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 687ccf19a8SNick Piggin static inline struct inode *wb_inode(struct list_head *head) 697ccf19a8SNick Piggin { 707ccf19a8SNick Piggin return list_entry(head, struct inode, i_wb_list); 717ccf19a8SNick Piggin } 727ccf19a8SNick Piggin 7315eb77a0SWu Fengguang /* 7415eb77a0SWu Fengguang * Include the creation of the trace points after defining the 7515eb77a0SWu Fengguang * wb_writeback_work structure and inline functions so that the definition 7615eb77a0SWu Fengguang * remains local to this file. 7715eb77a0SWu Fengguang */ 7815eb77a0SWu Fengguang #define CREATE_TRACE_POINTS 7915eb77a0SWu Fengguang #include <trace/events/writeback.h> 8015eb77a0SWu Fengguang 81774016b2SSteven Whitehouse EXPORT_TRACEPOINT_SYMBOL_GPL(wbc_writepage); 82774016b2SSteven Whitehouse 83d6c10f1fSTejun Heo static bool wb_io_lists_populated(struct bdi_writeback *wb) 84d6c10f1fSTejun Heo { 85d6c10f1fSTejun Heo if (wb_has_dirty_io(wb)) { 86d6c10f1fSTejun Heo return false; 87d6c10f1fSTejun Heo } else { 88d6c10f1fSTejun Heo set_bit(WB_has_dirty_io, &wb->state); 8995a46c65STejun Heo WARN_ON_ONCE(!wb->avg_write_bandwidth); 90766a9d6eSTejun Heo atomic_long_add(wb->avg_write_bandwidth, 91766a9d6eSTejun Heo &wb->bdi->tot_write_bandwidth); 92d6c10f1fSTejun Heo return true; 93d6c10f1fSTejun Heo } 94d6c10f1fSTejun Heo } 95d6c10f1fSTejun Heo 96d6c10f1fSTejun Heo static void wb_io_lists_depopulated(struct bdi_writeback *wb) 97d6c10f1fSTejun Heo { 98d6c10f1fSTejun Heo if (wb_has_dirty_io(wb) && list_empty(&wb->b_dirty) && 99766a9d6eSTejun Heo list_empty(&wb->b_io) && list_empty(&wb->b_more_io)) { 100d6c10f1fSTejun Heo clear_bit(WB_has_dirty_io, &wb->state); 10195a46c65STejun Heo WARN_ON_ONCE(atomic_long_sub_return(wb->avg_write_bandwidth, 10295a46c65STejun Heo &wb->bdi->tot_write_bandwidth) < 0); 103766a9d6eSTejun Heo } 104d6c10f1fSTejun Heo } 105d6c10f1fSTejun Heo 106d6c10f1fSTejun Heo /** 107d6c10f1fSTejun Heo * inode_wb_list_move_locked - move an inode onto a bdi_writeback IO list 108d6c10f1fSTejun Heo * @inode: inode to be moved 109d6c10f1fSTejun Heo * @wb: target bdi_writeback 110d6c10f1fSTejun Heo * @head: one of @wb->b_{dirty|io|more_io} 111d6c10f1fSTejun Heo * 112d6c10f1fSTejun Heo * Move @inode->i_wb_list to @list of @wb and set %WB_has_dirty_io. 113d6c10f1fSTejun Heo * Returns %true if @inode is the first occupant of the !dirty_time IO 114d6c10f1fSTejun Heo * lists; otherwise, %false. 115d6c10f1fSTejun Heo */ 116d6c10f1fSTejun Heo static bool inode_wb_list_move_locked(struct inode *inode, 117d6c10f1fSTejun Heo struct bdi_writeback *wb, 118d6c10f1fSTejun Heo struct list_head *head) 119d6c10f1fSTejun Heo { 120d6c10f1fSTejun Heo assert_spin_locked(&wb->list_lock); 121d6c10f1fSTejun Heo 122d6c10f1fSTejun Heo list_move(&inode->i_wb_list, head); 123d6c10f1fSTejun Heo 124d6c10f1fSTejun Heo /* dirty_time doesn't count as dirty_io until expiration */ 125d6c10f1fSTejun Heo if (head != &wb->b_dirty_time) 126d6c10f1fSTejun Heo return wb_io_lists_populated(wb); 127d6c10f1fSTejun Heo 128d6c10f1fSTejun Heo wb_io_lists_depopulated(wb); 129d6c10f1fSTejun Heo return false; 130d6c10f1fSTejun Heo } 131d6c10f1fSTejun Heo 132d6c10f1fSTejun Heo /** 133d6c10f1fSTejun Heo * inode_wb_list_del_locked - remove an inode from its bdi_writeback IO list 134d6c10f1fSTejun Heo * @inode: inode to be removed 135d6c10f1fSTejun Heo * @wb: bdi_writeback @inode is being removed from 136d6c10f1fSTejun Heo * 137d6c10f1fSTejun Heo * Remove @inode which may be on one of @wb->b_{dirty|io|more_io} lists and 138d6c10f1fSTejun Heo * clear %WB_has_dirty_io if all are empty afterwards. 139d6c10f1fSTejun Heo */ 140d6c10f1fSTejun Heo static void inode_wb_list_del_locked(struct inode *inode, 141d6c10f1fSTejun Heo struct bdi_writeback *wb) 142d6c10f1fSTejun Heo { 143d6c10f1fSTejun Heo assert_spin_locked(&wb->list_lock); 144d6c10f1fSTejun Heo 145d6c10f1fSTejun Heo list_del_init(&inode->i_wb_list); 146d6c10f1fSTejun Heo wb_io_lists_depopulated(wb); 147d6c10f1fSTejun Heo } 148d6c10f1fSTejun Heo 149f0054bb1STejun Heo static void wb_wakeup(struct bdi_writeback *wb) 1505acda9d1SJan Kara { 151f0054bb1STejun Heo spin_lock_bh(&wb->work_lock); 152f0054bb1STejun Heo if (test_bit(WB_registered, &wb->state)) 153f0054bb1STejun Heo mod_delayed_work(bdi_wq, &wb->dwork, 0); 154f0054bb1STejun Heo spin_unlock_bh(&wb->work_lock); 1555acda9d1SJan Kara } 1565acda9d1SJan Kara 157f0054bb1STejun Heo static void wb_queue_work(struct bdi_writeback *wb, 1586585027aSJan Kara struct wb_writeback_work *work) 1596585027aSJan Kara { 160f0054bb1STejun Heo trace_writeback_queue(wb->bdi, work); 1616585027aSJan Kara 162f0054bb1STejun Heo spin_lock_bh(&wb->work_lock); 163f0054bb1STejun Heo if (!test_bit(WB_registered, &wb->state)) { 1645acda9d1SJan Kara if (work->done) 1655acda9d1SJan Kara complete(work->done); 1665acda9d1SJan Kara goto out_unlock; 1675acda9d1SJan Kara } 168f0054bb1STejun Heo list_add_tail(&work->list, &wb->work_list); 169f0054bb1STejun Heo mod_delayed_work(bdi_wq, &wb->dwork, 0); 1705acda9d1SJan Kara out_unlock: 171f0054bb1STejun Heo spin_unlock_bh(&wb->work_lock); 17203ba3782SJens Axboe } 1731da177e4SLinus Torvalds 174703c2708STejun Heo #ifdef CONFIG_CGROUP_WRITEBACK 175703c2708STejun Heo 176703c2708STejun Heo /** 177703c2708STejun Heo * inode_congested - test whether an inode is congested 178703c2708STejun Heo * @inode: inode to test for congestion 179703c2708STejun Heo * @cong_bits: mask of WB_[a]sync_congested bits to test 180703c2708STejun Heo * 181703c2708STejun Heo * Tests whether @inode is congested. @cong_bits is the mask of congestion 182703c2708STejun Heo * bits to test and the return value is the mask of set bits. 183703c2708STejun Heo * 184703c2708STejun Heo * If cgroup writeback is enabled for @inode, the congestion state is 185703c2708STejun Heo * determined by whether the cgwb (cgroup bdi_writeback) for the blkcg 186703c2708STejun Heo * associated with @inode is congested; otherwise, the root wb's congestion 187703c2708STejun Heo * state is used. 188703c2708STejun Heo */ 189703c2708STejun Heo int inode_congested(struct inode *inode, int cong_bits) 190703c2708STejun Heo { 191703c2708STejun Heo if (inode) { 192703c2708STejun Heo struct bdi_writeback *wb = inode_to_wb(inode); 193703c2708STejun Heo if (wb) 194703c2708STejun Heo return wb_congested(wb, cong_bits); 195703c2708STejun Heo } 196703c2708STejun Heo 197703c2708STejun Heo return wb_congested(&inode_to_bdi(inode)->wb, cong_bits); 198703c2708STejun Heo } 199703c2708STejun Heo EXPORT_SYMBOL_GPL(inode_congested); 200703c2708STejun Heo 201703c2708STejun Heo #endif /* CONFIG_CGROUP_WRITEBACK */ 202703c2708STejun Heo 203c00ddad3STejun Heo void wb_start_writeback(struct bdi_writeback *wb, long nr_pages, 204c00ddad3STejun Heo bool range_cyclic, enum wb_reason reason) 205b6e51316SJens Axboe { 206c00ddad3STejun Heo struct wb_writeback_work *work; 207c00ddad3STejun Heo 208c00ddad3STejun Heo if (!wb_has_dirty_io(wb)) 209c00ddad3STejun Heo return; 210c00ddad3STejun Heo 211c00ddad3STejun Heo /* 212c00ddad3STejun Heo * This is WB_SYNC_NONE writeback, so if allocation fails just 213c00ddad3STejun Heo * wakeup the thread for old dirty data writeback 214c00ddad3STejun Heo */ 215c00ddad3STejun Heo work = kzalloc(sizeof(*work), GFP_ATOMIC); 216c00ddad3STejun Heo if (!work) { 217c00ddad3STejun Heo trace_writeback_nowork(wb->bdi); 218c00ddad3STejun Heo wb_wakeup(wb); 219c00ddad3STejun Heo return; 220c00ddad3STejun Heo } 221c00ddad3STejun Heo 222c00ddad3STejun Heo work->sync_mode = WB_SYNC_NONE; 223c00ddad3STejun Heo work->nr_pages = nr_pages; 224c00ddad3STejun Heo work->range_cyclic = range_cyclic; 225c00ddad3STejun Heo work->reason = reason; 226c00ddad3STejun Heo 227c00ddad3STejun Heo wb_queue_work(wb, work); 228d3ddec76SWu Fengguang } 229d3ddec76SWu Fengguang 230c5444198SChristoph Hellwig /** 2319ecf4866STejun Heo * wb_start_background_writeback - start background writeback 2329ecf4866STejun Heo * @wb: bdi_writback to write from 233c5444198SChristoph Hellwig * 234c5444198SChristoph Hellwig * Description: 2356585027aSJan Kara * This makes sure WB_SYNC_NONE background writeback happens. When 2369ecf4866STejun Heo * this function returns, it is only guaranteed that for given wb 2376585027aSJan Kara * some IO is happening if we are over background dirty threshold. 2386585027aSJan Kara * Caller need not hold sb s_umount semaphore. 239c5444198SChristoph Hellwig */ 2409ecf4866STejun Heo void wb_start_background_writeback(struct bdi_writeback *wb) 241c5444198SChristoph Hellwig { 2426585027aSJan Kara /* 2436585027aSJan Kara * We just wake up the flusher thread. It will perform background 2446585027aSJan Kara * writeback as soon as there is no other work to do. 2456585027aSJan Kara */ 2469ecf4866STejun Heo trace_writeback_wake_background(wb->bdi); 2479ecf4866STejun Heo wb_wakeup(wb); 2481da177e4SLinus Torvalds } 2491da177e4SLinus Torvalds 2501da177e4SLinus Torvalds /* 251a66979abSDave Chinner * Remove the inode from the writeback list it is on. 252a66979abSDave Chinner */ 253a66979abSDave Chinner void inode_wb_list_del(struct inode *inode) 254a66979abSDave Chinner { 25552ebea74STejun Heo struct bdi_writeback *wb = inode_to_wb(inode); 256a66979abSDave Chinner 25752ebea74STejun Heo spin_lock(&wb->list_lock); 258d6c10f1fSTejun Heo inode_wb_list_del_locked(inode, wb); 25952ebea74STejun Heo spin_unlock(&wb->list_lock); 260f758eeabSChristoph Hellwig } 261a66979abSDave Chinner 262a66979abSDave Chinner /* 2636610a0bcSAndrew Morton * Redirty an inode: set its when-it-was dirtied timestamp and move it to the 2646610a0bcSAndrew Morton * furthest end of its superblock's dirty-inode list. 2656610a0bcSAndrew Morton * 2666610a0bcSAndrew Morton * Before stamping the inode's ->dirtied_when, we check to see whether it is 26766f3b8e2SJens Axboe * already the most-recently-dirtied inode on the b_dirty list. If that is 2686610a0bcSAndrew Morton * the case then the inode must have been redirtied while it was being written 2696610a0bcSAndrew Morton * out and we don't reset its dirtied_when. 2706610a0bcSAndrew Morton */ 271f758eeabSChristoph Hellwig static void redirty_tail(struct inode *inode, struct bdi_writeback *wb) 2726610a0bcSAndrew Morton { 27303ba3782SJens Axboe if (!list_empty(&wb->b_dirty)) { 27466f3b8e2SJens Axboe struct inode *tail; 2756610a0bcSAndrew Morton 2767ccf19a8SNick Piggin tail = wb_inode(wb->b_dirty.next); 27766f3b8e2SJens Axboe if (time_before(inode->dirtied_when, tail->dirtied_when)) 2786610a0bcSAndrew Morton inode->dirtied_when = jiffies; 2796610a0bcSAndrew Morton } 280d6c10f1fSTejun Heo inode_wb_list_move_locked(inode, wb, &wb->b_dirty); 2816610a0bcSAndrew Morton } 2826610a0bcSAndrew Morton 2836610a0bcSAndrew Morton /* 28466f3b8e2SJens Axboe * requeue inode for re-scanning after bdi->b_io list is exhausted. 285c986d1e2SAndrew Morton */ 286f758eeabSChristoph Hellwig static void requeue_io(struct inode *inode, struct bdi_writeback *wb) 287c986d1e2SAndrew Morton { 288d6c10f1fSTejun Heo inode_wb_list_move_locked(inode, wb, &wb->b_more_io); 289c986d1e2SAndrew Morton } 290c986d1e2SAndrew Morton 2911c0eeaf5SJoern Engel static void inode_sync_complete(struct inode *inode) 2921c0eeaf5SJoern Engel { 293365b94aeSJan Kara inode->i_state &= ~I_SYNC; 2944eff96ddSJan Kara /* If inode is clean an unused, put it into LRU now... */ 2954eff96ddSJan Kara inode_add_lru(inode); 296365b94aeSJan Kara /* Waiters must see I_SYNC cleared before being woken up */ 2971c0eeaf5SJoern Engel smp_mb(); 2981c0eeaf5SJoern Engel wake_up_bit(&inode->i_state, __I_SYNC); 2991c0eeaf5SJoern Engel } 3001c0eeaf5SJoern Engel 301d2caa3c5SJeff Layton static bool inode_dirtied_after(struct inode *inode, unsigned long t) 302d2caa3c5SJeff Layton { 303d2caa3c5SJeff Layton bool ret = time_after(inode->dirtied_when, t); 304d2caa3c5SJeff Layton #ifndef CONFIG_64BIT 305d2caa3c5SJeff Layton /* 306d2caa3c5SJeff Layton * For inodes being constantly redirtied, dirtied_when can get stuck. 307d2caa3c5SJeff Layton * It _appears_ to be in the future, but is actually in distant past. 308d2caa3c5SJeff Layton * This test is necessary to prevent such wrapped-around relative times 3095b0830cbSJens Axboe * from permanently stopping the whole bdi writeback. 310d2caa3c5SJeff Layton */ 311d2caa3c5SJeff Layton ret = ret && time_before_eq(inode->dirtied_when, jiffies); 312d2caa3c5SJeff Layton #endif 313d2caa3c5SJeff Layton return ret; 314d2caa3c5SJeff Layton } 315d2caa3c5SJeff Layton 3160ae45f63STheodore Ts'o #define EXPIRE_DIRTY_ATIME 0x0001 3170ae45f63STheodore Ts'o 318c986d1e2SAndrew Morton /* 3190e2f2b23SWang Sheng-Hui * Move expired (dirtied before work->older_than_this) dirty inodes from 320697e6fedSJan Kara * @delaying_queue to @dispatch_queue. 3212c136579SFengguang Wu */ 322e84d0a4fSWu Fengguang static int move_expired_inodes(struct list_head *delaying_queue, 3232c136579SFengguang Wu struct list_head *dispatch_queue, 3240ae45f63STheodore Ts'o int flags, 325ad4e38ddSCurt Wohlgemuth struct wb_writeback_work *work) 3262c136579SFengguang Wu { 3270ae45f63STheodore Ts'o unsigned long *older_than_this = NULL; 3280ae45f63STheodore Ts'o unsigned long expire_time; 3295c03449dSShaohua Li LIST_HEAD(tmp); 3305c03449dSShaohua Li struct list_head *pos, *node; 331cf137307SJens Axboe struct super_block *sb = NULL; 3325c03449dSShaohua Li struct inode *inode; 333cf137307SJens Axboe int do_sb_sort = 0; 334e84d0a4fSWu Fengguang int moved = 0; 3355c03449dSShaohua Li 3360ae45f63STheodore Ts'o if ((flags & EXPIRE_DIRTY_ATIME) == 0) 3370ae45f63STheodore Ts'o older_than_this = work->older_than_this; 338a2f48706STheodore Ts'o else if (!work->for_sync) { 339a2f48706STheodore Ts'o expire_time = jiffies - (dirtytime_expire_interval * HZ); 3400ae45f63STheodore Ts'o older_than_this = &expire_time; 3410ae45f63STheodore Ts'o } 3422c136579SFengguang Wu while (!list_empty(delaying_queue)) { 3437ccf19a8SNick Piggin inode = wb_inode(delaying_queue->prev); 3440ae45f63STheodore Ts'o if (older_than_this && 3450ae45f63STheodore Ts'o inode_dirtied_after(inode, *older_than_this)) 3462c136579SFengguang Wu break; 347a8855990SJan Kara list_move(&inode->i_wb_list, &tmp); 348a8855990SJan Kara moved++; 3490ae45f63STheodore Ts'o if (flags & EXPIRE_DIRTY_ATIME) 3500ae45f63STheodore Ts'o set_bit(__I_DIRTY_TIME_EXPIRED, &inode->i_state); 351a8855990SJan Kara if (sb_is_blkdev_sb(inode->i_sb)) 352a8855990SJan Kara continue; 353cf137307SJens Axboe if (sb && sb != inode->i_sb) 354cf137307SJens Axboe do_sb_sort = 1; 355cf137307SJens Axboe sb = inode->i_sb; 3565c03449dSShaohua Li } 3575c03449dSShaohua Li 358cf137307SJens Axboe /* just one sb in list, splice to dispatch_queue and we're done */ 359cf137307SJens Axboe if (!do_sb_sort) { 360cf137307SJens Axboe list_splice(&tmp, dispatch_queue); 361e84d0a4fSWu Fengguang goto out; 362cf137307SJens Axboe } 363cf137307SJens Axboe 3645c03449dSShaohua Li /* Move inodes from one superblock together */ 3655c03449dSShaohua Li while (!list_empty(&tmp)) { 3667ccf19a8SNick Piggin sb = wb_inode(tmp.prev)->i_sb; 3675c03449dSShaohua Li list_for_each_prev_safe(pos, node, &tmp) { 3687ccf19a8SNick Piggin inode = wb_inode(pos); 3695c03449dSShaohua Li if (inode->i_sb == sb) 3707ccf19a8SNick Piggin list_move(&inode->i_wb_list, dispatch_queue); 3712c136579SFengguang Wu } 3722c136579SFengguang Wu } 373e84d0a4fSWu Fengguang out: 374e84d0a4fSWu Fengguang return moved; 3755c03449dSShaohua Li } 3762c136579SFengguang Wu 3772c136579SFengguang Wu /* 3782c136579SFengguang Wu * Queue all expired dirty inodes for io, eldest first. 3794ea879b9SWu Fengguang * Before 3804ea879b9SWu Fengguang * newly dirtied b_dirty b_io b_more_io 3814ea879b9SWu Fengguang * =============> gf edc BA 3824ea879b9SWu Fengguang * After 3834ea879b9SWu Fengguang * newly dirtied b_dirty b_io b_more_io 3844ea879b9SWu Fengguang * =============> g fBAedc 3854ea879b9SWu Fengguang * | 3864ea879b9SWu Fengguang * +--> dequeue for IO 3872c136579SFengguang Wu */ 388ad4e38ddSCurt Wohlgemuth static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work) 3892c136579SFengguang Wu { 390e84d0a4fSWu Fengguang int moved; 3910ae45f63STheodore Ts'o 392f758eeabSChristoph Hellwig assert_spin_locked(&wb->list_lock); 3934ea879b9SWu Fengguang list_splice_init(&wb->b_more_io, &wb->b_io); 3940ae45f63STheodore Ts'o moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, 0, work); 3950ae45f63STheodore Ts'o moved += move_expired_inodes(&wb->b_dirty_time, &wb->b_io, 3960ae45f63STheodore Ts'o EXPIRE_DIRTY_ATIME, work); 397d6c10f1fSTejun Heo if (moved) 398d6c10f1fSTejun Heo wb_io_lists_populated(wb); 399ad4e38ddSCurt Wohlgemuth trace_writeback_queue_io(wb, work, moved); 40066f3b8e2SJens Axboe } 40166f3b8e2SJens Axboe 402a9185b41SChristoph Hellwig static int write_inode(struct inode *inode, struct writeback_control *wbc) 40366f3b8e2SJens Axboe { 4049fb0a7daSTejun Heo int ret; 4059fb0a7daSTejun Heo 4069fb0a7daSTejun Heo if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) { 4079fb0a7daSTejun Heo trace_writeback_write_inode_start(inode, wbc); 4089fb0a7daSTejun Heo ret = inode->i_sb->s_op->write_inode(inode, wbc); 4099fb0a7daSTejun Heo trace_writeback_write_inode(inode, wbc); 4109fb0a7daSTejun Heo return ret; 4119fb0a7daSTejun Heo } 41203ba3782SJens Axboe return 0; 41366f3b8e2SJens Axboe } 41408d8e974SFengguang Wu 4152c136579SFengguang Wu /* 416169ebd90SJan Kara * Wait for writeback on an inode to complete. Called with i_lock held. 417169ebd90SJan Kara * Caller must make sure inode cannot go away when we drop i_lock. 41801c03194SChristoph Hellwig */ 419169ebd90SJan Kara static void __inode_wait_for_writeback(struct inode *inode) 420169ebd90SJan Kara __releases(inode->i_lock) 421169ebd90SJan Kara __acquires(inode->i_lock) 42201c03194SChristoph Hellwig { 42301c03194SChristoph Hellwig DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); 42401c03194SChristoph Hellwig wait_queue_head_t *wqh; 42501c03194SChristoph Hellwig 42601c03194SChristoph Hellwig wqh = bit_waitqueue(&inode->i_state, __I_SYNC); 42758a9d3d8SRichard Kennedy while (inode->i_state & I_SYNC) { 428250df6edSDave Chinner spin_unlock(&inode->i_lock); 42974316201SNeilBrown __wait_on_bit(wqh, &wq, bit_wait, 43074316201SNeilBrown TASK_UNINTERRUPTIBLE); 431250df6edSDave Chinner spin_lock(&inode->i_lock); 43258a9d3d8SRichard Kennedy } 43301c03194SChristoph Hellwig } 43401c03194SChristoph Hellwig 43501c03194SChristoph Hellwig /* 436169ebd90SJan Kara * Wait for writeback on an inode to complete. Caller must have inode pinned. 437169ebd90SJan Kara */ 438169ebd90SJan Kara void inode_wait_for_writeback(struct inode *inode) 439169ebd90SJan Kara { 440169ebd90SJan Kara spin_lock(&inode->i_lock); 441169ebd90SJan Kara __inode_wait_for_writeback(inode); 442169ebd90SJan Kara spin_unlock(&inode->i_lock); 443169ebd90SJan Kara } 444169ebd90SJan Kara 445169ebd90SJan Kara /* 446169ebd90SJan Kara * Sleep until I_SYNC is cleared. This function must be called with i_lock 447169ebd90SJan Kara * held and drops it. It is aimed for callers not holding any inode reference 448169ebd90SJan Kara * so once i_lock is dropped, inode can go away. 449169ebd90SJan Kara */ 450169ebd90SJan Kara static void inode_sleep_on_writeback(struct inode *inode) 451169ebd90SJan Kara __releases(inode->i_lock) 452169ebd90SJan Kara { 453169ebd90SJan Kara DEFINE_WAIT(wait); 454169ebd90SJan Kara wait_queue_head_t *wqh = bit_waitqueue(&inode->i_state, __I_SYNC); 455169ebd90SJan Kara int sleep; 456169ebd90SJan Kara 457169ebd90SJan Kara prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); 458169ebd90SJan Kara sleep = inode->i_state & I_SYNC; 459169ebd90SJan Kara spin_unlock(&inode->i_lock); 460169ebd90SJan Kara if (sleep) 461169ebd90SJan Kara schedule(); 462169ebd90SJan Kara finish_wait(wqh, &wait); 463169ebd90SJan Kara } 464169ebd90SJan Kara 465169ebd90SJan Kara /* 466ccb26b5aSJan Kara * Find proper writeback list for the inode depending on its current state and 467ccb26b5aSJan Kara * possibly also change of its state while we were doing writeback. Here we 468ccb26b5aSJan Kara * handle things such as livelock prevention or fairness of writeback among 469ccb26b5aSJan Kara * inodes. This function can be called only by flusher thread - noone else 470ccb26b5aSJan Kara * processes all inodes in writeback lists and requeueing inodes behind flusher 471ccb26b5aSJan Kara * thread's back can have unexpected consequences. 472ccb26b5aSJan Kara */ 473ccb26b5aSJan Kara static void requeue_inode(struct inode *inode, struct bdi_writeback *wb, 474ccb26b5aSJan Kara struct writeback_control *wbc) 475ccb26b5aSJan Kara { 476ccb26b5aSJan Kara if (inode->i_state & I_FREEING) 477ccb26b5aSJan Kara return; 478ccb26b5aSJan Kara 479ccb26b5aSJan Kara /* 480ccb26b5aSJan Kara * Sync livelock prevention. Each inode is tagged and synced in one 481ccb26b5aSJan Kara * shot. If still dirty, it will be redirty_tail()'ed below. Update 482ccb26b5aSJan Kara * the dirty time to prevent enqueue and sync it again. 483ccb26b5aSJan Kara */ 484ccb26b5aSJan Kara if ((inode->i_state & I_DIRTY) && 485ccb26b5aSJan Kara (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)) 486ccb26b5aSJan Kara inode->dirtied_when = jiffies; 487ccb26b5aSJan Kara 4884f8ad655SJan Kara if (wbc->pages_skipped) { 4894f8ad655SJan Kara /* 4904f8ad655SJan Kara * writeback is not making progress due to locked 4914f8ad655SJan Kara * buffers. Skip this inode for now. 4924f8ad655SJan Kara */ 4934f8ad655SJan Kara redirty_tail(inode, wb); 4944f8ad655SJan Kara return; 4954f8ad655SJan Kara } 4964f8ad655SJan Kara 497ccb26b5aSJan Kara if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) { 498ccb26b5aSJan Kara /* 499ccb26b5aSJan Kara * We didn't write back all the pages. nfs_writepages() 500ccb26b5aSJan Kara * sometimes bales out without doing anything. 501ccb26b5aSJan Kara */ 502ccb26b5aSJan Kara if (wbc->nr_to_write <= 0) { 503ccb26b5aSJan Kara /* Slice used up. Queue for next turn. */ 504ccb26b5aSJan Kara requeue_io(inode, wb); 505ccb26b5aSJan Kara } else { 506ccb26b5aSJan Kara /* 507ccb26b5aSJan Kara * Writeback blocked by something other than 508ccb26b5aSJan Kara * congestion. Delay the inode for some time to 509ccb26b5aSJan Kara * avoid spinning on the CPU (100% iowait) 510ccb26b5aSJan Kara * retrying writeback of the dirty page/inode 511ccb26b5aSJan Kara * that cannot be performed immediately. 512ccb26b5aSJan Kara */ 513ccb26b5aSJan Kara redirty_tail(inode, wb); 514ccb26b5aSJan Kara } 515ccb26b5aSJan Kara } else if (inode->i_state & I_DIRTY) { 516ccb26b5aSJan Kara /* 517ccb26b5aSJan Kara * Filesystems can dirty the inode during writeback operations, 518ccb26b5aSJan Kara * such as delayed allocation during submission or metadata 519ccb26b5aSJan Kara * updates after data IO completion. 520ccb26b5aSJan Kara */ 521ccb26b5aSJan Kara redirty_tail(inode, wb); 5220ae45f63STheodore Ts'o } else if (inode->i_state & I_DIRTY_TIME) { 523a2f48706STheodore Ts'o inode->dirtied_when = jiffies; 524d6c10f1fSTejun Heo inode_wb_list_move_locked(inode, wb, &wb->b_dirty_time); 525ccb26b5aSJan Kara } else { 526ccb26b5aSJan Kara /* The inode is clean. Remove from writeback lists. */ 527d6c10f1fSTejun Heo inode_wb_list_del_locked(inode, wb); 528ccb26b5aSJan Kara } 529ccb26b5aSJan Kara } 530ccb26b5aSJan Kara 531ccb26b5aSJan Kara /* 5324f8ad655SJan Kara * Write out an inode and its dirty pages. Do not update the writeback list 5334f8ad655SJan Kara * linkage. That is left to the caller. The caller is also responsible for 5344f8ad655SJan Kara * setting I_SYNC flag and calling inode_sync_complete() to clear it. 5351da177e4SLinus Torvalds */ 5361da177e4SLinus Torvalds static int 537cd8ed2a4SYan Hong __writeback_single_inode(struct inode *inode, struct writeback_control *wbc) 5381da177e4SLinus Torvalds { 5391da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 540251d6a47SWu Fengguang long nr_to_write = wbc->nr_to_write; 54101c03194SChristoph Hellwig unsigned dirty; 5421da177e4SLinus Torvalds int ret; 5431da177e4SLinus Torvalds 5444f8ad655SJan Kara WARN_ON(!(inode->i_state & I_SYNC)); 5451da177e4SLinus Torvalds 5469fb0a7daSTejun Heo trace_writeback_single_inode_start(inode, wbc, nr_to_write); 5479fb0a7daSTejun Heo 5481da177e4SLinus Torvalds ret = do_writepages(mapping, wbc); 5491da177e4SLinus Torvalds 55026821ed4SChristoph Hellwig /* 55126821ed4SChristoph Hellwig * Make sure to wait on the data before writing out the metadata. 55226821ed4SChristoph Hellwig * This is important for filesystems that modify metadata on data 5537747bd4bSDave Chinner * I/O completion. We don't do it for sync(2) writeback because it has a 5547747bd4bSDave Chinner * separate, external IO completion path and ->sync_fs for guaranteeing 5557747bd4bSDave Chinner * inode metadata is written back correctly. 55626821ed4SChristoph Hellwig */ 5577747bd4bSDave Chinner if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) { 55826821ed4SChristoph Hellwig int err = filemap_fdatawait(mapping); 5591da177e4SLinus Torvalds if (ret == 0) 5601da177e4SLinus Torvalds ret = err; 5611da177e4SLinus Torvalds } 5621da177e4SLinus Torvalds 5635547e8aaSDmitry Monakhov /* 5645547e8aaSDmitry Monakhov * Some filesystems may redirty the inode during the writeback 5655547e8aaSDmitry Monakhov * due to delalloc, clear dirty metadata flags right before 5665547e8aaSDmitry Monakhov * write_inode() 5675547e8aaSDmitry Monakhov */ 568250df6edSDave Chinner spin_lock(&inode->i_lock); 5699c6ac78eSTejun Heo 5705547e8aaSDmitry Monakhov dirty = inode->i_state & I_DIRTY; 571a2f48706STheodore Ts'o if (inode->i_state & I_DIRTY_TIME) { 572a2f48706STheodore Ts'o if ((dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) || 573a2f48706STheodore Ts'o unlikely(inode->i_state & I_DIRTY_TIME_EXPIRED) || 574a2f48706STheodore Ts'o unlikely(time_after(jiffies, 575a2f48706STheodore Ts'o (inode->dirtied_time_when + 576a2f48706STheodore Ts'o dirtytime_expire_interval * HZ)))) { 5770ae45f63STheodore Ts'o dirty |= I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED; 5780ae45f63STheodore Ts'o trace_writeback_lazytime(inode); 5790ae45f63STheodore Ts'o } 580a2f48706STheodore Ts'o } else 581a2f48706STheodore Ts'o inode->i_state &= ~I_DIRTY_TIME_EXPIRED; 5820ae45f63STheodore Ts'o inode->i_state &= ~dirty; 5839c6ac78eSTejun Heo 5849c6ac78eSTejun Heo /* 5859c6ac78eSTejun Heo * Paired with smp_mb() in __mark_inode_dirty(). This allows 5869c6ac78eSTejun Heo * __mark_inode_dirty() to test i_state without grabbing i_lock - 5879c6ac78eSTejun Heo * either they see the I_DIRTY bits cleared or we see the dirtied 5889c6ac78eSTejun Heo * inode. 5899c6ac78eSTejun Heo * 5909c6ac78eSTejun Heo * I_DIRTY_PAGES is always cleared together above even if @mapping 5919c6ac78eSTejun Heo * still has dirty pages. The flag is reinstated after smp_mb() if 5929c6ac78eSTejun Heo * necessary. This guarantees that either __mark_inode_dirty() 5939c6ac78eSTejun Heo * sees clear I_DIRTY_PAGES or we see PAGECACHE_TAG_DIRTY. 5949c6ac78eSTejun Heo */ 5959c6ac78eSTejun Heo smp_mb(); 5969c6ac78eSTejun Heo 5979c6ac78eSTejun Heo if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) 5989c6ac78eSTejun Heo inode->i_state |= I_DIRTY_PAGES; 5999c6ac78eSTejun Heo 600250df6edSDave Chinner spin_unlock(&inode->i_lock); 6019c6ac78eSTejun Heo 6020ae45f63STheodore Ts'o if (dirty & I_DIRTY_TIME) 6030ae45f63STheodore Ts'o mark_inode_dirty_sync(inode); 60426821ed4SChristoph Hellwig /* Don't write the inode if only I_DIRTY_PAGES was set */ 6050ae45f63STheodore Ts'o if (dirty & ~I_DIRTY_PAGES) { 606a9185b41SChristoph Hellwig int err = write_inode(inode, wbc); 6071da177e4SLinus Torvalds if (ret == 0) 6081da177e4SLinus Torvalds ret = err; 6091da177e4SLinus Torvalds } 6104f8ad655SJan Kara trace_writeback_single_inode(inode, wbc, nr_to_write); 6114f8ad655SJan Kara return ret; 6124f8ad655SJan Kara } 6134f8ad655SJan Kara 6144f8ad655SJan Kara /* 6154f8ad655SJan Kara * Write out an inode's dirty pages. Either the caller has an active reference 6164f8ad655SJan Kara * on the inode or the inode has I_WILL_FREE set. 6174f8ad655SJan Kara * 6184f8ad655SJan Kara * This function is designed to be called for writing back one inode which 6194f8ad655SJan Kara * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode() 6204f8ad655SJan Kara * and does more profound writeback list handling in writeback_sb_inodes(). 6214f8ad655SJan Kara */ 6224f8ad655SJan Kara static int 6234f8ad655SJan Kara writeback_single_inode(struct inode *inode, struct bdi_writeback *wb, 6244f8ad655SJan Kara struct writeback_control *wbc) 6254f8ad655SJan Kara { 6264f8ad655SJan Kara int ret = 0; 6274f8ad655SJan Kara 6284f8ad655SJan Kara spin_lock(&inode->i_lock); 6294f8ad655SJan Kara if (!atomic_read(&inode->i_count)) 6304f8ad655SJan Kara WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); 6314f8ad655SJan Kara else 6324f8ad655SJan Kara WARN_ON(inode->i_state & I_WILL_FREE); 6334f8ad655SJan Kara 6344f8ad655SJan Kara if (inode->i_state & I_SYNC) { 6354f8ad655SJan Kara if (wbc->sync_mode != WB_SYNC_ALL) 6364f8ad655SJan Kara goto out; 6374f8ad655SJan Kara /* 638169ebd90SJan Kara * It's a data-integrity sync. We must wait. Since callers hold 639169ebd90SJan Kara * inode reference or inode has I_WILL_FREE set, it cannot go 640169ebd90SJan Kara * away under us. 6414f8ad655SJan Kara */ 642169ebd90SJan Kara __inode_wait_for_writeback(inode); 6434f8ad655SJan Kara } 6444f8ad655SJan Kara WARN_ON(inode->i_state & I_SYNC); 6454f8ad655SJan Kara /* 646f9b0e058SJan Kara * Skip inode if it is clean and we have no outstanding writeback in 647f9b0e058SJan Kara * WB_SYNC_ALL mode. We don't want to mess with writeback lists in this 648f9b0e058SJan Kara * function since flusher thread may be doing for example sync in 649f9b0e058SJan Kara * parallel and if we move the inode, it could get skipped. So here we 650f9b0e058SJan Kara * make sure inode is on some writeback list and leave it there unless 651f9b0e058SJan Kara * we have completely cleaned the inode. 6524f8ad655SJan Kara */ 6530ae45f63STheodore Ts'o if (!(inode->i_state & I_DIRTY_ALL) && 654f9b0e058SJan Kara (wbc->sync_mode != WB_SYNC_ALL || 655f9b0e058SJan Kara !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK))) 6564f8ad655SJan Kara goto out; 6574f8ad655SJan Kara inode->i_state |= I_SYNC; 6584f8ad655SJan Kara spin_unlock(&inode->i_lock); 6594f8ad655SJan Kara 660cd8ed2a4SYan Hong ret = __writeback_single_inode(inode, wbc); 6611da177e4SLinus Torvalds 662f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 663250df6edSDave Chinner spin_lock(&inode->i_lock); 6644f8ad655SJan Kara /* 6654f8ad655SJan Kara * If inode is clean, remove it from writeback lists. Otherwise don't 6664f8ad655SJan Kara * touch it. See comment above for explanation. 6674f8ad655SJan Kara */ 6680ae45f63STheodore Ts'o if (!(inode->i_state & I_DIRTY_ALL)) 669d6c10f1fSTejun Heo inode_wb_list_del_locked(inode, wb); 6704f8ad655SJan Kara spin_unlock(&wb->list_lock); 6711c0eeaf5SJoern Engel inode_sync_complete(inode); 6724f8ad655SJan Kara out: 6734f8ad655SJan Kara spin_unlock(&inode->i_lock); 6741da177e4SLinus Torvalds return ret; 6751da177e4SLinus Torvalds } 6761da177e4SLinus Torvalds 677a88a341aSTejun Heo static long writeback_chunk_size(struct bdi_writeback *wb, 6781a12d8bdSWu Fengguang struct wb_writeback_work *work) 679d46db3d5SWu Fengguang { 680d46db3d5SWu Fengguang long pages; 681d46db3d5SWu Fengguang 682d46db3d5SWu Fengguang /* 683d46db3d5SWu Fengguang * WB_SYNC_ALL mode does livelock avoidance by syncing dirty 684d46db3d5SWu Fengguang * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX 685d46db3d5SWu Fengguang * here avoids calling into writeback_inodes_wb() more than once. 686d46db3d5SWu Fengguang * 687d46db3d5SWu Fengguang * The intended call sequence for WB_SYNC_ALL writeback is: 688d46db3d5SWu Fengguang * 689d46db3d5SWu Fengguang * wb_writeback() 690d46db3d5SWu Fengguang * writeback_sb_inodes() <== called only once 691d46db3d5SWu Fengguang * write_cache_pages() <== called once for each inode 692d46db3d5SWu Fengguang * (quickly) tag currently dirty pages 693d46db3d5SWu Fengguang * (maybe slowly) sync all tagged pages 694d46db3d5SWu Fengguang */ 695d46db3d5SWu Fengguang if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages) 696d46db3d5SWu Fengguang pages = LONG_MAX; 6971a12d8bdSWu Fengguang else { 698a88a341aSTejun Heo pages = min(wb->avg_write_bandwidth / 2, 6991a12d8bdSWu Fengguang global_dirty_limit / DIRTY_SCOPE); 7001a12d8bdSWu Fengguang pages = min(pages, work->nr_pages); 7011a12d8bdSWu Fengguang pages = round_down(pages + MIN_WRITEBACK_PAGES, 7021a12d8bdSWu Fengguang MIN_WRITEBACK_PAGES); 7031a12d8bdSWu Fengguang } 704d46db3d5SWu Fengguang 705d46db3d5SWu Fengguang return pages; 706d46db3d5SWu Fengguang } 707d46db3d5SWu Fengguang 70803ba3782SJens Axboe /* 709f11c9c5cSEdward Shishkin * Write a portion of b_io inodes which belong to @sb. 710edadfb10SChristoph Hellwig * 711d46db3d5SWu Fengguang * Return the number of pages and/or inodes written. 712f11c9c5cSEdward Shishkin */ 713d46db3d5SWu Fengguang static long writeback_sb_inodes(struct super_block *sb, 714d46db3d5SWu Fengguang struct bdi_writeback *wb, 715d46db3d5SWu Fengguang struct wb_writeback_work *work) 71603ba3782SJens Axboe { 717d46db3d5SWu Fengguang struct writeback_control wbc = { 718d46db3d5SWu Fengguang .sync_mode = work->sync_mode, 719d46db3d5SWu Fengguang .tagged_writepages = work->tagged_writepages, 720d46db3d5SWu Fengguang .for_kupdate = work->for_kupdate, 721d46db3d5SWu Fengguang .for_background = work->for_background, 7227747bd4bSDave Chinner .for_sync = work->for_sync, 723d46db3d5SWu Fengguang .range_cyclic = work->range_cyclic, 724d46db3d5SWu Fengguang .range_start = 0, 725d46db3d5SWu Fengguang .range_end = LLONG_MAX, 726d46db3d5SWu Fengguang }; 727d46db3d5SWu Fengguang unsigned long start_time = jiffies; 728d46db3d5SWu Fengguang long write_chunk; 729d46db3d5SWu Fengguang long wrote = 0; /* count both pages and inodes */ 730d46db3d5SWu Fengguang 73103ba3782SJens Axboe while (!list_empty(&wb->b_io)) { 7327ccf19a8SNick Piggin struct inode *inode = wb_inode(wb->b_io.prev); 733edadfb10SChristoph Hellwig 734edadfb10SChristoph Hellwig if (inode->i_sb != sb) { 735d46db3d5SWu Fengguang if (work->sb) { 736edadfb10SChristoph Hellwig /* 737edadfb10SChristoph Hellwig * We only want to write back data for this 738edadfb10SChristoph Hellwig * superblock, move all inodes not belonging 739edadfb10SChristoph Hellwig * to it back onto the dirty list. 740edadfb10SChristoph Hellwig */ 741f758eeabSChristoph Hellwig redirty_tail(inode, wb); 74266f3b8e2SJens Axboe continue; 74366f3b8e2SJens Axboe } 744edadfb10SChristoph Hellwig 745edadfb10SChristoph Hellwig /* 746edadfb10SChristoph Hellwig * The inode belongs to a different superblock. 747edadfb10SChristoph Hellwig * Bounce back to the caller to unpin this and 748edadfb10SChristoph Hellwig * pin the next superblock. 749edadfb10SChristoph Hellwig */ 750d46db3d5SWu Fengguang break; 751edadfb10SChristoph Hellwig } 752edadfb10SChristoph Hellwig 7539843b76aSChristoph Hellwig /* 754331cbdeeSWanpeng Li * Don't bother with new inodes or inodes being freed, first 755331cbdeeSWanpeng Li * kind does not need periodic writeout yet, and for the latter 7569843b76aSChristoph Hellwig * kind writeout is handled by the freer. 7579843b76aSChristoph Hellwig */ 758250df6edSDave Chinner spin_lock(&inode->i_lock); 7599843b76aSChristoph Hellwig if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { 760250df6edSDave Chinner spin_unlock(&inode->i_lock); 761fcc5c222SWu Fengguang redirty_tail(inode, wb); 7627ef0d737SNick Piggin continue; 7637ef0d737SNick Piggin } 764cc1676d9SJan Kara if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) { 765cc1676d9SJan Kara /* 766cc1676d9SJan Kara * If this inode is locked for writeback and we are not 767cc1676d9SJan Kara * doing writeback-for-data-integrity, move it to 768cc1676d9SJan Kara * b_more_io so that writeback can proceed with the 769cc1676d9SJan Kara * other inodes on s_io. 770cc1676d9SJan Kara * 771cc1676d9SJan Kara * We'll have another go at writing back this inode 772cc1676d9SJan Kara * when we completed a full scan of b_io. 773cc1676d9SJan Kara */ 774cc1676d9SJan Kara spin_unlock(&inode->i_lock); 775cc1676d9SJan Kara requeue_io(inode, wb); 776cc1676d9SJan Kara trace_writeback_sb_inodes_requeue(inode); 777cc1676d9SJan Kara continue; 778cc1676d9SJan Kara } 779f0d07b7fSJan Kara spin_unlock(&wb->list_lock); 780f0d07b7fSJan Kara 7814f8ad655SJan Kara /* 7824f8ad655SJan Kara * We already requeued the inode if it had I_SYNC set and we 7834f8ad655SJan Kara * are doing WB_SYNC_NONE writeback. So this catches only the 7844f8ad655SJan Kara * WB_SYNC_ALL case. 7854f8ad655SJan Kara */ 786169ebd90SJan Kara if (inode->i_state & I_SYNC) { 787169ebd90SJan Kara /* Wait for I_SYNC. This function drops i_lock... */ 788169ebd90SJan Kara inode_sleep_on_writeback(inode); 789169ebd90SJan Kara /* Inode may be gone, start again */ 790ead188f9SJan Kara spin_lock(&wb->list_lock); 791169ebd90SJan Kara continue; 792169ebd90SJan Kara } 7934f8ad655SJan Kara inode->i_state |= I_SYNC; 7944f8ad655SJan Kara spin_unlock(&inode->i_lock); 795169ebd90SJan Kara 796a88a341aSTejun Heo write_chunk = writeback_chunk_size(wb, work); 797d46db3d5SWu Fengguang wbc.nr_to_write = write_chunk; 798d46db3d5SWu Fengguang wbc.pages_skipped = 0; 799250df6edSDave Chinner 800169ebd90SJan Kara /* 801169ebd90SJan Kara * We use I_SYNC to pin the inode in memory. While it is set 802169ebd90SJan Kara * evict_inode() will wait so the inode cannot be freed. 803169ebd90SJan Kara */ 804cd8ed2a4SYan Hong __writeback_single_inode(inode, &wbc); 805d46db3d5SWu Fengguang 806d46db3d5SWu Fengguang work->nr_pages -= write_chunk - wbc.nr_to_write; 807d46db3d5SWu Fengguang wrote += write_chunk - wbc.nr_to_write; 8084f8ad655SJan Kara spin_lock(&wb->list_lock); 8094f8ad655SJan Kara spin_lock(&inode->i_lock); 8100ae45f63STheodore Ts'o if (!(inode->i_state & I_DIRTY_ALL)) 811d46db3d5SWu Fengguang wrote++; 8124f8ad655SJan Kara requeue_inode(inode, wb, &wbc); 8134f8ad655SJan Kara inode_sync_complete(inode); 8140f1b1fd8SDave Chinner spin_unlock(&inode->i_lock); 815169ebd90SJan Kara cond_resched_lock(&wb->list_lock); 816d46db3d5SWu Fengguang /* 817d46db3d5SWu Fengguang * bail out to wb_writeback() often enough to check 818d46db3d5SWu Fengguang * background threshold and other termination conditions. 819d46db3d5SWu Fengguang */ 820d46db3d5SWu Fengguang if (wrote) { 821d46db3d5SWu Fengguang if (time_is_before_jiffies(start_time + HZ / 10UL)) 822d46db3d5SWu Fengguang break; 823d46db3d5SWu Fengguang if (work->nr_pages <= 0) 824d46db3d5SWu Fengguang break; 8251da177e4SLinus Torvalds } 8268bc3be27SFengguang Wu } 827d46db3d5SWu Fengguang return wrote; 828f11c9c5cSEdward Shishkin } 82938f21977SNick Piggin 830d46db3d5SWu Fengguang static long __writeback_inodes_wb(struct bdi_writeback *wb, 831d46db3d5SWu Fengguang struct wb_writeback_work *work) 832f11c9c5cSEdward Shishkin { 833d46db3d5SWu Fengguang unsigned long start_time = jiffies; 834d46db3d5SWu Fengguang long wrote = 0; 835f11c9c5cSEdward Shishkin 836f11c9c5cSEdward Shishkin while (!list_empty(&wb->b_io)) { 8377ccf19a8SNick Piggin struct inode *inode = wb_inode(wb->b_io.prev); 838f11c9c5cSEdward Shishkin struct super_block *sb = inode->i_sb; 839f11c9c5cSEdward Shishkin 840eb6ef3dfSKonstantin Khlebnikov if (!trylock_super(sb)) { 8410e995816SWu Fengguang /* 842eb6ef3dfSKonstantin Khlebnikov * trylock_super() may fail consistently due to 8430e995816SWu Fengguang * s_umount being grabbed by someone else. Don't use 8440e995816SWu Fengguang * requeue_io() to avoid busy retrying the inode/sb. 8450e995816SWu Fengguang */ 8460e995816SWu Fengguang redirty_tail(inode, wb); 847d19de7edSChristoph Hellwig continue; 848334132aeSChristoph Hellwig } 849d46db3d5SWu Fengguang wrote += writeback_sb_inodes(sb, wb, work); 850eb6ef3dfSKonstantin Khlebnikov up_read(&sb->s_umount); 851f11c9c5cSEdward Shishkin 852d46db3d5SWu Fengguang /* refer to the same tests at the end of writeback_sb_inodes */ 853d46db3d5SWu Fengguang if (wrote) { 854d46db3d5SWu Fengguang if (time_is_before_jiffies(start_time + HZ / 10UL)) 855d46db3d5SWu Fengguang break; 856d46db3d5SWu Fengguang if (work->nr_pages <= 0) 857f11c9c5cSEdward Shishkin break; 858f11c9c5cSEdward Shishkin } 859d46db3d5SWu Fengguang } 86066f3b8e2SJens Axboe /* Leave any unwritten inodes on b_io */ 861d46db3d5SWu Fengguang return wrote; 86266f3b8e2SJens Axboe } 86366f3b8e2SJens Axboe 8647d9f073bSWanpeng Li static long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages, 8650e175a18SCurt Wohlgemuth enum wb_reason reason) 866edadfb10SChristoph Hellwig { 867d46db3d5SWu Fengguang struct wb_writeback_work work = { 868d46db3d5SWu Fengguang .nr_pages = nr_pages, 869d46db3d5SWu Fengguang .sync_mode = WB_SYNC_NONE, 870d46db3d5SWu Fengguang .range_cyclic = 1, 8710e175a18SCurt Wohlgemuth .reason = reason, 872d46db3d5SWu Fengguang }; 873edadfb10SChristoph Hellwig 874f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 875424b351fSWu Fengguang if (list_empty(&wb->b_io)) 876ad4e38ddSCurt Wohlgemuth queue_io(wb, &work); 877d46db3d5SWu Fengguang __writeback_inodes_wb(wb, &work); 878f758eeabSChristoph Hellwig spin_unlock(&wb->list_lock); 879edadfb10SChristoph Hellwig 880d46db3d5SWu Fengguang return nr_pages - work.nr_pages; 88166f3b8e2SJens Axboe } 88266f3b8e2SJens Axboe 883a88a341aSTejun Heo static bool over_bground_thresh(struct bdi_writeback *wb) 88403ba3782SJens Axboe { 88503ba3782SJens Axboe unsigned long background_thresh, dirty_thresh; 88603ba3782SJens Axboe 88716c4042fSWu Fengguang global_dirty_limits(&background_thresh, &dirty_thresh); 88803ba3782SJens Axboe 889b00949aaSWu Fengguang if (global_page_state(NR_FILE_DIRTY) + 890b00949aaSWu Fengguang global_page_state(NR_UNSTABLE_NFS) > background_thresh) 891b00949aaSWu Fengguang return true; 892b00949aaSWu Fengguang 893a88a341aSTejun Heo if (wb_stat(wb, WB_RECLAIMABLE) > wb_dirty_limit(wb, background_thresh)) 894b00949aaSWu Fengguang return true; 895b00949aaSWu Fengguang 896b00949aaSWu Fengguang return false; 89703ba3782SJens Axboe } 89803ba3782SJens Axboe 89903ba3782SJens Axboe /* 900e98be2d5SWu Fengguang * Called under wb->list_lock. If there are multiple wb per bdi, 901e98be2d5SWu Fengguang * only the flusher working on the first wb should do it. 902e98be2d5SWu Fengguang */ 903e98be2d5SWu Fengguang static void wb_update_bandwidth(struct bdi_writeback *wb, 904e98be2d5SWu Fengguang unsigned long start_time) 905e98be2d5SWu Fengguang { 906a88a341aSTejun Heo __wb_update_bandwidth(wb, 0, 0, 0, 0, 0, start_time); 907e98be2d5SWu Fengguang } 908e98be2d5SWu Fengguang 909e98be2d5SWu Fengguang /* 91003ba3782SJens Axboe * Explicit flushing or periodic writeback of "old" data. 91103ba3782SJens Axboe * 91203ba3782SJens Axboe * Define "old": the first time one of an inode's pages is dirtied, we mark the 91303ba3782SJens Axboe * dirtying-time in the inode's address_space. So this periodic writeback code 91403ba3782SJens Axboe * just walks the superblock inode list, writing back any inodes which are 91503ba3782SJens Axboe * older than a specific point in time. 91603ba3782SJens Axboe * 91703ba3782SJens Axboe * Try to run once per dirty_writeback_interval. But if a writeback event 91803ba3782SJens Axboe * takes longer than a dirty_writeback_interval interval, then leave a 91903ba3782SJens Axboe * one-second gap. 92003ba3782SJens Axboe * 92103ba3782SJens Axboe * older_than_this takes precedence over nr_to_write. So we'll only write back 92203ba3782SJens Axboe * all dirty pages if they are all attached to "old" mappings. 92303ba3782SJens Axboe */ 924c4a77a6cSJens Axboe static long wb_writeback(struct bdi_writeback *wb, 92583ba7b07SChristoph Hellwig struct wb_writeback_work *work) 92603ba3782SJens Axboe { 927e98be2d5SWu Fengguang unsigned long wb_start = jiffies; 928d46db3d5SWu Fengguang long nr_pages = work->nr_pages; 9290dc83bd3SJan Kara unsigned long oldest_jif; 930a5989bdcSJan Kara struct inode *inode; 931d46db3d5SWu Fengguang long progress; 93203ba3782SJens Axboe 9330dc83bd3SJan Kara oldest_jif = jiffies; 9340dc83bd3SJan Kara work->older_than_this = &oldest_jif; 93503ba3782SJens Axboe 936e8dfc305SWu Fengguang spin_lock(&wb->list_lock); 93703ba3782SJens Axboe for (;;) { 93803ba3782SJens Axboe /* 939d3ddec76SWu Fengguang * Stop writeback when nr_pages has been consumed 94003ba3782SJens Axboe */ 94183ba7b07SChristoph Hellwig if (work->nr_pages <= 0) 94203ba3782SJens Axboe break; 94303ba3782SJens Axboe 94403ba3782SJens Axboe /* 945aa373cf5SJan Kara * Background writeout and kupdate-style writeback may 946aa373cf5SJan Kara * run forever. Stop them if there is other work to do 947aa373cf5SJan Kara * so that e.g. sync can proceed. They'll be restarted 948aa373cf5SJan Kara * after the other works are all done. 949aa373cf5SJan Kara */ 950aa373cf5SJan Kara if ((work->for_background || work->for_kupdate) && 951f0054bb1STejun Heo !list_empty(&wb->work_list)) 952aa373cf5SJan Kara break; 953aa373cf5SJan Kara 954aa373cf5SJan Kara /* 955d3ddec76SWu Fengguang * For background writeout, stop when we are below the 956d3ddec76SWu Fengguang * background dirty threshold 95703ba3782SJens Axboe */ 958a88a341aSTejun Heo if (work->for_background && !over_bground_thresh(wb)) 95903ba3782SJens Axboe break; 96003ba3782SJens Axboe 9611bc36b64SJan Kara /* 9621bc36b64SJan Kara * Kupdate and background works are special and we want to 9631bc36b64SJan Kara * include all inodes that need writing. Livelock avoidance is 9641bc36b64SJan Kara * handled by these works yielding to any other work so we are 9651bc36b64SJan Kara * safe. 9661bc36b64SJan Kara */ 967ba9aa839SWu Fengguang if (work->for_kupdate) { 9680dc83bd3SJan Kara oldest_jif = jiffies - 969ba9aa839SWu Fengguang msecs_to_jiffies(dirty_expire_interval * 10); 9701bc36b64SJan Kara } else if (work->for_background) 9710dc83bd3SJan Kara oldest_jif = jiffies; 972028c2dd1SDave Chinner 973d46db3d5SWu Fengguang trace_writeback_start(wb->bdi, work); 974e8dfc305SWu Fengguang if (list_empty(&wb->b_io)) 975ad4e38ddSCurt Wohlgemuth queue_io(wb, work); 97683ba7b07SChristoph Hellwig if (work->sb) 977d46db3d5SWu Fengguang progress = writeback_sb_inodes(work->sb, wb, work); 978edadfb10SChristoph Hellwig else 979d46db3d5SWu Fengguang progress = __writeback_inodes_wb(wb, work); 980d46db3d5SWu Fengguang trace_writeback_written(wb->bdi, work); 981028c2dd1SDave Chinner 982e98be2d5SWu Fengguang wb_update_bandwidth(wb, wb_start); 98303ba3782SJens Axboe 98403ba3782SJens Axboe /* 98571fd05a8SJens Axboe * Did we write something? Try for more 986e6fb6da2SWu Fengguang * 987e6fb6da2SWu Fengguang * Dirty inodes are moved to b_io for writeback in batches. 988e6fb6da2SWu Fengguang * The completion of the current batch does not necessarily 989e6fb6da2SWu Fengguang * mean the overall work is done. So we keep looping as long 990e6fb6da2SWu Fengguang * as made some progress on cleaning pages or inodes. 99171fd05a8SJens Axboe */ 992d46db3d5SWu Fengguang if (progress) 99303ba3782SJens Axboe continue; 994a5989bdcSJan Kara /* 995e6fb6da2SWu Fengguang * No more inodes for IO, bail 996a5989bdcSJan Kara */ 997b7a2441fSWu Fengguang if (list_empty(&wb->b_more_io)) 99803ba3782SJens Axboe break; 99903ba3782SJens Axboe /* 10008010c3b6SJens Axboe * Nothing written. Wait for some inode to 10018010c3b6SJens Axboe * become available for writeback. Otherwise 10028010c3b6SJens Axboe * we'll just busyloop. 100303ba3782SJens Axboe */ 100403ba3782SJens Axboe if (!list_empty(&wb->b_more_io)) { 1005d46db3d5SWu Fengguang trace_writeback_wait(wb->bdi, work); 100603ba3782SJens Axboe inode = wb_inode(wb->b_more_io.prev); 1007250df6edSDave Chinner spin_lock(&inode->i_lock); 1008f0d07b7fSJan Kara spin_unlock(&wb->list_lock); 1009169ebd90SJan Kara /* This function drops i_lock... */ 1010169ebd90SJan Kara inode_sleep_on_writeback(inode); 1011f0d07b7fSJan Kara spin_lock(&wb->list_lock); 101203ba3782SJens Axboe } 101303ba3782SJens Axboe } 1014e8dfc305SWu Fengguang spin_unlock(&wb->list_lock); 101503ba3782SJens Axboe 1016d46db3d5SWu Fengguang return nr_pages - work->nr_pages; 101703ba3782SJens Axboe } 101803ba3782SJens Axboe 101903ba3782SJens Axboe /* 102083ba7b07SChristoph Hellwig * Return the next wb_writeback_work struct that hasn't been processed yet. 102103ba3782SJens Axboe */ 1022f0054bb1STejun Heo static struct wb_writeback_work *get_next_work_item(struct bdi_writeback *wb) 102303ba3782SJens Axboe { 102483ba7b07SChristoph Hellwig struct wb_writeback_work *work = NULL; 102503ba3782SJens Axboe 1026f0054bb1STejun Heo spin_lock_bh(&wb->work_lock); 1027f0054bb1STejun Heo if (!list_empty(&wb->work_list)) { 1028f0054bb1STejun Heo work = list_entry(wb->work_list.next, 102983ba7b07SChristoph Hellwig struct wb_writeback_work, list); 103083ba7b07SChristoph Hellwig list_del_init(&work->list); 103103ba3782SJens Axboe } 1032f0054bb1STejun Heo spin_unlock_bh(&wb->work_lock); 103383ba7b07SChristoph Hellwig return work; 103403ba3782SJens Axboe } 103503ba3782SJens Axboe 1036cdf01dd5SLinus Torvalds /* 1037cdf01dd5SLinus Torvalds * Add in the number of potentially dirty inodes, because each inode 1038cdf01dd5SLinus Torvalds * write can dirty pagecache in the underlying blockdev. 1039cdf01dd5SLinus Torvalds */ 1040cdf01dd5SLinus Torvalds static unsigned long get_nr_dirty_pages(void) 1041cdf01dd5SLinus Torvalds { 1042cdf01dd5SLinus Torvalds return global_page_state(NR_FILE_DIRTY) + 1043cdf01dd5SLinus Torvalds global_page_state(NR_UNSTABLE_NFS) + 1044cdf01dd5SLinus Torvalds get_nr_dirty_inodes(); 1045cdf01dd5SLinus Torvalds } 1046cdf01dd5SLinus Torvalds 10476585027aSJan Kara static long wb_check_background_flush(struct bdi_writeback *wb) 10486585027aSJan Kara { 1049a88a341aSTejun Heo if (over_bground_thresh(wb)) { 10506585027aSJan Kara 10516585027aSJan Kara struct wb_writeback_work work = { 10526585027aSJan Kara .nr_pages = LONG_MAX, 10536585027aSJan Kara .sync_mode = WB_SYNC_NONE, 10546585027aSJan Kara .for_background = 1, 10556585027aSJan Kara .range_cyclic = 1, 10560e175a18SCurt Wohlgemuth .reason = WB_REASON_BACKGROUND, 10576585027aSJan Kara }; 10586585027aSJan Kara 10596585027aSJan Kara return wb_writeback(wb, &work); 10606585027aSJan Kara } 10616585027aSJan Kara 10626585027aSJan Kara return 0; 10636585027aSJan Kara } 10646585027aSJan Kara 106503ba3782SJens Axboe static long wb_check_old_data_flush(struct bdi_writeback *wb) 106603ba3782SJens Axboe { 106703ba3782SJens Axboe unsigned long expired; 106803ba3782SJens Axboe long nr_pages; 106903ba3782SJens Axboe 107069b62d01SJens Axboe /* 107169b62d01SJens Axboe * When set to zero, disable periodic writeback 107269b62d01SJens Axboe */ 107369b62d01SJens Axboe if (!dirty_writeback_interval) 107469b62d01SJens Axboe return 0; 107569b62d01SJens Axboe 107603ba3782SJens Axboe expired = wb->last_old_flush + 107703ba3782SJens Axboe msecs_to_jiffies(dirty_writeback_interval * 10); 107803ba3782SJens Axboe if (time_before(jiffies, expired)) 107903ba3782SJens Axboe return 0; 108003ba3782SJens Axboe 108103ba3782SJens Axboe wb->last_old_flush = jiffies; 1082cdf01dd5SLinus Torvalds nr_pages = get_nr_dirty_pages(); 108303ba3782SJens Axboe 1084c4a77a6cSJens Axboe if (nr_pages) { 108583ba7b07SChristoph Hellwig struct wb_writeback_work work = { 1086c4a77a6cSJens Axboe .nr_pages = nr_pages, 1087c4a77a6cSJens Axboe .sync_mode = WB_SYNC_NONE, 1088c4a77a6cSJens Axboe .for_kupdate = 1, 1089c4a77a6cSJens Axboe .range_cyclic = 1, 10900e175a18SCurt Wohlgemuth .reason = WB_REASON_PERIODIC, 1091c4a77a6cSJens Axboe }; 1092c4a77a6cSJens Axboe 109383ba7b07SChristoph Hellwig return wb_writeback(wb, &work); 1094c4a77a6cSJens Axboe } 109503ba3782SJens Axboe 109603ba3782SJens Axboe return 0; 109703ba3782SJens Axboe } 109803ba3782SJens Axboe 109903ba3782SJens Axboe /* 110003ba3782SJens Axboe * Retrieve work items and do the writeback they describe 110103ba3782SJens Axboe */ 110225d130baSWanpeng Li static long wb_do_writeback(struct bdi_writeback *wb) 110303ba3782SJens Axboe { 110483ba7b07SChristoph Hellwig struct wb_writeback_work *work; 1105c4a77a6cSJens Axboe long wrote = 0; 110603ba3782SJens Axboe 11074452226eSTejun Heo set_bit(WB_writeback_running, &wb->state); 1108f0054bb1STejun Heo while ((work = get_next_work_item(wb)) != NULL) { 110983ba7b07SChristoph Hellwig 1110f0054bb1STejun Heo trace_writeback_exec(wb->bdi, work); 1111455b2864SDave Chinner 111283ba7b07SChristoph Hellwig wrote += wb_writeback(wb, work); 111303ba3782SJens Axboe 111403ba3782SJens Axboe /* 111583ba7b07SChristoph Hellwig * Notify the caller of completion if this is a synchronous 111683ba7b07SChristoph Hellwig * work item, otherwise just free it. 111703ba3782SJens Axboe */ 111883ba7b07SChristoph Hellwig if (work->done) 111983ba7b07SChristoph Hellwig complete(work->done); 112083ba7b07SChristoph Hellwig else 112183ba7b07SChristoph Hellwig kfree(work); 112203ba3782SJens Axboe } 112303ba3782SJens Axboe 112403ba3782SJens Axboe /* 112503ba3782SJens Axboe * Check for periodic writeback, kupdated() style 112603ba3782SJens Axboe */ 112703ba3782SJens Axboe wrote += wb_check_old_data_flush(wb); 11286585027aSJan Kara wrote += wb_check_background_flush(wb); 11294452226eSTejun Heo clear_bit(WB_writeback_running, &wb->state); 113003ba3782SJens Axboe 113103ba3782SJens Axboe return wrote; 113203ba3782SJens Axboe } 113303ba3782SJens Axboe 113403ba3782SJens Axboe /* 113503ba3782SJens Axboe * Handle writeback of dirty data for the device backed by this bdi. Also 1136839a8e86STejun Heo * reschedules periodically and does kupdated style flushing. 113703ba3782SJens Axboe */ 1138f0054bb1STejun Heo void wb_workfn(struct work_struct *work) 113903ba3782SJens Axboe { 1140839a8e86STejun Heo struct bdi_writeback *wb = container_of(to_delayed_work(work), 1141839a8e86STejun Heo struct bdi_writeback, dwork); 114203ba3782SJens Axboe long pages_written; 114303ba3782SJens Axboe 1144f0054bb1STejun Heo set_worker_desc("flush-%s", dev_name(wb->bdi->dev)); 1145766f9164SPeter Zijlstra current->flags |= PF_SWAPWRITE; 114603ba3782SJens Axboe 1147839a8e86STejun Heo if (likely(!current_is_workqueue_rescuer() || 11484452226eSTejun Heo !test_bit(WB_registered, &wb->state))) { 114903ba3782SJens Axboe /* 1150f0054bb1STejun Heo * The normal path. Keep writing back @wb until its 1151839a8e86STejun Heo * work_list is empty. Note that this path is also taken 1152f0054bb1STejun Heo * if @wb is shutting down even when we're running off the 1153839a8e86STejun Heo * rescuer as work_list needs to be drained. 115403ba3782SJens Axboe */ 1155839a8e86STejun Heo do { 115625d130baSWanpeng Li pages_written = wb_do_writeback(wb); 1157455b2864SDave Chinner trace_writeback_pages_written(pages_written); 1158f0054bb1STejun Heo } while (!list_empty(&wb->work_list)); 1159839a8e86STejun Heo } else { 1160253c34e9SArtem Bityutskiy /* 1161839a8e86STejun Heo * bdi_wq can't get enough workers and we're running off 1162839a8e86STejun Heo * the emergency worker. Don't hog it. Hopefully, 1024 is 1163839a8e86STejun Heo * enough for efficient IO. 1164253c34e9SArtem Bityutskiy */ 1165f0054bb1STejun Heo pages_written = writeback_inodes_wb(wb, 1024, 1166839a8e86STejun Heo WB_REASON_FORKER_THREAD); 1167839a8e86STejun Heo trace_writeback_pages_written(pages_written); 116803ba3782SJens Axboe } 116903ba3782SJens Axboe 1170f0054bb1STejun Heo if (!list_empty(&wb->work_list)) 11716ca738d6SDerek Basehore mod_delayed_work(bdi_wq, &wb->dwork, 0); 11726ca738d6SDerek Basehore else if (wb_has_dirty_io(wb) && dirty_writeback_interval) 1173f0054bb1STejun Heo wb_wakeup_delayed(wb); 1174455b2864SDave Chinner 1175839a8e86STejun Heo current->flags &= ~PF_SWAPWRITE; 117603ba3782SJens Axboe } 117703ba3782SJens Axboe 117803ba3782SJens Axboe /* 117903ba3782SJens Axboe * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back 118003ba3782SJens Axboe * the whole world. 118103ba3782SJens Axboe */ 11820e175a18SCurt Wohlgemuth void wakeup_flusher_threads(long nr_pages, enum wb_reason reason) 118303ba3782SJens Axboe { 1184b8c2f347SChristoph Hellwig struct backing_dev_info *bdi; 1185b8c2f347SChristoph Hellwig 118647df3ddeSJan Kara if (!nr_pages) 118747df3ddeSJan Kara nr_pages = get_nr_dirty_pages(); 1188b8c2f347SChristoph Hellwig 1189b8c2f347SChristoph Hellwig rcu_read_lock(); 1190e7972912STejun Heo list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) 1191c00ddad3STejun Heo wb_start_writeback(&bdi->wb, nr_pages, false, reason); 1192b8c2f347SChristoph Hellwig rcu_read_unlock(); 119303ba3782SJens Axboe } 119403ba3782SJens Axboe 1195a2f48706STheodore Ts'o /* 1196a2f48706STheodore Ts'o * Wake up bdi's periodically to make sure dirtytime inodes gets 1197a2f48706STheodore Ts'o * written back periodically. We deliberately do *not* check the 1198a2f48706STheodore Ts'o * b_dirtytime list in wb_has_dirty_io(), since this would cause the 1199a2f48706STheodore Ts'o * kernel to be constantly waking up once there are any dirtytime 1200a2f48706STheodore Ts'o * inodes on the system. So instead we define a separate delayed work 1201a2f48706STheodore Ts'o * function which gets called much more rarely. (By default, only 1202a2f48706STheodore Ts'o * once every 12 hours.) 1203a2f48706STheodore Ts'o * 1204a2f48706STheodore Ts'o * If there is any other write activity going on in the file system, 1205a2f48706STheodore Ts'o * this function won't be necessary. But if the only thing that has 1206a2f48706STheodore Ts'o * happened on the file system is a dirtytime inode caused by an atime 1207a2f48706STheodore Ts'o * update, we need this infrastructure below to make sure that inode 1208a2f48706STheodore Ts'o * eventually gets pushed out to disk. 1209a2f48706STheodore Ts'o */ 1210a2f48706STheodore Ts'o static void wakeup_dirtytime_writeback(struct work_struct *w); 1211a2f48706STheodore Ts'o static DECLARE_DELAYED_WORK(dirtytime_work, wakeup_dirtytime_writeback); 1212a2f48706STheodore Ts'o 1213a2f48706STheodore Ts'o static void wakeup_dirtytime_writeback(struct work_struct *w) 1214a2f48706STheodore Ts'o { 1215a2f48706STheodore Ts'o struct backing_dev_info *bdi; 1216a2f48706STheodore Ts'o 1217a2f48706STheodore Ts'o rcu_read_lock(); 1218a2f48706STheodore Ts'o list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { 1219a2f48706STheodore Ts'o if (list_empty(&bdi->wb.b_dirty_time)) 1220a2f48706STheodore Ts'o continue; 1221f0054bb1STejun Heo wb_wakeup(&bdi->wb); 1222a2f48706STheodore Ts'o } 1223a2f48706STheodore Ts'o rcu_read_unlock(); 1224a2f48706STheodore Ts'o schedule_delayed_work(&dirtytime_work, dirtytime_expire_interval * HZ); 1225a2f48706STheodore Ts'o } 1226a2f48706STheodore Ts'o 1227a2f48706STheodore Ts'o static int __init start_dirtytime_writeback(void) 1228a2f48706STheodore Ts'o { 1229a2f48706STheodore Ts'o schedule_delayed_work(&dirtytime_work, dirtytime_expire_interval * HZ); 1230a2f48706STheodore Ts'o return 0; 1231a2f48706STheodore Ts'o } 1232a2f48706STheodore Ts'o __initcall(start_dirtytime_writeback); 1233a2f48706STheodore Ts'o 12341efff914STheodore Ts'o int dirtytime_interval_handler(struct ctl_table *table, int write, 12351efff914STheodore Ts'o void __user *buffer, size_t *lenp, loff_t *ppos) 12361efff914STheodore Ts'o { 12371efff914STheodore Ts'o int ret; 12381efff914STheodore Ts'o 12391efff914STheodore Ts'o ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos); 12401efff914STheodore Ts'o if (ret == 0 && write) 12411efff914STheodore Ts'o mod_delayed_work(system_wq, &dirtytime_work, 0); 12421efff914STheodore Ts'o return ret; 12431efff914STheodore Ts'o } 12441efff914STheodore Ts'o 124503ba3782SJens Axboe static noinline void block_dump___mark_inode_dirty(struct inode *inode) 124603ba3782SJens Axboe { 124703ba3782SJens Axboe if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { 124803ba3782SJens Axboe struct dentry *dentry; 124903ba3782SJens Axboe const char *name = "?"; 125003ba3782SJens Axboe 125103ba3782SJens Axboe dentry = d_find_alias(inode); 125203ba3782SJens Axboe if (dentry) { 125303ba3782SJens Axboe spin_lock(&dentry->d_lock); 125403ba3782SJens Axboe name = (const char *) dentry->d_name.name; 125503ba3782SJens Axboe } 125603ba3782SJens Axboe printk(KERN_DEBUG 125703ba3782SJens Axboe "%s(%d): dirtied inode %lu (%s) on %s\n", 125803ba3782SJens Axboe current->comm, task_pid_nr(current), inode->i_ino, 125903ba3782SJens Axboe name, inode->i_sb->s_id); 126003ba3782SJens Axboe if (dentry) { 126103ba3782SJens Axboe spin_unlock(&dentry->d_lock); 126203ba3782SJens Axboe dput(dentry); 126303ba3782SJens Axboe } 126403ba3782SJens Axboe } 126503ba3782SJens Axboe } 126603ba3782SJens Axboe 126703ba3782SJens Axboe /** 126803ba3782SJens Axboe * __mark_inode_dirty - internal function 126903ba3782SJens Axboe * @inode: inode to mark 127003ba3782SJens Axboe * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) 127103ba3782SJens Axboe * Mark an inode as dirty. Callers should use mark_inode_dirty or 127203ba3782SJens Axboe * mark_inode_dirty_sync. 127303ba3782SJens Axboe * 127403ba3782SJens Axboe * Put the inode on the super block's dirty list. 127503ba3782SJens Axboe * 127603ba3782SJens Axboe * CAREFUL! We mark it dirty unconditionally, but move it onto the 127703ba3782SJens Axboe * dirty list only if it is hashed or if it refers to a blockdev. 127803ba3782SJens Axboe * If it was not hashed, it will never be added to the dirty list 127903ba3782SJens Axboe * even if it is later hashed, as it will have been marked dirty already. 128003ba3782SJens Axboe * 128103ba3782SJens Axboe * In short, make sure you hash any inodes _before_ you start marking 128203ba3782SJens Axboe * them dirty. 128303ba3782SJens Axboe * 128403ba3782SJens Axboe * Note that for blockdevs, inode->dirtied_when represents the dirtying time of 128503ba3782SJens Axboe * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of 128603ba3782SJens Axboe * the kernel-internal blockdev inode represents the dirtying time of the 128703ba3782SJens Axboe * blockdev's pages. This is why for I_DIRTY_PAGES we always use 128803ba3782SJens Axboe * page->mapping->host, so the page-dirtying time is recorded in the internal 128903ba3782SJens Axboe * blockdev inode. 129003ba3782SJens Axboe */ 12910ae45f63STheodore Ts'o #define I_DIRTY_INODE (I_DIRTY_SYNC | I_DIRTY_DATASYNC) 129203ba3782SJens Axboe void __mark_inode_dirty(struct inode *inode, int flags) 129303ba3782SJens Axboe { 129403ba3782SJens Axboe struct super_block *sb = inode->i_sb; 1295253c34e9SArtem Bityutskiy struct backing_dev_info *bdi = NULL; 12960ae45f63STheodore Ts'o int dirtytime; 12970ae45f63STheodore Ts'o 12980ae45f63STheodore Ts'o trace_writeback_mark_inode_dirty(inode, flags); 129903ba3782SJens Axboe 130003ba3782SJens Axboe /* 130103ba3782SJens Axboe * Don't do this for I_DIRTY_PAGES - that doesn't actually 130203ba3782SJens Axboe * dirty the inode itself 130303ba3782SJens Axboe */ 13040ae45f63STheodore Ts'o if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC | I_DIRTY_TIME)) { 13059fb0a7daSTejun Heo trace_writeback_dirty_inode_start(inode, flags); 13069fb0a7daSTejun Heo 130703ba3782SJens Axboe if (sb->s_op->dirty_inode) 1308aa385729SChristoph Hellwig sb->s_op->dirty_inode(inode, flags); 13099fb0a7daSTejun Heo 13109fb0a7daSTejun Heo trace_writeback_dirty_inode(inode, flags); 131103ba3782SJens Axboe } 13120ae45f63STheodore Ts'o if (flags & I_DIRTY_INODE) 13130ae45f63STheodore Ts'o flags &= ~I_DIRTY_TIME; 13140ae45f63STheodore Ts'o dirtytime = flags & I_DIRTY_TIME; 131503ba3782SJens Axboe 131603ba3782SJens Axboe /* 13179c6ac78eSTejun Heo * Paired with smp_mb() in __writeback_single_inode() for the 13189c6ac78eSTejun Heo * following lockless i_state test. See there for details. 131903ba3782SJens Axboe */ 132003ba3782SJens Axboe smp_mb(); 132103ba3782SJens Axboe 13220ae45f63STheodore Ts'o if (((inode->i_state & flags) == flags) || 13230ae45f63STheodore Ts'o (dirtytime && (inode->i_state & I_DIRTY_INODE))) 132403ba3782SJens Axboe return; 132503ba3782SJens Axboe 132603ba3782SJens Axboe if (unlikely(block_dump)) 132703ba3782SJens Axboe block_dump___mark_inode_dirty(inode); 132803ba3782SJens Axboe 1329250df6edSDave Chinner spin_lock(&inode->i_lock); 13300ae45f63STheodore Ts'o if (dirtytime && (inode->i_state & I_DIRTY_INODE)) 13310ae45f63STheodore Ts'o goto out_unlock_inode; 133203ba3782SJens Axboe if ((inode->i_state & flags) != flags) { 133303ba3782SJens Axboe const int was_dirty = inode->i_state & I_DIRTY; 133403ba3782SJens Axboe 133552ebea74STejun Heo inode_attach_wb(inode, NULL); 133652ebea74STejun Heo 13370ae45f63STheodore Ts'o if (flags & I_DIRTY_INODE) 13380ae45f63STheodore Ts'o inode->i_state &= ~I_DIRTY_TIME; 133903ba3782SJens Axboe inode->i_state |= flags; 134003ba3782SJens Axboe 134103ba3782SJens Axboe /* 134203ba3782SJens Axboe * If the inode is being synced, just update its dirty state. 134303ba3782SJens Axboe * The unlocker will place the inode on the appropriate 134403ba3782SJens Axboe * superblock list, based upon its state. 134503ba3782SJens Axboe */ 134603ba3782SJens Axboe if (inode->i_state & I_SYNC) 1347250df6edSDave Chinner goto out_unlock_inode; 134803ba3782SJens Axboe 134903ba3782SJens Axboe /* 135003ba3782SJens Axboe * Only add valid (hashed) inodes to the superblock's 135103ba3782SJens Axboe * dirty list. Add blockdev inodes as well. 135203ba3782SJens Axboe */ 135303ba3782SJens Axboe if (!S_ISBLK(inode->i_mode)) { 13541d3382cbSAl Viro if (inode_unhashed(inode)) 1355250df6edSDave Chinner goto out_unlock_inode; 135603ba3782SJens Axboe } 1357a4ffdde6SAl Viro if (inode->i_state & I_FREEING) 1358250df6edSDave Chinner goto out_unlock_inode; 135903ba3782SJens Axboe 136003ba3782SJens Axboe /* 136103ba3782SJens Axboe * If the inode was already on b_dirty/b_io/b_more_io, don't 136203ba3782SJens Axboe * reposition it (that would break b_dirty time-ordering). 136303ba3782SJens Axboe */ 136403ba3782SJens Axboe if (!was_dirty) { 1365d6c10f1fSTejun Heo struct list_head *dirty_list; 1366a66979abSDave Chinner bool wakeup_bdi = false; 1367253c34e9SArtem Bityutskiy bdi = inode_to_bdi(inode); 1368500b067cSJens Axboe 1369146d7009SJunxiao Bi spin_unlock(&inode->i_lock); 1370146d7009SJunxiao Bi spin_lock(&bdi->wb.list_lock); 1371253c34e9SArtem Bityutskiy 1372d6c10f1fSTejun Heo WARN(bdi_cap_writeback_dirty(bdi) && 1373d6c10f1fSTejun Heo !test_bit(WB_registered, &bdi->wb.state), 1374d6c10f1fSTejun Heo "bdi-%s not registered\n", bdi->name); 137503ba3782SJens Axboe 137603ba3782SJens Axboe inode->dirtied_when = jiffies; 1377a2f48706STheodore Ts'o if (dirtytime) 1378a2f48706STheodore Ts'o inode->dirtied_time_when = jiffies; 1379d6c10f1fSTejun Heo 1380a2f48706STheodore Ts'o if (inode->i_state & (I_DIRTY_INODE | I_DIRTY_PAGES)) 1381d6c10f1fSTejun Heo dirty_list = &bdi->wb.b_dirty; 1382a2f48706STheodore Ts'o else 1383d6c10f1fSTejun Heo dirty_list = &bdi->wb.b_dirty_time; 1384d6c10f1fSTejun Heo 1385d6c10f1fSTejun Heo wakeup_bdi = inode_wb_list_move_locked(inode, &bdi->wb, 1386d6c10f1fSTejun Heo dirty_list); 1387d6c10f1fSTejun Heo 1388f758eeabSChristoph Hellwig spin_unlock(&bdi->wb.list_lock); 13890ae45f63STheodore Ts'o trace_writeback_dirty_inode_enqueue(inode); 1390253c34e9SArtem Bityutskiy 1391d6c10f1fSTejun Heo /* 1392d6c10f1fSTejun Heo * If this is the first dirty inode for this bdi, 1393d6c10f1fSTejun Heo * we have to wake-up the corresponding bdi thread 1394d6c10f1fSTejun Heo * to make sure background write-back happens 1395d6c10f1fSTejun Heo * later. 1396d6c10f1fSTejun Heo */ 1397d6c10f1fSTejun Heo if (bdi_cap_writeback_dirty(bdi) && wakeup_bdi) 1398f0054bb1STejun Heo wb_wakeup_delayed(&bdi->wb); 1399a66979abSDave Chinner return; 1400a66979abSDave Chinner } 1401a66979abSDave Chinner } 1402a66979abSDave Chinner out_unlock_inode: 1403a66979abSDave Chinner spin_unlock(&inode->i_lock); 1404a66979abSDave Chinner 140503ba3782SJens Axboe } 140603ba3782SJens Axboe EXPORT_SYMBOL(__mark_inode_dirty); 140703ba3782SJens Axboe 1408b6e51316SJens Axboe static void wait_sb_inodes(struct super_block *sb) 140966f3b8e2SJens Axboe { 141038f21977SNick Piggin struct inode *inode, *old_inode = NULL; 141138f21977SNick Piggin 141203ba3782SJens Axboe /* 141303ba3782SJens Axboe * We need to be protected against the filesystem going from 141403ba3782SJens Axboe * r/o to r/w or vice versa. 141503ba3782SJens Axboe */ 1416b6e51316SJens Axboe WARN_ON(!rwsem_is_locked(&sb->s_umount)); 141703ba3782SJens Axboe 141855fa6091SDave Chinner spin_lock(&inode_sb_list_lock); 141966f3b8e2SJens Axboe 142038f21977SNick Piggin /* 142138f21977SNick Piggin * Data integrity sync. Must wait for all pages under writeback, 142238f21977SNick Piggin * because there may have been pages dirtied before our sync 142338f21977SNick Piggin * call, but which had writeout started before we write it out. 142438f21977SNick Piggin * In which case, the inode may not be on the dirty list, but 142538f21977SNick Piggin * we still have to wait for that writeout. 142638f21977SNick Piggin */ 1427b6e51316SJens Axboe list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { 1428250df6edSDave Chinner struct address_space *mapping = inode->i_mapping; 142938f21977SNick Piggin 1430250df6edSDave Chinner spin_lock(&inode->i_lock); 1431250df6edSDave Chinner if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) || 1432250df6edSDave Chinner (mapping->nrpages == 0)) { 1433250df6edSDave Chinner spin_unlock(&inode->i_lock); 143438f21977SNick Piggin continue; 1435250df6edSDave Chinner } 143638f21977SNick Piggin __iget(inode); 1437250df6edSDave Chinner spin_unlock(&inode->i_lock); 143855fa6091SDave Chinner spin_unlock(&inode_sb_list_lock); 143955fa6091SDave Chinner 144038f21977SNick Piggin /* 144155fa6091SDave Chinner * We hold a reference to 'inode' so it couldn't have been 144255fa6091SDave Chinner * removed from s_inodes list while we dropped the 144355fa6091SDave Chinner * inode_sb_list_lock. We cannot iput the inode now as we can 144455fa6091SDave Chinner * be holding the last reference and we cannot iput it under 144555fa6091SDave Chinner * inode_sb_list_lock. So we keep the reference and iput it 144655fa6091SDave Chinner * later. 144738f21977SNick Piggin */ 144838f21977SNick Piggin iput(old_inode); 144938f21977SNick Piggin old_inode = inode; 145038f21977SNick Piggin 145138f21977SNick Piggin filemap_fdatawait(mapping); 145238f21977SNick Piggin 145338f21977SNick Piggin cond_resched(); 145438f21977SNick Piggin 145555fa6091SDave Chinner spin_lock(&inode_sb_list_lock); 145638f21977SNick Piggin } 145755fa6091SDave Chinner spin_unlock(&inode_sb_list_lock); 145838f21977SNick Piggin iput(old_inode); 145966f3b8e2SJens Axboe } 14601da177e4SLinus Torvalds 1461d8a8559cSJens Axboe /** 14623259f8beSChris Mason * writeback_inodes_sb_nr - writeback dirty inodes from given super_block 1463d8a8559cSJens Axboe * @sb: the superblock 14643259f8beSChris Mason * @nr: the number of pages to write 1465786228abSMarcos Paulo de Souza * @reason: reason why some writeback work initiated 14661da177e4SLinus Torvalds * 1467d8a8559cSJens Axboe * Start writeback on some inodes on this super_block. No guarantees are made 1468d8a8559cSJens Axboe * on how many (if any) will be written, and this function does not wait 14693259f8beSChris Mason * for IO completion of submitted IO. 14701da177e4SLinus Torvalds */ 14710e175a18SCurt Wohlgemuth void writeback_inodes_sb_nr(struct super_block *sb, 14720e175a18SCurt Wohlgemuth unsigned long nr, 14730e175a18SCurt Wohlgemuth enum wb_reason reason) 14741da177e4SLinus Torvalds { 147583ba7b07SChristoph Hellwig DECLARE_COMPLETION_ONSTACK(done); 147683ba7b07SChristoph Hellwig struct wb_writeback_work work = { 14773c4d7165SChristoph Hellwig .sb = sb, 14783c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_NONE, 14796e6938b6SWu Fengguang .tagged_writepages = 1, 148083ba7b07SChristoph Hellwig .done = &done, 14813259f8beSChris Mason .nr_pages = nr, 14820e175a18SCurt Wohlgemuth .reason = reason, 14833c4d7165SChristoph Hellwig }; 1484e7972912STejun Heo struct backing_dev_info *bdi = sb->s_bdi; 14850e3c9a22SJens Axboe 1486e7972912STejun Heo if (!bdi_has_dirty_io(bdi) || bdi == &noop_backing_dev_info) 14876eedc701SJan Kara return; 1488cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 1489e7972912STejun Heo wb_queue_work(&bdi->wb, &work); 149083ba7b07SChristoph Hellwig wait_for_completion(&done); 14911da177e4SLinus Torvalds } 14923259f8beSChris Mason EXPORT_SYMBOL(writeback_inodes_sb_nr); 14933259f8beSChris Mason 14943259f8beSChris Mason /** 14953259f8beSChris Mason * writeback_inodes_sb - writeback dirty inodes from given super_block 14963259f8beSChris Mason * @sb: the superblock 1497786228abSMarcos Paulo de Souza * @reason: reason why some writeback work was initiated 14983259f8beSChris Mason * 14993259f8beSChris Mason * Start writeback on some inodes on this super_block. No guarantees are made 15003259f8beSChris Mason * on how many (if any) will be written, and this function does not wait 15013259f8beSChris Mason * for IO completion of submitted IO. 15023259f8beSChris Mason */ 15030e175a18SCurt Wohlgemuth void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) 15043259f8beSChris Mason { 15050e175a18SCurt Wohlgemuth return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); 15063259f8beSChris Mason } 1507d8a8559cSJens Axboe EXPORT_SYMBOL(writeback_inodes_sb); 1508d8a8559cSJens Axboe 1509d8a8559cSJens Axboe /** 151010ee27a0SMiao Xie * try_to_writeback_inodes_sb_nr - try to start writeback if none underway 15113259f8beSChris Mason * @sb: the superblock 15123259f8beSChris Mason * @nr: the number of pages to write 151310ee27a0SMiao Xie * @reason: the reason of writeback 15143259f8beSChris Mason * 151510ee27a0SMiao Xie * Invoke writeback_inodes_sb_nr if no writeback is currently underway. 15163259f8beSChris Mason * Returns 1 if writeback was started, 0 if not. 15173259f8beSChris Mason */ 151810ee27a0SMiao Xie int try_to_writeback_inodes_sb_nr(struct super_block *sb, 15190e175a18SCurt Wohlgemuth unsigned long nr, 15200e175a18SCurt Wohlgemuth enum wb_reason reason) 15213259f8beSChris Mason { 1522bc05873dSTejun Heo if (writeback_in_progress(&sb->s_bdi->wb)) 152310ee27a0SMiao Xie return 1; 152410ee27a0SMiao Xie 152510ee27a0SMiao Xie if (!down_read_trylock(&sb->s_umount)) 152610ee27a0SMiao Xie return 0; 152710ee27a0SMiao Xie 15280e175a18SCurt Wohlgemuth writeback_inodes_sb_nr(sb, nr, reason); 15293259f8beSChris Mason up_read(&sb->s_umount); 15303259f8beSChris Mason return 1; 15313259f8beSChris Mason } 153210ee27a0SMiao Xie EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr); 153310ee27a0SMiao Xie 153410ee27a0SMiao Xie /** 153510ee27a0SMiao Xie * try_to_writeback_inodes_sb - try to start writeback if none underway 153610ee27a0SMiao Xie * @sb: the superblock 153710ee27a0SMiao Xie * @reason: reason why some writeback work was initiated 153810ee27a0SMiao Xie * 153910ee27a0SMiao Xie * Implement by try_to_writeback_inodes_sb_nr() 154010ee27a0SMiao Xie * Returns 1 if writeback was started, 0 if not. 154110ee27a0SMiao Xie */ 154210ee27a0SMiao Xie int try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) 154310ee27a0SMiao Xie { 154410ee27a0SMiao Xie return try_to_writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); 154510ee27a0SMiao Xie } 154610ee27a0SMiao Xie EXPORT_SYMBOL(try_to_writeback_inodes_sb); 15473259f8beSChris Mason 15483259f8beSChris Mason /** 1549d8a8559cSJens Axboe * sync_inodes_sb - sync sb inode pages 1550d8a8559cSJens Axboe * @sb: the superblock 1551d8a8559cSJens Axboe * 1552d8a8559cSJens Axboe * This function writes and waits on any dirty inode belonging to this 15530dc83bd3SJan Kara * super_block. 1554d8a8559cSJens Axboe */ 15550dc83bd3SJan Kara void sync_inodes_sb(struct super_block *sb) 1556d8a8559cSJens Axboe { 155783ba7b07SChristoph Hellwig DECLARE_COMPLETION_ONSTACK(done); 155883ba7b07SChristoph Hellwig struct wb_writeback_work work = { 15593c4d7165SChristoph Hellwig .sb = sb, 15603c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_ALL, 15613c4d7165SChristoph Hellwig .nr_pages = LONG_MAX, 15623c4d7165SChristoph Hellwig .range_cyclic = 0, 156383ba7b07SChristoph Hellwig .done = &done, 15640e175a18SCurt Wohlgemuth .reason = WB_REASON_SYNC, 15657747bd4bSDave Chinner .for_sync = 1, 15663c4d7165SChristoph Hellwig }; 1567e7972912STejun Heo struct backing_dev_info *bdi = sb->s_bdi; 15683c4d7165SChristoph Hellwig 15696eedc701SJan Kara /* Nothing to do? */ 1570e7972912STejun Heo if (!bdi_has_dirty_io(bdi) || bdi == &noop_backing_dev_info) 15716eedc701SJan Kara return; 1572cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 1573cf37e972SChristoph Hellwig 1574e7972912STejun Heo wb_queue_work(&bdi->wb, &work); 157583ba7b07SChristoph Hellwig wait_for_completion(&done); 157683ba7b07SChristoph Hellwig 1577b6e51316SJens Axboe wait_sb_inodes(sb); 1578d8a8559cSJens Axboe } 1579d8a8559cSJens Axboe EXPORT_SYMBOL(sync_inodes_sb); 15801da177e4SLinus Torvalds 15811da177e4SLinus Torvalds /** 15821da177e4SLinus Torvalds * write_inode_now - write an inode to disk 15831da177e4SLinus Torvalds * @inode: inode to write to disk 15841da177e4SLinus Torvalds * @sync: whether the write should be synchronous or not 15851da177e4SLinus Torvalds * 15867f04c26dSAndrea Arcangeli * This function commits an inode to disk immediately if it is dirty. This is 15877f04c26dSAndrea Arcangeli * primarily needed by knfsd. 15887f04c26dSAndrea Arcangeli * 15897f04c26dSAndrea Arcangeli * The caller must either have a ref on the inode or must have set I_WILL_FREE. 15901da177e4SLinus Torvalds */ 15911da177e4SLinus Torvalds int write_inode_now(struct inode *inode, int sync) 15921da177e4SLinus Torvalds { 1593f758eeabSChristoph Hellwig struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; 15941da177e4SLinus Torvalds struct writeback_control wbc = { 15951da177e4SLinus Torvalds .nr_to_write = LONG_MAX, 159618914b18SMike Galbraith .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, 1597111ebb6eSOGAWA Hirofumi .range_start = 0, 1598111ebb6eSOGAWA Hirofumi .range_end = LLONG_MAX, 15991da177e4SLinus Torvalds }; 16001da177e4SLinus Torvalds 16011da177e4SLinus Torvalds if (!mapping_cap_writeback_dirty(inode->i_mapping)) 160249364ce2SAndrew Morton wbc.nr_to_write = 0; 16031da177e4SLinus Torvalds 16041da177e4SLinus Torvalds might_sleep(); 16054f8ad655SJan Kara return writeback_single_inode(inode, wb, &wbc); 16061da177e4SLinus Torvalds } 16071da177e4SLinus Torvalds EXPORT_SYMBOL(write_inode_now); 16081da177e4SLinus Torvalds 16091da177e4SLinus Torvalds /** 16101da177e4SLinus Torvalds * sync_inode - write an inode and its pages to disk. 16111da177e4SLinus Torvalds * @inode: the inode to sync 16121da177e4SLinus Torvalds * @wbc: controls the writeback mode 16131da177e4SLinus Torvalds * 16141da177e4SLinus Torvalds * sync_inode() will write an inode and its pages to disk. It will also 16151da177e4SLinus Torvalds * correctly update the inode on its superblock's dirty inode lists and will 16161da177e4SLinus Torvalds * update inode->i_state. 16171da177e4SLinus Torvalds * 16181da177e4SLinus Torvalds * The caller must have a ref on the inode. 16191da177e4SLinus Torvalds */ 16201da177e4SLinus Torvalds int sync_inode(struct inode *inode, struct writeback_control *wbc) 16211da177e4SLinus Torvalds { 16224f8ad655SJan Kara return writeback_single_inode(inode, &inode_to_bdi(inode)->wb, wbc); 16231da177e4SLinus Torvalds } 16241da177e4SLinus Torvalds EXPORT_SYMBOL(sync_inode); 1625c3765016SChristoph Hellwig 1626c3765016SChristoph Hellwig /** 1627c691b9d9SAndrew Morton * sync_inode_metadata - write an inode to disk 1628c3765016SChristoph Hellwig * @inode: the inode to sync 1629c3765016SChristoph Hellwig * @wait: wait for I/O to complete. 1630c3765016SChristoph Hellwig * 1631c691b9d9SAndrew Morton * Write an inode to disk and adjust its dirty state after completion. 1632c3765016SChristoph Hellwig * 1633c3765016SChristoph Hellwig * Note: only writes the actual inode, no associated data or other metadata. 1634c3765016SChristoph Hellwig */ 1635c3765016SChristoph Hellwig int sync_inode_metadata(struct inode *inode, int wait) 1636c3765016SChristoph Hellwig { 1637c3765016SChristoph Hellwig struct writeback_control wbc = { 1638c3765016SChristoph Hellwig .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE, 1639c3765016SChristoph Hellwig .nr_to_write = 0, /* metadata-only */ 1640c3765016SChristoph Hellwig }; 1641c3765016SChristoph Hellwig 1642c3765016SChristoph Hellwig return sync_inode(inode, &wbc); 1643c3765016SChristoph Hellwig } 1644c3765016SChristoph Hellwig EXPORT_SYMBOL(sync_inode_metadata); 1645