11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * fs/fs-writeback.c 31da177e4SLinus Torvalds * 41da177e4SLinus Torvalds * Copyright (C) 2002, Linus Torvalds. 51da177e4SLinus Torvalds * 61da177e4SLinus Torvalds * Contains all the functions related to writing back and waiting 71da177e4SLinus Torvalds * upon dirty inodes against superblocks, and writing back dirty 81da177e4SLinus Torvalds * pages against inodes. ie: data writeback. Writeout of the 91da177e4SLinus Torvalds * inode itself is not handled here. 101da177e4SLinus Torvalds * 11e1f8e874SFrancois Cami * 10Apr2002 Andrew Morton 121da177e4SLinus Torvalds * Split out of fs/inode.c 131da177e4SLinus Torvalds * Additions for address_space-based writeback 141da177e4SLinus Torvalds */ 151da177e4SLinus Torvalds 161da177e4SLinus Torvalds #include <linux/kernel.h> 17630d9c47SPaul Gortmaker #include <linux/export.h> 181da177e4SLinus Torvalds #include <linux/spinlock.h> 195a0e3ad6STejun Heo #include <linux/slab.h> 201da177e4SLinus Torvalds #include <linux/sched.h> 211da177e4SLinus Torvalds #include <linux/fs.h> 221da177e4SLinus Torvalds #include <linux/mm.h> 23bc31b86aSWu Fengguang #include <linux/pagemap.h> 2403ba3782SJens Axboe #include <linux/kthread.h> 251da177e4SLinus Torvalds #include <linux/writeback.h> 261da177e4SLinus Torvalds #include <linux/blkdev.h> 271da177e4SLinus Torvalds #include <linux/backing-dev.h> 28455b2864SDave Chinner #include <linux/tracepoint.h> 29719ea2fbSAl Viro #include <linux/device.h> 3007f3f05cSDavid Howells #include "internal.h" 311da177e4SLinus Torvalds 32d0bceac7SJens Axboe /* 33bc31b86aSWu Fengguang * 4MB minimal write chunk size 34bc31b86aSWu Fengguang */ 35bc31b86aSWu Fengguang #define MIN_WRITEBACK_PAGES (4096UL >> (PAGE_CACHE_SHIFT - 10)) 36bc31b86aSWu Fengguang 37bc31b86aSWu Fengguang /* 38c4a77a6cSJens Axboe * Passed into wb_writeback(), essentially a subset of writeback_control 39c4a77a6cSJens Axboe */ 4083ba7b07SChristoph Hellwig struct wb_writeback_work { 41c4a77a6cSJens Axboe long nr_pages; 42c4a77a6cSJens Axboe struct super_block *sb; 430dc83bd3SJan Kara unsigned long *older_than_this; 44c4a77a6cSJens Axboe enum writeback_sync_modes sync_mode; 456e6938b6SWu Fengguang unsigned int tagged_writepages:1; 4652957fe1SH Hartley Sweeten unsigned int for_kupdate:1; 4752957fe1SH Hartley Sweeten unsigned int range_cyclic:1; 4852957fe1SH Hartley Sweeten unsigned int for_background:1; 497747bd4bSDave Chinner unsigned int for_sync:1; /* sync(2) WB_SYNC_ALL writeback */ 500e175a18SCurt Wohlgemuth enum wb_reason reason; /* why was writeback initiated? */ 51c4a77a6cSJens Axboe 528010c3b6SJens Axboe struct list_head list; /* pending work list */ 5383ba7b07SChristoph Hellwig struct completion *done; /* set if the caller waits */ 5403ba3782SJens Axboe }; 5503ba3782SJens Axboe 56a2f48706STheodore Ts'o /* 57a2f48706STheodore Ts'o * If an inode is constantly having its pages dirtied, but then the 58a2f48706STheodore Ts'o * updates stop dirtytime_expire_interval seconds in the past, it's 59a2f48706STheodore Ts'o * possible for the worst case time between when an inode has its 60a2f48706STheodore Ts'o * timestamps updated and when they finally get written out to be two 61a2f48706STheodore Ts'o * dirtytime_expire_intervals. We set the default to 12 hours (in 62a2f48706STheodore Ts'o * seconds), which means most of the time inodes will have their 63a2f48706STheodore Ts'o * timestamps written to disk after 12 hours, but in the worst case a 64a2f48706STheodore Ts'o * few inodes might not their timestamps updated for 24 hours. 65a2f48706STheodore Ts'o */ 66a2f48706STheodore Ts'o unsigned int dirtytime_expire_interval = 12 * 60 * 60; 67a2f48706STheodore Ts'o 68f11b00f3SAdrian Bunk /** 69f11b00f3SAdrian Bunk * writeback_in_progress - determine whether there is writeback in progress 70f11b00f3SAdrian Bunk * @bdi: the device's backing_dev_info structure. 71f11b00f3SAdrian Bunk * 7203ba3782SJens Axboe * Determine whether there is writeback waiting to be handled against a 7303ba3782SJens Axboe * backing device. 74f11b00f3SAdrian Bunk */ 75f11b00f3SAdrian Bunk int writeback_in_progress(struct backing_dev_info *bdi) 76f11b00f3SAdrian Bunk { 774452226eSTejun Heo return test_bit(WB_writeback_running, &bdi->wb.state); 78f11b00f3SAdrian Bunk } 7900d4e736STheodore Ts'o EXPORT_SYMBOL(writeback_in_progress); 80f11b00f3SAdrian Bunk 817ccf19a8SNick Piggin static inline struct inode *wb_inode(struct list_head *head) 827ccf19a8SNick Piggin { 837ccf19a8SNick Piggin return list_entry(head, struct inode, i_wb_list); 847ccf19a8SNick Piggin } 857ccf19a8SNick Piggin 8615eb77a0SWu Fengguang /* 8715eb77a0SWu Fengguang * Include the creation of the trace points after defining the 8815eb77a0SWu Fengguang * wb_writeback_work structure and inline functions so that the definition 8915eb77a0SWu Fengguang * remains local to this file. 9015eb77a0SWu Fengguang */ 9115eb77a0SWu Fengguang #define CREATE_TRACE_POINTS 9215eb77a0SWu Fengguang #include <trace/events/writeback.h> 9315eb77a0SWu Fengguang 94774016b2SSteven Whitehouse EXPORT_TRACEPOINT_SYMBOL_GPL(wbc_writepage); 95774016b2SSteven Whitehouse 96d6c10f1fSTejun Heo static bool wb_io_lists_populated(struct bdi_writeback *wb) 97d6c10f1fSTejun Heo { 98d6c10f1fSTejun Heo if (wb_has_dirty_io(wb)) { 99d6c10f1fSTejun Heo return false; 100d6c10f1fSTejun Heo } else { 101d6c10f1fSTejun Heo set_bit(WB_has_dirty_io, &wb->state); 102d6c10f1fSTejun Heo return true; 103d6c10f1fSTejun Heo } 104d6c10f1fSTejun Heo } 105d6c10f1fSTejun Heo 106d6c10f1fSTejun Heo static void wb_io_lists_depopulated(struct bdi_writeback *wb) 107d6c10f1fSTejun Heo { 108d6c10f1fSTejun Heo if (wb_has_dirty_io(wb) && list_empty(&wb->b_dirty) && 109d6c10f1fSTejun Heo list_empty(&wb->b_io) && list_empty(&wb->b_more_io)) 110d6c10f1fSTejun Heo clear_bit(WB_has_dirty_io, &wb->state); 111d6c10f1fSTejun Heo } 112d6c10f1fSTejun Heo 113d6c10f1fSTejun Heo /** 114d6c10f1fSTejun Heo * inode_wb_list_move_locked - move an inode onto a bdi_writeback IO list 115d6c10f1fSTejun Heo * @inode: inode to be moved 116d6c10f1fSTejun Heo * @wb: target bdi_writeback 117d6c10f1fSTejun Heo * @head: one of @wb->b_{dirty|io|more_io} 118d6c10f1fSTejun Heo * 119d6c10f1fSTejun Heo * Move @inode->i_wb_list to @list of @wb and set %WB_has_dirty_io. 120d6c10f1fSTejun Heo * Returns %true if @inode is the first occupant of the !dirty_time IO 121d6c10f1fSTejun Heo * lists; otherwise, %false. 122d6c10f1fSTejun Heo */ 123d6c10f1fSTejun Heo static bool inode_wb_list_move_locked(struct inode *inode, 124d6c10f1fSTejun Heo struct bdi_writeback *wb, 125d6c10f1fSTejun Heo struct list_head *head) 126d6c10f1fSTejun Heo { 127d6c10f1fSTejun Heo assert_spin_locked(&wb->list_lock); 128d6c10f1fSTejun Heo 129d6c10f1fSTejun Heo list_move(&inode->i_wb_list, head); 130d6c10f1fSTejun Heo 131d6c10f1fSTejun Heo /* dirty_time doesn't count as dirty_io until expiration */ 132d6c10f1fSTejun Heo if (head != &wb->b_dirty_time) 133d6c10f1fSTejun Heo return wb_io_lists_populated(wb); 134d6c10f1fSTejun Heo 135d6c10f1fSTejun Heo wb_io_lists_depopulated(wb); 136d6c10f1fSTejun Heo return false; 137d6c10f1fSTejun Heo } 138d6c10f1fSTejun Heo 139d6c10f1fSTejun Heo /** 140d6c10f1fSTejun Heo * inode_wb_list_del_locked - remove an inode from its bdi_writeback IO list 141d6c10f1fSTejun Heo * @inode: inode to be removed 142d6c10f1fSTejun Heo * @wb: bdi_writeback @inode is being removed from 143d6c10f1fSTejun Heo * 144d6c10f1fSTejun Heo * Remove @inode which may be on one of @wb->b_{dirty|io|more_io} lists and 145d6c10f1fSTejun Heo * clear %WB_has_dirty_io if all are empty afterwards. 146d6c10f1fSTejun Heo */ 147d6c10f1fSTejun Heo static void inode_wb_list_del_locked(struct inode *inode, 148d6c10f1fSTejun Heo struct bdi_writeback *wb) 149d6c10f1fSTejun Heo { 150d6c10f1fSTejun Heo assert_spin_locked(&wb->list_lock); 151d6c10f1fSTejun Heo 152d6c10f1fSTejun Heo list_del_init(&inode->i_wb_list); 153d6c10f1fSTejun Heo wb_io_lists_depopulated(wb); 154d6c10f1fSTejun Heo } 155d6c10f1fSTejun Heo 156f0054bb1STejun Heo static void wb_wakeup(struct bdi_writeback *wb) 1575acda9d1SJan Kara { 158f0054bb1STejun Heo spin_lock_bh(&wb->work_lock); 159f0054bb1STejun Heo if (test_bit(WB_registered, &wb->state)) 160f0054bb1STejun Heo mod_delayed_work(bdi_wq, &wb->dwork, 0); 161f0054bb1STejun Heo spin_unlock_bh(&wb->work_lock); 1625acda9d1SJan Kara } 1635acda9d1SJan Kara 164f0054bb1STejun Heo static void wb_queue_work(struct bdi_writeback *wb, 1656585027aSJan Kara struct wb_writeback_work *work) 1666585027aSJan Kara { 167f0054bb1STejun Heo trace_writeback_queue(wb->bdi, work); 1686585027aSJan Kara 169f0054bb1STejun Heo spin_lock_bh(&wb->work_lock); 170f0054bb1STejun Heo if (!test_bit(WB_registered, &wb->state)) { 1715acda9d1SJan Kara if (work->done) 1725acda9d1SJan Kara complete(work->done); 1735acda9d1SJan Kara goto out_unlock; 1745acda9d1SJan Kara } 175f0054bb1STejun Heo list_add_tail(&work->list, &wb->work_list); 176f0054bb1STejun Heo mod_delayed_work(bdi_wq, &wb->dwork, 0); 1775acda9d1SJan Kara out_unlock: 178f0054bb1STejun Heo spin_unlock_bh(&wb->work_lock); 17903ba3782SJens Axboe } 1801da177e4SLinus Torvalds 181f0054bb1STejun Heo static void __wb_start_writeback(struct bdi_writeback *wb, long nr_pages, 1820e175a18SCurt Wohlgemuth bool range_cyclic, enum wb_reason reason) 1831da177e4SLinus Torvalds { 18483ba7b07SChristoph Hellwig struct wb_writeback_work *work; 18503ba3782SJens Axboe 186bcddc3f0SJens Axboe /* 187bcddc3f0SJens Axboe * This is WB_SYNC_NONE writeback, so if allocation fails just 188bcddc3f0SJens Axboe * wakeup the thread for old dirty data writeback 189bcddc3f0SJens Axboe */ 19083ba7b07SChristoph Hellwig work = kzalloc(sizeof(*work), GFP_ATOMIC); 19183ba7b07SChristoph Hellwig if (!work) { 192f0054bb1STejun Heo trace_writeback_nowork(wb->bdi); 193f0054bb1STejun Heo wb_wakeup(wb); 19483ba7b07SChristoph Hellwig return; 19583ba7b07SChristoph Hellwig } 19683ba7b07SChristoph Hellwig 19783ba7b07SChristoph Hellwig work->sync_mode = WB_SYNC_NONE; 19883ba7b07SChristoph Hellwig work->nr_pages = nr_pages; 19983ba7b07SChristoph Hellwig work->range_cyclic = range_cyclic; 2000e175a18SCurt Wohlgemuth work->reason = reason; 20183ba7b07SChristoph Hellwig 202f0054bb1STejun Heo wb_queue_work(wb, work); 20303ba3782SJens Axboe } 204b6e51316SJens Axboe 205703c2708STejun Heo #ifdef CONFIG_CGROUP_WRITEBACK 206703c2708STejun Heo 207703c2708STejun Heo /** 208703c2708STejun Heo * inode_congested - test whether an inode is congested 209703c2708STejun Heo * @inode: inode to test for congestion 210703c2708STejun Heo * @cong_bits: mask of WB_[a]sync_congested bits to test 211703c2708STejun Heo * 212703c2708STejun Heo * Tests whether @inode is congested. @cong_bits is the mask of congestion 213703c2708STejun Heo * bits to test and the return value is the mask of set bits. 214703c2708STejun Heo * 215703c2708STejun Heo * If cgroup writeback is enabled for @inode, the congestion state is 216703c2708STejun Heo * determined by whether the cgwb (cgroup bdi_writeback) for the blkcg 217703c2708STejun Heo * associated with @inode is congested; otherwise, the root wb's congestion 218703c2708STejun Heo * state is used. 219703c2708STejun Heo */ 220703c2708STejun Heo int inode_congested(struct inode *inode, int cong_bits) 221703c2708STejun Heo { 222703c2708STejun Heo if (inode) { 223703c2708STejun Heo struct bdi_writeback *wb = inode_to_wb(inode); 224703c2708STejun Heo if (wb) 225703c2708STejun Heo return wb_congested(wb, cong_bits); 226703c2708STejun Heo } 227703c2708STejun Heo 228703c2708STejun Heo return wb_congested(&inode_to_bdi(inode)->wb, cong_bits); 229703c2708STejun Heo } 230703c2708STejun Heo EXPORT_SYMBOL_GPL(inode_congested); 231703c2708STejun Heo 232703c2708STejun Heo #endif /* CONFIG_CGROUP_WRITEBACK */ 233703c2708STejun Heo 234b6e51316SJens Axboe /** 235b6e51316SJens Axboe * bdi_start_writeback - start writeback 236b6e51316SJens Axboe * @bdi: the backing device to write from 237b6e51316SJens Axboe * @nr_pages: the number of pages to write 238786228abSMarcos Paulo de Souza * @reason: reason why some writeback work was initiated 239b6e51316SJens Axboe * 240b6e51316SJens Axboe * Description: 241b6e51316SJens Axboe * This does WB_SYNC_NONE opportunistic writeback. The IO is only 24225985edcSLucas De Marchi * started when this function returns, we make no guarantees on 2430e3c9a22SJens Axboe * completion. Caller need not hold sb s_umount semaphore. 244b6e51316SJens Axboe * 245b6e51316SJens Axboe */ 2460e175a18SCurt Wohlgemuth void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, 2470e175a18SCurt Wohlgemuth enum wb_reason reason) 248b6e51316SJens Axboe { 249f0054bb1STejun Heo __wb_start_writeback(&bdi->wb, nr_pages, true, reason); 250d3ddec76SWu Fengguang } 251d3ddec76SWu Fengguang 252c5444198SChristoph Hellwig /** 253c5444198SChristoph Hellwig * bdi_start_background_writeback - start background writeback 254c5444198SChristoph Hellwig * @bdi: the backing device to write from 255c5444198SChristoph Hellwig * 256c5444198SChristoph Hellwig * Description: 2576585027aSJan Kara * This makes sure WB_SYNC_NONE background writeback happens. When 2586585027aSJan Kara * this function returns, it is only guaranteed that for given BDI 2596585027aSJan Kara * some IO is happening if we are over background dirty threshold. 2606585027aSJan Kara * Caller need not hold sb s_umount semaphore. 261c5444198SChristoph Hellwig */ 262c5444198SChristoph Hellwig void bdi_start_background_writeback(struct backing_dev_info *bdi) 263c5444198SChristoph Hellwig { 2646585027aSJan Kara /* 2656585027aSJan Kara * We just wake up the flusher thread. It will perform background 2666585027aSJan Kara * writeback as soon as there is no other work to do. 2676585027aSJan Kara */ 26871927e84SWu Fengguang trace_writeback_wake_background(bdi); 269f0054bb1STejun Heo wb_wakeup(&bdi->wb); 2701da177e4SLinus Torvalds } 2711da177e4SLinus Torvalds 2721da177e4SLinus Torvalds /* 273a66979abSDave Chinner * Remove the inode from the writeback list it is on. 274a66979abSDave Chinner */ 275a66979abSDave Chinner void inode_wb_list_del(struct inode *inode) 276a66979abSDave Chinner { 27752ebea74STejun Heo struct bdi_writeback *wb = inode_to_wb(inode); 278a66979abSDave Chinner 27952ebea74STejun Heo spin_lock(&wb->list_lock); 280d6c10f1fSTejun Heo inode_wb_list_del_locked(inode, wb); 28152ebea74STejun Heo spin_unlock(&wb->list_lock); 282f758eeabSChristoph Hellwig } 283a66979abSDave Chinner 284a66979abSDave Chinner /* 2856610a0bcSAndrew Morton * Redirty an inode: set its when-it-was dirtied timestamp and move it to the 2866610a0bcSAndrew Morton * furthest end of its superblock's dirty-inode list. 2876610a0bcSAndrew Morton * 2886610a0bcSAndrew Morton * Before stamping the inode's ->dirtied_when, we check to see whether it is 28966f3b8e2SJens Axboe * already the most-recently-dirtied inode on the b_dirty list. If that is 2906610a0bcSAndrew Morton * the case then the inode must have been redirtied while it was being written 2916610a0bcSAndrew Morton * out and we don't reset its dirtied_when. 2926610a0bcSAndrew Morton */ 293f758eeabSChristoph Hellwig static void redirty_tail(struct inode *inode, struct bdi_writeback *wb) 2946610a0bcSAndrew Morton { 29503ba3782SJens Axboe if (!list_empty(&wb->b_dirty)) { 29666f3b8e2SJens Axboe struct inode *tail; 2976610a0bcSAndrew Morton 2987ccf19a8SNick Piggin tail = wb_inode(wb->b_dirty.next); 29966f3b8e2SJens Axboe if (time_before(inode->dirtied_when, tail->dirtied_when)) 3006610a0bcSAndrew Morton inode->dirtied_when = jiffies; 3016610a0bcSAndrew Morton } 302d6c10f1fSTejun Heo inode_wb_list_move_locked(inode, wb, &wb->b_dirty); 3036610a0bcSAndrew Morton } 3046610a0bcSAndrew Morton 3056610a0bcSAndrew Morton /* 30666f3b8e2SJens Axboe * requeue inode for re-scanning after bdi->b_io list is exhausted. 307c986d1e2SAndrew Morton */ 308f758eeabSChristoph Hellwig static void requeue_io(struct inode *inode, struct bdi_writeback *wb) 309c986d1e2SAndrew Morton { 310d6c10f1fSTejun Heo inode_wb_list_move_locked(inode, wb, &wb->b_more_io); 311c986d1e2SAndrew Morton } 312c986d1e2SAndrew Morton 3131c0eeaf5SJoern Engel static void inode_sync_complete(struct inode *inode) 3141c0eeaf5SJoern Engel { 315365b94aeSJan Kara inode->i_state &= ~I_SYNC; 3164eff96ddSJan Kara /* If inode is clean an unused, put it into LRU now... */ 3174eff96ddSJan Kara inode_add_lru(inode); 318365b94aeSJan Kara /* Waiters must see I_SYNC cleared before being woken up */ 3191c0eeaf5SJoern Engel smp_mb(); 3201c0eeaf5SJoern Engel wake_up_bit(&inode->i_state, __I_SYNC); 3211c0eeaf5SJoern Engel } 3221c0eeaf5SJoern Engel 323d2caa3c5SJeff Layton static bool inode_dirtied_after(struct inode *inode, unsigned long t) 324d2caa3c5SJeff Layton { 325d2caa3c5SJeff Layton bool ret = time_after(inode->dirtied_when, t); 326d2caa3c5SJeff Layton #ifndef CONFIG_64BIT 327d2caa3c5SJeff Layton /* 328d2caa3c5SJeff Layton * For inodes being constantly redirtied, dirtied_when can get stuck. 329d2caa3c5SJeff Layton * It _appears_ to be in the future, but is actually in distant past. 330d2caa3c5SJeff Layton * This test is necessary to prevent such wrapped-around relative times 3315b0830cbSJens Axboe * from permanently stopping the whole bdi writeback. 332d2caa3c5SJeff Layton */ 333d2caa3c5SJeff Layton ret = ret && time_before_eq(inode->dirtied_when, jiffies); 334d2caa3c5SJeff Layton #endif 335d2caa3c5SJeff Layton return ret; 336d2caa3c5SJeff Layton } 337d2caa3c5SJeff Layton 3380ae45f63STheodore Ts'o #define EXPIRE_DIRTY_ATIME 0x0001 3390ae45f63STheodore Ts'o 340c986d1e2SAndrew Morton /* 3410e2f2b23SWang Sheng-Hui * Move expired (dirtied before work->older_than_this) dirty inodes from 342697e6fedSJan Kara * @delaying_queue to @dispatch_queue. 3432c136579SFengguang Wu */ 344e84d0a4fSWu Fengguang static int move_expired_inodes(struct list_head *delaying_queue, 3452c136579SFengguang Wu struct list_head *dispatch_queue, 3460ae45f63STheodore Ts'o int flags, 347ad4e38ddSCurt Wohlgemuth struct wb_writeback_work *work) 3482c136579SFengguang Wu { 3490ae45f63STheodore Ts'o unsigned long *older_than_this = NULL; 3500ae45f63STheodore Ts'o unsigned long expire_time; 3515c03449dSShaohua Li LIST_HEAD(tmp); 3525c03449dSShaohua Li struct list_head *pos, *node; 353cf137307SJens Axboe struct super_block *sb = NULL; 3545c03449dSShaohua Li struct inode *inode; 355cf137307SJens Axboe int do_sb_sort = 0; 356e84d0a4fSWu Fengguang int moved = 0; 3575c03449dSShaohua Li 3580ae45f63STheodore Ts'o if ((flags & EXPIRE_DIRTY_ATIME) == 0) 3590ae45f63STheodore Ts'o older_than_this = work->older_than_this; 360a2f48706STheodore Ts'o else if (!work->for_sync) { 361a2f48706STheodore Ts'o expire_time = jiffies - (dirtytime_expire_interval * HZ); 3620ae45f63STheodore Ts'o older_than_this = &expire_time; 3630ae45f63STheodore Ts'o } 3642c136579SFengguang Wu while (!list_empty(delaying_queue)) { 3657ccf19a8SNick Piggin inode = wb_inode(delaying_queue->prev); 3660ae45f63STheodore Ts'o if (older_than_this && 3670ae45f63STheodore Ts'o inode_dirtied_after(inode, *older_than_this)) 3682c136579SFengguang Wu break; 369a8855990SJan Kara list_move(&inode->i_wb_list, &tmp); 370a8855990SJan Kara moved++; 3710ae45f63STheodore Ts'o if (flags & EXPIRE_DIRTY_ATIME) 3720ae45f63STheodore Ts'o set_bit(__I_DIRTY_TIME_EXPIRED, &inode->i_state); 373a8855990SJan Kara if (sb_is_blkdev_sb(inode->i_sb)) 374a8855990SJan Kara continue; 375cf137307SJens Axboe if (sb && sb != inode->i_sb) 376cf137307SJens Axboe do_sb_sort = 1; 377cf137307SJens Axboe sb = inode->i_sb; 3785c03449dSShaohua Li } 3795c03449dSShaohua Li 380cf137307SJens Axboe /* just one sb in list, splice to dispatch_queue and we're done */ 381cf137307SJens Axboe if (!do_sb_sort) { 382cf137307SJens Axboe list_splice(&tmp, dispatch_queue); 383e84d0a4fSWu Fengguang goto out; 384cf137307SJens Axboe } 385cf137307SJens Axboe 3865c03449dSShaohua Li /* Move inodes from one superblock together */ 3875c03449dSShaohua Li while (!list_empty(&tmp)) { 3887ccf19a8SNick Piggin sb = wb_inode(tmp.prev)->i_sb; 3895c03449dSShaohua Li list_for_each_prev_safe(pos, node, &tmp) { 3907ccf19a8SNick Piggin inode = wb_inode(pos); 3915c03449dSShaohua Li if (inode->i_sb == sb) 3927ccf19a8SNick Piggin list_move(&inode->i_wb_list, dispatch_queue); 3932c136579SFengguang Wu } 3942c136579SFengguang Wu } 395e84d0a4fSWu Fengguang out: 396e84d0a4fSWu Fengguang return moved; 3975c03449dSShaohua Li } 3982c136579SFengguang Wu 3992c136579SFengguang Wu /* 4002c136579SFengguang Wu * Queue all expired dirty inodes for io, eldest first. 4014ea879b9SWu Fengguang * Before 4024ea879b9SWu Fengguang * newly dirtied b_dirty b_io b_more_io 4034ea879b9SWu Fengguang * =============> gf edc BA 4044ea879b9SWu Fengguang * After 4054ea879b9SWu Fengguang * newly dirtied b_dirty b_io b_more_io 4064ea879b9SWu Fengguang * =============> g fBAedc 4074ea879b9SWu Fengguang * | 4084ea879b9SWu Fengguang * +--> dequeue for IO 4092c136579SFengguang Wu */ 410ad4e38ddSCurt Wohlgemuth static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work) 4112c136579SFengguang Wu { 412e84d0a4fSWu Fengguang int moved; 4130ae45f63STheodore Ts'o 414f758eeabSChristoph Hellwig assert_spin_locked(&wb->list_lock); 4154ea879b9SWu Fengguang list_splice_init(&wb->b_more_io, &wb->b_io); 4160ae45f63STheodore Ts'o moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, 0, work); 4170ae45f63STheodore Ts'o moved += move_expired_inodes(&wb->b_dirty_time, &wb->b_io, 4180ae45f63STheodore Ts'o EXPIRE_DIRTY_ATIME, work); 419d6c10f1fSTejun Heo if (moved) 420d6c10f1fSTejun Heo wb_io_lists_populated(wb); 421ad4e38ddSCurt Wohlgemuth trace_writeback_queue_io(wb, work, moved); 42266f3b8e2SJens Axboe } 42366f3b8e2SJens Axboe 424a9185b41SChristoph Hellwig static int write_inode(struct inode *inode, struct writeback_control *wbc) 42566f3b8e2SJens Axboe { 4269fb0a7daSTejun Heo int ret; 4279fb0a7daSTejun Heo 4289fb0a7daSTejun Heo if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) { 4299fb0a7daSTejun Heo trace_writeback_write_inode_start(inode, wbc); 4309fb0a7daSTejun Heo ret = inode->i_sb->s_op->write_inode(inode, wbc); 4319fb0a7daSTejun Heo trace_writeback_write_inode(inode, wbc); 4329fb0a7daSTejun Heo return ret; 4339fb0a7daSTejun Heo } 43403ba3782SJens Axboe return 0; 43566f3b8e2SJens Axboe } 43608d8e974SFengguang Wu 4372c136579SFengguang Wu /* 438169ebd90SJan Kara * Wait for writeback on an inode to complete. Called with i_lock held. 439169ebd90SJan Kara * Caller must make sure inode cannot go away when we drop i_lock. 44001c03194SChristoph Hellwig */ 441169ebd90SJan Kara static void __inode_wait_for_writeback(struct inode *inode) 442169ebd90SJan Kara __releases(inode->i_lock) 443169ebd90SJan Kara __acquires(inode->i_lock) 44401c03194SChristoph Hellwig { 44501c03194SChristoph Hellwig DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); 44601c03194SChristoph Hellwig wait_queue_head_t *wqh; 44701c03194SChristoph Hellwig 44801c03194SChristoph Hellwig wqh = bit_waitqueue(&inode->i_state, __I_SYNC); 44958a9d3d8SRichard Kennedy while (inode->i_state & I_SYNC) { 450250df6edSDave Chinner spin_unlock(&inode->i_lock); 45174316201SNeilBrown __wait_on_bit(wqh, &wq, bit_wait, 45274316201SNeilBrown TASK_UNINTERRUPTIBLE); 453250df6edSDave Chinner spin_lock(&inode->i_lock); 45458a9d3d8SRichard Kennedy } 45501c03194SChristoph Hellwig } 45601c03194SChristoph Hellwig 45701c03194SChristoph Hellwig /* 458169ebd90SJan Kara * Wait for writeback on an inode to complete. Caller must have inode pinned. 459169ebd90SJan Kara */ 460169ebd90SJan Kara void inode_wait_for_writeback(struct inode *inode) 461169ebd90SJan Kara { 462169ebd90SJan Kara spin_lock(&inode->i_lock); 463169ebd90SJan Kara __inode_wait_for_writeback(inode); 464169ebd90SJan Kara spin_unlock(&inode->i_lock); 465169ebd90SJan Kara } 466169ebd90SJan Kara 467169ebd90SJan Kara /* 468169ebd90SJan Kara * Sleep until I_SYNC is cleared. This function must be called with i_lock 469169ebd90SJan Kara * held and drops it. It is aimed for callers not holding any inode reference 470169ebd90SJan Kara * so once i_lock is dropped, inode can go away. 471169ebd90SJan Kara */ 472169ebd90SJan Kara static void inode_sleep_on_writeback(struct inode *inode) 473169ebd90SJan Kara __releases(inode->i_lock) 474169ebd90SJan Kara { 475169ebd90SJan Kara DEFINE_WAIT(wait); 476169ebd90SJan Kara wait_queue_head_t *wqh = bit_waitqueue(&inode->i_state, __I_SYNC); 477169ebd90SJan Kara int sleep; 478169ebd90SJan Kara 479169ebd90SJan Kara prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); 480169ebd90SJan Kara sleep = inode->i_state & I_SYNC; 481169ebd90SJan Kara spin_unlock(&inode->i_lock); 482169ebd90SJan Kara if (sleep) 483169ebd90SJan Kara schedule(); 484169ebd90SJan Kara finish_wait(wqh, &wait); 485169ebd90SJan Kara } 486169ebd90SJan Kara 487169ebd90SJan Kara /* 488ccb26b5aSJan Kara * Find proper writeback list for the inode depending on its current state and 489ccb26b5aSJan Kara * possibly also change of its state while we were doing writeback. Here we 490ccb26b5aSJan Kara * handle things such as livelock prevention or fairness of writeback among 491ccb26b5aSJan Kara * inodes. This function can be called only by flusher thread - noone else 492ccb26b5aSJan Kara * processes all inodes in writeback lists and requeueing inodes behind flusher 493ccb26b5aSJan Kara * thread's back can have unexpected consequences. 494ccb26b5aSJan Kara */ 495ccb26b5aSJan Kara static void requeue_inode(struct inode *inode, struct bdi_writeback *wb, 496ccb26b5aSJan Kara struct writeback_control *wbc) 497ccb26b5aSJan Kara { 498ccb26b5aSJan Kara if (inode->i_state & I_FREEING) 499ccb26b5aSJan Kara return; 500ccb26b5aSJan Kara 501ccb26b5aSJan Kara /* 502ccb26b5aSJan Kara * Sync livelock prevention. Each inode is tagged and synced in one 503ccb26b5aSJan Kara * shot. If still dirty, it will be redirty_tail()'ed below. Update 504ccb26b5aSJan Kara * the dirty time to prevent enqueue and sync it again. 505ccb26b5aSJan Kara */ 506ccb26b5aSJan Kara if ((inode->i_state & I_DIRTY) && 507ccb26b5aSJan Kara (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)) 508ccb26b5aSJan Kara inode->dirtied_when = jiffies; 509ccb26b5aSJan Kara 5104f8ad655SJan Kara if (wbc->pages_skipped) { 5114f8ad655SJan Kara /* 5124f8ad655SJan Kara * writeback is not making progress due to locked 5134f8ad655SJan Kara * buffers. Skip this inode for now. 5144f8ad655SJan Kara */ 5154f8ad655SJan Kara redirty_tail(inode, wb); 5164f8ad655SJan Kara return; 5174f8ad655SJan Kara } 5184f8ad655SJan Kara 519ccb26b5aSJan Kara if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) { 520ccb26b5aSJan Kara /* 521ccb26b5aSJan Kara * We didn't write back all the pages. nfs_writepages() 522ccb26b5aSJan Kara * sometimes bales out without doing anything. 523ccb26b5aSJan Kara */ 524ccb26b5aSJan Kara if (wbc->nr_to_write <= 0) { 525ccb26b5aSJan Kara /* Slice used up. Queue for next turn. */ 526ccb26b5aSJan Kara requeue_io(inode, wb); 527ccb26b5aSJan Kara } else { 528ccb26b5aSJan Kara /* 529ccb26b5aSJan Kara * Writeback blocked by something other than 530ccb26b5aSJan Kara * congestion. Delay the inode for some time to 531ccb26b5aSJan Kara * avoid spinning on the CPU (100% iowait) 532ccb26b5aSJan Kara * retrying writeback of the dirty page/inode 533ccb26b5aSJan Kara * that cannot be performed immediately. 534ccb26b5aSJan Kara */ 535ccb26b5aSJan Kara redirty_tail(inode, wb); 536ccb26b5aSJan Kara } 537ccb26b5aSJan Kara } else if (inode->i_state & I_DIRTY) { 538ccb26b5aSJan Kara /* 539ccb26b5aSJan Kara * Filesystems can dirty the inode during writeback operations, 540ccb26b5aSJan Kara * such as delayed allocation during submission or metadata 541ccb26b5aSJan Kara * updates after data IO completion. 542ccb26b5aSJan Kara */ 543ccb26b5aSJan Kara redirty_tail(inode, wb); 5440ae45f63STheodore Ts'o } else if (inode->i_state & I_DIRTY_TIME) { 545a2f48706STheodore Ts'o inode->dirtied_when = jiffies; 546d6c10f1fSTejun Heo inode_wb_list_move_locked(inode, wb, &wb->b_dirty_time); 547ccb26b5aSJan Kara } else { 548ccb26b5aSJan Kara /* The inode is clean. Remove from writeback lists. */ 549d6c10f1fSTejun Heo inode_wb_list_del_locked(inode, wb); 550ccb26b5aSJan Kara } 551ccb26b5aSJan Kara } 552ccb26b5aSJan Kara 553ccb26b5aSJan Kara /* 5544f8ad655SJan Kara * Write out an inode and its dirty pages. Do not update the writeback list 5554f8ad655SJan Kara * linkage. That is left to the caller. The caller is also responsible for 5564f8ad655SJan Kara * setting I_SYNC flag and calling inode_sync_complete() to clear it. 5571da177e4SLinus Torvalds */ 5581da177e4SLinus Torvalds static int 559cd8ed2a4SYan Hong __writeback_single_inode(struct inode *inode, struct writeback_control *wbc) 5601da177e4SLinus Torvalds { 5611da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 562251d6a47SWu Fengguang long nr_to_write = wbc->nr_to_write; 56301c03194SChristoph Hellwig unsigned dirty; 5641da177e4SLinus Torvalds int ret; 5651da177e4SLinus Torvalds 5664f8ad655SJan Kara WARN_ON(!(inode->i_state & I_SYNC)); 5671da177e4SLinus Torvalds 5689fb0a7daSTejun Heo trace_writeback_single_inode_start(inode, wbc, nr_to_write); 5699fb0a7daSTejun Heo 5701da177e4SLinus Torvalds ret = do_writepages(mapping, wbc); 5711da177e4SLinus Torvalds 57226821ed4SChristoph Hellwig /* 57326821ed4SChristoph Hellwig * Make sure to wait on the data before writing out the metadata. 57426821ed4SChristoph Hellwig * This is important for filesystems that modify metadata on data 5757747bd4bSDave Chinner * I/O completion. We don't do it for sync(2) writeback because it has a 5767747bd4bSDave Chinner * separate, external IO completion path and ->sync_fs for guaranteeing 5777747bd4bSDave Chinner * inode metadata is written back correctly. 57826821ed4SChristoph Hellwig */ 5797747bd4bSDave Chinner if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) { 58026821ed4SChristoph Hellwig int err = filemap_fdatawait(mapping); 5811da177e4SLinus Torvalds if (ret == 0) 5821da177e4SLinus Torvalds ret = err; 5831da177e4SLinus Torvalds } 5841da177e4SLinus Torvalds 5855547e8aaSDmitry Monakhov /* 5865547e8aaSDmitry Monakhov * Some filesystems may redirty the inode during the writeback 5875547e8aaSDmitry Monakhov * due to delalloc, clear dirty metadata flags right before 5885547e8aaSDmitry Monakhov * write_inode() 5895547e8aaSDmitry Monakhov */ 590250df6edSDave Chinner spin_lock(&inode->i_lock); 5919c6ac78eSTejun Heo 5925547e8aaSDmitry Monakhov dirty = inode->i_state & I_DIRTY; 593a2f48706STheodore Ts'o if (inode->i_state & I_DIRTY_TIME) { 594a2f48706STheodore Ts'o if ((dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) || 595a2f48706STheodore Ts'o unlikely(inode->i_state & I_DIRTY_TIME_EXPIRED) || 596a2f48706STheodore Ts'o unlikely(time_after(jiffies, 597a2f48706STheodore Ts'o (inode->dirtied_time_when + 598a2f48706STheodore Ts'o dirtytime_expire_interval * HZ)))) { 5990ae45f63STheodore Ts'o dirty |= I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED; 6000ae45f63STheodore Ts'o trace_writeback_lazytime(inode); 6010ae45f63STheodore Ts'o } 602a2f48706STheodore Ts'o } else 603a2f48706STheodore Ts'o inode->i_state &= ~I_DIRTY_TIME_EXPIRED; 6040ae45f63STheodore Ts'o inode->i_state &= ~dirty; 6059c6ac78eSTejun Heo 6069c6ac78eSTejun Heo /* 6079c6ac78eSTejun Heo * Paired with smp_mb() in __mark_inode_dirty(). This allows 6089c6ac78eSTejun Heo * __mark_inode_dirty() to test i_state without grabbing i_lock - 6099c6ac78eSTejun Heo * either they see the I_DIRTY bits cleared or we see the dirtied 6109c6ac78eSTejun Heo * inode. 6119c6ac78eSTejun Heo * 6129c6ac78eSTejun Heo * I_DIRTY_PAGES is always cleared together above even if @mapping 6139c6ac78eSTejun Heo * still has dirty pages. The flag is reinstated after smp_mb() if 6149c6ac78eSTejun Heo * necessary. This guarantees that either __mark_inode_dirty() 6159c6ac78eSTejun Heo * sees clear I_DIRTY_PAGES or we see PAGECACHE_TAG_DIRTY. 6169c6ac78eSTejun Heo */ 6179c6ac78eSTejun Heo smp_mb(); 6189c6ac78eSTejun Heo 6199c6ac78eSTejun Heo if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) 6209c6ac78eSTejun Heo inode->i_state |= I_DIRTY_PAGES; 6219c6ac78eSTejun Heo 622250df6edSDave Chinner spin_unlock(&inode->i_lock); 6239c6ac78eSTejun Heo 6240ae45f63STheodore Ts'o if (dirty & I_DIRTY_TIME) 6250ae45f63STheodore Ts'o mark_inode_dirty_sync(inode); 62626821ed4SChristoph Hellwig /* Don't write the inode if only I_DIRTY_PAGES was set */ 6270ae45f63STheodore Ts'o if (dirty & ~I_DIRTY_PAGES) { 628a9185b41SChristoph Hellwig int err = write_inode(inode, wbc); 6291da177e4SLinus Torvalds if (ret == 0) 6301da177e4SLinus Torvalds ret = err; 6311da177e4SLinus Torvalds } 6324f8ad655SJan Kara trace_writeback_single_inode(inode, wbc, nr_to_write); 6334f8ad655SJan Kara return ret; 6344f8ad655SJan Kara } 6354f8ad655SJan Kara 6364f8ad655SJan Kara /* 6374f8ad655SJan Kara * Write out an inode's dirty pages. Either the caller has an active reference 6384f8ad655SJan Kara * on the inode or the inode has I_WILL_FREE set. 6394f8ad655SJan Kara * 6404f8ad655SJan Kara * This function is designed to be called for writing back one inode which 6414f8ad655SJan Kara * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode() 6424f8ad655SJan Kara * and does more profound writeback list handling in writeback_sb_inodes(). 6434f8ad655SJan Kara */ 6444f8ad655SJan Kara static int 6454f8ad655SJan Kara writeback_single_inode(struct inode *inode, struct bdi_writeback *wb, 6464f8ad655SJan Kara struct writeback_control *wbc) 6474f8ad655SJan Kara { 6484f8ad655SJan Kara int ret = 0; 6494f8ad655SJan Kara 6504f8ad655SJan Kara spin_lock(&inode->i_lock); 6514f8ad655SJan Kara if (!atomic_read(&inode->i_count)) 6524f8ad655SJan Kara WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); 6534f8ad655SJan Kara else 6544f8ad655SJan Kara WARN_ON(inode->i_state & I_WILL_FREE); 6554f8ad655SJan Kara 6564f8ad655SJan Kara if (inode->i_state & I_SYNC) { 6574f8ad655SJan Kara if (wbc->sync_mode != WB_SYNC_ALL) 6584f8ad655SJan Kara goto out; 6594f8ad655SJan Kara /* 660169ebd90SJan Kara * It's a data-integrity sync. We must wait. Since callers hold 661169ebd90SJan Kara * inode reference or inode has I_WILL_FREE set, it cannot go 662169ebd90SJan Kara * away under us. 6634f8ad655SJan Kara */ 664169ebd90SJan Kara __inode_wait_for_writeback(inode); 6654f8ad655SJan Kara } 6664f8ad655SJan Kara WARN_ON(inode->i_state & I_SYNC); 6674f8ad655SJan Kara /* 668f9b0e058SJan Kara * Skip inode if it is clean and we have no outstanding writeback in 669f9b0e058SJan Kara * WB_SYNC_ALL mode. We don't want to mess with writeback lists in this 670f9b0e058SJan Kara * function since flusher thread may be doing for example sync in 671f9b0e058SJan Kara * parallel and if we move the inode, it could get skipped. So here we 672f9b0e058SJan Kara * make sure inode is on some writeback list and leave it there unless 673f9b0e058SJan Kara * we have completely cleaned the inode. 6744f8ad655SJan Kara */ 6750ae45f63STheodore Ts'o if (!(inode->i_state & I_DIRTY_ALL) && 676f9b0e058SJan Kara (wbc->sync_mode != WB_SYNC_ALL || 677f9b0e058SJan Kara !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK))) 6784f8ad655SJan Kara goto out; 6794f8ad655SJan Kara inode->i_state |= I_SYNC; 6804f8ad655SJan Kara spin_unlock(&inode->i_lock); 6814f8ad655SJan Kara 682cd8ed2a4SYan Hong ret = __writeback_single_inode(inode, wbc); 6831da177e4SLinus Torvalds 684f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 685250df6edSDave Chinner spin_lock(&inode->i_lock); 6864f8ad655SJan Kara /* 6874f8ad655SJan Kara * If inode is clean, remove it from writeback lists. Otherwise don't 6884f8ad655SJan Kara * touch it. See comment above for explanation. 6894f8ad655SJan Kara */ 6900ae45f63STheodore Ts'o if (!(inode->i_state & I_DIRTY_ALL)) 691d6c10f1fSTejun Heo inode_wb_list_del_locked(inode, wb); 6924f8ad655SJan Kara spin_unlock(&wb->list_lock); 6931c0eeaf5SJoern Engel inode_sync_complete(inode); 6944f8ad655SJan Kara out: 6954f8ad655SJan Kara spin_unlock(&inode->i_lock); 6961da177e4SLinus Torvalds return ret; 6971da177e4SLinus Torvalds } 6981da177e4SLinus Torvalds 699a88a341aSTejun Heo static long writeback_chunk_size(struct bdi_writeback *wb, 7001a12d8bdSWu Fengguang struct wb_writeback_work *work) 701d46db3d5SWu Fengguang { 702d46db3d5SWu Fengguang long pages; 703d46db3d5SWu Fengguang 704d46db3d5SWu Fengguang /* 705d46db3d5SWu Fengguang * WB_SYNC_ALL mode does livelock avoidance by syncing dirty 706d46db3d5SWu Fengguang * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX 707d46db3d5SWu Fengguang * here avoids calling into writeback_inodes_wb() more than once. 708d46db3d5SWu Fengguang * 709d46db3d5SWu Fengguang * The intended call sequence for WB_SYNC_ALL writeback is: 710d46db3d5SWu Fengguang * 711d46db3d5SWu Fengguang * wb_writeback() 712d46db3d5SWu Fengguang * writeback_sb_inodes() <== called only once 713d46db3d5SWu Fengguang * write_cache_pages() <== called once for each inode 714d46db3d5SWu Fengguang * (quickly) tag currently dirty pages 715d46db3d5SWu Fengguang * (maybe slowly) sync all tagged pages 716d46db3d5SWu Fengguang */ 717d46db3d5SWu Fengguang if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages) 718d46db3d5SWu Fengguang pages = LONG_MAX; 7191a12d8bdSWu Fengguang else { 720a88a341aSTejun Heo pages = min(wb->avg_write_bandwidth / 2, 7211a12d8bdSWu Fengguang global_dirty_limit / DIRTY_SCOPE); 7221a12d8bdSWu Fengguang pages = min(pages, work->nr_pages); 7231a12d8bdSWu Fengguang pages = round_down(pages + MIN_WRITEBACK_PAGES, 7241a12d8bdSWu Fengguang MIN_WRITEBACK_PAGES); 7251a12d8bdSWu Fengguang } 726d46db3d5SWu Fengguang 727d46db3d5SWu Fengguang return pages; 728d46db3d5SWu Fengguang } 729d46db3d5SWu Fengguang 73003ba3782SJens Axboe /* 731f11c9c5cSEdward Shishkin * Write a portion of b_io inodes which belong to @sb. 732edadfb10SChristoph Hellwig * 733d46db3d5SWu Fengguang * Return the number of pages and/or inodes written. 734f11c9c5cSEdward Shishkin */ 735d46db3d5SWu Fengguang static long writeback_sb_inodes(struct super_block *sb, 736d46db3d5SWu Fengguang struct bdi_writeback *wb, 737d46db3d5SWu Fengguang struct wb_writeback_work *work) 73803ba3782SJens Axboe { 739d46db3d5SWu Fengguang struct writeback_control wbc = { 740d46db3d5SWu Fengguang .sync_mode = work->sync_mode, 741d46db3d5SWu Fengguang .tagged_writepages = work->tagged_writepages, 742d46db3d5SWu Fengguang .for_kupdate = work->for_kupdate, 743d46db3d5SWu Fengguang .for_background = work->for_background, 7447747bd4bSDave Chinner .for_sync = work->for_sync, 745d46db3d5SWu Fengguang .range_cyclic = work->range_cyclic, 746d46db3d5SWu Fengguang .range_start = 0, 747d46db3d5SWu Fengguang .range_end = LLONG_MAX, 748d46db3d5SWu Fengguang }; 749d46db3d5SWu Fengguang unsigned long start_time = jiffies; 750d46db3d5SWu Fengguang long write_chunk; 751d46db3d5SWu Fengguang long wrote = 0; /* count both pages and inodes */ 752d46db3d5SWu Fengguang 75303ba3782SJens Axboe while (!list_empty(&wb->b_io)) { 7547ccf19a8SNick Piggin struct inode *inode = wb_inode(wb->b_io.prev); 755edadfb10SChristoph Hellwig 756edadfb10SChristoph Hellwig if (inode->i_sb != sb) { 757d46db3d5SWu Fengguang if (work->sb) { 758edadfb10SChristoph Hellwig /* 759edadfb10SChristoph Hellwig * We only want to write back data for this 760edadfb10SChristoph Hellwig * superblock, move all inodes not belonging 761edadfb10SChristoph Hellwig * to it back onto the dirty list. 762edadfb10SChristoph Hellwig */ 763f758eeabSChristoph Hellwig redirty_tail(inode, wb); 76466f3b8e2SJens Axboe continue; 76566f3b8e2SJens Axboe } 766edadfb10SChristoph Hellwig 767edadfb10SChristoph Hellwig /* 768edadfb10SChristoph Hellwig * The inode belongs to a different superblock. 769edadfb10SChristoph Hellwig * Bounce back to the caller to unpin this and 770edadfb10SChristoph Hellwig * pin the next superblock. 771edadfb10SChristoph Hellwig */ 772d46db3d5SWu Fengguang break; 773edadfb10SChristoph Hellwig } 774edadfb10SChristoph Hellwig 7759843b76aSChristoph Hellwig /* 776331cbdeeSWanpeng Li * Don't bother with new inodes or inodes being freed, first 777331cbdeeSWanpeng Li * kind does not need periodic writeout yet, and for the latter 7789843b76aSChristoph Hellwig * kind writeout is handled by the freer. 7799843b76aSChristoph Hellwig */ 780250df6edSDave Chinner spin_lock(&inode->i_lock); 7819843b76aSChristoph Hellwig if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { 782250df6edSDave Chinner spin_unlock(&inode->i_lock); 783fcc5c222SWu Fengguang redirty_tail(inode, wb); 7847ef0d737SNick Piggin continue; 7857ef0d737SNick Piggin } 786cc1676d9SJan Kara if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) { 787cc1676d9SJan Kara /* 788cc1676d9SJan Kara * If this inode is locked for writeback and we are not 789cc1676d9SJan Kara * doing writeback-for-data-integrity, move it to 790cc1676d9SJan Kara * b_more_io so that writeback can proceed with the 791cc1676d9SJan Kara * other inodes on s_io. 792cc1676d9SJan Kara * 793cc1676d9SJan Kara * We'll have another go at writing back this inode 794cc1676d9SJan Kara * when we completed a full scan of b_io. 795cc1676d9SJan Kara */ 796cc1676d9SJan Kara spin_unlock(&inode->i_lock); 797cc1676d9SJan Kara requeue_io(inode, wb); 798cc1676d9SJan Kara trace_writeback_sb_inodes_requeue(inode); 799cc1676d9SJan Kara continue; 800cc1676d9SJan Kara } 801f0d07b7fSJan Kara spin_unlock(&wb->list_lock); 802f0d07b7fSJan Kara 8034f8ad655SJan Kara /* 8044f8ad655SJan Kara * We already requeued the inode if it had I_SYNC set and we 8054f8ad655SJan Kara * are doing WB_SYNC_NONE writeback. So this catches only the 8064f8ad655SJan Kara * WB_SYNC_ALL case. 8074f8ad655SJan Kara */ 808169ebd90SJan Kara if (inode->i_state & I_SYNC) { 809169ebd90SJan Kara /* Wait for I_SYNC. This function drops i_lock... */ 810169ebd90SJan Kara inode_sleep_on_writeback(inode); 811169ebd90SJan Kara /* Inode may be gone, start again */ 812ead188f9SJan Kara spin_lock(&wb->list_lock); 813169ebd90SJan Kara continue; 814169ebd90SJan Kara } 8154f8ad655SJan Kara inode->i_state |= I_SYNC; 8164f8ad655SJan Kara spin_unlock(&inode->i_lock); 817169ebd90SJan Kara 818a88a341aSTejun Heo write_chunk = writeback_chunk_size(wb, work); 819d46db3d5SWu Fengguang wbc.nr_to_write = write_chunk; 820d46db3d5SWu Fengguang wbc.pages_skipped = 0; 821250df6edSDave Chinner 822169ebd90SJan Kara /* 823169ebd90SJan Kara * We use I_SYNC to pin the inode in memory. While it is set 824169ebd90SJan Kara * evict_inode() will wait so the inode cannot be freed. 825169ebd90SJan Kara */ 826cd8ed2a4SYan Hong __writeback_single_inode(inode, &wbc); 827d46db3d5SWu Fengguang 828d46db3d5SWu Fengguang work->nr_pages -= write_chunk - wbc.nr_to_write; 829d46db3d5SWu Fengguang wrote += write_chunk - wbc.nr_to_write; 8304f8ad655SJan Kara spin_lock(&wb->list_lock); 8314f8ad655SJan Kara spin_lock(&inode->i_lock); 8320ae45f63STheodore Ts'o if (!(inode->i_state & I_DIRTY_ALL)) 833d46db3d5SWu Fengguang wrote++; 8344f8ad655SJan Kara requeue_inode(inode, wb, &wbc); 8354f8ad655SJan Kara inode_sync_complete(inode); 8360f1b1fd8SDave Chinner spin_unlock(&inode->i_lock); 837169ebd90SJan Kara cond_resched_lock(&wb->list_lock); 838d46db3d5SWu Fengguang /* 839d46db3d5SWu Fengguang * bail out to wb_writeback() often enough to check 840d46db3d5SWu Fengguang * background threshold and other termination conditions. 841d46db3d5SWu Fengguang */ 842d46db3d5SWu Fengguang if (wrote) { 843d46db3d5SWu Fengguang if (time_is_before_jiffies(start_time + HZ / 10UL)) 844d46db3d5SWu Fengguang break; 845d46db3d5SWu Fengguang if (work->nr_pages <= 0) 846d46db3d5SWu Fengguang break; 8471da177e4SLinus Torvalds } 8488bc3be27SFengguang Wu } 849d46db3d5SWu Fengguang return wrote; 850f11c9c5cSEdward Shishkin } 85138f21977SNick Piggin 852d46db3d5SWu Fengguang static long __writeback_inodes_wb(struct bdi_writeback *wb, 853d46db3d5SWu Fengguang struct wb_writeback_work *work) 854f11c9c5cSEdward Shishkin { 855d46db3d5SWu Fengguang unsigned long start_time = jiffies; 856d46db3d5SWu Fengguang long wrote = 0; 857f11c9c5cSEdward Shishkin 858f11c9c5cSEdward Shishkin while (!list_empty(&wb->b_io)) { 8597ccf19a8SNick Piggin struct inode *inode = wb_inode(wb->b_io.prev); 860f11c9c5cSEdward Shishkin struct super_block *sb = inode->i_sb; 861f11c9c5cSEdward Shishkin 862eb6ef3dfSKonstantin Khlebnikov if (!trylock_super(sb)) { 8630e995816SWu Fengguang /* 864eb6ef3dfSKonstantin Khlebnikov * trylock_super() may fail consistently due to 8650e995816SWu Fengguang * s_umount being grabbed by someone else. Don't use 8660e995816SWu Fengguang * requeue_io() to avoid busy retrying the inode/sb. 8670e995816SWu Fengguang */ 8680e995816SWu Fengguang redirty_tail(inode, wb); 869d19de7edSChristoph Hellwig continue; 870334132aeSChristoph Hellwig } 871d46db3d5SWu Fengguang wrote += writeback_sb_inodes(sb, wb, work); 872eb6ef3dfSKonstantin Khlebnikov up_read(&sb->s_umount); 873f11c9c5cSEdward Shishkin 874d46db3d5SWu Fengguang /* refer to the same tests at the end of writeback_sb_inodes */ 875d46db3d5SWu Fengguang if (wrote) { 876d46db3d5SWu Fengguang if (time_is_before_jiffies(start_time + HZ / 10UL)) 877d46db3d5SWu Fengguang break; 878d46db3d5SWu Fengguang if (work->nr_pages <= 0) 879f11c9c5cSEdward Shishkin break; 880f11c9c5cSEdward Shishkin } 881d46db3d5SWu Fengguang } 88266f3b8e2SJens Axboe /* Leave any unwritten inodes on b_io */ 883d46db3d5SWu Fengguang return wrote; 88466f3b8e2SJens Axboe } 88566f3b8e2SJens Axboe 8867d9f073bSWanpeng Li static long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages, 8870e175a18SCurt Wohlgemuth enum wb_reason reason) 888edadfb10SChristoph Hellwig { 889d46db3d5SWu Fengguang struct wb_writeback_work work = { 890d46db3d5SWu Fengguang .nr_pages = nr_pages, 891d46db3d5SWu Fengguang .sync_mode = WB_SYNC_NONE, 892d46db3d5SWu Fengguang .range_cyclic = 1, 8930e175a18SCurt Wohlgemuth .reason = reason, 894d46db3d5SWu Fengguang }; 895edadfb10SChristoph Hellwig 896f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 897424b351fSWu Fengguang if (list_empty(&wb->b_io)) 898ad4e38ddSCurt Wohlgemuth queue_io(wb, &work); 899d46db3d5SWu Fengguang __writeback_inodes_wb(wb, &work); 900f758eeabSChristoph Hellwig spin_unlock(&wb->list_lock); 901edadfb10SChristoph Hellwig 902d46db3d5SWu Fengguang return nr_pages - work.nr_pages; 90366f3b8e2SJens Axboe } 90466f3b8e2SJens Axboe 905a88a341aSTejun Heo static bool over_bground_thresh(struct bdi_writeback *wb) 90603ba3782SJens Axboe { 90703ba3782SJens Axboe unsigned long background_thresh, dirty_thresh; 90803ba3782SJens Axboe 90916c4042fSWu Fengguang global_dirty_limits(&background_thresh, &dirty_thresh); 91003ba3782SJens Axboe 911b00949aaSWu Fengguang if (global_page_state(NR_FILE_DIRTY) + 912b00949aaSWu Fengguang global_page_state(NR_UNSTABLE_NFS) > background_thresh) 913b00949aaSWu Fengguang return true; 914b00949aaSWu Fengguang 915a88a341aSTejun Heo if (wb_stat(wb, WB_RECLAIMABLE) > wb_dirty_limit(wb, background_thresh)) 916b00949aaSWu Fengguang return true; 917b00949aaSWu Fengguang 918b00949aaSWu Fengguang return false; 91903ba3782SJens Axboe } 92003ba3782SJens Axboe 92103ba3782SJens Axboe /* 922e98be2d5SWu Fengguang * Called under wb->list_lock. If there are multiple wb per bdi, 923e98be2d5SWu Fengguang * only the flusher working on the first wb should do it. 924e98be2d5SWu Fengguang */ 925e98be2d5SWu Fengguang static void wb_update_bandwidth(struct bdi_writeback *wb, 926e98be2d5SWu Fengguang unsigned long start_time) 927e98be2d5SWu Fengguang { 928a88a341aSTejun Heo __wb_update_bandwidth(wb, 0, 0, 0, 0, 0, start_time); 929e98be2d5SWu Fengguang } 930e98be2d5SWu Fengguang 931e98be2d5SWu Fengguang /* 93203ba3782SJens Axboe * Explicit flushing or periodic writeback of "old" data. 93303ba3782SJens Axboe * 93403ba3782SJens Axboe * Define "old": the first time one of an inode's pages is dirtied, we mark the 93503ba3782SJens Axboe * dirtying-time in the inode's address_space. So this periodic writeback code 93603ba3782SJens Axboe * just walks the superblock inode list, writing back any inodes which are 93703ba3782SJens Axboe * older than a specific point in time. 93803ba3782SJens Axboe * 93903ba3782SJens Axboe * Try to run once per dirty_writeback_interval. But if a writeback event 94003ba3782SJens Axboe * takes longer than a dirty_writeback_interval interval, then leave a 94103ba3782SJens Axboe * one-second gap. 94203ba3782SJens Axboe * 94303ba3782SJens Axboe * older_than_this takes precedence over nr_to_write. So we'll only write back 94403ba3782SJens Axboe * all dirty pages if they are all attached to "old" mappings. 94503ba3782SJens Axboe */ 946c4a77a6cSJens Axboe static long wb_writeback(struct bdi_writeback *wb, 94783ba7b07SChristoph Hellwig struct wb_writeback_work *work) 94803ba3782SJens Axboe { 949e98be2d5SWu Fengguang unsigned long wb_start = jiffies; 950d46db3d5SWu Fengguang long nr_pages = work->nr_pages; 9510dc83bd3SJan Kara unsigned long oldest_jif; 952a5989bdcSJan Kara struct inode *inode; 953d46db3d5SWu Fengguang long progress; 95403ba3782SJens Axboe 9550dc83bd3SJan Kara oldest_jif = jiffies; 9560dc83bd3SJan Kara work->older_than_this = &oldest_jif; 95703ba3782SJens Axboe 958e8dfc305SWu Fengguang spin_lock(&wb->list_lock); 95903ba3782SJens Axboe for (;;) { 96003ba3782SJens Axboe /* 961d3ddec76SWu Fengguang * Stop writeback when nr_pages has been consumed 96203ba3782SJens Axboe */ 96383ba7b07SChristoph Hellwig if (work->nr_pages <= 0) 96403ba3782SJens Axboe break; 96503ba3782SJens Axboe 96603ba3782SJens Axboe /* 967aa373cf5SJan Kara * Background writeout and kupdate-style writeback may 968aa373cf5SJan Kara * run forever. Stop them if there is other work to do 969aa373cf5SJan Kara * so that e.g. sync can proceed. They'll be restarted 970aa373cf5SJan Kara * after the other works are all done. 971aa373cf5SJan Kara */ 972aa373cf5SJan Kara if ((work->for_background || work->for_kupdate) && 973f0054bb1STejun Heo !list_empty(&wb->work_list)) 974aa373cf5SJan Kara break; 975aa373cf5SJan Kara 976aa373cf5SJan Kara /* 977d3ddec76SWu Fengguang * For background writeout, stop when we are below the 978d3ddec76SWu Fengguang * background dirty threshold 97903ba3782SJens Axboe */ 980a88a341aSTejun Heo if (work->for_background && !over_bground_thresh(wb)) 98103ba3782SJens Axboe break; 98203ba3782SJens Axboe 9831bc36b64SJan Kara /* 9841bc36b64SJan Kara * Kupdate and background works are special and we want to 9851bc36b64SJan Kara * include all inodes that need writing. Livelock avoidance is 9861bc36b64SJan Kara * handled by these works yielding to any other work so we are 9871bc36b64SJan Kara * safe. 9881bc36b64SJan Kara */ 989ba9aa839SWu Fengguang if (work->for_kupdate) { 9900dc83bd3SJan Kara oldest_jif = jiffies - 991ba9aa839SWu Fengguang msecs_to_jiffies(dirty_expire_interval * 10); 9921bc36b64SJan Kara } else if (work->for_background) 9930dc83bd3SJan Kara oldest_jif = jiffies; 994028c2dd1SDave Chinner 995d46db3d5SWu Fengguang trace_writeback_start(wb->bdi, work); 996e8dfc305SWu Fengguang if (list_empty(&wb->b_io)) 997ad4e38ddSCurt Wohlgemuth queue_io(wb, work); 99883ba7b07SChristoph Hellwig if (work->sb) 999d46db3d5SWu Fengguang progress = writeback_sb_inodes(work->sb, wb, work); 1000edadfb10SChristoph Hellwig else 1001d46db3d5SWu Fengguang progress = __writeback_inodes_wb(wb, work); 1002d46db3d5SWu Fengguang trace_writeback_written(wb->bdi, work); 1003028c2dd1SDave Chinner 1004e98be2d5SWu Fengguang wb_update_bandwidth(wb, wb_start); 100503ba3782SJens Axboe 100603ba3782SJens Axboe /* 100771fd05a8SJens Axboe * Did we write something? Try for more 1008e6fb6da2SWu Fengguang * 1009e6fb6da2SWu Fengguang * Dirty inodes are moved to b_io for writeback in batches. 1010e6fb6da2SWu Fengguang * The completion of the current batch does not necessarily 1011e6fb6da2SWu Fengguang * mean the overall work is done. So we keep looping as long 1012e6fb6da2SWu Fengguang * as made some progress on cleaning pages or inodes. 101371fd05a8SJens Axboe */ 1014d46db3d5SWu Fengguang if (progress) 101503ba3782SJens Axboe continue; 1016a5989bdcSJan Kara /* 1017e6fb6da2SWu Fengguang * No more inodes for IO, bail 1018a5989bdcSJan Kara */ 1019b7a2441fSWu Fengguang if (list_empty(&wb->b_more_io)) 102003ba3782SJens Axboe break; 102103ba3782SJens Axboe /* 10228010c3b6SJens Axboe * Nothing written. Wait for some inode to 10238010c3b6SJens Axboe * become available for writeback. Otherwise 10248010c3b6SJens Axboe * we'll just busyloop. 102503ba3782SJens Axboe */ 102603ba3782SJens Axboe if (!list_empty(&wb->b_more_io)) { 1027d46db3d5SWu Fengguang trace_writeback_wait(wb->bdi, work); 102803ba3782SJens Axboe inode = wb_inode(wb->b_more_io.prev); 1029250df6edSDave Chinner spin_lock(&inode->i_lock); 1030f0d07b7fSJan Kara spin_unlock(&wb->list_lock); 1031169ebd90SJan Kara /* This function drops i_lock... */ 1032169ebd90SJan Kara inode_sleep_on_writeback(inode); 1033f0d07b7fSJan Kara spin_lock(&wb->list_lock); 103403ba3782SJens Axboe } 103503ba3782SJens Axboe } 1036e8dfc305SWu Fengguang spin_unlock(&wb->list_lock); 103703ba3782SJens Axboe 1038d46db3d5SWu Fengguang return nr_pages - work->nr_pages; 103903ba3782SJens Axboe } 104003ba3782SJens Axboe 104103ba3782SJens Axboe /* 104283ba7b07SChristoph Hellwig * Return the next wb_writeback_work struct that hasn't been processed yet. 104303ba3782SJens Axboe */ 1044f0054bb1STejun Heo static struct wb_writeback_work *get_next_work_item(struct bdi_writeback *wb) 104503ba3782SJens Axboe { 104683ba7b07SChristoph Hellwig struct wb_writeback_work *work = NULL; 104703ba3782SJens Axboe 1048f0054bb1STejun Heo spin_lock_bh(&wb->work_lock); 1049f0054bb1STejun Heo if (!list_empty(&wb->work_list)) { 1050f0054bb1STejun Heo work = list_entry(wb->work_list.next, 105183ba7b07SChristoph Hellwig struct wb_writeback_work, list); 105283ba7b07SChristoph Hellwig list_del_init(&work->list); 105303ba3782SJens Axboe } 1054f0054bb1STejun Heo spin_unlock_bh(&wb->work_lock); 105583ba7b07SChristoph Hellwig return work; 105603ba3782SJens Axboe } 105703ba3782SJens Axboe 1058cdf01dd5SLinus Torvalds /* 1059cdf01dd5SLinus Torvalds * Add in the number of potentially dirty inodes, because each inode 1060cdf01dd5SLinus Torvalds * write can dirty pagecache in the underlying blockdev. 1061cdf01dd5SLinus Torvalds */ 1062cdf01dd5SLinus Torvalds static unsigned long get_nr_dirty_pages(void) 1063cdf01dd5SLinus Torvalds { 1064cdf01dd5SLinus Torvalds return global_page_state(NR_FILE_DIRTY) + 1065cdf01dd5SLinus Torvalds global_page_state(NR_UNSTABLE_NFS) + 1066cdf01dd5SLinus Torvalds get_nr_dirty_inodes(); 1067cdf01dd5SLinus Torvalds } 1068cdf01dd5SLinus Torvalds 10696585027aSJan Kara static long wb_check_background_flush(struct bdi_writeback *wb) 10706585027aSJan Kara { 1071a88a341aSTejun Heo if (over_bground_thresh(wb)) { 10726585027aSJan Kara 10736585027aSJan Kara struct wb_writeback_work work = { 10746585027aSJan Kara .nr_pages = LONG_MAX, 10756585027aSJan Kara .sync_mode = WB_SYNC_NONE, 10766585027aSJan Kara .for_background = 1, 10776585027aSJan Kara .range_cyclic = 1, 10780e175a18SCurt Wohlgemuth .reason = WB_REASON_BACKGROUND, 10796585027aSJan Kara }; 10806585027aSJan Kara 10816585027aSJan Kara return wb_writeback(wb, &work); 10826585027aSJan Kara } 10836585027aSJan Kara 10846585027aSJan Kara return 0; 10856585027aSJan Kara } 10866585027aSJan Kara 108703ba3782SJens Axboe static long wb_check_old_data_flush(struct bdi_writeback *wb) 108803ba3782SJens Axboe { 108903ba3782SJens Axboe unsigned long expired; 109003ba3782SJens Axboe long nr_pages; 109103ba3782SJens Axboe 109269b62d01SJens Axboe /* 109369b62d01SJens Axboe * When set to zero, disable periodic writeback 109469b62d01SJens Axboe */ 109569b62d01SJens Axboe if (!dirty_writeback_interval) 109669b62d01SJens Axboe return 0; 109769b62d01SJens Axboe 109803ba3782SJens Axboe expired = wb->last_old_flush + 109903ba3782SJens Axboe msecs_to_jiffies(dirty_writeback_interval * 10); 110003ba3782SJens Axboe if (time_before(jiffies, expired)) 110103ba3782SJens Axboe return 0; 110203ba3782SJens Axboe 110303ba3782SJens Axboe wb->last_old_flush = jiffies; 1104cdf01dd5SLinus Torvalds nr_pages = get_nr_dirty_pages(); 110503ba3782SJens Axboe 1106c4a77a6cSJens Axboe if (nr_pages) { 110783ba7b07SChristoph Hellwig struct wb_writeback_work work = { 1108c4a77a6cSJens Axboe .nr_pages = nr_pages, 1109c4a77a6cSJens Axboe .sync_mode = WB_SYNC_NONE, 1110c4a77a6cSJens Axboe .for_kupdate = 1, 1111c4a77a6cSJens Axboe .range_cyclic = 1, 11120e175a18SCurt Wohlgemuth .reason = WB_REASON_PERIODIC, 1113c4a77a6cSJens Axboe }; 1114c4a77a6cSJens Axboe 111583ba7b07SChristoph Hellwig return wb_writeback(wb, &work); 1116c4a77a6cSJens Axboe } 111703ba3782SJens Axboe 111803ba3782SJens Axboe return 0; 111903ba3782SJens Axboe } 112003ba3782SJens Axboe 112103ba3782SJens Axboe /* 112203ba3782SJens Axboe * Retrieve work items and do the writeback they describe 112303ba3782SJens Axboe */ 112425d130baSWanpeng Li static long wb_do_writeback(struct bdi_writeback *wb) 112503ba3782SJens Axboe { 112683ba7b07SChristoph Hellwig struct wb_writeback_work *work; 1127c4a77a6cSJens Axboe long wrote = 0; 112803ba3782SJens Axboe 11294452226eSTejun Heo set_bit(WB_writeback_running, &wb->state); 1130f0054bb1STejun Heo while ((work = get_next_work_item(wb)) != NULL) { 113183ba7b07SChristoph Hellwig 1132f0054bb1STejun Heo trace_writeback_exec(wb->bdi, work); 1133455b2864SDave Chinner 113483ba7b07SChristoph Hellwig wrote += wb_writeback(wb, work); 113503ba3782SJens Axboe 113603ba3782SJens Axboe /* 113783ba7b07SChristoph Hellwig * Notify the caller of completion if this is a synchronous 113883ba7b07SChristoph Hellwig * work item, otherwise just free it. 113903ba3782SJens Axboe */ 114083ba7b07SChristoph Hellwig if (work->done) 114183ba7b07SChristoph Hellwig complete(work->done); 114283ba7b07SChristoph Hellwig else 114383ba7b07SChristoph Hellwig kfree(work); 114403ba3782SJens Axboe } 114503ba3782SJens Axboe 114603ba3782SJens Axboe /* 114703ba3782SJens Axboe * Check for periodic writeback, kupdated() style 114803ba3782SJens Axboe */ 114903ba3782SJens Axboe wrote += wb_check_old_data_flush(wb); 11506585027aSJan Kara wrote += wb_check_background_flush(wb); 11514452226eSTejun Heo clear_bit(WB_writeback_running, &wb->state); 115203ba3782SJens Axboe 115303ba3782SJens Axboe return wrote; 115403ba3782SJens Axboe } 115503ba3782SJens Axboe 115603ba3782SJens Axboe /* 115703ba3782SJens Axboe * Handle writeback of dirty data for the device backed by this bdi. Also 1158839a8e86STejun Heo * reschedules periodically and does kupdated style flushing. 115903ba3782SJens Axboe */ 1160f0054bb1STejun Heo void wb_workfn(struct work_struct *work) 116103ba3782SJens Axboe { 1162839a8e86STejun Heo struct bdi_writeback *wb = container_of(to_delayed_work(work), 1163839a8e86STejun Heo struct bdi_writeback, dwork); 116403ba3782SJens Axboe long pages_written; 116503ba3782SJens Axboe 1166f0054bb1STejun Heo set_worker_desc("flush-%s", dev_name(wb->bdi->dev)); 1167766f9164SPeter Zijlstra current->flags |= PF_SWAPWRITE; 116803ba3782SJens Axboe 1169839a8e86STejun Heo if (likely(!current_is_workqueue_rescuer() || 11704452226eSTejun Heo !test_bit(WB_registered, &wb->state))) { 117103ba3782SJens Axboe /* 1172f0054bb1STejun Heo * The normal path. Keep writing back @wb until its 1173839a8e86STejun Heo * work_list is empty. Note that this path is also taken 1174f0054bb1STejun Heo * if @wb is shutting down even when we're running off the 1175839a8e86STejun Heo * rescuer as work_list needs to be drained. 117603ba3782SJens Axboe */ 1177839a8e86STejun Heo do { 117825d130baSWanpeng Li pages_written = wb_do_writeback(wb); 1179455b2864SDave Chinner trace_writeback_pages_written(pages_written); 1180f0054bb1STejun Heo } while (!list_empty(&wb->work_list)); 1181839a8e86STejun Heo } else { 1182253c34e9SArtem Bityutskiy /* 1183839a8e86STejun Heo * bdi_wq can't get enough workers and we're running off 1184839a8e86STejun Heo * the emergency worker. Don't hog it. Hopefully, 1024 is 1185839a8e86STejun Heo * enough for efficient IO. 1186253c34e9SArtem Bityutskiy */ 1187f0054bb1STejun Heo pages_written = writeback_inodes_wb(wb, 1024, 1188839a8e86STejun Heo WB_REASON_FORKER_THREAD); 1189839a8e86STejun Heo trace_writeback_pages_written(pages_written); 119003ba3782SJens Axboe } 119103ba3782SJens Axboe 1192f0054bb1STejun Heo if (!list_empty(&wb->work_list)) 11936ca738d6SDerek Basehore mod_delayed_work(bdi_wq, &wb->dwork, 0); 11946ca738d6SDerek Basehore else if (wb_has_dirty_io(wb) && dirty_writeback_interval) 1195f0054bb1STejun Heo wb_wakeup_delayed(wb); 1196455b2864SDave Chinner 1197839a8e86STejun Heo current->flags &= ~PF_SWAPWRITE; 119803ba3782SJens Axboe } 119903ba3782SJens Axboe 120003ba3782SJens Axboe /* 120103ba3782SJens Axboe * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back 120203ba3782SJens Axboe * the whole world. 120303ba3782SJens Axboe */ 12040e175a18SCurt Wohlgemuth void wakeup_flusher_threads(long nr_pages, enum wb_reason reason) 120503ba3782SJens Axboe { 1206b8c2f347SChristoph Hellwig struct backing_dev_info *bdi; 1207b8c2f347SChristoph Hellwig 120847df3ddeSJan Kara if (!nr_pages) 120947df3ddeSJan Kara nr_pages = get_nr_dirty_pages(); 1210b8c2f347SChristoph Hellwig 1211b8c2f347SChristoph Hellwig rcu_read_lock(); 1212b8c2f347SChristoph Hellwig list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { 1213b8c2f347SChristoph Hellwig if (!bdi_has_dirty_io(bdi)) 1214b8c2f347SChristoph Hellwig continue; 1215f0054bb1STejun Heo __wb_start_writeback(&bdi->wb, nr_pages, false, reason); 1216b8c2f347SChristoph Hellwig } 1217b8c2f347SChristoph Hellwig rcu_read_unlock(); 121803ba3782SJens Axboe } 121903ba3782SJens Axboe 1220a2f48706STheodore Ts'o /* 1221a2f48706STheodore Ts'o * Wake up bdi's periodically to make sure dirtytime inodes gets 1222a2f48706STheodore Ts'o * written back periodically. We deliberately do *not* check the 1223a2f48706STheodore Ts'o * b_dirtytime list in wb_has_dirty_io(), since this would cause the 1224a2f48706STheodore Ts'o * kernel to be constantly waking up once there are any dirtytime 1225a2f48706STheodore Ts'o * inodes on the system. So instead we define a separate delayed work 1226a2f48706STheodore Ts'o * function which gets called much more rarely. (By default, only 1227a2f48706STheodore Ts'o * once every 12 hours.) 1228a2f48706STheodore Ts'o * 1229a2f48706STheodore Ts'o * If there is any other write activity going on in the file system, 1230a2f48706STheodore Ts'o * this function won't be necessary. But if the only thing that has 1231a2f48706STheodore Ts'o * happened on the file system is a dirtytime inode caused by an atime 1232a2f48706STheodore Ts'o * update, we need this infrastructure below to make sure that inode 1233a2f48706STheodore Ts'o * eventually gets pushed out to disk. 1234a2f48706STheodore Ts'o */ 1235a2f48706STheodore Ts'o static void wakeup_dirtytime_writeback(struct work_struct *w); 1236a2f48706STheodore Ts'o static DECLARE_DELAYED_WORK(dirtytime_work, wakeup_dirtytime_writeback); 1237a2f48706STheodore Ts'o 1238a2f48706STheodore Ts'o static void wakeup_dirtytime_writeback(struct work_struct *w) 1239a2f48706STheodore Ts'o { 1240a2f48706STheodore Ts'o struct backing_dev_info *bdi; 1241a2f48706STheodore Ts'o 1242a2f48706STheodore Ts'o rcu_read_lock(); 1243a2f48706STheodore Ts'o list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { 1244a2f48706STheodore Ts'o if (list_empty(&bdi->wb.b_dirty_time)) 1245a2f48706STheodore Ts'o continue; 1246f0054bb1STejun Heo wb_wakeup(&bdi->wb); 1247a2f48706STheodore Ts'o } 1248a2f48706STheodore Ts'o rcu_read_unlock(); 1249a2f48706STheodore Ts'o schedule_delayed_work(&dirtytime_work, dirtytime_expire_interval * HZ); 1250a2f48706STheodore Ts'o } 1251a2f48706STheodore Ts'o 1252a2f48706STheodore Ts'o static int __init start_dirtytime_writeback(void) 1253a2f48706STheodore Ts'o { 1254a2f48706STheodore Ts'o schedule_delayed_work(&dirtytime_work, dirtytime_expire_interval * HZ); 1255a2f48706STheodore Ts'o return 0; 1256a2f48706STheodore Ts'o } 1257a2f48706STheodore Ts'o __initcall(start_dirtytime_writeback); 1258a2f48706STheodore Ts'o 12591efff914STheodore Ts'o int dirtytime_interval_handler(struct ctl_table *table, int write, 12601efff914STheodore Ts'o void __user *buffer, size_t *lenp, loff_t *ppos) 12611efff914STheodore Ts'o { 12621efff914STheodore Ts'o int ret; 12631efff914STheodore Ts'o 12641efff914STheodore Ts'o ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos); 12651efff914STheodore Ts'o if (ret == 0 && write) 12661efff914STheodore Ts'o mod_delayed_work(system_wq, &dirtytime_work, 0); 12671efff914STheodore Ts'o return ret; 12681efff914STheodore Ts'o } 12691efff914STheodore Ts'o 127003ba3782SJens Axboe static noinline void block_dump___mark_inode_dirty(struct inode *inode) 127103ba3782SJens Axboe { 127203ba3782SJens Axboe if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { 127303ba3782SJens Axboe struct dentry *dentry; 127403ba3782SJens Axboe const char *name = "?"; 127503ba3782SJens Axboe 127603ba3782SJens Axboe dentry = d_find_alias(inode); 127703ba3782SJens Axboe if (dentry) { 127803ba3782SJens Axboe spin_lock(&dentry->d_lock); 127903ba3782SJens Axboe name = (const char *) dentry->d_name.name; 128003ba3782SJens Axboe } 128103ba3782SJens Axboe printk(KERN_DEBUG 128203ba3782SJens Axboe "%s(%d): dirtied inode %lu (%s) on %s\n", 128303ba3782SJens Axboe current->comm, task_pid_nr(current), inode->i_ino, 128403ba3782SJens Axboe name, inode->i_sb->s_id); 128503ba3782SJens Axboe if (dentry) { 128603ba3782SJens Axboe spin_unlock(&dentry->d_lock); 128703ba3782SJens Axboe dput(dentry); 128803ba3782SJens Axboe } 128903ba3782SJens Axboe } 129003ba3782SJens Axboe } 129103ba3782SJens Axboe 129203ba3782SJens Axboe /** 129303ba3782SJens Axboe * __mark_inode_dirty - internal function 129403ba3782SJens Axboe * @inode: inode to mark 129503ba3782SJens Axboe * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) 129603ba3782SJens Axboe * Mark an inode as dirty. Callers should use mark_inode_dirty or 129703ba3782SJens Axboe * mark_inode_dirty_sync. 129803ba3782SJens Axboe * 129903ba3782SJens Axboe * Put the inode on the super block's dirty list. 130003ba3782SJens Axboe * 130103ba3782SJens Axboe * CAREFUL! We mark it dirty unconditionally, but move it onto the 130203ba3782SJens Axboe * dirty list only if it is hashed or if it refers to a blockdev. 130303ba3782SJens Axboe * If it was not hashed, it will never be added to the dirty list 130403ba3782SJens Axboe * even if it is later hashed, as it will have been marked dirty already. 130503ba3782SJens Axboe * 130603ba3782SJens Axboe * In short, make sure you hash any inodes _before_ you start marking 130703ba3782SJens Axboe * them dirty. 130803ba3782SJens Axboe * 130903ba3782SJens Axboe * Note that for blockdevs, inode->dirtied_when represents the dirtying time of 131003ba3782SJens Axboe * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of 131103ba3782SJens Axboe * the kernel-internal blockdev inode represents the dirtying time of the 131203ba3782SJens Axboe * blockdev's pages. This is why for I_DIRTY_PAGES we always use 131303ba3782SJens Axboe * page->mapping->host, so the page-dirtying time is recorded in the internal 131403ba3782SJens Axboe * blockdev inode. 131503ba3782SJens Axboe */ 13160ae45f63STheodore Ts'o #define I_DIRTY_INODE (I_DIRTY_SYNC | I_DIRTY_DATASYNC) 131703ba3782SJens Axboe void __mark_inode_dirty(struct inode *inode, int flags) 131803ba3782SJens Axboe { 131903ba3782SJens Axboe struct super_block *sb = inode->i_sb; 1320253c34e9SArtem Bityutskiy struct backing_dev_info *bdi = NULL; 13210ae45f63STheodore Ts'o int dirtytime; 13220ae45f63STheodore Ts'o 13230ae45f63STheodore Ts'o trace_writeback_mark_inode_dirty(inode, flags); 132403ba3782SJens Axboe 132503ba3782SJens Axboe /* 132603ba3782SJens Axboe * Don't do this for I_DIRTY_PAGES - that doesn't actually 132703ba3782SJens Axboe * dirty the inode itself 132803ba3782SJens Axboe */ 13290ae45f63STheodore Ts'o if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC | I_DIRTY_TIME)) { 13309fb0a7daSTejun Heo trace_writeback_dirty_inode_start(inode, flags); 13319fb0a7daSTejun Heo 133203ba3782SJens Axboe if (sb->s_op->dirty_inode) 1333aa385729SChristoph Hellwig sb->s_op->dirty_inode(inode, flags); 13349fb0a7daSTejun Heo 13359fb0a7daSTejun Heo trace_writeback_dirty_inode(inode, flags); 133603ba3782SJens Axboe } 13370ae45f63STheodore Ts'o if (flags & I_DIRTY_INODE) 13380ae45f63STheodore Ts'o flags &= ~I_DIRTY_TIME; 13390ae45f63STheodore Ts'o dirtytime = flags & I_DIRTY_TIME; 134003ba3782SJens Axboe 134103ba3782SJens Axboe /* 13429c6ac78eSTejun Heo * Paired with smp_mb() in __writeback_single_inode() for the 13439c6ac78eSTejun Heo * following lockless i_state test. See there for details. 134403ba3782SJens Axboe */ 134503ba3782SJens Axboe smp_mb(); 134603ba3782SJens Axboe 13470ae45f63STheodore Ts'o if (((inode->i_state & flags) == flags) || 13480ae45f63STheodore Ts'o (dirtytime && (inode->i_state & I_DIRTY_INODE))) 134903ba3782SJens Axboe return; 135003ba3782SJens Axboe 135103ba3782SJens Axboe if (unlikely(block_dump)) 135203ba3782SJens Axboe block_dump___mark_inode_dirty(inode); 135303ba3782SJens Axboe 1354250df6edSDave Chinner spin_lock(&inode->i_lock); 13550ae45f63STheodore Ts'o if (dirtytime && (inode->i_state & I_DIRTY_INODE)) 13560ae45f63STheodore Ts'o goto out_unlock_inode; 135703ba3782SJens Axboe if ((inode->i_state & flags) != flags) { 135803ba3782SJens Axboe const int was_dirty = inode->i_state & I_DIRTY; 135903ba3782SJens Axboe 136052ebea74STejun Heo inode_attach_wb(inode, NULL); 136152ebea74STejun Heo 13620ae45f63STheodore Ts'o if (flags & I_DIRTY_INODE) 13630ae45f63STheodore Ts'o inode->i_state &= ~I_DIRTY_TIME; 136403ba3782SJens Axboe inode->i_state |= flags; 136503ba3782SJens Axboe 136603ba3782SJens Axboe /* 136703ba3782SJens Axboe * If the inode is being synced, just update its dirty state. 136803ba3782SJens Axboe * The unlocker will place the inode on the appropriate 136903ba3782SJens Axboe * superblock list, based upon its state. 137003ba3782SJens Axboe */ 137103ba3782SJens Axboe if (inode->i_state & I_SYNC) 1372250df6edSDave Chinner goto out_unlock_inode; 137303ba3782SJens Axboe 137403ba3782SJens Axboe /* 137503ba3782SJens Axboe * Only add valid (hashed) inodes to the superblock's 137603ba3782SJens Axboe * dirty list. Add blockdev inodes as well. 137703ba3782SJens Axboe */ 137803ba3782SJens Axboe if (!S_ISBLK(inode->i_mode)) { 13791d3382cbSAl Viro if (inode_unhashed(inode)) 1380250df6edSDave Chinner goto out_unlock_inode; 138103ba3782SJens Axboe } 1382a4ffdde6SAl Viro if (inode->i_state & I_FREEING) 1383250df6edSDave Chinner goto out_unlock_inode; 138403ba3782SJens Axboe 138503ba3782SJens Axboe /* 138603ba3782SJens Axboe * If the inode was already on b_dirty/b_io/b_more_io, don't 138703ba3782SJens Axboe * reposition it (that would break b_dirty time-ordering). 138803ba3782SJens Axboe */ 138903ba3782SJens Axboe if (!was_dirty) { 1390d6c10f1fSTejun Heo struct list_head *dirty_list; 1391a66979abSDave Chinner bool wakeup_bdi = false; 1392253c34e9SArtem Bityutskiy bdi = inode_to_bdi(inode); 1393500b067cSJens Axboe 1394146d7009SJunxiao Bi spin_unlock(&inode->i_lock); 1395146d7009SJunxiao Bi spin_lock(&bdi->wb.list_lock); 1396253c34e9SArtem Bityutskiy 1397d6c10f1fSTejun Heo WARN(bdi_cap_writeback_dirty(bdi) && 1398d6c10f1fSTejun Heo !test_bit(WB_registered, &bdi->wb.state), 1399d6c10f1fSTejun Heo "bdi-%s not registered\n", bdi->name); 140003ba3782SJens Axboe 140103ba3782SJens Axboe inode->dirtied_when = jiffies; 1402a2f48706STheodore Ts'o if (dirtytime) 1403a2f48706STheodore Ts'o inode->dirtied_time_when = jiffies; 1404d6c10f1fSTejun Heo 1405a2f48706STheodore Ts'o if (inode->i_state & (I_DIRTY_INODE | I_DIRTY_PAGES)) 1406d6c10f1fSTejun Heo dirty_list = &bdi->wb.b_dirty; 1407a2f48706STheodore Ts'o else 1408d6c10f1fSTejun Heo dirty_list = &bdi->wb.b_dirty_time; 1409d6c10f1fSTejun Heo 1410d6c10f1fSTejun Heo wakeup_bdi = inode_wb_list_move_locked(inode, &bdi->wb, 1411d6c10f1fSTejun Heo dirty_list); 1412d6c10f1fSTejun Heo 1413f758eeabSChristoph Hellwig spin_unlock(&bdi->wb.list_lock); 14140ae45f63STheodore Ts'o trace_writeback_dirty_inode_enqueue(inode); 1415253c34e9SArtem Bityutskiy 1416d6c10f1fSTejun Heo /* 1417d6c10f1fSTejun Heo * If this is the first dirty inode for this bdi, 1418d6c10f1fSTejun Heo * we have to wake-up the corresponding bdi thread 1419d6c10f1fSTejun Heo * to make sure background write-back happens 1420d6c10f1fSTejun Heo * later. 1421d6c10f1fSTejun Heo */ 1422d6c10f1fSTejun Heo if (bdi_cap_writeback_dirty(bdi) && wakeup_bdi) 1423f0054bb1STejun Heo wb_wakeup_delayed(&bdi->wb); 1424a66979abSDave Chinner return; 1425a66979abSDave Chinner } 1426a66979abSDave Chinner } 1427a66979abSDave Chinner out_unlock_inode: 1428a66979abSDave Chinner spin_unlock(&inode->i_lock); 1429a66979abSDave Chinner 143003ba3782SJens Axboe } 143103ba3782SJens Axboe EXPORT_SYMBOL(__mark_inode_dirty); 143203ba3782SJens Axboe 1433b6e51316SJens Axboe static void wait_sb_inodes(struct super_block *sb) 143466f3b8e2SJens Axboe { 143538f21977SNick Piggin struct inode *inode, *old_inode = NULL; 143638f21977SNick Piggin 143703ba3782SJens Axboe /* 143803ba3782SJens Axboe * We need to be protected against the filesystem going from 143903ba3782SJens Axboe * r/o to r/w or vice versa. 144003ba3782SJens Axboe */ 1441b6e51316SJens Axboe WARN_ON(!rwsem_is_locked(&sb->s_umount)); 144203ba3782SJens Axboe 144355fa6091SDave Chinner spin_lock(&inode_sb_list_lock); 144466f3b8e2SJens Axboe 144538f21977SNick Piggin /* 144638f21977SNick Piggin * Data integrity sync. Must wait for all pages under writeback, 144738f21977SNick Piggin * because there may have been pages dirtied before our sync 144838f21977SNick Piggin * call, but which had writeout started before we write it out. 144938f21977SNick Piggin * In which case, the inode may not be on the dirty list, but 145038f21977SNick Piggin * we still have to wait for that writeout. 145138f21977SNick Piggin */ 1452b6e51316SJens Axboe list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { 1453250df6edSDave Chinner struct address_space *mapping = inode->i_mapping; 145438f21977SNick Piggin 1455250df6edSDave Chinner spin_lock(&inode->i_lock); 1456250df6edSDave Chinner if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) || 1457250df6edSDave Chinner (mapping->nrpages == 0)) { 1458250df6edSDave Chinner spin_unlock(&inode->i_lock); 145938f21977SNick Piggin continue; 1460250df6edSDave Chinner } 146138f21977SNick Piggin __iget(inode); 1462250df6edSDave Chinner spin_unlock(&inode->i_lock); 146355fa6091SDave Chinner spin_unlock(&inode_sb_list_lock); 146455fa6091SDave Chinner 146538f21977SNick Piggin /* 146655fa6091SDave Chinner * We hold a reference to 'inode' so it couldn't have been 146755fa6091SDave Chinner * removed from s_inodes list while we dropped the 146855fa6091SDave Chinner * inode_sb_list_lock. We cannot iput the inode now as we can 146955fa6091SDave Chinner * be holding the last reference and we cannot iput it under 147055fa6091SDave Chinner * inode_sb_list_lock. So we keep the reference and iput it 147155fa6091SDave Chinner * later. 147238f21977SNick Piggin */ 147338f21977SNick Piggin iput(old_inode); 147438f21977SNick Piggin old_inode = inode; 147538f21977SNick Piggin 147638f21977SNick Piggin filemap_fdatawait(mapping); 147738f21977SNick Piggin 147838f21977SNick Piggin cond_resched(); 147938f21977SNick Piggin 148055fa6091SDave Chinner spin_lock(&inode_sb_list_lock); 148138f21977SNick Piggin } 148255fa6091SDave Chinner spin_unlock(&inode_sb_list_lock); 148338f21977SNick Piggin iput(old_inode); 148466f3b8e2SJens Axboe } 14851da177e4SLinus Torvalds 1486d8a8559cSJens Axboe /** 14873259f8beSChris Mason * writeback_inodes_sb_nr - writeback dirty inodes from given super_block 1488d8a8559cSJens Axboe * @sb: the superblock 14893259f8beSChris Mason * @nr: the number of pages to write 1490786228abSMarcos Paulo de Souza * @reason: reason why some writeback work initiated 14911da177e4SLinus Torvalds * 1492d8a8559cSJens Axboe * Start writeback on some inodes on this super_block. No guarantees are made 1493d8a8559cSJens Axboe * on how many (if any) will be written, and this function does not wait 14943259f8beSChris Mason * for IO completion of submitted IO. 14951da177e4SLinus Torvalds */ 14960e175a18SCurt Wohlgemuth void writeback_inodes_sb_nr(struct super_block *sb, 14970e175a18SCurt Wohlgemuth unsigned long nr, 14980e175a18SCurt Wohlgemuth enum wb_reason reason) 14991da177e4SLinus Torvalds { 150083ba7b07SChristoph Hellwig DECLARE_COMPLETION_ONSTACK(done); 150183ba7b07SChristoph Hellwig struct wb_writeback_work work = { 15023c4d7165SChristoph Hellwig .sb = sb, 15033c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_NONE, 15046e6938b6SWu Fengguang .tagged_writepages = 1, 150583ba7b07SChristoph Hellwig .done = &done, 15063259f8beSChris Mason .nr_pages = nr, 15070e175a18SCurt Wohlgemuth .reason = reason, 15083c4d7165SChristoph Hellwig }; 15090e3c9a22SJens Axboe 15106eedc701SJan Kara if (sb->s_bdi == &noop_backing_dev_info) 15116eedc701SJan Kara return; 1512cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 1513f0054bb1STejun Heo wb_queue_work(&sb->s_bdi->wb, &work); 151483ba7b07SChristoph Hellwig wait_for_completion(&done); 15151da177e4SLinus Torvalds } 15163259f8beSChris Mason EXPORT_SYMBOL(writeback_inodes_sb_nr); 15173259f8beSChris Mason 15183259f8beSChris Mason /** 15193259f8beSChris Mason * writeback_inodes_sb - writeback dirty inodes from given super_block 15203259f8beSChris Mason * @sb: the superblock 1521786228abSMarcos Paulo de Souza * @reason: reason why some writeback work was initiated 15223259f8beSChris Mason * 15233259f8beSChris Mason * Start writeback on some inodes on this super_block. No guarantees are made 15243259f8beSChris Mason * on how many (if any) will be written, and this function does not wait 15253259f8beSChris Mason * for IO completion of submitted IO. 15263259f8beSChris Mason */ 15270e175a18SCurt Wohlgemuth void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) 15283259f8beSChris Mason { 15290e175a18SCurt Wohlgemuth return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); 15303259f8beSChris Mason } 1531d8a8559cSJens Axboe EXPORT_SYMBOL(writeback_inodes_sb); 1532d8a8559cSJens Axboe 1533d8a8559cSJens Axboe /** 153410ee27a0SMiao Xie * try_to_writeback_inodes_sb_nr - try to start writeback if none underway 15353259f8beSChris Mason * @sb: the superblock 15363259f8beSChris Mason * @nr: the number of pages to write 153710ee27a0SMiao Xie * @reason: the reason of writeback 15383259f8beSChris Mason * 153910ee27a0SMiao Xie * Invoke writeback_inodes_sb_nr if no writeback is currently underway. 15403259f8beSChris Mason * Returns 1 if writeback was started, 0 if not. 15413259f8beSChris Mason */ 154210ee27a0SMiao Xie int try_to_writeback_inodes_sb_nr(struct super_block *sb, 15430e175a18SCurt Wohlgemuth unsigned long nr, 15440e175a18SCurt Wohlgemuth enum wb_reason reason) 15453259f8beSChris Mason { 154610ee27a0SMiao Xie if (writeback_in_progress(sb->s_bdi)) 154710ee27a0SMiao Xie return 1; 154810ee27a0SMiao Xie 154910ee27a0SMiao Xie if (!down_read_trylock(&sb->s_umount)) 155010ee27a0SMiao Xie return 0; 155110ee27a0SMiao Xie 15520e175a18SCurt Wohlgemuth writeback_inodes_sb_nr(sb, nr, reason); 15533259f8beSChris Mason up_read(&sb->s_umount); 15543259f8beSChris Mason return 1; 15553259f8beSChris Mason } 155610ee27a0SMiao Xie EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr); 155710ee27a0SMiao Xie 155810ee27a0SMiao Xie /** 155910ee27a0SMiao Xie * try_to_writeback_inodes_sb - try to start writeback if none underway 156010ee27a0SMiao Xie * @sb: the superblock 156110ee27a0SMiao Xie * @reason: reason why some writeback work was initiated 156210ee27a0SMiao Xie * 156310ee27a0SMiao Xie * Implement by try_to_writeback_inodes_sb_nr() 156410ee27a0SMiao Xie * Returns 1 if writeback was started, 0 if not. 156510ee27a0SMiao Xie */ 156610ee27a0SMiao Xie int try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) 156710ee27a0SMiao Xie { 156810ee27a0SMiao Xie return try_to_writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); 156910ee27a0SMiao Xie } 157010ee27a0SMiao Xie EXPORT_SYMBOL(try_to_writeback_inodes_sb); 15713259f8beSChris Mason 15723259f8beSChris Mason /** 1573d8a8559cSJens Axboe * sync_inodes_sb - sync sb inode pages 1574d8a8559cSJens Axboe * @sb: the superblock 1575d8a8559cSJens Axboe * 1576d8a8559cSJens Axboe * This function writes and waits on any dirty inode belonging to this 15770dc83bd3SJan Kara * super_block. 1578d8a8559cSJens Axboe */ 15790dc83bd3SJan Kara void sync_inodes_sb(struct super_block *sb) 1580d8a8559cSJens Axboe { 158183ba7b07SChristoph Hellwig DECLARE_COMPLETION_ONSTACK(done); 158283ba7b07SChristoph Hellwig struct wb_writeback_work work = { 15833c4d7165SChristoph Hellwig .sb = sb, 15843c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_ALL, 15853c4d7165SChristoph Hellwig .nr_pages = LONG_MAX, 15863c4d7165SChristoph Hellwig .range_cyclic = 0, 158783ba7b07SChristoph Hellwig .done = &done, 15880e175a18SCurt Wohlgemuth .reason = WB_REASON_SYNC, 15897747bd4bSDave Chinner .for_sync = 1, 15903c4d7165SChristoph Hellwig }; 15913c4d7165SChristoph Hellwig 15926eedc701SJan Kara /* Nothing to do? */ 15936eedc701SJan Kara if (sb->s_bdi == &noop_backing_dev_info) 15946eedc701SJan Kara return; 1595cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 1596cf37e972SChristoph Hellwig 1597f0054bb1STejun Heo wb_queue_work(&sb->s_bdi->wb, &work); 159883ba7b07SChristoph Hellwig wait_for_completion(&done); 159983ba7b07SChristoph Hellwig 1600b6e51316SJens Axboe wait_sb_inodes(sb); 1601d8a8559cSJens Axboe } 1602d8a8559cSJens Axboe EXPORT_SYMBOL(sync_inodes_sb); 16031da177e4SLinus Torvalds 16041da177e4SLinus Torvalds /** 16051da177e4SLinus Torvalds * write_inode_now - write an inode to disk 16061da177e4SLinus Torvalds * @inode: inode to write to disk 16071da177e4SLinus Torvalds * @sync: whether the write should be synchronous or not 16081da177e4SLinus Torvalds * 16097f04c26dSAndrea Arcangeli * This function commits an inode to disk immediately if it is dirty. This is 16107f04c26dSAndrea Arcangeli * primarily needed by knfsd. 16117f04c26dSAndrea Arcangeli * 16127f04c26dSAndrea Arcangeli * The caller must either have a ref on the inode or must have set I_WILL_FREE. 16131da177e4SLinus Torvalds */ 16141da177e4SLinus Torvalds int write_inode_now(struct inode *inode, int sync) 16151da177e4SLinus Torvalds { 1616f758eeabSChristoph Hellwig struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; 16171da177e4SLinus Torvalds struct writeback_control wbc = { 16181da177e4SLinus Torvalds .nr_to_write = LONG_MAX, 161918914b18SMike Galbraith .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, 1620111ebb6eSOGAWA Hirofumi .range_start = 0, 1621111ebb6eSOGAWA Hirofumi .range_end = LLONG_MAX, 16221da177e4SLinus Torvalds }; 16231da177e4SLinus Torvalds 16241da177e4SLinus Torvalds if (!mapping_cap_writeback_dirty(inode->i_mapping)) 162549364ce2SAndrew Morton wbc.nr_to_write = 0; 16261da177e4SLinus Torvalds 16271da177e4SLinus Torvalds might_sleep(); 16284f8ad655SJan Kara return writeback_single_inode(inode, wb, &wbc); 16291da177e4SLinus Torvalds } 16301da177e4SLinus Torvalds EXPORT_SYMBOL(write_inode_now); 16311da177e4SLinus Torvalds 16321da177e4SLinus Torvalds /** 16331da177e4SLinus Torvalds * sync_inode - write an inode and its pages to disk. 16341da177e4SLinus Torvalds * @inode: the inode to sync 16351da177e4SLinus Torvalds * @wbc: controls the writeback mode 16361da177e4SLinus Torvalds * 16371da177e4SLinus Torvalds * sync_inode() will write an inode and its pages to disk. It will also 16381da177e4SLinus Torvalds * correctly update the inode on its superblock's dirty inode lists and will 16391da177e4SLinus Torvalds * update inode->i_state. 16401da177e4SLinus Torvalds * 16411da177e4SLinus Torvalds * The caller must have a ref on the inode. 16421da177e4SLinus Torvalds */ 16431da177e4SLinus Torvalds int sync_inode(struct inode *inode, struct writeback_control *wbc) 16441da177e4SLinus Torvalds { 16454f8ad655SJan Kara return writeback_single_inode(inode, &inode_to_bdi(inode)->wb, wbc); 16461da177e4SLinus Torvalds } 16471da177e4SLinus Torvalds EXPORT_SYMBOL(sync_inode); 1648c3765016SChristoph Hellwig 1649c3765016SChristoph Hellwig /** 1650c691b9d9SAndrew Morton * sync_inode_metadata - write an inode to disk 1651c3765016SChristoph Hellwig * @inode: the inode to sync 1652c3765016SChristoph Hellwig * @wait: wait for I/O to complete. 1653c3765016SChristoph Hellwig * 1654c691b9d9SAndrew Morton * Write an inode to disk and adjust its dirty state after completion. 1655c3765016SChristoph Hellwig * 1656c3765016SChristoph Hellwig * Note: only writes the actual inode, no associated data or other metadata. 1657c3765016SChristoph Hellwig */ 1658c3765016SChristoph Hellwig int sync_inode_metadata(struct inode *inode, int wait) 1659c3765016SChristoph Hellwig { 1660c3765016SChristoph Hellwig struct writeback_control wbc = { 1661c3765016SChristoph Hellwig .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE, 1662c3765016SChristoph Hellwig .nr_to_write = 0, /* metadata-only */ 1663c3765016SChristoph Hellwig }; 1664c3765016SChristoph Hellwig 1665c3765016SChristoph Hellwig return sync_inode(inode, &wbc); 1666c3765016SChristoph Hellwig } 1667c3765016SChristoph Hellwig EXPORT_SYMBOL(sync_inode_metadata); 1668