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 37cc395d7fSTejun Heo struct wb_completion { 38cc395d7fSTejun Heo atomic_t cnt; 39cc395d7fSTejun Heo }; 40cc395d7fSTejun Heo 41bc31b86aSWu Fengguang /* 42c4a77a6cSJens Axboe * Passed into wb_writeback(), essentially a subset of writeback_control 43c4a77a6cSJens Axboe */ 4483ba7b07SChristoph Hellwig struct wb_writeback_work { 45c4a77a6cSJens Axboe long nr_pages; 46c4a77a6cSJens Axboe struct super_block *sb; 470dc83bd3SJan Kara unsigned long *older_than_this; 48c4a77a6cSJens Axboe enum writeback_sync_modes sync_mode; 496e6938b6SWu Fengguang unsigned int tagged_writepages:1; 5052957fe1SH Hartley Sweeten unsigned int for_kupdate:1; 5152957fe1SH Hartley Sweeten unsigned int range_cyclic:1; 5252957fe1SH Hartley Sweeten unsigned int for_background:1; 537747bd4bSDave Chinner unsigned int for_sync:1; /* sync(2) WB_SYNC_ALL writeback */ 54ac7b19a3STejun Heo unsigned int auto_free:1; /* free on completion */ 550e175a18SCurt Wohlgemuth enum wb_reason reason; /* why was writeback initiated? */ 56c4a77a6cSJens Axboe 578010c3b6SJens Axboe struct list_head list; /* pending work list */ 58cc395d7fSTejun Heo struct wb_completion *done; /* set if the caller waits */ 5903ba3782SJens Axboe }; 6003ba3782SJens Axboe 61a2f48706STheodore Ts'o /* 62cc395d7fSTejun Heo * If one wants to wait for one or more wb_writeback_works, each work's 63cc395d7fSTejun Heo * ->done should be set to a wb_completion defined using the following 64cc395d7fSTejun Heo * macro. Once all work items are issued with wb_queue_work(), the caller 65cc395d7fSTejun Heo * can wait for the completion of all using wb_wait_for_completion(). Work 66cc395d7fSTejun Heo * items which are waited upon aren't freed automatically on completion. 67cc395d7fSTejun Heo */ 68cc395d7fSTejun Heo #define DEFINE_WB_COMPLETION_ONSTACK(cmpl) \ 69cc395d7fSTejun Heo struct wb_completion cmpl = { \ 70cc395d7fSTejun Heo .cnt = ATOMIC_INIT(1), \ 71cc395d7fSTejun Heo } 72cc395d7fSTejun Heo 73cc395d7fSTejun Heo 74cc395d7fSTejun Heo /* 75a2f48706STheodore Ts'o * If an inode is constantly having its pages dirtied, but then the 76a2f48706STheodore Ts'o * updates stop dirtytime_expire_interval seconds in the past, it's 77a2f48706STheodore Ts'o * possible for the worst case time between when an inode has its 78a2f48706STheodore Ts'o * timestamps updated and when they finally get written out to be two 79a2f48706STheodore Ts'o * dirtytime_expire_intervals. We set the default to 12 hours (in 80a2f48706STheodore Ts'o * seconds), which means most of the time inodes will have their 81a2f48706STheodore Ts'o * timestamps written to disk after 12 hours, but in the worst case a 82a2f48706STheodore Ts'o * few inodes might not their timestamps updated for 24 hours. 83a2f48706STheodore Ts'o */ 84a2f48706STheodore Ts'o unsigned int dirtytime_expire_interval = 12 * 60 * 60; 85a2f48706STheodore Ts'o 867ccf19a8SNick Piggin static inline struct inode *wb_inode(struct list_head *head) 877ccf19a8SNick Piggin { 887ccf19a8SNick Piggin return list_entry(head, struct inode, i_wb_list); 897ccf19a8SNick Piggin } 907ccf19a8SNick Piggin 9115eb77a0SWu Fengguang /* 9215eb77a0SWu Fengguang * Include the creation of the trace points after defining the 9315eb77a0SWu Fengguang * wb_writeback_work structure and inline functions so that the definition 9415eb77a0SWu Fengguang * remains local to this file. 9515eb77a0SWu Fengguang */ 9615eb77a0SWu Fengguang #define CREATE_TRACE_POINTS 9715eb77a0SWu Fengguang #include <trace/events/writeback.h> 9815eb77a0SWu Fengguang 99774016b2SSteven Whitehouse EXPORT_TRACEPOINT_SYMBOL_GPL(wbc_writepage); 100774016b2SSteven Whitehouse 101d6c10f1fSTejun Heo static bool wb_io_lists_populated(struct bdi_writeback *wb) 102d6c10f1fSTejun Heo { 103d6c10f1fSTejun Heo if (wb_has_dirty_io(wb)) { 104d6c10f1fSTejun Heo return false; 105d6c10f1fSTejun Heo } else { 106d6c10f1fSTejun Heo set_bit(WB_has_dirty_io, &wb->state); 10795a46c65STejun Heo WARN_ON_ONCE(!wb->avg_write_bandwidth); 108766a9d6eSTejun Heo atomic_long_add(wb->avg_write_bandwidth, 109766a9d6eSTejun Heo &wb->bdi->tot_write_bandwidth); 110d6c10f1fSTejun Heo return true; 111d6c10f1fSTejun Heo } 112d6c10f1fSTejun Heo } 113d6c10f1fSTejun Heo 114d6c10f1fSTejun Heo static void wb_io_lists_depopulated(struct bdi_writeback *wb) 115d6c10f1fSTejun Heo { 116d6c10f1fSTejun Heo if (wb_has_dirty_io(wb) && list_empty(&wb->b_dirty) && 117766a9d6eSTejun Heo list_empty(&wb->b_io) && list_empty(&wb->b_more_io)) { 118d6c10f1fSTejun Heo clear_bit(WB_has_dirty_io, &wb->state); 11995a46c65STejun Heo WARN_ON_ONCE(atomic_long_sub_return(wb->avg_write_bandwidth, 12095a46c65STejun Heo &wb->bdi->tot_write_bandwidth) < 0); 121766a9d6eSTejun Heo } 122d6c10f1fSTejun Heo } 123d6c10f1fSTejun Heo 124d6c10f1fSTejun Heo /** 125d6c10f1fSTejun Heo * inode_wb_list_move_locked - move an inode onto a bdi_writeback IO list 126d6c10f1fSTejun Heo * @inode: inode to be moved 127d6c10f1fSTejun Heo * @wb: target bdi_writeback 128d6c10f1fSTejun Heo * @head: one of @wb->b_{dirty|io|more_io} 129d6c10f1fSTejun Heo * 130d6c10f1fSTejun Heo * Move @inode->i_wb_list to @list of @wb and set %WB_has_dirty_io. 131d6c10f1fSTejun Heo * Returns %true if @inode is the first occupant of the !dirty_time IO 132d6c10f1fSTejun Heo * lists; otherwise, %false. 133d6c10f1fSTejun Heo */ 134d6c10f1fSTejun Heo static bool inode_wb_list_move_locked(struct inode *inode, 135d6c10f1fSTejun Heo struct bdi_writeback *wb, 136d6c10f1fSTejun Heo struct list_head *head) 137d6c10f1fSTejun Heo { 138d6c10f1fSTejun Heo assert_spin_locked(&wb->list_lock); 139d6c10f1fSTejun Heo 140d6c10f1fSTejun Heo list_move(&inode->i_wb_list, head); 141d6c10f1fSTejun Heo 142d6c10f1fSTejun Heo /* dirty_time doesn't count as dirty_io until expiration */ 143d6c10f1fSTejun Heo if (head != &wb->b_dirty_time) 144d6c10f1fSTejun Heo return wb_io_lists_populated(wb); 145d6c10f1fSTejun Heo 146d6c10f1fSTejun Heo wb_io_lists_depopulated(wb); 147d6c10f1fSTejun Heo return false; 148d6c10f1fSTejun Heo } 149d6c10f1fSTejun Heo 150d6c10f1fSTejun Heo /** 151d6c10f1fSTejun Heo * inode_wb_list_del_locked - remove an inode from its bdi_writeback IO list 152d6c10f1fSTejun Heo * @inode: inode to be removed 153d6c10f1fSTejun Heo * @wb: bdi_writeback @inode is being removed from 154d6c10f1fSTejun Heo * 155d6c10f1fSTejun Heo * Remove @inode which may be on one of @wb->b_{dirty|io|more_io} lists and 156d6c10f1fSTejun Heo * clear %WB_has_dirty_io if all are empty afterwards. 157d6c10f1fSTejun Heo */ 158d6c10f1fSTejun Heo static void inode_wb_list_del_locked(struct inode *inode, 159d6c10f1fSTejun Heo struct bdi_writeback *wb) 160d6c10f1fSTejun Heo { 161d6c10f1fSTejun Heo assert_spin_locked(&wb->list_lock); 162d6c10f1fSTejun Heo 163d6c10f1fSTejun Heo list_del_init(&inode->i_wb_list); 164d6c10f1fSTejun Heo wb_io_lists_depopulated(wb); 165d6c10f1fSTejun Heo } 166d6c10f1fSTejun Heo 167f0054bb1STejun Heo static void wb_wakeup(struct bdi_writeback *wb) 1685acda9d1SJan Kara { 169f0054bb1STejun Heo spin_lock_bh(&wb->work_lock); 170f0054bb1STejun Heo if (test_bit(WB_registered, &wb->state)) 171f0054bb1STejun Heo mod_delayed_work(bdi_wq, &wb->dwork, 0); 172f0054bb1STejun Heo spin_unlock_bh(&wb->work_lock); 1735acda9d1SJan Kara } 1745acda9d1SJan Kara 175f0054bb1STejun Heo static void wb_queue_work(struct bdi_writeback *wb, 1766585027aSJan Kara struct wb_writeback_work *work) 1776585027aSJan Kara { 178f0054bb1STejun Heo trace_writeback_queue(wb->bdi, work); 1796585027aSJan Kara 180f0054bb1STejun Heo spin_lock_bh(&wb->work_lock); 181cc395d7fSTejun Heo if (!test_bit(WB_registered, &wb->state)) 1825acda9d1SJan Kara goto out_unlock; 183cc395d7fSTejun Heo if (work->done) 184cc395d7fSTejun Heo atomic_inc(&work->done->cnt); 185f0054bb1STejun Heo list_add_tail(&work->list, &wb->work_list); 186f0054bb1STejun Heo mod_delayed_work(bdi_wq, &wb->dwork, 0); 1875acda9d1SJan Kara out_unlock: 188f0054bb1STejun Heo spin_unlock_bh(&wb->work_lock); 18903ba3782SJens Axboe } 1901da177e4SLinus Torvalds 191cc395d7fSTejun Heo /** 192cc395d7fSTejun Heo * wb_wait_for_completion - wait for completion of bdi_writeback_works 193cc395d7fSTejun Heo * @bdi: bdi work items were issued to 194cc395d7fSTejun Heo * @done: target wb_completion 195cc395d7fSTejun Heo * 196cc395d7fSTejun Heo * Wait for one or more work items issued to @bdi with their ->done field 197cc395d7fSTejun Heo * set to @done, which should have been defined with 198cc395d7fSTejun Heo * DEFINE_WB_COMPLETION_ONSTACK(). This function returns after all such 199cc395d7fSTejun Heo * work items are completed. Work items which are waited upon aren't freed 200cc395d7fSTejun Heo * automatically on completion. 201cc395d7fSTejun Heo */ 202cc395d7fSTejun Heo static void wb_wait_for_completion(struct backing_dev_info *bdi, 203cc395d7fSTejun Heo struct wb_completion *done) 204cc395d7fSTejun Heo { 205cc395d7fSTejun Heo atomic_dec(&done->cnt); /* put down the initial count */ 206cc395d7fSTejun Heo wait_event(bdi->wb_waitq, !atomic_read(&done->cnt)); 207cc395d7fSTejun Heo } 208cc395d7fSTejun Heo 209703c2708STejun Heo #ifdef CONFIG_CGROUP_WRITEBACK 210703c2708STejun Heo 211703c2708STejun Heo /** 212703c2708STejun Heo * inode_congested - test whether an inode is congested 213703c2708STejun Heo * @inode: inode to test for congestion 214703c2708STejun Heo * @cong_bits: mask of WB_[a]sync_congested bits to test 215703c2708STejun Heo * 216703c2708STejun Heo * Tests whether @inode is congested. @cong_bits is the mask of congestion 217703c2708STejun Heo * bits to test and the return value is the mask of set bits. 218703c2708STejun Heo * 219703c2708STejun Heo * If cgroup writeback is enabled for @inode, the congestion state is 220703c2708STejun Heo * determined by whether the cgwb (cgroup bdi_writeback) for the blkcg 221703c2708STejun Heo * associated with @inode is congested; otherwise, the root wb's congestion 222703c2708STejun Heo * state is used. 223703c2708STejun Heo */ 224703c2708STejun Heo int inode_congested(struct inode *inode, int cong_bits) 225703c2708STejun Heo { 226703c2708STejun Heo if (inode) { 227703c2708STejun Heo struct bdi_writeback *wb = inode_to_wb(inode); 228703c2708STejun Heo if (wb) 229703c2708STejun Heo return wb_congested(wb, cong_bits); 230703c2708STejun Heo } 231703c2708STejun Heo 232703c2708STejun Heo return wb_congested(&inode_to_bdi(inode)->wb, cong_bits); 233703c2708STejun Heo } 234703c2708STejun Heo EXPORT_SYMBOL_GPL(inode_congested); 235703c2708STejun Heo 236f2b65121STejun Heo /** 237f2b65121STejun Heo * wb_split_bdi_pages - split nr_pages to write according to bandwidth 238f2b65121STejun Heo * @wb: target bdi_writeback to split @nr_pages to 239f2b65121STejun Heo * @nr_pages: number of pages to write for the whole bdi 240f2b65121STejun Heo * 241f2b65121STejun Heo * Split @wb's portion of @nr_pages according to @wb's write bandwidth in 242f2b65121STejun Heo * relation to the total write bandwidth of all wb's w/ dirty inodes on 243f2b65121STejun Heo * @wb->bdi. 244f2b65121STejun Heo */ 245f2b65121STejun Heo static long wb_split_bdi_pages(struct bdi_writeback *wb, long nr_pages) 246f2b65121STejun Heo { 247f2b65121STejun Heo unsigned long this_bw = wb->avg_write_bandwidth; 248f2b65121STejun Heo unsigned long tot_bw = atomic_long_read(&wb->bdi->tot_write_bandwidth); 249f2b65121STejun Heo 250f2b65121STejun Heo if (nr_pages == LONG_MAX) 251f2b65121STejun Heo return LONG_MAX; 252f2b65121STejun Heo 253f2b65121STejun Heo /* 254f2b65121STejun Heo * This may be called on clean wb's and proportional distribution 255f2b65121STejun Heo * may not make sense, just use the original @nr_pages in those 256f2b65121STejun Heo * cases. In general, we wanna err on the side of writing more. 257f2b65121STejun Heo */ 258f2b65121STejun Heo if (!tot_bw || this_bw >= tot_bw) 259f2b65121STejun Heo return nr_pages; 260f2b65121STejun Heo else 261f2b65121STejun Heo return DIV_ROUND_UP_ULL((u64)nr_pages * this_bw, tot_bw); 262f2b65121STejun Heo } 263f2b65121STejun Heo 264f2b65121STejun Heo #else /* CONFIG_CGROUP_WRITEBACK */ 265f2b65121STejun Heo 266f2b65121STejun Heo static long wb_split_bdi_pages(struct bdi_writeback *wb, long nr_pages) 267f2b65121STejun Heo { 268f2b65121STejun Heo return nr_pages; 269f2b65121STejun Heo } 270f2b65121STejun Heo 271703c2708STejun Heo #endif /* CONFIG_CGROUP_WRITEBACK */ 272703c2708STejun Heo 273c00ddad3STejun Heo void wb_start_writeback(struct bdi_writeback *wb, long nr_pages, 274c00ddad3STejun Heo bool range_cyclic, enum wb_reason reason) 275b6e51316SJens Axboe { 276c00ddad3STejun Heo struct wb_writeback_work *work; 277c00ddad3STejun Heo 278c00ddad3STejun Heo if (!wb_has_dirty_io(wb)) 279c00ddad3STejun Heo return; 280c00ddad3STejun Heo 281c00ddad3STejun Heo /* 282c00ddad3STejun Heo * This is WB_SYNC_NONE writeback, so if allocation fails just 283c00ddad3STejun Heo * wakeup the thread for old dirty data writeback 284c00ddad3STejun Heo */ 285c00ddad3STejun Heo work = kzalloc(sizeof(*work), GFP_ATOMIC); 286c00ddad3STejun Heo if (!work) { 287c00ddad3STejun Heo trace_writeback_nowork(wb->bdi); 288c00ddad3STejun Heo wb_wakeup(wb); 289c00ddad3STejun Heo return; 290c00ddad3STejun Heo } 291c00ddad3STejun Heo 292c00ddad3STejun Heo work->sync_mode = WB_SYNC_NONE; 293c00ddad3STejun Heo work->nr_pages = nr_pages; 294c00ddad3STejun Heo work->range_cyclic = range_cyclic; 295c00ddad3STejun Heo work->reason = reason; 296ac7b19a3STejun Heo work->auto_free = 1; 297c00ddad3STejun Heo 298c00ddad3STejun Heo wb_queue_work(wb, work); 299d3ddec76SWu Fengguang } 300d3ddec76SWu Fengguang 301c5444198SChristoph Hellwig /** 3029ecf4866STejun Heo * wb_start_background_writeback - start background writeback 3039ecf4866STejun Heo * @wb: bdi_writback to write from 304c5444198SChristoph Hellwig * 305c5444198SChristoph Hellwig * Description: 3066585027aSJan Kara * This makes sure WB_SYNC_NONE background writeback happens. When 3079ecf4866STejun Heo * this function returns, it is only guaranteed that for given wb 3086585027aSJan Kara * some IO is happening if we are over background dirty threshold. 3096585027aSJan Kara * Caller need not hold sb s_umount semaphore. 310c5444198SChristoph Hellwig */ 3119ecf4866STejun Heo void wb_start_background_writeback(struct bdi_writeback *wb) 312c5444198SChristoph Hellwig { 3136585027aSJan Kara /* 3146585027aSJan Kara * We just wake up the flusher thread. It will perform background 3156585027aSJan Kara * writeback as soon as there is no other work to do. 3166585027aSJan Kara */ 3179ecf4866STejun Heo trace_writeback_wake_background(wb->bdi); 3189ecf4866STejun Heo wb_wakeup(wb); 3191da177e4SLinus Torvalds } 3201da177e4SLinus Torvalds 3211da177e4SLinus Torvalds /* 322a66979abSDave Chinner * Remove the inode from the writeback list it is on. 323a66979abSDave Chinner */ 324a66979abSDave Chinner void inode_wb_list_del(struct inode *inode) 325a66979abSDave Chinner { 32652ebea74STejun Heo struct bdi_writeback *wb = inode_to_wb(inode); 327a66979abSDave Chinner 32852ebea74STejun Heo spin_lock(&wb->list_lock); 329d6c10f1fSTejun Heo inode_wb_list_del_locked(inode, wb); 33052ebea74STejun Heo spin_unlock(&wb->list_lock); 331f758eeabSChristoph Hellwig } 332a66979abSDave Chinner 333a66979abSDave Chinner /* 3346610a0bcSAndrew Morton * Redirty an inode: set its when-it-was dirtied timestamp and move it to the 3356610a0bcSAndrew Morton * furthest end of its superblock's dirty-inode list. 3366610a0bcSAndrew Morton * 3376610a0bcSAndrew Morton * Before stamping the inode's ->dirtied_when, we check to see whether it is 33866f3b8e2SJens Axboe * already the most-recently-dirtied inode on the b_dirty list. If that is 3396610a0bcSAndrew Morton * the case then the inode must have been redirtied while it was being written 3406610a0bcSAndrew Morton * out and we don't reset its dirtied_when. 3416610a0bcSAndrew Morton */ 342f758eeabSChristoph Hellwig static void redirty_tail(struct inode *inode, struct bdi_writeback *wb) 3436610a0bcSAndrew Morton { 34403ba3782SJens Axboe if (!list_empty(&wb->b_dirty)) { 34566f3b8e2SJens Axboe struct inode *tail; 3466610a0bcSAndrew Morton 3477ccf19a8SNick Piggin tail = wb_inode(wb->b_dirty.next); 34866f3b8e2SJens Axboe if (time_before(inode->dirtied_when, tail->dirtied_when)) 3496610a0bcSAndrew Morton inode->dirtied_when = jiffies; 3506610a0bcSAndrew Morton } 351d6c10f1fSTejun Heo inode_wb_list_move_locked(inode, wb, &wb->b_dirty); 3526610a0bcSAndrew Morton } 3536610a0bcSAndrew Morton 3546610a0bcSAndrew Morton /* 35566f3b8e2SJens Axboe * requeue inode for re-scanning after bdi->b_io list is exhausted. 356c986d1e2SAndrew Morton */ 357f758eeabSChristoph Hellwig static void requeue_io(struct inode *inode, struct bdi_writeback *wb) 358c986d1e2SAndrew Morton { 359d6c10f1fSTejun Heo inode_wb_list_move_locked(inode, wb, &wb->b_more_io); 360c986d1e2SAndrew Morton } 361c986d1e2SAndrew Morton 3621c0eeaf5SJoern Engel static void inode_sync_complete(struct inode *inode) 3631c0eeaf5SJoern Engel { 364365b94aeSJan Kara inode->i_state &= ~I_SYNC; 3654eff96ddSJan Kara /* If inode is clean an unused, put it into LRU now... */ 3664eff96ddSJan Kara inode_add_lru(inode); 367365b94aeSJan Kara /* Waiters must see I_SYNC cleared before being woken up */ 3681c0eeaf5SJoern Engel smp_mb(); 3691c0eeaf5SJoern Engel wake_up_bit(&inode->i_state, __I_SYNC); 3701c0eeaf5SJoern Engel } 3711c0eeaf5SJoern Engel 372d2caa3c5SJeff Layton static bool inode_dirtied_after(struct inode *inode, unsigned long t) 373d2caa3c5SJeff Layton { 374d2caa3c5SJeff Layton bool ret = time_after(inode->dirtied_when, t); 375d2caa3c5SJeff Layton #ifndef CONFIG_64BIT 376d2caa3c5SJeff Layton /* 377d2caa3c5SJeff Layton * For inodes being constantly redirtied, dirtied_when can get stuck. 378d2caa3c5SJeff Layton * It _appears_ to be in the future, but is actually in distant past. 379d2caa3c5SJeff Layton * This test is necessary to prevent such wrapped-around relative times 3805b0830cbSJens Axboe * from permanently stopping the whole bdi writeback. 381d2caa3c5SJeff Layton */ 382d2caa3c5SJeff Layton ret = ret && time_before_eq(inode->dirtied_when, jiffies); 383d2caa3c5SJeff Layton #endif 384d2caa3c5SJeff Layton return ret; 385d2caa3c5SJeff Layton } 386d2caa3c5SJeff Layton 3870ae45f63STheodore Ts'o #define EXPIRE_DIRTY_ATIME 0x0001 3880ae45f63STheodore Ts'o 389c986d1e2SAndrew Morton /* 3900e2f2b23SWang Sheng-Hui * Move expired (dirtied before work->older_than_this) dirty inodes from 391697e6fedSJan Kara * @delaying_queue to @dispatch_queue. 3922c136579SFengguang Wu */ 393e84d0a4fSWu Fengguang static int move_expired_inodes(struct list_head *delaying_queue, 3942c136579SFengguang Wu struct list_head *dispatch_queue, 3950ae45f63STheodore Ts'o int flags, 396ad4e38ddSCurt Wohlgemuth struct wb_writeback_work *work) 3972c136579SFengguang Wu { 3980ae45f63STheodore Ts'o unsigned long *older_than_this = NULL; 3990ae45f63STheodore Ts'o unsigned long expire_time; 4005c03449dSShaohua Li LIST_HEAD(tmp); 4015c03449dSShaohua Li struct list_head *pos, *node; 402cf137307SJens Axboe struct super_block *sb = NULL; 4035c03449dSShaohua Li struct inode *inode; 404cf137307SJens Axboe int do_sb_sort = 0; 405e84d0a4fSWu Fengguang int moved = 0; 4065c03449dSShaohua Li 4070ae45f63STheodore Ts'o if ((flags & EXPIRE_DIRTY_ATIME) == 0) 4080ae45f63STheodore Ts'o older_than_this = work->older_than_this; 409a2f48706STheodore Ts'o else if (!work->for_sync) { 410a2f48706STheodore Ts'o expire_time = jiffies - (dirtytime_expire_interval * HZ); 4110ae45f63STheodore Ts'o older_than_this = &expire_time; 4120ae45f63STheodore Ts'o } 4132c136579SFengguang Wu while (!list_empty(delaying_queue)) { 4147ccf19a8SNick Piggin inode = wb_inode(delaying_queue->prev); 4150ae45f63STheodore Ts'o if (older_than_this && 4160ae45f63STheodore Ts'o inode_dirtied_after(inode, *older_than_this)) 4172c136579SFengguang Wu break; 418a8855990SJan Kara list_move(&inode->i_wb_list, &tmp); 419a8855990SJan Kara moved++; 4200ae45f63STheodore Ts'o if (flags & EXPIRE_DIRTY_ATIME) 4210ae45f63STheodore Ts'o set_bit(__I_DIRTY_TIME_EXPIRED, &inode->i_state); 422a8855990SJan Kara if (sb_is_blkdev_sb(inode->i_sb)) 423a8855990SJan Kara continue; 424cf137307SJens Axboe if (sb && sb != inode->i_sb) 425cf137307SJens Axboe do_sb_sort = 1; 426cf137307SJens Axboe sb = inode->i_sb; 4275c03449dSShaohua Li } 4285c03449dSShaohua Li 429cf137307SJens Axboe /* just one sb in list, splice to dispatch_queue and we're done */ 430cf137307SJens Axboe if (!do_sb_sort) { 431cf137307SJens Axboe list_splice(&tmp, dispatch_queue); 432e84d0a4fSWu Fengguang goto out; 433cf137307SJens Axboe } 434cf137307SJens Axboe 4355c03449dSShaohua Li /* Move inodes from one superblock together */ 4365c03449dSShaohua Li while (!list_empty(&tmp)) { 4377ccf19a8SNick Piggin sb = wb_inode(tmp.prev)->i_sb; 4385c03449dSShaohua Li list_for_each_prev_safe(pos, node, &tmp) { 4397ccf19a8SNick Piggin inode = wb_inode(pos); 4405c03449dSShaohua Li if (inode->i_sb == sb) 4417ccf19a8SNick Piggin list_move(&inode->i_wb_list, dispatch_queue); 4422c136579SFengguang Wu } 4432c136579SFengguang Wu } 444e84d0a4fSWu Fengguang out: 445e84d0a4fSWu Fengguang return moved; 4465c03449dSShaohua Li } 4472c136579SFengguang Wu 4482c136579SFengguang Wu /* 4492c136579SFengguang Wu * Queue all expired dirty inodes for io, eldest first. 4504ea879b9SWu Fengguang * Before 4514ea879b9SWu Fengguang * newly dirtied b_dirty b_io b_more_io 4524ea879b9SWu Fengguang * =============> gf edc BA 4534ea879b9SWu Fengguang * After 4544ea879b9SWu Fengguang * newly dirtied b_dirty b_io b_more_io 4554ea879b9SWu Fengguang * =============> g fBAedc 4564ea879b9SWu Fengguang * | 4574ea879b9SWu Fengguang * +--> dequeue for IO 4582c136579SFengguang Wu */ 459ad4e38ddSCurt Wohlgemuth static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work) 4602c136579SFengguang Wu { 461e84d0a4fSWu Fengguang int moved; 4620ae45f63STheodore Ts'o 463f758eeabSChristoph Hellwig assert_spin_locked(&wb->list_lock); 4644ea879b9SWu Fengguang list_splice_init(&wb->b_more_io, &wb->b_io); 4650ae45f63STheodore Ts'o moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, 0, work); 4660ae45f63STheodore Ts'o moved += move_expired_inodes(&wb->b_dirty_time, &wb->b_io, 4670ae45f63STheodore Ts'o EXPIRE_DIRTY_ATIME, work); 468d6c10f1fSTejun Heo if (moved) 469d6c10f1fSTejun Heo wb_io_lists_populated(wb); 470ad4e38ddSCurt Wohlgemuth trace_writeback_queue_io(wb, work, moved); 47166f3b8e2SJens Axboe } 47266f3b8e2SJens Axboe 473a9185b41SChristoph Hellwig static int write_inode(struct inode *inode, struct writeback_control *wbc) 47466f3b8e2SJens Axboe { 4759fb0a7daSTejun Heo int ret; 4769fb0a7daSTejun Heo 4779fb0a7daSTejun Heo if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) { 4789fb0a7daSTejun Heo trace_writeback_write_inode_start(inode, wbc); 4799fb0a7daSTejun Heo ret = inode->i_sb->s_op->write_inode(inode, wbc); 4809fb0a7daSTejun Heo trace_writeback_write_inode(inode, wbc); 4819fb0a7daSTejun Heo return ret; 4829fb0a7daSTejun Heo } 48303ba3782SJens Axboe return 0; 48466f3b8e2SJens Axboe } 48508d8e974SFengguang Wu 4862c136579SFengguang Wu /* 487169ebd90SJan Kara * Wait for writeback on an inode to complete. Called with i_lock held. 488169ebd90SJan Kara * Caller must make sure inode cannot go away when we drop i_lock. 48901c03194SChristoph Hellwig */ 490169ebd90SJan Kara static void __inode_wait_for_writeback(struct inode *inode) 491169ebd90SJan Kara __releases(inode->i_lock) 492169ebd90SJan Kara __acquires(inode->i_lock) 49301c03194SChristoph Hellwig { 49401c03194SChristoph Hellwig DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); 49501c03194SChristoph Hellwig wait_queue_head_t *wqh; 49601c03194SChristoph Hellwig 49701c03194SChristoph Hellwig wqh = bit_waitqueue(&inode->i_state, __I_SYNC); 49858a9d3d8SRichard Kennedy while (inode->i_state & I_SYNC) { 499250df6edSDave Chinner spin_unlock(&inode->i_lock); 50074316201SNeilBrown __wait_on_bit(wqh, &wq, bit_wait, 50174316201SNeilBrown TASK_UNINTERRUPTIBLE); 502250df6edSDave Chinner spin_lock(&inode->i_lock); 50358a9d3d8SRichard Kennedy } 50401c03194SChristoph Hellwig } 50501c03194SChristoph Hellwig 50601c03194SChristoph Hellwig /* 507169ebd90SJan Kara * Wait for writeback on an inode to complete. Caller must have inode pinned. 508169ebd90SJan Kara */ 509169ebd90SJan Kara void inode_wait_for_writeback(struct inode *inode) 510169ebd90SJan Kara { 511169ebd90SJan Kara spin_lock(&inode->i_lock); 512169ebd90SJan Kara __inode_wait_for_writeback(inode); 513169ebd90SJan Kara spin_unlock(&inode->i_lock); 514169ebd90SJan Kara } 515169ebd90SJan Kara 516169ebd90SJan Kara /* 517169ebd90SJan Kara * Sleep until I_SYNC is cleared. This function must be called with i_lock 518169ebd90SJan Kara * held and drops it. It is aimed for callers not holding any inode reference 519169ebd90SJan Kara * so once i_lock is dropped, inode can go away. 520169ebd90SJan Kara */ 521169ebd90SJan Kara static void inode_sleep_on_writeback(struct inode *inode) 522169ebd90SJan Kara __releases(inode->i_lock) 523169ebd90SJan Kara { 524169ebd90SJan Kara DEFINE_WAIT(wait); 525169ebd90SJan Kara wait_queue_head_t *wqh = bit_waitqueue(&inode->i_state, __I_SYNC); 526169ebd90SJan Kara int sleep; 527169ebd90SJan Kara 528169ebd90SJan Kara prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); 529169ebd90SJan Kara sleep = inode->i_state & I_SYNC; 530169ebd90SJan Kara spin_unlock(&inode->i_lock); 531169ebd90SJan Kara if (sleep) 532169ebd90SJan Kara schedule(); 533169ebd90SJan Kara finish_wait(wqh, &wait); 534169ebd90SJan Kara } 535169ebd90SJan Kara 536169ebd90SJan Kara /* 537ccb26b5aSJan Kara * Find proper writeback list for the inode depending on its current state and 538ccb26b5aSJan Kara * possibly also change of its state while we were doing writeback. Here we 539ccb26b5aSJan Kara * handle things such as livelock prevention or fairness of writeback among 540ccb26b5aSJan Kara * inodes. This function can be called only by flusher thread - noone else 541ccb26b5aSJan Kara * processes all inodes in writeback lists and requeueing inodes behind flusher 542ccb26b5aSJan Kara * thread's back can have unexpected consequences. 543ccb26b5aSJan Kara */ 544ccb26b5aSJan Kara static void requeue_inode(struct inode *inode, struct bdi_writeback *wb, 545ccb26b5aSJan Kara struct writeback_control *wbc) 546ccb26b5aSJan Kara { 547ccb26b5aSJan Kara if (inode->i_state & I_FREEING) 548ccb26b5aSJan Kara return; 549ccb26b5aSJan Kara 550ccb26b5aSJan Kara /* 551ccb26b5aSJan Kara * Sync livelock prevention. Each inode is tagged and synced in one 552ccb26b5aSJan Kara * shot. If still dirty, it will be redirty_tail()'ed below. Update 553ccb26b5aSJan Kara * the dirty time to prevent enqueue and sync it again. 554ccb26b5aSJan Kara */ 555ccb26b5aSJan Kara if ((inode->i_state & I_DIRTY) && 556ccb26b5aSJan Kara (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)) 557ccb26b5aSJan Kara inode->dirtied_when = jiffies; 558ccb26b5aSJan Kara 5594f8ad655SJan Kara if (wbc->pages_skipped) { 5604f8ad655SJan Kara /* 5614f8ad655SJan Kara * writeback is not making progress due to locked 5624f8ad655SJan Kara * buffers. Skip this inode for now. 5634f8ad655SJan Kara */ 5644f8ad655SJan Kara redirty_tail(inode, wb); 5654f8ad655SJan Kara return; 5664f8ad655SJan Kara } 5674f8ad655SJan Kara 568ccb26b5aSJan Kara if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) { 569ccb26b5aSJan Kara /* 570ccb26b5aSJan Kara * We didn't write back all the pages. nfs_writepages() 571ccb26b5aSJan Kara * sometimes bales out without doing anything. 572ccb26b5aSJan Kara */ 573ccb26b5aSJan Kara if (wbc->nr_to_write <= 0) { 574ccb26b5aSJan Kara /* Slice used up. Queue for next turn. */ 575ccb26b5aSJan Kara requeue_io(inode, wb); 576ccb26b5aSJan Kara } else { 577ccb26b5aSJan Kara /* 578ccb26b5aSJan Kara * Writeback blocked by something other than 579ccb26b5aSJan Kara * congestion. Delay the inode for some time to 580ccb26b5aSJan Kara * avoid spinning on the CPU (100% iowait) 581ccb26b5aSJan Kara * retrying writeback of the dirty page/inode 582ccb26b5aSJan Kara * that cannot be performed immediately. 583ccb26b5aSJan Kara */ 584ccb26b5aSJan Kara redirty_tail(inode, wb); 585ccb26b5aSJan Kara } 586ccb26b5aSJan Kara } else if (inode->i_state & I_DIRTY) { 587ccb26b5aSJan Kara /* 588ccb26b5aSJan Kara * Filesystems can dirty the inode during writeback operations, 589ccb26b5aSJan Kara * such as delayed allocation during submission or metadata 590ccb26b5aSJan Kara * updates after data IO completion. 591ccb26b5aSJan Kara */ 592ccb26b5aSJan Kara redirty_tail(inode, wb); 5930ae45f63STheodore Ts'o } else if (inode->i_state & I_DIRTY_TIME) { 594a2f48706STheodore Ts'o inode->dirtied_when = jiffies; 595d6c10f1fSTejun Heo inode_wb_list_move_locked(inode, wb, &wb->b_dirty_time); 596ccb26b5aSJan Kara } else { 597ccb26b5aSJan Kara /* The inode is clean. Remove from writeback lists. */ 598d6c10f1fSTejun Heo inode_wb_list_del_locked(inode, wb); 599ccb26b5aSJan Kara } 600ccb26b5aSJan Kara } 601ccb26b5aSJan Kara 602ccb26b5aSJan Kara /* 6034f8ad655SJan Kara * Write out an inode and its dirty pages. Do not update the writeback list 6044f8ad655SJan Kara * linkage. That is left to the caller. The caller is also responsible for 6054f8ad655SJan Kara * setting I_SYNC flag and calling inode_sync_complete() to clear it. 6061da177e4SLinus Torvalds */ 6071da177e4SLinus Torvalds static int 608cd8ed2a4SYan Hong __writeback_single_inode(struct inode *inode, struct writeback_control *wbc) 6091da177e4SLinus Torvalds { 6101da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 611251d6a47SWu Fengguang long nr_to_write = wbc->nr_to_write; 61201c03194SChristoph Hellwig unsigned dirty; 6131da177e4SLinus Torvalds int ret; 6141da177e4SLinus Torvalds 6154f8ad655SJan Kara WARN_ON(!(inode->i_state & I_SYNC)); 6161da177e4SLinus Torvalds 6179fb0a7daSTejun Heo trace_writeback_single_inode_start(inode, wbc, nr_to_write); 6189fb0a7daSTejun Heo 6191da177e4SLinus Torvalds ret = do_writepages(mapping, wbc); 6201da177e4SLinus Torvalds 62126821ed4SChristoph Hellwig /* 62226821ed4SChristoph Hellwig * Make sure to wait on the data before writing out the metadata. 62326821ed4SChristoph Hellwig * This is important for filesystems that modify metadata on data 6247747bd4bSDave Chinner * I/O completion. We don't do it for sync(2) writeback because it has a 6257747bd4bSDave Chinner * separate, external IO completion path and ->sync_fs for guaranteeing 6267747bd4bSDave Chinner * inode metadata is written back correctly. 62726821ed4SChristoph Hellwig */ 6287747bd4bSDave Chinner if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) { 62926821ed4SChristoph Hellwig int err = filemap_fdatawait(mapping); 6301da177e4SLinus Torvalds if (ret == 0) 6311da177e4SLinus Torvalds ret = err; 6321da177e4SLinus Torvalds } 6331da177e4SLinus Torvalds 6345547e8aaSDmitry Monakhov /* 6355547e8aaSDmitry Monakhov * Some filesystems may redirty the inode during the writeback 6365547e8aaSDmitry Monakhov * due to delalloc, clear dirty metadata flags right before 6375547e8aaSDmitry Monakhov * write_inode() 6385547e8aaSDmitry Monakhov */ 639250df6edSDave Chinner spin_lock(&inode->i_lock); 6409c6ac78eSTejun Heo 6415547e8aaSDmitry Monakhov dirty = inode->i_state & I_DIRTY; 642a2f48706STheodore Ts'o if (inode->i_state & I_DIRTY_TIME) { 643a2f48706STheodore Ts'o if ((dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) || 644a2f48706STheodore Ts'o unlikely(inode->i_state & I_DIRTY_TIME_EXPIRED) || 645a2f48706STheodore Ts'o unlikely(time_after(jiffies, 646a2f48706STheodore Ts'o (inode->dirtied_time_when + 647a2f48706STheodore Ts'o dirtytime_expire_interval * HZ)))) { 6480ae45f63STheodore Ts'o dirty |= I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED; 6490ae45f63STheodore Ts'o trace_writeback_lazytime(inode); 6500ae45f63STheodore Ts'o } 651a2f48706STheodore Ts'o } else 652a2f48706STheodore Ts'o inode->i_state &= ~I_DIRTY_TIME_EXPIRED; 6530ae45f63STheodore Ts'o inode->i_state &= ~dirty; 6549c6ac78eSTejun Heo 6559c6ac78eSTejun Heo /* 6569c6ac78eSTejun Heo * Paired with smp_mb() in __mark_inode_dirty(). This allows 6579c6ac78eSTejun Heo * __mark_inode_dirty() to test i_state without grabbing i_lock - 6589c6ac78eSTejun Heo * either they see the I_DIRTY bits cleared or we see the dirtied 6599c6ac78eSTejun Heo * inode. 6609c6ac78eSTejun Heo * 6619c6ac78eSTejun Heo * I_DIRTY_PAGES is always cleared together above even if @mapping 6629c6ac78eSTejun Heo * still has dirty pages. The flag is reinstated after smp_mb() if 6639c6ac78eSTejun Heo * necessary. This guarantees that either __mark_inode_dirty() 6649c6ac78eSTejun Heo * sees clear I_DIRTY_PAGES or we see PAGECACHE_TAG_DIRTY. 6659c6ac78eSTejun Heo */ 6669c6ac78eSTejun Heo smp_mb(); 6679c6ac78eSTejun Heo 6689c6ac78eSTejun Heo if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) 6699c6ac78eSTejun Heo inode->i_state |= I_DIRTY_PAGES; 6709c6ac78eSTejun Heo 671250df6edSDave Chinner spin_unlock(&inode->i_lock); 6729c6ac78eSTejun Heo 6730ae45f63STheodore Ts'o if (dirty & I_DIRTY_TIME) 6740ae45f63STheodore Ts'o mark_inode_dirty_sync(inode); 67526821ed4SChristoph Hellwig /* Don't write the inode if only I_DIRTY_PAGES was set */ 6760ae45f63STheodore Ts'o if (dirty & ~I_DIRTY_PAGES) { 677a9185b41SChristoph Hellwig int err = write_inode(inode, wbc); 6781da177e4SLinus Torvalds if (ret == 0) 6791da177e4SLinus Torvalds ret = err; 6801da177e4SLinus Torvalds } 6814f8ad655SJan Kara trace_writeback_single_inode(inode, wbc, nr_to_write); 6824f8ad655SJan Kara return ret; 6834f8ad655SJan Kara } 6844f8ad655SJan Kara 6854f8ad655SJan Kara /* 6864f8ad655SJan Kara * Write out an inode's dirty pages. Either the caller has an active reference 6874f8ad655SJan Kara * on the inode or the inode has I_WILL_FREE set. 6884f8ad655SJan Kara * 6894f8ad655SJan Kara * This function is designed to be called for writing back one inode which 6904f8ad655SJan Kara * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode() 6914f8ad655SJan Kara * and does more profound writeback list handling in writeback_sb_inodes(). 6924f8ad655SJan Kara */ 6934f8ad655SJan Kara static int 6944f8ad655SJan Kara writeback_single_inode(struct inode *inode, struct bdi_writeback *wb, 6954f8ad655SJan Kara struct writeback_control *wbc) 6964f8ad655SJan Kara { 6974f8ad655SJan Kara int ret = 0; 6984f8ad655SJan Kara 6994f8ad655SJan Kara spin_lock(&inode->i_lock); 7004f8ad655SJan Kara if (!atomic_read(&inode->i_count)) 7014f8ad655SJan Kara WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); 7024f8ad655SJan Kara else 7034f8ad655SJan Kara WARN_ON(inode->i_state & I_WILL_FREE); 7044f8ad655SJan Kara 7054f8ad655SJan Kara if (inode->i_state & I_SYNC) { 7064f8ad655SJan Kara if (wbc->sync_mode != WB_SYNC_ALL) 7074f8ad655SJan Kara goto out; 7084f8ad655SJan Kara /* 709169ebd90SJan Kara * It's a data-integrity sync. We must wait. Since callers hold 710169ebd90SJan Kara * inode reference or inode has I_WILL_FREE set, it cannot go 711169ebd90SJan Kara * away under us. 7124f8ad655SJan Kara */ 713169ebd90SJan Kara __inode_wait_for_writeback(inode); 7144f8ad655SJan Kara } 7154f8ad655SJan Kara WARN_ON(inode->i_state & I_SYNC); 7164f8ad655SJan Kara /* 717f9b0e058SJan Kara * Skip inode if it is clean and we have no outstanding writeback in 718f9b0e058SJan Kara * WB_SYNC_ALL mode. We don't want to mess with writeback lists in this 719f9b0e058SJan Kara * function since flusher thread may be doing for example sync in 720f9b0e058SJan Kara * parallel and if we move the inode, it could get skipped. So here we 721f9b0e058SJan Kara * make sure inode is on some writeback list and leave it there unless 722f9b0e058SJan Kara * we have completely cleaned the inode. 7234f8ad655SJan Kara */ 7240ae45f63STheodore Ts'o if (!(inode->i_state & I_DIRTY_ALL) && 725f9b0e058SJan Kara (wbc->sync_mode != WB_SYNC_ALL || 726f9b0e058SJan Kara !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK))) 7274f8ad655SJan Kara goto out; 7284f8ad655SJan Kara inode->i_state |= I_SYNC; 7294f8ad655SJan Kara spin_unlock(&inode->i_lock); 7304f8ad655SJan Kara 731cd8ed2a4SYan Hong ret = __writeback_single_inode(inode, wbc); 7321da177e4SLinus Torvalds 733f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 734250df6edSDave Chinner spin_lock(&inode->i_lock); 7354f8ad655SJan Kara /* 7364f8ad655SJan Kara * If inode is clean, remove it from writeback lists. Otherwise don't 7374f8ad655SJan Kara * touch it. See comment above for explanation. 7384f8ad655SJan Kara */ 7390ae45f63STheodore Ts'o if (!(inode->i_state & I_DIRTY_ALL)) 740d6c10f1fSTejun Heo inode_wb_list_del_locked(inode, wb); 7414f8ad655SJan Kara spin_unlock(&wb->list_lock); 7421c0eeaf5SJoern Engel inode_sync_complete(inode); 7434f8ad655SJan Kara out: 7444f8ad655SJan Kara spin_unlock(&inode->i_lock); 7451da177e4SLinus Torvalds return ret; 7461da177e4SLinus Torvalds } 7471da177e4SLinus Torvalds 748a88a341aSTejun Heo static long writeback_chunk_size(struct bdi_writeback *wb, 7491a12d8bdSWu Fengguang struct wb_writeback_work *work) 750d46db3d5SWu Fengguang { 751d46db3d5SWu Fengguang long pages; 752d46db3d5SWu Fengguang 753d46db3d5SWu Fengguang /* 754d46db3d5SWu Fengguang * WB_SYNC_ALL mode does livelock avoidance by syncing dirty 755d46db3d5SWu Fengguang * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX 756d46db3d5SWu Fengguang * here avoids calling into writeback_inodes_wb() more than once. 757d46db3d5SWu Fengguang * 758d46db3d5SWu Fengguang * The intended call sequence for WB_SYNC_ALL writeback is: 759d46db3d5SWu Fengguang * 760d46db3d5SWu Fengguang * wb_writeback() 761d46db3d5SWu Fengguang * writeback_sb_inodes() <== called only once 762d46db3d5SWu Fengguang * write_cache_pages() <== called once for each inode 763d46db3d5SWu Fengguang * (quickly) tag currently dirty pages 764d46db3d5SWu Fengguang * (maybe slowly) sync all tagged pages 765d46db3d5SWu Fengguang */ 766d46db3d5SWu Fengguang if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages) 767d46db3d5SWu Fengguang pages = LONG_MAX; 7681a12d8bdSWu Fengguang else { 769a88a341aSTejun Heo pages = min(wb->avg_write_bandwidth / 2, 7701a12d8bdSWu Fengguang global_dirty_limit / DIRTY_SCOPE); 7711a12d8bdSWu Fengguang pages = min(pages, work->nr_pages); 7721a12d8bdSWu Fengguang pages = round_down(pages + MIN_WRITEBACK_PAGES, 7731a12d8bdSWu Fengguang MIN_WRITEBACK_PAGES); 7741a12d8bdSWu Fengguang } 775d46db3d5SWu Fengguang 776d46db3d5SWu Fengguang return pages; 777d46db3d5SWu Fengguang } 778d46db3d5SWu Fengguang 77903ba3782SJens Axboe /* 780f11c9c5cSEdward Shishkin * Write a portion of b_io inodes which belong to @sb. 781edadfb10SChristoph Hellwig * 782d46db3d5SWu Fengguang * Return the number of pages and/or inodes written. 783f11c9c5cSEdward Shishkin */ 784d46db3d5SWu Fengguang static long writeback_sb_inodes(struct super_block *sb, 785d46db3d5SWu Fengguang struct bdi_writeback *wb, 786d46db3d5SWu Fengguang struct wb_writeback_work *work) 78703ba3782SJens Axboe { 788d46db3d5SWu Fengguang struct writeback_control wbc = { 789d46db3d5SWu Fengguang .sync_mode = work->sync_mode, 790d46db3d5SWu Fengguang .tagged_writepages = work->tagged_writepages, 791d46db3d5SWu Fengguang .for_kupdate = work->for_kupdate, 792d46db3d5SWu Fengguang .for_background = work->for_background, 7937747bd4bSDave Chinner .for_sync = work->for_sync, 794d46db3d5SWu Fengguang .range_cyclic = work->range_cyclic, 795d46db3d5SWu Fengguang .range_start = 0, 796d46db3d5SWu Fengguang .range_end = LLONG_MAX, 797d46db3d5SWu Fengguang }; 798d46db3d5SWu Fengguang unsigned long start_time = jiffies; 799d46db3d5SWu Fengguang long write_chunk; 800d46db3d5SWu Fengguang long wrote = 0; /* count both pages and inodes */ 801d46db3d5SWu Fengguang 80203ba3782SJens Axboe while (!list_empty(&wb->b_io)) { 8037ccf19a8SNick Piggin struct inode *inode = wb_inode(wb->b_io.prev); 804edadfb10SChristoph Hellwig 805edadfb10SChristoph Hellwig if (inode->i_sb != sb) { 806d46db3d5SWu Fengguang if (work->sb) { 807edadfb10SChristoph Hellwig /* 808edadfb10SChristoph Hellwig * We only want to write back data for this 809edadfb10SChristoph Hellwig * superblock, move all inodes not belonging 810edadfb10SChristoph Hellwig * to it back onto the dirty list. 811edadfb10SChristoph Hellwig */ 812f758eeabSChristoph Hellwig redirty_tail(inode, wb); 81366f3b8e2SJens Axboe continue; 81466f3b8e2SJens Axboe } 815edadfb10SChristoph Hellwig 816edadfb10SChristoph Hellwig /* 817edadfb10SChristoph Hellwig * The inode belongs to a different superblock. 818edadfb10SChristoph Hellwig * Bounce back to the caller to unpin this and 819edadfb10SChristoph Hellwig * pin the next superblock. 820edadfb10SChristoph Hellwig */ 821d46db3d5SWu Fengguang break; 822edadfb10SChristoph Hellwig } 823edadfb10SChristoph Hellwig 8249843b76aSChristoph Hellwig /* 825331cbdeeSWanpeng Li * Don't bother with new inodes or inodes being freed, first 826331cbdeeSWanpeng Li * kind does not need periodic writeout yet, and for the latter 8279843b76aSChristoph Hellwig * kind writeout is handled by the freer. 8289843b76aSChristoph Hellwig */ 829250df6edSDave Chinner spin_lock(&inode->i_lock); 8309843b76aSChristoph Hellwig if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { 831250df6edSDave Chinner spin_unlock(&inode->i_lock); 832fcc5c222SWu Fengguang redirty_tail(inode, wb); 8337ef0d737SNick Piggin continue; 8347ef0d737SNick Piggin } 835cc1676d9SJan Kara if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) { 836cc1676d9SJan Kara /* 837cc1676d9SJan Kara * If this inode is locked for writeback and we are not 838cc1676d9SJan Kara * doing writeback-for-data-integrity, move it to 839cc1676d9SJan Kara * b_more_io so that writeback can proceed with the 840cc1676d9SJan Kara * other inodes on s_io. 841cc1676d9SJan Kara * 842cc1676d9SJan Kara * We'll have another go at writing back this inode 843cc1676d9SJan Kara * when we completed a full scan of b_io. 844cc1676d9SJan Kara */ 845cc1676d9SJan Kara spin_unlock(&inode->i_lock); 846cc1676d9SJan Kara requeue_io(inode, wb); 847cc1676d9SJan Kara trace_writeback_sb_inodes_requeue(inode); 848cc1676d9SJan Kara continue; 849cc1676d9SJan Kara } 850f0d07b7fSJan Kara spin_unlock(&wb->list_lock); 851f0d07b7fSJan Kara 8524f8ad655SJan Kara /* 8534f8ad655SJan Kara * We already requeued the inode if it had I_SYNC set and we 8544f8ad655SJan Kara * are doing WB_SYNC_NONE writeback. So this catches only the 8554f8ad655SJan Kara * WB_SYNC_ALL case. 8564f8ad655SJan Kara */ 857169ebd90SJan Kara if (inode->i_state & I_SYNC) { 858169ebd90SJan Kara /* Wait for I_SYNC. This function drops i_lock... */ 859169ebd90SJan Kara inode_sleep_on_writeback(inode); 860169ebd90SJan Kara /* Inode may be gone, start again */ 861ead188f9SJan Kara spin_lock(&wb->list_lock); 862169ebd90SJan Kara continue; 863169ebd90SJan Kara } 8644f8ad655SJan Kara inode->i_state |= I_SYNC; 8654f8ad655SJan Kara spin_unlock(&inode->i_lock); 866169ebd90SJan Kara 867a88a341aSTejun Heo write_chunk = writeback_chunk_size(wb, work); 868d46db3d5SWu Fengguang wbc.nr_to_write = write_chunk; 869d46db3d5SWu Fengguang wbc.pages_skipped = 0; 870250df6edSDave Chinner 871169ebd90SJan Kara /* 872169ebd90SJan Kara * We use I_SYNC to pin the inode in memory. While it is set 873169ebd90SJan Kara * evict_inode() will wait so the inode cannot be freed. 874169ebd90SJan Kara */ 875cd8ed2a4SYan Hong __writeback_single_inode(inode, &wbc); 876d46db3d5SWu Fengguang 877d46db3d5SWu Fengguang work->nr_pages -= write_chunk - wbc.nr_to_write; 878d46db3d5SWu Fengguang wrote += write_chunk - wbc.nr_to_write; 8794f8ad655SJan Kara spin_lock(&wb->list_lock); 8804f8ad655SJan Kara spin_lock(&inode->i_lock); 8810ae45f63STheodore Ts'o if (!(inode->i_state & I_DIRTY_ALL)) 882d46db3d5SWu Fengguang wrote++; 8834f8ad655SJan Kara requeue_inode(inode, wb, &wbc); 8844f8ad655SJan Kara inode_sync_complete(inode); 8850f1b1fd8SDave Chinner spin_unlock(&inode->i_lock); 886169ebd90SJan Kara cond_resched_lock(&wb->list_lock); 887d46db3d5SWu Fengguang /* 888d46db3d5SWu Fengguang * bail out to wb_writeback() often enough to check 889d46db3d5SWu Fengguang * background threshold and other termination conditions. 890d46db3d5SWu Fengguang */ 891d46db3d5SWu Fengguang if (wrote) { 892d46db3d5SWu Fengguang if (time_is_before_jiffies(start_time + HZ / 10UL)) 893d46db3d5SWu Fengguang break; 894d46db3d5SWu Fengguang if (work->nr_pages <= 0) 895d46db3d5SWu Fengguang break; 8961da177e4SLinus Torvalds } 8978bc3be27SFengguang Wu } 898d46db3d5SWu Fengguang return wrote; 899f11c9c5cSEdward Shishkin } 90038f21977SNick Piggin 901d46db3d5SWu Fengguang static long __writeback_inodes_wb(struct bdi_writeback *wb, 902d46db3d5SWu Fengguang struct wb_writeback_work *work) 903f11c9c5cSEdward Shishkin { 904d46db3d5SWu Fengguang unsigned long start_time = jiffies; 905d46db3d5SWu Fengguang long wrote = 0; 906f11c9c5cSEdward Shishkin 907f11c9c5cSEdward Shishkin while (!list_empty(&wb->b_io)) { 9087ccf19a8SNick Piggin struct inode *inode = wb_inode(wb->b_io.prev); 909f11c9c5cSEdward Shishkin struct super_block *sb = inode->i_sb; 910f11c9c5cSEdward Shishkin 911eb6ef3dfSKonstantin Khlebnikov if (!trylock_super(sb)) { 9120e995816SWu Fengguang /* 913eb6ef3dfSKonstantin Khlebnikov * trylock_super() may fail consistently due to 9140e995816SWu Fengguang * s_umount being grabbed by someone else. Don't use 9150e995816SWu Fengguang * requeue_io() to avoid busy retrying the inode/sb. 9160e995816SWu Fengguang */ 9170e995816SWu Fengguang redirty_tail(inode, wb); 918d19de7edSChristoph Hellwig continue; 919334132aeSChristoph Hellwig } 920d46db3d5SWu Fengguang wrote += writeback_sb_inodes(sb, wb, work); 921eb6ef3dfSKonstantin Khlebnikov up_read(&sb->s_umount); 922f11c9c5cSEdward Shishkin 923d46db3d5SWu Fengguang /* refer to the same tests at the end of writeback_sb_inodes */ 924d46db3d5SWu Fengguang if (wrote) { 925d46db3d5SWu Fengguang if (time_is_before_jiffies(start_time + HZ / 10UL)) 926d46db3d5SWu Fengguang break; 927d46db3d5SWu Fengguang if (work->nr_pages <= 0) 928f11c9c5cSEdward Shishkin break; 929f11c9c5cSEdward Shishkin } 930d46db3d5SWu Fengguang } 93166f3b8e2SJens Axboe /* Leave any unwritten inodes on b_io */ 932d46db3d5SWu Fengguang return wrote; 93366f3b8e2SJens Axboe } 93466f3b8e2SJens Axboe 9357d9f073bSWanpeng Li static long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages, 9360e175a18SCurt Wohlgemuth enum wb_reason reason) 937edadfb10SChristoph Hellwig { 938d46db3d5SWu Fengguang struct wb_writeback_work work = { 939d46db3d5SWu Fengguang .nr_pages = nr_pages, 940d46db3d5SWu Fengguang .sync_mode = WB_SYNC_NONE, 941d46db3d5SWu Fengguang .range_cyclic = 1, 9420e175a18SCurt Wohlgemuth .reason = reason, 943d46db3d5SWu Fengguang }; 944edadfb10SChristoph Hellwig 945f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 946424b351fSWu Fengguang if (list_empty(&wb->b_io)) 947ad4e38ddSCurt Wohlgemuth queue_io(wb, &work); 948d46db3d5SWu Fengguang __writeback_inodes_wb(wb, &work); 949f758eeabSChristoph Hellwig spin_unlock(&wb->list_lock); 950edadfb10SChristoph Hellwig 951d46db3d5SWu Fengguang return nr_pages - work.nr_pages; 95266f3b8e2SJens Axboe } 95366f3b8e2SJens Axboe 954a88a341aSTejun Heo static bool over_bground_thresh(struct bdi_writeback *wb) 95503ba3782SJens Axboe { 95603ba3782SJens Axboe unsigned long background_thresh, dirty_thresh; 95703ba3782SJens Axboe 95816c4042fSWu Fengguang global_dirty_limits(&background_thresh, &dirty_thresh); 95903ba3782SJens Axboe 960b00949aaSWu Fengguang if (global_page_state(NR_FILE_DIRTY) + 961b00949aaSWu Fengguang global_page_state(NR_UNSTABLE_NFS) > background_thresh) 962b00949aaSWu Fengguang return true; 963b00949aaSWu Fengguang 964a88a341aSTejun Heo if (wb_stat(wb, WB_RECLAIMABLE) > wb_dirty_limit(wb, background_thresh)) 965b00949aaSWu Fengguang return true; 966b00949aaSWu Fengguang 967b00949aaSWu Fengguang return false; 96803ba3782SJens Axboe } 96903ba3782SJens Axboe 97003ba3782SJens Axboe /* 971e98be2d5SWu Fengguang * Called under wb->list_lock. If there are multiple wb per bdi, 972e98be2d5SWu Fengguang * only the flusher working on the first wb should do it. 973e98be2d5SWu Fengguang */ 974e98be2d5SWu Fengguang static void wb_update_bandwidth(struct bdi_writeback *wb, 975e98be2d5SWu Fengguang unsigned long start_time) 976e98be2d5SWu Fengguang { 977a88a341aSTejun Heo __wb_update_bandwidth(wb, 0, 0, 0, 0, 0, start_time); 978e98be2d5SWu Fengguang } 979e98be2d5SWu Fengguang 980e98be2d5SWu Fengguang /* 98103ba3782SJens Axboe * Explicit flushing or periodic writeback of "old" data. 98203ba3782SJens Axboe * 98303ba3782SJens Axboe * Define "old": the first time one of an inode's pages is dirtied, we mark the 98403ba3782SJens Axboe * dirtying-time in the inode's address_space. So this periodic writeback code 98503ba3782SJens Axboe * just walks the superblock inode list, writing back any inodes which are 98603ba3782SJens Axboe * older than a specific point in time. 98703ba3782SJens Axboe * 98803ba3782SJens Axboe * Try to run once per dirty_writeback_interval. But if a writeback event 98903ba3782SJens Axboe * takes longer than a dirty_writeback_interval interval, then leave a 99003ba3782SJens Axboe * one-second gap. 99103ba3782SJens Axboe * 99203ba3782SJens Axboe * older_than_this takes precedence over nr_to_write. So we'll only write back 99303ba3782SJens Axboe * all dirty pages if they are all attached to "old" mappings. 99403ba3782SJens Axboe */ 995c4a77a6cSJens Axboe static long wb_writeback(struct bdi_writeback *wb, 99683ba7b07SChristoph Hellwig struct wb_writeback_work *work) 99703ba3782SJens Axboe { 998e98be2d5SWu Fengguang unsigned long wb_start = jiffies; 999d46db3d5SWu Fengguang long nr_pages = work->nr_pages; 10000dc83bd3SJan Kara unsigned long oldest_jif; 1001a5989bdcSJan Kara struct inode *inode; 1002d46db3d5SWu Fengguang long progress; 100303ba3782SJens Axboe 10040dc83bd3SJan Kara oldest_jif = jiffies; 10050dc83bd3SJan Kara work->older_than_this = &oldest_jif; 100603ba3782SJens Axboe 1007e8dfc305SWu Fengguang spin_lock(&wb->list_lock); 100803ba3782SJens Axboe for (;;) { 100903ba3782SJens Axboe /* 1010d3ddec76SWu Fengguang * Stop writeback when nr_pages has been consumed 101103ba3782SJens Axboe */ 101283ba7b07SChristoph Hellwig if (work->nr_pages <= 0) 101303ba3782SJens Axboe break; 101403ba3782SJens Axboe 101503ba3782SJens Axboe /* 1016aa373cf5SJan Kara * Background writeout and kupdate-style writeback may 1017aa373cf5SJan Kara * run forever. Stop them if there is other work to do 1018aa373cf5SJan Kara * so that e.g. sync can proceed. They'll be restarted 1019aa373cf5SJan Kara * after the other works are all done. 1020aa373cf5SJan Kara */ 1021aa373cf5SJan Kara if ((work->for_background || work->for_kupdate) && 1022f0054bb1STejun Heo !list_empty(&wb->work_list)) 1023aa373cf5SJan Kara break; 1024aa373cf5SJan Kara 1025aa373cf5SJan Kara /* 1026d3ddec76SWu Fengguang * For background writeout, stop when we are below the 1027d3ddec76SWu Fengguang * background dirty threshold 102803ba3782SJens Axboe */ 1029a88a341aSTejun Heo if (work->for_background && !over_bground_thresh(wb)) 103003ba3782SJens Axboe break; 103103ba3782SJens Axboe 10321bc36b64SJan Kara /* 10331bc36b64SJan Kara * Kupdate and background works are special and we want to 10341bc36b64SJan Kara * include all inodes that need writing. Livelock avoidance is 10351bc36b64SJan Kara * handled by these works yielding to any other work so we are 10361bc36b64SJan Kara * safe. 10371bc36b64SJan Kara */ 1038ba9aa839SWu Fengguang if (work->for_kupdate) { 10390dc83bd3SJan Kara oldest_jif = jiffies - 1040ba9aa839SWu Fengguang msecs_to_jiffies(dirty_expire_interval * 10); 10411bc36b64SJan Kara } else if (work->for_background) 10420dc83bd3SJan Kara oldest_jif = jiffies; 1043028c2dd1SDave Chinner 1044d46db3d5SWu Fengguang trace_writeback_start(wb->bdi, work); 1045e8dfc305SWu Fengguang if (list_empty(&wb->b_io)) 1046ad4e38ddSCurt Wohlgemuth queue_io(wb, work); 104783ba7b07SChristoph Hellwig if (work->sb) 1048d46db3d5SWu Fengguang progress = writeback_sb_inodes(work->sb, wb, work); 1049edadfb10SChristoph Hellwig else 1050d46db3d5SWu Fengguang progress = __writeback_inodes_wb(wb, work); 1051d46db3d5SWu Fengguang trace_writeback_written(wb->bdi, work); 1052028c2dd1SDave Chinner 1053e98be2d5SWu Fengguang wb_update_bandwidth(wb, wb_start); 105403ba3782SJens Axboe 105503ba3782SJens Axboe /* 105671fd05a8SJens Axboe * Did we write something? Try for more 1057e6fb6da2SWu Fengguang * 1058e6fb6da2SWu Fengguang * Dirty inodes are moved to b_io for writeback in batches. 1059e6fb6da2SWu Fengguang * The completion of the current batch does not necessarily 1060e6fb6da2SWu Fengguang * mean the overall work is done. So we keep looping as long 1061e6fb6da2SWu Fengguang * as made some progress on cleaning pages or inodes. 106271fd05a8SJens Axboe */ 1063d46db3d5SWu Fengguang if (progress) 106403ba3782SJens Axboe continue; 1065a5989bdcSJan Kara /* 1066e6fb6da2SWu Fengguang * No more inodes for IO, bail 1067a5989bdcSJan Kara */ 1068b7a2441fSWu Fengguang if (list_empty(&wb->b_more_io)) 106903ba3782SJens Axboe break; 107003ba3782SJens Axboe /* 10718010c3b6SJens Axboe * Nothing written. Wait for some inode to 10728010c3b6SJens Axboe * become available for writeback. Otherwise 10738010c3b6SJens Axboe * we'll just busyloop. 107403ba3782SJens Axboe */ 107503ba3782SJens Axboe if (!list_empty(&wb->b_more_io)) { 1076d46db3d5SWu Fengguang trace_writeback_wait(wb->bdi, work); 107703ba3782SJens Axboe inode = wb_inode(wb->b_more_io.prev); 1078250df6edSDave Chinner spin_lock(&inode->i_lock); 1079f0d07b7fSJan Kara spin_unlock(&wb->list_lock); 1080169ebd90SJan Kara /* This function drops i_lock... */ 1081169ebd90SJan Kara inode_sleep_on_writeback(inode); 1082f0d07b7fSJan Kara spin_lock(&wb->list_lock); 108303ba3782SJens Axboe } 108403ba3782SJens Axboe } 1085e8dfc305SWu Fengguang spin_unlock(&wb->list_lock); 108603ba3782SJens Axboe 1087d46db3d5SWu Fengguang return nr_pages - work->nr_pages; 108803ba3782SJens Axboe } 108903ba3782SJens Axboe 109003ba3782SJens Axboe /* 109183ba7b07SChristoph Hellwig * Return the next wb_writeback_work struct that hasn't been processed yet. 109203ba3782SJens Axboe */ 1093f0054bb1STejun Heo static struct wb_writeback_work *get_next_work_item(struct bdi_writeback *wb) 109403ba3782SJens Axboe { 109583ba7b07SChristoph Hellwig struct wb_writeback_work *work = NULL; 109603ba3782SJens Axboe 1097f0054bb1STejun Heo spin_lock_bh(&wb->work_lock); 1098f0054bb1STejun Heo if (!list_empty(&wb->work_list)) { 1099f0054bb1STejun Heo work = list_entry(wb->work_list.next, 110083ba7b07SChristoph Hellwig struct wb_writeback_work, list); 110183ba7b07SChristoph Hellwig list_del_init(&work->list); 110203ba3782SJens Axboe } 1103f0054bb1STejun Heo spin_unlock_bh(&wb->work_lock); 110483ba7b07SChristoph Hellwig return work; 110503ba3782SJens Axboe } 110603ba3782SJens Axboe 1107cdf01dd5SLinus Torvalds /* 1108cdf01dd5SLinus Torvalds * Add in the number of potentially dirty inodes, because each inode 1109cdf01dd5SLinus Torvalds * write can dirty pagecache in the underlying blockdev. 1110cdf01dd5SLinus Torvalds */ 1111cdf01dd5SLinus Torvalds static unsigned long get_nr_dirty_pages(void) 1112cdf01dd5SLinus Torvalds { 1113cdf01dd5SLinus Torvalds return global_page_state(NR_FILE_DIRTY) + 1114cdf01dd5SLinus Torvalds global_page_state(NR_UNSTABLE_NFS) + 1115cdf01dd5SLinus Torvalds get_nr_dirty_inodes(); 1116cdf01dd5SLinus Torvalds } 1117cdf01dd5SLinus Torvalds 11186585027aSJan Kara static long wb_check_background_flush(struct bdi_writeback *wb) 11196585027aSJan Kara { 1120a88a341aSTejun Heo if (over_bground_thresh(wb)) { 11216585027aSJan Kara 11226585027aSJan Kara struct wb_writeback_work work = { 11236585027aSJan Kara .nr_pages = LONG_MAX, 11246585027aSJan Kara .sync_mode = WB_SYNC_NONE, 11256585027aSJan Kara .for_background = 1, 11266585027aSJan Kara .range_cyclic = 1, 11270e175a18SCurt Wohlgemuth .reason = WB_REASON_BACKGROUND, 11286585027aSJan Kara }; 11296585027aSJan Kara 11306585027aSJan Kara return wb_writeback(wb, &work); 11316585027aSJan Kara } 11326585027aSJan Kara 11336585027aSJan Kara return 0; 11346585027aSJan Kara } 11356585027aSJan Kara 113603ba3782SJens Axboe static long wb_check_old_data_flush(struct bdi_writeback *wb) 113703ba3782SJens Axboe { 113803ba3782SJens Axboe unsigned long expired; 113903ba3782SJens Axboe long nr_pages; 114003ba3782SJens Axboe 114169b62d01SJens Axboe /* 114269b62d01SJens Axboe * When set to zero, disable periodic writeback 114369b62d01SJens Axboe */ 114469b62d01SJens Axboe if (!dirty_writeback_interval) 114569b62d01SJens Axboe return 0; 114669b62d01SJens Axboe 114703ba3782SJens Axboe expired = wb->last_old_flush + 114803ba3782SJens Axboe msecs_to_jiffies(dirty_writeback_interval * 10); 114903ba3782SJens Axboe if (time_before(jiffies, expired)) 115003ba3782SJens Axboe return 0; 115103ba3782SJens Axboe 115203ba3782SJens Axboe wb->last_old_flush = jiffies; 1153cdf01dd5SLinus Torvalds nr_pages = get_nr_dirty_pages(); 115403ba3782SJens Axboe 1155c4a77a6cSJens Axboe if (nr_pages) { 115683ba7b07SChristoph Hellwig struct wb_writeback_work work = { 1157c4a77a6cSJens Axboe .nr_pages = nr_pages, 1158c4a77a6cSJens Axboe .sync_mode = WB_SYNC_NONE, 1159c4a77a6cSJens Axboe .for_kupdate = 1, 1160c4a77a6cSJens Axboe .range_cyclic = 1, 11610e175a18SCurt Wohlgemuth .reason = WB_REASON_PERIODIC, 1162c4a77a6cSJens Axboe }; 1163c4a77a6cSJens Axboe 116483ba7b07SChristoph Hellwig return wb_writeback(wb, &work); 1165c4a77a6cSJens Axboe } 116603ba3782SJens Axboe 116703ba3782SJens Axboe return 0; 116803ba3782SJens Axboe } 116903ba3782SJens Axboe 117003ba3782SJens Axboe /* 117103ba3782SJens Axboe * Retrieve work items and do the writeback they describe 117203ba3782SJens Axboe */ 117325d130baSWanpeng Li static long wb_do_writeback(struct bdi_writeback *wb) 117403ba3782SJens Axboe { 117583ba7b07SChristoph Hellwig struct wb_writeback_work *work; 1176c4a77a6cSJens Axboe long wrote = 0; 117703ba3782SJens Axboe 11784452226eSTejun Heo set_bit(WB_writeback_running, &wb->state); 1179f0054bb1STejun Heo while ((work = get_next_work_item(wb)) != NULL) { 1180cc395d7fSTejun Heo struct wb_completion *done = work->done; 118183ba7b07SChristoph Hellwig 1182f0054bb1STejun Heo trace_writeback_exec(wb->bdi, work); 1183455b2864SDave Chinner 118483ba7b07SChristoph Hellwig wrote += wb_writeback(wb, work); 118503ba3782SJens Axboe 1186ac7b19a3STejun Heo if (work->auto_free) 118783ba7b07SChristoph Hellwig kfree(work); 1188cc395d7fSTejun Heo if (done && atomic_dec_and_test(&done->cnt)) 1189cc395d7fSTejun Heo wake_up_all(&wb->bdi->wb_waitq); 119003ba3782SJens Axboe } 119103ba3782SJens Axboe 119203ba3782SJens Axboe /* 119303ba3782SJens Axboe * Check for periodic writeback, kupdated() style 119403ba3782SJens Axboe */ 119503ba3782SJens Axboe wrote += wb_check_old_data_flush(wb); 11966585027aSJan Kara wrote += wb_check_background_flush(wb); 11974452226eSTejun Heo clear_bit(WB_writeback_running, &wb->state); 119803ba3782SJens Axboe 119903ba3782SJens Axboe return wrote; 120003ba3782SJens Axboe } 120103ba3782SJens Axboe 120203ba3782SJens Axboe /* 120303ba3782SJens Axboe * Handle writeback of dirty data for the device backed by this bdi. Also 1204839a8e86STejun Heo * reschedules periodically and does kupdated style flushing. 120503ba3782SJens Axboe */ 1206f0054bb1STejun Heo void wb_workfn(struct work_struct *work) 120703ba3782SJens Axboe { 1208839a8e86STejun Heo struct bdi_writeback *wb = container_of(to_delayed_work(work), 1209839a8e86STejun Heo struct bdi_writeback, dwork); 121003ba3782SJens Axboe long pages_written; 121103ba3782SJens Axboe 1212f0054bb1STejun Heo set_worker_desc("flush-%s", dev_name(wb->bdi->dev)); 1213766f9164SPeter Zijlstra current->flags |= PF_SWAPWRITE; 121403ba3782SJens Axboe 1215839a8e86STejun Heo if (likely(!current_is_workqueue_rescuer() || 12164452226eSTejun Heo !test_bit(WB_registered, &wb->state))) { 121703ba3782SJens Axboe /* 1218f0054bb1STejun Heo * The normal path. Keep writing back @wb until its 1219839a8e86STejun Heo * work_list is empty. Note that this path is also taken 1220f0054bb1STejun Heo * if @wb is shutting down even when we're running off the 1221839a8e86STejun Heo * rescuer as work_list needs to be drained. 122203ba3782SJens Axboe */ 1223839a8e86STejun Heo do { 122425d130baSWanpeng Li pages_written = wb_do_writeback(wb); 1225455b2864SDave Chinner trace_writeback_pages_written(pages_written); 1226f0054bb1STejun Heo } while (!list_empty(&wb->work_list)); 1227839a8e86STejun Heo } else { 1228253c34e9SArtem Bityutskiy /* 1229839a8e86STejun Heo * bdi_wq can't get enough workers and we're running off 1230839a8e86STejun Heo * the emergency worker. Don't hog it. Hopefully, 1024 is 1231839a8e86STejun Heo * enough for efficient IO. 1232253c34e9SArtem Bityutskiy */ 1233f0054bb1STejun Heo pages_written = writeback_inodes_wb(wb, 1024, 1234839a8e86STejun Heo WB_REASON_FORKER_THREAD); 1235839a8e86STejun Heo trace_writeback_pages_written(pages_written); 123603ba3782SJens Axboe } 123703ba3782SJens Axboe 1238f0054bb1STejun Heo if (!list_empty(&wb->work_list)) 12396ca738d6SDerek Basehore mod_delayed_work(bdi_wq, &wb->dwork, 0); 12406ca738d6SDerek Basehore else if (wb_has_dirty_io(wb) && dirty_writeback_interval) 1241f0054bb1STejun Heo wb_wakeup_delayed(wb); 1242455b2864SDave Chinner 1243839a8e86STejun Heo current->flags &= ~PF_SWAPWRITE; 124403ba3782SJens Axboe } 124503ba3782SJens Axboe 124603ba3782SJens Axboe /* 124703ba3782SJens Axboe * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back 124803ba3782SJens Axboe * the whole world. 124903ba3782SJens Axboe */ 12500e175a18SCurt Wohlgemuth void wakeup_flusher_threads(long nr_pages, enum wb_reason reason) 125103ba3782SJens Axboe { 1252b8c2f347SChristoph Hellwig struct backing_dev_info *bdi; 1253b8c2f347SChristoph Hellwig 125447df3ddeSJan Kara if (!nr_pages) 125547df3ddeSJan Kara nr_pages = get_nr_dirty_pages(); 1256b8c2f347SChristoph Hellwig 1257b8c2f347SChristoph Hellwig rcu_read_lock(); 1258f2b65121STejun Heo list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { 1259f2b65121STejun Heo struct bdi_writeback *wb; 1260f2b65121STejun Heo struct wb_iter iter; 1261f2b65121STejun Heo 1262f2b65121STejun Heo if (!bdi_has_dirty_io(bdi)) 1263f2b65121STejun Heo continue; 1264f2b65121STejun Heo 1265f2b65121STejun Heo bdi_for_each_wb(wb, bdi, &iter, 0) 1266f2b65121STejun Heo wb_start_writeback(wb, wb_split_bdi_pages(wb, nr_pages), 1267f2b65121STejun Heo false, reason); 1268f2b65121STejun Heo } 1269b8c2f347SChristoph Hellwig rcu_read_unlock(); 127003ba3782SJens Axboe } 127103ba3782SJens Axboe 1272a2f48706STheodore Ts'o /* 1273a2f48706STheodore Ts'o * Wake up bdi's periodically to make sure dirtytime inodes gets 1274a2f48706STheodore Ts'o * written back periodically. We deliberately do *not* check the 1275a2f48706STheodore Ts'o * b_dirtytime list in wb_has_dirty_io(), since this would cause the 1276a2f48706STheodore Ts'o * kernel to be constantly waking up once there are any dirtytime 1277a2f48706STheodore Ts'o * inodes on the system. So instead we define a separate delayed work 1278a2f48706STheodore Ts'o * function which gets called much more rarely. (By default, only 1279a2f48706STheodore Ts'o * once every 12 hours.) 1280a2f48706STheodore Ts'o * 1281a2f48706STheodore Ts'o * If there is any other write activity going on in the file system, 1282a2f48706STheodore Ts'o * this function won't be necessary. But if the only thing that has 1283a2f48706STheodore Ts'o * happened on the file system is a dirtytime inode caused by an atime 1284a2f48706STheodore Ts'o * update, we need this infrastructure below to make sure that inode 1285a2f48706STheodore Ts'o * eventually gets pushed out to disk. 1286a2f48706STheodore Ts'o */ 1287a2f48706STheodore Ts'o static void wakeup_dirtytime_writeback(struct work_struct *w); 1288a2f48706STheodore Ts'o static DECLARE_DELAYED_WORK(dirtytime_work, wakeup_dirtytime_writeback); 1289a2f48706STheodore Ts'o 1290a2f48706STheodore Ts'o static void wakeup_dirtytime_writeback(struct work_struct *w) 1291a2f48706STheodore Ts'o { 1292a2f48706STheodore Ts'o struct backing_dev_info *bdi; 1293a2f48706STheodore Ts'o 1294a2f48706STheodore Ts'o rcu_read_lock(); 1295a2f48706STheodore Ts'o list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { 1296001fe6f6STejun Heo struct bdi_writeback *wb; 1297001fe6f6STejun Heo struct wb_iter iter; 1298001fe6f6STejun Heo 1299001fe6f6STejun Heo bdi_for_each_wb(wb, bdi, &iter, 0) 1300001fe6f6STejun Heo if (!list_empty(&bdi->wb.b_dirty_time)) 1301f0054bb1STejun Heo wb_wakeup(&bdi->wb); 1302a2f48706STheodore Ts'o } 1303a2f48706STheodore Ts'o rcu_read_unlock(); 1304a2f48706STheodore Ts'o schedule_delayed_work(&dirtytime_work, dirtytime_expire_interval * HZ); 1305a2f48706STheodore Ts'o } 1306a2f48706STheodore Ts'o 1307a2f48706STheodore Ts'o static int __init start_dirtytime_writeback(void) 1308a2f48706STheodore Ts'o { 1309a2f48706STheodore Ts'o schedule_delayed_work(&dirtytime_work, dirtytime_expire_interval * HZ); 1310a2f48706STheodore Ts'o return 0; 1311a2f48706STheodore Ts'o } 1312a2f48706STheodore Ts'o __initcall(start_dirtytime_writeback); 1313a2f48706STheodore Ts'o 13141efff914STheodore Ts'o int dirtytime_interval_handler(struct ctl_table *table, int write, 13151efff914STheodore Ts'o void __user *buffer, size_t *lenp, loff_t *ppos) 13161efff914STheodore Ts'o { 13171efff914STheodore Ts'o int ret; 13181efff914STheodore Ts'o 13191efff914STheodore Ts'o ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos); 13201efff914STheodore Ts'o if (ret == 0 && write) 13211efff914STheodore Ts'o mod_delayed_work(system_wq, &dirtytime_work, 0); 13221efff914STheodore Ts'o return ret; 13231efff914STheodore Ts'o } 13241efff914STheodore Ts'o 132503ba3782SJens Axboe static noinline void block_dump___mark_inode_dirty(struct inode *inode) 132603ba3782SJens Axboe { 132703ba3782SJens Axboe if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { 132803ba3782SJens Axboe struct dentry *dentry; 132903ba3782SJens Axboe const char *name = "?"; 133003ba3782SJens Axboe 133103ba3782SJens Axboe dentry = d_find_alias(inode); 133203ba3782SJens Axboe if (dentry) { 133303ba3782SJens Axboe spin_lock(&dentry->d_lock); 133403ba3782SJens Axboe name = (const char *) dentry->d_name.name; 133503ba3782SJens Axboe } 133603ba3782SJens Axboe printk(KERN_DEBUG 133703ba3782SJens Axboe "%s(%d): dirtied inode %lu (%s) on %s\n", 133803ba3782SJens Axboe current->comm, task_pid_nr(current), inode->i_ino, 133903ba3782SJens Axboe name, inode->i_sb->s_id); 134003ba3782SJens Axboe if (dentry) { 134103ba3782SJens Axboe spin_unlock(&dentry->d_lock); 134203ba3782SJens Axboe dput(dentry); 134303ba3782SJens Axboe } 134403ba3782SJens Axboe } 134503ba3782SJens Axboe } 134603ba3782SJens Axboe 134703ba3782SJens Axboe /** 134803ba3782SJens Axboe * __mark_inode_dirty - internal function 134903ba3782SJens Axboe * @inode: inode to mark 135003ba3782SJens Axboe * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) 135103ba3782SJens Axboe * Mark an inode as dirty. Callers should use mark_inode_dirty or 135203ba3782SJens Axboe * mark_inode_dirty_sync. 135303ba3782SJens Axboe * 135403ba3782SJens Axboe * Put the inode on the super block's dirty list. 135503ba3782SJens Axboe * 135603ba3782SJens Axboe * CAREFUL! We mark it dirty unconditionally, but move it onto the 135703ba3782SJens Axboe * dirty list only if it is hashed or if it refers to a blockdev. 135803ba3782SJens Axboe * If it was not hashed, it will never be added to the dirty list 135903ba3782SJens Axboe * even if it is later hashed, as it will have been marked dirty already. 136003ba3782SJens Axboe * 136103ba3782SJens Axboe * In short, make sure you hash any inodes _before_ you start marking 136203ba3782SJens Axboe * them dirty. 136303ba3782SJens Axboe * 136403ba3782SJens Axboe * Note that for blockdevs, inode->dirtied_when represents the dirtying time of 136503ba3782SJens Axboe * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of 136603ba3782SJens Axboe * the kernel-internal blockdev inode represents the dirtying time of the 136703ba3782SJens Axboe * blockdev's pages. This is why for I_DIRTY_PAGES we always use 136803ba3782SJens Axboe * page->mapping->host, so the page-dirtying time is recorded in the internal 136903ba3782SJens Axboe * blockdev inode. 137003ba3782SJens Axboe */ 13710ae45f63STheodore Ts'o #define I_DIRTY_INODE (I_DIRTY_SYNC | I_DIRTY_DATASYNC) 137203ba3782SJens Axboe void __mark_inode_dirty(struct inode *inode, int flags) 137303ba3782SJens Axboe { 137403ba3782SJens Axboe struct super_block *sb = inode->i_sb; 1375253c34e9SArtem Bityutskiy struct backing_dev_info *bdi = NULL; 13760ae45f63STheodore Ts'o int dirtytime; 13770ae45f63STheodore Ts'o 13780ae45f63STheodore Ts'o trace_writeback_mark_inode_dirty(inode, flags); 137903ba3782SJens Axboe 138003ba3782SJens Axboe /* 138103ba3782SJens Axboe * Don't do this for I_DIRTY_PAGES - that doesn't actually 138203ba3782SJens Axboe * dirty the inode itself 138303ba3782SJens Axboe */ 13840ae45f63STheodore Ts'o if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC | I_DIRTY_TIME)) { 13859fb0a7daSTejun Heo trace_writeback_dirty_inode_start(inode, flags); 13869fb0a7daSTejun Heo 138703ba3782SJens Axboe if (sb->s_op->dirty_inode) 1388aa385729SChristoph Hellwig sb->s_op->dirty_inode(inode, flags); 13899fb0a7daSTejun Heo 13909fb0a7daSTejun Heo trace_writeback_dirty_inode(inode, flags); 139103ba3782SJens Axboe } 13920ae45f63STheodore Ts'o if (flags & I_DIRTY_INODE) 13930ae45f63STheodore Ts'o flags &= ~I_DIRTY_TIME; 13940ae45f63STheodore Ts'o dirtytime = flags & I_DIRTY_TIME; 139503ba3782SJens Axboe 139603ba3782SJens Axboe /* 13979c6ac78eSTejun Heo * Paired with smp_mb() in __writeback_single_inode() for the 13989c6ac78eSTejun Heo * following lockless i_state test. See there for details. 139903ba3782SJens Axboe */ 140003ba3782SJens Axboe smp_mb(); 140103ba3782SJens Axboe 14020ae45f63STheodore Ts'o if (((inode->i_state & flags) == flags) || 14030ae45f63STheodore Ts'o (dirtytime && (inode->i_state & I_DIRTY_INODE))) 140403ba3782SJens Axboe return; 140503ba3782SJens Axboe 140603ba3782SJens Axboe if (unlikely(block_dump)) 140703ba3782SJens Axboe block_dump___mark_inode_dirty(inode); 140803ba3782SJens Axboe 1409250df6edSDave Chinner spin_lock(&inode->i_lock); 14100ae45f63STheodore Ts'o if (dirtytime && (inode->i_state & I_DIRTY_INODE)) 14110ae45f63STheodore Ts'o goto out_unlock_inode; 141203ba3782SJens Axboe if ((inode->i_state & flags) != flags) { 141303ba3782SJens Axboe const int was_dirty = inode->i_state & I_DIRTY; 141403ba3782SJens Axboe 141552ebea74STejun Heo inode_attach_wb(inode, NULL); 141652ebea74STejun Heo 14170ae45f63STheodore Ts'o if (flags & I_DIRTY_INODE) 14180ae45f63STheodore Ts'o inode->i_state &= ~I_DIRTY_TIME; 141903ba3782SJens Axboe inode->i_state |= flags; 142003ba3782SJens Axboe 142103ba3782SJens Axboe /* 142203ba3782SJens Axboe * If the inode is being synced, just update its dirty state. 142303ba3782SJens Axboe * The unlocker will place the inode on the appropriate 142403ba3782SJens Axboe * superblock list, based upon its state. 142503ba3782SJens Axboe */ 142603ba3782SJens Axboe if (inode->i_state & I_SYNC) 1427250df6edSDave Chinner goto out_unlock_inode; 142803ba3782SJens Axboe 142903ba3782SJens Axboe /* 143003ba3782SJens Axboe * Only add valid (hashed) inodes to the superblock's 143103ba3782SJens Axboe * dirty list. Add blockdev inodes as well. 143203ba3782SJens Axboe */ 143303ba3782SJens Axboe if (!S_ISBLK(inode->i_mode)) { 14341d3382cbSAl Viro if (inode_unhashed(inode)) 1435250df6edSDave Chinner goto out_unlock_inode; 143603ba3782SJens Axboe } 1437a4ffdde6SAl Viro if (inode->i_state & I_FREEING) 1438250df6edSDave Chinner goto out_unlock_inode; 143903ba3782SJens Axboe 144003ba3782SJens Axboe /* 144103ba3782SJens Axboe * If the inode was already on b_dirty/b_io/b_more_io, don't 144203ba3782SJens Axboe * reposition it (that would break b_dirty time-ordering). 144303ba3782SJens Axboe */ 144403ba3782SJens Axboe if (!was_dirty) { 1445d6c10f1fSTejun Heo struct list_head *dirty_list; 1446a66979abSDave Chinner bool wakeup_bdi = false; 1447253c34e9SArtem Bityutskiy bdi = inode_to_bdi(inode); 1448500b067cSJens Axboe 1449146d7009SJunxiao Bi spin_unlock(&inode->i_lock); 1450146d7009SJunxiao Bi spin_lock(&bdi->wb.list_lock); 1451253c34e9SArtem Bityutskiy 1452d6c10f1fSTejun Heo WARN(bdi_cap_writeback_dirty(bdi) && 1453d6c10f1fSTejun Heo !test_bit(WB_registered, &bdi->wb.state), 1454d6c10f1fSTejun Heo "bdi-%s not registered\n", bdi->name); 145503ba3782SJens Axboe 145603ba3782SJens Axboe inode->dirtied_when = jiffies; 1457a2f48706STheodore Ts'o if (dirtytime) 1458a2f48706STheodore Ts'o inode->dirtied_time_when = jiffies; 1459d6c10f1fSTejun Heo 1460a2f48706STheodore Ts'o if (inode->i_state & (I_DIRTY_INODE | I_DIRTY_PAGES)) 1461d6c10f1fSTejun Heo dirty_list = &bdi->wb.b_dirty; 1462a2f48706STheodore Ts'o else 1463d6c10f1fSTejun Heo dirty_list = &bdi->wb.b_dirty_time; 1464d6c10f1fSTejun Heo 1465d6c10f1fSTejun Heo wakeup_bdi = inode_wb_list_move_locked(inode, &bdi->wb, 1466d6c10f1fSTejun Heo dirty_list); 1467d6c10f1fSTejun Heo 1468f758eeabSChristoph Hellwig spin_unlock(&bdi->wb.list_lock); 14690ae45f63STheodore Ts'o trace_writeback_dirty_inode_enqueue(inode); 1470253c34e9SArtem Bityutskiy 1471d6c10f1fSTejun Heo /* 1472d6c10f1fSTejun Heo * If this is the first dirty inode for this bdi, 1473d6c10f1fSTejun Heo * we have to wake-up the corresponding bdi thread 1474d6c10f1fSTejun Heo * to make sure background write-back happens 1475d6c10f1fSTejun Heo * later. 1476d6c10f1fSTejun Heo */ 1477d6c10f1fSTejun Heo if (bdi_cap_writeback_dirty(bdi) && wakeup_bdi) 1478f0054bb1STejun Heo wb_wakeup_delayed(&bdi->wb); 1479a66979abSDave Chinner return; 1480a66979abSDave Chinner } 1481a66979abSDave Chinner } 1482a66979abSDave Chinner out_unlock_inode: 1483a66979abSDave Chinner spin_unlock(&inode->i_lock); 1484a66979abSDave Chinner 148503ba3782SJens Axboe } 148603ba3782SJens Axboe EXPORT_SYMBOL(__mark_inode_dirty); 148703ba3782SJens Axboe 1488b6e51316SJens Axboe static void wait_sb_inodes(struct super_block *sb) 148966f3b8e2SJens Axboe { 149038f21977SNick Piggin struct inode *inode, *old_inode = NULL; 149138f21977SNick Piggin 149203ba3782SJens Axboe /* 149303ba3782SJens Axboe * We need to be protected against the filesystem going from 149403ba3782SJens Axboe * r/o to r/w or vice versa. 149503ba3782SJens Axboe */ 1496b6e51316SJens Axboe WARN_ON(!rwsem_is_locked(&sb->s_umount)); 149703ba3782SJens Axboe 149855fa6091SDave Chinner spin_lock(&inode_sb_list_lock); 149966f3b8e2SJens Axboe 150038f21977SNick Piggin /* 150138f21977SNick Piggin * Data integrity sync. Must wait for all pages under writeback, 150238f21977SNick Piggin * because there may have been pages dirtied before our sync 150338f21977SNick Piggin * call, but which had writeout started before we write it out. 150438f21977SNick Piggin * In which case, the inode may not be on the dirty list, but 150538f21977SNick Piggin * we still have to wait for that writeout. 150638f21977SNick Piggin */ 1507b6e51316SJens Axboe list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { 1508250df6edSDave Chinner struct address_space *mapping = inode->i_mapping; 150938f21977SNick Piggin 1510250df6edSDave Chinner spin_lock(&inode->i_lock); 1511250df6edSDave Chinner if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) || 1512250df6edSDave Chinner (mapping->nrpages == 0)) { 1513250df6edSDave Chinner spin_unlock(&inode->i_lock); 151438f21977SNick Piggin continue; 1515250df6edSDave Chinner } 151638f21977SNick Piggin __iget(inode); 1517250df6edSDave Chinner spin_unlock(&inode->i_lock); 151855fa6091SDave Chinner spin_unlock(&inode_sb_list_lock); 151955fa6091SDave Chinner 152038f21977SNick Piggin /* 152155fa6091SDave Chinner * We hold a reference to 'inode' so it couldn't have been 152255fa6091SDave Chinner * removed from s_inodes list while we dropped the 152355fa6091SDave Chinner * inode_sb_list_lock. We cannot iput the inode now as we can 152455fa6091SDave Chinner * be holding the last reference and we cannot iput it under 152555fa6091SDave Chinner * inode_sb_list_lock. So we keep the reference and iput it 152655fa6091SDave Chinner * later. 152738f21977SNick Piggin */ 152838f21977SNick Piggin iput(old_inode); 152938f21977SNick Piggin old_inode = inode; 153038f21977SNick Piggin 153138f21977SNick Piggin filemap_fdatawait(mapping); 153238f21977SNick Piggin 153338f21977SNick Piggin cond_resched(); 153438f21977SNick Piggin 153555fa6091SDave Chinner spin_lock(&inode_sb_list_lock); 153638f21977SNick Piggin } 153755fa6091SDave Chinner spin_unlock(&inode_sb_list_lock); 153838f21977SNick Piggin iput(old_inode); 153966f3b8e2SJens Axboe } 15401da177e4SLinus Torvalds 1541d8a8559cSJens Axboe /** 15423259f8beSChris Mason * writeback_inodes_sb_nr - writeback dirty inodes from given super_block 1543d8a8559cSJens Axboe * @sb: the superblock 15443259f8beSChris Mason * @nr: the number of pages to write 1545786228abSMarcos Paulo de Souza * @reason: reason why some writeback work initiated 15461da177e4SLinus Torvalds * 1547d8a8559cSJens Axboe * Start writeback on some inodes on this super_block. No guarantees are made 1548d8a8559cSJens Axboe * on how many (if any) will be written, and this function does not wait 15493259f8beSChris Mason * for IO completion of submitted IO. 15501da177e4SLinus Torvalds */ 15510e175a18SCurt Wohlgemuth void writeback_inodes_sb_nr(struct super_block *sb, 15520e175a18SCurt Wohlgemuth unsigned long nr, 15530e175a18SCurt Wohlgemuth enum wb_reason reason) 15541da177e4SLinus Torvalds { 1555cc395d7fSTejun Heo DEFINE_WB_COMPLETION_ONSTACK(done); 155683ba7b07SChristoph Hellwig struct wb_writeback_work work = { 15573c4d7165SChristoph Hellwig .sb = sb, 15583c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_NONE, 15596e6938b6SWu Fengguang .tagged_writepages = 1, 156083ba7b07SChristoph Hellwig .done = &done, 15613259f8beSChris Mason .nr_pages = nr, 15620e175a18SCurt Wohlgemuth .reason = reason, 15633c4d7165SChristoph Hellwig }; 1564e7972912STejun Heo struct backing_dev_info *bdi = sb->s_bdi; 15650e3c9a22SJens Axboe 1566e7972912STejun Heo if (!bdi_has_dirty_io(bdi) || bdi == &noop_backing_dev_info) 15676eedc701SJan Kara return; 1568cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 1569e7972912STejun Heo wb_queue_work(&bdi->wb, &work); 1570cc395d7fSTejun Heo wb_wait_for_completion(bdi, &done); 15711da177e4SLinus Torvalds } 15723259f8beSChris Mason EXPORT_SYMBOL(writeback_inodes_sb_nr); 15733259f8beSChris Mason 15743259f8beSChris Mason /** 15753259f8beSChris Mason * writeback_inodes_sb - writeback dirty inodes from given super_block 15763259f8beSChris Mason * @sb: the superblock 1577786228abSMarcos Paulo de Souza * @reason: reason why some writeback work was initiated 15783259f8beSChris Mason * 15793259f8beSChris Mason * Start writeback on some inodes on this super_block. No guarantees are made 15803259f8beSChris Mason * on how many (if any) will be written, and this function does not wait 15813259f8beSChris Mason * for IO completion of submitted IO. 15823259f8beSChris Mason */ 15830e175a18SCurt Wohlgemuth void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) 15843259f8beSChris Mason { 15850e175a18SCurt Wohlgemuth return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); 15863259f8beSChris Mason } 1587d8a8559cSJens Axboe EXPORT_SYMBOL(writeback_inodes_sb); 1588d8a8559cSJens Axboe 1589d8a8559cSJens Axboe /** 159010ee27a0SMiao Xie * try_to_writeback_inodes_sb_nr - try to start writeback if none underway 15913259f8beSChris Mason * @sb: the superblock 15923259f8beSChris Mason * @nr: the number of pages to write 159310ee27a0SMiao Xie * @reason: the reason of writeback 15943259f8beSChris Mason * 159510ee27a0SMiao Xie * Invoke writeback_inodes_sb_nr if no writeback is currently underway. 15963259f8beSChris Mason * Returns 1 if writeback was started, 0 if not. 15973259f8beSChris Mason */ 159810ee27a0SMiao Xie int try_to_writeback_inodes_sb_nr(struct super_block *sb, 15990e175a18SCurt Wohlgemuth unsigned long nr, 16000e175a18SCurt Wohlgemuth enum wb_reason reason) 16013259f8beSChris Mason { 1602bc05873dSTejun Heo if (writeback_in_progress(&sb->s_bdi->wb)) 160310ee27a0SMiao Xie return 1; 160410ee27a0SMiao Xie 160510ee27a0SMiao Xie if (!down_read_trylock(&sb->s_umount)) 160610ee27a0SMiao Xie return 0; 160710ee27a0SMiao Xie 16080e175a18SCurt Wohlgemuth writeback_inodes_sb_nr(sb, nr, reason); 16093259f8beSChris Mason up_read(&sb->s_umount); 16103259f8beSChris Mason return 1; 16113259f8beSChris Mason } 161210ee27a0SMiao Xie EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr); 161310ee27a0SMiao Xie 161410ee27a0SMiao Xie /** 161510ee27a0SMiao Xie * try_to_writeback_inodes_sb - try to start writeback if none underway 161610ee27a0SMiao Xie * @sb: the superblock 161710ee27a0SMiao Xie * @reason: reason why some writeback work was initiated 161810ee27a0SMiao Xie * 161910ee27a0SMiao Xie * Implement by try_to_writeback_inodes_sb_nr() 162010ee27a0SMiao Xie * Returns 1 if writeback was started, 0 if not. 162110ee27a0SMiao Xie */ 162210ee27a0SMiao Xie int try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) 162310ee27a0SMiao Xie { 162410ee27a0SMiao Xie return try_to_writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); 162510ee27a0SMiao Xie } 162610ee27a0SMiao Xie EXPORT_SYMBOL(try_to_writeback_inodes_sb); 16273259f8beSChris Mason 16283259f8beSChris Mason /** 1629d8a8559cSJens Axboe * sync_inodes_sb - sync sb inode pages 1630d8a8559cSJens Axboe * @sb: the superblock 1631d8a8559cSJens Axboe * 1632d8a8559cSJens Axboe * This function writes and waits on any dirty inode belonging to this 16330dc83bd3SJan Kara * super_block. 1634d8a8559cSJens Axboe */ 16350dc83bd3SJan Kara void sync_inodes_sb(struct super_block *sb) 1636d8a8559cSJens Axboe { 1637cc395d7fSTejun Heo DEFINE_WB_COMPLETION_ONSTACK(done); 163883ba7b07SChristoph Hellwig struct wb_writeback_work work = { 16393c4d7165SChristoph Hellwig .sb = sb, 16403c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_ALL, 16413c4d7165SChristoph Hellwig .nr_pages = LONG_MAX, 16423c4d7165SChristoph Hellwig .range_cyclic = 0, 164383ba7b07SChristoph Hellwig .done = &done, 16440e175a18SCurt Wohlgemuth .reason = WB_REASON_SYNC, 16457747bd4bSDave Chinner .for_sync = 1, 16463c4d7165SChristoph Hellwig }; 1647e7972912STejun Heo struct backing_dev_info *bdi = sb->s_bdi; 16483c4d7165SChristoph Hellwig 16496eedc701SJan Kara /* Nothing to do? */ 1650e7972912STejun Heo if (!bdi_has_dirty_io(bdi) || bdi == &noop_backing_dev_info) 16516eedc701SJan Kara return; 1652cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 1653cf37e972SChristoph Hellwig 1654e7972912STejun Heo wb_queue_work(&bdi->wb, &work); 1655cc395d7fSTejun Heo wb_wait_for_completion(bdi, &done); 165683ba7b07SChristoph Hellwig 1657b6e51316SJens Axboe wait_sb_inodes(sb); 1658d8a8559cSJens Axboe } 1659d8a8559cSJens Axboe EXPORT_SYMBOL(sync_inodes_sb); 16601da177e4SLinus Torvalds 16611da177e4SLinus Torvalds /** 16621da177e4SLinus Torvalds * write_inode_now - write an inode to disk 16631da177e4SLinus Torvalds * @inode: inode to write to disk 16641da177e4SLinus Torvalds * @sync: whether the write should be synchronous or not 16651da177e4SLinus Torvalds * 16667f04c26dSAndrea Arcangeli * This function commits an inode to disk immediately if it is dirty. This is 16677f04c26dSAndrea Arcangeli * primarily needed by knfsd. 16687f04c26dSAndrea Arcangeli * 16697f04c26dSAndrea Arcangeli * The caller must either have a ref on the inode or must have set I_WILL_FREE. 16701da177e4SLinus Torvalds */ 16711da177e4SLinus Torvalds int write_inode_now(struct inode *inode, int sync) 16721da177e4SLinus Torvalds { 1673f758eeabSChristoph Hellwig struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; 16741da177e4SLinus Torvalds struct writeback_control wbc = { 16751da177e4SLinus Torvalds .nr_to_write = LONG_MAX, 167618914b18SMike Galbraith .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, 1677111ebb6eSOGAWA Hirofumi .range_start = 0, 1678111ebb6eSOGAWA Hirofumi .range_end = LLONG_MAX, 16791da177e4SLinus Torvalds }; 16801da177e4SLinus Torvalds 16811da177e4SLinus Torvalds if (!mapping_cap_writeback_dirty(inode->i_mapping)) 168249364ce2SAndrew Morton wbc.nr_to_write = 0; 16831da177e4SLinus Torvalds 16841da177e4SLinus Torvalds might_sleep(); 16854f8ad655SJan Kara return writeback_single_inode(inode, wb, &wbc); 16861da177e4SLinus Torvalds } 16871da177e4SLinus Torvalds EXPORT_SYMBOL(write_inode_now); 16881da177e4SLinus Torvalds 16891da177e4SLinus Torvalds /** 16901da177e4SLinus Torvalds * sync_inode - write an inode and its pages to disk. 16911da177e4SLinus Torvalds * @inode: the inode to sync 16921da177e4SLinus Torvalds * @wbc: controls the writeback mode 16931da177e4SLinus Torvalds * 16941da177e4SLinus Torvalds * sync_inode() will write an inode and its pages to disk. It will also 16951da177e4SLinus Torvalds * correctly update the inode on its superblock's dirty inode lists and will 16961da177e4SLinus Torvalds * update inode->i_state. 16971da177e4SLinus Torvalds * 16981da177e4SLinus Torvalds * The caller must have a ref on the inode. 16991da177e4SLinus Torvalds */ 17001da177e4SLinus Torvalds int sync_inode(struct inode *inode, struct writeback_control *wbc) 17011da177e4SLinus Torvalds { 17024f8ad655SJan Kara return writeback_single_inode(inode, &inode_to_bdi(inode)->wb, wbc); 17031da177e4SLinus Torvalds } 17041da177e4SLinus Torvalds EXPORT_SYMBOL(sync_inode); 1705c3765016SChristoph Hellwig 1706c3765016SChristoph Hellwig /** 1707c691b9d9SAndrew Morton * sync_inode_metadata - write an inode to disk 1708c3765016SChristoph Hellwig * @inode: the inode to sync 1709c3765016SChristoph Hellwig * @wait: wait for I/O to complete. 1710c3765016SChristoph Hellwig * 1711c691b9d9SAndrew Morton * Write an inode to disk and adjust its dirty state after completion. 1712c3765016SChristoph Hellwig * 1713c3765016SChristoph Hellwig * Note: only writes the actual inode, no associated data or other metadata. 1714c3765016SChristoph Hellwig */ 1715c3765016SChristoph Hellwig int sync_inode_metadata(struct inode *inode, int wait) 1716c3765016SChristoph Hellwig { 1717c3765016SChristoph Hellwig struct writeback_control wbc = { 1718c3765016SChristoph Hellwig .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE, 1719c3765016SChristoph Hellwig .nr_to_write = 0, /* metadata-only */ 1720c3765016SChristoph Hellwig }; 1721c3765016SChristoph Hellwig 1722c3765016SChristoph Hellwig return sync_inode(inode, &wbc); 1723c3765016SChristoph Hellwig } 1724c3765016SChristoph Hellwig EXPORT_SYMBOL(sync_inode_metadata); 1725