11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * fs/fs-writeback.c 31da177e4SLinus Torvalds * 41da177e4SLinus Torvalds * Copyright (C) 2002, Linus Torvalds. 51da177e4SLinus Torvalds * 61da177e4SLinus Torvalds * Contains all the functions related to writing back and waiting 71da177e4SLinus Torvalds * upon dirty inodes against superblocks, and writing back dirty 81da177e4SLinus Torvalds * pages against inodes. ie: data writeback. Writeout of the 91da177e4SLinus Torvalds * inode itself is not handled here. 101da177e4SLinus Torvalds * 11e1f8e874SFrancois Cami * 10Apr2002 Andrew Morton 121da177e4SLinus Torvalds * Split out of fs/inode.c 131da177e4SLinus Torvalds * Additions for address_space-based writeback 141da177e4SLinus Torvalds */ 151da177e4SLinus Torvalds 161da177e4SLinus Torvalds #include <linux/kernel.h> 17630d9c47SPaul Gortmaker #include <linux/export.h> 181da177e4SLinus Torvalds #include <linux/spinlock.h> 195a0e3ad6STejun Heo #include <linux/slab.h> 201da177e4SLinus Torvalds #include <linux/sched.h> 211da177e4SLinus Torvalds #include <linux/fs.h> 221da177e4SLinus Torvalds #include <linux/mm.h> 23bc31b86aSWu Fengguang #include <linux/pagemap.h> 2403ba3782SJens Axboe #include <linux/kthread.h> 251da177e4SLinus Torvalds #include <linux/writeback.h> 261da177e4SLinus Torvalds #include <linux/blkdev.h> 271da177e4SLinus Torvalds #include <linux/backing-dev.h> 28455b2864SDave Chinner #include <linux/tracepoint.h> 29719ea2fbSAl Viro #include <linux/device.h> 3007f3f05cSDavid Howells #include "internal.h" 311da177e4SLinus Torvalds 32d0bceac7SJens Axboe /* 33bc31b86aSWu Fengguang * 4MB minimal write chunk size 34bc31b86aSWu Fengguang */ 35bc31b86aSWu Fengguang #define MIN_WRITEBACK_PAGES (4096UL >> (PAGE_CACHE_SHIFT - 10)) 36bc31b86aSWu Fengguang 37bc31b86aSWu Fengguang /* 38c4a77a6cSJens Axboe * Passed into wb_writeback(), essentially a subset of writeback_control 39c4a77a6cSJens Axboe */ 4083ba7b07SChristoph Hellwig struct wb_writeback_work { 41c4a77a6cSJens Axboe long nr_pages; 42c4a77a6cSJens Axboe struct super_block *sb; 430dc83bd3SJan Kara unsigned long *older_than_this; 44c4a77a6cSJens Axboe enum writeback_sync_modes sync_mode; 456e6938b6SWu Fengguang unsigned int tagged_writepages:1; 4652957fe1SH Hartley Sweeten unsigned int for_kupdate:1; 4752957fe1SH Hartley Sweeten unsigned int range_cyclic:1; 4852957fe1SH Hartley Sweeten unsigned int for_background:1; 497747bd4bSDave Chinner unsigned int for_sync:1; /* sync(2) WB_SYNC_ALL writeback */ 500e175a18SCurt Wohlgemuth enum wb_reason reason; /* why was writeback initiated? */ 51c4a77a6cSJens Axboe 528010c3b6SJens Axboe struct list_head list; /* pending work list */ 5383ba7b07SChristoph Hellwig struct completion *done; /* set if the caller waits */ 5403ba3782SJens Axboe }; 5503ba3782SJens Axboe 56f11b00f3SAdrian Bunk /** 57f11b00f3SAdrian Bunk * writeback_in_progress - determine whether there is writeback in progress 58f11b00f3SAdrian Bunk * @bdi: the device's backing_dev_info structure. 59f11b00f3SAdrian Bunk * 6003ba3782SJens Axboe * Determine whether there is writeback waiting to be handled against a 6103ba3782SJens Axboe * backing device. 62f11b00f3SAdrian Bunk */ 63f11b00f3SAdrian Bunk int writeback_in_progress(struct backing_dev_info *bdi) 64f11b00f3SAdrian Bunk { 6581d73a32SJan Kara return test_bit(BDI_writeback_running, &bdi->state); 66f11b00f3SAdrian Bunk } 6700d4e736STheodore Ts'o EXPORT_SYMBOL(writeback_in_progress); 68f11b00f3SAdrian Bunk 69692ebd17SJan Kara static inline struct backing_dev_info *inode_to_bdi(struct inode *inode) 70692ebd17SJan Kara { 71692ebd17SJan Kara struct super_block *sb = inode->i_sb; 72692ebd17SJan Kara 73a8855990SJan Kara if (sb_is_blkdev_sb(sb)) 74aaead25bSChristoph Hellwig return inode->i_mapping->backing_dev_info; 75aaead25bSChristoph Hellwig 76692ebd17SJan Kara return sb->s_bdi; 77692ebd17SJan Kara } 78692ebd17SJan Kara 797ccf19a8SNick Piggin static inline struct inode *wb_inode(struct list_head *head) 807ccf19a8SNick Piggin { 817ccf19a8SNick Piggin return list_entry(head, struct inode, i_wb_list); 827ccf19a8SNick Piggin } 837ccf19a8SNick Piggin 8415eb77a0SWu Fengguang /* 8515eb77a0SWu Fengguang * Include the creation of the trace points after defining the 8615eb77a0SWu Fengguang * wb_writeback_work structure and inline functions so that the definition 8715eb77a0SWu Fengguang * remains local to this file. 8815eb77a0SWu Fengguang */ 8915eb77a0SWu Fengguang #define CREATE_TRACE_POINTS 9015eb77a0SWu Fengguang #include <trace/events/writeback.h> 9115eb77a0SWu Fengguang 92774016b2SSteven Whitehouse EXPORT_TRACEPOINT_SYMBOL_GPL(wbc_writepage); 93774016b2SSteven Whitehouse 945acda9d1SJan Kara static void bdi_wakeup_thread(struct backing_dev_info *bdi) 955acda9d1SJan Kara { 965acda9d1SJan Kara spin_lock_bh(&bdi->wb_lock); 975acda9d1SJan Kara if (test_bit(BDI_registered, &bdi->state)) 985acda9d1SJan Kara mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0); 995acda9d1SJan Kara spin_unlock_bh(&bdi->wb_lock); 1005acda9d1SJan Kara } 1015acda9d1SJan Kara 1026585027aSJan Kara static void bdi_queue_work(struct backing_dev_info *bdi, 1036585027aSJan Kara struct wb_writeback_work *work) 1046585027aSJan Kara { 1056585027aSJan Kara trace_writeback_queue(bdi, work); 1066585027aSJan Kara 1076585027aSJan Kara spin_lock_bh(&bdi->wb_lock); 1085acda9d1SJan Kara if (!test_bit(BDI_registered, &bdi->state)) { 1095acda9d1SJan Kara if (work->done) 1105acda9d1SJan Kara complete(work->done); 1115acda9d1SJan Kara goto out_unlock; 1125acda9d1SJan Kara } 1136585027aSJan Kara list_add_tail(&work->list, &bdi->work_list); 114839a8e86STejun Heo mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0); 1155acda9d1SJan Kara out_unlock: 1165acda9d1SJan Kara spin_unlock_bh(&bdi->wb_lock); 11703ba3782SJens Axboe } 1181da177e4SLinus Torvalds 11983ba7b07SChristoph Hellwig static void 12083ba7b07SChristoph Hellwig __bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, 1210e175a18SCurt Wohlgemuth bool range_cyclic, enum wb_reason reason) 1221da177e4SLinus Torvalds { 12383ba7b07SChristoph Hellwig struct wb_writeback_work *work; 12403ba3782SJens Axboe 125bcddc3f0SJens Axboe /* 126bcddc3f0SJens Axboe * This is WB_SYNC_NONE writeback, so if allocation fails just 127bcddc3f0SJens Axboe * wakeup the thread for old dirty data writeback 128bcddc3f0SJens Axboe */ 12983ba7b07SChristoph Hellwig work = kzalloc(sizeof(*work), GFP_ATOMIC); 13083ba7b07SChristoph Hellwig if (!work) { 131455b2864SDave Chinner trace_writeback_nowork(bdi); 1325acda9d1SJan Kara bdi_wakeup_thread(bdi); 13383ba7b07SChristoph Hellwig return; 13483ba7b07SChristoph Hellwig } 13583ba7b07SChristoph Hellwig 13683ba7b07SChristoph Hellwig work->sync_mode = WB_SYNC_NONE; 13783ba7b07SChristoph Hellwig work->nr_pages = nr_pages; 13883ba7b07SChristoph Hellwig work->range_cyclic = range_cyclic; 1390e175a18SCurt Wohlgemuth work->reason = reason; 14083ba7b07SChristoph Hellwig 141f11fcae8SJens Axboe bdi_queue_work(bdi, work); 14203ba3782SJens Axboe } 143b6e51316SJens Axboe 144b6e51316SJens Axboe /** 145b6e51316SJens Axboe * bdi_start_writeback - start writeback 146b6e51316SJens Axboe * @bdi: the backing device to write from 147b6e51316SJens Axboe * @nr_pages: the number of pages to write 148786228abSMarcos Paulo de Souza * @reason: reason why some writeback work was initiated 149b6e51316SJens Axboe * 150b6e51316SJens Axboe * Description: 151b6e51316SJens Axboe * This does WB_SYNC_NONE opportunistic writeback. The IO is only 15225985edcSLucas De Marchi * started when this function returns, we make no guarantees on 1530e3c9a22SJens Axboe * completion. Caller need not hold sb s_umount semaphore. 154b6e51316SJens Axboe * 155b6e51316SJens Axboe */ 1560e175a18SCurt Wohlgemuth void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, 1570e175a18SCurt Wohlgemuth enum wb_reason reason) 158b6e51316SJens Axboe { 1590e175a18SCurt Wohlgemuth __bdi_start_writeback(bdi, nr_pages, true, reason); 160d3ddec76SWu Fengguang } 161d3ddec76SWu Fengguang 162c5444198SChristoph Hellwig /** 163c5444198SChristoph Hellwig * bdi_start_background_writeback - start background writeback 164c5444198SChristoph Hellwig * @bdi: the backing device to write from 165c5444198SChristoph Hellwig * 166c5444198SChristoph Hellwig * Description: 1676585027aSJan Kara * This makes sure WB_SYNC_NONE background writeback happens. When 1686585027aSJan Kara * this function returns, it is only guaranteed that for given BDI 1696585027aSJan Kara * some IO is happening if we are over background dirty threshold. 1706585027aSJan Kara * Caller need not hold sb s_umount semaphore. 171c5444198SChristoph Hellwig */ 172c5444198SChristoph Hellwig void bdi_start_background_writeback(struct backing_dev_info *bdi) 173c5444198SChristoph Hellwig { 1746585027aSJan Kara /* 1756585027aSJan Kara * We just wake up the flusher thread. It will perform background 1766585027aSJan Kara * writeback as soon as there is no other work to do. 1776585027aSJan Kara */ 17871927e84SWu Fengguang trace_writeback_wake_background(bdi); 1795acda9d1SJan Kara bdi_wakeup_thread(bdi); 1801da177e4SLinus Torvalds } 1811da177e4SLinus Torvalds 1821da177e4SLinus Torvalds /* 183a66979abSDave Chinner * Remove the inode from the writeback list it is on. 184a66979abSDave Chinner */ 185a66979abSDave Chinner void inode_wb_list_del(struct inode *inode) 186a66979abSDave Chinner { 187f758eeabSChristoph Hellwig struct backing_dev_info *bdi = inode_to_bdi(inode); 188a66979abSDave Chinner 189f758eeabSChristoph Hellwig spin_lock(&bdi->wb.list_lock); 190f758eeabSChristoph Hellwig list_del_init(&inode->i_wb_list); 191f758eeabSChristoph Hellwig spin_unlock(&bdi->wb.list_lock); 192f758eeabSChristoph Hellwig } 193a66979abSDave Chinner 194a66979abSDave Chinner /* 1956610a0bcSAndrew Morton * Redirty an inode: set its when-it-was dirtied timestamp and move it to the 1966610a0bcSAndrew Morton * furthest end of its superblock's dirty-inode list. 1976610a0bcSAndrew Morton * 1986610a0bcSAndrew Morton * Before stamping the inode's ->dirtied_when, we check to see whether it is 19966f3b8e2SJens Axboe * already the most-recently-dirtied inode on the b_dirty list. If that is 2006610a0bcSAndrew Morton * the case then the inode must have been redirtied while it was being written 2016610a0bcSAndrew Morton * out and we don't reset its dirtied_when. 2026610a0bcSAndrew Morton */ 203f758eeabSChristoph Hellwig static void redirty_tail(struct inode *inode, struct bdi_writeback *wb) 2046610a0bcSAndrew Morton { 205f758eeabSChristoph Hellwig assert_spin_locked(&wb->list_lock); 20603ba3782SJens Axboe if (!list_empty(&wb->b_dirty)) { 20766f3b8e2SJens Axboe struct inode *tail; 2086610a0bcSAndrew Morton 2097ccf19a8SNick Piggin tail = wb_inode(wb->b_dirty.next); 21066f3b8e2SJens Axboe if (time_before(inode->dirtied_when, tail->dirtied_when)) 2116610a0bcSAndrew Morton inode->dirtied_when = jiffies; 2126610a0bcSAndrew Morton } 2137ccf19a8SNick Piggin list_move(&inode->i_wb_list, &wb->b_dirty); 2146610a0bcSAndrew Morton } 2156610a0bcSAndrew Morton 2166610a0bcSAndrew Morton /* 21766f3b8e2SJens Axboe * requeue inode for re-scanning after bdi->b_io list is exhausted. 218c986d1e2SAndrew Morton */ 219f758eeabSChristoph Hellwig static void requeue_io(struct inode *inode, struct bdi_writeback *wb) 220c986d1e2SAndrew Morton { 221f758eeabSChristoph Hellwig assert_spin_locked(&wb->list_lock); 2227ccf19a8SNick Piggin list_move(&inode->i_wb_list, &wb->b_more_io); 223c986d1e2SAndrew Morton } 224c986d1e2SAndrew Morton 2251c0eeaf5SJoern Engel static void inode_sync_complete(struct inode *inode) 2261c0eeaf5SJoern Engel { 227365b94aeSJan Kara inode->i_state &= ~I_SYNC; 2284eff96ddSJan Kara /* If inode is clean an unused, put it into LRU now... */ 2294eff96ddSJan Kara inode_add_lru(inode); 230365b94aeSJan Kara /* Waiters must see I_SYNC cleared before being woken up */ 2311c0eeaf5SJoern Engel smp_mb(); 2321c0eeaf5SJoern Engel wake_up_bit(&inode->i_state, __I_SYNC); 2331c0eeaf5SJoern Engel } 2341c0eeaf5SJoern Engel 235d2caa3c5SJeff Layton static bool inode_dirtied_after(struct inode *inode, unsigned long t) 236d2caa3c5SJeff Layton { 237d2caa3c5SJeff Layton bool ret = time_after(inode->dirtied_when, t); 238d2caa3c5SJeff Layton #ifndef CONFIG_64BIT 239d2caa3c5SJeff Layton /* 240d2caa3c5SJeff Layton * For inodes being constantly redirtied, dirtied_when can get stuck. 241d2caa3c5SJeff Layton * It _appears_ to be in the future, but is actually in distant past. 242d2caa3c5SJeff Layton * This test is necessary to prevent such wrapped-around relative times 2435b0830cbSJens Axboe * from permanently stopping the whole bdi writeback. 244d2caa3c5SJeff Layton */ 245d2caa3c5SJeff Layton ret = ret && time_before_eq(inode->dirtied_when, jiffies); 246d2caa3c5SJeff Layton #endif 247d2caa3c5SJeff Layton return ret; 248d2caa3c5SJeff Layton } 249d2caa3c5SJeff Layton 2500ae45f63STheodore Ts'o #define EXPIRE_DIRTY_ATIME 0x0001 2510ae45f63STheodore Ts'o 252c986d1e2SAndrew Morton /* 2530e2f2b23SWang Sheng-Hui * Move expired (dirtied before work->older_than_this) dirty inodes from 254697e6fedSJan Kara * @delaying_queue to @dispatch_queue. 2552c136579SFengguang Wu */ 256e84d0a4fSWu Fengguang static int move_expired_inodes(struct list_head *delaying_queue, 2572c136579SFengguang Wu struct list_head *dispatch_queue, 2580ae45f63STheodore Ts'o int flags, 259ad4e38ddSCurt Wohlgemuth struct wb_writeback_work *work) 2602c136579SFengguang Wu { 2610ae45f63STheodore Ts'o unsigned long *older_than_this = NULL; 2620ae45f63STheodore Ts'o unsigned long expire_time; 2635c03449dSShaohua Li LIST_HEAD(tmp); 2645c03449dSShaohua Li struct list_head *pos, *node; 265cf137307SJens Axboe struct super_block *sb = NULL; 2665c03449dSShaohua Li struct inode *inode; 267cf137307SJens Axboe int do_sb_sort = 0; 268e84d0a4fSWu Fengguang int moved = 0; 2695c03449dSShaohua Li 2700ae45f63STheodore Ts'o if ((flags & EXPIRE_DIRTY_ATIME) == 0) 2710ae45f63STheodore Ts'o older_than_this = work->older_than_this; 2720ae45f63STheodore Ts'o else if ((work->reason == WB_REASON_SYNC) == 0) { 2730ae45f63STheodore Ts'o expire_time = jiffies - (HZ * 86400); 2740ae45f63STheodore Ts'o older_than_this = &expire_time; 2750ae45f63STheodore Ts'o } 2762c136579SFengguang Wu while (!list_empty(delaying_queue)) { 2777ccf19a8SNick Piggin inode = wb_inode(delaying_queue->prev); 2780ae45f63STheodore Ts'o if (older_than_this && 2790ae45f63STheodore Ts'o inode_dirtied_after(inode, *older_than_this)) 2802c136579SFengguang Wu break; 281a8855990SJan Kara list_move(&inode->i_wb_list, &tmp); 282a8855990SJan Kara moved++; 2830ae45f63STheodore Ts'o if (flags & EXPIRE_DIRTY_ATIME) 2840ae45f63STheodore Ts'o set_bit(__I_DIRTY_TIME_EXPIRED, &inode->i_state); 285a8855990SJan Kara if (sb_is_blkdev_sb(inode->i_sb)) 286a8855990SJan Kara continue; 287cf137307SJens Axboe if (sb && sb != inode->i_sb) 288cf137307SJens Axboe do_sb_sort = 1; 289cf137307SJens Axboe sb = inode->i_sb; 2905c03449dSShaohua Li } 2915c03449dSShaohua Li 292cf137307SJens Axboe /* just one sb in list, splice to dispatch_queue and we're done */ 293cf137307SJens Axboe if (!do_sb_sort) { 294cf137307SJens Axboe list_splice(&tmp, dispatch_queue); 295e84d0a4fSWu Fengguang goto out; 296cf137307SJens Axboe } 297cf137307SJens Axboe 2985c03449dSShaohua Li /* Move inodes from one superblock together */ 2995c03449dSShaohua Li while (!list_empty(&tmp)) { 3007ccf19a8SNick Piggin sb = wb_inode(tmp.prev)->i_sb; 3015c03449dSShaohua Li list_for_each_prev_safe(pos, node, &tmp) { 3027ccf19a8SNick Piggin inode = wb_inode(pos); 3035c03449dSShaohua Li if (inode->i_sb == sb) 3047ccf19a8SNick Piggin list_move(&inode->i_wb_list, dispatch_queue); 3052c136579SFengguang Wu } 3062c136579SFengguang Wu } 307e84d0a4fSWu Fengguang out: 308e84d0a4fSWu Fengguang return moved; 3095c03449dSShaohua Li } 3102c136579SFengguang Wu 3112c136579SFengguang Wu /* 3122c136579SFengguang Wu * Queue all expired dirty inodes for io, eldest first. 3134ea879b9SWu Fengguang * Before 3144ea879b9SWu Fengguang * newly dirtied b_dirty b_io b_more_io 3154ea879b9SWu Fengguang * =============> gf edc BA 3164ea879b9SWu Fengguang * After 3174ea879b9SWu Fengguang * newly dirtied b_dirty b_io b_more_io 3184ea879b9SWu Fengguang * =============> g fBAedc 3194ea879b9SWu Fengguang * | 3204ea879b9SWu Fengguang * +--> dequeue for IO 3212c136579SFengguang Wu */ 322ad4e38ddSCurt Wohlgemuth static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work) 3232c136579SFengguang Wu { 324e84d0a4fSWu Fengguang int moved; 3250ae45f63STheodore Ts'o 326f758eeabSChristoph Hellwig assert_spin_locked(&wb->list_lock); 3274ea879b9SWu Fengguang list_splice_init(&wb->b_more_io, &wb->b_io); 3280ae45f63STheodore Ts'o moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, 0, work); 3290ae45f63STheodore Ts'o moved += move_expired_inodes(&wb->b_dirty_time, &wb->b_io, 3300ae45f63STheodore Ts'o EXPIRE_DIRTY_ATIME, work); 331ad4e38ddSCurt Wohlgemuth trace_writeback_queue_io(wb, work, moved); 33266f3b8e2SJens Axboe } 33366f3b8e2SJens Axboe 334a9185b41SChristoph Hellwig static int write_inode(struct inode *inode, struct writeback_control *wbc) 33566f3b8e2SJens Axboe { 3369fb0a7daSTejun Heo int ret; 3379fb0a7daSTejun Heo 3389fb0a7daSTejun Heo if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) { 3399fb0a7daSTejun Heo trace_writeback_write_inode_start(inode, wbc); 3409fb0a7daSTejun Heo ret = inode->i_sb->s_op->write_inode(inode, wbc); 3419fb0a7daSTejun Heo trace_writeback_write_inode(inode, wbc); 3429fb0a7daSTejun Heo return ret; 3439fb0a7daSTejun Heo } 34403ba3782SJens Axboe return 0; 34566f3b8e2SJens Axboe } 34608d8e974SFengguang Wu 3472c136579SFengguang Wu /* 348169ebd90SJan Kara * Wait for writeback on an inode to complete. Called with i_lock held. 349169ebd90SJan Kara * Caller must make sure inode cannot go away when we drop i_lock. 35001c03194SChristoph Hellwig */ 351169ebd90SJan Kara static void __inode_wait_for_writeback(struct inode *inode) 352169ebd90SJan Kara __releases(inode->i_lock) 353169ebd90SJan Kara __acquires(inode->i_lock) 35401c03194SChristoph Hellwig { 35501c03194SChristoph Hellwig DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); 35601c03194SChristoph Hellwig wait_queue_head_t *wqh; 35701c03194SChristoph Hellwig 35801c03194SChristoph Hellwig wqh = bit_waitqueue(&inode->i_state, __I_SYNC); 35958a9d3d8SRichard Kennedy while (inode->i_state & I_SYNC) { 360250df6edSDave Chinner spin_unlock(&inode->i_lock); 36174316201SNeilBrown __wait_on_bit(wqh, &wq, bit_wait, 36274316201SNeilBrown TASK_UNINTERRUPTIBLE); 363250df6edSDave Chinner spin_lock(&inode->i_lock); 36458a9d3d8SRichard Kennedy } 36501c03194SChristoph Hellwig } 36601c03194SChristoph Hellwig 36701c03194SChristoph Hellwig /* 368169ebd90SJan Kara * Wait for writeback on an inode to complete. Caller must have inode pinned. 369169ebd90SJan Kara */ 370169ebd90SJan Kara void inode_wait_for_writeback(struct inode *inode) 371169ebd90SJan Kara { 372169ebd90SJan Kara spin_lock(&inode->i_lock); 373169ebd90SJan Kara __inode_wait_for_writeback(inode); 374169ebd90SJan Kara spin_unlock(&inode->i_lock); 375169ebd90SJan Kara } 376169ebd90SJan Kara 377169ebd90SJan Kara /* 378169ebd90SJan Kara * Sleep until I_SYNC is cleared. This function must be called with i_lock 379169ebd90SJan Kara * held and drops it. It is aimed for callers not holding any inode reference 380169ebd90SJan Kara * so once i_lock is dropped, inode can go away. 381169ebd90SJan Kara */ 382169ebd90SJan Kara static void inode_sleep_on_writeback(struct inode *inode) 383169ebd90SJan Kara __releases(inode->i_lock) 384169ebd90SJan Kara { 385169ebd90SJan Kara DEFINE_WAIT(wait); 386169ebd90SJan Kara wait_queue_head_t *wqh = bit_waitqueue(&inode->i_state, __I_SYNC); 387169ebd90SJan Kara int sleep; 388169ebd90SJan Kara 389169ebd90SJan Kara prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); 390169ebd90SJan Kara sleep = inode->i_state & I_SYNC; 391169ebd90SJan Kara spin_unlock(&inode->i_lock); 392169ebd90SJan Kara if (sleep) 393169ebd90SJan Kara schedule(); 394169ebd90SJan Kara finish_wait(wqh, &wait); 395169ebd90SJan Kara } 396169ebd90SJan Kara 397169ebd90SJan Kara /* 398ccb26b5aSJan Kara * Find proper writeback list for the inode depending on its current state and 399ccb26b5aSJan Kara * possibly also change of its state while we were doing writeback. Here we 400ccb26b5aSJan Kara * handle things such as livelock prevention or fairness of writeback among 401ccb26b5aSJan Kara * inodes. This function can be called only by flusher thread - noone else 402ccb26b5aSJan Kara * processes all inodes in writeback lists and requeueing inodes behind flusher 403ccb26b5aSJan Kara * thread's back can have unexpected consequences. 404ccb26b5aSJan Kara */ 405ccb26b5aSJan Kara static void requeue_inode(struct inode *inode, struct bdi_writeback *wb, 406ccb26b5aSJan Kara struct writeback_control *wbc) 407ccb26b5aSJan Kara { 408ccb26b5aSJan Kara if (inode->i_state & I_FREEING) 409ccb26b5aSJan Kara return; 410ccb26b5aSJan Kara 411ccb26b5aSJan Kara /* 412ccb26b5aSJan Kara * Sync livelock prevention. Each inode is tagged and synced in one 413ccb26b5aSJan Kara * shot. If still dirty, it will be redirty_tail()'ed below. Update 414ccb26b5aSJan Kara * the dirty time to prevent enqueue and sync it again. 415ccb26b5aSJan Kara */ 416ccb26b5aSJan Kara if ((inode->i_state & I_DIRTY) && 417ccb26b5aSJan Kara (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)) 418ccb26b5aSJan Kara inode->dirtied_when = jiffies; 419ccb26b5aSJan Kara 4204f8ad655SJan Kara if (wbc->pages_skipped) { 4214f8ad655SJan Kara /* 4224f8ad655SJan Kara * writeback is not making progress due to locked 4234f8ad655SJan Kara * buffers. Skip this inode for now. 4244f8ad655SJan Kara */ 4254f8ad655SJan Kara redirty_tail(inode, wb); 4264f8ad655SJan Kara return; 4274f8ad655SJan Kara } 4284f8ad655SJan Kara 429ccb26b5aSJan Kara if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) { 430ccb26b5aSJan Kara /* 431ccb26b5aSJan Kara * We didn't write back all the pages. nfs_writepages() 432ccb26b5aSJan Kara * sometimes bales out without doing anything. 433ccb26b5aSJan Kara */ 434ccb26b5aSJan Kara if (wbc->nr_to_write <= 0) { 435ccb26b5aSJan Kara /* Slice used up. Queue for next turn. */ 436ccb26b5aSJan Kara requeue_io(inode, wb); 437ccb26b5aSJan Kara } else { 438ccb26b5aSJan Kara /* 439ccb26b5aSJan Kara * Writeback blocked by something other than 440ccb26b5aSJan Kara * congestion. Delay the inode for some time to 441ccb26b5aSJan Kara * avoid spinning on the CPU (100% iowait) 442ccb26b5aSJan Kara * retrying writeback of the dirty page/inode 443ccb26b5aSJan Kara * that cannot be performed immediately. 444ccb26b5aSJan Kara */ 445ccb26b5aSJan Kara redirty_tail(inode, wb); 446ccb26b5aSJan Kara } 447ccb26b5aSJan Kara } else if (inode->i_state & I_DIRTY) { 448ccb26b5aSJan Kara /* 449ccb26b5aSJan Kara * Filesystems can dirty the inode during writeback operations, 450ccb26b5aSJan Kara * such as delayed allocation during submission or metadata 451ccb26b5aSJan Kara * updates after data IO completion. 452ccb26b5aSJan Kara */ 453ccb26b5aSJan Kara redirty_tail(inode, wb); 4540ae45f63STheodore Ts'o } else if (inode->i_state & I_DIRTY_TIME) { 4550ae45f63STheodore Ts'o list_move(&inode->i_wb_list, &wb->b_dirty_time); 456ccb26b5aSJan Kara } else { 457ccb26b5aSJan Kara /* The inode is clean. Remove from writeback lists. */ 458ccb26b5aSJan Kara list_del_init(&inode->i_wb_list); 459ccb26b5aSJan Kara } 460ccb26b5aSJan Kara } 461ccb26b5aSJan Kara 462ccb26b5aSJan Kara /* 4634f8ad655SJan Kara * Write out an inode and its dirty pages. Do not update the writeback list 4644f8ad655SJan Kara * linkage. That is left to the caller. The caller is also responsible for 4654f8ad655SJan Kara * setting I_SYNC flag and calling inode_sync_complete() to clear it. 4661da177e4SLinus Torvalds */ 4671da177e4SLinus Torvalds static int 468cd8ed2a4SYan Hong __writeback_single_inode(struct inode *inode, struct writeback_control *wbc) 4691da177e4SLinus Torvalds { 4701da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 471251d6a47SWu Fengguang long nr_to_write = wbc->nr_to_write; 47201c03194SChristoph Hellwig unsigned dirty; 4731da177e4SLinus Torvalds int ret; 4741da177e4SLinus Torvalds 4754f8ad655SJan Kara WARN_ON(!(inode->i_state & I_SYNC)); 4761da177e4SLinus Torvalds 4779fb0a7daSTejun Heo trace_writeback_single_inode_start(inode, wbc, nr_to_write); 4789fb0a7daSTejun Heo 4791da177e4SLinus Torvalds ret = do_writepages(mapping, wbc); 4801da177e4SLinus Torvalds 48126821ed4SChristoph Hellwig /* 48226821ed4SChristoph Hellwig * Make sure to wait on the data before writing out the metadata. 48326821ed4SChristoph Hellwig * This is important for filesystems that modify metadata on data 4847747bd4bSDave Chinner * I/O completion. We don't do it for sync(2) writeback because it has a 4857747bd4bSDave Chinner * separate, external IO completion path and ->sync_fs for guaranteeing 4867747bd4bSDave Chinner * inode metadata is written back correctly. 48726821ed4SChristoph Hellwig */ 4887747bd4bSDave Chinner if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) { 48926821ed4SChristoph Hellwig int err = filemap_fdatawait(mapping); 4901da177e4SLinus Torvalds if (ret == 0) 4911da177e4SLinus Torvalds ret = err; 4921da177e4SLinus Torvalds } 4931da177e4SLinus Torvalds 4945547e8aaSDmitry Monakhov /* 4955547e8aaSDmitry Monakhov * Some filesystems may redirty the inode during the writeback 4965547e8aaSDmitry Monakhov * due to delalloc, clear dirty metadata flags right before 4975547e8aaSDmitry Monakhov * write_inode() 4985547e8aaSDmitry Monakhov */ 499250df6edSDave Chinner spin_lock(&inode->i_lock); 5009c6ac78eSTejun Heo 5015547e8aaSDmitry Monakhov dirty = inode->i_state & I_DIRTY; 5020ae45f63STheodore Ts'o if (((dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) && 5030ae45f63STheodore Ts'o (inode->i_state & I_DIRTY_TIME)) || 5040ae45f63STheodore Ts'o (inode->i_state & I_DIRTY_TIME_EXPIRED)) { 5050ae45f63STheodore Ts'o dirty |= I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED; 5060ae45f63STheodore Ts'o trace_writeback_lazytime(inode); 5070ae45f63STheodore Ts'o } 5080ae45f63STheodore Ts'o inode->i_state &= ~dirty; 5099c6ac78eSTejun Heo 5109c6ac78eSTejun Heo /* 5119c6ac78eSTejun Heo * Paired with smp_mb() in __mark_inode_dirty(). This allows 5129c6ac78eSTejun Heo * __mark_inode_dirty() to test i_state without grabbing i_lock - 5139c6ac78eSTejun Heo * either they see the I_DIRTY bits cleared or we see the dirtied 5149c6ac78eSTejun Heo * inode. 5159c6ac78eSTejun Heo * 5169c6ac78eSTejun Heo * I_DIRTY_PAGES is always cleared together above even if @mapping 5179c6ac78eSTejun Heo * still has dirty pages. The flag is reinstated after smp_mb() if 5189c6ac78eSTejun Heo * necessary. This guarantees that either __mark_inode_dirty() 5199c6ac78eSTejun Heo * sees clear I_DIRTY_PAGES or we see PAGECACHE_TAG_DIRTY. 5209c6ac78eSTejun Heo */ 5219c6ac78eSTejun Heo smp_mb(); 5229c6ac78eSTejun Heo 5239c6ac78eSTejun Heo if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) 5249c6ac78eSTejun Heo inode->i_state |= I_DIRTY_PAGES; 5259c6ac78eSTejun Heo 526250df6edSDave Chinner spin_unlock(&inode->i_lock); 5279c6ac78eSTejun Heo 5280ae45f63STheodore Ts'o if (dirty & I_DIRTY_TIME) 5290ae45f63STheodore Ts'o mark_inode_dirty_sync(inode); 53026821ed4SChristoph Hellwig /* Don't write the inode if only I_DIRTY_PAGES was set */ 5310ae45f63STheodore Ts'o if (dirty & ~I_DIRTY_PAGES) { 532a9185b41SChristoph Hellwig int err = write_inode(inode, wbc); 5331da177e4SLinus Torvalds if (ret == 0) 5341da177e4SLinus Torvalds ret = err; 5351da177e4SLinus Torvalds } 5364f8ad655SJan Kara trace_writeback_single_inode(inode, wbc, nr_to_write); 5374f8ad655SJan Kara return ret; 5384f8ad655SJan Kara } 5394f8ad655SJan Kara 5404f8ad655SJan Kara /* 5414f8ad655SJan Kara * Write out an inode's dirty pages. Either the caller has an active reference 5424f8ad655SJan Kara * on the inode or the inode has I_WILL_FREE set. 5434f8ad655SJan Kara * 5444f8ad655SJan Kara * This function is designed to be called for writing back one inode which 5454f8ad655SJan Kara * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode() 5464f8ad655SJan Kara * and does more profound writeback list handling in writeback_sb_inodes(). 5474f8ad655SJan Kara */ 5484f8ad655SJan Kara static int 5494f8ad655SJan Kara writeback_single_inode(struct inode *inode, struct bdi_writeback *wb, 5504f8ad655SJan Kara struct writeback_control *wbc) 5514f8ad655SJan Kara { 5524f8ad655SJan Kara int ret = 0; 5534f8ad655SJan Kara 5544f8ad655SJan Kara spin_lock(&inode->i_lock); 5554f8ad655SJan Kara if (!atomic_read(&inode->i_count)) 5564f8ad655SJan Kara WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); 5574f8ad655SJan Kara else 5584f8ad655SJan Kara WARN_ON(inode->i_state & I_WILL_FREE); 5594f8ad655SJan Kara 5604f8ad655SJan Kara if (inode->i_state & I_SYNC) { 5614f8ad655SJan Kara if (wbc->sync_mode != WB_SYNC_ALL) 5624f8ad655SJan Kara goto out; 5634f8ad655SJan Kara /* 564169ebd90SJan Kara * It's a data-integrity sync. We must wait. Since callers hold 565169ebd90SJan Kara * inode reference or inode has I_WILL_FREE set, it cannot go 566169ebd90SJan Kara * away under us. 5674f8ad655SJan Kara */ 568169ebd90SJan Kara __inode_wait_for_writeback(inode); 5694f8ad655SJan Kara } 5704f8ad655SJan Kara WARN_ON(inode->i_state & I_SYNC); 5714f8ad655SJan Kara /* 572f9b0e058SJan Kara * Skip inode if it is clean and we have no outstanding writeback in 573f9b0e058SJan Kara * WB_SYNC_ALL mode. We don't want to mess with writeback lists in this 574f9b0e058SJan Kara * function since flusher thread may be doing for example sync in 575f9b0e058SJan Kara * parallel and if we move the inode, it could get skipped. So here we 576f9b0e058SJan Kara * make sure inode is on some writeback list and leave it there unless 577f9b0e058SJan Kara * we have completely cleaned the inode. 5784f8ad655SJan Kara */ 5790ae45f63STheodore Ts'o if (!(inode->i_state & I_DIRTY_ALL) && 580f9b0e058SJan Kara (wbc->sync_mode != WB_SYNC_ALL || 581f9b0e058SJan Kara !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK))) 5824f8ad655SJan Kara goto out; 5834f8ad655SJan Kara inode->i_state |= I_SYNC; 5844f8ad655SJan Kara spin_unlock(&inode->i_lock); 5854f8ad655SJan Kara 586cd8ed2a4SYan Hong ret = __writeback_single_inode(inode, wbc); 5871da177e4SLinus Torvalds 588f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 589250df6edSDave Chinner spin_lock(&inode->i_lock); 5904f8ad655SJan Kara /* 5914f8ad655SJan Kara * If inode is clean, remove it from writeback lists. Otherwise don't 5924f8ad655SJan Kara * touch it. See comment above for explanation. 5934f8ad655SJan Kara */ 5940ae45f63STheodore Ts'o if (!(inode->i_state & I_DIRTY_ALL)) 5954f8ad655SJan Kara list_del_init(&inode->i_wb_list); 5964f8ad655SJan Kara spin_unlock(&wb->list_lock); 5971c0eeaf5SJoern Engel inode_sync_complete(inode); 5984f8ad655SJan Kara out: 5994f8ad655SJan Kara spin_unlock(&inode->i_lock); 6001da177e4SLinus Torvalds return ret; 6011da177e4SLinus Torvalds } 6021da177e4SLinus Torvalds 6031a12d8bdSWu Fengguang static long writeback_chunk_size(struct backing_dev_info *bdi, 6041a12d8bdSWu Fengguang struct wb_writeback_work *work) 605d46db3d5SWu Fengguang { 606d46db3d5SWu Fengguang long pages; 607d46db3d5SWu Fengguang 608d46db3d5SWu Fengguang /* 609d46db3d5SWu Fengguang * WB_SYNC_ALL mode does livelock avoidance by syncing dirty 610d46db3d5SWu Fengguang * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX 611d46db3d5SWu Fengguang * here avoids calling into writeback_inodes_wb() more than once. 612d46db3d5SWu Fengguang * 613d46db3d5SWu Fengguang * The intended call sequence for WB_SYNC_ALL writeback is: 614d46db3d5SWu Fengguang * 615d46db3d5SWu Fengguang * wb_writeback() 616d46db3d5SWu Fengguang * writeback_sb_inodes() <== called only once 617d46db3d5SWu Fengguang * write_cache_pages() <== called once for each inode 618d46db3d5SWu Fengguang * (quickly) tag currently dirty pages 619d46db3d5SWu Fengguang * (maybe slowly) sync all tagged pages 620d46db3d5SWu Fengguang */ 621d46db3d5SWu Fengguang if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages) 622d46db3d5SWu Fengguang pages = LONG_MAX; 6231a12d8bdSWu Fengguang else { 6241a12d8bdSWu Fengguang pages = min(bdi->avg_write_bandwidth / 2, 6251a12d8bdSWu Fengguang global_dirty_limit / DIRTY_SCOPE); 6261a12d8bdSWu Fengguang pages = min(pages, work->nr_pages); 6271a12d8bdSWu Fengguang pages = round_down(pages + MIN_WRITEBACK_PAGES, 6281a12d8bdSWu Fengguang MIN_WRITEBACK_PAGES); 6291a12d8bdSWu Fengguang } 630d46db3d5SWu Fengguang 631d46db3d5SWu Fengguang return pages; 632d46db3d5SWu Fengguang } 633d46db3d5SWu Fengguang 63403ba3782SJens Axboe /* 635f11c9c5cSEdward Shishkin * Write a portion of b_io inodes which belong to @sb. 636edadfb10SChristoph Hellwig * 637d46db3d5SWu Fengguang * Return the number of pages and/or inodes written. 638f11c9c5cSEdward Shishkin */ 639d46db3d5SWu Fengguang static long writeback_sb_inodes(struct super_block *sb, 640d46db3d5SWu Fengguang struct bdi_writeback *wb, 641d46db3d5SWu Fengguang struct wb_writeback_work *work) 64203ba3782SJens Axboe { 643d46db3d5SWu Fengguang struct writeback_control wbc = { 644d46db3d5SWu Fengguang .sync_mode = work->sync_mode, 645d46db3d5SWu Fengguang .tagged_writepages = work->tagged_writepages, 646d46db3d5SWu Fengguang .for_kupdate = work->for_kupdate, 647d46db3d5SWu Fengguang .for_background = work->for_background, 6487747bd4bSDave Chinner .for_sync = work->for_sync, 649d46db3d5SWu Fengguang .range_cyclic = work->range_cyclic, 650d46db3d5SWu Fengguang .range_start = 0, 651d46db3d5SWu Fengguang .range_end = LLONG_MAX, 652d46db3d5SWu Fengguang }; 653d46db3d5SWu Fengguang unsigned long start_time = jiffies; 654d46db3d5SWu Fengguang long write_chunk; 655d46db3d5SWu Fengguang long wrote = 0; /* count both pages and inodes */ 656d46db3d5SWu Fengguang 65703ba3782SJens Axboe while (!list_empty(&wb->b_io)) { 6587ccf19a8SNick Piggin struct inode *inode = wb_inode(wb->b_io.prev); 659edadfb10SChristoph Hellwig 660edadfb10SChristoph Hellwig if (inode->i_sb != sb) { 661d46db3d5SWu Fengguang if (work->sb) { 662edadfb10SChristoph Hellwig /* 663edadfb10SChristoph Hellwig * We only want to write back data for this 664edadfb10SChristoph Hellwig * superblock, move all inodes not belonging 665edadfb10SChristoph Hellwig * to it back onto the dirty list. 666edadfb10SChristoph Hellwig */ 667f758eeabSChristoph Hellwig redirty_tail(inode, wb); 66866f3b8e2SJens Axboe continue; 66966f3b8e2SJens Axboe } 670edadfb10SChristoph Hellwig 671edadfb10SChristoph Hellwig /* 672edadfb10SChristoph Hellwig * The inode belongs to a different superblock. 673edadfb10SChristoph Hellwig * Bounce back to the caller to unpin this and 674edadfb10SChristoph Hellwig * pin the next superblock. 675edadfb10SChristoph Hellwig */ 676d46db3d5SWu Fengguang break; 677edadfb10SChristoph Hellwig } 678edadfb10SChristoph Hellwig 6799843b76aSChristoph Hellwig /* 680331cbdeeSWanpeng Li * Don't bother with new inodes or inodes being freed, first 681331cbdeeSWanpeng Li * kind does not need periodic writeout yet, and for the latter 6829843b76aSChristoph Hellwig * kind writeout is handled by the freer. 6839843b76aSChristoph Hellwig */ 684250df6edSDave Chinner spin_lock(&inode->i_lock); 6859843b76aSChristoph Hellwig if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { 686250df6edSDave Chinner spin_unlock(&inode->i_lock); 687fcc5c222SWu Fengguang redirty_tail(inode, wb); 6887ef0d737SNick Piggin continue; 6897ef0d737SNick Piggin } 690cc1676d9SJan Kara if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) { 691cc1676d9SJan Kara /* 692cc1676d9SJan Kara * If this inode is locked for writeback and we are not 693cc1676d9SJan Kara * doing writeback-for-data-integrity, move it to 694cc1676d9SJan Kara * b_more_io so that writeback can proceed with the 695cc1676d9SJan Kara * other inodes on s_io. 696cc1676d9SJan Kara * 697cc1676d9SJan Kara * We'll have another go at writing back this inode 698cc1676d9SJan Kara * when we completed a full scan of b_io. 699cc1676d9SJan Kara */ 700cc1676d9SJan Kara spin_unlock(&inode->i_lock); 701cc1676d9SJan Kara requeue_io(inode, wb); 702cc1676d9SJan Kara trace_writeback_sb_inodes_requeue(inode); 703cc1676d9SJan Kara continue; 704cc1676d9SJan Kara } 705f0d07b7fSJan Kara spin_unlock(&wb->list_lock); 706f0d07b7fSJan Kara 7074f8ad655SJan Kara /* 7084f8ad655SJan Kara * We already requeued the inode if it had I_SYNC set and we 7094f8ad655SJan Kara * are doing WB_SYNC_NONE writeback. So this catches only the 7104f8ad655SJan Kara * WB_SYNC_ALL case. 7114f8ad655SJan Kara */ 712169ebd90SJan Kara if (inode->i_state & I_SYNC) { 713169ebd90SJan Kara /* Wait for I_SYNC. This function drops i_lock... */ 714169ebd90SJan Kara inode_sleep_on_writeback(inode); 715169ebd90SJan Kara /* Inode may be gone, start again */ 716ead188f9SJan Kara spin_lock(&wb->list_lock); 717169ebd90SJan Kara continue; 718169ebd90SJan Kara } 7194f8ad655SJan Kara inode->i_state |= I_SYNC; 7204f8ad655SJan Kara spin_unlock(&inode->i_lock); 721169ebd90SJan Kara 7221a12d8bdSWu Fengguang write_chunk = writeback_chunk_size(wb->bdi, work); 723d46db3d5SWu Fengguang wbc.nr_to_write = write_chunk; 724d46db3d5SWu Fengguang wbc.pages_skipped = 0; 725250df6edSDave Chinner 726169ebd90SJan Kara /* 727169ebd90SJan Kara * We use I_SYNC to pin the inode in memory. While it is set 728169ebd90SJan Kara * evict_inode() will wait so the inode cannot be freed. 729169ebd90SJan Kara */ 730cd8ed2a4SYan Hong __writeback_single_inode(inode, &wbc); 731d46db3d5SWu Fengguang 732d46db3d5SWu Fengguang work->nr_pages -= write_chunk - wbc.nr_to_write; 733d46db3d5SWu Fengguang wrote += write_chunk - wbc.nr_to_write; 7344f8ad655SJan Kara spin_lock(&wb->list_lock); 7354f8ad655SJan Kara spin_lock(&inode->i_lock); 7360ae45f63STheodore Ts'o if (!(inode->i_state & I_DIRTY_ALL)) 737d46db3d5SWu Fengguang wrote++; 7384f8ad655SJan Kara requeue_inode(inode, wb, &wbc); 7394f8ad655SJan Kara inode_sync_complete(inode); 7400f1b1fd8SDave Chinner spin_unlock(&inode->i_lock); 741169ebd90SJan Kara cond_resched_lock(&wb->list_lock); 742d46db3d5SWu Fengguang /* 743d46db3d5SWu Fengguang * bail out to wb_writeback() often enough to check 744d46db3d5SWu Fengguang * background threshold and other termination conditions. 745d46db3d5SWu Fengguang */ 746d46db3d5SWu Fengguang if (wrote) { 747d46db3d5SWu Fengguang if (time_is_before_jiffies(start_time + HZ / 10UL)) 748d46db3d5SWu Fengguang break; 749d46db3d5SWu Fengguang if (work->nr_pages <= 0) 750d46db3d5SWu Fengguang break; 7511da177e4SLinus Torvalds } 7528bc3be27SFengguang Wu } 753d46db3d5SWu Fengguang return wrote; 754f11c9c5cSEdward Shishkin } 75538f21977SNick Piggin 756d46db3d5SWu Fengguang static long __writeback_inodes_wb(struct bdi_writeback *wb, 757d46db3d5SWu Fengguang struct wb_writeback_work *work) 758f11c9c5cSEdward Shishkin { 759d46db3d5SWu Fengguang unsigned long start_time = jiffies; 760d46db3d5SWu Fengguang long wrote = 0; 761f11c9c5cSEdward Shishkin 762f11c9c5cSEdward Shishkin while (!list_empty(&wb->b_io)) { 7637ccf19a8SNick Piggin struct inode *inode = wb_inode(wb->b_io.prev); 764f11c9c5cSEdward Shishkin struct super_block *sb = inode->i_sb; 765f11c9c5cSEdward Shishkin 76612ad3ab6SDave Chinner if (!grab_super_passive(sb)) { 7670e995816SWu Fengguang /* 7680e995816SWu Fengguang * grab_super_passive() may fail consistently due to 7690e995816SWu Fengguang * s_umount being grabbed by someone else. Don't use 7700e995816SWu Fengguang * requeue_io() to avoid busy retrying the inode/sb. 7710e995816SWu Fengguang */ 7720e995816SWu Fengguang redirty_tail(inode, wb); 773d19de7edSChristoph Hellwig continue; 774334132aeSChristoph Hellwig } 775d46db3d5SWu Fengguang wrote += writeback_sb_inodes(sb, wb, work); 776d19de7edSChristoph Hellwig drop_super(sb); 777f11c9c5cSEdward Shishkin 778d46db3d5SWu Fengguang /* refer to the same tests at the end of writeback_sb_inodes */ 779d46db3d5SWu Fengguang if (wrote) { 780d46db3d5SWu Fengguang if (time_is_before_jiffies(start_time + HZ / 10UL)) 781d46db3d5SWu Fengguang break; 782d46db3d5SWu Fengguang if (work->nr_pages <= 0) 783f11c9c5cSEdward Shishkin break; 784f11c9c5cSEdward Shishkin } 785d46db3d5SWu Fengguang } 78666f3b8e2SJens Axboe /* Leave any unwritten inodes on b_io */ 787d46db3d5SWu Fengguang return wrote; 78866f3b8e2SJens Axboe } 78966f3b8e2SJens Axboe 7907d9f073bSWanpeng Li static long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages, 7910e175a18SCurt Wohlgemuth enum wb_reason reason) 792edadfb10SChristoph Hellwig { 793d46db3d5SWu Fengguang struct wb_writeback_work work = { 794d46db3d5SWu Fengguang .nr_pages = nr_pages, 795d46db3d5SWu Fengguang .sync_mode = WB_SYNC_NONE, 796d46db3d5SWu Fengguang .range_cyclic = 1, 7970e175a18SCurt Wohlgemuth .reason = reason, 798d46db3d5SWu Fengguang }; 799edadfb10SChristoph Hellwig 800f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 801424b351fSWu Fengguang if (list_empty(&wb->b_io)) 802ad4e38ddSCurt Wohlgemuth queue_io(wb, &work); 803d46db3d5SWu Fengguang __writeback_inodes_wb(wb, &work); 804f758eeabSChristoph Hellwig spin_unlock(&wb->list_lock); 805edadfb10SChristoph Hellwig 806d46db3d5SWu Fengguang return nr_pages - work.nr_pages; 80766f3b8e2SJens Axboe } 80866f3b8e2SJens Axboe 809b00949aaSWu Fengguang static bool over_bground_thresh(struct backing_dev_info *bdi) 81003ba3782SJens Axboe { 81103ba3782SJens Axboe unsigned long background_thresh, dirty_thresh; 81203ba3782SJens Axboe 81316c4042fSWu Fengguang global_dirty_limits(&background_thresh, &dirty_thresh); 81403ba3782SJens Axboe 815b00949aaSWu Fengguang if (global_page_state(NR_FILE_DIRTY) + 816b00949aaSWu Fengguang global_page_state(NR_UNSTABLE_NFS) > background_thresh) 817b00949aaSWu Fengguang return true; 818b00949aaSWu Fengguang 819b00949aaSWu Fengguang if (bdi_stat(bdi, BDI_RECLAIMABLE) > 820b00949aaSWu Fengguang bdi_dirty_limit(bdi, background_thresh)) 821b00949aaSWu Fengguang return true; 822b00949aaSWu Fengguang 823b00949aaSWu Fengguang return false; 82403ba3782SJens Axboe } 82503ba3782SJens Axboe 82603ba3782SJens Axboe /* 827e98be2d5SWu Fengguang * Called under wb->list_lock. If there are multiple wb per bdi, 828e98be2d5SWu Fengguang * only the flusher working on the first wb should do it. 829e98be2d5SWu Fengguang */ 830e98be2d5SWu Fengguang static void wb_update_bandwidth(struct bdi_writeback *wb, 831e98be2d5SWu Fengguang unsigned long start_time) 832e98be2d5SWu Fengguang { 833af6a3113SWu Fengguang __bdi_update_bandwidth(wb->bdi, 0, 0, 0, 0, 0, start_time); 834e98be2d5SWu Fengguang } 835e98be2d5SWu Fengguang 836e98be2d5SWu Fengguang /* 83703ba3782SJens Axboe * Explicit flushing or periodic writeback of "old" data. 83803ba3782SJens Axboe * 83903ba3782SJens Axboe * Define "old": the first time one of an inode's pages is dirtied, we mark the 84003ba3782SJens Axboe * dirtying-time in the inode's address_space. So this periodic writeback code 84103ba3782SJens Axboe * just walks the superblock inode list, writing back any inodes which are 84203ba3782SJens Axboe * older than a specific point in time. 84303ba3782SJens Axboe * 84403ba3782SJens Axboe * Try to run once per dirty_writeback_interval. But if a writeback event 84503ba3782SJens Axboe * takes longer than a dirty_writeback_interval interval, then leave a 84603ba3782SJens Axboe * one-second gap. 84703ba3782SJens Axboe * 84803ba3782SJens Axboe * older_than_this takes precedence over nr_to_write. So we'll only write back 84903ba3782SJens Axboe * all dirty pages if they are all attached to "old" mappings. 85003ba3782SJens Axboe */ 851c4a77a6cSJens Axboe static long wb_writeback(struct bdi_writeback *wb, 85283ba7b07SChristoph Hellwig struct wb_writeback_work *work) 85303ba3782SJens Axboe { 854e98be2d5SWu Fengguang unsigned long wb_start = jiffies; 855d46db3d5SWu Fengguang long nr_pages = work->nr_pages; 8560dc83bd3SJan Kara unsigned long oldest_jif; 857a5989bdcSJan Kara struct inode *inode; 858d46db3d5SWu Fengguang long progress; 85903ba3782SJens Axboe 8600dc83bd3SJan Kara oldest_jif = jiffies; 8610dc83bd3SJan Kara work->older_than_this = &oldest_jif; 86203ba3782SJens Axboe 863e8dfc305SWu Fengguang spin_lock(&wb->list_lock); 86403ba3782SJens Axboe for (;;) { 86503ba3782SJens Axboe /* 866d3ddec76SWu Fengguang * Stop writeback when nr_pages has been consumed 86703ba3782SJens Axboe */ 86883ba7b07SChristoph Hellwig if (work->nr_pages <= 0) 86903ba3782SJens Axboe break; 87003ba3782SJens Axboe 87103ba3782SJens Axboe /* 872aa373cf5SJan Kara * Background writeout and kupdate-style writeback may 873aa373cf5SJan Kara * run forever. Stop them if there is other work to do 874aa373cf5SJan Kara * so that e.g. sync can proceed. They'll be restarted 875aa373cf5SJan Kara * after the other works are all done. 876aa373cf5SJan Kara */ 877aa373cf5SJan Kara if ((work->for_background || work->for_kupdate) && 878aa373cf5SJan Kara !list_empty(&wb->bdi->work_list)) 879aa373cf5SJan Kara break; 880aa373cf5SJan Kara 881aa373cf5SJan Kara /* 882d3ddec76SWu Fengguang * For background writeout, stop when we are below the 883d3ddec76SWu Fengguang * background dirty threshold 88403ba3782SJens Axboe */ 885b00949aaSWu Fengguang if (work->for_background && !over_bground_thresh(wb->bdi)) 88603ba3782SJens Axboe break; 88703ba3782SJens Axboe 8881bc36b64SJan Kara /* 8891bc36b64SJan Kara * Kupdate and background works are special and we want to 8901bc36b64SJan Kara * include all inodes that need writing. Livelock avoidance is 8911bc36b64SJan Kara * handled by these works yielding to any other work so we are 8921bc36b64SJan Kara * safe. 8931bc36b64SJan Kara */ 894ba9aa839SWu Fengguang if (work->for_kupdate) { 8950dc83bd3SJan Kara oldest_jif = jiffies - 896ba9aa839SWu Fengguang msecs_to_jiffies(dirty_expire_interval * 10); 8971bc36b64SJan Kara } else if (work->for_background) 8980dc83bd3SJan Kara oldest_jif = jiffies; 899028c2dd1SDave Chinner 900d46db3d5SWu Fengguang trace_writeback_start(wb->bdi, work); 901e8dfc305SWu Fengguang if (list_empty(&wb->b_io)) 902ad4e38ddSCurt Wohlgemuth queue_io(wb, work); 90383ba7b07SChristoph Hellwig if (work->sb) 904d46db3d5SWu Fengguang progress = writeback_sb_inodes(work->sb, wb, work); 905edadfb10SChristoph Hellwig else 906d46db3d5SWu Fengguang progress = __writeback_inodes_wb(wb, work); 907d46db3d5SWu Fengguang trace_writeback_written(wb->bdi, work); 908028c2dd1SDave Chinner 909e98be2d5SWu Fengguang wb_update_bandwidth(wb, wb_start); 91003ba3782SJens Axboe 91103ba3782SJens Axboe /* 91271fd05a8SJens Axboe * Did we write something? Try for more 913e6fb6da2SWu Fengguang * 914e6fb6da2SWu Fengguang * Dirty inodes are moved to b_io for writeback in batches. 915e6fb6da2SWu Fengguang * The completion of the current batch does not necessarily 916e6fb6da2SWu Fengguang * mean the overall work is done. So we keep looping as long 917e6fb6da2SWu Fengguang * as made some progress on cleaning pages or inodes. 91871fd05a8SJens Axboe */ 919d46db3d5SWu Fengguang if (progress) 92003ba3782SJens Axboe continue; 921a5989bdcSJan Kara /* 922e6fb6da2SWu Fengguang * No more inodes for IO, bail 923a5989bdcSJan Kara */ 924b7a2441fSWu Fengguang if (list_empty(&wb->b_more_io)) 92503ba3782SJens Axboe break; 92603ba3782SJens Axboe /* 9278010c3b6SJens Axboe * Nothing written. Wait for some inode to 9288010c3b6SJens Axboe * become available for writeback. Otherwise 9298010c3b6SJens Axboe * we'll just busyloop. 93003ba3782SJens Axboe */ 93103ba3782SJens Axboe if (!list_empty(&wb->b_more_io)) { 932d46db3d5SWu Fengguang trace_writeback_wait(wb->bdi, work); 93303ba3782SJens Axboe inode = wb_inode(wb->b_more_io.prev); 934250df6edSDave Chinner spin_lock(&inode->i_lock); 935f0d07b7fSJan Kara spin_unlock(&wb->list_lock); 936169ebd90SJan Kara /* This function drops i_lock... */ 937169ebd90SJan Kara inode_sleep_on_writeback(inode); 938f0d07b7fSJan Kara spin_lock(&wb->list_lock); 93903ba3782SJens Axboe } 94003ba3782SJens Axboe } 941e8dfc305SWu Fengguang spin_unlock(&wb->list_lock); 94203ba3782SJens Axboe 943d46db3d5SWu Fengguang return nr_pages - work->nr_pages; 94403ba3782SJens Axboe } 94503ba3782SJens Axboe 94603ba3782SJens Axboe /* 94783ba7b07SChristoph Hellwig * Return the next wb_writeback_work struct that hasn't been processed yet. 94803ba3782SJens Axboe */ 94983ba7b07SChristoph Hellwig static struct wb_writeback_work * 95008852b6dSMinchan Kim get_next_work_item(struct backing_dev_info *bdi) 95103ba3782SJens Axboe { 95283ba7b07SChristoph Hellwig struct wb_writeback_work *work = NULL; 95303ba3782SJens Axboe 9546467716aSArtem Bityutskiy spin_lock_bh(&bdi->wb_lock); 95583ba7b07SChristoph Hellwig if (!list_empty(&bdi->work_list)) { 95683ba7b07SChristoph Hellwig work = list_entry(bdi->work_list.next, 95783ba7b07SChristoph Hellwig struct wb_writeback_work, list); 95883ba7b07SChristoph Hellwig list_del_init(&work->list); 95903ba3782SJens Axboe } 9606467716aSArtem Bityutskiy spin_unlock_bh(&bdi->wb_lock); 96183ba7b07SChristoph Hellwig return work; 96203ba3782SJens Axboe } 96303ba3782SJens Axboe 964cdf01dd5SLinus Torvalds /* 965cdf01dd5SLinus Torvalds * Add in the number of potentially dirty inodes, because each inode 966cdf01dd5SLinus Torvalds * write can dirty pagecache in the underlying blockdev. 967cdf01dd5SLinus Torvalds */ 968cdf01dd5SLinus Torvalds static unsigned long get_nr_dirty_pages(void) 969cdf01dd5SLinus Torvalds { 970cdf01dd5SLinus Torvalds return global_page_state(NR_FILE_DIRTY) + 971cdf01dd5SLinus Torvalds global_page_state(NR_UNSTABLE_NFS) + 972cdf01dd5SLinus Torvalds get_nr_dirty_inodes(); 973cdf01dd5SLinus Torvalds } 974cdf01dd5SLinus Torvalds 9756585027aSJan Kara static long wb_check_background_flush(struct bdi_writeback *wb) 9766585027aSJan Kara { 977b00949aaSWu Fengguang if (over_bground_thresh(wb->bdi)) { 9786585027aSJan Kara 9796585027aSJan Kara struct wb_writeback_work work = { 9806585027aSJan Kara .nr_pages = LONG_MAX, 9816585027aSJan Kara .sync_mode = WB_SYNC_NONE, 9826585027aSJan Kara .for_background = 1, 9836585027aSJan Kara .range_cyclic = 1, 9840e175a18SCurt Wohlgemuth .reason = WB_REASON_BACKGROUND, 9856585027aSJan Kara }; 9866585027aSJan Kara 9876585027aSJan Kara return wb_writeback(wb, &work); 9886585027aSJan Kara } 9896585027aSJan Kara 9906585027aSJan Kara return 0; 9916585027aSJan Kara } 9926585027aSJan Kara 99303ba3782SJens Axboe static long wb_check_old_data_flush(struct bdi_writeback *wb) 99403ba3782SJens Axboe { 99503ba3782SJens Axboe unsigned long expired; 99603ba3782SJens Axboe long nr_pages; 99703ba3782SJens Axboe 99869b62d01SJens Axboe /* 99969b62d01SJens Axboe * When set to zero, disable periodic writeback 100069b62d01SJens Axboe */ 100169b62d01SJens Axboe if (!dirty_writeback_interval) 100269b62d01SJens Axboe return 0; 100369b62d01SJens Axboe 100403ba3782SJens Axboe expired = wb->last_old_flush + 100503ba3782SJens Axboe msecs_to_jiffies(dirty_writeback_interval * 10); 100603ba3782SJens Axboe if (time_before(jiffies, expired)) 100703ba3782SJens Axboe return 0; 100803ba3782SJens Axboe 100903ba3782SJens Axboe wb->last_old_flush = jiffies; 1010cdf01dd5SLinus Torvalds nr_pages = get_nr_dirty_pages(); 101103ba3782SJens Axboe 1012c4a77a6cSJens Axboe if (nr_pages) { 101383ba7b07SChristoph Hellwig struct wb_writeback_work work = { 1014c4a77a6cSJens Axboe .nr_pages = nr_pages, 1015c4a77a6cSJens Axboe .sync_mode = WB_SYNC_NONE, 1016c4a77a6cSJens Axboe .for_kupdate = 1, 1017c4a77a6cSJens Axboe .range_cyclic = 1, 10180e175a18SCurt Wohlgemuth .reason = WB_REASON_PERIODIC, 1019c4a77a6cSJens Axboe }; 1020c4a77a6cSJens Axboe 102183ba7b07SChristoph Hellwig return wb_writeback(wb, &work); 1022c4a77a6cSJens Axboe } 102303ba3782SJens Axboe 102403ba3782SJens Axboe return 0; 102503ba3782SJens Axboe } 102603ba3782SJens Axboe 102703ba3782SJens Axboe /* 102803ba3782SJens Axboe * Retrieve work items and do the writeback they describe 102903ba3782SJens Axboe */ 103025d130baSWanpeng Li static long wb_do_writeback(struct bdi_writeback *wb) 103103ba3782SJens Axboe { 103203ba3782SJens Axboe struct backing_dev_info *bdi = wb->bdi; 103383ba7b07SChristoph Hellwig struct wb_writeback_work *work; 1034c4a77a6cSJens Axboe long wrote = 0; 103503ba3782SJens Axboe 103681d73a32SJan Kara set_bit(BDI_writeback_running, &wb->bdi->state); 103708852b6dSMinchan Kim while ((work = get_next_work_item(bdi)) != NULL) { 103883ba7b07SChristoph Hellwig 1039455b2864SDave Chinner trace_writeback_exec(bdi, work); 1040455b2864SDave Chinner 104183ba7b07SChristoph Hellwig wrote += wb_writeback(wb, work); 104203ba3782SJens Axboe 104303ba3782SJens Axboe /* 104483ba7b07SChristoph Hellwig * Notify the caller of completion if this is a synchronous 104583ba7b07SChristoph Hellwig * work item, otherwise just free it. 104603ba3782SJens Axboe */ 104783ba7b07SChristoph Hellwig if (work->done) 104883ba7b07SChristoph Hellwig complete(work->done); 104983ba7b07SChristoph Hellwig else 105083ba7b07SChristoph Hellwig kfree(work); 105103ba3782SJens Axboe } 105203ba3782SJens Axboe 105303ba3782SJens Axboe /* 105403ba3782SJens Axboe * Check for periodic writeback, kupdated() style 105503ba3782SJens Axboe */ 105603ba3782SJens Axboe wrote += wb_check_old_data_flush(wb); 10576585027aSJan Kara wrote += wb_check_background_flush(wb); 105881d73a32SJan Kara clear_bit(BDI_writeback_running, &wb->bdi->state); 105903ba3782SJens Axboe 106003ba3782SJens Axboe return wrote; 106103ba3782SJens Axboe } 106203ba3782SJens Axboe 106303ba3782SJens Axboe /* 106403ba3782SJens Axboe * Handle writeback of dirty data for the device backed by this bdi. Also 1065839a8e86STejun Heo * reschedules periodically and does kupdated style flushing. 106603ba3782SJens Axboe */ 1067839a8e86STejun Heo void bdi_writeback_workfn(struct work_struct *work) 106803ba3782SJens Axboe { 1069839a8e86STejun Heo struct bdi_writeback *wb = container_of(to_delayed_work(work), 1070839a8e86STejun Heo struct bdi_writeback, dwork); 107108243900SChristoph Hellwig struct backing_dev_info *bdi = wb->bdi; 107203ba3782SJens Axboe long pages_written; 107303ba3782SJens Axboe 1074ef3b1019STejun Heo set_worker_desc("flush-%s", dev_name(bdi->dev)); 1075766f9164SPeter Zijlstra current->flags |= PF_SWAPWRITE; 107603ba3782SJens Axboe 1077839a8e86STejun Heo if (likely(!current_is_workqueue_rescuer() || 10785acda9d1SJan Kara !test_bit(BDI_registered, &bdi->state))) { 107903ba3782SJens Axboe /* 1080839a8e86STejun Heo * The normal path. Keep writing back @bdi until its 1081839a8e86STejun Heo * work_list is empty. Note that this path is also taken 1082839a8e86STejun Heo * if @bdi is shutting down even when we're running off the 1083839a8e86STejun Heo * rescuer as work_list needs to be drained. 108403ba3782SJens Axboe */ 1085839a8e86STejun Heo do { 108625d130baSWanpeng Li pages_written = wb_do_writeback(wb); 1087455b2864SDave Chinner trace_writeback_pages_written(pages_written); 1088839a8e86STejun Heo } while (!list_empty(&bdi->work_list)); 1089839a8e86STejun Heo } else { 1090253c34e9SArtem Bityutskiy /* 1091839a8e86STejun Heo * bdi_wq can't get enough workers and we're running off 1092839a8e86STejun Heo * the emergency worker. Don't hog it. Hopefully, 1024 is 1093839a8e86STejun Heo * enough for efficient IO. 1094253c34e9SArtem Bityutskiy */ 1095839a8e86STejun Heo pages_written = writeback_inodes_wb(&bdi->wb, 1024, 1096839a8e86STejun Heo WB_REASON_FORKER_THREAD); 1097839a8e86STejun Heo trace_writeback_pages_written(pages_written); 109803ba3782SJens Axboe } 109903ba3782SJens Axboe 11006ca738d6SDerek Basehore if (!list_empty(&bdi->work_list)) 11016ca738d6SDerek Basehore mod_delayed_work(bdi_wq, &wb->dwork, 0); 11026ca738d6SDerek Basehore else if (wb_has_dirty_io(wb) && dirty_writeback_interval) 11036ca738d6SDerek Basehore bdi_wakeup_thread_delayed(bdi); 1104455b2864SDave Chinner 1105839a8e86STejun Heo current->flags &= ~PF_SWAPWRITE; 110603ba3782SJens Axboe } 110703ba3782SJens Axboe 110803ba3782SJens Axboe /* 110903ba3782SJens Axboe * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back 111003ba3782SJens Axboe * the whole world. 111103ba3782SJens Axboe */ 11120e175a18SCurt Wohlgemuth void wakeup_flusher_threads(long nr_pages, enum wb_reason reason) 111303ba3782SJens Axboe { 1114b8c2f347SChristoph Hellwig struct backing_dev_info *bdi; 1115b8c2f347SChristoph Hellwig 111647df3ddeSJan Kara if (!nr_pages) 111747df3ddeSJan Kara nr_pages = get_nr_dirty_pages(); 1118b8c2f347SChristoph Hellwig 1119b8c2f347SChristoph Hellwig rcu_read_lock(); 1120b8c2f347SChristoph Hellwig list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { 1121b8c2f347SChristoph Hellwig if (!bdi_has_dirty_io(bdi)) 1122b8c2f347SChristoph Hellwig continue; 11230e175a18SCurt Wohlgemuth __bdi_start_writeback(bdi, nr_pages, false, reason); 1124b8c2f347SChristoph Hellwig } 1125b8c2f347SChristoph Hellwig rcu_read_unlock(); 112603ba3782SJens Axboe } 112703ba3782SJens Axboe 112803ba3782SJens Axboe static noinline void block_dump___mark_inode_dirty(struct inode *inode) 112903ba3782SJens Axboe { 113003ba3782SJens Axboe if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { 113103ba3782SJens Axboe struct dentry *dentry; 113203ba3782SJens Axboe const char *name = "?"; 113303ba3782SJens Axboe 113403ba3782SJens Axboe dentry = d_find_alias(inode); 113503ba3782SJens Axboe if (dentry) { 113603ba3782SJens Axboe spin_lock(&dentry->d_lock); 113703ba3782SJens Axboe name = (const char *) dentry->d_name.name; 113803ba3782SJens Axboe } 113903ba3782SJens Axboe printk(KERN_DEBUG 114003ba3782SJens Axboe "%s(%d): dirtied inode %lu (%s) on %s\n", 114103ba3782SJens Axboe current->comm, task_pid_nr(current), inode->i_ino, 114203ba3782SJens Axboe name, inode->i_sb->s_id); 114303ba3782SJens Axboe if (dentry) { 114403ba3782SJens Axboe spin_unlock(&dentry->d_lock); 114503ba3782SJens Axboe dput(dentry); 114603ba3782SJens Axboe } 114703ba3782SJens Axboe } 114803ba3782SJens Axboe } 114903ba3782SJens Axboe 115003ba3782SJens Axboe /** 115103ba3782SJens Axboe * __mark_inode_dirty - internal function 115203ba3782SJens Axboe * @inode: inode to mark 115303ba3782SJens Axboe * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) 115403ba3782SJens Axboe * Mark an inode as dirty. Callers should use mark_inode_dirty or 115503ba3782SJens Axboe * mark_inode_dirty_sync. 115603ba3782SJens Axboe * 115703ba3782SJens Axboe * Put the inode on the super block's dirty list. 115803ba3782SJens Axboe * 115903ba3782SJens Axboe * CAREFUL! We mark it dirty unconditionally, but move it onto the 116003ba3782SJens Axboe * dirty list only if it is hashed or if it refers to a blockdev. 116103ba3782SJens Axboe * If it was not hashed, it will never be added to the dirty list 116203ba3782SJens Axboe * even if it is later hashed, as it will have been marked dirty already. 116303ba3782SJens Axboe * 116403ba3782SJens Axboe * In short, make sure you hash any inodes _before_ you start marking 116503ba3782SJens Axboe * them dirty. 116603ba3782SJens Axboe * 116703ba3782SJens Axboe * Note that for blockdevs, inode->dirtied_when represents the dirtying time of 116803ba3782SJens Axboe * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of 116903ba3782SJens Axboe * the kernel-internal blockdev inode represents the dirtying time of the 117003ba3782SJens Axboe * blockdev's pages. This is why for I_DIRTY_PAGES we always use 117103ba3782SJens Axboe * page->mapping->host, so the page-dirtying time is recorded in the internal 117203ba3782SJens Axboe * blockdev inode. 117303ba3782SJens Axboe */ 11740ae45f63STheodore Ts'o #define I_DIRTY_INODE (I_DIRTY_SYNC | I_DIRTY_DATASYNC) 117503ba3782SJens Axboe void __mark_inode_dirty(struct inode *inode, int flags) 117603ba3782SJens Axboe { 117703ba3782SJens Axboe struct super_block *sb = inode->i_sb; 1178253c34e9SArtem Bityutskiy struct backing_dev_info *bdi = NULL; 11790ae45f63STheodore Ts'o int dirtytime; 11800ae45f63STheodore Ts'o 11810ae45f63STheodore Ts'o trace_writeback_mark_inode_dirty(inode, flags); 118203ba3782SJens Axboe 118303ba3782SJens Axboe /* 118403ba3782SJens Axboe * Don't do this for I_DIRTY_PAGES - that doesn't actually 118503ba3782SJens Axboe * dirty the inode itself 118603ba3782SJens Axboe */ 11870ae45f63STheodore Ts'o if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC | I_DIRTY_TIME)) { 11889fb0a7daSTejun Heo trace_writeback_dirty_inode_start(inode, flags); 11899fb0a7daSTejun Heo 119003ba3782SJens Axboe if (sb->s_op->dirty_inode) 1191aa385729SChristoph Hellwig sb->s_op->dirty_inode(inode, flags); 11929fb0a7daSTejun Heo 11939fb0a7daSTejun Heo trace_writeback_dirty_inode(inode, flags); 119403ba3782SJens Axboe } 11950ae45f63STheodore Ts'o if (flags & I_DIRTY_INODE) 11960ae45f63STheodore Ts'o flags &= ~I_DIRTY_TIME; 11970ae45f63STheodore Ts'o dirtytime = flags & I_DIRTY_TIME; 119803ba3782SJens Axboe 119903ba3782SJens Axboe /* 12009c6ac78eSTejun Heo * Paired with smp_mb() in __writeback_single_inode() for the 12019c6ac78eSTejun Heo * following lockless i_state test. See there for details. 120203ba3782SJens Axboe */ 120303ba3782SJens Axboe smp_mb(); 120403ba3782SJens Axboe 12050ae45f63STheodore Ts'o if (((inode->i_state & flags) == flags) || 12060ae45f63STheodore Ts'o (dirtytime && (inode->i_state & I_DIRTY_INODE))) 120703ba3782SJens Axboe return; 120803ba3782SJens Axboe 120903ba3782SJens Axboe if (unlikely(block_dump)) 121003ba3782SJens Axboe block_dump___mark_inode_dirty(inode); 121103ba3782SJens Axboe 1212250df6edSDave Chinner spin_lock(&inode->i_lock); 12130ae45f63STheodore Ts'o if (dirtytime && (inode->i_state & I_DIRTY_INODE)) 12140ae45f63STheodore Ts'o goto out_unlock_inode; 121503ba3782SJens Axboe if ((inode->i_state & flags) != flags) { 121603ba3782SJens Axboe const int was_dirty = inode->i_state & I_DIRTY; 121703ba3782SJens Axboe 12180ae45f63STheodore Ts'o if (flags & I_DIRTY_INODE) 12190ae45f63STheodore Ts'o inode->i_state &= ~I_DIRTY_TIME; 122003ba3782SJens Axboe inode->i_state |= flags; 122103ba3782SJens Axboe 122203ba3782SJens Axboe /* 122303ba3782SJens Axboe * If the inode is being synced, just update its dirty state. 122403ba3782SJens Axboe * The unlocker will place the inode on the appropriate 122503ba3782SJens Axboe * superblock list, based upon its state. 122603ba3782SJens Axboe */ 122703ba3782SJens Axboe if (inode->i_state & I_SYNC) 1228250df6edSDave Chinner goto out_unlock_inode; 122903ba3782SJens Axboe 123003ba3782SJens Axboe /* 123103ba3782SJens Axboe * Only add valid (hashed) inodes to the superblock's 123203ba3782SJens Axboe * dirty list. Add blockdev inodes as well. 123303ba3782SJens Axboe */ 123403ba3782SJens Axboe if (!S_ISBLK(inode->i_mode)) { 12351d3382cbSAl Viro if (inode_unhashed(inode)) 1236250df6edSDave Chinner goto out_unlock_inode; 123703ba3782SJens Axboe } 1238a4ffdde6SAl Viro if (inode->i_state & I_FREEING) 1239250df6edSDave Chinner goto out_unlock_inode; 124003ba3782SJens Axboe 124103ba3782SJens Axboe /* 124203ba3782SJens Axboe * If the inode was already on b_dirty/b_io/b_more_io, don't 124303ba3782SJens Axboe * reposition it (that would break b_dirty time-ordering). 124403ba3782SJens Axboe */ 124503ba3782SJens Axboe if (!was_dirty) { 1246a66979abSDave Chinner bool wakeup_bdi = false; 1247253c34e9SArtem Bityutskiy bdi = inode_to_bdi(inode); 1248500b067cSJens Axboe 1249146d7009SJunxiao Bi spin_unlock(&inode->i_lock); 1250146d7009SJunxiao Bi spin_lock(&bdi->wb.list_lock); 1251253c34e9SArtem Bityutskiy if (bdi_cap_writeback_dirty(bdi)) { 1252253c34e9SArtem Bityutskiy WARN(!test_bit(BDI_registered, &bdi->state), 1253253c34e9SArtem Bityutskiy "bdi-%s not registered\n", bdi->name); 1254253c34e9SArtem Bityutskiy 1255253c34e9SArtem Bityutskiy /* 1256253c34e9SArtem Bityutskiy * If this is the first dirty inode for this 1257253c34e9SArtem Bityutskiy * bdi, we have to wake-up the corresponding 1258253c34e9SArtem Bityutskiy * bdi thread to make sure background 1259253c34e9SArtem Bityutskiy * write-back happens later. 1260253c34e9SArtem Bityutskiy */ 1261253c34e9SArtem Bityutskiy if (!wb_has_dirty_io(&bdi->wb)) 1262253c34e9SArtem Bityutskiy wakeup_bdi = true; 1263500b067cSJens Axboe } 126403ba3782SJens Axboe 126503ba3782SJens Axboe inode->dirtied_when = jiffies; 12660ae45f63STheodore Ts'o list_move(&inode->i_wb_list, dirtytime ? 12670ae45f63STheodore Ts'o &bdi->wb.b_dirty_time : &bdi->wb.b_dirty); 1268f758eeabSChristoph Hellwig spin_unlock(&bdi->wb.list_lock); 12690ae45f63STheodore Ts'o trace_writeback_dirty_inode_enqueue(inode); 1270253c34e9SArtem Bityutskiy 1271253c34e9SArtem Bityutskiy if (wakeup_bdi) 12726467716aSArtem Bityutskiy bdi_wakeup_thread_delayed(bdi); 1273a66979abSDave Chinner return; 1274a66979abSDave Chinner } 1275a66979abSDave Chinner } 1276a66979abSDave Chinner out_unlock_inode: 1277a66979abSDave Chinner spin_unlock(&inode->i_lock); 1278a66979abSDave Chinner 127903ba3782SJens Axboe } 128003ba3782SJens Axboe EXPORT_SYMBOL(__mark_inode_dirty); 128103ba3782SJens Axboe 1282b6e51316SJens Axboe static void wait_sb_inodes(struct super_block *sb) 128366f3b8e2SJens Axboe { 128438f21977SNick Piggin struct inode *inode, *old_inode = NULL; 128538f21977SNick Piggin 128603ba3782SJens Axboe /* 128703ba3782SJens Axboe * We need to be protected against the filesystem going from 128803ba3782SJens Axboe * r/o to r/w or vice versa. 128903ba3782SJens Axboe */ 1290b6e51316SJens Axboe WARN_ON(!rwsem_is_locked(&sb->s_umount)); 129103ba3782SJens Axboe 129255fa6091SDave Chinner spin_lock(&inode_sb_list_lock); 129366f3b8e2SJens Axboe 129438f21977SNick Piggin /* 129538f21977SNick Piggin * Data integrity sync. Must wait for all pages under writeback, 129638f21977SNick Piggin * because there may have been pages dirtied before our sync 129738f21977SNick Piggin * call, but which had writeout started before we write it out. 129838f21977SNick Piggin * In which case, the inode may not be on the dirty list, but 129938f21977SNick Piggin * we still have to wait for that writeout. 130038f21977SNick Piggin */ 1301b6e51316SJens Axboe list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { 1302250df6edSDave Chinner struct address_space *mapping = inode->i_mapping; 130338f21977SNick Piggin 1304250df6edSDave Chinner spin_lock(&inode->i_lock); 1305250df6edSDave Chinner if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) || 1306250df6edSDave Chinner (mapping->nrpages == 0)) { 1307250df6edSDave Chinner spin_unlock(&inode->i_lock); 130838f21977SNick Piggin continue; 1309250df6edSDave Chinner } 131038f21977SNick Piggin __iget(inode); 1311250df6edSDave Chinner spin_unlock(&inode->i_lock); 131255fa6091SDave Chinner spin_unlock(&inode_sb_list_lock); 131355fa6091SDave Chinner 131438f21977SNick Piggin /* 131555fa6091SDave Chinner * We hold a reference to 'inode' so it couldn't have been 131655fa6091SDave Chinner * removed from s_inodes list while we dropped the 131755fa6091SDave Chinner * inode_sb_list_lock. We cannot iput the inode now as we can 131855fa6091SDave Chinner * be holding the last reference and we cannot iput it under 131955fa6091SDave Chinner * inode_sb_list_lock. So we keep the reference and iput it 132055fa6091SDave Chinner * later. 132138f21977SNick Piggin */ 132238f21977SNick Piggin iput(old_inode); 132338f21977SNick Piggin old_inode = inode; 132438f21977SNick Piggin 132538f21977SNick Piggin filemap_fdatawait(mapping); 132638f21977SNick Piggin 132738f21977SNick Piggin cond_resched(); 132838f21977SNick Piggin 132955fa6091SDave Chinner spin_lock(&inode_sb_list_lock); 133038f21977SNick Piggin } 133155fa6091SDave Chinner spin_unlock(&inode_sb_list_lock); 133238f21977SNick Piggin iput(old_inode); 133366f3b8e2SJens Axboe } 13341da177e4SLinus Torvalds 1335d8a8559cSJens Axboe /** 13363259f8beSChris Mason * writeback_inodes_sb_nr - writeback dirty inodes from given super_block 1337d8a8559cSJens Axboe * @sb: the superblock 13383259f8beSChris Mason * @nr: the number of pages to write 1339786228abSMarcos Paulo de Souza * @reason: reason why some writeback work initiated 13401da177e4SLinus Torvalds * 1341d8a8559cSJens Axboe * Start writeback on some inodes on this super_block. No guarantees are made 1342d8a8559cSJens Axboe * on how many (if any) will be written, and this function does not wait 13433259f8beSChris Mason * for IO completion of submitted IO. 13441da177e4SLinus Torvalds */ 13450e175a18SCurt Wohlgemuth void writeback_inodes_sb_nr(struct super_block *sb, 13460e175a18SCurt Wohlgemuth unsigned long nr, 13470e175a18SCurt Wohlgemuth enum wb_reason reason) 13481da177e4SLinus Torvalds { 134983ba7b07SChristoph Hellwig DECLARE_COMPLETION_ONSTACK(done); 135083ba7b07SChristoph Hellwig struct wb_writeback_work work = { 13513c4d7165SChristoph Hellwig .sb = sb, 13523c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_NONE, 13536e6938b6SWu Fengguang .tagged_writepages = 1, 135483ba7b07SChristoph Hellwig .done = &done, 13553259f8beSChris Mason .nr_pages = nr, 13560e175a18SCurt Wohlgemuth .reason = reason, 13573c4d7165SChristoph Hellwig }; 13580e3c9a22SJens Axboe 13596eedc701SJan Kara if (sb->s_bdi == &noop_backing_dev_info) 13606eedc701SJan Kara return; 1361cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 136283ba7b07SChristoph Hellwig bdi_queue_work(sb->s_bdi, &work); 136383ba7b07SChristoph Hellwig wait_for_completion(&done); 13641da177e4SLinus Torvalds } 13653259f8beSChris Mason EXPORT_SYMBOL(writeback_inodes_sb_nr); 13663259f8beSChris Mason 13673259f8beSChris Mason /** 13683259f8beSChris Mason * writeback_inodes_sb - writeback dirty inodes from given super_block 13693259f8beSChris Mason * @sb: the superblock 1370786228abSMarcos Paulo de Souza * @reason: reason why some writeback work was initiated 13713259f8beSChris Mason * 13723259f8beSChris Mason * Start writeback on some inodes on this super_block. No guarantees are made 13733259f8beSChris Mason * on how many (if any) will be written, and this function does not wait 13743259f8beSChris Mason * for IO completion of submitted IO. 13753259f8beSChris Mason */ 13760e175a18SCurt Wohlgemuth void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) 13773259f8beSChris Mason { 13780e175a18SCurt Wohlgemuth return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); 13793259f8beSChris Mason } 1380d8a8559cSJens Axboe EXPORT_SYMBOL(writeback_inodes_sb); 1381d8a8559cSJens Axboe 1382d8a8559cSJens Axboe /** 138310ee27a0SMiao Xie * try_to_writeback_inodes_sb_nr - try to start writeback if none underway 13843259f8beSChris Mason * @sb: the superblock 13853259f8beSChris Mason * @nr: the number of pages to write 138610ee27a0SMiao Xie * @reason: the reason of writeback 13873259f8beSChris Mason * 138810ee27a0SMiao Xie * Invoke writeback_inodes_sb_nr if no writeback is currently underway. 13893259f8beSChris Mason * Returns 1 if writeback was started, 0 if not. 13903259f8beSChris Mason */ 139110ee27a0SMiao Xie int try_to_writeback_inodes_sb_nr(struct super_block *sb, 13920e175a18SCurt Wohlgemuth unsigned long nr, 13930e175a18SCurt Wohlgemuth enum wb_reason reason) 13943259f8beSChris Mason { 139510ee27a0SMiao Xie if (writeback_in_progress(sb->s_bdi)) 139610ee27a0SMiao Xie return 1; 139710ee27a0SMiao Xie 139810ee27a0SMiao Xie if (!down_read_trylock(&sb->s_umount)) 139910ee27a0SMiao Xie return 0; 140010ee27a0SMiao Xie 14010e175a18SCurt Wohlgemuth writeback_inodes_sb_nr(sb, nr, reason); 14023259f8beSChris Mason up_read(&sb->s_umount); 14033259f8beSChris Mason return 1; 14043259f8beSChris Mason } 140510ee27a0SMiao Xie EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr); 140610ee27a0SMiao Xie 140710ee27a0SMiao Xie /** 140810ee27a0SMiao Xie * try_to_writeback_inodes_sb - try to start writeback if none underway 140910ee27a0SMiao Xie * @sb: the superblock 141010ee27a0SMiao Xie * @reason: reason why some writeback work was initiated 141110ee27a0SMiao Xie * 141210ee27a0SMiao Xie * Implement by try_to_writeback_inodes_sb_nr() 141310ee27a0SMiao Xie * Returns 1 if writeback was started, 0 if not. 141410ee27a0SMiao Xie */ 141510ee27a0SMiao Xie int try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) 141610ee27a0SMiao Xie { 141710ee27a0SMiao Xie return try_to_writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); 141810ee27a0SMiao Xie } 141910ee27a0SMiao Xie EXPORT_SYMBOL(try_to_writeback_inodes_sb); 14203259f8beSChris Mason 14213259f8beSChris Mason /** 1422d8a8559cSJens Axboe * sync_inodes_sb - sync sb inode pages 1423d8a8559cSJens Axboe * @sb: the superblock 1424d8a8559cSJens Axboe * 1425d8a8559cSJens Axboe * This function writes and waits on any dirty inode belonging to this 14260dc83bd3SJan Kara * super_block. 1427d8a8559cSJens Axboe */ 14280dc83bd3SJan Kara void sync_inodes_sb(struct super_block *sb) 1429d8a8559cSJens Axboe { 143083ba7b07SChristoph Hellwig DECLARE_COMPLETION_ONSTACK(done); 143183ba7b07SChristoph Hellwig struct wb_writeback_work work = { 14323c4d7165SChristoph Hellwig .sb = sb, 14333c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_ALL, 14343c4d7165SChristoph Hellwig .nr_pages = LONG_MAX, 14353c4d7165SChristoph Hellwig .range_cyclic = 0, 143683ba7b07SChristoph Hellwig .done = &done, 14370e175a18SCurt Wohlgemuth .reason = WB_REASON_SYNC, 14387747bd4bSDave Chinner .for_sync = 1, 14393c4d7165SChristoph Hellwig }; 14403c4d7165SChristoph Hellwig 14416eedc701SJan Kara /* Nothing to do? */ 14426eedc701SJan Kara if (sb->s_bdi == &noop_backing_dev_info) 14436eedc701SJan Kara return; 1444cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 1445cf37e972SChristoph Hellwig 144683ba7b07SChristoph Hellwig bdi_queue_work(sb->s_bdi, &work); 144783ba7b07SChristoph Hellwig wait_for_completion(&done); 144883ba7b07SChristoph Hellwig 1449b6e51316SJens Axboe wait_sb_inodes(sb); 1450d8a8559cSJens Axboe } 1451d8a8559cSJens Axboe EXPORT_SYMBOL(sync_inodes_sb); 14521da177e4SLinus Torvalds 14531da177e4SLinus Torvalds /** 14541da177e4SLinus Torvalds * write_inode_now - write an inode to disk 14551da177e4SLinus Torvalds * @inode: inode to write to disk 14561da177e4SLinus Torvalds * @sync: whether the write should be synchronous or not 14571da177e4SLinus Torvalds * 14587f04c26dSAndrea Arcangeli * This function commits an inode to disk immediately if it is dirty. This is 14597f04c26dSAndrea Arcangeli * primarily needed by knfsd. 14607f04c26dSAndrea Arcangeli * 14617f04c26dSAndrea Arcangeli * The caller must either have a ref on the inode or must have set I_WILL_FREE. 14621da177e4SLinus Torvalds */ 14631da177e4SLinus Torvalds int write_inode_now(struct inode *inode, int sync) 14641da177e4SLinus Torvalds { 1465f758eeabSChristoph Hellwig struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; 14661da177e4SLinus Torvalds struct writeback_control wbc = { 14671da177e4SLinus Torvalds .nr_to_write = LONG_MAX, 146818914b18SMike Galbraith .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, 1469111ebb6eSOGAWA Hirofumi .range_start = 0, 1470111ebb6eSOGAWA Hirofumi .range_end = LLONG_MAX, 14711da177e4SLinus Torvalds }; 14721da177e4SLinus Torvalds 14731da177e4SLinus Torvalds if (!mapping_cap_writeback_dirty(inode->i_mapping)) 147449364ce2SAndrew Morton wbc.nr_to_write = 0; 14751da177e4SLinus Torvalds 14761da177e4SLinus Torvalds might_sleep(); 14774f8ad655SJan Kara return writeback_single_inode(inode, wb, &wbc); 14781da177e4SLinus Torvalds } 14791da177e4SLinus Torvalds EXPORT_SYMBOL(write_inode_now); 14801da177e4SLinus Torvalds 14811da177e4SLinus Torvalds /** 14821da177e4SLinus Torvalds * sync_inode - write an inode and its pages to disk. 14831da177e4SLinus Torvalds * @inode: the inode to sync 14841da177e4SLinus Torvalds * @wbc: controls the writeback mode 14851da177e4SLinus Torvalds * 14861da177e4SLinus Torvalds * sync_inode() will write an inode and its pages to disk. It will also 14871da177e4SLinus Torvalds * correctly update the inode on its superblock's dirty inode lists and will 14881da177e4SLinus Torvalds * update inode->i_state. 14891da177e4SLinus Torvalds * 14901da177e4SLinus Torvalds * The caller must have a ref on the inode. 14911da177e4SLinus Torvalds */ 14921da177e4SLinus Torvalds int sync_inode(struct inode *inode, struct writeback_control *wbc) 14931da177e4SLinus Torvalds { 14944f8ad655SJan Kara return writeback_single_inode(inode, &inode_to_bdi(inode)->wb, wbc); 14951da177e4SLinus Torvalds } 14961da177e4SLinus Torvalds EXPORT_SYMBOL(sync_inode); 1497c3765016SChristoph Hellwig 1498c3765016SChristoph Hellwig /** 1499c691b9d9SAndrew Morton * sync_inode_metadata - write an inode to disk 1500c3765016SChristoph Hellwig * @inode: the inode to sync 1501c3765016SChristoph Hellwig * @wait: wait for I/O to complete. 1502c3765016SChristoph Hellwig * 1503c691b9d9SAndrew Morton * Write an inode to disk and adjust its dirty state after completion. 1504c3765016SChristoph Hellwig * 1505c3765016SChristoph Hellwig * Note: only writes the actual inode, no associated data or other metadata. 1506c3765016SChristoph Hellwig */ 1507c3765016SChristoph Hellwig int sync_inode_metadata(struct inode *inode, int wait) 1508c3765016SChristoph Hellwig { 1509c3765016SChristoph Hellwig struct writeback_control wbc = { 1510c3765016SChristoph Hellwig .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE, 1511c3765016SChristoph Hellwig .nr_to_write = 0, /* metadata-only */ 1512c3765016SChristoph Hellwig }; 1513c3765016SChristoph Hellwig 1514c3765016SChristoph Hellwig return sync_inode(inode, &wbc); 1515c3765016SChristoph Hellwig } 1516c3765016SChristoph Hellwig EXPORT_SYMBOL(sync_inode_metadata); 1517