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 56a2f48706STheodore Ts'o /* 57a2f48706STheodore Ts'o * If an inode is constantly having its pages dirtied, but then the 58a2f48706STheodore Ts'o * updates stop dirtytime_expire_interval seconds in the past, it's 59a2f48706STheodore Ts'o * possible for the worst case time between when an inode has its 60a2f48706STheodore Ts'o * timestamps updated and when they finally get written out to be two 61a2f48706STheodore Ts'o * dirtytime_expire_intervals. We set the default to 12 hours (in 62a2f48706STheodore Ts'o * seconds), which means most of the time inodes will have their 63a2f48706STheodore Ts'o * timestamps written to disk after 12 hours, but in the worst case a 64a2f48706STheodore Ts'o * few inodes might not their timestamps updated for 24 hours. 65a2f48706STheodore Ts'o */ 66a2f48706STheodore Ts'o unsigned int dirtytime_expire_interval = 12 * 60 * 60; 67a2f48706STheodore Ts'o 68f11b00f3SAdrian Bunk /** 69f11b00f3SAdrian Bunk * writeback_in_progress - determine whether there is writeback in progress 70f11b00f3SAdrian Bunk * @bdi: the device's backing_dev_info structure. 71f11b00f3SAdrian Bunk * 7203ba3782SJens Axboe * Determine whether there is writeback waiting to be handled against a 7303ba3782SJens Axboe * backing device. 74f11b00f3SAdrian Bunk */ 75f11b00f3SAdrian Bunk int writeback_in_progress(struct backing_dev_info *bdi) 76f11b00f3SAdrian Bunk { 774452226eSTejun Heo return test_bit(WB_writeback_running, &bdi->wb.state); 78f11b00f3SAdrian Bunk } 7900d4e736STheodore Ts'o EXPORT_SYMBOL(writeback_in_progress); 80f11b00f3SAdrian Bunk 81de1414a6SChristoph Hellwig struct backing_dev_info *inode_to_bdi(struct inode *inode) 82692ebd17SJan Kara { 83b520252aSJens Axboe struct super_block *sb; 84b520252aSJens Axboe 85b520252aSJens Axboe if (!inode) 86b520252aSJens Axboe return &noop_backing_dev_info; 87b520252aSJens Axboe 88b520252aSJens Axboe sb = inode->i_sb; 89495a276eSChristoph Hellwig #ifdef CONFIG_BLOCK 90a8855990SJan Kara if (sb_is_blkdev_sb(sb)) 91495a276eSChristoph Hellwig return blk_get_backing_dev_info(I_BDEV(inode)); 92495a276eSChristoph Hellwig #endif 93692ebd17SJan Kara return sb->s_bdi; 94692ebd17SJan Kara } 95de1414a6SChristoph Hellwig EXPORT_SYMBOL_GPL(inode_to_bdi); 96692ebd17SJan Kara 977ccf19a8SNick Piggin static inline struct inode *wb_inode(struct list_head *head) 987ccf19a8SNick Piggin { 997ccf19a8SNick Piggin return list_entry(head, struct inode, i_wb_list); 1007ccf19a8SNick Piggin } 1017ccf19a8SNick Piggin 10215eb77a0SWu Fengguang /* 10315eb77a0SWu Fengguang * Include the creation of the trace points after defining the 10415eb77a0SWu Fengguang * wb_writeback_work structure and inline functions so that the definition 10515eb77a0SWu Fengguang * remains local to this file. 10615eb77a0SWu Fengguang */ 10715eb77a0SWu Fengguang #define CREATE_TRACE_POINTS 10815eb77a0SWu Fengguang #include <trace/events/writeback.h> 10915eb77a0SWu Fengguang 110774016b2SSteven Whitehouse EXPORT_TRACEPOINT_SYMBOL_GPL(wbc_writepage); 111774016b2SSteven Whitehouse 112f0054bb1STejun Heo static void wb_wakeup(struct bdi_writeback *wb) 1135acda9d1SJan Kara { 114f0054bb1STejun Heo spin_lock_bh(&wb->work_lock); 115f0054bb1STejun Heo if (test_bit(WB_registered, &wb->state)) 116f0054bb1STejun Heo mod_delayed_work(bdi_wq, &wb->dwork, 0); 117f0054bb1STejun Heo spin_unlock_bh(&wb->work_lock); 1185acda9d1SJan Kara } 1195acda9d1SJan Kara 120f0054bb1STejun Heo static void wb_queue_work(struct bdi_writeback *wb, 1216585027aSJan Kara struct wb_writeback_work *work) 1226585027aSJan Kara { 123f0054bb1STejun Heo trace_writeback_queue(wb->bdi, work); 1246585027aSJan Kara 125f0054bb1STejun Heo spin_lock_bh(&wb->work_lock); 126f0054bb1STejun Heo if (!test_bit(WB_registered, &wb->state)) { 1275acda9d1SJan Kara if (work->done) 1285acda9d1SJan Kara complete(work->done); 1295acda9d1SJan Kara goto out_unlock; 1305acda9d1SJan Kara } 131f0054bb1STejun Heo list_add_tail(&work->list, &wb->work_list); 132f0054bb1STejun Heo mod_delayed_work(bdi_wq, &wb->dwork, 0); 1335acda9d1SJan Kara out_unlock: 134f0054bb1STejun Heo spin_unlock_bh(&wb->work_lock); 13503ba3782SJens Axboe } 1361da177e4SLinus Torvalds 137f0054bb1STejun Heo static void __wb_start_writeback(struct bdi_writeback *wb, long nr_pages, 1380e175a18SCurt Wohlgemuth bool range_cyclic, enum wb_reason reason) 1391da177e4SLinus Torvalds { 14083ba7b07SChristoph Hellwig struct wb_writeback_work *work; 14103ba3782SJens Axboe 142bcddc3f0SJens Axboe /* 143bcddc3f0SJens Axboe * This is WB_SYNC_NONE writeback, so if allocation fails just 144bcddc3f0SJens Axboe * wakeup the thread for old dirty data writeback 145bcddc3f0SJens Axboe */ 14683ba7b07SChristoph Hellwig work = kzalloc(sizeof(*work), GFP_ATOMIC); 14783ba7b07SChristoph Hellwig if (!work) { 148f0054bb1STejun Heo trace_writeback_nowork(wb->bdi); 149f0054bb1STejun Heo wb_wakeup(wb); 15083ba7b07SChristoph Hellwig return; 15183ba7b07SChristoph Hellwig } 15283ba7b07SChristoph Hellwig 15383ba7b07SChristoph Hellwig work->sync_mode = WB_SYNC_NONE; 15483ba7b07SChristoph Hellwig work->nr_pages = nr_pages; 15583ba7b07SChristoph Hellwig work->range_cyclic = range_cyclic; 1560e175a18SCurt Wohlgemuth work->reason = reason; 15783ba7b07SChristoph Hellwig 158f0054bb1STejun Heo wb_queue_work(wb, work); 15903ba3782SJens Axboe } 160b6e51316SJens Axboe 161b6e51316SJens Axboe /** 162b6e51316SJens Axboe * bdi_start_writeback - start writeback 163b6e51316SJens Axboe * @bdi: the backing device to write from 164b6e51316SJens Axboe * @nr_pages: the number of pages to write 165786228abSMarcos Paulo de Souza * @reason: reason why some writeback work was initiated 166b6e51316SJens Axboe * 167b6e51316SJens Axboe * Description: 168b6e51316SJens Axboe * This does WB_SYNC_NONE opportunistic writeback. The IO is only 16925985edcSLucas De Marchi * started when this function returns, we make no guarantees on 1700e3c9a22SJens Axboe * completion. Caller need not hold sb s_umount semaphore. 171b6e51316SJens Axboe * 172b6e51316SJens Axboe */ 1730e175a18SCurt Wohlgemuth void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, 1740e175a18SCurt Wohlgemuth enum wb_reason reason) 175b6e51316SJens Axboe { 176f0054bb1STejun Heo __wb_start_writeback(&bdi->wb, nr_pages, true, reason); 177d3ddec76SWu Fengguang } 178d3ddec76SWu Fengguang 179c5444198SChristoph Hellwig /** 180c5444198SChristoph Hellwig * bdi_start_background_writeback - start background writeback 181c5444198SChristoph Hellwig * @bdi: the backing device to write from 182c5444198SChristoph Hellwig * 183c5444198SChristoph Hellwig * Description: 1846585027aSJan Kara * This makes sure WB_SYNC_NONE background writeback happens. When 1856585027aSJan Kara * this function returns, it is only guaranteed that for given BDI 1866585027aSJan Kara * some IO is happening if we are over background dirty threshold. 1876585027aSJan Kara * Caller need not hold sb s_umount semaphore. 188c5444198SChristoph Hellwig */ 189c5444198SChristoph Hellwig void bdi_start_background_writeback(struct backing_dev_info *bdi) 190c5444198SChristoph Hellwig { 1916585027aSJan Kara /* 1926585027aSJan Kara * We just wake up the flusher thread. It will perform background 1936585027aSJan Kara * writeback as soon as there is no other work to do. 1946585027aSJan Kara */ 19571927e84SWu Fengguang trace_writeback_wake_background(bdi); 196f0054bb1STejun Heo wb_wakeup(&bdi->wb); 1971da177e4SLinus Torvalds } 1981da177e4SLinus Torvalds 1991da177e4SLinus Torvalds /* 200a66979abSDave Chinner * Remove the inode from the writeback list it is on. 201a66979abSDave Chinner */ 202a66979abSDave Chinner void inode_wb_list_del(struct inode *inode) 203a66979abSDave Chinner { 204f758eeabSChristoph Hellwig struct backing_dev_info *bdi = inode_to_bdi(inode); 205a66979abSDave Chinner 206f758eeabSChristoph Hellwig spin_lock(&bdi->wb.list_lock); 207f758eeabSChristoph Hellwig list_del_init(&inode->i_wb_list); 208f758eeabSChristoph Hellwig spin_unlock(&bdi->wb.list_lock); 209f758eeabSChristoph Hellwig } 210a66979abSDave Chinner 211a66979abSDave Chinner /* 2126610a0bcSAndrew Morton * Redirty an inode: set its when-it-was dirtied timestamp and move it to the 2136610a0bcSAndrew Morton * furthest end of its superblock's dirty-inode list. 2146610a0bcSAndrew Morton * 2156610a0bcSAndrew Morton * Before stamping the inode's ->dirtied_when, we check to see whether it is 21666f3b8e2SJens Axboe * already the most-recently-dirtied inode on the b_dirty list. If that is 2176610a0bcSAndrew Morton * the case then the inode must have been redirtied while it was being written 2186610a0bcSAndrew Morton * out and we don't reset its dirtied_when. 2196610a0bcSAndrew Morton */ 220f758eeabSChristoph Hellwig static void redirty_tail(struct inode *inode, struct bdi_writeback *wb) 2216610a0bcSAndrew Morton { 222f758eeabSChristoph Hellwig assert_spin_locked(&wb->list_lock); 22303ba3782SJens Axboe if (!list_empty(&wb->b_dirty)) { 22466f3b8e2SJens Axboe struct inode *tail; 2256610a0bcSAndrew Morton 2267ccf19a8SNick Piggin tail = wb_inode(wb->b_dirty.next); 22766f3b8e2SJens Axboe if (time_before(inode->dirtied_when, tail->dirtied_when)) 2286610a0bcSAndrew Morton inode->dirtied_when = jiffies; 2296610a0bcSAndrew Morton } 2307ccf19a8SNick Piggin list_move(&inode->i_wb_list, &wb->b_dirty); 2316610a0bcSAndrew Morton } 2326610a0bcSAndrew Morton 2336610a0bcSAndrew Morton /* 23466f3b8e2SJens Axboe * requeue inode for re-scanning after bdi->b_io list is exhausted. 235c986d1e2SAndrew Morton */ 236f758eeabSChristoph Hellwig static void requeue_io(struct inode *inode, struct bdi_writeback *wb) 237c986d1e2SAndrew Morton { 238f758eeabSChristoph Hellwig assert_spin_locked(&wb->list_lock); 2397ccf19a8SNick Piggin list_move(&inode->i_wb_list, &wb->b_more_io); 240c986d1e2SAndrew Morton } 241c986d1e2SAndrew Morton 2421c0eeaf5SJoern Engel static void inode_sync_complete(struct inode *inode) 2431c0eeaf5SJoern Engel { 244365b94aeSJan Kara inode->i_state &= ~I_SYNC; 2454eff96ddSJan Kara /* If inode is clean an unused, put it into LRU now... */ 2464eff96ddSJan Kara inode_add_lru(inode); 247365b94aeSJan Kara /* Waiters must see I_SYNC cleared before being woken up */ 2481c0eeaf5SJoern Engel smp_mb(); 2491c0eeaf5SJoern Engel wake_up_bit(&inode->i_state, __I_SYNC); 2501c0eeaf5SJoern Engel } 2511c0eeaf5SJoern Engel 252d2caa3c5SJeff Layton static bool inode_dirtied_after(struct inode *inode, unsigned long t) 253d2caa3c5SJeff Layton { 254d2caa3c5SJeff Layton bool ret = time_after(inode->dirtied_when, t); 255d2caa3c5SJeff Layton #ifndef CONFIG_64BIT 256d2caa3c5SJeff Layton /* 257d2caa3c5SJeff Layton * For inodes being constantly redirtied, dirtied_when can get stuck. 258d2caa3c5SJeff Layton * It _appears_ to be in the future, but is actually in distant past. 259d2caa3c5SJeff Layton * This test is necessary to prevent such wrapped-around relative times 2605b0830cbSJens Axboe * from permanently stopping the whole bdi writeback. 261d2caa3c5SJeff Layton */ 262d2caa3c5SJeff Layton ret = ret && time_before_eq(inode->dirtied_when, jiffies); 263d2caa3c5SJeff Layton #endif 264d2caa3c5SJeff Layton return ret; 265d2caa3c5SJeff Layton } 266d2caa3c5SJeff Layton 2670ae45f63STheodore Ts'o #define EXPIRE_DIRTY_ATIME 0x0001 2680ae45f63STheodore Ts'o 269c986d1e2SAndrew Morton /* 2700e2f2b23SWang Sheng-Hui * Move expired (dirtied before work->older_than_this) dirty inodes from 271697e6fedSJan Kara * @delaying_queue to @dispatch_queue. 2722c136579SFengguang Wu */ 273e84d0a4fSWu Fengguang static int move_expired_inodes(struct list_head *delaying_queue, 2742c136579SFengguang Wu struct list_head *dispatch_queue, 2750ae45f63STheodore Ts'o int flags, 276ad4e38ddSCurt Wohlgemuth struct wb_writeback_work *work) 2772c136579SFengguang Wu { 2780ae45f63STheodore Ts'o unsigned long *older_than_this = NULL; 2790ae45f63STheodore Ts'o unsigned long expire_time; 2805c03449dSShaohua Li LIST_HEAD(tmp); 2815c03449dSShaohua Li struct list_head *pos, *node; 282cf137307SJens Axboe struct super_block *sb = NULL; 2835c03449dSShaohua Li struct inode *inode; 284cf137307SJens Axboe int do_sb_sort = 0; 285e84d0a4fSWu Fengguang int moved = 0; 2865c03449dSShaohua Li 2870ae45f63STheodore Ts'o if ((flags & EXPIRE_DIRTY_ATIME) == 0) 2880ae45f63STheodore Ts'o older_than_this = work->older_than_this; 289a2f48706STheodore Ts'o else if (!work->for_sync) { 290a2f48706STheodore Ts'o expire_time = jiffies - (dirtytime_expire_interval * HZ); 2910ae45f63STheodore Ts'o older_than_this = &expire_time; 2920ae45f63STheodore Ts'o } 2932c136579SFengguang Wu while (!list_empty(delaying_queue)) { 2947ccf19a8SNick Piggin inode = wb_inode(delaying_queue->prev); 2950ae45f63STheodore Ts'o if (older_than_this && 2960ae45f63STheodore Ts'o inode_dirtied_after(inode, *older_than_this)) 2972c136579SFengguang Wu break; 298a8855990SJan Kara list_move(&inode->i_wb_list, &tmp); 299a8855990SJan Kara moved++; 3000ae45f63STheodore Ts'o if (flags & EXPIRE_DIRTY_ATIME) 3010ae45f63STheodore Ts'o set_bit(__I_DIRTY_TIME_EXPIRED, &inode->i_state); 302a8855990SJan Kara if (sb_is_blkdev_sb(inode->i_sb)) 303a8855990SJan Kara continue; 304cf137307SJens Axboe if (sb && sb != inode->i_sb) 305cf137307SJens Axboe do_sb_sort = 1; 306cf137307SJens Axboe sb = inode->i_sb; 3075c03449dSShaohua Li } 3085c03449dSShaohua Li 309cf137307SJens Axboe /* just one sb in list, splice to dispatch_queue and we're done */ 310cf137307SJens Axboe if (!do_sb_sort) { 311cf137307SJens Axboe list_splice(&tmp, dispatch_queue); 312e84d0a4fSWu Fengguang goto out; 313cf137307SJens Axboe } 314cf137307SJens Axboe 3155c03449dSShaohua Li /* Move inodes from one superblock together */ 3165c03449dSShaohua Li while (!list_empty(&tmp)) { 3177ccf19a8SNick Piggin sb = wb_inode(tmp.prev)->i_sb; 3185c03449dSShaohua Li list_for_each_prev_safe(pos, node, &tmp) { 3197ccf19a8SNick Piggin inode = wb_inode(pos); 3205c03449dSShaohua Li if (inode->i_sb == sb) 3217ccf19a8SNick Piggin list_move(&inode->i_wb_list, dispatch_queue); 3222c136579SFengguang Wu } 3232c136579SFengguang Wu } 324e84d0a4fSWu Fengguang out: 325e84d0a4fSWu Fengguang return moved; 3265c03449dSShaohua Li } 3272c136579SFengguang Wu 3282c136579SFengguang Wu /* 3292c136579SFengguang Wu * Queue all expired dirty inodes for io, eldest first. 3304ea879b9SWu Fengguang * Before 3314ea879b9SWu Fengguang * newly dirtied b_dirty b_io b_more_io 3324ea879b9SWu Fengguang * =============> gf edc BA 3334ea879b9SWu Fengguang * After 3344ea879b9SWu Fengguang * newly dirtied b_dirty b_io b_more_io 3354ea879b9SWu Fengguang * =============> g fBAedc 3364ea879b9SWu Fengguang * | 3374ea879b9SWu Fengguang * +--> dequeue for IO 3382c136579SFengguang Wu */ 339ad4e38ddSCurt Wohlgemuth static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work) 3402c136579SFengguang Wu { 341e84d0a4fSWu Fengguang int moved; 3420ae45f63STheodore Ts'o 343f758eeabSChristoph Hellwig assert_spin_locked(&wb->list_lock); 3444ea879b9SWu Fengguang list_splice_init(&wb->b_more_io, &wb->b_io); 3450ae45f63STheodore Ts'o moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, 0, work); 3460ae45f63STheodore Ts'o moved += move_expired_inodes(&wb->b_dirty_time, &wb->b_io, 3470ae45f63STheodore Ts'o EXPIRE_DIRTY_ATIME, work); 348ad4e38ddSCurt Wohlgemuth trace_writeback_queue_io(wb, work, moved); 34966f3b8e2SJens Axboe } 35066f3b8e2SJens Axboe 351a9185b41SChristoph Hellwig static int write_inode(struct inode *inode, struct writeback_control *wbc) 35266f3b8e2SJens Axboe { 3539fb0a7daSTejun Heo int ret; 3549fb0a7daSTejun Heo 3559fb0a7daSTejun Heo if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) { 3569fb0a7daSTejun Heo trace_writeback_write_inode_start(inode, wbc); 3579fb0a7daSTejun Heo ret = inode->i_sb->s_op->write_inode(inode, wbc); 3589fb0a7daSTejun Heo trace_writeback_write_inode(inode, wbc); 3599fb0a7daSTejun Heo return ret; 3609fb0a7daSTejun Heo } 36103ba3782SJens Axboe return 0; 36266f3b8e2SJens Axboe } 36308d8e974SFengguang Wu 3642c136579SFengguang Wu /* 365169ebd90SJan Kara * Wait for writeback on an inode to complete. Called with i_lock held. 366169ebd90SJan Kara * Caller must make sure inode cannot go away when we drop i_lock. 36701c03194SChristoph Hellwig */ 368169ebd90SJan Kara static void __inode_wait_for_writeback(struct inode *inode) 369169ebd90SJan Kara __releases(inode->i_lock) 370169ebd90SJan Kara __acquires(inode->i_lock) 37101c03194SChristoph Hellwig { 37201c03194SChristoph Hellwig DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); 37301c03194SChristoph Hellwig wait_queue_head_t *wqh; 37401c03194SChristoph Hellwig 37501c03194SChristoph Hellwig wqh = bit_waitqueue(&inode->i_state, __I_SYNC); 37658a9d3d8SRichard Kennedy while (inode->i_state & I_SYNC) { 377250df6edSDave Chinner spin_unlock(&inode->i_lock); 37874316201SNeilBrown __wait_on_bit(wqh, &wq, bit_wait, 37974316201SNeilBrown TASK_UNINTERRUPTIBLE); 380250df6edSDave Chinner spin_lock(&inode->i_lock); 38158a9d3d8SRichard Kennedy } 38201c03194SChristoph Hellwig } 38301c03194SChristoph Hellwig 38401c03194SChristoph Hellwig /* 385169ebd90SJan Kara * Wait for writeback on an inode to complete. Caller must have inode pinned. 386169ebd90SJan Kara */ 387169ebd90SJan Kara void inode_wait_for_writeback(struct inode *inode) 388169ebd90SJan Kara { 389169ebd90SJan Kara spin_lock(&inode->i_lock); 390169ebd90SJan Kara __inode_wait_for_writeback(inode); 391169ebd90SJan Kara spin_unlock(&inode->i_lock); 392169ebd90SJan Kara } 393169ebd90SJan Kara 394169ebd90SJan Kara /* 395169ebd90SJan Kara * Sleep until I_SYNC is cleared. This function must be called with i_lock 396169ebd90SJan Kara * held and drops it. It is aimed for callers not holding any inode reference 397169ebd90SJan Kara * so once i_lock is dropped, inode can go away. 398169ebd90SJan Kara */ 399169ebd90SJan Kara static void inode_sleep_on_writeback(struct inode *inode) 400169ebd90SJan Kara __releases(inode->i_lock) 401169ebd90SJan Kara { 402169ebd90SJan Kara DEFINE_WAIT(wait); 403169ebd90SJan Kara wait_queue_head_t *wqh = bit_waitqueue(&inode->i_state, __I_SYNC); 404169ebd90SJan Kara int sleep; 405169ebd90SJan Kara 406169ebd90SJan Kara prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); 407169ebd90SJan Kara sleep = inode->i_state & I_SYNC; 408169ebd90SJan Kara spin_unlock(&inode->i_lock); 409169ebd90SJan Kara if (sleep) 410169ebd90SJan Kara schedule(); 411169ebd90SJan Kara finish_wait(wqh, &wait); 412169ebd90SJan Kara } 413169ebd90SJan Kara 414169ebd90SJan Kara /* 415ccb26b5aSJan Kara * Find proper writeback list for the inode depending on its current state and 416ccb26b5aSJan Kara * possibly also change of its state while we were doing writeback. Here we 417ccb26b5aSJan Kara * handle things such as livelock prevention or fairness of writeback among 418ccb26b5aSJan Kara * inodes. This function can be called only by flusher thread - noone else 419ccb26b5aSJan Kara * processes all inodes in writeback lists and requeueing inodes behind flusher 420ccb26b5aSJan Kara * thread's back can have unexpected consequences. 421ccb26b5aSJan Kara */ 422ccb26b5aSJan Kara static void requeue_inode(struct inode *inode, struct bdi_writeback *wb, 423ccb26b5aSJan Kara struct writeback_control *wbc) 424ccb26b5aSJan Kara { 425ccb26b5aSJan Kara if (inode->i_state & I_FREEING) 426ccb26b5aSJan Kara return; 427ccb26b5aSJan Kara 428ccb26b5aSJan Kara /* 429ccb26b5aSJan Kara * Sync livelock prevention. Each inode is tagged and synced in one 430ccb26b5aSJan Kara * shot. If still dirty, it will be redirty_tail()'ed below. Update 431ccb26b5aSJan Kara * the dirty time to prevent enqueue and sync it again. 432ccb26b5aSJan Kara */ 433ccb26b5aSJan Kara if ((inode->i_state & I_DIRTY) && 434ccb26b5aSJan Kara (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)) 435ccb26b5aSJan Kara inode->dirtied_when = jiffies; 436ccb26b5aSJan Kara 4374f8ad655SJan Kara if (wbc->pages_skipped) { 4384f8ad655SJan Kara /* 4394f8ad655SJan Kara * writeback is not making progress due to locked 4404f8ad655SJan Kara * buffers. Skip this inode for now. 4414f8ad655SJan Kara */ 4424f8ad655SJan Kara redirty_tail(inode, wb); 4434f8ad655SJan Kara return; 4444f8ad655SJan Kara } 4454f8ad655SJan Kara 446ccb26b5aSJan Kara if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) { 447ccb26b5aSJan Kara /* 448ccb26b5aSJan Kara * We didn't write back all the pages. nfs_writepages() 449ccb26b5aSJan Kara * sometimes bales out without doing anything. 450ccb26b5aSJan Kara */ 451ccb26b5aSJan Kara if (wbc->nr_to_write <= 0) { 452ccb26b5aSJan Kara /* Slice used up. Queue for next turn. */ 453ccb26b5aSJan Kara requeue_io(inode, wb); 454ccb26b5aSJan Kara } else { 455ccb26b5aSJan Kara /* 456ccb26b5aSJan Kara * Writeback blocked by something other than 457ccb26b5aSJan Kara * congestion. Delay the inode for some time to 458ccb26b5aSJan Kara * avoid spinning on the CPU (100% iowait) 459ccb26b5aSJan Kara * retrying writeback of the dirty page/inode 460ccb26b5aSJan Kara * that cannot be performed immediately. 461ccb26b5aSJan Kara */ 462ccb26b5aSJan Kara redirty_tail(inode, wb); 463ccb26b5aSJan Kara } 464ccb26b5aSJan Kara } else if (inode->i_state & I_DIRTY) { 465ccb26b5aSJan Kara /* 466ccb26b5aSJan Kara * Filesystems can dirty the inode during writeback operations, 467ccb26b5aSJan Kara * such as delayed allocation during submission or metadata 468ccb26b5aSJan Kara * updates after data IO completion. 469ccb26b5aSJan Kara */ 470ccb26b5aSJan Kara redirty_tail(inode, wb); 4710ae45f63STheodore Ts'o } else if (inode->i_state & I_DIRTY_TIME) { 472a2f48706STheodore Ts'o inode->dirtied_when = jiffies; 4730ae45f63STheodore Ts'o list_move(&inode->i_wb_list, &wb->b_dirty_time); 474ccb26b5aSJan Kara } else { 475ccb26b5aSJan Kara /* The inode is clean. Remove from writeback lists. */ 476ccb26b5aSJan Kara list_del_init(&inode->i_wb_list); 477ccb26b5aSJan Kara } 478ccb26b5aSJan Kara } 479ccb26b5aSJan Kara 480ccb26b5aSJan Kara /* 4814f8ad655SJan Kara * Write out an inode and its dirty pages. Do not update the writeback list 4824f8ad655SJan Kara * linkage. That is left to the caller. The caller is also responsible for 4834f8ad655SJan Kara * setting I_SYNC flag and calling inode_sync_complete() to clear it. 4841da177e4SLinus Torvalds */ 4851da177e4SLinus Torvalds static int 486cd8ed2a4SYan Hong __writeback_single_inode(struct inode *inode, struct writeback_control *wbc) 4871da177e4SLinus Torvalds { 4881da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 489251d6a47SWu Fengguang long nr_to_write = wbc->nr_to_write; 49001c03194SChristoph Hellwig unsigned dirty; 4911da177e4SLinus Torvalds int ret; 4921da177e4SLinus Torvalds 4934f8ad655SJan Kara WARN_ON(!(inode->i_state & I_SYNC)); 4941da177e4SLinus Torvalds 4959fb0a7daSTejun Heo trace_writeback_single_inode_start(inode, wbc, nr_to_write); 4969fb0a7daSTejun Heo 4971da177e4SLinus Torvalds ret = do_writepages(mapping, wbc); 4981da177e4SLinus Torvalds 49926821ed4SChristoph Hellwig /* 50026821ed4SChristoph Hellwig * Make sure to wait on the data before writing out the metadata. 50126821ed4SChristoph Hellwig * This is important for filesystems that modify metadata on data 5027747bd4bSDave Chinner * I/O completion. We don't do it for sync(2) writeback because it has a 5037747bd4bSDave Chinner * separate, external IO completion path and ->sync_fs for guaranteeing 5047747bd4bSDave Chinner * inode metadata is written back correctly. 50526821ed4SChristoph Hellwig */ 5067747bd4bSDave Chinner if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) { 50726821ed4SChristoph Hellwig int err = filemap_fdatawait(mapping); 5081da177e4SLinus Torvalds if (ret == 0) 5091da177e4SLinus Torvalds ret = err; 5101da177e4SLinus Torvalds } 5111da177e4SLinus Torvalds 5125547e8aaSDmitry Monakhov /* 5135547e8aaSDmitry Monakhov * Some filesystems may redirty the inode during the writeback 5145547e8aaSDmitry Monakhov * due to delalloc, clear dirty metadata flags right before 5155547e8aaSDmitry Monakhov * write_inode() 5165547e8aaSDmitry Monakhov */ 517250df6edSDave Chinner spin_lock(&inode->i_lock); 5189c6ac78eSTejun Heo 5195547e8aaSDmitry Monakhov dirty = inode->i_state & I_DIRTY; 520a2f48706STheodore Ts'o if (inode->i_state & I_DIRTY_TIME) { 521a2f48706STheodore Ts'o if ((dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) || 522a2f48706STheodore Ts'o unlikely(inode->i_state & I_DIRTY_TIME_EXPIRED) || 523a2f48706STheodore Ts'o unlikely(time_after(jiffies, 524a2f48706STheodore Ts'o (inode->dirtied_time_when + 525a2f48706STheodore Ts'o dirtytime_expire_interval * HZ)))) { 5260ae45f63STheodore Ts'o dirty |= I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED; 5270ae45f63STheodore Ts'o trace_writeback_lazytime(inode); 5280ae45f63STheodore Ts'o } 529a2f48706STheodore Ts'o } else 530a2f48706STheodore Ts'o inode->i_state &= ~I_DIRTY_TIME_EXPIRED; 5310ae45f63STheodore Ts'o inode->i_state &= ~dirty; 5329c6ac78eSTejun Heo 5339c6ac78eSTejun Heo /* 5349c6ac78eSTejun Heo * Paired with smp_mb() in __mark_inode_dirty(). This allows 5359c6ac78eSTejun Heo * __mark_inode_dirty() to test i_state without grabbing i_lock - 5369c6ac78eSTejun Heo * either they see the I_DIRTY bits cleared or we see the dirtied 5379c6ac78eSTejun Heo * inode. 5389c6ac78eSTejun Heo * 5399c6ac78eSTejun Heo * I_DIRTY_PAGES is always cleared together above even if @mapping 5409c6ac78eSTejun Heo * still has dirty pages. The flag is reinstated after smp_mb() if 5419c6ac78eSTejun Heo * necessary. This guarantees that either __mark_inode_dirty() 5429c6ac78eSTejun Heo * sees clear I_DIRTY_PAGES or we see PAGECACHE_TAG_DIRTY. 5439c6ac78eSTejun Heo */ 5449c6ac78eSTejun Heo smp_mb(); 5459c6ac78eSTejun Heo 5469c6ac78eSTejun Heo if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) 5479c6ac78eSTejun Heo inode->i_state |= I_DIRTY_PAGES; 5489c6ac78eSTejun Heo 549250df6edSDave Chinner spin_unlock(&inode->i_lock); 5509c6ac78eSTejun Heo 5510ae45f63STheodore Ts'o if (dirty & I_DIRTY_TIME) 5520ae45f63STheodore Ts'o mark_inode_dirty_sync(inode); 55326821ed4SChristoph Hellwig /* Don't write the inode if only I_DIRTY_PAGES was set */ 5540ae45f63STheodore Ts'o if (dirty & ~I_DIRTY_PAGES) { 555a9185b41SChristoph Hellwig int err = write_inode(inode, wbc); 5561da177e4SLinus Torvalds if (ret == 0) 5571da177e4SLinus Torvalds ret = err; 5581da177e4SLinus Torvalds } 5594f8ad655SJan Kara trace_writeback_single_inode(inode, wbc, nr_to_write); 5604f8ad655SJan Kara return ret; 5614f8ad655SJan Kara } 5624f8ad655SJan Kara 5634f8ad655SJan Kara /* 5644f8ad655SJan Kara * Write out an inode's dirty pages. Either the caller has an active reference 5654f8ad655SJan Kara * on the inode or the inode has I_WILL_FREE set. 5664f8ad655SJan Kara * 5674f8ad655SJan Kara * This function is designed to be called for writing back one inode which 5684f8ad655SJan Kara * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode() 5694f8ad655SJan Kara * and does more profound writeback list handling in writeback_sb_inodes(). 5704f8ad655SJan Kara */ 5714f8ad655SJan Kara static int 5724f8ad655SJan Kara writeback_single_inode(struct inode *inode, struct bdi_writeback *wb, 5734f8ad655SJan Kara struct writeback_control *wbc) 5744f8ad655SJan Kara { 5754f8ad655SJan Kara int ret = 0; 5764f8ad655SJan Kara 5774f8ad655SJan Kara spin_lock(&inode->i_lock); 5784f8ad655SJan Kara if (!atomic_read(&inode->i_count)) 5794f8ad655SJan Kara WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); 5804f8ad655SJan Kara else 5814f8ad655SJan Kara WARN_ON(inode->i_state & I_WILL_FREE); 5824f8ad655SJan Kara 5834f8ad655SJan Kara if (inode->i_state & I_SYNC) { 5844f8ad655SJan Kara if (wbc->sync_mode != WB_SYNC_ALL) 5854f8ad655SJan Kara goto out; 5864f8ad655SJan Kara /* 587169ebd90SJan Kara * It's a data-integrity sync. We must wait. Since callers hold 588169ebd90SJan Kara * inode reference or inode has I_WILL_FREE set, it cannot go 589169ebd90SJan Kara * away under us. 5904f8ad655SJan Kara */ 591169ebd90SJan Kara __inode_wait_for_writeback(inode); 5924f8ad655SJan Kara } 5934f8ad655SJan Kara WARN_ON(inode->i_state & I_SYNC); 5944f8ad655SJan Kara /* 595f9b0e058SJan Kara * Skip inode if it is clean and we have no outstanding writeback in 596f9b0e058SJan Kara * WB_SYNC_ALL mode. We don't want to mess with writeback lists in this 597f9b0e058SJan Kara * function since flusher thread may be doing for example sync in 598f9b0e058SJan Kara * parallel and if we move the inode, it could get skipped. So here we 599f9b0e058SJan Kara * make sure inode is on some writeback list and leave it there unless 600f9b0e058SJan Kara * we have completely cleaned the inode. 6014f8ad655SJan Kara */ 6020ae45f63STheodore Ts'o if (!(inode->i_state & I_DIRTY_ALL) && 603f9b0e058SJan Kara (wbc->sync_mode != WB_SYNC_ALL || 604f9b0e058SJan Kara !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK))) 6054f8ad655SJan Kara goto out; 6064f8ad655SJan Kara inode->i_state |= I_SYNC; 6074f8ad655SJan Kara spin_unlock(&inode->i_lock); 6084f8ad655SJan Kara 609cd8ed2a4SYan Hong ret = __writeback_single_inode(inode, wbc); 6101da177e4SLinus Torvalds 611f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 612250df6edSDave Chinner spin_lock(&inode->i_lock); 6134f8ad655SJan Kara /* 6144f8ad655SJan Kara * If inode is clean, remove it from writeback lists. Otherwise don't 6154f8ad655SJan Kara * touch it. See comment above for explanation. 6164f8ad655SJan Kara */ 6170ae45f63STheodore Ts'o if (!(inode->i_state & I_DIRTY_ALL)) 6184f8ad655SJan Kara list_del_init(&inode->i_wb_list); 6194f8ad655SJan Kara spin_unlock(&wb->list_lock); 6201c0eeaf5SJoern Engel inode_sync_complete(inode); 6214f8ad655SJan Kara out: 6224f8ad655SJan Kara spin_unlock(&inode->i_lock); 6231da177e4SLinus Torvalds return ret; 6241da177e4SLinus Torvalds } 6251da177e4SLinus Torvalds 626a88a341aSTejun Heo static long writeback_chunk_size(struct bdi_writeback *wb, 6271a12d8bdSWu Fengguang struct wb_writeback_work *work) 628d46db3d5SWu Fengguang { 629d46db3d5SWu Fengguang long pages; 630d46db3d5SWu Fengguang 631d46db3d5SWu Fengguang /* 632d46db3d5SWu Fengguang * WB_SYNC_ALL mode does livelock avoidance by syncing dirty 633d46db3d5SWu Fengguang * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX 634d46db3d5SWu Fengguang * here avoids calling into writeback_inodes_wb() more than once. 635d46db3d5SWu Fengguang * 636d46db3d5SWu Fengguang * The intended call sequence for WB_SYNC_ALL writeback is: 637d46db3d5SWu Fengguang * 638d46db3d5SWu Fengguang * wb_writeback() 639d46db3d5SWu Fengguang * writeback_sb_inodes() <== called only once 640d46db3d5SWu Fengguang * write_cache_pages() <== called once for each inode 641d46db3d5SWu Fengguang * (quickly) tag currently dirty pages 642d46db3d5SWu Fengguang * (maybe slowly) sync all tagged pages 643d46db3d5SWu Fengguang */ 644d46db3d5SWu Fengguang if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages) 645d46db3d5SWu Fengguang pages = LONG_MAX; 6461a12d8bdSWu Fengguang else { 647a88a341aSTejun Heo pages = min(wb->avg_write_bandwidth / 2, 6481a12d8bdSWu Fengguang global_dirty_limit / DIRTY_SCOPE); 6491a12d8bdSWu Fengguang pages = min(pages, work->nr_pages); 6501a12d8bdSWu Fengguang pages = round_down(pages + MIN_WRITEBACK_PAGES, 6511a12d8bdSWu Fengguang MIN_WRITEBACK_PAGES); 6521a12d8bdSWu Fengguang } 653d46db3d5SWu Fengguang 654d46db3d5SWu Fengguang return pages; 655d46db3d5SWu Fengguang } 656d46db3d5SWu Fengguang 65703ba3782SJens Axboe /* 658f11c9c5cSEdward Shishkin * Write a portion of b_io inodes which belong to @sb. 659edadfb10SChristoph Hellwig * 660d46db3d5SWu Fengguang * Return the number of pages and/or inodes written. 661f11c9c5cSEdward Shishkin */ 662d46db3d5SWu Fengguang static long writeback_sb_inodes(struct super_block *sb, 663d46db3d5SWu Fengguang struct bdi_writeback *wb, 664d46db3d5SWu Fengguang struct wb_writeback_work *work) 66503ba3782SJens Axboe { 666d46db3d5SWu Fengguang struct writeback_control wbc = { 667d46db3d5SWu Fengguang .sync_mode = work->sync_mode, 668d46db3d5SWu Fengguang .tagged_writepages = work->tagged_writepages, 669d46db3d5SWu Fengguang .for_kupdate = work->for_kupdate, 670d46db3d5SWu Fengguang .for_background = work->for_background, 6717747bd4bSDave Chinner .for_sync = work->for_sync, 672d46db3d5SWu Fengguang .range_cyclic = work->range_cyclic, 673d46db3d5SWu Fengguang .range_start = 0, 674d46db3d5SWu Fengguang .range_end = LLONG_MAX, 675d46db3d5SWu Fengguang }; 676d46db3d5SWu Fengguang unsigned long start_time = jiffies; 677d46db3d5SWu Fengguang long write_chunk; 678d46db3d5SWu Fengguang long wrote = 0; /* count both pages and inodes */ 679d46db3d5SWu Fengguang 68003ba3782SJens Axboe while (!list_empty(&wb->b_io)) { 6817ccf19a8SNick Piggin struct inode *inode = wb_inode(wb->b_io.prev); 682edadfb10SChristoph Hellwig 683edadfb10SChristoph Hellwig if (inode->i_sb != sb) { 684d46db3d5SWu Fengguang if (work->sb) { 685edadfb10SChristoph Hellwig /* 686edadfb10SChristoph Hellwig * We only want to write back data for this 687edadfb10SChristoph Hellwig * superblock, move all inodes not belonging 688edadfb10SChristoph Hellwig * to it back onto the dirty list. 689edadfb10SChristoph Hellwig */ 690f758eeabSChristoph Hellwig redirty_tail(inode, wb); 69166f3b8e2SJens Axboe continue; 69266f3b8e2SJens Axboe } 693edadfb10SChristoph Hellwig 694edadfb10SChristoph Hellwig /* 695edadfb10SChristoph Hellwig * The inode belongs to a different superblock. 696edadfb10SChristoph Hellwig * Bounce back to the caller to unpin this and 697edadfb10SChristoph Hellwig * pin the next superblock. 698edadfb10SChristoph Hellwig */ 699d46db3d5SWu Fengguang break; 700edadfb10SChristoph Hellwig } 701edadfb10SChristoph Hellwig 7029843b76aSChristoph Hellwig /* 703331cbdeeSWanpeng Li * Don't bother with new inodes or inodes being freed, first 704331cbdeeSWanpeng Li * kind does not need periodic writeout yet, and for the latter 7059843b76aSChristoph Hellwig * kind writeout is handled by the freer. 7069843b76aSChristoph Hellwig */ 707250df6edSDave Chinner spin_lock(&inode->i_lock); 7089843b76aSChristoph Hellwig if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { 709250df6edSDave Chinner spin_unlock(&inode->i_lock); 710fcc5c222SWu Fengguang redirty_tail(inode, wb); 7117ef0d737SNick Piggin continue; 7127ef0d737SNick Piggin } 713cc1676d9SJan Kara if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) { 714cc1676d9SJan Kara /* 715cc1676d9SJan Kara * If this inode is locked for writeback and we are not 716cc1676d9SJan Kara * doing writeback-for-data-integrity, move it to 717cc1676d9SJan Kara * b_more_io so that writeback can proceed with the 718cc1676d9SJan Kara * other inodes on s_io. 719cc1676d9SJan Kara * 720cc1676d9SJan Kara * We'll have another go at writing back this inode 721cc1676d9SJan Kara * when we completed a full scan of b_io. 722cc1676d9SJan Kara */ 723cc1676d9SJan Kara spin_unlock(&inode->i_lock); 724cc1676d9SJan Kara requeue_io(inode, wb); 725cc1676d9SJan Kara trace_writeback_sb_inodes_requeue(inode); 726cc1676d9SJan Kara continue; 727cc1676d9SJan Kara } 728f0d07b7fSJan Kara spin_unlock(&wb->list_lock); 729f0d07b7fSJan Kara 7304f8ad655SJan Kara /* 7314f8ad655SJan Kara * We already requeued the inode if it had I_SYNC set and we 7324f8ad655SJan Kara * are doing WB_SYNC_NONE writeback. So this catches only the 7334f8ad655SJan Kara * WB_SYNC_ALL case. 7344f8ad655SJan Kara */ 735169ebd90SJan Kara if (inode->i_state & I_SYNC) { 736169ebd90SJan Kara /* Wait for I_SYNC. This function drops i_lock... */ 737169ebd90SJan Kara inode_sleep_on_writeback(inode); 738169ebd90SJan Kara /* Inode may be gone, start again */ 739ead188f9SJan Kara spin_lock(&wb->list_lock); 740169ebd90SJan Kara continue; 741169ebd90SJan Kara } 7424f8ad655SJan Kara inode->i_state |= I_SYNC; 7434f8ad655SJan Kara spin_unlock(&inode->i_lock); 744169ebd90SJan Kara 745a88a341aSTejun Heo write_chunk = writeback_chunk_size(wb, work); 746d46db3d5SWu Fengguang wbc.nr_to_write = write_chunk; 747d46db3d5SWu Fengguang wbc.pages_skipped = 0; 748250df6edSDave Chinner 749169ebd90SJan Kara /* 750169ebd90SJan Kara * We use I_SYNC to pin the inode in memory. While it is set 751169ebd90SJan Kara * evict_inode() will wait so the inode cannot be freed. 752169ebd90SJan Kara */ 753cd8ed2a4SYan Hong __writeback_single_inode(inode, &wbc); 754d46db3d5SWu Fengguang 755d46db3d5SWu Fengguang work->nr_pages -= write_chunk - wbc.nr_to_write; 756d46db3d5SWu Fengguang wrote += write_chunk - wbc.nr_to_write; 7574f8ad655SJan Kara spin_lock(&wb->list_lock); 7584f8ad655SJan Kara spin_lock(&inode->i_lock); 7590ae45f63STheodore Ts'o if (!(inode->i_state & I_DIRTY_ALL)) 760d46db3d5SWu Fengguang wrote++; 7614f8ad655SJan Kara requeue_inode(inode, wb, &wbc); 7624f8ad655SJan Kara inode_sync_complete(inode); 7630f1b1fd8SDave Chinner spin_unlock(&inode->i_lock); 764169ebd90SJan Kara cond_resched_lock(&wb->list_lock); 765d46db3d5SWu Fengguang /* 766d46db3d5SWu Fengguang * bail out to wb_writeback() often enough to check 767d46db3d5SWu Fengguang * background threshold and other termination conditions. 768d46db3d5SWu Fengguang */ 769d46db3d5SWu Fengguang if (wrote) { 770d46db3d5SWu Fengguang if (time_is_before_jiffies(start_time + HZ / 10UL)) 771d46db3d5SWu Fengguang break; 772d46db3d5SWu Fengguang if (work->nr_pages <= 0) 773d46db3d5SWu Fengguang break; 7741da177e4SLinus Torvalds } 7758bc3be27SFengguang Wu } 776d46db3d5SWu Fengguang return wrote; 777f11c9c5cSEdward Shishkin } 77838f21977SNick Piggin 779d46db3d5SWu Fengguang static long __writeback_inodes_wb(struct bdi_writeback *wb, 780d46db3d5SWu Fengguang struct wb_writeback_work *work) 781f11c9c5cSEdward Shishkin { 782d46db3d5SWu Fengguang unsigned long start_time = jiffies; 783d46db3d5SWu Fengguang long wrote = 0; 784f11c9c5cSEdward Shishkin 785f11c9c5cSEdward Shishkin while (!list_empty(&wb->b_io)) { 7867ccf19a8SNick Piggin struct inode *inode = wb_inode(wb->b_io.prev); 787f11c9c5cSEdward Shishkin struct super_block *sb = inode->i_sb; 788f11c9c5cSEdward Shishkin 789eb6ef3dfSKonstantin Khlebnikov if (!trylock_super(sb)) { 7900e995816SWu Fengguang /* 791eb6ef3dfSKonstantin Khlebnikov * trylock_super() may fail consistently due to 7920e995816SWu Fengguang * s_umount being grabbed by someone else. Don't use 7930e995816SWu Fengguang * requeue_io() to avoid busy retrying the inode/sb. 7940e995816SWu Fengguang */ 7950e995816SWu Fengguang redirty_tail(inode, wb); 796d19de7edSChristoph Hellwig continue; 797334132aeSChristoph Hellwig } 798d46db3d5SWu Fengguang wrote += writeback_sb_inodes(sb, wb, work); 799eb6ef3dfSKonstantin Khlebnikov up_read(&sb->s_umount); 800f11c9c5cSEdward Shishkin 801d46db3d5SWu Fengguang /* refer to the same tests at the end of writeback_sb_inodes */ 802d46db3d5SWu Fengguang if (wrote) { 803d46db3d5SWu Fengguang if (time_is_before_jiffies(start_time + HZ / 10UL)) 804d46db3d5SWu Fengguang break; 805d46db3d5SWu Fengguang if (work->nr_pages <= 0) 806f11c9c5cSEdward Shishkin break; 807f11c9c5cSEdward Shishkin } 808d46db3d5SWu Fengguang } 80966f3b8e2SJens Axboe /* Leave any unwritten inodes on b_io */ 810d46db3d5SWu Fengguang return wrote; 81166f3b8e2SJens Axboe } 81266f3b8e2SJens Axboe 8137d9f073bSWanpeng Li static long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages, 8140e175a18SCurt Wohlgemuth enum wb_reason reason) 815edadfb10SChristoph Hellwig { 816d46db3d5SWu Fengguang struct wb_writeback_work work = { 817d46db3d5SWu Fengguang .nr_pages = nr_pages, 818d46db3d5SWu Fengguang .sync_mode = WB_SYNC_NONE, 819d46db3d5SWu Fengguang .range_cyclic = 1, 8200e175a18SCurt Wohlgemuth .reason = reason, 821d46db3d5SWu Fengguang }; 822edadfb10SChristoph Hellwig 823f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 824424b351fSWu Fengguang if (list_empty(&wb->b_io)) 825ad4e38ddSCurt Wohlgemuth queue_io(wb, &work); 826d46db3d5SWu Fengguang __writeback_inodes_wb(wb, &work); 827f758eeabSChristoph Hellwig spin_unlock(&wb->list_lock); 828edadfb10SChristoph Hellwig 829d46db3d5SWu Fengguang return nr_pages - work.nr_pages; 83066f3b8e2SJens Axboe } 83166f3b8e2SJens Axboe 832a88a341aSTejun Heo static bool over_bground_thresh(struct bdi_writeback *wb) 83303ba3782SJens Axboe { 83403ba3782SJens Axboe unsigned long background_thresh, dirty_thresh; 83503ba3782SJens Axboe 83616c4042fSWu Fengguang global_dirty_limits(&background_thresh, &dirty_thresh); 83703ba3782SJens Axboe 838b00949aaSWu Fengguang if (global_page_state(NR_FILE_DIRTY) + 839b00949aaSWu Fengguang global_page_state(NR_UNSTABLE_NFS) > background_thresh) 840b00949aaSWu Fengguang return true; 841b00949aaSWu Fengguang 842a88a341aSTejun Heo if (wb_stat(wb, WB_RECLAIMABLE) > wb_dirty_limit(wb, background_thresh)) 843b00949aaSWu Fengguang return true; 844b00949aaSWu Fengguang 845b00949aaSWu Fengguang return false; 84603ba3782SJens Axboe } 84703ba3782SJens Axboe 84803ba3782SJens Axboe /* 849e98be2d5SWu Fengguang * Called under wb->list_lock. If there are multiple wb per bdi, 850e98be2d5SWu Fengguang * only the flusher working on the first wb should do it. 851e98be2d5SWu Fengguang */ 852e98be2d5SWu Fengguang static void wb_update_bandwidth(struct bdi_writeback *wb, 853e98be2d5SWu Fengguang unsigned long start_time) 854e98be2d5SWu Fengguang { 855a88a341aSTejun Heo __wb_update_bandwidth(wb, 0, 0, 0, 0, 0, start_time); 856e98be2d5SWu Fengguang } 857e98be2d5SWu Fengguang 858e98be2d5SWu Fengguang /* 85903ba3782SJens Axboe * Explicit flushing or periodic writeback of "old" data. 86003ba3782SJens Axboe * 86103ba3782SJens Axboe * Define "old": the first time one of an inode's pages is dirtied, we mark the 86203ba3782SJens Axboe * dirtying-time in the inode's address_space. So this periodic writeback code 86303ba3782SJens Axboe * just walks the superblock inode list, writing back any inodes which are 86403ba3782SJens Axboe * older than a specific point in time. 86503ba3782SJens Axboe * 86603ba3782SJens Axboe * Try to run once per dirty_writeback_interval. But if a writeback event 86703ba3782SJens Axboe * takes longer than a dirty_writeback_interval interval, then leave a 86803ba3782SJens Axboe * one-second gap. 86903ba3782SJens Axboe * 87003ba3782SJens Axboe * older_than_this takes precedence over nr_to_write. So we'll only write back 87103ba3782SJens Axboe * all dirty pages if they are all attached to "old" mappings. 87203ba3782SJens Axboe */ 873c4a77a6cSJens Axboe static long wb_writeback(struct bdi_writeback *wb, 87483ba7b07SChristoph Hellwig struct wb_writeback_work *work) 87503ba3782SJens Axboe { 876e98be2d5SWu Fengguang unsigned long wb_start = jiffies; 877d46db3d5SWu Fengguang long nr_pages = work->nr_pages; 8780dc83bd3SJan Kara unsigned long oldest_jif; 879a5989bdcSJan Kara struct inode *inode; 880d46db3d5SWu Fengguang long progress; 88103ba3782SJens Axboe 8820dc83bd3SJan Kara oldest_jif = jiffies; 8830dc83bd3SJan Kara work->older_than_this = &oldest_jif; 88403ba3782SJens Axboe 885e8dfc305SWu Fengguang spin_lock(&wb->list_lock); 88603ba3782SJens Axboe for (;;) { 88703ba3782SJens Axboe /* 888d3ddec76SWu Fengguang * Stop writeback when nr_pages has been consumed 88903ba3782SJens Axboe */ 89083ba7b07SChristoph Hellwig if (work->nr_pages <= 0) 89103ba3782SJens Axboe break; 89203ba3782SJens Axboe 89303ba3782SJens Axboe /* 894aa373cf5SJan Kara * Background writeout and kupdate-style writeback may 895aa373cf5SJan Kara * run forever. Stop them if there is other work to do 896aa373cf5SJan Kara * so that e.g. sync can proceed. They'll be restarted 897aa373cf5SJan Kara * after the other works are all done. 898aa373cf5SJan Kara */ 899aa373cf5SJan Kara if ((work->for_background || work->for_kupdate) && 900f0054bb1STejun Heo !list_empty(&wb->work_list)) 901aa373cf5SJan Kara break; 902aa373cf5SJan Kara 903aa373cf5SJan Kara /* 904d3ddec76SWu Fengguang * For background writeout, stop when we are below the 905d3ddec76SWu Fengguang * background dirty threshold 90603ba3782SJens Axboe */ 907a88a341aSTejun Heo if (work->for_background && !over_bground_thresh(wb)) 90803ba3782SJens Axboe break; 90903ba3782SJens Axboe 9101bc36b64SJan Kara /* 9111bc36b64SJan Kara * Kupdate and background works are special and we want to 9121bc36b64SJan Kara * include all inodes that need writing. Livelock avoidance is 9131bc36b64SJan Kara * handled by these works yielding to any other work so we are 9141bc36b64SJan Kara * safe. 9151bc36b64SJan Kara */ 916ba9aa839SWu Fengguang if (work->for_kupdate) { 9170dc83bd3SJan Kara oldest_jif = jiffies - 918ba9aa839SWu Fengguang msecs_to_jiffies(dirty_expire_interval * 10); 9191bc36b64SJan Kara } else if (work->for_background) 9200dc83bd3SJan Kara oldest_jif = jiffies; 921028c2dd1SDave Chinner 922d46db3d5SWu Fengguang trace_writeback_start(wb->bdi, work); 923e8dfc305SWu Fengguang if (list_empty(&wb->b_io)) 924ad4e38ddSCurt Wohlgemuth queue_io(wb, work); 92583ba7b07SChristoph Hellwig if (work->sb) 926d46db3d5SWu Fengguang progress = writeback_sb_inodes(work->sb, wb, work); 927edadfb10SChristoph Hellwig else 928d46db3d5SWu Fengguang progress = __writeback_inodes_wb(wb, work); 929d46db3d5SWu Fengguang trace_writeback_written(wb->bdi, work); 930028c2dd1SDave Chinner 931e98be2d5SWu Fengguang wb_update_bandwidth(wb, wb_start); 93203ba3782SJens Axboe 93303ba3782SJens Axboe /* 93471fd05a8SJens Axboe * Did we write something? Try for more 935e6fb6da2SWu Fengguang * 936e6fb6da2SWu Fengguang * Dirty inodes are moved to b_io for writeback in batches. 937e6fb6da2SWu Fengguang * The completion of the current batch does not necessarily 938e6fb6da2SWu Fengguang * mean the overall work is done. So we keep looping as long 939e6fb6da2SWu Fengguang * as made some progress on cleaning pages or inodes. 94071fd05a8SJens Axboe */ 941d46db3d5SWu Fengguang if (progress) 94203ba3782SJens Axboe continue; 943a5989bdcSJan Kara /* 944e6fb6da2SWu Fengguang * No more inodes for IO, bail 945a5989bdcSJan Kara */ 946b7a2441fSWu Fengguang if (list_empty(&wb->b_more_io)) 94703ba3782SJens Axboe break; 94803ba3782SJens Axboe /* 9498010c3b6SJens Axboe * Nothing written. Wait for some inode to 9508010c3b6SJens Axboe * become available for writeback. Otherwise 9518010c3b6SJens Axboe * we'll just busyloop. 95203ba3782SJens Axboe */ 95303ba3782SJens Axboe if (!list_empty(&wb->b_more_io)) { 954d46db3d5SWu Fengguang trace_writeback_wait(wb->bdi, work); 95503ba3782SJens Axboe inode = wb_inode(wb->b_more_io.prev); 956250df6edSDave Chinner spin_lock(&inode->i_lock); 957f0d07b7fSJan Kara spin_unlock(&wb->list_lock); 958169ebd90SJan Kara /* This function drops i_lock... */ 959169ebd90SJan Kara inode_sleep_on_writeback(inode); 960f0d07b7fSJan Kara spin_lock(&wb->list_lock); 96103ba3782SJens Axboe } 96203ba3782SJens Axboe } 963e8dfc305SWu Fengguang spin_unlock(&wb->list_lock); 96403ba3782SJens Axboe 965d46db3d5SWu Fengguang return nr_pages - work->nr_pages; 96603ba3782SJens Axboe } 96703ba3782SJens Axboe 96803ba3782SJens Axboe /* 96983ba7b07SChristoph Hellwig * Return the next wb_writeback_work struct that hasn't been processed yet. 97003ba3782SJens Axboe */ 971f0054bb1STejun Heo static struct wb_writeback_work *get_next_work_item(struct bdi_writeback *wb) 97203ba3782SJens Axboe { 97383ba7b07SChristoph Hellwig struct wb_writeback_work *work = NULL; 97403ba3782SJens Axboe 975f0054bb1STejun Heo spin_lock_bh(&wb->work_lock); 976f0054bb1STejun Heo if (!list_empty(&wb->work_list)) { 977f0054bb1STejun Heo work = list_entry(wb->work_list.next, 97883ba7b07SChristoph Hellwig struct wb_writeback_work, list); 97983ba7b07SChristoph Hellwig list_del_init(&work->list); 98003ba3782SJens Axboe } 981f0054bb1STejun Heo spin_unlock_bh(&wb->work_lock); 98283ba7b07SChristoph Hellwig return work; 98303ba3782SJens Axboe } 98403ba3782SJens Axboe 985cdf01dd5SLinus Torvalds /* 986cdf01dd5SLinus Torvalds * Add in the number of potentially dirty inodes, because each inode 987cdf01dd5SLinus Torvalds * write can dirty pagecache in the underlying blockdev. 988cdf01dd5SLinus Torvalds */ 989cdf01dd5SLinus Torvalds static unsigned long get_nr_dirty_pages(void) 990cdf01dd5SLinus Torvalds { 991cdf01dd5SLinus Torvalds return global_page_state(NR_FILE_DIRTY) + 992cdf01dd5SLinus Torvalds global_page_state(NR_UNSTABLE_NFS) + 993cdf01dd5SLinus Torvalds get_nr_dirty_inodes(); 994cdf01dd5SLinus Torvalds } 995cdf01dd5SLinus Torvalds 9966585027aSJan Kara static long wb_check_background_flush(struct bdi_writeback *wb) 9976585027aSJan Kara { 998a88a341aSTejun Heo if (over_bground_thresh(wb)) { 9996585027aSJan Kara 10006585027aSJan Kara struct wb_writeback_work work = { 10016585027aSJan Kara .nr_pages = LONG_MAX, 10026585027aSJan Kara .sync_mode = WB_SYNC_NONE, 10036585027aSJan Kara .for_background = 1, 10046585027aSJan Kara .range_cyclic = 1, 10050e175a18SCurt Wohlgemuth .reason = WB_REASON_BACKGROUND, 10066585027aSJan Kara }; 10076585027aSJan Kara 10086585027aSJan Kara return wb_writeback(wb, &work); 10096585027aSJan Kara } 10106585027aSJan Kara 10116585027aSJan Kara return 0; 10126585027aSJan Kara } 10136585027aSJan Kara 101403ba3782SJens Axboe static long wb_check_old_data_flush(struct bdi_writeback *wb) 101503ba3782SJens Axboe { 101603ba3782SJens Axboe unsigned long expired; 101703ba3782SJens Axboe long nr_pages; 101803ba3782SJens Axboe 101969b62d01SJens Axboe /* 102069b62d01SJens Axboe * When set to zero, disable periodic writeback 102169b62d01SJens Axboe */ 102269b62d01SJens Axboe if (!dirty_writeback_interval) 102369b62d01SJens Axboe return 0; 102469b62d01SJens Axboe 102503ba3782SJens Axboe expired = wb->last_old_flush + 102603ba3782SJens Axboe msecs_to_jiffies(dirty_writeback_interval * 10); 102703ba3782SJens Axboe if (time_before(jiffies, expired)) 102803ba3782SJens Axboe return 0; 102903ba3782SJens Axboe 103003ba3782SJens Axboe wb->last_old_flush = jiffies; 1031cdf01dd5SLinus Torvalds nr_pages = get_nr_dirty_pages(); 103203ba3782SJens Axboe 1033c4a77a6cSJens Axboe if (nr_pages) { 103483ba7b07SChristoph Hellwig struct wb_writeback_work work = { 1035c4a77a6cSJens Axboe .nr_pages = nr_pages, 1036c4a77a6cSJens Axboe .sync_mode = WB_SYNC_NONE, 1037c4a77a6cSJens Axboe .for_kupdate = 1, 1038c4a77a6cSJens Axboe .range_cyclic = 1, 10390e175a18SCurt Wohlgemuth .reason = WB_REASON_PERIODIC, 1040c4a77a6cSJens Axboe }; 1041c4a77a6cSJens Axboe 104283ba7b07SChristoph Hellwig return wb_writeback(wb, &work); 1043c4a77a6cSJens Axboe } 104403ba3782SJens Axboe 104503ba3782SJens Axboe return 0; 104603ba3782SJens Axboe } 104703ba3782SJens Axboe 104803ba3782SJens Axboe /* 104903ba3782SJens Axboe * Retrieve work items and do the writeback they describe 105003ba3782SJens Axboe */ 105125d130baSWanpeng Li static long wb_do_writeback(struct bdi_writeback *wb) 105203ba3782SJens Axboe { 105383ba7b07SChristoph Hellwig struct wb_writeback_work *work; 1054c4a77a6cSJens Axboe long wrote = 0; 105503ba3782SJens Axboe 10564452226eSTejun Heo set_bit(WB_writeback_running, &wb->state); 1057f0054bb1STejun Heo while ((work = get_next_work_item(wb)) != NULL) { 105883ba7b07SChristoph Hellwig 1059f0054bb1STejun Heo trace_writeback_exec(wb->bdi, work); 1060455b2864SDave Chinner 106183ba7b07SChristoph Hellwig wrote += wb_writeback(wb, work); 106203ba3782SJens Axboe 106303ba3782SJens Axboe /* 106483ba7b07SChristoph Hellwig * Notify the caller of completion if this is a synchronous 106583ba7b07SChristoph Hellwig * work item, otherwise just free it. 106603ba3782SJens Axboe */ 106783ba7b07SChristoph Hellwig if (work->done) 106883ba7b07SChristoph Hellwig complete(work->done); 106983ba7b07SChristoph Hellwig else 107083ba7b07SChristoph Hellwig kfree(work); 107103ba3782SJens Axboe } 107203ba3782SJens Axboe 107303ba3782SJens Axboe /* 107403ba3782SJens Axboe * Check for periodic writeback, kupdated() style 107503ba3782SJens Axboe */ 107603ba3782SJens Axboe wrote += wb_check_old_data_flush(wb); 10776585027aSJan Kara wrote += wb_check_background_flush(wb); 10784452226eSTejun Heo clear_bit(WB_writeback_running, &wb->state); 107903ba3782SJens Axboe 108003ba3782SJens Axboe return wrote; 108103ba3782SJens Axboe } 108203ba3782SJens Axboe 108303ba3782SJens Axboe /* 108403ba3782SJens Axboe * Handle writeback of dirty data for the device backed by this bdi. Also 1085839a8e86STejun Heo * reschedules periodically and does kupdated style flushing. 108603ba3782SJens Axboe */ 1087f0054bb1STejun Heo void wb_workfn(struct work_struct *work) 108803ba3782SJens Axboe { 1089839a8e86STejun Heo struct bdi_writeback *wb = container_of(to_delayed_work(work), 1090839a8e86STejun Heo struct bdi_writeback, dwork); 109103ba3782SJens Axboe long pages_written; 109203ba3782SJens Axboe 1093f0054bb1STejun Heo set_worker_desc("flush-%s", dev_name(wb->bdi->dev)); 1094766f9164SPeter Zijlstra current->flags |= PF_SWAPWRITE; 109503ba3782SJens Axboe 1096839a8e86STejun Heo if (likely(!current_is_workqueue_rescuer() || 10974452226eSTejun Heo !test_bit(WB_registered, &wb->state))) { 109803ba3782SJens Axboe /* 1099f0054bb1STejun Heo * The normal path. Keep writing back @wb until its 1100839a8e86STejun Heo * work_list is empty. Note that this path is also taken 1101f0054bb1STejun Heo * if @wb is shutting down even when we're running off the 1102839a8e86STejun Heo * rescuer as work_list needs to be drained. 110303ba3782SJens Axboe */ 1104839a8e86STejun Heo do { 110525d130baSWanpeng Li pages_written = wb_do_writeback(wb); 1106455b2864SDave Chinner trace_writeback_pages_written(pages_written); 1107f0054bb1STejun Heo } while (!list_empty(&wb->work_list)); 1108839a8e86STejun Heo } else { 1109253c34e9SArtem Bityutskiy /* 1110839a8e86STejun Heo * bdi_wq can't get enough workers and we're running off 1111839a8e86STejun Heo * the emergency worker. Don't hog it. Hopefully, 1024 is 1112839a8e86STejun Heo * enough for efficient IO. 1113253c34e9SArtem Bityutskiy */ 1114f0054bb1STejun Heo pages_written = writeback_inodes_wb(wb, 1024, 1115839a8e86STejun Heo WB_REASON_FORKER_THREAD); 1116839a8e86STejun Heo trace_writeback_pages_written(pages_written); 111703ba3782SJens Axboe } 111803ba3782SJens Axboe 1119f0054bb1STejun Heo if (!list_empty(&wb->work_list)) 11206ca738d6SDerek Basehore mod_delayed_work(bdi_wq, &wb->dwork, 0); 11216ca738d6SDerek Basehore else if (wb_has_dirty_io(wb) && dirty_writeback_interval) 1122f0054bb1STejun Heo wb_wakeup_delayed(wb); 1123455b2864SDave Chinner 1124839a8e86STejun Heo current->flags &= ~PF_SWAPWRITE; 112503ba3782SJens Axboe } 112603ba3782SJens Axboe 112703ba3782SJens Axboe /* 112803ba3782SJens Axboe * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back 112903ba3782SJens Axboe * the whole world. 113003ba3782SJens Axboe */ 11310e175a18SCurt Wohlgemuth void wakeup_flusher_threads(long nr_pages, enum wb_reason reason) 113203ba3782SJens Axboe { 1133b8c2f347SChristoph Hellwig struct backing_dev_info *bdi; 1134b8c2f347SChristoph Hellwig 113547df3ddeSJan Kara if (!nr_pages) 113647df3ddeSJan Kara nr_pages = get_nr_dirty_pages(); 1137b8c2f347SChristoph Hellwig 1138b8c2f347SChristoph Hellwig rcu_read_lock(); 1139b8c2f347SChristoph Hellwig list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { 1140b8c2f347SChristoph Hellwig if (!bdi_has_dirty_io(bdi)) 1141b8c2f347SChristoph Hellwig continue; 1142f0054bb1STejun Heo __wb_start_writeback(&bdi->wb, nr_pages, false, reason); 1143b8c2f347SChristoph Hellwig } 1144b8c2f347SChristoph Hellwig rcu_read_unlock(); 114503ba3782SJens Axboe } 114603ba3782SJens Axboe 1147a2f48706STheodore Ts'o /* 1148a2f48706STheodore Ts'o * Wake up bdi's periodically to make sure dirtytime inodes gets 1149a2f48706STheodore Ts'o * written back periodically. We deliberately do *not* check the 1150a2f48706STheodore Ts'o * b_dirtytime list in wb_has_dirty_io(), since this would cause the 1151a2f48706STheodore Ts'o * kernel to be constantly waking up once there are any dirtytime 1152a2f48706STheodore Ts'o * inodes on the system. So instead we define a separate delayed work 1153a2f48706STheodore Ts'o * function which gets called much more rarely. (By default, only 1154a2f48706STheodore Ts'o * once every 12 hours.) 1155a2f48706STheodore Ts'o * 1156a2f48706STheodore Ts'o * If there is any other write activity going on in the file system, 1157a2f48706STheodore Ts'o * this function won't be necessary. But if the only thing that has 1158a2f48706STheodore Ts'o * happened on the file system is a dirtytime inode caused by an atime 1159a2f48706STheodore Ts'o * update, we need this infrastructure below to make sure that inode 1160a2f48706STheodore Ts'o * eventually gets pushed out to disk. 1161a2f48706STheodore Ts'o */ 1162a2f48706STheodore Ts'o static void wakeup_dirtytime_writeback(struct work_struct *w); 1163a2f48706STheodore Ts'o static DECLARE_DELAYED_WORK(dirtytime_work, wakeup_dirtytime_writeback); 1164a2f48706STheodore Ts'o 1165a2f48706STheodore Ts'o static void wakeup_dirtytime_writeback(struct work_struct *w) 1166a2f48706STheodore Ts'o { 1167a2f48706STheodore Ts'o struct backing_dev_info *bdi; 1168a2f48706STheodore Ts'o 1169a2f48706STheodore Ts'o rcu_read_lock(); 1170a2f48706STheodore Ts'o list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { 1171a2f48706STheodore Ts'o if (list_empty(&bdi->wb.b_dirty_time)) 1172a2f48706STheodore Ts'o continue; 1173f0054bb1STejun Heo wb_wakeup(&bdi->wb); 1174a2f48706STheodore Ts'o } 1175a2f48706STheodore Ts'o rcu_read_unlock(); 1176a2f48706STheodore Ts'o schedule_delayed_work(&dirtytime_work, dirtytime_expire_interval * HZ); 1177a2f48706STheodore Ts'o } 1178a2f48706STheodore Ts'o 1179a2f48706STheodore Ts'o static int __init start_dirtytime_writeback(void) 1180a2f48706STheodore Ts'o { 1181a2f48706STheodore Ts'o schedule_delayed_work(&dirtytime_work, dirtytime_expire_interval * HZ); 1182a2f48706STheodore Ts'o return 0; 1183a2f48706STheodore Ts'o } 1184a2f48706STheodore Ts'o __initcall(start_dirtytime_writeback); 1185a2f48706STheodore Ts'o 11861efff914STheodore Ts'o int dirtytime_interval_handler(struct ctl_table *table, int write, 11871efff914STheodore Ts'o void __user *buffer, size_t *lenp, loff_t *ppos) 11881efff914STheodore Ts'o { 11891efff914STheodore Ts'o int ret; 11901efff914STheodore Ts'o 11911efff914STheodore Ts'o ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos); 11921efff914STheodore Ts'o if (ret == 0 && write) 11931efff914STheodore Ts'o mod_delayed_work(system_wq, &dirtytime_work, 0); 11941efff914STheodore Ts'o return ret; 11951efff914STheodore Ts'o } 11961efff914STheodore Ts'o 119703ba3782SJens Axboe static noinline void block_dump___mark_inode_dirty(struct inode *inode) 119803ba3782SJens Axboe { 119903ba3782SJens Axboe if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { 120003ba3782SJens Axboe struct dentry *dentry; 120103ba3782SJens Axboe const char *name = "?"; 120203ba3782SJens Axboe 120303ba3782SJens Axboe dentry = d_find_alias(inode); 120403ba3782SJens Axboe if (dentry) { 120503ba3782SJens Axboe spin_lock(&dentry->d_lock); 120603ba3782SJens Axboe name = (const char *) dentry->d_name.name; 120703ba3782SJens Axboe } 120803ba3782SJens Axboe printk(KERN_DEBUG 120903ba3782SJens Axboe "%s(%d): dirtied inode %lu (%s) on %s\n", 121003ba3782SJens Axboe current->comm, task_pid_nr(current), inode->i_ino, 121103ba3782SJens Axboe name, inode->i_sb->s_id); 121203ba3782SJens Axboe if (dentry) { 121303ba3782SJens Axboe spin_unlock(&dentry->d_lock); 121403ba3782SJens Axboe dput(dentry); 121503ba3782SJens Axboe } 121603ba3782SJens Axboe } 121703ba3782SJens Axboe } 121803ba3782SJens Axboe 121903ba3782SJens Axboe /** 122003ba3782SJens Axboe * __mark_inode_dirty - internal function 122103ba3782SJens Axboe * @inode: inode to mark 122203ba3782SJens Axboe * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) 122303ba3782SJens Axboe * Mark an inode as dirty. Callers should use mark_inode_dirty or 122403ba3782SJens Axboe * mark_inode_dirty_sync. 122503ba3782SJens Axboe * 122603ba3782SJens Axboe * Put the inode on the super block's dirty list. 122703ba3782SJens Axboe * 122803ba3782SJens Axboe * CAREFUL! We mark it dirty unconditionally, but move it onto the 122903ba3782SJens Axboe * dirty list only if it is hashed or if it refers to a blockdev. 123003ba3782SJens Axboe * If it was not hashed, it will never be added to the dirty list 123103ba3782SJens Axboe * even if it is later hashed, as it will have been marked dirty already. 123203ba3782SJens Axboe * 123303ba3782SJens Axboe * In short, make sure you hash any inodes _before_ you start marking 123403ba3782SJens Axboe * them dirty. 123503ba3782SJens Axboe * 123603ba3782SJens Axboe * Note that for blockdevs, inode->dirtied_when represents the dirtying time of 123703ba3782SJens Axboe * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of 123803ba3782SJens Axboe * the kernel-internal blockdev inode represents the dirtying time of the 123903ba3782SJens Axboe * blockdev's pages. This is why for I_DIRTY_PAGES we always use 124003ba3782SJens Axboe * page->mapping->host, so the page-dirtying time is recorded in the internal 124103ba3782SJens Axboe * blockdev inode. 124203ba3782SJens Axboe */ 12430ae45f63STheodore Ts'o #define I_DIRTY_INODE (I_DIRTY_SYNC | I_DIRTY_DATASYNC) 124403ba3782SJens Axboe void __mark_inode_dirty(struct inode *inode, int flags) 124503ba3782SJens Axboe { 124603ba3782SJens Axboe struct super_block *sb = inode->i_sb; 1247253c34e9SArtem Bityutskiy struct backing_dev_info *bdi = NULL; 12480ae45f63STheodore Ts'o int dirtytime; 12490ae45f63STheodore Ts'o 12500ae45f63STheodore Ts'o trace_writeback_mark_inode_dirty(inode, flags); 125103ba3782SJens Axboe 125203ba3782SJens Axboe /* 125303ba3782SJens Axboe * Don't do this for I_DIRTY_PAGES - that doesn't actually 125403ba3782SJens Axboe * dirty the inode itself 125503ba3782SJens Axboe */ 12560ae45f63STheodore Ts'o if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC | I_DIRTY_TIME)) { 12579fb0a7daSTejun Heo trace_writeback_dirty_inode_start(inode, flags); 12589fb0a7daSTejun Heo 125903ba3782SJens Axboe if (sb->s_op->dirty_inode) 1260aa385729SChristoph Hellwig sb->s_op->dirty_inode(inode, flags); 12619fb0a7daSTejun Heo 12629fb0a7daSTejun Heo trace_writeback_dirty_inode(inode, flags); 126303ba3782SJens Axboe } 12640ae45f63STheodore Ts'o if (flags & I_DIRTY_INODE) 12650ae45f63STheodore Ts'o flags &= ~I_DIRTY_TIME; 12660ae45f63STheodore Ts'o dirtytime = flags & I_DIRTY_TIME; 126703ba3782SJens Axboe 126803ba3782SJens Axboe /* 12699c6ac78eSTejun Heo * Paired with smp_mb() in __writeback_single_inode() for the 12709c6ac78eSTejun Heo * following lockless i_state test. See there for details. 127103ba3782SJens Axboe */ 127203ba3782SJens Axboe smp_mb(); 127303ba3782SJens Axboe 12740ae45f63STheodore Ts'o if (((inode->i_state & flags) == flags) || 12750ae45f63STheodore Ts'o (dirtytime && (inode->i_state & I_DIRTY_INODE))) 127603ba3782SJens Axboe return; 127703ba3782SJens Axboe 127803ba3782SJens Axboe if (unlikely(block_dump)) 127903ba3782SJens Axboe block_dump___mark_inode_dirty(inode); 128003ba3782SJens Axboe 1281250df6edSDave Chinner spin_lock(&inode->i_lock); 12820ae45f63STheodore Ts'o if (dirtytime && (inode->i_state & I_DIRTY_INODE)) 12830ae45f63STheodore Ts'o goto out_unlock_inode; 128403ba3782SJens Axboe if ((inode->i_state & flags) != flags) { 128503ba3782SJens Axboe const int was_dirty = inode->i_state & I_DIRTY; 128603ba3782SJens Axboe 12870ae45f63STheodore Ts'o if (flags & I_DIRTY_INODE) 12880ae45f63STheodore Ts'o inode->i_state &= ~I_DIRTY_TIME; 128903ba3782SJens Axboe inode->i_state |= flags; 129003ba3782SJens Axboe 129103ba3782SJens Axboe /* 129203ba3782SJens Axboe * If the inode is being synced, just update its dirty state. 129303ba3782SJens Axboe * The unlocker will place the inode on the appropriate 129403ba3782SJens Axboe * superblock list, based upon its state. 129503ba3782SJens Axboe */ 129603ba3782SJens Axboe if (inode->i_state & I_SYNC) 1297250df6edSDave Chinner goto out_unlock_inode; 129803ba3782SJens Axboe 129903ba3782SJens Axboe /* 130003ba3782SJens Axboe * Only add valid (hashed) inodes to the superblock's 130103ba3782SJens Axboe * dirty list. Add blockdev inodes as well. 130203ba3782SJens Axboe */ 130303ba3782SJens Axboe if (!S_ISBLK(inode->i_mode)) { 13041d3382cbSAl Viro if (inode_unhashed(inode)) 1305250df6edSDave Chinner goto out_unlock_inode; 130603ba3782SJens Axboe } 1307a4ffdde6SAl Viro if (inode->i_state & I_FREEING) 1308250df6edSDave Chinner goto out_unlock_inode; 130903ba3782SJens Axboe 131003ba3782SJens Axboe /* 131103ba3782SJens Axboe * If the inode was already on b_dirty/b_io/b_more_io, don't 131203ba3782SJens Axboe * reposition it (that would break b_dirty time-ordering). 131303ba3782SJens Axboe */ 131403ba3782SJens Axboe if (!was_dirty) { 1315a66979abSDave Chinner bool wakeup_bdi = false; 1316253c34e9SArtem Bityutskiy bdi = inode_to_bdi(inode); 1317500b067cSJens Axboe 1318146d7009SJunxiao Bi spin_unlock(&inode->i_lock); 1319146d7009SJunxiao Bi spin_lock(&bdi->wb.list_lock); 1320253c34e9SArtem Bityutskiy if (bdi_cap_writeback_dirty(bdi)) { 13214452226eSTejun Heo WARN(!test_bit(WB_registered, &bdi->wb.state), 1322253c34e9SArtem Bityutskiy "bdi-%s not registered\n", bdi->name); 1323253c34e9SArtem Bityutskiy 1324253c34e9SArtem Bityutskiy /* 1325253c34e9SArtem Bityutskiy * If this is the first dirty inode for this 1326253c34e9SArtem Bityutskiy * bdi, we have to wake-up the corresponding 1327253c34e9SArtem Bityutskiy * bdi thread to make sure background 1328253c34e9SArtem Bityutskiy * write-back happens later. 1329253c34e9SArtem Bityutskiy */ 1330253c34e9SArtem Bityutskiy if (!wb_has_dirty_io(&bdi->wb)) 1331253c34e9SArtem Bityutskiy wakeup_bdi = true; 1332500b067cSJens Axboe } 133303ba3782SJens Axboe 133403ba3782SJens Axboe inode->dirtied_when = jiffies; 1335a2f48706STheodore Ts'o if (dirtytime) 1336a2f48706STheodore Ts'o inode->dirtied_time_when = jiffies; 1337a2f48706STheodore Ts'o if (inode->i_state & (I_DIRTY_INODE | I_DIRTY_PAGES)) 1338a2f48706STheodore Ts'o list_move(&inode->i_wb_list, &bdi->wb.b_dirty); 1339a2f48706STheodore Ts'o else 1340a2f48706STheodore Ts'o list_move(&inode->i_wb_list, 1341a2f48706STheodore Ts'o &bdi->wb.b_dirty_time); 1342f758eeabSChristoph Hellwig spin_unlock(&bdi->wb.list_lock); 13430ae45f63STheodore Ts'o trace_writeback_dirty_inode_enqueue(inode); 1344253c34e9SArtem Bityutskiy 1345253c34e9SArtem Bityutskiy if (wakeup_bdi) 1346f0054bb1STejun Heo wb_wakeup_delayed(&bdi->wb); 1347a66979abSDave Chinner return; 1348a66979abSDave Chinner } 1349a66979abSDave Chinner } 1350a66979abSDave Chinner out_unlock_inode: 1351a66979abSDave Chinner spin_unlock(&inode->i_lock); 1352a66979abSDave Chinner 135303ba3782SJens Axboe } 135403ba3782SJens Axboe EXPORT_SYMBOL(__mark_inode_dirty); 135503ba3782SJens Axboe 1356b6e51316SJens Axboe static void wait_sb_inodes(struct super_block *sb) 135766f3b8e2SJens Axboe { 135838f21977SNick Piggin struct inode *inode, *old_inode = NULL; 135938f21977SNick Piggin 136003ba3782SJens Axboe /* 136103ba3782SJens Axboe * We need to be protected against the filesystem going from 136203ba3782SJens Axboe * r/o to r/w or vice versa. 136303ba3782SJens Axboe */ 1364b6e51316SJens Axboe WARN_ON(!rwsem_is_locked(&sb->s_umount)); 136503ba3782SJens Axboe 136655fa6091SDave Chinner spin_lock(&inode_sb_list_lock); 136766f3b8e2SJens Axboe 136838f21977SNick Piggin /* 136938f21977SNick Piggin * Data integrity sync. Must wait for all pages under writeback, 137038f21977SNick Piggin * because there may have been pages dirtied before our sync 137138f21977SNick Piggin * call, but which had writeout started before we write it out. 137238f21977SNick Piggin * In which case, the inode may not be on the dirty list, but 137338f21977SNick Piggin * we still have to wait for that writeout. 137438f21977SNick Piggin */ 1375b6e51316SJens Axboe list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { 1376250df6edSDave Chinner struct address_space *mapping = inode->i_mapping; 137738f21977SNick Piggin 1378250df6edSDave Chinner spin_lock(&inode->i_lock); 1379250df6edSDave Chinner if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) || 1380250df6edSDave Chinner (mapping->nrpages == 0)) { 1381250df6edSDave Chinner spin_unlock(&inode->i_lock); 138238f21977SNick Piggin continue; 1383250df6edSDave Chinner } 138438f21977SNick Piggin __iget(inode); 1385250df6edSDave Chinner spin_unlock(&inode->i_lock); 138655fa6091SDave Chinner spin_unlock(&inode_sb_list_lock); 138755fa6091SDave Chinner 138838f21977SNick Piggin /* 138955fa6091SDave Chinner * We hold a reference to 'inode' so it couldn't have been 139055fa6091SDave Chinner * removed from s_inodes list while we dropped the 139155fa6091SDave Chinner * inode_sb_list_lock. We cannot iput the inode now as we can 139255fa6091SDave Chinner * be holding the last reference and we cannot iput it under 139355fa6091SDave Chinner * inode_sb_list_lock. So we keep the reference and iput it 139455fa6091SDave Chinner * later. 139538f21977SNick Piggin */ 139638f21977SNick Piggin iput(old_inode); 139738f21977SNick Piggin old_inode = inode; 139838f21977SNick Piggin 139938f21977SNick Piggin filemap_fdatawait(mapping); 140038f21977SNick Piggin 140138f21977SNick Piggin cond_resched(); 140238f21977SNick Piggin 140355fa6091SDave Chinner spin_lock(&inode_sb_list_lock); 140438f21977SNick Piggin } 140555fa6091SDave Chinner spin_unlock(&inode_sb_list_lock); 140638f21977SNick Piggin iput(old_inode); 140766f3b8e2SJens Axboe } 14081da177e4SLinus Torvalds 1409d8a8559cSJens Axboe /** 14103259f8beSChris Mason * writeback_inodes_sb_nr - writeback dirty inodes from given super_block 1411d8a8559cSJens Axboe * @sb: the superblock 14123259f8beSChris Mason * @nr: the number of pages to write 1413786228abSMarcos Paulo de Souza * @reason: reason why some writeback work initiated 14141da177e4SLinus Torvalds * 1415d8a8559cSJens Axboe * Start writeback on some inodes on this super_block. No guarantees are made 1416d8a8559cSJens Axboe * on how many (if any) will be written, and this function does not wait 14173259f8beSChris Mason * for IO completion of submitted IO. 14181da177e4SLinus Torvalds */ 14190e175a18SCurt Wohlgemuth void writeback_inodes_sb_nr(struct super_block *sb, 14200e175a18SCurt Wohlgemuth unsigned long nr, 14210e175a18SCurt Wohlgemuth enum wb_reason reason) 14221da177e4SLinus Torvalds { 142383ba7b07SChristoph Hellwig DECLARE_COMPLETION_ONSTACK(done); 142483ba7b07SChristoph Hellwig struct wb_writeback_work work = { 14253c4d7165SChristoph Hellwig .sb = sb, 14263c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_NONE, 14276e6938b6SWu Fengguang .tagged_writepages = 1, 142883ba7b07SChristoph Hellwig .done = &done, 14293259f8beSChris Mason .nr_pages = nr, 14300e175a18SCurt Wohlgemuth .reason = reason, 14313c4d7165SChristoph Hellwig }; 14320e3c9a22SJens Axboe 14336eedc701SJan Kara if (sb->s_bdi == &noop_backing_dev_info) 14346eedc701SJan Kara return; 1435cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 1436f0054bb1STejun Heo wb_queue_work(&sb->s_bdi->wb, &work); 143783ba7b07SChristoph Hellwig wait_for_completion(&done); 14381da177e4SLinus Torvalds } 14393259f8beSChris Mason EXPORT_SYMBOL(writeback_inodes_sb_nr); 14403259f8beSChris Mason 14413259f8beSChris Mason /** 14423259f8beSChris Mason * writeback_inodes_sb - writeback dirty inodes from given super_block 14433259f8beSChris Mason * @sb: the superblock 1444786228abSMarcos Paulo de Souza * @reason: reason why some writeback work was initiated 14453259f8beSChris Mason * 14463259f8beSChris Mason * Start writeback on some inodes on this super_block. No guarantees are made 14473259f8beSChris Mason * on how many (if any) will be written, and this function does not wait 14483259f8beSChris Mason * for IO completion of submitted IO. 14493259f8beSChris Mason */ 14500e175a18SCurt Wohlgemuth void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) 14513259f8beSChris Mason { 14520e175a18SCurt Wohlgemuth return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); 14533259f8beSChris Mason } 1454d8a8559cSJens Axboe EXPORT_SYMBOL(writeback_inodes_sb); 1455d8a8559cSJens Axboe 1456d8a8559cSJens Axboe /** 145710ee27a0SMiao Xie * try_to_writeback_inodes_sb_nr - try to start writeback if none underway 14583259f8beSChris Mason * @sb: the superblock 14593259f8beSChris Mason * @nr: the number of pages to write 146010ee27a0SMiao Xie * @reason: the reason of writeback 14613259f8beSChris Mason * 146210ee27a0SMiao Xie * Invoke writeback_inodes_sb_nr if no writeback is currently underway. 14633259f8beSChris Mason * Returns 1 if writeback was started, 0 if not. 14643259f8beSChris Mason */ 146510ee27a0SMiao Xie int try_to_writeback_inodes_sb_nr(struct super_block *sb, 14660e175a18SCurt Wohlgemuth unsigned long nr, 14670e175a18SCurt Wohlgemuth enum wb_reason reason) 14683259f8beSChris Mason { 146910ee27a0SMiao Xie if (writeback_in_progress(sb->s_bdi)) 147010ee27a0SMiao Xie return 1; 147110ee27a0SMiao Xie 147210ee27a0SMiao Xie if (!down_read_trylock(&sb->s_umount)) 147310ee27a0SMiao Xie return 0; 147410ee27a0SMiao Xie 14750e175a18SCurt Wohlgemuth writeback_inodes_sb_nr(sb, nr, reason); 14763259f8beSChris Mason up_read(&sb->s_umount); 14773259f8beSChris Mason return 1; 14783259f8beSChris Mason } 147910ee27a0SMiao Xie EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr); 148010ee27a0SMiao Xie 148110ee27a0SMiao Xie /** 148210ee27a0SMiao Xie * try_to_writeback_inodes_sb - try to start writeback if none underway 148310ee27a0SMiao Xie * @sb: the superblock 148410ee27a0SMiao Xie * @reason: reason why some writeback work was initiated 148510ee27a0SMiao Xie * 148610ee27a0SMiao Xie * Implement by try_to_writeback_inodes_sb_nr() 148710ee27a0SMiao Xie * Returns 1 if writeback was started, 0 if not. 148810ee27a0SMiao Xie */ 148910ee27a0SMiao Xie int try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) 149010ee27a0SMiao Xie { 149110ee27a0SMiao Xie return try_to_writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); 149210ee27a0SMiao Xie } 149310ee27a0SMiao Xie EXPORT_SYMBOL(try_to_writeback_inodes_sb); 14943259f8beSChris Mason 14953259f8beSChris Mason /** 1496d8a8559cSJens Axboe * sync_inodes_sb - sync sb inode pages 1497d8a8559cSJens Axboe * @sb: the superblock 1498d8a8559cSJens Axboe * 1499d8a8559cSJens Axboe * This function writes and waits on any dirty inode belonging to this 15000dc83bd3SJan Kara * super_block. 1501d8a8559cSJens Axboe */ 15020dc83bd3SJan Kara void sync_inodes_sb(struct super_block *sb) 1503d8a8559cSJens Axboe { 150483ba7b07SChristoph Hellwig DECLARE_COMPLETION_ONSTACK(done); 150583ba7b07SChristoph Hellwig struct wb_writeback_work work = { 15063c4d7165SChristoph Hellwig .sb = sb, 15073c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_ALL, 15083c4d7165SChristoph Hellwig .nr_pages = LONG_MAX, 15093c4d7165SChristoph Hellwig .range_cyclic = 0, 151083ba7b07SChristoph Hellwig .done = &done, 15110e175a18SCurt Wohlgemuth .reason = WB_REASON_SYNC, 15127747bd4bSDave Chinner .for_sync = 1, 15133c4d7165SChristoph Hellwig }; 15143c4d7165SChristoph Hellwig 15156eedc701SJan Kara /* Nothing to do? */ 15166eedc701SJan Kara if (sb->s_bdi == &noop_backing_dev_info) 15176eedc701SJan Kara return; 1518cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 1519cf37e972SChristoph Hellwig 1520f0054bb1STejun Heo wb_queue_work(&sb->s_bdi->wb, &work); 152183ba7b07SChristoph Hellwig wait_for_completion(&done); 152283ba7b07SChristoph Hellwig 1523b6e51316SJens Axboe wait_sb_inodes(sb); 1524d8a8559cSJens Axboe } 1525d8a8559cSJens Axboe EXPORT_SYMBOL(sync_inodes_sb); 15261da177e4SLinus Torvalds 15271da177e4SLinus Torvalds /** 15281da177e4SLinus Torvalds * write_inode_now - write an inode to disk 15291da177e4SLinus Torvalds * @inode: inode to write to disk 15301da177e4SLinus Torvalds * @sync: whether the write should be synchronous or not 15311da177e4SLinus Torvalds * 15327f04c26dSAndrea Arcangeli * This function commits an inode to disk immediately if it is dirty. This is 15337f04c26dSAndrea Arcangeli * primarily needed by knfsd. 15347f04c26dSAndrea Arcangeli * 15357f04c26dSAndrea Arcangeli * The caller must either have a ref on the inode or must have set I_WILL_FREE. 15361da177e4SLinus Torvalds */ 15371da177e4SLinus Torvalds int write_inode_now(struct inode *inode, int sync) 15381da177e4SLinus Torvalds { 1539f758eeabSChristoph Hellwig struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; 15401da177e4SLinus Torvalds struct writeback_control wbc = { 15411da177e4SLinus Torvalds .nr_to_write = LONG_MAX, 154218914b18SMike Galbraith .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, 1543111ebb6eSOGAWA Hirofumi .range_start = 0, 1544111ebb6eSOGAWA Hirofumi .range_end = LLONG_MAX, 15451da177e4SLinus Torvalds }; 15461da177e4SLinus Torvalds 15471da177e4SLinus Torvalds if (!mapping_cap_writeback_dirty(inode->i_mapping)) 154849364ce2SAndrew Morton wbc.nr_to_write = 0; 15491da177e4SLinus Torvalds 15501da177e4SLinus Torvalds might_sleep(); 15514f8ad655SJan Kara return writeback_single_inode(inode, wb, &wbc); 15521da177e4SLinus Torvalds } 15531da177e4SLinus Torvalds EXPORT_SYMBOL(write_inode_now); 15541da177e4SLinus Torvalds 15551da177e4SLinus Torvalds /** 15561da177e4SLinus Torvalds * sync_inode - write an inode and its pages to disk. 15571da177e4SLinus Torvalds * @inode: the inode to sync 15581da177e4SLinus Torvalds * @wbc: controls the writeback mode 15591da177e4SLinus Torvalds * 15601da177e4SLinus Torvalds * sync_inode() will write an inode and its pages to disk. It will also 15611da177e4SLinus Torvalds * correctly update the inode on its superblock's dirty inode lists and will 15621da177e4SLinus Torvalds * update inode->i_state. 15631da177e4SLinus Torvalds * 15641da177e4SLinus Torvalds * The caller must have a ref on the inode. 15651da177e4SLinus Torvalds */ 15661da177e4SLinus Torvalds int sync_inode(struct inode *inode, struct writeback_control *wbc) 15671da177e4SLinus Torvalds { 15684f8ad655SJan Kara return writeback_single_inode(inode, &inode_to_bdi(inode)->wb, wbc); 15691da177e4SLinus Torvalds } 15701da177e4SLinus Torvalds EXPORT_SYMBOL(sync_inode); 1571c3765016SChristoph Hellwig 1572c3765016SChristoph Hellwig /** 1573c691b9d9SAndrew Morton * sync_inode_metadata - write an inode to disk 1574c3765016SChristoph Hellwig * @inode: the inode to sync 1575c3765016SChristoph Hellwig * @wait: wait for I/O to complete. 1576c3765016SChristoph Hellwig * 1577c691b9d9SAndrew Morton * Write an inode to disk and adjust its dirty state after completion. 1578c3765016SChristoph Hellwig * 1579c3765016SChristoph Hellwig * Note: only writes the actual inode, no associated data or other metadata. 1580c3765016SChristoph Hellwig */ 1581c3765016SChristoph Hellwig int sync_inode_metadata(struct inode *inode, int wait) 1582c3765016SChristoph Hellwig { 1583c3765016SChristoph Hellwig struct writeback_control wbc = { 1584c3765016SChristoph Hellwig .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE, 1585c3765016SChristoph Hellwig .nr_to_write = 0, /* metadata-only */ 1586c3765016SChristoph Hellwig }; 1587c3765016SChristoph Hellwig 1588c3765016SChristoph Hellwig return sync_inode(inode, &wbc); 1589c3765016SChristoph Hellwig } 1590c3765016SChristoph Hellwig EXPORT_SYMBOL(sync_inode_metadata); 1591