blob: 51ea267d444c4d7aef1259521aa1f06154f26d13 [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;
231 /* Waiters must see I_SYNC cleared before being woken up */
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700232 smp_mb();
233 wake_up_bit(&inode->i_state, __I_SYNC);
234}
235
Jeff Laytond2caa3c52009-04-02 16:56:37 -0700236static bool inode_dirtied_after(struct inode *inode, unsigned long t)
237{
238 bool ret = time_after(inode->dirtied_when, t);
239#ifndef CONFIG_64BIT
240 /*
241 * For inodes being constantly redirtied, dirtied_when can get stuck.
242 * It _appears_ to be in the future, but is actually in distant past.
243 * This test is necessary to prevent such wrapped-around relative times
Jens Axboe5b0830c2009-09-23 19:37:09 +0200244 * from permanently stopping the whole bdi writeback.
Jeff Laytond2caa3c52009-04-02 16:56:37 -0700245 */
246 ret = ret && time_before_eq(inode->dirtied_when, jiffies);
247#endif
248 return ret;
249}
250
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700251/*
Wang Sheng-Hui0e2f2b22012-09-11 08:28:18 +0800252 * Move expired (dirtied before work->older_than_this) dirty inodes from
Jan Kara697e6fe2012-03-09 07:26:22 -0800253 * @delaying_queue to @dispatch_queue.
Fengguang Wu2c136572007-10-16 23:30:39 -0700254 */
Wu Fengguange84d0a42011-04-23 12:27:27 -0600255static int move_expired_inodes(struct list_head *delaying_queue,
Fengguang Wu2c136572007-10-16 23:30:39 -0700256 struct list_head *dispatch_queue,
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600257 struct wb_writeback_work *work)
Fengguang Wu2c136572007-10-16 23:30:39 -0700258{
Shaohua Li5c034492009-09-24 14:42:33 +0200259 LIST_HEAD(tmp);
260 struct list_head *pos, *node;
Jens Axboecf137302009-09-24 15:12:57 +0200261 struct super_block *sb = NULL;
Shaohua Li5c034492009-09-24 14:42:33 +0200262 struct inode *inode;
Jens Axboecf137302009-09-24 15:12:57 +0200263 int do_sb_sort = 0;
Wu Fengguange84d0a42011-04-23 12:27:27 -0600264 int moved = 0;
Shaohua Li5c034492009-09-24 14:42:33 +0200265
Fengguang Wu2c136572007-10-16 23:30:39 -0700266 while (!list_empty(delaying_queue)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100267 inode = wb_inode(delaying_queue->prev);
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600268 if (work->older_than_this &&
269 inode_dirtied_after(inode, *work->older_than_this))
Fengguang Wu2c136572007-10-16 23:30:39 -0700270 break;
Jens Axboecf137302009-09-24 15:12:57 +0200271 if (sb && sb != inode->i_sb)
272 do_sb_sort = 1;
273 sb = inode->i_sb;
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100274 list_move(&inode->i_wb_list, &tmp);
Wu Fengguange84d0a42011-04-23 12:27:27 -0600275 moved++;
Shaohua Li5c034492009-09-24 14:42:33 +0200276 }
277
Jens Axboecf137302009-09-24 15:12:57 +0200278 /* just one sb in list, splice to dispatch_queue and we're done */
279 if (!do_sb_sort) {
280 list_splice(&tmp, dispatch_queue);
Wu Fengguange84d0a42011-04-23 12:27:27 -0600281 goto out;
Jens Axboecf137302009-09-24 15:12:57 +0200282 }
283
Shaohua Li5c034492009-09-24 14:42:33 +0200284 /* Move inodes from one superblock together */
285 while (!list_empty(&tmp)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100286 sb = wb_inode(tmp.prev)->i_sb;
Shaohua Li5c034492009-09-24 14:42:33 +0200287 list_for_each_prev_safe(pos, node, &tmp) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100288 inode = wb_inode(pos);
Shaohua Li5c034492009-09-24 14:42:33 +0200289 if (inode->i_sb == sb)
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100290 list_move(&inode->i_wb_list, dispatch_queue);
Shaohua Li5c034492009-09-24 14:42:33 +0200291 }
Fengguang Wu2c136572007-10-16 23:30:39 -0700292 }
Wu Fengguange84d0a42011-04-23 12:27:27 -0600293out:
294 return moved;
Fengguang Wu2c136572007-10-16 23:30:39 -0700295}
296
297/*
298 * Queue all expired dirty inodes for io, eldest first.
Wu Fengguang4ea879b2010-08-11 14:17:42 -0700299 * Before
300 * newly dirtied b_dirty b_io b_more_io
301 * =============> gf edc BA
302 * After
303 * newly dirtied b_dirty b_io b_more_io
304 * =============> g fBAedc
305 * |
306 * +--> dequeue for IO
Fengguang Wu2c136572007-10-16 23:30:39 -0700307 */
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600308static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work)
Fengguang Wu2c136572007-10-16 23:30:39 -0700309{
Wu Fengguange84d0a42011-04-23 12:27:27 -0600310 int moved;
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600311 assert_spin_locked(&wb->list_lock);
Wu Fengguang4ea879b2010-08-11 14:17:42 -0700312 list_splice_init(&wb->b_more_io, &wb->b_io);
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600313 moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, work);
314 trace_writeback_queue_io(wb, work, moved);
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200315}
316
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100317static int write_inode(struct inode *inode, struct writeback_control *wbc)
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200318{
Jens Axboe03ba3782009-09-09 09:08:54 +0200319 if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode))
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100320 return inode->i_sb->s_op->write_inode(inode, wbc);
Jens Axboe03ba3782009-09-09 09:08:54 +0200321 return 0;
Fengguang Wu2c136572007-10-16 23:30:39 -0700322}
323
324/*
Jan Kara169ebd92012-05-03 14:48:03 +0200325 * Wait for writeback on an inode to complete. Called with i_lock held.
326 * Caller must make sure inode cannot go away when we drop i_lock.
Christoph Hellwig01c03192009-06-08 13:35:40 +0200327 */
Jan Kara169ebd92012-05-03 14:48:03 +0200328static void __inode_wait_for_writeback(struct inode *inode)
329 __releases(inode->i_lock)
330 __acquires(inode->i_lock)
Christoph Hellwig01c03192009-06-08 13:35:40 +0200331{
332 DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC);
333 wait_queue_head_t *wqh;
334
335 wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
Dave Chinner250df6e2011-03-22 22:23:36 +1100336 while (inode->i_state & I_SYNC) {
337 spin_unlock(&inode->i_lock);
Christoph Hellwig01c03192009-06-08 13:35:40 +0200338 __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE);
Dave Chinner250df6e2011-03-22 22:23:36 +1100339 spin_lock(&inode->i_lock);
Richard Kennedy58a9d3d82010-05-24 14:32:38 -0700340 }
Christoph Hellwig01c03192009-06-08 13:35:40 +0200341}
342
343/*
Jan Kara169ebd92012-05-03 14:48:03 +0200344 * Wait for writeback on an inode to complete. Caller must have inode pinned.
345 */
346void inode_wait_for_writeback(struct inode *inode)
347{
348 spin_lock(&inode->i_lock);
349 __inode_wait_for_writeback(inode);
350 spin_unlock(&inode->i_lock);
351}
352
353/*
354 * Sleep until I_SYNC is cleared. This function must be called with i_lock
355 * held and drops it. It is aimed for callers not holding any inode reference
356 * so once i_lock is dropped, inode can go away.
357 */
358static void inode_sleep_on_writeback(struct inode *inode)
359 __releases(inode->i_lock)
360{
361 DEFINE_WAIT(wait);
362 wait_queue_head_t *wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
363 int sleep;
364
365 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
366 sleep = inode->i_state & I_SYNC;
367 spin_unlock(&inode->i_lock);
368 if (sleep)
369 schedule();
370 finish_wait(wqh, &wait);
371}
372
373/*
Jan Karaccb26b52012-05-03 14:47:58 +0200374 * Find proper writeback list for the inode depending on its current state and
375 * possibly also change of its state while we were doing writeback. Here we
376 * handle things such as livelock prevention or fairness of writeback among
377 * inodes. This function can be called only by flusher thread - noone else
378 * processes all inodes in writeback lists and requeueing inodes behind flusher
379 * thread's back can have unexpected consequences.
380 */
381static void requeue_inode(struct inode *inode, struct bdi_writeback *wb,
382 struct writeback_control *wbc)
383{
384 if (inode->i_state & I_FREEING)
385 return;
386
387 /*
388 * Sync livelock prevention. Each inode is tagged and synced in one
389 * shot. If still dirty, it will be redirty_tail()'ed below. Update
390 * the dirty time to prevent enqueue and sync it again.
391 */
392 if ((inode->i_state & I_DIRTY) &&
393 (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages))
394 inode->dirtied_when = jiffies;
395
Jan Kara4f8ad652012-05-03 14:48:00 +0200396 if (wbc->pages_skipped) {
397 /*
398 * writeback is not making progress due to locked
399 * buffers. Skip this inode for now.
400 */
401 redirty_tail(inode, wb);
402 return;
403 }
404
Jan Karaccb26b52012-05-03 14:47:58 +0200405 if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) {
406 /*
407 * We didn't write back all the pages. nfs_writepages()
408 * sometimes bales out without doing anything.
409 */
410 if (wbc->nr_to_write <= 0) {
411 /* Slice used up. Queue for next turn. */
412 requeue_io(inode, wb);
413 } else {
414 /*
415 * Writeback blocked by something other than
416 * congestion. Delay the inode for some time to
417 * avoid spinning on the CPU (100% iowait)
418 * retrying writeback of the dirty page/inode
419 * that cannot be performed immediately.
420 */
421 redirty_tail(inode, wb);
422 }
423 } else if (inode->i_state & I_DIRTY) {
424 /*
425 * Filesystems can dirty the inode during writeback operations,
426 * such as delayed allocation during submission or metadata
427 * updates after data IO completion.
428 */
429 redirty_tail(inode, wb);
430 } else {
431 /* The inode is clean. Remove from writeback lists. */
432 list_del_init(&inode->i_wb_list);
433 }
434}
435
436/*
Jan Kara4f8ad652012-05-03 14:48:00 +0200437 * Write out an inode and its dirty pages. Do not update the writeback list
438 * linkage. That is left to the caller. The caller is also responsible for
439 * setting I_SYNC flag and calling inode_sync_complete() to clear it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440 */
441static int
Yan Hongcd8ed2a2012-10-08 16:33:45 -0700442__writeback_single_inode(struct inode *inode, struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444 struct address_space *mapping = inode->i_mapping;
Wu Fengguang251d6a42010-12-01 17:33:37 -0600445 long nr_to_write = wbc->nr_to_write;
Christoph Hellwig01c03192009-06-08 13:35:40 +0200446 unsigned dirty;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447 int ret;
448
Jan Kara4f8ad652012-05-03 14:48:00 +0200449 WARN_ON(!(inode->i_state & I_SYNC));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450
451 ret = do_writepages(mapping, wbc);
452
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100453 /*
454 * Make sure to wait on the data before writing out the metadata.
455 * This is important for filesystems that modify metadata on data
456 * I/O completion.
457 */
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100458 if (wbc->sync_mode == WB_SYNC_ALL) {
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100459 int err = filemap_fdatawait(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460 if (ret == 0)
461 ret = err;
462 }
463
Dmitry Monakhov5547e8a2010-05-07 13:35:44 +0400464 /*
465 * Some filesystems may redirty the inode during the writeback
466 * due to delalloc, clear dirty metadata flags right before
467 * write_inode()
468 */
Dave Chinner250df6e2011-03-22 22:23:36 +1100469 spin_lock(&inode->i_lock);
Jan Kara6290be12012-05-03 14:47:57 +0200470 /* Clear I_DIRTY_PAGES if we've written out all dirty pages */
471 if (!mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
472 inode->i_state &= ~I_DIRTY_PAGES;
Dmitry Monakhov5547e8a2010-05-07 13:35:44 +0400473 dirty = inode->i_state & I_DIRTY;
474 inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC);
Dave Chinner250df6e2011-03-22 22:23:36 +1100475 spin_unlock(&inode->i_lock);
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100476 /* Don't write the inode if only I_DIRTY_PAGES was set */
477 if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100478 int err = write_inode(inode, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479 if (ret == 0)
480 ret = err;
481 }
Jan Kara4f8ad652012-05-03 14:48:00 +0200482 trace_writeback_single_inode(inode, wbc, nr_to_write);
483 return ret;
484}
485
486/*
487 * Write out an inode's dirty pages. Either the caller has an active reference
488 * on the inode or the inode has I_WILL_FREE set.
489 *
490 * This function is designed to be called for writing back one inode which
491 * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode()
492 * and does more profound writeback list handling in writeback_sb_inodes().
493 */
494static int
495writeback_single_inode(struct inode *inode, struct bdi_writeback *wb,
496 struct writeback_control *wbc)
497{
498 int ret = 0;
499
500 spin_lock(&inode->i_lock);
501 if (!atomic_read(&inode->i_count))
502 WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING)));
503 else
504 WARN_ON(inode->i_state & I_WILL_FREE);
505
506 if (inode->i_state & I_SYNC) {
507 if (wbc->sync_mode != WB_SYNC_ALL)
508 goto out;
509 /*
Jan Kara169ebd92012-05-03 14:48:03 +0200510 * It's a data-integrity sync. We must wait. Since callers hold
511 * inode reference or inode has I_WILL_FREE set, it cannot go
512 * away under us.
Jan Kara4f8ad652012-05-03 14:48:00 +0200513 */
Jan Kara169ebd92012-05-03 14:48:03 +0200514 __inode_wait_for_writeback(inode);
Jan Kara4f8ad652012-05-03 14:48:00 +0200515 }
516 WARN_ON(inode->i_state & I_SYNC);
517 /*
518 * Skip inode if it is clean. We don't want to mess with writeback
519 * lists in this function since flusher thread may be doing for example
520 * sync in parallel and if we move the inode, it could get skipped. So
521 * here we make sure inode is on some writeback list and leave it there
522 * unless we have completely cleaned the inode.
523 */
524 if (!(inode->i_state & I_DIRTY))
525 goto out;
526 inode->i_state |= I_SYNC;
527 spin_unlock(&inode->i_lock);
528
Yan Hongcd8ed2a2012-10-08 16:33:45 -0700529 ret = __writeback_single_inode(inode, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600531 spin_lock(&wb->list_lock);
Dave Chinner250df6e2011-03-22 22:23:36 +1100532 spin_lock(&inode->i_lock);
Jan Kara4f8ad652012-05-03 14:48:00 +0200533 /*
534 * If inode is clean, remove it from writeback lists. Otherwise don't
535 * touch it. See comment above for explanation.
536 */
537 if (!(inode->i_state & I_DIRTY))
538 list_del_init(&inode->i_wb_list);
539 spin_unlock(&wb->list_lock);
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700540 inode_sync_complete(inode);
Jan Kara4f8ad652012-05-03 14:48:00 +0200541out:
542 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543 return ret;
544}
545
Wu Fengguang1a12d8b2010-08-29 13:28:09 -0600546static long writeback_chunk_size(struct backing_dev_info *bdi,
547 struct wb_writeback_work *work)
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600548{
549 long pages;
550
551 /*
552 * WB_SYNC_ALL mode does livelock avoidance by syncing dirty
553 * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX
554 * here avoids calling into writeback_inodes_wb() more than once.
555 *
556 * The intended call sequence for WB_SYNC_ALL writeback is:
557 *
558 * wb_writeback()
559 * writeback_sb_inodes() <== called only once
560 * write_cache_pages() <== called once for each inode
561 * (quickly) tag currently dirty pages
562 * (maybe slowly) sync all tagged pages
563 */
564 if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages)
565 pages = LONG_MAX;
Wu Fengguang1a12d8b2010-08-29 13:28:09 -0600566 else {
567 pages = min(bdi->avg_write_bandwidth / 2,
568 global_dirty_limit / DIRTY_SCOPE);
569 pages = min(pages, work->nr_pages);
570 pages = round_down(pages + MIN_WRITEBACK_PAGES,
571 MIN_WRITEBACK_PAGES);
572 }
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600573
574 return pages;
575}
576
Jens Axboe03ba3782009-09-09 09:08:54 +0200577/*
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800578 * Write a portion of b_io inodes which belong to @sb.
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200579 *
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600580 * Return the number of pages and/or inodes written.
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800581 */
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600582static long writeback_sb_inodes(struct super_block *sb,
583 struct bdi_writeback *wb,
584 struct wb_writeback_work *work)
Jens Axboe03ba3782009-09-09 09:08:54 +0200585{
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600586 struct writeback_control wbc = {
587 .sync_mode = work->sync_mode,
588 .tagged_writepages = work->tagged_writepages,
589 .for_kupdate = work->for_kupdate,
590 .for_background = work->for_background,
591 .range_cyclic = work->range_cyclic,
592 .range_start = 0,
593 .range_end = LLONG_MAX,
594 };
595 unsigned long start_time = jiffies;
596 long write_chunk;
597 long wrote = 0; /* count both pages and inodes */
598
Jens Axboe03ba3782009-09-09 09:08:54 +0200599 while (!list_empty(&wb->b_io)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100600 struct inode *inode = wb_inode(wb->b_io.prev);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200601
602 if (inode->i_sb != sb) {
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600603 if (work->sb) {
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200604 /*
605 * We only want to write back data for this
606 * superblock, move all inodes not belonging
607 * to it back onto the dirty list.
608 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600609 redirty_tail(inode, wb);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200610 continue;
611 }
612
613 /*
614 * The inode belongs to a different superblock.
615 * Bounce back to the caller to unpin this and
616 * pin the next superblock.
617 */
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600618 break;
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200619 }
620
Christoph Hellwig9843b762010-10-24 19:40:46 +0200621 /*
Wanpeng Li331cbde2012-06-09 11:10:55 +0800622 * Don't bother with new inodes or inodes being freed, first
623 * kind does not need periodic writeout yet, and for the latter
Christoph Hellwig9843b762010-10-24 19:40:46 +0200624 * kind writeout is handled by the freer.
625 */
Dave Chinner250df6e2011-03-22 22:23:36 +1100626 spin_lock(&inode->i_lock);
Christoph Hellwig9843b762010-10-24 19:40:46 +0200627 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
Dave Chinner250df6e2011-03-22 22:23:36 +1100628 spin_unlock(&inode->i_lock);
Wu Fengguangfcc5c222011-07-11 23:08:50 -0700629 redirty_tail(inode, wb);
Nick Piggin7ef0d732009-03-12 14:31:38 -0700630 continue;
631 }
Jan Karacc1676d2012-05-03 14:47:56 +0200632 if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) {
633 /*
634 * If this inode is locked for writeback and we are not
635 * doing writeback-for-data-integrity, move it to
636 * b_more_io so that writeback can proceed with the
637 * other inodes on s_io.
638 *
639 * We'll have another go at writing back this inode
640 * when we completed a full scan of b_io.
641 */
642 spin_unlock(&inode->i_lock);
643 requeue_io(inode, wb);
644 trace_writeback_sb_inodes_requeue(inode);
645 continue;
646 }
Jan Karaf0d07b72012-05-03 14:47:59 +0200647 spin_unlock(&wb->list_lock);
648
Jan Kara4f8ad652012-05-03 14:48:00 +0200649 /*
650 * We already requeued the inode if it had I_SYNC set and we
651 * are doing WB_SYNC_NONE writeback. So this catches only the
652 * WB_SYNC_ALL case.
653 */
Jan Kara169ebd92012-05-03 14:48:03 +0200654 if (inode->i_state & I_SYNC) {
655 /* Wait for I_SYNC. This function drops i_lock... */
656 inode_sleep_on_writeback(inode);
657 /* Inode may be gone, start again */
Jan Karaead188f2012-06-08 17:07:36 +0200658 spin_lock(&wb->list_lock);
Jan Kara169ebd92012-05-03 14:48:03 +0200659 continue;
660 }
Jan Kara4f8ad652012-05-03 14:48:00 +0200661 inode->i_state |= I_SYNC;
662 spin_unlock(&inode->i_lock);
Jan Kara169ebd92012-05-03 14:48:03 +0200663
Wu Fengguang1a12d8b2010-08-29 13:28:09 -0600664 write_chunk = writeback_chunk_size(wb->bdi, work);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600665 wbc.nr_to_write = write_chunk;
666 wbc.pages_skipped = 0;
Dave Chinner250df6e2011-03-22 22:23:36 +1100667
Jan Kara169ebd92012-05-03 14:48:03 +0200668 /*
669 * We use I_SYNC to pin the inode in memory. While it is set
670 * evict_inode() will wait so the inode cannot be freed.
671 */
Yan Hongcd8ed2a2012-10-08 16:33:45 -0700672 __writeback_single_inode(inode, &wbc);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600673
674 work->nr_pages -= write_chunk - wbc.nr_to_write;
675 wrote += write_chunk - wbc.nr_to_write;
Jan Kara4f8ad652012-05-03 14:48:00 +0200676 spin_lock(&wb->list_lock);
677 spin_lock(&inode->i_lock);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600678 if (!(inode->i_state & I_DIRTY))
679 wrote++;
Jan Kara4f8ad652012-05-03 14:48:00 +0200680 requeue_inode(inode, wb, &wbc);
681 inode_sync_complete(inode);
Dave Chinner0f1b1fd2011-03-22 22:23:43 +1100682 spin_unlock(&inode->i_lock);
Jan Kara169ebd92012-05-03 14:48:03 +0200683 cond_resched_lock(&wb->list_lock);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600684 /*
685 * bail out to wb_writeback() often enough to check
686 * background threshold and other termination conditions.
687 */
688 if (wrote) {
689 if (time_is_before_jiffies(start_time + HZ / 10UL))
690 break;
691 if (work->nr_pages <= 0)
692 break;
Fengguang Wu8bc3be22008-02-04 22:29:36 -0800693 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694 }
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600695 return wrote;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800696}
Nick Piggin38f21972009-01-06 14:40:25 -0800697
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600698static long __writeback_inodes_wb(struct bdi_writeback *wb,
699 struct wb_writeback_work *work)
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800700{
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600701 unsigned long start_time = jiffies;
702 long wrote = 0;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800703
704 while (!list_empty(&wb->b_io)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100705 struct inode *inode = wb_inode(wb->b_io.prev);
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800706 struct super_block *sb = inode->i_sb;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800707
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000708 if (!grab_super_passive(sb)) {
Wu Fengguang0e9958162011-07-29 22:14:35 -0600709 /*
710 * grab_super_passive() may fail consistently due to
711 * s_umount being grabbed by someone else. Don't use
712 * requeue_io() to avoid busy retrying the inode/sb.
713 */
714 redirty_tail(inode, wb);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200715 continue;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800716 }
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600717 wrote += writeback_sb_inodes(sb, wb, work);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200718 drop_super(sb);
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800719
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600720 /* refer to the same tests at the end of writeback_sb_inodes */
721 if (wrote) {
722 if (time_is_before_jiffies(start_time + HZ / 10UL))
723 break;
724 if (work->nr_pages <= 0)
725 break;
726 }
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800727 }
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200728 /* Leave any unwritten inodes on b_io */
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600729 return wrote;
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200730}
731
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600732long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages,
733 enum wb_reason reason)
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200734{
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600735 struct wb_writeback_work work = {
736 .nr_pages = nr_pages,
737 .sync_mode = WB_SYNC_NONE,
738 .range_cyclic = 1,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600739 .reason = reason,
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600740 };
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200741
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600742 spin_lock(&wb->list_lock);
Wu Fengguang424b3512010-07-21 20:11:53 -0600743 if (list_empty(&wb->b_io))
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600744 queue_io(wb, &work);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600745 __writeback_inodes_wb(wb, &work);
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600746 spin_unlock(&wb->list_lock);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200747
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600748 return nr_pages - work.nr_pages;
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200749}
750
Wu Fengguangb00949a2010-11-18 14:38:33 -0600751static bool over_bground_thresh(struct backing_dev_info *bdi)
Jens Axboe03ba3782009-09-09 09:08:54 +0200752{
753 unsigned long background_thresh, dirty_thresh;
754
Wu Fengguang16c40422010-08-11 14:17:39 -0700755 global_dirty_limits(&background_thresh, &dirty_thresh);
Jens Axboe03ba3782009-09-09 09:08:54 +0200756
Wu Fengguangb00949a2010-11-18 14:38:33 -0600757 if (global_page_state(NR_FILE_DIRTY) +
758 global_page_state(NR_UNSTABLE_NFS) > background_thresh)
759 return true;
760
761 if (bdi_stat(bdi, BDI_RECLAIMABLE) >
762 bdi_dirty_limit(bdi, background_thresh))
763 return true;
764
765 return false;
Jens Axboe03ba3782009-09-09 09:08:54 +0200766}
767
768/*
Wu Fengguange98be2d2010-08-29 11:22:30 -0600769 * Called under wb->list_lock. If there are multiple wb per bdi,
770 * only the flusher working on the first wb should do it.
771 */
772static void wb_update_bandwidth(struct bdi_writeback *wb,
773 unsigned long start_time)
774{
Wu Fengguangaf6a3112011-10-03 20:46:17 -0600775 __bdi_update_bandwidth(wb->bdi, 0, 0, 0, 0, 0, start_time);
Wu Fengguange98be2d2010-08-29 11:22:30 -0600776}
777
778/*
Jens Axboe03ba3782009-09-09 09:08:54 +0200779 * Explicit flushing or periodic writeback of "old" data.
780 *
781 * Define "old": the first time one of an inode's pages is dirtied, we mark the
782 * dirtying-time in the inode's address_space. So this periodic writeback code
783 * just walks the superblock inode list, writing back any inodes which are
784 * older than a specific point in time.
785 *
786 * Try to run once per dirty_writeback_interval. But if a writeback event
787 * takes longer than a dirty_writeback_interval interval, then leave a
788 * one-second gap.
789 *
790 * older_than_this takes precedence over nr_to_write. So we'll only write back
791 * all dirty pages if they are all attached to "old" mappings.
792 */
Jens Axboec4a77a62009-09-16 15:18:25 +0200793static long wb_writeback(struct bdi_writeback *wb,
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200794 struct wb_writeback_work *work)
Jens Axboe03ba3782009-09-09 09:08:54 +0200795{
Wu Fengguange98be2d2010-08-29 11:22:30 -0600796 unsigned long wb_start = jiffies;
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600797 long nr_pages = work->nr_pages;
Jens Axboe03ba3782009-09-09 09:08:54 +0200798 unsigned long oldest_jif;
Jan Karaa5989bd2009-09-16 19:22:48 +0200799 struct inode *inode;
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600800 long progress;
Jens Axboe03ba3782009-09-09 09:08:54 +0200801
Wu Fengguange185dda2011-04-23 11:26:07 -0600802 oldest_jif = jiffies;
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600803 work->older_than_this = &oldest_jif;
Jens Axboe03ba3782009-09-09 09:08:54 +0200804
Wu Fengguange8dfc302011-04-21 12:06:32 -0600805 spin_lock(&wb->list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200806 for (;;) {
807 /*
Wu Fengguangd3ddec72009-09-23 20:33:40 +0800808 * Stop writeback when nr_pages has been consumed
Jens Axboe03ba3782009-09-09 09:08:54 +0200809 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200810 if (work->nr_pages <= 0)
Jens Axboe03ba3782009-09-09 09:08:54 +0200811 break;
812
813 /*
Jan Karaaa373cf2011-01-13 15:45:47 -0800814 * Background writeout and kupdate-style writeback may
815 * run forever. Stop them if there is other work to do
816 * so that e.g. sync can proceed. They'll be restarted
817 * after the other works are all done.
818 */
819 if ((work->for_background || work->for_kupdate) &&
820 !list_empty(&wb->bdi->work_list))
821 break;
822
823 /*
Wu Fengguangd3ddec72009-09-23 20:33:40 +0800824 * For background writeout, stop when we are below the
825 * background dirty threshold
Jens Axboe03ba3782009-09-09 09:08:54 +0200826 */
Wu Fengguangb00949a2010-11-18 14:38:33 -0600827 if (work->for_background && !over_bground_thresh(wb->bdi))
Jens Axboe03ba3782009-09-09 09:08:54 +0200828 break;
829
Jan Kara1bc36b62011-10-19 11:44:41 +0200830 /*
831 * Kupdate and background works are special and we want to
832 * include all inodes that need writing. Livelock avoidance is
833 * handled by these works yielding to any other work so we are
834 * safe.
835 */
Wu Fengguangba9aa832010-07-21 20:32:30 -0600836 if (work->for_kupdate) {
837 oldest_jif = jiffies -
838 msecs_to_jiffies(dirty_expire_interval * 10);
Jan Kara1bc36b62011-10-19 11:44:41 +0200839 } else if (work->for_background)
840 oldest_jif = jiffies;
Dave Chinner028c2dd2010-07-07 13:24:07 +1000841
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600842 trace_writeback_start(wb->bdi, work);
Wu Fengguange8dfc302011-04-21 12:06:32 -0600843 if (list_empty(&wb->b_io))
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600844 queue_io(wb, work);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200845 if (work->sb)
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600846 progress = writeback_sb_inodes(work->sb, wb, work);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200847 else
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600848 progress = __writeback_inodes_wb(wb, work);
849 trace_writeback_written(wb->bdi, work);
Dave Chinner028c2dd2010-07-07 13:24:07 +1000850
Wu Fengguange98be2d2010-08-29 11:22:30 -0600851 wb_update_bandwidth(wb, wb_start);
Jens Axboe03ba3782009-09-09 09:08:54 +0200852
853 /*
Jens Axboe71fd05a2009-09-23 19:32:26 +0200854 * Did we write something? Try for more
Wu Fengguange6fb6da2010-07-22 10:23:44 -0600855 *
856 * Dirty inodes are moved to b_io for writeback in batches.
857 * The completion of the current batch does not necessarily
858 * mean the overall work is done. So we keep looping as long
859 * as made some progress on cleaning pages or inodes.
Jens Axboe71fd05a2009-09-23 19:32:26 +0200860 */
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600861 if (progress)
Jens Axboe71fd05a2009-09-23 19:32:26 +0200862 continue;
863 /*
Wu Fengguange6fb6da2010-07-22 10:23:44 -0600864 * No more inodes for IO, bail
Jens Axboe71fd05a2009-09-23 19:32:26 +0200865 */
Wu Fengguangb7a24412010-07-21 22:19:51 -0600866 if (list_empty(&wb->b_more_io))
Jens Axboe03ba3782009-09-09 09:08:54 +0200867 break;
868 /*
Jens Axboe8010c3b2009-09-15 20:04:57 +0200869 * Nothing written. Wait for some inode to
870 * become available for writeback. Otherwise
871 * we'll just busyloop.
Jens Axboe03ba3782009-09-09 09:08:54 +0200872 */
Jens Axboe03ba3782009-09-09 09:08:54 +0200873 if (!list_empty(&wb->b_more_io)) {
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600874 trace_writeback_wait(wb->bdi, work);
Jens Axboe03ba3782009-09-09 09:08:54 +0200875 inode = wb_inode(wb->b_more_io.prev);
Dave Chinner250df6e2011-03-22 22:23:36 +1100876 spin_lock(&inode->i_lock);
Jan Karaf0d07b72012-05-03 14:47:59 +0200877 spin_unlock(&wb->list_lock);
Jan Kara169ebd92012-05-03 14:48:03 +0200878 /* This function drops i_lock... */
879 inode_sleep_on_writeback(inode);
Jan Karaf0d07b72012-05-03 14:47:59 +0200880 spin_lock(&wb->list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200881 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200882 }
Wu Fengguange8dfc302011-04-21 12:06:32 -0600883 spin_unlock(&wb->list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200884
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600885 return nr_pages - work->nr_pages;
Jens Axboe03ba3782009-09-09 09:08:54 +0200886}
887
888/*
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200889 * Return the next wb_writeback_work struct that hasn't been processed yet.
Jens Axboe03ba3782009-09-09 09:08:54 +0200890 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200891static struct wb_writeback_work *
Minchan Kim08852b62010-08-03 12:51:16 +0200892get_next_work_item(struct backing_dev_info *bdi)
Jens Axboe03ba3782009-09-09 09:08:54 +0200893{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200894 struct wb_writeback_work *work = NULL;
Jens Axboe03ba3782009-09-09 09:08:54 +0200895
Artem Bityutskiy64677162010-07-25 14:29:22 +0300896 spin_lock_bh(&bdi->wb_lock);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200897 if (!list_empty(&bdi->work_list)) {
898 work = list_entry(bdi->work_list.next,
899 struct wb_writeback_work, list);
900 list_del_init(&work->list);
Jens Axboe03ba3782009-09-09 09:08:54 +0200901 }
Artem Bityutskiy64677162010-07-25 14:29:22 +0300902 spin_unlock_bh(&bdi->wb_lock);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200903 return work;
Jens Axboe03ba3782009-09-09 09:08:54 +0200904}
905
Linus Torvaldscdf01dd2010-10-30 08:55:52 -0700906/*
907 * Add in the number of potentially dirty inodes, because each inode
908 * write can dirty pagecache in the underlying blockdev.
909 */
910static unsigned long get_nr_dirty_pages(void)
911{
912 return global_page_state(NR_FILE_DIRTY) +
913 global_page_state(NR_UNSTABLE_NFS) +
914 get_nr_dirty_inodes();
915}
916
Jan Kara65850272011-01-13 15:45:44 -0800917static long wb_check_background_flush(struct bdi_writeback *wb)
918{
Wu Fengguangb00949a2010-11-18 14:38:33 -0600919 if (over_bground_thresh(wb->bdi)) {
Jan Kara65850272011-01-13 15:45:44 -0800920
921 struct wb_writeback_work work = {
922 .nr_pages = LONG_MAX,
923 .sync_mode = WB_SYNC_NONE,
924 .for_background = 1,
925 .range_cyclic = 1,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600926 .reason = WB_REASON_BACKGROUND,
Jan Kara65850272011-01-13 15:45:44 -0800927 };
928
929 return wb_writeback(wb, &work);
930 }
931
932 return 0;
933}
934
Jens Axboe03ba3782009-09-09 09:08:54 +0200935static long wb_check_old_data_flush(struct bdi_writeback *wb)
936{
937 unsigned long expired;
938 long nr_pages;
939
Jens Axboe69b62d02010-05-17 12:51:03 +0200940 /*
941 * When set to zero, disable periodic writeback
942 */
943 if (!dirty_writeback_interval)
944 return 0;
945
Jens Axboe03ba3782009-09-09 09:08:54 +0200946 expired = wb->last_old_flush +
947 msecs_to_jiffies(dirty_writeback_interval * 10);
948 if (time_before(jiffies, expired))
949 return 0;
950
951 wb->last_old_flush = jiffies;
Linus Torvaldscdf01dd2010-10-30 08:55:52 -0700952 nr_pages = get_nr_dirty_pages();
Jens Axboe03ba3782009-09-09 09:08:54 +0200953
Jens Axboec4a77a62009-09-16 15:18:25 +0200954 if (nr_pages) {
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200955 struct wb_writeback_work work = {
Jens Axboec4a77a62009-09-16 15:18:25 +0200956 .nr_pages = nr_pages,
957 .sync_mode = WB_SYNC_NONE,
958 .for_kupdate = 1,
959 .range_cyclic = 1,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600960 .reason = WB_REASON_PERIODIC,
Jens Axboec4a77a62009-09-16 15:18:25 +0200961 };
962
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200963 return wb_writeback(wb, &work);
Jens Axboec4a77a62009-09-16 15:18:25 +0200964 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200965
966 return 0;
967}
968
969/*
970 * Retrieve work items and do the writeback they describe
971 */
972long wb_do_writeback(struct bdi_writeback *wb, int force_wait)
973{
974 struct backing_dev_info *bdi = wb->bdi;
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200975 struct wb_writeback_work *work;
Jens Axboec4a77a62009-09-16 15:18:25 +0200976 long wrote = 0;
Jens Axboe03ba3782009-09-09 09:08:54 +0200977
Jan Kara81d73a32010-08-11 14:17:44 -0700978 set_bit(BDI_writeback_running, &wb->bdi->state);
Minchan Kim08852b62010-08-03 12:51:16 +0200979 while ((work = get_next_work_item(bdi)) != NULL) {
Jens Axboe03ba3782009-09-09 09:08:54 +0200980 /*
981 * Override sync mode, in case we must wait for completion
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200982 * because this thread is exiting now.
Jens Axboe03ba3782009-09-09 09:08:54 +0200983 */
984 if (force_wait)
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200985 work->sync_mode = WB_SYNC_ALL;
986
Dave Chinner455b2862010-07-07 13:24:06 +1000987 trace_writeback_exec(bdi, work);
988
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200989 wrote += wb_writeback(wb, work);
Jens Axboe03ba3782009-09-09 09:08:54 +0200990
991 /*
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200992 * Notify the caller of completion if this is a synchronous
993 * work item, otherwise just free it.
Jens Axboe03ba3782009-09-09 09:08:54 +0200994 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200995 if (work->done)
996 complete(work->done);
997 else
998 kfree(work);
Jens Axboe03ba3782009-09-09 09:08:54 +0200999 }
1000
1001 /*
1002 * Check for periodic writeback, kupdated() style
1003 */
1004 wrote += wb_check_old_data_flush(wb);
Jan Kara65850272011-01-13 15:45:44 -08001005 wrote += wb_check_background_flush(wb);
Jan Kara81d73a32010-08-11 14:17:44 -07001006 clear_bit(BDI_writeback_running, &wb->bdi->state);
Jens Axboe03ba3782009-09-09 09:08:54 +02001007
1008 return wrote;
1009}
1010
1011/*
1012 * Handle writeback of dirty data for the device backed by this bdi. Also
1013 * wakes up periodically and does kupdated style flushing.
1014 */
Christoph Hellwig08243902010-06-19 23:08:22 +02001015int bdi_writeback_thread(void *data)
Jens Axboe03ba3782009-09-09 09:08:54 +02001016{
Christoph Hellwig08243902010-06-19 23:08:22 +02001017 struct bdi_writeback *wb = data;
1018 struct backing_dev_info *bdi = wb->bdi;
Jens Axboe03ba3782009-09-09 09:08:54 +02001019 long pages_written;
1020
Peter Zijlstra766f9162010-10-26 14:22:45 -07001021 current->flags |= PF_SWAPWRITE;
Christoph Hellwig08243902010-06-19 23:08:22 +02001022 set_freezable();
Artem Bityutskiyecd58402010-07-25 14:29:18 +03001023 wb->last_active = jiffies;
Christoph Hellwig08243902010-06-19 23:08:22 +02001024
1025 /*
1026 * Our parent may run at a different priority, just set us to normal
1027 */
1028 set_user_nice(current, 0);
1029
Dave Chinner455b2862010-07-07 13:24:06 +10001030 trace_writeback_thread_start(bdi);
1031
Tejun Heo8a32c442011-11-21 12:32:23 -08001032 while (!kthread_freezable_should_stop(NULL)) {
Artem Bityutskiy64677162010-07-25 14:29:22 +03001033 /*
1034 * Remove own delayed wake-up timer, since we are already awake
1035 * and we'll take care of the preriodic write-back.
1036 */
1037 del_timer(&wb->wakeup_timer);
1038
Jens Axboe03ba3782009-09-09 09:08:54 +02001039 pages_written = wb_do_writeback(wb, 0);
1040
Dave Chinner455b2862010-07-07 13:24:06 +10001041 trace_writeback_pages_written(pages_written);
Jens Axboe03ba3782009-09-09 09:08:54 +02001042
Jens Axboe03ba3782009-09-09 09:08:54 +02001043 if (pages_written)
Artem Bityutskiyecd58402010-07-25 14:29:18 +03001044 wb->last_active = jiffies;
Jens Axboe03ba3782009-09-09 09:08:54 +02001045
Artem Bityutskiy297252c2010-07-25 14:29:15 +03001046 set_current_state(TASK_INTERRUPTIBLE);
J. Bruce Fieldsb76b4012010-08-28 08:52:10 +02001047 if (!list_empty(&bdi->work_list) || kthread_should_stop()) {
Artem Bityutskiy297252c2010-07-25 14:29:15 +03001048 __set_current_state(TASK_RUNNING);
1049 continue;
Jens Axboe03ba3782009-09-09 09:08:54 +02001050 }
1051
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001052 if (wb_has_dirty_io(wb) && dirty_writeback_interval)
Artem Bityutskiyfff5b852010-07-25 14:29:20 +03001053 schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10));
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001054 else {
1055 /*
1056 * We have nothing to do, so can go sleep without any
1057 * timeout and save power. When a work is queued or
1058 * something is made dirty - we will be woken up.
1059 */
Artem Bityutskiy297252c2010-07-25 14:29:15 +03001060 schedule();
Jens Axboef9eadbb2010-05-18 14:31:45 +02001061 }
Jens Axboe03ba3782009-09-09 09:08:54 +02001062 }
1063
Artem Bityutskiyfff5b852010-07-25 14:29:20 +03001064 /* Flush any work that raced with us exiting */
Christoph Hellwig08243902010-06-19 23:08:22 +02001065 if (!list_empty(&bdi->work_list))
1066 wb_do_writeback(wb, 1);
Dave Chinner455b2862010-07-07 13:24:06 +10001067
1068 trace_writeback_thread_stop(bdi);
Jens Axboe03ba3782009-09-09 09:08:54 +02001069 return 0;
1070}
1071
Christoph Hellwig08243902010-06-19 23:08:22 +02001072
Jens Axboe03ba3782009-09-09 09:08:54 +02001073/*
Jens Axboe03ba3782009-09-09 09:08:54 +02001074 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
1075 * the whole world.
1076 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001077void wakeup_flusher_threads(long nr_pages, enum wb_reason reason)
Jens Axboe03ba3782009-09-09 09:08:54 +02001078{
Christoph Hellwigb8c2f342010-06-08 18:15:07 +02001079 struct backing_dev_info *bdi;
Christoph Hellwigb8c2f342010-06-08 18:15:07 +02001080
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001081 if (!nr_pages) {
1082 nr_pages = global_page_state(NR_FILE_DIRTY) +
Jens Axboe03ba3782009-09-09 09:08:54 +02001083 global_page_state(NR_UNSTABLE_NFS);
Christoph Hellwigb8c2f342010-06-08 18:15:07 +02001084 }
1085
1086 rcu_read_lock();
1087 list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) {
1088 if (!bdi_has_dirty_io(bdi))
1089 continue;
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001090 __bdi_start_writeback(bdi, nr_pages, false, reason);
Christoph Hellwigb8c2f342010-06-08 18:15:07 +02001091 }
1092 rcu_read_unlock();
Jens Axboe03ba3782009-09-09 09:08:54 +02001093}
1094
1095static noinline void block_dump___mark_inode_dirty(struct inode *inode)
1096{
1097 if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) {
1098 struct dentry *dentry;
1099 const char *name = "?";
1100
1101 dentry = d_find_alias(inode);
1102 if (dentry) {
1103 spin_lock(&dentry->d_lock);
1104 name = (const char *) dentry->d_name.name;
1105 }
1106 printk(KERN_DEBUG
1107 "%s(%d): dirtied inode %lu (%s) on %s\n",
1108 current->comm, task_pid_nr(current), inode->i_ino,
1109 name, inode->i_sb->s_id);
1110 if (dentry) {
1111 spin_unlock(&dentry->d_lock);
1112 dput(dentry);
1113 }
1114 }
1115}
1116
1117/**
1118 * __mark_inode_dirty - internal function
1119 * @inode: inode to mark
1120 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
1121 * Mark an inode as dirty. Callers should use mark_inode_dirty or
1122 * mark_inode_dirty_sync.
1123 *
1124 * Put the inode on the super block's dirty list.
1125 *
1126 * CAREFUL! We mark it dirty unconditionally, but move it onto the
1127 * dirty list only if it is hashed or if it refers to a blockdev.
1128 * If it was not hashed, it will never be added to the dirty list
1129 * even if it is later hashed, as it will have been marked dirty already.
1130 *
1131 * In short, make sure you hash any inodes _before_ you start marking
1132 * them dirty.
1133 *
Jens Axboe03ba3782009-09-09 09:08:54 +02001134 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
1135 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
1136 * the kernel-internal blockdev inode represents the dirtying time of the
1137 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
1138 * page->mapping->host, so the page-dirtying time is recorded in the internal
1139 * blockdev inode.
1140 */
1141void __mark_inode_dirty(struct inode *inode, int flags)
1142{
1143 struct super_block *sb = inode->i_sb;
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001144 struct backing_dev_info *bdi = NULL;
Jens Axboe03ba3782009-09-09 09:08:54 +02001145
1146 /*
1147 * Don't do this for I_DIRTY_PAGES - that doesn't actually
1148 * dirty the inode itself
1149 */
1150 if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
1151 if (sb->s_op->dirty_inode)
Christoph Hellwigaa385722011-05-27 06:53:02 -04001152 sb->s_op->dirty_inode(inode, flags);
Jens Axboe03ba3782009-09-09 09:08:54 +02001153 }
1154
1155 /*
1156 * make sure that changes are seen by all cpus before we test i_state
1157 * -- mikulas
1158 */
1159 smp_mb();
1160
1161 /* avoid the locking if we can */
1162 if ((inode->i_state & flags) == flags)
1163 return;
1164
1165 if (unlikely(block_dump))
1166 block_dump___mark_inode_dirty(inode);
1167
Dave Chinner250df6e2011-03-22 22:23:36 +11001168 spin_lock(&inode->i_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001169 if ((inode->i_state & flags) != flags) {
1170 const int was_dirty = inode->i_state & I_DIRTY;
1171
1172 inode->i_state |= flags;
1173
1174 /*
1175 * If the inode is being synced, just update its dirty state.
1176 * The unlocker will place the inode on the appropriate
1177 * superblock list, based upon its state.
1178 */
1179 if (inode->i_state & I_SYNC)
Dave Chinner250df6e2011-03-22 22:23:36 +11001180 goto out_unlock_inode;
Jens Axboe03ba3782009-09-09 09:08:54 +02001181
1182 /*
1183 * Only add valid (hashed) inodes to the superblock's
1184 * dirty list. Add blockdev inodes as well.
1185 */
1186 if (!S_ISBLK(inode->i_mode)) {
Al Viro1d3382cb2010-10-23 15:19:20 -04001187 if (inode_unhashed(inode))
Dave Chinner250df6e2011-03-22 22:23:36 +11001188 goto out_unlock_inode;
Jens Axboe03ba3782009-09-09 09:08:54 +02001189 }
Al Viroa4ffdde2010-06-02 17:38:30 -04001190 if (inode->i_state & I_FREEING)
Dave Chinner250df6e2011-03-22 22:23:36 +11001191 goto out_unlock_inode;
Jens Axboe03ba3782009-09-09 09:08:54 +02001192
1193 /*
1194 * If the inode was already on b_dirty/b_io/b_more_io, don't
1195 * reposition it (that would break b_dirty time-ordering).
1196 */
1197 if (!was_dirty) {
Dave Chinnera66979a2011-03-22 22:23:41 +11001198 bool wakeup_bdi = false;
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001199 bdi = inode_to_bdi(inode);
Jens Axboe500b0672009-09-09 09:10:25 +02001200
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001201 if (bdi_cap_writeback_dirty(bdi)) {
1202 WARN(!test_bit(BDI_registered, &bdi->state),
1203 "bdi-%s not registered\n", bdi->name);
1204
1205 /*
1206 * If this is the first dirty inode for this
1207 * bdi, we have to wake-up the corresponding
1208 * bdi thread to make sure background
1209 * write-back happens later.
1210 */
1211 if (!wb_has_dirty_io(&bdi->wb))
1212 wakeup_bdi = true;
Jens Axboe500b0672009-09-09 09:10:25 +02001213 }
Jens Axboe03ba3782009-09-09 09:08:54 +02001214
Dave Chinnera66979a2011-03-22 22:23:41 +11001215 spin_unlock(&inode->i_lock);
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001216 spin_lock(&bdi->wb.list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001217 inode->dirtied_when = jiffies;
Nick Piggin7ccf19a2010-10-21 11:49:30 +11001218 list_move(&inode->i_wb_list, &bdi->wb.b_dirty);
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001219 spin_unlock(&bdi->wb.list_lock);
Dave Chinnera66979a2011-03-22 22:23:41 +11001220
1221 if (wakeup_bdi)
1222 bdi_wakeup_thread_delayed(bdi);
1223 return;
Jens Axboe03ba3782009-09-09 09:08:54 +02001224 }
1225 }
Dave Chinner250df6e2011-03-22 22:23:36 +11001226out_unlock_inode:
1227 spin_unlock(&inode->i_lock);
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001228
Jens Axboe03ba3782009-09-09 09:08:54 +02001229}
1230EXPORT_SYMBOL(__mark_inode_dirty);
1231
Jens Axboeb6e51312009-09-16 15:13:54 +02001232static void wait_sb_inodes(struct super_block *sb)
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001233{
Jens Axboe03ba3782009-09-09 09:08:54 +02001234 struct inode *inode, *old_inode = NULL;
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001235
Jens Axboe03ba3782009-09-09 09:08:54 +02001236 /*
1237 * We need to be protected against the filesystem going from
1238 * r/o to r/w or vice versa.
1239 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001240 WARN_ON(!rwsem_is_locked(&sb->s_umount));
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001241
Dave Chinner55fa6092011-03-22 22:23:40 +11001242 spin_lock(&inode_sb_list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001243
1244 /*
1245 * Data integrity sync. Must wait for all pages under writeback,
1246 * because there may have been pages dirtied before our sync
1247 * call, but which had writeout started before we write it out.
1248 * In which case, the inode may not be on the dirty list, but
1249 * we still have to wait for that writeout.
1250 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001251 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
Dave Chinner250df6e2011-03-22 22:23:36 +11001252 struct address_space *mapping = inode->i_mapping;
Jens Axboe03ba3782009-09-09 09:08:54 +02001253
Dave Chinner250df6e2011-03-22 22:23:36 +11001254 spin_lock(&inode->i_lock);
1255 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
1256 (mapping->nrpages == 0)) {
1257 spin_unlock(&inode->i_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001258 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +11001259 }
Jens Axboe03ba3782009-09-09 09:08:54 +02001260 __iget(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +11001261 spin_unlock(&inode->i_lock);
Dave Chinner55fa6092011-03-22 22:23:40 +11001262 spin_unlock(&inode_sb_list_lock);
1263
Jens Axboe03ba3782009-09-09 09:08:54 +02001264 /*
Dave Chinner55fa6092011-03-22 22:23:40 +11001265 * We hold a reference to 'inode' so it couldn't have been
1266 * removed from s_inodes list while we dropped the
1267 * inode_sb_list_lock. We cannot iput the inode now as we can
1268 * be holding the last reference and we cannot iput it under
1269 * inode_sb_list_lock. So we keep the reference and iput it
1270 * later.
Jens Axboe03ba3782009-09-09 09:08:54 +02001271 */
1272 iput(old_inode);
1273 old_inode = inode;
1274
1275 filemap_fdatawait(mapping);
1276
1277 cond_resched();
Nick Piggin38f21972009-01-06 14:40:25 -08001278
Dave Chinner55fa6092011-03-22 22:23:40 +11001279 spin_lock(&inode_sb_list_lock);
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001280 }
Dave Chinner55fa6092011-03-22 22:23:40 +11001281 spin_unlock(&inode_sb_list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001282 iput(old_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283}
1284
Jens Axboed8a85592009-09-02 12:34:32 +02001285/**
Chris Mason3259f8b2010-10-29 11:16:17 -04001286 * writeback_inodes_sb_nr - writeback dirty inodes from given super_block
Jens Axboed8a85592009-09-02 12:34:32 +02001287 * @sb: the superblock
Chris Mason3259f8b2010-10-29 11:16:17 -04001288 * @nr: the number of pages to write
Marcos Paulo de Souza786228a2011-11-23 20:56:45 +08001289 * @reason: reason why some writeback work initiated
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290 *
Jens Axboed8a85592009-09-02 12:34:32 +02001291 * Start writeback on some inodes on this super_block. No guarantees are made
1292 * on how many (if any) will be written, and this function does not wait
Chris Mason3259f8b2010-10-29 11:16:17 -04001293 * for IO completion of submitted IO.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001295void writeback_inodes_sb_nr(struct super_block *sb,
1296 unsigned long nr,
1297 enum wb_reason reason)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001299 DECLARE_COMPLETION_ONSTACK(done);
1300 struct wb_writeback_work work = {
Wu Fengguang6e6938b2010-06-06 10:38:15 -06001301 .sb = sb,
1302 .sync_mode = WB_SYNC_NONE,
1303 .tagged_writepages = 1,
1304 .done = &done,
1305 .nr_pages = nr,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001306 .reason = reason,
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001307 };
Jens Axboe0e3c9a22010-06-01 11:08:43 +02001308
Jan Kara6eedc702012-07-03 16:45:27 +02001309 if (sb->s_bdi == &noop_backing_dev_info)
1310 return;
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001311 WARN_ON(!rwsem_is_locked(&sb->s_umount));
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001312 bdi_queue_work(sb->s_bdi, &work);
1313 wait_for_completion(&done);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314}
Chris Mason3259f8b2010-10-29 11:16:17 -04001315EXPORT_SYMBOL(writeback_inodes_sb_nr);
1316
1317/**
1318 * writeback_inodes_sb - writeback dirty inodes from given super_block
1319 * @sb: the superblock
Marcos Paulo de Souza786228a2011-11-23 20:56:45 +08001320 * @reason: reason why some writeback work was initiated
Chris Mason3259f8b2010-10-29 11:16:17 -04001321 *
1322 * Start writeback on some inodes on this super_block. No guarantees are made
1323 * on how many (if any) will be written, and this function does not wait
1324 * for IO completion of submitted IO.
1325 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001326void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason)
Chris Mason3259f8b2010-10-29 11:16:17 -04001327{
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001328 return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason);
Chris Mason3259f8b2010-10-29 11:16:17 -04001329}
Jens Axboed8a85592009-09-02 12:34:32 +02001330EXPORT_SYMBOL(writeback_inodes_sb);
1331
1332/**
Eric Sandeen17bd55d2009-12-23 07:57:07 -05001333 * writeback_inodes_sb_if_idle - start writeback if none underway
1334 * @sb: the superblock
Marcos Paulo de Souza786228a2011-11-23 20:56:45 +08001335 * @reason: reason why some writeback work was initiated
Eric Sandeen17bd55d2009-12-23 07:57:07 -05001336 *
1337 * Invoke writeback_inodes_sb if no writeback is currently underway.
1338 * Returns 1 if writeback was started, 0 if not.
1339 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001340int writeback_inodes_sb_if_idle(struct super_block *sb, enum wb_reason reason)
Eric Sandeen17bd55d2009-12-23 07:57:07 -05001341{
1342 if (!writeback_in_progress(sb->s_bdi)) {
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001343 down_read(&sb->s_umount);
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001344 writeback_inodes_sb(sb, reason);
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001345 up_read(&sb->s_umount);
Eric Sandeen17bd55d2009-12-23 07:57:07 -05001346 return 1;
1347 } else
1348 return 0;
1349}
1350EXPORT_SYMBOL(writeback_inodes_sb_if_idle);
1351
1352/**
Fengguang Wuc097b2c2012-03-05 15:08:06 -08001353 * writeback_inodes_sb_nr_if_idle - start writeback if none underway
Chris Mason3259f8b2010-10-29 11:16:17 -04001354 * @sb: the superblock
1355 * @nr: the number of pages to write
Marcos Paulo de Souza786228a2011-11-23 20:56:45 +08001356 * @reason: reason why some writeback work was initiated
Chris Mason3259f8b2010-10-29 11:16:17 -04001357 *
1358 * Invoke writeback_inodes_sb if no writeback is currently underway.
1359 * Returns 1 if writeback was started, 0 if not.
1360 */
1361int writeback_inodes_sb_nr_if_idle(struct super_block *sb,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001362 unsigned long nr,
1363 enum wb_reason reason)
Chris Mason3259f8b2010-10-29 11:16:17 -04001364{
1365 if (!writeback_in_progress(sb->s_bdi)) {
1366 down_read(&sb->s_umount);
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001367 writeback_inodes_sb_nr(sb, nr, reason);
Chris Mason3259f8b2010-10-29 11:16:17 -04001368 up_read(&sb->s_umount);
1369 return 1;
1370 } else
1371 return 0;
1372}
1373EXPORT_SYMBOL(writeback_inodes_sb_nr_if_idle);
1374
1375/**
Jens Axboed8a85592009-09-02 12:34:32 +02001376 * sync_inodes_sb - sync sb inode pages
1377 * @sb: the superblock
1378 *
1379 * This function writes and waits on any dirty inode belonging to this
Stefan Hajnoczicb9ef8d2011-01-13 15:47:26 -08001380 * super_block.
Jens Axboed8a85592009-09-02 12:34:32 +02001381 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001382void sync_inodes_sb(struct super_block *sb)
Jens Axboed8a85592009-09-02 12:34:32 +02001383{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001384 DECLARE_COMPLETION_ONSTACK(done);
1385 struct wb_writeback_work work = {
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001386 .sb = sb,
1387 .sync_mode = WB_SYNC_ALL,
1388 .nr_pages = LONG_MAX,
1389 .range_cyclic = 0,
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001390 .done = &done,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001391 .reason = WB_REASON_SYNC,
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001392 };
1393
Jan Kara6eedc702012-07-03 16:45:27 +02001394 /* Nothing to do? */
1395 if (sb->s_bdi == &noop_backing_dev_info)
1396 return;
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001397 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1398
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001399 bdi_queue_work(sb->s_bdi, &work);
1400 wait_for_completion(&done);
1401
Jens Axboeb6e51312009-09-16 15:13:54 +02001402 wait_sb_inodes(sb);
Jens Axboed8a85592009-09-02 12:34:32 +02001403}
1404EXPORT_SYMBOL(sync_inodes_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406/**
Andrea Arcangeli7f04c262005-10-30 15:03:05 -08001407 * write_inode_now - write an inode to disk
1408 * @inode: inode to write to disk
1409 * @sync: whether the write should be synchronous or not
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410 *
Andrea Arcangeli7f04c262005-10-30 15:03:05 -08001411 * This function commits an inode to disk immediately if it is dirty. This is
1412 * primarily needed by knfsd.
1413 *
1414 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416int write_inode_now(struct inode *inode, int sync)
1417{
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001418 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419 struct writeback_control wbc = {
1420 .nr_to_write = LONG_MAX,
Mike Galbraith18914b12008-02-08 04:20:23 -08001421 .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -07001422 .range_start = 0,
1423 .range_end = LLONG_MAX,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424 };
1425
1426 if (!mapping_cap_writeback_dirty(inode->i_mapping))
Andrew Morton49364ce2005-11-07 00:59:15 -08001427 wbc.nr_to_write = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428
1429 might_sleep();
Jan Kara4f8ad652012-05-03 14:48:00 +02001430 return writeback_single_inode(inode, wb, &wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431}
1432EXPORT_SYMBOL(write_inode_now);
1433
1434/**
1435 * sync_inode - write an inode and its pages to disk.
1436 * @inode: the inode to sync
1437 * @wbc: controls the writeback mode
1438 *
1439 * sync_inode() will write an inode and its pages to disk. It will also
1440 * correctly update the inode on its superblock's dirty inode lists and will
1441 * update inode->i_state.
1442 *
1443 * The caller must have a ref on the inode.
1444 */
1445int sync_inode(struct inode *inode, struct writeback_control *wbc)
1446{
Jan Kara4f8ad652012-05-03 14:48:00 +02001447 return writeback_single_inode(inode, &inode_to_bdi(inode)->wb, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448}
1449EXPORT_SYMBOL(sync_inode);
Christoph Hellwigc37650162010-10-06 10:48:20 +02001450
1451/**
Andrew Mortonc691b9d2011-01-13 15:45:48 -08001452 * sync_inode_metadata - write an inode to disk
Christoph Hellwigc37650162010-10-06 10:48:20 +02001453 * @inode: the inode to sync
1454 * @wait: wait for I/O to complete.
1455 *
Andrew Mortonc691b9d2011-01-13 15:45:48 -08001456 * Write an inode to disk and adjust its dirty state after completion.
Christoph Hellwigc37650162010-10-06 10:48:20 +02001457 *
1458 * Note: only writes the actual inode, no associated data or other metadata.
1459 */
1460int sync_inode_metadata(struct inode *inode, int wait)
1461{
1462 struct writeback_control wbc = {
1463 .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE,
1464 .nr_to_write = 0, /* metadata-only */
1465 };
1466
1467 return sync_inode(inode, &wbc);
1468}
1469EXPORT_SYMBOL(sync_inode_metadata);