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 817ccf19a8SNick Piggin static inline struct inode *wb_inode(struct list_head *head) 827ccf19a8SNick Piggin { 837ccf19a8SNick Piggin return list_entry(head, struct inode, i_wb_list); 847ccf19a8SNick Piggin } 857ccf19a8SNick Piggin 8615eb77a0SWu Fengguang /* 8715eb77a0SWu Fengguang * Include the creation of the trace points after defining the 8815eb77a0SWu Fengguang * wb_writeback_work structure and inline functions so that the definition 8915eb77a0SWu Fengguang * remains local to this file. 9015eb77a0SWu Fengguang */ 9115eb77a0SWu Fengguang #define CREATE_TRACE_POINTS 9215eb77a0SWu Fengguang #include <trace/events/writeback.h> 9315eb77a0SWu Fengguang 94774016b2SSteven Whitehouse EXPORT_TRACEPOINT_SYMBOL_GPL(wbc_writepage); 95774016b2SSteven Whitehouse 96f0054bb1STejun Heo static void wb_wakeup(struct bdi_writeback *wb) 975acda9d1SJan Kara { 98f0054bb1STejun Heo spin_lock_bh(&wb->work_lock); 99f0054bb1STejun Heo if (test_bit(WB_registered, &wb->state)) 100f0054bb1STejun Heo mod_delayed_work(bdi_wq, &wb->dwork, 0); 101f0054bb1STejun Heo spin_unlock_bh(&wb->work_lock); 1025acda9d1SJan Kara } 1035acda9d1SJan Kara 104f0054bb1STejun Heo static void wb_queue_work(struct bdi_writeback *wb, 1056585027aSJan Kara struct wb_writeback_work *work) 1066585027aSJan Kara { 107f0054bb1STejun Heo trace_writeback_queue(wb->bdi, work); 1086585027aSJan Kara 109f0054bb1STejun Heo spin_lock_bh(&wb->work_lock); 110f0054bb1STejun Heo if (!test_bit(WB_registered, &wb->state)) { 1115acda9d1SJan Kara if (work->done) 1125acda9d1SJan Kara complete(work->done); 1135acda9d1SJan Kara goto out_unlock; 1145acda9d1SJan Kara } 115f0054bb1STejun Heo list_add_tail(&work->list, &wb->work_list); 116f0054bb1STejun Heo mod_delayed_work(bdi_wq, &wb->dwork, 0); 1175acda9d1SJan Kara out_unlock: 118f0054bb1STejun Heo spin_unlock_bh(&wb->work_lock); 11903ba3782SJens Axboe } 1201da177e4SLinus Torvalds 121f0054bb1STejun Heo static void __wb_start_writeback(struct bdi_writeback *wb, long nr_pages, 1220e175a18SCurt Wohlgemuth bool range_cyclic, enum wb_reason reason) 1231da177e4SLinus Torvalds { 12483ba7b07SChristoph Hellwig struct wb_writeback_work *work; 12503ba3782SJens Axboe 126bcddc3f0SJens Axboe /* 127bcddc3f0SJens Axboe * This is WB_SYNC_NONE writeback, so if allocation fails just 128bcddc3f0SJens Axboe * wakeup the thread for old dirty data writeback 129bcddc3f0SJens Axboe */ 13083ba7b07SChristoph Hellwig work = kzalloc(sizeof(*work), GFP_ATOMIC); 13183ba7b07SChristoph Hellwig if (!work) { 132f0054bb1STejun Heo trace_writeback_nowork(wb->bdi); 133f0054bb1STejun Heo wb_wakeup(wb); 13483ba7b07SChristoph Hellwig return; 13583ba7b07SChristoph Hellwig } 13683ba7b07SChristoph Hellwig 13783ba7b07SChristoph Hellwig work->sync_mode = WB_SYNC_NONE; 13883ba7b07SChristoph Hellwig work->nr_pages = nr_pages; 13983ba7b07SChristoph Hellwig work->range_cyclic = range_cyclic; 1400e175a18SCurt Wohlgemuth work->reason = reason; 14183ba7b07SChristoph Hellwig 142f0054bb1STejun Heo wb_queue_work(wb, work); 14303ba3782SJens Axboe } 144b6e51316SJens Axboe 145b6e51316SJens Axboe /** 146b6e51316SJens Axboe * bdi_start_writeback - start writeback 147b6e51316SJens Axboe * @bdi: the backing device to write from 148b6e51316SJens Axboe * @nr_pages: the number of pages to write 149786228abSMarcos Paulo de Souza * @reason: reason why some writeback work was initiated 150b6e51316SJens Axboe * 151b6e51316SJens Axboe * Description: 152b6e51316SJens Axboe * This does WB_SYNC_NONE opportunistic writeback. The IO is only 15325985edcSLucas De Marchi * started when this function returns, we make no guarantees on 1540e3c9a22SJens Axboe * completion. Caller need not hold sb s_umount semaphore. 155b6e51316SJens Axboe * 156b6e51316SJens Axboe */ 1570e175a18SCurt Wohlgemuth void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, 1580e175a18SCurt Wohlgemuth enum wb_reason reason) 159b6e51316SJens Axboe { 160f0054bb1STejun Heo __wb_start_writeback(&bdi->wb, nr_pages, true, reason); 161d3ddec76SWu Fengguang } 162d3ddec76SWu Fengguang 163c5444198SChristoph Hellwig /** 164c5444198SChristoph Hellwig * bdi_start_background_writeback - start background writeback 165c5444198SChristoph Hellwig * @bdi: the backing device to write from 166c5444198SChristoph Hellwig * 167c5444198SChristoph Hellwig * Description: 1686585027aSJan Kara * This makes sure WB_SYNC_NONE background writeback happens. When 1696585027aSJan Kara * this function returns, it is only guaranteed that for given BDI 1706585027aSJan Kara * some IO is happening if we are over background dirty threshold. 1716585027aSJan Kara * Caller need not hold sb s_umount semaphore. 172c5444198SChristoph Hellwig */ 173c5444198SChristoph Hellwig void bdi_start_background_writeback(struct backing_dev_info *bdi) 174c5444198SChristoph Hellwig { 1756585027aSJan Kara /* 1766585027aSJan Kara * We just wake up the flusher thread. It will perform background 1776585027aSJan Kara * writeback as soon as there is no other work to do. 1786585027aSJan Kara */ 17971927e84SWu Fengguang trace_writeback_wake_background(bdi); 180f0054bb1STejun Heo wb_wakeup(&bdi->wb); 1811da177e4SLinus Torvalds } 1821da177e4SLinus Torvalds 1831da177e4SLinus Torvalds /* 184a66979abSDave Chinner * Remove the inode from the writeback list it is on. 185a66979abSDave Chinner */ 186a66979abSDave Chinner void inode_wb_list_del(struct inode *inode) 187a66979abSDave Chinner { 18852ebea74STejun Heo struct bdi_writeback *wb = inode_to_wb(inode); 189a66979abSDave Chinner 19052ebea74STejun Heo spin_lock(&wb->list_lock); 191f758eeabSChristoph Hellwig list_del_init(&inode->i_wb_list); 19252ebea74STejun Heo spin_unlock(&wb->list_lock); 193f758eeabSChristoph Hellwig } 194a66979abSDave Chinner 195a66979abSDave Chinner /* 1966610a0bcSAndrew Morton * Redirty an inode: set its when-it-was dirtied timestamp and move it to the 1976610a0bcSAndrew Morton * furthest end of its superblock's dirty-inode list. 1986610a0bcSAndrew Morton * 1996610a0bcSAndrew Morton * Before stamping the inode's ->dirtied_when, we check to see whether it is 20066f3b8e2SJens Axboe * already the most-recently-dirtied inode on the b_dirty list. If that is 2016610a0bcSAndrew Morton * the case then the inode must have been redirtied while it was being written 2026610a0bcSAndrew Morton * out and we don't reset its dirtied_when. 2036610a0bcSAndrew Morton */ 204f758eeabSChristoph Hellwig static void redirty_tail(struct inode *inode, struct bdi_writeback *wb) 2056610a0bcSAndrew Morton { 206f758eeabSChristoph Hellwig assert_spin_locked(&wb->list_lock); 20703ba3782SJens Axboe if (!list_empty(&wb->b_dirty)) { 20866f3b8e2SJens Axboe struct inode *tail; 2096610a0bcSAndrew Morton 2107ccf19a8SNick Piggin tail = wb_inode(wb->b_dirty.next); 21166f3b8e2SJens Axboe if (time_before(inode->dirtied_when, tail->dirtied_when)) 2126610a0bcSAndrew Morton inode->dirtied_when = jiffies; 2136610a0bcSAndrew Morton } 2147ccf19a8SNick Piggin list_move(&inode->i_wb_list, &wb->b_dirty); 2156610a0bcSAndrew Morton } 2166610a0bcSAndrew Morton 2176610a0bcSAndrew Morton /* 21866f3b8e2SJens Axboe * requeue inode for re-scanning after bdi->b_io list is exhausted. 219c986d1e2SAndrew Morton */ 220f758eeabSChristoph Hellwig static void requeue_io(struct inode *inode, struct bdi_writeback *wb) 221c986d1e2SAndrew Morton { 222f758eeabSChristoph Hellwig assert_spin_locked(&wb->list_lock); 2237ccf19a8SNick Piggin list_move(&inode->i_wb_list, &wb->b_more_io); 224c986d1e2SAndrew Morton } 225c986d1e2SAndrew Morton 2261c0eeaf5SJoern Engel static void inode_sync_complete(struct inode *inode) 2271c0eeaf5SJoern Engel { 228365b94aeSJan Kara inode->i_state &= ~I_SYNC; 2294eff96ddSJan Kara /* If inode is clean an unused, put it into LRU now... */ 2304eff96ddSJan Kara inode_add_lru(inode); 231365b94aeSJan Kara /* Waiters must see I_SYNC cleared before being woken up */ 2321c0eeaf5SJoern Engel smp_mb(); 2331c0eeaf5SJoern Engel wake_up_bit(&inode->i_state, __I_SYNC); 2341c0eeaf5SJoern Engel } 2351c0eeaf5SJoern Engel 236d2caa3c5SJeff Layton static bool inode_dirtied_after(struct inode *inode, unsigned long t) 237d2caa3c5SJeff Layton { 238d2caa3c5SJeff Layton bool ret = time_after(inode->dirtied_when, t); 239d2caa3c5SJeff Layton #ifndef CONFIG_64BIT 240d2caa3c5SJeff Layton /* 241d2caa3c5SJeff Layton * For inodes being constantly redirtied, dirtied_when can get stuck. 242d2caa3c5SJeff Layton * It _appears_ to be in the future, but is actually in distant past. 243d2caa3c5SJeff Layton * This test is necessary to prevent such wrapped-around relative times 2445b0830cbSJens Axboe * from permanently stopping the whole bdi writeback. 245d2caa3c5SJeff Layton */ 246d2caa3c5SJeff Layton ret = ret && time_before_eq(inode->dirtied_when, jiffies); 247d2caa3c5SJeff Layton #endif 248d2caa3c5SJeff Layton return ret; 249d2caa3c5SJeff Layton } 250d2caa3c5SJeff Layton 2510ae45f63STheodore Ts'o #define EXPIRE_DIRTY_ATIME 0x0001 2520ae45f63STheodore Ts'o 253c986d1e2SAndrew Morton /* 2540e2f2b23SWang Sheng-Hui * Move expired (dirtied before work->older_than_this) dirty inodes from 255697e6fedSJan Kara * @delaying_queue to @dispatch_queue. 2562c136579SFengguang Wu */ 257e84d0a4fSWu Fengguang static int move_expired_inodes(struct list_head *delaying_queue, 2582c136579SFengguang Wu struct list_head *dispatch_queue, 2590ae45f63STheodore Ts'o int flags, 260ad4e38ddSCurt Wohlgemuth struct wb_writeback_work *work) 2612c136579SFengguang Wu { 2620ae45f63STheodore Ts'o unsigned long *older_than_this = NULL; 2630ae45f63STheodore Ts'o unsigned long expire_time; 2645c03449dSShaohua Li LIST_HEAD(tmp); 2655c03449dSShaohua Li struct list_head *pos, *node; 266cf137307SJens Axboe struct super_block *sb = NULL; 2675c03449dSShaohua Li struct inode *inode; 268cf137307SJens Axboe int do_sb_sort = 0; 269e84d0a4fSWu Fengguang int moved = 0; 2705c03449dSShaohua Li 2710ae45f63STheodore Ts'o if ((flags & EXPIRE_DIRTY_ATIME) == 0) 2720ae45f63STheodore Ts'o older_than_this = work->older_than_this; 273a2f48706STheodore Ts'o else if (!work->for_sync) { 274a2f48706STheodore Ts'o expire_time = jiffies - (dirtytime_expire_interval * HZ); 2750ae45f63STheodore Ts'o older_than_this = &expire_time; 2760ae45f63STheodore Ts'o } 2772c136579SFengguang Wu while (!list_empty(delaying_queue)) { 2787ccf19a8SNick Piggin inode = wb_inode(delaying_queue->prev); 2790ae45f63STheodore Ts'o if (older_than_this && 2800ae45f63STheodore Ts'o inode_dirtied_after(inode, *older_than_this)) 2812c136579SFengguang Wu break; 282a8855990SJan Kara list_move(&inode->i_wb_list, &tmp); 283a8855990SJan Kara moved++; 2840ae45f63STheodore Ts'o if (flags & EXPIRE_DIRTY_ATIME) 2850ae45f63STheodore Ts'o set_bit(__I_DIRTY_TIME_EXPIRED, &inode->i_state); 286a8855990SJan Kara if (sb_is_blkdev_sb(inode->i_sb)) 287a8855990SJan Kara continue; 288cf137307SJens Axboe if (sb && sb != inode->i_sb) 289cf137307SJens Axboe do_sb_sort = 1; 290cf137307SJens Axboe sb = inode->i_sb; 2915c03449dSShaohua Li } 2925c03449dSShaohua Li 293cf137307SJens Axboe /* just one sb in list, splice to dispatch_queue and we're done */ 294cf137307SJens Axboe if (!do_sb_sort) { 295cf137307SJens Axboe list_splice(&tmp, dispatch_queue); 296e84d0a4fSWu Fengguang goto out; 297cf137307SJens Axboe } 298cf137307SJens Axboe 2995c03449dSShaohua Li /* Move inodes from one superblock together */ 3005c03449dSShaohua Li while (!list_empty(&tmp)) { 3017ccf19a8SNick Piggin sb = wb_inode(tmp.prev)->i_sb; 3025c03449dSShaohua Li list_for_each_prev_safe(pos, node, &tmp) { 3037ccf19a8SNick Piggin inode = wb_inode(pos); 3045c03449dSShaohua Li if (inode->i_sb == sb) 3057ccf19a8SNick Piggin list_move(&inode->i_wb_list, dispatch_queue); 3062c136579SFengguang Wu } 3072c136579SFengguang Wu } 308e84d0a4fSWu Fengguang out: 309e84d0a4fSWu Fengguang return moved; 3105c03449dSShaohua Li } 3112c136579SFengguang Wu 3122c136579SFengguang Wu /* 3132c136579SFengguang Wu * Queue all expired dirty inodes for io, eldest first. 3144ea879b9SWu Fengguang * Before 3154ea879b9SWu Fengguang * newly dirtied b_dirty b_io b_more_io 3164ea879b9SWu Fengguang * =============> gf edc BA 3174ea879b9SWu Fengguang * After 3184ea879b9SWu Fengguang * newly dirtied b_dirty b_io b_more_io 3194ea879b9SWu Fengguang * =============> g fBAedc 3204ea879b9SWu Fengguang * | 3214ea879b9SWu Fengguang * +--> dequeue for IO 3222c136579SFengguang Wu */ 323ad4e38ddSCurt Wohlgemuth static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work) 3242c136579SFengguang Wu { 325e84d0a4fSWu Fengguang int moved; 3260ae45f63STheodore Ts'o 327f758eeabSChristoph Hellwig assert_spin_locked(&wb->list_lock); 3284ea879b9SWu Fengguang list_splice_init(&wb->b_more_io, &wb->b_io); 3290ae45f63STheodore Ts'o moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, 0, work); 3300ae45f63STheodore Ts'o moved += move_expired_inodes(&wb->b_dirty_time, &wb->b_io, 3310ae45f63STheodore Ts'o EXPIRE_DIRTY_ATIME, work); 332ad4e38ddSCurt Wohlgemuth trace_writeback_queue_io(wb, work, moved); 33366f3b8e2SJens Axboe } 33466f3b8e2SJens Axboe 335a9185b41SChristoph Hellwig static int write_inode(struct inode *inode, struct writeback_control *wbc) 33666f3b8e2SJens Axboe { 3379fb0a7daSTejun Heo int ret; 3389fb0a7daSTejun Heo 3399fb0a7daSTejun Heo if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) { 3409fb0a7daSTejun Heo trace_writeback_write_inode_start(inode, wbc); 3419fb0a7daSTejun Heo ret = inode->i_sb->s_op->write_inode(inode, wbc); 3429fb0a7daSTejun Heo trace_writeback_write_inode(inode, wbc); 3439fb0a7daSTejun Heo return ret; 3449fb0a7daSTejun Heo } 34503ba3782SJens Axboe return 0; 34666f3b8e2SJens Axboe } 34708d8e974SFengguang Wu 3482c136579SFengguang Wu /* 349169ebd90SJan Kara * Wait for writeback on an inode to complete. Called with i_lock held. 350169ebd90SJan Kara * Caller must make sure inode cannot go away when we drop i_lock. 35101c03194SChristoph Hellwig */ 352169ebd90SJan Kara static void __inode_wait_for_writeback(struct inode *inode) 353169ebd90SJan Kara __releases(inode->i_lock) 354169ebd90SJan Kara __acquires(inode->i_lock) 35501c03194SChristoph Hellwig { 35601c03194SChristoph Hellwig DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); 35701c03194SChristoph Hellwig wait_queue_head_t *wqh; 35801c03194SChristoph Hellwig 35901c03194SChristoph Hellwig wqh = bit_waitqueue(&inode->i_state, __I_SYNC); 36058a9d3d8SRichard Kennedy while (inode->i_state & I_SYNC) { 361250df6edSDave Chinner spin_unlock(&inode->i_lock); 36274316201SNeilBrown __wait_on_bit(wqh, &wq, bit_wait, 36374316201SNeilBrown TASK_UNINTERRUPTIBLE); 364250df6edSDave Chinner spin_lock(&inode->i_lock); 36558a9d3d8SRichard Kennedy } 36601c03194SChristoph Hellwig } 36701c03194SChristoph Hellwig 36801c03194SChristoph Hellwig /* 369169ebd90SJan Kara * Wait for writeback on an inode to complete. Caller must have inode pinned. 370169ebd90SJan Kara */ 371169ebd90SJan Kara void inode_wait_for_writeback(struct inode *inode) 372169ebd90SJan Kara { 373169ebd90SJan Kara spin_lock(&inode->i_lock); 374169ebd90SJan Kara __inode_wait_for_writeback(inode); 375169ebd90SJan Kara spin_unlock(&inode->i_lock); 376169ebd90SJan Kara } 377169ebd90SJan Kara 378169ebd90SJan Kara /* 379169ebd90SJan Kara * Sleep until I_SYNC is cleared. This function must be called with i_lock 380169ebd90SJan Kara * held and drops it. It is aimed for callers not holding any inode reference 381169ebd90SJan Kara * so once i_lock is dropped, inode can go away. 382169ebd90SJan Kara */ 383169ebd90SJan Kara static void inode_sleep_on_writeback(struct inode *inode) 384169ebd90SJan Kara __releases(inode->i_lock) 385169ebd90SJan Kara { 386169ebd90SJan Kara DEFINE_WAIT(wait); 387169ebd90SJan Kara wait_queue_head_t *wqh = bit_waitqueue(&inode->i_state, __I_SYNC); 388169ebd90SJan Kara int sleep; 389169ebd90SJan Kara 390169ebd90SJan Kara prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); 391169ebd90SJan Kara sleep = inode->i_state & I_SYNC; 392169ebd90SJan Kara spin_unlock(&inode->i_lock); 393169ebd90SJan Kara if (sleep) 394169ebd90SJan Kara schedule(); 395169ebd90SJan Kara finish_wait(wqh, &wait); 396169ebd90SJan Kara } 397169ebd90SJan Kara 398169ebd90SJan Kara /* 399ccb26b5aSJan Kara * Find proper writeback list for the inode depending on its current state and 400ccb26b5aSJan Kara * possibly also change of its state while we were doing writeback. Here we 401ccb26b5aSJan Kara * handle things such as livelock prevention or fairness of writeback among 402ccb26b5aSJan Kara * inodes. This function can be called only by flusher thread - noone else 403ccb26b5aSJan Kara * processes all inodes in writeback lists and requeueing inodes behind flusher 404ccb26b5aSJan Kara * thread's back can have unexpected consequences. 405ccb26b5aSJan Kara */ 406ccb26b5aSJan Kara static void requeue_inode(struct inode *inode, struct bdi_writeback *wb, 407ccb26b5aSJan Kara struct writeback_control *wbc) 408ccb26b5aSJan Kara { 409ccb26b5aSJan Kara if (inode->i_state & I_FREEING) 410ccb26b5aSJan Kara return; 411ccb26b5aSJan Kara 412ccb26b5aSJan Kara /* 413ccb26b5aSJan Kara * Sync livelock prevention. Each inode is tagged and synced in one 414ccb26b5aSJan Kara * shot. If still dirty, it will be redirty_tail()'ed below. Update 415ccb26b5aSJan Kara * the dirty time to prevent enqueue and sync it again. 416ccb26b5aSJan Kara */ 417ccb26b5aSJan Kara if ((inode->i_state & I_DIRTY) && 418ccb26b5aSJan Kara (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)) 419ccb26b5aSJan Kara inode->dirtied_when = jiffies; 420ccb26b5aSJan Kara 4214f8ad655SJan Kara if (wbc->pages_skipped) { 4224f8ad655SJan Kara /* 4234f8ad655SJan Kara * writeback is not making progress due to locked 4244f8ad655SJan Kara * buffers. Skip this inode for now. 4254f8ad655SJan Kara */ 4264f8ad655SJan Kara redirty_tail(inode, wb); 4274f8ad655SJan Kara return; 4284f8ad655SJan Kara } 4294f8ad655SJan Kara 430ccb26b5aSJan Kara if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) { 431ccb26b5aSJan Kara /* 432ccb26b5aSJan Kara * We didn't write back all the pages. nfs_writepages() 433ccb26b5aSJan Kara * sometimes bales out without doing anything. 434ccb26b5aSJan Kara */ 435ccb26b5aSJan Kara if (wbc->nr_to_write <= 0) { 436ccb26b5aSJan Kara /* Slice used up. Queue for next turn. */ 437ccb26b5aSJan Kara requeue_io(inode, wb); 438ccb26b5aSJan Kara } else { 439ccb26b5aSJan Kara /* 440ccb26b5aSJan Kara * Writeback blocked by something other than 441ccb26b5aSJan Kara * congestion. Delay the inode for some time to 442ccb26b5aSJan Kara * avoid spinning on the CPU (100% iowait) 443ccb26b5aSJan Kara * retrying writeback of the dirty page/inode 444ccb26b5aSJan Kara * that cannot be performed immediately. 445ccb26b5aSJan Kara */ 446ccb26b5aSJan Kara redirty_tail(inode, wb); 447ccb26b5aSJan Kara } 448ccb26b5aSJan Kara } else if (inode->i_state & I_DIRTY) { 449ccb26b5aSJan Kara /* 450ccb26b5aSJan Kara * Filesystems can dirty the inode during writeback operations, 451ccb26b5aSJan Kara * such as delayed allocation during submission or metadata 452ccb26b5aSJan Kara * updates after data IO completion. 453ccb26b5aSJan Kara */ 454ccb26b5aSJan Kara redirty_tail(inode, wb); 4550ae45f63STheodore Ts'o } else if (inode->i_state & I_DIRTY_TIME) { 456a2f48706STheodore Ts'o inode->dirtied_when = jiffies; 4570ae45f63STheodore Ts'o list_move(&inode->i_wb_list, &wb->b_dirty_time); 458ccb26b5aSJan Kara } else { 459ccb26b5aSJan Kara /* The inode is clean. Remove from writeback lists. */ 460ccb26b5aSJan Kara list_del_init(&inode->i_wb_list); 461ccb26b5aSJan Kara } 462ccb26b5aSJan Kara } 463ccb26b5aSJan Kara 464ccb26b5aSJan Kara /* 4654f8ad655SJan Kara * Write out an inode and its dirty pages. Do not update the writeback list 4664f8ad655SJan Kara * linkage. That is left to the caller. The caller is also responsible for 4674f8ad655SJan Kara * setting I_SYNC flag and calling inode_sync_complete() to clear it. 4681da177e4SLinus Torvalds */ 4691da177e4SLinus Torvalds static int 470cd8ed2a4SYan Hong __writeback_single_inode(struct inode *inode, struct writeback_control *wbc) 4711da177e4SLinus Torvalds { 4721da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 473251d6a47SWu Fengguang long nr_to_write = wbc->nr_to_write; 47401c03194SChristoph Hellwig unsigned dirty; 4751da177e4SLinus Torvalds int ret; 4761da177e4SLinus Torvalds 4774f8ad655SJan Kara WARN_ON(!(inode->i_state & I_SYNC)); 4781da177e4SLinus Torvalds 4799fb0a7daSTejun Heo trace_writeback_single_inode_start(inode, wbc, nr_to_write); 4809fb0a7daSTejun Heo 4811da177e4SLinus Torvalds ret = do_writepages(mapping, wbc); 4821da177e4SLinus Torvalds 48326821ed4SChristoph Hellwig /* 48426821ed4SChristoph Hellwig * Make sure to wait on the data before writing out the metadata. 48526821ed4SChristoph Hellwig * This is important for filesystems that modify metadata on data 4867747bd4bSDave Chinner * I/O completion. We don't do it for sync(2) writeback because it has a 4877747bd4bSDave Chinner * separate, external IO completion path and ->sync_fs for guaranteeing 4887747bd4bSDave Chinner * inode metadata is written back correctly. 48926821ed4SChristoph Hellwig */ 4907747bd4bSDave Chinner if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) { 49126821ed4SChristoph Hellwig int err = filemap_fdatawait(mapping); 4921da177e4SLinus Torvalds if (ret == 0) 4931da177e4SLinus Torvalds ret = err; 4941da177e4SLinus Torvalds } 4951da177e4SLinus Torvalds 4965547e8aaSDmitry Monakhov /* 4975547e8aaSDmitry Monakhov * Some filesystems may redirty the inode during the writeback 4985547e8aaSDmitry Monakhov * due to delalloc, clear dirty metadata flags right before 4995547e8aaSDmitry Monakhov * write_inode() 5005547e8aaSDmitry Monakhov */ 501250df6edSDave Chinner spin_lock(&inode->i_lock); 5029c6ac78eSTejun Heo 5035547e8aaSDmitry Monakhov dirty = inode->i_state & I_DIRTY; 504a2f48706STheodore Ts'o if (inode->i_state & I_DIRTY_TIME) { 505a2f48706STheodore Ts'o if ((dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) || 506a2f48706STheodore Ts'o unlikely(inode->i_state & I_DIRTY_TIME_EXPIRED) || 507a2f48706STheodore Ts'o unlikely(time_after(jiffies, 508a2f48706STheodore Ts'o (inode->dirtied_time_when + 509a2f48706STheodore Ts'o dirtytime_expire_interval * HZ)))) { 5100ae45f63STheodore Ts'o dirty |= I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED; 5110ae45f63STheodore Ts'o trace_writeback_lazytime(inode); 5120ae45f63STheodore Ts'o } 513a2f48706STheodore Ts'o } else 514a2f48706STheodore Ts'o inode->i_state &= ~I_DIRTY_TIME_EXPIRED; 5150ae45f63STheodore Ts'o inode->i_state &= ~dirty; 5169c6ac78eSTejun Heo 5179c6ac78eSTejun Heo /* 5189c6ac78eSTejun Heo * Paired with smp_mb() in __mark_inode_dirty(). This allows 5199c6ac78eSTejun Heo * __mark_inode_dirty() to test i_state without grabbing i_lock - 5209c6ac78eSTejun Heo * either they see the I_DIRTY bits cleared or we see the dirtied 5219c6ac78eSTejun Heo * inode. 5229c6ac78eSTejun Heo * 5239c6ac78eSTejun Heo * I_DIRTY_PAGES is always cleared together above even if @mapping 5249c6ac78eSTejun Heo * still has dirty pages. The flag is reinstated after smp_mb() if 5259c6ac78eSTejun Heo * necessary. This guarantees that either __mark_inode_dirty() 5269c6ac78eSTejun Heo * sees clear I_DIRTY_PAGES or we see PAGECACHE_TAG_DIRTY. 5279c6ac78eSTejun Heo */ 5289c6ac78eSTejun Heo smp_mb(); 5299c6ac78eSTejun Heo 5309c6ac78eSTejun Heo if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) 5319c6ac78eSTejun Heo inode->i_state |= I_DIRTY_PAGES; 5329c6ac78eSTejun Heo 533250df6edSDave Chinner spin_unlock(&inode->i_lock); 5349c6ac78eSTejun Heo 5350ae45f63STheodore Ts'o if (dirty & I_DIRTY_TIME) 5360ae45f63STheodore Ts'o mark_inode_dirty_sync(inode); 53726821ed4SChristoph Hellwig /* Don't write the inode if only I_DIRTY_PAGES was set */ 5380ae45f63STheodore Ts'o if (dirty & ~I_DIRTY_PAGES) { 539a9185b41SChristoph Hellwig int err = write_inode(inode, wbc); 5401da177e4SLinus Torvalds if (ret == 0) 5411da177e4SLinus Torvalds ret = err; 5421da177e4SLinus Torvalds } 5434f8ad655SJan Kara trace_writeback_single_inode(inode, wbc, nr_to_write); 5444f8ad655SJan Kara return ret; 5454f8ad655SJan Kara } 5464f8ad655SJan Kara 5474f8ad655SJan Kara /* 5484f8ad655SJan Kara * Write out an inode's dirty pages. Either the caller has an active reference 5494f8ad655SJan Kara * on the inode or the inode has I_WILL_FREE set. 5504f8ad655SJan Kara * 5514f8ad655SJan Kara * This function is designed to be called for writing back one inode which 5524f8ad655SJan Kara * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode() 5534f8ad655SJan Kara * and does more profound writeback list handling in writeback_sb_inodes(). 5544f8ad655SJan Kara */ 5554f8ad655SJan Kara static int 5564f8ad655SJan Kara writeback_single_inode(struct inode *inode, struct bdi_writeback *wb, 5574f8ad655SJan Kara struct writeback_control *wbc) 5584f8ad655SJan Kara { 5594f8ad655SJan Kara int ret = 0; 5604f8ad655SJan Kara 5614f8ad655SJan Kara spin_lock(&inode->i_lock); 5624f8ad655SJan Kara if (!atomic_read(&inode->i_count)) 5634f8ad655SJan Kara WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); 5644f8ad655SJan Kara else 5654f8ad655SJan Kara WARN_ON(inode->i_state & I_WILL_FREE); 5664f8ad655SJan Kara 5674f8ad655SJan Kara if (inode->i_state & I_SYNC) { 5684f8ad655SJan Kara if (wbc->sync_mode != WB_SYNC_ALL) 5694f8ad655SJan Kara goto out; 5704f8ad655SJan Kara /* 571169ebd90SJan Kara * It's a data-integrity sync. We must wait. Since callers hold 572169ebd90SJan Kara * inode reference or inode has I_WILL_FREE set, it cannot go 573169ebd90SJan Kara * away under us. 5744f8ad655SJan Kara */ 575169ebd90SJan Kara __inode_wait_for_writeback(inode); 5764f8ad655SJan Kara } 5774f8ad655SJan Kara WARN_ON(inode->i_state & I_SYNC); 5784f8ad655SJan Kara /* 579f9b0e058SJan Kara * Skip inode if it is clean and we have no outstanding writeback in 580f9b0e058SJan Kara * WB_SYNC_ALL mode. We don't want to mess with writeback lists in this 581f9b0e058SJan Kara * function since flusher thread may be doing for example sync in 582f9b0e058SJan Kara * parallel and if we move the inode, it could get skipped. So here we 583f9b0e058SJan Kara * make sure inode is on some writeback list and leave it there unless 584f9b0e058SJan Kara * we have completely cleaned the inode. 5854f8ad655SJan Kara */ 5860ae45f63STheodore Ts'o if (!(inode->i_state & I_DIRTY_ALL) && 587f9b0e058SJan Kara (wbc->sync_mode != WB_SYNC_ALL || 588f9b0e058SJan Kara !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK))) 5894f8ad655SJan Kara goto out; 5904f8ad655SJan Kara inode->i_state |= I_SYNC; 5914f8ad655SJan Kara spin_unlock(&inode->i_lock); 5924f8ad655SJan Kara 593cd8ed2a4SYan Hong ret = __writeback_single_inode(inode, wbc); 5941da177e4SLinus Torvalds 595f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 596250df6edSDave Chinner spin_lock(&inode->i_lock); 5974f8ad655SJan Kara /* 5984f8ad655SJan Kara * If inode is clean, remove it from writeback lists. Otherwise don't 5994f8ad655SJan Kara * touch it. See comment above for explanation. 6004f8ad655SJan Kara */ 6010ae45f63STheodore Ts'o if (!(inode->i_state & I_DIRTY_ALL)) 6024f8ad655SJan Kara list_del_init(&inode->i_wb_list); 6034f8ad655SJan Kara spin_unlock(&wb->list_lock); 6041c0eeaf5SJoern Engel inode_sync_complete(inode); 6054f8ad655SJan Kara out: 6064f8ad655SJan Kara spin_unlock(&inode->i_lock); 6071da177e4SLinus Torvalds return ret; 6081da177e4SLinus Torvalds } 6091da177e4SLinus Torvalds 610a88a341aSTejun Heo static long writeback_chunk_size(struct bdi_writeback *wb, 6111a12d8bdSWu Fengguang struct wb_writeback_work *work) 612d46db3d5SWu Fengguang { 613d46db3d5SWu Fengguang long pages; 614d46db3d5SWu Fengguang 615d46db3d5SWu Fengguang /* 616d46db3d5SWu Fengguang * WB_SYNC_ALL mode does livelock avoidance by syncing dirty 617d46db3d5SWu Fengguang * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX 618d46db3d5SWu Fengguang * here avoids calling into writeback_inodes_wb() more than once. 619d46db3d5SWu Fengguang * 620d46db3d5SWu Fengguang * The intended call sequence for WB_SYNC_ALL writeback is: 621d46db3d5SWu Fengguang * 622d46db3d5SWu Fengguang * wb_writeback() 623d46db3d5SWu Fengguang * writeback_sb_inodes() <== called only once 624d46db3d5SWu Fengguang * write_cache_pages() <== called once for each inode 625d46db3d5SWu Fengguang * (quickly) tag currently dirty pages 626d46db3d5SWu Fengguang * (maybe slowly) sync all tagged pages 627d46db3d5SWu Fengguang */ 628d46db3d5SWu Fengguang if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages) 629d46db3d5SWu Fengguang pages = LONG_MAX; 6301a12d8bdSWu Fengguang else { 631a88a341aSTejun Heo pages = min(wb->avg_write_bandwidth / 2, 6321a12d8bdSWu Fengguang global_dirty_limit / DIRTY_SCOPE); 6331a12d8bdSWu Fengguang pages = min(pages, work->nr_pages); 6341a12d8bdSWu Fengguang pages = round_down(pages + MIN_WRITEBACK_PAGES, 6351a12d8bdSWu Fengguang MIN_WRITEBACK_PAGES); 6361a12d8bdSWu Fengguang } 637d46db3d5SWu Fengguang 638d46db3d5SWu Fengguang return pages; 639d46db3d5SWu Fengguang } 640d46db3d5SWu Fengguang 64103ba3782SJens Axboe /* 642f11c9c5cSEdward Shishkin * Write a portion of b_io inodes which belong to @sb. 643edadfb10SChristoph Hellwig * 644d46db3d5SWu Fengguang * Return the number of pages and/or inodes written. 645f11c9c5cSEdward Shishkin */ 646d46db3d5SWu Fengguang static long writeback_sb_inodes(struct super_block *sb, 647d46db3d5SWu Fengguang struct bdi_writeback *wb, 648d46db3d5SWu Fengguang struct wb_writeback_work *work) 64903ba3782SJens Axboe { 650d46db3d5SWu Fengguang struct writeback_control wbc = { 651d46db3d5SWu Fengguang .sync_mode = work->sync_mode, 652d46db3d5SWu Fengguang .tagged_writepages = work->tagged_writepages, 653d46db3d5SWu Fengguang .for_kupdate = work->for_kupdate, 654d46db3d5SWu Fengguang .for_background = work->for_background, 6557747bd4bSDave Chinner .for_sync = work->for_sync, 656d46db3d5SWu Fengguang .range_cyclic = work->range_cyclic, 657d46db3d5SWu Fengguang .range_start = 0, 658d46db3d5SWu Fengguang .range_end = LLONG_MAX, 659d46db3d5SWu Fengguang }; 660d46db3d5SWu Fengguang unsigned long start_time = jiffies; 661d46db3d5SWu Fengguang long write_chunk; 662d46db3d5SWu Fengguang long wrote = 0; /* count both pages and inodes */ 663d46db3d5SWu Fengguang 66403ba3782SJens Axboe while (!list_empty(&wb->b_io)) { 6657ccf19a8SNick Piggin struct inode *inode = wb_inode(wb->b_io.prev); 666edadfb10SChristoph Hellwig 667edadfb10SChristoph Hellwig if (inode->i_sb != sb) { 668d46db3d5SWu Fengguang if (work->sb) { 669edadfb10SChristoph Hellwig /* 670edadfb10SChristoph Hellwig * We only want to write back data for this 671edadfb10SChristoph Hellwig * superblock, move all inodes not belonging 672edadfb10SChristoph Hellwig * to it back onto the dirty list. 673edadfb10SChristoph Hellwig */ 674f758eeabSChristoph Hellwig redirty_tail(inode, wb); 67566f3b8e2SJens Axboe continue; 67666f3b8e2SJens Axboe } 677edadfb10SChristoph Hellwig 678edadfb10SChristoph Hellwig /* 679edadfb10SChristoph Hellwig * The inode belongs to a different superblock. 680edadfb10SChristoph Hellwig * Bounce back to the caller to unpin this and 681edadfb10SChristoph Hellwig * pin the next superblock. 682edadfb10SChristoph Hellwig */ 683d46db3d5SWu Fengguang break; 684edadfb10SChristoph Hellwig } 685edadfb10SChristoph Hellwig 6869843b76aSChristoph Hellwig /* 687331cbdeeSWanpeng Li * Don't bother with new inodes or inodes being freed, first 688331cbdeeSWanpeng Li * kind does not need periodic writeout yet, and for the latter 6899843b76aSChristoph Hellwig * kind writeout is handled by the freer. 6909843b76aSChristoph Hellwig */ 691250df6edSDave Chinner spin_lock(&inode->i_lock); 6929843b76aSChristoph Hellwig if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { 693250df6edSDave Chinner spin_unlock(&inode->i_lock); 694fcc5c222SWu Fengguang redirty_tail(inode, wb); 6957ef0d737SNick Piggin continue; 6967ef0d737SNick Piggin } 697cc1676d9SJan Kara if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) { 698cc1676d9SJan Kara /* 699cc1676d9SJan Kara * If this inode is locked for writeback and we are not 700cc1676d9SJan Kara * doing writeback-for-data-integrity, move it to 701cc1676d9SJan Kara * b_more_io so that writeback can proceed with the 702cc1676d9SJan Kara * other inodes on s_io. 703cc1676d9SJan Kara * 704cc1676d9SJan Kara * We'll have another go at writing back this inode 705cc1676d9SJan Kara * when we completed a full scan of b_io. 706cc1676d9SJan Kara */ 707cc1676d9SJan Kara spin_unlock(&inode->i_lock); 708cc1676d9SJan Kara requeue_io(inode, wb); 709cc1676d9SJan Kara trace_writeback_sb_inodes_requeue(inode); 710cc1676d9SJan Kara continue; 711cc1676d9SJan Kara } 712f0d07b7fSJan Kara spin_unlock(&wb->list_lock); 713f0d07b7fSJan Kara 7144f8ad655SJan Kara /* 7154f8ad655SJan Kara * We already requeued the inode if it had I_SYNC set and we 7164f8ad655SJan Kara * are doing WB_SYNC_NONE writeback. So this catches only the 7174f8ad655SJan Kara * WB_SYNC_ALL case. 7184f8ad655SJan Kara */ 719169ebd90SJan Kara if (inode->i_state & I_SYNC) { 720169ebd90SJan Kara /* Wait for I_SYNC. This function drops i_lock... */ 721169ebd90SJan Kara inode_sleep_on_writeback(inode); 722169ebd90SJan Kara /* Inode may be gone, start again */ 723ead188f9SJan Kara spin_lock(&wb->list_lock); 724169ebd90SJan Kara continue; 725169ebd90SJan Kara } 7264f8ad655SJan Kara inode->i_state |= I_SYNC; 7274f8ad655SJan Kara spin_unlock(&inode->i_lock); 728169ebd90SJan Kara 729a88a341aSTejun Heo write_chunk = writeback_chunk_size(wb, work); 730d46db3d5SWu Fengguang wbc.nr_to_write = write_chunk; 731d46db3d5SWu Fengguang wbc.pages_skipped = 0; 732250df6edSDave Chinner 733169ebd90SJan Kara /* 734169ebd90SJan Kara * We use I_SYNC to pin the inode in memory. While it is set 735169ebd90SJan Kara * evict_inode() will wait so the inode cannot be freed. 736169ebd90SJan Kara */ 737cd8ed2a4SYan Hong __writeback_single_inode(inode, &wbc); 738d46db3d5SWu Fengguang 739d46db3d5SWu Fengguang work->nr_pages -= write_chunk - wbc.nr_to_write; 740d46db3d5SWu Fengguang wrote += write_chunk - wbc.nr_to_write; 7414f8ad655SJan Kara spin_lock(&wb->list_lock); 7424f8ad655SJan Kara spin_lock(&inode->i_lock); 7430ae45f63STheodore Ts'o if (!(inode->i_state & I_DIRTY_ALL)) 744d46db3d5SWu Fengguang wrote++; 7454f8ad655SJan Kara requeue_inode(inode, wb, &wbc); 7464f8ad655SJan Kara inode_sync_complete(inode); 7470f1b1fd8SDave Chinner spin_unlock(&inode->i_lock); 748169ebd90SJan Kara cond_resched_lock(&wb->list_lock); 749d46db3d5SWu Fengguang /* 750d46db3d5SWu Fengguang * bail out to wb_writeback() often enough to check 751d46db3d5SWu Fengguang * background threshold and other termination conditions. 752d46db3d5SWu Fengguang */ 753d46db3d5SWu Fengguang if (wrote) { 754d46db3d5SWu Fengguang if (time_is_before_jiffies(start_time + HZ / 10UL)) 755d46db3d5SWu Fengguang break; 756d46db3d5SWu Fengguang if (work->nr_pages <= 0) 757d46db3d5SWu Fengguang break; 7581da177e4SLinus Torvalds } 7598bc3be27SFengguang Wu } 760d46db3d5SWu Fengguang return wrote; 761f11c9c5cSEdward Shishkin } 76238f21977SNick Piggin 763d46db3d5SWu Fengguang static long __writeback_inodes_wb(struct bdi_writeback *wb, 764d46db3d5SWu Fengguang struct wb_writeback_work *work) 765f11c9c5cSEdward Shishkin { 766d46db3d5SWu Fengguang unsigned long start_time = jiffies; 767d46db3d5SWu Fengguang long wrote = 0; 768f11c9c5cSEdward Shishkin 769f11c9c5cSEdward Shishkin while (!list_empty(&wb->b_io)) { 7707ccf19a8SNick Piggin struct inode *inode = wb_inode(wb->b_io.prev); 771f11c9c5cSEdward Shishkin struct super_block *sb = inode->i_sb; 772f11c9c5cSEdward Shishkin 773eb6ef3dfSKonstantin Khlebnikov if (!trylock_super(sb)) { 7740e995816SWu Fengguang /* 775eb6ef3dfSKonstantin Khlebnikov * trylock_super() may fail consistently due to 7760e995816SWu Fengguang * s_umount being grabbed by someone else. Don't use 7770e995816SWu Fengguang * requeue_io() to avoid busy retrying the inode/sb. 7780e995816SWu Fengguang */ 7790e995816SWu Fengguang redirty_tail(inode, wb); 780d19de7edSChristoph Hellwig continue; 781334132aeSChristoph Hellwig } 782d46db3d5SWu Fengguang wrote += writeback_sb_inodes(sb, wb, work); 783eb6ef3dfSKonstantin Khlebnikov up_read(&sb->s_umount); 784f11c9c5cSEdward Shishkin 785d46db3d5SWu Fengguang /* refer to the same tests at the end of writeback_sb_inodes */ 786d46db3d5SWu Fengguang if (wrote) { 787d46db3d5SWu Fengguang if (time_is_before_jiffies(start_time + HZ / 10UL)) 788d46db3d5SWu Fengguang break; 789d46db3d5SWu Fengguang if (work->nr_pages <= 0) 790f11c9c5cSEdward Shishkin break; 791f11c9c5cSEdward Shishkin } 792d46db3d5SWu Fengguang } 79366f3b8e2SJens Axboe /* Leave any unwritten inodes on b_io */ 794d46db3d5SWu Fengguang return wrote; 79566f3b8e2SJens Axboe } 79666f3b8e2SJens Axboe 7977d9f073bSWanpeng Li static long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages, 7980e175a18SCurt Wohlgemuth enum wb_reason reason) 799edadfb10SChristoph Hellwig { 800d46db3d5SWu Fengguang struct wb_writeback_work work = { 801d46db3d5SWu Fengguang .nr_pages = nr_pages, 802d46db3d5SWu Fengguang .sync_mode = WB_SYNC_NONE, 803d46db3d5SWu Fengguang .range_cyclic = 1, 8040e175a18SCurt Wohlgemuth .reason = reason, 805d46db3d5SWu Fengguang }; 806edadfb10SChristoph Hellwig 807f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 808424b351fSWu Fengguang if (list_empty(&wb->b_io)) 809ad4e38ddSCurt Wohlgemuth queue_io(wb, &work); 810d46db3d5SWu Fengguang __writeback_inodes_wb(wb, &work); 811f758eeabSChristoph Hellwig spin_unlock(&wb->list_lock); 812edadfb10SChristoph Hellwig 813d46db3d5SWu Fengguang return nr_pages - work.nr_pages; 81466f3b8e2SJens Axboe } 81566f3b8e2SJens Axboe 816a88a341aSTejun Heo static bool over_bground_thresh(struct bdi_writeback *wb) 81703ba3782SJens Axboe { 81803ba3782SJens Axboe unsigned long background_thresh, dirty_thresh; 81903ba3782SJens Axboe 82016c4042fSWu Fengguang global_dirty_limits(&background_thresh, &dirty_thresh); 82103ba3782SJens Axboe 822b00949aaSWu Fengguang if (global_page_state(NR_FILE_DIRTY) + 823b00949aaSWu Fengguang global_page_state(NR_UNSTABLE_NFS) > background_thresh) 824b00949aaSWu Fengguang return true; 825b00949aaSWu Fengguang 826a88a341aSTejun Heo if (wb_stat(wb, WB_RECLAIMABLE) > wb_dirty_limit(wb, background_thresh)) 827b00949aaSWu Fengguang return true; 828b00949aaSWu Fengguang 829b00949aaSWu Fengguang return false; 83003ba3782SJens Axboe } 83103ba3782SJens Axboe 83203ba3782SJens Axboe /* 833e98be2d5SWu Fengguang * Called under wb->list_lock. If there are multiple wb per bdi, 834e98be2d5SWu Fengguang * only the flusher working on the first wb should do it. 835e98be2d5SWu Fengguang */ 836e98be2d5SWu Fengguang static void wb_update_bandwidth(struct bdi_writeback *wb, 837e98be2d5SWu Fengguang unsigned long start_time) 838e98be2d5SWu Fengguang { 839a88a341aSTejun Heo __wb_update_bandwidth(wb, 0, 0, 0, 0, 0, start_time); 840e98be2d5SWu Fengguang } 841e98be2d5SWu Fengguang 842e98be2d5SWu Fengguang /* 84303ba3782SJens Axboe * Explicit flushing or periodic writeback of "old" data. 84403ba3782SJens Axboe * 84503ba3782SJens Axboe * Define "old": the first time one of an inode's pages is dirtied, we mark the 84603ba3782SJens Axboe * dirtying-time in the inode's address_space. So this periodic writeback code 84703ba3782SJens Axboe * just walks the superblock inode list, writing back any inodes which are 84803ba3782SJens Axboe * older than a specific point in time. 84903ba3782SJens Axboe * 85003ba3782SJens Axboe * Try to run once per dirty_writeback_interval. But if a writeback event 85103ba3782SJens Axboe * takes longer than a dirty_writeback_interval interval, then leave a 85203ba3782SJens Axboe * one-second gap. 85303ba3782SJens Axboe * 85403ba3782SJens Axboe * older_than_this takes precedence over nr_to_write. So we'll only write back 85503ba3782SJens Axboe * all dirty pages if they are all attached to "old" mappings. 85603ba3782SJens Axboe */ 857c4a77a6cSJens Axboe static long wb_writeback(struct bdi_writeback *wb, 85883ba7b07SChristoph Hellwig struct wb_writeback_work *work) 85903ba3782SJens Axboe { 860e98be2d5SWu Fengguang unsigned long wb_start = jiffies; 861d46db3d5SWu Fengguang long nr_pages = work->nr_pages; 8620dc83bd3SJan Kara unsigned long oldest_jif; 863a5989bdcSJan Kara struct inode *inode; 864d46db3d5SWu Fengguang long progress; 86503ba3782SJens Axboe 8660dc83bd3SJan Kara oldest_jif = jiffies; 8670dc83bd3SJan Kara work->older_than_this = &oldest_jif; 86803ba3782SJens Axboe 869e8dfc305SWu Fengguang spin_lock(&wb->list_lock); 87003ba3782SJens Axboe for (;;) { 87103ba3782SJens Axboe /* 872d3ddec76SWu Fengguang * Stop writeback when nr_pages has been consumed 87303ba3782SJens Axboe */ 87483ba7b07SChristoph Hellwig if (work->nr_pages <= 0) 87503ba3782SJens Axboe break; 87603ba3782SJens Axboe 87703ba3782SJens Axboe /* 878aa373cf5SJan Kara * Background writeout and kupdate-style writeback may 879aa373cf5SJan Kara * run forever. Stop them if there is other work to do 880aa373cf5SJan Kara * so that e.g. sync can proceed. They'll be restarted 881aa373cf5SJan Kara * after the other works are all done. 882aa373cf5SJan Kara */ 883aa373cf5SJan Kara if ((work->for_background || work->for_kupdate) && 884f0054bb1STejun Heo !list_empty(&wb->work_list)) 885aa373cf5SJan Kara break; 886aa373cf5SJan Kara 887aa373cf5SJan Kara /* 888d3ddec76SWu Fengguang * For background writeout, stop when we are below the 889d3ddec76SWu Fengguang * background dirty threshold 89003ba3782SJens Axboe */ 891a88a341aSTejun Heo if (work->for_background && !over_bground_thresh(wb)) 89203ba3782SJens Axboe break; 89303ba3782SJens Axboe 8941bc36b64SJan Kara /* 8951bc36b64SJan Kara * Kupdate and background works are special and we want to 8961bc36b64SJan Kara * include all inodes that need writing. Livelock avoidance is 8971bc36b64SJan Kara * handled by these works yielding to any other work so we are 8981bc36b64SJan Kara * safe. 8991bc36b64SJan Kara */ 900ba9aa839SWu Fengguang if (work->for_kupdate) { 9010dc83bd3SJan Kara oldest_jif = jiffies - 902ba9aa839SWu Fengguang msecs_to_jiffies(dirty_expire_interval * 10); 9031bc36b64SJan Kara } else if (work->for_background) 9040dc83bd3SJan Kara oldest_jif = jiffies; 905028c2dd1SDave Chinner 906d46db3d5SWu Fengguang trace_writeback_start(wb->bdi, work); 907e8dfc305SWu Fengguang if (list_empty(&wb->b_io)) 908ad4e38ddSCurt Wohlgemuth queue_io(wb, work); 90983ba7b07SChristoph Hellwig if (work->sb) 910d46db3d5SWu Fengguang progress = writeback_sb_inodes(work->sb, wb, work); 911edadfb10SChristoph Hellwig else 912d46db3d5SWu Fengguang progress = __writeback_inodes_wb(wb, work); 913d46db3d5SWu Fengguang trace_writeback_written(wb->bdi, work); 914028c2dd1SDave Chinner 915e98be2d5SWu Fengguang wb_update_bandwidth(wb, wb_start); 91603ba3782SJens Axboe 91703ba3782SJens Axboe /* 91871fd05a8SJens Axboe * Did we write something? Try for more 919e6fb6da2SWu Fengguang * 920e6fb6da2SWu Fengguang * Dirty inodes are moved to b_io for writeback in batches. 921e6fb6da2SWu Fengguang * The completion of the current batch does not necessarily 922e6fb6da2SWu Fengguang * mean the overall work is done. So we keep looping as long 923e6fb6da2SWu Fengguang * as made some progress on cleaning pages or inodes. 92471fd05a8SJens Axboe */ 925d46db3d5SWu Fengguang if (progress) 92603ba3782SJens Axboe continue; 927a5989bdcSJan Kara /* 928e6fb6da2SWu Fengguang * No more inodes for IO, bail 929a5989bdcSJan Kara */ 930b7a2441fSWu Fengguang if (list_empty(&wb->b_more_io)) 93103ba3782SJens Axboe break; 93203ba3782SJens Axboe /* 9338010c3b6SJens Axboe * Nothing written. Wait for some inode to 9348010c3b6SJens Axboe * become available for writeback. Otherwise 9358010c3b6SJens Axboe * we'll just busyloop. 93603ba3782SJens Axboe */ 93703ba3782SJens Axboe if (!list_empty(&wb->b_more_io)) { 938d46db3d5SWu Fengguang trace_writeback_wait(wb->bdi, work); 93903ba3782SJens Axboe inode = wb_inode(wb->b_more_io.prev); 940250df6edSDave Chinner spin_lock(&inode->i_lock); 941f0d07b7fSJan Kara spin_unlock(&wb->list_lock); 942169ebd90SJan Kara /* This function drops i_lock... */ 943169ebd90SJan Kara inode_sleep_on_writeback(inode); 944f0d07b7fSJan Kara spin_lock(&wb->list_lock); 94503ba3782SJens Axboe } 94603ba3782SJens Axboe } 947e8dfc305SWu Fengguang spin_unlock(&wb->list_lock); 94803ba3782SJens Axboe 949d46db3d5SWu Fengguang return nr_pages - work->nr_pages; 95003ba3782SJens Axboe } 95103ba3782SJens Axboe 95203ba3782SJens Axboe /* 95383ba7b07SChristoph Hellwig * Return the next wb_writeback_work struct that hasn't been processed yet. 95403ba3782SJens Axboe */ 955f0054bb1STejun Heo static struct wb_writeback_work *get_next_work_item(struct bdi_writeback *wb) 95603ba3782SJens Axboe { 95783ba7b07SChristoph Hellwig struct wb_writeback_work *work = NULL; 95803ba3782SJens Axboe 959f0054bb1STejun Heo spin_lock_bh(&wb->work_lock); 960f0054bb1STejun Heo if (!list_empty(&wb->work_list)) { 961f0054bb1STejun Heo work = list_entry(wb->work_list.next, 96283ba7b07SChristoph Hellwig struct wb_writeback_work, list); 96383ba7b07SChristoph Hellwig list_del_init(&work->list); 96403ba3782SJens Axboe } 965f0054bb1STejun Heo spin_unlock_bh(&wb->work_lock); 96683ba7b07SChristoph Hellwig return work; 96703ba3782SJens Axboe } 96803ba3782SJens Axboe 969cdf01dd5SLinus Torvalds /* 970cdf01dd5SLinus Torvalds * Add in the number of potentially dirty inodes, because each inode 971cdf01dd5SLinus Torvalds * write can dirty pagecache in the underlying blockdev. 972cdf01dd5SLinus Torvalds */ 973cdf01dd5SLinus Torvalds static unsigned long get_nr_dirty_pages(void) 974cdf01dd5SLinus Torvalds { 975cdf01dd5SLinus Torvalds return global_page_state(NR_FILE_DIRTY) + 976cdf01dd5SLinus Torvalds global_page_state(NR_UNSTABLE_NFS) + 977cdf01dd5SLinus Torvalds get_nr_dirty_inodes(); 978cdf01dd5SLinus Torvalds } 979cdf01dd5SLinus Torvalds 9806585027aSJan Kara static long wb_check_background_flush(struct bdi_writeback *wb) 9816585027aSJan Kara { 982a88a341aSTejun Heo if (over_bground_thresh(wb)) { 9836585027aSJan Kara 9846585027aSJan Kara struct wb_writeback_work work = { 9856585027aSJan Kara .nr_pages = LONG_MAX, 9866585027aSJan Kara .sync_mode = WB_SYNC_NONE, 9876585027aSJan Kara .for_background = 1, 9886585027aSJan Kara .range_cyclic = 1, 9890e175a18SCurt Wohlgemuth .reason = WB_REASON_BACKGROUND, 9906585027aSJan Kara }; 9916585027aSJan Kara 9926585027aSJan Kara return wb_writeback(wb, &work); 9936585027aSJan Kara } 9946585027aSJan Kara 9956585027aSJan Kara return 0; 9966585027aSJan Kara } 9976585027aSJan Kara 99803ba3782SJens Axboe static long wb_check_old_data_flush(struct bdi_writeback *wb) 99903ba3782SJens Axboe { 100003ba3782SJens Axboe unsigned long expired; 100103ba3782SJens Axboe long nr_pages; 100203ba3782SJens Axboe 100369b62d01SJens Axboe /* 100469b62d01SJens Axboe * When set to zero, disable periodic writeback 100569b62d01SJens Axboe */ 100669b62d01SJens Axboe if (!dirty_writeback_interval) 100769b62d01SJens Axboe return 0; 100869b62d01SJens Axboe 100903ba3782SJens Axboe expired = wb->last_old_flush + 101003ba3782SJens Axboe msecs_to_jiffies(dirty_writeback_interval * 10); 101103ba3782SJens Axboe if (time_before(jiffies, expired)) 101203ba3782SJens Axboe return 0; 101303ba3782SJens Axboe 101403ba3782SJens Axboe wb->last_old_flush = jiffies; 1015cdf01dd5SLinus Torvalds nr_pages = get_nr_dirty_pages(); 101603ba3782SJens Axboe 1017c4a77a6cSJens Axboe if (nr_pages) { 101883ba7b07SChristoph Hellwig struct wb_writeback_work work = { 1019c4a77a6cSJens Axboe .nr_pages = nr_pages, 1020c4a77a6cSJens Axboe .sync_mode = WB_SYNC_NONE, 1021c4a77a6cSJens Axboe .for_kupdate = 1, 1022c4a77a6cSJens Axboe .range_cyclic = 1, 10230e175a18SCurt Wohlgemuth .reason = WB_REASON_PERIODIC, 1024c4a77a6cSJens Axboe }; 1025c4a77a6cSJens Axboe 102683ba7b07SChristoph Hellwig return wb_writeback(wb, &work); 1027c4a77a6cSJens Axboe } 102803ba3782SJens Axboe 102903ba3782SJens Axboe return 0; 103003ba3782SJens Axboe } 103103ba3782SJens Axboe 103203ba3782SJens Axboe /* 103303ba3782SJens Axboe * Retrieve work items and do the writeback they describe 103403ba3782SJens Axboe */ 103525d130baSWanpeng Li static long wb_do_writeback(struct bdi_writeback *wb) 103603ba3782SJens Axboe { 103783ba7b07SChristoph Hellwig struct wb_writeback_work *work; 1038c4a77a6cSJens Axboe long wrote = 0; 103903ba3782SJens Axboe 10404452226eSTejun Heo set_bit(WB_writeback_running, &wb->state); 1041f0054bb1STejun Heo while ((work = get_next_work_item(wb)) != NULL) { 104283ba7b07SChristoph Hellwig 1043f0054bb1STejun Heo trace_writeback_exec(wb->bdi, work); 1044455b2864SDave Chinner 104583ba7b07SChristoph Hellwig wrote += wb_writeback(wb, work); 104603ba3782SJens Axboe 104703ba3782SJens Axboe /* 104883ba7b07SChristoph Hellwig * Notify the caller of completion if this is a synchronous 104983ba7b07SChristoph Hellwig * work item, otherwise just free it. 105003ba3782SJens Axboe */ 105183ba7b07SChristoph Hellwig if (work->done) 105283ba7b07SChristoph Hellwig complete(work->done); 105383ba7b07SChristoph Hellwig else 105483ba7b07SChristoph Hellwig kfree(work); 105503ba3782SJens Axboe } 105603ba3782SJens Axboe 105703ba3782SJens Axboe /* 105803ba3782SJens Axboe * Check for periodic writeback, kupdated() style 105903ba3782SJens Axboe */ 106003ba3782SJens Axboe wrote += wb_check_old_data_flush(wb); 10616585027aSJan Kara wrote += wb_check_background_flush(wb); 10624452226eSTejun Heo clear_bit(WB_writeback_running, &wb->state); 106303ba3782SJens Axboe 106403ba3782SJens Axboe return wrote; 106503ba3782SJens Axboe } 106603ba3782SJens Axboe 106703ba3782SJens Axboe /* 106803ba3782SJens Axboe * Handle writeback of dirty data for the device backed by this bdi. Also 1069839a8e86STejun Heo * reschedules periodically and does kupdated style flushing. 107003ba3782SJens Axboe */ 1071f0054bb1STejun Heo void wb_workfn(struct work_struct *work) 107203ba3782SJens Axboe { 1073839a8e86STejun Heo struct bdi_writeback *wb = container_of(to_delayed_work(work), 1074839a8e86STejun Heo struct bdi_writeback, dwork); 107503ba3782SJens Axboe long pages_written; 107603ba3782SJens Axboe 1077f0054bb1STejun Heo set_worker_desc("flush-%s", dev_name(wb->bdi->dev)); 1078766f9164SPeter Zijlstra current->flags |= PF_SWAPWRITE; 107903ba3782SJens Axboe 1080839a8e86STejun Heo if (likely(!current_is_workqueue_rescuer() || 10814452226eSTejun Heo !test_bit(WB_registered, &wb->state))) { 108203ba3782SJens Axboe /* 1083f0054bb1STejun Heo * The normal path. Keep writing back @wb until its 1084839a8e86STejun Heo * work_list is empty. Note that this path is also taken 1085f0054bb1STejun Heo * if @wb is shutting down even when we're running off the 1086839a8e86STejun Heo * rescuer as work_list needs to be drained. 108703ba3782SJens Axboe */ 1088839a8e86STejun Heo do { 108925d130baSWanpeng Li pages_written = wb_do_writeback(wb); 1090455b2864SDave Chinner trace_writeback_pages_written(pages_written); 1091f0054bb1STejun Heo } while (!list_empty(&wb->work_list)); 1092839a8e86STejun Heo } else { 1093253c34e9SArtem Bityutskiy /* 1094839a8e86STejun Heo * bdi_wq can't get enough workers and we're running off 1095839a8e86STejun Heo * the emergency worker. Don't hog it. Hopefully, 1024 is 1096839a8e86STejun Heo * enough for efficient IO. 1097253c34e9SArtem Bityutskiy */ 1098f0054bb1STejun Heo pages_written = writeback_inodes_wb(wb, 1024, 1099839a8e86STejun Heo WB_REASON_FORKER_THREAD); 1100839a8e86STejun Heo trace_writeback_pages_written(pages_written); 110103ba3782SJens Axboe } 110203ba3782SJens Axboe 1103f0054bb1STejun Heo if (!list_empty(&wb->work_list)) 11046ca738d6SDerek Basehore mod_delayed_work(bdi_wq, &wb->dwork, 0); 11056ca738d6SDerek Basehore else if (wb_has_dirty_io(wb) && dirty_writeback_interval) 1106f0054bb1STejun Heo wb_wakeup_delayed(wb); 1107455b2864SDave Chinner 1108839a8e86STejun Heo current->flags &= ~PF_SWAPWRITE; 110903ba3782SJens Axboe } 111003ba3782SJens Axboe 111103ba3782SJens Axboe /* 111203ba3782SJens Axboe * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back 111303ba3782SJens Axboe * the whole world. 111403ba3782SJens Axboe */ 11150e175a18SCurt Wohlgemuth void wakeup_flusher_threads(long nr_pages, enum wb_reason reason) 111603ba3782SJens Axboe { 1117b8c2f347SChristoph Hellwig struct backing_dev_info *bdi; 1118b8c2f347SChristoph Hellwig 111947df3ddeSJan Kara if (!nr_pages) 112047df3ddeSJan Kara nr_pages = get_nr_dirty_pages(); 1121b8c2f347SChristoph Hellwig 1122b8c2f347SChristoph Hellwig rcu_read_lock(); 1123b8c2f347SChristoph Hellwig list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { 1124b8c2f347SChristoph Hellwig if (!bdi_has_dirty_io(bdi)) 1125b8c2f347SChristoph Hellwig continue; 1126f0054bb1STejun Heo __wb_start_writeback(&bdi->wb, nr_pages, false, reason); 1127b8c2f347SChristoph Hellwig } 1128b8c2f347SChristoph Hellwig rcu_read_unlock(); 112903ba3782SJens Axboe } 113003ba3782SJens Axboe 1131a2f48706STheodore Ts'o /* 1132a2f48706STheodore Ts'o * Wake up bdi's periodically to make sure dirtytime inodes gets 1133a2f48706STheodore Ts'o * written back periodically. We deliberately do *not* check the 1134a2f48706STheodore Ts'o * b_dirtytime list in wb_has_dirty_io(), since this would cause the 1135a2f48706STheodore Ts'o * kernel to be constantly waking up once there are any dirtytime 1136a2f48706STheodore Ts'o * inodes on the system. So instead we define a separate delayed work 1137a2f48706STheodore Ts'o * function which gets called much more rarely. (By default, only 1138a2f48706STheodore Ts'o * once every 12 hours.) 1139a2f48706STheodore Ts'o * 1140a2f48706STheodore Ts'o * If there is any other write activity going on in the file system, 1141a2f48706STheodore Ts'o * this function won't be necessary. But if the only thing that has 1142a2f48706STheodore Ts'o * happened on the file system is a dirtytime inode caused by an atime 1143a2f48706STheodore Ts'o * update, we need this infrastructure below to make sure that inode 1144a2f48706STheodore Ts'o * eventually gets pushed out to disk. 1145a2f48706STheodore Ts'o */ 1146a2f48706STheodore Ts'o static void wakeup_dirtytime_writeback(struct work_struct *w); 1147a2f48706STheodore Ts'o static DECLARE_DELAYED_WORK(dirtytime_work, wakeup_dirtytime_writeback); 1148a2f48706STheodore Ts'o 1149a2f48706STheodore Ts'o static void wakeup_dirtytime_writeback(struct work_struct *w) 1150a2f48706STheodore Ts'o { 1151a2f48706STheodore Ts'o struct backing_dev_info *bdi; 1152a2f48706STheodore Ts'o 1153a2f48706STheodore Ts'o rcu_read_lock(); 1154a2f48706STheodore Ts'o list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { 1155a2f48706STheodore Ts'o if (list_empty(&bdi->wb.b_dirty_time)) 1156a2f48706STheodore Ts'o continue; 1157f0054bb1STejun Heo wb_wakeup(&bdi->wb); 1158a2f48706STheodore Ts'o } 1159a2f48706STheodore Ts'o rcu_read_unlock(); 1160a2f48706STheodore Ts'o schedule_delayed_work(&dirtytime_work, dirtytime_expire_interval * HZ); 1161a2f48706STheodore Ts'o } 1162a2f48706STheodore Ts'o 1163a2f48706STheodore Ts'o static int __init start_dirtytime_writeback(void) 1164a2f48706STheodore Ts'o { 1165a2f48706STheodore Ts'o schedule_delayed_work(&dirtytime_work, dirtytime_expire_interval * HZ); 1166a2f48706STheodore Ts'o return 0; 1167a2f48706STheodore Ts'o } 1168a2f48706STheodore Ts'o __initcall(start_dirtytime_writeback); 1169a2f48706STheodore Ts'o 11701efff914STheodore Ts'o int dirtytime_interval_handler(struct ctl_table *table, int write, 11711efff914STheodore Ts'o void __user *buffer, size_t *lenp, loff_t *ppos) 11721efff914STheodore Ts'o { 11731efff914STheodore Ts'o int ret; 11741efff914STheodore Ts'o 11751efff914STheodore Ts'o ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos); 11761efff914STheodore Ts'o if (ret == 0 && write) 11771efff914STheodore Ts'o mod_delayed_work(system_wq, &dirtytime_work, 0); 11781efff914STheodore Ts'o return ret; 11791efff914STheodore Ts'o } 11801efff914STheodore Ts'o 118103ba3782SJens Axboe static noinline void block_dump___mark_inode_dirty(struct inode *inode) 118203ba3782SJens Axboe { 118303ba3782SJens Axboe if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { 118403ba3782SJens Axboe struct dentry *dentry; 118503ba3782SJens Axboe const char *name = "?"; 118603ba3782SJens Axboe 118703ba3782SJens Axboe dentry = d_find_alias(inode); 118803ba3782SJens Axboe if (dentry) { 118903ba3782SJens Axboe spin_lock(&dentry->d_lock); 119003ba3782SJens Axboe name = (const char *) dentry->d_name.name; 119103ba3782SJens Axboe } 119203ba3782SJens Axboe printk(KERN_DEBUG 119303ba3782SJens Axboe "%s(%d): dirtied inode %lu (%s) on %s\n", 119403ba3782SJens Axboe current->comm, task_pid_nr(current), inode->i_ino, 119503ba3782SJens Axboe name, inode->i_sb->s_id); 119603ba3782SJens Axboe if (dentry) { 119703ba3782SJens Axboe spin_unlock(&dentry->d_lock); 119803ba3782SJens Axboe dput(dentry); 119903ba3782SJens Axboe } 120003ba3782SJens Axboe } 120103ba3782SJens Axboe } 120203ba3782SJens Axboe 120303ba3782SJens Axboe /** 120403ba3782SJens Axboe * __mark_inode_dirty - internal function 120503ba3782SJens Axboe * @inode: inode to mark 120603ba3782SJens Axboe * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) 120703ba3782SJens Axboe * Mark an inode as dirty. Callers should use mark_inode_dirty or 120803ba3782SJens Axboe * mark_inode_dirty_sync. 120903ba3782SJens Axboe * 121003ba3782SJens Axboe * Put the inode on the super block's dirty list. 121103ba3782SJens Axboe * 121203ba3782SJens Axboe * CAREFUL! We mark it dirty unconditionally, but move it onto the 121303ba3782SJens Axboe * dirty list only if it is hashed or if it refers to a blockdev. 121403ba3782SJens Axboe * If it was not hashed, it will never be added to the dirty list 121503ba3782SJens Axboe * even if it is later hashed, as it will have been marked dirty already. 121603ba3782SJens Axboe * 121703ba3782SJens Axboe * In short, make sure you hash any inodes _before_ you start marking 121803ba3782SJens Axboe * them dirty. 121903ba3782SJens Axboe * 122003ba3782SJens Axboe * Note that for blockdevs, inode->dirtied_when represents the dirtying time of 122103ba3782SJens Axboe * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of 122203ba3782SJens Axboe * the kernel-internal blockdev inode represents the dirtying time of the 122303ba3782SJens Axboe * blockdev's pages. This is why for I_DIRTY_PAGES we always use 122403ba3782SJens Axboe * page->mapping->host, so the page-dirtying time is recorded in the internal 122503ba3782SJens Axboe * blockdev inode. 122603ba3782SJens Axboe */ 12270ae45f63STheodore Ts'o #define I_DIRTY_INODE (I_DIRTY_SYNC | I_DIRTY_DATASYNC) 122803ba3782SJens Axboe void __mark_inode_dirty(struct inode *inode, int flags) 122903ba3782SJens Axboe { 123003ba3782SJens Axboe struct super_block *sb = inode->i_sb; 1231253c34e9SArtem Bityutskiy struct backing_dev_info *bdi = NULL; 12320ae45f63STheodore Ts'o int dirtytime; 12330ae45f63STheodore Ts'o 12340ae45f63STheodore Ts'o trace_writeback_mark_inode_dirty(inode, flags); 123503ba3782SJens Axboe 123603ba3782SJens Axboe /* 123703ba3782SJens Axboe * Don't do this for I_DIRTY_PAGES - that doesn't actually 123803ba3782SJens Axboe * dirty the inode itself 123903ba3782SJens Axboe */ 12400ae45f63STheodore Ts'o if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC | I_DIRTY_TIME)) { 12419fb0a7daSTejun Heo trace_writeback_dirty_inode_start(inode, flags); 12429fb0a7daSTejun Heo 124303ba3782SJens Axboe if (sb->s_op->dirty_inode) 1244aa385729SChristoph Hellwig sb->s_op->dirty_inode(inode, flags); 12459fb0a7daSTejun Heo 12469fb0a7daSTejun Heo trace_writeback_dirty_inode(inode, flags); 124703ba3782SJens Axboe } 12480ae45f63STheodore Ts'o if (flags & I_DIRTY_INODE) 12490ae45f63STheodore Ts'o flags &= ~I_DIRTY_TIME; 12500ae45f63STheodore Ts'o dirtytime = flags & I_DIRTY_TIME; 125103ba3782SJens Axboe 125203ba3782SJens Axboe /* 12539c6ac78eSTejun Heo * Paired with smp_mb() in __writeback_single_inode() for the 12549c6ac78eSTejun Heo * following lockless i_state test. See there for details. 125503ba3782SJens Axboe */ 125603ba3782SJens Axboe smp_mb(); 125703ba3782SJens Axboe 12580ae45f63STheodore Ts'o if (((inode->i_state & flags) == flags) || 12590ae45f63STheodore Ts'o (dirtytime && (inode->i_state & I_DIRTY_INODE))) 126003ba3782SJens Axboe return; 126103ba3782SJens Axboe 126203ba3782SJens Axboe if (unlikely(block_dump)) 126303ba3782SJens Axboe block_dump___mark_inode_dirty(inode); 126403ba3782SJens Axboe 1265250df6edSDave Chinner spin_lock(&inode->i_lock); 12660ae45f63STheodore Ts'o if (dirtytime && (inode->i_state & I_DIRTY_INODE)) 12670ae45f63STheodore Ts'o goto out_unlock_inode; 126803ba3782SJens Axboe if ((inode->i_state & flags) != flags) { 126903ba3782SJens Axboe const int was_dirty = inode->i_state & I_DIRTY; 127003ba3782SJens Axboe 127152ebea74STejun Heo inode_attach_wb(inode, NULL); 127252ebea74STejun Heo 12730ae45f63STheodore Ts'o if (flags & I_DIRTY_INODE) 12740ae45f63STheodore Ts'o inode->i_state &= ~I_DIRTY_TIME; 127503ba3782SJens Axboe inode->i_state |= flags; 127603ba3782SJens Axboe 127703ba3782SJens Axboe /* 127803ba3782SJens Axboe * If the inode is being synced, just update its dirty state. 127903ba3782SJens Axboe * The unlocker will place the inode on the appropriate 128003ba3782SJens Axboe * superblock list, based upon its state. 128103ba3782SJens Axboe */ 128203ba3782SJens Axboe if (inode->i_state & I_SYNC) 1283250df6edSDave Chinner goto out_unlock_inode; 128403ba3782SJens Axboe 128503ba3782SJens Axboe /* 128603ba3782SJens Axboe * Only add valid (hashed) inodes to the superblock's 128703ba3782SJens Axboe * dirty list. Add blockdev inodes as well. 128803ba3782SJens Axboe */ 128903ba3782SJens Axboe if (!S_ISBLK(inode->i_mode)) { 12901d3382cbSAl Viro if (inode_unhashed(inode)) 1291250df6edSDave Chinner goto out_unlock_inode; 129203ba3782SJens Axboe } 1293a4ffdde6SAl Viro if (inode->i_state & I_FREEING) 1294250df6edSDave Chinner goto out_unlock_inode; 129503ba3782SJens Axboe 129603ba3782SJens Axboe /* 129703ba3782SJens Axboe * If the inode was already on b_dirty/b_io/b_more_io, don't 129803ba3782SJens Axboe * reposition it (that would break b_dirty time-ordering). 129903ba3782SJens Axboe */ 130003ba3782SJens Axboe if (!was_dirty) { 1301a66979abSDave Chinner bool wakeup_bdi = false; 1302253c34e9SArtem Bityutskiy bdi = inode_to_bdi(inode); 1303500b067cSJens Axboe 1304146d7009SJunxiao Bi spin_unlock(&inode->i_lock); 1305146d7009SJunxiao Bi spin_lock(&bdi->wb.list_lock); 1306253c34e9SArtem Bityutskiy if (bdi_cap_writeback_dirty(bdi)) { 13074452226eSTejun Heo WARN(!test_bit(WB_registered, &bdi->wb.state), 1308253c34e9SArtem Bityutskiy "bdi-%s not registered\n", bdi->name); 1309253c34e9SArtem Bityutskiy 1310253c34e9SArtem Bityutskiy /* 1311253c34e9SArtem Bityutskiy * If this is the first dirty inode for this 1312253c34e9SArtem Bityutskiy * bdi, we have to wake-up the corresponding 1313253c34e9SArtem Bityutskiy * bdi thread to make sure background 1314253c34e9SArtem Bityutskiy * write-back happens later. 1315253c34e9SArtem Bityutskiy */ 1316253c34e9SArtem Bityutskiy if (!wb_has_dirty_io(&bdi->wb)) 1317253c34e9SArtem Bityutskiy wakeup_bdi = true; 1318500b067cSJens Axboe } 131903ba3782SJens Axboe 132003ba3782SJens Axboe inode->dirtied_when = jiffies; 1321a2f48706STheodore Ts'o if (dirtytime) 1322a2f48706STheodore Ts'o inode->dirtied_time_when = jiffies; 1323a2f48706STheodore Ts'o if (inode->i_state & (I_DIRTY_INODE | I_DIRTY_PAGES)) 1324a2f48706STheodore Ts'o list_move(&inode->i_wb_list, &bdi->wb.b_dirty); 1325a2f48706STheodore Ts'o else 1326a2f48706STheodore Ts'o list_move(&inode->i_wb_list, 1327a2f48706STheodore Ts'o &bdi->wb.b_dirty_time); 1328f758eeabSChristoph Hellwig spin_unlock(&bdi->wb.list_lock); 13290ae45f63STheodore Ts'o trace_writeback_dirty_inode_enqueue(inode); 1330253c34e9SArtem Bityutskiy 1331253c34e9SArtem Bityutskiy if (wakeup_bdi) 1332f0054bb1STejun Heo wb_wakeup_delayed(&bdi->wb); 1333a66979abSDave Chinner return; 1334a66979abSDave Chinner } 1335a66979abSDave Chinner } 1336a66979abSDave Chinner out_unlock_inode: 1337a66979abSDave Chinner spin_unlock(&inode->i_lock); 1338a66979abSDave Chinner 133903ba3782SJens Axboe } 134003ba3782SJens Axboe EXPORT_SYMBOL(__mark_inode_dirty); 134103ba3782SJens Axboe 1342b6e51316SJens Axboe static void wait_sb_inodes(struct super_block *sb) 134366f3b8e2SJens Axboe { 134438f21977SNick Piggin struct inode *inode, *old_inode = NULL; 134538f21977SNick Piggin 134603ba3782SJens Axboe /* 134703ba3782SJens Axboe * We need to be protected against the filesystem going from 134803ba3782SJens Axboe * r/o to r/w or vice versa. 134903ba3782SJens Axboe */ 1350b6e51316SJens Axboe WARN_ON(!rwsem_is_locked(&sb->s_umount)); 135103ba3782SJens Axboe 135255fa6091SDave Chinner spin_lock(&inode_sb_list_lock); 135366f3b8e2SJens Axboe 135438f21977SNick Piggin /* 135538f21977SNick Piggin * Data integrity sync. Must wait for all pages under writeback, 135638f21977SNick Piggin * because there may have been pages dirtied before our sync 135738f21977SNick Piggin * call, but which had writeout started before we write it out. 135838f21977SNick Piggin * In which case, the inode may not be on the dirty list, but 135938f21977SNick Piggin * we still have to wait for that writeout. 136038f21977SNick Piggin */ 1361b6e51316SJens Axboe list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { 1362250df6edSDave Chinner struct address_space *mapping = inode->i_mapping; 136338f21977SNick Piggin 1364250df6edSDave Chinner spin_lock(&inode->i_lock); 1365250df6edSDave Chinner if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) || 1366250df6edSDave Chinner (mapping->nrpages == 0)) { 1367250df6edSDave Chinner spin_unlock(&inode->i_lock); 136838f21977SNick Piggin continue; 1369250df6edSDave Chinner } 137038f21977SNick Piggin __iget(inode); 1371250df6edSDave Chinner spin_unlock(&inode->i_lock); 137255fa6091SDave Chinner spin_unlock(&inode_sb_list_lock); 137355fa6091SDave Chinner 137438f21977SNick Piggin /* 137555fa6091SDave Chinner * We hold a reference to 'inode' so it couldn't have been 137655fa6091SDave Chinner * removed from s_inodes list while we dropped the 137755fa6091SDave Chinner * inode_sb_list_lock. We cannot iput the inode now as we can 137855fa6091SDave Chinner * be holding the last reference and we cannot iput it under 137955fa6091SDave Chinner * inode_sb_list_lock. So we keep the reference and iput it 138055fa6091SDave Chinner * later. 138138f21977SNick Piggin */ 138238f21977SNick Piggin iput(old_inode); 138338f21977SNick Piggin old_inode = inode; 138438f21977SNick Piggin 138538f21977SNick Piggin filemap_fdatawait(mapping); 138638f21977SNick Piggin 138738f21977SNick Piggin cond_resched(); 138838f21977SNick Piggin 138955fa6091SDave Chinner spin_lock(&inode_sb_list_lock); 139038f21977SNick Piggin } 139155fa6091SDave Chinner spin_unlock(&inode_sb_list_lock); 139238f21977SNick Piggin iput(old_inode); 139366f3b8e2SJens Axboe } 13941da177e4SLinus Torvalds 1395d8a8559cSJens Axboe /** 13963259f8beSChris Mason * writeback_inodes_sb_nr - writeback dirty inodes from given super_block 1397d8a8559cSJens Axboe * @sb: the superblock 13983259f8beSChris Mason * @nr: the number of pages to write 1399786228abSMarcos Paulo de Souza * @reason: reason why some writeback work initiated 14001da177e4SLinus Torvalds * 1401d8a8559cSJens Axboe * Start writeback on some inodes on this super_block. No guarantees are made 1402d8a8559cSJens Axboe * on how many (if any) will be written, and this function does not wait 14033259f8beSChris Mason * for IO completion of submitted IO. 14041da177e4SLinus Torvalds */ 14050e175a18SCurt Wohlgemuth void writeback_inodes_sb_nr(struct super_block *sb, 14060e175a18SCurt Wohlgemuth unsigned long nr, 14070e175a18SCurt Wohlgemuth enum wb_reason reason) 14081da177e4SLinus Torvalds { 140983ba7b07SChristoph Hellwig DECLARE_COMPLETION_ONSTACK(done); 141083ba7b07SChristoph Hellwig struct wb_writeback_work work = { 14113c4d7165SChristoph Hellwig .sb = sb, 14123c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_NONE, 14136e6938b6SWu Fengguang .tagged_writepages = 1, 141483ba7b07SChristoph Hellwig .done = &done, 14153259f8beSChris Mason .nr_pages = nr, 14160e175a18SCurt Wohlgemuth .reason = reason, 14173c4d7165SChristoph Hellwig }; 14180e3c9a22SJens Axboe 14196eedc701SJan Kara if (sb->s_bdi == &noop_backing_dev_info) 14206eedc701SJan Kara return; 1421cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 1422f0054bb1STejun Heo wb_queue_work(&sb->s_bdi->wb, &work); 142383ba7b07SChristoph Hellwig wait_for_completion(&done); 14241da177e4SLinus Torvalds } 14253259f8beSChris Mason EXPORT_SYMBOL(writeback_inodes_sb_nr); 14263259f8beSChris Mason 14273259f8beSChris Mason /** 14283259f8beSChris Mason * writeback_inodes_sb - writeback dirty inodes from given super_block 14293259f8beSChris Mason * @sb: the superblock 1430786228abSMarcos Paulo de Souza * @reason: reason why some writeback work was initiated 14313259f8beSChris Mason * 14323259f8beSChris Mason * Start writeback on some inodes on this super_block. No guarantees are made 14333259f8beSChris Mason * on how many (if any) will be written, and this function does not wait 14343259f8beSChris Mason * for IO completion of submitted IO. 14353259f8beSChris Mason */ 14360e175a18SCurt Wohlgemuth void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) 14373259f8beSChris Mason { 14380e175a18SCurt Wohlgemuth return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); 14393259f8beSChris Mason } 1440d8a8559cSJens Axboe EXPORT_SYMBOL(writeback_inodes_sb); 1441d8a8559cSJens Axboe 1442d8a8559cSJens Axboe /** 144310ee27a0SMiao Xie * try_to_writeback_inodes_sb_nr - try to start writeback if none underway 14443259f8beSChris Mason * @sb: the superblock 14453259f8beSChris Mason * @nr: the number of pages to write 144610ee27a0SMiao Xie * @reason: the reason of writeback 14473259f8beSChris Mason * 144810ee27a0SMiao Xie * Invoke writeback_inodes_sb_nr if no writeback is currently underway. 14493259f8beSChris Mason * Returns 1 if writeback was started, 0 if not. 14503259f8beSChris Mason */ 145110ee27a0SMiao Xie int try_to_writeback_inodes_sb_nr(struct super_block *sb, 14520e175a18SCurt Wohlgemuth unsigned long nr, 14530e175a18SCurt Wohlgemuth enum wb_reason reason) 14543259f8beSChris Mason { 145510ee27a0SMiao Xie if (writeback_in_progress(sb->s_bdi)) 145610ee27a0SMiao Xie return 1; 145710ee27a0SMiao Xie 145810ee27a0SMiao Xie if (!down_read_trylock(&sb->s_umount)) 145910ee27a0SMiao Xie return 0; 146010ee27a0SMiao Xie 14610e175a18SCurt Wohlgemuth writeback_inodes_sb_nr(sb, nr, reason); 14623259f8beSChris Mason up_read(&sb->s_umount); 14633259f8beSChris Mason return 1; 14643259f8beSChris Mason } 146510ee27a0SMiao Xie EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr); 146610ee27a0SMiao Xie 146710ee27a0SMiao Xie /** 146810ee27a0SMiao Xie * try_to_writeback_inodes_sb - try to start writeback if none underway 146910ee27a0SMiao Xie * @sb: the superblock 147010ee27a0SMiao Xie * @reason: reason why some writeback work was initiated 147110ee27a0SMiao Xie * 147210ee27a0SMiao Xie * Implement by try_to_writeback_inodes_sb_nr() 147310ee27a0SMiao Xie * Returns 1 if writeback was started, 0 if not. 147410ee27a0SMiao Xie */ 147510ee27a0SMiao Xie int try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) 147610ee27a0SMiao Xie { 147710ee27a0SMiao Xie return try_to_writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); 147810ee27a0SMiao Xie } 147910ee27a0SMiao Xie EXPORT_SYMBOL(try_to_writeback_inodes_sb); 14803259f8beSChris Mason 14813259f8beSChris Mason /** 1482d8a8559cSJens Axboe * sync_inodes_sb - sync sb inode pages 1483d8a8559cSJens Axboe * @sb: the superblock 1484d8a8559cSJens Axboe * 1485d8a8559cSJens Axboe * This function writes and waits on any dirty inode belonging to this 14860dc83bd3SJan Kara * super_block. 1487d8a8559cSJens Axboe */ 14880dc83bd3SJan Kara void sync_inodes_sb(struct super_block *sb) 1489d8a8559cSJens Axboe { 149083ba7b07SChristoph Hellwig DECLARE_COMPLETION_ONSTACK(done); 149183ba7b07SChristoph Hellwig struct wb_writeback_work work = { 14923c4d7165SChristoph Hellwig .sb = sb, 14933c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_ALL, 14943c4d7165SChristoph Hellwig .nr_pages = LONG_MAX, 14953c4d7165SChristoph Hellwig .range_cyclic = 0, 149683ba7b07SChristoph Hellwig .done = &done, 14970e175a18SCurt Wohlgemuth .reason = WB_REASON_SYNC, 14987747bd4bSDave Chinner .for_sync = 1, 14993c4d7165SChristoph Hellwig }; 15003c4d7165SChristoph Hellwig 15016eedc701SJan Kara /* Nothing to do? */ 15026eedc701SJan Kara if (sb->s_bdi == &noop_backing_dev_info) 15036eedc701SJan Kara return; 1504cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 1505cf37e972SChristoph Hellwig 1506f0054bb1STejun Heo wb_queue_work(&sb->s_bdi->wb, &work); 150783ba7b07SChristoph Hellwig wait_for_completion(&done); 150883ba7b07SChristoph Hellwig 1509b6e51316SJens Axboe wait_sb_inodes(sb); 1510d8a8559cSJens Axboe } 1511d8a8559cSJens Axboe EXPORT_SYMBOL(sync_inodes_sb); 15121da177e4SLinus Torvalds 15131da177e4SLinus Torvalds /** 15141da177e4SLinus Torvalds * write_inode_now - write an inode to disk 15151da177e4SLinus Torvalds * @inode: inode to write to disk 15161da177e4SLinus Torvalds * @sync: whether the write should be synchronous or not 15171da177e4SLinus Torvalds * 15187f04c26dSAndrea Arcangeli * This function commits an inode to disk immediately if it is dirty. This is 15197f04c26dSAndrea Arcangeli * primarily needed by knfsd. 15207f04c26dSAndrea Arcangeli * 15217f04c26dSAndrea Arcangeli * The caller must either have a ref on the inode or must have set I_WILL_FREE. 15221da177e4SLinus Torvalds */ 15231da177e4SLinus Torvalds int write_inode_now(struct inode *inode, int sync) 15241da177e4SLinus Torvalds { 1525f758eeabSChristoph Hellwig struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; 15261da177e4SLinus Torvalds struct writeback_control wbc = { 15271da177e4SLinus Torvalds .nr_to_write = LONG_MAX, 152818914b18SMike Galbraith .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, 1529111ebb6eSOGAWA Hirofumi .range_start = 0, 1530111ebb6eSOGAWA Hirofumi .range_end = LLONG_MAX, 15311da177e4SLinus Torvalds }; 15321da177e4SLinus Torvalds 15331da177e4SLinus Torvalds if (!mapping_cap_writeback_dirty(inode->i_mapping)) 153449364ce2SAndrew Morton wbc.nr_to_write = 0; 15351da177e4SLinus Torvalds 15361da177e4SLinus Torvalds might_sleep(); 15374f8ad655SJan Kara return writeback_single_inode(inode, wb, &wbc); 15381da177e4SLinus Torvalds } 15391da177e4SLinus Torvalds EXPORT_SYMBOL(write_inode_now); 15401da177e4SLinus Torvalds 15411da177e4SLinus Torvalds /** 15421da177e4SLinus Torvalds * sync_inode - write an inode and its pages to disk. 15431da177e4SLinus Torvalds * @inode: the inode to sync 15441da177e4SLinus Torvalds * @wbc: controls the writeback mode 15451da177e4SLinus Torvalds * 15461da177e4SLinus Torvalds * sync_inode() will write an inode and its pages to disk. It will also 15471da177e4SLinus Torvalds * correctly update the inode on its superblock's dirty inode lists and will 15481da177e4SLinus Torvalds * update inode->i_state. 15491da177e4SLinus Torvalds * 15501da177e4SLinus Torvalds * The caller must have a ref on the inode. 15511da177e4SLinus Torvalds */ 15521da177e4SLinus Torvalds int sync_inode(struct inode *inode, struct writeback_control *wbc) 15531da177e4SLinus Torvalds { 15544f8ad655SJan Kara return writeback_single_inode(inode, &inode_to_bdi(inode)->wb, wbc); 15551da177e4SLinus Torvalds } 15561da177e4SLinus Torvalds EXPORT_SYMBOL(sync_inode); 1557c3765016SChristoph Hellwig 1558c3765016SChristoph Hellwig /** 1559c691b9d9SAndrew Morton * sync_inode_metadata - write an inode to disk 1560c3765016SChristoph Hellwig * @inode: the inode to sync 1561c3765016SChristoph Hellwig * @wait: wait for I/O to complete. 1562c3765016SChristoph Hellwig * 1563c691b9d9SAndrew Morton * Write an inode to disk and adjust its dirty state after completion. 1564c3765016SChristoph Hellwig * 1565c3765016SChristoph Hellwig * Note: only writes the actual inode, no associated data or other metadata. 1566c3765016SChristoph Hellwig */ 1567c3765016SChristoph Hellwig int sync_inode_metadata(struct inode *inode, int wait) 1568c3765016SChristoph Hellwig { 1569c3765016SChristoph Hellwig struct writeback_control wbc = { 1570c3765016SChristoph Hellwig .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE, 1571c3765016SChristoph Hellwig .nr_to_write = 0, /* metadata-only */ 1572c3765016SChristoph Hellwig }; 1573c3765016SChristoph Hellwig 1574c3765016SChristoph Hellwig return sync_inode(inode, &wbc); 1575c3765016SChristoph Hellwig } 1576c3765016SChristoph Hellwig EXPORT_SYMBOL(sync_inode_metadata); 1577