blob: fca43d4d7bf4f0657edcf41def8ee1281d822f29 [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
55#define inode_to_bdi(inode) ((inode)->i_mapping->backing_dev_info)
56
57/*
58 * We don't actually have pdflush, but this one is exported though /proc...
59 */
60int nr_pdflush_threads;
61
Adrian Bunkf11b00f2008-04-29 00:58:56 -070062/**
63 * writeback_in_progress - determine whether there is writeback in progress
64 * @bdi: the device's backing_dev_info structure.
65 *
Jens Axboe03ba3782009-09-09 09:08:54 +020066 * Determine whether there is writeback waiting to be handled against a
67 * backing device.
Adrian Bunkf11b00f2008-04-29 00:58:56 -070068 */
69int writeback_in_progress(struct backing_dev_info *bdi)
70{
Jens Axboe03ba3782009-09-09 09:08:54 +020071 return !list_empty(&bdi->work_list);
Adrian Bunkf11b00f2008-04-29 00:58:56 -070072}
73
Christoph Hellwig83ba7b02010-07-06 08:59:53 +020074static void bdi_queue_work(struct backing_dev_info *bdi,
75 struct wb_writeback_work *work)
Nick Piggin4195f732009-05-28 09:01:15 +020076{
Dave Chinner455b2862010-07-07 13:24:06 +100077 trace_writeback_queue(bdi, work);
78
Artem Bityutskiy64677162010-07-25 14:29:22 +030079 spin_lock_bh(&bdi->wb_lock);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +020080 list_add_tail(&work->list, &bdi->work_list);
Artem Bityutskiyfff5b852010-07-25 14:29:20 +030081 if (bdi->wb.task) {
82 wake_up_process(bdi->wb.task);
83 } else {
84 /*
85 * The bdi thread isn't there, wake up the forker thread which
86 * will create and run it.
87 */
Dave Chinner455b2862010-07-07 13:24:06 +100088 trace_writeback_nothread(bdi, work);
Jens Axboe03ba3782009-09-09 09:08:54 +020089 wake_up_process(default_backing_dev_info.wb.task);
Linus Torvalds1da177e2005-04-16 15:20:36 -070090 }
Artem Bityutskiy64677162010-07-25 14:29:22 +030091 spin_unlock_bh(&bdi->wb_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070092}
93
Christoph Hellwig83ba7b02010-07-06 08:59:53 +020094static void
95__bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages,
96 bool range_cyclic, bool for_background)
Linus Torvalds1da177e2005-04-16 15:20:36 -070097{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +020098 struct wb_writeback_work *work;
Jens Axboe03ba3782009-09-09 09:08:54 +020099
Jens Axboebcddc3f2009-09-13 20:07:36 +0200100 /*
101 * This is WB_SYNC_NONE writeback, so if allocation fails just
102 * wakeup the thread for old dirty data writeback
103 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200104 work = kzalloc(sizeof(*work), GFP_ATOMIC);
105 if (!work) {
Dave Chinner455b2862010-07-07 13:24:06 +1000106 if (bdi->wb.task) {
107 trace_writeback_nowork(bdi);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200108 wake_up_process(bdi->wb.task);
Dave Chinner455b2862010-07-07 13:24:06 +1000109 }
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200110 return;
Jens Axboebcddc3f2009-09-13 20:07:36 +0200111 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200112
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200113 work->sync_mode = WB_SYNC_NONE;
114 work->nr_pages = nr_pages;
115 work->range_cyclic = range_cyclic;
116 work->for_background = for_background;
Christoph Hellwigf0fad8a2009-09-11 09:47:56 +0200117
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200118 bdi_queue_work(bdi, work);
Jens Axboeb6e51312009-09-16 15:13:54 +0200119}
120
121/**
122 * bdi_start_writeback - start writeback
123 * @bdi: the backing device to write from
124 * @nr_pages: the number of pages to write
125 *
126 * Description:
127 * This does WB_SYNC_NONE opportunistic writeback. The IO is only
128 * started when this function returns, we make no guarentees on
Jens Axboe0e3c9a22010-06-01 11:08:43 +0200129 * completion. Caller need not hold sb s_umount semaphore.
Jens Axboeb6e51312009-09-16 15:13:54 +0200130 *
131 */
Christoph Hellwigc5444192010-06-08 18:15:15 +0200132void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages)
Jens Axboeb6e51312009-09-16 15:13:54 +0200133{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200134 __bdi_start_writeback(bdi, nr_pages, true, false);
Christoph Hellwigc5444192010-06-08 18:15:15 +0200135}
Wu Fengguangd3ddec72009-09-23 20:33:40 +0800136
Christoph Hellwigc5444192010-06-08 18:15:15 +0200137/**
138 * bdi_start_background_writeback - start background writeback
139 * @bdi: the backing device to write from
140 *
141 * Description:
142 * This does WB_SYNC_NONE background writeback. The IO is only
143 * started when this function returns, we make no guarentees on
144 * completion. Caller need not hold sb s_umount semaphore.
145 */
146void bdi_start_background_writeback(struct backing_dev_info *bdi)
147{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200148 __bdi_start_writeback(bdi, LONG_MAX, true, true);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149}
150
151/*
Andrew Morton6610a0b2007-10-16 23:30:32 -0700152 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
153 * furthest end of its superblock's dirty-inode list.
154 *
155 * Before stamping the inode's ->dirtied_when, we check to see whether it is
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200156 * already the most-recently-dirtied inode on the b_dirty list. If that is
Andrew Morton6610a0b2007-10-16 23:30:32 -0700157 * the case then the inode must have been redirtied while it was being written
158 * out and we don't reset its dirtied_when.
159 */
160static void redirty_tail(struct inode *inode)
161{
Jens Axboe03ba3782009-09-09 09:08:54 +0200162 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
Andrew Morton6610a0b2007-10-16 23:30:32 -0700163
Jens Axboe03ba3782009-09-09 09:08:54 +0200164 if (!list_empty(&wb->b_dirty)) {
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200165 struct inode *tail;
Andrew Morton6610a0b2007-10-16 23:30:32 -0700166
Jens Axboe03ba3782009-09-09 09:08:54 +0200167 tail = list_entry(wb->b_dirty.next, struct inode, i_list);
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200168 if (time_before(inode->dirtied_when, tail->dirtied_when))
Andrew Morton6610a0b2007-10-16 23:30:32 -0700169 inode->dirtied_when = jiffies;
170 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200171 list_move(&inode->i_list, &wb->b_dirty);
Andrew Morton6610a0b2007-10-16 23:30:32 -0700172}
173
174/*
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200175 * requeue inode for re-scanning after bdi->b_io list is exhausted.
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700176 */
Ken Chen0e0f4fc2007-10-16 23:30:38 -0700177static void requeue_io(struct inode *inode)
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700178{
Jens Axboe03ba3782009-09-09 09:08:54 +0200179 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
180
181 list_move(&inode->i_list, &wb->b_more_io);
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700182}
183
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700184static void inode_sync_complete(struct inode *inode)
185{
186 /*
187 * Prevent speculative execution through spin_unlock(&inode_lock);
188 */
189 smp_mb();
190 wake_up_bit(&inode->i_state, __I_SYNC);
191}
192
Jeff Laytond2caa3c2009-04-02 16:56:37 -0700193static bool inode_dirtied_after(struct inode *inode, unsigned long t)
194{
195 bool ret = time_after(inode->dirtied_when, t);
196#ifndef CONFIG_64BIT
197 /*
198 * For inodes being constantly redirtied, dirtied_when can get stuck.
199 * It _appears_ to be in the future, but is actually in distant past.
200 * This test is necessary to prevent such wrapped-around relative times
Jens Axboe5b0830c2009-09-23 19:37:09 +0200201 * from permanently stopping the whole bdi writeback.
Jeff Laytond2caa3c2009-04-02 16:56:37 -0700202 */
203 ret = ret && time_before_eq(inode->dirtied_when, jiffies);
204#endif
205 return ret;
206}
207
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700208/*
Fengguang Wu2c136572007-10-16 23:30:39 -0700209 * Move expired dirty inodes from @delaying_queue to @dispatch_queue.
210 */
211static void move_expired_inodes(struct list_head *delaying_queue,
212 struct list_head *dispatch_queue,
213 unsigned long *older_than_this)
214{
Shaohua Li5c034492009-09-24 14:42:33 +0200215 LIST_HEAD(tmp);
216 struct list_head *pos, *node;
Jens Axboecf137302009-09-24 15:12:57 +0200217 struct super_block *sb = NULL;
Shaohua Li5c034492009-09-24 14:42:33 +0200218 struct inode *inode;
Jens Axboecf137302009-09-24 15:12:57 +0200219 int do_sb_sort = 0;
Shaohua Li5c034492009-09-24 14:42:33 +0200220
Fengguang Wu2c136572007-10-16 23:30:39 -0700221 while (!list_empty(delaying_queue)) {
Shaohua Li5c034492009-09-24 14:42:33 +0200222 inode = list_entry(delaying_queue->prev, struct inode, i_list);
Fengguang Wu2c136572007-10-16 23:30:39 -0700223 if (older_than_this &&
Jeff Laytond2caa3c2009-04-02 16:56:37 -0700224 inode_dirtied_after(inode, *older_than_this))
Fengguang Wu2c136572007-10-16 23:30:39 -0700225 break;
Jens Axboecf137302009-09-24 15:12:57 +0200226 if (sb && sb != inode->i_sb)
227 do_sb_sort = 1;
228 sb = inode->i_sb;
Shaohua Li5c034492009-09-24 14:42:33 +0200229 list_move(&inode->i_list, &tmp);
230 }
231
Jens Axboecf137302009-09-24 15:12:57 +0200232 /* just one sb in list, splice to dispatch_queue and we're done */
233 if (!do_sb_sort) {
234 list_splice(&tmp, dispatch_queue);
235 return;
236 }
237
Shaohua Li5c034492009-09-24 14:42:33 +0200238 /* Move inodes from one superblock together */
239 while (!list_empty(&tmp)) {
240 inode = list_entry(tmp.prev, struct inode, i_list);
241 sb = inode->i_sb;
242 list_for_each_prev_safe(pos, node, &tmp) {
243 inode = list_entry(pos, struct inode, i_list);
244 if (inode->i_sb == sb)
245 list_move(&inode->i_list, dispatch_queue);
246 }
Fengguang Wu2c136572007-10-16 23:30:39 -0700247 }
248}
249
250/*
251 * Queue all expired dirty inodes for io, eldest first.
252 */
Jens Axboe03ba3782009-09-09 09:08:54 +0200253static void queue_io(struct bdi_writeback *wb, unsigned long *older_than_this)
Fengguang Wu2c136572007-10-16 23:30:39 -0700254{
Jens Axboe03ba3782009-09-09 09:08:54 +0200255 list_splice_init(&wb->b_more_io, wb->b_io.prev);
256 move_expired_inodes(&wb->b_dirty, &wb->b_io, older_than_this);
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200257}
258
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100259static int write_inode(struct inode *inode, struct writeback_control *wbc)
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200260{
Jens Axboe03ba3782009-09-09 09:08:54 +0200261 if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode))
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100262 return inode->i_sb->s_op->write_inode(inode, wbc);
Jens Axboe03ba3782009-09-09 09:08:54 +0200263 return 0;
Fengguang Wu2c136572007-10-16 23:30:39 -0700264}
265
266/*
Christoph Hellwig01c03192009-06-08 13:35:40 +0200267 * Wait for writeback on an inode to complete.
268 */
269static void inode_wait_for_writeback(struct inode *inode)
270{
271 DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC);
272 wait_queue_head_t *wqh;
273
274 wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
Richard Kennedy58a9d3d2010-05-24 14:32:38 -0700275 while (inode->i_state & I_SYNC) {
Christoph Hellwig01c03192009-06-08 13:35:40 +0200276 spin_unlock(&inode_lock);
277 __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE);
278 spin_lock(&inode_lock);
Richard Kennedy58a9d3d2010-05-24 14:32:38 -0700279 }
Christoph Hellwig01c03192009-06-08 13:35:40 +0200280}
281
282/*
283 * Write out an inode's dirty pages. Called under inode_lock. Either the
284 * caller has ref on the inode (either via __iget or via syscall against an fd)
285 * or the inode has I_WILL_FREE set (via generic_forget_inode)
286 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 * If `wait' is set, wait on the writeout.
288 *
289 * The whole writeout design is quite complex and fragile. We want to avoid
290 * starvation of particular inodes when others are being redirtied, prevent
291 * livelocks, etc.
292 *
293 * Called under inode_lock.
294 */
295static int
Christoph Hellwig01c03192009-06-08 13:35:40 +0200296writeback_single_inode(struct inode *inode, struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298 struct address_space *mapping = inode->i_mapping;
Christoph Hellwig01c03192009-06-08 13:35:40 +0200299 unsigned dirty;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300 int ret;
301
Christoph Hellwig01c03192009-06-08 13:35:40 +0200302 if (!atomic_read(&inode->i_count))
303 WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING)));
304 else
305 WARN_ON(inode->i_state & I_WILL_FREE);
306
307 if (inode->i_state & I_SYNC) {
308 /*
309 * If this inode is locked for writeback and we are not doing
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200310 * writeback-for-data-integrity, move it to b_more_io so that
Christoph Hellwig01c03192009-06-08 13:35:40 +0200311 * writeback can proceed with the other inodes on s_io.
312 *
313 * We'll have another go at writing back this inode when we
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200314 * completed a full scan of b_io.
Christoph Hellwig01c03192009-06-08 13:35:40 +0200315 */
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100316 if (wbc->sync_mode != WB_SYNC_ALL) {
Christoph Hellwig01c03192009-06-08 13:35:40 +0200317 requeue_io(inode);
318 return 0;
319 }
320
321 /*
322 * It's a data-integrity sync. We must wait.
323 */
324 inode_wait_for_writeback(inode);
325 }
326
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700327 BUG_ON(inode->i_state & I_SYNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328
Dmitry Monakhov5547e8a2010-05-07 13:35:44 +0400329 /* Set I_SYNC, reset I_DIRTY_PAGES */
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700330 inode->i_state |= I_SYNC;
Dmitry Monakhov5547e8a2010-05-07 13:35:44 +0400331 inode->i_state &= ~I_DIRTY_PAGES;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 spin_unlock(&inode_lock);
333
334 ret = do_writepages(mapping, wbc);
335
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100336 /*
337 * Make sure to wait on the data before writing out the metadata.
338 * This is important for filesystems that modify metadata on data
339 * I/O completion.
340 */
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100341 if (wbc->sync_mode == WB_SYNC_ALL) {
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100342 int err = filemap_fdatawait(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343 if (ret == 0)
344 ret = err;
345 }
346
Dmitry Monakhov5547e8a2010-05-07 13:35:44 +0400347 /*
348 * Some filesystems may redirty the inode during the writeback
349 * due to delalloc, clear dirty metadata flags right before
350 * write_inode()
351 */
352 spin_lock(&inode_lock);
353 dirty = inode->i_state & I_DIRTY;
354 inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC);
355 spin_unlock(&inode_lock);
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100356 /* Don't write the inode if only I_DIRTY_PAGES was set */
357 if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100358 int err = write_inode(inode, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359 if (ret == 0)
360 ret = err;
361 }
362
363 spin_lock(&inode_lock);
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700364 inode->i_state &= ~I_SYNC;
Al Viroa4ffdde2010-06-02 17:38:30 -0400365 if (!(inode->i_state & I_FREEING)) {
Wu Fengguangb3af9462009-09-25 06:04:10 +0200366 if ((inode->i_state & I_DIRTY_PAGES) && wbc->for_kupdate) {
Wu Fengguangae1b7f72009-09-23 20:33:42 +0800367 /*
Wu Fengguangb3af9462009-09-25 06:04:10 +0200368 * More pages get dirtied by a fast dirtier.
369 */
370 goto select_queue;
371 } else if (inode->i_state & I_DIRTY) {
372 /*
373 * At least XFS will redirty the inode during the
374 * writeback (delalloc) and on io completion (isize).
Wu Fengguangae1b7f72009-09-23 20:33:42 +0800375 */
376 redirty_tail(inode);
377 } else if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378 /*
379 * We didn't write back all the pages. nfs_writepages()
380 * sometimes bales out without doing anything. Redirty
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200381 * the inode; Move it from b_io onto b_more_io/b_dirty.
Andrew Morton1b43ef92007-10-16 23:30:35 -0700382 */
383 /*
384 * akpm: if the caller was the kupdate function we put
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200385 * this inode at the head of b_dirty so it gets first
Andrew Morton1b43ef92007-10-16 23:30:35 -0700386 * consideration. Otherwise, move it to the tail, for
387 * the reasons described there. I'm not really sure
388 * how much sense this makes. Presumably I had a good
389 * reasons for doing it this way, and I'd rather not
390 * muck with it at present.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391 */
392 if (wbc->for_kupdate) {
393 /*
Fengguang Wu2c136572007-10-16 23:30:39 -0700394 * For the kupdate function we move the inode
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200395 * to b_more_io so it will get more writeout as
Fengguang Wu2c136572007-10-16 23:30:39 -0700396 * soon as the queue becomes uncongested.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397 */
398 inode->i_state |= I_DIRTY_PAGES;
Wu Fengguangb3af9462009-09-25 06:04:10 +0200399select_queue:
Fengguang Wu8bc3be22008-02-04 22:29:36 -0800400 if (wbc->nr_to_write <= 0) {
401 /*
402 * slice used up: queue for next turn
403 */
404 requeue_io(inode);
405 } else {
406 /*
407 * somehow blocked: retry later
408 */
409 redirty_tail(inode);
410 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 } else {
412 /*
413 * Otherwise fully redirty the inode so that
414 * other inodes on this superblock will get some
415 * writeout. Otherwise heavy writing to one
416 * file would indefinitely suspend writeout of
417 * all the other files.
418 */
419 inode->i_state |= I_DIRTY_PAGES;
Andrew Morton1b43ef92007-10-16 23:30:35 -0700420 redirty_tail(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422 } else if (atomic_read(&inode->i_count)) {
423 /*
424 * The inode is clean, inuse
425 */
426 list_move(&inode->i_list, &inode_in_use);
427 } else {
428 /*
429 * The inode is clean, unused
430 */
431 list_move(&inode->i_list, &inode_unused);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432 }
433 }
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700434 inode_sync_complete(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435 return ret;
436}
437
Jens Axboe03ba3782009-09-09 09:08:54 +0200438/*
Christoph Hellwigd19de7e2010-06-08 18:14:58 +0200439 * For background writeback the caller does not have the sb pinned
Jens Axboe03ba3782009-09-09 09:08:54 +0200440 * before calling writeback. So make sure that we do pin it, so it doesn't
441 * go away while we are writing inodes from it.
Jens Axboe03ba3782009-09-09 09:08:54 +0200442 */
Christoph Hellwigd19de7e2010-06-08 18:14:58 +0200443static bool pin_sb_for_writeback(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444{
Jens Axboe03ba3782009-09-09 09:08:54 +0200445 spin_lock(&sb_lock);
Christoph Hellwig29cb4852010-06-09 15:31:01 +0200446 if (list_empty(&sb->s_instances)) {
447 spin_unlock(&sb_lock);
448 return false;
449 }
450
Jens Axboe03ba3782009-09-09 09:08:54 +0200451 sb->s_count++;
Christoph Hellwig29cb4852010-06-09 15:31:01 +0200452 spin_unlock(&sb_lock);
453
Jens Axboe03ba3782009-09-09 09:08:54 +0200454 if (down_read_trylock(&sb->s_umount)) {
Christoph Hellwig29cb4852010-06-09 15:31:01 +0200455 if (sb->s_root)
Christoph Hellwigd19de7e2010-06-08 18:14:58 +0200456 return true;
Jens Axboe03ba3782009-09-09 09:08:54 +0200457 up_read(&sb->s_umount);
458 }
Christoph Hellwig29cb4852010-06-09 15:31:01 +0200459
460 put_super(sb);
Christoph Hellwigd19de7e2010-06-08 18:14:58 +0200461 return false;
Jens Axboe03ba3782009-09-09 09:08:54 +0200462}
463
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800464/*
465 * Write a portion of b_io inodes which belong to @sb.
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200466 *
467 * If @only_this_sb is true, then find and write all such
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800468 * inodes. Otherwise write only ones which go sequentially
469 * in reverse order.
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200470 *
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800471 * Return 1, if the caller writeback routine should be
472 * interrupted. Otherwise return 0.
473 */
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200474static int writeback_sb_inodes(struct super_block *sb, struct bdi_writeback *wb,
475 struct writeback_control *wbc, bool only_this_sb)
Jens Axboe03ba3782009-09-09 09:08:54 +0200476{
Jens Axboe03ba3782009-09-09 09:08:54 +0200477 while (!list_empty(&wb->b_io)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 long pages_skipped;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800479 struct inode *inode = list_entry(wb->b_io.prev,
480 struct inode, i_list);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200481
482 if (inode->i_sb != sb) {
483 if (only_this_sb) {
484 /*
485 * We only want to write back data for this
486 * superblock, move all inodes not belonging
487 * to it back onto the dirty list.
488 */
489 redirty_tail(inode);
490 continue;
491 }
492
493 /*
494 * The inode belongs to a different superblock.
495 * Bounce back to the caller to unpin this and
496 * pin the next superblock.
497 */
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800498 return 0;
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200499 }
500
Wu Fengguang84a89242009-06-16 15:33:17 -0700501 if (inode->i_state & (I_NEW | I_WILL_FREE)) {
Nick Piggin7ef0d732009-03-12 14:31:38 -0700502 requeue_io(inode);
503 continue;
504 }
Jeff Laytond2caa3c2009-04-02 16:56:37 -0700505 /*
506 * Was this inode dirtied after sync_sb_inodes was called?
507 * This keeps sync from extra jobs and livelock.
508 */
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800509 if (inode_dirtied_after(inode, wbc->wb_start))
510 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511
Al Viroa4ffdde2010-06-02 17:38:30 -0400512 BUG_ON(inode->i_state & I_FREEING);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513 __iget(inode);
514 pages_skipped = wbc->pages_skipped;
Christoph Hellwig01c03192009-06-08 13:35:40 +0200515 writeback_single_inode(inode, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516 if (wbc->pages_skipped != pages_skipped) {
517 /*
518 * writeback is not making progress due to locked
519 * buffers. Skip this inode for now.
520 */
Andrew Mortonf57b9b72007-10-16 23:30:34 -0700521 redirty_tail(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522 }
523 spin_unlock(&inode_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524 iput(inode);
OGAWA Hirofumi4ffc8442006-03-25 03:07:44 -0800525 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526 spin_lock(&inode_lock);
Fengguang Wu8bc3be22008-02-04 22:29:36 -0800527 if (wbc->nr_to_write <= 0) {
528 wbc->more_io = 1;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800529 return 1;
Fengguang Wu8bc3be22008-02-04 22:29:36 -0800530 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200531 if (!list_empty(&wb->b_more_io))
Fengguang Wu8bc3be22008-02-04 22:29:36 -0800532 wbc->more_io = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533 }
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800534 /* b_io is empty */
535 return 1;
536}
Nick Piggin38f21972009-01-06 14:40:25 -0800537
Christoph Hellwig9c3a8ee2010-06-10 12:07:27 +0200538void writeback_inodes_wb(struct bdi_writeback *wb,
539 struct writeback_control *wbc)
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800540{
541 int ret = 0;
Jens Axboe9ecc2732009-09-24 15:25:11 +0200542
Jan Kara7624ee72010-08-09 17:20:03 -0700543 if (!wbc->wb_start)
544 wbc->wb_start = jiffies; /* livelock avoidance */
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800545 spin_lock(&inode_lock);
546 if (!wbc->for_kupdate || list_empty(&wb->b_io))
547 queue_io(wb, wbc->older_than_this);
548
549 while (!list_empty(&wb->b_io)) {
550 struct inode *inode = list_entry(wb->b_io.prev,
551 struct inode, i_list);
552 struct super_block *sb = inode->i_sb;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800553
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200554 if (!pin_sb_for_writeback(sb)) {
555 requeue_io(inode);
556 continue;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800557 }
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200558 ret = writeback_sb_inodes(sb, wb, wbc, false);
559 drop_super(sb);
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800560
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800561 if (ret)
562 break;
563 }
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200564 spin_unlock(&inode_lock);
565 /* Leave any unwritten inodes on b_io */
566}
567
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200568static void __writeback_inodes_sb(struct super_block *sb,
569 struct bdi_writeback *wb, struct writeback_control *wbc)
570{
571 WARN_ON(!rwsem_is_locked(&sb->s_umount));
572
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200573 spin_lock(&inode_lock);
574 if (!wbc->for_kupdate || list_empty(&wb->b_io))
575 queue_io(wb, wbc->older_than_this);
576 writeback_sb_inodes(sb, wb, wbc, true);
577 spin_unlock(&inode_lock);
578}
579
Jens Axboe03ba3782009-09-09 09:08:54 +0200580/*
581 * The maximum number of pages to writeout in a single bdi flush/kupdate
582 * operation. We do this so we don't hold I_SYNC against an inode for
583 * enormous amounts of time, which would block a userspace task which has
584 * been forced to throttle against that inode. Also, the code reevaluates
585 * the dirty each time it has written this many pages.
586 */
587#define MAX_WRITEBACK_PAGES 1024
588
589static inline bool over_bground_thresh(void)
590{
591 unsigned long background_thresh, dirty_thresh;
592
Wu Fengguang16c40422010-08-11 14:17:39 -0700593 global_dirty_limits(&background_thresh, &dirty_thresh);
Jens Axboe03ba3782009-09-09 09:08:54 +0200594
595 return (global_page_state(NR_FILE_DIRTY) +
596 global_page_state(NR_UNSTABLE_NFS) >= background_thresh);
597}
598
599/*
600 * Explicit flushing or periodic writeback of "old" data.
601 *
602 * Define "old": the first time one of an inode's pages is dirtied, we mark the
603 * dirtying-time in the inode's address_space. So this periodic writeback code
604 * just walks the superblock inode list, writing back any inodes which are
605 * older than a specific point in time.
606 *
607 * Try to run once per dirty_writeback_interval. But if a writeback event
608 * takes longer than a dirty_writeback_interval interval, then leave a
609 * one-second gap.
610 *
611 * older_than_this takes precedence over nr_to_write. So we'll only write back
612 * all dirty pages if they are all attached to "old" mappings.
613 */
Jens Axboec4a77a62009-09-16 15:18:25 +0200614static long wb_writeback(struct bdi_writeback *wb,
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200615 struct wb_writeback_work *work)
Jens Axboe03ba3782009-09-09 09:08:54 +0200616{
617 struct writeback_control wbc = {
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200618 .sync_mode = work->sync_mode,
Jens Axboe03ba3782009-09-09 09:08:54 +0200619 .older_than_this = NULL,
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200620 .for_kupdate = work->for_kupdate,
621 .for_background = work->for_background,
622 .range_cyclic = work->range_cyclic,
Jens Axboe03ba3782009-09-09 09:08:54 +0200623 };
624 unsigned long oldest_jif;
625 long wrote = 0;
Jan Karaa5989bd2009-09-16 19:22:48 +0200626 struct inode *inode;
Jens Axboe03ba3782009-09-09 09:08:54 +0200627
628 if (wbc.for_kupdate) {
629 wbc.older_than_this = &oldest_jif;
630 oldest_jif = jiffies -
631 msecs_to_jiffies(dirty_expire_interval * 10);
632 }
Jens Axboec4a77a62009-09-16 15:18:25 +0200633 if (!wbc.range_cyclic) {
634 wbc.range_start = 0;
635 wbc.range_end = LLONG_MAX;
636 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200637
Jan Kara7624ee72010-08-09 17:20:03 -0700638 wbc.wb_start = jiffies; /* livelock avoidance */
Jens Axboe03ba3782009-09-09 09:08:54 +0200639 for (;;) {
640 /*
Wu Fengguangd3ddec72009-09-23 20:33:40 +0800641 * Stop writeback when nr_pages has been consumed
Jens Axboe03ba3782009-09-09 09:08:54 +0200642 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200643 if (work->nr_pages <= 0)
Jens Axboe03ba3782009-09-09 09:08:54 +0200644 break;
645
646 /*
Wu Fengguangd3ddec72009-09-23 20:33:40 +0800647 * For background writeout, stop when we are below the
648 * background dirty threshold
Jens Axboe03ba3782009-09-09 09:08:54 +0200649 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200650 if (work->for_background && !over_bground_thresh())
Jens Axboe03ba3782009-09-09 09:08:54 +0200651 break;
652
653 wbc.more_io = 0;
Jens Axboe03ba3782009-09-09 09:08:54 +0200654 wbc.nr_to_write = MAX_WRITEBACK_PAGES;
655 wbc.pages_skipped = 0;
Dave Chinner028c2dd2010-07-07 13:24:07 +1000656
657 trace_wbc_writeback_start(&wbc, wb->bdi);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200658 if (work->sb)
659 __writeback_inodes_sb(work->sb, wb, &wbc);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200660 else
661 writeback_inodes_wb(wb, &wbc);
Dave Chinner028c2dd2010-07-07 13:24:07 +1000662 trace_wbc_writeback_written(&wbc, wb->bdi);
663
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200664 work->nr_pages -= MAX_WRITEBACK_PAGES - wbc.nr_to_write;
Jens Axboe03ba3782009-09-09 09:08:54 +0200665 wrote += MAX_WRITEBACK_PAGES - wbc.nr_to_write;
666
667 /*
Jens Axboe71fd05a2009-09-23 19:32:26 +0200668 * If we consumed everything, see if we have more
Jens Axboe03ba3782009-09-09 09:08:54 +0200669 */
Jens Axboe71fd05a2009-09-23 19:32:26 +0200670 if (wbc.nr_to_write <= 0)
671 continue;
672 /*
673 * Didn't write everything and we don't have more IO, bail
674 */
675 if (!wbc.more_io)
Jens Axboe03ba3782009-09-09 09:08:54 +0200676 break;
Jens Axboe71fd05a2009-09-23 19:32:26 +0200677 /*
678 * Did we write something? Try for more
679 */
680 if (wbc.nr_to_write < MAX_WRITEBACK_PAGES)
681 continue;
682 /*
683 * Nothing written. Wait for some inode to
684 * become available for writeback. Otherwise
685 * we'll just busyloop.
686 */
687 spin_lock(&inode_lock);
688 if (!list_empty(&wb->b_more_io)) {
689 inode = list_entry(wb->b_more_io.prev,
690 struct inode, i_list);
Dave Chinner028c2dd2010-07-07 13:24:07 +1000691 trace_wbc_writeback_wait(&wbc, wb->bdi);
Jens Axboe71fd05a2009-09-23 19:32:26 +0200692 inode_wait_for_writeback(inode);
Jens Axboe03ba3782009-09-09 09:08:54 +0200693 }
Jens Axboe71fd05a2009-09-23 19:32:26 +0200694 spin_unlock(&inode_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200695 }
696
697 return wrote;
698}
699
700/*
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200701 * Return the next wb_writeback_work struct that hasn't been processed yet.
Jens Axboe03ba3782009-09-09 09:08:54 +0200702 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200703static struct wb_writeback_work *
Minchan Kim08852b62010-08-03 12:51:16 +0200704get_next_work_item(struct backing_dev_info *bdi)
Jens Axboe03ba3782009-09-09 09:08:54 +0200705{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200706 struct wb_writeback_work *work = NULL;
Jens Axboe03ba3782009-09-09 09:08:54 +0200707
Artem Bityutskiy64677162010-07-25 14:29:22 +0300708 spin_lock_bh(&bdi->wb_lock);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200709 if (!list_empty(&bdi->work_list)) {
710 work = list_entry(bdi->work_list.next,
711 struct wb_writeback_work, list);
712 list_del_init(&work->list);
Jens Axboe03ba3782009-09-09 09:08:54 +0200713 }
Artem Bityutskiy64677162010-07-25 14:29:22 +0300714 spin_unlock_bh(&bdi->wb_lock);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200715 return work;
Jens Axboe03ba3782009-09-09 09:08:54 +0200716}
717
718static long wb_check_old_data_flush(struct bdi_writeback *wb)
719{
720 unsigned long expired;
721 long nr_pages;
722
Jens Axboe69b62d02010-05-17 12:51:03 +0200723 /*
724 * When set to zero, disable periodic writeback
725 */
726 if (!dirty_writeback_interval)
727 return 0;
728
Jens Axboe03ba3782009-09-09 09:08:54 +0200729 expired = wb->last_old_flush +
730 msecs_to_jiffies(dirty_writeback_interval * 10);
731 if (time_before(jiffies, expired))
732 return 0;
733
734 wb->last_old_flush = jiffies;
735 nr_pages = global_page_state(NR_FILE_DIRTY) +
736 global_page_state(NR_UNSTABLE_NFS) +
737 (inodes_stat.nr_inodes - inodes_stat.nr_unused);
738
Jens Axboec4a77a62009-09-16 15:18:25 +0200739 if (nr_pages) {
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200740 struct wb_writeback_work work = {
Jens Axboec4a77a62009-09-16 15:18:25 +0200741 .nr_pages = nr_pages,
742 .sync_mode = WB_SYNC_NONE,
743 .for_kupdate = 1,
744 .range_cyclic = 1,
745 };
746
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200747 return wb_writeback(wb, &work);
Jens Axboec4a77a62009-09-16 15:18:25 +0200748 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200749
750 return 0;
751}
752
753/*
754 * Retrieve work items and do the writeback they describe
755 */
756long wb_do_writeback(struct bdi_writeback *wb, int force_wait)
757{
758 struct backing_dev_info *bdi = wb->bdi;
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200759 struct wb_writeback_work *work;
Jens Axboec4a77a62009-09-16 15:18:25 +0200760 long wrote = 0;
Jens Axboe03ba3782009-09-09 09:08:54 +0200761
Minchan Kim08852b62010-08-03 12:51:16 +0200762 while ((work = get_next_work_item(bdi)) != NULL) {
Jens Axboe03ba3782009-09-09 09:08:54 +0200763 /*
764 * Override sync mode, in case we must wait for completion
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200765 * because this thread is exiting now.
Jens Axboe03ba3782009-09-09 09:08:54 +0200766 */
767 if (force_wait)
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200768 work->sync_mode = WB_SYNC_ALL;
769
Dave Chinner455b2862010-07-07 13:24:06 +1000770 trace_writeback_exec(bdi, work);
771
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200772 wrote += wb_writeback(wb, work);
Jens Axboe03ba3782009-09-09 09:08:54 +0200773
774 /*
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200775 * Notify the caller of completion if this is a synchronous
776 * work item, otherwise just free it.
Jens Axboe03ba3782009-09-09 09:08:54 +0200777 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200778 if (work->done)
779 complete(work->done);
780 else
781 kfree(work);
Jens Axboe03ba3782009-09-09 09:08:54 +0200782 }
783
784 /*
785 * Check for periodic writeback, kupdated() style
786 */
787 wrote += wb_check_old_data_flush(wb);
788
789 return wrote;
790}
791
792/*
793 * Handle writeback of dirty data for the device backed by this bdi. Also
794 * wakes up periodically and does kupdated style flushing.
795 */
Christoph Hellwig08243902010-06-19 23:08:22 +0200796int bdi_writeback_thread(void *data)
Jens Axboe03ba3782009-09-09 09:08:54 +0200797{
Christoph Hellwig08243902010-06-19 23:08:22 +0200798 struct bdi_writeback *wb = data;
799 struct backing_dev_info *bdi = wb->bdi;
Jens Axboe03ba3782009-09-09 09:08:54 +0200800 long pages_written;
801
Christoph Hellwig08243902010-06-19 23:08:22 +0200802 current->flags |= PF_FLUSHER | PF_SWAPWRITE;
803 set_freezable();
Artem Bityutskiyecd58402010-07-25 14:29:18 +0300804 wb->last_active = jiffies;
Christoph Hellwig08243902010-06-19 23:08:22 +0200805
806 /*
807 * Our parent may run at a different priority, just set us to normal
808 */
809 set_user_nice(current, 0);
810
Dave Chinner455b2862010-07-07 13:24:06 +1000811 trace_writeback_thread_start(bdi);
812
Jens Axboe03ba3782009-09-09 09:08:54 +0200813 while (!kthread_should_stop()) {
Artem Bityutskiy64677162010-07-25 14:29:22 +0300814 /*
815 * Remove own delayed wake-up timer, since we are already awake
816 * and we'll take care of the preriodic write-back.
817 */
818 del_timer(&wb->wakeup_timer);
819
Jens Axboe03ba3782009-09-09 09:08:54 +0200820 pages_written = wb_do_writeback(wb, 0);
821
Dave Chinner455b2862010-07-07 13:24:06 +1000822 trace_writeback_pages_written(pages_written);
Jens Axboe03ba3782009-09-09 09:08:54 +0200823
Jens Axboe03ba3782009-09-09 09:08:54 +0200824 if (pages_written)
Artem Bityutskiyecd58402010-07-25 14:29:18 +0300825 wb->last_active = jiffies;
Jens Axboe03ba3782009-09-09 09:08:54 +0200826
Artem Bityutskiy297252c2010-07-25 14:29:15 +0300827 set_current_state(TASK_INTERRUPTIBLE);
828 if (!list_empty(&bdi->work_list)) {
829 __set_current_state(TASK_RUNNING);
830 continue;
Jens Axboe03ba3782009-09-09 09:08:54 +0200831 }
832
Artem Bityutskiy253c34e2010-07-25 14:29:21 +0300833 if (wb_has_dirty_io(wb) && dirty_writeback_interval)
Artem Bityutskiyfff5b852010-07-25 14:29:20 +0300834 schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10));
Artem Bityutskiy253c34e2010-07-25 14:29:21 +0300835 else {
836 /*
837 * We have nothing to do, so can go sleep without any
838 * timeout and save power. When a work is queued or
839 * something is made dirty - we will be woken up.
840 */
Artem Bityutskiy297252c2010-07-25 14:29:15 +0300841 schedule();
Jens Axboef9eadbb2010-05-18 14:31:45 +0200842 }
Jens Axboe69b62d02010-05-17 12:51:03 +0200843
Jens Axboe03ba3782009-09-09 09:08:54 +0200844 try_to_freeze();
845 }
846
Artem Bityutskiyfff5b852010-07-25 14:29:20 +0300847 /* Flush any work that raced with us exiting */
Christoph Hellwig08243902010-06-19 23:08:22 +0200848 if (!list_empty(&bdi->work_list))
849 wb_do_writeback(wb, 1);
Dave Chinner455b2862010-07-07 13:24:06 +1000850
851 trace_writeback_thread_stop(bdi);
Jens Axboe03ba3782009-09-09 09:08:54 +0200852 return 0;
853}
854
Christoph Hellwig08243902010-06-19 23:08:22 +0200855
Jens Axboe03ba3782009-09-09 09:08:54 +0200856/*
Jens Axboe03ba3782009-09-09 09:08:54 +0200857 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
858 * the whole world.
859 */
860void wakeup_flusher_threads(long nr_pages)
861{
Christoph Hellwigb8c2f342010-06-08 18:15:07 +0200862 struct backing_dev_info *bdi;
Christoph Hellwigb8c2f342010-06-08 18:15:07 +0200863
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200864 if (!nr_pages) {
865 nr_pages = global_page_state(NR_FILE_DIRTY) +
Jens Axboe03ba3782009-09-09 09:08:54 +0200866 global_page_state(NR_UNSTABLE_NFS);
Christoph Hellwigb8c2f342010-06-08 18:15:07 +0200867 }
868
869 rcu_read_lock();
870 list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) {
871 if (!bdi_has_dirty_io(bdi))
872 continue;
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200873 __bdi_start_writeback(bdi, nr_pages, false, false);
Christoph Hellwigb8c2f342010-06-08 18:15:07 +0200874 }
875 rcu_read_unlock();
Jens Axboe03ba3782009-09-09 09:08:54 +0200876}
877
878static noinline void block_dump___mark_inode_dirty(struct inode *inode)
879{
880 if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) {
881 struct dentry *dentry;
882 const char *name = "?";
883
884 dentry = d_find_alias(inode);
885 if (dentry) {
886 spin_lock(&dentry->d_lock);
887 name = (const char *) dentry->d_name.name;
888 }
889 printk(KERN_DEBUG
890 "%s(%d): dirtied inode %lu (%s) on %s\n",
891 current->comm, task_pid_nr(current), inode->i_ino,
892 name, inode->i_sb->s_id);
893 if (dentry) {
894 spin_unlock(&dentry->d_lock);
895 dput(dentry);
896 }
897 }
898}
899
900/**
901 * __mark_inode_dirty - internal function
902 * @inode: inode to mark
903 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
904 * Mark an inode as dirty. Callers should use mark_inode_dirty or
905 * mark_inode_dirty_sync.
906 *
907 * Put the inode on the super block's dirty list.
908 *
909 * CAREFUL! We mark it dirty unconditionally, but move it onto the
910 * dirty list only if it is hashed or if it refers to a blockdev.
911 * If it was not hashed, it will never be added to the dirty list
912 * even if it is later hashed, as it will have been marked dirty already.
913 *
914 * In short, make sure you hash any inodes _before_ you start marking
915 * them dirty.
916 *
917 * This function *must* be atomic for the I_DIRTY_PAGES case -
918 * set_page_dirty() is called under spinlock in several places.
919 *
920 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
921 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
922 * the kernel-internal blockdev inode represents the dirtying time of the
923 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
924 * page->mapping->host, so the page-dirtying time is recorded in the internal
925 * blockdev inode.
926 */
927void __mark_inode_dirty(struct inode *inode, int flags)
928{
929 struct super_block *sb = inode->i_sb;
Artem Bityutskiy253c34e2010-07-25 14:29:21 +0300930 struct backing_dev_info *bdi = NULL;
931 bool wakeup_bdi = false;
Jens Axboe03ba3782009-09-09 09:08:54 +0200932
933 /*
934 * Don't do this for I_DIRTY_PAGES - that doesn't actually
935 * dirty the inode itself
936 */
937 if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
938 if (sb->s_op->dirty_inode)
939 sb->s_op->dirty_inode(inode);
940 }
941
942 /*
943 * make sure that changes are seen by all cpus before we test i_state
944 * -- mikulas
945 */
946 smp_mb();
947
948 /* avoid the locking if we can */
949 if ((inode->i_state & flags) == flags)
950 return;
951
952 if (unlikely(block_dump))
953 block_dump___mark_inode_dirty(inode);
954
955 spin_lock(&inode_lock);
956 if ((inode->i_state & flags) != flags) {
957 const int was_dirty = inode->i_state & I_DIRTY;
958
959 inode->i_state |= flags;
960
961 /*
962 * If the inode is being synced, just update its dirty state.
963 * The unlocker will place the inode on the appropriate
964 * superblock list, based upon its state.
965 */
966 if (inode->i_state & I_SYNC)
967 goto out;
968
969 /*
970 * Only add valid (hashed) inodes to the superblock's
971 * dirty list. Add blockdev inodes as well.
972 */
973 if (!S_ISBLK(inode->i_mode)) {
974 if (hlist_unhashed(&inode->i_hash))
975 goto out;
976 }
Al Viroa4ffdde2010-06-02 17:38:30 -0400977 if (inode->i_state & I_FREEING)
Jens Axboe03ba3782009-09-09 09:08:54 +0200978 goto out;
979
980 /*
981 * If the inode was already on b_dirty/b_io/b_more_io, don't
982 * reposition it (that would break b_dirty time-ordering).
983 */
984 if (!was_dirty) {
Artem Bityutskiy253c34e2010-07-25 14:29:21 +0300985 bdi = inode_to_bdi(inode);
Jens Axboe500b0672009-09-09 09:10:25 +0200986
Artem Bityutskiy253c34e2010-07-25 14:29:21 +0300987 if (bdi_cap_writeback_dirty(bdi)) {
988 WARN(!test_bit(BDI_registered, &bdi->state),
989 "bdi-%s not registered\n", bdi->name);
990
991 /*
992 * If this is the first dirty inode for this
993 * bdi, we have to wake-up the corresponding
994 * bdi thread to make sure background
995 * write-back happens later.
996 */
997 if (!wb_has_dirty_io(&bdi->wb))
998 wakeup_bdi = true;
Jens Axboe500b0672009-09-09 09:10:25 +0200999 }
Jens Axboe03ba3782009-09-09 09:08:54 +02001000
1001 inode->dirtied_when = jiffies;
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001002 list_move(&inode->i_list, &bdi->wb.b_dirty);
Jens Axboe03ba3782009-09-09 09:08:54 +02001003 }
1004 }
1005out:
1006 spin_unlock(&inode_lock);
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001007
1008 if (wakeup_bdi)
Artem Bityutskiy64677162010-07-25 14:29:22 +03001009 bdi_wakeup_thread_delayed(bdi);
Jens Axboe03ba3782009-09-09 09:08:54 +02001010}
1011EXPORT_SYMBOL(__mark_inode_dirty);
1012
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001013/*
1014 * Write out a superblock's list of dirty inodes. A wait will be performed
1015 * upon no inodes, all inodes or the final one, depending upon sync_mode.
1016 *
1017 * If older_than_this is non-NULL, then only write out inodes which
1018 * had their first dirtying at a time earlier than *older_than_this.
1019 *
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001020 * If `bdi' is non-zero then we're being asked to writeback a specific queue.
1021 * This function assumes that the blockdev superblock's inodes are backed by
1022 * a variety of queues, so all inodes are searched. For other superblocks,
1023 * assume that all inodes are backed by the same queue.
1024 *
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001025 * The inodes to be written are parked on bdi->b_io. They are moved back onto
1026 * bdi->b_dirty as they are selected for writing. This way, none can be missed
1027 * on the writer throttling path, and we get decent balancing between many
1028 * throttled threads: we don't want them all piling up on inode_sync_wait.
1029 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001030static void wait_sb_inodes(struct super_block *sb)
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001031{
Jens Axboe03ba3782009-09-09 09:08:54 +02001032 struct inode *inode, *old_inode = NULL;
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001033
Jens Axboe03ba3782009-09-09 09:08:54 +02001034 /*
1035 * We need to be protected against the filesystem going from
1036 * r/o to r/w or vice versa.
1037 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001038 WARN_ON(!rwsem_is_locked(&sb->s_umount));
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001039
Jens Axboe03ba3782009-09-09 09:08:54 +02001040 spin_lock(&inode_lock);
1041
1042 /*
1043 * Data integrity sync. Must wait for all pages under writeback,
1044 * because there may have been pages dirtied before our sync
1045 * call, but which had writeout started before we write it out.
1046 * In which case, the inode may not be on the dirty list, but
1047 * we still have to wait for that writeout.
1048 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001049 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
Jens Axboe03ba3782009-09-09 09:08:54 +02001050 struct address_space *mapping;
1051
Al Viroa4ffdde2010-06-02 17:38:30 -04001052 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW))
Jens Axboe03ba3782009-09-09 09:08:54 +02001053 continue;
1054 mapping = inode->i_mapping;
1055 if (mapping->nrpages == 0)
1056 continue;
1057 __iget(inode);
1058 spin_unlock(&inode_lock);
1059 /*
1060 * We hold a reference to 'inode' so it couldn't have
1061 * been removed from s_inodes list while we dropped the
1062 * inode_lock. We cannot iput the inode now as we can
1063 * be holding the last reference and we cannot iput it
1064 * under inode_lock. So we keep the reference and iput
1065 * it later.
1066 */
1067 iput(old_inode);
1068 old_inode = inode;
1069
1070 filemap_fdatawait(mapping);
1071
1072 cond_resched();
Nick Piggin38f21972009-01-06 14:40:25 -08001073
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001074 spin_lock(&inode_lock);
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001075 }
Jens Axboe03ba3782009-09-09 09:08:54 +02001076 spin_unlock(&inode_lock);
1077 iput(old_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078}
1079
Jens Axboed8a85592009-09-02 12:34:32 +02001080/**
1081 * writeback_inodes_sb - writeback dirty inodes from given super_block
1082 * @sb: the superblock
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083 *
Jens Axboed8a85592009-09-02 12:34:32 +02001084 * Start writeback on some inodes on this super_block. No guarantees are made
1085 * on how many (if any) will be written, and this function does not wait
1086 * for IO completion of submitted IO. The number of pages submitted is
1087 * returned.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001089void writeback_inodes_sb(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090{
Jens Axboe0e3c9a22010-06-01 11:08:43 +02001091 unsigned long nr_dirty = global_page_state(NR_FILE_DIRTY);
1092 unsigned long nr_unstable = global_page_state(NR_UNSTABLE_NFS);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001093 DECLARE_COMPLETION_ONSTACK(done);
1094 struct wb_writeback_work work = {
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001095 .sb = sb,
1096 .sync_mode = WB_SYNC_NONE,
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001097 .done = &done,
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001098 };
Jens Axboe0e3c9a22010-06-01 11:08:43 +02001099
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001100 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1101
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001102 work.nr_pages = nr_dirty + nr_unstable +
Jens Axboe0e3c9a22010-06-01 11:08:43 +02001103 (inodes_stat.nr_inodes - inodes_stat.nr_unused);
1104
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001105 bdi_queue_work(sb->s_bdi, &work);
1106 wait_for_completion(&done);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107}
Jens Axboed8a85592009-09-02 12:34:32 +02001108EXPORT_SYMBOL(writeback_inodes_sb);
1109
1110/**
Eric Sandeen17bd55d2009-12-23 07:57:07 -05001111 * writeback_inodes_sb_if_idle - start writeback if none underway
1112 * @sb: the superblock
1113 *
1114 * Invoke writeback_inodes_sb if no writeback is currently underway.
1115 * Returns 1 if writeback was started, 0 if not.
1116 */
1117int writeback_inodes_sb_if_idle(struct super_block *sb)
1118{
1119 if (!writeback_in_progress(sb->s_bdi)) {
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001120 down_read(&sb->s_umount);
Eric Sandeen17bd55d2009-12-23 07:57:07 -05001121 writeback_inodes_sb(sb);
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001122 up_read(&sb->s_umount);
Eric Sandeen17bd55d2009-12-23 07:57:07 -05001123 return 1;
1124 } else
1125 return 0;
1126}
1127EXPORT_SYMBOL(writeback_inodes_sb_if_idle);
1128
1129/**
Jens Axboed8a85592009-09-02 12:34:32 +02001130 * sync_inodes_sb - sync sb inode pages
1131 * @sb: the superblock
1132 *
1133 * This function writes and waits on any dirty inode belonging to this
1134 * super_block. The number of pages synced is returned.
1135 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001136void sync_inodes_sb(struct super_block *sb)
Jens Axboed8a85592009-09-02 12:34:32 +02001137{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001138 DECLARE_COMPLETION_ONSTACK(done);
1139 struct wb_writeback_work work = {
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001140 .sb = sb,
1141 .sync_mode = WB_SYNC_ALL,
1142 .nr_pages = LONG_MAX,
1143 .range_cyclic = 0,
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001144 .done = &done,
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001145 };
1146
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001147 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1148
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001149 bdi_queue_work(sb->s_bdi, &work);
1150 wait_for_completion(&done);
1151
Jens Axboeb6e51312009-09-16 15:13:54 +02001152 wait_sb_inodes(sb);
Jens Axboed8a85592009-09-02 12:34:32 +02001153}
1154EXPORT_SYMBOL(sync_inodes_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156/**
Andrea Arcangeli7f04c262005-10-30 15:03:05 -08001157 * write_inode_now - write an inode to disk
1158 * @inode: inode to write to disk
1159 * @sync: whether the write should be synchronous or not
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160 *
Andrea Arcangeli7f04c262005-10-30 15:03:05 -08001161 * This function commits an inode to disk immediately if it is dirty. This is
1162 * primarily needed by knfsd.
1163 *
1164 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166int write_inode_now(struct inode *inode, int sync)
1167{
1168 int ret;
1169 struct writeback_control wbc = {
1170 .nr_to_write = LONG_MAX,
Mike Galbraith18914b12008-02-08 04:20:23 -08001171 .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -07001172 .range_start = 0,
1173 .range_end = LLONG_MAX,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174 };
1175
1176 if (!mapping_cap_writeback_dirty(inode->i_mapping))
Andrew Morton49364ce2005-11-07 00:59:15 -08001177 wbc.nr_to_write = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178
1179 might_sleep();
1180 spin_lock(&inode_lock);
Christoph Hellwig01c03192009-06-08 13:35:40 +02001181 ret = writeback_single_inode(inode, &wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182 spin_unlock(&inode_lock);
1183 if (sync)
Joern Engel1c0eeaf2007-10-16 23:30:44 -07001184 inode_sync_wait(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185 return ret;
1186}
1187EXPORT_SYMBOL(write_inode_now);
1188
1189/**
1190 * sync_inode - write an inode and its pages to disk.
1191 * @inode: the inode to sync
1192 * @wbc: controls the writeback mode
1193 *
1194 * sync_inode() will write an inode and its pages to disk. It will also
1195 * correctly update the inode on its superblock's dirty inode lists and will
1196 * update inode->i_state.
1197 *
1198 * The caller must have a ref on the inode.
1199 */
1200int sync_inode(struct inode *inode, struct writeback_control *wbc)
1201{
1202 int ret;
1203
1204 spin_lock(&inode_lock);
Christoph Hellwig01c03192009-06-08 13:35:40 +02001205 ret = writeback_single_inode(inode, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206 spin_unlock(&inode_lock);
1207 return ret;
1208}
1209EXPORT_SYMBOL(sync_inode);