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> 191da177e4SLinus Torvalds #include <linux/sched.h> 201da177e4SLinus Torvalds #include <linux/fs.h> 211da177e4SLinus Torvalds #include <linux/mm.h> 2203ba3782SJens Axboe #include <linux/kthread.h> 2303ba3782SJens Axboe #include <linux/freezer.h> 241da177e4SLinus Torvalds #include <linux/writeback.h> 251da177e4SLinus Torvalds #include <linux/blkdev.h> 261da177e4SLinus Torvalds #include <linux/backing-dev.h> 271da177e4SLinus Torvalds #include <linux/buffer_head.h> 2807f3f05cSDavid Howells #include "internal.h" 291da177e4SLinus Torvalds 3066f3b8e2SJens Axboe #define inode_to_bdi(inode) ((inode)->i_mapping->backing_dev_info) 31f11b00f3SAdrian Bunk 3203ba3782SJens Axboe /* 33d0bceac7SJens Axboe * We don't actually have pdflush, but this one is exported though /proc... 34d0bceac7SJens Axboe */ 35d0bceac7SJens Axboe int nr_pdflush_threads; 36d0bceac7SJens Axboe 37d0bceac7SJens Axboe /* 38c4a77a6cSJens Axboe * Passed into wb_writeback(), essentially a subset of writeback_control 39c4a77a6cSJens Axboe */ 40c4a77a6cSJens Axboe struct wb_writeback_args { 41c4a77a6cSJens Axboe long nr_pages; 42c4a77a6cSJens Axboe struct super_block *sb; 43c4a77a6cSJens Axboe enum writeback_sync_modes sync_mode; 44d3ddec76SWu Fengguang int for_kupdate:1; 45d3ddec76SWu Fengguang int range_cyclic:1; 46d3ddec76SWu Fengguang int for_background:1; 47c4a77a6cSJens Axboe }; 48c4a77a6cSJens Axboe 49c4a77a6cSJens Axboe /* 5003ba3782SJens Axboe * Work items for the bdi_writeback threads 51f11b00f3SAdrian Bunk */ 5203ba3782SJens Axboe struct bdi_work { 538010c3b6SJens Axboe struct list_head list; /* pending work list */ 548010c3b6SJens Axboe struct rcu_head rcu_head; /* for RCU free/clear of work */ 5503ba3782SJens Axboe 568010c3b6SJens Axboe unsigned long seen; /* threads that have seen this work */ 578010c3b6SJens Axboe atomic_t pending; /* number of threads still to do work */ 5803ba3782SJens Axboe 598010c3b6SJens Axboe struct wb_writeback_args args; /* writeback arguments */ 6003ba3782SJens Axboe 618010c3b6SJens Axboe unsigned long state; /* flag bits, see WS_* */ 6203ba3782SJens Axboe }; 6303ba3782SJens Axboe 6403ba3782SJens Axboe enum { 6503ba3782SJens Axboe WS_USED_B = 0, 6603ba3782SJens Axboe WS_ONSTACK_B, 6703ba3782SJens Axboe }; 6803ba3782SJens Axboe 6903ba3782SJens Axboe #define WS_USED (1 << WS_USED_B) 7003ba3782SJens Axboe #define WS_ONSTACK (1 << WS_ONSTACK_B) 7103ba3782SJens Axboe 7203ba3782SJens Axboe static inline bool bdi_work_on_stack(struct bdi_work *work) 73f11b00f3SAdrian Bunk { 7403ba3782SJens Axboe return test_bit(WS_ONSTACK_B, &work->state); 7503ba3782SJens Axboe } 7603ba3782SJens Axboe 7703ba3782SJens Axboe static inline void bdi_work_init(struct bdi_work *work, 78b6e51316SJens Axboe struct wb_writeback_args *args) 7903ba3782SJens Axboe { 8003ba3782SJens Axboe INIT_RCU_HEAD(&work->rcu_head); 81b6e51316SJens Axboe work->args = *args; 8203ba3782SJens Axboe work->state = WS_USED; 8303ba3782SJens Axboe } 8403ba3782SJens Axboe 85f11b00f3SAdrian Bunk /** 86f11b00f3SAdrian Bunk * writeback_in_progress - determine whether there is writeback in progress 87f11b00f3SAdrian Bunk * @bdi: the device's backing_dev_info structure. 88f11b00f3SAdrian Bunk * 8903ba3782SJens Axboe * Determine whether there is writeback waiting to be handled against a 9003ba3782SJens Axboe * backing device. 91f11b00f3SAdrian Bunk */ 92f11b00f3SAdrian Bunk int writeback_in_progress(struct backing_dev_info *bdi) 93f11b00f3SAdrian Bunk { 9403ba3782SJens Axboe return !list_empty(&bdi->work_list); 95f11b00f3SAdrian Bunk } 96f11b00f3SAdrian Bunk 9703ba3782SJens Axboe static void bdi_work_clear(struct bdi_work *work) 98f11b00f3SAdrian Bunk { 9903ba3782SJens Axboe clear_bit(WS_USED_B, &work->state); 10003ba3782SJens Axboe smp_mb__after_clear_bit(); 1011ef7d9aaSNick Piggin /* 1021ef7d9aaSNick Piggin * work can have disappeared at this point. bit waitq functions 1031ef7d9aaSNick Piggin * should be able to tolerate this, provided bdi_sched_wait does 1041ef7d9aaSNick Piggin * not dereference it's pointer argument. 1051ef7d9aaSNick Piggin */ 10603ba3782SJens Axboe wake_up_bit(&work->state, WS_USED_B); 107f11b00f3SAdrian Bunk } 108f11b00f3SAdrian Bunk 10903ba3782SJens Axboe static void bdi_work_free(struct rcu_head *head) 1104195f73dSNick Piggin { 11103ba3782SJens Axboe struct bdi_work *work = container_of(head, struct bdi_work, rcu_head); 1124195f73dSNick Piggin 11303ba3782SJens Axboe if (!bdi_work_on_stack(work)) 11403ba3782SJens Axboe kfree(work); 11503ba3782SJens Axboe else 11603ba3782SJens Axboe bdi_work_clear(work); 1174195f73dSNick Piggin } 1184195f73dSNick Piggin 11903ba3782SJens Axboe static void wb_work_complete(struct bdi_work *work) 1201da177e4SLinus Torvalds { 121c4a77a6cSJens Axboe const enum writeback_sync_modes sync_mode = work->args.sync_mode; 12277b9d059SNick Piggin int onstack = bdi_work_on_stack(work); 1231da177e4SLinus Torvalds 1241da177e4SLinus Torvalds /* 12503ba3782SJens Axboe * For allocated work, we can clear the done/seen bit right here. 12603ba3782SJens Axboe * For on-stack work, we need to postpone both the clear and free 12703ba3782SJens Axboe * to after the RCU grace period, since the stack could be invalidated 12803ba3782SJens Axboe * as soon as bdi_work_clear() has done the wakeup. 1291da177e4SLinus Torvalds */ 13077b9d059SNick Piggin if (!onstack) 13103ba3782SJens Axboe bdi_work_clear(work); 13277b9d059SNick Piggin if (sync_mode == WB_SYNC_NONE || onstack) 13303ba3782SJens Axboe call_rcu(&work->rcu_head, bdi_work_free); 1341da177e4SLinus Torvalds } 1351da177e4SLinus Torvalds 13603ba3782SJens Axboe static void wb_clear_pending(struct bdi_writeback *wb, struct bdi_work *work) 13703ba3782SJens Axboe { 1381da177e4SLinus Torvalds /* 13903ba3782SJens Axboe * The caller has retrieved the work arguments from this work, 14003ba3782SJens Axboe * drop our reference. If this is the last ref, delete and free it 14103ba3782SJens Axboe */ 14203ba3782SJens Axboe if (atomic_dec_and_test(&work->pending)) { 14303ba3782SJens Axboe struct backing_dev_info *bdi = wb->bdi; 14403ba3782SJens Axboe 14503ba3782SJens Axboe spin_lock(&bdi->wb_lock); 14603ba3782SJens Axboe list_del_rcu(&work->list); 14703ba3782SJens Axboe spin_unlock(&bdi->wb_lock); 14803ba3782SJens Axboe 14903ba3782SJens Axboe wb_work_complete(work); 15003ba3782SJens Axboe } 15103ba3782SJens Axboe } 15203ba3782SJens Axboe 15303ba3782SJens Axboe static void bdi_queue_work(struct backing_dev_info *bdi, struct bdi_work *work) 15403ba3782SJens Axboe { 15503ba3782SJens Axboe work->seen = bdi->wb_mask; 15603ba3782SJens Axboe BUG_ON(!work->seen); 15703ba3782SJens Axboe atomic_set(&work->pending, bdi->wb_cnt); 15803ba3782SJens Axboe BUG_ON(!bdi->wb_cnt); 15903ba3782SJens Axboe 16003ba3782SJens Axboe /* 161deed62edSNick Piggin * list_add_tail_rcu() contains the necessary barriers to 162deed62edSNick Piggin * make sure the above stores are seen before the item is 163deed62edSNick Piggin * noticed on the list 1641da177e4SLinus Torvalds */ 16503ba3782SJens Axboe spin_lock(&bdi->wb_lock); 16603ba3782SJens Axboe list_add_tail_rcu(&work->list, &bdi->work_list); 16703ba3782SJens Axboe spin_unlock(&bdi->wb_lock); 1681da177e4SLinus Torvalds 1691da177e4SLinus Torvalds /* 17003ba3782SJens Axboe * If the default thread isn't there, make sure we add it. When 17103ba3782SJens Axboe * it gets created and wakes up, we'll run this work. 1721da177e4SLinus Torvalds */ 17303ba3782SJens Axboe if (unlikely(list_empty_careful(&bdi->wb_list))) 17403ba3782SJens Axboe wake_up_process(default_backing_dev_info.wb.task); 17503ba3782SJens Axboe else { 17603ba3782SJens Axboe struct bdi_writeback *wb = &bdi->wb; 1771da177e4SLinus Torvalds 1781ef7d9aaSNick Piggin if (wb->task) 17903ba3782SJens Axboe wake_up_process(wb->task); 1801da177e4SLinus Torvalds } 18103ba3782SJens Axboe } 1821da177e4SLinus Torvalds 1831da177e4SLinus Torvalds /* 18403ba3782SJens Axboe * Used for on-stack allocated work items. The caller needs to wait until 18503ba3782SJens Axboe * the wb threads have acked the work before it's safe to continue. 1861da177e4SLinus Torvalds */ 18703ba3782SJens Axboe static void bdi_wait_on_work_clear(struct bdi_work *work) 1881da177e4SLinus Torvalds { 18903ba3782SJens Axboe wait_on_bit(&work->state, WS_USED_B, bdi_sched_wait, 19003ba3782SJens Axboe TASK_UNINTERRUPTIBLE); 19103ba3782SJens Axboe } 19203ba3782SJens Axboe 193f11fcae8SJens Axboe static void bdi_alloc_queue_work(struct backing_dev_info *bdi, 194b6e51316SJens Axboe struct wb_writeback_args *args) 19503ba3782SJens Axboe { 19603ba3782SJens Axboe struct bdi_work *work; 19703ba3782SJens Axboe 198bcddc3f0SJens Axboe /* 199bcddc3f0SJens Axboe * This is WB_SYNC_NONE writeback, so if allocation fails just 200bcddc3f0SJens Axboe * wakeup the thread for old dirty data writeback 201bcddc3f0SJens Axboe */ 20203ba3782SJens Axboe work = kmalloc(sizeof(*work), GFP_ATOMIC); 203bcddc3f0SJens Axboe if (work) { 204b6e51316SJens Axboe bdi_work_init(work, args); 205f11fcae8SJens Axboe bdi_queue_work(bdi, work); 206bcddc3f0SJens Axboe } else { 207bcddc3f0SJens Axboe struct bdi_writeback *wb = &bdi->wb; 208bcddc3f0SJens Axboe 209bcddc3f0SJens Axboe if (wb->task) 210bcddc3f0SJens Axboe wake_up_process(wb->task); 211bcddc3f0SJens Axboe } 21203ba3782SJens Axboe } 21303ba3782SJens Axboe 214b6e51316SJens Axboe /** 215b6e51316SJens Axboe * bdi_sync_writeback - start and wait for writeback 216b6e51316SJens Axboe * @bdi: the backing device to write from 217b6e51316SJens Axboe * @sb: write inodes from this super_block 218b6e51316SJens Axboe * 219b6e51316SJens Axboe * Description: 220b6e51316SJens Axboe * This does WB_SYNC_ALL data integrity writeback and waits for the 221b6e51316SJens Axboe * IO to complete. Callers must hold the sb s_umount semaphore for 222b6e51316SJens Axboe * reading, to avoid having the super disappear before we are done. 22303ba3782SJens Axboe */ 224b6e51316SJens Axboe static void bdi_sync_writeback(struct backing_dev_info *bdi, 225b6e51316SJens Axboe struct super_block *sb) 226b6e51316SJens Axboe { 227b6e51316SJens Axboe struct wb_writeback_args args = { 228b6e51316SJens Axboe .sb = sb, 229b6e51316SJens Axboe .sync_mode = WB_SYNC_ALL, 230b6e51316SJens Axboe .nr_pages = LONG_MAX, 231b6e51316SJens Axboe .range_cyclic = 0, 232b6e51316SJens Axboe }; 233f0fad8a5SChristoph Hellwig struct bdi_work work; 234f0fad8a5SChristoph Hellwig 235b6e51316SJens Axboe bdi_work_init(&work, &args); 236f0fad8a5SChristoph Hellwig work.state |= WS_ONSTACK; 237f0fad8a5SChristoph Hellwig 238b6e51316SJens Axboe bdi_queue_work(bdi, &work); 239f0fad8a5SChristoph Hellwig bdi_wait_on_work_clear(&work); 24003ba3782SJens Axboe } 241b6e51316SJens Axboe 242b6e51316SJens Axboe /** 243b6e51316SJens Axboe * bdi_start_writeback - start writeback 244b6e51316SJens Axboe * @bdi: the backing device to write from 245b6e51316SJens Axboe * @nr_pages: the number of pages to write 246b6e51316SJens Axboe * 247b6e51316SJens Axboe * Description: 248b6e51316SJens Axboe * This does WB_SYNC_NONE opportunistic writeback. The IO is only 249b6e51316SJens Axboe * started when this function returns, we make no guarentees on 250b6e51316SJens Axboe * completion. Caller need not hold sb s_umount semaphore. 251b6e51316SJens Axboe * 252b6e51316SJens Axboe */ 253b6e51316SJens Axboe void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages) 254b6e51316SJens Axboe { 255b6e51316SJens Axboe struct wb_writeback_args args = { 256b6e51316SJens Axboe .sync_mode = WB_SYNC_NONE, 257b6e51316SJens Axboe .nr_pages = nr_pages, 258b6e51316SJens Axboe .range_cyclic = 1, 259b6e51316SJens Axboe }; 260b6e51316SJens Axboe 261d3ddec76SWu Fengguang /* 262d3ddec76SWu Fengguang * We treat @nr_pages=0 as the special case to do background writeback, 263d3ddec76SWu Fengguang * ie. to sync pages until the background dirty threshold is reached. 264d3ddec76SWu Fengguang */ 265d3ddec76SWu Fengguang if (!nr_pages) { 266d3ddec76SWu Fengguang args.nr_pages = LONG_MAX; 267d3ddec76SWu Fengguang args.for_background = 1; 268d3ddec76SWu Fengguang } 269d3ddec76SWu Fengguang 270b6e51316SJens Axboe bdi_alloc_queue_work(bdi, &args); 2711da177e4SLinus Torvalds } 2721da177e4SLinus Torvalds 2731da177e4SLinus Torvalds /* 2746610a0bcSAndrew Morton * Redirty an inode: set its when-it-was dirtied timestamp and move it to the 2756610a0bcSAndrew Morton * furthest end of its superblock's dirty-inode list. 2766610a0bcSAndrew Morton * 2776610a0bcSAndrew Morton * Before stamping the inode's ->dirtied_when, we check to see whether it is 27866f3b8e2SJens Axboe * already the most-recently-dirtied inode on the b_dirty list. If that is 2796610a0bcSAndrew Morton * the case then the inode must have been redirtied while it was being written 2806610a0bcSAndrew Morton * out and we don't reset its dirtied_when. 2816610a0bcSAndrew Morton */ 2826610a0bcSAndrew Morton static void redirty_tail(struct inode *inode) 2836610a0bcSAndrew Morton { 28403ba3782SJens Axboe struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; 2856610a0bcSAndrew Morton 28603ba3782SJens Axboe if (!list_empty(&wb->b_dirty)) { 28766f3b8e2SJens Axboe struct inode *tail; 2886610a0bcSAndrew Morton 28903ba3782SJens Axboe tail = list_entry(wb->b_dirty.next, struct inode, i_list); 29066f3b8e2SJens Axboe if (time_before(inode->dirtied_when, tail->dirtied_when)) 2916610a0bcSAndrew Morton inode->dirtied_when = jiffies; 2926610a0bcSAndrew Morton } 29303ba3782SJens Axboe list_move(&inode->i_list, &wb->b_dirty); 2946610a0bcSAndrew Morton } 2956610a0bcSAndrew Morton 2966610a0bcSAndrew Morton /* 29766f3b8e2SJens Axboe * requeue inode for re-scanning after bdi->b_io list is exhausted. 298c986d1e2SAndrew Morton */ 2990e0f4fc2SKen Chen static void requeue_io(struct inode *inode) 300c986d1e2SAndrew Morton { 30103ba3782SJens Axboe struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; 30203ba3782SJens Axboe 30303ba3782SJens Axboe list_move(&inode->i_list, &wb->b_more_io); 304c986d1e2SAndrew Morton } 305c986d1e2SAndrew Morton 3061c0eeaf5SJoern Engel static void inode_sync_complete(struct inode *inode) 3071c0eeaf5SJoern Engel { 3081c0eeaf5SJoern Engel /* 3091c0eeaf5SJoern Engel * Prevent speculative execution through spin_unlock(&inode_lock); 3101c0eeaf5SJoern Engel */ 3111c0eeaf5SJoern Engel smp_mb(); 3121c0eeaf5SJoern Engel wake_up_bit(&inode->i_state, __I_SYNC); 3131c0eeaf5SJoern Engel } 3141c0eeaf5SJoern Engel 315d2caa3c5SJeff Layton static bool inode_dirtied_after(struct inode *inode, unsigned long t) 316d2caa3c5SJeff Layton { 317d2caa3c5SJeff Layton bool ret = time_after(inode->dirtied_when, t); 318d2caa3c5SJeff Layton #ifndef CONFIG_64BIT 319d2caa3c5SJeff Layton /* 320d2caa3c5SJeff Layton * For inodes being constantly redirtied, dirtied_when can get stuck. 321d2caa3c5SJeff Layton * It _appears_ to be in the future, but is actually in distant past. 322d2caa3c5SJeff Layton * This test is necessary to prevent such wrapped-around relative times 3235b0830cbSJens Axboe * from permanently stopping the whole bdi writeback. 324d2caa3c5SJeff Layton */ 325d2caa3c5SJeff Layton ret = ret && time_before_eq(inode->dirtied_when, jiffies); 326d2caa3c5SJeff Layton #endif 327d2caa3c5SJeff Layton return ret; 328d2caa3c5SJeff Layton } 329d2caa3c5SJeff Layton 330c986d1e2SAndrew Morton /* 3312c136579SFengguang Wu * Move expired dirty inodes from @delaying_queue to @dispatch_queue. 3322c136579SFengguang Wu */ 3332c136579SFengguang Wu static void move_expired_inodes(struct list_head *delaying_queue, 3342c136579SFengguang Wu struct list_head *dispatch_queue, 3352c136579SFengguang Wu unsigned long *older_than_this) 3362c136579SFengguang Wu { 3375c03449dSShaohua Li LIST_HEAD(tmp); 3385c03449dSShaohua Li struct list_head *pos, *node; 339cf137307SJens Axboe struct super_block *sb = NULL; 3405c03449dSShaohua Li struct inode *inode; 341cf137307SJens Axboe int do_sb_sort = 0; 3425c03449dSShaohua Li 3432c136579SFengguang Wu while (!list_empty(delaying_queue)) { 3445c03449dSShaohua Li inode = list_entry(delaying_queue->prev, struct inode, i_list); 3452c136579SFengguang Wu if (older_than_this && 346d2caa3c5SJeff Layton inode_dirtied_after(inode, *older_than_this)) 3472c136579SFengguang Wu break; 348cf137307SJens Axboe if (sb && sb != inode->i_sb) 349cf137307SJens Axboe do_sb_sort = 1; 350cf137307SJens Axboe sb = inode->i_sb; 3515c03449dSShaohua Li list_move(&inode->i_list, &tmp); 3525c03449dSShaohua Li } 3535c03449dSShaohua Li 354cf137307SJens Axboe /* just one sb in list, splice to dispatch_queue and we're done */ 355cf137307SJens Axboe if (!do_sb_sort) { 356cf137307SJens Axboe list_splice(&tmp, dispatch_queue); 357cf137307SJens Axboe return; 358cf137307SJens Axboe } 359cf137307SJens Axboe 3605c03449dSShaohua Li /* Move inodes from one superblock together */ 3615c03449dSShaohua Li while (!list_empty(&tmp)) { 3625c03449dSShaohua Li inode = list_entry(tmp.prev, struct inode, i_list); 3635c03449dSShaohua Li sb = inode->i_sb; 3645c03449dSShaohua Li list_for_each_prev_safe(pos, node, &tmp) { 3655c03449dSShaohua Li inode = list_entry(pos, struct inode, i_list); 3665c03449dSShaohua Li if (inode->i_sb == sb) 3672c136579SFengguang Wu list_move(&inode->i_list, dispatch_queue); 3682c136579SFengguang Wu } 3692c136579SFengguang Wu } 3705c03449dSShaohua Li } 3712c136579SFengguang Wu 3722c136579SFengguang Wu /* 3732c136579SFengguang Wu * Queue all expired dirty inodes for io, eldest first. 3742c136579SFengguang Wu */ 37503ba3782SJens Axboe static void queue_io(struct bdi_writeback *wb, unsigned long *older_than_this) 3762c136579SFengguang Wu { 37703ba3782SJens Axboe list_splice_init(&wb->b_more_io, wb->b_io.prev); 37803ba3782SJens Axboe move_expired_inodes(&wb->b_dirty, &wb->b_io, older_than_this); 37966f3b8e2SJens Axboe } 38066f3b8e2SJens Axboe 38103ba3782SJens Axboe static int write_inode(struct inode *inode, int sync) 38266f3b8e2SJens Axboe { 38303ba3782SJens Axboe if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) 38403ba3782SJens Axboe return inode->i_sb->s_op->write_inode(inode, sync); 38503ba3782SJens Axboe return 0; 38666f3b8e2SJens Axboe } 38708d8e974SFengguang Wu 3882c136579SFengguang Wu /* 38901c03194SChristoph Hellwig * Wait for writeback on an inode to complete. 39001c03194SChristoph Hellwig */ 39101c03194SChristoph Hellwig static void inode_wait_for_writeback(struct inode *inode) 39201c03194SChristoph Hellwig { 39301c03194SChristoph Hellwig DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); 39401c03194SChristoph Hellwig wait_queue_head_t *wqh; 39501c03194SChristoph Hellwig 39601c03194SChristoph Hellwig wqh = bit_waitqueue(&inode->i_state, __I_SYNC); 39701c03194SChristoph Hellwig do { 39801c03194SChristoph Hellwig spin_unlock(&inode_lock); 39901c03194SChristoph Hellwig __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE); 40001c03194SChristoph Hellwig spin_lock(&inode_lock); 40101c03194SChristoph Hellwig } while (inode->i_state & I_SYNC); 40201c03194SChristoph Hellwig } 40301c03194SChristoph Hellwig 40401c03194SChristoph Hellwig /* 40501c03194SChristoph Hellwig * Write out an inode's dirty pages. Called under inode_lock. Either the 40601c03194SChristoph Hellwig * caller has ref on the inode (either via __iget or via syscall against an fd) 40701c03194SChristoph Hellwig * or the inode has I_WILL_FREE set (via generic_forget_inode) 40801c03194SChristoph Hellwig * 4091da177e4SLinus Torvalds * If `wait' is set, wait on the writeout. 4101da177e4SLinus Torvalds * 4111da177e4SLinus Torvalds * The whole writeout design is quite complex and fragile. We want to avoid 4121da177e4SLinus Torvalds * starvation of particular inodes when others are being redirtied, prevent 4131da177e4SLinus Torvalds * livelocks, etc. 4141da177e4SLinus Torvalds * 4151da177e4SLinus Torvalds * Called under inode_lock. 4161da177e4SLinus Torvalds */ 4171da177e4SLinus Torvalds static int 41801c03194SChristoph Hellwig writeback_single_inode(struct inode *inode, struct writeback_control *wbc) 4191da177e4SLinus Torvalds { 4201da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 4211da177e4SLinus Torvalds int wait = wbc->sync_mode == WB_SYNC_ALL; 42201c03194SChristoph Hellwig unsigned dirty; 4231da177e4SLinus Torvalds int ret; 4241da177e4SLinus Torvalds 42501c03194SChristoph Hellwig if (!atomic_read(&inode->i_count)) 42601c03194SChristoph Hellwig WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); 42701c03194SChristoph Hellwig else 42801c03194SChristoph Hellwig WARN_ON(inode->i_state & I_WILL_FREE); 42901c03194SChristoph Hellwig 43001c03194SChristoph Hellwig if (inode->i_state & I_SYNC) { 43101c03194SChristoph Hellwig /* 43201c03194SChristoph Hellwig * If this inode is locked for writeback and we are not doing 43366f3b8e2SJens Axboe * writeback-for-data-integrity, move it to b_more_io so that 43401c03194SChristoph Hellwig * writeback can proceed with the other inodes on s_io. 43501c03194SChristoph Hellwig * 43601c03194SChristoph Hellwig * We'll have another go at writing back this inode when we 43766f3b8e2SJens Axboe * completed a full scan of b_io. 43801c03194SChristoph Hellwig */ 43901c03194SChristoph Hellwig if (!wait) { 44001c03194SChristoph Hellwig requeue_io(inode); 44101c03194SChristoph Hellwig return 0; 44201c03194SChristoph Hellwig } 44301c03194SChristoph Hellwig 44401c03194SChristoph Hellwig /* 44501c03194SChristoph Hellwig * It's a data-integrity sync. We must wait. 44601c03194SChristoph Hellwig */ 44701c03194SChristoph Hellwig inode_wait_for_writeback(inode); 44801c03194SChristoph Hellwig } 44901c03194SChristoph Hellwig 4501c0eeaf5SJoern Engel BUG_ON(inode->i_state & I_SYNC); 4511da177e4SLinus Torvalds 4521c0eeaf5SJoern Engel /* Set I_SYNC, reset I_DIRTY */ 4531da177e4SLinus Torvalds dirty = inode->i_state & I_DIRTY; 4541c0eeaf5SJoern Engel inode->i_state |= I_SYNC; 4551da177e4SLinus Torvalds inode->i_state &= ~I_DIRTY; 4561da177e4SLinus Torvalds 4571da177e4SLinus Torvalds spin_unlock(&inode_lock); 4581da177e4SLinus Torvalds 4591da177e4SLinus Torvalds ret = do_writepages(mapping, wbc); 4601da177e4SLinus Torvalds 4611da177e4SLinus Torvalds /* Don't write the inode if only I_DIRTY_PAGES was set */ 4621da177e4SLinus Torvalds if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { 4631da177e4SLinus Torvalds int err = write_inode(inode, wait); 4641da177e4SLinus Torvalds if (ret == 0) 4651da177e4SLinus Torvalds ret = err; 4661da177e4SLinus Torvalds } 4671da177e4SLinus Torvalds 4681da177e4SLinus Torvalds if (wait) { 4691da177e4SLinus Torvalds int err = filemap_fdatawait(mapping); 4701da177e4SLinus Torvalds if (ret == 0) 4711da177e4SLinus Torvalds ret = err; 4721da177e4SLinus Torvalds } 4731da177e4SLinus Torvalds 4741da177e4SLinus Torvalds spin_lock(&inode_lock); 4751c0eeaf5SJoern Engel inode->i_state &= ~I_SYNC; 47684a89245SWu Fengguang if (!(inode->i_state & (I_FREEING | I_CLEAR))) { 477ae1b7f7dSWu Fengguang if (inode->i_state & I_DIRTY) { 478ae1b7f7dSWu Fengguang /* 479ae1b7f7dSWu Fengguang * Someone redirtied the inode while were writing back 480ae1b7f7dSWu Fengguang * the pages. 481ae1b7f7dSWu Fengguang */ 482ae1b7f7dSWu Fengguang redirty_tail(inode); 483ae1b7f7dSWu Fengguang } else if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { 4841da177e4SLinus Torvalds /* 4851da177e4SLinus Torvalds * We didn't write back all the pages. nfs_writepages() 4861da177e4SLinus Torvalds * sometimes bales out without doing anything. Redirty 48766f3b8e2SJens Axboe * the inode; Move it from b_io onto b_more_io/b_dirty. 4881b43ef91SAndrew Morton */ 4891b43ef91SAndrew Morton /* 4901b43ef91SAndrew Morton * akpm: if the caller was the kupdate function we put 49166f3b8e2SJens Axboe * this inode at the head of b_dirty so it gets first 4921b43ef91SAndrew Morton * consideration. Otherwise, move it to the tail, for 4931b43ef91SAndrew Morton * the reasons described there. I'm not really sure 4941b43ef91SAndrew Morton * how much sense this makes. Presumably I had a good 4951b43ef91SAndrew Morton * reasons for doing it this way, and I'd rather not 4961b43ef91SAndrew Morton * muck with it at present. 4971da177e4SLinus Torvalds */ 4981da177e4SLinus Torvalds if (wbc->for_kupdate) { 4991da177e4SLinus Torvalds /* 5002c136579SFengguang Wu * For the kupdate function we move the inode 50166f3b8e2SJens Axboe * to b_more_io so it will get more writeout as 5022c136579SFengguang Wu * soon as the queue becomes uncongested. 5031da177e4SLinus Torvalds */ 5041da177e4SLinus Torvalds inode->i_state |= I_DIRTY_PAGES; 5058bc3be27SFengguang Wu if (wbc->nr_to_write <= 0) { 5068bc3be27SFengguang Wu /* 5078bc3be27SFengguang Wu * slice used up: queue for next turn 5088bc3be27SFengguang Wu */ 5090e0f4fc2SKen Chen requeue_io(inode); 5101da177e4SLinus Torvalds } else { 5111da177e4SLinus Torvalds /* 5128bc3be27SFengguang Wu * somehow blocked: retry later 5138bc3be27SFengguang Wu */ 5148bc3be27SFengguang Wu redirty_tail(inode); 5158bc3be27SFengguang Wu } 5168bc3be27SFengguang Wu } else { 5178bc3be27SFengguang Wu /* 5181da177e4SLinus Torvalds * Otherwise fully redirty the inode so that 5191da177e4SLinus Torvalds * other inodes on this superblock will get some 5201da177e4SLinus Torvalds * writeout. Otherwise heavy writing to one 5211da177e4SLinus Torvalds * file would indefinitely suspend writeout of 5221da177e4SLinus Torvalds * all the other files. 5231da177e4SLinus Torvalds */ 5241da177e4SLinus Torvalds inode->i_state |= I_DIRTY_PAGES; 5251b43ef91SAndrew Morton redirty_tail(inode); 5261da177e4SLinus Torvalds } 5271da177e4SLinus Torvalds } else if (atomic_read(&inode->i_count)) { 5281da177e4SLinus Torvalds /* 5291da177e4SLinus Torvalds * The inode is clean, inuse 5301da177e4SLinus Torvalds */ 5311da177e4SLinus Torvalds list_move(&inode->i_list, &inode_in_use); 5321da177e4SLinus Torvalds } else { 5331da177e4SLinus Torvalds /* 5341da177e4SLinus Torvalds * The inode is clean, unused 5351da177e4SLinus Torvalds */ 5361da177e4SLinus Torvalds list_move(&inode->i_list, &inode_unused); 5371da177e4SLinus Torvalds } 5381da177e4SLinus Torvalds } 5391c0eeaf5SJoern Engel inode_sync_complete(inode); 5401da177e4SLinus Torvalds return ret; 5411da177e4SLinus Torvalds } 5421da177e4SLinus Torvalds 54303ba3782SJens Axboe /* 54403ba3782SJens Axboe * For WB_SYNC_NONE writeback, the caller does not have the sb pinned 54503ba3782SJens Axboe * before calling writeback. So make sure that we do pin it, so it doesn't 54603ba3782SJens Axboe * go away while we are writing inodes from it. 54703ba3782SJens Axboe * 54803ba3782SJens Axboe * Returns 0 if the super was successfully pinned (or pinning wasn't needed), 54903ba3782SJens Axboe * 1 if we failed. 55003ba3782SJens Axboe */ 55103ba3782SJens Axboe static int pin_sb_for_writeback(struct writeback_control *wbc, 55203ba3782SJens Axboe struct inode *inode) 5531da177e4SLinus Torvalds { 55403ba3782SJens Axboe struct super_block *sb = inode->i_sb; 55503ba3782SJens Axboe 55603ba3782SJens Axboe /* 55703ba3782SJens Axboe * Caller must already hold the ref for this 55803ba3782SJens Axboe */ 55903ba3782SJens Axboe if (wbc->sync_mode == WB_SYNC_ALL) { 56003ba3782SJens Axboe WARN_ON(!rwsem_is_locked(&sb->s_umount)); 56103ba3782SJens Axboe return 0; 56203ba3782SJens Axboe } 56303ba3782SJens Axboe 56403ba3782SJens Axboe spin_lock(&sb_lock); 56503ba3782SJens Axboe sb->s_count++; 56603ba3782SJens Axboe if (down_read_trylock(&sb->s_umount)) { 56703ba3782SJens Axboe if (sb->s_root) { 56803ba3782SJens Axboe spin_unlock(&sb_lock); 56903ba3782SJens Axboe return 0; 57003ba3782SJens Axboe } 57103ba3782SJens Axboe /* 57203ba3782SJens Axboe * umounted, drop rwsem again and fall through to failure 57303ba3782SJens Axboe */ 57403ba3782SJens Axboe up_read(&sb->s_umount); 57503ba3782SJens Axboe } 57603ba3782SJens Axboe 57703ba3782SJens Axboe sb->s_count--; 57803ba3782SJens Axboe spin_unlock(&sb_lock); 57903ba3782SJens Axboe return 1; 58003ba3782SJens Axboe } 58103ba3782SJens Axboe 58203ba3782SJens Axboe static void unpin_sb_for_writeback(struct writeback_control *wbc, 58303ba3782SJens Axboe struct inode *inode) 58403ba3782SJens Axboe { 58503ba3782SJens Axboe struct super_block *sb = inode->i_sb; 58603ba3782SJens Axboe 58703ba3782SJens Axboe if (wbc->sync_mode == WB_SYNC_ALL) 58803ba3782SJens Axboe return; 58903ba3782SJens Axboe 59003ba3782SJens Axboe up_read(&sb->s_umount); 59103ba3782SJens Axboe put_super(sb); 59203ba3782SJens Axboe } 59303ba3782SJens Axboe 59403ba3782SJens Axboe static void writeback_inodes_wb(struct bdi_writeback *wb, 59503ba3782SJens Axboe struct writeback_control *wbc) 59603ba3782SJens Axboe { 59703ba3782SJens Axboe struct super_block *sb = wbc->sb; 59866f3b8e2SJens Axboe const int is_blkdev_sb = sb_is_blkdev_sb(sb); 5991da177e4SLinus Torvalds const unsigned long start = jiffies; /* livelock avoidance */ 6001da177e4SLinus Torvalds 601ae8547b0SHans Reiser spin_lock(&inode_lock); 6021da177e4SLinus Torvalds 60303ba3782SJens Axboe if (!wbc->for_kupdate || list_empty(&wb->b_io)) 60403ba3782SJens Axboe queue_io(wb, wbc->older_than_this); 60566f3b8e2SJens Axboe 60603ba3782SJens Axboe while (!list_empty(&wb->b_io)) { 60703ba3782SJens Axboe struct inode *inode = list_entry(wb->b_io.prev, 6081da177e4SLinus Torvalds struct inode, i_list); 6091da177e4SLinus Torvalds long pages_skipped; 6101da177e4SLinus Torvalds 61166f3b8e2SJens Axboe /* 61266f3b8e2SJens Axboe * super block given and doesn't match, skip this inode 61366f3b8e2SJens Axboe */ 61466f3b8e2SJens Axboe if (sb && sb != inode->i_sb) { 61566f3b8e2SJens Axboe redirty_tail(inode); 61666f3b8e2SJens Axboe continue; 61766f3b8e2SJens Axboe } 61866f3b8e2SJens Axboe 61903ba3782SJens Axboe if (!bdi_cap_writeback_dirty(wb->bdi)) { 6209852a0e7SAndrew Morton redirty_tail(inode); 62166f3b8e2SJens Axboe if (is_blkdev_sb) { 6221da177e4SLinus Torvalds /* 6231da177e4SLinus Torvalds * Dirty memory-backed blockdev: the ramdisk 6241da177e4SLinus Torvalds * driver does this. Skip just this inode 6251da177e4SLinus Torvalds */ 6261da177e4SLinus Torvalds continue; 6271da177e4SLinus Torvalds } 6281da177e4SLinus Torvalds /* 6291da177e4SLinus Torvalds * Dirty memory-backed inode against a filesystem other 6301da177e4SLinus Torvalds * than the kernel-internal bdev filesystem. Skip the 6311da177e4SLinus Torvalds * entire superblock. 6321da177e4SLinus Torvalds */ 6331da177e4SLinus Torvalds break; 6341da177e4SLinus Torvalds } 6351da177e4SLinus Torvalds 63684a89245SWu Fengguang if (inode->i_state & (I_NEW | I_WILL_FREE)) { 6377ef0d737SNick Piggin requeue_io(inode); 6387ef0d737SNick Piggin continue; 6397ef0d737SNick Piggin } 6407ef0d737SNick Piggin 64103ba3782SJens Axboe if (wbc->nonblocking && bdi_write_congested(wb->bdi)) { 6421da177e4SLinus Torvalds wbc->encountered_congestion = 1; 64366f3b8e2SJens Axboe if (!is_blkdev_sb) 6441da177e4SLinus Torvalds break; /* Skip a congested fs */ 6450e0f4fc2SKen Chen requeue_io(inode); 6461da177e4SLinus Torvalds continue; /* Skip a congested blockdev */ 6471da177e4SLinus Torvalds } 6481da177e4SLinus Torvalds 649d2caa3c5SJeff Layton /* 650d2caa3c5SJeff Layton * Was this inode dirtied after sync_sb_inodes was called? 651d2caa3c5SJeff Layton * This keeps sync from extra jobs and livelock. 652d2caa3c5SJeff Layton */ 653d2caa3c5SJeff Layton if (inode_dirtied_after(inode, start)) 6541da177e4SLinus Torvalds break; 6551da177e4SLinus Torvalds 65603ba3782SJens Axboe if (pin_sb_for_writeback(wbc, inode)) { 65703ba3782SJens Axboe requeue_io(inode); 65803ba3782SJens Axboe continue; 65903ba3782SJens Axboe } 6601da177e4SLinus Torvalds 66184a89245SWu Fengguang BUG_ON(inode->i_state & (I_FREEING | I_CLEAR)); 6621da177e4SLinus Torvalds __iget(inode); 6631da177e4SLinus Torvalds pages_skipped = wbc->pages_skipped; 66401c03194SChristoph Hellwig writeback_single_inode(inode, wbc); 66503ba3782SJens Axboe unpin_sb_for_writeback(wbc, inode); 6661da177e4SLinus Torvalds if (wbc->pages_skipped != pages_skipped) { 6671da177e4SLinus Torvalds /* 6681da177e4SLinus Torvalds * writeback is not making progress due to locked 6691da177e4SLinus Torvalds * buffers. Skip this inode for now. 6701da177e4SLinus Torvalds */ 671f57b9b7bSAndrew Morton redirty_tail(inode); 6721da177e4SLinus Torvalds } 6731da177e4SLinus Torvalds spin_unlock(&inode_lock); 6741da177e4SLinus Torvalds iput(inode); 6754ffc8444SOGAWA Hirofumi cond_resched(); 6761da177e4SLinus Torvalds spin_lock(&inode_lock); 6778bc3be27SFengguang Wu if (wbc->nr_to_write <= 0) { 6788bc3be27SFengguang Wu wbc->more_io = 1; 6791da177e4SLinus Torvalds break; 6801da177e4SLinus Torvalds } 68103ba3782SJens Axboe if (!list_empty(&wb->b_more_io)) 6828bc3be27SFengguang Wu wbc->more_io = 1; 6838bc3be27SFengguang Wu } 68438f21977SNick Piggin 68566f3b8e2SJens Axboe spin_unlock(&inode_lock); 68666f3b8e2SJens Axboe /* Leave any unwritten inodes on b_io */ 68766f3b8e2SJens Axboe } 68866f3b8e2SJens Axboe 68903ba3782SJens Axboe void writeback_inodes_wbc(struct writeback_control *wbc) 69003ba3782SJens Axboe { 69103ba3782SJens Axboe struct backing_dev_info *bdi = wbc->bdi; 69203ba3782SJens Axboe 69303ba3782SJens Axboe writeback_inodes_wb(&bdi->wb, wbc); 69403ba3782SJens Axboe } 69503ba3782SJens Axboe 69603ba3782SJens Axboe /* 69703ba3782SJens Axboe * The maximum number of pages to writeout in a single bdi flush/kupdate 69803ba3782SJens Axboe * operation. We do this so we don't hold I_SYNC against an inode for 69903ba3782SJens Axboe * enormous amounts of time, which would block a userspace task which has 70003ba3782SJens Axboe * been forced to throttle against that inode. Also, the code reevaluates 70103ba3782SJens Axboe * the dirty each time it has written this many pages. 70203ba3782SJens Axboe */ 70303ba3782SJens Axboe #define MAX_WRITEBACK_PAGES 1024 70403ba3782SJens Axboe 70503ba3782SJens Axboe static inline bool over_bground_thresh(void) 70603ba3782SJens Axboe { 70703ba3782SJens Axboe unsigned long background_thresh, dirty_thresh; 70803ba3782SJens Axboe 70903ba3782SJens Axboe get_dirty_limits(&background_thresh, &dirty_thresh, NULL, NULL); 71003ba3782SJens Axboe 71103ba3782SJens Axboe return (global_page_state(NR_FILE_DIRTY) + 71203ba3782SJens Axboe global_page_state(NR_UNSTABLE_NFS) >= background_thresh); 71303ba3782SJens Axboe } 71403ba3782SJens Axboe 71503ba3782SJens Axboe /* 71603ba3782SJens Axboe * Explicit flushing or periodic writeback of "old" data. 71703ba3782SJens Axboe * 71803ba3782SJens Axboe * Define "old": the first time one of an inode's pages is dirtied, we mark the 71903ba3782SJens Axboe * dirtying-time in the inode's address_space. So this periodic writeback code 72003ba3782SJens Axboe * just walks the superblock inode list, writing back any inodes which are 72103ba3782SJens Axboe * older than a specific point in time. 72203ba3782SJens Axboe * 72303ba3782SJens Axboe * Try to run once per dirty_writeback_interval. But if a writeback event 72403ba3782SJens Axboe * takes longer than a dirty_writeback_interval interval, then leave a 72503ba3782SJens Axboe * one-second gap. 72603ba3782SJens Axboe * 72703ba3782SJens Axboe * older_than_this takes precedence over nr_to_write. So we'll only write back 72803ba3782SJens Axboe * all dirty pages if they are all attached to "old" mappings. 72903ba3782SJens Axboe */ 730c4a77a6cSJens Axboe static long wb_writeback(struct bdi_writeback *wb, 731c4a77a6cSJens Axboe struct wb_writeback_args *args) 73203ba3782SJens Axboe { 73303ba3782SJens Axboe struct writeback_control wbc = { 73403ba3782SJens Axboe .bdi = wb->bdi, 735c4a77a6cSJens Axboe .sb = args->sb, 736c4a77a6cSJens Axboe .sync_mode = args->sync_mode, 73703ba3782SJens Axboe .older_than_this = NULL, 738c4a77a6cSJens Axboe .for_kupdate = args->for_kupdate, 739c4a77a6cSJens Axboe .range_cyclic = args->range_cyclic, 74003ba3782SJens Axboe }; 74103ba3782SJens Axboe unsigned long oldest_jif; 74203ba3782SJens Axboe long wrote = 0; 743a5989bdcSJan Kara struct inode *inode; 74403ba3782SJens Axboe 74503ba3782SJens Axboe if (wbc.for_kupdate) { 74603ba3782SJens Axboe wbc.older_than_this = &oldest_jif; 74703ba3782SJens Axboe oldest_jif = jiffies - 74803ba3782SJens Axboe msecs_to_jiffies(dirty_expire_interval * 10); 74903ba3782SJens Axboe } 750c4a77a6cSJens Axboe if (!wbc.range_cyclic) { 751c4a77a6cSJens Axboe wbc.range_start = 0; 752c4a77a6cSJens Axboe wbc.range_end = LLONG_MAX; 753c4a77a6cSJens Axboe } 75403ba3782SJens Axboe 75503ba3782SJens Axboe for (;;) { 75603ba3782SJens Axboe /* 757d3ddec76SWu Fengguang * Stop writeback when nr_pages has been consumed 75803ba3782SJens Axboe */ 759d3ddec76SWu Fengguang if (args->nr_pages <= 0) 76003ba3782SJens Axboe break; 76103ba3782SJens Axboe 76203ba3782SJens Axboe /* 763d3ddec76SWu Fengguang * For background writeout, stop when we are below the 764d3ddec76SWu Fengguang * background dirty threshold 76503ba3782SJens Axboe */ 766d3ddec76SWu Fengguang if (args->for_background && !over_bground_thresh()) 76703ba3782SJens Axboe break; 76803ba3782SJens Axboe 76903ba3782SJens Axboe wbc.more_io = 0; 77003ba3782SJens Axboe wbc.encountered_congestion = 0; 77103ba3782SJens Axboe wbc.nr_to_write = MAX_WRITEBACK_PAGES; 77203ba3782SJens Axboe wbc.pages_skipped = 0; 77303ba3782SJens Axboe writeback_inodes_wb(wb, &wbc); 774c4a77a6cSJens Axboe args->nr_pages -= MAX_WRITEBACK_PAGES - wbc.nr_to_write; 77503ba3782SJens Axboe wrote += MAX_WRITEBACK_PAGES - wbc.nr_to_write; 77603ba3782SJens Axboe 77703ba3782SJens Axboe /* 77871fd05a8SJens Axboe * If we consumed everything, see if we have more 77903ba3782SJens Axboe */ 78071fd05a8SJens Axboe if (wbc.nr_to_write <= 0) 78171fd05a8SJens Axboe continue; 78271fd05a8SJens Axboe /* 78371fd05a8SJens Axboe * Didn't write everything and we don't have more IO, bail 78471fd05a8SJens Axboe */ 78571fd05a8SJens Axboe if (!wbc.more_io) 78671fd05a8SJens Axboe break; 78771fd05a8SJens Axboe /* 78871fd05a8SJens Axboe * Did we write something? Try for more 78971fd05a8SJens Axboe */ 790a5989bdcSJan Kara if (wbc.nr_to_write < MAX_WRITEBACK_PAGES) 79103ba3782SJens Axboe continue; 792a5989bdcSJan Kara /* 793a5989bdcSJan Kara * Nothing written. Wait for some inode to 794a5989bdcSJan Kara * become available for writeback. Otherwise 795a5989bdcSJan Kara * we'll just busyloop. 796a5989bdcSJan Kara */ 797a5989bdcSJan Kara spin_lock(&inode_lock); 798a5989bdcSJan Kara if (!list_empty(&wb->b_more_io)) { 79971fd05a8SJens Axboe inode = list_entry(wb->b_more_io.prev, 800a5989bdcSJan Kara struct inode, i_list); 801a5989bdcSJan Kara inode_wait_for_writeback(inode); 802a5989bdcSJan Kara } 803a5989bdcSJan Kara spin_unlock(&inode_lock); 80403ba3782SJens Axboe } 80503ba3782SJens Axboe 80603ba3782SJens Axboe return wrote; 80703ba3782SJens Axboe } 80803ba3782SJens Axboe 80903ba3782SJens Axboe /* 81003ba3782SJens Axboe * Return the next bdi_work struct that hasn't been processed by this 8118010c3b6SJens Axboe * wb thread yet. ->seen is initially set for each thread that exists 8128010c3b6SJens Axboe * for this device, when a thread first notices a piece of work it 8138010c3b6SJens Axboe * clears its bit. Depending on writeback type, the thread will notify 8148010c3b6SJens Axboe * completion on either receiving the work (WB_SYNC_NONE) or after 8158010c3b6SJens Axboe * it is done (WB_SYNC_ALL). 81603ba3782SJens Axboe */ 81703ba3782SJens Axboe static struct bdi_work *get_next_work_item(struct backing_dev_info *bdi, 81803ba3782SJens Axboe struct bdi_writeback *wb) 81903ba3782SJens Axboe { 82003ba3782SJens Axboe struct bdi_work *work, *ret = NULL; 82103ba3782SJens Axboe 82203ba3782SJens Axboe rcu_read_lock(); 82303ba3782SJens Axboe 82403ba3782SJens Axboe list_for_each_entry_rcu(work, &bdi->work_list, list) { 82577fad5e6SNick Piggin if (!test_bit(wb->nr, &work->seen)) 82603ba3782SJens Axboe continue; 82777fad5e6SNick Piggin clear_bit(wb->nr, &work->seen); 82803ba3782SJens Axboe 82903ba3782SJens Axboe ret = work; 83003ba3782SJens Axboe break; 83103ba3782SJens Axboe } 83203ba3782SJens Axboe 83303ba3782SJens Axboe rcu_read_unlock(); 83403ba3782SJens Axboe return ret; 83503ba3782SJens Axboe } 83603ba3782SJens Axboe 83703ba3782SJens Axboe static long wb_check_old_data_flush(struct bdi_writeback *wb) 83803ba3782SJens Axboe { 83903ba3782SJens Axboe unsigned long expired; 84003ba3782SJens Axboe long nr_pages; 84103ba3782SJens Axboe 84203ba3782SJens Axboe expired = wb->last_old_flush + 84303ba3782SJens Axboe msecs_to_jiffies(dirty_writeback_interval * 10); 84403ba3782SJens Axboe if (time_before(jiffies, expired)) 84503ba3782SJens Axboe return 0; 84603ba3782SJens Axboe 84703ba3782SJens Axboe wb->last_old_flush = jiffies; 84803ba3782SJens Axboe nr_pages = global_page_state(NR_FILE_DIRTY) + 84903ba3782SJens Axboe global_page_state(NR_UNSTABLE_NFS) + 85003ba3782SJens Axboe (inodes_stat.nr_inodes - inodes_stat.nr_unused); 85103ba3782SJens Axboe 852c4a77a6cSJens Axboe if (nr_pages) { 853c4a77a6cSJens Axboe struct wb_writeback_args args = { 854c4a77a6cSJens Axboe .nr_pages = nr_pages, 855c4a77a6cSJens Axboe .sync_mode = WB_SYNC_NONE, 856c4a77a6cSJens Axboe .for_kupdate = 1, 857c4a77a6cSJens Axboe .range_cyclic = 1, 858c4a77a6cSJens Axboe }; 859c4a77a6cSJens Axboe 860c4a77a6cSJens Axboe return wb_writeback(wb, &args); 861c4a77a6cSJens Axboe } 86203ba3782SJens Axboe 86303ba3782SJens Axboe return 0; 86403ba3782SJens Axboe } 86503ba3782SJens Axboe 86603ba3782SJens Axboe /* 86703ba3782SJens Axboe * Retrieve work items and do the writeback they describe 86803ba3782SJens Axboe */ 86903ba3782SJens Axboe long wb_do_writeback(struct bdi_writeback *wb, int force_wait) 87003ba3782SJens Axboe { 87103ba3782SJens Axboe struct backing_dev_info *bdi = wb->bdi; 87203ba3782SJens Axboe struct bdi_work *work; 873c4a77a6cSJens Axboe long wrote = 0; 87403ba3782SJens Axboe 87503ba3782SJens Axboe while ((work = get_next_work_item(bdi, wb)) != NULL) { 876c4a77a6cSJens Axboe struct wb_writeback_args args = work->args; 87703ba3782SJens Axboe 87803ba3782SJens Axboe /* 87903ba3782SJens Axboe * Override sync mode, in case we must wait for completion 88003ba3782SJens Axboe */ 88103ba3782SJens Axboe if (force_wait) 882c4a77a6cSJens Axboe work->args.sync_mode = args.sync_mode = WB_SYNC_ALL; 88303ba3782SJens Axboe 88403ba3782SJens Axboe /* 88503ba3782SJens Axboe * If this isn't a data integrity operation, just notify 88603ba3782SJens Axboe * that we have seen this work and we are now starting it. 88703ba3782SJens Axboe */ 888c4a77a6cSJens Axboe if (args.sync_mode == WB_SYNC_NONE) 88903ba3782SJens Axboe wb_clear_pending(wb, work); 89003ba3782SJens Axboe 891c4a77a6cSJens Axboe wrote += wb_writeback(wb, &args); 89203ba3782SJens Axboe 89303ba3782SJens Axboe /* 89403ba3782SJens Axboe * This is a data integrity writeback, so only do the 89503ba3782SJens Axboe * notification when we have completed the work. 89603ba3782SJens Axboe */ 897c4a77a6cSJens Axboe if (args.sync_mode == WB_SYNC_ALL) 89803ba3782SJens Axboe wb_clear_pending(wb, work); 89903ba3782SJens Axboe } 90003ba3782SJens Axboe 90103ba3782SJens Axboe /* 90203ba3782SJens Axboe * Check for periodic writeback, kupdated() style 90303ba3782SJens Axboe */ 90403ba3782SJens Axboe wrote += wb_check_old_data_flush(wb); 90503ba3782SJens Axboe 90603ba3782SJens Axboe return wrote; 90703ba3782SJens Axboe } 90803ba3782SJens Axboe 90903ba3782SJens Axboe /* 91003ba3782SJens Axboe * Handle writeback of dirty data for the device backed by this bdi. Also 91103ba3782SJens Axboe * wakes up periodically and does kupdated style flushing. 91203ba3782SJens Axboe */ 91303ba3782SJens Axboe int bdi_writeback_task(struct bdi_writeback *wb) 91403ba3782SJens Axboe { 91503ba3782SJens Axboe unsigned long last_active = jiffies; 91603ba3782SJens Axboe unsigned long wait_jiffies = -1UL; 91703ba3782SJens Axboe long pages_written; 91803ba3782SJens Axboe 91903ba3782SJens Axboe while (!kthread_should_stop()) { 92003ba3782SJens Axboe pages_written = wb_do_writeback(wb, 0); 92103ba3782SJens Axboe 92203ba3782SJens Axboe if (pages_written) 92303ba3782SJens Axboe last_active = jiffies; 92403ba3782SJens Axboe else if (wait_jiffies != -1UL) { 92503ba3782SJens Axboe unsigned long max_idle; 92603ba3782SJens Axboe 92703ba3782SJens Axboe /* 92803ba3782SJens Axboe * Longest period of inactivity that we tolerate. If we 92903ba3782SJens Axboe * see dirty data again later, the task will get 93003ba3782SJens Axboe * recreated automatically. 93103ba3782SJens Axboe */ 93203ba3782SJens Axboe max_idle = max(5UL * 60 * HZ, wait_jiffies); 93303ba3782SJens Axboe if (time_after(jiffies, max_idle + last_active)) 93403ba3782SJens Axboe break; 93503ba3782SJens Axboe } 93603ba3782SJens Axboe 93703ba3782SJens Axboe wait_jiffies = msecs_to_jiffies(dirty_writeback_interval * 10); 93849db0414SJens Axboe schedule_timeout_interruptible(wait_jiffies); 93903ba3782SJens Axboe try_to_freeze(); 94003ba3782SJens Axboe } 94103ba3782SJens Axboe 94203ba3782SJens Axboe return 0; 94303ba3782SJens Axboe } 94403ba3782SJens Axboe 94503ba3782SJens Axboe /* 946b6e51316SJens Axboe * Schedule writeback for all backing devices. This does WB_SYNC_NONE 947b6e51316SJens Axboe * writeback, for integrity writeback see bdi_sync_writeback(). 94803ba3782SJens Axboe */ 949b6e51316SJens Axboe static void bdi_writeback_all(struct super_block *sb, long nr_pages) 95003ba3782SJens Axboe { 951b6e51316SJens Axboe struct wb_writeback_args args = { 952b6e51316SJens Axboe .sb = sb, 953b6e51316SJens Axboe .nr_pages = nr_pages, 954b6e51316SJens Axboe .sync_mode = WB_SYNC_NONE, 955b6e51316SJens Axboe }; 95603ba3782SJens Axboe struct backing_dev_info *bdi; 95703ba3782SJens Axboe 958cfc4ba53SJens Axboe rcu_read_lock(); 95903ba3782SJens Axboe 960cfc4ba53SJens Axboe list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { 96103ba3782SJens Axboe if (!bdi_has_dirty_io(bdi)) 96203ba3782SJens Axboe continue; 96303ba3782SJens Axboe 964b6e51316SJens Axboe bdi_alloc_queue_work(bdi, &args); 96503ba3782SJens Axboe } 96603ba3782SJens Axboe 967cfc4ba53SJens Axboe rcu_read_unlock(); 96803ba3782SJens Axboe } 96903ba3782SJens Axboe 97003ba3782SJens Axboe /* 97103ba3782SJens Axboe * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back 97203ba3782SJens Axboe * the whole world. 97303ba3782SJens Axboe */ 97403ba3782SJens Axboe void wakeup_flusher_threads(long nr_pages) 97503ba3782SJens Axboe { 97603ba3782SJens Axboe if (nr_pages == 0) 97703ba3782SJens Axboe nr_pages = global_page_state(NR_FILE_DIRTY) + 97803ba3782SJens Axboe global_page_state(NR_UNSTABLE_NFS); 979b6e51316SJens Axboe bdi_writeback_all(NULL, nr_pages); 98003ba3782SJens Axboe } 98103ba3782SJens Axboe 98203ba3782SJens Axboe static noinline void block_dump___mark_inode_dirty(struct inode *inode) 98303ba3782SJens Axboe { 98403ba3782SJens Axboe if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { 98503ba3782SJens Axboe struct dentry *dentry; 98603ba3782SJens Axboe const char *name = "?"; 98703ba3782SJens Axboe 98803ba3782SJens Axboe dentry = d_find_alias(inode); 98903ba3782SJens Axboe if (dentry) { 99003ba3782SJens Axboe spin_lock(&dentry->d_lock); 99103ba3782SJens Axboe name = (const char *) dentry->d_name.name; 99203ba3782SJens Axboe } 99303ba3782SJens Axboe printk(KERN_DEBUG 99403ba3782SJens Axboe "%s(%d): dirtied inode %lu (%s) on %s\n", 99503ba3782SJens Axboe current->comm, task_pid_nr(current), inode->i_ino, 99603ba3782SJens Axboe name, inode->i_sb->s_id); 99703ba3782SJens Axboe if (dentry) { 99803ba3782SJens Axboe spin_unlock(&dentry->d_lock); 99903ba3782SJens Axboe dput(dentry); 100003ba3782SJens Axboe } 100103ba3782SJens Axboe } 100203ba3782SJens Axboe } 100303ba3782SJens Axboe 100403ba3782SJens Axboe /** 100503ba3782SJens Axboe * __mark_inode_dirty - internal function 100603ba3782SJens Axboe * @inode: inode to mark 100703ba3782SJens Axboe * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) 100803ba3782SJens Axboe * Mark an inode as dirty. Callers should use mark_inode_dirty or 100903ba3782SJens Axboe * mark_inode_dirty_sync. 101003ba3782SJens Axboe * 101103ba3782SJens Axboe * Put the inode on the super block's dirty list. 101203ba3782SJens Axboe * 101303ba3782SJens Axboe * CAREFUL! We mark it dirty unconditionally, but move it onto the 101403ba3782SJens Axboe * dirty list only if it is hashed or if it refers to a blockdev. 101503ba3782SJens Axboe * If it was not hashed, it will never be added to the dirty list 101603ba3782SJens Axboe * even if it is later hashed, as it will have been marked dirty already. 101703ba3782SJens Axboe * 101803ba3782SJens Axboe * In short, make sure you hash any inodes _before_ you start marking 101903ba3782SJens Axboe * them dirty. 102003ba3782SJens Axboe * 102103ba3782SJens Axboe * This function *must* be atomic for the I_DIRTY_PAGES case - 102203ba3782SJens Axboe * set_page_dirty() is called under spinlock in several places. 102303ba3782SJens Axboe * 102403ba3782SJens Axboe * Note that for blockdevs, inode->dirtied_when represents the dirtying time of 102503ba3782SJens Axboe * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of 102603ba3782SJens Axboe * the kernel-internal blockdev inode represents the dirtying time of the 102703ba3782SJens Axboe * blockdev's pages. This is why for I_DIRTY_PAGES we always use 102803ba3782SJens Axboe * page->mapping->host, so the page-dirtying time is recorded in the internal 102903ba3782SJens Axboe * blockdev inode. 103003ba3782SJens Axboe */ 103103ba3782SJens Axboe void __mark_inode_dirty(struct inode *inode, int flags) 103203ba3782SJens Axboe { 103303ba3782SJens Axboe struct super_block *sb = inode->i_sb; 103403ba3782SJens Axboe 103503ba3782SJens Axboe /* 103603ba3782SJens Axboe * Don't do this for I_DIRTY_PAGES - that doesn't actually 103703ba3782SJens Axboe * dirty the inode itself 103803ba3782SJens Axboe */ 103903ba3782SJens Axboe if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { 104003ba3782SJens Axboe if (sb->s_op->dirty_inode) 104103ba3782SJens Axboe sb->s_op->dirty_inode(inode); 104203ba3782SJens Axboe } 104303ba3782SJens Axboe 104403ba3782SJens Axboe /* 104503ba3782SJens Axboe * make sure that changes are seen by all cpus before we test i_state 104603ba3782SJens Axboe * -- mikulas 104703ba3782SJens Axboe */ 104803ba3782SJens Axboe smp_mb(); 104903ba3782SJens Axboe 105003ba3782SJens Axboe /* avoid the locking if we can */ 105103ba3782SJens Axboe if ((inode->i_state & flags) == flags) 105203ba3782SJens Axboe return; 105303ba3782SJens Axboe 105403ba3782SJens Axboe if (unlikely(block_dump)) 105503ba3782SJens Axboe block_dump___mark_inode_dirty(inode); 105603ba3782SJens Axboe 105703ba3782SJens Axboe spin_lock(&inode_lock); 105803ba3782SJens Axboe if ((inode->i_state & flags) != flags) { 105903ba3782SJens Axboe const int was_dirty = inode->i_state & I_DIRTY; 106003ba3782SJens Axboe 106103ba3782SJens Axboe inode->i_state |= flags; 106203ba3782SJens Axboe 106303ba3782SJens Axboe /* 106403ba3782SJens Axboe * If the inode is being synced, just update its dirty state. 106503ba3782SJens Axboe * The unlocker will place the inode on the appropriate 106603ba3782SJens Axboe * superblock list, based upon its state. 106703ba3782SJens Axboe */ 106803ba3782SJens Axboe if (inode->i_state & I_SYNC) 106903ba3782SJens Axboe goto out; 107003ba3782SJens Axboe 107103ba3782SJens Axboe /* 107203ba3782SJens Axboe * Only add valid (hashed) inodes to the superblock's 107303ba3782SJens Axboe * dirty list. Add blockdev inodes as well. 107403ba3782SJens Axboe */ 107503ba3782SJens Axboe if (!S_ISBLK(inode->i_mode)) { 107603ba3782SJens Axboe if (hlist_unhashed(&inode->i_hash)) 107703ba3782SJens Axboe goto out; 107803ba3782SJens Axboe } 107903ba3782SJens Axboe if (inode->i_state & (I_FREEING|I_CLEAR)) 108003ba3782SJens Axboe goto out; 108103ba3782SJens Axboe 108203ba3782SJens Axboe /* 108303ba3782SJens Axboe * If the inode was already on b_dirty/b_io/b_more_io, don't 108403ba3782SJens Axboe * reposition it (that would break b_dirty time-ordering). 108503ba3782SJens Axboe */ 108603ba3782SJens Axboe if (!was_dirty) { 108703ba3782SJens Axboe struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; 1088500b067cSJens Axboe struct backing_dev_info *bdi = wb->bdi; 1089500b067cSJens Axboe 1090500b067cSJens Axboe if (bdi_cap_writeback_dirty(bdi) && 1091500b067cSJens Axboe !test_bit(BDI_registered, &bdi->state)) { 1092500b067cSJens Axboe WARN_ON(1); 1093500b067cSJens Axboe printk(KERN_ERR "bdi-%s not registered\n", 1094500b067cSJens Axboe bdi->name); 1095500b067cSJens Axboe } 109603ba3782SJens Axboe 109703ba3782SJens Axboe inode->dirtied_when = jiffies; 109803ba3782SJens Axboe list_move(&inode->i_list, &wb->b_dirty); 109903ba3782SJens Axboe } 110003ba3782SJens Axboe } 110103ba3782SJens Axboe out: 110203ba3782SJens Axboe spin_unlock(&inode_lock); 110303ba3782SJens Axboe } 110403ba3782SJens Axboe EXPORT_SYMBOL(__mark_inode_dirty); 110503ba3782SJens Axboe 110666f3b8e2SJens Axboe /* 110766f3b8e2SJens Axboe * Write out a superblock's list of dirty inodes. A wait will be performed 110866f3b8e2SJens Axboe * upon no inodes, all inodes or the final one, depending upon sync_mode. 110966f3b8e2SJens Axboe * 111066f3b8e2SJens Axboe * If older_than_this is non-NULL, then only write out inodes which 111166f3b8e2SJens Axboe * had their first dirtying at a time earlier than *older_than_this. 111266f3b8e2SJens Axboe * 111366f3b8e2SJens Axboe * If `bdi' is non-zero then we're being asked to writeback a specific queue. 111466f3b8e2SJens Axboe * This function assumes that the blockdev superblock's inodes are backed by 111566f3b8e2SJens Axboe * a variety of queues, so all inodes are searched. For other superblocks, 111666f3b8e2SJens Axboe * assume that all inodes are backed by the same queue. 111766f3b8e2SJens Axboe * 111866f3b8e2SJens Axboe * The inodes to be written are parked on bdi->b_io. They are moved back onto 111966f3b8e2SJens Axboe * bdi->b_dirty as they are selected for writing. This way, none can be missed 112066f3b8e2SJens Axboe * on the writer throttling path, and we get decent balancing between many 112166f3b8e2SJens Axboe * throttled threads: we don't want them all piling up on inode_sync_wait. 112266f3b8e2SJens Axboe */ 1123b6e51316SJens Axboe static void wait_sb_inodes(struct super_block *sb) 112466f3b8e2SJens Axboe { 112538f21977SNick Piggin struct inode *inode, *old_inode = NULL; 112638f21977SNick Piggin 112703ba3782SJens Axboe /* 112803ba3782SJens Axboe * We need to be protected against the filesystem going from 112903ba3782SJens Axboe * r/o to r/w or vice versa. 113003ba3782SJens Axboe */ 1131b6e51316SJens Axboe WARN_ON(!rwsem_is_locked(&sb->s_umount)); 113203ba3782SJens Axboe 113366f3b8e2SJens Axboe spin_lock(&inode_lock); 113466f3b8e2SJens Axboe 113538f21977SNick Piggin /* 113638f21977SNick Piggin * Data integrity sync. Must wait for all pages under writeback, 113738f21977SNick Piggin * because there may have been pages dirtied before our sync 113838f21977SNick Piggin * call, but which had writeout started before we write it out. 113938f21977SNick Piggin * In which case, the inode may not be on the dirty list, but 114038f21977SNick Piggin * we still have to wait for that writeout. 114138f21977SNick Piggin */ 1142b6e51316SJens Axboe list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { 114338f21977SNick Piggin struct address_space *mapping; 114438f21977SNick Piggin 114503ba3782SJens Axboe if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE|I_NEW)) 114638f21977SNick Piggin continue; 114738f21977SNick Piggin mapping = inode->i_mapping; 114838f21977SNick Piggin if (mapping->nrpages == 0) 114938f21977SNick Piggin continue; 115038f21977SNick Piggin __iget(inode); 1151ae8547b0SHans Reiser spin_unlock(&inode_lock); 115238f21977SNick Piggin /* 115338f21977SNick Piggin * We hold a reference to 'inode' so it couldn't have 115438f21977SNick Piggin * been removed from s_inodes list while we dropped the 115538f21977SNick Piggin * inode_lock. We cannot iput the inode now as we can 115638f21977SNick Piggin * be holding the last reference and we cannot iput it 115738f21977SNick Piggin * under inode_lock. So we keep the reference and iput 115838f21977SNick Piggin * it later. 115938f21977SNick Piggin */ 116038f21977SNick Piggin iput(old_inode); 116138f21977SNick Piggin old_inode = inode; 116238f21977SNick Piggin 116338f21977SNick Piggin filemap_fdatawait(mapping); 116438f21977SNick Piggin 116538f21977SNick Piggin cond_resched(); 116638f21977SNick Piggin 116738f21977SNick Piggin spin_lock(&inode_lock); 116838f21977SNick Piggin } 116938f21977SNick Piggin spin_unlock(&inode_lock); 117038f21977SNick Piggin iput(old_inode); 117166f3b8e2SJens Axboe } 11721da177e4SLinus Torvalds 1173d8a8559cSJens Axboe /** 1174d8a8559cSJens Axboe * writeback_inodes_sb - writeback dirty inodes from given super_block 1175d8a8559cSJens Axboe * @sb: the superblock 11761da177e4SLinus Torvalds * 1177d8a8559cSJens Axboe * Start writeback on some inodes on this super_block. No guarantees are made 1178d8a8559cSJens Axboe * on how many (if any) will be written, and this function does not wait 1179d8a8559cSJens Axboe * for IO completion of submitted IO. The number of pages submitted is 1180d8a8559cSJens Axboe * returned. 11811da177e4SLinus Torvalds */ 1182b6e51316SJens Axboe void writeback_inodes_sb(struct super_block *sb) 11831da177e4SLinus Torvalds { 1184b1e7a8fdSChristoph Lameter unsigned long nr_dirty = global_page_state(NR_FILE_DIRTY); 1185fd39fc85SChristoph Lameter unsigned long nr_unstable = global_page_state(NR_UNSTABLE_NFS); 1186d8a8559cSJens Axboe long nr_to_write; 11871da177e4SLinus Torvalds 1188d8a8559cSJens Axboe nr_to_write = nr_dirty + nr_unstable + 118938f21977SNick Piggin (inodes_stat.nr_inodes - inodes_stat.nr_unused); 119038f21977SNick Piggin 1191b6e51316SJens Axboe bdi_writeback_all(sb, nr_to_write); 11921da177e4SLinus Torvalds } 1193d8a8559cSJens Axboe EXPORT_SYMBOL(writeback_inodes_sb); 1194d8a8559cSJens Axboe 1195d8a8559cSJens Axboe /** 1196d8a8559cSJens Axboe * sync_inodes_sb - sync sb inode pages 1197d8a8559cSJens Axboe * @sb: the superblock 1198d8a8559cSJens Axboe * 1199d8a8559cSJens Axboe * This function writes and waits on any dirty inode belonging to this 1200d8a8559cSJens Axboe * super_block. The number of pages synced is returned. 1201d8a8559cSJens Axboe */ 1202b6e51316SJens Axboe void sync_inodes_sb(struct super_block *sb) 1203d8a8559cSJens Axboe { 1204b6e51316SJens Axboe bdi_sync_writeback(sb->s_bdi, sb); 1205b6e51316SJens Axboe wait_sb_inodes(sb); 1206d8a8559cSJens Axboe } 1207d8a8559cSJens Axboe EXPORT_SYMBOL(sync_inodes_sb); 12081da177e4SLinus Torvalds 12091da177e4SLinus Torvalds /** 12101da177e4SLinus Torvalds * write_inode_now - write an inode to disk 12111da177e4SLinus Torvalds * @inode: inode to write to disk 12121da177e4SLinus Torvalds * @sync: whether the write should be synchronous or not 12131da177e4SLinus Torvalds * 12147f04c26dSAndrea Arcangeli * This function commits an inode to disk immediately if it is dirty. This is 12157f04c26dSAndrea Arcangeli * primarily needed by knfsd. 12167f04c26dSAndrea Arcangeli * 12177f04c26dSAndrea Arcangeli * The caller must either have a ref on the inode or must have set I_WILL_FREE. 12181da177e4SLinus Torvalds */ 12191da177e4SLinus Torvalds int write_inode_now(struct inode *inode, int sync) 12201da177e4SLinus Torvalds { 12211da177e4SLinus Torvalds int ret; 12221da177e4SLinus Torvalds struct writeback_control wbc = { 12231da177e4SLinus Torvalds .nr_to_write = LONG_MAX, 122418914b18SMike Galbraith .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, 1225111ebb6eSOGAWA Hirofumi .range_start = 0, 1226111ebb6eSOGAWA Hirofumi .range_end = LLONG_MAX, 12271da177e4SLinus Torvalds }; 12281da177e4SLinus Torvalds 12291da177e4SLinus Torvalds if (!mapping_cap_writeback_dirty(inode->i_mapping)) 123049364ce2SAndrew Morton wbc.nr_to_write = 0; 12311da177e4SLinus Torvalds 12321da177e4SLinus Torvalds might_sleep(); 12331da177e4SLinus Torvalds spin_lock(&inode_lock); 123401c03194SChristoph Hellwig ret = writeback_single_inode(inode, &wbc); 12351da177e4SLinus Torvalds spin_unlock(&inode_lock); 12361da177e4SLinus Torvalds if (sync) 12371c0eeaf5SJoern Engel inode_sync_wait(inode); 12381da177e4SLinus Torvalds return ret; 12391da177e4SLinus Torvalds } 12401da177e4SLinus Torvalds EXPORT_SYMBOL(write_inode_now); 12411da177e4SLinus Torvalds 12421da177e4SLinus Torvalds /** 12431da177e4SLinus Torvalds * sync_inode - write an inode and its pages to disk. 12441da177e4SLinus Torvalds * @inode: the inode to sync 12451da177e4SLinus Torvalds * @wbc: controls the writeback mode 12461da177e4SLinus Torvalds * 12471da177e4SLinus Torvalds * sync_inode() will write an inode and its pages to disk. It will also 12481da177e4SLinus Torvalds * correctly update the inode on its superblock's dirty inode lists and will 12491da177e4SLinus Torvalds * update inode->i_state. 12501da177e4SLinus Torvalds * 12511da177e4SLinus Torvalds * The caller must have a ref on the inode. 12521da177e4SLinus Torvalds */ 12531da177e4SLinus Torvalds int sync_inode(struct inode *inode, struct writeback_control *wbc) 12541da177e4SLinus Torvalds { 12551da177e4SLinus Torvalds int ret; 12561da177e4SLinus Torvalds 12571da177e4SLinus Torvalds spin_lock(&inode_lock); 125801c03194SChristoph Hellwig ret = writeback_single_inode(inode, wbc); 12591da177e4SLinus Torvalds spin_unlock(&inode_lock); 12601da177e4SLinus Torvalds return ret; 12611da177e4SLinus Torvalds } 12621da177e4SLinus Torvalds EXPORT_SYMBOL(sync_inode); 1263