11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * fs/fs-writeback.c 31da177e4SLinus Torvalds * 41da177e4SLinus Torvalds * Copyright (C) 2002, Linus Torvalds. 51da177e4SLinus Torvalds * 61da177e4SLinus Torvalds * Contains all the functions related to writing back and waiting 71da177e4SLinus Torvalds * upon dirty inodes against superblocks, and writing back dirty 81da177e4SLinus Torvalds * pages against inodes. ie: data writeback. Writeout of the 91da177e4SLinus Torvalds * inode itself is not handled here. 101da177e4SLinus Torvalds * 11e1f8e874SFrancois Cami * 10Apr2002 Andrew Morton 121da177e4SLinus Torvalds * Split out of fs/inode.c 131da177e4SLinus Torvalds * Additions for address_space-based writeback 141da177e4SLinus Torvalds */ 151da177e4SLinus Torvalds 161da177e4SLinus Torvalds #include <linux/kernel.h> 17630d9c47SPaul Gortmaker #include <linux/export.h> 181da177e4SLinus Torvalds #include <linux/spinlock.h> 195a0e3ad6STejun Heo #include <linux/slab.h> 201da177e4SLinus Torvalds #include <linux/sched.h> 211da177e4SLinus Torvalds #include <linux/fs.h> 221da177e4SLinus Torvalds #include <linux/mm.h> 23bc31b86aSWu Fengguang #include <linux/pagemap.h> 2403ba3782SJens Axboe #include <linux/kthread.h> 251da177e4SLinus Torvalds #include <linux/writeback.h> 261da177e4SLinus Torvalds #include <linux/blkdev.h> 271da177e4SLinus Torvalds #include <linux/backing-dev.h> 28455b2864SDave Chinner #include <linux/tracepoint.h> 29719ea2fbSAl Viro #include <linux/device.h> 3007f3f05cSDavid Howells #include "internal.h" 311da177e4SLinus Torvalds 32d0bceac7SJens Axboe /* 33bc31b86aSWu Fengguang * 4MB minimal write chunk size 34bc31b86aSWu Fengguang */ 35bc31b86aSWu Fengguang #define MIN_WRITEBACK_PAGES (4096UL >> (PAGE_CACHE_SHIFT - 10)) 36bc31b86aSWu Fengguang 37bc31b86aSWu Fengguang /* 38c4a77a6cSJens Axboe * Passed into wb_writeback(), essentially a subset of writeback_control 39c4a77a6cSJens Axboe */ 4083ba7b07SChristoph Hellwig struct wb_writeback_work { 41c4a77a6cSJens Axboe long nr_pages; 42c4a77a6cSJens Axboe struct super_block *sb; 430dc83bd3SJan Kara unsigned long *older_than_this; 44c4a77a6cSJens Axboe enum writeback_sync_modes sync_mode; 456e6938b6SWu Fengguang unsigned int tagged_writepages:1; 4652957fe1SH Hartley Sweeten unsigned int for_kupdate:1; 4752957fe1SH Hartley Sweeten unsigned int range_cyclic:1; 4852957fe1SH Hartley Sweeten unsigned int for_background:1; 497747bd4bSDave Chinner unsigned int for_sync:1; /* sync(2) WB_SYNC_ALL writeback */ 500e175a18SCurt Wohlgemuth enum wb_reason reason; /* why was writeback initiated? */ 51c4a77a6cSJens Axboe 528010c3b6SJens Axboe struct list_head list; /* pending work list */ 5383ba7b07SChristoph Hellwig struct completion *done; /* set if the caller waits */ 5403ba3782SJens Axboe }; 5503ba3782SJens Axboe 56a2f48706STheodore Ts'o /* 57a2f48706STheodore Ts'o * If an inode is constantly having its pages dirtied, but then the 58a2f48706STheodore Ts'o * updates stop dirtytime_expire_interval seconds in the past, it's 59a2f48706STheodore Ts'o * possible for the worst case time between when an inode has its 60a2f48706STheodore Ts'o * timestamps updated and when they finally get written out to be two 61a2f48706STheodore Ts'o * dirtytime_expire_intervals. We set the default to 12 hours (in 62a2f48706STheodore Ts'o * seconds), which means most of the time inodes will have their 63a2f48706STheodore Ts'o * timestamps written to disk after 12 hours, but in the worst case a 64a2f48706STheodore Ts'o * few inodes might not their timestamps updated for 24 hours. 65a2f48706STheodore Ts'o */ 66a2f48706STheodore Ts'o unsigned int dirtytime_expire_interval = 12 * 60 * 60; 67a2f48706STheodore Ts'o 687ccf19a8SNick Piggin static inline struct inode *wb_inode(struct list_head *head) 697ccf19a8SNick Piggin { 707ccf19a8SNick Piggin return list_entry(head, struct inode, i_wb_list); 717ccf19a8SNick Piggin } 727ccf19a8SNick Piggin 7315eb77a0SWu Fengguang /* 7415eb77a0SWu Fengguang * Include the creation of the trace points after defining the 7515eb77a0SWu Fengguang * wb_writeback_work structure and inline functions so that the definition 7615eb77a0SWu Fengguang * remains local to this file. 7715eb77a0SWu Fengguang */ 7815eb77a0SWu Fengguang #define CREATE_TRACE_POINTS 7915eb77a0SWu Fengguang #include <trace/events/writeback.h> 8015eb77a0SWu Fengguang 81774016b2SSteven Whitehouse EXPORT_TRACEPOINT_SYMBOL_GPL(wbc_writepage); 82774016b2SSteven Whitehouse 83d6c10f1fSTejun Heo static bool wb_io_lists_populated(struct bdi_writeback *wb) 84d6c10f1fSTejun Heo { 85d6c10f1fSTejun Heo if (wb_has_dirty_io(wb)) { 86d6c10f1fSTejun Heo return false; 87d6c10f1fSTejun Heo } else { 88d6c10f1fSTejun Heo set_bit(WB_has_dirty_io, &wb->state); 8995a46c65STejun Heo WARN_ON_ONCE(!wb->avg_write_bandwidth); 90766a9d6eSTejun Heo atomic_long_add(wb->avg_write_bandwidth, 91766a9d6eSTejun Heo &wb->bdi->tot_write_bandwidth); 92d6c10f1fSTejun Heo return true; 93d6c10f1fSTejun Heo } 94d6c10f1fSTejun Heo } 95d6c10f1fSTejun Heo 96d6c10f1fSTejun Heo static void wb_io_lists_depopulated(struct bdi_writeback *wb) 97d6c10f1fSTejun Heo { 98d6c10f1fSTejun Heo if (wb_has_dirty_io(wb) && list_empty(&wb->b_dirty) && 99766a9d6eSTejun Heo list_empty(&wb->b_io) && list_empty(&wb->b_more_io)) { 100d6c10f1fSTejun Heo clear_bit(WB_has_dirty_io, &wb->state); 10195a46c65STejun Heo WARN_ON_ONCE(atomic_long_sub_return(wb->avg_write_bandwidth, 10295a46c65STejun Heo &wb->bdi->tot_write_bandwidth) < 0); 103766a9d6eSTejun Heo } 104d6c10f1fSTejun Heo } 105d6c10f1fSTejun Heo 106d6c10f1fSTejun Heo /** 107d6c10f1fSTejun Heo * inode_wb_list_move_locked - move an inode onto a bdi_writeback IO list 108d6c10f1fSTejun Heo * @inode: inode to be moved 109d6c10f1fSTejun Heo * @wb: target bdi_writeback 110d6c10f1fSTejun Heo * @head: one of @wb->b_{dirty|io|more_io} 111d6c10f1fSTejun Heo * 112d6c10f1fSTejun Heo * Move @inode->i_wb_list to @list of @wb and set %WB_has_dirty_io. 113d6c10f1fSTejun Heo * Returns %true if @inode is the first occupant of the !dirty_time IO 114d6c10f1fSTejun Heo * lists; otherwise, %false. 115d6c10f1fSTejun Heo */ 116d6c10f1fSTejun Heo static bool inode_wb_list_move_locked(struct inode *inode, 117d6c10f1fSTejun Heo struct bdi_writeback *wb, 118d6c10f1fSTejun Heo struct list_head *head) 119d6c10f1fSTejun Heo { 120d6c10f1fSTejun Heo assert_spin_locked(&wb->list_lock); 121d6c10f1fSTejun Heo 122d6c10f1fSTejun Heo list_move(&inode->i_wb_list, head); 123d6c10f1fSTejun Heo 124d6c10f1fSTejun Heo /* dirty_time doesn't count as dirty_io until expiration */ 125d6c10f1fSTejun Heo if (head != &wb->b_dirty_time) 126d6c10f1fSTejun Heo return wb_io_lists_populated(wb); 127d6c10f1fSTejun Heo 128d6c10f1fSTejun Heo wb_io_lists_depopulated(wb); 129d6c10f1fSTejun Heo return false; 130d6c10f1fSTejun Heo } 131d6c10f1fSTejun Heo 132d6c10f1fSTejun Heo /** 133d6c10f1fSTejun Heo * inode_wb_list_del_locked - remove an inode from its bdi_writeback IO list 134d6c10f1fSTejun Heo * @inode: inode to be removed 135d6c10f1fSTejun Heo * @wb: bdi_writeback @inode is being removed from 136d6c10f1fSTejun Heo * 137d6c10f1fSTejun Heo * Remove @inode which may be on one of @wb->b_{dirty|io|more_io} lists and 138d6c10f1fSTejun Heo * clear %WB_has_dirty_io if all are empty afterwards. 139d6c10f1fSTejun Heo */ 140d6c10f1fSTejun Heo static void inode_wb_list_del_locked(struct inode *inode, 141d6c10f1fSTejun Heo struct bdi_writeback *wb) 142d6c10f1fSTejun Heo { 143d6c10f1fSTejun Heo assert_spin_locked(&wb->list_lock); 144d6c10f1fSTejun Heo 145d6c10f1fSTejun Heo list_del_init(&inode->i_wb_list); 146d6c10f1fSTejun Heo wb_io_lists_depopulated(wb); 147d6c10f1fSTejun Heo } 148d6c10f1fSTejun Heo 149f0054bb1STejun Heo static void wb_wakeup(struct bdi_writeback *wb) 1505acda9d1SJan Kara { 151f0054bb1STejun Heo spin_lock_bh(&wb->work_lock); 152f0054bb1STejun Heo if (test_bit(WB_registered, &wb->state)) 153f0054bb1STejun Heo mod_delayed_work(bdi_wq, &wb->dwork, 0); 154f0054bb1STejun Heo spin_unlock_bh(&wb->work_lock); 1555acda9d1SJan Kara } 1565acda9d1SJan Kara 157f0054bb1STejun Heo static void wb_queue_work(struct bdi_writeback *wb, 1586585027aSJan Kara struct wb_writeback_work *work) 1596585027aSJan Kara { 160f0054bb1STejun Heo trace_writeback_queue(wb->bdi, work); 1616585027aSJan Kara 162f0054bb1STejun Heo spin_lock_bh(&wb->work_lock); 163f0054bb1STejun Heo if (!test_bit(WB_registered, &wb->state)) { 1645acda9d1SJan Kara if (work->done) 1655acda9d1SJan Kara complete(work->done); 1665acda9d1SJan Kara goto out_unlock; 1675acda9d1SJan Kara } 168f0054bb1STejun Heo list_add_tail(&work->list, &wb->work_list); 169f0054bb1STejun Heo mod_delayed_work(bdi_wq, &wb->dwork, 0); 1705acda9d1SJan Kara out_unlock: 171f0054bb1STejun Heo spin_unlock_bh(&wb->work_lock); 17203ba3782SJens Axboe } 1731da177e4SLinus Torvalds 174703c2708STejun Heo #ifdef CONFIG_CGROUP_WRITEBACK 175703c2708STejun Heo 176703c2708STejun Heo /** 177703c2708STejun Heo * inode_congested - test whether an inode is congested 178703c2708STejun Heo * @inode: inode to test for congestion 179703c2708STejun Heo * @cong_bits: mask of WB_[a]sync_congested bits to test 180703c2708STejun Heo * 181703c2708STejun Heo * Tests whether @inode is congested. @cong_bits is the mask of congestion 182703c2708STejun Heo * bits to test and the return value is the mask of set bits. 183703c2708STejun Heo * 184703c2708STejun Heo * If cgroup writeback is enabled for @inode, the congestion state is 185703c2708STejun Heo * determined by whether the cgwb (cgroup bdi_writeback) for the blkcg 186703c2708STejun Heo * associated with @inode is congested; otherwise, the root wb's congestion 187703c2708STejun Heo * state is used. 188703c2708STejun Heo */ 189703c2708STejun Heo int inode_congested(struct inode *inode, int cong_bits) 190703c2708STejun Heo { 191703c2708STejun Heo if (inode) { 192703c2708STejun Heo struct bdi_writeback *wb = inode_to_wb(inode); 193703c2708STejun Heo if (wb) 194703c2708STejun Heo return wb_congested(wb, cong_bits); 195703c2708STejun Heo } 196703c2708STejun Heo 197703c2708STejun Heo return wb_congested(&inode_to_bdi(inode)->wb, cong_bits); 198703c2708STejun Heo } 199703c2708STejun Heo EXPORT_SYMBOL_GPL(inode_congested); 200703c2708STejun Heo 201f2b65121STejun Heo /** 202f2b65121STejun Heo * wb_split_bdi_pages - split nr_pages to write according to bandwidth 203f2b65121STejun Heo * @wb: target bdi_writeback to split @nr_pages to 204f2b65121STejun Heo * @nr_pages: number of pages to write for the whole bdi 205f2b65121STejun Heo * 206f2b65121STejun Heo * Split @wb's portion of @nr_pages according to @wb's write bandwidth in 207f2b65121STejun Heo * relation to the total write bandwidth of all wb's w/ dirty inodes on 208f2b65121STejun Heo * @wb->bdi. 209f2b65121STejun Heo */ 210f2b65121STejun Heo static long wb_split_bdi_pages(struct bdi_writeback *wb, long nr_pages) 211f2b65121STejun Heo { 212f2b65121STejun Heo unsigned long this_bw = wb->avg_write_bandwidth; 213f2b65121STejun Heo unsigned long tot_bw = atomic_long_read(&wb->bdi->tot_write_bandwidth); 214f2b65121STejun Heo 215f2b65121STejun Heo if (nr_pages == LONG_MAX) 216f2b65121STejun Heo return LONG_MAX; 217f2b65121STejun Heo 218f2b65121STejun Heo /* 219f2b65121STejun Heo * This may be called on clean wb's and proportional distribution 220f2b65121STejun Heo * may not make sense, just use the original @nr_pages in those 221f2b65121STejun Heo * cases. In general, we wanna err on the side of writing more. 222f2b65121STejun Heo */ 223f2b65121STejun Heo if (!tot_bw || this_bw >= tot_bw) 224f2b65121STejun Heo return nr_pages; 225f2b65121STejun Heo else 226f2b65121STejun Heo return DIV_ROUND_UP_ULL((u64)nr_pages * this_bw, tot_bw); 227f2b65121STejun Heo } 228f2b65121STejun Heo 229f2b65121STejun Heo #else /* CONFIG_CGROUP_WRITEBACK */ 230f2b65121STejun Heo 231f2b65121STejun Heo static long wb_split_bdi_pages(struct bdi_writeback *wb, long nr_pages) 232f2b65121STejun Heo { 233f2b65121STejun Heo return nr_pages; 234f2b65121STejun Heo } 235f2b65121STejun Heo 236703c2708STejun Heo #endif /* CONFIG_CGROUP_WRITEBACK */ 237703c2708STejun Heo 238c00ddad3STejun Heo void wb_start_writeback(struct bdi_writeback *wb, long nr_pages, 239c00ddad3STejun Heo bool range_cyclic, enum wb_reason reason) 240b6e51316SJens Axboe { 241c00ddad3STejun Heo struct wb_writeback_work *work; 242c00ddad3STejun Heo 243c00ddad3STejun Heo if (!wb_has_dirty_io(wb)) 244c00ddad3STejun Heo return; 245c00ddad3STejun Heo 246c00ddad3STejun Heo /* 247c00ddad3STejun Heo * This is WB_SYNC_NONE writeback, so if allocation fails just 248c00ddad3STejun Heo * wakeup the thread for old dirty data writeback 249c00ddad3STejun Heo */ 250c00ddad3STejun Heo work = kzalloc(sizeof(*work), GFP_ATOMIC); 251c00ddad3STejun Heo if (!work) { 252c00ddad3STejun Heo trace_writeback_nowork(wb->bdi); 253c00ddad3STejun Heo wb_wakeup(wb); 254c00ddad3STejun Heo return; 255c00ddad3STejun Heo } 256c00ddad3STejun Heo 257c00ddad3STejun Heo work->sync_mode = WB_SYNC_NONE; 258c00ddad3STejun Heo work->nr_pages = nr_pages; 259c00ddad3STejun Heo work->range_cyclic = range_cyclic; 260c00ddad3STejun Heo work->reason = reason; 261c00ddad3STejun Heo 262c00ddad3STejun Heo wb_queue_work(wb, work); 263d3ddec76SWu Fengguang } 264d3ddec76SWu Fengguang 265c5444198SChristoph Hellwig /** 2669ecf4866STejun Heo * wb_start_background_writeback - start background writeback 2679ecf4866STejun Heo * @wb: bdi_writback to write from 268c5444198SChristoph Hellwig * 269c5444198SChristoph Hellwig * Description: 2706585027aSJan Kara * This makes sure WB_SYNC_NONE background writeback happens. When 2719ecf4866STejun Heo * this function returns, it is only guaranteed that for given wb 2726585027aSJan Kara * some IO is happening if we are over background dirty threshold. 2736585027aSJan Kara * Caller need not hold sb s_umount semaphore. 274c5444198SChristoph Hellwig */ 2759ecf4866STejun Heo void wb_start_background_writeback(struct bdi_writeback *wb) 276c5444198SChristoph Hellwig { 2776585027aSJan Kara /* 2786585027aSJan Kara * We just wake up the flusher thread. It will perform background 2796585027aSJan Kara * writeback as soon as there is no other work to do. 2806585027aSJan Kara */ 2819ecf4866STejun Heo trace_writeback_wake_background(wb->bdi); 2829ecf4866STejun Heo wb_wakeup(wb); 2831da177e4SLinus Torvalds } 2841da177e4SLinus Torvalds 2851da177e4SLinus Torvalds /* 286a66979abSDave Chinner * Remove the inode from the writeback list it is on. 287a66979abSDave Chinner */ 288a66979abSDave Chinner void inode_wb_list_del(struct inode *inode) 289a66979abSDave Chinner { 29052ebea74STejun Heo struct bdi_writeback *wb = inode_to_wb(inode); 291a66979abSDave Chinner 29252ebea74STejun Heo spin_lock(&wb->list_lock); 293d6c10f1fSTejun Heo inode_wb_list_del_locked(inode, wb); 29452ebea74STejun Heo spin_unlock(&wb->list_lock); 295f758eeabSChristoph Hellwig } 296a66979abSDave Chinner 297a66979abSDave Chinner /* 2986610a0bcSAndrew Morton * Redirty an inode: set its when-it-was dirtied timestamp and move it to the 2996610a0bcSAndrew Morton * furthest end of its superblock's dirty-inode list. 3006610a0bcSAndrew Morton * 3016610a0bcSAndrew Morton * Before stamping the inode's ->dirtied_when, we check to see whether it is 30266f3b8e2SJens Axboe * already the most-recently-dirtied inode on the b_dirty list. If that is 3036610a0bcSAndrew Morton * the case then the inode must have been redirtied while it was being written 3046610a0bcSAndrew Morton * out and we don't reset its dirtied_when. 3056610a0bcSAndrew Morton */ 306f758eeabSChristoph Hellwig static void redirty_tail(struct inode *inode, struct bdi_writeback *wb) 3076610a0bcSAndrew Morton { 30803ba3782SJens Axboe if (!list_empty(&wb->b_dirty)) { 30966f3b8e2SJens Axboe struct inode *tail; 3106610a0bcSAndrew Morton 3117ccf19a8SNick Piggin tail = wb_inode(wb->b_dirty.next); 31266f3b8e2SJens Axboe if (time_before(inode->dirtied_when, tail->dirtied_when)) 3136610a0bcSAndrew Morton inode->dirtied_when = jiffies; 3146610a0bcSAndrew Morton } 315d6c10f1fSTejun Heo inode_wb_list_move_locked(inode, wb, &wb->b_dirty); 3166610a0bcSAndrew Morton } 3176610a0bcSAndrew Morton 3186610a0bcSAndrew Morton /* 31966f3b8e2SJens Axboe * requeue inode for re-scanning after bdi->b_io list is exhausted. 320c986d1e2SAndrew Morton */ 321f758eeabSChristoph Hellwig static void requeue_io(struct inode *inode, struct bdi_writeback *wb) 322c986d1e2SAndrew Morton { 323d6c10f1fSTejun Heo inode_wb_list_move_locked(inode, wb, &wb->b_more_io); 324c986d1e2SAndrew Morton } 325c986d1e2SAndrew Morton 3261c0eeaf5SJoern Engel static void inode_sync_complete(struct inode *inode) 3271c0eeaf5SJoern Engel { 328365b94aeSJan Kara inode->i_state &= ~I_SYNC; 3294eff96ddSJan Kara /* If inode is clean an unused, put it into LRU now... */ 3304eff96ddSJan Kara inode_add_lru(inode); 331365b94aeSJan Kara /* Waiters must see I_SYNC cleared before being woken up */ 3321c0eeaf5SJoern Engel smp_mb(); 3331c0eeaf5SJoern Engel wake_up_bit(&inode->i_state, __I_SYNC); 3341c0eeaf5SJoern Engel } 3351c0eeaf5SJoern Engel 336d2caa3c5SJeff Layton static bool inode_dirtied_after(struct inode *inode, unsigned long t) 337d2caa3c5SJeff Layton { 338d2caa3c5SJeff Layton bool ret = time_after(inode->dirtied_when, t); 339d2caa3c5SJeff Layton #ifndef CONFIG_64BIT 340d2caa3c5SJeff Layton /* 341d2caa3c5SJeff Layton * For inodes being constantly redirtied, dirtied_when can get stuck. 342d2caa3c5SJeff Layton * It _appears_ to be in the future, but is actually in distant past. 343d2caa3c5SJeff Layton * This test is necessary to prevent such wrapped-around relative times 3445b0830cbSJens Axboe * from permanently stopping the whole bdi writeback. 345d2caa3c5SJeff Layton */ 346d2caa3c5SJeff Layton ret = ret && time_before_eq(inode->dirtied_when, jiffies); 347d2caa3c5SJeff Layton #endif 348d2caa3c5SJeff Layton return ret; 349d2caa3c5SJeff Layton } 350d2caa3c5SJeff Layton 3510ae45f63STheodore Ts'o #define EXPIRE_DIRTY_ATIME 0x0001 3520ae45f63STheodore Ts'o 353c986d1e2SAndrew Morton /* 3540e2f2b23SWang Sheng-Hui * Move expired (dirtied before work->older_than_this) dirty inodes from 355697e6fedSJan Kara * @delaying_queue to @dispatch_queue. 3562c136579SFengguang Wu */ 357e84d0a4fSWu Fengguang static int move_expired_inodes(struct list_head *delaying_queue, 3582c136579SFengguang Wu struct list_head *dispatch_queue, 3590ae45f63STheodore Ts'o int flags, 360ad4e38ddSCurt Wohlgemuth struct wb_writeback_work *work) 3612c136579SFengguang Wu { 3620ae45f63STheodore Ts'o unsigned long *older_than_this = NULL; 3630ae45f63STheodore Ts'o unsigned long expire_time; 3645c03449dSShaohua Li LIST_HEAD(tmp); 3655c03449dSShaohua Li struct list_head *pos, *node; 366cf137307SJens Axboe struct super_block *sb = NULL; 3675c03449dSShaohua Li struct inode *inode; 368cf137307SJens Axboe int do_sb_sort = 0; 369e84d0a4fSWu Fengguang int moved = 0; 3705c03449dSShaohua Li 3710ae45f63STheodore Ts'o if ((flags & EXPIRE_DIRTY_ATIME) == 0) 3720ae45f63STheodore Ts'o older_than_this = work->older_than_this; 373a2f48706STheodore Ts'o else if (!work->for_sync) { 374a2f48706STheodore Ts'o expire_time = jiffies - (dirtytime_expire_interval * HZ); 3750ae45f63STheodore Ts'o older_than_this = &expire_time; 3760ae45f63STheodore Ts'o } 3772c136579SFengguang Wu while (!list_empty(delaying_queue)) { 3787ccf19a8SNick Piggin inode = wb_inode(delaying_queue->prev); 3790ae45f63STheodore Ts'o if (older_than_this && 3800ae45f63STheodore Ts'o inode_dirtied_after(inode, *older_than_this)) 3812c136579SFengguang Wu break; 382a8855990SJan Kara list_move(&inode->i_wb_list, &tmp); 383a8855990SJan Kara moved++; 3840ae45f63STheodore Ts'o if (flags & EXPIRE_DIRTY_ATIME) 3850ae45f63STheodore Ts'o set_bit(__I_DIRTY_TIME_EXPIRED, &inode->i_state); 386a8855990SJan Kara if (sb_is_blkdev_sb(inode->i_sb)) 387a8855990SJan Kara continue; 388cf137307SJens Axboe if (sb && sb != inode->i_sb) 389cf137307SJens Axboe do_sb_sort = 1; 390cf137307SJens Axboe sb = inode->i_sb; 3915c03449dSShaohua Li } 3925c03449dSShaohua Li 393cf137307SJens Axboe /* just one sb in list, splice to dispatch_queue and we're done */ 394cf137307SJens Axboe if (!do_sb_sort) { 395cf137307SJens Axboe list_splice(&tmp, dispatch_queue); 396e84d0a4fSWu Fengguang goto out; 397cf137307SJens Axboe } 398cf137307SJens Axboe 3995c03449dSShaohua Li /* Move inodes from one superblock together */ 4005c03449dSShaohua Li while (!list_empty(&tmp)) { 4017ccf19a8SNick Piggin sb = wb_inode(tmp.prev)->i_sb; 4025c03449dSShaohua Li list_for_each_prev_safe(pos, node, &tmp) { 4037ccf19a8SNick Piggin inode = wb_inode(pos); 4045c03449dSShaohua Li if (inode->i_sb == sb) 4057ccf19a8SNick Piggin list_move(&inode->i_wb_list, dispatch_queue); 4062c136579SFengguang Wu } 4072c136579SFengguang Wu } 408e84d0a4fSWu Fengguang out: 409e84d0a4fSWu Fengguang return moved; 4105c03449dSShaohua Li } 4112c136579SFengguang Wu 4122c136579SFengguang Wu /* 4132c136579SFengguang Wu * Queue all expired dirty inodes for io, eldest first. 4144ea879b9SWu Fengguang * Before 4154ea879b9SWu Fengguang * newly dirtied b_dirty b_io b_more_io 4164ea879b9SWu Fengguang * =============> gf edc BA 4174ea879b9SWu Fengguang * After 4184ea879b9SWu Fengguang * newly dirtied b_dirty b_io b_more_io 4194ea879b9SWu Fengguang * =============> g fBAedc 4204ea879b9SWu Fengguang * | 4214ea879b9SWu Fengguang * +--> dequeue for IO 4222c136579SFengguang Wu */ 423ad4e38ddSCurt Wohlgemuth static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work) 4242c136579SFengguang Wu { 425e84d0a4fSWu Fengguang int moved; 4260ae45f63STheodore Ts'o 427f758eeabSChristoph Hellwig assert_spin_locked(&wb->list_lock); 4284ea879b9SWu Fengguang list_splice_init(&wb->b_more_io, &wb->b_io); 4290ae45f63STheodore Ts'o moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, 0, work); 4300ae45f63STheodore Ts'o moved += move_expired_inodes(&wb->b_dirty_time, &wb->b_io, 4310ae45f63STheodore Ts'o EXPIRE_DIRTY_ATIME, work); 432d6c10f1fSTejun Heo if (moved) 433d6c10f1fSTejun Heo wb_io_lists_populated(wb); 434ad4e38ddSCurt Wohlgemuth trace_writeback_queue_io(wb, work, moved); 43566f3b8e2SJens Axboe } 43666f3b8e2SJens Axboe 437a9185b41SChristoph Hellwig static int write_inode(struct inode *inode, struct writeback_control *wbc) 43866f3b8e2SJens Axboe { 4399fb0a7daSTejun Heo int ret; 4409fb0a7daSTejun Heo 4419fb0a7daSTejun Heo if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) { 4429fb0a7daSTejun Heo trace_writeback_write_inode_start(inode, wbc); 4439fb0a7daSTejun Heo ret = inode->i_sb->s_op->write_inode(inode, wbc); 4449fb0a7daSTejun Heo trace_writeback_write_inode(inode, wbc); 4459fb0a7daSTejun Heo return ret; 4469fb0a7daSTejun Heo } 44703ba3782SJens Axboe return 0; 44866f3b8e2SJens Axboe } 44908d8e974SFengguang Wu 4502c136579SFengguang Wu /* 451169ebd90SJan Kara * Wait for writeback on an inode to complete. Called with i_lock held. 452169ebd90SJan Kara * Caller must make sure inode cannot go away when we drop i_lock. 45301c03194SChristoph Hellwig */ 454169ebd90SJan Kara static void __inode_wait_for_writeback(struct inode *inode) 455169ebd90SJan Kara __releases(inode->i_lock) 456169ebd90SJan Kara __acquires(inode->i_lock) 45701c03194SChristoph Hellwig { 45801c03194SChristoph Hellwig DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); 45901c03194SChristoph Hellwig wait_queue_head_t *wqh; 46001c03194SChristoph Hellwig 46101c03194SChristoph Hellwig wqh = bit_waitqueue(&inode->i_state, __I_SYNC); 46258a9d3d8SRichard Kennedy while (inode->i_state & I_SYNC) { 463250df6edSDave Chinner spin_unlock(&inode->i_lock); 46474316201SNeilBrown __wait_on_bit(wqh, &wq, bit_wait, 46574316201SNeilBrown TASK_UNINTERRUPTIBLE); 466250df6edSDave Chinner spin_lock(&inode->i_lock); 46758a9d3d8SRichard Kennedy } 46801c03194SChristoph Hellwig } 46901c03194SChristoph Hellwig 47001c03194SChristoph Hellwig /* 471169ebd90SJan Kara * Wait for writeback on an inode to complete. Caller must have inode pinned. 472169ebd90SJan Kara */ 473169ebd90SJan Kara void inode_wait_for_writeback(struct inode *inode) 474169ebd90SJan Kara { 475169ebd90SJan Kara spin_lock(&inode->i_lock); 476169ebd90SJan Kara __inode_wait_for_writeback(inode); 477169ebd90SJan Kara spin_unlock(&inode->i_lock); 478169ebd90SJan Kara } 479169ebd90SJan Kara 480169ebd90SJan Kara /* 481169ebd90SJan Kara * Sleep until I_SYNC is cleared. This function must be called with i_lock 482169ebd90SJan Kara * held and drops it. It is aimed for callers not holding any inode reference 483169ebd90SJan Kara * so once i_lock is dropped, inode can go away. 484169ebd90SJan Kara */ 485169ebd90SJan Kara static void inode_sleep_on_writeback(struct inode *inode) 486169ebd90SJan Kara __releases(inode->i_lock) 487169ebd90SJan Kara { 488169ebd90SJan Kara DEFINE_WAIT(wait); 489169ebd90SJan Kara wait_queue_head_t *wqh = bit_waitqueue(&inode->i_state, __I_SYNC); 490169ebd90SJan Kara int sleep; 491169ebd90SJan Kara 492169ebd90SJan Kara prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); 493169ebd90SJan Kara sleep = inode->i_state & I_SYNC; 494169ebd90SJan Kara spin_unlock(&inode->i_lock); 495169ebd90SJan Kara if (sleep) 496169ebd90SJan Kara schedule(); 497169ebd90SJan Kara finish_wait(wqh, &wait); 498169ebd90SJan Kara } 499169ebd90SJan Kara 500169ebd90SJan Kara /* 501ccb26b5aSJan Kara * Find proper writeback list for the inode depending on its current state and 502ccb26b5aSJan Kara * possibly also change of its state while we were doing writeback. Here we 503ccb26b5aSJan Kara * handle things such as livelock prevention or fairness of writeback among 504ccb26b5aSJan Kara * inodes. This function can be called only by flusher thread - noone else 505ccb26b5aSJan Kara * processes all inodes in writeback lists and requeueing inodes behind flusher 506ccb26b5aSJan Kara * thread's back can have unexpected consequences. 507ccb26b5aSJan Kara */ 508ccb26b5aSJan Kara static void requeue_inode(struct inode *inode, struct bdi_writeback *wb, 509ccb26b5aSJan Kara struct writeback_control *wbc) 510ccb26b5aSJan Kara { 511ccb26b5aSJan Kara if (inode->i_state & I_FREEING) 512ccb26b5aSJan Kara return; 513ccb26b5aSJan Kara 514ccb26b5aSJan Kara /* 515ccb26b5aSJan Kara * Sync livelock prevention. Each inode is tagged and synced in one 516ccb26b5aSJan Kara * shot. If still dirty, it will be redirty_tail()'ed below. Update 517ccb26b5aSJan Kara * the dirty time to prevent enqueue and sync it again. 518ccb26b5aSJan Kara */ 519ccb26b5aSJan Kara if ((inode->i_state & I_DIRTY) && 520ccb26b5aSJan Kara (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)) 521ccb26b5aSJan Kara inode->dirtied_when = jiffies; 522ccb26b5aSJan Kara 5234f8ad655SJan Kara if (wbc->pages_skipped) { 5244f8ad655SJan Kara /* 5254f8ad655SJan Kara * writeback is not making progress due to locked 5264f8ad655SJan Kara * buffers. Skip this inode for now. 5274f8ad655SJan Kara */ 5284f8ad655SJan Kara redirty_tail(inode, wb); 5294f8ad655SJan Kara return; 5304f8ad655SJan Kara } 5314f8ad655SJan Kara 532ccb26b5aSJan Kara if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) { 533ccb26b5aSJan Kara /* 534ccb26b5aSJan Kara * We didn't write back all the pages. nfs_writepages() 535ccb26b5aSJan Kara * sometimes bales out without doing anything. 536ccb26b5aSJan Kara */ 537ccb26b5aSJan Kara if (wbc->nr_to_write <= 0) { 538ccb26b5aSJan Kara /* Slice used up. Queue for next turn. */ 539ccb26b5aSJan Kara requeue_io(inode, wb); 540ccb26b5aSJan Kara } else { 541ccb26b5aSJan Kara /* 542ccb26b5aSJan Kara * Writeback blocked by something other than 543ccb26b5aSJan Kara * congestion. Delay the inode for some time to 544ccb26b5aSJan Kara * avoid spinning on the CPU (100% iowait) 545ccb26b5aSJan Kara * retrying writeback of the dirty page/inode 546ccb26b5aSJan Kara * that cannot be performed immediately. 547ccb26b5aSJan Kara */ 548ccb26b5aSJan Kara redirty_tail(inode, wb); 549ccb26b5aSJan Kara } 550ccb26b5aSJan Kara } else if (inode->i_state & I_DIRTY) { 551ccb26b5aSJan Kara /* 552ccb26b5aSJan Kara * Filesystems can dirty the inode during writeback operations, 553ccb26b5aSJan Kara * such as delayed allocation during submission or metadata 554ccb26b5aSJan Kara * updates after data IO completion. 555ccb26b5aSJan Kara */ 556ccb26b5aSJan Kara redirty_tail(inode, wb); 5570ae45f63STheodore Ts'o } else if (inode->i_state & I_DIRTY_TIME) { 558a2f48706STheodore Ts'o inode->dirtied_when = jiffies; 559d6c10f1fSTejun Heo inode_wb_list_move_locked(inode, wb, &wb->b_dirty_time); 560ccb26b5aSJan Kara } else { 561ccb26b5aSJan Kara /* The inode is clean. Remove from writeback lists. */ 562d6c10f1fSTejun Heo inode_wb_list_del_locked(inode, wb); 563ccb26b5aSJan Kara } 564ccb26b5aSJan Kara } 565ccb26b5aSJan Kara 566ccb26b5aSJan Kara /* 5674f8ad655SJan Kara * Write out an inode and its dirty pages. Do not update the writeback list 5684f8ad655SJan Kara * linkage. That is left to the caller. The caller is also responsible for 5694f8ad655SJan Kara * setting I_SYNC flag and calling inode_sync_complete() to clear it. 5701da177e4SLinus Torvalds */ 5711da177e4SLinus Torvalds static int 572cd8ed2a4SYan Hong __writeback_single_inode(struct inode *inode, struct writeback_control *wbc) 5731da177e4SLinus Torvalds { 5741da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 575251d6a47SWu Fengguang long nr_to_write = wbc->nr_to_write; 57601c03194SChristoph Hellwig unsigned dirty; 5771da177e4SLinus Torvalds int ret; 5781da177e4SLinus Torvalds 5794f8ad655SJan Kara WARN_ON(!(inode->i_state & I_SYNC)); 5801da177e4SLinus Torvalds 5819fb0a7daSTejun Heo trace_writeback_single_inode_start(inode, wbc, nr_to_write); 5829fb0a7daSTejun Heo 5831da177e4SLinus Torvalds ret = do_writepages(mapping, wbc); 5841da177e4SLinus Torvalds 58526821ed4SChristoph Hellwig /* 58626821ed4SChristoph Hellwig * Make sure to wait on the data before writing out the metadata. 58726821ed4SChristoph Hellwig * This is important for filesystems that modify metadata on data 5887747bd4bSDave Chinner * I/O completion. We don't do it for sync(2) writeback because it has a 5897747bd4bSDave Chinner * separate, external IO completion path and ->sync_fs for guaranteeing 5907747bd4bSDave Chinner * inode metadata is written back correctly. 59126821ed4SChristoph Hellwig */ 5927747bd4bSDave Chinner if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) { 59326821ed4SChristoph Hellwig int err = filemap_fdatawait(mapping); 5941da177e4SLinus Torvalds if (ret == 0) 5951da177e4SLinus Torvalds ret = err; 5961da177e4SLinus Torvalds } 5971da177e4SLinus Torvalds 5985547e8aaSDmitry Monakhov /* 5995547e8aaSDmitry Monakhov * Some filesystems may redirty the inode during the writeback 6005547e8aaSDmitry Monakhov * due to delalloc, clear dirty metadata flags right before 6015547e8aaSDmitry Monakhov * write_inode() 6025547e8aaSDmitry Monakhov */ 603250df6edSDave Chinner spin_lock(&inode->i_lock); 6049c6ac78eSTejun Heo 6055547e8aaSDmitry Monakhov dirty = inode->i_state & I_DIRTY; 606a2f48706STheodore Ts'o if (inode->i_state & I_DIRTY_TIME) { 607a2f48706STheodore Ts'o if ((dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) || 608a2f48706STheodore Ts'o unlikely(inode->i_state & I_DIRTY_TIME_EXPIRED) || 609a2f48706STheodore Ts'o unlikely(time_after(jiffies, 610a2f48706STheodore Ts'o (inode->dirtied_time_when + 611a2f48706STheodore Ts'o dirtytime_expire_interval * HZ)))) { 6120ae45f63STheodore Ts'o dirty |= I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED; 6130ae45f63STheodore Ts'o trace_writeback_lazytime(inode); 6140ae45f63STheodore Ts'o } 615a2f48706STheodore Ts'o } else 616a2f48706STheodore Ts'o inode->i_state &= ~I_DIRTY_TIME_EXPIRED; 6170ae45f63STheodore Ts'o inode->i_state &= ~dirty; 6189c6ac78eSTejun Heo 6199c6ac78eSTejun Heo /* 6209c6ac78eSTejun Heo * Paired with smp_mb() in __mark_inode_dirty(). This allows 6219c6ac78eSTejun Heo * __mark_inode_dirty() to test i_state without grabbing i_lock - 6229c6ac78eSTejun Heo * either they see the I_DIRTY bits cleared or we see the dirtied 6239c6ac78eSTejun Heo * inode. 6249c6ac78eSTejun Heo * 6259c6ac78eSTejun Heo * I_DIRTY_PAGES is always cleared together above even if @mapping 6269c6ac78eSTejun Heo * still has dirty pages. The flag is reinstated after smp_mb() if 6279c6ac78eSTejun Heo * necessary. This guarantees that either __mark_inode_dirty() 6289c6ac78eSTejun Heo * sees clear I_DIRTY_PAGES or we see PAGECACHE_TAG_DIRTY. 6299c6ac78eSTejun Heo */ 6309c6ac78eSTejun Heo smp_mb(); 6319c6ac78eSTejun Heo 6329c6ac78eSTejun Heo if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) 6339c6ac78eSTejun Heo inode->i_state |= I_DIRTY_PAGES; 6349c6ac78eSTejun Heo 635250df6edSDave Chinner spin_unlock(&inode->i_lock); 6369c6ac78eSTejun Heo 6370ae45f63STheodore Ts'o if (dirty & I_DIRTY_TIME) 6380ae45f63STheodore Ts'o mark_inode_dirty_sync(inode); 63926821ed4SChristoph Hellwig /* Don't write the inode if only I_DIRTY_PAGES was set */ 6400ae45f63STheodore Ts'o if (dirty & ~I_DIRTY_PAGES) { 641a9185b41SChristoph Hellwig int err = write_inode(inode, wbc); 6421da177e4SLinus Torvalds if (ret == 0) 6431da177e4SLinus Torvalds ret = err; 6441da177e4SLinus Torvalds } 6454f8ad655SJan Kara trace_writeback_single_inode(inode, wbc, nr_to_write); 6464f8ad655SJan Kara return ret; 6474f8ad655SJan Kara } 6484f8ad655SJan Kara 6494f8ad655SJan Kara /* 6504f8ad655SJan Kara * Write out an inode's dirty pages. Either the caller has an active reference 6514f8ad655SJan Kara * on the inode or the inode has I_WILL_FREE set. 6524f8ad655SJan Kara * 6534f8ad655SJan Kara * This function is designed to be called for writing back one inode which 6544f8ad655SJan Kara * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode() 6554f8ad655SJan Kara * and does more profound writeback list handling in writeback_sb_inodes(). 6564f8ad655SJan Kara */ 6574f8ad655SJan Kara static int 6584f8ad655SJan Kara writeback_single_inode(struct inode *inode, struct bdi_writeback *wb, 6594f8ad655SJan Kara struct writeback_control *wbc) 6604f8ad655SJan Kara { 6614f8ad655SJan Kara int ret = 0; 6624f8ad655SJan Kara 6634f8ad655SJan Kara spin_lock(&inode->i_lock); 6644f8ad655SJan Kara if (!atomic_read(&inode->i_count)) 6654f8ad655SJan Kara WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); 6664f8ad655SJan Kara else 6674f8ad655SJan Kara WARN_ON(inode->i_state & I_WILL_FREE); 6684f8ad655SJan Kara 6694f8ad655SJan Kara if (inode->i_state & I_SYNC) { 6704f8ad655SJan Kara if (wbc->sync_mode != WB_SYNC_ALL) 6714f8ad655SJan Kara goto out; 6724f8ad655SJan Kara /* 673169ebd90SJan Kara * It's a data-integrity sync. We must wait. Since callers hold 674169ebd90SJan Kara * inode reference or inode has I_WILL_FREE set, it cannot go 675169ebd90SJan Kara * away under us. 6764f8ad655SJan Kara */ 677169ebd90SJan Kara __inode_wait_for_writeback(inode); 6784f8ad655SJan Kara } 6794f8ad655SJan Kara WARN_ON(inode->i_state & I_SYNC); 6804f8ad655SJan Kara /* 681f9b0e058SJan Kara * Skip inode if it is clean and we have no outstanding writeback in 682f9b0e058SJan Kara * WB_SYNC_ALL mode. We don't want to mess with writeback lists in this 683f9b0e058SJan Kara * function since flusher thread may be doing for example sync in 684f9b0e058SJan Kara * parallel and if we move the inode, it could get skipped. So here we 685f9b0e058SJan Kara * make sure inode is on some writeback list and leave it there unless 686f9b0e058SJan Kara * we have completely cleaned the inode. 6874f8ad655SJan Kara */ 6880ae45f63STheodore Ts'o if (!(inode->i_state & I_DIRTY_ALL) && 689f9b0e058SJan Kara (wbc->sync_mode != WB_SYNC_ALL || 690f9b0e058SJan Kara !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK))) 6914f8ad655SJan Kara goto out; 6924f8ad655SJan Kara inode->i_state |= I_SYNC; 6934f8ad655SJan Kara spin_unlock(&inode->i_lock); 6944f8ad655SJan Kara 695cd8ed2a4SYan Hong ret = __writeback_single_inode(inode, wbc); 6961da177e4SLinus Torvalds 697f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 698250df6edSDave Chinner spin_lock(&inode->i_lock); 6994f8ad655SJan Kara /* 7004f8ad655SJan Kara * If inode is clean, remove it from writeback lists. Otherwise don't 7014f8ad655SJan Kara * touch it. See comment above for explanation. 7024f8ad655SJan Kara */ 7030ae45f63STheodore Ts'o if (!(inode->i_state & I_DIRTY_ALL)) 704d6c10f1fSTejun Heo inode_wb_list_del_locked(inode, wb); 7054f8ad655SJan Kara spin_unlock(&wb->list_lock); 7061c0eeaf5SJoern Engel inode_sync_complete(inode); 7074f8ad655SJan Kara out: 7084f8ad655SJan Kara spin_unlock(&inode->i_lock); 7091da177e4SLinus Torvalds return ret; 7101da177e4SLinus Torvalds } 7111da177e4SLinus Torvalds 712a88a341aSTejun Heo static long writeback_chunk_size(struct bdi_writeback *wb, 7131a12d8bdSWu Fengguang struct wb_writeback_work *work) 714d46db3d5SWu Fengguang { 715d46db3d5SWu Fengguang long pages; 716d46db3d5SWu Fengguang 717d46db3d5SWu Fengguang /* 718d46db3d5SWu Fengguang * WB_SYNC_ALL mode does livelock avoidance by syncing dirty 719d46db3d5SWu Fengguang * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX 720d46db3d5SWu Fengguang * here avoids calling into writeback_inodes_wb() more than once. 721d46db3d5SWu Fengguang * 722d46db3d5SWu Fengguang * The intended call sequence for WB_SYNC_ALL writeback is: 723d46db3d5SWu Fengguang * 724d46db3d5SWu Fengguang * wb_writeback() 725d46db3d5SWu Fengguang * writeback_sb_inodes() <== called only once 726d46db3d5SWu Fengguang * write_cache_pages() <== called once for each inode 727d46db3d5SWu Fengguang * (quickly) tag currently dirty pages 728d46db3d5SWu Fengguang * (maybe slowly) sync all tagged pages 729d46db3d5SWu Fengguang */ 730d46db3d5SWu Fengguang if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages) 731d46db3d5SWu Fengguang pages = LONG_MAX; 7321a12d8bdSWu Fengguang else { 733a88a341aSTejun Heo pages = min(wb->avg_write_bandwidth / 2, 7341a12d8bdSWu Fengguang global_dirty_limit / DIRTY_SCOPE); 7351a12d8bdSWu Fengguang pages = min(pages, work->nr_pages); 7361a12d8bdSWu Fengguang pages = round_down(pages + MIN_WRITEBACK_PAGES, 7371a12d8bdSWu Fengguang MIN_WRITEBACK_PAGES); 7381a12d8bdSWu Fengguang } 739d46db3d5SWu Fengguang 740d46db3d5SWu Fengguang return pages; 741d46db3d5SWu Fengguang } 742d46db3d5SWu Fengguang 74303ba3782SJens Axboe /* 744f11c9c5cSEdward Shishkin * Write a portion of b_io inodes which belong to @sb. 745edadfb10SChristoph Hellwig * 746d46db3d5SWu Fengguang * Return the number of pages and/or inodes written. 747f11c9c5cSEdward Shishkin */ 748d46db3d5SWu Fengguang static long writeback_sb_inodes(struct super_block *sb, 749d46db3d5SWu Fengguang struct bdi_writeback *wb, 750d46db3d5SWu Fengguang struct wb_writeback_work *work) 75103ba3782SJens Axboe { 752d46db3d5SWu Fengguang struct writeback_control wbc = { 753d46db3d5SWu Fengguang .sync_mode = work->sync_mode, 754d46db3d5SWu Fengguang .tagged_writepages = work->tagged_writepages, 755d46db3d5SWu Fengguang .for_kupdate = work->for_kupdate, 756d46db3d5SWu Fengguang .for_background = work->for_background, 7577747bd4bSDave Chinner .for_sync = work->for_sync, 758d46db3d5SWu Fengguang .range_cyclic = work->range_cyclic, 759d46db3d5SWu Fengguang .range_start = 0, 760d46db3d5SWu Fengguang .range_end = LLONG_MAX, 761d46db3d5SWu Fengguang }; 762d46db3d5SWu Fengguang unsigned long start_time = jiffies; 763d46db3d5SWu Fengguang long write_chunk; 764d46db3d5SWu Fengguang long wrote = 0; /* count both pages and inodes */ 765d46db3d5SWu Fengguang 76603ba3782SJens Axboe while (!list_empty(&wb->b_io)) { 7677ccf19a8SNick Piggin struct inode *inode = wb_inode(wb->b_io.prev); 768edadfb10SChristoph Hellwig 769edadfb10SChristoph Hellwig if (inode->i_sb != sb) { 770d46db3d5SWu Fengguang if (work->sb) { 771edadfb10SChristoph Hellwig /* 772edadfb10SChristoph Hellwig * We only want to write back data for this 773edadfb10SChristoph Hellwig * superblock, move all inodes not belonging 774edadfb10SChristoph Hellwig * to it back onto the dirty list. 775edadfb10SChristoph Hellwig */ 776f758eeabSChristoph Hellwig redirty_tail(inode, wb); 77766f3b8e2SJens Axboe continue; 77866f3b8e2SJens Axboe } 779edadfb10SChristoph Hellwig 780edadfb10SChristoph Hellwig /* 781edadfb10SChristoph Hellwig * The inode belongs to a different superblock. 782edadfb10SChristoph Hellwig * Bounce back to the caller to unpin this and 783edadfb10SChristoph Hellwig * pin the next superblock. 784edadfb10SChristoph Hellwig */ 785d46db3d5SWu Fengguang break; 786edadfb10SChristoph Hellwig } 787edadfb10SChristoph Hellwig 7889843b76aSChristoph Hellwig /* 789331cbdeeSWanpeng Li * Don't bother with new inodes or inodes being freed, first 790331cbdeeSWanpeng Li * kind does not need periodic writeout yet, and for the latter 7919843b76aSChristoph Hellwig * kind writeout is handled by the freer. 7929843b76aSChristoph Hellwig */ 793250df6edSDave Chinner spin_lock(&inode->i_lock); 7949843b76aSChristoph Hellwig if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { 795250df6edSDave Chinner spin_unlock(&inode->i_lock); 796fcc5c222SWu Fengguang redirty_tail(inode, wb); 7977ef0d737SNick Piggin continue; 7987ef0d737SNick Piggin } 799cc1676d9SJan Kara if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) { 800cc1676d9SJan Kara /* 801cc1676d9SJan Kara * If this inode is locked for writeback and we are not 802cc1676d9SJan Kara * doing writeback-for-data-integrity, move it to 803cc1676d9SJan Kara * b_more_io so that writeback can proceed with the 804cc1676d9SJan Kara * other inodes on s_io. 805cc1676d9SJan Kara * 806cc1676d9SJan Kara * We'll have another go at writing back this inode 807cc1676d9SJan Kara * when we completed a full scan of b_io. 808cc1676d9SJan Kara */ 809cc1676d9SJan Kara spin_unlock(&inode->i_lock); 810cc1676d9SJan Kara requeue_io(inode, wb); 811cc1676d9SJan Kara trace_writeback_sb_inodes_requeue(inode); 812cc1676d9SJan Kara continue; 813cc1676d9SJan Kara } 814f0d07b7fSJan Kara spin_unlock(&wb->list_lock); 815f0d07b7fSJan Kara 8164f8ad655SJan Kara /* 8174f8ad655SJan Kara * We already requeued the inode if it had I_SYNC set and we 8184f8ad655SJan Kara * are doing WB_SYNC_NONE writeback. So this catches only the 8194f8ad655SJan Kara * WB_SYNC_ALL case. 8204f8ad655SJan Kara */ 821169ebd90SJan Kara if (inode->i_state & I_SYNC) { 822169ebd90SJan Kara /* Wait for I_SYNC. This function drops i_lock... */ 823169ebd90SJan Kara inode_sleep_on_writeback(inode); 824169ebd90SJan Kara /* Inode may be gone, start again */ 825ead188f9SJan Kara spin_lock(&wb->list_lock); 826169ebd90SJan Kara continue; 827169ebd90SJan Kara } 8284f8ad655SJan Kara inode->i_state |= I_SYNC; 8294f8ad655SJan Kara spin_unlock(&inode->i_lock); 830169ebd90SJan Kara 831a88a341aSTejun Heo write_chunk = writeback_chunk_size(wb, work); 832d46db3d5SWu Fengguang wbc.nr_to_write = write_chunk; 833d46db3d5SWu Fengguang wbc.pages_skipped = 0; 834250df6edSDave Chinner 835169ebd90SJan Kara /* 836169ebd90SJan Kara * We use I_SYNC to pin the inode in memory. While it is set 837169ebd90SJan Kara * evict_inode() will wait so the inode cannot be freed. 838169ebd90SJan Kara */ 839cd8ed2a4SYan Hong __writeback_single_inode(inode, &wbc); 840d46db3d5SWu Fengguang 841d46db3d5SWu Fengguang work->nr_pages -= write_chunk - wbc.nr_to_write; 842d46db3d5SWu Fengguang wrote += write_chunk - wbc.nr_to_write; 8434f8ad655SJan Kara spin_lock(&wb->list_lock); 8444f8ad655SJan Kara spin_lock(&inode->i_lock); 8450ae45f63STheodore Ts'o if (!(inode->i_state & I_DIRTY_ALL)) 846d46db3d5SWu Fengguang wrote++; 8474f8ad655SJan Kara requeue_inode(inode, wb, &wbc); 8484f8ad655SJan Kara inode_sync_complete(inode); 8490f1b1fd8SDave Chinner spin_unlock(&inode->i_lock); 850169ebd90SJan Kara cond_resched_lock(&wb->list_lock); 851d46db3d5SWu Fengguang /* 852d46db3d5SWu Fengguang * bail out to wb_writeback() often enough to check 853d46db3d5SWu Fengguang * background threshold and other termination conditions. 854d46db3d5SWu Fengguang */ 855d46db3d5SWu Fengguang if (wrote) { 856d46db3d5SWu Fengguang if (time_is_before_jiffies(start_time + HZ / 10UL)) 857d46db3d5SWu Fengguang break; 858d46db3d5SWu Fengguang if (work->nr_pages <= 0) 859d46db3d5SWu Fengguang break; 8601da177e4SLinus Torvalds } 8618bc3be27SFengguang Wu } 862d46db3d5SWu Fengguang return wrote; 863f11c9c5cSEdward Shishkin } 86438f21977SNick Piggin 865d46db3d5SWu Fengguang static long __writeback_inodes_wb(struct bdi_writeback *wb, 866d46db3d5SWu Fengguang struct wb_writeback_work *work) 867f11c9c5cSEdward Shishkin { 868d46db3d5SWu Fengguang unsigned long start_time = jiffies; 869d46db3d5SWu Fengguang long wrote = 0; 870f11c9c5cSEdward Shishkin 871f11c9c5cSEdward Shishkin while (!list_empty(&wb->b_io)) { 8727ccf19a8SNick Piggin struct inode *inode = wb_inode(wb->b_io.prev); 873f11c9c5cSEdward Shishkin struct super_block *sb = inode->i_sb; 874f11c9c5cSEdward Shishkin 875eb6ef3dfSKonstantin Khlebnikov if (!trylock_super(sb)) { 8760e995816SWu Fengguang /* 877eb6ef3dfSKonstantin Khlebnikov * trylock_super() may fail consistently due to 8780e995816SWu Fengguang * s_umount being grabbed by someone else. Don't use 8790e995816SWu Fengguang * requeue_io() to avoid busy retrying the inode/sb. 8800e995816SWu Fengguang */ 8810e995816SWu Fengguang redirty_tail(inode, wb); 882d19de7edSChristoph Hellwig continue; 883334132aeSChristoph Hellwig } 884d46db3d5SWu Fengguang wrote += writeback_sb_inodes(sb, wb, work); 885eb6ef3dfSKonstantin Khlebnikov up_read(&sb->s_umount); 886f11c9c5cSEdward Shishkin 887d46db3d5SWu Fengguang /* refer to the same tests at the end of writeback_sb_inodes */ 888d46db3d5SWu Fengguang if (wrote) { 889d46db3d5SWu Fengguang if (time_is_before_jiffies(start_time + HZ / 10UL)) 890d46db3d5SWu Fengguang break; 891d46db3d5SWu Fengguang if (work->nr_pages <= 0) 892f11c9c5cSEdward Shishkin break; 893f11c9c5cSEdward Shishkin } 894d46db3d5SWu Fengguang } 89566f3b8e2SJens Axboe /* Leave any unwritten inodes on b_io */ 896d46db3d5SWu Fengguang return wrote; 89766f3b8e2SJens Axboe } 89866f3b8e2SJens Axboe 8997d9f073bSWanpeng Li static long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages, 9000e175a18SCurt Wohlgemuth enum wb_reason reason) 901edadfb10SChristoph Hellwig { 902d46db3d5SWu Fengguang struct wb_writeback_work work = { 903d46db3d5SWu Fengguang .nr_pages = nr_pages, 904d46db3d5SWu Fengguang .sync_mode = WB_SYNC_NONE, 905d46db3d5SWu Fengguang .range_cyclic = 1, 9060e175a18SCurt Wohlgemuth .reason = reason, 907d46db3d5SWu Fengguang }; 908edadfb10SChristoph Hellwig 909f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 910424b351fSWu Fengguang if (list_empty(&wb->b_io)) 911ad4e38ddSCurt Wohlgemuth queue_io(wb, &work); 912d46db3d5SWu Fengguang __writeback_inodes_wb(wb, &work); 913f758eeabSChristoph Hellwig spin_unlock(&wb->list_lock); 914edadfb10SChristoph Hellwig 915d46db3d5SWu Fengguang return nr_pages - work.nr_pages; 91666f3b8e2SJens Axboe } 91766f3b8e2SJens Axboe 918a88a341aSTejun Heo static bool over_bground_thresh(struct bdi_writeback *wb) 91903ba3782SJens Axboe { 92003ba3782SJens Axboe unsigned long background_thresh, dirty_thresh; 92103ba3782SJens Axboe 92216c4042fSWu Fengguang global_dirty_limits(&background_thresh, &dirty_thresh); 92303ba3782SJens Axboe 924b00949aaSWu Fengguang if (global_page_state(NR_FILE_DIRTY) + 925b00949aaSWu Fengguang global_page_state(NR_UNSTABLE_NFS) > background_thresh) 926b00949aaSWu Fengguang return true; 927b00949aaSWu Fengguang 928a88a341aSTejun Heo if (wb_stat(wb, WB_RECLAIMABLE) > wb_dirty_limit(wb, background_thresh)) 929b00949aaSWu Fengguang return true; 930b00949aaSWu Fengguang 931b00949aaSWu Fengguang return false; 93203ba3782SJens Axboe } 93303ba3782SJens Axboe 93403ba3782SJens Axboe /* 935e98be2d5SWu Fengguang * Called under wb->list_lock. If there are multiple wb per bdi, 936e98be2d5SWu Fengguang * only the flusher working on the first wb should do it. 937e98be2d5SWu Fengguang */ 938e98be2d5SWu Fengguang static void wb_update_bandwidth(struct bdi_writeback *wb, 939e98be2d5SWu Fengguang unsigned long start_time) 940e98be2d5SWu Fengguang { 941a88a341aSTejun Heo __wb_update_bandwidth(wb, 0, 0, 0, 0, 0, start_time); 942e98be2d5SWu Fengguang } 943e98be2d5SWu Fengguang 944e98be2d5SWu Fengguang /* 94503ba3782SJens Axboe * Explicit flushing or periodic writeback of "old" data. 94603ba3782SJens Axboe * 94703ba3782SJens Axboe * Define "old": the first time one of an inode's pages is dirtied, we mark the 94803ba3782SJens Axboe * dirtying-time in the inode's address_space. So this periodic writeback code 94903ba3782SJens Axboe * just walks the superblock inode list, writing back any inodes which are 95003ba3782SJens Axboe * older than a specific point in time. 95103ba3782SJens Axboe * 95203ba3782SJens Axboe * Try to run once per dirty_writeback_interval. But if a writeback event 95303ba3782SJens Axboe * takes longer than a dirty_writeback_interval interval, then leave a 95403ba3782SJens Axboe * one-second gap. 95503ba3782SJens Axboe * 95603ba3782SJens Axboe * older_than_this takes precedence over nr_to_write. So we'll only write back 95703ba3782SJens Axboe * all dirty pages if they are all attached to "old" mappings. 95803ba3782SJens Axboe */ 959c4a77a6cSJens Axboe static long wb_writeback(struct bdi_writeback *wb, 96083ba7b07SChristoph Hellwig struct wb_writeback_work *work) 96103ba3782SJens Axboe { 962e98be2d5SWu Fengguang unsigned long wb_start = jiffies; 963d46db3d5SWu Fengguang long nr_pages = work->nr_pages; 9640dc83bd3SJan Kara unsigned long oldest_jif; 965a5989bdcSJan Kara struct inode *inode; 966d46db3d5SWu Fengguang long progress; 96703ba3782SJens Axboe 9680dc83bd3SJan Kara oldest_jif = jiffies; 9690dc83bd3SJan Kara work->older_than_this = &oldest_jif; 97003ba3782SJens Axboe 971e8dfc305SWu Fengguang spin_lock(&wb->list_lock); 97203ba3782SJens Axboe for (;;) { 97303ba3782SJens Axboe /* 974d3ddec76SWu Fengguang * Stop writeback when nr_pages has been consumed 97503ba3782SJens Axboe */ 97683ba7b07SChristoph Hellwig if (work->nr_pages <= 0) 97703ba3782SJens Axboe break; 97803ba3782SJens Axboe 97903ba3782SJens Axboe /* 980aa373cf5SJan Kara * Background writeout and kupdate-style writeback may 981aa373cf5SJan Kara * run forever. Stop them if there is other work to do 982aa373cf5SJan Kara * so that e.g. sync can proceed. They'll be restarted 983aa373cf5SJan Kara * after the other works are all done. 984aa373cf5SJan Kara */ 985aa373cf5SJan Kara if ((work->for_background || work->for_kupdate) && 986f0054bb1STejun Heo !list_empty(&wb->work_list)) 987aa373cf5SJan Kara break; 988aa373cf5SJan Kara 989aa373cf5SJan Kara /* 990d3ddec76SWu Fengguang * For background writeout, stop when we are below the 991d3ddec76SWu Fengguang * background dirty threshold 99203ba3782SJens Axboe */ 993a88a341aSTejun Heo if (work->for_background && !over_bground_thresh(wb)) 99403ba3782SJens Axboe break; 99503ba3782SJens Axboe 9961bc36b64SJan Kara /* 9971bc36b64SJan Kara * Kupdate and background works are special and we want to 9981bc36b64SJan Kara * include all inodes that need writing. Livelock avoidance is 9991bc36b64SJan Kara * handled by these works yielding to any other work so we are 10001bc36b64SJan Kara * safe. 10011bc36b64SJan Kara */ 1002ba9aa839SWu Fengguang if (work->for_kupdate) { 10030dc83bd3SJan Kara oldest_jif = jiffies - 1004ba9aa839SWu Fengguang msecs_to_jiffies(dirty_expire_interval * 10); 10051bc36b64SJan Kara } else if (work->for_background) 10060dc83bd3SJan Kara oldest_jif = jiffies; 1007028c2dd1SDave Chinner 1008d46db3d5SWu Fengguang trace_writeback_start(wb->bdi, work); 1009e8dfc305SWu Fengguang if (list_empty(&wb->b_io)) 1010ad4e38ddSCurt Wohlgemuth queue_io(wb, work); 101183ba7b07SChristoph Hellwig if (work->sb) 1012d46db3d5SWu Fengguang progress = writeback_sb_inodes(work->sb, wb, work); 1013edadfb10SChristoph Hellwig else 1014d46db3d5SWu Fengguang progress = __writeback_inodes_wb(wb, work); 1015d46db3d5SWu Fengguang trace_writeback_written(wb->bdi, work); 1016028c2dd1SDave Chinner 1017e98be2d5SWu Fengguang wb_update_bandwidth(wb, wb_start); 101803ba3782SJens Axboe 101903ba3782SJens Axboe /* 102071fd05a8SJens Axboe * Did we write something? Try for more 1021e6fb6da2SWu Fengguang * 1022e6fb6da2SWu Fengguang * Dirty inodes are moved to b_io for writeback in batches. 1023e6fb6da2SWu Fengguang * The completion of the current batch does not necessarily 1024e6fb6da2SWu Fengguang * mean the overall work is done. So we keep looping as long 1025e6fb6da2SWu Fengguang * as made some progress on cleaning pages or inodes. 102671fd05a8SJens Axboe */ 1027d46db3d5SWu Fengguang if (progress) 102803ba3782SJens Axboe continue; 1029a5989bdcSJan Kara /* 1030e6fb6da2SWu Fengguang * No more inodes for IO, bail 1031a5989bdcSJan Kara */ 1032b7a2441fSWu Fengguang if (list_empty(&wb->b_more_io)) 103303ba3782SJens Axboe break; 103403ba3782SJens Axboe /* 10358010c3b6SJens Axboe * Nothing written. Wait for some inode to 10368010c3b6SJens Axboe * become available for writeback. Otherwise 10378010c3b6SJens Axboe * we'll just busyloop. 103803ba3782SJens Axboe */ 103903ba3782SJens Axboe if (!list_empty(&wb->b_more_io)) { 1040d46db3d5SWu Fengguang trace_writeback_wait(wb->bdi, work); 104103ba3782SJens Axboe inode = wb_inode(wb->b_more_io.prev); 1042250df6edSDave Chinner spin_lock(&inode->i_lock); 1043f0d07b7fSJan Kara spin_unlock(&wb->list_lock); 1044169ebd90SJan Kara /* This function drops i_lock... */ 1045169ebd90SJan Kara inode_sleep_on_writeback(inode); 1046f0d07b7fSJan Kara spin_lock(&wb->list_lock); 104703ba3782SJens Axboe } 104803ba3782SJens Axboe } 1049e8dfc305SWu Fengguang spin_unlock(&wb->list_lock); 105003ba3782SJens Axboe 1051d46db3d5SWu Fengguang return nr_pages - work->nr_pages; 105203ba3782SJens Axboe } 105303ba3782SJens Axboe 105403ba3782SJens Axboe /* 105583ba7b07SChristoph Hellwig * Return the next wb_writeback_work struct that hasn't been processed yet. 105603ba3782SJens Axboe */ 1057f0054bb1STejun Heo static struct wb_writeback_work *get_next_work_item(struct bdi_writeback *wb) 105803ba3782SJens Axboe { 105983ba7b07SChristoph Hellwig struct wb_writeback_work *work = NULL; 106003ba3782SJens Axboe 1061f0054bb1STejun Heo spin_lock_bh(&wb->work_lock); 1062f0054bb1STejun Heo if (!list_empty(&wb->work_list)) { 1063f0054bb1STejun Heo work = list_entry(wb->work_list.next, 106483ba7b07SChristoph Hellwig struct wb_writeback_work, list); 106583ba7b07SChristoph Hellwig list_del_init(&work->list); 106603ba3782SJens Axboe } 1067f0054bb1STejun Heo spin_unlock_bh(&wb->work_lock); 106883ba7b07SChristoph Hellwig return work; 106903ba3782SJens Axboe } 107003ba3782SJens Axboe 1071cdf01dd5SLinus Torvalds /* 1072cdf01dd5SLinus Torvalds * Add in the number of potentially dirty inodes, because each inode 1073cdf01dd5SLinus Torvalds * write can dirty pagecache in the underlying blockdev. 1074cdf01dd5SLinus Torvalds */ 1075cdf01dd5SLinus Torvalds static unsigned long get_nr_dirty_pages(void) 1076cdf01dd5SLinus Torvalds { 1077cdf01dd5SLinus Torvalds return global_page_state(NR_FILE_DIRTY) + 1078cdf01dd5SLinus Torvalds global_page_state(NR_UNSTABLE_NFS) + 1079cdf01dd5SLinus Torvalds get_nr_dirty_inodes(); 1080cdf01dd5SLinus Torvalds } 1081cdf01dd5SLinus Torvalds 10826585027aSJan Kara static long wb_check_background_flush(struct bdi_writeback *wb) 10836585027aSJan Kara { 1084a88a341aSTejun Heo if (over_bground_thresh(wb)) { 10856585027aSJan Kara 10866585027aSJan Kara struct wb_writeback_work work = { 10876585027aSJan Kara .nr_pages = LONG_MAX, 10886585027aSJan Kara .sync_mode = WB_SYNC_NONE, 10896585027aSJan Kara .for_background = 1, 10906585027aSJan Kara .range_cyclic = 1, 10910e175a18SCurt Wohlgemuth .reason = WB_REASON_BACKGROUND, 10926585027aSJan Kara }; 10936585027aSJan Kara 10946585027aSJan Kara return wb_writeback(wb, &work); 10956585027aSJan Kara } 10966585027aSJan Kara 10976585027aSJan Kara return 0; 10986585027aSJan Kara } 10996585027aSJan Kara 110003ba3782SJens Axboe static long wb_check_old_data_flush(struct bdi_writeback *wb) 110103ba3782SJens Axboe { 110203ba3782SJens Axboe unsigned long expired; 110303ba3782SJens Axboe long nr_pages; 110403ba3782SJens Axboe 110569b62d01SJens Axboe /* 110669b62d01SJens Axboe * When set to zero, disable periodic writeback 110769b62d01SJens Axboe */ 110869b62d01SJens Axboe if (!dirty_writeback_interval) 110969b62d01SJens Axboe return 0; 111069b62d01SJens Axboe 111103ba3782SJens Axboe expired = wb->last_old_flush + 111203ba3782SJens Axboe msecs_to_jiffies(dirty_writeback_interval * 10); 111303ba3782SJens Axboe if (time_before(jiffies, expired)) 111403ba3782SJens Axboe return 0; 111503ba3782SJens Axboe 111603ba3782SJens Axboe wb->last_old_flush = jiffies; 1117cdf01dd5SLinus Torvalds nr_pages = get_nr_dirty_pages(); 111803ba3782SJens Axboe 1119c4a77a6cSJens Axboe if (nr_pages) { 112083ba7b07SChristoph Hellwig struct wb_writeback_work work = { 1121c4a77a6cSJens Axboe .nr_pages = nr_pages, 1122c4a77a6cSJens Axboe .sync_mode = WB_SYNC_NONE, 1123c4a77a6cSJens Axboe .for_kupdate = 1, 1124c4a77a6cSJens Axboe .range_cyclic = 1, 11250e175a18SCurt Wohlgemuth .reason = WB_REASON_PERIODIC, 1126c4a77a6cSJens Axboe }; 1127c4a77a6cSJens Axboe 112883ba7b07SChristoph Hellwig return wb_writeback(wb, &work); 1129c4a77a6cSJens Axboe } 113003ba3782SJens Axboe 113103ba3782SJens Axboe return 0; 113203ba3782SJens Axboe } 113303ba3782SJens Axboe 113403ba3782SJens Axboe /* 113503ba3782SJens Axboe * Retrieve work items and do the writeback they describe 113603ba3782SJens Axboe */ 113725d130baSWanpeng Li static long wb_do_writeback(struct bdi_writeback *wb) 113803ba3782SJens Axboe { 113983ba7b07SChristoph Hellwig struct wb_writeback_work *work; 1140c4a77a6cSJens Axboe long wrote = 0; 114103ba3782SJens Axboe 11424452226eSTejun Heo set_bit(WB_writeback_running, &wb->state); 1143f0054bb1STejun Heo while ((work = get_next_work_item(wb)) != NULL) { 114483ba7b07SChristoph Hellwig 1145f0054bb1STejun Heo trace_writeback_exec(wb->bdi, work); 1146455b2864SDave Chinner 114783ba7b07SChristoph Hellwig wrote += wb_writeback(wb, work); 114803ba3782SJens Axboe 114903ba3782SJens Axboe /* 115083ba7b07SChristoph Hellwig * Notify the caller of completion if this is a synchronous 115183ba7b07SChristoph Hellwig * work item, otherwise just free it. 115203ba3782SJens Axboe */ 115383ba7b07SChristoph Hellwig if (work->done) 115483ba7b07SChristoph Hellwig complete(work->done); 115583ba7b07SChristoph Hellwig else 115683ba7b07SChristoph Hellwig kfree(work); 115703ba3782SJens Axboe } 115803ba3782SJens Axboe 115903ba3782SJens Axboe /* 116003ba3782SJens Axboe * Check for periodic writeback, kupdated() style 116103ba3782SJens Axboe */ 116203ba3782SJens Axboe wrote += wb_check_old_data_flush(wb); 11636585027aSJan Kara wrote += wb_check_background_flush(wb); 11644452226eSTejun Heo clear_bit(WB_writeback_running, &wb->state); 116503ba3782SJens Axboe 116603ba3782SJens Axboe return wrote; 116703ba3782SJens Axboe } 116803ba3782SJens Axboe 116903ba3782SJens Axboe /* 117003ba3782SJens Axboe * Handle writeback of dirty data for the device backed by this bdi. Also 1171839a8e86STejun Heo * reschedules periodically and does kupdated style flushing. 117203ba3782SJens Axboe */ 1173f0054bb1STejun Heo void wb_workfn(struct work_struct *work) 117403ba3782SJens Axboe { 1175839a8e86STejun Heo struct bdi_writeback *wb = container_of(to_delayed_work(work), 1176839a8e86STejun Heo struct bdi_writeback, dwork); 117703ba3782SJens Axboe long pages_written; 117803ba3782SJens Axboe 1179f0054bb1STejun Heo set_worker_desc("flush-%s", dev_name(wb->bdi->dev)); 1180766f9164SPeter Zijlstra current->flags |= PF_SWAPWRITE; 118103ba3782SJens Axboe 1182839a8e86STejun Heo if (likely(!current_is_workqueue_rescuer() || 11834452226eSTejun Heo !test_bit(WB_registered, &wb->state))) { 118403ba3782SJens Axboe /* 1185f0054bb1STejun Heo * The normal path. Keep writing back @wb until its 1186839a8e86STejun Heo * work_list is empty. Note that this path is also taken 1187f0054bb1STejun Heo * if @wb is shutting down even when we're running off the 1188839a8e86STejun Heo * rescuer as work_list needs to be drained. 118903ba3782SJens Axboe */ 1190839a8e86STejun Heo do { 119125d130baSWanpeng Li pages_written = wb_do_writeback(wb); 1192455b2864SDave Chinner trace_writeback_pages_written(pages_written); 1193f0054bb1STejun Heo } while (!list_empty(&wb->work_list)); 1194839a8e86STejun Heo } else { 1195253c34e9SArtem Bityutskiy /* 1196839a8e86STejun Heo * bdi_wq can't get enough workers and we're running off 1197839a8e86STejun Heo * the emergency worker. Don't hog it. Hopefully, 1024 is 1198839a8e86STejun Heo * enough for efficient IO. 1199253c34e9SArtem Bityutskiy */ 1200f0054bb1STejun Heo pages_written = writeback_inodes_wb(wb, 1024, 1201839a8e86STejun Heo WB_REASON_FORKER_THREAD); 1202839a8e86STejun Heo trace_writeback_pages_written(pages_written); 120303ba3782SJens Axboe } 120403ba3782SJens Axboe 1205f0054bb1STejun Heo if (!list_empty(&wb->work_list)) 12066ca738d6SDerek Basehore mod_delayed_work(bdi_wq, &wb->dwork, 0); 12076ca738d6SDerek Basehore else if (wb_has_dirty_io(wb) && dirty_writeback_interval) 1208f0054bb1STejun Heo wb_wakeup_delayed(wb); 1209455b2864SDave Chinner 1210839a8e86STejun Heo current->flags &= ~PF_SWAPWRITE; 121103ba3782SJens Axboe } 121203ba3782SJens Axboe 121303ba3782SJens Axboe /* 121403ba3782SJens Axboe * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back 121503ba3782SJens Axboe * the whole world. 121603ba3782SJens Axboe */ 12170e175a18SCurt Wohlgemuth void wakeup_flusher_threads(long nr_pages, enum wb_reason reason) 121803ba3782SJens Axboe { 1219b8c2f347SChristoph Hellwig struct backing_dev_info *bdi; 1220b8c2f347SChristoph Hellwig 122147df3ddeSJan Kara if (!nr_pages) 122247df3ddeSJan Kara nr_pages = get_nr_dirty_pages(); 1223b8c2f347SChristoph Hellwig 1224b8c2f347SChristoph Hellwig rcu_read_lock(); 1225f2b65121STejun Heo list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { 1226f2b65121STejun Heo struct bdi_writeback *wb; 1227f2b65121STejun Heo struct wb_iter iter; 1228f2b65121STejun Heo 1229f2b65121STejun Heo if (!bdi_has_dirty_io(bdi)) 1230f2b65121STejun Heo continue; 1231f2b65121STejun Heo 1232f2b65121STejun Heo bdi_for_each_wb(wb, bdi, &iter, 0) 1233f2b65121STejun Heo wb_start_writeback(wb, wb_split_bdi_pages(wb, nr_pages), 1234f2b65121STejun Heo false, reason); 1235f2b65121STejun Heo } 1236b8c2f347SChristoph Hellwig rcu_read_unlock(); 123703ba3782SJens Axboe } 123803ba3782SJens Axboe 1239a2f48706STheodore Ts'o /* 1240a2f48706STheodore Ts'o * Wake up bdi's periodically to make sure dirtytime inodes gets 1241a2f48706STheodore Ts'o * written back periodically. We deliberately do *not* check the 1242a2f48706STheodore Ts'o * b_dirtytime list in wb_has_dirty_io(), since this would cause the 1243a2f48706STheodore Ts'o * kernel to be constantly waking up once there are any dirtytime 1244a2f48706STheodore Ts'o * inodes on the system. So instead we define a separate delayed work 1245a2f48706STheodore Ts'o * function which gets called much more rarely. (By default, only 1246a2f48706STheodore Ts'o * once every 12 hours.) 1247a2f48706STheodore Ts'o * 1248a2f48706STheodore Ts'o * If there is any other write activity going on in the file system, 1249a2f48706STheodore Ts'o * this function won't be necessary. But if the only thing that has 1250a2f48706STheodore Ts'o * happened on the file system is a dirtytime inode caused by an atime 1251a2f48706STheodore Ts'o * update, we need this infrastructure below to make sure that inode 1252a2f48706STheodore Ts'o * eventually gets pushed out to disk. 1253a2f48706STheodore Ts'o */ 1254a2f48706STheodore Ts'o static void wakeup_dirtytime_writeback(struct work_struct *w); 1255a2f48706STheodore Ts'o static DECLARE_DELAYED_WORK(dirtytime_work, wakeup_dirtytime_writeback); 1256a2f48706STheodore Ts'o 1257a2f48706STheodore Ts'o static void wakeup_dirtytime_writeback(struct work_struct *w) 1258a2f48706STheodore Ts'o { 1259a2f48706STheodore Ts'o struct backing_dev_info *bdi; 1260a2f48706STheodore Ts'o 1261a2f48706STheodore Ts'o rcu_read_lock(); 1262a2f48706STheodore Ts'o list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { 1263001fe6f6STejun Heo struct bdi_writeback *wb; 1264001fe6f6STejun Heo struct wb_iter iter; 1265001fe6f6STejun Heo 1266001fe6f6STejun Heo bdi_for_each_wb(wb, bdi, &iter, 0) 1267001fe6f6STejun Heo if (!list_empty(&bdi->wb.b_dirty_time)) 1268f0054bb1STejun Heo wb_wakeup(&bdi->wb); 1269a2f48706STheodore Ts'o } 1270a2f48706STheodore Ts'o rcu_read_unlock(); 1271a2f48706STheodore Ts'o schedule_delayed_work(&dirtytime_work, dirtytime_expire_interval * HZ); 1272a2f48706STheodore Ts'o } 1273a2f48706STheodore Ts'o 1274a2f48706STheodore Ts'o static int __init start_dirtytime_writeback(void) 1275a2f48706STheodore Ts'o { 1276a2f48706STheodore Ts'o schedule_delayed_work(&dirtytime_work, dirtytime_expire_interval * HZ); 1277a2f48706STheodore Ts'o return 0; 1278a2f48706STheodore Ts'o } 1279a2f48706STheodore Ts'o __initcall(start_dirtytime_writeback); 1280a2f48706STheodore Ts'o 12811efff914STheodore Ts'o int dirtytime_interval_handler(struct ctl_table *table, int write, 12821efff914STheodore Ts'o void __user *buffer, size_t *lenp, loff_t *ppos) 12831efff914STheodore Ts'o { 12841efff914STheodore Ts'o int ret; 12851efff914STheodore Ts'o 12861efff914STheodore Ts'o ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos); 12871efff914STheodore Ts'o if (ret == 0 && write) 12881efff914STheodore Ts'o mod_delayed_work(system_wq, &dirtytime_work, 0); 12891efff914STheodore Ts'o return ret; 12901efff914STheodore Ts'o } 12911efff914STheodore Ts'o 129203ba3782SJens Axboe static noinline void block_dump___mark_inode_dirty(struct inode *inode) 129303ba3782SJens Axboe { 129403ba3782SJens Axboe if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { 129503ba3782SJens Axboe struct dentry *dentry; 129603ba3782SJens Axboe const char *name = "?"; 129703ba3782SJens Axboe 129803ba3782SJens Axboe dentry = d_find_alias(inode); 129903ba3782SJens Axboe if (dentry) { 130003ba3782SJens Axboe spin_lock(&dentry->d_lock); 130103ba3782SJens Axboe name = (const char *) dentry->d_name.name; 130203ba3782SJens Axboe } 130303ba3782SJens Axboe printk(KERN_DEBUG 130403ba3782SJens Axboe "%s(%d): dirtied inode %lu (%s) on %s\n", 130503ba3782SJens Axboe current->comm, task_pid_nr(current), inode->i_ino, 130603ba3782SJens Axboe name, inode->i_sb->s_id); 130703ba3782SJens Axboe if (dentry) { 130803ba3782SJens Axboe spin_unlock(&dentry->d_lock); 130903ba3782SJens Axboe dput(dentry); 131003ba3782SJens Axboe } 131103ba3782SJens Axboe } 131203ba3782SJens Axboe } 131303ba3782SJens Axboe 131403ba3782SJens Axboe /** 131503ba3782SJens Axboe * __mark_inode_dirty - internal function 131603ba3782SJens Axboe * @inode: inode to mark 131703ba3782SJens Axboe * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) 131803ba3782SJens Axboe * Mark an inode as dirty. Callers should use mark_inode_dirty or 131903ba3782SJens Axboe * mark_inode_dirty_sync. 132003ba3782SJens Axboe * 132103ba3782SJens Axboe * Put the inode on the super block's dirty list. 132203ba3782SJens Axboe * 132303ba3782SJens Axboe * CAREFUL! We mark it dirty unconditionally, but move it onto the 132403ba3782SJens Axboe * dirty list only if it is hashed or if it refers to a blockdev. 132503ba3782SJens Axboe * If it was not hashed, it will never be added to the dirty list 132603ba3782SJens Axboe * even if it is later hashed, as it will have been marked dirty already. 132703ba3782SJens Axboe * 132803ba3782SJens Axboe * In short, make sure you hash any inodes _before_ you start marking 132903ba3782SJens Axboe * them dirty. 133003ba3782SJens Axboe * 133103ba3782SJens Axboe * Note that for blockdevs, inode->dirtied_when represents the dirtying time of 133203ba3782SJens Axboe * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of 133303ba3782SJens Axboe * the kernel-internal blockdev inode represents the dirtying time of the 133403ba3782SJens Axboe * blockdev's pages. This is why for I_DIRTY_PAGES we always use 133503ba3782SJens Axboe * page->mapping->host, so the page-dirtying time is recorded in the internal 133603ba3782SJens Axboe * blockdev inode. 133703ba3782SJens Axboe */ 13380ae45f63STheodore Ts'o #define I_DIRTY_INODE (I_DIRTY_SYNC | I_DIRTY_DATASYNC) 133903ba3782SJens Axboe void __mark_inode_dirty(struct inode *inode, int flags) 134003ba3782SJens Axboe { 134103ba3782SJens Axboe struct super_block *sb = inode->i_sb; 1342253c34e9SArtem Bityutskiy struct backing_dev_info *bdi = NULL; 13430ae45f63STheodore Ts'o int dirtytime; 13440ae45f63STheodore Ts'o 13450ae45f63STheodore Ts'o trace_writeback_mark_inode_dirty(inode, flags); 134603ba3782SJens Axboe 134703ba3782SJens Axboe /* 134803ba3782SJens Axboe * Don't do this for I_DIRTY_PAGES - that doesn't actually 134903ba3782SJens Axboe * dirty the inode itself 135003ba3782SJens Axboe */ 13510ae45f63STheodore Ts'o if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC | I_DIRTY_TIME)) { 13529fb0a7daSTejun Heo trace_writeback_dirty_inode_start(inode, flags); 13539fb0a7daSTejun Heo 135403ba3782SJens Axboe if (sb->s_op->dirty_inode) 1355aa385729SChristoph Hellwig sb->s_op->dirty_inode(inode, flags); 13569fb0a7daSTejun Heo 13579fb0a7daSTejun Heo trace_writeback_dirty_inode(inode, flags); 135803ba3782SJens Axboe } 13590ae45f63STheodore Ts'o if (flags & I_DIRTY_INODE) 13600ae45f63STheodore Ts'o flags &= ~I_DIRTY_TIME; 13610ae45f63STheodore Ts'o dirtytime = flags & I_DIRTY_TIME; 136203ba3782SJens Axboe 136303ba3782SJens Axboe /* 13649c6ac78eSTejun Heo * Paired with smp_mb() in __writeback_single_inode() for the 13659c6ac78eSTejun Heo * following lockless i_state test. See there for details. 136603ba3782SJens Axboe */ 136703ba3782SJens Axboe smp_mb(); 136803ba3782SJens Axboe 13690ae45f63STheodore Ts'o if (((inode->i_state & flags) == flags) || 13700ae45f63STheodore Ts'o (dirtytime && (inode->i_state & I_DIRTY_INODE))) 137103ba3782SJens Axboe return; 137203ba3782SJens Axboe 137303ba3782SJens Axboe if (unlikely(block_dump)) 137403ba3782SJens Axboe block_dump___mark_inode_dirty(inode); 137503ba3782SJens Axboe 1376250df6edSDave Chinner spin_lock(&inode->i_lock); 13770ae45f63STheodore Ts'o if (dirtytime && (inode->i_state & I_DIRTY_INODE)) 13780ae45f63STheodore Ts'o goto out_unlock_inode; 137903ba3782SJens Axboe if ((inode->i_state & flags) != flags) { 138003ba3782SJens Axboe const int was_dirty = inode->i_state & I_DIRTY; 138103ba3782SJens Axboe 138252ebea74STejun Heo inode_attach_wb(inode, NULL); 138352ebea74STejun Heo 13840ae45f63STheodore Ts'o if (flags & I_DIRTY_INODE) 13850ae45f63STheodore Ts'o inode->i_state &= ~I_DIRTY_TIME; 138603ba3782SJens Axboe inode->i_state |= flags; 138703ba3782SJens Axboe 138803ba3782SJens Axboe /* 138903ba3782SJens Axboe * If the inode is being synced, just update its dirty state. 139003ba3782SJens Axboe * The unlocker will place the inode on the appropriate 139103ba3782SJens Axboe * superblock list, based upon its state. 139203ba3782SJens Axboe */ 139303ba3782SJens Axboe if (inode->i_state & I_SYNC) 1394250df6edSDave Chinner goto out_unlock_inode; 139503ba3782SJens Axboe 139603ba3782SJens Axboe /* 139703ba3782SJens Axboe * Only add valid (hashed) inodes to the superblock's 139803ba3782SJens Axboe * dirty list. Add blockdev inodes as well. 139903ba3782SJens Axboe */ 140003ba3782SJens Axboe if (!S_ISBLK(inode->i_mode)) { 14011d3382cbSAl Viro if (inode_unhashed(inode)) 1402250df6edSDave Chinner goto out_unlock_inode; 140303ba3782SJens Axboe } 1404a4ffdde6SAl Viro if (inode->i_state & I_FREEING) 1405250df6edSDave Chinner goto out_unlock_inode; 140603ba3782SJens Axboe 140703ba3782SJens Axboe /* 140803ba3782SJens Axboe * If the inode was already on b_dirty/b_io/b_more_io, don't 140903ba3782SJens Axboe * reposition it (that would break b_dirty time-ordering). 141003ba3782SJens Axboe */ 141103ba3782SJens Axboe if (!was_dirty) { 1412d6c10f1fSTejun Heo struct list_head *dirty_list; 1413a66979abSDave Chinner bool wakeup_bdi = false; 1414253c34e9SArtem Bityutskiy bdi = inode_to_bdi(inode); 1415500b067cSJens Axboe 1416146d7009SJunxiao Bi spin_unlock(&inode->i_lock); 1417146d7009SJunxiao Bi spin_lock(&bdi->wb.list_lock); 1418253c34e9SArtem Bityutskiy 1419d6c10f1fSTejun Heo WARN(bdi_cap_writeback_dirty(bdi) && 1420d6c10f1fSTejun Heo !test_bit(WB_registered, &bdi->wb.state), 1421d6c10f1fSTejun Heo "bdi-%s not registered\n", bdi->name); 142203ba3782SJens Axboe 142303ba3782SJens Axboe inode->dirtied_when = jiffies; 1424a2f48706STheodore Ts'o if (dirtytime) 1425a2f48706STheodore Ts'o inode->dirtied_time_when = jiffies; 1426d6c10f1fSTejun Heo 1427a2f48706STheodore Ts'o if (inode->i_state & (I_DIRTY_INODE | I_DIRTY_PAGES)) 1428d6c10f1fSTejun Heo dirty_list = &bdi->wb.b_dirty; 1429a2f48706STheodore Ts'o else 1430d6c10f1fSTejun Heo dirty_list = &bdi->wb.b_dirty_time; 1431d6c10f1fSTejun Heo 1432d6c10f1fSTejun Heo wakeup_bdi = inode_wb_list_move_locked(inode, &bdi->wb, 1433d6c10f1fSTejun Heo dirty_list); 1434d6c10f1fSTejun Heo 1435f758eeabSChristoph Hellwig spin_unlock(&bdi->wb.list_lock); 14360ae45f63STheodore Ts'o trace_writeback_dirty_inode_enqueue(inode); 1437253c34e9SArtem Bityutskiy 1438d6c10f1fSTejun Heo /* 1439d6c10f1fSTejun Heo * If this is the first dirty inode for this bdi, 1440d6c10f1fSTejun Heo * we have to wake-up the corresponding bdi thread 1441d6c10f1fSTejun Heo * to make sure background write-back happens 1442d6c10f1fSTejun Heo * later. 1443d6c10f1fSTejun Heo */ 1444d6c10f1fSTejun Heo if (bdi_cap_writeback_dirty(bdi) && wakeup_bdi) 1445f0054bb1STejun Heo wb_wakeup_delayed(&bdi->wb); 1446a66979abSDave Chinner return; 1447a66979abSDave Chinner } 1448a66979abSDave Chinner } 1449a66979abSDave Chinner out_unlock_inode: 1450a66979abSDave Chinner spin_unlock(&inode->i_lock); 1451a66979abSDave Chinner 145203ba3782SJens Axboe } 145303ba3782SJens Axboe EXPORT_SYMBOL(__mark_inode_dirty); 145403ba3782SJens Axboe 1455b6e51316SJens Axboe static void wait_sb_inodes(struct super_block *sb) 145666f3b8e2SJens Axboe { 145738f21977SNick Piggin struct inode *inode, *old_inode = NULL; 145838f21977SNick Piggin 145903ba3782SJens Axboe /* 146003ba3782SJens Axboe * We need to be protected against the filesystem going from 146103ba3782SJens Axboe * r/o to r/w or vice versa. 146203ba3782SJens Axboe */ 1463b6e51316SJens Axboe WARN_ON(!rwsem_is_locked(&sb->s_umount)); 146403ba3782SJens Axboe 146555fa6091SDave Chinner spin_lock(&inode_sb_list_lock); 146666f3b8e2SJens Axboe 146738f21977SNick Piggin /* 146838f21977SNick Piggin * Data integrity sync. Must wait for all pages under writeback, 146938f21977SNick Piggin * because there may have been pages dirtied before our sync 147038f21977SNick Piggin * call, but which had writeout started before we write it out. 147138f21977SNick Piggin * In which case, the inode may not be on the dirty list, but 147238f21977SNick Piggin * we still have to wait for that writeout. 147338f21977SNick Piggin */ 1474b6e51316SJens Axboe list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { 1475250df6edSDave Chinner struct address_space *mapping = inode->i_mapping; 147638f21977SNick Piggin 1477250df6edSDave Chinner spin_lock(&inode->i_lock); 1478250df6edSDave Chinner if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) || 1479250df6edSDave Chinner (mapping->nrpages == 0)) { 1480250df6edSDave Chinner spin_unlock(&inode->i_lock); 148138f21977SNick Piggin continue; 1482250df6edSDave Chinner } 148338f21977SNick Piggin __iget(inode); 1484250df6edSDave Chinner spin_unlock(&inode->i_lock); 148555fa6091SDave Chinner spin_unlock(&inode_sb_list_lock); 148655fa6091SDave Chinner 148738f21977SNick Piggin /* 148855fa6091SDave Chinner * We hold a reference to 'inode' so it couldn't have been 148955fa6091SDave Chinner * removed from s_inodes list while we dropped the 149055fa6091SDave Chinner * inode_sb_list_lock. We cannot iput the inode now as we can 149155fa6091SDave Chinner * be holding the last reference and we cannot iput it under 149255fa6091SDave Chinner * inode_sb_list_lock. So we keep the reference and iput it 149355fa6091SDave Chinner * later. 149438f21977SNick Piggin */ 149538f21977SNick Piggin iput(old_inode); 149638f21977SNick Piggin old_inode = inode; 149738f21977SNick Piggin 149838f21977SNick Piggin filemap_fdatawait(mapping); 149938f21977SNick Piggin 150038f21977SNick Piggin cond_resched(); 150138f21977SNick Piggin 150255fa6091SDave Chinner spin_lock(&inode_sb_list_lock); 150338f21977SNick Piggin } 150455fa6091SDave Chinner spin_unlock(&inode_sb_list_lock); 150538f21977SNick Piggin iput(old_inode); 150666f3b8e2SJens Axboe } 15071da177e4SLinus Torvalds 1508d8a8559cSJens Axboe /** 15093259f8beSChris Mason * writeback_inodes_sb_nr - writeback dirty inodes from given super_block 1510d8a8559cSJens Axboe * @sb: the superblock 15113259f8beSChris Mason * @nr: the number of pages to write 1512786228abSMarcos Paulo de Souza * @reason: reason why some writeback work initiated 15131da177e4SLinus Torvalds * 1514d8a8559cSJens Axboe * Start writeback on some inodes on this super_block. No guarantees are made 1515d8a8559cSJens Axboe * on how many (if any) will be written, and this function does not wait 15163259f8beSChris Mason * for IO completion of submitted IO. 15171da177e4SLinus Torvalds */ 15180e175a18SCurt Wohlgemuth void writeback_inodes_sb_nr(struct super_block *sb, 15190e175a18SCurt Wohlgemuth unsigned long nr, 15200e175a18SCurt Wohlgemuth enum wb_reason reason) 15211da177e4SLinus Torvalds { 152283ba7b07SChristoph Hellwig DECLARE_COMPLETION_ONSTACK(done); 152383ba7b07SChristoph Hellwig struct wb_writeback_work work = { 15243c4d7165SChristoph Hellwig .sb = sb, 15253c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_NONE, 15266e6938b6SWu Fengguang .tagged_writepages = 1, 152783ba7b07SChristoph Hellwig .done = &done, 15283259f8beSChris Mason .nr_pages = nr, 15290e175a18SCurt Wohlgemuth .reason = reason, 15303c4d7165SChristoph Hellwig }; 1531e7972912STejun Heo struct backing_dev_info *bdi = sb->s_bdi; 15320e3c9a22SJens Axboe 1533e7972912STejun Heo if (!bdi_has_dirty_io(bdi) || bdi == &noop_backing_dev_info) 15346eedc701SJan Kara return; 1535cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 1536e7972912STejun Heo wb_queue_work(&bdi->wb, &work); 153783ba7b07SChristoph Hellwig wait_for_completion(&done); 15381da177e4SLinus Torvalds } 15393259f8beSChris Mason EXPORT_SYMBOL(writeback_inodes_sb_nr); 15403259f8beSChris Mason 15413259f8beSChris Mason /** 15423259f8beSChris Mason * writeback_inodes_sb - writeback dirty inodes from given super_block 15433259f8beSChris Mason * @sb: the superblock 1544786228abSMarcos Paulo de Souza * @reason: reason why some writeback work was initiated 15453259f8beSChris Mason * 15463259f8beSChris Mason * Start writeback on some inodes on this super_block. No guarantees are made 15473259f8beSChris Mason * on how many (if any) will be written, and this function does not wait 15483259f8beSChris Mason * for IO completion of submitted IO. 15493259f8beSChris Mason */ 15500e175a18SCurt Wohlgemuth void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) 15513259f8beSChris Mason { 15520e175a18SCurt Wohlgemuth return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); 15533259f8beSChris Mason } 1554d8a8559cSJens Axboe EXPORT_SYMBOL(writeback_inodes_sb); 1555d8a8559cSJens Axboe 1556d8a8559cSJens Axboe /** 155710ee27a0SMiao Xie * try_to_writeback_inodes_sb_nr - try to start writeback if none underway 15583259f8beSChris Mason * @sb: the superblock 15593259f8beSChris Mason * @nr: the number of pages to write 156010ee27a0SMiao Xie * @reason: the reason of writeback 15613259f8beSChris Mason * 156210ee27a0SMiao Xie * Invoke writeback_inodes_sb_nr if no writeback is currently underway. 15633259f8beSChris Mason * Returns 1 if writeback was started, 0 if not. 15643259f8beSChris Mason */ 156510ee27a0SMiao Xie int try_to_writeback_inodes_sb_nr(struct super_block *sb, 15660e175a18SCurt Wohlgemuth unsigned long nr, 15670e175a18SCurt Wohlgemuth enum wb_reason reason) 15683259f8beSChris Mason { 1569bc05873dSTejun Heo if (writeback_in_progress(&sb->s_bdi->wb)) 157010ee27a0SMiao Xie return 1; 157110ee27a0SMiao Xie 157210ee27a0SMiao Xie if (!down_read_trylock(&sb->s_umount)) 157310ee27a0SMiao Xie return 0; 157410ee27a0SMiao Xie 15750e175a18SCurt Wohlgemuth writeback_inodes_sb_nr(sb, nr, reason); 15763259f8beSChris Mason up_read(&sb->s_umount); 15773259f8beSChris Mason return 1; 15783259f8beSChris Mason } 157910ee27a0SMiao Xie EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr); 158010ee27a0SMiao Xie 158110ee27a0SMiao Xie /** 158210ee27a0SMiao Xie * try_to_writeback_inodes_sb - try to start writeback if none underway 158310ee27a0SMiao Xie * @sb: the superblock 158410ee27a0SMiao Xie * @reason: reason why some writeback work was initiated 158510ee27a0SMiao Xie * 158610ee27a0SMiao Xie * Implement by try_to_writeback_inodes_sb_nr() 158710ee27a0SMiao Xie * Returns 1 if writeback was started, 0 if not. 158810ee27a0SMiao Xie */ 158910ee27a0SMiao Xie int try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) 159010ee27a0SMiao Xie { 159110ee27a0SMiao Xie return try_to_writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); 159210ee27a0SMiao Xie } 159310ee27a0SMiao Xie EXPORT_SYMBOL(try_to_writeback_inodes_sb); 15943259f8beSChris Mason 15953259f8beSChris Mason /** 1596d8a8559cSJens Axboe * sync_inodes_sb - sync sb inode pages 1597d8a8559cSJens Axboe * @sb: the superblock 1598d8a8559cSJens Axboe * 1599d8a8559cSJens Axboe * This function writes and waits on any dirty inode belonging to this 16000dc83bd3SJan Kara * super_block. 1601d8a8559cSJens Axboe */ 16020dc83bd3SJan Kara void sync_inodes_sb(struct super_block *sb) 1603d8a8559cSJens Axboe { 160483ba7b07SChristoph Hellwig DECLARE_COMPLETION_ONSTACK(done); 160583ba7b07SChristoph Hellwig struct wb_writeback_work work = { 16063c4d7165SChristoph Hellwig .sb = sb, 16073c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_ALL, 16083c4d7165SChristoph Hellwig .nr_pages = LONG_MAX, 16093c4d7165SChristoph Hellwig .range_cyclic = 0, 161083ba7b07SChristoph Hellwig .done = &done, 16110e175a18SCurt Wohlgemuth .reason = WB_REASON_SYNC, 16127747bd4bSDave Chinner .for_sync = 1, 16133c4d7165SChristoph Hellwig }; 1614e7972912STejun Heo struct backing_dev_info *bdi = sb->s_bdi; 16153c4d7165SChristoph Hellwig 16166eedc701SJan Kara /* Nothing to do? */ 1617e7972912STejun Heo if (!bdi_has_dirty_io(bdi) || bdi == &noop_backing_dev_info) 16186eedc701SJan Kara return; 1619cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 1620cf37e972SChristoph Hellwig 1621e7972912STejun Heo wb_queue_work(&bdi->wb, &work); 162283ba7b07SChristoph Hellwig wait_for_completion(&done); 162383ba7b07SChristoph Hellwig 1624b6e51316SJens Axboe wait_sb_inodes(sb); 1625d8a8559cSJens Axboe } 1626d8a8559cSJens Axboe EXPORT_SYMBOL(sync_inodes_sb); 16271da177e4SLinus Torvalds 16281da177e4SLinus Torvalds /** 16291da177e4SLinus Torvalds * write_inode_now - write an inode to disk 16301da177e4SLinus Torvalds * @inode: inode to write to disk 16311da177e4SLinus Torvalds * @sync: whether the write should be synchronous or not 16321da177e4SLinus Torvalds * 16337f04c26dSAndrea Arcangeli * This function commits an inode to disk immediately if it is dirty. This is 16347f04c26dSAndrea Arcangeli * primarily needed by knfsd. 16357f04c26dSAndrea Arcangeli * 16367f04c26dSAndrea Arcangeli * The caller must either have a ref on the inode or must have set I_WILL_FREE. 16371da177e4SLinus Torvalds */ 16381da177e4SLinus Torvalds int write_inode_now(struct inode *inode, int sync) 16391da177e4SLinus Torvalds { 1640f758eeabSChristoph Hellwig struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; 16411da177e4SLinus Torvalds struct writeback_control wbc = { 16421da177e4SLinus Torvalds .nr_to_write = LONG_MAX, 164318914b18SMike Galbraith .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, 1644111ebb6eSOGAWA Hirofumi .range_start = 0, 1645111ebb6eSOGAWA Hirofumi .range_end = LLONG_MAX, 16461da177e4SLinus Torvalds }; 16471da177e4SLinus Torvalds 16481da177e4SLinus Torvalds if (!mapping_cap_writeback_dirty(inode->i_mapping)) 164949364ce2SAndrew Morton wbc.nr_to_write = 0; 16501da177e4SLinus Torvalds 16511da177e4SLinus Torvalds might_sleep(); 16524f8ad655SJan Kara return writeback_single_inode(inode, wb, &wbc); 16531da177e4SLinus Torvalds } 16541da177e4SLinus Torvalds EXPORT_SYMBOL(write_inode_now); 16551da177e4SLinus Torvalds 16561da177e4SLinus Torvalds /** 16571da177e4SLinus Torvalds * sync_inode - write an inode and its pages to disk. 16581da177e4SLinus Torvalds * @inode: the inode to sync 16591da177e4SLinus Torvalds * @wbc: controls the writeback mode 16601da177e4SLinus Torvalds * 16611da177e4SLinus Torvalds * sync_inode() will write an inode and its pages to disk. It will also 16621da177e4SLinus Torvalds * correctly update the inode on its superblock's dirty inode lists and will 16631da177e4SLinus Torvalds * update inode->i_state. 16641da177e4SLinus Torvalds * 16651da177e4SLinus Torvalds * The caller must have a ref on the inode. 16661da177e4SLinus Torvalds */ 16671da177e4SLinus Torvalds int sync_inode(struct inode *inode, struct writeback_control *wbc) 16681da177e4SLinus Torvalds { 16694f8ad655SJan Kara return writeback_single_inode(inode, &inode_to_bdi(inode)->wb, wbc); 16701da177e4SLinus Torvalds } 16711da177e4SLinus Torvalds EXPORT_SYMBOL(sync_inode); 1672c3765016SChristoph Hellwig 1673c3765016SChristoph Hellwig /** 1674c691b9d9SAndrew Morton * sync_inode_metadata - write an inode to disk 1675c3765016SChristoph Hellwig * @inode: the inode to sync 1676c3765016SChristoph Hellwig * @wait: wait for I/O to complete. 1677c3765016SChristoph Hellwig * 1678c691b9d9SAndrew Morton * Write an inode to disk and adjust its dirty state after completion. 1679c3765016SChristoph Hellwig * 1680c3765016SChristoph Hellwig * Note: only writes the actual inode, no associated data or other metadata. 1681c3765016SChristoph Hellwig */ 1682c3765016SChristoph Hellwig int sync_inode_metadata(struct inode *inode, int wait) 1683c3765016SChristoph Hellwig { 1684c3765016SChristoph Hellwig struct writeback_control wbc = { 1685c3765016SChristoph Hellwig .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE, 1686c3765016SChristoph Hellwig .nr_to_write = 0, /* metadata-only */ 1687c3765016SChristoph Hellwig }; 1688c3765016SChristoph Hellwig 1689c3765016SChristoph Hellwig return sync_inode(inode, &wbc); 1690c3765016SChristoph Hellwig } 1691c3765016SChristoph Hellwig EXPORT_SYMBOL(sync_inode_metadata); 1692