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> 17f5ff8422SJens Axboe #include <linux/module.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> 2303ba3782SJens Axboe #include <linux/kthread.h> 2403ba3782SJens Axboe #include <linux/freezer.h> 251da177e4SLinus Torvalds #include <linux/writeback.h> 261da177e4SLinus Torvalds #include <linux/blkdev.h> 271da177e4SLinus Torvalds #include <linux/backing-dev.h> 281da177e4SLinus Torvalds #include <linux/buffer_head.h> 2907f3f05cSDavid Howells #include "internal.h" 301da177e4SLinus Torvalds 3166f3b8e2SJens Axboe #define inode_to_bdi(inode) ((inode)->i_mapping->backing_dev_info) 32f11b00f3SAdrian Bunk 3303ba3782SJens Axboe /* 34d0bceac7SJens Axboe * We don't actually have pdflush, but this one is exported though /proc... 35d0bceac7SJens Axboe */ 36d0bceac7SJens Axboe int nr_pdflush_threads; 37d0bceac7SJens Axboe 38d0bceac7SJens Axboe /* 39c4a77a6cSJens Axboe * Passed into wb_writeback(), essentially a subset of writeback_control 40c4a77a6cSJens Axboe */ 41c4a77a6cSJens Axboe struct wb_writeback_args { 42c4a77a6cSJens Axboe long nr_pages; 43c4a77a6cSJens Axboe struct super_block *sb; 44c4a77a6cSJens Axboe enum writeback_sync_modes sync_mode; 4552957fe1SH Hartley Sweeten unsigned int for_kupdate:1; 4652957fe1SH Hartley Sweeten unsigned int range_cyclic:1; 4752957fe1SH Hartley Sweeten unsigned int for_background:1; 48c4a77a6cSJens Axboe }; 49c4a77a6cSJens Axboe 50c4a77a6cSJens Axboe /* 5103ba3782SJens Axboe * Work items for the bdi_writeback threads 52f11b00f3SAdrian Bunk */ 5303ba3782SJens Axboe struct bdi_work { 548010c3b6SJens Axboe struct list_head list; /* pending work list */ 558010c3b6SJens Axboe struct rcu_head rcu_head; /* for RCU free/clear of work */ 5603ba3782SJens Axboe 578010c3b6SJens Axboe unsigned long seen; /* threads that have seen this work */ 588010c3b6SJens Axboe atomic_t pending; /* number of threads still to do work */ 5903ba3782SJens Axboe 608010c3b6SJens Axboe struct wb_writeback_args args; /* writeback arguments */ 6103ba3782SJens Axboe 628010c3b6SJens Axboe unsigned long state; /* flag bits, see WS_* */ 6303ba3782SJens Axboe }; 6403ba3782SJens Axboe 6503ba3782SJens Axboe enum { 6603ba3782SJens Axboe WS_USED_B = 0, 6703ba3782SJens Axboe WS_ONSTACK_B, 6803ba3782SJens Axboe }; 6903ba3782SJens Axboe 7003ba3782SJens Axboe #define WS_USED (1 << WS_USED_B) 7103ba3782SJens Axboe #define WS_ONSTACK (1 << WS_ONSTACK_B) 7203ba3782SJens Axboe 7303ba3782SJens Axboe static inline bool bdi_work_on_stack(struct bdi_work *work) 74f11b00f3SAdrian Bunk { 7503ba3782SJens Axboe return test_bit(WS_ONSTACK_B, &work->state); 7603ba3782SJens Axboe } 7703ba3782SJens Axboe 7803ba3782SJens Axboe static inline void bdi_work_init(struct bdi_work *work, 79b6e51316SJens Axboe struct wb_writeback_args *args) 8003ba3782SJens Axboe { 8103ba3782SJens Axboe INIT_RCU_HEAD(&work->rcu_head); 82b6e51316SJens Axboe work->args = *args; 8303ba3782SJens Axboe work->state = WS_USED; 8403ba3782SJens Axboe } 8503ba3782SJens Axboe 86f11b00f3SAdrian Bunk /** 87f11b00f3SAdrian Bunk * writeback_in_progress - determine whether there is writeback in progress 88f11b00f3SAdrian Bunk * @bdi: the device's backing_dev_info structure. 89f11b00f3SAdrian Bunk * 9003ba3782SJens Axboe * Determine whether there is writeback waiting to be handled against a 9103ba3782SJens Axboe * backing device. 92f11b00f3SAdrian Bunk */ 93f11b00f3SAdrian Bunk int writeback_in_progress(struct backing_dev_info *bdi) 94f11b00f3SAdrian Bunk { 9503ba3782SJens Axboe return !list_empty(&bdi->work_list); 96f11b00f3SAdrian Bunk } 97f11b00f3SAdrian Bunk 9803ba3782SJens Axboe static void bdi_work_clear(struct bdi_work *work) 99f11b00f3SAdrian Bunk { 10003ba3782SJens Axboe clear_bit(WS_USED_B, &work->state); 10103ba3782SJens Axboe smp_mb__after_clear_bit(); 1021ef7d9aaSNick Piggin /* 1031ef7d9aaSNick Piggin * work can have disappeared at this point. bit waitq functions 1041ef7d9aaSNick Piggin * should be able to tolerate this, provided bdi_sched_wait does 1051ef7d9aaSNick Piggin * not dereference it's pointer argument. 1061ef7d9aaSNick Piggin */ 10703ba3782SJens Axboe wake_up_bit(&work->state, WS_USED_B); 108f11b00f3SAdrian Bunk } 109f11b00f3SAdrian Bunk 11003ba3782SJens Axboe static void bdi_work_free(struct rcu_head *head) 1114195f73dSNick Piggin { 11203ba3782SJens Axboe struct bdi_work *work = container_of(head, struct bdi_work, rcu_head); 1134195f73dSNick Piggin 11403ba3782SJens Axboe if (!bdi_work_on_stack(work)) 11503ba3782SJens Axboe kfree(work); 11603ba3782SJens Axboe else 11703ba3782SJens Axboe bdi_work_clear(work); 1184195f73dSNick Piggin } 1194195f73dSNick Piggin 12003ba3782SJens Axboe static void wb_work_complete(struct bdi_work *work) 1211da177e4SLinus Torvalds { 122c4a77a6cSJens Axboe const enum writeback_sync_modes sync_mode = work->args.sync_mode; 12377b9d059SNick Piggin int onstack = bdi_work_on_stack(work); 1241da177e4SLinus Torvalds 1251da177e4SLinus Torvalds /* 12603ba3782SJens Axboe * For allocated work, we can clear the done/seen bit right here. 12703ba3782SJens Axboe * For on-stack work, we need to postpone both the clear and free 12803ba3782SJens Axboe * to after the RCU grace period, since the stack could be invalidated 12903ba3782SJens Axboe * as soon as bdi_work_clear() has done the wakeup. 1301da177e4SLinus Torvalds */ 13177b9d059SNick Piggin if (!onstack) 13203ba3782SJens Axboe bdi_work_clear(work); 13377b9d059SNick Piggin if (sync_mode == WB_SYNC_NONE || onstack) 13403ba3782SJens Axboe call_rcu(&work->rcu_head, bdi_work_free); 1351da177e4SLinus Torvalds } 1361da177e4SLinus Torvalds 13703ba3782SJens Axboe static void wb_clear_pending(struct bdi_writeback *wb, struct bdi_work *work) 13803ba3782SJens Axboe { 1391da177e4SLinus Torvalds /* 14003ba3782SJens Axboe * The caller has retrieved the work arguments from this work, 14103ba3782SJens Axboe * drop our reference. If this is the last ref, delete and free it 14203ba3782SJens Axboe */ 14303ba3782SJens Axboe if (atomic_dec_and_test(&work->pending)) { 14403ba3782SJens Axboe struct backing_dev_info *bdi = wb->bdi; 14503ba3782SJens Axboe 14603ba3782SJens Axboe spin_lock(&bdi->wb_lock); 14703ba3782SJens Axboe list_del_rcu(&work->list); 14803ba3782SJens Axboe spin_unlock(&bdi->wb_lock); 14903ba3782SJens Axboe 15003ba3782SJens Axboe wb_work_complete(work); 15103ba3782SJens Axboe } 15203ba3782SJens Axboe } 15303ba3782SJens Axboe 15403ba3782SJens Axboe static void bdi_queue_work(struct backing_dev_info *bdi, struct bdi_work *work) 15503ba3782SJens Axboe { 15603ba3782SJens Axboe work->seen = bdi->wb_mask; 15703ba3782SJens Axboe BUG_ON(!work->seen); 15803ba3782SJens Axboe atomic_set(&work->pending, bdi->wb_cnt); 15903ba3782SJens Axboe BUG_ON(!bdi->wb_cnt); 16003ba3782SJens Axboe 16103ba3782SJens Axboe /* 162deed62edSNick Piggin * list_add_tail_rcu() contains the necessary barriers to 163deed62edSNick Piggin * make sure the above stores are seen before the item is 164deed62edSNick Piggin * noticed on the list 1651da177e4SLinus Torvalds */ 16603ba3782SJens Axboe spin_lock(&bdi->wb_lock); 16703ba3782SJens Axboe list_add_tail_rcu(&work->list, &bdi->work_list); 16803ba3782SJens Axboe spin_unlock(&bdi->wb_lock); 1691da177e4SLinus Torvalds 1701da177e4SLinus Torvalds /* 17103ba3782SJens Axboe * If the default thread isn't there, make sure we add it. When 17203ba3782SJens Axboe * it gets created and wakes up, we'll run this work. 1731da177e4SLinus Torvalds */ 17403ba3782SJens Axboe if (unlikely(list_empty_careful(&bdi->wb_list))) 17503ba3782SJens Axboe wake_up_process(default_backing_dev_info.wb.task); 17603ba3782SJens Axboe else { 17703ba3782SJens Axboe struct bdi_writeback *wb = &bdi->wb; 1781da177e4SLinus Torvalds 1791ef7d9aaSNick Piggin if (wb->task) 18003ba3782SJens Axboe wake_up_process(wb->task); 1811da177e4SLinus Torvalds } 18203ba3782SJens Axboe } 1831da177e4SLinus Torvalds 1841da177e4SLinus Torvalds /* 18503ba3782SJens Axboe * Used for on-stack allocated work items. The caller needs to wait until 18603ba3782SJens Axboe * the wb threads have acked the work before it's safe to continue. 1871da177e4SLinus Torvalds */ 18803ba3782SJens Axboe static void bdi_wait_on_work_clear(struct bdi_work *work) 1891da177e4SLinus Torvalds { 19003ba3782SJens Axboe wait_on_bit(&work->state, WS_USED_B, bdi_sched_wait, 19103ba3782SJens Axboe TASK_UNINTERRUPTIBLE); 19203ba3782SJens Axboe } 19303ba3782SJens Axboe 194f11fcae8SJens Axboe static void bdi_alloc_queue_work(struct backing_dev_info *bdi, 195f17625b3SJens Axboe struct wb_writeback_args *args) 19603ba3782SJens Axboe { 19703ba3782SJens Axboe struct bdi_work *work; 19803ba3782SJens Axboe 199bcddc3f0SJens Axboe /* 200bcddc3f0SJens Axboe * This is WB_SYNC_NONE writeback, so if allocation fails just 201bcddc3f0SJens Axboe * wakeup the thread for old dirty data writeback 202bcddc3f0SJens Axboe */ 20303ba3782SJens Axboe work = kmalloc(sizeof(*work), GFP_ATOMIC); 204bcddc3f0SJens Axboe if (work) { 205b6e51316SJens Axboe bdi_work_init(work, args); 206f11fcae8SJens Axboe bdi_queue_work(bdi, work); 207bcddc3f0SJens Axboe } else { 208bcddc3f0SJens Axboe struct bdi_writeback *wb = &bdi->wb; 209bcddc3f0SJens Axboe 210bcddc3f0SJens Axboe if (wb->task) 211bcddc3f0SJens Axboe wake_up_process(wb->task); 212bcddc3f0SJens Axboe } 21303ba3782SJens Axboe } 21403ba3782SJens Axboe 215b6e51316SJens Axboe /** 216b6e51316SJens Axboe * bdi_sync_writeback - start and wait for writeback 217b6e51316SJens Axboe * @bdi: the backing device to write from 218b6e51316SJens Axboe * @sb: write inodes from this super_block 219b6e51316SJens Axboe * 220b6e51316SJens Axboe * Description: 221b6e51316SJens Axboe * This does WB_SYNC_ALL data integrity writeback and waits for the 222b6e51316SJens Axboe * IO to complete. Callers must hold the sb s_umount semaphore for 223b6e51316SJens Axboe * reading, to avoid having the super disappear before we are done. 22403ba3782SJens Axboe */ 225b6e51316SJens Axboe static void bdi_sync_writeback(struct backing_dev_info *bdi, 226b6e51316SJens Axboe struct super_block *sb) 227b6e51316SJens Axboe { 228b6e51316SJens Axboe struct wb_writeback_args args = { 229b6e51316SJens Axboe .sb = sb, 230b6e51316SJens Axboe .sync_mode = WB_SYNC_ALL, 231b6e51316SJens Axboe .nr_pages = LONG_MAX, 232b6e51316SJens Axboe .range_cyclic = 0, 233b6e51316SJens Axboe }; 234f0fad8a5SChristoph Hellwig struct bdi_work work; 235f0fad8a5SChristoph Hellwig 236b6e51316SJens Axboe bdi_work_init(&work, &args); 237f0fad8a5SChristoph Hellwig work.state |= WS_ONSTACK; 238f0fad8a5SChristoph Hellwig 239b6e51316SJens Axboe bdi_queue_work(bdi, &work); 240f0fad8a5SChristoph Hellwig bdi_wait_on_work_clear(&work); 24103ba3782SJens Axboe } 242b6e51316SJens Axboe 243b6e51316SJens Axboe /** 244b6e51316SJens Axboe * bdi_start_writeback - start writeback 245b6e51316SJens Axboe * @bdi: the backing device to write from 2464b6764faSJaswinder Singh Rajput * @sb: write inodes from this super_block 247b6e51316SJens Axboe * @nr_pages: the number of pages to write 248b6e51316SJens Axboe * 249b6e51316SJens Axboe * Description: 250b6e51316SJens Axboe * This does WB_SYNC_NONE opportunistic writeback. The IO is only 251b6e51316SJens Axboe * started when this function returns, we make no guarentees on 2520e3c9a22SJens Axboe * completion. Caller need not hold sb s_umount semaphore. 253b6e51316SJens Axboe * 254b6e51316SJens Axboe */ 255a72bfd4dSJens Axboe void bdi_start_writeback(struct backing_dev_info *bdi, struct super_block *sb, 2560e3c9a22SJens Axboe long nr_pages) 257b6e51316SJens Axboe { 258b6e51316SJens Axboe struct wb_writeback_args args = { 259a72bfd4dSJens Axboe .sb = sb, 260b6e51316SJens Axboe .sync_mode = WB_SYNC_NONE, 261b6e51316SJens Axboe .nr_pages = nr_pages, 262b6e51316SJens Axboe .range_cyclic = 1, 263b6e51316SJens Axboe }; 264b6e51316SJens Axboe 265d3ddec76SWu Fengguang /* 266d3ddec76SWu Fengguang * We treat @nr_pages=0 as the special case to do background writeback, 267d3ddec76SWu Fengguang * ie. to sync pages until the background dirty threshold is reached. 268d3ddec76SWu Fengguang */ 269d3ddec76SWu Fengguang if (!nr_pages) { 270d3ddec76SWu Fengguang args.nr_pages = LONG_MAX; 271d3ddec76SWu Fengguang args.for_background = 1; 272d3ddec76SWu Fengguang } 273d3ddec76SWu Fengguang 274f17625b3SJens Axboe bdi_alloc_queue_work(bdi, &args); 2751da177e4SLinus Torvalds } 2761da177e4SLinus Torvalds 2771da177e4SLinus Torvalds /* 2786610a0bcSAndrew Morton * Redirty an inode: set its when-it-was dirtied timestamp and move it to the 2796610a0bcSAndrew Morton * furthest end of its superblock's dirty-inode list. 2806610a0bcSAndrew Morton * 2816610a0bcSAndrew Morton * Before stamping the inode's ->dirtied_when, we check to see whether it is 28266f3b8e2SJens Axboe * already the most-recently-dirtied inode on the b_dirty list. If that is 2836610a0bcSAndrew Morton * the case then the inode must have been redirtied while it was being written 2846610a0bcSAndrew Morton * out and we don't reset its dirtied_when. 2856610a0bcSAndrew Morton */ 2866610a0bcSAndrew Morton static void redirty_tail(struct inode *inode) 2876610a0bcSAndrew Morton { 28803ba3782SJens Axboe struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; 2896610a0bcSAndrew Morton 29003ba3782SJens Axboe if (!list_empty(&wb->b_dirty)) { 29166f3b8e2SJens Axboe struct inode *tail; 2926610a0bcSAndrew Morton 29303ba3782SJens Axboe tail = list_entry(wb->b_dirty.next, struct inode, i_list); 29466f3b8e2SJens Axboe if (time_before(inode->dirtied_when, tail->dirtied_when)) 2956610a0bcSAndrew Morton inode->dirtied_when = jiffies; 2966610a0bcSAndrew Morton } 29703ba3782SJens Axboe list_move(&inode->i_list, &wb->b_dirty); 2986610a0bcSAndrew Morton } 2996610a0bcSAndrew Morton 3006610a0bcSAndrew Morton /* 30166f3b8e2SJens Axboe * requeue inode for re-scanning after bdi->b_io list is exhausted. 302c986d1e2SAndrew Morton */ 3030e0f4fc2SKen Chen static void requeue_io(struct inode *inode) 304c986d1e2SAndrew Morton { 30503ba3782SJens Axboe struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; 30603ba3782SJens Axboe 30703ba3782SJens Axboe list_move(&inode->i_list, &wb->b_more_io); 308c986d1e2SAndrew Morton } 309c986d1e2SAndrew Morton 3101c0eeaf5SJoern Engel static void inode_sync_complete(struct inode *inode) 3111c0eeaf5SJoern Engel { 3121c0eeaf5SJoern Engel /* 3131c0eeaf5SJoern Engel * Prevent speculative execution through spin_unlock(&inode_lock); 3141c0eeaf5SJoern Engel */ 3151c0eeaf5SJoern Engel smp_mb(); 3161c0eeaf5SJoern Engel wake_up_bit(&inode->i_state, __I_SYNC); 3171c0eeaf5SJoern Engel } 3181c0eeaf5SJoern Engel 319d2caa3c5SJeff Layton static bool inode_dirtied_after(struct inode *inode, unsigned long t) 320d2caa3c5SJeff Layton { 321d2caa3c5SJeff Layton bool ret = time_after(inode->dirtied_when, t); 322d2caa3c5SJeff Layton #ifndef CONFIG_64BIT 323d2caa3c5SJeff Layton /* 324d2caa3c5SJeff Layton * For inodes being constantly redirtied, dirtied_when can get stuck. 325d2caa3c5SJeff Layton * It _appears_ to be in the future, but is actually in distant past. 326d2caa3c5SJeff Layton * This test is necessary to prevent such wrapped-around relative times 3275b0830cbSJens Axboe * from permanently stopping the whole bdi writeback. 328d2caa3c5SJeff Layton */ 329d2caa3c5SJeff Layton ret = ret && time_before_eq(inode->dirtied_when, jiffies); 330d2caa3c5SJeff Layton #endif 331d2caa3c5SJeff Layton return ret; 332d2caa3c5SJeff Layton } 333d2caa3c5SJeff Layton 334c986d1e2SAndrew Morton /* 3352c136579SFengguang Wu * Move expired dirty inodes from @delaying_queue to @dispatch_queue. 3362c136579SFengguang Wu */ 3372c136579SFengguang Wu static void move_expired_inodes(struct list_head *delaying_queue, 3382c136579SFengguang Wu struct list_head *dispatch_queue, 3392c136579SFengguang Wu unsigned long *older_than_this) 3402c136579SFengguang Wu { 3415c03449dSShaohua Li LIST_HEAD(tmp); 3425c03449dSShaohua Li struct list_head *pos, *node; 343cf137307SJens Axboe struct super_block *sb = NULL; 3445c03449dSShaohua Li struct inode *inode; 345cf137307SJens Axboe int do_sb_sort = 0; 3465c03449dSShaohua Li 3472c136579SFengguang Wu while (!list_empty(delaying_queue)) { 3485c03449dSShaohua Li inode = list_entry(delaying_queue->prev, struct inode, i_list); 3492c136579SFengguang Wu if (older_than_this && 350d2caa3c5SJeff Layton inode_dirtied_after(inode, *older_than_this)) 3512c136579SFengguang Wu break; 352cf137307SJens Axboe if (sb && sb != inode->i_sb) 353cf137307SJens Axboe do_sb_sort = 1; 354cf137307SJens Axboe sb = inode->i_sb; 3555c03449dSShaohua Li list_move(&inode->i_list, &tmp); 3565c03449dSShaohua Li } 3575c03449dSShaohua Li 358cf137307SJens Axboe /* just one sb in list, splice to dispatch_queue and we're done */ 359cf137307SJens Axboe if (!do_sb_sort) { 360cf137307SJens Axboe list_splice(&tmp, dispatch_queue); 361cf137307SJens Axboe return; 362cf137307SJens Axboe } 363cf137307SJens Axboe 3645c03449dSShaohua Li /* Move inodes from one superblock together */ 3655c03449dSShaohua Li while (!list_empty(&tmp)) { 3665c03449dSShaohua Li inode = list_entry(tmp.prev, struct inode, i_list); 3675c03449dSShaohua Li sb = inode->i_sb; 3685c03449dSShaohua Li list_for_each_prev_safe(pos, node, &tmp) { 3695c03449dSShaohua Li inode = list_entry(pos, struct inode, i_list); 3705c03449dSShaohua Li if (inode->i_sb == sb) 3712c136579SFengguang Wu list_move(&inode->i_list, dispatch_queue); 3722c136579SFengguang Wu } 3732c136579SFengguang Wu } 3745c03449dSShaohua Li } 3752c136579SFengguang Wu 3762c136579SFengguang Wu /* 3772c136579SFengguang Wu * Queue all expired dirty inodes for io, eldest first. 3782c136579SFengguang Wu */ 37903ba3782SJens Axboe static void queue_io(struct bdi_writeback *wb, unsigned long *older_than_this) 3802c136579SFengguang Wu { 38103ba3782SJens Axboe list_splice_init(&wb->b_more_io, wb->b_io.prev); 38203ba3782SJens Axboe move_expired_inodes(&wb->b_dirty, &wb->b_io, older_than_this); 38366f3b8e2SJens Axboe } 38466f3b8e2SJens Axboe 385a9185b41SChristoph Hellwig static int write_inode(struct inode *inode, struct writeback_control *wbc) 38666f3b8e2SJens Axboe { 38703ba3782SJens Axboe if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) 388a9185b41SChristoph Hellwig return inode->i_sb->s_op->write_inode(inode, wbc); 38903ba3782SJens Axboe return 0; 39066f3b8e2SJens Axboe } 39108d8e974SFengguang Wu 3922c136579SFengguang Wu /* 39301c03194SChristoph Hellwig * Wait for writeback on an inode to complete. 39401c03194SChristoph Hellwig */ 39501c03194SChristoph Hellwig static void inode_wait_for_writeback(struct inode *inode) 39601c03194SChristoph Hellwig { 39701c03194SChristoph Hellwig DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); 39801c03194SChristoph Hellwig wait_queue_head_t *wqh; 39901c03194SChristoph Hellwig 40001c03194SChristoph Hellwig wqh = bit_waitqueue(&inode->i_state, __I_SYNC); 40101c03194SChristoph Hellwig do { 40201c03194SChristoph Hellwig spin_unlock(&inode_lock); 40301c03194SChristoph Hellwig __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE); 40401c03194SChristoph Hellwig spin_lock(&inode_lock); 40501c03194SChristoph Hellwig } while (inode->i_state & I_SYNC); 40601c03194SChristoph Hellwig } 40701c03194SChristoph Hellwig 40801c03194SChristoph Hellwig /* 40901c03194SChristoph Hellwig * Write out an inode's dirty pages. Called under inode_lock. Either the 41001c03194SChristoph Hellwig * caller has ref on the inode (either via __iget or via syscall against an fd) 41101c03194SChristoph Hellwig * or the inode has I_WILL_FREE set (via generic_forget_inode) 41201c03194SChristoph Hellwig * 4131da177e4SLinus Torvalds * If `wait' is set, wait on the writeout. 4141da177e4SLinus Torvalds * 4151da177e4SLinus Torvalds * The whole writeout design is quite complex and fragile. We want to avoid 4161da177e4SLinus Torvalds * starvation of particular inodes when others are being redirtied, prevent 4171da177e4SLinus Torvalds * livelocks, etc. 4181da177e4SLinus Torvalds * 4191da177e4SLinus Torvalds * Called under inode_lock. 4201da177e4SLinus Torvalds */ 4211da177e4SLinus Torvalds static int 42201c03194SChristoph Hellwig writeback_single_inode(struct inode *inode, struct writeback_control *wbc) 4231da177e4SLinus Torvalds { 4241da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 42501c03194SChristoph Hellwig unsigned dirty; 4261da177e4SLinus Torvalds int ret; 4271da177e4SLinus Torvalds 42801c03194SChristoph Hellwig if (!atomic_read(&inode->i_count)) 42901c03194SChristoph Hellwig WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); 43001c03194SChristoph Hellwig else 43101c03194SChristoph Hellwig WARN_ON(inode->i_state & I_WILL_FREE); 43201c03194SChristoph Hellwig 43301c03194SChristoph Hellwig if (inode->i_state & I_SYNC) { 43401c03194SChristoph Hellwig /* 43501c03194SChristoph Hellwig * If this inode is locked for writeback and we are not doing 43666f3b8e2SJens Axboe * writeback-for-data-integrity, move it to b_more_io so that 43701c03194SChristoph Hellwig * writeback can proceed with the other inodes on s_io. 43801c03194SChristoph Hellwig * 43901c03194SChristoph Hellwig * We'll have another go at writing back this inode when we 44066f3b8e2SJens Axboe * completed a full scan of b_io. 44101c03194SChristoph Hellwig */ 442a9185b41SChristoph Hellwig if (wbc->sync_mode != WB_SYNC_ALL) { 44301c03194SChristoph Hellwig requeue_io(inode); 44401c03194SChristoph Hellwig return 0; 44501c03194SChristoph Hellwig } 44601c03194SChristoph Hellwig 44701c03194SChristoph Hellwig /* 44801c03194SChristoph Hellwig * It's a data-integrity sync. We must wait. 44901c03194SChristoph Hellwig */ 45001c03194SChristoph Hellwig inode_wait_for_writeback(inode); 45101c03194SChristoph Hellwig } 45201c03194SChristoph Hellwig 4531c0eeaf5SJoern Engel BUG_ON(inode->i_state & I_SYNC); 4541da177e4SLinus Torvalds 4555547e8aaSDmitry Monakhov /* Set I_SYNC, reset I_DIRTY_PAGES */ 4561c0eeaf5SJoern Engel inode->i_state |= I_SYNC; 4575547e8aaSDmitry Monakhov inode->i_state &= ~I_DIRTY_PAGES; 4581da177e4SLinus Torvalds spin_unlock(&inode_lock); 4591da177e4SLinus Torvalds 4601da177e4SLinus Torvalds ret = do_writepages(mapping, wbc); 4611da177e4SLinus Torvalds 46226821ed4SChristoph Hellwig /* 46326821ed4SChristoph Hellwig * Make sure to wait on the data before writing out the metadata. 46426821ed4SChristoph Hellwig * This is important for filesystems that modify metadata on data 46526821ed4SChristoph Hellwig * I/O completion. 46626821ed4SChristoph Hellwig */ 467a9185b41SChristoph Hellwig if (wbc->sync_mode == WB_SYNC_ALL) { 46826821ed4SChristoph Hellwig int err = filemap_fdatawait(mapping); 4691da177e4SLinus Torvalds if (ret == 0) 4701da177e4SLinus Torvalds ret = err; 4711da177e4SLinus Torvalds } 4721da177e4SLinus Torvalds 4735547e8aaSDmitry Monakhov /* 4745547e8aaSDmitry Monakhov * Some filesystems may redirty the inode during the writeback 4755547e8aaSDmitry Monakhov * due to delalloc, clear dirty metadata flags right before 4765547e8aaSDmitry Monakhov * write_inode() 4775547e8aaSDmitry Monakhov */ 4785547e8aaSDmitry Monakhov spin_lock(&inode_lock); 4795547e8aaSDmitry Monakhov dirty = inode->i_state & I_DIRTY; 4805547e8aaSDmitry Monakhov inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC); 4815547e8aaSDmitry Monakhov spin_unlock(&inode_lock); 48226821ed4SChristoph Hellwig /* Don't write the inode if only I_DIRTY_PAGES was set */ 48326821ed4SChristoph Hellwig if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { 484a9185b41SChristoph Hellwig int err = write_inode(inode, wbc); 4851da177e4SLinus Torvalds if (ret == 0) 4861da177e4SLinus Torvalds ret = err; 4871da177e4SLinus Torvalds } 4881da177e4SLinus Torvalds 4891da177e4SLinus Torvalds spin_lock(&inode_lock); 4901c0eeaf5SJoern Engel inode->i_state &= ~I_SYNC; 49184a89245SWu Fengguang if (!(inode->i_state & (I_FREEING | I_CLEAR))) { 492b3af9468SWu Fengguang if ((inode->i_state & I_DIRTY_PAGES) && wbc->for_kupdate) { 493ae1b7f7dSWu Fengguang /* 494b3af9468SWu Fengguang * More pages get dirtied by a fast dirtier. 495b3af9468SWu Fengguang */ 496b3af9468SWu Fengguang goto select_queue; 497b3af9468SWu Fengguang } else if (inode->i_state & I_DIRTY) { 498b3af9468SWu Fengguang /* 499b3af9468SWu Fengguang * At least XFS will redirty the inode during the 500b3af9468SWu Fengguang * writeback (delalloc) and on io completion (isize). 501ae1b7f7dSWu Fengguang */ 502ae1b7f7dSWu Fengguang redirty_tail(inode); 503ae1b7f7dSWu Fengguang } else if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { 5041da177e4SLinus Torvalds /* 5051da177e4SLinus Torvalds * We didn't write back all the pages. nfs_writepages() 5061da177e4SLinus Torvalds * sometimes bales out without doing anything. Redirty 50766f3b8e2SJens Axboe * the inode; Move it from b_io onto b_more_io/b_dirty. 5081b43ef91SAndrew Morton */ 5091b43ef91SAndrew Morton /* 5101b43ef91SAndrew Morton * akpm: if the caller was the kupdate function we put 51166f3b8e2SJens Axboe * this inode at the head of b_dirty so it gets first 5121b43ef91SAndrew Morton * consideration. Otherwise, move it to the tail, for 5131b43ef91SAndrew Morton * the reasons described there. I'm not really sure 5141b43ef91SAndrew Morton * how much sense this makes. Presumably I had a good 5151b43ef91SAndrew Morton * reasons for doing it this way, and I'd rather not 5161b43ef91SAndrew Morton * muck with it at present. 5171da177e4SLinus Torvalds */ 5181da177e4SLinus Torvalds if (wbc->for_kupdate) { 5191da177e4SLinus Torvalds /* 5202c136579SFengguang Wu * For the kupdate function we move the inode 52166f3b8e2SJens Axboe * to b_more_io so it will get more writeout as 5222c136579SFengguang Wu * soon as the queue becomes uncongested. 5231da177e4SLinus Torvalds */ 5241da177e4SLinus Torvalds inode->i_state |= I_DIRTY_PAGES; 525b3af9468SWu Fengguang select_queue: 5268bc3be27SFengguang Wu if (wbc->nr_to_write <= 0) { 5278bc3be27SFengguang Wu /* 5288bc3be27SFengguang Wu * slice used up: queue for next turn 5298bc3be27SFengguang Wu */ 5300e0f4fc2SKen Chen requeue_io(inode); 5311da177e4SLinus Torvalds } else { 5321da177e4SLinus Torvalds /* 5338bc3be27SFengguang Wu * somehow blocked: retry later 5348bc3be27SFengguang Wu */ 5358bc3be27SFengguang Wu redirty_tail(inode); 5368bc3be27SFengguang Wu } 5378bc3be27SFengguang Wu } else { 5388bc3be27SFengguang Wu /* 5391da177e4SLinus Torvalds * Otherwise fully redirty the inode so that 5401da177e4SLinus Torvalds * other inodes on this superblock will get some 5411da177e4SLinus Torvalds * writeout. Otherwise heavy writing to one 5421da177e4SLinus Torvalds * file would indefinitely suspend writeout of 5431da177e4SLinus Torvalds * all the other files. 5441da177e4SLinus Torvalds */ 5451da177e4SLinus Torvalds inode->i_state |= I_DIRTY_PAGES; 5461b43ef91SAndrew Morton redirty_tail(inode); 5471da177e4SLinus Torvalds } 5481da177e4SLinus Torvalds } else if (atomic_read(&inode->i_count)) { 5491da177e4SLinus Torvalds /* 5501da177e4SLinus Torvalds * The inode is clean, inuse 5511da177e4SLinus Torvalds */ 5521da177e4SLinus Torvalds list_move(&inode->i_list, &inode_in_use); 5531da177e4SLinus Torvalds } else { 5541da177e4SLinus Torvalds /* 5551da177e4SLinus Torvalds * The inode is clean, unused 5561da177e4SLinus Torvalds */ 5571da177e4SLinus Torvalds list_move(&inode->i_list, &inode_unused); 5581da177e4SLinus Torvalds } 5591da177e4SLinus Torvalds } 5601c0eeaf5SJoern Engel inode_sync_complete(inode); 5611da177e4SLinus Torvalds return ret; 5621da177e4SLinus Torvalds } 5631da177e4SLinus Torvalds 564f11c9c5cSEdward Shishkin static void unpin_sb_for_writeback(struct super_block *sb) 5659ecc2738SJens Axboe { 5669ecc2738SJens Axboe up_read(&sb->s_umount); 5679ecc2738SJens Axboe put_super(sb); 5689ecc2738SJens Axboe } 569f11c9c5cSEdward Shishkin 570f11c9c5cSEdward Shishkin enum sb_pin_state { 571f11c9c5cSEdward Shishkin SB_PINNED, 572f11c9c5cSEdward Shishkin SB_NOT_PINNED, 573f11c9c5cSEdward Shishkin SB_PIN_FAILED 574f11c9c5cSEdward Shishkin }; 5759ecc2738SJens Axboe 57603ba3782SJens Axboe /* 57703ba3782SJens Axboe * For WB_SYNC_NONE writeback, the caller does not have the sb pinned 57803ba3782SJens Axboe * before calling writeback. So make sure that we do pin it, so it doesn't 57903ba3782SJens Axboe * go away while we are writing inodes from it. 58003ba3782SJens Axboe */ 581f11c9c5cSEdward Shishkin static enum sb_pin_state pin_sb_for_writeback(struct writeback_control *wbc, 582f11c9c5cSEdward Shishkin struct super_block *sb) 5831da177e4SLinus Torvalds { 5849ecc2738SJens Axboe /* 58503ba3782SJens Axboe * Caller must already hold the ref for this 58603ba3782SJens Axboe */ 5870e3c9a22SJens Axboe if (wbc->sync_mode == WB_SYNC_ALL) { 58803ba3782SJens Axboe WARN_ON(!rwsem_is_locked(&sb->s_umount)); 589f11c9c5cSEdward Shishkin return SB_NOT_PINNED; 59003ba3782SJens Axboe } 59103ba3782SJens Axboe spin_lock(&sb_lock); 59203ba3782SJens Axboe sb->s_count++; 59303ba3782SJens Axboe if (down_read_trylock(&sb->s_umount)) { 59403ba3782SJens Axboe if (sb->s_root) { 59503ba3782SJens Axboe spin_unlock(&sb_lock); 596f11c9c5cSEdward Shishkin return SB_PINNED; 59703ba3782SJens Axboe } 59803ba3782SJens Axboe /* 59903ba3782SJens Axboe * umounted, drop rwsem again and fall through to failure 60003ba3782SJens Axboe */ 60103ba3782SJens Axboe up_read(&sb->s_umount); 60203ba3782SJens Axboe } 60303ba3782SJens Axboe sb->s_count--; 60403ba3782SJens Axboe spin_unlock(&sb_lock); 605f11c9c5cSEdward Shishkin return SB_PIN_FAILED; 60603ba3782SJens Axboe } 60703ba3782SJens Axboe 608f11c9c5cSEdward Shishkin /* 609f11c9c5cSEdward Shishkin * Write a portion of b_io inodes which belong to @sb. 610f11c9c5cSEdward Shishkin * If @wbc->sb != NULL, then find and write all such 611f11c9c5cSEdward Shishkin * inodes. Otherwise write only ones which go sequentially 612f11c9c5cSEdward Shishkin * in reverse order. 613f11c9c5cSEdward Shishkin * Return 1, if the caller writeback routine should be 614f11c9c5cSEdward Shishkin * interrupted. Otherwise return 0. 615f11c9c5cSEdward Shishkin */ 616f11c9c5cSEdward Shishkin static int writeback_sb_inodes(struct super_block *sb, 617f11c9c5cSEdward Shishkin struct bdi_writeback *wb, 61803ba3782SJens Axboe struct writeback_control *wbc) 61903ba3782SJens Axboe { 62003ba3782SJens Axboe while (!list_empty(&wb->b_io)) { 621f11c9c5cSEdward Shishkin long pages_skipped; 62203ba3782SJens Axboe struct inode *inode = list_entry(wb->b_io.prev, 6231da177e4SLinus Torvalds struct inode, i_list); 624f11c9c5cSEdward Shishkin if (wbc->sb && sb != inode->i_sb) { 625f11c9c5cSEdward Shishkin /* super block given and doesn't 626f11c9c5cSEdward Shishkin match, skip this inode */ 62766f3b8e2SJens Axboe redirty_tail(inode); 62866f3b8e2SJens Axboe continue; 62966f3b8e2SJens Axboe } 630f11c9c5cSEdward Shishkin if (sb != inode->i_sb) 631f11c9c5cSEdward Shishkin /* finish with this superblock */ 632f11c9c5cSEdward Shishkin return 0; 63384a89245SWu Fengguang if (inode->i_state & (I_NEW | I_WILL_FREE)) { 6347ef0d737SNick Piggin requeue_io(inode); 6357ef0d737SNick Piggin continue; 6367ef0d737SNick Piggin } 637d2caa3c5SJeff Layton /* 638d2caa3c5SJeff Layton * Was this inode dirtied after sync_sb_inodes was called? 639d2caa3c5SJeff Layton * This keeps sync from extra jobs and livelock. 640d2caa3c5SJeff Layton */ 641f11c9c5cSEdward Shishkin if (inode_dirtied_after(inode, wbc->wb_start)) 642f11c9c5cSEdward Shishkin return 1; 6431da177e4SLinus Torvalds 64484a89245SWu Fengguang BUG_ON(inode->i_state & (I_FREEING | I_CLEAR)); 6451da177e4SLinus Torvalds __iget(inode); 6461da177e4SLinus Torvalds pages_skipped = wbc->pages_skipped; 64701c03194SChristoph Hellwig writeback_single_inode(inode, wbc); 6481da177e4SLinus Torvalds if (wbc->pages_skipped != pages_skipped) { 6491da177e4SLinus Torvalds /* 6501da177e4SLinus Torvalds * writeback is not making progress due to locked 6511da177e4SLinus Torvalds * buffers. Skip this inode for now. 6521da177e4SLinus Torvalds */ 653f57b9b7bSAndrew Morton redirty_tail(inode); 6541da177e4SLinus Torvalds } 6551da177e4SLinus Torvalds spin_unlock(&inode_lock); 6561da177e4SLinus Torvalds iput(inode); 6574ffc8444SOGAWA Hirofumi cond_resched(); 6581da177e4SLinus Torvalds spin_lock(&inode_lock); 6598bc3be27SFengguang Wu if (wbc->nr_to_write <= 0) { 6608bc3be27SFengguang Wu wbc->more_io = 1; 661f11c9c5cSEdward Shishkin return 1; 6621da177e4SLinus Torvalds } 66303ba3782SJens Axboe if (!list_empty(&wb->b_more_io)) 6648bc3be27SFengguang Wu wbc->more_io = 1; 6658bc3be27SFengguang Wu } 666f11c9c5cSEdward Shishkin /* b_io is empty */ 667f11c9c5cSEdward Shishkin return 1; 668f11c9c5cSEdward Shishkin } 66938f21977SNick Piggin 670f11c9c5cSEdward Shishkin static void writeback_inodes_wb(struct bdi_writeback *wb, 671f11c9c5cSEdward Shishkin struct writeback_control *wbc) 672f11c9c5cSEdward Shishkin { 673f11c9c5cSEdward Shishkin int ret = 0; 6749ecc2738SJens Axboe 675f11c9c5cSEdward Shishkin wbc->wb_start = jiffies; /* livelock avoidance */ 676f11c9c5cSEdward Shishkin spin_lock(&inode_lock); 677f11c9c5cSEdward Shishkin if (!wbc->for_kupdate || list_empty(&wb->b_io)) 678f11c9c5cSEdward Shishkin queue_io(wb, wbc->older_than_this); 679f11c9c5cSEdward Shishkin 680f11c9c5cSEdward Shishkin while (!list_empty(&wb->b_io)) { 681f11c9c5cSEdward Shishkin struct inode *inode = list_entry(wb->b_io.prev, 682f11c9c5cSEdward Shishkin struct inode, i_list); 683f11c9c5cSEdward Shishkin struct super_block *sb = inode->i_sb; 684f11c9c5cSEdward Shishkin enum sb_pin_state state; 685f11c9c5cSEdward Shishkin 686f11c9c5cSEdward Shishkin if (wbc->sb && sb != wbc->sb) { 687f11c9c5cSEdward Shishkin /* super block given and doesn't 688f11c9c5cSEdward Shishkin match, skip this inode */ 689f11c9c5cSEdward Shishkin redirty_tail(inode); 690f11c9c5cSEdward Shishkin continue; 691f11c9c5cSEdward Shishkin } 692f11c9c5cSEdward Shishkin state = pin_sb_for_writeback(wbc, sb); 693f11c9c5cSEdward Shishkin 694f11c9c5cSEdward Shishkin if (state == SB_PIN_FAILED) { 695f11c9c5cSEdward Shishkin requeue_io(inode); 696f11c9c5cSEdward Shishkin continue; 697f11c9c5cSEdward Shishkin } 698f11c9c5cSEdward Shishkin ret = writeback_sb_inodes(sb, wb, wbc); 699f11c9c5cSEdward Shishkin 700f11c9c5cSEdward Shishkin if (state == SB_PINNED) 701f11c9c5cSEdward Shishkin unpin_sb_for_writeback(sb); 702f11c9c5cSEdward Shishkin if (ret) 703f11c9c5cSEdward Shishkin break; 704f11c9c5cSEdward Shishkin } 70566f3b8e2SJens Axboe spin_unlock(&inode_lock); 70666f3b8e2SJens Axboe /* Leave any unwritten inodes on b_io */ 70766f3b8e2SJens Axboe } 70866f3b8e2SJens Axboe 70903ba3782SJens Axboe void writeback_inodes_wbc(struct writeback_control *wbc) 71003ba3782SJens Axboe { 71103ba3782SJens Axboe struct backing_dev_info *bdi = wbc->bdi; 71203ba3782SJens Axboe 71303ba3782SJens Axboe writeback_inodes_wb(&bdi->wb, wbc); 71403ba3782SJens Axboe } 71503ba3782SJens Axboe 71603ba3782SJens Axboe /* 71703ba3782SJens Axboe * The maximum number of pages to writeout in a single bdi flush/kupdate 71803ba3782SJens Axboe * operation. We do this so we don't hold I_SYNC against an inode for 71903ba3782SJens Axboe * enormous amounts of time, which would block a userspace task which has 72003ba3782SJens Axboe * been forced to throttle against that inode. Also, the code reevaluates 72103ba3782SJens Axboe * the dirty each time it has written this many pages. 72203ba3782SJens Axboe */ 72303ba3782SJens Axboe #define MAX_WRITEBACK_PAGES 1024 72403ba3782SJens Axboe 72503ba3782SJens Axboe static inline bool over_bground_thresh(void) 72603ba3782SJens Axboe { 72703ba3782SJens Axboe unsigned long background_thresh, dirty_thresh; 72803ba3782SJens Axboe 72903ba3782SJens Axboe get_dirty_limits(&background_thresh, &dirty_thresh, NULL, NULL); 73003ba3782SJens Axboe 73103ba3782SJens Axboe return (global_page_state(NR_FILE_DIRTY) + 73203ba3782SJens Axboe global_page_state(NR_UNSTABLE_NFS) >= background_thresh); 73303ba3782SJens Axboe } 73403ba3782SJens Axboe 73503ba3782SJens Axboe /* 73603ba3782SJens Axboe * Explicit flushing or periodic writeback of "old" data. 73703ba3782SJens Axboe * 73803ba3782SJens Axboe * Define "old": the first time one of an inode's pages is dirtied, we mark the 73903ba3782SJens Axboe * dirtying-time in the inode's address_space. So this periodic writeback code 74003ba3782SJens Axboe * just walks the superblock inode list, writing back any inodes which are 74103ba3782SJens Axboe * older than a specific point in time. 74203ba3782SJens Axboe * 74303ba3782SJens Axboe * Try to run once per dirty_writeback_interval. But if a writeback event 74403ba3782SJens Axboe * takes longer than a dirty_writeback_interval interval, then leave a 74503ba3782SJens Axboe * one-second gap. 74603ba3782SJens Axboe * 74703ba3782SJens Axboe * older_than_this takes precedence over nr_to_write. So we'll only write back 74803ba3782SJens Axboe * all dirty pages if they are all attached to "old" mappings. 74903ba3782SJens Axboe */ 750c4a77a6cSJens Axboe static long wb_writeback(struct bdi_writeback *wb, 751c4a77a6cSJens Axboe struct wb_writeback_args *args) 75203ba3782SJens Axboe { 75303ba3782SJens Axboe struct writeback_control wbc = { 75403ba3782SJens Axboe .bdi = wb->bdi, 755c4a77a6cSJens Axboe .sb = args->sb, 756c4a77a6cSJens Axboe .sync_mode = args->sync_mode, 75703ba3782SJens Axboe .older_than_this = NULL, 758c4a77a6cSJens Axboe .for_kupdate = args->for_kupdate, 759b17621feSWu Fengguang .for_background = args->for_background, 760c4a77a6cSJens Axboe .range_cyclic = args->range_cyclic, 76103ba3782SJens Axboe }; 76203ba3782SJens Axboe unsigned long oldest_jif; 76303ba3782SJens Axboe long wrote = 0; 764a5989bdcSJan Kara struct inode *inode; 76503ba3782SJens Axboe 76603ba3782SJens Axboe if (wbc.for_kupdate) { 76703ba3782SJens Axboe wbc.older_than_this = &oldest_jif; 76803ba3782SJens Axboe oldest_jif = jiffies - 76903ba3782SJens Axboe msecs_to_jiffies(dirty_expire_interval * 10); 77003ba3782SJens Axboe } 771c4a77a6cSJens Axboe if (!wbc.range_cyclic) { 772c4a77a6cSJens Axboe wbc.range_start = 0; 773c4a77a6cSJens Axboe wbc.range_end = LLONG_MAX; 774c4a77a6cSJens Axboe } 77503ba3782SJens Axboe 77603ba3782SJens Axboe for (;;) { 77703ba3782SJens Axboe /* 778d3ddec76SWu Fengguang * Stop writeback when nr_pages has been consumed 77903ba3782SJens Axboe */ 780d3ddec76SWu Fengguang if (args->nr_pages <= 0) 78103ba3782SJens Axboe break; 78203ba3782SJens Axboe 78303ba3782SJens Axboe /* 784d3ddec76SWu Fengguang * For background writeout, stop when we are below the 785d3ddec76SWu Fengguang * background dirty threshold 78603ba3782SJens Axboe */ 787d3ddec76SWu Fengguang if (args->for_background && !over_bground_thresh()) 78803ba3782SJens Axboe break; 78903ba3782SJens Axboe 79003ba3782SJens Axboe wbc.more_io = 0; 79103ba3782SJens Axboe wbc.nr_to_write = MAX_WRITEBACK_PAGES; 79203ba3782SJens Axboe wbc.pages_skipped = 0; 79303ba3782SJens Axboe writeback_inodes_wb(wb, &wbc); 794c4a77a6cSJens Axboe args->nr_pages -= MAX_WRITEBACK_PAGES - wbc.nr_to_write; 79503ba3782SJens Axboe wrote += MAX_WRITEBACK_PAGES - wbc.nr_to_write; 79603ba3782SJens Axboe 79703ba3782SJens Axboe /* 79871fd05a8SJens Axboe * If we consumed everything, see if we have more 79903ba3782SJens Axboe */ 80071fd05a8SJens Axboe if (wbc.nr_to_write <= 0) 80171fd05a8SJens Axboe continue; 80271fd05a8SJens Axboe /* 80371fd05a8SJens Axboe * Didn't write everything and we don't have more IO, bail 80471fd05a8SJens Axboe */ 80571fd05a8SJens Axboe if (!wbc.more_io) 80671fd05a8SJens Axboe break; 80771fd05a8SJens Axboe /* 80871fd05a8SJens Axboe * Did we write something? Try for more 80971fd05a8SJens Axboe */ 810a5989bdcSJan Kara if (wbc.nr_to_write < MAX_WRITEBACK_PAGES) 81103ba3782SJens Axboe continue; 812a5989bdcSJan Kara /* 813a5989bdcSJan Kara * Nothing written. Wait for some inode to 814a5989bdcSJan Kara * become available for writeback. Otherwise 815a5989bdcSJan Kara * we'll just busyloop. 816a5989bdcSJan Kara */ 817a5989bdcSJan Kara spin_lock(&inode_lock); 818a5989bdcSJan Kara if (!list_empty(&wb->b_more_io)) { 81971fd05a8SJens Axboe inode = list_entry(wb->b_more_io.prev, 820a5989bdcSJan Kara struct inode, i_list); 821a5989bdcSJan Kara inode_wait_for_writeback(inode); 822a5989bdcSJan Kara } 823a5989bdcSJan Kara spin_unlock(&inode_lock); 82403ba3782SJens Axboe } 82503ba3782SJens Axboe 82603ba3782SJens Axboe return wrote; 82703ba3782SJens Axboe } 82803ba3782SJens Axboe 82903ba3782SJens Axboe /* 83003ba3782SJens Axboe * Return the next bdi_work struct that hasn't been processed by this 8318010c3b6SJens Axboe * wb thread yet. ->seen is initially set for each thread that exists 8328010c3b6SJens Axboe * for this device, when a thread first notices a piece of work it 8338010c3b6SJens Axboe * clears its bit. Depending on writeback type, the thread will notify 8348010c3b6SJens Axboe * completion on either receiving the work (WB_SYNC_NONE) or after 8358010c3b6SJens Axboe * it is done (WB_SYNC_ALL). 83603ba3782SJens Axboe */ 83703ba3782SJens Axboe static struct bdi_work *get_next_work_item(struct backing_dev_info *bdi, 83803ba3782SJens Axboe struct bdi_writeback *wb) 83903ba3782SJens Axboe { 84003ba3782SJens Axboe struct bdi_work *work, *ret = NULL; 84103ba3782SJens Axboe 84203ba3782SJens Axboe rcu_read_lock(); 84303ba3782SJens Axboe 84403ba3782SJens Axboe list_for_each_entry_rcu(work, &bdi->work_list, list) { 84577fad5e6SNick Piggin if (!test_bit(wb->nr, &work->seen)) 84603ba3782SJens Axboe continue; 84777fad5e6SNick Piggin clear_bit(wb->nr, &work->seen); 84803ba3782SJens Axboe 84903ba3782SJens Axboe ret = work; 85003ba3782SJens Axboe break; 85103ba3782SJens Axboe } 85203ba3782SJens Axboe 85303ba3782SJens Axboe rcu_read_unlock(); 85403ba3782SJens Axboe return ret; 85503ba3782SJens Axboe } 85603ba3782SJens Axboe 85703ba3782SJens Axboe static long wb_check_old_data_flush(struct bdi_writeback *wb) 85803ba3782SJens Axboe { 85903ba3782SJens Axboe unsigned long expired; 86003ba3782SJens Axboe long nr_pages; 86103ba3782SJens Axboe 86269b62d01SJens Axboe /* 86369b62d01SJens Axboe * When set to zero, disable periodic writeback 86469b62d01SJens Axboe */ 86569b62d01SJens Axboe if (!dirty_writeback_interval) 86669b62d01SJens Axboe return 0; 86769b62d01SJens Axboe 86803ba3782SJens Axboe expired = wb->last_old_flush + 86903ba3782SJens Axboe msecs_to_jiffies(dirty_writeback_interval * 10); 87003ba3782SJens Axboe if (time_before(jiffies, expired)) 87103ba3782SJens Axboe return 0; 87203ba3782SJens Axboe 87303ba3782SJens Axboe wb->last_old_flush = jiffies; 87403ba3782SJens Axboe nr_pages = global_page_state(NR_FILE_DIRTY) + 87503ba3782SJens Axboe global_page_state(NR_UNSTABLE_NFS) + 87603ba3782SJens Axboe (inodes_stat.nr_inodes - inodes_stat.nr_unused); 87703ba3782SJens Axboe 878c4a77a6cSJens Axboe if (nr_pages) { 879c4a77a6cSJens Axboe struct wb_writeback_args args = { 880c4a77a6cSJens Axboe .nr_pages = nr_pages, 881c4a77a6cSJens Axboe .sync_mode = WB_SYNC_NONE, 882c4a77a6cSJens Axboe .for_kupdate = 1, 883c4a77a6cSJens Axboe .range_cyclic = 1, 884c4a77a6cSJens Axboe }; 885c4a77a6cSJens Axboe 886c4a77a6cSJens Axboe return wb_writeback(wb, &args); 887c4a77a6cSJens Axboe } 88803ba3782SJens Axboe 88903ba3782SJens Axboe return 0; 89003ba3782SJens Axboe } 89103ba3782SJens Axboe 89203ba3782SJens Axboe /* 89303ba3782SJens Axboe * Retrieve work items and do the writeback they describe 89403ba3782SJens Axboe */ 89503ba3782SJens Axboe long wb_do_writeback(struct bdi_writeback *wb, int force_wait) 89603ba3782SJens Axboe { 89703ba3782SJens Axboe struct backing_dev_info *bdi = wb->bdi; 89803ba3782SJens Axboe struct bdi_work *work; 899c4a77a6cSJens Axboe long wrote = 0; 90003ba3782SJens Axboe 90103ba3782SJens Axboe while ((work = get_next_work_item(bdi, wb)) != NULL) { 902c4a77a6cSJens Axboe struct wb_writeback_args args = work->args; 90303ba3782SJens Axboe 90403ba3782SJens Axboe /* 90503ba3782SJens Axboe * Override sync mode, in case we must wait for completion 90603ba3782SJens Axboe */ 90703ba3782SJens Axboe if (force_wait) 908c4a77a6cSJens Axboe work->args.sync_mode = args.sync_mode = WB_SYNC_ALL; 90903ba3782SJens Axboe 91003ba3782SJens Axboe /* 91103ba3782SJens Axboe * If this isn't a data integrity operation, just notify 91203ba3782SJens Axboe * that we have seen this work and we are now starting it. 91303ba3782SJens Axboe */ 914f17625b3SJens Axboe if (args.sync_mode == WB_SYNC_NONE) 91503ba3782SJens Axboe wb_clear_pending(wb, work); 91603ba3782SJens Axboe 917c4a77a6cSJens Axboe wrote += wb_writeback(wb, &args); 91803ba3782SJens Axboe 91903ba3782SJens Axboe /* 92003ba3782SJens Axboe * This is a data integrity writeback, so only do the 92103ba3782SJens Axboe * notification when we have completed the work. 92203ba3782SJens Axboe */ 923f17625b3SJens Axboe if (args.sync_mode == WB_SYNC_ALL) 92403ba3782SJens Axboe wb_clear_pending(wb, work); 92503ba3782SJens Axboe } 92603ba3782SJens Axboe 92703ba3782SJens Axboe /* 92803ba3782SJens Axboe * Check for periodic writeback, kupdated() style 92903ba3782SJens Axboe */ 93003ba3782SJens Axboe wrote += wb_check_old_data_flush(wb); 93103ba3782SJens Axboe 93203ba3782SJens Axboe return wrote; 93303ba3782SJens Axboe } 93403ba3782SJens Axboe 93503ba3782SJens Axboe /* 93603ba3782SJens Axboe * Handle writeback of dirty data for the device backed by this bdi. Also 93703ba3782SJens Axboe * wakes up periodically and does kupdated style flushing. 93803ba3782SJens Axboe */ 93903ba3782SJens Axboe int bdi_writeback_task(struct bdi_writeback *wb) 94003ba3782SJens Axboe { 94103ba3782SJens Axboe unsigned long last_active = jiffies; 94203ba3782SJens Axboe unsigned long wait_jiffies = -1UL; 94303ba3782SJens Axboe long pages_written; 94403ba3782SJens Axboe 94503ba3782SJens Axboe while (!kthread_should_stop()) { 94603ba3782SJens Axboe pages_written = wb_do_writeback(wb, 0); 94703ba3782SJens Axboe 94803ba3782SJens Axboe if (pages_written) 94903ba3782SJens Axboe last_active = jiffies; 95003ba3782SJens Axboe else if (wait_jiffies != -1UL) { 95103ba3782SJens Axboe unsigned long max_idle; 95203ba3782SJens Axboe 95303ba3782SJens Axboe /* 95403ba3782SJens Axboe * Longest period of inactivity that we tolerate. If we 95503ba3782SJens Axboe * see dirty data again later, the task will get 95603ba3782SJens Axboe * recreated automatically. 95703ba3782SJens Axboe */ 95803ba3782SJens Axboe max_idle = max(5UL * 60 * HZ, wait_jiffies); 95903ba3782SJens Axboe if (time_after(jiffies, max_idle + last_active)) 96003ba3782SJens Axboe break; 96103ba3782SJens Axboe } 96203ba3782SJens Axboe 96369b62d01SJens Axboe if (dirty_writeback_interval) { 96403ba3782SJens Axboe wait_jiffies = msecs_to_jiffies(dirty_writeback_interval * 10); 96549db0414SJens Axboe schedule_timeout_interruptible(wait_jiffies); 966f9eadbbdSJens Axboe } else { 967f9eadbbdSJens Axboe set_current_state(TASK_INTERRUPTIBLE); 968f9eadbbdSJens Axboe if (list_empty_careful(&wb->bdi->work_list) && 969f9eadbbdSJens Axboe !kthread_should_stop()) 97069b62d01SJens Axboe schedule(); 971f9eadbbdSJens Axboe __set_current_state(TASK_RUNNING); 972f9eadbbdSJens Axboe } 97369b62d01SJens Axboe 97403ba3782SJens Axboe try_to_freeze(); 97503ba3782SJens Axboe } 97603ba3782SJens Axboe 97703ba3782SJens Axboe return 0; 97803ba3782SJens Axboe } 97903ba3782SJens Axboe 98003ba3782SJens Axboe /* 981b6e51316SJens Axboe * Schedule writeback for all backing devices. This does WB_SYNC_NONE 982b6e51316SJens Axboe * writeback, for integrity writeback see bdi_sync_writeback(). 98303ba3782SJens Axboe */ 984b6e51316SJens Axboe static void bdi_writeback_all(struct super_block *sb, long nr_pages) 98503ba3782SJens Axboe { 986b6e51316SJens Axboe struct wb_writeback_args args = { 987b6e51316SJens Axboe .sb = sb, 988b6e51316SJens Axboe .nr_pages = nr_pages, 989b6e51316SJens Axboe .sync_mode = WB_SYNC_NONE, 990b6e51316SJens Axboe }; 99103ba3782SJens Axboe struct backing_dev_info *bdi; 99203ba3782SJens Axboe 993cfc4ba53SJens Axboe rcu_read_lock(); 99403ba3782SJens Axboe 995cfc4ba53SJens Axboe list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { 99603ba3782SJens Axboe if (!bdi_has_dirty_io(bdi)) 99703ba3782SJens Axboe continue; 99803ba3782SJens Axboe 999f17625b3SJens Axboe bdi_alloc_queue_work(bdi, &args); 100003ba3782SJens Axboe } 100103ba3782SJens Axboe 1002cfc4ba53SJens Axboe rcu_read_unlock(); 100303ba3782SJens Axboe } 100403ba3782SJens Axboe 100503ba3782SJens Axboe /* 100603ba3782SJens Axboe * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back 100703ba3782SJens Axboe * the whole world. 100803ba3782SJens Axboe */ 100903ba3782SJens Axboe void wakeup_flusher_threads(long nr_pages) 101003ba3782SJens Axboe { 101103ba3782SJens Axboe if (nr_pages == 0) 101203ba3782SJens Axboe nr_pages = global_page_state(NR_FILE_DIRTY) + 101303ba3782SJens Axboe global_page_state(NR_UNSTABLE_NFS); 1014b6e51316SJens Axboe bdi_writeback_all(NULL, nr_pages); 101503ba3782SJens Axboe } 101603ba3782SJens Axboe 101703ba3782SJens Axboe static noinline void block_dump___mark_inode_dirty(struct inode *inode) 101803ba3782SJens Axboe { 101903ba3782SJens Axboe if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { 102003ba3782SJens Axboe struct dentry *dentry; 102103ba3782SJens Axboe const char *name = "?"; 102203ba3782SJens Axboe 102303ba3782SJens Axboe dentry = d_find_alias(inode); 102403ba3782SJens Axboe if (dentry) { 102503ba3782SJens Axboe spin_lock(&dentry->d_lock); 102603ba3782SJens Axboe name = (const char *) dentry->d_name.name; 102703ba3782SJens Axboe } 102803ba3782SJens Axboe printk(KERN_DEBUG 102903ba3782SJens Axboe "%s(%d): dirtied inode %lu (%s) on %s\n", 103003ba3782SJens Axboe current->comm, task_pid_nr(current), inode->i_ino, 103103ba3782SJens Axboe name, inode->i_sb->s_id); 103203ba3782SJens Axboe if (dentry) { 103303ba3782SJens Axboe spin_unlock(&dentry->d_lock); 103403ba3782SJens Axboe dput(dentry); 103503ba3782SJens Axboe } 103603ba3782SJens Axboe } 103703ba3782SJens Axboe } 103803ba3782SJens Axboe 103903ba3782SJens Axboe /** 104003ba3782SJens Axboe * __mark_inode_dirty - internal function 104103ba3782SJens Axboe * @inode: inode to mark 104203ba3782SJens Axboe * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) 104303ba3782SJens Axboe * Mark an inode as dirty. Callers should use mark_inode_dirty or 104403ba3782SJens Axboe * mark_inode_dirty_sync. 104503ba3782SJens Axboe * 104603ba3782SJens Axboe * Put the inode on the super block's dirty list. 104703ba3782SJens Axboe * 104803ba3782SJens Axboe * CAREFUL! We mark it dirty unconditionally, but move it onto the 104903ba3782SJens Axboe * dirty list only if it is hashed or if it refers to a blockdev. 105003ba3782SJens Axboe * If it was not hashed, it will never be added to the dirty list 105103ba3782SJens Axboe * even if it is later hashed, as it will have been marked dirty already. 105203ba3782SJens Axboe * 105303ba3782SJens Axboe * In short, make sure you hash any inodes _before_ you start marking 105403ba3782SJens Axboe * them dirty. 105503ba3782SJens Axboe * 105603ba3782SJens Axboe * This function *must* be atomic for the I_DIRTY_PAGES case - 105703ba3782SJens Axboe * set_page_dirty() is called under spinlock in several places. 105803ba3782SJens Axboe * 105903ba3782SJens Axboe * Note that for blockdevs, inode->dirtied_when represents the dirtying time of 106003ba3782SJens Axboe * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of 106103ba3782SJens Axboe * the kernel-internal blockdev inode represents the dirtying time of the 106203ba3782SJens Axboe * blockdev's pages. This is why for I_DIRTY_PAGES we always use 106303ba3782SJens Axboe * page->mapping->host, so the page-dirtying time is recorded in the internal 106403ba3782SJens Axboe * blockdev inode. 106503ba3782SJens Axboe */ 106603ba3782SJens Axboe void __mark_inode_dirty(struct inode *inode, int flags) 106703ba3782SJens Axboe { 106803ba3782SJens Axboe struct super_block *sb = inode->i_sb; 106903ba3782SJens Axboe 107003ba3782SJens Axboe /* 107103ba3782SJens Axboe * Don't do this for I_DIRTY_PAGES - that doesn't actually 107203ba3782SJens Axboe * dirty the inode itself 107303ba3782SJens Axboe */ 107403ba3782SJens Axboe if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { 107503ba3782SJens Axboe if (sb->s_op->dirty_inode) 107603ba3782SJens Axboe sb->s_op->dirty_inode(inode); 107703ba3782SJens Axboe } 107803ba3782SJens Axboe 107903ba3782SJens Axboe /* 108003ba3782SJens Axboe * make sure that changes are seen by all cpus before we test i_state 108103ba3782SJens Axboe * -- mikulas 108203ba3782SJens Axboe */ 108303ba3782SJens Axboe smp_mb(); 108403ba3782SJens Axboe 108503ba3782SJens Axboe /* avoid the locking if we can */ 108603ba3782SJens Axboe if ((inode->i_state & flags) == flags) 108703ba3782SJens Axboe return; 108803ba3782SJens Axboe 108903ba3782SJens Axboe if (unlikely(block_dump)) 109003ba3782SJens Axboe block_dump___mark_inode_dirty(inode); 109103ba3782SJens Axboe 109203ba3782SJens Axboe spin_lock(&inode_lock); 109303ba3782SJens Axboe if ((inode->i_state & flags) != flags) { 109403ba3782SJens Axboe const int was_dirty = inode->i_state & I_DIRTY; 109503ba3782SJens Axboe 109603ba3782SJens Axboe inode->i_state |= flags; 109703ba3782SJens Axboe 109803ba3782SJens Axboe /* 109903ba3782SJens Axboe * If the inode is being synced, just update its dirty state. 110003ba3782SJens Axboe * The unlocker will place the inode on the appropriate 110103ba3782SJens Axboe * superblock list, based upon its state. 110203ba3782SJens Axboe */ 110303ba3782SJens Axboe if (inode->i_state & I_SYNC) 110403ba3782SJens Axboe goto out; 110503ba3782SJens Axboe 110603ba3782SJens Axboe /* 110703ba3782SJens Axboe * Only add valid (hashed) inodes to the superblock's 110803ba3782SJens Axboe * dirty list. Add blockdev inodes as well. 110903ba3782SJens Axboe */ 111003ba3782SJens Axboe if (!S_ISBLK(inode->i_mode)) { 111103ba3782SJens Axboe if (hlist_unhashed(&inode->i_hash)) 111203ba3782SJens Axboe goto out; 111303ba3782SJens Axboe } 111403ba3782SJens Axboe if (inode->i_state & (I_FREEING|I_CLEAR)) 111503ba3782SJens Axboe goto out; 111603ba3782SJens Axboe 111703ba3782SJens Axboe /* 111803ba3782SJens Axboe * If the inode was already on b_dirty/b_io/b_more_io, don't 111903ba3782SJens Axboe * reposition it (that would break b_dirty time-ordering). 112003ba3782SJens Axboe */ 112103ba3782SJens Axboe if (!was_dirty) { 112203ba3782SJens Axboe struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; 1123500b067cSJens Axboe struct backing_dev_info *bdi = wb->bdi; 1124500b067cSJens Axboe 1125500b067cSJens Axboe if (bdi_cap_writeback_dirty(bdi) && 1126500b067cSJens Axboe !test_bit(BDI_registered, &bdi->state)) { 1127500b067cSJens Axboe WARN_ON(1); 1128500b067cSJens Axboe printk(KERN_ERR "bdi-%s not registered\n", 1129500b067cSJens Axboe bdi->name); 1130500b067cSJens Axboe } 113103ba3782SJens Axboe 113203ba3782SJens Axboe inode->dirtied_when = jiffies; 113303ba3782SJens Axboe list_move(&inode->i_list, &wb->b_dirty); 113403ba3782SJens Axboe } 113503ba3782SJens Axboe } 113603ba3782SJens Axboe out: 113703ba3782SJens Axboe spin_unlock(&inode_lock); 113803ba3782SJens Axboe } 113903ba3782SJens Axboe EXPORT_SYMBOL(__mark_inode_dirty); 114003ba3782SJens Axboe 114166f3b8e2SJens Axboe /* 114266f3b8e2SJens Axboe * Write out a superblock's list of dirty inodes. A wait will be performed 114366f3b8e2SJens Axboe * upon no inodes, all inodes or the final one, depending upon sync_mode. 114466f3b8e2SJens Axboe * 114566f3b8e2SJens Axboe * If older_than_this is non-NULL, then only write out inodes which 114666f3b8e2SJens Axboe * had their first dirtying at a time earlier than *older_than_this. 114766f3b8e2SJens Axboe * 114866f3b8e2SJens Axboe * If `bdi' is non-zero then we're being asked to writeback a specific queue. 114966f3b8e2SJens Axboe * This function assumes that the blockdev superblock's inodes are backed by 115066f3b8e2SJens Axboe * a variety of queues, so all inodes are searched. For other superblocks, 115166f3b8e2SJens Axboe * assume that all inodes are backed by the same queue. 115266f3b8e2SJens Axboe * 115366f3b8e2SJens Axboe * The inodes to be written are parked on bdi->b_io. They are moved back onto 115466f3b8e2SJens Axboe * bdi->b_dirty as they are selected for writing. This way, none can be missed 115566f3b8e2SJens Axboe * on the writer throttling path, and we get decent balancing between many 115666f3b8e2SJens Axboe * throttled threads: we don't want them all piling up on inode_sync_wait. 115766f3b8e2SJens Axboe */ 1158b6e51316SJens Axboe static void wait_sb_inodes(struct super_block *sb) 115966f3b8e2SJens Axboe { 116038f21977SNick Piggin struct inode *inode, *old_inode = NULL; 116138f21977SNick Piggin 116203ba3782SJens Axboe /* 116303ba3782SJens Axboe * We need to be protected against the filesystem going from 116403ba3782SJens Axboe * r/o to r/w or vice versa. 116503ba3782SJens Axboe */ 1166b6e51316SJens Axboe WARN_ON(!rwsem_is_locked(&sb->s_umount)); 116703ba3782SJens Axboe 116866f3b8e2SJens Axboe spin_lock(&inode_lock); 116966f3b8e2SJens Axboe 117038f21977SNick Piggin /* 117138f21977SNick Piggin * Data integrity sync. Must wait for all pages under writeback, 117238f21977SNick Piggin * because there may have been pages dirtied before our sync 117338f21977SNick Piggin * call, but which had writeout started before we write it out. 117438f21977SNick Piggin * In which case, the inode may not be on the dirty list, but 117538f21977SNick Piggin * we still have to wait for that writeout. 117638f21977SNick Piggin */ 1177b6e51316SJens Axboe list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { 117838f21977SNick Piggin struct address_space *mapping; 117938f21977SNick Piggin 118003ba3782SJens Axboe if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE|I_NEW)) 118138f21977SNick Piggin continue; 118238f21977SNick Piggin mapping = inode->i_mapping; 118338f21977SNick Piggin if (mapping->nrpages == 0) 118438f21977SNick Piggin continue; 118538f21977SNick Piggin __iget(inode); 1186ae8547b0SHans Reiser spin_unlock(&inode_lock); 118738f21977SNick Piggin /* 118838f21977SNick Piggin * We hold a reference to 'inode' so it couldn't have 118938f21977SNick Piggin * been removed from s_inodes list while we dropped the 119038f21977SNick Piggin * inode_lock. We cannot iput the inode now as we can 119138f21977SNick Piggin * be holding the last reference and we cannot iput it 119238f21977SNick Piggin * under inode_lock. So we keep the reference and iput 119338f21977SNick Piggin * it later. 119438f21977SNick Piggin */ 119538f21977SNick Piggin iput(old_inode); 119638f21977SNick Piggin old_inode = inode; 119738f21977SNick Piggin 119838f21977SNick Piggin filemap_fdatawait(mapping); 119938f21977SNick Piggin 120038f21977SNick Piggin cond_resched(); 120138f21977SNick Piggin 120238f21977SNick Piggin spin_lock(&inode_lock); 120338f21977SNick Piggin } 120438f21977SNick Piggin spin_unlock(&inode_lock); 120538f21977SNick Piggin iput(old_inode); 120666f3b8e2SJens Axboe } 12071da177e4SLinus Torvalds 1208d8a8559cSJens Axboe /** 1209d8a8559cSJens Axboe * writeback_inodes_sb - writeback dirty inodes from given super_block 1210d8a8559cSJens Axboe * @sb: the superblock 12111da177e4SLinus Torvalds * 1212d8a8559cSJens Axboe * Start writeback on some inodes on this super_block. No guarantees are made 1213d8a8559cSJens Axboe * on how many (if any) will be written, and this function does not wait 1214d8a8559cSJens Axboe * for IO completion of submitted IO. The number of pages submitted is 1215d8a8559cSJens Axboe * returned. 12161da177e4SLinus Torvalds */ 1217b6e51316SJens Axboe void writeback_inodes_sb(struct super_block *sb) 12181da177e4SLinus Torvalds { 12190e3c9a22SJens Axboe unsigned long nr_dirty = global_page_state(NR_FILE_DIRTY); 12200e3c9a22SJens Axboe unsigned long nr_unstable = global_page_state(NR_UNSTABLE_NFS); 12210e3c9a22SJens Axboe long nr_to_write; 12220e3c9a22SJens Axboe 12230e3c9a22SJens Axboe nr_to_write = nr_dirty + nr_unstable + 12240e3c9a22SJens Axboe (inodes_stat.nr_inodes - inodes_stat.nr_unused); 12250e3c9a22SJens Axboe 12260e3c9a22SJens Axboe bdi_start_writeback(sb->s_bdi, sb, nr_to_write); 12271da177e4SLinus Torvalds } 1228d8a8559cSJens Axboe EXPORT_SYMBOL(writeback_inodes_sb); 1229d8a8559cSJens Axboe 1230d8a8559cSJens Axboe /** 123117bd55d0SEric Sandeen * writeback_inodes_sb_if_idle - start writeback if none underway 123217bd55d0SEric Sandeen * @sb: the superblock 123317bd55d0SEric Sandeen * 123417bd55d0SEric Sandeen * Invoke writeback_inodes_sb if no writeback is currently underway. 123517bd55d0SEric Sandeen * Returns 1 if writeback was started, 0 if not. 123617bd55d0SEric Sandeen */ 123717bd55d0SEric Sandeen int writeback_inodes_sb_if_idle(struct super_block *sb) 123817bd55d0SEric Sandeen { 123917bd55d0SEric Sandeen if (!writeback_in_progress(sb->s_bdi)) { 124017bd55d0SEric Sandeen writeback_inodes_sb(sb); 124117bd55d0SEric Sandeen return 1; 124217bd55d0SEric Sandeen } else 124317bd55d0SEric Sandeen return 0; 124417bd55d0SEric Sandeen } 124517bd55d0SEric Sandeen EXPORT_SYMBOL(writeback_inodes_sb_if_idle); 124617bd55d0SEric Sandeen 124717bd55d0SEric Sandeen /** 1248d8a8559cSJens Axboe * sync_inodes_sb - sync sb inode pages 1249d8a8559cSJens Axboe * @sb: the superblock 1250d8a8559cSJens Axboe * 1251d8a8559cSJens Axboe * This function writes and waits on any dirty inode belonging to this 1252d8a8559cSJens Axboe * super_block. The number of pages synced is returned. 1253d8a8559cSJens Axboe */ 1254b6e51316SJens Axboe void sync_inodes_sb(struct super_block *sb) 1255d8a8559cSJens Axboe { 1256b6e51316SJens Axboe bdi_sync_writeback(sb->s_bdi, sb); 1257b6e51316SJens Axboe wait_sb_inodes(sb); 1258d8a8559cSJens Axboe } 1259d8a8559cSJens Axboe EXPORT_SYMBOL(sync_inodes_sb); 12601da177e4SLinus Torvalds 12611da177e4SLinus Torvalds /** 12621da177e4SLinus Torvalds * write_inode_now - write an inode to disk 12631da177e4SLinus Torvalds * @inode: inode to write to disk 12641da177e4SLinus Torvalds * @sync: whether the write should be synchronous or not 12651da177e4SLinus Torvalds * 12667f04c26dSAndrea Arcangeli * This function commits an inode to disk immediately if it is dirty. This is 12677f04c26dSAndrea Arcangeli * primarily needed by knfsd. 12687f04c26dSAndrea Arcangeli * 12697f04c26dSAndrea Arcangeli * The caller must either have a ref on the inode or must have set I_WILL_FREE. 12701da177e4SLinus Torvalds */ 12711da177e4SLinus Torvalds int write_inode_now(struct inode *inode, int sync) 12721da177e4SLinus Torvalds { 12731da177e4SLinus Torvalds int ret; 12741da177e4SLinus Torvalds struct writeback_control wbc = { 12751da177e4SLinus Torvalds .nr_to_write = LONG_MAX, 127618914b18SMike Galbraith .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, 1277111ebb6eSOGAWA Hirofumi .range_start = 0, 1278111ebb6eSOGAWA Hirofumi .range_end = LLONG_MAX, 12791da177e4SLinus Torvalds }; 12801da177e4SLinus Torvalds 12811da177e4SLinus Torvalds if (!mapping_cap_writeback_dirty(inode->i_mapping)) 128249364ce2SAndrew Morton wbc.nr_to_write = 0; 12831da177e4SLinus Torvalds 12841da177e4SLinus Torvalds might_sleep(); 12851da177e4SLinus Torvalds spin_lock(&inode_lock); 128601c03194SChristoph Hellwig ret = writeback_single_inode(inode, &wbc); 12871da177e4SLinus Torvalds spin_unlock(&inode_lock); 12881da177e4SLinus Torvalds if (sync) 12891c0eeaf5SJoern Engel inode_sync_wait(inode); 12901da177e4SLinus Torvalds return ret; 12911da177e4SLinus Torvalds } 12921da177e4SLinus Torvalds EXPORT_SYMBOL(write_inode_now); 12931da177e4SLinus Torvalds 12941da177e4SLinus Torvalds /** 12951da177e4SLinus Torvalds * sync_inode - write an inode and its pages to disk. 12961da177e4SLinus Torvalds * @inode: the inode to sync 12971da177e4SLinus Torvalds * @wbc: controls the writeback mode 12981da177e4SLinus Torvalds * 12991da177e4SLinus Torvalds * sync_inode() will write an inode and its pages to disk. It will also 13001da177e4SLinus Torvalds * correctly update the inode on its superblock's dirty inode lists and will 13011da177e4SLinus Torvalds * update inode->i_state. 13021da177e4SLinus Torvalds * 13031da177e4SLinus Torvalds * The caller must have a ref on the inode. 13041da177e4SLinus Torvalds */ 13051da177e4SLinus Torvalds int sync_inode(struct inode *inode, struct writeback_control *wbc) 13061da177e4SLinus Torvalds { 13071da177e4SLinus Torvalds int ret; 13081da177e4SLinus Torvalds 13091da177e4SLinus Torvalds spin_lock(&inode_lock); 131001c03194SChristoph Hellwig ret = writeback_single_inode(inode, wbc); 13111da177e4SLinus Torvalds spin_unlock(&inode_lock); 13121da177e4SLinus Torvalds return ret; 13131da177e4SLinus Torvalds } 13141da177e4SLinus Torvalds EXPORT_SYMBOL(sync_inode); 1315