11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * fs/fs-writeback.c 31da177e4SLinus Torvalds * 41da177e4SLinus Torvalds * Copyright (C) 2002, Linus Torvalds. 51da177e4SLinus Torvalds * 61da177e4SLinus Torvalds * Contains all the functions related to writing back and waiting 71da177e4SLinus Torvalds * upon dirty inodes against superblocks, and writing back dirty 81da177e4SLinus Torvalds * pages against inodes. ie: data writeback. Writeout of the 91da177e4SLinus Torvalds * inode itself is not handled here. 101da177e4SLinus Torvalds * 11e1f8e874SFrancois Cami * 10Apr2002 Andrew Morton 121da177e4SLinus Torvalds * Split out of fs/inode.c 131da177e4SLinus Torvalds * Additions for address_space-based writeback 141da177e4SLinus Torvalds */ 151da177e4SLinus Torvalds 161da177e4SLinus Torvalds #include <linux/kernel.h> 17f5ff8422SJens Axboe #include <linux/module.h> 181da177e4SLinus Torvalds #include <linux/spinlock.h> 195a0e3ad6STejun Heo #include <linux/slab.h> 201da177e4SLinus Torvalds #include <linux/sched.h> 211da177e4SLinus Torvalds #include <linux/fs.h> 221da177e4SLinus Torvalds #include <linux/mm.h> 2303ba3782SJens Axboe #include <linux/kthread.h> 2403ba3782SJens Axboe #include <linux/freezer.h> 251da177e4SLinus Torvalds #include <linux/writeback.h> 261da177e4SLinus Torvalds #include <linux/blkdev.h> 271da177e4SLinus Torvalds #include <linux/backing-dev.h> 281da177e4SLinus Torvalds #include <linux/buffer_head.h> 29455b2864SDave Chinner #include <linux/tracepoint.h> 3007f3f05cSDavid Howells #include "internal.h" 311da177e4SLinus Torvalds 32d0bceac7SJens Axboe /* 33c4a77a6cSJens Axboe * Passed into wb_writeback(), essentially a subset of writeback_control 34c4a77a6cSJens Axboe */ 3583ba7b07SChristoph Hellwig struct wb_writeback_work { 36c4a77a6cSJens Axboe long nr_pages; 37c4a77a6cSJens Axboe struct super_block *sb; 38d46db3d5SWu Fengguang unsigned long *older_than_this; 39c4a77a6cSJens Axboe enum writeback_sync_modes sync_mode; 406e6938b6SWu Fengguang unsigned int tagged_writepages:1; 4152957fe1SH Hartley Sweeten unsigned int for_kupdate:1; 4252957fe1SH Hartley Sweeten unsigned int range_cyclic:1; 4352957fe1SH Hartley Sweeten unsigned int for_background:1; 44c4a77a6cSJens Axboe 458010c3b6SJens Axboe struct list_head list; /* pending work list */ 4683ba7b07SChristoph Hellwig struct completion *done; /* set if the caller waits */ 4703ba3782SJens Axboe }; 4803ba3782SJens Axboe 49455b2864SDave Chinner /* 50455b2864SDave Chinner * Include the creation of the trace points after defining the 51455b2864SDave Chinner * wb_writeback_work structure so that the definition remains local to this 52455b2864SDave Chinner * file. 53455b2864SDave Chinner */ 54455b2864SDave Chinner #define CREATE_TRACE_POINTS 55455b2864SDave Chinner #include <trace/events/writeback.h> 56455b2864SDave Chinner 57455b2864SDave Chinner /* 58455b2864SDave Chinner * We don't actually have pdflush, but this one is exported though /proc... 59455b2864SDave Chinner */ 60455b2864SDave Chinner int nr_pdflush_threads; 61455b2864SDave Chinner 62f11b00f3SAdrian Bunk /** 63f11b00f3SAdrian Bunk * writeback_in_progress - determine whether there is writeback in progress 64f11b00f3SAdrian Bunk * @bdi: the device's backing_dev_info structure. 65f11b00f3SAdrian Bunk * 6603ba3782SJens Axboe * Determine whether there is writeback waiting to be handled against a 6703ba3782SJens Axboe * backing device. 68f11b00f3SAdrian Bunk */ 69f11b00f3SAdrian Bunk int writeback_in_progress(struct backing_dev_info *bdi) 70f11b00f3SAdrian Bunk { 7181d73a32SJan Kara return test_bit(BDI_writeback_running, &bdi->state); 72f11b00f3SAdrian Bunk } 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 78aaead25bSChristoph Hellwig if (strcmp(sb->s_type->name, "bdev") == 0) 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 896585027aSJan Kara /* Wakeup flusher thread or forker thread to fork it. Requires bdi->wb_lock. */ 906585027aSJan Kara static void bdi_wakeup_flusher(struct backing_dev_info *bdi) 914195f73dSNick Piggin { 92fff5b85aSArtem Bityutskiy if (bdi->wb.task) { 93fff5b85aSArtem Bityutskiy wake_up_process(bdi->wb.task); 94fff5b85aSArtem Bityutskiy } else { 951da177e4SLinus Torvalds /* 96fff5b85aSArtem Bityutskiy * The bdi thread isn't there, wake up the forker thread which 97fff5b85aSArtem Bityutskiy * will create and run it. 981da177e4SLinus Torvalds */ 9903ba3782SJens Axboe wake_up_process(default_backing_dev_info.wb.task); 1001da177e4SLinus Torvalds } 1016585027aSJan Kara } 1026585027aSJan Kara 1036585027aSJan Kara static void bdi_queue_work(struct backing_dev_info *bdi, 1046585027aSJan Kara struct wb_writeback_work *work) 1056585027aSJan Kara { 1066585027aSJan Kara trace_writeback_queue(bdi, work); 1076585027aSJan Kara 1086585027aSJan Kara spin_lock_bh(&bdi->wb_lock); 1096585027aSJan Kara list_add_tail(&work->list, &bdi->work_list); 1106585027aSJan Kara if (!bdi->wb.task) 1116585027aSJan Kara trace_writeback_nothread(bdi, work); 1126585027aSJan Kara bdi_wakeup_flusher(bdi); 1136467716aSArtem Bityutskiy spin_unlock_bh(&bdi->wb_lock); 11403ba3782SJens Axboe } 1151da177e4SLinus Torvalds 11683ba7b07SChristoph Hellwig static void 11783ba7b07SChristoph Hellwig __bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, 1186585027aSJan Kara bool range_cyclic) 1191da177e4SLinus Torvalds { 12083ba7b07SChristoph Hellwig struct wb_writeback_work *work; 12103ba3782SJens Axboe 122bcddc3f0SJens Axboe /* 123bcddc3f0SJens Axboe * This is WB_SYNC_NONE writeback, so if allocation fails just 124bcddc3f0SJens Axboe * wakeup the thread for old dirty data writeback 125bcddc3f0SJens Axboe */ 12683ba7b07SChristoph Hellwig work = kzalloc(sizeof(*work), GFP_ATOMIC); 12783ba7b07SChristoph Hellwig if (!work) { 128455b2864SDave Chinner if (bdi->wb.task) { 129455b2864SDave Chinner trace_writeback_nowork(bdi); 13083ba7b07SChristoph Hellwig wake_up_process(bdi->wb.task); 131455b2864SDave Chinner } 13283ba7b07SChristoph Hellwig return; 13383ba7b07SChristoph Hellwig } 13483ba7b07SChristoph Hellwig 13583ba7b07SChristoph Hellwig work->sync_mode = WB_SYNC_NONE; 13683ba7b07SChristoph Hellwig work->nr_pages = nr_pages; 13783ba7b07SChristoph Hellwig work->range_cyclic = range_cyclic; 13883ba7b07SChristoph Hellwig 139f11fcae8SJens Axboe bdi_queue_work(bdi, work); 14003ba3782SJens Axboe } 141b6e51316SJens Axboe 142b6e51316SJens Axboe /** 143b6e51316SJens Axboe * bdi_start_writeback - start writeback 144b6e51316SJens Axboe * @bdi: the backing device to write from 145b6e51316SJens Axboe * @nr_pages: the number of pages to write 146b6e51316SJens Axboe * 147b6e51316SJens Axboe * Description: 148b6e51316SJens Axboe * This does WB_SYNC_NONE opportunistic writeback. The IO is only 14925985edcSLucas De Marchi * started when this function returns, we make no guarantees on 1500e3c9a22SJens Axboe * completion. Caller need not hold sb s_umount semaphore. 151b6e51316SJens Axboe * 152b6e51316SJens Axboe */ 153c5444198SChristoph Hellwig void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages) 154b6e51316SJens Axboe { 1556585027aSJan Kara __bdi_start_writeback(bdi, nr_pages, true); 156d3ddec76SWu Fengguang } 157d3ddec76SWu Fengguang 158c5444198SChristoph Hellwig /** 159c5444198SChristoph Hellwig * bdi_start_background_writeback - start background writeback 160c5444198SChristoph Hellwig * @bdi: the backing device to write from 161c5444198SChristoph Hellwig * 162c5444198SChristoph Hellwig * Description: 1636585027aSJan Kara * This makes sure WB_SYNC_NONE background writeback happens. When 1646585027aSJan Kara * this function returns, it is only guaranteed that for given BDI 1656585027aSJan Kara * some IO is happening if we are over background dirty threshold. 1666585027aSJan Kara * Caller need not hold sb s_umount semaphore. 167c5444198SChristoph Hellwig */ 168c5444198SChristoph Hellwig void bdi_start_background_writeback(struct backing_dev_info *bdi) 169c5444198SChristoph Hellwig { 1706585027aSJan Kara /* 1716585027aSJan Kara * We just wake up the flusher thread. It will perform background 1726585027aSJan Kara * writeback as soon as there is no other work to do. 1736585027aSJan Kara */ 17471927e84SWu Fengguang trace_writeback_wake_background(bdi); 1756585027aSJan Kara spin_lock_bh(&bdi->wb_lock); 1766585027aSJan Kara bdi_wakeup_flusher(bdi); 1776585027aSJan Kara spin_unlock_bh(&bdi->wb_lock); 1781da177e4SLinus Torvalds } 1791da177e4SLinus Torvalds 1801da177e4SLinus Torvalds /* 181a66979abSDave Chinner * Remove the inode from the writeback list it is on. 182a66979abSDave Chinner */ 183a66979abSDave Chinner void inode_wb_list_del(struct inode *inode) 184a66979abSDave Chinner { 185f758eeabSChristoph Hellwig struct backing_dev_info *bdi = inode_to_bdi(inode); 186a66979abSDave Chinner 187f758eeabSChristoph Hellwig spin_lock(&bdi->wb.list_lock); 188f758eeabSChristoph Hellwig list_del_init(&inode->i_wb_list); 189f758eeabSChristoph Hellwig spin_unlock(&bdi->wb.list_lock); 190f758eeabSChristoph Hellwig } 191a66979abSDave Chinner 192a66979abSDave Chinner /* 1936610a0bcSAndrew Morton * Redirty an inode: set its when-it-was dirtied timestamp and move it to the 1946610a0bcSAndrew Morton * furthest end of its superblock's dirty-inode list. 1956610a0bcSAndrew Morton * 1966610a0bcSAndrew Morton * Before stamping the inode's ->dirtied_when, we check to see whether it is 19766f3b8e2SJens Axboe * already the most-recently-dirtied inode on the b_dirty list. If that is 1986610a0bcSAndrew Morton * the case then the inode must have been redirtied while it was being written 1996610a0bcSAndrew Morton * out and we don't reset its dirtied_when. 2006610a0bcSAndrew Morton */ 201f758eeabSChristoph Hellwig static void redirty_tail(struct inode *inode, struct bdi_writeback *wb) 2026610a0bcSAndrew Morton { 203f758eeabSChristoph Hellwig assert_spin_locked(&wb->list_lock); 20403ba3782SJens Axboe if (!list_empty(&wb->b_dirty)) { 20566f3b8e2SJens Axboe struct inode *tail; 2066610a0bcSAndrew Morton 2077ccf19a8SNick Piggin tail = wb_inode(wb->b_dirty.next); 20866f3b8e2SJens Axboe if (time_before(inode->dirtied_when, tail->dirtied_when)) 2096610a0bcSAndrew Morton inode->dirtied_when = jiffies; 2106610a0bcSAndrew Morton } 2117ccf19a8SNick Piggin list_move(&inode->i_wb_list, &wb->b_dirty); 2126610a0bcSAndrew Morton } 2136610a0bcSAndrew Morton 2146610a0bcSAndrew Morton /* 21566f3b8e2SJens Axboe * requeue inode for re-scanning after bdi->b_io list is exhausted. 216c986d1e2SAndrew Morton */ 217f758eeabSChristoph Hellwig static void requeue_io(struct inode *inode, struct bdi_writeback *wb) 218c986d1e2SAndrew Morton { 219f758eeabSChristoph Hellwig assert_spin_locked(&wb->list_lock); 2207ccf19a8SNick Piggin list_move(&inode->i_wb_list, &wb->b_more_io); 221c986d1e2SAndrew Morton } 222c986d1e2SAndrew Morton 2231c0eeaf5SJoern Engel static void inode_sync_complete(struct inode *inode) 2241c0eeaf5SJoern Engel { 2251c0eeaf5SJoern Engel /* 226a66979abSDave Chinner * Prevent speculative execution through 227f758eeabSChristoph Hellwig * spin_unlock(&wb->list_lock); 2281c0eeaf5SJoern Engel */ 229a66979abSDave Chinner 2301c0eeaf5SJoern Engel smp_mb(); 2311c0eeaf5SJoern Engel wake_up_bit(&inode->i_state, __I_SYNC); 2321c0eeaf5SJoern Engel } 2331c0eeaf5SJoern Engel 234d2caa3c5SJeff Layton static bool inode_dirtied_after(struct inode *inode, unsigned long t) 235d2caa3c5SJeff Layton { 236d2caa3c5SJeff Layton bool ret = time_after(inode->dirtied_when, t); 237d2caa3c5SJeff Layton #ifndef CONFIG_64BIT 238d2caa3c5SJeff Layton /* 239d2caa3c5SJeff Layton * For inodes being constantly redirtied, dirtied_when can get stuck. 240d2caa3c5SJeff Layton * It _appears_ to be in the future, but is actually in distant past. 241d2caa3c5SJeff Layton * This test is necessary to prevent such wrapped-around relative times 2425b0830cbSJens Axboe * from permanently stopping the whole bdi writeback. 243d2caa3c5SJeff Layton */ 244d2caa3c5SJeff Layton ret = ret && time_before_eq(inode->dirtied_when, jiffies); 245d2caa3c5SJeff Layton #endif 246d2caa3c5SJeff Layton return ret; 247d2caa3c5SJeff Layton } 248d2caa3c5SJeff Layton 249c986d1e2SAndrew Morton /* 2502c136579SFengguang Wu * Move expired dirty inodes from @delaying_queue to @dispatch_queue. 2512c136579SFengguang Wu */ 252e84d0a4fSWu Fengguang static int move_expired_inodes(struct list_head *delaying_queue, 2532c136579SFengguang Wu struct list_head *dispatch_queue, 2542c136579SFengguang Wu unsigned long *older_than_this) 2552c136579SFengguang Wu { 2565c03449dSShaohua Li LIST_HEAD(tmp); 2575c03449dSShaohua Li struct list_head *pos, *node; 258cf137307SJens Axboe struct super_block *sb = NULL; 2595c03449dSShaohua Li struct inode *inode; 260cf137307SJens Axboe int do_sb_sort = 0; 261e84d0a4fSWu Fengguang int moved = 0; 2625c03449dSShaohua Li 2632c136579SFengguang Wu while (!list_empty(delaying_queue)) { 2647ccf19a8SNick Piggin inode = wb_inode(delaying_queue->prev); 2652c136579SFengguang Wu if (older_than_this && 266d2caa3c5SJeff Layton inode_dirtied_after(inode, *older_than_this)) 2672c136579SFengguang Wu break; 268cf137307SJens Axboe if (sb && sb != inode->i_sb) 269cf137307SJens Axboe do_sb_sort = 1; 270cf137307SJens Axboe sb = inode->i_sb; 2717ccf19a8SNick Piggin list_move(&inode->i_wb_list, &tmp); 272e84d0a4fSWu Fengguang moved++; 2735c03449dSShaohua Li } 2745c03449dSShaohua Li 275cf137307SJens Axboe /* just one sb in list, splice to dispatch_queue and we're done */ 276cf137307SJens Axboe if (!do_sb_sort) { 277cf137307SJens Axboe list_splice(&tmp, dispatch_queue); 278e84d0a4fSWu Fengguang goto out; 279cf137307SJens Axboe } 280cf137307SJens Axboe 2815c03449dSShaohua Li /* Move inodes from one superblock together */ 2825c03449dSShaohua Li while (!list_empty(&tmp)) { 2837ccf19a8SNick Piggin sb = wb_inode(tmp.prev)->i_sb; 2845c03449dSShaohua Li list_for_each_prev_safe(pos, node, &tmp) { 2857ccf19a8SNick Piggin inode = wb_inode(pos); 2865c03449dSShaohua Li if (inode->i_sb == sb) 2877ccf19a8SNick Piggin list_move(&inode->i_wb_list, dispatch_queue); 2882c136579SFengguang Wu } 2892c136579SFengguang Wu } 290e84d0a4fSWu Fengguang out: 291e84d0a4fSWu Fengguang return moved; 2925c03449dSShaohua Li } 2932c136579SFengguang Wu 2942c136579SFengguang Wu /* 2952c136579SFengguang Wu * Queue all expired dirty inodes for io, eldest first. 2964ea879b9SWu Fengguang * Before 2974ea879b9SWu Fengguang * newly dirtied b_dirty b_io b_more_io 2984ea879b9SWu Fengguang * =============> gf edc BA 2994ea879b9SWu Fengguang * After 3004ea879b9SWu Fengguang * newly dirtied b_dirty b_io b_more_io 3014ea879b9SWu Fengguang * =============> g fBAedc 3024ea879b9SWu Fengguang * | 3034ea879b9SWu Fengguang * +--> dequeue for IO 3042c136579SFengguang Wu */ 30503ba3782SJens Axboe static void queue_io(struct bdi_writeback *wb, unsigned long *older_than_this) 3062c136579SFengguang Wu { 307e84d0a4fSWu Fengguang int moved; 308f758eeabSChristoph Hellwig assert_spin_locked(&wb->list_lock); 3094ea879b9SWu Fengguang list_splice_init(&wb->b_more_io, &wb->b_io); 310e84d0a4fSWu Fengguang moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, older_than_this); 311e84d0a4fSWu Fengguang trace_writeback_queue_io(wb, older_than_this, moved); 31266f3b8e2SJens Axboe } 31366f3b8e2SJens Axboe 314a9185b41SChristoph Hellwig static int write_inode(struct inode *inode, struct writeback_control *wbc) 31566f3b8e2SJens Axboe { 31603ba3782SJens Axboe if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) 317a9185b41SChristoph Hellwig return inode->i_sb->s_op->write_inode(inode, wbc); 31803ba3782SJens Axboe return 0; 31966f3b8e2SJens Axboe } 32008d8e974SFengguang Wu 3212c136579SFengguang Wu /* 32201c03194SChristoph Hellwig * Wait for writeback on an inode to complete. 32301c03194SChristoph Hellwig */ 324f758eeabSChristoph Hellwig static void inode_wait_for_writeback(struct inode *inode, 325f758eeabSChristoph Hellwig struct bdi_writeback *wb) 32601c03194SChristoph Hellwig { 32701c03194SChristoph Hellwig DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); 32801c03194SChristoph Hellwig wait_queue_head_t *wqh; 32901c03194SChristoph Hellwig 33001c03194SChristoph Hellwig wqh = bit_waitqueue(&inode->i_state, __I_SYNC); 33158a9d3d8SRichard Kennedy while (inode->i_state & I_SYNC) { 332250df6edSDave Chinner spin_unlock(&inode->i_lock); 333f758eeabSChristoph Hellwig spin_unlock(&wb->list_lock); 33401c03194SChristoph Hellwig __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE); 335f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 336250df6edSDave Chinner spin_lock(&inode->i_lock); 33758a9d3d8SRichard Kennedy } 33801c03194SChristoph Hellwig } 33901c03194SChristoph Hellwig 34001c03194SChristoph Hellwig /* 341f758eeabSChristoph Hellwig * Write out an inode's dirty pages. Called under wb->list_lock and 3420f1b1fd8SDave Chinner * inode->i_lock. Either the caller has an active reference on the inode or 3430f1b1fd8SDave Chinner * the inode has I_WILL_FREE set. 34401c03194SChristoph Hellwig * 3451da177e4SLinus Torvalds * If `wait' is set, wait on the writeout. 3461da177e4SLinus Torvalds * 3471da177e4SLinus Torvalds * The whole writeout design is quite complex and fragile. We want to avoid 3481da177e4SLinus Torvalds * starvation of particular inodes when others are being redirtied, prevent 3491da177e4SLinus Torvalds * livelocks, etc. 3501da177e4SLinus Torvalds */ 3511da177e4SLinus Torvalds static int 352f758eeabSChristoph Hellwig writeback_single_inode(struct inode *inode, struct bdi_writeback *wb, 353f758eeabSChristoph Hellwig struct writeback_control *wbc) 3541da177e4SLinus Torvalds { 3551da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 356251d6a47SWu Fengguang long nr_to_write = wbc->nr_to_write; 35701c03194SChristoph Hellwig unsigned dirty; 3581da177e4SLinus Torvalds int ret; 3591da177e4SLinus Torvalds 360f758eeabSChristoph Hellwig assert_spin_locked(&wb->list_lock); 3610f1b1fd8SDave Chinner assert_spin_locked(&inode->i_lock); 3620f1b1fd8SDave Chinner 36301c03194SChristoph Hellwig if (!atomic_read(&inode->i_count)) 36401c03194SChristoph Hellwig WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); 36501c03194SChristoph Hellwig else 36601c03194SChristoph Hellwig WARN_ON(inode->i_state & I_WILL_FREE); 36701c03194SChristoph Hellwig 36801c03194SChristoph Hellwig if (inode->i_state & I_SYNC) { 36901c03194SChristoph Hellwig /* 37001c03194SChristoph Hellwig * If this inode is locked for writeback and we are not doing 37166f3b8e2SJens Axboe * writeback-for-data-integrity, move it to b_more_io so that 37201c03194SChristoph Hellwig * writeback can proceed with the other inodes on s_io. 37301c03194SChristoph Hellwig * 37401c03194SChristoph Hellwig * We'll have another go at writing back this inode when we 37566f3b8e2SJens Axboe * completed a full scan of b_io. 37601c03194SChristoph Hellwig */ 377a9185b41SChristoph Hellwig if (wbc->sync_mode != WB_SYNC_ALL) { 378f758eeabSChristoph Hellwig requeue_io(inode, wb); 379251d6a47SWu Fengguang trace_writeback_single_inode_requeue(inode, wbc, 380251d6a47SWu Fengguang nr_to_write); 38101c03194SChristoph Hellwig return 0; 38201c03194SChristoph Hellwig } 38301c03194SChristoph Hellwig 38401c03194SChristoph Hellwig /* 38501c03194SChristoph Hellwig * It's a data-integrity sync. We must wait. 38601c03194SChristoph Hellwig */ 387f758eeabSChristoph Hellwig inode_wait_for_writeback(inode, wb); 38801c03194SChristoph Hellwig } 38901c03194SChristoph Hellwig 3901c0eeaf5SJoern Engel BUG_ON(inode->i_state & I_SYNC); 3911da177e4SLinus Torvalds 3925547e8aaSDmitry Monakhov /* Set I_SYNC, reset I_DIRTY_PAGES */ 3931c0eeaf5SJoern Engel inode->i_state |= I_SYNC; 3945547e8aaSDmitry Monakhov inode->i_state &= ~I_DIRTY_PAGES; 395250df6edSDave Chinner spin_unlock(&inode->i_lock); 396f758eeabSChristoph Hellwig spin_unlock(&wb->list_lock); 3971da177e4SLinus Torvalds 3981da177e4SLinus Torvalds ret = do_writepages(mapping, wbc); 3991da177e4SLinus Torvalds 40026821ed4SChristoph Hellwig /* 40126821ed4SChristoph Hellwig * Make sure to wait on the data before writing out the metadata. 40226821ed4SChristoph Hellwig * This is important for filesystems that modify metadata on data 40326821ed4SChristoph Hellwig * I/O completion. 40426821ed4SChristoph Hellwig */ 405a9185b41SChristoph Hellwig if (wbc->sync_mode == WB_SYNC_ALL) { 40626821ed4SChristoph Hellwig int err = filemap_fdatawait(mapping); 4071da177e4SLinus Torvalds if (ret == 0) 4081da177e4SLinus Torvalds ret = err; 4091da177e4SLinus Torvalds } 4101da177e4SLinus Torvalds 4115547e8aaSDmitry Monakhov /* 4125547e8aaSDmitry Monakhov * Some filesystems may redirty the inode during the writeback 4135547e8aaSDmitry Monakhov * due to delalloc, clear dirty metadata flags right before 4145547e8aaSDmitry Monakhov * write_inode() 4155547e8aaSDmitry Monakhov */ 416250df6edSDave Chinner spin_lock(&inode->i_lock); 4175547e8aaSDmitry Monakhov dirty = inode->i_state & I_DIRTY; 4185547e8aaSDmitry Monakhov inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC); 419250df6edSDave Chinner spin_unlock(&inode->i_lock); 42026821ed4SChristoph Hellwig /* Don't write the inode if only I_DIRTY_PAGES was set */ 42126821ed4SChristoph Hellwig if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { 422a9185b41SChristoph Hellwig int err = write_inode(inode, wbc); 4231da177e4SLinus Torvalds if (ret == 0) 4241da177e4SLinus Torvalds ret = err; 4251da177e4SLinus Torvalds } 4261da177e4SLinus Torvalds 427f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 428250df6edSDave Chinner spin_lock(&inode->i_lock); 4291c0eeaf5SJoern Engel inode->i_state &= ~I_SYNC; 430a4ffdde6SAl Viro if (!(inode->i_state & I_FREEING)) { 43194c3dcbbSWu Fengguang /* 43294c3dcbbSWu Fengguang * Sync livelock prevention. Each inode is tagged and synced in 43394c3dcbbSWu Fengguang * one shot. If still dirty, it will be redirty_tail()'ed below. 43494c3dcbbSWu Fengguang * Update the dirty time to prevent enqueue and sync it again. 43594c3dcbbSWu Fengguang */ 43694c3dcbbSWu Fengguang if ((inode->i_state & I_DIRTY) && 43794c3dcbbSWu Fengguang (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)) 43894c3dcbbSWu Fengguang inode->dirtied_when = jiffies; 43994c3dcbbSWu Fengguang 44023539afcSWu Fengguang if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { 4411da177e4SLinus Torvalds /* 4421da177e4SLinus Torvalds * We didn't write back all the pages. nfs_writepages() 443a50aeb40SWu Fengguang * sometimes bales out without doing anything. 4441da177e4SLinus Torvalds */ 4451da177e4SLinus Torvalds inode->i_state |= I_DIRTY_PAGES; 4468bc3be27SFengguang Wu if (wbc->nr_to_write <= 0) { 4478bc3be27SFengguang Wu /* 4488bc3be27SFengguang Wu * slice used up: queue for next turn 4498bc3be27SFengguang Wu */ 450f758eeabSChristoph Hellwig requeue_io(inode, wb); 4511da177e4SLinus Torvalds } else { 4521da177e4SLinus Torvalds /* 453a50aeb40SWu Fengguang * Writeback blocked by something other than 454a50aeb40SWu Fengguang * congestion. Delay the inode for some time to 455a50aeb40SWu Fengguang * avoid spinning on the CPU (100% iowait) 456a50aeb40SWu Fengguang * retrying writeback of the dirty page/inode 457a50aeb40SWu Fengguang * that cannot be performed immediately. 4588bc3be27SFengguang Wu */ 459f758eeabSChristoph Hellwig redirty_tail(inode, wb); 4608bc3be27SFengguang Wu } 46123539afcSWu Fengguang } else if (inode->i_state & I_DIRTY) { 46223539afcSWu Fengguang /* 46323539afcSWu Fengguang * Filesystems can dirty the inode during writeback 46423539afcSWu Fengguang * operations, such as delayed allocation during 46523539afcSWu Fengguang * submission or metadata updates after data IO 46623539afcSWu Fengguang * completion. 46723539afcSWu Fengguang */ 468f758eeabSChristoph Hellwig redirty_tail(inode, wb); 4691da177e4SLinus Torvalds } else { 4701da177e4SLinus Torvalds /* 4719e38d86fSNick Piggin * The inode is clean. At this point we either have 4729e38d86fSNick Piggin * a reference to the inode or it's on it's way out. 4739e38d86fSNick Piggin * No need to add it back to the LRU. 4741da177e4SLinus Torvalds */ 4757ccf19a8SNick Piggin list_del_init(&inode->i_wb_list); 4761da177e4SLinus Torvalds } 4771da177e4SLinus Torvalds } 4781c0eeaf5SJoern Engel inode_sync_complete(inode); 479251d6a47SWu Fengguang trace_writeback_single_inode(inode, wbc, nr_to_write); 4801da177e4SLinus Torvalds return ret; 4811da177e4SLinus Torvalds } 4821da177e4SLinus Torvalds 48303ba3782SJens Axboe /* 484d19de7edSChristoph Hellwig * For background writeback the caller does not have the sb pinned 48503ba3782SJens Axboe * before calling writeback. So make sure that we do pin it, so it doesn't 48603ba3782SJens Axboe * go away while we are writing inodes from it. 48703ba3782SJens Axboe */ 488d19de7edSChristoph Hellwig static bool pin_sb_for_writeback(struct super_block *sb) 4891da177e4SLinus Torvalds { 49003ba3782SJens Axboe spin_lock(&sb_lock); 49129cb4859SChristoph Hellwig if (list_empty(&sb->s_instances)) { 49203ba3782SJens Axboe spin_unlock(&sb_lock); 49329cb4859SChristoph Hellwig return false; 49403ba3782SJens Axboe } 49529cb4859SChristoph Hellwig 49629cb4859SChristoph Hellwig sb->s_count++; 49729cb4859SChristoph Hellwig spin_unlock(&sb_lock); 49829cb4859SChristoph Hellwig 49929cb4859SChristoph Hellwig if (down_read_trylock(&sb->s_umount)) { 50029cb4859SChristoph Hellwig if (sb->s_root) 50129cb4859SChristoph Hellwig return true; 50203ba3782SJens Axboe up_read(&sb->s_umount); 50303ba3782SJens Axboe } 50429cb4859SChristoph Hellwig 50529cb4859SChristoph Hellwig put_super(sb); 506d19de7edSChristoph Hellwig return false; 50703ba3782SJens Axboe } 50803ba3782SJens Axboe 5091a12d8bdSWu Fengguang static long writeback_chunk_size(struct backing_dev_info *bdi, 5101a12d8bdSWu Fengguang struct wb_writeback_work *work) 511d46db3d5SWu Fengguang { 512d46db3d5SWu Fengguang long pages; 513d46db3d5SWu Fengguang 514d46db3d5SWu Fengguang /* 515d46db3d5SWu Fengguang * WB_SYNC_ALL mode does livelock avoidance by syncing dirty 516d46db3d5SWu Fengguang * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX 517d46db3d5SWu Fengguang * here avoids calling into writeback_inodes_wb() more than once. 518d46db3d5SWu Fengguang * 519d46db3d5SWu Fengguang * The intended call sequence for WB_SYNC_ALL writeback is: 520d46db3d5SWu Fengguang * 521d46db3d5SWu Fengguang * wb_writeback() 522d46db3d5SWu Fengguang * writeback_sb_inodes() <== called only once 523d46db3d5SWu Fengguang * write_cache_pages() <== called once for each inode 524d46db3d5SWu Fengguang * (quickly) tag currently dirty pages 525d46db3d5SWu Fengguang * (maybe slowly) sync all tagged pages 526d46db3d5SWu Fengguang */ 527d46db3d5SWu Fengguang if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages) 528d46db3d5SWu Fengguang pages = LONG_MAX; 5291a12d8bdSWu Fengguang else { 5301a12d8bdSWu Fengguang pages = min(bdi->avg_write_bandwidth / 2, 5311a12d8bdSWu Fengguang global_dirty_limit / DIRTY_SCOPE); 5321a12d8bdSWu Fengguang pages = min(pages, work->nr_pages); 5331a12d8bdSWu Fengguang pages = round_down(pages + MIN_WRITEBACK_PAGES, 5341a12d8bdSWu Fengguang MIN_WRITEBACK_PAGES); 5351a12d8bdSWu Fengguang } 536d46db3d5SWu Fengguang 537d46db3d5SWu Fengguang return pages; 538d46db3d5SWu Fengguang } 539d46db3d5SWu Fengguang 540f11c9c5cSEdward Shishkin /* 541f11c9c5cSEdward Shishkin * Write a portion of b_io inodes which belong to @sb. 542edadfb10SChristoph Hellwig * 543edadfb10SChristoph Hellwig * If @only_this_sb is true, then find and write all such 544f11c9c5cSEdward Shishkin * inodes. Otherwise write only ones which go sequentially 545f11c9c5cSEdward Shishkin * in reverse order. 546edadfb10SChristoph Hellwig * 547d46db3d5SWu Fengguang * Return the number of pages and/or inodes written. 548f11c9c5cSEdward Shishkin */ 549d46db3d5SWu Fengguang static long writeback_sb_inodes(struct super_block *sb, 550d46db3d5SWu Fengguang struct bdi_writeback *wb, 551d46db3d5SWu Fengguang struct wb_writeback_work *work) 55203ba3782SJens Axboe { 553d46db3d5SWu Fengguang struct writeback_control wbc = { 554d46db3d5SWu Fengguang .sync_mode = work->sync_mode, 555d46db3d5SWu Fengguang .tagged_writepages = work->tagged_writepages, 556d46db3d5SWu Fengguang .for_kupdate = work->for_kupdate, 557d46db3d5SWu Fengguang .for_background = work->for_background, 558d46db3d5SWu Fengguang .range_cyclic = work->range_cyclic, 559d46db3d5SWu Fengguang .range_start = 0, 560d46db3d5SWu Fengguang .range_end = LLONG_MAX, 561d46db3d5SWu Fengguang }; 562d46db3d5SWu Fengguang unsigned long start_time = jiffies; 563d46db3d5SWu Fengguang long write_chunk; 564d46db3d5SWu Fengguang long wrote = 0; /* count both pages and inodes */ 565d46db3d5SWu Fengguang 56603ba3782SJens Axboe while (!list_empty(&wb->b_io)) { 5677ccf19a8SNick Piggin struct inode *inode = wb_inode(wb->b_io.prev); 568edadfb10SChristoph Hellwig 569edadfb10SChristoph Hellwig if (inode->i_sb != sb) { 570d46db3d5SWu Fengguang if (work->sb) { 571edadfb10SChristoph Hellwig /* 572edadfb10SChristoph Hellwig * We only want to write back data for this 573edadfb10SChristoph Hellwig * superblock, move all inodes not belonging 574edadfb10SChristoph Hellwig * to it back onto the dirty list. 575edadfb10SChristoph Hellwig */ 576f758eeabSChristoph Hellwig redirty_tail(inode, wb); 57766f3b8e2SJens Axboe continue; 57866f3b8e2SJens Axboe } 579edadfb10SChristoph Hellwig 580edadfb10SChristoph Hellwig /* 581edadfb10SChristoph Hellwig * The inode belongs to a different superblock. 582edadfb10SChristoph Hellwig * Bounce back to the caller to unpin this and 583edadfb10SChristoph Hellwig * pin the next superblock. 584edadfb10SChristoph Hellwig */ 585d46db3d5SWu Fengguang break; 586edadfb10SChristoph Hellwig } 587edadfb10SChristoph Hellwig 5889843b76aSChristoph Hellwig /* 5899843b76aSChristoph Hellwig * Don't bother with new inodes or inodes beeing freed, first 5909843b76aSChristoph Hellwig * kind does not need peridic writeout yet, and for the latter 5919843b76aSChristoph Hellwig * kind writeout is handled by the freer. 5929843b76aSChristoph Hellwig */ 593250df6edSDave Chinner spin_lock(&inode->i_lock); 5949843b76aSChristoph Hellwig if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { 595250df6edSDave Chinner spin_unlock(&inode->i_lock); 596fcc5c222SWu Fengguang redirty_tail(inode, wb); 5977ef0d737SNick Piggin continue; 5987ef0d737SNick Piggin } 5991da177e4SLinus Torvalds __iget(inode); 6001a12d8bdSWu Fengguang write_chunk = writeback_chunk_size(wb->bdi, work); 601d46db3d5SWu Fengguang wbc.nr_to_write = write_chunk; 602d46db3d5SWu Fengguang wbc.pages_skipped = 0; 603250df6edSDave Chinner 604d46db3d5SWu Fengguang writeback_single_inode(inode, wb, &wbc); 605d46db3d5SWu Fengguang 606d46db3d5SWu Fengguang work->nr_pages -= write_chunk - wbc.nr_to_write; 607d46db3d5SWu Fengguang wrote += write_chunk - wbc.nr_to_write; 608d46db3d5SWu Fengguang if (!(inode->i_state & I_DIRTY)) 609d46db3d5SWu Fengguang wrote++; 610d46db3d5SWu Fengguang if (wbc.pages_skipped) { 6111da177e4SLinus Torvalds /* 6121da177e4SLinus Torvalds * writeback is not making progress due to locked 6131da177e4SLinus Torvalds * buffers. Skip this inode for now. 6141da177e4SLinus Torvalds */ 615f758eeabSChristoph Hellwig redirty_tail(inode, wb); 6161da177e4SLinus Torvalds } 6170f1b1fd8SDave Chinner spin_unlock(&inode->i_lock); 618f758eeabSChristoph Hellwig spin_unlock(&wb->list_lock); 6191da177e4SLinus Torvalds iput(inode); 6204ffc8444SOGAWA Hirofumi cond_resched(); 621f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 622d46db3d5SWu Fengguang /* 623d46db3d5SWu Fengguang * bail out to wb_writeback() often enough to check 624d46db3d5SWu Fengguang * background threshold and other termination conditions. 625d46db3d5SWu Fengguang */ 626d46db3d5SWu Fengguang if (wrote) { 627d46db3d5SWu Fengguang if (time_is_before_jiffies(start_time + HZ / 10UL)) 628d46db3d5SWu Fengguang break; 629d46db3d5SWu Fengguang if (work->nr_pages <= 0) 630d46db3d5SWu Fengguang break; 6311da177e4SLinus Torvalds } 632d46db3d5SWu Fengguang } 633d46db3d5SWu Fengguang return wrote; 634f11c9c5cSEdward Shishkin } 63538f21977SNick Piggin 636d46db3d5SWu Fengguang static long __writeback_inodes_wb(struct bdi_writeback *wb, 637d46db3d5SWu Fengguang struct wb_writeback_work *work) 638f11c9c5cSEdward Shishkin { 639d46db3d5SWu Fengguang unsigned long start_time = jiffies; 640d46db3d5SWu Fengguang long wrote = 0; 6419ecc2738SJens Axboe 642f11c9c5cSEdward Shishkin while (!list_empty(&wb->b_io)) { 6437ccf19a8SNick Piggin struct inode *inode = wb_inode(wb->b_io.prev); 644f11c9c5cSEdward Shishkin struct super_block *sb = inode->i_sb; 645f11c9c5cSEdward Shishkin 646334132aeSChristoph Hellwig if (!pin_sb_for_writeback(sb)) { 647f758eeabSChristoph Hellwig requeue_io(inode, wb); 648d19de7edSChristoph Hellwig continue; 649334132aeSChristoph Hellwig } 650d46db3d5SWu Fengguang wrote += writeback_sb_inodes(sb, wb, work); 651d19de7edSChristoph Hellwig drop_super(sb); 652f11c9c5cSEdward Shishkin 653d46db3d5SWu Fengguang /* refer to the same tests at the end of writeback_sb_inodes */ 654d46db3d5SWu Fengguang if (wrote) { 655d46db3d5SWu Fengguang if (time_is_before_jiffies(start_time + HZ / 10UL)) 656d46db3d5SWu Fengguang break; 657d46db3d5SWu Fengguang if (work->nr_pages <= 0) 658f11c9c5cSEdward Shishkin break; 659f11c9c5cSEdward Shishkin } 660d46db3d5SWu Fengguang } 66166f3b8e2SJens Axboe /* Leave any unwritten inodes on b_io */ 662d46db3d5SWu Fengguang return wrote; 66366f3b8e2SJens Axboe } 66466f3b8e2SJens Axboe 665d46db3d5SWu Fengguang long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages) 666edadfb10SChristoph Hellwig { 667d46db3d5SWu Fengguang struct wb_writeback_work work = { 668d46db3d5SWu Fengguang .nr_pages = nr_pages, 669d46db3d5SWu Fengguang .sync_mode = WB_SYNC_NONE, 670d46db3d5SWu Fengguang .range_cyclic = 1, 671d46db3d5SWu Fengguang }; 672d46db3d5SWu Fengguang 673f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 674424b351fSWu Fengguang if (list_empty(&wb->b_io)) 675d46db3d5SWu Fengguang queue_io(wb, NULL); 676d46db3d5SWu Fengguang __writeback_inodes_wb(wb, &work); 677f758eeabSChristoph Hellwig spin_unlock(&wb->list_lock); 678edadfb10SChristoph Hellwig 679d46db3d5SWu Fengguang return nr_pages - work.nr_pages; 680d46db3d5SWu Fengguang } 68103ba3782SJens Axboe 68203ba3782SJens Axboe static inline bool over_bground_thresh(void) 68303ba3782SJens Axboe { 68403ba3782SJens Axboe unsigned long background_thresh, dirty_thresh; 68503ba3782SJens Axboe 68616c4042fSWu Fengguang global_dirty_limits(&background_thresh, &dirty_thresh); 68703ba3782SJens Axboe 68803ba3782SJens Axboe return (global_page_state(NR_FILE_DIRTY) + 6894cbec4c8SWu Fengguang global_page_state(NR_UNSTABLE_NFS) > background_thresh); 69003ba3782SJens Axboe } 69103ba3782SJens Axboe 69203ba3782SJens Axboe /* 693e98be2d5SWu Fengguang * Called under wb->list_lock. If there are multiple wb per bdi, 694e98be2d5SWu Fengguang * only the flusher working on the first wb should do it. 695e98be2d5SWu Fengguang */ 696e98be2d5SWu Fengguang static void wb_update_bandwidth(struct bdi_writeback *wb, 697e98be2d5SWu Fengguang unsigned long start_time) 698e98be2d5SWu Fengguang { 699c42843f2SWu Fengguang __bdi_update_bandwidth(wb->bdi, 0, 0, 0, 0, start_time); 700e98be2d5SWu Fengguang } 701e98be2d5SWu Fengguang 702e98be2d5SWu Fengguang /* 70303ba3782SJens Axboe * Explicit flushing or periodic writeback of "old" data. 70403ba3782SJens Axboe * 70503ba3782SJens Axboe * Define "old": the first time one of an inode's pages is dirtied, we mark the 70603ba3782SJens Axboe * dirtying-time in the inode's address_space. So this periodic writeback code 70703ba3782SJens Axboe * just walks the superblock inode list, writing back any inodes which are 70803ba3782SJens Axboe * older than a specific point in time. 70903ba3782SJens Axboe * 71003ba3782SJens Axboe * Try to run once per dirty_writeback_interval. But if a writeback event 71103ba3782SJens Axboe * takes longer than a dirty_writeback_interval interval, then leave a 71203ba3782SJens Axboe * one-second gap. 71303ba3782SJens Axboe * 71403ba3782SJens Axboe * older_than_this takes precedence over nr_to_write. So we'll only write back 71503ba3782SJens Axboe * all dirty pages if they are all attached to "old" mappings. 71603ba3782SJens Axboe */ 717c4a77a6cSJens Axboe static long wb_writeback(struct bdi_writeback *wb, 71883ba7b07SChristoph Hellwig struct wb_writeback_work *work) 71903ba3782SJens Axboe { 720e98be2d5SWu Fengguang unsigned long wb_start = jiffies; 721d46db3d5SWu Fengguang long nr_pages = work->nr_pages; 72203ba3782SJens Axboe unsigned long oldest_jif; 723a5989bdcSJan Kara struct inode *inode; 724d46db3d5SWu Fengguang long progress; 725b9543dacSJan Kara 726e185dda8SWu Fengguang oldest_jif = jiffies; 727d46db3d5SWu Fengguang work->older_than_this = &oldest_jif; 728e185dda8SWu Fengguang 729e8dfc305SWu Fengguang spin_lock(&wb->list_lock); 73003ba3782SJens Axboe for (;;) { 73103ba3782SJens Axboe /* 732d3ddec76SWu Fengguang * Stop writeback when nr_pages has been consumed 73303ba3782SJens Axboe */ 73483ba7b07SChristoph Hellwig if (work->nr_pages <= 0) 73503ba3782SJens Axboe break; 73603ba3782SJens Axboe 73703ba3782SJens Axboe /* 738aa373cf5SJan Kara * Background writeout and kupdate-style writeback may 739aa373cf5SJan Kara * run forever. Stop them if there is other work to do 740aa373cf5SJan Kara * so that e.g. sync can proceed. They'll be restarted 741aa373cf5SJan Kara * after the other works are all done. 742aa373cf5SJan Kara */ 743aa373cf5SJan Kara if ((work->for_background || work->for_kupdate) && 744aa373cf5SJan Kara !list_empty(&wb->bdi->work_list)) 745aa373cf5SJan Kara break; 746aa373cf5SJan Kara 747aa373cf5SJan Kara /* 748d3ddec76SWu Fengguang * For background writeout, stop when we are below the 749d3ddec76SWu Fengguang * background dirty threshold 75003ba3782SJens Axboe */ 75183ba7b07SChristoph Hellwig if (work->for_background && !over_bground_thresh()) 75203ba3782SJens Axboe break; 75303ba3782SJens Axboe 754ba9aa839SWu Fengguang if (work->for_kupdate) { 755ba9aa839SWu Fengguang oldest_jif = jiffies - 756ba9aa839SWu Fengguang msecs_to_jiffies(dirty_expire_interval * 10); 757d46db3d5SWu Fengguang work->older_than_this = &oldest_jif; 758ba9aa839SWu Fengguang } 759ba9aa839SWu Fengguang 760d46db3d5SWu Fengguang trace_writeback_start(wb->bdi, work); 761e8dfc305SWu Fengguang if (list_empty(&wb->b_io)) 762d46db3d5SWu Fengguang queue_io(wb, work->older_than_this); 76383ba7b07SChristoph Hellwig if (work->sb) 764d46db3d5SWu Fengguang progress = writeback_sb_inodes(work->sb, wb, work); 765edadfb10SChristoph Hellwig else 766d46db3d5SWu Fengguang progress = __writeback_inodes_wb(wb, work); 767d46db3d5SWu Fengguang trace_writeback_written(wb->bdi, work); 76803ba3782SJens Axboe 769e98be2d5SWu Fengguang wb_update_bandwidth(wb, wb_start); 770e98be2d5SWu Fengguang 77103ba3782SJens Axboe /* 772e6fb6da2SWu Fengguang * Did we write something? Try for more 773e6fb6da2SWu Fengguang * 774e6fb6da2SWu Fengguang * Dirty inodes are moved to b_io for writeback in batches. 775e6fb6da2SWu Fengguang * The completion of the current batch does not necessarily 776e6fb6da2SWu Fengguang * mean the overall work is done. So we keep looping as long 777e6fb6da2SWu Fengguang * as made some progress on cleaning pages or inodes. 77803ba3782SJens Axboe */ 779d46db3d5SWu Fengguang if (progress) 780cb9bd115SWu Fengguang continue; 78171fd05a8SJens Axboe /* 782e6fb6da2SWu Fengguang * No more inodes for IO, bail 78371fd05a8SJens Axboe */ 784b7a2441fSWu Fengguang if (list_empty(&wb->b_more_io)) 78571fd05a8SJens Axboe break; 78671fd05a8SJens Axboe /* 787a5989bdcSJan Kara * Nothing written. Wait for some inode to 788a5989bdcSJan Kara * become available for writeback. Otherwise 789a5989bdcSJan Kara * we'll just busyloop. 790a5989bdcSJan Kara */ 791a5989bdcSJan Kara if (!list_empty(&wb->b_more_io)) { 792d46db3d5SWu Fengguang trace_writeback_wait(wb->bdi, work); 7937ccf19a8SNick Piggin inode = wb_inode(wb->b_more_io.prev); 794250df6edSDave Chinner spin_lock(&inode->i_lock); 795f758eeabSChristoph Hellwig inode_wait_for_writeback(inode, wb); 796250df6edSDave Chinner spin_unlock(&inode->i_lock); 797a5989bdcSJan Kara } 79803ba3782SJens Axboe } 799e8dfc305SWu Fengguang spin_unlock(&wb->list_lock); 80003ba3782SJens Axboe 801d46db3d5SWu Fengguang return nr_pages - work->nr_pages; 80203ba3782SJens Axboe } 80303ba3782SJens Axboe 80403ba3782SJens Axboe /* 80583ba7b07SChristoph Hellwig * Return the next wb_writeback_work struct that hasn't been processed yet. 80603ba3782SJens Axboe */ 80783ba7b07SChristoph Hellwig static struct wb_writeback_work * 80808852b6dSMinchan Kim get_next_work_item(struct backing_dev_info *bdi) 80903ba3782SJens Axboe { 81083ba7b07SChristoph Hellwig struct wb_writeback_work *work = NULL; 81103ba3782SJens Axboe 8126467716aSArtem Bityutskiy spin_lock_bh(&bdi->wb_lock); 81383ba7b07SChristoph Hellwig if (!list_empty(&bdi->work_list)) { 81483ba7b07SChristoph Hellwig work = list_entry(bdi->work_list.next, 81583ba7b07SChristoph Hellwig struct wb_writeback_work, list); 81683ba7b07SChristoph Hellwig list_del_init(&work->list); 81703ba3782SJens Axboe } 8186467716aSArtem Bityutskiy spin_unlock_bh(&bdi->wb_lock); 81983ba7b07SChristoph Hellwig return work; 82003ba3782SJens Axboe } 82103ba3782SJens Axboe 822cdf01dd5SLinus Torvalds /* 823cdf01dd5SLinus Torvalds * Add in the number of potentially dirty inodes, because each inode 824cdf01dd5SLinus Torvalds * write can dirty pagecache in the underlying blockdev. 825cdf01dd5SLinus Torvalds */ 826cdf01dd5SLinus Torvalds static unsigned long get_nr_dirty_pages(void) 827cdf01dd5SLinus Torvalds { 828cdf01dd5SLinus Torvalds return global_page_state(NR_FILE_DIRTY) + 829cdf01dd5SLinus Torvalds global_page_state(NR_UNSTABLE_NFS) + 830cdf01dd5SLinus Torvalds get_nr_dirty_inodes(); 831cdf01dd5SLinus Torvalds } 832cdf01dd5SLinus Torvalds 8336585027aSJan Kara static long wb_check_background_flush(struct bdi_writeback *wb) 8346585027aSJan Kara { 8356585027aSJan Kara if (over_bground_thresh()) { 8366585027aSJan Kara 8376585027aSJan Kara struct wb_writeback_work work = { 8386585027aSJan Kara .nr_pages = LONG_MAX, 8396585027aSJan Kara .sync_mode = WB_SYNC_NONE, 8406585027aSJan Kara .for_background = 1, 8416585027aSJan Kara .range_cyclic = 1, 8426585027aSJan Kara }; 8436585027aSJan Kara 8446585027aSJan Kara return wb_writeback(wb, &work); 8456585027aSJan Kara } 8466585027aSJan Kara 8476585027aSJan Kara return 0; 8486585027aSJan Kara } 8496585027aSJan Kara 85003ba3782SJens Axboe static long wb_check_old_data_flush(struct bdi_writeback *wb) 85103ba3782SJens Axboe { 85203ba3782SJens Axboe unsigned long expired; 85303ba3782SJens Axboe long nr_pages; 85403ba3782SJens Axboe 85569b62d01SJens Axboe /* 85669b62d01SJens Axboe * When set to zero, disable periodic writeback 85769b62d01SJens Axboe */ 85869b62d01SJens Axboe if (!dirty_writeback_interval) 85969b62d01SJens Axboe return 0; 86069b62d01SJens Axboe 86103ba3782SJens Axboe expired = wb->last_old_flush + 86203ba3782SJens Axboe msecs_to_jiffies(dirty_writeback_interval * 10); 86303ba3782SJens Axboe if (time_before(jiffies, expired)) 86403ba3782SJens Axboe return 0; 86503ba3782SJens Axboe 86603ba3782SJens Axboe wb->last_old_flush = jiffies; 867cdf01dd5SLinus Torvalds nr_pages = get_nr_dirty_pages(); 86803ba3782SJens Axboe 869c4a77a6cSJens Axboe if (nr_pages) { 87083ba7b07SChristoph Hellwig struct wb_writeback_work work = { 871c4a77a6cSJens Axboe .nr_pages = nr_pages, 872c4a77a6cSJens Axboe .sync_mode = WB_SYNC_NONE, 873c4a77a6cSJens Axboe .for_kupdate = 1, 874c4a77a6cSJens Axboe .range_cyclic = 1, 875c4a77a6cSJens Axboe }; 876c4a77a6cSJens Axboe 87783ba7b07SChristoph Hellwig return wb_writeback(wb, &work); 878c4a77a6cSJens Axboe } 87903ba3782SJens Axboe 88003ba3782SJens Axboe return 0; 88103ba3782SJens Axboe } 88203ba3782SJens Axboe 88303ba3782SJens Axboe /* 88403ba3782SJens Axboe * Retrieve work items and do the writeback they describe 88503ba3782SJens Axboe */ 88603ba3782SJens Axboe long wb_do_writeback(struct bdi_writeback *wb, int force_wait) 88703ba3782SJens Axboe { 88803ba3782SJens Axboe struct backing_dev_info *bdi = wb->bdi; 88983ba7b07SChristoph Hellwig struct wb_writeback_work *work; 890c4a77a6cSJens Axboe long wrote = 0; 89103ba3782SJens Axboe 89281d73a32SJan Kara set_bit(BDI_writeback_running, &wb->bdi->state); 89308852b6dSMinchan Kim while ((work = get_next_work_item(bdi)) != NULL) { 89403ba3782SJens Axboe /* 89503ba3782SJens Axboe * Override sync mode, in case we must wait for completion 89683ba7b07SChristoph Hellwig * because this thread is exiting now. 89703ba3782SJens Axboe */ 89803ba3782SJens Axboe if (force_wait) 89983ba7b07SChristoph Hellwig work->sync_mode = WB_SYNC_ALL; 90083ba7b07SChristoph Hellwig 901455b2864SDave Chinner trace_writeback_exec(bdi, work); 902455b2864SDave Chinner 90383ba7b07SChristoph Hellwig wrote += wb_writeback(wb, work); 90403ba3782SJens Axboe 90503ba3782SJens Axboe /* 90683ba7b07SChristoph Hellwig * Notify the caller of completion if this is a synchronous 90783ba7b07SChristoph Hellwig * work item, otherwise just free it. 90803ba3782SJens Axboe */ 90983ba7b07SChristoph Hellwig if (work->done) 91083ba7b07SChristoph Hellwig complete(work->done); 91183ba7b07SChristoph Hellwig else 91283ba7b07SChristoph Hellwig kfree(work); 91303ba3782SJens Axboe } 91403ba3782SJens Axboe 91503ba3782SJens Axboe /* 91603ba3782SJens Axboe * Check for periodic writeback, kupdated() style 91703ba3782SJens Axboe */ 91803ba3782SJens Axboe wrote += wb_check_old_data_flush(wb); 9196585027aSJan Kara wrote += wb_check_background_flush(wb); 92081d73a32SJan Kara clear_bit(BDI_writeback_running, &wb->bdi->state); 92103ba3782SJens Axboe 92203ba3782SJens Axboe return wrote; 92303ba3782SJens Axboe } 92403ba3782SJens Axboe 92503ba3782SJens Axboe /* 92603ba3782SJens Axboe * Handle writeback of dirty data for the device backed by this bdi. Also 92703ba3782SJens Axboe * wakes up periodically and does kupdated style flushing. 92803ba3782SJens Axboe */ 92908243900SChristoph Hellwig int bdi_writeback_thread(void *data) 93003ba3782SJens Axboe { 93108243900SChristoph Hellwig struct bdi_writeback *wb = data; 93208243900SChristoph Hellwig struct backing_dev_info *bdi = wb->bdi; 93303ba3782SJens Axboe long pages_written; 93403ba3782SJens Axboe 935766f9164SPeter Zijlstra current->flags |= PF_SWAPWRITE; 93608243900SChristoph Hellwig set_freezable(); 937ecd58403SArtem Bityutskiy wb->last_active = jiffies; 93803ba3782SJens Axboe 93903ba3782SJens Axboe /* 94008243900SChristoph Hellwig * Our parent may run at a different priority, just set us to normal 94103ba3782SJens Axboe */ 94208243900SChristoph Hellwig set_user_nice(current, 0); 94308243900SChristoph Hellwig 944455b2864SDave Chinner trace_writeback_thread_start(bdi); 945455b2864SDave Chinner 94603ba3782SJens Axboe while (!kthread_should_stop()) { 9476467716aSArtem Bityutskiy /* 9486467716aSArtem Bityutskiy * Remove own delayed wake-up timer, since we are already awake 9496467716aSArtem Bityutskiy * and we'll take care of the preriodic write-back. 9506467716aSArtem Bityutskiy */ 9516467716aSArtem Bityutskiy del_timer(&wb->wakeup_timer); 9526467716aSArtem Bityutskiy 95303ba3782SJens Axboe pages_written = wb_do_writeback(wb, 0); 95403ba3782SJens Axboe 955455b2864SDave Chinner trace_writeback_pages_written(pages_written); 956455b2864SDave Chinner 95703ba3782SJens Axboe if (pages_written) 958ecd58403SArtem Bityutskiy wb->last_active = jiffies; 95903ba3782SJens Axboe 960297252c8SArtem Bityutskiy set_current_state(TASK_INTERRUPTIBLE); 961b76b4014SJ. Bruce Fields if (!list_empty(&bdi->work_list) || kthread_should_stop()) { 962297252c8SArtem Bityutskiy __set_current_state(TASK_RUNNING); 963297252c8SArtem Bityutskiy continue; 96403ba3782SJens Axboe } 96503ba3782SJens Axboe 966253c34e9SArtem Bityutskiy if (wb_has_dirty_io(wb) && dirty_writeback_interval) 967fff5b85aSArtem Bityutskiy schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10)); 968253c34e9SArtem Bityutskiy else { 969253c34e9SArtem Bityutskiy /* 970253c34e9SArtem Bityutskiy * We have nothing to do, so can go sleep without any 971253c34e9SArtem Bityutskiy * timeout and save power. When a work is queued or 972253c34e9SArtem Bityutskiy * something is made dirty - we will be woken up. 973253c34e9SArtem Bityutskiy */ 97469b62d01SJens Axboe schedule(); 975f9eadbbdSJens Axboe } 97669b62d01SJens Axboe 97703ba3782SJens Axboe try_to_freeze(); 97803ba3782SJens Axboe } 97903ba3782SJens Axboe 980fff5b85aSArtem Bityutskiy /* Flush any work that raced with us exiting */ 98108243900SChristoph Hellwig if (!list_empty(&bdi->work_list)) 98208243900SChristoph Hellwig wb_do_writeback(wb, 1); 983455b2864SDave Chinner 984455b2864SDave Chinner trace_writeback_thread_stop(bdi); 98503ba3782SJens Axboe return 0; 98603ba3782SJens Axboe } 98703ba3782SJens Axboe 98808243900SChristoph Hellwig 98903ba3782SJens Axboe /* 99003ba3782SJens Axboe * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back 99103ba3782SJens Axboe * the whole world. 99203ba3782SJens Axboe */ 99303ba3782SJens Axboe void wakeup_flusher_threads(long nr_pages) 99403ba3782SJens Axboe { 995b8c2f347SChristoph Hellwig struct backing_dev_info *bdi; 996b8c2f347SChristoph Hellwig 99783ba7b07SChristoph Hellwig if (!nr_pages) { 99883ba7b07SChristoph Hellwig nr_pages = global_page_state(NR_FILE_DIRTY) + 99903ba3782SJens Axboe global_page_state(NR_UNSTABLE_NFS); 1000b8c2f347SChristoph Hellwig } 1001b8c2f347SChristoph Hellwig 1002b8c2f347SChristoph Hellwig rcu_read_lock(); 1003b8c2f347SChristoph Hellwig list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { 1004b8c2f347SChristoph Hellwig if (!bdi_has_dirty_io(bdi)) 1005b8c2f347SChristoph Hellwig continue; 10066585027aSJan Kara __bdi_start_writeback(bdi, nr_pages, false); 1007b8c2f347SChristoph Hellwig } 1008b8c2f347SChristoph Hellwig rcu_read_unlock(); 100903ba3782SJens Axboe } 101003ba3782SJens Axboe 101103ba3782SJens Axboe static noinline void block_dump___mark_inode_dirty(struct inode *inode) 101203ba3782SJens Axboe { 101303ba3782SJens Axboe if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { 101403ba3782SJens Axboe struct dentry *dentry; 101503ba3782SJens Axboe const char *name = "?"; 101603ba3782SJens Axboe 101703ba3782SJens Axboe dentry = d_find_alias(inode); 101803ba3782SJens Axboe if (dentry) { 101903ba3782SJens Axboe spin_lock(&dentry->d_lock); 102003ba3782SJens Axboe name = (const char *) dentry->d_name.name; 102103ba3782SJens Axboe } 102203ba3782SJens Axboe printk(KERN_DEBUG 102303ba3782SJens Axboe "%s(%d): dirtied inode %lu (%s) on %s\n", 102403ba3782SJens Axboe current->comm, task_pid_nr(current), inode->i_ino, 102503ba3782SJens Axboe name, inode->i_sb->s_id); 102603ba3782SJens Axboe if (dentry) { 102703ba3782SJens Axboe spin_unlock(&dentry->d_lock); 102803ba3782SJens Axboe dput(dentry); 102903ba3782SJens Axboe } 103003ba3782SJens Axboe } 103103ba3782SJens Axboe } 103203ba3782SJens Axboe 103303ba3782SJens Axboe /** 103403ba3782SJens Axboe * __mark_inode_dirty - internal function 103503ba3782SJens Axboe * @inode: inode to mark 103603ba3782SJens Axboe * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) 103703ba3782SJens Axboe * Mark an inode as dirty. Callers should use mark_inode_dirty or 103803ba3782SJens Axboe * mark_inode_dirty_sync. 103903ba3782SJens Axboe * 104003ba3782SJens Axboe * Put the inode on the super block's dirty list. 104103ba3782SJens Axboe * 104203ba3782SJens Axboe * CAREFUL! We mark it dirty unconditionally, but move it onto the 104303ba3782SJens Axboe * dirty list only if it is hashed or if it refers to a blockdev. 104403ba3782SJens Axboe * If it was not hashed, it will never be added to the dirty list 104503ba3782SJens Axboe * even if it is later hashed, as it will have been marked dirty already. 104603ba3782SJens Axboe * 104703ba3782SJens Axboe * In short, make sure you hash any inodes _before_ you start marking 104803ba3782SJens Axboe * them dirty. 104903ba3782SJens Axboe * 105003ba3782SJens Axboe * Note that for blockdevs, inode->dirtied_when represents the dirtying time of 105103ba3782SJens Axboe * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of 105203ba3782SJens Axboe * the kernel-internal blockdev inode represents the dirtying time of the 105303ba3782SJens Axboe * blockdev's pages. This is why for I_DIRTY_PAGES we always use 105403ba3782SJens Axboe * page->mapping->host, so the page-dirtying time is recorded in the internal 105503ba3782SJens Axboe * blockdev inode. 105603ba3782SJens Axboe */ 105703ba3782SJens Axboe void __mark_inode_dirty(struct inode *inode, int flags) 105803ba3782SJens Axboe { 105903ba3782SJens Axboe struct super_block *sb = inode->i_sb; 1060253c34e9SArtem Bityutskiy struct backing_dev_info *bdi = NULL; 106103ba3782SJens Axboe 106203ba3782SJens Axboe /* 106303ba3782SJens Axboe * Don't do this for I_DIRTY_PAGES - that doesn't actually 106403ba3782SJens Axboe * dirty the inode itself 106503ba3782SJens Axboe */ 106603ba3782SJens Axboe if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { 106703ba3782SJens Axboe if (sb->s_op->dirty_inode) 1068aa385729SChristoph Hellwig sb->s_op->dirty_inode(inode, flags); 106903ba3782SJens Axboe } 107003ba3782SJens Axboe 107103ba3782SJens Axboe /* 107203ba3782SJens Axboe * make sure that changes are seen by all cpus before we test i_state 107303ba3782SJens Axboe * -- mikulas 107403ba3782SJens Axboe */ 107503ba3782SJens Axboe smp_mb(); 107603ba3782SJens Axboe 107703ba3782SJens Axboe /* avoid the locking if we can */ 107803ba3782SJens Axboe if ((inode->i_state & flags) == flags) 107903ba3782SJens Axboe return; 108003ba3782SJens Axboe 108103ba3782SJens Axboe if (unlikely(block_dump)) 108203ba3782SJens Axboe block_dump___mark_inode_dirty(inode); 108303ba3782SJens Axboe 1084250df6edSDave Chinner spin_lock(&inode->i_lock); 108503ba3782SJens Axboe if ((inode->i_state & flags) != flags) { 108603ba3782SJens Axboe const int was_dirty = inode->i_state & I_DIRTY; 108703ba3782SJens Axboe 108803ba3782SJens Axboe inode->i_state |= flags; 108903ba3782SJens Axboe 109003ba3782SJens Axboe /* 109103ba3782SJens Axboe * If the inode is being synced, just update its dirty state. 109203ba3782SJens Axboe * The unlocker will place the inode on the appropriate 109303ba3782SJens Axboe * superblock list, based upon its state. 109403ba3782SJens Axboe */ 109503ba3782SJens Axboe if (inode->i_state & I_SYNC) 1096250df6edSDave Chinner goto out_unlock_inode; 109703ba3782SJens Axboe 109803ba3782SJens Axboe /* 109903ba3782SJens Axboe * Only add valid (hashed) inodes to the superblock's 110003ba3782SJens Axboe * dirty list. Add blockdev inodes as well. 110103ba3782SJens Axboe */ 110203ba3782SJens Axboe if (!S_ISBLK(inode->i_mode)) { 11031d3382cbSAl Viro if (inode_unhashed(inode)) 1104250df6edSDave Chinner goto out_unlock_inode; 110503ba3782SJens Axboe } 1106a4ffdde6SAl Viro if (inode->i_state & I_FREEING) 1107250df6edSDave Chinner goto out_unlock_inode; 110803ba3782SJens Axboe 110903ba3782SJens Axboe /* 111003ba3782SJens Axboe * If the inode was already on b_dirty/b_io/b_more_io, don't 111103ba3782SJens Axboe * reposition it (that would break b_dirty time-ordering). 111203ba3782SJens Axboe */ 111303ba3782SJens Axboe if (!was_dirty) { 1114a66979abSDave Chinner bool wakeup_bdi = false; 1115253c34e9SArtem Bityutskiy bdi = inode_to_bdi(inode); 1116500b067cSJens Axboe 1117253c34e9SArtem Bityutskiy if (bdi_cap_writeback_dirty(bdi)) { 1118253c34e9SArtem Bityutskiy WARN(!test_bit(BDI_registered, &bdi->state), 1119253c34e9SArtem Bityutskiy "bdi-%s not registered\n", bdi->name); 1120253c34e9SArtem Bityutskiy 1121253c34e9SArtem Bityutskiy /* 1122253c34e9SArtem Bityutskiy * If this is the first dirty inode for this 1123253c34e9SArtem Bityutskiy * bdi, we have to wake-up the corresponding 1124253c34e9SArtem Bityutskiy * bdi thread to make sure background 1125253c34e9SArtem Bityutskiy * write-back happens later. 1126253c34e9SArtem Bityutskiy */ 1127253c34e9SArtem Bityutskiy if (!wb_has_dirty_io(&bdi->wb)) 1128253c34e9SArtem Bityutskiy wakeup_bdi = true; 1129500b067cSJens Axboe } 113003ba3782SJens Axboe 1131a66979abSDave Chinner spin_unlock(&inode->i_lock); 1132f758eeabSChristoph Hellwig spin_lock(&bdi->wb.list_lock); 113303ba3782SJens Axboe inode->dirtied_when = jiffies; 11347ccf19a8SNick Piggin list_move(&inode->i_wb_list, &bdi->wb.b_dirty); 1135f758eeabSChristoph Hellwig spin_unlock(&bdi->wb.list_lock); 1136253c34e9SArtem Bityutskiy 1137253c34e9SArtem Bityutskiy if (wakeup_bdi) 11386467716aSArtem Bityutskiy bdi_wakeup_thread_delayed(bdi); 1139a66979abSDave Chinner return; 1140a66979abSDave Chinner } 1141a66979abSDave Chinner } 1142a66979abSDave Chinner out_unlock_inode: 1143a66979abSDave Chinner spin_unlock(&inode->i_lock); 1144a66979abSDave Chinner 114503ba3782SJens Axboe } 114603ba3782SJens Axboe EXPORT_SYMBOL(__mark_inode_dirty); 114703ba3782SJens Axboe 114866f3b8e2SJens Axboe /* 114966f3b8e2SJens Axboe * Write out a superblock's list of dirty inodes. A wait will be performed 115066f3b8e2SJens Axboe * upon no inodes, all inodes or the final one, depending upon sync_mode. 115166f3b8e2SJens Axboe * 115266f3b8e2SJens Axboe * If older_than_this is non-NULL, then only write out inodes which 115366f3b8e2SJens Axboe * had their first dirtying at a time earlier than *older_than_this. 115466f3b8e2SJens Axboe * 115566f3b8e2SJens Axboe * If `bdi' is non-zero then we're being asked to writeback a specific queue. 115666f3b8e2SJens Axboe * This function assumes that the blockdev superblock's inodes are backed by 115766f3b8e2SJens Axboe * a variety of queues, so all inodes are searched. For other superblocks, 115866f3b8e2SJens Axboe * assume that all inodes are backed by the same queue. 115966f3b8e2SJens Axboe * 116066f3b8e2SJens Axboe * The inodes to be written are parked on bdi->b_io. They are moved back onto 116166f3b8e2SJens Axboe * bdi->b_dirty as they are selected for writing. This way, none can be missed 116266f3b8e2SJens Axboe * on the writer throttling path, and we get decent balancing between many 116366f3b8e2SJens Axboe * throttled threads: we don't want them all piling up on inode_sync_wait. 116466f3b8e2SJens Axboe */ 1165b6e51316SJens Axboe static void wait_sb_inodes(struct super_block *sb) 116666f3b8e2SJens Axboe { 116738f21977SNick Piggin struct inode *inode, *old_inode = NULL; 116838f21977SNick Piggin 116903ba3782SJens Axboe /* 117003ba3782SJens Axboe * We need to be protected against the filesystem going from 117103ba3782SJens Axboe * r/o to r/w or vice versa. 117203ba3782SJens Axboe */ 1173b6e51316SJens Axboe WARN_ON(!rwsem_is_locked(&sb->s_umount)); 117403ba3782SJens Axboe 117555fa6091SDave Chinner spin_lock(&inode_sb_list_lock); 117666f3b8e2SJens Axboe 117738f21977SNick Piggin /* 117838f21977SNick Piggin * Data integrity sync. Must wait for all pages under writeback, 117938f21977SNick Piggin * because there may have been pages dirtied before our sync 118038f21977SNick Piggin * call, but which had writeout started before we write it out. 118138f21977SNick Piggin * In which case, the inode may not be on the dirty list, but 118238f21977SNick Piggin * we still have to wait for that writeout. 118338f21977SNick Piggin */ 1184b6e51316SJens Axboe list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { 1185250df6edSDave Chinner struct address_space *mapping = inode->i_mapping; 118638f21977SNick Piggin 1187250df6edSDave Chinner spin_lock(&inode->i_lock); 1188250df6edSDave Chinner if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) || 1189250df6edSDave Chinner (mapping->nrpages == 0)) { 1190250df6edSDave Chinner spin_unlock(&inode->i_lock); 119138f21977SNick Piggin continue; 1192250df6edSDave Chinner } 119338f21977SNick Piggin __iget(inode); 1194250df6edSDave Chinner spin_unlock(&inode->i_lock); 119555fa6091SDave Chinner spin_unlock(&inode_sb_list_lock); 119655fa6091SDave Chinner 119738f21977SNick Piggin /* 119855fa6091SDave Chinner * We hold a reference to 'inode' so it couldn't have been 119955fa6091SDave Chinner * removed from s_inodes list while we dropped the 120055fa6091SDave Chinner * inode_sb_list_lock. We cannot iput the inode now as we can 120155fa6091SDave Chinner * be holding the last reference and we cannot iput it under 120255fa6091SDave Chinner * inode_sb_list_lock. So we keep the reference and iput it 120355fa6091SDave Chinner * later. 120438f21977SNick Piggin */ 120538f21977SNick Piggin iput(old_inode); 120638f21977SNick Piggin old_inode = inode; 120738f21977SNick Piggin 120838f21977SNick Piggin filemap_fdatawait(mapping); 120938f21977SNick Piggin 121038f21977SNick Piggin cond_resched(); 121138f21977SNick Piggin 121255fa6091SDave Chinner spin_lock(&inode_sb_list_lock); 121338f21977SNick Piggin } 121455fa6091SDave Chinner spin_unlock(&inode_sb_list_lock); 121538f21977SNick Piggin iput(old_inode); 121666f3b8e2SJens Axboe } 12171da177e4SLinus Torvalds 1218d8a8559cSJens Axboe /** 12193259f8beSChris Mason * writeback_inodes_sb_nr - writeback dirty inodes from given super_block 1220d8a8559cSJens Axboe * @sb: the superblock 12213259f8beSChris Mason * @nr: the number of pages to write 12221da177e4SLinus Torvalds * 1223d8a8559cSJens Axboe * Start writeback on some inodes on this super_block. No guarantees are made 1224d8a8559cSJens Axboe * on how many (if any) will be written, and this function does not wait 12253259f8beSChris Mason * for IO completion of submitted IO. 12261da177e4SLinus Torvalds */ 12273259f8beSChris Mason void writeback_inodes_sb_nr(struct super_block *sb, unsigned long nr) 12281da177e4SLinus Torvalds { 122983ba7b07SChristoph Hellwig DECLARE_COMPLETION_ONSTACK(done); 123083ba7b07SChristoph Hellwig struct wb_writeback_work work = { 12313c4d7165SChristoph Hellwig .sb = sb, 12323c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_NONE, 12336e6938b6SWu Fengguang .tagged_writepages = 1, 123483ba7b07SChristoph Hellwig .done = &done, 12353259f8beSChris Mason .nr_pages = nr, 12363c4d7165SChristoph Hellwig }; 12370e3c9a22SJens Axboe 1238cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 123983ba7b07SChristoph Hellwig bdi_queue_work(sb->s_bdi, &work); 124083ba7b07SChristoph Hellwig wait_for_completion(&done); 12411da177e4SLinus Torvalds } 12423259f8beSChris Mason EXPORT_SYMBOL(writeback_inodes_sb_nr); 12433259f8beSChris Mason 12443259f8beSChris Mason /** 12453259f8beSChris Mason * writeback_inodes_sb - writeback dirty inodes from given super_block 12463259f8beSChris Mason * @sb: the superblock 12473259f8beSChris Mason * 12483259f8beSChris Mason * Start writeback on some inodes on this super_block. No guarantees are made 12493259f8beSChris Mason * on how many (if any) will be written, and this function does not wait 12503259f8beSChris Mason * for IO completion of submitted IO. 12513259f8beSChris Mason */ 12523259f8beSChris Mason void writeback_inodes_sb(struct super_block *sb) 12533259f8beSChris Mason { 1254925d169fSLinus Torvalds return writeback_inodes_sb_nr(sb, get_nr_dirty_pages()); 12553259f8beSChris Mason } 1256d8a8559cSJens Axboe EXPORT_SYMBOL(writeback_inodes_sb); 1257d8a8559cSJens Axboe 1258d8a8559cSJens Axboe /** 125917bd55d0SEric Sandeen * writeback_inodes_sb_if_idle - start writeback if none underway 126017bd55d0SEric Sandeen * @sb: the superblock 126117bd55d0SEric Sandeen * 126217bd55d0SEric Sandeen * Invoke writeback_inodes_sb if no writeback is currently underway. 126317bd55d0SEric Sandeen * Returns 1 if writeback was started, 0 if not. 126417bd55d0SEric Sandeen */ 126517bd55d0SEric Sandeen int writeback_inodes_sb_if_idle(struct super_block *sb) 126617bd55d0SEric Sandeen { 126717bd55d0SEric Sandeen if (!writeback_in_progress(sb->s_bdi)) { 1268cf37e972SChristoph Hellwig down_read(&sb->s_umount); 126917bd55d0SEric Sandeen writeback_inodes_sb(sb); 1270cf37e972SChristoph Hellwig up_read(&sb->s_umount); 127117bd55d0SEric Sandeen return 1; 127217bd55d0SEric Sandeen } else 127317bd55d0SEric Sandeen return 0; 127417bd55d0SEric Sandeen } 127517bd55d0SEric Sandeen EXPORT_SYMBOL(writeback_inodes_sb_if_idle); 127617bd55d0SEric Sandeen 127717bd55d0SEric Sandeen /** 12783259f8beSChris Mason * writeback_inodes_sb_if_idle - start writeback if none underway 12793259f8beSChris Mason * @sb: the superblock 12803259f8beSChris Mason * @nr: the number of pages to write 12813259f8beSChris Mason * 12823259f8beSChris Mason * Invoke writeback_inodes_sb if no writeback is currently underway. 12833259f8beSChris Mason * Returns 1 if writeback was started, 0 if not. 12843259f8beSChris Mason */ 12853259f8beSChris Mason int writeback_inodes_sb_nr_if_idle(struct super_block *sb, 12863259f8beSChris Mason unsigned long nr) 12873259f8beSChris Mason { 12883259f8beSChris Mason if (!writeback_in_progress(sb->s_bdi)) { 12893259f8beSChris Mason down_read(&sb->s_umount); 12903259f8beSChris Mason writeback_inodes_sb_nr(sb, nr); 12913259f8beSChris Mason up_read(&sb->s_umount); 12923259f8beSChris Mason return 1; 12933259f8beSChris Mason } else 12943259f8beSChris Mason return 0; 12953259f8beSChris Mason } 12963259f8beSChris Mason EXPORT_SYMBOL(writeback_inodes_sb_nr_if_idle); 12973259f8beSChris Mason 12983259f8beSChris Mason /** 1299d8a8559cSJens Axboe * sync_inodes_sb - sync sb inode pages 1300d8a8559cSJens Axboe * @sb: the superblock 1301d8a8559cSJens Axboe * 1302d8a8559cSJens Axboe * This function writes and waits on any dirty inode belonging to this 1303cb9ef8d5SStefan Hajnoczi * super_block. 1304d8a8559cSJens Axboe */ 1305b6e51316SJens Axboe void sync_inodes_sb(struct super_block *sb) 1306d8a8559cSJens Axboe { 130783ba7b07SChristoph Hellwig DECLARE_COMPLETION_ONSTACK(done); 130883ba7b07SChristoph Hellwig struct wb_writeback_work work = { 13093c4d7165SChristoph Hellwig .sb = sb, 13103c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_ALL, 13113c4d7165SChristoph Hellwig .nr_pages = LONG_MAX, 13123c4d7165SChristoph Hellwig .range_cyclic = 0, 131383ba7b07SChristoph Hellwig .done = &done, 13143c4d7165SChristoph Hellwig }; 13153c4d7165SChristoph Hellwig 1316cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 1317cf37e972SChristoph Hellwig 131883ba7b07SChristoph Hellwig bdi_queue_work(sb->s_bdi, &work); 131983ba7b07SChristoph Hellwig wait_for_completion(&done); 132083ba7b07SChristoph Hellwig 1321b6e51316SJens Axboe wait_sb_inodes(sb); 1322d8a8559cSJens Axboe } 1323d8a8559cSJens Axboe EXPORT_SYMBOL(sync_inodes_sb); 13241da177e4SLinus Torvalds 13251da177e4SLinus Torvalds /** 13261da177e4SLinus Torvalds * write_inode_now - write an inode to disk 13271da177e4SLinus Torvalds * @inode: inode to write to disk 13281da177e4SLinus Torvalds * @sync: whether the write should be synchronous or not 13291da177e4SLinus Torvalds * 13307f04c26dSAndrea Arcangeli * This function commits an inode to disk immediately if it is dirty. This is 13317f04c26dSAndrea Arcangeli * primarily needed by knfsd. 13327f04c26dSAndrea Arcangeli * 13337f04c26dSAndrea Arcangeli * The caller must either have a ref on the inode or must have set I_WILL_FREE. 13341da177e4SLinus Torvalds */ 13351da177e4SLinus Torvalds int write_inode_now(struct inode *inode, int sync) 13361da177e4SLinus Torvalds { 1337f758eeabSChristoph Hellwig struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; 13381da177e4SLinus Torvalds int ret; 13391da177e4SLinus Torvalds struct writeback_control wbc = { 13401da177e4SLinus Torvalds .nr_to_write = LONG_MAX, 134118914b18SMike Galbraith .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, 1342111ebb6eSOGAWA Hirofumi .range_start = 0, 1343111ebb6eSOGAWA Hirofumi .range_end = LLONG_MAX, 13441da177e4SLinus Torvalds }; 13451da177e4SLinus Torvalds 13461da177e4SLinus Torvalds if (!mapping_cap_writeback_dirty(inode->i_mapping)) 134749364ce2SAndrew Morton wbc.nr_to_write = 0; 13481da177e4SLinus Torvalds 13491da177e4SLinus Torvalds might_sleep(); 1350f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 13510f1b1fd8SDave Chinner spin_lock(&inode->i_lock); 1352f758eeabSChristoph Hellwig ret = writeback_single_inode(inode, wb, &wbc); 13530f1b1fd8SDave Chinner spin_unlock(&inode->i_lock); 1354f758eeabSChristoph Hellwig spin_unlock(&wb->list_lock); 13551da177e4SLinus Torvalds if (sync) 13561c0eeaf5SJoern Engel inode_sync_wait(inode); 13571da177e4SLinus Torvalds return ret; 13581da177e4SLinus Torvalds } 13591da177e4SLinus Torvalds EXPORT_SYMBOL(write_inode_now); 13601da177e4SLinus Torvalds 13611da177e4SLinus Torvalds /** 13621da177e4SLinus Torvalds * sync_inode - write an inode and its pages to disk. 13631da177e4SLinus Torvalds * @inode: the inode to sync 13641da177e4SLinus Torvalds * @wbc: controls the writeback mode 13651da177e4SLinus Torvalds * 13661da177e4SLinus Torvalds * sync_inode() will write an inode and its pages to disk. It will also 13671da177e4SLinus Torvalds * correctly update the inode on its superblock's dirty inode lists and will 13681da177e4SLinus Torvalds * update inode->i_state. 13691da177e4SLinus Torvalds * 13701da177e4SLinus Torvalds * The caller must have a ref on the inode. 13711da177e4SLinus Torvalds */ 13721da177e4SLinus Torvalds int sync_inode(struct inode *inode, struct writeback_control *wbc) 13731da177e4SLinus Torvalds { 1374f758eeabSChristoph Hellwig struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; 13751da177e4SLinus Torvalds int ret; 13761da177e4SLinus Torvalds 1377f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 13780f1b1fd8SDave Chinner spin_lock(&inode->i_lock); 1379f758eeabSChristoph Hellwig ret = writeback_single_inode(inode, wb, wbc); 13800f1b1fd8SDave Chinner spin_unlock(&inode->i_lock); 1381f758eeabSChristoph Hellwig spin_unlock(&wb->list_lock); 13821da177e4SLinus Torvalds return ret; 13831da177e4SLinus Torvalds } 13841da177e4SLinus Torvalds EXPORT_SYMBOL(sync_inode); 1385c3765016SChristoph Hellwig 1386c3765016SChristoph Hellwig /** 1387c691b9d9SAndrew Morton * sync_inode_metadata - write an inode to disk 1388c3765016SChristoph Hellwig * @inode: the inode to sync 1389c3765016SChristoph Hellwig * @wait: wait for I/O to complete. 1390c3765016SChristoph Hellwig * 1391c691b9d9SAndrew Morton * Write an inode to disk and adjust its dirty state after completion. 1392c3765016SChristoph Hellwig * 1393c3765016SChristoph Hellwig * Note: only writes the actual inode, no associated data or other metadata. 1394c3765016SChristoph Hellwig */ 1395c3765016SChristoph Hellwig int sync_inode_metadata(struct inode *inode, int wait) 1396c3765016SChristoph Hellwig { 1397c3765016SChristoph Hellwig struct writeback_control wbc = { 1398c3765016SChristoph Hellwig .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE, 1399c3765016SChristoph Hellwig .nr_to_write = 0, /* metadata-only */ 1400c3765016SChristoph Hellwig }; 1401c3765016SChristoph Hellwig 1402c3765016SChristoph Hellwig return sync_inode(inode, &wbc); 1403c3765016SChristoph Hellwig } 1404c3765016SChristoph Hellwig EXPORT_SYMBOL(sync_inode_metadata); 1405