blob: 0f015a0468de53e7c4b27a9a5efdfeb9fa4cd67b [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;
38 enum writeback_sync_modes sync_mode;
H Hartley Sweeten52957fe2010-04-01 20:36:30 -050039 unsigned int for_kupdate:1;
40 unsigned int range_cyclic:1;
41 unsigned int for_background:1;
Jens Axboec4a77a62009-09-16 15:18:25 +020042
Jens Axboe8010c3b2009-09-15 20:04:57 +020043 struct list_head list; /* pending work list */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +020044 struct completion *done; /* set if the caller waits */
Jens Axboe03ba3782009-09-09 09:08:54 +020045};
46
Dave Chinner455b2862010-07-07 13:24:06 +100047/*
48 * Include the creation of the trace points after defining the
49 * wb_writeback_work structure so that the definition remains local to this
50 * file.
51 */
52#define CREATE_TRACE_POINTS
53#include <trace/events/writeback.h>
54
Dave Chinner455b2862010-07-07 13:24:06 +100055/*
56 * We don't actually have pdflush, but this one is exported though /proc...
57 */
58int nr_pdflush_threads;
59
Adrian Bunkf11b00f2008-04-29 00:58:56 -070060/**
61 * writeback_in_progress - determine whether there is writeback in progress
62 * @bdi: the device's backing_dev_info structure.
63 *
Jens Axboe03ba3782009-09-09 09:08:54 +020064 * Determine whether there is writeback waiting to be handled against a
65 * backing device.
Adrian Bunkf11b00f2008-04-29 00:58:56 -070066 */
67int writeback_in_progress(struct backing_dev_info *bdi)
68{
Jan Kara81d73a32010-08-11 14:17:44 -070069 return test_bit(BDI_writeback_running, &bdi->state);
Adrian Bunkf11b00f2008-04-29 00:58:56 -070070}
71
Jan Kara692ebd12010-09-21 11:51:01 +020072static inline struct backing_dev_info *inode_to_bdi(struct inode *inode)
73{
74 struct super_block *sb = inode->i_sb;
Jan Kara692ebd12010-09-21 11:51:01 +020075
Christoph Hellwigaaead252010-10-04 14:25:33 +020076 if (strcmp(sb->s_type->name, "bdev") == 0)
77 return inode->i_mapping->backing_dev_info;
78
79 return sb->s_bdi;
Jan Kara692ebd12010-09-21 11:51:01 +020080}
81
Nick Piggin7ccf19a2010-10-21 11:49:30 +110082static inline struct inode *wb_inode(struct list_head *head)
83{
84 return list_entry(head, struct inode, i_wb_list);
85}
86
Jan Kara65850272011-01-13 15:45:44 -080087/* Wakeup flusher thread or forker thread to fork it. Requires bdi->wb_lock. */
88static void bdi_wakeup_flusher(struct backing_dev_info *bdi)
Nick Piggin4195f732009-05-28 09:01:15 +020089{
Artem Bityutskiyfff5b852010-07-25 14:29:20 +030090 if (bdi->wb.task) {
91 wake_up_process(bdi->wb.task);
92 } else {
93 /*
94 * The bdi thread isn't there, wake up the forker thread which
95 * will create and run it.
96 */
Jens Axboe03ba3782009-09-09 09:08:54 +020097 wake_up_process(default_backing_dev_info.wb.task);
Linus Torvalds1da177e2005-04-16 15:20:36 -070098 }
Jan Kara65850272011-01-13 15:45:44 -080099}
100
101static void bdi_queue_work(struct backing_dev_info *bdi,
102 struct wb_writeback_work *work)
103{
104 trace_writeback_queue(bdi, work);
105
106 spin_lock_bh(&bdi->wb_lock);
107 list_add_tail(&work->list, &bdi->work_list);
108 if (!bdi->wb.task)
109 trace_writeback_nothread(bdi, work);
110 bdi_wakeup_flusher(bdi);
Artem Bityutskiy64677162010-07-25 14:29:22 +0300111 spin_unlock_bh(&bdi->wb_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112}
113
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200114static void
115__bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages,
Jan Kara65850272011-01-13 15:45:44 -0800116 bool range_cyclic)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200118 struct wb_writeback_work *work;
Jens Axboe03ba3782009-09-09 09:08:54 +0200119
Jens Axboebcddc3f2009-09-13 20:07:36 +0200120 /*
121 * This is WB_SYNC_NONE writeback, so if allocation fails just
122 * wakeup the thread for old dirty data writeback
123 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200124 work = kzalloc(sizeof(*work), GFP_ATOMIC);
125 if (!work) {
Dave Chinner455b2862010-07-07 13:24:06 +1000126 if (bdi->wb.task) {
127 trace_writeback_nowork(bdi);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200128 wake_up_process(bdi->wb.task);
Dave Chinner455b2862010-07-07 13:24:06 +1000129 }
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200130 return;
Jens Axboebcddc3f2009-09-13 20:07:36 +0200131 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200132
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200133 work->sync_mode = WB_SYNC_NONE;
134 work->nr_pages = nr_pages;
135 work->range_cyclic = range_cyclic;
Christoph Hellwigf0fad8a2009-09-11 09:47:56 +0200136
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200137 bdi_queue_work(bdi, work);
Jens Axboeb6e51312009-09-16 15:13:54 +0200138}
139
140/**
141 * bdi_start_writeback - start writeback
142 * @bdi: the backing device to write from
143 * @nr_pages: the number of pages to write
144 *
145 * Description:
146 * This does WB_SYNC_NONE opportunistic writeback. The IO is only
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300147 * started when this function returns, we make no guarantees on
Jens Axboe0e3c9a22010-06-01 11:08:43 +0200148 * completion. Caller need not hold sb s_umount semaphore.
Jens Axboeb6e51312009-09-16 15:13:54 +0200149 *
150 */
Christoph Hellwigc5444192010-06-08 18:15:15 +0200151void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages)
Jens Axboeb6e51312009-09-16 15:13:54 +0200152{
Jan Kara65850272011-01-13 15:45:44 -0800153 __bdi_start_writeback(bdi, nr_pages, true);
Christoph Hellwigc5444192010-06-08 18:15:15 +0200154}
Wu Fengguangd3ddec72009-09-23 20:33:40 +0800155
Christoph Hellwigc5444192010-06-08 18:15:15 +0200156/**
157 * bdi_start_background_writeback - start background writeback
158 * @bdi: the backing device to write from
159 *
160 * Description:
Jan Kara65850272011-01-13 15:45:44 -0800161 * This makes sure WB_SYNC_NONE background writeback happens. When
162 * this function returns, it is only guaranteed that for given BDI
163 * some IO is happening if we are over background dirty threshold.
164 * Caller need not hold sb s_umount semaphore.
Christoph Hellwigc5444192010-06-08 18:15:15 +0200165 */
166void bdi_start_background_writeback(struct backing_dev_info *bdi)
167{
Jan Kara65850272011-01-13 15:45:44 -0800168 /*
169 * We just wake up the flusher thread. It will perform background
170 * writeback as soon as there is no other work to do.
171 */
Wu Fengguang71927e82011-01-13 15:45:46 -0800172 trace_writeback_wake_background(bdi);
Jan Kara65850272011-01-13 15:45:44 -0800173 spin_lock_bh(&bdi->wb_lock);
174 bdi_wakeup_flusher(bdi);
175 spin_unlock_bh(&bdi->wb_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176}
177
178/*
Dave Chinnera66979a2011-03-22 22:23:41 +1100179 * Remove the inode from the writeback list it is on.
180 */
181void inode_wb_list_del(struct inode *inode)
182{
183 spin_lock(&inode_wb_list_lock);
184 list_del_init(&inode->i_wb_list);
185 spin_unlock(&inode_wb_list_lock);
186}
187
188
189/*
Andrew Morton6610a0b2007-10-16 23:30:32 -0700190 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
191 * furthest end of its superblock's dirty-inode list.
192 *
193 * Before stamping the inode's ->dirtied_when, we check to see whether it is
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200194 * already the most-recently-dirtied inode on the b_dirty list. If that is
Andrew Morton6610a0b2007-10-16 23:30:32 -0700195 * the case then the inode must have been redirtied while it was being written
196 * out and we don't reset its dirtied_when.
197 */
198static void redirty_tail(struct inode *inode)
199{
Jens Axboe03ba3782009-09-09 09:08:54 +0200200 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
Andrew Morton6610a0b2007-10-16 23:30:32 -0700201
Dave Chinnera66979a2011-03-22 22:23:41 +1100202 assert_spin_locked(&inode_wb_list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200203 if (!list_empty(&wb->b_dirty)) {
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200204 struct inode *tail;
Andrew Morton6610a0b2007-10-16 23:30:32 -0700205
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100206 tail = wb_inode(wb->b_dirty.next);
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200207 if (time_before(inode->dirtied_when, tail->dirtied_when))
Andrew Morton6610a0b2007-10-16 23:30:32 -0700208 inode->dirtied_when = jiffies;
209 }
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100210 list_move(&inode->i_wb_list, &wb->b_dirty);
Andrew Morton6610a0b2007-10-16 23:30:32 -0700211}
212
213/*
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200214 * requeue inode for re-scanning after bdi->b_io list is exhausted.
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700215 */
Ken Chen0e0f4fc2007-10-16 23:30:38 -0700216static void requeue_io(struct inode *inode)
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700217{
Jens Axboe03ba3782009-09-09 09:08:54 +0200218 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
219
Dave Chinnera66979a2011-03-22 22:23:41 +1100220 assert_spin_locked(&inode_wb_list_lock);
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100221 list_move(&inode->i_wb_list, &wb->b_more_io);
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700222}
223
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700224static void inode_sync_complete(struct inode *inode)
225{
226 /*
Dave Chinnera66979a2011-03-22 22:23:41 +1100227 * Prevent speculative execution through
228 * spin_unlock(&inode_wb_list_lock);
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700229 */
Dave Chinnera66979a2011-03-22 22:23:41 +1100230
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700231 smp_mb();
232 wake_up_bit(&inode->i_state, __I_SYNC);
233}
234
Jeff Laytond2caa3c52009-04-02 16:56:37 -0700235static bool inode_dirtied_after(struct inode *inode, unsigned long t)
236{
237 bool ret = time_after(inode->dirtied_when, t);
238#ifndef CONFIG_64BIT
239 /*
240 * For inodes being constantly redirtied, dirtied_when can get stuck.
241 * It _appears_ to be in the future, but is actually in distant past.
242 * This test is necessary to prevent such wrapped-around relative times
Jens Axboe5b0830c2009-09-23 19:37:09 +0200243 * from permanently stopping the whole bdi writeback.
Jeff Laytond2caa3c52009-04-02 16:56:37 -0700244 */
245 ret = ret && time_before_eq(inode->dirtied_when, jiffies);
246#endif
247 return ret;
248}
249
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700250/*
Fengguang Wu2c136572007-10-16 23:30:39 -0700251 * Move expired dirty inodes from @delaying_queue to @dispatch_queue.
252 */
253static void move_expired_inodes(struct list_head *delaying_queue,
254 struct list_head *dispatch_queue,
255 unsigned long *older_than_this)
256{
Shaohua Li5c034492009-09-24 14:42:33 +0200257 LIST_HEAD(tmp);
258 struct list_head *pos, *node;
Jens Axboecf137302009-09-24 15:12:57 +0200259 struct super_block *sb = NULL;
Shaohua Li5c034492009-09-24 14:42:33 +0200260 struct inode *inode;
Jens Axboecf137302009-09-24 15:12:57 +0200261 int do_sb_sort = 0;
Shaohua Li5c034492009-09-24 14:42:33 +0200262
Fengguang Wu2c136572007-10-16 23:30:39 -0700263 while (!list_empty(delaying_queue)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100264 inode = wb_inode(delaying_queue->prev);
Fengguang Wu2c136572007-10-16 23:30:39 -0700265 if (older_than_this &&
Jeff Laytond2caa3c52009-04-02 16:56:37 -0700266 inode_dirtied_after(inode, *older_than_this))
Fengguang Wu2c136572007-10-16 23:30:39 -0700267 break;
Jens Axboecf137302009-09-24 15:12:57 +0200268 if (sb && sb != inode->i_sb)
269 do_sb_sort = 1;
270 sb = inode->i_sb;
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100271 list_move(&inode->i_wb_list, &tmp);
Shaohua Li5c034492009-09-24 14:42:33 +0200272 }
273
Jens Axboecf137302009-09-24 15:12:57 +0200274 /* just one sb in list, splice to dispatch_queue and we're done */
275 if (!do_sb_sort) {
276 list_splice(&tmp, dispatch_queue);
277 return;
278 }
279
Shaohua Li5c034492009-09-24 14:42:33 +0200280 /* Move inodes from one superblock together */
281 while (!list_empty(&tmp)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100282 sb = wb_inode(tmp.prev)->i_sb;
Shaohua Li5c034492009-09-24 14:42:33 +0200283 list_for_each_prev_safe(pos, node, &tmp) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100284 inode = wb_inode(pos);
Shaohua Li5c034492009-09-24 14:42:33 +0200285 if (inode->i_sb == sb)
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100286 list_move(&inode->i_wb_list, dispatch_queue);
Shaohua Li5c034492009-09-24 14:42:33 +0200287 }
Fengguang Wu2c136572007-10-16 23:30:39 -0700288 }
289}
290
291/*
292 * Queue all expired dirty inodes for io, eldest first.
Wu Fengguang4ea879b2010-08-11 14:17:42 -0700293 * Before
294 * newly dirtied b_dirty b_io b_more_io
295 * =============> gf edc BA
296 * After
297 * newly dirtied b_dirty b_io b_more_io
298 * =============> g fBAedc
299 * |
300 * +--> dequeue for IO
Fengguang Wu2c136572007-10-16 23:30:39 -0700301 */
Jens Axboe03ba3782009-09-09 09:08:54 +0200302static void queue_io(struct bdi_writeback *wb, unsigned long *older_than_this)
Fengguang Wu2c136572007-10-16 23:30:39 -0700303{
Dave Chinnera66979a2011-03-22 22:23:41 +1100304 assert_spin_locked(&inode_wb_list_lock);
Wu Fengguang4ea879b2010-08-11 14:17:42 -0700305 list_splice_init(&wb->b_more_io, &wb->b_io);
Jens Axboe03ba3782009-09-09 09:08:54 +0200306 move_expired_inodes(&wb->b_dirty, &wb->b_io, older_than_this);
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200307}
308
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100309static int write_inode(struct inode *inode, struct writeback_control *wbc)
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200310{
Jens Axboe03ba3782009-09-09 09:08:54 +0200311 if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode))
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100312 return inode->i_sb->s_op->write_inode(inode, wbc);
Jens Axboe03ba3782009-09-09 09:08:54 +0200313 return 0;
Fengguang Wu2c136572007-10-16 23:30:39 -0700314}
315
316/*
Christoph Hellwig01c03192009-06-08 13:35:40 +0200317 * Wait for writeback on an inode to complete.
318 */
319static void inode_wait_for_writeback(struct inode *inode)
320{
321 DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC);
322 wait_queue_head_t *wqh;
323
324 wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
Dave Chinner250df6e2011-03-22 22:23:36 +1100325 while (inode->i_state & I_SYNC) {
326 spin_unlock(&inode->i_lock);
Dave Chinnera66979a2011-03-22 22:23:41 +1100327 spin_unlock(&inode_wb_list_lock);
Christoph Hellwig01c03192009-06-08 13:35:40 +0200328 __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE);
Dave Chinnera66979a2011-03-22 22:23:41 +1100329 spin_lock(&inode_wb_list_lock);
Dave Chinner250df6e2011-03-22 22:23:36 +1100330 spin_lock(&inode->i_lock);
Richard Kennedy58a9d3d82010-05-24 14:32:38 -0700331 }
Christoph Hellwig01c03192009-06-08 13:35:40 +0200332}
333
334/*
Dave Chinner0f1b1fd2011-03-22 22:23:43 +1100335 * Write out an inode's dirty pages. Called under inode_wb_list_lock and
336 * inode->i_lock. Either the caller has an active reference on the inode or
337 * the inode has I_WILL_FREE set.
Christoph Hellwig01c03192009-06-08 13:35:40 +0200338 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339 * If `wait' is set, wait on the writeout.
340 *
341 * The whole writeout design is quite complex and fragile. We want to avoid
342 * starvation of particular inodes when others are being redirtied, prevent
343 * livelocks, etc.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344 */
345static int
Christoph Hellwig01c03192009-06-08 13:35:40 +0200346writeback_single_inode(struct inode *inode, struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348 struct address_space *mapping = inode->i_mapping;
Christoph Hellwig01c03192009-06-08 13:35:40 +0200349 unsigned dirty;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350 int ret;
351
Dave Chinner0f1b1fd2011-03-22 22:23:43 +1100352 assert_spin_locked(&inode_wb_list_lock);
353 assert_spin_locked(&inode->i_lock);
354
Christoph Hellwig01c03192009-06-08 13:35:40 +0200355 if (!atomic_read(&inode->i_count))
356 WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING)));
357 else
358 WARN_ON(inode->i_state & I_WILL_FREE);
359
360 if (inode->i_state & I_SYNC) {
361 /*
362 * If this inode is locked for writeback and we are not doing
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200363 * writeback-for-data-integrity, move it to b_more_io so that
Christoph Hellwig01c03192009-06-08 13:35:40 +0200364 * writeback can proceed with the other inodes on s_io.
365 *
366 * We'll have another go at writing back this inode when we
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200367 * completed a full scan of b_io.
Christoph Hellwig01c03192009-06-08 13:35:40 +0200368 */
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100369 if (wbc->sync_mode != WB_SYNC_ALL) {
Christoph Hellwig01c03192009-06-08 13:35:40 +0200370 requeue_io(inode);
371 return 0;
372 }
373
374 /*
375 * It's a data-integrity sync. We must wait.
376 */
377 inode_wait_for_writeback(inode);
378 }
379
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700380 BUG_ON(inode->i_state & I_SYNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381
Dmitry Monakhov5547e8a2010-05-07 13:35:44 +0400382 /* Set I_SYNC, reset I_DIRTY_PAGES */
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700383 inode->i_state |= I_SYNC;
Dmitry Monakhov5547e8a2010-05-07 13:35:44 +0400384 inode->i_state &= ~I_DIRTY_PAGES;
Dave Chinner250df6e2011-03-22 22:23:36 +1100385 spin_unlock(&inode->i_lock);
Dave Chinnera66979a2011-03-22 22:23:41 +1100386 spin_unlock(&inode_wb_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387
388 ret = do_writepages(mapping, wbc);
389
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100390 /*
391 * Make sure to wait on the data before writing out the metadata.
392 * This is important for filesystems that modify metadata on data
393 * I/O completion.
394 */
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100395 if (wbc->sync_mode == WB_SYNC_ALL) {
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100396 int err = filemap_fdatawait(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397 if (ret == 0)
398 ret = err;
399 }
400
Dmitry Monakhov5547e8a2010-05-07 13:35:44 +0400401 /*
402 * Some filesystems may redirty the inode during the writeback
403 * due to delalloc, clear dirty metadata flags right before
404 * write_inode()
405 */
Dave Chinner250df6e2011-03-22 22:23:36 +1100406 spin_lock(&inode->i_lock);
Dmitry Monakhov5547e8a2010-05-07 13:35:44 +0400407 dirty = inode->i_state & I_DIRTY;
408 inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC);
Dave Chinner250df6e2011-03-22 22:23:36 +1100409 spin_unlock(&inode->i_lock);
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100410 /* Don't write the inode if only I_DIRTY_PAGES was set */
411 if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100412 int err = write_inode(inode, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413 if (ret == 0)
414 ret = err;
415 }
416
Dave Chinnera66979a2011-03-22 22:23:41 +1100417 spin_lock(&inode_wb_list_lock);
Dave Chinner250df6e2011-03-22 22:23:36 +1100418 spin_lock(&inode->i_lock);
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700419 inode->i_state &= ~I_SYNC;
Al Viroa4ffdde2010-06-02 17:38:30 -0400420 if (!(inode->i_state & I_FREEING)) {
Wu Fengguang23539af2010-08-11 14:17:41 -0700421 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422 /*
423 * We didn't write back all the pages. nfs_writepages()
Wu Fengguanga50aeb42010-08-11 14:17:43 -0700424 * sometimes bales out without doing anything.
Andrew Morton1b43ef92007-10-16 23:30:35 -0700425 */
Wu Fengguanga50aeb42010-08-11 14:17:43 -0700426 inode->i_state |= I_DIRTY_PAGES;
427 if (wbc->nr_to_write <= 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428 /*
Wu Fengguanga50aeb42010-08-11 14:17:43 -0700429 * slice used up: queue for next turn
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430 */
Wu Fengguanga50aeb42010-08-11 14:17:43 -0700431 requeue_io(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432 } else {
433 /*
Wu Fengguanga50aeb42010-08-11 14:17:43 -0700434 * Writeback blocked by something other than
435 * congestion. Delay the inode for some time to
436 * avoid spinning on the CPU (100% iowait)
437 * retrying writeback of the dirty page/inode
438 * that cannot be performed immediately.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439 */
Andrew Morton1b43ef92007-10-16 23:30:35 -0700440 redirty_tail(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 }
Wu Fengguang23539af2010-08-11 14:17:41 -0700442 } else if (inode->i_state & I_DIRTY) {
443 /*
444 * Filesystems can dirty the inode during writeback
445 * operations, such as delayed allocation during
446 * submission or metadata updates after data IO
447 * completion.
448 */
449 redirty_tail(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450 } else {
451 /*
Nick Piggin9e38d862010-10-23 06:55:17 -0400452 * The inode is clean. At this point we either have
453 * a reference to the inode or it's on it's way out.
454 * No need to add it back to the LRU.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455 */
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100456 list_del_init(&inode->i_wb_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457 }
458 }
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700459 inode_sync_complete(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460 return ret;
461}
462
Jens Axboe03ba3782009-09-09 09:08:54 +0200463/*
Christoph Hellwigd19de7e2010-06-08 18:14:58 +0200464 * For background writeback the caller does not have the sb pinned
Jens Axboe03ba3782009-09-09 09:08:54 +0200465 * before calling writeback. So make sure that we do pin it, so it doesn't
466 * go away while we are writing inodes from it.
Jens Axboe03ba3782009-09-09 09:08:54 +0200467 */
Christoph Hellwigd19de7e2010-06-08 18:14:58 +0200468static bool pin_sb_for_writeback(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469{
Jens Axboe03ba3782009-09-09 09:08:54 +0200470 spin_lock(&sb_lock);
Christoph Hellwig29cb48592010-06-09 15:31:01 +0200471 if (list_empty(&sb->s_instances)) {
472 spin_unlock(&sb_lock);
473 return false;
474 }
475
Jens Axboe03ba3782009-09-09 09:08:54 +0200476 sb->s_count++;
Christoph Hellwig29cb48592010-06-09 15:31:01 +0200477 spin_unlock(&sb_lock);
478
Jens Axboe03ba3782009-09-09 09:08:54 +0200479 if (down_read_trylock(&sb->s_umount)) {
Christoph Hellwig29cb48592010-06-09 15:31:01 +0200480 if (sb->s_root)
Christoph Hellwigd19de7e2010-06-08 18:14:58 +0200481 return true;
Jens Axboe03ba3782009-09-09 09:08:54 +0200482 up_read(&sb->s_umount);
483 }
Christoph Hellwig29cb48592010-06-09 15:31:01 +0200484
485 put_super(sb);
Christoph Hellwigd19de7e2010-06-08 18:14:58 +0200486 return false;
Jens Axboe03ba3782009-09-09 09:08:54 +0200487}
488
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800489/*
490 * Write a portion of b_io inodes which belong to @sb.
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200491 *
492 * If @only_this_sb is true, then find and write all such
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800493 * inodes. Otherwise write only ones which go sequentially
494 * in reverse order.
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200495 *
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800496 * Return 1, if the caller writeback routine should be
497 * interrupted. Otherwise return 0.
498 */
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200499static int writeback_sb_inodes(struct super_block *sb, struct bdi_writeback *wb,
500 struct writeback_control *wbc, bool only_this_sb)
Jens Axboe03ba3782009-09-09 09:08:54 +0200501{
Jens Axboe03ba3782009-09-09 09:08:54 +0200502 while (!list_empty(&wb->b_io)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503 long pages_skipped;
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100504 struct inode *inode = wb_inode(wb->b_io.prev);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200505
506 if (inode->i_sb != sb) {
507 if (only_this_sb) {
508 /*
509 * We only want to write back data for this
510 * superblock, move all inodes not belonging
511 * to it back onto the dirty list.
512 */
513 redirty_tail(inode);
514 continue;
515 }
516
517 /*
518 * The inode belongs to a different superblock.
519 * Bounce back to the caller to unpin this and
520 * pin the next superblock.
521 */
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800522 return 0;
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200523 }
524
Christoph Hellwig9843b76a2010-10-24 19:40:46 +0200525 /*
526 * Don't bother with new inodes or inodes beeing freed, first
527 * kind does not need peridic writeout yet, and for the latter
528 * kind writeout is handled by the freer.
529 */
Dave Chinner250df6e2011-03-22 22:23:36 +1100530 spin_lock(&inode->i_lock);
Christoph Hellwig9843b76a2010-10-24 19:40:46 +0200531 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
Dave Chinner250df6e2011-03-22 22:23:36 +1100532 spin_unlock(&inode->i_lock);
Nick Piggin7ef0d732009-03-12 14:31:38 -0700533 requeue_io(inode);
534 continue;
535 }
Christoph Hellwig9843b76a2010-10-24 19:40:46 +0200536
Jeff Laytond2caa3c52009-04-02 16:56:37 -0700537 /*
538 * Was this inode dirtied after sync_sb_inodes was called?
539 * This keeps sync from extra jobs and livelock.
540 */
Dave Chinner250df6e2011-03-22 22:23:36 +1100541 if (inode_dirtied_after(inode, wbc->wb_start)) {
542 spin_unlock(&inode->i_lock);
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800543 return 1;
Dave Chinner250df6e2011-03-22 22:23:36 +1100544 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546 __iget(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +1100547
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548 pages_skipped = wbc->pages_skipped;
Christoph Hellwig01c03192009-06-08 13:35:40 +0200549 writeback_single_inode(inode, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550 if (wbc->pages_skipped != pages_skipped) {
551 /*
552 * writeback is not making progress due to locked
553 * buffers. Skip this inode for now.
554 */
Andrew Mortonf57b9b72007-10-16 23:30:34 -0700555 redirty_tail(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556 }
Dave Chinner0f1b1fd2011-03-22 22:23:43 +1100557 spin_unlock(&inode->i_lock);
Dave Chinnera66979a2011-03-22 22:23:41 +1100558 spin_unlock(&inode_wb_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559 iput(inode);
OGAWA Hirofumi4ffc8442006-03-25 03:07:44 -0800560 cond_resched();
Dave Chinnera66979a2011-03-22 22:23:41 +1100561 spin_lock(&inode_wb_list_lock);
Fengguang Wu8bc3be22008-02-04 22:29:36 -0800562 if (wbc->nr_to_write <= 0) {
563 wbc->more_io = 1;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800564 return 1;
Fengguang Wu8bc3be22008-02-04 22:29:36 -0800565 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200566 if (!list_empty(&wb->b_more_io))
Fengguang Wu8bc3be22008-02-04 22:29:36 -0800567 wbc->more_io = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568 }
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800569 /* b_io is empty */
570 return 1;
571}
Nick Piggin38f21972009-01-06 14:40:25 -0800572
Christoph Hellwig9c3a8ee2010-06-10 12:07:27 +0200573void writeback_inodes_wb(struct bdi_writeback *wb,
574 struct writeback_control *wbc)
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800575{
576 int ret = 0;
Jens Axboe9ecc2732009-09-24 15:25:11 +0200577
Jan Kara7624ee722010-08-09 17:20:03 -0700578 if (!wbc->wb_start)
579 wbc->wb_start = jiffies; /* livelock avoidance */
Dave Chinnera66979a2011-03-22 22:23:41 +1100580 spin_lock(&inode_wb_list_lock);
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800581 if (!wbc->for_kupdate || list_empty(&wb->b_io))
582 queue_io(wb, wbc->older_than_this);
583
584 while (!list_empty(&wb->b_io)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100585 struct inode *inode = wb_inode(wb->b_io.prev);
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800586 struct super_block *sb = inode->i_sb;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800587
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200588 if (!pin_sb_for_writeback(sb)) {
589 requeue_io(inode);
590 continue;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800591 }
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200592 ret = writeback_sb_inodes(sb, wb, wbc, false);
593 drop_super(sb);
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800594
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800595 if (ret)
596 break;
597 }
Dave Chinnera66979a2011-03-22 22:23:41 +1100598 spin_unlock(&inode_wb_list_lock);
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200599 /* Leave any unwritten inodes on b_io */
600}
601
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200602static void __writeback_inodes_sb(struct super_block *sb,
603 struct bdi_writeback *wb, struct writeback_control *wbc)
604{
605 WARN_ON(!rwsem_is_locked(&sb->s_umount));
606
Dave Chinnera66979a2011-03-22 22:23:41 +1100607 spin_lock(&inode_wb_list_lock);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200608 if (!wbc->for_kupdate || list_empty(&wb->b_io))
609 queue_io(wb, wbc->older_than_this);
610 writeback_sb_inodes(sb, wb, wbc, true);
Dave Chinnera66979a2011-03-22 22:23:41 +1100611 spin_unlock(&inode_wb_list_lock);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200612}
613
Jens Axboe03ba3782009-09-09 09:08:54 +0200614/*
615 * The maximum number of pages to writeout in a single bdi flush/kupdate
616 * operation. We do this so we don't hold I_SYNC against an inode for
617 * enormous amounts of time, which would block a userspace task which has
618 * been forced to throttle against that inode. Also, the code reevaluates
619 * the dirty each time it has written this many pages.
620 */
621#define MAX_WRITEBACK_PAGES 1024
622
623static inline bool over_bground_thresh(void)
624{
625 unsigned long background_thresh, dirty_thresh;
626
Wu Fengguang16c40422010-08-11 14:17:39 -0700627 global_dirty_limits(&background_thresh, &dirty_thresh);
Jens Axboe03ba3782009-09-09 09:08:54 +0200628
629 return (global_page_state(NR_FILE_DIRTY) +
Wu Fengguang4cbec4c2010-10-26 14:21:45 -0700630 global_page_state(NR_UNSTABLE_NFS) > background_thresh);
Jens Axboe03ba3782009-09-09 09:08:54 +0200631}
632
633/*
634 * Explicit flushing or periodic writeback of "old" data.
635 *
636 * Define "old": the first time one of an inode's pages is dirtied, we mark the
637 * dirtying-time in the inode's address_space. So this periodic writeback code
638 * just walks the superblock inode list, writing back any inodes which are
639 * older than a specific point in time.
640 *
641 * Try to run once per dirty_writeback_interval. But if a writeback event
642 * takes longer than a dirty_writeback_interval interval, then leave a
643 * one-second gap.
644 *
645 * older_than_this takes precedence over nr_to_write. So we'll only write back
646 * all dirty pages if they are all attached to "old" mappings.
647 */
Jens Axboec4a77a62009-09-16 15:18:25 +0200648static long wb_writeback(struct bdi_writeback *wb,
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200649 struct wb_writeback_work *work)
Jens Axboe03ba3782009-09-09 09:08:54 +0200650{
651 struct writeback_control wbc = {
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200652 .sync_mode = work->sync_mode,
Jens Axboe03ba3782009-09-09 09:08:54 +0200653 .older_than_this = NULL,
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200654 .for_kupdate = work->for_kupdate,
655 .for_background = work->for_background,
656 .range_cyclic = work->range_cyclic,
Jens Axboe03ba3782009-09-09 09:08:54 +0200657 };
658 unsigned long oldest_jif;
659 long wrote = 0;
Jan Karab9543da2011-01-13 15:45:48 -0800660 long write_chunk;
Jan Karaa5989bd2009-09-16 19:22:48 +0200661 struct inode *inode;
Jens Axboe03ba3782009-09-09 09:08:54 +0200662
663 if (wbc.for_kupdate) {
664 wbc.older_than_this = &oldest_jif;
665 oldest_jif = jiffies -
666 msecs_to_jiffies(dirty_expire_interval * 10);
667 }
Jens Axboec4a77a62009-09-16 15:18:25 +0200668 if (!wbc.range_cyclic) {
669 wbc.range_start = 0;
670 wbc.range_end = LLONG_MAX;
671 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200672
Jan Karab9543da2011-01-13 15:45:48 -0800673 /*
674 * WB_SYNC_ALL mode does livelock avoidance by syncing dirty
675 * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX
676 * here avoids calling into writeback_inodes_wb() more than once.
677 *
678 * The intended call sequence for WB_SYNC_ALL writeback is:
679 *
680 * wb_writeback()
681 * __writeback_inodes_sb() <== called only once
682 * write_cache_pages() <== called once for each inode
683 * (quickly) tag currently dirty pages
684 * (maybe slowly) sync all tagged pages
685 */
686 if (wbc.sync_mode == WB_SYNC_NONE)
687 write_chunk = MAX_WRITEBACK_PAGES;
688 else
689 write_chunk = LONG_MAX;
690
Jan Kara7624ee722010-08-09 17:20:03 -0700691 wbc.wb_start = jiffies; /* livelock avoidance */
Jens Axboe03ba3782009-09-09 09:08:54 +0200692 for (;;) {
693 /*
Wu Fengguangd3ddec72009-09-23 20:33:40 +0800694 * Stop writeback when nr_pages has been consumed
Jens Axboe03ba3782009-09-09 09:08:54 +0200695 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200696 if (work->nr_pages <= 0)
Jens Axboe03ba3782009-09-09 09:08:54 +0200697 break;
698
699 /*
Jan Karaaa373cf2011-01-13 15:45:47 -0800700 * Background writeout and kupdate-style writeback may
701 * run forever. Stop them if there is other work to do
702 * so that e.g. sync can proceed. They'll be restarted
703 * after the other works are all done.
704 */
705 if ((work->for_background || work->for_kupdate) &&
706 !list_empty(&wb->bdi->work_list))
707 break;
708
709 /*
Wu Fengguangd3ddec72009-09-23 20:33:40 +0800710 * For background writeout, stop when we are below the
711 * background dirty threshold
Jens Axboe03ba3782009-09-09 09:08:54 +0200712 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200713 if (work->for_background && !over_bground_thresh())
Jens Axboe03ba3782009-09-09 09:08:54 +0200714 break;
715
716 wbc.more_io = 0;
Jan Karab9543da2011-01-13 15:45:48 -0800717 wbc.nr_to_write = write_chunk;
Jens Axboe03ba3782009-09-09 09:08:54 +0200718 wbc.pages_skipped = 0;
Dave Chinner028c2dd2010-07-07 13:24:07 +1000719
720 trace_wbc_writeback_start(&wbc, wb->bdi);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200721 if (work->sb)
722 __writeback_inodes_sb(work->sb, wb, &wbc);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200723 else
724 writeback_inodes_wb(wb, &wbc);
Dave Chinner028c2dd2010-07-07 13:24:07 +1000725 trace_wbc_writeback_written(&wbc, wb->bdi);
726
Jan Karab9543da2011-01-13 15:45:48 -0800727 work->nr_pages -= write_chunk - wbc.nr_to_write;
728 wrote += write_chunk - wbc.nr_to_write;
Jens Axboe03ba3782009-09-09 09:08:54 +0200729
730 /*
Jens Axboe71fd05a2009-09-23 19:32:26 +0200731 * If we consumed everything, see if we have more
Jens Axboe03ba3782009-09-09 09:08:54 +0200732 */
Jens Axboe71fd05a2009-09-23 19:32:26 +0200733 if (wbc.nr_to_write <= 0)
734 continue;
735 /*
736 * Didn't write everything and we don't have more IO, bail
737 */
738 if (!wbc.more_io)
Jens Axboe03ba3782009-09-09 09:08:54 +0200739 break;
Jens Axboe71fd05a2009-09-23 19:32:26 +0200740 /*
741 * Did we write something? Try for more
742 */
Jan Karab9543da2011-01-13 15:45:48 -0800743 if (wbc.nr_to_write < write_chunk)
Jens Axboe71fd05a2009-09-23 19:32:26 +0200744 continue;
745 /*
746 * Nothing written. Wait for some inode to
747 * become available for writeback. Otherwise
748 * we'll just busyloop.
749 */
Dave Chinnera66979a2011-03-22 22:23:41 +1100750 spin_lock(&inode_wb_list_lock);
Jens Axboe71fd05a2009-09-23 19:32:26 +0200751 if (!list_empty(&wb->b_more_io)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100752 inode = wb_inode(wb->b_more_io.prev);
Dave Chinner028c2dd2010-07-07 13:24:07 +1000753 trace_wbc_writeback_wait(&wbc, wb->bdi);
Dave Chinner250df6e2011-03-22 22:23:36 +1100754 spin_lock(&inode->i_lock);
Jens Axboe71fd05a2009-09-23 19:32:26 +0200755 inode_wait_for_writeback(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +1100756 spin_unlock(&inode->i_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200757 }
Dave Chinnera66979a2011-03-22 22:23:41 +1100758 spin_unlock(&inode_wb_list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200759 }
760
761 return wrote;
762}
763
764/*
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200765 * Return the next wb_writeback_work struct that hasn't been processed yet.
Jens Axboe03ba3782009-09-09 09:08:54 +0200766 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200767static struct wb_writeback_work *
Minchan Kim08852b62010-08-03 12:51:16 +0200768get_next_work_item(struct backing_dev_info *bdi)
Jens Axboe03ba3782009-09-09 09:08:54 +0200769{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200770 struct wb_writeback_work *work = NULL;
Jens Axboe03ba3782009-09-09 09:08:54 +0200771
Artem Bityutskiy64677162010-07-25 14:29:22 +0300772 spin_lock_bh(&bdi->wb_lock);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200773 if (!list_empty(&bdi->work_list)) {
774 work = list_entry(bdi->work_list.next,
775 struct wb_writeback_work, list);
776 list_del_init(&work->list);
Jens Axboe03ba3782009-09-09 09:08:54 +0200777 }
Artem Bityutskiy64677162010-07-25 14:29:22 +0300778 spin_unlock_bh(&bdi->wb_lock);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200779 return work;
Jens Axboe03ba3782009-09-09 09:08:54 +0200780}
781
Linus Torvaldscdf01dd2010-10-30 08:55:52 -0700782/*
783 * Add in the number of potentially dirty inodes, because each inode
784 * write can dirty pagecache in the underlying blockdev.
785 */
786static unsigned long get_nr_dirty_pages(void)
787{
788 return global_page_state(NR_FILE_DIRTY) +
789 global_page_state(NR_UNSTABLE_NFS) +
790 get_nr_dirty_inodes();
791}
792
Jan Kara65850272011-01-13 15:45:44 -0800793static long wb_check_background_flush(struct bdi_writeback *wb)
794{
795 if (over_bground_thresh()) {
796
797 struct wb_writeback_work work = {
798 .nr_pages = LONG_MAX,
799 .sync_mode = WB_SYNC_NONE,
800 .for_background = 1,
801 .range_cyclic = 1,
802 };
803
804 return wb_writeback(wb, &work);
805 }
806
807 return 0;
808}
809
Jens Axboe03ba3782009-09-09 09:08:54 +0200810static long wb_check_old_data_flush(struct bdi_writeback *wb)
811{
812 unsigned long expired;
813 long nr_pages;
814
Jens Axboe69b62d02010-05-17 12:51:03 +0200815 /*
816 * When set to zero, disable periodic writeback
817 */
818 if (!dirty_writeback_interval)
819 return 0;
820
Jens Axboe03ba3782009-09-09 09:08:54 +0200821 expired = wb->last_old_flush +
822 msecs_to_jiffies(dirty_writeback_interval * 10);
823 if (time_before(jiffies, expired))
824 return 0;
825
826 wb->last_old_flush = jiffies;
Linus Torvaldscdf01dd2010-10-30 08:55:52 -0700827 nr_pages = get_nr_dirty_pages();
Jens Axboe03ba3782009-09-09 09:08:54 +0200828
Jens Axboec4a77a62009-09-16 15:18:25 +0200829 if (nr_pages) {
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200830 struct wb_writeback_work work = {
Jens Axboec4a77a62009-09-16 15:18:25 +0200831 .nr_pages = nr_pages,
832 .sync_mode = WB_SYNC_NONE,
833 .for_kupdate = 1,
834 .range_cyclic = 1,
835 };
836
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200837 return wb_writeback(wb, &work);
Jens Axboec4a77a62009-09-16 15:18:25 +0200838 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200839
840 return 0;
841}
842
843/*
844 * Retrieve work items and do the writeback they describe
845 */
846long wb_do_writeback(struct bdi_writeback *wb, int force_wait)
847{
848 struct backing_dev_info *bdi = wb->bdi;
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200849 struct wb_writeback_work *work;
Jens Axboec4a77a62009-09-16 15:18:25 +0200850 long wrote = 0;
Jens Axboe03ba3782009-09-09 09:08:54 +0200851
Jan Kara81d73a32010-08-11 14:17:44 -0700852 set_bit(BDI_writeback_running, &wb->bdi->state);
Minchan Kim08852b62010-08-03 12:51:16 +0200853 while ((work = get_next_work_item(bdi)) != NULL) {
Jens Axboe03ba3782009-09-09 09:08:54 +0200854 /*
855 * Override sync mode, in case we must wait for completion
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200856 * because this thread is exiting now.
Jens Axboe03ba3782009-09-09 09:08:54 +0200857 */
858 if (force_wait)
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200859 work->sync_mode = WB_SYNC_ALL;
860
Dave Chinner455b2862010-07-07 13:24:06 +1000861 trace_writeback_exec(bdi, work);
862
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200863 wrote += wb_writeback(wb, work);
Jens Axboe03ba3782009-09-09 09:08:54 +0200864
865 /*
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200866 * Notify the caller of completion if this is a synchronous
867 * work item, otherwise just free it.
Jens Axboe03ba3782009-09-09 09:08:54 +0200868 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200869 if (work->done)
870 complete(work->done);
871 else
872 kfree(work);
Jens Axboe03ba3782009-09-09 09:08:54 +0200873 }
874
875 /*
876 * Check for periodic writeback, kupdated() style
877 */
878 wrote += wb_check_old_data_flush(wb);
Jan Kara65850272011-01-13 15:45:44 -0800879 wrote += wb_check_background_flush(wb);
Jan Kara81d73a32010-08-11 14:17:44 -0700880 clear_bit(BDI_writeback_running, &wb->bdi->state);
Jens Axboe03ba3782009-09-09 09:08:54 +0200881
882 return wrote;
883}
884
885/*
886 * Handle writeback of dirty data for the device backed by this bdi. Also
887 * wakes up periodically and does kupdated style flushing.
888 */
Christoph Hellwig08243902010-06-19 23:08:22 +0200889int bdi_writeback_thread(void *data)
Jens Axboe03ba3782009-09-09 09:08:54 +0200890{
Christoph Hellwig08243902010-06-19 23:08:22 +0200891 struct bdi_writeback *wb = data;
892 struct backing_dev_info *bdi = wb->bdi;
Jens Axboe03ba3782009-09-09 09:08:54 +0200893 long pages_written;
894
Peter Zijlstra766f9162010-10-26 14:22:45 -0700895 current->flags |= PF_SWAPWRITE;
Christoph Hellwig08243902010-06-19 23:08:22 +0200896 set_freezable();
Artem Bityutskiyecd58402010-07-25 14:29:18 +0300897 wb->last_active = jiffies;
Christoph Hellwig08243902010-06-19 23:08:22 +0200898
899 /*
900 * Our parent may run at a different priority, just set us to normal
901 */
902 set_user_nice(current, 0);
903
Dave Chinner455b2862010-07-07 13:24:06 +1000904 trace_writeback_thread_start(bdi);
905
Jens Axboe03ba3782009-09-09 09:08:54 +0200906 while (!kthread_should_stop()) {
Artem Bityutskiy64677162010-07-25 14:29:22 +0300907 /*
908 * Remove own delayed wake-up timer, since we are already awake
909 * and we'll take care of the preriodic write-back.
910 */
911 del_timer(&wb->wakeup_timer);
912
Jens Axboe03ba3782009-09-09 09:08:54 +0200913 pages_written = wb_do_writeback(wb, 0);
914
Dave Chinner455b2862010-07-07 13:24:06 +1000915 trace_writeback_pages_written(pages_written);
Jens Axboe03ba3782009-09-09 09:08:54 +0200916
Jens Axboe03ba3782009-09-09 09:08:54 +0200917 if (pages_written)
Artem Bityutskiyecd58402010-07-25 14:29:18 +0300918 wb->last_active = jiffies;
Jens Axboe03ba3782009-09-09 09:08:54 +0200919
Artem Bityutskiy297252c2010-07-25 14:29:15 +0300920 set_current_state(TASK_INTERRUPTIBLE);
J. Bruce Fieldsb76b4012010-08-28 08:52:10 +0200921 if (!list_empty(&bdi->work_list) || kthread_should_stop()) {
Artem Bityutskiy297252c2010-07-25 14:29:15 +0300922 __set_current_state(TASK_RUNNING);
923 continue;
Jens Axboe03ba3782009-09-09 09:08:54 +0200924 }
925
Artem Bityutskiy253c34e2010-07-25 14:29:21 +0300926 if (wb_has_dirty_io(wb) && dirty_writeback_interval)
Artem Bityutskiyfff5b852010-07-25 14:29:20 +0300927 schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10));
Artem Bityutskiy253c34e2010-07-25 14:29:21 +0300928 else {
929 /*
930 * We have nothing to do, so can go sleep without any
931 * timeout and save power. When a work is queued or
932 * something is made dirty - we will be woken up.
933 */
Artem Bityutskiy297252c2010-07-25 14:29:15 +0300934 schedule();
Jens Axboef9eadbb2010-05-18 14:31:45 +0200935 }
Jens Axboe69b62d02010-05-17 12:51:03 +0200936
Jens Axboe03ba3782009-09-09 09:08:54 +0200937 try_to_freeze();
938 }
939
Artem Bityutskiyfff5b852010-07-25 14:29:20 +0300940 /* Flush any work that raced with us exiting */
Christoph Hellwig08243902010-06-19 23:08:22 +0200941 if (!list_empty(&bdi->work_list))
942 wb_do_writeback(wb, 1);
Dave Chinner455b2862010-07-07 13:24:06 +1000943
944 trace_writeback_thread_stop(bdi);
Jens Axboe03ba3782009-09-09 09:08:54 +0200945 return 0;
946}
947
Christoph Hellwig08243902010-06-19 23:08:22 +0200948
Jens Axboe03ba3782009-09-09 09:08:54 +0200949/*
Jens Axboe03ba3782009-09-09 09:08:54 +0200950 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
951 * the whole world.
952 */
953void wakeup_flusher_threads(long nr_pages)
954{
Christoph Hellwigb8c2f342010-06-08 18:15:07 +0200955 struct backing_dev_info *bdi;
Christoph Hellwigb8c2f342010-06-08 18:15:07 +0200956
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200957 if (!nr_pages) {
958 nr_pages = global_page_state(NR_FILE_DIRTY) +
Jens Axboe03ba3782009-09-09 09:08:54 +0200959 global_page_state(NR_UNSTABLE_NFS);
Christoph Hellwigb8c2f342010-06-08 18:15:07 +0200960 }
961
962 rcu_read_lock();
963 list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) {
964 if (!bdi_has_dirty_io(bdi))
965 continue;
Jan Kara65850272011-01-13 15:45:44 -0800966 __bdi_start_writeback(bdi, nr_pages, false);
Christoph Hellwigb8c2f342010-06-08 18:15:07 +0200967 }
968 rcu_read_unlock();
Jens Axboe03ba3782009-09-09 09:08:54 +0200969}
970
971static noinline void block_dump___mark_inode_dirty(struct inode *inode)
972{
973 if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) {
974 struct dentry *dentry;
975 const char *name = "?";
976
977 dentry = d_find_alias(inode);
978 if (dentry) {
979 spin_lock(&dentry->d_lock);
980 name = (const char *) dentry->d_name.name;
981 }
982 printk(KERN_DEBUG
983 "%s(%d): dirtied inode %lu (%s) on %s\n",
984 current->comm, task_pid_nr(current), inode->i_ino,
985 name, inode->i_sb->s_id);
986 if (dentry) {
987 spin_unlock(&dentry->d_lock);
988 dput(dentry);
989 }
990 }
991}
992
993/**
994 * __mark_inode_dirty - internal function
995 * @inode: inode to mark
996 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
997 * Mark an inode as dirty. Callers should use mark_inode_dirty or
998 * mark_inode_dirty_sync.
999 *
1000 * Put the inode on the super block's dirty list.
1001 *
1002 * CAREFUL! We mark it dirty unconditionally, but move it onto the
1003 * dirty list only if it is hashed or if it refers to a blockdev.
1004 * If it was not hashed, it will never be added to the dirty list
1005 * even if it is later hashed, as it will have been marked dirty already.
1006 *
1007 * In short, make sure you hash any inodes _before_ you start marking
1008 * them dirty.
1009 *
Jens Axboe03ba3782009-09-09 09:08:54 +02001010 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
1011 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
1012 * the kernel-internal blockdev inode represents the dirtying time of the
1013 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
1014 * page->mapping->host, so the page-dirtying time is recorded in the internal
1015 * blockdev inode.
1016 */
1017void __mark_inode_dirty(struct inode *inode, int flags)
1018{
1019 struct super_block *sb = inode->i_sb;
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001020 struct backing_dev_info *bdi = NULL;
Jens Axboe03ba3782009-09-09 09:08:54 +02001021
1022 /*
1023 * Don't do this for I_DIRTY_PAGES - that doesn't actually
1024 * dirty the inode itself
1025 */
1026 if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
1027 if (sb->s_op->dirty_inode)
Christoph Hellwigaa385722011-05-27 06:53:02 -04001028 sb->s_op->dirty_inode(inode, flags);
Jens Axboe03ba3782009-09-09 09:08:54 +02001029 }
1030
1031 /*
1032 * make sure that changes are seen by all cpus before we test i_state
1033 * -- mikulas
1034 */
1035 smp_mb();
1036
1037 /* avoid the locking if we can */
1038 if ((inode->i_state & flags) == flags)
1039 return;
1040
1041 if (unlikely(block_dump))
1042 block_dump___mark_inode_dirty(inode);
1043
Dave Chinner250df6e2011-03-22 22:23:36 +11001044 spin_lock(&inode->i_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001045 if ((inode->i_state & flags) != flags) {
1046 const int was_dirty = inode->i_state & I_DIRTY;
1047
1048 inode->i_state |= flags;
1049
1050 /*
1051 * If the inode is being synced, just update its dirty state.
1052 * The unlocker will place the inode on the appropriate
1053 * superblock list, based upon its state.
1054 */
1055 if (inode->i_state & I_SYNC)
Dave Chinner250df6e2011-03-22 22:23:36 +11001056 goto out_unlock_inode;
Jens Axboe03ba3782009-09-09 09:08:54 +02001057
1058 /*
1059 * Only add valid (hashed) inodes to the superblock's
1060 * dirty list. Add blockdev inodes as well.
1061 */
1062 if (!S_ISBLK(inode->i_mode)) {
Al Viro1d3382cb2010-10-23 15:19:20 -04001063 if (inode_unhashed(inode))
Dave Chinner250df6e2011-03-22 22:23:36 +11001064 goto out_unlock_inode;
Jens Axboe03ba3782009-09-09 09:08:54 +02001065 }
Al Viroa4ffdde2010-06-02 17:38:30 -04001066 if (inode->i_state & I_FREEING)
Dave Chinner250df6e2011-03-22 22:23:36 +11001067 goto out_unlock_inode;
Jens Axboe03ba3782009-09-09 09:08:54 +02001068
1069 /*
1070 * If the inode was already on b_dirty/b_io/b_more_io, don't
1071 * reposition it (that would break b_dirty time-ordering).
1072 */
1073 if (!was_dirty) {
Dave Chinnera66979a2011-03-22 22:23:41 +11001074 bool wakeup_bdi = false;
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001075 bdi = inode_to_bdi(inode);
Jens Axboe500b0672009-09-09 09:10:25 +02001076
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001077 if (bdi_cap_writeback_dirty(bdi)) {
1078 WARN(!test_bit(BDI_registered, &bdi->state),
1079 "bdi-%s not registered\n", bdi->name);
1080
1081 /*
1082 * If this is the first dirty inode for this
1083 * bdi, we have to wake-up the corresponding
1084 * bdi thread to make sure background
1085 * write-back happens later.
1086 */
1087 if (!wb_has_dirty_io(&bdi->wb))
1088 wakeup_bdi = true;
Jens Axboe500b0672009-09-09 09:10:25 +02001089 }
Jens Axboe03ba3782009-09-09 09:08:54 +02001090
Dave Chinnera66979a2011-03-22 22:23:41 +11001091 spin_unlock(&inode->i_lock);
1092 spin_lock(&inode_wb_list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001093 inode->dirtied_when = jiffies;
Nick Piggin7ccf19a2010-10-21 11:49:30 +11001094 list_move(&inode->i_wb_list, &bdi->wb.b_dirty);
Dave Chinnera66979a2011-03-22 22:23:41 +11001095 spin_unlock(&inode_wb_list_lock);
1096
1097 if (wakeup_bdi)
1098 bdi_wakeup_thread_delayed(bdi);
1099 return;
Jens Axboe03ba3782009-09-09 09:08:54 +02001100 }
1101 }
Dave Chinner250df6e2011-03-22 22:23:36 +11001102out_unlock_inode:
1103 spin_unlock(&inode->i_lock);
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001104
Jens Axboe03ba3782009-09-09 09:08:54 +02001105}
1106EXPORT_SYMBOL(__mark_inode_dirty);
1107
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001108/*
1109 * Write out a superblock's list of dirty inodes. A wait will be performed
1110 * upon no inodes, all inodes or the final one, depending upon sync_mode.
1111 *
1112 * If older_than_this is non-NULL, then only write out inodes which
1113 * had their first dirtying at a time earlier than *older_than_this.
1114 *
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001115 * If `bdi' is non-zero then we're being asked to writeback a specific queue.
1116 * This function assumes that the blockdev superblock's inodes are backed by
1117 * a variety of queues, so all inodes are searched. For other superblocks,
1118 * assume that all inodes are backed by the same queue.
1119 *
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001120 * The inodes to be written are parked on bdi->b_io. They are moved back onto
1121 * bdi->b_dirty as they are selected for writing. This way, none can be missed
1122 * on the writer throttling path, and we get decent balancing between many
1123 * throttled threads: we don't want them all piling up on inode_sync_wait.
1124 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001125static void wait_sb_inodes(struct super_block *sb)
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001126{
Jens Axboe03ba3782009-09-09 09:08:54 +02001127 struct inode *inode, *old_inode = NULL;
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001128
Jens Axboe03ba3782009-09-09 09:08:54 +02001129 /*
1130 * We need to be protected against the filesystem going from
1131 * r/o to r/w or vice versa.
1132 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001133 WARN_ON(!rwsem_is_locked(&sb->s_umount));
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001134
Dave Chinner55fa6092011-03-22 22:23:40 +11001135 spin_lock(&inode_sb_list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001136
1137 /*
1138 * Data integrity sync. Must wait for all pages under writeback,
1139 * because there may have been pages dirtied before our sync
1140 * call, but which had writeout started before we write it out.
1141 * In which case, the inode may not be on the dirty list, but
1142 * we still have to wait for that writeout.
1143 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001144 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
Dave Chinner250df6e2011-03-22 22:23:36 +11001145 struct address_space *mapping = inode->i_mapping;
Jens Axboe03ba3782009-09-09 09:08:54 +02001146
Dave Chinner250df6e2011-03-22 22:23:36 +11001147 spin_lock(&inode->i_lock);
1148 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
1149 (mapping->nrpages == 0)) {
1150 spin_unlock(&inode->i_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001151 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +11001152 }
Jens Axboe03ba3782009-09-09 09:08:54 +02001153 __iget(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +11001154 spin_unlock(&inode->i_lock);
Dave Chinner55fa6092011-03-22 22:23:40 +11001155 spin_unlock(&inode_sb_list_lock);
1156
Jens Axboe03ba3782009-09-09 09:08:54 +02001157 /*
Dave Chinner55fa6092011-03-22 22:23:40 +11001158 * We hold a reference to 'inode' so it couldn't have been
1159 * removed from s_inodes list while we dropped the
1160 * inode_sb_list_lock. We cannot iput the inode now as we can
1161 * be holding the last reference and we cannot iput it under
1162 * inode_sb_list_lock. So we keep the reference and iput it
1163 * later.
Jens Axboe03ba3782009-09-09 09:08:54 +02001164 */
1165 iput(old_inode);
1166 old_inode = inode;
1167
1168 filemap_fdatawait(mapping);
1169
1170 cond_resched();
Nick Piggin38f21972009-01-06 14:40:25 -08001171
Dave Chinner55fa6092011-03-22 22:23:40 +11001172 spin_lock(&inode_sb_list_lock);
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001173 }
Dave Chinner55fa6092011-03-22 22:23:40 +11001174 spin_unlock(&inode_sb_list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001175 iput(old_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176}
1177
Jens Axboed8a85592009-09-02 12:34:32 +02001178/**
Chris Mason3259f8b2010-10-29 11:16:17 -04001179 * writeback_inodes_sb_nr - writeback dirty inodes from given super_block
Jens Axboed8a85592009-09-02 12:34:32 +02001180 * @sb: the superblock
Chris Mason3259f8b2010-10-29 11:16:17 -04001181 * @nr: the number of pages to write
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182 *
Jens Axboed8a85592009-09-02 12:34:32 +02001183 * Start writeback on some inodes on this super_block. No guarantees are made
1184 * on how many (if any) will be written, and this function does not wait
Chris Mason3259f8b2010-10-29 11:16:17 -04001185 * for IO completion of submitted IO.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186 */
Chris Mason3259f8b2010-10-29 11:16:17 -04001187void writeback_inodes_sb_nr(struct super_block *sb, unsigned long nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001189 DECLARE_COMPLETION_ONSTACK(done);
1190 struct wb_writeback_work work = {
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001191 .sb = sb,
1192 .sync_mode = WB_SYNC_NONE,
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001193 .done = &done,
Chris Mason3259f8b2010-10-29 11:16:17 -04001194 .nr_pages = nr,
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001195 };
Jens Axboe0e3c9a22010-06-01 11:08:43 +02001196
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001197 WARN_ON(!rwsem_is_locked(&sb->s_umount));
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001198 bdi_queue_work(sb->s_bdi, &work);
1199 wait_for_completion(&done);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200}
Chris Mason3259f8b2010-10-29 11:16:17 -04001201EXPORT_SYMBOL(writeback_inodes_sb_nr);
1202
1203/**
1204 * writeback_inodes_sb - writeback dirty inodes from given super_block
1205 * @sb: the superblock
1206 *
1207 * Start writeback on some inodes on this super_block. No guarantees are made
1208 * on how many (if any) will be written, and this function does not wait
1209 * for IO completion of submitted IO.
1210 */
1211void writeback_inodes_sb(struct super_block *sb)
1212{
Linus Torvalds925d1692010-10-30 09:05:48 -07001213 return writeback_inodes_sb_nr(sb, get_nr_dirty_pages());
Chris Mason3259f8b2010-10-29 11:16:17 -04001214}
Jens Axboed8a85592009-09-02 12:34:32 +02001215EXPORT_SYMBOL(writeback_inodes_sb);
1216
1217/**
Eric Sandeen17bd55d2009-12-23 07:57:07 -05001218 * writeback_inodes_sb_if_idle - start writeback if none underway
1219 * @sb: the superblock
1220 *
1221 * Invoke writeback_inodes_sb if no writeback is currently underway.
1222 * Returns 1 if writeback was started, 0 if not.
1223 */
1224int writeback_inodes_sb_if_idle(struct super_block *sb)
1225{
1226 if (!writeback_in_progress(sb->s_bdi)) {
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001227 down_read(&sb->s_umount);
Eric Sandeen17bd55d2009-12-23 07:57:07 -05001228 writeback_inodes_sb(sb);
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001229 up_read(&sb->s_umount);
Eric Sandeen17bd55d2009-12-23 07:57:07 -05001230 return 1;
1231 } else
1232 return 0;
1233}
1234EXPORT_SYMBOL(writeback_inodes_sb_if_idle);
1235
1236/**
Chris Mason3259f8b2010-10-29 11:16:17 -04001237 * writeback_inodes_sb_if_idle - start writeback if none underway
1238 * @sb: the superblock
1239 * @nr: the number of pages to write
1240 *
1241 * Invoke writeback_inodes_sb if no writeback is currently underway.
1242 * Returns 1 if writeback was started, 0 if not.
1243 */
1244int writeback_inodes_sb_nr_if_idle(struct super_block *sb,
1245 unsigned long nr)
1246{
1247 if (!writeback_in_progress(sb->s_bdi)) {
1248 down_read(&sb->s_umount);
1249 writeback_inodes_sb_nr(sb, nr);
1250 up_read(&sb->s_umount);
1251 return 1;
1252 } else
1253 return 0;
1254}
1255EXPORT_SYMBOL(writeback_inodes_sb_nr_if_idle);
1256
1257/**
Jens Axboed8a85592009-09-02 12:34:32 +02001258 * sync_inodes_sb - sync sb inode pages
1259 * @sb: the superblock
1260 *
1261 * This function writes and waits on any dirty inode belonging to this
Stefan Hajnoczicb9ef8d2011-01-13 15:47:26 -08001262 * super_block.
Jens Axboed8a85592009-09-02 12:34:32 +02001263 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001264void sync_inodes_sb(struct super_block *sb)
Jens Axboed8a85592009-09-02 12:34:32 +02001265{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001266 DECLARE_COMPLETION_ONSTACK(done);
1267 struct wb_writeback_work work = {
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001268 .sb = sb,
1269 .sync_mode = WB_SYNC_ALL,
1270 .nr_pages = LONG_MAX,
1271 .range_cyclic = 0,
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001272 .done = &done,
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001273 };
1274
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001275 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1276
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001277 bdi_queue_work(sb->s_bdi, &work);
1278 wait_for_completion(&done);
1279
Jens Axboeb6e51312009-09-16 15:13:54 +02001280 wait_sb_inodes(sb);
Jens Axboed8a85592009-09-02 12:34:32 +02001281}
1282EXPORT_SYMBOL(sync_inodes_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284/**
Andrea Arcangeli7f04c262005-10-30 15:03:05 -08001285 * write_inode_now - write an inode to disk
1286 * @inode: inode to write to disk
1287 * @sync: whether the write should be synchronous or not
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288 *
Andrea Arcangeli7f04c262005-10-30 15:03:05 -08001289 * This function commits an inode to disk immediately if it is dirty. This is
1290 * primarily needed by knfsd.
1291 *
1292 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294int write_inode_now(struct inode *inode, int sync)
1295{
1296 int ret;
1297 struct writeback_control wbc = {
1298 .nr_to_write = LONG_MAX,
Mike Galbraith18914b12008-02-08 04:20:23 -08001299 .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -07001300 .range_start = 0,
1301 .range_end = LLONG_MAX,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302 };
1303
1304 if (!mapping_cap_writeback_dirty(inode->i_mapping))
Andrew Morton49364ce2005-11-07 00:59:15 -08001305 wbc.nr_to_write = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306
1307 might_sleep();
Dave Chinnera66979a2011-03-22 22:23:41 +11001308 spin_lock(&inode_wb_list_lock);
Dave Chinner0f1b1fd2011-03-22 22:23:43 +11001309 spin_lock(&inode->i_lock);
Christoph Hellwig01c03192009-06-08 13:35:40 +02001310 ret = writeback_single_inode(inode, &wbc);
Dave Chinner0f1b1fd2011-03-22 22:23:43 +11001311 spin_unlock(&inode->i_lock);
Dave Chinnera66979a2011-03-22 22:23:41 +11001312 spin_unlock(&inode_wb_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313 if (sync)
Joern Engel1c0eeaf2007-10-16 23:30:44 -07001314 inode_sync_wait(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315 return ret;
1316}
1317EXPORT_SYMBOL(write_inode_now);
1318
1319/**
1320 * sync_inode - write an inode and its pages to disk.
1321 * @inode: the inode to sync
1322 * @wbc: controls the writeback mode
1323 *
1324 * sync_inode() will write an inode and its pages to disk. It will also
1325 * correctly update the inode on its superblock's dirty inode lists and will
1326 * update inode->i_state.
1327 *
1328 * The caller must have a ref on the inode.
1329 */
1330int sync_inode(struct inode *inode, struct writeback_control *wbc)
1331{
1332 int ret;
1333
Dave Chinnera66979a2011-03-22 22:23:41 +11001334 spin_lock(&inode_wb_list_lock);
Dave Chinner0f1b1fd2011-03-22 22:23:43 +11001335 spin_lock(&inode->i_lock);
Christoph Hellwig01c03192009-06-08 13:35:40 +02001336 ret = writeback_single_inode(inode, wbc);
Dave Chinner0f1b1fd2011-03-22 22:23:43 +11001337 spin_unlock(&inode->i_lock);
Dave Chinnera66979a2011-03-22 22:23:41 +11001338 spin_unlock(&inode_wb_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339 return ret;
1340}
1341EXPORT_SYMBOL(sync_inode);
Christoph Hellwigc37650162010-10-06 10:48:20 +02001342
1343/**
Andrew Mortonc691b9d2011-01-13 15:45:48 -08001344 * sync_inode_metadata - write an inode to disk
Christoph Hellwigc37650162010-10-06 10:48:20 +02001345 * @inode: the inode to sync
1346 * @wait: wait for I/O to complete.
1347 *
Andrew Mortonc691b9d2011-01-13 15:45:48 -08001348 * Write an inode to disk and adjust its dirty state after completion.
Christoph Hellwigc37650162010-10-06 10:48:20 +02001349 *
1350 * Note: only writes the actual inode, no associated data or other metadata.
1351 */
1352int sync_inode_metadata(struct inode *inode, int wait)
1353{
1354 struct writeback_control wbc = {
1355 .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE,
1356 .nr_to_write = 0, /* metadata-only */
1357 };
1358
1359 return sync_inode(inode, &wbc);
1360}
1361EXPORT_SYMBOL(sync_inode_metadata);