blob: 21f46fb3a10193a966c18879c5bea794fd9a494e [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>
25#include <linux/freezer.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include <linux/writeback.h>
27#include <linux/blkdev.h>
28#include <linux/backing-dev.h>
Dave Chinner455b2862010-07-07 13:24:06 +100029#include <linux/tracepoint.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;
Wu Fengguangd46db3d2011-05-04 19:54:37 -060043 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;
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -060049 enum wb_reason reason; /* why was writeback initiated? */
Jens Axboec4a77a62009-09-16 15:18:25 +020050
Jens Axboe8010c3b2009-09-15 20:04:57 +020051 struct list_head list; /* pending work list */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +020052 struct completion *done; /* set if the caller waits */
Jens Axboe03ba3782009-09-09 09:08:54 +020053};
54
Adrian Bunkf11b00f2008-04-29 00:58:56 -070055/**
56 * writeback_in_progress - determine whether there is writeback in progress
57 * @bdi: the device's backing_dev_info structure.
58 *
Jens Axboe03ba3782009-09-09 09:08:54 +020059 * Determine whether there is writeback waiting to be handled against a
60 * backing device.
Adrian Bunkf11b00f2008-04-29 00:58:56 -070061 */
62int writeback_in_progress(struct backing_dev_info *bdi)
63{
Jan Kara81d73a32010-08-11 14:17:44 -070064 return test_bit(BDI_writeback_running, &bdi->state);
Adrian Bunkf11b00f2008-04-29 00:58:56 -070065}
Theodore Ts'o00d4e732012-09-19 22:42:36 -040066EXPORT_SYMBOL(writeback_in_progress);
Adrian Bunkf11b00f2008-04-29 00:58:56 -070067
Jan Kara692ebd12010-09-21 11:51:01 +020068static inline struct backing_dev_info *inode_to_bdi(struct inode *inode)
69{
70 struct super_block *sb = inode->i_sb;
Jan Kara692ebd12010-09-21 11:51:01 +020071
Christoph Hellwigaaead252010-10-04 14:25:33 +020072 if (strcmp(sb->s_type->name, "bdev") == 0)
73 return inode->i_mapping->backing_dev_info;
74
75 return sb->s_bdi;
Jan Kara692ebd12010-09-21 11:51:01 +020076}
77
Nick Piggin7ccf19a2010-10-21 11:49:30 +110078static inline struct inode *wb_inode(struct list_head *head)
79{
80 return list_entry(head, struct inode, i_wb_list);
81}
82
Wu Fengguang15eb77a2012-01-17 11:18:56 -060083/*
84 * Include the creation of the trace points after defining the
85 * wb_writeback_work structure and inline functions so that the definition
86 * remains local to this file.
87 */
88#define CREATE_TRACE_POINTS
89#include <trace/events/writeback.h>
90
Jan Kara65850272011-01-13 15:45:44 -080091/* Wakeup flusher thread or forker thread to fork it. Requires bdi->wb_lock. */
92static void bdi_wakeup_flusher(struct backing_dev_info *bdi)
Nick Piggin4195f732009-05-28 09:01:15 +020093{
Artem Bityutskiyfff5b852010-07-25 14:29:20 +030094 if (bdi->wb.task) {
95 wake_up_process(bdi->wb.task);
96 } else {
97 /*
98 * The bdi thread isn't there, wake up the forker thread which
99 * will create and run it.
100 */
Jens Axboe03ba3782009-09-09 09:08:54 +0200101 wake_up_process(default_backing_dev_info.wb.task);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102 }
Jan Kara65850272011-01-13 15:45:44 -0800103}
104
105static void bdi_queue_work(struct backing_dev_info *bdi,
106 struct wb_writeback_work *work)
107{
108 trace_writeback_queue(bdi, work);
109
110 spin_lock_bh(&bdi->wb_lock);
111 list_add_tail(&work->list, &bdi->work_list);
112 if (!bdi->wb.task)
113 trace_writeback_nothread(bdi, work);
114 bdi_wakeup_flusher(bdi);
Artem Bityutskiy64677162010-07-25 14:29:22 +0300115 spin_unlock_bh(&bdi->wb_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116}
117
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200118static void
119__bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600120 bool range_cyclic, enum wb_reason reason)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200122 struct wb_writeback_work *work;
Jens Axboe03ba3782009-09-09 09:08:54 +0200123
Jens Axboebcddc3f2009-09-13 20:07:36 +0200124 /*
125 * This is WB_SYNC_NONE writeback, so if allocation fails just
126 * wakeup the thread for old dirty data writeback
127 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200128 work = kzalloc(sizeof(*work), GFP_ATOMIC);
129 if (!work) {
Dave Chinner455b2862010-07-07 13:24:06 +1000130 if (bdi->wb.task) {
131 trace_writeback_nowork(bdi);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200132 wake_up_process(bdi->wb.task);
Dave Chinner455b2862010-07-07 13:24:06 +1000133 }
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200134 return;
Jens Axboebcddc3f2009-09-13 20:07:36 +0200135 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200136
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200137 work->sync_mode = WB_SYNC_NONE;
138 work->nr_pages = nr_pages;
139 work->range_cyclic = range_cyclic;
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600140 work->reason = reason;
Christoph Hellwigf0fad8a2009-09-11 09:47:56 +0200141
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200142 bdi_queue_work(bdi, work);
Jens Axboeb6e51312009-09-16 15:13:54 +0200143}
144
145/**
146 * bdi_start_writeback - start writeback
147 * @bdi: the backing device to write from
148 * @nr_pages: the number of pages to write
Marcos Paulo de Souza786228a2011-11-23 20:56:45 +0800149 * @reason: reason why some writeback work was initiated
Jens Axboeb6e51312009-09-16 15:13:54 +0200150 *
151 * Description:
152 * This does WB_SYNC_NONE opportunistic writeback. The IO is only
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300153 * started when this function returns, we make no guarantees on
Jens Axboe0e3c9a22010-06-01 11:08:43 +0200154 * completion. Caller need not hold sb s_umount semaphore.
Jens Axboeb6e51312009-09-16 15:13:54 +0200155 *
156 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600157void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages,
158 enum wb_reason reason)
Jens Axboeb6e51312009-09-16 15:13:54 +0200159{
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600160 __bdi_start_writeback(bdi, nr_pages, true, reason);
Christoph Hellwigc5444192010-06-08 18:15:15 +0200161}
Wu Fengguangd3ddec72009-09-23 20:33:40 +0800162
Christoph Hellwigc5444192010-06-08 18:15:15 +0200163/**
164 * bdi_start_background_writeback - start background writeback
165 * @bdi: the backing device to write from
166 *
167 * Description:
Jan Kara65850272011-01-13 15:45:44 -0800168 * This makes sure WB_SYNC_NONE background writeback happens. When
169 * this function returns, it is only guaranteed that for given BDI
170 * some IO is happening if we are over background dirty threshold.
171 * Caller need not hold sb s_umount semaphore.
Christoph Hellwigc5444192010-06-08 18:15:15 +0200172 */
173void bdi_start_background_writeback(struct backing_dev_info *bdi)
174{
Jan Kara65850272011-01-13 15:45:44 -0800175 /*
176 * We just wake up the flusher thread. It will perform background
177 * writeback as soon as there is no other work to do.
178 */
Wu Fengguang71927e82011-01-13 15:45:46 -0800179 trace_writeback_wake_background(bdi);
Jan Kara65850272011-01-13 15:45:44 -0800180 spin_lock_bh(&bdi->wb_lock);
181 bdi_wakeup_flusher(bdi);
182 spin_unlock_bh(&bdi->wb_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183}
184
185/*
Dave Chinnera66979a2011-03-22 22:23:41 +1100186 * Remove the inode from the writeback list it is on.
187 */
188void inode_wb_list_del(struct inode *inode)
189{
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600190 struct backing_dev_info *bdi = inode_to_bdi(inode);
Dave Chinnera66979a2011-03-22 22:23:41 +1100191
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600192 spin_lock(&bdi->wb.list_lock);
193 list_del_init(&inode->i_wb_list);
194 spin_unlock(&bdi->wb.list_lock);
195}
Dave Chinnera66979a2011-03-22 22:23:41 +1100196
197/*
Andrew Morton6610a0b2007-10-16 23:30:32 -0700198 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
199 * furthest end of its superblock's dirty-inode list.
200 *
201 * Before stamping the inode's ->dirtied_when, we check to see whether it is
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200202 * already the most-recently-dirtied inode on the b_dirty list. If that is
Andrew Morton6610a0b2007-10-16 23:30:32 -0700203 * the case then the inode must have been redirtied while it was being written
204 * out and we don't reset its dirtied_when.
205 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600206static void redirty_tail(struct inode *inode, struct bdi_writeback *wb)
Andrew Morton6610a0b2007-10-16 23:30:32 -0700207{
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600208 assert_spin_locked(&wb->list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200209 if (!list_empty(&wb->b_dirty)) {
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200210 struct inode *tail;
Andrew Morton6610a0b2007-10-16 23:30:32 -0700211
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100212 tail = wb_inode(wb->b_dirty.next);
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200213 if (time_before(inode->dirtied_when, tail->dirtied_when))
Andrew Morton6610a0b2007-10-16 23:30:32 -0700214 inode->dirtied_when = jiffies;
215 }
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100216 list_move(&inode->i_wb_list, &wb->b_dirty);
Andrew Morton6610a0b2007-10-16 23:30:32 -0700217}
218
219/*
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200220 * requeue inode for re-scanning after bdi->b_io list is exhausted.
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700221 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600222static void requeue_io(struct inode *inode, struct bdi_writeback *wb)
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700223{
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600224 assert_spin_locked(&wb->list_lock);
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100225 list_move(&inode->i_wb_list, &wb->b_more_io);
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700226}
227
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700228static void inode_sync_complete(struct inode *inode)
229{
Jan Kara365b94ae2012-05-03 14:47:55 +0200230 inode->i_state &= ~I_SYNC;
Jan Kara4eff96d2012-11-26 16:29:51 -0800231 /* If inode is clean an unused, put it into LRU now... */
232 inode_add_lru(inode);
Jan Kara365b94ae2012-05-03 14:47:55 +0200233 /* Waiters must see I_SYNC cleared before being woken up */
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700234 smp_mb();
235 wake_up_bit(&inode->i_state, __I_SYNC);
236}
237
Jeff Laytond2caa3c52009-04-02 16:56:37 -0700238static bool inode_dirtied_after(struct inode *inode, unsigned long t)
239{
240 bool ret = time_after(inode->dirtied_when, t);
241#ifndef CONFIG_64BIT
242 /*
243 * For inodes being constantly redirtied, dirtied_when can get stuck.
244 * It _appears_ to be in the future, but is actually in distant past.
245 * This test is necessary to prevent such wrapped-around relative times
Jens Axboe5b0830c2009-09-23 19:37:09 +0200246 * from permanently stopping the whole bdi writeback.
Jeff Laytond2caa3c52009-04-02 16:56:37 -0700247 */
248 ret = ret && time_before_eq(inode->dirtied_when, jiffies);
249#endif
250 return ret;
251}
252
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700253/*
Wang Sheng-Hui0e2f2b22012-09-11 08:28:18 +0800254 * Move expired (dirtied before work->older_than_this) dirty inodes from
Jan Kara697e6fe2012-03-09 07:26:22 -0800255 * @delaying_queue to @dispatch_queue.
Fengguang Wu2c136572007-10-16 23:30:39 -0700256 */
Wu Fengguange84d0a42011-04-23 12:27:27 -0600257static int move_expired_inodes(struct list_head *delaying_queue,
Fengguang Wu2c136572007-10-16 23:30:39 -0700258 struct list_head *dispatch_queue,
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600259 struct wb_writeback_work *work)
Fengguang Wu2c136572007-10-16 23:30:39 -0700260{
Shaohua Li5c034492009-09-24 14:42:33 +0200261 LIST_HEAD(tmp);
262 struct list_head *pos, *node;
Jens Axboecf137302009-09-24 15:12:57 +0200263 struct super_block *sb = NULL;
Shaohua Li5c034492009-09-24 14:42:33 +0200264 struct inode *inode;
Jens Axboecf137302009-09-24 15:12:57 +0200265 int do_sb_sort = 0;
Wu Fengguange84d0a42011-04-23 12:27:27 -0600266 int moved = 0;
Shaohua Li5c034492009-09-24 14:42:33 +0200267
Fengguang Wu2c136572007-10-16 23:30:39 -0700268 while (!list_empty(delaying_queue)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100269 inode = wb_inode(delaying_queue->prev);
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600270 if (work->older_than_this &&
271 inode_dirtied_after(inode, *work->older_than_this))
Fengguang Wu2c136572007-10-16 23:30:39 -0700272 break;
Jens Axboecf137302009-09-24 15:12:57 +0200273 if (sb && sb != inode->i_sb)
274 do_sb_sort = 1;
275 sb = inode->i_sb;
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100276 list_move(&inode->i_wb_list, &tmp);
Wu Fengguange84d0a42011-04-23 12:27:27 -0600277 moved++;
Shaohua Li5c034492009-09-24 14:42:33 +0200278 }
279
Jens Axboecf137302009-09-24 15:12:57 +0200280 /* just one sb in list, splice to dispatch_queue and we're done */
281 if (!do_sb_sort) {
282 list_splice(&tmp, dispatch_queue);
Wu Fengguange84d0a42011-04-23 12:27:27 -0600283 goto out;
Jens Axboecf137302009-09-24 15:12:57 +0200284 }
285
Shaohua Li5c034492009-09-24 14:42:33 +0200286 /* Move inodes from one superblock together */
287 while (!list_empty(&tmp)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100288 sb = wb_inode(tmp.prev)->i_sb;
Shaohua Li5c034492009-09-24 14:42:33 +0200289 list_for_each_prev_safe(pos, node, &tmp) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100290 inode = wb_inode(pos);
Shaohua Li5c034492009-09-24 14:42:33 +0200291 if (inode->i_sb == sb)
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100292 list_move(&inode->i_wb_list, dispatch_queue);
Shaohua Li5c034492009-09-24 14:42:33 +0200293 }
Fengguang Wu2c136572007-10-16 23:30:39 -0700294 }
Wu Fengguange84d0a42011-04-23 12:27:27 -0600295out:
296 return moved;
Fengguang Wu2c136572007-10-16 23:30:39 -0700297}
298
299/*
300 * Queue all expired dirty inodes for io, eldest first.
Wu Fengguang4ea879b2010-08-11 14:17:42 -0700301 * Before
302 * newly dirtied b_dirty b_io b_more_io
303 * =============> gf edc BA
304 * After
305 * newly dirtied b_dirty b_io b_more_io
306 * =============> g fBAedc
307 * |
308 * +--> dequeue for IO
Fengguang Wu2c136572007-10-16 23:30:39 -0700309 */
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600310static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work)
Fengguang Wu2c136572007-10-16 23:30:39 -0700311{
Wu Fengguange84d0a42011-04-23 12:27:27 -0600312 int moved;
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600313 assert_spin_locked(&wb->list_lock);
Wu Fengguang4ea879b2010-08-11 14:17:42 -0700314 list_splice_init(&wb->b_more_io, &wb->b_io);
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600315 moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, work);
316 trace_writeback_queue_io(wb, work, moved);
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200317}
318
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100319static int write_inode(struct inode *inode, struct writeback_control *wbc)
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200320{
Tejun Heo9fb0a7d2013-01-11 13:06:37 -0800321 int ret;
322
323 if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) {
324 trace_writeback_write_inode_start(inode, wbc);
325 ret = inode->i_sb->s_op->write_inode(inode, wbc);
326 trace_writeback_write_inode(inode, wbc);
327 return ret;
328 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200329 return 0;
Fengguang Wu2c136572007-10-16 23:30:39 -0700330}
331
332/*
Jan Kara169ebd92012-05-03 14:48:03 +0200333 * Wait for writeback on an inode to complete. Called with i_lock held.
334 * Caller must make sure inode cannot go away when we drop i_lock.
Christoph Hellwig01c03192009-06-08 13:35:40 +0200335 */
Jan Kara169ebd92012-05-03 14:48:03 +0200336static void __inode_wait_for_writeback(struct inode *inode)
337 __releases(inode->i_lock)
338 __acquires(inode->i_lock)
Christoph Hellwig01c03192009-06-08 13:35:40 +0200339{
340 DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC);
341 wait_queue_head_t *wqh;
342
343 wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
Dave Chinner250df6e2011-03-22 22:23:36 +1100344 while (inode->i_state & I_SYNC) {
345 spin_unlock(&inode->i_lock);
Christoph Hellwig01c03192009-06-08 13:35:40 +0200346 __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE);
Dave Chinner250df6e2011-03-22 22:23:36 +1100347 spin_lock(&inode->i_lock);
Richard Kennedy58a9d3d82010-05-24 14:32:38 -0700348 }
Christoph Hellwig01c03192009-06-08 13:35:40 +0200349}
350
351/*
Jan Kara169ebd92012-05-03 14:48:03 +0200352 * Wait for writeback on an inode to complete. Caller must have inode pinned.
353 */
354void inode_wait_for_writeback(struct inode *inode)
355{
356 spin_lock(&inode->i_lock);
357 __inode_wait_for_writeback(inode);
358 spin_unlock(&inode->i_lock);
359}
360
361/*
362 * Sleep until I_SYNC is cleared. This function must be called with i_lock
363 * held and drops it. It is aimed for callers not holding any inode reference
364 * so once i_lock is dropped, inode can go away.
365 */
366static void inode_sleep_on_writeback(struct inode *inode)
367 __releases(inode->i_lock)
368{
369 DEFINE_WAIT(wait);
370 wait_queue_head_t *wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
371 int sleep;
372
373 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
374 sleep = inode->i_state & I_SYNC;
375 spin_unlock(&inode->i_lock);
376 if (sleep)
377 schedule();
378 finish_wait(wqh, &wait);
379}
380
381/*
Jan Karaccb26b52012-05-03 14:47:58 +0200382 * Find proper writeback list for the inode depending on its current state and
383 * possibly also change of its state while we were doing writeback. Here we
384 * handle things such as livelock prevention or fairness of writeback among
385 * inodes. This function can be called only by flusher thread - noone else
386 * processes all inodes in writeback lists and requeueing inodes behind flusher
387 * thread's back can have unexpected consequences.
388 */
389static void requeue_inode(struct inode *inode, struct bdi_writeback *wb,
390 struct writeback_control *wbc)
391{
392 if (inode->i_state & I_FREEING)
393 return;
394
395 /*
396 * Sync livelock prevention. Each inode is tagged and synced in one
397 * shot. If still dirty, it will be redirty_tail()'ed below. Update
398 * the dirty time to prevent enqueue and sync it again.
399 */
400 if ((inode->i_state & I_DIRTY) &&
401 (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages))
402 inode->dirtied_when = jiffies;
403
Jan Kara4f8ad652012-05-03 14:48:00 +0200404 if (wbc->pages_skipped) {
405 /*
406 * writeback is not making progress due to locked
407 * buffers. Skip this inode for now.
408 */
409 redirty_tail(inode, wb);
410 return;
411 }
412
Jan Karaccb26b52012-05-03 14:47:58 +0200413 if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) {
414 /*
415 * We didn't write back all the pages. nfs_writepages()
416 * sometimes bales out without doing anything.
417 */
418 if (wbc->nr_to_write <= 0) {
419 /* Slice used up. Queue for next turn. */
420 requeue_io(inode, wb);
421 } else {
422 /*
423 * Writeback blocked by something other than
424 * congestion. Delay the inode for some time to
425 * avoid spinning on the CPU (100% iowait)
426 * retrying writeback of the dirty page/inode
427 * that cannot be performed immediately.
428 */
429 redirty_tail(inode, wb);
430 }
431 } else if (inode->i_state & I_DIRTY) {
432 /*
433 * Filesystems can dirty the inode during writeback operations,
434 * such as delayed allocation during submission or metadata
435 * updates after data IO completion.
436 */
437 redirty_tail(inode, wb);
438 } else {
439 /* The inode is clean. Remove from writeback lists. */
440 list_del_init(&inode->i_wb_list);
441 }
442}
443
444/*
Jan Kara4f8ad652012-05-03 14:48:00 +0200445 * Write out an inode and its dirty pages. Do not update the writeback list
446 * linkage. That is left to the caller. The caller is also responsible for
447 * setting I_SYNC flag and calling inode_sync_complete() to clear it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448 */
449static int
Yan Hongcd8ed2a2012-10-08 16:33:45 -0700450__writeback_single_inode(struct inode *inode, struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452 struct address_space *mapping = inode->i_mapping;
Wu Fengguang251d6a42010-12-01 17:33:37 -0600453 long nr_to_write = wbc->nr_to_write;
Christoph Hellwig01c03192009-06-08 13:35:40 +0200454 unsigned dirty;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455 int ret;
456
Jan Kara4f8ad652012-05-03 14:48:00 +0200457 WARN_ON(!(inode->i_state & I_SYNC));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458
Tejun Heo9fb0a7d2013-01-11 13:06:37 -0800459 trace_writeback_single_inode_start(inode, wbc, nr_to_write);
460
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 ret = do_writepages(mapping, wbc);
462
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100463 /*
464 * Make sure to wait on the data before writing out the metadata.
465 * This is important for filesystems that modify metadata on data
466 * I/O completion.
467 */
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100468 if (wbc->sync_mode == WB_SYNC_ALL) {
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100469 int err = filemap_fdatawait(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470 if (ret == 0)
471 ret = err;
472 }
473
Dmitry Monakhov5547e8a2010-05-07 13:35:44 +0400474 /*
475 * Some filesystems may redirty the inode during the writeback
476 * due to delalloc, clear dirty metadata flags right before
477 * write_inode()
478 */
Dave Chinner250df6e2011-03-22 22:23:36 +1100479 spin_lock(&inode->i_lock);
Jan Kara6290be12012-05-03 14:47:57 +0200480 /* Clear I_DIRTY_PAGES if we've written out all dirty pages */
481 if (!mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
482 inode->i_state &= ~I_DIRTY_PAGES;
Dmitry Monakhov5547e8a2010-05-07 13:35:44 +0400483 dirty = inode->i_state & I_DIRTY;
484 inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC);
Dave Chinner250df6e2011-03-22 22:23:36 +1100485 spin_unlock(&inode->i_lock);
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100486 /* Don't write the inode if only I_DIRTY_PAGES was set */
487 if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100488 int err = write_inode(inode, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489 if (ret == 0)
490 ret = err;
491 }
Jan Kara4f8ad652012-05-03 14:48:00 +0200492 trace_writeback_single_inode(inode, wbc, nr_to_write);
493 return ret;
494}
495
496/*
497 * Write out an inode's dirty pages. Either the caller has an active reference
498 * on the inode or the inode has I_WILL_FREE set.
499 *
500 * This function is designed to be called for writing back one inode which
501 * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode()
502 * and does more profound writeback list handling in writeback_sb_inodes().
503 */
504static int
505writeback_single_inode(struct inode *inode, struct bdi_writeback *wb,
506 struct writeback_control *wbc)
507{
508 int ret = 0;
509
510 spin_lock(&inode->i_lock);
511 if (!atomic_read(&inode->i_count))
512 WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING)));
513 else
514 WARN_ON(inode->i_state & I_WILL_FREE);
515
516 if (inode->i_state & I_SYNC) {
517 if (wbc->sync_mode != WB_SYNC_ALL)
518 goto out;
519 /*
Jan Kara169ebd92012-05-03 14:48:03 +0200520 * It's a data-integrity sync. We must wait. Since callers hold
521 * inode reference or inode has I_WILL_FREE set, it cannot go
522 * away under us.
Jan Kara4f8ad652012-05-03 14:48:00 +0200523 */
Jan Kara169ebd92012-05-03 14:48:03 +0200524 __inode_wait_for_writeback(inode);
Jan Kara4f8ad652012-05-03 14:48:00 +0200525 }
526 WARN_ON(inode->i_state & I_SYNC);
527 /*
528 * Skip inode if it is clean. We don't want to mess with writeback
529 * lists in this function since flusher thread may be doing for example
530 * sync in parallel and if we move the inode, it could get skipped. So
531 * here we make sure inode is on some writeback list and leave it there
532 * unless we have completely cleaned the inode.
533 */
534 if (!(inode->i_state & I_DIRTY))
535 goto out;
536 inode->i_state |= I_SYNC;
537 spin_unlock(&inode->i_lock);
538
Yan Hongcd8ed2a2012-10-08 16:33:45 -0700539 ret = __writeback_single_inode(inode, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600541 spin_lock(&wb->list_lock);
Dave Chinner250df6e2011-03-22 22:23:36 +1100542 spin_lock(&inode->i_lock);
Jan Kara4f8ad652012-05-03 14:48:00 +0200543 /*
544 * If inode is clean, remove it from writeback lists. Otherwise don't
545 * touch it. See comment above for explanation.
546 */
547 if (!(inode->i_state & I_DIRTY))
548 list_del_init(&inode->i_wb_list);
549 spin_unlock(&wb->list_lock);
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700550 inode_sync_complete(inode);
Jan Kara4f8ad652012-05-03 14:48:00 +0200551out:
552 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553 return ret;
554}
555
Wu Fengguang1a12d8b2010-08-29 13:28:09 -0600556static long writeback_chunk_size(struct backing_dev_info *bdi,
557 struct wb_writeback_work *work)
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600558{
559 long pages;
560
561 /*
562 * WB_SYNC_ALL mode does livelock avoidance by syncing dirty
563 * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX
564 * here avoids calling into writeback_inodes_wb() more than once.
565 *
566 * The intended call sequence for WB_SYNC_ALL writeback is:
567 *
568 * wb_writeback()
569 * writeback_sb_inodes() <== called only once
570 * write_cache_pages() <== called once for each inode
571 * (quickly) tag currently dirty pages
572 * (maybe slowly) sync all tagged pages
573 */
574 if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages)
575 pages = LONG_MAX;
Wu Fengguang1a12d8b2010-08-29 13:28:09 -0600576 else {
577 pages = min(bdi->avg_write_bandwidth / 2,
578 global_dirty_limit / DIRTY_SCOPE);
579 pages = min(pages, work->nr_pages);
580 pages = round_down(pages + MIN_WRITEBACK_PAGES,
581 MIN_WRITEBACK_PAGES);
582 }
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600583
584 return pages;
585}
586
Jens Axboe03ba3782009-09-09 09:08:54 +0200587/*
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800588 * Write a portion of b_io inodes which belong to @sb.
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200589 *
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600590 * Return the number of pages and/or inodes written.
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800591 */
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600592static long writeback_sb_inodes(struct super_block *sb,
593 struct bdi_writeback *wb,
594 struct wb_writeback_work *work)
Jens Axboe03ba3782009-09-09 09:08:54 +0200595{
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600596 struct writeback_control wbc = {
597 .sync_mode = work->sync_mode,
598 .tagged_writepages = work->tagged_writepages,
599 .for_kupdate = work->for_kupdate,
600 .for_background = work->for_background,
601 .range_cyclic = work->range_cyclic,
602 .range_start = 0,
603 .range_end = LLONG_MAX,
604 };
605 unsigned long start_time = jiffies;
606 long write_chunk;
607 long wrote = 0; /* count both pages and inodes */
608
Jens Axboe03ba3782009-09-09 09:08:54 +0200609 while (!list_empty(&wb->b_io)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100610 struct inode *inode = wb_inode(wb->b_io.prev);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200611
612 if (inode->i_sb != sb) {
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600613 if (work->sb) {
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200614 /*
615 * We only want to write back data for this
616 * superblock, move all inodes not belonging
617 * to it back onto the dirty list.
618 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600619 redirty_tail(inode, wb);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200620 continue;
621 }
622
623 /*
624 * The inode belongs to a different superblock.
625 * Bounce back to the caller to unpin this and
626 * pin the next superblock.
627 */
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600628 break;
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200629 }
630
Christoph Hellwig9843b762010-10-24 19:40:46 +0200631 /*
Wanpeng Li331cbde2012-06-09 11:10:55 +0800632 * Don't bother with new inodes or inodes being freed, first
633 * kind does not need periodic writeout yet, and for the latter
Christoph Hellwig9843b762010-10-24 19:40:46 +0200634 * kind writeout is handled by the freer.
635 */
Dave Chinner250df6e2011-03-22 22:23:36 +1100636 spin_lock(&inode->i_lock);
Christoph Hellwig9843b762010-10-24 19:40:46 +0200637 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
Dave Chinner250df6e2011-03-22 22:23:36 +1100638 spin_unlock(&inode->i_lock);
Wu Fengguangfcc5c222011-07-11 23:08:50 -0700639 redirty_tail(inode, wb);
Nick Piggin7ef0d732009-03-12 14:31:38 -0700640 continue;
641 }
Jan Karacc1676d2012-05-03 14:47:56 +0200642 if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) {
643 /*
644 * If this inode is locked for writeback and we are not
645 * doing writeback-for-data-integrity, move it to
646 * b_more_io so that writeback can proceed with the
647 * other inodes on s_io.
648 *
649 * We'll have another go at writing back this inode
650 * when we completed a full scan of b_io.
651 */
652 spin_unlock(&inode->i_lock);
653 requeue_io(inode, wb);
654 trace_writeback_sb_inodes_requeue(inode);
655 continue;
656 }
Jan Karaf0d07b72012-05-03 14:47:59 +0200657 spin_unlock(&wb->list_lock);
658
Jan Kara4f8ad652012-05-03 14:48:00 +0200659 /*
660 * We already requeued the inode if it had I_SYNC set and we
661 * are doing WB_SYNC_NONE writeback. So this catches only the
662 * WB_SYNC_ALL case.
663 */
Jan Kara169ebd92012-05-03 14:48:03 +0200664 if (inode->i_state & I_SYNC) {
665 /* Wait for I_SYNC. This function drops i_lock... */
666 inode_sleep_on_writeback(inode);
667 /* Inode may be gone, start again */
Jan Karaead188f2012-06-08 17:07:36 +0200668 spin_lock(&wb->list_lock);
Jan Kara169ebd92012-05-03 14:48:03 +0200669 continue;
670 }
Jan Kara4f8ad652012-05-03 14:48:00 +0200671 inode->i_state |= I_SYNC;
672 spin_unlock(&inode->i_lock);
Jan Kara169ebd92012-05-03 14:48:03 +0200673
Wu Fengguang1a12d8b2010-08-29 13:28:09 -0600674 write_chunk = writeback_chunk_size(wb->bdi, work);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600675 wbc.nr_to_write = write_chunk;
676 wbc.pages_skipped = 0;
Dave Chinner250df6e2011-03-22 22:23:36 +1100677
Jan Kara169ebd92012-05-03 14:48:03 +0200678 /*
679 * We use I_SYNC to pin the inode in memory. While it is set
680 * evict_inode() will wait so the inode cannot be freed.
681 */
Yan Hongcd8ed2a2012-10-08 16:33:45 -0700682 __writeback_single_inode(inode, &wbc);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600683
684 work->nr_pages -= write_chunk - wbc.nr_to_write;
685 wrote += write_chunk - wbc.nr_to_write;
Jan Kara4f8ad652012-05-03 14:48:00 +0200686 spin_lock(&wb->list_lock);
687 spin_lock(&inode->i_lock);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600688 if (!(inode->i_state & I_DIRTY))
689 wrote++;
Jan Kara4f8ad652012-05-03 14:48:00 +0200690 requeue_inode(inode, wb, &wbc);
691 inode_sync_complete(inode);
Dave Chinner0f1b1fd2011-03-22 22:23:43 +1100692 spin_unlock(&inode->i_lock);
Jan Kara169ebd92012-05-03 14:48:03 +0200693 cond_resched_lock(&wb->list_lock);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600694 /*
695 * bail out to wb_writeback() often enough to check
696 * background threshold and other termination conditions.
697 */
698 if (wrote) {
699 if (time_is_before_jiffies(start_time + HZ / 10UL))
700 break;
701 if (work->nr_pages <= 0)
702 break;
Fengguang Wu8bc3be22008-02-04 22:29:36 -0800703 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704 }
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600705 return wrote;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800706}
Nick Piggin38f21972009-01-06 14:40:25 -0800707
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600708static long __writeback_inodes_wb(struct bdi_writeback *wb,
709 struct wb_writeback_work *work)
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800710{
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600711 unsigned long start_time = jiffies;
712 long wrote = 0;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800713
714 while (!list_empty(&wb->b_io)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100715 struct inode *inode = wb_inode(wb->b_io.prev);
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800716 struct super_block *sb = inode->i_sb;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800717
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000718 if (!grab_super_passive(sb)) {
Wu Fengguang0e9958162011-07-29 22:14:35 -0600719 /*
720 * grab_super_passive() may fail consistently due to
721 * s_umount being grabbed by someone else. Don't use
722 * requeue_io() to avoid busy retrying the inode/sb.
723 */
724 redirty_tail(inode, wb);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200725 continue;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800726 }
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600727 wrote += writeback_sb_inodes(sb, wb, work);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200728 drop_super(sb);
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800729
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600730 /* refer to the same tests at the end of writeback_sb_inodes */
731 if (wrote) {
732 if (time_is_before_jiffies(start_time + HZ / 10UL))
733 break;
734 if (work->nr_pages <= 0)
735 break;
736 }
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800737 }
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200738 /* Leave any unwritten inodes on b_io */
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600739 return wrote;
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200740}
741
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600742long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages,
743 enum wb_reason reason)
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200744{
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600745 struct wb_writeback_work work = {
746 .nr_pages = nr_pages,
747 .sync_mode = WB_SYNC_NONE,
748 .range_cyclic = 1,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600749 .reason = reason,
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600750 };
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200751
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600752 spin_lock(&wb->list_lock);
Wu Fengguang424b3512010-07-21 20:11:53 -0600753 if (list_empty(&wb->b_io))
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600754 queue_io(wb, &work);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600755 __writeback_inodes_wb(wb, &work);
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600756 spin_unlock(&wb->list_lock);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200757
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600758 return nr_pages - work.nr_pages;
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200759}
760
Wu Fengguangb00949a2010-11-18 14:38:33 -0600761static bool over_bground_thresh(struct backing_dev_info *bdi)
Jens Axboe03ba3782009-09-09 09:08:54 +0200762{
763 unsigned long background_thresh, dirty_thresh;
764
Wu Fengguang16c40422010-08-11 14:17:39 -0700765 global_dirty_limits(&background_thresh, &dirty_thresh);
Jens Axboe03ba3782009-09-09 09:08:54 +0200766
Wu Fengguangb00949a2010-11-18 14:38:33 -0600767 if (global_page_state(NR_FILE_DIRTY) +
768 global_page_state(NR_UNSTABLE_NFS) > background_thresh)
769 return true;
770
771 if (bdi_stat(bdi, BDI_RECLAIMABLE) >
772 bdi_dirty_limit(bdi, background_thresh))
773 return true;
774
775 return false;
Jens Axboe03ba3782009-09-09 09:08:54 +0200776}
777
778/*
Wu Fengguange98be2d2010-08-29 11:22:30 -0600779 * Called under wb->list_lock. If there are multiple wb per bdi,
780 * only the flusher working on the first wb should do it.
781 */
782static void wb_update_bandwidth(struct bdi_writeback *wb,
783 unsigned long start_time)
784{
Wu Fengguangaf6a3112011-10-03 20:46:17 -0600785 __bdi_update_bandwidth(wb->bdi, 0, 0, 0, 0, 0, start_time);
Wu Fengguange98be2d2010-08-29 11:22:30 -0600786}
787
788/*
Jens Axboe03ba3782009-09-09 09:08:54 +0200789 * Explicit flushing or periodic writeback of "old" data.
790 *
791 * Define "old": the first time one of an inode's pages is dirtied, we mark the
792 * dirtying-time in the inode's address_space. So this periodic writeback code
793 * just walks the superblock inode list, writing back any inodes which are
794 * older than a specific point in time.
795 *
796 * Try to run once per dirty_writeback_interval. But if a writeback event
797 * takes longer than a dirty_writeback_interval interval, then leave a
798 * one-second gap.
799 *
800 * older_than_this takes precedence over nr_to_write. So we'll only write back
801 * all dirty pages if they are all attached to "old" mappings.
802 */
Jens Axboec4a77a62009-09-16 15:18:25 +0200803static long wb_writeback(struct bdi_writeback *wb,
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200804 struct wb_writeback_work *work)
Jens Axboe03ba3782009-09-09 09:08:54 +0200805{
Wu Fengguange98be2d2010-08-29 11:22:30 -0600806 unsigned long wb_start = jiffies;
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600807 long nr_pages = work->nr_pages;
Jens Axboe03ba3782009-09-09 09:08:54 +0200808 unsigned long oldest_jif;
Jan Karaa5989bd2009-09-16 19:22:48 +0200809 struct inode *inode;
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600810 long progress;
Jens Axboe03ba3782009-09-09 09:08:54 +0200811
Wu Fengguange185dda2011-04-23 11:26:07 -0600812 oldest_jif = jiffies;
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600813 work->older_than_this = &oldest_jif;
Jens Axboe03ba3782009-09-09 09:08:54 +0200814
Wu Fengguange8dfc302011-04-21 12:06:32 -0600815 spin_lock(&wb->list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200816 for (;;) {
817 /*
Wu Fengguangd3ddec72009-09-23 20:33:40 +0800818 * Stop writeback when nr_pages has been consumed
Jens Axboe03ba3782009-09-09 09:08:54 +0200819 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200820 if (work->nr_pages <= 0)
Jens Axboe03ba3782009-09-09 09:08:54 +0200821 break;
822
823 /*
Jan Karaaa373cf2011-01-13 15:45:47 -0800824 * Background writeout and kupdate-style writeback may
825 * run forever. Stop them if there is other work to do
826 * so that e.g. sync can proceed. They'll be restarted
827 * after the other works are all done.
828 */
829 if ((work->for_background || work->for_kupdate) &&
830 !list_empty(&wb->bdi->work_list))
831 break;
832
833 /*
Wu Fengguangd3ddec72009-09-23 20:33:40 +0800834 * For background writeout, stop when we are below the
835 * background dirty threshold
Jens Axboe03ba3782009-09-09 09:08:54 +0200836 */
Wu Fengguangb00949a2010-11-18 14:38:33 -0600837 if (work->for_background && !over_bground_thresh(wb->bdi))
Jens Axboe03ba3782009-09-09 09:08:54 +0200838 break;
839
Jan Kara1bc36b62011-10-19 11:44:41 +0200840 /*
841 * Kupdate and background works are special and we want to
842 * include all inodes that need writing. Livelock avoidance is
843 * handled by these works yielding to any other work so we are
844 * safe.
845 */
Wu Fengguangba9aa832010-07-21 20:32:30 -0600846 if (work->for_kupdate) {
847 oldest_jif = jiffies -
848 msecs_to_jiffies(dirty_expire_interval * 10);
Jan Kara1bc36b62011-10-19 11:44:41 +0200849 } else if (work->for_background)
850 oldest_jif = jiffies;
Dave Chinner028c2dd2010-07-07 13:24:07 +1000851
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600852 trace_writeback_start(wb->bdi, work);
Wu Fengguange8dfc302011-04-21 12:06:32 -0600853 if (list_empty(&wb->b_io))
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600854 queue_io(wb, work);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200855 if (work->sb)
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600856 progress = writeback_sb_inodes(work->sb, wb, work);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200857 else
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600858 progress = __writeback_inodes_wb(wb, work);
859 trace_writeback_written(wb->bdi, work);
Dave Chinner028c2dd2010-07-07 13:24:07 +1000860
Wu Fengguange98be2d2010-08-29 11:22:30 -0600861 wb_update_bandwidth(wb, wb_start);
Jens Axboe03ba3782009-09-09 09:08:54 +0200862
863 /*
Jens Axboe71fd05a2009-09-23 19:32:26 +0200864 * Did we write something? Try for more
Wu Fengguange6fb6da2010-07-22 10:23:44 -0600865 *
866 * Dirty inodes are moved to b_io for writeback in batches.
867 * The completion of the current batch does not necessarily
868 * mean the overall work is done. So we keep looping as long
869 * as made some progress on cleaning pages or inodes.
Jens Axboe71fd05a2009-09-23 19:32:26 +0200870 */
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600871 if (progress)
Jens Axboe71fd05a2009-09-23 19:32:26 +0200872 continue;
873 /*
Wu Fengguange6fb6da2010-07-22 10:23:44 -0600874 * No more inodes for IO, bail
Jens Axboe71fd05a2009-09-23 19:32:26 +0200875 */
Wu Fengguangb7a24412010-07-21 22:19:51 -0600876 if (list_empty(&wb->b_more_io))
Jens Axboe03ba3782009-09-09 09:08:54 +0200877 break;
878 /*
Jens Axboe8010c3b2009-09-15 20:04:57 +0200879 * Nothing written. Wait for some inode to
880 * become available for writeback. Otherwise
881 * we'll just busyloop.
Jens Axboe03ba3782009-09-09 09:08:54 +0200882 */
Jens Axboe03ba3782009-09-09 09:08:54 +0200883 if (!list_empty(&wb->b_more_io)) {
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600884 trace_writeback_wait(wb->bdi, work);
Jens Axboe03ba3782009-09-09 09:08:54 +0200885 inode = wb_inode(wb->b_more_io.prev);
Dave Chinner250df6e2011-03-22 22:23:36 +1100886 spin_lock(&inode->i_lock);
Jan Karaf0d07b72012-05-03 14:47:59 +0200887 spin_unlock(&wb->list_lock);
Jan Kara169ebd92012-05-03 14:48:03 +0200888 /* This function drops i_lock... */
889 inode_sleep_on_writeback(inode);
Jan Karaf0d07b72012-05-03 14:47:59 +0200890 spin_lock(&wb->list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200891 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200892 }
Wu Fengguange8dfc302011-04-21 12:06:32 -0600893 spin_unlock(&wb->list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200894
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600895 return nr_pages - work->nr_pages;
Jens Axboe03ba3782009-09-09 09:08:54 +0200896}
897
898/*
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200899 * Return the next wb_writeback_work struct that hasn't been processed yet.
Jens Axboe03ba3782009-09-09 09:08:54 +0200900 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200901static struct wb_writeback_work *
Minchan Kim08852b62010-08-03 12:51:16 +0200902get_next_work_item(struct backing_dev_info *bdi)
Jens Axboe03ba3782009-09-09 09:08:54 +0200903{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200904 struct wb_writeback_work *work = NULL;
Jens Axboe03ba3782009-09-09 09:08:54 +0200905
Artem Bityutskiy64677162010-07-25 14:29:22 +0300906 spin_lock_bh(&bdi->wb_lock);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200907 if (!list_empty(&bdi->work_list)) {
908 work = list_entry(bdi->work_list.next,
909 struct wb_writeback_work, list);
910 list_del_init(&work->list);
Jens Axboe03ba3782009-09-09 09:08:54 +0200911 }
Artem Bityutskiy64677162010-07-25 14:29:22 +0300912 spin_unlock_bh(&bdi->wb_lock);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200913 return work;
Jens Axboe03ba3782009-09-09 09:08:54 +0200914}
915
Linus Torvaldscdf01dd2010-10-30 08:55:52 -0700916/*
917 * Add in the number of potentially dirty inodes, because each inode
918 * write can dirty pagecache in the underlying blockdev.
919 */
920static unsigned long get_nr_dirty_pages(void)
921{
922 return global_page_state(NR_FILE_DIRTY) +
923 global_page_state(NR_UNSTABLE_NFS) +
924 get_nr_dirty_inodes();
925}
926
Jan Kara65850272011-01-13 15:45:44 -0800927static long wb_check_background_flush(struct bdi_writeback *wb)
928{
Wu Fengguangb00949a2010-11-18 14:38:33 -0600929 if (over_bground_thresh(wb->bdi)) {
Jan Kara65850272011-01-13 15:45:44 -0800930
931 struct wb_writeback_work work = {
932 .nr_pages = LONG_MAX,
933 .sync_mode = WB_SYNC_NONE,
934 .for_background = 1,
935 .range_cyclic = 1,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600936 .reason = WB_REASON_BACKGROUND,
Jan Kara65850272011-01-13 15:45:44 -0800937 };
938
939 return wb_writeback(wb, &work);
940 }
941
942 return 0;
943}
944
Jens Axboe03ba3782009-09-09 09:08:54 +0200945static long wb_check_old_data_flush(struct bdi_writeback *wb)
946{
947 unsigned long expired;
948 long nr_pages;
949
Jens Axboe69b62d02010-05-17 12:51:03 +0200950 /*
951 * When set to zero, disable periodic writeback
952 */
953 if (!dirty_writeback_interval)
954 return 0;
955
Jens Axboe03ba3782009-09-09 09:08:54 +0200956 expired = wb->last_old_flush +
957 msecs_to_jiffies(dirty_writeback_interval * 10);
958 if (time_before(jiffies, expired))
959 return 0;
960
961 wb->last_old_flush = jiffies;
Linus Torvaldscdf01dd2010-10-30 08:55:52 -0700962 nr_pages = get_nr_dirty_pages();
Jens Axboe03ba3782009-09-09 09:08:54 +0200963
Jens Axboec4a77a62009-09-16 15:18:25 +0200964 if (nr_pages) {
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200965 struct wb_writeback_work work = {
Jens Axboec4a77a62009-09-16 15:18:25 +0200966 .nr_pages = nr_pages,
967 .sync_mode = WB_SYNC_NONE,
968 .for_kupdate = 1,
969 .range_cyclic = 1,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600970 .reason = WB_REASON_PERIODIC,
Jens Axboec4a77a62009-09-16 15:18:25 +0200971 };
972
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200973 return wb_writeback(wb, &work);
Jens Axboec4a77a62009-09-16 15:18:25 +0200974 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200975
976 return 0;
977}
978
979/*
980 * Retrieve work items and do the writeback they describe
981 */
982long wb_do_writeback(struct bdi_writeback *wb, int force_wait)
983{
984 struct backing_dev_info *bdi = wb->bdi;
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200985 struct wb_writeback_work *work;
Jens Axboec4a77a62009-09-16 15:18:25 +0200986 long wrote = 0;
Jens Axboe03ba3782009-09-09 09:08:54 +0200987
Jan Kara81d73a32010-08-11 14:17:44 -0700988 set_bit(BDI_writeback_running, &wb->bdi->state);
Minchan Kim08852b62010-08-03 12:51:16 +0200989 while ((work = get_next_work_item(bdi)) != NULL) {
Jens Axboe03ba3782009-09-09 09:08:54 +0200990 /*
991 * Override sync mode, in case we must wait for completion
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200992 * because this thread is exiting now.
Jens Axboe03ba3782009-09-09 09:08:54 +0200993 */
994 if (force_wait)
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200995 work->sync_mode = WB_SYNC_ALL;
996
Dave Chinner455b2862010-07-07 13:24:06 +1000997 trace_writeback_exec(bdi, work);
998
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200999 wrote += wb_writeback(wb, work);
Jens Axboe03ba3782009-09-09 09:08:54 +02001000
1001 /*
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001002 * Notify the caller of completion if this is a synchronous
1003 * work item, otherwise just free it.
Jens Axboe03ba3782009-09-09 09:08:54 +02001004 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001005 if (work->done)
1006 complete(work->done);
1007 else
1008 kfree(work);
Jens Axboe03ba3782009-09-09 09:08:54 +02001009 }
1010
1011 /*
1012 * Check for periodic writeback, kupdated() style
1013 */
1014 wrote += wb_check_old_data_flush(wb);
Jan Kara65850272011-01-13 15:45:44 -08001015 wrote += wb_check_background_flush(wb);
Jan Kara81d73a32010-08-11 14:17:44 -07001016 clear_bit(BDI_writeback_running, &wb->bdi->state);
Jens Axboe03ba3782009-09-09 09:08:54 +02001017
1018 return wrote;
1019}
1020
1021/*
1022 * Handle writeback of dirty data for the device backed by this bdi. Also
1023 * wakes up periodically and does kupdated style flushing.
1024 */
Christoph Hellwig08243902010-06-19 23:08:22 +02001025int bdi_writeback_thread(void *data)
Jens Axboe03ba3782009-09-09 09:08:54 +02001026{
Christoph Hellwig08243902010-06-19 23:08:22 +02001027 struct bdi_writeback *wb = data;
1028 struct backing_dev_info *bdi = wb->bdi;
Jens Axboe03ba3782009-09-09 09:08:54 +02001029 long pages_written;
1030
Peter Zijlstra766f9162010-10-26 14:22:45 -07001031 current->flags |= PF_SWAPWRITE;
Christoph Hellwig08243902010-06-19 23:08:22 +02001032 set_freezable();
Artem Bityutskiyecd58402010-07-25 14:29:18 +03001033 wb->last_active = jiffies;
Christoph Hellwig08243902010-06-19 23:08:22 +02001034
1035 /*
1036 * Our parent may run at a different priority, just set us to normal
1037 */
1038 set_user_nice(current, 0);
1039
Dave Chinner455b2862010-07-07 13:24:06 +10001040 trace_writeback_thread_start(bdi);
1041
Tejun Heo8a32c442011-11-21 12:32:23 -08001042 while (!kthread_freezable_should_stop(NULL)) {
Artem Bityutskiy64677162010-07-25 14:29:22 +03001043 /*
1044 * Remove own delayed wake-up timer, since we are already awake
Yan Hong5aaea512012-12-12 13:52:14 -08001045 * and we'll take care of the periodic write-back.
Artem Bityutskiy64677162010-07-25 14:29:22 +03001046 */
1047 del_timer(&wb->wakeup_timer);
1048
Jens Axboe03ba3782009-09-09 09:08:54 +02001049 pages_written = wb_do_writeback(wb, 0);
1050
Dave Chinner455b2862010-07-07 13:24:06 +10001051 trace_writeback_pages_written(pages_written);
Jens Axboe03ba3782009-09-09 09:08:54 +02001052
Jens Axboe03ba3782009-09-09 09:08:54 +02001053 if (pages_written)
Artem Bityutskiyecd58402010-07-25 14:29:18 +03001054 wb->last_active = jiffies;
Jens Axboe03ba3782009-09-09 09:08:54 +02001055
Artem Bityutskiy297252c2010-07-25 14:29:15 +03001056 set_current_state(TASK_INTERRUPTIBLE);
J. Bruce Fieldsb76b4012010-08-28 08:52:10 +02001057 if (!list_empty(&bdi->work_list) || kthread_should_stop()) {
Artem Bityutskiy297252c2010-07-25 14:29:15 +03001058 __set_current_state(TASK_RUNNING);
1059 continue;
Jens Axboe03ba3782009-09-09 09:08:54 +02001060 }
1061
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001062 if (wb_has_dirty_io(wb) && dirty_writeback_interval)
Artem Bityutskiyfff5b852010-07-25 14:29:20 +03001063 schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10));
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001064 else {
1065 /*
1066 * We have nothing to do, so can go sleep without any
1067 * timeout and save power. When a work is queued or
1068 * something is made dirty - we will be woken up.
1069 */
Artem Bityutskiy297252c2010-07-25 14:29:15 +03001070 schedule();
Jens Axboef9eadbb2010-05-18 14:31:45 +02001071 }
Jens Axboe03ba3782009-09-09 09:08:54 +02001072 }
1073
Artem Bityutskiyfff5b852010-07-25 14:29:20 +03001074 /* Flush any work that raced with us exiting */
Christoph Hellwig08243902010-06-19 23:08:22 +02001075 if (!list_empty(&bdi->work_list))
1076 wb_do_writeback(wb, 1);
Dave Chinner455b2862010-07-07 13:24:06 +10001077
1078 trace_writeback_thread_stop(bdi);
Jens Axboe03ba3782009-09-09 09:08:54 +02001079 return 0;
1080}
1081
Christoph Hellwig08243902010-06-19 23:08:22 +02001082
Jens Axboe03ba3782009-09-09 09:08:54 +02001083/*
Jens Axboe03ba3782009-09-09 09:08:54 +02001084 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
1085 * the whole world.
1086 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001087void wakeup_flusher_threads(long nr_pages, enum wb_reason reason)
Jens Axboe03ba3782009-09-09 09:08:54 +02001088{
Christoph Hellwigb8c2f342010-06-08 18:15:07 +02001089 struct backing_dev_info *bdi;
Christoph Hellwigb8c2f342010-06-08 18:15:07 +02001090
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001091 if (!nr_pages) {
1092 nr_pages = global_page_state(NR_FILE_DIRTY) +
Jens Axboe03ba3782009-09-09 09:08:54 +02001093 global_page_state(NR_UNSTABLE_NFS);
Christoph Hellwigb8c2f342010-06-08 18:15:07 +02001094 }
1095
1096 rcu_read_lock();
1097 list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) {
1098 if (!bdi_has_dirty_io(bdi))
1099 continue;
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001100 __bdi_start_writeback(bdi, nr_pages, false, reason);
Christoph Hellwigb8c2f342010-06-08 18:15:07 +02001101 }
1102 rcu_read_unlock();
Jens Axboe03ba3782009-09-09 09:08:54 +02001103}
1104
1105static noinline void block_dump___mark_inode_dirty(struct inode *inode)
1106{
1107 if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) {
1108 struct dentry *dentry;
1109 const char *name = "?";
1110
1111 dentry = d_find_alias(inode);
1112 if (dentry) {
1113 spin_lock(&dentry->d_lock);
1114 name = (const char *) dentry->d_name.name;
1115 }
1116 printk(KERN_DEBUG
1117 "%s(%d): dirtied inode %lu (%s) on %s\n",
1118 current->comm, task_pid_nr(current), inode->i_ino,
1119 name, inode->i_sb->s_id);
1120 if (dentry) {
1121 spin_unlock(&dentry->d_lock);
1122 dput(dentry);
1123 }
1124 }
1125}
1126
1127/**
1128 * __mark_inode_dirty - internal function
1129 * @inode: inode to mark
1130 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
1131 * Mark an inode as dirty. Callers should use mark_inode_dirty or
1132 * mark_inode_dirty_sync.
1133 *
1134 * Put the inode on the super block's dirty list.
1135 *
1136 * CAREFUL! We mark it dirty unconditionally, but move it onto the
1137 * dirty list only if it is hashed or if it refers to a blockdev.
1138 * If it was not hashed, it will never be added to the dirty list
1139 * even if it is later hashed, as it will have been marked dirty already.
1140 *
1141 * In short, make sure you hash any inodes _before_ you start marking
1142 * them dirty.
1143 *
Jens Axboe03ba3782009-09-09 09:08:54 +02001144 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
1145 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
1146 * the kernel-internal blockdev inode represents the dirtying time of the
1147 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
1148 * page->mapping->host, so the page-dirtying time is recorded in the internal
1149 * blockdev inode.
1150 */
1151void __mark_inode_dirty(struct inode *inode, int flags)
1152{
1153 struct super_block *sb = inode->i_sb;
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001154 struct backing_dev_info *bdi = NULL;
Jens Axboe03ba3782009-09-09 09:08:54 +02001155
1156 /*
1157 * Don't do this for I_DIRTY_PAGES - that doesn't actually
1158 * dirty the inode itself
1159 */
1160 if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
Tejun Heo9fb0a7d2013-01-11 13:06:37 -08001161 trace_writeback_dirty_inode_start(inode, flags);
1162
Jens Axboe03ba3782009-09-09 09:08:54 +02001163 if (sb->s_op->dirty_inode)
Christoph Hellwigaa385722011-05-27 06:53:02 -04001164 sb->s_op->dirty_inode(inode, flags);
Tejun Heo9fb0a7d2013-01-11 13:06:37 -08001165
1166 trace_writeback_dirty_inode(inode, flags);
Jens Axboe03ba3782009-09-09 09:08:54 +02001167 }
1168
1169 /*
1170 * make sure that changes are seen by all cpus before we test i_state
1171 * -- mikulas
1172 */
1173 smp_mb();
1174
1175 /* avoid the locking if we can */
1176 if ((inode->i_state & flags) == flags)
1177 return;
1178
1179 if (unlikely(block_dump))
1180 block_dump___mark_inode_dirty(inode);
1181
Dave Chinner250df6e2011-03-22 22:23:36 +11001182 spin_lock(&inode->i_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001183 if ((inode->i_state & flags) != flags) {
1184 const int was_dirty = inode->i_state & I_DIRTY;
1185
1186 inode->i_state |= flags;
1187
1188 /*
1189 * If the inode is being synced, just update its dirty state.
1190 * The unlocker will place the inode on the appropriate
1191 * superblock list, based upon its state.
1192 */
1193 if (inode->i_state & I_SYNC)
Dave Chinner250df6e2011-03-22 22:23:36 +11001194 goto out_unlock_inode;
Jens Axboe03ba3782009-09-09 09:08:54 +02001195
1196 /*
1197 * Only add valid (hashed) inodes to the superblock's
1198 * dirty list. Add blockdev inodes as well.
1199 */
1200 if (!S_ISBLK(inode->i_mode)) {
Al Viro1d3382cb2010-10-23 15:19:20 -04001201 if (inode_unhashed(inode))
Dave Chinner250df6e2011-03-22 22:23:36 +11001202 goto out_unlock_inode;
Jens Axboe03ba3782009-09-09 09:08:54 +02001203 }
Al Viroa4ffdde2010-06-02 17:38:30 -04001204 if (inode->i_state & I_FREEING)
Dave Chinner250df6e2011-03-22 22:23:36 +11001205 goto out_unlock_inode;
Jens Axboe03ba3782009-09-09 09:08:54 +02001206
1207 /*
1208 * If the inode was already on b_dirty/b_io/b_more_io, don't
1209 * reposition it (that would break b_dirty time-ordering).
1210 */
1211 if (!was_dirty) {
Dave Chinnera66979a2011-03-22 22:23:41 +11001212 bool wakeup_bdi = false;
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001213 bdi = inode_to_bdi(inode);
Jens Axboe500b0672009-09-09 09:10:25 +02001214
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001215 if (bdi_cap_writeback_dirty(bdi)) {
1216 WARN(!test_bit(BDI_registered, &bdi->state),
1217 "bdi-%s not registered\n", bdi->name);
1218
1219 /*
1220 * If this is the first dirty inode for this
1221 * bdi, we have to wake-up the corresponding
1222 * bdi thread to make sure background
1223 * write-back happens later.
1224 */
1225 if (!wb_has_dirty_io(&bdi->wb))
1226 wakeup_bdi = true;
Jens Axboe500b0672009-09-09 09:10:25 +02001227 }
Jens Axboe03ba3782009-09-09 09:08:54 +02001228
Dave Chinnera66979a2011-03-22 22:23:41 +11001229 spin_unlock(&inode->i_lock);
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001230 spin_lock(&bdi->wb.list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001231 inode->dirtied_when = jiffies;
Nick Piggin7ccf19a2010-10-21 11:49:30 +11001232 list_move(&inode->i_wb_list, &bdi->wb.b_dirty);
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001233 spin_unlock(&bdi->wb.list_lock);
Dave Chinnera66979a2011-03-22 22:23:41 +11001234
1235 if (wakeup_bdi)
1236 bdi_wakeup_thread_delayed(bdi);
1237 return;
Jens Axboe03ba3782009-09-09 09:08:54 +02001238 }
1239 }
Dave Chinner250df6e2011-03-22 22:23:36 +11001240out_unlock_inode:
1241 spin_unlock(&inode->i_lock);
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001242
Jens Axboe03ba3782009-09-09 09:08:54 +02001243}
1244EXPORT_SYMBOL(__mark_inode_dirty);
1245
Jens Axboeb6e51312009-09-16 15:13:54 +02001246static void wait_sb_inodes(struct super_block *sb)
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001247{
Jens Axboe03ba3782009-09-09 09:08:54 +02001248 struct inode *inode, *old_inode = NULL;
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001249
Jens Axboe03ba3782009-09-09 09:08:54 +02001250 /*
1251 * We need to be protected against the filesystem going from
1252 * r/o to r/w or vice versa.
1253 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001254 WARN_ON(!rwsem_is_locked(&sb->s_umount));
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001255
Dave Chinner55fa6092011-03-22 22:23:40 +11001256 spin_lock(&inode_sb_list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001257
1258 /*
1259 * Data integrity sync. Must wait for all pages under writeback,
1260 * because there may have been pages dirtied before our sync
1261 * call, but which had writeout started before we write it out.
1262 * In which case, the inode may not be on the dirty list, but
1263 * we still have to wait for that writeout.
1264 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001265 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
Dave Chinner250df6e2011-03-22 22:23:36 +11001266 struct address_space *mapping = inode->i_mapping;
Jens Axboe03ba3782009-09-09 09:08:54 +02001267
Dave Chinner250df6e2011-03-22 22:23:36 +11001268 spin_lock(&inode->i_lock);
1269 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
1270 (mapping->nrpages == 0)) {
1271 spin_unlock(&inode->i_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001272 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +11001273 }
Jens Axboe03ba3782009-09-09 09:08:54 +02001274 __iget(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +11001275 spin_unlock(&inode->i_lock);
Dave Chinner55fa6092011-03-22 22:23:40 +11001276 spin_unlock(&inode_sb_list_lock);
1277
Jens Axboe03ba3782009-09-09 09:08:54 +02001278 /*
Dave Chinner55fa6092011-03-22 22:23:40 +11001279 * We hold a reference to 'inode' so it couldn't have been
1280 * removed from s_inodes list while we dropped the
1281 * inode_sb_list_lock. We cannot iput the inode now as we can
1282 * be holding the last reference and we cannot iput it under
1283 * inode_sb_list_lock. So we keep the reference and iput it
1284 * later.
Jens Axboe03ba3782009-09-09 09:08:54 +02001285 */
1286 iput(old_inode);
1287 old_inode = inode;
1288
1289 filemap_fdatawait(mapping);
1290
1291 cond_resched();
Nick Piggin38f21972009-01-06 14:40:25 -08001292
Dave Chinner55fa6092011-03-22 22:23:40 +11001293 spin_lock(&inode_sb_list_lock);
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001294 }
Dave Chinner55fa6092011-03-22 22:23:40 +11001295 spin_unlock(&inode_sb_list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001296 iput(old_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297}
1298
Jens Axboed8a85592009-09-02 12:34:32 +02001299/**
Chris Mason3259f8b2010-10-29 11:16:17 -04001300 * writeback_inodes_sb_nr - writeback dirty inodes from given super_block
Jens Axboed8a85592009-09-02 12:34:32 +02001301 * @sb: the superblock
Chris Mason3259f8b2010-10-29 11:16:17 -04001302 * @nr: the number of pages to write
Marcos Paulo de Souza786228a2011-11-23 20:56:45 +08001303 * @reason: reason why some writeback work initiated
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304 *
Jens Axboed8a85592009-09-02 12:34:32 +02001305 * Start writeback on some inodes on this super_block. No guarantees are made
1306 * on how many (if any) will be written, and this function does not wait
Chris Mason3259f8b2010-10-29 11:16:17 -04001307 * for IO completion of submitted IO.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001309void writeback_inodes_sb_nr(struct super_block *sb,
1310 unsigned long nr,
1311 enum wb_reason reason)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001313 DECLARE_COMPLETION_ONSTACK(done);
1314 struct wb_writeback_work work = {
Wu Fengguang6e6938b2010-06-06 10:38:15 -06001315 .sb = sb,
1316 .sync_mode = WB_SYNC_NONE,
1317 .tagged_writepages = 1,
1318 .done = &done,
1319 .nr_pages = nr,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001320 .reason = reason,
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001321 };
Jens Axboe0e3c9a22010-06-01 11:08:43 +02001322
Jan Kara6eedc702012-07-03 16:45:27 +02001323 if (sb->s_bdi == &noop_backing_dev_info)
1324 return;
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001325 WARN_ON(!rwsem_is_locked(&sb->s_umount));
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001326 bdi_queue_work(sb->s_bdi, &work);
1327 wait_for_completion(&done);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328}
Chris Mason3259f8b2010-10-29 11:16:17 -04001329EXPORT_SYMBOL(writeback_inodes_sb_nr);
1330
1331/**
1332 * writeback_inodes_sb - writeback dirty inodes from given super_block
1333 * @sb: the superblock
Marcos Paulo de Souza786228a2011-11-23 20:56:45 +08001334 * @reason: reason why some writeback work was initiated
Chris Mason3259f8b2010-10-29 11:16:17 -04001335 *
1336 * Start writeback on some inodes on this super_block. No guarantees are made
1337 * on how many (if any) will be written, and this function does not wait
1338 * for IO completion of submitted IO.
1339 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001340void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason)
Chris Mason3259f8b2010-10-29 11:16:17 -04001341{
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001342 return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason);
Chris Mason3259f8b2010-10-29 11:16:17 -04001343}
Jens Axboed8a85592009-09-02 12:34:32 +02001344EXPORT_SYMBOL(writeback_inodes_sb);
1345
1346/**
Miao Xie10ee27a2013-01-10 13:47:57 +08001347 * try_to_writeback_inodes_sb_nr - try to start writeback if none underway
Chris Mason3259f8b2010-10-29 11:16:17 -04001348 * @sb: the superblock
1349 * @nr: the number of pages to write
Miao Xie10ee27a2013-01-10 13:47:57 +08001350 * @reason: the reason of writeback
Chris Mason3259f8b2010-10-29 11:16:17 -04001351 *
Miao Xie10ee27a2013-01-10 13:47:57 +08001352 * Invoke writeback_inodes_sb_nr if no writeback is currently underway.
Chris Mason3259f8b2010-10-29 11:16:17 -04001353 * Returns 1 if writeback was started, 0 if not.
1354 */
Miao Xie10ee27a2013-01-10 13:47:57 +08001355int try_to_writeback_inodes_sb_nr(struct super_block *sb,
1356 unsigned long nr,
1357 enum wb_reason reason)
Chris Mason3259f8b2010-10-29 11:16:17 -04001358{
Miao Xie10ee27a2013-01-10 13:47:57 +08001359 if (writeback_in_progress(sb->s_bdi))
Chris Mason3259f8b2010-10-29 11:16:17 -04001360 return 1;
Miao Xie10ee27a2013-01-10 13:47:57 +08001361
1362 if (!down_read_trylock(&sb->s_umount))
Chris Mason3259f8b2010-10-29 11:16:17 -04001363 return 0;
Miao Xie10ee27a2013-01-10 13:47:57 +08001364
1365 writeback_inodes_sb_nr(sb, nr, reason);
1366 up_read(&sb->s_umount);
1367 return 1;
Chris Mason3259f8b2010-10-29 11:16:17 -04001368}
Miao Xie10ee27a2013-01-10 13:47:57 +08001369EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr);
1370
1371/**
1372 * try_to_writeback_inodes_sb - try to start writeback if none underway
1373 * @sb: the superblock
1374 * @reason: reason why some writeback work was initiated
1375 *
1376 * Implement by try_to_writeback_inodes_sb_nr()
1377 * Returns 1 if writeback was started, 0 if not.
1378 */
1379int try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason)
1380{
1381 return try_to_writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason);
1382}
1383EXPORT_SYMBOL(try_to_writeback_inodes_sb);
Chris Mason3259f8b2010-10-29 11:16:17 -04001384
1385/**
Jens Axboed8a85592009-09-02 12:34:32 +02001386 * sync_inodes_sb - sync sb inode pages
1387 * @sb: the superblock
1388 *
1389 * This function writes and waits on any dirty inode belonging to this
Stefan Hajnoczicb9ef8d2011-01-13 15:47:26 -08001390 * super_block.
Jens Axboed8a85592009-09-02 12:34:32 +02001391 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001392void sync_inodes_sb(struct super_block *sb)
Jens Axboed8a85592009-09-02 12:34:32 +02001393{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001394 DECLARE_COMPLETION_ONSTACK(done);
1395 struct wb_writeback_work work = {
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001396 .sb = sb,
1397 .sync_mode = WB_SYNC_ALL,
1398 .nr_pages = LONG_MAX,
1399 .range_cyclic = 0,
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001400 .done = &done,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001401 .reason = WB_REASON_SYNC,
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001402 };
1403
Jan Kara6eedc702012-07-03 16:45:27 +02001404 /* Nothing to do? */
1405 if (sb->s_bdi == &noop_backing_dev_info)
1406 return;
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001407 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1408
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001409 bdi_queue_work(sb->s_bdi, &work);
1410 wait_for_completion(&done);
1411
Jens Axboeb6e51312009-09-16 15:13:54 +02001412 wait_sb_inodes(sb);
Jens Axboed8a85592009-09-02 12:34:32 +02001413}
1414EXPORT_SYMBOL(sync_inodes_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416/**
Andrea Arcangeli7f04c262005-10-30 15:03:05 -08001417 * write_inode_now - write an inode to disk
1418 * @inode: inode to write to disk
1419 * @sync: whether the write should be synchronous or not
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420 *
Andrea Arcangeli7f04c262005-10-30 15:03:05 -08001421 * This function commits an inode to disk immediately if it is dirty. This is
1422 * primarily needed by knfsd.
1423 *
1424 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426int write_inode_now(struct inode *inode, int sync)
1427{
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001428 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429 struct writeback_control wbc = {
1430 .nr_to_write = LONG_MAX,
Mike Galbraith18914b12008-02-08 04:20:23 -08001431 .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -07001432 .range_start = 0,
1433 .range_end = LLONG_MAX,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434 };
1435
1436 if (!mapping_cap_writeback_dirty(inode->i_mapping))
Andrew Morton49364ce2005-11-07 00:59:15 -08001437 wbc.nr_to_write = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438
1439 might_sleep();
Jan Kara4f8ad652012-05-03 14:48:00 +02001440 return writeback_single_inode(inode, wb, &wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441}
1442EXPORT_SYMBOL(write_inode_now);
1443
1444/**
1445 * sync_inode - write an inode and its pages to disk.
1446 * @inode: the inode to sync
1447 * @wbc: controls the writeback mode
1448 *
1449 * sync_inode() will write an inode and its pages to disk. It will also
1450 * correctly update the inode on its superblock's dirty inode lists and will
1451 * update inode->i_state.
1452 *
1453 * The caller must have a ref on the inode.
1454 */
1455int sync_inode(struct inode *inode, struct writeback_control *wbc)
1456{
Jan Kara4f8ad652012-05-03 14:48:00 +02001457 return writeback_single_inode(inode, &inode_to_bdi(inode)->wb, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458}
1459EXPORT_SYMBOL(sync_inode);
Christoph Hellwigc37650162010-10-06 10:48:20 +02001460
1461/**
Andrew Mortonc691b9d2011-01-13 15:45:48 -08001462 * sync_inode_metadata - write an inode to disk
Christoph Hellwigc37650162010-10-06 10:48:20 +02001463 * @inode: the inode to sync
1464 * @wait: wait for I/O to complete.
1465 *
Andrew Mortonc691b9d2011-01-13 15:45:48 -08001466 * Write an inode to disk and adjust its dirty state after completion.
Christoph Hellwigc37650162010-10-06 10:48:20 +02001467 *
1468 * Note: only writes the actual inode, no associated data or other metadata.
1469 */
1470int sync_inode_metadata(struct inode *inode, int wait)
1471{
1472 struct writeback_control wbc = {
1473 .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE,
1474 .nr_to_write = 0, /* metadata-only */
1475 };
1476
1477 return sync_inode(inode, &wbc);
1478}
1479EXPORT_SYMBOL(sync_inode_metadata);