages() 407ccb26b5aSJan Kara * sometimes bales out without doing anything. 408ccb26b5aSJan Kara */ 409ccb26b5aSJan Kara if (wbc->nr_to_write <= 0) { 410ccb26b5aSJan Kara /* Slice used up. Queue for next turn. */ 411ccb26b5aSJan Kara requeue_io(inode, wb); 412ccb26b5aSJan Kara } else { 413ccb26b5aSJan Kara /* 414ccb26b5aSJan Kara * Writeback blocked by something other than 415ccb26b5aSJan Kara * congestion. Delay the inode for some time to 416ccb26b5aSJan Kara * avoid spinning on the CPU (100% iowait) 417ccb26b5aSJan Kara * retrying writeback of the dirty page/inode 418ccb26b5aSJan Kara * that cannot be performed immediately. 419ccb26b5aSJan Kara */ 420ccb26b5aSJan Kara redirty_tail(inode, wb); 421ccb26b5aSJan Kara } 422ccb26b5aSJan Kara } else if (inode->i_state & I_DIRTY) { 423ccb26b5aSJan Kara /* 424ccb26b5aSJan Kara * Filesystems can dirty the inode during writeback operations, 425ccb26b5aSJan Kara * such as delayed allocation during submission or metadata 426ccb26b5aSJan Kara * updates after data IO completion. 427ccb26b5aSJan Kara */ 428ccb26b5aSJan Kara redirty_tail(inode, wb); 429ccb26b5aSJan Kara } else { 430ccb26b5aSJan Kara /* The inode is clean. Remove from writeback lists. */ 431ccb26b5aSJan Kara list_del_init(&inode->i_wb_list); 432ccb26b5aSJan Kara } 433ccb26b5aSJan Kara } 434ccb26b5aSJan Kara 435ccb26b5aSJan Kara /* 4364f8ad655SJan Kara * Write out an inode and its dirty pages. Do not update the writeback list 4374f8ad655SJan Kara * linkage. That is left to the caller. The caller is also responsible for 4384f8ad655SJan Kara * setting I_SYNC flag and calling inode_sync_complete() to clear it. 4391da177e4SLinus Torvalds */ 4401da177e4SLinus Torvalds static int 4414f8ad655SJan Kara __writeback_single_inode(struct inode *inode, struct bdi_writeback *wb, 442f758eeabSChristoph Hellwig struct writeback_control *wbc) 4431da177e4SLinus Torvalds { 4441da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 445251d6a47SWu Fengguang long nr_to_write = wbc->nr_to_write; 44601c03194SChristoph Hellwig unsigned dirty; 4471da177e4SLinus Torvalds int ret; 4481da177e4SLinus Torvalds 4494f8ad655SJan Kara WARN_ON(!(inode->i_state & I_SYNC)); 4501da177e4SLinus Torvalds 4511da177e4SLinus Torvalds ret = do_writepages(mapping, wbc); 4521da177e4SLinus Torvalds 45326821ed4SChristoph Hellwig /* 45426821ed4SChristoph Hellwig * Make sure to wait on the data before writing out the metadata. 45526821ed4SChristoph Hellwig * This is important for filesystems that modify metadata on data 45626821ed4SChristoph Hellwig * I/O completion. 45726821ed4SChristoph Hellwig */ 458a9185b41SChristoph Hellwig if (wbc->sync_mode == WB_SYNC_ALL) { 45926821ed4SChristoph Hellwig int err = filemap_fdatawait(mapping); 4601da177e4SLinus Torvalds if (ret == 0) 4611da177e4SLinus Torvalds ret = err; 4621da177e4SLinus Torvalds } 4631da177e4SLinus Torvalds 4645547e8aaSDmitry Monakhov /* 4655547e8aaSDmitry Monakhov * Some filesystems may redirty the inode during the writeback 4665547e8aaSDmitry Monakhov * due to delalloc, clear dirty metadata flags right before 4675547e8aaSDmitry Monakhov * write_inode() 4685547e8aaSDmitry Monakhov */ 469250df6edSDave Chinner spin_lock(&inode->i_lock); 4706290be1cSJan Kara /* Clear I_DIRTY_PAGES if we've written out all dirty pages */ 4716290be1cSJan Kara if (!mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) 4726290be1cSJan Kara inode->i_state &= ~I_DIRTY_PAGES; 4735547e8aaSDmitry Monakhov dirty = inode->i_state & I_DIRTY; 4745547e8aaSDmitry Monakhov inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC); 475250df6edSDave Chinner spin_unlock(&inode->i_lock); 47626821ed4SChristoph Hellwig /* Don't write the inode if only I_DIRTY_PAGES was set */ 47726821ed4SChristoph Hellwig if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { 478a9185b41SChristoph Hellwig int err = write_inode(inode, wbc); 4791da177e4SLinus Torvalds if (ret == 0) 4801da177e4SLinus Torvalds ret = err; 4811da177e4SLinus Torvalds } 4824f8ad655SJan Kara trace_writeback_single_inode(inode, wbc, nr_to_write); 4834f8ad655SJan Kara return ret; 4844f8ad655SJan Kara } 4854f8ad655SJan Kara 4864f8ad655SJan Kara /* 4874f8ad655SJan Kara * Write out an inode's dirty pages. Either the caller has an active reference 4884f8ad655SJan Kara * on the inode or the inode has I_WILL_FREE set. 4894f8ad655SJan Kara * 4904f8ad655SJan Kara * This function is designed to be called for writing back one inode which 4914f8ad655SJan Kara * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode() 4924f8ad655SJan Kara * and does more profound writeback list handling in writeback_sb_inodes(). 4934f8ad655SJan Kara */ 4944f8ad655SJan Kara static int 4954f8ad655SJan Kara writeback_single_inode(struct inode *inode, struct bdi_writeback *wb, 4964f8ad655SJan Kara struct writeback_control *wbc) 4974f8ad655SJan Kara { 4984f8ad655SJan Kara int ret = 0; 4994f8ad655SJan Kara 5004f8ad655SJan Kara spin_lock(&inode->i_lock); 5014f8ad655SJan Kara if (!atomic_read(&inode->i_count)) 5024f8ad655SJan Kara WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); 5034f8ad655SJan Kara else 5044f8ad655SJan Kara WARN_ON(inode->i_state & I_WILL_FREE); 5054f8ad655SJan Kara 5064f8ad655SJan Kara if (inode->i_state & I_SYNC) { 5074f8ad655SJan Kara if (wbc->sync_mode != WB_SYNC_ALL) 5084f8ad655SJan Kara goto out; 5094f8ad655SJan Kara /* 510169ebd90SJan Kara * It's a data-integrity sync. We must wait. Since callers hold 511169ebd90SJan Kara * inode reference or inode has I_WILL_FREE set, it cannot go 512169ebd90SJan Kara * away under us. 5134f8ad655SJan Kara */ 514169ebd90SJan Kara __inode_wait_for_writeback(inode); 5154f8ad655SJan Kara } 5164f8ad655SJan Kara WARN_ON(inode->i_state & I_SYNC); 5174f8ad655SJan Kara /* 5184f8ad655SJan Kara * Skip inode if it is clean. We don't want to mess with writeback 5194f8ad655SJan Kara * lists in this function since flusher thread may be doing for example 5204f8ad655SJan Kara * sync in parallel and if we move the inode, it could get skipped. So 5214f8ad655SJan Kara * here we make sure inode is on some writeback list and leave it there 5224f8ad655SJan Kara * unless we have completely cleaned the inode. 5234f8ad655SJan Kara */ 5244f8ad655SJan Kara if (!(inode->i_state & I_DIRTY)) 5254f8ad655SJan Kara goto out; 5264f8ad655SJan Kara inode->i_state |= I_SYNC; 5274f8ad655SJan Kara spin_unlock(&inode->i_lock); 5284f8ad655SJan Kara 5294f8ad655SJan Kara ret = __writeback_single_inode(inode, wb, wbc); 5301da177e4SLinus Torvalds 531f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 532250df6edSDave Chinner spin_lock(&inode->i_lock); 5334f8ad655SJan Kara /* 5344f8ad655SJan Kara * If inode is clean, remove it from writeback lists. Otherwise don't 5354f8ad655SJan Kara * touch it. See comment above for explanation. 5364f8ad655SJan Kara */ 5374f8ad655SJan Kara if (!(inode->i_state & I_DIRTY)) 5384f8ad655SJan Kara list_del_init(&inode->i_wb_list); 5394f8ad655SJan Kara spin_unlock(&wb->list_lock); 5401c0eeaf5SJoern Engel inode_sync_complete(inode); 5414f8ad655SJan Kara out: 5424f8ad655SJan Kara spin_unlock(&inode->i_lock); 5431da177e4SLinus Torvalds return ret; 5441da177e4SLinus Torvalds } 5451da177e4SLinus Torvalds 5461a12d8bdSWu Fengguang static long writeback_chunk_size(struct backing_dev_info *bdi, 5471a12d8bdSWu Fengguang struct wb_writeback_work *work) 548d46db3d5SWu Fengguang { 549d46db3d5SWu Fengguang long pages; 550d46db3d5SWu Fengguang 551d46db3d5SWu Fengguang /* 552d46db3d5SWu Fengguang * WB_SYNC_ALL mode does livelock avoidance by syncing dirty 553d46db3d5SWu Fengguang * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX 554d46db3d5SWu Fengguang * here avoids calling into writeback_inodes_wb() more than once. 555d46db3d5SWu Fengguang * 556d46db3d5SWu Fengguang * The intended call sequence for WB_SYNC_ALL writeback is: 557d46db3d5SWu Fengguang * 558d46db3d5SWu Fengguang * wb_writeback() 559d46db3d5SWu Fengguang * writeback_sb_inodes() <== called only once 560d46db3d5SWu Fengguang * write_cache_pages() <== called once for each inode 561d46db3d5SWu Fengguang * (quickly) tag currently dirty pages 562d46db3d5SWu Fengguang * (maybe slowly) sync all tagged pages 563d46db3d5SWu Fengguang */ 564d46db3d5SWu Fengguang if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages) 565d46db3d5SWu Fengguang pages = LONG_MAX; 5661a12d8bdSWu Fengguang else { 5671a12d8bdSWu Fengguang pages = min(bdi->avg_write_bandwidth / 2, 5681a12d8bdSWu Fengguang global_dirty_limit / DIRTY_SCOPE); 5691a12d8bdSWu Fengguang pages = min(pages, work->nr_pages); 5701a12d8bdSWu Fengguang pages = round_down(pages + MIN_WRITEBACK_PAGES, 5711a12d8bdSWu Fengguang MIN_WRITEBACK_PAGES); 5721a12d8bdSWu Fengguang } 573d46db3d5SWu Fengguang 574d46db3d5SWu Fengguang return pages; 575d46db3d5SWu Fengguang } 576d46db3d5SWu Fengguang 57703ba3782SJens Axboe /* 578f11c9c5cSEdward Shishkin * Write a portion of b_io inodes which belong to @sb. 579edadfb10SChristoph Hellwig * 580edadfb10SChristoph Hellwig * If @only_this_sb is true, then find and write all such 581f11c9c5cSEdward Shishkin * inodes. Otherwise write only ones which go sequentially 582f11c9c5cSEdward Shishkin * in reverse order. 583edadfb10SChristoph Hellwig * 584d46db3d5SWu Fengguang * Return the number of pages and/or inodes written. 585f11c9c5cSEdward Shishkin */ 586d46db3d5SWu Fengguang static long writeback_sb_inodes(struct super_block *sb, 587d46db3d5SWu Fengguang struct bdi_writeback *wb, 588d46db3d5SWu Fengguang struct wb_writeback_work *work) 58903ba3782SJens Axboe { 590d46db3d5SWu Fengguang struct writeback_control wbc = { 591d46db3d5SWu Fengguang .sync_mode = work->sync_mode, 592d46db3d5SWu Fengguang .tagged_writepages = work->tagged_writepages, 593d46db3d5SWu Fengguang .for_kupdate = work->for_kupdate, 594d46db3d5SWu Fengguang .for_background = work->for_background, 595d46db3d5SWu Fengguang .range_cyclic = work->range_cyclic, 596d46db3d5SWu Fengguang .range_start = 0, 597d46db3d5SWu Fengguang .range_end = LLONG_MAX, 598d46db3d5SWu Fengguang }; 599d46db3d5SWu Fengguang unsigned long start_time = jiffies; 600d46db3d5SWu Fengguang long write_chunk; 601d46db3d5SWu Fengguang long wrote = 0; /* count both pages and inodes */ 602d46db3d5SWu Fengguang 60303ba3782SJens Axboe while (!list_empty(&wb->b_io)) { 6047ccf19a8SNick Piggin struct inode *inode = wb_inode(wb->b_io.prev); 605edadfb10SChristoph Hellwig 606edadfb10SChristoph Hellwig if (inode->i_sb != sb) { 607d46db3d5SWu Fengguang if (work->sb) { 608edadfb10SChristoph Hellwig /* 609edadfb10SChristoph Hellwig * We only want to write back data for this 610edadfb10SChristoph Hellwig * superblock, move all inodes not belonging 611edadfb10SChristoph Hellwig * to it back onto the dirty list. 612edadfb10SChristoph Hellwig */ 613f758eeabSChristoph Hellwig redirty_tail(inode, wb); 61466f3b8e2SJens Axboe continue; 61566f3b8e2SJens Axboe } 616edadfb10SChristoph Hellwig 617edadfb10SChristoph Hellwig /* 618edadfb10SChristoph Hellwig * The inode belongs to a different superblock. 619edadfb10SChristoph Hellwig * Bounce back to the caller to unpin this and 620edadfb10SChristoph Hellwig * pin the next superblock. 621edadfb10SChristoph Hellwig */ 622d46db3d5SWu Fengguang break; 623edadfb10SChristoph Hellwig } 624edadfb10SChristoph Hellwig 6259843b76aSChristoph Hellwig /* 626331cbdeeSWanpeng Li * Don't bother with new inodes or inodes being freed, first 627331cbdeeSWanpeng Li * kind does not need periodic writeout yet, and for the latter 6289843b76aSChristoph Hellwig * kind writeout is handled by the freer. 6299843b76aSChristoph Hellwig */ 630250df6edSDave Chinner spin_lock(&inode->i_lock); 6319843b76aSChristoph Hellwig if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { 632250df6edSDave Chinner spin_unlock(&inode->i_lock); 633fcc5c222SWu Fengguang redirty_tail(inode, wb); 6347ef0d737SNick Piggin continue; 6357ef0d737SNick Piggin } 636cc1676d9SJan Kara if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) { 637cc1676d9SJan Kara /* 638cc1676d9SJan Kara * If this inode is locked for writeback and we are not 639cc1676d9SJan Kara * doing writeback-for-data-integrity, move it to 640cc1676d9SJan Kara * b_more_io so that writeback can proceed with the 641cc1676d9SJan Kara * other inodes on s_io. 642cc1676d9SJan Kara * 643cc1676d9SJan Kara * We'll have another go at writing back this inode 644cc1676d9SJan Kara * when we completed a full scan of b_io. 645cc1676d9SJan Kara */ 646cc1676d9SJan Kara spin_unlock(&inode->i_lock); 647cc1676d9SJan Kara requeue_io(inode, wb); 648cc1676d9SJan Kara trace_writeback_sb_inodes_requeue(inode); 649cc1676d9SJan Kara continue; 650cc1676d9SJan Kara } 651f0d07b7fSJan Kara spin_unlock(&wb->list_lock); 652f0d07b7fSJan Kara 6534f8ad655SJan Kara /* 6544f8ad655SJan Kara * We already requeued the inode if it had I_SYNC set and we 6554f8ad655SJan Kara * are doing WB_SYNC_NONE writeback. So this catches only the 6564f8ad655SJan Kara * WB_SYNC_ALL case. 6574f8ad655SJan Kara */ 658169ebd90SJan Kara if (inode->i_state & I_SYNC) { 659169ebd90SJan Kara /* Wait for I_SYNC. This function drops i_lock... */ 660169ebd90SJan Kara inode_sleep_on_writeback(inode); 661169ebd90SJan Kara /* Inode may be gone, start again */ 662ead188f9SJan Kara spin_lock(&wb->list_lock); 663169ebd90SJan Kara continue; 664169ebd90SJan Kara } 6654f8ad655SJan Kara inode->i_state |= I_SYNC; 6664f8ad655SJan Kara spin_unlock(&inode->i_lock); 667169ebd90SJan Kara 6681a12d8bdSWu Fengguang write_chunk = writeback_chunk_size(wb->bdi, work); 669d46db3d5SWu Fengguang wbc.nr_to_write = write_chunk; 670d46db3d5SWu Fengguang wbc.pages_skipped = 0; 671250df6edSDave Chinner 672169ebd90SJan Kara /* 673169ebd90SJan Kara * We use I_SYNC to pin the inode in memory. While it is set 674169ebd90SJan Kara * evict_inode() will wait so the inode cannot be freed. 675169ebd90SJan Kara */ 6764f8ad655SJan Kara __writeback_single_inode(inode, wb, &wbc); 677d46db3d5SWu Fengguang 678d46db3d5SWu Fengguang work->nr_pages -= write_chunk - wbc.nr_to_write; 679d46db3d5SWu Fengguang wrote += write_chunk - wbc.nr_to_write; 6804f8ad655SJan Kara spin_lock(&wb->list_lock); 6814f8ad655SJan Kara spin_lock(&inode->i_lock); 682d46db3d5SWu Fengguang if (!(inode->i_state & I_DIRTY)) 683d46db3d5SWu Fengguang wrote++; 6844f8ad655SJan Kara requeue_inode(inode, wb, &wbc); 6854f8ad655SJan Kara inode_sync_complete(inode); 6860f1b1fd8SDave Chinner spin_unlock(&inode->i_lock); 687169ebd90SJan Kara cond_resched_lock(&wb->list_lock); 688d46db3d5SWu Fengguang /* 689d46db3d5SWu Fengguang * bail out to wb_writeback() often enough to check 690d46db3d5SWu Fengguang * background threshold and other termination conditions. 691d46db3d5SWu Fengguang */ 692d46db3d5SWu Fengguang if (wrote) { 693d46db3d5SWu Fengguang if (time_is_before_jiffies(start_time + HZ / 10UL)) 694d46db3d5SWu Fengguang break; 695d46db3d5SWu Fengguang if (work->nr_pages <= 0) 696d46db3d5SWu Fengguang break; 6971da177e4SLinus Torvalds } 6988bc3be27SFengguang Wu } 699d46db3d5SWu Fengguang return wrote; 700f11c9c5cSEdward Shishkin } 70138f21977SNick Piggin 702d46db3d5SWu Fengguang static long __writeback_inodes_wb(struct bdi_writeback *wb, 703d46db3d5SWu Fengguang struct wb_writeback_work *work) 704f11c9c5cSEdward Shishkin { 705d46db3d5SWu Fengguang unsigned long start_time = jiffies; 706d46db3d5SWu Fengguang long wrote = 0; 707f11c9c5cSEdward Shishkin 708f11c9c5cSEdward Shishkin while (!list_empty(&wb->b_io)) { 7097ccf19a8SNick Piggin struct inode *inode = wb_inode(wb->b_io.prev); 710f11c9c5cSEdward Shishkin struct super_block *sb = inode->i_sb; 711f11c9c5cSEdward Shishkin 71212ad3ab6SDave Chinner if (!grab_super_passive(sb)) { 7130e995816SWu Fengguang /* 7140e995816SWu Fengguang * grab_super_passive() may fail consistently due to 7150e995816SWu Fengguang * s_umount being grabbed by someone else. Don't use 7160e995816SWu Fengguang * requeue_io() to avoid busy retrying the inode/sb. 7170e995816SWu Fengguang */ 7180e995816SWu Fengguang redirty_tail(inode, wb); 719d19de7edSChristoph Hellwig continue; 720334132aeSChristoph Hellwig } 721d46db3d5SWu Fengguang wrote += writeback_sb_inodes(sb, wb, work); 722d19de7edSChristoph Hellwig drop_super(sb); 723f11c9c5cSEdward Shishkin 724d46db3d5SWu Fengguang /* refer to the same tests at the end of writeback_sb_inodes */ 725d46db3d5SWu Fengguang if (wrote) { 726d46db3d5SWu Fengguang if (time_is_before_jiffies(start_time + HZ / 10UL)) 727d46db3d5SWu Fengguang break; 728d46db3d5SWu Fengguang if (work->nr_pages <= 0) 729f11c9c5cSEdward Shishkin break; 730f11c9c5cSEdward Shishkin } 731d46db3d5SWu Fengguang } 73266f3b8e2SJens Axboe /* Leave any unwritten inodes on b_io */ 733d46db3d5SWu Fengguang return wrote; 73466f3b8e2SJens Axboe } 73566f3b8e2SJens Axboe 7360e175a18SCurt Wohlgemuth long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages, 7370e175a18SCurt Wohlgemuth enum wb_reason reason) 738edadfb10SChristoph Hellwig { 739d46db3d5SWu Fengguang struct wb_writeback_work work = { 740d46db3d5SWu Fengguang .nr_pages = nr_pages, 741d46db3d5SWu Fengguang .sync_mode = WB_SYNC_NONE, 742d46db3d5SWu Fengguang .range_cyclic = 1, 7430e175a18SCurt Wohlgemuth .reason = reason, 744d46db3d5SWu Fengguang }; 745edadfb10SChristoph Hellwig 746f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 747424b351fSWu Fengguang if (list_empty(&wb->b_io)) 748ad4e38ddSCurt Wohlgemuth queue_io(wb, &work); 749d46db3d5SWu Fengguang __writeback_inodes_wb(wb, &work); 750f758eeabSChristoph Hellwig spin_unlock(&wb->list_lock); 751edadfb10SChristoph Hellwig 752d46db3d5SWu Fengguang return nr_pages - work.nr_pages; 75366f3b8e2SJens Axboe } 75466f3b8e2SJens Axboe 755b00949aaSWu Fengguang static bool over_bground_thresh(struct backing_dev_info *bdi) 75603ba3782SJens Axboe { 75703ba3782SJens Axboe unsigned long background_thresh, dirty_thresh; 75803ba3782SJens Axboe 75916c4042fSWu Fengguang global_dirty_limits(&background_thresh, &dirty_thresh); 76003ba3782SJens Axboe 761b00949aaSWu Fengguang if (global_page_state(NR_FILE_DIRTY) + 762b00949aaSWu Fengguang global_page_state(NR_UNSTABLE_NFS) > background_thresh) 763b00949aaSWu Fengguang return true; 764b00949aaSWu Fengguang 765b00949aaSWu Fengguang if (bdi_stat(bdi, BDI_RECLAIMABLE) > 766b00949aaSWu Fengguang bdi_dirty_limit(bdi, background_thresh)) 767b00949aaSWu Fengguang return true; 768b00949aaSWu Fengguang 769b00949aaSWu Fengguang return false; 77003ba3782SJens Axboe } 77103ba3782SJens Axboe 77203ba3782SJens Axboe /* 773e98be2d5SWu Fengguang * Called under wb->list_lock. If there are multiple wb per bdi, 774e98be2d5SWu Fengguang * only the flusher working on the first wb should do it. 775e98be2d5SWu Fengguang */ 776e98be2d5SWu Fengguang static void wb_update_bandwidth(struct bdi_writeback *wb, 777e98be2d5SWu Fengguang unsigned long start_time) 778e98be2d5SWu Fengguang { 779af6a3113SWu Fengguang __bdi_update_bandwidth(wb->bdi, 0, 0, 0, 0, 0, start_time); 780e98be2d5SWu Fengguang } 781e98be2d5SWu Fengguang 782e98be2d5SWu Fengguang /* 78303ba3782SJens Axboe * Explicit flushing or periodic writeback of "old" data. 78403ba3782SJens Axboe * 78503ba3782SJens Axboe * Define "old": the first time one of an inode's pages is dirtied, we mark the 78603ba3782SJens Axboe * dirtying-time in the inode's address_space. So this periodic writeback code 78703ba3782SJens Axboe * just walks the superblock inode list, writing back any inodes which are 78803ba3782SJens Axboe * older than a specific point in time. 78903ba3782SJens Axboe * 79003ba3782SJens Axboe * Try to run once per dirty_writeback_interval. But if a writeback event 79103ba3782SJens Axboe * takes longer than a dirty_writeback_interval interval, then leave a 79203ba3782SJens Axboe * one-second gap. 79303ba3782SJens Axboe * 79403ba3782SJens Axboe * older_than_this takes precedence over nr_to_write. So we'll only write back 79503ba3782SJens Axboe * all dirty pages if they are all attached to "old" mappings. 79603ba3782SJens Axboe */ 797c4a77a6cSJens Axboe static long wb_writeback(struct bdi_writeback *wb, 79883ba7b07SChristoph Hellwig struct wb_writeback_work *work) 79903ba3782SJens Axboe { 800e98be2d5SWu Fengguang unsigned long wb_start = jiffies; 801d46db3d5SWu Fengguang long nr_pages = work->nr_pages; 80203ba3782SJens Axboe unsigned long oldest_jif; 803a5989bdcSJan Kara struct inode *inode; 804d46db3d5SWu Fengguang long progress; 80503ba3782SJens Axboe 806e185dda8SWu Fengguang oldest_jif = jiffies; 807d46db3d5SWu Fengguang work->older_than_this = &oldest_jif; 80803ba3782SJens Axboe 809e8dfc305SWu Fengguang spin_lock(&wb->list_lock); 81003ba3782SJens Axboe for (;;) { 81103ba3782SJens Axboe /* 812d3ddec76SWu Fengguang * Stop writeback when nr_pages has been consumed 81303ba3782SJens Axboe */ 81483ba7b07SChristoph Hellwig if (work->nr_pages <= 0) 81503ba3782SJens Axboe break; 81603ba3782SJens Axboe 81703ba3782SJens Axboe /* 818aa373cf5SJan Kara * Background writeout and kupdate-style writeback may 819aa373cf5SJan Kara * run forever. Stop them if there is other work to do 820aa373cf5SJan Kara * so that e.g. sync can proceed. They'll be restarted 821aa373cf5SJan Kara * after the other works are all done. 822aa373cf5SJan Kara */ 823aa373cf5SJan Kara if ((work->for_background || work->for_kupdate) && 824aa373cf5SJan Kara !list_empty(&wb->bdi->work_list)) 825aa373cf5SJan Kara break; 826aa373cf5SJan Kara 827aa373cf5SJan Kara /* 828d3ddec76SWu Fengguang * For background writeout, stop when we are below the 829d3ddec76SWu Fengguang * background dirty threshold 83003ba3782SJens Axboe */ 831b00949aaSWu Fengguang if (work->for_background && !over_bground_thresh(wb->bdi)) 83203ba3782SJens Axboe break; 83303ba3782SJens Axboe 8341bc36b64SJan Kara /* 8351bc36b64SJan Kara * Kupdate and background works are special and we want to 8361bc36b64SJan Kara * include all inodes that need writing. Livelock avoidance is 8371bc36b64SJan Kara * handled by these works yielding to any other work so we are 8381bc36b64SJan Kara * safe. 8391bc36b64SJan Kara */ 840ba9aa839SWu Fengguang if (work->for_kupdate) { 841ba9aa839SWu Fengguang oldest_jif = jiffies - 842ba9aa839SWu Fengguang msecs_to_jiffies(dirty_expire_interval * 10); 8431bc36b64SJan Kara } else if (work->for_background) 8441bc36b64SJan Kara oldest_jif = jiffies; 845028c2dd1SDave Chinner 846d46db3d5SWu Fengguang trace_writeback_start(wb->bdi, work); 847e8dfc305SWu Fengguang if (list_empty(&wb->b_io)) 848ad4e38ddSCurt Wohlgemuth queue_io(wb, work); 84983ba7b07SChristoph Hellwig if (work->sb) 850d46db3d5SWu Fengguang progress = writeback_sb_inodes(work->sb, wb, work); 851edadfb10SChristoph Hellwig else 852d46db3d5SWu Fengguang progress = __writeback_inodes_wb(wb, work); 853d46db3d5SWu Fengguang trace_writeback_written(wb->bdi, work); 854028c2dd1SDave Chinner 855e98be2d5SWu Fengguang wb_update_bandwidth(wb, wb_start); 85603ba3782SJens Axboe 85703ba3782SJens Axboe /* 85871fd05a8SJens Axboe * Did we write something? Try for more 859e6fb6da2SWu Fengguang * 860e6fb6da2SWu Fengguang * Dirty inodes are moved to b_io for writeback in batches. 861e6fb6da2SWu Fengguang * The completion of the current batch does not necessarily 862e6fb6da2SWu Fengguang * mean the overall work is done. So we keep looping as long 863e6fb6da2SWu Fengguang * as made some progress on cleaning pages or inodes. 86471fd05a8SJens Axboe */ 865d46db3d5SWu Fengguang if (progress) 86603ba3782SJens Axboe continue; 867a5989bdcSJan Kara /* 868e6fb6da2SWu Fengguang * No more inodes for IO, bail 869a5989bdcSJan Kara */ 870b7a2441fSWu Fengguang if (list_empty(&wb->b_more_io)) 87103ba3782SJens Axboe break; 87203ba3782SJens Axboe /* 8738010c3b6SJens Axboe * Nothing written. Wait for some inode to 8748010c3b6SJens Axboe * become available for writeback. Otherwise 8758010c3b6SJens Axboe * we'll just busyloop. 87603ba3782SJens Axboe */ 87703ba3782SJens Axboe if (!list_empty(&wb->b_more_io)) { 878d46db3d5SWu Fengguang trace_writeback_wait(wb->bdi, work); 87903ba3782SJens Axboe inode = wb_inode(wb->b_more_io.prev); 880250df6edSDave Chinner spin_lock(&inode->i_lock); 881f0d07b7fSJan Kara spin_unlock(&wb->list_lock); 882169ebd90SJan Kara /* This function drops i_lock... */ 883169ebd90SJan Kara inode_sleep_on_writeback(inode); 884f0d07b7fSJan Kara spin_lock(&wb->list_lock); 88503ba3782SJens Axboe } 88603ba3782SJens Axboe } 887e8dfc305SWu Fengguang spin_unlock(&wb->list_lock); 88803ba3782SJens Axboe 889d46db3d5SWu Fengguang return nr_pages - work->nr_pages; 89003ba3782SJens Axboe } 89103ba3782SJens Axboe 89203ba3782SJens Axboe /* 89383ba7b07SChristoph Hellwig * Return the next wb_writeback_work struct that hasn't been processed yet. 89403ba3782SJens Axboe */ 89583ba7b07SChristoph Hellwig static struct wb_writeback_work * 89608852b6dSMinchan Kim get_next_work_item(struct backing_dev_info *bdi) 89703ba3782SJens Axboe { 89883ba7b07SChristoph Hellwig struct wb_writeback_work *work = NULL; 89903ba3782SJens Axboe 9006467716aSArtem Bityutskiy spin_lock_bh(&bdi->wb_lock); 90183ba7b07SChristoph Hellwig if (!list_empty(&bdi->work_list)) { 90283ba7b07SChristoph Hellwig work = list_entry(bdi->work_list.next, 90383ba7b07SChristoph Hellwig struct wb_writeback_work, list); 90483ba7b07SChristoph Hellwig list_del_init(&work->list); 90503ba3782SJens Axboe } 9066467716aSArtem Bityutskiy spin_unlock_bh(&bdi->wb_lock); 90783ba7b07SChristoph Hellwig return work; 90803ba3782SJens Axboe } 90903ba3782SJens Axboe 910cdf01dd5SLinus Torvalds /* 911cdf01dd5SLinus Torvalds * Add in the number of potentially dirty inodes, because each inode 912cdf01dd5SLinus Torvalds * write can dirty pagecache in the underlying blockdev. 913cdf01dd5SLinus Torvalds */ 914cdf01dd5SLinus Torvalds static unsigned long get_nr_dirty_pages(void) 915cdf01dd5SLinus Torvalds { 916cdf01dd5SLinus Torvalds return global_page_state(NR_FILE_DIRTY) + 917cdf01dd5SLinus Torvalds global_page_state(NR_UNSTABLE_NFS) + 918cdf01dd5SLinus Torvalds get_nr_dirty_inodes(); 919cdf01dd5SLinus Torvalds } 920cdf01dd5SLinus Torvalds 9216585027aSJan Kara static long wb_check_background_flush(struct bdi_writeback *wb) 9226585027aSJan Kara { 923b00949aaSWu Fengguang if (over_bground_thresh(wb->bdi)) { 9246585027aSJan Kara 9256585027aSJan Kara struct wb_writeback_work work = { 9266585027aSJan Kara .nr_pages = LONG_MAX, 9276585027aSJan Kara .sync_mode = WB_SYNC_NONE, 9286585027aSJan Kara .for_background = 1, 9296585027aSJan Kara .range_cyclic = 1, 9300e175a18SCurt Wohlgemuth .reason = WB_REASON_BACKGROUND, 9316585027aSJan Kara }; 9326585027aSJan Kara 9336585027aSJan Kara return wb_writeback(wb, &work); 9346585027aSJan Kara } 9356585027aSJan Kara 9366585027aSJan Kara return 0; 9376585027aSJan Kara } 9386585027aSJan Kara 93903ba3782SJens Axboe static long wb_check_old_data_flush(struct bdi_writeback *wb) 94003ba3782SJens Axboe { 94103ba3782SJens Axboe unsigned long expired; 94203ba3782SJens Axboe long nr_pages; 94303ba3782SJens Axboe 94469b62d01SJens Axboe /* 94569b62d01SJens Axboe * When set to zero, disable periodic writeback 94669b62d01SJens Axboe */ 94769b62d01SJens Axboe if (!dirty_writeback_interval) 94869b62d01SJens Axboe return 0; 94969b62d01SJens Axboe 95003ba3782SJens Axboe expired = wb->last_old_flush + 95103ba3782SJens Axboe msecs_to_jiffies(dirty_writeback_interval * 10); 95203ba3782SJens Axboe if (time_before(jiffies, expired)) 95303ba3782SJens Axboe return 0; 95403ba3782SJens Axboe 95503ba3782SJens Axboe wb->last_old_flush = jiffies; 956cdf01dd5SLinus Torvalds nr_pages = get_nr_dirty_pages(); 95703ba3782SJens Axboe 958c4a77a6cSJens Axboe if (nr_pages) { 95983ba7b07SChristoph Hellwig struct wb_writeback_work work = { 960c4a77a6cSJens Axboe .nr_pages = nr_pages, 961c4a77a6cSJens Axboe .sync_mode = WB_SYNC_NONE, 962c4a77a6cSJens Axboe .for_kupdate = 1, 963c4a77a6cSJens Axboe .range_cyclic = 1, 9640e175a18SCurt Wohlgemuth .reason = WB_REASON_PERIODIC, 965c4a77a6cSJens Axboe }; 966c4a77a6cSJens Axboe 96783ba7b07SChristoph Hellwig return wb_writeback(wb, &work); 968c4a77a6cSJens Axboe } 96903ba3782SJens Axboe 97003ba3782SJens Axboe return 0; 97103ba3782SJens Axboe } 97203ba3782SJens Axboe 97303ba3782SJens Axboe /* 97403ba3782SJens Axboe * Retrieve work items and do the writeback they describe 97503ba3782SJens Axboe */ 97603ba3782SJens Axboe long wb_do_writeback(struct bdi_writeback *wb, int force_wait) 97703ba3782SJens Axboe { 97803ba3782SJens Axboe struct backing_dev_info *bdi = wb->bdi; 97983ba7b07SChristoph Hellwig struct wb_writeback_work *work; 980c4a77a6cSJens Axboe long wrote = 0; 98103ba3782SJens Axboe 98281d73a32SJan Kara set_bit(BDI_writeback_running, &wb->bdi->state); 98308852b6dSMinchan Kim while ((work = get_next_work_item(bdi)) != NULL) { 98403ba3782SJens Axboe /* 98503ba3782SJens Axboe * Override sync mode, in case we must wait for completion 98683ba7b07SChristoph Hellwig * because this thread is exiting now. 98703ba3782SJens Axboe */ 98803ba3782SJens Axboe if (force_wait) 98983ba7b07SChristoph Hellwig work->sync_mode = WB_SYNC_ALL; 99083ba7b07SChristoph Hellwig 991455b2864SDave Chinner trace_writeback_exec(bdi, work); 992455b2864SDave Chinner 99383ba7b07SChristoph Hellwig wrote += wb_writeback(wb, work); 99403ba3782SJens Axboe 99503ba3782SJens Axboe /* 99683ba7b07SChristoph Hellwig * Notify the caller of completion if this is a synchronous 99783ba7b07SChristoph Hellwig * work item, otherwise just free it. 99803ba3782SJens Axboe */ 99983ba7b07SChristoph Hellwig if (work->done) 100083ba7b07SChristoph Hellwig complete(work->done); 100183ba7b07SChristoph Hellwig else 100283ba7b07SChristoph Hellwig kfree(work); 100303ba3782SJens Axboe } 100403ba3782SJens Axboe 100503ba3782SJens Axboe /* 100603ba3782SJens Axboe * Check for periodic writeback, kupdated() style 100703ba3782SJens Axboe */ 100803ba3782SJens Axboe wrote += wb_check_old_data_flush(wb); 10096585027aSJan Kara wrote += wb_check_background_flush(wb); 101081d73a32SJan Kara clear_bit(BDI_writeback_running, &wb->bdi->state); 101103ba3782SJens Axboe 101203ba3782SJens Axboe return wrote; 101303ba3782SJens Axboe } 101403ba3782SJens Axboe 101503ba3782SJens Axboe /* 101603ba3782SJens Axboe * Handle writeback of dirty data for the device backed by this bdi. Also 101703ba3782SJens Axboe * wakes up periodically and does kupdated style flushing. 101803ba3782SJens Axboe */ 101908243900SChristoph Hellwig int bdi_writeback_thread(void *data) 102003ba3782SJens Axboe { 102108243900SChristoph Hellwig struct bdi_writeback *wb = data; 102208243900SChristoph Hellwig struct backing_dev_info *bdi = wb->bdi; 102303ba3782SJens Axboe long pages_written; 102403ba3782SJens Axboe 1025766f9164SPeter Zijlstra current->flags |= PF_SWAPWRITE; 102608243900SChristoph Hellwig set_freezable(); 1027ecd58403SArtem Bityutskiy wb->last_active = jiffies; 102803ba3782SJens Axboe 102903ba3782SJens Axboe /* 103008243900SChristoph Hellwig * Our parent may run at a different priority, just set us to normal 103103ba3782SJens Axboe */ 103208243900SChristoph Hellwig set_user_nice(current, 0); 103308243900SChristoph Hellwig 1034455b2864SDave Chinner trace_writeback_thread_start(bdi); 1035455b2864SDave Chinner 10368a32c441STejun Heo while (!kthread_freezable_should_stop(NULL)) { 10376467716aSArtem Bityutskiy /* 10386467716aSArtem Bityutskiy * Remove own delayed wake-up timer, since we are already awake 10396467716aSArtem Bityutskiy * and we'll take care of the preriodic write-back. 10406467716aSArtem Bityutskiy */ 10416467716aSArtem Bityutskiy del_timer(&wb->wakeup_timer); 10426467716aSArtem Bityutskiy 104303ba3782SJens Axboe pages_written = wb_do_writeback(wb, 0); 104403ba3782SJens Axboe 1045455b2864SDave Chinner trace_writeback_pages_written(pages_written); 1046455b2864SDave Chinner 104703ba3782SJens Axboe if (pages_written) 1048ecd58403SArtem Bityutskiy wb->last_active = jiffies; 104903ba3782SJens Axboe 1050297252c8SArtem Bityutskiy set_current_state(TASK_INTERRUPTIBLE); 1051b76b4014SJ. Bruce Fields if (!list_empty(&bdi->work_list) || kthread_should_stop()) { 1052297252c8SArtem Bityutskiy __set_current_state(TASK_RUNNING); 1053297252c8SArtem Bityutskiy continue; 105403ba3782SJens Axboe } 105503ba3782SJens Axboe 1056253c34e9SArtem Bityutskiy if (wb_has_dirty_io(wb) && dirty_writeback_interval) 1057fff5b85aSArtem Bityutskiy schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10)); 1058253c34e9SArtem Bityutskiy else { 1059253c34e9SArtem Bityutskiy /* 1060253c34e9SArtem Bityutskiy * We have nothing to do, so can go sleep without any 1061253c34e9SArtem Bityutskiy * timeout and save power. When a work is queued or 1062253c34e9SArtem Bityutskiy * something is made dirty - we will be woken up. 1063253c34e9SArtem Bityutskiy */ 106469b62d01SJens Axboe schedule(); 1065f9eadbbdSJens Axboe } 106603ba3782SJens Axboe } 106703ba3782SJens Axboe 1068fff5b85aSArtem Bityutskiy /* Flush any work that raced with us exiting */ 106908243900SChristoph Hellwig if (!list_empty(&bdi->work_list)) 107008243900SChristoph Hellwig wb_do_writeback(wb, 1); 1071455b2864SDave Chinner 1072455b2864SDave Chinner trace_writeback_thread_stop(bdi); 107303ba3782SJens Axboe return 0; 107403ba3782SJens Axboe } 107503ba3782SJens Axboe 107608243900SChristoph Hellwig 107703ba3782SJens Axboe /* 107803ba3782SJens Axboe * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back 107903ba3782SJens Axboe * the whole world. 108003ba3782SJens Axboe */ 10810e175a18SCurt Wohlgemuth void wakeup_flusher_threads(long nr_pages, enum wb_reason reason) 108203ba3782SJens Axboe { 1083b8c2f347SChristoph Hellwig struct backing_dev_info *bdi; 1084b8c2f347SChristoph Hellwig 108583ba7b07SChristoph Hellwig if (!nr_pages) { 108683ba7b07SChristoph Hellwig nr_pages = global_page_state(NR_FILE_DIRTY) + 108703ba3782SJens Axboe global_page_state(NR_UNSTABLE_NFS); 1088b8c2f347SChristoph Hellwig } 1089b8c2f347SChristoph Hellwig 1090b8c2f347SChristoph Hellwig rcu_read_lock(); 1091b8c2f347SChristoph Hellwig list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { 1092b8c2f347SChristoph Hellwig if (!bdi_has_dirty_io(bdi)) 1093b8c2f347SChristoph Hellwig continue; 10940e175a18SCurt Wohlgemuth __bdi_start_writeback(bdi, nr_pages, false, reason); 1095b8c2f347SChristoph Hellwig } 1096b8c2f347SChristoph Hellwig rcu_read_unlock(); 109703ba3782SJens Axboe } 109803ba3782SJens Axboe 109903ba3782SJens Axboe static noinline void block_dump___mark_inode_dirty(struct inode *inode) 110003ba3782SJens Axboe { 110103ba3782SJens Axboe if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { 110203ba3782SJens Axboe struct dentry *dentry; 110303ba3782SJens Axboe const char *name = "?"; 110403ba3782SJens Axboe 110503ba3782SJens Axboe dentry = d_find_alias(inode); 110603ba3782SJens Axboe if (dentry) { 110703ba3782SJens Axboe spin_lock(&dentry->d_lock); 110803ba3782SJens Axboe name = (const char *) dentry->d_name.name; 110903ba3782SJens Axboe } 111003ba3782SJens Axboe printk(KERN_DEBUG 111103ba3782SJens Axboe "%s(%d): dirtied inode %lu (%s) on %s\n", 111203ba3782SJens Axboe current->comm, task_pid_nr(current), inode->i_ino, 111303ba3782SJens Axboe name, inode->i_sb->s_id); 111403ba3782SJens Axboe if (dentry) { 111503ba3782SJens Axboe spin_unlock(&dentry->d_lock); 111603ba3782SJens Axboe dput(dentry); 111703ba3782SJens Axboe } 111803ba3782SJens Axboe } 111903ba3782SJens Axboe } 112003ba3782SJens Axboe 112103ba3782SJens Axboe /** 112203ba3782SJens Axboe * __mark_inode_dirty - internal function 112303ba3782SJens Axboe * @inode: inode to mark 112403ba3782SJens Axboe * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) 112503ba3782SJens Axboe * Mark an inode as dirty. Callers should use mark_inode_dirty or 112603ba3782SJens Axboe * mark_inode_dirty_sync. 112703ba3782SJens Axboe * 112803ba3782SJens Axboe * Put the inode on the super block's dirty list. 112903ba3782SJens Axboe * 113003ba3782SJens Axboe * CAREFUL! We mark it dirty unconditionally, but move it onto the 113103ba3782SJens Axboe * dirty list only if it is hashed or if it refers to a blockdev. 113203ba3782SJens Axboe * If it was not hashed, it will never be added to the dirty list 113303ba3782SJens Axboe * even if it is later hashed, as it will have been marked dirty already. 113403ba3782SJens Axboe * 113503ba3782SJens Axboe * In short, make sure you hash any inodes _before_ you start marking 113603ba3782SJens Axboe * them dirty. 113703ba3782SJens Axboe * 113803ba3782SJens Axboe * Note that for blockdevs, inode->dirtied_when represents the dirtying time of 113903ba3782SJens Axboe * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of 114003ba3782SJens Axboe * the kernel-internal blockdev inode represents the dirtying time of the 114103ba3782SJens Axboe * blockdev's pages. This is why for I_DIRTY_PAGES we always use 114203ba3782SJens Axboe * page->mapping->host, so the page-dirtying time is recorded in the internal 114303ba3782SJens Axboe * blockdev inode. 114403ba3782SJens Axboe */ 114503ba3782SJens Axboe void __mark_inode_dirty(struct inode *inode, int flags) 114603ba3782SJens Axboe { 114703ba3782SJens Axboe struct super_block *sb = inode->i_sb; 1148253c34e9SArtem Bityutskiy struct backing_dev_info *bdi = NULL; 114903ba3782SJens Axboe 115003ba3782SJens Axboe /* 115103ba3782SJens Axboe * Don't do this for I_DIRTY_PAGES - that doesn't actually 115203ba3782SJens Axboe * dirty the inode itself 115303ba3782SJens Axboe */ 115403ba3782SJens Axboe if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { 115503ba3782SJens Axboe if (sb->s_op->dirty_inode) 1156aa385729SChristoph Hellwig sb->s_op->dirty_inode(inode, flags); 115703ba3782SJens Axboe } 115803ba3782SJens Axboe 115903ba3782SJens Axboe /* 116003ba3782SJens Axboe * make sure that changes are seen by all cpus before we test i_state 116103ba3782SJens Axboe * -- mikulas 116203ba3782SJens Axboe */ 116303ba3782SJens Axboe smp_mb(); 116403ba3782SJens Axboe 116503ba3782SJens Axboe /* avoid the locking if we can */ 116603ba3782SJens Axboe if ((inode->i_state & flags) == flags) 116703ba3782SJens Axboe return; 116803ba3782SJens Axboe 116903ba3782SJens Axboe if (unlikely(block_dump)) 117003ba3782SJens Axboe block_dump___mark_inode_dirty(inode); 117103ba3782SJens Axboe 1172250df6edSDave Chinner spin_lock(&inode->i_lock); 117303ba3782SJens Axboe if ((inode->i_state & flags) != flags) { 117403ba3782SJens Axboe const int was_dirty = inode->i_state & I_DIRTY; 117503ba3782SJens Axboe 117603ba3782SJens Axboe inode->i_state |= flags; 117703ba3782SJens Axboe 117803ba3782SJens Axboe /* 117903ba3782SJens Axboe * If the inode is being synced, just update its dirty state. 118003ba3782SJens Axboe * The unlocker will place the inode on the appropriate 118103ba3782SJens Axboe * superblock list, based upon its state. 118203ba3782SJens Axboe */ 118303ba3782SJens Axboe if (inode->i_state & I_SYNC) 1184250df6edSDave Chinner goto out_unlock_inode; 118503ba3782SJens Axboe 118603ba3782SJens Axboe /* 118703ba3782SJens Axboe * Only add valid (hashed) inodes to the superblock's 118803ba3782SJens Axboe * dirty list. Add blockdev inodes as well. 118903ba3782SJens Axboe */ 119003ba3782SJens Axboe if (!S_ISBLK(inode->i_mode)) { 11911d3382cbSAl Viro if (inode_unhashed(inode)) 1192250df6edSDave Chinner goto out_unlock_inode; 119303ba3782SJens Axboe } 1194a4ffdde6SAl Viro if (inode->i_state & I_FREEING) 1195250df6edSDave Chinner goto out_unlock_inode; 119603ba3782SJens Axboe 119703ba3782SJens Axboe /* 119803ba3782SJens Axboe * If the inode was already on b_dirty/b_io/b_more_io, don't 119903ba3782SJens Axboe * reposition it (that would break b_dirty time-ordering). 120003ba3782SJens Axboe */ 120103ba3782SJens Axboe if (!was_dirty) { 1202a66979abSDave Chinner bool wakeup_bdi = false; 1203253c34e9SArtem Bityutskiy bdi = inode_to_bdi(inode); 1204500b067cSJens Axboe 1205253c34e9SArtem Bityutskiy if (bdi_cap_writeback_dirty(bdi)) { 1206253c34e9SArtem Bityutskiy WARN(!test_bit(BDI_registered, &bdi->state), 1207253c34e9SArtem Bityutskiy "bdi-%s not registered\n", bdi->name); 1208253c34e9SArtem Bityutskiy 1209253c34e9SArtem Bityutskiy /* 1210253c34e9SArtem Bityutskiy * If this is the first dirty inode for this 1211253c34e9SArtem Bityutskiy * bdi, we have to wake-up the corresponding 1212253c34e9SArtem Bityutskiy * bdi thread to make sure background 1213253c34e9SArtem Bityutskiy * write-back happens later. 1214253c34e9SArtem Bityutskiy */ 1215253c34e9SArtem Bityutskiy if (!wb_has_dirty_io(&bdi->wb)) 1216253c34e9SArtem Bityutskiy wakeup_bdi = true; 1217500b067cSJens Axboe } 121803ba3782SJens Axboe 1219a66979abSDave Chinner spin_unlock(&inode->i_lock); 1220f758eeabSChristoph Hellwig spin_lock(&bdi->wb.list_lock); 122103ba3782SJens Axboe inode->dirtied_when = jiffies; 12227ccf19a8SNick Piggin list_move(&inode->i_wb_list, &bdi->wb.b_dirty); 1223f758eeabSChristoph Hellwig spin_unlock(&bdi->wb.list_lock); 1224253c34e9SArtem Bityutskiy 1225253c34e9SArtem Bityutskiy if (wakeup_bdi) 12266467716aSArtem Bityutskiy bdi_wakeup_thread_delayed(bdi); 1227a66979abSDave Chinner return; 1228a66979abSDave Chinner } 1229a66979abSDave Chinner } 1230a66979abSDave Chinner out_unlock_inode: 1231a66979abSDave Chinner spin_unlock(&inode->i_lock); 1232a66979abSDave Chinner 123303ba3782SJens Axboe } 123403ba3782SJens Axboe EXPORT_SYMBOL(__mark_inode_dirty); 123503ba3782SJens Axboe 1236b6e51316SJens Axboe static void wait_sb_inodes(struct super_block *sb) 123766f3b8e2SJens Axboe { 123838f21977SNick Piggin struct inode *inode, *old_inode = NULL; 123938f21977SNick Piggin 124003ba3782SJens Axboe /* 124103ba3782SJens Axboe * We need to be protected against the filesystem going from 124203ba3782SJens Axboe * r/o to r/w or vice versa. 124303ba3782SJens Axboe */ 1244b6e51316SJens Axboe WARN_ON(!rwsem_is_locked(&sb->s_umount)); 124503ba3782SJens Axboe 124655fa6091SDave Chinner spin_lock(&inode_sb_list_lock); 124766f3b8e2SJens Axboe 124838f21977SNick Piggin /* 124938f21977SNick Piggin * Data integrity sync. Must wait for all pages under writeback, 125038f21977SNick Piggin * because there may have been pages dirtied before our sync 125138f21977SNick Piggin * call, but which had writeout started before we write it out. 125238f21977SNick Piggin * In which case, the inode may not be on the dirty list, but 125338f21977SNick Piggin * we still have to wait for that writeout. 125438f21977SNick Piggin */ 1255b6e51316SJens Axboe list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { 1256250df6edSDave Chinner struct address_space *mapping = inode->i_mapping; 125738f21977SNick Piggin 1258250df6edSDave Chinner spin_lock(&inode->i_lock); 1259250df6edSDave Chinner if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) || 1260250df6edSDave Chinner (mapping->nrpages == 0)) { 1261250df6edSDave Chinner spin_unlock(&inode->i_lock); 126238f21977SNick Piggin continue; 1263250df6edSDave Chinner } 126438f21977SNick Piggin __iget(inode); 1265250df6edSDave Chinner spin_unlock(&inode->i_lock); 126655fa6091SDave Chinner spin_unlock(&inode_sb_list_lock); 126755fa6091SDave Chinner 126838f21977SNick Piggin /* 126955fa6091SDave Chinner * We hold a reference to 'inode' so it couldn't have been 127055fa6091SDave Chinner * removed from s_inodes list while we dropped the 127155fa6091SDave Chinner * inode_sb_list_lock. We cannot iput the inode now as we can 127255fa6091SDave Chinner * be holding the last reference and we cannot iput it under 127355fa6091SDave Chinner * inode_sb_list_lock. So we keep the reference and iput it 127455fa6091SDave Chinner * later. 127538f21977SNick Piggin */ 127638f21977SNick Piggin iput(old_inode); 127738f21977SNick Piggin old_inode = inode; 127838f21977SNick Piggin 127938f21977SNick Piggin filemap_fdatawait(mapping); 128038f21977SNick Piggin 128138f21977SNick Piggin cond_resched(); 128238f21977SNick Piggin 128355fa6091SDave Chinner spin_lock(&inode_sb_list_lock); 128438f21977SNick Piggin } 128555fa6091SDave Chinner spin_unlock(&inode_sb_list_lock); 128638f21977SNick Piggin iput(old_inode); 128766f3b8e2SJens Axboe } 12881da177e4SLinus Torvalds 1289d8a8559cSJens Axboe /** 12903259f8beSChris Mason * writeback_inodes_sb_nr - writeback dirty inodes from given super_block 1291d8a8559cSJens Axboe * @sb: the superblock 12923259f8beSChris Mason * @nr: the number of pages to write 1293786228abSMarcos Paulo de Souza * @reason: reason why some writeback work initiated 12941da177e4SLinus Torvalds * 1295d8a8559cSJens Axboe * Start writeback on some inodes on this super_block. No guarantees are made 1296d8a8559cSJens Axboe * on how many (if any) will be written, and this function does not wait 12973259f8beSChris Mason * for IO completion of submitted IO. 12981da177e4SLinus Torvalds */ 12990e175a18SCurt Wohlgemuth void writeback_inodes_sb_nr(struct super_block *sb, 13000e175a18SCurt Wohlgemuth unsigned long nr, 13010e175a18SCurt Wohlgemuth enum wb_reason reason) 13021da177e4SLinus Torvalds { 130383ba7b07SChristoph Hellwig DECLARE_COMPLETION_ONSTACK(done); 130483ba7b07SChristoph Hellwig struct wb_writeback_work work = { 13053c4d7165SChristoph Hellwig .sb = sb, 13063c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_NONE, 13076e6938b6SWu Fengguang .tagged_writepages = 1, 130883ba7b07SChristoph Hellwig .done = &done, 13093259f8beSChris Mason .nr_pages = nr, 13100e175a18SCurt Wohlgemuth .reason = reason, 13113c4d7165SChristoph Hellwig }; 13120e3c9a22SJens Axboe 13136eedc701SJan Kara if (sb->s_bdi == &noop_backing_dev_info) 13146eedc701SJan Kara return; 1315cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 131683ba7b07SChristoph Hellwig bdi_queue_work(sb->s_bdi, &work); 131783ba7b07SChristoph Hellwig wait_for_completion(&done); 13181da177e4SLinus Torvalds } 13193259f8beSChris Mason EXPORT_SYMBOL(writeback_inodes_sb_nr); 13203259f8beSChris Mason 13213259f8beSChris Mason /** 13223259f8beSChris Mason * writeback_inodes_sb - writeback dirty inodes from given super_block 13233259f8beSChris Mason * @sb: the superblock 1324786228abSMarcos Paulo de Souza * @reason: reason why some writeback work was initiated 13253259f8beSChris Mason * 13263259f8beSChris Mason * Start writeback on some inodes on this super_block. No guarantees are made 13273259f8beSChris Mason * on how many (if any) will be written, and this function does not wait 13283259f8beSChris Mason * for IO completion of submitted IO. 13293259f8beSChris Mason */ 13300e175a18SCurt Wohlgemuth void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) 13313259f8beSChris Mason { 13320e175a18SCurt Wohlgemuth return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); 13333259f8beSChris Mason } 1334d8a8559cSJens Axboe EXPORT_SYMBOL(writeback_inodes_sb); 1335d8a8559cSJens Axboe 1336d8a8559cSJens Axboe /** 133717bd55d0SEric Sandeen * writeback_inodes_sb_if_idle - start writeback if none underway 133817bd55d0SEric Sandeen * @sb: the superblock 1339786228abSMarcos Paulo de Souza * @reason: reason why some writeback work was initiated 134017bd55d0SEric Sandeen * 134117bd55d0SEric Sandeen * Invoke writeback_inodes_sb if no writeback is currently underway. 134217bd55d0SEric Sandeen * Returns 1 if writeback was started, 0 if not. 134317bd55d0SEric Sandeen */ 13440e175a18SCurt Wohlgemuth int writeback_inodes_sb_if_idle(struct super_block *sb, enum wb_reason reason) 134517bd55d0SEric Sandeen { 134617bd55d0SEric Sandeen if (!writeback_in_progress(sb->s_bdi)) { 1347cf37e972SChristoph Hellwig down_read(&sb->s_umount); 13480e175a18SCurt Wohlgemuth writeback_inodes_sb(sb, reason); 1349cf37e972SChristoph Hellwig up_read(&sb->s_umount); 135017bd55d0SEric Sandeen return 1; 135117bd55d0SEric Sandeen } else 135217bd55d0SEric Sandeen return 0; 135317bd55d0SEric Sandeen } 135417bd55d0SEric Sandeen EXPORT_SYMBOL(writeback_inodes_sb_if_idle); 135517bd55d0SEric Sandeen 135617bd55d0SEric Sandeen /** 1357c097b2caSFengguang Wu * writeback_inodes_sb_nr_if_idle - start writeback if none underway 13583259f8beSChris Mason * @sb: the superblock 13593259f8beSChris Mason * @nr: the number of pages to write 1360786228abSMarcos Paulo de Souza * @reason: reason why some writeback work was initiated 13613259f8beSChris Mason * 13623259f8beSChris Mason * Invoke writeback_inodes_sb if no writeback is currently underway. 13633259f8beSChris Mason * Returns 1 if writeback was started, 0 if not. 13643259f8beSChris Mason */ 13653259f8beSChris Mason int writeback_inodes_sb_nr_if_idle(struct super_block *sb, 13660e175a18SCurt Wohlgemuth unsigned long nr, 13670e175a18SCurt Wohlgemuth enum wb_reason reason) 13683259f8beSChris Mason { 13693259f8beSChris Mason if (!writeback_in_progress(sb->s_bdi)) { 13703259f8beSChris Mason down_read(&sb->s_umount); 13710e175a18SCurt Wohlgemuth writeback_inodes_sb_nr(sb, nr, reason); 13723259f8beSChris Mason up_read(&sb->s_umount); 13733259f8beSChris Mason return 1; 13743259f8beSChris Mason } else 13753259f8beSChris Mason return 0; 13763259f8beSChris Mason } 13773259f8beSChris Mason EXPORT_SYMBOL(writeback_inodes_sb_nr_if_idle); 13783259f8beSChris Mason 13793259f8beSChris Mason /** 1380d8a8559cSJens Axboe * sync_inodes_sb - sync sb inode pages 1381d8a8559cSJens Axboe * @sb: the superblock 1382d8a8559cSJens Axboe * 1383d8a8559cSJens Axboe * This function writes and waits on any dirty inode belonging to this 1384cb9ef8d5SStefan Hajnoczi * super_block. 1385d8a8559cSJens Axboe */ 1386b6e51316SJens Axboe void sync_inodes_sb(struct super_block *sb) 1387d8a8559cSJens Axboe { 138883ba7b07SChristoph Hellwig DECLARE_COMPLETION_ONSTACK(done); 138983ba7b07SChristoph Hellwig struct wb_writeback_work work = { 13903c4d7165SChristoph Hellwig .sb = sb, 13913c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_ALL, 13923c4d7165SChristoph Hellwig .nr_pages = LONG_MAX, 13933c4d7165SChristoph Hellwig .range_cyclic = 0, 139483ba7b07SChristoph Hellwig .done = &done, 13950e175a18SCurt Wohlgemuth .reason = WB_REASON_SYNC, 13963c4d7165SChristoph Hellwig }; 13973c4d7165SChristoph Hellwig 13986eedc701SJan Kara /* Nothing to do? */ 13996eedc701SJan Kara if (sb->s_bdi == &noop_backing_dev_info) 14006eedc701SJan Kara return; 1401cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 1402cf37e972SChristoph Hellwig 140383ba7b07SChristoph Hellwig bdi_queue_work(sb->s_bdi, &work); 140483ba7b07SChristoph Hellwig wait_for_completion(&done); 140583ba7b07SChristoph Hellwig 1406b6e51316SJens Axboe wait_sb_inodes(sb); 1407d8a8559cSJens Axboe } 1408d8a8559cSJens Axboe EXPORT_SYMBOL(sync_inodes_sb); 14091da177e4SLinus Torvalds 14101da177e4SLinus Torvalds /** 14111da177e4SLinus Torvalds * write_inode_now - write an inode to disk 14121da177e4SLinus Torvalds * @inode: inode to write to disk 14131da177e4SLinus Torvalds * @sync: whether the write should be synchronous or not 14141da177e4SLinus Torvalds * 14157f04c26dSAndrea Arcangeli * This function commits an inode to disk immediately if it is dirty. This is 14167f04c26dSAndrea Arcangeli * primarily needed by knfsd. 14177f04c26dSAndrea Arcangeli * 14187f04c26dSAndrea Arcangeli * The caller must either have a ref on the inode or must have set I_WILL_FREE. 14191da177e4SLinus Torvalds */ 14201da177e4SLinus Torvalds int write_inode_now(struct inode *inode, int sync) 14211da177e4SLinus Torvalds { 1422f758eeabSChristoph Hellwig struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; 14231da177e4SLinus Torvalds struct writeback_control wbc = { 14241da177e4SLinus Torvalds .nr_to_write = LONG_MAX, 142518914b18SMike Galbraith .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, 1426111ebb6eSOGAWA Hirofumi .range_start = 0, 1427111ebb6eSOGAWA Hirofumi .range_end = LLONG_MAX, 14281da177e4SLinus Torvalds }; 14291da177e4SLinus Torvalds 14301da177e4SLinus Torvalds if (!mapping_cap_writeback_dirty(inode->i_mapping)) 143149364ce2SAndrew Morton wbc.nr_to_write = 0; 14321da177e4SLinus Torvalds 14331da177e4SLinus Torvalds might_sleep(); 14344f8ad655SJan Kara return writeback_single_inode(inode, wb, &wbc); 14351da177e4SLinus Torvalds } 14361da177e4SLinus Torvalds EXPORT_SYMBOL(write_inode_now); 14371da177e4SLinus Torvalds 14381da177e4SLinus Torvalds /** 14391da177e4SLinus Torvalds * sync_inode - write an inode and its pages to disk. 14401da177e4SLinus Torvalds * @inode: the inode to sync 14411da177e4SLinus Torvalds * @wbc: controls the writeback mode 14421da177e4SLinus Torvalds * 14431da177e4SLinus Torvalds * sync_inode() will write an inode and its pages to disk. It will also 14441da177e4SLinus Torvalds * correctly update the inode on its superblock's dirty inode lists and will 14451da177e4SLinus Torvalds * update inode->i_state. 14461da177e4SLinus Torvalds * 14471da177e4SLinus Torvalds * The caller must have a ref on the inode. 14481da177e4SLinus Torvalds */ 14491da177e4SLinus Torvalds int sync_inode(struct inode *inode, struct writeback_control *wbc) 14501da177e4SLinus Torvalds { 14514f8ad655SJan Kara return writeback_single_inode(inode, &inode_to_bdi(inode)->wb, wbc); 14521da177e4SLinus Torvalds } 14531da177e4SLinus Torvalds EXPORT_SYMBOL(sync_inode); 1454c3765016SChristoph Hellwig 1455c3765016SChristoph Hellwig /** 1456c691b9d9SAndrew Morton * sync_inode_metadata - write an inode to disk 1457c3765016SChristoph Hellwig * @inode: the inode to sync 1458c3765016SChristoph Hellwig * @wait: wait for I/O to complete. 1459c3765016SChristoph Hellwig * 1460c691b9d9SAndrew Morton * Write an inode to disk and adjust its dirty state after completion. 1461c3765016SChristoph Hellwig * 1462c3765016SChristoph Hellwig * Note: only writes the actual inode, no associated data or other metadata. 1463c3765016SChristoph Hellwig */ 1464c3765016SChristoph Hellwig int sync_inode_metadata(struct inode *inode, int wait) 1465c3765016SChristoph Hellwig { 1466c3765016SChristoph Hellwig struct writeback_control wbc = { 1467c3765016SChristoph Hellwig .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE, 1468c3765016SChristoph Hellwig .nr_to_write = 0, /* metadata-only */ 1469c3765016SChristoph Hellwig }; 1470c3765016SChristoph Hellwig 1471c3765016SChristoph Hellwig return sync_inode(inode, &wbc); 1472c3765016SChristoph Hellwig } 1473c3765016SChristoph Hellwig EXPORT_SYMBOL(sync_inode_metadata); 1474