11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * fs/fs-writeback.c 31da177e4SLinus Torvalds * 41da177e4SLinus Torvalds * Copyright (C) 2002, Linus Torvalds. 51da177e4SLinus Torvalds * 61da177e4SLinus Torvalds * Contains all the functions related to writing back and waiting 71da177e4SLinus Torvalds * upon dirty inodes against superblocks, and writing back dirty 81da177e4SLinus Torvalds * pages against inodes. ie: data writeback. Writeout of the 91da177e4SLinus Torvalds * inode itself is not handled here. 101da177e4SLinus Torvalds * 11e1f8e874SFrancois Cami * 10Apr2002 Andrew Morton 121da177e4SLinus Torvalds * Split out of fs/inode.c 131da177e4SLinus Torvalds * Additions for address_space-based writeback 141da177e4SLinus Torvalds */ 151da177e4SLinus Torvalds 161da177e4SLinus Torvalds #include <linux/kernel.h> 17630d9c47SPaul Gortmaker #include <linux/export.h> 181da177e4SLinus Torvalds #include <linux/spinlock.h> 195a0e3ad6STejun Heo #include <linux/slab.h> 201da177e4SLinus Torvalds #include <linux/sched.h> 211da177e4SLinus Torvalds #include <linux/fs.h> 221da177e4SLinus Torvalds #include <linux/mm.h> 23bc31b86aSWu Fengguang #include <linux/pagemap.h> 2403ba3782SJens Axboe #include <linux/kthread.h> 251da177e4SLinus Torvalds #include <linux/writeback.h> 261da177e4SLinus Torvalds #include <linux/blkdev.h> 271da177e4SLinus Torvalds #include <linux/backing-dev.h> 28455b2864SDave Chinner #include <linux/tracepoint.h> 29719ea2fbSAl Viro #include <linux/device.h> 3007f3f05cSDavid Howells #include "internal.h" 311da177e4SLinus Torvalds 32d0bceac7SJens Axboe /* 33bc31b86aSWu Fengguang * 4MB minimal write chunk size 34bc31b86aSWu Fengguang */ 35bc31b86aSWu Fengguang #define MIN_WRITEBACK_PAGES (4096UL >> (PAGE_CACHE_SHIFT - 10)) 36bc31b86aSWu Fengguang 37bc31b86aSWu Fengguang /* 38c4a77a6cSJens Axboe * Passed into wb_writeback(), essentially a subset of writeback_control 39c4a77a6cSJens Axboe */ 4083ba7b07SChristoph Hellwig struct wb_writeback_work { 41c4a77a6cSJens Axboe long nr_pages; 42c4a77a6cSJens Axboe struct super_block *sb; 430dc83bd3SJan Kara unsigned long *older_than_this; 44c4a77a6cSJens Axboe enum writeback_sync_modes sync_mode; 456e6938b6SWu Fengguang unsigned int tagged_writepages:1; 4652957fe1SH Hartley Sweeten unsigned int for_kupdate:1; 4752957fe1SH Hartley Sweeten unsigned int range_cyclic:1; 4852957fe1SH Hartley Sweeten unsigned int for_background:1; 497747bd4bSDave Chinner unsigned int for_sync:1; /* sync(2) WB_SYNC_ALL writeback */ 500e175a18SCurt Wohlgemuth enum wb_reason reason; /* why was writeback initiated? */ 51c4a77a6cSJens Axboe 528010c3b6SJens Axboe struct list_head list; /* pending work list */ 5383ba7b07SChristoph Hellwig struct completion *done; /* set if the caller waits */ 5403ba3782SJens Axboe }; 5503ba3782SJens Axboe 56f11b00f3SAdrian Bunk /** 57f11b00f3SAdrian Bunk * writeback_in_progress - determine whether there is writeback in progress 58f11b00f3SAdrian Bunk * @bdi: the device's backing_dev_info structure. 59f11b00f3SAdrian Bunk * 6003ba3782SJens Axboe * Determine whether there is writeback waiting to be handled against a 6103ba3782SJens Axboe * backing device. 62f11b00f3SAdrian Bunk */ 63f11b00f3SAdrian Bunk int writeback_in_progress(struct backing_dev_info *bdi) 64f11b00f3SAdrian Bunk { 6581d73a32SJan Kara return test_bit(BDI_writeback_running, &bdi->state); 66f11b00f3SAdrian Bunk } 6700d4e736STheodore Ts'o EXPORT_SYMBOL(writeback_in_progress); 68f11b00f3SAdrian Bunk 69692ebd17SJan Kara static inline struct backing_dev_info *inode_to_bdi(struct inode *inode) 70692ebd17SJan Kara { 71692ebd17SJan Kara struct super_block *sb = inode->i_sb; 72692ebd17SJan Kara 73a8855990SJan Kara if (sb_is_blkdev_sb(sb)) 74aaead25bSChristoph Hellwig return inode->i_mapping->backing_dev_info; 75aaead25bSChristoph Hellwig 76692ebd17SJan Kara return sb->s_bdi; 77692ebd17SJan Kara } 78692ebd17SJan Kara 797ccf19a8SNick Piggin static inline struct inode *wb_inode(struct list_head *head) 807ccf19a8SNick Piggin { 817ccf19a8SNick Piggin return list_entry(head, struct inode, i_wb_list); 827ccf19a8SNick Piggin } 837ccf19a8SNick Piggin 8415eb77a0SWu Fengguang /* 8515eb77a0SWu Fengguang * Include the creation of the trace points after defining the 8615eb77a0SWu Fengguang * wb_writeback_work structure and inline functions so that the definition 8715eb77a0SWu Fengguang * remains local to this file. 8815eb77a0SWu Fengguang */ 8915eb77a0SWu Fengguang #define CREATE_TRACE_POINTS 9015eb77a0SWu Fengguang #include <trace/events/writeback.h> 9115eb77a0SWu Fengguang 925acda9d1SJan Kara static void bdi_wakeup_thread(struct backing_dev_info *bdi) 935acda9d1SJan Kara { 945acda9d1SJan Kara spin_lock_bh(&bdi->wb_lock); 955acda9d1SJan Kara if (test_bit(BDI_registered, &bdi->state)) 965acda9d1SJan Kara mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0); 975acda9d1SJan Kara spin_unlock_bh(&bdi->wb_lock); 985acda9d1SJan Kara } 995acda9d1SJan Kara 1006585027aSJan Kara static void bdi_queue_work(struct backing_dev_info *bdi, 1016585027aSJan Kara struct wb_writeback_work *work) 1026585027aSJan Kara { 1036585027aSJan Kara trace_writeback_queue(bdi, work); 1046585027aSJan Kara 1056585027aSJan Kara spin_lock_bh(&bdi->wb_lock); 1065acda9d1SJan Kara if (!test_bit(BDI_registered, &bdi->state)) { 1075acda9d1SJan Kara if (work->done) 1085acda9d1SJan Kara complete(work->done); 1095acda9d1SJan Kara goto out_unlock; 1105acda9d1SJan Kara } 1116585027aSJan Kara list_add_tail(&work->list, &bdi->work_list); 112839a8e86STejun Heo mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0); 1135acda9d1SJan Kara out_unlock: 1145acda9d1SJan Kara 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 trace_writeback_nowork(bdi); 1305acda9d1SJan Kara bdi_wakeup_thread(bdi); 13183ba7b07SChristoph Hellwig return; 13283ba7b07SChristoph Hellwig } 13383ba7b07SChristoph Hellwig 13483ba7b07SChristoph Hellwig work->sync_mode = WB_SYNC_NONE; 13583ba7b07SChristoph Hellwig work->nr_pages = nr_pages; 13683ba7b07SChristoph Hellwig work->range_cyclic = range_cyclic; 1370e175a18SCurt Wohlgemuth work->reason = reason; 13883ba7b07SChristoph Hellwig 139f11fcae8SJens Axboe bdi_queue_work(bdi, work); 14003ba3782SJens Axboe } 141b6e51316SJens Axboe 142b6e51316SJens Axboe /** 143b6e51316SJens Axboe * bdi_start_writeback - start writeback 144b6e51316SJens Axboe * @bdi: the backing device to write from 145b6e51316SJens Axboe * @nr_pages: the number of pages to write 146786228abSMarcos Paulo de Souza * @reason: reason why some writeback work was initiated 147b6e51316SJens Axboe * 148b6e51316SJens Axboe * Description: 149b6e51316SJens Axboe * This does WB_SYNC_NONE opportunistic writeback. The IO is only 15025985edcSLucas De Marchi * started when this function returns, we make no guarantees on 1510e3c9a22SJens Axboe * completion. Caller need not hold sb s_umount semaphore. 152b6e51316SJens Axboe * 153b6e51316SJens Axboe */ 1540e175a18SCurt Wohlgemuth void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, 1550e175a18SCurt Wohlgemuth enum wb_reason reason) 156b6e51316SJens Axboe { 1570e175a18SCurt Wohlgemuth __bdi_start_writeback(bdi, nr_pages, true, reason); 158d3ddec76SWu Fengguang } 159d3ddec76SWu Fengguang 160c5444198SChristoph Hellwig /** 161c5444198SChristoph Hellwig * bdi_start_background_writeback - start background writeback 162c5444198SChristoph Hellwig * @bdi: the backing device to write from 163c5444198SChristoph Hellwig * 164c5444198SChristoph Hellwig * Description: 1656585027aSJan Kara * This makes sure WB_SYNC_NONE background writeback happens. When 1666585027aSJan Kara * this function returns, it is only guaranteed that for given BDI 1676585027aSJan Kara * some IO is happening if we are over background dirty threshold. 1686585027aSJan Kara * Caller need not hold sb s_umount semaphore. 169c5444198SChristoph Hellwig */ 170c5444198SChristoph Hellwig void bdi_start_background_writeback(struct backing_dev_info *bdi) 171c5444198SChristoph Hellwig { 1726585027aSJan Kara /* 1736585027aSJan Kara * We just wake up the flusher thread. It will perform background 1746585027aSJan Kara * writeback as soon as there is no other work to do. 1756585027aSJan Kara */ 17671927e84SWu Fengguang trace_writeback_wake_background(bdi); 1775acda9d1SJan Kara bdi_wakeup_thread(bdi); 1781da177e4SLinus Torvalds } 1791da177e4SLinus Torvalds 1801da177e4SLinus Torvalds /* 181a66979abSDave Chinner * Remove the inode from the writeback list it is on. 182a66979abSDave Chinner */ 183a66979abSDave Chinner void inode_wb_list_del(struct inode *inode) 184a66979abSDave Chinner { 185f758eeabSChristoph Hellwig struct backing_dev_info *bdi = inode_to_bdi(inode); 186a66979abSDave Chinner 187f758eeabSChristoph Hellwig spin_lock(&bdi->wb.list_lock); 188f758eeabSChristoph Hellwig list_del_init(&inode->i_wb_list); 189f758eeabSChristoph Hellwig spin_unlock(&bdi->wb.list_lock); 190f758eeabSChristoph Hellwig } 191a66979abSDave Chinner 192a66979abSDave Chinner /* 1936610a0bcSAndrew Morton * Redirty an inode: set its when-it-was dirtied timestamp and move it to the 1946610a0bcSAndrew Morton * furthest end of its superblock's dirty-inode list. 1956610a0bcSAndrew Morton * 1966610a0bcSAndrew Morton * Before stamping the inode's ->dirtied_when, we check to see whether it is 19766f3b8e2SJens Axboe * already the most-recently-dirtied inode on the b_dirty list. If that is 1986610a0bcSAndrew Morton * the case then the inode must have been redirtied while it was being written 1996610a0bcSAndrew Morton * out and we don't reset its dirtied_when. 2006610a0bcSAndrew Morton */ 201f758eeabSChristoph Hellwig static void redirty_tail(struct inode *inode, struct bdi_writeback *wb) 2026610a0bcSAndrew Morton { 203f758eeabSChristoph Hellwig assert_spin_locked(&wb->list_lock); 20403ba3782SJens Axboe if (!list_empty(&wb->b_dirty)) { 20566f3b8e2SJens Axboe struct inode *tail; 2066610a0bcSAndrew Morton 2077ccf19a8SNick Piggin tail = wb_inode(wb->b_dirty.next); 20866f3b8e2SJens Axboe if (time_before(inode->dirtied_when, tail->dirtied_when)) 2096610a0bcSAndrew Morton inode->dirtied_when = jiffies; 2106610a0bcSAndrew Morton } 2117ccf19a8SNick Piggin list_move(&inode->i_wb_list, &wb->b_dirty); 2126610a0bcSAndrew Morton } 2136610a0bcSAndrew Morton 2146610a0bcSAndrew Morton /* 21566f3b8e2SJens Axboe * requeue inode for re-scanning after bdi->b_io list is exhausted. 216c986d1e2SAndrew Morton */ 217f758eeabSChristoph Hellwig static void requeue_io(struct inode *inode, struct bdi_writeback *wb) 218c986d1e2SAndrew Morton { 219f758eeabSChristoph Hellwig assert_spin_locked(&wb->list_lock); 2207ccf19a8SNick Piggin list_move(&inode->i_wb_list, &wb->b_more_io); 221c986d1e2SAndrew Morton } 222c986d1e2SAndrew Morton 2231c0eeaf5SJoern Engel static void inode_sync_complete(struct inode *inode) 2241c0eeaf5SJoern Engel { 225365b94aeSJan Kara inode->i_state &= ~I_SYNC; 2264eff96ddSJan Kara /* If inode is clean an unused, put it into LRU now... */ 2274eff96ddSJan Kara inode_add_lru(inode); 228365b94aeSJan Kara /* Waiters must see I_SYNC cleared before being woken up */ 2291c0eeaf5SJoern Engel smp_mb(); 2301c0eeaf5SJoern Engel wake_up_bit(&inode->i_state, __I_SYNC); 2311c0eeaf5SJoern Engel } 2321c0eeaf5SJoern Engel 233d2caa3c5SJeff Layton static bool inode_dirtied_after(struct inode *inode, unsigned long t) 234d2caa3c5SJeff Layton { 235d2caa3c5SJeff Layton bool ret = time_after(inode->dirtied_when, t); 236d2caa3c5SJeff Layton #ifndef CONFIG_64BIT 237d2caa3c5SJeff Layton /* 238d2caa3c5SJeff Layton * For inodes being constantly redirtied, dirtied_when can get stuck. 239d2caa3c5SJeff Layton * It _appears_ to be in the future, but is actually in distant past. 240d2caa3c5SJeff Layton * This test is necessary to prevent such wrapped-around relative times 2415b0830cbSJens Axboe * from permanently stopping the whole bdi writeback. 242d2caa3c5SJeff Layton */ 243d2caa3c5SJeff Layton ret = ret && time_before_eq(inode->dirtied_when, jiffies); 244d2caa3c5SJeff Layton #endif 245d2caa3c5SJeff Layton return ret; 246d2caa3c5SJeff Layton } 247d2caa3c5SJeff Layton 248c986d1e2SAndrew Morton /* 2490e2f2b23SWang Sheng-Hui * Move expired (dirtied before work->older_than_this) dirty inodes from 250697e6fedSJan Kara * @delaying_queue to @dispatch_queue. 2512c136579SFengguang Wu */ 252e84d0a4fSWu Fengguang static int move_expired_inodes(struct list_head *delaying_queue, 2532c136579SFengguang Wu struct list_head *dispatch_queue, 254ad4e38ddSCurt Wohlgemuth struct wb_writeback_work *work) 2552c136579SFengguang Wu { 2565c03449dSShaohua Li LIST_HEAD(tmp); 2575c03449dSShaohua Li struct list_head *pos, *node; 258cf137307SJens Axboe struct super_block *sb = NULL; 2595c03449dSShaohua Li struct inode *inode; 260cf137307SJens Axboe int do_sb_sort = 0; 261e84d0a4fSWu Fengguang int moved = 0; 2625c03449dSShaohua Li 2632c136579SFengguang Wu while (!list_empty(delaying_queue)) { 2647ccf19a8SNick Piggin inode = wb_inode(delaying_queue->prev); 2650dc83bd3SJan Kara if (work->older_than_this && 2660dc83bd3SJan Kara inode_dirtied_after(inode, *work->older_than_this)) 2672c136579SFengguang Wu break; 268a8855990SJan Kara list_move(&inode->i_wb_list, &tmp); 269a8855990SJan Kara moved++; 270a8855990SJan Kara if (sb_is_blkdev_sb(inode->i_sb)) 271a8855990SJan Kara continue; 272cf137307SJens Axboe if (sb && sb != inode->i_sb) 273cf137307SJens Axboe do_sb_sort = 1; 274cf137307SJens Axboe sb = inode->i_sb; 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 { 3189fb0a7daSTejun Heo int ret; 3199fb0a7daSTejun Heo 3209fb0a7daSTejun Heo if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) { 3219fb0a7daSTejun Heo trace_writeback_write_inode_start(inode, wbc); 3229fb0a7daSTejun Heo ret = inode->i_sb->s_op->write_inode(inode, wbc); 3239fb0a7daSTejun Heo trace_writeback_write_inode(inode, wbc); 3249fb0a7daSTejun Heo return ret; 3259fb0a7daSTejun Heo } 32603ba3782SJens Axboe return 0; 32766f3b8e2SJens Axboe } 32808d8e974SFengguang Wu 3292c136579SFengguang Wu /* 330169ebd90SJan Kara * Wait for writeback on an inode to complete. Called with i_lock held. 331169ebd90SJan Kara * Caller must make sure inode cannot go away when we drop i_lock. 33201c03194SChristoph Hellwig */ 333169ebd90SJan Kara static void __inode_wait_for_writeback(struct inode *inode) 334169ebd90SJan Kara __releases(inode->i_lock) 335169ebd90SJan Kara __acquires(inode->i_lock) 33601c03194SChristoph Hellwig { 33701c03194SChristoph Hellwig DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); 33801c03194SChristoph Hellwig wait_queue_head_t *wqh; 33901c03194SChristoph Hellwig 34001c03194SChristoph Hellwig wqh = bit_waitqueue(&inode->i_state, __I_SYNC); 34158a9d3d8SRichard Kennedy while (inode->i_state & I_SYNC) { 342250df6edSDave Chinner spin_unlock(&inode->i_lock); 34301c03194SChristoph Hellwig __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE); 344250df6edSDave Chinner spin_lock(&inode->i_lock); 34558a9d3d8SRichard Kennedy } 34601c03194SChristoph Hellwig } 34701c03194SChristoph Hellwig 34801c03194SChristoph Hellwig /* 349169ebd90SJan Kara * Wait for writeback on an inode to complete. Caller must have inode pinned. 350169ebd90SJan Kara */ 351169ebd90SJan Kara void inode_wait_for_writeback(struct inode *inode) 352169ebd90SJan Kara { 353169ebd90SJan Kara spin_lock(&inode->i_lock); 354169ebd90SJan Kara __inode_wait_for_writeback(inode); 355169ebd90SJan Kara spin_unlock(&inode->i_lock); 356169ebd90SJan Kara } 357169ebd90SJan Kara 358169ebd90SJan Kara /* 359169ebd90SJan Kara * Sleep until I_SYNC is cleared. This function must be called with i_lock 360169ebd90SJan Kara * held and drops it. It is aimed for callers not holding any inode reference 361169ebd90SJan Kara * so once i_lock is dropped, inode can go away. 362169ebd90SJan Kara */ 363169ebd90SJan Kara static void inode_sleep_on_writeback(struct inode *inode) 364169ebd90SJan Kara __releases(inode->i_lock) 365169ebd90SJan Kara { 366169ebd90SJan Kara DEFINE_WAIT(wait); 367169ebd90SJan Kara wait_queue_head_t *wqh = bit_waitqueue(&inode->i_state, __I_SYNC); 368169ebd90SJan Kara int sleep; 369169ebd90SJan Kara 370169ebd90SJan Kara prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); 371169ebd90SJan Kara sleep = inode->i_state & I_SYNC; 372169ebd90SJan Kara spin_unlock(&inode->i_lock); 373169ebd90SJan Kara if (sleep) 374169ebd90SJan Kara schedule(); 375169ebd90SJan Kara finish_wait(wqh, &wait); 376169ebd90SJan Kara } 377169ebd90SJan Kara 378169ebd90SJan Kara /* 379ccb26b5aSJan Kara * Find proper writeback list for the inode depending on its current state and 380ccb26b5aSJan Kara * possibly also change of its state while we were doing writeback. Here we 381ccb26b5aSJan Kara * handle things such as livelock prevention or fairness of writeback among 382ccb26b5aSJan Kara * inodes. This function can be called only by flusher thread - noone else 383ccb26b5aSJan Kara * processes all inodes in writeback lists and requeueing inodes behind flusher 384ccb26b5aSJan Kara * thread's back can have unexpected consequences. 385ccb26b5aSJan Kara */ 386ccb26b5aSJan Kara static void requeue_inode(struct inode *inode, struct bdi_writeback *wb, 387ccb26b5aSJan Kara struct writeback_control *wbc) 388ccb26b5aSJan Kara { 389ccb26b5aSJan Kara if (inode->i_state & I_FREEING) 390ccb26b5aSJan Kara return; 391ccb26b5aSJan Kara 392ccb26b5aSJan Kara /* 393ccb26b5aSJan Kara * Sync livelock prevention. Each inode is tagged and synced in one 394ccb26b5aSJan Kara * shot. If still dirty, it will be redirty_tail()'ed below. Update 395ccb26b5aSJan Kara * the dirty time to prevent enqueue and sync it again. 396ccb26b5aSJan Kara */ 397ccb26b5aSJan Kara if ((inode->i_state & I_DIRTY) && 398ccb26b5aSJan Kara (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)) 399ccb26b5aSJan Kara inode->dirtied_when = jiffies; 400ccb26b5aSJan Kara 4014f8ad655SJan Kara if (wbc->pages_skipped) { 4024f8ad655SJan Kara /* 4034f8ad655SJan Kara * writeback is not making progress due to locked 4044f8ad655SJan Kara * buffers. Skip this inode for now. 4054f8ad655SJan Kara */ 4064f8ad655SJan Kara redirty_tail(inode, wb); 4074f8ad655SJan Kara return; 4084f8ad655SJan Kara } 4094f8ad655SJan Kara 410ccb26b5aSJan Kara if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) { 411ccb26b5aSJan Kara /* 412ccb26b5aSJan Kara * We didn't write back all the pages. nfs_writepages() 413ccb26b5aSJan Kara * sometimes bales out without doing anything. 414ccb26b5aSJan Kara */ 415ccb26b5aSJan Kara if (wbc->nr_to_write <= 0) { 416ccb26b5aSJan Kara /* Slice used up. Queue for next turn. */ 417ccb26b5aSJan Kara requeue_io(inode, wb); 418ccb26b5aSJan Kara } else { 419ccb26b5aSJan Kara /* 420ccb26b5aSJan Kara * Writeback blocked by something other than 421ccb26b5aSJan Kara * congestion. Delay the inode for some time to 422ccb26b5aSJan Kara * avoid spinning on the CPU (100% iowait) 423ccb26b5aSJan Kara * retrying writeback of the dirty page/inode 424ccb26b5aSJan Kara * that cannot be performed immediately. 425ccb26b5aSJan Kara */ 426ccb26b5aSJan Kara redirty_tail(inode, wb); 427ccb26b5aSJan Kara } 428ccb26b5aSJan Kara } else if (inode->i_state & I_DIRTY) { 429ccb26b5aSJan Kara /* 430ccb26b5aSJan Kara * Filesystems can dirty the inode during writeback operations, 431ccb26b5aSJan Kara * such as delayed allocation during submission or metadata 432ccb26b5aSJan Kara * updates after data IO completion. 433ccb26b5aSJan Kara */ 434ccb26b5aSJan Kara redirty_tail(inode, wb); 435ccb26b5aSJan Kara } else { 436ccb26b5aSJan Kara /* The inode is clean. Remove from writeback lists. */ 437ccb26b5aSJan Kara list_del_init(&inode->i_wb_list); 438ccb26b5aSJan Kara } 439ccb26b5aSJan Kara } 440ccb26b5aSJan Kara 441ccb26b5aSJan Kara /* 4424f8ad655SJan Kara * Write out an inode and its dirty pages. Do not update the writeback list 4434f8ad655SJan Kara * linkage. That is left to the caller. The caller is also responsible for 4444f8ad655SJan Kara * setting I_SYNC flag and calling inode_sync_complete() to clear it. 4451da177e4SLinus Torvalds */ 4461da177e4SLinus Torvalds static int 447cd8ed2a4SYan Hong __writeback_single_inode(struct inode *inode, 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 4569fb0a7daSTejun Heo trace_writeback_single_inode_start(inode, wbc, nr_to_write); 4579fb0a7daSTejun Heo 4581da177e4SLinus Torvalds ret = do_writepages(mapping, wbc); 4591da177e4SLinus Torvalds 46026821ed4SChristoph Hellwig /* 46126821ed4SChristoph Hellwig * Make sure to wait on the data before writing out the metadata. 46226821ed4SChristoph Hellwig * This is important for filesystems that modify metadata on data 4637747bd4bSDave Chinner * I/O completion. We don't do it for sync(2) writeback because it has a 4647747bd4bSDave Chinner * separate, external IO completion path and ->sync_fs for guaranteeing 4657747bd4bSDave Chinner * inode metadata is written back correctly. 46626821ed4SChristoph Hellwig */ 4677747bd4bSDave Chinner if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) { 46826821ed4SChristoph Hellwig int err = filemap_fdatawait(mapping); 4691da177e4SLinus Torvalds if (ret == 0) 4701da177e4SLinus Torvalds ret = err; 4711da177e4SLinus Torvalds } 4721da177e4SLinus Torvalds 4735547e8aaSDmitry Monakhov /* 4745547e8aaSDmitry Monakhov * Some filesystems may redirty the inode during the writeback 4755547e8aaSDmitry Monakhov * due to delalloc, clear dirty metadata flags right before 4765547e8aaSDmitry Monakhov * write_inode() 4775547e8aaSDmitry Monakhov */ 478250df6edSDave Chinner spin_lock(&inode->i_lock); 4796290be1cSJan Kara /* Clear I_DIRTY_PAGES if we've written out all dirty pages */ 4806290be1cSJan Kara if (!mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) 4816290be1cSJan Kara inode->i_state &= ~I_DIRTY_PAGES; 4825547e8aaSDmitry Monakhov dirty = inode->i_state & I_DIRTY; 4835547e8aaSDmitry Monakhov inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC); 484250df6edSDave Chinner spin_unlock(&inode->i_lock); 48526821ed4SChristoph Hellwig /* Don't write the inode if only I_DIRTY_PAGES was set */ 48626821ed4SChristoph Hellwig if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { 487a9185b41SChristoph Hellwig int err = write_inode(inode, wbc); 4881da177e4SLinus Torvalds if (ret == 0) 4891da177e4SLinus Torvalds ret = err; 4901da177e4SLinus Torvalds } 4914f8ad655SJan Kara trace_writeback_single_inode(inode, wbc, nr_to_write); 4924f8ad655SJan Kara return ret; 4934f8ad655SJan Kara } 4944f8ad655SJan Kara 4954f8ad655SJan Kara /* 4964f8ad655SJan Kara * Write out an inode's dirty pages. Either the caller has an active reference 4974f8ad655SJan Kara * on the inode or the inode has I_WILL_FREE set. 4984f8ad655SJan Kara * 4994f8ad655SJan Kara * This function is designed to be called for writing back one inode which 5004f8ad655SJan Kara * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode() 5014f8ad655SJan Kara * and does more profound writeback list handling in writeback_sb_inodes(). 5024f8ad655SJan Kara */ 5034f8ad655SJan Kara static int 5044f8ad655SJan Kara writeback_single_inode(struct inode *inode, struct bdi_writeback *wb, 5054f8ad655SJan Kara struct writeback_control *wbc) 5064f8ad655SJan Kara { 5074f8ad655SJan Kara int ret = 0; 5084f8ad655SJan Kara 5094f8ad655SJan Kara spin_lock(&inode->i_lock); 5104f8ad655SJan Kara if (!atomic_read(&inode->i_count)) 5114f8ad655SJan Kara WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); 5124f8ad655SJan Kara else 5134f8ad655SJan Kara WARN_ON(inode->i_state & I_WILL_FREE); 5144f8ad655SJan Kara 5154f8ad655SJan Kara if (inode->i_state & I_SYNC) { 5164f8ad655SJan Kara if (wbc->sync_mode != WB_SYNC_ALL) 5174f8ad655SJan Kara goto out; 5184f8ad655SJan Kara /* 519169ebd90SJan Kara * It's a data-integrity sync. We must wait. Since callers hold 520169ebd90SJan Kara * inode reference or inode has I_WILL_FREE set, it cannot go 521169ebd90SJan Kara * away under us. 5224f8ad655SJan Kara */ 523169ebd90SJan Kara __inode_wait_for_writeback(inode); 5244f8ad655SJan Kara } 5254f8ad655SJan Kara WARN_ON(inode->i_state & I_SYNC); 5264f8ad655SJan Kara /* 527f9b0e058SJan Kara * Skip inode if it is clean and we have no outstanding writeback in 528f9b0e058SJan Kara * WB_SYNC_ALL mode. We don't want to mess with writeback lists in this 529f9b0e058SJan Kara * function since flusher thread may be doing for example sync in 530f9b0e058SJan Kara * parallel and if we move the inode, it could get skipped. So here we 531f9b0e058SJan Kara * make sure inode is on some writeback list and leave it there unless 532f9b0e058SJan Kara * we have completely cleaned the inode. 5334f8ad655SJan Kara */ 534f9b0e058SJan Kara if (!(inode->i_state & I_DIRTY) && 535f9b0e058SJan Kara (wbc->sync_mode != WB_SYNC_ALL || 536f9b0e058SJan Kara !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK))) 5374f8ad655SJan Kara goto out; 5384f8ad655SJan Kara inode->i_state |= I_SYNC; 5394f8ad655SJan Kara spin_unlock(&inode->i_lock); 5404f8ad655SJan Kara 541cd8ed2a4SYan Hong ret = __writeback_single_inode(inode, wbc); 5421da177e4SLinus Torvalds 543f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 544250df6edSDave Chinner spin_lock(&inode->i_lock); 5454f8ad655SJan Kara /* 5464f8ad655SJan Kara * If inode is clean, remove it from writeback lists. Otherwise don't 5474f8ad655SJan Kara * touch it. See comment above for explanation. 5484f8ad655SJan Kara */ 5494f8ad655SJan Kara if (!(inode->i_state & I_DIRTY)) 5504f8ad655SJan Kara list_del_init(&inode->i_wb_list); 5514f8ad655SJan Kara spin_unlock(&wb->list_lock); 5521c0eeaf5SJoern Engel inode_sync_complete(inode); 5534f8ad655SJan Kara out: 5544f8ad655SJan Kara spin_unlock(&inode->i_lock); 5551da177e4SLinus Torvalds return ret; 5561da177e4SLinus Torvalds } 5571da177e4SLinus Torvalds 5581a12d8bdSWu Fengguang static long writeback_chunk_size(struct backing_dev_info *bdi, 5591a12d8bdSWu Fengguang struct wb_writeback_work *work) 560d46db3d5SWu Fengguang { 561d46db3d5SWu Fengguang long pages; 562d46db3d5SWu Fengguang 563d46db3d5SWu Fengguang /* 564d46db3d5SWu Fengguang * WB_SYNC_ALL mode does livelock avoidance by syncing dirty 565d46db3d5SWu Fengguang * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX 566d46db3d5SWu Fengguang * here avoids calling into writeback_inodes_wb() more than once. 567d46db3d5SWu Fengguang * 568d46db3d5SWu Fengguang * The intended call sequence for WB_SYNC_ALL writeback is: 569d46db3d5SWu Fengguang * 570d46db3d5SWu Fengguang * wb_writeback() 571d46db3d5SWu Fengguang * writeback_sb_inodes() <== called only once 572d46db3d5SWu Fengguang * write_cache_pages() <== called once for each inode 573d46db3d5SWu Fengguang * (quickly) tag currently dirty pages 574d46db3d5SWu Fengguang * (maybe slowly) sync all tagged pages 575d46db3d5SWu Fengguang */ 576d46db3d5SWu Fengguang if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages) 577d46db3d5SWu Fengguang pages = LONG_MAX; 5781a12d8bdSWu Fengguang else { 5791a12d8bdSWu Fengguang pages = min(bdi->avg_write_bandwidth / 2, 5801a12d8bdSWu Fengguang global_dirty_limit / DIRTY_SCOPE); 5811a12d8bdSWu Fengguang pages = min(pages, work->nr_pages); 5821a12d8bdSWu Fengguang pages = round_down(pages + MIN_WRITEBACK_PAGES, 5831a12d8bdSWu Fengguang MIN_WRITEBACK_PAGES); 5841a12d8bdSWu Fengguang } 585d46db3d5SWu Fengguang 586d46db3d5SWu Fengguang return pages; 587d46db3d5SWu Fengguang } 588d46db3d5SWu Fengguang 58903ba3782SJens Axboe /* 590f11c9c5cSEdward Shishkin * Write a portion of b_io inodes which belong to @sb. 591edadfb10SChristoph Hellwig * 592d46db3d5SWu Fengguang * Return the number of pages and/or inodes written. 593f11c9c5cSEdward Shishkin */ 594d46db3d5SWu Fengguang static long writeback_sb_inodes(struct super_block *sb, 595d46db3d5SWu Fengguang struct bdi_writeback *wb, 596d46db3d5SWu Fengguang struct wb_writeback_work *work) 59703ba3782SJens Axboe { 598d46db3d5SWu Fengguang struct writeback_control wbc = { 599d46db3d5SWu Fengguang .sync_mode = work->sync_mode, 600d46db3d5SWu Fengguang .tagged_writepages = work->tagged_writepages, 601d46db3d5SWu Fengguang .for_kupdate = work->for_kupdate, 602d46db3d5SWu Fengguang .for_background = work->for_background, 6037747bd4bSDave Chinner .for_sync = work->for_sync, 604d46db3d5SWu Fengguang .range_cyclic = work->range_cyclic, 605d46db3d5SWu Fengguang .range_start = 0, 606d46db3d5SWu Fengguang .range_end = LLONG_MAX, 607d46db3d5SWu Fengguang }; 608d46db3d5SWu Fengguang unsigned long start_time = jiffies; 609d46db3d5SWu Fengguang long write_chunk; 610d46db3d5SWu Fengguang long wrote = 0; /* count both pages and inodes */ 611d46db3d5SWu Fengguang 61203ba3782SJens Axboe while (!list_empty(&wb->b_io)) { 6137ccf19a8SNick Piggin struct inode *inode = wb_inode(wb->b_io.prev); 614edadfb10SChristoph Hellwig 615edadfb10SChristoph Hellwig if (inode->i_sb != sb) { 616d46db3d5SWu Fengguang if (work->sb) { 617edadfb10SChristoph Hellwig /* 618edadfb10SChristoph Hellwig * We only want to write back data for this 619edadfb10SChristoph Hellwig * superblock, move all inodes not belonging 620edadfb10SChristoph Hellwig * to it back onto the dirty list. 621edadfb10SChristoph Hellwig */ 622f758eeabSChristoph Hellwig redirty_tail(inode, wb); 62366f3b8e2SJens Axboe continue; 62466f3b8e2SJens Axboe } 625edadfb10SChristoph Hellwig 626edadfb10SChristoph Hellwig /* 627edadfb10SChristoph Hellwig * The inode belongs to a different superblock. 628edadfb10SChristoph Hellwig * Bounce back to the caller to unpin this and 629edadfb10SChristoph Hellwig * pin the next superblock. 630edadfb10SChristoph Hellwig */ 631d46db3d5SWu Fengguang break; 632edadfb10SChristoph Hellwig } 633edadfb10SChristoph Hellwig 6349843b76aSChristoph Hellwig /* 635331cbdeeSWanpeng Li * Don't bother with new inodes or inodes being freed, first 636331cbdeeSWanpeng Li * kind does not need periodic writeout yet, and for the latter 6379843b76aSChristoph Hellwig * kind writeout is handled by the freer. 6389843b76aSChristoph Hellwig */ 639250df6edSDave Chinner spin_lock(&inode->i_lock); 6409843b76aSChristoph Hellwig if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { 641250df6edSDave Chinner spin_unlock(&inode->i_lock); 642fcc5c222SWu Fengguang redirty_tail(inode, wb); 6437ef0d737SNick Piggin continue; 6447ef0d737SNick Piggin } 645cc1676d9SJan Kara if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) { 646cc1676d9SJan Kara /* 647cc1676d9SJan Kara * If this inode is locked for writeback and we are not 648cc1676d9SJan Kara * doing writeback-for-data-integrity, move it to 649cc1676d9SJan Kara * b_more_io so that writeback can proceed with the 650cc1676d9SJan Kara * other inodes on s_io. 651cc1676d9SJan Kara * 652cc1676d9SJan Kara * We'll have another go at writing back this inode 653cc1676d9SJan Kara * when we completed a full scan of b_io. 654cc1676d9SJan Kara */ 655cc1676d9SJan Kara spin_unlock(&inode->i_lock); 656cc1676d9SJan Kara requeue_io(inode, wb); 657cc1676d9SJan Kara trace_writeback_sb_inodes_requeue(inode); 658cc1676d9SJan Kara continue; 659cc1676d9SJan Kara } 660f0d07b7fSJan Kara spin_unlock(&wb->list_lock); 661f0d07b7fSJan Kara 6624f8ad655SJan Kara /* 6634f8ad655SJan Kara * We already requeued the inode if it had I_SYNC set and we 6644f8ad655SJan Kara * are doing WB_SYNC_NONE writeback. So this catches only the 6654f8ad655SJan Kara * WB_SYNC_ALL case. 6664f8ad655SJan Kara */ 667169ebd90SJan Kara if (inode->i_state & I_SYNC) { 668169ebd90SJan Kara /* Wait for I_SYNC. This function drops i_lock... */ 669169ebd90SJan Kara inode_sleep_on_writeback(inode); 670169ebd90SJan Kara /* Inode may be gone, start again */ 671ead188f9SJan Kara spin_lock(&wb->list_lock); 672169ebd90SJan Kara continue; 673169ebd90SJan Kara } 6744f8ad655SJan Kara inode->i_state |= I_SYNC; 6754f8ad655SJan Kara spin_unlock(&inode->i_lock); 676169ebd90SJan Kara 6771a12d8bdSWu Fengguang write_chunk = writeback_chunk_size(wb->bdi, work); 678d46db3d5SWu Fengguang wbc.nr_to_write = write_chunk; 679d46db3d5SWu Fengguang wbc.pages_skipped = 0; 680250df6edSDave Chinner 681169ebd90SJan Kara /* 682169ebd90SJan Kara * We use I_SYNC to pin the inode in memory. While it is set 683169ebd90SJan Kara * evict_inode() will wait so the inode cannot be freed. 684169ebd90SJan Kara */ 685cd8ed2a4SYan Hong __writeback_single_inode(inode, &wbc); 686d46db3d5SWu Fengguang 687d46db3d5SWu Fengguang work->nr_pages -= write_chunk - wbc.nr_to_write; 688d46db3d5SWu Fengguang wrote += write_chunk - wbc.nr_to_write; 6894f8ad655SJan Kara spin_lock(&wb->list_lock); 6904f8ad655SJan Kara spin_lock(&inode->i_lock); 691d46db3d5SWu Fengguang if (!(inode->i_state & I_DIRTY)) 692d46db3d5SWu Fengguang wrote++; 6934f8ad655SJan Kara requeue_inode(inode, wb, &wbc); 6944f8ad655SJan Kara inode_sync_complete(inode); 6950f1b1fd8SDave Chinner spin_unlock(&inode->i_lock); 696169ebd90SJan Kara cond_resched_lock(&wb->list_lock); 697d46db3d5SWu Fengguang /* 698d46db3d5SWu Fengguang * bail out to wb_writeback() often enough to check 699d46db3d5SWu Fengguang * background threshold and other termination conditions. 700d46db3d5SWu Fengguang */ 701d46db3d5SWu Fengguang if (wrote) { 702d46db3d5SWu Fengguang if (time_is_before_jiffies(start_time + HZ / 10UL)) 703d46db3d5SWu Fengguang break; 704d46db3d5SWu Fengguang if (work->nr_pages <= 0) 705d46db3d5SWu Fengguang break; 7061da177e4SLinus Torvalds } 7078bc3be27SFengguang Wu } 708d46db3d5SWu Fengguang return wrote; 709f11c9c5cSEdward Shishkin } 71038f21977SNick Piggin 711d46db3d5SWu Fengguang static long __writeback_inodes_wb(struct bdi_writeback *wb, 712d46db3d5SWu Fengguang struct wb_writeback_work *work) 713f11c9c5cSEdward Shishkin { 714d46db3d5SWu Fengguang unsigned long start_time = jiffies; 715d46db3d5SWu Fengguang long wrote = 0; 716f11c9c5cSEdward Shishkin 717f11c9c5cSEdward Shishkin while (!list_empty(&wb->b_io)) { 7187ccf19a8SNick Piggin struct inode *inode = wb_inode(wb->b_io.prev); 719f11c9c5cSEdward Shishkin struct super_block *sb = inode->i_sb; 720f11c9c5cSEdward Shishkin 72112ad3ab6SDave Chinner if (!grab_super_passive(sb)) { 7220e995816SWu Fengguang /* 7230e995816SWu Fengguang * grab_super_passive() may fail consistently due to 7240e995816SWu Fengguang * s_umount being grabbed by someone else. Don't use 7250e995816SWu Fengguang * requeue_io() to avoid busy retrying the inode/sb. 7260e995816SWu Fengguang */ 7270e995816SWu Fengguang redirty_tail(inode, wb); 728d19de7edSChristoph Hellwig continue; 729334132aeSChristoph Hellwig } 730d46db3d5SWu Fengguang wrote += writeback_sb_inodes(sb, wb, work); 731d19de7edSChristoph Hellwig drop_super(sb); 732f11c9c5cSEdward Shishkin 733d46db3d5SWu Fengguang /* refer to the same tests at the end of writeback_sb_inodes */ 734d46db3d5SWu Fengguang if (wrote) { 735d46db3d5SWu Fengguang if (time_is_before_jiffies(start_time + HZ / 10UL)) 736d46db3d5SWu Fengguang break; 737d46db3d5SWu Fengguang if (work->nr_pages <= 0) 738f11c9c5cSEdward Shishkin break; 739f11c9c5cSEdward Shishkin } 740d46db3d5SWu Fengguang } 74166f3b8e2SJens Axboe /* Leave any unwritten inodes on b_io */ 742d46db3d5SWu Fengguang return wrote; 74366f3b8e2SJens Axboe } 74466f3b8e2SJens Axboe 7457d9f073bSWanpeng Li static long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages, 7460e175a18SCurt Wohlgemuth enum wb_reason reason) 747edadfb10SChristoph Hellwig { 748d46db3d5SWu Fengguang struct wb_writeback_work work = { 749d46db3d5SWu Fengguang .nr_pages = nr_pages, 750d46db3d5SWu Fengguang .sync_mode = WB_SYNC_NONE, 751d46db3d5SWu Fengguang .range_cyclic = 1, 7520e175a18SCurt Wohlgemuth .reason = reason, 753d46db3d5SWu Fengguang }; 754edadfb10SChristoph Hellwig 755f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 756424b351fSWu Fengguang if (list_empty(&wb->b_io)) 757ad4e38ddSCurt Wohlgemuth queue_io(wb, &work); 758d46db3d5SWu Fengguang __writeback_inodes_wb(wb, &work); 759f758eeabSChristoph Hellwig spin_unlock(&wb->list_lock); 760edadfb10SChristoph Hellwig 761d46db3d5SWu Fengguang return nr_pages - work.nr_pages; 76266f3b8e2SJens Axboe } 76366f3b8e2SJens Axboe 764b00949aaSWu Fengguang static bool over_bground_thresh(struct backing_dev_info *bdi) 76503ba3782SJens Axboe { 76603ba3782SJens Axboe unsigned long background_thresh, dirty_thresh; 76703ba3782SJens Axboe 76816c4042fSWu Fengguang global_dirty_limits(&background_thresh, &dirty_thresh); 76903ba3782SJens Axboe 770b00949aaSWu Fengguang if (global_page_state(NR_FILE_DIRTY) + 771b00949aaSWu Fengguang global_page_state(NR_UNSTABLE_NFS) > background_thresh) 772b00949aaSWu Fengguang return true; 773b00949aaSWu Fengguang 774b00949aaSWu Fengguang if (bdi_stat(bdi, BDI_RECLAIMABLE) > 775b00949aaSWu Fengguang bdi_dirty_limit(bdi, background_thresh)) 776b00949aaSWu Fengguang return true; 777b00949aaSWu Fengguang 778b00949aaSWu Fengguang return false; 77903ba3782SJens Axboe } 78003ba3782SJens Axboe 78103ba3782SJens Axboe /* 782e98be2d5SWu Fengguang * Called under wb->list_lock. If there are multiple wb per bdi, 783e98be2d5SWu Fengguang * only the flusher working on the first wb should do it. 784e98be2d5SWu Fengguang */ 785e98be2d5SWu Fengguang static void wb_update_bandwidth(struct bdi_writeback *wb, 786e98be2d5SWu Fengguang unsigned long start_time) 787e98be2d5SWu Fengguang { 788af6a3113SWu Fengguang __bdi_update_bandwidth(wb->bdi, 0, 0, 0, 0, 0, start_time); 789e98be2d5SWu Fengguang } 790e98be2d5SWu Fengguang 791e98be2d5SWu Fengguang /* 79203ba3782SJens Axboe * Explicit flushing or periodic writeback of "old" data. 79303ba3782SJens Axboe * 79403ba3782SJens Axboe * Define "old": the first time one of an inode's pages is dirtied, we mark the 79503ba3782SJens Axboe * dirtying-time in the inode's address_space. So this periodic writeback code 79603ba3782SJens Axboe * just walks the superblock inode list, writing back any inodes which are 79703ba3782SJens Axboe * older than a specific point in time. 79803ba3782SJens Axboe * 79903ba3782SJens Axboe * Try to run once per dirty_writeback_interval. But if a writeback event 80003ba3782SJens Axboe * takes longer than a dirty_writeback_interval interval, then leave a 80103ba3782SJens Axboe * one-second gap. 80203ba3782SJens Axboe * 80303ba3782SJens Axboe * older_than_this takes precedence over nr_to_write. So we'll only write back 80403ba3782SJens Axboe * all dirty pages if they are all attached to "old" mappings. 80503ba3782SJens Axboe */ 806c4a77a6cSJens Axboe static long wb_writeback(struct bdi_writeback *wb, 80783ba7b07SChristoph Hellwig struct wb_writeback_work *work) 80803ba3782SJens Axboe { 809e98be2d5SWu Fengguang unsigned long wb_start = jiffies; 810d46db3d5SWu Fengguang long nr_pages = work->nr_pages; 8110dc83bd3SJan Kara unsigned long oldest_jif; 812a5989bdcSJan Kara struct inode *inode; 813d46db3d5SWu Fengguang long progress; 81403ba3782SJens Axboe 8150dc83bd3SJan Kara oldest_jif = jiffies; 8160dc83bd3SJan Kara work->older_than_this = &oldest_jif; 81703ba3782SJens Axboe 818e8dfc305SWu Fengguang spin_lock(&wb->list_lock); 81903ba3782SJens Axboe for (;;) { 82003ba3782SJens Axboe /* 821d3ddec76SWu Fengguang * Stop writeback when nr_pages has been consumed 82203ba3782SJens Axboe */ 82383ba7b07SChristoph Hellwig if (work->nr_pages <= 0) 82403ba3782SJens Axboe break; 82503ba3782SJens Axboe 82603ba3782SJens Axboe /* 827aa373cf5SJan Kara * Background writeout and kupdate-style writeback may 828aa373cf5SJan Kara * run forever. Stop them if there is other work to do 829aa373cf5SJan Kara * so that e.g. sync can proceed. They'll be restarted 830aa373cf5SJan Kara * after the other works are all done. 831aa373cf5SJan Kara */ 832aa373cf5SJan Kara if ((work->for_background || work->for_kupdate) && 833aa373cf5SJan Kara !list_empty(&wb->bdi->work_list)) 834aa373cf5SJan Kara break; 835aa373cf5SJan Kara 836aa373cf5SJan Kara /* 837d3ddec76SWu Fengguang * For background writeout, stop when we are below the 838d3ddec76SWu Fengguang * background dirty threshold 83903ba3782SJens Axboe */ 840b00949aaSWu Fengguang if (work->for_background && !over_bground_thresh(wb->bdi)) 84103ba3782SJens Axboe break; 84203ba3782SJens Axboe 8431bc36b64SJan Kara /* 8441bc36b64SJan Kara * Kupdate and background works are special and we want to 8451bc36b64SJan Kara * include all inodes that need writing. Livelock avoidance is 8461bc36b64SJan Kara * handled by these works yielding to any other work so we are 8471bc36b64SJan Kara * safe. 8481bc36b64SJan Kara */ 849ba9aa839SWu Fengguang if (work->for_kupdate) { 8500dc83bd3SJan Kara oldest_jif = jiffies - 851ba9aa839SWu Fengguang msecs_to_jiffies(dirty_expire_interval * 10); 8521bc36b64SJan Kara } else if (work->for_background) 8530dc83bd3SJan Kara oldest_jif = jiffies; 854028c2dd1SDave Chinner 855d46db3d5SWu Fengguang trace_writeback_start(wb->bdi, work); 856e8dfc305SWu Fengguang if (list_empty(&wb->b_io)) 857ad4e38ddSCurt Wohlgemuth queue_io(wb, work); 85883ba7b07SChristoph Hellwig if (work->sb) 859d46db3d5SWu Fengguang progress = writeback_sb_inodes(work->sb, wb, work); 860edadfb10SChristoph Hellwig else 861d46db3d5SWu Fengguang progress = __writeback_inodes_wb(wb, work); 862d46db3d5SWu Fengguang trace_writeback_written(wb->bdi, work); 863028c2dd1SDave Chinner 864e98be2d5SWu Fengguang wb_update_bandwidth(wb, wb_start); 86503ba3782SJens Axboe 86603ba3782SJens Axboe /* 86771fd05a8SJens Axboe * Did we write something? Try for more 868e6fb6da2SWu Fengguang * 869e6fb6da2SWu Fengguang * Dirty inodes are moved to b_io for writeback in batches. 870e6fb6da2SWu Fengguang * The completion of the current batch does not necessarily 871e6fb6da2SWu Fengguang * mean the overall work is done. So we keep looping as long 872e6fb6da2SWu Fengguang * as made some progress on cleaning pages or inodes. 87371fd05a8SJens Axboe */ 874d46db3d5SWu Fengguang if (progress) 87503ba3782SJens Axboe continue; 876a5989bdcSJan Kara /* 877e6fb6da2SWu Fengguang * No more inodes for IO, bail 878a5989bdcSJan Kara */ 879b7a2441fSWu Fengguang if (list_empty(&wb->b_more_io)) 88003ba3782SJens Axboe break; 88103ba3782SJens Axboe /* 8828010c3b6SJens Axboe * Nothing written. Wait for some inode to 8838010c3b6SJens Axboe * become available for writeback. Otherwise 8848010c3b6SJens Axboe * we'll just busyloop. 88503ba3782SJens Axboe */ 88603ba3782SJens Axboe if (!list_empty(&wb->b_more_io)) { 887d46db3d5SWu Fengguang trace_writeback_wait(wb->bdi, work); 88803ba3782SJens Axboe inode = wb_inode(wb->b_more_io.prev); 889250df6edSDave Chinner spin_lock(&inode->i_lock); 890f0d07b7fSJan Kara spin_unlock(&wb->list_lock); 891169ebd90SJan Kara /* This function drops i_lock... */ 892169ebd90SJan Kara inode_sleep_on_writeback(inode); 893f0d07b7fSJan Kara spin_lock(&wb->list_lock); 89403ba3782SJens Axboe } 89503ba3782SJens Axboe } 896e8dfc305SWu Fengguang spin_unlock(&wb->list_lock); 89703ba3782SJens Axboe 898d46db3d5SWu Fengguang return nr_pages - work->nr_pages; 89903ba3782SJens Axboe } 90003ba3782SJens Axboe 90103ba3782SJens Axboe /* 90283ba7b07SChristoph Hellwig * Return the next wb_writeback_work struct that hasn't been processed yet. 90303ba3782SJens Axboe */ 90483ba7b07SChristoph Hellwig static struct wb_writeback_work * 90508852b6dSMinchan Kim get_next_work_item(struct backing_dev_info *bdi) 90603ba3782SJens Axboe { 90783ba7b07SChristoph Hellwig struct wb_writeback_work *work = NULL; 90803ba3782SJens Axboe 9096467716aSArtem Bityutskiy spin_lock_bh(&bdi->wb_lock); 91083ba7b07SChristoph Hellwig if (!list_empty(&bdi->work_list)) { 91183ba7b07SChristoph Hellwig work = list_entry(bdi->work_list.next, 91283ba7b07SChristoph Hellwig struct wb_writeback_work, list); 91383ba7b07SChristoph Hellwig list_del_init(&work->list); 91403ba3782SJens Axboe } 9156467716aSArtem Bityutskiy spin_unlock_bh(&bdi->wb_lock); 91683ba7b07SChristoph Hellwig return work; 91703ba3782SJens Axboe } 91803ba3782SJens Axboe 919cdf01dd5SLinus Torvalds /* 920cdf01dd5SLinus Torvalds * Add in the number of potentially dirty inodes, because each inode 921cdf01dd5SLinus Torvalds * write can dirty pagecache in the underlying blockdev. 922cdf01dd5SLinus Torvalds */ 923cdf01dd5SLinus Torvalds static unsigned long get_nr_dirty_pages(void) 924cdf01dd5SLinus Torvalds { 925cdf01dd5SLinus Torvalds return global_page_state(NR_FILE_DIRTY) + 926cdf01dd5SLinus Torvalds global_page_state(NR_UNSTABLE_NFS) + 927cdf01dd5SLinus Torvalds get_nr_dirty_inodes(); 928cdf01dd5SLinus Torvalds } 929cdf01dd5SLinus Torvalds 9306585027aSJan Kara static long wb_check_background_flush(struct bdi_writeback *wb) 9316585027aSJan Kara { 932b00949aaSWu Fengguang if (over_bground_thresh(wb->bdi)) { 9336585027aSJan Kara 9346585027aSJan Kara struct wb_writeback_work work = { 9356585027aSJan Kara .nr_pages = LONG_MAX, 9366585027aSJan Kara .sync_mode = WB_SYNC_NONE, 9376585027aSJan Kara .for_background = 1, 9386585027aSJan Kara .range_cyclic = 1, 9390e175a18SCurt Wohlgemuth .reason = WB_REASON_BACKGROUND, 9406585027aSJan Kara }; 9416585027aSJan Kara 9426585027aSJan Kara return wb_writeback(wb, &work); 9436585027aSJan Kara } 9446585027aSJan Kara 9456585027aSJan Kara return 0; 9466585027aSJan Kara } 9476585027aSJan Kara 94803ba3782SJens Axboe static long wb_check_old_data_flush(struct bdi_writeback *wb) 94903ba3782SJens Axboe { 95003ba3782SJens Axboe unsigned long expired; 95103ba3782SJens Axboe long nr_pages; 95203ba3782SJens Axboe 95369b62d01SJens Axboe /* 95469b62d01SJens Axboe * When set to zero, disable periodic writeback 95569b62d01SJens Axboe */ 95669b62d01SJens Axboe if (!dirty_writeback_interval) 95769b62d01SJens Axboe return 0; 95869b62d01SJens Axboe 95903ba3782SJens Axboe expired = wb->last_old_flush + 96003ba3782SJens Axboe msecs_to_jiffies(dirty_writeback_interval * 10); 96103ba3782SJens Axboe if (time_before(jiffies, expired)) 96203ba3782SJens Axboe return 0; 96303ba3782SJens Axboe 96403ba3782SJens Axboe wb->last_old_flush = jiffies; 965cdf01dd5SLinus Torvalds nr_pages = get_nr_dirty_pages(); 96603ba3782SJens Axboe 967c4a77a6cSJens Axboe if (nr_pages) { 96883ba7b07SChristoph Hellwig struct wb_writeback_work work = { 969c4a77a6cSJens Axboe .nr_pages = nr_pages, 970c4a77a6cSJens Axboe .sync_mode = WB_SYNC_NONE, 971c4a77a6cSJens Axboe .for_kupdate = 1, 972c4a77a6cSJens Axboe .range_cyclic = 1, 9730e175a18SCurt Wohlgemuth .reason = WB_REASON_PERIODIC, 974c4a77a6cSJens Axboe }; 975c4a77a6cSJens Axboe 97683ba7b07SChristoph Hellwig return wb_writeback(wb, &work); 977c4a77a6cSJens Axboe } 97803ba3782SJens Axboe 97903ba3782SJens Axboe return 0; 98003ba3782SJens Axboe } 98103ba3782SJens Axboe 98203ba3782SJens Axboe /* 98303ba3782SJens Axboe * Retrieve work items and do the writeback they describe 98403ba3782SJens Axboe */ 98525d130baSWanpeng Li static long wb_do_writeback(struct bdi_writeback *wb) 98603ba3782SJens Axboe { 98703ba3782SJens Axboe struct backing_dev_info *bdi = wb->bdi; 98883ba7b07SChristoph Hellwig struct wb_writeback_work *work; 989c4a77a6cSJens Axboe long wrote = 0; 99003ba3782SJens Axboe 99181d73a32SJan Kara set_bit(BDI_writeback_running, &wb->bdi->state); 99208852b6dSMinchan Kim while ((work = get_next_work_item(bdi)) != NULL) { 99383ba7b07SChristoph Hellwig 994455b2864SDave Chinner trace_writeback_exec(bdi, work); 995455b2864SDave Chinner 99683ba7b07SChristoph Hellwig wrote += wb_writeback(wb, work); 99703ba3782SJens Axboe 99803ba3782SJens Axboe /* 99983ba7b07SChristoph Hellwig * Notify the caller of completion if this is a synchronous 100083ba7b07SChristoph Hellwig * work item, otherwise just free it. 100103ba3782SJens Axboe */ 100283ba7b07SChristoph Hellwig if (work->done) 100383ba7b07SChristoph Hellwig complete(work->done); 100483ba7b07SChristoph Hellwig else 100583ba7b07SChristoph Hellwig kfree(work); 100603ba3782SJens Axboe } 100703ba3782SJens Axboe 100803ba3782SJens Axboe /* 100903ba3782SJens Axboe * Check for periodic writeback, kupdated() style 101003ba3782SJens Axboe */ 101103ba3782SJens Axboe wrote += wb_check_old_data_flush(wb); 10126585027aSJan Kara wrote += wb_check_background_flush(wb); 101381d73a32SJan Kara clear_bit(BDI_writeback_running, &wb->bdi->state); 101403ba3782SJens Axboe 101503ba3782SJens Axboe return wrote; 101603ba3782SJens Axboe } 101703ba3782SJens Axboe 101803ba3782SJens Axboe /* 101903ba3782SJens Axboe * Handle writeback of dirty data for the device backed by this bdi. Also 1020839a8e86STejun Heo * reschedules periodically and does kupdated style flushing. 102103ba3782SJens Axboe */ 1022839a8e86STejun Heo void bdi_writeback_workfn(struct work_struct *work) 102303ba3782SJens Axboe { 1024839a8e86STejun Heo struct bdi_writeback *wb = container_of(to_delayed_work(work), 1025839a8e86STejun Heo struct bdi_writeback, dwork); 102608243900SChristoph Hellwig struct backing_dev_info *bdi = wb->bdi; 102703ba3782SJens Axboe long pages_written; 102803ba3782SJens Axboe 1029ef3b1019STejun Heo set_worker_desc("flush-%s", dev_name(bdi->dev)); 1030766f9164SPeter Zijlstra current->flags |= PF_SWAPWRITE; 103103ba3782SJens Axboe 1032839a8e86STejun Heo if (likely(!current_is_workqueue_rescuer() || 10335acda9d1SJan Kara !test_bit(BDI_registered, &bdi->state))) { 103403ba3782SJens Axboe /* 1035839a8e86STejun Heo * The normal path. Keep writing back @bdi until its 1036839a8e86STejun Heo * work_list is empty. Note that this path is also taken 1037839a8e86STejun Heo * if @bdi is shutting down even when we're running off the 1038839a8e86STejun Heo * rescuer as work_list needs to be drained. 103903ba3782SJens Axboe */ 1040839a8e86STejun Heo do { 104125d130baSWanpeng Li pages_written = wb_do_writeback(wb); 1042455b2864SDave Chinner trace_writeback_pages_written(pages_written); 1043839a8e86STejun Heo } while (!list_empty(&bdi->work_list)); 1044839a8e86STejun Heo } else { 1045253c34e9SArtem Bityutskiy /* 1046839a8e86STejun Heo * bdi_wq can't get enough workers and we're running off 1047839a8e86STejun Heo * the emergency worker. Don't hog it. Hopefully, 1024 is 1048839a8e86STejun Heo * enough for efficient IO. 1049253c34e9SArtem Bityutskiy */ 1050839a8e86STejun Heo pages_written = writeback_inodes_wb(&bdi->wb, 1024, 1051839a8e86STejun Heo WB_REASON_FORKER_THREAD); 1052839a8e86STejun Heo trace_writeback_pages_written(pages_written); 105303ba3782SJens Axboe } 105403ba3782SJens Axboe 10556ca738d6SDerek Basehore if (!list_empty(&bdi->work_list)) 10566ca738d6SDerek Basehore mod_delayed_work(bdi_wq, &wb->dwork, 0); 10576ca738d6SDerek Basehore else if (wb_has_dirty_io(wb) && dirty_writeback_interval) 10586ca738d6SDerek Basehore bdi_wakeup_thread_delayed(bdi); 1059455b2864SDave Chinner 1060839a8e86STejun Heo current->flags &= ~PF_SWAPWRITE; 106103ba3782SJens Axboe } 106203ba3782SJens Axboe 106303ba3782SJens Axboe /* 106403ba3782SJens Axboe * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back 106503ba3782SJens Axboe * the whole world. 106603ba3782SJens Axboe */ 10670e175a18SCurt Wohlgemuth void wakeup_flusher_threads(long nr_pages, enum wb_reason reason) 106803ba3782SJens Axboe { 1069b8c2f347SChristoph Hellwig struct backing_dev_info *bdi; 1070b8c2f347SChristoph Hellwig 107147df3ddeSJan Kara if (!nr_pages) 107247df3ddeSJan Kara nr_pages = get_nr_dirty_pages(); 1073b8c2f347SChristoph Hellwig 1074b8c2f347SChristoph Hellwig rcu_read_lock(); 1075b8c2f347SChristoph Hellwig list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { 1076b8c2f347SChristoph Hellwig if (!bdi_has_dirty_io(bdi)) 1077b8c2f347SChristoph Hellwig continue; 10780e175a18SCurt Wohlgemuth __bdi_start_writeback(bdi, nr_pages, false, reason); 1079b8c2f347SChristoph Hellwig } 1080b8c2f347SChristoph Hellwig rcu_read_unlock(); 108103ba3782SJens Axboe } 108203ba3782SJens Axboe 108303ba3782SJens Axboe static noinline void block_dump___mark_inode_dirty(struct inode *inode) 108403ba3782SJens Axboe { 108503ba3782SJens Axboe if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { 108603ba3782SJens Axboe struct dentry *dentry; 108703ba3782SJens Axboe const char *name = "?"; 108803ba3782SJens Axboe 108903ba3782SJens Axboe dentry = d_find_alias(inode); 109003ba3782SJens Axboe if (dentry) { 109103ba3782SJens Axboe spin_lock(&dentry->d_lock); 109203ba3782SJens Axboe name = (const char *) dentry->d_name.name; 109303ba3782SJens Axboe } 109403ba3782SJens Axboe printk(KERN_DEBUG 109503ba3782SJens Axboe "%s(%d): dirtied inode %lu (%s) on %s\n", 109603ba3782SJens Axboe current->comm, task_pid_nr(current), inode->i_ino, 109703ba3782SJens Axboe name, inode->i_sb->s_id); 109803ba3782SJens Axboe if (dentry) { 109903ba3782SJens Axboe spin_unlock(&dentry->d_lock); 110003ba3782SJens Axboe dput(dentry); 110103ba3782SJens Axboe } 110203ba3782SJens Axboe } 110303ba3782SJens Axboe } 110403ba3782SJens Axboe 110503ba3782SJens Axboe /** 110603ba3782SJens Axboe * __mark_inode_dirty - internal function 110703ba3782SJens Axboe * @inode: inode to mark 110803ba3782SJens Axboe * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) 110903ba3782SJens Axboe * Mark an inode as dirty. Callers should use mark_inode_dirty or 111003ba3782SJens Axboe * mark_inode_dirty_sync. 111103ba3782SJens Axboe * 111203ba3782SJens Axboe * Put the inode on the super block's dirty list. 111303ba3782SJens Axboe * 111403ba3782SJens Axboe * CAREFUL! We mark it dirty unconditionally, but move it onto the 111503ba3782SJens Axboe * dirty list only if it is hashed or if it refers to a blockdev. 111603ba3782SJens Axboe * If it was not hashed, it will never be added to the dirty list 111703ba3782SJens Axboe * even if it is later hashed, as it will have been marked dirty already. 111803ba3782SJens Axboe * 111903ba3782SJens Axboe * In short, make sure you hash any inodes _before_ you start marking 112003ba3782SJens Axboe * them dirty. 112103ba3782SJens Axboe * 112203ba3782SJens Axboe * Note that for blockdevs, inode->dirtied_when represents the dirtying time of 112303ba3782SJens Axboe * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of 112403ba3782SJens Axboe * the kernel-internal blockdev inode represents the dirtying time of the 112503ba3782SJens Axboe * blockdev's pages. This is why for I_DIRTY_PAGES we always use 112603ba3782SJens Axboe * page->mapping->host, so the page-dirtying time is recorded in the internal 112703ba3782SJens Axboe * blockdev inode. 112803ba3782SJens Axboe */ 112903ba3782SJens Axboe void __mark_inode_dirty(struct inode *inode, int flags) 113003ba3782SJens Axboe { 113103ba3782SJens Axboe struct super_block *sb = inode->i_sb; 1132253c34e9SArtem Bityutskiy struct backing_dev_info *bdi = NULL; 113303ba3782SJens Axboe 113403ba3782SJens Axboe /* 113503ba3782SJens Axboe * Don't do this for I_DIRTY_PAGES - that doesn't actually 113603ba3782SJens Axboe * dirty the inode itself 113703ba3782SJens Axboe */ 113803ba3782SJens Axboe if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { 11399fb0a7daSTejun Heo trace_writeback_dirty_inode_start(inode, flags); 11409fb0a7daSTejun Heo 114103ba3782SJens Axboe if (sb->s_op->dirty_inode) 1142aa385729SChristoph Hellwig sb->s_op->dirty_inode(inode, flags); 11439fb0a7daSTejun Heo 11449fb0a7daSTejun Heo trace_writeback_dirty_inode(inode, flags); 114503ba3782SJens Axboe } 114603ba3782SJens Axboe 114703ba3782SJens Axboe /* 114803ba3782SJens Axboe * make sure that changes are seen by all cpus before we test i_state 114903ba3782SJens Axboe * -- mikulas 115003ba3782SJens Axboe */ 115103ba3782SJens Axboe smp_mb(); 115203ba3782SJens Axboe 115303ba3782SJens Axboe /* avoid the locking if we can */ 115403ba3782SJens Axboe if ((inode->i_state & flags) == flags) 115503ba3782SJens Axboe return; 115603ba3782SJens Axboe 115703ba3782SJens Axboe if (unlikely(block_dump)) 115803ba3782SJens Axboe block_dump___mark_inode_dirty(inode); 115903ba3782SJens Axboe 1160250df6edSDave Chinner spin_lock(&inode->i_lock); 116103ba3782SJens Axboe if ((inode->i_state & flags) != flags) { 116203ba3782SJens Axboe const int was_dirty = inode->i_state & I_DIRTY; 116303ba3782SJens Axboe 116403ba3782SJens Axboe inode->i_state |= flags; 116503ba3782SJens Axboe 116603ba3782SJens Axboe /* 116703ba3782SJens Axboe * If the inode is being synced, just update its dirty state. 116803ba3782SJens Axboe * The unlocker will place the inode on the appropriate 116903ba3782SJens Axboe * superblock list, based upon its state. 117003ba3782SJens Axboe */ 117103ba3782SJens Axboe if (inode->i_state & I_SYNC) 1172250df6edSDave Chinner goto out_unlock_inode; 117303ba3782SJens Axboe 117403ba3782SJens Axboe /* 117503ba3782SJens Axboe * Only add valid (hashed) inodes to the superblock's 117603ba3782SJens Axboe * dirty list. Add blockdev inodes as well. 117703ba3782SJens Axboe */ 117803ba3782SJens Axboe if (!S_ISBLK(inode->i_mode)) { 11791d3382cbSAl Viro if (inode_unhashed(inode)) 1180250df6edSDave Chinner goto out_unlock_inode; 118103ba3782SJens Axboe } 1182a4ffdde6SAl Viro if (inode->i_state & I_FREEING) 1183250df6edSDave Chinner goto out_unlock_inode; 118403ba3782SJens Axboe 118503ba3782SJens Axboe /* 118603ba3782SJens Axboe * If the inode was already on b_dirty/b_io/b_more_io, don't 118703ba3782SJens Axboe * reposition it (that would break b_dirty time-ordering). 118803ba3782SJens Axboe */ 118903ba3782SJens Axboe if (!was_dirty) { 1190a66979abSDave Chinner bool wakeup_bdi = false; 1191253c34e9SArtem Bityutskiy bdi = inode_to_bdi(inode); 1192500b067cSJens Axboe 1193146d7009SJunxiao Bi spin_unlock(&inode->i_lock); 1194146d7009SJunxiao Bi spin_lock(&bdi->wb.list_lock); 1195253c34e9SArtem Bityutskiy if (bdi_cap_writeback_dirty(bdi)) { 1196253c34e9SArtem Bityutskiy WARN(!test_bit(BDI_registered, &bdi->state), 1197253c34e9SArtem Bityutskiy "bdi-%s not registered\n", bdi->name); 1198253c34e9SArtem Bityutskiy 1199253c34e9SArtem Bityutskiy /* 1200253c34e9SArtem Bityutskiy * If this is the first dirty inode for this 1201253c34e9SArtem Bityutskiy * bdi, we have to wake-up the corresponding 1202253c34e9SArtem Bityutskiy * bdi thread to make sure background 1203253c34e9SArtem Bityutskiy * write-back happens later. 1204253c34e9SArtem Bityutskiy */ 1205253c34e9SArtem Bityutskiy if (!wb_has_dirty_io(&bdi->wb)) 1206253c34e9SArtem Bityutskiy wakeup_bdi = true; 1207500b067cSJens Axboe } 120803ba3782SJens Axboe 120903ba3782SJens Axboe inode->dirtied_when = jiffies; 12107ccf19a8SNick Piggin list_move(&inode->i_wb_list, &bdi->wb.b_dirty); 1211f758eeabSChristoph Hellwig spin_unlock(&bdi->wb.list_lock); 1212253c34e9SArtem Bityutskiy 1213253c34e9SArtem Bityutskiy if (wakeup_bdi) 12146467716aSArtem Bityutskiy bdi_wakeup_thread_delayed(bdi); 1215a66979abSDave Chinner return; 1216a66979abSDave Chinner } 1217a66979abSDave Chinner } 1218a66979abSDave Chinner out_unlock_inode: 1219a66979abSDave Chinner spin_unlock(&inode->i_lock); 1220a66979abSDave Chinner 122103ba3782SJens Axboe } 122203ba3782SJens Axboe EXPORT_SYMBOL(__mark_inode_dirty); 122303ba3782SJens Axboe 1224b6e51316SJens Axboe static void wait_sb_inodes(struct super_block *sb) 122566f3b8e2SJens Axboe { 122638f21977SNick Piggin struct inode *inode, *old_inode = NULL; 122738f21977SNick Piggin 122803ba3782SJens Axboe /* 122903ba3782SJens Axboe * We need to be protected against the filesystem going from 123003ba3782SJens Axboe * r/o to r/w or vice versa. 123103ba3782SJens Axboe */ 1232b6e51316SJens Axboe WARN_ON(!rwsem_is_locked(&sb->s_umount)); 123303ba3782SJens Axboe 123455fa6091SDave Chinner spin_lock(&inode_sb_list_lock); 123566f3b8e2SJens Axboe 123638f21977SNick Piggin /* 123738f21977SNick Piggin * Data integrity sync. Must wait for all pages under writeback, 123838f21977SNick Piggin * because there may have been pages dirtied before our sync 123938f21977SNick Piggin * call, but which had writeout started before we write it out. 124038f21977SNick Piggin * In which case, the inode may not be on the dirty list, but 124138f21977SNick Piggin * we still have to wait for that writeout. 124238f21977SNick Piggin */ 1243b6e51316SJens Axboe list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { 1244250df6edSDave Chinner struct address_space *mapping = inode->i_mapping; 124538f21977SNick Piggin 1246250df6edSDave Chinner spin_lock(&inode->i_lock); 1247250df6edSDave Chinner if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) || 1248250df6edSDave Chinner (mapping->nrpages == 0)) { 1249250df6edSDave Chinner spin_unlock(&inode->i_lock); 125038f21977SNick Piggin continue; 1251250df6edSDave Chinner } 125238f21977SNick Piggin __iget(inode); 1253250df6edSDave Chinner spin_unlock(&inode->i_lock); 125455fa6091SDave Chinner spin_unlock(&inode_sb_list_lock); 125555fa6091SDave Chinner 125638f21977SNick Piggin /* 125755fa6091SDave Chinner * We hold a reference to 'inode' so it couldn't have been 125855fa6091SDave Chinner * removed from s_inodes list while we dropped the 125955fa6091SDave Chinner * inode_sb_list_lock. We cannot iput the inode now as we can 126055fa6091SDave Chinner * be holding the last reference and we cannot iput it under 126155fa6091SDave Chinner * inode_sb_list_lock. So we keep the reference and iput it 126255fa6091SDave Chinner * later. 126338f21977SNick Piggin */ 126438f21977SNick Piggin iput(old_inode); 126538f21977SNick Piggin old_inode = inode; 126638f21977SNick Piggin 126738f21977SNick Piggin filemap_fdatawait(mapping); 126838f21977SNick Piggin 126938f21977SNick Piggin cond_resched(); 127038f21977SNick Piggin 127155fa6091SDave Chinner spin_lock(&inode_sb_list_lock); 127238f21977SNick Piggin } 127355fa6091SDave Chinner spin_unlock(&inode_sb_list_lock); 127438f21977SNick Piggin iput(old_inode); 127566f3b8e2SJens Axboe } 12761da177e4SLinus Torvalds 1277d8a8559cSJens Axboe /** 12783259f8beSChris Mason * writeback_inodes_sb_nr - writeback dirty inodes from given super_block 1279d8a8559cSJens Axboe * @sb: the superblock 12803259f8beSChris Mason * @nr: the number of pages to write 1281786228abSMarcos Paulo de Souza * @reason: reason why some writeback work initiated 12821da177e4SLinus Torvalds * 1283d8a8559cSJens Axboe * Start writeback on some inodes on this super_block. No guarantees are made 1284d8a8559cSJens Axboe * on how many (if any) will be written, and this function does not wait 12853259f8beSChris Mason * for IO completion of submitted IO. 12861da177e4SLinus Torvalds */ 12870e175a18SCurt Wohlgemuth void writeback_inodes_sb_nr(struct super_block *sb, 12880e175a18SCurt Wohlgemuth unsigned long nr, 12890e175a18SCurt Wohlgemuth enum wb_reason reason) 12901da177e4SLinus Torvalds { 129183ba7b07SChristoph Hellwig DECLARE_COMPLETION_ONSTACK(done); 129283ba7b07SChristoph Hellwig struct wb_writeback_work work = { 12933c4d7165SChristoph Hellwig .sb = sb, 12943c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_NONE, 12956e6938b6SWu Fengguang .tagged_writepages = 1, 129683ba7b07SChristoph Hellwig .done = &done, 12973259f8beSChris Mason .nr_pages = nr, 12980e175a18SCurt Wohlgemuth .reason = reason, 12993c4d7165SChristoph Hellwig }; 13000e3c9a22SJens Axboe 13016eedc701SJan Kara if (sb->s_bdi == &noop_backing_dev_info) 13026eedc701SJan Kara return; 1303cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 130483ba7b07SChristoph Hellwig bdi_queue_work(sb->s_bdi, &work); 130583ba7b07SChristoph Hellwig wait_for_completion(&done); 13061da177e4SLinus Torvalds } 13073259f8beSChris Mason EXPORT_SYMBOL(writeback_inodes_sb_nr); 13083259f8beSChris Mason 13093259f8beSChris Mason /** 13103259f8beSChris Mason * writeback_inodes_sb - writeback dirty inodes from given super_block 13113259f8beSChris Mason * @sb: the superblock 1312786228abSMarcos Paulo de Souza * @reason: reason why some writeback work was initiated 13133259f8beSChris Mason * 13143259f8beSChris Mason * Start writeback on some inodes on this super_block. No guarantees are made 13153259f8beSChris Mason * on how many (if any) will be written, and this function does not wait 13163259f8beSChris Mason * for IO completion of submitted IO. 13173259f8beSChris Mason */ 13180e175a18SCurt Wohlgemuth void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) 13193259f8beSChris Mason { 13200e175a18SCurt Wohlgemuth return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); 13213259f8beSChris Mason } 1322d8a8559cSJens Axboe EXPORT_SYMBOL(writeback_inodes_sb); 1323d8a8559cSJens Axboe 1324d8a8559cSJens Axboe /** 132510ee27a0SMiao Xie * try_to_writeback_inodes_sb_nr - try to start writeback if none underway 13263259f8beSChris Mason * @sb: the superblock 13273259f8beSChris Mason * @nr: the number of pages to write 132810ee27a0SMiao Xie * @reason: the reason of writeback 13293259f8beSChris Mason * 133010ee27a0SMiao Xie * Invoke writeback_inodes_sb_nr if no writeback is currently underway. 13313259f8beSChris Mason * Returns 1 if writeback was started, 0 if not. 13323259f8beSChris Mason */ 133310ee27a0SMiao Xie int try_to_writeback_inodes_sb_nr(struct super_block *sb, 13340e175a18SCurt Wohlgemuth unsigned long nr, 13350e175a18SCurt Wohlgemuth enum wb_reason reason) 13363259f8beSChris Mason { 133710ee27a0SMiao Xie if (writeback_in_progress(sb->s_bdi)) 133810ee27a0SMiao Xie return 1; 133910ee27a0SMiao Xie 134010ee27a0SMiao Xie if (!down_read_trylock(&sb->s_umount)) 134110ee27a0SMiao Xie return 0; 134210ee27a0SMiao Xie 13430e175a18SCurt Wohlgemuth writeback_inodes_sb_nr(sb, nr, reason); 13443259f8beSChris Mason up_read(&sb->s_umount); 13453259f8beSChris Mason return 1; 13463259f8beSChris Mason } 134710ee27a0SMiao Xie EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr); 134810ee27a0SMiao Xie 134910ee27a0SMiao Xie /** 135010ee27a0SMiao Xie * try_to_writeback_inodes_sb - try to start writeback if none underway 135110ee27a0SMiao Xie * @sb: the superblock 135210ee27a0SMiao Xie * @reason: reason why some writeback work was initiated 135310ee27a0SMiao Xie * 135410ee27a0SMiao Xie * Implement by try_to_writeback_inodes_sb_nr() 135510ee27a0SMiao Xie * Returns 1 if writeback was started, 0 if not. 135610ee27a0SMiao Xie */ 135710ee27a0SMiao Xie int try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) 135810ee27a0SMiao Xie { 135910ee27a0SMiao Xie return try_to_writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); 136010ee27a0SMiao Xie } 136110ee27a0SMiao Xie EXPORT_SYMBOL(try_to_writeback_inodes_sb); 13623259f8beSChris Mason 13633259f8beSChris Mason /** 1364d8a8559cSJens Axboe * sync_inodes_sb - sync sb inode pages 1365d8a8559cSJens Axboe * @sb: the superblock 1366d8a8559cSJens Axboe * 1367d8a8559cSJens Axboe * This function writes and waits on any dirty inode belonging to this 13680dc83bd3SJan Kara * super_block. 1369d8a8559cSJens Axboe */ 13700dc83bd3SJan Kara void sync_inodes_sb(struct super_block *sb) 1371d8a8559cSJens Axboe { 137283ba7b07SChristoph Hellwig DECLARE_COMPLETION_ONSTACK(done); 137383ba7b07SChristoph Hellwig struct wb_writeback_work work = { 13743c4d7165SChristoph Hellwig .sb = sb, 13753c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_ALL, 13763c4d7165SChristoph Hellwig .nr_pages = LONG_MAX, 13773c4d7165SChristoph Hellwig .range_cyclic = 0, 137883ba7b07SChristoph Hellwig .done = &done, 13790e175a18SCurt Wohlgemuth .reason = WB_REASON_SYNC, 13807747bd4bSDave Chinner .for_sync = 1, 13813c4d7165SChristoph Hellwig }; 13823c4d7165SChristoph Hellwig 13836eedc701SJan Kara /* Nothing to do? */ 13846eedc701SJan Kara if (sb->s_bdi == &noop_backing_dev_info) 13856eedc701SJan Kara return; 1386cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 1387cf37e972SChristoph Hellwig 138883ba7b07SChristoph Hellwig bdi_queue_work(sb->s_bdi, &work); 138983ba7b07SChristoph Hellwig wait_for_completion(&done); 139083ba7b07SChristoph Hellwig 1391b6e51316SJens Axboe wait_sb_inodes(sb); 1392d8a8559cSJens Axboe } 1393d8a8559cSJens Axboe EXPORT_SYMBOL(sync_inodes_sb); 13941da177e4SLinus Torvalds 13951da177e4SLinus Torvalds /** 13961da177e4SLinus Torvalds * write_inode_now - write an inode to disk 13971da177e4SLinus Torvalds * @inode: inode to write to disk 13981da177e4SLinus Torvalds * @sync: whether the write should be synchronous or not 13991da177e4SLinus Torvalds * 14007f04c26dSAndrea Arcangeli * This function commits an inode to disk immediately if it is dirty. This is 14017f04c26dSAndrea Arcangeli * primarily needed by knfsd. 14027f04c26dSAndrea Arcangeli * 14037f04c26dSAndrea Arcangeli * The caller must either have a ref on the inode or must have set I_WILL_FREE. 14041da177e4SLinus Torvalds */ 14051da177e4SLinus Torvalds int write_inode_now(struct inode *inode, int sync) 14061da177e4SLinus Torvalds { 1407f758eeabSChristoph Hellwig struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; 14081da177e4SLinus Torvalds struct writeback_control wbc = { 14091da177e4SLinus Torvalds .nr_to_write = LONG_MAX, 141018914b18SMike Galbraith .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, 1411111ebb6eSOGAWA Hirofumi .range_start = 0, 1412111ebb6eSOGAWA Hirofumi .range_end = LLONG_MAX, 14131da177e4SLinus Torvalds }; 14141da177e4SLinus Torvalds 14151da177e4SLinus Torvalds if (!mapping_cap_writeback_dirty(inode->i_mapping)) 141649364ce2SAndrew Morton wbc.nr_to_write = 0; 14171da177e4SLinus Torvalds 14181da177e4SLinus Torvalds might_sleep(); 14194f8ad655SJan Kara return writeback_single_inode(inode, wb, &wbc); 14201da177e4SLinus Torvalds } 14211da177e4SLinus Torvalds EXPORT_SYMBOL(write_inode_now); 14221da177e4SLinus Torvalds 14231da177e4SLinus Torvalds /** 14241da177e4SLinus Torvalds * sync_inode - write an inode and its pages to disk. 14251da177e4SLinus Torvalds * @inode: the inode to sync 14261da177e4SLinus Torvalds * @wbc: controls the writeback mode 14271da177e4SLinus Torvalds * 14281da177e4SLinus Torvalds * sync_inode() will write an inode and its pages to disk. It will also 14291da177e4SLinus Torvalds * correctly update the inode on its superblock's dirty inode lists and will 14301da177e4SLinus Torvalds * update inode->i_state. 14311da177e4SLinus Torvalds * 14321da177e4SLinus Torvalds * The caller must have a ref on the inode. 14331da177e4SLinus Torvalds */ 14341da177e4SLinus Torvalds int sync_inode(struct inode *inode, struct writeback_control *wbc) 14351da177e4SLinus Torvalds { 14364f8ad655SJan Kara return writeback_single_inode(inode, &inode_to_bdi(inode)->wb, wbc); 14371da177e4SLinus Torvalds } 14381da177e4SLinus Torvalds EXPORT_SYMBOL(sync_inode); 1439c3765016SChristoph Hellwig 1440c3765016SChristoph Hellwig /** 1441c691b9d9SAndrew Morton * sync_inode_metadata - write an inode to disk 1442c3765016SChristoph Hellwig * @inode: the inode to sync 1443c3765016SChristoph Hellwig * @wait: wait for I/O to complete. 1444c3765016SChristoph Hellwig * 1445c691b9d9SAndrew Morton * Write an inode to disk and adjust its dirty state after completion. 1446c3765016SChristoph Hellwig * 1447c3765016SChristoph Hellwig * Note: only writes the actual inode, no associated data or other metadata. 1448c3765016SChristoph Hellwig */ 1449c3765016SChristoph Hellwig int sync_inode_metadata(struct inode *inode, int wait) 1450c3765016SChristoph Hellwig { 1451c3765016SChristoph Hellwig struct writeback_control wbc = { 1452c3765016SChristoph Hellwig .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE, 1453c3765016SChristoph Hellwig .nr_to_write = 0, /* metadata-only */ 1454c3765016SChristoph Hellwig }; 1455c3765016SChristoph Hellwig 1456c3765016SChristoph Hellwig return sync_inode(inode, &wbc); 1457c3765016SChristoph Hellwig } 1458c3765016SChristoph Hellwig EXPORT_SYMBOL(sync_inode_metadata); 1459