blob: 517f211a3bd45c60f607e5da7d957bc739d55d47 [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>
Jens Axboef5ff8422007-09-21 09:19:54 +020017#include <linux/module.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>
Jens Axboe03ba3782009-09-09 09:08:54 +020023#include <linux/kthread.h>
24#include <linux/freezer.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/writeback.h>
26#include <linux/blkdev.h>
27#include <linux/backing-dev.h>
28#include <linux/buffer_head.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/*
Jens Axboec4a77a62009-09-16 15:18:25 +020033 * Passed into wb_writeback(), essentially a subset of writeback_control
34 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +020035struct wb_writeback_work {
Jens Axboec4a77a62009-09-16 15:18:25 +020036 long nr_pages;
37 struct super_block *sb;
Wu Fengguangd46db3d2011-05-04 19:54:37 -060038 unsigned long *older_than_this;
Jens Axboec4a77a62009-09-16 15:18:25 +020039 enum writeback_sync_modes sync_mode;
Wu Fengguang6e6938b2010-06-06 10:38:15 -060040 unsigned int tagged_writepages:1;
H Hartley Sweeten52957fe2010-04-01 20:36:30 -050041 unsigned int for_kupdate:1;
42 unsigned int range_cyclic:1;
43 unsigned int for_background:1;
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -060044 enum wb_reason reason; /* why was writeback initiated? */
Jens Axboec4a77a62009-09-16 15:18:25 +020045
Jens Axboe8010c3b2009-09-15 20:04:57 +020046 struct list_head list; /* pending work list */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +020047 struct completion *done; /* set if the caller waits */
Jens Axboe03ba3782009-09-09 09:08:54 +020048};
49
Dave Chinner455b2862010-07-07 13:24:06 +100050/*
51 * Include the creation of the trace points after defining the
52 * wb_writeback_work structure so that the definition remains local to this
53 * file.
54 */
55#define CREATE_TRACE_POINTS
56#include <trace/events/writeback.h>
57
Dave Chinner455b2862010-07-07 13:24:06 +100058/*
59 * We don't actually have pdflush, but this one is exported though /proc...
60 */
61int nr_pdflush_threads;
62
Adrian Bunkf11b00f2008-04-29 00:58:56 -070063/**
64 * writeback_in_progress - determine whether there is writeback in progress
65 * @bdi: the device's backing_dev_info structure.
66 *
Jens Axboe03ba3782009-09-09 09:08:54 +020067 * Determine whether there is writeback waiting to be handled against a
68 * backing device.
Adrian Bunkf11b00f2008-04-29 00:58:56 -070069 */
70int writeback_in_progress(struct backing_dev_info *bdi)
71{
Jan Kara81d73a32010-08-11 14:17:44 -070072 return test_bit(BDI_writeback_running, &bdi->state);
Adrian Bunkf11b00f2008-04-29 00:58:56 -070073}
74
Jan Kara692ebd12010-09-21 11:51:01 +020075static inline struct backing_dev_info *inode_to_bdi(struct inode *inode)
76{
77 struct super_block *sb = inode->i_sb;
Jan Kara692ebd12010-09-21 11:51:01 +020078
Christoph Hellwigaaead252010-10-04 14:25:33 +020079 if (strcmp(sb->s_type->name, "bdev") == 0)
80 return inode->i_mapping->backing_dev_info;
81
82 return sb->s_bdi;
Jan Kara692ebd12010-09-21 11:51:01 +020083}
84
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
Jan Kara65850272011-01-13 15:45:44 -080090/* Wakeup flusher thread or forker thread to fork it. Requires bdi->wb_lock. */
91static void bdi_wakeup_flusher(struct backing_dev_info *bdi)
Nick Piggin4195f732009-05-28 09:01:15 +020092{
Artem Bityutskiyfff5b852010-07-25 14:29:20 +030093 if (bdi->wb.task) {
94 wake_up_process(bdi->wb.task);
95 } else {
96 /*
97 * The bdi thread isn't there, wake up the forker thread which
98 * will create and run it.
99 */
Jens Axboe03ba3782009-09-09 09:08:54 +0200100 wake_up_process(default_backing_dev_info.wb.task);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101 }
Jan Kara65850272011-01-13 15:45:44 -0800102}
103
104static void bdi_queue_work(struct backing_dev_info *bdi,
105 struct wb_writeback_work *work)
106{
107 trace_writeback_queue(bdi, work);
108
109 spin_lock_bh(&bdi->wb_lock);
110 list_add_tail(&work->list, &bdi->work_list);
111 if (!bdi->wb.task)
112 trace_writeback_nothread(bdi, work);
113 bdi_wakeup_flusher(bdi);
Artem Bityutskiy64677162010-07-25 14:29:22 +0300114 spin_unlock_bh(&bdi->wb_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115}
116
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200117static void
118__bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600119 bool range_cyclic, enum wb_reason reason)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200121 struct wb_writeback_work *work;
Jens Axboe03ba3782009-09-09 09:08:54 +0200122
Jens Axboebcddc3f2009-09-13 20:07:36 +0200123 /*
124 * This is WB_SYNC_NONE writeback, so if allocation fails just
125 * wakeup the thread for old dirty data writeback
126 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200127 work = kzalloc(sizeof(*work), GFP_ATOMIC);
128 if (!work) {
Dave Chinner455b2862010-07-07 13:24:06 +1000129 if (bdi->wb.task) {
130 trace_writeback_nowork(bdi);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200131 wake_up_process(bdi->wb.task);
Dave Chinner455b2862010-07-07 13:24:06 +1000132 }
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200133 return;
Jens Axboebcddc3f2009-09-13 20:07:36 +0200134 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200135
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200136 work->sync_mode = WB_SYNC_NONE;
137 work->nr_pages = nr_pages;
138 work->range_cyclic = range_cyclic;
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600139 work->reason = reason;
Christoph Hellwigf0fad8a2009-09-11 09:47:56 +0200140
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200141 bdi_queue_work(bdi, work);
Jens Axboeb6e51312009-09-16 15:13:54 +0200142}
143
144/**
145 * bdi_start_writeback - start writeback
146 * @bdi: the backing device to write from
147 * @nr_pages: the number of pages to write
Marcos Paulo de Souza786228a2011-11-23 20:56:45 +0800148 * @reason: reason why some writeback work was initiated
Jens Axboeb6e51312009-09-16 15:13:54 +0200149 *
150 * Description:
151 * This does WB_SYNC_NONE opportunistic writeback. The IO is only
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300152 * started when this function returns, we make no guarantees on
Jens Axboe0e3c9a22010-06-01 11:08:43 +0200153 * completion. Caller need not hold sb s_umount semaphore.
Jens Axboeb6e51312009-09-16 15:13:54 +0200154 *
155 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600156void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages,
157 enum wb_reason reason)
Jens Axboeb6e51312009-09-16 15:13:54 +0200158{
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600159 __bdi_start_writeback(bdi, nr_pages, true, reason);
Christoph Hellwigc5444192010-06-08 18:15:15 +0200160}
Wu Fengguangd3ddec72009-09-23 20:33:40 +0800161
Christoph Hellwigc5444192010-06-08 18:15:15 +0200162/**
163 * bdi_start_background_writeback - start background writeback
164 * @bdi: the backing device to write from
165 *
166 * Description:
Jan Kara65850272011-01-13 15:45:44 -0800167 * This makes sure WB_SYNC_NONE background writeback happens. When
168 * this function returns, it is only guaranteed that for given BDI
169 * some IO is happening if we are over background dirty threshold.
170 * Caller need not hold sb s_umount semaphore.
Christoph Hellwigc5444192010-06-08 18:15:15 +0200171 */
172void bdi_start_background_writeback(struct backing_dev_info *bdi)
173{
Jan Kara65850272011-01-13 15:45:44 -0800174 /*
175 * We just wake up the flusher thread. It will perform background
176 * writeback as soon as there is no other work to do.
177 */
Wu Fengguang71927e82011-01-13 15:45:46 -0800178 trace_writeback_wake_background(bdi);
Jan Kara65850272011-01-13 15:45:44 -0800179 spin_lock_bh(&bdi->wb_lock);
180 bdi_wakeup_flusher(bdi);
181 spin_unlock_bh(&bdi->wb_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182}
183
184/*
Dave Chinnera66979a2011-03-22 22:23:41 +1100185 * Remove the inode from the writeback list it is on.
186 */
187void inode_wb_list_del(struct inode *inode)
188{
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600189 struct backing_dev_info *bdi = inode_to_bdi(inode);
Dave Chinnera66979a2011-03-22 22:23:41 +1100190
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600191 spin_lock(&bdi->wb.list_lock);
192 list_del_init(&inode->i_wb_list);
193 spin_unlock(&bdi->wb.list_lock);
194}
Dave Chinnera66979a2011-03-22 22:23:41 +1100195
196/*
Andrew Morton6610a0b2007-10-16 23:30:32 -0700197 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
198 * furthest end of its superblock's dirty-inode list.
199 *
200 * Before stamping the inode's ->dirtied_when, we check to see whether it is
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200201 * already the most-recently-dirtied inode on the b_dirty list. If that is
Andrew Morton6610a0b2007-10-16 23:30:32 -0700202 * the case then the inode must have been redirtied while it was being written
203 * out and we don't reset its dirtied_when.
204 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600205static void redirty_tail(struct inode *inode, struct bdi_writeback *wb)
Andrew Morton6610a0b2007-10-16 23:30:32 -0700206{
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600207 assert_spin_locked(&wb->list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200208 if (!list_empty(&wb->b_dirty)) {
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200209 struct inode *tail;
Andrew Morton6610a0b2007-10-16 23:30:32 -0700210
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100211 tail = wb_inode(wb->b_dirty.next);
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200212 if (time_before(inode->dirtied_when, tail->dirtied_when))
Andrew Morton6610a0b2007-10-16 23:30:32 -0700213 inode->dirtied_when = jiffies;
214 }
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100215 list_move(&inode->i_wb_list, &wb->b_dirty);
Andrew Morton6610a0b2007-10-16 23:30:32 -0700216}
217
218/*
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200219 * requeue inode for re-scanning after bdi->b_io list is exhausted.
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700220 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600221static void requeue_io(struct inode *inode, struct bdi_writeback *wb)
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700222{
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600223 assert_spin_locked(&wb->list_lock);
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100224 list_move(&inode->i_wb_list, &wb->b_more_io);
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700225}
226
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700227static void inode_sync_complete(struct inode *inode)
228{
229 /*
Dave Chinnera66979a2011-03-22 22:23:41 +1100230 * Prevent speculative execution through
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600231 * spin_unlock(&wb->list_lock);
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700232 */
Dave Chinnera66979a2011-03-22 22:23:41 +1100233
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700234 smp_mb();
235 wake_up_bit(&inode->i_state, __I_SYNC);
236}
237
Jeff Laytond2caa3c2009-04-02 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 Laytond2caa3c2009-04-02 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/*
Fengguang Wu2c136572007-10-16 23:30:39 -0700254 * Move expired dirty inodes from @delaying_queue to @dispatch_queue.
255 */
Wu Fengguange84d0a42011-04-23 12:27:27 -0600256static int move_expired_inodes(struct list_head *delaying_queue,
Fengguang Wu2c136572007-10-16 23:30:39 -0700257 struct list_head *dispatch_queue,
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600258 struct wb_writeback_work *work)
Fengguang Wu2c136572007-10-16 23:30:39 -0700259{
Shaohua Li5c034492009-09-24 14:42:33 +0200260 LIST_HEAD(tmp);
261 struct list_head *pos, *node;
Jens Axboecf137302009-09-24 15:12:57 +0200262 struct super_block *sb = NULL;
Shaohua Li5c034492009-09-24 14:42:33 +0200263 struct inode *inode;
Jens Axboecf137302009-09-24 15:12:57 +0200264 int do_sb_sort = 0;
Wu Fengguange84d0a42011-04-23 12:27:27 -0600265 int moved = 0;
Shaohua Li5c034492009-09-24 14:42:33 +0200266
Fengguang Wu2c136572007-10-16 23:30:39 -0700267 while (!list_empty(delaying_queue)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100268 inode = wb_inode(delaying_queue->prev);
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600269 if (work->older_than_this &&
270 inode_dirtied_after(inode, *work->older_than_this))
Fengguang Wu2c136572007-10-16 23:30:39 -0700271 break;
Jens Axboecf137302009-09-24 15:12:57 +0200272 if (sb && sb != inode->i_sb)
273 do_sb_sort = 1;
274 sb = inode->i_sb;
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100275 list_move(&inode->i_wb_list, &tmp);
Wu Fengguange84d0a42011-04-23 12:27:27 -0600276 moved++;
Shaohua Li5c034492009-09-24 14:42:33 +0200277 }
278
Jens Axboecf137302009-09-24 15:12:57 +0200279 /* just one sb in list, splice to dispatch_queue and we're done */
280 if (!do_sb_sort) {
281 list_splice(&tmp, dispatch_queue);
Wu Fengguange84d0a42011-04-23 12:27:27 -0600282 goto out;
Jens Axboecf137302009-09-24 15:12:57 +0200283 }
284
Shaohua Li5c034492009-09-24 14:42:33 +0200285 /* Move inodes from one superblock together */
286 while (!list_empty(&tmp)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100287 sb = wb_inode(tmp.prev)->i_sb;
Shaohua Li5c034492009-09-24 14:42:33 +0200288 list_for_each_prev_safe(pos, node, &tmp) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100289 inode = wb_inode(pos);
Shaohua Li5c034492009-09-24 14:42:33 +0200290 if (inode->i_sb == sb)
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100291 list_move(&inode->i_wb_list, dispatch_queue);
Shaohua Li5c034492009-09-24 14:42:33 +0200292 }
Fengguang Wu2c136572007-10-16 23:30:39 -0700293 }
Wu Fengguange84d0a42011-04-23 12:27:27 -0600294out:
295 return moved;
Fengguang Wu2c136572007-10-16 23:30:39 -0700296}
297
298/*
299 * Queue all expired dirty inodes for io, eldest first.
Wu Fengguang4ea879b2010-08-11 14:17:42 -0700300 * Before
301 * newly dirtied b_dirty b_io b_more_io
302 * =============> gf edc BA
303 * After
304 * newly dirtied b_dirty b_io b_more_io
305 * =============> g fBAedc
306 * |
307 * +--> dequeue for IO
Fengguang Wu2c136572007-10-16 23:30:39 -0700308 */
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600309static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work)
Fengguang Wu2c136572007-10-16 23:30:39 -0700310{
Wu Fengguange84d0a42011-04-23 12:27:27 -0600311 int moved;
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600312 assert_spin_locked(&wb->list_lock);
Wu Fengguang4ea879b2010-08-11 14:17:42 -0700313 list_splice_init(&wb->b_more_io, &wb->b_io);
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600314 moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, work);
315 trace_writeback_queue_io(wb, work, moved);
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200316}
317
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100318static int write_inode(struct inode *inode, struct writeback_control *wbc)
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200319{
Jens Axboe03ba3782009-09-09 09:08:54 +0200320 if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode))
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100321 return inode->i_sb->s_op->write_inode(inode, wbc);
Jens Axboe03ba3782009-09-09 09:08:54 +0200322 return 0;
Fengguang Wu2c136572007-10-16 23:30:39 -0700323}
324
325/*
Christoph Hellwig01c03192009-06-08 13:35:40 +0200326 * Wait for writeback on an inode to complete.
327 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600328static void inode_wait_for_writeback(struct inode *inode,
329 struct bdi_writeback *wb)
Christoph Hellwig01c03192009-06-08 13:35:40 +0200330{
331 DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC);
332 wait_queue_head_t *wqh;
333
334 wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
Dave Chinner250df6e2011-03-22 22:23:36 +1100335 while (inode->i_state & I_SYNC) {
336 spin_unlock(&inode->i_lock);
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600337 spin_unlock(&wb->list_lock);
Christoph Hellwig01c03192009-06-08 13:35:40 +0200338 __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE);
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600339 spin_lock(&wb->list_lock);
Dave Chinner250df6e2011-03-22 22:23:36 +1100340 spin_lock(&inode->i_lock);
Richard Kennedy58a9d3d2010-05-24 14:32:38 -0700341 }
Christoph Hellwig01c03192009-06-08 13:35:40 +0200342}
343
344/*
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600345 * Write out an inode's dirty pages. Called under wb->list_lock and
Dave Chinner0f1b1fd2011-03-22 22:23:43 +1100346 * inode->i_lock. Either the caller has an active reference on the inode or
347 * the inode has I_WILL_FREE set.
Christoph Hellwig01c03192009-06-08 13:35:40 +0200348 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349 * If `wait' is set, wait on the writeout.
350 *
351 * The whole writeout design is quite complex and fragile. We want to avoid
352 * starvation of particular inodes when others are being redirtied, prevent
353 * livelocks, etc.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354 */
355static int
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600356writeback_single_inode(struct inode *inode, struct bdi_writeback *wb,
357 struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359 struct address_space *mapping = inode->i_mapping;
Wu Fengguang251d6a42010-12-01 17:33:37 -0600360 long nr_to_write = wbc->nr_to_write;
Christoph Hellwig01c03192009-06-08 13:35:40 +0200361 unsigned dirty;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362 int ret;
363
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600364 assert_spin_locked(&wb->list_lock);
Dave Chinner0f1b1fd2011-03-22 22:23:43 +1100365 assert_spin_locked(&inode->i_lock);
366
Christoph Hellwig01c03192009-06-08 13:35:40 +0200367 if (!atomic_read(&inode->i_count))
368 WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING)));
369 else
370 WARN_ON(inode->i_state & I_WILL_FREE);
371
372 if (inode->i_state & I_SYNC) {
373 /*
374 * If this inode is locked for writeback and we are not doing
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200375 * writeback-for-data-integrity, move it to b_more_io so that
Christoph Hellwig01c03192009-06-08 13:35:40 +0200376 * writeback can proceed with the other inodes on s_io.
377 *
378 * We'll have another go at writing back this inode when we
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200379 * completed a full scan of b_io.
Christoph Hellwig01c03192009-06-08 13:35:40 +0200380 */
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100381 if (wbc->sync_mode != WB_SYNC_ALL) {
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600382 requeue_io(inode, wb);
Wu Fengguang251d6a42010-12-01 17:33:37 -0600383 trace_writeback_single_inode_requeue(inode, wbc,
384 nr_to_write);
Christoph Hellwig01c03192009-06-08 13:35:40 +0200385 return 0;
386 }
387
388 /*
389 * It's a data-integrity sync. We must wait.
390 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600391 inode_wait_for_writeback(inode, wb);
Christoph Hellwig01c03192009-06-08 13:35:40 +0200392 }
393
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700394 BUG_ON(inode->i_state & I_SYNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395
Dmitry Monakhov5547e8a2010-05-07 13:35:44 +0400396 /* Set I_SYNC, reset I_DIRTY_PAGES */
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700397 inode->i_state |= I_SYNC;
Dmitry Monakhov5547e8a2010-05-07 13:35:44 +0400398 inode->i_state &= ~I_DIRTY_PAGES;
Dave Chinner250df6e2011-03-22 22:23:36 +1100399 spin_unlock(&inode->i_lock);
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600400 spin_unlock(&wb->list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401
402 ret = do_writepages(mapping, wbc);
403
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100404 /*
405 * Make sure to wait on the data before writing out the metadata.
406 * This is important for filesystems that modify metadata on data
407 * I/O completion.
408 */
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100409 if (wbc->sync_mode == WB_SYNC_ALL) {
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100410 int err = filemap_fdatawait(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 if (ret == 0)
412 ret = err;
413 }
414
Dmitry Monakhov5547e8a2010-05-07 13:35:44 +0400415 /*
416 * Some filesystems may redirty the inode during the writeback
417 * due to delalloc, clear dirty metadata flags right before
418 * write_inode()
419 */
Dave Chinner250df6e2011-03-22 22:23:36 +1100420 spin_lock(&inode->i_lock);
Dmitry Monakhov5547e8a2010-05-07 13:35:44 +0400421 dirty = inode->i_state & I_DIRTY;
422 inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC);
Dave Chinner250df6e2011-03-22 22:23:36 +1100423 spin_unlock(&inode->i_lock);
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100424 /* Don't write the inode if only I_DIRTY_PAGES was set */
425 if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100426 int err = write_inode(inode, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427 if (ret == 0)
428 ret = err;
429 }
430
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600431 spin_lock(&wb->list_lock);
Dave Chinner250df6e2011-03-22 22:23:36 +1100432 spin_lock(&inode->i_lock);
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700433 inode->i_state &= ~I_SYNC;
Al Viroa4ffdde2010-06-02 17:38:30 -0400434 if (!(inode->i_state & I_FREEING)) {
Wu Fengguang94c3dcb2011-04-27 19:05:21 -0600435 /*
436 * Sync livelock prevention. Each inode is tagged and synced in
437 * one shot. If still dirty, it will be redirty_tail()'ed below.
438 * Update the dirty time to prevent enqueue and sync it again.
439 */
440 if ((inode->i_state & I_DIRTY) &&
441 (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages))
442 inode->dirtied_when = jiffies;
443
Wu Fengguang23539af2010-08-11 14:17:41 -0700444 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445 /*
446 * We didn't write back all the pages. nfs_writepages()
Wu Fengguanga50aeb42010-08-11 14:17:43 -0700447 * sometimes bales out without doing anything.
Andrew Morton1b43ef92007-10-16 23:30:35 -0700448 */
Wu Fengguanga50aeb42010-08-11 14:17:43 -0700449 inode->i_state |= I_DIRTY_PAGES;
450 if (wbc->nr_to_write <= 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451 /*
Wu Fengguanga50aeb42010-08-11 14:17:43 -0700452 * slice used up: queue for next turn
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600454 requeue_io(inode, wb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455 } else {
456 /*
Wu Fengguanga50aeb42010-08-11 14:17:43 -0700457 * Writeback blocked by something other than
458 * congestion. Delay the inode for some time to
459 * avoid spinning on the CPU (100% iowait)
460 * retrying writeback of the dirty page/inode
461 * that cannot be performed immediately.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600463 redirty_tail(inode, wb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 }
Wu Fengguang23539af2010-08-11 14:17:41 -0700465 } else if (inode->i_state & I_DIRTY) {
466 /*
467 * Filesystems can dirty the inode during writeback
468 * operations, such as delayed allocation during
469 * submission or metadata updates after data IO
470 * completion.
471 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600472 redirty_tail(inode, wb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473 } else {
474 /*
Nick Piggin9e38d862010-10-23 06:55:17 -0400475 * The inode is clean. At this point we either have
476 * a reference to the inode or it's on it's way out.
477 * No need to add it back to the LRU.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 */
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100479 list_del_init(&inode->i_wb_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480 }
481 }
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700482 inode_sync_complete(inode);
Wu Fengguang251d6a42010-12-01 17:33:37 -0600483 trace_writeback_single_inode(inode, wbc, nr_to_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484 return ret;
485}
486
Wu Fengguang1a12d8b2010-08-29 13:28:09 -0600487static long writeback_chunk_size(struct backing_dev_info *bdi,
488 struct wb_writeback_work *work)
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600489{
490 long pages;
491
492 /*
493 * WB_SYNC_ALL mode does livelock avoidance by syncing dirty
494 * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX
495 * here avoids calling into writeback_inodes_wb() more than once.
496 *
497 * The intended call sequence for WB_SYNC_ALL writeback is:
498 *
499 * wb_writeback()
500 * writeback_sb_inodes() <== called only once
501 * write_cache_pages() <== called once for each inode
502 * (quickly) tag currently dirty pages
503 * (maybe slowly) sync all tagged pages
504 */
505 if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages)
506 pages = LONG_MAX;
Wu Fengguang1a12d8b2010-08-29 13:28:09 -0600507 else {
508 pages = min(bdi->avg_write_bandwidth / 2,
509 global_dirty_limit / DIRTY_SCOPE);
510 pages = min(pages, work->nr_pages);
511 pages = round_down(pages + MIN_WRITEBACK_PAGES,
512 MIN_WRITEBACK_PAGES);
513 }
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600514
515 return pages;
516}
517
Jens Axboe03ba3782009-09-09 09:08:54 +0200518/*
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800519 * Write a portion of b_io inodes which belong to @sb.
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200520 *
521 * If @only_this_sb is true, then find and write all such
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800522 * inodes. Otherwise write only ones which go sequentially
523 * in reverse order.
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200524 *
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600525 * Return the number of pages and/or inodes written.
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800526 */
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600527static long writeback_sb_inodes(struct super_block *sb,
528 struct bdi_writeback *wb,
529 struct wb_writeback_work *work)
Jens Axboe03ba3782009-09-09 09:08:54 +0200530{
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600531 struct writeback_control wbc = {
532 .sync_mode = work->sync_mode,
533 .tagged_writepages = work->tagged_writepages,
534 .for_kupdate = work->for_kupdate,
535 .for_background = work->for_background,
536 .range_cyclic = work->range_cyclic,
537 .range_start = 0,
538 .range_end = LLONG_MAX,
539 };
540 unsigned long start_time = jiffies;
541 long write_chunk;
542 long wrote = 0; /* count both pages and inodes */
543
Jens Axboe03ba3782009-09-09 09:08:54 +0200544 while (!list_empty(&wb->b_io)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100545 struct inode *inode = wb_inode(wb->b_io.prev);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200546
547 if (inode->i_sb != sb) {
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600548 if (work->sb) {
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200549 /*
550 * We only want to write back data for this
551 * superblock, move all inodes not belonging
552 * to it back onto the dirty list.
553 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600554 redirty_tail(inode, wb);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200555 continue;
556 }
557
558 /*
559 * The inode belongs to a different superblock.
560 * Bounce back to the caller to unpin this and
561 * pin the next superblock.
562 */
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600563 break;
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200564 }
565
Christoph Hellwig9843b762010-10-24 19:40:46 +0200566 /*
567 * Don't bother with new inodes or inodes beeing freed, first
568 * kind does not need peridic writeout yet, and for the latter
569 * kind writeout is handled by the freer.
570 */
Dave Chinner250df6e2011-03-22 22:23:36 +1100571 spin_lock(&inode->i_lock);
Christoph Hellwig9843b762010-10-24 19:40:46 +0200572 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
Dave Chinner250df6e2011-03-22 22:23:36 +1100573 spin_unlock(&inode->i_lock);
Wu Fengguangfcc5c222011-07-11 23:08:50 -0700574 redirty_tail(inode, wb);
Nick Piggin7ef0d732009-03-12 14:31:38 -0700575 continue;
576 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577 __iget(inode);
Wu Fengguang1a12d8b2010-08-29 13:28:09 -0600578 write_chunk = writeback_chunk_size(wb->bdi, work);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600579 wbc.nr_to_write = write_chunk;
580 wbc.pages_skipped = 0;
Dave Chinner250df6e2011-03-22 22:23:36 +1100581
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600582 writeback_single_inode(inode, wb, &wbc);
583
584 work->nr_pages -= write_chunk - wbc.nr_to_write;
585 wrote += write_chunk - wbc.nr_to_write;
586 if (!(inode->i_state & I_DIRTY))
587 wrote++;
588 if (wbc.pages_skipped) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589 /*
590 * writeback is not making progress due to locked
591 * buffers. Skip this inode for now.
592 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600593 redirty_tail(inode, wb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 }
Dave Chinner0f1b1fd2011-03-22 22:23:43 +1100595 spin_unlock(&inode->i_lock);
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600596 spin_unlock(&wb->list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 iput(inode);
OGAWA Hirofumi4ffc8442006-03-25 03:07:44 -0800598 cond_resched();
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600599 spin_lock(&wb->list_lock);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600600 /*
601 * bail out to wb_writeback() often enough to check
602 * background threshold and other termination conditions.
603 */
604 if (wrote) {
605 if (time_is_before_jiffies(start_time + HZ / 10UL))
606 break;
607 if (work->nr_pages <= 0)
608 break;
Fengguang Wu8bc3be22008-02-04 22:29:36 -0800609 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 }
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600611 return wrote;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800612}
Nick Piggin38f21972009-01-06 14:40:25 -0800613
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600614static long __writeback_inodes_wb(struct bdi_writeback *wb,
615 struct wb_writeback_work *work)
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800616{
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600617 unsigned long start_time = jiffies;
618 long wrote = 0;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800619
620 while (!list_empty(&wb->b_io)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100621 struct inode *inode = wb_inode(wb->b_io.prev);
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800622 struct super_block *sb = inode->i_sb;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800623
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000624 if (!grab_super_passive(sb)) {
Wu Fengguang0e995812011-07-29 22:14:35 -0600625 /*
626 * grab_super_passive() may fail consistently due to
627 * s_umount being grabbed by someone else. Don't use
628 * requeue_io() to avoid busy retrying the inode/sb.
629 */
630 redirty_tail(inode, wb);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200631 continue;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800632 }
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600633 wrote += writeback_sb_inodes(sb, wb, work);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200634 drop_super(sb);
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800635
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600636 /* refer to the same tests at the end of writeback_sb_inodes */
637 if (wrote) {
638 if (time_is_before_jiffies(start_time + HZ / 10UL))
639 break;
640 if (work->nr_pages <= 0)
641 break;
642 }
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800643 }
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200644 /* Leave any unwritten inodes on b_io */
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600645 return wrote;
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200646}
647
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600648long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages,
649 enum wb_reason reason)
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200650{
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600651 struct wb_writeback_work work = {
652 .nr_pages = nr_pages,
653 .sync_mode = WB_SYNC_NONE,
654 .range_cyclic = 1,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600655 .reason = reason,
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600656 };
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200657
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600658 spin_lock(&wb->list_lock);
Wu Fengguang424b3512010-07-21 20:11:53 -0600659 if (list_empty(&wb->b_io))
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600660 queue_io(wb, &work);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600661 __writeback_inodes_wb(wb, &work);
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600662 spin_unlock(&wb->list_lock);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200663
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600664 return nr_pages - work.nr_pages;
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200665}
666
Wu Fengguangb00949a2010-11-18 14:38:33 -0600667static bool over_bground_thresh(struct backing_dev_info *bdi)
Jens Axboe03ba3782009-09-09 09:08:54 +0200668{
669 unsigned long background_thresh, dirty_thresh;
670
Wu Fengguang16c40422010-08-11 14:17:39 -0700671 global_dirty_limits(&background_thresh, &dirty_thresh);
Jens Axboe03ba3782009-09-09 09:08:54 +0200672
Wu Fengguangb00949a2010-11-18 14:38:33 -0600673 if (global_page_state(NR_FILE_DIRTY) +
674 global_page_state(NR_UNSTABLE_NFS) > background_thresh)
675 return true;
676
677 if (bdi_stat(bdi, BDI_RECLAIMABLE) >
678 bdi_dirty_limit(bdi, background_thresh))
679 return true;
680
681 return false;
Jens Axboe03ba3782009-09-09 09:08:54 +0200682}
683
684/*
Wu Fengguange98be2d2010-08-29 11:22:30 -0600685 * Called under wb->list_lock. If there are multiple wb per bdi,
686 * only the flusher working on the first wb should do it.
687 */
688static void wb_update_bandwidth(struct bdi_writeback *wb,
689 unsigned long start_time)
690{
Wu Fengguangaf6a3112011-10-03 20:46:17 -0600691 __bdi_update_bandwidth(wb->bdi, 0, 0, 0, 0, 0, start_time);
Wu Fengguange98be2d2010-08-29 11:22:30 -0600692}
693
694/*
Jens Axboe03ba3782009-09-09 09:08:54 +0200695 * Explicit flushing or periodic writeback of "old" data.
696 *
697 * Define "old": the first time one of an inode's pages is dirtied, we mark the
698 * dirtying-time in the inode's address_space. So this periodic writeback code
699 * just walks the superblock inode list, writing back any inodes which are
700 * older than a specific point in time.
701 *
702 * Try to run once per dirty_writeback_interval. But if a writeback event
703 * takes longer than a dirty_writeback_interval interval, then leave a
704 * one-second gap.
705 *
706 * older_than_this takes precedence over nr_to_write. So we'll only write back
707 * all dirty pages if they are all attached to "old" mappings.
708 */
Jens Axboec4a77a62009-09-16 15:18:25 +0200709static long wb_writeback(struct bdi_writeback *wb,
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200710 struct wb_writeback_work *work)
Jens Axboe03ba3782009-09-09 09:08:54 +0200711{
Wu Fengguange98be2d2010-08-29 11:22:30 -0600712 unsigned long wb_start = jiffies;
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600713 long nr_pages = work->nr_pages;
Jens Axboe03ba3782009-09-09 09:08:54 +0200714 unsigned long oldest_jif;
Jan Karaa5989bd2009-09-16 19:22:48 +0200715 struct inode *inode;
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600716 long progress;
Jens Axboe03ba3782009-09-09 09:08:54 +0200717
Wu Fengguange185dda2011-04-23 11:26:07 -0600718 oldest_jif = jiffies;
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600719 work->older_than_this = &oldest_jif;
Jens Axboe03ba3782009-09-09 09:08:54 +0200720
Wu Fengguange8dfc302011-04-21 12:06:32 -0600721 spin_lock(&wb->list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200722 for (;;) {
723 /*
Wu Fengguangd3ddec72009-09-23 20:33:40 +0800724 * Stop writeback when nr_pages has been consumed
Jens Axboe03ba3782009-09-09 09:08:54 +0200725 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200726 if (work->nr_pages <= 0)
Jens Axboe03ba3782009-09-09 09:08:54 +0200727 break;
728
729 /*
Jan Karaaa373cf2011-01-13 15:45:47 -0800730 * Background writeout and kupdate-style writeback may
731 * run forever. Stop them if there is other work to do
732 * so that e.g. sync can proceed. They'll be restarted
733 * after the other works are all done.
734 */
735 if ((work->for_background || work->for_kupdate) &&
736 !list_empty(&wb->bdi->work_list))
737 break;
738
739 /*
Wu Fengguangd3ddec72009-09-23 20:33:40 +0800740 * For background writeout, stop when we are below the
741 * background dirty threshold
Jens Axboe03ba3782009-09-09 09:08:54 +0200742 */
Wu Fengguangb00949a2010-11-18 14:38:33 -0600743 if (work->for_background && !over_bground_thresh(wb->bdi))
Jens Axboe03ba3782009-09-09 09:08:54 +0200744 break;
745
Wu Fengguangba9aa832010-07-21 20:32:30 -0600746 if (work->for_kupdate) {
747 oldest_jif = jiffies -
748 msecs_to_jiffies(dirty_expire_interval * 10);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600749 work->older_than_this = &oldest_jif;
Wu Fengguangba9aa832010-07-21 20:32:30 -0600750 }
Dave Chinner028c2dd2010-07-07 13:24:07 +1000751
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600752 trace_writeback_start(wb->bdi, work);
Wu Fengguange8dfc302011-04-21 12:06:32 -0600753 if (list_empty(&wb->b_io))
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600754 queue_io(wb, work);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200755 if (work->sb)
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600756 progress = writeback_sb_inodes(work->sb, wb, work);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200757 else
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600758 progress = __writeback_inodes_wb(wb, work);
759 trace_writeback_written(wb->bdi, work);
Dave Chinner028c2dd2010-07-07 13:24:07 +1000760
Wu Fengguange98be2d2010-08-29 11:22:30 -0600761 wb_update_bandwidth(wb, wb_start);
Jens Axboe03ba3782009-09-09 09:08:54 +0200762
763 /*
Jens Axboe71fd05a2009-09-23 19:32:26 +0200764 * Did we write something? Try for more
Wu Fengguange6fb6da2010-07-22 10:23:44 -0600765 *
766 * Dirty inodes are moved to b_io for writeback in batches.
767 * The completion of the current batch does not necessarily
768 * mean the overall work is done. So we keep looping as long
769 * as made some progress on cleaning pages or inodes.
Jens Axboe71fd05a2009-09-23 19:32:26 +0200770 */
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600771 if (progress)
Jens Axboe71fd05a2009-09-23 19:32:26 +0200772 continue;
773 /*
Wu Fengguange6fb6da2010-07-22 10:23:44 -0600774 * No more inodes for IO, bail
Jens Axboe71fd05a2009-09-23 19:32:26 +0200775 */
Wu Fengguangb7a24412010-07-21 22:19:51 -0600776 if (list_empty(&wb->b_more_io))
Jens Axboe03ba3782009-09-09 09:08:54 +0200777 break;
778 /*
Jens Axboe8010c3b2009-09-15 20:04:57 +0200779 * Nothing written. Wait for some inode to
780 * become available for writeback. Otherwise
781 * we'll just busyloop.
Jens Axboe03ba3782009-09-09 09:08:54 +0200782 */
Jens Axboe03ba3782009-09-09 09:08:54 +0200783 if (!list_empty(&wb->b_more_io)) {
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600784 trace_writeback_wait(wb->bdi, work);
Jens Axboe03ba3782009-09-09 09:08:54 +0200785 inode = wb_inode(wb->b_more_io.prev);
Dave Chinner250df6e2011-03-22 22:23:36 +1100786 spin_lock(&inode->i_lock);
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600787 inode_wait_for_writeback(inode, wb);
Dave Chinner250df6e2011-03-22 22:23:36 +1100788 spin_unlock(&inode->i_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200789 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200790 }
Wu Fengguange8dfc302011-04-21 12:06:32 -0600791 spin_unlock(&wb->list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200792
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600793 return nr_pages - work->nr_pages;
Jens Axboe03ba3782009-09-09 09:08:54 +0200794}
795
796/*
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200797 * Return the next wb_writeback_work struct that hasn't been processed yet.
Jens Axboe03ba3782009-09-09 09:08:54 +0200798 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200799static struct wb_writeback_work *
Minchan Kim08852b62010-08-03 12:51:16 +0200800get_next_work_item(struct backing_dev_info *bdi)
Jens Axboe03ba3782009-09-09 09:08:54 +0200801{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200802 struct wb_writeback_work *work = NULL;
Jens Axboe03ba3782009-09-09 09:08:54 +0200803
Artem Bityutskiy64677162010-07-25 14:29:22 +0300804 spin_lock_bh(&bdi->wb_lock);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200805 if (!list_empty(&bdi->work_list)) {
806 work = list_entry(bdi->work_list.next,
807 struct wb_writeback_work, list);
808 list_del_init(&work->list);
Jens Axboe03ba3782009-09-09 09:08:54 +0200809 }
Artem Bityutskiy64677162010-07-25 14:29:22 +0300810 spin_unlock_bh(&bdi->wb_lock);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200811 return work;
Jens Axboe03ba3782009-09-09 09:08:54 +0200812}
813
Linus Torvaldscdf01dd2010-10-30 08:55:52 -0700814/*
815 * Add in the number of potentially dirty inodes, because each inode
816 * write can dirty pagecache in the underlying blockdev.
817 */
818static unsigned long get_nr_dirty_pages(void)
819{
820 return global_page_state(NR_FILE_DIRTY) +
821 global_page_state(NR_UNSTABLE_NFS) +
822 get_nr_dirty_inodes();
823}
824
Jan Kara65850272011-01-13 15:45:44 -0800825static long wb_check_background_flush(struct bdi_writeback *wb)
826{
Wu Fengguangb00949a2010-11-18 14:38:33 -0600827 if (over_bground_thresh(wb->bdi)) {
Jan Kara65850272011-01-13 15:45:44 -0800828
829 struct wb_writeback_work work = {
830 .nr_pages = LONG_MAX,
831 .sync_mode = WB_SYNC_NONE,
832 .for_background = 1,
833 .range_cyclic = 1,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600834 .reason = WB_REASON_BACKGROUND,
Jan Kara65850272011-01-13 15:45:44 -0800835 };
836
837 return wb_writeback(wb, &work);
838 }
839
840 return 0;
841}
842
Jens Axboe03ba3782009-09-09 09:08:54 +0200843static long wb_check_old_data_flush(struct bdi_writeback *wb)
844{
845 unsigned long expired;
846 long nr_pages;
847
Jens Axboe69b62d02010-05-17 12:51:03 +0200848 /*
849 * When set to zero, disable periodic writeback
850 */
851 if (!dirty_writeback_interval)
852 return 0;
853
Jens Axboe03ba3782009-09-09 09:08:54 +0200854 expired = wb->last_old_flush +
855 msecs_to_jiffies(dirty_writeback_interval * 10);
856 if (time_before(jiffies, expired))
857 return 0;
858
859 wb->last_old_flush = jiffies;
Linus Torvaldscdf01dd2010-10-30 08:55:52 -0700860 nr_pages = get_nr_dirty_pages();
Jens Axboe03ba3782009-09-09 09:08:54 +0200861
Jens Axboec4a77a62009-09-16 15:18:25 +0200862 if (nr_pages) {
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200863 struct wb_writeback_work work = {
Jens Axboec4a77a62009-09-16 15:18:25 +0200864 .nr_pages = nr_pages,
865 .sync_mode = WB_SYNC_NONE,
866 .for_kupdate = 1,
867 .range_cyclic = 1,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600868 .reason = WB_REASON_PERIODIC,
Jens Axboec4a77a62009-09-16 15:18:25 +0200869 };
870
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200871 return wb_writeback(wb, &work);
Jens Axboec4a77a62009-09-16 15:18:25 +0200872 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200873
874 return 0;
875}
876
877/*
878 * Retrieve work items and do the writeback they describe
879 */
880long wb_do_writeback(struct bdi_writeback *wb, int force_wait)
881{
882 struct backing_dev_info *bdi = wb->bdi;
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200883 struct wb_writeback_work *work;
Jens Axboec4a77a62009-09-16 15:18:25 +0200884 long wrote = 0;
Jens Axboe03ba3782009-09-09 09:08:54 +0200885
Jan Kara81d73a32010-08-11 14:17:44 -0700886 set_bit(BDI_writeback_running, &wb->bdi->state);
Minchan Kim08852b62010-08-03 12:51:16 +0200887 while ((work = get_next_work_item(bdi)) != NULL) {
Jens Axboe03ba3782009-09-09 09:08:54 +0200888 /*
889 * Override sync mode, in case we must wait for completion
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200890 * because this thread is exiting now.
Jens Axboe03ba3782009-09-09 09:08:54 +0200891 */
892 if (force_wait)
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200893 work->sync_mode = WB_SYNC_ALL;
894
Dave Chinner455b2862010-07-07 13:24:06 +1000895 trace_writeback_exec(bdi, work);
896
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200897 wrote += wb_writeback(wb, work);
Jens Axboe03ba3782009-09-09 09:08:54 +0200898
899 /*
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200900 * Notify the caller of completion if this is a synchronous
901 * work item, otherwise just free it.
Jens Axboe03ba3782009-09-09 09:08:54 +0200902 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200903 if (work->done)
904 complete(work->done);
905 else
906 kfree(work);
Jens Axboe03ba3782009-09-09 09:08:54 +0200907 }
908
909 /*
910 * Check for periodic writeback, kupdated() style
911 */
912 wrote += wb_check_old_data_flush(wb);
Jan Kara65850272011-01-13 15:45:44 -0800913 wrote += wb_check_background_flush(wb);
Jan Kara81d73a32010-08-11 14:17:44 -0700914 clear_bit(BDI_writeback_running, &wb->bdi->state);
Jens Axboe03ba3782009-09-09 09:08:54 +0200915
916 return wrote;
917}
918
919/*
920 * Handle writeback of dirty data for the device backed by this bdi. Also
921 * wakes up periodically and does kupdated style flushing.
922 */
Christoph Hellwig08243902010-06-19 23:08:22 +0200923int bdi_writeback_thread(void *data)
Jens Axboe03ba3782009-09-09 09:08:54 +0200924{
Christoph Hellwig08243902010-06-19 23:08:22 +0200925 struct bdi_writeback *wb = data;
926 struct backing_dev_info *bdi = wb->bdi;
Jens Axboe03ba3782009-09-09 09:08:54 +0200927 long pages_written;
928
Peter Zijlstra766f9162010-10-26 14:22:45 -0700929 current->flags |= PF_SWAPWRITE;
Christoph Hellwig08243902010-06-19 23:08:22 +0200930 set_freezable();
Artem Bityutskiyecd58402010-07-25 14:29:18 +0300931 wb->last_active = jiffies;
Christoph Hellwig08243902010-06-19 23:08:22 +0200932
933 /*
934 * Our parent may run at a different priority, just set us to normal
935 */
936 set_user_nice(current, 0);
937
Dave Chinner455b2862010-07-07 13:24:06 +1000938 trace_writeback_thread_start(bdi);
939
Jens Axboe03ba3782009-09-09 09:08:54 +0200940 while (!kthread_should_stop()) {
Artem Bityutskiy64677162010-07-25 14:29:22 +0300941 /*
942 * Remove own delayed wake-up timer, since we are already awake
943 * and we'll take care of the preriodic write-back.
944 */
945 del_timer(&wb->wakeup_timer);
946
Jens Axboe03ba3782009-09-09 09:08:54 +0200947 pages_written = wb_do_writeback(wb, 0);
948
Dave Chinner455b2862010-07-07 13:24:06 +1000949 trace_writeback_pages_written(pages_written);
Jens Axboe03ba3782009-09-09 09:08:54 +0200950
Jens Axboe03ba3782009-09-09 09:08:54 +0200951 if (pages_written)
Artem Bityutskiyecd58402010-07-25 14:29:18 +0300952 wb->last_active = jiffies;
Jens Axboe03ba3782009-09-09 09:08:54 +0200953
Artem Bityutskiy297252c2010-07-25 14:29:15 +0300954 set_current_state(TASK_INTERRUPTIBLE);
J. Bruce Fieldsb76b4012010-08-28 08:52:10 +0200955 if (!list_empty(&bdi->work_list) || kthread_should_stop()) {
Artem Bityutskiy297252c2010-07-25 14:29:15 +0300956 __set_current_state(TASK_RUNNING);
957 continue;
Jens Axboe03ba3782009-09-09 09:08:54 +0200958 }
959
Artem Bityutskiy253c34e2010-07-25 14:29:21 +0300960 if (wb_has_dirty_io(wb) && dirty_writeback_interval)
Artem Bityutskiyfff5b852010-07-25 14:29:20 +0300961 schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10));
Artem Bityutskiy253c34e2010-07-25 14:29:21 +0300962 else {
963 /*
964 * We have nothing to do, so can go sleep without any
965 * timeout and save power. When a work is queued or
966 * something is made dirty - we will be woken up.
967 */
Artem Bityutskiy297252c2010-07-25 14:29:15 +0300968 schedule();
Jens Axboef9eadbb2010-05-18 14:31:45 +0200969 }
Jens Axboe69b62d02010-05-17 12:51:03 +0200970
Jens Axboe03ba3782009-09-09 09:08:54 +0200971 try_to_freeze();
972 }
973
Artem Bityutskiyfff5b852010-07-25 14:29:20 +0300974 /* Flush any work that raced with us exiting */
Christoph Hellwig08243902010-06-19 23:08:22 +0200975 if (!list_empty(&bdi->work_list))
976 wb_do_writeback(wb, 1);
Dave Chinner455b2862010-07-07 13:24:06 +1000977
978 trace_writeback_thread_stop(bdi);
Jens Axboe03ba3782009-09-09 09:08:54 +0200979 return 0;
980}
981
Christoph Hellwig08243902010-06-19 23:08:22 +0200982
Jens Axboe03ba3782009-09-09 09:08:54 +0200983/*
Jens Axboe03ba3782009-09-09 09:08:54 +0200984 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
985 * the whole world.
986 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600987void wakeup_flusher_threads(long nr_pages, enum wb_reason reason)
Jens Axboe03ba3782009-09-09 09:08:54 +0200988{
Christoph Hellwigb8c2f342010-06-08 18:15:07 +0200989 struct backing_dev_info *bdi;
Christoph Hellwigb8c2f342010-06-08 18:15:07 +0200990
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200991 if (!nr_pages) {
992 nr_pages = global_page_state(NR_FILE_DIRTY) +
Jens Axboe03ba3782009-09-09 09:08:54 +0200993 global_page_state(NR_UNSTABLE_NFS);
Christoph Hellwigb8c2f342010-06-08 18:15:07 +0200994 }
995
996 rcu_read_lock();
997 list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) {
998 if (!bdi_has_dirty_io(bdi))
999 continue;
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001000 __bdi_start_writeback(bdi, nr_pages, false, reason);
Christoph Hellwigb8c2f342010-06-08 18:15:07 +02001001 }
1002 rcu_read_unlock();
Jens Axboe03ba3782009-09-09 09:08:54 +02001003}
1004
1005static noinline void block_dump___mark_inode_dirty(struct inode *inode)
1006{
1007 if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) {
1008 struct dentry *dentry;
1009 const char *name = "?";
1010
1011 dentry = d_find_alias(inode);
1012 if (dentry) {
1013 spin_lock(&dentry->d_lock);
1014 name = (const char *) dentry->d_name.name;
1015 }
1016 printk(KERN_DEBUG
1017 "%s(%d): dirtied inode %lu (%s) on %s\n",
1018 current->comm, task_pid_nr(current), inode->i_ino,
1019 name, inode->i_sb->s_id);
1020 if (dentry) {
1021 spin_unlock(&dentry->d_lock);
1022 dput(dentry);
1023 }
1024 }
1025}
1026
1027/**
1028 * __mark_inode_dirty - internal function
1029 * @inode: inode to mark
1030 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
1031 * Mark an inode as dirty. Callers should use mark_inode_dirty or
1032 * mark_inode_dirty_sync.
1033 *
1034 * Put the inode on the super block's dirty list.
1035 *
1036 * CAREFUL! We mark it dirty unconditionally, but move it onto the
1037 * dirty list only if it is hashed or if it refers to a blockdev.
1038 * If it was not hashed, it will never be added to the dirty list
1039 * even if it is later hashed, as it will have been marked dirty already.
1040 *
1041 * In short, make sure you hash any inodes _before_ you start marking
1042 * them dirty.
1043 *
Jens Axboe03ba3782009-09-09 09:08:54 +02001044 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
1045 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
1046 * the kernel-internal blockdev inode represents the dirtying time of the
1047 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
1048 * page->mapping->host, so the page-dirtying time is recorded in the internal
1049 * blockdev inode.
1050 */
1051void __mark_inode_dirty(struct inode *inode, int flags)
1052{
1053 struct super_block *sb = inode->i_sb;
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001054 struct backing_dev_info *bdi = NULL;
Jens Axboe03ba3782009-09-09 09:08:54 +02001055
1056 /*
1057 * Don't do this for I_DIRTY_PAGES - that doesn't actually
1058 * dirty the inode itself
1059 */
1060 if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
1061 if (sb->s_op->dirty_inode)
Christoph Hellwigaa385722011-05-27 06:53:02 -04001062 sb->s_op->dirty_inode(inode, flags);
Jens Axboe03ba3782009-09-09 09:08:54 +02001063 }
1064
1065 /*
1066 * make sure that changes are seen by all cpus before we test i_state
1067 * -- mikulas
1068 */
1069 smp_mb();
1070
1071 /* avoid the locking if we can */
1072 if ((inode->i_state & flags) == flags)
1073 return;
1074
1075 if (unlikely(block_dump))
1076 block_dump___mark_inode_dirty(inode);
1077
Dave Chinner250df6e2011-03-22 22:23:36 +11001078 spin_lock(&inode->i_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001079 if ((inode->i_state & flags) != flags) {
1080 const int was_dirty = inode->i_state & I_DIRTY;
1081
1082 inode->i_state |= flags;
1083
1084 /*
1085 * If the inode is being synced, just update its dirty state.
1086 * The unlocker will place the inode on the appropriate
1087 * superblock list, based upon its state.
1088 */
1089 if (inode->i_state & I_SYNC)
Dave Chinner250df6e2011-03-22 22:23:36 +11001090 goto out_unlock_inode;
Jens Axboe03ba3782009-09-09 09:08:54 +02001091
1092 /*
1093 * Only add valid (hashed) inodes to the superblock's
1094 * dirty list. Add blockdev inodes as well.
1095 */
1096 if (!S_ISBLK(inode->i_mode)) {
Al Viro1d3382c2010-10-23 15:19:20 -04001097 if (inode_unhashed(inode))
Dave Chinner250df6e2011-03-22 22:23:36 +11001098 goto out_unlock_inode;
Jens Axboe03ba3782009-09-09 09:08:54 +02001099 }
Al Viroa4ffdde2010-06-02 17:38:30 -04001100 if (inode->i_state & I_FREEING)
Dave Chinner250df6e2011-03-22 22:23:36 +11001101 goto out_unlock_inode;
Jens Axboe03ba3782009-09-09 09:08:54 +02001102
1103 /*
1104 * If the inode was already on b_dirty/b_io/b_more_io, don't
1105 * reposition it (that would break b_dirty time-ordering).
1106 */
1107 if (!was_dirty) {
Dave Chinnera66979a2011-03-22 22:23:41 +11001108 bool wakeup_bdi = false;
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001109 bdi = inode_to_bdi(inode);
Jens Axboe500b0672009-09-09 09:10:25 +02001110
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001111 if (bdi_cap_writeback_dirty(bdi)) {
1112 WARN(!test_bit(BDI_registered, &bdi->state),
1113 "bdi-%s not registered\n", bdi->name);
1114
1115 /*
1116 * If this is the first dirty inode for this
1117 * bdi, we have to wake-up the corresponding
1118 * bdi thread to make sure background
1119 * write-back happens later.
1120 */
1121 if (!wb_has_dirty_io(&bdi->wb))
1122 wakeup_bdi = true;
Jens Axboe500b0672009-09-09 09:10:25 +02001123 }
Jens Axboe03ba3782009-09-09 09:08:54 +02001124
Dave Chinnera66979a2011-03-22 22:23:41 +11001125 spin_unlock(&inode->i_lock);
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001126 spin_lock(&bdi->wb.list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001127 inode->dirtied_when = jiffies;
Nick Piggin7ccf19a2010-10-21 11:49:30 +11001128 list_move(&inode->i_wb_list, &bdi->wb.b_dirty);
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001129 spin_unlock(&bdi->wb.list_lock);
Dave Chinnera66979a2011-03-22 22:23:41 +11001130
1131 if (wakeup_bdi)
1132 bdi_wakeup_thread_delayed(bdi);
1133 return;
Jens Axboe03ba3782009-09-09 09:08:54 +02001134 }
1135 }
Dave Chinner250df6e2011-03-22 22:23:36 +11001136out_unlock_inode:
1137 spin_unlock(&inode->i_lock);
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001138
Jens Axboe03ba3782009-09-09 09:08:54 +02001139}
1140EXPORT_SYMBOL(__mark_inode_dirty);
1141
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001142/*
1143 * Write out a superblock's list of dirty inodes. A wait will be performed
1144 * upon no inodes, all inodes or the final one, depending upon sync_mode.
1145 *
1146 * If older_than_this is non-NULL, then only write out inodes which
1147 * had their first dirtying at a time earlier than *older_than_this.
1148 *
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001149 * If `bdi' is non-zero then we're being asked to writeback a specific queue.
1150 * This function assumes that the blockdev superblock's inodes are backed by
1151 * a variety of queues, so all inodes are searched. For other superblocks,
1152 * assume that all inodes are backed by the same queue.
1153 *
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001154 * The inodes to be written are parked on bdi->b_io. They are moved back onto
1155 * bdi->b_dirty as they are selected for writing. This way, none can be missed
1156 * on the writer throttling path, and we get decent balancing between many
1157 * throttled threads: we don't want them all piling up on inode_sync_wait.
1158 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001159static void wait_sb_inodes(struct super_block *sb)
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001160{
Jens Axboe03ba3782009-09-09 09:08:54 +02001161 struct inode *inode, *old_inode = NULL;
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001162
Jens Axboe03ba3782009-09-09 09:08:54 +02001163 /*
1164 * We need to be protected against the filesystem going from
1165 * r/o to r/w or vice versa.
1166 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001167 WARN_ON(!rwsem_is_locked(&sb->s_umount));
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001168
Dave Chinner55fa6092011-03-22 22:23:40 +11001169 spin_lock(&inode_sb_list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001170
1171 /*
1172 * Data integrity sync. Must wait for all pages under writeback,
1173 * because there may have been pages dirtied before our sync
1174 * call, but which had writeout started before we write it out.
1175 * In which case, the inode may not be on the dirty list, but
1176 * we still have to wait for that writeout.
1177 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001178 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
Dave Chinner250df6e2011-03-22 22:23:36 +11001179 struct address_space *mapping = inode->i_mapping;
Jens Axboe03ba3782009-09-09 09:08:54 +02001180
Dave Chinner250df6e2011-03-22 22:23:36 +11001181 spin_lock(&inode->i_lock);
1182 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
1183 (mapping->nrpages == 0)) {
1184 spin_unlock(&inode->i_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001185 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +11001186 }
Jens Axboe03ba3782009-09-09 09:08:54 +02001187 __iget(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +11001188 spin_unlock(&inode->i_lock);
Dave Chinner55fa6092011-03-22 22:23:40 +11001189 spin_unlock(&inode_sb_list_lock);
1190
Jens Axboe03ba3782009-09-09 09:08:54 +02001191 /*
Dave Chinner55fa6092011-03-22 22:23:40 +11001192 * We hold a reference to 'inode' so it couldn't have been
1193 * removed from s_inodes list while we dropped the
1194 * inode_sb_list_lock. We cannot iput the inode now as we can
1195 * be holding the last reference and we cannot iput it under
1196 * inode_sb_list_lock. So we keep the reference and iput it
1197 * later.
Jens Axboe03ba3782009-09-09 09:08:54 +02001198 */
1199 iput(old_inode);
1200 old_inode = inode;
1201
1202 filemap_fdatawait(mapping);
1203
1204 cond_resched();
Nick Piggin38f21972009-01-06 14:40:25 -08001205
Dave Chinner55fa6092011-03-22 22:23:40 +11001206 spin_lock(&inode_sb_list_lock);
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001207 }
Dave Chinner55fa6092011-03-22 22:23:40 +11001208 spin_unlock(&inode_sb_list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001209 iput(old_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210}
1211
Jens Axboed8a85592009-09-02 12:34:32 +02001212/**
Chris Mason3259f8b2010-10-29 11:16:17 -04001213 * writeback_inodes_sb_nr - writeback dirty inodes from given super_block
Jens Axboed8a85592009-09-02 12:34:32 +02001214 * @sb: the superblock
Chris Mason3259f8b2010-10-29 11:16:17 -04001215 * @nr: the number of pages to write
Marcos Paulo de Souza786228a2011-11-23 20:56:45 +08001216 * @reason: reason why some writeback work initiated
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217 *
Jens Axboed8a85592009-09-02 12:34:32 +02001218 * Start writeback on some inodes on this super_block. No guarantees are made
1219 * on how many (if any) will be written, and this function does not wait
Chris Mason3259f8b2010-10-29 11:16:17 -04001220 * for IO completion of submitted IO.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001222void writeback_inodes_sb_nr(struct super_block *sb,
1223 unsigned long nr,
1224 enum wb_reason reason)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001226 DECLARE_COMPLETION_ONSTACK(done);
1227 struct wb_writeback_work work = {
Wu Fengguang6e6938b2010-06-06 10:38:15 -06001228 .sb = sb,
1229 .sync_mode = WB_SYNC_NONE,
1230 .tagged_writepages = 1,
1231 .done = &done,
1232 .nr_pages = nr,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001233 .reason = reason,
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001234 };
Jens Axboe0e3c9a22010-06-01 11:08:43 +02001235
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001236 WARN_ON(!rwsem_is_locked(&sb->s_umount));
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001237 bdi_queue_work(sb->s_bdi, &work);
1238 wait_for_completion(&done);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239}
Chris Mason3259f8b2010-10-29 11:16:17 -04001240EXPORT_SYMBOL(writeback_inodes_sb_nr);
1241
1242/**
1243 * writeback_inodes_sb - writeback dirty inodes from given super_block
1244 * @sb: the superblock
Marcos Paulo de Souza786228a2011-11-23 20:56:45 +08001245 * @reason: reason why some writeback work was initiated
Chris Mason3259f8b2010-10-29 11:16:17 -04001246 *
1247 * Start writeback on some inodes on this super_block. No guarantees are made
1248 * on how many (if any) will be written, and this function does not wait
1249 * for IO completion of submitted IO.
1250 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001251void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason)
Chris Mason3259f8b2010-10-29 11:16:17 -04001252{
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001253 return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason);
Chris Mason3259f8b2010-10-29 11:16:17 -04001254}
Jens Axboed8a85592009-09-02 12:34:32 +02001255EXPORT_SYMBOL(writeback_inodes_sb);
1256
1257/**
Eric Sandeen17bd55d2009-12-23 07:57:07 -05001258 * writeback_inodes_sb_if_idle - start writeback if none underway
1259 * @sb: the superblock
Marcos Paulo de Souza786228a2011-11-23 20:56:45 +08001260 * @reason: reason why some writeback work was initiated
Eric Sandeen17bd55d2009-12-23 07:57:07 -05001261 *
1262 * Invoke writeback_inodes_sb if no writeback is currently underway.
1263 * Returns 1 if writeback was started, 0 if not.
1264 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001265int writeback_inodes_sb_if_idle(struct super_block *sb, enum wb_reason reason)
Eric Sandeen17bd55d2009-12-23 07:57:07 -05001266{
1267 if (!writeback_in_progress(sb->s_bdi)) {
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001268 down_read(&sb->s_umount);
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001269 writeback_inodes_sb(sb, reason);
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001270 up_read(&sb->s_umount);
Eric Sandeen17bd55d2009-12-23 07:57:07 -05001271 return 1;
1272 } else
1273 return 0;
1274}
1275EXPORT_SYMBOL(writeback_inodes_sb_if_idle);
1276
1277/**
Chris Mason3259f8b2010-10-29 11:16:17 -04001278 * writeback_inodes_sb_if_idle - start writeback if none underway
1279 * @sb: the superblock
1280 * @nr: the number of pages to write
Marcos Paulo de Souza786228a2011-11-23 20:56:45 +08001281 * @reason: reason why some writeback work was initiated
Chris Mason3259f8b2010-10-29 11:16:17 -04001282 *
1283 * Invoke writeback_inodes_sb if no writeback is currently underway.
1284 * Returns 1 if writeback was started, 0 if not.
1285 */
1286int writeback_inodes_sb_nr_if_idle(struct super_block *sb,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001287 unsigned long nr,
1288 enum wb_reason reason)
Chris Mason3259f8b2010-10-29 11:16:17 -04001289{
1290 if (!writeback_in_progress(sb->s_bdi)) {
1291 down_read(&sb->s_umount);
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001292 writeback_inodes_sb_nr(sb, nr, reason);
Chris Mason3259f8b2010-10-29 11:16:17 -04001293 up_read(&sb->s_umount);
1294 return 1;
1295 } else
1296 return 0;
1297}
1298EXPORT_SYMBOL(writeback_inodes_sb_nr_if_idle);
1299
1300/**
Jens Axboed8a85592009-09-02 12:34:32 +02001301 * sync_inodes_sb - sync sb inode pages
1302 * @sb: the superblock
1303 *
1304 * This function writes and waits on any dirty inode belonging to this
Stefan Hajnoczicb9ef8d2011-01-13 15:47:26 -08001305 * super_block.
Jens Axboed8a85592009-09-02 12:34:32 +02001306 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001307void sync_inodes_sb(struct super_block *sb)
Jens Axboed8a85592009-09-02 12:34:32 +02001308{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001309 DECLARE_COMPLETION_ONSTACK(done);
1310 struct wb_writeback_work work = {
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001311 .sb = sb,
1312 .sync_mode = WB_SYNC_ALL,
1313 .nr_pages = LONG_MAX,
1314 .range_cyclic = 0,
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001315 .done = &done,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001316 .reason = WB_REASON_SYNC,
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001317 };
1318
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001319 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1320
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001321 bdi_queue_work(sb->s_bdi, &work);
1322 wait_for_completion(&done);
1323
Jens Axboeb6e51312009-09-16 15:13:54 +02001324 wait_sb_inodes(sb);
Jens Axboed8a85592009-09-02 12:34:32 +02001325}
1326EXPORT_SYMBOL(sync_inodes_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328/**
Andrea Arcangeli7f04c262005-10-30 15:03:05 -08001329 * write_inode_now - write an inode to disk
1330 * @inode: inode to write to disk
1331 * @sync: whether the write should be synchronous or not
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332 *
Andrea Arcangeli7f04c262005-10-30 15:03:05 -08001333 * This function commits an inode to disk immediately if it is dirty. This is
1334 * primarily needed by knfsd.
1335 *
1336 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338int write_inode_now(struct inode *inode, int sync)
1339{
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001340 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341 int ret;
1342 struct writeback_control wbc = {
1343 .nr_to_write = LONG_MAX,
Mike Galbraith18914b12008-02-08 04:20:23 -08001344 .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -07001345 .range_start = 0,
1346 .range_end = LLONG_MAX,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347 };
1348
1349 if (!mapping_cap_writeback_dirty(inode->i_mapping))
Andrew Morton49364ce2005-11-07 00:59:15 -08001350 wbc.nr_to_write = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351
1352 might_sleep();
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001353 spin_lock(&wb->list_lock);
Dave Chinner0f1b1fd2011-03-22 22:23:43 +11001354 spin_lock(&inode->i_lock);
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001355 ret = writeback_single_inode(inode, wb, &wbc);
Dave Chinner0f1b1fd2011-03-22 22:23:43 +11001356 spin_unlock(&inode->i_lock);
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001357 spin_unlock(&wb->list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358 if (sync)
Joern Engel1c0eeaf2007-10-16 23:30:44 -07001359 inode_sync_wait(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360 return ret;
1361}
1362EXPORT_SYMBOL(write_inode_now);
1363
1364/**
1365 * sync_inode - write an inode and its pages to disk.
1366 * @inode: the inode to sync
1367 * @wbc: controls the writeback mode
1368 *
1369 * sync_inode() will write an inode and its pages to disk. It will also
1370 * correctly update the inode on its superblock's dirty inode lists and will
1371 * update inode->i_state.
1372 *
1373 * The caller must have a ref on the inode.
1374 */
1375int sync_inode(struct inode *inode, struct writeback_control *wbc)
1376{
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001377 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378 int ret;
1379
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001380 spin_lock(&wb->list_lock);
Dave Chinner0f1b1fd2011-03-22 22:23:43 +11001381 spin_lock(&inode->i_lock);
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001382 ret = writeback_single_inode(inode, wb, wbc);
Dave Chinner0f1b1fd2011-03-22 22:23:43 +11001383 spin_unlock(&inode->i_lock);
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001384 spin_unlock(&wb->list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385 return ret;
1386}
1387EXPORT_SYMBOL(sync_inode);
Christoph Hellwigc3765012010-10-06 10:48:20 +02001388
1389/**
Andrew Mortonc691b9d2011-01-13 15:45:48 -08001390 * sync_inode_metadata - write an inode to disk
Christoph Hellwigc3765012010-10-06 10:48:20 +02001391 * @inode: the inode to sync
1392 * @wait: wait for I/O to complete.
1393 *
Andrew Mortonc691b9d2011-01-13 15:45:48 -08001394 * Write an inode to disk and adjust its dirty state after completion.
Christoph Hellwigc3765012010-10-06 10:48:20 +02001395 *
1396 * Note: only writes the actual inode, no associated data or other metadata.
1397 */
1398int sync_inode_metadata(struct inode *inode, int wait)
1399{
1400 struct writeback_control wbc = {
1401 .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE,
1402 .nr_to_write = 0, /* metadata-only */
1403 };
1404
1405 return sync_inode(inode, &wbc);
1406}
1407EXPORT_SYMBOL(sync_inode_metadata);