blob: 9e72d04e706e06dfd0885432cec6dc20e15374c8 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * fs/fs-writeback.c
3 *
4 * Copyright (C) 2002, Linus Torvalds.
5 *
6 * Contains all the functions related to writing back and waiting
7 * upon dirty inodes against superblocks, and writing back dirty
8 * pages against inodes. ie: data writeback. Writeout of the
9 * inode itself is not handled here.
10 *
Francois Camie1f8e872008-10-15 22:01:59 -070011 * 10Apr2002 Andrew Morton
Linus Torvalds1da177e2005-04-16 15:20:36 -070012 * Split out of fs/inode.c
13 * Additions for address_space-based writeback
14 */
15
16#include <linux/kernel.h>
Jens Axboef5ff8422007-09-21 09:19:54 +020017#include <linux/module.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include <linux/spinlock.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090019#include <linux/slab.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/sched.h>
21#include <linux/fs.h>
22#include <linux/mm.h>
Jens Axboe03ba3782009-09-09 09:08:54 +020023#include <linux/kthread.h>
24#include <linux/freezer.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/writeback.h>
26#include <linux/blkdev.h>
27#include <linux/backing-dev.h>
28#include <linux/buffer_head.h>
Dave Chinner455b2862010-07-07 13:24:06 +100029#include <linux/tracepoint.h>
David Howells07f3f052006-09-30 20:52:18 +020030#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070031
Jens Axboed0bceac2009-05-18 08:20:32 +020032/*
Jens Axboec4a77a62009-09-16 15:18:25 +020033 * Passed into wb_writeback(), essentially a subset of writeback_control
34 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +020035struct wb_writeback_work {
Jens Axboec4a77a62009-09-16 15:18:25 +020036 long nr_pages;
37 struct super_block *sb;
38 enum writeback_sync_modes sync_mode;
H Hartley Sweeten52957fe2010-04-01 20:36:30 -050039 unsigned int for_kupdate:1;
40 unsigned int range_cyclic:1;
41 unsigned int for_background:1;
Jens Axboec4a77a62009-09-16 15:18:25 +020042
Jens Axboe8010c3b2009-09-15 20:04:57 +020043 struct list_head list; /* pending work list */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +020044 struct completion *done; /* set if the caller waits */
Jens Axboe03ba3782009-09-09 09:08:54 +020045};
46
Dave Chinner455b2862010-07-07 13:24:06 +100047/*
48 * Include the creation of the trace points after defining the
49 * wb_writeback_work structure so that the definition remains local to this
50 * file.
51 */
52#define CREATE_TRACE_POINTS
53#include <trace/events/writeback.h>
54
Dave Chinner455b2862010-07-07 13:24:06 +100055/*
56 * We don't actually have pdflush, but this one is exported though /proc...
57 */
58int nr_pdflush_threads;
59
Adrian Bunkf11b00f2008-04-29 00:58:56 -070060/**
61 * writeback_in_progress - determine whether there is writeback in progress
62 * @bdi: the device's backing_dev_info structure.
63 *
Jens Axboe03ba3782009-09-09 09:08:54 +020064 * Determine whether there is writeback waiting to be handled against a
65 * backing device.
Adrian Bunkf11b00f2008-04-29 00:58:56 -070066 */
67int writeback_in_progress(struct backing_dev_info *bdi)
68{
Jan Kara81d73a32010-08-11 14:17:44 -070069 return test_bit(BDI_writeback_running, &bdi->state);
Adrian Bunkf11b00f2008-04-29 00:58:56 -070070}
71
Jan Kara692ebd12010-09-21 11:51:01 +020072static inline struct backing_dev_info *inode_to_bdi(struct inode *inode)
73{
74 struct super_block *sb = inode->i_sb;
Jan Kara692ebd12010-09-21 11:51:01 +020075
Christoph Hellwigaaead252010-10-04 14:25:33 +020076 if (strcmp(sb->s_type->name, "bdev") == 0)
77 return inode->i_mapping->backing_dev_info;
78
79 return sb->s_bdi;
Jan Kara692ebd12010-09-21 11:51:01 +020080}
81
Nick Piggin7ccf19a2010-10-21 11:49:30 +110082static inline struct inode *wb_inode(struct list_head *head)
83{
84 return list_entry(head, struct inode, i_wb_list);
85}
86
Jan Kara65850272011-01-13 15:45:44 -080087/* Wakeup flusher thread or forker thread to fork it. Requires bdi->wb_lock. */
88static void bdi_wakeup_flusher(struct backing_dev_info *bdi)
Nick Piggin4195f732009-05-28 09:01:15 +020089{
Artem Bityutskiyfff5b852010-07-25 14:29:20 +030090 if (bdi->wb.task) {
91 wake_up_process(bdi->wb.task);
92 } else {
93 /*
94 * The bdi thread isn't there, wake up the forker thread which
95 * will create and run it.
96 */
Jens Axboe03ba3782009-09-09 09:08:54 +020097 wake_up_process(default_backing_dev_info.wb.task);
Linus Torvalds1da177e2005-04-16 15:20:36 -070098 }
Jan Kara65850272011-01-13 15:45:44 -080099}
100
101static void bdi_queue_work(struct backing_dev_info *bdi,
102 struct wb_writeback_work *work)
103{
104 trace_writeback_queue(bdi, work);
105
106 spin_lock_bh(&bdi->wb_lock);
107 list_add_tail(&work->list, &bdi->work_list);
108 if (!bdi->wb.task)
109 trace_writeback_nothread(bdi, work);
110 bdi_wakeup_flusher(bdi);
Artem Bityutskiy64677162010-07-25 14:29:22 +0300111 spin_unlock_bh(&bdi->wb_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112}
113
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200114static void
115__bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages,
Jan Kara65850272011-01-13 15:45:44 -0800116 bool range_cyclic)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200118 struct wb_writeback_work *work;
Jens Axboe03ba3782009-09-09 09:08:54 +0200119
Jens Axboebcddc3f2009-09-13 20:07:36 +0200120 /*
121 * This is WB_SYNC_NONE writeback, so if allocation fails just
122 * wakeup the thread for old dirty data writeback
123 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200124 work = kzalloc(sizeof(*work), GFP_ATOMIC);
125 if (!work) {
Dave Chinner455b2862010-07-07 13:24:06 +1000126 if (bdi->wb.task) {
127 trace_writeback_nowork(bdi);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200128 wake_up_process(bdi->wb.task);
Dave Chinner455b2862010-07-07 13:24:06 +1000129 }
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200130 return;
Jens Axboebcddc3f2009-09-13 20:07:36 +0200131 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200132
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200133 work->sync_mode = WB_SYNC_NONE;
134 work->nr_pages = nr_pages;
135 work->range_cyclic = range_cyclic;
Christoph Hellwigf0fad8a2009-09-11 09:47:56 +0200136
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200137 bdi_queue_work(bdi, work);
Jens Axboeb6e51312009-09-16 15:13:54 +0200138}
139
140/**
141 * bdi_start_writeback - start writeback
142 * @bdi: the backing device to write from
143 * @nr_pages: the number of pages to write
144 *
145 * Description:
146 * This does WB_SYNC_NONE opportunistic writeback. The IO is only
147 * started when this function returns, we make no guarentees on
Jens Axboe0e3c9a22010-06-01 11:08:43 +0200148 * completion. Caller need not hold sb s_umount semaphore.
Jens Axboeb6e51312009-09-16 15:13:54 +0200149 *
150 */
Christoph Hellwigc5444192010-06-08 18:15:15 +0200151void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages)
Jens Axboeb6e51312009-09-16 15:13:54 +0200152{
Jan Kara65850272011-01-13 15:45:44 -0800153 __bdi_start_writeback(bdi, nr_pages, true);
Christoph Hellwigc5444192010-06-08 18:15:15 +0200154}
Wu Fengguangd3ddec72009-09-23 20:33:40 +0800155
Christoph Hellwigc5444192010-06-08 18:15:15 +0200156/**
157 * bdi_start_background_writeback - start background writeback
158 * @bdi: the backing device to write from
159 *
160 * Description:
Jan Kara65850272011-01-13 15:45:44 -0800161 * This makes sure WB_SYNC_NONE background writeback happens. When
162 * this function returns, it is only guaranteed that for given BDI
163 * some IO is happening if we are over background dirty threshold.
164 * Caller need not hold sb s_umount semaphore.
Christoph Hellwigc5444192010-06-08 18:15:15 +0200165 */
166void bdi_start_background_writeback(struct backing_dev_info *bdi)
167{
Jan Kara65850272011-01-13 15:45:44 -0800168 /*
169 * We just wake up the flusher thread. It will perform background
170 * writeback as soon as there is no other work to do.
171 */
Wu Fengguang71927e82011-01-13 15:45:46 -0800172 trace_writeback_wake_background(bdi);
Jan Kara65850272011-01-13 15:45:44 -0800173 spin_lock_bh(&bdi->wb_lock);
174 bdi_wakeup_flusher(bdi);
175 spin_unlock_bh(&bdi->wb_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176}
177
178/*
Andrew Morton6610a0b2007-10-16 23:30:32 -0700179 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
180 * furthest end of its superblock's dirty-inode list.
181 *
182 * Before stamping the inode's ->dirtied_when, we check to see whether it is
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200183 * already the most-recently-dirtied inode on the b_dirty list. If that is
Andrew Morton6610a0b2007-10-16 23:30:32 -0700184 * the case then the inode must have been redirtied while it was being written
185 * out and we don't reset its dirtied_when.
186 */
187static void redirty_tail(struct inode *inode)
188{
Jens Axboe03ba3782009-09-09 09:08:54 +0200189 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
Andrew Morton6610a0b2007-10-16 23:30:32 -0700190
Jens Axboe03ba3782009-09-09 09:08:54 +0200191 if (!list_empty(&wb->b_dirty)) {
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200192 struct inode *tail;
Andrew Morton6610a0b2007-10-16 23:30:32 -0700193
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100194 tail = wb_inode(wb->b_dirty.next);
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200195 if (time_before(inode->dirtied_when, tail->dirtied_when))
Andrew Morton6610a0b2007-10-16 23:30:32 -0700196 inode->dirtied_when = jiffies;
197 }
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100198 list_move(&inode->i_wb_list, &wb->b_dirty);
Andrew Morton6610a0b2007-10-16 23:30:32 -0700199}
200
201/*
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200202 * requeue inode for re-scanning after bdi->b_io list is exhausted.
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700203 */
Ken Chen0e0f4fc2007-10-16 23:30:38 -0700204static void requeue_io(struct inode *inode)
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700205{
Jens Axboe03ba3782009-09-09 09:08:54 +0200206 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
207
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100208 list_move(&inode->i_wb_list, &wb->b_more_io);
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700209}
210
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700211static void inode_sync_complete(struct inode *inode)
212{
213 /*
214 * Prevent speculative execution through spin_unlock(&inode_lock);
215 */
216 smp_mb();
217 wake_up_bit(&inode->i_state, __I_SYNC);
218}
219
Jeff Laytond2caa3c52009-04-02 16:56:37 -0700220static bool inode_dirtied_after(struct inode *inode, unsigned long t)
221{
222 bool ret = time_after(inode->dirtied_when, t);
223#ifndef CONFIG_64BIT
224 /*
225 * For inodes being constantly redirtied, dirtied_when can get stuck.
226 * It _appears_ to be in the future, but is actually in distant past.
227 * This test is necessary to prevent such wrapped-around relative times
Jens Axboe5b0830c2009-09-23 19:37:09 +0200228 * from permanently stopping the whole bdi writeback.
Jeff Laytond2caa3c52009-04-02 16:56:37 -0700229 */
230 ret = ret && time_before_eq(inode->dirtied_when, jiffies);
231#endif
232 return ret;
233}
234
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700235/*
Fengguang Wu2c136572007-10-16 23:30:39 -0700236 * Move expired dirty inodes from @delaying_queue to @dispatch_queue.
237 */
238static void move_expired_inodes(struct list_head *delaying_queue,
239 struct list_head *dispatch_queue,
240 unsigned long *older_than_this)
241{
Shaohua Li5c034492009-09-24 14:42:33 +0200242 LIST_HEAD(tmp);
243 struct list_head *pos, *node;
Jens Axboecf137302009-09-24 15:12:57 +0200244 struct super_block *sb = NULL;
Shaohua Li5c034492009-09-24 14:42:33 +0200245 struct inode *inode;
Jens Axboecf137302009-09-24 15:12:57 +0200246 int do_sb_sort = 0;
Shaohua Li5c034492009-09-24 14:42:33 +0200247
Fengguang Wu2c136572007-10-16 23:30:39 -0700248 while (!list_empty(delaying_queue)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100249 inode = wb_inode(delaying_queue->prev);
Fengguang Wu2c136572007-10-16 23:30:39 -0700250 if (older_than_this &&
Jeff Laytond2caa3c52009-04-02 16:56:37 -0700251 inode_dirtied_after(inode, *older_than_this))
Fengguang Wu2c136572007-10-16 23:30:39 -0700252 break;
Jens Axboecf137302009-09-24 15:12:57 +0200253 if (sb && sb != inode->i_sb)
254 do_sb_sort = 1;
255 sb = inode->i_sb;
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100256 list_move(&inode->i_wb_list, &tmp);
Shaohua Li5c034492009-09-24 14:42:33 +0200257 }
258
Jens Axboecf137302009-09-24 15:12:57 +0200259 /* just one sb in list, splice to dispatch_queue and we're done */
260 if (!do_sb_sort) {
261 list_splice(&tmp, dispatch_queue);
262 return;
263 }
264
Shaohua Li5c034492009-09-24 14:42:33 +0200265 /* Move inodes from one superblock together */
266 while (!list_empty(&tmp)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100267 sb = wb_inode(tmp.prev)->i_sb;
Shaohua Li5c034492009-09-24 14:42:33 +0200268 list_for_each_prev_safe(pos, node, &tmp) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100269 inode = wb_inode(pos);
Shaohua Li5c034492009-09-24 14:42:33 +0200270 if (inode->i_sb == sb)
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100271 list_move(&inode->i_wb_list, dispatch_queue);
Shaohua Li5c034492009-09-24 14:42:33 +0200272 }
Fengguang Wu2c136572007-10-16 23:30:39 -0700273 }
274}
275
276/*
277 * Queue all expired dirty inodes for io, eldest first.
Wu Fengguang4ea879b2010-08-11 14:17:42 -0700278 * Before
279 * newly dirtied b_dirty b_io b_more_io
280 * =============> gf edc BA
281 * After
282 * newly dirtied b_dirty b_io b_more_io
283 * =============> g fBAedc
284 * |
285 * +--> dequeue for IO
Fengguang Wu2c136572007-10-16 23:30:39 -0700286 */
Jens Axboe03ba3782009-09-09 09:08:54 +0200287static void queue_io(struct bdi_writeback *wb, unsigned long *older_than_this)
Fengguang Wu2c136572007-10-16 23:30:39 -0700288{
Wu Fengguang4ea879b2010-08-11 14:17:42 -0700289 list_splice_init(&wb->b_more_io, &wb->b_io);
Jens Axboe03ba3782009-09-09 09:08:54 +0200290 move_expired_inodes(&wb->b_dirty, &wb->b_io, older_than_this);
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200291}
292
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100293static int write_inode(struct inode *inode, struct writeback_control *wbc)
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200294{
Jens Axboe03ba3782009-09-09 09:08:54 +0200295 if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode))
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100296 return inode->i_sb->s_op->write_inode(inode, wbc);
Jens Axboe03ba3782009-09-09 09:08:54 +0200297 return 0;
Fengguang Wu2c136572007-10-16 23:30:39 -0700298}
299
300/*
Christoph Hellwig01c03192009-06-08 13:35:40 +0200301 * Wait for writeback on an inode to complete.
302 */
303static void inode_wait_for_writeback(struct inode *inode)
304{
305 DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC);
306 wait_queue_head_t *wqh;
307
308 wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
Richard Kennedy58a9d3d82010-05-24 14:32:38 -0700309 while (inode->i_state & I_SYNC) {
Christoph Hellwig01c03192009-06-08 13:35:40 +0200310 spin_unlock(&inode_lock);
311 __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE);
312 spin_lock(&inode_lock);
Richard Kennedy58a9d3d82010-05-24 14:32:38 -0700313 }
Christoph Hellwig01c03192009-06-08 13:35:40 +0200314}
315
316/*
317 * Write out an inode's dirty pages. Called under inode_lock. Either the
318 * caller has ref on the inode (either via __iget or via syscall against an fd)
319 * or the inode has I_WILL_FREE set (via generic_forget_inode)
320 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321 * If `wait' is set, wait on the writeout.
322 *
323 * The whole writeout design is quite complex and fragile. We want to avoid
324 * starvation of particular inodes when others are being redirtied, prevent
325 * livelocks, etc.
326 *
327 * Called under inode_lock.
328 */
329static int
Christoph Hellwig01c03192009-06-08 13:35:40 +0200330writeback_single_inode(struct inode *inode, struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 struct address_space *mapping = inode->i_mapping;
Christoph Hellwig01c03192009-06-08 13:35:40 +0200333 unsigned dirty;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 int ret;
335
Christoph Hellwig01c03192009-06-08 13:35:40 +0200336 if (!atomic_read(&inode->i_count))
337 WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING)));
338 else
339 WARN_ON(inode->i_state & I_WILL_FREE);
340
341 if (inode->i_state & I_SYNC) {
342 /*
343 * If this inode is locked for writeback and we are not doing
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200344 * writeback-for-data-integrity, move it to b_more_io so that
Christoph Hellwig01c03192009-06-08 13:35:40 +0200345 * writeback can proceed with the other inodes on s_io.
346 *
347 * We'll have another go at writing back this inode when we
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200348 * completed a full scan of b_io.
Christoph Hellwig01c03192009-06-08 13:35:40 +0200349 */
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100350 if (wbc->sync_mode != WB_SYNC_ALL) {
Christoph Hellwig01c03192009-06-08 13:35:40 +0200351 requeue_io(inode);
352 return 0;
353 }
354
355 /*
356 * It's a data-integrity sync. We must wait.
357 */
358 inode_wait_for_writeback(inode);
359 }
360
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700361 BUG_ON(inode->i_state & I_SYNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362
Dmitry Monakhov5547e8a2010-05-07 13:35:44 +0400363 /* Set I_SYNC, reset I_DIRTY_PAGES */
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700364 inode->i_state |= I_SYNC;
Dmitry Monakhov5547e8a2010-05-07 13:35:44 +0400365 inode->i_state &= ~I_DIRTY_PAGES;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366 spin_unlock(&inode_lock);
367
368 ret = do_writepages(mapping, wbc);
369
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100370 /*
371 * Make sure to wait on the data before writing out the metadata.
372 * This is important for filesystems that modify metadata on data
373 * I/O completion.
374 */
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100375 if (wbc->sync_mode == WB_SYNC_ALL) {
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100376 int err = filemap_fdatawait(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377 if (ret == 0)
378 ret = err;
379 }
380
Dmitry Monakhov5547e8a2010-05-07 13:35:44 +0400381 /*
382 * Some filesystems may redirty the inode during the writeback
383 * due to delalloc, clear dirty metadata flags right before
384 * write_inode()
385 */
386 spin_lock(&inode_lock);
387 dirty = inode->i_state & I_DIRTY;
388 inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC);
389 spin_unlock(&inode_lock);
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100390 /* Don't write the inode if only I_DIRTY_PAGES was set */
391 if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100392 int err = write_inode(inode, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393 if (ret == 0)
394 ret = err;
395 }
396
397 spin_lock(&inode_lock);
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700398 inode->i_state &= ~I_SYNC;
Al Viroa4ffdde2010-06-02 17:38:30 -0400399 if (!(inode->i_state & I_FREEING)) {
Wu Fengguang23539af2010-08-11 14:17:41 -0700400 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401 /*
402 * We didn't write back all the pages. nfs_writepages()
Wu Fengguanga50aeb42010-08-11 14:17:43 -0700403 * sometimes bales out without doing anything.
Andrew Morton1b43ef92007-10-16 23:30:35 -0700404 */
Wu Fengguanga50aeb42010-08-11 14:17:43 -0700405 inode->i_state |= I_DIRTY_PAGES;
406 if (wbc->nr_to_write <= 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407 /*
Wu Fengguanga50aeb42010-08-11 14:17:43 -0700408 * slice used up: queue for next turn
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409 */
Wu Fengguanga50aeb42010-08-11 14:17:43 -0700410 requeue_io(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 } else {
412 /*
Wu Fengguanga50aeb42010-08-11 14:17:43 -0700413 * Writeback blocked by something other than
414 * congestion. Delay the inode for some time to
415 * avoid spinning on the CPU (100% iowait)
416 * retrying writeback of the dirty page/inode
417 * that cannot be performed immediately.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418 */
Andrew Morton1b43ef92007-10-16 23:30:35 -0700419 redirty_tail(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420 }
Wu Fengguang23539af2010-08-11 14:17:41 -0700421 } else if (inode->i_state & I_DIRTY) {
422 /*
423 * Filesystems can dirty the inode during writeback
424 * operations, such as delayed allocation during
425 * submission or metadata updates after data IO
426 * completion.
427 */
428 redirty_tail(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429 } else {
430 /*
Nick Piggin9e38d862010-10-23 06:55:17 -0400431 * The inode is clean. At this point we either have
432 * a reference to the inode or it's on it's way out.
433 * No need to add it back to the LRU.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434 */
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100435 list_del_init(&inode->i_wb_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436 }
437 }
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700438 inode_sync_complete(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439 return ret;
440}
441
Jens Axboe03ba3782009-09-09 09:08:54 +0200442/*
Christoph Hellwigd19de7e2010-06-08 18:14:58 +0200443 * For background writeback the caller does not have the sb pinned
Jens Axboe03ba3782009-09-09 09:08:54 +0200444 * before calling writeback. So make sure that we do pin it, so it doesn't
445 * go away while we are writing inodes from it.
Jens Axboe03ba3782009-09-09 09:08:54 +0200446 */
Christoph Hellwigd19de7e2010-06-08 18:14:58 +0200447static bool pin_sb_for_writeback(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448{
Jens Axboe03ba3782009-09-09 09:08:54 +0200449 spin_lock(&sb_lock);
Christoph Hellwig29cb48592010-06-09 15:31:01 +0200450 if (list_empty(&sb->s_instances)) {
451 spin_unlock(&sb_lock);
452 return false;
453 }
454
Jens Axboe03ba3782009-09-09 09:08:54 +0200455 sb->s_count++;
Christoph Hellwig29cb48592010-06-09 15:31:01 +0200456 spin_unlock(&sb_lock);
457
Jens Axboe03ba3782009-09-09 09:08:54 +0200458 if (down_read_trylock(&sb->s_umount)) {
Christoph Hellwig29cb48592010-06-09 15:31:01 +0200459 if (sb->s_root)
Christoph Hellwigd19de7e2010-06-08 18:14:58 +0200460 return true;
Jens Axboe03ba3782009-09-09 09:08:54 +0200461 up_read(&sb->s_umount);
462 }
Christoph Hellwig29cb48592010-06-09 15:31:01 +0200463
464 put_super(sb);
Christoph Hellwigd19de7e2010-06-08 18:14:58 +0200465 return false;
Jens Axboe03ba3782009-09-09 09:08:54 +0200466}
467
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800468/*
469 * Write a portion of b_io inodes which belong to @sb.
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200470 *
471 * If @only_this_sb is true, then find and write all such
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800472 * inodes. Otherwise write only ones which go sequentially
473 * in reverse order.
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200474 *
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800475 * Return 1, if the caller writeback routine should be
476 * interrupted. Otherwise return 0.
477 */
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200478static int writeback_sb_inodes(struct super_block *sb, struct bdi_writeback *wb,
479 struct writeback_control *wbc, bool only_this_sb)
Jens Axboe03ba3782009-09-09 09:08:54 +0200480{
Jens Axboe03ba3782009-09-09 09:08:54 +0200481 while (!list_empty(&wb->b_io)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482 long pages_skipped;
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100483 struct inode *inode = wb_inode(wb->b_io.prev);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200484
485 if (inode->i_sb != sb) {
486 if (only_this_sb) {
487 /*
488 * We only want to write back data for this
489 * superblock, move all inodes not belonging
490 * to it back onto the dirty list.
491 */
492 redirty_tail(inode);
493 continue;
494 }
495
496 /*
497 * The inode belongs to a different superblock.
498 * Bounce back to the caller to unpin this and
499 * pin the next superblock.
500 */
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800501 return 0;
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200502 }
503
Christoph Hellwig9843b762010-10-24 19:40:46 +0200504 /*
505 * Don't bother with new inodes or inodes beeing freed, first
506 * kind does not need peridic writeout yet, and for the latter
507 * kind writeout is handled by the freer.
508 */
509 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
Nick Piggin7ef0d732009-03-12 14:31:38 -0700510 requeue_io(inode);
511 continue;
512 }
Christoph Hellwig9843b762010-10-24 19:40:46 +0200513
Jeff Laytond2caa3c52009-04-02 16:56:37 -0700514 /*
515 * Was this inode dirtied after sync_sb_inodes was called?
516 * This keeps sync from extra jobs and livelock.
517 */
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800518 if (inode_dirtied_after(inode, wbc->wb_start))
519 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521 __iget(inode);
522 pages_skipped = wbc->pages_skipped;
Christoph Hellwig01c03192009-06-08 13:35:40 +0200523 writeback_single_inode(inode, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524 if (wbc->pages_skipped != pages_skipped) {
525 /*
526 * writeback is not making progress due to locked
527 * buffers. Skip this inode for now.
528 */
Andrew Mortonf57b9b72007-10-16 23:30:34 -0700529 redirty_tail(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530 }
531 spin_unlock(&inode_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 iput(inode);
OGAWA Hirofumi4ffc8442006-03-25 03:07:44 -0800533 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 spin_lock(&inode_lock);
Fengguang Wu8bc3be22008-02-04 22:29:36 -0800535 if (wbc->nr_to_write <= 0) {
536 wbc->more_io = 1;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800537 return 1;
Fengguang Wu8bc3be22008-02-04 22:29:36 -0800538 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200539 if (!list_empty(&wb->b_more_io))
Fengguang Wu8bc3be22008-02-04 22:29:36 -0800540 wbc->more_io = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541 }
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800542 /* b_io is empty */
543 return 1;
544}
Nick Piggin38f21972009-01-06 14:40:25 -0800545
Christoph Hellwig9c3a8ee2010-06-10 12:07:27 +0200546void writeback_inodes_wb(struct bdi_writeback *wb,
547 struct writeback_control *wbc)
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800548{
549 int ret = 0;
Jens Axboe9ecc2732009-09-24 15:25:11 +0200550
Jan Kara7624ee72010-08-09 17:20:03 -0700551 if (!wbc->wb_start)
552 wbc->wb_start = jiffies; /* livelock avoidance */
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800553 spin_lock(&inode_lock);
554 if (!wbc->for_kupdate || list_empty(&wb->b_io))
555 queue_io(wb, wbc->older_than_this);
556
557 while (!list_empty(&wb->b_io)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100558 struct inode *inode = wb_inode(wb->b_io.prev);
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800559 struct super_block *sb = inode->i_sb;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800560
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200561 if (!pin_sb_for_writeback(sb)) {
562 requeue_io(inode);
563 continue;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800564 }
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200565 ret = writeback_sb_inodes(sb, wb, wbc, false);
566 drop_super(sb);
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800567
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800568 if (ret)
569 break;
570 }
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200571 spin_unlock(&inode_lock);
572 /* Leave any unwritten inodes on b_io */
573}
574
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200575static void __writeback_inodes_sb(struct super_block *sb,
576 struct bdi_writeback *wb, struct writeback_control *wbc)
577{
578 WARN_ON(!rwsem_is_locked(&sb->s_umount));
579
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200580 spin_lock(&inode_lock);
581 if (!wbc->for_kupdate || list_empty(&wb->b_io))
582 queue_io(wb, wbc->older_than_this);
583 writeback_sb_inodes(sb, wb, wbc, true);
584 spin_unlock(&inode_lock);
585}
586
Jens Axboe03ba3782009-09-09 09:08:54 +0200587/*
588 * The maximum number of pages to writeout in a single bdi flush/kupdate
589 * operation. We do this so we don't hold I_SYNC against an inode for
590 * enormous amounts of time, which would block a userspace task which has
591 * been forced to throttle against that inode. Also, the code reevaluates
592 * the dirty each time it has written this many pages.
593 */
594#define MAX_WRITEBACK_PAGES 1024
595
596static inline bool over_bground_thresh(void)
597{
598 unsigned long background_thresh, dirty_thresh;
599
Wu Fengguang16c40422010-08-11 14:17:39 -0700600 global_dirty_limits(&background_thresh, &dirty_thresh);
Jens Axboe03ba3782009-09-09 09:08:54 +0200601
602 return (global_page_state(NR_FILE_DIRTY) +
Wu Fengguang4cbec4c2010-10-26 14:21:45 -0700603 global_page_state(NR_UNSTABLE_NFS) > background_thresh);
Jens Axboe03ba3782009-09-09 09:08:54 +0200604}
605
606/*
607 * Explicit flushing or periodic writeback of "old" data.
608 *
609 * Define "old": the first time one of an inode's pages is dirtied, we mark the
610 * dirtying-time in the inode's address_space. So this periodic writeback code
611 * just walks the superblock inode list, writing back any inodes which are
612 * older than a specific point in time.
613 *
614 * Try to run once per dirty_writeback_interval. But if a writeback event
615 * takes longer than a dirty_writeback_interval interval, then leave a
616 * one-second gap.
617 *
618 * older_than_this takes precedence over nr_to_write. So we'll only write back
619 * all dirty pages if they are all attached to "old" mappings.
620 */
Jens Axboec4a77a62009-09-16 15:18:25 +0200621static long wb_writeback(struct bdi_writeback *wb,
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200622 struct wb_writeback_work *work)
Jens Axboe03ba3782009-09-09 09:08:54 +0200623{
624 struct writeback_control wbc = {
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200625 .sync_mode = work->sync_mode,
Jens Axboe03ba3782009-09-09 09:08:54 +0200626 .older_than_this = NULL,
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200627 .for_kupdate = work->for_kupdate,
628 .for_background = work->for_background,
629 .range_cyclic = work->range_cyclic,
Jens Axboe03ba3782009-09-09 09:08:54 +0200630 };
631 unsigned long oldest_jif;
632 long wrote = 0;
Jan Karaa5989bd2009-09-16 19:22:48 +0200633 struct inode *inode;
Jens Axboe03ba3782009-09-09 09:08:54 +0200634
635 if (wbc.for_kupdate) {
636 wbc.older_than_this = &oldest_jif;
637 oldest_jif = jiffies -
638 msecs_to_jiffies(dirty_expire_interval * 10);
639 }
Jens Axboec4a77a62009-09-16 15:18:25 +0200640 if (!wbc.range_cyclic) {
641 wbc.range_start = 0;
642 wbc.range_end = LLONG_MAX;
643 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200644
Jan Kara7624ee72010-08-09 17:20:03 -0700645 wbc.wb_start = jiffies; /* livelock avoidance */
Jens Axboe03ba3782009-09-09 09:08:54 +0200646 for (;;) {
647 /*
Wu Fengguangd3ddec72009-09-23 20:33:40 +0800648 * Stop writeback when nr_pages has been consumed
Jens Axboe03ba3782009-09-09 09:08:54 +0200649 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200650 if (work->nr_pages <= 0)
Jens Axboe03ba3782009-09-09 09:08:54 +0200651 break;
652
653 /*
Jan Karaaa373cf2011-01-13 15:45:47 -0800654 * Background writeout and kupdate-style writeback may
655 * run forever. Stop them if there is other work to do
656 * so that e.g. sync can proceed. They'll be restarted
657 * after the other works are all done.
658 */
659 if ((work->for_background || work->for_kupdate) &&
660 !list_empty(&wb->bdi->work_list))
661 break;
662
663 /*
Wu Fengguangd3ddec72009-09-23 20:33:40 +0800664 * For background writeout, stop when we are below the
665 * background dirty threshold
Jens Axboe03ba3782009-09-09 09:08:54 +0200666 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200667 if (work->for_background && !over_bground_thresh())
Jens Axboe03ba3782009-09-09 09:08:54 +0200668 break;
669
670 wbc.more_io = 0;
Jens Axboe03ba3782009-09-09 09:08:54 +0200671 wbc.nr_to_write = MAX_WRITEBACK_PAGES;
672 wbc.pages_skipped = 0;
Dave Chinner028c2dd2010-07-07 13:24:07 +1000673
674 trace_wbc_writeback_start(&wbc, wb->bdi);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200675 if (work->sb)
676 __writeback_inodes_sb(work->sb, wb, &wbc);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200677 else
678 writeback_inodes_wb(wb, &wbc);
Dave Chinner028c2dd2010-07-07 13:24:07 +1000679 trace_wbc_writeback_written(&wbc, wb->bdi);
680
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200681 work->nr_pages -= MAX_WRITEBACK_PAGES - wbc.nr_to_write;
Jens Axboe03ba3782009-09-09 09:08:54 +0200682 wrote += MAX_WRITEBACK_PAGES - wbc.nr_to_write;
683
684 /*
Jens Axboe71fd05a2009-09-23 19:32:26 +0200685 * If we consumed everything, see if we have more
Jens Axboe03ba3782009-09-09 09:08:54 +0200686 */
Jens Axboe71fd05a2009-09-23 19:32:26 +0200687 if (wbc.nr_to_write <= 0)
688 continue;
689 /*
690 * Didn't write everything and we don't have more IO, bail
691 */
692 if (!wbc.more_io)
Jens Axboe03ba3782009-09-09 09:08:54 +0200693 break;
Jens Axboe71fd05a2009-09-23 19:32:26 +0200694 /*
695 * Did we write something? Try for more
696 */
697 if (wbc.nr_to_write < MAX_WRITEBACK_PAGES)
698 continue;
699 /*
700 * Nothing written. Wait for some inode to
701 * become available for writeback. Otherwise
702 * we'll just busyloop.
703 */
704 spin_lock(&inode_lock);
705 if (!list_empty(&wb->b_more_io)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100706 inode = wb_inode(wb->b_more_io.prev);
Dave Chinner028c2dd2010-07-07 13:24:07 +1000707 trace_wbc_writeback_wait(&wbc, wb->bdi);
Jens Axboe71fd05a2009-09-23 19:32:26 +0200708 inode_wait_for_writeback(inode);
Jens Axboe03ba3782009-09-09 09:08:54 +0200709 }
Jens Axboe71fd05a2009-09-23 19:32:26 +0200710 spin_unlock(&inode_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200711 }
712
713 return wrote;
714}
715
716/*
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200717 * Return the next wb_writeback_work struct that hasn't been processed yet.
Jens Axboe03ba3782009-09-09 09:08:54 +0200718 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200719static struct wb_writeback_work *
Minchan Kim08852b62010-08-03 12:51:16 +0200720get_next_work_item(struct backing_dev_info *bdi)
Jens Axboe03ba3782009-09-09 09:08:54 +0200721{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200722 struct wb_writeback_work *work = NULL;
Jens Axboe03ba3782009-09-09 09:08:54 +0200723
Artem Bityutskiy64677162010-07-25 14:29:22 +0300724 spin_lock_bh(&bdi->wb_lock);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200725 if (!list_empty(&bdi->work_list)) {
726 work = list_entry(bdi->work_list.next,
727 struct wb_writeback_work, list);
728 list_del_init(&work->list);
Jens Axboe03ba3782009-09-09 09:08:54 +0200729 }
Artem Bityutskiy64677162010-07-25 14:29:22 +0300730 spin_unlock_bh(&bdi->wb_lock);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200731 return work;
Jens Axboe03ba3782009-09-09 09:08:54 +0200732}
733
Linus Torvaldscdf01dd2010-10-30 08:55:52 -0700734/*
735 * Add in the number of potentially dirty inodes, because each inode
736 * write can dirty pagecache in the underlying blockdev.
737 */
738static unsigned long get_nr_dirty_pages(void)
739{
740 return global_page_state(NR_FILE_DIRTY) +
741 global_page_state(NR_UNSTABLE_NFS) +
742 get_nr_dirty_inodes();
743}
744
Jan Kara65850272011-01-13 15:45:44 -0800745static long wb_check_background_flush(struct bdi_writeback *wb)
746{
747 if (over_bground_thresh()) {
748
749 struct wb_writeback_work work = {
750 .nr_pages = LONG_MAX,
751 .sync_mode = WB_SYNC_NONE,
752 .for_background = 1,
753 .range_cyclic = 1,
754 };
755
756 return wb_writeback(wb, &work);
757 }
758
759 return 0;
760}
761
Jens Axboe03ba3782009-09-09 09:08:54 +0200762static long wb_check_old_data_flush(struct bdi_writeback *wb)
763{
764 unsigned long expired;
765 long nr_pages;
766
Jens Axboe69b62d02010-05-17 12:51:03 +0200767 /*
768 * When set to zero, disable periodic writeback
769 */
770 if (!dirty_writeback_interval)
771 return 0;
772
Jens Axboe03ba3782009-09-09 09:08:54 +0200773 expired = wb->last_old_flush +
774 msecs_to_jiffies(dirty_writeback_interval * 10);
775 if (time_before(jiffies, expired))
776 return 0;
777
778 wb->last_old_flush = jiffies;
Linus Torvaldscdf01dd2010-10-30 08:55:52 -0700779 nr_pages = get_nr_dirty_pages();
Jens Axboe03ba3782009-09-09 09:08:54 +0200780
Jens Axboec4a77a62009-09-16 15:18:25 +0200781 if (nr_pages) {
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200782 struct wb_writeback_work work = {
Jens Axboec4a77a62009-09-16 15:18:25 +0200783 .nr_pages = nr_pages,
784 .sync_mode = WB_SYNC_NONE,
785 .for_kupdate = 1,
786 .range_cyclic = 1,
787 };
788
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200789 return wb_writeback(wb, &work);
Jens Axboec4a77a62009-09-16 15:18:25 +0200790 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200791
792 return 0;
793}
794
795/*
796 * Retrieve work items and do the writeback they describe
797 */
798long wb_do_writeback(struct bdi_writeback *wb, int force_wait)
799{
800 struct backing_dev_info *bdi = wb->bdi;
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200801 struct wb_writeback_work *work;
Jens Axboec4a77a62009-09-16 15:18:25 +0200802 long wrote = 0;
Jens Axboe03ba3782009-09-09 09:08:54 +0200803
Jan Kara81d73a32010-08-11 14:17:44 -0700804 set_bit(BDI_writeback_running, &wb->bdi->state);
Minchan Kim08852b62010-08-03 12:51:16 +0200805 while ((work = get_next_work_item(bdi)) != NULL) {
Jens Axboe03ba3782009-09-09 09:08:54 +0200806 /*
807 * Override sync mode, in case we must wait for completion
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200808 * because this thread is exiting now.
Jens Axboe03ba3782009-09-09 09:08:54 +0200809 */
810 if (force_wait)
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200811 work->sync_mode = WB_SYNC_ALL;
812
Dave Chinner455b2862010-07-07 13:24:06 +1000813 trace_writeback_exec(bdi, work);
814
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200815 wrote += wb_writeback(wb, work);
Jens Axboe03ba3782009-09-09 09:08:54 +0200816
817 /*
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200818 * Notify the caller of completion if this is a synchronous
819 * work item, otherwise just free it.
Jens Axboe03ba3782009-09-09 09:08:54 +0200820 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200821 if (work->done)
822 complete(work->done);
823 else
824 kfree(work);
Jens Axboe03ba3782009-09-09 09:08:54 +0200825 }
826
827 /*
828 * Check for periodic writeback, kupdated() style
829 */
830 wrote += wb_check_old_data_flush(wb);
Jan Kara65850272011-01-13 15:45:44 -0800831 wrote += wb_check_background_flush(wb);
Jan Kara81d73a32010-08-11 14:17:44 -0700832 clear_bit(BDI_writeback_running, &wb->bdi->state);
Jens Axboe03ba3782009-09-09 09:08:54 +0200833
834 return wrote;
835}
836
837/*
838 * Handle writeback of dirty data for the device backed by this bdi. Also
839 * wakes up periodically and does kupdated style flushing.
840 */
Christoph Hellwig08243902010-06-19 23:08:22 +0200841int bdi_writeback_thread(void *data)
Jens Axboe03ba3782009-09-09 09:08:54 +0200842{
Christoph Hellwig08243902010-06-19 23:08:22 +0200843 struct bdi_writeback *wb = data;
844 struct backing_dev_info *bdi = wb->bdi;
Jens Axboe03ba3782009-09-09 09:08:54 +0200845 long pages_written;
846
Peter Zijlstra766f9162010-10-26 14:22:45 -0700847 current->flags |= PF_SWAPWRITE;
Christoph Hellwig08243902010-06-19 23:08:22 +0200848 set_freezable();
Artem Bityutskiyecd58402010-07-25 14:29:18 +0300849 wb->last_active = jiffies;
Christoph Hellwig08243902010-06-19 23:08:22 +0200850
851 /*
852 * Our parent may run at a different priority, just set us to normal
853 */
854 set_user_nice(current, 0);
855
Dave Chinner455b2862010-07-07 13:24:06 +1000856 trace_writeback_thread_start(bdi);
857
Jens Axboe03ba3782009-09-09 09:08:54 +0200858 while (!kthread_should_stop()) {
Artem Bityutskiy64677162010-07-25 14:29:22 +0300859 /*
860 * Remove own delayed wake-up timer, since we are already awake
861 * and we'll take care of the preriodic write-back.
862 */
863 del_timer(&wb->wakeup_timer);
864
Jens Axboe03ba3782009-09-09 09:08:54 +0200865 pages_written = wb_do_writeback(wb, 0);
866
Dave Chinner455b2862010-07-07 13:24:06 +1000867 trace_writeback_pages_written(pages_written);
Jens Axboe03ba3782009-09-09 09:08:54 +0200868
Jens Axboe03ba3782009-09-09 09:08:54 +0200869 if (pages_written)
Artem Bityutskiyecd58402010-07-25 14:29:18 +0300870 wb->last_active = jiffies;
Jens Axboe03ba3782009-09-09 09:08:54 +0200871
Artem Bityutskiy297252c2010-07-25 14:29:15 +0300872 set_current_state(TASK_INTERRUPTIBLE);
J. Bruce Fieldsb76b4012010-08-28 08:52:10 +0200873 if (!list_empty(&bdi->work_list) || kthread_should_stop()) {
Artem Bityutskiy297252c2010-07-25 14:29:15 +0300874 __set_current_state(TASK_RUNNING);
875 continue;
Jens Axboe03ba3782009-09-09 09:08:54 +0200876 }
877
Artem Bityutskiy253c34e2010-07-25 14:29:21 +0300878 if (wb_has_dirty_io(wb) && dirty_writeback_interval)
Artem Bityutskiyfff5b852010-07-25 14:29:20 +0300879 schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10));
Artem Bityutskiy253c34e2010-07-25 14:29:21 +0300880 else {
881 /*
882 * We have nothing to do, so can go sleep without any
883 * timeout and save power. When a work is queued or
884 * something is made dirty - we will be woken up.
885 */
Artem Bityutskiy297252c2010-07-25 14:29:15 +0300886 schedule();
Jens Axboef9eadbb2010-05-18 14:31:45 +0200887 }
Jens Axboe69b62d02010-05-17 12:51:03 +0200888
Jens Axboe03ba3782009-09-09 09:08:54 +0200889 try_to_freeze();
890 }
891
Artem Bityutskiyfff5b852010-07-25 14:29:20 +0300892 /* Flush any work that raced with us exiting */
Christoph Hellwig08243902010-06-19 23:08:22 +0200893 if (!list_empty(&bdi->work_list))
894 wb_do_writeback(wb, 1);
Dave Chinner455b2862010-07-07 13:24:06 +1000895
896 trace_writeback_thread_stop(bdi);
Jens Axboe03ba3782009-09-09 09:08:54 +0200897 return 0;
898}
899
Christoph Hellwig08243902010-06-19 23:08:22 +0200900
Jens Axboe03ba3782009-09-09 09:08:54 +0200901/*
Jens Axboe03ba3782009-09-09 09:08:54 +0200902 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
903 * the whole world.
904 */
905void wakeup_flusher_threads(long nr_pages)
906{
Christoph Hellwigb8c2f342010-06-08 18:15:07 +0200907 struct backing_dev_info *bdi;
Christoph Hellwigb8c2f342010-06-08 18:15:07 +0200908
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200909 if (!nr_pages) {
910 nr_pages = global_page_state(NR_FILE_DIRTY) +
Jens Axboe03ba3782009-09-09 09:08:54 +0200911 global_page_state(NR_UNSTABLE_NFS);
Christoph Hellwigb8c2f342010-06-08 18:15:07 +0200912 }
913
914 rcu_read_lock();
915 list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) {
916 if (!bdi_has_dirty_io(bdi))
917 continue;
Jan Kara65850272011-01-13 15:45:44 -0800918 __bdi_start_writeback(bdi, nr_pages, false);
Christoph Hellwigb8c2f342010-06-08 18:15:07 +0200919 }
920 rcu_read_unlock();
Jens Axboe03ba3782009-09-09 09:08:54 +0200921}
922
923static noinline void block_dump___mark_inode_dirty(struct inode *inode)
924{
925 if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) {
926 struct dentry *dentry;
927 const char *name = "?";
928
929 dentry = d_find_alias(inode);
930 if (dentry) {
931 spin_lock(&dentry->d_lock);
932 name = (const char *) dentry->d_name.name;
933 }
934 printk(KERN_DEBUG
935 "%s(%d): dirtied inode %lu (%s) on %s\n",
936 current->comm, task_pid_nr(current), inode->i_ino,
937 name, inode->i_sb->s_id);
938 if (dentry) {
939 spin_unlock(&dentry->d_lock);
940 dput(dentry);
941 }
942 }
943}
944
945/**
946 * __mark_inode_dirty - internal function
947 * @inode: inode to mark
948 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
949 * Mark an inode as dirty. Callers should use mark_inode_dirty or
950 * mark_inode_dirty_sync.
951 *
952 * Put the inode on the super block's dirty list.
953 *
954 * CAREFUL! We mark it dirty unconditionally, but move it onto the
955 * dirty list only if it is hashed or if it refers to a blockdev.
956 * If it was not hashed, it will never be added to the dirty list
957 * even if it is later hashed, as it will have been marked dirty already.
958 *
959 * In short, make sure you hash any inodes _before_ you start marking
960 * them dirty.
961 *
962 * This function *must* be atomic for the I_DIRTY_PAGES case -
963 * set_page_dirty() is called under spinlock in several places.
964 *
965 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
966 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
967 * the kernel-internal blockdev inode represents the dirtying time of the
968 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
969 * page->mapping->host, so the page-dirtying time is recorded in the internal
970 * blockdev inode.
971 */
972void __mark_inode_dirty(struct inode *inode, int flags)
973{
974 struct super_block *sb = inode->i_sb;
Artem Bityutskiy253c34e2010-07-25 14:29:21 +0300975 struct backing_dev_info *bdi = NULL;
976 bool wakeup_bdi = false;
Jens Axboe03ba3782009-09-09 09:08:54 +0200977
978 /*
979 * Don't do this for I_DIRTY_PAGES - that doesn't actually
980 * dirty the inode itself
981 */
982 if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
983 if (sb->s_op->dirty_inode)
984 sb->s_op->dirty_inode(inode);
985 }
986
987 /*
988 * make sure that changes are seen by all cpus before we test i_state
989 * -- mikulas
990 */
991 smp_mb();
992
993 /* avoid the locking if we can */
994 if ((inode->i_state & flags) == flags)
995 return;
996
997 if (unlikely(block_dump))
998 block_dump___mark_inode_dirty(inode);
999
1000 spin_lock(&inode_lock);
1001 if ((inode->i_state & flags) != flags) {
1002 const int was_dirty = inode->i_state & I_DIRTY;
1003
1004 inode->i_state |= flags;
1005
1006 /*
1007 * If the inode is being synced, just update its dirty state.
1008 * The unlocker will place the inode on the appropriate
1009 * superblock list, based upon its state.
1010 */
1011 if (inode->i_state & I_SYNC)
1012 goto out;
1013
1014 /*
1015 * Only add valid (hashed) inodes to the superblock's
1016 * dirty list. Add blockdev inodes as well.
1017 */
1018 if (!S_ISBLK(inode->i_mode)) {
Al Viro1d3382cb2010-10-23 15:19:20 -04001019 if (inode_unhashed(inode))
Jens Axboe03ba3782009-09-09 09:08:54 +02001020 goto out;
1021 }
Al Viroa4ffdde2010-06-02 17:38:30 -04001022 if (inode->i_state & I_FREEING)
Jens Axboe03ba3782009-09-09 09:08:54 +02001023 goto out;
1024
1025 /*
1026 * If the inode was already on b_dirty/b_io/b_more_io, don't
1027 * reposition it (that would break b_dirty time-ordering).
1028 */
1029 if (!was_dirty) {
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001030 bdi = inode_to_bdi(inode);
Jens Axboe500b0672009-09-09 09:10:25 +02001031
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001032 if (bdi_cap_writeback_dirty(bdi)) {
1033 WARN(!test_bit(BDI_registered, &bdi->state),
1034 "bdi-%s not registered\n", bdi->name);
1035
1036 /*
1037 * If this is the first dirty inode for this
1038 * bdi, we have to wake-up the corresponding
1039 * bdi thread to make sure background
1040 * write-back happens later.
1041 */
1042 if (!wb_has_dirty_io(&bdi->wb))
1043 wakeup_bdi = true;
Jens Axboe500b0672009-09-09 09:10:25 +02001044 }
Jens Axboe03ba3782009-09-09 09:08:54 +02001045
1046 inode->dirtied_when = jiffies;
Nick Piggin7ccf19a2010-10-21 11:49:30 +11001047 list_move(&inode->i_wb_list, &bdi->wb.b_dirty);
Jens Axboe03ba3782009-09-09 09:08:54 +02001048 }
1049 }
1050out:
1051 spin_unlock(&inode_lock);
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001052
1053 if (wakeup_bdi)
Artem Bityutskiy64677162010-07-25 14:29:22 +03001054 bdi_wakeup_thread_delayed(bdi);
Jens Axboe03ba3782009-09-09 09:08:54 +02001055}
1056EXPORT_SYMBOL(__mark_inode_dirty);
1057
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001058/*
1059 * Write out a superblock's list of dirty inodes. A wait will be performed
1060 * upon no inodes, all inodes or the final one, depending upon sync_mode.
1061 *
1062 * If older_than_this is non-NULL, then only write out inodes which
1063 * had their first dirtying at a time earlier than *older_than_this.
1064 *
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001065 * If `bdi' is non-zero then we're being asked to writeback a specific queue.
1066 * This function assumes that the blockdev superblock's inodes are backed by
1067 * a variety of queues, so all inodes are searched. For other superblocks,
1068 * assume that all inodes are backed by the same queue.
1069 *
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001070 * The inodes to be written are parked on bdi->b_io. They are moved back onto
1071 * bdi->b_dirty as they are selected for writing. This way, none can be missed
1072 * on the writer throttling path, and we get decent balancing between many
1073 * throttled threads: we don't want them all piling up on inode_sync_wait.
1074 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001075static void wait_sb_inodes(struct super_block *sb)
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001076{
Jens Axboe03ba3782009-09-09 09:08:54 +02001077 struct inode *inode, *old_inode = NULL;
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001078
Jens Axboe03ba3782009-09-09 09:08:54 +02001079 /*
1080 * We need to be protected against the filesystem going from
1081 * r/o to r/w or vice versa.
1082 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001083 WARN_ON(!rwsem_is_locked(&sb->s_umount));
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001084
Jens Axboe03ba3782009-09-09 09:08:54 +02001085 spin_lock(&inode_lock);
1086
1087 /*
1088 * Data integrity sync. Must wait for all pages under writeback,
1089 * because there may have been pages dirtied before our sync
1090 * call, but which had writeout started before we write it out.
1091 * In which case, the inode may not be on the dirty list, but
1092 * we still have to wait for that writeout.
1093 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001094 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
Jens Axboe03ba3782009-09-09 09:08:54 +02001095 struct address_space *mapping;
1096
Al Viroa4ffdde2010-06-02 17:38:30 -04001097 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW))
Jens Axboe03ba3782009-09-09 09:08:54 +02001098 continue;
1099 mapping = inode->i_mapping;
1100 if (mapping->nrpages == 0)
1101 continue;
1102 __iget(inode);
1103 spin_unlock(&inode_lock);
1104 /*
1105 * We hold a reference to 'inode' so it couldn't have
1106 * been removed from s_inodes list while we dropped the
1107 * inode_lock. We cannot iput the inode now as we can
1108 * be holding the last reference and we cannot iput it
1109 * under inode_lock. So we keep the reference and iput
1110 * it later.
1111 */
1112 iput(old_inode);
1113 old_inode = inode;
1114
1115 filemap_fdatawait(mapping);
1116
1117 cond_resched();
Nick Piggin38f21972009-01-06 14:40:25 -08001118
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001119 spin_lock(&inode_lock);
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001120 }
Jens Axboe03ba3782009-09-09 09:08:54 +02001121 spin_unlock(&inode_lock);
1122 iput(old_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123}
1124
Jens Axboed8a85592009-09-02 12:34:32 +02001125/**
Chris Mason3259f8b2010-10-29 11:16:17 -04001126 * writeback_inodes_sb_nr - writeback dirty inodes from given super_block
Jens Axboed8a85592009-09-02 12:34:32 +02001127 * @sb: the superblock
Chris Mason3259f8b2010-10-29 11:16:17 -04001128 * @nr: the number of pages to write
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129 *
Jens Axboed8a85592009-09-02 12:34:32 +02001130 * Start writeback on some inodes on this super_block. No guarantees are made
1131 * on how many (if any) will be written, and this function does not wait
Chris Mason3259f8b2010-10-29 11:16:17 -04001132 * for IO completion of submitted IO.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133 */
Chris Mason3259f8b2010-10-29 11:16:17 -04001134void writeback_inodes_sb_nr(struct super_block *sb, unsigned long nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001136 DECLARE_COMPLETION_ONSTACK(done);
1137 struct wb_writeback_work work = {
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001138 .sb = sb,
1139 .sync_mode = WB_SYNC_NONE,
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001140 .done = &done,
Chris Mason3259f8b2010-10-29 11:16:17 -04001141 .nr_pages = nr,
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001142 };
Jens Axboe0e3c9a22010-06-01 11:08:43 +02001143
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001144 WARN_ON(!rwsem_is_locked(&sb->s_umount));
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001145 bdi_queue_work(sb->s_bdi, &work);
1146 wait_for_completion(&done);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147}
Chris Mason3259f8b2010-10-29 11:16:17 -04001148EXPORT_SYMBOL(writeback_inodes_sb_nr);
1149
1150/**
1151 * writeback_inodes_sb - writeback dirty inodes from given super_block
1152 * @sb: the superblock
1153 *
1154 * Start writeback on some inodes on this super_block. No guarantees are made
1155 * on how many (if any) will be written, and this function does not wait
1156 * for IO completion of submitted IO.
1157 */
1158void writeback_inodes_sb(struct super_block *sb)
1159{
Linus Torvalds925d1692010-10-30 09:05:48 -07001160 return writeback_inodes_sb_nr(sb, get_nr_dirty_pages());
Chris Mason3259f8b2010-10-29 11:16:17 -04001161}
Jens Axboed8a85592009-09-02 12:34:32 +02001162EXPORT_SYMBOL(writeback_inodes_sb);
1163
1164/**
Eric Sandeen17bd55d2009-12-23 07:57:07 -05001165 * writeback_inodes_sb_if_idle - start writeback if none underway
1166 * @sb: the superblock
1167 *
1168 * Invoke writeback_inodes_sb if no writeback is currently underway.
1169 * Returns 1 if writeback was started, 0 if not.
1170 */
1171int writeback_inodes_sb_if_idle(struct super_block *sb)
1172{
1173 if (!writeback_in_progress(sb->s_bdi)) {
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001174 down_read(&sb->s_umount);
Eric Sandeen17bd55d2009-12-23 07:57:07 -05001175 writeback_inodes_sb(sb);
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001176 up_read(&sb->s_umount);
Eric Sandeen17bd55d2009-12-23 07:57:07 -05001177 return 1;
1178 } else
1179 return 0;
1180}
1181EXPORT_SYMBOL(writeback_inodes_sb_if_idle);
1182
1183/**
Chris Mason3259f8b2010-10-29 11:16:17 -04001184 * writeback_inodes_sb_if_idle - start writeback if none underway
1185 * @sb: the superblock
1186 * @nr: the number of pages to write
1187 *
1188 * Invoke writeback_inodes_sb if no writeback is currently underway.
1189 * Returns 1 if writeback was started, 0 if not.
1190 */
1191int writeback_inodes_sb_nr_if_idle(struct super_block *sb,
1192 unsigned long nr)
1193{
1194 if (!writeback_in_progress(sb->s_bdi)) {
1195 down_read(&sb->s_umount);
1196 writeback_inodes_sb_nr(sb, nr);
1197 up_read(&sb->s_umount);
1198 return 1;
1199 } else
1200 return 0;
1201}
1202EXPORT_SYMBOL(writeback_inodes_sb_nr_if_idle);
1203
1204/**
Jens Axboed8a85592009-09-02 12:34:32 +02001205 * sync_inodes_sb - sync sb inode pages
1206 * @sb: the superblock
1207 *
1208 * This function writes and waits on any dirty inode belonging to this
1209 * super_block. The number of pages synced is returned.
1210 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001211void sync_inodes_sb(struct super_block *sb)
Jens Axboed8a85592009-09-02 12:34:32 +02001212{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001213 DECLARE_COMPLETION_ONSTACK(done);
1214 struct wb_writeback_work work = {
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001215 .sb = sb,
1216 .sync_mode = WB_SYNC_ALL,
1217 .nr_pages = LONG_MAX,
1218 .range_cyclic = 0,
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001219 .done = &done,
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001220 };
1221
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001222 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1223
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001224 bdi_queue_work(sb->s_bdi, &work);
1225 wait_for_completion(&done);
1226
Jens Axboeb6e51312009-09-16 15:13:54 +02001227 wait_sb_inodes(sb);
Jens Axboed8a85592009-09-02 12:34:32 +02001228}
1229EXPORT_SYMBOL(sync_inodes_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231/**
Andrea Arcangeli7f04c262005-10-30 15:03:05 -08001232 * write_inode_now - write an inode to disk
1233 * @inode: inode to write to disk
1234 * @sync: whether the write should be synchronous or not
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235 *
Andrea Arcangeli7f04c262005-10-30 15:03:05 -08001236 * This function commits an inode to disk immediately if it is dirty. This is
1237 * primarily needed by knfsd.
1238 *
1239 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241int write_inode_now(struct inode *inode, int sync)
1242{
1243 int ret;
1244 struct writeback_control wbc = {
1245 .nr_to_write = LONG_MAX,
Mike Galbraith18914b12008-02-08 04:20:23 -08001246 .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -07001247 .range_start = 0,
1248 .range_end = LLONG_MAX,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249 };
1250
1251 if (!mapping_cap_writeback_dirty(inode->i_mapping))
Andrew Morton49364ce2005-11-07 00:59:15 -08001252 wbc.nr_to_write = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253
1254 might_sleep();
1255 spin_lock(&inode_lock);
Christoph Hellwig01c03192009-06-08 13:35:40 +02001256 ret = writeback_single_inode(inode, &wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257 spin_unlock(&inode_lock);
1258 if (sync)
Joern Engel1c0eeaf2007-10-16 23:30:44 -07001259 inode_sync_wait(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260 return ret;
1261}
1262EXPORT_SYMBOL(write_inode_now);
1263
1264/**
1265 * sync_inode - write an inode and its pages to disk.
1266 * @inode: the inode to sync
1267 * @wbc: controls the writeback mode
1268 *
1269 * sync_inode() will write an inode and its pages to disk. It will also
1270 * correctly update the inode on its superblock's dirty inode lists and will
1271 * update inode->i_state.
1272 *
1273 * The caller must have a ref on the inode.
1274 */
1275int sync_inode(struct inode *inode, struct writeback_control *wbc)
1276{
1277 int ret;
1278
1279 spin_lock(&inode_lock);
Christoph Hellwig01c03192009-06-08 13:35:40 +02001280 ret = writeback_single_inode(inode, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281 spin_unlock(&inode_lock);
1282 return ret;
1283}
1284EXPORT_SYMBOL(sync_inode);
Christoph Hellwigc37650162010-10-06 10:48:20 +02001285
1286/**
1287 * sync_inode - write an inode to disk
1288 * @inode: the inode to sync
1289 * @wait: wait for I/O to complete.
1290 *
1291 * Write an inode to disk and adjust it's dirty state after completion.
1292 *
1293 * Note: only writes the actual inode, no associated data or other metadata.
1294 */
1295int sync_inode_metadata(struct inode *inode, int wait)
1296{
1297 struct writeback_control wbc = {
1298 .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE,
1299 .nr_to_write = 0, /* metadata-only */
1300 };
1301
1302 return sync_inode(inode, &wbc);
1303}
1304EXPORT_SYMBOL(sync_inode_metadata);