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> 2503ba3782SJens Axboe #include <linux/freezer.h> 261da177e4SLinus Torvalds #include <linux/writeback.h> 271da177e4SLinus Torvalds #include <linux/blkdev.h> 281da177e4SLinus Torvalds #include <linux/backing-dev.h> 29455b2864SDave Chinner #include <linux/tracepoint.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; 43d46db3d5SWu Fengguang unsigned long *older_than_this; 44c4a77a6cSJens Axboe enum writeback_sync_modes sync_mode; 456e6938b6SWu Fengguang unsigned int tagged_writepages:1; 4652957fe1SH Hartley Sweeten unsigned int for_kupdate:1; 4752957fe1SH Hartley Sweeten unsigned int range_cyclic:1; 4852957fe1SH Hartley Sweeten unsigned int for_background:1; 490e175a18SCurt Wohlgemuth enum wb_reason reason; /* why was writeback initiated? */ 50c4a77a6cSJens Axboe 518010c3b6SJens Axboe struct list_head list; /* pending work list */ 5283ba7b07SChristoph Hellwig struct completion *done; /* set if the caller waits */ 5303ba3782SJens Axboe }; 5403ba3782SJens Axboe 55455b2864SDave Chinner /* 56455b2864SDave Chinner * We don't actually have pdflush, but this one is exported though /proc... 57455b2864SDave Chinner */ 58455b2864SDave Chinner int nr_pdflush_threads; 59455b2864SDave Chinner 60f11b00f3SAdrian Bunk /** 61f11b00f3SAdrian Bunk * writeback_in_progress - determine whether there is writeback in progress 62f11b00f3SAdrian Bunk * @bdi: the device's backing_dev_info structure. 63f11b00f3SAdrian Bunk * 6403ba3782SJens Axboe * Determine whether there is writeback waiting to be handled against a 6503ba3782SJens Axboe * backing device. 66f11b00f3SAdrian Bunk */ 67f11b00f3SAdrian Bunk int writeback_in_progress(struct backing_dev_info *bdi) 68f11b00f3SAdrian Bunk { 6981d73a32SJan Kara return test_bit(BDI_writeback_running, &bdi->state); 70f11b00f3SAdrian Bunk } 71f11b00f3SAdrian Bunk 72692ebd17SJan Kara static inline struct backing_dev_info *inode_to_bdi(struct inode *inode) 73692ebd17SJan Kara { 74692ebd17SJan Kara struct super_block *sb = inode->i_sb; 75692ebd17SJan Kara 76aaead25bSChristoph Hellwig if (strcmp(sb->s_type->name, "bdev") == 0) 77aaead25bSChristoph Hellwig return inode->i_mapping->backing_dev_info; 78aaead25bSChristoph Hellwig 79692ebd17SJan Kara return sb->s_bdi; 80692ebd17SJan Kara } 81692ebd17SJan Kara 827ccf19a8SNick Piggin static inline struct inode *wb_inode(struct list_head *head) 837ccf19a8SNick Piggin { 847ccf19a8SNick Piggin return list_entry(head, struct inode, i_wb_list); 857ccf19a8SNick Piggin } 867ccf19a8SNick Piggin 8715eb77a0SWu Fengguang /* 8815eb77a0SWu Fengguang * Include the creation of the trace points after defining the 8915eb77a0SWu Fengguang * wb_writeback_work structure and inline functions so that the definition 9015eb77a0SWu Fengguang * remains local to this file. 9115eb77a0SWu Fengguang */ 9215eb77a0SWu Fengguang #define CREATE_TRACE_POINTS 9315eb77a0SWu Fengguang #include <trace/events/writeback.h> 9415eb77a0SWu Fengguang 956585027aSJan Kara /* Wakeup flusher thread or forker thread to fork it. Requires bdi->wb_lock. */ 966585027aSJan Kara static void bdi_wakeup_flusher(struct backing_dev_info *bdi) 974195f73dSNick Piggin { 98fff5b85aSArtem Bityutskiy if (bdi->wb.task) { 99fff5b85aSArtem Bityutskiy wake_up_process(bdi->wb.task); 100fff5b85aSArtem Bityutskiy } else { 1011da177e4SLinus Torvalds /* 102fff5b85aSArtem Bityutskiy * The bdi thread isn't there, wake up the forker thread which 103fff5b85aSArtem Bityutskiy * will create and run it. 1041da177e4SLinus Torvalds */ 10503ba3782SJens Axboe wake_up_process(default_backing_dev_info.wb.task); 1061da177e4SLinus Torvalds } 1076585027aSJan Kara } 1086585027aSJan Kara 1096585027aSJan Kara static void bdi_queue_work(struct backing_dev_info *bdi, 1106585027aSJan Kara struct wb_writeback_work *work) 1116585027aSJan Kara { 1126585027aSJan Kara trace_writeback_queue(bdi, work); 1136585027aSJan Kara 1146585027aSJan Kara spin_lock_bh(&bdi->wb_lock); 1156585027aSJan Kara list_add_tail(&work->list, &bdi->work_list); 1166585027aSJan Kara if (!bdi->wb.task) 1176585027aSJan Kara trace_writeback_nothread(bdi, work); 1186585027aSJan Kara bdi_wakeup_flusher(bdi); 1196467716aSArtem Bityutskiy spin_unlock_bh(&bdi->wb_lock); 12003ba3782SJens Axboe } 1211da177e4SLinus Torvalds 12283ba7b07SChristoph Hellwig static void 12383ba7b07SChristoph Hellwig __bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, 1240e175a18SCurt Wohlgemuth bool range_cyclic, enum wb_reason reason) 1251da177e4SLinus Torvalds { 12683ba7b07SChristoph Hellwig struct wb_writeback_work *work; 12703ba3782SJens Axboe 128bcddc3f0SJens Axboe /* 129bcddc3f0SJens Axboe * This is WB_SYNC_NONE writeback, so if allocation fails just 130bcddc3f0SJens Axboe * wakeup the thread for old dirty data writeback 131bcddc3f0SJens Axboe */ 13283ba7b07SChristoph Hellwig work = kzalloc(sizeof(*work), GFP_ATOMIC); 13383ba7b07SChristoph Hellwig if (!work) { 134455b2864SDave Chinner if (bdi->wb.task) { 135455b2864SDave Chinner trace_writeback_nowork(bdi); 13683ba7b07SChristoph Hellwig wake_up_process(bdi->wb.task); 137455b2864SDave Chinner } 13883ba7b07SChristoph Hellwig return; 13983ba7b07SChristoph Hellwig } 14083ba7b07SChristoph Hellwig 14183ba7b07SChristoph Hellwig work->sync_mode = WB_SYNC_NONE; 14283ba7b07SChristoph Hellwig work->nr_pages = nr_pages; 14383ba7b07SChristoph Hellwig work->range_cyclic = range_cyclic; 1440e175a18SCurt Wohlgemuth work->reason = reason; 14583ba7b07SChristoph Hellwig 146f11fcae8SJens Axboe bdi_queue_work(bdi, work); 14703ba3782SJens Axboe } 148b6e51316SJens Axboe 149b6e51316SJens Axboe /** 150b6e51316SJens Axboe * bdi_start_writeback - start writeback 151b6e51316SJens Axboe * @bdi: the backing device to write from 152b6e51316SJens Axboe * @nr_pages: the number of pages to write 153786228abSMarcos Paulo de Souza * @reason: reason why some writeback work was initiated 154b6e51316SJens Axboe * 155b6e51316SJens Axboe * Description: 156b6e51316SJens Axboe * This does WB_SYNC_NONE opportunistic writeback. The IO is only 15725985edcSLucas De Marchi * started when this function returns, we make no guarantees on 1580e3c9a22SJens Axboe * completion. Caller need not hold sb s_umount semaphore. 159b6e51316SJens Axboe * 160b6e51316SJens Axboe */ 1610e175a18SCurt Wohlgemuth void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, 1620e175a18SCurt Wohlgemuth enum wb_reason reason) 163b6e51316SJens Axboe { 1640e175a18SCurt Wohlgemuth __bdi_start_writeback(bdi, nr_pages, true, reason); 165d3ddec76SWu Fengguang } 166d3ddec76SWu Fengguang 167c5444198SChristoph Hellwig /** 168c5444198SChristoph Hellwig * bdi_start_background_writeback - start background writeback 169c5444198SChristoph Hellwig * @bdi: the backing device to write from 170c5444198SChristoph Hellwig * 171c5444198SChristoph Hellwig * Description: 1726585027aSJan Kara * This makes sure WB_SYNC_NONE background writeback happens. When 1736585027aSJan Kara * this function returns, it is only guaranteed that for given BDI 1746585027aSJan Kara * some IO is happening if we are over background dirty threshold. 1756585027aSJan Kara * Caller need not hold sb s_umount semaphore. 176c5444198SChristoph Hellwig */ 177c5444198SChristoph Hellwig void bdi_start_background_writeback(struct backing_dev_info *bdi) 178c5444198SChristoph Hellwig { 1796585027aSJan Kara /* 1806585027aSJan Kara * We just wake up the flusher thread. It will perform background 1816585027aSJan Kara * writeback as soon as there is no other work to do. 1826585027aSJan Kara */ 18371927e84SWu Fengguang trace_writeback_wake_background(bdi); 1846585027aSJan Kara spin_lock_bh(&bdi->wb_lock); 1856585027aSJan Kara bdi_wakeup_flusher(bdi); 1866585027aSJan Kara spin_unlock_bh(&bdi->wb_lock); 1871da177e4SLinus Torvalds } 1881da177e4SLinus Torvalds 1891da177e4SLinus Torvalds /* 190a66979abSDave Chinner * Remove the inode from the writeback list it is on. 191a66979abSDave Chinner */ 192a66979abSDave Chinner void inode_wb_list_del(struct inode *inode) 193a66979abSDave Chinner { 194f758eeabSChristoph Hellwig struct backing_dev_info *bdi = inode_to_bdi(inode); 195a66979abSDave Chinner 196f758eeabSChristoph Hellwig spin_lock(&bdi->wb.list_lock); 197f758eeabSChristoph Hellwig list_del_init(&inode->i_wb_list); 198f758eeabSChristoph Hellwig spin_unlock(&bdi->wb.list_lock); 199f758eeabSChristoph Hellwig } 200a66979abSDave Chinner 201a66979abSDave Chinner /* 2026610a0bcSAndrew Morton * Redirty an inode: set its when-it-was dirtied timestamp and move it to the 2036610a0bcSAndrew Morton * furthest end of its superblock's dirty-inode list. 2046610a0bcSAndrew Morton * 2056610a0bcSAndrew Morton * Before stamping the inode's ->dirtied_when, we check to see whether it is 20666f3b8e2SJens Axboe * already the most-recently-dirtied inode on the b_dirty list. If that is 2076610a0bcSAndrew Morton * the case then the inode must have been redirtied while it was being written 2086610a0bcSAndrew Morton * out and we don't reset its dirtied_when. 2096610a0bcSAndrew Morton */ 210f758eeabSChristoph Hellwig static void redirty_tail(struct inode *inode, struct bdi_writeback *wb) 2116610a0bcSAndrew Morton { 212f758eeabSChristoph Hellwig assert_spin_locked(&wb->list_lock); 21303ba3782SJens Axboe if (!list_empty(&wb->b_dirty)) { 21466f3b8e2SJens Axboe struct inode *tail; 2156610a0bcSAndrew Morton 2167ccf19a8SNick Piggin tail = wb_inode(wb->b_dirty.next); 21766f3b8e2SJens Axboe if (time_before(inode->dirtied_when, tail->dirtied_when)) 2186610a0bcSAndrew Morton inode->dirtied_when = jiffies; 2196610a0bcSAndrew Morton } 2207ccf19a8SNick Piggin list_move(&inode->i_wb_list, &wb->b_dirty); 2216610a0bcSAndrew Morton } 2226610a0bcSAndrew Morton 2236610a0bcSAndrew Morton /* 22466f3b8e2SJens Axboe * requeue inode for re-scanning after bdi->b_io list is exhausted. 225c986d1e2SAndrew Morton */ 226f758eeabSChristoph Hellwig static void requeue_io(struct inode *inode, struct bdi_writeback *wb) 227c986d1e2SAndrew Morton { 228f758eeabSChristoph Hellwig assert_spin_locked(&wb->list_lock); 2297ccf19a8SNick Piggin list_move(&inode->i_wb_list, &wb->b_more_io); 230c986d1e2SAndrew Morton } 231c986d1e2SAndrew Morton 2321c0eeaf5SJoern Engel static void inode_sync_complete(struct inode *inode) 2331c0eeaf5SJoern Engel { 234365b94aeSJan Kara inode->i_state &= ~I_SYNC; 235365b94aeSJan Kara /* Waiters must see I_SYNC cleared before being woken up */ 2361c0eeaf5SJoern Engel smp_mb(); 2371c0eeaf5SJoern Engel wake_up_bit(&inode->i_state, __I_SYNC); 2381c0eeaf5SJoern Engel } 2391c0eeaf5SJoern Engel 240d2caa3c5SJeff Layton static bool inode_dirtied_after(struct inode *inode, unsigned long t) 241d2caa3c5SJeff Layton { 242d2caa3c5SJeff Layton bool ret = time_after(inode->dirtied_when, t); 243d2caa3c5SJeff Layton #ifndef CONFIG_64BIT 244d2caa3c5SJeff Layton /* 245d2caa3c5SJeff Layton * For inodes being constantly redirtied, dirtied_when can get stuck. 246d2caa3c5SJeff Layton * It _appears_ to be in the future, but is actually in distant past. 247d2caa3c5SJeff Layton * This test is necessary to prevent such wrapped-around relative times 2485b0830cbSJens Axboe * from permanently stopping the whole bdi writeback. 249d2caa3c5SJeff Layton */ 250d2caa3c5SJeff Layton ret = ret && time_before_eq(inode->dirtied_when, jiffies); 251d2caa3c5SJeff Layton #endif 252d2caa3c5SJeff Layton return ret; 253d2caa3c5SJeff Layton } 254d2caa3c5SJeff Layton 255c986d1e2SAndrew Morton /* 256697e6fedSJan Kara * Move expired (dirtied after work->older_than_this) dirty inodes from 257697e6fedSJan Kara * @delaying_queue to @dispatch_queue. 2582c136579SFengguang Wu */ 259e84d0a4fSWu Fengguang static int move_expired_inodes(struct list_head *delaying_queue, 2602c136579SFengguang Wu struct list_head *dispatch_queue, 261ad4e38ddSCurt Wohlgemuth struct wb_writeback_work *work) 2622c136579SFengguang Wu { 2635c03449dSShaohua Li LIST_HEAD(tmp); 2645c03449dSShaohua Li struct list_head *pos, *node; 265cf137307SJens Axboe struct super_block *sb = NULL; 2665c03449dSShaohua Li struct inode *inode; 267cf137307SJens Axboe int do_sb_sort = 0; 268e84d0a4fSWu Fengguang int moved = 0; 2695c03449dSShaohua Li 2702c136579SFengguang Wu while (!list_empty(delaying_queue)) { 2717ccf19a8SNick Piggin inode = wb_inode(delaying_queue->prev); 272ad4e38ddSCurt Wohlgemuth if (work->older_than_this && 273ad4e38ddSCurt Wohlgemuth inode_dirtied_after(inode, *work->older_than_this)) 2742c136579SFengguang Wu break; 275cf137307SJens Axboe if (sb && sb != inode->i_sb) 276cf137307SJens Axboe do_sb_sort = 1; 277cf137307SJens Axboe sb = inode->i_sb; 2787ccf19a8SNick Piggin list_move(&inode->i_wb_list, &tmp); 279e84d0a4fSWu Fengguang moved++; 2805c03449dSShaohua Li } 2815c03449dSShaohua Li 282cf137307SJens Axboe /* just one sb in list, splice to dispatch_queue and we're done */ 283cf137307SJens Axboe if (!do_sb_sort) { 284cf137307SJens Axboe list_splice(&tmp, dispatch_queue); 285e84d0a4fSWu Fengguang goto out; 286cf137307SJens Axboe } 287cf137307SJens Axboe 2885c03449dSShaohua Li /* Move inodes from one superblock together */ 2895c03449dSShaohua Li while (!list_empty(&tmp)) { 2907ccf19a8SNick Piggin sb = wb_inode(tmp.prev)->i_sb; 2915c03449dSShaohua Li list_for_each_prev_safe(pos, node, &tmp) { 2927ccf19a8SNick Piggin inode = wb_inode(pos); 2935c03449dSShaohua Li if (inode->i_sb == sb) 2947ccf19a8SNick Piggin list_move(&inode->i_wb_list, dispatch_queue); 2952c136579SFengguang Wu } 2962c136579SFengguang Wu } 297e84d0a4fSWu Fengguang out: 298e84d0a4fSWu Fengguang return moved; 2995c03449dSShaohua Li } 3002c136579SFengguang Wu 3012c136579SFengguang Wu /* 3022c136579SFengguang Wu * Queue all expired dirty inodes for io, eldest first. 3034ea879b9SWu Fengguang * Before 3044ea879b9SWu Fengguang * newly dirtied b_dirty b_io b_more_io 3054ea879b9SWu Fengguang * =============> gf edc BA 3064ea879b9SWu Fengguang * After 3074ea879b9SWu Fengguang * newly dirtied b_dirty b_io b_more_io 3084ea879b9SWu Fengguang * =============> g fBAedc 3094ea879b9SWu Fengguang * | 3104ea879b9SWu Fengguang * +--> dequeue for IO 3112c136579SFengguang Wu */ 312ad4e38ddSCurt Wohlgemuth static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work) 3132c136579SFengguang Wu { 314e84d0a4fSWu Fengguang int moved; 315f758eeabSChristoph Hellwig assert_spin_locked(&wb->list_lock); 3164ea879b9SWu Fengguang list_splice_init(&wb->b_more_io, &wb->b_io); 317ad4e38ddSCurt Wohlgemuth moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, work); 318ad4e38ddSCurt Wohlgemuth trace_writeback_queue_io(wb, work, moved); 31966f3b8e2SJens Axboe } 32066f3b8e2SJens Axboe 321a9185b41SChristoph Hellwig static int write_inode(struct inode *inode, struct writeback_control *wbc) 32266f3b8e2SJens Axboe { 32303ba3782SJens Axboe if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) 324a9185b41SChristoph Hellwig return inode->i_sb->s_op->write_inode(inode, wbc); 32503ba3782SJens Axboe return 0; 32666f3b8e2SJens Axboe } 32708d8e974SFengguang Wu 3282c136579SFengguang Wu /* 32901c03194SChristoph Hellwig * Wait for writeback on an inode to complete. 33001c03194SChristoph Hellwig */ 331f758eeabSChristoph Hellwig static void inode_wait_for_writeback(struct inode *inode, 332f758eeabSChristoph Hellwig struct bdi_writeback *wb) 33301c03194SChristoph Hellwig { 33401c03194SChristoph Hellwig DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); 33501c03194SChristoph Hellwig wait_queue_head_t *wqh; 33601c03194SChristoph Hellwig 33701c03194SChristoph Hellwig wqh = bit_waitqueue(&inode->i_state, __I_SYNC); 33858a9d3d8SRichard Kennedy while (inode->i_state & I_SYNC) { 339250df6edSDave Chinner spin_unlock(&inode->i_lock); 340f758eeabSChristoph Hellwig spin_unlock(&wb->list_lock); 34101c03194SChristoph Hellwig __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE); 342f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 343250df6edSDave Chinner spin_lock(&inode->i_lock); 34458a9d3d8SRichard Kennedy } 34501c03194SChristoph Hellwig } 34601c03194SChristoph Hellwig 34701c03194SChristoph Hellwig /* 348ccb26b5aSJan Kara * Find proper writeback list for the inode depending on its current state and 349ccb26b5aSJan Kara * possibly also change of its state while we were doing writeback. Here we 350ccb26b5aSJan Kara * handle things such as livelock prevention or fairness of writeback among 351ccb26b5aSJan Kara * inodes. This function can be called only by flusher thread - noone else 352ccb26b5aSJan Kara * processes all inodes in writeback lists and requeueing inodes behind flusher 353ccb26b5aSJan Kara * thread's back can have unexpected consequences. 354ccb26b5aSJan Kara */ 355ccb26b5aSJan Kara static void requeue_inode(struct inode *inode, struct bdi_writeback *wb, 356ccb26b5aSJan Kara struct writeback_control *wbc) 357ccb26b5aSJan Kara { 358ccb26b5aSJan Kara if (inode->i_state & I_FREEING) 359ccb26b5aSJan Kara return; 360ccb26b5aSJan Kara 361ccb26b5aSJan Kara /* 362ccb26b5aSJan Kara * Sync livelock prevention. Each inode is tagged and synced in one 363ccb26b5aSJan Kara * shot. If still dirty, it will be redirty_tail()'ed below. Update 364ccb26b5aSJan Kara * the dirty time to prevent enqueue and sync it again. 365ccb26b5aSJan Kara */ 366ccb26b5aSJan Kara if ((inode->i_state & I_DIRTY) && 367ccb26b5aSJan Kara (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)) 368ccb26b5aSJan Kara inode->dirtied_when = jiffies; 369ccb26b5aSJan Kara 370ccb26b5aSJan Kara if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) { 371ccb26b5aSJan Kara /* 372ccb26b5aSJan Kara * We didn't write back all the pages. nfs_writepages() 373ccb26b5aSJan Kara * sometimes bales out without doing anything. 374ccb26b5aSJan Kara */ 375ccb26b5aSJan Kara if (wbc->nr_to_write <= 0) { 376ccb26b5aSJan Kara /* Slice used up. Queue for next turn. */ 377ccb26b5aSJan Kara requeue_io(inode, wb); 378ccb26b5aSJan Kara } else { 379ccb26b5aSJan Kara /* 380ccb26b5aSJan Kara * Writeback blocked by something other than 381ccb26b5aSJan Kara * congestion. Delay the inode for some time to 382ccb26b5aSJan Kara * avoid spinning on the CPU (100% iowait) 383ccb26b5aSJan Kara * retrying writeback of the dirty page/inode 384ccb26b5aSJan Kara * that cannot be performed immediately. 385ccb26b5aSJan Kara */ 386ccb26b5aSJan Kara redirty_tail(inode, wb); 387ccb26b5aSJan Kara } 388ccb26b5aSJan Kara } else if (inode->i_state & I_DIRTY) { 389ccb26b5aSJan Kara /* 390ccb26b5aSJan Kara * Filesystems can dirty the inode during writeback operations, 391ccb26b5aSJan Kara * such as delayed allocation during submission or metadata 392ccb26b5aSJan Kara * updates after data IO completion. 393ccb26b5aSJan Kara */ 394ccb26b5aSJan Kara redirty_tail(inode, wb); 395ccb26b5aSJan Kara } else { 396ccb26b5aSJan Kara /* The inode is clean. Remove from writeback lists. */ 397ccb26b5aSJan Kara list_del_init(&inode->i_wb_list); 398ccb26b5aSJan Kara } 399ccb26b5aSJan Kara } 400ccb26b5aSJan Kara 401ccb26b5aSJan Kara /* 402f758eeabSChristoph Hellwig * Write out an inode's dirty pages. Called under wb->list_lock and 4030f1b1fd8SDave Chinner * inode->i_lock. Either the caller has an active reference on the inode or 4040f1b1fd8SDave Chinner * the inode has I_WILL_FREE set. 40501c03194SChristoph Hellwig * 4061da177e4SLinus Torvalds * If `wait' is set, wait on the writeout. 4071da177e4SLinus Torvalds * 4081da177e4SLinus Torvalds * The whole writeout design is quite complex and fragile. We want to avoid 4091da177e4SLinus Torvalds * starvation of particular inodes when others are being redirtied, prevent 4101da177e4SLinus Torvalds * livelocks, etc. 4111da177e4SLinus Torvalds */ 4121da177e4SLinus Torvalds static int 413f758eeabSChristoph Hellwig writeback_single_inode(struct inode *inode, struct bdi_writeback *wb, 414f758eeabSChristoph Hellwig struct writeback_control *wbc) 4151da177e4SLinus Torvalds { 4161da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 417251d6a47SWu Fengguang long nr_to_write = wbc->nr_to_write; 41801c03194SChristoph Hellwig unsigned dirty; 4191da177e4SLinus Torvalds int ret; 4201da177e4SLinus Torvalds 421f758eeabSChristoph Hellwig assert_spin_locked(&wb->list_lock); 4220f1b1fd8SDave Chinner assert_spin_locked(&inode->i_lock); 4230f1b1fd8SDave Chinner 42401c03194SChristoph Hellwig if (!atomic_read(&inode->i_count)) 42501c03194SChristoph Hellwig WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); 42601c03194SChristoph Hellwig else 42701c03194SChristoph Hellwig WARN_ON(inode->i_state & I_WILL_FREE); 42801c03194SChristoph Hellwig 42901c03194SChristoph Hellwig if (inode->i_state & I_SYNC) { 430cc1676d9SJan Kara if (wbc->sync_mode != WB_SYNC_ALL) 43101c03194SChristoph Hellwig return 0; 43201c03194SChristoph Hellwig /* 43301c03194SChristoph Hellwig * It's a data-integrity sync. We must wait. 43401c03194SChristoph Hellwig */ 435f758eeabSChristoph Hellwig inode_wait_for_writeback(inode, wb); 43601c03194SChristoph Hellwig } 43701c03194SChristoph Hellwig 4381c0eeaf5SJoern Engel BUG_ON(inode->i_state & I_SYNC); 4391da177e4SLinus Torvalds 4405547e8aaSDmitry Monakhov /* Set I_SYNC, reset I_DIRTY_PAGES */ 4411c0eeaf5SJoern Engel inode->i_state |= I_SYNC; 442250df6edSDave Chinner spin_unlock(&inode->i_lock); 443f758eeabSChristoph Hellwig spin_unlock(&wb->list_lock); 4441da177e4SLinus Torvalds 4451da177e4SLinus Torvalds ret = do_writepages(mapping, wbc); 4461da177e4SLinus Torvalds 44726821ed4SChristoph Hellwig /* 44826821ed4SChristoph Hellwig * Make sure to wait on the data before writing out the metadata. 44926821ed4SChristoph Hellwig * This is important for filesystems that modify metadata on data 45026821ed4SChristoph Hellwig * I/O completion. 45126821ed4SChristoph Hellwig */ 452a9185b41SChristoph Hellwig if (wbc->sync_mode == WB_SYNC_ALL) { 45326821ed4SChristoph Hellwig int err = filemap_fdatawait(mapping); 4541da177e4SLinus Torvalds if (ret == 0) 4551da177e4SLinus Torvalds ret = err; 4561da177e4SLinus Torvalds } 4571da177e4SLinus Torvalds 4585547e8aaSDmitry Monakhov /* 4595547e8aaSDmitry Monakhov * Some filesystems may redirty the inode during the writeback 4605547e8aaSDmitry Monakhov * due to delalloc, clear dirty metadata flags right before 4615547e8aaSDmitry Monakhov * write_inode() 4625547e8aaSDmitry Monakhov */ 463250df6edSDave Chinner spin_lock(&inode->i_lock); 4646290be1cSJan Kara /* Clear I_DIRTY_PAGES if we've written out all dirty pages */ 4656290be1cSJan Kara if (!mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) 4666290be1cSJan Kara inode->i_state &= ~I_DIRTY_PAGES; 4675547e8aaSDmitry Monakhov dirty = inode->i_state & I_DIRTY; 4685547e8aaSDmitry Monakhov inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC); 469250df6edSDave Chinner spin_unlock(&inode->i_lock); 47026821ed4SChristoph Hellwig /* Don't write the inode if only I_DIRTY_PAGES was set */ 47126821ed4SChristoph Hellwig if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { 472a9185b41SChristoph Hellwig int err = write_inode(inode, wbc); 4731da177e4SLinus Torvalds if (ret == 0) 4741da177e4SLinus Torvalds ret = err; 4751da177e4SLinus Torvalds } 4761da177e4SLinus Torvalds 477f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 478250df6edSDave Chinner spin_lock(&inode->i_lock); 479ccb26b5aSJan Kara requeue_inode(inode, wb, wbc); 4801c0eeaf5SJoern Engel inode_sync_complete(inode); 481251d6a47SWu Fengguang trace_writeback_single_inode(inode, wbc, nr_to_write); 4821da177e4SLinus Torvalds return ret; 4831da177e4SLinus Torvalds } 4841da177e4SLinus Torvalds 4851a12d8bdSWu Fengguang static long writeback_chunk_size(struct backing_dev_info *bdi, 4861a12d8bdSWu Fengguang struct wb_writeback_work *work) 487d46db3d5SWu Fengguang { 488d46db3d5SWu Fengguang long pages; 489d46db3d5SWu Fengguang 490d46db3d5SWu Fengguang /* 491d46db3d5SWu Fengguang * WB_SYNC_ALL mode does livelock avoidance by syncing dirty 492d46db3d5SWu Fengguang * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX 493d46db3d5SWu Fengguang * here avoids calling into writeback_inodes_wb() more than once. 494d46db3d5SWu Fengguang * 495d46db3d5SWu Fengguang * The intended call sequence for WB_SYNC_ALL writeback is: 496d46db3d5SWu Fengguang * 497d46db3d5SWu Fengguang * wb_writeback() 498d46db3d5SWu Fengguang * writeback_sb_inodes() <== called only once 499d46db3d5SWu Fengguang * write_cache_pages() <== called once for each inode 500d46db3d5SWu Fengguang * (quickly) tag currently dirty pages 501d46db3d5SWu Fengguang * (maybe slowly) sync all tagged pages 502d46db3d5SWu Fengguang */ 503d46db3d5SWu Fengguang if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages) 504d46db3d5SWu Fengguang pages = LONG_MAX; 5051a12d8bdSWu Fengguang else { 5061a12d8bdSWu Fengguang pages = min(bdi->avg_write_bandwidth / 2, 5071a12d8bdSWu Fengguang global_dirty_limit / DIRTY_SCOPE); 5081a12d8bdSWu Fengguang pages = min(pages, work->nr_pages); 5091a12d8bdSWu Fengguang pages = round_down(pages + MIN_WRITEBACK_PAGES, 5101a12d8bdSWu Fengguang MIN_WRITEBACK_PAGES); 5111a12d8bdSWu Fengguang } 512d46db3d5SWu Fengguang 513d46db3d5SWu Fengguang return pages; 514d46db3d5SWu Fengguang } 515d46db3d5SWu Fengguang 51603ba3782SJens Axboe /* 517f11c9c5cSEdward Shishkin * Write a portion of b_io inodes which belong to @sb. 518edadfb10SChristoph Hellwig * 519edadfb10SChristoph Hellwig * If @only_this_sb is true, then find and write all such 520f11c9c5cSEdward Shishkin * inodes. Otherwise write only ones which go sequentially 521f11c9c5cSEdward Shishkin * in reverse order. 522edadfb10SChristoph Hellwig * 523d46db3d5SWu Fengguang * Return the number of pages and/or inodes written. 524f11c9c5cSEdward Shishkin */ 525d46db3d5SWu Fengguang static long writeback_sb_inodes(struct super_block *sb, 526d46db3d5SWu Fengguang struct bdi_writeback *wb, 527d46db3d5SWu Fengguang struct wb_writeback_work *work) 52803ba3782SJens Axboe { 529d46db3d5SWu Fengguang struct writeback_control wbc = { 530d46db3d5SWu Fengguang .sync_mode = work->sync_mode, 531d46db3d5SWu Fengguang .tagged_writepages = work->tagged_writepages, 532d46db3d5SWu Fengguang .for_kupdate = work->for_kupdate, 533d46db3d5SWu Fengguang .for_background = work->for_background, 534d46db3d5SWu Fengguang .range_cyclic = work->range_cyclic, 535d46db3d5SWu Fengguang .range_start = 0, 536d46db3d5SWu Fengguang .range_end = LLONG_MAX, 537d46db3d5SWu Fengguang }; 538d46db3d5SWu Fengguang unsigned long start_time = jiffies; 539d46db3d5SWu Fengguang long write_chunk; 540d46db3d5SWu Fengguang long wrote = 0; /* count both pages and inodes */ 541d46db3d5SWu Fengguang 54203ba3782SJens Axboe while (!list_empty(&wb->b_io)) { 5437ccf19a8SNick Piggin struct inode *inode = wb_inode(wb->b_io.prev); 544edadfb10SChristoph Hellwig 545edadfb10SChristoph Hellwig if (inode->i_sb != sb) { 546d46db3d5SWu Fengguang if (work->sb) { 547edadfb10SChristoph Hellwig /* 548edadfb10SChristoph Hellwig * We only want to write back data for this 549edadfb10SChristoph Hellwig * superblock, move all inodes not belonging 550edadfb10SChristoph Hellwig * to it back onto the dirty list. 551edadfb10SChristoph Hellwig */ 552f758eeabSChristoph Hellwig redirty_tail(inode, wb); 55366f3b8e2SJens Axboe continue; 55466f3b8e2SJens Axboe } 555edadfb10SChristoph Hellwig 556edadfb10SChristoph Hellwig /* 557edadfb10SChristoph Hellwig * The inode belongs to a different superblock. 558edadfb10SChristoph Hellwig * Bounce back to the caller to unpin this and 559edadfb10SChristoph Hellwig * pin the next superblock. 560edadfb10SChristoph Hellwig */ 561d46db3d5SWu Fengguang break; 562edadfb10SChristoph Hellwig } 563edadfb10SChristoph Hellwig 5649843b76aSChristoph Hellwig /* 5659843b76aSChristoph Hellwig * Don't bother with new inodes or inodes beeing freed, first 5669843b76aSChristoph Hellwig * kind does not need peridic writeout yet, and for the latter 5679843b76aSChristoph Hellwig * kind writeout is handled by the freer. 5689843b76aSChristoph Hellwig */ 569250df6edSDave Chinner spin_lock(&inode->i_lock); 5709843b76aSChristoph Hellwig if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { 571250df6edSDave Chinner spin_unlock(&inode->i_lock); 572fcc5c222SWu Fengguang redirty_tail(inode, wb); 5737ef0d737SNick Piggin continue; 5747ef0d737SNick Piggin } 575cc1676d9SJan Kara if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) { 576cc1676d9SJan Kara /* 577cc1676d9SJan Kara * If this inode is locked for writeback and we are not 578cc1676d9SJan Kara * doing writeback-for-data-integrity, move it to 579cc1676d9SJan Kara * b_more_io so that writeback can proceed with the 580cc1676d9SJan Kara * other inodes on s_io. 581cc1676d9SJan Kara * 582cc1676d9SJan Kara * We'll have another go at writing back this inode 583cc1676d9SJan Kara * when we completed a full scan of b_io. 584cc1676d9SJan Kara */ 585cc1676d9SJan Kara spin_unlock(&inode->i_lock); 586cc1676d9SJan Kara requeue_io(inode, wb); 587cc1676d9SJan Kara trace_writeback_sb_inodes_requeue(inode); 588cc1676d9SJan Kara continue; 589cc1676d9SJan Kara } 5901da177e4SLinus Torvalds __iget(inode); 5911a12d8bdSWu Fengguang write_chunk = writeback_chunk_size(wb->bdi, work); 592d46db3d5SWu Fengguang wbc.nr_to_write = write_chunk; 593d46db3d5SWu Fengguang wbc.pages_skipped = 0; 594250df6edSDave Chinner 595d46db3d5SWu Fengguang writeback_single_inode(inode, wb, &wbc); 596d46db3d5SWu Fengguang 597d46db3d5SWu Fengguang work->nr_pages -= write_chunk - wbc.nr_to_write; 598d46db3d5SWu Fengguang wrote += write_chunk - wbc.nr_to_write; 599d46db3d5SWu Fengguang if (!(inode->i_state & I_DIRTY)) 600d46db3d5SWu Fengguang wrote++; 601d46db3d5SWu Fengguang if (wbc.pages_skipped) { 6021da177e4SLinus Torvalds /* 6031da177e4SLinus Torvalds * writeback is not making progress due to locked 6041da177e4SLinus Torvalds * buffers. Skip this inode for now. 6051da177e4SLinus Torvalds */ 606f758eeabSChristoph Hellwig redirty_tail(inode, wb); 6071da177e4SLinus Torvalds } 6080f1b1fd8SDave Chinner spin_unlock(&inode->i_lock); 609f758eeabSChristoph Hellwig spin_unlock(&wb->list_lock); 6101da177e4SLinus Torvalds iput(inode); 6114ffc8444SOGAWA Hirofumi cond_resched(); 612f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 613d46db3d5SWu Fengguang /* 614d46db3d5SWu Fengguang * bail out to wb_writeback() often enough to check 615d46db3d5SWu Fengguang * background threshold and other termination conditions. 616d46db3d5SWu Fengguang */ 617d46db3d5SWu Fengguang if (wrote) { 618d46db3d5SWu Fengguang if (time_is_before_jiffies(start_time + HZ / 10UL)) 619d46db3d5SWu Fengguang break; 620d46db3d5SWu Fengguang if (work->nr_pages <= 0) 621d46db3d5SWu Fengguang break; 6221da177e4SLinus Torvalds } 6238bc3be27SFengguang Wu } 624d46db3d5SWu Fengguang return wrote; 625f11c9c5cSEdward Shishkin } 62638f21977SNick Piggin 627d46db3d5SWu Fengguang static long __writeback_inodes_wb(struct bdi_writeback *wb, 628d46db3d5SWu Fengguang struct wb_writeback_work *work) 629f11c9c5cSEdward Shishkin { 630d46db3d5SWu Fengguang unsigned long start_time = jiffies; 631d46db3d5SWu Fengguang long wrote = 0; 632f11c9c5cSEdward Shishkin 633f11c9c5cSEdward Shishkin while (!list_empty(&wb->b_io)) { 6347ccf19a8SNick Piggin struct inode *inode = wb_inode(wb->b_io.prev); 635f11c9c5cSEdward Shishkin struct super_block *sb = inode->i_sb; 636f11c9c5cSEdward Shishkin 63712ad3ab6SDave Chinner if (!grab_super_passive(sb)) { 6380e995816SWu Fengguang /* 6390e995816SWu Fengguang * grab_super_passive() may fail consistently due to 6400e995816SWu Fengguang * s_umount being grabbed by someone else. Don't use 6410e995816SWu Fengguang * requeue_io() to avoid busy retrying the inode/sb. 6420e995816SWu Fengguang */ 6430e995816SWu Fengguang redirty_tail(inode, wb); 644d19de7edSChristoph Hellwig continue; 645334132aeSChristoph Hellwig } 646d46db3d5SWu Fengguang wrote += writeback_sb_inodes(sb, wb, work); 647d19de7edSChristoph Hellwig drop_super(sb); 648f11c9c5cSEdward Shishkin 649d46db3d5SWu Fengguang /* refer to the same tests at the end of writeback_sb_inodes */ 650d46db3d5SWu Fengguang if (wrote) { 651d46db3d5SWu Fengguang if (time_is_before_jiffies(start_time + HZ / 10UL)) 652d46db3d5SWu Fengguang break; 653d46db3d5SWu Fengguang if (work->nr_pages <= 0) 654f11c9c5cSEdward Shishkin break; 655f11c9c5cSEdward Shishkin } 656d46db3d5SWu Fengguang } 65766f3b8e2SJens Axboe /* Leave any unwritten inodes on b_io */ 658d46db3d5SWu Fengguang return wrote; 65966f3b8e2SJens Axboe } 66066f3b8e2SJens Axboe 6610e175a18SCurt Wohlgemuth long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages, 6620e175a18SCurt Wohlgemuth enum wb_reason reason) 663edadfb10SChristoph Hellwig { 664d46db3d5SWu Fengguang struct wb_writeback_work work = { 665d46db3d5SWu Fengguang .nr_pages = nr_pages, 666d46db3d5SWu Fengguang .sync_mode = WB_SYNC_NONE, 667d46db3d5SWu Fengguang .range_cyclic = 1, 6680e175a18SCurt Wohlgemuth .reason = reason, 669d46db3d5SWu Fengguang }; 670edadfb10SChristoph Hellwig 671f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 672424b351fSWu Fengguang if (list_empty(&wb->b_io)) 673ad4e38ddSCurt Wohlgemuth queue_io(wb, &work); 674d46db3d5SWu Fengguang __writeback_inodes_wb(wb, &work); 675f758eeabSChristoph Hellwig spin_unlock(&wb->list_lock); 676edadfb10SChristoph Hellwig 677d46db3d5SWu Fengguang return nr_pages - work.nr_pages; 67866f3b8e2SJens Axboe } 67966f3b8e2SJens Axboe 680b00949aaSWu Fengguang static bool over_bground_thresh(struct backing_dev_info *bdi) 68103ba3782SJens Axboe { 68203ba3782SJens Axboe unsigned long background_thresh, dirty_thresh; 68303ba3782SJens Axboe 68416c4042fSWu Fengguang global_dirty_limits(&background_thresh, &dirty_thresh); 68503ba3782SJens Axboe 686b00949aaSWu Fengguang if (global_page_state(NR_FILE_DIRTY) + 687b00949aaSWu Fengguang global_page_state(NR_UNSTABLE_NFS) > background_thresh) 688b00949aaSWu Fengguang return true; 689b00949aaSWu Fengguang 690b00949aaSWu Fengguang if (bdi_stat(bdi, BDI_RECLAIMABLE) > 691b00949aaSWu Fengguang bdi_dirty_limit(bdi, background_thresh)) 692b00949aaSWu Fengguang return true; 693b00949aaSWu Fengguang 694b00949aaSWu Fengguang return false; 69503ba3782SJens Axboe } 69603ba3782SJens Axboe 69703ba3782SJens Axboe /* 698e98be2d5SWu Fengguang * Called under wb->list_lock. If there are multiple wb per bdi, 699e98be2d5SWu Fengguang * only the flusher working on the first wb should do it. 700e98be2d5SWu Fengguang */ 701e98be2d5SWu Fengguang static void wb_update_bandwidth(struct bdi_writeback *wb, 702e98be2d5SWu Fengguang unsigned long start_time) 703e98be2d5SWu Fengguang { 704af6a3113SWu Fengguang __bdi_update_bandwidth(wb->bdi, 0, 0, 0, 0, 0, start_time); 705e98be2d5SWu Fengguang } 706e98be2d5SWu Fengguang 707e98be2d5SWu Fengguang /* 70803ba3782SJens Axboe * Explicit flushing or periodic writeback of "old" data. 70903ba3782SJens Axboe * 71003ba3782SJens Axboe * Define "old": the first time one of an inode's pages is dirtied, we mark the 71103ba3782SJens Axboe * dirtying-time in the inode's address_space. So this periodic writeback code 71203ba3782SJens Axboe * just walks the superblock inode list, writing back any inodes which are 71303ba3782SJens Axboe * older than a specific point in time. 71403ba3782SJens Axboe * 71503ba3782SJens Axboe * Try to run once per dirty_writeback_interval. But if a writeback event 71603ba3782SJens Axboe * takes longer than a dirty_writeback_interval interval, then leave a 71703ba3782SJens Axboe * one-second gap. 71803ba3782SJens Axboe * 71903ba3782SJens Axboe * older_than_this takes precedence over nr_to_write. So we'll only write back 72003ba3782SJens Axboe * all dirty pages if they are all attached to "old" mappings. 72103ba3782SJens Axboe */ 722c4a77a6cSJens Axboe static long wb_writeback(struct bdi_writeback *wb, 72383ba7b07SChristoph Hellwig struct wb_writeback_work *work) 72403ba3782SJens Axboe { 725e98be2d5SWu Fengguang unsigned long wb_start = jiffies; 726d46db3d5SWu Fengguang long nr_pages = work->nr_pages; 72703ba3782SJens Axboe unsigned long oldest_jif; 728a5989bdcSJan Kara struct inode *inode; 729d46db3d5SWu Fengguang long progress; 73003ba3782SJens Axboe 731e185dda8SWu Fengguang oldest_jif = jiffies; 732d46db3d5SWu Fengguang work->older_than_this = &oldest_jif; 73303ba3782SJens Axboe 734e8dfc305SWu Fengguang spin_lock(&wb->list_lock); 73503ba3782SJens Axboe for (;;) { 73603ba3782SJens Axboe /* 737d3ddec76SWu Fengguang * Stop writeback when nr_pages has been consumed 73803ba3782SJens Axboe */ 73983ba7b07SChristoph Hellwig if (work->nr_pages <= 0) 74003ba3782SJens Axboe break; 74103ba3782SJens Axboe 74203ba3782SJens Axboe /* 743aa373cf5SJan Kara * Background writeout and kupdate-style writeback may 744aa373cf5SJan Kara * run forever. Stop them if there is other work to do 745aa373cf5SJan Kara * so that e.g. sync can proceed. They'll be restarted 746aa373cf5SJan Kara * after the other works are all done. 747aa373cf5SJan Kara */ 748aa373cf5SJan Kara if ((work->for_background || work->for_kupdate) && 749aa373cf5SJan Kara !list_empty(&wb->bdi->work_list)) 750aa373cf5SJan Kara break; 751aa373cf5SJan Kara 752aa373cf5SJan Kara /* 753d3ddec76SWu Fengguang * For background writeout, stop when we are below the 754d3ddec76SWu Fengguang * background dirty threshold 75503ba3782SJens Axboe */ 756b00949aaSWu Fengguang if (work->for_background && !over_bground_thresh(wb->bdi)) 75703ba3782SJens Axboe break; 75803ba3782SJens Axboe 7591bc36b64SJan Kara /* 7601bc36b64SJan Kara * Kupdate and background works are special and we want to 7611bc36b64SJan Kara * include all inodes that need writing. Livelock avoidance is 7621bc36b64SJan Kara * handled by these works yielding to any other work so we are 7631bc36b64SJan Kara * safe. 7641bc36b64SJan Kara */ 765ba9aa839SWu Fengguang if (work->for_kupdate) { 766ba9aa839SWu Fengguang oldest_jif = jiffies - 767ba9aa839SWu Fengguang msecs_to_jiffies(dirty_expire_interval * 10); 7681bc36b64SJan Kara } else if (work->for_background) 7691bc36b64SJan Kara oldest_jif = jiffies; 770028c2dd1SDave Chinner 771d46db3d5SWu Fengguang trace_writeback_start(wb->bdi, work); 772e8dfc305SWu Fengguang if (list_empty(&wb->b_io)) 773ad4e38ddSCurt Wohlgemuth queue_io(wb, work); 77483ba7b07SChristoph Hellwig if (work->sb) 775d46db3d5SWu Fengguang progress = writeback_sb_inodes(work->sb, wb, work); 776edadfb10SChristoph Hellwig else 777d46db3d5SWu Fengguang progress = __writeback_inodes_wb(wb, work); 778d46db3d5SWu Fengguang trace_writeback_written(wb->bdi, work); 779028c2dd1SDave Chinner 780e98be2d5SWu Fengguang wb_update_bandwidth(wb, wb_start); 78103ba3782SJens Axboe 78203ba3782SJens Axboe /* 78371fd05a8SJens Axboe * Did we write something? Try for more 784e6fb6da2SWu Fengguang * 785e6fb6da2SWu Fengguang * Dirty inodes are moved to b_io for writeback in batches. 786e6fb6da2SWu Fengguang * The completion of the current batch does not necessarily 787e6fb6da2SWu Fengguang * mean the overall work is done. So we keep looping as long 788e6fb6da2SWu Fengguang * as made some progress on cleaning pages or inodes. 78971fd05a8SJens Axboe */ 790d46db3d5SWu Fengguang if (progress) 79103ba3782SJens Axboe continue; 792a5989bdcSJan Kara /* 793e6fb6da2SWu Fengguang * No more inodes for IO, bail 794a5989bdcSJan Kara */ 795b7a2441fSWu Fengguang if (list_empty(&wb->b_more_io)) 79603ba3782SJens Axboe break; 79703ba3782SJens Axboe /* 7988010c3b6SJens Axboe * Nothing written. Wait for some inode to 7998010c3b6SJens Axboe * become available for writeback. Otherwise 8008010c3b6SJens Axboe * we'll just busyloop. 80103ba3782SJens Axboe */ 80203ba3782SJens Axboe if (!list_empty(&wb->b_more_io)) { 803d46db3d5SWu Fengguang trace_writeback_wait(wb->bdi, work); 80403ba3782SJens Axboe inode = wb_inode(wb->b_more_io.prev); 805250df6edSDave Chinner spin_lock(&inode->i_lock); 806f758eeabSChristoph Hellwig inode_wait_for_writeback(inode, wb); 807250df6edSDave Chinner spin_unlock(&inode->i_lock); 80803ba3782SJens Axboe } 80903ba3782SJens Axboe } 810e8dfc305SWu Fengguang spin_unlock(&wb->list_lock); 81103ba3782SJens Axboe 812d46db3d5SWu Fengguang return nr_pages - work->nr_pages; 81303ba3782SJens Axboe } 81403ba3782SJens Axboe 81503ba3782SJens Axboe /* 81683ba7b07SChristoph Hellwig * Return the next wb_writeback_work struct that hasn't been processed yet. 81703ba3782SJens Axboe */ 81883ba7b07SChristoph Hellwig static struct wb_writeback_work * 81908852b6dSMinchan Kim get_next_work_item(struct backing_dev_info *bdi) 82003ba3782SJens Axboe { 82183ba7b07SChristoph Hellwig struct wb_writeback_work *work = NULL; 82203ba3782SJens Axboe 8236467716aSArtem Bityutskiy spin_lock_bh(&bdi->wb_lock); 82483ba7b07SChristoph Hellwig if (!list_empty(&bdi->work_list)) { 82583ba7b07SChristoph Hellwig work = list_entry(bdi->work_list.next, 82683ba7b07SChristoph Hellwig struct wb_writeback_work, list); 82783ba7b07SChristoph Hellwig list_del_init(&work->list); 82803ba3782SJens Axboe } 8296467716aSArtem Bityutskiy spin_unlock_bh(&bdi->wb_lock); 83083ba7b07SChristoph Hellwig return work; 83103ba3782SJens Axboe } 83203ba3782SJens Axboe 833cdf01dd5SLinus Torvalds /* 834cdf01dd5SLinus Torvalds * Add in the number of potentially dirty inodes, because each inode 835cdf01dd5SLinus Torvalds * write can dirty pagecache in the underlying blockdev. 836cdf01dd5SLinus Torvalds */ 837cdf01dd5SLinus Torvalds static unsigned long get_nr_dirty_pages(void) 838cdf01dd5SLinus Torvalds { 839cdf01dd5SLinus Torvalds return global_page_state(NR_FILE_DIRTY) + 840cdf01dd5SLinus Torvalds global_page_state(NR_UNSTABLE_NFS) + 841cdf01dd5SLinus Torvalds get_nr_dirty_inodes(); 842cdf01dd5SLinus Torvalds } 843cdf01dd5SLinus Torvalds 8446585027aSJan Kara static long wb_check_background_flush(struct bdi_writeback *wb) 8456585027aSJan Kara { 846b00949aaSWu Fengguang if (over_bground_thresh(wb->bdi)) { 8476585027aSJan Kara 8486585027aSJan Kara struct wb_writeback_work work = { 8496585027aSJan Kara .nr_pages = LONG_MAX, 8506585027aSJan Kara .sync_mode = WB_SYNC_NONE, 8516585027aSJan Kara .for_background = 1, 8526585027aSJan Kara .range_cyclic = 1, 8530e175a18SCurt Wohlgemuth .reason = WB_REASON_BACKGROUND, 8546585027aSJan Kara }; 8556585027aSJan Kara 8566585027aSJan Kara return wb_writeback(wb, &work); 8576585027aSJan Kara } 8586585027aSJan Kara 8596585027aSJan Kara return 0; 8606585027aSJan Kara } 8616585027aSJan Kara 86203ba3782SJens Axboe static long wb_check_old_data_flush(struct bdi_writeback *wb) 86303ba3782SJens Axboe { 86403ba3782SJens Axboe unsigned long expired; 86503ba3782SJens Axboe long nr_pages; 86603ba3782SJens Axboe 86769b62d01SJens Axboe /* 86869b62d01SJens Axboe * When set to zero, disable periodic writeback 86969b62d01SJens Axboe */ 87069b62d01SJens Axboe if (!dirty_writeback_interval) 87169b62d01SJens Axboe return 0; 87269b62d01SJens Axboe 87303ba3782SJens Axboe expired = wb->last_old_flush + 87403ba3782SJens Axboe msecs_to_jiffies(dirty_writeback_interval * 10); 87503ba3782SJens Axboe if (time_before(jiffies, expired)) 87603ba3782SJens Axboe return 0; 87703ba3782SJens Axboe 87803ba3782SJens Axboe wb->last_old_flush = jiffies; 879cdf01dd5SLinus Torvalds nr_pages = get_nr_dirty_pages(); 88003ba3782SJens Axboe 881c4a77a6cSJens Axboe if (nr_pages) { 88283ba7b07SChristoph Hellwig struct wb_writeback_work work = { 883c4a77a6cSJens Axboe .nr_pages = nr_pages, 884c4a77a6cSJens Axboe .sync_mode = WB_SYNC_NONE, 885c4a77a6cSJens Axboe .for_kupdate = 1, 886c4a77a6cSJens Axboe .range_cyclic = 1, 8870e175a18SCurt Wohlgemuth .reason = WB_REASON_PERIODIC, 888c4a77a6cSJens Axboe }; 889c4a77a6cSJens Axboe 89083ba7b07SChristoph Hellwig return wb_writeback(wb, &work); 891c4a77a6cSJens Axboe } 89203ba3782SJens Axboe 89303ba3782SJens Axboe return 0; 89403ba3782SJens Axboe } 89503ba3782SJens Axboe 89603ba3782SJens Axboe /* 89703ba3782SJens Axboe * Retrieve work items and do the writeback they describe 89803ba3782SJens Axboe */ 89903ba3782SJens Axboe long wb_do_writeback(struct bdi_writeback *wb, int force_wait) 90003ba3782SJens Axboe { 90103ba3782SJens Axboe struct backing_dev_info *bdi = wb->bdi; 90283ba7b07SChristoph Hellwig struct wb_writeback_work *work; 903c4a77a6cSJens Axboe long wrote = 0; 90403ba3782SJens Axboe 90581d73a32SJan Kara set_bit(BDI_writeback_running, &wb->bdi->state); 90608852b6dSMinchan Kim while ((work = get_next_work_item(bdi)) != NULL) { 90703ba3782SJens Axboe /* 90803ba3782SJens Axboe * Override sync mode, in case we must wait for completion 90983ba7b07SChristoph Hellwig * because this thread is exiting now. 91003ba3782SJens Axboe */ 91103ba3782SJens Axboe if (force_wait) 91283ba7b07SChristoph Hellwig work->sync_mode = WB_SYNC_ALL; 91383ba7b07SChristoph Hellwig 914455b2864SDave Chinner trace_writeback_exec(bdi, work); 915455b2864SDave Chinner 91683ba7b07SChristoph Hellwig wrote += wb_writeback(wb, work); 91703ba3782SJens Axboe 91803ba3782SJens Axboe /* 91983ba7b07SChristoph Hellwig * Notify the caller of completion if this is a synchronous 92083ba7b07SChristoph Hellwig * work item, otherwise just free it. 92103ba3782SJens Axboe */ 92283ba7b07SChristoph Hellwig if (work->done) 92383ba7b07SChristoph Hellwig complete(work->done); 92483ba7b07SChristoph Hellwig else 92583ba7b07SChristoph Hellwig kfree(work); 92603ba3782SJens Axboe } 92703ba3782SJens Axboe 92803ba3782SJens Axboe /* 92903ba3782SJens Axboe * Check for periodic writeback, kupdated() style 93003ba3782SJens Axboe */ 93103ba3782SJens Axboe wrote += wb_check_old_data_flush(wb); 9326585027aSJan Kara wrote += wb_check_background_flush(wb); 93381d73a32SJan Kara clear_bit(BDI_writeback_running, &wb->bdi->state); 93403ba3782SJens Axboe 93503ba3782SJens Axboe return wrote; 93603ba3782SJens Axboe } 93703ba3782SJens Axboe 93803ba3782SJens Axboe /* 93903ba3782SJens Axboe * Handle writeback of dirty data for the device backed by this bdi. Also 94003ba3782SJens Axboe * wakes up periodically and does kupdated style flushing. 94103ba3782SJens Axboe */ 94208243900SChristoph Hellwig int bdi_writeback_thread(void *data) 94303ba3782SJens Axboe { 94408243900SChristoph Hellwig struct bdi_writeback *wb = data; 94508243900SChristoph Hellwig struct backing_dev_info *bdi = wb->bdi; 94603ba3782SJens Axboe long pages_written; 94703ba3782SJens Axboe 948766f9164SPeter Zijlstra current->flags |= PF_SWAPWRITE; 94908243900SChristoph Hellwig set_freezable(); 950ecd58403SArtem Bityutskiy wb->last_active = jiffies; 95103ba3782SJens Axboe 95203ba3782SJens Axboe /* 95308243900SChristoph Hellwig * Our parent may run at a different priority, just set us to normal 95403ba3782SJens Axboe */ 95508243900SChristoph Hellwig set_user_nice(current, 0); 95608243900SChristoph Hellwig 957455b2864SDave Chinner trace_writeback_thread_start(bdi); 958455b2864SDave Chinner 9598a32c441STejun Heo while (!kthread_freezable_should_stop(NULL)) { 9606467716aSArtem Bityutskiy /* 9616467716aSArtem Bityutskiy * Remove own delayed wake-up timer, since we are already awake 9626467716aSArtem Bityutskiy * and we'll take care of the preriodic write-back. 9636467716aSArtem Bityutskiy */ 9646467716aSArtem Bityutskiy del_timer(&wb->wakeup_timer); 9656467716aSArtem Bityutskiy 96603ba3782SJens Axboe pages_written = wb_do_writeback(wb, 0); 96703ba3782SJens Axboe 968455b2864SDave Chinner trace_writeback_pages_written(pages_written); 969455b2864SDave Chinner 97003ba3782SJens Axboe if (pages_written) 971ecd58403SArtem Bityutskiy wb->last_active = jiffies; 97203ba3782SJens Axboe 973297252c8SArtem Bityutskiy set_current_state(TASK_INTERRUPTIBLE); 974b76b4014SJ. Bruce Fields if (!list_empty(&bdi->work_list) || kthread_should_stop()) { 975297252c8SArtem Bityutskiy __set_current_state(TASK_RUNNING); 976297252c8SArtem Bityutskiy continue; 97703ba3782SJens Axboe } 97803ba3782SJens Axboe 979253c34e9SArtem Bityutskiy if (wb_has_dirty_io(wb) && dirty_writeback_interval) 980fff5b85aSArtem Bityutskiy schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10)); 981253c34e9SArtem Bityutskiy else { 982253c34e9SArtem Bityutskiy /* 983253c34e9SArtem Bityutskiy * We have nothing to do, so can go sleep without any 984253c34e9SArtem Bityutskiy * timeout and save power. When a work is queued or 985253c34e9SArtem Bityutskiy * something is made dirty - we will be woken up. 986253c34e9SArtem Bityutskiy */ 98769b62d01SJens Axboe schedule(); 988f9eadbbdSJens Axboe } 98903ba3782SJens Axboe } 99003ba3782SJens Axboe 991fff5b85aSArtem Bityutskiy /* Flush any work that raced with us exiting */ 99208243900SChristoph Hellwig if (!list_empty(&bdi->work_list)) 99308243900SChristoph Hellwig wb_do_writeback(wb, 1); 994455b2864SDave Chinner 995455b2864SDave Chinner trace_writeback_thread_stop(bdi); 99603ba3782SJens Axboe return 0; 99703ba3782SJens Axboe } 99803ba3782SJens Axboe 99908243900SChristoph Hellwig 100003ba3782SJens Axboe /* 100103ba3782SJens Axboe * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back 100203ba3782SJens Axboe * the whole world. 100303ba3782SJens Axboe */ 10040e175a18SCurt Wohlgemuth void wakeup_flusher_threads(long nr_pages, enum wb_reason reason) 100503ba3782SJens Axboe { 1006b8c2f347SChristoph Hellwig struct backing_dev_info *bdi; 1007b8c2f347SChristoph Hellwig 100883ba7b07SChristoph Hellwig if (!nr_pages) { 100983ba7b07SChristoph Hellwig nr_pages = global_page_state(NR_FILE_DIRTY) + 101003ba3782SJens Axboe global_page_state(NR_UNSTABLE_NFS); 1011b8c2f347SChristoph Hellwig } 1012b8c2f347SChristoph Hellwig 1013b8c2f347SChristoph Hellwig rcu_read_lock(); 1014b8c2f347SChristoph Hellwig list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { 1015b8c2f347SChristoph Hellwig if (!bdi_has_dirty_io(bdi)) 1016b8c2f347SChristoph Hellwig continue; 10170e175a18SCurt Wohlgemuth __bdi_start_writeback(bdi, nr_pages, false, reason); 1018b8c2f347SChristoph Hellwig } 1019b8c2f347SChristoph Hellwig rcu_read_unlock(); 102003ba3782SJens Axboe } 102103ba3782SJens Axboe 102203ba3782SJens Axboe static noinline void block_dump___mark_inode_dirty(struct inode *inode) 102303ba3782SJens Axboe { 102403ba3782SJens Axboe if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { 102503ba3782SJens Axboe struct dentry *dentry; 102603ba3782SJens Axboe const char *name = "?"; 102703ba3782SJens Axboe 102803ba3782SJens Axboe dentry = d_find_alias(inode); 102903ba3782SJens Axboe if (dentry) { 103003ba3782SJens Axboe spin_lock(&dentry->d_lock); 103103ba3782SJens Axboe name = (const char *) dentry->d_name.name; 103203ba3782SJens Axboe } 103303ba3782SJens Axboe printk(KERN_DEBUG 103403ba3782SJens Axboe "%s(%d): dirtied inode %lu (%s) on %s\n", 103503ba3782SJens Axboe current->comm, task_pid_nr(current), inode->i_ino, 103603ba3782SJens Axboe name, inode->i_sb->s_id); 103703ba3782SJens Axboe if (dentry) { 103803ba3782SJens Axboe spin_unlock(&dentry->d_lock); 103903ba3782SJens Axboe dput(dentry); 104003ba3782SJens Axboe } 104103ba3782SJens Axboe } 104203ba3782SJens Axboe } 104303ba3782SJens Axboe 104403ba3782SJens Axboe /** 104503ba3782SJens Axboe * __mark_inode_dirty - internal function 104603ba3782SJens Axboe * @inode: inode to mark 104703ba3782SJens Axboe * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) 104803ba3782SJens Axboe * Mark an inode as dirty. Callers should use mark_inode_dirty or 104903ba3782SJens Axboe * mark_inode_dirty_sync. 105003ba3782SJens Axboe * 105103ba3782SJens Axboe * Put the inode on the super block's dirty list. 105203ba3782SJens Axboe * 105303ba3782SJens Axboe * CAREFUL! We mark it dirty unconditionally, but move it onto the 105403ba3782SJens Axboe * dirty list only if it is hashed or if it refers to a blockdev. 105503ba3782SJens Axboe * If it was not hashed, it will never be added to the dirty list 105603ba3782SJens Axboe * even if it is later hashed, as it will have been marked dirty already. 105703ba3782SJens Axboe * 105803ba3782SJens Axboe * In short, make sure you hash any inodes _before_ you start marking 105903ba3782SJens Axboe * them dirty. 106003ba3782SJens Axboe * 106103ba3782SJens Axboe * Note that for blockdevs, inode->dirtied_when represents the dirtying time of 106203ba3782SJens Axboe * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of 106303ba3782SJens Axboe * the kernel-internal blockdev inode represents the dirtying time of the 106403ba3782SJens Axboe * blockdev's pages. This is why for I_DIRTY_PAGES we always use 106503ba3782SJens Axboe * page->mapping->host, so the page-dirtying time is recorded in the internal 106603ba3782SJens Axboe * blockdev inode. 106703ba3782SJens Axboe */ 106803ba3782SJens Axboe void __mark_inode_dirty(struct inode *inode, int flags) 106903ba3782SJens Axboe { 107003ba3782SJens Axboe struct super_block *sb = inode->i_sb; 1071253c34e9SArtem Bityutskiy struct backing_dev_info *bdi = NULL; 107203ba3782SJens Axboe 107303ba3782SJens Axboe /* 107403ba3782SJens Axboe * Don't do this for I_DIRTY_PAGES - that doesn't actually 107503ba3782SJens Axboe * dirty the inode itself 107603ba3782SJens Axboe */ 107703ba3782SJens Axboe if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { 107803ba3782SJens Axboe if (sb->s_op->dirty_inode) 1079aa385729SChristoph Hellwig sb->s_op->dirty_inode(inode, flags); 108003ba3782SJens Axboe } 108103ba3782SJens Axboe 108203ba3782SJens Axboe /* 108303ba3782SJens Axboe * make sure that changes are seen by all cpus before we test i_state 108403ba3782SJens Axboe * -- mikulas 108503ba3782SJens Axboe */ 108603ba3782SJens Axboe smp_mb(); 108703ba3782SJens Axboe 108803ba3782SJens Axboe /* avoid the locking if we can */ 108903ba3782SJens Axboe if ((inode->i_state & flags) == flags) 109003ba3782SJens Axboe return; 109103ba3782SJens Axboe 109203ba3782SJens Axboe if (unlikely(block_dump)) 109303ba3782SJens Axboe block_dump___mark_inode_dirty(inode); 109403ba3782SJens Axboe 1095250df6edSDave Chinner spin_lock(&inode->i_lock); 109603ba3782SJens Axboe if ((inode->i_state & flags) != flags) { 109703ba3782SJens Axboe const int was_dirty = inode->i_state & I_DIRTY; 109803ba3782SJens Axboe 109903ba3782SJens Axboe inode->i_state |= flags; 110003ba3782SJens Axboe 110103ba3782SJens Axboe /* 110203ba3782SJens Axboe * If the inode is being synced, just update its dirty state. 110303ba3782SJens Axboe * The unlocker will place the inode on the appropriate 110403ba3782SJens Axboe * superblock list, based upon its state. 110503ba3782SJens Axboe */ 110603ba3782SJens Axboe if (inode->i_state & I_SYNC) 1107250df6edSDave Chinner goto out_unlock_inode; 110803ba3782SJens Axboe 110903ba3782SJens Axboe /* 111003ba3782SJens Axboe * Only add valid (hashed) inodes to the superblock's 111103ba3782SJens Axboe * dirty list. Add blockdev inodes as well. 111203ba3782SJens Axboe */ 111303ba3782SJens Axboe if (!S_ISBLK(inode->i_mode)) { 11141d3382cbSAl Viro if (inode_unhashed(inode)) 1115250df6edSDave Chinner goto out_unlock_inode; 111603ba3782SJens Axboe } 1117a4ffdde6SAl Viro if (inode->i_state & I_FREEING) 1118250df6edSDave Chinner goto out_unlock_inode; 111903ba3782SJens Axboe 112003ba3782SJens Axboe /* 112103ba3782SJens Axboe * If the inode was already on b_dirty/b_io/b_more_io, don't 112203ba3782SJens Axboe * reposition it (that would break b_dirty time-ordering). 112303ba3782SJens Axboe */ 112403ba3782SJens Axboe if (!was_dirty) { 1125a66979abSDave Chinner bool wakeup_bdi = false; 1126253c34e9SArtem Bityutskiy bdi = inode_to_bdi(inode); 1127500b067cSJens Axboe 1128253c34e9SArtem Bityutskiy if (bdi_cap_writeback_dirty(bdi)) { 1129253c34e9SArtem Bityutskiy WARN(!test_bit(BDI_registered, &bdi->state), 1130253c34e9SArtem Bityutskiy "bdi-%s not registered\n", bdi->name); 1131253c34e9SArtem Bityutskiy 1132253c34e9SArtem Bityutskiy /* 1133253c34e9SArtem Bityutskiy * If this is the first dirty inode for this 1134253c34e9SArtem Bityutskiy * bdi, we have to wake-up the corresponding 1135253c34e9SArtem Bityutskiy * bdi thread to make sure background 1136253c34e9SArtem Bityutskiy * write-back happens later. 1137253c34e9SArtem Bityutskiy */ 1138253c34e9SArtem Bityutskiy if (!wb_has_dirty_io(&bdi->wb)) 1139253c34e9SArtem Bityutskiy wakeup_bdi = true; 1140500b067cSJens Axboe } 114103ba3782SJens Axboe 1142a66979abSDave Chinner spin_unlock(&inode->i_lock); 1143f758eeabSChristoph Hellwig spin_lock(&bdi->wb.list_lock); 114403ba3782SJens Axboe inode->dirtied_when = jiffies; 11457ccf19a8SNick Piggin list_move(&inode->i_wb_list, &bdi->wb.b_dirty); 1146f758eeabSChristoph Hellwig spin_unlock(&bdi->wb.list_lock); 1147253c34e9SArtem Bityutskiy 1148253c34e9SArtem Bityutskiy if (wakeup_bdi) 11496467716aSArtem Bityutskiy bdi_wakeup_thread_delayed(bdi); 1150a66979abSDave Chinner return; 1151a66979abSDave Chinner } 1152a66979abSDave Chinner } 1153a66979abSDave Chinner out_unlock_inode: 1154a66979abSDave Chinner spin_unlock(&inode->i_lock); 1155a66979abSDave Chinner 115603ba3782SJens Axboe } 115703ba3782SJens Axboe EXPORT_SYMBOL(__mark_inode_dirty); 115803ba3782SJens Axboe 1159b6e51316SJens Axboe static void wait_sb_inodes(struct super_block *sb) 116066f3b8e2SJens Axboe { 116138f21977SNick Piggin struct inode *inode, *old_inode = NULL; 116238f21977SNick Piggin 116303ba3782SJens Axboe /* 116403ba3782SJens Axboe * We need to be protected against the filesystem going from 116503ba3782SJens Axboe * r/o to r/w or vice versa. 116603ba3782SJens Axboe */ 1167b6e51316SJens Axboe WARN_ON(!rwsem_is_locked(&sb->s_umount)); 116803ba3782SJens Axboe 116955fa6091SDave Chinner spin_lock(&inode_sb_list_lock); 117066f3b8e2SJens Axboe 117138f21977SNick Piggin /* 117238f21977SNick Piggin * Data integrity sync. Must wait for all pages under writeback, 117338f21977SNick Piggin * because there may have been pages dirtied before our sync 117438f21977SNick Piggin * call, but which had writeout started before we write it out. 117538f21977SNick Piggin * In which case, the inode may not be on the dirty list, but 117638f21977SNick Piggin * we still have to wait for that writeout. 117738f21977SNick Piggin */ 1178b6e51316SJens Axboe list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { 1179250df6edSDave Chinner struct address_space *mapping = inode->i_mapping; 118038f21977SNick Piggin 1181250df6edSDave Chinner spin_lock(&inode->i_lock); 1182250df6edSDave Chinner if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) || 1183250df6edSDave Chinner (mapping->nrpages == 0)) { 1184250df6edSDave Chinner spin_unlock(&inode->i_lock); 118538f21977SNick Piggin continue; 1186250df6edSDave Chinner } 118738f21977SNick Piggin __iget(inode); 1188250df6edSDave Chinner spin_unlock(&inode->i_lock); 118955fa6091SDave Chinner spin_unlock(&inode_sb_list_lock); 119055fa6091SDave Chinner 119138f21977SNick Piggin /* 119255fa6091SDave Chinner * We hold a reference to 'inode' so it couldn't have been 119355fa6091SDave Chinner * removed from s_inodes list while we dropped the 119455fa6091SDave Chinner * inode_sb_list_lock. We cannot iput the inode now as we can 119555fa6091SDave Chinner * be holding the last reference and we cannot iput it under 119655fa6091SDave Chinner * inode_sb_list_lock. So we keep the reference and iput it 119755fa6091SDave Chinner * later. 119838f21977SNick Piggin */ 119938f21977SNick Piggin iput(old_inode); 120038f21977SNick Piggin old_inode = inode; 120138f21977SNick Piggin 120238f21977SNick Piggin filemap_fdatawait(mapping); 120338f21977SNick Piggin 120438f21977SNick Piggin cond_resched(); 120538f21977SNick Piggin 120655fa6091SDave Chinner spin_lock(&inode_sb_list_lock); 120738f21977SNick Piggin } 120855fa6091SDave Chinner spin_unlock(&inode_sb_list_lock); 120938f21977SNick Piggin iput(old_inode); 121066f3b8e2SJens Axboe } 12111da177e4SLinus Torvalds 1212d8a8559cSJens Axboe /** 12133259f8beSChris Mason * writeback_inodes_sb_nr - writeback dirty inodes from given super_block 1214d8a8559cSJens Axboe * @sb: the superblock 12153259f8beSChris Mason * @nr: the number of pages to write 1216786228abSMarcos Paulo de Souza * @reason: reason why some writeback work initiated 12171da177e4SLinus Torvalds * 1218d8a8559cSJens Axboe * Start writeback on some inodes on this super_block. No guarantees are made 1219d8a8559cSJens Axboe * on how many (if any) will be written, and this function does not wait 12203259f8beSChris Mason * for IO completion of submitted IO. 12211da177e4SLinus Torvalds */ 12220e175a18SCurt Wohlgemuth void writeback_inodes_sb_nr(struct super_block *sb, 12230e175a18SCurt Wohlgemuth unsigned long nr, 12240e175a18SCurt Wohlgemuth enum wb_reason reason) 12251da177e4SLinus Torvalds { 122683ba7b07SChristoph Hellwig DECLARE_COMPLETION_ONSTACK(done); 122783ba7b07SChristoph Hellwig struct wb_writeback_work work = { 12283c4d7165SChristoph Hellwig .sb = sb, 12293c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_NONE, 12306e6938b6SWu Fengguang .tagged_writepages = 1, 123183ba7b07SChristoph Hellwig .done = &done, 12323259f8beSChris Mason .nr_pages = nr, 12330e175a18SCurt Wohlgemuth .reason = reason, 12343c4d7165SChristoph Hellwig }; 12350e3c9a22SJens Axboe 1236cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 123783ba7b07SChristoph Hellwig bdi_queue_work(sb->s_bdi, &work); 123883ba7b07SChristoph Hellwig wait_for_completion(&done); 12391da177e4SLinus Torvalds } 12403259f8beSChris Mason EXPORT_SYMBOL(writeback_inodes_sb_nr); 12413259f8beSChris Mason 12423259f8beSChris Mason /** 12433259f8beSChris Mason * writeback_inodes_sb - writeback dirty inodes from given super_block 12443259f8beSChris Mason * @sb: the superblock 1245786228abSMarcos Paulo de Souza * @reason: reason why some writeback work was initiated 12463259f8beSChris Mason * 12473259f8beSChris Mason * Start writeback on some inodes on this super_block. No guarantees are made 12483259f8beSChris Mason * on how many (if any) will be written, and this function does not wait 12493259f8beSChris Mason * for IO completion of submitted IO. 12503259f8beSChris Mason */ 12510e175a18SCurt Wohlgemuth void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) 12523259f8beSChris Mason { 12530e175a18SCurt Wohlgemuth return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); 12543259f8beSChris Mason } 1255d8a8559cSJens Axboe EXPORT_SYMBOL(writeback_inodes_sb); 1256d8a8559cSJens Axboe 1257d8a8559cSJens Axboe /** 125817bd55d0SEric Sandeen * writeback_inodes_sb_if_idle - start writeback if none underway 125917bd55d0SEric Sandeen * @sb: the superblock 1260786228abSMarcos Paulo de Souza * @reason: reason why some writeback work was initiated 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 */ 12650e175a18SCurt Wohlgemuth int writeback_inodes_sb_if_idle(struct super_block *sb, enum wb_reason reason) 126617bd55d0SEric Sandeen { 126717bd55d0SEric Sandeen if (!writeback_in_progress(sb->s_bdi)) { 1268cf37e972SChristoph Hellwig down_read(&sb->s_umount); 12690e175a18SCurt Wohlgemuth writeback_inodes_sb(sb, reason); 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 /** 1278c097b2caSFengguang Wu * writeback_inodes_sb_nr_if_idle - start writeback if none underway 12793259f8beSChris Mason * @sb: the superblock 12803259f8beSChris Mason * @nr: the number of pages to write 1281786228abSMarcos Paulo de Souza * @reason: reason why some writeback work was initiated 12823259f8beSChris Mason * 12833259f8beSChris Mason * Invoke writeback_inodes_sb if no writeback is currently underway. 12843259f8beSChris Mason * Returns 1 if writeback was started, 0 if not. 12853259f8beSChris Mason */ 12863259f8beSChris Mason int writeback_inodes_sb_nr_if_idle(struct super_block *sb, 12870e175a18SCurt Wohlgemuth unsigned long nr, 12880e175a18SCurt Wohlgemuth enum wb_reason reason) 12893259f8beSChris Mason { 12903259f8beSChris Mason if (!writeback_in_progress(sb->s_bdi)) { 12913259f8beSChris Mason down_read(&sb->s_umount); 12920e175a18SCurt Wohlgemuth writeback_inodes_sb_nr(sb, nr, reason); 12933259f8beSChris Mason up_read(&sb->s_umount); 12943259f8beSChris Mason return 1; 12953259f8beSChris Mason } else 12963259f8beSChris Mason return 0; 12973259f8beSChris Mason } 12983259f8beSChris Mason EXPORT_SYMBOL(writeback_inodes_sb_nr_if_idle); 12993259f8beSChris Mason 13003259f8beSChris Mason /** 1301d8a8559cSJens Axboe * sync_inodes_sb - sync sb inode pages 1302d8a8559cSJens Axboe * @sb: the superblock 1303d8a8559cSJens Axboe * 1304d8a8559cSJens Axboe * This function writes and waits on any dirty inode belonging to this 1305cb9ef8d5SStefan Hajnoczi * super_block. 1306d8a8559cSJens Axboe */ 1307b6e51316SJens Axboe void sync_inodes_sb(struct super_block *sb) 1308d8a8559cSJens Axboe { 130983ba7b07SChristoph Hellwig DECLARE_COMPLETION_ONSTACK(done); 131083ba7b07SChristoph Hellwig struct wb_writeback_work work = { 13113c4d7165SChristoph Hellwig .sb = sb, 13123c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_ALL, 13133c4d7165SChristoph Hellwig .nr_pages = LONG_MAX, 13143c4d7165SChristoph Hellwig .range_cyclic = 0, 131583ba7b07SChristoph Hellwig .done = &done, 13160e175a18SCurt Wohlgemuth .reason = WB_REASON_SYNC, 13173c4d7165SChristoph Hellwig }; 13183c4d7165SChristoph Hellwig 1319cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 1320cf37e972SChristoph Hellwig 132183ba7b07SChristoph Hellwig bdi_queue_work(sb->s_bdi, &work); 132283ba7b07SChristoph Hellwig wait_for_completion(&done); 132383ba7b07SChristoph Hellwig 1324b6e51316SJens Axboe wait_sb_inodes(sb); 1325d8a8559cSJens Axboe } 1326d8a8559cSJens Axboe EXPORT_SYMBOL(sync_inodes_sb); 13271da177e4SLinus Torvalds 13281da177e4SLinus Torvalds /** 13291da177e4SLinus Torvalds * write_inode_now - write an inode to disk 13301da177e4SLinus Torvalds * @inode: inode to write to disk 13311da177e4SLinus Torvalds * @sync: whether the write should be synchronous or not 13321da177e4SLinus Torvalds * 13337f04c26dSAndrea Arcangeli * This function commits an inode to disk immediately if it is dirty. This is 13347f04c26dSAndrea Arcangeli * primarily needed by knfsd. 13357f04c26dSAndrea Arcangeli * 13367f04c26dSAndrea Arcangeli * The caller must either have a ref on the inode or must have set I_WILL_FREE. 13371da177e4SLinus Torvalds */ 13381da177e4SLinus Torvalds int write_inode_now(struct inode *inode, int sync) 13391da177e4SLinus Torvalds { 1340f758eeabSChristoph Hellwig struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; 13411da177e4SLinus Torvalds int ret; 13421da177e4SLinus Torvalds struct writeback_control wbc = { 13431da177e4SLinus Torvalds .nr_to_write = LONG_MAX, 134418914b18SMike Galbraith .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, 1345111ebb6eSOGAWA Hirofumi .range_start = 0, 1346111ebb6eSOGAWA Hirofumi .range_end = LLONG_MAX, 13471da177e4SLinus Torvalds }; 13481da177e4SLinus Torvalds 13491da177e4SLinus Torvalds if (!mapping_cap_writeback_dirty(inode->i_mapping)) 135049364ce2SAndrew Morton wbc.nr_to_write = 0; 13511da177e4SLinus Torvalds 13521da177e4SLinus Torvalds might_sleep(); 1353f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 13540f1b1fd8SDave Chinner spin_lock(&inode->i_lock); 1355f758eeabSChristoph Hellwig ret = writeback_single_inode(inode, wb, &wbc); 13560f1b1fd8SDave Chinner spin_unlock(&inode->i_lock); 1357f758eeabSChristoph Hellwig spin_unlock(&wb->list_lock); 13581da177e4SLinus Torvalds return ret; 13591da177e4SLinus Torvalds } 13601da177e4SLinus Torvalds EXPORT_SYMBOL(write_inode_now); 13611da177e4SLinus Torvalds 13621da177e4SLinus Torvalds /** 13631da177e4SLinus Torvalds * sync_inode - write an inode and its pages to disk. 13641da177e4SLinus Torvalds * @inode: the inode to sync 13651da177e4SLinus Torvalds * @wbc: controls the writeback mode 13661da177e4SLinus Torvalds * 13671da177e4SLinus Torvalds * sync_inode() will write an inode and its pages to disk. It will also 13681da177e4SLinus Torvalds * correctly update the inode on its superblock's dirty inode lists and will 13691da177e4SLinus Torvalds * update inode->i_state. 13701da177e4SLinus Torvalds * 13711da177e4SLinus Torvalds * The caller must have a ref on the inode. 13721da177e4SLinus Torvalds */ 13731da177e4SLinus Torvalds int sync_inode(struct inode *inode, struct writeback_control *wbc) 13741da177e4SLinus Torvalds { 1375f758eeabSChristoph Hellwig struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; 13761da177e4SLinus Torvalds int ret; 13771da177e4SLinus Torvalds 1378f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 13790f1b1fd8SDave Chinner spin_lock(&inode->i_lock); 1380f758eeabSChristoph Hellwig ret = writeback_single_inode(inode, wb, wbc); 13810f1b1fd8SDave Chinner spin_unlock(&inode->i_lock); 1382f758eeabSChristoph Hellwig spin_unlock(&wb->list_lock); 13831da177e4SLinus Torvalds return ret; 13841da177e4SLinus Torvalds } 13851da177e4SLinus Torvalds EXPORT_SYMBOL(sync_inode); 1386c3765016SChristoph Hellwig 1387c3765016SChristoph Hellwig /** 1388c691b9d9SAndrew Morton * sync_inode_metadata - write an inode to disk 1389c3765016SChristoph Hellwig * @inode: the inode to sync 1390c3765016SChristoph Hellwig * @wait: wait for I/O to complete. 1391c3765016SChristoph Hellwig * 1392c691b9d9SAndrew Morton * Write an inode to disk and adjust its dirty state after completion. 1393c3765016SChristoph Hellwig * 1394c3765016SChristoph Hellwig * Note: only writes the actual inode, no associated data or other metadata. 1395c3765016SChristoph Hellwig */ 1396c3765016SChristoph Hellwig int sync_inode_metadata(struct inode *inode, int wait) 1397c3765016SChristoph Hellwig { 1398c3765016SChristoph Hellwig struct writeback_control wbc = { 1399c3765016SChristoph Hellwig .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE, 1400c3765016SChristoph Hellwig .nr_to_write = 0, /* metadata-only */ 1401c3765016SChristoph Hellwig }; 1402c3765016SChristoph Hellwig 1403c3765016SChristoph Hellwig return sync_inode(inode, &wbc); 1404c3765016SChristoph Hellwig } 1405c3765016SChristoph Hellwig EXPORT_SYMBOL(sync_inode_metadata); 1406