blob: c399152de397b75dba35bafc7a13b77c86db3aa1 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * fs/fs-writeback.c
3 *
4 * Copyright (C) 2002, Linus Torvalds.
5 *
6 * Contains all the functions related to writing back and waiting
7 * upon dirty inodes against superblocks, and writing back dirty
8 * pages against inodes. ie: data writeback. Writeout of the
9 * inode itself is not handled here.
10 *
Francois Camie1f8e872008-10-15 22:01:59 -070011 * 10Apr2002 Andrew Morton
Linus Torvalds1da177e2005-04-16 15:20:36 -070012 * Split out of fs/inode.c
13 * Additions for address_space-based writeback
14 */
15
16#include <linux/kernel.h>
Paul Gortmaker630d9c42011-11-16 23:57:37 -050017#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include <linux/spinlock.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090019#include <linux/slab.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/sched.h>
21#include <linux/fs.h>
22#include <linux/mm.h>
Wu Fengguangbc31b862012-01-07 20:41:55 -060023#include <linux/pagemap.h>
Jens Axboe03ba3782009-09-09 09:08:54 +020024#include <linux/kthread.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/writeback.h>
26#include <linux/blkdev.h>
27#include <linux/backing-dev.h>
Dave Chinner455b2862010-07-07 13:24:06 +100028#include <linux/tracepoint.h>
Al Viro719ea2f2013-09-29 11:24:49 -040029#include <linux/device.h>
David Howells07f3f052006-09-30 20:52:18 +020030#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070031
Jens Axboed0bceac2009-05-18 08:20:32 +020032/*
Wu Fengguangbc31b862012-01-07 20:41:55 -060033 * 4MB minimal write chunk size
34 */
35#define MIN_WRITEBACK_PAGES (4096UL >> (PAGE_CACHE_SHIFT - 10))
36
37/*
Jens Axboec4a77a62009-09-16 15:18:25 +020038 * Passed into wb_writeback(), essentially a subset of writeback_control
39 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +020040struct wb_writeback_work {
Jens Axboec4a77a62009-09-16 15:18:25 +020041 long nr_pages;
42 struct super_block *sb;
Jan Kara0dc83bd2014-02-21 11:19:04 +010043 unsigned long *older_than_this;
Jens Axboec4a77a62009-09-16 15:18:25 +020044 enum writeback_sync_modes sync_mode;
Wu Fengguang6e6938b2010-06-06 10:38:15 -060045 unsigned int tagged_writepages:1;
H Hartley Sweeten52957fe2010-04-01 20:36:30 -050046 unsigned int for_kupdate:1;
47 unsigned int range_cyclic:1;
48 unsigned int for_background:1;
Dave Chinner7747bd42013-07-02 22:38:35 +100049 unsigned int for_sync:1; /* sync(2) WB_SYNC_ALL writeback */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -060050 enum wb_reason reason; /* why was writeback initiated? */
Jens Axboec4a77a62009-09-16 15:18:25 +020051
Jens Axboe8010c3b2009-09-15 20:04:57 +020052 struct list_head list; /* pending work list */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +020053 struct completion *done; /* set if the caller waits */
Jens Axboe03ba3782009-09-09 09:08:54 +020054};
55
Adrian Bunkf11b00f2008-04-29 00:58:56 -070056/**
57 * writeback_in_progress - determine whether there is writeback in progress
58 * @bdi: the device's backing_dev_info structure.
59 *
Jens Axboe03ba3782009-09-09 09:08:54 +020060 * Determine whether there is writeback waiting to be handled against a
61 * backing device.
Adrian Bunkf11b00f2008-04-29 00:58:56 -070062 */
63int writeback_in_progress(struct backing_dev_info *bdi)
64{
Jan Kara81d73a32010-08-11 14:17:44 -070065 return test_bit(BDI_writeback_running, &bdi->state);
Adrian Bunkf11b00f2008-04-29 00:58:56 -070066}
Theodore Ts'o00d4e732012-09-19 22:42:36 -040067EXPORT_SYMBOL(writeback_in_progress);
Adrian Bunkf11b00f2008-04-29 00:58:56 -070068
Christoph Hellwigde1414a2015-01-14 10:42:36 +010069struct backing_dev_info *inode_to_bdi(struct inode *inode)
Jan Kara692ebd12010-09-21 11:51:01 +020070{
Jens Axboeb5202522015-01-22 08:13:17 -070071 struct super_block *sb;
72
73 if (!inode)
74 return &noop_backing_dev_info;
75
76 sb = inode->i_sb;
Christoph Hellwig495a2762015-01-14 10:42:34 +010077#ifdef CONFIG_BLOCK
Jan Karaa8855992013-07-09 22:36:45 +080078 if (sb_is_blkdev_sb(sb))
Christoph Hellwig495a2762015-01-14 10:42:34 +010079 return blk_get_backing_dev_info(I_BDEV(inode));
80#endif
Christoph Hellwigaaead252010-10-04 14:25:33 +020081 return sb->s_bdi;
Jan Kara692ebd12010-09-21 11:51:01 +020082}
Christoph Hellwigde1414a2015-01-14 10:42:36 +010083EXPORT_SYMBOL_GPL(inode_to_bdi);
Jan Kara692ebd12010-09-21 11:51:01 +020084
Nick Piggin7ccf19a2010-10-21 11:49:30 +110085static inline struct inode *wb_inode(struct list_head *head)
86{
87 return list_entry(head, struct inode, i_wb_list);
88}
89
Wu Fengguang15eb77a2012-01-17 11:18:56 -060090/*
91 * Include the creation of the trace points after defining the
92 * wb_writeback_work structure and inline functions so that the definition
93 * remains local to this file.
94 */
95#define CREATE_TRACE_POINTS
96#include <trace/events/writeback.h>
97
Steven Whitehouse774016b2014-02-06 15:47:47 +000098EXPORT_TRACEPOINT_SYMBOL_GPL(wbc_writepage);
99
Jan Kara5acda9d2014-04-03 14:46:23 -0700100static void bdi_wakeup_thread(struct backing_dev_info *bdi)
101{
102 spin_lock_bh(&bdi->wb_lock);
103 if (test_bit(BDI_registered, &bdi->state))
104 mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0);
105 spin_unlock_bh(&bdi->wb_lock);
106}
107
Jan Kara65850272011-01-13 15:45:44 -0800108static void bdi_queue_work(struct backing_dev_info *bdi,
109 struct wb_writeback_work *work)
110{
111 trace_writeback_queue(bdi, work);
112
113 spin_lock_bh(&bdi->wb_lock);
Jan Kara5acda9d2014-04-03 14:46:23 -0700114 if (!test_bit(BDI_registered, &bdi->state)) {
115 if (work->done)
116 complete(work->done);
117 goto out_unlock;
118 }
Jan Kara65850272011-01-13 15:45:44 -0800119 list_add_tail(&work->list, &bdi->work_list);
Tejun Heo839a8e82013-04-01 19:08:06 -0700120 mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0);
Jan Kara5acda9d2014-04-03 14:46:23 -0700121out_unlock:
122 spin_unlock_bh(&bdi->wb_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123}
124
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200125static void
126__bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600127 bool range_cyclic, enum wb_reason reason)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200129 struct wb_writeback_work *work;
Jens Axboe03ba3782009-09-09 09:08:54 +0200130
Jens Axboebcddc3f2009-09-13 20:07:36 +0200131 /*
132 * This is WB_SYNC_NONE writeback, so if allocation fails just
133 * wakeup the thread for old dirty data writeback
134 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200135 work = kzalloc(sizeof(*work), GFP_ATOMIC);
136 if (!work) {
Tejun Heo839a8e82013-04-01 19:08:06 -0700137 trace_writeback_nowork(bdi);
Jan Kara5acda9d2014-04-03 14:46:23 -0700138 bdi_wakeup_thread(bdi);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200139 return;
Jens Axboebcddc3f2009-09-13 20:07:36 +0200140 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200141
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200142 work->sync_mode = WB_SYNC_NONE;
143 work->nr_pages = nr_pages;
144 work->range_cyclic = range_cyclic;
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600145 work->reason = reason;
Christoph Hellwigf0fad8a2009-09-11 09:47:56 +0200146
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200147 bdi_queue_work(bdi, work);
Jens Axboeb6e51312009-09-16 15:13:54 +0200148}
149
150/**
151 * bdi_start_writeback - start writeback
152 * @bdi: the backing device to write from
153 * @nr_pages: the number of pages to write
Marcos Paulo de Souza786228a2011-11-23 20:56:45 +0800154 * @reason: reason why some writeback work was initiated
Jens Axboeb6e51312009-09-16 15:13:54 +0200155 *
156 * Description:
157 * This does WB_SYNC_NONE opportunistic writeback. The IO is only
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300158 * started when this function returns, we make no guarantees on
Jens Axboe0e3c9a22010-06-01 11:08:43 +0200159 * completion. Caller need not hold sb s_umount semaphore.
Jens Axboeb6e51312009-09-16 15:13:54 +0200160 *
161 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600162void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages,
163 enum wb_reason reason)
Jens Axboeb6e51312009-09-16 15:13:54 +0200164{
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600165 __bdi_start_writeback(bdi, nr_pages, true, reason);
Christoph Hellwigc5444192010-06-08 18:15:15 +0200166}
Wu Fengguangd3ddec72009-09-23 20:33:40 +0800167
Christoph Hellwigc5444192010-06-08 18:15:15 +0200168/**
169 * bdi_start_background_writeback - start background writeback
170 * @bdi: the backing device to write from
171 *
172 * Description:
Jan Kara65850272011-01-13 15:45:44 -0800173 * This makes sure WB_SYNC_NONE background writeback happens. When
174 * this function returns, it is only guaranteed that for given BDI
175 * some IO is happening if we are over background dirty threshold.
176 * Caller need not hold sb s_umount semaphore.
Christoph Hellwigc5444192010-06-08 18:15:15 +0200177 */
178void bdi_start_background_writeback(struct backing_dev_info *bdi)
179{
Jan Kara65850272011-01-13 15:45:44 -0800180 /*
181 * We just wake up the flusher thread. It will perform background
182 * writeback as soon as there is no other work to do.
183 */
Wu Fengguang71927e82011-01-13 15:45:46 -0800184 trace_writeback_wake_background(bdi);
Jan Kara5acda9d2014-04-03 14:46:23 -0700185 bdi_wakeup_thread(bdi);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186}
187
188/*
Dave Chinnera66979a2011-03-22 22:23:41 +1100189 * Remove the inode from the writeback list it is on.
190 */
191void inode_wb_list_del(struct inode *inode)
192{
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600193 struct backing_dev_info *bdi = inode_to_bdi(inode);
Dave Chinnera66979a2011-03-22 22:23:41 +1100194
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600195 spin_lock(&bdi->wb.list_lock);
196 list_del_init(&inode->i_wb_list);
197 spin_unlock(&bdi->wb.list_lock);
198}
Dave Chinnera66979a2011-03-22 22:23:41 +1100199
200/*
Andrew Morton6610a0b2007-10-16 23:30:32 -0700201 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
202 * furthest end of its superblock's dirty-inode list.
203 *
204 * Before stamping the inode's ->dirtied_when, we check to see whether it is
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200205 * already the most-recently-dirtied inode on the b_dirty list. If that is
Andrew Morton6610a0b2007-10-16 23:30:32 -0700206 * the case then the inode must have been redirtied while it was being written
207 * out and we don't reset its dirtied_when.
208 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600209static void redirty_tail(struct inode *inode, struct bdi_writeback *wb)
Andrew Morton6610a0b2007-10-16 23:30:32 -0700210{
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600211 assert_spin_locked(&wb->list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200212 if (!list_empty(&wb->b_dirty)) {
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200213 struct inode *tail;
Andrew Morton6610a0b2007-10-16 23:30:32 -0700214
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100215 tail = wb_inode(wb->b_dirty.next);
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200216 if (time_before(inode->dirtied_when, tail->dirtied_when))
Andrew Morton6610a0b2007-10-16 23:30:32 -0700217 inode->dirtied_when = jiffies;
218 }
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100219 list_move(&inode->i_wb_list, &wb->b_dirty);
Andrew Morton6610a0b2007-10-16 23:30:32 -0700220}
221
222/*
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200223 * requeue inode for re-scanning after bdi->b_io list is exhausted.
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700224 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600225static void requeue_io(struct inode *inode, struct bdi_writeback *wb)
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700226{
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600227 assert_spin_locked(&wb->list_lock);
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100228 list_move(&inode->i_wb_list, &wb->b_more_io);
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700229}
230
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700231static void inode_sync_complete(struct inode *inode)
232{
Jan Kara365b94ae2012-05-03 14:47:55 +0200233 inode->i_state &= ~I_SYNC;
Jan Kara4eff96d2012-11-26 16:29:51 -0800234 /* If inode is clean an unused, put it into LRU now... */
235 inode_add_lru(inode);
Jan Kara365b94ae2012-05-03 14:47:55 +0200236 /* Waiters must see I_SYNC cleared before being woken up */
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700237 smp_mb();
238 wake_up_bit(&inode->i_state, __I_SYNC);
239}
240
Jeff Laytond2caa3c52009-04-02 16:56:37 -0700241static bool inode_dirtied_after(struct inode *inode, unsigned long t)
242{
243 bool ret = time_after(inode->dirtied_when, t);
244#ifndef CONFIG_64BIT
245 /*
246 * For inodes being constantly redirtied, dirtied_when can get stuck.
247 * It _appears_ to be in the future, but is actually in distant past.
248 * This test is necessary to prevent such wrapped-around relative times
Jens Axboe5b0830c2009-09-23 19:37:09 +0200249 * from permanently stopping the whole bdi writeback.
Jeff Laytond2caa3c52009-04-02 16:56:37 -0700250 */
251 ret = ret && time_before_eq(inode->dirtied_when, jiffies);
252#endif
253 return ret;
254}
255
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700256/*
Wang Sheng-Hui0e2f2b22012-09-11 08:28:18 +0800257 * Move expired (dirtied before work->older_than_this) dirty inodes from
Jan Kara697e6fe2012-03-09 07:26:22 -0800258 * @delaying_queue to @dispatch_queue.
Fengguang Wu2c136572007-10-16 23:30:39 -0700259 */
Wu Fengguange84d0a42011-04-23 12:27:27 -0600260static int move_expired_inodes(struct list_head *delaying_queue,
Fengguang Wu2c136572007-10-16 23:30:39 -0700261 struct list_head *dispatch_queue,
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600262 struct wb_writeback_work *work)
Fengguang Wu2c136572007-10-16 23:30:39 -0700263{
Shaohua Li5c034492009-09-24 14:42:33 +0200264 LIST_HEAD(tmp);
265 struct list_head *pos, *node;
Jens Axboecf137302009-09-24 15:12:57 +0200266 struct super_block *sb = NULL;
Shaohua Li5c034492009-09-24 14:42:33 +0200267 struct inode *inode;
Jens Axboecf137302009-09-24 15:12:57 +0200268 int do_sb_sort = 0;
Wu Fengguange84d0a42011-04-23 12:27:27 -0600269 int moved = 0;
Shaohua Li5c034492009-09-24 14:42:33 +0200270
Fengguang Wu2c136572007-10-16 23:30:39 -0700271 while (!list_empty(delaying_queue)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100272 inode = wb_inode(delaying_queue->prev);
Jan Kara0dc83bd2014-02-21 11:19:04 +0100273 if (work->older_than_this &&
274 inode_dirtied_after(inode, *work->older_than_this))
Fengguang Wu2c136572007-10-16 23:30:39 -0700275 break;
Jan Karaa8855992013-07-09 22:36:45 +0800276 list_move(&inode->i_wb_list, &tmp);
277 moved++;
278 if (sb_is_blkdev_sb(inode->i_sb))
279 continue;
Jens Axboecf137302009-09-24 15:12:57 +0200280 if (sb && sb != inode->i_sb)
281 do_sb_sort = 1;
282 sb = inode->i_sb;
Shaohua Li5c034492009-09-24 14:42:33 +0200283 }
284
Jens Axboecf137302009-09-24 15:12:57 +0200285 /* just one sb in list, splice to dispatch_queue and we're done */
286 if (!do_sb_sort) {
287 list_splice(&tmp, dispatch_queue);
Wu Fengguange84d0a42011-04-23 12:27:27 -0600288 goto out;
Jens Axboecf137302009-09-24 15:12:57 +0200289 }
290
Shaohua Li5c034492009-09-24 14:42:33 +0200291 /* Move inodes from one superblock together */
292 while (!list_empty(&tmp)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100293 sb = wb_inode(tmp.prev)->i_sb;
Shaohua Li5c034492009-09-24 14:42:33 +0200294 list_for_each_prev_safe(pos, node, &tmp) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100295 inode = wb_inode(pos);
Shaohua Li5c034492009-09-24 14:42:33 +0200296 if (inode->i_sb == sb)
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100297 list_move(&inode->i_wb_list, dispatch_queue);
Shaohua Li5c034492009-09-24 14:42:33 +0200298 }
Fengguang Wu2c136572007-10-16 23:30:39 -0700299 }
Wu Fengguange84d0a42011-04-23 12:27:27 -0600300out:
301 return moved;
Fengguang Wu2c136572007-10-16 23:30:39 -0700302}
303
304/*
305 * Queue all expired dirty inodes for io, eldest first.
Wu Fengguang4ea879b2010-08-11 14:17:42 -0700306 * Before
307 * newly dirtied b_dirty b_io b_more_io
308 * =============> gf edc BA
309 * After
310 * newly dirtied b_dirty b_io b_more_io
311 * =============> g fBAedc
312 * |
313 * +--> dequeue for IO
Fengguang Wu2c136572007-10-16 23:30:39 -0700314 */
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600315static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work)
Fengguang Wu2c136572007-10-16 23:30:39 -0700316{
Wu Fengguange84d0a42011-04-23 12:27:27 -0600317 int moved;
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600318 assert_spin_locked(&wb->list_lock);
Wu Fengguang4ea879b2010-08-11 14:17:42 -0700319 list_splice_init(&wb->b_more_io, &wb->b_io);
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600320 moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, work);
321 trace_writeback_queue_io(wb, work, moved);
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200322}
323
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100324static int write_inode(struct inode *inode, struct writeback_control *wbc)
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200325{
Tejun Heo9fb0a7d2013-01-11 13:06:37 -0800326 int ret;
327
328 if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) {
329 trace_writeback_write_inode_start(inode, wbc);
330 ret = inode->i_sb->s_op->write_inode(inode, wbc);
331 trace_writeback_write_inode(inode, wbc);
332 return ret;
333 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200334 return 0;
Fengguang Wu2c136572007-10-16 23:30:39 -0700335}
336
337/*
Jan Kara169ebd92012-05-03 14:48:03 +0200338 * Wait for writeback on an inode to complete. Called with i_lock held.
339 * Caller must make sure inode cannot go away when we drop i_lock.
Christoph Hellwig01c03192009-06-08 13:35:40 +0200340 */
Jan Kara169ebd92012-05-03 14:48:03 +0200341static void __inode_wait_for_writeback(struct inode *inode)
342 __releases(inode->i_lock)
343 __acquires(inode->i_lock)
Christoph Hellwig01c03192009-06-08 13:35:40 +0200344{
345 DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC);
346 wait_queue_head_t *wqh;
347
348 wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
Dave Chinner250df6e2011-03-22 22:23:36 +1100349 while (inode->i_state & I_SYNC) {
350 spin_unlock(&inode->i_lock);
NeilBrown74316202014-07-07 15:16:04 +1000351 __wait_on_bit(wqh, &wq, bit_wait,
352 TASK_UNINTERRUPTIBLE);
Dave Chinner250df6e2011-03-22 22:23:36 +1100353 spin_lock(&inode->i_lock);
Richard Kennedy58a9d3d82010-05-24 14:32:38 -0700354 }
Christoph Hellwig01c03192009-06-08 13:35:40 +0200355}
356
357/*
Jan Kara169ebd92012-05-03 14:48:03 +0200358 * Wait for writeback on an inode to complete. Caller must have inode pinned.
359 */
360void inode_wait_for_writeback(struct inode *inode)
361{
362 spin_lock(&inode->i_lock);
363 __inode_wait_for_writeback(inode);
364 spin_unlock(&inode->i_lock);
365}
366
367/*
368 * Sleep until I_SYNC is cleared. This function must be called with i_lock
369 * held and drops it. It is aimed for callers not holding any inode reference
370 * so once i_lock is dropped, inode can go away.
371 */
372static void inode_sleep_on_writeback(struct inode *inode)
373 __releases(inode->i_lock)
374{
375 DEFINE_WAIT(wait);
376 wait_queue_head_t *wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
377 int sleep;
378
379 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
380 sleep = inode->i_state & I_SYNC;
381 spin_unlock(&inode->i_lock);
382 if (sleep)
383 schedule();
384 finish_wait(wqh, &wait);
385}
386
387/*
Jan Karaccb26b52012-05-03 14:47:58 +0200388 * Find proper writeback list for the inode depending on its current state and
389 * possibly also change of its state while we were doing writeback. Here we
390 * handle things such as livelock prevention or fairness of writeback among
391 * inodes. This function can be called only by flusher thread - noone else
392 * processes all inodes in writeback lists and requeueing inodes behind flusher
393 * thread's back can have unexpected consequences.
394 */
395static void requeue_inode(struct inode *inode, struct bdi_writeback *wb,
396 struct writeback_control *wbc)
397{
398 if (inode->i_state & I_FREEING)
399 return;
400
401 /*
402 * Sync livelock prevention. Each inode is tagged and synced in one
403 * shot. If still dirty, it will be redirty_tail()'ed below. Update
404 * the dirty time to prevent enqueue and sync it again.
405 */
406 if ((inode->i_state & I_DIRTY) &&
407 (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages))
408 inode->dirtied_when = jiffies;
409
Jan Kara4f8ad652012-05-03 14:48:00 +0200410 if (wbc->pages_skipped) {
411 /*
412 * writeback is not making progress due to locked
413 * buffers. Skip this inode for now.
414 */
415 redirty_tail(inode, wb);
416 return;
417 }
418
Jan Karaccb26b52012-05-03 14:47:58 +0200419 if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) {
420 /*
421 * We didn't write back all the pages. nfs_writepages()
422 * sometimes bales out without doing anything.
423 */
424 if (wbc->nr_to_write <= 0) {
425 /* Slice used up. Queue for next turn. */
426 requeue_io(inode, wb);
427 } else {
428 /*
429 * Writeback blocked by something other than
430 * congestion. Delay the inode for some time to
431 * avoid spinning on the CPU (100% iowait)
432 * retrying writeback of the dirty page/inode
433 * that cannot be performed immediately.
434 */
435 redirty_tail(inode, wb);
436 }
437 } else if (inode->i_state & I_DIRTY) {
438 /*
439 * Filesystems can dirty the inode during writeback operations,
440 * such as delayed allocation during submission or metadata
441 * updates after data IO completion.
442 */
443 redirty_tail(inode, wb);
444 } else {
445 /* The inode is clean. Remove from writeback lists. */
446 list_del_init(&inode->i_wb_list);
447 }
448}
449
450/*
Jan Kara4f8ad652012-05-03 14:48:00 +0200451 * Write out an inode and its dirty pages. Do not update the writeback list
452 * linkage. That is left to the caller. The caller is also responsible for
453 * setting I_SYNC flag and calling inode_sync_complete() to clear it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454 */
455static int
Yan Hongcd8ed2a2012-10-08 16:33:45 -0700456__writeback_single_inode(struct inode *inode, struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458 struct address_space *mapping = inode->i_mapping;
Wu Fengguang251d6a42010-12-01 17:33:37 -0600459 long nr_to_write = wbc->nr_to_write;
Christoph Hellwig01c03192009-06-08 13:35:40 +0200460 unsigned dirty;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 int ret;
462
Jan Kara4f8ad652012-05-03 14:48:00 +0200463 WARN_ON(!(inode->i_state & I_SYNC));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464
Tejun Heo9fb0a7d2013-01-11 13:06:37 -0800465 trace_writeback_single_inode_start(inode, wbc, nr_to_write);
466
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467 ret = do_writepages(mapping, wbc);
468
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100469 /*
470 * Make sure to wait on the data before writing out the metadata.
471 * This is important for filesystems that modify metadata on data
Dave Chinner7747bd42013-07-02 22:38:35 +1000472 * I/O completion. We don't do it for sync(2) writeback because it has a
473 * separate, external IO completion path and ->sync_fs for guaranteeing
474 * inode metadata is written back correctly.
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100475 */
Dave Chinner7747bd42013-07-02 22:38:35 +1000476 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) {
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100477 int err = filemap_fdatawait(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 if (ret == 0)
479 ret = err;
480 }
481
Dmitry Monakhov5547e8a2010-05-07 13:35:44 +0400482 /*
483 * Some filesystems may redirty the inode during the writeback
484 * due to delalloc, clear dirty metadata flags right before
485 * write_inode()
486 */
Dave Chinner250df6e2011-03-22 22:23:36 +1100487 spin_lock(&inode->i_lock);
Tejun Heo9c6ac782014-10-24 15:38:21 -0400488
Dmitry Monakhov5547e8a2010-05-07 13:35:44 +0400489 dirty = inode->i_state & I_DIRTY;
Tejun Heo9c6ac782014-10-24 15:38:21 -0400490 inode->i_state &= ~I_DIRTY;
491
492 /*
493 * Paired with smp_mb() in __mark_inode_dirty(). This allows
494 * __mark_inode_dirty() to test i_state without grabbing i_lock -
495 * either they see the I_DIRTY bits cleared or we see the dirtied
496 * inode.
497 *
498 * I_DIRTY_PAGES is always cleared together above even if @mapping
499 * still has dirty pages. The flag is reinstated after smp_mb() if
500 * necessary. This guarantees that either __mark_inode_dirty()
501 * sees clear I_DIRTY_PAGES or we see PAGECACHE_TAG_DIRTY.
502 */
503 smp_mb();
504
505 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
506 inode->i_state |= I_DIRTY_PAGES;
507
Dave Chinner250df6e2011-03-22 22:23:36 +1100508 spin_unlock(&inode->i_lock);
Tejun Heo9c6ac782014-10-24 15:38:21 -0400509
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100510 /* Don't write the inode if only I_DIRTY_PAGES was set */
511 if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100512 int err = write_inode(inode, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513 if (ret == 0)
514 ret = err;
515 }
Jan Kara4f8ad652012-05-03 14:48:00 +0200516 trace_writeback_single_inode(inode, wbc, nr_to_write);
517 return ret;
518}
519
520/*
521 * Write out an inode's dirty pages. Either the caller has an active reference
522 * on the inode or the inode has I_WILL_FREE set.
523 *
524 * This function is designed to be called for writing back one inode which
525 * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode()
526 * and does more profound writeback list handling in writeback_sb_inodes().
527 */
528static int
529writeback_single_inode(struct inode *inode, struct bdi_writeback *wb,
530 struct writeback_control *wbc)
531{
532 int ret = 0;
533
534 spin_lock(&inode->i_lock);
535 if (!atomic_read(&inode->i_count))
536 WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING)));
537 else
538 WARN_ON(inode->i_state & I_WILL_FREE);
539
540 if (inode->i_state & I_SYNC) {
541 if (wbc->sync_mode != WB_SYNC_ALL)
542 goto out;
543 /*
Jan Kara169ebd92012-05-03 14:48:03 +0200544 * It's a data-integrity sync. We must wait. Since callers hold
545 * inode reference or inode has I_WILL_FREE set, it cannot go
546 * away under us.
Jan Kara4f8ad652012-05-03 14:48:00 +0200547 */
Jan Kara169ebd92012-05-03 14:48:03 +0200548 __inode_wait_for_writeback(inode);
Jan Kara4f8ad652012-05-03 14:48:00 +0200549 }
550 WARN_ON(inode->i_state & I_SYNC);
551 /*
Jan Karaf9b0e052013-12-14 04:21:26 +0800552 * Skip inode if it is clean and we have no outstanding writeback in
553 * WB_SYNC_ALL mode. We don't want to mess with writeback lists in this
554 * function since flusher thread may be doing for example sync in
555 * parallel and if we move the inode, it could get skipped. So here we
556 * make sure inode is on some writeback list and leave it there unless
557 * we have completely cleaned the inode.
Jan Kara4f8ad652012-05-03 14:48:00 +0200558 */
Jan Karaf9b0e052013-12-14 04:21:26 +0800559 if (!(inode->i_state & I_DIRTY) &&
560 (wbc->sync_mode != WB_SYNC_ALL ||
561 !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK)))
Jan Kara4f8ad652012-05-03 14:48:00 +0200562 goto out;
563 inode->i_state |= I_SYNC;
564 spin_unlock(&inode->i_lock);
565
Yan Hongcd8ed2a2012-10-08 16:33:45 -0700566 ret = __writeback_single_inode(inode, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600568 spin_lock(&wb->list_lock);
Dave Chinner250df6e2011-03-22 22:23:36 +1100569 spin_lock(&inode->i_lock);
Jan Kara4f8ad652012-05-03 14:48:00 +0200570 /*
571 * If inode is clean, remove it from writeback lists. Otherwise don't
572 * touch it. See comment above for explanation.
573 */
574 if (!(inode->i_state & I_DIRTY))
575 list_del_init(&inode->i_wb_list);
576 spin_unlock(&wb->list_lock);
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700577 inode_sync_complete(inode);
Jan Kara4f8ad652012-05-03 14:48:00 +0200578out:
579 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580 return ret;
581}
582
Wu Fengguang1a12d8b2010-08-29 13:28:09 -0600583static long writeback_chunk_size(struct backing_dev_info *bdi,
584 struct wb_writeback_work *work)
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600585{
586 long pages;
587
588 /*
589 * WB_SYNC_ALL mode does livelock avoidance by syncing dirty
590 * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX
591 * here avoids calling into writeback_inodes_wb() more than once.
592 *
593 * The intended call sequence for WB_SYNC_ALL writeback is:
594 *
595 * wb_writeback()
596 * writeback_sb_inodes() <== called only once
597 * write_cache_pages() <== called once for each inode
598 * (quickly) tag currently dirty pages
599 * (maybe slowly) sync all tagged pages
600 */
601 if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages)
602 pages = LONG_MAX;
Wu Fengguang1a12d8b2010-08-29 13:28:09 -0600603 else {
604 pages = min(bdi->avg_write_bandwidth / 2,
605 global_dirty_limit / DIRTY_SCOPE);
606 pages = min(pages, work->nr_pages);
607 pages = round_down(pages + MIN_WRITEBACK_PAGES,
608 MIN_WRITEBACK_PAGES);
609 }
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600610
611 return pages;
612}
613
Jens Axboe03ba3782009-09-09 09:08:54 +0200614/*
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800615 * Write a portion of b_io inodes which belong to @sb.
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200616 *
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600617 * Return the number of pages and/or inodes written.
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800618 */
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600619static long writeback_sb_inodes(struct super_block *sb,
620 struct bdi_writeback *wb,
621 struct wb_writeback_work *work)
Jens Axboe03ba3782009-09-09 09:08:54 +0200622{
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600623 struct writeback_control wbc = {
624 .sync_mode = work->sync_mode,
625 .tagged_writepages = work->tagged_writepages,
626 .for_kupdate = work->for_kupdate,
627 .for_background = work->for_background,
Dave Chinner7747bd42013-07-02 22:38:35 +1000628 .for_sync = work->for_sync,
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600629 .range_cyclic = work->range_cyclic,
630 .range_start = 0,
631 .range_end = LLONG_MAX,
632 };
633 unsigned long start_time = jiffies;
634 long write_chunk;
635 long wrote = 0; /* count both pages and inodes */
636
Jens Axboe03ba3782009-09-09 09:08:54 +0200637 while (!list_empty(&wb->b_io)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100638 struct inode *inode = wb_inode(wb->b_io.prev);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200639
640 if (inode->i_sb != sb) {
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600641 if (work->sb) {
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200642 /*
643 * We only want to write back data for this
644 * superblock, move all inodes not belonging
645 * to it back onto the dirty list.
646 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600647 redirty_tail(inode, wb);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200648 continue;
649 }
650
651 /*
652 * The inode belongs to a different superblock.
653 * Bounce back to the caller to unpin this and
654 * pin the next superblock.
655 */
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600656 break;
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200657 }
658
Christoph Hellwig9843b762010-10-24 19:40:46 +0200659 /*
Wanpeng Li331cbde2012-06-09 11:10:55 +0800660 * Don't bother with new inodes or inodes being freed, first
661 * kind does not need periodic writeout yet, and for the latter
Christoph Hellwig9843b762010-10-24 19:40:46 +0200662 * kind writeout is handled by the freer.
663 */
Dave Chinner250df6e2011-03-22 22:23:36 +1100664 spin_lock(&inode->i_lock);
Christoph Hellwig9843b762010-10-24 19:40:46 +0200665 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
Dave Chinner250df6e2011-03-22 22:23:36 +1100666 spin_unlock(&inode->i_lock);
Wu Fengguangfcc5c222011-07-11 23:08:50 -0700667 redirty_tail(inode, wb);
Nick Piggin7ef0d732009-03-12 14:31:38 -0700668 continue;
669 }
Jan Karacc1676d2012-05-03 14:47:56 +0200670 if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) {
671 /*
672 * If this inode is locked for writeback and we are not
673 * doing writeback-for-data-integrity, move it to
674 * b_more_io so that writeback can proceed with the
675 * other inodes on s_io.
676 *
677 * We'll have another go at writing back this inode
678 * when we completed a full scan of b_io.
679 */
680 spin_unlock(&inode->i_lock);
681 requeue_io(inode, wb);
682 trace_writeback_sb_inodes_requeue(inode);
683 continue;
684 }
Jan Karaf0d07b72012-05-03 14:47:59 +0200685 spin_unlock(&wb->list_lock);
686
Jan Kara4f8ad652012-05-03 14:48:00 +0200687 /*
688 * We already requeued the inode if it had I_SYNC set and we
689 * are doing WB_SYNC_NONE writeback. So this catches only the
690 * WB_SYNC_ALL case.
691 */
Jan Kara169ebd92012-05-03 14:48:03 +0200692 if (inode->i_state & I_SYNC) {
693 /* Wait for I_SYNC. This function drops i_lock... */
694 inode_sleep_on_writeback(inode);
695 /* Inode may be gone, start again */
Jan Karaead188f2012-06-08 17:07:36 +0200696 spin_lock(&wb->list_lock);
Jan Kara169ebd92012-05-03 14:48:03 +0200697 continue;
698 }
Jan Kara4f8ad652012-05-03 14:48:00 +0200699 inode->i_state |= I_SYNC;
700 spin_unlock(&inode->i_lock);
Jan Kara169ebd92012-05-03 14:48:03 +0200701
Wu Fengguang1a12d8b2010-08-29 13:28:09 -0600702 write_chunk = writeback_chunk_size(wb->bdi, work);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600703 wbc.nr_to_write = write_chunk;
704 wbc.pages_skipped = 0;
Dave Chinner250df6e2011-03-22 22:23:36 +1100705
Jan Kara169ebd92012-05-03 14:48:03 +0200706 /*
707 * We use I_SYNC to pin the inode in memory. While it is set
708 * evict_inode() will wait so the inode cannot be freed.
709 */
Yan Hongcd8ed2a2012-10-08 16:33:45 -0700710 __writeback_single_inode(inode, &wbc);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600711
712 work->nr_pages -= write_chunk - wbc.nr_to_write;
713 wrote += write_chunk - wbc.nr_to_write;
Jan Kara4f8ad652012-05-03 14:48:00 +0200714 spin_lock(&wb->list_lock);
715 spin_lock(&inode->i_lock);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600716 if (!(inode->i_state & I_DIRTY))
717 wrote++;
Jan Kara4f8ad652012-05-03 14:48:00 +0200718 requeue_inode(inode, wb, &wbc);
719 inode_sync_complete(inode);
Dave Chinner0f1b1fd2011-03-22 22:23:43 +1100720 spin_unlock(&inode->i_lock);
Jan Kara169ebd92012-05-03 14:48:03 +0200721 cond_resched_lock(&wb->list_lock);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600722 /*
723 * bail out to wb_writeback() often enough to check
724 * background threshold and other termination conditions.
725 */
726 if (wrote) {
727 if (time_is_before_jiffies(start_time + HZ / 10UL))
728 break;
729 if (work->nr_pages <= 0)
730 break;
Fengguang Wu8bc3be22008-02-04 22:29:36 -0800731 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732 }
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600733 return wrote;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800734}
Nick Piggin38f21972009-01-06 14:40:25 -0800735
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600736static long __writeback_inodes_wb(struct bdi_writeback *wb,
737 struct wb_writeback_work *work)
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800738{
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600739 unsigned long start_time = jiffies;
740 long wrote = 0;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800741
742 while (!list_empty(&wb->b_io)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100743 struct inode *inode = wb_inode(wb->b_io.prev);
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800744 struct super_block *sb = inode->i_sb;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800745
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000746 if (!grab_super_passive(sb)) {
Wu Fengguang0e9958162011-07-29 22:14:35 -0600747 /*
748 * grab_super_passive() may fail consistently due to
749 * s_umount being grabbed by someone else. Don't use
750 * requeue_io() to avoid busy retrying the inode/sb.
751 */
752 redirty_tail(inode, wb);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200753 continue;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800754 }
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600755 wrote += writeback_sb_inodes(sb, wb, work);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200756 drop_super(sb);
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800757
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600758 /* refer to the same tests at the end of writeback_sb_inodes */
759 if (wrote) {
760 if (time_is_before_jiffies(start_time + HZ / 10UL))
761 break;
762 if (work->nr_pages <= 0)
763 break;
764 }
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800765 }
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200766 /* Leave any unwritten inodes on b_io */
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600767 return wrote;
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200768}
769
Wanpeng Li7d9f0732013-09-11 14:22:40 -0700770static long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600771 enum wb_reason reason)
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200772{
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600773 struct wb_writeback_work work = {
774 .nr_pages = nr_pages,
775 .sync_mode = WB_SYNC_NONE,
776 .range_cyclic = 1,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600777 .reason = reason,
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600778 };
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200779
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600780 spin_lock(&wb->list_lock);
Wu Fengguang424b3512010-07-21 20:11:53 -0600781 if (list_empty(&wb->b_io))
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600782 queue_io(wb, &work);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600783 __writeback_inodes_wb(wb, &work);
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600784 spin_unlock(&wb->list_lock);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200785
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600786 return nr_pages - work.nr_pages;
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200787}
788
Wu Fengguangb00949a2010-11-18 14:38:33 -0600789static bool over_bground_thresh(struct backing_dev_info *bdi)
Jens Axboe03ba3782009-09-09 09:08:54 +0200790{
791 unsigned long background_thresh, dirty_thresh;
792
Wu Fengguang16c40422010-08-11 14:17:39 -0700793 global_dirty_limits(&background_thresh, &dirty_thresh);
Jens Axboe03ba3782009-09-09 09:08:54 +0200794
Wu Fengguangb00949a2010-11-18 14:38:33 -0600795 if (global_page_state(NR_FILE_DIRTY) +
796 global_page_state(NR_UNSTABLE_NFS) > background_thresh)
797 return true;
798
799 if (bdi_stat(bdi, BDI_RECLAIMABLE) >
800 bdi_dirty_limit(bdi, background_thresh))
801 return true;
802
803 return false;
Jens Axboe03ba3782009-09-09 09:08:54 +0200804}
805
806/*
Wu Fengguange98be2d2010-08-29 11:22:30 -0600807 * Called under wb->list_lock. If there are multiple wb per bdi,
808 * only the flusher working on the first wb should do it.
809 */
810static void wb_update_bandwidth(struct bdi_writeback *wb,
811 unsigned long start_time)
812{
Wu Fengguangaf6a3112011-10-03 20:46:17 -0600813 __bdi_update_bandwidth(wb->bdi, 0, 0, 0, 0, 0, start_time);
Wu Fengguange98be2d2010-08-29 11:22:30 -0600814}
815
816/*
Jens Axboe03ba3782009-09-09 09:08:54 +0200817 * Explicit flushing or periodic writeback of "old" data.
818 *
819 * Define "old": the first time one of an inode's pages is dirtied, we mark the
820 * dirtying-time in the inode's address_space. So this periodic writeback code
821 * just walks the superblock inode list, writing back any inodes which are
822 * older than a specific point in time.
823 *
824 * Try to run once per dirty_writeback_interval. But if a writeback event
825 * takes longer than a dirty_writeback_interval interval, then leave a
826 * one-second gap.
827 *
828 * older_than_this takes precedence over nr_to_write. So we'll only write back
829 * all dirty pages if they are all attached to "old" mappings.
830 */
Jens Axboec4a77a62009-09-16 15:18:25 +0200831static long wb_writeback(struct bdi_writeback *wb,
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200832 struct wb_writeback_work *work)
Jens Axboe03ba3782009-09-09 09:08:54 +0200833{
Wu Fengguange98be2d2010-08-29 11:22:30 -0600834 unsigned long wb_start = jiffies;
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600835 long nr_pages = work->nr_pages;
Jan Kara0dc83bd2014-02-21 11:19:04 +0100836 unsigned long oldest_jif;
Jan Karaa5989bd2009-09-16 19:22:48 +0200837 struct inode *inode;
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600838 long progress;
Jens Axboe03ba3782009-09-09 09:08:54 +0200839
Jan Kara0dc83bd2014-02-21 11:19:04 +0100840 oldest_jif = jiffies;
841 work->older_than_this = &oldest_jif;
Jens Axboe03ba3782009-09-09 09:08:54 +0200842
Wu Fengguange8dfc302011-04-21 12:06:32 -0600843 spin_lock(&wb->list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200844 for (;;) {
845 /*
Wu Fengguangd3ddec72009-09-23 20:33:40 +0800846 * Stop writeback when nr_pages has been consumed
Jens Axboe03ba3782009-09-09 09:08:54 +0200847 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200848 if (work->nr_pages <= 0)
Jens Axboe03ba3782009-09-09 09:08:54 +0200849 break;
850
851 /*
Jan Karaaa373cf2011-01-13 15:45:47 -0800852 * Background writeout and kupdate-style writeback may
853 * run forever. Stop them if there is other work to do
854 * so that e.g. sync can proceed. They'll be restarted
855 * after the other works are all done.
856 */
857 if ((work->for_background || work->for_kupdate) &&
858 !list_empty(&wb->bdi->work_list))
859 break;
860
861 /*
Wu Fengguangd3ddec72009-09-23 20:33:40 +0800862 * For background writeout, stop when we are below the
863 * background dirty threshold
Jens Axboe03ba3782009-09-09 09:08:54 +0200864 */
Wu Fengguangb00949a2010-11-18 14:38:33 -0600865 if (work->for_background && !over_bground_thresh(wb->bdi))
Jens Axboe03ba3782009-09-09 09:08:54 +0200866 break;
867
Jan Kara1bc36b62011-10-19 11:44:41 +0200868 /*
869 * Kupdate and background works are special and we want to
870 * include all inodes that need writing. Livelock avoidance is
871 * handled by these works yielding to any other work so we are
872 * safe.
873 */
Wu Fengguangba9aa832010-07-21 20:32:30 -0600874 if (work->for_kupdate) {
Jan Kara0dc83bd2014-02-21 11:19:04 +0100875 oldest_jif = jiffies -
Wu Fengguangba9aa832010-07-21 20:32:30 -0600876 msecs_to_jiffies(dirty_expire_interval * 10);
Jan Kara1bc36b62011-10-19 11:44:41 +0200877 } else if (work->for_background)
Jan Kara0dc83bd2014-02-21 11:19:04 +0100878 oldest_jif = jiffies;
Dave Chinner028c2dd2010-07-07 13:24:07 +1000879
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600880 trace_writeback_start(wb->bdi, work);
Wu Fengguange8dfc302011-04-21 12:06:32 -0600881 if (list_empty(&wb->b_io))
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600882 queue_io(wb, work);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200883 if (work->sb)
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600884 progress = writeback_sb_inodes(work->sb, wb, work);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200885 else
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600886 progress = __writeback_inodes_wb(wb, work);
887 trace_writeback_written(wb->bdi, work);
Dave Chinner028c2dd2010-07-07 13:24:07 +1000888
Wu Fengguange98be2d2010-08-29 11:22:30 -0600889 wb_update_bandwidth(wb, wb_start);
Jens Axboe03ba3782009-09-09 09:08:54 +0200890
891 /*
Jens Axboe71fd05a2009-09-23 19:32:26 +0200892 * Did we write something? Try for more
Wu Fengguange6fb6da2010-07-22 10:23:44 -0600893 *
894 * Dirty inodes are moved to b_io for writeback in batches.
895 * The completion of the current batch does not necessarily
896 * mean the overall work is done. So we keep looping as long
897 * as made some progress on cleaning pages or inodes.
Jens Axboe71fd05a2009-09-23 19:32:26 +0200898 */
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600899 if (progress)
Jens Axboe71fd05a2009-09-23 19:32:26 +0200900 continue;
901 /*
Wu Fengguange6fb6da2010-07-22 10:23:44 -0600902 * No more inodes for IO, bail
Jens Axboe71fd05a2009-09-23 19:32:26 +0200903 */
Wu Fengguangb7a24412010-07-21 22:19:51 -0600904 if (list_empty(&wb->b_more_io))
Jens Axboe03ba3782009-09-09 09:08:54 +0200905 break;
906 /*
Jens Axboe8010c3b2009-09-15 20:04:57 +0200907 * Nothing written. Wait for some inode to
908 * become available for writeback. Otherwise
909 * we'll just busyloop.
Jens Axboe03ba3782009-09-09 09:08:54 +0200910 */
Jens Axboe03ba3782009-09-09 09:08:54 +0200911 if (!list_empty(&wb->b_more_io)) {
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600912 trace_writeback_wait(wb->bdi, work);
Jens Axboe03ba3782009-09-09 09:08:54 +0200913 inode = wb_inode(wb->b_more_io.prev);
Dave Chinner250df6e2011-03-22 22:23:36 +1100914 spin_lock(&inode->i_lock);
Jan Karaf0d07b72012-05-03 14:47:59 +0200915 spin_unlock(&wb->list_lock);
Jan Kara169ebd92012-05-03 14:48:03 +0200916 /* This function drops i_lock... */
917 inode_sleep_on_writeback(inode);
Jan Karaf0d07b72012-05-03 14:47:59 +0200918 spin_lock(&wb->list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200919 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200920 }
Wu Fengguange8dfc302011-04-21 12:06:32 -0600921 spin_unlock(&wb->list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200922
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600923 return nr_pages - work->nr_pages;
Jens Axboe03ba3782009-09-09 09:08:54 +0200924}
925
926/*
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200927 * Return the next wb_writeback_work struct that hasn't been processed yet.
Jens Axboe03ba3782009-09-09 09:08:54 +0200928 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200929static struct wb_writeback_work *
Minchan Kim08852b62010-08-03 12:51:16 +0200930get_next_work_item(struct backing_dev_info *bdi)
Jens Axboe03ba3782009-09-09 09:08:54 +0200931{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200932 struct wb_writeback_work *work = NULL;
Jens Axboe03ba3782009-09-09 09:08:54 +0200933
Artem Bityutskiy64677162010-07-25 14:29:22 +0300934 spin_lock_bh(&bdi->wb_lock);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200935 if (!list_empty(&bdi->work_list)) {
936 work = list_entry(bdi->work_list.next,
937 struct wb_writeback_work, list);
938 list_del_init(&work->list);
Jens Axboe03ba3782009-09-09 09:08:54 +0200939 }
Artem Bityutskiy64677162010-07-25 14:29:22 +0300940 spin_unlock_bh(&bdi->wb_lock);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200941 return work;
Jens Axboe03ba3782009-09-09 09:08:54 +0200942}
943
Linus Torvaldscdf01dd2010-10-30 08:55:52 -0700944/*
945 * Add in the number of potentially dirty inodes, because each inode
946 * write can dirty pagecache in the underlying blockdev.
947 */
948static unsigned long get_nr_dirty_pages(void)
949{
950 return global_page_state(NR_FILE_DIRTY) +
951 global_page_state(NR_UNSTABLE_NFS) +
952 get_nr_dirty_inodes();
953}
954
Jan Kara65850272011-01-13 15:45:44 -0800955static long wb_check_background_flush(struct bdi_writeback *wb)
956{
Wu Fengguangb00949a2010-11-18 14:38:33 -0600957 if (over_bground_thresh(wb->bdi)) {
Jan Kara65850272011-01-13 15:45:44 -0800958
959 struct wb_writeback_work work = {
960 .nr_pages = LONG_MAX,
961 .sync_mode = WB_SYNC_NONE,
962 .for_background = 1,
963 .range_cyclic = 1,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600964 .reason = WB_REASON_BACKGROUND,
Jan Kara65850272011-01-13 15:45:44 -0800965 };
966
967 return wb_writeback(wb, &work);
968 }
969
970 return 0;
971}
972
Jens Axboe03ba3782009-09-09 09:08:54 +0200973static long wb_check_old_data_flush(struct bdi_writeback *wb)
974{
975 unsigned long expired;
976 long nr_pages;
977
Jens Axboe69b62d02010-05-17 12:51:03 +0200978 /*
979 * When set to zero, disable periodic writeback
980 */
981 if (!dirty_writeback_interval)
982 return 0;
983
Jens Axboe03ba3782009-09-09 09:08:54 +0200984 expired = wb->last_old_flush +
985 msecs_to_jiffies(dirty_writeback_interval * 10);
986 if (time_before(jiffies, expired))
987 return 0;
988
989 wb->last_old_flush = jiffies;
Linus Torvaldscdf01dd2010-10-30 08:55:52 -0700990 nr_pages = get_nr_dirty_pages();
Jens Axboe03ba3782009-09-09 09:08:54 +0200991
Jens Axboec4a77a62009-09-16 15:18:25 +0200992 if (nr_pages) {
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200993 struct wb_writeback_work work = {
Jens Axboec4a77a62009-09-16 15:18:25 +0200994 .nr_pages = nr_pages,
995 .sync_mode = WB_SYNC_NONE,
996 .for_kupdate = 1,
997 .range_cyclic = 1,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600998 .reason = WB_REASON_PERIODIC,
Jens Axboec4a77a62009-09-16 15:18:25 +0200999 };
1000
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001001 return wb_writeback(wb, &work);
Jens Axboec4a77a62009-09-16 15:18:25 +02001002 }
Jens Axboe03ba3782009-09-09 09:08:54 +02001003
1004 return 0;
1005}
1006
1007/*
1008 * Retrieve work items and do the writeback they describe
1009 */
Wanpeng Li25d130b2013-07-08 16:00:14 -07001010static long wb_do_writeback(struct bdi_writeback *wb)
Jens Axboe03ba3782009-09-09 09:08:54 +02001011{
1012 struct backing_dev_info *bdi = wb->bdi;
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001013 struct wb_writeback_work *work;
Jens Axboec4a77a62009-09-16 15:18:25 +02001014 long wrote = 0;
Jens Axboe03ba3782009-09-09 09:08:54 +02001015
Jan Kara81d73a32010-08-11 14:17:44 -07001016 set_bit(BDI_writeback_running, &wb->bdi->state);
Minchan Kim08852b62010-08-03 12:51:16 +02001017 while ((work = get_next_work_item(bdi)) != NULL) {
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001018
Dave Chinner455b2862010-07-07 13:24:06 +10001019 trace_writeback_exec(bdi, work);
1020
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001021 wrote += wb_writeback(wb, work);
Jens Axboe03ba3782009-09-09 09:08:54 +02001022
1023 /*
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001024 * Notify the caller of completion if this is a synchronous
1025 * work item, otherwise just free it.
Jens Axboe03ba3782009-09-09 09:08:54 +02001026 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001027 if (work->done)
1028 complete(work->done);
1029 else
1030 kfree(work);
Jens Axboe03ba3782009-09-09 09:08:54 +02001031 }
1032
1033 /*
1034 * Check for periodic writeback, kupdated() style
1035 */
1036 wrote += wb_check_old_data_flush(wb);
Jan Kara65850272011-01-13 15:45:44 -08001037 wrote += wb_check_background_flush(wb);
Jan Kara81d73a32010-08-11 14:17:44 -07001038 clear_bit(BDI_writeback_running, &wb->bdi->state);
Jens Axboe03ba3782009-09-09 09:08:54 +02001039
1040 return wrote;
1041}
1042
1043/*
1044 * Handle writeback of dirty data for the device backed by this bdi. Also
Tejun Heo839a8e82013-04-01 19:08:06 -07001045 * reschedules periodically and does kupdated style flushing.
Jens Axboe03ba3782009-09-09 09:08:54 +02001046 */
Tejun Heo839a8e82013-04-01 19:08:06 -07001047void bdi_writeback_workfn(struct work_struct *work)
Jens Axboe03ba3782009-09-09 09:08:54 +02001048{
Tejun Heo839a8e82013-04-01 19:08:06 -07001049 struct bdi_writeback *wb = container_of(to_delayed_work(work),
1050 struct bdi_writeback, dwork);
Christoph Hellwig08243902010-06-19 23:08:22 +02001051 struct backing_dev_info *bdi = wb->bdi;
Jens Axboe03ba3782009-09-09 09:08:54 +02001052 long pages_written;
1053
Tejun Heoef3b1012013-04-30 15:27:24 -07001054 set_worker_desc("flush-%s", dev_name(bdi->dev));
Peter Zijlstra766f9162010-10-26 14:22:45 -07001055 current->flags |= PF_SWAPWRITE;
Christoph Hellwig08243902010-06-19 23:08:22 +02001056
Tejun Heo839a8e82013-04-01 19:08:06 -07001057 if (likely(!current_is_workqueue_rescuer() ||
Jan Kara5acda9d2014-04-03 14:46:23 -07001058 !test_bit(BDI_registered, &bdi->state))) {
Artem Bityutskiy64677162010-07-25 14:29:22 +03001059 /*
Tejun Heo839a8e82013-04-01 19:08:06 -07001060 * The normal path. Keep writing back @bdi until its
1061 * work_list is empty. Note that this path is also taken
1062 * if @bdi is shutting down even when we're running off the
1063 * rescuer as work_list needs to be drained.
Artem Bityutskiy64677162010-07-25 14:29:22 +03001064 */
Tejun Heo839a8e82013-04-01 19:08:06 -07001065 do {
Wanpeng Li25d130b2013-07-08 16:00:14 -07001066 pages_written = wb_do_writeback(wb);
Tejun Heo839a8e82013-04-01 19:08:06 -07001067 trace_writeback_pages_written(pages_written);
1068 } while (!list_empty(&bdi->work_list));
1069 } else {
1070 /*
1071 * bdi_wq can't get enough workers and we're running off
1072 * the emergency worker. Don't hog it. Hopefully, 1024 is
1073 * enough for efficient IO.
1074 */
1075 pages_written = writeback_inodes_wb(&bdi->wb, 1024,
1076 WB_REASON_FORKER_THREAD);
Dave Chinner455b2862010-07-07 13:24:06 +10001077 trace_writeback_pages_written(pages_written);
Jens Axboe03ba3782009-09-09 09:08:54 +02001078 }
1079
Derek Basehore6ca738d2014-04-03 14:46:22 -07001080 if (!list_empty(&bdi->work_list))
1081 mod_delayed_work(bdi_wq, &wb->dwork, 0);
1082 else if (wb_has_dirty_io(wb) && dirty_writeback_interval)
1083 bdi_wakeup_thread_delayed(bdi);
Dave Chinner455b2862010-07-07 13:24:06 +10001084
Tejun Heo839a8e82013-04-01 19:08:06 -07001085 current->flags &= ~PF_SWAPWRITE;
Jens Axboe03ba3782009-09-09 09:08:54 +02001086}
1087
1088/*
Jens Axboe03ba3782009-09-09 09:08:54 +02001089 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
1090 * the whole world.
1091 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001092void wakeup_flusher_threads(long nr_pages, enum wb_reason reason)
Jens Axboe03ba3782009-09-09 09:08:54 +02001093{
Christoph Hellwigb8c2f342010-06-08 18:15:07 +02001094 struct backing_dev_info *bdi;
Christoph Hellwigb8c2f342010-06-08 18:15:07 +02001095
Jan Kara47df3dd2013-09-11 14:22:22 -07001096 if (!nr_pages)
1097 nr_pages = get_nr_dirty_pages();
Christoph Hellwigb8c2f342010-06-08 18:15:07 +02001098
1099 rcu_read_lock();
1100 list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) {
1101 if (!bdi_has_dirty_io(bdi))
1102 continue;
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001103 __bdi_start_writeback(bdi, nr_pages, false, reason);
Christoph Hellwigb8c2f342010-06-08 18:15:07 +02001104 }
1105 rcu_read_unlock();
Jens Axboe03ba3782009-09-09 09:08:54 +02001106}
1107
1108static noinline void block_dump___mark_inode_dirty(struct inode *inode)
1109{
1110 if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) {
1111 struct dentry *dentry;
1112 const char *name = "?";
1113
1114 dentry = d_find_alias(inode);
1115 if (dentry) {
1116 spin_lock(&dentry->d_lock);
1117 name = (const char *) dentry->d_name.name;
1118 }
1119 printk(KERN_DEBUG
1120 "%s(%d): dirtied inode %lu (%s) on %s\n",
1121 current->comm, task_pid_nr(current), inode->i_ino,
1122 name, inode->i_sb->s_id);
1123 if (dentry) {
1124 spin_unlock(&dentry->d_lock);
1125 dput(dentry);
1126 }
1127 }
1128}
1129
1130/**
1131 * __mark_inode_dirty - internal function
1132 * @inode: inode to mark
1133 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
1134 * Mark an inode as dirty. Callers should use mark_inode_dirty or
1135 * mark_inode_dirty_sync.
1136 *
1137 * Put the inode on the super block's dirty list.
1138 *
1139 * CAREFUL! We mark it dirty unconditionally, but move it onto the
1140 * dirty list only if it is hashed or if it refers to a blockdev.
1141 * If it was not hashed, it will never be added to the dirty list
1142 * even if it is later hashed, as it will have been marked dirty already.
1143 *
1144 * In short, make sure you hash any inodes _before_ you start marking
1145 * them dirty.
1146 *
Jens Axboe03ba3782009-09-09 09:08:54 +02001147 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
1148 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
1149 * the kernel-internal blockdev inode represents the dirtying time of the
1150 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
1151 * page->mapping->host, so the page-dirtying time is recorded in the internal
1152 * blockdev inode.
1153 */
1154void __mark_inode_dirty(struct inode *inode, int flags)
1155{
1156 struct super_block *sb = inode->i_sb;
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001157 struct backing_dev_info *bdi = NULL;
Jens Axboe03ba3782009-09-09 09:08:54 +02001158
1159 /*
1160 * Don't do this for I_DIRTY_PAGES - that doesn't actually
1161 * dirty the inode itself
1162 */
1163 if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
Tejun Heo9fb0a7d2013-01-11 13:06:37 -08001164 trace_writeback_dirty_inode_start(inode, flags);
1165
Jens Axboe03ba3782009-09-09 09:08:54 +02001166 if (sb->s_op->dirty_inode)
Christoph Hellwigaa385722011-05-27 06:53:02 -04001167 sb->s_op->dirty_inode(inode, flags);
Tejun Heo9fb0a7d2013-01-11 13:06:37 -08001168
1169 trace_writeback_dirty_inode(inode, flags);
Jens Axboe03ba3782009-09-09 09:08:54 +02001170 }
1171
1172 /*
Tejun Heo9c6ac782014-10-24 15:38:21 -04001173 * Paired with smp_mb() in __writeback_single_inode() for the
1174 * following lockless i_state test. See there for details.
Jens Axboe03ba3782009-09-09 09:08:54 +02001175 */
1176 smp_mb();
1177
Jens Axboe03ba3782009-09-09 09:08:54 +02001178 if ((inode->i_state & flags) == flags)
1179 return;
1180
1181 if (unlikely(block_dump))
1182 block_dump___mark_inode_dirty(inode);
1183
Dave Chinner250df6e2011-03-22 22:23:36 +11001184 spin_lock(&inode->i_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001185 if ((inode->i_state & flags) != flags) {
1186 const int was_dirty = inode->i_state & I_DIRTY;
1187
1188 inode->i_state |= flags;
1189
1190 /*
1191 * If the inode is being synced, just update its dirty state.
1192 * The unlocker will place the inode on the appropriate
1193 * superblock list, based upon its state.
1194 */
1195 if (inode->i_state & I_SYNC)
Dave Chinner250df6e2011-03-22 22:23:36 +11001196 goto out_unlock_inode;
Jens Axboe03ba3782009-09-09 09:08:54 +02001197
1198 /*
1199 * Only add valid (hashed) inodes to the superblock's
1200 * dirty list. Add blockdev inodes as well.
1201 */
1202 if (!S_ISBLK(inode->i_mode)) {
Al Viro1d3382cb2010-10-23 15:19:20 -04001203 if (inode_unhashed(inode))
Dave Chinner250df6e2011-03-22 22:23:36 +11001204 goto out_unlock_inode;
Jens Axboe03ba3782009-09-09 09:08:54 +02001205 }
Al Viroa4ffdde2010-06-02 17:38:30 -04001206 if (inode->i_state & I_FREEING)
Dave Chinner250df6e2011-03-22 22:23:36 +11001207 goto out_unlock_inode;
Jens Axboe03ba3782009-09-09 09:08:54 +02001208
1209 /*
1210 * If the inode was already on b_dirty/b_io/b_more_io, don't
1211 * reposition it (that would break b_dirty time-ordering).
1212 */
1213 if (!was_dirty) {
Dave Chinnera66979a2011-03-22 22:23:41 +11001214 bool wakeup_bdi = false;
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001215 bdi = inode_to_bdi(inode);
Jens Axboe500b0672009-09-09 09:10:25 +02001216
Junxiao Bi146d7002013-09-11 14:23:04 -07001217 spin_unlock(&inode->i_lock);
1218 spin_lock(&bdi->wb.list_lock);
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001219 if (bdi_cap_writeback_dirty(bdi)) {
1220 WARN(!test_bit(BDI_registered, &bdi->state),
1221 "bdi-%s not registered\n", bdi->name);
1222
1223 /*
1224 * If this is the first dirty inode for this
1225 * bdi, we have to wake-up the corresponding
1226 * bdi thread to make sure background
1227 * write-back happens later.
1228 */
1229 if (!wb_has_dirty_io(&bdi->wb))
1230 wakeup_bdi = true;
Jens Axboe500b0672009-09-09 09:10:25 +02001231 }
Jens Axboe03ba3782009-09-09 09:08:54 +02001232
1233 inode->dirtied_when = jiffies;
Nick Piggin7ccf19a2010-10-21 11:49:30 +11001234 list_move(&inode->i_wb_list, &bdi->wb.b_dirty);
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001235 spin_unlock(&bdi->wb.list_lock);
Dave Chinnera66979a2011-03-22 22:23:41 +11001236
1237 if (wakeup_bdi)
1238 bdi_wakeup_thread_delayed(bdi);
1239 return;
Jens Axboe03ba3782009-09-09 09:08:54 +02001240 }
1241 }
Dave Chinner250df6e2011-03-22 22:23:36 +11001242out_unlock_inode:
1243 spin_unlock(&inode->i_lock);
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001244
Jens Axboe03ba3782009-09-09 09:08:54 +02001245}
1246EXPORT_SYMBOL(__mark_inode_dirty);
1247
Jens Axboeb6e51312009-09-16 15:13:54 +02001248static void wait_sb_inodes(struct super_block *sb)
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001249{
Jens Axboe03ba3782009-09-09 09:08:54 +02001250 struct inode *inode, *old_inode = NULL;
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001251
Jens Axboe03ba3782009-09-09 09:08:54 +02001252 /*
1253 * We need to be protected against the filesystem going from
1254 * r/o to r/w or vice versa.
1255 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001256 WARN_ON(!rwsem_is_locked(&sb->s_umount));
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001257
Dave Chinner55fa6092011-03-22 22:23:40 +11001258 spin_lock(&inode_sb_list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001259
1260 /*
1261 * Data integrity sync. Must wait for all pages under writeback,
1262 * because there may have been pages dirtied before our sync
1263 * call, but which had writeout started before we write it out.
1264 * In which case, the inode may not be on the dirty list, but
1265 * we still have to wait for that writeout.
1266 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001267 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
Dave Chinner250df6e2011-03-22 22:23:36 +11001268 struct address_space *mapping = inode->i_mapping;
Jens Axboe03ba3782009-09-09 09:08:54 +02001269
Dave Chinner250df6e2011-03-22 22:23:36 +11001270 spin_lock(&inode->i_lock);
1271 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
1272 (mapping->nrpages == 0)) {
1273 spin_unlock(&inode->i_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001274 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +11001275 }
Jens Axboe03ba3782009-09-09 09:08:54 +02001276 __iget(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +11001277 spin_unlock(&inode->i_lock);
Dave Chinner55fa6092011-03-22 22:23:40 +11001278 spin_unlock(&inode_sb_list_lock);
1279
Jens Axboe03ba3782009-09-09 09:08:54 +02001280 /*
Dave Chinner55fa6092011-03-22 22:23:40 +11001281 * We hold a reference to 'inode' so it couldn't have been
1282 * removed from s_inodes list while we dropped the
1283 * inode_sb_list_lock. We cannot iput the inode now as we can
1284 * be holding the last reference and we cannot iput it under
1285 * inode_sb_list_lock. So we keep the reference and iput it
1286 * later.
Jens Axboe03ba3782009-09-09 09:08:54 +02001287 */
1288 iput(old_inode);
1289 old_inode = inode;
1290
1291 filemap_fdatawait(mapping);
1292
1293 cond_resched();
Nick Piggin38f21972009-01-06 14:40:25 -08001294
Dave Chinner55fa6092011-03-22 22:23:40 +11001295 spin_lock(&inode_sb_list_lock);
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001296 }
Dave Chinner55fa6092011-03-22 22:23:40 +11001297 spin_unlock(&inode_sb_list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001298 iput(old_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299}
1300
Jens Axboed8a85592009-09-02 12:34:32 +02001301/**
Chris Mason3259f8b2010-10-29 11:16:17 -04001302 * writeback_inodes_sb_nr - writeback dirty inodes from given super_block
Jens Axboed8a85592009-09-02 12:34:32 +02001303 * @sb: the superblock
Chris Mason3259f8b2010-10-29 11:16:17 -04001304 * @nr: the number of pages to write
Marcos Paulo de Souza786228a2011-11-23 20:56:45 +08001305 * @reason: reason why some writeback work initiated
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306 *
Jens Axboed8a85592009-09-02 12:34:32 +02001307 * Start writeback on some inodes on this super_block. No guarantees are made
1308 * on how many (if any) will be written, and this function does not wait
Chris Mason3259f8b2010-10-29 11:16:17 -04001309 * for IO completion of submitted IO.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001311void writeback_inodes_sb_nr(struct super_block *sb,
1312 unsigned long nr,
1313 enum wb_reason reason)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001315 DECLARE_COMPLETION_ONSTACK(done);
1316 struct wb_writeback_work work = {
Wu Fengguang6e6938b2010-06-06 10:38:15 -06001317 .sb = sb,
1318 .sync_mode = WB_SYNC_NONE,
1319 .tagged_writepages = 1,
1320 .done = &done,
1321 .nr_pages = nr,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001322 .reason = reason,
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001323 };
Jens Axboe0e3c9a22010-06-01 11:08:43 +02001324
Jan Kara6eedc702012-07-03 16:45:27 +02001325 if (sb->s_bdi == &noop_backing_dev_info)
1326 return;
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001327 WARN_ON(!rwsem_is_locked(&sb->s_umount));
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001328 bdi_queue_work(sb->s_bdi, &work);
1329 wait_for_completion(&done);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330}
Chris Mason3259f8b2010-10-29 11:16:17 -04001331EXPORT_SYMBOL(writeback_inodes_sb_nr);
1332
1333/**
1334 * writeback_inodes_sb - writeback dirty inodes from given super_block
1335 * @sb: the superblock
Marcos Paulo de Souza786228a2011-11-23 20:56:45 +08001336 * @reason: reason why some writeback work was initiated
Chris Mason3259f8b2010-10-29 11:16:17 -04001337 *
1338 * Start writeback on some inodes on this super_block. No guarantees are made
1339 * on how many (if any) will be written, and this function does not wait
1340 * for IO completion of submitted IO.
1341 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001342void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason)
Chris Mason3259f8b2010-10-29 11:16:17 -04001343{
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001344 return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason);
Chris Mason3259f8b2010-10-29 11:16:17 -04001345}
Jens Axboed8a85592009-09-02 12:34:32 +02001346EXPORT_SYMBOL(writeback_inodes_sb);
1347
1348/**
Miao Xie10ee27a2013-01-10 13:47:57 +08001349 * try_to_writeback_inodes_sb_nr - try to start writeback if none underway
Chris Mason3259f8b2010-10-29 11:16:17 -04001350 * @sb: the superblock
1351 * @nr: the number of pages to write
Miao Xie10ee27a2013-01-10 13:47:57 +08001352 * @reason: the reason of writeback
Chris Mason3259f8b2010-10-29 11:16:17 -04001353 *
Miao Xie10ee27a2013-01-10 13:47:57 +08001354 * Invoke writeback_inodes_sb_nr if no writeback is currently underway.
Chris Mason3259f8b2010-10-29 11:16:17 -04001355 * Returns 1 if writeback was started, 0 if not.
1356 */
Miao Xie10ee27a2013-01-10 13:47:57 +08001357int try_to_writeback_inodes_sb_nr(struct super_block *sb,
1358 unsigned long nr,
1359 enum wb_reason reason)
Chris Mason3259f8b2010-10-29 11:16:17 -04001360{
Miao Xie10ee27a2013-01-10 13:47:57 +08001361 if (writeback_in_progress(sb->s_bdi))
Chris Mason3259f8b2010-10-29 11:16:17 -04001362 return 1;
Miao Xie10ee27a2013-01-10 13:47:57 +08001363
1364 if (!down_read_trylock(&sb->s_umount))
Chris Mason3259f8b2010-10-29 11:16:17 -04001365 return 0;
Miao Xie10ee27a2013-01-10 13:47:57 +08001366
1367 writeback_inodes_sb_nr(sb, nr, reason);
1368 up_read(&sb->s_umount);
1369 return 1;
Chris Mason3259f8b2010-10-29 11:16:17 -04001370}
Miao Xie10ee27a2013-01-10 13:47:57 +08001371EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr);
1372
1373/**
1374 * try_to_writeback_inodes_sb - try to start writeback if none underway
1375 * @sb: the superblock
1376 * @reason: reason why some writeback work was initiated
1377 *
1378 * Implement by try_to_writeback_inodes_sb_nr()
1379 * Returns 1 if writeback was started, 0 if not.
1380 */
1381int try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason)
1382{
1383 return try_to_writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason);
1384}
1385EXPORT_SYMBOL(try_to_writeback_inodes_sb);
Chris Mason3259f8b2010-10-29 11:16:17 -04001386
1387/**
Jens Axboed8a85592009-09-02 12:34:32 +02001388 * sync_inodes_sb - sync sb inode pages
Jan Kara0dc83bd2014-02-21 11:19:04 +01001389 * @sb: the superblock
Jens Axboed8a85592009-09-02 12:34:32 +02001390 *
1391 * This function writes and waits on any dirty inode belonging to this
Jan Kara0dc83bd2014-02-21 11:19:04 +01001392 * super_block.
Jens Axboed8a85592009-09-02 12:34:32 +02001393 */
Jan Kara0dc83bd2014-02-21 11:19:04 +01001394void sync_inodes_sb(struct super_block *sb)
Jens Axboed8a85592009-09-02 12:34:32 +02001395{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001396 DECLARE_COMPLETION_ONSTACK(done);
1397 struct wb_writeback_work work = {
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001398 .sb = sb,
1399 .sync_mode = WB_SYNC_ALL,
1400 .nr_pages = LONG_MAX,
1401 .range_cyclic = 0,
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001402 .done = &done,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001403 .reason = WB_REASON_SYNC,
Dave Chinner7747bd42013-07-02 22:38:35 +10001404 .for_sync = 1,
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001405 };
1406
Jan Kara6eedc702012-07-03 16:45:27 +02001407 /* Nothing to do? */
1408 if (sb->s_bdi == &noop_backing_dev_info)
1409 return;
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001410 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1411
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001412 bdi_queue_work(sb->s_bdi, &work);
1413 wait_for_completion(&done);
1414
Jens Axboeb6e51312009-09-16 15:13:54 +02001415 wait_sb_inodes(sb);
Jens Axboed8a85592009-09-02 12:34:32 +02001416}
1417EXPORT_SYMBOL(sync_inodes_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419/**
Andrea Arcangeli7f04c262005-10-30 15:03:05 -08001420 * write_inode_now - write an inode to disk
1421 * @inode: inode to write to disk
1422 * @sync: whether the write should be synchronous or not
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423 *
Andrea Arcangeli7f04c262005-10-30 15:03:05 -08001424 * This function commits an inode to disk immediately if it is dirty. This is
1425 * primarily needed by knfsd.
1426 *
1427 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429int write_inode_now(struct inode *inode, int sync)
1430{
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001431 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432 struct writeback_control wbc = {
1433 .nr_to_write = LONG_MAX,
Mike Galbraith18914b12008-02-08 04:20:23 -08001434 .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -07001435 .range_start = 0,
1436 .range_end = LLONG_MAX,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437 };
1438
1439 if (!mapping_cap_writeback_dirty(inode->i_mapping))
Andrew Morton49364ce2005-11-07 00:59:15 -08001440 wbc.nr_to_write = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441
1442 might_sleep();
Jan Kara4f8ad652012-05-03 14:48:00 +02001443 return writeback_single_inode(inode, wb, &wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444}
1445EXPORT_SYMBOL(write_inode_now);
1446
1447/**
1448 * sync_inode - write an inode and its pages to disk.
1449 * @inode: the inode to sync
1450 * @wbc: controls the writeback mode
1451 *
1452 * sync_inode() will write an inode and its pages to disk. It will also
1453 * correctly update the inode on its superblock's dirty inode lists and will
1454 * update inode->i_state.
1455 *
1456 * The caller must have a ref on the inode.
1457 */
1458int sync_inode(struct inode *inode, struct writeback_control *wbc)
1459{
Jan Kara4f8ad652012-05-03 14:48:00 +02001460 return writeback_single_inode(inode, &inode_to_bdi(inode)->wb, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461}
1462EXPORT_SYMBOL(sync_inode);
Christoph Hellwigc37650162010-10-06 10:48:20 +02001463
1464/**
Andrew Mortonc691b9d2011-01-13 15:45:48 -08001465 * sync_inode_metadata - write an inode to disk
Christoph Hellwigc37650162010-10-06 10:48:20 +02001466 * @inode: the inode to sync
1467 * @wait: wait for I/O to complete.
1468 *
Andrew Mortonc691b9d2011-01-13 15:45:48 -08001469 * Write an inode to disk and adjust its dirty state after completion.
Christoph Hellwigc37650162010-10-06 10:48:20 +02001470 *
1471 * Note: only writes the actual inode, no associated data or other metadata.
1472 */
1473int sync_inode_metadata(struct inode *inode, int wait)
1474{
1475 struct writeback_control wbc = {
1476 .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE,
1477 .nr_to_write = 0, /* metadata-only */
1478 };
1479
1480 return sync_inode(inode, &wbc);
1481}
1482EXPORT_SYMBOL(sync_inode_metadata);