blob: 310972b72a6654c1597fc11f717954606561c47e [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>
25#include <linux/freezer.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include <linux/writeback.h>
27#include <linux/blkdev.h>
28#include <linux/backing-dev.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/*
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;
Wu Fengguangd46db3d2011-05-04 19:54:37 -060043 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;
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -060049 enum wb_reason reason; /* why was writeback initiated? */
Jens Axboec4a77a62009-09-16 15:18:25 +020050
Jens Axboe8010c3b2009-09-15 20:04:57 +020051 struct list_head list; /* pending work list */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +020052 struct completion *done; /* set if the caller waits */
Jens Axboe03ba3782009-09-09 09:08:54 +020053};
54
Adrian Bunkf11b00f2008-04-29 00:58:56 -070055/**
56 * writeback_in_progress - determine whether there is writeback in progress
57 * @bdi: the device's backing_dev_info structure.
58 *
Jens Axboe03ba3782009-09-09 09:08:54 +020059 * Determine whether there is writeback waiting to be handled against a
60 * backing device.
Adrian Bunkf11b00f2008-04-29 00:58:56 -070061 */
62int writeback_in_progress(struct backing_dev_info *bdi)
63{
Jan Kara81d73a32010-08-11 14:17:44 -070064 return test_bit(BDI_writeback_running, &bdi->state);
Adrian Bunkf11b00f2008-04-29 00:58:56 -070065}
Theodore Ts'o00d4e732012-09-19 22:42:36 -040066EXPORT_SYMBOL(writeback_in_progress);
Adrian Bunkf11b00f2008-04-29 00:58:56 -070067
Jan Kara692ebd12010-09-21 11:51:01 +020068static inline struct backing_dev_info *inode_to_bdi(struct inode *inode)
69{
70 struct super_block *sb = inode->i_sb;
Jan Kara692ebd12010-09-21 11:51:01 +020071
Christoph Hellwigaaead252010-10-04 14:25:33 +020072 if (strcmp(sb->s_type->name, "bdev") == 0)
73 return inode->i_mapping->backing_dev_info;
74
75 return sb->s_bdi;
Jan Kara692ebd12010-09-21 11:51:01 +020076}
77
Nick Piggin7ccf19a2010-10-21 11:49:30 +110078static inline struct inode *wb_inode(struct list_head *head)
79{
80 return list_entry(head, struct inode, i_wb_list);
81}
82
Wu Fengguang15eb77a2012-01-17 11:18:56 -060083/*
84 * Include the creation of the trace points after defining the
85 * wb_writeback_work structure and inline functions so that the definition
86 * remains local to this file.
87 */
88#define CREATE_TRACE_POINTS
89#include <trace/events/writeback.h>
90
Jan Kara65850272011-01-13 15:45:44 -080091/* Wakeup flusher thread or forker thread to fork it. Requires bdi->wb_lock. */
92static void bdi_wakeup_flusher(struct backing_dev_info *bdi)
Nick Piggin4195f732009-05-28 09:01:15 +020093{
Artem Bityutskiyfff5b852010-07-25 14:29:20 +030094 if (bdi->wb.task) {
95 wake_up_process(bdi->wb.task);
96 } else {
97 /*
98 * The bdi thread isn't there, wake up the forker thread which
99 * will create and run it.
100 */
Jens Axboe03ba3782009-09-09 09:08:54 +0200101 wake_up_process(default_backing_dev_info.wb.task);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102 }
Jan Kara65850272011-01-13 15:45:44 -0800103}
104
105static void bdi_queue_work(struct backing_dev_info *bdi,
106 struct wb_writeback_work *work)
107{
108 trace_writeback_queue(bdi, work);
109
110 spin_lock_bh(&bdi->wb_lock);
111 list_add_tail(&work->list, &bdi->work_list);
112 if (!bdi->wb.task)
113 trace_writeback_nothread(bdi, work);
114 bdi_wakeup_flusher(bdi);
Artem Bityutskiy64677162010-07-25 14:29:22 +0300115 spin_unlock_bh(&bdi->wb_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116}
117
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200118static void
119__bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600120 bool range_cyclic, enum wb_reason reason)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200122 struct wb_writeback_work *work;
Jens Axboe03ba3782009-09-09 09:08:54 +0200123
Jens Axboebcddc3f2009-09-13 20:07:36 +0200124 /*
125 * This is WB_SYNC_NONE writeback, so if allocation fails just
126 * wakeup the thread for old dirty data writeback
127 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200128 work = kzalloc(sizeof(*work), GFP_ATOMIC);
129 if (!work) {
Dave Chinner455b2862010-07-07 13:24:06 +1000130 if (bdi->wb.task) {
131 trace_writeback_nowork(bdi);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200132 wake_up_process(bdi->wb.task);
Dave Chinner455b2862010-07-07 13:24:06 +1000133 }
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200134 return;
Jens Axboebcddc3f2009-09-13 20:07:36 +0200135 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200136
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200137 work->sync_mode = WB_SYNC_NONE;
138 work->nr_pages = nr_pages;
139 work->range_cyclic = range_cyclic;
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600140 work->reason = reason;
Christoph Hellwigf0fad8a2009-09-11 09:47:56 +0200141
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200142 bdi_queue_work(bdi, work);
Jens Axboeb6e51312009-09-16 15:13:54 +0200143}
144
145/**
146 * bdi_start_writeback - start writeback
147 * @bdi: the backing device to write from
148 * @nr_pages: the number of pages to write
Marcos Paulo de Souza786228a2011-11-23 20:56:45 +0800149 * @reason: reason why some writeback work was initiated
Jens Axboeb6e51312009-09-16 15:13:54 +0200150 *
151 * Description:
152 * This does WB_SYNC_NONE opportunistic writeback. The IO is only
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300153 * started when this function returns, we make no guarantees on
Jens Axboe0e3c9a22010-06-01 11:08:43 +0200154 * completion. Caller need not hold sb s_umount semaphore.
Jens Axboeb6e51312009-09-16 15:13:54 +0200155 *
156 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600157void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages,
158 enum wb_reason reason)
Jens Axboeb6e51312009-09-16 15:13:54 +0200159{
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600160 __bdi_start_writeback(bdi, nr_pages, true, reason);
Christoph Hellwigc5444192010-06-08 18:15:15 +0200161}
Wu Fengguangd3ddec72009-09-23 20:33:40 +0800162
Christoph Hellwigc5444192010-06-08 18:15:15 +0200163/**
164 * bdi_start_background_writeback - start background writeback
165 * @bdi: the backing device to write from
166 *
167 * Description:
Jan Kara65850272011-01-13 15:45:44 -0800168 * This makes sure WB_SYNC_NONE background writeback happens. When
169 * this function returns, it is only guaranteed that for given BDI
170 * some IO is happening if we are over background dirty threshold.
171 * Caller need not hold sb s_umount semaphore.
Christoph Hellwigc5444192010-06-08 18:15:15 +0200172 */
173void bdi_start_background_writeback(struct backing_dev_info *bdi)
174{
Jan Kara65850272011-01-13 15:45:44 -0800175 /*
176 * We just wake up the flusher thread. It will perform background
177 * writeback as soon as there is no other work to do.
178 */
Wu Fengguang71927e82011-01-13 15:45:46 -0800179 trace_writeback_wake_background(bdi);
Jan Kara65850272011-01-13 15:45:44 -0800180 spin_lock_bh(&bdi->wb_lock);
181 bdi_wakeup_flusher(bdi);
182 spin_unlock_bh(&bdi->wb_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183}
184
185/*
Dave Chinnera66979a2011-03-22 22:23:41 +1100186 * Remove the inode from the writeback list it is on.
187 */
188void inode_wb_list_del(struct inode *inode)
189{
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600190 struct backing_dev_info *bdi = inode_to_bdi(inode);
Dave Chinnera66979a2011-03-22 22:23:41 +1100191
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600192 spin_lock(&bdi->wb.list_lock);
193 list_del_init(&inode->i_wb_list);
194 spin_unlock(&bdi->wb.list_lock);
195}
Dave Chinnera66979a2011-03-22 22:23:41 +1100196
197/*
Andrew Morton6610a0b2007-10-16 23:30:32 -0700198 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
199 * furthest end of its superblock's dirty-inode list.
200 *
201 * Before stamping the inode's ->dirtied_when, we check to see whether it is
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200202 * already the most-recently-dirtied inode on the b_dirty list. If that is
Andrew Morton6610a0b2007-10-16 23:30:32 -0700203 * the case then the inode must have been redirtied while it was being written
204 * out and we don't reset its dirtied_when.
205 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600206static void redirty_tail(struct inode *inode, struct bdi_writeback *wb)
Andrew Morton6610a0b2007-10-16 23:30:32 -0700207{
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600208 assert_spin_locked(&wb->list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200209 if (!list_empty(&wb->b_dirty)) {
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200210 struct inode *tail;
Andrew Morton6610a0b2007-10-16 23:30:32 -0700211
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100212 tail = wb_inode(wb->b_dirty.next);
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200213 if (time_before(inode->dirtied_when, tail->dirtied_when))
Andrew Morton6610a0b2007-10-16 23:30:32 -0700214 inode->dirtied_when = jiffies;
215 }
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100216 list_move(&inode->i_wb_list, &wb->b_dirty);
Andrew Morton6610a0b2007-10-16 23:30:32 -0700217}
218
219/*
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200220 * requeue inode for re-scanning after bdi->b_io list is exhausted.
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700221 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600222static void requeue_io(struct inode *inode, struct bdi_writeback *wb)
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700223{
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600224 assert_spin_locked(&wb->list_lock);
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100225 list_move(&inode->i_wb_list, &wb->b_more_io);
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700226}
227
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700228static void inode_sync_complete(struct inode *inode)
229{
Jan Kara365b94ae2012-05-03 14:47:55 +0200230 inode->i_state &= ~I_SYNC;
Jan Kara4eff96d2012-11-26 16:29:51 -0800231 /* If inode is clean an unused, put it into LRU now... */
232 inode_add_lru(inode);
Jan Kara365b94ae2012-05-03 14:47:55 +0200233 /* Waiters must see I_SYNC cleared before being woken up */
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700234 smp_mb();
235 wake_up_bit(&inode->i_state, __I_SYNC);
236}
237
Jeff Laytond2caa3c52009-04-02 16:56:37 -0700238static bool inode_dirtied_after(struct inode *inode, unsigned long t)
239{
240 bool ret = time_after(inode->dirtied_when, t);
241#ifndef CONFIG_64BIT
242 /*
243 * For inodes being constantly redirtied, dirtied_when can get stuck.
244 * It _appears_ to be in the future, but is actually in distant past.
245 * This test is necessary to prevent such wrapped-around relative times
Jens Axboe5b0830c2009-09-23 19:37:09 +0200246 * from permanently stopping the whole bdi writeback.
Jeff Laytond2caa3c52009-04-02 16:56:37 -0700247 */
248 ret = ret && time_before_eq(inode->dirtied_when, jiffies);
249#endif
250 return ret;
251}
252
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700253/*
Wang Sheng-Hui0e2f2b22012-09-11 08:28:18 +0800254 * Move expired (dirtied before work->older_than_this) dirty inodes from
Jan Kara697e6fe2012-03-09 07:26:22 -0800255 * @delaying_queue to @dispatch_queue.
Fengguang Wu2c136572007-10-16 23:30:39 -0700256 */
Wu Fengguange84d0a42011-04-23 12:27:27 -0600257static int move_expired_inodes(struct list_head *delaying_queue,
Fengguang Wu2c136572007-10-16 23:30:39 -0700258 struct list_head *dispatch_queue,
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600259 struct wb_writeback_work *work)
Fengguang Wu2c136572007-10-16 23:30:39 -0700260{
Shaohua Li5c034492009-09-24 14:42:33 +0200261 LIST_HEAD(tmp);
262 struct list_head *pos, *node;
Jens Axboecf137302009-09-24 15:12:57 +0200263 struct super_block *sb = NULL;
Shaohua Li5c034492009-09-24 14:42:33 +0200264 struct inode *inode;
Jens Axboecf137302009-09-24 15:12:57 +0200265 int do_sb_sort = 0;
Wu Fengguange84d0a42011-04-23 12:27:27 -0600266 int moved = 0;
Shaohua Li5c034492009-09-24 14:42:33 +0200267
Fengguang Wu2c136572007-10-16 23:30:39 -0700268 while (!list_empty(delaying_queue)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100269 inode = wb_inode(delaying_queue->prev);
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600270 if (work->older_than_this &&
271 inode_dirtied_after(inode, *work->older_than_this))
Fengguang Wu2c136572007-10-16 23:30:39 -0700272 break;
Jens Axboecf137302009-09-24 15:12:57 +0200273 if (sb && sb != inode->i_sb)
274 do_sb_sort = 1;
275 sb = inode->i_sb;
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100276 list_move(&inode->i_wb_list, &tmp);
Wu Fengguange84d0a42011-04-23 12:27:27 -0600277 moved++;
Shaohua Li5c034492009-09-24 14:42:33 +0200278 }
279
Jens Axboecf137302009-09-24 15:12:57 +0200280 /* just one sb in list, splice to dispatch_queue and we're done */
281 if (!do_sb_sort) {
282 list_splice(&tmp, dispatch_queue);
Wu Fengguange84d0a42011-04-23 12:27:27 -0600283 goto out;
Jens Axboecf137302009-09-24 15:12:57 +0200284 }
285
Shaohua Li5c034492009-09-24 14:42:33 +0200286 /* Move inodes from one superblock together */
287 while (!list_empty(&tmp)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100288 sb = wb_inode(tmp.prev)->i_sb;
Shaohua Li5c034492009-09-24 14:42:33 +0200289 list_for_each_prev_safe(pos, node, &tmp) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100290 inode = wb_inode(pos);
Shaohua Li5c034492009-09-24 14:42:33 +0200291 if (inode->i_sb == sb)
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100292 list_move(&inode->i_wb_list, dispatch_queue);
Shaohua Li5c034492009-09-24 14:42:33 +0200293 }
Fengguang Wu2c136572007-10-16 23:30:39 -0700294 }
Wu Fengguange84d0a42011-04-23 12:27:27 -0600295out:
296 return moved;
Fengguang Wu2c136572007-10-16 23:30:39 -0700297}
298
299/*
300 * Queue all expired dirty inodes for io, eldest first.
Wu Fengguang4ea879b2010-08-11 14:17:42 -0700301 * Before
302 * newly dirtied b_dirty b_io b_more_io
303 * =============> gf edc BA
304 * After
305 * newly dirtied b_dirty b_io b_more_io
306 * =============> g fBAedc
307 * |
308 * +--> dequeue for IO
Fengguang Wu2c136572007-10-16 23:30:39 -0700309 */
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600310static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work)
Fengguang Wu2c136572007-10-16 23:30:39 -0700311{
Wu Fengguange84d0a42011-04-23 12:27:27 -0600312 int moved;
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600313 assert_spin_locked(&wb->list_lock);
Wu Fengguang4ea879b2010-08-11 14:17:42 -0700314 list_splice_init(&wb->b_more_io, &wb->b_io);
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600315 moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, work);
316 trace_writeback_queue_io(wb, work, moved);
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200317}
318
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100319static int write_inode(struct inode *inode, struct writeback_control *wbc)
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200320{
Jens Axboe03ba3782009-09-09 09:08:54 +0200321 if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode))
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100322 return inode->i_sb->s_op->write_inode(inode, wbc);
Jens Axboe03ba3782009-09-09 09:08:54 +0200323 return 0;
Fengguang Wu2c136572007-10-16 23:30:39 -0700324}
325
326/*
Jan Kara169ebd92012-05-03 14:48:03 +0200327 * Wait for writeback on an inode to complete. Called with i_lock held.
328 * Caller must make sure inode cannot go away when we drop i_lock.
Christoph Hellwig01c03192009-06-08 13:35:40 +0200329 */
Jan Kara169ebd92012-05-03 14:48:03 +0200330static void __inode_wait_for_writeback(struct inode *inode)
331 __releases(inode->i_lock)
332 __acquires(inode->i_lock)
Christoph Hellwig01c03192009-06-08 13:35:40 +0200333{
334 DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC);
335 wait_queue_head_t *wqh;
336
337 wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
Dave Chinner250df6e2011-03-22 22:23:36 +1100338 while (inode->i_state & I_SYNC) {
339 spin_unlock(&inode->i_lock);
Christoph Hellwig01c03192009-06-08 13:35:40 +0200340 __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE);
Dave Chinner250df6e2011-03-22 22:23:36 +1100341 spin_lock(&inode->i_lock);
Richard Kennedy58a9d3d82010-05-24 14:32:38 -0700342 }
Christoph Hellwig01c03192009-06-08 13:35:40 +0200343}
344
345/*
Jan Kara169ebd92012-05-03 14:48:03 +0200346 * Wait for writeback on an inode to complete. Caller must have inode pinned.
347 */
348void inode_wait_for_writeback(struct inode *inode)
349{
350 spin_lock(&inode->i_lock);
351 __inode_wait_for_writeback(inode);
352 spin_unlock(&inode->i_lock);
353}
354
355/*
356 * Sleep until I_SYNC is cleared. This function must be called with i_lock
357 * held and drops it. It is aimed for callers not holding any inode reference
358 * so once i_lock is dropped, inode can go away.
359 */
360static void inode_sleep_on_writeback(struct inode *inode)
361 __releases(inode->i_lock)
362{
363 DEFINE_WAIT(wait);
364 wait_queue_head_t *wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
365 int sleep;
366
367 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
368 sleep = inode->i_state & I_SYNC;
369 spin_unlock(&inode->i_lock);
370 if (sleep)
371 schedule();
372 finish_wait(wqh, &wait);
373}
374
375/*
Jan Karaccb26b52012-05-03 14:47:58 +0200376 * Find proper writeback list for the inode depending on its current state and
377 * possibly also change of its state while we were doing writeback. Here we
378 * handle things such as livelock prevention or fairness of writeback among
379 * inodes. This function can be called only by flusher thread - noone else
380 * processes all inodes in writeback lists and requeueing inodes behind flusher
381 * thread's back can have unexpected consequences.
382 */
383static void requeue_inode(struct inode *inode, struct bdi_writeback *wb,
384 struct writeback_control *wbc)
385{
386 if (inode->i_state & I_FREEING)
387 return;
388
389 /*
390 * Sync livelock prevention. Each inode is tagged and synced in one
391 * shot. If still dirty, it will be redirty_tail()'ed below. Update
392 * the dirty time to prevent enqueue and sync it again.
393 */
394 if ((inode->i_state & I_DIRTY) &&
395 (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages))
396 inode->dirtied_when = jiffies;
397
Jan Kara4f8ad652012-05-03 14:48:00 +0200398 if (wbc->pages_skipped) {
399 /*
400 * writeback is not making progress due to locked
401 * buffers. Skip this inode for now.
402 */
403 redirty_tail(inode, wb);
404 return;
405 }
406
Jan Karaccb26b52012-05-03 14:47:58 +0200407 if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) {
408 /*
409 * We didn't write back all the pages. nfs_writepages()
410 * sometimes bales out without doing anything.
411 */
412 if (wbc->nr_to_write <= 0) {
413 /* Slice used up. Queue for next turn. */
414 requeue_io(inode, wb);
415 } else {
416 /*
417 * Writeback blocked by something other than
418 * congestion. Delay the inode for some time to
419 * avoid spinning on the CPU (100% iowait)
420 * retrying writeback of the dirty page/inode
421 * that cannot be performed immediately.
422 */
423 redirty_tail(inode, wb);
424 }
425 } else if (inode->i_state & I_DIRTY) {
426 /*
427 * Filesystems can dirty the inode during writeback operations,
428 * such as delayed allocation during submission or metadata
429 * updates after data IO completion.
430 */
431 redirty_tail(inode, wb);
432 } else {
433 /* The inode is clean. Remove from writeback lists. */
434 list_del_init(&inode->i_wb_list);
435 }
436}
437
438/*
Jan Kara4f8ad652012-05-03 14:48:00 +0200439 * Write out an inode and its dirty pages. Do not update the writeback list
440 * linkage. That is left to the caller. The caller is also responsible for
441 * setting I_SYNC flag and calling inode_sync_complete() to clear it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 */
443static int
Yan Hongcd8ed2a2012-10-08 16:33:45 -0700444__writeback_single_inode(struct inode *inode, struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446 struct address_space *mapping = inode->i_mapping;
Wu Fengguang251d6a42010-12-01 17:33:37 -0600447 long nr_to_write = wbc->nr_to_write;
Christoph Hellwig01c03192009-06-08 13:35:40 +0200448 unsigned dirty;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449 int ret;
450
Jan Kara4f8ad652012-05-03 14:48:00 +0200451 WARN_ON(!(inode->i_state & I_SYNC));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452
453 ret = do_writepages(mapping, wbc);
454
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100455 /*
456 * Make sure to wait on the data before writing out the metadata.
457 * This is important for filesystems that modify metadata on data
458 * I/O completion.
459 */
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100460 if (wbc->sync_mode == WB_SYNC_ALL) {
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100461 int err = filemap_fdatawait(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462 if (ret == 0)
463 ret = err;
464 }
465
Dmitry Monakhov5547e8a2010-05-07 13:35:44 +0400466 /*
467 * Some filesystems may redirty the inode during the writeback
468 * due to delalloc, clear dirty metadata flags right before
469 * write_inode()
470 */
Dave Chinner250df6e2011-03-22 22:23:36 +1100471 spin_lock(&inode->i_lock);
Jan Kara6290be12012-05-03 14:47:57 +0200472 /* Clear I_DIRTY_PAGES if we've written out all dirty pages */
473 if (!mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
474 inode->i_state &= ~I_DIRTY_PAGES;
Dmitry Monakhov5547e8a2010-05-07 13:35:44 +0400475 dirty = inode->i_state & I_DIRTY;
476 inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC);
Dave Chinner250df6e2011-03-22 22:23:36 +1100477 spin_unlock(&inode->i_lock);
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100478 /* Don't write the inode if only I_DIRTY_PAGES was set */
479 if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100480 int err = write_inode(inode, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481 if (ret == 0)
482 ret = err;
483 }
Jan Kara4f8ad652012-05-03 14:48:00 +0200484 trace_writeback_single_inode(inode, wbc, nr_to_write);
485 return ret;
486}
487
488/*
489 * Write out an inode's dirty pages. Either the caller has an active reference
490 * on the inode or the inode has I_WILL_FREE set.
491 *
492 * This function is designed to be called for writing back one inode which
493 * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode()
494 * and does more profound writeback list handling in writeback_sb_inodes().
495 */
496static int
497writeback_single_inode(struct inode *inode, struct bdi_writeback *wb,
498 struct writeback_control *wbc)
499{
500 int ret = 0;
501
502 spin_lock(&inode->i_lock);
503 if (!atomic_read(&inode->i_count))
504 WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING)));
505 else
506 WARN_ON(inode->i_state & I_WILL_FREE);
507
508 if (inode->i_state & I_SYNC) {
509 if (wbc->sync_mode != WB_SYNC_ALL)
510 goto out;
511 /*
Jan Kara169ebd92012-05-03 14:48:03 +0200512 * It's a data-integrity sync. We must wait. Since callers hold
513 * inode reference or inode has I_WILL_FREE set, it cannot go
514 * away under us.
Jan Kara4f8ad652012-05-03 14:48:00 +0200515 */
Jan Kara169ebd92012-05-03 14:48:03 +0200516 __inode_wait_for_writeback(inode);
Jan Kara4f8ad652012-05-03 14:48:00 +0200517 }
518 WARN_ON(inode->i_state & I_SYNC);
519 /*
520 * Skip inode if it is clean. We don't want to mess with writeback
521 * lists in this function since flusher thread may be doing for example
522 * sync in parallel and if we move the inode, it could get skipped. So
523 * here we make sure inode is on some writeback list and leave it there
524 * unless we have completely cleaned the inode.
525 */
526 if (!(inode->i_state & I_DIRTY))
527 goto out;
528 inode->i_state |= I_SYNC;
529 spin_unlock(&inode->i_lock);
530
Yan Hongcd8ed2a2012-10-08 16:33:45 -0700531 ret = __writeback_single_inode(inode, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600533 spin_lock(&wb->list_lock);
Dave Chinner250df6e2011-03-22 22:23:36 +1100534 spin_lock(&inode->i_lock);
Jan Kara4f8ad652012-05-03 14:48:00 +0200535 /*
536 * If inode is clean, remove it from writeback lists. Otherwise don't
537 * touch it. See comment above for explanation.
538 */
539 if (!(inode->i_state & I_DIRTY))
540 list_del_init(&inode->i_wb_list);
541 spin_unlock(&wb->list_lock);
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700542 inode_sync_complete(inode);
Jan Kara4f8ad652012-05-03 14:48:00 +0200543out:
544 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545 return ret;
546}
547
Wu Fengguang1a12d8b2010-08-29 13:28:09 -0600548static long writeback_chunk_size(struct backing_dev_info *bdi,
549 struct wb_writeback_work *work)
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600550{
551 long pages;
552
553 /*
554 * WB_SYNC_ALL mode does livelock avoidance by syncing dirty
555 * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX
556 * here avoids calling into writeback_inodes_wb() more than once.
557 *
558 * The intended call sequence for WB_SYNC_ALL writeback is:
559 *
560 * wb_writeback()
561 * writeback_sb_inodes() <== called only once
562 * write_cache_pages() <== called once for each inode
563 * (quickly) tag currently dirty pages
564 * (maybe slowly) sync all tagged pages
565 */
566 if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages)
567 pages = LONG_MAX;
Wu Fengguang1a12d8b2010-08-29 13:28:09 -0600568 else {
569 pages = min(bdi->avg_write_bandwidth / 2,
570 global_dirty_limit / DIRTY_SCOPE);
571 pages = min(pages, work->nr_pages);
572 pages = round_down(pages + MIN_WRITEBACK_PAGES,
573 MIN_WRITEBACK_PAGES);
574 }
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600575
576 return pages;
577}
578
Jens Axboe03ba3782009-09-09 09:08:54 +0200579/*
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800580 * Write a portion of b_io inodes which belong to @sb.
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200581 *
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600582 * Return the number of pages and/or inodes written.
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800583 */
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600584static long writeback_sb_inodes(struct super_block *sb,
585 struct bdi_writeback *wb,
586 struct wb_writeback_work *work)
Jens Axboe03ba3782009-09-09 09:08:54 +0200587{
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600588 struct writeback_control wbc = {
589 .sync_mode = work->sync_mode,
590 .tagged_writepages = work->tagged_writepages,
591 .for_kupdate = work->for_kupdate,
592 .for_background = work->for_background,
593 .range_cyclic = work->range_cyclic,
594 .range_start = 0,
595 .range_end = LLONG_MAX,
596 };
597 unsigned long start_time = jiffies;
598 long write_chunk;
599 long wrote = 0; /* count both pages and inodes */
600
Jens Axboe03ba3782009-09-09 09:08:54 +0200601 while (!list_empty(&wb->b_io)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100602 struct inode *inode = wb_inode(wb->b_io.prev);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200603
604 if (inode->i_sb != sb) {
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600605 if (work->sb) {
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200606 /*
607 * We only want to write back data for this
608 * superblock, move all inodes not belonging
609 * to it back onto the dirty list.
610 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600611 redirty_tail(inode, wb);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200612 continue;
613 }
614
615 /*
616 * The inode belongs to a different superblock.
617 * Bounce back to the caller to unpin this and
618 * pin the next superblock.
619 */
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600620 break;
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200621 }
622
Christoph Hellwig9843b762010-10-24 19:40:46 +0200623 /*
Wanpeng Li331cbde2012-06-09 11:10:55 +0800624 * Don't bother with new inodes or inodes being freed, first
625 * kind does not need periodic writeout yet, and for the latter
Christoph Hellwig9843b762010-10-24 19:40:46 +0200626 * kind writeout is handled by the freer.
627 */
Dave Chinner250df6e2011-03-22 22:23:36 +1100628 spin_lock(&inode->i_lock);
Christoph Hellwig9843b762010-10-24 19:40:46 +0200629 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
Dave Chinner250df6e2011-03-22 22:23:36 +1100630 spin_unlock(&inode->i_lock);
Wu Fengguangfcc5c222011-07-11 23:08:50 -0700631 redirty_tail(inode, wb);
Nick Piggin7ef0d732009-03-12 14:31:38 -0700632 continue;
633 }
Jan Karacc1676d2012-05-03 14:47:56 +0200634 if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) {
635 /*
636 * If this inode is locked for writeback and we are not
637 * doing writeback-for-data-integrity, move it to
638 * b_more_io so that writeback can proceed with the
639 * other inodes on s_io.
640 *
641 * We'll have another go at writing back this inode
642 * when we completed a full scan of b_io.
643 */
644 spin_unlock(&inode->i_lock);
645 requeue_io(inode, wb);
646 trace_writeback_sb_inodes_requeue(inode);
647 continue;
648 }
Jan Karaf0d07b72012-05-03 14:47:59 +0200649 spin_unlock(&wb->list_lock);
650
Jan Kara4f8ad652012-05-03 14:48:00 +0200651 /*
652 * We already requeued the inode if it had I_SYNC set and we
653 * are doing WB_SYNC_NONE writeback. So this catches only the
654 * WB_SYNC_ALL case.
655 */
Jan Kara169ebd92012-05-03 14:48:03 +0200656 if (inode->i_state & I_SYNC) {
657 /* Wait for I_SYNC. This function drops i_lock... */
658 inode_sleep_on_writeback(inode);
659 /* Inode may be gone, start again */
Jan Karaead188f2012-06-08 17:07:36 +0200660 spin_lock(&wb->list_lock);
Jan Kara169ebd92012-05-03 14:48:03 +0200661 continue;
662 }
Jan Kara4f8ad652012-05-03 14:48:00 +0200663 inode->i_state |= I_SYNC;
664 spin_unlock(&inode->i_lock);
Jan Kara169ebd92012-05-03 14:48:03 +0200665
Wu Fengguang1a12d8b2010-08-29 13:28:09 -0600666 write_chunk = writeback_chunk_size(wb->bdi, work);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600667 wbc.nr_to_write = write_chunk;
668 wbc.pages_skipped = 0;
Dave Chinner250df6e2011-03-22 22:23:36 +1100669
Jan Kara169ebd92012-05-03 14:48:03 +0200670 /*
671 * We use I_SYNC to pin the inode in memory. While it is set
672 * evict_inode() will wait so the inode cannot be freed.
673 */
Yan Hongcd8ed2a2012-10-08 16:33:45 -0700674 __writeback_single_inode(inode, &wbc);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600675
676 work->nr_pages -= write_chunk - wbc.nr_to_write;
677 wrote += write_chunk - wbc.nr_to_write;
Jan Kara4f8ad652012-05-03 14:48:00 +0200678 spin_lock(&wb->list_lock);
679 spin_lock(&inode->i_lock);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600680 if (!(inode->i_state & I_DIRTY))
681 wrote++;
Jan Kara4f8ad652012-05-03 14:48:00 +0200682 requeue_inode(inode, wb, &wbc);
683 inode_sync_complete(inode);
Dave Chinner0f1b1fd2011-03-22 22:23:43 +1100684 spin_unlock(&inode->i_lock);
Jan Kara169ebd92012-05-03 14:48:03 +0200685 cond_resched_lock(&wb->list_lock);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600686 /*
687 * bail out to wb_writeback() often enough to check
688 * background threshold and other termination conditions.
689 */
690 if (wrote) {
691 if (time_is_before_jiffies(start_time + HZ / 10UL))
692 break;
693 if (work->nr_pages <= 0)
694 break;
Fengguang Wu8bc3be22008-02-04 22:29:36 -0800695 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696 }
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600697 return wrote;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800698}
Nick Piggin38f21972009-01-06 14:40:25 -0800699
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600700static long __writeback_inodes_wb(struct bdi_writeback *wb,
701 struct wb_writeback_work *work)
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800702{
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600703 unsigned long start_time = jiffies;
704 long wrote = 0;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800705
706 while (!list_empty(&wb->b_io)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100707 struct inode *inode = wb_inode(wb->b_io.prev);
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800708 struct super_block *sb = inode->i_sb;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800709
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000710 if (!grab_super_passive(sb)) {
Wu Fengguang0e9958162011-07-29 22:14:35 -0600711 /*
712 * grab_super_passive() may fail consistently due to
713 * s_umount being grabbed by someone else. Don't use
714 * requeue_io() to avoid busy retrying the inode/sb.
715 */
716 redirty_tail(inode, wb);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200717 continue;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800718 }
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600719 wrote += writeback_sb_inodes(sb, wb, work);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200720 drop_super(sb);
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800721
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600722 /* refer to the same tests at the end of writeback_sb_inodes */
723 if (wrote) {
724 if (time_is_before_jiffies(start_time + HZ / 10UL))
725 break;
726 if (work->nr_pages <= 0)
727 break;
728 }
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800729 }
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200730 /* Leave any unwritten inodes on b_io */
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600731 return wrote;
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200732}
733
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600734long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages,
735 enum wb_reason reason)
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200736{
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600737 struct wb_writeback_work work = {
738 .nr_pages = nr_pages,
739 .sync_mode = WB_SYNC_NONE,
740 .range_cyclic = 1,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600741 .reason = reason,
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600742 };
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200743
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600744 spin_lock(&wb->list_lock);
Wu Fengguang424b3512010-07-21 20:11:53 -0600745 if (list_empty(&wb->b_io))
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600746 queue_io(wb, &work);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600747 __writeback_inodes_wb(wb, &work);
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600748 spin_unlock(&wb->list_lock);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200749
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600750 return nr_pages - work.nr_pages;
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200751}
752
Wu Fengguangb00949a2010-11-18 14:38:33 -0600753static bool over_bground_thresh(struct backing_dev_info *bdi)
Jens Axboe03ba3782009-09-09 09:08:54 +0200754{
755 unsigned long background_thresh, dirty_thresh;
756
Wu Fengguang16c40422010-08-11 14:17:39 -0700757 global_dirty_limits(&background_thresh, &dirty_thresh);
Jens Axboe03ba3782009-09-09 09:08:54 +0200758
Wu Fengguangb00949a2010-11-18 14:38:33 -0600759 if (global_page_state(NR_FILE_DIRTY) +
760 global_page_state(NR_UNSTABLE_NFS) > background_thresh)
761 return true;
762
763 if (bdi_stat(bdi, BDI_RECLAIMABLE) >
764 bdi_dirty_limit(bdi, background_thresh))
765 return true;
766
767 return false;
Jens Axboe03ba3782009-09-09 09:08:54 +0200768}
769
770/*
Wu Fengguange98be2d2010-08-29 11:22:30 -0600771 * Called under wb->list_lock. If there are multiple wb per bdi,
772 * only the flusher working on the first wb should do it.
773 */
774static void wb_update_bandwidth(struct bdi_writeback *wb,
775 unsigned long start_time)
776{
Wu Fengguangaf6a3112011-10-03 20:46:17 -0600777 __bdi_update_bandwidth(wb->bdi, 0, 0, 0, 0, 0, start_time);
Wu Fengguange98be2d2010-08-29 11:22:30 -0600778}
779
780/*
Jens Axboe03ba3782009-09-09 09:08:54 +0200781 * Explicit flushing or periodic writeback of "old" data.
782 *
783 * Define "old": the first time one of an inode's pages is dirtied, we mark the
784 * dirtying-time in the inode's address_space. So this periodic writeback code
785 * just walks the superblock inode list, writing back any inodes which are
786 * older than a specific point in time.
787 *
788 * Try to run once per dirty_writeback_interval. But if a writeback event
789 * takes longer than a dirty_writeback_interval interval, then leave a
790 * one-second gap.
791 *
792 * older_than_this takes precedence over nr_to_write. So we'll only write back
793 * all dirty pages if they are all attached to "old" mappings.
794 */
Jens Axboec4a77a62009-09-16 15:18:25 +0200795static long wb_writeback(struct bdi_writeback *wb,
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200796 struct wb_writeback_work *work)
Jens Axboe03ba3782009-09-09 09:08:54 +0200797{
Wu Fengguange98be2d2010-08-29 11:22:30 -0600798 unsigned long wb_start = jiffies;
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600799 long nr_pages = work->nr_pages;
Jens Axboe03ba3782009-09-09 09:08:54 +0200800 unsigned long oldest_jif;
Jan Karaa5989bd2009-09-16 19:22:48 +0200801 struct inode *inode;
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600802 long progress;
Jens Axboe03ba3782009-09-09 09:08:54 +0200803
Wu Fengguange185dda2011-04-23 11:26:07 -0600804 oldest_jif = jiffies;
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600805 work->older_than_this = &oldest_jif;
Jens Axboe03ba3782009-09-09 09:08:54 +0200806
Wu Fengguange8dfc302011-04-21 12:06:32 -0600807 spin_lock(&wb->list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200808 for (;;) {
809 /*
Wu Fengguangd3ddec72009-09-23 20:33:40 +0800810 * Stop writeback when nr_pages has been consumed
Jens Axboe03ba3782009-09-09 09:08:54 +0200811 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200812 if (work->nr_pages <= 0)
Jens Axboe03ba3782009-09-09 09:08:54 +0200813 break;
814
815 /*
Jan Karaaa373cf2011-01-13 15:45:47 -0800816 * Background writeout and kupdate-style writeback may
817 * run forever. Stop them if there is other work to do
818 * so that e.g. sync can proceed. They'll be restarted
819 * after the other works are all done.
820 */
821 if ((work->for_background || work->for_kupdate) &&
822 !list_empty(&wb->bdi->work_list))
823 break;
824
825 /*
Wu Fengguangd3ddec72009-09-23 20:33:40 +0800826 * For background writeout, stop when we are below the
827 * background dirty threshold
Jens Axboe03ba3782009-09-09 09:08:54 +0200828 */
Wu Fengguangb00949a2010-11-18 14:38:33 -0600829 if (work->for_background && !over_bground_thresh(wb->bdi))
Jens Axboe03ba3782009-09-09 09:08:54 +0200830 break;
831
Jan Kara1bc36b62011-10-19 11:44:41 +0200832 /*
833 * Kupdate and background works are special and we want to
834 * include all inodes that need writing. Livelock avoidance is
835 * handled by these works yielding to any other work so we are
836 * safe.
837 */
Wu Fengguangba9aa832010-07-21 20:32:30 -0600838 if (work->for_kupdate) {
839 oldest_jif = jiffies -
840 msecs_to_jiffies(dirty_expire_interval * 10);
Jan Kara1bc36b62011-10-19 11:44:41 +0200841 } else if (work->for_background)
842 oldest_jif = jiffies;
Dave Chinner028c2dd2010-07-07 13:24:07 +1000843
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600844 trace_writeback_start(wb->bdi, work);
Wu Fengguange8dfc302011-04-21 12:06:32 -0600845 if (list_empty(&wb->b_io))
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600846 queue_io(wb, work);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200847 if (work->sb)
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600848 progress = writeback_sb_inodes(work->sb, wb, work);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200849 else
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600850 progress = __writeback_inodes_wb(wb, work);
851 trace_writeback_written(wb->bdi, work);
Dave Chinner028c2dd2010-07-07 13:24:07 +1000852
Wu Fengguange98be2d2010-08-29 11:22:30 -0600853 wb_update_bandwidth(wb, wb_start);
Jens Axboe03ba3782009-09-09 09:08:54 +0200854
855 /*
Jens Axboe71fd05a2009-09-23 19:32:26 +0200856 * Did we write something? Try for more
Wu Fengguange6fb6da2010-07-22 10:23:44 -0600857 *
858 * Dirty inodes are moved to b_io for writeback in batches.
859 * The completion of the current batch does not necessarily
860 * mean the overall work is done. So we keep looping as long
861 * as made some progress on cleaning pages or inodes.
Jens Axboe71fd05a2009-09-23 19:32:26 +0200862 */
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600863 if (progress)
Jens Axboe71fd05a2009-09-23 19:32:26 +0200864 continue;
865 /*
Wu Fengguange6fb6da2010-07-22 10:23:44 -0600866 * No more inodes for IO, bail
Jens Axboe71fd05a2009-09-23 19:32:26 +0200867 */
Wu Fengguangb7a24412010-07-21 22:19:51 -0600868 if (list_empty(&wb->b_more_io))
Jens Axboe03ba3782009-09-09 09:08:54 +0200869 break;
870 /*
Jens Axboe8010c3b2009-09-15 20:04:57 +0200871 * Nothing written. Wait for some inode to
872 * become available for writeback. Otherwise
873 * we'll just busyloop.
Jens Axboe03ba3782009-09-09 09:08:54 +0200874 */
Jens Axboe03ba3782009-09-09 09:08:54 +0200875 if (!list_empty(&wb->b_more_io)) {
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600876 trace_writeback_wait(wb->bdi, work);
Jens Axboe03ba3782009-09-09 09:08:54 +0200877 inode = wb_inode(wb->b_more_io.prev);
Dave Chinner250df6e2011-03-22 22:23:36 +1100878 spin_lock(&inode->i_lock);
Jan Karaf0d07b72012-05-03 14:47:59 +0200879 spin_unlock(&wb->list_lock);
Jan Kara169ebd92012-05-03 14:48:03 +0200880 /* This function drops i_lock... */
881 inode_sleep_on_writeback(inode);
Jan Karaf0d07b72012-05-03 14:47:59 +0200882 spin_lock(&wb->list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200883 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200884 }
Wu Fengguange8dfc302011-04-21 12:06:32 -0600885 spin_unlock(&wb->list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200886
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600887 return nr_pages - work->nr_pages;
Jens Axboe03ba3782009-09-09 09:08:54 +0200888}
889
890/*
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200891 * Return the next wb_writeback_work struct that hasn't been processed yet.
Jens Axboe03ba3782009-09-09 09:08:54 +0200892 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200893static struct wb_writeback_work *
Minchan Kim08852b62010-08-03 12:51:16 +0200894get_next_work_item(struct backing_dev_info *bdi)
Jens Axboe03ba3782009-09-09 09:08:54 +0200895{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200896 struct wb_writeback_work *work = NULL;
Jens Axboe03ba3782009-09-09 09:08:54 +0200897
Artem Bityutskiy64677162010-07-25 14:29:22 +0300898 spin_lock_bh(&bdi->wb_lock);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200899 if (!list_empty(&bdi->work_list)) {
900 work = list_entry(bdi->work_list.next,
901 struct wb_writeback_work, list);
902 list_del_init(&work->list);
Jens Axboe03ba3782009-09-09 09:08:54 +0200903 }
Artem Bityutskiy64677162010-07-25 14:29:22 +0300904 spin_unlock_bh(&bdi->wb_lock);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200905 return work;
Jens Axboe03ba3782009-09-09 09:08:54 +0200906}
907
Linus Torvaldscdf01dd2010-10-30 08:55:52 -0700908/*
909 * Add in the number of potentially dirty inodes, because each inode
910 * write can dirty pagecache in the underlying blockdev.
911 */
912static unsigned long get_nr_dirty_pages(void)
913{
914 return global_page_state(NR_FILE_DIRTY) +
915 global_page_state(NR_UNSTABLE_NFS) +
916 get_nr_dirty_inodes();
917}
918
Jan Kara65850272011-01-13 15:45:44 -0800919static long wb_check_background_flush(struct bdi_writeback *wb)
920{
Wu Fengguangb00949a2010-11-18 14:38:33 -0600921 if (over_bground_thresh(wb->bdi)) {
Jan Kara65850272011-01-13 15:45:44 -0800922
923 struct wb_writeback_work work = {
924 .nr_pages = LONG_MAX,
925 .sync_mode = WB_SYNC_NONE,
926 .for_background = 1,
927 .range_cyclic = 1,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600928 .reason = WB_REASON_BACKGROUND,
Jan Kara65850272011-01-13 15:45:44 -0800929 };
930
931 return wb_writeback(wb, &work);
932 }
933
934 return 0;
935}
936
Jens Axboe03ba3782009-09-09 09:08:54 +0200937static long wb_check_old_data_flush(struct bdi_writeback *wb)
938{
939 unsigned long expired;
940 long nr_pages;
941
Jens Axboe69b62d02010-05-17 12:51:03 +0200942 /*
943 * When set to zero, disable periodic writeback
944 */
945 if (!dirty_writeback_interval)
946 return 0;
947
Jens Axboe03ba3782009-09-09 09:08:54 +0200948 expired = wb->last_old_flush +
949 msecs_to_jiffies(dirty_writeback_interval * 10);
950 if (time_before(jiffies, expired))
951 return 0;
952
953 wb->last_old_flush = jiffies;
Linus Torvaldscdf01dd2010-10-30 08:55:52 -0700954 nr_pages = get_nr_dirty_pages();
Jens Axboe03ba3782009-09-09 09:08:54 +0200955
Jens Axboec4a77a62009-09-16 15:18:25 +0200956 if (nr_pages) {
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200957 struct wb_writeback_work work = {
Jens Axboec4a77a62009-09-16 15:18:25 +0200958 .nr_pages = nr_pages,
959 .sync_mode = WB_SYNC_NONE,
960 .for_kupdate = 1,
961 .range_cyclic = 1,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600962 .reason = WB_REASON_PERIODIC,
Jens Axboec4a77a62009-09-16 15:18:25 +0200963 };
964
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200965 return wb_writeback(wb, &work);
Jens Axboec4a77a62009-09-16 15:18:25 +0200966 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200967
968 return 0;
969}
970
971/*
972 * Retrieve work items and do the writeback they describe
973 */
974long wb_do_writeback(struct bdi_writeback *wb, int force_wait)
975{
976 struct backing_dev_info *bdi = wb->bdi;
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200977 struct wb_writeback_work *work;
Jens Axboec4a77a62009-09-16 15:18:25 +0200978 long wrote = 0;
Jens Axboe03ba3782009-09-09 09:08:54 +0200979
Jan Kara81d73a32010-08-11 14:17:44 -0700980 set_bit(BDI_writeback_running, &wb->bdi->state);
Minchan Kim08852b62010-08-03 12:51:16 +0200981 while ((work = get_next_work_item(bdi)) != NULL) {
Jens Axboe03ba3782009-09-09 09:08:54 +0200982 /*
983 * Override sync mode, in case we must wait for completion
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200984 * because this thread is exiting now.
Jens Axboe03ba3782009-09-09 09:08:54 +0200985 */
986 if (force_wait)
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200987 work->sync_mode = WB_SYNC_ALL;
988
Dave Chinner455b2862010-07-07 13:24:06 +1000989 trace_writeback_exec(bdi, work);
990
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200991 wrote += wb_writeback(wb, work);
Jens Axboe03ba3782009-09-09 09:08:54 +0200992
993 /*
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200994 * Notify the caller of completion if this is a synchronous
995 * work item, otherwise just free it.
Jens Axboe03ba3782009-09-09 09:08:54 +0200996 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200997 if (work->done)
998 complete(work->done);
999 else
1000 kfree(work);
Jens Axboe03ba3782009-09-09 09:08:54 +02001001 }
1002
1003 /*
1004 * Check for periodic writeback, kupdated() style
1005 */
1006 wrote += wb_check_old_data_flush(wb);
Jan Kara65850272011-01-13 15:45:44 -08001007 wrote += wb_check_background_flush(wb);
Jan Kara81d73a32010-08-11 14:17:44 -07001008 clear_bit(BDI_writeback_running, &wb->bdi->state);
Jens Axboe03ba3782009-09-09 09:08:54 +02001009
1010 return wrote;
1011}
1012
1013/*
1014 * Handle writeback of dirty data for the device backed by this bdi. Also
1015 * wakes up periodically and does kupdated style flushing.
1016 */
Christoph Hellwig08243902010-06-19 23:08:22 +02001017int bdi_writeback_thread(void *data)
Jens Axboe03ba3782009-09-09 09:08:54 +02001018{
Christoph Hellwig08243902010-06-19 23:08:22 +02001019 struct bdi_writeback *wb = data;
1020 struct backing_dev_info *bdi = wb->bdi;
Jens Axboe03ba3782009-09-09 09:08:54 +02001021 long pages_written;
1022
Peter Zijlstra766f9162010-10-26 14:22:45 -07001023 current->flags |= PF_SWAPWRITE;
Christoph Hellwig08243902010-06-19 23:08:22 +02001024 set_freezable();
Artem Bityutskiyecd58402010-07-25 14:29:18 +03001025 wb->last_active = jiffies;
Christoph Hellwig08243902010-06-19 23:08:22 +02001026
1027 /*
1028 * Our parent may run at a different priority, just set us to normal
1029 */
1030 set_user_nice(current, 0);
1031
Dave Chinner455b2862010-07-07 13:24:06 +10001032 trace_writeback_thread_start(bdi);
1033
Tejun Heo8a32c442011-11-21 12:32:23 -08001034 while (!kthread_freezable_should_stop(NULL)) {
Artem Bityutskiy64677162010-07-25 14:29:22 +03001035 /*
1036 * Remove own delayed wake-up timer, since we are already awake
Yan Hong5aaea512012-12-12 13:52:14 -08001037 * and we'll take care of the periodic write-back.
Artem Bityutskiy64677162010-07-25 14:29:22 +03001038 */
1039 del_timer(&wb->wakeup_timer);
1040
Jens Axboe03ba3782009-09-09 09:08:54 +02001041 pages_written = wb_do_writeback(wb, 0);
1042
Dave Chinner455b2862010-07-07 13:24:06 +10001043 trace_writeback_pages_written(pages_written);
Jens Axboe03ba3782009-09-09 09:08:54 +02001044
Jens Axboe03ba3782009-09-09 09:08:54 +02001045 if (pages_written)
Artem Bityutskiyecd58402010-07-25 14:29:18 +03001046 wb->last_active = jiffies;
Jens Axboe03ba3782009-09-09 09:08:54 +02001047
Artem Bityutskiy297252c2010-07-25 14:29:15 +03001048 set_current_state(TASK_INTERRUPTIBLE);
J. Bruce Fieldsb76b4012010-08-28 08:52:10 +02001049 if (!list_empty(&bdi->work_list) || kthread_should_stop()) {
Artem Bityutskiy297252c2010-07-25 14:29:15 +03001050 __set_current_state(TASK_RUNNING);
1051 continue;
Jens Axboe03ba3782009-09-09 09:08:54 +02001052 }
1053
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001054 if (wb_has_dirty_io(wb) && dirty_writeback_interval)
Artem Bityutskiyfff5b852010-07-25 14:29:20 +03001055 schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10));
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001056 else {
1057 /*
1058 * We have nothing to do, so can go sleep without any
1059 * timeout and save power. When a work is queued or
1060 * something is made dirty - we will be woken up.
1061 */
Artem Bityutskiy297252c2010-07-25 14:29:15 +03001062 schedule();
Jens Axboef9eadbb2010-05-18 14:31:45 +02001063 }
Jens Axboe03ba3782009-09-09 09:08:54 +02001064 }
1065
Artem Bityutskiyfff5b852010-07-25 14:29:20 +03001066 /* Flush any work that raced with us exiting */
Christoph Hellwig08243902010-06-19 23:08:22 +02001067 if (!list_empty(&bdi->work_list))
1068 wb_do_writeback(wb, 1);
Dave Chinner455b2862010-07-07 13:24:06 +10001069
1070 trace_writeback_thread_stop(bdi);
Jens Axboe03ba3782009-09-09 09:08:54 +02001071 return 0;
1072}
1073
Christoph Hellwig08243902010-06-19 23:08:22 +02001074
Jens Axboe03ba3782009-09-09 09:08:54 +02001075/*
Jens Axboe03ba3782009-09-09 09:08:54 +02001076 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
1077 * the whole world.
1078 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001079void wakeup_flusher_threads(long nr_pages, enum wb_reason reason)
Jens Axboe03ba3782009-09-09 09:08:54 +02001080{
Christoph Hellwigb8c2f342010-06-08 18:15:07 +02001081 struct backing_dev_info *bdi;
Christoph Hellwigb8c2f342010-06-08 18:15:07 +02001082
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001083 if (!nr_pages) {
1084 nr_pages = global_page_state(NR_FILE_DIRTY) +
Jens Axboe03ba3782009-09-09 09:08:54 +02001085 global_page_state(NR_UNSTABLE_NFS);
Christoph Hellwigb8c2f342010-06-08 18:15:07 +02001086 }
1087
1088 rcu_read_lock();
1089 list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) {
1090 if (!bdi_has_dirty_io(bdi))
1091 continue;
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001092 __bdi_start_writeback(bdi, nr_pages, false, reason);
Christoph Hellwigb8c2f342010-06-08 18:15:07 +02001093 }
1094 rcu_read_unlock();
Jens Axboe03ba3782009-09-09 09:08:54 +02001095}
1096
1097static noinline void block_dump___mark_inode_dirty(struct inode *inode)
1098{
1099 if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) {
1100 struct dentry *dentry;
1101 const char *name = "?";
1102
1103 dentry = d_find_alias(inode);
1104 if (dentry) {
1105 spin_lock(&dentry->d_lock);
1106 name = (const char *) dentry->d_name.name;
1107 }
1108 printk(KERN_DEBUG
1109 "%s(%d): dirtied inode %lu (%s) on %s\n",
1110 current->comm, task_pid_nr(current), inode->i_ino,
1111 name, inode->i_sb->s_id);
1112 if (dentry) {
1113 spin_unlock(&dentry->d_lock);
1114 dput(dentry);
1115 }
1116 }
1117}
1118
1119/**
1120 * __mark_inode_dirty - internal function
1121 * @inode: inode to mark
1122 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
1123 * Mark an inode as dirty. Callers should use mark_inode_dirty or
1124 * mark_inode_dirty_sync.
1125 *
1126 * Put the inode on the super block's dirty list.
1127 *
1128 * CAREFUL! We mark it dirty unconditionally, but move it onto the
1129 * dirty list only if it is hashed or if it refers to a blockdev.
1130 * If it was not hashed, it will never be added to the dirty list
1131 * even if it is later hashed, as it will have been marked dirty already.
1132 *
1133 * In short, make sure you hash any inodes _before_ you start marking
1134 * them dirty.
1135 *
Jens Axboe03ba3782009-09-09 09:08:54 +02001136 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
1137 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
1138 * the kernel-internal blockdev inode represents the dirtying time of the
1139 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
1140 * page->mapping->host, so the page-dirtying time is recorded in the internal
1141 * blockdev inode.
1142 */
1143void __mark_inode_dirty(struct inode *inode, int flags)
1144{
1145 struct super_block *sb = inode->i_sb;
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001146 struct backing_dev_info *bdi = NULL;
Jens Axboe03ba3782009-09-09 09:08:54 +02001147
1148 /*
1149 * Don't do this for I_DIRTY_PAGES - that doesn't actually
1150 * dirty the inode itself
1151 */
1152 if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
1153 if (sb->s_op->dirty_inode)
Christoph Hellwigaa385722011-05-27 06:53:02 -04001154 sb->s_op->dirty_inode(inode, flags);
Jens Axboe03ba3782009-09-09 09:08:54 +02001155 }
1156
1157 /*
1158 * make sure that changes are seen by all cpus before we test i_state
1159 * -- mikulas
1160 */
1161 smp_mb();
1162
1163 /* avoid the locking if we can */
1164 if ((inode->i_state & flags) == flags)
1165 return;
1166
1167 if (unlikely(block_dump))
1168 block_dump___mark_inode_dirty(inode);
1169
Dave Chinner250df6e2011-03-22 22:23:36 +11001170 spin_lock(&inode->i_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001171 if ((inode->i_state & flags) != flags) {
1172 const int was_dirty = inode->i_state & I_DIRTY;
1173
1174 inode->i_state |= flags;
1175
1176 /*
1177 * If the inode is being synced, just update its dirty state.
1178 * The unlocker will place the inode on the appropriate
1179 * superblock list, based upon its state.
1180 */
1181 if (inode->i_state & I_SYNC)
Dave Chinner250df6e2011-03-22 22:23:36 +11001182 goto out_unlock_inode;
Jens Axboe03ba3782009-09-09 09:08:54 +02001183
1184 /*
1185 * Only add valid (hashed) inodes to the superblock's
1186 * dirty list. Add blockdev inodes as well.
1187 */
1188 if (!S_ISBLK(inode->i_mode)) {
Al Viro1d3382cb2010-10-23 15:19:20 -04001189 if (inode_unhashed(inode))
Dave Chinner250df6e2011-03-22 22:23:36 +11001190 goto out_unlock_inode;
Jens Axboe03ba3782009-09-09 09:08:54 +02001191 }
Al Viroa4ffdde2010-06-02 17:38:30 -04001192 if (inode->i_state & I_FREEING)
Dave Chinner250df6e2011-03-22 22:23:36 +11001193 goto out_unlock_inode;
Jens Axboe03ba3782009-09-09 09:08:54 +02001194
1195 /*
1196 * If the inode was already on b_dirty/b_io/b_more_io, don't
1197 * reposition it (that would break b_dirty time-ordering).
1198 */
1199 if (!was_dirty) {
Dave Chinnera66979a2011-03-22 22:23:41 +11001200 bool wakeup_bdi = false;
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001201 bdi = inode_to_bdi(inode);
Jens Axboe500b0672009-09-09 09:10:25 +02001202
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001203 if (bdi_cap_writeback_dirty(bdi)) {
1204 WARN(!test_bit(BDI_registered, &bdi->state),
1205 "bdi-%s not registered\n", bdi->name);
1206
1207 /*
1208 * If this is the first dirty inode for this
1209 * bdi, we have to wake-up the corresponding
1210 * bdi thread to make sure background
1211 * write-back happens later.
1212 */
1213 if (!wb_has_dirty_io(&bdi->wb))
1214 wakeup_bdi = true;
Jens Axboe500b0672009-09-09 09:10:25 +02001215 }
Jens Axboe03ba3782009-09-09 09:08:54 +02001216
Dave Chinnera66979a2011-03-22 22:23:41 +11001217 spin_unlock(&inode->i_lock);
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001218 spin_lock(&bdi->wb.list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001219 inode->dirtied_when = jiffies;
Nick Piggin7ccf19a2010-10-21 11:49:30 +11001220 list_move(&inode->i_wb_list, &bdi->wb.b_dirty);
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001221 spin_unlock(&bdi->wb.list_lock);
Dave Chinnera66979a2011-03-22 22:23:41 +11001222
1223 if (wakeup_bdi)
1224 bdi_wakeup_thread_delayed(bdi);
1225 return;
Jens Axboe03ba3782009-09-09 09:08:54 +02001226 }
1227 }
Dave Chinner250df6e2011-03-22 22:23:36 +11001228out_unlock_inode:
1229 spin_unlock(&inode->i_lock);
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001230
Jens Axboe03ba3782009-09-09 09:08:54 +02001231}
1232EXPORT_SYMBOL(__mark_inode_dirty);
1233
Jens Axboeb6e51312009-09-16 15:13:54 +02001234static void wait_sb_inodes(struct super_block *sb)
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001235{
Jens Axboe03ba3782009-09-09 09:08:54 +02001236 struct inode *inode, *old_inode = NULL;
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001237
Jens Axboe03ba3782009-09-09 09:08:54 +02001238 /*
1239 * We need to be protected against the filesystem going from
1240 * r/o to r/w or vice versa.
1241 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001242 WARN_ON(!rwsem_is_locked(&sb->s_umount));
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001243
Dave Chinner55fa6092011-03-22 22:23:40 +11001244 spin_lock(&inode_sb_list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001245
1246 /*
1247 * Data integrity sync. Must wait for all pages under writeback,
1248 * because there may have been pages dirtied before our sync
1249 * call, but which had writeout started before we write it out.
1250 * In which case, the inode may not be on the dirty list, but
1251 * we still have to wait for that writeout.
1252 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001253 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
Dave Chinner250df6e2011-03-22 22:23:36 +11001254 struct address_space *mapping = inode->i_mapping;
Jens Axboe03ba3782009-09-09 09:08:54 +02001255
Dave Chinner250df6e2011-03-22 22:23:36 +11001256 spin_lock(&inode->i_lock);
1257 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
1258 (mapping->nrpages == 0)) {
1259 spin_unlock(&inode->i_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001260 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +11001261 }
Jens Axboe03ba3782009-09-09 09:08:54 +02001262 __iget(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +11001263 spin_unlock(&inode->i_lock);
Dave Chinner55fa6092011-03-22 22:23:40 +11001264 spin_unlock(&inode_sb_list_lock);
1265
Jens Axboe03ba3782009-09-09 09:08:54 +02001266 /*
Dave Chinner55fa6092011-03-22 22:23:40 +11001267 * We hold a reference to 'inode' so it couldn't have been
1268 * removed from s_inodes list while we dropped the
1269 * inode_sb_list_lock. We cannot iput the inode now as we can
1270 * be holding the last reference and we cannot iput it under
1271 * inode_sb_list_lock. So we keep the reference and iput it
1272 * later.
Jens Axboe03ba3782009-09-09 09:08:54 +02001273 */
1274 iput(old_inode);
1275 old_inode = inode;
1276
1277 filemap_fdatawait(mapping);
1278
1279 cond_resched();
Nick Piggin38f21972009-01-06 14:40:25 -08001280
Dave Chinner55fa6092011-03-22 22:23:40 +11001281 spin_lock(&inode_sb_list_lock);
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001282 }
Dave Chinner55fa6092011-03-22 22:23:40 +11001283 spin_unlock(&inode_sb_list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001284 iput(old_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285}
1286
Jens Axboed8a85592009-09-02 12:34:32 +02001287/**
Chris Mason3259f8b2010-10-29 11:16:17 -04001288 * writeback_inodes_sb_nr - writeback dirty inodes from given super_block
Jens Axboed8a85592009-09-02 12:34:32 +02001289 * @sb: the superblock
Chris Mason3259f8b2010-10-29 11:16:17 -04001290 * @nr: the number of pages to write
Marcos Paulo de Souza786228a2011-11-23 20:56:45 +08001291 * @reason: reason why some writeback work initiated
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292 *
Jens Axboed8a85592009-09-02 12:34:32 +02001293 * Start writeback on some inodes on this super_block. No guarantees are made
1294 * on how many (if any) will be written, and this function does not wait
Chris Mason3259f8b2010-10-29 11:16:17 -04001295 * for IO completion of submitted IO.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001297void writeback_inodes_sb_nr(struct super_block *sb,
1298 unsigned long nr,
1299 enum wb_reason reason)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001301 DECLARE_COMPLETION_ONSTACK(done);
1302 struct wb_writeback_work work = {
Wu Fengguang6e6938b2010-06-06 10:38:15 -06001303 .sb = sb,
1304 .sync_mode = WB_SYNC_NONE,
1305 .tagged_writepages = 1,
1306 .done = &done,
1307 .nr_pages = nr,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001308 .reason = reason,
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001309 };
Jens Axboe0e3c9a22010-06-01 11:08:43 +02001310
Jan Kara6eedc702012-07-03 16:45:27 +02001311 if (sb->s_bdi == &noop_backing_dev_info)
1312 return;
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001313 WARN_ON(!rwsem_is_locked(&sb->s_umount));
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001314 bdi_queue_work(sb->s_bdi, &work);
1315 wait_for_completion(&done);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316}
Chris Mason3259f8b2010-10-29 11:16:17 -04001317EXPORT_SYMBOL(writeback_inodes_sb_nr);
1318
1319/**
1320 * writeback_inodes_sb - writeback dirty inodes from given super_block
1321 * @sb: the superblock
Marcos Paulo de Souza786228a2011-11-23 20:56:45 +08001322 * @reason: reason why some writeback work was initiated
Chris Mason3259f8b2010-10-29 11:16:17 -04001323 *
1324 * Start writeback on some inodes on this super_block. No guarantees are made
1325 * on how many (if any) will be written, and this function does not wait
1326 * for IO completion of submitted IO.
1327 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001328void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason)
Chris Mason3259f8b2010-10-29 11:16:17 -04001329{
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001330 return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason);
Chris Mason3259f8b2010-10-29 11:16:17 -04001331}
Jens Axboed8a85592009-09-02 12:34:32 +02001332EXPORT_SYMBOL(writeback_inodes_sb);
1333
1334/**
Eric Sandeen17bd55d2009-12-23 07:57:07 -05001335 * writeback_inodes_sb_if_idle - start writeback if none underway
1336 * @sb: the superblock
Marcos Paulo de Souza786228a2011-11-23 20:56:45 +08001337 * @reason: reason why some writeback work was initiated
Eric Sandeen17bd55d2009-12-23 07:57:07 -05001338 *
1339 * Invoke writeback_inodes_sb if no writeback is currently underway.
1340 * Returns 1 if writeback was started, 0 if not.
1341 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001342int writeback_inodes_sb_if_idle(struct super_block *sb, enum wb_reason reason)
Eric Sandeen17bd55d2009-12-23 07:57:07 -05001343{
1344 if (!writeback_in_progress(sb->s_bdi)) {
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001345 down_read(&sb->s_umount);
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001346 writeback_inodes_sb(sb, reason);
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001347 up_read(&sb->s_umount);
Eric Sandeen17bd55d2009-12-23 07:57:07 -05001348 return 1;
1349 } else
1350 return 0;
1351}
1352EXPORT_SYMBOL(writeback_inodes_sb_if_idle);
1353
1354/**
Fengguang Wuc097b2c2012-03-05 15:08:06 -08001355 * writeback_inodes_sb_nr_if_idle - start writeback if none underway
Chris Mason3259f8b2010-10-29 11:16:17 -04001356 * @sb: the superblock
1357 * @nr: the number of pages to write
Marcos Paulo de Souza786228a2011-11-23 20:56:45 +08001358 * @reason: reason why some writeback work was initiated
Chris Mason3259f8b2010-10-29 11:16:17 -04001359 *
1360 * Invoke writeback_inodes_sb if no writeback is currently underway.
1361 * Returns 1 if writeback was started, 0 if not.
1362 */
1363int writeback_inodes_sb_nr_if_idle(struct super_block *sb,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001364 unsigned long nr,
1365 enum wb_reason reason)
Chris Mason3259f8b2010-10-29 11:16:17 -04001366{
1367 if (!writeback_in_progress(sb->s_bdi)) {
1368 down_read(&sb->s_umount);
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001369 writeback_inodes_sb_nr(sb, nr, reason);
Chris Mason3259f8b2010-10-29 11:16:17 -04001370 up_read(&sb->s_umount);
1371 return 1;
1372 } else
1373 return 0;
1374}
1375EXPORT_SYMBOL(writeback_inodes_sb_nr_if_idle);
1376
1377/**
Jens Axboed8a85592009-09-02 12:34:32 +02001378 * sync_inodes_sb - sync sb inode pages
1379 * @sb: the superblock
1380 *
1381 * This function writes and waits on any dirty inode belonging to this
Stefan Hajnoczicb9ef8d2011-01-13 15:47:26 -08001382 * super_block.
Jens Axboed8a85592009-09-02 12:34:32 +02001383 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001384void sync_inodes_sb(struct super_block *sb)
Jens Axboed8a85592009-09-02 12:34:32 +02001385{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001386 DECLARE_COMPLETION_ONSTACK(done);
1387 struct wb_writeback_work work = {
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001388 .sb = sb,
1389 .sync_mode = WB_SYNC_ALL,
1390 .nr_pages = LONG_MAX,
1391 .range_cyclic = 0,
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001392 .done = &done,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001393 .reason = WB_REASON_SYNC,
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001394 };
1395
Jan Kara6eedc702012-07-03 16:45:27 +02001396 /* Nothing to do? */
1397 if (sb->s_bdi == &noop_backing_dev_info)
1398 return;
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001399 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1400
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001401 bdi_queue_work(sb->s_bdi, &work);
1402 wait_for_completion(&done);
1403
Jens Axboeb6e51312009-09-16 15:13:54 +02001404 wait_sb_inodes(sb);
Jens Axboed8a85592009-09-02 12:34:32 +02001405}
1406EXPORT_SYMBOL(sync_inodes_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408/**
Andrea Arcangeli7f04c262005-10-30 15:03:05 -08001409 * write_inode_now - write an inode to disk
1410 * @inode: inode to write to disk
1411 * @sync: whether the write should be synchronous or not
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412 *
Andrea Arcangeli7f04c262005-10-30 15:03:05 -08001413 * This function commits an inode to disk immediately if it is dirty. This is
1414 * primarily needed by knfsd.
1415 *
1416 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418int write_inode_now(struct inode *inode, int sync)
1419{
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001420 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421 struct writeback_control wbc = {
1422 .nr_to_write = LONG_MAX,
Mike Galbraith18914b12008-02-08 04:20:23 -08001423 .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -07001424 .range_start = 0,
1425 .range_end = LLONG_MAX,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426 };
1427
1428 if (!mapping_cap_writeback_dirty(inode->i_mapping))
Andrew Morton49364ce2005-11-07 00:59:15 -08001429 wbc.nr_to_write = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430
1431 might_sleep();
Jan Kara4f8ad652012-05-03 14:48:00 +02001432 return writeback_single_inode(inode, wb, &wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433}
1434EXPORT_SYMBOL(write_inode_now);
1435
1436/**
1437 * sync_inode - write an inode and its pages to disk.
1438 * @inode: the inode to sync
1439 * @wbc: controls the writeback mode
1440 *
1441 * sync_inode() will write an inode and its pages to disk. It will also
1442 * correctly update the inode on its superblock's dirty inode lists and will
1443 * update inode->i_state.
1444 *
1445 * The caller must have a ref on the inode.
1446 */
1447int sync_inode(struct inode *inode, struct writeback_control *wbc)
1448{
Jan Kara4f8ad652012-05-03 14:48:00 +02001449 return writeback_single_inode(inode, &inode_to_bdi(inode)->wb, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450}
1451EXPORT_SYMBOL(sync_inode);
Christoph Hellwigc37650162010-10-06 10:48:20 +02001452
1453/**
Andrew Mortonc691b9d2011-01-13 15:45:48 -08001454 * sync_inode_metadata - write an inode to disk
Christoph Hellwigc37650162010-10-06 10:48:20 +02001455 * @inode: the inode to sync
1456 * @wait: wait for I/O to complete.
1457 *
Andrew Mortonc691b9d2011-01-13 15:45:48 -08001458 * Write an inode to disk and adjust its dirty state after completion.
Christoph Hellwigc37650162010-10-06 10:48:20 +02001459 *
1460 * Note: only writes the actual inode, no associated data or other metadata.
1461 */
1462int sync_inode_metadata(struct inode *inode, int wait)
1463{
1464 struct writeback_control wbc = {
1465 .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE,
1466 .nr_to_write = 0, /* metadata-only */
1467 };
1468
1469 return sync_inode(inode, &wbc);
1470}
1471EXPORT_SYMBOL(sync_inode_metadata);