blob: c5102969360071a39b2f3a05a34861cd82217d54 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * fs/fs-writeback.c
3 *
4 * Copyright (C) 2002, Linus Torvalds.
5 *
6 * Contains all the functions related to writing back and waiting
7 * upon dirty inodes against superblocks, and writing back dirty
8 * pages against inodes. ie: data writeback. Writeout of the
9 * inode itself is not handled here.
10 *
Francois Camie1f8e872008-10-15 22:01:59 -070011 * 10Apr2002 Andrew Morton
Linus Torvalds1da177e2005-04-16 15:20:36 -070012 * Split out of fs/inode.c
13 * Additions for address_space-based writeback
14 */
15
16#include <linux/kernel.h>
Jens Axboef5ff8422007-09-21 09:19:54 +020017#include <linux/module.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include <linux/spinlock.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090019#include <linux/slab.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/sched.h>
21#include <linux/fs.h>
22#include <linux/mm.h>
Jens Axboe03ba3782009-09-09 09:08:54 +020023#include <linux/kthread.h>
24#include <linux/freezer.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/writeback.h>
26#include <linux/blkdev.h>
27#include <linux/backing-dev.h>
28#include <linux/buffer_head.h>
Dave Chinner455b2862010-07-07 13:24:06 +100029#include <linux/tracepoint.h>
David Howells07f3f052006-09-30 20:52:18 +020030#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070031
Jens Axboed0bceac2009-05-18 08:20:32 +020032/*
Jens Axboec4a77a62009-09-16 15:18:25 +020033 * Passed into wb_writeback(), essentially a subset of writeback_control
34 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +020035struct wb_writeback_work {
Jens Axboec4a77a62009-09-16 15:18:25 +020036 long nr_pages;
37 struct super_block *sb;
Wu Fengguangd46db3d2011-05-04 19:54:37 -060038 unsigned long *older_than_this;
Jens Axboec4a77a62009-09-16 15:18:25 +020039 enum writeback_sync_modes sync_mode;
Wu Fengguang6e6938b2010-06-06 10:38:15 -060040 unsigned int tagged_writepages:1;
H Hartley Sweeten52957fe2010-04-01 20:36:30 -050041 unsigned int for_kupdate:1;
42 unsigned int range_cyclic:1;
43 unsigned int for_background:1;
Jens Axboec4a77a62009-09-16 15:18:25 +020044
Jens Axboe8010c3b2009-09-15 20:04:57 +020045 struct list_head list; /* pending work list */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +020046 struct completion *done; /* set if the caller waits */
Jens Axboe03ba3782009-09-09 09:08:54 +020047};
48
Dave Chinner455b2862010-07-07 13:24:06 +100049/*
50 * Include the creation of the trace points after defining the
51 * wb_writeback_work structure so that the definition remains local to this
52 * file.
53 */
54#define CREATE_TRACE_POINTS
55#include <trace/events/writeback.h>
56
Dave Chinner455b2862010-07-07 13:24:06 +100057/*
58 * We don't actually have pdflush, but this one is exported though /proc...
59 */
60int nr_pdflush_threads;
61
Adrian Bunkf11b00f2008-04-29 00:58:56 -070062/**
63 * writeback_in_progress - determine whether there is writeback in progress
64 * @bdi: the device's backing_dev_info structure.
65 *
Jens Axboe03ba3782009-09-09 09:08:54 +020066 * Determine whether there is writeback waiting to be handled against a
67 * backing device.
Adrian Bunkf11b00f2008-04-29 00:58:56 -070068 */
69int writeback_in_progress(struct backing_dev_info *bdi)
70{
Jan Kara81d73a32010-08-11 14:17:44 -070071 return test_bit(BDI_writeback_running, &bdi->state);
Adrian Bunkf11b00f2008-04-29 00:58:56 -070072}
73
Jan Kara692ebd12010-09-21 11:51:01 +020074static inline struct backing_dev_info *inode_to_bdi(struct inode *inode)
75{
76 struct super_block *sb = inode->i_sb;
Jan Kara692ebd12010-09-21 11:51:01 +020077
Christoph Hellwigaaead252010-10-04 14:25:33 +020078 if (strcmp(sb->s_type->name, "bdev") == 0)
79 return inode->i_mapping->backing_dev_info;
80
81 return sb->s_bdi;
Jan Kara692ebd12010-09-21 11:51:01 +020082}
83
Nick Piggin7ccf19a2010-10-21 11:49:30 +110084static inline struct inode *wb_inode(struct list_head *head)
85{
86 return list_entry(head, struct inode, i_wb_list);
87}
88
Jan Kara65850272011-01-13 15:45:44 -080089/* Wakeup flusher thread or forker thread to fork it. Requires bdi->wb_lock. */
90static void bdi_wakeup_flusher(struct backing_dev_info *bdi)
Nick Piggin4195f732009-05-28 09:01:15 +020091{
Artem Bityutskiyfff5b852010-07-25 14:29:20 +030092 if (bdi->wb.task) {
93 wake_up_process(bdi->wb.task);
94 } else {
95 /*
96 * The bdi thread isn't there, wake up the forker thread which
97 * will create and run it.
98 */
Jens Axboe03ba3782009-09-09 09:08:54 +020099 wake_up_process(default_backing_dev_info.wb.task);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100 }
Jan Kara65850272011-01-13 15:45:44 -0800101}
102
103static void bdi_queue_work(struct backing_dev_info *bdi,
104 struct wb_writeback_work *work)
105{
106 trace_writeback_queue(bdi, work);
107
108 spin_lock_bh(&bdi->wb_lock);
109 list_add_tail(&work->list, &bdi->work_list);
110 if (!bdi->wb.task)
111 trace_writeback_nothread(bdi, work);
112 bdi_wakeup_flusher(bdi);
Artem Bityutskiy64677162010-07-25 14:29:22 +0300113 spin_unlock_bh(&bdi->wb_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114}
115
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200116static void
117__bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages,
Jan Kara65850272011-01-13 15:45:44 -0800118 bool range_cyclic)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200120 struct wb_writeback_work *work;
Jens Axboe03ba3782009-09-09 09:08:54 +0200121
Jens Axboebcddc3f2009-09-13 20:07:36 +0200122 /*
123 * This is WB_SYNC_NONE writeback, so if allocation fails just
124 * wakeup the thread for old dirty data writeback
125 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200126 work = kzalloc(sizeof(*work), GFP_ATOMIC);
127 if (!work) {
Dave Chinner455b2862010-07-07 13:24:06 +1000128 if (bdi->wb.task) {
129 trace_writeback_nowork(bdi);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200130 wake_up_process(bdi->wb.task);
Dave Chinner455b2862010-07-07 13:24:06 +1000131 }
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200132 return;
Jens Axboebcddc3f2009-09-13 20:07:36 +0200133 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200134
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200135 work->sync_mode = WB_SYNC_NONE;
136 work->nr_pages = nr_pages;
137 work->range_cyclic = range_cyclic;
Christoph Hellwigf0fad8a2009-09-11 09:47:56 +0200138
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200139 bdi_queue_work(bdi, work);
Jens Axboeb6e51312009-09-16 15:13:54 +0200140}
141
142/**
143 * bdi_start_writeback - start writeback
144 * @bdi: the backing device to write from
145 * @nr_pages: the number of pages to write
146 *
147 * Description:
148 * This does WB_SYNC_NONE opportunistic writeback. The IO is only
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300149 * started when this function returns, we make no guarantees on
Jens Axboe0e3c9a22010-06-01 11:08:43 +0200150 * completion. Caller need not hold sb s_umount semaphore.
Jens Axboeb6e51312009-09-16 15:13:54 +0200151 *
152 */
Christoph Hellwigc5444192010-06-08 18:15:15 +0200153void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages)
Jens Axboeb6e51312009-09-16 15:13:54 +0200154{
Jan Kara65850272011-01-13 15:45:44 -0800155 __bdi_start_writeback(bdi, nr_pages, true);
Christoph Hellwigc5444192010-06-08 18:15:15 +0200156}
Wu Fengguangd3ddec72009-09-23 20:33:40 +0800157
Christoph Hellwigc5444192010-06-08 18:15:15 +0200158/**
159 * bdi_start_background_writeback - start background writeback
160 * @bdi: the backing device to write from
161 *
162 * Description:
Jan Kara65850272011-01-13 15:45:44 -0800163 * This makes sure WB_SYNC_NONE background writeback happens. When
164 * this function returns, it is only guaranteed that for given BDI
165 * some IO is happening if we are over background dirty threshold.
166 * Caller need not hold sb s_umount semaphore.
Christoph Hellwigc5444192010-06-08 18:15:15 +0200167 */
168void bdi_start_background_writeback(struct backing_dev_info *bdi)
169{
Jan Kara65850272011-01-13 15:45:44 -0800170 /*
171 * We just wake up the flusher thread. It will perform background
172 * writeback as soon as there is no other work to do.
173 */
Wu Fengguang71927e82011-01-13 15:45:46 -0800174 trace_writeback_wake_background(bdi);
Jan Kara65850272011-01-13 15:45:44 -0800175 spin_lock_bh(&bdi->wb_lock);
176 bdi_wakeup_flusher(bdi);
177 spin_unlock_bh(&bdi->wb_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178}
179
180/*
Dave Chinnera66979a2011-03-22 22:23:41 +1100181 * Remove the inode from the writeback list it is on.
182 */
183void inode_wb_list_del(struct inode *inode)
184{
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600185 struct backing_dev_info *bdi = inode_to_bdi(inode);
Dave Chinnera66979a2011-03-22 22:23:41 +1100186
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600187 spin_lock(&bdi->wb.list_lock);
188 list_del_init(&inode->i_wb_list);
189 spin_unlock(&bdi->wb.list_lock);
190}
Dave Chinnera66979a2011-03-22 22:23:41 +1100191
192/*
Andrew Morton6610a0b2007-10-16 23:30:32 -0700193 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
194 * furthest end of its superblock's dirty-inode list.
195 *
196 * Before stamping the inode's ->dirtied_when, we check to see whether it is
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200197 * already the most-recently-dirtied inode on the b_dirty list. If that is
Andrew Morton6610a0b2007-10-16 23:30:32 -0700198 * the case then the inode must have been redirtied while it was being written
199 * out and we don't reset its dirtied_when.
200 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600201static void redirty_tail(struct inode *inode, struct bdi_writeback *wb)
Andrew Morton6610a0b2007-10-16 23:30:32 -0700202{
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600203 assert_spin_locked(&wb->list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200204 if (!list_empty(&wb->b_dirty)) {
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200205 struct inode *tail;
Andrew Morton6610a0b2007-10-16 23:30:32 -0700206
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100207 tail = wb_inode(wb->b_dirty.next);
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200208 if (time_before(inode->dirtied_when, tail->dirtied_when))
Andrew Morton6610a0b2007-10-16 23:30:32 -0700209 inode->dirtied_when = jiffies;
210 }
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100211 list_move(&inode->i_wb_list, &wb->b_dirty);
Andrew Morton6610a0b2007-10-16 23:30:32 -0700212}
213
214/*
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200215 * requeue inode for re-scanning after bdi->b_io list is exhausted.
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700216 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600217static void requeue_io(struct inode *inode, struct bdi_writeback *wb)
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700218{
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600219 assert_spin_locked(&wb->list_lock);
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100220 list_move(&inode->i_wb_list, &wb->b_more_io);
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700221}
222
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700223static void inode_sync_complete(struct inode *inode)
224{
225 /*
Dave Chinnera66979a2011-03-22 22:23:41 +1100226 * Prevent speculative execution through
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600227 * spin_unlock(&wb->list_lock);
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700228 */
Dave Chinnera66979a2011-03-22 22:23:41 +1100229
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700230 smp_mb();
231 wake_up_bit(&inode->i_state, __I_SYNC);
232}
233
Jeff Laytond2caa3c52009-04-02 16:56:37 -0700234static bool inode_dirtied_after(struct inode *inode, unsigned long t)
235{
236 bool ret = time_after(inode->dirtied_when, t);
237#ifndef CONFIG_64BIT
238 /*
239 * For inodes being constantly redirtied, dirtied_when can get stuck.
240 * It _appears_ to be in the future, but is actually in distant past.
241 * This test is necessary to prevent such wrapped-around relative times
Jens Axboe5b0830c2009-09-23 19:37:09 +0200242 * from permanently stopping the whole bdi writeback.
Jeff Laytond2caa3c52009-04-02 16:56:37 -0700243 */
244 ret = ret && time_before_eq(inode->dirtied_when, jiffies);
245#endif
246 return ret;
247}
248
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700249/*
Fengguang Wu2c136572007-10-16 23:30:39 -0700250 * Move expired dirty inodes from @delaying_queue to @dispatch_queue.
251 */
Wu Fengguange84d0a42011-04-23 12:27:27 -0600252static int move_expired_inodes(struct list_head *delaying_queue,
Fengguang Wu2c136572007-10-16 23:30:39 -0700253 struct list_head *dispatch_queue,
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600254 struct wb_writeback_work *work)
Fengguang Wu2c136572007-10-16 23:30:39 -0700255{
Shaohua Li5c034492009-09-24 14:42:33 +0200256 LIST_HEAD(tmp);
257 struct list_head *pos, *node;
Jens Axboecf137302009-09-24 15:12:57 +0200258 struct super_block *sb = NULL;
Shaohua Li5c034492009-09-24 14:42:33 +0200259 struct inode *inode;
Jens Axboecf137302009-09-24 15:12:57 +0200260 int do_sb_sort = 0;
Wu Fengguange84d0a42011-04-23 12:27:27 -0600261 int moved = 0;
Shaohua Li5c034492009-09-24 14:42:33 +0200262
Fengguang Wu2c136572007-10-16 23:30:39 -0700263 while (!list_empty(delaying_queue)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100264 inode = wb_inode(delaying_queue->prev);
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600265 if (work->older_than_this &&
266 inode_dirtied_after(inode, *work->older_than_this))
Fengguang Wu2c136572007-10-16 23:30:39 -0700267 break;
Jens Axboecf137302009-09-24 15:12:57 +0200268 if (sb && sb != inode->i_sb)
269 do_sb_sort = 1;
270 sb = inode->i_sb;
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100271 list_move(&inode->i_wb_list, &tmp);
Wu Fengguange84d0a42011-04-23 12:27:27 -0600272 moved++;
Shaohua Li5c034492009-09-24 14:42:33 +0200273 }
274
Jens Axboecf137302009-09-24 15:12:57 +0200275 /* just one sb in list, splice to dispatch_queue and we're done */
276 if (!do_sb_sort) {
277 list_splice(&tmp, dispatch_queue);
Wu Fengguange84d0a42011-04-23 12:27:27 -0600278 goto out;
Jens Axboecf137302009-09-24 15:12:57 +0200279 }
280
Shaohua Li5c034492009-09-24 14:42:33 +0200281 /* Move inodes from one superblock together */
282 while (!list_empty(&tmp)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100283 sb = wb_inode(tmp.prev)->i_sb;
Shaohua Li5c034492009-09-24 14:42:33 +0200284 list_for_each_prev_safe(pos, node, &tmp) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100285 inode = wb_inode(pos);
Shaohua Li5c034492009-09-24 14:42:33 +0200286 if (inode->i_sb == sb)
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100287 list_move(&inode->i_wb_list, dispatch_queue);
Shaohua Li5c034492009-09-24 14:42:33 +0200288 }
Fengguang Wu2c136572007-10-16 23:30:39 -0700289 }
Wu Fengguange84d0a42011-04-23 12:27:27 -0600290out:
291 return moved;
Fengguang Wu2c136572007-10-16 23:30:39 -0700292}
293
294/*
295 * Queue all expired dirty inodes for io, eldest first.
Wu Fengguang4ea879b2010-08-11 14:17:42 -0700296 * Before
297 * newly dirtied b_dirty b_io b_more_io
298 * =============> gf edc BA
299 * After
300 * newly dirtied b_dirty b_io b_more_io
301 * =============> g fBAedc
302 * |
303 * +--> dequeue for IO
Fengguang Wu2c136572007-10-16 23:30:39 -0700304 */
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600305static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work)
Fengguang Wu2c136572007-10-16 23:30:39 -0700306{
Wu Fengguange84d0a42011-04-23 12:27:27 -0600307 int moved;
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600308 assert_spin_locked(&wb->list_lock);
Wu Fengguang4ea879b2010-08-11 14:17:42 -0700309 list_splice_init(&wb->b_more_io, &wb->b_io);
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600310 moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, work);
311 trace_writeback_queue_io(wb, work, moved);
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200312}
313
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100314static int write_inode(struct inode *inode, struct writeback_control *wbc)
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200315{
Jens Axboe03ba3782009-09-09 09:08:54 +0200316 if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode))
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100317 return inode->i_sb->s_op->write_inode(inode, wbc);
Jens Axboe03ba3782009-09-09 09:08:54 +0200318 return 0;
Fengguang Wu2c136572007-10-16 23:30:39 -0700319}
320
321/*
Christoph Hellwig01c03192009-06-08 13:35:40 +0200322 * Wait for writeback on an inode to complete.
323 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600324static void inode_wait_for_writeback(struct inode *inode,
325 struct bdi_writeback *wb)
Christoph Hellwig01c03192009-06-08 13:35:40 +0200326{
327 DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC);
328 wait_queue_head_t *wqh;
329
330 wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
Dave Chinner250df6e2011-03-22 22:23:36 +1100331 while (inode->i_state & I_SYNC) {
332 spin_unlock(&inode->i_lock);
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600333 spin_unlock(&wb->list_lock);
Christoph Hellwig01c03192009-06-08 13:35:40 +0200334 __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE);
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600335 spin_lock(&wb->list_lock);
Dave Chinner250df6e2011-03-22 22:23:36 +1100336 spin_lock(&inode->i_lock);
Richard Kennedy58a9d3d82010-05-24 14:32:38 -0700337 }
Christoph Hellwig01c03192009-06-08 13:35:40 +0200338}
339
340/*
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600341 * Write out an inode's dirty pages. Called under wb->list_lock and
Dave Chinner0f1b1fd2011-03-22 22:23:43 +1100342 * inode->i_lock. Either the caller has an active reference on the inode or
343 * the inode has I_WILL_FREE set.
Christoph Hellwig01c03192009-06-08 13:35:40 +0200344 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345 * If `wait' is set, wait on the writeout.
346 *
347 * The whole writeout design is quite complex and fragile. We want to avoid
348 * starvation of particular inodes when others are being redirtied, prevent
349 * livelocks, etc.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350 */
351static int
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600352writeback_single_inode(struct inode *inode, struct bdi_writeback *wb,
353 struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355 struct address_space *mapping = inode->i_mapping;
Wu Fengguang251d6a42010-12-01 17:33:37 -0600356 long nr_to_write = wbc->nr_to_write;
Christoph Hellwig01c03192009-06-08 13:35:40 +0200357 unsigned dirty;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358 int ret;
359
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600360 assert_spin_locked(&wb->list_lock);
Dave Chinner0f1b1fd2011-03-22 22:23:43 +1100361 assert_spin_locked(&inode->i_lock);
362
Christoph Hellwig01c03192009-06-08 13:35:40 +0200363 if (!atomic_read(&inode->i_count))
364 WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING)));
365 else
366 WARN_ON(inode->i_state & I_WILL_FREE);
367
368 if (inode->i_state & I_SYNC) {
369 /*
370 * If this inode is locked for writeback and we are not doing
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200371 * writeback-for-data-integrity, move it to b_more_io so that
Christoph Hellwig01c03192009-06-08 13:35:40 +0200372 * writeback can proceed with the other inodes on s_io.
373 *
374 * We'll have another go at writing back this inode when we
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200375 * completed a full scan of b_io.
Christoph Hellwig01c03192009-06-08 13:35:40 +0200376 */
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100377 if (wbc->sync_mode != WB_SYNC_ALL) {
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600378 requeue_io(inode, wb);
Wu Fengguang251d6a42010-12-01 17:33:37 -0600379 trace_writeback_single_inode_requeue(inode, wbc,
380 nr_to_write);
Christoph Hellwig01c03192009-06-08 13:35:40 +0200381 return 0;
382 }
383
384 /*
385 * It's a data-integrity sync. We must wait.
386 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600387 inode_wait_for_writeback(inode, wb);
Christoph Hellwig01c03192009-06-08 13:35:40 +0200388 }
389
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700390 BUG_ON(inode->i_state & I_SYNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391
Dmitry Monakhov5547e8a2010-05-07 13:35:44 +0400392 /* Set I_SYNC, reset I_DIRTY_PAGES */
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700393 inode->i_state |= I_SYNC;
Dmitry Monakhov5547e8a2010-05-07 13:35:44 +0400394 inode->i_state &= ~I_DIRTY_PAGES;
Dave Chinner250df6e2011-03-22 22:23:36 +1100395 spin_unlock(&inode->i_lock);
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600396 spin_unlock(&wb->list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397
398 ret = do_writepages(mapping, wbc);
399
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100400 /*
401 * Make sure to wait on the data before writing out the metadata.
402 * This is important for filesystems that modify metadata on data
403 * I/O completion.
404 */
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100405 if (wbc->sync_mode == WB_SYNC_ALL) {
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100406 int err = filemap_fdatawait(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407 if (ret == 0)
408 ret = err;
409 }
410
Dmitry Monakhov5547e8a2010-05-07 13:35:44 +0400411 /*
412 * Some filesystems may redirty the inode during the writeback
413 * due to delalloc, clear dirty metadata flags right before
414 * write_inode()
415 */
Dave Chinner250df6e2011-03-22 22:23:36 +1100416 spin_lock(&inode->i_lock);
Dmitry Monakhov5547e8a2010-05-07 13:35:44 +0400417 dirty = inode->i_state & I_DIRTY;
418 inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC);
Dave Chinner250df6e2011-03-22 22:23:36 +1100419 spin_unlock(&inode->i_lock);
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100420 /* Don't write the inode if only I_DIRTY_PAGES was set */
421 if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100422 int err = write_inode(inode, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423 if (ret == 0)
424 ret = err;
425 }
426
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600427 spin_lock(&wb->list_lock);
Dave Chinner250df6e2011-03-22 22:23:36 +1100428 spin_lock(&inode->i_lock);
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700429 inode->i_state &= ~I_SYNC;
Al Viroa4ffdde2010-06-02 17:38:30 -0400430 if (!(inode->i_state & I_FREEING)) {
Wu Fengguang94c3dcb2011-04-27 19:05:21 -0600431 /*
432 * Sync livelock prevention. Each inode is tagged and synced in
433 * one shot. If still dirty, it will be redirty_tail()'ed below.
434 * Update the dirty time to prevent enqueue and sync it again.
435 */
436 if ((inode->i_state & I_DIRTY) &&
437 (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages))
438 inode->dirtied_when = jiffies;
439
Wu Fengguang23539af2010-08-11 14:17:41 -0700440 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 /*
442 * We didn't write back all the pages. nfs_writepages()
Wu Fengguanga50aeb42010-08-11 14:17:43 -0700443 * sometimes bales out without doing anything.
Andrew Morton1b43ef92007-10-16 23:30:35 -0700444 */
Wu Fengguanga50aeb42010-08-11 14:17:43 -0700445 inode->i_state |= I_DIRTY_PAGES;
446 if (wbc->nr_to_write <= 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447 /*
Wu Fengguanga50aeb42010-08-11 14:17:43 -0700448 * slice used up: queue for next turn
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600450 requeue_io(inode, wb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451 } else {
452 /*
Wu Fengguanga50aeb42010-08-11 14:17:43 -0700453 * Writeback blocked by something other than
454 * congestion. Delay the inode for some time to
455 * avoid spinning on the CPU (100% iowait)
456 * retrying writeback of the dirty page/inode
457 * that cannot be performed immediately.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600459 redirty_tail(inode, wb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460 }
Wu Fengguang23539af2010-08-11 14:17:41 -0700461 } else if (inode->i_state & I_DIRTY) {
462 /*
463 * Filesystems can dirty the inode during writeback
464 * operations, such as delayed allocation during
465 * submission or metadata updates after data IO
466 * completion.
467 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600468 redirty_tail(inode, wb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469 } else {
470 /*
Nick Piggin9e38d862010-10-23 06:55:17 -0400471 * The inode is clean. At this point we either have
472 * a reference to the inode or it's on it's way out.
473 * No need to add it back to the LRU.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474 */
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100475 list_del_init(&inode->i_wb_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476 }
477 }
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700478 inode_sync_complete(inode);
Wu Fengguang251d6a42010-12-01 17:33:37 -0600479 trace_writeback_single_inode(inode, wbc, nr_to_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480 return ret;
481}
482
Wu Fengguang1a12d8b2010-08-29 13:28:09 -0600483static long writeback_chunk_size(struct backing_dev_info *bdi,
484 struct wb_writeback_work *work)
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600485{
486 long pages;
487
488 /*
489 * WB_SYNC_ALL mode does livelock avoidance by syncing dirty
490 * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX
491 * here avoids calling into writeback_inodes_wb() more than once.
492 *
493 * The intended call sequence for WB_SYNC_ALL writeback is:
494 *
495 * wb_writeback()
496 * writeback_sb_inodes() <== called only once
497 * write_cache_pages() <== called once for each inode
498 * (quickly) tag currently dirty pages
499 * (maybe slowly) sync all tagged pages
500 */
501 if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages)
502 pages = LONG_MAX;
Wu Fengguang1a12d8b2010-08-29 13:28:09 -0600503 else {
504 pages = min(bdi->avg_write_bandwidth / 2,
505 global_dirty_limit / DIRTY_SCOPE);
506 pages = min(pages, work->nr_pages);
507 pages = round_down(pages + MIN_WRITEBACK_PAGES,
508 MIN_WRITEBACK_PAGES);
509 }
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600510
511 return pages;
512}
513
Jens Axboe03ba3782009-09-09 09:08:54 +0200514/*
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800515 * Write a portion of b_io inodes which belong to @sb.
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200516 *
517 * If @only_this_sb is true, then find and write all such
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800518 * inodes. Otherwise write only ones which go sequentially
519 * in reverse order.
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200520 *
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600521 * Return the number of pages and/or inodes written.
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800522 */
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600523static long writeback_sb_inodes(struct super_block *sb,
524 struct bdi_writeback *wb,
525 struct wb_writeback_work *work)
Jens Axboe03ba3782009-09-09 09:08:54 +0200526{
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600527 struct writeback_control wbc = {
528 .sync_mode = work->sync_mode,
529 .tagged_writepages = work->tagged_writepages,
530 .for_kupdate = work->for_kupdate,
531 .for_background = work->for_background,
532 .range_cyclic = work->range_cyclic,
533 .range_start = 0,
534 .range_end = LLONG_MAX,
535 };
536 unsigned long start_time = jiffies;
537 long write_chunk;
538 long wrote = 0; /* count both pages and inodes */
539
Jens Axboe03ba3782009-09-09 09:08:54 +0200540 while (!list_empty(&wb->b_io)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100541 struct inode *inode = wb_inode(wb->b_io.prev);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200542
543 if (inode->i_sb != sb) {
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600544 if (work->sb) {
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200545 /*
546 * We only want to write back data for this
547 * superblock, move all inodes not belonging
548 * to it back onto the dirty list.
549 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600550 redirty_tail(inode, wb);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200551 continue;
552 }
553
554 /*
555 * The inode belongs to a different superblock.
556 * Bounce back to the caller to unpin this and
557 * pin the next superblock.
558 */
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600559 break;
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200560 }
561
Christoph Hellwig9843b762010-10-24 19:40:46 +0200562 /*
563 * Don't bother with new inodes or inodes beeing freed, first
564 * kind does not need peridic writeout yet, and for the latter
565 * kind writeout is handled by the freer.
566 */
Dave Chinner250df6e2011-03-22 22:23:36 +1100567 spin_lock(&inode->i_lock);
Christoph Hellwig9843b762010-10-24 19:40:46 +0200568 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
Dave Chinner250df6e2011-03-22 22:23:36 +1100569 spin_unlock(&inode->i_lock);
Wu Fengguangfcc5c222011-07-11 23:08:50 -0700570 redirty_tail(inode, wb);
Nick Piggin7ef0d732009-03-12 14:31:38 -0700571 continue;
572 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573 __iget(inode);
Wu Fengguang1a12d8b2010-08-29 13:28:09 -0600574 write_chunk = writeback_chunk_size(wb->bdi, work);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600575 wbc.nr_to_write = write_chunk;
576 wbc.pages_skipped = 0;
Dave Chinner250df6e2011-03-22 22:23:36 +1100577
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600578 writeback_single_inode(inode, wb, &wbc);
579
580 work->nr_pages -= write_chunk - wbc.nr_to_write;
581 wrote += write_chunk - wbc.nr_to_write;
582 if (!(inode->i_state & I_DIRTY))
583 wrote++;
584 if (wbc.pages_skipped) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585 /*
586 * writeback is not making progress due to locked
587 * buffers. Skip this inode for now.
588 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600589 redirty_tail(inode, wb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590 }
Dave Chinner0f1b1fd2011-03-22 22:23:43 +1100591 spin_unlock(&inode->i_lock);
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600592 spin_unlock(&wb->list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593 iput(inode);
OGAWA Hirofumi4ffc8442006-03-25 03:07:44 -0800594 cond_resched();
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600595 spin_lock(&wb->list_lock);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600596 /*
597 * bail out to wb_writeback() often enough to check
598 * background threshold and other termination conditions.
599 */
600 if (wrote) {
601 if (time_is_before_jiffies(start_time + HZ / 10UL))
602 break;
603 if (work->nr_pages <= 0)
604 break;
Fengguang Wu8bc3be22008-02-04 22:29:36 -0800605 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 }
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600607 return wrote;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800608}
Nick Piggin38f21972009-01-06 14:40:25 -0800609
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600610static long __writeback_inodes_wb(struct bdi_writeback *wb,
611 struct wb_writeback_work *work)
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800612{
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600613 unsigned long start_time = jiffies;
614 long wrote = 0;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800615
616 while (!list_empty(&wb->b_io)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100617 struct inode *inode = wb_inode(wb->b_io.prev);
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800618 struct super_block *sb = inode->i_sb;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800619
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000620 if (!grab_super_passive(sb)) {
Wu Fengguang0e9958162011-07-29 22:14:35 -0600621 /*
622 * grab_super_passive() may fail consistently due to
623 * s_umount being grabbed by someone else. Don't use
624 * requeue_io() to avoid busy retrying the inode/sb.
625 */
626 redirty_tail(inode, wb);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200627 continue;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800628 }
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600629 wrote += writeback_sb_inodes(sb, wb, work);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200630 drop_super(sb);
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800631
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600632 /* refer to the same tests at the end of writeback_sb_inodes */
633 if (wrote) {
634 if (time_is_before_jiffies(start_time + HZ / 10UL))
635 break;
636 if (work->nr_pages <= 0)
637 break;
638 }
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800639 }
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200640 /* Leave any unwritten inodes on b_io */
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600641 return wrote;
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200642}
643
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600644long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages)
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200645{
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600646 struct wb_writeback_work work = {
647 .nr_pages = nr_pages,
648 .sync_mode = WB_SYNC_NONE,
649 .range_cyclic = 1,
650 };
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200651
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600652 spin_lock(&wb->list_lock);
Wu Fengguang424b3512010-07-21 20:11:53 -0600653 if (list_empty(&wb->b_io))
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600654 queue_io(wb, &work);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600655 __writeback_inodes_wb(wb, &work);
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600656 spin_unlock(&wb->list_lock);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200657
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600658 return nr_pages - work.nr_pages;
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200659}
660
Wu Fengguangb00949a2010-11-18 14:38:33 -0600661static bool over_bground_thresh(struct backing_dev_info *bdi)
Jens Axboe03ba3782009-09-09 09:08:54 +0200662{
663 unsigned long background_thresh, dirty_thresh;
664
Wu Fengguang16c40422010-08-11 14:17:39 -0700665 global_dirty_limits(&background_thresh, &dirty_thresh);
Jens Axboe03ba3782009-09-09 09:08:54 +0200666
Wu Fengguangb00949a2010-11-18 14:38:33 -0600667 if (global_page_state(NR_FILE_DIRTY) +
668 global_page_state(NR_UNSTABLE_NFS) > background_thresh)
669 return true;
670
671 if (bdi_stat(bdi, BDI_RECLAIMABLE) >
672 bdi_dirty_limit(bdi, background_thresh))
673 return true;
674
675 return false;
Jens Axboe03ba3782009-09-09 09:08:54 +0200676}
677
678/*
Wu Fengguange98be2d2010-08-29 11:22:30 -0600679 * Called under wb->list_lock. If there are multiple wb per bdi,
680 * only the flusher working on the first wb should do it.
681 */
682static void wb_update_bandwidth(struct bdi_writeback *wb,
683 unsigned long start_time)
684{
Wu Fengguangaf6a3112011-10-03 20:46:17 -0600685 __bdi_update_bandwidth(wb->bdi, 0, 0, 0, 0, 0, start_time);
Wu Fengguange98be2d2010-08-29 11:22:30 -0600686}
687
688/*
Jens Axboe03ba3782009-09-09 09:08:54 +0200689 * Explicit flushing or periodic writeback of "old" data.
690 *
691 * Define "old": the first time one of an inode's pages is dirtied, we mark the
692 * dirtying-time in the inode's address_space. So this periodic writeback code
693 * just walks the superblock inode list, writing back any inodes which are
694 * older than a specific point in time.
695 *
696 * Try to run once per dirty_writeback_interval. But if a writeback event
697 * takes longer than a dirty_writeback_interval interval, then leave a
698 * one-second gap.
699 *
700 * older_than_this takes precedence over nr_to_write. So we'll only write back
701 * all dirty pages if they are all attached to "old" mappings.
702 */
Jens Axboec4a77a62009-09-16 15:18:25 +0200703static long wb_writeback(struct bdi_writeback *wb,
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200704 struct wb_writeback_work *work)
Jens Axboe03ba3782009-09-09 09:08:54 +0200705{
Wu Fengguange98be2d2010-08-29 11:22:30 -0600706 unsigned long wb_start = jiffies;
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600707 long nr_pages = work->nr_pages;
Jens Axboe03ba3782009-09-09 09:08:54 +0200708 unsigned long oldest_jif;
Jan Karaa5989bd2009-09-16 19:22:48 +0200709 struct inode *inode;
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600710 long progress;
Jens Axboe03ba3782009-09-09 09:08:54 +0200711
Wu Fengguange185dda2011-04-23 11:26:07 -0600712 oldest_jif = jiffies;
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600713 work->older_than_this = &oldest_jif;
Jens Axboe03ba3782009-09-09 09:08:54 +0200714
Wu Fengguange8dfc302011-04-21 12:06:32 -0600715 spin_lock(&wb->list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200716 for (;;) {
717 /*
Wu Fengguangd3ddec72009-09-23 20:33:40 +0800718 * Stop writeback when nr_pages has been consumed
Jens Axboe03ba3782009-09-09 09:08:54 +0200719 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200720 if (work->nr_pages <= 0)
Jens Axboe03ba3782009-09-09 09:08:54 +0200721 break;
722
723 /*
Jan Karaaa373cf2011-01-13 15:45:47 -0800724 * Background writeout and kupdate-style writeback may
725 * run forever. Stop them if there is other work to do
726 * so that e.g. sync can proceed. They'll be restarted
727 * after the other works are all done.
728 */
729 if ((work->for_background || work->for_kupdate) &&
730 !list_empty(&wb->bdi->work_list))
731 break;
732
733 /*
Wu Fengguangd3ddec72009-09-23 20:33:40 +0800734 * For background writeout, stop when we are below the
735 * background dirty threshold
Jens Axboe03ba3782009-09-09 09:08:54 +0200736 */
Wu Fengguangb00949a2010-11-18 14:38:33 -0600737 if (work->for_background && !over_bground_thresh(wb->bdi))
Jens Axboe03ba3782009-09-09 09:08:54 +0200738 break;
739
Wu Fengguangba9aa832010-07-21 20:32:30 -0600740 if (work->for_kupdate) {
741 oldest_jif = jiffies -
742 msecs_to_jiffies(dirty_expire_interval * 10);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600743 work->older_than_this = &oldest_jif;
Wu Fengguangba9aa832010-07-21 20:32:30 -0600744 }
Dave Chinner028c2dd2010-07-07 13:24:07 +1000745
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600746 trace_writeback_start(wb->bdi, work);
Wu Fengguange8dfc302011-04-21 12:06:32 -0600747 if (list_empty(&wb->b_io))
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600748 queue_io(wb, work);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200749 if (work->sb)
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600750 progress = writeback_sb_inodes(work->sb, wb, work);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200751 else
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600752 progress = __writeback_inodes_wb(wb, work);
753 trace_writeback_written(wb->bdi, work);
Dave Chinner028c2dd2010-07-07 13:24:07 +1000754
Wu Fengguange98be2d2010-08-29 11:22:30 -0600755 wb_update_bandwidth(wb, wb_start);
Jens Axboe03ba3782009-09-09 09:08:54 +0200756
757 /*
Jens Axboe71fd05a2009-09-23 19:32:26 +0200758 * Did we write something? Try for more
Wu Fengguange6fb6da2010-07-22 10:23:44 -0600759 *
760 * Dirty inodes are moved to b_io for writeback in batches.
761 * The completion of the current batch does not necessarily
762 * mean the overall work is done. So we keep looping as long
763 * as made some progress on cleaning pages or inodes.
Jens Axboe71fd05a2009-09-23 19:32:26 +0200764 */
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600765 if (progress)
Jens Axboe71fd05a2009-09-23 19:32:26 +0200766 continue;
767 /*
Wu Fengguange6fb6da2010-07-22 10:23:44 -0600768 * No more inodes for IO, bail
Jens Axboe71fd05a2009-09-23 19:32:26 +0200769 */
Wu Fengguangb7a24412010-07-21 22:19:51 -0600770 if (list_empty(&wb->b_more_io))
Jens Axboe03ba3782009-09-09 09:08:54 +0200771 break;
772 /*
Jens Axboe8010c3b2009-09-15 20:04:57 +0200773 * Nothing written. Wait for some inode to
774 * become available for writeback. Otherwise
775 * we'll just busyloop.
Jens Axboe03ba3782009-09-09 09:08:54 +0200776 */
Jens Axboe03ba3782009-09-09 09:08:54 +0200777 if (!list_empty(&wb->b_more_io)) {
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600778 trace_writeback_wait(wb->bdi, work);
Jens Axboe03ba3782009-09-09 09:08:54 +0200779 inode = wb_inode(wb->b_more_io.prev);
Dave Chinner250df6e2011-03-22 22:23:36 +1100780 spin_lock(&inode->i_lock);
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600781 inode_wait_for_writeback(inode, wb);
Dave Chinner250df6e2011-03-22 22:23:36 +1100782 spin_unlock(&inode->i_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200783 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200784 }
Wu Fengguange8dfc302011-04-21 12:06:32 -0600785 spin_unlock(&wb->list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200786
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600787 return nr_pages - work->nr_pages;
Jens Axboe03ba3782009-09-09 09:08:54 +0200788}
789
790/*
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200791 * Return the next wb_writeback_work struct that hasn't been processed yet.
Jens Axboe03ba3782009-09-09 09:08:54 +0200792 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200793static struct wb_writeback_work *
Minchan Kim08852b62010-08-03 12:51:16 +0200794get_next_work_item(struct backing_dev_info *bdi)
Jens Axboe03ba3782009-09-09 09:08:54 +0200795{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200796 struct wb_writeback_work *work = NULL;
Jens Axboe03ba3782009-09-09 09:08:54 +0200797
Artem Bityutskiy64677162010-07-25 14:29:22 +0300798 spin_lock_bh(&bdi->wb_lock);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200799 if (!list_empty(&bdi->work_list)) {
800 work = list_entry(bdi->work_list.next,
801 struct wb_writeback_work, list);
802 list_del_init(&work->list);
Jens Axboe03ba3782009-09-09 09:08:54 +0200803 }
Artem Bityutskiy64677162010-07-25 14:29:22 +0300804 spin_unlock_bh(&bdi->wb_lock);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200805 return work;
Jens Axboe03ba3782009-09-09 09:08:54 +0200806}
807
Linus Torvaldscdf01dd2010-10-30 08:55:52 -0700808/*
809 * Add in the number of potentially dirty inodes, because each inode
810 * write can dirty pagecache in the underlying blockdev.
811 */
812static unsigned long get_nr_dirty_pages(void)
813{
814 return global_page_state(NR_FILE_DIRTY) +
815 global_page_state(NR_UNSTABLE_NFS) +
816 get_nr_dirty_inodes();
817}
818
Jan Kara65850272011-01-13 15:45:44 -0800819static long wb_check_background_flush(struct bdi_writeback *wb)
820{
Wu Fengguangb00949a2010-11-18 14:38:33 -0600821 if (over_bground_thresh(wb->bdi)) {
Jan Kara65850272011-01-13 15:45:44 -0800822
823 struct wb_writeback_work work = {
824 .nr_pages = LONG_MAX,
825 .sync_mode = WB_SYNC_NONE,
826 .for_background = 1,
827 .range_cyclic = 1,
828 };
829
830 return wb_writeback(wb, &work);
831 }
832
833 return 0;
834}
835
Jens Axboe03ba3782009-09-09 09:08:54 +0200836static long wb_check_old_data_flush(struct bdi_writeback *wb)
837{
838 unsigned long expired;
839 long nr_pages;
840
Jens Axboe69b62d02010-05-17 12:51:03 +0200841 /*
842 * When set to zero, disable periodic writeback
843 */
844 if (!dirty_writeback_interval)
845 return 0;
846
Jens Axboe03ba3782009-09-09 09:08:54 +0200847 expired = wb->last_old_flush +
848 msecs_to_jiffies(dirty_writeback_interval * 10);
849 if (time_before(jiffies, expired))
850 return 0;
851
852 wb->last_old_flush = jiffies;
Linus Torvaldscdf01dd2010-10-30 08:55:52 -0700853 nr_pages = get_nr_dirty_pages();
Jens Axboe03ba3782009-09-09 09:08:54 +0200854
Jens Axboec4a77a62009-09-16 15:18:25 +0200855 if (nr_pages) {
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200856 struct wb_writeback_work work = {
Jens Axboec4a77a62009-09-16 15:18:25 +0200857 .nr_pages = nr_pages,
858 .sync_mode = WB_SYNC_NONE,
859 .for_kupdate = 1,
860 .range_cyclic = 1,
861 };
862
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200863 return wb_writeback(wb, &work);
Jens Axboec4a77a62009-09-16 15:18:25 +0200864 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200865
866 return 0;
867}
868
869/*
870 * Retrieve work items and do the writeback they describe
871 */
872long wb_do_writeback(struct bdi_writeback *wb, int force_wait)
873{
874 struct backing_dev_info *bdi = wb->bdi;
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200875 struct wb_writeback_work *work;
Jens Axboec4a77a62009-09-16 15:18:25 +0200876 long wrote = 0;
Jens Axboe03ba3782009-09-09 09:08:54 +0200877
Jan Kara81d73a32010-08-11 14:17:44 -0700878 set_bit(BDI_writeback_running, &wb->bdi->state);
Minchan Kim08852b62010-08-03 12:51:16 +0200879 while ((work = get_next_work_item(bdi)) != NULL) {
Jens Axboe03ba3782009-09-09 09:08:54 +0200880 /*
881 * Override sync mode, in case we must wait for completion
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200882 * because this thread is exiting now.
Jens Axboe03ba3782009-09-09 09:08:54 +0200883 */
884 if (force_wait)
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200885 work->sync_mode = WB_SYNC_ALL;
886
Dave Chinner455b2862010-07-07 13:24:06 +1000887 trace_writeback_exec(bdi, work);
888
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200889 wrote += wb_writeback(wb, work);
Jens Axboe03ba3782009-09-09 09:08:54 +0200890
891 /*
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200892 * Notify the caller of completion if this is a synchronous
893 * work item, otherwise just free it.
Jens Axboe03ba3782009-09-09 09:08:54 +0200894 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200895 if (work->done)
896 complete(work->done);
897 else
898 kfree(work);
Jens Axboe03ba3782009-09-09 09:08:54 +0200899 }
900
901 /*
902 * Check for periodic writeback, kupdated() style
903 */
904 wrote += wb_check_old_data_flush(wb);
Jan Kara65850272011-01-13 15:45:44 -0800905 wrote += wb_check_background_flush(wb);
Jan Kara81d73a32010-08-11 14:17:44 -0700906 clear_bit(BDI_writeback_running, &wb->bdi->state);
Jens Axboe03ba3782009-09-09 09:08:54 +0200907
908 return wrote;
909}
910
911/*
912 * Handle writeback of dirty data for the device backed by this bdi. Also
913 * wakes up periodically and does kupdated style flushing.
914 */
Christoph Hellwig08243902010-06-19 23:08:22 +0200915int bdi_writeback_thread(void *data)
Jens Axboe03ba3782009-09-09 09:08:54 +0200916{
Christoph Hellwig08243902010-06-19 23:08:22 +0200917 struct bdi_writeback *wb = data;
918 struct backing_dev_info *bdi = wb->bdi;
Jens Axboe03ba3782009-09-09 09:08:54 +0200919 long pages_written;
920
Peter Zijlstra766f9162010-10-26 14:22:45 -0700921 current->flags |= PF_SWAPWRITE;
Christoph Hellwig08243902010-06-19 23:08:22 +0200922 set_freezable();
Artem Bityutskiyecd58402010-07-25 14:29:18 +0300923 wb->last_active = jiffies;
Christoph Hellwig08243902010-06-19 23:08:22 +0200924
925 /*
926 * Our parent may run at a different priority, just set us to normal
927 */
928 set_user_nice(current, 0);
929
Dave Chinner455b2862010-07-07 13:24:06 +1000930 trace_writeback_thread_start(bdi);
931
Jens Axboe03ba3782009-09-09 09:08:54 +0200932 while (!kthread_should_stop()) {
Artem Bityutskiy64677162010-07-25 14:29:22 +0300933 /*
934 * Remove own delayed wake-up timer, since we are already awake
935 * and we'll take care of the preriodic write-back.
936 */
937 del_timer(&wb->wakeup_timer);
938
Jens Axboe03ba3782009-09-09 09:08:54 +0200939 pages_written = wb_do_writeback(wb, 0);
940
Dave Chinner455b2862010-07-07 13:24:06 +1000941 trace_writeback_pages_written(pages_written);
Jens Axboe03ba3782009-09-09 09:08:54 +0200942
Jens Axboe03ba3782009-09-09 09:08:54 +0200943 if (pages_written)
Artem Bityutskiyecd58402010-07-25 14:29:18 +0300944 wb->last_active = jiffies;
Jens Axboe03ba3782009-09-09 09:08:54 +0200945
Artem Bityutskiy297252c2010-07-25 14:29:15 +0300946 set_current_state(TASK_INTERRUPTIBLE);
J. Bruce Fieldsb76b4012010-08-28 08:52:10 +0200947 if (!list_empty(&bdi->work_list) || kthread_should_stop()) {
Artem Bityutskiy297252c2010-07-25 14:29:15 +0300948 __set_current_state(TASK_RUNNING);
949 continue;
Jens Axboe03ba3782009-09-09 09:08:54 +0200950 }
951
Artem Bityutskiy253c34e2010-07-25 14:29:21 +0300952 if (wb_has_dirty_io(wb) && dirty_writeback_interval)
Artem Bityutskiyfff5b852010-07-25 14:29:20 +0300953 schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10));
Artem Bityutskiy253c34e2010-07-25 14:29:21 +0300954 else {
955 /*
956 * We have nothing to do, so can go sleep without any
957 * timeout and save power. When a work is queued or
958 * something is made dirty - we will be woken up.
959 */
Artem Bityutskiy297252c2010-07-25 14:29:15 +0300960 schedule();
Jens Axboef9eadbb2010-05-18 14:31:45 +0200961 }
Jens Axboe69b62d02010-05-17 12:51:03 +0200962
Jens Axboe03ba3782009-09-09 09:08:54 +0200963 try_to_freeze();
964 }
965
Artem Bityutskiyfff5b852010-07-25 14:29:20 +0300966 /* Flush any work that raced with us exiting */
Christoph Hellwig08243902010-06-19 23:08:22 +0200967 if (!list_empty(&bdi->work_list))
968 wb_do_writeback(wb, 1);
Dave Chinner455b2862010-07-07 13:24:06 +1000969
970 trace_writeback_thread_stop(bdi);
Jens Axboe03ba3782009-09-09 09:08:54 +0200971 return 0;
972}
973
Christoph Hellwig08243902010-06-19 23:08:22 +0200974
Jens Axboe03ba3782009-09-09 09:08:54 +0200975/*
Jens Axboe03ba3782009-09-09 09:08:54 +0200976 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
977 * the whole world.
978 */
979void wakeup_flusher_threads(long nr_pages)
980{
Christoph Hellwigb8c2f342010-06-08 18:15:07 +0200981 struct backing_dev_info *bdi;
Christoph Hellwigb8c2f342010-06-08 18:15:07 +0200982
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200983 if (!nr_pages) {
984 nr_pages = global_page_state(NR_FILE_DIRTY) +
Jens Axboe03ba3782009-09-09 09:08:54 +0200985 global_page_state(NR_UNSTABLE_NFS);
Christoph Hellwigb8c2f342010-06-08 18:15:07 +0200986 }
987
988 rcu_read_lock();
989 list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) {
990 if (!bdi_has_dirty_io(bdi))
991 continue;
Jan Kara65850272011-01-13 15:45:44 -0800992 __bdi_start_writeback(bdi, nr_pages, false);
Christoph Hellwigb8c2f342010-06-08 18:15:07 +0200993 }
994 rcu_read_unlock();
Jens Axboe03ba3782009-09-09 09:08:54 +0200995}
996
997static noinline void block_dump___mark_inode_dirty(struct inode *inode)
998{
999 if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) {
1000 struct dentry *dentry;
1001 const char *name = "?";
1002
1003 dentry = d_find_alias(inode);
1004 if (dentry) {
1005 spin_lock(&dentry->d_lock);
1006 name = (const char *) dentry->d_name.name;
1007 }
1008 printk(KERN_DEBUG
1009 "%s(%d): dirtied inode %lu (%s) on %s\n",
1010 current->comm, task_pid_nr(current), inode->i_ino,
1011 name, inode->i_sb->s_id);
1012 if (dentry) {
1013 spin_unlock(&dentry->d_lock);
1014 dput(dentry);
1015 }
1016 }
1017}
1018
1019/**
1020 * __mark_inode_dirty - internal function
1021 * @inode: inode to mark
1022 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
1023 * Mark an inode as dirty. Callers should use mark_inode_dirty or
1024 * mark_inode_dirty_sync.
1025 *
1026 * Put the inode on the super block's dirty list.
1027 *
1028 * CAREFUL! We mark it dirty unconditionally, but move it onto the
1029 * dirty list only if it is hashed or if it refers to a blockdev.
1030 * If it was not hashed, it will never be added to the dirty list
1031 * even if it is later hashed, as it will have been marked dirty already.
1032 *
1033 * In short, make sure you hash any inodes _before_ you start marking
1034 * them dirty.
1035 *
Jens Axboe03ba3782009-09-09 09:08:54 +02001036 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
1037 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
1038 * the kernel-internal blockdev inode represents the dirtying time of the
1039 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
1040 * page->mapping->host, so the page-dirtying time is recorded in the internal
1041 * blockdev inode.
1042 */
1043void __mark_inode_dirty(struct inode *inode, int flags)
1044{
1045 struct super_block *sb = inode->i_sb;
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001046 struct backing_dev_info *bdi = NULL;
Jens Axboe03ba3782009-09-09 09:08:54 +02001047
1048 /*
1049 * Don't do this for I_DIRTY_PAGES - that doesn't actually
1050 * dirty the inode itself
1051 */
1052 if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
1053 if (sb->s_op->dirty_inode)
Christoph Hellwigaa385722011-05-27 06:53:02 -04001054 sb->s_op->dirty_inode(inode, flags);
Jens Axboe03ba3782009-09-09 09:08:54 +02001055 }
1056
1057 /*
1058 * make sure that changes are seen by all cpus before we test i_state
1059 * -- mikulas
1060 */
1061 smp_mb();
1062
1063 /* avoid the locking if we can */
1064 if ((inode->i_state & flags) == flags)
1065 return;
1066
1067 if (unlikely(block_dump))
1068 block_dump___mark_inode_dirty(inode);
1069
Dave Chinner250df6e2011-03-22 22:23:36 +11001070 spin_lock(&inode->i_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001071 if ((inode->i_state & flags) != flags) {
1072 const int was_dirty = inode->i_state & I_DIRTY;
1073
1074 inode->i_state |= flags;
1075
1076 /*
1077 * If the inode is being synced, just update its dirty state.
1078 * The unlocker will place the inode on the appropriate
1079 * superblock list, based upon its state.
1080 */
1081 if (inode->i_state & I_SYNC)
Dave Chinner250df6e2011-03-22 22:23:36 +11001082 goto out_unlock_inode;
Jens Axboe03ba3782009-09-09 09:08:54 +02001083
1084 /*
1085 * Only add valid (hashed) inodes to the superblock's
1086 * dirty list. Add blockdev inodes as well.
1087 */
1088 if (!S_ISBLK(inode->i_mode)) {
Al Viro1d3382cb2010-10-23 15:19:20 -04001089 if (inode_unhashed(inode))
Dave Chinner250df6e2011-03-22 22:23:36 +11001090 goto out_unlock_inode;
Jens Axboe03ba3782009-09-09 09:08:54 +02001091 }
Al Viroa4ffdde2010-06-02 17:38:30 -04001092 if (inode->i_state & I_FREEING)
Dave Chinner250df6e2011-03-22 22:23:36 +11001093 goto out_unlock_inode;
Jens Axboe03ba3782009-09-09 09:08:54 +02001094
1095 /*
1096 * If the inode was already on b_dirty/b_io/b_more_io, don't
1097 * reposition it (that would break b_dirty time-ordering).
1098 */
1099 if (!was_dirty) {
Dave Chinnera66979a2011-03-22 22:23:41 +11001100 bool wakeup_bdi = false;
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001101 bdi = inode_to_bdi(inode);
Jens Axboe500b0672009-09-09 09:10:25 +02001102
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001103 if (bdi_cap_writeback_dirty(bdi)) {
1104 WARN(!test_bit(BDI_registered, &bdi->state),
1105 "bdi-%s not registered\n", bdi->name);
1106
1107 /*
1108 * If this is the first dirty inode for this
1109 * bdi, we have to wake-up the corresponding
1110 * bdi thread to make sure background
1111 * write-back happens later.
1112 */
1113 if (!wb_has_dirty_io(&bdi->wb))
1114 wakeup_bdi = true;
Jens Axboe500b0672009-09-09 09:10:25 +02001115 }
Jens Axboe03ba3782009-09-09 09:08:54 +02001116
Dave Chinnera66979a2011-03-22 22:23:41 +11001117 spin_unlock(&inode->i_lock);
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001118 spin_lock(&bdi->wb.list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001119 inode->dirtied_when = jiffies;
Nick Piggin7ccf19a2010-10-21 11:49:30 +11001120 list_move(&inode->i_wb_list, &bdi->wb.b_dirty);
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001121 spin_unlock(&bdi->wb.list_lock);
Dave Chinnera66979a2011-03-22 22:23:41 +11001122
1123 if (wakeup_bdi)
1124 bdi_wakeup_thread_delayed(bdi);
1125 return;
Jens Axboe03ba3782009-09-09 09:08:54 +02001126 }
1127 }
Dave Chinner250df6e2011-03-22 22:23:36 +11001128out_unlock_inode:
1129 spin_unlock(&inode->i_lock);
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001130
Jens Axboe03ba3782009-09-09 09:08:54 +02001131}
1132EXPORT_SYMBOL(__mark_inode_dirty);
1133
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001134/*
1135 * Write out a superblock's list of dirty inodes. A wait will be performed
1136 * upon no inodes, all inodes or the final one, depending upon sync_mode.
1137 *
1138 * If older_than_this is non-NULL, then only write out inodes which
1139 * had their first dirtying at a time earlier than *older_than_this.
1140 *
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001141 * If `bdi' is non-zero then we're being asked to writeback a specific queue.
1142 * This function assumes that the blockdev superblock's inodes are backed by
1143 * a variety of queues, so all inodes are searched. For other superblocks,
1144 * assume that all inodes are backed by the same queue.
1145 *
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001146 * The inodes to be written are parked on bdi->b_io. They are moved back onto
1147 * bdi->b_dirty as they are selected for writing. This way, none can be missed
1148 * on the writer throttling path, and we get decent balancing between many
1149 * throttled threads: we don't want them all piling up on inode_sync_wait.
1150 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001151static void wait_sb_inodes(struct super_block *sb)
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001152{
Jens Axboe03ba3782009-09-09 09:08:54 +02001153 struct inode *inode, *old_inode = NULL;
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001154
Jens Axboe03ba3782009-09-09 09:08:54 +02001155 /*
1156 * We need to be protected against the filesystem going from
1157 * r/o to r/w or vice versa.
1158 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001159 WARN_ON(!rwsem_is_locked(&sb->s_umount));
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001160
Dave Chinner55fa6092011-03-22 22:23:40 +11001161 spin_lock(&inode_sb_list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001162
1163 /*
1164 * Data integrity sync. Must wait for all pages under writeback,
1165 * because there may have been pages dirtied before our sync
1166 * call, but which had writeout started before we write it out.
1167 * In which case, the inode may not be on the dirty list, but
1168 * we still have to wait for that writeout.
1169 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001170 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
Dave Chinner250df6e2011-03-22 22:23:36 +11001171 struct address_space *mapping = inode->i_mapping;
Jens Axboe03ba3782009-09-09 09:08:54 +02001172
Dave Chinner250df6e2011-03-22 22:23:36 +11001173 spin_lock(&inode->i_lock);
1174 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
1175 (mapping->nrpages == 0)) {
1176 spin_unlock(&inode->i_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001177 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +11001178 }
Jens Axboe03ba3782009-09-09 09:08:54 +02001179 __iget(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +11001180 spin_unlock(&inode->i_lock);
Dave Chinner55fa6092011-03-22 22:23:40 +11001181 spin_unlock(&inode_sb_list_lock);
1182
Jens Axboe03ba3782009-09-09 09:08:54 +02001183 /*
Dave Chinner55fa6092011-03-22 22:23:40 +11001184 * We hold a reference to 'inode' so it couldn't have been
1185 * removed from s_inodes list while we dropped the
1186 * inode_sb_list_lock. We cannot iput the inode now as we can
1187 * be holding the last reference and we cannot iput it under
1188 * inode_sb_list_lock. So we keep the reference and iput it
1189 * later.
Jens Axboe03ba3782009-09-09 09:08:54 +02001190 */
1191 iput(old_inode);
1192 old_inode = inode;
1193
1194 filemap_fdatawait(mapping);
1195
1196 cond_resched();
Nick Piggin38f21972009-01-06 14:40:25 -08001197
Dave Chinner55fa6092011-03-22 22:23:40 +11001198 spin_lock(&inode_sb_list_lock);
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001199 }
Dave Chinner55fa6092011-03-22 22:23:40 +11001200 spin_unlock(&inode_sb_list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001201 iput(old_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202}
1203
Jens Axboed8a85592009-09-02 12:34:32 +02001204/**
Chris Mason3259f8b2010-10-29 11:16:17 -04001205 * writeback_inodes_sb_nr - writeback dirty inodes from given super_block
Jens Axboed8a85592009-09-02 12:34:32 +02001206 * @sb: the superblock
Chris Mason3259f8b2010-10-29 11:16:17 -04001207 * @nr: the number of pages to write
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208 *
Jens Axboed8a85592009-09-02 12:34:32 +02001209 * Start writeback on some inodes on this super_block. No guarantees are made
1210 * on how many (if any) will be written, and this function does not wait
Chris Mason3259f8b2010-10-29 11:16:17 -04001211 * for IO completion of submitted IO.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212 */
Chris Mason3259f8b2010-10-29 11:16:17 -04001213void writeback_inodes_sb_nr(struct super_block *sb, unsigned long nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001215 DECLARE_COMPLETION_ONSTACK(done);
1216 struct wb_writeback_work work = {
Wu Fengguang6e6938b2010-06-06 10:38:15 -06001217 .sb = sb,
1218 .sync_mode = WB_SYNC_NONE,
1219 .tagged_writepages = 1,
1220 .done = &done,
1221 .nr_pages = nr,
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001222 };
Jens Axboe0e3c9a22010-06-01 11:08:43 +02001223
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001224 WARN_ON(!rwsem_is_locked(&sb->s_umount));
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001225 bdi_queue_work(sb->s_bdi, &work);
1226 wait_for_completion(&done);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227}
Chris Mason3259f8b2010-10-29 11:16:17 -04001228EXPORT_SYMBOL(writeback_inodes_sb_nr);
1229
1230/**
1231 * writeback_inodes_sb - writeback dirty inodes from given super_block
1232 * @sb: the superblock
1233 *
1234 * Start writeback on some inodes on this super_block. No guarantees are made
1235 * on how many (if any) will be written, and this function does not wait
1236 * for IO completion of submitted IO.
1237 */
1238void writeback_inodes_sb(struct super_block *sb)
1239{
Linus Torvalds925d1692010-10-30 09:05:48 -07001240 return writeback_inodes_sb_nr(sb, get_nr_dirty_pages());
Chris Mason3259f8b2010-10-29 11:16:17 -04001241}
Jens Axboed8a85592009-09-02 12:34:32 +02001242EXPORT_SYMBOL(writeback_inodes_sb);
1243
1244/**
Eric Sandeen17bd55d2009-12-23 07:57:07 -05001245 * writeback_inodes_sb_if_idle - start writeback if none underway
1246 * @sb: the superblock
1247 *
1248 * Invoke writeback_inodes_sb if no writeback is currently underway.
1249 * Returns 1 if writeback was started, 0 if not.
1250 */
1251int writeback_inodes_sb_if_idle(struct super_block *sb)
1252{
1253 if (!writeback_in_progress(sb->s_bdi)) {
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001254 down_read(&sb->s_umount);
Eric Sandeen17bd55d2009-12-23 07:57:07 -05001255 writeback_inodes_sb(sb);
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001256 up_read(&sb->s_umount);
Eric Sandeen17bd55d2009-12-23 07:57:07 -05001257 return 1;
1258 } else
1259 return 0;
1260}
1261EXPORT_SYMBOL(writeback_inodes_sb_if_idle);
1262
1263/**
Chris Mason3259f8b2010-10-29 11:16:17 -04001264 * writeback_inodes_sb_if_idle - start writeback if none underway
1265 * @sb: the superblock
1266 * @nr: the number of pages to write
1267 *
1268 * Invoke writeback_inodes_sb if no writeback is currently underway.
1269 * Returns 1 if writeback was started, 0 if not.
1270 */
1271int writeback_inodes_sb_nr_if_idle(struct super_block *sb,
1272 unsigned long nr)
1273{
1274 if (!writeback_in_progress(sb->s_bdi)) {
1275 down_read(&sb->s_umount);
1276 writeback_inodes_sb_nr(sb, nr);
1277 up_read(&sb->s_umount);
1278 return 1;
1279 } else
1280 return 0;
1281}
1282EXPORT_SYMBOL(writeback_inodes_sb_nr_if_idle);
1283
1284/**
Jens Axboed8a85592009-09-02 12:34:32 +02001285 * sync_inodes_sb - sync sb inode pages
1286 * @sb: the superblock
1287 *
1288 * This function writes and waits on any dirty inode belonging to this
Stefan Hajnoczicb9ef8d2011-01-13 15:47:26 -08001289 * super_block.
Jens Axboed8a85592009-09-02 12:34:32 +02001290 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001291void sync_inodes_sb(struct super_block *sb)
Jens Axboed8a85592009-09-02 12:34:32 +02001292{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001293 DECLARE_COMPLETION_ONSTACK(done);
1294 struct wb_writeback_work work = {
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001295 .sb = sb,
1296 .sync_mode = WB_SYNC_ALL,
1297 .nr_pages = LONG_MAX,
1298 .range_cyclic = 0,
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001299 .done = &done,
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001300 };
1301
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001302 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1303
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001304 bdi_queue_work(sb->s_bdi, &work);
1305 wait_for_completion(&done);
1306
Jens Axboeb6e51312009-09-16 15:13:54 +02001307 wait_sb_inodes(sb);
Jens Axboed8a85592009-09-02 12:34:32 +02001308}
1309EXPORT_SYMBOL(sync_inodes_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311/**
Andrea Arcangeli7f04c262005-10-30 15:03:05 -08001312 * write_inode_now - write an inode to disk
1313 * @inode: inode to write to disk
1314 * @sync: whether the write should be synchronous or not
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315 *
Andrea Arcangeli7f04c262005-10-30 15:03:05 -08001316 * This function commits an inode to disk immediately if it is dirty. This is
1317 * primarily needed by knfsd.
1318 *
1319 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321int write_inode_now(struct inode *inode, int sync)
1322{
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001323 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324 int ret;
1325 struct writeback_control wbc = {
1326 .nr_to_write = LONG_MAX,
Mike Galbraith18914b12008-02-08 04:20:23 -08001327 .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -07001328 .range_start = 0,
1329 .range_end = LLONG_MAX,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330 };
1331
1332 if (!mapping_cap_writeback_dirty(inode->i_mapping))
Andrew Morton49364ce2005-11-07 00:59:15 -08001333 wbc.nr_to_write = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334
1335 might_sleep();
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001336 spin_lock(&wb->list_lock);
Dave Chinner0f1b1fd2011-03-22 22:23:43 +11001337 spin_lock(&inode->i_lock);
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001338 ret = writeback_single_inode(inode, wb, &wbc);
Dave Chinner0f1b1fd2011-03-22 22:23:43 +11001339 spin_unlock(&inode->i_lock);
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001340 spin_unlock(&wb->list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341 if (sync)
Joern Engel1c0eeaf2007-10-16 23:30:44 -07001342 inode_sync_wait(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 return ret;
1344}
1345EXPORT_SYMBOL(write_inode_now);
1346
1347/**
1348 * sync_inode - write an inode and its pages to disk.
1349 * @inode: the inode to sync
1350 * @wbc: controls the writeback mode
1351 *
1352 * sync_inode() will write an inode and its pages to disk. It will also
1353 * correctly update the inode on its superblock's dirty inode lists and will
1354 * update inode->i_state.
1355 *
1356 * The caller must have a ref on the inode.
1357 */
1358int sync_inode(struct inode *inode, struct writeback_control *wbc)
1359{
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001360 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361 int ret;
1362
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001363 spin_lock(&wb->list_lock);
Dave Chinner0f1b1fd2011-03-22 22:23:43 +11001364 spin_lock(&inode->i_lock);
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001365 ret = writeback_single_inode(inode, wb, wbc);
Dave Chinner0f1b1fd2011-03-22 22:23:43 +11001366 spin_unlock(&inode->i_lock);
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001367 spin_unlock(&wb->list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368 return ret;
1369}
1370EXPORT_SYMBOL(sync_inode);
Christoph Hellwigc37650162010-10-06 10:48:20 +02001371
1372/**
Andrew Mortonc691b9d2011-01-13 15:45:48 -08001373 * sync_inode_metadata - write an inode to disk
Christoph Hellwigc37650162010-10-06 10:48:20 +02001374 * @inode: the inode to sync
1375 * @wait: wait for I/O to complete.
1376 *
Andrew Mortonc691b9d2011-01-13 15:45:48 -08001377 * Write an inode to disk and adjust its dirty state after completion.
Christoph Hellwigc37650162010-10-06 10:48:20 +02001378 *
1379 * Note: only writes the actual inode, no associated data or other metadata.
1380 */
1381int sync_inode_metadata(struct inode *inode, int wait)
1382{
1383 struct writeback_control wbc = {
1384 .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE,
1385 .nr_to_write = 0, /* metadata-only */
1386 };
1387
1388 return sync_inode(inode, &wbc);
1389}
1390EXPORT_SYMBOL(sync_inode_metadata);