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 96d6c10f1fSTejun Heo static bool wb_io_lists_populated(struct bdi_writeback *wb) 97d6c10f1fSTejun Heo { 98d6c10f1fSTejun Heo if (wb_has_dirty_io(wb)) { 99d6c10f1fSTejun Heo return false; 100d6c10f1fSTejun Heo } else { 101d6c10f1fSTejun Heo set_bit(WB_has_dirty_io, &wb->state); 102766a9d6eSTejun Heo atomic_long_add(wb->avg_write_bandwidth, 103766a9d6eSTejun Heo &wb->bdi->tot_write_bandwidth); 104d6c10f1fSTejun Heo return true; 105d6c10f1fSTejun Heo } 106d6c10f1fSTejun Heo } 107d6c10f1fSTejun Heo 108d6c10f1fSTejun Heo static void wb_io_lists_depopulated(struct bdi_writeback *wb) 109d6c10f1fSTejun Heo { 110d6c10f1fSTejun Heo if (wb_has_dirty_io(wb) && list_empty(&wb->b_dirty) && 111766a9d6eSTejun Heo list_empty(&wb->b_io) && list_empty(&wb->b_more_io)) { 112d6c10f1fSTejun Heo clear_bit(WB_has_dirty_io, &wb->state); 113766a9d6eSTejun Heo atomic_long_sub(wb->avg_write_bandwidth, 114766a9d6eSTejun Heo &wb->bdi->tot_write_bandwidth); 115766a9d6eSTejun Heo } 116d6c10f1fSTejun Heo } 117d6c10f1fSTejun Heo 118d6c10f1fSTejun Heo /** 119d6c10f1fSTejun Heo * inode_wb_list_move_locked - move an inode onto a bdi_writeback IO list 120d6c10f1fSTejun Heo * @inode: inode to be moved 121d6c10f1fSTejun Heo * @wb: target bdi_writeback 122d6c10f1fSTejun Heo * @head: one of @wb->b_{dirty|io|more_io} 123d6c10f1fSTejun Heo * 124d6c10f1fSTejun Heo * Move @inode->i_wb_list to @list of @wb and set %WB_has_dirty_io. 125d6c10f1fSTejun Heo * Returns %true if @inode is the first occupant of the !dirty_time IO 126d6c10f1fSTejun Heo * lists; otherwise, %false. 127d6c10f1fSTejun Heo */ 128d6c10f1fSTejun Heo static bool inode_wb_list_move_locked(struct inode *inode, 129d6c10f1fSTejun Heo struct bdi_writeback *wb, 130d6c10f1fSTejun Heo struct list_head *head) 131d6c10f1fSTejun Heo { 132d6c10f1fSTejun Heo assert_spin_locked(&wb->list_lock); 133d6c10f1fSTejun Heo 134d6c10f1fSTejun Heo list_move(&inode->i_wb_list, head); 135d6c10f1fSTejun Heo 136d6c10f1fSTejun Heo /* dirty_time doesn't count as dirty_io until expiration */ 137d6c10f1fSTejun Heo if (head != &wb->b_dirty_time) 138d6c10f1fSTejun Heo return wb_io_lists_populated(wb); 139d6c10f1fSTejun Heo 140d6c10f1fSTejun Heo wb_io_lists_depopulated(wb); 141d6c10f1fSTejun Heo return false; 142d6c10f1fSTejun Heo } 143d6c10f1fSTejun Heo 144d6c10f1fSTejun Heo /** 145d6c10f1fSTejun Heo * inode_wb_list_del_locked - remove an inode from its bdi_writeback IO list 146d6c10f1fSTejun Heo * @inode: inode to be removed 147d6c10f1fSTejun Heo * @wb: bdi_writeback @inode is being removed from 148d6c10f1fSTejun Heo * 149d6c10f1fSTejun Heo * Remove @inode which may be on one of @wb->b_{dirty|io|more_io} lists and 150d6c10f1fSTejun Heo * clear %WB_has_dirty_io if all are empty afterwards. 151d6c10f1fSTejun Heo */ 152d6c10f1fSTejun Heo static void inode_wb_list_del_locked(struct inode *inode, 153d6c10f1fSTejun Heo struct bdi_writeback *wb) 154d6c10f1fSTejun Heo { 155d6c10f1fSTejun Heo assert_spin_locked(&wb->list_lock); 156d6c10f1fSTejun Heo 157d6c10f1fSTejun Heo list_del_init(&inode->i_wb_list); 158d6c10f1fSTejun Heo wb_io_lists_depopulated(wb); 159d6c10f1fSTejun Heo } 160d6c10f1fSTejun Heo 161f0054bb1STejun Heo static void wb_wakeup(struct bdi_writeback *wb) 1625acda9d1SJan Kara { 163f0054bb1STejun Heo spin_lock_bh(&wb->work_lock); 164f0054bb1STejun Heo if (test_bit(WB_registered, &wb->state)) 165f0054bb1STejun Heo mod_delayed_work(bdi_wq, &wb->dwork, 0); 166f0054bb1STejun Heo spin_unlock_bh(&wb->work_lock); 1675acda9d1SJan Kara } 1685acda9d1SJan Kara 169f0054bb1STejun Heo static void wb_queue_work(struct bdi_writeback *wb, 1706585027aSJan Kara struct wb_writeback_work *work) 1716585027aSJan Kara { 172f0054bb1STejun Heo trace_writeback_queue(wb->bdi, work); 1736585027aSJan Kara 174f0054bb1STejun Heo spin_lock_bh(&wb->work_lock); 175f0054bb1STejun Heo if (!test_bit(WB_registered, &wb->state)) { 1765acda9d1SJan Kara if (work->done) 1775acda9d1SJan Kara complete(work->done); 1785acda9d1SJan Kara goto out_unlock; 1795acda9d1SJan Kara } 180f0054bb1STejun Heo list_add_tail(&work->list, &wb->work_list); 181f0054bb1STejun Heo mod_delayed_work(bdi_wq, &wb->dwork, 0); 1825acda9d1SJan Kara out_unlock: 183f0054bb1STejun Heo spin_unlock_bh(&wb->work_lock); 18403ba3782SJens Axboe } 1851da177e4SLinus Torvalds 186f0054bb1STejun Heo static void __wb_start_writeback(struct bdi_writeback *wb, long nr_pages, 1870e175a18SCurt Wohlgemuth bool range_cyclic, enum wb_reason reason) 1881da177e4SLinus Torvalds { 18983ba7b07SChristoph Hellwig struct wb_writeback_work *work; 19003ba3782SJens Axboe 191bcddc3f0SJens Axboe /* 192bcddc3f0SJens Axboe * This is WB_SYNC_NONE writeback, so if allocation fails just 193bcddc3f0SJens Axboe * wakeup the thread for old dirty data writeback 194bcddc3f0SJens Axboe */ 19583ba7b07SChristoph Hellwig work = kzalloc(sizeof(*work), GFP_ATOMIC); 19683ba7b07SChristoph Hellwig if (!work) { 197f0054bb1STejun Heo trace_writeback_nowork(wb->bdi); 198f0054bb1STejun Heo wb_wakeup(wb); 19983ba7b07SChristoph Hellwig return; 20083ba7b07SChristoph Hellwig } 20183ba7b07SChristoph Hellwig 20283ba7b07SChristoph Hellwig work->sync_mode = WB_SYNC_NONE; 20383ba7b07SChristoph Hellwig work->nr_pages = nr_pages; 20483ba7b07SChristoph Hellwig work->range_cyclic = range_cyclic; 2050e175a18SCurt Wohlgemuth work->reason = reason; 20683ba7b07SChristoph Hellwig 207f0054bb1STejun Heo wb_queue_work(wb, work); 20803ba3782SJens Axboe } 209b6e51316SJens Axboe 210703c2708STejun Heo #ifdef CONFIG_CGROUP_WRITEBACK 211703c2708STejun Heo 212703c2708STejun Heo /** 213703c2708STejun Heo * inode_congested - test whether an inode is congested 214703c2708STejun Heo * @inode: inode to test for congestion 215703c2708STejun Heo * @cong_bits: mask of WB_[a]sync_congested bits to test 216703c2708STejun Heo * 217703c2708STejun Heo * Tests whether @inode is congested. @cong_bits is the mask of congestion 218703c2708STejun Heo * bits to test and the return value is the mask of set bits. 219703c2708STejun Heo * 220703c2708STejun Heo * If cgroup writeback is enabled for @inode, the congestion state is 221703c2708STejun Heo * determined by whether the cgwb (cgroup bdi_writeback) for the blkcg 222703c2708STejun Heo * associated with @inode is congested; otherwise, the root wb's congestion 223703c2708STejun Heo * state is used. 224703c2708STejun Heo */ 225703c2708STejun Heo int inode_congested(struct inode *inode, int cong_bits) 226703c2708STejun Heo { 227703c2708STejun Heo if (inode) { 228703c2708STejun Heo struct bdi_writeback *wb = inode_to_wb(inode); 229703c2708STejun Heo if (wb) 230703c2708STejun Heo return wb_congested(wb, cong_bits); 231703c2708STejun Heo } 232703c2708STejun Heo 233703c2708STejun Heo return wb_congested(&inode_to_bdi(inode)->wb, cong_bits); 234703c2708STejun Heo } 235703c2708STejun Heo EXPORT_SYMBOL_GPL(inode_congested); 236703c2708STejun Heo 237703c2708STejun Heo #endif /* CONFIG_CGROUP_WRITEBACK */ 238703c2708STejun Heo 239b6e51316SJens Axboe /** 240b6e51316SJens Axboe * bdi_start_writeback - start writeback 241b6e51316SJens Axboe * @bdi: the backing device to write from 242b6e51316SJens Axboe * @nr_pages: the number of pages to write 243786228abSMarcos Paulo de Souza * @reason: reason why some writeback work was initiated 244b6e51316SJens Axboe * 245b6e51316SJens Axboe * Description: 246b6e51316SJens Axboe * This does WB_SYNC_NONE opportunistic writeback. The IO is only 24725985edcSLucas De Marchi * started when this function returns, we make no guarantees on 2480e3c9a22SJens Axboe * completion. Caller need not hold sb s_umount semaphore. 249b6e51316SJens Axboe * 250b6e51316SJens Axboe */ 2510e175a18SCurt Wohlgemuth void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, 2520e175a18SCurt Wohlgemuth enum wb_reason reason) 253b6e51316SJens Axboe { 254f0054bb1STejun Heo __wb_start_writeback(&bdi->wb, nr_pages, true, reason); 255d3ddec76SWu Fengguang } 256d3ddec76SWu Fengguang 257c5444198SChristoph Hellwig /** 258c5444198SChristoph Hellwig * bdi_start_background_writeback - start background writeback 259c5444198SChristoph Hellwig * @bdi: the backing device to write from 260c5444198SChristoph Hellwig * 261c5444198SChristoph Hellwig * Description: 2626585027aSJan Kara * This makes sure WB_SYNC_NONE background writeback happens. When 2636585027aSJan Kara * this function returns, it is only guaranteed that for given BDI 2646585027aSJan Kara * some IO is happening if we are over background dirty threshold. 2656585027aSJan Kara * Caller need not hold sb s_umount semaphore. 266c5444198SChristoph Hellwig */ 267c5444198SChristoph Hellwig void bdi_start_background_writeback(struct backing_dev_info *bdi) 268c5444198SChristoph Hellwig { 2696585027aSJan Kara /* 2706585027aSJan Kara * We just wake up the flusher thread. It will perform background 2716585027aSJan Kara * writeback as soon as there is no other work to do. 2726585027aSJan Kara */ 27371927e84SWu Fengguang trace_writeback_wake_background(bdi); 274f0054bb1STejun Heo wb_wakeup(&bdi->wb); 2751da177e4SLinus Torvalds } 2761da177e4SLinus Torvalds 2771da177e4SLinus Torvalds /* 278a66979abSDave Chinner * Remove the inode from the writeback list it is on. 279a66979abSDave Chinner */ 280a66979abSDave Chinner void inode_wb_list_del(struct inode *inode) 281a66979abSDave Chinner { 28252ebea74STejun Heo struct bdi_writeback *wb = inode_to_wb(inode); 283a66979abSDave Chinner 28452ebea74STejun Heo spin_lock(&wb->list_lock); 285d6c10f1fSTejun Heo inode_wb_list_del_locked(inode, wb); 28652ebea74STejun Heo spin_unlock(&wb->list_lock); 287f758eeabSChristoph Hellwig } 288a66979abSDave Chinner 289a66979abSDave Chinner /* 2906610a0bcSAndrew Morton * Redirty an inode: set its when-it-was dirtied timestamp and move it to the 2916610a0bcSAndrew Morton * furthest end of its superblock's dirty-inode list. 2926610a0bcSAndrew Morton * 2936610a0bcSAndrew Morton * Before stamping the inode's ->dirtied_when, we check to see whether it is 29466f3b8e2SJens Axboe * already the most-recently-dirtied inode on the b_dirty list. If that is 2956610a0bcSAndrew Morton * the case then the inode must have been redirtied while it was being written 2966610a0bcSAndrew Morton * out and we don't reset its dirtied_when. 2976610a0bcSAndrew Morton */ 298f758eeabSChristoph Hellwig static void redirty_tail(struct inode *inode, struct bdi_writeback *wb) 2996610a0bcSAndrew Morton { 30003ba3782SJens Axboe if (!list_empty(&wb->b_dirty)) { 30166f3b8e2SJens Axboe struct inode *tail; 3026610a0bcSAndrew Morton 3037ccf19a8SNick Piggin tail = wb_inode(wb->b_dirty.next); 30466f3b8e2SJens Axboe if (time_before(inode->dirtied_when, tail->dirtied_when)) 3056610a0bcSAndrew Morton inode->dirtied_when = jiffies; 3066610a0bcSAndrew Morton } 307d6c10f1fSTejun Heo inode_wb_list_move_locked(inode, wb, &wb->b_dirty); 3086610a0bcSAndrew Morton } 3096610a0bcSAndrew Morton 3106610a0bcSAndrew Morton /* 31166f3b8e2SJens Axboe * requeue inode for re-scanning after bdi->b_io list is exhausted. 312c986d1e2SAndrew Morton */ 313f758eeabSChristoph Hellwig static void requeue_io(struct inode *inode, struct bdi_writeback *wb) 314c986d1e2SAndrew Morton { 315d6c10f1fSTejun Heo inode_wb_list_move_locked(inode, wb, &wb->b_more_io); 316c986d1e2SAndrew Morton } 317c986d1e2SAndrew Morton 3181c0eeaf5SJoern Engel static void inode_sync_complete(struct inode *inode) 3191c0eeaf5SJoern Engel { 320365b94aeSJan Kara inode->i_state &= ~I_SYNC; 3214eff96ddSJan Kara /* If inode is clean an unused, put it into LRU now... */ 3224eff96ddSJan Kara inode_add_lru(inode); 323365b94aeSJan Kara /* Waiters must see I_SYNC cleared before being woken up */ 3241c0eeaf5SJoern Engel smp_mb(); 3251c0eeaf5SJoern Engel wake_up_bit(&inode->i_state, __I_SYNC); 3261c0eeaf5SJoern Engel } 3271c0eeaf5SJoern Engel 328d2caa3c5SJeff Layton static bool inode_dirtied_after(struct inode *inode, unsigned long t) 329d2caa3c5SJeff Layton { 330d2caa3c5SJeff Layton bool ret = time_after(inode->dirtied_when, t); 331d2caa3c5SJeff Layton #ifndef CONFIG_64BIT 332d2caa3c5SJeff Layton /* 333d2caa3c5SJeff Layton * For inodes being constantly redirtied, dirtied_when can get stuck. 334d2caa3c5SJeff Layton * It _appears_ to be in the future, but is actually in distant past. 335d2caa3c5SJeff Layton * This test is necessary to prevent such wrapped-around relative times 3365b0830cbSJens Axboe * from permanently stopping the whole bdi writeback. 337d2caa3c5SJeff Layton */ 338d2caa3c5SJeff Layton ret = ret && time_before_eq(inode->dirtied_when, jiffies); 339d2caa3c5SJeff Layton #endif 340d2caa3c5SJeff Layton return ret; 341d2caa3c5SJeff Layton } 342d2caa3c5SJeff Layton 3430ae45f63STheodore Ts'o #define EXPIRE_DIRTY_ATIME 0x0001 3440ae45f63STheodore Ts'o 345c986d1e2SAndrew Morton /* 3460e2f2b23SWang Sheng-Hui * Move expired (dirtied before work->older_than_this) dirty inodes from 347697e6fedSJan Kara * @delaying_queue to @dispatch_queue. 3482c136579SFengguang Wu */ 349e84d0a4fSWu Fengguang static int move_expired_inodes(struct list_head *delaying_queue, 3502c136579SFengguang Wu struct list_head *dispatch_queue, 3510ae45f63STheodore Ts'o int flags, 352ad4e38ddSCurt Wohlgemuth struct wb_writeback_work *work) 3532c136579SFengguang Wu { 3540ae45f63STheodore Ts'o unsigned long *older_than_this = NULL; 3550ae45f63STheodore Ts'o unsigned long expire_time; 3565c03449dSShaohua Li LIST_HEAD(tmp); 3575c03449dSShaohua Li struct list_head *pos, *node; 358cf137307SJens Axboe struct super_block *sb = NULL; 3595c03449dSShaohua Li struct inode *inode; 360cf137307SJens Axboe int do_sb_sort = 0; 361e84d0a4fSWu Fengguang int moved = 0; 3625c03449dSShaohua Li 3630ae45f63STheodore Ts'o if ((flags & EXPIRE_DIRTY_ATIME) == 0) 3640ae45f63STheodore Ts'o older_than_this = work->older_than_this; 365a2f48706STheodore Ts'o else if (!work->for_sync) { 366a2f48706STheodore Ts'o expire_time = jiffies - (dirtytime_expire_interval * HZ); 3670ae45f63STheodore Ts'o older_than_this = &expire_time; 3680ae45f63STheodore Ts'o } 3692c136579SFengguang Wu while (!list_empty(delaying_queue)) { 3707ccf19a8SNick Piggin inode = wb_inode(delaying_queue->prev); 3710ae45f63STheodore Ts'o if (older_than_this && 3720ae45f63STheodore Ts'o inode_dirtied_after(inode, *older_than_this)) 3732c136579SFengguang Wu break; 374a8855990SJan Kara list_move(&inode->i_wb_list, &tmp); 375a8855990SJan Kara moved++; 3760ae45f63STheodore Ts'o if (flags & EXPIRE_DIRTY_ATIME) 3770ae45f63STheodore Ts'o set_bit(__I_DIRTY_TIME_EXPIRED, &inode->i_state); 378a8855990SJan Kara if (sb_is_blkdev_sb(inode->i_sb)) 379a8855990SJan Kara continue; 380cf137307SJens Axboe if (sb && sb != inode->i_sb) 381cf137307SJens Axboe do_sb_sort = 1; 382cf137307SJens Axboe sb = inode->i_sb; 3835c03449dSShaohua Li } 3845c03449dSShaohua Li 385cf137307SJens Axboe /* just one sb in list, splice to dispatch_queue and we're done */ 386cf137307SJens Axboe if (!do_sb_sort) { 387cf137307SJens Axboe list_splice(&tmp, dispatch_queue); 388e84d0a4fSWu Fengguang goto out; 389cf137307SJens Axboe } 390cf137307SJens Axboe 3915c03449dSShaohua Li /* Move inodes from one superblock together */ 3925c03449dSShaohua Li while (!list_empty(&tmp)) { 3937ccf19a8SNick Piggin sb = wb_inode(tmp.prev)->i_sb; 3945c03449dSShaohua Li list_for_each_prev_safe(pos, node, &tmp) { 3957ccf19a8SNick Piggin inode = wb_inode(pos); 3965c03449dSShaohua Li if (inode->i_sb == sb) 3977ccf19a8SNick Piggin list_move(&inode->i_wb_list, dispatch_queue); 3982c136579SFengguang Wu } 3992c136579SFengguang Wu } 400e84d0a4fSWu Fengguang out: 401e84d0a4fSWu Fengguang return moved; 4025c03449dSShaohua Li } 4032c136579SFengguang Wu 4042c136579SFengguang Wu /* 4052c136579SFengguang Wu * Queue all expired dirty inodes for io, eldest first. 4064ea879b9SWu Fengguang * Before 4074ea879b9SWu Fengguang * newly dirtied b_dirty b_io b_more_io 4084ea879b9SWu Fengguang * =============> gf edc BA 4094ea879b9SWu Fengguang * After 4104ea879b9SWu Fengguang * newly dirtied b_dirty b_io b_more_io 4114ea879b9SWu Fengguang * =============> g fBAedc 4124ea879b9SWu Fengguang * | 4134ea879b9SWu Fengguang * +--> dequeue for IO 4142c136579SFengguang Wu */ 415ad4e38ddSCurt Wohlgemuth static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work) 4162c136579SFengguang Wu { 417e84d0a4fSWu Fengguang int moved; 4180ae45f63STheodore Ts'o 419f758eeabSChristoph Hellwig assert_spin_locked(&wb->list_lock); 4204ea879b9SWu Fengguang list_splice_init(&wb->b_more_io, &wb->b_io); 4210ae45f63STheodore Ts'o moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, 0, work); 4220ae45f63STheodore Ts'o moved += move_expired_inodes(&wb->b_dirty_time, &wb->b_io, 4230ae45f63STheodore Ts'o EXPIRE_DIRTY_ATIME, work); 424d6c10f1fSTejun Heo if (moved) 425d6c10f1fSTejun Heo wb_io_lists_populated(wb); 426ad4e38ddSCurt Wohlgemuth trace_writeback_queue_io(wb, work, moved); 42766f3b8e2SJens Axboe } 42866f3b8e2SJens Axboe 429a9185b41SChristoph Hellwig static int write_inode(struct inode *inode, struct writeback_control *wbc) 43066f3b8e2SJens Axboe { 4319fb0a7daSTejun Heo int ret; 4329fb0a7daSTejun Heo 4339fb0a7daSTejun Heo if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) { 4349fb0a7daSTejun Heo trace_writeback_write_inode_start(inode, wbc); 4359fb0a7daSTejun Heo ret = inode->i_sb->s_op->write_inode(inode, wbc); 4369fb0a7daSTejun Heo trace_writeback_write_inode(inode, wbc); 4379fb0a7daSTejun Heo return ret; 4389fb0a7daSTejun Heo } 43903ba3782SJens Axboe return 0; 44066f3b8e2SJens Axboe } 44108d8e974SFengguang Wu 4422c136579SFengguang Wu /* 443169ebd90SJan Kara * Wait for writeback on an inode to complete. Called with i_lock held. 444169ebd90SJan Kara * Caller must make sure inode cannot go away when we drop i_lock. 44501c03194SChristoph Hellwig */ 446169ebd90SJan Kara static void __inode_wait_for_writeback(struct inode *inode) 447169ebd90SJan Kara __releases(inode->i_lock) 448169ebd90SJan Kara __acquires(inode->i_lock) 44901c03194SChristoph Hellwig { 45001c03194SChristoph Hellwig DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); 45101c03194SChristoph Hellwig wait_queue_head_t *wqh; 45201c03194SChristoph Hellwig 45301c03194SChristoph Hellwig wqh = bit_waitqueue(&inode->i_state, __I_SYNC); 45458a9d3d8SRichard Kennedy while (inode->i_state & I_SYNC) { 455250df6edSDave Chinner spin_unlock(&inode->i_lock); 45674316201SNeilBrown __wait_on_bit(wqh, &wq, bit_wait, 45774316201SNeilBrown TASK_UNINTERRUPTIBLE); 458250df6edSDave Chinner spin_lock(&inode->i_lock); 45958a9d3d8SRichard Kennedy } 46001c03194SChristoph Hellwig } 46101c03194SChristoph Hellwig 46201c03194SChristoph Hellwig /* 463169ebd90SJan Kara * Wait for writeback on an inode to complete. Caller must have inode pinned. 464169ebd90SJan Kara */ 465169ebd90SJan Kara void inode_wait_for_writeback(struct inode *inode) 466169ebd90SJan Kara { 467169ebd90SJan Kara spin_lock(&inode->i_lock); 468169ebd90SJan Kara __inode_wait_for_writeback(inode); 469169ebd90SJan Kara spin_unlock(&inode->i_lock); 470169ebd90SJan Kara } 471169ebd90SJan Kara 472169ebd90SJan Kara /* 473169ebd90SJan Kara * Sleep until I_SYNC is cleared. This function must be called with i_lock 474169ebd90SJan Kara * held and drops it. It is aimed for callers not holding any inode reference 475169ebd90SJan Kara * so once i_lock is dropped, inode can go away. 476169ebd90SJan Kara */ 477169ebd90SJan Kara static void inode_sleep_on_writeback(struct inode *inode) 478169ebd90SJan Kara __releases(inode->i_lock) 479169ebd90SJan Kara { 480169ebd90SJan Kara DEFINE_WAIT(wait); 481169ebd90SJan Kara wait_queue_head_t *wqh = bit_waitqueue(&inode->i_state, __I_SYNC); 482169ebd90SJan Kara int sleep; 483169ebd90SJan Kara 484169ebd90SJan Kara prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); 485169ebd90SJan Kara sleep = inode->i_state & I_SYNC; 486169ebd90SJan Kara spin_unlock(&inode->i_lock); 487169ebd90SJan Kara if (sleep) 488169ebd90SJan Kara schedule(); 489169ebd90SJan Kara finish_wait(wqh, &wait); 490169ebd90SJan Kara } 491169ebd90SJan Kara 492169ebd90SJan Kara /* 493ccb26b5aSJan Kara * Find proper writeback list for the inode depending on its current state and 494ccb26b5aSJan Kara * possibly also change of its state while we were doing writeback. Here we 495ccb26b5aSJan Kara * handle things such as livelock prevention or fairness of writeback among 496ccb26b5aSJan Kara * inodes. This function can be called only by flusher thread - noone else 497ccb26b5aSJan Kara * processes all inodes in writeback lists and requeueing inodes behind flusher 498ccb26b5aSJan Kara * thread's back can have unexpected consequences. 499ccb26b5aSJan Kara */ 500ccb26b5aSJan Kara static void requeue_inode(struct inode *inode, struct bdi_writeback *wb, 501ccb26b5aSJan Kara struct writeback_control *wbc) 502ccb26b5aSJan Kara { 503ccb26b5aSJan Kara if (inode->i_state & I_FREEING) 504ccb26b5aSJan Kara return; 505ccb26b5aSJan Kara 506ccb26b5aSJan Kara /* 507ccb26b5aSJan Kara * Sync livelock prevention. Each inode is tagged and synced in one 508ccb26b5aSJan Kara * shot. If still dirty, it will be redirty_tail()'ed below. Update 509ccb26b5aSJan Kara * the dirty time to prevent enqueue and sync it again. 510ccb26b5aSJan Kara */ 511ccb26b5aSJan Kara if ((inode->i_state & I_DIRTY) && 512ccb26b5aSJan Kara (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)) 513ccb26b5aSJan Kara inode->dirtied_when = jiffies; 514ccb26b5aSJan Kara 5154f8ad655SJan Kara if (wbc->pages_skipped) { 5164f8ad655SJan Kara /* 5174f8ad655SJan Kara * writeback is not making progress due to locked 5184f8ad655SJan Kara * buffers. Skip this inode for now. 5194f8ad655SJan Kara */ 5204f8ad655SJan Kara redirty_tail(inode, wb); 5214f8ad655SJan Kara return; 5224f8ad655SJan Kara } 5234f8ad655SJan Kara 524ccb26b5aSJan Kara if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) { 525ccb26b5aSJan Kara /* 526ccb26b5aSJan Kara * We didn't write back all the pages. nfs_writepages() 527ccb26b5aSJan Kara * sometimes bales out without doing anything. 528ccb26b5aSJan Kara */ 529ccb26b5aSJan Kara if (wbc->nr_to_write <= 0) { 530ccb26b5aSJan Kara /* Slice used up. Queue for next turn. */ 531ccb26b5aSJan Kara requeue_io(inode, wb); 532ccb26b5aSJan Kara } else { 533ccb26b5aSJan Kara /* 534ccb26b5aSJan Kara * Writeback blocked by something other than 535ccb26b5aSJan Kara * congestion. Delay the inode for some time to 536ccb26b5aSJan Kara * avoid spinning on the CPU (100% iowait) 537ccb26b5aSJan Kara * retrying writeback of the dirty page/inode 538ccb26b5aSJan Kara * that cannot be performed immediately. 539ccb26b5aSJan Kara */ 540ccb26b5aSJan Kara redirty_tail(inode, wb); 541ccb26b5aSJan Kara } 542ccb26b5aSJan Kara } else if (inode->i_state & I_DIRTY) { 543ccb26b5aSJan Kara /* 544ccb26b5aSJan Kara * Filesystems can dirty the inode during writeback operations, 545ccb26b5aSJan Kara * such as delayed allocation during submission or metadata 546ccb26b5aSJan Kara * updates after data IO completion. 547ccb26b5aSJan Kara */ 548ccb26b5aSJan Kara redirty_tail(inode, wb); 5490ae45f63STheodore Ts'o } else if (inode->i_state & I_DIRTY_TIME) { 550a2f48706STheodore Ts'o inode->dirtied_when = jiffies; 551d6c10f1fSTejun Heo inode_wb_list_move_locked(inode, wb, &wb->b_dirty_time); 552ccb26b5aSJan Kara } else { 553ccb26b5aSJan Kara /* The inode is clean. Remove from writeback lists. */ 554d6c10f1fSTejun Heo inode_wb_list_del_locked(inode, wb); 555ccb26b5aSJan Kara } 556ccb26b5aSJan Kara } 557ccb26b5aSJan Kara 558ccb26b5aSJan Kara /* 5594f8ad655SJan Kara * Write out an inode and its dirty pages. Do not update the writeback list 5604f8ad655SJan Kara * linkage. That is left to the caller. The caller is also responsible for 5614f8ad655SJan Kara * setting I_SYNC flag and calling inode_sync_complete() to clear it. 5621da177e4SLinus Torvalds */ 5631da177e4SLinus Torvalds static int 564cd8ed2a4SYan Hong __writeback_single_inode(struct inode *inode, struct writeback_control *wbc) 5651da177e4SLinus Torvalds { 5661da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 567251d6a47SWu Fengguang long nr_to_write = wbc->nr_to_write; 56801c03194SChristoph Hellwig unsigned dirty; 5691da177e4SLinus Torvalds int ret; 5701da177e4SLinus Torvalds 5714f8ad655SJan Kara WARN_ON(!(inode->i_state & I_SYNC)); 5721da177e4SLinus Torvalds 5739fb0a7daSTejun Heo trace_writeback_single_inode_start(inode, wbc, nr_to_write); 5749fb0a7daSTejun Heo 5751da177e4SLinus Torvalds ret = do_writepages(mapping, wbc); 5761da177e4SLinus Torvalds 57726821ed4SChristoph Hellwig /* 57826821ed4SChristoph Hellwig * Make sure to wait on the data before writing out the metadata. 57926821ed4SChristoph Hellwig * This is important for filesystems that modify metadata on data 5807747bd4bSDave Chinner * I/O completion. We don't do it for sync(2) writeback because it has a 5817747bd4bSDave Chinner * separate, external IO completion path and ->sync_fs for guaranteeing 5827747bd4bSDave Chinner * inode metadata is written back correctly. 58326821ed4SChristoph Hellwig */ 5847747bd4bSDave Chinner if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) { 58526821ed4SChristoph Hellwig int err = filemap_fdatawait(mapping); 5861da177e4SLinus Torvalds if (ret == 0) 5871da177e4SLinus Torvalds ret = err; 5881da177e4SLinus Torvalds } 5891da177e4SLinus Torvalds 5905547e8aaSDmitry Monakhov /* 5915547e8aaSDmitry Monakhov * Some filesystems may redirty the inode during the writeback 5925547e8aaSDmitry Monakhov * due to delalloc, clear dirty metadata flags right before 5935547e8aaSDmitry Monakhov * write_inode() 5945547e8aaSDmitry Monakhov */ 595250df6edSDave Chinner spin_lock(&inode->i_lock); 5969c6ac78eSTejun Heo 5975547e8aaSDmitry Monakhov dirty = inode->i_state & I_DIRTY; 598a2f48706STheodore Ts'o if (inode->i_state & I_DIRTY_TIME) { 599a2f48706STheodore Ts'o if ((dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) || 600a2f48706STheodore Ts'o unlikely(inode->i_state & I_DIRTY_TIME_EXPIRED) || 601a2f48706STheodore Ts'o unlikely(time_after(jiffies, 602a2f48706STheodore Ts'o (inode->dirtied_time_when + 603a2f48706STheodore Ts'o dirtytime_expire_interval * HZ)))) { 6040ae45f63STheodore Ts'o dirty |= I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED; 6050ae45f63STheodore Ts'o trace_writeback_lazytime(inode); 6060ae45f63STheodore Ts'o } 607a2f48706STheodore Ts'o } else 608a2f48706STheodore Ts'o inode->i_state &= ~I_DIRTY_TIME_EXPIRED; 6090ae45f63STheodore Ts'o inode->i_state &= ~dirty; 6109c6ac78eSTejun Heo 6119c6ac78eSTejun Heo /* 6129c6ac78eSTejun Heo * Paired with smp_mb() in __mark_inode_dirty(). This allows 6139c6ac78eSTejun Heo * __mark_inode_dirty() to test i_state without grabbing i_lock - 6149c6ac78eSTejun Heo * either they see the I_DIRTY bits cleared or we see the dirtied 6159c6ac78eSTejun Heo * inode. 6169c6ac78eSTejun Heo * 6179c6ac78eSTejun Heo * I_DIRTY_PAGES is always cleared together above even if @mapping 6189c6ac78eSTejun Heo * still has dirty pages. The flag is reinstated after smp_mb() if 6199c6ac78eSTejun Heo * necessary. This guarantees that either __mark_inode_dirty() 6209c6ac78eSTejun Heo * sees clear I_DIRTY_PAGES or we see PAGECACHE_TAG_DIRTY. 6219c6ac78eSTejun Heo */ 6229c6ac78eSTejun Heo smp_mb(); 6239c6ac78eSTejun Heo 6249c6ac78eSTejun Heo if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) 6259c6ac78eSTejun Heo inode->i_state |= I_DIRTY_PAGES; 6269c6ac78eSTejun Heo 627250df6edSDave Chinner spin_unlock(&inode->i_lock); 6289c6ac78eSTejun Heo 6290ae45f63STheodore Ts'o if (dirty & I_DIRTY_TIME) 6300ae45f63STheodore Ts'o mark_inode_dirty_sync(inode); 63126821ed4SChristoph Hellwig /* Don't write the inode if only I_DIRTY_PAGES was set */ 6320ae45f63STheodore Ts'o if (dirty & ~I_DIRTY_PAGES) { 633a9185b41SChristoph Hellwig int err = write_inode(inode, wbc); 6341da177e4SLinus Torvalds if (ret == 0) 6351da177e4SLinus Torvalds ret = err; 6361da177e4SLinus Torvalds } 6374f8ad655SJan Kara trace_writeback_single_inode(inode, wbc, nr_to_write); 6384f8ad655SJan Kara return ret; 6394f8ad655SJan Kara } 6404f8ad655SJan Kara 6414f8ad655SJan Kara /* 6424f8ad655SJan Kara * Write out an inode's dirty pages. Either the caller has an active reference 6434f8ad655SJan Kara * on the inode or the inode has I_WILL_FREE set. 6444f8ad655SJan Kara * 6454f8ad655SJan Kara * This function is designed to be called for writing back one inode which 6464f8ad655SJan Kara * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode() 6474f8ad655SJan Kara * and does more profound writeback list handling in writeback_sb_inodes(). 6484f8ad655SJan Kara */ 6494f8ad655SJan Kara static int 6504f8ad655SJan Kara writeback_single_inode(struct inode *inode, struct bdi_writeback *wb, 6514f8ad655SJan Kara struct writeback_control *wbc) 6524f8ad655SJan Kara { 6534f8ad655SJan Kara int ret = 0; 6544f8ad655SJan Kara 6554f8ad655SJan Kara spin_lock(&inode->i_lock); 6564f8ad655SJan Kara if (!atomic_read(&inode->i_count)) 6574f8ad655SJan Kara WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); 6584f8ad655SJan Kara else 6594f8ad655SJan Kara WARN_ON(inode->i_state & I_WILL_FREE); 6604f8ad655SJan Kara 6614f8ad655SJan Kara if (inode->i_state & I_SYNC) { 6624f8ad655SJan Kara if (wbc->sync_mode != WB_SYNC_ALL) 6634f8ad655SJan Kara goto out; 6644f8ad655SJan Kara /* 665169ebd90SJan Kara * It's a data-integrity sync. We must wait. Since callers hold 666169ebd90SJan Kara * inode reference or inode has I_WILL_FREE set, it cannot go 667169ebd90SJan Kara * away under us. 6684f8ad655SJan Kara */ 669169ebd90SJan Kara __inode_wait_for_writeback(inode); 6704f8ad655SJan Kara } 6714f8ad655SJan Kara WARN_ON(inode->i_state & I_SYNC); 6724f8ad655SJan Kara /* 673f9b0e058SJan Kara * Skip inode if it is clean and we have no outstanding writeback in 674f9b0e058SJan Kara * WB_SYNC_ALL mode. We don't want to mess with writeback lists in this 675f9b0e058SJan Kara * function since flusher thread may be doing for example sync in 676f9b0e058SJan Kara * parallel and if we move the inode, it could get skipped. So here we 677f9b0e058SJan Kara * make sure inode is on some writeback list and leave it there unless 678f9b0e058SJan Kara * we have completely cleaned the inode. 6794f8ad655SJan Kara */ 6800ae45f63STheodore Ts'o if (!(inode->i_state & I_DIRTY_ALL) && 681f9b0e058SJan Kara (wbc->sync_mode != WB_SYNC_ALL || 682f9b0e058SJan Kara !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK))) 6834f8ad655SJan Kara goto out; 6844f8ad655SJan Kara inode->i_state |= I_SYNC; 6854f8ad655SJan Kara spin_unlock(&inode->i_lock); 6864f8ad655SJan Kara 687cd8ed2a4SYan Hong ret = __writeback_single_inode(inode, wbc); 6881da177e4SLinus Torvalds 689f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 690250df6edSDave Chinner spin_lock(&inode->i_lock); 6914f8ad655SJan Kara /* 6924f8ad655SJan Kara * If inode is clean, remove it from writeback lists. Otherwise don't 6934f8ad655SJan Kara * touch it. See comment above for explanation. 6944f8ad655SJan Kara */ 6950ae45f63STheodore Ts'o if (!(inode->i_state & I_DIRTY_ALL)) 696d6c10f1fSTejun Heo inode_wb_list_del_locked(inode, wb); 6974f8ad655SJan Kara spin_unlock(&wb->list_lock); 6981c0eeaf5SJoern Engel inode_sync_complete(inode); 6994f8ad655SJan Kara out: 7004f8ad655SJan Kara spin_unlock(&inode->i_lock); 7011da177e4SLinus Torvalds return ret; 7021da177e4SLinus Torvalds } 7031da177e4SLinus Torvalds 704a88a341aSTejun Heo static long writeback_chunk_size(struct bdi_writeback *wb, 7051a12d8bdSWu Fengguang struct wb_writeback_work *work) 706d46db3d5SWu Fengguang { 707d46db3d5SWu Fengguang long pages; 708d46db3d5SWu Fengguang 709d46db3d5SWu Fengguang /* 710d46db3d5SWu Fengguang * WB_SYNC_ALL mode does livelock avoidance by syncing dirty 711d46db3d5SWu Fengguang * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX 712d46db3d5SWu Fengguang * here avoids calling into writeback_inodes_wb() more than once. 713d46db3d5SWu Fengguang * 714d46db3d5SWu Fengguang * The intended call sequence for WB_SYNC_ALL writeback is: 715d46db3d5SWu Fengguang * 716d46db3d5SWu Fengguang * wb_writeback() 717d46db3d5SWu Fengguang * writeback_sb_inodes() <== called only once 718d46db3d5SWu Fengguang * write_cache_pages() <== called once for each inode 719d46db3d5SWu Fengguang * (quickly) tag currently dirty pages 720d46db3d5SWu Fengguang * (maybe slowly) sync all tagged pages 721d46db3d5SWu Fengguang */ 722d46db3d5SWu Fengguang if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages) 723d46db3d5SWu Fengguang pages = LONG_MAX; 7241a12d8bdSWu Fengguang else { 725a88a341aSTejun Heo pages = min(wb->avg_write_bandwidth / 2, 7261a12d8bdSWu Fengguang global_dirty_limit / DIRTY_SCOPE); 7271a12d8bdSWu Fengguang pages = min(pages, work->nr_pages); 7281a12d8bdSWu Fengguang pages = round_down(pages + MIN_WRITEBACK_PAGES, 7291a12d8bdSWu Fengguang MIN_WRITEBACK_PAGES); 7301a12d8bdSWu Fengguang } 731d46db3d5SWu Fengguang 732d46db3d5SWu Fengguang return pages; 733d46db3d5SWu Fengguang } 734d46db3d5SWu Fengguang 73503ba3782SJens Axboe /* 736f11c9c5cSEdward Shishkin * Write a portion of b_io inodes which belong to @sb. 737edadfb10SChristoph Hellwig * 738d46db3d5SWu Fengguang * Return the number of pages and/or inodes written. 739f11c9c5cSEdward Shishkin */ 740d46db3d5SWu Fengguang static long writeback_sb_inodes(struct super_block *sb, 741d46db3d5SWu Fengguang struct bdi_writeback *wb, 742d46db3d5SWu Fengguang struct wb_writeback_work *work) 74303ba3782SJens Axboe { 744d46db3d5SWu Fengguang struct writeback_control wbc = { 745d46db3d5SWu Fengguang .sync_mode = work->sync_mode, 746d46db3d5SWu Fengguang .tagged_writepages = work->tagged_writepages, 747d46db3d5SWu Fengguang .for_kupdate = work->for_kupdate, 748d46db3d5SWu Fengguang .for_background = work->for_background, 7497747bd4bSDave Chinner .for_sync = work->for_sync, 750d46db3d5SWu Fengguang .range_cyclic = work->range_cyclic, 751d46db3d5SWu Fengguang .range_start = 0, 752d46db3d5SWu Fengguang .range_end = LLONG_MAX, 753d46db3d5SWu Fengguang }; 754d46db3d5SWu Fengguang unsigned long start_time = jiffies; 755d46db3d5SWu Fengguang long write_chunk; 756d46db3d5SWu Fengguang long wrote = 0; /* count both pages and inodes */ 757d46db3d5SWu Fengguang 75803ba3782SJens Axboe while (!list_empty(&wb->b_io)) { 7597ccf19a8SNick Piggin struct inode *inode = wb_inode(wb->b_io.prev); 760edadfb10SChristoph Hellwig 761edadfb10SChristoph Hellwig if (inode->i_sb != sb) { 762d46db3d5SWu Fengguang if (work->sb) { 763edadfb10SChristoph Hellwig /* 764edadfb10SChristoph Hellwig * We only want to write back data for this 765edadfb10SChristoph Hellwig * superblock, move all inodes not belonging 766edadfb10SChristoph Hellwig * to it back onto the dirty list. 767edadfb10SChristoph Hellwig */ 768f758eeabSChristoph Hellwig redirty_tail(inode, wb); 76966f3b8e2SJens Axboe continue; 77066f3b8e2SJens Axboe } 771edadfb10SChristoph Hellwig 772edadfb10SChristoph Hellwig /* 773edadfb10SChristoph Hellwig * The inode belongs to a different superblock. 774edadfb10SChristoph Hellwig * Bounce back to the caller to unpin this and 775edadfb10SChristoph Hellwig * pin the next superblock. 776edadfb10SChristoph Hellwig */ 777d46db3d5SWu Fengguang break; 778edadfb10SChristoph Hellwig } 779edadfb10SChristoph Hellwig 7809843b76aSChristoph Hellwig /* 781331cbdeeSWanpeng Li * Don't bother with new inodes or inodes being freed, first 782331cbdeeSWanpeng Li * kind does not need periodic writeout yet, and for the latter 7839843b76aSChristoph Hellwig * kind writeout is handled by the freer. 7849843b76aSChristoph Hellwig */ 785250df6edSDave Chinner spin_lock(&inode->i_lock); 7869843b76aSChristoph Hellwig if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { 787250df6edSDave Chinner spin_unlock(&inode->i_lock); 788fcc5c222SWu Fengguang redirty_tail(inode, wb); 7897ef0d737SNick Piggin continue; 7907ef0d737SNick Piggin } 791cc1676d9SJan Kara if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) { 792cc1676d9SJan Kara /* 793cc1676d9SJan Kara * If this inode is locked for writeback and we are not 794cc1676d9SJan Kara * doing writeback-for-data-integrity, move it to 795cc1676d9SJan Kara * b_more_io so that writeback can proceed with the 796cc1676d9SJan Kara * other inodes on s_io. 797cc1676d9SJan Kara * 798cc1676d9SJan Kara * We'll have another go at writing back this inode 799cc1676d9SJan Kara * when we completed a full scan of b_io. 800cc1676d9SJan Kara */ 801cc1676d9SJan Kara spin_unlock(&inode->i_lock); 802cc1676d9SJan Kara requeue_io(inode, wb); 803cc1676d9SJan Kara trace_writeback_sb_inodes_requeue(inode); 804cc1676d9SJan Kara continue; 805cc1676d9SJan Kara } 806f0d07b7fSJan Kara spin_unlock(&wb->list_lock); 807f0d07b7fSJan Kara 8084f8ad655SJan Kara /* 8094f8ad655SJan Kara * We already requeued the inode if it had I_SYNC set and we 8104f8ad655SJan Kara * are doing WB_SYNC_NONE writeback. So this catches only the 8114f8ad655SJan Kara * WB_SYNC_ALL case. 8124f8ad655SJan Kara */ 813169ebd90SJan Kara if (inode->i_state & I_SYNC) { 814169ebd90SJan Kara /* Wait for I_SYNC. This function drops i_lock... */ 815169ebd90SJan Kara inode_sleep_on_writeback(inode); 816169ebd90SJan Kara /* Inode may be gone, start again */ 817ead188f9SJan Kara spin_lock(&wb->list_lock); 818169ebd90SJan Kara continue; 819169ebd90SJan Kara } 8204f8ad655SJan Kara inode->i_state |= I_SYNC; 8214f8ad655SJan Kara spin_unlock(&inode->i_lock); 822169ebd90SJan Kara 823a88a341aSTejun Heo write_chunk = writeback_chunk_size(wb, work); 824d46db3d5SWu Fengguang wbc.nr_to_write = write_chunk; 825d46db3d5SWu Fengguang wbc.pages_skipped = 0; 826250df6edSDave Chinner 827169ebd90SJan Kara /* 828169ebd90SJan Kara * We use I_SYNC to pin the inode in memory. While it is set 829169ebd90SJan Kara * evict_inode() will wait so the inode cannot be freed. 830169ebd90SJan Kara */ 831cd8ed2a4SYan Hong __writeback_single_inode(inode, &wbc); 832d46db3d5SWu Fengguang 833d46db3d5SWu Fengguang work->nr_pages -= write_chunk - wbc.nr_to_write; 834d46db3d5SWu Fengguang wrote += write_chunk - wbc.nr_to_write; 8354f8ad655SJan Kara spin_lock(&wb->list_lock); 8364f8ad655SJan Kara spin_lock(&inode->i_lock); 8370ae45f63STheodore Ts'o if (!(inode->i_state & I_DIRTY_ALL)) 838d46db3d5SWu Fengguang wrote++; 8394f8ad655SJan Kara requeue_inode(inode, wb, &wbc); 8404f8ad655SJan Kara inode_sync_complete(inode); 8410f1b1fd8SDave Chinner spin_unlock(&inode->i_lock); 842169ebd90SJan Kara cond_resched_lock(&wb->list_lock); 843d46db3d5SWu Fengguang /* 844d46db3d5SWu Fengguang * bail out to wb_writeback() often enough to check 845d46db3d5SWu Fengguang * background threshold and other termination conditions. 846d46db3d5SWu Fengguang */ 847d46db3d5SWu Fengguang if (wrote) { 848d46db3d5SWu Fengguang if (time_is_before_jiffies(start_time + HZ / 10UL)) 849d46db3d5SWu Fengguang break; 850d46db3d5SWu Fengguang if (work->nr_pages <= 0) 851d46db3d5SWu Fengguang break; 8521da177e4SLinus Torvalds } 8538bc3be27SFengguang Wu } 854d46db3d5SWu Fengguang return wrote; 855f11c9c5cSEdward Shishkin } 85638f21977SNick Piggin 857d46db3d5SWu Fengguang static long __writeback_inodes_wb(struct bdi_writeback *wb, 858d46db3d5SWu Fengguang struct wb_writeback_work *work) 859f11c9c5cSEdward Shishkin { 860d46db3d5SWu Fengguang unsigned long start_time = jiffies; 861d46db3d5SWu Fengguang long wrote = 0; 862f11c9c5cSEdward Shishkin 863f11c9c5cSEdward Shishkin while (!list_empty(&wb->b_io)) { 8647ccf19a8SNick Piggin struct inode *inode = wb_inode(wb->b_io.prev); 865f11c9c5cSEdward Shishkin struct super_block *sb = inode->i_sb; 866f11c9c5cSEdward Shishkin 867eb6ef3dfSKonstantin Khlebnikov if (!trylock_super(sb)) { 8680e995816SWu Fengguang /* 869eb6ef3dfSKonstantin Khlebnikov * trylock_super() may fail consistently due to 8700e995816SWu Fengguang * s_umount being grabbed by someone else. Don't use 8710e995816SWu Fengguang * requeue_io() to avoid busy retrying the inode/sb. 8720e995816SWu Fengguang */ 8730e995816SWu Fengguang redirty_tail(inode, wb); 874d19de7edSChristoph Hellwig continue; 875334132aeSChristoph Hellwig } 876d46db3d5SWu Fengguang wrote += writeback_sb_inodes(sb, wb, work); 877eb6ef3dfSKonstantin Khlebnikov up_read(&sb->s_umount); 878f11c9c5cSEdward Shishkin 879d46db3d5SWu Fengguang /* refer to the same tests at the end of writeback_sb_inodes */ 880d46db3d5SWu Fengguang if (wrote) { 881d46db3d5SWu Fengguang if (time_is_before_jiffies(start_time + HZ / 10UL)) 882d46db3d5SWu Fengguang break; 883d46db3d5SWu Fengguang if (work->nr_pages <= 0) 884f11c9c5cSEdward Shishkin break; 885f11c9c5cSEdward Shishkin } 886d46db3d5SWu Fengguang } 88766f3b8e2SJens Axboe /* Leave any unwritten inodes on b_io */ 888d46db3d5SWu Fengguang return wrote; 88966f3b8e2SJens Axboe } 89066f3b8e2SJens Axboe 8917d9f073bSWanpeng Li static long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages, 8920e175a18SCurt Wohlgemuth enum wb_reason reason) 893edadfb10SChristoph Hellwig { 894d46db3d5SWu Fengguang struct wb_writeback_work work = { 895d46db3d5SWu Fengguang .nr_pages = nr_pages, 896d46db3d5SWu Fengguang .sync_mode = WB_SYNC_NONE, 897d46db3d5SWu Fengguang .range_cyclic = 1, 8980e175a18SCurt Wohlgemuth .reason = reason, 899d46db3d5SWu Fengguang }; 900edadfb10SChristoph Hellwig 901f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 902424b351fSWu Fengguang if (list_empty(&wb->b_io)) 903ad4e38ddSCurt Wohlgemuth queue_io(wb, &work); 904d46db3d5SWu Fengguang __writeback_inodes_wb(wb, &work); 905f758eeabSChristoph Hellwig spin_unlock(&wb->list_lock); 906edadfb10SChristoph Hellwig 907d46db3d5SWu Fengguang return nr_pages - work.nr_pages; 90866f3b8e2SJens Axboe } 90966f3b8e2SJens Axboe 910a88a341aSTejun Heo static bool over_bground_thresh(struct bdi_writeback *wb) 91103ba3782SJens Axboe { 91203ba3782SJens Axboe unsigned long background_thresh, dirty_thresh; 91303ba3782SJens Axboe 91416c4042fSWu Fengguang global_dirty_limits(&background_thresh, &dirty_thresh); 91503ba3782SJens Axboe 916b00949aaSWu Fengguang if (global_page_state(NR_FILE_DIRTY) + 917b00949aaSWu Fengguang global_page_state(NR_UNSTABLE_NFS) > background_thresh) 918b00949aaSWu Fengguang return true; 919b00949aaSWu Fengguang 920a88a341aSTejun Heo if (wb_stat(wb, WB_RECLAIMABLE) > wb_dirty_limit(wb, background_thresh)) 921b00949aaSWu Fengguang return true; 922b00949aaSWu Fengguang 923b00949aaSWu Fengguang return false; 92403ba3782SJens Axboe } 92503ba3782SJens Axboe 92603ba3782SJens Axboe /* 927e98be2d5SWu Fengguang * Called under wb->list_lock. If there are multiple wb per bdi, 928e98be2d5SWu Fengguang * only the flusher working on the first wb should do it. 929e98be2d5SWu Fengguang */ 930e98be2d5SWu Fengguang static void wb_update_bandwidth(struct bdi_writeback *wb, 931e98be2d5SWu Fengguang unsigned long start_time) 932e98be2d5SWu Fengguang { 933a88a341aSTejun Heo __wb_update_bandwidth(wb, 0, 0, 0, 0, 0, start_time); 934e98be2d5SWu Fengguang } 935e98be2d5SWu Fengguang 936e98be2d5SWu Fengguang /* 93703ba3782SJens Axboe * Explicit flushing or periodic writeback of "old" data. 93803ba3782SJens Axboe * 93903ba3782SJens Axboe * Define "old": the first time one of an inode's pages is dirtied, we mark the 94003ba3782SJens Axboe * dirtying-time in the inode's address_space. So this periodic writeback code 94103ba3782SJens Axboe * just walks the superblock inode list, writing back any inodes which are 94203ba3782SJens Axboe * older than a specific point in time. 94303ba3782SJens Axboe * 94403ba3782SJens Axboe * Try to run once per dirty_writeback_interval. But if a writeback event 94503ba3782SJens Axboe * takes longer than a dirty_writeback_interval interval, then leave a 94603ba3782SJens Axboe * one-second gap. 94703ba3782SJens Axboe * 94803ba3782SJens Axboe * older_than_this takes precedence over nr_to_write. So we'll only write back 94903ba3782SJens Axboe * all dirty pages if they are all attached to "old" mappings. 95003ba3782SJens Axboe */ 951c4a77a6cSJens Axboe static long wb_writeback(struct bdi_writeback *wb, 95283ba7b07SChristoph Hellwig struct wb_writeback_work *work) 95303ba3782SJens Axboe { 954e98be2d5SWu Fengguang unsigned long wb_start = jiffies; 955d46db3d5SWu Fengguang long nr_pages = work->nr_pages; 9560dc83bd3SJan Kara unsigned long oldest_jif; 957a5989bdcSJan Kara struct inode *inode; 958d46db3d5SWu Fengguang long progress; 95903ba3782SJens Axboe 9600dc83bd3SJan Kara oldest_jif = jiffies; 9610dc83bd3SJan Kara work->older_than_this = &oldest_jif; 96203ba3782SJens Axboe 963e8dfc305SWu Fengguang spin_lock(&wb->list_lock); 96403ba3782SJens Axboe for (;;) { 96503ba3782SJens Axboe /* 966d3ddec76SWu Fengguang * Stop writeback when nr_pages has been consumed 96703ba3782SJens Axboe */ 96883ba7b07SChristoph Hellwig if (work->nr_pages <= 0) 96903ba3782SJens Axboe break; 97003ba3782SJens Axboe 97103ba3782SJens Axboe /* 972aa373cf5SJan Kara * Background writeout and kupdate-style writeback may 973aa373cf5SJan Kara * run forever. Stop them if there is other work to do 974aa373cf5SJan Kara * so that e.g. sync can proceed. They'll be restarted 975aa373cf5SJan Kara * after the other works are all done. 976aa373cf5SJan Kara */ 977aa373cf5SJan Kara if ((work->for_background || work->for_kupdate) && 978f0054bb1STejun Heo !list_empty(&wb->work_list)) 979aa373cf5SJan Kara break; 980aa373cf5SJan Kara 981aa373cf5SJan Kara /* 982d3ddec76SWu Fengguang * For background writeout, stop when we are below the 983d3ddec76SWu Fengguang * background dirty threshold 98403ba3782SJens Axboe */ 985a88a341aSTejun Heo if (work->for_background && !over_bground_thresh(wb)) 98603ba3782SJens Axboe break; 98703ba3782SJens Axboe 9881bc36b64SJan Kara /* 9891bc36b64SJan Kara * Kupdate and background works are special and we want to 9901bc36b64SJan Kara * include all inodes that need writing. Livelock avoidance is 9911bc36b64SJan Kara * handled by these works yielding to any other work so we are 9921bc36b64SJan Kara * safe. 9931bc36b64SJan Kara */ 994ba9aa839SWu Fengguang if (work->for_kupdate) { 9950dc83bd3SJan Kara oldest_jif = jiffies - 996ba9aa839SWu Fengguang msecs_to_jiffies(dirty_expire_interval * 10); 9971bc36b64SJan Kara } else if (work->for_background) 9980dc83bd3SJan Kara oldest_jif = jiffies; 999028c2dd1SDave Chinner 1000d46db3d5SWu Fengguang trace_writeback_start(wb->bdi, work); 1001e8dfc305SWu Fengguang if (list_empty(&wb->b_io)) 1002ad4e38ddSCurt Wohlgemuth queue_io(wb, work); 100383ba7b07SChristoph Hellwig if (work->sb) 1004d46db3d5SWu Fengguang progress = writeback_sb_inodes(work->sb, wb, work); 1005edadfb10SChristoph Hellwig else 1006d46db3d5SWu Fengguang progress = __writeback_inodes_wb(wb, work); 1007d46db3d5SWu Fengguang trace_writeback_written(wb->bdi, work); 1008028c2dd1SDave Chinner 1009e98be2d5SWu Fengguang wb_update_bandwidth(wb, wb_start); 101003ba3782SJens Axboe 101103ba3782SJens Axboe /* 101271fd05a8SJens Axboe * Did we write something? Try for more 1013e6fb6da2SWu Fengguang * 1014e6fb6da2SWu Fengguang * Dirty inodes are moved to b_io for writeback in batches. 1015e6fb6da2SWu Fengguang * The completion of the current batch does not necessarily 1016e6fb6da2SWu Fengguang * mean the overall work is done. So we keep looping as long 1017e6fb6da2SWu Fengguang * as made some progress on cleaning pages or inodes. 101871fd05a8SJens Axboe */ 1019d46db3d5SWu Fengguang if (progress) 102003ba3782SJens Axboe continue; 1021a5989bdcSJan Kara /* 1022e6fb6da2SWu Fengguang * No more inodes for IO, bail 1023a5989bdcSJan Kara */ 1024b7a2441fSWu Fengguang if (list_empty(&wb->b_more_io)) 102503ba3782SJens Axboe break; 102603ba3782SJens Axboe /* 10278010c3b6SJens Axboe * Nothing written. Wait for some inode to 10288010c3b6SJens Axboe * become available for writeback. Otherwise 10298010c3b6SJens Axboe * we'll just busyloop. 103003ba3782SJens Axboe */ 103103ba3782SJens Axboe if (!list_empty(&wb->b_more_io)) { 1032d46db3d5SWu Fengguang trace_writeback_wait(wb->bdi, work); 103303ba3782SJens Axboe inode = wb_inode(wb->b_more_io.prev); 1034250df6edSDave Chinner spin_lock(&inode->i_lock); 1035f0d07b7fSJan Kara spin_unlock(&wb->list_lock); 1036169ebd90SJan Kara /* This function drops i_lock... */ 1037169ebd90SJan Kara inode_sleep_on_writeback(inode); 1038f0d07b7fSJan Kara spin_lock(&wb->list_lock); 103903ba3782SJens Axboe } 104003ba3782SJens Axboe } 1041e8dfc305SWu Fengguang spin_unlock(&wb->list_lock); 104203ba3782SJens Axboe 1043d46db3d5SWu Fengguang return nr_pages - work->nr_pages; 104403ba3782SJens Axboe } 104503ba3782SJens Axboe 104603ba3782SJens Axboe /* 104783ba7b07SChristoph Hellwig * Return the next wb_writeback_work struct that hasn't been processed yet. 104803ba3782SJens Axboe */ 1049f0054bb1STejun Heo static struct wb_writeback_work *get_next_work_item(struct bdi_writeback *wb) 105003ba3782SJens Axboe { 105183ba7b07SChristoph Hellwig struct wb_writeback_work *work = NULL; 105203ba3782SJens Axboe 1053f0054bb1STejun Heo spin_lock_bh(&wb->work_lock); 1054f0054bb1STejun Heo if (!list_empty(&wb->work_list)) { 1055f0054bb1STejun Heo work = list_entry(wb->work_list.next, 105683ba7b07SChristoph Hellwig struct wb_writeback_work, list); 105783ba7b07SChristoph Hellwig list_del_init(&work->list); 105803ba3782SJens Axboe } 1059f0054bb1STejun Heo spin_unlock_bh(&wb->work_lock); 106083ba7b07SChristoph Hellwig return work; 106103ba3782SJens Axboe } 106203ba3782SJens Axboe 1063cdf01dd5SLinus Torvalds /* 1064cdf01dd5SLinus Torvalds * Add in the number of potentially dirty inodes, because each inode 1065cdf01dd5SLinus Torvalds * write can dirty pagecache in the underlying blockdev. 1066cdf01dd5SLinus Torvalds */ 1067cdf01dd5SLinus Torvalds static unsigned long get_nr_dirty_pages(void) 1068cdf01dd5SLinus Torvalds { 1069cdf01dd5SLinus Torvalds return global_page_state(NR_FILE_DIRTY) + 1070cdf01dd5SLinus Torvalds global_page_state(NR_UNSTABLE_NFS) + 1071cdf01dd5SLinus Torvalds get_nr_dirty_inodes(); 1072cdf01dd5SLinus Torvalds } 1073cdf01dd5SLinus Torvalds 10746585027aSJan Kara static long wb_check_background_flush(struct bdi_writeback *wb) 10756585027aSJan Kara { 1076a88a341aSTejun Heo if (over_bground_thresh(wb)) { 10776585027aSJan Kara 10786585027aSJan Kara struct wb_writeback_work work = { 10796585027aSJan Kara .nr_pages = LONG_MAX, 10806585027aSJan Kara .sync_mode = WB_SYNC_NONE, 10816585027aSJan Kara .for_background = 1, 10826585027aSJan Kara .range_cyclic = 1, 10830e175a18SCurt Wohlgemuth .reason = WB_REASON_BACKGROUND, 10846585027aSJan Kara }; 10856585027aSJan Kara 10866585027aSJan Kara return wb_writeback(wb, &work); 10876585027aSJan Kara } 10886585027aSJan Kara 10896585027aSJan Kara return 0; 10906585027aSJan Kara } 10916585027aSJan Kara 109203ba3782SJens Axboe static long wb_check_old_data_flush(struct bdi_writeback *wb) 109303ba3782SJens Axboe { 109403ba3782SJens Axboe unsigned long expired; 109503ba3782SJens Axboe long nr_pages; 109603ba3782SJens Axboe 109769b62d01SJens Axboe /* 109869b62d01SJens Axboe * When set to zero, disable periodic writeback 109969b62d01SJens Axboe */ 110069b62d01SJens Axboe if (!dirty_writeback_interval) 110169b62d01SJens Axboe return 0; 110269b62d01SJens Axboe 110303ba3782SJens Axboe expired = wb->last_old_flush + 110403ba3782SJens Axboe msecs_to_jiffies(dirty_writeback_interval * 10); 110503ba3782SJens Axboe if (time_before(jiffies, expired)) 110603ba3782SJens Axboe return 0; 110703ba3782SJens Axboe 110803ba3782SJens Axboe wb->last_old_flush = jiffies; 1109cdf01dd5SLinus Torvalds nr_pages = get_nr_dirty_pages(); 111003ba3782SJens Axboe 1111c4a77a6cSJens Axboe if (nr_pages) { 111283ba7b07SChristoph Hellwig struct wb_writeback_work work = { 1113c4a77a6cSJens Axboe .nr_pages = nr_pages, 1114c4a77a6cSJens Axboe .sync_mode = WB_SYNC_NONE, 1115c4a77a6cSJens Axboe .for_kupdate = 1, 1116c4a77a6cSJens Axboe .range_cyclic = 1, 11170e175a18SCurt Wohlgemuth .reason = WB_REASON_PERIODIC, 1118c4a77a6cSJens Axboe }; 1119c4a77a6cSJens Axboe 112083ba7b07SChristoph Hellwig return wb_writeback(wb, &work); 1121c4a77a6cSJens Axboe } 112203ba3782SJens Axboe 112303ba3782SJens Axboe return 0; 112403ba3782SJens Axboe } 112503ba3782SJens Axboe 112603ba3782SJens Axboe /* 112703ba3782SJens Axboe * Retrieve work items and do the writeback they describe 112803ba3782SJens Axboe */ 112925d130baSWanpeng Li static long wb_do_writeback(struct bdi_writeback *wb) 113003ba3782SJens Axboe { 113183ba7b07SChristoph Hellwig struct wb_writeback_work *work; 1132c4a77a6cSJens Axboe long wrote = 0; 113303ba3782SJens Axboe 11344452226eSTejun Heo set_bit(WB_writeback_running, &wb->state); 1135f0054bb1STejun Heo while ((work = get_next_work_item(wb)) != NULL) { 113683ba7b07SChristoph Hellwig 1137f0054bb1STejun Heo trace_writeback_exec(wb->bdi, work); 1138455b2864SDave Chinner 113983ba7b07SChristoph Hellwig wrote += wb_writeback(wb, work); 114003ba3782SJens Axboe 114103ba3782SJens Axboe /* 114283ba7b07SChristoph Hellwig * Notify the caller of completion if this is a synchronous 114383ba7b07SChristoph Hellwig * work item, otherwise just free it. 114403ba3782SJens Axboe */ 114583ba7b07SChristoph Hellwig if (work->done) 114683ba7b07SChristoph Hellwig complete(work->done); 114783ba7b07SChristoph Hellwig else 114883ba7b07SChristoph Hellwig kfree(work); 114903ba3782SJens Axboe } 115003ba3782SJens Axboe 115103ba3782SJens Axboe /* 115203ba3782SJens Axboe * Check for periodic writeback, kupdated() style 115303ba3782SJens Axboe */ 115403ba3782SJens Axboe wrote += wb_check_old_data_flush(wb); 11556585027aSJan Kara wrote += wb_check_background_flush(wb); 11564452226eSTejun Heo clear_bit(WB_writeback_running, &wb->state); 115703ba3782SJens Axboe 115803ba3782SJens Axboe return wrote; 115903ba3782SJens Axboe } 116003ba3782SJens Axboe 116103ba3782SJens Axboe /* 116203ba3782SJens Axboe * Handle writeback of dirty data for the device backed by this bdi. Also 1163839a8e86STejun Heo * reschedules periodically and does kupdated style flushing. 116403ba3782SJens Axboe */ 1165f0054bb1STejun Heo void wb_workfn(struct work_struct *work) 116603ba3782SJens Axboe { 1167839a8e86STejun Heo struct bdi_writeback *wb = container_of(to_delayed_work(work), 1168839a8e86STejun Heo struct bdi_writeback, dwork); 116903ba3782SJens Axboe long pages_written; 117003ba3782SJens Axboe 1171f0054bb1STejun Heo set_worker_desc("flush-%s", dev_name(wb->bdi->dev)); 1172766f9164SPeter Zijlstra current->flags |= PF_SWAPWRITE; 117303ba3782SJens Axboe 1174839a8e86STejun Heo if (likely(!current_is_workqueue_rescuer() || 11754452226eSTejun Heo !test_bit(WB_registered, &wb->state))) { 117603ba3782SJens Axboe /* 1177f0054bb1STejun Heo * The normal path. Keep writing back @wb until its 1178839a8e86STejun Heo * work_list is empty. Note that this path is also taken 1179f0054bb1STejun Heo * if @wb is shutting down even when we're running off the 1180839a8e86STejun Heo * rescuer as work_list needs to be drained. 118103ba3782SJens Axboe */ 1182839a8e86STejun Heo do { 118325d130baSWanpeng Li pages_written = wb_do_writeback(wb); 1184455b2864SDave Chinner trace_writeback_pages_written(pages_written); 1185f0054bb1STejun Heo } while (!list_empty(&wb->work_list)); 1186839a8e86STejun Heo } else { 1187253c34e9SArtem Bityutskiy /* 1188839a8e86STejun Heo * bdi_wq can't get enough workers and we're running off 1189839a8e86STejun Heo * the emergency worker. Don't hog it. Hopefully, 1024 is 1190839a8e86STejun Heo * enough for efficient IO. 1191253c34e9SArtem Bityutskiy */ 1192f0054bb1STejun Heo pages_written = writeback_inodes_wb(wb, 1024, 1193839a8e86STejun Heo WB_REASON_FORKER_THREAD); 1194839a8e86STejun Heo trace_writeback_pages_written(pages_written); 119503ba3782SJens Axboe } 119603ba3782SJens Axboe 1197f0054bb1STejun Heo if (!list_empty(&wb->work_list)) 11986ca738d6SDerek Basehore mod_delayed_work(bdi_wq, &wb->dwork, 0); 11996ca738d6SDerek Basehore else if (wb_has_dirty_io(wb) && dirty_writeback_interval) 1200f0054bb1STejun Heo wb_wakeup_delayed(wb); 1201455b2864SDave Chinner 1202839a8e86STejun Heo current->flags &= ~PF_SWAPWRITE; 120303ba3782SJens Axboe } 120403ba3782SJens Axboe 120503ba3782SJens Axboe /* 120603ba3782SJens Axboe * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back 120703ba3782SJens Axboe * the whole world. 120803ba3782SJens Axboe */ 12090e175a18SCurt Wohlgemuth void wakeup_flusher_threads(long nr_pages, enum wb_reason reason) 121003ba3782SJens Axboe { 1211b8c2f347SChristoph Hellwig struct backing_dev_info *bdi; 1212b8c2f347SChristoph Hellwig 121347df3ddeSJan Kara if (!nr_pages) 121447df3ddeSJan Kara nr_pages = get_nr_dirty_pages(); 1215b8c2f347SChristoph Hellwig 1216b8c2f347SChristoph Hellwig rcu_read_lock(); 1217b8c2f347SChristoph Hellwig list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { 1218b8c2f347SChristoph Hellwig if (!bdi_has_dirty_io(bdi)) 1219b8c2f347SChristoph Hellwig continue; 1220f0054bb1STejun Heo __wb_start_writeback(&bdi->wb, nr_pages, false, reason); 1221b8c2f347SChristoph Hellwig } 1222b8c2f347SChristoph Hellwig rcu_read_unlock(); 122303ba3782SJens Axboe } 122403ba3782SJens Axboe 1225a2f48706STheodore Ts'o /* 1226a2f48706STheodore Ts'o * Wake up bdi's periodically to make sure dirtytime inodes gets 1227a2f48706STheodore Ts'o * written back periodically. We deliberately do *not* check the 1228a2f48706STheodore Ts'o * b_dirtytime list in wb_has_dirty_io(), since this would cause the 1229a2f48706STheodore Ts'o * kernel to be constantly waking up once there are any dirtytime 1230a2f48706STheodore Ts'o * inodes on the system. So instead we define a separate delayed work 1231a2f48706STheodore Ts'o * function which gets called much more rarely. (By default, only 1232a2f48706STheodore Ts'o * once every 12 hours.) 1233a2f48706STheodore Ts'o * 1234a2f48706STheodore Ts'o * If there is any other write activity going on in the file system, 1235a2f48706STheodore Ts'o * this function won't be necessary. But if the only thing that has 1236a2f48706STheodore Ts'o * happened on the file system is a dirtytime inode caused by an atime 1237a2f48706STheodore Ts'o * update, we need this infrastructure below to make sure that inode 1238a2f48706STheodore Ts'o * eventually gets pushed out to disk. 1239a2f48706STheodore Ts'o */ 1240a2f48706STheodore Ts'o static void wakeup_dirtytime_writeback(struct work_struct *w); 1241a2f48706STheodore Ts'o static DECLARE_DELAYED_WORK(dirtytime_work, wakeup_dirtytime_writeback); 1242a2f48706STheodore Ts'o 1243a2f48706STheodore Ts'o static void wakeup_dirtytime_writeback(struct work_struct *w) 1244a2f48706STheodore Ts'o { 1245a2f48706STheodore Ts'o struct backing_dev_info *bdi; 1246a2f48706STheodore Ts'o 1247a2f48706STheodore Ts'o rcu_read_lock(); 1248a2f48706STheodore Ts'o list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { 1249a2f48706STheodore Ts'o if (list_empty(&bdi->wb.b_dirty_time)) 1250a2f48706STheodore Ts'o continue; 1251f0054bb1STejun Heo wb_wakeup(&bdi->wb); 1252a2f48706STheodore Ts'o } 1253a2f48706STheodore Ts'o rcu_read_unlock(); 1254a2f48706STheodore Ts'o schedule_delayed_work(&dirtytime_work, dirtytime_expire_interval * HZ); 1255a2f48706STheodore Ts'o } 1256a2f48706STheodore Ts'o 1257a2f48706STheodore Ts'o static int __init start_dirtytime_writeback(void) 1258a2f48706STheodore Ts'o { 1259a2f48706STheodore Ts'o schedule_delayed_work(&dirtytime_work, dirtytime_expire_interval * HZ); 1260a2f48706STheodore Ts'o return 0; 1261a2f48706STheodore Ts'o } 1262a2f48706STheodore Ts'o __initcall(start_dirtytime_writeback); 1263a2f48706STheodore Ts'o 12641efff914STheodore Ts'o int dirtytime_interval_handler(struct ctl_table *table, int write, 12651efff914STheodore Ts'o void __user *buffer, size_t *lenp, loff_t *ppos) 12661efff914STheodore Ts'o { 12671efff914STheodore Ts'o int ret; 12681efff914STheodore Ts'o 12691efff914STheodore Ts'o ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos); 12701efff914STheodore Ts'o if (ret == 0 && write) 12711efff914STheodore Ts'o mod_delayed_work(system_wq, &dirtytime_work, 0); 12721efff914STheodore Ts'o return ret; 12731efff914STheodore Ts'o } 12741efff914STheodore Ts'o 127503ba3782SJens Axboe static noinline void block_dump___mark_inode_dirty(struct inode *inode) 127603ba3782SJens Axboe { 127703ba3782SJens Axboe if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { 127803ba3782SJens Axboe struct dentry *dentry; 127903ba3782SJens Axboe const char *name = "?"; 128003ba3782SJens Axboe 128103ba3782SJens Axboe dentry = d_find_alias(inode); 128203ba3782SJens Axboe if (dentry) { 128303ba3782SJens Axboe spin_lock(&dentry->d_lock); 128403ba3782SJens Axboe name = (const char *) dentry->d_name.name; 128503ba3782SJens Axboe } 128603ba3782SJens Axboe printk(KERN_DEBUG 128703ba3782SJens Axboe "%s(%d): dirtied inode %lu (%s) on %s\n", 128803ba3782SJens Axboe current->comm, task_pid_nr(current), inode->i_ino, 128903ba3782SJens Axboe name, inode->i_sb->s_id); 129003ba3782SJens Axboe if (dentry) { 129103ba3782SJens Axboe spin_unlock(&dentry->d_lock); 129203ba3782SJens Axboe dput(dentry); 129303ba3782SJens Axboe } 129403ba3782SJens Axboe } 129503ba3782SJens Axboe } 129603ba3782SJens Axboe 129703ba3782SJens Axboe /** 129803ba3782SJens Axboe * __mark_inode_dirty - internal function 129903ba3782SJens Axboe * @inode: inode to mark 130003ba3782SJens Axboe * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) 130103ba3782SJens Axboe * Mark an inode as dirty. Callers should use mark_inode_dirty or 130203ba3782SJens Axboe * mark_inode_dirty_sync. 130303ba3782SJens Axboe * 130403ba3782SJens Axboe * Put the inode on the super block's dirty list. 130503ba3782SJens Axboe * 130603ba3782SJens Axboe * CAREFUL! We mark it dirty unconditionally, but move it onto the 130703ba3782SJens Axboe * dirty list only if it is hashed or if it refers to a blockdev. 130803ba3782SJens Axboe * If it was not hashed, it will never be added to the dirty list 130903ba3782SJens Axboe * even if it is later hashed, as it will have been marked dirty already. 131003ba3782SJens Axboe * 131103ba3782SJens Axboe * In short, make sure you hash any inodes _before_ you start marking 131203ba3782SJens Axboe * them dirty. 131303ba3782SJens Axboe * 131403ba3782SJens Axboe * Note that for blockdevs, inode->dirtied_when represents the dirtying time of 131503ba3782SJens Axboe * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of 131603ba3782SJens Axboe * the kernel-internal blockdev inode represents the dirtying time of the 131703ba3782SJens Axboe * blockdev's pages. This is why for I_DIRTY_PAGES we always use 131803ba3782SJens Axboe * page->mapping->host, so the page-dirtying time is recorded in the internal 131903ba3782SJens Axboe * blockdev inode. 132003ba3782SJens Axboe */ 13210ae45f63STheodore Ts'o #define I_DIRTY_INODE (I_DIRTY_SYNC | I_DIRTY_DATASYNC) 132203ba3782SJens Axboe void __mark_inode_dirty(struct inode *inode, int flags) 132303ba3782SJens Axboe { 132403ba3782SJens Axboe struct super_block *sb = inode->i_sb; 1325253c34e9SArtem Bityutskiy struct backing_dev_info *bdi = NULL; 13260ae45f63STheodore Ts'o int dirtytime; 13270ae45f63STheodore Ts'o 13280ae45f63STheodore Ts'o trace_writeback_mark_inode_dirty(inode, flags); 132903ba3782SJens Axboe 133003ba3782SJens Axboe /* 133103ba3782SJens Axboe * Don't do this for I_DIRTY_PAGES - that doesn't actually 133203ba3782SJens Axboe * dirty the inode itself 133303ba3782SJens Axboe */ 13340ae45f63STheodore Ts'o if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC | I_DIRTY_TIME)) { 13359fb0a7daSTejun Heo trace_writeback_dirty_inode_start(inode, flags); 13369fb0a7daSTejun Heo 133703ba3782SJens Axboe if (sb->s_op->dirty_inode) 1338aa385729SChristoph Hellwig sb->s_op->dirty_inode(inode, flags); 13399fb0a7daSTejun Heo 13409fb0a7daSTejun Heo trace_writeback_dirty_inode(inode, flags); 134103ba3782SJens Axboe } 13420ae45f63STheodore Ts'o if (flags & I_DIRTY_INODE) 13430ae45f63STheodore Ts'o flags &= ~I_DIRTY_TIME; 13440ae45f63STheodore Ts'o dirtytime = flags & I_DIRTY_TIME; 134503ba3782SJens Axboe 134603ba3782SJens Axboe /* 13479c6ac78eSTejun Heo * Paired with smp_mb() in __writeback_single_inode() for the 13489c6ac78eSTejun Heo * following lockless i_state test. See there for details. 134903ba3782SJens Axboe */ 135003ba3782SJens Axboe smp_mb(); 135103ba3782SJens Axboe 13520ae45f63STheodore Ts'o if (((inode->i_state & flags) == flags) || 13530ae45f63STheodore Ts'o (dirtytime && (inode->i_state & I_DIRTY_INODE))) 135403ba3782SJens Axboe return; 135503ba3782SJens Axboe 135603ba3782SJens Axboe if (unlikely(block_dump)) 135703ba3782SJens Axboe block_dump___mark_inode_dirty(inode); 135803ba3782SJens Axboe 1359250df6edSDave Chinner spin_lock(&inode->i_lock); 13600ae45f63STheodore Ts'o if (dirtytime && (inode->i_state & I_DIRTY_INODE)) 13610ae45f63STheodore Ts'o goto out_unlock_inode; 136203ba3782SJens Axboe if ((inode->i_state & flags) != flags) { 136303ba3782SJens Axboe const int was_dirty = inode->i_state & I_DIRTY; 136403ba3782SJens Axboe 136552ebea74STejun Heo inode_attach_wb(inode, NULL); 136652ebea74STejun Heo 13670ae45f63STheodore Ts'o if (flags & I_DIRTY_INODE) 13680ae45f63STheodore Ts'o inode->i_state &= ~I_DIRTY_TIME; 136903ba3782SJens Axboe inode->i_state |= flags; 137003ba3782SJens Axboe 137103ba3782SJens Axboe /* 137203ba3782SJens Axboe * If the inode is being synced, just update its dirty state. 137303ba3782SJens Axboe * The unlocker will place the inode on the appropriate 137403ba3782SJens Axboe * superblock list, based upon its state. 137503ba3782SJens Axboe */ 137603ba3782SJens Axboe if (inode->i_state & I_SYNC) 1377250df6edSDave Chinner goto out_unlock_inode; 137803ba3782SJens Axboe 137903ba3782SJens Axboe /* 138003ba3782SJens Axboe * Only add valid (hashed) inodes to the superblock's 138103ba3782SJens Axboe * dirty list. Add blockdev inodes as well. 138203ba3782SJens Axboe */ 138303ba3782SJens Axboe if (!S_ISBLK(inode->i_mode)) { 13841d3382cbSAl Viro if (inode_unhashed(inode)) 1385250df6edSDave Chinner goto out_unlock_inode; 138603ba3782SJens Axboe } 1387a4ffdde6SAl Viro if (inode->i_state & I_FREEING) 1388250df6edSDave Chinner goto out_unlock_inode; 138903ba3782SJens Axboe 139003ba3782SJens Axboe /* 139103ba3782SJens Axboe * If the inode was already on b_dirty/b_io/b_more_io, don't 139203ba3782SJens Axboe * reposition it (that would break b_dirty time-ordering). 139303ba3782SJens Axboe */ 139403ba3782SJens Axboe if (!was_dirty) { 1395d6c10f1fSTejun Heo struct list_head *dirty_list; 1396a66979abSDave Chinner bool wakeup_bdi = false; 1397253c34e9SArtem Bityutskiy bdi = inode_to_bdi(inode); 1398500b067cSJens Axboe 1399146d7009SJunxiao Bi spin_unlock(&inode->i_lock); 1400146d7009SJunxiao Bi spin_lock(&bdi->wb.list_lock); 1401253c34e9SArtem Bityutskiy 1402d6c10f1fSTejun Heo WARN(bdi_cap_writeback_dirty(bdi) && 1403d6c10f1fSTejun Heo !test_bit(WB_registered, &bdi->wb.state), 1404d6c10f1fSTejun Heo "bdi-%s not registered\n", bdi->name); 140503ba3782SJens Axboe 140603ba3782SJens Axboe inode->dirtied_when = jiffies; 1407a2f48706STheodore Ts'o if (dirtytime) 1408a2f48706STheodore Ts'o inode->dirtied_time_when = jiffies; 1409d6c10f1fSTejun Heo 1410a2f48706STheodore Ts'o if (inode->i_state & (I_DIRTY_INODE | I_DIRTY_PAGES)) 1411d6c10f1fSTejun Heo dirty_list = &bdi->wb.b_dirty; 1412a2f48706STheodore Ts'o else 1413d6c10f1fSTejun Heo dirty_list = &bdi->wb.b_dirty_time; 1414d6c10f1fSTejun Heo 1415d6c10f1fSTejun Heo wakeup_bdi = inode_wb_list_move_locked(inode, &bdi->wb, 1416d6c10f1fSTejun Heo dirty_list); 1417d6c10f1fSTejun Heo 1418f758eeabSChristoph Hellwig spin_unlock(&bdi->wb.list_lock); 14190ae45f63STheodore Ts'o trace_writeback_dirty_inode_enqueue(inode); 1420253c34e9SArtem Bityutskiy 1421d6c10f1fSTejun Heo /* 1422d6c10f1fSTejun Heo * If this is the first dirty inode for this bdi, 1423d6c10f1fSTejun Heo * we have to wake-up the corresponding bdi thread 1424d6c10f1fSTejun Heo * to make sure background write-back happens 1425d6c10f1fSTejun Heo * later. 1426d6c10f1fSTejun Heo */ 1427d6c10f1fSTejun Heo if (bdi_cap_writeback_dirty(bdi) && wakeup_bdi) 1428f0054bb1STejun Heo wb_wakeup_delayed(&bdi->wb); 1429a66979abSDave Chinner return; 1430a66979abSDave Chinner } 1431a66979abSDave Chinner } 1432a66979abSDave Chinner out_unlock_inode: 1433a66979abSDave Chinner spin_unlock(&inode->i_lock); 1434a66979abSDave Chinner 143503ba3782SJens Axboe } 143603ba3782SJens Axboe EXPORT_SYMBOL(__mark_inode_dirty); 143703ba3782SJens Axboe 1438b6e51316SJens Axboe static void wait_sb_inodes(struct super_block *sb) 143966f3b8e2SJens Axboe { 144038f21977SNick Piggin struct inode *inode, *old_inode = NULL; 144138f21977SNick Piggin 144203ba3782SJens Axboe /* 144303ba3782SJens Axboe * We need to be protected against the filesystem going from 144403ba3782SJens Axboe * r/o to r/w or vice versa. 144503ba3782SJens Axboe */ 1446b6e51316SJens Axboe WARN_ON(!rwsem_is_locked(&sb->s_umount)); 144703ba3782SJens Axboe 144855fa6091SDave Chinner spin_lock(&inode_sb_list_lock); 144966f3b8e2SJens Axboe 145038f21977SNick Piggin /* 145138f21977SNick Piggin * Data integrity sync. Must wait for all pages under writeback, 145238f21977SNick Piggin * because there may have been pages dirtied before our sync 145338f21977SNick Piggin * call, but which had writeout started before we write it out. 145438f21977SNick Piggin * In which case, the inode may not be on the dirty list, but 145538f21977SNick Piggin * we still have to wait for that writeout. 145638f21977SNick Piggin */ 1457b6e51316SJens Axboe list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { 1458250df6edSDave Chinner struct address_space *mapping = inode->i_mapping; 145938f21977SNick Piggin 1460250df6edSDave Chinner spin_lock(&inode->i_lock); 1461250df6edSDave Chinner if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) || 1462250df6edSDave Chinner (mapping->nrpages == 0)) { 1463250df6edSDave Chinner spin_unlock(&inode->i_lock); 146438f21977SNick Piggin continue; 1465250df6edSDave Chinner } 146638f21977SNick Piggin __iget(inode); 1467250df6edSDave Chinner spin_unlock(&inode->i_lock); 146855fa6091SDave Chinner spin_unlock(&inode_sb_list_lock); 146955fa6091SDave Chinner 147038f21977SNick Piggin /* 147155fa6091SDave Chinner * We hold a reference to 'inode' so it couldn't have been 147255fa6091SDave Chinner * removed from s_inodes list while we dropped the 147355fa6091SDave Chinner * inode_sb_list_lock. We cannot iput the inode now as we can 147455fa6091SDave Chinner * be holding the last reference and we cannot iput it under 147555fa6091SDave Chinner * inode_sb_list_lock. So we keep the reference and iput it 147655fa6091SDave Chinner * later. 147738f21977SNick Piggin */ 147838f21977SNick Piggin iput(old_inode); 147938f21977SNick Piggin old_inode = inode; 148038f21977SNick Piggin 148138f21977SNick Piggin filemap_fdatawait(mapping); 148238f21977SNick Piggin 148338f21977SNick Piggin cond_resched(); 148438f21977SNick Piggin 148555fa6091SDave Chinner spin_lock(&inode_sb_list_lock); 148638f21977SNick Piggin } 148755fa6091SDave Chinner spin_unlock(&inode_sb_list_lock); 148838f21977SNick Piggin iput(old_inode); 148966f3b8e2SJens Axboe } 14901da177e4SLinus Torvalds 1491d8a8559cSJens Axboe /** 14923259f8beSChris Mason * writeback_inodes_sb_nr - writeback dirty inodes from given super_block 1493d8a8559cSJens Axboe * @sb: the superblock 14943259f8beSChris Mason * @nr: the number of pages to write 1495786228abSMarcos Paulo de Souza * @reason: reason why some writeback work initiated 14961da177e4SLinus Torvalds * 1497d8a8559cSJens Axboe * Start writeback on some inodes on this super_block. No guarantees are made 1498d8a8559cSJens Axboe * on how many (if any) will be written, and this function does not wait 14993259f8beSChris Mason * for IO completion of submitted IO. 15001da177e4SLinus Torvalds */ 15010e175a18SCurt Wohlgemuth void writeback_inodes_sb_nr(struct super_block *sb, 15020e175a18SCurt Wohlgemuth unsigned long nr, 15030e175a18SCurt Wohlgemuth enum wb_reason reason) 15041da177e4SLinus Torvalds { 150583ba7b07SChristoph Hellwig DECLARE_COMPLETION_ONSTACK(done); 150683ba7b07SChristoph Hellwig struct wb_writeback_work work = { 15073c4d7165SChristoph Hellwig .sb = sb, 15083c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_NONE, 15096e6938b6SWu Fengguang .tagged_writepages = 1, 151083ba7b07SChristoph Hellwig .done = &done, 15113259f8beSChris Mason .nr_pages = nr, 15120e175a18SCurt Wohlgemuth .reason = reason, 15133c4d7165SChristoph Hellwig }; 15140e3c9a22SJens Axboe 15156eedc701SJan Kara if (sb->s_bdi == &noop_backing_dev_info) 15166eedc701SJan Kara return; 1517cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 1518f0054bb1STejun Heo wb_queue_work(&sb->s_bdi->wb, &work); 151983ba7b07SChristoph Hellwig wait_for_completion(&done); 15201da177e4SLinus Torvalds } 15213259f8beSChris Mason EXPORT_SYMBOL(writeback_inodes_sb_nr); 15223259f8beSChris Mason 15233259f8beSChris Mason /** 15243259f8beSChris Mason * writeback_inodes_sb - writeback dirty inodes from given super_block 15253259f8beSChris Mason * @sb: the superblock 1526786228abSMarcos Paulo de Souza * @reason: reason why some writeback work was initiated 15273259f8beSChris Mason * 15283259f8beSChris Mason * Start writeback on some inodes on this super_block. No guarantees are made 15293259f8beSChris Mason * on how many (if any) will be written, and this function does not wait 15303259f8beSChris Mason * for IO completion of submitted IO. 15313259f8beSChris Mason */ 15320e175a18SCurt Wohlgemuth void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) 15333259f8beSChris Mason { 15340e175a18SCurt Wohlgemuth return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); 15353259f8beSChris Mason } 1536d8a8559cSJens Axboe EXPORT_SYMBOL(writeback_inodes_sb); 1537d8a8559cSJens Axboe 1538d8a8559cSJens Axboe /** 153910ee27a0SMiao Xie * try_to_writeback_inodes_sb_nr - try to start writeback if none underway 15403259f8beSChris Mason * @sb: the superblock 15413259f8beSChris Mason * @nr: the number of pages to write 154210ee27a0SMiao Xie * @reason: the reason of writeback 15433259f8beSChris Mason * 154410ee27a0SMiao Xie * Invoke writeback_inodes_sb_nr if no writeback is currently underway. 15453259f8beSChris Mason * Returns 1 if writeback was started, 0 if not. 15463259f8beSChris Mason */ 154710ee27a0SMiao Xie int try_to_writeback_inodes_sb_nr(struct super_block *sb, 15480e175a18SCurt Wohlgemuth unsigned long nr, 15490e175a18SCurt Wohlgemuth enum wb_reason reason) 15503259f8beSChris Mason { 155110ee27a0SMiao Xie if (writeback_in_progress(sb->s_bdi)) 155210ee27a0SMiao Xie return 1; 155310ee27a0SMiao Xie 155410ee27a0SMiao Xie if (!down_read_trylock(&sb->s_umount)) 155510ee27a0SMiao Xie return 0; 155610ee27a0SMiao Xie 15570e175a18SCurt Wohlgemuth writeback_inodes_sb_nr(sb, nr, reason); 15583259f8beSChris Mason up_read(&sb->s_umount); 15593259f8beSChris Mason return 1; 15603259f8beSChris Mason } 156110ee27a0SMiao Xie EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr); 156210ee27a0SMiao Xie 156310ee27a0SMiao Xie /** 156410ee27a0SMiao Xie * try_to_writeback_inodes_sb - try to start writeback if none underway 156510ee27a0SMiao Xie * @sb: the superblock 156610ee27a0SMiao Xie * @reason: reason why some writeback work was initiated 156710ee27a0SMiao Xie * 156810ee27a0SMiao Xie * Implement by try_to_writeback_inodes_sb_nr() 156910ee27a0SMiao Xie * Returns 1 if writeback was started, 0 if not. 157010ee27a0SMiao Xie */ 157110ee27a0SMiao Xie int try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) 157210ee27a0SMiao Xie { 157310ee27a0SMiao Xie return try_to_writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); 157410ee27a0SMiao Xie } 157510ee27a0SMiao Xie EXPORT_SYMBOL(try_to_writeback_inodes_sb); 15763259f8beSChris Mason 15773259f8beSChris Mason /** 1578d8a8559cSJens Axboe * sync_inodes_sb - sync sb inode pages 1579d8a8559cSJens Axboe * @sb: the superblock 1580d8a8559cSJens Axboe * 1581d8a8559cSJens Axboe * This function writes and waits on any dirty inode belonging to this 15820dc83bd3SJan Kara * super_block. 1583d8a8559cSJens Axboe */ 15840dc83bd3SJan Kara void sync_inodes_sb(struct super_block *sb) 1585d8a8559cSJens Axboe { 158683ba7b07SChristoph Hellwig DECLARE_COMPLETION_ONSTACK(done); 158783ba7b07SChristoph Hellwig struct wb_writeback_work work = { 15883c4d7165SChristoph Hellwig .sb = sb, 15893c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_ALL, 15903c4d7165SChristoph Hellwig .nr_pages = LONG_MAX, 15913c4d7165SChristoph Hellwig .range_cyclic = 0, 159283ba7b07SChristoph Hellwig .done = &done, 15930e175a18SCurt Wohlgemuth .reason = WB_REASON_SYNC, 15947747bd4bSDave Chinner .for_sync = 1, 15953c4d7165SChristoph Hellwig }; 15963c4d7165SChristoph Hellwig 15976eedc701SJan Kara /* Nothing to do? */ 15986eedc701SJan Kara if (sb->s_bdi == &noop_backing_dev_info) 15996eedc701SJan Kara return; 1600cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 1601cf37e972SChristoph Hellwig 1602f0054bb1STejun Heo wb_queue_work(&sb->s_bdi->wb, &work); 160383ba7b07SChristoph Hellwig wait_for_completion(&done); 160483ba7b07SChristoph Hellwig 1605b6e51316SJens Axboe wait_sb_inodes(sb); 1606d8a8559cSJens Axboe } 1607d8a8559cSJens Axboe EXPORT_SYMBOL(sync_inodes_sb); 16081da177e4SLinus Torvalds 16091da177e4SLinus Torvalds /** 16101da177e4SLinus Torvalds * write_inode_now - write an inode to disk 16111da177e4SLinus Torvalds * @inode: inode to write to disk 16121da177e4SLinus Torvalds * @sync: whether the write should be synchronous or not 16131da177e4SLinus Torvalds * 16147f04c26dSAndrea Arcangeli * This function commits an inode to disk immediately if it is dirty. This is 16157f04c26dSAndrea Arcangeli * primarily needed by knfsd. 16167f04c26dSAndrea Arcangeli * 16177f04c26dSAndrea Arcangeli * The caller must either have a ref on the inode or must have set I_WILL_FREE. 16181da177e4SLinus Torvalds */ 16191da177e4SLinus Torvalds int write_inode_now(struct inode *inode, int sync) 16201da177e4SLinus Torvalds { 1621f758eeabSChristoph Hellwig struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; 16221da177e4SLinus Torvalds struct writeback_control wbc = { 16231da177e4SLinus Torvalds .nr_to_write = LONG_MAX, 162418914b18SMike Galbraith .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, 1625111ebb6eSOGAWA Hirofumi .range_start = 0, 1626111ebb6eSOGAWA Hirofumi .range_end = LLONG_MAX, 16271da177e4SLinus Torvalds }; 16281da177e4SLinus Torvalds 16291da177e4SLinus Torvalds if (!mapping_cap_writeback_dirty(inode->i_mapping)) 163049364ce2SAndrew Morton wbc.nr_to_write = 0; 16311da177e4SLinus Torvalds 16321da177e4SLinus Torvalds might_sleep(); 16334f8ad655SJan Kara return writeback_single_inode(inode, wb, &wbc); 16341da177e4SLinus Torvalds } 16351da177e4SLinus Torvalds EXPORT_SYMBOL(write_inode_now); 16361da177e4SLinus Torvalds 16371da177e4SLinus Torvalds /** 16381da177e4SLinus Torvalds * sync_inode - write an inode and its pages to disk. 16391da177e4SLinus Torvalds * @inode: the inode to sync 16401da177e4SLinus Torvalds * @wbc: controls the writeback mode 16411da177e4SLinus Torvalds * 16421da177e4SLinus Torvalds * sync_inode() will write an inode and its pages to disk. It will also 16431da177e4SLinus Torvalds * correctly update the inode on its superblock's dirty inode lists and will 16441da177e4SLinus Torvalds * update inode->i_state. 16451da177e4SLinus Torvalds * 16461da177e4SLinus Torvalds * The caller must have a ref on the inode. 16471da177e4SLinus Torvalds */ 16481da177e4SLinus Torvalds int sync_inode(struct inode *inode, struct writeback_control *wbc) 16491da177e4SLinus Torvalds { 16504f8ad655SJan Kara return writeback_single_inode(inode, &inode_to_bdi(inode)->wb, wbc); 16511da177e4SLinus Torvalds } 16521da177e4SLinus Torvalds EXPORT_SYMBOL(sync_inode); 1653c3765016SChristoph Hellwig 1654c3765016SChristoph Hellwig /** 1655c691b9d9SAndrew Morton * sync_inode_metadata - write an inode to disk 1656c3765016SChristoph Hellwig * @inode: the inode to sync 1657c3765016SChristoph Hellwig * @wait: wait for I/O to complete. 1658c3765016SChristoph Hellwig * 1659c691b9d9SAndrew Morton * Write an inode to disk and adjust its dirty state after completion. 1660c3765016SChristoph Hellwig * 1661c3765016SChristoph Hellwig * Note: only writes the actual inode, no associated data or other metadata. 1662c3765016SChristoph Hellwig */ 1663c3765016SChristoph Hellwig int sync_inode_metadata(struct inode *inode, int wait) 1664c3765016SChristoph Hellwig { 1665c3765016SChristoph Hellwig struct writeback_control wbc = { 1666c3765016SChristoph Hellwig .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE, 1667c3765016SChristoph Hellwig .nr_to_write = 0, /* metadata-only */ 1668c3765016SChristoph Hellwig }; 1669c3765016SChristoph Hellwig 1670c3765016SChristoph Hellwig return sync_inode(inode, &wbc); 1671c3765016SChristoph Hellwig } 1672c3765016SChristoph Hellwig EXPORT_SYMBOL(sync_inode_metadata); 1673