11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * fs/fs-writeback.c 31da177e4SLinus Torvalds * 41da177e4SLinus Torvalds * Copyright (C) 2002, Linus Torvalds. 51da177e4SLinus Torvalds * 61da177e4SLinus Torvalds * Contains all the functions related to writing back and waiting 71da177e4SLinus Torvalds * upon dirty inodes against superblocks, and writing back dirty 81da177e4SLinus Torvalds * pages against inodes. ie: data writeback. Writeout of the 91da177e4SLinus Torvalds * inode itself is not handled here. 101da177e4SLinus Torvalds * 11e1f8e874SFrancois Cami * 10Apr2002 Andrew Morton 121da177e4SLinus Torvalds * Split out of fs/inode.c 131da177e4SLinus Torvalds * Additions for address_space-based writeback 141da177e4SLinus Torvalds */ 151da177e4SLinus Torvalds 161da177e4SLinus Torvalds #include <linux/kernel.h> 17630d9c47SPaul Gortmaker #include <linux/export.h> 181da177e4SLinus Torvalds #include <linux/spinlock.h> 195a0e3ad6STejun Heo #include <linux/slab.h> 201da177e4SLinus Torvalds #include <linux/sched.h> 211da177e4SLinus Torvalds #include <linux/fs.h> 221da177e4SLinus Torvalds #include <linux/mm.h> 23bc31b86aSWu Fengguang #include <linux/pagemap.h> 2403ba3782SJens Axboe #include <linux/kthread.h> 251da177e4SLinus Torvalds #include <linux/writeback.h> 261da177e4SLinus Torvalds #include <linux/blkdev.h> 271da177e4SLinus Torvalds #include <linux/backing-dev.h> 28455b2864SDave Chinner #include <linux/tracepoint.h> 29719ea2fbSAl Viro #include <linux/device.h> 3007f3f05cSDavid Howells #include "internal.h" 311da177e4SLinus Torvalds 32d0bceac7SJens Axboe /* 33bc31b86aSWu Fengguang * 4MB minimal write chunk size 34bc31b86aSWu Fengguang */ 35bc31b86aSWu Fengguang #define MIN_WRITEBACK_PAGES (4096UL >> (PAGE_CACHE_SHIFT - 10)) 36bc31b86aSWu Fengguang 37bc31b86aSWu Fengguang /* 38c4a77a6cSJens Axboe * Passed into wb_writeback(), essentially a subset of writeback_control 39c4a77a6cSJens Axboe */ 4083ba7b07SChristoph Hellwig struct wb_writeback_work { 41c4a77a6cSJens Axboe long nr_pages; 42c4a77a6cSJens Axboe struct super_block *sb; 430dc83bd3SJan Kara 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; 497747bd4bSDave Chinner unsigned int for_sync:1; /* sync(2) WB_SYNC_ALL writeback */ 500e175a18SCurt Wohlgemuth enum wb_reason reason; /* why was writeback initiated? */ 51c4a77a6cSJens Axboe 528010c3b6SJens Axboe struct list_head list; /* pending work list */ 5383ba7b07SChristoph Hellwig struct completion *done; /* set if the caller waits */ 5403ba3782SJens Axboe }; 5503ba3782SJens Axboe 56f11b00f3SAdrian Bunk /** 57f11b00f3SAdrian Bunk * writeback_in_progress - determine whether there is writeback in progress 58f11b00f3SAdrian Bunk * @bdi: the device's backing_dev_info structure. 59f11b00f3SAdrian Bunk * 6003ba3782SJens Axboe * Determine whether there is writeback waiting to be handled against a 6103ba3782SJens Axboe * backing device. 62f11b00f3SAdrian Bunk */ 63f11b00f3SAdrian Bunk int writeback_in_progress(struct backing_dev_info *bdi) 64f11b00f3SAdrian Bunk { 6581d73a32SJan Kara return test_bit(BDI_writeback_running, &bdi->state); 66f11b00f3SAdrian Bunk } 6700d4e736STheodore Ts'o EXPORT_SYMBOL(writeback_in_progress); 68f11b00f3SAdrian Bunk 69de1414a6SChristoph Hellwig struct backing_dev_info *inode_to_bdi(struct inode *inode) 70692ebd17SJan Kara { 71b520252aSJens Axboe struct super_block *sb; 72b520252aSJens Axboe 73b520252aSJens Axboe if (!inode) 74b520252aSJens Axboe return &noop_backing_dev_info; 75b520252aSJens Axboe 76b520252aSJens Axboe sb = inode->i_sb; 77495a276eSChristoph Hellwig #ifdef CONFIG_BLOCK 78a8855990SJan Kara if (sb_is_blkdev_sb(sb)) 79495a276eSChristoph Hellwig return blk_get_backing_dev_info(I_BDEV(inode)); 80495a276eSChristoph Hellwig #endif 81692ebd17SJan Kara return sb->s_bdi; 82692ebd17SJan Kara } 83de1414a6SChristoph Hellwig EXPORT_SYMBOL_GPL(inode_to_bdi); 84692ebd17SJan Kara 857ccf19a8SNick Piggin static inline struct inode *wb_inode(struct list_head *head) 867ccf19a8SNick Piggin { 877ccf19a8SNick Piggin return list_entry(head, struct inode, i_wb_list); 887ccf19a8SNick Piggin } 897ccf19a8SNick Piggin 9015eb77a0SWu Fengguang /* 9115eb77a0SWu Fengguang * Include the creation of the trace points after defining the 9215eb77a0SWu Fengguang * wb_writeback_work structure and inline functions so that the definition 9315eb77a0SWu Fengguang * remains local to this file. 9415eb77a0SWu Fengguang */ 9515eb77a0SWu Fengguang #define CREATE_TRACE_POINTS 9615eb77a0SWu Fengguang #include <trace/events/writeback.h> 9715eb77a0SWu Fengguang 98774016b2SSteven Whitehouse EXPORT_TRACEPOINT_SYMBOL_GPL(wbc_writepage); 99774016b2SSteven Whitehouse 1005acda9d1SJan Kara static void bdi_wakeup_thread(struct backing_dev_info *bdi) 1015acda9d1SJan Kara { 1025acda9d1SJan Kara spin_lock_bh(&bdi->wb_lock); 1035acda9d1SJan Kara if (test_bit(BDI_registered, &bdi->state)) 1045acda9d1SJan Kara mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0); 1055acda9d1SJan Kara spin_unlock_bh(&bdi->wb_lock); 1065acda9d1SJan Kara } 1075acda9d1SJan Kara 1086585027aSJan Kara static void bdi_queue_work(struct backing_dev_info *bdi, 1096585027aSJan Kara struct wb_writeback_work *work) 1106585027aSJan Kara { 1116585027aSJan Kara trace_writeback_queue(bdi, work); 1126585027aSJan Kara 1136585027aSJan Kara spin_lock_bh(&bdi->wb_lock); 1145acda9d1SJan Kara if (!test_bit(BDI_registered, &bdi->state)) { 1155acda9d1SJan Kara if (work->done) 1165acda9d1SJan Kara complete(work->done); 1175acda9d1SJan Kara goto out_unlock; 1185acda9d1SJan Kara } 1196585027aSJan Kara list_add_tail(&work->list, &bdi->work_list); 120839a8e86STejun Heo mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0); 1215acda9d1SJan Kara out_unlock: 1225acda9d1SJan Kara spin_unlock_bh(&bdi->wb_lock); 12303ba3782SJens Axboe } 1241da177e4SLinus Torvalds 12583ba7b07SChristoph Hellwig static void 12683ba7b07SChristoph Hellwig __bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, 1270e175a18SCurt Wohlgemuth bool range_cyclic, enum wb_reason reason) 1281da177e4SLinus Torvalds { 12983ba7b07SChristoph Hellwig struct wb_writeback_work *work; 13003ba3782SJens Axboe 131bcddc3f0SJens Axboe /* 132bcddc3f0SJens Axboe * This is WB_SYNC_NONE writeback, so if allocation fails just 133bcddc3f0SJens Axboe * wakeup the thread for old dirty data writeback 134bcddc3f0SJens Axboe */ 13583ba7b07SChristoph Hellwig work = kzalloc(sizeof(*work), GFP_ATOMIC); 13683ba7b07SChristoph Hellwig if (!work) { 137455b2864SDave Chinner trace_writeback_nowork(bdi); 1385acda9d1SJan Kara bdi_wakeup_thread(bdi); 13983ba7b07SChristoph Hellwig return; 14083ba7b07SChristoph Hellwig } 14183ba7b07SChristoph Hellwig 14283ba7b07SChristoph Hellwig work->sync_mode = WB_SYNC_NONE; 14383ba7b07SChristoph Hellwig work->nr_pages = nr_pages; 14483ba7b07SChristoph Hellwig work->range_cyclic = range_cyclic; 1450e175a18SCurt Wohlgemuth work->reason = reason; 14683ba7b07SChristoph Hellwig 147f11fcae8SJens Axboe bdi_queue_work(bdi, work); 14803ba3782SJens Axboe } 149b6e51316SJens Axboe 150b6e51316SJens Axboe /** 151b6e51316SJens Axboe * bdi_start_writeback - start writeback 152b6e51316SJens Axboe * @bdi: the backing device to write from 153b6e51316SJens Axboe * @nr_pages: the number of pages to write 154786228abSMarcos Paulo de Souza * @reason: reason why some writeback work was initiated 155b6e51316SJens Axboe * 156b6e51316SJens Axboe * Description: 157b6e51316SJens Axboe * This does WB_SYNC_NONE opportunistic writeback. The IO is only 15825985edcSLucas De Marchi * started when this function returns, we make no guarantees on 1590e3c9a22SJens Axboe * completion. Caller need not hold sb s_umount semaphore. 160b6e51316SJens Axboe * 161b6e51316SJens Axboe */ 1620e175a18SCurt Wohlgemuth void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, 1630e175a18SCurt Wohlgemuth enum wb_reason reason) 164b6e51316SJens Axboe { 1650e175a18SCurt Wohlgemuth __bdi_start_writeback(bdi, nr_pages, true, reason); 166d3ddec76SWu Fengguang } 167d3ddec76SWu Fengguang 168c5444198SChristoph Hellwig /** 169c5444198SChristoph Hellwig * bdi_start_background_writeback - start background writeback 170c5444198SChristoph Hellwig * @bdi: the backing device to write from 171c5444198SChristoph Hellwig * 172c5444198SChristoph Hellwig * Description: 1736585027aSJan Kara * This makes sure WB_SYNC_NONE background writeback happens. When 1746585027aSJan Kara * this function returns, it is only guaranteed that for given BDI 1756585027aSJan Kara * some IO is happening if we are over background dirty threshold. 1766585027aSJan Kara * Caller need not hold sb s_umount semaphore. 177c5444198SChristoph Hellwig */ 178c5444198SChristoph Hellwig void bdi_start_background_writeback(struct backing_dev_info *bdi) 179c5444198SChristoph Hellwig { 1806585027aSJan Kara /* 1816585027aSJan Kara * We just wake up the flusher thread. It will perform background 1826585027aSJan Kara * writeback as soon as there is no other work to do. 1836585027aSJan Kara */ 18471927e84SWu Fengguang trace_writeback_wake_background(bdi); 1855acda9d1SJan Kara bdi_wakeup_thread(bdi); 1861da177e4SLinus Torvalds } 1871da177e4SLinus Torvalds 1881da177e4SLinus Torvalds /* 189a66979abSDave Chinner * Remove the inode from the writeback list it is on. 190a66979abSDave Chinner */ 191a66979abSDave Chinner void inode_wb_list_del(struct inode *inode) 192a66979abSDave Chinner { 193f758eeabSChristoph Hellwig struct backing_dev_info *bdi = inode_to_bdi(inode); 194a66979abSDave Chinner 195f758eeabSChristoph Hellwig spin_lock(&bdi->wb.list_lock); 196f758eeabSChristoph Hellwig list_del_init(&inode->i_wb_list); 197f758eeabSChristoph Hellwig spin_unlock(&bdi->wb.list_lock); 198f758eeabSChristoph Hellwig } 199a66979abSDave Chinner 200a66979abSDave Chinner /* 2016610a0bcSAndrew Morton * Redirty an inode: set its when-it-was dirtied timestamp and move it to the 2026610a0bcSAndrew Morton * furthest end of its superblock's dirty-inode list. 2036610a0bcSAndrew Morton * 2046610a0bcSAndrew Morton * Before stamping the inode's ->dirtied_when, we check to see whether it is 20566f3b8e2SJens Axboe * already the most-recently-dirtied inode on the b_dirty list. If that is 2066610a0bcSAndrew Morton * the case then the inode must have been redirtied while it was being written 2076610a0bcSAndrew Morton * out and we don't reset its dirtied_when. 2086610a0bcSAndrew Morton */ 209f758eeabSChristoph Hellwig static void redirty_tail(struct inode *inode, struct bdi_writeback *wb) 2106610a0bcSAndrew Morton { 211f758eeabSChristoph Hellwig assert_spin_locked(&wb->list_lock); 21203ba3782SJens Axboe if (!list_empty(&wb->b_dirty)) { 21366f3b8e2SJens Axboe struct inode *tail; 2146610a0bcSAndrew Morton 2157ccf19a8SNick Piggin tail = wb_inode(wb->b_dirty.next); 21666f3b8e2SJens Axboe if (time_before(inode->dirtied_when, tail->dirtied_when)) 2176610a0bcSAndrew Morton inode->dirtied_when = jiffies; 2186610a0bcSAndrew Morton } 2197ccf19a8SNick Piggin list_move(&inode->i_wb_list, &wb->b_dirty); 2206610a0bcSAndrew Morton } 2216610a0bcSAndrew Morton 2226610a0bcSAndrew Morton /* 22366f3b8e2SJens Axboe * requeue inode for re-scanning after bdi->b_io list is exhausted. 224c986d1e2SAndrew Morton */ 225f758eeabSChristoph Hellwig static void requeue_io(struct inode *inode, struct bdi_writeback *wb) 226c986d1e2SAndrew Morton { 227f758eeabSChristoph Hellwig assert_spin_locked(&wb->list_lock); 2287ccf19a8SNick Piggin list_move(&inode->i_wb_list, &wb->b_more_io); 229c986d1e2SAndrew Morton } 230c986d1e2SAndrew Morton 2311c0eeaf5SJoern Engel static void inode_sync_complete(struct inode *inode) 2321c0eeaf5SJoern Engel { 233365b94aeSJan Kara inode->i_state &= ~I_SYNC; 2344eff96ddSJan Kara /* If inode is clean an unused, put it into LRU now... */ 2354eff96ddSJan Kara inode_add_lru(inode); 236365b94aeSJan Kara /* Waiters must see I_SYNC cleared before being woken up */ 2371c0eeaf5SJoern Engel smp_mb(); 2381c0eeaf5SJoern Engel wake_up_bit(&inode->i_state, __I_SYNC); 2391c0eeaf5SJoern Engel } 2401c0eeaf5SJoern Engel 241d2caa3c5SJeff Layton static bool inode_dirtied_after(struct inode *inode, unsigned long t) 242d2caa3c5SJeff Layton { 243d2caa3c5SJeff Layton bool ret = time_after(inode->dirtied_when, t); 244d2caa3c5SJeff Layton #ifndef CONFIG_64BIT 245d2caa3c5SJeff Layton /* 246d2caa3c5SJeff Layton * For inodes being constantly redirtied, dirtied_when can get stuck. 247d2caa3c5SJeff Layton * It _appears_ to be in the future, but is actually in distant past. 248d2caa3c5SJeff Layton * This test is necessary to prevent such wrapped-around relative times 2495b0830cbSJens Axboe * from permanently stopping the whole bdi writeback. 250d2caa3c5SJeff Layton */ 251d2caa3c5SJeff Layton ret = ret && time_before_eq(inode->dirtied_when, jiffies); 252d2caa3c5SJeff Layton #endif 253d2caa3c5SJeff Layton return ret; 254d2caa3c5SJeff Layton } 255d2caa3c5SJeff Layton 2560ae45f63STheodore Ts'o #define EXPIRE_DIRTY_ATIME 0x0001 2570ae45f63STheodore Ts'o 258c986d1e2SAndrew Morton /* 2590e2f2b23SWang Sheng-Hui * Move expired (dirtied before work->older_than_this) dirty inodes from 260697e6fedSJan Kara * @delaying_queue to @dispatch_queue. 2612c136579SFengguang Wu */ 262e84d0a4fSWu Fengguang static int move_expired_inodes(struct list_head *delaying_queue, 2632c136579SFengguang Wu struct list_head *dispatch_queue, 2640ae45f63STheodore Ts'o int flags, 265ad4e38ddSCurt Wohlgemuth struct wb_writeback_work *work) 2662c136579SFengguang Wu { 2670ae45f63STheodore Ts'o unsigned long *older_than_this = NULL; 2680ae45f63STheodore Ts'o unsigned long expire_time; 2695c03449dSShaohua Li LIST_HEAD(tmp); 2705c03449dSShaohua Li struct list_head *pos, *node; 271cf137307SJens Axboe struct super_block *sb = NULL; 2725c03449dSShaohua Li struct inode *inode; 273cf137307SJens Axboe int do_sb_sort = 0; 274e84d0a4fSWu Fengguang int moved = 0; 2755c03449dSShaohua Li 2760ae45f63STheodore Ts'o if ((flags & EXPIRE_DIRTY_ATIME) == 0) 2770ae45f63STheodore Ts'o older_than_this = work->older_than_this; 2780ae45f63STheodore Ts'o else if ((work->reason == WB_REASON_SYNC) == 0) { 2790ae45f63STheodore Ts'o expire_time = jiffies - (HZ * 86400); 2800ae45f63STheodore Ts'o older_than_this = &expire_time; 2810ae45f63STheodore Ts'o } 2822c136579SFengguang Wu while (!list_empty(delaying_queue)) { 2837ccf19a8SNick Piggin inode = wb_inode(delaying_queue->prev); 2840ae45f63STheodore Ts'o if (older_than_this && 2850ae45f63STheodore Ts'o inode_dirtied_after(inode, *older_than_this)) 2862c136579SFengguang Wu break; 287a8855990SJan Kara list_move(&inode->i_wb_list, &tmp); 288a8855990SJan Kara moved++; 2890ae45f63STheodore Ts'o if (flags & EXPIRE_DIRTY_ATIME) 2900ae45f63STheodore Ts'o set_bit(__I_DIRTY_TIME_EXPIRED, &inode->i_state); 291a8855990SJan Kara if (sb_is_blkdev_sb(inode->i_sb)) 292a8855990SJan Kara continue; 293cf137307SJens Axboe if (sb && sb != inode->i_sb) 294cf137307SJens Axboe do_sb_sort = 1; 295cf137307SJens Axboe sb = inode->i_sb; 2965c03449dSShaohua Li } 2975c03449dSShaohua Li 298cf137307SJens Axboe /* just one sb in list, splice to dispatch_queue and we're done */ 299cf137307SJens Axboe if (!do_sb_sort) { 300cf137307SJens Axboe list_splice(&tmp, dispatch_queue); 301e84d0a4fSWu Fengguang goto out; 302cf137307SJens Axboe } 303cf137307SJens Axboe 3045c03449dSShaohua Li /* Move inodes from one superblock together */ 3055c03449dSShaohua Li while (!list_empty(&tmp)) { 3067ccf19a8SNick Piggin sb = wb_inode(tmp.prev)->i_sb; 3075c03449dSShaohua Li list_for_each_prev_safe(pos, node, &tmp) { 3087ccf19a8SNick Piggin inode = wb_inode(pos); 3095c03449dSShaohua Li if (inode->i_sb == sb) 3107ccf19a8SNick Piggin list_move(&inode->i_wb_list, dispatch_queue); 3112c136579SFengguang Wu } 3122c136579SFengguang Wu } 313e84d0a4fSWu Fengguang out: 314e84d0a4fSWu Fengguang return moved; 3155c03449dSShaohua Li } 3162c136579SFengguang Wu 3172c136579SFengguang Wu /* 3182c136579SFengguang Wu * Queue all expired dirty inodes for io, eldest first. 3194ea879b9SWu Fengguang * Before 3204ea879b9SWu Fengguang * newly dirtied b_dirty b_io b_more_io 3214ea879b9SWu Fengguang * =============> gf edc BA 3224ea879b9SWu Fengguang * After 3234ea879b9SWu Fengguang * newly dirtied b_dirty b_io b_more_io 3244ea879b9SWu Fengguang * =============> g fBAedc 3254ea879b9SWu Fengguang * | 3264ea879b9SWu Fengguang * +--> dequeue for IO 3272c136579SFengguang Wu */ 328ad4e38ddSCurt Wohlgemuth static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work) 3292c136579SFengguang Wu { 330e84d0a4fSWu Fengguang int moved; 3310ae45f63STheodore Ts'o 332f758eeabSChristoph Hellwig assert_spin_locked(&wb->list_lock); 3334ea879b9SWu Fengguang list_splice_init(&wb->b_more_io, &wb->b_io); 3340ae45f63STheodore Ts'o moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, 0, work); 3350ae45f63STheodore Ts'o moved += move_expired_inodes(&wb->b_dirty_time, &wb->b_io, 3360ae45f63STheodore Ts'o EXPIRE_DIRTY_ATIME, work); 337ad4e38ddSCurt Wohlgemuth trace_writeback_queue_io(wb, work, moved); 33866f3b8e2SJens Axboe } 33966f3b8e2SJens Axboe 340a9185b41SChristoph Hellwig static int write_inode(struct inode *inode, struct writeback_control *wbc) 34166f3b8e2SJens Axboe { 3429fb0a7daSTejun Heo int ret; 3439fb0a7daSTejun Heo 3449fb0a7daSTejun Heo if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) { 3459fb0a7daSTejun Heo trace_writeback_write_inode_start(inode, wbc); 3469fb0a7daSTejun Heo ret = inode->i_sb->s_op->write_inode(inode, wbc); 3479fb0a7daSTejun Heo trace_writeback_write_inode(inode, wbc); 3489fb0a7daSTejun Heo return ret; 3499fb0a7daSTejun Heo } 35003ba3782SJens Axboe return 0; 35166f3b8e2SJens Axboe } 35208d8e974SFengguang Wu 3532c136579SFengguang Wu /* 354169ebd90SJan Kara * Wait for writeback on an inode to complete. Called with i_lock held. 355169ebd90SJan Kara * Caller must make sure inode cannot go away when we drop i_lock. 35601c03194SChristoph Hellwig */ 357169ebd90SJan Kara static void __inode_wait_for_writeback(struct inode *inode) 358169ebd90SJan Kara __releases(inode->i_lock) 359169ebd90SJan Kara __acquires(inode->i_lock) 36001c03194SChristoph Hellwig { 36101c03194SChristoph Hellwig DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); 36201c03194SChristoph Hellwig wait_queue_head_t *wqh; 36301c03194SChristoph Hellwig 36401c03194SChristoph Hellwig wqh = bit_waitqueue(&inode->i_state, __I_SYNC); 36558a9d3d8SRichard Kennedy while (inode->i_state & I_SYNC) { 366250df6edSDave Chinner spin_unlock(&inode->i_lock); 36774316201SNeilBrown __wait_on_bit(wqh, &wq, bit_wait, 36874316201SNeilBrown TASK_UNINTERRUPTIBLE); 369250df6edSDave Chinner spin_lock(&inode->i_lock); 37058a9d3d8SRichard Kennedy } 37101c03194SChristoph Hellwig } 37201c03194SChristoph Hellwig 37301c03194SChristoph Hellwig /* 374169ebd90SJan Kara * Wait for writeback on an inode to complete. Caller must have inode pinned. 375169ebd90SJan Kara */ 376169ebd90SJan Kara void inode_wait_for_writeback(struct inode *inode) 377169ebd90SJan Kara { 378169ebd90SJan Kara spin_lock(&inode->i_lock); 379169ebd90SJan Kara __inode_wait_for_writeback(inode); 380169ebd90SJan Kara spin_unlock(&inode->i_lock); 381169ebd90SJan Kara } 382169ebd90SJan Kara 383169ebd90SJan Kara /* 384169ebd90SJan Kara * Sleep until I_SYNC is cleared. This function must be called with i_lock 385169ebd90SJan Kara * held and drops it. It is aimed for callers not holding any inode reference 386169ebd90SJan Kara * so once i_lock is dropped, inode can go away. 387169ebd90SJan Kara */ 388169ebd90SJan Kara static void inode_sleep_on_writeback(struct inode *inode) 389169ebd90SJan Kara __releases(inode->i_lock) 390169ebd90SJan Kara { 391169ebd90SJan Kara DEFINE_WAIT(wait); 392169ebd90SJan Kara wait_queue_head_t *wqh = bit_waitqueue(&inode->i_state, __I_SYNC); 393169ebd90SJan Kara int sleep; 394169ebd90SJan Kara 395169ebd90SJan Kara prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); 396169ebd90SJan Kara sleep = inode->i_state & I_SYNC; 397169ebd90SJan Kara spin_unlock(&inode->i_lock); 398169ebd90SJan Kara if (sleep) 399169ebd90SJan Kara schedule(); 400169ebd90SJan Kara finish_wait(wqh, &wait); 401169ebd90SJan Kara } 402169ebd90SJan Kara 403169ebd90SJan Kara /* 404ccb26b5aSJan Kara * Find proper writeback list for the inode depending on its current state and 405ccb26b5aSJan Kara * possibly also change of its state while we were doing writeback. Here we 406ccb26b5aSJan Kara * handle things such as livelock prevention or fairness of writeback among 407ccb26b5aSJan Kara * inodes. This function can be called only by flusher thread - noone else 408ccb26b5aSJan Kara * processes all inodes in writeback lists and requeueing inodes behind flusher 409ccb26b5aSJan Kara * thread's back can have unexpected consequences. 410ccb26b5aSJan Kara */ 411ccb26b5aSJan Kara static void requeue_inode(struct inode *inode, struct bdi_writeback *wb, 412ccb26b5aSJan Kara struct writeback_control *wbc) 413ccb26b5aSJan Kara { 414ccb26b5aSJan Kara if (inode->i_state & I_FREEING) 415ccb26b5aSJan Kara return; 416ccb26b5aSJan Kara 417ccb26b5aSJan Kara /* 418ccb26b5aSJan Kara * Sync livelock prevention. Each inode is tagged and synced in one 419ccb26b5aSJan Kara * shot. If still dirty, it will be redirty_tail()'ed below. Update 420ccb26b5aSJan Kara * the dirty time to prevent enqueue and sync it again. 421ccb26b5aSJan Kara */ 422ccb26b5aSJan Kara if ((inode->i_state & I_DIRTY) && 423ccb26b5aSJan Kara (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)) 424ccb26b5aSJan Kara inode->dirtied_when = jiffies; 425ccb26b5aSJan Kara 4264f8ad655SJan Kara if (wbc->pages_skipped) { 4274f8ad655SJan Kara /* 4284f8ad655SJan Kara * writeback is not making progress due to locked 4294f8ad655SJan Kara * buffers. Skip this inode for now. 4304f8ad655SJan Kara */ 4314f8ad655SJan Kara redirty_tail(inode, wb); 4324f8ad655SJan Kara return; 4334f8ad655SJan Kara } 4344f8ad655SJan Kara 435ccb26b5aSJan Kara if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) { 436ccb26b5aSJan Kara /* 437ccb26b5aSJan Kara * We didn't write back all the pages. nfs_writepages() 438ccb26b5aSJan Kara * sometimes bales out without doing anything. 439ccb26b5aSJan Kara */ 440ccb26b5aSJan Kara if (wbc->nr_to_write <= 0) { 441ccb26b5aSJan Kara /* Slice used up. Queue for next turn. */ 442ccb26b5aSJan Kara requeue_io(inode, wb); 443ccb26b5aSJan Kara } else { 444ccb26b5aSJan Kara /* 445ccb26b5aSJan Kara * Writeback blocked by something other than 446ccb26b5aSJan Kara * congestion. Delay the inode for some time to 447ccb26b5aSJan Kara * avoid spinning on the CPU (100% iowait) 448ccb26b5aSJan Kara * retrying writeback of the dirty page/inode 449ccb26b5aSJan Kara * that cannot be performed immediately. 450ccb26b5aSJan Kara */ 451ccb26b5aSJan Kara redirty_tail(inode, wb); 452ccb26b5aSJan Kara } 453ccb26b5aSJan Kara } else if (inode->i_state & I_DIRTY) { 454ccb26b5aSJan Kara /* 455ccb26b5aSJan Kara * Filesystems can dirty the inode during writeback operations, 456ccb26b5aSJan Kara * such as delayed allocation during submission or metadata 457ccb26b5aSJan Kara * updates after data IO completion. 458ccb26b5aSJan Kara */ 459ccb26b5aSJan Kara redirty_tail(inode, wb); 4600ae45f63STheodore Ts'o } else if (inode->i_state & I_DIRTY_TIME) { 4610ae45f63STheodore Ts'o list_move(&inode->i_wb_list, &wb->b_dirty_time); 462ccb26b5aSJan Kara } else { 463ccb26b5aSJan Kara /* The inode is clean. Remove from writeback lists. */ 464ccb26b5aSJan Kara list_del_init(&inode->i_wb_list); 465ccb26b5aSJan Kara } 466ccb26b5aSJan Kara } 467ccb26b5aSJan Kara 468ccb26b5aSJan Kara /* 4694f8ad655SJan Kara * Write out an inode and its dirty pages. Do not update the writeback list 4704f8ad655SJan Kara * linkage. That is left to the caller. The caller is also responsible for 4714f8ad655SJan Kara * setting I_SYNC flag and calling inode_sync_complete() to clear it. 4721da177e4SLinus Torvalds */ 4731da177e4SLinus Torvalds static int 474cd8ed2a4SYan Hong __writeback_single_inode(struct inode *inode, struct writeback_control *wbc) 4751da177e4SLinus Torvalds { 4761da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 477251d6a47SWu Fengguang long nr_to_write = wbc->nr_to_write; 47801c03194SChristoph Hellwig unsigned dirty; 4791da177e4SLinus Torvalds int ret; 4801da177e4SLinus Torvalds 4814f8ad655SJan Kara WARN_ON(!(inode->i_state & I_SYNC)); 4821da177e4SLinus Torvalds 4839fb0a7daSTejun Heo trace_writeback_single_inode_start(inode, wbc, nr_to_write); 4849fb0a7daSTejun Heo 4851da177e4SLinus Torvalds ret = do_writepages(mapping, wbc); 4861da177e4SLinus Torvalds 48726821ed4SChristoph Hellwig /* 48826821ed4SChristoph Hellwig * Make sure to wait on the data before writing out the metadata. 48926821ed4SChristoph Hellwig * This is important for filesystems that modify metadata on data 4907747bd4bSDave Chinner * I/O completion. We don't do it for sync(2) writeback because it has a 4917747bd4bSDave Chinner * separate, external IO completion path and ->sync_fs for guaranteeing 4927747bd4bSDave Chinner * inode metadata is written back correctly. 49326821ed4SChristoph Hellwig */ 4947747bd4bSDave Chinner if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) { 49526821ed4SChristoph Hellwig int err = filemap_fdatawait(mapping); 4961da177e4SLinus Torvalds if (ret == 0) 4971da177e4SLinus Torvalds ret = err; 4981da177e4SLinus Torvalds } 4991da177e4SLinus Torvalds 5005547e8aaSDmitry Monakhov /* 5015547e8aaSDmitry Monakhov * Some filesystems may redirty the inode during the writeback 5025547e8aaSDmitry Monakhov * due to delalloc, clear dirty metadata flags right before 5035547e8aaSDmitry Monakhov * write_inode() 5045547e8aaSDmitry Monakhov */ 505250df6edSDave Chinner spin_lock(&inode->i_lock); 5069c6ac78eSTejun Heo 5075547e8aaSDmitry Monakhov dirty = inode->i_state & I_DIRTY; 5080ae45f63STheodore Ts'o if (((dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) && 5090ae45f63STheodore Ts'o (inode->i_state & I_DIRTY_TIME)) || 5100ae45f63STheodore Ts'o (inode->i_state & I_DIRTY_TIME_EXPIRED)) { 5110ae45f63STheodore Ts'o dirty |= I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED; 5120ae45f63STheodore Ts'o trace_writeback_lazytime(inode); 5130ae45f63STheodore Ts'o } 5140ae45f63STheodore Ts'o inode->i_state &= ~dirty; 5159c6ac78eSTejun Heo 5169c6ac78eSTejun Heo /* 5179c6ac78eSTejun Heo * Paired with smp_mb() in __mark_inode_dirty(). This allows 5189c6ac78eSTejun Heo * __mark_inode_dirty() to test i_state without grabbing i_lock - 5199c6ac78eSTejun Heo * either they see the I_DIRTY bits cleared or we see the dirtied 5209c6ac78eSTejun Heo * inode. 5219c6ac78eSTejun Heo * 5229c6ac78eSTejun Heo * I_DIRTY_PAGES is always cleared together above even if @mapping 5239c6ac78eSTejun Heo * still has dirty pages. The flag is reinstated after smp_mb() if 5249c6ac78eSTejun Heo * necessary. This guarantees that either __mark_inode_dirty() 5259c6ac78eSTejun Heo * sees clear I_DIRTY_PAGES or we see PAGECACHE_TAG_DIRTY. 5269c6ac78eSTejun Heo */ 5279c6ac78eSTejun Heo smp_mb(); 5289c6ac78eSTejun Heo 5299c6ac78eSTejun Heo if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) 5309c6ac78eSTejun Heo inode->i_state |= I_DIRTY_PAGES; 5319c6ac78eSTejun Heo 532250df6edSDave Chinner spin_unlock(&inode->i_lock); 5339c6ac78eSTejun Heo 5340ae45f63STheodore Ts'o if (dirty & I_DIRTY_TIME) 5350ae45f63STheodore Ts'o mark_inode_dirty_sync(inode); 53626821ed4SChristoph Hellwig /* Don't write the inode if only I_DIRTY_PAGES was set */ 5370ae45f63STheodore Ts'o if (dirty & ~I_DIRTY_PAGES) { 538a9185b41SChristoph Hellwig int err = write_inode(inode, wbc); 5391da177e4SLinus Torvalds if (ret == 0) 5401da177e4SLinus Torvalds ret = err; 5411da177e4SLinus Torvalds } 5424f8ad655SJan Kara trace_writeback_single_inode(inode, wbc, nr_to_write); 5434f8ad655SJan Kara return ret; 5444f8ad655SJan Kara } 5454f8ad655SJan Kara 5464f8ad655SJan Kara /* 5474f8ad655SJan Kara * Write out an inode's dirty pages. Either the caller has an active reference 5484f8ad655SJan Kara * on the inode or the inode has I_WILL_FREE set. 5494f8ad655SJan Kara * 5504f8ad655SJan Kara * This function is designed to be called for writing back one inode which 5514f8ad655SJan Kara * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode() 5524f8ad655SJan Kara * and does more profound writeback list handling in writeback_sb_inodes(). 5534f8ad655SJan Kara */ 5544f8ad655SJan Kara static int 5554f8ad655SJan Kara writeback_single_inode(struct inode *inode, struct bdi_writeback *wb, 5564f8ad655SJan Kara struct writeback_control *wbc) 5574f8ad655SJan Kara { 5584f8ad655SJan Kara int ret = 0; 5594f8ad655SJan Kara 5604f8ad655SJan Kara spin_lock(&inode->i_lock); 5614f8ad655SJan Kara if (!atomic_read(&inode->i_count)) 5624f8ad655SJan Kara WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); 5634f8ad655SJan Kara else 5644f8ad655SJan Kara WARN_ON(inode->i_state & I_WILL_FREE); 5654f8ad655SJan Kara 5664f8ad655SJan Kara if (inode->i_state & I_SYNC) { 5674f8ad655SJan Kara if (wbc->sync_mode != WB_SYNC_ALL) 5684f8ad655SJan Kara goto out; 5694f8ad655SJan Kara /* 570169ebd90SJan Kara * It's a data-integrity sync. We must wait. Since callers hold 571169ebd90SJan Kara * inode reference or inode has I_WILL_FREE set, it cannot go 572169ebd90SJan Kara * away under us. 5734f8ad655SJan Kara */ 574169ebd90SJan Kara __inode_wait_for_writeback(inode); 5754f8ad655SJan Kara } 5764f8ad655SJan Kara WARN_ON(inode->i_state & I_SYNC); 5774f8ad655SJan Kara /* 578f9b0e058SJan Kara * Skip inode if it is clean and we have no outstanding writeback in 579f9b0e058SJan Kara * WB_SYNC_ALL mode. We don't want to mess with writeback lists in this 580f9b0e058SJan Kara * function since flusher thread may be doing for example sync in 581f9b0e058SJan Kara * parallel and if we move the inode, it could get skipped. So here we 582f9b0e058SJan Kara * make sure inode is on some writeback list and leave it there unless 583f9b0e058SJan Kara * we have completely cleaned the inode. 5844f8ad655SJan Kara */ 5850ae45f63STheodore Ts'o if (!(inode->i_state & I_DIRTY_ALL) && 586f9b0e058SJan Kara (wbc->sync_mode != WB_SYNC_ALL || 587f9b0e058SJan Kara !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK))) 5884f8ad655SJan Kara goto out; 5894f8ad655SJan Kara inode->i_state |= I_SYNC; 5904f8ad655SJan Kara spin_unlock(&inode->i_lock); 5914f8ad655SJan Kara 592cd8ed2a4SYan Hong ret = __writeback_single_inode(inode, wbc); 5931da177e4SLinus Torvalds 594f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 595250df6edSDave Chinner spin_lock(&inode->i_lock); 5964f8ad655SJan Kara /* 5974f8ad655SJan Kara * If inode is clean, remove it from writeback lists. Otherwise don't 5984f8ad655SJan Kara * touch it. See comment above for explanation. 5994f8ad655SJan Kara */ 6000ae45f63STheodore Ts'o if (!(inode->i_state & I_DIRTY_ALL)) 6014f8ad655SJan Kara list_del_init(&inode->i_wb_list); 6024f8ad655SJan Kara spin_unlock(&wb->list_lock); 6031c0eeaf5SJoern Engel inode_sync_complete(inode); 6044f8ad655SJan Kara out: 6054f8ad655SJan Kara spin_unlock(&inode->i_lock); 6061da177e4SLinus Torvalds return ret; 6071da177e4SLinus Torvalds } 6081da177e4SLinus Torvalds 6091a12d8bdSWu Fengguang static long writeback_chunk_size(struct backing_dev_info *bdi, 6101a12d8bdSWu Fengguang struct wb_writeback_work *work) 611d46db3d5SWu Fengguang { 612d46db3d5SWu Fengguang long pages; 613d46db3d5SWu Fengguang 614d46db3d5SWu Fengguang /* 615d46db3d5SWu Fengguang * WB_SYNC_ALL mode does livelock avoidance by syncing dirty 616d46db3d5SWu Fengguang * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX 617d46db3d5SWu Fengguang * here avoids calling into writeback_inodes_wb() more than once. 618d46db3d5SWu Fengguang * 619d46db3d5SWu Fengguang * The intended call sequence for WB_SYNC_ALL writeback is: 620d46db3d5SWu Fengguang * 621d46db3d5SWu Fengguang * wb_writeback() 622d46db3d5SWu Fengguang * writeback_sb_inodes() <== called only once 623d46db3d5SWu Fengguang * write_cache_pages() <== called once for each inode 624d46db3d5SWu Fengguang * (quickly) tag currently dirty pages 625d46db3d5SWu Fengguang * (maybe slowly) sync all tagged pages 626d46db3d5SWu Fengguang */ 627d46db3d5SWu Fengguang if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages) 628d46db3d5SWu Fengguang pages = LONG_MAX; 6291a12d8bdSWu Fengguang else { 6301a12d8bdSWu Fengguang pages = min(bdi->avg_write_bandwidth / 2, 6311a12d8bdSWu Fengguang global_dirty_limit / DIRTY_SCOPE); 6321a12d8bdSWu Fengguang pages = min(pages, work->nr_pages); 6331a12d8bdSWu Fengguang pages = round_down(pages + MIN_WRITEBACK_PAGES, 6341a12d8bdSWu Fengguang MIN_WRITEBACK_PAGES); 6351a12d8bdSWu Fengguang } 636d46db3d5SWu Fengguang 637d46db3d5SWu Fengguang return pages; 638d46db3d5SWu Fengguang } 639d46db3d5SWu Fengguang 64003ba3782SJens Axboe /* 641f11c9c5cSEdward Shishkin * Write a portion of b_io inodes which belong to @sb. 642edadfb10SChristoph Hellwig * 643d46db3d5SWu Fengguang * Return the number of pages and/or inodes written. 644f11c9c5cSEdward Shishkin */ 645d46db3d5SWu Fengguang static long writeback_sb_inodes(struct super_block *sb, 646d46db3d5SWu Fengguang struct bdi_writeback *wb, 647d46db3d5SWu Fengguang struct wb_writeback_work *work) 64803ba3782SJens Axboe { 649d46db3d5SWu Fengguang struct writeback_control wbc = { 650d46db3d5SWu Fengguang .sync_mode = work->sync_mode, 651d46db3d5SWu Fengguang .tagged_writepages = work->tagged_writepages, 652d46db3d5SWu Fengguang .for_kupdate = work->for_kupdate, 653d46db3d5SWu Fengguang .for_background = work->for_background, 6547747bd4bSDave Chinner .for_sync = work->for_sync, 655d46db3d5SWu Fengguang .range_cyclic = work->range_cyclic, 656d46db3d5SWu Fengguang .range_start = 0, 657d46db3d5SWu Fengguang .range_end = LLONG_MAX, 658d46db3d5SWu Fengguang }; 659d46db3d5SWu Fengguang unsigned long start_time = jiffies; 660d46db3d5SWu Fengguang long write_chunk; 661d46db3d5SWu Fengguang long wrote = 0; /* count both pages and inodes */ 662d46db3d5SWu Fengguang 66303ba3782SJens Axboe while (!list_empty(&wb->b_io)) { 6647ccf19a8SNick Piggin struct inode *inode = wb_inode(wb->b_io.prev); 665edadfb10SChristoph Hellwig 666edadfb10SChristoph Hellwig if (inode->i_sb != sb) { 667d46db3d5SWu Fengguang if (work->sb) { 668edadfb10SChristoph Hellwig /* 669edadfb10SChristoph Hellwig * We only want to write back data for this 670edadfb10SChristoph Hellwig * superblock, move all inodes not belonging 671edadfb10SChristoph Hellwig * to it back onto the dirty list. 672edadfb10SChristoph Hellwig */ 673f758eeabSChristoph Hellwig redirty_tail(inode, wb); 67466f3b8e2SJens Axboe continue; 67566f3b8e2SJens Axboe } 676edadfb10SChristoph Hellwig 677edadfb10SChristoph Hellwig /* 678edadfb10SChristoph Hellwig * The inode belongs to a different superblock. 679edadfb10SChristoph Hellwig * Bounce back to the caller to unpin this and 680edadfb10SChristoph Hellwig * pin the next superblock. 681edadfb10SChristoph Hellwig */ 682d46db3d5SWu Fengguang break; 683edadfb10SChristoph Hellwig } 684edadfb10SChristoph Hellwig 6859843b76aSChristoph Hellwig /* 686331cbdeeSWanpeng Li * Don't bother with new inodes or inodes being freed, first 687331cbdeeSWanpeng Li * kind does not need periodic writeout yet, and for the latter 6889843b76aSChristoph Hellwig * kind writeout is handled by the freer. 6899843b76aSChristoph Hellwig */ 690250df6edSDave Chinner spin_lock(&inode->i_lock); 6919843b76aSChristoph Hellwig if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { 692250df6edSDave Chinner spin_unlock(&inode->i_lock); 693fcc5c222SWu Fengguang redirty_tail(inode, wb); 6947ef0d737SNick Piggin continue; 6957ef0d737SNick Piggin } 696cc1676d9SJan Kara if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) { 697cc1676d9SJan Kara /* 698cc1676d9SJan Kara * If this inode is locked for writeback and we are not 699cc1676d9SJan Kara * doing writeback-for-data-integrity, move it to 700cc1676d9SJan Kara * b_more_io so that writeback can proceed with the 701cc1676d9SJan Kara * other inodes on s_io. 702cc1676d9SJan Kara * 703cc1676d9SJan Kara * We'll have another go at writing back this inode 704cc1676d9SJan Kara * when we completed a full scan of b_io. 705cc1676d9SJan Kara */ 706cc1676d9SJan Kara spin_unlock(&inode->i_lock); 707cc1676d9SJan Kara requeue_io(inode, wb); 708cc1676d9SJan Kara trace_writeback_sb_inodes_requeue(inode); 709cc1676d9SJan Kara continue; 710cc1676d9SJan Kara } 711f0d07b7fSJan Kara spin_unlock(&wb->list_lock); 712f0d07b7fSJan Kara 7134f8ad655SJan Kara /* 7144f8ad655SJan Kara * We already requeued the inode if it had I_SYNC set and we 7154f8ad655SJan Kara * are doing WB_SYNC_NONE writeback. So this catches only the 7164f8ad655SJan Kara * WB_SYNC_ALL case. 7174f8ad655SJan Kara */ 718169ebd90SJan Kara if (inode->i_state & I_SYNC) { 719169ebd90SJan Kara /* Wait for I_SYNC. This function drops i_lock... */ 720169ebd90SJan Kara inode_sleep_on_writeback(inode); 721169ebd90SJan Kara /* Inode may be gone, start again */ 722ead188f9SJan Kara spin_lock(&wb->list_lock); 723169ebd90SJan Kara continue; 724169ebd90SJan Kara } 7254f8ad655SJan Kara inode->i_state |= I_SYNC; 7264f8ad655SJan Kara spin_unlock(&inode->i_lock); 727169ebd90SJan Kara 7281a12d8bdSWu Fengguang write_chunk = writeback_chunk_size(wb->bdi, work); 729d46db3d5SWu Fengguang wbc.nr_to_write = write_chunk; 730d46db3d5SWu Fengguang wbc.pages_skipped = 0; 731250df6edSDave Chinner 732169ebd90SJan Kara /* 733169ebd90SJan Kara * We use I_SYNC to pin the inode in memory. While it is set 734169ebd90SJan Kara * evict_inode() will wait so the inode cannot be freed. 735169ebd90SJan Kara */ 736cd8ed2a4SYan Hong __writeback_single_inode(inode, &wbc); 737d46db3d5SWu Fengguang 738d46db3d5SWu Fengguang work->nr_pages -= write_chunk - wbc.nr_to_write; 739d46db3d5SWu Fengguang wrote += write_chunk - wbc.nr_to_write; 7404f8ad655SJan Kara spin_lock(&wb->list_lock); 7414f8ad655SJan Kara spin_lock(&inode->i_lock); 7420ae45f63STheodore Ts'o if (!(inode->i_state & I_DIRTY_ALL)) 743d46db3d5SWu Fengguang wrote++; 7444f8ad655SJan Kara requeue_inode(inode, wb, &wbc); 7454f8ad655SJan Kara inode_sync_complete(inode); 7460f1b1fd8SDave Chinner spin_unlock(&inode->i_lock); 747169ebd90SJan Kara cond_resched_lock(&wb->list_lock); 748d46db3d5SWu Fengguang /* 749d46db3d5SWu Fengguang * bail out to wb_writeback() often enough to check 750d46db3d5SWu Fengguang * background threshold and other termination conditions. 751d46db3d5SWu Fengguang */ 752d46db3d5SWu Fengguang if (wrote) { 753d46db3d5SWu Fengguang if (time_is_before_jiffies(start_time + HZ / 10UL)) 754d46db3d5SWu Fengguang break; 755d46db3d5SWu Fengguang if (work->nr_pages <= 0) 756d46db3d5SWu Fengguang break; 7571da177e4SLinus Torvalds } 7588bc3be27SFengguang Wu } 759d46db3d5SWu Fengguang return wrote; 760f11c9c5cSEdward Shishkin } 76138f21977SNick Piggin 762d46db3d5SWu Fengguang static long __writeback_inodes_wb(struct bdi_writeback *wb, 763d46db3d5SWu Fengguang struct wb_writeback_work *work) 764f11c9c5cSEdward Shishkin { 765d46db3d5SWu Fengguang unsigned long start_time = jiffies; 766d46db3d5SWu Fengguang long wrote = 0; 767f11c9c5cSEdward Shishkin 768f11c9c5cSEdward Shishkin while (!list_empty(&wb->b_io)) { 7697ccf19a8SNick Piggin struct inode *inode = wb_inode(wb->b_io.prev); 770f11c9c5cSEdward Shishkin struct super_block *sb = inode->i_sb; 771f11c9c5cSEdward Shishkin 772eb6ef3dfSKonstantin Khlebnikov if (!trylock_super(sb)) { 7730e995816SWu Fengguang /* 774eb6ef3dfSKonstantin Khlebnikov * trylock_super() may fail consistently due to 7750e995816SWu Fengguang * s_umount being grabbed by someone else. Don't use 7760e995816SWu Fengguang * requeue_io() to avoid busy retrying the inode/sb. 7770e995816SWu Fengguang */ 7780e995816SWu Fengguang redirty_tail(inode, wb); 779d19de7edSChristoph Hellwig continue; 780334132aeSChristoph Hellwig } 781d46db3d5SWu Fengguang wrote += writeback_sb_inodes(sb, wb, work); 782eb6ef3dfSKonstantin Khlebnikov up_read(&sb->s_umount); 783f11c9c5cSEdward Shishkin 784d46db3d5SWu Fengguang /* refer to the same tests at the end of writeback_sb_inodes */ 785d46db3d5SWu Fengguang if (wrote) { 786d46db3d5SWu Fengguang if (time_is_before_jiffies(start_time + HZ / 10UL)) 787d46db3d5SWu Fengguang break; 788d46db3d5SWu Fengguang if (work->nr_pages <= 0) 789f11c9c5cSEdward Shishkin break; 790f11c9c5cSEdward Shishkin } 791d46db3d5SWu Fengguang } 79266f3b8e2SJens Axboe /* Leave any unwritten inodes on b_io */ 793d46db3d5SWu Fengguang return wrote; 79466f3b8e2SJens Axboe } 79566f3b8e2SJens Axboe 7967d9f073bSWanpeng Li static long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages, 7970e175a18SCurt Wohlgemuth enum wb_reason reason) 798edadfb10SChristoph Hellwig { 799d46db3d5SWu Fengguang struct wb_writeback_work work = { 800d46db3d5SWu Fengguang .nr_pages = nr_pages, 801d46db3d5SWu Fengguang .sync_mode = WB_SYNC_NONE, 802d46db3d5SWu Fengguang .range_cyclic = 1, 8030e175a18SCurt Wohlgemuth .reason = reason, 804d46db3d5SWu Fengguang }; 805edadfb10SChristoph Hellwig 806f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 807424b351fSWu Fengguang if (list_empty(&wb->b_io)) 808ad4e38ddSCurt Wohlgemuth queue_io(wb, &work); 809d46db3d5SWu Fengguang __writeback_inodes_wb(wb, &work); 810f758eeabSChristoph Hellwig spin_unlock(&wb->list_lock); 811edadfb10SChristoph Hellwig 812d46db3d5SWu Fengguang return nr_pages - work.nr_pages; 81366f3b8e2SJens Axboe } 81466f3b8e2SJens Axboe 815b00949aaSWu Fengguang static bool over_bground_thresh(struct backing_dev_info *bdi) 81603ba3782SJens Axboe { 81703ba3782SJens Axboe unsigned long background_thresh, dirty_thresh; 81803ba3782SJens Axboe 81916c4042fSWu Fengguang global_dirty_limits(&background_thresh, &dirty_thresh); 82003ba3782SJens Axboe 821b00949aaSWu Fengguang if (global_page_state(NR_FILE_DIRTY) + 822b00949aaSWu Fengguang global_page_state(NR_UNSTABLE_NFS) > background_thresh) 823b00949aaSWu Fengguang return true; 824b00949aaSWu Fengguang 825b00949aaSWu Fengguang if (bdi_stat(bdi, BDI_RECLAIMABLE) > 826b00949aaSWu Fengguang bdi_dirty_limit(bdi, background_thresh)) 827b00949aaSWu Fengguang return true; 828b00949aaSWu Fengguang 829b00949aaSWu Fengguang return false; 83003ba3782SJens Axboe } 83103ba3782SJens Axboe 83203ba3782SJens Axboe /* 833e98be2d5SWu Fengguang * Called under wb->list_lock. If there are multiple wb per bdi, 834e98be2d5SWu Fengguang * only the flusher working on the first wb should do it. 835e98be2d5SWu Fengguang */ 836e98be2d5SWu Fengguang static void wb_update_bandwidth(struct bdi_writeback *wb, 837e98be2d5SWu Fengguang unsigned long start_time) 838e98be2d5SWu Fengguang { 839af6a3113SWu Fengguang __bdi_update_bandwidth(wb->bdi, 0, 0, 0, 0, 0, start_time); 840e98be2d5SWu Fengguang } 841e98be2d5SWu Fengguang 842e98be2d5SWu Fengguang /* 84303ba3782SJens Axboe * Explicit flushing or periodic writeback of "old" data. 84403ba3782SJens Axboe * 84503ba3782SJens Axboe * Define "old": the first time one of an inode's pages is dirtied, we mark the 84603ba3782SJens Axboe * dirtying-time in the inode's address_space. So this periodic writeback code 84703ba3782SJens Axboe * just walks the superblock inode list, writing back any inodes which are 84803ba3782SJens Axboe * older than a specific point in time. 84903ba3782SJens Axboe * 85003ba3782SJens Axboe * Try to run once per dirty_writeback_interval. But if a writeback event 85103ba3782SJens Axboe * takes longer than a dirty_writeback_interval interval, then leave a 85203ba3782SJens Axboe * one-second gap. 85303ba3782SJens Axboe * 85403ba3782SJens Axboe * older_than_this takes precedence over nr_to_write. So we'll only write back 85503ba3782SJens Axboe * all dirty pages if they are all attached to "old" mappings. 85603ba3782SJens Axboe */ 857c4a77a6cSJens Axboe static long wb_writeback(struct bdi_writeback *wb, 85883ba7b07SChristoph Hellwig struct wb_writeback_work *work) 85903ba3782SJens Axboe { 860e98be2d5SWu Fengguang unsigned long wb_start = jiffies; 861d46db3d5SWu Fengguang long nr_pages = work->nr_pages; 8620dc83bd3SJan Kara unsigned long oldest_jif; 863a5989bdcSJan Kara struct inode *inode; 864d46db3d5SWu Fengguang long progress; 86503ba3782SJens Axboe 8660dc83bd3SJan Kara oldest_jif = jiffies; 8670dc83bd3SJan Kara work->older_than_this = &oldest_jif; 86803ba3782SJens Axboe 869e8dfc305SWu Fengguang spin_lock(&wb->list_lock); 87003ba3782SJens Axboe for (;;) { 87103ba3782SJens Axboe /* 872d3ddec76SWu Fengguang * Stop writeback when nr_pages has been consumed 87303ba3782SJens Axboe */ 87483ba7b07SChristoph Hellwig if (work->nr_pages <= 0) 87503ba3782SJens Axboe break; 87603ba3782SJens Axboe 87703ba3782SJens Axboe /* 878aa373cf5SJan Kara * Background writeout and kupdate-style writeback may 879aa373cf5SJan Kara * run forever. Stop them if there is other work to do 880aa373cf5SJan Kara * so that e.g. sync can proceed. They'll be restarted 881aa373cf5SJan Kara * after the other works are all done. 882aa373cf5SJan Kara */ 883aa373cf5SJan Kara if ((work->for_background || work->for_kupdate) && 884aa373cf5SJan Kara !list_empty(&wb->bdi->work_list)) 885aa373cf5SJan Kara break; 886aa373cf5SJan Kara 887aa373cf5SJan Kara /* 888d3ddec76SWu Fengguang * For background writeout, stop when we are below the 889d3ddec76SWu Fengguang * background dirty threshold 89003ba3782SJens Axboe */ 891b00949aaSWu Fengguang if (work->for_background && !over_bground_thresh(wb->bdi)) 89203ba3782SJens Axboe break; 89303ba3782SJens Axboe 8941bc36b64SJan Kara /* 8951bc36b64SJan Kara * Kupdate and background works are special and we want to 8961bc36b64SJan Kara * include all inodes that need writing. Livelock avoidance is 8971bc36b64SJan Kara * handled by these works yielding to any other work so we are 8981bc36b64SJan Kara * safe. 8991bc36b64SJan Kara */ 900ba9aa839SWu Fengguang if (work->for_kupdate) { 9010dc83bd3SJan Kara oldest_jif = jiffies - 902ba9aa839SWu Fengguang msecs_to_jiffies(dirty_expire_interval * 10); 9031bc36b64SJan Kara } else if (work->for_background) 9040dc83bd3SJan Kara oldest_jif = jiffies; 905028c2dd1SDave Chinner 906d46db3d5SWu Fengguang trace_writeback_start(wb->bdi, work); 907e8dfc305SWu Fengguang if (list_empty(&wb->b_io)) 908ad4e38ddSCurt Wohlgemuth queue_io(wb, work); 90983ba7b07SChristoph Hellwig if (work->sb) 910d46db3d5SWu Fengguang progress = writeback_sb_inodes(work->sb, wb, work); 911edadfb10SChristoph Hellwig else 912d46db3d5SWu Fengguang progress = __writeback_inodes_wb(wb, work); 913d46db3d5SWu Fengguang trace_writeback_written(wb->bdi, work); 914028c2dd1SDave Chinner 915e98be2d5SWu Fengguang wb_update_bandwidth(wb, wb_start); 91603ba3782SJens Axboe 91703ba3782SJens Axboe /* 91871fd05a8SJens Axboe * Did we write something? Try for more 919e6fb6da2SWu Fengguang * 920e6fb6da2SWu Fengguang * Dirty inodes are moved to b_io for writeback in batches. 921e6fb6da2SWu Fengguang * The completion of the current batch does not necessarily 922e6fb6da2SWu Fengguang * mean the overall work is done. So we keep looping as long 923e6fb6da2SWu Fengguang * as made some progress on cleaning pages or inodes. 92471fd05a8SJens Axboe */ 925d46db3d5SWu Fengguang if (progress) 92603ba3782SJens Axboe continue; 927a5989bdcSJan Kara /* 928e6fb6da2SWu Fengguang * No more inodes for IO, bail 929a5989bdcSJan Kara */ 930b7a2441fSWu Fengguang if (list_empty(&wb->b_more_io)) 93103ba3782SJens Axboe break; 93203ba3782SJens Axboe /* 9338010c3b6SJens Axboe * Nothing written. Wait for some inode to 9348010c3b6SJens Axboe * become available for writeback. Otherwise 9358010c3b6SJens Axboe * we'll just busyloop. 93603ba3782SJens Axboe */ 93703ba3782SJens Axboe if (!list_empty(&wb->b_more_io)) { 938d46db3d5SWu Fengguang trace_writeback_wait(wb->bdi, work); 93903ba3782SJens Axboe inode = wb_inode(wb->b_more_io.prev); 940250df6edSDave Chinner spin_lock(&inode->i_lock); 941f0d07b7fSJan Kara spin_unlock(&wb->list_lock); 942169ebd90SJan Kara /* This function drops i_lock... */ 943169ebd90SJan Kara inode_sleep_on_writeback(inode); 944f0d07b7fSJan Kara spin_lock(&wb->list_lock); 94503ba3782SJens Axboe } 94603ba3782SJens Axboe } 947e8dfc305SWu Fengguang spin_unlock(&wb->list_lock); 94803ba3782SJens Axboe 949d46db3d5SWu Fengguang return nr_pages - work->nr_pages; 95003ba3782SJens Axboe } 95103ba3782SJens Axboe 95203ba3782SJens Axboe /* 95383ba7b07SChristoph Hellwig * Return the next wb_writeback_work struct that hasn't been processed yet. 95403ba3782SJens Axboe */ 95583ba7b07SChristoph Hellwig static struct wb_writeback_work * 95608852b6dSMinchan Kim get_next_work_item(struct backing_dev_info *bdi) 95703ba3782SJens Axboe { 95883ba7b07SChristoph Hellwig struct wb_writeback_work *work = NULL; 95903ba3782SJens Axboe 9606467716aSArtem Bityutskiy spin_lock_bh(&bdi->wb_lock); 96183ba7b07SChristoph Hellwig if (!list_empty(&bdi->work_list)) { 96283ba7b07SChristoph Hellwig work = list_entry(bdi->work_list.next, 96383ba7b07SChristoph Hellwig struct wb_writeback_work, list); 96483ba7b07SChristoph Hellwig list_del_init(&work->list); 96503ba3782SJens Axboe } 9666467716aSArtem Bityutskiy spin_unlock_bh(&bdi->wb_lock); 96783ba7b07SChristoph Hellwig return work; 96803ba3782SJens Axboe } 96903ba3782SJens Axboe 970cdf01dd5SLinus Torvalds /* 971cdf01dd5SLinus Torvalds * Add in the number of potentially dirty inodes, because each inode 972cdf01dd5SLinus Torvalds * write can dirty pagecache in the underlying blockdev. 973cdf01dd5SLinus Torvalds */ 974cdf01dd5SLinus Torvalds static unsigned long get_nr_dirty_pages(void) 975cdf01dd5SLinus Torvalds { 976cdf01dd5SLinus Torvalds return global_page_state(NR_FILE_DIRTY) + 977cdf01dd5SLinus Torvalds global_page_state(NR_UNSTABLE_NFS) + 978cdf01dd5SLinus Torvalds get_nr_dirty_inodes(); 979cdf01dd5SLinus Torvalds } 980cdf01dd5SLinus Torvalds 9816585027aSJan Kara static long wb_check_background_flush(struct bdi_writeback *wb) 9826585027aSJan Kara { 983b00949aaSWu Fengguang if (over_bground_thresh(wb->bdi)) { 9846585027aSJan Kara 9856585027aSJan Kara struct wb_writeback_work work = { 9866585027aSJan Kara .nr_pages = LONG_MAX, 9876585027aSJan Kara .sync_mode = WB_SYNC_NONE, 9886585027aSJan Kara .for_background = 1, 9896585027aSJan Kara .range_cyclic = 1, 9900e175a18SCurt Wohlgemuth .reason = WB_REASON_BACKGROUND, 9916585027aSJan Kara }; 9926585027aSJan Kara 9936585027aSJan Kara return wb_writeback(wb, &work); 9946585027aSJan Kara } 9956585027aSJan Kara 9966585027aSJan Kara return 0; 9976585027aSJan Kara } 9986585027aSJan Kara 99903ba3782SJens Axboe static long wb_check_old_data_flush(struct bdi_writeback *wb) 100003ba3782SJens Axboe { 100103ba3782SJens Axboe unsigned long expired; 100203ba3782SJens Axboe long nr_pages; 100303ba3782SJens Axboe 100469b62d01SJens Axboe /* 100569b62d01SJens Axboe * When set to zero, disable periodic writeback 100669b62d01SJens Axboe */ 100769b62d01SJens Axboe if (!dirty_writeback_interval) 100869b62d01SJens Axboe return 0; 100969b62d01SJens Axboe 101003ba3782SJens Axboe expired = wb->last_old_flush + 101103ba3782SJens Axboe msecs_to_jiffies(dirty_writeback_interval * 10); 101203ba3782SJens Axboe if (time_before(jiffies, expired)) 101303ba3782SJens Axboe return 0; 101403ba3782SJens Axboe 101503ba3782SJens Axboe wb->last_old_flush = jiffies; 1016cdf01dd5SLinus Torvalds nr_pages = get_nr_dirty_pages(); 101703ba3782SJens Axboe 1018c4a77a6cSJens Axboe if (nr_pages) { 101983ba7b07SChristoph Hellwig struct wb_writeback_work work = { 1020c4a77a6cSJens Axboe .nr_pages = nr_pages, 1021c4a77a6cSJens Axboe .sync_mode = WB_SYNC_NONE, 1022c4a77a6cSJens Axboe .for_kupdate = 1, 1023c4a77a6cSJens Axboe .range_cyclic = 1, 10240e175a18SCurt Wohlgemuth .reason = WB_REASON_PERIODIC, 1025c4a77a6cSJens Axboe }; 1026c4a77a6cSJens Axboe 102783ba7b07SChristoph Hellwig return wb_writeback(wb, &work); 1028c4a77a6cSJens Axboe } 102903ba3782SJens Axboe 103003ba3782SJens Axboe return 0; 103103ba3782SJens Axboe } 103203ba3782SJens Axboe 103303ba3782SJens Axboe /* 103403ba3782SJens Axboe * Retrieve work items and do the writeback they describe 103503ba3782SJens Axboe */ 103625d130baSWanpeng Li static long wb_do_writeback(struct bdi_writeback *wb) 103703ba3782SJens Axboe { 103803ba3782SJens Axboe struct backing_dev_info *bdi = wb->bdi; 103983ba7b07SChristoph Hellwig struct wb_writeback_work *work; 1040c4a77a6cSJens Axboe long wrote = 0; 104103ba3782SJens Axboe 104281d73a32SJan Kara set_bit(BDI_writeback_running, &wb->bdi->state); 104308852b6dSMinchan Kim while ((work = get_next_work_item(bdi)) != NULL) { 104483ba7b07SChristoph Hellwig 1045455b2864SDave Chinner trace_writeback_exec(bdi, work); 1046455b2864SDave Chinner 104783ba7b07SChristoph Hellwig wrote += wb_writeback(wb, work); 104803ba3782SJens Axboe 104903ba3782SJens Axboe /* 105083ba7b07SChristoph Hellwig * Notify the caller of completion if this is a synchronous 105183ba7b07SChristoph Hellwig * work item, otherwise just free it. 105203ba3782SJens Axboe */ 105383ba7b07SChristoph Hellwig if (work->done) 105483ba7b07SChristoph Hellwig complete(work->done); 105583ba7b07SChristoph Hellwig else 105683ba7b07SChristoph Hellwig kfree(work); 105703ba3782SJens Axboe } 105803ba3782SJens Axboe 105903ba3782SJens Axboe /* 106003ba3782SJens Axboe * Check for periodic writeback, kupdated() style 106103ba3782SJens Axboe */ 106203ba3782SJens Axboe wrote += wb_check_old_data_flush(wb); 10636585027aSJan Kara wrote += wb_check_background_flush(wb); 106481d73a32SJan Kara clear_bit(BDI_writeback_running, &wb->bdi->state); 106503ba3782SJens Axboe 106603ba3782SJens Axboe return wrote; 106703ba3782SJens Axboe } 106803ba3782SJens Axboe 106903ba3782SJens Axboe /* 107003ba3782SJens Axboe * Handle writeback of dirty data for the device backed by this bdi. Also 1071839a8e86STejun Heo * reschedules periodically and does kupdated style flushing. 107203ba3782SJens Axboe */ 1073839a8e86STejun Heo void bdi_writeback_workfn(struct work_struct *work) 107403ba3782SJens Axboe { 1075839a8e86STejun Heo struct bdi_writeback *wb = container_of(to_delayed_work(work), 1076839a8e86STejun Heo struct bdi_writeback, dwork); 107708243900SChristoph Hellwig struct backing_dev_info *bdi = wb->bdi; 107803ba3782SJens Axboe long pages_written; 107903ba3782SJens Axboe 1080ef3b1019STejun Heo set_worker_desc("flush-%s", dev_name(bdi->dev)); 1081766f9164SPeter Zijlstra current->flags |= PF_SWAPWRITE; 108203ba3782SJens Axboe 1083839a8e86STejun Heo if (likely(!current_is_workqueue_rescuer() || 10845acda9d1SJan Kara !test_bit(BDI_registered, &bdi->state))) { 108503ba3782SJens Axboe /* 1086839a8e86STejun Heo * The normal path. Keep writing back @bdi until its 1087839a8e86STejun Heo * work_list is empty. Note that this path is also taken 1088839a8e86STejun Heo * if @bdi is shutting down even when we're running off the 1089839a8e86STejun Heo * rescuer as work_list needs to be drained. 109003ba3782SJens Axboe */ 1091839a8e86STejun Heo do { 109225d130baSWanpeng Li pages_written = wb_do_writeback(wb); 1093455b2864SDave Chinner trace_writeback_pages_written(pages_written); 1094839a8e86STejun Heo } while (!list_empty(&bdi->work_list)); 1095839a8e86STejun Heo } else { 1096253c34e9SArtem Bityutskiy /* 1097839a8e86STejun Heo * bdi_wq can't get enough workers and we're running off 1098839a8e86STejun Heo * the emergency worker. Don't hog it. Hopefully, 1024 is 1099839a8e86STejun Heo * enough for efficient IO. 1100253c34e9SArtem Bityutskiy */ 1101839a8e86STejun Heo pages_written = writeback_inodes_wb(&bdi->wb, 1024, 1102839a8e86STejun Heo WB_REASON_FORKER_THREAD); 1103839a8e86STejun Heo trace_writeback_pages_written(pages_written); 110403ba3782SJens Axboe } 110503ba3782SJens Axboe 11066ca738d6SDerek Basehore if (!list_empty(&bdi->work_list)) 11076ca738d6SDerek Basehore mod_delayed_work(bdi_wq, &wb->dwork, 0); 11086ca738d6SDerek Basehore else if (wb_has_dirty_io(wb) && dirty_writeback_interval) 11096ca738d6SDerek Basehore bdi_wakeup_thread_delayed(bdi); 1110455b2864SDave Chinner 1111839a8e86STejun Heo current->flags &= ~PF_SWAPWRITE; 111203ba3782SJens Axboe } 111303ba3782SJens Axboe 111403ba3782SJens Axboe /* 111503ba3782SJens Axboe * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back 111603ba3782SJens Axboe * the whole world. 111703ba3782SJens Axboe */ 11180e175a18SCurt Wohlgemuth void wakeup_flusher_threads(long nr_pages, enum wb_reason reason) 111903ba3782SJens Axboe { 1120b8c2f347SChristoph Hellwig struct backing_dev_info *bdi; 1121b8c2f347SChristoph Hellwig 112247df3ddeSJan Kara if (!nr_pages) 112347df3ddeSJan Kara nr_pages = get_nr_dirty_pages(); 1124b8c2f347SChristoph Hellwig 1125b8c2f347SChristoph Hellwig rcu_read_lock(); 1126b8c2f347SChristoph Hellwig list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { 1127b8c2f347SChristoph Hellwig if (!bdi_has_dirty_io(bdi)) 1128b8c2f347SChristoph Hellwig continue; 11290e175a18SCurt Wohlgemuth __bdi_start_writeback(bdi, nr_pages, false, reason); 1130b8c2f347SChristoph Hellwig } 1131b8c2f347SChristoph Hellwig rcu_read_unlock(); 113203ba3782SJens Axboe } 113303ba3782SJens Axboe 113403ba3782SJens Axboe static noinline void block_dump___mark_inode_dirty(struct inode *inode) 113503ba3782SJens Axboe { 113603ba3782SJens Axboe if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { 113703ba3782SJens Axboe struct dentry *dentry; 113803ba3782SJens Axboe const char *name = "?"; 113903ba3782SJens Axboe 114003ba3782SJens Axboe dentry = d_find_alias(inode); 114103ba3782SJens Axboe if (dentry) { 114203ba3782SJens Axboe spin_lock(&dentry->d_lock); 114303ba3782SJens Axboe name = (const char *) dentry->d_name.name; 114403ba3782SJens Axboe } 114503ba3782SJens Axboe printk(KERN_DEBUG 114603ba3782SJens Axboe "%s(%d): dirtied inode %lu (%s) on %s\n", 114703ba3782SJens Axboe current->comm, task_pid_nr(current), inode->i_ino, 114803ba3782SJens Axboe name, inode->i_sb->s_id); 114903ba3782SJens Axboe if (dentry) { 115003ba3782SJens Axboe spin_unlock(&dentry->d_lock); 115103ba3782SJens Axboe dput(dentry); 115203ba3782SJens Axboe } 115303ba3782SJens Axboe } 115403ba3782SJens Axboe } 115503ba3782SJens Axboe 115603ba3782SJens Axboe /** 115703ba3782SJens Axboe * __mark_inode_dirty - internal function 115803ba3782SJens Axboe * @inode: inode to mark 115903ba3782SJens Axboe * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) 116003ba3782SJens Axboe * Mark an inode as dirty. Callers should use mark_inode_dirty or 116103ba3782SJens Axboe * mark_inode_dirty_sync. 116203ba3782SJens Axboe * 116303ba3782SJens Axboe * Put the inode on the super block's dirty list. 116403ba3782SJens Axboe * 116503ba3782SJens Axboe * CAREFUL! We mark it dirty unconditionally, but move it onto the 116603ba3782SJens Axboe * dirty list only if it is hashed or if it refers to a blockdev. 116703ba3782SJens Axboe * If it was not hashed, it will never be added to the dirty list 116803ba3782SJens Axboe * even if it is later hashed, as it will have been marked dirty already. 116903ba3782SJens Axboe * 117003ba3782SJens Axboe * In short, make sure you hash any inodes _before_ you start marking 117103ba3782SJens Axboe * them dirty. 117203ba3782SJens Axboe * 117303ba3782SJens Axboe * Note that for blockdevs, inode->dirtied_when represents the dirtying time of 117403ba3782SJens Axboe * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of 117503ba3782SJens Axboe * the kernel-internal blockdev inode represents the dirtying time of the 117603ba3782SJens Axboe * blockdev's pages. This is why for I_DIRTY_PAGES we always use 117703ba3782SJens Axboe * page->mapping->host, so the page-dirtying time is recorded in the internal 117803ba3782SJens Axboe * blockdev inode. 117903ba3782SJens Axboe */ 11800ae45f63STheodore Ts'o #define I_DIRTY_INODE (I_DIRTY_SYNC | I_DIRTY_DATASYNC) 118103ba3782SJens Axboe void __mark_inode_dirty(struct inode *inode, int flags) 118203ba3782SJens Axboe { 118303ba3782SJens Axboe struct super_block *sb = inode->i_sb; 1184253c34e9SArtem Bityutskiy struct backing_dev_info *bdi = NULL; 11850ae45f63STheodore Ts'o int dirtytime; 11860ae45f63STheodore Ts'o 11870ae45f63STheodore Ts'o trace_writeback_mark_inode_dirty(inode, flags); 118803ba3782SJens Axboe 118903ba3782SJens Axboe /* 119003ba3782SJens Axboe * Don't do this for I_DIRTY_PAGES - that doesn't actually 119103ba3782SJens Axboe * dirty the inode itself 119203ba3782SJens Axboe */ 11930ae45f63STheodore Ts'o if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC | I_DIRTY_TIME)) { 11949fb0a7daSTejun Heo trace_writeback_dirty_inode_start(inode, flags); 11959fb0a7daSTejun Heo 119603ba3782SJens Axboe if (sb->s_op->dirty_inode) 1197aa385729SChristoph Hellwig sb->s_op->dirty_inode(inode, flags); 11989fb0a7daSTejun Heo 11999fb0a7daSTejun Heo trace_writeback_dirty_inode(inode, flags); 120003ba3782SJens Axboe } 12010ae45f63STheodore Ts'o if (flags & I_DIRTY_INODE) 12020ae45f63STheodore Ts'o flags &= ~I_DIRTY_TIME; 12030ae45f63STheodore Ts'o dirtytime = flags & I_DIRTY_TIME; 120403ba3782SJens Axboe 120503ba3782SJens Axboe /* 12069c6ac78eSTejun Heo * Paired with smp_mb() in __writeback_single_inode() for the 12079c6ac78eSTejun Heo * following lockless i_state test. See there for details. 120803ba3782SJens Axboe */ 120903ba3782SJens Axboe smp_mb(); 121003ba3782SJens Axboe 12110ae45f63STheodore Ts'o if (((inode->i_state & flags) == flags) || 12120ae45f63STheodore Ts'o (dirtytime && (inode->i_state & I_DIRTY_INODE))) 121303ba3782SJens Axboe return; 121403ba3782SJens Axboe 121503ba3782SJens Axboe if (unlikely(block_dump)) 121603ba3782SJens Axboe block_dump___mark_inode_dirty(inode); 121703ba3782SJens Axboe 1218250df6edSDave Chinner spin_lock(&inode->i_lock); 12190ae45f63STheodore Ts'o if (dirtytime && (inode->i_state & I_DIRTY_INODE)) 12200ae45f63STheodore Ts'o goto out_unlock_inode; 122103ba3782SJens Axboe if ((inode->i_state & flags) != flags) { 122203ba3782SJens Axboe const int was_dirty = inode->i_state & I_DIRTY; 122303ba3782SJens Axboe 12240ae45f63STheodore Ts'o if (flags & I_DIRTY_INODE) 12250ae45f63STheodore Ts'o inode->i_state &= ~I_DIRTY_TIME; 122603ba3782SJens Axboe inode->i_state |= flags; 122703ba3782SJens Axboe 122803ba3782SJens Axboe /* 122903ba3782SJens Axboe * If the inode is being synced, just update its dirty state. 123003ba3782SJens Axboe * The unlocker will place the inode on the appropriate 123103ba3782SJens Axboe * superblock list, based upon its state. 123203ba3782SJens Axboe */ 123303ba3782SJens Axboe if (inode->i_state & I_SYNC) 1234250df6edSDave Chinner goto out_unlock_inode; 123503ba3782SJens Axboe 123603ba3782SJens Axboe /* 123703ba3782SJens Axboe * Only add valid (hashed) inodes to the superblock's 123803ba3782SJens Axboe * dirty list. Add blockdev inodes as well. 123903ba3782SJens Axboe */ 124003ba3782SJens Axboe if (!S_ISBLK(inode->i_mode)) { 12411d3382cbSAl Viro if (inode_unhashed(inode)) 1242250df6edSDave Chinner goto out_unlock_inode; 124303ba3782SJens Axboe } 1244a4ffdde6SAl Viro if (inode->i_state & I_FREEING) 1245250df6edSDave Chinner goto out_unlock_inode; 124603ba3782SJens Axboe 124703ba3782SJens Axboe /* 124803ba3782SJens Axboe * If the inode was already on b_dirty/b_io/b_more_io, don't 124903ba3782SJens Axboe * reposition it (that would break b_dirty time-ordering). 125003ba3782SJens Axboe */ 125103ba3782SJens Axboe if (!was_dirty) { 1252a66979abSDave Chinner bool wakeup_bdi = false; 1253253c34e9SArtem Bityutskiy bdi = inode_to_bdi(inode); 1254500b067cSJens Axboe 1255146d7009SJunxiao Bi spin_unlock(&inode->i_lock); 1256146d7009SJunxiao Bi spin_lock(&bdi->wb.list_lock); 1257253c34e9SArtem Bityutskiy if (bdi_cap_writeback_dirty(bdi)) { 1258253c34e9SArtem Bityutskiy WARN(!test_bit(BDI_registered, &bdi->state), 1259253c34e9SArtem Bityutskiy "bdi-%s not registered\n", bdi->name); 1260253c34e9SArtem Bityutskiy 1261253c34e9SArtem Bityutskiy /* 1262253c34e9SArtem Bityutskiy * If this is the first dirty inode for this 1263253c34e9SArtem Bityutskiy * bdi, we have to wake-up the corresponding 1264253c34e9SArtem Bityutskiy * bdi thread to make sure background 1265253c34e9SArtem Bityutskiy * write-back happens later. 1266253c34e9SArtem Bityutskiy */ 1267253c34e9SArtem Bityutskiy if (!wb_has_dirty_io(&bdi->wb)) 1268253c34e9SArtem Bityutskiy wakeup_bdi = true; 1269500b067cSJens Axboe } 127003ba3782SJens Axboe 127103ba3782SJens Axboe inode->dirtied_when = jiffies; 12720ae45f63STheodore Ts'o list_move(&inode->i_wb_list, dirtytime ? 12730ae45f63STheodore Ts'o &bdi->wb.b_dirty_time : &bdi->wb.b_dirty); 1274f758eeabSChristoph Hellwig spin_unlock(&bdi->wb.list_lock); 12750ae45f63STheodore Ts'o trace_writeback_dirty_inode_enqueue(inode); 1276253c34e9SArtem Bityutskiy 1277253c34e9SArtem Bityutskiy if (wakeup_bdi) 12786467716aSArtem Bityutskiy bdi_wakeup_thread_delayed(bdi); 1279a66979abSDave Chinner return; 1280a66979abSDave Chinner } 1281a66979abSDave Chinner } 1282a66979abSDave Chinner out_unlock_inode: 1283a66979abSDave Chinner spin_unlock(&inode->i_lock); 1284a66979abSDave Chinner 128503ba3782SJens Axboe } 128603ba3782SJens Axboe EXPORT_SYMBOL(__mark_inode_dirty); 128703ba3782SJens Axboe 1288b6e51316SJens Axboe static void wait_sb_inodes(struct super_block *sb) 128966f3b8e2SJens Axboe { 129038f21977SNick Piggin struct inode *inode, *old_inode = NULL; 129138f21977SNick Piggin 129203ba3782SJens Axboe /* 129303ba3782SJens Axboe * We need to be protected against the filesystem going from 129403ba3782SJens Axboe * r/o to r/w or vice versa. 129503ba3782SJens Axboe */ 1296b6e51316SJens Axboe WARN_ON(!rwsem_is_locked(&sb->s_umount)); 129703ba3782SJens Axboe 129855fa6091SDave Chinner spin_lock(&inode_sb_list_lock); 129966f3b8e2SJens Axboe 130038f21977SNick Piggin /* 130138f21977SNick Piggin * Data integrity sync. Must wait for all pages under writeback, 130238f21977SNick Piggin * because there may have been pages dirtied before our sync 130338f21977SNick Piggin * call, but which had writeout started before we write it out. 130438f21977SNick Piggin * In which case, the inode may not be on the dirty list, but 130538f21977SNick Piggin * we still have to wait for that writeout. 130638f21977SNick Piggin */ 1307b6e51316SJens Axboe list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { 1308250df6edSDave Chinner struct address_space *mapping = inode->i_mapping; 130938f21977SNick Piggin 1310250df6edSDave Chinner spin_lock(&inode->i_lock); 1311250df6edSDave Chinner if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) || 1312250df6edSDave Chinner (mapping->nrpages == 0)) { 1313250df6edSDave Chinner spin_unlock(&inode->i_lock); 131438f21977SNick Piggin continue; 1315250df6edSDave Chinner } 131638f21977SNick Piggin __iget(inode); 1317250df6edSDave Chinner spin_unlock(&inode->i_lock); 131855fa6091SDave Chinner spin_unlock(&inode_sb_list_lock); 131955fa6091SDave Chinner 132038f21977SNick Piggin /* 132155fa6091SDave Chinner * We hold a reference to 'inode' so it couldn't have been 132255fa6091SDave Chinner * removed from s_inodes list while we dropped the 132355fa6091SDave Chinner * inode_sb_list_lock. We cannot iput the inode now as we can 132455fa6091SDave Chinner * be holding the last reference and we cannot iput it under 132555fa6091SDave Chinner * inode_sb_list_lock. So we keep the reference and iput it 132655fa6091SDave Chinner * later. 132738f21977SNick Piggin */ 132838f21977SNick Piggin iput(old_inode); 132938f21977SNick Piggin old_inode = inode; 133038f21977SNick Piggin 133138f21977SNick Piggin filemap_fdatawait(mapping); 133238f21977SNick Piggin 133338f21977SNick Piggin cond_resched(); 133438f21977SNick Piggin 133555fa6091SDave Chinner spin_lock(&inode_sb_list_lock); 133638f21977SNick Piggin } 133755fa6091SDave Chinner spin_unlock(&inode_sb_list_lock); 133838f21977SNick Piggin iput(old_inode); 133966f3b8e2SJens Axboe } 13401da177e4SLinus Torvalds 1341d8a8559cSJens Axboe /** 13423259f8beSChris Mason * writeback_inodes_sb_nr - writeback dirty inodes from given super_block 1343d8a8559cSJens Axboe * @sb: the superblock 13443259f8beSChris Mason * @nr: the number of pages to write 1345786228abSMarcos Paulo de Souza * @reason: reason why some writeback work initiated 13461da177e4SLinus Torvalds * 1347d8a8559cSJens Axboe * Start writeback on some inodes on this super_block. No guarantees are made 1348d8a8559cSJens Axboe * on how many (if any) will be written, and this function does not wait 13493259f8beSChris Mason * for IO completion of submitted IO. 13501da177e4SLinus Torvalds */ 13510e175a18SCurt Wohlgemuth void writeback_inodes_sb_nr(struct super_block *sb, 13520e175a18SCurt Wohlgemuth unsigned long nr, 13530e175a18SCurt Wohlgemuth enum wb_reason reason) 13541da177e4SLinus Torvalds { 135583ba7b07SChristoph Hellwig DECLARE_COMPLETION_ONSTACK(done); 135683ba7b07SChristoph Hellwig struct wb_writeback_work work = { 13573c4d7165SChristoph Hellwig .sb = sb, 13583c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_NONE, 13596e6938b6SWu Fengguang .tagged_writepages = 1, 136083ba7b07SChristoph Hellwig .done = &done, 13613259f8beSChris Mason .nr_pages = nr, 13620e175a18SCurt Wohlgemuth .reason = reason, 13633c4d7165SChristoph Hellwig }; 13640e3c9a22SJens Axboe 13656eedc701SJan Kara if (sb->s_bdi == &noop_backing_dev_info) 13666eedc701SJan Kara return; 1367cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 136883ba7b07SChristoph Hellwig bdi_queue_work(sb->s_bdi, &work); 136983ba7b07SChristoph Hellwig wait_for_completion(&done); 13701da177e4SLinus Torvalds } 13713259f8beSChris Mason EXPORT_SYMBOL(writeback_inodes_sb_nr); 13723259f8beSChris Mason 13733259f8beSChris Mason /** 13743259f8beSChris Mason * writeback_inodes_sb - writeback dirty inodes from given super_block 13753259f8beSChris Mason * @sb: the superblock 1376786228abSMarcos Paulo de Souza * @reason: reason why some writeback work was initiated 13773259f8beSChris Mason * 13783259f8beSChris Mason * Start writeback on some inodes on this super_block. No guarantees are made 13793259f8beSChris Mason * on how many (if any) will be written, and this function does not wait 13803259f8beSChris Mason * for IO completion of submitted IO. 13813259f8beSChris Mason */ 13820e175a18SCurt Wohlgemuth void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) 13833259f8beSChris Mason { 13840e175a18SCurt Wohlgemuth return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); 13853259f8beSChris Mason } 1386d8a8559cSJens Axboe EXPORT_SYMBOL(writeback_inodes_sb); 1387d8a8559cSJens Axboe 1388d8a8559cSJens Axboe /** 138910ee27a0SMiao Xie * try_to_writeback_inodes_sb_nr - try to start writeback if none underway 13903259f8beSChris Mason * @sb: the superblock 13913259f8beSChris Mason * @nr: the number of pages to write 139210ee27a0SMiao Xie * @reason: the reason of writeback 13933259f8beSChris Mason * 139410ee27a0SMiao Xie * Invoke writeback_inodes_sb_nr if no writeback is currently underway. 13953259f8beSChris Mason * Returns 1 if writeback was started, 0 if not. 13963259f8beSChris Mason */ 139710ee27a0SMiao Xie int try_to_writeback_inodes_sb_nr(struct super_block *sb, 13980e175a18SCurt Wohlgemuth unsigned long nr, 13990e175a18SCurt Wohlgemuth enum wb_reason reason) 14003259f8beSChris Mason { 140110ee27a0SMiao Xie if (writeback_in_progress(sb->s_bdi)) 140210ee27a0SMiao Xie return 1; 140310ee27a0SMiao Xie 140410ee27a0SMiao Xie if (!down_read_trylock(&sb->s_umount)) 140510ee27a0SMiao Xie return 0; 140610ee27a0SMiao Xie 14070e175a18SCurt Wohlgemuth writeback_inodes_sb_nr(sb, nr, reason); 14083259f8beSChris Mason up_read(&sb->s_umount); 14093259f8beSChris Mason return 1; 14103259f8beSChris Mason } 141110ee27a0SMiao Xie EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr); 141210ee27a0SMiao Xie 141310ee27a0SMiao Xie /** 141410ee27a0SMiao Xie * try_to_writeback_inodes_sb - try to start writeback if none underway 141510ee27a0SMiao Xie * @sb: the superblock 141610ee27a0SMiao Xie * @reason: reason why some writeback work was initiated 141710ee27a0SMiao Xie * 141810ee27a0SMiao Xie * Implement by try_to_writeback_inodes_sb_nr() 141910ee27a0SMiao Xie * Returns 1 if writeback was started, 0 if not. 142010ee27a0SMiao Xie */ 142110ee27a0SMiao Xie int try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) 142210ee27a0SMiao Xie { 142310ee27a0SMiao Xie return try_to_writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); 142410ee27a0SMiao Xie } 142510ee27a0SMiao Xie EXPORT_SYMBOL(try_to_writeback_inodes_sb); 14263259f8beSChris Mason 14273259f8beSChris Mason /** 1428d8a8559cSJens Axboe * sync_inodes_sb - sync sb inode pages 1429d8a8559cSJens Axboe * @sb: the superblock 1430d8a8559cSJens Axboe * 1431d8a8559cSJens Axboe * This function writes and waits on any dirty inode belonging to this 14320dc83bd3SJan Kara * super_block. 1433d8a8559cSJens Axboe */ 14340dc83bd3SJan Kara void sync_inodes_sb(struct super_block *sb) 1435d8a8559cSJens Axboe { 143683ba7b07SChristoph Hellwig DECLARE_COMPLETION_ONSTACK(done); 143783ba7b07SChristoph Hellwig struct wb_writeback_work work = { 14383c4d7165SChristoph Hellwig .sb = sb, 14393c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_ALL, 14403c4d7165SChristoph Hellwig .nr_pages = LONG_MAX, 14413c4d7165SChristoph Hellwig .range_cyclic = 0, 144283ba7b07SChristoph Hellwig .done = &done, 14430e175a18SCurt Wohlgemuth .reason = WB_REASON_SYNC, 14447747bd4bSDave Chinner .for_sync = 1, 14453c4d7165SChristoph Hellwig }; 14463c4d7165SChristoph Hellwig 14476eedc701SJan Kara /* Nothing to do? */ 14486eedc701SJan Kara if (sb->s_bdi == &noop_backing_dev_info) 14496eedc701SJan Kara return; 1450cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 1451cf37e972SChristoph Hellwig 145283ba7b07SChristoph Hellwig bdi_queue_work(sb->s_bdi, &work); 145383ba7b07SChristoph Hellwig wait_for_completion(&done); 145483ba7b07SChristoph Hellwig 1455b6e51316SJens Axboe wait_sb_inodes(sb); 1456d8a8559cSJens Axboe } 1457d8a8559cSJens Axboe EXPORT_SYMBOL(sync_inodes_sb); 14581da177e4SLinus Torvalds 14591da177e4SLinus Torvalds /** 14601da177e4SLinus Torvalds * write_inode_now - write an inode to disk 14611da177e4SLinus Torvalds * @inode: inode to write to disk 14621da177e4SLinus Torvalds * @sync: whether the write should be synchronous or not 14631da177e4SLinus Torvalds * 14647f04c26dSAndrea Arcangeli * This function commits an inode to disk immediately if it is dirty. This is 14657f04c26dSAndrea Arcangeli * primarily needed by knfsd. 14667f04c26dSAndrea Arcangeli * 14677f04c26dSAndrea Arcangeli * The caller must either have a ref on the inode or must have set I_WILL_FREE. 14681da177e4SLinus Torvalds */ 14691da177e4SLinus Torvalds int write_inode_now(struct inode *inode, int sync) 14701da177e4SLinus Torvalds { 1471f758eeabSChristoph Hellwig struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; 14721da177e4SLinus Torvalds struct writeback_control wbc = { 14731da177e4SLinus Torvalds .nr_to_write = LONG_MAX, 147418914b18SMike Galbraith .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, 1475111ebb6eSOGAWA Hirofumi .range_start = 0, 1476111ebb6eSOGAWA Hirofumi .range_end = LLONG_MAX, 14771da177e4SLinus Torvalds }; 14781da177e4SLinus Torvalds 14791da177e4SLinus Torvalds if (!mapping_cap_writeback_dirty(inode->i_mapping)) 148049364ce2SAndrew Morton wbc.nr_to_write = 0; 14811da177e4SLinus Torvalds 14821da177e4SLinus Torvalds might_sleep(); 14834f8ad655SJan Kara return writeback_single_inode(inode, wb, &wbc); 14841da177e4SLinus Torvalds } 14851da177e4SLinus Torvalds EXPORT_SYMBOL(write_inode_now); 14861da177e4SLinus Torvalds 14871da177e4SLinus Torvalds /** 14881da177e4SLinus Torvalds * sync_inode - write an inode and its pages to disk. 14891da177e4SLinus Torvalds * @inode: the inode to sync 14901da177e4SLinus Torvalds * @wbc: controls the writeback mode 14911da177e4SLinus Torvalds * 14921da177e4SLinus Torvalds * sync_inode() will write an inode and its pages to disk. It will also 14931da177e4SLinus Torvalds * correctly update the inode on its superblock's dirty inode lists and will 14941da177e4SLinus Torvalds * update inode->i_state. 14951da177e4SLinus Torvalds * 14961da177e4SLinus Torvalds * The caller must have a ref on the inode. 14971da177e4SLinus Torvalds */ 14981da177e4SLinus Torvalds int sync_inode(struct inode *inode, struct writeback_control *wbc) 14991da177e4SLinus Torvalds { 15004f8ad655SJan Kara return writeback_single_inode(inode, &inode_to_bdi(inode)->wb, wbc); 15011da177e4SLinus Torvalds } 15021da177e4SLinus Torvalds EXPORT_SYMBOL(sync_inode); 1503c3765016SChristoph Hellwig 1504c3765016SChristoph Hellwig /** 1505c691b9d9SAndrew Morton * sync_inode_metadata - write an inode to disk 1506c3765016SChristoph Hellwig * @inode: the inode to sync 1507c3765016SChristoph Hellwig * @wait: wait for I/O to complete. 1508c3765016SChristoph Hellwig * 1509c691b9d9SAndrew Morton * Write an inode to disk and adjust its dirty state after completion. 1510c3765016SChristoph Hellwig * 1511c3765016SChristoph Hellwig * Note: only writes the actual inode, no associated data or other metadata. 1512c3765016SChristoph Hellwig */ 1513c3765016SChristoph Hellwig int sync_inode_metadata(struct inode *inode, int wait) 1514c3765016SChristoph Hellwig { 1515c3765016SChristoph Hellwig struct writeback_control wbc = { 1516c3765016SChristoph Hellwig .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE, 1517c3765016SChristoph Hellwig .nr_to_write = 0, /* metadata-only */ 1518c3765016SChristoph Hellwig }; 1519c3765016SChristoph Hellwig 1520c3765016SChristoph Hellwig return sync_inode(inode, &wbc); 1521c3765016SChristoph Hellwig } 1522c3765016SChristoph Hellwig EXPORT_SYMBOL(sync_inode_metadata); 1523