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; 43c4a391b5SJan Kara /* 44c4a391b5SJan Kara * Write only inodes dirtied before this time. Don't forget to set 45c4a391b5SJan Kara * older_than_this_is_set when you set this. 46c4a391b5SJan Kara */ 47c4a391b5SJan Kara unsigned long older_than_this; 48c4a77a6cSJens Axboe enum writeback_sync_modes sync_mode; 496e6938b6SWu Fengguang unsigned int tagged_writepages:1; 5052957fe1SH Hartley Sweeten unsigned int for_kupdate:1; 5152957fe1SH Hartley Sweeten unsigned int range_cyclic:1; 5252957fe1SH Hartley Sweeten unsigned int for_background:1; 537747bd4bSDave Chinner unsigned int for_sync:1; /* sync(2) WB_SYNC_ALL writeback */ 54c4a391b5SJan Kara unsigned int older_than_this_is_set:1; 550e175a18SCurt Wohlgemuth enum wb_reason reason; /* why was writeback initiated? */ 56c4a77a6cSJens Axboe 578010c3b6SJens Axboe struct list_head list; /* pending work list */ 5883ba7b07SChristoph Hellwig struct completion *done; /* set if the caller waits */ 5903ba3782SJens Axboe }; 6003ba3782SJens Axboe 61f11b00f3SAdrian Bunk /** 62f11b00f3SAdrian Bunk * writeback_in_progress - determine whether there is writeback in progress 63f11b00f3SAdrian Bunk * @bdi: the device's backing_dev_info structure. 64f11b00f3SAdrian Bunk * 6503ba3782SJens Axboe * Determine whether there is writeback waiting to be handled against a 6603ba3782SJens Axboe * backing device. 67f11b00f3SAdrian Bunk */ 68f11b00f3SAdrian Bunk int writeback_in_progress(struct backing_dev_info *bdi) 69f11b00f3SAdrian Bunk { 7081d73a32SJan Kara return test_bit(BDI_writeback_running, &bdi->state); 71f11b00f3SAdrian Bunk } 7200d4e736STheodore Ts'o EXPORT_SYMBOL(writeback_in_progress); 73f11b00f3SAdrian Bunk 74692ebd17SJan Kara static inline struct backing_dev_info *inode_to_bdi(struct inode *inode) 75692ebd17SJan Kara { 76692ebd17SJan Kara struct super_block *sb = inode->i_sb; 77692ebd17SJan Kara 78a8855990SJan Kara if (sb_is_blkdev_sb(sb)) 79aaead25bSChristoph Hellwig return inode->i_mapping->backing_dev_info; 80aaead25bSChristoph Hellwig 81692ebd17SJan Kara return sb->s_bdi; 82692ebd17SJan Kara } 83692ebd17SJan Kara 847ccf19a8SNick Piggin static inline struct inode *wb_inode(struct list_head *head) 857ccf19a8SNick Piggin { 867ccf19a8SNick Piggin return list_entry(head, struct inode, i_wb_list); 877ccf19a8SNick Piggin } 887ccf19a8SNick Piggin 8915eb77a0SWu Fengguang /* 9015eb77a0SWu Fengguang * Include the creation of the trace points after defining the 9115eb77a0SWu Fengguang * wb_writeback_work structure and inline functions so that the definition 9215eb77a0SWu Fengguang * remains local to this file. 9315eb77a0SWu Fengguang */ 9415eb77a0SWu Fengguang #define CREATE_TRACE_POINTS 9515eb77a0SWu Fengguang #include <trace/events/writeback.h> 9615eb77a0SWu Fengguang 97774016b2SSteven Whitehouse EXPORT_TRACEPOINT_SYMBOL_GPL(wbc_writepage); 98774016b2SSteven Whitehouse 996585027aSJan Kara static void bdi_queue_work(struct backing_dev_info *bdi, 1006585027aSJan Kara struct wb_writeback_work *work) 1016585027aSJan Kara { 1026585027aSJan Kara trace_writeback_queue(bdi, work); 1036585027aSJan Kara 1046585027aSJan Kara spin_lock_bh(&bdi->wb_lock); 1056585027aSJan Kara list_add_tail(&work->list, &bdi->work_list); 1066467716aSArtem Bityutskiy spin_unlock_bh(&bdi->wb_lock); 107839a8e86STejun Heo 108839a8e86STejun Heo mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0); 10903ba3782SJens Axboe } 1101da177e4SLinus Torvalds 11183ba7b07SChristoph Hellwig static void 11283ba7b07SChristoph Hellwig __bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, 1130e175a18SCurt Wohlgemuth bool range_cyclic, enum wb_reason reason) 1141da177e4SLinus Torvalds { 11583ba7b07SChristoph Hellwig struct wb_writeback_work *work; 11603ba3782SJens Axboe 117bcddc3f0SJens Axboe /* 118bcddc3f0SJens Axboe * This is WB_SYNC_NONE writeback, so if allocation fails just 119bcddc3f0SJens Axboe * wakeup the thread for old dirty data writeback 120bcddc3f0SJens Axboe */ 12183ba7b07SChristoph Hellwig work = kzalloc(sizeof(*work), GFP_ATOMIC); 12283ba7b07SChristoph Hellwig if (!work) { 123455b2864SDave Chinner trace_writeback_nowork(bdi); 124839a8e86STejun Heo mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0); 12583ba7b07SChristoph Hellwig return; 12683ba7b07SChristoph Hellwig } 12783ba7b07SChristoph Hellwig 12883ba7b07SChristoph Hellwig work->sync_mode = WB_SYNC_NONE; 12983ba7b07SChristoph Hellwig work->nr_pages = nr_pages; 13083ba7b07SChristoph Hellwig work->range_cyclic = range_cyclic; 1310e175a18SCurt Wohlgemuth work->reason = reason; 13283ba7b07SChristoph Hellwig 133f11fcae8SJens Axboe bdi_queue_work(bdi, work); 13403ba3782SJens Axboe } 135b6e51316SJens Axboe 136b6e51316SJens Axboe /** 137b6e51316SJens Axboe * bdi_start_writeback - start writeback 138b6e51316SJens Axboe * @bdi: the backing device to write from 139b6e51316SJens Axboe * @nr_pages: the number of pages to write 140786228abSMarcos Paulo de Souza * @reason: reason why some writeback work was initiated 141b6e51316SJens Axboe * 142b6e51316SJens Axboe * Description: 143b6e51316SJens Axboe * This does WB_SYNC_NONE opportunistic writeback. The IO is only 14425985edcSLucas De Marchi * started when this function returns, we make no guarantees on 1450e3c9a22SJens Axboe * completion. Caller need not hold sb s_umount semaphore. 146b6e51316SJens Axboe * 147b6e51316SJens Axboe */ 1480e175a18SCurt Wohlgemuth void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, 1490e175a18SCurt Wohlgemuth enum wb_reason reason) 150b6e51316SJens Axboe { 1510e175a18SCurt Wohlgemuth __bdi_start_writeback(bdi, nr_pages, true, reason); 152d3ddec76SWu Fengguang } 153d3ddec76SWu Fengguang 154c5444198SChristoph Hellwig /** 155c5444198SChristoph Hellwig * bdi_start_background_writeback - start background writeback 156c5444198SChristoph Hellwig * @bdi: the backing device to write from 157c5444198SChristoph Hellwig * 158c5444198SChristoph Hellwig * Description: 1596585027aSJan Kara * This makes sure WB_SYNC_NONE background writeback happens. When 1606585027aSJan Kara * this function returns, it is only guaranteed that for given BDI 1616585027aSJan Kara * some IO is happening if we are over background dirty threshold. 1626585027aSJan Kara * Caller need not hold sb s_umount semaphore. 163c5444198SChristoph Hellwig */ 164c5444198SChristoph Hellwig void bdi_start_background_writeback(struct backing_dev_info *bdi) 165c5444198SChristoph Hellwig { 1666585027aSJan Kara /* 1676585027aSJan Kara * We just wake up the flusher thread. It will perform background 1686585027aSJan Kara * writeback as soon as there is no other work to do. 1696585027aSJan Kara */ 17071927e84SWu Fengguang trace_writeback_wake_background(bdi); 171839a8e86STejun Heo mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0); 1721da177e4SLinus Torvalds } 1731da177e4SLinus Torvalds 1741da177e4SLinus Torvalds /* 175a66979abSDave Chinner * Remove the inode from the writeback list it is on. 176a66979abSDave Chinner */ 177a66979abSDave Chinner void inode_wb_list_del(struct inode *inode) 178a66979abSDave Chinner { 179f758eeabSChristoph Hellwig struct backing_dev_info *bdi = inode_to_bdi(inode); 180a66979abSDave Chinner 181f758eeabSChristoph Hellwig spin_lock(&bdi->wb.list_lock); 182f758eeabSChristoph Hellwig list_del_init(&inode->i_wb_list); 183f758eeabSChristoph Hellwig spin_unlock(&bdi->wb.list_lock); 184f758eeabSChristoph Hellwig } 185a66979abSDave Chinner 186a66979abSDave Chinner /* 1876610a0bcSAndrew Morton * Redirty an inode: set its when-it-was dirtied timestamp and move it to the 1886610a0bcSAndrew Morton * furthest end of its superblock's dirty-inode list. 1896610a0bcSAndrew Morton * 1906610a0bcSAndrew Morton * Before stamping the inode's ->dirtied_when, we check to see whether it is 19166f3b8e2SJens Axboe * already the most-recently-dirtied inode on the b_dirty list. If that is 1926610a0bcSAndrew Morton * the case then the inode must have been redirtied while it was being written 1936610a0bcSAndrew Morton * out and we don't reset its dirtied_when. 1946610a0bcSAndrew Morton */ 195f758eeabSChristoph Hellwig static void redirty_tail(struct inode *inode, struct bdi_writeback *wb) 1966610a0bcSAndrew Morton { 197f758eeabSChristoph Hellwig assert_spin_locked(&wb->list_lock); 19803ba3782SJens Axboe if (!list_empty(&wb->b_dirty)) { 19966f3b8e2SJens Axboe struct inode *tail; 2006610a0bcSAndrew Morton 2017ccf19a8SNick Piggin tail = wb_inode(wb->b_dirty.next); 20266f3b8e2SJens Axboe if (time_before(inode->dirtied_when, tail->dirtied_when)) 2036610a0bcSAndrew Morton inode->dirtied_when = jiffies; 2046610a0bcSAndrew Morton } 2057ccf19a8SNick Piggin list_move(&inode->i_wb_list, &wb->b_dirty); 2066610a0bcSAndrew Morton } 2076610a0bcSAndrew Morton 2086610a0bcSAndrew Morton /* 20966f3b8e2SJens Axboe * requeue inode for re-scanning after bdi->b_io list is exhausted. 210c986d1e2SAndrew Morton */ 211f758eeabSChristoph Hellwig static void requeue_io(struct inode *inode, struct bdi_writeback *wb) 212c986d1e2SAndrew Morton { 213f758eeabSChristoph Hellwig assert_spin_locked(&wb->list_lock); 2147ccf19a8SNick Piggin list_move(&inode->i_wb_list, &wb->b_more_io); 215c986d1e2SAndrew Morton } 216c986d1e2SAndrew Morton 2171c0eeaf5SJoern Engel static void inode_sync_complete(struct inode *inode) 2181c0eeaf5SJoern Engel { 219365b94aeSJan Kara inode->i_state &= ~I_SYNC; 2204eff96ddSJan Kara /* If inode is clean an unused, put it into LRU now... */ 2214eff96ddSJan Kara inode_add_lru(inode); 222365b94aeSJan Kara /* Waiters must see I_SYNC cleared before being woken up */ 2231c0eeaf5SJoern Engel smp_mb(); 2241c0eeaf5SJoern Engel wake_up_bit(&inode->i_state, __I_SYNC); 2251c0eeaf5SJoern Engel } 2261c0eeaf5SJoern Engel 227d2caa3c5SJeff Layton static bool inode_dirtied_after(struct inode *inode, unsigned long t) 228d2caa3c5SJeff Layton { 229d2caa3c5SJeff Layton bool ret = time_after(inode->dirtied_when, t); 230d2caa3c5SJeff Layton #ifndef CONFIG_64BIT 231d2caa3c5SJeff Layton /* 232d2caa3c5SJeff Layton * For inodes being constantly redirtied, dirtied_when can get stuck. 233d2caa3c5SJeff Layton * It _appears_ to be in the future, but is actually in distant past. 234d2caa3c5SJeff Layton * This test is necessary to prevent such wrapped-around relative times 2355b0830cbSJens Axboe * from permanently stopping the whole bdi writeback. 236d2caa3c5SJeff Layton */ 237d2caa3c5SJeff Layton ret = ret && time_before_eq(inode->dirtied_when, jiffies); 238d2caa3c5SJeff Layton #endif 239d2caa3c5SJeff Layton return ret; 240d2caa3c5SJeff Layton } 241d2caa3c5SJeff Layton 242c986d1e2SAndrew Morton /* 2430e2f2b23SWang Sheng-Hui * Move expired (dirtied before work->older_than_this) dirty inodes from 244697e6fedSJan Kara * @delaying_queue to @dispatch_queue. 2452c136579SFengguang Wu */ 246e84d0a4fSWu Fengguang static int move_expired_inodes(struct list_head *delaying_queue, 2472c136579SFengguang Wu struct list_head *dispatch_queue, 248ad4e38ddSCurt Wohlgemuth struct wb_writeback_work *work) 2492c136579SFengguang Wu { 2505c03449dSShaohua Li LIST_HEAD(tmp); 2515c03449dSShaohua Li struct list_head *pos, *node; 252cf137307SJens Axboe struct super_block *sb = NULL; 2535c03449dSShaohua Li struct inode *inode; 254cf137307SJens Axboe int do_sb_sort = 0; 255e84d0a4fSWu Fengguang int moved = 0; 2565c03449dSShaohua Li 257c4a391b5SJan Kara WARN_ON_ONCE(!work->older_than_this_is_set); 2582c136579SFengguang Wu while (!list_empty(delaying_queue)) { 2597ccf19a8SNick Piggin inode = wb_inode(delaying_queue->prev); 260c4a391b5SJan Kara if (inode_dirtied_after(inode, work->older_than_this)) 2612c136579SFengguang Wu break; 262a8855990SJan Kara list_move(&inode->i_wb_list, &tmp); 263a8855990SJan Kara moved++; 264a8855990SJan Kara if (sb_is_blkdev_sb(inode->i_sb)) 265a8855990SJan Kara continue; 266cf137307SJens Axboe if (sb && sb != inode->i_sb) 267cf137307SJens Axboe do_sb_sort = 1; 268cf137307SJens Axboe sb = inode->i_sb; 2695c03449dSShaohua Li } 2705c03449dSShaohua Li 271cf137307SJens Axboe /* just one sb in list, splice to dispatch_queue and we're done */ 272cf137307SJens Axboe if (!do_sb_sort) { 273cf137307SJens Axboe list_splice(&tmp, dispatch_queue); 274e84d0a4fSWu Fengguang goto out; 275cf137307SJens Axboe } 276cf137307SJens Axboe 2775c03449dSShaohua Li /* Move inodes from one superblock together */ 2785c03449dSShaohua Li while (!list_empty(&tmp)) { 2797ccf19a8SNick Piggin sb = wb_inode(tmp.prev)->i_sb; 2805c03449dSShaohua Li list_for_each_prev_safe(pos, node, &tmp) { 2817ccf19a8SNick Piggin inode = wb_inode(pos); 2825c03449dSShaohua Li if (inode->i_sb == sb) 2837ccf19a8SNick Piggin list_move(&inode->i_wb_list, dispatch_queue); 2842c136579SFengguang Wu } 2852c136579SFengguang Wu } 286e84d0a4fSWu Fengguang out: 287e84d0a4fSWu Fengguang return moved; 2885c03449dSShaohua Li } 2892c136579SFengguang Wu 2902c136579SFengguang Wu /* 2912c136579SFengguang Wu * Queue all expired dirty inodes for io, eldest first. 2924ea879b9SWu Fengguang * Before 2934ea879b9SWu Fengguang * newly dirtied b_dirty b_io b_more_io 2944ea879b9SWu Fengguang * =============> gf edc BA 2954ea879b9SWu Fengguang * After 2964ea879b9SWu Fengguang * newly dirtied b_dirty b_io b_more_io 2974ea879b9SWu Fengguang * =============> g fBAedc 2984ea879b9SWu Fengguang * | 2994ea879b9SWu Fengguang * +--> dequeue for IO 3002c136579SFengguang Wu */ 301ad4e38ddSCurt Wohlgemuth static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work) 3022c136579SFengguang Wu { 303e84d0a4fSWu Fengguang int moved; 304f758eeabSChristoph Hellwig assert_spin_locked(&wb->list_lock); 3054ea879b9SWu Fengguang list_splice_init(&wb->b_more_io, &wb->b_io); 306ad4e38ddSCurt Wohlgemuth moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, work); 307ad4e38ddSCurt Wohlgemuth trace_writeback_queue_io(wb, work, moved); 30866f3b8e2SJens Axboe } 30966f3b8e2SJens Axboe 310a9185b41SChristoph Hellwig static int write_inode(struct inode *inode, struct writeback_control *wbc) 31166f3b8e2SJens Axboe { 3129fb0a7daSTejun Heo int ret; 3139fb0a7daSTejun Heo 3149fb0a7daSTejun Heo if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) { 3159fb0a7daSTejun Heo trace_writeback_write_inode_start(inode, wbc); 3169fb0a7daSTejun Heo ret = inode->i_sb->s_op->write_inode(inode, wbc); 3179fb0a7daSTejun Heo trace_writeback_write_inode(inode, wbc); 3189fb0a7daSTejun Heo return ret; 3199fb0a7daSTejun Heo } 32003ba3782SJens Axboe return 0; 32166f3b8e2SJens Axboe } 32208d8e974SFengguang Wu 3232c136579SFengguang Wu /* 324169ebd90SJan Kara * Wait for writeback on an inode to complete. Called with i_lock held. 325169ebd90SJan Kara * Caller must make sure inode cannot go away when we drop i_lock. 32601c03194SChristoph Hellwig */ 327169ebd90SJan Kara static void __inode_wait_for_writeback(struct inode *inode) 328169ebd90SJan Kara __releases(inode->i_lock) 329169ebd90SJan Kara __acquires(inode->i_lock) 33001c03194SChristoph Hellwig { 33101c03194SChristoph Hellwig DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); 33201c03194SChristoph Hellwig wait_queue_head_t *wqh; 33301c03194SChristoph Hellwig 33401c03194SChristoph Hellwig wqh = bit_waitqueue(&inode->i_state, __I_SYNC); 33558a9d3d8SRichard Kennedy while (inode->i_state & I_SYNC) { 336250df6edSDave Chinner spin_unlock(&inode->i_lock); 33701c03194SChristoph Hellwig __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE); 338250df6edSDave Chinner spin_lock(&inode->i_lock); 33958a9d3d8SRichard Kennedy } 34001c03194SChristoph Hellwig } 34101c03194SChristoph Hellwig 34201c03194SChristoph Hellwig /* 343169ebd90SJan Kara * Wait for writeback on an inode to complete. Caller must have inode pinned. 344169ebd90SJan Kara */ 345169ebd90SJan Kara void inode_wait_for_writeback(struct inode *inode) 346169ebd90SJan Kara { 347169ebd90SJan Kara spin_lock(&inode->i_lock); 348169ebd90SJan Kara __inode_wait_for_writeback(inode); 349169ebd90SJan Kara spin_unlock(&inode->i_lock); 350169ebd90SJan Kara } 351169ebd90SJan Kara 352169ebd90SJan Kara /* 353169ebd90SJan Kara * Sleep until I_SYNC is cleared. This function must be called with i_lock 354169ebd90SJan Kara * held and drops it. It is aimed for callers not holding any inode reference 355169ebd90SJan Kara * so once i_lock is dropped, inode can go away. 356169ebd90SJan Kara */ 357169ebd90SJan Kara static void inode_sleep_on_writeback(struct inode *inode) 358169ebd90SJan Kara __releases(inode->i_lock) 359169ebd90SJan Kara { 360169ebd90SJan Kara DEFINE_WAIT(wait); 361169ebd90SJan Kara wait_queue_head_t *wqh = bit_waitqueue(&inode->i_state, __I_SYNC); 362169ebd90SJan Kara int sleep; 363169ebd90SJan Kara 364169ebd90SJan Kara prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); 365169ebd90SJan Kara sleep = inode->i_state & I_SYNC; 366169ebd90SJan Kara spin_unlock(&inode->i_lock); 367169ebd90SJan Kara if (sleep) 368169ebd90SJan Kara schedule(); 369169ebd90SJan Kara finish_wait(wqh, &wait); 370169ebd90SJan Kara } 371169ebd90SJan Kara 372169ebd90SJan Kara /* 373ccb26b5aSJan Kara * Find proper writeback list for the inode depending on its current state and 374ccb26b5aSJan Kara * possibly also change of its state while we were doing writeback. Here we 375ccb26b5aSJan Kara * handle things such as livelock prevention or fairness of writeback among 376ccb26b5aSJan Kara * inodes. This function can be called only by flusher thread - noone else 377ccb26b5aSJan Kara * processes all inodes in writeback lists and requeueing inodes behind flusher 378ccb26b5aSJan Kara * thread's back can have unexpected consequences. 379ccb26b5aSJan Kara */ 380ccb26b5aSJan Kara static void requeue_inode(struct inode *inode, struct bdi_writeback *wb, 381ccb26b5aSJan Kara struct writeback_control *wbc) 382ccb26b5aSJan Kara { 383ccb26b5aSJan Kara if (inode->i_state & I_FREEING) 384ccb26b5aSJan Kara return; 385ccb26b5aSJan Kara 386ccb26b5aSJan Kara /* 387ccb26b5aSJan Kara * Sync livelock prevention. Each inode is tagged and synced in one 388ccb26b5aSJan Kara * shot. If still dirty, it will be redirty_tail()'ed below. Update 389ccb26b5aSJan Kara * the dirty time to prevent enqueue and sync it again. 390ccb26b5aSJan Kara */ 391ccb26b5aSJan Kara if ((inode->i_state & I_DIRTY) && 392ccb26b5aSJan Kara (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)) 393ccb26b5aSJan Kara inode->dirtied_when = jiffies; 394ccb26b5aSJan Kara 3954f8ad655SJan Kara if (wbc->pages_skipped) { 3964f8ad655SJan Kara /* 3974f8ad655SJan Kara * writeback is not making progress due to locked 3984f8ad655SJan Kara * buffers. Skip this inode for now. 3994f8ad655SJan Kara */ 4004f8ad655SJan Kara redirty_tail(inode, wb); 4014f8ad655SJan Kara return; 4024f8ad655SJan Kara } 4034f8ad655SJan Kara 404ccb26b5aSJan Kara if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) { 405ccb26b5aSJan Kara /* 406ccb26b5aSJan Kara * We didn't write back all the pages. nfs_writepages() 407ccb26b5aSJan Kara * sometimes bales out without doing anything. 408ccb26b5aSJan Kara */ 409ccb26b5aSJan Kara if (wbc->nr_to_write <= 0) { 410ccb26b5aSJan Kara /* Slice used up. Queue for next turn. */ 411ccb26b5aSJan Kara requeue_io(inode, wb); 412ccb26b5aSJan Kara } else { 413ccb26b5aSJan Kara /* 414ccb26b5aSJan Kara * Writeback blocked by something other than 415ccb26b5aSJan Kara * congestion. Delay the inode for some time to 416ccb26b5aSJan Kara * avoid spinning on the CPU (100% iowait) 417ccb26b5aSJan Kara * retrying writeback of the dirty page/inode 418ccb26b5aSJan Kara * that cannot be performed immediately. 419ccb26b5aSJan Kara */ 420ccb26b5aSJan Kara redirty_tail(inode, wb); 421ccb26b5aSJan Kara } 422ccb26b5aSJan Kara } else if (inode->i_state & I_DIRTY) { 423ccb26b5aSJan Kara /* 424ccb26b5aSJan Kara * Filesystems can dirty the inode during writeback operations, 425ccb26b5aSJan Kara * such as delayed allocation during submission or metadata 426ccb26b5aSJan Kara * updates after data IO completion. 427ccb26b5aSJan Kara */ 428ccb26b5aSJan Kara redirty_tail(inode, wb); 429ccb26b5aSJan Kara } else { 430ccb26b5aSJan Kara /* The inode is clean. Remove from writeback lists. */ 431ccb26b5aSJan Kara list_del_init(&inode->i_wb_list); 432ccb26b5aSJan Kara } 433ccb26b5aSJan Kara } 434ccb26b5aSJan Kara 435ccb26b5aSJan Kara /* 4364f8ad655SJan Kara * Write out an inode and its dirty pages. Do not update the writeback list 4374f8ad655SJan Kara * linkage. That is left to the caller. The caller is also responsible for 4384f8ad655SJan Kara * setting I_SYNC flag and calling inode_sync_complete() to clear it. 4391da177e4SLinus Torvalds */ 4401da177e4SLinus Torvalds static int 441cd8ed2a4SYan Hong __writeback_single_inode(struct inode *inode, struct writeback_control *wbc) 4421da177e4SLinus Torvalds { 4431da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 444251d6a47SWu Fengguang long nr_to_write = wbc->nr_to_write; 44501c03194SChristoph Hellwig unsigned dirty; 4461da177e4SLinus Torvalds int ret; 4471da177e4SLinus Torvalds 4484f8ad655SJan Kara WARN_ON(!(inode->i_state & I_SYNC)); 4491da177e4SLinus Torvalds 4509fb0a7daSTejun Heo trace_writeback_single_inode_start(inode, wbc, nr_to_write); 4519fb0a7daSTejun Heo 4521da177e4SLinus Torvalds ret = do_writepages(mapping, wbc); 4531da177e4SLinus Torvalds 45426821ed4SChristoph Hellwig /* 45526821ed4SChristoph Hellwig * Make sure to wait on the data before writing out the metadata. 45626821ed4SChristoph Hellwig * This is important for filesystems that modify metadata on data 4577747bd4bSDave Chinner * I/O completion. We don't do it for sync(2) writeback because it has a 4587747bd4bSDave Chinner * separate, external IO completion path and ->sync_fs for guaranteeing 4597747bd4bSDave Chinner * inode metadata is written back correctly. 46026821ed4SChristoph Hellwig */ 4617747bd4bSDave Chinner if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) { 46226821ed4SChristoph Hellwig int err = filemap_fdatawait(mapping); 4631da177e4SLinus Torvalds if (ret == 0) 4641da177e4SLinus Torvalds ret = err; 4651da177e4SLinus Torvalds } 4661da177e4SLinus Torvalds 4675547e8aaSDmitry Monakhov /* 4685547e8aaSDmitry Monakhov * Some filesystems may redirty the inode during the writeback 4695547e8aaSDmitry Monakhov * due to delalloc, clear dirty metadata flags right before 4705547e8aaSDmitry Monakhov * write_inode() 4715547e8aaSDmitry Monakhov */ 472250df6edSDave Chinner spin_lock(&inode->i_lock); 4736290be1cSJan Kara /* Clear I_DIRTY_PAGES if we've written out all dirty pages */ 4746290be1cSJan Kara if (!mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) 4756290be1cSJan Kara inode->i_state &= ~I_DIRTY_PAGES; 4765547e8aaSDmitry Monakhov dirty = inode->i_state & I_DIRTY; 4775547e8aaSDmitry Monakhov inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC); 478250df6edSDave Chinner spin_unlock(&inode->i_lock); 47926821ed4SChristoph Hellwig /* Don't write the inode if only I_DIRTY_PAGES was set */ 48026821ed4SChristoph Hellwig if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { 481a9185b41SChristoph Hellwig int err = write_inode(inode, wbc); 4821da177e4SLinus Torvalds if (ret == 0) 4831da177e4SLinus Torvalds ret = err; 4841da177e4SLinus Torvalds } 4854f8ad655SJan Kara trace_writeback_single_inode(inode, wbc, nr_to_write); 4864f8ad655SJan Kara return ret; 4874f8ad655SJan Kara } 4884f8ad655SJan Kara 4894f8ad655SJan Kara /* 4904f8ad655SJan Kara * Write out an inode's dirty pages. Either the caller has an active reference 4914f8ad655SJan Kara * on the inode or the inode has I_WILL_FREE set. 4924f8ad655SJan Kara * 4934f8ad655SJan Kara * This function is designed to be called for writing back one inode which 4944f8ad655SJan Kara * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode() 4954f8ad655SJan Kara * and does more profound writeback list handling in writeback_sb_inodes(). 4964f8ad655SJan Kara */ 4974f8ad655SJan Kara static int 4984f8ad655SJan Kara writeback_single_inode(struct inode *inode, struct bdi_writeback *wb, 4994f8ad655SJan Kara struct writeback_control *wbc) 5004f8ad655SJan Kara { 5014f8ad655SJan Kara int ret = 0; 5024f8ad655SJan Kara 5034f8ad655SJan Kara spin_lock(&inode->i_lock); 5044f8ad655SJan Kara if (!atomic_read(&inode->i_count)) 5054f8ad655SJan Kara WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); 5064f8ad655SJan Kara else 5074f8ad655SJan Kara WARN_ON(inode->i_state & I_WILL_FREE); 5084f8ad655SJan Kara 5094f8ad655SJan Kara if (inode->i_state & I_SYNC) { 5104f8ad655SJan Kara if (wbc->sync_mode != WB_SYNC_ALL) 5114f8ad655SJan Kara goto out; 5124f8ad655SJan Kara /* 513169ebd90SJan Kara * It's a data-integrity sync. We must wait. Since callers hold 514169ebd90SJan Kara * inode reference or inode has I_WILL_FREE set, it cannot go 515169ebd90SJan Kara * away under us. 5164f8ad655SJan Kara */ 517169ebd90SJan Kara __inode_wait_for_writeback(inode); 5184f8ad655SJan Kara } 5194f8ad655SJan Kara WARN_ON(inode->i_state & I_SYNC); 5204f8ad655SJan Kara /* 521f9b0e058SJan Kara * Skip inode if it is clean and we have no outstanding writeback in 522f9b0e058SJan Kara * WB_SYNC_ALL mode. We don't want to mess with writeback lists in this 523f9b0e058SJan Kara * function since flusher thread may be doing for example sync in 524f9b0e058SJan Kara * parallel and if we move the inode, it could get skipped. So here we 525f9b0e058SJan Kara * make sure inode is on some writeback list and leave it there unless 526f9b0e058SJan Kara * we have completely cleaned the inode. 5274f8ad655SJan Kara */ 528f9b0e058SJan Kara if (!(inode->i_state & I_DIRTY) && 529f9b0e058SJan Kara (wbc->sync_mode != WB_SYNC_ALL || 530f9b0e058SJan Kara !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK))) 5314f8ad655SJan Kara goto out; 5324f8ad655SJan Kara inode->i_state |= I_SYNC; 5334f8ad655SJan Kara spin_unlock(&inode->i_lock); 5344f8ad655SJan Kara 535cd8ed2a4SYan Hong ret = __writeback_single_inode(inode, wbc); 5361da177e4SLinus Torvalds 537f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 538250df6edSDave Chinner spin_lock(&inode->i_lock); 5394f8ad655SJan Kara /* 5404f8ad655SJan Kara * If inode is clean, remove it from writeback lists. Otherwise don't 5414f8ad655SJan Kara * touch it. See comment above for explanation. 5424f8ad655SJan Kara */ 5434f8ad655SJan Kara if (!(inode->i_state & I_DIRTY)) 5444f8ad655SJan Kara list_del_init(&inode->i_wb_list); 5454f8ad655SJan Kara spin_unlock(&wb->list_lock); 5461c0eeaf5SJoern Engel inode_sync_complete(inode); 5474f8ad655SJan Kara out: 5484f8ad655SJan Kara spin_unlock(&inode->i_lock); 5491da177e4SLinus Torvalds return ret; 5501da177e4SLinus Torvalds } 5511da177e4SLinus Torvalds 5521a12d8bdSWu Fengguang static long writeback_chunk_size(struct backing_dev_info *bdi, 5531a12d8bdSWu Fengguang struct wb_writeback_work *work) 554d46db3d5SWu Fengguang { 555d46db3d5SWu Fengguang long pages; 556d46db3d5SWu Fengguang 557d46db3d5SWu Fengguang /* 558d46db3d5SWu Fengguang * WB_SYNC_ALL mode does livelock avoidance by syncing dirty 559d46db3d5SWu Fengguang * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX 560d46db3d5SWu Fengguang * here avoids calling into writeback_inodes_wb() more than once. 561d46db3d5SWu Fengguang * 562d46db3d5SWu Fengguang * The intended call sequence for WB_SYNC_ALL writeback is: 563d46db3d5SWu Fengguang * 564d46db3d5SWu Fengguang * wb_writeback() 565d46db3d5SWu Fengguang * writeback_sb_inodes() <== called only once 566d46db3d5SWu Fengguang * write_cache_pages() <== called once for each inode 567d46db3d5SWu Fengguang * (quickly) tag currently dirty pages 568d46db3d5SWu Fengguang * (maybe slowly) sync all tagged pages 569d46db3d5SWu Fengguang */ 570d46db3d5SWu Fengguang if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages) 571d46db3d5SWu Fengguang pages = LONG_MAX; 5721a12d8bdSWu Fengguang else { 5731a12d8bdSWu Fengguang pages = min(bdi->avg_write_bandwidth / 2, 5741a12d8bdSWu Fengguang global_dirty_limit / DIRTY_SCOPE); 5751a12d8bdSWu Fengguang pages = min(pages, work->nr_pages); 5761a12d8bdSWu Fengguang pages = round_down(pages + MIN_WRITEBACK_PAGES, 5771a12d8bdSWu Fengguang MIN_WRITEBACK_PAGES); 5781a12d8bdSWu Fengguang } 579d46db3d5SWu Fengguang 580d46db3d5SWu Fengguang return pages; 581d46db3d5SWu Fengguang } 582d46db3d5SWu Fengguang 58303ba3782SJens Axboe /* 584f11c9c5cSEdward Shishkin * Write a portion of b_io inodes which belong to @sb. 585edadfb10SChristoph Hellwig * 586d46db3d5SWu Fengguang * Return the number of pages and/or inodes written. 587f11c9c5cSEdward Shishkin */ 588d46db3d5SWu Fengguang static long writeback_sb_inodes(struct super_block *sb, 589d46db3d5SWu Fengguang struct bdi_writeback *wb, 590d46db3d5SWu Fengguang struct wb_writeback_work *work) 59103ba3782SJens Axboe { 592d46db3d5SWu Fengguang struct writeback_control wbc = { 593d46db3d5SWu Fengguang .sync_mode = work->sync_mode, 594d46db3d5SWu Fengguang .tagged_writepages = work->tagged_writepages, 595d46db3d5SWu Fengguang .for_kupdate = work->for_kupdate, 596d46db3d5SWu Fengguang .for_background = work->for_background, 5977747bd4bSDave Chinner .for_sync = work->for_sync, 598d46db3d5SWu Fengguang .range_cyclic = work->range_cyclic, 599d46db3d5SWu Fengguang .range_start = 0, 600d46db3d5SWu Fengguang .range_end = LLONG_MAX, 601d46db3d5SWu Fengguang }; 602d46db3d5SWu Fengguang unsigned long start_time = jiffies; 603d46db3d5SWu Fengguang long write_chunk; 604d46db3d5SWu Fengguang long wrote = 0; /* count both pages and inodes */ 605d46db3d5SWu Fengguang 60603ba3782SJens Axboe while (!list_empty(&wb->b_io)) { 6077ccf19a8SNick Piggin struct inode *inode = wb_inode(wb->b_io.prev); 608edadfb10SChristoph Hellwig 609edadfb10SChristoph Hellwig if (inode->i_sb != sb) { 610d46db3d5SWu Fengguang if (work->sb) { 611edadfb10SChristoph Hellwig /* 612edadfb10SChristoph Hellwig * We only want to write back data for this 613edadfb10SChristoph Hellwig * superblock, move all inodes not belonging 614edadfb10SChristoph Hellwig * to it back onto the dirty list. 615edadfb10SChristoph Hellwig */ 616f758eeabSChristoph Hellwig redirty_tail(inode, wb); 61766f3b8e2SJens Axboe continue; 61866f3b8e2SJens Axboe } 619edadfb10SChristoph Hellwig 620edadfb10SChristoph Hellwig /* 621edadfb10SChristoph Hellwig * The inode belongs to a different superblock. 622edadfb10SChristoph Hellwig * Bounce back to the caller to unpin this and 623edadfb10SChristoph Hellwig * pin the next superblock. 624edadfb10SChristoph Hellwig */ 625d46db3d5SWu Fengguang break; 626edadfb10SChristoph Hellwig } 627edadfb10SChristoph Hellwig 6289843b76aSChristoph Hellwig /* 629331cbdeeSWanpeng Li * Don't bother with new inodes or inodes being freed, first 630331cbdeeSWanpeng Li * kind does not need periodic writeout yet, and for the latter 6319843b76aSChristoph Hellwig * kind writeout is handled by the freer. 6329843b76aSChristoph Hellwig */ 633250df6edSDave Chinner spin_lock(&inode->i_lock); 6349843b76aSChristoph Hellwig if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { 635250df6edSDave Chinner spin_unlock(&inode->i_lock); 636fcc5c222SWu Fengguang redirty_tail(inode, wb); 6377ef0d737SNick Piggin continue; 6387ef0d737SNick Piggin } 639cc1676d9SJan Kara if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) { 640cc1676d9SJan Kara /* 641cc1676d9SJan Kara * If this inode is locked for writeback and we are not 642cc1676d9SJan Kara * doing writeback-for-data-integrity, move it to 643cc1676d9SJan Kara * b_more_io so that writeback can proceed with the 644cc1676d9SJan Kara * other inodes on s_io. 645cc1676d9SJan Kara * 646cc1676d9SJan Kara * We'll have another go at writing back this inode 647cc1676d9SJan Kara * when we completed a full scan of b_io. 648cc1676d9SJan Kara */ 649cc1676d9SJan Kara spin_unlock(&inode->i_lock); 650cc1676d9SJan Kara requeue_io(inode, wb); 651cc1676d9SJan Kara trace_writeback_sb_inodes_requeue(inode); 652cc1676d9SJan Kara continue; 653cc1676d9SJan Kara } 654f0d07b7fSJan Kara spin_unlock(&wb->list_lock); 655f0d07b7fSJan Kara 6564f8ad655SJan Kara /* 6574f8ad655SJan Kara * We already requeued the inode if it had I_SYNC set and we 6584f8ad655SJan Kara * are doing WB_SYNC_NONE writeback. So this catches only the 6594f8ad655SJan Kara * WB_SYNC_ALL case. 6604f8ad655SJan Kara */ 661169ebd90SJan Kara if (inode->i_state & I_SYNC) { 662169ebd90SJan Kara /* Wait for I_SYNC. This function drops i_lock... */ 663169ebd90SJan Kara inode_sleep_on_writeback(inode); 664169ebd90SJan Kara /* Inode may be gone, start again */ 665ead188f9SJan Kara spin_lock(&wb->list_lock); 666169ebd90SJan Kara continue; 667169ebd90SJan Kara } 6684f8ad655SJan Kara inode->i_state |= I_SYNC; 6694f8ad655SJan Kara spin_unlock(&inode->i_lock); 670169ebd90SJan Kara 6711a12d8bdSWu Fengguang write_chunk = writeback_chunk_size(wb->bdi, work); 672d46db3d5SWu Fengguang wbc.nr_to_write = write_chunk; 673d46db3d5SWu Fengguang wbc.pages_skipped = 0; 674250df6edSDave Chinner 675169ebd90SJan Kara /* 676169ebd90SJan Kara * We use I_SYNC to pin the inode in memory. While it is set 677169ebd90SJan Kara * evict_inode() will wait so the inode cannot be freed. 678169ebd90SJan Kara */ 679cd8ed2a4SYan Hong __writeback_single_inode(inode, &wbc); 680d46db3d5SWu Fengguang 681d46db3d5SWu Fengguang work->nr_pages -= write_chunk - wbc.nr_to_write; 682d46db3d5SWu Fengguang wrote += write_chunk - wbc.nr_to_write; 6834f8ad655SJan Kara spin_lock(&wb->list_lock); 6844f8ad655SJan Kara spin_lock(&inode->i_lock); 685d46db3d5SWu Fengguang if (!(inode->i_state & I_DIRTY)) 686d46db3d5SWu Fengguang wrote++; 6874f8ad655SJan Kara requeue_inode(inode, wb, &wbc); 6884f8ad655SJan Kara inode_sync_complete(inode); 6890f1b1fd8SDave Chinner spin_unlock(&inode->i_lock); 690169ebd90SJan Kara cond_resched_lock(&wb->list_lock); 691d46db3d5SWu Fengguang /* 692d46db3d5SWu Fengguang * bail out to wb_writeback() often enough to check 693d46db3d5SWu Fengguang * background threshold and other termination conditions. 694d46db3d5SWu Fengguang */ 695d46db3d5SWu Fengguang if (wrote) { 696d46db3d5SWu Fengguang if (time_is_before_jiffies(start_time + HZ / 10UL)) 697d46db3d5SWu Fengguang break; 698d46db3d5SWu Fengguang if (work->nr_pages <= 0) 699d46db3d5SWu Fengguang break; 7001da177e4SLinus Torvalds } 7018bc3be27SFengguang Wu } 702d46db3d5SWu Fengguang return wrote; 703f11c9c5cSEdward Shishkin } 70438f21977SNick Piggin 705d46db3d5SWu Fengguang static long __writeback_inodes_wb(struct bdi_writeback *wb, 706d46db3d5SWu Fengguang struct wb_writeback_work *work) 707f11c9c5cSEdward Shishkin { 708d46db3d5SWu Fengguang unsigned long start_time = jiffies; 709d46db3d5SWu Fengguang long wrote = 0; 710f11c9c5cSEdward Shishkin 711f11c9c5cSEdward Shishkin while (!list_empty(&wb->b_io)) { 7127ccf19a8SNick Piggin struct inode *inode = wb_inode(wb->b_io.prev); 713f11c9c5cSEdward Shishkin struct super_block *sb = inode->i_sb; 714f11c9c5cSEdward Shishkin 71512ad3ab6SDave Chinner if (!grab_super_passive(sb)) { 7160e995816SWu Fengguang /* 7170e995816SWu Fengguang * grab_super_passive() may fail consistently due to 7180e995816SWu Fengguang * s_umount being grabbed by someone else. Don't use 7190e995816SWu Fengguang * requeue_io() to avoid busy retrying the inode/sb. 7200e995816SWu Fengguang */ 7210e995816SWu Fengguang redirty_tail(inode, wb); 722d19de7edSChristoph Hellwig continue; 723334132aeSChristoph Hellwig } 724d46db3d5SWu Fengguang wrote += writeback_sb_inodes(sb, wb, work); 725d19de7edSChristoph Hellwig drop_super(sb); 726f11c9c5cSEdward Shishkin 727d46db3d5SWu Fengguang /* refer to the same tests at the end of writeback_sb_inodes */ 728d46db3d5SWu Fengguang if (wrote) { 729d46db3d5SWu Fengguang if (time_is_before_jiffies(start_time + HZ / 10UL)) 730d46db3d5SWu Fengguang break; 731d46db3d5SWu Fengguang if (work->nr_pages <= 0) 732f11c9c5cSEdward Shishkin break; 733f11c9c5cSEdward Shishkin } 734d46db3d5SWu Fengguang } 73566f3b8e2SJens Axboe /* Leave any unwritten inodes on b_io */ 736d46db3d5SWu Fengguang return wrote; 73766f3b8e2SJens Axboe } 73866f3b8e2SJens Axboe 7397d9f073bSWanpeng Li static long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages, 7400e175a18SCurt Wohlgemuth enum wb_reason reason) 741edadfb10SChristoph Hellwig { 742d46db3d5SWu Fengguang struct wb_writeback_work work = { 743d46db3d5SWu Fengguang .nr_pages = nr_pages, 744d46db3d5SWu Fengguang .sync_mode = WB_SYNC_NONE, 745d46db3d5SWu Fengguang .range_cyclic = 1, 7460e175a18SCurt Wohlgemuth .reason = reason, 747c4a391b5SJan Kara .older_than_this = jiffies, 748c4a391b5SJan Kara .older_than_this_is_set = 1, 749d46db3d5SWu Fengguang }; 750edadfb10SChristoph Hellwig 751f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 752424b351fSWu Fengguang if (list_empty(&wb->b_io)) 753ad4e38ddSCurt Wohlgemuth queue_io(wb, &work); 754d46db3d5SWu Fengguang __writeback_inodes_wb(wb, &work); 755f758eeabSChristoph Hellwig spin_unlock(&wb->list_lock); 756edadfb10SChristoph Hellwig 757d46db3d5SWu Fengguang return nr_pages - work.nr_pages; 75866f3b8e2SJens Axboe } 75966f3b8e2SJens Axboe 760b00949aaSWu Fengguang static bool over_bground_thresh(struct backing_dev_info *bdi) 76103ba3782SJens Axboe { 76203ba3782SJens Axboe unsigned long background_thresh, dirty_thresh; 76303ba3782SJens Axboe 76416c4042fSWu Fengguang global_dirty_limits(&background_thresh, &dirty_thresh); 76503ba3782SJens Axboe 766b00949aaSWu Fengguang if (global_page_state(NR_FILE_DIRTY) + 767b00949aaSWu Fengguang global_page_state(NR_UNSTABLE_NFS) > background_thresh) 768b00949aaSWu Fengguang return true; 769b00949aaSWu Fengguang 770b00949aaSWu Fengguang if (bdi_stat(bdi, BDI_RECLAIMABLE) > 771b00949aaSWu Fengguang bdi_dirty_limit(bdi, background_thresh)) 772b00949aaSWu Fengguang return true; 773b00949aaSWu Fengguang 774b00949aaSWu Fengguang return false; 77503ba3782SJens Axboe } 77603ba3782SJens Axboe 77703ba3782SJens Axboe /* 778e98be2d5SWu Fengguang * Called under wb->list_lock. If there are multiple wb per bdi, 779e98be2d5SWu Fengguang * only the flusher working on the first wb should do it. 780e98be2d5SWu Fengguang */ 781e98be2d5SWu Fengguang static void wb_update_bandwidth(struct bdi_writeback *wb, 782e98be2d5SWu Fengguang unsigned long start_time) 783e98be2d5SWu Fengguang { 784af6a3113SWu Fengguang __bdi_update_bandwidth(wb->bdi, 0, 0, 0, 0, 0, start_time); 785e98be2d5SWu Fengguang } 786e98be2d5SWu Fengguang 787e98be2d5SWu Fengguang /* 78803ba3782SJens Axboe * Explicit flushing or periodic writeback of "old" data. 78903ba3782SJens Axboe * 79003ba3782SJens Axboe * Define "old": the first time one of an inode's pages is dirtied, we mark the 79103ba3782SJens Axboe * dirtying-time in the inode's address_space. So this periodic writeback code 79203ba3782SJens Axboe * just walks the superblock inode list, writing back any inodes which are 79303ba3782SJens Axboe * older than a specific point in time. 79403ba3782SJens Axboe * 79503ba3782SJens Axboe * Try to run once per dirty_writeback_interval. But if a writeback event 79603ba3782SJens Axboe * takes longer than a dirty_writeback_interval interval, then leave a 79703ba3782SJens Axboe * one-second gap. 79803ba3782SJens Axboe * 79903ba3782SJens Axboe * older_than_this takes precedence over nr_to_write. So we'll only write back 80003ba3782SJens Axboe * all dirty pages if they are all attached to "old" mappings. 80103ba3782SJens Axboe */ 802c4a77a6cSJens Axboe static long wb_writeback(struct bdi_writeback *wb, 80383ba7b07SChristoph Hellwig struct wb_writeback_work *work) 80403ba3782SJens Axboe { 805e98be2d5SWu Fengguang unsigned long wb_start = jiffies; 806d46db3d5SWu Fengguang long nr_pages = work->nr_pages; 807a5989bdcSJan Kara struct inode *inode; 808d46db3d5SWu Fengguang long progress; 80903ba3782SJens Axboe 810c4a391b5SJan Kara if (!work->older_than_this_is_set) { 811c4a391b5SJan Kara work->older_than_this = jiffies; 812c4a391b5SJan Kara work->older_than_this_is_set = 1; 813c4a391b5SJan Kara } 81403ba3782SJens Axboe 815e8dfc305SWu Fengguang spin_lock(&wb->list_lock); 81603ba3782SJens Axboe for (;;) { 81703ba3782SJens Axboe /* 818d3ddec76SWu Fengguang * Stop writeback when nr_pages has been consumed 81903ba3782SJens Axboe */ 82083ba7b07SChristoph Hellwig if (work->nr_pages <= 0) 82103ba3782SJens Axboe break; 82203ba3782SJens Axboe 82303ba3782SJens Axboe /* 824aa373cf5SJan Kara * Background writeout and kupdate-style writeback may 825aa373cf5SJan Kara * run forever. Stop them if there is other work to do 826aa373cf5SJan Kara * so that e.g. sync can proceed. They'll be restarted 827aa373cf5SJan Kara * after the other works are all done. 828aa373cf5SJan Kara */ 829aa373cf5SJan Kara if ((work->for_background || work->for_kupdate) && 830aa373cf5SJan Kara !list_empty(&wb->bdi->work_list)) 831aa373cf5SJan Kara break; 832aa373cf5SJan Kara 833aa373cf5SJan Kara /* 834d3ddec76SWu Fengguang * For background writeout, stop when we are below the 835d3ddec76SWu Fengguang * background dirty threshold 83603ba3782SJens Axboe */ 837b00949aaSWu Fengguang if (work->for_background && !over_bground_thresh(wb->bdi)) 83803ba3782SJens Axboe break; 83903ba3782SJens Axboe 8401bc36b64SJan Kara /* 8411bc36b64SJan Kara * Kupdate and background works are special and we want to 8421bc36b64SJan Kara * include all inodes that need writing. Livelock avoidance is 8431bc36b64SJan Kara * handled by these works yielding to any other work so we are 8441bc36b64SJan Kara * safe. 8451bc36b64SJan Kara */ 846ba9aa839SWu Fengguang if (work->for_kupdate) { 847c4a391b5SJan Kara work->older_than_this = jiffies - 848ba9aa839SWu Fengguang msecs_to_jiffies(dirty_expire_interval * 10); 8491bc36b64SJan Kara } else if (work->for_background) 850c4a391b5SJan Kara work->older_than_this = jiffies; 851028c2dd1SDave Chinner 852d46db3d5SWu Fengguang trace_writeback_start(wb->bdi, work); 853e8dfc305SWu Fengguang if (list_empty(&wb->b_io)) 854ad4e38ddSCurt Wohlgemuth queue_io(wb, work); 85583ba7b07SChristoph Hellwig if (work->sb) 856d46db3d5SWu Fengguang progress = writeback_sb_inodes(work->sb, wb, work); 857edadfb10SChristoph Hellwig else 858d46db3d5SWu Fengguang progress = __writeback_inodes_wb(wb, work); 859d46db3d5SWu Fengguang trace_writeback_written(wb->bdi, work); 860028c2dd1SDave Chinner 861e98be2d5SWu Fengguang wb_update_bandwidth(wb, wb_start); 86203ba3782SJens Axboe 86303ba3782SJens Axboe /* 86471fd05a8SJens Axboe * Did we write something? Try for more 865e6fb6da2SWu Fengguang * 866e6fb6da2SWu Fengguang * Dirty inodes are moved to b_io for writeback in batches. 867e6fb6da2SWu Fengguang * The completion of the current batch does not necessarily 868e6fb6da2SWu Fengguang * mean the overall work is done. So we keep looping as long 869e6fb6da2SWu Fengguang * as made some progress on cleaning pages or inodes. 87071fd05a8SJens Axboe */ 871d46db3d5SWu Fengguang if (progress) 87203ba3782SJens Axboe continue; 873a5989bdcSJan Kara /* 874e6fb6da2SWu Fengguang * No more inodes for IO, bail 875a5989bdcSJan Kara */ 876b7a2441fSWu Fengguang if (list_empty(&wb->b_more_io)) 87703ba3782SJens Axboe break; 87803ba3782SJens Axboe /* 8798010c3b6SJens Axboe * Nothing written. Wait for some inode to 8808010c3b6SJens Axboe * become available for writeback. Otherwise 8818010c3b6SJens Axboe * we'll just busyloop. 88203ba3782SJens Axboe */ 88303ba3782SJens Axboe if (!list_empty(&wb->b_more_io)) { 884d46db3d5SWu Fengguang trace_writeback_wait(wb->bdi, work); 88503ba3782SJens Axboe inode = wb_inode(wb->b_more_io.prev); 886250df6edSDave Chinner spin_lock(&inode->i_lock); 887f0d07b7fSJan Kara spin_unlock(&wb->list_lock); 888169ebd90SJan Kara /* This function drops i_lock... */ 889169ebd90SJan Kara inode_sleep_on_writeback(inode); 890f0d07b7fSJan Kara spin_lock(&wb->list_lock); 89103ba3782SJens Axboe } 89203ba3782SJens Axboe } 893e8dfc305SWu Fengguang spin_unlock(&wb->list_lock); 89403ba3782SJens Axboe 895d46db3d5SWu Fengguang return nr_pages - work->nr_pages; 89603ba3782SJens Axboe } 89703ba3782SJens Axboe 89803ba3782SJens Axboe /* 89983ba7b07SChristoph Hellwig * Return the next wb_writeback_work struct that hasn't been processed yet. 90003ba3782SJens Axboe */ 90183ba7b07SChristoph Hellwig static struct wb_writeback_work * 90208852b6dSMinchan Kim get_next_work_item(struct backing_dev_info *bdi) 90303ba3782SJens Axboe { 90483ba7b07SChristoph Hellwig struct wb_writeback_work *work = NULL; 90503ba3782SJens Axboe 9066467716aSArtem Bityutskiy spin_lock_bh(&bdi->wb_lock); 90783ba7b07SChristoph Hellwig if (!list_empty(&bdi->work_list)) { 90883ba7b07SChristoph Hellwig work = list_entry(bdi->work_list.next, 90983ba7b07SChristoph Hellwig struct wb_writeback_work, list); 91083ba7b07SChristoph Hellwig list_del_init(&work->list); 91103ba3782SJens Axboe } 9126467716aSArtem Bityutskiy spin_unlock_bh(&bdi->wb_lock); 91383ba7b07SChristoph Hellwig return work; 91403ba3782SJens Axboe } 91503ba3782SJens Axboe 916cdf01dd5SLinus Torvalds /* 917cdf01dd5SLinus Torvalds * Add in the number of potentially dirty inodes, because each inode 918cdf01dd5SLinus Torvalds * write can dirty pagecache in the underlying blockdev. 919cdf01dd5SLinus Torvalds */ 920cdf01dd5SLinus Torvalds static unsigned long get_nr_dirty_pages(void) 921cdf01dd5SLinus Torvalds { 922cdf01dd5SLinus Torvalds return global_page_state(NR_FILE_DIRTY) + 923cdf01dd5SLinus Torvalds global_page_state(NR_UNSTABLE_NFS) + 924cdf01dd5SLinus Torvalds get_nr_dirty_inodes(); 925cdf01dd5SLinus Torvalds } 926cdf01dd5SLinus Torvalds 9276585027aSJan Kara static long wb_check_background_flush(struct bdi_writeback *wb) 9286585027aSJan Kara { 929b00949aaSWu Fengguang if (over_bground_thresh(wb->bdi)) { 9306585027aSJan Kara 9316585027aSJan Kara struct wb_writeback_work work = { 9326585027aSJan Kara .nr_pages = LONG_MAX, 9336585027aSJan Kara .sync_mode = WB_SYNC_NONE, 9346585027aSJan Kara .for_background = 1, 9356585027aSJan Kara .range_cyclic = 1, 9360e175a18SCurt Wohlgemuth .reason = WB_REASON_BACKGROUND, 9376585027aSJan Kara }; 9386585027aSJan Kara 9396585027aSJan Kara return wb_writeback(wb, &work); 9406585027aSJan Kara } 9416585027aSJan Kara 9426585027aSJan Kara return 0; 9436585027aSJan Kara } 9446585027aSJan Kara 94503ba3782SJens Axboe static long wb_check_old_data_flush(struct bdi_writeback *wb) 94603ba3782SJens Axboe { 94703ba3782SJens Axboe unsigned long expired; 94803ba3782SJens Axboe long nr_pages; 94903ba3782SJens Axboe 95069b62d01SJens Axboe /* 95169b62d01SJens Axboe * When set to zero, disable periodic writeback 95269b62d01SJens Axboe */ 95369b62d01SJens Axboe if (!dirty_writeback_interval) 95469b62d01SJens Axboe return 0; 95569b62d01SJens Axboe 95603ba3782SJens Axboe expired = wb->last_old_flush + 95703ba3782SJens Axboe msecs_to_jiffies(dirty_writeback_interval * 10); 95803ba3782SJens Axboe if (time_before(jiffies, expired)) 95903ba3782SJens Axboe return 0; 96003ba3782SJens Axboe 96103ba3782SJens Axboe wb->last_old_flush = jiffies; 962cdf01dd5SLinus Torvalds nr_pages = get_nr_dirty_pages(); 96303ba3782SJens Axboe 964c4a77a6cSJens Axboe if (nr_pages) { 96583ba7b07SChristoph Hellwig struct wb_writeback_work work = { 966c4a77a6cSJens Axboe .nr_pages = nr_pages, 967c4a77a6cSJens Axboe .sync_mode = WB_SYNC_NONE, 968c4a77a6cSJens Axboe .for_kupdate = 1, 969c4a77a6cSJens Axboe .range_cyclic = 1, 9700e175a18SCurt Wohlgemuth .reason = WB_REASON_PERIODIC, 971c4a77a6cSJens Axboe }; 972c4a77a6cSJens Axboe 97383ba7b07SChristoph Hellwig return wb_writeback(wb, &work); 974c4a77a6cSJens Axboe } 97503ba3782SJens Axboe 97603ba3782SJens Axboe return 0; 97703ba3782SJens Axboe } 97803ba3782SJens Axboe 97903ba3782SJens Axboe /* 98003ba3782SJens Axboe * Retrieve work items and do the writeback they describe 98103ba3782SJens Axboe */ 98225d130baSWanpeng Li static long wb_do_writeback(struct bdi_writeback *wb) 98303ba3782SJens Axboe { 98403ba3782SJens Axboe struct backing_dev_info *bdi = wb->bdi; 98583ba7b07SChristoph Hellwig struct wb_writeback_work *work; 986c4a77a6cSJens Axboe long wrote = 0; 98703ba3782SJens Axboe 98881d73a32SJan Kara set_bit(BDI_writeback_running, &wb->bdi->state); 98908852b6dSMinchan Kim while ((work = get_next_work_item(bdi)) != NULL) { 99083ba7b07SChristoph Hellwig 991455b2864SDave Chinner trace_writeback_exec(bdi, work); 992455b2864SDave Chinner 99383ba7b07SChristoph Hellwig wrote += wb_writeback(wb, work); 99403ba3782SJens Axboe 99503ba3782SJens Axboe /* 99683ba7b07SChristoph Hellwig * Notify the caller of completion if this is a synchronous 99783ba7b07SChristoph Hellwig * work item, otherwise just free it. 99803ba3782SJens Axboe */ 99983ba7b07SChristoph Hellwig if (work->done) 100083ba7b07SChristoph Hellwig complete(work->done); 100183ba7b07SChristoph Hellwig else 100283ba7b07SChristoph Hellwig kfree(work); 100303ba3782SJens Axboe } 100403ba3782SJens Axboe 100503ba3782SJens Axboe /* 100603ba3782SJens Axboe * Check for periodic writeback, kupdated() style 100703ba3782SJens Axboe */ 100803ba3782SJens Axboe wrote += wb_check_old_data_flush(wb); 10096585027aSJan Kara wrote += wb_check_background_flush(wb); 101081d73a32SJan Kara clear_bit(BDI_writeback_running, &wb->bdi->state); 101103ba3782SJens Axboe 101203ba3782SJens Axboe return wrote; 101303ba3782SJens Axboe } 101403ba3782SJens Axboe 101503ba3782SJens Axboe /* 101603ba3782SJens Axboe * Handle writeback of dirty data for the device backed by this bdi. Also 1017839a8e86STejun Heo * reschedules periodically and does kupdated style flushing. 101803ba3782SJens Axboe */ 1019839a8e86STejun Heo void bdi_writeback_workfn(struct work_struct *work) 102003ba3782SJens Axboe { 1021839a8e86STejun Heo struct bdi_writeback *wb = container_of(to_delayed_work(work), 1022839a8e86STejun Heo struct bdi_writeback, dwork); 102308243900SChristoph Hellwig struct backing_dev_info *bdi = wb->bdi; 102403ba3782SJens Axboe long pages_written; 102503ba3782SJens Axboe 1026ef3b1019STejun Heo set_worker_desc("flush-%s", dev_name(bdi->dev)); 1027766f9164SPeter Zijlstra current->flags |= PF_SWAPWRITE; 102803ba3782SJens Axboe 1029839a8e86STejun Heo if (likely(!current_is_workqueue_rescuer() || 1030839a8e86STejun Heo list_empty(&bdi->bdi_list))) { 103103ba3782SJens Axboe /* 1032839a8e86STejun Heo * The normal path. Keep writing back @bdi until its 1033839a8e86STejun Heo * work_list is empty. Note that this path is also taken 1034839a8e86STejun Heo * if @bdi is shutting down even when we're running off the 1035839a8e86STejun Heo * rescuer as work_list needs to be drained. 103603ba3782SJens Axboe */ 1037839a8e86STejun Heo do { 103825d130baSWanpeng Li pages_written = wb_do_writeback(wb); 1039455b2864SDave Chinner trace_writeback_pages_written(pages_written); 1040839a8e86STejun Heo } while (!list_empty(&bdi->work_list)); 1041839a8e86STejun Heo } else { 1042253c34e9SArtem Bityutskiy /* 1043839a8e86STejun Heo * bdi_wq can't get enough workers and we're running off 1044839a8e86STejun Heo * the emergency worker. Don't hog it. Hopefully, 1024 is 1045839a8e86STejun Heo * enough for efficient IO. 1046253c34e9SArtem Bityutskiy */ 1047839a8e86STejun Heo pages_written = writeback_inodes_wb(&bdi->wb, 1024, 1048839a8e86STejun Heo WB_REASON_FORKER_THREAD); 1049839a8e86STejun Heo trace_writeback_pages_written(pages_written); 105003ba3782SJens Axboe } 105103ba3782SJens Axboe 1052839a8e86STejun Heo if (!list_empty(&bdi->work_list) || 1053839a8e86STejun Heo (wb_has_dirty_io(wb) && dirty_writeback_interval)) 1054839a8e86STejun Heo queue_delayed_work(bdi_wq, &wb->dwork, 1055839a8e86STejun Heo msecs_to_jiffies(dirty_writeback_interval * 10)); 1056455b2864SDave Chinner 1057839a8e86STejun Heo current->flags &= ~PF_SWAPWRITE; 105803ba3782SJens Axboe } 105903ba3782SJens Axboe 106003ba3782SJens Axboe /* 106103ba3782SJens Axboe * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back 106203ba3782SJens Axboe * the whole world. 106303ba3782SJens Axboe */ 10640e175a18SCurt Wohlgemuth void wakeup_flusher_threads(long nr_pages, enum wb_reason reason) 106503ba3782SJens Axboe { 1066b8c2f347SChristoph Hellwig struct backing_dev_info *bdi; 1067b8c2f347SChristoph Hellwig 106847df3ddeSJan Kara if (!nr_pages) 106947df3ddeSJan Kara nr_pages = get_nr_dirty_pages(); 1070b8c2f347SChristoph Hellwig 1071b8c2f347SChristoph Hellwig rcu_read_lock(); 1072b8c2f347SChristoph Hellwig list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { 1073b8c2f347SChristoph Hellwig if (!bdi_has_dirty_io(bdi)) 1074b8c2f347SChristoph Hellwig continue; 10750e175a18SCurt Wohlgemuth __bdi_start_writeback(bdi, nr_pages, false, reason); 1076b8c2f347SChristoph Hellwig } 1077b8c2f347SChristoph Hellwig rcu_read_unlock(); 107803ba3782SJens Axboe } 107903ba3782SJens Axboe 108003ba3782SJens Axboe static noinline void block_dump___mark_inode_dirty(struct inode *inode) 108103ba3782SJens Axboe { 108203ba3782SJens Axboe if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { 108303ba3782SJens Axboe struct dentry *dentry; 108403ba3782SJens Axboe const char *name = "?"; 108503ba3782SJens Axboe 108603ba3782SJens Axboe dentry = d_find_alias(inode); 108703ba3782SJens Axboe if (dentry) { 108803ba3782SJens Axboe spin_lock(&dentry->d_lock); 108903ba3782SJens Axboe name = (const char *) dentry->d_name.name; 109003ba3782SJens Axboe } 109103ba3782SJens Axboe printk(KERN_DEBUG 109203ba3782SJens Axboe "%s(%d): dirtied inode %lu (%s) on %s\n", 109303ba3782SJens Axboe current->comm, task_pid_nr(current), inode->i_ino, 109403ba3782SJens Axboe name, inode->i_sb->s_id); 109503ba3782SJens Axboe if (dentry) { 109603ba3782SJens Axboe spin_unlock(&dentry->d_lock); 109703ba3782SJens Axboe dput(dentry); 109803ba3782SJens Axboe } 109903ba3782SJens Axboe } 110003ba3782SJens Axboe } 110103ba3782SJens Axboe 110203ba3782SJens Axboe /** 110303ba3782SJens Axboe * __mark_inode_dirty - internal function 110403ba3782SJens Axboe * @inode: inode to mark 110503ba3782SJens Axboe * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) 110603ba3782SJens Axboe * Mark an inode as dirty. Callers should use mark_inode_dirty or 110703ba3782SJens Axboe * mark_inode_dirty_sync. 110803ba3782SJens Axboe * 110903ba3782SJens Axboe * Put the inode on the super block's dirty list. 111003ba3782SJens Axboe * 111103ba3782SJens Axboe * CAREFUL! We mark it dirty unconditionally, but move it onto the 111203ba3782SJens Axboe * dirty list only if it is hashed or if it refers to a blockdev. 111303ba3782SJens Axboe * If it was not hashed, it will never be added to the dirty list 111403ba3782SJens Axboe * even if it is later hashed, as it will have been marked dirty already. 111503ba3782SJens Axboe * 111603ba3782SJens Axboe * In short, make sure you hash any inodes _before_ you start marking 111703ba3782SJens Axboe * them dirty. 111803ba3782SJens Axboe * 111903ba3782SJens Axboe * Note that for blockdevs, inode->dirtied_when represents the dirtying time of 112003ba3782SJens Axboe * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of 112103ba3782SJens Axboe * the kernel-internal blockdev inode represents the dirtying time of the 112203ba3782SJens Axboe * blockdev's pages. This is why for I_DIRTY_PAGES we always use 112303ba3782SJens Axboe * page->mapping->host, so the page-dirtying time is recorded in the internal 112403ba3782SJens Axboe * blockdev inode. 112503ba3782SJens Axboe */ 112603ba3782SJens Axboe void __mark_inode_dirty(struct inode *inode, int flags) 112703ba3782SJens Axboe { 112803ba3782SJens Axboe struct super_block *sb = inode->i_sb; 1129253c34e9SArtem Bityutskiy struct backing_dev_info *bdi = NULL; 113003ba3782SJens Axboe 113103ba3782SJens Axboe /* 113203ba3782SJens Axboe * Don't do this for I_DIRTY_PAGES - that doesn't actually 113303ba3782SJens Axboe * dirty the inode itself 113403ba3782SJens Axboe */ 113503ba3782SJens Axboe if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { 11369fb0a7daSTejun Heo trace_writeback_dirty_inode_start(inode, flags); 11379fb0a7daSTejun Heo 113803ba3782SJens Axboe if (sb->s_op->dirty_inode) 1139aa385729SChristoph Hellwig sb->s_op->dirty_inode(inode, flags); 11409fb0a7daSTejun Heo 11419fb0a7daSTejun Heo trace_writeback_dirty_inode(inode, flags); 114203ba3782SJens Axboe } 114303ba3782SJens Axboe 114403ba3782SJens Axboe /* 114503ba3782SJens Axboe * make sure that changes are seen by all cpus before we test i_state 114603ba3782SJens Axboe * -- mikulas 114703ba3782SJens Axboe */ 114803ba3782SJens Axboe smp_mb(); 114903ba3782SJens Axboe 115003ba3782SJens Axboe /* avoid the locking if we can */ 115103ba3782SJens Axboe if ((inode->i_state & flags) == flags) 115203ba3782SJens Axboe return; 115303ba3782SJens Axboe 115403ba3782SJens Axboe if (unlikely(block_dump)) 115503ba3782SJens Axboe block_dump___mark_inode_dirty(inode); 115603ba3782SJens Axboe 1157250df6edSDave Chinner spin_lock(&inode->i_lock); 115803ba3782SJens Axboe if ((inode->i_state & flags) != flags) { 115903ba3782SJens Axboe const int was_dirty = inode->i_state & I_DIRTY; 116003ba3782SJens Axboe 116103ba3782SJens Axboe inode->i_state |= flags; 116203ba3782SJens Axboe 116303ba3782SJens Axboe /* 116403ba3782SJens Axboe * If the inode is being synced, just update its dirty state. 116503ba3782SJens Axboe * The unlocker will place the inode on the appropriate 116603ba3782SJens Axboe * superblock list, based upon its state. 116703ba3782SJens Axboe */ 116803ba3782SJens Axboe if (inode->i_state & I_SYNC) 1169250df6edSDave Chinner goto out_unlock_inode; 117003ba3782SJens Axboe 117103ba3782SJens Axboe /* 117203ba3782SJens Axboe * Only add valid (hashed) inodes to the superblock's 117303ba3782SJens Axboe * dirty list. Add blockdev inodes as well. 117403ba3782SJens Axboe */ 117503ba3782SJens Axboe if (!S_ISBLK(inode->i_mode)) { 11761d3382cbSAl Viro if (inode_unhashed(inode)) 1177250df6edSDave Chinner goto out_unlock_inode; 117803ba3782SJens Axboe } 1179a4ffdde6SAl Viro if (inode->i_state & I_FREEING) 1180250df6edSDave Chinner goto out_unlock_inode; 118103ba3782SJens Axboe 118203ba3782SJens Axboe /* 118303ba3782SJens Axboe * If the inode was already on b_dirty/b_io/b_more_io, don't 118403ba3782SJens Axboe * reposition it (that would break b_dirty time-ordering). 118503ba3782SJens Axboe */ 118603ba3782SJens Axboe if (!was_dirty) { 1187a66979abSDave Chinner bool wakeup_bdi = false; 1188253c34e9SArtem Bityutskiy bdi = inode_to_bdi(inode); 1189500b067cSJens Axboe 1190146d7009SJunxiao Bi spin_unlock(&inode->i_lock); 1191146d7009SJunxiao Bi spin_lock(&bdi->wb.list_lock); 1192253c34e9SArtem Bityutskiy if (bdi_cap_writeback_dirty(bdi)) { 1193253c34e9SArtem Bityutskiy WARN(!test_bit(BDI_registered, &bdi->state), 1194253c34e9SArtem Bityutskiy "bdi-%s not registered\n", bdi->name); 1195253c34e9SArtem Bityutskiy 1196253c34e9SArtem Bityutskiy /* 1197253c34e9SArtem Bityutskiy * If this is the first dirty inode for this 1198253c34e9SArtem Bityutskiy * bdi, we have to wake-up the corresponding 1199253c34e9SArtem Bityutskiy * bdi thread to make sure background 1200253c34e9SArtem Bityutskiy * write-back happens later. 1201253c34e9SArtem Bityutskiy */ 1202253c34e9SArtem Bityutskiy if (!wb_has_dirty_io(&bdi->wb)) 1203253c34e9SArtem Bityutskiy wakeup_bdi = true; 1204500b067cSJens Axboe } 120503ba3782SJens Axboe 120603ba3782SJens Axboe inode->dirtied_when = jiffies; 12077ccf19a8SNick Piggin list_move(&inode->i_wb_list, &bdi->wb.b_dirty); 1208f758eeabSChristoph Hellwig spin_unlock(&bdi->wb.list_lock); 1209253c34e9SArtem Bityutskiy 1210253c34e9SArtem Bityutskiy if (wakeup_bdi) 12116467716aSArtem Bityutskiy bdi_wakeup_thread_delayed(bdi); 1212a66979abSDave Chinner return; 1213a66979abSDave Chinner } 1214a66979abSDave Chinner } 1215a66979abSDave Chinner out_unlock_inode: 1216a66979abSDave Chinner spin_unlock(&inode->i_lock); 1217a66979abSDave Chinner 121803ba3782SJens Axboe } 121903ba3782SJens Axboe EXPORT_SYMBOL(__mark_inode_dirty); 122003ba3782SJens Axboe 1221b6e51316SJens Axboe static void wait_sb_inodes(struct super_block *sb) 122266f3b8e2SJens Axboe { 122338f21977SNick Piggin struct inode *inode, *old_inode = NULL; 122438f21977SNick Piggin 122503ba3782SJens Axboe /* 122603ba3782SJens Axboe * We need to be protected against the filesystem going from 122703ba3782SJens Axboe * r/o to r/w or vice versa. 122803ba3782SJens Axboe */ 1229b6e51316SJens Axboe WARN_ON(!rwsem_is_locked(&sb->s_umount)); 123003ba3782SJens Axboe 123155fa6091SDave Chinner spin_lock(&inode_sb_list_lock); 123266f3b8e2SJens Axboe 123338f21977SNick Piggin /* 123438f21977SNick Piggin * Data integrity sync. Must wait for all pages under writeback, 123538f21977SNick Piggin * because there may have been pages dirtied before our sync 123638f21977SNick Piggin * call, but which had writeout started before we write it out. 123738f21977SNick Piggin * In which case, the inode may not be on the dirty list, but 123838f21977SNick Piggin * we still have to wait for that writeout. 123938f21977SNick Piggin */ 1240b6e51316SJens Axboe list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { 1241250df6edSDave Chinner struct address_space *mapping = inode->i_mapping; 124238f21977SNick Piggin 1243250df6edSDave Chinner spin_lock(&inode->i_lock); 1244250df6edSDave Chinner if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) || 1245250df6edSDave Chinner (mapping->nrpages == 0)) { 1246250df6edSDave Chinner spin_unlock(&inode->i_lock); 124738f21977SNick Piggin continue; 1248250df6edSDave Chinner } 124938f21977SNick Piggin __iget(inode); 1250250df6edSDave Chinner spin_unlock(&inode->i_lock); 125155fa6091SDave Chinner spin_unlock(&inode_sb_list_lock); 125255fa6091SDave Chinner 125338f21977SNick Piggin /* 125455fa6091SDave Chinner * We hold a reference to 'inode' so it couldn't have been 125555fa6091SDave Chinner * removed from s_inodes list while we dropped the 125655fa6091SDave Chinner * inode_sb_list_lock. We cannot iput the inode now as we can 125755fa6091SDave Chinner * be holding the last reference and we cannot iput it under 125855fa6091SDave Chinner * inode_sb_list_lock. So we keep the reference and iput it 125955fa6091SDave Chinner * later. 126038f21977SNick Piggin */ 126138f21977SNick Piggin iput(old_inode); 126238f21977SNick Piggin old_inode = inode; 126338f21977SNick Piggin 126438f21977SNick Piggin filemap_fdatawait(mapping); 126538f21977SNick Piggin 126638f21977SNick Piggin cond_resched(); 126738f21977SNick Piggin 126855fa6091SDave Chinner spin_lock(&inode_sb_list_lock); 126938f21977SNick Piggin } 127055fa6091SDave Chinner spin_unlock(&inode_sb_list_lock); 127138f21977SNick Piggin iput(old_inode); 127266f3b8e2SJens Axboe } 12731da177e4SLinus Torvalds 1274d8a8559cSJens Axboe /** 12753259f8beSChris Mason * writeback_inodes_sb_nr - writeback dirty inodes from given super_block 1276d8a8559cSJens Axboe * @sb: the superblock 12773259f8beSChris Mason * @nr: the number of pages to write 1278786228abSMarcos Paulo de Souza * @reason: reason why some writeback work initiated 12791da177e4SLinus Torvalds * 1280d8a8559cSJens Axboe * Start writeback on some inodes on this super_block. No guarantees are made 1281d8a8559cSJens Axboe * on how many (if any) will be written, and this function does not wait 12823259f8beSChris Mason * for IO completion of submitted IO. 12831da177e4SLinus Torvalds */ 12840e175a18SCurt Wohlgemuth void writeback_inodes_sb_nr(struct super_block *sb, 12850e175a18SCurt Wohlgemuth unsigned long nr, 12860e175a18SCurt Wohlgemuth enum wb_reason reason) 12871da177e4SLinus Torvalds { 128883ba7b07SChristoph Hellwig DECLARE_COMPLETION_ONSTACK(done); 128983ba7b07SChristoph Hellwig struct wb_writeback_work work = { 12903c4d7165SChristoph Hellwig .sb = sb, 12913c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_NONE, 12926e6938b6SWu Fengguang .tagged_writepages = 1, 129383ba7b07SChristoph Hellwig .done = &done, 12943259f8beSChris Mason .nr_pages = nr, 12950e175a18SCurt Wohlgemuth .reason = reason, 12963c4d7165SChristoph Hellwig }; 12970e3c9a22SJens Axboe 12986eedc701SJan Kara if (sb->s_bdi == &noop_backing_dev_info) 12996eedc701SJan Kara return; 1300cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 130183ba7b07SChristoph Hellwig bdi_queue_work(sb->s_bdi, &work); 130283ba7b07SChristoph Hellwig wait_for_completion(&done); 13031da177e4SLinus Torvalds } 13043259f8beSChris Mason EXPORT_SYMBOL(writeback_inodes_sb_nr); 13053259f8beSChris Mason 13063259f8beSChris Mason /** 13073259f8beSChris Mason * writeback_inodes_sb - writeback dirty inodes from given super_block 13083259f8beSChris Mason * @sb: the superblock 1309786228abSMarcos Paulo de Souza * @reason: reason why some writeback work was initiated 13103259f8beSChris Mason * 13113259f8beSChris Mason * Start writeback on some inodes on this super_block. No guarantees are made 13123259f8beSChris Mason * on how many (if any) will be written, and this function does not wait 13133259f8beSChris Mason * for IO completion of submitted IO. 13143259f8beSChris Mason */ 13150e175a18SCurt Wohlgemuth void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) 13163259f8beSChris Mason { 13170e175a18SCurt Wohlgemuth return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); 13183259f8beSChris Mason } 1319d8a8559cSJens Axboe EXPORT_SYMBOL(writeback_inodes_sb); 1320d8a8559cSJens Axboe 1321d8a8559cSJens Axboe /** 132210ee27a0SMiao Xie * try_to_writeback_inodes_sb_nr - try to start writeback if none underway 13233259f8beSChris Mason * @sb: the superblock 13243259f8beSChris Mason * @nr: the number of pages to write 132510ee27a0SMiao Xie * @reason: the reason of writeback 13263259f8beSChris Mason * 132710ee27a0SMiao Xie * Invoke writeback_inodes_sb_nr if no writeback is currently underway. 13283259f8beSChris Mason * Returns 1 if writeback was started, 0 if not. 13293259f8beSChris Mason */ 133010ee27a0SMiao Xie int try_to_writeback_inodes_sb_nr(struct super_block *sb, 13310e175a18SCurt Wohlgemuth unsigned long nr, 13320e175a18SCurt Wohlgemuth enum wb_reason reason) 13333259f8beSChris Mason { 133410ee27a0SMiao Xie if (writeback_in_progress(sb->s_bdi)) 133510ee27a0SMiao Xie return 1; 133610ee27a0SMiao Xie 133710ee27a0SMiao Xie if (!down_read_trylock(&sb->s_umount)) 133810ee27a0SMiao Xie return 0; 133910ee27a0SMiao Xie 13400e175a18SCurt Wohlgemuth writeback_inodes_sb_nr(sb, nr, reason); 13413259f8beSChris Mason up_read(&sb->s_umount); 13423259f8beSChris Mason return 1; 13433259f8beSChris Mason } 134410ee27a0SMiao Xie EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr); 134510ee27a0SMiao Xie 134610ee27a0SMiao Xie /** 134710ee27a0SMiao Xie * try_to_writeback_inodes_sb - try to start writeback if none underway 134810ee27a0SMiao Xie * @sb: the superblock 134910ee27a0SMiao Xie * @reason: reason why some writeback work was initiated 135010ee27a0SMiao Xie * 135110ee27a0SMiao Xie * Implement by try_to_writeback_inodes_sb_nr() 135210ee27a0SMiao Xie * Returns 1 if writeback was started, 0 if not. 135310ee27a0SMiao Xie */ 135410ee27a0SMiao Xie int try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) 135510ee27a0SMiao Xie { 135610ee27a0SMiao Xie return try_to_writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); 135710ee27a0SMiao Xie } 135810ee27a0SMiao Xie EXPORT_SYMBOL(try_to_writeback_inodes_sb); 13593259f8beSChris Mason 13603259f8beSChris Mason /** 1361d8a8559cSJens Axboe * sync_inodes_sb - sync sb inode pages 1362d8a8559cSJens Axboe * @sb: the superblock 1363c4a391b5SJan Kara * @older_than_this: timestamp 1364d8a8559cSJens Axboe * 1365d8a8559cSJens Axboe * This function writes and waits on any dirty inode belonging to this 1366c4a391b5SJan Kara * superblock that has been dirtied before given timestamp. 1367d8a8559cSJens Axboe */ 1368c4a391b5SJan Kara void sync_inodes_sb(struct super_block *sb, unsigned long older_than_this) 1369d8a8559cSJens Axboe { 137083ba7b07SChristoph Hellwig DECLARE_COMPLETION_ONSTACK(done); 137183ba7b07SChristoph Hellwig struct wb_writeback_work work = { 13723c4d7165SChristoph Hellwig .sb = sb, 13733c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_ALL, 13743c4d7165SChristoph Hellwig .nr_pages = LONG_MAX, 1375c4a391b5SJan Kara .older_than_this = older_than_this, 1376c4a391b5SJan Kara .older_than_this_is_set = 1, 13773c4d7165SChristoph Hellwig .range_cyclic = 0, 137883ba7b07SChristoph Hellwig .done = &done, 13790e175a18SCurt Wohlgemuth .reason = WB_REASON_SYNC, 13807747bd4bSDave Chinner .for_sync = 1, 13813c4d7165SChristoph Hellwig }; 13823c4d7165SChristoph Hellwig 13836eedc701SJan Kara /* Nothing to do? */ 13846eedc701SJan Kara if (sb->s_bdi == &noop_backing_dev_info) 13856eedc701SJan Kara return; 1386cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 1387cf37e972SChristoph Hellwig 138883ba7b07SChristoph Hellwig bdi_queue_work(sb->s_bdi, &work); 138983ba7b07SChristoph Hellwig wait_for_completion(&done); 139083ba7b07SChristoph Hellwig 1391b6e51316SJens Axboe wait_sb_inodes(sb); 1392d8a8559cSJens Axboe } 1393d8a8559cSJens Axboe EXPORT_SYMBOL(sync_inodes_sb); 13941da177e4SLinus Torvalds 13951da177e4SLinus Torvalds /** 13961da177e4SLinus Torvalds * write_inode_now - write an inode to disk 13971da177e4SLinus Torvalds * @inode: inode to write to disk 13981da177e4SLinus Torvalds * @sync: whether the write should be synchronous or not 13991da177e4SLinus Torvalds * 14007f04c26dSAndrea Arcangeli * This function commits an inode to disk immediately if it is dirty. This is 14017f04c26dSAndrea Arcangeli * primarily needed by knfsd. 14027f04c26dSAndrea Arcangeli * 14037f04c26dSAndrea Arcangeli * The caller must either have a ref on the inode or must have set I_WILL_FREE. 14041da177e4SLinus Torvalds */ 14051da177e4SLinus Torvalds int write_inode_now(struct inode *inode, int sync) 14061da177e4SLinus Torvalds { 1407f758eeabSChristoph Hellwig struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; 14081da177e4SLinus Torvalds struct writeback_control wbc = { 14091da177e4SLinus Torvalds .nr_to_write = LONG_MAX, 141018914b18SMike Galbraith .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, 1411111ebb6eSOGAWA Hirofumi .range_start = 0, 1412111ebb6eSOGAWA Hirofumi .range_end = LLONG_MAX, 14131da177e4SLinus Torvalds }; 14141da177e4SLinus Torvalds 14151da177e4SLinus Torvalds if (!mapping_cap_writeback_dirty(inode->i_mapping)) 141649364ce2SAndrew Morton wbc.nr_to_write = 0; 14171da177e4SLinus Torvalds 14181da177e4SLinus Torvalds might_sleep(); 14194f8ad655SJan Kara return writeback_single_inode(inode, wb, &wbc); 14201da177e4SLinus Torvalds } 14211da177e4SLinus Torvalds EXPORT_SYMBOL(write_inode_now); 14221da177e4SLinus Torvalds 14231da177e4SLinus Torvalds /** 14241da177e4SLinus Torvalds * sync_inode - write an inode and its pages to disk. 14251da177e4SLinus Torvalds * @inode: the inode to sync 14261da177e4SLinus Torvalds * @wbc: controls the writeback mode 14271da177e4SLinus Torvalds * 14281da177e4SLinus Torvalds * sync_inode() will write an inode and its pages to disk. It will also 14291da177e4SLinus Torvalds * correctly update the inode on its superblock's dirty inode lists and will 14301da177e4SLinus Torvalds * update inode->i_state. 14311da177e4SLinus Torvalds * 14321da177e4SLinus Torvalds * The caller must have a ref on the inode. 14331da177e4SLinus Torvalds */ 14341da177e4SLinus Torvalds int sync_inode(struct inode *inode, struct writeback_control *wbc) 14351da177e4SLinus Torvalds { 14364f8ad655SJan Kara return writeback_single_inode(inode, &inode_to_bdi(inode)->wb, wbc); 14371da177e4SLinus Torvalds } 14381da177e4SLinus Torvalds EXPORT_SYMBOL(sync_inode); 1439c3765016SChristoph Hellwig 1440c3765016SChristoph Hellwig /** 1441c691b9d9SAndrew Morton * sync_inode_metadata - write an inode to disk 1442c3765016SChristoph Hellwig * @inode: the inode to sync 1443c3765016SChristoph Hellwig * @wait: wait for I/O to complete. 1444c3765016SChristoph Hellwig * 1445c691b9d9SAndrew Morton * Write an inode to disk and adjust its dirty state after completion. 1446c3765016SChristoph Hellwig * 1447c3765016SChristoph Hellwig * Note: only writes the actual inode, no associated data or other metadata. 1448c3765016SChristoph Hellwig */ 1449c3765016SChristoph Hellwig int sync_inode_metadata(struct inode *inode, int wait) 1450c3765016SChristoph Hellwig { 1451c3765016SChristoph Hellwig struct writeback_control wbc = { 1452c3765016SChristoph Hellwig .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE, 1453c3765016SChristoph Hellwig .nr_to_write = 0, /* metadata-only */ 1454c3765016SChristoph Hellwig }; 1455c3765016SChristoph Hellwig 1456c3765016SChristoph Hellwig return sync_inode(inode, &wbc); 1457c3765016SChristoph Hellwig } 1458c3765016SChristoph Hellwig EXPORT_SYMBOL(sync_inode_metadata); 1459