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 22621c6321fSTejun Heo void __inode_attach_wb(struct inode *inode, struct page *page) 22721c6321fSTejun Heo { 22821c6321fSTejun Heo struct backing_dev_info *bdi = inode_to_bdi(inode); 22921c6321fSTejun Heo struct bdi_writeback *wb = NULL; 23021c6321fSTejun Heo 23121c6321fSTejun Heo if (inode_cgwb_enabled(inode)) { 23221c6321fSTejun Heo struct cgroup_subsys_state *memcg_css; 23321c6321fSTejun Heo 23421c6321fSTejun Heo if (page) { 23521c6321fSTejun Heo memcg_css = mem_cgroup_css_from_page(page); 23621c6321fSTejun Heo wb = wb_get_create(bdi, memcg_css, GFP_ATOMIC); 23721c6321fSTejun Heo } else { 23821c6321fSTejun Heo /* must pin memcg_css, see wb_get_create() */ 23921c6321fSTejun Heo memcg_css = task_get_css(current, memory_cgrp_id); 24021c6321fSTejun Heo wb = wb_get_create(bdi, memcg_css, GFP_ATOMIC); 24121c6321fSTejun Heo css_put(memcg_css); 24221c6321fSTejun Heo } 24321c6321fSTejun Heo } 24421c6321fSTejun Heo 24521c6321fSTejun Heo if (!wb) 24621c6321fSTejun Heo wb = &bdi->wb; 24721c6321fSTejun Heo 24821c6321fSTejun Heo /* 24921c6321fSTejun Heo * There may be multiple instances of this function racing to 25021c6321fSTejun Heo * update the same inode. Use cmpxchg() to tell the winner. 25121c6321fSTejun Heo */ 25221c6321fSTejun Heo if (unlikely(cmpxchg(&inode->i_wb, NULL, wb))) 25321c6321fSTejun Heo wb_put(wb); 25421c6321fSTejun Heo } 25521c6321fSTejun Heo 256703c2708STejun Heo /** 25787e1d789STejun Heo * locked_inode_to_wb_and_lock_list - determine a locked inode's wb and lock it 25887e1d789STejun Heo * @inode: inode of interest with i_lock held 25987e1d789STejun Heo * 26087e1d789STejun Heo * Returns @inode's wb with its list_lock held. @inode->i_lock must be 26187e1d789STejun Heo * held on entry and is released on return. The returned wb is guaranteed 26287e1d789STejun Heo * to stay @inode's associated wb until its list_lock is released. 26387e1d789STejun Heo */ 26487e1d789STejun Heo static struct bdi_writeback * 26587e1d789STejun Heo locked_inode_to_wb_and_lock_list(struct inode *inode) 26687e1d789STejun Heo __releases(&inode->i_lock) 26787e1d789STejun Heo __acquires(&wb->list_lock) 26887e1d789STejun Heo { 26987e1d789STejun Heo while (true) { 27087e1d789STejun Heo struct bdi_writeback *wb = inode_to_wb(inode); 27187e1d789STejun Heo 27287e1d789STejun Heo /* 27387e1d789STejun Heo * inode_to_wb() association is protected by both 27487e1d789STejun Heo * @inode->i_lock and @wb->list_lock but list_lock nests 27587e1d789STejun Heo * outside i_lock. Drop i_lock and verify that the 27687e1d789STejun Heo * association hasn't changed after acquiring list_lock. 27787e1d789STejun Heo */ 27887e1d789STejun Heo wb_get(wb); 27987e1d789STejun Heo spin_unlock(&inode->i_lock); 28087e1d789STejun Heo spin_lock(&wb->list_lock); 28187e1d789STejun Heo wb_put(wb); /* not gonna deref it anymore */ 28287e1d789STejun Heo 283aaa2cacfSTejun Heo /* i_wb may have changed inbetween, can't use inode_to_wb() */ 284aaa2cacfSTejun Heo if (likely(wb == inode->i_wb)) 28587e1d789STejun Heo return wb; /* @inode already has ref */ 28687e1d789STejun Heo 28787e1d789STejun Heo spin_unlock(&wb->list_lock); 28887e1d789STejun Heo cpu_relax(); 28987e1d789STejun Heo spin_lock(&inode->i_lock); 29087e1d789STejun Heo } 29187e1d789STejun Heo } 29287e1d789STejun Heo 29387e1d789STejun Heo /** 29487e1d789STejun Heo * inode_to_wb_and_lock_list - determine an inode's wb and lock it 29587e1d789STejun Heo * @inode: inode of interest 29687e1d789STejun Heo * 29787e1d789STejun Heo * Same as locked_inode_to_wb_and_lock_list() but @inode->i_lock isn't held 29887e1d789STejun Heo * on entry. 29987e1d789STejun Heo */ 30087e1d789STejun Heo static struct bdi_writeback *inode_to_wb_and_lock_list(struct inode *inode) 30187e1d789STejun Heo __acquires(&wb->list_lock) 30287e1d789STejun Heo { 30387e1d789STejun Heo spin_lock(&inode->i_lock); 30487e1d789STejun Heo return locked_inode_to_wb_and_lock_list(inode); 30587e1d789STejun Heo } 30687e1d789STejun Heo 307682aa8e1STejun Heo struct inode_switch_wbs_context { 308682aa8e1STejun Heo struct inode *inode; 309682aa8e1STejun Heo struct bdi_writeback *new_wb; 310682aa8e1STejun Heo 311682aa8e1STejun Heo struct rcu_head rcu_head; 312682aa8e1STejun Heo struct work_struct work; 313682aa8e1STejun Heo }; 314682aa8e1STejun Heo 315682aa8e1STejun Heo static void inode_switch_wbs_work_fn(struct work_struct *work) 316682aa8e1STejun Heo { 317682aa8e1STejun Heo struct inode_switch_wbs_context *isw = 318682aa8e1STejun Heo container_of(work, struct inode_switch_wbs_context, work); 319682aa8e1STejun Heo struct inode *inode = isw->inode; 320d10c8095STejun Heo struct address_space *mapping = inode->i_mapping; 321d10c8095STejun Heo struct bdi_writeback *old_wb = inode->i_wb; 322682aa8e1STejun Heo struct bdi_writeback *new_wb = isw->new_wb; 323d10c8095STejun Heo struct radix_tree_iter iter; 324d10c8095STejun Heo bool switched = false; 325d10c8095STejun Heo void **slot; 326682aa8e1STejun Heo 327682aa8e1STejun Heo /* 328682aa8e1STejun Heo * By the time control reaches here, RCU grace period has passed 329682aa8e1STejun Heo * since I_WB_SWITCH assertion and all wb stat update transactions 330682aa8e1STejun Heo * between unlocked_inode_to_wb_begin/end() are guaranteed to be 331682aa8e1STejun Heo * synchronizing against mapping->tree_lock. 332d10c8095STejun Heo * 333d10c8095STejun Heo * Grabbing old_wb->list_lock, inode->i_lock and mapping->tree_lock 334d10c8095STejun Heo * gives us exclusion against all wb related operations on @inode 335d10c8095STejun Heo * including IO list manipulations and stat updates. 336682aa8e1STejun Heo */ 337d10c8095STejun Heo if (old_wb < new_wb) { 338d10c8095STejun Heo spin_lock(&old_wb->list_lock); 339d10c8095STejun Heo spin_lock_nested(&new_wb->list_lock, SINGLE_DEPTH_NESTING); 340d10c8095STejun Heo } else { 341d10c8095STejun Heo spin_lock(&new_wb->list_lock); 342d10c8095STejun Heo spin_lock_nested(&old_wb->list_lock, SINGLE_DEPTH_NESTING); 343d10c8095STejun Heo } 344682aa8e1STejun Heo spin_lock(&inode->i_lock); 345d10c8095STejun Heo spin_lock_irq(&mapping->tree_lock); 346682aa8e1STejun Heo 347d10c8095STejun Heo /* 348d10c8095STejun Heo * Once I_FREEING is visible under i_lock, the eviction path owns 349c7f54084SDave Chinner * the inode and we shouldn't modify ->i_io_list. 350d10c8095STejun Heo */ 351d10c8095STejun Heo if (unlikely(inode->i_state & I_FREEING)) 352d10c8095STejun Heo goto skip_switch; 353d10c8095STejun Heo 354d10c8095STejun Heo /* 355d10c8095STejun Heo * Count and transfer stats. Note that PAGECACHE_TAG_DIRTY points 356d10c8095STejun Heo * to possibly dirty pages while PAGECACHE_TAG_WRITEBACK points to 357d10c8095STejun Heo * pages actually under underwriteback. 358d10c8095STejun Heo */ 359d10c8095STejun Heo radix_tree_for_each_tagged(slot, &mapping->page_tree, &iter, 0, 360d10c8095STejun Heo PAGECACHE_TAG_DIRTY) { 361d10c8095STejun Heo struct page *page = radix_tree_deref_slot_protected(slot, 362d10c8095STejun Heo &mapping->tree_lock); 363d10c8095STejun Heo if (likely(page) && PageDirty(page)) { 364d10c8095STejun Heo __dec_wb_stat(old_wb, WB_RECLAIMABLE); 365d10c8095STejun Heo __inc_wb_stat(new_wb, WB_RECLAIMABLE); 366d10c8095STejun Heo } 367d10c8095STejun Heo } 368d10c8095STejun Heo 369d10c8095STejun Heo radix_tree_for_each_tagged(slot, &mapping->page_tree, &iter, 0, 370d10c8095STejun Heo PAGECACHE_TAG_WRITEBACK) { 371d10c8095STejun Heo struct page *page = radix_tree_deref_slot_protected(slot, 372d10c8095STejun Heo &mapping->tree_lock); 373d10c8095STejun Heo if (likely(page)) { 374d10c8095STejun Heo WARN_ON_ONCE(!PageWriteback(page)); 375d10c8095STejun Heo __dec_wb_stat(old_wb, WB_WRITEBACK); 376d10c8095STejun Heo __inc_wb_stat(new_wb, WB_WRITEBACK); 377d10c8095STejun Heo } 378d10c8095STejun Heo } 379d10c8095STejun Heo 380d10c8095STejun Heo wb_get(new_wb); 381d10c8095STejun Heo 382d10c8095STejun Heo /* 383d10c8095STejun Heo * Transfer to @new_wb's IO list if necessary. The specific list 384d10c8095STejun Heo * @inode was on is ignored and the inode is put on ->b_dirty which 385d10c8095STejun Heo * is always correct including from ->b_dirty_time. The transfer 386d10c8095STejun Heo * preserves @inode->dirtied_when ordering. 387d10c8095STejun Heo */ 388c7f54084SDave Chinner if (!list_empty(&inode->i_io_list)) { 389d10c8095STejun Heo struct inode *pos; 390d10c8095STejun Heo 391c7f54084SDave Chinner inode_io_list_del_locked(inode, old_wb); 392d10c8095STejun Heo inode->i_wb = new_wb; 393c7f54084SDave Chinner list_for_each_entry(pos, &new_wb->b_dirty, i_io_list) 394d10c8095STejun Heo if (time_after_eq(inode->dirtied_when, 395d10c8095STejun Heo pos->dirtied_when)) 396d10c8095STejun Heo break; 397c7f54084SDave Chinner inode_io_list_move_locked(inode, new_wb, pos->i_io_list.prev); 398d10c8095STejun Heo } else { 399d10c8095STejun Heo inode->i_wb = new_wb; 400d10c8095STejun Heo } 401d10c8095STejun Heo 402d10c8095STejun Heo /* ->i_wb_frn updates may race wbc_detach_inode() but doesn't matter */ 403682aa8e1STejun Heo inode->i_wb_frn_winner = 0; 404682aa8e1STejun Heo inode->i_wb_frn_avg_time = 0; 405682aa8e1STejun Heo inode->i_wb_frn_history = 0; 406d10c8095STejun Heo switched = true; 407d10c8095STejun Heo skip_switch: 408682aa8e1STejun Heo /* 409682aa8e1STejun Heo * Paired with load_acquire in unlocked_inode_to_wb_begin() and 410682aa8e1STejun Heo * ensures that the new wb is visible if they see !I_WB_SWITCH. 411682aa8e1STejun Heo */ 412682aa8e1STejun Heo smp_store_release(&inode->i_state, inode->i_state & ~I_WB_SWITCH); 413682aa8e1STejun Heo 414d10c8095STejun Heo spin_unlock_irq(&mapping->tree_lock); 415682aa8e1STejun Heo spin_unlock(&inode->i_lock); 416d10c8095STejun Heo spin_unlock(&new_wb->list_lock); 417d10c8095STejun Heo spin_unlock(&old_wb->list_lock); 418d10c8095STejun Heo 419d10c8095STejun Heo if (switched) { 420d10c8095STejun Heo wb_wakeup(new_wb); 421d10c8095STejun Heo wb_put(old_wb); 422d10c8095STejun Heo } 423d10c8095STejun Heo wb_put(new_wb); 424682aa8e1STejun Heo 425682aa8e1STejun Heo iput(inode); 426682aa8e1STejun Heo kfree(isw); 427682aa8e1STejun Heo } 428682aa8e1STejun Heo 429682aa8e1STejun Heo static void inode_switch_wbs_rcu_fn(struct rcu_head *rcu_head) 430682aa8e1STejun Heo { 431682aa8e1STejun Heo struct inode_switch_wbs_context *isw = container_of(rcu_head, 432682aa8e1STejun Heo struct inode_switch_wbs_context, rcu_head); 433682aa8e1STejun Heo 434682aa8e1STejun Heo /* needs to grab bh-unsafe locks, bounce to work item */ 435682aa8e1STejun Heo INIT_WORK(&isw->work, inode_switch_wbs_work_fn); 436682aa8e1STejun Heo schedule_work(&isw->work); 437682aa8e1STejun Heo } 438682aa8e1STejun Heo 439682aa8e1STejun Heo /** 440682aa8e1STejun Heo * inode_switch_wbs - change the wb association of an inode 441682aa8e1STejun Heo * @inode: target inode 442682aa8e1STejun Heo * @new_wb_id: ID of the new wb 443682aa8e1STejun Heo * 444682aa8e1STejun Heo * Switch @inode's wb association to the wb identified by @new_wb_id. The 445682aa8e1STejun Heo * switching is performed asynchronously and may fail silently. 446682aa8e1STejun Heo */ 447682aa8e1STejun Heo static void inode_switch_wbs(struct inode *inode, int new_wb_id) 448682aa8e1STejun Heo { 449682aa8e1STejun Heo struct backing_dev_info *bdi = inode_to_bdi(inode); 450682aa8e1STejun Heo struct cgroup_subsys_state *memcg_css; 451682aa8e1STejun Heo struct inode_switch_wbs_context *isw; 452682aa8e1STejun Heo 453682aa8e1STejun Heo /* noop if seems to be already in progress */ 454682aa8e1STejun Heo if (inode->i_state & I_WB_SWITCH) 455682aa8e1STejun Heo return; 456682aa8e1STejun Heo 457682aa8e1STejun Heo isw = kzalloc(sizeof(*isw), GFP_ATOMIC); 458682aa8e1STejun Heo if (!isw) 459682aa8e1STejun Heo return; 460682aa8e1STejun Heo 461682aa8e1STejun Heo /* find and pin the new wb */ 462682aa8e1STejun Heo rcu_read_lock(); 463682aa8e1STejun Heo memcg_css = css_from_id(new_wb_id, &memory_cgrp_subsys); 464682aa8e1STejun Heo if (memcg_css) 465682aa8e1STejun Heo isw->new_wb = wb_get_create(bdi, memcg_css, GFP_ATOMIC); 466682aa8e1STejun Heo rcu_read_unlock(); 467682aa8e1STejun Heo if (!isw->new_wb) 468682aa8e1STejun Heo goto out_free; 469682aa8e1STejun Heo 470682aa8e1STejun Heo /* while holding I_WB_SWITCH, no one else can update the association */ 471682aa8e1STejun Heo spin_lock(&inode->i_lock); 472682aa8e1STejun Heo if (inode->i_state & (I_WB_SWITCH | I_FREEING) || 473682aa8e1STejun Heo inode_to_wb(inode) == isw->new_wb) { 474682aa8e1STejun Heo spin_unlock(&inode->i_lock); 475682aa8e1STejun Heo goto out_free; 476682aa8e1STejun Heo } 477682aa8e1STejun Heo inode->i_state |= I_WB_SWITCH; 478682aa8e1STejun Heo spin_unlock(&inode->i_lock); 479682aa8e1STejun Heo 480682aa8e1STejun Heo ihold(inode); 481682aa8e1STejun Heo isw->inode = inode; 482682aa8e1STejun Heo 483682aa8e1STejun Heo /* 484682aa8e1STejun Heo * In addition to synchronizing among switchers, I_WB_SWITCH tells 485682aa8e1STejun Heo * the RCU protected stat update paths to grab the mapping's 486682aa8e1STejun Heo * tree_lock so that stat transfer can synchronize against them. 487682aa8e1STejun Heo * Let's continue after I_WB_SWITCH is guaranteed to be visible. 488682aa8e1STejun Heo */ 489682aa8e1STejun Heo call_rcu(&isw->rcu_head, inode_switch_wbs_rcu_fn); 490682aa8e1STejun Heo return; 491682aa8e1STejun Heo 492682aa8e1STejun Heo out_free: 493682aa8e1STejun Heo if (isw->new_wb) 494682aa8e1STejun Heo wb_put(isw->new_wb); 495682aa8e1STejun Heo kfree(isw); 496682aa8e1STejun Heo } 497682aa8e1STejun Heo 49887e1d789STejun Heo /** 499b16b1debSTejun Heo * wbc_attach_and_unlock_inode - associate wbc with target inode and unlock it 500b16b1debSTejun Heo * @wbc: writeback_control of interest 501b16b1debSTejun Heo * @inode: target inode 502b16b1debSTejun Heo * 503b16b1debSTejun Heo * @inode is locked and about to be written back under the control of @wbc. 504b16b1debSTejun Heo * Record @inode's writeback context into @wbc and unlock the i_lock. On 505b16b1debSTejun Heo * writeback completion, wbc_detach_inode() should be called. This is used 506b16b1debSTejun Heo * to track the cgroup writeback context. 507b16b1debSTejun Heo */ 508b16b1debSTejun Heo void wbc_attach_and_unlock_inode(struct writeback_control *wbc, 509b16b1debSTejun Heo struct inode *inode) 510b16b1debSTejun Heo { 511dd73e4b7STejun Heo if (!inode_cgwb_enabled(inode)) { 512dd73e4b7STejun Heo spin_unlock(&inode->i_lock); 513dd73e4b7STejun Heo return; 514dd73e4b7STejun Heo } 515dd73e4b7STejun Heo 516b16b1debSTejun Heo wbc->wb = inode_to_wb(inode); 5172a814908STejun Heo wbc->inode = inode; 5182a814908STejun Heo 5192a814908STejun Heo wbc->wb_id = wbc->wb->memcg_css->id; 5202a814908STejun Heo wbc->wb_lcand_id = inode->i_wb_frn_winner; 5212a814908STejun Heo wbc->wb_tcand_id = 0; 5222a814908STejun Heo wbc->wb_bytes = 0; 5232a814908STejun Heo wbc->wb_lcand_bytes = 0; 5242a814908STejun Heo wbc->wb_tcand_bytes = 0; 5252a814908STejun Heo 526b16b1debSTejun Heo wb_get(wbc->wb); 527b16b1debSTejun Heo spin_unlock(&inode->i_lock); 528e8a7abf5STejun Heo 529e8a7abf5STejun Heo /* 530e8a7abf5STejun Heo * A dying wb indicates that the memcg-blkcg mapping has changed 531e8a7abf5STejun Heo * and a new wb is already serving the memcg. Switch immediately. 532e8a7abf5STejun Heo */ 533e8a7abf5STejun Heo if (unlikely(wb_dying(wbc->wb))) 534e8a7abf5STejun Heo inode_switch_wbs(inode, wbc->wb_id); 535b16b1debSTejun Heo } 536b16b1debSTejun Heo 537b16b1debSTejun Heo /** 5382a814908STejun Heo * wbc_detach_inode - disassociate wbc from inode and perform foreign detection 5392a814908STejun Heo * @wbc: writeback_control of the just finished writeback 540b16b1debSTejun Heo * 541b16b1debSTejun Heo * To be called after a writeback attempt of an inode finishes and undoes 542b16b1debSTejun Heo * wbc_attach_and_unlock_inode(). Can be called under any context. 5432a814908STejun Heo * 5442a814908STejun Heo * As concurrent write sharing of an inode is expected to be very rare and 5452a814908STejun Heo * memcg only tracks page ownership on first-use basis severely confining 5462a814908STejun Heo * the usefulness of such sharing, cgroup writeback tracks ownership 5472a814908STejun Heo * per-inode. While the support for concurrent write sharing of an inode 5482a814908STejun Heo * is deemed unnecessary, an inode being written to by different cgroups at 5492a814908STejun Heo * different points in time is a lot more common, and, more importantly, 5502a814908STejun Heo * charging only by first-use can too readily lead to grossly incorrect 5512a814908STejun Heo * behaviors (single foreign page can lead to gigabytes of writeback to be 5522a814908STejun Heo * incorrectly attributed). 5532a814908STejun Heo * 5542a814908STejun Heo * To resolve this issue, cgroup writeback detects the majority dirtier of 5552a814908STejun Heo * an inode and transfers the ownership to it. To avoid unnnecessary 5562a814908STejun Heo * oscillation, the detection mechanism keeps track of history and gives 5572a814908STejun Heo * out the switch verdict only if the foreign usage pattern is stable over 5582a814908STejun Heo * a certain amount of time and/or writeback attempts. 5592a814908STejun Heo * 5602a814908STejun Heo * On each writeback attempt, @wbc tries to detect the majority writer 5612a814908STejun Heo * using Boyer-Moore majority vote algorithm. In addition to the byte 5622a814908STejun Heo * count from the majority voting, it also counts the bytes written for the 5632a814908STejun Heo * current wb and the last round's winner wb (max of last round's current 5642a814908STejun Heo * wb, the winner from two rounds ago, and the last round's majority 5652a814908STejun Heo * candidate). Keeping track of the historical winner helps the algorithm 5662a814908STejun Heo * to semi-reliably detect the most active writer even when it's not the 5672a814908STejun Heo * absolute majority. 5682a814908STejun Heo * 5692a814908STejun Heo * Once the winner of the round is determined, whether the winner is 5702a814908STejun Heo * foreign or not and how much IO time the round consumed is recorded in 5712a814908STejun Heo * inode->i_wb_frn_history. If the amount of recorded foreign IO time is 5722a814908STejun Heo * over a certain threshold, the switch verdict is given. 573b16b1debSTejun Heo */ 574b16b1debSTejun Heo void wbc_detach_inode(struct writeback_control *wbc) 575b16b1debSTejun Heo { 5762a814908STejun Heo struct bdi_writeback *wb = wbc->wb; 5772a814908STejun Heo struct inode *inode = wbc->inode; 578dd73e4b7STejun Heo unsigned long avg_time, max_bytes, max_time; 579dd73e4b7STejun Heo u16 history; 5802a814908STejun Heo int max_id; 5812a814908STejun Heo 582dd73e4b7STejun Heo if (!wb) 583dd73e4b7STejun Heo return; 584dd73e4b7STejun Heo 585dd73e4b7STejun Heo history = inode->i_wb_frn_history; 586dd73e4b7STejun Heo avg_time = inode->i_wb_frn_avg_time; 587dd73e4b7STejun Heo 5882a814908STejun Heo /* pick the winner of this round */ 5892a814908STejun Heo if (wbc->wb_bytes >= wbc->wb_lcand_bytes && 5902a814908STejun Heo wbc->wb_bytes >= wbc->wb_tcand_bytes) { 5912a814908STejun Heo max_id = wbc->wb_id; 5922a814908STejun Heo max_bytes = wbc->wb_bytes; 5932a814908STejun Heo } else if (wbc->wb_lcand_bytes >= wbc->wb_tcand_bytes) { 5942a814908STejun Heo max_id = wbc->wb_lcand_id; 5952a814908STejun Heo max_bytes = wbc->wb_lcand_bytes; 5962a814908STejun Heo } else { 5972a814908STejun Heo max_id = wbc->wb_tcand_id; 5982a814908STejun Heo max_bytes = wbc->wb_tcand_bytes; 5992a814908STejun Heo } 6002a814908STejun Heo 6012a814908STejun Heo /* 6022a814908STejun Heo * Calculate the amount of IO time the winner consumed and fold it 6032a814908STejun Heo * into the running average kept per inode. If the consumed IO 6042a814908STejun Heo * time is lower than avag / WB_FRN_TIME_CUT_DIV, ignore it for 6052a814908STejun Heo * deciding whether to switch or not. This is to prevent one-off 6062a814908STejun Heo * small dirtiers from skewing the verdict. 6072a814908STejun Heo */ 6082a814908STejun Heo max_time = DIV_ROUND_UP((max_bytes >> PAGE_SHIFT) << WB_FRN_TIME_SHIFT, 6092a814908STejun Heo wb->avg_write_bandwidth); 6102a814908STejun Heo if (avg_time) 6112a814908STejun Heo avg_time += (max_time >> WB_FRN_TIME_AVG_SHIFT) - 6122a814908STejun Heo (avg_time >> WB_FRN_TIME_AVG_SHIFT); 6132a814908STejun Heo else 6142a814908STejun Heo avg_time = max_time; /* immediate catch up on first run */ 6152a814908STejun Heo 6162a814908STejun Heo if (max_time >= avg_time / WB_FRN_TIME_CUT_DIV) { 6172a814908STejun Heo int slots; 6182a814908STejun Heo 6192a814908STejun Heo /* 6202a814908STejun Heo * The switch verdict is reached if foreign wb's consume 6212a814908STejun Heo * more than a certain proportion of IO time in a 6222a814908STejun Heo * WB_FRN_TIME_PERIOD. This is loosely tracked by 16 slot 6232a814908STejun Heo * history mask where each bit represents one sixteenth of 6242a814908STejun Heo * the period. Determine the number of slots to shift into 6252a814908STejun Heo * history from @max_time. 6262a814908STejun Heo */ 6272a814908STejun Heo slots = min(DIV_ROUND_UP(max_time, WB_FRN_HIST_UNIT), 6282a814908STejun Heo (unsigned long)WB_FRN_HIST_MAX_SLOTS); 6292a814908STejun Heo history <<= slots; 6302a814908STejun Heo if (wbc->wb_id != max_id) 6312a814908STejun Heo history |= (1U << slots) - 1; 6322a814908STejun Heo 6332a814908STejun Heo /* 6342a814908STejun Heo * Switch if the current wb isn't the consistent winner. 6352a814908STejun Heo * If there are multiple closely competing dirtiers, the 6362a814908STejun Heo * inode may switch across them repeatedly over time, which 6372a814908STejun Heo * is okay. The main goal is avoiding keeping an inode on 6382a814908STejun Heo * the wrong wb for an extended period of time. 6392a814908STejun Heo */ 640682aa8e1STejun Heo if (hweight32(history) > WB_FRN_HIST_THR_SLOTS) 641682aa8e1STejun Heo inode_switch_wbs(inode, max_id); 6422a814908STejun Heo } 6432a814908STejun Heo 6442a814908STejun Heo /* 6452a814908STejun Heo * Multiple instances of this function may race to update the 6462a814908STejun Heo * following fields but we don't mind occassional inaccuracies. 6472a814908STejun Heo */ 6482a814908STejun Heo inode->i_wb_frn_winner = max_id; 6492a814908STejun Heo inode->i_wb_frn_avg_time = min(avg_time, (unsigned long)U16_MAX); 6502a814908STejun Heo inode->i_wb_frn_history = history; 6512a814908STejun Heo 652b16b1debSTejun Heo wb_put(wbc->wb); 653b16b1debSTejun Heo wbc->wb = NULL; 654b16b1debSTejun Heo } 655b16b1debSTejun Heo 656b16b1debSTejun Heo /** 6572a814908STejun Heo * wbc_account_io - account IO issued during writeback 6582a814908STejun Heo * @wbc: writeback_control of the writeback in progress 6592a814908STejun Heo * @page: page being written out 6602a814908STejun Heo * @bytes: number of bytes being written out 6612a814908STejun Heo * 6622a814908STejun Heo * @bytes from @page are about to written out during the writeback 6632a814908STejun Heo * controlled by @wbc. Keep the book for foreign inode detection. See 6642a814908STejun Heo * wbc_detach_inode(). 6652a814908STejun Heo */ 6662a814908STejun Heo void wbc_account_io(struct writeback_control *wbc, struct page *page, 6672a814908STejun Heo size_t bytes) 6682a814908STejun Heo { 6692a814908STejun Heo int id; 6702a814908STejun Heo 6712a814908STejun Heo /* 6722a814908STejun Heo * pageout() path doesn't attach @wbc to the inode being written 6732a814908STejun Heo * out. This is intentional as we don't want the function to block 6742a814908STejun Heo * behind a slow cgroup. Ultimately, we want pageout() to kick off 6752a814908STejun Heo * regular writeback instead of writing things out itself. 6762a814908STejun Heo */ 6772a814908STejun Heo if (!wbc->wb) 6782a814908STejun Heo return; 6792a814908STejun Heo 6802a814908STejun Heo rcu_read_lock(); 6812a814908STejun Heo id = mem_cgroup_css_from_page(page)->id; 6822a814908STejun Heo rcu_read_unlock(); 6832a814908STejun Heo 6842a814908STejun Heo if (id == wbc->wb_id) { 6852a814908STejun Heo wbc->wb_bytes += bytes; 6862a814908STejun Heo return; 6872a814908STejun Heo } 6882a814908STejun Heo 6892a814908STejun Heo if (id == wbc->wb_lcand_id) 6902a814908STejun Heo wbc->wb_lcand_bytes += bytes; 6912a814908STejun Heo 6922a814908STejun Heo /* Boyer-Moore majority vote algorithm */ 6932a814908STejun Heo if (!wbc->wb_tcand_bytes) 6942a814908STejun Heo wbc->wb_tcand_id = id; 6952a814908STejun Heo if (id == wbc->wb_tcand_id) 6962a814908STejun Heo wbc->wb_tcand_bytes += bytes; 6972a814908STejun Heo else 6982a814908STejun Heo wbc->wb_tcand_bytes -= min(bytes, wbc->wb_tcand_bytes); 6992a814908STejun Heo } 7005aa2a96bSTejun Heo EXPORT_SYMBOL_GPL(wbc_account_io); 7012a814908STejun Heo 7022a814908STejun Heo /** 703703c2708STejun Heo * inode_congested - test whether an inode is congested 70460292bccSTejun Heo * @inode: inode to test for congestion (may be NULL) 705703c2708STejun Heo * @cong_bits: mask of WB_[a]sync_congested bits to test 706703c2708STejun Heo * 707703c2708STejun Heo * Tests whether @inode is congested. @cong_bits is the mask of congestion 708703c2708STejun Heo * bits to test and the return value is the mask of set bits. 709703c2708STejun Heo * 710703c2708STejun Heo * If cgroup writeback is enabled for @inode, the congestion state is 711703c2708STejun Heo * determined by whether the cgwb (cgroup bdi_writeback) for the blkcg 712703c2708STejun Heo * associated with @inode is congested; otherwise, the root wb's congestion 713703c2708STejun Heo * state is used. 71460292bccSTejun Heo * 71560292bccSTejun Heo * @inode is allowed to be NULL as this function is often called on 71660292bccSTejun Heo * mapping->host which is NULL for the swapper space. 717703c2708STejun Heo */ 718703c2708STejun Heo int inode_congested(struct inode *inode, int cong_bits) 719703c2708STejun Heo { 7205cb8b824STejun Heo /* 7215cb8b824STejun Heo * Once set, ->i_wb never becomes NULL while the inode is alive. 7225cb8b824STejun Heo * Start transaction iff ->i_wb is visible. 7235cb8b824STejun Heo */ 724aaa2cacfSTejun Heo if (inode && inode_to_wb_is_valid(inode)) { 7255cb8b824STejun Heo struct bdi_writeback *wb; 7265cb8b824STejun Heo bool locked, congested; 7275cb8b824STejun Heo 7285cb8b824STejun Heo wb = unlocked_inode_to_wb_begin(inode, &locked); 7295cb8b824STejun Heo congested = wb_congested(wb, cong_bits); 7305cb8b824STejun Heo unlocked_inode_to_wb_end(inode, locked); 7315cb8b824STejun Heo return congested; 732703c2708STejun Heo } 733703c2708STejun Heo 734703c2708STejun Heo return wb_congested(&inode_to_bdi(inode)->wb, cong_bits); 735703c2708STejun Heo } 736703c2708STejun Heo EXPORT_SYMBOL_GPL(inode_congested); 737703c2708STejun Heo 738f2b65121STejun Heo /** 739f2b65121STejun Heo * wb_split_bdi_pages - split nr_pages to write according to bandwidth 740f2b65121STejun Heo * @wb: target bdi_writeback to split @nr_pages to 741f2b65121STejun Heo * @nr_pages: number of pages to write for the whole bdi 742f2b65121STejun Heo * 743f2b65121STejun Heo * Split @wb's portion of @nr_pages according to @wb's write bandwidth in 744f2b65121STejun Heo * relation to the total write bandwidth of all wb's w/ dirty inodes on 745f2b65121STejun Heo * @wb->bdi. 746f2b65121STejun Heo */ 747f2b65121STejun Heo static long wb_split_bdi_pages(struct bdi_writeback *wb, long nr_pages) 748f2b65121STejun Heo { 749f2b65121STejun Heo unsigned long this_bw = wb->avg_write_bandwidth; 750f2b65121STejun Heo unsigned long tot_bw = atomic_long_read(&wb->bdi->tot_write_bandwidth); 751f2b65121STejun Heo 752f2b65121STejun Heo if (nr_pages == LONG_MAX) 753f2b65121STejun Heo return LONG_MAX; 754f2b65121STejun Heo 755f2b65121STejun Heo /* 756f2b65121STejun Heo * This may be called on clean wb's and proportional distribution 757f2b65121STejun Heo * may not make sense, just use the original @nr_pages in those 758f2b65121STejun Heo * cases. In general, we wanna err on the side of writing more. 759f2b65121STejun Heo */ 760f2b65121STejun Heo if (!tot_bw || this_bw >= tot_bw) 761f2b65121STejun Heo return nr_pages; 762f2b65121STejun Heo else 763f2b65121STejun Heo return DIV_ROUND_UP_ULL((u64)nr_pages * this_bw, tot_bw); 764f2b65121STejun Heo } 765f2b65121STejun Heo 766db125360STejun Heo /** 767db125360STejun Heo * bdi_split_work_to_wbs - split a wb_writeback_work to all wb's of a bdi 768db125360STejun Heo * @bdi: target backing_dev_info 769db125360STejun Heo * @base_work: wb_writeback_work to issue 770db125360STejun Heo * @skip_if_busy: skip wb's which already have writeback in progress 771db125360STejun Heo * 772db125360STejun Heo * Split and issue @base_work to all wb's (bdi_writeback's) of @bdi which 773db125360STejun Heo * have dirty inodes. If @base_work->nr_page isn't %LONG_MAX, it's 774db125360STejun Heo * distributed to the busy wbs according to each wb's proportion in the 775db125360STejun Heo * total active write bandwidth of @bdi. 776db125360STejun Heo */ 777db125360STejun Heo static void bdi_split_work_to_wbs(struct backing_dev_info *bdi, 778db125360STejun Heo struct wb_writeback_work *base_work, 779db125360STejun Heo bool skip_if_busy) 780db125360STejun Heo { 781b817525aSTejun Heo struct bdi_writeback *last_wb = NULL; 782b33e18f6STejun Heo struct bdi_writeback *wb = list_entry(&bdi->wb_list, 783b817525aSTejun Heo struct bdi_writeback, bdi_node); 784db125360STejun Heo 785db125360STejun Heo might_sleep(); 786db125360STejun Heo restart: 787db125360STejun Heo rcu_read_lock(); 788b817525aSTejun Heo list_for_each_entry_continue_rcu(wb, &bdi->wb_list, bdi_node) { 7898a1270cdSTejun Heo DEFINE_WB_COMPLETION_ONSTACK(fallback_work_done); 7908a1270cdSTejun Heo struct wb_writeback_work fallback_work; 7918a1270cdSTejun Heo struct wb_writeback_work *work; 7928a1270cdSTejun Heo long nr_pages; 7938a1270cdSTejun Heo 794b817525aSTejun Heo if (last_wb) { 795b817525aSTejun Heo wb_put(last_wb); 796b817525aSTejun Heo last_wb = NULL; 797b817525aSTejun Heo } 798b817525aSTejun Heo 799006a0973STejun Heo /* SYNC_ALL writes out I_DIRTY_TIME too */ 800006a0973STejun Heo if (!wb_has_dirty_io(wb) && 801006a0973STejun Heo (base_work->sync_mode == WB_SYNC_NONE || 802006a0973STejun Heo list_empty(&wb->b_dirty_time))) 803006a0973STejun Heo continue; 804006a0973STejun Heo if (skip_if_busy && writeback_in_progress(wb)) 805db125360STejun Heo continue; 806db125360STejun Heo 8078a1270cdSTejun Heo nr_pages = wb_split_bdi_pages(wb, base_work->nr_pages); 8088a1270cdSTejun Heo 8098a1270cdSTejun Heo work = kmalloc(sizeof(*work), GFP_ATOMIC); 8108a1270cdSTejun Heo if (work) { 8118a1270cdSTejun Heo *work = *base_work; 8128a1270cdSTejun Heo work->nr_pages = nr_pages; 8138a1270cdSTejun Heo work->auto_free = 1; 8148a1270cdSTejun Heo wb_queue_work(wb, work); 8158a1270cdSTejun Heo continue; 816db125360STejun Heo } 8178a1270cdSTejun Heo 8188a1270cdSTejun Heo /* alloc failed, execute synchronously using on-stack fallback */ 8198a1270cdSTejun Heo work = &fallback_work; 8208a1270cdSTejun Heo *work = *base_work; 8218a1270cdSTejun Heo work->nr_pages = nr_pages; 8228a1270cdSTejun Heo work->auto_free = 0; 8238a1270cdSTejun Heo work->done = &fallback_work_done; 8248a1270cdSTejun Heo 8258a1270cdSTejun Heo wb_queue_work(wb, work); 8268a1270cdSTejun Heo 827b817525aSTejun Heo /* 828b817525aSTejun Heo * Pin @wb so that it stays on @bdi->wb_list. This allows 829b817525aSTejun Heo * continuing iteration from @wb after dropping and 830b817525aSTejun Heo * regrabbing rcu read lock. 831b817525aSTejun Heo */ 832b817525aSTejun Heo wb_get(wb); 833b817525aSTejun Heo last_wb = wb; 834b817525aSTejun Heo 835db125360STejun Heo rcu_read_unlock(); 8368a1270cdSTejun Heo wb_wait_for_completion(bdi, &fallback_work_done); 837db125360STejun Heo goto restart; 838db125360STejun Heo } 839db125360STejun Heo rcu_read_unlock(); 840b817525aSTejun Heo 841b817525aSTejun Heo if (last_wb) 842b817525aSTejun Heo wb_put(last_wb); 843db125360STejun Heo } 844db125360STejun Heo 845f2b65121STejun Heo #else /* CONFIG_CGROUP_WRITEBACK */ 846f2b65121STejun Heo 84787e1d789STejun Heo static struct bdi_writeback * 84887e1d789STejun Heo locked_inode_to_wb_and_lock_list(struct inode *inode) 84987e1d789STejun Heo __releases(&inode->i_lock) 85087e1d789STejun Heo __acquires(&wb->list_lock) 85187e1d789STejun Heo { 85287e1d789STejun Heo struct bdi_writeback *wb = inode_to_wb(inode); 85387e1d789STejun Heo 85487e1d789STejun Heo spin_unlock(&inode->i_lock); 85587e1d789STejun Heo spin_lock(&wb->list_lock); 85687e1d789STejun Heo return wb; 85787e1d789STejun Heo } 85887e1d789STejun Heo 85987e1d789STejun Heo static struct bdi_writeback *inode_to_wb_and_lock_list(struct inode *inode) 86087e1d789STejun Heo __acquires(&wb->list_lock) 86187e1d789STejun Heo { 86287e1d789STejun Heo struct bdi_writeback *wb = inode_to_wb(inode); 86387e1d789STejun Heo 86487e1d789STejun Heo spin_lock(&wb->list_lock); 86587e1d789STejun Heo return wb; 86687e1d789STejun Heo } 86787e1d789STejun Heo 868f2b65121STejun Heo static long wb_split_bdi_pages(struct bdi_writeback *wb, long nr_pages) 869f2b65121STejun Heo { 870f2b65121STejun Heo return nr_pages; 871f2b65121STejun Heo } 872f2b65121STejun Heo 873db125360STejun Heo static void bdi_split_work_to_wbs(struct backing_dev_info *bdi, 874db125360STejun Heo struct wb_writeback_work *base_work, 875db125360STejun Heo bool skip_if_busy) 876db125360STejun Heo { 877db125360STejun Heo might_sleep(); 878db125360STejun Heo 879006a0973STejun Heo if (!skip_if_busy || !writeback_in_progress(&bdi->wb)) { 880db125360STejun Heo base_work->auto_free = 0; 881db125360STejun Heo wb_queue_work(&bdi->wb, base_work); 882db125360STejun Heo } 883db125360STejun Heo } 884db125360STejun Heo 885703c2708STejun Heo #endif /* CONFIG_CGROUP_WRITEBACK */ 886703c2708STejun Heo 887c00ddad3STejun Heo void wb_start_writeback(struct bdi_writeback *wb, long nr_pages, 888c00ddad3STejun Heo bool range_cyclic, enum wb_reason reason) 889b6e51316SJens Axboe { 890c00ddad3STejun Heo struct wb_writeback_work *work; 891c00ddad3STejun Heo 892c00ddad3STejun Heo if (!wb_has_dirty_io(wb)) 893c00ddad3STejun Heo return; 894c00ddad3STejun Heo 895c00ddad3STejun Heo /* 896c00ddad3STejun Heo * This is WB_SYNC_NONE writeback, so if allocation fails just 897c00ddad3STejun Heo * wakeup the thread for old dirty data writeback 898c00ddad3STejun Heo */ 899c00ddad3STejun Heo work = kzalloc(sizeof(*work), GFP_ATOMIC); 900c00ddad3STejun Heo if (!work) { 9015634cc2aSTejun Heo trace_writeback_nowork(wb); 902c00ddad3STejun Heo wb_wakeup(wb); 903c00ddad3STejun Heo return; 904c00ddad3STejun Heo } 905c00ddad3STejun Heo 906c00ddad3STejun Heo work->sync_mode = WB_SYNC_NONE; 907c00ddad3STejun Heo work->nr_pages = nr_pages; 908c00ddad3STejun Heo work->range_cyclic = range_cyclic; 909c00ddad3STejun Heo work->reason = reason; 910ac7b19a3STejun Heo work->auto_free = 1; 911c00ddad3STejun Heo 912c00ddad3STejun Heo wb_queue_work(wb, work); 913d3ddec76SWu Fengguang } 914d3ddec76SWu Fengguang 915c5444198SChristoph Hellwig /** 9169ecf4866STejun Heo * wb_start_background_writeback - start background writeback 9179ecf4866STejun Heo * @wb: bdi_writback to write from 918c5444198SChristoph Hellwig * 919c5444198SChristoph Hellwig * Description: 9206585027aSJan Kara * This makes sure WB_SYNC_NONE background writeback happens. When 9219ecf4866STejun Heo * this function returns, it is only guaranteed that for given wb 9226585027aSJan Kara * some IO is happening if we are over background dirty threshold. 9236585027aSJan Kara * Caller need not hold sb s_umount semaphore. 924c5444198SChristoph Hellwig */ 9259ecf4866STejun Heo void wb_start_background_writeback(struct bdi_writeback *wb) 926c5444198SChristoph Hellwig { 9276585027aSJan Kara /* 9286585027aSJan Kara * We just wake up the flusher thread. It will perform background 9296585027aSJan Kara * writeback as soon as there is no other work to do. 9306585027aSJan Kara */ 9315634cc2aSTejun Heo trace_writeback_wake_background(wb); 9329ecf4866STejun Heo wb_wakeup(wb); 9331da177e4SLinus Torvalds } 9341da177e4SLinus Torvalds 9351da177e4SLinus Torvalds /* 936a66979abSDave Chinner * Remove the inode from the writeback list it is on. 937a66979abSDave Chinner */ 938c7f54084SDave Chinner void inode_io_list_del(struct inode *inode) 939a66979abSDave Chinner { 94087e1d789STejun Heo struct bdi_writeback *wb; 941a66979abSDave Chinner 94287e1d789STejun Heo wb = inode_to_wb_and_lock_list(inode); 943c7f54084SDave Chinner inode_io_list_del_locked(inode, wb); 94452ebea74STejun Heo spin_unlock(&wb->list_lock); 945f758eeabSChristoph Hellwig } 946a66979abSDave Chinner 947a66979abSDave Chinner /* 9486610a0bcSAndrew Morton * Redirty an inode: set its when-it-was dirtied timestamp and move it to the 9496610a0bcSAndrew Morton * furthest end of its superblock's dirty-inode list. 9506610a0bcSAndrew Morton * 9516610a0bcSAndrew Morton * Before stamping the inode's ->dirtied_when, we check to see whether it is 95266f3b8e2SJens Axboe * already the most-recently-dirtied inode on the b_dirty list. If that is 9536610a0bcSAndrew Morton * the case then the inode must have been redirtied while it was being written 9546610a0bcSAndrew Morton * out and we don't reset its dirtied_when. 9556610a0bcSAndrew Morton */ 956f758eeabSChristoph Hellwig static void redirty_tail(struct inode *inode, struct bdi_writeback *wb) 9576610a0bcSAndrew Morton { 95803ba3782SJens Axboe if (!list_empty(&wb->b_dirty)) { 95966f3b8e2SJens Axboe struct inode *tail; 9606610a0bcSAndrew Morton 9617ccf19a8SNick Piggin tail = wb_inode(wb->b_dirty.next); 96266f3b8e2SJens Axboe if (time_before(inode->dirtied_when, tail->dirtied_when)) 9636610a0bcSAndrew Morton inode->dirtied_when = jiffies; 9646610a0bcSAndrew Morton } 965c7f54084SDave Chinner inode_io_list_move_locked(inode, wb, &wb->b_dirty); 9666610a0bcSAndrew Morton } 9676610a0bcSAndrew Morton 9686610a0bcSAndrew Morton /* 96966f3b8e2SJens Axboe * requeue inode for re-scanning after bdi->b_io list is exhausted. 970c986d1e2SAndrew Morton */ 971f758eeabSChristoph Hellwig static void requeue_io(struct inode *inode, struct bdi_writeback *wb) 972c986d1e2SAndrew Morton { 973c7f54084SDave Chinner inode_io_list_move_locked(inode, wb, &wb->b_more_io); 974c986d1e2SAndrew Morton } 975c986d1e2SAndrew Morton 9761c0eeaf5SJoern Engel static void inode_sync_complete(struct inode *inode) 9771c0eeaf5SJoern Engel { 978365b94aeSJan Kara inode->i_state &= ~I_SYNC; 9794eff96ddSJan Kara /* If inode is clean an unused, put it into LRU now... */ 9804eff96ddSJan Kara inode_add_lru(inode); 981365b94aeSJan Kara /* Waiters must see I_SYNC cleared before being woken up */ 9821c0eeaf5SJoern Engel smp_mb(); 9831c0eeaf5SJoern Engel wake_up_bit(&inode->i_state, __I_SYNC); 9841c0eeaf5SJoern Engel } 9851c0eeaf5SJoern Engel 986d2caa3c5SJeff Layton static bool inode_dirtied_after(struct inode *inode, unsigned long t) 987d2caa3c5SJeff Layton { 988d2caa3c5SJeff Layton bool ret = time_after(inode->dirtied_when, t); 989d2caa3c5SJeff Layton #ifndef CONFIG_64BIT 990d2caa3c5SJeff Layton /* 991d2caa3c5SJeff Layton * For inodes being constantly redirtied, dirtied_when can get stuck. 992d2caa3c5SJeff Layton * It _appears_ to be in the future, but is actually in distant past. 993d2caa3c5SJeff Layton * This test is necessary to prevent such wrapped-around relative times 9945b0830cbSJens Axboe * from permanently stopping the whole bdi writeback. 995d2caa3c5SJeff Layton */ 996d2caa3c5SJeff Layton ret = ret && time_before_eq(inode->dirtied_when, jiffies); 997d2caa3c5SJeff Layton #endif 998d2caa3c5SJeff Layton return ret; 999d2caa3c5SJeff Layton } 1000d2caa3c5SJeff Layton 10010ae45f63STheodore Ts'o #define EXPIRE_DIRTY_ATIME 0x0001 10020ae45f63STheodore Ts'o 1003c986d1e2SAndrew Morton /* 10040e2f2b23SWang Sheng-Hui * Move expired (dirtied before work->older_than_this) dirty inodes from 1005697e6fedSJan Kara * @delaying_queue to @dispatch_queue. 10062c136579SFengguang Wu */ 1007e84d0a4fSWu Fengguang static int move_expired_inodes(struct list_head *delaying_queue, 10082c136579SFengguang Wu struct list_head *dispatch_queue, 10090ae45f63STheodore Ts'o int flags, 1010ad4e38ddSCurt Wohlgemuth struct wb_writeback_work *work) 10112c136579SFengguang Wu { 10120ae45f63STheodore Ts'o unsigned long *older_than_this = NULL; 10130ae45f63STheodore Ts'o unsigned long expire_time; 10145c03449dSShaohua Li LIST_HEAD(tmp); 10155c03449dSShaohua Li struct list_head *pos, *node; 1016cf137307SJens Axboe struct super_block *sb = NULL; 10175c03449dSShaohua Li struct inode *inode; 1018cf137307SJens Axboe int do_sb_sort = 0; 1019e84d0a4fSWu Fengguang int moved = 0; 10205c03449dSShaohua Li 10210ae45f63STheodore Ts'o if ((flags & EXPIRE_DIRTY_ATIME) == 0) 10220ae45f63STheodore Ts'o older_than_this = work->older_than_this; 1023a2f48706STheodore Ts'o else if (!work->for_sync) { 1024a2f48706STheodore Ts'o expire_time = jiffies - (dirtytime_expire_interval * HZ); 10250ae45f63STheodore Ts'o older_than_this = &expire_time; 10260ae45f63STheodore Ts'o } 10272c136579SFengguang Wu while (!list_empty(delaying_queue)) { 10287ccf19a8SNick Piggin inode = wb_inode(delaying_queue->prev); 10290ae45f63STheodore Ts'o if (older_than_this && 10300ae45f63STheodore Ts'o inode_dirtied_after(inode, *older_than_this)) 10312c136579SFengguang Wu break; 1032c7f54084SDave Chinner list_move(&inode->i_io_list, &tmp); 1033a8855990SJan Kara moved++; 10340ae45f63STheodore Ts'o if (flags & EXPIRE_DIRTY_ATIME) 10350ae45f63STheodore Ts'o set_bit(__I_DIRTY_TIME_EXPIRED, &inode->i_state); 1036a8855990SJan Kara if (sb_is_blkdev_sb(inode->i_sb)) 1037a8855990SJan Kara continue; 1038cf137307SJens Axboe if (sb && sb != inode->i_sb) 1039cf137307SJens Axboe do_sb_sort = 1; 1040cf137307SJens Axboe sb = inode->i_sb; 10415c03449dSShaohua Li } 10425c03449dSShaohua Li 1043cf137307SJens Axboe /* just one sb in list, splice to dispatch_queue and we're done */ 1044cf137307SJens Axboe if (!do_sb_sort) { 1045cf137307SJens Axboe list_splice(&tmp, dispatch_queue); 1046e84d0a4fSWu Fengguang goto out; 1047cf137307SJens Axboe } 1048cf137307SJens Axboe 10495c03449dSShaohua Li /* Move inodes from one superblock together */ 10505c03449dSShaohua Li while (!list_empty(&tmp)) { 10517ccf19a8SNick Piggin sb = wb_inode(tmp.prev)->i_sb; 10525c03449dSShaohua Li list_for_each_prev_safe(pos, node, &tmp) { 10537ccf19a8SNick Piggin inode = wb_inode(pos); 10545c03449dSShaohua Li if (inode->i_sb == sb) 1055c7f54084SDave Chinner list_move(&inode->i_io_list, dispatch_queue); 10562c136579SFengguang Wu } 10572c136579SFengguang Wu } 1058e84d0a4fSWu Fengguang out: 1059e84d0a4fSWu Fengguang return moved; 10605c03449dSShaohua Li } 10612c136579SFengguang Wu 10622c136579SFengguang Wu /* 10632c136579SFengguang Wu * Queue all expired dirty inodes for io, eldest first. 10644ea879b9SWu Fengguang * Before 10654ea879b9SWu Fengguang * newly dirtied b_dirty b_io b_more_io 10664ea879b9SWu Fengguang * =============> gf edc BA 10674ea879b9SWu Fengguang * After 10684ea879b9SWu Fengguang * newly dirtied b_dirty b_io b_more_io 10694ea879b9SWu Fengguang * =============> g fBAedc 10704ea879b9SWu Fengguang * | 10714ea879b9SWu Fengguang * +--> dequeue for IO 10722c136579SFengguang Wu */ 1073ad4e38ddSCurt Wohlgemuth static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work) 10742c136579SFengguang Wu { 1075e84d0a4fSWu Fengguang int moved; 10760ae45f63STheodore Ts'o 1077f758eeabSChristoph Hellwig assert_spin_locked(&wb->list_lock); 10784ea879b9SWu Fengguang list_splice_init(&wb->b_more_io, &wb->b_io); 10790ae45f63STheodore Ts'o moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, 0, work); 10800ae45f63STheodore Ts'o moved += move_expired_inodes(&wb->b_dirty_time, &wb->b_io, 10810ae45f63STheodore Ts'o EXPIRE_DIRTY_ATIME, work); 1082d6c10f1fSTejun Heo if (moved) 1083d6c10f1fSTejun Heo wb_io_lists_populated(wb); 1084ad4e38ddSCurt Wohlgemuth trace_writeback_queue_io(wb, work, moved); 108566f3b8e2SJens Axboe } 108666f3b8e2SJens Axboe 1087a9185b41SChristoph Hellwig static int write_inode(struct inode *inode, struct writeback_control *wbc) 108866f3b8e2SJens Axboe { 10899fb0a7daSTejun Heo int ret; 10909fb0a7daSTejun Heo 10919fb0a7daSTejun Heo if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) { 10929fb0a7daSTejun Heo trace_writeback_write_inode_start(inode, wbc); 10939fb0a7daSTejun Heo ret = inode->i_sb->s_op->write_inode(inode, wbc); 10949fb0a7daSTejun Heo trace_writeback_write_inode(inode, wbc); 10959fb0a7daSTejun Heo return ret; 10969fb0a7daSTejun Heo } 109703ba3782SJens Axboe return 0; 109866f3b8e2SJens Axboe } 109908d8e974SFengguang Wu 11002c136579SFengguang Wu /* 1101169ebd90SJan Kara * Wait for writeback on an inode to complete. Called with i_lock held. 1102169ebd90SJan Kara * Caller must make sure inode cannot go away when we drop i_lock. 110301c03194SChristoph Hellwig */ 1104169ebd90SJan Kara static void __inode_wait_for_writeback(struct inode *inode) 1105169ebd90SJan Kara __releases(inode->i_lock) 1106169ebd90SJan Kara __acquires(inode->i_lock) 110701c03194SChristoph Hellwig { 110801c03194SChristoph Hellwig DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); 110901c03194SChristoph Hellwig wait_queue_head_t *wqh; 111001c03194SChristoph Hellwig 111101c03194SChristoph Hellwig wqh = bit_waitqueue(&inode->i_state, __I_SYNC); 111258a9d3d8SRichard Kennedy while (inode->i_state & I_SYNC) { 1113250df6edSDave Chinner spin_unlock(&inode->i_lock); 111474316201SNeilBrown __wait_on_bit(wqh, &wq, bit_wait, 111574316201SNeilBrown TASK_UNINTERRUPTIBLE); 1116250df6edSDave Chinner spin_lock(&inode->i_lock); 111758a9d3d8SRichard Kennedy } 111801c03194SChristoph Hellwig } 111901c03194SChristoph Hellwig 112001c03194SChristoph Hellwig /* 1121169ebd90SJan Kara * Wait for writeback on an inode to complete. Caller must have inode pinned. 1122169ebd90SJan Kara */ 1123169ebd90SJan Kara void inode_wait_for_writeback(struct inode *inode) 1124169ebd90SJan Kara { 1125169ebd90SJan Kara spin_lock(&inode->i_lock); 1126169ebd90SJan Kara __inode_wait_for_writeback(inode); 1127169ebd90SJan Kara spin_unlock(&inode->i_lock); 1128169ebd90SJan Kara } 1129169ebd90SJan Kara 1130169ebd90SJan Kara /* 1131169ebd90SJan Kara * Sleep until I_SYNC is cleared. This function must be called with i_lock 1132169ebd90SJan Kara * held and drops it. It is aimed for callers not holding any inode reference 1133169ebd90SJan Kara * so once i_lock is dropped, inode can go away. 1134169ebd90SJan Kara */ 1135169ebd90SJan Kara static void inode_sleep_on_writeback(struct inode *inode) 1136169ebd90SJan Kara __releases(inode->i_lock) 1137169ebd90SJan Kara { 1138169ebd90SJan Kara DEFINE_WAIT(wait); 1139169ebd90SJan Kara wait_queue_head_t *wqh = bit_waitqueue(&inode->i_state, __I_SYNC); 1140169ebd90SJan Kara int sleep; 1141169ebd90SJan Kara 1142169ebd90SJan Kara prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); 1143169ebd90SJan Kara sleep = inode->i_state & I_SYNC; 1144169ebd90SJan Kara spin_unlock(&inode->i_lock); 1145169ebd90SJan Kara if (sleep) 1146169ebd90SJan Kara schedule(); 1147169ebd90SJan Kara finish_wait(wqh, &wait); 1148169ebd90SJan Kara } 1149169ebd90SJan Kara 1150169ebd90SJan Kara /* 1151ccb26b5aSJan Kara * Find proper writeback list for the inode depending on its current state and 1152ccb26b5aSJan Kara * possibly also change of its state while we were doing writeback. Here we 1153ccb26b5aSJan Kara * handle things such as livelock prevention or fairness of writeback among 1154ccb26b5aSJan Kara * inodes. This function can be called only by flusher thread - noone else 1155ccb26b5aSJan Kara * processes all inodes in writeback lists and requeueing inodes behind flusher 1156ccb26b5aSJan Kara * thread's back can have unexpected consequences. 1157ccb26b5aSJan Kara */ 1158ccb26b5aSJan Kara static void requeue_inode(struct inode *inode, struct bdi_writeback *wb, 1159ccb26b5aSJan Kara struct writeback_control *wbc) 1160ccb26b5aSJan Kara { 1161ccb26b5aSJan Kara if (inode->i_state & I_FREEING) 1162ccb26b5aSJan Kara return; 1163ccb26b5aSJan Kara 1164ccb26b5aSJan Kara /* 1165ccb26b5aSJan Kara * Sync livelock prevention. Each inode is tagged and synced in one 1166ccb26b5aSJan Kara * shot. If still dirty, it will be redirty_tail()'ed below. Update 1167ccb26b5aSJan Kara * the dirty time to prevent enqueue and sync it again. 1168ccb26b5aSJan Kara */ 1169ccb26b5aSJan Kara if ((inode->i_state & I_DIRTY) && 1170ccb26b5aSJan Kara (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)) 1171ccb26b5aSJan Kara inode->dirtied_when = jiffies; 1172ccb26b5aSJan Kara 11734f8ad655SJan Kara if (wbc->pages_skipped) { 11744f8ad655SJan Kara /* 11754f8ad655SJan Kara * writeback is not making progress due to locked 11764f8ad655SJan Kara * buffers. Skip this inode for now. 11774f8ad655SJan Kara */ 11784f8ad655SJan Kara redirty_tail(inode, wb); 11794f8ad655SJan Kara return; 11804f8ad655SJan Kara } 11814f8ad655SJan Kara 1182ccb26b5aSJan Kara if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) { 1183ccb26b5aSJan Kara /* 1184ccb26b5aSJan Kara * We didn't write back all the pages. nfs_writepages() 1185ccb26b5aSJan Kara * sometimes bales out without doing anything. 1186ccb26b5aSJan Kara */ 1187ccb26b5aSJan Kara if (wbc->nr_to_write <= 0) { 1188ccb26b5aSJan Kara /* Slice used up. Queue for next turn. */ 1189ccb26b5aSJan Kara requeue_io(inode, wb); 1190ccb26b5aSJan Kara } else { 1191ccb26b5aSJan Kara /* 1192ccb26b5aSJan Kara * Writeback blocked by something other than 1193ccb26b5aSJan Kara * congestion. Delay the inode for some time to 1194ccb26b5aSJan Kara * avoid spinning on the CPU (100% iowait) 1195ccb26b5aSJan Kara * retrying writeback of the dirty page/inode 1196ccb26b5aSJan Kara * that cannot be performed immediately. 1197ccb26b5aSJan Kara */ 1198ccb26b5aSJan Kara redirty_tail(inode, wb); 1199ccb26b5aSJan Kara } 1200ccb26b5aSJan Kara } else if (inode->i_state & I_DIRTY) { 1201ccb26b5aSJan Kara /* 1202ccb26b5aSJan Kara * Filesystems can dirty the inode during writeback operations, 1203ccb26b5aSJan Kara * such as delayed allocation during submission or metadata 1204ccb26b5aSJan Kara * updates after data IO completion. 1205ccb26b5aSJan Kara */ 1206ccb26b5aSJan Kara redirty_tail(inode, wb); 12070ae45f63STheodore Ts'o } else if (inode->i_state & I_DIRTY_TIME) { 1208a2f48706STheodore Ts'o inode->dirtied_when = jiffies; 1209c7f54084SDave Chinner inode_io_list_move_locked(inode, wb, &wb->b_dirty_time); 1210ccb26b5aSJan Kara } else { 1211ccb26b5aSJan Kara /* The inode is clean. Remove from writeback lists. */ 1212c7f54084SDave Chinner inode_io_list_del_locked(inode, wb); 1213ccb26b5aSJan Kara } 1214ccb26b5aSJan Kara } 1215ccb26b5aSJan Kara 1216ccb26b5aSJan Kara /* 12174f8ad655SJan Kara * Write out an inode and its dirty pages. Do not update the writeback list 12184f8ad655SJan Kara * linkage. That is left to the caller. The caller is also responsible for 12194f8ad655SJan Kara * setting I_SYNC flag and calling inode_sync_complete() to clear it. 12201da177e4SLinus Torvalds */ 12211da177e4SLinus Torvalds static int 1222cd8ed2a4SYan Hong __writeback_single_inode(struct inode *inode, struct writeback_control *wbc) 12231da177e4SLinus Torvalds { 12241da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 1225251d6a47SWu Fengguang long nr_to_write = wbc->nr_to_write; 122601c03194SChristoph Hellwig unsigned dirty; 12271da177e4SLinus Torvalds int ret; 12281da177e4SLinus Torvalds 12294f8ad655SJan Kara WARN_ON(!(inode->i_state & I_SYNC)); 12301da177e4SLinus Torvalds 12319fb0a7daSTejun Heo trace_writeback_single_inode_start(inode, wbc, nr_to_write); 12329fb0a7daSTejun Heo 12331da177e4SLinus Torvalds ret = do_writepages(mapping, wbc); 12341da177e4SLinus Torvalds 123526821ed4SChristoph Hellwig /* 123626821ed4SChristoph Hellwig * Make sure to wait on the data before writing out the metadata. 123726821ed4SChristoph Hellwig * This is important for filesystems that modify metadata on data 12387747bd4bSDave Chinner * I/O completion. We don't do it for sync(2) writeback because it has a 12397747bd4bSDave Chinner * separate, external IO completion path and ->sync_fs for guaranteeing 12407747bd4bSDave Chinner * inode metadata is written back correctly. 124126821ed4SChristoph Hellwig */ 12427747bd4bSDave Chinner if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) { 124326821ed4SChristoph Hellwig int err = filemap_fdatawait(mapping); 12441da177e4SLinus Torvalds if (ret == 0) 12451da177e4SLinus Torvalds ret = err; 12461da177e4SLinus Torvalds } 12471da177e4SLinus Torvalds 12485547e8aaSDmitry Monakhov /* 12495547e8aaSDmitry Monakhov * Some filesystems may redirty the inode during the writeback 12505547e8aaSDmitry Monakhov * due to delalloc, clear dirty metadata flags right before 12515547e8aaSDmitry Monakhov * write_inode() 12525547e8aaSDmitry Monakhov */ 1253250df6edSDave Chinner spin_lock(&inode->i_lock); 12549c6ac78eSTejun Heo 12555547e8aaSDmitry Monakhov dirty = inode->i_state & I_DIRTY; 1256a2f48706STheodore Ts'o if (inode->i_state & I_DIRTY_TIME) { 1257a2f48706STheodore Ts'o if ((dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) || 1258a2f48706STheodore Ts'o unlikely(inode->i_state & I_DIRTY_TIME_EXPIRED) || 1259a2f48706STheodore Ts'o unlikely(time_after(jiffies, 1260a2f48706STheodore Ts'o (inode->dirtied_time_when + 1261a2f48706STheodore Ts'o dirtytime_expire_interval * HZ)))) { 12620ae45f63STheodore Ts'o dirty |= I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED; 12630ae45f63STheodore Ts'o trace_writeback_lazytime(inode); 12640ae45f63STheodore Ts'o } 1265a2f48706STheodore Ts'o } else 1266a2f48706STheodore Ts'o inode->i_state &= ~I_DIRTY_TIME_EXPIRED; 12670ae45f63STheodore Ts'o inode->i_state &= ~dirty; 12689c6ac78eSTejun Heo 12699c6ac78eSTejun Heo /* 12709c6ac78eSTejun Heo * Paired with smp_mb() in __mark_inode_dirty(). This allows 12719c6ac78eSTejun Heo * __mark_inode_dirty() to test i_state without grabbing i_lock - 12729c6ac78eSTejun Heo * either they see the I_DIRTY bits cleared or we see the dirtied 12739c6ac78eSTejun Heo * inode. 12749c6ac78eSTejun Heo * 12759c6ac78eSTejun Heo * I_DIRTY_PAGES is always cleared together above even if @mapping 12769c6ac78eSTejun Heo * still has dirty pages. The flag is reinstated after smp_mb() if 12779c6ac78eSTejun Heo * necessary. This guarantees that either __mark_inode_dirty() 12789c6ac78eSTejun Heo * sees clear I_DIRTY_PAGES or we see PAGECACHE_TAG_DIRTY. 12799c6ac78eSTejun Heo */ 12809c6ac78eSTejun Heo smp_mb(); 12819c6ac78eSTejun Heo 12829c6ac78eSTejun Heo if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) 12839c6ac78eSTejun Heo inode->i_state |= I_DIRTY_PAGES; 12849c6ac78eSTejun Heo 1285250df6edSDave Chinner spin_unlock(&inode->i_lock); 12869c6ac78eSTejun Heo 12870ae45f63STheodore Ts'o if (dirty & I_DIRTY_TIME) 12880ae45f63STheodore Ts'o mark_inode_dirty_sync(inode); 128926821ed4SChristoph Hellwig /* Don't write the inode if only I_DIRTY_PAGES was set */ 12900ae45f63STheodore Ts'o if (dirty & ~I_DIRTY_PAGES) { 1291a9185b41SChristoph Hellwig int err = write_inode(inode, wbc); 12921da177e4SLinus Torvalds if (ret == 0) 12931da177e4SLinus Torvalds ret = err; 12941da177e4SLinus Torvalds } 12954f8ad655SJan Kara trace_writeback_single_inode(inode, wbc, nr_to_write); 12964f8ad655SJan Kara return ret; 12974f8ad655SJan Kara } 12984f8ad655SJan Kara 12994f8ad655SJan Kara /* 13004f8ad655SJan Kara * Write out an inode's dirty pages. Either the caller has an active reference 13014f8ad655SJan Kara * on the inode or the inode has I_WILL_FREE set. 13024f8ad655SJan Kara * 13034f8ad655SJan Kara * This function is designed to be called for writing back one inode which 13044f8ad655SJan Kara * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode() 13054f8ad655SJan Kara * and does more profound writeback list handling in writeback_sb_inodes(). 13064f8ad655SJan Kara */ 13074f8ad655SJan Kara static int 13084f8ad655SJan Kara writeback_single_inode(struct inode *inode, struct bdi_writeback *wb, 13094f8ad655SJan Kara struct writeback_control *wbc) 13104f8ad655SJan Kara { 13114f8ad655SJan Kara int ret = 0; 13124f8ad655SJan Kara 13134f8ad655SJan Kara spin_lock(&inode->i_lock); 13144f8ad655SJan Kara if (!atomic_read(&inode->i_count)) 13154f8ad655SJan Kara WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); 13164f8ad655SJan Kara else 13174f8ad655SJan Kara WARN_ON(inode->i_state & I_WILL_FREE); 13184f8ad655SJan Kara 13194f8ad655SJan Kara if (inode->i_state & I_SYNC) { 13204f8ad655SJan Kara if (wbc->sync_mode != WB_SYNC_ALL) 13214f8ad655SJan Kara goto out; 13224f8ad655SJan Kara /* 1323169ebd90SJan Kara * It's a data-integrity sync. We must wait. Since callers hold 1324169ebd90SJan Kara * inode reference or inode has I_WILL_FREE set, it cannot go 1325169ebd90SJan Kara * away under us. 13264f8ad655SJan Kara */ 1327169ebd90SJan Kara __inode_wait_for_writeback(inode); 13284f8ad655SJan Kara } 13294f8ad655SJan Kara WARN_ON(inode->i_state & I_SYNC); 13304f8ad655SJan Kara /* 1331f9b0e058SJan Kara * Skip inode if it is clean and we have no outstanding writeback in 1332f9b0e058SJan Kara * WB_SYNC_ALL mode. We don't want to mess with writeback lists in this 1333f9b0e058SJan Kara * function since flusher thread may be doing for example sync in 1334f9b0e058SJan Kara * parallel and if we move the inode, it could get skipped. So here we 1335f9b0e058SJan Kara * make sure inode is on some writeback list and leave it there unless 1336f9b0e058SJan Kara * we have completely cleaned the inode. 13374f8ad655SJan Kara */ 13380ae45f63STheodore Ts'o if (!(inode->i_state & I_DIRTY_ALL) && 1339f9b0e058SJan Kara (wbc->sync_mode != WB_SYNC_ALL || 1340f9b0e058SJan Kara !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK))) 13414f8ad655SJan Kara goto out; 13424f8ad655SJan Kara inode->i_state |= I_SYNC; 1343b16b1debSTejun Heo wbc_attach_and_unlock_inode(wbc, inode); 13444f8ad655SJan Kara 1345cd8ed2a4SYan Hong ret = __writeback_single_inode(inode, wbc); 13461da177e4SLinus Torvalds 1347b16b1debSTejun Heo wbc_detach_inode(wbc); 1348f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 1349250df6edSDave Chinner spin_lock(&inode->i_lock); 13504f8ad655SJan Kara /* 13514f8ad655SJan Kara * If inode is clean, remove it from writeback lists. Otherwise don't 13524f8ad655SJan Kara * touch it. See comment above for explanation. 13534f8ad655SJan Kara */ 13540ae45f63STheodore Ts'o if (!(inode->i_state & I_DIRTY_ALL)) 1355c7f54084SDave Chinner inode_io_list_del_locked(inode, wb); 13564f8ad655SJan Kara spin_unlock(&wb->list_lock); 13571c0eeaf5SJoern Engel inode_sync_complete(inode); 13584f8ad655SJan Kara out: 13594f8ad655SJan Kara spin_unlock(&inode->i_lock); 13601da177e4SLinus Torvalds return ret; 13611da177e4SLinus Torvalds } 13621da177e4SLinus Torvalds 1363a88a341aSTejun Heo static long writeback_chunk_size(struct bdi_writeback *wb, 13641a12d8bdSWu Fengguang struct wb_writeback_work *work) 1365d46db3d5SWu Fengguang { 1366d46db3d5SWu Fengguang long pages; 1367d46db3d5SWu Fengguang 1368d46db3d5SWu Fengguang /* 1369d46db3d5SWu Fengguang * WB_SYNC_ALL mode does livelock avoidance by syncing dirty 1370d46db3d5SWu Fengguang * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX 1371d46db3d5SWu Fengguang * here avoids calling into writeback_inodes_wb() more than once. 1372d46db3d5SWu Fengguang * 1373d46db3d5SWu Fengguang * The intended call sequence for WB_SYNC_ALL writeback is: 1374d46db3d5SWu Fengguang * 1375d46db3d5SWu Fengguang * wb_writeback() 1376d46db3d5SWu Fengguang * writeback_sb_inodes() <== called only once 1377d46db3d5SWu Fengguang * write_cache_pages() <== called once for each inode 1378d46db3d5SWu Fengguang * (quickly) tag currently dirty pages 1379d46db3d5SWu Fengguang * (maybe slowly) sync all tagged pages 1380d46db3d5SWu Fengguang */ 1381d46db3d5SWu Fengguang if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages) 1382d46db3d5SWu Fengguang pages = LONG_MAX; 13831a12d8bdSWu Fengguang else { 1384a88a341aSTejun Heo pages = min(wb->avg_write_bandwidth / 2, 1385dcc25ae7STejun Heo global_wb_domain.dirty_limit / DIRTY_SCOPE); 13861a12d8bdSWu Fengguang pages = min(pages, work->nr_pages); 13871a12d8bdSWu Fengguang pages = round_down(pages + MIN_WRITEBACK_PAGES, 13881a12d8bdSWu Fengguang MIN_WRITEBACK_PAGES); 13891a12d8bdSWu Fengguang } 1390d46db3d5SWu Fengguang 1391d46db3d5SWu Fengguang return pages; 1392d46db3d5SWu Fengguang } 1393d46db3d5SWu Fengguang 139403ba3782SJens Axboe /* 1395f11c9c5cSEdward Shishkin * Write a portion of b_io inodes which belong to @sb. 1396edadfb10SChristoph Hellwig * 1397d46db3d5SWu Fengguang * Return the number of pages and/or inodes written. 13980ba13fd1SLinus Torvalds * 13990ba13fd1SLinus Torvalds * NOTE! This is called with wb->list_lock held, and will 14000ba13fd1SLinus Torvalds * unlock and relock that for each inode it ends up doing 14010ba13fd1SLinus Torvalds * IO for. 1402f11c9c5cSEdward Shishkin */ 1403d46db3d5SWu Fengguang static long writeback_sb_inodes(struct super_block *sb, 1404d46db3d5SWu Fengguang struct bdi_writeback *wb, 1405d46db3d5SWu Fengguang struct wb_writeback_work *work) 140603ba3782SJens Axboe { 1407d46db3d5SWu Fengguang struct writeback_control wbc = { 1408d46db3d5SWu Fengguang .sync_mode = work->sync_mode, 1409d46db3d5SWu Fengguang .tagged_writepages = work->tagged_writepages, 1410d46db3d5SWu Fengguang .for_kupdate = work->for_kupdate, 1411d46db3d5SWu Fengguang .for_background = work->for_background, 14127747bd4bSDave Chinner .for_sync = work->for_sync, 1413d46db3d5SWu Fengguang .range_cyclic = work->range_cyclic, 1414d46db3d5SWu Fengguang .range_start = 0, 1415d46db3d5SWu Fengguang .range_end = LLONG_MAX, 1416d46db3d5SWu Fengguang }; 1417d46db3d5SWu Fengguang unsigned long start_time = jiffies; 1418d46db3d5SWu Fengguang long write_chunk; 1419d46db3d5SWu Fengguang long wrote = 0; /* count both pages and inodes */ 1420d46db3d5SWu Fengguang 142103ba3782SJens Axboe while (!list_empty(&wb->b_io)) { 14227ccf19a8SNick Piggin struct inode *inode = wb_inode(wb->b_io.prev); 1423edadfb10SChristoph Hellwig 1424edadfb10SChristoph Hellwig if (inode->i_sb != sb) { 1425d46db3d5SWu Fengguang if (work->sb) { 1426edadfb10SChristoph Hellwig /* 1427edadfb10SChristoph Hellwig * We only want to write back data for this 1428edadfb10SChristoph Hellwig * superblock, move all inodes not belonging 1429edadfb10SChristoph Hellwig * to it back onto the dirty list. 1430edadfb10SChristoph Hellwig */ 1431f758eeabSChristoph Hellwig redirty_tail(inode, wb); 143266f3b8e2SJens Axboe continue; 143366f3b8e2SJens Axboe } 1434edadfb10SChristoph Hellwig 1435edadfb10SChristoph Hellwig /* 1436edadfb10SChristoph Hellwig * The inode belongs to a different superblock. 1437edadfb10SChristoph Hellwig * Bounce back to the caller to unpin this and 1438edadfb10SChristoph Hellwig * pin the next superblock. 1439edadfb10SChristoph Hellwig */ 1440d46db3d5SWu Fengguang break; 1441edadfb10SChristoph Hellwig } 1442edadfb10SChristoph Hellwig 14439843b76aSChristoph Hellwig /* 1444331cbdeeSWanpeng Li * Don't bother with new inodes or inodes being freed, first 1445331cbdeeSWanpeng Li * kind does not need periodic writeout yet, and for the latter 14469843b76aSChristoph Hellwig * kind writeout is handled by the freer. 14479843b76aSChristoph Hellwig */ 1448250df6edSDave Chinner spin_lock(&inode->i_lock); 14499843b76aSChristoph Hellwig if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { 1450250df6edSDave Chinner spin_unlock(&inode->i_lock); 1451fcc5c222SWu Fengguang redirty_tail(inode, wb); 14527ef0d737SNick Piggin continue; 14537ef0d737SNick Piggin } 1454cc1676d9SJan Kara if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) { 1455cc1676d9SJan Kara /* 1456cc1676d9SJan Kara * If this inode is locked for writeback and we are not 1457cc1676d9SJan Kara * doing writeback-for-data-integrity, move it to 1458cc1676d9SJan Kara * b_more_io so that writeback can proceed with the 1459cc1676d9SJan Kara * other inodes on s_io. 1460cc1676d9SJan Kara * 1461cc1676d9SJan Kara * We'll have another go at writing back this inode 1462cc1676d9SJan Kara * when we completed a full scan of b_io. 1463cc1676d9SJan Kara */ 1464cc1676d9SJan Kara spin_unlock(&inode->i_lock); 1465cc1676d9SJan Kara requeue_io(inode, wb); 1466cc1676d9SJan Kara trace_writeback_sb_inodes_requeue(inode); 1467cc1676d9SJan Kara continue; 1468cc1676d9SJan Kara } 1469f0d07b7fSJan Kara spin_unlock(&wb->list_lock); 1470f0d07b7fSJan Kara 14714f8ad655SJan Kara /* 14724f8ad655SJan Kara * We already requeued the inode if it had I_SYNC set and we 14734f8ad655SJan Kara * are doing WB_SYNC_NONE writeback. So this catches only the 14744f8ad655SJan Kara * WB_SYNC_ALL case. 14754f8ad655SJan Kara */ 1476169ebd90SJan Kara if (inode->i_state & I_SYNC) { 1477169ebd90SJan Kara /* Wait for I_SYNC. This function drops i_lock... */ 1478169ebd90SJan Kara inode_sleep_on_writeback(inode); 1479169ebd90SJan Kara /* Inode may be gone, start again */ 1480ead188f9SJan Kara spin_lock(&wb->list_lock); 1481169ebd90SJan Kara continue; 1482169ebd90SJan Kara } 14834f8ad655SJan Kara inode->i_state |= I_SYNC; 1484b16b1debSTejun Heo wbc_attach_and_unlock_inode(&wbc, inode); 1485169ebd90SJan Kara 1486a88a341aSTejun Heo write_chunk = writeback_chunk_size(wb, work); 1487d46db3d5SWu Fengguang wbc.nr_to_write = write_chunk; 1488d46db3d5SWu Fengguang wbc.pages_skipped = 0; 1489250df6edSDave Chinner 1490169ebd90SJan Kara /* 1491169ebd90SJan Kara * We use I_SYNC to pin the inode in memory. While it is set 1492169ebd90SJan Kara * evict_inode() will wait so the inode cannot be freed. 1493169ebd90SJan Kara */ 1494cd8ed2a4SYan Hong __writeback_single_inode(inode, &wbc); 1495d46db3d5SWu Fengguang 1496b16b1debSTejun Heo wbc_detach_inode(&wbc); 1497d46db3d5SWu Fengguang work->nr_pages -= write_chunk - wbc.nr_to_write; 1498d46db3d5SWu Fengguang wrote += write_chunk - wbc.nr_to_write; 1499590dca3aSChris Mason 1500590dca3aSChris Mason if (need_resched()) { 1501590dca3aSChris Mason /* 1502590dca3aSChris Mason * We're trying to balance between building up a nice 1503590dca3aSChris Mason * long list of IOs to improve our merge rate, and 1504590dca3aSChris Mason * getting those IOs out quickly for anyone throttling 1505590dca3aSChris Mason * in balance_dirty_pages(). cond_resched() doesn't 1506590dca3aSChris Mason * unplug, so get our IOs out the door before we 1507590dca3aSChris Mason * give up the CPU. 1508590dca3aSChris Mason */ 1509590dca3aSChris Mason blk_flush_plug(current); 1510590dca3aSChris Mason cond_resched(); 1511590dca3aSChris Mason } 1512590dca3aSChris Mason 1513590dca3aSChris Mason 15144f8ad655SJan Kara spin_lock(&wb->list_lock); 15154f8ad655SJan Kara spin_lock(&inode->i_lock); 15160ae45f63STheodore Ts'o if (!(inode->i_state & I_DIRTY_ALL)) 1517d46db3d5SWu Fengguang wrote++; 15184f8ad655SJan Kara requeue_inode(inode, wb, &wbc); 15194f8ad655SJan Kara inode_sync_complete(inode); 15200f1b1fd8SDave Chinner spin_unlock(&inode->i_lock); 1521590dca3aSChris Mason 1522d46db3d5SWu Fengguang /* 1523d46db3d5SWu Fengguang * bail out to wb_writeback() often enough to check 1524d46db3d5SWu Fengguang * background threshold and other termination conditions. 1525d46db3d5SWu Fengguang */ 1526d46db3d5SWu Fengguang if (wrote) { 1527d46db3d5SWu Fengguang if (time_is_before_jiffies(start_time + HZ / 10UL)) 1528d46db3d5SWu Fengguang break; 1529d46db3d5SWu Fengguang if (work->nr_pages <= 0) 1530d46db3d5SWu Fengguang break; 15311da177e4SLinus Torvalds } 15328bc3be27SFengguang Wu } 1533d46db3d5SWu Fengguang return wrote; 1534f11c9c5cSEdward Shishkin } 153538f21977SNick Piggin 1536d46db3d5SWu Fengguang static long __writeback_inodes_wb(struct bdi_writeback *wb, 1537d46db3d5SWu Fengguang struct wb_writeback_work *work) 1538f11c9c5cSEdward Shishkin { 1539d46db3d5SWu Fengguang unsigned long start_time = jiffies; 1540d46db3d5SWu Fengguang long wrote = 0; 1541f11c9c5cSEdward Shishkin 1542f11c9c5cSEdward Shishkin while (!list_empty(&wb->b_io)) { 15437ccf19a8SNick Piggin struct inode *inode = wb_inode(wb->b_io.prev); 1544f11c9c5cSEdward Shishkin struct super_block *sb = inode->i_sb; 1545f11c9c5cSEdward Shishkin 1546eb6ef3dfSKonstantin Khlebnikov if (!trylock_super(sb)) { 15470e995816SWu Fengguang /* 1548eb6ef3dfSKonstantin Khlebnikov * trylock_super() may fail consistently due to 15490e995816SWu Fengguang * s_umount being grabbed by someone else. Don't use 15500e995816SWu Fengguang * requeue_io() to avoid busy retrying the inode/sb. 15510e995816SWu Fengguang */ 15520e995816SWu Fengguang redirty_tail(inode, wb); 1553d19de7edSChristoph Hellwig continue; 1554334132aeSChristoph Hellwig } 1555d46db3d5SWu Fengguang wrote += writeback_sb_inodes(sb, wb, work); 1556eb6ef3dfSKonstantin Khlebnikov up_read(&sb->s_umount); 1557f11c9c5cSEdward Shishkin 1558d46db3d5SWu Fengguang /* refer to the same tests at the end of writeback_sb_inodes */ 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) 1563f11c9c5cSEdward Shishkin break; 1564f11c9c5cSEdward Shishkin } 1565d46db3d5SWu Fengguang } 156666f3b8e2SJens Axboe /* Leave any unwritten inodes on b_io */ 1567d46db3d5SWu Fengguang return wrote; 156866f3b8e2SJens Axboe } 156966f3b8e2SJens Axboe 15707d9f073bSWanpeng Li static long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages, 15710e175a18SCurt Wohlgemuth enum wb_reason reason) 1572edadfb10SChristoph Hellwig { 1573d46db3d5SWu Fengguang struct wb_writeback_work work = { 1574d46db3d5SWu Fengguang .nr_pages = nr_pages, 1575d46db3d5SWu Fengguang .sync_mode = WB_SYNC_NONE, 1576d46db3d5SWu Fengguang .range_cyclic = 1, 15770e175a18SCurt Wohlgemuth .reason = reason, 1578d46db3d5SWu Fengguang }; 1579505a666eSLinus Torvalds struct blk_plug plug; 1580edadfb10SChristoph Hellwig 1581505a666eSLinus Torvalds blk_start_plug(&plug); 1582f758eeabSChristoph Hellwig spin_lock(&wb->list_lock); 1583424b351fSWu Fengguang if (list_empty(&wb->b_io)) 1584ad4e38ddSCurt Wohlgemuth queue_io(wb, &work); 1585d46db3d5SWu Fengguang __writeback_inodes_wb(wb, &work); 1586f758eeabSChristoph Hellwig spin_unlock(&wb->list_lock); 1587505a666eSLinus Torvalds blk_finish_plug(&plug); 1588edadfb10SChristoph Hellwig 1589d46db3d5SWu Fengguang return nr_pages - work.nr_pages; 159066f3b8e2SJens Axboe } 159166f3b8e2SJens Axboe 159203ba3782SJens Axboe /* 159303ba3782SJens Axboe * Explicit flushing or periodic writeback of "old" data. 159403ba3782SJens Axboe * 159503ba3782SJens Axboe * Define "old": the first time one of an inode's pages is dirtied, we mark the 159603ba3782SJens Axboe * dirtying-time in the inode's address_space. So this periodic writeback code 159703ba3782SJens Axboe * just walks the superblock inode list, writing back any inodes which are 159803ba3782SJens Axboe * older than a specific point in time. 159903ba3782SJens Axboe * 160003ba3782SJens Axboe * Try to run once per dirty_writeback_interval. But if a writeback event 160103ba3782SJens Axboe * takes longer than a dirty_writeback_interval interval, then leave a 160203ba3782SJens Axboe * one-second gap. 160303ba3782SJens Axboe * 160403ba3782SJens Axboe * older_than_this takes precedence over nr_to_write. So we'll only write back 160503ba3782SJens Axboe * all dirty pages if they are all attached to "old" mappings. 160603ba3782SJens Axboe */ 1607c4a77a6cSJens Axboe static long wb_writeback(struct bdi_writeback *wb, 160883ba7b07SChristoph Hellwig struct wb_writeback_work *work) 160903ba3782SJens Axboe { 1610e98be2d5SWu Fengguang unsigned long wb_start = jiffies; 1611d46db3d5SWu Fengguang long nr_pages = work->nr_pages; 16120dc83bd3SJan Kara unsigned long oldest_jif; 1613a5989bdcSJan Kara struct inode *inode; 1614d46db3d5SWu Fengguang long progress; 1615505a666eSLinus Torvalds struct blk_plug plug; 161603ba3782SJens Axboe 16170dc83bd3SJan Kara oldest_jif = jiffies; 16180dc83bd3SJan Kara work->older_than_this = &oldest_jif; 161903ba3782SJens Axboe 1620505a666eSLinus Torvalds blk_start_plug(&plug); 1621e8dfc305SWu Fengguang spin_lock(&wb->list_lock); 162203ba3782SJens Axboe for (;;) { 162303ba3782SJens Axboe /* 1624d3ddec76SWu Fengguang * Stop writeback when nr_pages has been consumed 162503ba3782SJens Axboe */ 162683ba7b07SChristoph Hellwig if (work->nr_pages <= 0) 162703ba3782SJens Axboe break; 162803ba3782SJens Axboe 162903ba3782SJens Axboe /* 1630aa373cf5SJan Kara * Background writeout and kupdate-style writeback may 1631aa373cf5SJan Kara * run forever. Stop them if there is other work to do 1632aa373cf5SJan Kara * so that e.g. sync can proceed. They'll be restarted 1633aa373cf5SJan Kara * after the other works are all done. 1634aa373cf5SJan Kara */ 1635aa373cf5SJan Kara if ((work->for_background || work->for_kupdate) && 1636f0054bb1STejun Heo !list_empty(&wb->work_list)) 1637aa373cf5SJan Kara break; 1638aa373cf5SJan Kara 1639aa373cf5SJan Kara /* 1640d3ddec76SWu Fengguang * For background writeout, stop when we are below the 1641d3ddec76SWu Fengguang * background dirty threshold 164203ba3782SJens Axboe */ 1643aa661bbeSTejun Heo if (work->for_background && !wb_over_bg_thresh(wb)) 164403ba3782SJens Axboe break; 164503ba3782SJens Axboe 16461bc36b64SJan Kara /* 16471bc36b64SJan Kara * Kupdate and background works are special and we want to 16481bc36b64SJan Kara * include all inodes that need writing. Livelock avoidance is 16491bc36b64SJan Kara * handled by these works yielding to any other work so we are 16501bc36b64SJan Kara * safe. 16511bc36b64SJan Kara */ 1652ba9aa839SWu Fengguang if (work->for_kupdate) { 16530dc83bd3SJan Kara oldest_jif = jiffies - 1654ba9aa839SWu Fengguang msecs_to_jiffies(dirty_expire_interval * 10); 16551bc36b64SJan Kara } else if (work->for_background) 16560dc83bd3SJan Kara oldest_jif = jiffies; 1657028c2dd1SDave Chinner 16585634cc2aSTejun Heo trace_writeback_start(wb, work); 1659e8dfc305SWu Fengguang if (list_empty(&wb->b_io)) 1660ad4e38ddSCurt Wohlgemuth queue_io(wb, work); 166183ba7b07SChristoph Hellwig if (work->sb) 1662d46db3d5SWu Fengguang progress = writeback_sb_inodes(work->sb, wb, work); 1663edadfb10SChristoph Hellwig else 1664d46db3d5SWu Fengguang progress = __writeback_inodes_wb(wb, work); 16655634cc2aSTejun Heo trace_writeback_written(wb, work); 1666028c2dd1SDave Chinner 1667e98be2d5SWu Fengguang wb_update_bandwidth(wb, wb_start); 166803ba3782SJens Axboe 166903ba3782SJens Axboe /* 167071fd05a8SJens Axboe * Did we write something? Try for more 1671e6fb6da2SWu Fengguang * 1672e6fb6da2SWu Fengguang * Dirty inodes are moved to b_io for writeback in batches. 1673e6fb6da2SWu Fengguang * The completion of the current batch does not necessarily 1674e6fb6da2SWu Fengguang * mean the overall work is done. So we keep looping as long 1675e6fb6da2SWu Fengguang * as made some progress on cleaning pages or inodes. 167671fd05a8SJens Axboe */ 1677d46db3d5SWu Fengguang if (progress) 167803ba3782SJens Axboe continue; 1679a5989bdcSJan Kara /* 1680e6fb6da2SWu Fengguang * No more inodes for IO, bail 1681a5989bdcSJan Kara */ 1682b7a2441fSWu Fengguang if (list_empty(&wb->b_more_io)) 168303ba3782SJens Axboe break; 168403ba3782SJens Axboe /* 16858010c3b6SJens Axboe * Nothing written. Wait for some inode to 16868010c3b6SJens Axboe * become available for writeback. Otherwise 16878010c3b6SJens Axboe * we'll just busyloop. 168803ba3782SJens Axboe */ 168903ba3782SJens Axboe if (!list_empty(&wb->b_more_io)) { 16905634cc2aSTejun Heo trace_writeback_wait(wb, work); 169103ba3782SJens Axboe inode = wb_inode(wb->b_more_io.prev); 1692250df6edSDave Chinner spin_lock(&inode->i_lock); 1693f0d07b7fSJan Kara spin_unlock(&wb->list_lock); 1694169ebd90SJan Kara /* This function drops i_lock... */ 1695169ebd90SJan Kara inode_sleep_on_writeback(inode); 1696f0d07b7fSJan Kara spin_lock(&wb->list_lock); 169703ba3782SJens Axboe } 169803ba3782SJens Axboe } 1699e8dfc305SWu Fengguang spin_unlock(&wb->list_lock); 1700505a666eSLinus Torvalds blk_finish_plug(&plug); 170103ba3782SJens Axboe 1702d46db3d5SWu Fengguang return nr_pages - work->nr_pages; 170303ba3782SJens Axboe } 170403ba3782SJens Axboe 170503ba3782SJens Axboe /* 170683ba7b07SChristoph Hellwig * Return the next wb_writeback_work struct that hasn't been processed yet. 170703ba3782SJens Axboe */ 1708f0054bb1STejun Heo static struct wb_writeback_work *get_next_work_item(struct bdi_writeback *wb) 170903ba3782SJens Axboe { 171083ba7b07SChristoph Hellwig struct wb_writeback_work *work = NULL; 171103ba3782SJens Axboe 1712f0054bb1STejun Heo spin_lock_bh(&wb->work_lock); 1713f0054bb1STejun Heo if (!list_empty(&wb->work_list)) { 1714f0054bb1STejun Heo work = list_entry(wb->work_list.next, 171583ba7b07SChristoph Hellwig struct wb_writeback_work, list); 171683ba7b07SChristoph Hellwig list_del_init(&work->list); 171703ba3782SJens Axboe } 1718f0054bb1STejun Heo spin_unlock_bh(&wb->work_lock); 171983ba7b07SChristoph Hellwig return work; 172003ba3782SJens Axboe } 172103ba3782SJens Axboe 1722cdf01dd5SLinus Torvalds /* 1723cdf01dd5SLinus Torvalds * Add in the number of potentially dirty inodes, because each inode 1724cdf01dd5SLinus Torvalds * write can dirty pagecache in the underlying blockdev. 1725cdf01dd5SLinus Torvalds */ 1726cdf01dd5SLinus Torvalds static unsigned long get_nr_dirty_pages(void) 1727cdf01dd5SLinus Torvalds { 1728cdf01dd5SLinus Torvalds return global_page_state(NR_FILE_DIRTY) + 1729cdf01dd5SLinus Torvalds global_page_state(NR_UNSTABLE_NFS) + 1730cdf01dd5SLinus Torvalds get_nr_dirty_inodes(); 1731cdf01dd5SLinus Torvalds } 1732cdf01dd5SLinus Torvalds 17336585027aSJan Kara static long wb_check_background_flush(struct bdi_writeback *wb) 17346585027aSJan Kara { 1735aa661bbeSTejun Heo if (wb_over_bg_thresh(wb)) { 17366585027aSJan Kara 17376585027aSJan Kara struct wb_writeback_work work = { 17386585027aSJan Kara .nr_pages = LONG_MAX, 17396585027aSJan Kara .sync_mode = WB_SYNC_NONE, 17406585027aSJan Kara .for_background = 1, 17416585027aSJan Kara .range_cyclic = 1, 17420e175a18SCurt Wohlgemuth .reason = WB_REASON_BACKGROUND, 17436585027aSJan Kara }; 17446585027aSJan Kara 17456585027aSJan Kara return wb_writeback(wb, &work); 17466585027aSJan Kara } 17476585027aSJan Kara 17486585027aSJan Kara return 0; 17496585027aSJan Kara } 17506585027aSJan Kara 175103ba3782SJens Axboe static long wb_check_old_data_flush(struct bdi_writeback *wb) 175203ba3782SJens Axboe { 175303ba3782SJens Axboe unsigned long expired; 175403ba3782SJens Axboe long nr_pages; 175503ba3782SJens Axboe 175669b62d01SJens Axboe /* 175769b62d01SJens Axboe * When set to zero, disable periodic writeback 175869b62d01SJens Axboe */ 175969b62d01SJens Axboe if (!dirty_writeback_interval) 176069b62d01SJens Axboe return 0; 176169b62d01SJens Axboe 176203ba3782SJens Axboe expired = wb->last_old_flush + 176303ba3782SJens Axboe msecs_to_jiffies(dirty_writeback_interval * 10); 176403ba3782SJens Axboe if (time_before(jiffies, expired)) 176503ba3782SJens Axboe return 0; 176603ba3782SJens Axboe 176703ba3782SJens Axboe wb->last_old_flush = jiffies; 1768cdf01dd5SLinus Torvalds nr_pages = get_nr_dirty_pages(); 176903ba3782SJens Axboe 1770c4a77a6cSJens Axboe if (nr_pages) { 177183ba7b07SChristoph Hellwig struct wb_writeback_work work = { 1772c4a77a6cSJens Axboe .nr_pages = nr_pages, 1773c4a77a6cSJens Axboe .sync_mode = WB_SYNC_NONE, 1774c4a77a6cSJens Axboe .for_kupdate = 1, 1775c4a77a6cSJens Axboe .range_cyclic = 1, 17760e175a18SCurt Wohlgemuth .reason = WB_REASON_PERIODIC, 1777c4a77a6cSJens Axboe }; 1778c4a77a6cSJens Axboe 177983ba7b07SChristoph Hellwig return wb_writeback(wb, &work); 1780c4a77a6cSJens Axboe } 178103ba3782SJens Axboe 178203ba3782SJens Axboe return 0; 178303ba3782SJens Axboe } 178403ba3782SJens Axboe 178503ba3782SJens Axboe /* 178603ba3782SJens Axboe * Retrieve work items and do the writeback they describe 178703ba3782SJens Axboe */ 178825d130baSWanpeng Li static long wb_do_writeback(struct bdi_writeback *wb) 178903ba3782SJens Axboe { 179083ba7b07SChristoph Hellwig struct wb_writeback_work *work; 1791c4a77a6cSJens Axboe long wrote = 0; 179203ba3782SJens Axboe 17934452226eSTejun Heo set_bit(WB_writeback_running, &wb->state); 1794f0054bb1STejun Heo while ((work = get_next_work_item(wb)) != NULL) { 1795cc395d7fSTejun Heo struct wb_completion *done = work->done; 179683ba7b07SChristoph Hellwig 17975634cc2aSTejun Heo trace_writeback_exec(wb, work); 1798455b2864SDave Chinner 179983ba7b07SChristoph Hellwig wrote += wb_writeback(wb, work); 180003ba3782SJens Axboe 18018a1270cdSTejun Heo if (work->auto_free) 180283ba7b07SChristoph Hellwig kfree(work); 1803cc395d7fSTejun Heo if (done && atomic_dec_and_test(&done->cnt)) 1804cc395d7fSTejun Heo wake_up_all(&wb->bdi->wb_waitq); 180503ba3782SJens Axboe } 180603ba3782SJens Axboe 180703ba3782SJens Axboe /* 180803ba3782SJens Axboe * Check for periodic writeback, kupdated() style 180903ba3782SJens Axboe */ 181003ba3782SJens Axboe wrote += wb_check_old_data_flush(wb); 18116585027aSJan Kara wrote += wb_check_background_flush(wb); 18124452226eSTejun Heo clear_bit(WB_writeback_running, &wb->state); 181303ba3782SJens Axboe 181403ba3782SJens Axboe return wrote; 181503ba3782SJens Axboe } 181603ba3782SJens Axboe 181703ba3782SJens Axboe /* 181803ba3782SJens Axboe * Handle writeback of dirty data for the device backed by this bdi. Also 1819839a8e86STejun Heo * reschedules periodically and does kupdated style flushing. 182003ba3782SJens Axboe */ 1821f0054bb1STejun Heo void wb_workfn(struct work_struct *work) 182203ba3782SJens Axboe { 1823839a8e86STejun Heo struct bdi_writeback *wb = container_of(to_delayed_work(work), 1824839a8e86STejun Heo struct bdi_writeback, dwork); 182503ba3782SJens Axboe long pages_written; 182603ba3782SJens Axboe 1827f0054bb1STejun Heo set_worker_desc("flush-%s", dev_name(wb->bdi->dev)); 1828766f9164SPeter Zijlstra current->flags |= PF_SWAPWRITE; 182903ba3782SJens Axboe 1830839a8e86STejun Heo if (likely(!current_is_workqueue_rescuer() || 18314452226eSTejun Heo !test_bit(WB_registered, &wb->state))) { 183203ba3782SJens Axboe /* 1833f0054bb1STejun Heo * The normal path. Keep writing back @wb until its 1834839a8e86STejun Heo * work_list is empty. Note that this path is also taken 1835f0054bb1STejun Heo * if @wb is shutting down even when we're running off the 1836839a8e86STejun Heo * rescuer as work_list needs to be drained. 183703ba3782SJens Axboe */ 1838839a8e86STejun Heo do { 183925d130baSWanpeng Li pages_written = wb_do_writeback(wb); 1840455b2864SDave Chinner trace_writeback_pages_written(pages_written); 1841f0054bb1STejun Heo } while (!list_empty(&wb->work_list)); 1842839a8e86STejun Heo } else { 1843253c34e9SArtem Bityutskiy /* 1844839a8e86STejun Heo * bdi_wq can't get enough workers and we're running off 1845839a8e86STejun Heo * the emergency worker. Don't hog it. Hopefully, 1024 is 1846839a8e86STejun Heo * enough for efficient IO. 1847253c34e9SArtem Bityutskiy */ 1848f0054bb1STejun Heo pages_written = writeback_inodes_wb(wb, 1024, 1849839a8e86STejun Heo WB_REASON_FORKER_THREAD); 1850839a8e86STejun Heo trace_writeback_pages_written(pages_written); 185103ba3782SJens Axboe } 185203ba3782SJens Axboe 1853f0054bb1STejun Heo if (!list_empty(&wb->work_list)) 18546ca738d6SDerek Basehore mod_delayed_work(bdi_wq, &wb->dwork, 0); 18556ca738d6SDerek Basehore else if (wb_has_dirty_io(wb) && dirty_writeback_interval) 1856f0054bb1STejun Heo wb_wakeup_delayed(wb); 1857455b2864SDave Chinner 1858839a8e86STejun Heo current->flags &= ~PF_SWAPWRITE; 185903ba3782SJens Axboe } 186003ba3782SJens Axboe 186103ba3782SJens Axboe /* 186203ba3782SJens Axboe * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back 186303ba3782SJens Axboe * the whole world. 186403ba3782SJens Axboe */ 18650e175a18SCurt Wohlgemuth void wakeup_flusher_threads(long nr_pages, enum wb_reason reason) 186603ba3782SJens Axboe { 1867b8c2f347SChristoph Hellwig struct backing_dev_info *bdi; 1868b8c2f347SChristoph Hellwig 186947df3ddeSJan Kara if (!nr_pages) 187047df3ddeSJan Kara nr_pages = get_nr_dirty_pages(); 1871b8c2f347SChristoph Hellwig 1872b8c2f347SChristoph Hellwig rcu_read_lock(); 1873f2b65121STejun Heo list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { 1874f2b65121STejun Heo struct bdi_writeback *wb; 1875f2b65121STejun Heo 1876f2b65121STejun Heo if (!bdi_has_dirty_io(bdi)) 1877f2b65121STejun Heo continue; 1878f2b65121STejun Heo 1879b817525aSTejun Heo list_for_each_entry_rcu(wb, &bdi->wb_list, bdi_node) 1880f2b65121STejun Heo wb_start_writeback(wb, wb_split_bdi_pages(wb, nr_pages), 1881f2b65121STejun Heo false, reason); 1882f2b65121STejun Heo } 1883b8c2f347SChristoph Hellwig rcu_read_unlock(); 188403ba3782SJens Axboe } 188503ba3782SJens Axboe 1886a2f48706STheodore Ts'o /* 1887a2f48706STheodore Ts'o * Wake up bdi's periodically to make sure dirtytime inodes gets 1888a2f48706STheodore Ts'o * written back periodically. We deliberately do *not* check the 1889a2f48706STheodore Ts'o * b_dirtytime list in wb_has_dirty_io(), since this would cause the 1890a2f48706STheodore Ts'o * kernel to be constantly waking up once there are any dirtytime 1891a2f48706STheodore Ts'o * inodes on the system. So instead we define a separate delayed work 1892a2f48706STheodore Ts'o * function which gets called much more rarely. (By default, only 1893a2f48706STheodore Ts'o * once every 12 hours.) 1894a2f48706STheodore Ts'o * 1895a2f48706STheodore Ts'o * If there is any other write activity going on in the file system, 1896a2f48706STheodore Ts'o * this function won't be necessary. But if the only thing that has 1897a2f48706STheodore Ts'o * happened on the file system is a dirtytime inode caused by an atime 1898a2f48706STheodore Ts'o * update, we need this infrastructure below to make sure that inode 1899a2f48706STheodore Ts'o * eventually gets pushed out to disk. 1900a2f48706STheodore Ts'o */ 1901a2f48706STheodore Ts'o static void wakeup_dirtytime_writeback(struct work_struct *w); 1902a2f48706STheodore Ts'o static DECLARE_DELAYED_WORK(dirtytime_work, wakeup_dirtytime_writeback); 1903a2f48706STheodore Ts'o 1904a2f48706STheodore Ts'o static void wakeup_dirtytime_writeback(struct work_struct *w) 1905a2f48706STheodore Ts'o { 1906a2f48706STheodore Ts'o struct backing_dev_info *bdi; 1907a2f48706STheodore Ts'o 1908a2f48706STheodore Ts'o rcu_read_lock(); 1909a2f48706STheodore Ts'o list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { 1910001fe6f6STejun Heo struct bdi_writeback *wb; 1911001fe6f6STejun Heo 1912b817525aSTejun Heo list_for_each_entry_rcu(wb, &bdi->wb_list, bdi_node) 19136fdf860fSTejun Heo if (!list_empty(&wb->b_dirty_time)) 19146fdf860fSTejun Heo wb_wakeup(wb); 1915a2f48706STheodore Ts'o } 1916a2f48706STheodore Ts'o rcu_read_unlock(); 1917a2f48706STheodore Ts'o schedule_delayed_work(&dirtytime_work, dirtytime_expire_interval * HZ); 1918a2f48706STheodore Ts'o } 1919a2f48706STheodore Ts'o 1920a2f48706STheodore Ts'o static int __init start_dirtytime_writeback(void) 1921a2f48706STheodore Ts'o { 1922a2f48706STheodore Ts'o schedule_delayed_work(&dirtytime_work, dirtytime_expire_interval * HZ); 1923a2f48706STheodore Ts'o return 0; 1924a2f48706STheodore Ts'o } 1925a2f48706STheodore Ts'o __initcall(start_dirtytime_writeback); 1926a2f48706STheodore Ts'o 19271efff914STheodore Ts'o int dirtytime_interval_handler(struct ctl_table *table, int write, 19281efff914STheodore Ts'o void __user *buffer, size_t *lenp, loff_t *ppos) 19291efff914STheodore Ts'o { 19301efff914STheodore Ts'o int ret; 19311efff914STheodore Ts'o 19321efff914STheodore Ts'o ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos); 19331efff914STheodore Ts'o if (ret == 0 && write) 19341efff914STheodore Ts'o mod_delayed_work(system_wq, &dirtytime_work, 0); 19351efff914STheodore Ts'o return ret; 19361efff914STheodore Ts'o } 19371efff914STheodore Ts'o 193803ba3782SJens Axboe static noinline void block_dump___mark_inode_dirty(struct inode *inode) 193903ba3782SJens Axboe { 194003ba3782SJens Axboe if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { 194103ba3782SJens Axboe struct dentry *dentry; 194203ba3782SJens Axboe const char *name = "?"; 194303ba3782SJens Axboe 194403ba3782SJens Axboe dentry = d_find_alias(inode); 194503ba3782SJens Axboe if (dentry) { 194603ba3782SJens Axboe spin_lock(&dentry->d_lock); 194703ba3782SJens Axboe name = (const char *) dentry->d_name.name; 194803ba3782SJens Axboe } 194903ba3782SJens Axboe printk(KERN_DEBUG 195003ba3782SJens Axboe "%s(%d): dirtied inode %lu (%s) on %s\n", 195103ba3782SJens Axboe current->comm, task_pid_nr(current), inode->i_ino, 195203ba3782SJens Axboe name, inode->i_sb->s_id); 195303ba3782SJens Axboe if (dentry) { 195403ba3782SJens Axboe spin_unlock(&dentry->d_lock); 195503ba3782SJens Axboe dput(dentry); 195603ba3782SJens Axboe } 195703ba3782SJens Axboe } 195803ba3782SJens Axboe } 195903ba3782SJens Axboe 196003ba3782SJens Axboe /** 196103ba3782SJens Axboe * __mark_inode_dirty - internal function 196203ba3782SJens Axboe * @inode: inode to mark 196303ba3782SJens Axboe * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) 196403ba3782SJens Axboe * Mark an inode as dirty. Callers should use mark_inode_dirty or 196503ba3782SJens Axboe * mark_inode_dirty_sync. 196603ba3782SJens Axboe * 196703ba3782SJens Axboe * Put the inode on the super block's dirty list. 196803ba3782SJens Axboe * 196903ba3782SJens Axboe * CAREFUL! We mark it dirty unconditionally, but move it onto the 197003ba3782SJens Axboe * dirty list only if it is hashed or if it refers to a blockdev. 197103ba3782SJens Axboe * If it was not hashed, it will never be added to the dirty list 197203ba3782SJens Axboe * even if it is later hashed, as it will have been marked dirty already. 197303ba3782SJens Axboe * 197403ba3782SJens Axboe * In short, make sure you hash any inodes _before_ you start marking 197503ba3782SJens Axboe * them dirty. 197603ba3782SJens Axboe * 197703ba3782SJens Axboe * Note that for blockdevs, inode->dirtied_when represents the dirtying time of 197803ba3782SJens Axboe * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of 197903ba3782SJens Axboe * the kernel-internal blockdev inode represents the dirtying time of the 198003ba3782SJens Axboe * blockdev's pages. This is why for I_DIRTY_PAGES we always use 198103ba3782SJens Axboe * page->mapping->host, so the page-dirtying time is recorded in the internal 198203ba3782SJens Axboe * blockdev inode. 198303ba3782SJens Axboe */ 19840ae45f63STheodore Ts'o #define I_DIRTY_INODE (I_DIRTY_SYNC | I_DIRTY_DATASYNC) 198503ba3782SJens Axboe void __mark_inode_dirty(struct inode *inode, int flags) 198603ba3782SJens Axboe { 198703ba3782SJens Axboe struct super_block *sb = inode->i_sb; 19880ae45f63STheodore Ts'o int dirtytime; 19890ae45f63STheodore Ts'o 19900ae45f63STheodore Ts'o trace_writeback_mark_inode_dirty(inode, flags); 199103ba3782SJens Axboe 199203ba3782SJens Axboe /* 199303ba3782SJens Axboe * Don't do this for I_DIRTY_PAGES - that doesn't actually 199403ba3782SJens Axboe * dirty the inode itself 199503ba3782SJens Axboe */ 19960ae45f63STheodore Ts'o if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC | I_DIRTY_TIME)) { 19979fb0a7daSTejun Heo trace_writeback_dirty_inode_start(inode, flags); 19989fb0a7daSTejun Heo 199903ba3782SJens Axboe if (sb->s_op->dirty_inode) 2000aa385729SChristoph Hellwig sb->s_op->dirty_inode(inode, flags); 20019fb0a7daSTejun Heo 20029fb0a7daSTejun Heo trace_writeback_dirty_inode(inode, flags); 200303ba3782SJens Axboe } 20040ae45f63STheodore Ts'o if (flags & I_DIRTY_INODE) 20050ae45f63STheodore Ts'o flags &= ~I_DIRTY_TIME; 20060ae45f63STheodore Ts'o dirtytime = flags & I_DIRTY_TIME; 200703ba3782SJens Axboe 200803ba3782SJens Axboe /* 20099c6ac78eSTejun Heo * Paired with smp_mb() in __writeback_single_inode() for the 20109c6ac78eSTejun Heo * following lockless i_state test. See there for details. 201103ba3782SJens Axboe */ 201203ba3782SJens Axboe smp_mb(); 201303ba3782SJens Axboe 20140ae45f63STheodore Ts'o if (((inode->i_state & flags) == flags) || 20150ae45f63STheodore Ts'o (dirtytime && (inode->i_state & I_DIRTY_INODE))) 201603ba3782SJens Axboe return; 201703ba3782SJens Axboe 201803ba3782SJens Axboe if (unlikely(block_dump)) 201903ba3782SJens Axboe block_dump___mark_inode_dirty(inode); 202003ba3782SJens Axboe 2021250df6edSDave Chinner spin_lock(&inode->i_lock); 20220ae45f63STheodore Ts'o if (dirtytime && (inode->i_state & I_DIRTY_INODE)) 20230ae45f63STheodore Ts'o goto out_unlock_inode; 202403ba3782SJens Axboe if ((inode->i_state & flags) != flags) { 202503ba3782SJens Axboe const int was_dirty = inode->i_state & I_DIRTY; 202603ba3782SJens Axboe 202752ebea74STejun Heo inode_attach_wb(inode, NULL); 202852ebea74STejun Heo 20290ae45f63STheodore Ts'o if (flags & I_DIRTY_INODE) 20300ae45f63STheodore Ts'o inode->i_state &= ~I_DIRTY_TIME; 203103ba3782SJens Axboe inode->i_state |= flags; 203203ba3782SJens Axboe 203303ba3782SJens Axboe /* 203403ba3782SJens Axboe * If the inode is being synced, just update its dirty state. 203503ba3782SJens Axboe * The unlocker will place the inode on the appropriate 203603ba3782SJens Axboe * superblock list, based upon its state. 203703ba3782SJens Axboe */ 203803ba3782SJens Axboe if (inode->i_state & I_SYNC) 2039250df6edSDave Chinner goto out_unlock_inode; 204003ba3782SJens Axboe 204103ba3782SJens Axboe /* 204203ba3782SJens Axboe * Only add valid (hashed) inodes to the superblock's 204303ba3782SJens Axboe * dirty list. Add blockdev inodes as well. 204403ba3782SJens Axboe */ 204503ba3782SJens Axboe if (!S_ISBLK(inode->i_mode)) { 20461d3382cbSAl Viro if (inode_unhashed(inode)) 2047250df6edSDave Chinner goto out_unlock_inode; 204803ba3782SJens Axboe } 2049a4ffdde6SAl Viro if (inode->i_state & I_FREEING) 2050250df6edSDave Chinner goto out_unlock_inode; 205103ba3782SJens Axboe 205203ba3782SJens Axboe /* 205303ba3782SJens Axboe * If the inode was already on b_dirty/b_io/b_more_io, don't 205403ba3782SJens Axboe * reposition it (that would break b_dirty time-ordering). 205503ba3782SJens Axboe */ 205603ba3782SJens Axboe if (!was_dirty) { 205787e1d789STejun Heo struct bdi_writeback *wb; 2058d6c10f1fSTejun Heo struct list_head *dirty_list; 2059a66979abSDave Chinner bool wakeup_bdi = false; 2060500b067cSJens Axboe 206187e1d789STejun Heo wb = locked_inode_to_wb_and_lock_list(inode); 2062253c34e9SArtem Bityutskiy 20630747259dSTejun Heo WARN(bdi_cap_writeback_dirty(wb->bdi) && 20640747259dSTejun Heo !test_bit(WB_registered, &wb->state), 20650747259dSTejun Heo "bdi-%s not registered\n", wb->bdi->name); 206603ba3782SJens Axboe 206703ba3782SJens Axboe inode->dirtied_when = jiffies; 2068a2f48706STheodore Ts'o if (dirtytime) 2069a2f48706STheodore Ts'o inode->dirtied_time_when = jiffies; 2070d6c10f1fSTejun Heo 2071a2f48706STheodore Ts'o if (inode->i_state & (I_DIRTY_INODE | I_DIRTY_PAGES)) 20720747259dSTejun Heo dirty_list = &wb->b_dirty; 2073a2f48706STheodore Ts'o else 20740747259dSTejun Heo dirty_list = &wb->b_dirty_time; 2075d6c10f1fSTejun Heo 2076c7f54084SDave Chinner wakeup_bdi = inode_io_list_move_locked(inode, wb, 2077d6c10f1fSTejun Heo dirty_list); 2078d6c10f1fSTejun Heo 20790747259dSTejun Heo spin_unlock(&wb->list_lock); 20800ae45f63STheodore Ts'o trace_writeback_dirty_inode_enqueue(inode); 2081253c34e9SArtem Bityutskiy 2082d6c10f1fSTejun Heo /* 2083d6c10f1fSTejun Heo * If this is the first dirty inode for this bdi, 2084d6c10f1fSTejun Heo * we have to wake-up the corresponding bdi thread 2085d6c10f1fSTejun Heo * to make sure background write-back happens 2086d6c10f1fSTejun Heo * later. 2087d6c10f1fSTejun Heo */ 20880747259dSTejun Heo if (bdi_cap_writeback_dirty(wb->bdi) && wakeup_bdi) 20890747259dSTejun Heo wb_wakeup_delayed(wb); 2090a66979abSDave Chinner return; 2091a66979abSDave Chinner } 2092a66979abSDave Chinner } 2093a66979abSDave Chinner out_unlock_inode: 2094a66979abSDave Chinner spin_unlock(&inode->i_lock); 2095a66979abSDave Chinner 209603ba3782SJens Axboe } 209703ba3782SJens Axboe EXPORT_SYMBOL(__mark_inode_dirty); 209803ba3782SJens Axboe 2099e97fedb9SDave Chinner /* 2100e97fedb9SDave Chinner * The @s_sync_lock is used to serialise concurrent sync operations 2101e97fedb9SDave Chinner * to avoid lock contention problems with concurrent wait_sb_inodes() calls. 2102e97fedb9SDave Chinner * Concurrent callers will block on the s_sync_lock rather than doing contending 2103e97fedb9SDave Chinner * walks. The queueing maintains sync(2) required behaviour as all the IO that 2104e97fedb9SDave Chinner * has been issued up to the time this function is enter is guaranteed to be 2105e97fedb9SDave Chinner * completed by the time we have gained the lock and waited for all IO that is 2106e97fedb9SDave Chinner * in progress regardless of the order callers are granted the lock. 2107e97fedb9SDave Chinner */ 2108b6e51316SJens Axboe static void wait_sb_inodes(struct super_block *sb) 210966f3b8e2SJens Axboe { 211038f21977SNick Piggin struct inode *inode, *old_inode = NULL; 211138f21977SNick Piggin 211203ba3782SJens Axboe /* 211303ba3782SJens Axboe * We need to be protected against the filesystem going from 211403ba3782SJens Axboe * r/o to r/w or vice versa. 211503ba3782SJens Axboe */ 2116b6e51316SJens Axboe WARN_ON(!rwsem_is_locked(&sb->s_umount)); 211703ba3782SJens Axboe 2118e97fedb9SDave Chinner mutex_lock(&sb->s_sync_lock); 211974278da9SDave Chinner spin_lock(&sb->s_inode_list_lock); 212066f3b8e2SJens Axboe 212138f21977SNick Piggin /* 212238f21977SNick Piggin * Data integrity sync. Must wait for all pages under writeback, 212338f21977SNick Piggin * because there may have been pages dirtied before our sync 212438f21977SNick Piggin * call, but which had writeout started before we write it out. 212538f21977SNick Piggin * In which case, the inode may not be on the dirty list, but 212638f21977SNick Piggin * we still have to wait for that writeout. 212738f21977SNick Piggin */ 2128b6e51316SJens Axboe list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { 2129250df6edSDave Chinner struct address_space *mapping = inode->i_mapping; 213038f21977SNick Piggin 2131250df6edSDave Chinner spin_lock(&inode->i_lock); 2132250df6edSDave Chinner if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) || 2133250df6edSDave Chinner (mapping->nrpages == 0)) { 2134250df6edSDave Chinner spin_unlock(&inode->i_lock); 213538f21977SNick Piggin continue; 2136250df6edSDave Chinner } 213738f21977SNick Piggin __iget(inode); 2138250df6edSDave Chinner spin_unlock(&inode->i_lock); 213974278da9SDave Chinner spin_unlock(&sb->s_inode_list_lock); 214055fa6091SDave Chinner 214138f21977SNick Piggin /* 214255fa6091SDave Chinner * We hold a reference to 'inode' so it couldn't have been 214355fa6091SDave Chinner * removed from s_inodes list while we dropped the 214474278da9SDave Chinner * s_inode_list_lock. We cannot iput the inode now as we can 214555fa6091SDave Chinner * be holding the last reference and we cannot iput it under 214674278da9SDave Chinner * s_inode_list_lock. So we keep the reference and iput it 214755fa6091SDave Chinner * later. 214838f21977SNick Piggin */ 214938f21977SNick Piggin iput(old_inode); 215038f21977SNick Piggin old_inode = inode; 215138f21977SNick Piggin 215238f21977SNick Piggin filemap_fdatawait(mapping); 215338f21977SNick Piggin 215438f21977SNick Piggin cond_resched(); 215538f21977SNick Piggin 215674278da9SDave Chinner spin_lock(&sb->s_inode_list_lock); 215738f21977SNick Piggin } 215874278da9SDave Chinner spin_unlock(&sb->s_inode_list_lock); 215938f21977SNick Piggin iput(old_inode); 2160e97fedb9SDave Chinner mutex_unlock(&sb->s_sync_lock); 216166f3b8e2SJens Axboe } 21621da177e4SLinus Torvalds 2163f30a7d0cSTejun Heo static void __writeback_inodes_sb_nr(struct super_block *sb, unsigned long nr, 2164f30a7d0cSTejun Heo enum wb_reason reason, bool skip_if_busy) 21651da177e4SLinus Torvalds { 2166cc395d7fSTejun Heo DEFINE_WB_COMPLETION_ONSTACK(done); 216783ba7b07SChristoph Hellwig struct wb_writeback_work work = { 21683c4d7165SChristoph Hellwig .sb = sb, 21693c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_NONE, 21706e6938b6SWu Fengguang .tagged_writepages = 1, 217183ba7b07SChristoph Hellwig .done = &done, 21723259f8beSChris Mason .nr_pages = nr, 21730e175a18SCurt Wohlgemuth .reason = reason, 21743c4d7165SChristoph Hellwig }; 2175e7972912STejun Heo struct backing_dev_info *bdi = sb->s_bdi; 21760e3c9a22SJens Axboe 2177e7972912STejun Heo if (!bdi_has_dirty_io(bdi) || bdi == &noop_backing_dev_info) 21786eedc701SJan Kara return; 2179cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 2180f30a7d0cSTejun Heo 2181db125360STejun Heo bdi_split_work_to_wbs(sb->s_bdi, &work, skip_if_busy); 2182cc395d7fSTejun Heo wb_wait_for_completion(bdi, &done); 21831da177e4SLinus Torvalds } 2184f30a7d0cSTejun Heo 2185f30a7d0cSTejun Heo /** 2186f30a7d0cSTejun Heo * writeback_inodes_sb_nr - writeback dirty inodes from given super_block 2187f30a7d0cSTejun Heo * @sb: the superblock 2188f30a7d0cSTejun Heo * @nr: the number of pages to write 2189f30a7d0cSTejun Heo * @reason: reason why some writeback work initiated 2190f30a7d0cSTejun Heo * 2191f30a7d0cSTejun Heo * Start writeback on some inodes on this super_block. No guarantees are made 2192f30a7d0cSTejun Heo * on how many (if any) will be written, and this function does not wait 2193f30a7d0cSTejun Heo * for IO completion of submitted IO. 2194f30a7d0cSTejun Heo */ 2195f30a7d0cSTejun Heo void writeback_inodes_sb_nr(struct super_block *sb, 2196f30a7d0cSTejun Heo unsigned long nr, 2197f30a7d0cSTejun Heo enum wb_reason reason) 2198f30a7d0cSTejun Heo { 2199f30a7d0cSTejun Heo __writeback_inodes_sb_nr(sb, nr, reason, false); 2200f30a7d0cSTejun Heo } 22013259f8beSChris Mason EXPORT_SYMBOL(writeback_inodes_sb_nr); 22023259f8beSChris Mason 22033259f8beSChris Mason /** 22043259f8beSChris Mason * writeback_inodes_sb - writeback dirty inodes from given super_block 22053259f8beSChris Mason * @sb: the superblock 2206786228abSMarcos Paulo de Souza * @reason: reason why some writeback work was initiated 22073259f8beSChris Mason * 22083259f8beSChris Mason * Start writeback on some inodes on this super_block. No guarantees are made 22093259f8beSChris Mason * on how many (if any) will be written, and this function does not wait 22103259f8beSChris Mason * for IO completion of submitted IO. 22113259f8beSChris Mason */ 22120e175a18SCurt Wohlgemuth void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) 22133259f8beSChris Mason { 22140e175a18SCurt Wohlgemuth return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); 22153259f8beSChris Mason } 2216d8a8559cSJens Axboe EXPORT_SYMBOL(writeback_inodes_sb); 2217d8a8559cSJens Axboe 2218d8a8559cSJens Axboe /** 221910ee27a0SMiao Xie * try_to_writeback_inodes_sb_nr - try to start writeback if none underway 22203259f8beSChris Mason * @sb: the superblock 22213259f8beSChris Mason * @nr: the number of pages to write 222210ee27a0SMiao Xie * @reason: the reason of writeback 22233259f8beSChris Mason * 222410ee27a0SMiao Xie * Invoke writeback_inodes_sb_nr if no writeback is currently underway. 22253259f8beSChris Mason * Returns 1 if writeback was started, 0 if not. 22263259f8beSChris Mason */ 2227f30a7d0cSTejun Heo bool try_to_writeback_inodes_sb_nr(struct super_block *sb, unsigned long nr, 22280e175a18SCurt Wohlgemuth enum wb_reason reason) 22293259f8beSChris Mason { 223010ee27a0SMiao Xie if (!down_read_trylock(&sb->s_umount)) 2231f30a7d0cSTejun Heo return false; 223210ee27a0SMiao Xie 2233f30a7d0cSTejun Heo __writeback_inodes_sb_nr(sb, nr, reason, true); 22343259f8beSChris Mason up_read(&sb->s_umount); 2235f30a7d0cSTejun Heo return true; 22363259f8beSChris Mason } 223710ee27a0SMiao Xie EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr); 223810ee27a0SMiao Xie 223910ee27a0SMiao Xie /** 224010ee27a0SMiao Xie * try_to_writeback_inodes_sb - try to start writeback if none underway 224110ee27a0SMiao Xie * @sb: the superblock 224210ee27a0SMiao Xie * @reason: reason why some writeback work was initiated 224310ee27a0SMiao Xie * 224410ee27a0SMiao Xie * Implement by try_to_writeback_inodes_sb_nr() 224510ee27a0SMiao Xie * Returns 1 if writeback was started, 0 if not. 224610ee27a0SMiao Xie */ 2247f30a7d0cSTejun Heo bool try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) 224810ee27a0SMiao Xie { 224910ee27a0SMiao Xie return try_to_writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); 225010ee27a0SMiao Xie } 225110ee27a0SMiao Xie EXPORT_SYMBOL(try_to_writeback_inodes_sb); 22523259f8beSChris Mason 22533259f8beSChris Mason /** 2254d8a8559cSJens Axboe * sync_inodes_sb - sync sb inode pages 2255d8a8559cSJens Axboe * @sb: the superblock 2256d8a8559cSJens Axboe * 2257d8a8559cSJens Axboe * This function writes and waits on any dirty inode belonging to this 22580dc83bd3SJan Kara * super_block. 2259d8a8559cSJens Axboe */ 22600dc83bd3SJan Kara void sync_inodes_sb(struct super_block *sb) 2261d8a8559cSJens Axboe { 2262cc395d7fSTejun Heo DEFINE_WB_COMPLETION_ONSTACK(done); 226383ba7b07SChristoph Hellwig struct wb_writeback_work work = { 22643c4d7165SChristoph Hellwig .sb = sb, 22653c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_ALL, 22663c4d7165SChristoph Hellwig .nr_pages = LONG_MAX, 22673c4d7165SChristoph Hellwig .range_cyclic = 0, 226883ba7b07SChristoph Hellwig .done = &done, 22690e175a18SCurt Wohlgemuth .reason = WB_REASON_SYNC, 22707747bd4bSDave Chinner .for_sync = 1, 22713c4d7165SChristoph Hellwig }; 2272e7972912STejun Heo struct backing_dev_info *bdi = sb->s_bdi; 22733c4d7165SChristoph Hellwig 2274006a0973STejun Heo /* 2275006a0973STejun Heo * Can't skip on !bdi_has_dirty() because we should wait for !dirty 2276006a0973STejun Heo * inodes under writeback and I_DIRTY_TIME inodes ignored by 2277006a0973STejun Heo * bdi_has_dirty() need to be written out too. 2278006a0973STejun Heo */ 2279006a0973STejun Heo if (bdi == &noop_backing_dev_info) 22806eedc701SJan Kara return; 2281cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 2282cf37e972SChristoph Hellwig 2283db125360STejun Heo bdi_split_work_to_wbs(bdi, &work, false); 2284cc395d7fSTejun Heo wb_wait_for_completion(bdi, &done); 228583ba7b07SChristoph Hellwig 2286b6e51316SJens Axboe wait_sb_inodes(sb); 2287d8a8559cSJens Axboe } 2288d8a8559cSJens Axboe EXPORT_SYMBOL(sync_inodes_sb); 22891da177e4SLinus Torvalds 22901da177e4SLinus Torvalds /** 22911da177e4SLinus Torvalds * write_inode_now - write an inode to disk 22921da177e4SLinus Torvalds * @inode: inode to write to disk 22931da177e4SLinus Torvalds * @sync: whether the write should be synchronous or not 22941da177e4SLinus Torvalds * 22957f04c26dSAndrea Arcangeli * This function commits an inode to disk immediately if it is dirty. This is 22967f04c26dSAndrea Arcangeli * primarily needed by knfsd. 22977f04c26dSAndrea Arcangeli * 22987f04c26dSAndrea Arcangeli * The caller must either have a ref on the inode or must have set I_WILL_FREE. 22991da177e4SLinus Torvalds */ 23001da177e4SLinus Torvalds int write_inode_now(struct inode *inode, int sync) 23011da177e4SLinus Torvalds { 2302f758eeabSChristoph Hellwig struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; 23031da177e4SLinus Torvalds struct writeback_control wbc = { 23041da177e4SLinus Torvalds .nr_to_write = LONG_MAX, 230518914b18SMike Galbraith .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, 2306111ebb6eSOGAWA Hirofumi .range_start = 0, 2307111ebb6eSOGAWA Hirofumi .range_end = LLONG_MAX, 23081da177e4SLinus Torvalds }; 23091da177e4SLinus Torvalds 23101da177e4SLinus Torvalds if (!mapping_cap_writeback_dirty(inode->i_mapping)) 231149364ce2SAndrew Morton wbc.nr_to_write = 0; 23121da177e4SLinus Torvalds 23131da177e4SLinus Torvalds might_sleep(); 23144f8ad655SJan Kara return writeback_single_inode(inode, wb, &wbc); 23151da177e4SLinus Torvalds } 23161da177e4SLinus Torvalds EXPORT_SYMBOL(write_inode_now); 23171da177e4SLinus Torvalds 23181da177e4SLinus Torvalds /** 23191da177e4SLinus Torvalds * sync_inode - write an inode and its pages to disk. 23201da177e4SLinus Torvalds * @inode: the inode to sync 23211da177e4SLinus Torvalds * @wbc: controls the writeback mode 23221da177e4SLinus Torvalds * 23231da177e4SLinus Torvalds * sync_inode() will write an inode and its pages to disk. It will also 23241da177e4SLinus Torvalds * correctly update the inode on its superblock's dirty inode lists and will 23251da177e4SLinus Torvalds * update inode->i_state. 23261da177e4SLinus Torvalds * 23271da177e4SLinus Torvalds * The caller must have a ref on the inode. 23281da177e4SLinus Torvalds */ 23291da177e4SLinus Torvalds int sync_inode(struct inode *inode, struct writeback_control *wbc) 23301da177e4SLinus Torvalds { 23314f8ad655SJan Kara return writeback_single_inode(inode, &inode_to_bdi(inode)->wb, wbc); 23321da177e4SLinus Torvalds } 23331da177e4SLinus Torvalds EXPORT_SYMBOL(sync_inode); 2334c3765016SChristoph Hellwig 2335c3765016SChristoph Hellwig /** 2336c691b9d9SAndrew Morton * sync_inode_metadata - write an inode to disk 2337c3765016SChristoph Hellwig * @inode: the inode to sync 2338c3765016SChristoph Hellwig * @wait: wait for I/O to complete. 2339c3765016SChristoph Hellwig * 2340c691b9d9SAndrew Morton * Write an inode to disk and adjust its dirty state after completion. 2341c3765016SChristoph Hellwig * 2342c3765016SChristoph Hellwig * Note: only writes the actual inode, no associated data or other metadata. 2343c3765016SChristoph Hellwig */ 2344c3765016SChristoph Hellwig int sync_inode_metadata(struct inode *inode, int wait) 2345c3765016SChristoph Hellwig { 2346c3765016SChristoph Hellwig struct writeback_control wbc = { 2347c3765016SChristoph Hellwig .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE, 2348c3765016SChristoph Hellwig .nr_to_write = 0, /* metadata-only */ 2349c3765016SChristoph Hellwig }; 2350c3765016SChristoph Hellwig 2351c3765016SChristoph Hellwig return sync_inode(inode, &wbc); 2352c3765016SChristoph Hellwig } 2353c3765016SChristoph Hellwig EXPORT_SYMBOL(sync_inode_metadata); 2354