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> 3021c6321fSTejun Heo #include <linux/memcontrol.h> 3107f3f05cSDavid Howells #include "internal.h" 321da177e4SLinus Torvalds 33d0bceac7SJens Axboe /* 34bc31b86aSWu Fengguang * 4MB minimal write chunk size 35bc31b86aSWu Fengguang */ 36bc31b86aSWu Fengguang #define MIN_WRITEBACK_PAGES (4096UL >> (PAGE_CACHE_SHIFT - 10)) 37bc31b86aSWu Fengguang 38cc395d7fSTejun Heo struct wb_completion { 39cc395d7fSTejun Heo atomic_t cnt; 40cc395d7fSTejun Heo }; 41cc395d7fSTejun Heo 42bc31b86aSWu Fengguang /* 43c4a77a6cSJens Axboe * Passed into wb_writeback(), essentially a subset of writeback_control 44c4a77a6cSJens Axboe */ 4583ba7b07SChristoph Hellwig struct wb_writeback_work { 46c4a77a6cSJens Axboe long nr_pages; 47c4a77a6cSJens Axboe struct super_block *sb; 480dc83bd3SJan Kara unsigned long *older_than_this; 49c4a77a6cSJens Axboe enum writeback_sync_modes sync_mode; 506e6938b6SWu Fengguang unsigned int tagged_writepages:1; 5152957fe1SH Hartley Sweeten unsigned int for_kupdate:1; 5252957fe1SH Hartley Sweeten unsigned int range_cyclic:1; 5352957fe1SH Hartley Sweeten unsigned int for_background:1; 547747bd4bSDave Chinner unsigned int for_sync:1; /* sync(2) WB_SYNC_ALL writeback */ 55ac7b19a3STejun Heo unsigned int auto_free:1; /* free on completion */ 560e175a18SCurt Wohlgemuth enum wb_reason reason; /* why was writeback initiated? */ 57c4a77a6cSJens Axboe 588010c3b6SJens Axboe struct list_head list; /* pending work list */ 59cc395d7fSTejun Heo struct wb_completion *done; /* set if the caller waits */ 6003ba3782SJens Axboe }; 6103ba3782SJens Axboe 62a2f48706STheodore Ts'o /* 63cc395d7fSTejun Heo * If one wants to wait for one or more wb_writeback_works, each work's 64cc395d7fSTejun Heo * ->done should be set to a wb_completion defined using the following 65cc395d7fSTejun Heo * macro. Once all work items are issued with wb_queue_work(), the caller 66cc395d7fSTejun Heo * can wait for the completion of all using wb_wait_for_completion(). Work 67cc395d7fSTejun Heo * items which are waited upon aren't freed automatically on completion. 68cc395d7fSTejun Heo */ 69cc395d7fSTejun Heo #define DEFINE_WB_COMPLETION_ONSTACK(cmpl) \ 70cc395d7fSTejun Heo struct wb_completion cmpl = { \ 71cc395d7fSTejun Heo .cnt = ATOMIC_INIT(1), \ 72cc395d7fSTejun Heo } 73cc395d7fSTejun Heo 74cc395d7fSTejun Heo 75cc395d7fSTejun Heo /* 76a2f48706STheodore Ts'o * If an inode is constantly having its pages dirtied, but then the 77a2f48706STheodore Ts'o * updates stop dirtytime_expire_interval seconds in the past, it's 78a2f48706STheodore Ts'o * possible for the worst case time between when an inode has its 79a2f48706STheodore Ts'o * timestamps updated and when they finally get written out to be two 80a2f48706STheodore Ts'o * dirtytime_expire_intervals. We set the default to 12 hours (in 81a2f48706STheodore Ts'o * seconds), which means most of the time inodes will have their 82a2f48706STheodore Ts'o * timestamps written to disk after 12 hours, but in the worst case a 83a2f48706STheodore Ts'o * few inodes might not their timestamps updated for 24 hours. 84a2f48706STheodore Ts'o */ 85a2f48706STheodore Ts'o unsigned int dirtytime_expire_interval = 12 * 60 * 60; 86a2f48706STheodore Ts'o 877ccf19a8SNick Piggin static inline struct inode *wb_inode(struct list_head *head) 887ccf19a8SNick Piggin { 89c7f54084SDave Chinner return list_entry(head, struct inode, i_io_list); 907ccf19a8SNick Piggin } 917ccf19a8SNick Piggin 9215eb77a0SWu Fengguang /* 9315eb77a0SWu Fengguang * Include the creation of the trace points after defining the 9415eb77a0SWu Fengguang * wb_writeback_work structure and inline functions so that the definition 9515eb77a0SWu Fengguang * remains local to this file. 9615eb77a0SWu Fengguang */ 9715eb77a0SWu Fengguang #define CREATE_TRACE_POINTS 9815eb77a0SWu Fengguang #include <trace/events/writeback.h> 9915eb77a0SWu Fengguang 100774016b2SSteven Whitehouse EXPORT_TRACEPOINT_SYMBOL_GPL(wbc_writepage); 101774016b2SSteven Whitehouse 102d6c10f1fSTejun Heo static bool wb_io_lists_populated(struct bdi_writeback *wb) 103d6c10f1fSTejun Heo { 104d6c10f1fSTejun Heo if (wb_has_dirty_io(wb)) { 105d6c10f1fSTejun Heo return false; 106d6c10f1fSTejun Heo } else { 107d6c10f1fSTejun Heo set_bit(WB_has_dirty_io, &wb->state); 10895a46c65STejun Heo WARN_ON_ONCE(!wb->avg_write_bandwidth); 109766a9d6eSTejun Heo atomic_long_add(wb->avg_write_bandwidth, 110766a9d6eSTejun Heo &wb->bdi->tot_write_bandwidth); 111d6c10f1fSTejun Heo return true; 112d6c10f1fSTejun Heo } 113d6c10f1fSTejun Heo } 114d6c10f1fSTejun Heo 115d6c10f1fSTejun Heo static void wb_io_lists_depopulated(struct bdi_writeback *wb) 116d6c10f1fSTejun Heo { 117d6c10f1fSTejun Heo if (wb_has_dirty_io(wb) && list_empty(&wb->b_dirty) && 118766a9d6eSTejun Heo list_empty(&wb->b_io) && list_empty(&wb->b_more_io)) { 119d6c10f1fSTejun Heo clear_bit(WB_has_dirty_io, &wb->state); 12095a46c65STejun Heo WARN_ON_ONCE(atomic_long_sub_return(wb->avg_write_bandwidth, 12195a46c65STejun Heo &wb->bdi->tot_write_bandwidth) < 0); 122766a9d6eSTejun Heo } 123d6c10f1fSTejun Heo } 124d6c10f1fSTejun Heo 125d6c10f1fSTejun Heo /** 126c7f54084SDave Chinner * inode_io_list_move_locked - move an inode onto a bdi_writeback IO list 127d6c10f1fSTejun Heo * @inode: inode to be moved 128d6c10f1fSTejun Heo * @wb: target bdi_writeback 129d6c10f1fSTejun Heo * @head: one of @wb->b_{dirty|io|more_io} 130d6c10f1fSTejun Heo * 131c7f54084SDave Chinner * Move @inode->i_io_list to @list of @wb and set %WB_has_dirty_io. 132d6c10f1fSTejun Heo * Returns %true if @inode is the first occupant of the !dirty_time IO 133d6c10f1fSTejun Heo * lists; otherwise, %false. 134d6c10f1fSTejun Heo */ 135c7f54084SDave Chinner static bool inode_io_list_move_locked(struct inode *inode, 136d6c10f1fSTejun Heo struct bdi_writeback *wb, 137d6c10f1fSTejun Heo struct list_head *head) 138d6c10f1fSTejun Heo { 139d6c10f1fSTejun Heo assert_spin_locked(&wb->list_lock); 140d6c10f1fSTejun Heo 141c7f54084SDave Chinner list_move(&inode->i_io_list, head); 142d6c10f1fSTejun Heo 143d6c10f1fSTejun Heo /* dirty_time doesn't count as dirty_io until expiration */ 144d6c10f1fSTejun Heo if (head != &wb->b_dirty_time) 145d6c10f1fSTejun Heo return wb_io_lists_populated(wb); 146d6c10f1fSTejun Heo 147d6c10f1fSTejun Heo wb_io_lists_depopulated(wb); 148d6c10f1fSTejun Heo return false; 149d6c10f1fSTejun Heo } 150d6c10f1fSTejun Heo 151d6c10f1fSTejun Heo /** 152c7f54084SDave Chinner * inode_io_list_del_locked - remove an inode from its bdi_writeback IO list 153d6c10f1fSTejun Heo * @inode: inode to be removed 154d6c10f1fSTejun Heo * @wb: bdi_writeback @inode is being removed from 155d6c10f1fSTejun Heo * 156d6c10f1fSTejun Heo * Remove @inode which may be on one of @wb->b_{dirty|io|more_io} lists and 157d6c10f1fSTejun Heo * clear %WB_has_dirty_io if all are empty afterwards. 158d6c10f1fSTejun Heo */ 159c7f54084SDave Chinner static void inode_io_list_del_locked(struct inode *inode, 160d6c10f1fSTejun Heo struct bdi_writeback *wb) 161d6c10f1fSTejun Heo { 162d6c10f1fSTejun Heo assert_spin_locked(&wb->list_lock); 163d6c10f1fSTejun Heo 164c7f54084SDave Chinner list_del_init(&inode->i_io_list); 165d6c10f1fSTejun Heo wb_io_lists_depopulated(wb); 166d6c10f1fSTejun Heo } 167d6c10f1fSTejun Heo 168f0054bb1STejun Heo static void wb_wakeup(struct bdi_writeback *wb) 1695acda9d1SJan Kara { 170f0054bb1STejun Heo spin_lock_bh(&wb->work_lock); 171f0054bb1STejun Heo if (test_bit(WB_registered, &wb->state)) 172f0054bb1STejun Heo mod_delayed_work(bdi_wq, &wb->dwork, 0); 173f0054bb1STejun Heo spin_unlock_bh(&wb->work_lock); 1745acda9d1SJan Kara } 1755acda9d1SJan Kara 176f0054bb1STejun Heo static void wb_queue_work(struct bdi_writeback *wb, 1776585027aSJan Kara struct wb_writeback_work *work) 1786585027aSJan Kara { 1795634cc2aSTejun Heo trace_writeback_queue(wb, work); 1806585027aSJan Kara 181f0054bb1STejun Heo spin_lock_bh(&wb->work_lock); 1828a1270cdSTejun Heo if (!test_bit(WB_registered, &wb->state)) 1835acda9d1SJan Kara goto out_unlock; 184cc395d7fSTejun Heo if (work->done) 185cc395d7fSTejun Heo atomic_inc(&work->done->cnt); 186f0054bb1STejun Heo list_add_tail(&work->list, &wb->work_list); 187f0054bb1STejun Heo mod_delayed_work(bdi_wq, &wb->dwork, 0); 1885acda9d1SJan Kara out_unlock: 189f0054bb1STejun Heo spin_unlock_bh(&wb->work_lock); 19003ba3782SJens Axboe } 1911da177e4SLinus Torvalds 192cc395d7fSTejun Heo /** 193cc395d7fSTejun Heo * wb_wait_for_completion - wait for completion of bdi_writeback_works 194cc395d7fSTejun Heo * @bdi: bdi work items were issued to 195cc395d7fSTejun Heo * @done: target wb_completion 196cc395d7fSTejun Heo * 197cc395d7fSTejun Heo * Wait for one or more work items issued to @bdi with their ->done field 198cc395d7fSTejun Heo * set to @done, which should have been defined with 199cc395d7fSTejun Heo * DEFINE_WB_COMPLETION_ONSTACK(). This function returns after all such 200cc395d7fSTejun Heo * work items are completed. Work items which are waited upon aren't freed 201cc395d7fSTejun Heo * automatically on completion. 202cc395d7fSTejun Heo */ 203cc395d7fSTejun Heo static void wb_wait_for_completion(struct backing_dev_info *bdi, 204cc395d7fSTejun Heo struct wb_completion *done) 205cc395d7fSTejun Heo { 206cc395d7fSTejun Heo atomic_dec(&done->cnt); /* put down the initial count */ 207cc395d7fSTejun Heo wait_event(bdi->wb_waitq, !atomic_read(&done->cnt)); 208cc395d7fSTejun Heo } 209cc395d7fSTejun Heo 210703c2708STejun Heo #ifdef CONFIG_CGROUP_WRITEBACK 211703c2708STejun Heo 2122a814908STejun Heo /* parameters for foreign inode detection, see wb_detach_inode() */ 2132a814908STejun Heo #define WB_FRN_TIME_SHIFT 13 /* 1s = 2^13, upto 8 secs w/ 16bit */ 2142a814908STejun Heo #define WB_FRN_TIME_AVG_SHIFT 3 /* avg = avg * 7/8 + new * 1/8 */ 2152a814908STejun Heo #define WB_FRN_TIME_CUT_DIV 2 /* ignore rounds < avg / 2 */ 2162a814908STejun Heo #define WB_FRN_TIME_PERIOD (2 * (1 << WB_FRN_TIME_SHIFT)) /* 2s */ 2172a814908STejun Heo 2182a814908STejun Heo #define WB_FRN_HIST_SLOTS 16 /* inode->i_wb_frn_history is 16bit */ 2192a814908STejun Heo #define WB_FRN_HIST_UNIT (WB_FRN_TIME_PERIOD / WB_FRN_HIST_SLOTS) 2202a814908STejun Heo /* each slot's duration is 2s / 16 */ 2212a814908STejun Heo #define WB_FRN_HIST_THR_SLOTS (WB_FRN_HIST_SLOTS / 2) 2222a814908STejun Heo /* if foreign slots >= 8, switch */ 2232a814908STejun Heo #define WB_FRN_HIST_MAX_SLOTS (WB_FRN_HIST_THR_SLOTS / 2 + 1) 2242a814908STejun Heo /* one round can affect upto 5 slots */ 2252a814908STejun Heo 226a1a0e23eSTejun Heo static atomic_t isw_nr_in_flight = ATOMIC_INIT(0); 227a1a0e23eSTejun Heo static struct workqueue_struct *isw_wq; 228a1a0e23eSTejun Heo 22921c6321fSTejun Heo void __inode_attach_wb(struct inode *inode, struct page *page) 23021c6321fSTejun Heo { 23121c6321fSTejun Heo struct backing_dev_info *bdi = inode_to_bdi(inode); 23221c6321fSTejun Heo struct bdi_writeback *wb = NULL; 23321c6321fSTejun Heo 23421c6321fSTejun Heo if (inode_cgwb_enabled(inode)) { 23521c6321fSTejun Heo struct cgroup_subsys_state *memcg_css; 23621c6321fSTejun Heo 23721c6321fSTejun Heo if (page) { 23821c6321fSTejun Heo memcg_css = mem_cgroup_css_from_page(page); 23921c6321fSTejun Heo wb = wb_get_create(bdi, memcg_css, GFP_ATOMIC); 24021c6321fSTejun Heo } else { 24121c6321fSTejun Heo /* must pin memcg_css, see wb_get_create() */ 24221c6321fSTejun Heo memcg_css = task_get_css(current, memory_cgrp_id); 24321c6321fSTejun Heo wb = wb_get_create(bdi, memcg_css, GFP_ATOMIC); 24421c6321fSTejun Heo css_put(memcg_css); 24521c6321fSTejun Heo } 24621c6321fSTejun Heo } 24721c6321fSTejun Heo 24821c6321fSTejun Heo if (!wb) 24921c6321fSTejun Heo wb = &bdi->wb; 25021c6321fSTejun Heo 25121c6321fSTejun Heo /* 25221c6321fSTejun Heo * There may be multiple instances of this function racing to 25321c6321fSTejun Heo * update the same inode. Use cmpxchg() to tell the winner. 25421c6321fSTejun Heo */ 25521c6321fSTejun Heo if (unlikely(cmpxchg(&inode->i_wb, NULL, wb))) 25621c6321fSTejun Heo wb_put(wb); 25721c6321fSTejun Heo } 25821c6321fSTejun Heo 259703c2708STejun Heo /** 26087e1d789STejun Heo * locked_inode_to_wb_and_lock_list - determine a locked inode's wb and lock it 26187e1d789STejun Heo * @inode: inode of interest with i_lock held 26287e1d789STejun Heo * 26387e1d789STejun Heo * Returns @inode's wb with its list_lock held. @inode->i_lock must be 26487e1d789STejun Heo * held on entry and is released on return. The returned wb is guaranteed 26587e1d789STejun Heo * to stay @inode's associated wb until its list_lock is released. 26687e1d789STejun Heo */ 26787e1d789STejun Heo static struct bdi_writeback * 26887e1d789STejun Heo locked_inode_to_wb_and_lock_list(struct inode *inode) 26987e1d789STejun Heo __releases(&inode->i_lock) 27087e1d789STejun Heo __acquires(&wb->list_lock) 27187e1d789STejun Heo { 27287e1d789STejun Heo while (true) { 27387e1d789STejun Heo struct bdi_writeback *wb = inode_to_wb(inode); 27487e1d789STejun Heo 27587e1d789STejun Heo /* 27687e1d789STejun Heo * inode_to_wb() association is protected by both 27787e1d789STejun Heo * @inode->i_lock and @wb->list_lock but list_lock nests 27887e1d789STejun Heo * outside i_lock. Drop i_lock and verify that the 27987e1d789STejun Heo * association hasn't changed after acquiring list_lock. 28087e1d789STejun Heo */ 28187e1d789STejun Heo wb_get(wb); 28287e1d789STejun Heo spin_unlock(&inode->i_lock); 28387e1d789STejun Heo spin_lock(&wb->list_lock); 28487e1d789STejun Heo wb_put(wb); /* not gonna deref it anymore */ 28587e1d789STejun Heo 286aaa2cacfSTejun Heo /* i_wb may have changed inbetween, can't use inode_to_wb() */ 287aaa2cacfSTejun Heo if (likely(wb == inode->i_wb)) 28887e1d789STejun Heo return wb; /* @inode already has ref */ 28987e1d789STejun Heo 29087e1d789STejun Heo spin_unlock(&wb->list_lock); 29187e1d789STejun Heo cpu_relax(); 29287e1d789STejun Heo spin_lock(&inode->i_lock); 29387e1d789STejun Heo } 29487e1d789STejun Heo } 29587e1d789STejun Heo 29687e1d789STejun Heo /** 29787e1d789STejun Heo * inode_to_wb_and_lock_list - determine an inode's wb and lock it 29887e1d789STejun Heo * @inode: inode of interest 29987e1d789STejun Heo * 30087e1d789STejun Heo * Same as locked_inode_to_wb_and_lock_list() but @inode->i_lock isn't held 30187e1d789STejun Heo * on entry. 30287e1d789STejun Heo */ 30387e1d789STejun Heo static struct bdi_writeback *inode_to_wb_and_lock_list(struct inode *inode) 30487e1d789STejun Heo __acquires(&wb->list_lock) 30587e1d789STejun Heo { 30687e1d789STejun Heo spin_lock(&inode->i_lock); 30787e1d789STejun Heo return locked_inode_to_wb_and_lock_list(inode); 30887e1d789STejun Heo } 30987e1d789STejun Heo 310682aa8e1STejun Heo struct inode_switch_wbs_context { 311682aa8e1STejun Heo struct inode *inode; 312682aa8e1STejun Heo struct bdi_writeback *new_wb; 313682aa8e1STejun Heo 314682aa8e1STejun Heo struct rcu_head rcu_head; 315682aa8e1STejun Heo struct work_struct work; 316682aa8e1STejun Heo }; 317682aa8e1STejun Heo 318682aa8e1STejun Heo static void inode_switch_wbs_work_fn(struct work_struct *work) 319682aa8e1STejun Heo { 320682aa8e1STejun Heo struct inode_switch_wbs_context *isw = 321682aa8e1STejun Heo container_of(work, struct inode_switch_wbs_context, work); 322682aa8e1STejun Heo struct inode *inode = isw->inode; 323d10c8095STejun Heo struct address_space *mapping = inode->i_mapping; 324d10c8095STejun Heo struct bdi_writeback *old_wb = inode->i_wb; 325682aa8e1STejun Heo struct bdi_writeback *new_wb = isw->new_wb; 326d10c8095STejun Heo struct radix_tree_iter iter; 327d10c8095STejun Heo bool switched = false; 328d10c8095STejun Heo void **slot; 329682aa8e1STejun Heo 330682aa8e1STejun Heo /* 331682aa8e1STejun Heo * By the time control reaches here, RCU grace period has passed 332682aa8e1STejun Heo * since I_WB_SWITCH assertion and all wb stat update transactions 333682aa8e1STejun Heo * between unlocked_inode_to_wb_begin/end() are guaranteed to be 334682aa8e1STejun Heo * synchronizing against mapping->tree_lock. 335d10c8095STejun Heo * 336d10c8095STejun Heo * Grabbing old_wb->list_lock, inode->i_lock and mapping->tree_lock 337d10c8095STejun Heo * gives us exclusion against all wb related operations on @inode 338d10c8095STejun Heo * including IO list manipulations and stat updates. 339682aa8e1STejun Heo */ 340d10c8095STejun Heo if (old_wb < new_wb) { 341d10c8095STejun Heo spin_lock(&old_wb->list_lock); 342d10c8095STejun Heo spin_lock_nested(&new_wb->list_lock, SINGLE_DEPTH_NESTING); 343d10c8095STejun Heo } else { 344d10c8095STejun Heo spin_lock(&new_wb->list_lock); 345d10c8095STejun Heo spin_lock_nested(&old_wb->list_lock, SINGLE_DEPTH_NESTING); 346d10c8095STejun Heo } 347682aa8e1STejun Heo spin_lock(&inode->i_lock); 348d10c8095STejun Heo spin_lock_irq(&mapping->tree_lock); 349682aa8e1STejun Heo 350d10c8095STejun Heo /* 351d10c8095STejun Heo * Once I_FREEING is visible under i_lock, the eviction path owns 352c7f54084SDave Chinner * the inode and we shouldn't modify ->i_io_list. 353d10c8095STejun Heo */ 354d10c8095STejun Heo if (unlikely(inode->i_state & I_FREEING)) 355d10c8095STejun Heo goto skip_switch; 356d10c8095STejun Heo 357d10c8095STejun Heo /* 358d10c8095STejun Heo * Count and transfer stats. Note that PAGECACHE_TAG_DIRTY points 359d10c8095STejun Heo * to possibly dirty pages while PAGECACHE_TAG_WRITEBACK points to 360d10c8095STejun Heo * pages actually under underwriteback. 361d10c8095STejun Heo */ 362d10c8095STejun Heo radix_tree_for_each_tagged(slot, &mapping->page_tree, &iter, 0, 363d10c8095STejun Heo PAGECACHE_TAG_DIRTY) { 364d10c8095STejun Heo struct page *page = radix_tree_deref_slot_protected(slot, 365d10c8095STejun Heo &mapping->tree_lock); 366d10c8095STejun Heo if (likely(page) && PageDirty(page)) { 367d10c8095STejun Heo __dec_wb_stat(old_wb, WB_RECLAIMABLE); 368d10c8095STejun Heo __inc_wb_stat(new_wb, WB_RECLAIMABLE); 369d10c8095STejun Heo } 370d10c8095STejun Heo } 371d10c8095STejun Heo 372d10c8095STejun Heo radix_tree_for_each_tagged(slot, &mapping->page_tree, &iter, 0, 373d10c8095STejun Heo PAGECACHE_TAG_WRITEBACK) { 374d10c8095STejun Heo struct page *page = radix_tree_deref_slot_protected(slot, 375d10c8095STejun Heo &mapping->tree_lock); 376d10c8095STejun Heo if (likely(page)) { 377d10c8095STejun Heo WARN_ON_ONCE(!PageWriteback(page)); 378d10c8095STejun Heo __dec_wb_stat(old_wb, WB_WRITEBACK); 379d10c8095STejun Heo __inc_wb_stat(new_wb, WB_WRITEBACK); 380d10c8095STejun Heo } 381d10c8095STejun Heo } 382d10c8095STejun Heo 383d10c8095STejun Heo wb_get(new_wb); 384d10c8095STejun Heo 385d10c8095STejun Heo /* 386d10c8095STejun Heo * Transfer to @new_wb's IO list if necessary. The specific list 387d10c8095STejun Heo * @inode was on is ignored and the inode is put on ->b_dirty which 388d10c8095STejun Heo * is always correct including from ->b_dirty_time. The transfer 389d10c8095STejun Heo * preserves @inode->dirtied_when ordering. 390d10c8095STejun Heo */ 391c7f54084SDave Chinner if (!list_empty(&inode->i_io_list)) { 392d10c8095STejun Heo struct inode *pos; 393d10c8095STejun Heo 394c7f54084SDave Chinner inode_io_list_del_locked(inode, old_wb); 395d10c8095STejun Heo inode->i_wb = new_wb; 396c7f54084SDave Chinner list_for_each_entry(pos, &new_wb->b_dirty, i_io_list) 397d10c8095STejun Heo if (time_after_eq(inode->dirtied_when, 398d10c8095STejun Heo pos->dirtied_when)) 399d10c8095STejun Heo break; 400c7f54084SDave Chinner inode_io_list_move_locked(inode, new_wb, pos->i_io_list.prev); 401d10c8095STejun Heo } else { 402d10c8095STejun Heo inode->i_wb = new_wb; 403d10c8095STejun Heo } 404d10c8095STejun Heo 405d10c8095STejun Heo /* ->i_wb_frn updates may race wbc_detach_inode() but doesn't matter */ 406682aa8e1STejun Heo inode->i_wb_frn_winner = 0; 407682aa8e1STejun Heo inode->i_wb_frn_avg_time = 0; 408682aa8e1STejun Heo inode->i_wb_frn_history = 0; 409d10c8095STejun Heo switched = true; 410d10c8095STejun Heo skip_switch: 411682aa8e1STejun Heo /* 412682aa8e1STejun Heo * Paired with load_acquire in unlocked_inode_to_wb_begin() and 413682aa8e1STejun Heo * ensures that the new wb is visible if they see !I_WB_SWITCH. 414682aa8e1STejun Heo */ 415682aa8e1STejun Heo smp_store_release(&inode->i_state, inode->i_state & ~I_WB_SWITCH); 416682aa8e1STejun Heo 417d10c8095STejun Heo spin_unlock_irq(&mapping->tree_lock); 418682aa8e1STejun Heo spin_unlock(&inode->i_lock); 419d10c8095STejun Heo spin_unlock(&new_wb->list_lock); 420d10c8095STejun Heo spin_unlock(&old_wb->list_lock); 421d10c8095STejun Heo 422d10c8095STejun Heo if (switched) { 423d10c8095STejun Heo wb_wakeup(new_wb); 424d10c8095STejun Heo wb_put(old_wb); 425d10c8095STejun Heo } 426d10c8095STejun Heo wb_put(new_wb); 427682aa8e1STejun Heo 428682aa8e1STejun Heo iput(inode); 429682aa8e1STejun Heo kfree(isw); 430a1a0e23eSTejun Heo 431a1a0e23eSTejun Heo atomic_dec(&isw_nr_in_flight); 432682aa8e1STejun Heo } 433682aa8e1STejun Heo 434682aa8e1STejun Heo static void inode_switch_wbs_rcu_fn(struct rcu_head *rcu_head) 435682aa8e1STejun Heo { 436682aa8e1STejun Heo struct inode_switch_wbs_context *isw = container_of(rcu_head, 437682aa8e1STejun Heo struct inode_switch_wbs_context, rcu_head); 438682aa8e1STejun Heo 439682aa8e1STejun Heo /* needs to grab bh-unsafe locks, bounce to work item */ 440682aa8e1STejun Heo INIT_WORK(&isw->work, inode_switch_wbs_work_fn); 441a1a0e23eSTejun Heo queue_work(isw_wq, &isw->work); 442682aa8e1STejun Heo } 443682aa8e1STejun Heo 444682aa8e1STejun Heo /** 445682aa8e1STejun Heo * inode_switch_wbs - change the wb association of an inode 446682aa8e1STejun Heo * @inode: target inode 447682aa8e1STejun Heo * @new_wb_id: ID of the new wb 448682aa8e1STejun Heo * 449682aa8e1STejun Heo * Switch @inode's wb association to the wb identified by @new_wb_id. The 450682aa8e1STejun Heo * switching is performed asynchronously and may fail silently. 451682aa8e1STejun Heo */ 452682aa8e1STejun Heo static void inode_switch_wbs(struct inode *inode, int new_wb_id) 453682aa8e1STejun Heo { 454682aa8e1STejun Heo struct backing_dev_info *bdi = inode_to_bdi(inode); 455682aa8e1STejun Heo struct cgroup_subsys_state *memcg_css; 456682aa8e1STejun Heo struct inode_switch_wbs_context *isw; 457682aa8e1STejun Heo 458682aa8e1STejun Heo /* noop if seems to be already in progress */ 459682aa8e1STejun Heo if (inode->i_state & I_WB_SWITCH) 460682aa8e1STejun Heo return; 461682aa8e1STejun Heo 462682aa8e1STejun Heo isw = kzalloc(sizeof(*isw), GFP_ATOMIC); 463682aa8e1STejun Heo if (!isw) 464682aa8e1STejun Heo return; 465682aa8e1STejun Heo 466682aa8e1STejun Heo /* find and pin the new wb */ 467682aa8e1STejun Heo rcu_read_lock(); 468682aa8e1STejun Heo memcg_css = css_from_id(new_wb_id, &memory_cgrp_subsys); 469682aa8e1STejun Heo if (memcg_css) 470682aa8e1STejun Heo isw->new_wb = wb_get_create(bdi, memcg_css, GFP_ATOMIC); 471682aa8e1STejun Heo rcu_read_unlock(); 472682aa8e1STejun Heo if (!isw->new_wb) 473682aa8e1STejun Heo goto out_free; 474682aa8e1STejun Heo 475682aa8e1STejun Heo /* while holding I_WB_SWITCH, no one else can update the association */ 476682aa8e1STejun Heo spin_lock(&inode->i_lock); 477a1a0e23eSTejun Heo if (!(inode->i_sb->s_flags & MS_ACTIVE) || 478a1a0e23eSTejun Heo inode->i_state & (I_WB_SWITCH | I_FREEING) || 479a1a0e23eSTejun Heo inode_to_wb(inode) == isw->new_wb) { 480a1a0e23eSTejun Heo spin_unlock(&inode->i_lock); 481a1a0e23eSTejun Heo goto out_free; 482a1a0e23eSTejun Heo } 483682aa8e1STejun Heo inode->i_state |= I_WB_SWITCH; 484682aa8e1STejun Heo spin_unlock(&inode->i_lock); 485682aa8e1STejun Heo 486682aa8e1STejun Heo ihold(inode); 487682aa8e1STejun Heo isw->inode = inode; 488682aa8e1STejun Heo 489a1a0e23eSTejun Heo atomic_inc(&isw_nr_in_flight); 490a1a0e23eSTejun Heo 491682aa8e1STejun Heo /* 492682aa8e1STejun Heo * In addition to synchronizing among switchers, I_WB_SWITCH tells 493682aa8e1STejun Heo * the RCU protected stat update paths to grab the mapping's 494682aa8e1STejun Heo * tree_lock so that stat transfer can synchronize against them. 495682aa8e1STejun Heo * Let's continue after I_WB_SWITCH is guaranteed to be visible. 496682aa8e1STejun Heo */ 497682aa8e1STejun Heo call_rcu(&isw->rcu_head, inode_switch_wbs_rcu_fn); 498682aa8e1STejun Heo return; 499682aa8e1STejun Heo 500682aa8e1STejun Heo out_free: 501682aa8e1STejun Heo if (isw->new_wb) 502682aa8e1STejun Heo wb_put(isw->new_wb); 503682aa8e1STejun Heo kfree(isw); 504682aa8e1STejun Heo } 505682aa8e1STejun Heo 50687e1d789STejun Heo /** 507b16b1debSTejun Heo * wbc_attach_and_unlock_inode - associate wbc with target inode and unlock it 508b16b1debSTejun Heo * @wbc: writeback_control of interest 509b16b1debSTejun Heo * @inode: target inode 510b16b1debSTejun Heo * 511b16b1debSTejun Heo * @inode is locked and about to be written back under the control of @wbc. 512b16b1debSTejun Heo * Record @inode's writeback context into @wbc and unlock the i_lock. On 513b16b1debSTejun Heo * writeback completion, wbc_detach_inode() should be called. This is used 514b16b1debSTejun Heo * to track the cgroup writeback context. 515b16b1debSTejun Heo */ 516b16b1debSTejun Heo void wbc_attach_and_unlock_inode(struct writeback_control *wbc, 517b16b1debSTejun Heo struct inode *inode) 518b16b1debSTejun Heo { 519dd73e4b7STejun Heo if (!inode_cgwb_enabled(inode)) { 520dd73e4b7STejun Heo spin_unlock(&inode->i_lock); 521dd73e4b7STejun Heo return; 522dd73e4b7STejun Heo } 523dd73e4b7STejun Heo 524b16b1debSTejun Heo wbc->wb = inode_to_wb(inode); 5252a814908STejun Heo wbc->inode = inode; 5262a814908STejun Heo 5272a814908STejun Heo wbc->wb_id = wbc->wb->memcg_css->id; 5282a814908STejun Heo wbc->wb_lcand_id = inode->i_wb_frn_winner; 5292a814908STejun Heo wbc->wb_tcand_id = 0; 5302a814908STejun Heo wbc->wb_bytes = 0; 5312a814908STejun Heo wbc->wb_lcand_bytes = 0; 5322a814908STejun Heo wbc->wb_tcand_bytes = 0; 5332a814908STejun Heo 534b16b1debSTejun Heo wb_get(wbc->wb); 535b16b1debSTejun Heo spin_unlock(&inode->i_lock); 536e8a7abf5STejun Heo 537e8a7abf5STejun Heo /* 538e8a7abf5STejun Heo * A dying wb indicates that the memcg-blkcg mapping has changed 539e8a7abf5STejun Heo * and a new wb is already serving the memcg. Switch immediately. 540e8a7abf5STejun Heo */ 541e8a7abf5STejun Heo if (unlikely(wb_dying(wbc->wb))) 542e8a7abf5STejun Heo inode_switch_wbs(inode, wbc->wb_id); 543b16b1debSTejun Heo } 544b16b1debSTejun Heo 545b16b1debSTejun Heo /** 5462a814908STejun Heo * wbc_detach_inode - disassociate wbc from inode and perform foreign detection 5472a814908STejun Heo * @wbc: writeback_control of the just finished writeback 548b16b1debSTejun Heo * 549b16b1debSTejun Heo * To be called after a writeback attempt of an inode finishes and undoes 550b16b1debSTejun Heo * wbc_attach_and_unlock_inode(). Can be called under any context. 5512a814908STejun Heo * 5522a814908STejun Heo * As concurrent write sharing of an inode is expected to be very rare and 5532a814908STejun Heo * memcg only tracks page ownership on first-use basis severely confining 5542a814908STejun Heo * the usefulness of such sharing, cgroup writeback tracks ownership 5552a814908STejun Heo * per-inode. While the support for concurrent write sharing of an inode 5562a814908STejun Heo * is deemed unnecessary, an inode being written to by different cgroups at 5572a814908STejun Heo * different points in time is a lot more common, and, more importantly, 5582a814908STejun Heo * charging only by first-use can too readily lead to grossly incorrect 5592a814908STejun Heo * behaviors (single foreign page can lead to gigabytes of writeback to be 5602a814908STejun Heo * incorrectly attributed). 5612a814908STejun Heo * 5622a814908STejun Heo * To resolve this issue, cgroup writeback detects the majority dirtier of 5632a814908STejun Heo * an inode and transfers the ownership to it. To avoid unnnecessary 5642a814908STejun Heo * oscillation, the detection mechanism keeps track of history and gives 5652a814908STejun Heo * out the switch verdict only if the foreign usage pattern is stable over 5662a814908STejun Heo * a certain amount of time and/or writeback attempts. 5672a814908STejun Heo * 5682a814908STejun Heo * On each writeback attempt, @wbc tries to detect the majority writer 5692a814908STejun Heo * using Boyer-Moore majority vote algorithm. In addition to the byte 5702a814908STejun Heo * count from the majority voting, it also counts the bytes written for the 5712a814908STejun Heo * current wb and the last round's winner wb (max of last round's current 5722a814908STejun Heo * wb, the winner from two rounds ago, and the last round's majority 5732a814908STejun Heo * candidate). Keeping track of the historical winner helps the algorithm 5742a814908STejun Heo * to semi-reliably detect the most active writer even when it's not the 5752a814908STejun Heo * absolute majority. 5762a814908STejun Heo * 5772a814908STejun Heo * Once the winner of the round is determined, whether the winner is 5782a814908STejun Heo * foreign or not and how much IO time the round consumed is recorded in 5792a814908STejun Heo * inode->i_wb_frn_history. If the amount of recorded foreign IO time is 5802a814908STejun Heo * over a certain threshold, the switch verdict is given. 581b16b1debSTejun Heo */ 582b16b1debSTejun Heo void wbc_detach_inode(struct writeback_control *wbc) 583b16b1debSTejun Heo { 5842a814908STejun Heo struct bdi_writeback *wb = wbc->wb; 5852a814908STejun Heo struct inode *inode = wbc->inode; 586dd73e4b7STejun Heo unsigned long avg_time, max_bytes, max_time; 587dd73e4b7STejun Heo u16 history; 5882a814908STejun Heo int max_id; 5892a814908STejun Heo 590dd73e4b7STejun Heo if (!wb) 591dd73e4b7STejun Heo return; 592dd73e4b7STejun Heo 593dd73e4b7STejun Heo history = inode->i_wb_frn_history; 594dd73e4b7STejun Heo avg_time = inode->i_wb_frn_avg_time; 595dd73e4b7STejun Heo 5962a814908STejun Heo /* pick the winner of this round */ 5972a814908STejun Heo if (wbc->wb_bytes >= wbc->wb_lcand_bytes && 5982a814908STejun Heo wbc->wb_bytes >= wbc->wb_tcand_bytes) { 5992a814908STejun Heo max_id = wbc->wb_id; 6002a814908STejun Heo max_bytes = wbc->wb_bytes; 6012a814908STejun Heo } else if (wbc->wb_lcand_bytes >= wbc->wb_tcand_bytes) { 6022a814908STejun Heo max_id = wbc->wb_lcand_id; 6032a814908STejun Heo max_bytes = wbc->wb_lcand_bytes; 6042a814908STejun Heo } else { 6052a814908STejun Heo max_id = wbc->wb_tcand_id; 6062a814908STejun Heo max_bytes = wbc->wb_tcand_bytes; 6072a814908STejun Heo } 6082a814908STejun Heo 6092a814908STejun Heo /* 6102a814908STejun Heo * Calculate the amount of IO time the winner consumed and fold it 6112a814908STejun Heo * into the running average kept per inode. If the consumed IO 6122a814908STejun Heo * time is lower than avag / WB_FRN_TIME_CUT_DIV, ignore it for 6132a814908STejun Heo * deciding whether to switch or not. This is to prevent one-off 6142a814908STejun Heo * small dirtiers from skewing the verdict. 6152a814908STejun Heo */ 6162a814908STejun Heo max_time = DIV_ROUND_UP((max_bytes >> PAGE_SHIFT) << WB_FRN_TIME_SHIFT, 6172a814908STejun Heo wb->avg_write_bandwidth); 6182a814908STejun Heo if (avg_time) 6192a814908STejun Heo avg_time += (max_time >> WB_FRN_TIME_AVG_SHIFT) - 6202a814908STejun Heo (avg_time >> WB_FRN_TIME_AVG_SHIFT); 6212a814908STejun Heo else 6222a814908STejun Heo avg_time = max_time; /* immediate catch up on first run */ 6232a814908STejun Heo 6242a814908STejun Heo if (max_time >= avg_time / WB_FRN_TIME_CUT_DIV) { 6252a814908STejun Heo int slots; 6262a814908STejun Heo 6272a814908STejun Heo /* 6282a814908STejun Heo * The switch verdict is reached if foreign wb's consume 6292a814908STejun Heo * more than a certain proportion of IO time in a 6302a814908STejun Heo * WB_FRN_TIME_PERIOD. This is loosely tracked by 16 slot 6312a814908STejun Heo * history mask where each bit represents one sixteenth of 6322a814908STejun Heo * the period. Determine the number of slots to shift into 6332a814908STejun Heo * history from @max_time. 6342a814908STejun Heo */ 6352a814908STejun Heo slots = min(DIV_ROUND_UP(max_time, WB_FRN_HIST_UNIT), 6362a814908STejun Heo (unsigned long)WB_FRN_HIST_MAX_SLOTS); 6372a814908STejun Heo history <<= slots; 6382a814908STejun Heo if (wbc->wb_id != max_id) 6392a814908STejun Heo history |= (1U << slots) - 1; 6402a814908STejun Heo 6412a814908STejun Heo /* 6422a814908STejun Heo * Switch if the current wb isn't the consistent winner. 6432a814908STejun Heo * If there are multiple closely competing dirtiers, the 6442a814908STejun Heo * inode may switch across them repeatedly over time, which 6452a814908STejun Heo * is okay. The main goal is avoiding keeping an inode on 6462a814908STejun Heo * the wrong wb for an extended period of time. 6472a814908STejun Heo */ 648682aa8e1STejun Heo if (hweight32(history) > WB_FRN_HIST_THR_SLOTS) 649682aa8e1STejun Heo inode_switch_wbs(inode, max_id); 6502a814908STejun Heo } 6512a814908STejun Heo 6522a814908STejun Heo /* 6532a814908STejun Heo * Multiple instances of this function may race to update the 6542a814908STejun Heo * following fields but we don't mind occassional inaccuracies. 6552a814908STejun Heo */ 6562a814908STejun Heo inode->i_wb_frn_winner = max_id; 6572a814908STejun Heo inode->i_wb_frn_avg_time = min(avg_time, (unsigned long)U16_MAX); 6582a814908STejun Heo inode->i_wb_frn_history = history; 6592a814908STejun Heo 660b16b1debSTejun Heo wb_put(wbc->wb); 661b16b1debSTejun Heo wbc->wb = NULL; 662b16b1debSTejun Heo } 663b16b1debSTejun Heo 664b16b1debSTejun Heo /** 6652a814908STejun Heo * wbc_account_io - account IO issued during writeback 6662a814908STejun Heo * @wbc: writeback_control of the writeback in progress 6672a814908STejun Heo * @page: page being written out 6682a814908STejun Heo * @bytes: number of bytes being written out 6692a814908STejun Heo * 6702a814908STejun Heo * @bytes from @page are about to written out during the writeback 6712a814908STejun Heo * controlled by @wbc. Keep the book for foreign inode detection. See 6722a814908STejun Heo * wbc_detach_inode(). 6732a814908STejun Heo */ 6742a814908STejun Heo void wbc_account_io(struct writeback_control *wbc, struct page *page, 6752a814908STejun Heo size_t bytes) 6762a814908STejun Heo { 6772a814908STejun Heo int id; 6782a814908STejun Heo 6792a814908STejun Heo /* 6802a814908STejun Heo * pageout() path doesn't attach @wbc to the inode being written 6812a814908STejun Heo * out. This is intentional as we don't want the function to block 6822a814908STejun Heo * behind a slow cgroup. Ultimately, we want pageout() to kick off 6832a814908STejun Heo * regular writeback instead of writing things out itself. 6842a814908STejun Heo */ 6852a814908STejun Heo if (!wbc->wb) 6862a814908STejun Heo return; 6872a814908STejun Heo 6882a814908STejun Heo id = mem_cgroup_css_from_page(page)->id; 6892a814908STejun Heo 6902a814908STejun Heo if (id == wbc->wb_id) { 6912a814908STejun Heo wbc->wb_bytes += bytes; 6922a814908STejun Heo return; 6932a814908STejun Heo } 6942a814908STejun Heo 6952a814908STejun Heo if (id == wbc->wb_lcand_id) 6962a814908STejun Heo wbc->wb_lcand_bytes += bytes; 6972a814908STejun Heo 6982a814908STejun Heo /* Boyer-Moore majority vote algorithm */ 6992a814908STejun Heo if (!wbc->wb_tcand_bytes) 7002a814908STejun Heo wbc->wb_tcand_id = id; 7012a814908STejun Heo if (id == wbc->wb_tcand_id) 7022a814908STejun Heo wbc->wb_tcand_bytes += bytes; 7032a814908STejun Heo else 7042a814908STejun Heo wbc->wb_tcand_bytes -= min(bytes, wbc->wb_tcand_bytes); 7052a814908STejun Heo } 7065aa2a96bSTejun Heo EXPORT_SYMBOL_GPL(wbc_account_io); 7072a814908STejun Heo 7082a814908STejun Heo /** 709703c2708STejun Heo * inode_congested - test whether an inode is congested 71060292bccSTejun Heo * @inode: inode to test for congestion (may be NULL) 711703c2708STejun Heo * @cong_bits: mask of WB_[a]sync_congested bits to test 712703c2708STejun Heo * 713703c2708STejun Heo * Tests whether @inode is congested. @cong_bits is the mask of congestion 714703c2708STejun Heo * bits to test and the return value is the mask of set bits. 715703c2708STejun Heo * 716703c2708STejun Heo * If cgroup writeback is enabled for @inode, the congestion state is 717703c2708STejun Heo * determined by whether the cgwb (cgroup bdi_writeback) for the blkcg 718703c2708STejun Heo * associated with @inode is congested; otherwise, the root wb's congestion 719703c2708STejun Heo * state is used. 72060292bccSTejun Heo * 72160292bccSTejun Heo * @inode is allowed to be NULL as this function is often called on 72260292bccSTejun Heo * mapping->host which is NULL for the swapper space. 723703c2708STejun Heo */ 724703c2708STejun Heo int inode_congested(struct inode *inode, int cong_bits) 725703c2708STejun Heo { 7265cb8b824STejun Heo /* 7275cb8b824STejun Heo * Once set, ->i_wb never becomes NULL while the inode is alive. 7285cb8b824STejun Heo * Start transaction iff ->i_wb is visible. 7295cb8b824STejun Heo */ 730aaa2cacfSTejun Heo if (inode && inode_to_wb_is_valid(inode)) { 7315cb8b824STejun Heo struct bdi_writeback *wb; 7325cb8b824STejun Heo bool locked, congested; 7335cb8b824STejun Heo 7345cb8b824STejun Heo wb = unlocked_inode_to_wb_begin(inode, &locked); 7355cb8b824STejun Heo congested = wb_congested(wb, cong_bits); 7365cb8b824STejun Heo unlocked_inode_to_wb_end(inode, locked); 7375cb8b824STejun Heo return congested; 738703c2708STejun Heo } 739703c2708STejun Heo 740703c2708STejun Heo return wb_congested(&inode_to_bdi(inode)->wb, cong_bits); 741703c2708STejun Heo } 742703c2708STejun Heo EXPORT_SYMBOL_GPL(inode_congested); 743703c2708STejun Heo 744f2b65121STejun Heo /** 745f2b65121STejun Heo * wb_split_bdi_pages - split nr_pages to write according to bandwidth 746f2b65121STejun Heo * @wb: target bdi_writeback to split @nr_pages to 747f2b65121STejun Heo * @nr_pages: number of pages to write for the whole bdi 748f2b65121STejun Heo * 749f2b65121STejun Heo * Split @wb's portion of @nr_pages according to @wb's write bandwidth in 750f2b65121STejun Heo * relation to the total write bandwidth of all wb's w/ dirty inodes on 751f2b65121STejun Heo * @wb->bdi. 752f2b65121STejun Heo */ 753f2b65121STejun Heo static long wb_split_bdi_pages(struct bdi_writeback *wb, long nr_pages) 754f2b65121STejun Heo { 755f2b65121STejun Heo unsigned long this_bw = wb->avg_write_bandwidth; 756f2b65121STejun Heo unsigned long tot_bw = atomic_long_read(&wb->bdi->tot_write_bandwidth); 757f2b65121STejun Heo 758f2b65121STejun Heo if (nr_pages == LONG_MAX) 759f2b65121STejun Heo return LONG_MAX; 760f2b65121STejun Heo 761f2b65121STejun Heo /* 762f2b65121STejun Heo * This may be called on clean wb's and proportional distribution 763f2b65121STejun Heo * may not make sense, just use the original @nr_pages in those 764f2b65121STejun Heo * cases. In general, we wanna err on the side of writing more. 765f2b65121STejun Heo */ 766f2b65121STejun Heo if (!tot_bw || this_bw >= tot_bw) 767f2b65121STejun Heo return nr_pages; 768f2b65121STejun Heo else 769f2b65121STejun Heo return DIV_ROUND_UP_ULL((u64)nr_pages * this_bw, tot_bw); 770f2b65121STejun Heo } 771f2b65121STejun Heo 772db125360STejun Heo /** 773db125360STejun Heo * bdi_split_work_to_wbs - split a wb_writeback_work to all wb's of a bdi 774db125360STejun Heo * @bdi: target backing_dev_info 775db125360STejun Heo * @base_work: wb_writeback_work to issue 776db125360STejun Heo * @skip_if_busy: skip wb's which already have writeback in progress 777db125360STejun Heo * 778db125360STejun Heo * Split and issue @base_work to all wb's (bdi_writeback's) of @bdi which 779db125360STejun Heo * have dirty inodes. If @base_work->nr_page isn't %LONG_MAX, it's 780db125360STejun Heo * distributed to the busy wbs according to each wb's proportion in the 781db125360STejun Heo * total active write bandwidth of @bdi. 782db125360STejun Heo */ 783db125360STejun Heo static void bdi_split_work_to_wbs(struct backing_dev_info *bdi, 784db125360STejun Heo struct wb_writeback_work *base_work, 785db125360STejun Heo bool skip_if_busy) 786db125360STejun Heo { 787b817525aSTejun Heo struct bdi_writeback *last_wb = NULL; 788b33e18f6STejun Heo struct bdi_writeback *wb = list_entry(&bdi->wb_list, 789b817525aSTejun Heo struct bdi_writeback, bdi_node); 790db125360STejun Heo 791db125360STejun Heo might_sleep(); 792db125360STejun Heo restart: 793db125360STejun Heo rcu_read_lock(); 794b817525aSTejun Heo list_for_each_entry_continue_rcu(wb, &bdi->wb_list, bdi_node) { 7958a1270cdSTejun Heo DEFINE_WB_COMPLETION_ONSTACK(fallback_work_done); 7968a1270cdSTejun Heo struct wb_writeback_work fallback_work; 7978a1270cdSTejun Heo struct wb_writeback_work *work; 7988a1270cdSTejun Heo long nr_pages; 7998a1270cdSTejun Heo 800b817525aSTejun Heo if (last_wb) { 801b817525aSTejun Heo wb_put(last_wb); 802b817525aSTejun Heo last_wb = NULL; 803b817525aSTejun Heo } 804b817525aSTejun Heo 805006a0973STejun Heo /* SYNC_ALL writes out I_DIRTY_TIME too */ 806006a0973STejun Heo if (!wb_has_dirty_io(wb) && 807006a0973STejun Heo (base_work->sync_mode == WB_SYNC_NONE || 808006a0973STejun Heo list_empty(&wb->b_dirty_time))) 809006a0973STejun Heo continue; 810006a0973STejun Heo if (skip_if_busy && writeback_in_progress(wb)) 811db125360STejun Heo continue; 812db125360STejun Heo 8138a1270cdSTejun Heo nr_pages = wb_split_bdi_pages(wb, base_work->nr_pages); 8148a1270cdSTejun Heo 8158a1270cdSTejun Heo work = kmalloc(sizeof(*work), GFP_ATOMIC); 8168a1270cdSTejun Heo if (work) { 8178a1270cdSTejun Heo *work = *base_work; 8188a1270cdSTejun Heo work->nr_pages = nr_pages; 8198a1270cdSTejun Heo work->auto_free = 1; 8208a1270cdSTejun Heo wb_queue_work(wb, work); 8218a1270cdSTejun Heo continue; 822db125360STejun Heo } 8238a1270cdSTejun Heo 8248a1270cdSTejun Heo /* alloc failed, execute synchronously using on-stack fallback */ 8258a1270cdSTejun Heo work = &fallback_work; 8268a1270cdSTejun Heo *work = *base_work; 8278a1270cdSTejun Heo work->nr_pages = nr_pages; 8288a1270cdSTejun Heo work->auto_free = 0; 8298a1270cdSTejun Heo work->done = &fallback_work_done; 8308a1270cdSTejun Heo 8318a1270cdSTejun Heo wb_queue_work(wb, work); 8328a1270cdSTejun Heo 833b817525aSTejun Heo /* 834b817525aSTejun Heo * Pin @wb so that it stays on @bdi->wb_list. This allows 835b817525aSTejun Heo * continuing iteration from @wb after dropping and 836b817525aSTejun Heo * regrabbing rcu read lock. 837b817525aSTejun Heo */ 838b817525aSTejun Heo wb_get(wb); 839b817525aSTejun Heo last_wb = wb; 840b817525aSTejun Heo 841db125360STejun Heo rcu_read_unlock(); 8428a1270cdSTejun Heo wb_wait_for_completion(bdi, &fallback_work_done); 843db125360STejun Heo goto restart; 844db125360STejun Heo } 845db125360STejun Heo rcu_read_unlock(); 846b817525aSTejun Heo 847b817525aSTejun Heo if (last_wb) 848b817525aSTejun Heo wb_put(last_wb); 849db125360STejun Heo } 850db125360STejun Heo 851a1a0e23eSTejun Heo /** 852a1a0e23eSTejun Heo * cgroup_writeback_umount - flush inode wb switches for umount 853a1a0e23eSTejun Heo * 854a1a0e23eSTejun Heo * This function is called when a super_block is about to be destroyed and 855a1a0e23eSTejun Heo * flushes in-flight inode wb switches. An inode wb switch goes through 856a1a0e23eSTejun Heo * RCU and then workqueue, so the two need to be flushed in order to ensure 857a1a0e23eSTejun Heo * that all previously scheduled switches are finished. As wb switches are 858a1a0e23eSTejun Heo * rare occurrences and synchronize_rcu() can take a while, perform 859a1a0e23eSTejun Heo * flushing iff wb switches are in flight. 860a1a0e23eSTejun Heo */ 861a1a0e23eSTejun Heo void cgroup_writeback_umount(void) 862a1a0e23eSTejun Heo { 863a1a0e23eSTejun Heo if (atomic_read(&isw_nr_in_flight)) { 864a1a0e23eSTejun Heo synchronize_rcu(); 865a1a0e23eSTejun Heo flush_workqueue(isw_wq); 866a1a0e23eSTejun Heo } 867a1a0e23eSTejun Heo } 868a1a0e23eSTejun Heo 869a1a0e23eSTejun Heo static int __init cgroup_writeback_init(void) 870a1a0e23eSTejun Heo { 871a1a0e23eSTejun Heo isw_wq = alloc_workqueue("inode_switch_wbs", 0, 0); 872a1a0e23eSTejun Heo if (!isw_wq) 873a1a0e23eSTejun Heo return -ENOMEM; 874a1a0e23eSTejun Heo return 0; 875a1a0e23eSTejun Heo } 876a1a0e23eSTejun Heo fs_initcall(cgroup_writeback_init); 877a1a0e23eSTejun Heo 878f2b65121STejun Heo #else /* CONFIG_CGROUP_WRITEBACK */ 879f2b65121STejun Heo 88087e1d789STejun Heo static struct bdi_writeback * 88187e1d789STejun Heo locked_inode_to_wb_and_lock_list(struct inode *inode) 88287e1d789STejun Heo __releases(&inode->i_lock) 88387e1d789STejun Heo __acquires(&wb->list_lock) 88487e1d789STejun Heo { 88587e1d789STejun Heo struct bdi_writeback *wb = inode_to_wb(inode); 88687e1d789STejun Heo 88787e1d789STejun Heo spin_unlock(&inode->i_lock); 88887e1d789STejun Heo spin_lock(&wb->list_lock); 88987e1d789STejun Heo return wb; 89087e1d789STejun Heo } 89187e1d789STejun Heo 89287e1d789STejun Heo static struct bdi_writeback *inode_to_wb_and_lock_list(struct inode *inode) 89387e1d789STejun Heo __acquires(&wb->list_lock) 89487e1d789STejun Heo { 89587e1d789STejun Heo struct bdi_writeback *wb = inode_to_wb(inode); 89687e1d789STejun Heo 89787e1d789STejun Heo spin_lock(&wb->list_lock); 89887e1d789STejun Heo return wb; 89987e1d789STejun Heo } 90087e1d789STejun Heo 901f2b65121STejun Heo static long wb_split_bdi_pages(struct bdi_writeback *wb, long nr_pages) 902f2b65121STejun Heo { 903f2b65121STejun Heo return nr_pages; 904f2b65121STejun Heo } 905f2b65121STejun Heo 906db125360STejun Heo static void bdi_split_work_to_wbs(struct backing_dev_info *bdi, 907db125360STejun Heo struct wb_writeback_work *base_work, 908db125360STejun Heo bool skip_if_busy) 909db125360STejun Heo { 910db125360STejun Heo might_sleep(); 911db125360STejun Heo 912006a0973STejun Heo if (!skip_if_busy || !writeback_in_progress(&bdi->wb)) { 913db125360STejun Heo base_work->auto_free = 0; 914db125360STejun Heo wb_queue_work(&bdi->wb, base_work); 915db125360STejun Heo } 916db125360STejun Heo } 917db125360STejun Heo 918703c2708STejun Heo #endif /* CONFIG_CGROUP_WRITEBACK */ 919703c2708STejun Heo 920c00ddad3STejun Heo void wb_start_writeback(struct bdi_writeback *wb, long nr_pages, 921c00ddad3STejun Heo bool range_cyclic, enum wb_reason reason) 922b6e51316SJens Axboe { 923c00ddad3STejun Heo struct wb_writeback_work *work; 924c00ddad3STejun Heo 925c00ddad3STejun Heo if (!wb_has_dirty_io(wb)) 926c00ddad3STejun Heo return; 927c00ddad3STejun Heo 928c00ddad3STejun Heo /* 929c00ddad3STejun Heo * This is WB_SYNC_NONE writeback, so if allocation fails just 930c00ddad3STejun Heo * wakeup the thread for old dirty data writeback 931c00ddad3STejun Heo */ 932c00ddad3STejun Heo work = kzalloc(sizeof(*work), GFP_ATOMIC); 933c00ddad3STejun Heo if (!work) { 9345634cc2aSTejun Heo trace_writeback_nowork(wb); 935c00ddad3STejun Heo wb_wakeup(wb); 936c00ddad3STejun Heo return; 937c00ddad3STejun Heo } 938c00ddad3STejun Heo 939c00ddad3STejun Heo work->sync_mode = WB_SYNC_NONE; 940c00ddad3STejun Heo work->nr_pages = nr_pages; 941c00ddad3STejun Heo work->range_cyclic = range_cyclic; 942c00ddad3STejun Heo work->reason = reason; 943ac7b19a3STejun Heo work->auto_free = 1; 944c00ddad3STejun Heo 945c00ddad3STejun Heo wb_queue_work(wb, work); 946d3ddec76SWu Fengguang } 947d3ddec76SWu Fengguang 948c5444198SChristoph Hellwig /** 9499ecf4866STejun Heo * wb_start_background_writeback - start background writeback 9509ecf4866STejun Heo * @wb: bdi_writback to write from 951c5444198SChristoph Hellwig * 952c5444198SChristoph Hellwig * Description: 9536585027aSJan Kara * This makes sure WB_SYNC_NONE background writeback happens. When 9549ecf4866STejun Heo * this function returns, it is only guaranteed that for given wb 9556585027aSJan Kara * some IO is happening if we are over background dirty threshold. 9566585027aSJan Kara * Caller need not hold sb s_umount semaphore. 957c5444198SChristoph Hellwig */ 9589ecf4866STejun Heo void wb_start_background_writeback(struct bdi_writeback *wb) 959c5444198SChristoph Hellwig { 9606585027aSJan Kara /* 9616585027aSJan Kara * We just wake up the flusher thread. It will perform background 9626585027aSJan Kara * writeback as soon as there is no other work to do. 9636585027aSJan Kara */ 9645634cc2aSTejun Heo trace_writeback_wake_background(wb); 9659ecf4866STejun Heo wb_wakeup(wb); 9661da177e4SLinus Torvalds } 9671da177e4SLinus Torvalds 9681da177e4SLinus Torvalds /* 969a66979abSDave Chinner * Remove the inode from the writeback list it is on. 970a66979abSDave Chinner */ 971c7f54084SDave Chinner void inode_io_list_del(struct inode *inode) 972a66979abSDave Chinner { 97387e1d789STejun Heo struct bdi_writeback *wb; 974a66979abSDave Chinner 97587e1d789STejun Heo wb = inode_to_wb_and_lock_list(inode); 976c7f54084SDave Chinner inode_io_list_del_locked(inode, wb); 97752ebea74STejun Heo spin_unlock(&wb->list_lock); 978f758eeabSChristoph Hellwig } 979a66979abSDave Chinner 980a66979abSDave Chinner /* 9816610a0bcSAndrew Morton * Redirty an inode: set its when-it-was dirtied timestamp and move it to the 9826610a0bcSAndrew Morton * furthest end of its superblock's dirty-inode list. 9836610a0bcSAndrew Morton * 9846610a0bcSAndrew Morton * Before stamping the inode's ->dirtied_when, we check to see whether it is 98566f3b8e2SJens Axboe * already the most-recently-dirtied inode on the b_dirty list. If that is 9866610a0bcSAndrew Morton * the case then the inode must have been redirtied while it was being written 9876610a0bcSAndrew Morton * out and we don't reset its dirtied_when. 9886610a0bcSAndrew Morton */ 989f758eeabSChristoph Hellwig static void redirty_tail(struct inode *inode, struct bdi_writeback *wb) 9906610a0bcSAndrew Morton { 99103ba3782SJens Axboe if (!list_empty(&wb->b_dirty)) { 99266f3b8e2SJens Axboe struct inode *tail; 9936610a0bcSAndrew Morton 9947ccf19a8SNick Piggin tail = wb_inode(wb->b_dirty.next); 99566f3b8e2SJens Axboe if (time_before(inode->dirtied_when, tail->dirtied_when)) 9966610a0bcSAndrew Morton inode->dirtied_when = jiffies; 9976610a0bcSAndrew Morton } 998c7f54084SDave Chinner inode_io_list_move_locked(inode, wb, &wb->b_dirty); 9996610a0bcSAndrew Morton } 10006610a0bcSAndrew Morton 10016610a0bcSAndrew Morton /* 100266f3b8e2SJens Axboe * requeue inode for re-scanning after bdi->b_io list is exhausted. 1003c986d1e2SAndrew Morton */ 1004f758eeabSChristoph Hellwig static void requeue_io(struct inode *inode, struct bdi_writeback *wb) 1005c986d1e2SAndrew Morton { 1006c7f54084SDave Chinner inode_io_list_move_locked(inode, wb, &wb->b_more_io); 1007c986d1e2SAndrew Morton } 1008c986d1e2SAndrew Morton 10091c0eeaf5SJoern Engel static void inode_sync_complete(struct inode *inode) 10101c0eeaf5SJoern Engel { 1011365b94aeSJan Kara inode->i_state &= ~I_SYNC; 10124eff96ddSJan Kara /* If inode is clean an unused, put it into LRU now... */ 10134eff96ddSJan Kara inode_add_lru(inode); 1014365b94aeSJan Kara /* Waiters must see I_SYNC cleared before being woken up */ 10151c0eeaf5SJoern Engel smp_mb(); 10161c0eeaf5SJoern Engel wake_up_bit(&inode->i_state, __I_SYNC); 10171c0eeaf5SJoern Engel } 10181c0eeaf5SJoern Engel 1019d2caa3c5SJeff Layton static bool inode_dirtied_after(struct inode *inode, unsigned long t) 1020d2caa3c5SJeff Layton { 1021d2caa3c5SJeff Layton bool ret = time_after(inode->dirtied_when, t); 1022d2caa3c5SJeff Layton #ifndef CONFIG_64BIT 1023d2caa3c5SJeff Layton /* 1024d2caa3c5SJeff Layton * For inodes being constantly redirtied, dirtied_when can get stuck. 1025d2caa3c5SJeff Layton * It _appears_ to be in the future, but is actually in distant past. 1026d2caa3c5SJeff Layton * This test is necessary to prevent such wrapped-around relative times 10275b0830cbSJens Axboe * from permanently stopping the whole bdi writeback. 1028d2caa3c5SJeff Layton */ 1029d2caa3c5SJeff Layton ret = ret && time_before_eq(inode->dirtied_when, jiffies); 1030d2caa3c5SJeff Layton #endif 1031d2caa3c5SJeff Layton return ret; 1032d2caa3c5SJeff Layton } 1033d2caa3c5SJeff Layton 10340ae45f63STheodore Ts'o #define EXPIRE_DIRTY_ATIME 0x0001 10350ae45f63STheodore Ts'o 1036c986d1e2SAndrew Morton /* 10370e2f2b23SWang Sheng-Hui * Move expired (dirtied before work->older_than_this) dirty inodes from 1038697e6fedSJan Kara * @delaying_queue to @dispatch_queue. 10392c136579SFengguang Wu */ 1040e84d0a4fSWu Fengguang static int move_expired_inodes(struct list_head *delaying_queue, 10412c136579SFengguang Wu struct list_head *dispatch_queue, 10420ae45f63STheodore Ts'o int flags, 1043ad4e38ddSCurt Wohlgemuth struct wb_writeback_work *work) 10442c136579SFengguang Wu { 10450ae45f63STheodore Ts'o unsigned long *older_than_this = NULL; 10460ae45f63STheodore Ts'o unsigned long expire_time; 10475c03449dSShaohua Li LIST_HEAD(tmp); 10485c03449dSShaohua Li struct list_head *pos, *node; 1049cf137307SJens Axboe struct super_block *sb = NULL; 10505c03449dSShaohua Li struct inode *inode; 1051cf137307SJens Axboe int do_sb_sort = 0; 1052e84d0a4fSWu Fengguang int moved = 0; 10535c03449dSShaohua Li 10540ae45f63STheodore Ts'o if ((flags & EXPIRE_DIRTY_ATIME) == 0) 10550ae45f63STheodore Ts'o older_than_this = work->older_than_this; 1056a2f48706STheodore Ts'o else if (!work->for_sync) { 1057a2f48706STheodore Ts'o expire_time = jiffies - (dirtytime_expire_interval * HZ); 10580ae45f63STheodore Ts'o older_than_this = &expire_time; 10590ae45f63STheodore Ts'o } 10602c136579SFengguang Wu while (!list_empty(delaying_queue)) { 10617ccf19a8SNick Piggin inode = wb_inode(delaying_queue->prev); 10620ae45f63STheodore Ts'o if (older_than_this && 10630ae45f63STheodore Ts'o inode_dirtied_after(inode, *older_than_this)) 10642c136579SFengguang Wu break; 1065c7f54084SDave Chinner list_move(&inode->i_io_list, &tmp); 1066a8855990SJan Kara moved++; 10670ae45f63STheodore Ts'o if (flags & EXPIRE_DIRTY_ATIME) 10680ae45f63STheodore Ts'o set_bit(__I_DIRTY_TIME_EXPIRED, &inode->i_state); 1069a8855990SJan Kara if (sb_is_blkdev_sb(inode->i_sb)) 1070a8855990SJan Kara continue; 1071cf137307SJens Axboe if (sb && sb != inode->i_sb) 1072cf137307SJens Axboe do_sb_sort = 1; 1073cf137307SJens Axboe sb = inode->i_sb; 10745c03449dSShaohua Li } 10755c03449dSShaohua Li 1076cf137307SJens Axboe /* just one sb in list, splice to dispatch_queue and we're done */ 1077cf137307SJens Axboe if (!do_sb_sort) { 1078cf137307SJens Axboe list_splice(&tmp, dispatch_queue); 1079e84d0a4fSWu Fengguang goto out; 1080cf137307SJens Axboe } 1081cf137307SJens Axboe 10825c03449dSShaohua Li /* Move inodes from one superblock together */ 10835c03449dSShaohua Li while (!list_empty(&tmp)) { 10847ccf19a8SNick Piggin sb = wb_inode(tmp.prev)->i_sb; 10855c03449dSShaohua Li list_for_each_prev_safe(pos, node, &tmp) { 10867ccf19a8SNick Piggin inode = wb_inode(pos); 10875c03449dSShaohua Li if (inode->i_sb == sb) 1088c7f54084SDave Chinner list_move(&inode->i_io_list, dispatch_queue); 10892c136579SFengguang Wu } 10902c136579SFengguang Wu } 1091e84d0a4fSWu Fengguang out: 1092e84d0a4fSWu Fengguang return moved; 10935c03449dSShaohua Li } 10942c136579SFengguang Wu 10952c136579SFengguang Wu /* 10962c136579SFengguang Wu * Queue all expired dirty inodes for io, eldest first. 10974ea879b9SWu Fengguang * Before 10984ea879b9SWu Fengguang * newly dirtied b_dirty b_io b_more_io 10994ea879b9SWu Fengguang * =============> gf edc BA 11004ea879b9SWu Fengguang * After 11014ea879b9SWu Fengguang * newly dirtied b_dirty b_io b_more_io 11024ea879b9SWu Fengguang * =============> g fBAedc 11034ea879b9SWu Fengguang * | 11044ea879b9SWu Fengguang * +--> dequeue for IO 11052c136579SFengguang Wu */ 1106ad4e38ddSCurt Wohlgemuth static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work) 11072c136579SFengguang Wu { 1108e84d0a4fSWu Fengguang int moved; 11090ae45f63STheodore Ts'o 1110f758eeabSChristoph Hellwig assert_spin_locked(&wb->list_lock); 11114ea879b9SWu Fengguang list_splice_init(&wb->b_more_io, &wb->b_io); 11120ae45f63STheodore Ts'o moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, 0, work); 11130ae45f63STheodore Ts'o moved += move_expired_inodes(&wb->b_dirty_time, &wb->b_io, 11140ae45f63STheodore Ts'o EXPIRE_DIRTY_ATIME, work); 1115d6c10f1fSTejun Heo if (moved) 1116d6c10f1fSTejun Heo wb_io_lists_populated(wb); 1117ad4e38ddSCurt Wohlgemuth trace_writeback_queue_io(wb, work, moved); 111866f3b8e2SJens Axboe } 111966f3b8e2SJens Axboe 1120a9185b41SChristoph Hellwig static int write_inode(struct inode *inode, struct writeback_control *wbc) 112166f3b8e2SJens Axboe { 11229fb0a7daSTejun Heo int ret; 11239fb0a7daSTejun Heo 11249fb0a7daSTejun Heo if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) { 11259fb0a7daSTejun Heo trace_writeback_write_inode_start(inode, wbc); 11269fb0a7daSTejun Heo ret = inode->i_sb->s_op->write_inode(inode, wbc); 11279fb0a7daSTejun Heo trace_writeback_write_inode(inode, wbc); 11289fb0a7daSTejun Heo return ret; 11299fb0a7daSTejun Heo } 113003ba3782SJens Axboe return 0; 113166f3b8e2SJens Axboe } 113208d8e974SFengguang Wu 11332c136579SFengguang Wu /* 1134169ebd90SJan Kara * Wait for writeback on an inode to complete. Called with i_lock held. 1135169ebd90SJan Kara * Caller must make sure inode cannot go away when we drop i_lock. 113601c03194SChristoph Hellwig */ 1137169ebd90SJan Kara static void __inode_wait_for_writeback(struct inode *inode) 1138169ebd90SJan Kara __releases(inode->i_lock) 1139169ebd90SJan Kara __acquires(inode->i_lock) 114001c03194SChristoph Hellwig { 114101c03194SChristoph Hellwig DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); 114201c03194SChristoph Hellwig wait_queue_head_t *wqh; 114301c03194SChristoph Hellwig 114401c03194SChristoph Hellwig wqh = bit_waitqueue(&inode->i_state, __I_SYNC); 114558a9d3d8SRichard Kennedy while (inode->i_state & I_SYNC) { 1146250df6edSDave Chinner spin_unlock(&inode->i_lock); 114774316201SNeilBrown __wait_on_bit(wqh, &wq, bit_wait, 114874316201SNeilBrown TASK_UNINTERRUPTIBLE); 1149250df6edSDave Chinner spin_lock(&inode->i_lock); 115058a9d3d8SRichard Kennedy } 115101c03194SChristoph Hellwig } 115201c03194SChristoph Hellwig 115301c03194SChristoph Hellwig /* 1154169ebd90SJan Kara * Wait for writeback on an inode to complete. Caller must have inode pinned. 1155169ebd90SJan Kara */ 1156169ebd90SJan Kara void inode_wait_for_writeback(struct inode *inode) 1157169ebd90SJan Kara { 1158169ebd90SJan Kara spin_lock(&inode->i_lock); 1159169ebd90SJan Kara __inode_wait_for_writeback(inode); 1160169ebd90SJan Kara spin_unlock(&inode->i_lock); 1161169ebd90SJan Kara } 1162169ebd90SJan Kara 1163169ebd90SJan Kara /* 1164169ebd90SJan Kara * Sleep until I_SYNC is cleared. This function must be called with i_lock 1165169ebd90SJan Kara * held and drops it. It is aimed for callers not holding any inode reference 1166169ebd90SJan Kara * so once i_lock is dropped, inode can go away. 1167169ebd90SJan Kara */ 1168169ebd90SJan Kara static void inode_sleep_on_writeback(struct inode *inode) 1169169ebd90SJan Kara __releases(inode->i_lock) 1170169ebd90SJan Kara { 1171169ebd90SJan Kara DEFINE_WAIT(wait); 1172169ebd90SJan Kara wait_queue_head_t *wqh = bit_waitqueue(&inode->i_state, __I_SYNC); 1173169ebd90SJan Kara int sleep; 1174169ebd90SJan Kara 1175169ebd90SJan Kara prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); 1176169ebd90SJan Kara sleep = inode->i_state & I_SYNC; 1177169ebd90SJan Kara spin_unlock(&inode->i_lock); 1178169ebd90SJan Kara if (sleep) 1179169ebd90SJan Kara schedule(); 1180169ebd90SJan Kara finish_wait(wqh, &wait); 1181169ebd90SJan Kara } 1182169ebd90SJan Kara 1183169ebd90SJan Kara /* 1184ccb26b5aSJan Kara * Find proper writeback list for the inode depending on its current state and 1185ccb26b5aSJan Kara * possibly also change of its state while we were doing writeback. Here we 1186ccb26b5aSJan Kara * handle things such as livelock prevention or fairness of writeback among 1187ccb26b5aSJan Kara * inodes. This function can be called only by flusher thread - noone else 1188ccb26b5aSJan Kara * processes all inodes in writeback lists and requeueing inodes behind flusher 1189ccb26b5aSJan Kara * thread's back can have unexpected consequences. 1190ccb26b5aSJan Kara */ 1191ccb26b5aSJan Kara static void requeue_inode(struct inode *inode, struct bdi_writeback *wb, 1192ccb26b5aSJan Kara struct writeback_control *wbc) 1193ccb26b5aSJan Kara { 1194ccb26b5aSJan Kara if (inode->i_state & I_FREEING) 1195ccb26b5aSJan Kara return; 1196ccb26b5aSJan Kara 1197ccb26b5aSJan Kara /* 1198ccb26b5aSJan Kara * Sync livelock prevention. Each inode is tagged and synced in one 1199ccb26b5aSJan Kara * shot. If still dirty, it will be redirty_tail()'ed below. Update 1200ccb26b5aSJan Kara * the dirty time to prevent enqueue and sync it again. 1201ccb26b5aSJan Kara */ 1202ccb26b5aSJan Kara if ((inode->i_state & I_DIRTY) && 1203ccb26b5aSJan Kara (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)) 1204ccb26b5aSJan Kara inode->dirtied_when = jiffies; 1205ccb26b5aSJan Kara 12064f8ad655SJan Kara if (wbc->pages_skipped) { 12074f8ad655SJan Kara /* 12084f8ad655SJan Kara * writeback is not making progress due to locked 12094f8ad655SJan Kara * buffers. Skip this inode for now. 12104f8ad655SJan Kara */ 12114f8ad655SJan Kara redirty_tail(inode, wb); 12124f8ad655SJan Kara return; 12134f8ad655SJan Kara } 12144f8ad655SJan Kara 1215ccb26b5aSJan Kara if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) { 1216ccb26b5aSJan Kara /* 1217ccb26b5aSJan Kara * We didn't write back all the pages. nfs_writepages() 1218ccb26b5aSJan Kara * sometimes bales out without doing anything. 1219ccb26b5aSJan Kara */ 1220ccb26b5aSJan Kara if (wbc->nr_to_write <= 0) { 1221ccb26b5aSJan Kara /* Slice used up. Queue for next turn. */ 1222ccb26b5aSJan Kara requeue_io(inode, wb); 1223ccb26b5aSJan Kara } else { 1224ccb26b5aSJan Kara /* 1225ccb26b5aSJan Kara * Writeback blocked by something other than 1226ccb26b5aSJan Kara * congestion. Delay the inode for some time to 1227ccb26b5aSJan Kara * avoid spinning on the CPU (100% iowait) 1228ccb26b5aSJan Kara * retrying writeback of the dirty page/inode 1229ccb26b5aSJan Kara * that cannot be performed immediately. 1230ccb26b5aSJan Kara */ 1231ccb26b5aSJan Kara redirty_tail(inode, wb); 1232ccb26b5aSJan Kara } 1233ccb26b5aSJan Kara } else if (inode->i_state & I_DIRTY) { 1234ccb26b5aSJan Kara /* 1235ccb26b5aSJan Kara * Filesystems can dirty the inode during writeback operations, 1236ccb26b5aSJan Kara * such as delayed allocation during submission or metadata 1237ccb26b5aSJan Kara * updates after data IO completion. 1238ccb26b5aSJan Kara */ 1239ccb26b5aSJan Kara redirty_tail(inode, wb); 12400ae45f63STheodore Ts'o } else if (inode->i_state & I_DIRTY_TIME) { 1241a2f48706STheodore Ts'o inode->dirtied_when = jiffies; 1242c7f54084SDave Chinner inode_io_list_move_locked(inode, wb, &wb->b_dirty_time); 1243ccb26b5aSJan Kara } else { 1244ccb26b5aSJan Kara /* The inode is clean. Remove from writeback lists. */ 1245c7f54084SDave Chinner inode_io_list_del_locked(inode, wb); 1246ccb26b5aSJan Kara } 1247ccb26b5aSJan Kara } 1248ccb26b5aSJan Kara 1249ccb26b5aSJan Kara /* 12504f8ad655SJan Kara * Write out an inode and its dirty pages. Do not update the writeback list 12514f8ad655SJan Kara * linkage. That is left to the caller. The caller is also responsible for 12524f8ad655SJan Kara * setting I_SYNC flag and calling inode_sync_complete() to clear it. 12531da177e4SLinus Torvalds */ 12541da177e4SLinus Torvalds static int 1255cd8ed2a4SYan Hong __writeback_single_inode(struct inode *inode, struct writeback_control *wbc) 12561da177e4SLinus Torvalds { 12571da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 1258251d6a47SWu Fengguang long nr_to_write = wbc->nr_to_write; 125901c03194SChristoph Hellwig unsigned dirty; 12601da177e4SLinus Torvalds int ret; 12611da177e4SLinus Torvalds 12624f8ad655SJan Kara WARN_ON(!(inode->i_state & I_SYNC)); 12631da177e4SLinus Torvalds 12649fb0a7daSTejun Heo trace_writeback_single_inode_start(inode, wbc, nr_to_write); 12659fb0a7daSTejun Heo 12661da177e4SLinus Torvalds ret = do_writepages(mapping, wbc); 12671da177e4SLinus Torvalds 126826821ed4SChristoph Hellwig /* 126926821ed4SChristoph Hellwig * Make sure to wait on the data before writing out the metadata. 127026821ed4SChristoph Hellwig * This is important for filesystems that modify metadata on data 12717747bd4bSDave Chinner * I/O completion. We don't do it for sync(2) writeback because it has a 12727747bd4bSDave Chinner * separate, external IO completion path and ->sync_fs for guaranteeing 12737747bd4bSDave Chinner * inode metadata is written back correctly. 127426821ed4SChristoph Hellwig */ 12757747bd4bSDave Chinner if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) { 127626821ed4SChristoph Hellwig int err = filemap_fdatawait(mapping); 12771da177e4SLinus Torvalds if (ret == 0) 12781da177e4SLinus Torvalds ret = err; 12791da177e4SLinus Torvalds } 12801da177e4SLinus Torvalds 12815547e8aaSDmitry Monakhov /* 12825547e8aaSDmitry Monakhov * Some filesystems may redirty the inode during the writeback 12835547e8aaSDmitry Monakhov * due to delalloc, clear dirty metadata flags right before 12845547e8aaSDmitry Monakhov * write_inode() 12855547e8aaSDmitry Monakhov */ 1286250df6edSDave Chinner spin_lock(&inode->i_lock); 12879c6ac78eSTejun Heo 12885547e8aaSDmitry Monakhov dirty = inode->i_state & I_DIRTY; 1289a2f48706STheodore Ts'o if (inode->i_state & I_DIRTY_TIME) { 1290a2f48706STheodore Ts'o if ((dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) || 1291a2f48706STheodore Ts'o unlikely(inode->i_state & I_DIRTY_TIME_EXPIRED) || 1292a2f48706STheodore Ts'o unlikely(time_after(jiffies, 1293a2f48706STheodore Ts'o (inode->dirtied_time_when + 1294a2f48706STheodore Ts'o dirtytime_expire_interval * HZ)))) { 12950ae45f63STheodore Ts'o dirty |= I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED; 12960ae45f63STheodore Ts'o trace_writeback_lazytime(inode); 12970ae45f63STheodore Ts'o } 1298a2f48706STheodore Ts'o } else 1299a2f48706STheodore Ts'o inode->i_state &= ~I_DIRTY_TIME_EXPIRED; 13000ae45f63STheodore Ts'o inode->i_state &= ~dirty; 13019c6ac78eSTejun Heo 13029c6ac78eSTejun Heo /* 13039c6ac78eSTejun Heo * Paired with smp_mb() in __mark_inode_dirty(). This allows 13049c6ac78eSTejun Heo * __mark_inode_dirty() to test i_state without grabbing i_lock - 13059c6ac78eSTejun Heo * either they see the I_DIRTY bits cleared or we see the dirtied 13069c6ac78eSTejun Heo * inode. 13079c6ac78eSTejun Heo * 13089c6ac78eSTejun Heo * I_DIRTY_PAGES is always cleared together above even if @mapping 13099c6ac78eSTejun Heo * still has dirty pages. The flag is reinstated after smp_mb() if 13109c6ac78eSTejun Heo * necessary. This guarantees that either __mark_inode_dirty() 13119c6ac78eSTejun Heo * sees clear I_DIRTY_PAGES or we see PAGECACHE_TAG_DIRTY. 13129c6ac78eSTejun Heo */ 13139c6ac78eSTejun Heo smp_mb(); 13149c6ac78eSTejun Heo 13159c6ac78eSTejun Heo if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) 13169c6ac78eSTejun Heo inode->i_state |= I_DIRTY_PAGES; 13179c6ac78eSTejun Heo 1318250df6edSDave Chinner spin_unlock(&inode->i_lock); 13199c6ac78eSTejun Heo 13200ae45f63STheodore Ts'o if (dirty & I_DIRTY_TIME) 13210ae45f63STheodore Ts'o mark_inode_dirty_sync(inode); 132226821ed4SChristoph Hellwig /* Don't write the inode if only I_DIRTY_PAGES was set */ 13230ae45f63STheodore Ts'o if (dirty & ~I_DIRTY_PAGES) { 1324a9185b41SChristoph Hellwig int err = write_inode(inode, wbc); 13251da177e4SLinus Torvalds if (ret == 0) 13261da177e4SLinus Torvalds ret = err; 13271da177e4SLinus Torvalds } 13284f8ad655SJan Kara trace_writeback_single_inode(inode, wbc, nr_to_write); 13294f8ad655SJan Kara return ret; 13304f8ad655SJan Kara } 13314f8ad655SJan Kara 13324f8ad655SJan Kara /* 13334f8ad655SJan Kara * Write out an inode's dirty pages. Either the caller has an active reference 13344f8ad655SJan Kara * on the inode or the inode has I_WILL_FREE set. 13354f8ad655SJan Kara * 13364f8ad655SJan Kara * This function is designed to be called for writing back one inode which 13374f8ad655SJan Kara * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode() 13384f8ad655SJan Kara * and does more profound writeback list handling in writeback_sb_inodes(). 13394f8ad655SJan Kara */ 13404f8ad655SJan Kara static int 13414f8ad655SJan Kara writeback_single_inode(struct inode *inode, struct bdi_writeback *wb, 13424f8ad655SJan Kara struct writeback_control *wbc) 13434f8ad655SJan Kara { 13444f8ad655SJan Kara int ret = 0; 13454f8ad655SJan Kara 13464f8ad655SJan Kara spin_lock(&inode->i_lock); 13474f8ad655SJan Kara if (!atomic_read(&inode->i_count)) 13484f8ad655SJan Kara WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); 13494f8ad655SJan Kara else 13504f8ad655SJan Kara WARN_ON(inode->i_state & I_WILL_FREE); 13514f8ad655SJan Kara 13524f8ad655SJan Kara if (inode->i_state & I_SYNC) { 13534f8ad655SJan Kara if (wbc->sync_mode != WB_SYNC_ALL) 13544f8ad655SJan Kara goto out; 13554f8ad655SJan Kara /* 1356169ebd90SJan Kara * It's a data-integrity sync. We must wait. Since callers hold 1357169ebd90SJan Kara * inode reference or inode has I_WILL_FREE set, it cannot go 1358169ebd90SJan Kara * away under us. 13594f8ad655SJan Kara */ 1360169ebd90SJan Kara __inode_wait_for_writeback(inode); 13614f8ad655SJan Kara } 13624f8ad655SJan Kara WARN_ON(inode->i_state & I_SYNC); 13634f8ad655SJan Kara /* 1364f9b0e058SJan Kara * Skip inode if it is clean and we have no outstanding writeback in 1365f9b0e058SJan Kara * WB_SYNC_ALL mode. We don't want to mess with writeback lists in this 1366f9b0e058SJan Kara * function since flusher thread may be doing for example sync in 1367f9b0e058SJan Kara * parallel and if we move the inode, it could get skipped. So here we 1368f9b0e058SJan Kara * make sure inode is on some writeback list and leave it there unless 1369f9b0e058SJan Kara * we have completely cleaned the inode. 13704f8ad655SJan Kara */ 13710ae45f63STheodore Ts'o if (!(inode->i_state & I_DIRTY_ALL) && 1372f9b0e058SJan Kara (wbc->sync_mode != WB_SYNC_ALL || 1373f9b0e058SJan Kara !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK))) 13744f8ad655SJan Kara goto out; 13754f8ad655SJan Kara inode->i_state |= I_SYNC; 1376b16b1debSTejun Heo wbc_attach_and_unlock_inode(wbc, inode); 13774f8ad655SJan Kara 1378cd8ed2a4SYan Hong ret = __writeback_single_inode(inode, wbc); 13791da177e4SLinus Torvalds 1380b16b1debSTejun Heo wbc_detach_inode(wbc); 1381f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 1382250df6edSDave Chinner spin_lock(&inode->i_lock); 13834f8ad655SJan Kara /* 13844f8ad655SJan Kara * If inode is clean, remove it from writeback lists. Otherwise don't 13854f8ad655SJan Kara * touch it. See comment above for explanation. 13864f8ad655SJan Kara */ 13870ae45f63STheodore Ts'o if (!(inode->i_state & I_DIRTY_ALL)) 1388c7f54084SDave Chinner inode_io_list_del_locked(inode, wb); 13894f8ad655SJan Kara spin_unlock(&wb->list_lock); 13901c0eeaf5SJoern Engel inode_sync_complete(inode); 13914f8ad655SJan Kara out: 13924f8ad655SJan Kara spin_unlock(&inode->i_lock); 13931da177e4SLinus Torvalds return ret; 13941da177e4SLinus Torvalds } 13951da177e4SLinus Torvalds 1396a88a341aSTejun Heo static long writeback_chunk_size(struct bdi_writeback *wb, 13971a12d8bdSWu Fengguang struct wb_writeback_work *work) 1398d46db3d5SWu Fengguang { 1399d46db3d5SWu Fengguang long pages; 1400d46db3d5SWu Fengguang 1401d46db3d5SWu Fengguang /* 1402d46db3d5SWu Fengguang * WB_SYNC_ALL mode does livelock avoidance by syncing dirty 1403d46db3d5SWu Fengguang * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX 1404d46db3d5SWu Fengguang * here avoids calling into writeback_inodes_wb() more than once. 1405d46db3d5SWu Fengguang * 1406d46db3d5SWu Fengguang * The intended call sequence for WB_SYNC_ALL writeback is: 1407d46db3d5SWu Fengguang * 1408d46db3d5SWu Fengguang * wb_writeback() 1409d46db3d5SWu Fengguang * writeback_sb_inodes() <== called only once 1410d46db3d5SWu Fengguang * write_cache_pages() <== called once for each inode 1411d46db3d5SWu Fengguang * (quickly) tag currently dirty pages 1412d46db3d5SWu Fengguang * (maybe slowly) sync all tagged pages 1413d46db3d5SWu Fengguang */ 1414d46db3d5SWu Fengguang if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages) 1415d46db3d5SWu Fengguang pages = LONG_MAX; 14161a12d8bdSWu Fengguang else { 1417a88a341aSTejun Heo pages = min(wb->avg_write_bandwidth / 2, 1418dcc25ae7STejun Heo global_wb_domain.dirty_limit / DIRTY_SCOPE); 14191a12d8bdSWu Fengguang pages = min(pages, work->nr_pages); 14201a12d8bdSWu Fengguang pages = round_down(pages + MIN_WRITEBACK_PAGES, 14211a12d8bdSWu Fengguang MIN_WRITEBACK_PAGES); 14221a12d8bdSWu Fengguang } 1423d46db3d5SWu Fengguang 1424d46db3d5SWu Fengguang return pages; 1425d46db3d5SWu Fengguang } 1426d46db3d5SWu Fengguang 142703ba3782SJens Axboe /* 1428f11c9c5cSEdward Shishkin * Write a portion of b_io inodes which belong to @sb. 1429edadfb10SChristoph Hellwig * 1430d46db3d5SWu Fengguang * Return the number of pages and/or inodes written. 14310ba13fd1SLinus Torvalds * 14320ba13fd1SLinus Torvalds * NOTE! This is called with wb->list_lock held, and will 14330ba13fd1SLinus Torvalds * unlock and relock that for each inode it ends up doing 14340ba13fd1SLinus Torvalds * IO for. 1435f11c9c5cSEdward Shishkin */ 1436d46db3d5SWu Fengguang static long writeback_sb_inodes(struct super_block *sb, 1437d46db3d5SWu Fengguang struct bdi_writeback *wb, 1438d46db3d5SWu Fengguang struct wb_writeback_work *work) 143903ba3782SJens Axboe { 1440d46db3d5SWu Fengguang struct writeback_control wbc = { 1441d46db3d5SWu Fengguang .sync_mode = work->sync_mode, 1442d46db3d5SWu Fengguang .tagged_writepages = work->tagged_writepages, 1443d46db3d5SWu Fengguang .for_kupdate = work->for_kupdate, 1444d46db3d5SWu Fengguang .for_background = work->for_background, 14457747bd4bSDave Chinner .for_sync = work->for_sync, 1446d46db3d5SWu Fengguang .range_cyclic = work->range_cyclic, 1447d46db3d5SWu Fengguang .range_start = 0, 1448d46db3d5SWu Fengguang .range_end = LLONG_MAX, 1449d46db3d5SWu Fengguang }; 1450d46db3d5SWu Fengguang unsigned long start_time = jiffies; 1451d46db3d5SWu Fengguang long write_chunk; 1452d46db3d5SWu Fengguang long wrote = 0; /* count both pages and inodes */ 1453d46db3d5SWu Fengguang 145403ba3782SJens Axboe while (!list_empty(&wb->b_io)) { 14557ccf19a8SNick Piggin struct inode *inode = wb_inode(wb->b_io.prev); 1456edadfb10SChristoph Hellwig 1457edadfb10SChristoph Hellwig if (inode->i_sb != sb) { 1458d46db3d5SWu Fengguang if (work->sb) { 1459edadfb10SChristoph Hellwig /* 1460edadfb10SChristoph Hellwig * We only want to write back data for this 1461edadfb10SChristoph Hellwig * superblock, move all inodes not belonging 1462edadfb10SChristoph Hellwig * to it back onto the dirty list. 1463edadfb10SChristoph Hellwig */ 1464f758eeabSChristoph Hellwig redirty_tail(inode, wb); 146566f3b8e2SJens Axboe continue; 146666f3b8e2SJens Axboe } 1467edadfb10SChristoph Hellwig 1468edadfb10SChristoph Hellwig /* 1469edadfb10SChristoph Hellwig * The inode belongs to a different superblock. 1470edadfb10SChristoph Hellwig * Bounce back to the caller to unpin this and 1471edadfb10SChristoph Hellwig * pin the next superblock. 1472edadfb10SChristoph Hellwig */ 1473d46db3d5SWu Fengguang break; 1474edadfb10SChristoph Hellwig } 1475edadfb10SChristoph Hellwig 14769843b76aSChristoph Hellwig /* 1477331cbdeeSWanpeng Li * Don't bother with new inodes or inodes being freed, first 1478331cbdeeSWanpeng Li * kind does not need periodic writeout yet, and for the latter 14799843b76aSChristoph Hellwig * kind writeout is handled by the freer. 14809843b76aSChristoph Hellwig */ 1481250df6edSDave Chinner spin_lock(&inode->i_lock); 14829843b76aSChristoph Hellwig if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { 1483250df6edSDave Chinner spin_unlock(&inode->i_lock); 1484fcc5c222SWu Fengguang redirty_tail(inode, wb); 14857ef0d737SNick Piggin continue; 14867ef0d737SNick Piggin } 1487cc1676d9SJan Kara if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) { 1488cc1676d9SJan Kara /* 1489cc1676d9SJan Kara * If this inode is locked for writeback and we are not 1490cc1676d9SJan Kara * doing writeback-for-data-integrity, move it to 1491cc1676d9SJan Kara * b_more_io so that writeback can proceed with the 1492cc1676d9SJan Kara * other inodes on s_io. 1493cc1676d9SJan Kara * 1494cc1676d9SJan Kara * We'll have another go at writing back this inode 1495cc1676d9SJan Kara * when we completed a full scan of b_io. 1496cc1676d9SJan Kara */ 1497cc1676d9SJan Kara spin_unlock(&inode->i_lock); 1498cc1676d9SJan Kara requeue_io(inode, wb); 1499cc1676d9SJan Kara trace_writeback_sb_inodes_requeue(inode); 1500cc1676d9SJan Kara continue; 1501cc1676d9SJan Kara } 1502f0d07b7fSJan Kara spin_unlock(&wb->list_lock); 1503f0d07b7fSJan Kara 15044f8ad655SJan Kara /* 15054f8ad655SJan Kara * We already requeued the inode if it had I_SYNC set and we 15064f8ad655SJan Kara * are doing WB_SYNC_NONE writeback. So this catches only the 15074f8ad655SJan Kara * WB_SYNC_ALL case. 15084f8ad655SJan Kara */ 1509169ebd90SJan Kara if (inode->i_state & I_SYNC) { 1510169ebd90SJan Kara /* Wait for I_SYNC. This function drops i_lock... */ 1511169ebd90SJan Kara inode_sleep_on_writeback(inode); 1512169ebd90SJan Kara /* Inode may be gone, start again */ 1513ead188f9SJan Kara spin_lock(&wb->list_lock); 1514169ebd90SJan Kara continue; 1515169ebd90SJan Kara } 15164f8ad655SJan Kara inode->i_state |= I_SYNC; 1517b16b1debSTejun Heo wbc_attach_and_unlock_inode(&wbc, inode); 1518169ebd90SJan Kara 1519a88a341aSTejun Heo write_chunk = writeback_chunk_size(wb, work); 1520d46db3d5SWu Fengguang wbc.nr_to_write = write_chunk; 1521d46db3d5SWu Fengguang wbc.pages_skipped = 0; 1522250df6edSDave Chinner 1523169ebd90SJan Kara /* 1524169ebd90SJan Kara * We use I_SYNC to pin the inode in memory. While it is set 1525169ebd90SJan Kara * evict_inode() will wait so the inode cannot be freed. 1526169ebd90SJan Kara */ 1527cd8ed2a4SYan Hong __writeback_single_inode(inode, &wbc); 1528d46db3d5SWu Fengguang 1529b16b1debSTejun Heo wbc_detach_inode(&wbc); 1530d46db3d5SWu Fengguang work->nr_pages -= write_chunk - wbc.nr_to_write; 1531d46db3d5SWu Fengguang wrote += write_chunk - wbc.nr_to_write; 1532590dca3aSChris Mason 1533590dca3aSChris Mason if (need_resched()) { 1534590dca3aSChris Mason /* 1535590dca3aSChris Mason * We're trying to balance between building up a nice 1536590dca3aSChris Mason * long list of IOs to improve our merge rate, and 1537590dca3aSChris Mason * getting those IOs out quickly for anyone throttling 1538590dca3aSChris Mason * in balance_dirty_pages(). cond_resched() doesn't 1539590dca3aSChris Mason * unplug, so get our IOs out the door before we 1540590dca3aSChris Mason * give up the CPU. 1541590dca3aSChris Mason */ 1542590dca3aSChris Mason blk_flush_plug(current); 1543590dca3aSChris Mason cond_resched(); 1544590dca3aSChris Mason } 1545590dca3aSChris Mason 1546590dca3aSChris Mason 15474f8ad655SJan Kara spin_lock(&wb->list_lock); 15484f8ad655SJan Kara spin_lock(&inode->i_lock); 15490ae45f63STheodore Ts'o if (!(inode->i_state & I_DIRTY_ALL)) 1550d46db3d5SWu Fengguang wrote++; 15514f8ad655SJan Kara requeue_inode(inode, wb, &wbc); 15524f8ad655SJan Kara inode_sync_complete(inode); 15530f1b1fd8SDave Chinner spin_unlock(&inode->i_lock); 1554590dca3aSChris Mason 1555d46db3d5SWu Fengguang /* 1556d46db3d5SWu Fengguang * bail out to wb_writeback() often enough to check 1557d46db3d5SWu Fengguang * background threshold and other termination conditions. 1558d46db3d5SWu Fengguang */ 1559d46db3d5SWu Fengguang if (wrote) { 1560d46db3d5SWu Fengguang if (time_is_before_jiffies(start_time + HZ / 10UL)) 1561d46db3d5SWu Fengguang break; 1562d46db3d5SWu Fengguang if (work->nr_pages <= 0) 1563d46db3d5SWu Fengguang break; 15641da177e4SLinus Torvalds } 15658bc3be27SFengguang Wu } 1566d46db3d5SWu Fengguang return wrote; 1567f11c9c5cSEdward Shishkin } 156838f21977SNick Piggin 1569d46db3d5SWu Fengguang static long __writeback_inodes_wb(struct bdi_writeback *wb, 1570d46db3d5SWu Fengguang struct wb_writeback_work *work) 1571f11c9c5cSEdward Shishkin { 1572d46db3d5SWu Fengguang unsigned long start_time = jiffies; 1573d46db3d5SWu Fengguang long wrote = 0; 1574f11c9c5cSEdward Shishkin 1575f11c9c5cSEdward Shishkin while (!list_empty(&wb->b_io)) { 15767ccf19a8SNick Piggin struct inode *inode = wb_inode(wb->b_io.prev); 1577f11c9c5cSEdward Shishkin struct super_block *sb = inode->i_sb; 1578f11c9c5cSEdward Shishkin 1579eb6ef3dfSKonstantin Khlebnikov if (!trylock_super(sb)) { 15800e995816SWu Fengguang /* 1581eb6ef3dfSKonstantin Khlebnikov * trylock_super() may fail consistently due to 15820e995816SWu Fengguang * s_umount being grabbed by someone else. Don't use 15830e995816SWu Fengguang * requeue_io() to avoid busy retrying the inode/sb. 15840e995816SWu Fengguang */ 15850e995816SWu Fengguang redirty_tail(inode, wb); 1586d19de7edSChristoph Hellwig continue; 1587334132aeSChristoph Hellwig } 1588d46db3d5SWu Fengguang wrote += writeback_sb_inodes(sb, wb, work); 1589eb6ef3dfSKonstantin Khlebnikov up_read(&sb->s_umount); 1590f11c9c5cSEdward Shishkin 1591d46db3d5SWu Fengguang /* refer to the same tests at the end of writeback_sb_inodes */ 1592d46db3d5SWu Fengguang if (wrote) { 1593d46db3d5SWu Fengguang if (time_is_before_jiffies(start_time + HZ / 10UL)) 1594d46db3d5SWu Fengguang break; 1595d46db3d5SWu Fengguang if (work->nr_pages <= 0) 1596f11c9c5cSEdward Shishkin break; 1597f11c9c5cSEdward Shishkin } 1598d46db3d5SWu Fengguang } 159966f3b8e2SJens Axboe /* Leave any unwritten inodes on b_io */ 1600d46db3d5SWu Fengguang return wrote; 160166f3b8e2SJens Axboe } 160266f3b8e2SJens Axboe 16037d9f073bSWanpeng Li static long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages, 16040e175a18SCurt Wohlgemuth enum wb_reason reason) 1605edadfb10SChristoph Hellwig { 1606d46db3d5SWu Fengguang struct wb_writeback_work work = { 1607d46db3d5SWu Fengguang .nr_pages = nr_pages, 1608d46db3d5SWu Fengguang .sync_mode = WB_SYNC_NONE, 1609d46db3d5SWu Fengguang .range_cyclic = 1, 16100e175a18SCurt Wohlgemuth .reason = reason, 1611d46db3d5SWu Fengguang }; 1612505a666eSLinus Torvalds struct blk_plug plug; 1613edadfb10SChristoph Hellwig 1614505a666eSLinus Torvalds blk_start_plug(&plug); 1615f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 1616424b351fSWu Fengguang if (list_empty(&wb->b_io)) 1617ad4e38ddSCurt Wohlgemuth queue_io(wb, &work); 1618d46db3d5SWu Fengguang __writeback_inodes_wb(wb, &work); 1619f758eeabSChristoph Hellwig spin_unlock(&wb->list_lock); 1620505a666eSLinus Torvalds blk_finish_plug(&plug); 1621edadfb10SChristoph Hellwig 1622d46db3d5SWu Fengguang return nr_pages - work.nr_pages; 162366f3b8e2SJens Axboe } 162466f3b8e2SJens Axboe 162503ba3782SJens Axboe /* 162603ba3782SJens Axboe * Explicit flushing or periodic writeback of "old" data. 162703ba3782SJens Axboe * 162803ba3782SJens Axboe * Define "old": the first time one of an inode's pages is dirtied, we mark the 162903ba3782SJens Axboe * dirtying-time in the inode's address_space. So this periodic writeback code 163003ba3782SJens Axboe * just walks the superblock inode list, writing back any inodes which are 163103ba3782SJens Axboe * older than a specific point in time. 163203ba3782SJens Axboe * 163303ba3782SJens Axboe * Try to run once per dirty_writeback_interval. But if a writeback event 163403ba3782SJens Axboe * takes longer than a dirty_writeback_interval interval, then leave a 163503ba3782SJens Axboe * one-second gap. 163603ba3782SJens Axboe * 163703ba3782SJens Axboe * older_than_this takes precedence over nr_to_write. So we'll only write back 163803ba3782SJens Axboe * all dirty pages if they are all attached to "old" mappings. 163903ba3782SJens Axboe */ 1640c4a77a6cSJens Axboe static long wb_writeback(struct bdi_writeback *wb, 164183ba7b07SChristoph Hellwig struct wb_writeback_work *work) 164203ba3782SJens Axboe { 1643e98be2d5SWu Fengguang unsigned long wb_start = jiffies; 1644d46db3d5SWu Fengguang long nr_pages = work->nr_pages; 16450dc83bd3SJan Kara unsigned long oldest_jif; 1646a5989bdcSJan Kara struct inode *inode; 1647d46db3d5SWu Fengguang long progress; 1648505a666eSLinus Torvalds struct blk_plug plug; 164903ba3782SJens Axboe 16500dc83bd3SJan Kara oldest_jif = jiffies; 16510dc83bd3SJan Kara work->older_than_this = &oldest_jif; 165203ba3782SJens Axboe 1653505a666eSLinus Torvalds blk_start_plug(&plug); 1654e8dfc305SWu Fengguang spin_lock(&wb->list_lock); 165503ba3782SJens Axboe for (;;) { 165603ba3782SJens Axboe /* 1657d3ddec76SWu Fengguang * Stop writeback when nr_pages has been consumed 165803ba3782SJens Axboe */ 165983ba7b07SChristoph Hellwig if (work->nr_pages <= 0) 166003ba3782SJens Axboe break; 166103ba3782SJens Axboe 166203ba3782SJens Axboe /* 1663aa373cf5SJan Kara * Background writeout and kupdate-style writeback may 1664aa373cf5SJan Kara * run forever. Stop them if there is other work to do 1665aa373cf5SJan Kara * so that e.g. sync can proceed. They'll be restarted 1666aa373cf5SJan Kara * after the other works are all done. 1667aa373cf5SJan Kara */ 1668aa373cf5SJan Kara if ((work->for_background || work->for_kupdate) && 1669f0054bb1STejun Heo !list_empty(&wb->work_list)) 1670aa373cf5SJan Kara break; 1671aa373cf5SJan Kara 1672aa373cf5SJan Kara /* 1673d3ddec76SWu Fengguang * For background writeout, stop when we are below the 1674d3ddec76SWu Fengguang * background dirty threshold 167503ba3782SJens Axboe */ 1676aa661bbeSTejun Heo if (work->for_background && !wb_over_bg_thresh(wb)) 167703ba3782SJens Axboe break; 167803ba3782SJens Axboe 16791bc36b64SJan Kara /* 16801bc36b64SJan Kara * Kupdate and background works are special and we want to 16811bc36b64SJan Kara * include all inodes that need writing. Livelock avoidance is 16821bc36b64SJan Kara * handled by these works yielding to any other work so we are 16831bc36b64SJan Kara * safe. 16841bc36b64SJan Kara */ 1685ba9aa839SWu Fengguang if (work->for_kupdate) { 16860dc83bd3SJan Kara oldest_jif = jiffies - 1687ba9aa839SWu Fengguang msecs_to_jiffies(dirty_expire_interval * 10); 16881bc36b64SJan Kara } else if (work->for_background) 16890dc83bd3SJan Kara oldest_jif = jiffies; 1690028c2dd1SDave Chinner 16915634cc2aSTejun Heo trace_writeback_start(wb, work); 1692e8dfc305SWu Fengguang if (list_empty(&wb->b_io)) 1693ad4e38ddSCurt Wohlgemuth queue_io(wb, work); 169483ba7b07SChristoph Hellwig if (work->sb) 1695d46db3d5SWu Fengguang progress = writeback_sb_inodes(work->sb, wb, work); 1696edadfb10SChristoph Hellwig else 1697d46db3d5SWu Fengguang progress = __writeback_inodes_wb(wb, work); 16985634cc2aSTejun Heo trace_writeback_written(wb, work); 1699028c2dd1SDave Chinner 1700e98be2d5SWu Fengguang wb_update_bandwidth(wb, wb_start); 170103ba3782SJens Axboe 170203ba3782SJens Axboe /* 170371fd05a8SJens Axboe * Did we write something? Try for more 1704e6fb6da2SWu Fengguang * 1705e6fb6da2SWu Fengguang * Dirty inodes are moved to b_io for writeback in batches. 1706e6fb6da2SWu Fengguang * The completion of the current batch does not necessarily 1707e6fb6da2SWu Fengguang * mean the overall work is done. So we keep looping as long 1708e6fb6da2SWu Fengguang * as made some progress on cleaning pages or inodes. 170971fd05a8SJens Axboe */ 1710d46db3d5SWu Fengguang if (progress) 171103ba3782SJens Axboe continue; 1712a5989bdcSJan Kara /* 1713e6fb6da2SWu Fengguang * No more inodes for IO, bail 1714a5989bdcSJan Kara */ 1715b7a2441fSWu Fengguang if (list_empty(&wb->b_more_io)) 171603ba3782SJens Axboe break; 171703ba3782SJens Axboe /* 17188010c3b6SJens Axboe * Nothing written. Wait for some inode to 17198010c3b6SJens Axboe * become available for writeback. Otherwise 17208010c3b6SJens Axboe * we'll just busyloop. 172103ba3782SJens Axboe */ 172203ba3782SJens Axboe if (!list_empty(&wb->b_more_io)) { 17235634cc2aSTejun Heo trace_writeback_wait(wb, work); 172403ba3782SJens Axboe inode = wb_inode(wb->b_more_io.prev); 1725250df6edSDave Chinner spin_lock(&inode->i_lock); 1726f0d07b7fSJan Kara spin_unlock(&wb->list_lock); 1727169ebd90SJan Kara /* This function drops i_lock... */ 1728169ebd90SJan Kara inode_sleep_on_writeback(inode); 1729f0d07b7fSJan Kara spin_lock(&wb->list_lock); 173003ba3782SJens Axboe } 173103ba3782SJens Axboe } 1732e8dfc305SWu Fengguang spin_unlock(&wb->list_lock); 1733505a666eSLinus Torvalds blk_finish_plug(&plug); 173403ba3782SJens Axboe 1735d46db3d5SWu Fengguang return nr_pages - work->nr_pages; 173603ba3782SJens Axboe } 173703ba3782SJens Axboe 173803ba3782SJens Axboe /* 173983ba7b07SChristoph Hellwig * Return the next wb_writeback_work struct that hasn't been processed yet. 174003ba3782SJens Axboe */ 1741f0054bb1STejun Heo static struct wb_writeback_work *get_next_work_item(struct bdi_writeback *wb) 174203ba3782SJens Axboe { 174383ba7b07SChristoph Hellwig struct wb_writeback_work *work = NULL; 174403ba3782SJens Axboe 1745f0054bb1STejun Heo spin_lock_bh(&wb->work_lock); 1746f0054bb1STejun Heo if (!list_empty(&wb->work_list)) { 1747f0054bb1STejun Heo work = list_entry(wb->work_list.next, 174883ba7b07SChristoph Hellwig struct wb_writeback_work, list); 174983ba7b07SChristoph Hellwig list_del_init(&work->list); 175003ba3782SJens Axboe } 1751f0054bb1STejun Heo spin_unlock_bh(&wb->work_lock); 175283ba7b07SChristoph Hellwig return work; 175303ba3782SJens Axboe } 175403ba3782SJens Axboe 1755cdf01dd5SLinus Torvalds /* 1756cdf01dd5SLinus Torvalds * Add in the number of potentially dirty inodes, because each inode 1757cdf01dd5SLinus Torvalds * write can dirty pagecache in the underlying blockdev. 1758cdf01dd5SLinus Torvalds */ 1759cdf01dd5SLinus Torvalds static unsigned long get_nr_dirty_pages(void) 1760cdf01dd5SLinus Torvalds { 1761cdf01dd5SLinus Torvalds return global_page_state(NR_FILE_DIRTY) + 1762cdf01dd5SLinus Torvalds global_page_state(NR_UNSTABLE_NFS) + 1763cdf01dd5SLinus Torvalds get_nr_dirty_inodes(); 1764cdf01dd5SLinus Torvalds } 1765cdf01dd5SLinus Torvalds 17666585027aSJan Kara static long wb_check_background_flush(struct bdi_writeback *wb) 17676585027aSJan Kara { 1768aa661bbeSTejun Heo if (wb_over_bg_thresh(wb)) { 17696585027aSJan Kara 17706585027aSJan Kara struct wb_writeback_work work = { 17716585027aSJan Kara .nr_pages = LONG_MAX, 17726585027aSJan Kara .sync_mode = WB_SYNC_NONE, 17736585027aSJan Kara .for_background = 1, 17746585027aSJan Kara .range_cyclic = 1, 17750e175a18SCurt Wohlgemuth .reason = WB_REASON_BACKGROUND, 17766585027aSJan Kara }; 17776585027aSJan Kara 17786585027aSJan Kara return wb_writeback(wb, &work); 17796585027aSJan Kara } 17806585027aSJan Kara 17816585027aSJan Kara return 0; 17826585027aSJan Kara } 17836585027aSJan Kara 178403ba3782SJens Axboe static long wb_check_old_data_flush(struct bdi_writeback *wb) 178503ba3782SJens Axboe { 178603ba3782SJens Axboe unsigned long expired; 178703ba3782SJens Axboe long nr_pages; 178803ba3782SJens Axboe 178969b62d01SJens Axboe /* 179069b62d01SJens Axboe * When set to zero, disable periodic writeback 179169b62d01SJens Axboe */ 179269b62d01SJens Axboe if (!dirty_writeback_interval) 179369b62d01SJens Axboe return 0; 179469b62d01SJens Axboe 179503ba3782SJens Axboe expired = wb->last_old_flush + 179603ba3782SJens Axboe msecs_to_jiffies(dirty_writeback_interval * 10); 179703ba3782SJens Axboe if (time_before(jiffies, expired)) 179803ba3782SJens Axboe return 0; 179903ba3782SJens Axboe 180003ba3782SJens Axboe wb->last_old_flush = jiffies; 1801cdf01dd5SLinus Torvalds nr_pages = get_nr_dirty_pages(); 180203ba3782SJens Axboe 1803c4a77a6cSJens Axboe if (nr_pages) { 180483ba7b07SChristoph Hellwig struct wb_writeback_work work = { 1805c4a77a6cSJens Axboe .nr_pages = nr_pages, 1806c4a77a6cSJens Axboe .sync_mode = WB_SYNC_NONE, 1807c4a77a6cSJens Axboe .for_kupdate = 1, 1808c4a77a6cSJens Axboe .range_cyclic = 1, 18090e175a18SCurt Wohlgemuth .reason = WB_REASON_PERIODIC, 1810c4a77a6cSJens Axboe }; 1811c4a77a6cSJens Axboe 181283ba7b07SChristoph Hellwig return wb_writeback(wb, &work); 1813c4a77a6cSJens Axboe } 181403ba3782SJens Axboe 181503ba3782SJens Axboe return 0; 181603ba3782SJens Axboe } 181703ba3782SJens Axboe 181803ba3782SJens Axboe /* 181903ba3782SJens Axboe * Retrieve work items and do the writeback they describe 182003ba3782SJens Axboe */ 182125d130baSWanpeng Li static long wb_do_writeback(struct bdi_writeback *wb) 182203ba3782SJens Axboe { 182383ba7b07SChristoph Hellwig struct wb_writeback_work *work; 1824c4a77a6cSJens Axboe long wrote = 0; 182503ba3782SJens Axboe 18264452226eSTejun Heo set_bit(WB_writeback_running, &wb->state); 1827f0054bb1STejun Heo while ((work = get_next_work_item(wb)) != NULL) { 1828cc395d7fSTejun Heo struct wb_completion *done = work->done; 182983ba7b07SChristoph Hellwig 18305634cc2aSTejun Heo trace_writeback_exec(wb, work); 1831455b2864SDave Chinner 183283ba7b07SChristoph Hellwig wrote += wb_writeback(wb, work); 183303ba3782SJens Axboe 18348a1270cdSTejun Heo if (work->auto_free) 183583ba7b07SChristoph Hellwig kfree(work); 1836cc395d7fSTejun Heo if (done && atomic_dec_and_test(&done->cnt)) 1837cc395d7fSTejun Heo wake_up_all(&wb->bdi->wb_waitq); 183803ba3782SJens Axboe } 183903ba3782SJens Axboe 184003ba3782SJens Axboe /* 184103ba3782SJens Axboe * Check for periodic writeback, kupdated() style 184203ba3782SJens Axboe */ 184303ba3782SJens Axboe wrote += wb_check_old_data_flush(wb); 18446585027aSJan Kara wrote += wb_check_background_flush(wb); 18454452226eSTejun Heo clear_bit(WB_writeback_running, &wb->state); 184603ba3782SJens Axboe 184703ba3782SJens Axboe return wrote; 184803ba3782SJens Axboe } 184903ba3782SJens Axboe 185003ba3782SJens Axboe /* 185103ba3782SJens Axboe * Handle writeback of dirty data for the device backed by this bdi. Also 1852839a8e86STejun Heo * reschedules periodically and does kupdated style flushing. 185303ba3782SJens Axboe */ 1854f0054bb1STejun Heo void wb_workfn(struct work_struct *work) 185503ba3782SJens Axboe { 1856839a8e86STejun Heo struct bdi_writeback *wb = container_of(to_delayed_work(work), 1857839a8e86STejun Heo struct bdi_writeback, dwork); 185803ba3782SJens Axboe long pages_written; 185903ba3782SJens Axboe 1860f0054bb1STejun Heo set_worker_desc("flush-%s", dev_name(wb->bdi->dev)); 1861766f9164SPeter Zijlstra current->flags |= PF_SWAPWRITE; 186203ba3782SJens Axboe 1863839a8e86STejun Heo if (likely(!current_is_workqueue_rescuer() || 18644452226eSTejun Heo !test_bit(WB_registered, &wb->state))) { 186503ba3782SJens Axboe /* 1866f0054bb1STejun Heo * The normal path. Keep writing back @wb until its 1867839a8e86STejun Heo * work_list is empty. Note that this path is also taken 1868f0054bb1STejun Heo * if @wb is shutting down even when we're running off the 1869839a8e86STejun Heo * rescuer as work_list needs to be drained. 187003ba3782SJens Axboe */ 1871839a8e86STejun Heo do { 187225d130baSWanpeng Li pages_written = wb_do_writeback(wb); 1873455b2864SDave Chinner trace_writeback_pages_written(pages_written); 1874f0054bb1STejun Heo } while (!list_empty(&wb->work_list)); 1875839a8e86STejun Heo } else { 1876253c34e9SArtem Bityutskiy /* 1877839a8e86STejun Heo * bdi_wq can't get enough workers and we're running off 1878839a8e86STejun Heo * the emergency worker. Don't hog it. Hopefully, 1024 is 1879839a8e86STejun Heo * enough for efficient IO. 1880253c34e9SArtem Bityutskiy */ 1881f0054bb1STejun Heo pages_written = writeback_inodes_wb(wb, 1024, 1882839a8e86STejun Heo WB_REASON_FORKER_THREAD); 1883839a8e86STejun Heo trace_writeback_pages_written(pages_written); 188403ba3782SJens Axboe } 188503ba3782SJens Axboe 1886f0054bb1STejun Heo if (!list_empty(&wb->work_list)) 18876ca738d6SDerek Basehore mod_delayed_work(bdi_wq, &wb->dwork, 0); 18886ca738d6SDerek Basehore else if (wb_has_dirty_io(wb) && dirty_writeback_interval) 1889f0054bb1STejun Heo wb_wakeup_delayed(wb); 1890455b2864SDave Chinner 1891839a8e86STejun Heo current->flags &= ~PF_SWAPWRITE; 189203ba3782SJens Axboe } 189303ba3782SJens Axboe 189403ba3782SJens Axboe /* 189503ba3782SJens Axboe * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back 189603ba3782SJens Axboe * the whole world. 189703ba3782SJens Axboe */ 18980e175a18SCurt Wohlgemuth void wakeup_flusher_threads(long nr_pages, enum wb_reason reason) 189903ba3782SJens Axboe { 1900b8c2f347SChristoph Hellwig struct backing_dev_info *bdi; 1901b8c2f347SChristoph Hellwig 190247df3ddeSJan Kara if (!nr_pages) 190347df3ddeSJan Kara nr_pages = get_nr_dirty_pages(); 1904b8c2f347SChristoph Hellwig 1905b8c2f347SChristoph Hellwig rcu_read_lock(); 1906f2b65121STejun Heo list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { 1907f2b65121STejun Heo struct bdi_writeback *wb; 1908f2b65121STejun Heo 1909f2b65121STejun Heo if (!bdi_has_dirty_io(bdi)) 1910f2b65121STejun Heo continue; 1911f2b65121STejun Heo 1912b817525aSTejun Heo list_for_each_entry_rcu(wb, &bdi->wb_list, bdi_node) 1913f2b65121STejun Heo wb_start_writeback(wb, wb_split_bdi_pages(wb, nr_pages), 1914f2b65121STejun Heo false, reason); 1915f2b65121STejun Heo } 1916b8c2f347SChristoph Hellwig rcu_read_unlock(); 191703ba3782SJens Axboe } 191803ba3782SJens Axboe 1919a2f48706STheodore Ts'o /* 1920a2f48706STheodore Ts'o * Wake up bdi's periodically to make sure dirtytime inodes gets 1921a2f48706STheodore Ts'o * written back periodically. We deliberately do *not* check the 1922a2f48706STheodore Ts'o * b_dirtytime list in wb_has_dirty_io(), since this would cause the 1923a2f48706STheodore Ts'o * kernel to be constantly waking up once there are any dirtytime 1924a2f48706STheodore Ts'o * inodes on the system. So instead we define a separate delayed work 1925a2f48706STheodore Ts'o * function which gets called much more rarely. (By default, only 1926a2f48706STheodore Ts'o * once every 12 hours.) 1927a2f48706STheodore Ts'o * 1928a2f48706STheodore Ts'o * If there is any other write activity going on in the file system, 1929a2f48706STheodore Ts'o * this function won't be necessary. But if the only thing that has 1930a2f48706STheodore Ts'o * happened on the file system is a dirtytime inode caused by an atime 1931a2f48706STheodore Ts'o * update, we need this infrastructure below to make sure that inode 1932a2f48706STheodore Ts'o * eventually gets pushed out to disk. 1933a2f48706STheodore Ts'o */ 1934a2f48706STheodore Ts'o static void wakeup_dirtytime_writeback(struct work_struct *w); 1935a2f48706STheodore Ts'o static DECLARE_DELAYED_WORK(dirtytime_work, wakeup_dirtytime_writeback); 1936a2f48706STheodore Ts'o 1937a2f48706STheodore Ts'o static void wakeup_dirtytime_writeback(struct work_struct *w) 1938a2f48706STheodore Ts'o { 1939a2f48706STheodore Ts'o struct backing_dev_info *bdi; 1940a2f48706STheodore Ts'o 1941a2f48706STheodore Ts'o rcu_read_lock(); 1942a2f48706STheodore Ts'o list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { 1943001fe6f6STejun Heo struct bdi_writeback *wb; 1944001fe6f6STejun Heo 1945b817525aSTejun Heo list_for_each_entry_rcu(wb, &bdi->wb_list, bdi_node) 19466fdf860fSTejun Heo if (!list_empty(&wb->b_dirty_time)) 19476fdf860fSTejun Heo wb_wakeup(wb); 1948a2f48706STheodore Ts'o } 1949a2f48706STheodore Ts'o rcu_read_unlock(); 1950a2f48706STheodore Ts'o schedule_delayed_work(&dirtytime_work, dirtytime_expire_interval * HZ); 1951a2f48706STheodore Ts'o } 1952a2f48706STheodore Ts'o 1953a2f48706STheodore Ts'o static int __init start_dirtytime_writeback(void) 1954a2f48706STheodore Ts'o { 1955a2f48706STheodore Ts'o schedule_delayed_work(&dirtytime_work, dirtytime_expire_interval * HZ); 1956a2f48706STheodore Ts'o return 0; 1957a2f48706STheodore Ts'o } 1958a2f48706STheodore Ts'o __initcall(start_dirtytime_writeback); 1959a2f48706STheodore Ts'o 19601efff914STheodore Ts'o int dirtytime_interval_handler(struct ctl_table *table, int write, 19611efff914STheodore Ts'o void __user *buffer, size_t *lenp, loff_t *ppos) 19621efff914STheodore Ts'o { 19631efff914STheodore Ts'o int ret; 19641efff914STheodore Ts'o 19651efff914STheodore Ts'o ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos); 19661efff914STheodore Ts'o if (ret == 0 && write) 19671efff914STheodore Ts'o mod_delayed_work(system_wq, &dirtytime_work, 0); 19681efff914STheodore Ts'o return ret; 19691efff914STheodore Ts'o } 19701efff914STheodore Ts'o 197103ba3782SJens Axboe static noinline void block_dump___mark_inode_dirty(struct inode *inode) 197203ba3782SJens Axboe { 197303ba3782SJens Axboe if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { 197403ba3782SJens Axboe struct dentry *dentry; 197503ba3782SJens Axboe const char *name = "?"; 197603ba3782SJens Axboe 197703ba3782SJens Axboe dentry = d_find_alias(inode); 197803ba3782SJens Axboe if (dentry) { 197903ba3782SJens Axboe spin_lock(&dentry->d_lock); 198003ba3782SJens Axboe name = (const char *) dentry->d_name.name; 198103ba3782SJens Axboe } 198203ba3782SJens Axboe printk(KERN_DEBUG 198303ba3782SJens Axboe "%s(%d): dirtied inode %lu (%s) on %s\n", 198403ba3782SJens Axboe current->comm, task_pid_nr(current), inode->i_ino, 198503ba3782SJens Axboe name, inode->i_sb->s_id); 198603ba3782SJens Axboe if (dentry) { 198703ba3782SJens Axboe spin_unlock(&dentry->d_lock); 198803ba3782SJens Axboe dput(dentry); 198903ba3782SJens Axboe } 199003ba3782SJens Axboe } 199103ba3782SJens Axboe } 199203ba3782SJens Axboe 199303ba3782SJens Axboe /** 199403ba3782SJens Axboe * __mark_inode_dirty - internal function 199503ba3782SJens Axboe * @inode: inode to mark 199603ba3782SJens Axboe * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) 199703ba3782SJens Axboe * Mark an inode as dirty. Callers should use mark_inode_dirty or 199803ba3782SJens Axboe * mark_inode_dirty_sync. 199903ba3782SJens Axboe * 200003ba3782SJens Axboe * Put the inode on the super block's dirty list. 200103ba3782SJens Axboe * 200203ba3782SJens Axboe * CAREFUL! We mark it dirty unconditionally, but move it onto the 200303ba3782SJens Axboe * dirty list only if it is hashed or if it refers to a blockdev. 200403ba3782SJens Axboe * If it was not hashed, it will never be added to the dirty list 200503ba3782SJens Axboe * even if it is later hashed, as it will have been marked dirty already. 200603ba3782SJens Axboe * 200703ba3782SJens Axboe * In short, make sure you hash any inodes _before_ you start marking 200803ba3782SJens Axboe * them dirty. 200903ba3782SJens Axboe * 201003ba3782SJens Axboe * Note that for blockdevs, inode->dirtied_when represents the dirtying time of 201103ba3782SJens Axboe * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of 201203ba3782SJens Axboe * the kernel-internal blockdev inode represents the dirtying time of the 201303ba3782SJens Axboe * blockdev's pages. This is why for I_DIRTY_PAGES we always use 201403ba3782SJens Axboe * page->mapping->host, so the page-dirtying time is recorded in the internal 201503ba3782SJens Axboe * blockdev inode. 201603ba3782SJens Axboe */ 201703ba3782SJens Axboe void __mark_inode_dirty(struct inode *inode, int flags) 201803ba3782SJens Axboe { 2019dbce03b9SRandy Dunlap #define I_DIRTY_INODE (I_DIRTY_SYNC | I_DIRTY_DATASYNC) 202003ba3782SJens Axboe struct super_block *sb = inode->i_sb; 20210ae45f63STheodore Ts'o int dirtytime; 20220ae45f63STheodore Ts'o 20230ae45f63STheodore Ts'o trace_writeback_mark_inode_dirty(inode, flags); 202403ba3782SJens Axboe 202503ba3782SJens Axboe /* 202603ba3782SJens Axboe * Don't do this for I_DIRTY_PAGES - that doesn't actually 202703ba3782SJens Axboe * dirty the inode itself 202803ba3782SJens Axboe */ 20290ae45f63STheodore Ts'o if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC | I_DIRTY_TIME)) { 20309fb0a7daSTejun Heo trace_writeback_dirty_inode_start(inode, flags); 20319fb0a7daSTejun Heo 203203ba3782SJens Axboe if (sb->s_op->dirty_inode) 2033aa385729SChristoph Hellwig sb->s_op->dirty_inode(inode, flags); 20349fb0a7daSTejun Heo 20359fb0a7daSTejun Heo trace_writeback_dirty_inode(inode, flags); 203603ba3782SJens Axboe } 20370ae45f63STheodore Ts'o if (flags & I_DIRTY_INODE) 20380ae45f63STheodore Ts'o flags &= ~I_DIRTY_TIME; 20390ae45f63STheodore Ts'o dirtytime = flags & I_DIRTY_TIME; 204003ba3782SJens Axboe 204103ba3782SJens Axboe /* 20429c6ac78eSTejun Heo * Paired with smp_mb() in __writeback_single_inode() for the 20439c6ac78eSTejun Heo * following lockless i_state test. See there for details. 204403ba3782SJens Axboe */ 204503ba3782SJens Axboe smp_mb(); 204603ba3782SJens Axboe 20470ae45f63STheodore Ts'o if (((inode->i_state & flags) == flags) || 20480ae45f63STheodore Ts'o (dirtytime && (inode->i_state & I_DIRTY_INODE))) 204903ba3782SJens Axboe return; 205003ba3782SJens Axboe 205103ba3782SJens Axboe if (unlikely(block_dump)) 205203ba3782SJens Axboe block_dump___mark_inode_dirty(inode); 205303ba3782SJens Axboe 2054250df6edSDave Chinner spin_lock(&inode->i_lock); 20550ae45f63STheodore Ts'o if (dirtytime && (inode->i_state & I_DIRTY_INODE)) 20560ae45f63STheodore Ts'o goto out_unlock_inode; 205703ba3782SJens Axboe if ((inode->i_state & flags) != flags) { 205803ba3782SJens Axboe const int was_dirty = inode->i_state & I_DIRTY; 205903ba3782SJens Axboe 206052ebea74STejun Heo inode_attach_wb(inode, NULL); 206152ebea74STejun Heo 20620ae45f63STheodore Ts'o if (flags & I_DIRTY_INODE) 20630ae45f63STheodore Ts'o inode->i_state &= ~I_DIRTY_TIME; 206403ba3782SJens Axboe inode->i_state |= flags; 206503ba3782SJens Axboe 206603ba3782SJens Axboe /* 206703ba3782SJens Axboe * If the inode is being synced, just update its dirty state. 206803ba3782SJens Axboe * The unlocker will place the inode on the appropriate 206903ba3782SJens Axboe * superblock list, based upon its state. 207003ba3782SJens Axboe */ 207103ba3782SJens Axboe if (inode->i_state & I_SYNC) 2072250df6edSDave Chinner goto out_unlock_inode; 207303ba3782SJens Axboe 207403ba3782SJens Axboe /* 207503ba3782SJens Axboe * Only add valid (hashed) inodes to the superblock's 207603ba3782SJens Axboe * dirty list. Add blockdev inodes as well. 207703ba3782SJens Axboe */ 207803ba3782SJens Axboe if (!S_ISBLK(inode->i_mode)) { 20791d3382cbSAl Viro if (inode_unhashed(inode)) 2080250df6edSDave Chinner goto out_unlock_inode; 208103ba3782SJens Axboe } 2082a4ffdde6SAl Viro if (inode->i_state & I_FREEING) 2083250df6edSDave Chinner goto out_unlock_inode; 208403ba3782SJens Axboe 208503ba3782SJens Axboe /* 208603ba3782SJens Axboe * If the inode was already on b_dirty/b_io/b_more_io, don't 208703ba3782SJens Axboe * reposition it (that would break b_dirty time-ordering). 208803ba3782SJens Axboe */ 208903ba3782SJens Axboe if (!was_dirty) { 209087e1d789STejun Heo struct bdi_writeback *wb; 2091d6c10f1fSTejun Heo struct list_head *dirty_list; 2092a66979abSDave Chinner bool wakeup_bdi = false; 2093500b067cSJens Axboe 209487e1d789STejun Heo wb = locked_inode_to_wb_and_lock_list(inode); 2095253c34e9SArtem Bityutskiy 20960747259dSTejun Heo WARN(bdi_cap_writeback_dirty(wb->bdi) && 20970747259dSTejun Heo !test_bit(WB_registered, &wb->state), 20980747259dSTejun Heo "bdi-%s not registered\n", wb->bdi->name); 209903ba3782SJens Axboe 210003ba3782SJens Axboe inode->dirtied_when = jiffies; 2101a2f48706STheodore Ts'o if (dirtytime) 2102a2f48706STheodore Ts'o inode->dirtied_time_when = jiffies; 2103d6c10f1fSTejun Heo 2104a2f48706STheodore Ts'o if (inode->i_state & (I_DIRTY_INODE | I_DIRTY_PAGES)) 21050747259dSTejun Heo dirty_list = &wb->b_dirty; 2106a2f48706STheodore Ts'o else 21070747259dSTejun Heo dirty_list = &wb->b_dirty_time; 2108d6c10f1fSTejun Heo 2109c7f54084SDave Chinner wakeup_bdi = inode_io_list_move_locked(inode, wb, 2110d6c10f1fSTejun Heo dirty_list); 2111d6c10f1fSTejun Heo 21120747259dSTejun Heo spin_unlock(&wb->list_lock); 21130ae45f63STheodore Ts'o trace_writeback_dirty_inode_enqueue(inode); 2114253c34e9SArtem Bityutskiy 2115d6c10f1fSTejun Heo /* 2116d6c10f1fSTejun Heo * If this is the first dirty inode for this bdi, 2117d6c10f1fSTejun Heo * we have to wake-up the corresponding bdi thread 2118d6c10f1fSTejun Heo * to make sure background write-back happens 2119d6c10f1fSTejun Heo * later. 2120d6c10f1fSTejun Heo */ 21210747259dSTejun Heo if (bdi_cap_writeback_dirty(wb->bdi) && wakeup_bdi) 21220747259dSTejun Heo wb_wakeup_delayed(wb); 2123a66979abSDave Chinner return; 2124a66979abSDave Chinner } 2125a66979abSDave Chinner } 2126a66979abSDave Chinner out_unlock_inode: 2127a66979abSDave Chinner spin_unlock(&inode->i_lock); 2128a66979abSDave Chinner 2129dbce03b9SRandy Dunlap #undef I_DIRTY_INODE 213003ba3782SJens Axboe } 213103ba3782SJens Axboe EXPORT_SYMBOL(__mark_inode_dirty); 213203ba3782SJens Axboe 2133e97fedb9SDave Chinner /* 2134e97fedb9SDave Chinner * The @s_sync_lock is used to serialise concurrent sync operations 2135e97fedb9SDave Chinner * to avoid lock contention problems with concurrent wait_sb_inodes() calls. 2136e97fedb9SDave Chinner * Concurrent callers will block on the s_sync_lock rather than doing contending 2137e97fedb9SDave Chinner * walks. The queueing maintains sync(2) required behaviour as all the IO that 2138e97fedb9SDave Chinner * has been issued up to the time this function is enter is guaranteed to be 2139e97fedb9SDave Chinner * completed by the time we have gained the lock and waited for all IO that is 2140e97fedb9SDave Chinner * in progress regardless of the order callers are granted the lock. 2141e97fedb9SDave Chinner */ 2142b6e51316SJens Axboe static void wait_sb_inodes(struct super_block *sb) 214366f3b8e2SJens Axboe { 214438f21977SNick Piggin struct inode *inode, *old_inode = NULL; 214538f21977SNick Piggin 214603ba3782SJens Axboe /* 214703ba3782SJens Axboe * We need to be protected against the filesystem going from 214803ba3782SJens Axboe * r/o to r/w or vice versa. 214903ba3782SJens Axboe */ 2150b6e51316SJens Axboe WARN_ON(!rwsem_is_locked(&sb->s_umount)); 215103ba3782SJens Axboe 2152e97fedb9SDave Chinner mutex_lock(&sb->s_sync_lock); 215374278da9SDave Chinner spin_lock(&sb->s_inode_list_lock); 215466f3b8e2SJens Axboe 215538f21977SNick Piggin /* 215638f21977SNick Piggin * Data integrity sync. Must wait for all pages under writeback, 215738f21977SNick Piggin * because there may have been pages dirtied before our sync 215838f21977SNick Piggin * call, but which had writeout started before we write it out. 215938f21977SNick Piggin * In which case, the inode may not be on the dirty list, but 216038f21977SNick Piggin * we still have to wait for that writeout. 216138f21977SNick Piggin */ 2162b6e51316SJens Axboe list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { 2163250df6edSDave Chinner struct address_space *mapping = inode->i_mapping; 216438f21977SNick Piggin 2165250df6edSDave Chinner spin_lock(&inode->i_lock); 2166250df6edSDave Chinner if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) || 2167250df6edSDave Chinner (mapping->nrpages == 0)) { 2168250df6edSDave Chinner spin_unlock(&inode->i_lock); 216938f21977SNick Piggin continue; 2170250df6edSDave Chinner } 217138f21977SNick Piggin __iget(inode); 2172250df6edSDave Chinner spin_unlock(&inode->i_lock); 217374278da9SDave Chinner spin_unlock(&sb->s_inode_list_lock); 217455fa6091SDave Chinner 217538f21977SNick Piggin /* 217655fa6091SDave Chinner * We hold a reference to 'inode' so it couldn't have been 217755fa6091SDave Chinner * removed from s_inodes list while we dropped the 217874278da9SDave Chinner * s_inode_list_lock. We cannot iput the inode now as we can 217955fa6091SDave Chinner * be holding the last reference and we cannot iput it under 218074278da9SDave Chinner * s_inode_list_lock. So we keep the reference and iput it 218155fa6091SDave Chinner * later. 218238f21977SNick Piggin */ 218338f21977SNick Piggin iput(old_inode); 218438f21977SNick Piggin old_inode = inode; 218538f21977SNick Piggin 2186aa750fd7SJunichi Nomura /* 2187aa750fd7SJunichi Nomura * We keep the error status of individual mapping so that 2188aa750fd7SJunichi Nomura * applications can catch the writeback error using fsync(2). 2189aa750fd7SJunichi Nomura * See filemap_fdatawait_keep_errors() for details. 2190aa750fd7SJunichi Nomura */ 2191aa750fd7SJunichi Nomura filemap_fdatawait_keep_errors(mapping); 219238f21977SNick Piggin 219338f21977SNick Piggin cond_resched(); 219438f21977SNick Piggin 219574278da9SDave Chinner spin_lock(&sb->s_inode_list_lock); 219638f21977SNick Piggin } 219774278da9SDave Chinner spin_unlock(&sb->s_inode_list_lock); 219838f21977SNick Piggin iput(old_inode); 2199e97fedb9SDave Chinner mutex_unlock(&sb->s_sync_lock); 220066f3b8e2SJens Axboe } 22011da177e4SLinus Torvalds 2202f30a7d0cSTejun Heo static void __writeback_inodes_sb_nr(struct super_block *sb, unsigned long nr, 2203f30a7d0cSTejun Heo enum wb_reason reason, bool skip_if_busy) 22041da177e4SLinus Torvalds { 2205cc395d7fSTejun Heo DEFINE_WB_COMPLETION_ONSTACK(done); 220683ba7b07SChristoph Hellwig struct wb_writeback_work work = { 22073c4d7165SChristoph Hellwig .sb = sb, 22083c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_NONE, 22096e6938b6SWu Fengguang .tagged_writepages = 1, 221083ba7b07SChristoph Hellwig .done = &done, 22113259f8beSChris Mason .nr_pages = nr, 22120e175a18SCurt Wohlgemuth .reason = reason, 22133c4d7165SChristoph Hellwig }; 2214e7972912STejun Heo struct backing_dev_info *bdi = sb->s_bdi; 22150e3c9a22SJens Axboe 2216e7972912STejun Heo if (!bdi_has_dirty_io(bdi) || bdi == &noop_backing_dev_info) 22176eedc701SJan Kara return; 2218cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 2219f30a7d0cSTejun Heo 2220db125360STejun Heo bdi_split_work_to_wbs(sb->s_bdi, &work, skip_if_busy); 2221cc395d7fSTejun Heo wb_wait_for_completion(bdi, &done); 22221da177e4SLinus Torvalds } 2223f30a7d0cSTejun Heo 2224f30a7d0cSTejun Heo /** 2225f30a7d0cSTejun Heo * writeback_inodes_sb_nr - writeback dirty inodes from given super_block 2226f30a7d0cSTejun Heo * @sb: the superblock 2227f30a7d0cSTejun Heo * @nr: the number of pages to write 2228f30a7d0cSTejun Heo * @reason: reason why some writeback work initiated 2229f30a7d0cSTejun Heo * 2230f30a7d0cSTejun Heo * Start writeback on some inodes on this super_block. No guarantees are made 2231f30a7d0cSTejun Heo * on how many (if any) will be written, and this function does not wait 2232f30a7d0cSTejun Heo * for IO completion of submitted IO. 2233f30a7d0cSTejun Heo */ 2234f30a7d0cSTejun Heo void writeback_inodes_sb_nr(struct super_block *sb, 2235f30a7d0cSTejun Heo unsigned long nr, 2236f30a7d0cSTejun Heo enum wb_reason reason) 2237f30a7d0cSTejun Heo { 2238f30a7d0cSTejun Heo __writeback_inodes_sb_nr(sb, nr, reason, false); 2239f30a7d0cSTejun Heo } 22403259f8beSChris Mason EXPORT_SYMBOL(writeback_inodes_sb_nr); 22413259f8beSChris Mason 22423259f8beSChris Mason /** 22433259f8beSChris Mason * writeback_inodes_sb - writeback dirty inodes from given super_block 22443259f8beSChris Mason * @sb: the superblock 2245786228abSMarcos Paulo de Souza * @reason: reason why some writeback work was initiated 22463259f8beSChris Mason * 22473259f8beSChris Mason * Start writeback on some inodes on this super_block. No guarantees are made 22483259f8beSChris Mason * on how many (if any) will be written, and this function does not wait 22493259f8beSChris Mason * for IO completion of submitted IO. 22503259f8beSChris Mason */ 22510e175a18SCurt Wohlgemuth void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) 22523259f8beSChris Mason { 22530e175a18SCurt Wohlgemuth return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); 22543259f8beSChris Mason } 2255d8a8559cSJens Axboe EXPORT_SYMBOL(writeback_inodes_sb); 2256d8a8559cSJens Axboe 2257d8a8559cSJens Axboe /** 225810ee27a0SMiao Xie * try_to_writeback_inodes_sb_nr - try to start writeback if none underway 22593259f8beSChris Mason * @sb: the superblock 22603259f8beSChris Mason * @nr: the number of pages to write 226110ee27a0SMiao Xie * @reason: the reason of writeback 22623259f8beSChris Mason * 226310ee27a0SMiao Xie * Invoke writeback_inodes_sb_nr if no writeback is currently underway. 22643259f8beSChris Mason * Returns 1 if writeback was started, 0 if not. 22653259f8beSChris Mason */ 2266f30a7d0cSTejun Heo bool try_to_writeback_inodes_sb_nr(struct super_block *sb, unsigned long nr, 22670e175a18SCurt Wohlgemuth enum wb_reason reason) 22683259f8beSChris Mason { 226910ee27a0SMiao Xie if (!down_read_trylock(&sb->s_umount)) 2270f30a7d0cSTejun Heo return false; 227110ee27a0SMiao Xie 2272f30a7d0cSTejun Heo __writeback_inodes_sb_nr(sb, nr, reason, true); 22733259f8beSChris Mason up_read(&sb->s_umount); 2274f30a7d0cSTejun Heo return true; 22753259f8beSChris Mason } 227610ee27a0SMiao Xie EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr); 227710ee27a0SMiao Xie 227810ee27a0SMiao Xie /** 227910ee27a0SMiao Xie * try_to_writeback_inodes_sb - try to start writeback if none underway 228010ee27a0SMiao Xie * @sb: the superblock 228110ee27a0SMiao Xie * @reason: reason why some writeback work was initiated 228210ee27a0SMiao Xie * 228310ee27a0SMiao Xie * Implement by try_to_writeback_inodes_sb_nr() 228410ee27a0SMiao Xie * Returns 1 if writeback was started, 0 if not. 228510ee27a0SMiao Xie */ 2286f30a7d0cSTejun Heo bool try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) 228710ee27a0SMiao Xie { 228810ee27a0SMiao Xie return try_to_writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); 228910ee27a0SMiao Xie } 229010ee27a0SMiao Xie EXPORT_SYMBOL(try_to_writeback_inodes_sb); 22913259f8beSChris Mason 22923259f8beSChris Mason /** 2293d8a8559cSJens Axboe * sync_inodes_sb - sync sb inode pages 2294d8a8559cSJens Axboe * @sb: the superblock 2295d8a8559cSJens Axboe * 2296d8a8559cSJens Axboe * This function writes and waits on any dirty inode belonging to this 22970dc83bd3SJan Kara * super_block. 2298d8a8559cSJens Axboe */ 22990dc83bd3SJan Kara void sync_inodes_sb(struct super_block *sb) 2300d8a8559cSJens Axboe { 2301cc395d7fSTejun Heo DEFINE_WB_COMPLETION_ONSTACK(done); 230283ba7b07SChristoph Hellwig struct wb_writeback_work work = { 23033c4d7165SChristoph Hellwig .sb = sb, 23043c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_ALL, 23053c4d7165SChristoph Hellwig .nr_pages = LONG_MAX, 23063c4d7165SChristoph Hellwig .range_cyclic = 0, 230783ba7b07SChristoph Hellwig .done = &done, 23080e175a18SCurt Wohlgemuth .reason = WB_REASON_SYNC, 23097747bd4bSDave Chinner .for_sync = 1, 23103c4d7165SChristoph Hellwig }; 2311e7972912STejun Heo struct backing_dev_info *bdi = sb->s_bdi; 23123c4d7165SChristoph Hellwig 2313006a0973STejun Heo /* 2314006a0973STejun Heo * Can't skip on !bdi_has_dirty() because we should wait for !dirty 2315006a0973STejun Heo * inodes under writeback and I_DIRTY_TIME inodes ignored by 2316006a0973STejun Heo * bdi_has_dirty() need to be written out too. 2317006a0973STejun Heo */ 2318006a0973STejun Heo if (bdi == &noop_backing_dev_info) 23196eedc701SJan Kara return; 2320cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 2321cf37e972SChristoph Hellwig 2322db125360STejun Heo bdi_split_work_to_wbs(bdi, &work, false); 2323cc395d7fSTejun Heo wb_wait_for_completion(bdi, &done); 232483ba7b07SChristoph Hellwig 2325b6e51316SJens Axboe wait_sb_inodes(sb); 2326d8a8559cSJens Axboe } 2327d8a8559cSJens Axboe EXPORT_SYMBOL(sync_inodes_sb); 23281da177e4SLinus Torvalds 23291da177e4SLinus Torvalds /** 23301da177e4SLinus Torvalds * write_inode_now - write an inode to disk 23311da177e4SLinus Torvalds * @inode: inode to write to disk 23321da177e4SLinus Torvalds * @sync: whether the write should be synchronous or not 23331da177e4SLinus Torvalds * 23347f04c26dSAndrea Arcangeli * This function commits an inode to disk immediately if it is dirty. This is 23357f04c26dSAndrea Arcangeli * primarily needed by knfsd. 23367f04c26dSAndrea Arcangeli * 23377f04c26dSAndrea Arcangeli * The caller must either have a ref on the inode or must have set I_WILL_FREE. 23381da177e4SLinus Torvalds */ 23391da177e4SLinus Torvalds int write_inode_now(struct inode *inode, int sync) 23401da177e4SLinus Torvalds { 2341f758eeabSChristoph Hellwig struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; 23421da177e4SLinus Torvalds struct writeback_control wbc = { 23431da177e4SLinus Torvalds .nr_to_write = LONG_MAX, 234418914b18SMike Galbraith .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, 2345111ebb6eSOGAWA Hirofumi .range_start = 0, 2346111ebb6eSOGAWA Hirofumi .range_end = LLONG_MAX, 23471da177e4SLinus Torvalds }; 23481da177e4SLinus Torvalds 23491da177e4SLinus Torvalds if (!mapping_cap_writeback_dirty(inode->i_mapping)) 235049364ce2SAndrew Morton wbc.nr_to_write = 0; 23511da177e4SLinus Torvalds 23521da177e4SLinus Torvalds might_sleep(); 23534f8ad655SJan Kara return writeback_single_inode(inode, wb, &wbc); 23541da177e4SLinus Torvalds } 23551da177e4SLinus Torvalds EXPORT_SYMBOL(write_inode_now); 23561da177e4SLinus Torvalds 23571da177e4SLinus Torvalds /** 23581da177e4SLinus Torvalds * sync_inode - write an inode and its pages to disk. 23591da177e4SLinus Torvalds * @inode: the inode to sync 23601da177e4SLinus Torvalds * @wbc: controls the writeback mode 23611da177e4SLinus Torvalds * 23621da177e4SLinus Torvalds * sync_inode() will write an inode and its pages to disk. It will also 23631da177e4SLinus Torvalds * correctly update the inode on its superblock's dirty inode lists and will 23641da177e4SLinus Torvalds * update inode->i_state. 23651da177e4SLinus Torvalds * 23661da177e4SLinus Torvalds * The caller must have a ref on the inode. 23671da177e4SLinus Torvalds */ 23681da177e4SLinus Torvalds int sync_inode(struct inode *inode, struct writeback_control *wbc) 23691da177e4SLinus Torvalds { 23704f8ad655SJan Kara return writeback_single_inode(inode, &inode_to_bdi(inode)->wb, wbc); 23711da177e4SLinus Torvalds } 23721da177e4SLinus Torvalds EXPORT_SYMBOL(sync_inode); 2373c3765016SChristoph Hellwig 2374c3765016SChristoph Hellwig /** 2375c691b9d9SAndrew Morton * sync_inode_metadata - write an inode to disk 2376c3765016SChristoph Hellwig * @inode: the inode to sync 2377c3765016SChristoph Hellwig * @wait: wait for I/O to complete. 2378c3765016SChristoph Hellwig * 2379c691b9d9SAndrew Morton * Write an inode to disk and adjust its dirty state after completion. 2380c3765016SChristoph Hellwig * 2381c3765016SChristoph Hellwig * Note: only writes the actual inode, no associated data or other metadata. 2382c3765016SChristoph Hellwig */ 2383c3765016SChristoph Hellwig int sync_inode_metadata(struct inode *inode, int wait) 2384c3765016SChristoph Hellwig { 2385c3765016SChristoph Hellwig struct writeback_control wbc = { 2386c3765016SChristoph Hellwig .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE, 2387c3765016SChristoph Hellwig .nr_to_write = 0, /* metadata-only */ 2388c3765016SChristoph Hellwig }; 2389c3765016SChristoph Hellwig 2390c3765016SChristoph Hellwig return sync_inode(inode, &wbc); 2391c3765016SChristoph Hellwig } 2392c3765016SChristoph Hellwig EXPORT_SYMBOL(sync_inode_metadata); 2393