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 976585027aSJan Kara static void bdi_queue_work(struct backing_dev_info *bdi, 986585027aSJan Kara struct wb_writeback_work *work) 996585027aSJan Kara { 1006585027aSJan Kara trace_writeback_queue(bdi, work); 1016585027aSJan Kara 1026585027aSJan Kara spin_lock_bh(&bdi->wb_lock); 1036585027aSJan Kara list_add_tail(&work->list, &bdi->work_list); 1046467716aSArtem Bityutskiy spin_unlock_bh(&bdi->wb_lock); 105839a8e86STejun Heo 106839a8e86STejun Heo mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0); 10703ba3782SJens Axboe } 1081da177e4SLinus Torvalds 10983ba7b07SChristoph Hellwig static void 11083ba7b07SChristoph Hellwig __bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, 1110e175a18SCurt Wohlgemuth bool range_cyclic, enum wb_reason reason) 1121da177e4SLinus Torvalds { 11383ba7b07SChristoph Hellwig struct wb_writeback_work *work; 11403ba3782SJens Axboe 115bcddc3f0SJens Axboe /* 116bcddc3f0SJens Axboe * This is WB_SYNC_NONE writeback, so if allocation fails just 117bcddc3f0SJens Axboe * wakeup the thread for old dirty data writeback 118bcddc3f0SJens Axboe */ 11983ba7b07SChristoph Hellwig work = kzalloc(sizeof(*work), GFP_ATOMIC); 12083ba7b07SChristoph Hellwig if (!work) { 121455b2864SDave Chinner trace_writeback_nowork(bdi); 122839a8e86STejun Heo mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0); 12383ba7b07SChristoph Hellwig return; 12483ba7b07SChristoph Hellwig } 12583ba7b07SChristoph Hellwig 12683ba7b07SChristoph Hellwig work->sync_mode = WB_SYNC_NONE; 12783ba7b07SChristoph Hellwig work->nr_pages = nr_pages; 12883ba7b07SChristoph Hellwig work->range_cyclic = range_cyclic; 1290e175a18SCurt Wohlgemuth work->reason = reason; 13083ba7b07SChristoph Hellwig 131f11fcae8SJens Axboe bdi_queue_work(bdi, work); 13203ba3782SJens Axboe } 133b6e51316SJens Axboe 134b6e51316SJens Axboe /** 135b6e51316SJens Axboe * bdi_start_writeback - start writeback 136b6e51316SJens Axboe * @bdi: the backing device to write from 137b6e51316SJens Axboe * @nr_pages: the number of pages to write 138786228abSMarcos Paulo de Souza * @reason: reason why some writeback work was initiated 139b6e51316SJens Axboe * 140b6e51316SJens Axboe * Description: 141b6e51316SJens Axboe * This does WB_SYNC_NONE opportunistic writeback. The IO is only 14225985edcSLucas De Marchi * started when this function returns, we make no guarantees on 1430e3c9a22SJens Axboe * completion. Caller need not hold sb s_umount semaphore. 144b6e51316SJens Axboe * 145b6e51316SJens Axboe */ 1460e175a18SCurt Wohlgemuth void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, 1470e175a18SCurt Wohlgemuth enum wb_reason reason) 148b6e51316SJens Axboe { 1490e175a18SCurt Wohlgemuth __bdi_start_writeback(bdi, nr_pages, true, reason); 150d3ddec76SWu Fengguang } 151d3ddec76SWu Fengguang 152c5444198SChristoph Hellwig /** 153c5444198SChristoph Hellwig * bdi_start_background_writeback - start background writeback 154c5444198SChristoph Hellwig * @bdi: the backing device to write from 155c5444198SChristoph Hellwig * 156c5444198SChristoph Hellwig * Description: 1576585027aSJan Kara * This makes sure WB_SYNC_NONE background writeback happens. When 1586585027aSJan Kara * this function returns, it is only guaranteed that for given BDI 1596585027aSJan Kara * some IO is happening if we are over background dirty threshold. 1606585027aSJan Kara * Caller need not hold sb s_umount semaphore. 161c5444198SChristoph Hellwig */ 162c5444198SChristoph Hellwig void bdi_start_background_writeback(struct backing_dev_info *bdi) 163c5444198SChristoph Hellwig { 1646585027aSJan Kara /* 1656585027aSJan Kara * We just wake up the flusher thread. It will perform background 1666585027aSJan Kara * writeback as soon as there is no other work to do. 1676585027aSJan Kara */ 16871927e84SWu Fengguang trace_writeback_wake_background(bdi); 169839a8e86STejun Heo mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0); 1701da177e4SLinus Torvalds } 1711da177e4SLinus Torvalds 1721da177e4SLinus Torvalds /* 173a66979abSDave Chinner * Remove the inode from the writeback list it is on. 174a66979abSDave Chinner */ 175a66979abSDave Chinner void inode_wb_list_del(struct inode *inode) 176a66979abSDave Chinner { 177f758eeabSChristoph Hellwig struct backing_dev_info *bdi = inode_to_bdi(inode); 178a66979abSDave Chinner 179f758eeabSChristoph Hellwig spin_lock(&bdi->wb.list_lock); 180f758eeabSChristoph Hellwig list_del_init(&inode->i_wb_list); 181f758eeabSChristoph Hellwig spin_unlock(&bdi->wb.list_lock); 182f758eeabSChristoph Hellwig } 183a66979abSDave Chinner 184a66979abSDave Chinner /* 1856610a0bcSAndrew Morton * Redirty an inode: set its when-it-was dirtied timestamp and move it to the 1866610a0bcSAndrew Morton * furthest end of its superblock's dirty-inode list. 1876610a0bcSAndrew Morton * 1886610a0bcSAndrew Morton * Before stamping the inode's ->dirtied_when, we check to see whether it is 18966f3b8e2SJens Axboe * already the most-recently-dirtied inode on the b_dirty list. If that is 1906610a0bcSAndrew Morton * the case then the inode must have been redirtied while it was being written 1916610a0bcSAndrew Morton * out and we don't reset its dirtied_when. 1926610a0bcSAndrew Morton */ 193f758eeabSChristoph Hellwig static void redirty_tail(struct inode *inode, struct bdi_writeback *wb) 1946610a0bcSAndrew Morton { 195f758eeabSChristoph Hellwig assert_spin_locked(&wb->list_lock); 19603ba3782SJens Axboe if (!list_empty(&wb->b_dirty)) { 19766f3b8e2SJens Axboe struct inode *tail; 1986610a0bcSAndrew Morton 1997ccf19a8SNick Piggin tail = wb_inode(wb->b_dirty.next); 20066f3b8e2SJens Axboe if (time_before(inode->dirtied_when, tail->dirtied_when)) 2016610a0bcSAndrew Morton inode->dirtied_when = jiffies; 2026610a0bcSAndrew Morton } 2037ccf19a8SNick Piggin list_move(&inode->i_wb_list, &wb->b_dirty); 2046610a0bcSAndrew Morton } 2056610a0bcSAndrew Morton 2066610a0bcSAndrew Morton /* 20766f3b8e2SJens Axboe * requeue inode for re-scanning after bdi->b_io list is exhausted. 208c986d1e2SAndrew Morton */ 209f758eeabSChristoph Hellwig static void requeue_io(struct inode *inode, struct bdi_writeback *wb) 210c986d1e2SAndrew Morton { 211f758eeabSChristoph Hellwig assert_spin_locked(&wb->list_lock); 2127ccf19a8SNick Piggin list_move(&inode->i_wb_list, &wb->b_more_io); 213c986d1e2SAndrew Morton } 214c986d1e2SAndrew Morton 2151c0eeaf5SJoern Engel static void inode_sync_complete(struct inode *inode) 2161c0eeaf5SJoern Engel { 217365b94aeSJan Kara inode->i_state &= ~I_SYNC; 2184eff96ddSJan Kara /* If inode is clean an unused, put it into LRU now... */ 2194eff96ddSJan Kara inode_add_lru(inode); 220365b94aeSJan Kara /* Waiters must see I_SYNC cleared before being woken up */ 2211c0eeaf5SJoern Engel smp_mb(); 2221c0eeaf5SJoern Engel wake_up_bit(&inode->i_state, __I_SYNC); 2231c0eeaf5SJoern Engel } 2241c0eeaf5SJoern Engel 225d2caa3c5SJeff Layton static bool inode_dirtied_after(struct inode *inode, unsigned long t) 226d2caa3c5SJeff Layton { 227d2caa3c5SJeff Layton bool ret = time_after(inode->dirtied_when, t); 228d2caa3c5SJeff Layton #ifndef CONFIG_64BIT 229d2caa3c5SJeff Layton /* 230d2caa3c5SJeff Layton * For inodes being constantly redirtied, dirtied_when can get stuck. 231d2caa3c5SJeff Layton * It _appears_ to be in the future, but is actually in distant past. 232d2caa3c5SJeff Layton * This test is necessary to prevent such wrapped-around relative times 2335b0830cbSJens Axboe * from permanently stopping the whole bdi writeback. 234d2caa3c5SJeff Layton */ 235d2caa3c5SJeff Layton ret = ret && time_before_eq(inode->dirtied_when, jiffies); 236d2caa3c5SJeff Layton #endif 237d2caa3c5SJeff Layton return ret; 238d2caa3c5SJeff Layton } 239d2caa3c5SJeff Layton 240c986d1e2SAndrew Morton /* 2410e2f2b23SWang Sheng-Hui * Move expired (dirtied before work->older_than_this) dirty inodes from 242697e6fedSJan Kara * @delaying_queue to @dispatch_queue. 2432c136579SFengguang Wu */ 244e84d0a4fSWu Fengguang static int move_expired_inodes(struct list_head *delaying_queue, 2452c136579SFengguang Wu struct list_head *dispatch_queue, 246ad4e38ddSCurt Wohlgemuth struct wb_writeback_work *work) 2472c136579SFengguang Wu { 2485c03449dSShaohua Li LIST_HEAD(tmp); 2495c03449dSShaohua Li struct list_head *pos, *node; 250cf137307SJens Axboe struct super_block *sb = NULL; 2515c03449dSShaohua Li struct inode *inode; 252cf137307SJens Axboe int do_sb_sort = 0; 253e84d0a4fSWu Fengguang int moved = 0; 2545c03449dSShaohua Li 255c4a391b5SJan Kara WARN_ON_ONCE(!work->older_than_this_is_set); 2562c136579SFengguang Wu while (!list_empty(delaying_queue)) { 2577ccf19a8SNick Piggin inode = wb_inode(delaying_queue->prev); 258c4a391b5SJan Kara if (inode_dirtied_after(inode, work->older_than_this)) 2592c136579SFengguang Wu break; 260a8855990SJan Kara list_move(&inode->i_wb_list, &tmp); 261a8855990SJan Kara moved++; 262a8855990SJan Kara if (sb_is_blkdev_sb(inode->i_sb)) 263a8855990SJan Kara continue; 264cf137307SJens Axboe if (sb && sb != inode->i_sb) 265cf137307SJens Axboe do_sb_sort = 1; 266cf137307SJens Axboe sb = inode->i_sb; 2675c03449dSShaohua Li } 2685c03449dSShaohua Li 269cf137307SJens Axboe /* just one sb in list, splice to dispatch_queue and we're done */ 270cf137307SJens Axboe if (!do_sb_sort) { 271cf137307SJens Axboe list_splice(&tmp, dispatch_queue); 272e84d0a4fSWu Fengguang goto out; 273cf137307SJens Axboe } 274cf137307SJens Axboe 2755c03449dSShaohua Li /* Move inodes from one superblock together */ 2765c03449dSShaohua Li while (!list_empty(&tmp)) { 2777ccf19a8SNick Piggin sb = wb_inode(tmp.prev)->i_sb; 2785c03449dSShaohua Li list_for_each_prev_safe(pos, node, &tmp) { 2797ccf19a8SNick Piggin inode = wb_inode(pos); 2805c03449dSShaohua Li if (inode->i_sb == sb) 2817ccf19a8SNick Piggin list_move(&inode->i_wb_list, dispatch_queue); 2822c136579SFengguang Wu } 2832c136579SFengguang Wu } 284e84d0a4fSWu Fengguang out: 285e84d0a4fSWu Fengguang return moved; 2865c03449dSShaohua Li } 2872c136579SFengguang Wu 2882c136579SFengguang Wu /* 2892c136579SFengguang Wu * Queue all expired dirty inodes for io, eldest first. 2904ea879b9SWu Fengguang * Before 2914ea879b9SWu Fengguang * newly dirtied b_dirty b_io b_more_io 2924ea879b9SWu Fengguang * =============> gf edc BA 2934ea879b9SWu Fengguang * After 2944ea879b9SWu Fengguang * newly dirtied b_dirty b_io b_more_io 2954ea879b9SWu Fengguang * =============> g fBAedc 2964ea879b9SWu Fengguang * | 2974ea879b9SWu Fengguang * +--> dequeue for IO 2982c136579SFengguang Wu */ 299ad4e38ddSCurt Wohlgemuth static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work) 3002c136579SFengguang Wu { 301e84d0a4fSWu Fengguang int moved; 302f758eeabSChristoph Hellwig assert_spin_locked(&wb->list_lock); 3034ea879b9SWu Fengguang list_splice_init(&wb->b_more_io, &wb->b_io); 304ad4e38ddSCurt Wohlgemuth moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, work); 305ad4e38ddSCurt Wohlgemuth trace_writeback_queue_io(wb, work, moved); 30666f3b8e2SJens Axboe } 30766f3b8e2SJens Axboe 308a9185b41SChristoph Hellwig static int write_inode(struct inode *inode, struct writeback_control *wbc) 30966f3b8e2SJens Axboe { 3109fb0a7daSTejun Heo int ret; 3119fb0a7daSTejun Heo 3129fb0a7daSTejun Heo if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) { 3139fb0a7daSTejun Heo trace_writeback_write_inode_start(inode, wbc); 3149fb0a7daSTejun Heo ret = inode->i_sb->s_op->write_inode(inode, wbc); 3159fb0a7daSTejun Heo trace_writeback_write_inode(inode, wbc); 3169fb0a7daSTejun Heo return ret; 3179fb0a7daSTejun Heo } 31803ba3782SJens Axboe return 0; 31966f3b8e2SJens Axboe } 32008d8e974SFengguang Wu 3212c136579SFengguang Wu /* 322169ebd90SJan Kara * Wait for writeback on an inode to complete. Called with i_lock held. 323169ebd90SJan Kara * Caller must make sure inode cannot go away when we drop i_lock. 32401c03194SChristoph Hellwig */ 325169ebd90SJan Kara static void __inode_wait_for_writeback(struct inode *inode) 326169ebd90SJan Kara __releases(inode->i_lock) 327169ebd90SJan Kara __acquires(inode->i_lock) 32801c03194SChristoph Hellwig { 32901c03194SChristoph Hellwig DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); 33001c03194SChristoph Hellwig wait_queue_head_t *wqh; 33101c03194SChristoph Hellwig 33201c03194SChristoph Hellwig wqh = bit_waitqueue(&inode->i_state, __I_SYNC); 33358a9d3d8SRichard Kennedy while (inode->i_state & I_SYNC) { 334250df6edSDave Chinner spin_unlock(&inode->i_lock); 33501c03194SChristoph Hellwig __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE); 336250df6edSDave Chinner spin_lock(&inode->i_lock); 33758a9d3d8SRichard Kennedy } 33801c03194SChristoph Hellwig } 33901c03194SChristoph Hellwig 34001c03194SChristoph Hellwig /* 341169ebd90SJan Kara * Wait for writeback on an inode to complete. Caller must have inode pinned. 342169ebd90SJan Kara */ 343169ebd90SJan Kara void inode_wait_for_writeback(struct inode *inode) 344169ebd90SJan Kara { 345169ebd90SJan Kara spin_lock(&inode->i_lock); 346169ebd90SJan Kara __inode_wait_for_writeback(inode); 347169ebd90SJan Kara spin_unlock(&inode->i_lock); 348169ebd90SJan Kara } 349169ebd90SJan Kara 350169ebd90SJan Kara /* 351169ebd90SJan Kara * Sleep until I_SYNC is cleared. This function must be called with i_lock 352169ebd90SJan Kara * held and drops it. It is aimed for callers not holding any inode reference 353169ebd90SJan Kara * so once i_lock is dropped, inode can go away. 354169ebd90SJan Kara */ 355169ebd90SJan Kara static void inode_sleep_on_writeback(struct inode *inode) 356169ebd90SJan Kara __releases(inode->i_lock) 357169ebd90SJan Kara { 358169ebd90SJan Kara DEFINE_WAIT(wait); 359169ebd90SJan Kara wait_queue_head_t *wqh = bit_waitqueue(&inode->i_state, __I_SYNC); 360169ebd90SJan Kara int sleep; 361169ebd90SJan Kara 362169ebd90SJan Kara prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); 363169ebd90SJan Kara sleep = inode->i_state & I_SYNC; 364169ebd90SJan Kara spin_unlock(&inode->i_lock); 365169ebd90SJan Kara if (sleep) 366169ebd90SJan Kara schedule(); 367169ebd90SJan Kara finish_wait(wqh, &wait); 368169ebd90SJan Kara } 369169ebd90SJan Kara 370169ebd90SJan Kara /* 371ccb26b5aSJan Kara * Find proper writeback list for the inode depending on its current state and 372ccb26b5aSJan Kara * possibly also change of its state while we were doing writeback. Here we 373ccb26b5aSJan Kara * handle things such as livelock prevention or fairness of writeback among 374ccb26b5aSJan Kara * inodes. This function can be called only by flusher thread - noone else 375ccb26b5aSJan Kara * processes all inodes in writeback lists and requeueing inodes behind flusher 376ccb26b5aSJan Kara * thread's back can have unexpected consequences. 377ccb26b5aSJan Kara */ 378ccb26b5aSJan Kara static void requeue_inode(struct inode *inode, struct bdi_writeback *wb, 379ccb26b5aSJan Kara struct writeback_control *wbc) 380ccb26b5aSJan Kara { 381ccb26b5aSJan Kara if (inode->i_state & I_FREEING) 382ccb26b5aSJan Kara return; 383ccb26b5aSJan Kara 384ccb26b5aSJan Kara /* 385ccb26b5aSJan Kara * Sync livelock prevention. Each inode is tagged and synced in one 386ccb26b5aSJan Kara * shot. If still dirty, it will be redirty_tail()'ed below. Update 387ccb26b5aSJan Kara * the dirty time to prevent enqueue and sync it again. 388ccb26b5aSJan Kara */ 389ccb26b5aSJan Kara if ((inode->i_state & I_DIRTY) && 390ccb26b5aSJan Kara (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)) 391ccb26b5aSJan Kara inode->dirtied_when = jiffies; 392ccb26b5aSJan Kara 3934f8ad655SJan Kara if (wbc->pages_skipped) { 3944f8ad655SJan Kara /* 3954f8ad655SJan Kara * writeback is not making progress due to locked 3964f8ad655SJan Kara * buffers. Skip this inode for now. 3974f8ad655SJan Kara */ 3984f8ad655SJan Kara redirty_tail(inode, wb); 3994f8ad655SJan Kara return; 4004f8ad655SJan Kara } 4014f8ad655SJan Kara 402ccb26b5aSJan Kara if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) { 403ccb26b5aSJan Kara /* 404ccb26b5aSJan Kara * We didn't write back all the pages. nfs_writepages() 405ccb26b5aSJan Kara * sometimes bales out without doing anything. 406ccb26b5aSJan Kara */ 407ccb26b5aSJan Kara if (wbc->nr_to_write <= 0) { 408ccb26b5aSJan Kara /* Slice used up. Queue for next turn. */ 409ccb26b5aSJan Kara requeue_io(inode, wb); 410ccb26b5aSJan Kara } else { 411ccb26b5aSJan Kara /* 412ccb26b5aSJan Kara * Writeback blocked by something other than 413ccb26b5aSJan Kara * congestion. Delay the inode for some time to 414ccb26b5aSJan Kara * avoid spinning on the CPU (100% iowait) 415ccb26b5aSJan Kara * retrying writeback of the dirty page/inode 416ccb26b5aSJan Kara * that cannot be performed immediately. 417ccb26b5aSJan Kara */ 418ccb26b5aSJan Kara redirty_tail(inode, wb); 419ccb26b5aSJan Kara } 420ccb26b5aSJan Kara } else if (inode->i_state & I_DIRTY) { 421ccb26b5aSJan Kara /* 422ccb26b5aSJan Kara * Filesystems can dirty the inode during writeback operations, 423ccb26b5aSJan Kara * such as delayed allocation during submission or metadata 424ccb26b5aSJan Kara * updates after data IO completion. 425ccb26b5aSJan Kara */ 426ccb26b5aSJan Kara redirty_tail(inode, wb); 427ccb26b5aSJan Kara } else { 428ccb26b5aSJan Kara /* The inode is clean. Remove from writeback lists. */ 429ccb26b5aSJan Kara list_del_init(&inode->i_wb_list); 430ccb26b5aSJan Kara } 431ccb26b5aSJan Kara } 432ccb26b5aSJan Kara 433ccb26b5aSJan Kara /* 4344f8ad655SJan Kara * Write out an inode and its dirty pages. Do not update the writeback list 4354f8ad655SJan Kara * linkage. That is left to the caller. The caller is also responsible for 4364f8ad655SJan Kara * setting I_SYNC flag and calling inode_sync_complete() to clear it. 4371da177e4SLinus Torvalds */ 4381da177e4SLinus Torvalds static int 439cd8ed2a4SYan Hong __writeback_single_inode(struct inode *inode, struct writeback_control *wbc) 4401da177e4SLinus Torvalds { 4411da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 442251d6a47SWu Fengguang long nr_to_write = wbc->nr_to_write; 44301c03194SChristoph Hellwig unsigned dirty; 4441da177e4SLinus Torvalds int ret; 4451da177e4SLinus Torvalds 4464f8ad655SJan Kara WARN_ON(!(inode->i_state & I_SYNC)); 4471da177e4SLinus Torvalds 4489fb0a7daSTejun Heo trace_writeback_single_inode_start(inode, wbc, nr_to_write); 4499fb0a7daSTejun Heo 4501da177e4SLinus Torvalds ret = do_writepages(mapping, wbc); 4511da177e4SLinus Torvalds 45226821ed4SChristoph Hellwig /* 45326821ed4SChristoph Hellwig * Make sure to wait on the data before writing out the metadata. 45426821ed4SChristoph Hellwig * This is important for filesystems that modify metadata on data 4557747bd4bSDave Chinner * I/O completion. We don't do it for sync(2) writeback because it has a 4567747bd4bSDave Chinner * separate, external IO completion path and ->sync_fs for guaranteeing 4577747bd4bSDave Chinner * inode metadata is written back correctly. 45826821ed4SChristoph Hellwig */ 4597747bd4bSDave Chinner if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) { 46026821ed4SChristoph Hellwig int err = filemap_fdatawait(mapping); 4611da177e4SLinus Torvalds if (ret == 0) 4621da177e4SLinus Torvalds ret = err; 4631da177e4SLinus Torvalds } 4641da177e4SLinus Torvalds 4655547e8aaSDmitry Monakhov /* 4665547e8aaSDmitry Monakhov * Some filesystems may redirty the inode during the writeback 4675547e8aaSDmitry Monakhov * due to delalloc, clear dirty metadata flags right before 4685547e8aaSDmitry Monakhov * write_inode() 4695547e8aaSDmitry Monakhov */ 470250df6edSDave Chinner spin_lock(&inode->i_lock); 4716290be1cSJan Kara /* Clear I_DIRTY_PAGES if we've written out all dirty pages */ 4726290be1cSJan Kara if (!mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) 4736290be1cSJan Kara inode->i_state &= ~I_DIRTY_PAGES; 4745547e8aaSDmitry Monakhov dirty = inode->i_state & I_DIRTY; 4755547e8aaSDmitry Monakhov inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC); 476250df6edSDave Chinner spin_unlock(&inode->i_lock); 47726821ed4SChristoph Hellwig /* Don't write the inode if only I_DIRTY_PAGES was set */ 47826821ed4SChristoph Hellwig if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { 479a9185b41SChristoph Hellwig int err = write_inode(inode, wbc); 4801da177e4SLinus Torvalds if (ret == 0) 4811da177e4SLinus Torvalds ret = err; 4821da177e4SLinus Torvalds } 4834f8ad655SJan Kara trace_writeback_single_inode(inode, wbc, nr_to_write); 4844f8ad655SJan Kara return ret; 4854f8ad655SJan Kara } 4864f8ad655SJan Kara 4874f8ad655SJan Kara /* 4884f8ad655SJan Kara * Write out an inode's dirty pages. Either the caller has an active reference 4894f8ad655SJan Kara * on the inode or the inode has I_WILL_FREE set. 4904f8ad655SJan Kara * 4914f8ad655SJan Kara * This function is designed to be called for writing back one inode which 4924f8ad655SJan Kara * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode() 4934f8ad655SJan Kara * and does more profound writeback list handling in writeback_sb_inodes(). 4944f8ad655SJan Kara */ 4954f8ad655SJan Kara static int 4964f8ad655SJan Kara writeback_single_inode(struct inode *inode, struct bdi_writeback *wb, 4974f8ad655SJan Kara struct writeback_control *wbc) 4984f8ad655SJan Kara { 4994f8ad655SJan Kara int ret = 0; 5004f8ad655SJan Kara 5014f8ad655SJan Kara spin_lock(&inode->i_lock); 5024f8ad655SJan Kara if (!atomic_read(&inode->i_count)) 5034f8ad655SJan Kara WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); 5044f8ad655SJan Kara else 5054f8ad655SJan Kara WARN_ON(inode->i_state & I_WILL_FREE); 5064f8ad655SJan Kara 5074f8ad655SJan Kara if (inode->i_state & I_SYNC) { 5084f8ad655SJan Kara if (wbc->sync_mode != WB_SYNC_ALL) 5094f8ad655SJan Kara goto out; 5104f8ad655SJan Kara /* 511169ebd90SJan Kara * It's a data-integrity sync. We must wait. Since callers hold 512169ebd90SJan Kara * inode reference or inode has I_WILL_FREE set, it cannot go 513169ebd90SJan Kara * away under us. 5144f8ad655SJan Kara */ 515169ebd90SJan Kara __inode_wait_for_writeback(inode); 5164f8ad655SJan Kara } 5174f8ad655SJan Kara WARN_ON(inode->i_state & I_SYNC); 5184f8ad655SJan Kara /* 519f9b0e058SJan Kara * Skip inode if it is clean and we have no outstanding writeback in 520f9b0e058SJan Kara * WB_SYNC_ALL mode. We don't want to mess with writeback lists in this 521f9b0e058SJan Kara * function since flusher thread may be doing for example sync in 522f9b0e058SJan Kara * parallel and if we move the inode, it could get skipped. So here we 523f9b0e058SJan Kara * make sure inode is on some writeback list and leave it there unless 524f9b0e058SJan Kara * we have completely cleaned the inode. 5254f8ad655SJan Kara */ 526f9b0e058SJan Kara if (!(inode->i_state & I_DIRTY) && 527f9b0e058SJan Kara (wbc->sync_mode != WB_SYNC_ALL || 528f9b0e058SJan Kara !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK))) 5294f8ad655SJan Kara goto out; 5304f8ad655SJan Kara inode->i_state |= I_SYNC; 5314f8ad655SJan Kara spin_unlock(&inode->i_lock); 5324f8ad655SJan Kara 533cd8ed2a4SYan Hong ret = __writeback_single_inode(inode, wbc); 5341da177e4SLinus Torvalds 535f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 536250df6edSDave Chinner spin_lock(&inode->i_lock); 5374f8ad655SJan Kara /* 5384f8ad655SJan Kara * If inode is clean, remove it from writeback lists. Otherwise don't 5394f8ad655SJan Kara * touch it. See comment above for explanation. 5404f8ad655SJan Kara */ 5414f8ad655SJan Kara if (!(inode->i_state & I_DIRTY)) 5424f8ad655SJan Kara list_del_init(&inode->i_wb_list); 5434f8ad655SJan Kara spin_unlock(&wb->list_lock); 5441c0eeaf5SJoern Engel inode_sync_complete(inode); 5454f8ad655SJan Kara out: 5464f8ad655SJan Kara spin_unlock(&inode->i_lock); 5471da177e4SLinus Torvalds return ret; 5481da177e4SLinus Torvalds } 5491da177e4SLinus Torvalds 5501a12d8bdSWu Fengguang static long writeback_chunk_size(struct backing_dev_info *bdi, 5511a12d8bdSWu Fengguang struct wb_writeback_work *work) 552d46db3d5SWu Fengguang { 553d46db3d5SWu Fengguang long pages; 554d46db3d5SWu Fengguang 555d46db3d5SWu Fengguang /* 556d46db3d5SWu Fengguang * WB_SYNC_ALL mode does livelock avoidance by syncing dirty 557d46db3d5SWu Fengguang * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX 558d46db3d5SWu Fengguang * here avoids calling into writeback_inodes_wb() more than once. 559d46db3d5SWu Fengguang * 560d46db3d5SWu Fengguang * The intended call sequence for WB_SYNC_ALL writeback is: 561d46db3d5SWu Fengguang * 562d46db3d5SWu Fengguang * wb_writeback() 563d46db3d5SWu Fengguang * writeback_sb_inodes() <== called only once 564d46db3d5SWu Fengguang * write_cache_pages() <== called once for each inode 565d46db3d5SWu Fengguang * (quickly) tag currently dirty pages 566d46db3d5SWu Fengguang * (maybe slowly) sync all tagged pages 567d46db3d5SWu Fengguang */ 568d46db3d5SWu Fengguang if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages) 569d46db3d5SWu Fengguang pages = LONG_MAX; 5701a12d8bdSWu Fengguang else { 5711a12d8bdSWu Fengguang pages = min(bdi->avg_write_bandwidth / 2, 5721a12d8bdSWu Fengguang global_dirty_limit / DIRTY_SCOPE); 5731a12d8bdSWu Fengguang pages = min(pages, work->nr_pages); 5741a12d8bdSWu Fengguang pages = round_down(pages + MIN_WRITEBACK_PAGES, 5751a12d8bdSWu Fengguang MIN_WRITEBACK_PAGES); 5761a12d8bdSWu Fengguang } 577d46db3d5SWu Fengguang 578d46db3d5SWu Fengguang return pages; 579d46db3d5SWu Fengguang } 580d46db3d5SWu Fengguang 58103ba3782SJens Axboe /* 582f11c9c5cSEdward Shishkin * Write a portion of b_io inodes which belong to @sb. 583edadfb10SChristoph Hellwig * 584d46db3d5SWu Fengguang * Return the number of pages and/or inodes written. 585f11c9c5cSEdward Shishkin */ 586d46db3d5SWu Fengguang static long writeback_sb_inodes(struct super_block *sb, 587d46db3d5SWu Fengguang struct bdi_writeback *wb, 588d46db3d5SWu Fengguang struct wb_writeback_work *work) 58903ba3782SJens Axboe { 590d46db3d5SWu Fengguang struct writeback_control wbc = { 591d46db3d5SWu Fengguang .sync_mode = work->sync_mode, 592d46db3d5SWu Fengguang .tagged_writepages = work->tagged_writepages, 593d46db3d5SWu Fengguang .for_kupdate = work->for_kupdate, 594d46db3d5SWu Fengguang .for_background = work->for_background, 5957747bd4bSDave Chinner .for_sync = work->for_sync, 596d46db3d5SWu Fengguang .range_cyclic = work->range_cyclic, 597d46db3d5SWu Fengguang .range_start = 0, 598d46db3d5SWu Fengguang .range_end = LLONG_MAX, 599d46db3d5SWu Fengguang }; 600d46db3d5SWu Fengguang unsigned long start_time = jiffies; 601d46db3d5SWu Fengguang long write_chunk; 602d46db3d5SWu Fengguang long wrote = 0; /* count both pages and inodes */ 603d46db3d5SWu Fengguang 60403ba3782SJens Axboe while (!list_empty(&wb->b_io)) { 6057ccf19a8SNick Piggin struct inode *inode = wb_inode(wb->b_io.prev); 606edadfb10SChristoph Hellwig 607edadfb10SChristoph Hellwig if (inode->i_sb != sb) { 608d46db3d5SWu Fengguang if (work->sb) { 609edadfb10SChristoph Hellwig /* 610edadfb10SChristoph Hellwig * We only want to write back data for this 611edadfb10SChristoph Hellwig * superblock, move all inodes not belonging 612edadfb10SChristoph Hellwig * to it back onto the dirty list. 613edadfb10SChristoph Hellwig */ 614f758eeabSChristoph Hellwig redirty_tail(inode, wb); 61566f3b8e2SJens Axboe continue; 61666f3b8e2SJens Axboe } 617edadfb10SChristoph Hellwig 618edadfb10SChristoph Hellwig /* 619edadfb10SChristoph Hellwig * The inode belongs to a different superblock. 620edadfb10SChristoph Hellwig * Bounce back to the caller to unpin this and 621edadfb10SChristoph Hellwig * pin the next superblock. 622edadfb10SChristoph Hellwig */ 623d46db3d5SWu Fengguang break; 624edadfb10SChristoph Hellwig } 625edadfb10SChristoph Hellwig 6269843b76aSChristoph Hellwig /* 627331cbdeeSWanpeng Li * Don't bother with new inodes or inodes being freed, first 628331cbdeeSWanpeng Li * kind does not need periodic writeout yet, and for the latter 6299843b76aSChristoph Hellwig * kind writeout is handled by the freer. 6309843b76aSChristoph Hellwig */ 631250df6edSDave Chinner spin_lock(&inode->i_lock); 6329843b76aSChristoph Hellwig if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { 633250df6edSDave Chinner spin_unlock(&inode->i_lock); 634fcc5c222SWu Fengguang redirty_tail(inode, wb); 6357ef0d737SNick Piggin continue; 6367ef0d737SNick Piggin } 637cc1676d9SJan Kara if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) { 638cc1676d9SJan Kara /* 639cc1676d9SJan Kara * If this inode is locked for writeback and we are not 640cc1676d9SJan Kara * doing writeback-for-data-integrity, move it to 641cc1676d9SJan Kara * b_more_io so that writeback can proceed with the 642cc1676d9SJan Kara * other inodes on s_io. 643cc1676d9SJan Kara * 644cc1676d9SJan Kara * We'll have another go at writing back this inode 645cc1676d9SJan Kara * when we completed a full scan of b_io. 646cc1676d9SJan Kara */ 647cc1676d9SJan Kara spin_unlock(&inode->i_lock); 648cc1676d9SJan Kara requeue_io(inode, wb); 649cc1676d9SJan Kara trace_writeback_sb_inodes_requeue(inode); 650cc1676d9SJan Kara continue; 651cc1676d9SJan Kara } 652f0d07b7fSJan Kara spin_unlock(&wb->list_lock); 653f0d07b7fSJan Kara 6544f8ad655SJan Kara /* 6554f8ad655SJan Kara * We already requeued the inode if it had I_SYNC set and we 6564f8ad655SJan Kara * are doing WB_SYNC_NONE writeback. So this catches only the 6574f8ad655SJan Kara * WB_SYNC_ALL case. 6584f8ad655SJan Kara */ 659169ebd90SJan Kara if (inode->i_state & I_SYNC) { 660169ebd90SJan Kara /* Wait for I_SYNC. This function drops i_lock... */ 661169ebd90SJan Kara inode_sleep_on_writeback(inode); 662169ebd90SJan Kara /* Inode may be gone, start again */ 663ead188f9SJan Kara spin_lock(&wb->list_lock); 664169ebd90SJan Kara continue; 665169ebd90SJan Kara } 6664f8ad655SJan Kara inode->i_state |= I_SYNC; 6674f8ad655SJan Kara spin_unlock(&inode->i_lock); 668169ebd90SJan Kara 6691a12d8bdSWu Fengguang write_chunk = writeback_chunk_size(wb->bdi, work); 670d46db3d5SWu Fengguang wbc.nr_to_write = write_chunk; 671d46db3d5SWu Fengguang wbc.pages_skipped = 0; 672250df6edSDave Chinner 673169ebd90SJan Kara /* 674169ebd90SJan Kara * We use I_SYNC to pin the inode in memory. While it is set 675169ebd90SJan Kara * evict_inode() will wait so the inode cannot be freed. 676169ebd90SJan Kara */ 677cd8ed2a4SYan Hong __writeback_single_inode(inode, &wbc); 678d46db3d5SWu Fengguang 679d46db3d5SWu Fengguang work->nr_pages -= write_chunk - wbc.nr_to_write; 680d46db3d5SWu Fengguang wrote += write_chunk - wbc.nr_to_write; 6814f8ad655SJan Kara spin_lock(&wb->list_lock); 6824f8ad655SJan Kara spin_lock(&inode->i_lock); 683d46db3d5SWu Fengguang if (!(inode->i_state & I_DIRTY)) 684d46db3d5SWu Fengguang wrote++; 6854f8ad655SJan Kara requeue_inode(inode, wb, &wbc); 6864f8ad655SJan Kara inode_sync_complete(inode); 6870f1b1fd8SDave Chinner spin_unlock(&inode->i_lock); 688169ebd90SJan Kara cond_resched_lock(&wb->list_lock); 689d46db3d5SWu Fengguang /* 690d46db3d5SWu Fengguang * bail out to wb_writeback() often enough to check 691d46db3d5SWu Fengguang * background threshold and other termination conditions. 692d46db3d5SWu Fengguang */ 693d46db3d5SWu Fengguang if (wrote) { 694d46db3d5SWu Fengguang if (time_is_before_jiffies(start_time + HZ / 10UL)) 695d46db3d5SWu Fengguang break; 696d46db3d5SWu Fengguang if (work->nr_pages <= 0) 697d46db3d5SWu Fengguang break; 6981da177e4SLinus Torvalds } 6998bc3be27SFengguang Wu } 700d46db3d5SWu Fengguang return wrote; 701f11c9c5cSEdward Shishkin } 70238f21977SNick Piggin 703d46db3d5SWu Fengguang static long __writeback_inodes_wb(struct bdi_writeback *wb, 704d46db3d5SWu Fengguang struct wb_writeback_work *work) 705f11c9c5cSEdward Shishkin { 706d46db3d5SWu Fengguang unsigned long start_time = jiffies; 707d46db3d5SWu Fengguang long wrote = 0; 708f11c9c5cSEdward Shishkin 709f11c9c5cSEdward Shishkin while (!list_empty(&wb->b_io)) { 7107ccf19a8SNick Piggin struct inode *inode = wb_inode(wb->b_io.prev); 711f11c9c5cSEdward Shishkin struct super_block *sb = inode->i_sb; 712f11c9c5cSEdward Shishkin 71312ad3ab6SDave Chinner if (!grab_super_passive(sb)) { 7140e995816SWu Fengguang /* 7150e995816SWu Fengguang * grab_super_passive() may fail consistently due to 7160e995816SWu Fengguang * s_umount being grabbed by someone else. Don't use 7170e995816SWu Fengguang * requeue_io() to avoid busy retrying the inode/sb. 7180e995816SWu Fengguang */ 7190e995816SWu Fengguang redirty_tail(inode, wb); 720d19de7edSChristoph Hellwig continue; 721334132aeSChristoph Hellwig } 722d46db3d5SWu Fengguang wrote += writeback_sb_inodes(sb, wb, work); 723d19de7edSChristoph Hellwig drop_super(sb); 724f11c9c5cSEdward Shishkin 725d46db3d5SWu Fengguang /* refer to the same tests at the end of writeback_sb_inodes */ 726d46db3d5SWu Fengguang if (wrote) { 727d46db3d5SWu Fengguang if (time_is_before_jiffies(start_time + HZ / 10UL)) 728d46db3d5SWu Fengguang break; 729d46db3d5SWu Fengguang if (work->nr_pages <= 0) 730f11c9c5cSEdward Shishkin break; 731f11c9c5cSEdward Shishkin } 732d46db3d5SWu Fengguang } 73366f3b8e2SJens Axboe /* Leave any unwritten inodes on b_io */ 734d46db3d5SWu Fengguang return wrote; 73566f3b8e2SJens Axboe } 73666f3b8e2SJens Axboe 7377d9f073bSWanpeng Li static long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages, 7380e175a18SCurt Wohlgemuth enum wb_reason reason) 739edadfb10SChristoph Hellwig { 740d46db3d5SWu Fengguang struct wb_writeback_work work = { 741d46db3d5SWu Fengguang .nr_pages = nr_pages, 742d46db3d5SWu Fengguang .sync_mode = WB_SYNC_NONE, 743d46db3d5SWu Fengguang .range_cyclic = 1, 7440e175a18SCurt Wohlgemuth .reason = reason, 745c4a391b5SJan Kara .older_than_this = jiffies, 746c4a391b5SJan Kara .older_than_this_is_set = 1, 747d46db3d5SWu Fengguang }; 748edadfb10SChristoph Hellwig 749f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 750424b351fSWu Fengguang if (list_empty(&wb->b_io)) 751ad4e38ddSCurt Wohlgemuth queue_io(wb, &work); 752d46db3d5SWu Fengguang __writeback_inodes_wb(wb, &work); 753f758eeabSChristoph Hellwig spin_unlock(&wb->list_lock); 754edadfb10SChristoph Hellwig 755d46db3d5SWu Fengguang return nr_pages - work.nr_pages; 75666f3b8e2SJens Axboe } 75766f3b8e2SJens Axboe 758b00949aaSWu Fengguang static bool over_bground_thresh(struct backing_dev_info *bdi) 75903ba3782SJens Axboe { 76003ba3782SJens Axboe unsigned long background_thresh, dirty_thresh; 76103ba3782SJens Axboe 76216c4042fSWu Fengguang global_dirty_limits(&background_thresh, &dirty_thresh); 76303ba3782SJens Axboe 764b00949aaSWu Fengguang if (global_page_state(NR_FILE_DIRTY) + 765b00949aaSWu Fengguang global_page_state(NR_UNSTABLE_NFS) > background_thresh) 766b00949aaSWu Fengguang return true; 767b00949aaSWu Fengguang 768b00949aaSWu Fengguang if (bdi_stat(bdi, BDI_RECLAIMABLE) > 769b00949aaSWu Fengguang bdi_dirty_limit(bdi, background_thresh)) 770b00949aaSWu Fengguang return true; 771b00949aaSWu Fengguang 772b00949aaSWu Fengguang return false; 77303ba3782SJens Axboe } 77403ba3782SJens Axboe 77503ba3782SJens Axboe /* 776e98be2d5SWu Fengguang * Called under wb->list_lock. If there are multiple wb per bdi, 777e98be2d5SWu Fengguang * only the flusher working on the first wb should do it. 778e98be2d5SWu Fengguang */ 779e98be2d5SWu Fengguang static void wb_update_bandwidth(struct bdi_writeback *wb, 780e98be2d5SWu Fengguang unsigned long start_time) 781e98be2d5SWu Fengguang { 782af6a3113SWu Fengguang __bdi_update_bandwidth(wb->bdi, 0, 0, 0, 0, 0, start_time); 783e98be2d5SWu Fengguang } 784e98be2d5SWu Fengguang 785e98be2d5SWu Fengguang /* 78603ba3782SJens Axboe * Explicit flushing or periodic writeback of "old" data. 78703ba3782SJens Axboe * 78803ba3782SJens Axboe * Define "old": the first time one of an inode's pages is dirtied, we mark the 78903ba3782SJens Axboe * dirtying-time in the inode's address_space. So this periodic writeback code 79003ba3782SJens Axboe * just walks the superblock inode list, writing back any inodes which are 79103ba3782SJens Axboe * older than a specific point in time. 79203ba3782SJens Axboe * 79303ba3782SJens Axboe * Try to run once per dirty_writeback_interval. But if a writeback event 79403ba3782SJens Axboe * takes longer than a dirty_writeback_interval interval, then leave a 79503ba3782SJens Axboe * one-second gap. 79603ba3782SJens Axboe * 79703ba3782SJens Axboe * older_than_this takes precedence over nr_to_write. So we'll only write back 79803ba3782SJens Axboe * all dirty pages if they are all attached to "old" mappings. 79903ba3782SJens Axboe */ 800c4a77a6cSJens Axboe static long wb_writeback(struct bdi_writeback *wb, 80183ba7b07SChristoph Hellwig struct wb_writeback_work *work) 80203ba3782SJens Axboe { 803e98be2d5SWu Fengguang unsigned long wb_start = jiffies; 804d46db3d5SWu Fengguang long nr_pages = work->nr_pages; 805a5989bdcSJan Kara struct inode *inode; 806d46db3d5SWu Fengguang long progress; 80703ba3782SJens Axboe 808c4a391b5SJan Kara if (!work->older_than_this_is_set) { 809c4a391b5SJan Kara work->older_than_this = jiffies; 810c4a391b5SJan Kara work->older_than_this_is_set = 1; 811c4a391b5SJan Kara } 81203ba3782SJens Axboe 813e8dfc305SWu Fengguang spin_lock(&wb->list_lock); 81403ba3782SJens Axboe for (;;) { 81503ba3782SJens Axboe /* 816d3ddec76SWu Fengguang * Stop writeback when nr_pages has been consumed 81703ba3782SJens Axboe */ 81883ba7b07SChristoph Hellwig if (work->nr_pages <= 0) 81903ba3782SJens Axboe break; 82003ba3782SJens Axboe 82103ba3782SJens Axboe /* 822aa373cf5SJan Kara * Background writeout and kupdate-style writeback may 823aa373cf5SJan Kara * run forever. Stop them if there is other work to do 824aa373cf5SJan Kara * so that e.g. sync can proceed. They'll be restarted 825aa373cf5SJan Kara * after the other works are all done. 826aa373cf5SJan Kara */ 827aa373cf5SJan Kara if ((work->for_background || work->for_kupdate) && 828aa373cf5SJan Kara !list_empty(&wb->bdi->work_list)) 829aa373cf5SJan Kara break; 830aa373cf5SJan Kara 831aa373cf5SJan Kara /* 832d3ddec76SWu Fengguang * For background writeout, stop when we are below the 833d3ddec76SWu Fengguang * background dirty threshold 83403ba3782SJens Axboe */ 835b00949aaSWu Fengguang if (work->for_background && !over_bground_thresh(wb->bdi)) 83603ba3782SJens Axboe break; 83703ba3782SJens Axboe 8381bc36b64SJan Kara /* 8391bc36b64SJan Kara * Kupdate and background works are special and we want to 8401bc36b64SJan Kara * include all inodes that need writing. Livelock avoidance is 8411bc36b64SJan Kara * handled by these works yielding to any other work so we are 8421bc36b64SJan Kara * safe. 8431bc36b64SJan Kara */ 844ba9aa839SWu Fengguang if (work->for_kupdate) { 845c4a391b5SJan Kara work->older_than_this = jiffies - 846ba9aa839SWu Fengguang msecs_to_jiffies(dirty_expire_interval * 10); 8471bc36b64SJan Kara } else if (work->for_background) 848c4a391b5SJan Kara work->older_than_this = jiffies; 849028c2dd1SDave Chinner 850d46db3d5SWu Fengguang trace_writeback_start(wb->bdi, work); 851e8dfc305SWu Fengguang if (list_empty(&wb->b_io)) 852ad4e38ddSCurt Wohlgemuth queue_io(wb, work); 85383ba7b07SChristoph Hellwig if (work->sb) 854d46db3d5SWu Fengguang progress = writeback_sb_inodes(work->sb, wb, work); 855edadfb10SChristoph Hellwig else 856d46db3d5SWu Fengguang progress = __writeback_inodes_wb(wb, work); 857d46db3d5SWu Fengguang trace_writeback_written(wb->bdi, work); 858028c2dd1SDave Chinner 859e98be2d5SWu Fengguang wb_update_bandwidth(wb, wb_start); 86003ba3782SJens Axboe 86103ba3782SJens Axboe /* 86271fd05a8SJens Axboe * Did we write something? Try for more 863e6fb6da2SWu Fengguang * 864e6fb6da2SWu Fengguang * Dirty inodes are moved to b_io for writeback in batches. 865e6fb6da2SWu Fengguang * The completion of the current batch does not necessarily 866e6fb6da2SWu Fengguang * mean the overall work is done. So we keep looping as long 867e6fb6da2SWu Fengguang * as made some progress on cleaning pages or inodes. 86871fd05a8SJens Axboe */ 869d46db3d5SWu Fengguang if (progress) 87003ba3782SJens Axboe continue; 871a5989bdcSJan Kara /* 872e6fb6da2SWu Fengguang * No more inodes for IO, bail 873a5989bdcSJan Kara */ 874b7a2441fSWu Fengguang if (list_empty(&wb->b_more_io)) 87503ba3782SJens Axboe break; 87603ba3782SJens Axboe /* 8778010c3b6SJens Axboe * Nothing written. Wait for some inode to 8788010c3b6SJens Axboe * become available for writeback. Otherwise 8798010c3b6SJens Axboe * we'll just busyloop. 88003ba3782SJens Axboe */ 88103ba3782SJens Axboe if (!list_empty(&wb->b_more_io)) { 882d46db3d5SWu Fengguang trace_writeback_wait(wb->bdi, work); 88303ba3782SJens Axboe inode = wb_inode(wb->b_more_io.prev); 884250df6edSDave Chinner spin_lock(&inode->i_lock); 885f0d07b7fSJan Kara spin_unlock(&wb->list_lock); 886169ebd90SJan Kara /* This function drops i_lock... */ 887169ebd90SJan Kara inode_sleep_on_writeback(inode); 888f0d07b7fSJan Kara spin_lock(&wb->list_lock); 88903ba3782SJens Axboe } 89003ba3782SJens Axboe } 891e8dfc305SWu Fengguang spin_unlock(&wb->list_lock); 89203ba3782SJens Axboe 893d46db3d5SWu Fengguang return nr_pages - work->nr_pages; 89403ba3782SJens Axboe } 89503ba3782SJens Axboe 89603ba3782SJens Axboe /* 89783ba7b07SChristoph Hellwig * Return the next wb_writeback_work struct that hasn't been processed yet. 89803ba3782SJens Axboe */ 89983ba7b07SChristoph Hellwig static struct wb_writeback_work * 90008852b6dSMinchan Kim get_next_work_item(struct backing_dev_info *bdi) 90103ba3782SJens Axboe { 90283ba7b07SChristoph Hellwig struct wb_writeback_work *work = NULL; 90303ba3782SJens Axboe 9046467716aSArtem Bityutskiy spin_lock_bh(&bdi->wb_lock); 90583ba7b07SChristoph Hellwig if (!list_empty(&bdi->work_list)) { 90683ba7b07SChristoph Hellwig work = list_entry(bdi->work_list.next, 90783ba7b07SChristoph Hellwig struct wb_writeback_work, list); 90883ba7b07SChristoph Hellwig list_del_init(&work->list); 90903ba3782SJens Axboe } 9106467716aSArtem Bityutskiy spin_unlock_bh(&bdi->wb_lock); 91183ba7b07SChristoph Hellwig return work; 91203ba3782SJens Axboe } 91303ba3782SJens Axboe 914cdf01dd5SLinus Torvalds /* 915cdf01dd5SLinus Torvalds * Add in the number of potentially dirty inodes, because each inode 916cdf01dd5SLinus Torvalds * write can dirty pagecache in the underlying blockdev. 917cdf01dd5SLinus Torvalds */ 918cdf01dd5SLinus Torvalds static unsigned long get_nr_dirty_pages(void) 919cdf01dd5SLinus Torvalds { 920cdf01dd5SLinus Torvalds return global_page_state(NR_FILE_DIRTY) + 921cdf01dd5SLinus Torvalds global_page_state(NR_UNSTABLE_NFS) + 922cdf01dd5SLinus Torvalds get_nr_dirty_inodes(); 923cdf01dd5SLinus Torvalds } 924cdf01dd5SLinus Torvalds 9256585027aSJan Kara static long wb_check_background_flush(struct bdi_writeback *wb) 9266585027aSJan Kara { 927b00949aaSWu Fengguang if (over_bground_thresh(wb->bdi)) { 9286585027aSJan Kara 9296585027aSJan Kara struct wb_writeback_work work = { 9306585027aSJan Kara .nr_pages = LONG_MAX, 9316585027aSJan Kara .sync_mode = WB_SYNC_NONE, 9326585027aSJan Kara .for_background = 1, 9336585027aSJan Kara .range_cyclic = 1, 9340e175a18SCurt Wohlgemuth .reason = WB_REASON_BACKGROUND, 9356585027aSJan Kara }; 9366585027aSJan Kara 9376585027aSJan Kara return wb_writeback(wb, &work); 9386585027aSJan Kara } 9396585027aSJan Kara 9406585027aSJan Kara return 0; 9416585027aSJan Kara } 9426585027aSJan Kara 94303ba3782SJens Axboe static long wb_check_old_data_flush(struct bdi_writeback *wb) 94403ba3782SJens Axboe { 94503ba3782SJens Axboe unsigned long expired; 94603ba3782SJens Axboe long nr_pages; 94703ba3782SJens Axboe 94869b62d01SJens Axboe /* 94969b62d01SJens Axboe * When set to zero, disable periodic writeback 95069b62d01SJens Axboe */ 95169b62d01SJens Axboe if (!dirty_writeback_interval) 95269b62d01SJens Axboe return 0; 95369b62d01SJens Axboe 95403ba3782SJens Axboe expired = wb->last_old_flush + 95503ba3782SJens Axboe msecs_to_jiffies(dirty_writeback_interval * 10); 95603ba3782SJens Axboe if (time_before(jiffies, expired)) 95703ba3782SJens Axboe return 0; 95803ba3782SJens Axboe 95903ba3782SJens Axboe wb->last_old_flush = jiffies; 960cdf01dd5SLinus Torvalds nr_pages = get_nr_dirty_pages(); 96103ba3782SJens Axboe 962c4a77a6cSJens Axboe if (nr_pages) { 96383ba7b07SChristoph Hellwig struct wb_writeback_work work = { 964c4a77a6cSJens Axboe .nr_pages = nr_pages, 965c4a77a6cSJens Axboe .sync_mode = WB_SYNC_NONE, 966c4a77a6cSJens Axboe .for_kupdate = 1, 967c4a77a6cSJens Axboe .range_cyclic = 1, 9680e175a18SCurt Wohlgemuth .reason = WB_REASON_PERIODIC, 969c4a77a6cSJens Axboe }; 970c4a77a6cSJens Axboe 97183ba7b07SChristoph Hellwig return wb_writeback(wb, &work); 972c4a77a6cSJens Axboe } 97303ba3782SJens Axboe 97403ba3782SJens Axboe return 0; 97503ba3782SJens Axboe } 97603ba3782SJens Axboe 97703ba3782SJens Axboe /* 97803ba3782SJens Axboe * Retrieve work items and do the writeback they describe 97903ba3782SJens Axboe */ 98025d130baSWanpeng Li static long wb_do_writeback(struct bdi_writeback *wb) 98103ba3782SJens Axboe { 98203ba3782SJens Axboe struct backing_dev_info *bdi = wb->bdi; 98383ba7b07SChristoph Hellwig struct wb_writeback_work *work; 984c4a77a6cSJens Axboe long wrote = 0; 98503ba3782SJens Axboe 98681d73a32SJan Kara set_bit(BDI_writeback_running, &wb->bdi->state); 98708852b6dSMinchan Kim while ((work = get_next_work_item(bdi)) != NULL) { 98883ba7b07SChristoph Hellwig 989455b2864SDave Chinner trace_writeback_exec(bdi, work); 990455b2864SDave Chinner 99183ba7b07SChristoph Hellwig wrote += wb_writeback(wb, work); 99203ba3782SJens Axboe 99303ba3782SJens Axboe /* 99483ba7b07SChristoph Hellwig * Notify the caller of completion if this is a synchronous 99583ba7b07SChristoph Hellwig * work item, otherwise just free it. 99603ba3782SJens Axboe */ 99783ba7b07SChristoph Hellwig if (work->done) 99883ba7b07SChristoph Hellwig complete(work->done); 99983ba7b07SChristoph Hellwig else 100083ba7b07SChristoph Hellwig kfree(work); 100103ba3782SJens Axboe } 100203ba3782SJens Axboe 100303ba3782SJens Axboe /* 100403ba3782SJens Axboe * Check for periodic writeback, kupdated() style 100503ba3782SJens Axboe */ 100603ba3782SJens Axboe wrote += wb_check_old_data_flush(wb); 10076585027aSJan Kara wrote += wb_check_background_flush(wb); 100881d73a32SJan Kara clear_bit(BDI_writeback_running, &wb->bdi->state); 100903ba3782SJens Axboe 101003ba3782SJens Axboe return wrote; 101103ba3782SJens Axboe } 101203ba3782SJens Axboe 101303ba3782SJens Axboe /* 101403ba3782SJens Axboe * Handle writeback of dirty data for the device backed by this bdi. Also 1015839a8e86STejun Heo * reschedules periodically and does kupdated style flushing. 101603ba3782SJens Axboe */ 1017839a8e86STejun Heo void bdi_writeback_workfn(struct work_struct *work) 101803ba3782SJens Axboe { 1019839a8e86STejun Heo struct bdi_writeback *wb = container_of(to_delayed_work(work), 1020839a8e86STejun Heo struct bdi_writeback, dwork); 102108243900SChristoph Hellwig struct backing_dev_info *bdi = wb->bdi; 102203ba3782SJens Axboe long pages_written; 102303ba3782SJens Axboe 1024ef3b1019STejun Heo set_worker_desc("flush-%s", dev_name(bdi->dev)); 1025766f9164SPeter Zijlstra current->flags |= PF_SWAPWRITE; 102603ba3782SJens Axboe 1027839a8e86STejun Heo if (likely(!current_is_workqueue_rescuer() || 1028839a8e86STejun Heo list_empty(&bdi->bdi_list))) { 102903ba3782SJens Axboe /* 1030839a8e86STejun Heo * The normal path. Keep writing back @bdi until its 1031839a8e86STejun Heo * work_list is empty. Note that this path is also taken 1032839a8e86STejun Heo * if @bdi is shutting down even when we're running off the 1033839a8e86STejun Heo * rescuer as work_list needs to be drained. 103403ba3782SJens Axboe */ 1035839a8e86STejun Heo do { 103625d130baSWanpeng Li pages_written = wb_do_writeback(wb); 1037455b2864SDave Chinner trace_writeback_pages_written(pages_written); 1038839a8e86STejun Heo } while (!list_empty(&bdi->work_list)); 1039839a8e86STejun Heo } else { 1040253c34e9SArtem Bityutskiy /* 1041839a8e86STejun Heo * bdi_wq can't get enough workers and we're running off 1042839a8e86STejun Heo * the emergency worker. Don't hog it. Hopefully, 1024 is 1043839a8e86STejun Heo * enough for efficient IO. 1044253c34e9SArtem Bityutskiy */ 1045839a8e86STejun Heo pages_written = writeback_inodes_wb(&bdi->wb, 1024, 1046839a8e86STejun Heo WB_REASON_FORKER_THREAD); 1047839a8e86STejun Heo trace_writeback_pages_written(pages_written); 104803ba3782SJens Axboe } 104903ba3782SJens Axboe 1050839a8e86STejun Heo if (!list_empty(&bdi->work_list) || 1051839a8e86STejun Heo (wb_has_dirty_io(wb) && dirty_writeback_interval)) 1052839a8e86STejun Heo queue_delayed_work(bdi_wq, &wb->dwork, 1053839a8e86STejun Heo msecs_to_jiffies(dirty_writeback_interval * 10)); 1054455b2864SDave Chinner 1055839a8e86STejun Heo current->flags &= ~PF_SWAPWRITE; 105603ba3782SJens Axboe } 105703ba3782SJens Axboe 105803ba3782SJens Axboe /* 105903ba3782SJens Axboe * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back 106003ba3782SJens Axboe * the whole world. 106103ba3782SJens Axboe */ 10620e175a18SCurt Wohlgemuth void wakeup_flusher_threads(long nr_pages, enum wb_reason reason) 106303ba3782SJens Axboe { 1064b8c2f347SChristoph Hellwig struct backing_dev_info *bdi; 1065b8c2f347SChristoph Hellwig 106647df3ddeSJan Kara if (!nr_pages) 106747df3ddeSJan Kara nr_pages = get_nr_dirty_pages(); 1068b8c2f347SChristoph Hellwig 1069b8c2f347SChristoph Hellwig rcu_read_lock(); 1070b8c2f347SChristoph Hellwig list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { 1071b8c2f347SChristoph Hellwig if (!bdi_has_dirty_io(bdi)) 1072b8c2f347SChristoph Hellwig continue; 10730e175a18SCurt Wohlgemuth __bdi_start_writeback(bdi, nr_pages, false, reason); 1074b8c2f347SChristoph Hellwig } 1075b8c2f347SChristoph Hellwig rcu_read_unlock(); 107603ba3782SJens Axboe } 107703ba3782SJens Axboe 107803ba3782SJens Axboe static noinline void block_dump___mark_inode_dirty(struct inode *inode) 107903ba3782SJens Axboe { 108003ba3782SJens Axboe if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { 108103ba3782SJens Axboe struct dentry *dentry; 108203ba3782SJens Axboe const char *name = "?"; 108303ba3782SJens Axboe 108403ba3782SJens Axboe dentry = d_find_alias(inode); 108503ba3782SJens Axboe if (dentry) { 108603ba3782SJens Axboe spin_lock(&dentry->d_lock); 108703ba3782SJens Axboe name = (const char *) dentry->d_name.name; 108803ba3782SJens Axboe } 108903ba3782SJens Axboe printk(KERN_DEBUG 109003ba3782SJens Axboe "%s(%d): dirtied inode %lu (%s) on %s\n", 109103ba3782SJens Axboe current->comm, task_pid_nr(current), inode->i_ino, 109203ba3782SJens Axboe name, inode->i_sb->s_id); 109303ba3782SJens Axboe if (dentry) { 109403ba3782SJens Axboe spin_unlock(&dentry->d_lock); 109503ba3782SJens Axboe dput(dentry); 109603ba3782SJens Axboe } 109703ba3782SJens Axboe } 109803ba3782SJens Axboe } 109903ba3782SJens Axboe 110003ba3782SJens Axboe /** 110103ba3782SJens Axboe * __mark_inode_dirty - internal function 110203ba3782SJens Axboe * @inode: inode to mark 110303ba3782SJens Axboe * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) 110403ba3782SJens Axboe * Mark an inode as dirty. Callers should use mark_inode_dirty or 110503ba3782SJens Axboe * mark_inode_dirty_sync. 110603ba3782SJens Axboe * 110703ba3782SJens Axboe * Put the inode on the super block's dirty list. 110803ba3782SJens Axboe * 110903ba3782SJens Axboe * CAREFUL! We mark it dirty unconditionally, but move it onto the 111003ba3782SJens Axboe * dirty list only if it is hashed or if it refers to a blockdev. 111103ba3782SJens Axboe * If it was not hashed, it will never be added to the dirty list 111203ba3782SJens Axboe * even if it is later hashed, as it will have been marked dirty already. 111303ba3782SJens Axboe * 111403ba3782SJens Axboe * In short, make sure you hash any inodes _before_ you start marking 111503ba3782SJens Axboe * them dirty. 111603ba3782SJens Axboe * 111703ba3782SJens Axboe * Note that for blockdevs, inode->dirtied_when represents the dirtying time of 111803ba3782SJens Axboe * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of 111903ba3782SJens Axboe * the kernel-internal blockdev inode represents the dirtying time of the 112003ba3782SJens Axboe * blockdev's pages. This is why for I_DIRTY_PAGES we always use 112103ba3782SJens Axboe * page->mapping->host, so the page-dirtying time is recorded in the internal 112203ba3782SJens Axboe * blockdev inode. 112303ba3782SJens Axboe */ 112403ba3782SJens Axboe void __mark_inode_dirty(struct inode *inode, int flags) 112503ba3782SJens Axboe { 112603ba3782SJens Axboe struct super_block *sb = inode->i_sb; 1127253c34e9SArtem Bityutskiy struct backing_dev_info *bdi = NULL; 112803ba3782SJens Axboe 112903ba3782SJens Axboe /* 113003ba3782SJens Axboe * Don't do this for I_DIRTY_PAGES - that doesn't actually 113103ba3782SJens Axboe * dirty the inode itself 113203ba3782SJens Axboe */ 113303ba3782SJens Axboe if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { 11349fb0a7daSTejun Heo trace_writeback_dirty_inode_start(inode, flags); 11359fb0a7daSTejun Heo 113603ba3782SJens Axboe if (sb->s_op->dirty_inode) 1137aa385729SChristoph Hellwig sb->s_op->dirty_inode(inode, flags); 11389fb0a7daSTejun Heo 11399fb0a7daSTejun Heo trace_writeback_dirty_inode(inode, flags); 114003ba3782SJens Axboe } 114103ba3782SJens Axboe 114203ba3782SJens Axboe /* 114303ba3782SJens Axboe * make sure that changes are seen by all cpus before we test i_state 114403ba3782SJens Axboe * -- mikulas 114503ba3782SJens Axboe */ 114603ba3782SJens Axboe smp_mb(); 114703ba3782SJens Axboe 114803ba3782SJens Axboe /* avoid the locking if we can */ 114903ba3782SJens Axboe if ((inode->i_state & flags) == flags) 115003ba3782SJens Axboe return; 115103ba3782SJens Axboe 115203ba3782SJens Axboe if (unlikely(block_dump)) 115303ba3782SJens Axboe block_dump___mark_inode_dirty(inode); 115403ba3782SJens Axboe 1155250df6edSDave Chinner spin_lock(&inode->i_lock); 115603ba3782SJens Axboe if ((inode->i_state & flags) != flags) { 115703ba3782SJens Axboe const int was_dirty = inode->i_state & I_DIRTY; 115803ba3782SJens Axboe 115903ba3782SJens Axboe inode->i_state |= flags; 116003ba3782SJens Axboe 116103ba3782SJens Axboe /* 116203ba3782SJens Axboe * If the inode is being synced, just update its dirty state. 116303ba3782SJens Axboe * The unlocker will place the inode on the appropriate 116403ba3782SJens Axboe * superblock list, based upon its state. 116503ba3782SJens Axboe */ 116603ba3782SJens Axboe if (inode->i_state & I_SYNC) 1167250df6edSDave Chinner goto out_unlock_inode; 116803ba3782SJens Axboe 116903ba3782SJens Axboe /* 117003ba3782SJens Axboe * Only add valid (hashed) inodes to the superblock's 117103ba3782SJens Axboe * dirty list. Add blockdev inodes as well. 117203ba3782SJens Axboe */ 117303ba3782SJens Axboe if (!S_ISBLK(inode->i_mode)) { 11741d3382cbSAl Viro if (inode_unhashed(inode)) 1175250df6edSDave Chinner goto out_unlock_inode; 117603ba3782SJens Axboe } 1177a4ffdde6SAl Viro if (inode->i_state & I_FREEING) 1178250df6edSDave Chinner goto out_unlock_inode; 117903ba3782SJens Axboe 118003ba3782SJens Axboe /* 118103ba3782SJens Axboe * If the inode was already on b_dirty/b_io/b_more_io, don't 118203ba3782SJens Axboe * reposition it (that would break b_dirty time-ordering). 118303ba3782SJens Axboe */ 118403ba3782SJens Axboe if (!was_dirty) { 1185a66979abSDave Chinner bool wakeup_bdi = false; 1186253c34e9SArtem Bityutskiy bdi = inode_to_bdi(inode); 1187500b067cSJens Axboe 1188146d7009SJunxiao Bi spin_unlock(&inode->i_lock); 1189146d7009SJunxiao Bi spin_lock(&bdi->wb.list_lock); 1190253c34e9SArtem Bityutskiy if (bdi_cap_writeback_dirty(bdi)) { 1191253c34e9SArtem Bityutskiy WARN(!test_bit(BDI_registered, &bdi->state), 1192253c34e9SArtem Bityutskiy "bdi-%s not registered\n", bdi->name); 1193253c34e9SArtem Bityutskiy 1194253c34e9SArtem Bityutskiy /* 1195253c34e9SArtem Bityutskiy * If this is the first dirty inode for this 1196253c34e9SArtem Bityutskiy * bdi, we have to wake-up the corresponding 1197253c34e9SArtem Bityutskiy * bdi thread to make sure background 1198253c34e9SArtem Bityutskiy * write-back happens later. 1199253c34e9SArtem Bityutskiy */ 1200253c34e9SArtem Bityutskiy if (!wb_has_dirty_io(&bdi->wb)) 1201253c34e9SArtem Bityutskiy wakeup_bdi = true; 1202500b067cSJens Axboe } 120303ba3782SJens Axboe 120403ba3782SJens Axboe inode->dirtied_when = jiffies; 12057ccf19a8SNick Piggin list_move(&inode->i_wb_list, &bdi->wb.b_dirty); 1206f758eeabSChristoph Hellwig spin_unlock(&bdi->wb.list_lock); 1207253c34e9SArtem Bityutskiy 1208253c34e9SArtem Bityutskiy if (wakeup_bdi) 12096467716aSArtem Bityutskiy bdi_wakeup_thread_delayed(bdi); 1210a66979abSDave Chinner return; 1211a66979abSDave Chinner } 1212a66979abSDave Chinner } 1213a66979abSDave Chinner out_unlock_inode: 1214a66979abSDave Chinner spin_unlock(&inode->i_lock); 1215a66979abSDave Chinner 121603ba3782SJens Axboe } 121703ba3782SJens Axboe EXPORT_SYMBOL(__mark_inode_dirty); 121803ba3782SJens Axboe 1219b6e51316SJens Axboe static void wait_sb_inodes(struct super_block *sb) 122066f3b8e2SJens Axboe { 122138f21977SNick Piggin struct inode *inode, *old_inode = NULL; 122238f21977SNick Piggin 122303ba3782SJens Axboe /* 122403ba3782SJens Axboe * We need to be protected against the filesystem going from 122503ba3782SJens Axboe * r/o to r/w or vice versa. 122603ba3782SJens Axboe */ 1227b6e51316SJens Axboe WARN_ON(!rwsem_is_locked(&sb->s_umount)); 122803ba3782SJens Axboe 122955fa6091SDave Chinner spin_lock(&inode_sb_list_lock); 123066f3b8e2SJens Axboe 123138f21977SNick Piggin /* 123238f21977SNick Piggin * Data integrity sync. Must wait for all pages under writeback, 123338f21977SNick Piggin * because there may have been pages dirtied before our sync 123438f21977SNick Piggin * call, but which had writeout started before we write it out. 123538f21977SNick Piggin * In which case, the inode may not be on the dirty list, but 123638f21977SNick Piggin * we still have to wait for that writeout. 123738f21977SNick Piggin */ 1238b6e51316SJens Axboe list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { 1239250df6edSDave Chinner struct address_space *mapping = inode->i_mapping; 124038f21977SNick Piggin 1241250df6edSDave Chinner spin_lock(&inode->i_lock); 1242250df6edSDave Chinner if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) || 1243250df6edSDave Chinner (mapping->nrpages == 0)) { 1244250df6edSDave Chinner spin_unlock(&inode->i_lock); 124538f21977SNick Piggin continue; 1246250df6edSDave Chinner } 124738f21977SNick Piggin __iget(inode); 1248250df6edSDave Chinner spin_unlock(&inode->i_lock); 124955fa6091SDave Chinner spin_unlock(&inode_sb_list_lock); 125055fa6091SDave Chinner 125138f21977SNick Piggin /* 125255fa6091SDave Chinner * We hold a reference to 'inode' so it couldn't have been 125355fa6091SDave Chinner * removed from s_inodes list while we dropped the 125455fa6091SDave Chinner * inode_sb_list_lock. We cannot iput the inode now as we can 125555fa6091SDave Chinner * be holding the last reference and we cannot iput it under 125655fa6091SDave Chinner * inode_sb_list_lock. So we keep the reference and iput it 125755fa6091SDave Chinner * later. 125838f21977SNick Piggin */ 125938f21977SNick Piggin iput(old_inode); 126038f21977SNick Piggin old_inode = inode; 126138f21977SNick Piggin 126238f21977SNick Piggin filemap_fdatawait(mapping); 126338f21977SNick Piggin 126438f21977SNick Piggin cond_resched(); 126538f21977SNick Piggin 126655fa6091SDave Chinner spin_lock(&inode_sb_list_lock); 126738f21977SNick Piggin } 126855fa6091SDave Chinner spin_unlock(&inode_sb_list_lock); 126938f21977SNick Piggin iput(old_inode); 127066f3b8e2SJens Axboe } 12711da177e4SLinus Torvalds 1272d8a8559cSJens Axboe /** 12733259f8beSChris Mason * writeback_inodes_sb_nr - writeback dirty inodes from given super_block 1274d8a8559cSJens Axboe * @sb: the superblock 12753259f8beSChris Mason * @nr: the number of pages to write 1276786228abSMarcos Paulo de Souza * @reason: reason why some writeback work initiated 12771da177e4SLinus Torvalds * 1278d8a8559cSJens Axboe * Start writeback on some inodes on this super_block. No guarantees are made 1279d8a8559cSJens Axboe * on how many (if any) will be written, and this function does not wait 12803259f8beSChris Mason * for IO completion of submitted IO. 12811da177e4SLinus Torvalds */ 12820e175a18SCurt Wohlgemuth void writeback_inodes_sb_nr(struct super_block *sb, 12830e175a18SCurt Wohlgemuth unsigned long nr, 12840e175a18SCurt Wohlgemuth enum wb_reason reason) 12851da177e4SLinus Torvalds { 128683ba7b07SChristoph Hellwig DECLARE_COMPLETION_ONSTACK(done); 128783ba7b07SChristoph Hellwig struct wb_writeback_work work = { 12883c4d7165SChristoph Hellwig .sb = sb, 12893c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_NONE, 12906e6938b6SWu Fengguang .tagged_writepages = 1, 129183ba7b07SChristoph Hellwig .done = &done, 12923259f8beSChris Mason .nr_pages = nr, 12930e175a18SCurt Wohlgemuth .reason = reason, 12943c4d7165SChristoph Hellwig }; 12950e3c9a22SJens Axboe 12966eedc701SJan Kara if (sb->s_bdi == &noop_backing_dev_info) 12976eedc701SJan Kara return; 1298cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 129983ba7b07SChristoph Hellwig bdi_queue_work(sb->s_bdi, &work); 130083ba7b07SChristoph Hellwig wait_for_completion(&done); 13011da177e4SLinus Torvalds } 13023259f8beSChris Mason EXPORT_SYMBOL(writeback_inodes_sb_nr); 13033259f8beSChris Mason 13043259f8beSChris Mason /** 13053259f8beSChris Mason * writeback_inodes_sb - writeback dirty inodes from given super_block 13063259f8beSChris Mason * @sb: the superblock 1307786228abSMarcos Paulo de Souza * @reason: reason why some writeback work was initiated 13083259f8beSChris Mason * 13093259f8beSChris Mason * Start writeback on some inodes on this super_block. No guarantees are made 13103259f8beSChris Mason * on how many (if any) will be written, and this function does not wait 13113259f8beSChris Mason * for IO completion of submitted IO. 13123259f8beSChris Mason */ 13130e175a18SCurt Wohlgemuth void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) 13143259f8beSChris Mason { 13150e175a18SCurt Wohlgemuth return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); 13163259f8beSChris Mason } 1317d8a8559cSJens Axboe EXPORT_SYMBOL(writeback_inodes_sb); 1318d8a8559cSJens Axboe 1319d8a8559cSJens Axboe /** 132010ee27a0SMiao Xie * try_to_writeback_inodes_sb_nr - try to start writeback if none underway 13213259f8beSChris Mason * @sb: the superblock 13223259f8beSChris Mason * @nr: the number of pages to write 132310ee27a0SMiao Xie * @reason: the reason of writeback 13243259f8beSChris Mason * 132510ee27a0SMiao Xie * Invoke writeback_inodes_sb_nr if no writeback is currently underway. 13263259f8beSChris Mason * Returns 1 if writeback was started, 0 if not. 13273259f8beSChris Mason */ 132810ee27a0SMiao Xie int try_to_writeback_inodes_sb_nr(struct super_block *sb, 13290e175a18SCurt Wohlgemuth unsigned long nr, 13300e175a18SCurt Wohlgemuth enum wb_reason reason) 13313259f8beSChris Mason { 133210ee27a0SMiao Xie if (writeback_in_progress(sb->s_bdi)) 133310ee27a0SMiao Xie return 1; 133410ee27a0SMiao Xie 133510ee27a0SMiao Xie if (!down_read_trylock(&sb->s_umount)) 133610ee27a0SMiao Xie return 0; 133710ee27a0SMiao Xie 13380e175a18SCurt Wohlgemuth writeback_inodes_sb_nr(sb, nr, reason); 13393259f8beSChris Mason up_read(&sb->s_umount); 13403259f8beSChris Mason return 1; 13413259f8beSChris Mason } 134210ee27a0SMiao Xie EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr); 134310ee27a0SMiao Xie 134410ee27a0SMiao Xie /** 134510ee27a0SMiao Xie * try_to_writeback_inodes_sb - try to start writeback if none underway 134610ee27a0SMiao Xie * @sb: the superblock 134710ee27a0SMiao Xie * @reason: reason why some writeback work was initiated 134810ee27a0SMiao Xie * 134910ee27a0SMiao Xie * Implement by try_to_writeback_inodes_sb_nr() 135010ee27a0SMiao Xie * Returns 1 if writeback was started, 0 if not. 135110ee27a0SMiao Xie */ 135210ee27a0SMiao Xie int try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) 135310ee27a0SMiao Xie { 135410ee27a0SMiao Xie return try_to_writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); 135510ee27a0SMiao Xie } 135610ee27a0SMiao Xie EXPORT_SYMBOL(try_to_writeback_inodes_sb); 13573259f8beSChris Mason 13583259f8beSChris Mason /** 1359d8a8559cSJens Axboe * sync_inodes_sb - sync sb inode pages 1360d8a8559cSJens Axboe * @sb: the superblock 1361c4a391b5SJan Kara * @older_than_this: timestamp 1362d8a8559cSJens Axboe * 1363d8a8559cSJens Axboe * This function writes and waits on any dirty inode belonging to this 1364c4a391b5SJan Kara * superblock that has been dirtied before given timestamp. 1365d8a8559cSJens Axboe */ 1366c4a391b5SJan Kara void sync_inodes_sb(struct super_block *sb, unsigned long older_than_this) 1367d8a8559cSJens Axboe { 136883ba7b07SChristoph Hellwig DECLARE_COMPLETION_ONSTACK(done); 136983ba7b07SChristoph Hellwig struct wb_writeback_work work = { 13703c4d7165SChristoph Hellwig .sb = sb, 13713c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_ALL, 13723c4d7165SChristoph Hellwig .nr_pages = LONG_MAX, 1373c4a391b5SJan Kara .older_than_this = older_than_this, 1374c4a391b5SJan Kara .older_than_this_is_set = 1, 13753c4d7165SChristoph Hellwig .range_cyclic = 0, 137683ba7b07SChristoph Hellwig .done = &done, 13770e175a18SCurt Wohlgemuth .reason = WB_REASON_SYNC, 13787747bd4bSDave Chinner .for_sync = 1, 13793c4d7165SChristoph Hellwig }; 13803c4d7165SChristoph Hellwig 13816eedc701SJan Kara /* Nothing to do? */ 13826eedc701SJan Kara if (sb->s_bdi == &noop_backing_dev_info) 13836eedc701SJan Kara return; 1384cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 1385cf37e972SChristoph Hellwig 138683ba7b07SChristoph Hellwig bdi_queue_work(sb->s_bdi, &work); 138783ba7b07SChristoph Hellwig wait_for_completion(&done); 138883ba7b07SChristoph Hellwig 1389b6e51316SJens Axboe wait_sb_inodes(sb); 1390d8a8559cSJens Axboe } 1391d8a8559cSJens Axboe EXPORT_SYMBOL(sync_inodes_sb); 13921da177e4SLinus Torvalds 13931da177e4SLinus Torvalds /** 13941da177e4SLinus Torvalds * write_inode_now - write an inode to disk 13951da177e4SLinus Torvalds * @inode: inode to write to disk 13961da177e4SLinus Torvalds * @sync: whether the write should be synchronous or not 13971da177e4SLinus Torvalds * 13987f04c26dSAndrea Arcangeli * This function commits an inode to disk immediately if it is dirty. This is 13997f04c26dSAndrea Arcangeli * primarily needed by knfsd. 14007f04c26dSAndrea Arcangeli * 14017f04c26dSAndrea Arcangeli * The caller must either have a ref on the inode or must have set I_WILL_FREE. 14021da177e4SLinus Torvalds */ 14031da177e4SLinus Torvalds int write_inode_now(struct inode *inode, int sync) 14041da177e4SLinus Torvalds { 1405f758eeabSChristoph Hellwig struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; 14061da177e4SLinus Torvalds struct writeback_control wbc = { 14071da177e4SLinus Torvalds .nr_to_write = LONG_MAX, 140818914b18SMike Galbraith .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, 1409111ebb6eSOGAWA Hirofumi .range_start = 0, 1410111ebb6eSOGAWA Hirofumi .range_end = LLONG_MAX, 14111da177e4SLinus Torvalds }; 14121da177e4SLinus Torvalds 14131da177e4SLinus Torvalds if (!mapping_cap_writeback_dirty(inode->i_mapping)) 141449364ce2SAndrew Morton wbc.nr_to_write = 0; 14151da177e4SLinus Torvalds 14161da177e4SLinus Torvalds might_sleep(); 14174f8ad655SJan Kara return writeback_single_inode(inode, wb, &wbc); 14181da177e4SLinus Torvalds } 14191da177e4SLinus Torvalds EXPORT_SYMBOL(write_inode_now); 14201da177e4SLinus Torvalds 14211da177e4SLinus Torvalds /** 14221da177e4SLinus Torvalds * sync_inode - write an inode and its pages to disk. 14231da177e4SLinus Torvalds * @inode: the inode to sync 14241da177e4SLinus Torvalds * @wbc: controls the writeback mode 14251da177e4SLinus Torvalds * 14261da177e4SLinus Torvalds * sync_inode() will write an inode and its pages to disk. It will also 14271da177e4SLinus Torvalds * correctly update the inode on its superblock's dirty inode lists and will 14281da177e4SLinus Torvalds * update inode->i_state. 14291da177e4SLinus Torvalds * 14301da177e4SLinus Torvalds * The caller must have a ref on the inode. 14311da177e4SLinus Torvalds */ 14321da177e4SLinus Torvalds int sync_inode(struct inode *inode, struct writeback_control *wbc) 14331da177e4SLinus Torvalds { 14344f8ad655SJan Kara return writeback_single_inode(inode, &inode_to_bdi(inode)->wb, wbc); 14351da177e4SLinus Torvalds } 14361da177e4SLinus Torvalds EXPORT_SYMBOL(sync_inode); 1437c3765016SChristoph Hellwig 1438c3765016SChristoph Hellwig /** 1439c691b9d9SAndrew Morton * sync_inode_metadata - write an inode to disk 1440c3765016SChristoph Hellwig * @inode: the inode to sync 1441c3765016SChristoph Hellwig * @wait: wait for I/O to complete. 1442c3765016SChristoph Hellwig * 1443c691b9d9SAndrew Morton * Write an inode to disk and adjust its dirty state after completion. 1444c3765016SChristoph Hellwig * 1445c3765016SChristoph Hellwig * Note: only writes the actual inode, no associated data or other metadata. 1446c3765016SChristoph Hellwig */ 1447c3765016SChristoph Hellwig int sync_inode_metadata(struct inode *inode, int wait) 1448c3765016SChristoph Hellwig { 1449c3765016SChristoph Hellwig struct writeback_control wbc = { 1450c3765016SChristoph Hellwig .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE, 1451c3765016SChristoph Hellwig .nr_to_write = 0, /* metadata-only */ 1452c3765016SChristoph Hellwig }; 1453c3765016SChristoph Hellwig 1454c3765016SChristoph Hellwig return sync_inode(inode, &wbc); 1455c3765016SChristoph Hellwig } 1456c3765016SChristoph Hellwig EXPORT_SYMBOL(sync_inode_metadata); 1457