blob: 32a8bbd7a9ad1121f9b100da08f80500736bcebf [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>
Paul Gortmaker630d9c42011-11-16 23:57:37 -050017#include <linux/export.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>
Wu Fengguangbc31b862012-01-07 20:41:55 -060023#include <linux/pagemap.h>
Jens Axboe03ba3782009-09-09 09:08:54 +020024#include <linux/kthread.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/writeback.h>
26#include <linux/blkdev.h>
27#include <linux/backing-dev.h>
Dave Chinner455b2862010-07-07 13:24:06 +100028#include <linux/tracepoint.h>
Al Viro719ea2f2013-09-29 11:24:49 -040029#include <linux/device.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/*
Wu Fengguangbc31b862012-01-07 20:41:55 -060033 * 4MB minimal write chunk size
34 */
35#define MIN_WRITEBACK_PAGES (4096UL >> (PAGE_CACHE_SHIFT - 10))
36
37/*
Jens Axboec4a77a62009-09-16 15:18:25 +020038 * Passed into wb_writeback(), essentially a subset of writeback_control
39 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +020040struct wb_writeback_work {
Jens Axboec4a77a62009-09-16 15:18:25 +020041 long nr_pages;
42 struct super_block *sb;
Jan Kara0dc83bd2014-02-21 11:19:04 +010043 unsigned long *older_than_this;
Jens Axboec4a77a62009-09-16 15:18:25 +020044 enum writeback_sync_modes sync_mode;
Wu Fengguang6e6938b2010-06-06 10:38:15 -060045 unsigned int tagged_writepages:1;
H Hartley Sweeten52957fe2010-04-01 20:36:30 -050046 unsigned int for_kupdate:1;
47 unsigned int range_cyclic:1;
48 unsigned int for_background:1;
Dave Chinner7747bd42013-07-02 22:38:35 +100049 unsigned int for_sync:1; /* sync(2) WB_SYNC_ALL writeback */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -060050 enum wb_reason reason; /* why was writeback initiated? */
Jens Axboec4a77a62009-09-16 15:18:25 +020051
Jens Axboe8010c3b2009-09-15 20:04:57 +020052 struct list_head list; /* pending work list */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +020053 struct completion *done; /* set if the caller waits */
Jens Axboe03ba3782009-09-09 09:08:54 +020054};
55
Theodore Ts'oa2f48702015-03-17 12:23:19 -040056/*
57 * If an inode is constantly having its pages dirtied, but then the
58 * updates stop dirtytime_expire_interval seconds in the past, it's
59 * possible for the worst case time between when an inode has its
60 * timestamps updated and when they finally get written out to be two
61 * dirtytime_expire_intervals. We set the default to 12 hours (in
62 * seconds), which means most of the time inodes will have their
63 * timestamps written to disk after 12 hours, but in the worst case a
64 * few inodes might not their timestamps updated for 24 hours.
65 */
66unsigned int dirtytime_expire_interval = 12 * 60 * 60;
67
Adrian Bunkf11b00f2008-04-29 00:58:56 -070068/**
69 * writeback_in_progress - determine whether there is writeback in progress
70 * @bdi: the device's backing_dev_info structure.
71 *
Jens Axboe03ba3782009-09-09 09:08:54 +020072 * Determine whether there is writeback waiting to be handled against a
73 * backing device.
Adrian Bunkf11b00f2008-04-29 00:58:56 -070074 */
75int writeback_in_progress(struct backing_dev_info *bdi)
76{
Jan Kara81d73a32010-08-11 14:17:44 -070077 return test_bit(BDI_writeback_running, &bdi->state);
Adrian Bunkf11b00f2008-04-29 00:58:56 -070078}
Theodore Ts'o00d4e732012-09-19 22:42:36 -040079EXPORT_SYMBOL(writeback_in_progress);
Adrian Bunkf11b00f2008-04-29 00:58:56 -070080
Christoph Hellwigde1414a2015-01-14 10:42:36 +010081struct backing_dev_info *inode_to_bdi(struct inode *inode)
Jan Kara692ebd12010-09-21 11:51:01 +020082{
Jens Axboeb5202522015-01-22 08:13:17 -070083 struct super_block *sb;
84
85 if (!inode)
86 return &noop_backing_dev_info;
87
88 sb = inode->i_sb;
Christoph Hellwig495a2762015-01-14 10:42:34 +010089#ifdef CONFIG_BLOCK
Jan Karaa8855992013-07-09 22:36:45 +080090 if (sb_is_blkdev_sb(sb))
Christoph Hellwig495a2762015-01-14 10:42:34 +010091 return blk_get_backing_dev_info(I_BDEV(inode));
92#endif
Christoph Hellwigaaead252010-10-04 14:25:33 +020093 return sb->s_bdi;
Jan Kara692ebd12010-09-21 11:51:01 +020094}
Christoph Hellwigde1414a2015-01-14 10:42:36 +010095EXPORT_SYMBOL_GPL(inode_to_bdi);
Jan Kara692ebd12010-09-21 11:51:01 +020096
Nick Piggin7ccf19a2010-10-21 11:49:30 +110097static inline struct inode *wb_inode(struct list_head *head)
98{
99 return list_entry(head, struct inode, i_wb_list);
100}
101
Wu Fengguang15eb77a2012-01-17 11:18:56 -0600102/*
103 * Include the creation of the trace points after defining the
104 * wb_writeback_work structure and inline functions so that the definition
105 * remains local to this file.
106 */
107#define CREATE_TRACE_POINTS
108#include <trace/events/writeback.h>
109
Steven Whitehouse774016b2014-02-06 15:47:47 +0000110EXPORT_TRACEPOINT_SYMBOL_GPL(wbc_writepage);
111
Jan Kara5acda9d2014-04-03 14:46:23 -0700112static void bdi_wakeup_thread(struct backing_dev_info *bdi)
113{
114 spin_lock_bh(&bdi->wb_lock);
115 if (test_bit(BDI_registered, &bdi->state))
116 mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0);
117 spin_unlock_bh(&bdi->wb_lock);
118}
119
Jan Kara65850272011-01-13 15:45:44 -0800120static void bdi_queue_work(struct backing_dev_info *bdi,
121 struct wb_writeback_work *work)
122{
123 trace_writeback_queue(bdi, work);
124
125 spin_lock_bh(&bdi->wb_lock);
Jan Kara5acda9d2014-04-03 14:46:23 -0700126 if (!test_bit(BDI_registered, &bdi->state)) {
127 if (work->done)
128 complete(work->done);
129 goto out_unlock;
130 }
Jan Kara65850272011-01-13 15:45:44 -0800131 list_add_tail(&work->list, &bdi->work_list);
Tejun Heo839a8e82013-04-01 19:08:06 -0700132 mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0);
Jan Kara5acda9d2014-04-03 14:46:23 -0700133out_unlock:
134 spin_unlock_bh(&bdi->wb_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135}
136
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200137static void
138__bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600139 bool range_cyclic, enum wb_reason reason)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200141 struct wb_writeback_work *work;
Jens Axboe03ba3782009-09-09 09:08:54 +0200142
Jens Axboebcddc3f2009-09-13 20:07:36 +0200143 /*
144 * This is WB_SYNC_NONE writeback, so if allocation fails just
145 * wakeup the thread for old dirty data writeback
146 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200147 work = kzalloc(sizeof(*work), GFP_ATOMIC);
148 if (!work) {
Tejun Heo839a8e82013-04-01 19:08:06 -0700149 trace_writeback_nowork(bdi);
Jan Kara5acda9d2014-04-03 14:46:23 -0700150 bdi_wakeup_thread(bdi);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200151 return;
Jens Axboebcddc3f2009-09-13 20:07:36 +0200152 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200153
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200154 work->sync_mode = WB_SYNC_NONE;
155 work->nr_pages = nr_pages;
156 work->range_cyclic = range_cyclic;
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600157 work->reason = reason;
Christoph Hellwigf0fad8a2009-09-11 09:47:56 +0200158
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200159 bdi_queue_work(bdi, work);
Jens Axboeb6e51312009-09-16 15:13:54 +0200160}
161
162/**
163 * bdi_start_writeback - start writeback
164 * @bdi: the backing device to write from
165 * @nr_pages: the number of pages to write
Marcos Paulo de Souza786228a2011-11-23 20:56:45 +0800166 * @reason: reason why some writeback work was initiated
Jens Axboeb6e51312009-09-16 15:13:54 +0200167 *
168 * Description:
169 * This does WB_SYNC_NONE opportunistic writeback. The IO is only
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300170 * started when this function returns, we make no guarantees on
Jens Axboe0e3c9a22010-06-01 11:08:43 +0200171 * completion. Caller need not hold sb s_umount semaphore.
Jens Axboeb6e51312009-09-16 15:13:54 +0200172 *
173 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600174void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages,
175 enum wb_reason reason)
Jens Axboeb6e51312009-09-16 15:13:54 +0200176{
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600177 __bdi_start_writeback(bdi, nr_pages, true, reason);
Christoph Hellwigc5444192010-06-08 18:15:15 +0200178}
Wu Fengguangd3ddec72009-09-23 20:33:40 +0800179
Christoph Hellwigc5444192010-06-08 18:15:15 +0200180/**
181 * bdi_start_background_writeback - start background writeback
182 * @bdi: the backing device to write from
183 *
184 * Description:
Jan Kara65850272011-01-13 15:45:44 -0800185 * This makes sure WB_SYNC_NONE background writeback happens. When
186 * this function returns, it is only guaranteed that for given BDI
187 * some IO is happening if we are over background dirty threshold.
188 * Caller need not hold sb s_umount semaphore.
Christoph Hellwigc5444192010-06-08 18:15:15 +0200189 */
190void bdi_start_background_writeback(struct backing_dev_info *bdi)
191{
Jan Kara65850272011-01-13 15:45:44 -0800192 /*
193 * We just wake up the flusher thread. It will perform background
194 * writeback as soon as there is no other work to do.
195 */
Wu Fengguang71927e82011-01-13 15:45:46 -0800196 trace_writeback_wake_background(bdi);
Jan Kara5acda9d2014-04-03 14:46:23 -0700197 bdi_wakeup_thread(bdi);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198}
199
200/*
Dave Chinnera66979a2011-03-22 22:23:41 +1100201 * Remove the inode from the writeback list it is on.
202 */
203void inode_wb_list_del(struct inode *inode)
204{
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600205 struct backing_dev_info *bdi = inode_to_bdi(inode);
Dave Chinnera66979a2011-03-22 22:23:41 +1100206
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600207 spin_lock(&bdi->wb.list_lock);
208 list_del_init(&inode->i_wb_list);
209 spin_unlock(&bdi->wb.list_lock);
210}
Dave Chinnera66979a2011-03-22 22:23:41 +1100211
212/*
Andrew Morton6610a0b2007-10-16 23:30:32 -0700213 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
214 * furthest end of its superblock's dirty-inode list.
215 *
216 * Before stamping the inode's ->dirtied_when, we check to see whether it is
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200217 * already the most-recently-dirtied inode on the b_dirty list. If that is
Andrew Morton6610a0b2007-10-16 23:30:32 -0700218 * the case then the inode must have been redirtied while it was being written
219 * out and we don't reset its dirtied_when.
220 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600221static void redirty_tail(struct inode *inode, struct bdi_writeback *wb)
Andrew Morton6610a0b2007-10-16 23:30:32 -0700222{
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600223 assert_spin_locked(&wb->list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200224 if (!list_empty(&wb->b_dirty)) {
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200225 struct inode *tail;
Andrew Morton6610a0b2007-10-16 23:30:32 -0700226
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100227 tail = wb_inode(wb->b_dirty.next);
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200228 if (time_before(inode->dirtied_when, tail->dirtied_when))
Andrew Morton6610a0b2007-10-16 23:30:32 -0700229 inode->dirtied_when = jiffies;
230 }
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100231 list_move(&inode->i_wb_list, &wb->b_dirty);
Andrew Morton6610a0b2007-10-16 23:30:32 -0700232}
233
234/*
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200235 * requeue inode for re-scanning after bdi->b_io list is exhausted.
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700236 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600237static void requeue_io(struct inode *inode, struct bdi_writeback *wb)
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700238{
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600239 assert_spin_locked(&wb->list_lock);
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100240 list_move(&inode->i_wb_list, &wb->b_more_io);
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700241}
242
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700243static void inode_sync_complete(struct inode *inode)
244{
Jan Kara365b94ae2012-05-03 14:47:55 +0200245 inode->i_state &= ~I_SYNC;
Jan Kara4eff96d2012-11-26 16:29:51 -0800246 /* If inode is clean an unused, put it into LRU now... */
247 inode_add_lru(inode);
Jan Kara365b94ae2012-05-03 14:47:55 +0200248 /* Waiters must see I_SYNC cleared before being woken up */
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700249 smp_mb();
250 wake_up_bit(&inode->i_state, __I_SYNC);
251}
252
Jeff Laytond2caa3c52009-04-02 16:56:37 -0700253static bool inode_dirtied_after(struct inode *inode, unsigned long t)
254{
255 bool ret = time_after(inode->dirtied_when, t);
256#ifndef CONFIG_64BIT
257 /*
258 * For inodes being constantly redirtied, dirtied_when can get stuck.
259 * It _appears_ to be in the future, but is actually in distant past.
260 * This test is necessary to prevent such wrapped-around relative times
Jens Axboe5b0830c2009-09-23 19:37:09 +0200261 * from permanently stopping the whole bdi writeback.
Jeff Laytond2caa3c52009-04-02 16:56:37 -0700262 */
263 ret = ret && time_before_eq(inode->dirtied_when, jiffies);
264#endif
265 return ret;
266}
267
Theodore Ts'o0ae45f62015-02-02 00:37:00 -0500268#define EXPIRE_DIRTY_ATIME 0x0001
269
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700270/*
Wang Sheng-Hui0e2f2b22012-09-11 08:28:18 +0800271 * Move expired (dirtied before work->older_than_this) dirty inodes from
Jan Kara697e6fe2012-03-09 07:26:22 -0800272 * @delaying_queue to @dispatch_queue.
Fengguang Wu2c136572007-10-16 23:30:39 -0700273 */
Wu Fengguange84d0a42011-04-23 12:27:27 -0600274static int move_expired_inodes(struct list_head *delaying_queue,
Fengguang Wu2c136572007-10-16 23:30:39 -0700275 struct list_head *dispatch_queue,
Theodore Ts'o0ae45f62015-02-02 00:37:00 -0500276 int flags,
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600277 struct wb_writeback_work *work)
Fengguang Wu2c136572007-10-16 23:30:39 -0700278{
Theodore Ts'o0ae45f62015-02-02 00:37:00 -0500279 unsigned long *older_than_this = NULL;
280 unsigned long expire_time;
Shaohua Li5c034492009-09-24 14:42:33 +0200281 LIST_HEAD(tmp);
282 struct list_head *pos, *node;
Jens Axboecf137302009-09-24 15:12:57 +0200283 struct super_block *sb = NULL;
Shaohua Li5c034492009-09-24 14:42:33 +0200284 struct inode *inode;
Jens Axboecf137302009-09-24 15:12:57 +0200285 int do_sb_sort = 0;
Wu Fengguange84d0a42011-04-23 12:27:27 -0600286 int moved = 0;
Shaohua Li5c034492009-09-24 14:42:33 +0200287
Theodore Ts'o0ae45f62015-02-02 00:37:00 -0500288 if ((flags & EXPIRE_DIRTY_ATIME) == 0)
289 older_than_this = work->older_than_this;
Theodore Ts'oa2f48702015-03-17 12:23:19 -0400290 else if (!work->for_sync) {
291 expire_time = jiffies - (dirtytime_expire_interval * HZ);
Theodore Ts'o0ae45f62015-02-02 00:37:00 -0500292 older_than_this = &expire_time;
293 }
Fengguang Wu2c136572007-10-16 23:30:39 -0700294 while (!list_empty(delaying_queue)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100295 inode = wb_inode(delaying_queue->prev);
Theodore Ts'o0ae45f62015-02-02 00:37:00 -0500296 if (older_than_this &&
297 inode_dirtied_after(inode, *older_than_this))
Fengguang Wu2c136572007-10-16 23:30:39 -0700298 break;
Jan Karaa8855992013-07-09 22:36:45 +0800299 list_move(&inode->i_wb_list, &tmp);
300 moved++;
Theodore Ts'o0ae45f62015-02-02 00:37:00 -0500301 if (flags & EXPIRE_DIRTY_ATIME)
302 set_bit(__I_DIRTY_TIME_EXPIRED, &inode->i_state);
Jan Karaa8855992013-07-09 22:36:45 +0800303 if (sb_is_blkdev_sb(inode->i_sb))
304 continue;
Jens Axboecf137302009-09-24 15:12:57 +0200305 if (sb && sb != inode->i_sb)
306 do_sb_sort = 1;
307 sb = inode->i_sb;
Shaohua Li5c034492009-09-24 14:42:33 +0200308 }
309
Jens Axboecf137302009-09-24 15:12:57 +0200310 /* just one sb in list, splice to dispatch_queue and we're done */
311 if (!do_sb_sort) {
312 list_splice(&tmp, dispatch_queue);
Wu Fengguange84d0a42011-04-23 12:27:27 -0600313 goto out;
Jens Axboecf137302009-09-24 15:12:57 +0200314 }
315
Shaohua Li5c034492009-09-24 14:42:33 +0200316 /* Move inodes from one superblock together */
317 while (!list_empty(&tmp)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100318 sb = wb_inode(tmp.prev)->i_sb;
Shaohua Li5c034492009-09-24 14:42:33 +0200319 list_for_each_prev_safe(pos, node, &tmp) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100320 inode = wb_inode(pos);
Shaohua Li5c034492009-09-24 14:42:33 +0200321 if (inode->i_sb == sb)
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100322 list_move(&inode->i_wb_list, dispatch_queue);
Shaohua Li5c034492009-09-24 14:42:33 +0200323 }
Fengguang Wu2c136572007-10-16 23:30:39 -0700324 }
Wu Fengguange84d0a42011-04-23 12:27:27 -0600325out:
326 return moved;
Fengguang Wu2c136572007-10-16 23:30:39 -0700327}
328
329/*
330 * Queue all expired dirty inodes for io, eldest first.
Wu Fengguang4ea879b2010-08-11 14:17:42 -0700331 * Before
332 * newly dirtied b_dirty b_io b_more_io
333 * =============> gf edc BA
334 * After
335 * newly dirtied b_dirty b_io b_more_io
336 * =============> g fBAedc
337 * |
338 * +--> dequeue for IO
Fengguang Wu2c136572007-10-16 23:30:39 -0700339 */
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600340static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work)
Fengguang Wu2c136572007-10-16 23:30:39 -0700341{
Wu Fengguange84d0a42011-04-23 12:27:27 -0600342 int moved;
Theodore Ts'o0ae45f62015-02-02 00:37:00 -0500343
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600344 assert_spin_locked(&wb->list_lock);
Wu Fengguang4ea879b2010-08-11 14:17:42 -0700345 list_splice_init(&wb->b_more_io, &wb->b_io);
Theodore Ts'o0ae45f62015-02-02 00:37:00 -0500346 moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, 0, work);
347 moved += move_expired_inodes(&wb->b_dirty_time, &wb->b_io,
348 EXPIRE_DIRTY_ATIME, work);
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600349 trace_writeback_queue_io(wb, work, moved);
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200350}
351
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100352static int write_inode(struct inode *inode, struct writeback_control *wbc)
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200353{
Tejun Heo9fb0a7d2013-01-11 13:06:37 -0800354 int ret;
355
356 if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) {
357 trace_writeback_write_inode_start(inode, wbc);
358 ret = inode->i_sb->s_op->write_inode(inode, wbc);
359 trace_writeback_write_inode(inode, wbc);
360 return ret;
361 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200362 return 0;
Fengguang Wu2c136572007-10-16 23:30:39 -0700363}
364
365/*
Jan Kara169ebd92012-05-03 14:48:03 +0200366 * Wait for writeback on an inode to complete. Called with i_lock held.
367 * Caller must make sure inode cannot go away when we drop i_lock.
Christoph Hellwig01c03192009-06-08 13:35:40 +0200368 */
Jan Kara169ebd92012-05-03 14:48:03 +0200369static void __inode_wait_for_writeback(struct inode *inode)
370 __releases(inode->i_lock)
371 __acquires(inode->i_lock)
Christoph Hellwig01c03192009-06-08 13:35:40 +0200372{
373 DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC);
374 wait_queue_head_t *wqh;
375
376 wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
Dave Chinner250df6e2011-03-22 22:23:36 +1100377 while (inode->i_state & I_SYNC) {
378 spin_unlock(&inode->i_lock);
NeilBrown74316202014-07-07 15:16:04 +1000379 __wait_on_bit(wqh, &wq, bit_wait,
380 TASK_UNINTERRUPTIBLE);
Dave Chinner250df6e2011-03-22 22:23:36 +1100381 spin_lock(&inode->i_lock);
Richard Kennedy58a9d3d82010-05-24 14:32:38 -0700382 }
Christoph Hellwig01c03192009-06-08 13:35:40 +0200383}
384
385/*
Jan Kara169ebd92012-05-03 14:48:03 +0200386 * Wait for writeback on an inode to complete. Caller must have inode pinned.
387 */
388void inode_wait_for_writeback(struct inode *inode)
389{
390 spin_lock(&inode->i_lock);
391 __inode_wait_for_writeback(inode);
392 spin_unlock(&inode->i_lock);
393}
394
395/*
396 * Sleep until I_SYNC is cleared. This function must be called with i_lock
397 * held and drops it. It is aimed for callers not holding any inode reference
398 * so once i_lock is dropped, inode can go away.
399 */
400static void inode_sleep_on_writeback(struct inode *inode)
401 __releases(inode->i_lock)
402{
403 DEFINE_WAIT(wait);
404 wait_queue_head_t *wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
405 int sleep;
406
407 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
408 sleep = inode->i_state & I_SYNC;
409 spin_unlock(&inode->i_lock);
410 if (sleep)
411 schedule();
412 finish_wait(wqh, &wait);
413}
414
415/*
Jan Karaccb26b52012-05-03 14:47:58 +0200416 * Find proper writeback list for the inode depending on its current state and
417 * possibly also change of its state while we were doing writeback. Here we
418 * handle things such as livelock prevention or fairness of writeback among
419 * inodes. This function can be called only by flusher thread - noone else
420 * processes all inodes in writeback lists and requeueing inodes behind flusher
421 * thread's back can have unexpected consequences.
422 */
423static void requeue_inode(struct inode *inode, struct bdi_writeback *wb,
424 struct writeback_control *wbc)
425{
426 if (inode->i_state & I_FREEING)
427 return;
428
429 /*
430 * Sync livelock prevention. Each inode is tagged and synced in one
431 * shot. If still dirty, it will be redirty_tail()'ed below. Update
432 * the dirty time to prevent enqueue and sync it again.
433 */
434 if ((inode->i_state & I_DIRTY) &&
435 (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages))
436 inode->dirtied_when = jiffies;
437
Jan Kara4f8ad652012-05-03 14:48:00 +0200438 if (wbc->pages_skipped) {
439 /*
440 * writeback is not making progress due to locked
441 * buffers. Skip this inode for now.
442 */
443 redirty_tail(inode, wb);
444 return;
445 }
446
Jan Karaccb26b52012-05-03 14:47:58 +0200447 if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) {
448 /*
449 * We didn't write back all the pages. nfs_writepages()
450 * sometimes bales out without doing anything.
451 */
452 if (wbc->nr_to_write <= 0) {
453 /* Slice used up. Queue for next turn. */
454 requeue_io(inode, wb);
455 } else {
456 /*
457 * Writeback blocked by something other than
458 * congestion. Delay the inode for some time to
459 * avoid spinning on the CPU (100% iowait)
460 * retrying writeback of the dirty page/inode
461 * that cannot be performed immediately.
462 */
463 redirty_tail(inode, wb);
464 }
465 } else if (inode->i_state & I_DIRTY) {
466 /*
467 * Filesystems can dirty the inode during writeback operations,
468 * such as delayed allocation during submission or metadata
469 * updates after data IO completion.
470 */
471 redirty_tail(inode, wb);
Theodore Ts'o0ae45f62015-02-02 00:37:00 -0500472 } else if (inode->i_state & I_DIRTY_TIME) {
Theodore Ts'oa2f48702015-03-17 12:23:19 -0400473 inode->dirtied_when = jiffies;
Theodore Ts'o0ae45f62015-02-02 00:37:00 -0500474 list_move(&inode->i_wb_list, &wb->b_dirty_time);
Jan Karaccb26b52012-05-03 14:47:58 +0200475 } else {
476 /* The inode is clean. Remove from writeback lists. */
477 list_del_init(&inode->i_wb_list);
478 }
479}
480
481/*
Jan Kara4f8ad652012-05-03 14:48:00 +0200482 * Write out an inode and its dirty pages. Do not update the writeback list
483 * linkage. That is left to the caller. The caller is also responsible for
484 * setting I_SYNC flag and calling inode_sync_complete() to clear it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485 */
486static int
Yan Hongcd8ed2a2012-10-08 16:33:45 -0700487__writeback_single_inode(struct inode *inode, struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489 struct address_space *mapping = inode->i_mapping;
Wu Fengguang251d6a42010-12-01 17:33:37 -0600490 long nr_to_write = wbc->nr_to_write;
Christoph Hellwig01c03192009-06-08 13:35:40 +0200491 unsigned dirty;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492 int ret;
493
Jan Kara4f8ad652012-05-03 14:48:00 +0200494 WARN_ON(!(inode->i_state & I_SYNC));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495
Tejun Heo9fb0a7d2013-01-11 13:06:37 -0800496 trace_writeback_single_inode_start(inode, wbc, nr_to_write);
497
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498 ret = do_writepages(mapping, wbc);
499
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100500 /*
501 * Make sure to wait on the data before writing out the metadata.
502 * This is important for filesystems that modify metadata on data
Dave Chinner7747bd42013-07-02 22:38:35 +1000503 * I/O completion. We don't do it for sync(2) writeback because it has a
504 * separate, external IO completion path and ->sync_fs for guaranteeing
505 * inode metadata is written back correctly.
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100506 */
Dave Chinner7747bd42013-07-02 22:38:35 +1000507 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) {
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100508 int err = filemap_fdatawait(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509 if (ret == 0)
510 ret = err;
511 }
512
Dmitry Monakhov5547e8a2010-05-07 13:35:44 +0400513 /*
514 * Some filesystems may redirty the inode during the writeback
515 * due to delalloc, clear dirty metadata flags right before
516 * write_inode()
517 */
Dave Chinner250df6e2011-03-22 22:23:36 +1100518 spin_lock(&inode->i_lock);
Tejun Heo9c6ac782014-10-24 15:38:21 -0400519
Dmitry Monakhov5547e8a2010-05-07 13:35:44 +0400520 dirty = inode->i_state & I_DIRTY;
Theodore Ts'oa2f48702015-03-17 12:23:19 -0400521 if (inode->i_state & I_DIRTY_TIME) {
522 if ((dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) ||
523 unlikely(inode->i_state & I_DIRTY_TIME_EXPIRED) ||
524 unlikely(time_after(jiffies,
525 (inode->dirtied_time_when +
526 dirtytime_expire_interval * HZ)))) {
527 dirty |= I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED;
528 trace_writeback_lazytime(inode);
529 }
530 } else
531 inode->i_state &= ~I_DIRTY_TIME_EXPIRED;
Theodore Ts'o0ae45f62015-02-02 00:37:00 -0500532 inode->i_state &= ~dirty;
Tejun Heo9c6ac782014-10-24 15:38:21 -0400533
534 /*
535 * Paired with smp_mb() in __mark_inode_dirty(). This allows
536 * __mark_inode_dirty() to test i_state without grabbing i_lock -
537 * either they see the I_DIRTY bits cleared or we see the dirtied
538 * inode.
539 *
540 * I_DIRTY_PAGES is always cleared together above even if @mapping
541 * still has dirty pages. The flag is reinstated after smp_mb() if
542 * necessary. This guarantees that either __mark_inode_dirty()
543 * sees clear I_DIRTY_PAGES or we see PAGECACHE_TAG_DIRTY.
544 */
545 smp_mb();
546
547 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
548 inode->i_state |= I_DIRTY_PAGES;
549
Dave Chinner250df6e2011-03-22 22:23:36 +1100550 spin_unlock(&inode->i_lock);
Tejun Heo9c6ac782014-10-24 15:38:21 -0400551
Theodore Ts'o0ae45f62015-02-02 00:37:00 -0500552 if (dirty & I_DIRTY_TIME)
553 mark_inode_dirty_sync(inode);
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100554 /* Don't write the inode if only I_DIRTY_PAGES was set */
Theodore Ts'o0ae45f62015-02-02 00:37:00 -0500555 if (dirty & ~I_DIRTY_PAGES) {
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100556 int err = write_inode(inode, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557 if (ret == 0)
558 ret = err;
559 }
Jan Kara4f8ad652012-05-03 14:48:00 +0200560 trace_writeback_single_inode(inode, wbc, nr_to_write);
561 return ret;
562}
563
564/*
565 * Write out an inode's dirty pages. Either the caller has an active reference
566 * on the inode or the inode has I_WILL_FREE set.
567 *
568 * This function is designed to be called for writing back one inode which
569 * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode()
570 * and does more profound writeback list handling in writeback_sb_inodes().
571 */
572static int
573writeback_single_inode(struct inode *inode, struct bdi_writeback *wb,
574 struct writeback_control *wbc)
575{
576 int ret = 0;
577
578 spin_lock(&inode->i_lock);
579 if (!atomic_read(&inode->i_count))
580 WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING)));
581 else
582 WARN_ON(inode->i_state & I_WILL_FREE);
583
584 if (inode->i_state & I_SYNC) {
585 if (wbc->sync_mode != WB_SYNC_ALL)
586 goto out;
587 /*
Jan Kara169ebd92012-05-03 14:48:03 +0200588 * It's a data-integrity sync. We must wait. Since callers hold
589 * inode reference or inode has I_WILL_FREE set, it cannot go
590 * away under us.
Jan Kara4f8ad652012-05-03 14:48:00 +0200591 */
Jan Kara169ebd92012-05-03 14:48:03 +0200592 __inode_wait_for_writeback(inode);
Jan Kara4f8ad652012-05-03 14:48:00 +0200593 }
594 WARN_ON(inode->i_state & I_SYNC);
595 /*
Jan Karaf9b0e052013-12-14 04:21:26 +0800596 * Skip inode if it is clean and we have no outstanding writeback in
597 * WB_SYNC_ALL mode. We don't want to mess with writeback lists in this
598 * function since flusher thread may be doing for example sync in
599 * parallel and if we move the inode, it could get skipped. So here we
600 * make sure inode is on some writeback list and leave it there unless
601 * we have completely cleaned the inode.
Jan Kara4f8ad652012-05-03 14:48:00 +0200602 */
Theodore Ts'o0ae45f62015-02-02 00:37:00 -0500603 if (!(inode->i_state & I_DIRTY_ALL) &&
Jan Karaf9b0e052013-12-14 04:21:26 +0800604 (wbc->sync_mode != WB_SYNC_ALL ||
605 !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK)))
Jan Kara4f8ad652012-05-03 14:48:00 +0200606 goto out;
607 inode->i_state |= I_SYNC;
608 spin_unlock(&inode->i_lock);
609
Yan Hongcd8ed2a2012-10-08 16:33:45 -0700610 ret = __writeback_single_inode(inode, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600612 spin_lock(&wb->list_lock);
Dave Chinner250df6e2011-03-22 22:23:36 +1100613 spin_lock(&inode->i_lock);
Jan Kara4f8ad652012-05-03 14:48:00 +0200614 /*
615 * If inode is clean, remove it from writeback lists. Otherwise don't
616 * touch it. See comment above for explanation.
617 */
Theodore Ts'o0ae45f62015-02-02 00:37:00 -0500618 if (!(inode->i_state & I_DIRTY_ALL))
Jan Kara4f8ad652012-05-03 14:48:00 +0200619 list_del_init(&inode->i_wb_list);
620 spin_unlock(&wb->list_lock);
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700621 inode_sync_complete(inode);
Jan Kara4f8ad652012-05-03 14:48:00 +0200622out:
623 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 return ret;
625}
626
Wu Fengguang1a12d8b2010-08-29 13:28:09 -0600627static long writeback_chunk_size(struct backing_dev_info *bdi,
628 struct wb_writeback_work *work)
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600629{
630 long pages;
631
632 /*
633 * WB_SYNC_ALL mode does livelock avoidance by syncing dirty
634 * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX
635 * here avoids calling into writeback_inodes_wb() more than once.
636 *
637 * The intended call sequence for WB_SYNC_ALL writeback is:
638 *
639 * wb_writeback()
640 * writeback_sb_inodes() <== called only once
641 * write_cache_pages() <== called once for each inode
642 * (quickly) tag currently dirty pages
643 * (maybe slowly) sync all tagged pages
644 */
645 if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages)
646 pages = LONG_MAX;
Wu Fengguang1a12d8b2010-08-29 13:28:09 -0600647 else {
648 pages = min(bdi->avg_write_bandwidth / 2,
649 global_dirty_limit / DIRTY_SCOPE);
650 pages = min(pages, work->nr_pages);
651 pages = round_down(pages + MIN_WRITEBACK_PAGES,
652 MIN_WRITEBACK_PAGES);
653 }
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600654
655 return pages;
656}
657
Jens Axboe03ba3782009-09-09 09:08:54 +0200658/*
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800659 * Write a portion of b_io inodes which belong to @sb.
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200660 *
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600661 * Return the number of pages and/or inodes written.
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800662 */
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600663static long writeback_sb_inodes(struct super_block *sb,
664 struct bdi_writeback *wb,
665 struct wb_writeback_work *work)
Jens Axboe03ba3782009-09-09 09:08:54 +0200666{
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600667 struct writeback_control wbc = {
668 .sync_mode = work->sync_mode,
669 .tagged_writepages = work->tagged_writepages,
670 .for_kupdate = work->for_kupdate,
671 .for_background = work->for_background,
Dave Chinner7747bd42013-07-02 22:38:35 +1000672 .for_sync = work->for_sync,
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600673 .range_cyclic = work->range_cyclic,
674 .range_start = 0,
675 .range_end = LLONG_MAX,
676 };
677 unsigned long start_time = jiffies;
678 long write_chunk;
679 long wrote = 0; /* count both pages and inodes */
680
Jens Axboe03ba3782009-09-09 09:08:54 +0200681 while (!list_empty(&wb->b_io)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100682 struct inode *inode = wb_inode(wb->b_io.prev);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200683
684 if (inode->i_sb != sb) {
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600685 if (work->sb) {
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200686 /*
687 * We only want to write back data for this
688 * superblock, move all inodes not belonging
689 * to it back onto the dirty list.
690 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600691 redirty_tail(inode, wb);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200692 continue;
693 }
694
695 /*
696 * The inode belongs to a different superblock.
697 * Bounce back to the caller to unpin this and
698 * pin the next superblock.
699 */
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600700 break;
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200701 }
702
Christoph Hellwig9843b762010-10-24 19:40:46 +0200703 /*
Wanpeng Li331cbde2012-06-09 11:10:55 +0800704 * Don't bother with new inodes or inodes being freed, first
705 * kind does not need periodic writeout yet, and for the latter
Christoph Hellwig9843b762010-10-24 19:40:46 +0200706 * kind writeout is handled by the freer.
707 */
Dave Chinner250df6e2011-03-22 22:23:36 +1100708 spin_lock(&inode->i_lock);
Christoph Hellwig9843b762010-10-24 19:40:46 +0200709 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
Dave Chinner250df6e2011-03-22 22:23:36 +1100710 spin_unlock(&inode->i_lock);
Wu Fengguangfcc5c222011-07-11 23:08:50 -0700711 redirty_tail(inode, wb);
Nick Piggin7ef0d732009-03-12 14:31:38 -0700712 continue;
713 }
Jan Karacc1676d2012-05-03 14:47:56 +0200714 if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) {
715 /*
716 * If this inode is locked for writeback and we are not
717 * doing writeback-for-data-integrity, move it to
718 * b_more_io so that writeback can proceed with the
719 * other inodes on s_io.
720 *
721 * We'll have another go at writing back this inode
722 * when we completed a full scan of b_io.
723 */
724 spin_unlock(&inode->i_lock);
725 requeue_io(inode, wb);
726 trace_writeback_sb_inodes_requeue(inode);
727 continue;
728 }
Jan Karaf0d07b72012-05-03 14:47:59 +0200729 spin_unlock(&wb->list_lock);
730
Jan Kara4f8ad652012-05-03 14:48:00 +0200731 /*
732 * We already requeued the inode if it had I_SYNC set and we
733 * are doing WB_SYNC_NONE writeback. So this catches only the
734 * WB_SYNC_ALL case.
735 */
Jan Kara169ebd92012-05-03 14:48:03 +0200736 if (inode->i_state & I_SYNC) {
737 /* Wait for I_SYNC. This function drops i_lock... */
738 inode_sleep_on_writeback(inode);
739 /* Inode may be gone, start again */
Jan Karaead188f2012-06-08 17:07:36 +0200740 spin_lock(&wb->list_lock);
Jan Kara169ebd92012-05-03 14:48:03 +0200741 continue;
742 }
Jan Kara4f8ad652012-05-03 14:48:00 +0200743 inode->i_state |= I_SYNC;
744 spin_unlock(&inode->i_lock);
Jan Kara169ebd92012-05-03 14:48:03 +0200745
Wu Fengguang1a12d8b2010-08-29 13:28:09 -0600746 write_chunk = writeback_chunk_size(wb->bdi, work);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600747 wbc.nr_to_write = write_chunk;
748 wbc.pages_skipped = 0;
Dave Chinner250df6e2011-03-22 22:23:36 +1100749
Jan Kara169ebd92012-05-03 14:48:03 +0200750 /*
751 * We use I_SYNC to pin the inode in memory. While it is set
752 * evict_inode() will wait so the inode cannot be freed.
753 */
Yan Hongcd8ed2a2012-10-08 16:33:45 -0700754 __writeback_single_inode(inode, &wbc);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600755
756 work->nr_pages -= write_chunk - wbc.nr_to_write;
757 wrote += write_chunk - wbc.nr_to_write;
Jan Kara4f8ad652012-05-03 14:48:00 +0200758 spin_lock(&wb->list_lock);
759 spin_lock(&inode->i_lock);
Theodore Ts'o0ae45f62015-02-02 00:37:00 -0500760 if (!(inode->i_state & I_DIRTY_ALL))
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600761 wrote++;
Jan Kara4f8ad652012-05-03 14:48:00 +0200762 requeue_inode(inode, wb, &wbc);
763 inode_sync_complete(inode);
Dave Chinner0f1b1fd2011-03-22 22:23:43 +1100764 spin_unlock(&inode->i_lock);
Jan Kara169ebd92012-05-03 14:48:03 +0200765 cond_resched_lock(&wb->list_lock);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600766 /*
767 * bail out to wb_writeback() often enough to check
768 * background threshold and other termination conditions.
769 */
770 if (wrote) {
771 if (time_is_before_jiffies(start_time + HZ / 10UL))
772 break;
773 if (work->nr_pages <= 0)
774 break;
Fengguang Wu8bc3be22008-02-04 22:29:36 -0800775 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 }
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600777 return wrote;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800778}
Nick Piggin38f21972009-01-06 14:40:25 -0800779
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600780static long __writeback_inodes_wb(struct bdi_writeback *wb,
781 struct wb_writeback_work *work)
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800782{
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600783 unsigned long start_time = jiffies;
784 long wrote = 0;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800785
786 while (!list_empty(&wb->b_io)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100787 struct inode *inode = wb_inode(wb->b_io.prev);
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800788 struct super_block *sb = inode->i_sb;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800789
Konstantin Khlebnikoveb6ef3d2015-02-19 20:19:35 +0300790 if (!trylock_super(sb)) {
Wu Fengguang0e9958162011-07-29 22:14:35 -0600791 /*
Konstantin Khlebnikoveb6ef3d2015-02-19 20:19:35 +0300792 * trylock_super() may fail consistently due to
Wu Fengguang0e9958162011-07-29 22:14:35 -0600793 * s_umount being grabbed by someone else. Don't use
794 * requeue_io() to avoid busy retrying the inode/sb.
795 */
796 redirty_tail(inode, wb);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200797 continue;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800798 }
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600799 wrote += writeback_sb_inodes(sb, wb, work);
Konstantin Khlebnikoveb6ef3d2015-02-19 20:19:35 +0300800 up_read(&sb->s_umount);
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800801
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600802 /* refer to the same tests at the end of writeback_sb_inodes */
803 if (wrote) {
804 if (time_is_before_jiffies(start_time + HZ / 10UL))
805 break;
806 if (work->nr_pages <= 0)
807 break;
808 }
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800809 }
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200810 /* Leave any unwritten inodes on b_io */
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600811 return wrote;
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200812}
813
Wanpeng Li7d9f0732013-09-11 14:22:40 -0700814static long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600815 enum wb_reason reason)
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200816{
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600817 struct wb_writeback_work work = {
818 .nr_pages = nr_pages,
819 .sync_mode = WB_SYNC_NONE,
820 .range_cyclic = 1,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600821 .reason = reason,
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600822 };
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200823
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600824 spin_lock(&wb->list_lock);
Wu Fengguang424b3512010-07-21 20:11:53 -0600825 if (list_empty(&wb->b_io))
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600826 queue_io(wb, &work);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600827 __writeback_inodes_wb(wb, &work);
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600828 spin_unlock(&wb->list_lock);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200829
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600830 return nr_pages - work.nr_pages;
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200831}
832
Wu Fengguangb00949a2010-11-18 14:38:33 -0600833static bool over_bground_thresh(struct backing_dev_info *bdi)
Jens Axboe03ba3782009-09-09 09:08:54 +0200834{
835 unsigned long background_thresh, dirty_thresh;
836
Wu Fengguang16c40422010-08-11 14:17:39 -0700837 global_dirty_limits(&background_thresh, &dirty_thresh);
Jens Axboe03ba3782009-09-09 09:08:54 +0200838
Wu Fengguangb00949a2010-11-18 14:38:33 -0600839 if (global_page_state(NR_FILE_DIRTY) +
840 global_page_state(NR_UNSTABLE_NFS) > background_thresh)
841 return true;
842
843 if (bdi_stat(bdi, BDI_RECLAIMABLE) >
844 bdi_dirty_limit(bdi, background_thresh))
845 return true;
846
847 return false;
Jens Axboe03ba3782009-09-09 09:08:54 +0200848}
849
850/*
Wu Fengguange98be2d2010-08-29 11:22:30 -0600851 * Called under wb->list_lock. If there are multiple wb per bdi,
852 * only the flusher working on the first wb should do it.
853 */
854static void wb_update_bandwidth(struct bdi_writeback *wb,
855 unsigned long start_time)
856{
Wu Fengguangaf6a3112011-10-03 20:46:17 -0600857 __bdi_update_bandwidth(wb->bdi, 0, 0, 0, 0, 0, start_time);
Wu Fengguange98be2d2010-08-29 11:22:30 -0600858}
859
860/*
Jens Axboe03ba3782009-09-09 09:08:54 +0200861 * Explicit flushing or periodic writeback of "old" data.
862 *
863 * Define "old": the first time one of an inode's pages is dirtied, we mark the
864 * dirtying-time in the inode's address_space. So this periodic writeback code
865 * just walks the superblock inode list, writing back any inodes which are
866 * older than a specific point in time.
867 *
868 * Try to run once per dirty_writeback_interval. But if a writeback event
869 * takes longer than a dirty_writeback_interval interval, then leave a
870 * one-second gap.
871 *
872 * older_than_this takes precedence over nr_to_write. So we'll only write back
873 * all dirty pages if they are all attached to "old" mappings.
874 */
Jens Axboec4a77a62009-09-16 15:18:25 +0200875static long wb_writeback(struct bdi_writeback *wb,
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200876 struct wb_writeback_work *work)
Jens Axboe03ba3782009-09-09 09:08:54 +0200877{
Wu Fengguange98be2d2010-08-29 11:22:30 -0600878 unsigned long wb_start = jiffies;
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600879 long nr_pages = work->nr_pages;
Jan Kara0dc83bd2014-02-21 11:19:04 +0100880 unsigned long oldest_jif;
Jan Karaa5989bd2009-09-16 19:22:48 +0200881 struct inode *inode;
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600882 long progress;
Jens Axboe03ba3782009-09-09 09:08:54 +0200883
Jan Kara0dc83bd2014-02-21 11:19:04 +0100884 oldest_jif = jiffies;
885 work->older_than_this = &oldest_jif;
Jens Axboe03ba3782009-09-09 09:08:54 +0200886
Wu Fengguange8dfc302011-04-21 12:06:32 -0600887 spin_lock(&wb->list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200888 for (;;) {
889 /*
Wu Fengguangd3ddec72009-09-23 20:33:40 +0800890 * Stop writeback when nr_pages has been consumed
Jens Axboe03ba3782009-09-09 09:08:54 +0200891 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200892 if (work->nr_pages <= 0)
Jens Axboe03ba3782009-09-09 09:08:54 +0200893 break;
894
895 /*
Jan Karaaa373cf2011-01-13 15:45:47 -0800896 * Background writeout and kupdate-style writeback may
897 * run forever. Stop them if there is other work to do
898 * so that e.g. sync can proceed. They'll be restarted
899 * after the other works are all done.
900 */
901 if ((work->for_background || work->for_kupdate) &&
902 !list_empty(&wb->bdi->work_list))
903 break;
904
905 /*
Wu Fengguangd3ddec72009-09-23 20:33:40 +0800906 * For background writeout, stop when we are below the
907 * background dirty threshold
Jens Axboe03ba3782009-09-09 09:08:54 +0200908 */
Wu Fengguangb00949a2010-11-18 14:38:33 -0600909 if (work->for_background && !over_bground_thresh(wb->bdi))
Jens Axboe03ba3782009-09-09 09:08:54 +0200910 break;
911
Jan Kara1bc36b62011-10-19 11:44:41 +0200912 /*
913 * Kupdate and background works are special and we want to
914 * include all inodes that need writing. Livelock avoidance is
915 * handled by these works yielding to any other work so we are
916 * safe.
917 */
Wu Fengguangba9aa832010-07-21 20:32:30 -0600918 if (work->for_kupdate) {
Jan Kara0dc83bd2014-02-21 11:19:04 +0100919 oldest_jif = jiffies -
Wu Fengguangba9aa832010-07-21 20:32:30 -0600920 msecs_to_jiffies(dirty_expire_interval * 10);
Jan Kara1bc36b62011-10-19 11:44:41 +0200921 } else if (work->for_background)
Jan Kara0dc83bd2014-02-21 11:19:04 +0100922 oldest_jif = jiffies;
Dave Chinner028c2dd2010-07-07 13:24:07 +1000923
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600924 trace_writeback_start(wb->bdi, work);
Wu Fengguange8dfc302011-04-21 12:06:32 -0600925 if (list_empty(&wb->b_io))
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600926 queue_io(wb, work);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200927 if (work->sb)
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600928 progress = writeback_sb_inodes(work->sb, wb, work);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200929 else
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600930 progress = __writeback_inodes_wb(wb, work);
931 trace_writeback_written(wb->bdi, work);
Dave Chinner028c2dd2010-07-07 13:24:07 +1000932
Wu Fengguange98be2d2010-08-29 11:22:30 -0600933 wb_update_bandwidth(wb, wb_start);
Jens Axboe03ba3782009-09-09 09:08:54 +0200934
935 /*
Jens Axboe71fd05a2009-09-23 19:32:26 +0200936 * Did we write something? Try for more
Wu Fengguange6fb6da2010-07-22 10:23:44 -0600937 *
938 * Dirty inodes are moved to b_io for writeback in batches.
939 * The completion of the current batch does not necessarily
940 * mean the overall work is done. So we keep looping as long
941 * as made some progress on cleaning pages or inodes.
Jens Axboe71fd05a2009-09-23 19:32:26 +0200942 */
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600943 if (progress)
Jens Axboe71fd05a2009-09-23 19:32:26 +0200944 continue;
945 /*
Wu Fengguange6fb6da2010-07-22 10:23:44 -0600946 * No more inodes for IO, bail
Jens Axboe71fd05a2009-09-23 19:32:26 +0200947 */
Wu Fengguangb7a24412010-07-21 22:19:51 -0600948 if (list_empty(&wb->b_more_io))
Jens Axboe03ba3782009-09-09 09:08:54 +0200949 break;
950 /*
Jens Axboe8010c3b2009-09-15 20:04:57 +0200951 * Nothing written. Wait for some inode to
952 * become available for writeback. Otherwise
953 * we'll just busyloop.
Jens Axboe03ba3782009-09-09 09:08:54 +0200954 */
Jens Axboe03ba3782009-09-09 09:08:54 +0200955 if (!list_empty(&wb->b_more_io)) {
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600956 trace_writeback_wait(wb->bdi, work);
Jens Axboe03ba3782009-09-09 09:08:54 +0200957 inode = wb_inode(wb->b_more_io.prev);
Dave Chinner250df6e2011-03-22 22:23:36 +1100958 spin_lock(&inode->i_lock);
Jan Karaf0d07b72012-05-03 14:47:59 +0200959 spin_unlock(&wb->list_lock);
Jan Kara169ebd92012-05-03 14:48:03 +0200960 /* This function drops i_lock... */
961 inode_sleep_on_writeback(inode);
Jan Karaf0d07b72012-05-03 14:47:59 +0200962 spin_lock(&wb->list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200963 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200964 }
Wu Fengguange8dfc302011-04-21 12:06:32 -0600965 spin_unlock(&wb->list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200966
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600967 return nr_pages - work->nr_pages;
Jens Axboe03ba3782009-09-09 09:08:54 +0200968}
969
970/*
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200971 * Return the next wb_writeback_work struct that hasn't been processed yet.
Jens Axboe03ba3782009-09-09 09:08:54 +0200972 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200973static struct wb_writeback_work *
Minchan Kim08852b62010-08-03 12:51:16 +0200974get_next_work_item(struct backing_dev_info *bdi)
Jens Axboe03ba3782009-09-09 09:08:54 +0200975{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200976 struct wb_writeback_work *work = NULL;
Jens Axboe03ba3782009-09-09 09:08:54 +0200977
Artem Bityutskiy64677162010-07-25 14:29:22 +0300978 spin_lock_bh(&bdi->wb_lock);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200979 if (!list_empty(&bdi->work_list)) {
980 work = list_entry(bdi->work_list.next,
981 struct wb_writeback_work, list);
982 list_del_init(&work->list);
Jens Axboe03ba3782009-09-09 09:08:54 +0200983 }
Artem Bityutskiy64677162010-07-25 14:29:22 +0300984 spin_unlock_bh(&bdi->wb_lock);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200985 return work;
Jens Axboe03ba3782009-09-09 09:08:54 +0200986}
987
Linus Torvaldscdf01dd2010-10-30 08:55:52 -0700988/*
989 * Add in the number of potentially dirty inodes, because each inode
990 * write can dirty pagecache in the underlying blockdev.
991 */
992static unsigned long get_nr_dirty_pages(void)
993{
994 return global_page_state(NR_FILE_DIRTY) +
995 global_page_state(NR_UNSTABLE_NFS) +
996 get_nr_dirty_inodes();
997}
998
Jan Kara65850272011-01-13 15:45:44 -0800999static long wb_check_background_flush(struct bdi_writeback *wb)
1000{
Wu Fengguangb00949a2010-11-18 14:38:33 -06001001 if (over_bground_thresh(wb->bdi)) {
Jan Kara65850272011-01-13 15:45:44 -08001002
1003 struct wb_writeback_work work = {
1004 .nr_pages = LONG_MAX,
1005 .sync_mode = WB_SYNC_NONE,
1006 .for_background = 1,
1007 .range_cyclic = 1,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001008 .reason = WB_REASON_BACKGROUND,
Jan Kara65850272011-01-13 15:45:44 -08001009 };
1010
1011 return wb_writeback(wb, &work);
1012 }
1013
1014 return 0;
1015}
1016
Jens Axboe03ba3782009-09-09 09:08:54 +02001017static long wb_check_old_data_flush(struct bdi_writeback *wb)
1018{
1019 unsigned long expired;
1020 long nr_pages;
1021
Jens Axboe69b62d02010-05-17 12:51:03 +02001022 /*
1023 * When set to zero, disable periodic writeback
1024 */
1025 if (!dirty_writeback_interval)
1026 return 0;
1027
Jens Axboe03ba3782009-09-09 09:08:54 +02001028 expired = wb->last_old_flush +
1029 msecs_to_jiffies(dirty_writeback_interval * 10);
1030 if (time_before(jiffies, expired))
1031 return 0;
1032
1033 wb->last_old_flush = jiffies;
Linus Torvaldscdf01dd2010-10-30 08:55:52 -07001034 nr_pages = get_nr_dirty_pages();
Jens Axboe03ba3782009-09-09 09:08:54 +02001035
Jens Axboec4a77a62009-09-16 15:18:25 +02001036 if (nr_pages) {
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001037 struct wb_writeback_work work = {
Jens Axboec4a77a62009-09-16 15:18:25 +02001038 .nr_pages = nr_pages,
1039 .sync_mode = WB_SYNC_NONE,
1040 .for_kupdate = 1,
1041 .range_cyclic = 1,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001042 .reason = WB_REASON_PERIODIC,
Jens Axboec4a77a62009-09-16 15:18:25 +02001043 };
1044
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001045 return wb_writeback(wb, &work);
Jens Axboec4a77a62009-09-16 15:18:25 +02001046 }
Jens Axboe03ba3782009-09-09 09:08:54 +02001047
1048 return 0;
1049}
1050
1051/*
1052 * Retrieve work items and do the writeback they describe
1053 */
Wanpeng Li25d130b2013-07-08 16:00:14 -07001054static long wb_do_writeback(struct bdi_writeback *wb)
Jens Axboe03ba3782009-09-09 09:08:54 +02001055{
1056 struct backing_dev_info *bdi = wb->bdi;
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001057 struct wb_writeback_work *work;
Jens Axboec4a77a62009-09-16 15:18:25 +02001058 long wrote = 0;
Jens Axboe03ba3782009-09-09 09:08:54 +02001059
Jan Kara81d73a32010-08-11 14:17:44 -07001060 set_bit(BDI_writeback_running, &wb->bdi->state);
Minchan Kim08852b62010-08-03 12:51:16 +02001061 while ((work = get_next_work_item(bdi)) != NULL) {
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001062
Dave Chinner455b2862010-07-07 13:24:06 +10001063 trace_writeback_exec(bdi, work);
1064
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001065 wrote += wb_writeback(wb, work);
Jens Axboe03ba3782009-09-09 09:08:54 +02001066
1067 /*
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001068 * Notify the caller of completion if this is a synchronous
1069 * work item, otherwise just free it.
Jens Axboe03ba3782009-09-09 09:08:54 +02001070 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001071 if (work->done)
1072 complete(work->done);
1073 else
1074 kfree(work);
Jens Axboe03ba3782009-09-09 09:08:54 +02001075 }
1076
1077 /*
1078 * Check for periodic writeback, kupdated() style
1079 */
1080 wrote += wb_check_old_data_flush(wb);
Jan Kara65850272011-01-13 15:45:44 -08001081 wrote += wb_check_background_flush(wb);
Jan Kara81d73a32010-08-11 14:17:44 -07001082 clear_bit(BDI_writeback_running, &wb->bdi->state);
Jens Axboe03ba3782009-09-09 09:08:54 +02001083
1084 return wrote;
1085}
1086
1087/*
1088 * Handle writeback of dirty data for the device backed by this bdi. Also
Tejun Heo839a8e82013-04-01 19:08:06 -07001089 * reschedules periodically and does kupdated style flushing.
Jens Axboe03ba3782009-09-09 09:08:54 +02001090 */
Tejun Heo839a8e82013-04-01 19:08:06 -07001091void bdi_writeback_workfn(struct work_struct *work)
Jens Axboe03ba3782009-09-09 09:08:54 +02001092{
Tejun Heo839a8e82013-04-01 19:08:06 -07001093 struct bdi_writeback *wb = container_of(to_delayed_work(work),
1094 struct bdi_writeback, dwork);
Christoph Hellwig08243902010-06-19 23:08:22 +02001095 struct backing_dev_info *bdi = wb->bdi;
Jens Axboe03ba3782009-09-09 09:08:54 +02001096 long pages_written;
1097
Tejun Heoef3b1012013-04-30 15:27:24 -07001098 set_worker_desc("flush-%s", dev_name(bdi->dev));
Peter Zijlstra766f9162010-10-26 14:22:45 -07001099 current->flags |= PF_SWAPWRITE;
Christoph Hellwig08243902010-06-19 23:08:22 +02001100
Tejun Heo839a8e82013-04-01 19:08:06 -07001101 if (likely(!current_is_workqueue_rescuer() ||
Jan Kara5acda9d2014-04-03 14:46:23 -07001102 !test_bit(BDI_registered, &bdi->state))) {
Artem Bityutskiy64677162010-07-25 14:29:22 +03001103 /*
Tejun Heo839a8e82013-04-01 19:08:06 -07001104 * The normal path. Keep writing back @bdi until its
1105 * work_list is empty. Note that this path is also taken
1106 * if @bdi is shutting down even when we're running off the
1107 * rescuer as work_list needs to be drained.
Artem Bityutskiy64677162010-07-25 14:29:22 +03001108 */
Tejun Heo839a8e82013-04-01 19:08:06 -07001109 do {
Wanpeng Li25d130b2013-07-08 16:00:14 -07001110 pages_written = wb_do_writeback(wb);
Tejun Heo839a8e82013-04-01 19:08:06 -07001111 trace_writeback_pages_written(pages_written);
1112 } while (!list_empty(&bdi->work_list));
1113 } else {
1114 /*
1115 * bdi_wq can't get enough workers and we're running off
1116 * the emergency worker. Don't hog it. Hopefully, 1024 is
1117 * enough for efficient IO.
1118 */
1119 pages_written = writeback_inodes_wb(&bdi->wb, 1024,
1120 WB_REASON_FORKER_THREAD);
Dave Chinner455b2862010-07-07 13:24:06 +10001121 trace_writeback_pages_written(pages_written);
Jens Axboe03ba3782009-09-09 09:08:54 +02001122 }
1123
Derek Basehore6ca738d2014-04-03 14:46:22 -07001124 if (!list_empty(&bdi->work_list))
1125 mod_delayed_work(bdi_wq, &wb->dwork, 0);
1126 else if (wb_has_dirty_io(wb) && dirty_writeback_interval)
1127 bdi_wakeup_thread_delayed(bdi);
Dave Chinner455b2862010-07-07 13:24:06 +10001128
Tejun Heo839a8e82013-04-01 19:08:06 -07001129 current->flags &= ~PF_SWAPWRITE;
Jens Axboe03ba3782009-09-09 09:08:54 +02001130}
1131
1132/*
Jens Axboe03ba3782009-09-09 09:08:54 +02001133 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
1134 * the whole world.
1135 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001136void wakeup_flusher_threads(long nr_pages, enum wb_reason reason)
Jens Axboe03ba3782009-09-09 09:08:54 +02001137{
Christoph Hellwigb8c2f342010-06-08 18:15:07 +02001138 struct backing_dev_info *bdi;
Christoph Hellwigb8c2f342010-06-08 18:15:07 +02001139
Jan Kara47df3dd2013-09-11 14:22:22 -07001140 if (!nr_pages)
1141 nr_pages = get_nr_dirty_pages();
Christoph Hellwigb8c2f342010-06-08 18:15:07 +02001142
1143 rcu_read_lock();
1144 list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) {
1145 if (!bdi_has_dirty_io(bdi))
1146 continue;
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001147 __bdi_start_writeback(bdi, nr_pages, false, reason);
Christoph Hellwigb8c2f342010-06-08 18:15:07 +02001148 }
1149 rcu_read_unlock();
Jens Axboe03ba3782009-09-09 09:08:54 +02001150}
1151
Theodore Ts'oa2f48702015-03-17 12:23:19 -04001152/*
1153 * Wake up bdi's periodically to make sure dirtytime inodes gets
1154 * written back periodically. We deliberately do *not* check the
1155 * b_dirtytime list in wb_has_dirty_io(), since this would cause the
1156 * kernel to be constantly waking up once there are any dirtytime
1157 * inodes on the system. So instead we define a separate delayed work
1158 * function which gets called much more rarely. (By default, only
1159 * once every 12 hours.)
1160 *
1161 * If there is any other write activity going on in the file system,
1162 * this function won't be necessary. But if the only thing that has
1163 * happened on the file system is a dirtytime inode caused by an atime
1164 * update, we need this infrastructure below to make sure that inode
1165 * eventually gets pushed out to disk.
1166 */
1167static void wakeup_dirtytime_writeback(struct work_struct *w);
1168static DECLARE_DELAYED_WORK(dirtytime_work, wakeup_dirtytime_writeback);
1169
1170static void wakeup_dirtytime_writeback(struct work_struct *w)
1171{
1172 struct backing_dev_info *bdi;
1173
1174 rcu_read_lock();
1175 list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) {
1176 if (list_empty(&bdi->wb.b_dirty_time))
1177 continue;
1178 bdi_wakeup_thread(bdi);
1179 }
1180 rcu_read_unlock();
1181 schedule_delayed_work(&dirtytime_work, dirtytime_expire_interval * HZ);
1182}
1183
1184static int __init start_dirtytime_writeback(void)
1185{
1186 schedule_delayed_work(&dirtytime_work, dirtytime_expire_interval * HZ);
1187 return 0;
1188}
1189__initcall(start_dirtytime_writeback);
1190
Theodore Ts'o1efff912015-03-17 12:23:32 -04001191int dirtytime_interval_handler(struct ctl_table *table, int write,
1192 void __user *buffer, size_t *lenp, loff_t *ppos)
1193{
1194 int ret;
1195
1196 ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
1197 if (ret == 0 && write)
1198 mod_delayed_work(system_wq, &dirtytime_work, 0);
1199 return ret;
1200}
1201
Jens Axboe03ba3782009-09-09 09:08:54 +02001202static noinline void block_dump___mark_inode_dirty(struct inode *inode)
1203{
1204 if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) {
1205 struct dentry *dentry;
1206 const char *name = "?";
1207
1208 dentry = d_find_alias(inode);
1209 if (dentry) {
1210 spin_lock(&dentry->d_lock);
1211 name = (const char *) dentry->d_name.name;
1212 }
1213 printk(KERN_DEBUG
1214 "%s(%d): dirtied inode %lu (%s) on %s\n",
1215 current->comm, task_pid_nr(current), inode->i_ino,
1216 name, inode->i_sb->s_id);
1217 if (dentry) {
1218 spin_unlock(&dentry->d_lock);
1219 dput(dentry);
1220 }
1221 }
1222}
1223
1224/**
1225 * __mark_inode_dirty - internal function
1226 * @inode: inode to mark
1227 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
1228 * Mark an inode as dirty. Callers should use mark_inode_dirty or
1229 * mark_inode_dirty_sync.
1230 *
1231 * Put the inode on the super block's dirty list.
1232 *
1233 * CAREFUL! We mark it dirty unconditionally, but move it onto the
1234 * dirty list only if it is hashed or if it refers to a blockdev.
1235 * If it was not hashed, it will never be added to the dirty list
1236 * even if it is later hashed, as it will have been marked dirty already.
1237 *
1238 * In short, make sure you hash any inodes _before_ you start marking
1239 * them dirty.
1240 *
Jens Axboe03ba3782009-09-09 09:08:54 +02001241 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
1242 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
1243 * the kernel-internal blockdev inode represents the dirtying time of the
1244 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
1245 * page->mapping->host, so the page-dirtying time is recorded in the internal
1246 * blockdev inode.
1247 */
Theodore Ts'o0ae45f62015-02-02 00:37:00 -05001248#define I_DIRTY_INODE (I_DIRTY_SYNC | I_DIRTY_DATASYNC)
Jens Axboe03ba3782009-09-09 09:08:54 +02001249void __mark_inode_dirty(struct inode *inode, int flags)
1250{
1251 struct super_block *sb = inode->i_sb;
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001252 struct backing_dev_info *bdi = NULL;
Theodore Ts'o0ae45f62015-02-02 00:37:00 -05001253 int dirtytime;
1254
1255 trace_writeback_mark_inode_dirty(inode, flags);
Jens Axboe03ba3782009-09-09 09:08:54 +02001256
1257 /*
1258 * Don't do this for I_DIRTY_PAGES - that doesn't actually
1259 * dirty the inode itself
1260 */
Theodore Ts'o0ae45f62015-02-02 00:37:00 -05001261 if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC | I_DIRTY_TIME)) {
Tejun Heo9fb0a7d2013-01-11 13:06:37 -08001262 trace_writeback_dirty_inode_start(inode, flags);
1263
Jens Axboe03ba3782009-09-09 09:08:54 +02001264 if (sb->s_op->dirty_inode)
Christoph Hellwigaa385722011-05-27 06:53:02 -04001265 sb->s_op->dirty_inode(inode, flags);
Tejun Heo9fb0a7d2013-01-11 13:06:37 -08001266
1267 trace_writeback_dirty_inode(inode, flags);
Jens Axboe03ba3782009-09-09 09:08:54 +02001268 }
Theodore Ts'o0ae45f62015-02-02 00:37:00 -05001269 if (flags & I_DIRTY_INODE)
1270 flags &= ~I_DIRTY_TIME;
1271 dirtytime = flags & I_DIRTY_TIME;
Jens Axboe03ba3782009-09-09 09:08:54 +02001272
1273 /*
Tejun Heo9c6ac782014-10-24 15:38:21 -04001274 * Paired with smp_mb() in __writeback_single_inode() for the
1275 * following lockless i_state test. See there for details.
Jens Axboe03ba3782009-09-09 09:08:54 +02001276 */
1277 smp_mb();
1278
Theodore Ts'o0ae45f62015-02-02 00:37:00 -05001279 if (((inode->i_state & flags) == flags) ||
1280 (dirtytime && (inode->i_state & I_DIRTY_INODE)))
Jens Axboe03ba3782009-09-09 09:08:54 +02001281 return;
1282
1283 if (unlikely(block_dump))
1284 block_dump___mark_inode_dirty(inode);
1285
Dave Chinner250df6e2011-03-22 22:23:36 +11001286 spin_lock(&inode->i_lock);
Theodore Ts'o0ae45f62015-02-02 00:37:00 -05001287 if (dirtytime && (inode->i_state & I_DIRTY_INODE))
1288 goto out_unlock_inode;
Jens Axboe03ba3782009-09-09 09:08:54 +02001289 if ((inode->i_state & flags) != flags) {
1290 const int was_dirty = inode->i_state & I_DIRTY;
1291
Theodore Ts'o0ae45f62015-02-02 00:37:00 -05001292 if (flags & I_DIRTY_INODE)
1293 inode->i_state &= ~I_DIRTY_TIME;
Jens Axboe03ba3782009-09-09 09:08:54 +02001294 inode->i_state |= flags;
1295
1296 /*
1297 * If the inode is being synced, just update its dirty state.
1298 * The unlocker will place the inode on the appropriate
1299 * superblock list, based upon its state.
1300 */
1301 if (inode->i_state & I_SYNC)
Dave Chinner250df6e2011-03-22 22:23:36 +11001302 goto out_unlock_inode;
Jens Axboe03ba3782009-09-09 09:08:54 +02001303
1304 /*
1305 * Only add valid (hashed) inodes to the superblock's
1306 * dirty list. Add blockdev inodes as well.
1307 */
1308 if (!S_ISBLK(inode->i_mode)) {
Al Viro1d3382cb2010-10-23 15:19:20 -04001309 if (inode_unhashed(inode))
Dave Chinner250df6e2011-03-22 22:23:36 +11001310 goto out_unlock_inode;
Jens Axboe03ba3782009-09-09 09:08:54 +02001311 }
Al Viroa4ffdde2010-06-02 17:38:30 -04001312 if (inode->i_state & I_FREEING)
Dave Chinner250df6e2011-03-22 22:23:36 +11001313 goto out_unlock_inode;
Jens Axboe03ba3782009-09-09 09:08:54 +02001314
1315 /*
1316 * If the inode was already on b_dirty/b_io/b_more_io, don't
1317 * reposition it (that would break b_dirty time-ordering).
1318 */
1319 if (!was_dirty) {
Dave Chinnera66979a2011-03-22 22:23:41 +11001320 bool wakeup_bdi = false;
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001321 bdi = inode_to_bdi(inode);
Jens Axboe500b0672009-09-09 09:10:25 +02001322
Junxiao Bi146d7002013-09-11 14:23:04 -07001323 spin_unlock(&inode->i_lock);
1324 spin_lock(&bdi->wb.list_lock);
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001325 if (bdi_cap_writeback_dirty(bdi)) {
1326 WARN(!test_bit(BDI_registered, &bdi->state),
1327 "bdi-%s not registered\n", bdi->name);
1328
1329 /*
1330 * If this is the first dirty inode for this
1331 * bdi, we have to wake-up the corresponding
1332 * bdi thread to make sure background
1333 * write-back happens later.
1334 */
1335 if (!wb_has_dirty_io(&bdi->wb))
1336 wakeup_bdi = true;
Jens Axboe500b0672009-09-09 09:10:25 +02001337 }
Jens Axboe03ba3782009-09-09 09:08:54 +02001338
1339 inode->dirtied_when = jiffies;
Theodore Ts'oa2f48702015-03-17 12:23:19 -04001340 if (dirtytime)
1341 inode->dirtied_time_when = jiffies;
1342 if (inode->i_state & (I_DIRTY_INODE | I_DIRTY_PAGES))
1343 list_move(&inode->i_wb_list, &bdi->wb.b_dirty);
1344 else
1345 list_move(&inode->i_wb_list,
1346 &bdi->wb.b_dirty_time);
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001347 spin_unlock(&bdi->wb.list_lock);
Theodore Ts'o0ae45f62015-02-02 00:37:00 -05001348 trace_writeback_dirty_inode_enqueue(inode);
Dave Chinnera66979a2011-03-22 22:23:41 +11001349
1350 if (wakeup_bdi)
1351 bdi_wakeup_thread_delayed(bdi);
1352 return;
Jens Axboe03ba3782009-09-09 09:08:54 +02001353 }
1354 }
Dave Chinner250df6e2011-03-22 22:23:36 +11001355out_unlock_inode:
1356 spin_unlock(&inode->i_lock);
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001357
Jens Axboe03ba3782009-09-09 09:08:54 +02001358}
1359EXPORT_SYMBOL(__mark_inode_dirty);
1360
Jens Axboeb6e51312009-09-16 15:13:54 +02001361static void wait_sb_inodes(struct super_block *sb)
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001362{
Jens Axboe03ba3782009-09-09 09:08:54 +02001363 struct inode *inode, *old_inode = NULL;
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001364
Jens Axboe03ba3782009-09-09 09:08:54 +02001365 /*
1366 * We need to be protected against the filesystem going from
1367 * r/o to r/w or vice versa.
1368 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001369 WARN_ON(!rwsem_is_locked(&sb->s_umount));
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001370
Dave Chinner55fa6092011-03-22 22:23:40 +11001371 spin_lock(&inode_sb_list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001372
1373 /*
1374 * Data integrity sync. Must wait for all pages under writeback,
1375 * because there may have been pages dirtied before our sync
1376 * call, but which had writeout started before we write it out.
1377 * In which case, the inode may not be on the dirty list, but
1378 * we still have to wait for that writeout.
1379 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001380 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
Dave Chinner250df6e2011-03-22 22:23:36 +11001381 struct address_space *mapping = inode->i_mapping;
Jens Axboe03ba3782009-09-09 09:08:54 +02001382
Dave Chinner250df6e2011-03-22 22:23:36 +11001383 spin_lock(&inode->i_lock);
1384 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
1385 (mapping->nrpages == 0)) {
1386 spin_unlock(&inode->i_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001387 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +11001388 }
Jens Axboe03ba3782009-09-09 09:08:54 +02001389 __iget(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +11001390 spin_unlock(&inode->i_lock);
Dave Chinner55fa6092011-03-22 22:23:40 +11001391 spin_unlock(&inode_sb_list_lock);
1392
Jens Axboe03ba3782009-09-09 09:08:54 +02001393 /*
Dave Chinner55fa6092011-03-22 22:23:40 +11001394 * We hold a reference to 'inode' so it couldn't have been
1395 * removed from s_inodes list while we dropped the
1396 * inode_sb_list_lock. We cannot iput the inode now as we can
1397 * be holding the last reference and we cannot iput it under
1398 * inode_sb_list_lock. So we keep the reference and iput it
1399 * later.
Jens Axboe03ba3782009-09-09 09:08:54 +02001400 */
1401 iput(old_inode);
1402 old_inode = inode;
1403
1404 filemap_fdatawait(mapping);
1405
1406 cond_resched();
Nick Piggin38f21972009-01-06 14:40:25 -08001407
Dave Chinner55fa6092011-03-22 22:23:40 +11001408 spin_lock(&inode_sb_list_lock);
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001409 }
Dave Chinner55fa6092011-03-22 22:23:40 +11001410 spin_unlock(&inode_sb_list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001411 iput(old_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412}
1413
Jens Axboed8a85592009-09-02 12:34:32 +02001414/**
Chris Mason3259f8b2010-10-29 11:16:17 -04001415 * writeback_inodes_sb_nr - writeback dirty inodes from given super_block
Jens Axboed8a85592009-09-02 12:34:32 +02001416 * @sb: the superblock
Chris Mason3259f8b2010-10-29 11:16:17 -04001417 * @nr: the number of pages to write
Marcos Paulo de Souza786228a2011-11-23 20:56:45 +08001418 * @reason: reason why some writeback work initiated
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419 *
Jens Axboed8a85592009-09-02 12:34:32 +02001420 * Start writeback on some inodes on this super_block. No guarantees are made
1421 * on how many (if any) will be written, and this function does not wait
Chris Mason3259f8b2010-10-29 11:16:17 -04001422 * for IO completion of submitted IO.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001424void writeback_inodes_sb_nr(struct super_block *sb,
1425 unsigned long nr,
1426 enum wb_reason reason)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001428 DECLARE_COMPLETION_ONSTACK(done);
1429 struct wb_writeback_work work = {
Wu Fengguang6e6938b2010-06-06 10:38:15 -06001430 .sb = sb,
1431 .sync_mode = WB_SYNC_NONE,
1432 .tagged_writepages = 1,
1433 .done = &done,
1434 .nr_pages = nr,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001435 .reason = reason,
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001436 };
Jens Axboe0e3c9a22010-06-01 11:08:43 +02001437
Jan Kara6eedc702012-07-03 16:45:27 +02001438 if (sb->s_bdi == &noop_backing_dev_info)
1439 return;
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001440 WARN_ON(!rwsem_is_locked(&sb->s_umount));
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001441 bdi_queue_work(sb->s_bdi, &work);
1442 wait_for_completion(&done);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443}
Chris Mason3259f8b2010-10-29 11:16:17 -04001444EXPORT_SYMBOL(writeback_inodes_sb_nr);
1445
1446/**
1447 * writeback_inodes_sb - writeback dirty inodes from given super_block
1448 * @sb: the superblock
Marcos Paulo de Souza786228a2011-11-23 20:56:45 +08001449 * @reason: reason why some writeback work was initiated
Chris Mason3259f8b2010-10-29 11:16:17 -04001450 *
1451 * Start writeback on some inodes on this super_block. No guarantees are made
1452 * on how many (if any) will be written, and this function does not wait
1453 * for IO completion of submitted IO.
1454 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001455void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason)
Chris Mason3259f8b2010-10-29 11:16:17 -04001456{
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001457 return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason);
Chris Mason3259f8b2010-10-29 11:16:17 -04001458}
Jens Axboed8a85592009-09-02 12:34:32 +02001459EXPORT_SYMBOL(writeback_inodes_sb);
1460
1461/**
Miao Xie10ee27a2013-01-10 13:47:57 +08001462 * try_to_writeback_inodes_sb_nr - try to start writeback if none underway
Chris Mason3259f8b2010-10-29 11:16:17 -04001463 * @sb: the superblock
1464 * @nr: the number of pages to write
Miao Xie10ee27a2013-01-10 13:47:57 +08001465 * @reason: the reason of writeback
Chris Mason3259f8b2010-10-29 11:16:17 -04001466 *
Miao Xie10ee27a2013-01-10 13:47:57 +08001467 * Invoke writeback_inodes_sb_nr if no writeback is currently underway.
Chris Mason3259f8b2010-10-29 11:16:17 -04001468 * Returns 1 if writeback was started, 0 if not.
1469 */
Miao Xie10ee27a2013-01-10 13:47:57 +08001470int try_to_writeback_inodes_sb_nr(struct super_block *sb,
1471 unsigned long nr,
1472 enum wb_reason reason)
Chris Mason3259f8b2010-10-29 11:16:17 -04001473{
Miao Xie10ee27a2013-01-10 13:47:57 +08001474 if (writeback_in_progress(sb->s_bdi))
Chris Mason3259f8b2010-10-29 11:16:17 -04001475 return 1;
Miao Xie10ee27a2013-01-10 13:47:57 +08001476
1477 if (!down_read_trylock(&sb->s_umount))
Chris Mason3259f8b2010-10-29 11:16:17 -04001478 return 0;
Miao Xie10ee27a2013-01-10 13:47:57 +08001479
1480 writeback_inodes_sb_nr(sb, nr, reason);
1481 up_read(&sb->s_umount);
1482 return 1;
Chris Mason3259f8b2010-10-29 11:16:17 -04001483}
Miao Xie10ee27a2013-01-10 13:47:57 +08001484EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr);
1485
1486/**
1487 * try_to_writeback_inodes_sb - try to start writeback if none underway
1488 * @sb: the superblock
1489 * @reason: reason why some writeback work was initiated
1490 *
1491 * Implement by try_to_writeback_inodes_sb_nr()
1492 * Returns 1 if writeback was started, 0 if not.
1493 */
1494int try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason)
1495{
1496 return try_to_writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason);
1497}
1498EXPORT_SYMBOL(try_to_writeback_inodes_sb);
Chris Mason3259f8b2010-10-29 11:16:17 -04001499
1500/**
Jens Axboed8a85592009-09-02 12:34:32 +02001501 * sync_inodes_sb - sync sb inode pages
Jan Kara0dc83bd2014-02-21 11:19:04 +01001502 * @sb: the superblock
Jens Axboed8a85592009-09-02 12:34:32 +02001503 *
1504 * This function writes and waits on any dirty inode belonging to this
Jan Kara0dc83bd2014-02-21 11:19:04 +01001505 * super_block.
Jens Axboed8a85592009-09-02 12:34:32 +02001506 */
Jan Kara0dc83bd2014-02-21 11:19:04 +01001507void sync_inodes_sb(struct super_block *sb)
Jens Axboed8a85592009-09-02 12:34:32 +02001508{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001509 DECLARE_COMPLETION_ONSTACK(done);
1510 struct wb_writeback_work work = {
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001511 .sb = sb,
1512 .sync_mode = WB_SYNC_ALL,
1513 .nr_pages = LONG_MAX,
1514 .range_cyclic = 0,
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001515 .done = &done,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001516 .reason = WB_REASON_SYNC,
Dave Chinner7747bd42013-07-02 22:38:35 +10001517 .for_sync = 1,
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001518 };
1519
Jan Kara6eedc702012-07-03 16:45:27 +02001520 /* Nothing to do? */
1521 if (sb->s_bdi == &noop_backing_dev_info)
1522 return;
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001523 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1524
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001525 bdi_queue_work(sb->s_bdi, &work);
1526 wait_for_completion(&done);
1527
Jens Axboeb6e51312009-09-16 15:13:54 +02001528 wait_sb_inodes(sb);
Jens Axboed8a85592009-09-02 12:34:32 +02001529}
1530EXPORT_SYMBOL(sync_inodes_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532/**
Andrea Arcangeli7f04c262005-10-30 15:03:05 -08001533 * write_inode_now - write an inode to disk
1534 * @inode: inode to write to disk
1535 * @sync: whether the write should be synchronous or not
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536 *
Andrea Arcangeli7f04c262005-10-30 15:03:05 -08001537 * This function commits an inode to disk immediately if it is dirty. This is
1538 * primarily needed by knfsd.
1539 *
1540 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542int write_inode_now(struct inode *inode, int sync)
1543{
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001544 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545 struct writeback_control wbc = {
1546 .nr_to_write = LONG_MAX,
Mike Galbraith18914b12008-02-08 04:20:23 -08001547 .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -07001548 .range_start = 0,
1549 .range_end = LLONG_MAX,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550 };
1551
1552 if (!mapping_cap_writeback_dirty(inode->i_mapping))
Andrew Morton49364ce2005-11-07 00:59:15 -08001553 wbc.nr_to_write = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554
1555 might_sleep();
Jan Kara4f8ad652012-05-03 14:48:00 +02001556 return writeback_single_inode(inode, wb, &wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557}
1558EXPORT_SYMBOL(write_inode_now);
1559
1560/**
1561 * sync_inode - write an inode and its pages to disk.
1562 * @inode: the inode to sync
1563 * @wbc: controls the writeback mode
1564 *
1565 * sync_inode() will write an inode and its pages to disk. It will also
1566 * correctly update the inode on its superblock's dirty inode lists and will
1567 * update inode->i_state.
1568 *
1569 * The caller must have a ref on the inode.
1570 */
1571int sync_inode(struct inode *inode, struct writeback_control *wbc)
1572{
Jan Kara4f8ad652012-05-03 14:48:00 +02001573 return writeback_single_inode(inode, &inode_to_bdi(inode)->wb, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574}
1575EXPORT_SYMBOL(sync_inode);
Christoph Hellwigc37650162010-10-06 10:48:20 +02001576
1577/**
Andrew Mortonc691b9d2011-01-13 15:45:48 -08001578 * sync_inode_metadata - write an inode to disk
Christoph Hellwigc37650162010-10-06 10:48:20 +02001579 * @inode: the inode to sync
1580 * @wait: wait for I/O to complete.
1581 *
Andrew Mortonc691b9d2011-01-13 15:45:48 -08001582 * Write an inode to disk and adjust its dirty state after completion.
Christoph Hellwigc37650162010-10-06 10:48:20 +02001583 *
1584 * Note: only writes the actual inode, no associated data or other metadata.
1585 */
1586int sync_inode_metadata(struct inode *inode, int wait)
1587{
1588 struct writeback_control wbc = {
1589 .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE,
1590 .nr_to_write = 0, /* metadata-only */
1591 };
1592
1593 return sync_inode(inode, &wbc);
1594}
1595EXPORT_SYMBOL(sync_inode_metadata);