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> 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 { 2341c0eeaf5SJoern Engel /* 235a66979abSDave Chinner * Prevent speculative execution through 236f758eeabSChristoph Hellwig * spin_unlock(&wb->list_lock); 2371c0eeaf5SJoern Engel */ 238a66979abSDave Chinner 2391c0eeaf5SJoern Engel smp_mb(); 2401c0eeaf5SJoern Engel wake_up_bit(&inode->i_state, __I_SYNC); 2411c0eeaf5SJoern Engel } 2421c0eeaf5SJoern Engel 243d2caa3c5SJeff Layton static bool inode_dirtied_after(struct inode *inode, unsigned long t) 244d2caa3c5SJeff Layton { 245d2caa3c5SJeff Layton bool ret = time_after(inode->dirtied_when, t); 246d2caa3c5SJeff Layton #ifndef CONFIG_64BIT 247d2caa3c5SJeff Layton /* 248d2caa3c5SJeff Layton * For inodes being constantly redirtied, dirtied_when can get stuck. 249d2caa3c5SJeff Layton * It _appears_ to be in the future, but is actually in distant past. 250d2caa3c5SJeff Layton * This test is necessary to prevent such wrapped-around relative times 2515b0830cbSJens Axboe * from permanently stopping the whole bdi writeback. 252d2caa3c5SJeff Layton */ 253d2caa3c5SJeff Layton ret = ret && time_before_eq(inode->dirtied_when, jiffies); 254d2caa3c5SJeff Layton #endif 255d2caa3c5SJeff Layton return ret; 256d2caa3c5SJeff Layton } 257d2caa3c5SJeff Layton 258c986d1e2SAndrew Morton /* 2592c136579SFengguang Wu * Move expired dirty inodes from @delaying_queue to @dispatch_queue. 2602c136579SFengguang Wu */ 261e84d0a4fSWu Fengguang static int move_expired_inodes(struct list_head *delaying_queue, 2622c136579SFengguang Wu struct list_head *dispatch_queue, 263ad4e38ddSCurt Wohlgemuth struct wb_writeback_work *work) 2642c136579SFengguang Wu { 2655c03449dSShaohua Li LIST_HEAD(tmp); 2665c03449dSShaohua Li struct list_head *pos, *node; 267cf137307SJens Axboe struct super_block *sb = NULL; 2685c03449dSShaohua Li struct inode *inode; 269cf137307SJens Axboe int do_sb_sort = 0; 270e84d0a4fSWu Fengguang int moved = 0; 2715c03449dSShaohua Li 2722c136579SFengguang Wu while (!list_empty(delaying_queue)) { 2737ccf19a8SNick Piggin inode = wb_inode(delaying_queue->prev); 274ad4e38ddSCurt Wohlgemuth if (work->older_than_this && 275ad4e38ddSCurt Wohlgemuth inode_dirtied_after(inode, *work->older_than_this)) 2762c136579SFengguang Wu break; 277cf137307SJens Axboe if (sb && sb != inode->i_sb) 278cf137307SJens Axboe do_sb_sort = 1; 279cf137307SJens Axboe sb = inode->i_sb; 2807ccf19a8SNick Piggin list_move(&inode->i_wb_list, &tmp); 281e84d0a4fSWu Fengguang moved++; 2825c03449dSShaohua Li } 2835c03449dSShaohua Li 284cf137307SJens Axboe /* just one sb in list, splice to dispatch_queue and we're done */ 285cf137307SJens Axboe if (!do_sb_sort) { 286cf137307SJens Axboe list_splice(&tmp, dispatch_queue); 287e84d0a4fSWu Fengguang goto out; 288cf137307SJens Axboe } 289cf137307SJens Axboe 2905c03449dSShaohua Li /* Move inodes from one superblock together */ 2915c03449dSShaohua Li while (!list_empty(&tmp)) { 2927ccf19a8SNick Piggin sb = wb_inode(tmp.prev)->i_sb; 2935c03449dSShaohua Li list_for_each_prev_safe(pos, node, &tmp) { 2947ccf19a8SNick Piggin inode = wb_inode(pos); 2955c03449dSShaohua Li if (inode->i_sb == sb) 2967ccf19a8SNick Piggin list_move(&inode->i_wb_list, dispatch_queue); 2972c136579SFengguang Wu } 2982c136579SFengguang Wu } 299e84d0a4fSWu Fengguang out: 300e84d0a4fSWu Fengguang return moved; 3015c03449dSShaohua Li } 3022c136579SFengguang Wu 3032c136579SFengguang Wu /* 3042c136579SFengguang Wu * Queue all expired dirty inodes for io, eldest first. 3054ea879b9SWu Fengguang * Before 3064ea879b9SWu Fengguang * newly dirtied b_dirty b_io b_more_io 3074ea879b9SWu Fengguang * =============> gf edc BA 3084ea879b9SWu Fengguang * After 3094ea879b9SWu Fengguang * newly dirtied b_dirty b_io b_more_io 3104ea879b9SWu Fengguang * =============> g fBAedc 3114ea879b9SWu Fengguang * | 3124ea879b9SWu Fengguang * +--> dequeue for IO 3132c136579SFengguang Wu */ 314ad4e38ddSCurt Wohlgemuth static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work) 3152c136579SFengguang Wu { 316e84d0a4fSWu Fengguang int moved; 317f758eeabSChristoph Hellwig assert_spin_locked(&wb->list_lock); 3184ea879b9SWu Fengguang list_splice_init(&wb->b_more_io, &wb->b_io); 319ad4e38ddSCurt Wohlgemuth moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, work); 320ad4e38ddSCurt Wohlgemuth trace_writeback_queue_io(wb, work, moved); 32166f3b8e2SJens Axboe } 32266f3b8e2SJens Axboe 323a9185b41SChristoph Hellwig static int write_inode(struct inode *inode, struct writeback_control *wbc) 32466f3b8e2SJens Axboe { 32503ba3782SJens Axboe if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) 326a9185b41SChristoph Hellwig return inode->i_sb->s_op->write_inode(inode, wbc); 32703ba3782SJens Axboe return 0; 32866f3b8e2SJens Axboe } 32908d8e974SFengguang Wu 3302c136579SFengguang Wu /* 33101c03194SChristoph Hellwig * Wait for writeback on an inode to complete. 33201c03194SChristoph Hellwig */ 333f758eeabSChristoph Hellwig static void inode_wait_for_writeback(struct inode *inode, 334f758eeabSChristoph Hellwig struct bdi_writeback *wb) 33501c03194SChristoph Hellwig { 33601c03194SChristoph Hellwig DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); 33701c03194SChristoph Hellwig wait_queue_head_t *wqh; 33801c03194SChristoph Hellwig 33901c03194SChristoph Hellwig wqh = bit_waitqueue(&inode->i_state, __I_SYNC); 34058a9d3d8SRichard Kennedy while (inode->i_state & I_SYNC) { 341250df6edSDave Chinner spin_unlock(&inode->i_lock); 342f758eeabSChristoph Hellwig spin_unlock(&wb->list_lock); 34301c03194SChristoph Hellwig __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE); 344f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 345250df6edSDave Chinner spin_lock(&inode->i_lock); 34658a9d3d8SRichard Kennedy } 34701c03194SChristoph Hellwig } 34801c03194SChristoph Hellwig 34901c03194SChristoph Hellwig /* 350f758eeabSChristoph Hellwig * Write out an inode's dirty pages. Called under wb->list_lock and 3510f1b1fd8SDave Chinner * inode->i_lock. Either the caller has an active reference on the inode or 3520f1b1fd8SDave Chinner * the inode has I_WILL_FREE set. 35301c03194SChristoph Hellwig * 3541da177e4SLinus Torvalds * If `wait' is set, wait on the writeout. 3551da177e4SLinus Torvalds * 3561da177e4SLinus Torvalds * The whole writeout design is quite complex and fragile. We want to avoid 3571da177e4SLinus Torvalds * starvation of particular inodes when others are being redirtied, prevent 3581da177e4SLinus Torvalds * livelocks, etc. 3591da177e4SLinus Torvalds */ 3601da177e4SLinus Torvalds static int 361f758eeabSChristoph Hellwig writeback_single_inode(struct inode *inode, struct bdi_writeback *wb, 362f758eeabSChristoph Hellwig struct writeback_control *wbc) 3631da177e4SLinus Torvalds { 3641da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 365251d6a47SWu Fengguang long nr_to_write = wbc->nr_to_write; 36601c03194SChristoph Hellwig unsigned dirty; 3671da177e4SLinus Torvalds int ret; 3681da177e4SLinus Torvalds 369f758eeabSChristoph Hellwig assert_spin_locked(&wb->list_lock); 3700f1b1fd8SDave Chinner assert_spin_locked(&inode->i_lock); 3710f1b1fd8SDave Chinner 37201c03194SChristoph Hellwig if (!atomic_read(&inode->i_count)) 37301c03194SChristoph Hellwig WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); 37401c03194SChristoph Hellwig else 37501c03194SChristoph Hellwig WARN_ON(inode->i_state & I_WILL_FREE); 37601c03194SChristoph Hellwig 37701c03194SChristoph Hellwig if (inode->i_state & I_SYNC) { 37801c03194SChristoph Hellwig /* 37901c03194SChristoph Hellwig * If this inode is locked for writeback and we are not doing 38066f3b8e2SJens Axboe * writeback-for-data-integrity, move it to b_more_io so that 38101c03194SChristoph Hellwig * writeback can proceed with the other inodes on s_io. 38201c03194SChristoph Hellwig * 38301c03194SChristoph Hellwig * We'll have another go at writing back this inode when we 38466f3b8e2SJens Axboe * completed a full scan of b_io. 38501c03194SChristoph Hellwig */ 386a9185b41SChristoph Hellwig if (wbc->sync_mode != WB_SYNC_ALL) { 387f758eeabSChristoph Hellwig requeue_io(inode, wb); 388251d6a47SWu Fengguang trace_writeback_single_inode_requeue(inode, wbc, 389251d6a47SWu Fengguang nr_to_write); 39001c03194SChristoph Hellwig return 0; 39101c03194SChristoph Hellwig } 39201c03194SChristoph Hellwig 39301c03194SChristoph Hellwig /* 39401c03194SChristoph Hellwig * It's a data-integrity sync. We must wait. 39501c03194SChristoph Hellwig */ 396f758eeabSChristoph Hellwig inode_wait_for_writeback(inode, wb); 39701c03194SChristoph Hellwig } 39801c03194SChristoph Hellwig 3991c0eeaf5SJoern Engel BUG_ON(inode->i_state & I_SYNC); 4001da177e4SLinus Torvalds 4015547e8aaSDmitry Monakhov /* Set I_SYNC, reset I_DIRTY_PAGES */ 4021c0eeaf5SJoern Engel inode->i_state |= I_SYNC; 4035547e8aaSDmitry Monakhov inode->i_state &= ~I_DIRTY_PAGES; 404250df6edSDave Chinner spin_unlock(&inode->i_lock); 405f758eeabSChristoph Hellwig spin_unlock(&wb->list_lock); 4061da177e4SLinus Torvalds 4071da177e4SLinus Torvalds ret = do_writepages(mapping, wbc); 4081da177e4SLinus Torvalds 40926821ed4SChristoph Hellwig /* 41026821ed4SChristoph Hellwig * Make sure to wait on the data before writing out the metadata. 41126821ed4SChristoph Hellwig * This is important for filesystems that modify metadata on data 41226821ed4SChristoph Hellwig * I/O completion. 41326821ed4SChristoph Hellwig */ 414a9185b41SChristoph Hellwig if (wbc->sync_mode == WB_SYNC_ALL) { 41526821ed4SChristoph Hellwig int err = filemap_fdatawait(mapping); 4161da177e4SLinus Torvalds if (ret == 0) 4171da177e4SLinus Torvalds ret = err; 4181da177e4SLinus Torvalds } 4191da177e4SLinus Torvalds 4205547e8aaSDmitry Monakhov /* 4215547e8aaSDmitry Monakhov * Some filesystems may redirty the inode during the writeback 4225547e8aaSDmitry Monakhov * due to delalloc, clear dirty metadata flags right before 4235547e8aaSDmitry Monakhov * write_inode() 4245547e8aaSDmitry Monakhov */ 425250df6edSDave Chinner spin_lock(&inode->i_lock); 4265547e8aaSDmitry Monakhov dirty = inode->i_state & I_DIRTY; 4275547e8aaSDmitry Monakhov inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC); 428250df6edSDave Chinner spin_unlock(&inode->i_lock); 42926821ed4SChristoph Hellwig /* Don't write the inode if only I_DIRTY_PAGES was set */ 43026821ed4SChristoph Hellwig if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { 431a9185b41SChristoph Hellwig int err = write_inode(inode, wbc); 4321da177e4SLinus Torvalds if (ret == 0) 4331da177e4SLinus Torvalds ret = err; 4341da177e4SLinus Torvalds } 4351da177e4SLinus Torvalds 436f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 437250df6edSDave Chinner spin_lock(&inode->i_lock); 4381c0eeaf5SJoern Engel inode->i_state &= ~I_SYNC; 439a4ffdde6SAl Viro if (!(inode->i_state & I_FREEING)) { 44094c3dcbbSWu Fengguang /* 44194c3dcbbSWu Fengguang * Sync livelock prevention. Each inode is tagged and synced in 44294c3dcbbSWu Fengguang * one shot. If still dirty, it will be redirty_tail()'ed below. 44394c3dcbbSWu Fengguang * Update the dirty time to prevent enqueue and sync it again. 44494c3dcbbSWu Fengguang */ 44594c3dcbbSWu Fengguang if ((inode->i_state & I_DIRTY) && 44694c3dcbbSWu Fengguang (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)) 44794c3dcbbSWu Fengguang inode->dirtied_when = jiffies; 44894c3dcbbSWu Fengguang 44923539afcSWu Fengguang if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { 4501da177e4SLinus Torvalds /* 4511da177e4SLinus Torvalds * We didn't write back all the pages. nfs_writepages() 452a50aeb40SWu Fengguang * sometimes bales out without doing anything. 4531da177e4SLinus Torvalds */ 4541da177e4SLinus Torvalds inode->i_state |= I_DIRTY_PAGES; 4558bc3be27SFengguang Wu if (wbc->nr_to_write <= 0) { 4568bc3be27SFengguang Wu /* 4578bc3be27SFengguang Wu * slice used up: queue for next turn 4588bc3be27SFengguang Wu */ 459f758eeabSChristoph Hellwig requeue_io(inode, wb); 4601da177e4SLinus Torvalds } else { 4611da177e4SLinus Torvalds /* 462a50aeb40SWu Fengguang * Writeback blocked by something other than 463a50aeb40SWu Fengguang * congestion. Delay the inode for some time to 464a50aeb40SWu Fengguang * avoid spinning on the CPU (100% iowait) 465a50aeb40SWu Fengguang * retrying writeback of the dirty page/inode 466a50aeb40SWu Fengguang * that cannot be performed immediately. 4678bc3be27SFengguang Wu */ 468f758eeabSChristoph Hellwig redirty_tail(inode, wb); 4698bc3be27SFengguang Wu } 47023539afcSWu Fengguang } else if (inode->i_state & I_DIRTY) { 47123539afcSWu Fengguang /* 47223539afcSWu Fengguang * Filesystems can dirty the inode during writeback 47323539afcSWu Fengguang * operations, such as delayed allocation during 47423539afcSWu Fengguang * submission or metadata updates after data IO 47523539afcSWu Fengguang * completion. 47623539afcSWu Fengguang */ 477f758eeabSChristoph Hellwig redirty_tail(inode, wb); 4781da177e4SLinus Torvalds } else { 4791da177e4SLinus Torvalds /* 4809e38d86fSNick Piggin * The inode is clean. At this point we either have 4819e38d86fSNick Piggin * a reference to the inode or it's on it's way out. 4829e38d86fSNick Piggin * No need to add it back to the LRU. 4831da177e4SLinus Torvalds */ 4847ccf19a8SNick Piggin list_del_init(&inode->i_wb_list); 4851da177e4SLinus Torvalds } 4861da177e4SLinus Torvalds } 4871c0eeaf5SJoern Engel inode_sync_complete(inode); 488251d6a47SWu Fengguang trace_writeback_single_inode(inode, wbc, nr_to_write); 4891da177e4SLinus Torvalds return ret; 4901da177e4SLinus Torvalds } 4911da177e4SLinus Torvalds 4921a12d8bdSWu Fengguang static long writeback_chunk_size(struct backing_dev_info *bdi, 4931a12d8bdSWu Fengguang struct wb_writeback_work *work) 494d46db3d5SWu Fengguang { 495d46db3d5SWu Fengguang long pages; 496d46db3d5SWu Fengguang 497d46db3d5SWu Fengguang /* 498d46db3d5SWu Fengguang * WB_SYNC_ALL mode does livelock avoidance by syncing dirty 499d46db3d5SWu Fengguang * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX 500d46db3d5SWu Fengguang * here avoids calling into writeback_inodes_wb() more than once. 501d46db3d5SWu Fengguang * 502d46db3d5SWu Fengguang * The intended call sequence for WB_SYNC_ALL writeback is: 503d46db3d5SWu Fengguang * 504d46db3d5SWu Fengguang * wb_writeback() 505d46db3d5SWu Fengguang * writeback_sb_inodes() <== called only once 506d46db3d5SWu Fengguang * write_cache_pages() <== called once for each inode 507d46db3d5SWu Fengguang * (quickly) tag currently dirty pages 508d46db3d5SWu Fengguang * (maybe slowly) sync all tagged pages 509d46db3d5SWu Fengguang */ 510d46db3d5SWu Fengguang if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages) 511d46db3d5SWu Fengguang pages = LONG_MAX; 5121a12d8bdSWu Fengguang else { 5131a12d8bdSWu Fengguang pages = min(bdi->avg_write_bandwidth / 2, 5141a12d8bdSWu Fengguang global_dirty_limit / DIRTY_SCOPE); 5151a12d8bdSWu Fengguang pages = min(pages, work->nr_pages); 5161a12d8bdSWu Fengguang pages = round_down(pages + MIN_WRITEBACK_PAGES, 5171a12d8bdSWu Fengguang MIN_WRITEBACK_PAGES); 5181a12d8bdSWu Fengguang } 519d46db3d5SWu Fengguang 520d46db3d5SWu Fengguang return pages; 521d46db3d5SWu Fengguang } 522d46db3d5SWu Fengguang 52303ba3782SJens Axboe /* 524f11c9c5cSEdward Shishkin * Write a portion of b_io inodes which belong to @sb. 525edadfb10SChristoph Hellwig * 526edadfb10SChristoph Hellwig * If @only_this_sb is true, then find and write all such 527f11c9c5cSEdward Shishkin * inodes. Otherwise write only ones which go sequentially 528f11c9c5cSEdward Shishkin * in reverse order. 529edadfb10SChristoph Hellwig * 530d46db3d5SWu Fengguang * Return the number of pages and/or inodes written. 531f11c9c5cSEdward Shishkin */ 532d46db3d5SWu Fengguang static long writeback_sb_inodes(struct super_block *sb, 533d46db3d5SWu Fengguang struct bdi_writeback *wb, 534d46db3d5SWu Fengguang struct wb_writeback_work *work) 53503ba3782SJens Axboe { 536d46db3d5SWu Fengguang struct writeback_control wbc = { 537d46db3d5SWu Fengguang .sync_mode = work->sync_mode, 538d46db3d5SWu Fengguang .tagged_writepages = work->tagged_writepages, 539d46db3d5SWu Fengguang .for_kupdate = work->for_kupdate, 540d46db3d5SWu Fengguang .for_background = work->for_background, 541d46db3d5SWu Fengguang .range_cyclic = work->range_cyclic, 542d46db3d5SWu Fengguang .range_start = 0, 543d46db3d5SWu Fengguang .range_end = LLONG_MAX, 544d46db3d5SWu Fengguang }; 545d46db3d5SWu Fengguang unsigned long start_time = jiffies; 546d46db3d5SWu Fengguang long write_chunk; 547d46db3d5SWu Fengguang long wrote = 0; /* count both pages and inodes */ 548d46db3d5SWu Fengguang 54903ba3782SJens Axboe while (!list_empty(&wb->b_io)) { 5507ccf19a8SNick Piggin struct inode *inode = wb_inode(wb->b_io.prev); 551edadfb10SChristoph Hellwig 552edadfb10SChristoph Hellwig if (inode->i_sb != sb) { 553d46db3d5SWu Fengguang if (work->sb) { 554edadfb10SChristoph Hellwig /* 555edadfb10SChristoph Hellwig * We only want to write back data for this 556edadfb10SChristoph Hellwig * superblock, move all inodes not belonging 557edadfb10SChristoph Hellwig * to it back onto the dirty list. 558edadfb10SChristoph Hellwig */ 559f758eeabSChristoph Hellwig redirty_tail(inode, wb); 56066f3b8e2SJens Axboe continue; 56166f3b8e2SJens Axboe } 562edadfb10SChristoph Hellwig 563edadfb10SChristoph Hellwig /* 564edadfb10SChristoph Hellwig * The inode belongs to a different superblock. 565edadfb10SChristoph Hellwig * Bounce back to the caller to unpin this and 566edadfb10SChristoph Hellwig * pin the next superblock. 567edadfb10SChristoph Hellwig */ 568d46db3d5SWu Fengguang break; 569edadfb10SChristoph Hellwig } 570edadfb10SChristoph Hellwig 5719843b76aSChristoph Hellwig /* 5729843b76aSChristoph Hellwig * Don't bother with new inodes or inodes beeing freed, first 5739843b76aSChristoph Hellwig * kind does not need peridic writeout yet, and for the latter 5749843b76aSChristoph Hellwig * kind writeout is handled by the freer. 5759843b76aSChristoph Hellwig */ 576250df6edSDave Chinner spin_lock(&inode->i_lock); 5779843b76aSChristoph Hellwig if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { 578250df6edSDave Chinner spin_unlock(&inode->i_lock); 579fcc5c222SWu Fengguang redirty_tail(inode, wb); 5807ef0d737SNick Piggin continue; 5817ef0d737SNick Piggin } 5821da177e4SLinus Torvalds __iget(inode); 5831a12d8bdSWu Fengguang write_chunk = writeback_chunk_size(wb->bdi, work); 584d46db3d5SWu Fengguang wbc.nr_to_write = write_chunk; 585d46db3d5SWu Fengguang wbc.pages_skipped = 0; 586250df6edSDave Chinner 587d46db3d5SWu Fengguang writeback_single_inode(inode, wb, &wbc); 588d46db3d5SWu Fengguang 589d46db3d5SWu Fengguang work->nr_pages -= write_chunk - wbc.nr_to_write; 590d46db3d5SWu Fengguang wrote += write_chunk - wbc.nr_to_write; 591d46db3d5SWu Fengguang if (!(inode->i_state & I_DIRTY)) 592d46db3d5SWu Fengguang wrote++; 593d46db3d5SWu Fengguang if (wbc.pages_skipped) { 5941da177e4SLinus Torvalds /* 5951da177e4SLinus Torvalds * writeback is not making progress due to locked 5961da177e4SLinus Torvalds * buffers. Skip this inode for now. 5971da177e4SLinus Torvalds */ 598f758eeabSChristoph Hellwig redirty_tail(inode, wb); 5991da177e4SLinus Torvalds } 6000f1b1fd8SDave Chinner spin_unlock(&inode->i_lock); 601f758eeabSChristoph Hellwig spin_unlock(&wb->list_lock); 6021da177e4SLinus Torvalds iput(inode); 6034ffc8444SOGAWA Hirofumi cond_resched(); 604f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 605d46db3d5SWu Fengguang /* 606d46db3d5SWu Fengguang * bail out to wb_writeback() often enough to check 607d46db3d5SWu Fengguang * background threshold and other termination conditions. 608d46db3d5SWu Fengguang */ 609d46db3d5SWu Fengguang if (wrote) { 610d46db3d5SWu Fengguang if (time_is_before_jiffies(start_time + HZ / 10UL)) 611d46db3d5SWu Fengguang break; 612d46db3d5SWu Fengguang if (work->nr_pages <= 0) 613d46db3d5SWu Fengguang break; 6141da177e4SLinus Torvalds } 6158bc3be27SFengguang Wu } 616d46db3d5SWu Fengguang return wrote; 617f11c9c5cSEdward Shishkin } 61838f21977SNick Piggin 619d46db3d5SWu Fengguang static long __writeback_inodes_wb(struct bdi_writeback *wb, 620d46db3d5SWu Fengguang struct wb_writeback_work *work) 621f11c9c5cSEdward Shishkin { 622d46db3d5SWu Fengguang unsigned long start_time = jiffies; 623d46db3d5SWu Fengguang long wrote = 0; 624f11c9c5cSEdward Shishkin 625f11c9c5cSEdward Shishkin while (!list_empty(&wb->b_io)) { 6267ccf19a8SNick Piggin struct inode *inode = wb_inode(wb->b_io.prev); 627f11c9c5cSEdward Shishkin struct super_block *sb = inode->i_sb; 628f11c9c5cSEdward Shishkin 62912ad3ab6SDave Chinner if (!grab_super_passive(sb)) { 6300e995816SWu Fengguang /* 6310e995816SWu Fengguang * grab_super_passive() may fail consistently due to 6320e995816SWu Fengguang * s_umount being grabbed by someone else. Don't use 6330e995816SWu Fengguang * requeue_io() to avoid busy retrying the inode/sb. 6340e995816SWu Fengguang */ 6350e995816SWu Fengguang redirty_tail(inode, wb); 636d19de7edSChristoph Hellwig continue; 637334132aeSChristoph Hellwig } 638d46db3d5SWu Fengguang wrote += writeback_sb_inodes(sb, wb, work); 639d19de7edSChristoph Hellwig drop_super(sb); 640f11c9c5cSEdward Shishkin 641d46db3d5SWu Fengguang /* refer to the same tests at the end of writeback_sb_inodes */ 642d46db3d5SWu Fengguang if (wrote) { 643d46db3d5SWu Fengguang if (time_is_before_jiffies(start_time + HZ / 10UL)) 644d46db3d5SWu Fengguang break; 645d46db3d5SWu Fengguang if (work->nr_pages <= 0) 646f11c9c5cSEdward Shishkin break; 647f11c9c5cSEdward Shishkin } 648d46db3d5SWu Fengguang } 64966f3b8e2SJens Axboe /* Leave any unwritten inodes on b_io */ 650d46db3d5SWu Fengguang return wrote; 65166f3b8e2SJens Axboe } 65266f3b8e2SJens Axboe 6530e175a18SCurt Wohlgemuth long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages, 6540e175a18SCurt Wohlgemuth enum wb_reason reason) 655edadfb10SChristoph Hellwig { 656d46db3d5SWu Fengguang struct wb_writeback_work work = { 657d46db3d5SWu Fengguang .nr_pages = nr_pages, 658d46db3d5SWu Fengguang .sync_mode = WB_SYNC_NONE, 659d46db3d5SWu Fengguang .range_cyclic = 1, 6600e175a18SCurt Wohlgemuth .reason = reason, 661d46db3d5SWu Fengguang }; 662edadfb10SChristoph Hellwig 663f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 664424b351fSWu Fengguang if (list_empty(&wb->b_io)) 665ad4e38ddSCurt Wohlgemuth queue_io(wb, &work); 666d46db3d5SWu Fengguang __writeback_inodes_wb(wb, &work); 667f758eeabSChristoph Hellwig spin_unlock(&wb->list_lock); 668edadfb10SChristoph Hellwig 669d46db3d5SWu Fengguang return nr_pages - work.nr_pages; 67066f3b8e2SJens Axboe } 67166f3b8e2SJens Axboe 672b00949aaSWu Fengguang static bool over_bground_thresh(struct backing_dev_info *bdi) 67303ba3782SJens Axboe { 67403ba3782SJens Axboe unsigned long background_thresh, dirty_thresh; 67503ba3782SJens Axboe 67616c4042fSWu Fengguang global_dirty_limits(&background_thresh, &dirty_thresh); 67703ba3782SJens Axboe 678b00949aaSWu Fengguang if (global_page_state(NR_FILE_DIRTY) + 679b00949aaSWu Fengguang global_page_state(NR_UNSTABLE_NFS) > background_thresh) 680b00949aaSWu Fengguang return true; 681b00949aaSWu Fengguang 682b00949aaSWu Fengguang if (bdi_stat(bdi, BDI_RECLAIMABLE) > 683b00949aaSWu Fengguang bdi_dirty_limit(bdi, background_thresh)) 684b00949aaSWu Fengguang return true; 685b00949aaSWu Fengguang 686b00949aaSWu Fengguang return false; 68703ba3782SJens Axboe } 68803ba3782SJens Axboe 68903ba3782SJens Axboe /* 690e98be2d5SWu Fengguang * Called under wb->list_lock. If there are multiple wb per bdi, 691e98be2d5SWu Fengguang * only the flusher working on the first wb should do it. 692e98be2d5SWu Fengguang */ 693e98be2d5SWu Fengguang static void wb_update_bandwidth(struct bdi_writeback *wb, 694e98be2d5SWu Fengguang unsigned long start_time) 695e98be2d5SWu Fengguang { 696af6a3113SWu Fengguang __bdi_update_bandwidth(wb->bdi, 0, 0, 0, 0, 0, start_time); 697e98be2d5SWu Fengguang } 698e98be2d5SWu Fengguang 699e98be2d5SWu Fengguang /* 70003ba3782SJens Axboe * Explicit flushing or periodic writeback of "old" data. 70103ba3782SJens Axboe * 70203ba3782SJens Axboe * Define "old": the first time one of an inode's pages is dirtied, we mark the 70303ba3782SJens Axboe * dirtying-time in the inode's address_space. So this periodic writeback code 70403ba3782SJens Axboe * just walks the superblock inode list, writing back any inodes which are 70503ba3782SJens Axboe * older than a specific point in time. 70603ba3782SJens Axboe * 70703ba3782SJens Axboe * Try to run once per dirty_writeback_interval. But if a writeback event 70803ba3782SJens Axboe * takes longer than a dirty_writeback_interval interval, then leave a 70903ba3782SJens Axboe * one-second gap. 71003ba3782SJens Axboe * 71103ba3782SJens Axboe * older_than_this takes precedence over nr_to_write. So we'll only write back 71203ba3782SJens Axboe * all dirty pages if they are all attached to "old" mappings. 71303ba3782SJens Axboe */ 714c4a77a6cSJens Axboe static long wb_writeback(struct bdi_writeback *wb, 71583ba7b07SChristoph Hellwig struct wb_writeback_work *work) 71603ba3782SJens Axboe { 717e98be2d5SWu Fengguang unsigned long wb_start = jiffies; 718d46db3d5SWu Fengguang long nr_pages = work->nr_pages; 71903ba3782SJens Axboe unsigned long oldest_jif; 720a5989bdcSJan Kara struct inode *inode; 721d46db3d5SWu Fengguang long progress; 72203ba3782SJens Axboe 723e185dda8SWu Fengguang oldest_jif = jiffies; 724d46db3d5SWu Fengguang work->older_than_this = &oldest_jif; 72503ba3782SJens Axboe 726e8dfc305SWu Fengguang spin_lock(&wb->list_lock); 72703ba3782SJens Axboe for (;;) { 72803ba3782SJens Axboe /* 729d3ddec76SWu Fengguang * Stop writeback when nr_pages has been consumed 73003ba3782SJens Axboe */ 73183ba7b07SChristoph Hellwig if (work->nr_pages <= 0) 73203ba3782SJens Axboe break; 73303ba3782SJens Axboe 73403ba3782SJens Axboe /* 735aa373cf5SJan Kara * Background writeout and kupdate-style writeback may 736aa373cf5SJan Kara * run forever. Stop them if there is other work to do 737aa373cf5SJan Kara * so that e.g. sync can proceed. They'll be restarted 738aa373cf5SJan Kara * after the other works are all done. 739aa373cf5SJan Kara */ 740aa373cf5SJan Kara if ((work->for_background || work->for_kupdate) && 741aa373cf5SJan Kara !list_empty(&wb->bdi->work_list)) 742aa373cf5SJan Kara break; 743aa373cf5SJan Kara 744aa373cf5SJan Kara /* 745d3ddec76SWu Fengguang * For background writeout, stop when we are below the 746d3ddec76SWu Fengguang * background dirty threshold 74703ba3782SJens Axboe */ 748b00949aaSWu Fengguang if (work->for_background && !over_bground_thresh(wb->bdi)) 74903ba3782SJens Axboe break; 75003ba3782SJens Axboe 7511bc36b64SJan Kara /* 7521bc36b64SJan Kara * Kupdate and background works are special and we want to 7531bc36b64SJan Kara * include all inodes that need writing. Livelock avoidance is 7541bc36b64SJan Kara * handled by these works yielding to any other work so we are 7551bc36b64SJan Kara * safe. 7561bc36b64SJan Kara */ 757ba9aa839SWu Fengguang if (work->for_kupdate) { 758ba9aa839SWu Fengguang oldest_jif = jiffies - 759ba9aa839SWu Fengguang msecs_to_jiffies(dirty_expire_interval * 10); 7601bc36b64SJan Kara } else if (work->for_background) 7611bc36b64SJan Kara oldest_jif = jiffies; 762028c2dd1SDave Chinner 763d46db3d5SWu Fengguang trace_writeback_start(wb->bdi, work); 764e8dfc305SWu Fengguang if (list_empty(&wb->b_io)) 765ad4e38ddSCurt Wohlgemuth queue_io(wb, work); 76683ba7b07SChristoph Hellwig if (work->sb) 767d46db3d5SWu Fengguang progress = writeback_sb_inodes(work->sb, wb, work); 768edadfb10SChristoph Hellwig else 769d46db3d5SWu Fengguang progress = __writeback_inodes_wb(wb, work); 770d46db3d5SWu Fengguang trace_writeback_written(wb->bdi, work); 771028c2dd1SDave Chinner 772e98be2d5SWu Fengguang wb_update_bandwidth(wb, wb_start); 77303ba3782SJens Axboe 77403ba3782SJens Axboe /* 77571fd05a8SJens Axboe * Did we write something? Try for more 776e6fb6da2SWu Fengguang * 777e6fb6da2SWu Fengguang * Dirty inodes are moved to b_io for writeback in batches. 778e6fb6da2SWu Fengguang * The completion of the current batch does not necessarily 779e6fb6da2SWu Fengguang * mean the overall work is done. So we keep looping as long 780e6fb6da2SWu Fengguang * as made some progress on cleaning pages or inodes. 78171fd05a8SJens Axboe */ 782d46db3d5SWu Fengguang if (progress) 78303ba3782SJens Axboe continue; 784a5989bdcSJan Kara /* 785e6fb6da2SWu Fengguang * No more inodes for IO, bail 786a5989bdcSJan Kara */ 787b7a2441fSWu Fengguang if (list_empty(&wb->b_more_io)) 78803ba3782SJens Axboe break; 78903ba3782SJens Axboe /* 7908010c3b6SJens Axboe * Nothing written. Wait for some inode to 7918010c3b6SJens Axboe * become available for writeback. Otherwise 7928010c3b6SJens Axboe * we'll just busyloop. 79303ba3782SJens Axboe */ 79403ba3782SJens Axboe if (!list_empty(&wb->b_more_io)) { 795d46db3d5SWu Fengguang trace_writeback_wait(wb->bdi, work); 79603ba3782SJens Axboe inode = wb_inode(wb->b_more_io.prev); 797250df6edSDave Chinner spin_lock(&inode->i_lock); 798f758eeabSChristoph Hellwig inode_wait_for_writeback(inode, wb); 799250df6edSDave Chinner spin_unlock(&inode->i_lock); 80003ba3782SJens Axboe } 80103ba3782SJens Axboe } 802e8dfc305SWu Fengguang spin_unlock(&wb->list_lock); 80303ba3782SJens Axboe 804d46db3d5SWu Fengguang return nr_pages - work->nr_pages; 80503ba3782SJens Axboe } 80603ba3782SJens Axboe 80703ba3782SJens Axboe /* 80883ba7b07SChristoph Hellwig * Return the next wb_writeback_work struct that hasn't been processed yet. 80903ba3782SJens Axboe */ 81083ba7b07SChristoph Hellwig static struct wb_writeback_work * 81108852b6dSMinchan Kim get_next_work_item(struct backing_dev_info *bdi) 81203ba3782SJens Axboe { 81383ba7b07SChristoph Hellwig struct wb_writeback_work *work = NULL; 81403ba3782SJens Axboe 8156467716aSArtem Bityutskiy spin_lock_bh(&bdi->wb_lock); 81683ba7b07SChristoph Hellwig if (!list_empty(&bdi->work_list)) { 81783ba7b07SChristoph Hellwig work = list_entry(bdi->work_list.next, 81883ba7b07SChristoph Hellwig struct wb_writeback_work, list); 81983ba7b07SChristoph Hellwig list_del_init(&work->list); 82003ba3782SJens Axboe } 8216467716aSArtem Bityutskiy spin_unlock_bh(&bdi->wb_lock); 82283ba7b07SChristoph Hellwig return work; 82303ba3782SJens Axboe } 82403ba3782SJens Axboe 825cdf01dd5SLinus Torvalds /* 826cdf01dd5SLinus Torvalds * Add in the number of potentially dirty inodes, because each inode 827cdf01dd5SLinus Torvalds * write can dirty pagecache in the underlying blockdev. 828cdf01dd5SLinus Torvalds */ 829cdf01dd5SLinus Torvalds static unsigned long get_nr_dirty_pages(void) 830cdf01dd5SLinus Torvalds { 831cdf01dd5SLinus Torvalds return global_page_state(NR_FILE_DIRTY) + 832cdf01dd5SLinus Torvalds global_page_state(NR_UNSTABLE_NFS) + 833cdf01dd5SLinus Torvalds get_nr_dirty_inodes(); 834cdf01dd5SLinus Torvalds } 835cdf01dd5SLinus Torvalds 8366585027aSJan Kara static long wb_check_background_flush(struct bdi_writeback *wb) 8376585027aSJan Kara { 838b00949aaSWu Fengguang if (over_bground_thresh(wb->bdi)) { 8396585027aSJan Kara 8406585027aSJan Kara struct wb_writeback_work work = { 8416585027aSJan Kara .nr_pages = LONG_MAX, 8426585027aSJan Kara .sync_mode = WB_SYNC_NONE, 8436585027aSJan Kara .for_background = 1, 8446585027aSJan Kara .range_cyclic = 1, 8450e175a18SCurt Wohlgemuth .reason = WB_REASON_BACKGROUND, 8466585027aSJan Kara }; 8476585027aSJan Kara 8486585027aSJan Kara return wb_writeback(wb, &work); 8496585027aSJan Kara } 8506585027aSJan Kara 8516585027aSJan Kara return 0; 8526585027aSJan Kara } 8536585027aSJan Kara 85403ba3782SJens Axboe static long wb_check_old_data_flush(struct bdi_writeback *wb) 85503ba3782SJens Axboe { 85603ba3782SJens Axboe unsigned long expired; 85703ba3782SJens Axboe long nr_pages; 85803ba3782SJens Axboe 85969b62d01SJens Axboe /* 86069b62d01SJens Axboe * When set to zero, disable periodic writeback 86169b62d01SJens Axboe */ 86269b62d01SJens Axboe if (!dirty_writeback_interval) 86369b62d01SJens Axboe return 0; 86469b62d01SJens Axboe 86503ba3782SJens Axboe expired = wb->last_old_flush + 86603ba3782SJens Axboe msecs_to_jiffies(dirty_writeback_interval * 10); 86703ba3782SJens Axboe if (time_before(jiffies, expired)) 86803ba3782SJens Axboe return 0; 86903ba3782SJens Axboe 87003ba3782SJens Axboe wb->last_old_flush = jiffies; 871cdf01dd5SLinus Torvalds nr_pages = get_nr_dirty_pages(); 87203ba3782SJens Axboe 873c4a77a6cSJens Axboe if (nr_pages) { 87483ba7b07SChristoph Hellwig struct wb_writeback_work work = { 875c4a77a6cSJens Axboe .nr_pages = nr_pages, 876c4a77a6cSJens Axboe .sync_mode = WB_SYNC_NONE, 877c4a77a6cSJens Axboe .for_kupdate = 1, 878c4a77a6cSJens Axboe .range_cyclic = 1, 8790e175a18SCurt Wohlgemuth .reason = WB_REASON_PERIODIC, 880c4a77a6cSJens Axboe }; 881c4a77a6cSJens Axboe 88283ba7b07SChristoph Hellwig return wb_writeback(wb, &work); 883c4a77a6cSJens Axboe } 88403ba3782SJens Axboe 88503ba3782SJens Axboe return 0; 88603ba3782SJens Axboe } 88703ba3782SJens Axboe 88803ba3782SJens Axboe /* 88903ba3782SJens Axboe * Retrieve work items and do the writeback they describe 89003ba3782SJens Axboe */ 89103ba3782SJens Axboe long wb_do_writeback(struct bdi_writeback *wb, int force_wait) 89203ba3782SJens Axboe { 89303ba3782SJens Axboe struct backing_dev_info *bdi = wb->bdi; 89483ba7b07SChristoph Hellwig struct wb_writeback_work *work; 895c4a77a6cSJens Axboe long wrote = 0; 89603ba3782SJens Axboe 89781d73a32SJan Kara set_bit(BDI_writeback_running, &wb->bdi->state); 89808852b6dSMinchan Kim while ((work = get_next_work_item(bdi)) != NULL) { 89903ba3782SJens Axboe /* 90003ba3782SJens Axboe * Override sync mode, in case we must wait for completion 90183ba7b07SChristoph Hellwig * because this thread is exiting now. 90203ba3782SJens Axboe */ 90303ba3782SJens Axboe if (force_wait) 90483ba7b07SChristoph Hellwig work->sync_mode = WB_SYNC_ALL; 90583ba7b07SChristoph Hellwig 906455b2864SDave Chinner trace_writeback_exec(bdi, work); 907455b2864SDave Chinner 90883ba7b07SChristoph Hellwig wrote += wb_writeback(wb, work); 90903ba3782SJens Axboe 91003ba3782SJens Axboe /* 91183ba7b07SChristoph Hellwig * Notify the caller of completion if this is a synchronous 91283ba7b07SChristoph Hellwig * work item, otherwise just free it. 91303ba3782SJens Axboe */ 91483ba7b07SChristoph Hellwig if (work->done) 91583ba7b07SChristoph Hellwig complete(work->done); 91683ba7b07SChristoph Hellwig else 91783ba7b07SChristoph Hellwig kfree(work); 91803ba3782SJens Axboe } 91903ba3782SJens Axboe 92003ba3782SJens Axboe /* 92103ba3782SJens Axboe * Check for periodic writeback, kupdated() style 92203ba3782SJens Axboe */ 92303ba3782SJens Axboe wrote += wb_check_old_data_flush(wb); 9246585027aSJan Kara wrote += wb_check_background_flush(wb); 92581d73a32SJan Kara clear_bit(BDI_writeback_running, &wb->bdi->state); 92603ba3782SJens Axboe 92703ba3782SJens Axboe return wrote; 92803ba3782SJens Axboe } 92903ba3782SJens Axboe 93003ba3782SJens Axboe /* 93103ba3782SJens Axboe * Handle writeback of dirty data for the device backed by this bdi. Also 93203ba3782SJens Axboe * wakes up periodically and does kupdated style flushing. 93303ba3782SJens Axboe */ 93408243900SChristoph Hellwig int bdi_writeback_thread(void *data) 93503ba3782SJens Axboe { 93608243900SChristoph Hellwig struct bdi_writeback *wb = data; 93708243900SChristoph Hellwig struct backing_dev_info *bdi = wb->bdi; 93803ba3782SJens Axboe long pages_written; 93903ba3782SJens Axboe 940766f9164SPeter Zijlstra current->flags |= PF_SWAPWRITE; 94108243900SChristoph Hellwig set_freezable(); 942ecd58403SArtem Bityutskiy wb->last_active = jiffies; 94303ba3782SJens Axboe 94403ba3782SJens Axboe /* 94508243900SChristoph Hellwig * Our parent may run at a different priority, just set us to normal 94603ba3782SJens Axboe */ 94708243900SChristoph Hellwig set_user_nice(current, 0); 94808243900SChristoph Hellwig 949455b2864SDave Chinner trace_writeback_thread_start(bdi); 950455b2864SDave Chinner 9518a32c441STejun Heo while (!kthread_freezable_should_stop(NULL)) { 9526467716aSArtem Bityutskiy /* 9536467716aSArtem Bityutskiy * Remove own delayed wake-up timer, since we are already awake 9546467716aSArtem Bityutskiy * and we'll take care of the preriodic write-back. 9556467716aSArtem Bityutskiy */ 9566467716aSArtem Bityutskiy del_timer(&wb->wakeup_timer); 9576467716aSArtem Bityutskiy 95803ba3782SJens Axboe pages_written = wb_do_writeback(wb, 0); 95903ba3782SJens Axboe 960455b2864SDave Chinner trace_writeback_pages_written(pages_written); 961455b2864SDave Chinner 96203ba3782SJens Axboe if (pages_written) 963ecd58403SArtem Bityutskiy wb->last_active = jiffies; 96403ba3782SJens Axboe 965297252c8SArtem Bityutskiy set_current_state(TASK_INTERRUPTIBLE); 966b76b4014SJ. Bruce Fields if (!list_empty(&bdi->work_list) || kthread_should_stop()) { 967297252c8SArtem Bityutskiy __set_current_state(TASK_RUNNING); 968297252c8SArtem Bityutskiy continue; 96903ba3782SJens Axboe } 97003ba3782SJens Axboe 971253c34e9SArtem Bityutskiy if (wb_has_dirty_io(wb) && dirty_writeback_interval) 972fff5b85aSArtem Bityutskiy schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10)); 973253c34e9SArtem Bityutskiy else { 974253c34e9SArtem Bityutskiy /* 975253c34e9SArtem Bityutskiy * We have nothing to do, so can go sleep without any 976253c34e9SArtem Bityutskiy * timeout and save power. When a work is queued or 977253c34e9SArtem Bityutskiy * something is made dirty - we will be woken up. 978253c34e9SArtem Bityutskiy */ 97969b62d01SJens Axboe schedule(); 980f9eadbbdSJens Axboe } 98103ba3782SJens Axboe } 98203ba3782SJens Axboe 983fff5b85aSArtem Bityutskiy /* Flush any work that raced with us exiting */ 98408243900SChristoph Hellwig if (!list_empty(&bdi->work_list)) 98508243900SChristoph Hellwig wb_do_writeback(wb, 1); 986455b2864SDave Chinner 987455b2864SDave Chinner trace_writeback_thread_stop(bdi); 98803ba3782SJens Axboe return 0; 98903ba3782SJens Axboe } 99003ba3782SJens Axboe 99108243900SChristoph Hellwig 99203ba3782SJens Axboe /* 99303ba3782SJens Axboe * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back 99403ba3782SJens Axboe * the whole world. 99503ba3782SJens Axboe */ 9960e175a18SCurt Wohlgemuth void wakeup_flusher_threads(long nr_pages, enum wb_reason reason) 99703ba3782SJens Axboe { 998b8c2f347SChristoph Hellwig struct backing_dev_info *bdi; 999b8c2f347SChristoph Hellwig 100083ba7b07SChristoph Hellwig if (!nr_pages) { 100183ba7b07SChristoph Hellwig nr_pages = global_page_state(NR_FILE_DIRTY) + 100203ba3782SJens Axboe global_page_state(NR_UNSTABLE_NFS); 1003b8c2f347SChristoph Hellwig } 1004b8c2f347SChristoph Hellwig 1005b8c2f347SChristoph Hellwig rcu_read_lock(); 1006b8c2f347SChristoph Hellwig list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { 1007b8c2f347SChristoph Hellwig if (!bdi_has_dirty_io(bdi)) 1008b8c2f347SChristoph Hellwig continue; 10090e175a18SCurt Wohlgemuth __bdi_start_writeback(bdi, nr_pages, false, reason); 1010b8c2f347SChristoph Hellwig } 1011b8c2f347SChristoph Hellwig rcu_read_unlock(); 101203ba3782SJens Axboe } 101303ba3782SJens Axboe 101403ba3782SJens Axboe static noinline void block_dump___mark_inode_dirty(struct inode *inode) 101503ba3782SJens Axboe { 101603ba3782SJens Axboe if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { 101703ba3782SJens Axboe struct dentry *dentry; 101803ba3782SJens Axboe const char *name = "?"; 101903ba3782SJens Axboe 102003ba3782SJens Axboe dentry = d_find_alias(inode); 102103ba3782SJens Axboe if (dentry) { 102203ba3782SJens Axboe spin_lock(&dentry->d_lock); 102303ba3782SJens Axboe name = (const char *) dentry->d_name.name; 102403ba3782SJens Axboe } 102503ba3782SJens Axboe printk(KERN_DEBUG 102603ba3782SJens Axboe "%s(%d): dirtied inode %lu (%s) on %s\n", 102703ba3782SJens Axboe current->comm, task_pid_nr(current), inode->i_ino, 102803ba3782SJens Axboe name, inode->i_sb->s_id); 102903ba3782SJens Axboe if (dentry) { 103003ba3782SJens Axboe spin_unlock(&dentry->d_lock); 103103ba3782SJens Axboe dput(dentry); 103203ba3782SJens Axboe } 103303ba3782SJens Axboe } 103403ba3782SJens Axboe } 103503ba3782SJens Axboe 103603ba3782SJens Axboe /** 103703ba3782SJens Axboe * __mark_inode_dirty - internal function 103803ba3782SJens Axboe * @inode: inode to mark 103903ba3782SJens Axboe * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) 104003ba3782SJens Axboe * Mark an inode as dirty. Callers should use mark_inode_dirty or 104103ba3782SJens Axboe * mark_inode_dirty_sync. 104203ba3782SJens Axboe * 104303ba3782SJens Axboe * Put the inode on the super block's dirty list. 104403ba3782SJens Axboe * 104503ba3782SJens Axboe * CAREFUL! We mark it dirty unconditionally, but move it onto the 104603ba3782SJens Axboe * dirty list only if it is hashed or if it refers to a blockdev. 104703ba3782SJens Axboe * If it was not hashed, it will never be added to the dirty list 104803ba3782SJens Axboe * even if it is later hashed, as it will have been marked dirty already. 104903ba3782SJens Axboe * 105003ba3782SJens Axboe * In short, make sure you hash any inodes _before_ you start marking 105103ba3782SJens Axboe * them dirty. 105203ba3782SJens Axboe * 105303ba3782SJens Axboe * Note that for blockdevs, inode->dirtied_when represents the dirtying time of 105403ba3782SJens Axboe * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of 105503ba3782SJens Axboe * the kernel-internal blockdev inode represents the dirtying time of the 105603ba3782SJens Axboe * blockdev's pages. This is why for I_DIRTY_PAGES we always use 105703ba3782SJens Axboe * page->mapping->host, so the page-dirtying time is recorded in the internal 105803ba3782SJens Axboe * blockdev inode. 105903ba3782SJens Axboe */ 106003ba3782SJens Axboe void __mark_inode_dirty(struct inode *inode, int flags) 106103ba3782SJens Axboe { 106203ba3782SJens Axboe struct super_block *sb = inode->i_sb; 1063253c34e9SArtem Bityutskiy struct backing_dev_info *bdi = NULL; 106403ba3782SJens Axboe 106503ba3782SJens Axboe /* 106603ba3782SJens Axboe * Don't do this for I_DIRTY_PAGES - that doesn't actually 106703ba3782SJens Axboe * dirty the inode itself 106803ba3782SJens Axboe */ 106903ba3782SJens Axboe if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { 107003ba3782SJens Axboe if (sb->s_op->dirty_inode) 1071aa385729SChristoph Hellwig sb->s_op->dirty_inode(inode, flags); 107203ba3782SJens Axboe } 107303ba3782SJens Axboe 107403ba3782SJens Axboe /* 107503ba3782SJens Axboe * make sure that changes are seen by all cpus before we test i_state 107603ba3782SJens Axboe * -- mikulas 107703ba3782SJens Axboe */ 107803ba3782SJens Axboe smp_mb(); 107903ba3782SJens Axboe 108003ba3782SJens Axboe /* avoid the locking if we can */ 108103ba3782SJens Axboe if ((inode->i_state & flags) == flags) 108203ba3782SJens Axboe return; 108303ba3782SJens Axboe 108403ba3782SJens Axboe if (unlikely(block_dump)) 108503ba3782SJens Axboe block_dump___mark_inode_dirty(inode); 108603ba3782SJens Axboe 1087250df6edSDave Chinner spin_lock(&inode->i_lock); 108803ba3782SJens Axboe if ((inode->i_state & flags) != flags) { 108903ba3782SJens Axboe const int was_dirty = inode->i_state & I_DIRTY; 109003ba3782SJens Axboe 109103ba3782SJens Axboe inode->i_state |= flags; 109203ba3782SJens Axboe 109303ba3782SJens Axboe /* 109403ba3782SJens Axboe * If the inode is being synced, just update its dirty state. 109503ba3782SJens Axboe * The unlocker will place the inode on the appropriate 109603ba3782SJens Axboe * superblock list, based upon its state. 109703ba3782SJens Axboe */ 109803ba3782SJens Axboe if (inode->i_state & I_SYNC) 1099250df6edSDave Chinner goto out_unlock_inode; 110003ba3782SJens Axboe 110103ba3782SJens Axboe /* 110203ba3782SJens Axboe * Only add valid (hashed) inodes to the superblock's 110303ba3782SJens Axboe * dirty list. Add blockdev inodes as well. 110403ba3782SJens Axboe */ 110503ba3782SJens Axboe if (!S_ISBLK(inode->i_mode)) { 11061d3382cbSAl Viro if (inode_unhashed(inode)) 1107250df6edSDave Chinner goto out_unlock_inode; 110803ba3782SJens Axboe } 1109a4ffdde6SAl Viro if (inode->i_state & I_FREEING) 1110250df6edSDave Chinner goto out_unlock_inode; 111103ba3782SJens Axboe 111203ba3782SJens Axboe /* 111303ba3782SJens Axboe * If the inode was already on b_dirty/b_io/b_more_io, don't 111403ba3782SJens Axboe * reposition it (that would break b_dirty time-ordering). 111503ba3782SJens Axboe */ 111603ba3782SJens Axboe if (!was_dirty) { 1117a66979abSDave Chinner bool wakeup_bdi = false; 1118253c34e9SArtem Bityutskiy bdi = inode_to_bdi(inode); 1119500b067cSJens Axboe 1120253c34e9SArtem Bityutskiy if (bdi_cap_writeback_dirty(bdi)) { 1121253c34e9SArtem Bityutskiy WARN(!test_bit(BDI_registered, &bdi->state), 1122253c34e9SArtem Bityutskiy "bdi-%s not registered\n", bdi->name); 1123253c34e9SArtem Bityutskiy 1124253c34e9SArtem Bityutskiy /* 1125253c34e9SArtem Bityutskiy * If this is the first dirty inode for this 1126253c34e9SArtem Bityutskiy * bdi, we have to wake-up the corresponding 1127253c34e9SArtem Bityutskiy * bdi thread to make sure background 1128253c34e9SArtem Bityutskiy * write-back happens later. 1129253c34e9SArtem Bityutskiy */ 1130253c34e9SArtem Bityutskiy if (!wb_has_dirty_io(&bdi->wb)) 1131253c34e9SArtem Bityutskiy wakeup_bdi = true; 1132500b067cSJens Axboe } 113303ba3782SJens Axboe 1134a66979abSDave Chinner spin_unlock(&inode->i_lock); 1135f758eeabSChristoph Hellwig spin_lock(&bdi->wb.list_lock); 113603ba3782SJens Axboe inode->dirtied_when = jiffies; 11377ccf19a8SNick Piggin list_move(&inode->i_wb_list, &bdi->wb.b_dirty); 1138f758eeabSChristoph Hellwig spin_unlock(&bdi->wb.list_lock); 1139253c34e9SArtem Bityutskiy 1140253c34e9SArtem Bityutskiy if (wakeup_bdi) 11416467716aSArtem Bityutskiy bdi_wakeup_thread_delayed(bdi); 1142a66979abSDave Chinner return; 1143a66979abSDave Chinner } 1144a66979abSDave Chinner } 1145a66979abSDave Chinner out_unlock_inode: 1146a66979abSDave Chinner spin_unlock(&inode->i_lock); 1147a66979abSDave Chinner 114803ba3782SJens Axboe } 114903ba3782SJens Axboe EXPORT_SYMBOL(__mark_inode_dirty); 115003ba3782SJens Axboe 115166f3b8e2SJens Axboe /* 115266f3b8e2SJens Axboe * Write out a superblock's list of dirty inodes. A wait will be performed 115366f3b8e2SJens Axboe * upon no inodes, all inodes or the final one, depending upon sync_mode. 115466f3b8e2SJens Axboe * 115566f3b8e2SJens Axboe * If older_than_this is non-NULL, then only write out inodes which 115666f3b8e2SJens Axboe * had their first dirtying at a time earlier than *older_than_this. 115766f3b8e2SJens Axboe * 115866f3b8e2SJens Axboe * If `bdi' is non-zero then we're being asked to writeback a specific queue. 115966f3b8e2SJens Axboe * This function assumes that the blockdev superblock's inodes are backed by 116066f3b8e2SJens Axboe * a variety of queues, so all inodes are searched. For other superblocks, 116166f3b8e2SJens Axboe * assume that all inodes are backed by the same queue. 116266f3b8e2SJens Axboe * 116366f3b8e2SJens Axboe * The inodes to be written are parked on bdi->b_io. They are moved back onto 116466f3b8e2SJens Axboe * bdi->b_dirty as they are selected for writing. This way, none can be missed 116566f3b8e2SJens Axboe * on the writer throttling path, and we get decent balancing between many 116666f3b8e2SJens Axboe * throttled threads: we don't want them all piling up on inode_sync_wait. 116766f3b8e2SJens Axboe */ 1168b6e51316SJens Axboe static void wait_sb_inodes(struct super_block *sb) 116966f3b8e2SJens Axboe { 117038f21977SNick Piggin struct inode *inode, *old_inode = NULL; 117138f21977SNick Piggin 117203ba3782SJens Axboe /* 117303ba3782SJens Axboe * We need to be protected against the filesystem going from 117403ba3782SJens Axboe * r/o to r/w or vice versa. 117503ba3782SJens Axboe */ 1176b6e51316SJens Axboe WARN_ON(!rwsem_is_locked(&sb->s_umount)); 117703ba3782SJens Axboe 117855fa6091SDave Chinner spin_lock(&inode_sb_list_lock); 117966f3b8e2SJens Axboe 118038f21977SNick Piggin /* 118138f21977SNick Piggin * Data integrity sync. Must wait for all pages under writeback, 118238f21977SNick Piggin * because there may have been pages dirtied before our sync 118338f21977SNick Piggin * call, but which had writeout started before we write it out. 118438f21977SNick Piggin * In which case, the inode may not be on the dirty list, but 118538f21977SNick Piggin * we still have to wait for that writeout. 118638f21977SNick Piggin */ 1187b6e51316SJens Axboe list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { 1188250df6edSDave Chinner struct address_space *mapping = inode->i_mapping; 118938f21977SNick Piggin 1190250df6edSDave Chinner spin_lock(&inode->i_lock); 1191250df6edSDave Chinner if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) || 1192250df6edSDave Chinner (mapping->nrpages == 0)) { 1193250df6edSDave Chinner spin_unlock(&inode->i_lock); 119438f21977SNick Piggin continue; 1195250df6edSDave Chinner } 119638f21977SNick Piggin __iget(inode); 1197250df6edSDave Chinner spin_unlock(&inode->i_lock); 119855fa6091SDave Chinner spin_unlock(&inode_sb_list_lock); 119955fa6091SDave Chinner 120038f21977SNick Piggin /* 120155fa6091SDave Chinner * We hold a reference to 'inode' so it couldn't have been 120255fa6091SDave Chinner * removed from s_inodes list while we dropped the 120355fa6091SDave Chinner * inode_sb_list_lock. We cannot iput the inode now as we can 120455fa6091SDave Chinner * be holding the last reference and we cannot iput it under 120555fa6091SDave Chinner * inode_sb_list_lock. So we keep the reference and iput it 120655fa6091SDave Chinner * later. 120738f21977SNick Piggin */ 120838f21977SNick Piggin iput(old_inode); 120938f21977SNick Piggin old_inode = inode; 121038f21977SNick Piggin 121138f21977SNick Piggin filemap_fdatawait(mapping); 121238f21977SNick Piggin 121338f21977SNick Piggin cond_resched(); 121438f21977SNick Piggin 121555fa6091SDave Chinner spin_lock(&inode_sb_list_lock); 121638f21977SNick Piggin } 121755fa6091SDave Chinner spin_unlock(&inode_sb_list_lock); 121838f21977SNick Piggin iput(old_inode); 121966f3b8e2SJens Axboe } 12201da177e4SLinus Torvalds 1221d8a8559cSJens Axboe /** 12223259f8beSChris Mason * writeback_inodes_sb_nr - writeback dirty inodes from given super_block 1223d8a8559cSJens Axboe * @sb: the superblock 12243259f8beSChris Mason * @nr: the number of pages to write 1225786228abSMarcos Paulo de Souza * @reason: reason why some writeback work initiated 12261da177e4SLinus Torvalds * 1227d8a8559cSJens Axboe * Start writeback on some inodes on this super_block. No guarantees are made 1228d8a8559cSJens Axboe * on how many (if any) will be written, and this function does not wait 12293259f8beSChris Mason * for IO completion of submitted IO. 12301da177e4SLinus Torvalds */ 12310e175a18SCurt Wohlgemuth void writeback_inodes_sb_nr(struct super_block *sb, 12320e175a18SCurt Wohlgemuth unsigned long nr, 12330e175a18SCurt Wohlgemuth enum wb_reason reason) 12341da177e4SLinus Torvalds { 123583ba7b07SChristoph Hellwig DECLARE_COMPLETION_ONSTACK(done); 123683ba7b07SChristoph Hellwig struct wb_writeback_work work = { 12373c4d7165SChristoph Hellwig .sb = sb, 12383c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_NONE, 12396e6938b6SWu Fengguang .tagged_writepages = 1, 124083ba7b07SChristoph Hellwig .done = &done, 12413259f8beSChris Mason .nr_pages = nr, 12420e175a18SCurt Wohlgemuth .reason = reason, 12433c4d7165SChristoph Hellwig }; 12440e3c9a22SJens Axboe 1245cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 124683ba7b07SChristoph Hellwig bdi_queue_work(sb->s_bdi, &work); 124783ba7b07SChristoph Hellwig wait_for_completion(&done); 12481da177e4SLinus Torvalds } 12493259f8beSChris Mason EXPORT_SYMBOL(writeback_inodes_sb_nr); 12503259f8beSChris Mason 12513259f8beSChris Mason /** 12523259f8beSChris Mason * writeback_inodes_sb - writeback dirty inodes from given super_block 12533259f8beSChris Mason * @sb: the superblock 1254786228abSMarcos Paulo de Souza * @reason: reason why some writeback work was initiated 12553259f8beSChris Mason * 12563259f8beSChris Mason * Start writeback on some inodes on this super_block. No guarantees are made 12573259f8beSChris Mason * on how many (if any) will be written, and this function does not wait 12583259f8beSChris Mason * for IO completion of submitted IO. 12593259f8beSChris Mason */ 12600e175a18SCurt Wohlgemuth void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) 12613259f8beSChris Mason { 12620e175a18SCurt Wohlgemuth return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); 12633259f8beSChris Mason } 1264d8a8559cSJens Axboe EXPORT_SYMBOL(writeback_inodes_sb); 1265d8a8559cSJens Axboe 1266d8a8559cSJens Axboe /** 126717bd55d0SEric Sandeen * writeback_inodes_sb_if_idle - start writeback if none underway 126817bd55d0SEric Sandeen * @sb: the superblock 1269786228abSMarcos Paulo de Souza * @reason: reason why some writeback work was initiated 127017bd55d0SEric Sandeen * 127117bd55d0SEric Sandeen * Invoke writeback_inodes_sb if no writeback is currently underway. 127217bd55d0SEric Sandeen * Returns 1 if writeback was started, 0 if not. 127317bd55d0SEric Sandeen */ 12740e175a18SCurt Wohlgemuth int writeback_inodes_sb_if_idle(struct super_block *sb, enum wb_reason reason) 127517bd55d0SEric Sandeen { 127617bd55d0SEric Sandeen if (!writeback_in_progress(sb->s_bdi)) { 1277cf37e972SChristoph Hellwig down_read(&sb->s_umount); 12780e175a18SCurt Wohlgemuth writeback_inodes_sb(sb, reason); 1279cf37e972SChristoph Hellwig up_read(&sb->s_umount); 128017bd55d0SEric Sandeen return 1; 128117bd55d0SEric Sandeen } else 128217bd55d0SEric Sandeen return 0; 128317bd55d0SEric Sandeen } 128417bd55d0SEric Sandeen EXPORT_SYMBOL(writeback_inodes_sb_if_idle); 128517bd55d0SEric Sandeen 128617bd55d0SEric Sandeen /** 12873259f8beSChris Mason * writeback_inodes_sb_if_idle - start writeback if none underway 12883259f8beSChris Mason * @sb: the superblock 12893259f8beSChris Mason * @nr: the number of pages to write 1290786228abSMarcos Paulo de Souza * @reason: reason why some writeback work was initiated 12913259f8beSChris Mason * 12923259f8beSChris Mason * Invoke writeback_inodes_sb if no writeback is currently underway. 12933259f8beSChris Mason * Returns 1 if writeback was started, 0 if not. 12943259f8beSChris Mason */ 12953259f8beSChris Mason int writeback_inodes_sb_nr_if_idle(struct super_block *sb, 12960e175a18SCurt Wohlgemuth unsigned long nr, 12970e175a18SCurt Wohlgemuth enum wb_reason reason) 12983259f8beSChris Mason { 12993259f8beSChris Mason if (!writeback_in_progress(sb->s_bdi)) { 13003259f8beSChris Mason down_read(&sb->s_umount); 13010e175a18SCurt Wohlgemuth writeback_inodes_sb_nr(sb, nr, reason); 13023259f8beSChris Mason up_read(&sb->s_umount); 13033259f8beSChris Mason return 1; 13043259f8beSChris Mason } else 13053259f8beSChris Mason return 0; 13063259f8beSChris Mason } 13073259f8beSChris Mason EXPORT_SYMBOL(writeback_inodes_sb_nr_if_idle); 13083259f8beSChris Mason 13093259f8beSChris Mason /** 1310d8a8559cSJens Axboe * sync_inodes_sb - sync sb inode pages 1311d8a8559cSJens Axboe * @sb: the superblock 1312d8a8559cSJens Axboe * 1313d8a8559cSJens Axboe * This function writes and waits on any dirty inode belonging to this 1314cb9ef8d5SStefan Hajnoczi * super_block. 1315d8a8559cSJens Axboe */ 1316b6e51316SJens Axboe void sync_inodes_sb(struct super_block *sb) 1317d8a8559cSJens Axboe { 131883ba7b07SChristoph Hellwig DECLARE_COMPLETION_ONSTACK(done); 131983ba7b07SChristoph Hellwig struct wb_writeback_work work = { 13203c4d7165SChristoph Hellwig .sb = sb, 13213c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_ALL, 13223c4d7165SChristoph Hellwig .nr_pages = LONG_MAX, 13233c4d7165SChristoph Hellwig .range_cyclic = 0, 132483ba7b07SChristoph Hellwig .done = &done, 13250e175a18SCurt Wohlgemuth .reason = WB_REASON_SYNC, 13263c4d7165SChristoph Hellwig }; 13273c4d7165SChristoph Hellwig 1328cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 1329cf37e972SChristoph Hellwig 133083ba7b07SChristoph Hellwig bdi_queue_work(sb->s_bdi, &work); 133183ba7b07SChristoph Hellwig wait_for_completion(&done); 133283ba7b07SChristoph Hellwig 1333b6e51316SJens Axboe wait_sb_inodes(sb); 1334d8a8559cSJens Axboe } 1335d8a8559cSJens Axboe EXPORT_SYMBOL(sync_inodes_sb); 13361da177e4SLinus Torvalds 13371da177e4SLinus Torvalds /** 13381da177e4SLinus Torvalds * write_inode_now - write an inode to disk 13391da177e4SLinus Torvalds * @inode: inode to write to disk 13401da177e4SLinus Torvalds * @sync: whether the write should be synchronous or not 13411da177e4SLinus Torvalds * 13427f04c26dSAndrea Arcangeli * This function commits an inode to disk immediately if it is dirty. This is 13437f04c26dSAndrea Arcangeli * primarily needed by knfsd. 13447f04c26dSAndrea Arcangeli * 13457f04c26dSAndrea Arcangeli * The caller must either have a ref on the inode or must have set I_WILL_FREE. 13461da177e4SLinus Torvalds */ 13471da177e4SLinus Torvalds int write_inode_now(struct inode *inode, int sync) 13481da177e4SLinus Torvalds { 1349f758eeabSChristoph Hellwig struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; 13501da177e4SLinus Torvalds int ret; 13511da177e4SLinus Torvalds struct writeback_control wbc = { 13521da177e4SLinus Torvalds .nr_to_write = LONG_MAX, 135318914b18SMike Galbraith .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, 1354111ebb6eSOGAWA Hirofumi .range_start = 0, 1355111ebb6eSOGAWA Hirofumi .range_end = LLONG_MAX, 13561da177e4SLinus Torvalds }; 13571da177e4SLinus Torvalds 13581da177e4SLinus Torvalds if (!mapping_cap_writeback_dirty(inode->i_mapping)) 135949364ce2SAndrew Morton wbc.nr_to_write = 0; 13601da177e4SLinus Torvalds 13611da177e4SLinus Torvalds might_sleep(); 1362f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 13630f1b1fd8SDave Chinner spin_lock(&inode->i_lock); 1364f758eeabSChristoph Hellwig ret = writeback_single_inode(inode, wb, &wbc); 13650f1b1fd8SDave Chinner spin_unlock(&inode->i_lock); 1366f758eeabSChristoph Hellwig spin_unlock(&wb->list_lock); 13671da177e4SLinus Torvalds if (sync) 13681c0eeaf5SJoern Engel inode_sync_wait(inode); 13691da177e4SLinus Torvalds return ret; 13701da177e4SLinus Torvalds } 13711da177e4SLinus Torvalds EXPORT_SYMBOL(write_inode_now); 13721da177e4SLinus Torvalds 13731da177e4SLinus Torvalds /** 13741da177e4SLinus Torvalds * sync_inode - write an inode and its pages to disk. 13751da177e4SLinus Torvalds * @inode: the inode to sync 13761da177e4SLinus Torvalds * @wbc: controls the writeback mode 13771da177e4SLinus Torvalds * 13781da177e4SLinus Torvalds * sync_inode() will write an inode and its pages to disk. It will also 13791da177e4SLinus Torvalds * correctly update the inode on its superblock's dirty inode lists and will 13801da177e4SLinus Torvalds * update inode->i_state. 13811da177e4SLinus Torvalds * 13821da177e4SLinus Torvalds * The caller must have a ref on the inode. 13831da177e4SLinus Torvalds */ 13841da177e4SLinus Torvalds int sync_inode(struct inode *inode, struct writeback_control *wbc) 13851da177e4SLinus Torvalds { 1386f758eeabSChristoph Hellwig struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; 13871da177e4SLinus Torvalds int ret; 13881da177e4SLinus Torvalds 1389f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 13900f1b1fd8SDave Chinner spin_lock(&inode->i_lock); 1391f758eeabSChristoph Hellwig ret = writeback_single_inode(inode, wb, wbc); 13920f1b1fd8SDave Chinner spin_unlock(&inode->i_lock); 1393f758eeabSChristoph Hellwig spin_unlock(&wb->list_lock); 13941da177e4SLinus Torvalds return ret; 13951da177e4SLinus Torvalds } 13961da177e4SLinus Torvalds EXPORT_SYMBOL(sync_inode); 1397c3765016SChristoph Hellwig 1398c3765016SChristoph Hellwig /** 1399c691b9d9SAndrew Morton * sync_inode_metadata - write an inode to disk 1400c3765016SChristoph Hellwig * @inode: the inode to sync 1401c3765016SChristoph Hellwig * @wait: wait for I/O to complete. 1402c3765016SChristoph Hellwig * 1403c691b9d9SAndrew Morton * Write an inode to disk and adjust its dirty state after completion. 1404c3765016SChristoph Hellwig * 1405c3765016SChristoph Hellwig * Note: only writes the actual inode, no associated data or other metadata. 1406c3765016SChristoph Hellwig */ 1407c3765016SChristoph Hellwig int sync_inode_metadata(struct inode *inode, int wait) 1408c3765016SChristoph Hellwig { 1409c3765016SChristoph Hellwig struct writeback_control wbc = { 1410c3765016SChristoph Hellwig .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE, 1411c3765016SChristoph Hellwig .nr_to_write = 0, /* metadata-only */ 1412c3765016SChristoph Hellwig }; 1413c3765016SChristoph Hellwig 1414c3765016SChristoph Hellwig return sync_inode(inode, &wbc); 1415c3765016SChristoph Hellwig } 1416c3765016SChristoph Hellwig EXPORT_SYMBOL(sync_inode_metadata); 1417