11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * fs/fs-writeback.c 31da177e4SLinus Torvalds * 41da177e4SLinus Torvalds * Copyright (C) 2002, Linus Torvalds. 51da177e4SLinus Torvalds * 61da177e4SLinus Torvalds * Contains all the functions related to writing back and waiting 71da177e4SLinus Torvalds * upon dirty inodes against superblocks, and writing back dirty 81da177e4SLinus Torvalds * pages against inodes. ie: data writeback. Writeout of the 91da177e4SLinus Torvalds * inode itself is not handled here. 101da177e4SLinus Torvalds * 11e1f8e874SFrancois Cami * 10Apr2002 Andrew Morton 121da177e4SLinus Torvalds * Split out of fs/inode.c 131da177e4SLinus Torvalds * Additions for address_space-based writeback 141da177e4SLinus Torvalds */ 151da177e4SLinus Torvalds 161da177e4SLinus Torvalds #include <linux/kernel.h> 17f5ff8422SJens Axboe #include <linux/module.h> 181da177e4SLinus Torvalds #include <linux/spinlock.h> 195a0e3ad6STejun Heo #include <linux/slab.h> 201da177e4SLinus Torvalds #include <linux/sched.h> 211da177e4SLinus Torvalds #include <linux/fs.h> 221da177e4SLinus Torvalds #include <linux/mm.h> 2303ba3782SJens Axboe #include <linux/kthread.h> 2403ba3782SJens Axboe #include <linux/freezer.h> 251da177e4SLinus Torvalds #include <linux/writeback.h> 261da177e4SLinus Torvalds #include <linux/blkdev.h> 271da177e4SLinus Torvalds #include <linux/backing-dev.h> 281da177e4SLinus Torvalds #include <linux/buffer_head.h> 29455b2864SDave Chinner #include <linux/tracepoint.h> 3007f3f05cSDavid Howells #include "internal.h" 311da177e4SLinus Torvalds 32d0bceac7SJens Axboe /* 33c4a77a6cSJens Axboe * Passed into wb_writeback(), essentially a subset of writeback_control 34c4a77a6cSJens Axboe */ 3583ba7b07SChristoph Hellwig struct wb_writeback_work { 36c4a77a6cSJens Axboe long nr_pages; 37c4a77a6cSJens Axboe struct super_block *sb; 38d46db3d5SWu Fengguang unsigned long *older_than_this; 39c4a77a6cSJens Axboe enum writeback_sync_modes sync_mode; 406e6938b6SWu Fengguang unsigned int tagged_writepages:1; 4152957fe1SH Hartley Sweeten unsigned int for_kupdate:1; 4252957fe1SH Hartley Sweeten unsigned int range_cyclic:1; 4352957fe1SH Hartley Sweeten unsigned int for_background:1; 440e175a18SCurt Wohlgemuth enum wb_reason reason; /* why was writeback initiated? */ 45c4a77a6cSJens Axboe 468010c3b6SJens Axboe struct list_head list; /* pending work list */ 4783ba7b07SChristoph Hellwig struct completion *done; /* set if the caller waits */ 4803ba3782SJens Axboe }; 4903ba3782SJens Axboe 500e175a18SCurt Wohlgemuth const char *wb_reason_name[] = { 510e175a18SCurt Wohlgemuth [WB_REASON_BACKGROUND] = "background", 520e175a18SCurt Wohlgemuth [WB_REASON_TRY_TO_FREE_PAGES] = "try_to_free_pages", 530e175a18SCurt Wohlgemuth [WB_REASON_SYNC] = "sync", 540e175a18SCurt Wohlgemuth [WB_REASON_PERIODIC] = "periodic", 550e175a18SCurt Wohlgemuth [WB_REASON_LAPTOP_TIMER] = "laptop_timer", 560e175a18SCurt Wohlgemuth [WB_REASON_FREE_MORE_MEM] = "free_more_memory", 570e175a18SCurt Wohlgemuth [WB_REASON_FS_FREE_SPACE] = "fs_free_space", 580e175a18SCurt Wohlgemuth [WB_REASON_FORKER_THREAD] = "forker_thread" 590e175a18SCurt Wohlgemuth }; 600e175a18SCurt Wohlgemuth 61455b2864SDave Chinner /* 62455b2864SDave Chinner * Include the creation of the trace points after defining the 63455b2864SDave Chinner * wb_writeback_work structure so that the definition remains local to this 64455b2864SDave Chinner * file. 65455b2864SDave Chinner */ 66455b2864SDave Chinner #define CREATE_TRACE_POINTS 67455b2864SDave Chinner #include <trace/events/writeback.h> 68455b2864SDave Chinner 69455b2864SDave Chinner /* 70455b2864SDave Chinner * We don't actually have pdflush, but this one is exported though /proc... 71455b2864SDave Chinner */ 72455b2864SDave Chinner int nr_pdflush_threads; 73455b2864SDave Chinner 74f11b00f3SAdrian Bunk /** 75f11b00f3SAdrian Bunk * writeback_in_progress - determine whether there is writeback in progress 76f11b00f3SAdrian Bunk * @bdi: the device's backing_dev_info structure. 77f11b00f3SAdrian Bunk * 7803ba3782SJens Axboe * Determine whether there is writeback waiting to be handled against a 7903ba3782SJens Axboe * backing device. 80f11b00f3SAdrian Bunk */ 81f11b00f3SAdrian Bunk int writeback_in_progress(struct backing_dev_info *bdi) 82f11b00f3SAdrian Bunk { 8381d73a32SJan Kara return test_bit(BDI_writeback_running, &bdi->state); 84f11b00f3SAdrian Bunk } 85f11b00f3SAdrian Bunk 86692ebd17SJan Kara static inline struct backing_dev_info *inode_to_bdi(struct inode *inode) 87692ebd17SJan Kara { 88692ebd17SJan Kara struct super_block *sb = inode->i_sb; 89692ebd17SJan Kara 90aaead25bSChristoph Hellwig if (strcmp(sb->s_type->name, "bdev") == 0) 91aaead25bSChristoph Hellwig return inode->i_mapping->backing_dev_info; 92aaead25bSChristoph Hellwig 93692ebd17SJan Kara return sb->s_bdi; 94692ebd17SJan Kara } 95692ebd17SJan Kara 967ccf19a8SNick Piggin static inline struct inode *wb_inode(struct list_head *head) 977ccf19a8SNick Piggin { 987ccf19a8SNick Piggin return list_entry(head, struct inode, i_wb_list); 997ccf19a8SNick Piggin } 1007ccf19a8SNick Piggin 1016585027aSJan Kara /* Wakeup flusher thread or forker thread to fork it. Requires bdi->wb_lock. */ 1026585027aSJan Kara static void bdi_wakeup_flusher(struct backing_dev_info *bdi) 1034195f73dSNick Piggin { 104fff5b85aSArtem Bityutskiy if (bdi->wb.task) { 105fff5b85aSArtem Bityutskiy wake_up_process(bdi->wb.task); 106fff5b85aSArtem Bityutskiy } else { 1071da177e4SLinus Torvalds /* 108fff5b85aSArtem Bityutskiy * The bdi thread isn't there, wake up the forker thread which 109fff5b85aSArtem Bityutskiy * will create and run it. 1101da177e4SLinus Torvalds */ 11103ba3782SJens Axboe wake_up_process(default_backing_dev_info.wb.task); 1121da177e4SLinus Torvalds } 1136585027aSJan Kara } 1146585027aSJan Kara 1156585027aSJan Kara static void bdi_queue_work(struct backing_dev_info *bdi, 1166585027aSJan Kara struct wb_writeback_work *work) 1176585027aSJan Kara { 1186585027aSJan Kara trace_writeback_queue(bdi, work); 1196585027aSJan Kara 1206585027aSJan Kara spin_lock_bh(&bdi->wb_lock); 1216585027aSJan Kara list_add_tail(&work->list, &bdi->work_list); 1226585027aSJan Kara if (!bdi->wb.task) 1236585027aSJan Kara trace_writeback_nothread(bdi, work); 1246585027aSJan Kara bdi_wakeup_flusher(bdi); 1256467716aSArtem Bityutskiy spin_unlock_bh(&bdi->wb_lock); 12603ba3782SJens Axboe } 1271da177e4SLinus Torvalds 12883ba7b07SChristoph Hellwig static void 12983ba7b07SChristoph Hellwig __bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, 1300e175a18SCurt Wohlgemuth bool range_cyclic, enum wb_reason reason) 1311da177e4SLinus Torvalds { 13283ba7b07SChristoph Hellwig struct wb_writeback_work *work; 13303ba3782SJens Axboe 134bcddc3f0SJens Axboe /* 135bcddc3f0SJens Axboe * This is WB_SYNC_NONE writeback, so if allocation fails just 136bcddc3f0SJens Axboe * wakeup the thread for old dirty data writeback 137bcddc3f0SJens Axboe */ 13883ba7b07SChristoph Hellwig work = kzalloc(sizeof(*work), GFP_ATOMIC); 13983ba7b07SChristoph Hellwig if (!work) { 140455b2864SDave Chinner if (bdi->wb.task) { 141455b2864SDave Chinner trace_writeback_nowork(bdi); 14283ba7b07SChristoph Hellwig wake_up_process(bdi->wb.task); 143455b2864SDave Chinner } 14483ba7b07SChristoph Hellwig return; 14583ba7b07SChristoph Hellwig } 14683ba7b07SChristoph Hellwig 14783ba7b07SChristoph Hellwig work->sync_mode = WB_SYNC_NONE; 14883ba7b07SChristoph Hellwig work->nr_pages = nr_pages; 14983ba7b07SChristoph Hellwig work->range_cyclic = range_cyclic; 1500e175a18SCurt Wohlgemuth work->reason = reason; 15183ba7b07SChristoph Hellwig 152f11fcae8SJens Axboe bdi_queue_work(bdi, work); 15303ba3782SJens Axboe } 154b6e51316SJens Axboe 155b6e51316SJens Axboe /** 156b6e51316SJens Axboe * bdi_start_writeback - start writeback 157b6e51316SJens Axboe * @bdi: the backing device to write from 158b6e51316SJens Axboe * @nr_pages: the number of pages to write 159786228abSMarcos Paulo de Souza * @reason: reason why some writeback work was initiated 160b6e51316SJens Axboe * 161b6e51316SJens Axboe * Description: 162b6e51316SJens Axboe * This does WB_SYNC_NONE opportunistic writeback. The IO is only 16325985edcSLucas De Marchi * started when this function returns, we make no guarantees on 1640e3c9a22SJens Axboe * completion. Caller need not hold sb s_umount semaphore. 165b6e51316SJens Axboe * 166b6e51316SJens Axboe */ 1670e175a18SCurt Wohlgemuth void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, 1680e175a18SCurt Wohlgemuth enum wb_reason reason) 169b6e51316SJens Axboe { 1700e175a18SCurt Wohlgemuth __bdi_start_writeback(bdi, nr_pages, true, reason); 171d3ddec76SWu Fengguang } 172d3ddec76SWu Fengguang 173c5444198SChristoph Hellwig /** 174c5444198SChristoph Hellwig * bdi_start_background_writeback - start background writeback 175c5444198SChristoph Hellwig * @bdi: the backing device to write from 176c5444198SChristoph Hellwig * 177c5444198SChristoph Hellwig * Description: 1786585027aSJan Kara * This makes sure WB_SYNC_NONE background writeback happens. When 1796585027aSJan Kara * this function returns, it is only guaranteed that for given BDI 1806585027aSJan Kara * some IO is happening if we are over background dirty threshold. 1816585027aSJan Kara * Caller need not hold sb s_umount semaphore. 182c5444198SChristoph Hellwig */ 183c5444198SChristoph Hellwig void bdi_start_background_writeback(struct backing_dev_info *bdi) 184c5444198SChristoph Hellwig { 1856585027aSJan Kara /* 1866585027aSJan Kara * We just wake up the flusher thread. It will perform background 1876585027aSJan Kara * writeback as soon as there is no other work to do. 1886585027aSJan Kara */ 18971927e84SWu Fengguang trace_writeback_wake_background(bdi); 1906585027aSJan Kara spin_lock_bh(&bdi->wb_lock); 1916585027aSJan Kara bdi_wakeup_flusher(bdi); 1926585027aSJan Kara spin_unlock_bh(&bdi->wb_lock); 1931da177e4SLinus Torvalds } 1941da177e4SLinus Torvalds 1951da177e4SLinus Torvalds /* 196a66979abSDave Chinner * Remove the inode from the writeback list it is on. 197a66979abSDave Chinner */ 198a66979abSDave Chinner void inode_wb_list_del(struct inode *inode) 199a66979abSDave Chinner { 200f758eeabSChristoph Hellwig struct backing_dev_info *bdi = inode_to_bdi(inode); 201a66979abSDave Chinner 202f758eeabSChristoph Hellwig spin_lock(&bdi->wb.list_lock); 203f758eeabSChristoph Hellwig list_del_init(&inode->i_wb_list); 204f758eeabSChristoph Hellwig spin_unlock(&bdi->wb.list_lock); 205f758eeabSChristoph Hellwig } 206a66979abSDave Chinner 207a66979abSDave Chinner /* 2086610a0bcSAndrew Morton * Redirty an inode: set its when-it-was dirtied timestamp and move it to the 2096610a0bcSAndrew Morton * furthest end of its superblock's dirty-inode list. 2106610a0bcSAndrew Morton * 2116610a0bcSAndrew Morton * Before stamping the inode's ->dirtied_when, we check to see whether it is 21266f3b8e2SJens Axboe * already the most-recently-dirtied inode on the b_dirty list. If that is 2136610a0bcSAndrew Morton * the case then the inode must have been redirtied while it was being written 2146610a0bcSAndrew Morton * out and we don't reset its dirtied_when. 2156610a0bcSAndrew Morton */ 216f758eeabSChristoph Hellwig static void redirty_tail(struct inode *inode, struct bdi_writeback *wb) 2176610a0bcSAndrew Morton { 218f758eeabSChristoph Hellwig assert_spin_locked(&wb->list_lock); 21903ba3782SJens Axboe if (!list_empty(&wb->b_dirty)) { 22066f3b8e2SJens Axboe struct inode *tail; 2216610a0bcSAndrew Morton 2227ccf19a8SNick Piggin tail = wb_inode(wb->b_dirty.next); 22366f3b8e2SJens Axboe if (time_before(inode->dirtied_when, tail->dirtied_when)) 2246610a0bcSAndrew Morton inode->dirtied_when = jiffies; 2256610a0bcSAndrew Morton } 2267ccf19a8SNick Piggin list_move(&inode->i_wb_list, &wb->b_dirty); 2276610a0bcSAndrew Morton } 2286610a0bcSAndrew Morton 2296610a0bcSAndrew Morton /* 23066f3b8e2SJens Axboe * requeue inode for re-scanning after bdi->b_io list is exhausted. 231c986d1e2SAndrew Morton */ 232f758eeabSChristoph Hellwig static void requeue_io(struct inode *inode, struct bdi_writeback *wb) 233c986d1e2SAndrew Morton { 234f758eeabSChristoph Hellwig assert_spin_locked(&wb->list_lock); 2357ccf19a8SNick Piggin list_move(&inode->i_wb_list, &wb->b_more_io); 236c986d1e2SAndrew Morton } 237c986d1e2SAndrew Morton 2381c0eeaf5SJoern Engel static void inode_sync_complete(struct inode *inode) 2391c0eeaf5SJoern Engel { 2401c0eeaf5SJoern Engel /* 241a66979abSDave Chinner * Prevent speculative execution through 242f758eeabSChristoph Hellwig * spin_unlock(&wb->list_lock); 2431c0eeaf5SJoern Engel */ 244a66979abSDave Chinner 2451c0eeaf5SJoern Engel smp_mb(); 2461c0eeaf5SJoern Engel wake_up_bit(&inode->i_state, __I_SYNC); 2471c0eeaf5SJoern Engel } 2481c0eeaf5SJoern Engel 249d2caa3c5SJeff Layton static bool inode_dirtied_after(struct inode *inode, unsigned long t) 250d2caa3c5SJeff Layton { 251d2caa3c5SJeff Layton bool ret = time_after(inode->dirtied_when, t); 252d2caa3c5SJeff Layton #ifndef CONFIG_64BIT 253d2caa3c5SJeff Layton /* 254d2caa3c5SJeff Layton * For inodes being constantly redirtied, dirtied_when can get stuck. 255d2caa3c5SJeff Layton * It _appears_ to be in the future, but is actually in distant past. 256d2caa3c5SJeff Layton * This test is necessary to prevent such wrapped-around relative times 2575b0830cbSJens Axboe * from permanently stopping the whole bdi writeback. 258d2caa3c5SJeff Layton */ 259d2caa3c5SJeff Layton ret = ret && time_before_eq(inode->dirtied_when, jiffies); 260d2caa3c5SJeff Layton #endif 261d2caa3c5SJeff Layton return ret; 262d2caa3c5SJeff Layton } 263d2caa3c5SJeff Layton 264c986d1e2SAndrew Morton /* 2652c136579SFengguang Wu * Move expired dirty inodes from @delaying_queue to @dispatch_queue. 2662c136579SFengguang Wu */ 267e84d0a4fSWu Fengguang static int move_expired_inodes(struct list_head *delaying_queue, 2682c136579SFengguang Wu struct list_head *dispatch_queue, 269ad4e38ddSCurt Wohlgemuth struct wb_writeback_work *work) 2702c136579SFengguang Wu { 2715c03449dSShaohua Li LIST_HEAD(tmp); 2725c03449dSShaohua Li struct list_head *pos, *node; 273cf137307SJens Axboe struct super_block *sb = NULL; 2745c03449dSShaohua Li struct inode *inode; 275cf137307SJens Axboe int do_sb_sort = 0; 276e84d0a4fSWu Fengguang int moved = 0; 2775c03449dSShaohua Li 2782c136579SFengguang Wu while (!list_empty(delaying_queue)) { 2797ccf19a8SNick Piggin inode = wb_inode(delaying_queue->prev); 280ad4e38ddSCurt Wohlgemuth if (work->older_than_this && 281ad4e38ddSCurt Wohlgemuth inode_dirtied_after(inode, *work->older_than_this)) 2822c136579SFengguang Wu break; 283cf137307SJens Axboe if (sb && sb != inode->i_sb) 284cf137307SJens Axboe do_sb_sort = 1; 285cf137307SJens Axboe sb = inode->i_sb; 2867ccf19a8SNick Piggin list_move(&inode->i_wb_list, &tmp); 287e84d0a4fSWu Fengguang moved++; 2885c03449dSShaohua Li } 2895c03449dSShaohua Li 290cf137307SJens Axboe /* just one sb in list, splice to dispatch_queue and we're done */ 291cf137307SJens Axboe if (!do_sb_sort) { 292cf137307SJens Axboe list_splice(&tmp, dispatch_queue); 293e84d0a4fSWu Fengguang goto out; 294cf137307SJens Axboe } 295cf137307SJens Axboe 2965c03449dSShaohua Li /* Move inodes from one superblock together */ 2975c03449dSShaohua Li while (!list_empty(&tmp)) { 2987ccf19a8SNick Piggin sb = wb_inode(tmp.prev)->i_sb; 2995c03449dSShaohua Li list_for_each_prev_safe(pos, node, &tmp) { 3007ccf19a8SNick Piggin inode = wb_inode(pos); 3015c03449dSShaohua Li if (inode->i_sb == sb) 3027ccf19a8SNick Piggin list_move(&inode->i_wb_list, dispatch_queue); 3032c136579SFengguang Wu } 3042c136579SFengguang Wu } 305e84d0a4fSWu Fengguang out: 306e84d0a4fSWu Fengguang return moved; 3075c03449dSShaohua Li } 3082c136579SFengguang Wu 3092c136579SFengguang Wu /* 3102c136579SFengguang Wu * Queue all expired dirty inodes for io, eldest first. 3114ea879b9SWu Fengguang * Before 3124ea879b9SWu Fengguang * newly dirtied b_dirty b_io b_more_io 3134ea879b9SWu Fengguang * =============> gf edc BA 3144ea879b9SWu Fengguang * After 3154ea879b9SWu Fengguang * newly dirtied b_dirty b_io b_more_io 3164ea879b9SWu Fengguang * =============> g fBAedc 3174ea879b9SWu Fengguang * | 3184ea879b9SWu Fengguang * +--> dequeue for IO 3192c136579SFengguang Wu */ 320ad4e38ddSCurt Wohlgemuth static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work) 3212c136579SFengguang Wu { 322e84d0a4fSWu Fengguang int moved; 323f758eeabSChristoph Hellwig assert_spin_locked(&wb->list_lock); 3244ea879b9SWu Fengguang list_splice_init(&wb->b_more_io, &wb->b_io); 325ad4e38ddSCurt Wohlgemuth moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, work); 326ad4e38ddSCurt Wohlgemuth trace_writeback_queue_io(wb, work, moved); 32766f3b8e2SJens Axboe } 32866f3b8e2SJens Axboe 329a9185b41SChristoph Hellwig static int write_inode(struct inode *inode, struct writeback_control *wbc) 33066f3b8e2SJens Axboe { 33103ba3782SJens Axboe if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) 332a9185b41SChristoph Hellwig return inode->i_sb->s_op->write_inode(inode, wbc); 33303ba3782SJens Axboe return 0; 33466f3b8e2SJens Axboe } 33508d8e974SFengguang Wu 3362c136579SFengguang Wu /* 33701c03194SChristoph Hellwig * Wait for writeback on an inode to complete. 33801c03194SChristoph Hellwig */ 339f758eeabSChristoph Hellwig static void inode_wait_for_writeback(struct inode *inode, 340f758eeabSChristoph Hellwig struct bdi_writeback *wb) 34101c03194SChristoph Hellwig { 34201c03194SChristoph Hellwig DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); 34301c03194SChristoph Hellwig wait_queue_head_t *wqh; 34401c03194SChristoph Hellwig 34501c03194SChristoph Hellwig wqh = bit_waitqueue(&inode->i_state, __I_SYNC); 34658a9d3d8SRichard Kennedy while (inode->i_state & I_SYNC) { 347250df6edSDave Chinner spin_unlock(&inode->i_lock); 348f758eeabSChristoph Hellwig spin_unlock(&wb->list_lock); 34901c03194SChristoph Hellwig __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE); 350f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 351250df6edSDave Chinner spin_lock(&inode->i_lock); 35258a9d3d8SRichard Kennedy } 35301c03194SChristoph Hellwig } 35401c03194SChristoph Hellwig 35501c03194SChristoph Hellwig /* 356f758eeabSChristoph Hellwig * Write out an inode's dirty pages. Called under wb->list_lock and 3570f1b1fd8SDave Chinner * inode->i_lock. Either the caller has an active reference on the inode or 3580f1b1fd8SDave Chinner * the inode has I_WILL_FREE set. 35901c03194SChristoph Hellwig * 3601da177e4SLinus Torvalds * If `wait' is set, wait on the writeout. 3611da177e4SLinus Torvalds * 3621da177e4SLinus Torvalds * The whole writeout design is quite complex and fragile. We want to avoid 3631da177e4SLinus Torvalds * starvation of particular inodes when others are being redirtied, prevent 3641da177e4SLinus Torvalds * livelocks, etc. 3651da177e4SLinus Torvalds */ 3661da177e4SLinus Torvalds static int 367f758eeabSChristoph Hellwig writeback_single_inode(struct inode *inode, struct bdi_writeback *wb, 368f758eeabSChristoph Hellwig struct writeback_control *wbc) 3691da177e4SLinus Torvalds { 3701da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 371251d6a47SWu Fengguang long nr_to_write = wbc->nr_to_write; 37201c03194SChristoph Hellwig unsigned dirty; 3731da177e4SLinus Torvalds int ret; 3741da177e4SLinus Torvalds 375f758eeabSChristoph Hellwig assert_spin_locked(&wb->list_lock); 3760f1b1fd8SDave Chinner assert_spin_locked(&inode->i_lock); 3770f1b1fd8SDave Chinner 37801c03194SChristoph Hellwig if (!atomic_read(&inode->i_count)) 37901c03194SChristoph Hellwig WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); 38001c03194SChristoph Hellwig else 38101c03194SChristoph Hellwig WARN_ON(inode->i_state & I_WILL_FREE); 38201c03194SChristoph Hellwig 38301c03194SChristoph Hellwig if (inode->i_state & I_SYNC) { 38401c03194SChristoph Hellwig /* 38501c03194SChristoph Hellwig * If this inode is locked for writeback and we are not doing 38666f3b8e2SJens Axboe * writeback-for-data-integrity, move it to b_more_io so that 38701c03194SChristoph Hellwig * writeback can proceed with the other inodes on s_io. 38801c03194SChristoph Hellwig * 38901c03194SChristoph Hellwig * We'll have another go at writing back this inode when we 39066f3b8e2SJens Axboe * completed a full scan of b_io. 39101c03194SChristoph Hellwig */ 392a9185b41SChristoph Hellwig if (wbc->sync_mode != WB_SYNC_ALL) { 393f758eeabSChristoph Hellwig requeue_io(inode, wb); 394251d6a47SWu Fengguang trace_writeback_single_inode_requeue(inode, wbc, 395251d6a47SWu Fengguang nr_to_write); 39601c03194SChristoph Hellwig return 0; 39701c03194SChristoph Hellwig } 39801c03194SChristoph Hellwig 39901c03194SChristoph Hellwig /* 40001c03194SChristoph Hellwig * It's a data-integrity sync. We must wait. 40101c03194SChristoph Hellwig */ 402f758eeabSChristoph Hellwig inode_wait_for_writeback(inode, wb); 40301c03194SChristoph Hellwig } 40401c03194SChristoph Hellwig 4051c0eeaf5SJoern Engel BUG_ON(inode->i_state & I_SYNC); 4061da177e4SLinus Torvalds 4075547e8aaSDmitry Monakhov /* Set I_SYNC, reset I_DIRTY_PAGES */ 4081c0eeaf5SJoern Engel inode->i_state |= I_SYNC; 4095547e8aaSDmitry Monakhov inode->i_state &= ~I_DIRTY_PAGES; 410250df6edSDave Chinner spin_unlock(&inode->i_lock); 411f758eeabSChristoph Hellwig spin_unlock(&wb->list_lock); 4121da177e4SLinus Torvalds 4131da177e4SLinus Torvalds ret = do_writepages(mapping, wbc); 4141da177e4SLinus Torvalds 41526821ed4SChristoph Hellwig /* 41626821ed4SChristoph Hellwig * Make sure to wait on the data before writing out the metadata. 41726821ed4SChristoph Hellwig * This is important for filesystems that modify metadata on data 41826821ed4SChristoph Hellwig * I/O completion. 41926821ed4SChristoph Hellwig */ 420a9185b41SChristoph Hellwig if (wbc->sync_mode == WB_SYNC_ALL) { 42126821ed4SChristoph Hellwig int err = filemap_fdatawait(mapping); 4221da177e4SLinus Torvalds if (ret == 0) 4231da177e4SLinus Torvalds ret = err; 4241da177e4SLinus Torvalds } 4251da177e4SLinus Torvalds 4265547e8aaSDmitry Monakhov /* 4275547e8aaSDmitry Monakhov * Some filesystems may redirty the inode during the writeback 4285547e8aaSDmitry Monakhov * due to delalloc, clear dirty metadata flags right before 4295547e8aaSDmitry Monakhov * write_inode() 4305547e8aaSDmitry Monakhov */ 431250df6edSDave Chinner spin_lock(&inode->i_lock); 4325547e8aaSDmitry Monakhov dirty = inode->i_state & I_DIRTY; 4335547e8aaSDmitry Monakhov inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC); 434250df6edSDave Chinner spin_unlock(&inode->i_lock); 43526821ed4SChristoph Hellwig /* Don't write the inode if only I_DIRTY_PAGES was set */ 43626821ed4SChristoph Hellwig if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { 437a9185b41SChristoph Hellwig int err = write_inode(inode, wbc); 4381da177e4SLinus Torvalds if (ret == 0) 4391da177e4SLinus Torvalds ret = err; 4401da177e4SLinus Torvalds } 4411da177e4SLinus Torvalds 442f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 443250df6edSDave Chinner spin_lock(&inode->i_lock); 4441c0eeaf5SJoern Engel inode->i_state &= ~I_SYNC; 445a4ffdde6SAl Viro if (!(inode->i_state & I_FREEING)) { 44694c3dcbbSWu Fengguang /* 44794c3dcbbSWu Fengguang * Sync livelock prevention. Each inode is tagged and synced in 44894c3dcbbSWu Fengguang * one shot. If still dirty, it will be redirty_tail()'ed below. 44994c3dcbbSWu Fengguang * Update the dirty time to prevent enqueue and sync it again. 45094c3dcbbSWu Fengguang */ 45194c3dcbbSWu Fengguang if ((inode->i_state & I_DIRTY) && 45294c3dcbbSWu Fengguang (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)) 45394c3dcbbSWu Fengguang inode->dirtied_when = jiffies; 45494c3dcbbSWu Fengguang 45523539afcSWu Fengguang if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { 4561da177e4SLinus Torvalds /* 4571da177e4SLinus Torvalds * We didn't write back all the pages. nfs_writepages() 458a50aeb40SWu Fengguang * sometimes bales out without doing anything. 4591da177e4SLinus Torvalds */ 4601da177e4SLinus Torvalds inode->i_state |= I_DIRTY_PAGES; 4618bc3be27SFengguang Wu if (wbc->nr_to_write <= 0) { 4628bc3be27SFengguang Wu /* 4638bc3be27SFengguang Wu * slice used up: queue for next turn 4648bc3be27SFengguang Wu */ 465f758eeabSChristoph Hellwig requeue_io(inode, wb); 4661da177e4SLinus Torvalds } else { 4671da177e4SLinus Torvalds /* 468a50aeb40SWu Fengguang * Writeback blocked by something other than 469a50aeb40SWu Fengguang * congestion. Delay the inode for some time to 470a50aeb40SWu Fengguang * avoid spinning on the CPU (100% iowait) 471a50aeb40SWu Fengguang * retrying writeback of the dirty page/inode 472a50aeb40SWu Fengguang * that cannot be performed immediately. 4738bc3be27SFengguang Wu */ 474f758eeabSChristoph Hellwig redirty_tail(inode, wb); 4758bc3be27SFengguang Wu } 47623539afcSWu Fengguang } else if (inode->i_state & I_DIRTY) { 47723539afcSWu Fengguang /* 47823539afcSWu Fengguang * Filesystems can dirty the inode during writeback 47923539afcSWu Fengguang * operations, such as delayed allocation during 48023539afcSWu Fengguang * submission or metadata updates after data IO 48123539afcSWu Fengguang * completion. 48223539afcSWu Fengguang */ 483f758eeabSChristoph Hellwig redirty_tail(inode, wb); 4841da177e4SLinus Torvalds } else { 4851da177e4SLinus Torvalds /* 4869e38d86fSNick Piggin * The inode is clean. At this point we either have 4879e38d86fSNick Piggin * a reference to the inode or it's on it's way out. 4889e38d86fSNick Piggin * No need to add it back to the LRU. 4891da177e4SLinus Torvalds */ 4907ccf19a8SNick Piggin list_del_init(&inode->i_wb_list); 4911da177e4SLinus Torvalds } 4921da177e4SLinus Torvalds } 4931c0eeaf5SJoern Engel inode_sync_complete(inode); 494251d6a47SWu Fengguang trace_writeback_single_inode(inode, wbc, nr_to_write); 4951da177e4SLinus Torvalds return ret; 4961da177e4SLinus Torvalds } 4971da177e4SLinus Torvalds 4981a12d8bdSWu Fengguang static long writeback_chunk_size(struct backing_dev_info *bdi, 4991a12d8bdSWu Fengguang struct wb_writeback_work *work) 500d46db3d5SWu Fengguang { 501d46db3d5SWu Fengguang long pages; 502d46db3d5SWu Fengguang 503d46db3d5SWu Fengguang /* 504d46db3d5SWu Fengguang * WB_SYNC_ALL mode does livelock avoidance by syncing dirty 505d46db3d5SWu Fengguang * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX 506d46db3d5SWu Fengguang * here avoids calling into writeback_inodes_wb() more than once. 507d46db3d5SWu Fengguang * 508d46db3d5SWu Fengguang * The intended call sequence for WB_SYNC_ALL writeback is: 509d46db3d5SWu Fengguang * 510d46db3d5SWu Fengguang * wb_writeback() 511d46db3d5SWu Fengguang * writeback_sb_inodes() <== called only once 512d46db3d5SWu Fengguang * write_cache_pages() <== called once for each inode 513d46db3d5SWu Fengguang * (quickly) tag currently dirty pages 514d46db3d5SWu Fengguang * (maybe slowly) sync all tagged pages 515d46db3d5SWu Fengguang */ 516d46db3d5SWu Fengguang if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages) 517d46db3d5SWu Fengguang pages = LONG_MAX; 5181a12d8bdSWu Fengguang else { 5191a12d8bdSWu Fengguang pages = min(bdi->avg_write_bandwidth / 2, 5201a12d8bdSWu Fengguang global_dirty_limit / DIRTY_SCOPE); 5211a12d8bdSWu Fengguang pages = min(pages, work->nr_pages); 5221a12d8bdSWu Fengguang pages = round_down(pages + MIN_WRITEBACK_PAGES, 5231a12d8bdSWu Fengguang MIN_WRITEBACK_PAGES); 5241a12d8bdSWu Fengguang } 525d46db3d5SWu Fengguang 526d46db3d5SWu Fengguang return pages; 527d46db3d5SWu Fengguang } 528d46db3d5SWu Fengguang 52903ba3782SJens Axboe /* 530f11c9c5cSEdward Shishkin * Write a portion of b_io inodes which belong to @sb. 531edadfb10SChristoph Hellwig * 532edadfb10SChristoph Hellwig * If @only_this_sb is true, then find and write all such 533f11c9c5cSEdward Shishkin * inodes. Otherwise write only ones which go sequentially 534f11c9c5cSEdward Shishkin * in reverse order. 535edadfb10SChristoph Hellwig * 536d46db3d5SWu Fengguang * Return the number of pages and/or inodes written. 537f11c9c5cSEdward Shishkin */ 538d46db3d5SWu Fengguang static long writeback_sb_inodes(struct super_block *sb, 539d46db3d5SWu Fengguang struct bdi_writeback *wb, 540d46db3d5SWu Fengguang struct wb_writeback_work *work) 54103ba3782SJens Axboe { 542d46db3d5SWu Fengguang struct writeback_control wbc = { 543d46db3d5SWu Fengguang .sync_mode = work->sync_mode, 544d46db3d5SWu Fengguang .tagged_writepages = work->tagged_writepages, 545d46db3d5SWu Fengguang .for_kupdate = work->for_kupdate, 546d46db3d5SWu Fengguang .for_background = work->for_background, 547d46db3d5SWu Fengguang .range_cyclic = work->range_cyclic, 548d46db3d5SWu Fengguang .range_start = 0, 549d46db3d5SWu Fengguang .range_end = LLONG_MAX, 550d46db3d5SWu Fengguang }; 551d46db3d5SWu Fengguang unsigned long start_time = jiffies; 552d46db3d5SWu Fengguang long write_chunk; 553d46db3d5SWu Fengguang long wrote = 0; /* count both pages and inodes */ 554d46db3d5SWu Fengguang 55503ba3782SJens Axboe while (!list_empty(&wb->b_io)) { 5567ccf19a8SNick Piggin struct inode *inode = wb_inode(wb->b_io.prev); 557edadfb10SChristoph Hellwig 558edadfb10SChristoph Hellwig if (inode->i_sb != sb) { 559d46db3d5SWu Fengguang if (work->sb) { 560edadfb10SChristoph Hellwig /* 561edadfb10SChristoph Hellwig * We only want to write back data for this 562edadfb10SChristoph Hellwig * superblock, move all inodes not belonging 563edadfb10SChristoph Hellwig * to it back onto the dirty list. 564edadfb10SChristoph Hellwig */ 565f758eeabSChristoph Hellwig redirty_tail(inode, wb); 56666f3b8e2SJens Axboe continue; 56766f3b8e2SJens Axboe } 568edadfb10SChristoph Hellwig 569edadfb10SChristoph Hellwig /* 570edadfb10SChristoph Hellwig * The inode belongs to a different superblock. 571edadfb10SChristoph Hellwig * Bounce back to the caller to unpin this and 572edadfb10SChristoph Hellwig * pin the next superblock. 573edadfb10SChristoph Hellwig */ 574d46db3d5SWu Fengguang break; 575edadfb10SChristoph Hellwig } 576edadfb10SChristoph Hellwig 5779843b76aSChristoph Hellwig /* 5789843b76aSChristoph Hellwig * Don't bother with new inodes or inodes beeing freed, first 5799843b76aSChristoph Hellwig * kind does not need peridic writeout yet, and for the latter 5809843b76aSChristoph Hellwig * kind writeout is handled by the freer. 5819843b76aSChristoph Hellwig */ 582250df6edSDave Chinner spin_lock(&inode->i_lock); 5839843b76aSChristoph Hellwig if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { 584250df6edSDave Chinner spin_unlock(&inode->i_lock); 585fcc5c222SWu Fengguang redirty_tail(inode, wb); 5867ef0d737SNick Piggin continue; 5877ef0d737SNick Piggin } 5881da177e4SLinus Torvalds __iget(inode); 5891a12d8bdSWu Fengguang write_chunk = writeback_chunk_size(wb->bdi, work); 590d46db3d5SWu Fengguang wbc.nr_to_write = write_chunk; 591d46db3d5SWu Fengguang wbc.pages_skipped = 0; 592250df6edSDave Chinner 593d46db3d5SWu Fengguang writeback_single_inode(inode, wb, &wbc); 594d46db3d5SWu Fengguang 595d46db3d5SWu Fengguang work->nr_pages -= write_chunk - wbc.nr_to_write; 596d46db3d5SWu Fengguang wrote += write_chunk - wbc.nr_to_write; 597d46db3d5SWu Fengguang if (!(inode->i_state & I_DIRTY)) 598d46db3d5SWu Fengguang wrote++; 599d46db3d5SWu Fengguang if (wbc.pages_skipped) { 6001da177e4SLinus Torvalds /* 6011da177e4SLinus Torvalds * writeback is not making progress due to locked 6021da177e4SLinus Torvalds * buffers. Skip this inode for now. 6031da177e4SLinus Torvalds */ 604f758eeabSChristoph Hellwig redirty_tail(inode, wb); 6051da177e4SLinus Torvalds } 6060f1b1fd8SDave Chinner spin_unlock(&inode->i_lock); 607f758eeabSChristoph Hellwig spin_unlock(&wb->list_lock); 6081da177e4SLinus Torvalds iput(inode); 6094ffc8444SOGAWA Hirofumi cond_resched(); 610f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 611d46db3d5SWu Fengguang /* 612d46db3d5SWu Fengguang * bail out to wb_writeback() often enough to check 613d46db3d5SWu Fengguang * background threshold and other termination conditions. 614d46db3d5SWu Fengguang */ 615d46db3d5SWu Fengguang if (wrote) { 616d46db3d5SWu Fengguang if (time_is_before_jiffies(start_time + HZ / 10UL)) 617d46db3d5SWu Fengguang break; 618d46db3d5SWu Fengguang if (work->nr_pages <= 0) 619d46db3d5SWu Fengguang break; 6201da177e4SLinus Torvalds } 6218bc3be27SFengguang Wu } 622d46db3d5SWu Fengguang return wrote; 623f11c9c5cSEdward Shishkin } 62438f21977SNick Piggin 625d46db3d5SWu Fengguang static long __writeback_inodes_wb(struct bdi_writeback *wb, 626d46db3d5SWu Fengguang struct wb_writeback_work *work) 627f11c9c5cSEdward Shishkin { 628d46db3d5SWu Fengguang unsigned long start_time = jiffies; 629d46db3d5SWu Fengguang long wrote = 0; 630f11c9c5cSEdward Shishkin 631f11c9c5cSEdward Shishkin while (!list_empty(&wb->b_io)) { 6327ccf19a8SNick Piggin struct inode *inode = wb_inode(wb->b_io.prev); 633f11c9c5cSEdward Shishkin struct super_block *sb = inode->i_sb; 634f11c9c5cSEdward Shishkin 63512ad3ab6SDave Chinner if (!grab_super_passive(sb)) { 6360e995816SWu Fengguang /* 6370e995816SWu Fengguang * grab_super_passive() may fail consistently due to 6380e995816SWu Fengguang * s_umount being grabbed by someone else. Don't use 6390e995816SWu Fengguang * requeue_io() to avoid busy retrying the inode/sb. 6400e995816SWu Fengguang */ 6410e995816SWu Fengguang redirty_tail(inode, wb); 642d19de7edSChristoph Hellwig continue; 643334132aeSChristoph Hellwig } 644d46db3d5SWu Fengguang wrote += writeback_sb_inodes(sb, wb, work); 645d19de7edSChristoph Hellwig drop_super(sb); 646f11c9c5cSEdward Shishkin 647d46db3d5SWu Fengguang /* refer to the same tests at the end of writeback_sb_inodes */ 648d46db3d5SWu Fengguang if (wrote) { 649d46db3d5SWu Fengguang if (time_is_before_jiffies(start_time + HZ / 10UL)) 650d46db3d5SWu Fengguang break; 651d46db3d5SWu Fengguang if (work->nr_pages <= 0) 652f11c9c5cSEdward Shishkin break; 653f11c9c5cSEdward Shishkin } 654d46db3d5SWu Fengguang } 65566f3b8e2SJens Axboe /* Leave any unwritten inodes on b_io */ 656d46db3d5SWu Fengguang return wrote; 65766f3b8e2SJens Axboe } 65866f3b8e2SJens Axboe 6590e175a18SCurt Wohlgemuth long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages, 6600e175a18SCurt Wohlgemuth enum wb_reason reason) 661edadfb10SChristoph Hellwig { 662d46db3d5SWu Fengguang struct wb_writeback_work work = { 663d46db3d5SWu Fengguang .nr_pages = nr_pages, 664d46db3d5SWu Fengguang .sync_mode = WB_SYNC_NONE, 665d46db3d5SWu Fengguang .range_cyclic = 1, 6660e175a18SCurt Wohlgemuth .reason = reason, 667d46db3d5SWu Fengguang }; 668edadfb10SChristoph Hellwig 669f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 670424b351fSWu Fengguang if (list_empty(&wb->b_io)) 671ad4e38ddSCurt Wohlgemuth queue_io(wb, &work); 672d46db3d5SWu Fengguang __writeback_inodes_wb(wb, &work); 673f758eeabSChristoph Hellwig spin_unlock(&wb->list_lock); 674edadfb10SChristoph Hellwig 675d46db3d5SWu Fengguang return nr_pages - work.nr_pages; 67666f3b8e2SJens Axboe } 67766f3b8e2SJens Axboe 678b00949aaSWu Fengguang static bool over_bground_thresh(struct backing_dev_info *bdi) 67903ba3782SJens Axboe { 68003ba3782SJens Axboe unsigned long background_thresh, dirty_thresh; 68103ba3782SJens Axboe 68216c4042fSWu Fengguang global_dirty_limits(&background_thresh, &dirty_thresh); 68303ba3782SJens Axboe 684b00949aaSWu Fengguang if (global_page_state(NR_FILE_DIRTY) + 685b00949aaSWu Fengguang global_page_state(NR_UNSTABLE_NFS) > background_thresh) 686b00949aaSWu Fengguang return true; 687b00949aaSWu Fengguang 688b00949aaSWu Fengguang if (bdi_stat(bdi, BDI_RECLAIMABLE) > 689b00949aaSWu Fengguang bdi_dirty_limit(bdi, background_thresh)) 690b00949aaSWu Fengguang return true; 691b00949aaSWu Fengguang 692b00949aaSWu Fengguang return false; 69303ba3782SJens Axboe } 69403ba3782SJens Axboe 69503ba3782SJens Axboe /* 696e98be2d5SWu Fengguang * Called under wb->list_lock. If there are multiple wb per bdi, 697e98be2d5SWu Fengguang * only the flusher working on the first wb should do it. 698e98be2d5SWu Fengguang */ 699e98be2d5SWu Fengguang static void wb_update_bandwidth(struct bdi_writeback *wb, 700e98be2d5SWu Fengguang unsigned long start_time) 701e98be2d5SWu Fengguang { 702af6a3113SWu Fengguang __bdi_update_bandwidth(wb->bdi, 0, 0, 0, 0, 0, start_time); 703e98be2d5SWu Fengguang } 704e98be2d5SWu Fengguang 705e98be2d5SWu Fengguang /* 70603ba3782SJens Axboe * Explicit flushing or periodic writeback of "old" data. 70703ba3782SJens Axboe * 70803ba3782SJens Axboe * Define "old": the first time one of an inode's pages is dirtied, we mark the 70903ba3782SJens Axboe * dirtying-time in the inode's address_space. So this periodic writeback code 71003ba3782SJens Axboe * just walks the superblock inode list, writing back any inodes which are 71103ba3782SJens Axboe * older than a specific point in time. 71203ba3782SJens Axboe * 71303ba3782SJens Axboe * Try to run once per dirty_writeback_interval. But if a writeback event 71403ba3782SJens Axboe * takes longer than a dirty_writeback_interval interval, then leave a 71503ba3782SJens Axboe * one-second gap. 71603ba3782SJens Axboe * 71703ba3782SJens Axboe * older_than_this takes precedence over nr_to_write. So we'll only write back 71803ba3782SJens Axboe * all dirty pages if they are all attached to "old" mappings. 71903ba3782SJens Axboe */ 720c4a77a6cSJens Axboe static long wb_writeback(struct bdi_writeback *wb, 72183ba7b07SChristoph Hellwig struct wb_writeback_work *work) 72203ba3782SJens Axboe { 723e98be2d5SWu Fengguang unsigned long wb_start = jiffies; 724d46db3d5SWu Fengguang long nr_pages = work->nr_pages; 72503ba3782SJens Axboe unsigned long oldest_jif; 726a5989bdcSJan Kara struct inode *inode; 727d46db3d5SWu Fengguang long progress; 72803ba3782SJens Axboe 729e185dda8SWu Fengguang oldest_jif = jiffies; 730d46db3d5SWu Fengguang work->older_than_this = &oldest_jif; 73103ba3782SJens Axboe 732e8dfc305SWu Fengguang spin_lock(&wb->list_lock); 73303ba3782SJens Axboe for (;;) { 73403ba3782SJens Axboe /* 735d3ddec76SWu Fengguang * Stop writeback when nr_pages has been consumed 73603ba3782SJens Axboe */ 73783ba7b07SChristoph Hellwig if (work->nr_pages <= 0) 73803ba3782SJens Axboe break; 73903ba3782SJens Axboe 74003ba3782SJens Axboe /* 741aa373cf5SJan Kara * Background writeout and kupdate-style writeback may 742aa373cf5SJan Kara * run forever. Stop them if there is other work to do 743aa373cf5SJan Kara * so that e.g. sync can proceed. They'll be restarted 744aa373cf5SJan Kara * after the other works are all done. 745aa373cf5SJan Kara */ 746aa373cf5SJan Kara if ((work->for_background || work->for_kupdate) && 747aa373cf5SJan Kara !list_empty(&wb->bdi->work_list)) 748aa373cf5SJan Kara break; 749aa373cf5SJan Kara 750aa373cf5SJan Kara /* 751d3ddec76SWu Fengguang * For background writeout, stop when we are below the 752d3ddec76SWu Fengguang * background dirty threshold 75303ba3782SJens Axboe */ 754b00949aaSWu Fengguang if (work->for_background && !over_bground_thresh(wb->bdi)) 75503ba3782SJens Axboe break; 75603ba3782SJens Axboe 757ba9aa839SWu Fengguang if (work->for_kupdate) { 758ba9aa839SWu Fengguang oldest_jif = jiffies - 759ba9aa839SWu Fengguang msecs_to_jiffies(dirty_expire_interval * 10); 760d46db3d5SWu Fengguang work->older_than_this = &oldest_jif; 761ba9aa839SWu Fengguang } 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 95103ba3782SJens Axboe while (!kthread_should_stop()) { 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 } 98169b62d01SJens Axboe 98203ba3782SJens Axboe try_to_freeze(); 98303ba3782SJens Axboe } 98403ba3782SJens Axboe 985fff5b85aSArtem Bityutskiy /* Flush any work that raced with us exiting */ 98608243900SChristoph Hellwig if (!list_empty(&bdi->work_list)) 98708243900SChristoph Hellwig wb_do_writeback(wb, 1); 988455b2864SDave Chinner 989455b2864SDave Chinner trace_writeback_thread_stop(bdi); 99003ba3782SJens Axboe return 0; 99103ba3782SJens Axboe } 99203ba3782SJens Axboe 99308243900SChristoph Hellwig 99403ba3782SJens Axboe /* 99503ba3782SJens Axboe * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back 99603ba3782SJens Axboe * the whole world. 99703ba3782SJens Axboe */ 9980e175a18SCurt Wohlgemuth void wakeup_flusher_threads(long nr_pages, enum wb_reason reason) 99903ba3782SJens Axboe { 1000b8c2f347SChristoph Hellwig struct backing_dev_info *bdi; 1001b8c2f347SChristoph Hellwig 100283ba7b07SChristoph Hellwig if (!nr_pages) { 100383ba7b07SChristoph Hellwig nr_pages = global_page_state(NR_FILE_DIRTY) + 100403ba3782SJens Axboe global_page_state(NR_UNSTABLE_NFS); 1005b8c2f347SChristoph Hellwig } 1006b8c2f347SChristoph Hellwig 1007b8c2f347SChristoph Hellwig rcu_read_lock(); 1008b8c2f347SChristoph Hellwig list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { 1009b8c2f347SChristoph Hellwig if (!bdi_has_dirty_io(bdi)) 1010b8c2f347SChristoph Hellwig continue; 10110e175a18SCurt Wohlgemuth __bdi_start_writeback(bdi, nr_pages, false, reason); 1012b8c2f347SChristoph Hellwig } 1013b8c2f347SChristoph Hellwig rcu_read_unlock(); 101403ba3782SJens Axboe } 101503ba3782SJens Axboe 101603ba3782SJens Axboe static noinline void block_dump___mark_inode_dirty(struct inode *inode) 101703ba3782SJens Axboe { 101803ba3782SJens Axboe if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { 101903ba3782SJens Axboe struct dentry *dentry; 102003ba3782SJens Axboe const char *name = "?"; 102103ba3782SJens Axboe 102203ba3782SJens Axboe dentry = d_find_alias(inode); 102303ba3782SJens Axboe if (dentry) { 102403ba3782SJens Axboe spin_lock(&dentry->d_lock); 102503ba3782SJens Axboe name = (const char *) dentry->d_name.name; 102603ba3782SJens Axboe } 102703ba3782SJens Axboe printk(KERN_DEBUG 102803ba3782SJens Axboe "%s(%d): dirtied inode %lu (%s) on %s\n", 102903ba3782SJens Axboe current->comm, task_pid_nr(current), inode->i_ino, 103003ba3782SJens Axboe name, inode->i_sb->s_id); 103103ba3782SJens Axboe if (dentry) { 103203ba3782SJens Axboe spin_unlock(&dentry->d_lock); 103303ba3782SJens Axboe dput(dentry); 103403ba3782SJens Axboe } 103503ba3782SJens Axboe } 103603ba3782SJens Axboe } 103703ba3782SJens Axboe 103803ba3782SJens Axboe /** 103903ba3782SJens Axboe * __mark_inode_dirty - internal function 104003ba3782SJens Axboe * @inode: inode to mark 104103ba3782SJens Axboe * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) 104203ba3782SJens Axboe * Mark an inode as dirty. Callers should use mark_inode_dirty or 104303ba3782SJens Axboe * mark_inode_dirty_sync. 104403ba3782SJens Axboe * 104503ba3782SJens Axboe * Put the inode on the super block's dirty list. 104603ba3782SJens Axboe * 104703ba3782SJens Axboe * CAREFUL! We mark it dirty unconditionally, but move it onto the 104803ba3782SJens Axboe * dirty list only if it is hashed or if it refers to a blockdev. 104903ba3782SJens Axboe * If it was not hashed, it will never be added to the dirty list 105003ba3782SJens Axboe * even if it is later hashed, as it will have been marked dirty already. 105103ba3782SJens Axboe * 105203ba3782SJens Axboe * In short, make sure you hash any inodes _before_ you start marking 105303ba3782SJens Axboe * them dirty. 105403ba3782SJens Axboe * 105503ba3782SJens Axboe * Note that for blockdevs, inode->dirtied_when represents the dirtying time of 105603ba3782SJens Axboe * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of 105703ba3782SJens Axboe * the kernel-internal blockdev inode represents the dirtying time of the 105803ba3782SJens Axboe * blockdev's pages. This is why for I_DIRTY_PAGES we always use 105903ba3782SJens Axboe * page->mapping->host, so the page-dirtying time is recorded in the internal 106003ba3782SJens Axboe * blockdev inode. 106103ba3782SJens Axboe */ 106203ba3782SJens Axboe void __mark_inode_dirty(struct inode *inode, int flags) 106303ba3782SJens Axboe { 106403ba3782SJens Axboe struct super_block *sb = inode->i_sb; 1065253c34e9SArtem Bityutskiy struct backing_dev_info *bdi = NULL; 106603ba3782SJens Axboe 106703ba3782SJens Axboe /* 106803ba3782SJens Axboe * Don't do this for I_DIRTY_PAGES - that doesn't actually 106903ba3782SJens Axboe * dirty the inode itself 107003ba3782SJens Axboe */ 107103ba3782SJens Axboe if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { 107203ba3782SJens Axboe if (sb->s_op->dirty_inode) 1073aa385729SChristoph Hellwig sb->s_op->dirty_inode(inode, flags); 107403ba3782SJens Axboe } 107503ba3782SJens Axboe 107603ba3782SJens Axboe /* 107703ba3782SJens Axboe * make sure that changes are seen by all cpus before we test i_state 107803ba3782SJens Axboe * -- mikulas 107903ba3782SJens Axboe */ 108003ba3782SJens Axboe smp_mb(); 108103ba3782SJens Axboe 108203ba3782SJens Axboe /* avoid the locking if we can */ 108303ba3782SJens Axboe if ((inode->i_state & flags) == flags) 108403ba3782SJens Axboe return; 108503ba3782SJens Axboe 108603ba3782SJens Axboe if (unlikely(block_dump)) 108703ba3782SJens Axboe block_dump___mark_inode_dirty(inode); 108803ba3782SJens Axboe 1089250df6edSDave Chinner spin_lock(&inode->i_lock); 109003ba3782SJens Axboe if ((inode->i_state & flags) != flags) { 109103ba3782SJens Axboe const int was_dirty = inode->i_state & I_DIRTY; 109203ba3782SJens Axboe 109303ba3782SJens Axboe inode->i_state |= flags; 109403ba3782SJens Axboe 109503ba3782SJens Axboe /* 109603ba3782SJens Axboe * If the inode is being synced, just update its dirty state. 109703ba3782SJens Axboe * The unlocker will place the inode on the appropriate 109803ba3782SJens Axboe * superblock list, based upon its state. 109903ba3782SJens Axboe */ 110003ba3782SJens Axboe if (inode->i_state & I_SYNC) 1101250df6edSDave Chinner goto out_unlock_inode; 110203ba3782SJens Axboe 110303ba3782SJens Axboe /* 110403ba3782SJens Axboe * Only add valid (hashed) inodes to the superblock's 110503ba3782SJens Axboe * dirty list. Add blockdev inodes as well. 110603ba3782SJens Axboe */ 110703ba3782SJens Axboe if (!S_ISBLK(inode->i_mode)) { 11081d3382cbSAl Viro if (inode_unhashed(inode)) 1109250df6edSDave Chinner goto out_unlock_inode; 111003ba3782SJens Axboe } 1111a4ffdde6SAl Viro if (inode->i_state & I_FREEING) 1112250df6edSDave Chinner goto out_unlock_inode; 111303ba3782SJens Axboe 111403ba3782SJens Axboe /* 111503ba3782SJens Axboe * If the inode was already on b_dirty/b_io/b_more_io, don't 111603ba3782SJens Axboe * reposition it (that would break b_dirty time-ordering). 111703ba3782SJens Axboe */ 111803ba3782SJens Axboe if (!was_dirty) { 1119a66979abSDave Chinner bool wakeup_bdi = false; 1120253c34e9SArtem Bityutskiy bdi = inode_to_bdi(inode); 1121500b067cSJens Axboe 1122253c34e9SArtem Bityutskiy if (bdi_cap_writeback_dirty(bdi)) { 1123253c34e9SArtem Bityutskiy WARN(!test_bit(BDI_registered, &bdi->state), 1124253c34e9SArtem Bityutskiy "bdi-%s not registered\n", bdi->name); 1125253c34e9SArtem Bityutskiy 1126253c34e9SArtem Bityutskiy /* 1127253c34e9SArtem Bityutskiy * If this is the first dirty inode for this 1128253c34e9SArtem Bityutskiy * bdi, we have to wake-up the corresponding 1129253c34e9SArtem Bityutskiy * bdi thread to make sure background 1130253c34e9SArtem Bityutskiy * write-back happens later. 1131253c34e9SArtem Bityutskiy */ 1132253c34e9SArtem Bityutskiy if (!wb_has_dirty_io(&bdi->wb)) 1133253c34e9SArtem Bityutskiy wakeup_bdi = true; 1134500b067cSJens Axboe } 113503ba3782SJens Axboe 1136a66979abSDave Chinner spin_unlock(&inode->i_lock); 1137f758eeabSChristoph Hellwig spin_lock(&bdi->wb.list_lock); 113803ba3782SJens Axboe inode->dirtied_when = jiffies; 11397ccf19a8SNick Piggin list_move(&inode->i_wb_list, &bdi->wb.b_dirty); 1140f758eeabSChristoph Hellwig spin_unlock(&bdi->wb.list_lock); 1141253c34e9SArtem Bityutskiy 1142253c34e9SArtem Bityutskiy if (wakeup_bdi) 11436467716aSArtem Bityutskiy bdi_wakeup_thread_delayed(bdi); 1144a66979abSDave Chinner return; 1145a66979abSDave Chinner } 1146a66979abSDave Chinner } 1147a66979abSDave Chinner out_unlock_inode: 1148a66979abSDave Chinner spin_unlock(&inode->i_lock); 1149a66979abSDave Chinner 115003ba3782SJens Axboe } 115103ba3782SJens Axboe EXPORT_SYMBOL(__mark_inode_dirty); 115203ba3782SJens Axboe 115366f3b8e2SJens Axboe /* 115466f3b8e2SJens Axboe * Write out a superblock's list of dirty inodes. A wait will be performed 115566f3b8e2SJens Axboe * upon no inodes, all inodes or the final one, depending upon sync_mode. 115666f3b8e2SJens Axboe * 115766f3b8e2SJens Axboe * If older_than_this is non-NULL, then only write out inodes which 115866f3b8e2SJens Axboe * had their first dirtying at a time earlier than *older_than_this. 115966f3b8e2SJens Axboe * 116066f3b8e2SJens Axboe * If `bdi' is non-zero then we're being asked to writeback a specific queue. 116166f3b8e2SJens Axboe * This function assumes that the blockdev superblock's inodes are backed by 116266f3b8e2SJens Axboe * a variety of queues, so all inodes are searched. For other superblocks, 116366f3b8e2SJens Axboe * assume that all inodes are backed by the same queue. 116466f3b8e2SJens Axboe * 116566f3b8e2SJens Axboe * The inodes to be written are parked on bdi->b_io. They are moved back onto 116666f3b8e2SJens Axboe * bdi->b_dirty as they are selected for writing. This way, none can be missed 116766f3b8e2SJens Axboe * on the writer throttling path, and we get decent balancing between many 116866f3b8e2SJens Axboe * throttled threads: we don't want them all piling up on inode_sync_wait. 116966f3b8e2SJens Axboe */ 1170b6e51316SJens Axboe static void wait_sb_inodes(struct super_block *sb) 117166f3b8e2SJens Axboe { 117238f21977SNick Piggin struct inode *inode, *old_inode = NULL; 117338f21977SNick Piggin 117403ba3782SJens Axboe /* 117503ba3782SJens Axboe * We need to be protected against the filesystem going from 117603ba3782SJens Axboe * r/o to r/w or vice versa. 117703ba3782SJens Axboe */ 1178b6e51316SJens Axboe WARN_ON(!rwsem_is_locked(&sb->s_umount)); 117903ba3782SJens Axboe 118055fa6091SDave Chinner spin_lock(&inode_sb_list_lock); 118166f3b8e2SJens Axboe 118238f21977SNick Piggin /* 118338f21977SNick Piggin * Data integrity sync. Must wait for all pages under writeback, 118438f21977SNick Piggin * because there may have been pages dirtied before our sync 118538f21977SNick Piggin * call, but which had writeout started before we write it out. 118638f21977SNick Piggin * In which case, the inode may not be on the dirty list, but 118738f21977SNick Piggin * we still have to wait for that writeout. 118838f21977SNick Piggin */ 1189b6e51316SJens Axboe list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { 1190250df6edSDave Chinner struct address_space *mapping = inode->i_mapping; 119138f21977SNick Piggin 1192250df6edSDave Chinner spin_lock(&inode->i_lock); 1193250df6edSDave Chinner if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) || 1194250df6edSDave Chinner (mapping->nrpages == 0)) { 1195250df6edSDave Chinner spin_unlock(&inode->i_lock); 119638f21977SNick Piggin continue; 1197250df6edSDave Chinner } 119838f21977SNick Piggin __iget(inode); 1199250df6edSDave Chinner spin_unlock(&inode->i_lock); 120055fa6091SDave Chinner spin_unlock(&inode_sb_list_lock); 120155fa6091SDave Chinner 120238f21977SNick Piggin /* 120355fa6091SDave Chinner * We hold a reference to 'inode' so it couldn't have been 120455fa6091SDave Chinner * removed from s_inodes list while we dropped the 120555fa6091SDave Chinner * inode_sb_list_lock. We cannot iput the inode now as we can 120655fa6091SDave Chinner * be holding the last reference and we cannot iput it under 120755fa6091SDave Chinner * inode_sb_list_lock. So we keep the reference and iput it 120855fa6091SDave Chinner * later. 120938f21977SNick Piggin */ 121038f21977SNick Piggin iput(old_inode); 121138f21977SNick Piggin old_inode = inode; 121238f21977SNick Piggin 121338f21977SNick Piggin filemap_fdatawait(mapping); 121438f21977SNick Piggin 121538f21977SNick Piggin cond_resched(); 121638f21977SNick Piggin 121755fa6091SDave Chinner spin_lock(&inode_sb_list_lock); 121838f21977SNick Piggin } 121955fa6091SDave Chinner spin_unlock(&inode_sb_list_lock); 122038f21977SNick Piggin iput(old_inode); 122166f3b8e2SJens Axboe } 12221da177e4SLinus Torvalds 1223d8a8559cSJens Axboe /** 12243259f8beSChris Mason * writeback_inodes_sb_nr - writeback dirty inodes from given super_block 1225d8a8559cSJens Axboe * @sb: the superblock 12263259f8beSChris Mason * @nr: the number of pages to write 1227786228abSMarcos Paulo de Souza * @reason: reason why some writeback work initiated 12281da177e4SLinus Torvalds * 1229d8a8559cSJens Axboe * Start writeback on some inodes on this super_block. No guarantees are made 1230d8a8559cSJens Axboe * on how many (if any) will be written, and this function does not wait 12313259f8beSChris Mason * for IO completion of submitted IO. 12321da177e4SLinus Torvalds */ 12330e175a18SCurt Wohlgemuth void writeback_inodes_sb_nr(struct super_block *sb, 12340e175a18SCurt Wohlgemuth unsigned long nr, 12350e175a18SCurt Wohlgemuth enum wb_reason reason) 12361da177e4SLinus Torvalds { 123783ba7b07SChristoph Hellwig DECLARE_COMPLETION_ONSTACK(done); 123883ba7b07SChristoph Hellwig struct wb_writeback_work work = { 12393c4d7165SChristoph Hellwig .sb = sb, 12403c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_NONE, 12416e6938b6SWu Fengguang .tagged_writepages = 1, 124283ba7b07SChristoph Hellwig .done = &done, 12433259f8beSChris Mason .nr_pages = nr, 12440e175a18SCurt Wohlgemuth .reason = reason, 12453c4d7165SChristoph Hellwig }; 12460e3c9a22SJens Axboe 1247cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 124883ba7b07SChristoph Hellwig bdi_queue_work(sb->s_bdi, &work); 124983ba7b07SChristoph Hellwig wait_for_completion(&done); 12501da177e4SLinus Torvalds } 12513259f8beSChris Mason EXPORT_SYMBOL(writeback_inodes_sb_nr); 12523259f8beSChris Mason 12533259f8beSChris Mason /** 12543259f8beSChris Mason * writeback_inodes_sb - writeback dirty inodes from given super_block 12553259f8beSChris Mason * @sb: the superblock 1256786228abSMarcos Paulo de Souza * @reason: reason why some writeback work was initiated 12573259f8beSChris Mason * 12583259f8beSChris Mason * Start writeback on some inodes on this super_block. No guarantees are made 12593259f8beSChris Mason * on how many (if any) will be written, and this function does not wait 12603259f8beSChris Mason * for IO completion of submitted IO. 12613259f8beSChris Mason */ 12620e175a18SCurt Wohlgemuth void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) 12633259f8beSChris Mason { 12640e175a18SCurt Wohlgemuth return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); 12653259f8beSChris Mason } 1266d8a8559cSJens Axboe EXPORT_SYMBOL(writeback_inodes_sb); 1267d8a8559cSJens Axboe 1268d8a8559cSJens Axboe /** 126917bd55d0SEric Sandeen * writeback_inodes_sb_if_idle - start writeback if none underway 127017bd55d0SEric Sandeen * @sb: the superblock 1271786228abSMarcos Paulo de Souza * @reason: reason why some writeback work was initiated 127217bd55d0SEric Sandeen * 127317bd55d0SEric Sandeen * Invoke writeback_inodes_sb if no writeback is currently underway. 127417bd55d0SEric Sandeen * Returns 1 if writeback was started, 0 if not. 127517bd55d0SEric Sandeen */ 12760e175a18SCurt Wohlgemuth int writeback_inodes_sb_if_idle(struct super_block *sb, enum wb_reason reason) 127717bd55d0SEric Sandeen { 127817bd55d0SEric Sandeen if (!writeback_in_progress(sb->s_bdi)) { 1279cf37e972SChristoph Hellwig down_read(&sb->s_umount); 12800e175a18SCurt Wohlgemuth writeback_inodes_sb(sb, reason); 1281cf37e972SChristoph Hellwig up_read(&sb->s_umount); 128217bd55d0SEric Sandeen return 1; 128317bd55d0SEric Sandeen } else 128417bd55d0SEric Sandeen return 0; 128517bd55d0SEric Sandeen } 128617bd55d0SEric Sandeen EXPORT_SYMBOL(writeback_inodes_sb_if_idle); 128717bd55d0SEric Sandeen 128817bd55d0SEric Sandeen /** 12893259f8beSChris Mason * writeback_inodes_sb_if_idle - start writeback if none underway 12903259f8beSChris Mason * @sb: the superblock 12913259f8beSChris Mason * @nr: the number of pages to write 1292786228abSMarcos Paulo de Souza * @reason: reason why some writeback work was initiated 12933259f8beSChris Mason * 12943259f8beSChris Mason * Invoke writeback_inodes_sb if no writeback is currently underway. 12953259f8beSChris Mason * Returns 1 if writeback was started, 0 if not. 12963259f8beSChris Mason */ 12973259f8beSChris Mason int writeback_inodes_sb_nr_if_idle(struct super_block *sb, 12980e175a18SCurt Wohlgemuth unsigned long nr, 12990e175a18SCurt Wohlgemuth enum wb_reason reason) 13003259f8beSChris Mason { 13013259f8beSChris Mason if (!writeback_in_progress(sb->s_bdi)) { 13023259f8beSChris Mason down_read(&sb->s_umount); 13030e175a18SCurt Wohlgemuth writeback_inodes_sb_nr(sb, nr, reason); 13043259f8beSChris Mason up_read(&sb->s_umount); 13053259f8beSChris Mason return 1; 13063259f8beSChris Mason } else 13073259f8beSChris Mason return 0; 13083259f8beSChris Mason } 13093259f8beSChris Mason EXPORT_SYMBOL(writeback_inodes_sb_nr_if_idle); 13103259f8beSChris Mason 13113259f8beSChris Mason /** 1312d8a8559cSJens Axboe * sync_inodes_sb - sync sb inode pages 1313d8a8559cSJens Axboe * @sb: the superblock 1314d8a8559cSJens Axboe * 1315d8a8559cSJens Axboe * This function writes and waits on any dirty inode belonging to this 1316cb9ef8d5SStefan Hajnoczi * super_block. 1317d8a8559cSJens Axboe */ 1318b6e51316SJens Axboe void sync_inodes_sb(struct super_block *sb) 1319d8a8559cSJens Axboe { 132083ba7b07SChristoph Hellwig DECLARE_COMPLETION_ONSTACK(done); 132183ba7b07SChristoph Hellwig struct wb_writeback_work work = { 13223c4d7165SChristoph Hellwig .sb = sb, 13233c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_ALL, 13243c4d7165SChristoph Hellwig .nr_pages = LONG_MAX, 13253c4d7165SChristoph Hellwig .range_cyclic = 0, 132683ba7b07SChristoph Hellwig .done = &done, 13270e175a18SCurt Wohlgemuth .reason = WB_REASON_SYNC, 13283c4d7165SChristoph Hellwig }; 13293c4d7165SChristoph Hellwig 1330cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 1331cf37e972SChristoph Hellwig 133283ba7b07SChristoph Hellwig bdi_queue_work(sb->s_bdi, &work); 133383ba7b07SChristoph Hellwig wait_for_completion(&done); 133483ba7b07SChristoph Hellwig 1335b6e51316SJens Axboe wait_sb_inodes(sb); 1336d8a8559cSJens Axboe } 1337d8a8559cSJens Axboe EXPORT_SYMBOL(sync_inodes_sb); 13381da177e4SLinus Torvalds 13391da177e4SLinus Torvalds /** 13401da177e4SLinus Torvalds * write_inode_now - write an inode to disk 13411da177e4SLinus Torvalds * @inode: inode to write to disk 13421da177e4SLinus Torvalds * @sync: whether the write should be synchronous or not 13431da177e4SLinus Torvalds * 13447f04c26dSAndrea Arcangeli * This function commits an inode to disk immediately if it is dirty. This is 13457f04c26dSAndrea Arcangeli * primarily needed by knfsd. 13467f04c26dSAndrea Arcangeli * 13477f04c26dSAndrea Arcangeli * The caller must either have a ref on the inode or must have set I_WILL_FREE. 13481da177e4SLinus Torvalds */ 13491da177e4SLinus Torvalds int write_inode_now(struct inode *inode, int sync) 13501da177e4SLinus Torvalds { 1351f758eeabSChristoph Hellwig struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; 13521da177e4SLinus Torvalds int ret; 13531da177e4SLinus Torvalds struct writeback_control wbc = { 13541da177e4SLinus Torvalds .nr_to_write = LONG_MAX, 135518914b18SMike Galbraith .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, 1356111ebb6eSOGAWA Hirofumi .range_start = 0, 1357111ebb6eSOGAWA Hirofumi .range_end = LLONG_MAX, 13581da177e4SLinus Torvalds }; 13591da177e4SLinus Torvalds 13601da177e4SLinus Torvalds if (!mapping_cap_writeback_dirty(inode->i_mapping)) 136149364ce2SAndrew Morton wbc.nr_to_write = 0; 13621da177e4SLinus Torvalds 13631da177e4SLinus Torvalds might_sleep(); 1364f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 13650f1b1fd8SDave Chinner spin_lock(&inode->i_lock); 1366f758eeabSChristoph Hellwig ret = writeback_single_inode(inode, wb, &wbc); 13670f1b1fd8SDave Chinner spin_unlock(&inode->i_lock); 1368f758eeabSChristoph Hellwig spin_unlock(&wb->list_lock); 13691da177e4SLinus Torvalds if (sync) 13701c0eeaf5SJoern Engel inode_sync_wait(inode); 13711da177e4SLinus Torvalds return ret; 13721da177e4SLinus Torvalds } 13731da177e4SLinus Torvalds EXPORT_SYMBOL(write_inode_now); 13741da177e4SLinus Torvalds 13751da177e4SLinus Torvalds /** 13761da177e4SLinus Torvalds * sync_inode - write an inode and its pages to disk. 13771da177e4SLinus Torvalds * @inode: the inode to sync 13781da177e4SLinus Torvalds * @wbc: controls the writeback mode 13791da177e4SLinus Torvalds * 13801da177e4SLinus Torvalds * sync_inode() will write an inode and its pages to disk. It will also 13811da177e4SLinus Torvalds * correctly update the inode on its superblock's dirty inode lists and will 13821da177e4SLinus Torvalds * update inode->i_state. 13831da177e4SLinus Torvalds * 13841da177e4SLinus Torvalds * The caller must have a ref on the inode. 13851da177e4SLinus Torvalds */ 13861da177e4SLinus Torvalds int sync_inode(struct inode *inode, struct writeback_control *wbc) 13871da177e4SLinus Torvalds { 1388f758eeabSChristoph Hellwig struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; 13891da177e4SLinus Torvalds int ret; 13901da177e4SLinus Torvalds 1391f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 13920f1b1fd8SDave Chinner spin_lock(&inode->i_lock); 1393f758eeabSChristoph Hellwig ret = writeback_single_inode(inode, wb, wbc); 13940f1b1fd8SDave Chinner spin_unlock(&inode->i_lock); 1395f758eeabSChristoph Hellwig spin_unlock(&wb->list_lock); 13961da177e4SLinus Torvalds return ret; 13971da177e4SLinus Torvalds } 13981da177e4SLinus Torvalds EXPORT_SYMBOL(sync_inode); 1399c3765016SChristoph Hellwig 1400c3765016SChristoph Hellwig /** 1401c691b9d9SAndrew Morton * sync_inode_metadata - write an inode to disk 1402c3765016SChristoph Hellwig * @inode: the inode to sync 1403c3765016SChristoph Hellwig * @wait: wait for I/O to complete. 1404c3765016SChristoph Hellwig * 1405c691b9d9SAndrew Morton * Write an inode to disk and adjust its dirty state after completion. 1406c3765016SChristoph Hellwig * 1407c3765016SChristoph Hellwig * Note: only writes the actual inode, no associated data or other metadata. 1408c3765016SChristoph Hellwig */ 1409c3765016SChristoph Hellwig int sync_inode_metadata(struct inode *inode, int wait) 1410c3765016SChristoph Hellwig { 1411c3765016SChristoph Hellwig struct writeback_control wbc = { 1412c3765016SChristoph Hellwig .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE, 1413c3765016SChristoph Hellwig .nr_to_write = 0, /* metadata-only */ 1414c3765016SChristoph Hellwig }; 1415c3765016SChristoph Hellwig 1416c3765016SChristoph Hellwig return sync_inode(inode, &wbc); 1417c3765016SChristoph Hellwig } 1418c3765016SChristoph Hellwig EXPORT_SYMBOL(sync_inode_metadata); 1419