blob: be3efc4f64f4b8c556e1d6de5c2ff400f05731ab [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 Howells07f3f05c2006-09-30 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}
66
Jan Kara692ebd12010-09-21 11:51:01 +020067static inline struct backing_dev_info *inode_to_bdi(struct inode *inode)
68{
69 struct super_block *sb = inode->i_sb;
Jan Kara692ebd12010-09-21 11:51:01 +020070
Christoph Hellwigaaead252010-10-04 14:25:33 +020071 if (strcmp(sb->s_type->name, "bdev") == 0)
72 return inode->i_mapping->backing_dev_info;
73
74 return sb->s_bdi;
Jan Kara692ebd12010-09-21 11:51:01 +020075}
76
Nick Piggin7ccf19a2010-10-21 11:49:30 +110077static inline struct inode *wb_inode(struct list_head *head)
78{
79 return list_entry(head, struct inode, i_wb_list);
80}
81
Wu Fengguang15eb77a2012-01-17 11:18:56 -060082/*
83 * Include the creation of the trace points after defining the
84 * wb_writeback_work structure and inline functions so that the definition
85 * remains local to this file.
86 */
87#define CREATE_TRACE_POINTS
88#include <trace/events/writeback.h>
89
Jan Kara65850272011-01-13 15:45:44 -080090/* Wakeup flusher thread or forker thread to fork it. Requires bdi->wb_lock. */
91static void bdi_wakeup_flusher(struct backing_dev_info *bdi)
Nick Piggin4195f732009-05-28 09:01:15 +020092{
Artem Bityutskiyfff5b852010-07-25 14:29:20 +030093 if (bdi->wb.task) {
94 wake_up_process(bdi->wb.task);
95 } else {
96 /*
97 * The bdi thread isn't there, wake up the forker thread which
98 * will create and run it.
99 */
Jens Axboe03ba3782009-09-09 09:08:54 +0200100 wake_up_process(default_backing_dev_info.wb.task);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101 }
Jan Kara65850272011-01-13 15:45:44 -0800102}
103
104static void bdi_queue_work(struct backing_dev_info *bdi,
105 struct wb_writeback_work *work)
106{
107 trace_writeback_queue(bdi, work);
108
109 spin_lock_bh(&bdi->wb_lock);
110 list_add_tail(&work->list, &bdi->work_list);
111 if (!bdi->wb.task)
112 trace_writeback_nothread(bdi, work);
113 bdi_wakeup_flusher(bdi);
Artem Bityutskiy64677162010-07-25 14:29:22 +0300114 spin_unlock_bh(&bdi->wb_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115}
116
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200117static void
118__bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600119 bool range_cyclic, enum wb_reason reason)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200121 struct wb_writeback_work *work;
Jens Axboe03ba3782009-09-09 09:08:54 +0200122
Jens Axboebcddc3f2009-09-13 20:07:36 +0200123 /*
124 * This is WB_SYNC_NONE writeback, so if allocation fails just
125 * wakeup the thread for old dirty data writeback
126 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200127 work = kzalloc(sizeof(*work), GFP_ATOMIC);
128 if (!work) {
Dave Chinner455b2862010-07-07 13:24:06 +1000129 if (bdi->wb.task) {
130 trace_writeback_nowork(bdi);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200131 wake_up_process(bdi->wb.task);
Dave Chinner455b2862010-07-07 13:24:06 +1000132 }
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200133 return;
Jens Axboebcddc3f2009-09-13 20:07:36 +0200134 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200135
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200136 work->sync_mode = WB_SYNC_NONE;
137 work->nr_pages = nr_pages;
138 work->range_cyclic = range_cyclic;
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600139 work->reason = reason;
Christoph Hellwigf0fad8a2009-09-11 09:47:56 +0200140
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200141 bdi_queue_work(bdi, work);
Jens Axboeb6e51312009-09-16 15:13:54 +0200142}
143
144/**
145 * bdi_start_writeback - start writeback
146 * @bdi: the backing device to write from
147 * @nr_pages: the number of pages to write
Marcos Paulo de Souza786228a2011-11-23 20:56:45 +0800148 * @reason: reason why some writeback work was initiated
Jens Axboeb6e51312009-09-16 15:13:54 +0200149 *
150 * Description:
151 * This does WB_SYNC_NONE opportunistic writeback. The IO is only
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300152 * started when this function returns, we make no guarantees on
Jens Axboe0e3c9a22010-06-01 11:08:43 +0200153 * completion. Caller need not hold sb s_umount semaphore.
Jens Axboeb6e51312009-09-16 15:13:54 +0200154 *
155 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600156void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages,
157 enum wb_reason reason)
Jens Axboeb6e51312009-09-16 15:13:54 +0200158{
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600159 __bdi_start_writeback(bdi, nr_pages, true, reason);
Christoph Hellwigc5444192010-06-08 18:15:15 +0200160}
Wu Fengguangd3ddec72009-09-23 20:33:40 +0800161
Christoph Hellwigc5444192010-06-08 18:15:15 +0200162/**
163 * bdi_start_background_writeback - start background writeback
164 * @bdi: the backing device to write from
165 *
166 * Description:
Jan Kara65850272011-01-13 15:45:44 -0800167 * This makes sure WB_SYNC_NONE background writeback happens. When
168 * this function returns, it is only guaranteed that for given BDI
169 * some IO is happening if we are over background dirty threshold.
170 * Caller need not hold sb s_umount semaphore.
Christoph Hellwigc5444192010-06-08 18:15:15 +0200171 */
172void bdi_start_background_writeback(struct backing_dev_info *bdi)
173{
Jan Kara65850272011-01-13 15:45:44 -0800174 /*
175 * We just wake up the flusher thread. It will perform background
176 * writeback as soon as there is no other work to do.
177 */
Wu Fengguang71927e82011-01-13 15:45:46 -0800178 trace_writeback_wake_background(bdi);
Jan Kara65850272011-01-13 15:45:44 -0800179 spin_lock_bh(&bdi->wb_lock);
180 bdi_wakeup_flusher(bdi);
181 spin_unlock_bh(&bdi->wb_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182}
183
184/*
Dave Chinnera66979a2011-03-22 22:23:41 +1100185 * Remove the inode from the writeback list it is on.
186 */
187void inode_wb_list_del(struct inode *inode)
188{
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600189 struct backing_dev_info *bdi = inode_to_bdi(inode);
Dave Chinnera66979a2011-03-22 22:23:41 +1100190
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600191 spin_lock(&bdi->wb.list_lock);
192 list_del_init(&inode->i_wb_list);
193 spin_unlock(&bdi->wb.list_lock);
194}
Dave Chinnera66979a2011-03-22 22:23:41 +1100195
196/*
Andrew Morton6610a0b2007-10-16 23:30:32 -0700197 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
198 * furthest end of its superblock's dirty-inode list.
199 *
200 * Before stamping the inode's ->dirtied_when, we check to see whether it is
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200201 * already the most-recently-dirtied inode on the b_dirty list. If that is
Andrew Morton6610a0b2007-10-16 23:30:32 -0700202 * the case then the inode must have been redirtied while it was being written
203 * out and we don't reset its dirtied_when.
204 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600205static void redirty_tail(struct inode *inode, struct bdi_writeback *wb)
Andrew Morton6610a0b2007-10-16 23:30:32 -0700206{
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600207 assert_spin_locked(&wb->list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200208 if (!list_empty(&wb->b_dirty)) {
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200209 struct inode *tail;
Andrew Morton6610a0b2007-10-16 23:30:32 -0700210
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100211 tail = wb_inode(wb->b_dirty.next);
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200212 if (time_before(inode->dirtied_when, tail->dirtied_when))
Andrew Morton6610a0b2007-10-16 23:30:32 -0700213 inode->dirtied_when = jiffies;
214 }
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100215 list_move(&inode->i_wb_list, &wb->b_dirty);
Andrew Morton6610a0b2007-10-16 23:30:32 -0700216}
217
218/*
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200219 * requeue inode for re-scanning after bdi->b_io list is exhausted.
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700220 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600221static void requeue_io(struct inode *inode, struct bdi_writeback *wb)
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700222{
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600223 assert_spin_locked(&wb->list_lock);
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100224 list_move(&inode->i_wb_list, &wb->b_more_io);
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700225}
226
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700227static void inode_sync_complete(struct inode *inode)
228{
Jan Kara365b94ae2012-05-03 14:47:55 +0200229 inode->i_state &= ~I_SYNC;
230 /* Waiters must see I_SYNC cleared before being woken up */
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700231 smp_mb();
232 wake_up_bit(&inode->i_state, __I_SYNC);
233}
234
Jeff Laytond2caa3c52009-04-02 16:56:37 -0700235static bool inode_dirtied_after(struct inode *inode, unsigned long t)
236{
237 bool ret = time_after(inode->dirtied_when, t);
238#ifndef CONFIG_64BIT
239 /*
240 * For inodes being constantly redirtied, dirtied_when can get stuck.
241 * It _appears_ to be in the future, but is actually in distant past.
242 * This test is necessary to prevent such wrapped-around relative times
Jens Axboe5b0830c2009-09-23 19:37:09 +0200243 * from permanently stopping the whole bdi writeback.
Jeff Laytond2caa3c52009-04-02 16:56:37 -0700244 */
245 ret = ret && time_before_eq(inode->dirtied_when, jiffies);
246#endif
247 return ret;
248}
249
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700250/*
Jan Kara697e6fe2012-03-09 07:26:22 -0800251 * Move expired (dirtied after work->older_than_this) dirty inodes from
252 * @delaying_queue to @dispatch_queue.
Fengguang Wu2c136572007-10-16 23:30:39 -0700253 */
Wu Fengguange84d0a42011-04-23 12:27:27 -0600254static int move_expired_inodes(struct list_head *delaying_queue,
Fengguang Wu2c136572007-10-16 23:30:39 -0700255 struct list_head *dispatch_queue,
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600256 struct wb_writeback_work *work)
Fengguang Wu2c136572007-10-16 23:30:39 -0700257{
Shaohua Li5c034492009-09-24 14:42:33 +0200258 LIST_HEAD(tmp);
259 struct list_head *pos, *node;
Jens Axboecf137302009-09-24 15:12:57 +0200260 struct super_block *sb = NULL;
Shaohua Li5c034492009-09-24 14:42:33 +0200261 struct inode *inode;
Jens Axboecf137302009-09-24 15:12:57 +0200262 int do_sb_sort = 0;
Wu Fengguange84d0a42011-04-23 12:27:27 -0600263 int moved = 0;
Shaohua Li5c034492009-09-24 14:42:33 +0200264
Fengguang Wu2c136572007-10-16 23:30:39 -0700265 while (!list_empty(delaying_queue)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100266 inode = wb_inode(delaying_queue->prev);
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600267 if (work->older_than_this &&
268 inode_dirtied_after(inode, *work->older_than_this))
Fengguang Wu2c136572007-10-16 23:30:39 -0700269 break;
Jens Axboecf137302009-09-24 15:12:57 +0200270 if (sb && sb != inode->i_sb)
271 do_sb_sort = 1;
272 sb = inode->i_sb;
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100273 list_move(&inode->i_wb_list, &tmp);
Wu Fengguange84d0a42011-04-23 12:27:27 -0600274 moved++;
Shaohua Li5c034492009-09-24 14:42:33 +0200275 }
276
Jens Axboecf137302009-09-24 15:12:57 +0200277 /* just one sb in list, splice to dispatch_queue and we're done */
278 if (!do_sb_sort) {
279 list_splice(&tmp, dispatch_queue);
Wu Fengguange84d0a42011-04-23 12:27:27 -0600280 goto out;
Jens Axboecf137302009-09-24 15:12:57 +0200281 }
282
Shaohua Li5c034492009-09-24 14:42:33 +0200283 /* Move inodes from one superblock together */
284 while (!list_empty(&tmp)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100285 sb = wb_inode(tmp.prev)->i_sb;
Shaohua Li5c034492009-09-24 14:42:33 +0200286 list_for_each_prev_safe(pos, node, &tmp) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100287 inode = wb_inode(pos);
Shaohua Li5c034492009-09-24 14:42:33 +0200288 if (inode->i_sb == sb)
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100289 list_move(&inode->i_wb_list, dispatch_queue);
Shaohua Li5c034492009-09-24 14:42:33 +0200290 }
Fengguang Wu2c136572007-10-16 23:30:39 -0700291 }
Wu Fengguange84d0a42011-04-23 12:27:27 -0600292out:
293 return moved;
Fengguang Wu2c136572007-10-16 23:30:39 -0700294}
295
296/*
297 * Queue all expired dirty inodes for io, eldest first.
Wu Fengguang4ea879b2010-08-11 14:17:42 -0700298 * Before
299 * newly dirtied b_dirty b_io b_more_io
300 * =============> gf edc BA
301 * After
302 * newly dirtied b_dirty b_io b_more_io
303 * =============> g fBAedc
304 * |
305 * +--> dequeue for IO
Fengguang Wu2c136572007-10-16 23:30:39 -0700306 */
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600307static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work)
Fengguang Wu2c136572007-10-16 23:30:39 -0700308{
Wu Fengguange84d0a42011-04-23 12:27:27 -0600309 int moved;
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600310 assert_spin_locked(&wb->list_lock);
Wu Fengguang4ea879b2010-08-11 14:17:42 -0700311 list_splice_init(&wb->b_more_io, &wb->b_io);
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600312 moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, work);
313 trace_writeback_queue_io(wb, work, moved);
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200314}
315
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100316static int write_inode(struct inode *inode, struct writeback_control *wbc)
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200317{
Jens Axboe03ba3782009-09-09 09:08:54 +0200318 if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode))
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100319 return inode->i_sb->s_op->write_inode(inode, wbc);
Jens Axboe03ba3782009-09-09 09:08:54 +0200320 return 0;
Fengguang Wu2c136572007-10-16 23:30:39 -0700321}
322
323/*
Jan Kara169ebd92012-05-03 14:48:03 +0200324 * Wait for writeback on an inode to complete. Called with i_lock held.
325 * Caller must make sure inode cannot go away when we drop i_lock.
Christoph Hellwig01c03192009-06-08 13:35:40 +0200326 */
Jan Kara169ebd92012-05-03 14:48:03 +0200327static void __inode_wait_for_writeback(struct inode *inode)
328 __releases(inode->i_lock)
329 __acquires(inode->i_lock)
Christoph Hellwig01c03192009-06-08 13:35:40 +0200330{
331 DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC);
332 wait_queue_head_t *wqh;
333
334 wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
Dave Chinner250df6e2011-03-22 22:23:36 +1100335 while (inode->i_state & I_SYNC) {
336 spin_unlock(&inode->i_lock);
Christoph Hellwig01c03192009-06-08 13:35:40 +0200337 __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE);
Dave Chinner250df6e2011-03-22 22:23:36 +1100338 spin_lock(&inode->i_lock);
Richard Kennedy58a9d3d82010-05-24 14:32:38 -0700339 }
Christoph Hellwig01c03192009-06-08 13:35:40 +0200340}
341
342/*
Jan Kara169ebd92012-05-03 14:48:03 +0200343 * Wait for writeback on an inode to complete. Caller must have inode pinned.
344 */
345void inode_wait_for_writeback(struct inode *inode)
346{
347 spin_lock(&inode->i_lock);
348 __inode_wait_for_writeback(inode);
349 spin_unlock(&inode->i_lock);
350}
351
352/*
353 * Sleep until I_SYNC is cleared. This function must be called with i_lock
354 * held and drops it. It is aimed for callers not holding any inode reference
355 * so once i_lock is dropped, inode can go away.
356 */
357static void inode_sleep_on_writeback(struct inode *inode)
358 __releases(inode->i_lock)
359{
360 DEFINE_WAIT(wait);
361 wait_queue_head_t *wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
362 int sleep;
363
364 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
365 sleep = inode->i_state & I_SYNC;
366 spin_unlock(&inode->i_lock);
367 if (sleep)
368 schedule();
369 finish_wait(wqh, &wait);
370}
371
372/*
Jan Karaccb26b52012-05-03 14:47:58 +0200373 * Find proper writeback list for the inode depending on its current state and
374 * possibly also change of its state while we were doing writeback. Here we
375 * handle things such as livelock prevention or fairness of writeback among
376 * inodes. This function can be called only by flusher thread - noone else
377 * processes all inodes in writeback lists and requeueing inodes behind flusher
378 * thread's back can have unexpected consequences.
379 */
380static void requeue_inode(struct inode *inode, struct bdi_writeback *wb,
381 struct writeback_control *wbc)
382{
383 if (inode->i_state & I_FREEING)
384 return;
385
386 /*
387 * Sync livelock prevention. Each inode is tagged and synced in one
388 * shot. If still dirty, it will be redirty_tail()'ed below. Update
389 * the dirty time to prevent enqueue and sync it again.
390 */
391 if ((inode->i_state & I_DIRTY) &&
392 (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages))
393 inode->dirtied_when = jiffies;
394
Jan Kara4f8ad652012-05-03 14:48:00 +0200395 if (wbc->pages_skipped) {
396 /*
397 * writeback is not making progress due to locked
398 * buffers. Skip this inode for now.
399 */
400 redirty_tail(inode, wb);
401 return;
402 }
403
Jan Karaccb26b52012-05-03 14:47:58 +0200404 if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) {
405 /*
406 * We didn't write back all the pages. nfs_writepages()
407 * sometimes bales out without doing anything.
408 */
409 if (wbc->nr_to_write <= 0) {
410 /* Slice used up. Queue for next turn. */
411 requeue_io(inode, wb);
412 } else {
413 /*
414 * Writeback blocked by something other than
415 * congestion. Delay the inode for some time to
416 * avoid spinning on the CPU (100% iowait)
417 * retrying writeback of the dirty page/inode
418 * that cannot be performed immediately.
419 */
420 redirty_tail(inode, wb);
421 }
422 } else if (inode->i_state & I_DIRTY) {
423 /*
424 * Filesystems can dirty the inode during writeback operations,
425 * such as delayed allocation during submission or metadata
426 * updates after data IO completion.
427 */
428 redirty_tail(inode, wb);
429 } else {
430 /* The inode is clean. Remove from writeback lists. */
431 list_del_init(&inode->i_wb_list);
432 }
433}
434
435/*
Jan Kara4f8ad652012-05-03 14:48:00 +0200436 * Write out an inode and its dirty pages. Do not update the writeback list
437 * linkage. That is left to the caller. The caller is also responsible for
438 * setting I_SYNC flag and calling inode_sync_complete() to clear it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439 */
440static int
Jan Kara4f8ad652012-05-03 14:48:00 +0200441__writeback_single_inode(struct inode *inode, struct bdi_writeback *wb,
442 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
529 ret = __writeback_single_inode(inode, wb, 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 *
580 * If @only_this_sb is true, then find and write all such
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800581 * inodes. Otherwise write only ones which go sequentially
582 * in reverse order.
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200583 *
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600584 * Return the number of pages and/or inodes written.
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800585 */
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600586static long writeback_sb_inodes(struct super_block *sb,
587 struct bdi_writeback *wb,
588 struct wb_writeback_work *work)
Jens Axboe03ba3782009-09-09 09:08:54 +0200589{
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600590 struct writeback_control wbc = {
591 .sync_mode = work->sync_mode,
592 .tagged_writepages = work->tagged_writepages,
593 .for_kupdate = work->for_kupdate,
594 .for_background = work->for_background,
595 .range_cyclic = work->range_cyclic,
596 .range_start = 0,
597 .range_end = LLONG_MAX,
598 };
599 unsigned long start_time = jiffies;
600 long write_chunk;
601 long wrote = 0; /* count both pages and inodes */
602
Jens Axboe03ba3782009-09-09 09:08:54 +0200603 while (!list_empty(&wb->b_io)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100604 struct inode *inode = wb_inode(wb->b_io.prev);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200605
606 if (inode->i_sb != sb) {
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600607 if (work->sb) {
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200608 /*
609 * We only want to write back data for this
610 * superblock, move all inodes not belonging
611 * to it back onto the dirty list.
612 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600613 redirty_tail(inode, wb);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200614 continue;
615 }
616
617 /*
618 * The inode belongs to a different superblock.
619 * Bounce back to the caller to unpin this and
620 * pin the next superblock.
621 */
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600622 break;
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200623 }
624
Christoph Hellwig9843b762010-10-24 19:40:46 +0200625 /*
Wanpeng Li331cbde2012-06-09 11:10:55 +0800626 * Don't bother with new inodes or inodes being freed, first
627 * kind does not need periodic writeout yet, and for the latter
Christoph Hellwig9843b762010-10-24 19:40:46 +0200628 * kind writeout is handled by the freer.
629 */
Dave Chinner250df6e2011-03-22 22:23:36 +1100630 spin_lock(&inode->i_lock);
Christoph Hellwig9843b762010-10-24 19:40:46 +0200631 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
Dave Chinner250df6e2011-03-22 22:23:36 +1100632 spin_unlock(&inode->i_lock);
Wu Fengguangfcc5c222011-07-11 23:08:50 -0700633 redirty_tail(inode, wb);
Nick Piggin7ef0d732009-03-12 14:31:38 -0700634 continue;
635 }
Jan Karacc1676d2012-05-03 14:47:56 +0200636 if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) {
637 /*
638 * If this inode is locked for writeback and we are not
639 * doing writeback-for-data-integrity, move it to
640 * b_more_io so that writeback can proceed with the
641 * other inodes on s_io.
642 *
643 * We'll have another go at writing back this inode
644 * when we completed a full scan of b_io.
645 */
646 spin_unlock(&inode->i_lock);
647 requeue_io(inode, wb);
648 trace_writeback_sb_inodes_requeue(inode);
649 continue;
650 }
Jan Karaf0d07b72012-05-03 14:47:59 +0200651 spin_unlock(&wb->list_lock);
652
Jan Kara4f8ad652012-05-03 14:48:00 +0200653 /*
654 * We already requeued the inode if it had I_SYNC set and we
655 * are doing WB_SYNC_NONE writeback. So this catches only the
656 * WB_SYNC_ALL case.
657 */
Jan Kara169ebd92012-05-03 14:48:03 +0200658 if (inode->i_state & I_SYNC) {
659 /* Wait for I_SYNC. This function drops i_lock... */
660 inode_sleep_on_writeback(inode);
661 /* Inode may be gone, start again */
Jan Karaead188f2012-06-08 17:07:36 +0200662 spin_lock(&wb->list_lock);
Jan Kara169ebd92012-05-03 14:48:03 +0200663 continue;
664 }
Jan Kara4f8ad652012-05-03 14:48:00 +0200665 inode->i_state |= I_SYNC;
666 spin_unlock(&inode->i_lock);
Jan Kara169ebd92012-05-03 14:48:03 +0200667
Wu Fengguang1a12d8b2010-08-29 13:28:09 -0600668 write_chunk = writeback_chunk_size(wb->bdi, work);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600669 wbc.nr_to_write = write_chunk;
670 wbc.pages_skipped = 0;
Dave Chinner250df6e2011-03-22 22:23:36 +1100671
Jan Kara169ebd92012-05-03 14:48:03 +0200672 /*
673 * We use I_SYNC to pin the inode in memory. While it is set
674 * evict_inode() will wait so the inode cannot be freed.
675 */
Jan Kara4f8ad652012-05-03 14:48:00 +0200676 __writeback_single_inode(inode, wb, &wbc);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600677
678 work->nr_pages -= write_chunk - wbc.nr_to_write;
679 wrote += write_chunk - wbc.nr_to_write;
Jan Kara4f8ad652012-05-03 14:48:00 +0200680 spin_lock(&wb->list_lock);
681 spin_lock(&inode->i_lock);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600682 if (!(inode->i_state & I_DIRTY))
683 wrote++;
Jan Kara4f8ad652012-05-03 14:48:00 +0200684 requeue_inode(inode, wb, &wbc);
685 inode_sync_complete(inode);
Dave Chinner0f1b1fd2011-03-22 22:23:43 +1100686 spin_unlock(&inode->i_lock);
Jan Kara169ebd92012-05-03 14:48:03 +0200687 cond_resched_lock(&wb->list_lock);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600688 /*
689 * bail out to wb_writeback() often enough to check
690 * background threshold and other termination conditions.
691 */
692 if (wrote) {
693 if (time_is_before_jiffies(start_time + HZ / 10UL))
694 break;
695 if (work->nr_pages <= 0)
696 break;
Fengguang Wu8bc3be22008-02-04 22:29:36 -0800697 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698 }
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600699 return wrote;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800700}
Nick Piggin38f21972009-01-06 14:40:25 -0800701
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600702static long __writeback_inodes_wb(struct bdi_writeback *wb,
703 struct wb_writeback_work *work)
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800704{
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600705 unsigned long start_time = jiffies;
706 long wrote = 0;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800707
708 while (!list_empty(&wb->b_io)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100709 struct inode *inode = wb_inode(wb->b_io.prev);
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800710 struct super_block *sb = inode->i_sb;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800711
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000712 if (!grab_super_passive(sb)) {
Wu Fengguang0e995812011-07-29 22:14:35 -0600713 /*
714 * grab_super_passive() may fail consistently due to
715 * s_umount being grabbed by someone else. Don't use
716 * requeue_io() to avoid busy retrying the inode/sb.
717 */
718 redirty_tail(inode, wb);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200719 continue;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800720 }
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600721 wrote += writeback_sb_inodes(sb, wb, work);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200722 drop_super(sb);
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800723
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600724 /* refer to the same tests at the end of writeback_sb_inodes */
725 if (wrote) {
726 if (time_is_before_jiffies(start_time + HZ / 10UL))
727 break;
728 if (work->nr_pages <= 0)
729 break;
730 }
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800731 }
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200732 /* Leave any unwritten inodes on b_io */
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600733 return wrote;
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200734}
735
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600736long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages,
737 enum wb_reason reason)
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200738{
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600739 struct wb_writeback_work work = {
740 .nr_pages = nr_pages,
741 .sync_mode = WB_SYNC_NONE,
742 .range_cyclic = 1,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600743 .reason = reason,
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600744 };
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200745
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600746 spin_lock(&wb->list_lock);
Wu Fengguang424b3512010-07-21 20:11:53 -0600747 if (list_empty(&wb->b_io))
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600748 queue_io(wb, &work);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600749 __writeback_inodes_wb(wb, &work);
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600750 spin_unlock(&wb->list_lock);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200751
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600752 return nr_pages - work.nr_pages;
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200753}
754
Wu Fengguangb00949a2010-11-18 14:38:33 -0600755static bool over_bground_thresh(struct backing_dev_info *bdi)
Jens Axboe03ba3782009-09-09 09:08:54 +0200756{
757 unsigned long background_thresh, dirty_thresh;
758
Wu Fengguang16c40422010-08-11 14:17:39 -0700759 global_dirty_limits(&background_thresh, &dirty_thresh);
Jens Axboe03ba3782009-09-09 09:08:54 +0200760
Wu Fengguangb00949a2010-11-18 14:38:33 -0600761 if (global_page_state(NR_FILE_DIRTY) +
762 global_page_state(NR_UNSTABLE_NFS) > background_thresh)
763 return true;
764
765 if (bdi_stat(bdi, BDI_RECLAIMABLE) >
766 bdi_dirty_limit(bdi, background_thresh))
767 return true;
768
769 return false;
Jens Axboe03ba3782009-09-09 09:08:54 +0200770}
771
772/*
Wu Fengguange98be2d2010-08-29 11:22:30 -0600773 * Called under wb->list_lock. If there are multiple wb per bdi,
774 * only the flusher working on the first wb should do it.
775 */
776static void wb_update_bandwidth(struct bdi_writeback *wb,
777 unsigned long start_time)
778{
Wu Fengguangaf6a3112011-10-03 20:46:17 -0600779 __bdi_update_bandwidth(wb->bdi, 0, 0, 0, 0, 0, start_time);
Wu Fengguange98be2d2010-08-29 11:22:30 -0600780}
781
782/*
Jens Axboe03ba3782009-09-09 09:08:54 +0200783 * Explicit flushing or periodic writeback of "old" data.
784 *
785 * Define "old": the first time one of an inode's pages is dirtied, we mark the
786 * dirtying-time in the inode's address_space. So this periodic writeback code
787 * just walks the superblock inode list, writing back any inodes which are
788 * older than a specific point in time.
789 *
790 * Try to run once per dirty_writeback_interval. But if a writeback event
791 * takes longer than a dirty_writeback_interval interval, then leave a
792 * one-second gap.
793 *
794 * older_than_this takes precedence over nr_to_write. So we'll only write back
795 * all dirty pages if they are all attached to "old" mappings.
796 */
Jens Axboec4a77a62009-09-16 15:18:25 +0200797static long wb_writeback(struct bdi_writeback *wb,
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200798 struct wb_writeback_work *work)
Jens Axboe03ba3782009-09-09 09:08:54 +0200799{
Wu Fengguange98be2d2010-08-29 11:22:30 -0600800 unsigned long wb_start = jiffies;
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600801 long nr_pages = work->nr_pages;
Jens Axboe03ba3782009-09-09 09:08:54 +0200802 unsigned long oldest_jif;
Jan Karaa5989bd2009-09-16 19:22:48 +0200803 struct inode *inode;
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600804 long progress;
Jens Axboe03ba3782009-09-09 09:08:54 +0200805
Wu Fengguange185dda2011-04-23 11:26:07 -0600806 oldest_jif = jiffies;
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600807 work->older_than_this = &oldest_jif;
Jens Axboe03ba3782009-09-09 09:08:54 +0200808
Wu Fengguange8dfc302011-04-21 12:06:32 -0600809 spin_lock(&wb->list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200810 for (;;) {
811 /*
Wu Fengguangd3ddec72009-09-23 20:33:40 +0800812 * Stop writeback when nr_pages has been consumed
Jens Axboe03ba3782009-09-09 09:08:54 +0200813 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200814 if (work->nr_pages <= 0)
Jens Axboe03ba3782009-09-09 09:08:54 +0200815 break;
816
817 /*
Jan Karaaa373cf2011-01-13 15:45:47 -0800818 * Background writeout and kupdate-style writeback may
819 * run forever. Stop them if there is other work to do
820 * so that e.g. sync can proceed. They'll be restarted
821 * after the other works are all done.
822 */
823 if ((work->for_background || work->for_kupdate) &&
824 !list_empty(&wb->bdi->work_list))
825 break;
826
827 /*
Wu Fengguangd3ddec72009-09-23 20:33:40 +0800828 * For background writeout, stop when we are below the
829 * background dirty threshold
Jens Axboe03ba3782009-09-09 09:08:54 +0200830 */
Wu Fengguangb00949a2010-11-18 14:38:33 -0600831 if (work->for_background && !over_bground_thresh(wb->bdi))
Jens Axboe03ba3782009-09-09 09:08:54 +0200832 break;
833
Jan Kara1bc36b62011-10-19 11:44:41 +0200834 /*
835 * Kupdate and background works are special and we want to
836 * include all inodes that need writing. Livelock avoidance is
837 * handled by these works yielding to any other work so we are
838 * safe.
839 */
Wu Fengguangba9aa832010-07-21 20:32:30 -0600840 if (work->for_kupdate) {
841 oldest_jif = jiffies -
842 msecs_to_jiffies(dirty_expire_interval * 10);
Jan Kara1bc36b62011-10-19 11:44:41 +0200843 } else if (work->for_background)
844 oldest_jif = jiffies;
Dave Chinner028c2dd2010-07-07 13:24:07 +1000845
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600846 trace_writeback_start(wb->bdi, work);
Wu Fengguange8dfc302011-04-21 12:06:32 -0600847 if (list_empty(&wb->b_io))
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600848 queue_io(wb, work);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200849 if (work->sb)
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600850 progress = writeback_sb_inodes(work->sb, wb, work);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200851 else
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600852 progress = __writeback_inodes_wb(wb, work);
853 trace_writeback_written(wb->bdi, work);
Dave Chinner028c2dd2010-07-07 13:24:07 +1000854
Wu Fengguange98be2d2010-08-29 11:22:30 -0600855 wb_update_bandwidth(wb, wb_start);
Jens Axboe03ba3782009-09-09 09:08:54 +0200856
857 /*
Jens Axboe71fd05a2009-09-23 19:32:26 +0200858 * Did we write something? Try for more
Wu Fengguange6fb6da2010-07-22 10:23:44 -0600859 *
860 * Dirty inodes are moved to b_io for writeback in batches.
861 * The completion of the current batch does not necessarily
862 * mean the overall work is done. So we keep looping as long
863 * as made some progress on cleaning pages or inodes.
Jens Axboe71fd05a2009-09-23 19:32:26 +0200864 */
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600865 if (progress)
Jens Axboe71fd05a2009-09-23 19:32:26 +0200866 continue;
867 /*
Wu Fengguange6fb6da2010-07-22 10:23:44 -0600868 * No more inodes for IO, bail
Jens Axboe71fd05a2009-09-23 19:32:26 +0200869 */
Wu Fengguangb7a24412010-07-21 22:19:51 -0600870 if (list_empty(&wb->b_more_io))
Jens Axboe03ba3782009-09-09 09:08:54 +0200871 break;
872 /*
Jens Axboe8010c3b2009-09-15 20:04:57 +0200873 * Nothing written. Wait for some inode to
874 * become available for writeback. Otherwise
875 * we'll just busyloop.
Jens Axboe03ba3782009-09-09 09:08:54 +0200876 */
Jens Axboe03ba3782009-09-09 09:08:54 +0200877 if (!list_empty(&wb->b_more_io)) {
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600878 trace_writeback_wait(wb->bdi, work);
Jens Axboe03ba3782009-09-09 09:08:54 +0200879 inode = wb_inode(wb->b_more_io.prev);
Dave Chinner250df6e2011-03-22 22:23:36 +1100880 spin_lock(&inode->i_lock);
Jan Karaf0d07b72012-05-03 14:47:59 +0200881 spin_unlock(&wb->list_lock);
Jan Kara169ebd92012-05-03 14:48:03 +0200882 /* This function drops i_lock... */
883 inode_sleep_on_writeback(inode);
Jan Karaf0d07b72012-05-03 14:47:59 +0200884 spin_lock(&wb->list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200885 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200886 }
Wu Fengguange8dfc302011-04-21 12:06:32 -0600887 spin_unlock(&wb->list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200888
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600889 return nr_pages - work->nr_pages;
Jens Axboe03ba3782009-09-09 09:08:54 +0200890}
891
892/*
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200893 * Return the next wb_writeback_work struct that hasn't been processed yet.
Jens Axboe03ba3782009-09-09 09:08:54 +0200894 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200895static struct wb_writeback_work *
Minchan Kim08852b62010-08-03 12:51:16 +0200896get_next_work_item(struct backing_dev_info *bdi)
Jens Axboe03ba3782009-09-09 09:08:54 +0200897{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200898 struct wb_writeback_work *work = NULL;
Jens Axboe03ba3782009-09-09 09:08:54 +0200899
Artem Bityutskiy64677162010-07-25 14:29:22 +0300900 spin_lock_bh(&bdi->wb_lock);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200901 if (!list_empty(&bdi->work_list)) {
902 work = list_entry(bdi->work_list.next,
903 struct wb_writeback_work, list);
904 list_del_init(&work->list);
Jens Axboe03ba3782009-09-09 09:08:54 +0200905 }
Artem Bityutskiy64677162010-07-25 14:29:22 +0300906 spin_unlock_bh(&bdi->wb_lock);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200907 return work;
Jens Axboe03ba3782009-09-09 09:08:54 +0200908}
909
Linus Torvaldscdf01dd2010-10-30 08:55:52 -0700910/*
911 * Add in the number of potentially dirty inodes, because each inode
912 * write can dirty pagecache in the underlying blockdev.
913 */
914static unsigned long get_nr_dirty_pages(void)
915{
916 return global_page_state(NR_FILE_DIRTY) +
917 global_page_state(NR_UNSTABLE_NFS) +
918 get_nr_dirty_inodes();
919}
920
Jan Kara65850272011-01-13 15:45:44 -0800921static long wb_check_background_flush(struct bdi_writeback *wb)
922{
Wu Fengguangb00949a2010-11-18 14:38:33 -0600923 if (over_bground_thresh(wb->bdi)) {
Jan Kara65850272011-01-13 15:45:44 -0800924
925 struct wb_writeback_work work = {
926 .nr_pages = LONG_MAX,
927 .sync_mode = WB_SYNC_NONE,
928 .for_background = 1,
929 .range_cyclic = 1,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600930 .reason = WB_REASON_BACKGROUND,
Jan Kara65850272011-01-13 15:45:44 -0800931 };
932
933 return wb_writeback(wb, &work);
934 }
935
936 return 0;
937}
938
Jens Axboe03ba3782009-09-09 09:08:54 +0200939static long wb_check_old_data_flush(struct bdi_writeback *wb)
940{
941 unsigned long expired;
942 long nr_pages;
943
Jens Axboe69b62d02010-05-17 12:51:03 +0200944 /*
945 * When set to zero, disable periodic writeback
946 */
947 if (!dirty_writeback_interval)
948 return 0;
949
Jens Axboe03ba3782009-09-09 09:08:54 +0200950 expired = wb->last_old_flush +
951 msecs_to_jiffies(dirty_writeback_interval * 10);
952 if (time_before(jiffies, expired))
953 return 0;
954
955 wb->last_old_flush = jiffies;
Linus Torvaldscdf01dd2010-10-30 08:55:52 -0700956 nr_pages = get_nr_dirty_pages();
Jens Axboe03ba3782009-09-09 09:08:54 +0200957
Jens Axboec4a77a62009-09-16 15:18:25 +0200958 if (nr_pages) {
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200959 struct wb_writeback_work work = {
Jens Axboec4a77a62009-09-16 15:18:25 +0200960 .nr_pages = nr_pages,
961 .sync_mode = WB_SYNC_NONE,
962 .for_kupdate = 1,
963 .range_cyclic = 1,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600964 .reason = WB_REASON_PERIODIC,
Jens Axboec4a77a62009-09-16 15:18:25 +0200965 };
966
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200967 return wb_writeback(wb, &work);
Jens Axboec4a77a62009-09-16 15:18:25 +0200968 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200969
970 return 0;
971}
972
973/*
974 * Retrieve work items and do the writeback they describe
975 */
976long wb_do_writeback(struct bdi_writeback *wb, int force_wait)
977{
978 struct backing_dev_info *bdi = wb->bdi;
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200979 struct wb_writeback_work *work;
Jens Axboec4a77a62009-09-16 15:18:25 +0200980 long wrote = 0;
Jens Axboe03ba3782009-09-09 09:08:54 +0200981
Jan Kara81d73a32010-08-11 14:17:44 -0700982 set_bit(BDI_writeback_running, &wb->bdi->state);
Minchan Kim08852b62010-08-03 12:51:16 +0200983 while ((work = get_next_work_item(bdi)) != NULL) {
Jens Axboe03ba3782009-09-09 09:08:54 +0200984 /*
985 * Override sync mode, in case we must wait for completion
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200986 * because this thread is exiting now.
Jens Axboe03ba3782009-09-09 09:08:54 +0200987 */
988 if (force_wait)
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200989 work->sync_mode = WB_SYNC_ALL;
990
Dave Chinner455b2862010-07-07 13:24:06 +1000991 trace_writeback_exec(bdi, work);
992
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200993 wrote += wb_writeback(wb, work);
Jens Axboe03ba3782009-09-09 09:08:54 +0200994
995 /*
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200996 * Notify the caller of completion if this is a synchronous
997 * work item, otherwise just free it.
Jens Axboe03ba3782009-09-09 09:08:54 +0200998 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200999 if (work->done)
1000 complete(work->done);
1001 else
1002 kfree(work);
Jens Axboe03ba3782009-09-09 09:08:54 +02001003 }
1004
1005 /*
1006 * Check for periodic writeback, kupdated() style
1007 */
1008 wrote += wb_check_old_data_flush(wb);
Jan Kara65850272011-01-13 15:45:44 -08001009 wrote += wb_check_background_flush(wb);
Jan Kara81d73a32010-08-11 14:17:44 -07001010 clear_bit(BDI_writeback_running, &wb->bdi->state);
Jens Axboe03ba3782009-09-09 09:08:54 +02001011
1012 return wrote;
1013}
1014
1015/*
1016 * Handle writeback of dirty data for the device backed by this bdi. Also
1017 * wakes up periodically and does kupdated style flushing.
1018 */
Christoph Hellwig08243902010-06-19 23:08:22 +02001019int bdi_writeback_thread(void *data)
Jens Axboe03ba3782009-09-09 09:08:54 +02001020{
Christoph Hellwig08243902010-06-19 23:08:22 +02001021 struct bdi_writeback *wb = data;
1022 struct backing_dev_info *bdi = wb->bdi;
Jens Axboe03ba3782009-09-09 09:08:54 +02001023 long pages_written;
1024
Peter Zijlstra766f9162010-10-26 14:22:45 -07001025 current->flags |= PF_SWAPWRITE;
Christoph Hellwig08243902010-06-19 23:08:22 +02001026 set_freezable();
Artem Bityutskiyecd58402010-07-25 14:29:18 +03001027 wb->last_active = jiffies;
Christoph Hellwig08243902010-06-19 23:08:22 +02001028
1029 /*
1030 * Our parent may run at a different priority, just set us to normal
1031 */
1032 set_user_nice(current, 0);
1033
Dave Chinner455b2862010-07-07 13:24:06 +10001034 trace_writeback_thread_start(bdi);
1035
Tejun Heo8a32c442011-11-21 12:32:23 -08001036 while (!kthread_freezable_should_stop(NULL)) {
Artem Bityutskiy64677162010-07-25 14:29:22 +03001037 /*
1038 * Remove own delayed wake-up timer, since we are already awake
1039 * and we'll take care of the preriodic write-back.
1040 */
1041 del_timer(&wb->wakeup_timer);
1042
Jens Axboe03ba3782009-09-09 09:08:54 +02001043 pages_written = wb_do_writeback(wb, 0);
1044
Dave Chinner455b2862010-07-07 13:24:06 +10001045 trace_writeback_pages_written(pages_written);
Jens Axboe03ba3782009-09-09 09:08:54 +02001046
Jens Axboe03ba3782009-09-09 09:08:54 +02001047 if (pages_written)
Artem Bityutskiyecd58402010-07-25 14:29:18 +03001048 wb->last_active = jiffies;
Jens Axboe03ba3782009-09-09 09:08:54 +02001049
Artem Bityutskiy297252c2010-07-25 14:29:15 +03001050 set_current_state(TASK_INTERRUPTIBLE);
J. Bruce Fieldsb76b4012010-08-28 08:52:10 +02001051 if (!list_empty(&bdi->work_list) || kthread_should_stop()) {
Artem Bityutskiy297252c2010-07-25 14:29:15 +03001052 __set_current_state(TASK_RUNNING);
1053 continue;
Jens Axboe03ba3782009-09-09 09:08:54 +02001054 }
1055
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001056 if (wb_has_dirty_io(wb) && dirty_writeback_interval)
Artem Bityutskiyfff5b852010-07-25 14:29:20 +03001057 schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10));
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001058 else {
1059 /*
1060 * We have nothing to do, so can go sleep without any
1061 * timeout and save power. When a work is queued or
1062 * something is made dirty - we will be woken up.
1063 */
Artem Bityutskiy297252c2010-07-25 14:29:15 +03001064 schedule();
Jens Axboef9eadbb2010-05-18 14:31:45 +02001065 }
Jens Axboe03ba3782009-09-09 09:08:54 +02001066 }
1067
Artem Bityutskiyfff5b852010-07-25 14:29:20 +03001068 /* Flush any work that raced with us exiting */
Christoph Hellwig08243902010-06-19 23:08:22 +02001069 if (!list_empty(&bdi->work_list))
1070 wb_do_writeback(wb, 1);
Dave Chinner455b2862010-07-07 13:24:06 +10001071
1072 trace_writeback_thread_stop(bdi);
Jens Axboe03ba3782009-09-09 09:08:54 +02001073 return 0;
1074}
1075
Christoph Hellwig08243902010-06-19 23:08:22 +02001076
Jens Axboe03ba3782009-09-09 09:08:54 +02001077/*
Jens Axboe03ba3782009-09-09 09:08:54 +02001078 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
1079 * the whole world.
1080 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001081void wakeup_flusher_threads(long nr_pages, enum wb_reason reason)
Jens Axboe03ba3782009-09-09 09:08:54 +02001082{
Christoph Hellwigb8c2f342010-06-08 18:15:07 +02001083 struct backing_dev_info *bdi;
Christoph Hellwigb8c2f342010-06-08 18:15:07 +02001084
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001085 if (!nr_pages) {
1086 nr_pages = global_page_state(NR_FILE_DIRTY) +
Jens Axboe03ba3782009-09-09 09:08:54 +02001087 global_page_state(NR_UNSTABLE_NFS);
Christoph Hellwigb8c2f342010-06-08 18:15:07 +02001088 }
1089
1090 rcu_read_lock();
1091 list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) {
1092 if (!bdi_has_dirty_io(bdi))
1093 continue;
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001094 __bdi_start_writeback(bdi, nr_pages, false, reason);
Christoph Hellwigb8c2f342010-06-08 18:15:07 +02001095 }
1096 rcu_read_unlock();
Jens Axboe03ba3782009-09-09 09:08:54 +02001097}
1098
1099static noinline void block_dump___mark_inode_dirty(struct inode *inode)
1100{
1101 if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) {
1102 struct dentry *dentry;
1103 const char *name = "?";
1104
1105 dentry = d_find_alias(inode);
1106 if (dentry) {
1107 spin_lock(&dentry->d_lock);
1108 name = (const char *) dentry->d_name.name;
1109 }
1110 printk(KERN_DEBUG
1111 "%s(%d): dirtied inode %lu (%s) on %s\n",
1112 current->comm, task_pid_nr(current), inode->i_ino,
1113 name, inode->i_sb->s_id);
1114 if (dentry) {
1115 spin_unlock(&dentry->d_lock);
1116 dput(dentry);
1117 }
1118 }
1119}
1120
1121/**
1122 * __mark_inode_dirty - internal function
1123 * @inode: inode to mark
1124 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
1125 * Mark an inode as dirty. Callers should use mark_inode_dirty or
1126 * mark_inode_dirty_sync.
1127 *
1128 * Put the inode on the super block's dirty list.
1129 *
1130 * CAREFUL! We mark it dirty unconditionally, but move it onto the
1131 * dirty list only if it is hashed or if it refers to a blockdev.
1132 * If it was not hashed, it will never be added to the dirty list
1133 * even if it is later hashed, as it will have been marked dirty already.
1134 *
1135 * In short, make sure you hash any inodes _before_ you start marking
1136 * them dirty.
1137 *
Jens Axboe03ba3782009-09-09 09:08:54 +02001138 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
1139 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
1140 * the kernel-internal blockdev inode represents the dirtying time of the
1141 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
1142 * page->mapping->host, so the page-dirtying time is recorded in the internal
1143 * blockdev inode.
1144 */
1145void __mark_inode_dirty(struct inode *inode, int flags)
1146{
1147 struct super_block *sb = inode->i_sb;
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001148 struct backing_dev_info *bdi = NULL;
Jens Axboe03ba3782009-09-09 09:08:54 +02001149
1150 /*
1151 * Don't do this for I_DIRTY_PAGES - that doesn't actually
1152 * dirty the inode itself
1153 */
1154 if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
1155 if (sb->s_op->dirty_inode)
Christoph Hellwigaa385722011-05-27 06:53:02 -04001156 sb->s_op->dirty_inode(inode, flags);
Jens Axboe03ba3782009-09-09 09:08:54 +02001157 }
1158
1159 /*
1160 * make sure that changes are seen by all cpus before we test i_state
1161 * -- mikulas
1162 */
1163 smp_mb();
1164
1165 /* avoid the locking if we can */
1166 if ((inode->i_state & flags) == flags)
1167 return;
1168
1169 if (unlikely(block_dump))
1170 block_dump___mark_inode_dirty(inode);
1171
Dave Chinner250df6e2011-03-22 22:23:36 +11001172 spin_lock(&inode->i_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001173 if ((inode->i_state & flags) != flags) {
1174 const int was_dirty = inode->i_state & I_DIRTY;
1175
1176 inode->i_state |= flags;
1177
1178 /*
1179 * If the inode is being synced, just update its dirty state.
1180 * The unlocker will place the inode on the appropriate
1181 * superblock list, based upon its state.
1182 */
1183 if (inode->i_state & I_SYNC)
Dave Chinner250df6e2011-03-22 22:23:36 +11001184 goto out_unlock_inode;
Jens Axboe03ba3782009-09-09 09:08:54 +02001185
1186 /*
1187 * Only add valid (hashed) inodes to the superblock's
1188 * dirty list. Add blockdev inodes as well.
1189 */
1190 if (!S_ISBLK(inode->i_mode)) {
Al Viro1d3382cb2010-10-23 15:19:20 -04001191 if (inode_unhashed(inode))
Dave Chinner250df6e2011-03-22 22:23:36 +11001192 goto out_unlock_inode;
Jens Axboe03ba3782009-09-09 09:08:54 +02001193 }
Al Viroa4ffdde2010-06-02 17:38:30 -04001194 if (inode->i_state & I_FREEING)
Dave Chinner250df6e2011-03-22 22:23:36 +11001195 goto out_unlock_inode;
Jens Axboe03ba3782009-09-09 09:08:54 +02001196
1197 /*
1198 * If the inode was already on b_dirty/b_io/b_more_io, don't
1199 * reposition it (that would break b_dirty time-ordering).
1200 */
1201 if (!was_dirty) {
Dave Chinnera66979a2011-03-22 22:23:41 +11001202 bool wakeup_bdi = false;
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001203 bdi = inode_to_bdi(inode);
Jens Axboe500b0672009-09-09 09:10:25 +02001204
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001205 if (bdi_cap_writeback_dirty(bdi)) {
1206 WARN(!test_bit(BDI_registered, &bdi->state),
1207 "bdi-%s not registered\n", bdi->name);
1208
1209 /*
1210 * If this is the first dirty inode for this
1211 * bdi, we have to wake-up the corresponding
1212 * bdi thread to make sure background
1213 * write-back happens later.
1214 */
1215 if (!wb_has_dirty_io(&bdi->wb))
1216 wakeup_bdi = true;
Jens Axboe500b0672009-09-09 09:10:25 +02001217 }
Jens Axboe03ba3782009-09-09 09:08:54 +02001218
Dave Chinnera66979a2011-03-22 22:23:41 +11001219 spin_unlock(&inode->i_lock);
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001220 spin_lock(&bdi->wb.list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001221 inode->dirtied_when = jiffies;
Nick Piggin7ccf19a2010-10-21 11:49:30 +11001222 list_move(&inode->i_wb_list, &bdi->wb.b_dirty);
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001223 spin_unlock(&bdi->wb.list_lock);
Dave Chinnera66979a2011-03-22 22:23:41 +11001224
1225 if (wakeup_bdi)
1226 bdi_wakeup_thread_delayed(bdi);
1227 return;
Jens Axboe03ba3782009-09-09 09:08:54 +02001228 }
1229 }
Dave Chinner250df6e2011-03-22 22:23:36 +11001230out_unlock_inode:
1231 spin_unlock(&inode->i_lock);
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001232
Jens Axboe03ba3782009-09-09 09:08:54 +02001233}
1234EXPORT_SYMBOL(__mark_inode_dirty);
1235
Jens Axboeb6e51312009-09-16 15:13:54 +02001236static void wait_sb_inodes(struct super_block *sb)
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001237{
Jens Axboe03ba3782009-09-09 09:08:54 +02001238 struct inode *inode, *old_inode = NULL;
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001239
Jens Axboe03ba3782009-09-09 09:08:54 +02001240 /*
1241 * We need to be protected against the filesystem going from
1242 * r/o to r/w or vice versa.
1243 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001244 WARN_ON(!rwsem_is_locked(&sb->s_umount));
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001245
Dave Chinner55fa6092011-03-22 22:23:40 +11001246 spin_lock(&inode_sb_list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001247
1248 /*
1249 * Data integrity sync. Must wait for all pages under writeback,
1250 * because there may have been pages dirtied before our sync
1251 * call, but which had writeout started before we write it out.
1252 * In which case, the inode may not be on the dirty list, but
1253 * we still have to wait for that writeout.
1254 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001255 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
Dave Chinner250df6e2011-03-22 22:23:36 +11001256 struct address_space *mapping = inode->i_mapping;
Jens Axboe03ba3782009-09-09 09:08:54 +02001257
Dave Chinner250df6e2011-03-22 22:23:36 +11001258 spin_lock(&inode->i_lock);
1259 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
1260 (mapping->nrpages == 0)) {
1261 spin_unlock(&inode->i_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001262 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +11001263 }
Jens Axboe03ba3782009-09-09 09:08:54 +02001264 __iget(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +11001265 spin_unlock(&inode->i_lock);
Dave Chinner55fa6092011-03-22 22:23:40 +11001266 spin_unlock(&inode_sb_list_lock);
1267
Jens Axboe03ba3782009-09-09 09:08:54 +02001268 /*
Dave Chinner55fa6092011-03-22 22:23:40 +11001269 * We hold a reference to 'inode' so it couldn't have been
1270 * removed from s_inodes list while we dropped the
1271 * inode_sb_list_lock. We cannot iput the inode now as we can
1272 * be holding the last reference and we cannot iput it under
1273 * inode_sb_list_lock. So we keep the reference and iput it
1274 * later.
Jens Axboe03ba3782009-09-09 09:08:54 +02001275 */
1276 iput(old_inode);
1277 old_inode = inode;
1278
1279 filemap_fdatawait(mapping);
1280
1281 cond_resched();
Nick Piggin38f21972009-01-06 14:40:25 -08001282
Dave Chinner55fa6092011-03-22 22:23:40 +11001283 spin_lock(&inode_sb_list_lock);
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001284 }
Dave Chinner55fa6092011-03-22 22:23:40 +11001285 spin_unlock(&inode_sb_list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001286 iput(old_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287}
1288
Jens Axboed8a85592009-09-02 12:34:32 +02001289/**
Chris Mason3259f8b2010-10-29 11:16:17 -04001290 * writeback_inodes_sb_nr - writeback dirty inodes from given super_block
Jens Axboed8a85592009-09-02 12:34:32 +02001291 * @sb: the superblock
Chris Mason3259f8b2010-10-29 11:16:17 -04001292 * @nr: the number of pages to write
Marcos Paulo de Souza786228a2011-11-23 20:56:45 +08001293 * @reason: reason why some writeback work initiated
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294 *
Jens Axboed8a85592009-09-02 12:34:32 +02001295 * Start writeback on some inodes on this super_block. No guarantees are made
1296 * on how many (if any) will be written, and this function does not wait
Chris Mason3259f8b2010-10-29 11:16:17 -04001297 * for IO completion of submitted IO.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001299void writeback_inodes_sb_nr(struct super_block *sb,
1300 unsigned long nr,
1301 enum wb_reason reason)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001303 DECLARE_COMPLETION_ONSTACK(done);
1304 struct wb_writeback_work work = {
Wu Fengguang6e6938b2010-06-06 10:38:15 -06001305 .sb = sb,
1306 .sync_mode = WB_SYNC_NONE,
1307 .tagged_writepages = 1,
1308 .done = &done,
1309 .nr_pages = nr,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001310 .reason = reason,
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001311 };
Jens Axboe0e3c9a22010-06-01 11:08:43 +02001312
Jan Kara6eedc702012-07-03 16:45:27 +02001313 if (sb->s_bdi == &noop_backing_dev_info)
1314 return;
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001315 WARN_ON(!rwsem_is_locked(&sb->s_umount));
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001316 bdi_queue_work(sb->s_bdi, &work);
1317 wait_for_completion(&done);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318}
Chris Mason3259f8b2010-10-29 11:16:17 -04001319EXPORT_SYMBOL(writeback_inodes_sb_nr);
1320
1321/**
1322 * writeback_inodes_sb - writeback dirty inodes from given super_block
1323 * @sb: the superblock
Marcos Paulo de Souza786228a2011-11-23 20:56:45 +08001324 * @reason: reason why some writeback work was initiated
Chris Mason3259f8b2010-10-29 11:16:17 -04001325 *
1326 * Start writeback on some inodes on this super_block. No guarantees are made
1327 * on how many (if any) will be written, and this function does not wait
1328 * for IO completion of submitted IO.
1329 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001330void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason)
Chris Mason3259f8b2010-10-29 11:16:17 -04001331{
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001332 return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason);
Chris Mason3259f8b2010-10-29 11:16:17 -04001333}
Jens Axboed8a85592009-09-02 12:34:32 +02001334EXPORT_SYMBOL(writeback_inodes_sb);
1335
1336/**
Eric Sandeen17bd55d2009-12-23 07:57:07 -05001337 * writeback_inodes_sb_if_idle - start writeback if none underway
1338 * @sb: the superblock
Marcos Paulo de Souza786228a2011-11-23 20:56:45 +08001339 * @reason: reason why some writeback work was initiated
Eric Sandeen17bd55d2009-12-23 07:57:07 -05001340 *
1341 * Invoke writeback_inodes_sb if no writeback is currently underway.
1342 * Returns 1 if writeback was started, 0 if not.
1343 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001344int writeback_inodes_sb_if_idle(struct super_block *sb, enum wb_reason reason)
Eric Sandeen17bd55d2009-12-23 07:57:07 -05001345{
1346 if (!writeback_in_progress(sb->s_bdi)) {
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001347 down_read(&sb->s_umount);
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001348 writeback_inodes_sb(sb, reason);
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001349 up_read(&sb->s_umount);
Eric Sandeen17bd55d2009-12-23 07:57:07 -05001350 return 1;
1351 } else
1352 return 0;
1353}
1354EXPORT_SYMBOL(writeback_inodes_sb_if_idle);
1355
1356/**
Fengguang Wuc097b2c2012-03-05 15:08:06 -08001357 * writeback_inodes_sb_nr_if_idle - start writeback if none underway
Chris Mason3259f8b2010-10-29 11:16:17 -04001358 * @sb: the superblock
1359 * @nr: the number of pages to write
Marcos Paulo de Souza786228a2011-11-23 20:56:45 +08001360 * @reason: reason why some writeback work was initiated
Chris Mason3259f8b2010-10-29 11:16:17 -04001361 *
1362 * Invoke writeback_inodes_sb if no writeback is currently underway.
1363 * Returns 1 if writeback was started, 0 if not.
1364 */
1365int writeback_inodes_sb_nr_if_idle(struct super_block *sb,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001366 unsigned long nr,
1367 enum wb_reason reason)
Chris Mason3259f8b2010-10-29 11:16:17 -04001368{
1369 if (!writeback_in_progress(sb->s_bdi)) {
1370 down_read(&sb->s_umount);
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001371 writeback_inodes_sb_nr(sb, nr, reason);
Chris Mason3259f8b2010-10-29 11:16:17 -04001372 up_read(&sb->s_umount);
1373 return 1;
1374 } else
1375 return 0;
1376}
1377EXPORT_SYMBOL(writeback_inodes_sb_nr_if_idle);
1378
1379/**
Jens Axboed8a85592009-09-02 12:34:32 +02001380 * sync_inodes_sb - sync sb inode pages
1381 * @sb: the superblock
1382 *
1383 * This function writes and waits on any dirty inode belonging to this
Stefan Hajnoczicb9ef8d2011-01-13 15:47:26 -08001384 * super_block.
Jens Axboed8a85592009-09-02 12:34:32 +02001385 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001386void sync_inodes_sb(struct super_block *sb)
Jens Axboed8a85592009-09-02 12:34:32 +02001387{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001388 DECLARE_COMPLETION_ONSTACK(done);
1389 struct wb_writeback_work work = {
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001390 .sb = sb,
1391 .sync_mode = WB_SYNC_ALL,
1392 .nr_pages = LONG_MAX,
1393 .range_cyclic = 0,
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001394 .done = &done,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001395 .reason = WB_REASON_SYNC,
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001396 };
1397
Jan Kara6eedc702012-07-03 16:45:27 +02001398 /* Nothing to do? */
1399 if (sb->s_bdi == &noop_backing_dev_info)
1400 return;
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001401 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1402
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001403 bdi_queue_work(sb->s_bdi, &work);
1404 wait_for_completion(&done);
1405
Jens Axboeb6e51312009-09-16 15:13:54 +02001406 wait_sb_inodes(sb);
Jens Axboed8a85592009-09-02 12:34:32 +02001407}
1408EXPORT_SYMBOL(sync_inodes_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410/**
Andrea Arcangeli7f04c262005-10-30 15:03:05 -08001411 * write_inode_now - write an inode to disk
1412 * @inode: inode to write to disk
1413 * @sync: whether the write should be synchronous or not
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414 *
Andrea Arcangeli7f04c262005-10-30 15:03:05 -08001415 * This function commits an inode to disk immediately if it is dirty. This is
1416 * primarily needed by knfsd.
1417 *
1418 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420int write_inode_now(struct inode *inode, int sync)
1421{
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001422 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423 struct writeback_control wbc = {
1424 .nr_to_write = LONG_MAX,
Mike Galbraith18914b12008-02-08 04:20:23 -08001425 .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -07001426 .range_start = 0,
1427 .range_end = LLONG_MAX,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428 };
1429
1430 if (!mapping_cap_writeback_dirty(inode->i_mapping))
Andrew Morton49364ce2005-11-07 00:59:15 -08001431 wbc.nr_to_write = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432
1433 might_sleep();
Jan Kara4f8ad652012-05-03 14:48:00 +02001434 return writeback_single_inode(inode, wb, &wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435}
1436EXPORT_SYMBOL(write_inode_now);
1437
1438/**
1439 * sync_inode - write an inode and its pages to disk.
1440 * @inode: the inode to sync
1441 * @wbc: controls the writeback mode
1442 *
1443 * sync_inode() will write an inode and its pages to disk. It will also
1444 * correctly update the inode on its superblock's dirty inode lists and will
1445 * update inode->i_state.
1446 *
1447 * The caller must have a ref on the inode.
1448 */
1449int sync_inode(struct inode *inode, struct writeback_control *wbc)
1450{
Jan Kara4f8ad652012-05-03 14:48:00 +02001451 return writeback_single_inode(inode, &inode_to_bdi(inode)->wb, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452}
1453EXPORT_SYMBOL(sync_inode);
Christoph Hellwigc37650162010-10-06 10:48:20 +02001454
1455/**
Andrew Mortonc691b9d2011-01-13 15:45:48 -08001456 * sync_inode_metadata - write an inode to disk
Christoph Hellwigc37650162010-10-06 10:48:20 +02001457 * @inode: the inode to sync
1458 * @wait: wait for I/O to complete.
1459 *
Andrew Mortonc691b9d2011-01-13 15:45:48 -08001460 * Write an inode to disk and adjust its dirty state after completion.
Christoph Hellwigc37650162010-10-06 10:48:20 +02001461 *
1462 * Note: only writes the actual inode, no associated data or other metadata.
1463 */
1464int sync_inode_metadata(struct inode *inode, int wait)
1465{
1466 struct writeback_control wbc = {
1467 .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE,
1468 .nr_to_write = 0, /* metadata-only */
1469 };
1470
1471 return sync_inode(inode, &wbc);
1472}
1473EXPORT_SYMBOL(sync_inode_metadata);