blob: 81e086d8aa5733c7a86a966827bf7ea18d29baa4 [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{
Jan Kara81d73a32010-08-11 14:17:44 -070071 return test_bit(BDI_writeback_running, &bdi->state);
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 Laytond2caa3c52009-04-02 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 Laytond2caa3c52009-04-02 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 Laytond2caa3c52009-04-02 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.
Wu Fengguang4ea879b2010-08-11 14:17:42 -0700252 * Before
253 * newly dirtied b_dirty b_io b_more_io
254 * =============> gf edc BA
255 * After
256 * newly dirtied b_dirty b_io b_more_io
257 * =============> g fBAedc
258 * |
259 * +--> dequeue for IO
Fengguang Wu2c136572007-10-16 23:30:39 -0700260 */
Jens Axboe03ba3782009-09-09 09:08:54 +0200261static void queue_io(struct bdi_writeback *wb, unsigned long *older_than_this)
Fengguang Wu2c136572007-10-16 23:30:39 -0700262{
Wu Fengguang4ea879b2010-08-11 14:17:42 -0700263 list_splice_init(&wb->b_more_io, &wb->b_io);
Jens Axboe03ba3782009-09-09 09:08:54 +0200264 move_expired_inodes(&wb->b_dirty, &wb->b_io, older_than_this);
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200265}
266
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100267static int write_inode(struct inode *inode, struct writeback_control *wbc)
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200268{
Jens Axboe03ba3782009-09-09 09:08:54 +0200269 if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode))
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100270 return inode->i_sb->s_op->write_inode(inode, wbc);
Jens Axboe03ba3782009-09-09 09:08:54 +0200271 return 0;
Fengguang Wu2c136572007-10-16 23:30:39 -0700272}
273
274/*
Christoph Hellwig01c03192009-06-08 13:35:40 +0200275 * Wait for writeback on an inode to complete.
276 */
277static void inode_wait_for_writeback(struct inode *inode)
278{
279 DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC);
280 wait_queue_head_t *wqh;
281
282 wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
Richard Kennedy58a9d3d82010-05-24 14:32:38 -0700283 while (inode->i_state & I_SYNC) {
Christoph Hellwig01c03192009-06-08 13:35:40 +0200284 spin_unlock(&inode_lock);
285 __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE);
286 spin_lock(&inode_lock);
Richard Kennedy58a9d3d82010-05-24 14:32:38 -0700287 }
Christoph Hellwig01c03192009-06-08 13:35:40 +0200288}
289
290/*
291 * Write out an inode's dirty pages. Called under inode_lock. Either the
292 * caller has ref on the inode (either via __iget or via syscall against an fd)
293 * or the inode has I_WILL_FREE set (via generic_forget_inode)
294 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 * If `wait' is set, wait on the writeout.
296 *
297 * The whole writeout design is quite complex and fragile. We want to avoid
298 * starvation of particular inodes when others are being redirtied, prevent
299 * livelocks, etc.
300 *
301 * Called under inode_lock.
302 */
303static int
Christoph Hellwig01c03192009-06-08 13:35:40 +0200304writeback_single_inode(struct inode *inode, struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306 struct address_space *mapping = inode->i_mapping;
Christoph Hellwig01c03192009-06-08 13:35:40 +0200307 unsigned dirty;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308 int ret;
309
Christoph Hellwig01c03192009-06-08 13:35:40 +0200310 if (!atomic_read(&inode->i_count))
311 WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING)));
312 else
313 WARN_ON(inode->i_state & I_WILL_FREE);
314
315 if (inode->i_state & I_SYNC) {
316 /*
317 * If this inode is locked for writeback and we are not doing
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200318 * writeback-for-data-integrity, move it to b_more_io so that
Christoph Hellwig01c03192009-06-08 13:35:40 +0200319 * writeback can proceed with the other inodes on s_io.
320 *
321 * We'll have another go at writing back this inode when we
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200322 * completed a full scan of b_io.
Christoph Hellwig01c03192009-06-08 13:35:40 +0200323 */
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100324 if (wbc->sync_mode != WB_SYNC_ALL) {
Christoph Hellwig01c03192009-06-08 13:35:40 +0200325 requeue_io(inode);
326 return 0;
327 }
328
329 /*
330 * It's a data-integrity sync. We must wait.
331 */
332 inode_wait_for_writeback(inode);
333 }
334
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700335 BUG_ON(inode->i_state & I_SYNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336
Dmitry Monakhov5547e8a2010-05-07 13:35:44 +0400337 /* Set I_SYNC, reset I_DIRTY_PAGES */
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700338 inode->i_state |= I_SYNC;
Dmitry Monakhov5547e8a2010-05-07 13:35:44 +0400339 inode->i_state &= ~I_DIRTY_PAGES;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340 spin_unlock(&inode_lock);
341
342 ret = do_writepages(mapping, wbc);
343
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100344 /*
345 * Make sure to wait on the data before writing out the metadata.
346 * This is important for filesystems that modify metadata on data
347 * I/O completion.
348 */
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100349 if (wbc->sync_mode == WB_SYNC_ALL) {
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100350 int err = filemap_fdatawait(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351 if (ret == 0)
352 ret = err;
353 }
354
Dmitry Monakhov5547e8a2010-05-07 13:35:44 +0400355 /*
356 * Some filesystems may redirty the inode during the writeback
357 * due to delalloc, clear dirty metadata flags right before
358 * write_inode()
359 */
360 spin_lock(&inode_lock);
361 dirty = inode->i_state & I_DIRTY;
362 inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC);
363 spin_unlock(&inode_lock);
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100364 /* Don't write the inode if only I_DIRTY_PAGES was set */
365 if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100366 int err = write_inode(inode, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367 if (ret == 0)
368 ret = err;
369 }
370
371 spin_lock(&inode_lock);
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700372 inode->i_state &= ~I_SYNC;
Al Viroa4ffdde2010-06-02 17:38:30 -0400373 if (!(inode->i_state & I_FREEING)) {
Wu Fengguang23539af2010-08-11 14:17:41 -0700374 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 /*
376 * We didn't write back all the pages. nfs_writepages()
Wu Fengguanga50aeb42010-08-11 14:17:43 -0700377 * sometimes bales out without doing anything.
Andrew Morton1b43ef92007-10-16 23:30:35 -0700378 */
Wu Fengguanga50aeb42010-08-11 14:17:43 -0700379 inode->i_state |= I_DIRTY_PAGES;
380 if (wbc->nr_to_write <= 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381 /*
Wu Fengguanga50aeb42010-08-11 14:17:43 -0700382 * slice used up: queue for next turn
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383 */
Wu Fengguanga50aeb42010-08-11 14:17:43 -0700384 requeue_io(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385 } else {
386 /*
Wu Fengguanga50aeb42010-08-11 14:17:43 -0700387 * Writeback blocked by something other than
388 * congestion. Delay the inode for some time to
389 * avoid spinning on the CPU (100% iowait)
390 * retrying writeback of the dirty page/inode
391 * that cannot be performed immediately.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392 */
Andrew Morton1b43ef92007-10-16 23:30:35 -0700393 redirty_tail(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394 }
Wu Fengguang23539af2010-08-11 14:17:41 -0700395 } else if (inode->i_state & I_DIRTY) {
396 /*
397 * Filesystems can dirty the inode during writeback
398 * operations, such as delayed allocation during
399 * submission or metadata updates after data IO
400 * completion.
401 */
402 redirty_tail(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403 } else if (atomic_read(&inode->i_count)) {
404 /*
405 * The inode is clean, inuse
406 */
407 list_move(&inode->i_list, &inode_in_use);
408 } else {
409 /*
410 * The inode is clean, unused
411 */
412 list_move(&inode->i_list, &inode_unused);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413 }
414 }
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700415 inode_sync_complete(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416 return ret;
417}
418
Jens Axboe03ba3782009-09-09 09:08:54 +0200419/*
Christoph Hellwigd19de7e2010-06-08 18:14:58 +0200420 * For background writeback the caller does not have the sb pinned
Jens Axboe03ba3782009-09-09 09:08:54 +0200421 * before calling writeback. So make sure that we do pin it, so it doesn't
422 * go away while we are writing inodes from it.
Jens Axboe03ba3782009-09-09 09:08:54 +0200423 */
Christoph Hellwigd19de7e2010-06-08 18:14:58 +0200424static bool pin_sb_for_writeback(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425{
Jens Axboe03ba3782009-09-09 09:08:54 +0200426 spin_lock(&sb_lock);
Christoph Hellwig29cb48592010-06-09 15:31:01 +0200427 if (list_empty(&sb->s_instances)) {
428 spin_unlock(&sb_lock);
429 return false;
430 }
431
Jens Axboe03ba3782009-09-09 09:08:54 +0200432 sb->s_count++;
Christoph Hellwig29cb48592010-06-09 15:31:01 +0200433 spin_unlock(&sb_lock);
434
Jens Axboe03ba3782009-09-09 09:08:54 +0200435 if (down_read_trylock(&sb->s_umount)) {
Christoph Hellwig29cb48592010-06-09 15:31:01 +0200436 if (sb->s_root)
Christoph Hellwigd19de7e2010-06-08 18:14:58 +0200437 return true;
Jens Axboe03ba3782009-09-09 09:08:54 +0200438 up_read(&sb->s_umount);
439 }
Christoph Hellwig29cb48592010-06-09 15:31:01 +0200440
441 put_super(sb);
Christoph Hellwigd19de7e2010-06-08 18:14:58 +0200442 return false;
Jens Axboe03ba3782009-09-09 09:08:54 +0200443}
444
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800445/*
446 * Write a portion of b_io inodes which belong to @sb.
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200447 *
448 * If @only_this_sb is true, then find and write all such
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800449 * inodes. Otherwise write only ones which go sequentially
450 * in reverse order.
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200451 *
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800452 * Return 1, if the caller writeback routine should be
453 * interrupted. Otherwise return 0.
454 */
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200455static int writeback_sb_inodes(struct super_block *sb, struct bdi_writeback *wb,
456 struct writeback_control *wbc, bool only_this_sb)
Jens Axboe03ba3782009-09-09 09:08:54 +0200457{
Jens Axboe03ba3782009-09-09 09:08:54 +0200458 while (!list_empty(&wb->b_io)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 long pages_skipped;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800460 struct inode *inode = list_entry(wb->b_io.prev,
461 struct inode, i_list);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200462
463 if (inode->i_sb != sb) {
464 if (only_this_sb) {
465 /*
466 * We only want to write back data for this
467 * superblock, move all inodes not belonging
468 * to it back onto the dirty list.
469 */
470 redirty_tail(inode);
471 continue;
472 }
473
474 /*
475 * The inode belongs to a different superblock.
476 * Bounce back to the caller to unpin this and
477 * pin the next superblock.
478 */
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800479 return 0;
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200480 }
481
Wu Fengguang84a89242009-06-16 15:33:17 -0700482 if (inode->i_state & (I_NEW | I_WILL_FREE)) {
Nick Piggin7ef0d732009-03-12 14:31:38 -0700483 requeue_io(inode);
484 continue;
485 }
Jeff Laytond2caa3c52009-04-02 16:56:37 -0700486 /*
487 * Was this inode dirtied after sync_sb_inodes was called?
488 * This keeps sync from extra jobs and livelock.
489 */
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800490 if (inode_dirtied_after(inode, wbc->wb_start))
491 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492
Al Viroa4ffdde2010-06-02 17:38:30 -0400493 BUG_ON(inode->i_state & I_FREEING);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494 __iget(inode);
495 pages_skipped = wbc->pages_skipped;
Christoph Hellwig01c03192009-06-08 13:35:40 +0200496 writeback_single_inode(inode, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497 if (wbc->pages_skipped != pages_skipped) {
498 /*
499 * writeback is not making progress due to locked
500 * buffers. Skip this inode for now.
501 */
Andrew Mortonf57b9b72007-10-16 23:30:34 -0700502 redirty_tail(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503 }
504 spin_unlock(&inode_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505 iput(inode);
OGAWA Hirofumi4ffc8442006-03-25 03:07:44 -0800506 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507 spin_lock(&inode_lock);
Fengguang Wu8bc3be22008-02-04 22:29:36 -0800508 if (wbc->nr_to_write <= 0) {
509 wbc->more_io = 1;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800510 return 1;
Fengguang Wu8bc3be22008-02-04 22:29:36 -0800511 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200512 if (!list_empty(&wb->b_more_io))
Fengguang Wu8bc3be22008-02-04 22:29:36 -0800513 wbc->more_io = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 }
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800515 /* b_io is empty */
516 return 1;
517}
Nick Piggin38f21972009-01-06 14:40:25 -0800518
Christoph Hellwig9c3a8ee2010-06-10 12:07:27 +0200519void writeback_inodes_wb(struct bdi_writeback *wb,
520 struct writeback_control *wbc)
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800521{
522 int ret = 0;
Jens Axboe9ecc2732009-09-24 15:25:11 +0200523
Jan Kara7624ee72010-08-09 17:20:03 -0700524 if (!wbc->wb_start)
525 wbc->wb_start = jiffies; /* livelock avoidance */
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800526 spin_lock(&inode_lock);
527 if (!wbc->for_kupdate || list_empty(&wb->b_io))
528 queue_io(wb, wbc->older_than_this);
529
530 while (!list_empty(&wb->b_io)) {
531 struct inode *inode = list_entry(wb->b_io.prev,
532 struct inode, i_list);
533 struct super_block *sb = inode->i_sb;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800534
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200535 if (!pin_sb_for_writeback(sb)) {
536 requeue_io(inode);
537 continue;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800538 }
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200539 ret = writeback_sb_inodes(sb, wb, wbc, false);
540 drop_super(sb);
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800541
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800542 if (ret)
543 break;
544 }
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200545 spin_unlock(&inode_lock);
546 /* Leave any unwritten inodes on b_io */
547}
548
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200549static void __writeback_inodes_sb(struct super_block *sb,
550 struct bdi_writeback *wb, struct writeback_control *wbc)
551{
552 WARN_ON(!rwsem_is_locked(&sb->s_umount));
553
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200554 spin_lock(&inode_lock);
555 if (!wbc->for_kupdate || list_empty(&wb->b_io))
556 queue_io(wb, wbc->older_than_this);
557 writeback_sb_inodes(sb, wb, wbc, true);
558 spin_unlock(&inode_lock);
559}
560
Jens Axboe03ba3782009-09-09 09:08:54 +0200561/*
562 * The maximum number of pages to writeout in a single bdi flush/kupdate
563 * operation. We do this so we don't hold I_SYNC against an inode for
564 * enormous amounts of time, which would block a userspace task which has
565 * been forced to throttle against that inode. Also, the code reevaluates
566 * the dirty each time it has written this many pages.
567 */
568#define MAX_WRITEBACK_PAGES 1024
569
570static inline bool over_bground_thresh(void)
571{
572 unsigned long background_thresh, dirty_thresh;
573
Wu Fengguang16c40422010-08-11 14:17:39 -0700574 global_dirty_limits(&background_thresh, &dirty_thresh);
Jens Axboe03ba3782009-09-09 09:08:54 +0200575
576 return (global_page_state(NR_FILE_DIRTY) +
577 global_page_state(NR_UNSTABLE_NFS) >= background_thresh);
578}
579
580/*
581 * Explicit flushing or periodic writeback of "old" data.
582 *
583 * Define "old": the first time one of an inode's pages is dirtied, we mark the
584 * dirtying-time in the inode's address_space. So this periodic writeback code
585 * just walks the superblock inode list, writing back any inodes which are
586 * older than a specific point in time.
587 *
588 * Try to run once per dirty_writeback_interval. But if a writeback event
589 * takes longer than a dirty_writeback_interval interval, then leave a
590 * one-second gap.
591 *
592 * older_than_this takes precedence over nr_to_write. So we'll only write back
593 * all dirty pages if they are all attached to "old" mappings.
594 */
Jens Axboec4a77a62009-09-16 15:18:25 +0200595static long wb_writeback(struct bdi_writeback *wb,
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200596 struct wb_writeback_work *work)
Jens Axboe03ba3782009-09-09 09:08:54 +0200597{
598 struct writeback_control wbc = {
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200599 .sync_mode = work->sync_mode,
Jens Axboe03ba3782009-09-09 09:08:54 +0200600 .older_than_this = NULL,
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200601 .for_kupdate = work->for_kupdate,
602 .for_background = work->for_background,
603 .range_cyclic = work->range_cyclic,
Jens Axboe03ba3782009-09-09 09:08:54 +0200604 };
605 unsigned long oldest_jif;
606 long wrote = 0;
Jan Karaa5989bd2009-09-16 19:22:48 +0200607 struct inode *inode;
Jens Axboe03ba3782009-09-09 09:08:54 +0200608
609 if (wbc.for_kupdate) {
610 wbc.older_than_this = &oldest_jif;
611 oldest_jif = jiffies -
612 msecs_to_jiffies(dirty_expire_interval * 10);
613 }
Jens Axboec4a77a62009-09-16 15:18:25 +0200614 if (!wbc.range_cyclic) {
615 wbc.range_start = 0;
616 wbc.range_end = LLONG_MAX;
617 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200618
Jan Kara7624ee72010-08-09 17:20:03 -0700619 wbc.wb_start = jiffies; /* livelock avoidance */
Jens Axboe03ba3782009-09-09 09:08:54 +0200620 for (;;) {
621 /*
Wu Fengguangd3ddec72009-09-23 20:33:40 +0800622 * Stop writeback when nr_pages has been consumed
Jens Axboe03ba3782009-09-09 09:08:54 +0200623 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200624 if (work->nr_pages <= 0)
Jens Axboe03ba3782009-09-09 09:08:54 +0200625 break;
626
627 /*
Wu Fengguangd3ddec72009-09-23 20:33:40 +0800628 * For background writeout, stop when we are below the
629 * background dirty threshold
Jens Axboe03ba3782009-09-09 09:08:54 +0200630 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200631 if (work->for_background && !over_bground_thresh())
Jens Axboe03ba3782009-09-09 09:08:54 +0200632 break;
633
634 wbc.more_io = 0;
Jens Axboe03ba3782009-09-09 09:08:54 +0200635 wbc.nr_to_write = MAX_WRITEBACK_PAGES;
636 wbc.pages_skipped = 0;
Dave Chinner028c2dd2010-07-07 13:24:07 +1000637
638 trace_wbc_writeback_start(&wbc, wb->bdi);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200639 if (work->sb)
640 __writeback_inodes_sb(work->sb, wb, &wbc);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200641 else
642 writeback_inodes_wb(wb, &wbc);
Dave Chinner028c2dd2010-07-07 13:24:07 +1000643 trace_wbc_writeback_written(&wbc, wb->bdi);
644
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200645 work->nr_pages -= MAX_WRITEBACK_PAGES - wbc.nr_to_write;
Jens Axboe03ba3782009-09-09 09:08:54 +0200646 wrote += MAX_WRITEBACK_PAGES - wbc.nr_to_write;
647
648 /*
Jens Axboe71fd05a2009-09-23 19:32:26 +0200649 * If we consumed everything, see if we have more
Jens Axboe03ba3782009-09-09 09:08:54 +0200650 */
Jens Axboe71fd05a2009-09-23 19:32:26 +0200651 if (wbc.nr_to_write <= 0)
652 continue;
653 /*
654 * Didn't write everything and we don't have more IO, bail
655 */
656 if (!wbc.more_io)
Jens Axboe03ba3782009-09-09 09:08:54 +0200657 break;
Jens Axboe71fd05a2009-09-23 19:32:26 +0200658 /*
659 * Did we write something? Try for more
660 */
661 if (wbc.nr_to_write < MAX_WRITEBACK_PAGES)
662 continue;
663 /*
664 * Nothing written. Wait for some inode to
665 * become available for writeback. Otherwise
666 * we'll just busyloop.
667 */
668 spin_lock(&inode_lock);
669 if (!list_empty(&wb->b_more_io)) {
670 inode = list_entry(wb->b_more_io.prev,
671 struct inode, i_list);
Dave Chinner028c2dd2010-07-07 13:24:07 +1000672 trace_wbc_writeback_wait(&wbc, wb->bdi);
Jens Axboe71fd05a2009-09-23 19:32:26 +0200673 inode_wait_for_writeback(inode);
Jens Axboe03ba3782009-09-09 09:08:54 +0200674 }
Jens Axboe71fd05a2009-09-23 19:32:26 +0200675 spin_unlock(&inode_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200676 }
677
678 return wrote;
679}
680
681/*
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200682 * Return the next wb_writeback_work struct that hasn't been processed yet.
Jens Axboe03ba3782009-09-09 09:08:54 +0200683 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200684static struct wb_writeback_work *
Minchan Kim08852b62010-08-03 12:51:16 +0200685get_next_work_item(struct backing_dev_info *bdi)
Jens Axboe03ba3782009-09-09 09:08:54 +0200686{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200687 struct wb_writeback_work *work = NULL;
Jens Axboe03ba3782009-09-09 09:08:54 +0200688
Artem Bityutskiy64677162010-07-25 14:29:22 +0300689 spin_lock_bh(&bdi->wb_lock);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200690 if (!list_empty(&bdi->work_list)) {
691 work = list_entry(bdi->work_list.next,
692 struct wb_writeback_work, list);
693 list_del_init(&work->list);
Jens Axboe03ba3782009-09-09 09:08:54 +0200694 }
Artem Bityutskiy64677162010-07-25 14:29:22 +0300695 spin_unlock_bh(&bdi->wb_lock);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200696 return work;
Jens Axboe03ba3782009-09-09 09:08:54 +0200697}
698
699static long wb_check_old_data_flush(struct bdi_writeback *wb)
700{
701 unsigned long expired;
702 long nr_pages;
703
Jens Axboe69b62d02010-05-17 12:51:03 +0200704 /*
705 * When set to zero, disable periodic writeback
706 */
707 if (!dirty_writeback_interval)
708 return 0;
709
Jens Axboe03ba3782009-09-09 09:08:54 +0200710 expired = wb->last_old_flush +
711 msecs_to_jiffies(dirty_writeback_interval * 10);
712 if (time_before(jiffies, expired))
713 return 0;
714
715 wb->last_old_flush = jiffies;
716 nr_pages = global_page_state(NR_FILE_DIRTY) +
717 global_page_state(NR_UNSTABLE_NFS) +
718 (inodes_stat.nr_inodes - inodes_stat.nr_unused);
719
Jens Axboec4a77a62009-09-16 15:18:25 +0200720 if (nr_pages) {
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200721 struct wb_writeback_work work = {
Jens Axboec4a77a62009-09-16 15:18:25 +0200722 .nr_pages = nr_pages,
723 .sync_mode = WB_SYNC_NONE,
724 .for_kupdate = 1,
725 .range_cyclic = 1,
726 };
727
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200728 return wb_writeback(wb, &work);
Jens Axboec4a77a62009-09-16 15:18:25 +0200729 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200730
731 return 0;
732}
733
734/*
735 * Retrieve work items and do the writeback they describe
736 */
737long wb_do_writeback(struct bdi_writeback *wb, int force_wait)
738{
739 struct backing_dev_info *bdi = wb->bdi;
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200740 struct wb_writeback_work *work;
Jens Axboec4a77a62009-09-16 15:18:25 +0200741 long wrote = 0;
Jens Axboe03ba3782009-09-09 09:08:54 +0200742
Jan Kara81d73a32010-08-11 14:17:44 -0700743 set_bit(BDI_writeback_running, &wb->bdi->state);
Minchan Kim08852b62010-08-03 12:51:16 +0200744 while ((work = get_next_work_item(bdi)) != NULL) {
Jens Axboe03ba3782009-09-09 09:08:54 +0200745 /*
746 * Override sync mode, in case we must wait for completion
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200747 * because this thread is exiting now.
Jens Axboe03ba3782009-09-09 09:08:54 +0200748 */
749 if (force_wait)
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200750 work->sync_mode = WB_SYNC_ALL;
751
Dave Chinner455b2862010-07-07 13:24:06 +1000752 trace_writeback_exec(bdi, work);
753
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200754 wrote += wb_writeback(wb, work);
Jens Axboe03ba3782009-09-09 09:08:54 +0200755
756 /*
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200757 * Notify the caller of completion if this is a synchronous
758 * work item, otherwise just free it.
Jens Axboe03ba3782009-09-09 09:08:54 +0200759 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200760 if (work->done)
761 complete(work->done);
762 else
763 kfree(work);
Jens Axboe03ba3782009-09-09 09:08:54 +0200764 }
765
766 /*
767 * Check for periodic writeback, kupdated() style
768 */
769 wrote += wb_check_old_data_flush(wb);
Jan Kara81d73a32010-08-11 14:17:44 -0700770 clear_bit(BDI_writeback_running, &wb->bdi->state);
Jens Axboe03ba3782009-09-09 09:08:54 +0200771
772 return wrote;
773}
774
775/*
776 * Handle writeback of dirty data for the device backed by this bdi. Also
777 * wakes up periodically and does kupdated style flushing.
778 */
Christoph Hellwig08243902010-06-19 23:08:22 +0200779int bdi_writeback_thread(void *data)
Jens Axboe03ba3782009-09-09 09:08:54 +0200780{
Christoph Hellwig08243902010-06-19 23:08:22 +0200781 struct bdi_writeback *wb = data;
782 struct backing_dev_info *bdi = wb->bdi;
Jens Axboe03ba3782009-09-09 09:08:54 +0200783 long pages_written;
784
Christoph Hellwig08243902010-06-19 23:08:22 +0200785 current->flags |= PF_FLUSHER | PF_SWAPWRITE;
786 set_freezable();
Artem Bityutskiyecd58402010-07-25 14:29:18 +0300787 wb->last_active = jiffies;
Christoph Hellwig08243902010-06-19 23:08:22 +0200788
789 /*
790 * Our parent may run at a different priority, just set us to normal
791 */
792 set_user_nice(current, 0);
793
Dave Chinner455b2862010-07-07 13:24:06 +1000794 trace_writeback_thread_start(bdi);
795
Jens Axboe03ba3782009-09-09 09:08:54 +0200796 while (!kthread_should_stop()) {
Artem Bityutskiy64677162010-07-25 14:29:22 +0300797 /*
798 * Remove own delayed wake-up timer, since we are already awake
799 * and we'll take care of the preriodic write-back.
800 */
801 del_timer(&wb->wakeup_timer);
802
Jens Axboe03ba3782009-09-09 09:08:54 +0200803 pages_written = wb_do_writeback(wb, 0);
804
Dave Chinner455b2862010-07-07 13:24:06 +1000805 trace_writeback_pages_written(pages_written);
Jens Axboe03ba3782009-09-09 09:08:54 +0200806
Jens Axboe03ba3782009-09-09 09:08:54 +0200807 if (pages_written)
Artem Bityutskiyecd58402010-07-25 14:29:18 +0300808 wb->last_active = jiffies;
Jens Axboe03ba3782009-09-09 09:08:54 +0200809
Artem Bityutskiy297252c2010-07-25 14:29:15 +0300810 set_current_state(TASK_INTERRUPTIBLE);
J. Bruce Fieldsb76b4012010-08-28 08:52:10 +0200811 if (!list_empty(&bdi->work_list) || kthread_should_stop()) {
Artem Bityutskiy297252c2010-07-25 14:29:15 +0300812 __set_current_state(TASK_RUNNING);
813 continue;
Jens Axboe03ba3782009-09-09 09:08:54 +0200814 }
815
Artem Bityutskiy253c34e2010-07-25 14:29:21 +0300816 if (wb_has_dirty_io(wb) && dirty_writeback_interval)
Artem Bityutskiyfff5b852010-07-25 14:29:20 +0300817 schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10));
Artem Bityutskiy253c34e2010-07-25 14:29:21 +0300818 else {
819 /*
820 * We have nothing to do, so can go sleep without any
821 * timeout and save power. When a work is queued or
822 * something is made dirty - we will be woken up.
823 */
Artem Bityutskiy297252c2010-07-25 14:29:15 +0300824 schedule();
Jens Axboef9eadbb2010-05-18 14:31:45 +0200825 }
Jens Axboe69b62d02010-05-17 12:51:03 +0200826
Jens Axboe03ba3782009-09-09 09:08:54 +0200827 try_to_freeze();
828 }
829
Artem Bityutskiyfff5b852010-07-25 14:29:20 +0300830 /* Flush any work that raced with us exiting */
Christoph Hellwig08243902010-06-19 23:08:22 +0200831 if (!list_empty(&bdi->work_list))
832 wb_do_writeback(wb, 1);
Dave Chinner455b2862010-07-07 13:24:06 +1000833
834 trace_writeback_thread_stop(bdi);
Jens Axboe03ba3782009-09-09 09:08:54 +0200835 return 0;
836}
837
Christoph Hellwig08243902010-06-19 23:08:22 +0200838
Jens Axboe03ba3782009-09-09 09:08:54 +0200839/*
Jens Axboe03ba3782009-09-09 09:08:54 +0200840 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
841 * the whole world.
842 */
843void wakeup_flusher_threads(long nr_pages)
844{
Christoph Hellwigb8c2f342010-06-08 18:15:07 +0200845 struct backing_dev_info *bdi;
Christoph Hellwigb8c2f342010-06-08 18:15:07 +0200846
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200847 if (!nr_pages) {
848 nr_pages = global_page_state(NR_FILE_DIRTY) +
Jens Axboe03ba3782009-09-09 09:08:54 +0200849 global_page_state(NR_UNSTABLE_NFS);
Christoph Hellwigb8c2f342010-06-08 18:15:07 +0200850 }
851
852 rcu_read_lock();
853 list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) {
854 if (!bdi_has_dirty_io(bdi))
855 continue;
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200856 __bdi_start_writeback(bdi, nr_pages, false, false);
Christoph Hellwigb8c2f342010-06-08 18:15:07 +0200857 }
858 rcu_read_unlock();
Jens Axboe03ba3782009-09-09 09:08:54 +0200859}
860
861static noinline void block_dump___mark_inode_dirty(struct inode *inode)
862{
863 if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) {
864 struct dentry *dentry;
865 const char *name = "?";
866
867 dentry = d_find_alias(inode);
868 if (dentry) {
869 spin_lock(&dentry->d_lock);
870 name = (const char *) dentry->d_name.name;
871 }
872 printk(KERN_DEBUG
873 "%s(%d): dirtied inode %lu (%s) on %s\n",
874 current->comm, task_pid_nr(current), inode->i_ino,
875 name, inode->i_sb->s_id);
876 if (dentry) {
877 spin_unlock(&dentry->d_lock);
878 dput(dentry);
879 }
880 }
881}
882
883/**
884 * __mark_inode_dirty - internal function
885 * @inode: inode to mark
886 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
887 * Mark an inode as dirty. Callers should use mark_inode_dirty or
888 * mark_inode_dirty_sync.
889 *
890 * Put the inode on the super block's dirty list.
891 *
892 * CAREFUL! We mark it dirty unconditionally, but move it onto the
893 * dirty list only if it is hashed or if it refers to a blockdev.
894 * If it was not hashed, it will never be added to the dirty list
895 * even if it is later hashed, as it will have been marked dirty already.
896 *
897 * In short, make sure you hash any inodes _before_ you start marking
898 * them dirty.
899 *
900 * This function *must* be atomic for the I_DIRTY_PAGES case -
901 * set_page_dirty() is called under spinlock in several places.
902 *
903 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
904 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
905 * the kernel-internal blockdev inode represents the dirtying time of the
906 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
907 * page->mapping->host, so the page-dirtying time is recorded in the internal
908 * blockdev inode.
909 */
910void __mark_inode_dirty(struct inode *inode, int flags)
911{
912 struct super_block *sb = inode->i_sb;
Artem Bityutskiy253c34e2010-07-25 14:29:21 +0300913 struct backing_dev_info *bdi = NULL;
914 bool wakeup_bdi = false;
Jens Axboe03ba3782009-09-09 09:08:54 +0200915
916 /*
917 * Don't do this for I_DIRTY_PAGES - that doesn't actually
918 * dirty the inode itself
919 */
920 if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
921 if (sb->s_op->dirty_inode)
922 sb->s_op->dirty_inode(inode);
923 }
924
925 /*
926 * make sure that changes are seen by all cpus before we test i_state
927 * -- mikulas
928 */
929 smp_mb();
930
931 /* avoid the locking if we can */
932 if ((inode->i_state & flags) == flags)
933 return;
934
935 if (unlikely(block_dump))
936 block_dump___mark_inode_dirty(inode);
937
938 spin_lock(&inode_lock);
939 if ((inode->i_state & flags) != flags) {
940 const int was_dirty = inode->i_state & I_DIRTY;
941
942 inode->i_state |= flags;
943
944 /*
945 * If the inode is being synced, just update its dirty state.
946 * The unlocker will place the inode on the appropriate
947 * superblock list, based upon its state.
948 */
949 if (inode->i_state & I_SYNC)
950 goto out;
951
952 /*
953 * Only add valid (hashed) inodes to the superblock's
954 * dirty list. Add blockdev inodes as well.
955 */
956 if (!S_ISBLK(inode->i_mode)) {
957 if (hlist_unhashed(&inode->i_hash))
958 goto out;
959 }
Al Viroa4ffdde2010-06-02 17:38:30 -0400960 if (inode->i_state & I_FREEING)
Jens Axboe03ba3782009-09-09 09:08:54 +0200961 goto out;
962
963 /*
964 * If the inode was already on b_dirty/b_io/b_more_io, don't
965 * reposition it (that would break b_dirty time-ordering).
966 */
967 if (!was_dirty) {
Artem Bityutskiy253c34e2010-07-25 14:29:21 +0300968 bdi = inode_to_bdi(inode);
Jens Axboe500b0672009-09-09 09:10:25 +0200969
Artem Bityutskiy253c34e2010-07-25 14:29:21 +0300970 if (bdi_cap_writeback_dirty(bdi)) {
971 WARN(!test_bit(BDI_registered, &bdi->state),
972 "bdi-%s not registered\n", bdi->name);
973
974 /*
975 * If this is the first dirty inode for this
976 * bdi, we have to wake-up the corresponding
977 * bdi thread to make sure background
978 * write-back happens later.
979 */
980 if (!wb_has_dirty_io(&bdi->wb))
981 wakeup_bdi = true;
Jens Axboe500b0672009-09-09 09:10:25 +0200982 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200983
984 inode->dirtied_when = jiffies;
Artem Bityutskiy253c34e2010-07-25 14:29:21 +0300985 list_move(&inode->i_list, &bdi->wb.b_dirty);
Jens Axboe03ba3782009-09-09 09:08:54 +0200986 }
987 }
988out:
989 spin_unlock(&inode_lock);
Artem Bityutskiy253c34e2010-07-25 14:29:21 +0300990
991 if (wakeup_bdi)
Artem Bityutskiy64677162010-07-25 14:29:22 +0300992 bdi_wakeup_thread_delayed(bdi);
Jens Axboe03ba3782009-09-09 09:08:54 +0200993}
994EXPORT_SYMBOL(__mark_inode_dirty);
995
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200996/*
997 * Write out a superblock's list of dirty inodes. A wait will be performed
998 * upon no inodes, all inodes or the final one, depending upon sync_mode.
999 *
1000 * If older_than_this is non-NULL, then only write out inodes which
1001 * had their first dirtying at a time earlier than *older_than_this.
1002 *
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001003 * If `bdi' is non-zero then we're being asked to writeback a specific queue.
1004 * This function assumes that the blockdev superblock's inodes are backed by
1005 * a variety of queues, so all inodes are searched. For other superblocks,
1006 * assume that all inodes are backed by the same queue.
1007 *
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001008 * The inodes to be written are parked on bdi->b_io. They are moved back onto
1009 * bdi->b_dirty as they are selected for writing. This way, none can be missed
1010 * on the writer throttling path, and we get decent balancing between many
1011 * throttled threads: we don't want them all piling up on inode_sync_wait.
1012 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001013static void wait_sb_inodes(struct super_block *sb)
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001014{
Jens Axboe03ba3782009-09-09 09:08:54 +02001015 struct inode *inode, *old_inode = NULL;
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001016
Jens Axboe03ba3782009-09-09 09:08:54 +02001017 /*
1018 * We need to be protected against the filesystem going from
1019 * r/o to r/w or vice versa.
1020 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001021 WARN_ON(!rwsem_is_locked(&sb->s_umount));
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001022
Jens Axboe03ba3782009-09-09 09:08:54 +02001023 spin_lock(&inode_lock);
1024
1025 /*
1026 * Data integrity sync. Must wait for all pages under writeback,
1027 * because there may have been pages dirtied before our sync
1028 * call, but which had writeout started before we write it out.
1029 * In which case, the inode may not be on the dirty list, but
1030 * we still have to wait for that writeout.
1031 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001032 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
Jens Axboe03ba3782009-09-09 09:08:54 +02001033 struct address_space *mapping;
1034
Al Viroa4ffdde2010-06-02 17:38:30 -04001035 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW))
Jens Axboe03ba3782009-09-09 09:08:54 +02001036 continue;
1037 mapping = inode->i_mapping;
1038 if (mapping->nrpages == 0)
1039 continue;
1040 __iget(inode);
1041 spin_unlock(&inode_lock);
1042 /*
1043 * We hold a reference to 'inode' so it couldn't have
1044 * been removed from s_inodes list while we dropped the
1045 * inode_lock. We cannot iput the inode now as we can
1046 * be holding the last reference and we cannot iput it
1047 * under inode_lock. So we keep the reference and iput
1048 * it later.
1049 */
1050 iput(old_inode);
1051 old_inode = inode;
1052
1053 filemap_fdatawait(mapping);
1054
1055 cond_resched();
Nick Piggin38f21972009-01-06 14:40:25 -08001056
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001057 spin_lock(&inode_lock);
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001058 }
Jens Axboe03ba3782009-09-09 09:08:54 +02001059 spin_unlock(&inode_lock);
1060 iput(old_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061}
1062
Jens Axboed8a85592009-09-02 12:34:32 +02001063/**
1064 * writeback_inodes_sb - writeback dirty inodes from given super_block
1065 * @sb: the superblock
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066 *
Jens Axboed8a85592009-09-02 12:34:32 +02001067 * Start writeback on some inodes on this super_block. No guarantees are made
1068 * on how many (if any) will be written, and this function does not wait
1069 * for IO completion of submitted IO. The number of pages submitted is
1070 * returned.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001072void writeback_inodes_sb(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073{
Jens Axboe0e3c9a22010-06-01 11:08:43 +02001074 unsigned long nr_dirty = global_page_state(NR_FILE_DIRTY);
1075 unsigned long nr_unstable = global_page_state(NR_UNSTABLE_NFS);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001076 DECLARE_COMPLETION_ONSTACK(done);
1077 struct wb_writeback_work work = {
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001078 .sb = sb,
1079 .sync_mode = WB_SYNC_NONE,
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001080 .done = &done,
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001081 };
Jens Axboe0e3c9a22010-06-01 11:08:43 +02001082
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001083 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1084
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001085 work.nr_pages = nr_dirty + nr_unstable +
Jens Axboe0e3c9a22010-06-01 11:08:43 +02001086 (inodes_stat.nr_inodes - inodes_stat.nr_unused);
1087
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001088 bdi_queue_work(sb->s_bdi, &work);
1089 wait_for_completion(&done);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090}
Jens Axboed8a85592009-09-02 12:34:32 +02001091EXPORT_SYMBOL(writeback_inodes_sb);
1092
1093/**
Eric Sandeen17bd55d2009-12-23 07:57:07 -05001094 * writeback_inodes_sb_if_idle - start writeback if none underway
1095 * @sb: the superblock
1096 *
1097 * Invoke writeback_inodes_sb if no writeback is currently underway.
1098 * Returns 1 if writeback was started, 0 if not.
1099 */
1100int writeback_inodes_sb_if_idle(struct super_block *sb)
1101{
1102 if (!writeback_in_progress(sb->s_bdi)) {
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001103 down_read(&sb->s_umount);
Eric Sandeen17bd55d2009-12-23 07:57:07 -05001104 writeback_inodes_sb(sb);
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001105 up_read(&sb->s_umount);
Eric Sandeen17bd55d2009-12-23 07:57:07 -05001106 return 1;
1107 } else
1108 return 0;
1109}
1110EXPORT_SYMBOL(writeback_inodes_sb_if_idle);
1111
1112/**
Jens Axboed8a85592009-09-02 12:34:32 +02001113 * sync_inodes_sb - sync sb inode pages
1114 * @sb: the superblock
1115 *
1116 * This function writes and waits on any dirty inode belonging to this
1117 * super_block. The number of pages synced is returned.
1118 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001119void sync_inodes_sb(struct super_block *sb)
Jens Axboed8a85592009-09-02 12:34:32 +02001120{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001121 DECLARE_COMPLETION_ONSTACK(done);
1122 struct wb_writeback_work work = {
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001123 .sb = sb,
1124 .sync_mode = WB_SYNC_ALL,
1125 .nr_pages = LONG_MAX,
1126 .range_cyclic = 0,
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001127 .done = &done,
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001128 };
1129
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001130 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1131
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001132 bdi_queue_work(sb->s_bdi, &work);
1133 wait_for_completion(&done);
1134
Jens Axboeb6e51312009-09-16 15:13:54 +02001135 wait_sb_inodes(sb);
Jens Axboed8a85592009-09-02 12:34:32 +02001136}
1137EXPORT_SYMBOL(sync_inodes_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139/**
Andrea Arcangeli7f04c262005-10-30 15:03:05 -08001140 * write_inode_now - write an inode to disk
1141 * @inode: inode to write to disk
1142 * @sync: whether the write should be synchronous or not
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143 *
Andrea Arcangeli7f04c262005-10-30 15:03:05 -08001144 * This function commits an inode to disk immediately if it is dirty. This is
1145 * primarily needed by knfsd.
1146 *
1147 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149int write_inode_now(struct inode *inode, int sync)
1150{
1151 int ret;
1152 struct writeback_control wbc = {
1153 .nr_to_write = LONG_MAX,
Mike Galbraith18914b12008-02-08 04:20:23 -08001154 .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -07001155 .range_start = 0,
1156 .range_end = LLONG_MAX,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157 };
1158
1159 if (!mapping_cap_writeback_dirty(inode->i_mapping))
Andrew Morton49364ce2005-11-07 00:59:15 -08001160 wbc.nr_to_write = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161
1162 might_sleep();
1163 spin_lock(&inode_lock);
Christoph Hellwig01c03192009-06-08 13:35:40 +02001164 ret = writeback_single_inode(inode, &wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165 spin_unlock(&inode_lock);
1166 if (sync)
Joern Engel1c0eeaf2007-10-16 23:30:44 -07001167 inode_sync_wait(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168 return ret;
1169}
1170EXPORT_SYMBOL(write_inode_now);
1171
1172/**
1173 * sync_inode - write an inode and its pages to disk.
1174 * @inode: the inode to sync
1175 * @wbc: controls the writeback mode
1176 *
1177 * sync_inode() will write an inode and its pages to disk. It will also
1178 * correctly update the inode on its superblock's dirty inode lists and will
1179 * update inode->i_state.
1180 *
1181 * The caller must have a ref on the inode.
1182 */
1183int sync_inode(struct inode *inode, struct writeback_control *wbc)
1184{
1185 int ret;
1186
1187 spin_lock(&inode_lock);
Christoph Hellwig01c03192009-06-08 13:35:40 +02001188 ret = writeback_single_inode(inode, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189 spin_unlock(&inode_lock);
1190 return ret;
1191}
1192EXPORT_SYMBOL(sync_inode);