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 */ 50ac7b19a3STejun Heo unsigned int auto_free:1; /* free on completion */ 510e175a18SCurt Wohlgemuth enum wb_reason reason; /* why was writeback initiated? */ 52c4a77a6cSJens Axboe 538010c3b6SJens Axboe struct list_head list; /* pending work list */ 5483ba7b07SChristoph Hellwig struct completion *done; /* set if the caller waits */ 5503ba3782SJens Axboe }; 5603ba3782SJens Axboe 57a2f48706STheodore Ts'o /* 58a2f48706STheodore Ts'o * If an inode is constantly having its pages dirtied, but then the 59a2f48706STheodore Ts'o * updates stop dirtytime_expire_interval seconds in the past, it's 60a2f48706STheodore Ts'o * possible for the worst case time between when an inode has its 61a2f48706STheodore Ts'o * timestamps updated and when they finally get written out to be two 62a2f48706STheodore Ts'o * dirtytime_expire_intervals. We set the default to 12 hours (in 63a2f48706STheodore Ts'o * seconds), which means most of the time inodes will have their 64a2f48706STheodore Ts'o * timestamps written to disk after 12 hours, but in the worst case a 65a2f48706STheodore Ts'o * few inodes might not their timestamps updated for 24 hours. 66a2f48706STheodore Ts'o */ 67a2f48706STheodore Ts'o unsigned int dirtytime_expire_interval = 12 * 60 * 60; 68a2f48706STheodore Ts'o 697ccf19a8SNick Piggin static inline struct inode *wb_inode(struct list_head *head) 707ccf19a8SNick Piggin { 717ccf19a8SNick Piggin return list_entry(head, struct inode, i_wb_list); 727ccf19a8SNick Piggin } 737ccf19a8SNick Piggin 7415eb77a0SWu Fengguang /* 7515eb77a0SWu Fengguang * Include the creation of the trace points after defining the 7615eb77a0SWu Fengguang * wb_writeback_work structure and inline functions so that the definition 7715eb77a0SWu Fengguang * remains local to this file. 7815eb77a0SWu Fengguang */ 7915eb77a0SWu Fengguang #define CREATE_TRACE_POINTS 8015eb77a0SWu Fengguang #include <trace/events/writeback.h> 8115eb77a0SWu Fengguang 82774016b2SSteven Whitehouse EXPORT_TRACEPOINT_SYMBOL_GPL(wbc_writepage); 83774016b2SSteven Whitehouse 84d6c10f1fSTejun Heo static bool wb_io_lists_populated(struct bdi_writeback *wb) 85d6c10f1fSTejun Heo { 86d6c10f1fSTejun Heo if (wb_has_dirty_io(wb)) { 87d6c10f1fSTejun Heo return false; 88d6c10f1fSTejun Heo } else { 89d6c10f1fSTejun Heo set_bit(WB_has_dirty_io, &wb->state); 9095a46c65STejun Heo WARN_ON_ONCE(!wb->avg_write_bandwidth); 91766a9d6eSTejun Heo atomic_long_add(wb->avg_write_bandwidth, 92766a9d6eSTejun Heo &wb->bdi->tot_write_bandwidth); 93d6c10f1fSTejun Heo return true; 94d6c10f1fSTejun Heo } 95d6c10f1fSTejun Heo } 96d6c10f1fSTejun Heo 97d6c10f1fSTejun Heo static void wb_io_lists_depopulated(struct bdi_writeback *wb) 98d6c10f1fSTejun Heo { 99d6c10f1fSTejun Heo if (wb_has_dirty_io(wb) && list_empty(&wb->b_dirty) && 100766a9d6eSTejun Heo list_empty(&wb->b_io) && list_empty(&wb->b_more_io)) { 101d6c10f1fSTejun Heo clear_bit(WB_has_dirty_io, &wb->state); 10295a46c65STejun Heo WARN_ON_ONCE(atomic_long_sub_return(wb->avg_write_bandwidth, 10395a46c65STejun Heo &wb->bdi->tot_write_bandwidth) < 0); 104766a9d6eSTejun Heo } 105d6c10f1fSTejun Heo } 106d6c10f1fSTejun Heo 107d6c10f1fSTejun Heo /** 108d6c10f1fSTejun Heo * inode_wb_list_move_locked - move an inode onto a bdi_writeback IO list 109d6c10f1fSTejun Heo * @inode: inode to be moved 110d6c10f1fSTejun Heo * @wb: target bdi_writeback 111d6c10f1fSTejun Heo * @head: one of @wb->b_{dirty|io|more_io} 112d6c10f1fSTejun Heo * 113d6c10f1fSTejun Heo * Move @inode->i_wb_list to @list of @wb and set %WB_has_dirty_io. 114d6c10f1fSTejun Heo * Returns %true if @inode is the first occupant of the !dirty_time IO 115d6c10f1fSTejun Heo * lists; otherwise, %false. 116d6c10f1fSTejun Heo */ 117d6c10f1fSTejun Heo static bool inode_wb_list_move_locked(struct inode *inode, 118d6c10f1fSTejun Heo struct bdi_writeback *wb, 119d6c10f1fSTejun Heo struct list_head *head) 120d6c10f1fSTejun Heo { 121d6c10f1fSTejun Heo assert_spin_locked(&wb->list_lock); 122d6c10f1fSTejun Heo 123d6c10f1fSTejun Heo list_move(&inode->i_wb_list, head); 124d6c10f1fSTejun Heo 125d6c10f1fSTejun Heo /* dirty_time doesn't count as dirty_io until expiration */ 126d6c10f1fSTejun Heo if (head != &wb->b_dirty_time) 127d6c10f1fSTejun Heo return wb_io_lists_populated(wb); 128d6c10f1fSTejun Heo 129d6c10f1fSTejun Heo wb_io_lists_depopulated(wb); 130d6c10f1fSTejun Heo return false; 131d6c10f1fSTejun Heo } 132d6c10f1fSTejun Heo 133d6c10f1fSTejun Heo /** 134d6c10f1fSTejun Heo * inode_wb_list_del_locked - remove an inode from its bdi_writeback IO list 135d6c10f1fSTejun Heo * @inode: inode to be removed 136d6c10f1fSTejun Heo * @wb: bdi_writeback @inode is being removed from 137d6c10f1fSTejun Heo * 138d6c10f1fSTejun Heo * Remove @inode which may be on one of @wb->b_{dirty|io|more_io} lists and 139d6c10f1fSTejun Heo * clear %WB_has_dirty_io if all are empty afterwards. 140d6c10f1fSTejun Heo */ 141d6c10f1fSTejun Heo static void inode_wb_list_del_locked(struct inode *inode, 142d6c10f1fSTejun Heo struct bdi_writeback *wb) 143d6c10f1fSTejun Heo { 144d6c10f1fSTejun Heo assert_spin_locked(&wb->list_lock); 145d6c10f1fSTejun Heo 146d6c10f1fSTejun Heo list_del_init(&inode->i_wb_list); 147d6c10f1fSTejun Heo wb_io_lists_depopulated(wb); 148d6c10f1fSTejun Heo } 149d6c10f1fSTejun Heo 150f0054bb1STejun Heo static void wb_wakeup(struct bdi_writeback *wb) 1515acda9d1SJan Kara { 152f0054bb1STejun Heo spin_lock_bh(&wb->work_lock); 153f0054bb1STejun Heo if (test_bit(WB_registered, &wb->state)) 154f0054bb1STejun Heo mod_delayed_work(bdi_wq, &wb->dwork, 0); 155f0054bb1STejun Heo spin_unlock_bh(&wb->work_lock); 1565acda9d1SJan Kara } 1575acda9d1SJan Kara 158f0054bb1STejun Heo static void wb_queue_work(struct bdi_writeback *wb, 1596585027aSJan Kara struct wb_writeback_work *work) 1606585027aSJan Kara { 161f0054bb1STejun Heo trace_writeback_queue(wb->bdi, work); 1626585027aSJan Kara 163f0054bb1STejun Heo spin_lock_bh(&wb->work_lock); 164f0054bb1STejun Heo if (!test_bit(WB_registered, &wb->state)) { 1655acda9d1SJan Kara if (work->done) 1665acda9d1SJan Kara complete(work->done); 1675acda9d1SJan Kara goto out_unlock; 1685acda9d1SJan Kara } 169f0054bb1STejun Heo list_add_tail(&work->list, &wb->work_list); 170f0054bb1STejun Heo mod_delayed_work(bdi_wq, &wb->dwork, 0); 1715acda9d1SJan Kara out_unlock: 172f0054bb1STejun Heo spin_unlock_bh(&wb->work_lock); 17303ba3782SJens Axboe } 1741da177e4SLinus Torvalds 175703c2708STejun Heo #ifdef CONFIG_CGROUP_WRITEBACK 176703c2708STejun Heo 177703c2708STejun Heo /** 178703c2708STejun Heo * inode_congested - test whether an inode is congested 179703c2708STejun Heo * @inode: inode to test for congestion 180703c2708STejun Heo * @cong_bits: mask of WB_[a]sync_congested bits to test 181703c2708STejun Heo * 182703c2708STejun Heo * Tests whether @inode is congested. @cong_bits is the mask of congestion 183703c2708STejun Heo * bits to test and the return value is the mask of set bits. 184703c2708STejun Heo * 185703c2708STejun Heo * If cgroup writeback is enabled for @inode, the congestion state is 186703c2708STejun Heo * determined by whether the cgwb (cgroup bdi_writeback) for the blkcg 187703c2708STejun Heo * associated with @inode is congested; otherwise, the root wb's congestion 188703c2708STejun Heo * state is used. 189703c2708STejun Heo */ 190703c2708STejun Heo int inode_congested(struct inode *inode, int cong_bits) 191703c2708STejun Heo { 192703c2708STejun Heo if (inode) { 193703c2708STejun Heo struct bdi_writeback *wb = inode_to_wb(inode); 194703c2708STejun Heo if (wb) 195703c2708STejun Heo return wb_congested(wb, cong_bits); 196703c2708STejun Heo } 197703c2708STejun Heo 198703c2708STejun Heo return wb_congested(&inode_to_bdi(inode)->wb, cong_bits); 199703c2708STejun Heo } 200703c2708STejun Heo EXPORT_SYMBOL_GPL(inode_congested); 201703c2708STejun Heo 202f2b65121STejun Heo /** 203f2b65121STejun Heo * wb_split_bdi_pages - split nr_pages to write according to bandwidth 204f2b65121STejun Heo * @wb: target bdi_writeback to split @nr_pages to 205f2b65121STejun Heo * @nr_pages: number of pages to write for the whole bdi 206f2b65121STejun Heo * 207f2b65121STejun Heo * Split @wb's portion of @nr_pages according to @wb's write bandwidth in 208f2b65121STejun Heo * relation to the total write bandwidth of all wb's w/ dirty inodes on 209f2b65121STejun Heo * @wb->bdi. 210f2b65121STejun Heo */ 211f2b65121STejun Heo static long wb_split_bdi_pages(struct bdi_writeback *wb, long nr_pages) 212f2b65121STejun Heo { 213f2b65121STejun Heo unsigned long this_bw = wb->avg_write_bandwidth; 214f2b65121STejun Heo unsigned long tot_bw = atomic_long_read(&wb->bdi->tot_write_bandwidth); 215f2b65121STejun Heo 216f2b65121STejun Heo if (nr_pages == LONG_MAX) 217f2b65121STejun Heo return LONG_MAX; 218f2b65121STejun Heo 219f2b65121STejun Heo /* 220f2b65121STejun Heo * This may be called on clean wb's and proportional distribution 221f2b65121STejun Heo * may not make sense, just use the original @nr_pages in those 222f2b65121STejun Heo * cases. In general, we wanna err on the side of writing more. 223f2b65121STejun Heo */ 224f2b65121STejun Heo if (!tot_bw || this_bw >= tot_bw) 225f2b65121STejun Heo return nr_pages; 226f2b65121STejun Heo else 227f2b65121STejun Heo return DIV_ROUND_UP_ULL((u64)nr_pages * this_bw, tot_bw); 228f2b65121STejun Heo } 229f2b65121STejun Heo 230f2b65121STejun Heo #else /* CONFIG_CGROUP_WRITEBACK */ 231f2b65121STejun Heo 232f2b65121STejun Heo static long wb_split_bdi_pages(struct bdi_writeback *wb, long nr_pages) 233f2b65121STejun Heo { 234f2b65121STejun Heo return nr_pages; 235f2b65121STejun Heo } 236f2b65121STejun Heo 237703c2708STejun Heo #endif /* CONFIG_CGROUP_WRITEBACK */ 238703c2708STejun Heo 239c00ddad3STejun Heo void wb_start_writeback(struct bdi_writeback *wb, long nr_pages, 240c00ddad3STejun Heo bool range_cyclic, enum wb_reason reason) 241b6e51316SJens Axboe { 242c00ddad3STejun Heo struct wb_writeback_work *work; 243c00ddad3STejun Heo 244c00ddad3STejun Heo if (!wb_has_dirty_io(wb)) 245c00ddad3STejun Heo return; 246c00ddad3STejun Heo 247c00ddad3STejun Heo /* 248c00ddad3STejun Heo * This is WB_SYNC_NONE writeback, so if allocation fails just 249c00ddad3STejun Heo * wakeup the thread for old dirty data writeback 250c00ddad3STejun Heo */ 251c00ddad3STejun Heo work = kzalloc(sizeof(*work), GFP_ATOMIC); 252c00ddad3STejun Heo if (!work) { 253c00ddad3STejun Heo trace_writeback_nowork(wb->bdi); 254c00ddad3STejun Heo wb_wakeup(wb); 255c00ddad3STejun Heo return; 256c00ddad3STejun Heo } 257c00ddad3STejun Heo 258c00ddad3STejun Heo work->sync_mode = WB_SYNC_NONE; 259c00ddad3STejun Heo work->nr_pages = nr_pages; 260c00ddad3STejun Heo work->range_cyclic = range_cyclic; 261c00ddad3STejun Heo work->reason = reason; 262ac7b19a3STejun Heo work->auto_free = 1; 263c00ddad3STejun Heo 264c00ddad3STejun Heo wb_queue_work(wb, work); 265d3ddec76SWu Fengguang } 266d3ddec76SWu Fengguang 267c5444198SChristoph Hellwig /** 2689ecf4866STejun Heo * wb_start_background_writeback - start background writeback 2699ecf4866STejun Heo * @wb: bdi_writback to write from 270c5444198SChristoph Hellwig * 271c5444198SChristoph Hellwig * Description: 2726585027aSJan Kara * This makes sure WB_SYNC_NONE background writeback happens. When 2739ecf4866STejun Heo * this function returns, it is only guaranteed that for given wb 2746585027aSJan Kara * some IO is happening if we are over background dirty threshold. 2756585027aSJan Kara * Caller need not hold sb s_umount semaphore. 276c5444198SChristoph Hellwig */ 2779ecf4866STejun Heo void wb_start_background_writeback(struct bdi_writeback *wb) 278c5444198SChristoph Hellwig { 2796585027aSJan Kara /* 2806585027aSJan Kara * We just wake up the flusher thread. It will perform background 2816585027aSJan Kara * writeback as soon as there is no other work to do. 2826585027aSJan Kara */ 2839ecf4866STejun Heo trace_writeback_wake_background(wb->bdi); 2849ecf4866STejun Heo wb_wakeup(wb); 2851da177e4SLinus Torvalds } 2861da177e4SLinus Torvalds 2871da177e4SLinus Torvalds /* 288a66979abSDave Chinner * Remove the inode from the writeback list it is on. 289a66979abSDave Chinner */ 290a66979abSDave Chinner void inode_wb_list_del(struct inode *inode) 291a66979abSDave Chinner { 29252ebea74STejun Heo struct bdi_writeback *wb = inode_to_wb(inode); 293a66979abSDave Chinner 29452ebea74STejun Heo spin_lock(&wb->list_lock); 295d6c10f1fSTejun Heo inode_wb_list_del_locked(inode, wb); 29652ebea74STejun Heo spin_unlock(&wb->list_lock); 297f758eeabSChristoph Hellwig } 298a66979abSDave Chinner 299a66979abSDave Chinner /* 3006610a0bcSAndrew Morton * Redirty an inode: set its when-it-was dirtied timestamp and move it to the 3016610a0bcSAndrew Morton * furthest end of its superblock's dirty-inode list. 3026610a0bcSAndrew Morton * 3036610a0bcSAndrew Morton * Before stamping the inode's ->dirtied_when, we check to see whether it is 30466f3b8e2SJens Axboe * already the most-recently-dirtied inode on the b_dirty list. If that is 3056610a0bcSAndrew Morton * the case then the inode must have been redirtied while it was being written 3066610a0bcSAndrew Morton * out and we don't reset its dirtied_when. 3076610a0bcSAndrew Morton */ 308f758eeabSChristoph Hellwig static void redirty_tail(struct inode *inode, struct bdi_writeback *wb) 3096610a0bcSAndrew Morton { 31003ba3782SJens Axboe if (!list_empty(&wb->b_dirty)) { 31166f3b8e2SJens Axboe struct inode *tail; 3126610a0bcSAndrew Morton 3137ccf19a8SNick Piggin tail = wb_inode(wb->b_dirty.next); 31466f3b8e2SJens Axboe if (time_before(inode->dirtied_when, tail->dirtied_when)) 3156610a0bcSAndrew Morton inode->dirtied_when = jiffies; 3166610a0bcSAndrew Morton } 317d6c10f1fSTejun Heo inode_wb_list_move_locked(inode, wb, &wb->b_dirty); 3186610a0bcSAndrew Morton } 3196610a0bcSAndrew Morton 3206610a0bcSAndrew Morton /* 32166f3b8e2SJens Axboe * requeue inode for re-scanning after bdi->b_io list is exhausted. 322c986d1e2SAndrew Morton */ 323f758eeabSChristoph Hellwig static void requeue_io(struct inode *inode, struct bdi_writeback *wb) 324c986d1e2SAndrew Morton { 325d6c10f1fSTejun Heo inode_wb_list_move_locked(inode, wb, &wb->b_more_io); 326c986d1e2SAndrew Morton } 327c986d1e2SAndrew Morton 3281c0eeaf5SJoern Engel static void inode_sync_complete(struct inode *inode) 3291c0eeaf5SJoern Engel { 330365b94aeSJan Kara inode->i_state &= ~I_SYNC; 3314eff96ddSJan Kara /* If inode is clean an unused, put it into LRU now... */ 3324eff96ddSJan Kara inode_add_lru(inode); 333365b94aeSJan Kara /* Waiters must see I_SYNC cleared before being woken up */ 3341c0eeaf5SJoern Engel smp_mb(); 3351c0eeaf5SJoern Engel wake_up_bit(&inode->i_state, __I_SYNC); 3361c0eeaf5SJoern Engel } 3371c0eeaf5SJoern Engel 338d2caa3c5SJeff Layton static bool inode_dirtied_after(struct inode *inode, unsigned long t) 339d2caa3c5SJeff Layton { 340d2caa3c5SJeff Layton bool ret = time_after(inode->dirtied_when, t); 341d2caa3c5SJeff Layton #ifndef CONFIG_64BIT 342d2caa3c5SJeff Layton /* 343d2caa3c5SJeff Layton * For inodes being constantly redirtied, dirtied_when can get stuck. 344d2caa3c5SJeff Layton * It _appears_ to be in the future, but is actually in distant past. 345d2caa3c5SJeff Layton * This test is necessary to prevent such wrapped-around relative times 3465b0830cbSJens Axboe * from permanently stopping the whole bdi writeback. 347d2caa3c5SJeff Layton */ 348d2caa3c5SJeff Layton ret = ret && time_before_eq(inode->dirtied_when, jiffies); 349d2caa3c5SJeff Layton #endif 350d2caa3c5SJeff Layton return ret; 351d2caa3c5SJeff Layton } 352d2caa3c5SJeff Layton 3530ae45f63STheodore Ts'o #define EXPIRE_DIRTY_ATIME 0x0001 3540ae45f63STheodore Ts'o 355c986d1e2SAndrew Morton /* 3560e2f2b23SWang Sheng-Hui * Move expired (dirtied before work->older_than_this) dirty inodes from 357697e6fedSJan Kara * @delaying_queue to @dispatch_queue. 3582c136579SFengguang Wu */ 359e84d0a4fSWu Fengguang static int move_expired_inodes(struct list_head *delaying_queue, 3602c136579SFengguang Wu struct list_head *dispatch_queue, 3610ae45f63STheodore Ts'o int flags, 362ad4e38ddSCurt Wohlgemuth struct wb_writeback_work *work) 3632c136579SFengguang Wu { 3640ae45f63STheodore Ts'o unsigned long *older_than_this = NULL; 3650ae45f63STheodore Ts'o unsigned long expire_time; 3665c03449dSShaohua Li LIST_HEAD(tmp); 3675c03449dSShaohua Li struct list_head *pos, *node; 368cf137307SJens Axboe struct super_block *sb = NULL; 3695c03449dSShaohua Li struct inode *inode; 370cf137307SJens Axboe int do_sb_sort = 0; 371e84d0a4fSWu Fengguang int moved = 0; 3725c03449dSShaohua Li 3730ae45f63STheodore Ts'o if ((flags & EXPIRE_DIRTY_ATIME) == 0) 3740ae45f63STheodore Ts'o older_than_this = work->older_than_this; 375a2f48706STheodore Ts'o else if (!work->for_sync) { 376a2f48706STheodore Ts'o expire_time = jiffies - (dirtytime_expire_interval * HZ); 3770ae45f63STheodore Ts'o older_than_this = &expire_time; 3780ae45f63STheodore Ts'o } 3792c136579SFengguang Wu while (!list_empty(delaying_queue)) { 3807ccf19a8SNick Piggin inode = wb_inode(delaying_queue->prev); 3810ae45f63STheodore Ts'o if (older_than_this && 3820ae45f63STheodore Ts'o inode_dirtied_after(inode, *older_than_this)) 3832c136579SFengguang Wu break; 384a8855990SJan Kara list_move(&inode->i_wb_list, &tmp); 385a8855990SJan Kara moved++; 3860ae45f63STheodore Ts'o if (flags & EXPIRE_DIRTY_ATIME) 3870ae45f63STheodore Ts'o set_bit(__I_DIRTY_TIME_EXPIRED, &inode->i_state); 388a8855990SJan Kara if (sb_is_blkdev_sb(inode->i_sb)) 389a8855990SJan Kara continue; 390cf137307SJens Axboe if (sb && sb != inode->i_sb) 391cf137307SJens Axboe do_sb_sort = 1; 392cf137307SJens Axboe sb = inode->i_sb; 3935c03449dSShaohua Li } 3945c03449dSShaohua Li 395cf137307SJens Axboe /* just one sb in list, splice to dispatch_queue and we're done */ 396cf137307SJens Axboe if (!do_sb_sort) { 397cf137307SJens Axboe list_splice(&tmp, dispatch_queue); 398e84d0a4fSWu Fengguang goto out; 399cf137307SJens Axboe } 400cf137307SJens Axboe 4015c03449dSShaohua Li /* Move inodes from one superblock together */ 4025c03449dSShaohua Li while (!list_empty(&tmp)) { 4037ccf19a8SNick Piggin sb = wb_inode(tmp.prev)->i_sb; 4045c03449dSShaohua Li list_for_each_prev_safe(pos, node, &tmp) { 4057ccf19a8SNick Piggin inode = wb_inode(pos); 4065c03449dSShaohua Li if (inode->i_sb == sb) 4077ccf19a8SNick Piggin list_move(&inode->i_wb_list, dispatch_queue); 4082c136579SFengguang Wu } 4092c136579SFengguang Wu } 410e84d0a4fSWu Fengguang out: 411e84d0a4fSWu Fengguang return moved; 4125c03449dSShaohua Li } 4132c136579SFengguang Wu 4142c136579SFengguang Wu /* 4152c136579SFengguang Wu * Queue all expired dirty inodes for io, eldest first. 4164ea879b9SWu Fengguang * Before 4174ea879b9SWu Fengguang * newly dirtied b_dirty b_io b_more_io 4184ea879b9SWu Fengguang * =============> gf edc BA 4194ea879b9SWu Fengguang * After 4204ea879b9SWu Fengguang * newly dirtied b_dirty b_io b_more_io 4214ea879b9SWu Fengguang * =============> g fBAedc 4224ea879b9SWu Fengguang * | 4234ea879b9SWu Fengguang * +--> dequeue for IO 4242c136579SFengguang Wu */ 425ad4e38ddSCurt Wohlgemuth static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work) 4262c136579SFengguang Wu { 427e84d0a4fSWu Fengguang int moved; 4280ae45f63STheodore Ts'o 429f758eeabSChristoph Hellwig assert_spin_locked(&wb->list_lock); 4304ea879b9SWu Fengguang list_splice_init(&wb->b_more_io, &wb->b_io); 4310ae45f63STheodore Ts'o moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, 0, work); 4320ae45f63STheodore Ts'o moved += move_expired_inodes(&wb->b_dirty_time, &wb->b_io, 4330ae45f63STheodore Ts'o EXPIRE_DIRTY_ATIME, work); 434d6c10f1fSTejun Heo if (moved) 435d6c10f1fSTejun Heo wb_io_lists_populated(wb); 436ad4e38ddSCurt Wohlgemuth trace_writeback_queue_io(wb, work, moved); 43766f3b8e2SJens Axboe } 43866f3b8e2SJens Axboe 439a9185b41SChristoph Hellwig static int write_inode(struct inode *inode, struct writeback_control *wbc) 44066f3b8e2SJens Axboe { 4419fb0a7daSTejun Heo int ret; 4429fb0a7daSTejun Heo 4439fb0a7daSTejun Heo if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) { 4449fb0a7daSTejun Heo trace_writeback_write_inode_start(inode, wbc); 4459fb0a7daSTejun Heo ret = inode->i_sb->s_op->write_inode(inode, wbc); 4469fb0a7daSTejun Heo trace_writeback_write_inode(inode, wbc); 4479fb0a7daSTejun Heo return ret; 4489fb0a7daSTejun Heo } 44903ba3782SJens Axboe return 0; 45066f3b8e2SJens Axboe } 45108d8e974SFengguang Wu 4522c136579SFengguang Wu /* 453169ebd90SJan Kara * Wait for writeback on an inode to complete. Called with i_lock held. 454169ebd90SJan Kara * Caller must make sure inode cannot go away when we drop i_lock. 45501c03194SChristoph Hellwig */ 456169ebd90SJan Kara static void __inode_wait_for_writeback(struct inode *inode) 457169ebd90SJan Kara __releases(inode->i_lock) 458169ebd90SJan Kara __acquires(inode->i_lock) 45901c03194SChristoph Hellwig { 46001c03194SChristoph Hellwig DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); 46101c03194SChristoph Hellwig wait_queue_head_t *wqh; 46201c03194SChristoph Hellwig 46301c03194SChristoph Hellwig wqh = bit_waitqueue(&inode->i_state, __I_SYNC); 46458a9d3d8SRichard Kennedy while (inode->i_state & I_SYNC) { 465250df6edSDave Chinner spin_unlock(&inode->i_lock); 46674316201SNeilBrown __wait_on_bit(wqh, &wq, bit_wait, 46774316201SNeilBrown TASK_UNINTERRUPTIBLE); 468250df6edSDave Chinner spin_lock(&inode->i_lock); 46958a9d3d8SRichard Kennedy } 47001c03194SChristoph Hellwig } 47101c03194SChristoph Hellwig 47201c03194SChristoph Hellwig /* 473169ebd90SJan Kara * Wait for writeback on an inode to complete. Caller must have inode pinned. 474169ebd90SJan Kara */ 475169ebd90SJan Kara void inode_wait_for_writeback(struct inode *inode) 476169ebd90SJan Kara { 477169ebd90SJan Kara spin_lock(&inode->i_lock); 478169ebd90SJan Kara __inode_wait_for_writeback(inode); 479169ebd90SJan Kara spin_unlock(&inode->i_lock); 480169ebd90SJan Kara } 481169ebd90SJan Kara 482169ebd90SJan Kara /* 483169ebd90SJan Kara * Sleep until I_SYNC is cleared. This function must be called with i_lock 484169ebd90SJan Kara * held and drops it. It is aimed for callers not holding any inode reference 485169ebd90SJan Kara * so once i_lock is dropped, inode can go away. 486169ebd90SJan Kara */ 487169ebd90SJan Kara static void inode_sleep_on_writeback(struct inode *inode) 488169ebd90SJan Kara __releases(inode->i_lock) 489169ebd90SJan Kara { 490169ebd90SJan Kara DEFINE_WAIT(wait); 491169ebd90SJan Kara wait_queue_head_t *wqh = bit_waitqueue(&inode->i_state, __I_SYNC); 492169ebd90SJan Kara int sleep; 493169ebd90SJan Kara 494169ebd90SJan Kara prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); 495169ebd90SJan Kara sleep = inode->i_state & I_SYNC; 496169ebd90SJan Kara spin_unlock(&inode->i_lock); 497169ebd90SJan Kara if (sleep) 498169ebd90SJan Kara schedule(); 499169ebd90SJan Kara finish_wait(wqh, &wait); 500169ebd90SJan Kara } 501169ebd90SJan Kara 502169ebd90SJan Kara /* 503ccb26b5aSJan Kara * Find proper writeback list for the inode depending on its current state and 504ccb26b5aSJan Kara * possibly also change of its state while we were doing writeback. Here we 505ccb26b5aSJan Kara * handle things such as livelock prevention or fairness of writeback among 506ccb26b5aSJan Kara * inodes. This function can be called only by flusher thread - noone else 507ccb26b5aSJan Kara * processes all inodes in writeback lists and requeueing inodes behind flusher 508ccb26b5aSJan Kara * thread's back can have unexpected consequences. 509ccb26b5aSJan Kara */ 510ccb26b5aSJan Kara static void requeue_inode(struct inode *inode, struct bdi_writeback *wb, 511ccb26b5aSJan Kara struct writeback_control *wbc) 512ccb26b5aSJan Kara { 513ccb26b5aSJan Kara if (inode->i_state & I_FREEING) 514ccb26b5aSJan Kara return; 515ccb26b5aSJan Kara 516ccb26b5aSJan Kara /* 517ccb26b5aSJan Kara * Sync livelock prevention. Each inode is tagged and synced in one 518ccb26b5aSJan Kara * shot. If still dirty, it will be redirty_tail()'ed below. Update 519ccb26b5aSJan Kara * the dirty time to prevent enqueue and sync it again. 520ccb26b5aSJan Kara */ 521ccb26b5aSJan Kara if ((inode->i_state & I_DIRTY) && 522ccb26b5aSJan Kara (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)) 523ccb26b5aSJan Kara inode->dirtied_when = jiffies; 524ccb26b5aSJan Kara 5254f8ad655SJan Kara if (wbc->pages_skipped) { 5264f8ad655SJan Kara /* 5274f8ad655SJan Kara * writeback is not making progress due to locked 5284f8ad655SJan Kara * buffers. Skip this inode for now. 5294f8ad655SJan Kara */ 5304f8ad655SJan Kara redirty_tail(inode, wb); 5314f8ad655SJan Kara return; 5324f8ad655SJan Kara } 5334f8ad655SJan Kara 534ccb26b5aSJan Kara if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) { 535ccb26b5aSJan Kara /* 536ccb26b5aSJan Kara * We didn't write back all the pages. nfs_writepages() 537ccb26b5aSJan Kara * sometimes bales out without doing anything. 538ccb26b5aSJan Kara */ 539ccb26b5aSJan Kara if (wbc->nr_to_write <= 0) { 540ccb26b5aSJan Kara /* Slice used up. Queue for next turn. */ 541ccb26b5aSJan Kara requeue_io(inode, wb); 542ccb26b5aSJan Kara } else { 543ccb26b5aSJan Kara /* 544ccb26b5aSJan Kara * Writeback blocked by something other than 545ccb26b5aSJan Kara * congestion. Delay the inode for some time to 546ccb26b5aSJan Kara * avoid spinning on the CPU (100% iowait) 547ccb26b5aSJan Kara * retrying writeback of the dirty page/inode 548ccb26b5aSJan Kara * that cannot be performed immediately. 549ccb26b5aSJan Kara */ 550ccb26b5aSJan Kara redirty_tail(inode, wb); 551ccb26b5aSJan Kara } 552ccb26b5aSJan Kara } else if (inode->i_state & I_DIRTY) { 553ccb26b5aSJan Kara /* 554ccb26b5aSJan Kara * Filesystems can dirty the inode during writeback operations, 555ccb26b5aSJan Kara * such as delayed allocation during submission or metadata 556ccb26b5aSJan Kara * updates after data IO completion. 557ccb26b5aSJan Kara */ 558ccb26b5aSJan Kara redirty_tail(inode, wb); 5590ae45f63STheodore Ts'o } else if (inode->i_state & I_DIRTY_TIME) { 560a2f48706STheodore Ts'o inode->dirtied_when = jiffies; 561d6c10f1fSTejun Heo inode_wb_list_move_locked(inode, wb, &wb->b_dirty_time); 562ccb26b5aSJan Kara } else { 563ccb26b5aSJan Kara /* The inode is clean. Remove from writeback lists. */ 564d6c10f1fSTejun Heo inode_wb_list_del_locked(inode, wb); 565ccb26b5aSJan Kara } 566ccb26b5aSJan Kara } 567ccb26b5aSJan Kara 568ccb26b5aSJan Kara /* 5694f8ad655SJan Kara * Write out an inode and its dirty pages. Do not update the writeback list 5704f8ad655SJan Kara * linkage. That is left to the caller. The caller is also responsible for 5714f8ad655SJan Kara * setting I_SYNC flag and calling inode_sync_complete() to clear it. 5721da177e4SLinus Torvalds */ 5731da177e4SLinus Torvalds static int 574cd8ed2a4SYan Hong __writeback_single_inode(struct inode *inode, struct writeback_control *wbc) 5751da177e4SLinus Torvalds { 5761da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 577251d6a47SWu Fengguang long nr_to_write = wbc->nr_to_write; 57801c03194SChristoph Hellwig unsigned dirty; 5791da177e4SLinus Torvalds int ret; 5801da177e4SLinus Torvalds 5814f8ad655SJan Kara WARN_ON(!(inode->i_state & I_SYNC)); 5821da177e4SLinus Torvalds 5839fb0a7daSTejun Heo trace_writeback_single_inode_start(inode, wbc, nr_to_write); 5849fb0a7daSTejun Heo 5851da177e4SLinus Torvalds ret = do_writepages(mapping, wbc); 5861da177e4SLinus Torvalds 58726821ed4SChristoph Hellwig /* 58826821ed4SChristoph Hellwig * Make sure to wait on the data before writing out the metadata. 58926821ed4SChristoph Hellwig * This is important for filesystems that modify metadata on data 5907747bd4bSDave Chinner * I/O completion. We don't do it for sync(2) writeback because it has a 5917747bd4bSDave Chinner * separate, external IO completion path and ->sync_fs for guaranteeing 5927747bd4bSDave Chinner * inode metadata is written back correctly. 59326821ed4SChristoph Hellwig */ 5947747bd4bSDave Chinner if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) { 59526821ed4SChristoph Hellwig int err = filemap_fdatawait(mapping); 5961da177e4SLinus Torvalds if (ret == 0) 5971da177e4SLinus Torvalds ret = err; 5981da177e4SLinus Torvalds } 5991da177e4SLinus Torvalds 6005547e8aaSDmitry Monakhov /* 6015547e8aaSDmitry Monakhov * Some filesystems may redirty the inode during the writeback 6025547e8aaSDmitry Monakhov * due to delalloc, clear dirty metadata flags right before 6035547e8aaSDmitry Monakhov * write_inode() 6045547e8aaSDmitry Monakhov */ 605250df6edSDave Chinner spin_lock(&inode->i_lock); 6069c6ac78eSTejun Heo 6075547e8aaSDmitry Monakhov dirty = inode->i_state & I_DIRTY; 608a2f48706STheodore Ts'o if (inode->i_state & I_DIRTY_TIME) { 609a2f48706STheodore Ts'o if ((dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) || 610a2f48706STheodore Ts'o unlikely(inode->i_state & I_DIRTY_TIME_EXPIRED) || 611a2f48706STheodore Ts'o unlikely(time_after(jiffies, 612a2f48706STheodore Ts'o (inode->dirtied_time_when + 613a2f48706STheodore Ts'o dirtytime_expire_interval * HZ)))) { 6140ae45f63STheodore Ts'o dirty |= I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED; 6150ae45f63STheodore Ts'o trace_writeback_lazytime(inode); 6160ae45f63STheodore Ts'o } 617a2f48706STheodore Ts'o } else 618a2f48706STheodore Ts'o inode->i_state &= ~I_DIRTY_TIME_EXPIRED; 6190ae45f63STheodore Ts'o inode->i_state &= ~dirty; 6209c6ac78eSTejun Heo 6219c6ac78eSTejun Heo /* 6229c6ac78eSTejun Heo * Paired with smp_mb() in __mark_inode_dirty(). This allows 6239c6ac78eSTejun Heo * __mark_inode_dirty() to test i_state without grabbing i_lock - 6249c6ac78eSTejun Heo * either they see the I_DIRTY bits cleared or we see the dirtied 6259c6ac78eSTejun Heo * inode. 6269c6ac78eSTejun Heo * 6279c6ac78eSTejun Heo * I_DIRTY_PAGES is always cleared together above even if @mapping 6289c6ac78eSTejun Heo * still has dirty pages. The flag is reinstated after smp_mb() if 6299c6ac78eSTejun Heo * necessary. This guarantees that either __mark_inode_dirty() 6309c6ac78eSTejun Heo * sees clear I_DIRTY_PAGES or we see PAGECACHE_TAG_DIRTY. 6319c6ac78eSTejun Heo */ 6329c6ac78eSTejun Heo smp_mb(); 6339c6ac78eSTejun Heo 6349c6ac78eSTejun Heo if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) 6359c6ac78eSTejun Heo inode->i_state |= I_DIRTY_PAGES; 6369c6ac78eSTejun Heo 637250df6edSDave Chinner spin_unlock(&inode->i_lock); 6389c6ac78eSTejun Heo 6390ae45f63STheodore Ts'o if (dirty & I_DIRTY_TIME) 6400ae45f63STheodore Ts'o mark_inode_dirty_sync(inode); 64126821ed4SChristoph Hellwig /* Don't write the inode if only I_DIRTY_PAGES was set */ 6420ae45f63STheodore Ts'o if (dirty & ~I_DIRTY_PAGES) { 643a9185b41SChristoph Hellwig int err = write_inode(inode, wbc); 6441da177e4SLinus Torvalds if (ret == 0) 6451da177e4SLinus Torvalds ret = err; 6461da177e4SLinus Torvalds } 6474f8ad655SJan Kara trace_writeback_single_inode(inode, wbc, nr_to_write); 6484f8ad655SJan Kara return ret; 6494f8ad655SJan Kara } 6504f8ad655SJan Kara 6514f8ad655SJan Kara /* 6524f8ad655SJan Kara * Write out an inode's dirty pages. Either the caller has an active reference 6534f8ad655SJan Kara * on the inode or the inode has I_WILL_FREE set. 6544f8ad655SJan Kara * 6554f8ad655SJan Kara * This function is designed to be called for writing back one inode which 6564f8ad655SJan Kara * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode() 6574f8ad655SJan Kara * and does more profound writeback list handling in writeback_sb_inodes(). 6584f8ad655SJan Kara */ 6594f8ad655SJan Kara static int 6604f8ad655SJan Kara writeback_single_inode(struct inode *inode, struct bdi_writeback *wb, 6614f8ad655SJan Kara struct writeback_control *wbc) 6624f8ad655SJan Kara { 6634f8ad655SJan Kara int ret = 0; 6644f8ad655SJan Kara 6654f8ad655SJan Kara spin_lock(&inode->i_lock); 6664f8ad655SJan Kara if (!atomic_read(&inode->i_count)) 6674f8ad655SJan Kara WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); 6684f8ad655SJan Kara else 6694f8ad655SJan Kara WARN_ON(inode->i_state & I_WILL_FREE); 6704f8ad655SJan Kara 6714f8ad655SJan Kara if (inode->i_state & I_SYNC) { 6724f8ad655SJan Kara if (wbc->sync_mode != WB_SYNC_ALL) 6734f8ad655SJan Kara goto out; 6744f8ad655SJan Kara /* 675169ebd90SJan Kara * It's a data-integrity sync. We must wait. Since callers hold 676169ebd90SJan Kara * inode reference or inode has I_WILL_FREE set, it cannot go 677169ebd90SJan Kara * away under us. 6784f8ad655SJan Kara */ 679169ebd90SJan Kara __inode_wait_for_writeback(inode); 6804f8ad655SJan Kara } 6814f8ad655SJan Kara WARN_ON(inode->i_state & I_SYNC); 6824f8ad655SJan Kara /* 683f9b0e058SJan Kara * Skip inode if it is clean and we have no outstanding writeback in 684f9b0e058SJan Kara * WB_SYNC_ALL mode. We don't want to mess with writeback lists in this 685f9b0e058SJan Kara * function since flusher thread may be doing for example sync in 686f9b0e058SJan Kara * parallel and if we move the inode, it could get skipped. So here we 687f9b0e058SJan Kara * make sure inode is on some writeback list and leave it there unless 688f9b0e058SJan Kara * we have completely cleaned the inode. 6894f8ad655SJan Kara */ 6900ae45f63STheodore Ts'o if (!(inode->i_state & I_DIRTY_ALL) && 691f9b0e058SJan Kara (wbc->sync_mode != WB_SYNC_ALL || 692f9b0e058SJan Kara !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK))) 6934f8ad655SJan Kara goto out; 6944f8ad655SJan Kara inode->i_state |= I_SYNC; 6954f8ad655SJan Kara spin_unlock(&inode->i_lock); 6964f8ad655SJan Kara 697cd8ed2a4SYan Hong ret = __writeback_single_inode(inode, wbc); 6981da177e4SLinus Torvalds 699f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 700250df6edSDave Chinner spin_lock(&inode->i_lock); 7014f8ad655SJan Kara /* 7024f8ad655SJan Kara * If inode is clean, remove it from writeback lists. Otherwise don't 7034f8ad655SJan Kara * touch it. See comment above for explanation. 7044f8ad655SJan Kara */ 7050ae45f63STheodore Ts'o if (!(inode->i_state & I_DIRTY_ALL)) 706d6c10f1fSTejun Heo inode_wb_list_del_locked(inode, wb); 7074f8ad655SJan Kara spin_unlock(&wb->list_lock); 7081c0eeaf5SJoern Engel inode_sync_complete(inode); 7094f8ad655SJan Kara out: 7104f8ad655SJan Kara spin_unlock(&inode->i_lock); 7111da177e4SLinus Torvalds return ret; 7121da177e4SLinus Torvalds } 7131da177e4SLinus Torvalds 714a88a341aSTejun Heo static long writeback_chunk_size(struct bdi_writeback *wb, 7151a12d8bdSWu Fengguang struct wb_writeback_work *work) 716d46db3d5SWu Fengguang { 717d46db3d5SWu Fengguang long pages; 718d46db3d5SWu Fengguang 719d46db3d5SWu Fengguang /* 720d46db3d5SWu Fengguang * WB_SYNC_ALL mode does livelock avoidance by syncing dirty 721d46db3d5SWu Fengguang * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX 722d46db3d5SWu Fengguang * here avoids calling into writeback_inodes_wb() more than once. 723d46db3d5SWu Fengguang * 724d46db3d5SWu Fengguang * The intended call sequence for WB_SYNC_ALL writeback is: 725d46db3d5SWu Fengguang * 726d46db3d5SWu Fengguang * wb_writeback() 727d46db3d5SWu Fengguang * writeback_sb_inodes() <== called only once 728d46db3d5SWu Fengguang * write_cache_pages() <== called once for each inode 729d46db3d5SWu Fengguang * (quickly) tag currently dirty pages 730d46db3d5SWu Fengguang * (maybe slowly) sync all tagged pages 731d46db3d5SWu Fengguang */ 732d46db3d5SWu Fengguang if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages) 733d46db3d5SWu Fengguang pages = LONG_MAX; 7341a12d8bdSWu Fengguang else { 735a88a341aSTejun Heo pages = min(wb->avg_write_bandwidth / 2, 7361a12d8bdSWu Fengguang global_dirty_limit / DIRTY_SCOPE); 7371a12d8bdSWu Fengguang pages = min(pages, work->nr_pages); 7381a12d8bdSWu Fengguang pages = round_down(pages + MIN_WRITEBACK_PAGES, 7391a12d8bdSWu Fengguang MIN_WRITEBACK_PAGES); 7401a12d8bdSWu Fengguang } 741d46db3d5SWu Fengguang 742d46db3d5SWu Fengguang return pages; 743d46db3d5SWu Fengguang } 744d46db3d5SWu Fengguang 74503ba3782SJens Axboe /* 746f11c9c5cSEdward Shishkin * Write a portion of b_io inodes which belong to @sb. 747edadfb10SChristoph Hellwig * 748d46db3d5SWu Fengguang * Return the number of pages and/or inodes written. 749f11c9c5cSEdward Shishkin */ 750d46db3d5SWu Fengguang static long writeback_sb_inodes(struct super_block *sb, 751d46db3d5SWu Fengguang struct bdi_writeback *wb, 752d46db3d5SWu Fengguang struct wb_writeback_work *work) 75303ba3782SJens Axboe { 754d46db3d5SWu Fengguang struct writeback_control wbc = { 755d46db3d5SWu Fengguang .sync_mode = work->sync_mode, 756d46db3d5SWu Fengguang .tagged_writepages = work->tagged_writepages, 757d46db3d5SWu Fengguang .for_kupdate = work->for_kupdate, 758d46db3d5SWu Fengguang .for_background = work->for_background, 7597747bd4bSDave Chinner .for_sync = work->for_sync, 760d46db3d5SWu Fengguang .range_cyclic = work->range_cyclic, 761d46db3d5SWu Fengguang .range_start = 0, 762d46db3d5SWu Fengguang .range_end = LLONG_MAX, 763d46db3d5SWu Fengguang }; 764d46db3d5SWu Fengguang unsigned long start_time = jiffies; 765d46db3d5SWu Fengguang long write_chunk; 766d46db3d5SWu Fengguang long wrote = 0; /* count both pages and inodes */ 767d46db3d5SWu Fengguang 76803ba3782SJens Axboe while (!list_empty(&wb->b_io)) { 7697ccf19a8SNick Piggin struct inode *inode = wb_inode(wb->b_io.prev); 770edadfb10SChristoph Hellwig 771edadfb10SChristoph Hellwig if (inode->i_sb != sb) { 772d46db3d5SWu Fengguang if (work->sb) { 773edadfb10SChristoph Hellwig /* 774edadfb10SChristoph Hellwig * We only want to write back data for this 775edadfb10SChristoph Hellwig * superblock, move all inodes not belonging 776edadfb10SChristoph Hellwig * to it back onto the dirty list. 777edadfb10SChristoph Hellwig */ 778f758eeabSChristoph Hellwig redirty_tail(inode, wb); 77966f3b8e2SJens Axboe continue; 78066f3b8e2SJens Axboe } 781edadfb10SChristoph Hellwig 782edadfb10SChristoph Hellwig /* 783edadfb10SChristoph Hellwig * The inode belongs to a different superblock. 784edadfb10SChristoph Hellwig * Bounce back to the caller to unpin this and 785edadfb10SChristoph Hellwig * pin the next superblock. 786edadfb10SChristoph Hellwig */ 787d46db3d5SWu Fengguang break; 788edadfb10SChristoph Hellwig } 789edadfb10SChristoph Hellwig 7909843b76aSChristoph Hellwig /* 791331cbdeeSWanpeng Li * Don't bother with new inodes or inodes being freed, first 792331cbdeeSWanpeng Li * kind does not need periodic writeout yet, and for the latter 7939843b76aSChristoph Hellwig * kind writeout is handled by the freer. 7949843b76aSChristoph Hellwig */ 795250df6edSDave Chinner spin_lock(&inode->i_lock); 7969843b76aSChristoph Hellwig if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { 797250df6edSDave Chinner spin_unlock(&inode->i_lock); 798fcc5c222SWu Fengguang redirty_tail(inode, wb); 7997ef0d737SNick Piggin continue; 8007ef0d737SNick Piggin } 801cc1676d9SJan Kara if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) { 802cc1676d9SJan Kara /* 803cc1676d9SJan Kara * If this inode is locked for writeback and we are not 804cc1676d9SJan Kara * doing writeback-for-data-integrity, move it to 805cc1676d9SJan Kara * b_more_io so that writeback can proceed with the 806cc1676d9SJan Kara * other inodes on s_io. 807cc1676d9SJan Kara * 808cc1676d9SJan Kara * We'll have another go at writing back this inode 809cc1676d9SJan Kara * when we completed a full scan of b_io. 810cc1676d9SJan Kara */ 811cc1676d9SJan Kara spin_unlock(&inode->i_lock); 812cc1676d9SJan Kara requeue_io(inode, wb); 813cc1676d9SJan Kara trace_writeback_sb_inodes_requeue(inode); 814cc1676d9SJan Kara continue; 815cc1676d9SJan Kara } 816f0d07b7fSJan Kara spin_unlock(&wb->list_lock); 817f0d07b7fSJan Kara 8184f8ad655SJan Kara /* 8194f8ad655SJan Kara * We already requeued the inode if it had I_SYNC set and we 8204f8ad655SJan Kara * are doing WB_SYNC_NONE writeback. So this catches only the 8214f8ad655SJan Kara * WB_SYNC_ALL case. 8224f8ad655SJan Kara */ 823169ebd90SJan Kara if (inode->i_state & I_SYNC) { 824169ebd90SJan Kara /* Wait for I_SYNC. This function drops i_lock... */ 825169ebd90SJan Kara inode_sleep_on_writeback(inode); 826169ebd90SJan Kara /* Inode may be gone, start again */ 827ead188f9SJan Kara spin_lock(&wb->list_lock); 828169ebd90SJan Kara continue; 829169ebd90SJan Kara } 8304f8ad655SJan Kara inode->i_state |= I_SYNC; 8314f8ad655SJan Kara spin_unlock(&inode->i_lock); 832169ebd90SJan Kara 833a88a341aSTejun Heo write_chunk = writeback_chunk_size(wb, work); 834d46db3d5SWu Fengguang wbc.nr_to_write = write_chunk; 835d46db3d5SWu Fengguang wbc.pages_skipped = 0; 836250df6edSDave Chinner 837169ebd90SJan Kara /* 838169ebd90SJan Kara * We use I_SYNC to pin the inode in memory. While it is set 839169ebd90SJan Kara * evict_inode() will wait so the inode cannot be freed. 840169ebd90SJan Kara */ 841cd8ed2a4SYan Hong __writeback_single_inode(inode, &wbc); 842d46db3d5SWu Fengguang 843d46db3d5SWu Fengguang work->nr_pages -= write_chunk - wbc.nr_to_write; 844d46db3d5SWu Fengguang wrote += write_chunk - wbc.nr_to_write; 8454f8ad655SJan Kara spin_lock(&wb->list_lock); 8464f8ad655SJan Kara spin_lock(&inode->i_lock); 8470ae45f63STheodore Ts'o if (!(inode->i_state & I_DIRTY_ALL)) 848d46db3d5SWu Fengguang wrote++; 8494f8ad655SJan Kara requeue_inode(inode, wb, &wbc); 8504f8ad655SJan Kara inode_sync_complete(inode); 8510f1b1fd8SDave Chinner spin_unlock(&inode->i_lock); 852169ebd90SJan Kara cond_resched_lock(&wb->list_lock); 853d46db3d5SWu Fengguang /* 854d46db3d5SWu Fengguang * bail out to wb_writeback() often enough to check 855d46db3d5SWu Fengguang * background threshold and other termination conditions. 856d46db3d5SWu Fengguang */ 857d46db3d5SWu Fengguang if (wrote) { 858d46db3d5SWu Fengguang if (time_is_before_jiffies(start_time + HZ / 10UL)) 859d46db3d5SWu Fengguang break; 860d46db3d5SWu Fengguang if (work->nr_pages <= 0) 861d46db3d5SWu Fengguang break; 8621da177e4SLinus Torvalds } 8638bc3be27SFengguang Wu } 864d46db3d5SWu Fengguang return wrote; 865f11c9c5cSEdward Shishkin } 86638f21977SNick Piggin 867d46db3d5SWu Fengguang static long __writeback_inodes_wb(struct bdi_writeback *wb, 868d46db3d5SWu Fengguang struct wb_writeback_work *work) 869f11c9c5cSEdward Shishkin { 870d46db3d5SWu Fengguang unsigned long start_time = jiffies; 871d46db3d5SWu Fengguang long wrote = 0; 872f11c9c5cSEdward Shishkin 873f11c9c5cSEdward Shishkin while (!list_empty(&wb->b_io)) { 8747ccf19a8SNick Piggin struct inode *inode = wb_inode(wb->b_io.prev); 875f11c9c5cSEdward Shishkin struct super_block *sb = inode->i_sb; 876f11c9c5cSEdward Shishkin 877eb6ef3dfSKonstantin Khlebnikov if (!trylock_super(sb)) { 8780e995816SWu Fengguang /* 879eb6ef3dfSKonstantin Khlebnikov * trylock_super() may fail consistently due to 8800e995816SWu Fengguang * s_umount being grabbed by someone else. Don't use 8810e995816SWu Fengguang * requeue_io() to avoid busy retrying the inode/sb. 8820e995816SWu Fengguang */ 8830e995816SWu Fengguang redirty_tail(inode, wb); 884d19de7edSChristoph Hellwig continue; 885334132aeSChristoph Hellwig } 886d46db3d5SWu Fengguang wrote += writeback_sb_inodes(sb, wb, work); 887eb6ef3dfSKonstantin Khlebnikov up_read(&sb->s_umount); 888f11c9c5cSEdward Shishkin 889d46db3d5SWu Fengguang /* refer to the same tests at the end of writeback_sb_inodes */ 890d46db3d5SWu Fengguang if (wrote) { 891d46db3d5SWu Fengguang if (time_is_before_jiffies(start_time + HZ / 10UL)) 892d46db3d5SWu Fengguang break; 893d46db3d5SWu Fengguang if (work->nr_pages <= 0) 894f11c9c5cSEdward Shishkin break; 895f11c9c5cSEdward Shishkin } 896d46db3d5SWu Fengguang } 89766f3b8e2SJens Axboe /* Leave any unwritten inodes on b_io */ 898d46db3d5SWu Fengguang return wrote; 89966f3b8e2SJens Axboe } 90066f3b8e2SJens Axboe 9017d9f073bSWanpeng Li static long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages, 9020e175a18SCurt Wohlgemuth enum wb_reason reason) 903edadfb10SChristoph Hellwig { 904d46db3d5SWu Fengguang struct wb_writeback_work work = { 905d46db3d5SWu Fengguang .nr_pages = nr_pages, 906d46db3d5SWu Fengguang .sync_mode = WB_SYNC_NONE, 907d46db3d5SWu Fengguang .range_cyclic = 1, 9080e175a18SCurt Wohlgemuth .reason = reason, 909d46db3d5SWu Fengguang }; 910edadfb10SChristoph Hellwig 911f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 912424b351fSWu Fengguang if (list_empty(&wb->b_io)) 913ad4e38ddSCurt Wohlgemuth queue_io(wb, &work); 914d46db3d5SWu Fengguang __writeback_inodes_wb(wb, &work); 915f758eeabSChristoph Hellwig spin_unlock(&wb->list_lock); 916edadfb10SChristoph Hellwig 917d46db3d5SWu Fengguang return nr_pages - work.nr_pages; 91866f3b8e2SJens Axboe } 91966f3b8e2SJens Axboe 920a88a341aSTejun Heo static bool over_bground_thresh(struct bdi_writeback *wb) 92103ba3782SJens Axboe { 92203ba3782SJens Axboe unsigned long background_thresh, dirty_thresh; 92303ba3782SJens Axboe 92416c4042fSWu Fengguang global_dirty_limits(&background_thresh, &dirty_thresh); 92503ba3782SJens Axboe 926b00949aaSWu Fengguang if (global_page_state(NR_FILE_DIRTY) + 927b00949aaSWu Fengguang global_page_state(NR_UNSTABLE_NFS) > background_thresh) 928b00949aaSWu Fengguang return true; 929b00949aaSWu Fengguang 930a88a341aSTejun Heo if (wb_stat(wb, WB_RECLAIMABLE) > wb_dirty_limit(wb, background_thresh)) 931b00949aaSWu Fengguang return true; 932b00949aaSWu Fengguang 933b00949aaSWu Fengguang return false; 93403ba3782SJens Axboe } 93503ba3782SJens Axboe 93603ba3782SJens Axboe /* 937e98be2d5SWu Fengguang * Called under wb->list_lock. If there are multiple wb per bdi, 938e98be2d5SWu Fengguang * only the flusher working on the first wb should do it. 939e98be2d5SWu Fengguang */ 940e98be2d5SWu Fengguang static void wb_update_bandwidth(struct bdi_writeback *wb, 941e98be2d5SWu Fengguang unsigned long start_time) 942e98be2d5SWu Fengguang { 943a88a341aSTejun Heo __wb_update_bandwidth(wb, 0, 0, 0, 0, 0, start_time); 944e98be2d5SWu Fengguang } 945e98be2d5SWu Fengguang 946e98be2d5SWu Fengguang /* 94703ba3782SJens Axboe * Explicit flushing or periodic writeback of "old" data. 94803ba3782SJens Axboe * 94903ba3782SJens Axboe * Define "old": the first time one of an inode's pages is dirtied, we mark the 95003ba3782SJens Axboe * dirtying-time in the inode's address_space. So this periodic writeback code 95103ba3782SJens Axboe * just walks the superblock inode list, writing back any inodes which are 95203ba3782SJens Axboe * older than a specific point in time. 95303ba3782SJens Axboe * 95403ba3782SJens Axboe * Try to run once per dirty_writeback_interval. But if a writeback event 95503ba3782SJens Axboe * takes longer than a dirty_writeback_interval interval, then leave a 95603ba3782SJens Axboe * one-second gap. 95703ba3782SJens Axboe * 95803ba3782SJens Axboe * older_than_this takes precedence over nr_to_write. So we'll only write back 95903ba3782SJens Axboe * all dirty pages if they are all attached to "old" mappings. 96003ba3782SJens Axboe */ 961c4a77a6cSJens Axboe static long wb_writeback(struct bdi_writeback *wb, 96283ba7b07SChristoph Hellwig struct wb_writeback_work *work) 96303ba3782SJens Axboe { 964e98be2d5SWu Fengguang unsigned long wb_start = jiffies; 965d46db3d5SWu Fengguang long nr_pages = work->nr_pages; 9660dc83bd3SJan Kara unsigned long oldest_jif; 967a5989bdcSJan Kara struct inode *inode; 968d46db3d5SWu Fengguang long progress; 96903ba3782SJens Axboe 9700dc83bd3SJan Kara oldest_jif = jiffies; 9710dc83bd3SJan Kara work->older_than_this = &oldest_jif; 97203ba3782SJens Axboe 973e8dfc305SWu Fengguang spin_lock(&wb->list_lock); 97403ba3782SJens Axboe for (;;) { 97503ba3782SJens Axboe /* 976d3ddec76SWu Fengguang * Stop writeback when nr_pages has been consumed 97703ba3782SJens Axboe */ 97883ba7b07SChristoph Hellwig if (work->nr_pages <= 0) 97903ba3782SJens Axboe break; 98003ba3782SJens Axboe 98103ba3782SJens Axboe /* 982aa373cf5SJan Kara * Background writeout and kupdate-style writeback may 983aa373cf5SJan Kara * run forever. Stop them if there is other work to do 984aa373cf5SJan Kara * so that e.g. sync can proceed. They'll be restarted 985aa373cf5SJan Kara * after the other works are all done. 986aa373cf5SJan Kara */ 987aa373cf5SJan Kara if ((work->for_background || work->for_kupdate) && 988f0054bb1STejun Heo !list_empty(&wb->work_list)) 989aa373cf5SJan Kara break; 990aa373cf5SJan Kara 991aa373cf5SJan Kara /* 992d3ddec76SWu Fengguang * For background writeout, stop when we are below the 993d3ddec76SWu Fengguang * background dirty threshold 99403ba3782SJens Axboe */ 995a88a341aSTejun Heo if (work->for_background && !over_bground_thresh(wb)) 99603ba3782SJens Axboe break; 99703ba3782SJens Axboe 9981bc36b64SJan Kara /* 9991bc36b64SJan Kara * Kupdate and background works are special and we want to 10001bc36b64SJan Kara * include all inodes that need writing. Livelock avoidance is 10011bc36b64SJan Kara * handled by these works yielding to any other work so we are 10021bc36b64SJan Kara * safe. 10031bc36b64SJan Kara */ 1004ba9aa839SWu Fengguang if (work->for_kupdate) { 10050dc83bd3SJan Kara oldest_jif = jiffies - 1006ba9aa839SWu Fengguang msecs_to_jiffies(dirty_expire_interval * 10); 10071bc36b64SJan Kara } else if (work->for_background) 10080dc83bd3SJan Kara oldest_jif = jiffies; 1009028c2dd1SDave Chinner 1010d46db3d5SWu Fengguang trace_writeback_start(wb->bdi, work); 1011e8dfc305SWu Fengguang if (list_empty(&wb->b_io)) 1012ad4e38ddSCurt Wohlgemuth queue_io(wb, work); 101383ba7b07SChristoph Hellwig if (work->sb) 1014d46db3d5SWu Fengguang progress = writeback_sb_inodes(work->sb, wb, work); 1015edadfb10SChristoph Hellwig else 1016d46db3d5SWu Fengguang progress = __writeback_inodes_wb(wb, work); 1017d46db3d5SWu Fengguang trace_writeback_written(wb->bdi, work); 1018028c2dd1SDave Chinner 1019e98be2d5SWu Fengguang wb_update_bandwidth(wb, wb_start); 102003ba3782SJens Axboe 102103ba3782SJens Axboe /* 102271fd05a8SJens Axboe * Did we write something? Try for more 1023e6fb6da2SWu Fengguang * 1024e6fb6da2SWu Fengguang * Dirty inodes are moved to b_io for writeback in batches. 1025e6fb6da2SWu Fengguang * The completion of the current batch does not necessarily 1026e6fb6da2SWu Fengguang * mean the overall work is done. So we keep looping as long 1027e6fb6da2SWu Fengguang * as made some progress on cleaning pages or inodes. 102871fd05a8SJens Axboe */ 1029d46db3d5SWu Fengguang if (progress) 103003ba3782SJens Axboe continue; 1031a5989bdcSJan Kara /* 1032e6fb6da2SWu Fengguang * No more inodes for IO, bail 1033a5989bdcSJan Kara */ 1034b7a2441fSWu Fengguang if (list_empty(&wb->b_more_io)) 103503ba3782SJens Axboe break; 103603ba3782SJens Axboe /* 10378010c3b6SJens Axboe * Nothing written. Wait for some inode to 10388010c3b6SJens Axboe * become available for writeback. Otherwise 10398010c3b6SJens Axboe * we'll just busyloop. 104003ba3782SJens Axboe */ 104103ba3782SJens Axboe if (!list_empty(&wb->b_more_io)) { 1042d46db3d5SWu Fengguang trace_writeback_wait(wb->bdi, work); 104303ba3782SJens Axboe inode = wb_inode(wb->b_more_io.prev); 1044250df6edSDave Chinner spin_lock(&inode->i_lock); 1045f0d07b7fSJan Kara spin_unlock(&wb->list_lock); 1046169ebd90SJan Kara /* This function drops i_lock... */ 1047169ebd90SJan Kara inode_sleep_on_writeback(inode); 1048f0d07b7fSJan Kara spin_lock(&wb->list_lock); 104903ba3782SJens Axboe } 105003ba3782SJens Axboe } 1051e8dfc305SWu Fengguang spin_unlock(&wb->list_lock); 105203ba3782SJens Axboe 1053d46db3d5SWu Fengguang return nr_pages - work->nr_pages; 105403ba3782SJens Axboe } 105503ba3782SJens Axboe 105603ba3782SJens Axboe /* 105783ba7b07SChristoph Hellwig * Return the next wb_writeback_work struct that hasn't been processed yet. 105803ba3782SJens Axboe */ 1059f0054bb1STejun Heo static struct wb_writeback_work *get_next_work_item(struct bdi_writeback *wb) 106003ba3782SJens Axboe { 106183ba7b07SChristoph Hellwig struct wb_writeback_work *work = NULL; 106203ba3782SJens Axboe 1063f0054bb1STejun Heo spin_lock_bh(&wb->work_lock); 1064f0054bb1STejun Heo if (!list_empty(&wb->work_list)) { 1065f0054bb1STejun Heo work = list_entry(wb->work_list.next, 106683ba7b07SChristoph Hellwig struct wb_writeback_work, list); 106783ba7b07SChristoph Hellwig list_del_init(&work->list); 106803ba3782SJens Axboe } 1069f0054bb1STejun Heo spin_unlock_bh(&wb->work_lock); 107083ba7b07SChristoph Hellwig return work; 107103ba3782SJens Axboe } 107203ba3782SJens Axboe 1073cdf01dd5SLinus Torvalds /* 1074cdf01dd5SLinus Torvalds * Add in the number of potentially dirty inodes, because each inode 1075cdf01dd5SLinus Torvalds * write can dirty pagecache in the underlying blockdev. 1076cdf01dd5SLinus Torvalds */ 1077cdf01dd5SLinus Torvalds static unsigned long get_nr_dirty_pages(void) 1078cdf01dd5SLinus Torvalds { 1079cdf01dd5SLinus Torvalds return global_page_state(NR_FILE_DIRTY) + 1080cdf01dd5SLinus Torvalds global_page_state(NR_UNSTABLE_NFS) + 1081cdf01dd5SLinus Torvalds get_nr_dirty_inodes(); 1082cdf01dd5SLinus Torvalds } 1083cdf01dd5SLinus Torvalds 10846585027aSJan Kara static long wb_check_background_flush(struct bdi_writeback *wb) 10856585027aSJan Kara { 1086a88a341aSTejun Heo if (over_bground_thresh(wb)) { 10876585027aSJan Kara 10886585027aSJan Kara struct wb_writeback_work work = { 10896585027aSJan Kara .nr_pages = LONG_MAX, 10906585027aSJan Kara .sync_mode = WB_SYNC_NONE, 10916585027aSJan Kara .for_background = 1, 10926585027aSJan Kara .range_cyclic = 1, 10930e175a18SCurt Wohlgemuth .reason = WB_REASON_BACKGROUND, 10946585027aSJan Kara }; 10956585027aSJan Kara 10966585027aSJan Kara return wb_writeback(wb, &work); 10976585027aSJan Kara } 10986585027aSJan Kara 10996585027aSJan Kara return 0; 11006585027aSJan Kara } 11016585027aSJan Kara 110203ba3782SJens Axboe static long wb_check_old_data_flush(struct bdi_writeback *wb) 110303ba3782SJens Axboe { 110403ba3782SJens Axboe unsigned long expired; 110503ba3782SJens Axboe long nr_pages; 110603ba3782SJens Axboe 110769b62d01SJens Axboe /* 110869b62d01SJens Axboe * When set to zero, disable periodic writeback 110969b62d01SJens Axboe */ 111069b62d01SJens Axboe if (!dirty_writeback_interval) 111169b62d01SJens Axboe return 0; 111269b62d01SJens Axboe 111303ba3782SJens Axboe expired = wb->last_old_flush + 111403ba3782SJens Axboe msecs_to_jiffies(dirty_writeback_interval * 10); 111503ba3782SJens Axboe if (time_before(jiffies, expired)) 111603ba3782SJens Axboe return 0; 111703ba3782SJens Axboe 111803ba3782SJens Axboe wb->last_old_flush = jiffies; 1119cdf01dd5SLinus Torvalds nr_pages = get_nr_dirty_pages(); 112003ba3782SJens Axboe 1121c4a77a6cSJens Axboe if (nr_pages) { 112283ba7b07SChristoph Hellwig struct wb_writeback_work work = { 1123c4a77a6cSJens Axboe .nr_pages = nr_pages, 1124c4a77a6cSJens Axboe .sync_mode = WB_SYNC_NONE, 1125c4a77a6cSJens Axboe .for_kupdate = 1, 1126c4a77a6cSJens Axboe .range_cyclic = 1, 11270e175a18SCurt Wohlgemuth .reason = WB_REASON_PERIODIC, 1128c4a77a6cSJens Axboe }; 1129c4a77a6cSJens Axboe 113083ba7b07SChristoph Hellwig return wb_writeback(wb, &work); 1131c4a77a6cSJens Axboe } 113203ba3782SJens Axboe 113303ba3782SJens Axboe return 0; 113403ba3782SJens Axboe } 113503ba3782SJens Axboe 113603ba3782SJens Axboe /* 113703ba3782SJens Axboe * Retrieve work items and do the writeback they describe 113803ba3782SJens Axboe */ 113925d130baSWanpeng Li static long wb_do_writeback(struct bdi_writeback *wb) 114003ba3782SJens Axboe { 114183ba7b07SChristoph Hellwig struct wb_writeback_work *work; 1142c4a77a6cSJens Axboe long wrote = 0; 114303ba3782SJens Axboe 11444452226eSTejun Heo set_bit(WB_writeback_running, &wb->state); 1145f0054bb1STejun Heo while ((work = get_next_work_item(wb)) != NULL) { 1146ac7b19a3STejun Heo struct completion *done = work->done; 114783ba7b07SChristoph Hellwig 1148f0054bb1STejun Heo trace_writeback_exec(wb->bdi, work); 1149455b2864SDave Chinner 115083ba7b07SChristoph Hellwig wrote += wb_writeback(wb, work); 115103ba3782SJens Axboe 1152ac7b19a3STejun Heo if (work->auto_free) 115383ba7b07SChristoph Hellwig kfree(work); 1154ac7b19a3STejun Heo if (done) 1155ac7b19a3STejun Heo complete(done); 115603ba3782SJens Axboe } 115703ba3782SJens Axboe 115803ba3782SJens Axboe /* 115903ba3782SJens Axboe * Check for periodic writeback, kupdated() style 116003ba3782SJens Axboe */ 116103ba3782SJens Axboe wrote += wb_check_old_data_flush(wb); 11626585027aSJan Kara wrote += wb_check_background_flush(wb); 11634452226eSTejun Heo clear_bit(WB_writeback_running, &wb->state); 116403ba3782SJens Axboe 116503ba3782SJens Axboe return wrote; 116603ba3782SJens Axboe } 116703ba3782SJens Axboe 116803ba3782SJens Axboe /* 116903ba3782SJens Axboe * Handle writeback of dirty data for the device backed by this bdi. Also 1170839a8e86STejun Heo * reschedules periodically and does kupdated style flushing. 117103ba3782SJens Axboe */ 1172f0054bb1STejun Heo void wb_workfn(struct work_struct *work) 117303ba3782SJens Axboe { 1174839a8e86STejun Heo struct bdi_writeback *wb = container_of(to_delayed_work(work), 1175839a8e86STejun Heo struct bdi_writeback, dwork); 117603ba3782SJens Axboe long pages_written; 117703ba3782SJens Axboe 1178f0054bb1STejun Heo set_worker_desc("flush-%s", dev_name(wb->bdi->dev)); 1179766f9164SPeter Zijlstra current->flags |= PF_SWAPWRITE; 118003ba3782SJens Axboe 1181839a8e86STejun Heo if (likely(!current_is_workqueue_rescuer() || 11824452226eSTejun Heo !test_bit(WB_registered, &wb->state))) { 118303ba3782SJens Axboe /* 1184f0054bb1STejun Heo * The normal path. Keep writing back @wb until its 1185839a8e86STejun Heo * work_list is empty. Note that this path is also taken 1186f0054bb1STejun Heo * if @wb is shutting down even when we're running off the 1187839a8e86STejun Heo * rescuer as work_list needs to be drained. 118803ba3782SJens Axboe */ 1189839a8e86STejun Heo do { 119025d130baSWanpeng Li pages_written = wb_do_writeback(wb); 1191455b2864SDave Chinner trace_writeback_pages_written(pages_written); 1192f0054bb1STejun Heo } while (!list_empty(&wb->work_list)); 1193839a8e86STejun Heo } else { 1194253c34e9SArtem Bityutskiy /* 1195839a8e86STejun Heo * bdi_wq can't get enough workers and we're running off 1196839a8e86STejun Heo * the emergency worker. Don't hog it. Hopefully, 1024 is 1197839a8e86STejun Heo * enough for efficient IO. 1198253c34e9SArtem Bityutskiy */ 1199f0054bb1STejun Heo pages_written = writeback_inodes_wb(wb, 1024, 1200839a8e86STejun Heo WB_REASON_FORKER_THREAD); 1201839a8e86STejun Heo trace_writeback_pages_written(pages_written); 120203ba3782SJens Axboe } 120303ba3782SJens Axboe 1204f0054bb1STejun Heo if (!list_empty(&wb->work_list)) 12056ca738d6SDerek Basehore mod_delayed_work(bdi_wq, &wb->dwork, 0); 12066ca738d6SDerek Basehore else if (wb_has_dirty_io(wb) && dirty_writeback_interval) 1207f0054bb1STejun Heo wb_wakeup_delayed(wb); 1208455b2864SDave Chinner 1209839a8e86STejun Heo current->flags &= ~PF_SWAPWRITE; 121003ba3782SJens Axboe } 121103ba3782SJens Axboe 121203ba3782SJens Axboe /* 121303ba3782SJens Axboe * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back 121403ba3782SJens Axboe * the whole world. 121503ba3782SJens Axboe */ 12160e175a18SCurt Wohlgemuth void wakeup_flusher_threads(long nr_pages, enum wb_reason reason) 121703ba3782SJens Axboe { 1218b8c2f347SChristoph Hellwig struct backing_dev_info *bdi; 1219b8c2f347SChristoph Hellwig 122047df3ddeSJan Kara if (!nr_pages) 122147df3ddeSJan Kara nr_pages = get_nr_dirty_pages(); 1222b8c2f347SChristoph Hellwig 1223b8c2f347SChristoph Hellwig rcu_read_lock(); 1224f2b65121STejun Heo list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { 1225f2b65121STejun Heo struct bdi_writeback *wb; 1226f2b65121STejun Heo struct wb_iter iter; 1227f2b65121STejun Heo 1228f2b65121STejun Heo if (!bdi_has_dirty_io(bdi)) 1229f2b65121STejun Heo continue; 1230f2b65121STejun Heo 1231f2b65121STejun Heo bdi_for_each_wb(wb, bdi, &iter, 0) 1232f2b65121STejun Heo wb_start_writeback(wb, wb_split_bdi_pages(wb, nr_pages), 1233f2b65121STejun Heo false, reason); 1234f2b65121STejun Heo } 1235b8c2f347SChristoph Hellwig rcu_read_unlock(); 123603ba3782SJens Axboe } 123703ba3782SJens Axboe 1238a2f48706STheodore Ts'o /* 1239a2f48706STheodore Ts'o * Wake up bdi's periodically to make sure dirtytime inodes gets 1240a2f48706STheodore Ts'o * written back periodically. We deliberately do *not* check the 1241a2f48706STheodore Ts'o * b_dirtytime list in wb_has_dirty_io(), since this would cause the 1242a2f48706STheodore Ts'o * kernel to be constantly waking up once there are any dirtytime 1243a2f48706STheodore Ts'o * inodes on the system. So instead we define a separate delayed work 1244a2f48706STheodore Ts'o * function which gets called much more rarely. (By default, only 1245a2f48706STheodore Ts'o * once every 12 hours.) 1246a2f48706STheodore Ts'o * 1247a2f48706STheodore Ts'o * If there is any other write activity going on in the file system, 1248a2f48706STheodore Ts'o * this function won't be necessary. But if the only thing that has 1249a2f48706STheodore Ts'o * happened on the file system is a dirtytime inode caused by an atime 1250a2f48706STheodore Ts'o * update, we need this infrastructure below to make sure that inode 1251a2f48706STheodore Ts'o * eventually gets pushed out to disk. 1252a2f48706STheodore Ts'o */ 1253a2f48706STheodore Ts'o static void wakeup_dirtytime_writeback(struct work_struct *w); 1254a2f48706STheodore Ts'o static DECLARE_DELAYED_WORK(dirtytime_work, wakeup_dirtytime_writeback); 1255a2f48706STheodore Ts'o 1256a2f48706STheodore Ts'o static void wakeup_dirtytime_writeback(struct work_struct *w) 1257a2f48706STheodore Ts'o { 1258a2f48706STheodore Ts'o struct backing_dev_info *bdi; 1259a2f48706STheodore Ts'o 1260a2f48706STheodore Ts'o rcu_read_lock(); 1261a2f48706STheodore Ts'o list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { 1262001fe6f6STejun Heo struct bdi_writeback *wb; 1263001fe6f6STejun Heo struct wb_iter iter; 1264001fe6f6STejun Heo 1265001fe6f6STejun Heo bdi_for_each_wb(wb, bdi, &iter, 0) 1266001fe6f6STejun Heo if (!list_empty(&bdi->wb.b_dirty_time)) 1267f0054bb1STejun Heo wb_wakeup(&bdi->wb); 1268a2f48706STheodore Ts'o } 1269a2f48706STheodore Ts'o rcu_read_unlock(); 1270a2f48706STheodore Ts'o schedule_delayed_work(&dirtytime_work, dirtytime_expire_interval * HZ); 1271a2f48706STheodore Ts'o } 1272a2f48706STheodore Ts'o 1273a2f48706STheodore Ts'o static int __init start_dirtytime_writeback(void) 1274a2f48706STheodore Ts'o { 1275a2f48706STheodore Ts'o schedule_delayed_work(&dirtytime_work, dirtytime_expire_interval * HZ); 1276a2f48706STheodore Ts'o return 0; 1277a2f48706STheodore Ts'o } 1278a2f48706STheodore Ts'o __initcall(start_dirtytime_writeback); 1279a2f48706STheodore Ts'o 12801efff914STheodore Ts'o int dirtytime_interval_handler(struct ctl_table *table, int write, 12811efff914STheodore Ts'o void __user *buffer, size_t *lenp, loff_t *ppos) 12821efff914STheodore Ts'o { 12831efff914STheodore Ts'o int ret; 12841efff914STheodore Ts'o 12851efff914STheodore Ts'o ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos); 12861efff914STheodore Ts'o if (ret == 0 && write) 12871efff914STheodore Ts'o mod_delayed_work(system_wq, &dirtytime_work, 0); 12881efff914STheodore Ts'o return ret; 12891efff914STheodore Ts'o } 12901efff914STheodore Ts'o 129103ba3782SJens Axboe static noinline void block_dump___mark_inode_dirty(struct inode *inode) 129203ba3782SJens Axboe { 129303ba3782SJens Axboe if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { 129403ba3782SJens Axboe struct dentry *dentry; 129503ba3782SJens Axboe const char *name = "?"; 129603ba3782SJens Axboe 129703ba3782SJens Axboe dentry = d_find_alias(inode); 129803ba3782SJens Axboe if (dentry) { 129903ba3782SJens Axboe spin_lock(&dentry->d_lock); 130003ba3782SJens Axboe name = (const char *) dentry->d_name.name; 130103ba3782SJens Axboe } 130203ba3782SJens Axboe printk(KERN_DEBUG 130303ba3782SJens Axboe "%s(%d): dirtied inode %lu (%s) on %s\n", 130403ba3782SJens Axboe current->comm, task_pid_nr(current), inode->i_ino, 130503ba3782SJens Axboe name, inode->i_sb->s_id); 130603ba3782SJens Axboe if (dentry) { 130703ba3782SJens Axboe spin_unlock(&dentry->d_lock); 130803ba3782SJens Axboe dput(dentry); 130903ba3782SJens Axboe } 131003ba3782SJens Axboe } 131103ba3782SJens Axboe } 131203ba3782SJens Axboe 131303ba3782SJens Axboe /** 131403ba3782SJens Axboe * __mark_inode_dirty - internal function 131503ba3782SJens Axboe * @inode: inode to mark 131603ba3782SJens Axboe * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) 131703ba3782SJens Axboe * Mark an inode as dirty. Callers should use mark_inode_dirty or 131803ba3782SJens Axboe * mark_inode_dirty_sync. 131903ba3782SJens Axboe * 132003ba3782SJens Axboe * Put the inode on the super block's dirty list. 132103ba3782SJens Axboe * 132203ba3782SJens Axboe * CAREFUL! We mark it dirty unconditionally, but move it onto the 132303ba3782SJens Axboe * dirty list only if it is hashed or if it refers to a blockdev. 132403ba3782SJens Axboe * If it was not hashed, it will never be added to the dirty list 132503ba3782SJens Axboe * even if it is later hashed, as it will have been marked dirty already. 132603ba3782SJens Axboe * 132703ba3782SJens Axboe * In short, make sure you hash any inodes _before_ you start marking 132803ba3782SJens Axboe * them dirty. 132903ba3782SJens Axboe * 133003ba3782SJens Axboe * Note that for blockdevs, inode->dirtied_when represents the dirtying time of 133103ba3782SJens Axboe * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of 133203ba3782SJens Axboe * the kernel-internal blockdev inode represents the dirtying time of the 133303ba3782SJens Axboe * blockdev's pages. This is why for I_DIRTY_PAGES we always use 133403ba3782SJens Axboe * page->mapping->host, so the page-dirtying time is recorded in the internal 133503ba3782SJens Axboe * blockdev inode. 133603ba3782SJens Axboe */ 13370ae45f63STheodore Ts'o #define I_DIRTY_INODE (I_DIRTY_SYNC | I_DIRTY_DATASYNC) 133803ba3782SJens Axboe void __mark_inode_dirty(struct inode *inode, int flags) 133903ba3782SJens Axboe { 134003ba3782SJens Axboe struct super_block *sb = inode->i_sb; 1341253c34e9SArtem Bityutskiy struct backing_dev_info *bdi = NULL; 13420ae45f63STheodore Ts'o int dirtytime; 13430ae45f63STheodore Ts'o 13440ae45f63STheodore Ts'o trace_writeback_mark_inode_dirty(inode, flags); 134503ba3782SJens Axboe 134603ba3782SJens Axboe /* 134703ba3782SJens Axboe * Don't do this for I_DIRTY_PAGES - that doesn't actually 134803ba3782SJens Axboe * dirty the inode itself 134903ba3782SJens Axboe */ 13500ae45f63STheodore Ts'o if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC | I_DIRTY_TIME)) { 13519fb0a7daSTejun Heo trace_writeback_dirty_inode_start(inode, flags); 13529fb0a7daSTejun Heo 135303ba3782SJens Axboe if (sb->s_op->dirty_inode) 1354aa385729SChristoph Hellwig sb->s_op->dirty_inode(inode, flags); 13559fb0a7daSTejun Heo 13569fb0a7daSTejun Heo trace_writeback_dirty_inode(inode, flags); 135703ba3782SJens Axboe } 13580ae45f63STheodore Ts'o if (flags & I_DIRTY_INODE) 13590ae45f63STheodore Ts'o flags &= ~I_DIRTY_TIME; 13600ae45f63STheodore Ts'o dirtytime = flags & I_DIRTY_TIME; 136103ba3782SJens Axboe 136203ba3782SJens Axboe /* 13639c6ac78eSTejun Heo * Paired with smp_mb() in __writeback_single_inode() for the 13649c6ac78eSTejun Heo * following lockless i_state test. See there for details. 136503ba3782SJens Axboe */ 136603ba3782SJens Axboe smp_mb(); 136703ba3782SJens Axboe 13680ae45f63STheodore Ts'o if (((inode->i_state & flags) == flags) || 13690ae45f63STheodore Ts'o (dirtytime && (inode->i_state & I_DIRTY_INODE))) 137003ba3782SJens Axboe return; 137103ba3782SJens Axboe 137203ba3782SJens Axboe if (unlikely(block_dump)) 137303ba3782SJens Axboe block_dump___mark_inode_dirty(inode); 137403ba3782SJens Axboe 1375250df6edSDave Chinner spin_lock(&inode->i_lock); 13760ae45f63STheodore Ts'o if (dirtytime && (inode->i_state & I_DIRTY_INODE)) 13770ae45f63STheodore Ts'o goto out_unlock_inode; 137803ba3782SJens Axboe if ((inode->i_state & flags) != flags) { 137903ba3782SJens Axboe const int was_dirty = inode->i_state & I_DIRTY; 138003ba3782SJens Axboe 138152ebea74STejun Heo inode_attach_wb(inode, NULL); 138252ebea74STejun Heo 13830ae45f63STheodore Ts'o if (flags & I_DIRTY_INODE) 13840ae45f63STheodore Ts'o inode->i_state &= ~I_DIRTY_TIME; 138503ba3782SJens Axboe inode->i_state |= flags; 138603ba3782SJens Axboe 138703ba3782SJens Axboe /* 138803ba3782SJens Axboe * If the inode is being synced, just update its dirty state. 138903ba3782SJens Axboe * The unlocker will place the inode on the appropriate 139003ba3782SJens Axboe * superblock list, based upon its state. 139103ba3782SJens Axboe */ 139203ba3782SJens Axboe if (inode->i_state & I_SYNC) 1393250df6edSDave Chinner goto out_unlock_inode; 139403ba3782SJens Axboe 139503ba3782SJens Axboe /* 139603ba3782SJens Axboe * Only add valid (hashed) inodes to the superblock's 139703ba3782SJens Axboe * dirty list. Add blockdev inodes as well. 139803ba3782SJens Axboe */ 139903ba3782SJens Axboe if (!S_ISBLK(inode->i_mode)) { 14001d3382cbSAl Viro if (inode_unhashed(inode)) 1401250df6edSDave Chinner goto out_unlock_inode; 140203ba3782SJens Axboe } 1403a4ffdde6SAl Viro if (inode->i_state & I_FREEING) 1404250df6edSDave Chinner goto out_unlock_inode; 140503ba3782SJens Axboe 140603ba3782SJens Axboe /* 140703ba3782SJens Axboe * If the inode was already on b_dirty/b_io/b_more_io, don't 140803ba3782SJens Axboe * reposition it (that would break b_dirty time-ordering). 140903ba3782SJens Axboe */ 141003ba3782SJens Axboe if (!was_dirty) { 1411d6c10f1fSTejun Heo struct list_head *dirty_list; 1412a66979abSDave Chinner bool wakeup_bdi = false; 1413253c34e9SArtem Bityutskiy bdi = inode_to_bdi(inode); 1414500b067cSJens Axboe 1415146d7009SJunxiao Bi spin_unlock(&inode->i_lock); 1416146d7009SJunxiao Bi spin_lock(&bdi->wb.list_lock); 1417253c34e9SArtem Bityutskiy 1418d6c10f1fSTejun Heo WARN(bdi_cap_writeback_dirty(bdi) && 1419d6c10f1fSTejun Heo !test_bit(WB_registered, &bdi->wb.state), 1420d6c10f1fSTejun Heo "bdi-%s not registered\n", bdi->name); 142103ba3782SJens Axboe 142203ba3782SJens Axboe inode->dirtied_when = jiffies; 1423a2f48706STheodore Ts'o if (dirtytime) 1424a2f48706STheodore Ts'o inode->dirtied_time_when = jiffies; 1425d6c10f1fSTejun Heo 1426a2f48706STheodore Ts'o if (inode->i_state & (I_DIRTY_INODE | I_DIRTY_PAGES)) 1427d6c10f1fSTejun Heo dirty_list = &bdi->wb.b_dirty; 1428a2f48706STheodore Ts'o else 1429d6c10f1fSTejun Heo dirty_list = &bdi->wb.b_dirty_time; 1430d6c10f1fSTejun Heo 1431d6c10f1fSTejun Heo wakeup_bdi = inode_wb_list_move_locked(inode, &bdi->wb, 1432d6c10f1fSTejun Heo dirty_list); 1433d6c10f1fSTejun Heo 1434f758eeabSChristoph Hellwig spin_unlock(&bdi->wb.list_lock); 14350ae45f63STheodore Ts'o trace_writeback_dirty_inode_enqueue(inode); 1436253c34e9SArtem Bityutskiy 1437d6c10f1fSTejun Heo /* 1438d6c10f1fSTejun Heo * If this is the first dirty inode for this bdi, 1439d6c10f1fSTejun Heo * we have to wake-up the corresponding bdi thread 1440d6c10f1fSTejun Heo * to make sure background write-back happens 1441d6c10f1fSTejun Heo * later. 1442d6c10f1fSTejun Heo */ 1443d6c10f1fSTejun Heo if (bdi_cap_writeback_dirty(bdi) && wakeup_bdi) 1444f0054bb1STejun Heo wb_wakeup_delayed(&bdi->wb); 1445a66979abSDave Chinner return; 1446a66979abSDave Chinner } 1447a66979abSDave Chinner } 1448a66979abSDave Chinner out_unlock_inode: 1449a66979abSDave Chinner spin_unlock(&inode->i_lock); 1450a66979abSDave Chinner 145103ba3782SJens Axboe } 145203ba3782SJens Axboe EXPORT_SYMBOL(__mark_inode_dirty); 145303ba3782SJens Axboe 1454b6e51316SJens Axboe static void wait_sb_inodes(struct super_block *sb) 145566f3b8e2SJens Axboe { 145638f21977SNick Piggin struct inode *inode, *old_inode = NULL; 145738f21977SNick Piggin 145803ba3782SJens Axboe /* 145903ba3782SJens Axboe * We need to be protected against the filesystem going from 146003ba3782SJens Axboe * r/o to r/w or vice versa. 146103ba3782SJens Axboe */ 1462b6e51316SJens Axboe WARN_ON(!rwsem_is_locked(&sb->s_umount)); 146303ba3782SJens Axboe 146455fa6091SDave Chinner spin_lock(&inode_sb_list_lock); 146566f3b8e2SJens Axboe 146638f21977SNick Piggin /* 146738f21977SNick Piggin * Data integrity sync. Must wait for all pages under writeback, 146838f21977SNick Piggin * because there may have been pages dirtied before our sync 146938f21977SNick Piggin * call, but which had writeout started before we write it out. 147038f21977SNick Piggin * In which case, the inode may not be on the dirty list, but 147138f21977SNick Piggin * we still have to wait for that writeout. 147238f21977SNick Piggin */ 1473b6e51316SJens Axboe list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { 1474250df6edSDave Chinner struct address_space *mapping = inode->i_mapping; 147538f21977SNick Piggin 1476250df6edSDave Chinner spin_lock(&inode->i_lock); 1477250df6edSDave Chinner if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) || 1478250df6edSDave Chinner (mapping->nrpages == 0)) { 1479250df6edSDave Chinner spin_unlock(&inode->i_lock); 148038f21977SNick Piggin continue; 1481250df6edSDave Chinner } 148238f21977SNick Piggin __iget(inode); 1483250df6edSDave Chinner spin_unlock(&inode->i_lock); 148455fa6091SDave Chinner spin_unlock(&inode_sb_list_lock); 148555fa6091SDave Chinner 148638f21977SNick Piggin /* 148755fa6091SDave Chinner * We hold a reference to 'inode' so it couldn't have been 148855fa6091SDave Chinner * removed from s_inodes list while we dropped the 148955fa6091SDave Chinner * inode_sb_list_lock. We cannot iput the inode now as we can 149055fa6091SDave Chinner * be holding the last reference and we cannot iput it under 149155fa6091SDave Chinner * inode_sb_list_lock. So we keep the reference and iput it 149255fa6091SDave Chinner * later. 149338f21977SNick Piggin */ 149438f21977SNick Piggin iput(old_inode); 149538f21977SNick Piggin old_inode = inode; 149638f21977SNick Piggin 149738f21977SNick Piggin filemap_fdatawait(mapping); 149838f21977SNick Piggin 149938f21977SNick Piggin cond_resched(); 150038f21977SNick Piggin 150155fa6091SDave Chinner spin_lock(&inode_sb_list_lock); 150238f21977SNick Piggin } 150355fa6091SDave Chinner spin_unlock(&inode_sb_list_lock); 150438f21977SNick Piggin iput(old_inode); 150566f3b8e2SJens Axboe } 15061da177e4SLinus Torvalds 1507d8a8559cSJens Axboe /** 15083259f8beSChris Mason * writeback_inodes_sb_nr - writeback dirty inodes from given super_block 1509d8a8559cSJens Axboe * @sb: the superblock 15103259f8beSChris Mason * @nr: the number of pages to write 1511786228abSMarcos Paulo de Souza * @reason: reason why some writeback work initiated 15121da177e4SLinus Torvalds * 1513d8a8559cSJens Axboe * Start writeback on some inodes on this super_block. No guarantees are made 1514d8a8559cSJens Axboe * on how many (if any) will be written, and this function does not wait 15153259f8beSChris Mason * for IO completion of submitted IO. 15161da177e4SLinus Torvalds */ 15170e175a18SCurt Wohlgemuth void writeback_inodes_sb_nr(struct super_block *sb, 15180e175a18SCurt Wohlgemuth unsigned long nr, 15190e175a18SCurt Wohlgemuth enum wb_reason reason) 15201da177e4SLinus Torvalds { 152183ba7b07SChristoph Hellwig DECLARE_COMPLETION_ONSTACK(done); 152283ba7b07SChristoph Hellwig struct wb_writeback_work work = { 15233c4d7165SChristoph Hellwig .sb = sb, 15243c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_NONE, 15256e6938b6SWu Fengguang .tagged_writepages = 1, 152683ba7b07SChristoph Hellwig .done = &done, 15273259f8beSChris Mason .nr_pages = nr, 15280e175a18SCurt Wohlgemuth .reason = reason, 15293c4d7165SChristoph Hellwig }; 1530e7972912STejun Heo struct backing_dev_info *bdi = sb->s_bdi; 15310e3c9a22SJens Axboe 1532e7972912STejun Heo if (!bdi_has_dirty_io(bdi) || bdi == &noop_backing_dev_info) 15336eedc701SJan Kara return; 1534cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 1535e7972912STejun Heo wb_queue_work(&bdi->wb, &work); 153683ba7b07SChristoph Hellwig wait_for_completion(&done); 15371da177e4SLinus Torvalds } 15383259f8beSChris Mason EXPORT_SYMBOL(writeback_inodes_sb_nr); 15393259f8beSChris Mason 15403259f8beSChris Mason /** 15413259f8beSChris Mason * writeback_inodes_sb - writeback dirty inodes from given super_block 15423259f8beSChris Mason * @sb: the superblock 1543786228abSMarcos Paulo de Souza * @reason: reason why some writeback work was initiated 15443259f8beSChris Mason * 15453259f8beSChris Mason * Start writeback on some inodes on this super_block. No guarantees are made 15463259f8beSChris Mason * on how many (if any) will be written, and this function does not wait 15473259f8beSChris Mason * for IO completion of submitted IO. 15483259f8beSChris Mason */ 15490e175a18SCurt Wohlgemuth void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) 15503259f8beSChris Mason { 15510e175a18SCurt Wohlgemuth return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); 15523259f8beSChris Mason } 1553d8a8559cSJens Axboe EXPORT_SYMBOL(writeback_inodes_sb); 1554d8a8559cSJens Axboe 1555d8a8559cSJens Axboe /** 155610ee27a0SMiao Xie * try_to_writeback_inodes_sb_nr - try to start writeback if none underway 15573259f8beSChris Mason * @sb: the superblock 15583259f8beSChris Mason * @nr: the number of pages to write 155910ee27a0SMiao Xie * @reason: the reason of writeback 15603259f8beSChris Mason * 156110ee27a0SMiao Xie * Invoke writeback_inodes_sb_nr if no writeback is currently underway. 15623259f8beSChris Mason * Returns 1 if writeback was started, 0 if not. 15633259f8beSChris Mason */ 156410ee27a0SMiao Xie int try_to_writeback_inodes_sb_nr(struct super_block *sb, 15650e175a18SCurt Wohlgemuth unsigned long nr, 15660e175a18SCurt Wohlgemuth enum wb_reason reason) 15673259f8beSChris Mason { 1568bc05873dSTejun Heo if (writeback_in_progress(&sb->s_bdi->wb)) 156910ee27a0SMiao Xie return 1; 157010ee27a0SMiao Xie 157110ee27a0SMiao Xie if (!down_read_trylock(&sb->s_umount)) 157210ee27a0SMiao Xie return 0; 157310ee27a0SMiao Xie 15740e175a18SCurt Wohlgemuth writeback_inodes_sb_nr(sb, nr, reason); 15753259f8beSChris Mason up_read(&sb->s_umount); 15763259f8beSChris Mason return 1; 15773259f8beSChris Mason } 157810ee27a0SMiao Xie EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr); 157910ee27a0SMiao Xie 158010ee27a0SMiao Xie /** 158110ee27a0SMiao Xie * try_to_writeback_inodes_sb - try to start writeback if none underway 158210ee27a0SMiao Xie * @sb: the superblock 158310ee27a0SMiao Xie * @reason: reason why some writeback work was initiated 158410ee27a0SMiao Xie * 158510ee27a0SMiao Xie * Implement by try_to_writeback_inodes_sb_nr() 158610ee27a0SMiao Xie * Returns 1 if writeback was started, 0 if not. 158710ee27a0SMiao Xie */ 158810ee27a0SMiao Xie int try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) 158910ee27a0SMiao Xie { 159010ee27a0SMiao Xie return try_to_writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); 159110ee27a0SMiao Xie } 159210ee27a0SMiao Xie EXPORT_SYMBOL(try_to_writeback_inodes_sb); 15933259f8beSChris Mason 15943259f8beSChris Mason /** 1595d8a8559cSJens Axboe * sync_inodes_sb - sync sb inode pages 1596d8a8559cSJens Axboe * @sb: the superblock 1597d8a8559cSJens Axboe * 1598d8a8559cSJens Axboe * This function writes and waits on any dirty inode belonging to this 15990dc83bd3SJan Kara * super_block. 1600d8a8559cSJens Axboe */ 16010dc83bd3SJan Kara void sync_inodes_sb(struct super_block *sb) 1602d8a8559cSJens Axboe { 160383ba7b07SChristoph Hellwig DECLARE_COMPLETION_ONSTACK(done); 160483ba7b07SChristoph Hellwig struct wb_writeback_work work = { 16053c4d7165SChristoph Hellwig .sb = sb, 16063c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_ALL, 16073c4d7165SChristoph Hellwig .nr_pages = LONG_MAX, 16083c4d7165SChristoph Hellwig .range_cyclic = 0, 160983ba7b07SChristoph Hellwig .done = &done, 16100e175a18SCurt Wohlgemuth .reason = WB_REASON_SYNC, 16117747bd4bSDave Chinner .for_sync = 1, 16123c4d7165SChristoph Hellwig }; 1613e7972912STejun Heo struct backing_dev_info *bdi = sb->s_bdi; 16143c4d7165SChristoph Hellwig 16156eedc701SJan Kara /* Nothing to do? */ 1616e7972912STejun Heo if (!bdi_has_dirty_io(bdi) || bdi == &noop_backing_dev_info) 16176eedc701SJan Kara return; 1618cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 1619cf37e972SChristoph Hellwig 1620e7972912STejun Heo wb_queue_work(&bdi->wb, &work); 162183ba7b07SChristoph Hellwig wait_for_completion(&done); 162283ba7b07SChristoph Hellwig 1623b6e51316SJens Axboe wait_sb_inodes(sb); 1624d8a8559cSJens Axboe } 1625d8a8559cSJens Axboe EXPORT_SYMBOL(sync_inodes_sb); 16261da177e4SLinus Torvalds 16271da177e4SLinus Torvalds /** 16281da177e4SLinus Torvalds * write_inode_now - write an inode to disk 16291da177e4SLinus Torvalds * @inode: inode to write to disk 16301da177e4SLinus Torvalds * @sync: whether the write should be synchronous or not 16311da177e4SLinus Torvalds * 16327f04c26dSAndrea Arcangeli * This function commits an inode to disk immediately if it is dirty. This is 16337f04c26dSAndrea Arcangeli * primarily needed by knfsd. 16347f04c26dSAndrea Arcangeli * 16357f04c26dSAndrea Arcangeli * The caller must either have a ref on the inode or must have set I_WILL_FREE. 16361da177e4SLinus Torvalds */ 16371da177e4SLinus Torvalds int write_inode_now(struct inode *inode, int sync) 16381da177e4SLinus Torvalds { 1639f758eeabSChristoph Hellwig struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; 16401da177e4SLinus Torvalds struct writeback_control wbc = { 16411da177e4SLinus Torvalds .nr_to_write = LONG_MAX, 164218914b18SMike Galbraith .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, 1643111ebb6eSOGAWA Hirofumi .range_start = 0, 1644111ebb6eSOGAWA Hirofumi .range_end = LLONG_MAX, 16451da177e4SLinus Torvalds }; 16461da177e4SLinus Torvalds 16471da177e4SLinus Torvalds if (!mapping_cap_writeback_dirty(inode->i_mapping)) 164849364ce2SAndrew Morton wbc.nr_to_write = 0; 16491da177e4SLinus Torvalds 16501da177e4SLinus Torvalds might_sleep(); 16514f8ad655SJan Kara return writeback_single_inode(inode, wb, &wbc); 16521da177e4SLinus Torvalds } 16531da177e4SLinus Torvalds EXPORT_SYMBOL(write_inode_now); 16541da177e4SLinus Torvalds 16551da177e4SLinus Torvalds /** 16561da177e4SLinus Torvalds * sync_inode - write an inode and its pages to disk. 16571da177e4SLinus Torvalds * @inode: the inode to sync 16581da177e4SLinus Torvalds * @wbc: controls the writeback mode 16591da177e4SLinus Torvalds * 16601da177e4SLinus Torvalds * sync_inode() will write an inode and its pages to disk. It will also 16611da177e4SLinus Torvalds * correctly update the inode on its superblock's dirty inode lists and will 16621da177e4SLinus Torvalds * update inode->i_state. 16631da177e4SLinus Torvalds * 16641da177e4SLinus Torvalds * The caller must have a ref on the inode. 16651da177e4SLinus Torvalds */ 16661da177e4SLinus Torvalds int sync_inode(struct inode *inode, struct writeback_control *wbc) 16671da177e4SLinus Torvalds { 16684f8ad655SJan Kara return writeback_single_inode(inode, &inode_to_bdi(inode)->wb, wbc); 16691da177e4SLinus Torvalds } 16701da177e4SLinus Torvalds EXPORT_SYMBOL(sync_inode); 1671c3765016SChristoph Hellwig 1672c3765016SChristoph Hellwig /** 1673c691b9d9SAndrew Morton * sync_inode_metadata - write an inode to disk 1674c3765016SChristoph Hellwig * @inode: the inode to sync 1675c3765016SChristoph Hellwig * @wait: wait for I/O to complete. 1676c3765016SChristoph Hellwig * 1677c691b9d9SAndrew Morton * Write an inode to disk and adjust its dirty state after completion. 1678c3765016SChristoph Hellwig * 1679c3765016SChristoph Hellwig * Note: only writes the actual inode, no associated data or other metadata. 1680c3765016SChristoph Hellwig */ 1681c3765016SChristoph Hellwig int sync_inode_metadata(struct inode *inode, int wait) 1682c3765016SChristoph Hellwig { 1683c3765016SChristoph Hellwig struct writeback_control wbc = { 1684c3765016SChristoph Hellwig .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE, 1685c3765016SChristoph Hellwig .nr_to_write = 0, /* metadata-only */ 1686c3765016SChristoph Hellwig }; 1687c3765016SChristoph Hellwig 1688c3765016SChristoph Hellwig return sync_inode(inode, &wbc); 1689c3765016SChristoph Hellwig } 1690c3765016SChristoph Hellwig EXPORT_SYMBOL(sync_inode_metadata); 1691