blob: 539f36cf3e4a3b574f6bb977bcbdf387571675b3 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * fs/fs-writeback.c
3 *
4 * Copyright (C) 2002, Linus Torvalds.
5 *
6 * Contains all the functions related to writing back and waiting
7 * upon dirty inodes against superblocks, and writing back dirty
8 * pages against inodes. ie: data writeback. Writeout of the
9 * inode itself is not handled here.
10 *
Francois Camie1f8e872008-10-15 22:01:59 -070011 * 10Apr2002 Andrew Morton
Linus Torvalds1da177e2005-04-16 15:20:36 -070012 * Split out of fs/inode.c
13 * Additions for address_space-based writeback
14 */
15
16#include <linux/kernel.h>
Paul Gortmaker630d9c42011-11-16 23:57:37 -050017#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include <linux/spinlock.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090019#include <linux/slab.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/sched.h>
21#include <linux/fs.h>
22#include <linux/mm.h>
Wu Fengguangbc31b862012-01-07 20:41:55 -060023#include <linux/pagemap.h>
Jens Axboe03ba3782009-09-09 09:08:54 +020024#include <linux/kthread.h>
25#include <linux/freezer.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include <linux/writeback.h>
27#include <linux/blkdev.h>
28#include <linux/backing-dev.h>
Dave Chinner455b2862010-07-07 13:24:06 +100029#include <linux/tracepoint.h>
David Howells07f3f052006-09-30 20:52:18 +020030#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070031
Jens Axboed0bceac2009-05-18 08:20:32 +020032/*
Wu Fengguangbc31b862012-01-07 20:41:55 -060033 * 4MB minimal write chunk size
34 */
35#define MIN_WRITEBACK_PAGES (4096UL >> (PAGE_CACHE_SHIFT - 10))
36
37/*
Jens Axboec4a77a62009-09-16 15:18:25 +020038 * Passed into wb_writeback(), essentially a subset of writeback_control
39 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +020040struct wb_writeback_work {
Jens Axboec4a77a62009-09-16 15:18:25 +020041 long nr_pages;
42 struct super_block *sb;
Wu Fengguangd46db3d2011-05-04 19:54:37 -060043 unsigned long *older_than_this;
Jens Axboec4a77a62009-09-16 15:18:25 +020044 enum writeback_sync_modes sync_mode;
Wu Fengguang6e6938b2010-06-06 10:38:15 -060045 unsigned int tagged_writepages:1;
H Hartley Sweeten52957fe2010-04-01 20:36:30 -050046 unsigned int for_kupdate:1;
47 unsigned int range_cyclic:1;
48 unsigned int for_background:1;
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -060049 enum wb_reason reason; /* why was writeback initiated? */
Jens Axboec4a77a62009-09-16 15:18:25 +020050
Jens Axboe8010c3b2009-09-15 20:04:57 +020051 struct list_head list; /* pending work list */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +020052 struct completion *done; /* set if the caller waits */
Jens Axboe03ba3782009-09-09 09:08:54 +020053};
54
Dave Chinner455b2862010-07-07 13:24:06 +100055/*
Dave Chinner455b2862010-07-07 13:24:06 +100056 * We don't actually have pdflush, but this one is exported though /proc...
57 */
58int nr_pdflush_threads;
59
Adrian Bunkf11b00f2008-04-29 00:58:56 -070060/**
61 * writeback_in_progress - determine whether there is writeback in progress
62 * @bdi: the device's backing_dev_info structure.
63 *
Jens Axboe03ba3782009-09-09 09:08:54 +020064 * Determine whether there is writeback waiting to be handled against a
65 * backing device.
Adrian Bunkf11b00f2008-04-29 00:58:56 -070066 */
67int writeback_in_progress(struct backing_dev_info *bdi)
68{
Jan Kara81d73a32010-08-11 14:17:44 -070069 return test_bit(BDI_writeback_running, &bdi->state);
Adrian Bunkf11b00f2008-04-29 00:58:56 -070070}
71
Jan Kara692ebd12010-09-21 11:51:01 +020072static inline struct backing_dev_info *inode_to_bdi(struct inode *inode)
73{
74 struct super_block *sb = inode->i_sb;
Jan Kara692ebd12010-09-21 11:51:01 +020075
Christoph Hellwigaaead252010-10-04 14:25:33 +020076 if (strcmp(sb->s_type->name, "bdev") == 0)
77 return inode->i_mapping->backing_dev_info;
78
79 return sb->s_bdi;
Jan Kara692ebd12010-09-21 11:51:01 +020080}
81
Nick Piggin7ccf19a2010-10-21 11:49:30 +110082static inline struct inode *wb_inode(struct list_head *head)
83{
84 return list_entry(head, struct inode, i_wb_list);
85}
86
Wu Fengguang15eb77a2012-01-17 11:18:56 -060087/*
88 * Include the creation of the trace points after defining the
89 * wb_writeback_work structure and inline functions so that the definition
90 * remains local to this file.
91 */
92#define CREATE_TRACE_POINTS
93#include <trace/events/writeback.h>
94
Jan Kara65850272011-01-13 15:45:44 -080095/* Wakeup flusher thread or forker thread to fork it. Requires bdi->wb_lock. */
96static void bdi_wakeup_flusher(struct backing_dev_info *bdi)
Nick Piggin4195f732009-05-28 09:01:15 +020097{
Artem Bityutskiyfff5b852010-07-25 14:29:20 +030098 if (bdi->wb.task) {
99 wake_up_process(bdi->wb.task);
100 } else {
101 /*
102 * The bdi thread isn't there, wake up the forker thread which
103 * will create and run it.
104 */
Jens Axboe03ba3782009-09-09 09:08:54 +0200105 wake_up_process(default_backing_dev_info.wb.task);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106 }
Jan Kara65850272011-01-13 15:45:44 -0800107}
108
109static void bdi_queue_work(struct backing_dev_info *bdi,
110 struct wb_writeback_work *work)
111{
112 trace_writeback_queue(bdi, work);
113
114 spin_lock_bh(&bdi->wb_lock);
115 list_add_tail(&work->list, &bdi->work_list);
116 if (!bdi->wb.task)
117 trace_writeback_nothread(bdi, work);
118 bdi_wakeup_flusher(bdi);
Artem Bityutskiy64677162010-07-25 14:29:22 +0300119 spin_unlock_bh(&bdi->wb_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120}
121
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200122static void
123__bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600124 bool range_cyclic, enum wb_reason reason)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200126 struct wb_writeback_work *work;
Jens Axboe03ba3782009-09-09 09:08:54 +0200127
Jens Axboebcddc3f2009-09-13 20:07:36 +0200128 /*
129 * This is WB_SYNC_NONE writeback, so if allocation fails just
130 * wakeup the thread for old dirty data writeback
131 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200132 work = kzalloc(sizeof(*work), GFP_ATOMIC);
133 if (!work) {
Dave Chinner455b2862010-07-07 13:24:06 +1000134 if (bdi->wb.task) {
135 trace_writeback_nowork(bdi);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200136 wake_up_process(bdi->wb.task);
Dave Chinner455b2862010-07-07 13:24:06 +1000137 }
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200138 return;
Jens Axboebcddc3f2009-09-13 20:07:36 +0200139 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200140
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200141 work->sync_mode = WB_SYNC_NONE;
142 work->nr_pages = nr_pages;
143 work->range_cyclic = range_cyclic;
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600144 work->reason = reason;
Christoph Hellwigf0fad8a2009-09-11 09:47:56 +0200145
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200146 bdi_queue_work(bdi, work);
Jens Axboeb6e51312009-09-16 15:13:54 +0200147}
148
149/**
150 * bdi_start_writeback - start writeback
151 * @bdi: the backing device to write from
152 * @nr_pages: the number of pages to write
Marcos Paulo de Souza786228a2011-11-23 20:56:45 +0800153 * @reason: reason why some writeback work was initiated
Jens Axboeb6e51312009-09-16 15:13:54 +0200154 *
155 * Description:
156 * This does WB_SYNC_NONE opportunistic writeback. The IO is only
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300157 * started when this function returns, we make no guarantees on
Jens Axboe0e3c9a22010-06-01 11:08:43 +0200158 * completion. Caller need not hold sb s_umount semaphore.
Jens Axboeb6e51312009-09-16 15:13:54 +0200159 *
160 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600161void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages,
162 enum wb_reason reason)
Jens Axboeb6e51312009-09-16 15:13:54 +0200163{
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600164 __bdi_start_writeback(bdi, nr_pages, true, reason);
Christoph Hellwigc5444192010-06-08 18:15:15 +0200165}
Wu Fengguangd3ddec72009-09-23 20:33:40 +0800166
Christoph Hellwigc5444192010-06-08 18:15:15 +0200167/**
168 * bdi_start_background_writeback - start background writeback
169 * @bdi: the backing device to write from
170 *
171 * Description:
Jan Kara65850272011-01-13 15:45:44 -0800172 * This makes sure WB_SYNC_NONE background writeback happens. When
173 * this function returns, it is only guaranteed that for given BDI
174 * some IO is happening if we are over background dirty threshold.
175 * Caller need not hold sb s_umount semaphore.
Christoph Hellwigc5444192010-06-08 18:15:15 +0200176 */
177void bdi_start_background_writeback(struct backing_dev_info *bdi)
178{
Jan Kara65850272011-01-13 15:45:44 -0800179 /*
180 * We just wake up the flusher thread. It will perform background
181 * writeback as soon as there is no other work to do.
182 */
Wu Fengguang71927e82011-01-13 15:45:46 -0800183 trace_writeback_wake_background(bdi);
Jan Kara65850272011-01-13 15:45:44 -0800184 spin_lock_bh(&bdi->wb_lock);
185 bdi_wakeup_flusher(bdi);
186 spin_unlock_bh(&bdi->wb_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187}
188
189/*
Dave Chinnera66979a2011-03-22 22:23:41 +1100190 * Remove the inode from the writeback list it is on.
191 */
192void inode_wb_list_del(struct inode *inode)
193{
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600194 struct backing_dev_info *bdi = inode_to_bdi(inode);
Dave Chinnera66979a2011-03-22 22:23:41 +1100195
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600196 spin_lock(&bdi->wb.list_lock);
197 list_del_init(&inode->i_wb_list);
198 spin_unlock(&bdi->wb.list_lock);
199}
Dave Chinnera66979a2011-03-22 22:23:41 +1100200
201/*
Andrew Morton6610a0b2007-10-16 23:30:32 -0700202 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
203 * furthest end of its superblock's dirty-inode list.
204 *
205 * Before stamping the inode's ->dirtied_when, we check to see whether it is
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200206 * already the most-recently-dirtied inode on the b_dirty list. If that is
Andrew Morton6610a0b2007-10-16 23:30:32 -0700207 * the case then the inode must have been redirtied while it was being written
208 * out and we don't reset its dirtied_when.
209 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600210static void redirty_tail(struct inode *inode, struct bdi_writeback *wb)
Andrew Morton6610a0b2007-10-16 23:30:32 -0700211{
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600212 assert_spin_locked(&wb->list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200213 if (!list_empty(&wb->b_dirty)) {
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200214 struct inode *tail;
Andrew Morton6610a0b2007-10-16 23:30:32 -0700215
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100216 tail = wb_inode(wb->b_dirty.next);
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200217 if (time_before(inode->dirtied_when, tail->dirtied_when))
Andrew Morton6610a0b2007-10-16 23:30:32 -0700218 inode->dirtied_when = jiffies;
219 }
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100220 list_move(&inode->i_wb_list, &wb->b_dirty);
Andrew Morton6610a0b2007-10-16 23:30:32 -0700221}
222
223/*
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200224 * requeue inode for re-scanning after bdi->b_io list is exhausted.
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700225 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600226static void requeue_io(struct inode *inode, struct bdi_writeback *wb)
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700227{
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600228 assert_spin_locked(&wb->list_lock);
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100229 list_move(&inode->i_wb_list, &wb->b_more_io);
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700230}
231
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700232static void inode_sync_complete(struct inode *inode)
233{
234 /*
Dave Chinnera66979a2011-03-22 22:23:41 +1100235 * Prevent speculative execution through
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600236 * spin_unlock(&wb->list_lock);
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700237 */
Dave Chinnera66979a2011-03-22 22:23:41 +1100238
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700239 smp_mb();
240 wake_up_bit(&inode->i_state, __I_SYNC);
241}
242
Jeff Laytond2caa3c2009-04-02 16:56:37 -0700243static bool inode_dirtied_after(struct inode *inode, unsigned long t)
244{
245 bool ret = time_after(inode->dirtied_when, t);
246#ifndef CONFIG_64BIT
247 /*
248 * For inodes being constantly redirtied, dirtied_when can get stuck.
249 * It _appears_ to be in the future, but is actually in distant past.
250 * This test is necessary to prevent such wrapped-around relative times
Jens Axboe5b0830c2009-09-23 19:37:09 +0200251 * from permanently stopping the whole bdi writeback.
Jeff Laytond2caa3c2009-04-02 16:56:37 -0700252 */
253 ret = ret && time_before_eq(inode->dirtied_when, jiffies);
254#endif
255 return ret;
256}
257
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700258/*
Jan Kara697e6fe2012-03-09 07:26:22 -0800259 * Move expired (dirtied after work->older_than_this) dirty inodes from
260 * @delaying_queue to @dispatch_queue.
Fengguang Wu2c136572007-10-16 23:30:39 -0700261 */
Wu Fengguange84d0a42011-04-23 12:27:27 -0600262static int move_expired_inodes(struct list_head *delaying_queue,
Fengguang Wu2c136572007-10-16 23:30:39 -0700263 struct list_head *dispatch_queue,
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600264 struct wb_writeback_work *work)
Fengguang Wu2c136572007-10-16 23:30:39 -0700265{
Shaohua Li5c034492009-09-24 14:42:33 +0200266 LIST_HEAD(tmp);
267 struct list_head *pos, *node;
Jens Axboecf137302009-09-24 15:12:57 +0200268 struct super_block *sb = NULL;
Shaohua Li5c034492009-09-24 14:42:33 +0200269 struct inode *inode;
Jens Axboecf137302009-09-24 15:12:57 +0200270 int do_sb_sort = 0;
Wu Fengguange84d0a42011-04-23 12:27:27 -0600271 int moved = 0;
Shaohua Li5c034492009-09-24 14:42:33 +0200272
Fengguang Wu2c136572007-10-16 23:30:39 -0700273 while (!list_empty(delaying_queue)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100274 inode = wb_inode(delaying_queue->prev);
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600275 if (work->older_than_this &&
276 inode_dirtied_after(inode, *work->older_than_this))
Fengguang Wu2c136572007-10-16 23:30:39 -0700277 break;
Jens Axboecf137302009-09-24 15:12:57 +0200278 if (sb && sb != inode->i_sb)
279 do_sb_sort = 1;
280 sb = inode->i_sb;
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100281 list_move(&inode->i_wb_list, &tmp);
Wu Fengguange84d0a42011-04-23 12:27:27 -0600282 moved++;
Shaohua Li5c034492009-09-24 14:42:33 +0200283 }
284
Jens Axboecf137302009-09-24 15:12:57 +0200285 /* just one sb in list, splice to dispatch_queue and we're done */
286 if (!do_sb_sort) {
287 list_splice(&tmp, dispatch_queue);
Wu Fengguange84d0a42011-04-23 12:27:27 -0600288 goto out;
Jens Axboecf137302009-09-24 15:12:57 +0200289 }
290
Shaohua Li5c034492009-09-24 14:42:33 +0200291 /* Move inodes from one superblock together */
292 while (!list_empty(&tmp)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100293 sb = wb_inode(tmp.prev)->i_sb;
Shaohua Li5c034492009-09-24 14:42:33 +0200294 list_for_each_prev_safe(pos, node, &tmp) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100295 inode = wb_inode(pos);
Shaohua Li5c034492009-09-24 14:42:33 +0200296 if (inode->i_sb == sb)
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100297 list_move(&inode->i_wb_list, dispatch_queue);
Shaohua Li5c034492009-09-24 14:42:33 +0200298 }
Fengguang Wu2c136572007-10-16 23:30:39 -0700299 }
Wu Fengguange84d0a42011-04-23 12:27:27 -0600300out:
301 return moved;
Fengguang Wu2c136572007-10-16 23:30:39 -0700302}
303
304/*
305 * Queue all expired dirty inodes for io, eldest first.
Wu Fengguang4ea879b2010-08-11 14:17:42 -0700306 * Before
307 * newly dirtied b_dirty b_io b_more_io
308 * =============> gf edc BA
309 * After
310 * newly dirtied b_dirty b_io b_more_io
311 * =============> g fBAedc
312 * |
313 * +--> dequeue for IO
Fengguang Wu2c136572007-10-16 23:30:39 -0700314 */
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600315static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work)
Fengguang Wu2c136572007-10-16 23:30:39 -0700316{
Wu Fengguange84d0a42011-04-23 12:27:27 -0600317 int moved;
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600318 assert_spin_locked(&wb->list_lock);
Wu Fengguang4ea879b2010-08-11 14:17:42 -0700319 list_splice_init(&wb->b_more_io, &wb->b_io);
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600320 moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, work);
321 trace_writeback_queue_io(wb, work, moved);
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200322}
323
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100324static int write_inode(struct inode *inode, struct writeback_control *wbc)
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200325{
Jens Axboe03ba3782009-09-09 09:08:54 +0200326 if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode))
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100327 return inode->i_sb->s_op->write_inode(inode, wbc);
Jens Axboe03ba3782009-09-09 09:08:54 +0200328 return 0;
Fengguang Wu2c136572007-10-16 23:30:39 -0700329}
330
331/*
Christoph Hellwig01c03192009-06-08 13:35:40 +0200332 * Wait for writeback on an inode to complete.
333 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600334static void inode_wait_for_writeback(struct inode *inode,
335 struct bdi_writeback *wb)
Christoph Hellwig01c03192009-06-08 13:35:40 +0200336{
337 DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC);
338 wait_queue_head_t *wqh;
339
340 wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
Dave Chinner250df6e2011-03-22 22:23:36 +1100341 while (inode->i_state & I_SYNC) {
342 spin_unlock(&inode->i_lock);
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600343 spin_unlock(&wb->list_lock);
Christoph Hellwig01c03192009-06-08 13:35:40 +0200344 __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE);
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600345 spin_lock(&wb->list_lock);
Dave Chinner250df6e2011-03-22 22:23:36 +1100346 spin_lock(&inode->i_lock);
Richard Kennedy58a9d3d2010-05-24 14:32:38 -0700347 }
Christoph Hellwig01c03192009-06-08 13:35:40 +0200348}
349
350/*
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600351 * Write out an inode's dirty pages. Called under wb->list_lock and
Dave Chinner0f1b1fd2011-03-22 22:23:43 +1100352 * inode->i_lock. Either the caller has an active reference on the inode or
353 * the inode has I_WILL_FREE set.
Christoph Hellwig01c03192009-06-08 13:35:40 +0200354 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355 * If `wait' is set, wait on the writeout.
356 *
357 * The whole writeout design is quite complex and fragile. We want to avoid
358 * starvation of particular inodes when others are being redirtied, prevent
359 * livelocks, etc.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360 */
361static int
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600362writeback_single_inode(struct inode *inode, struct bdi_writeback *wb,
363 struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365 struct address_space *mapping = inode->i_mapping;
Wu Fengguang251d6a42010-12-01 17:33:37 -0600366 long nr_to_write = wbc->nr_to_write;
Christoph Hellwig01c03192009-06-08 13:35:40 +0200367 unsigned dirty;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368 int ret;
369
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600370 assert_spin_locked(&wb->list_lock);
Dave Chinner0f1b1fd2011-03-22 22:23:43 +1100371 assert_spin_locked(&inode->i_lock);
372
Christoph Hellwig01c03192009-06-08 13:35:40 +0200373 if (!atomic_read(&inode->i_count))
374 WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING)));
375 else
376 WARN_ON(inode->i_state & I_WILL_FREE);
377
378 if (inode->i_state & I_SYNC) {
379 /*
380 * If this inode is locked for writeback and we are not doing
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200381 * writeback-for-data-integrity, move it to b_more_io so that
Christoph Hellwig01c03192009-06-08 13:35:40 +0200382 * writeback can proceed with the other inodes on s_io.
383 *
384 * We'll have another go at writing back this inode when we
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200385 * completed a full scan of b_io.
Christoph Hellwig01c03192009-06-08 13:35:40 +0200386 */
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100387 if (wbc->sync_mode != WB_SYNC_ALL) {
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600388 requeue_io(inode, wb);
Wu Fengguang251d6a42010-12-01 17:33:37 -0600389 trace_writeback_single_inode_requeue(inode, wbc,
390 nr_to_write);
Christoph Hellwig01c03192009-06-08 13:35:40 +0200391 return 0;
392 }
393
394 /*
395 * It's a data-integrity sync. We must wait.
396 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600397 inode_wait_for_writeback(inode, wb);
Christoph Hellwig01c03192009-06-08 13:35:40 +0200398 }
399
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700400 BUG_ON(inode->i_state & I_SYNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401
Dmitry Monakhov5547e8a2010-05-07 13:35:44 +0400402 /* Set I_SYNC, reset I_DIRTY_PAGES */
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700403 inode->i_state |= I_SYNC;
Dmitry Monakhov5547e8a2010-05-07 13:35:44 +0400404 inode->i_state &= ~I_DIRTY_PAGES;
Dave Chinner250df6e2011-03-22 22:23:36 +1100405 spin_unlock(&inode->i_lock);
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600406 spin_unlock(&wb->list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407
408 ret = do_writepages(mapping, wbc);
409
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100410 /*
411 * Make sure to wait on the data before writing out the metadata.
412 * This is important for filesystems that modify metadata on data
413 * I/O completion.
414 */
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100415 if (wbc->sync_mode == WB_SYNC_ALL) {
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100416 int err = filemap_fdatawait(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417 if (ret == 0)
418 ret = err;
419 }
420
Dmitry Monakhov5547e8a2010-05-07 13:35:44 +0400421 /*
422 * Some filesystems may redirty the inode during the writeback
423 * due to delalloc, clear dirty metadata flags right before
424 * write_inode()
425 */
Dave Chinner250df6e2011-03-22 22:23:36 +1100426 spin_lock(&inode->i_lock);
Dmitry Monakhov5547e8a2010-05-07 13:35:44 +0400427 dirty = inode->i_state & I_DIRTY;
428 inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC);
Dave Chinner250df6e2011-03-22 22:23:36 +1100429 spin_unlock(&inode->i_lock);
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100430 /* Don't write the inode if only I_DIRTY_PAGES was set */
431 if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100432 int err = write_inode(inode, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 if (ret == 0)
434 ret = err;
435 }
436
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600437 spin_lock(&wb->list_lock);
Dave Chinner250df6e2011-03-22 22:23:36 +1100438 spin_lock(&inode->i_lock);
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700439 inode->i_state &= ~I_SYNC;
Al Viroa4ffdde2010-06-02 17:38:30 -0400440 if (!(inode->i_state & I_FREEING)) {
Wu Fengguang94c3dcb2011-04-27 19:05:21 -0600441 /*
442 * Sync livelock prevention. Each inode is tagged and synced in
443 * one shot. If still dirty, it will be redirty_tail()'ed below.
444 * Update the dirty time to prevent enqueue and sync it again.
445 */
446 if ((inode->i_state & I_DIRTY) &&
447 (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages))
448 inode->dirtied_when = jiffies;
449
Wu Fengguang23539af2010-08-11 14:17:41 -0700450 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451 /*
452 * We didn't write back all the pages. nfs_writepages()
Wu Fengguanga50aeb42010-08-11 14:17:43 -0700453 * sometimes bales out without doing anything.
Andrew Morton1b43ef92007-10-16 23:30:35 -0700454 */
Wu Fengguanga50aeb42010-08-11 14:17:43 -0700455 inode->i_state |= I_DIRTY_PAGES;
456 if (wbc->nr_to_write <= 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457 /*
Wu Fengguanga50aeb42010-08-11 14:17:43 -0700458 * slice used up: queue for next turn
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600460 requeue_io(inode, wb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 } else {
462 /*
Wu Fengguanga50aeb42010-08-11 14:17:43 -0700463 * Writeback blocked by something other than
464 * congestion. Delay the inode for some time to
465 * avoid spinning on the CPU (100% iowait)
466 * retrying writeback of the dirty page/inode
467 * that cannot be performed immediately.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600469 redirty_tail(inode, wb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470 }
Wu Fengguang23539af2010-08-11 14:17:41 -0700471 } else if (inode->i_state & I_DIRTY) {
472 /*
473 * Filesystems can dirty the inode during writeback
474 * operations, such as delayed allocation during
475 * submission or metadata updates after data IO
476 * completion.
477 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600478 redirty_tail(inode, wb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479 } else {
480 /*
Nick Piggin9e38d862010-10-23 06:55:17 -0400481 * The inode is clean. At this point we either have
482 * a reference to the inode or it's on it's way out.
483 * No need to add it back to the LRU.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484 */
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100485 list_del_init(&inode->i_wb_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486 }
487 }
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700488 inode_sync_complete(inode);
Wu Fengguang251d6a42010-12-01 17:33:37 -0600489 trace_writeback_single_inode(inode, wbc, nr_to_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490 return ret;
491}
492
Wu Fengguang1a12d8b2010-08-29 13:28:09 -0600493static long writeback_chunk_size(struct backing_dev_info *bdi,
494 struct wb_writeback_work *work)
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600495{
496 long pages;
497
498 /*
499 * WB_SYNC_ALL mode does livelock avoidance by syncing dirty
500 * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX
501 * here avoids calling into writeback_inodes_wb() more than once.
502 *
503 * The intended call sequence for WB_SYNC_ALL writeback is:
504 *
505 * wb_writeback()
506 * writeback_sb_inodes() <== called only once
507 * write_cache_pages() <== called once for each inode
508 * (quickly) tag currently dirty pages
509 * (maybe slowly) sync all tagged pages
510 */
511 if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages)
512 pages = LONG_MAX;
Wu Fengguang1a12d8b2010-08-29 13:28:09 -0600513 else {
514 pages = min(bdi->avg_write_bandwidth / 2,
515 global_dirty_limit / DIRTY_SCOPE);
516 pages = min(pages, work->nr_pages);
517 pages = round_down(pages + MIN_WRITEBACK_PAGES,
518 MIN_WRITEBACK_PAGES);
519 }
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600520
521 return pages;
522}
523
Jens Axboe03ba3782009-09-09 09:08:54 +0200524/*
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800525 * Write a portion of b_io inodes which belong to @sb.
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200526 *
527 * If @only_this_sb is true, then find and write all such
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800528 * inodes. Otherwise write only ones which go sequentially
529 * in reverse order.
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200530 *
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600531 * Return the number of pages and/or inodes written.
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800532 */
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600533static long writeback_sb_inodes(struct super_block *sb,
534 struct bdi_writeback *wb,
535 struct wb_writeback_work *work)
Jens Axboe03ba3782009-09-09 09:08:54 +0200536{
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600537 struct writeback_control wbc = {
538 .sync_mode = work->sync_mode,
539 .tagged_writepages = work->tagged_writepages,
540 .for_kupdate = work->for_kupdate,
541 .for_background = work->for_background,
542 .range_cyclic = work->range_cyclic,
543 .range_start = 0,
544 .range_end = LLONG_MAX,
545 };
546 unsigned long start_time = jiffies;
547 long write_chunk;
548 long wrote = 0; /* count both pages and inodes */
549
Jens Axboe03ba3782009-09-09 09:08:54 +0200550 while (!list_empty(&wb->b_io)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100551 struct inode *inode = wb_inode(wb->b_io.prev);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200552
553 if (inode->i_sb != sb) {
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600554 if (work->sb) {
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200555 /*
556 * We only want to write back data for this
557 * superblock, move all inodes not belonging
558 * to it back onto the dirty list.
559 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600560 redirty_tail(inode, wb);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200561 continue;
562 }
563
564 /*
565 * The inode belongs to a different superblock.
566 * Bounce back to the caller to unpin this and
567 * pin the next superblock.
568 */
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600569 break;
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200570 }
571
Christoph Hellwig9843b762010-10-24 19:40:46 +0200572 /*
573 * Don't bother with new inodes or inodes beeing freed, first
574 * kind does not need peridic writeout yet, and for the latter
575 * kind writeout is handled by the freer.
576 */
Dave Chinner250df6e2011-03-22 22:23:36 +1100577 spin_lock(&inode->i_lock);
Christoph Hellwig9843b762010-10-24 19:40:46 +0200578 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
Dave Chinner250df6e2011-03-22 22:23:36 +1100579 spin_unlock(&inode->i_lock);
Wu Fengguangfcc5c222011-07-11 23:08:50 -0700580 redirty_tail(inode, wb);
Nick Piggin7ef0d732009-03-12 14:31:38 -0700581 continue;
582 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 __iget(inode);
Wu Fengguang1a12d8b2010-08-29 13:28:09 -0600584 write_chunk = writeback_chunk_size(wb->bdi, work);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600585 wbc.nr_to_write = write_chunk;
586 wbc.pages_skipped = 0;
Dave Chinner250df6e2011-03-22 22:23:36 +1100587
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600588 writeback_single_inode(inode, wb, &wbc);
589
590 work->nr_pages -= write_chunk - wbc.nr_to_write;
591 wrote += write_chunk - wbc.nr_to_write;
592 if (!(inode->i_state & I_DIRTY))
593 wrote++;
594 if (wbc.pages_skipped) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 /*
596 * writeback is not making progress due to locked
597 * buffers. Skip this inode for now.
598 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600599 redirty_tail(inode, wb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 }
Dave Chinner0f1b1fd2011-03-22 22:23:43 +1100601 spin_unlock(&inode->i_lock);
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600602 spin_unlock(&wb->list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 iput(inode);
OGAWA Hirofumi4ffc8442006-03-25 03:07:44 -0800604 cond_resched();
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600605 spin_lock(&wb->list_lock);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600606 /*
607 * bail out to wb_writeback() often enough to check
608 * background threshold and other termination conditions.
609 */
610 if (wrote) {
611 if (time_is_before_jiffies(start_time + HZ / 10UL))
612 break;
613 if (work->nr_pages <= 0)
614 break;
Fengguang Wu8bc3be22008-02-04 22:29:36 -0800615 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616 }
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600617 return wrote;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800618}
Nick Piggin38f21972009-01-06 14:40:25 -0800619
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600620static long __writeback_inodes_wb(struct bdi_writeback *wb,
621 struct wb_writeback_work *work)
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800622{
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600623 unsigned long start_time = jiffies;
624 long wrote = 0;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800625
626 while (!list_empty(&wb->b_io)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100627 struct inode *inode = wb_inode(wb->b_io.prev);
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800628 struct super_block *sb = inode->i_sb;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800629
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000630 if (!grab_super_passive(sb)) {
Wu Fengguang0e995812011-07-29 22:14:35 -0600631 /*
632 * grab_super_passive() may fail consistently due to
633 * s_umount being grabbed by someone else. Don't use
634 * requeue_io() to avoid busy retrying the inode/sb.
635 */
636 redirty_tail(inode, wb);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200637 continue;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800638 }
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600639 wrote += writeback_sb_inodes(sb, wb, work);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200640 drop_super(sb);
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800641
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600642 /* refer to the same tests at the end of writeback_sb_inodes */
643 if (wrote) {
644 if (time_is_before_jiffies(start_time + HZ / 10UL))
645 break;
646 if (work->nr_pages <= 0)
647 break;
648 }
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800649 }
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200650 /* Leave any unwritten inodes on b_io */
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600651 return wrote;
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200652}
653
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600654long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages,
655 enum wb_reason reason)
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200656{
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600657 struct wb_writeback_work work = {
658 .nr_pages = nr_pages,
659 .sync_mode = WB_SYNC_NONE,
660 .range_cyclic = 1,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600661 .reason = reason,
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600662 };
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200663
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600664 spin_lock(&wb->list_lock);
Wu Fengguang424b3512010-07-21 20:11:53 -0600665 if (list_empty(&wb->b_io))
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600666 queue_io(wb, &work);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600667 __writeback_inodes_wb(wb, &work);
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600668 spin_unlock(&wb->list_lock);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200669
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600670 return nr_pages - work.nr_pages;
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200671}
672
Wu Fengguangb00949a2010-11-18 14:38:33 -0600673static bool over_bground_thresh(struct backing_dev_info *bdi)
Jens Axboe03ba3782009-09-09 09:08:54 +0200674{
675 unsigned long background_thresh, dirty_thresh;
676
Wu Fengguang16c40422010-08-11 14:17:39 -0700677 global_dirty_limits(&background_thresh, &dirty_thresh);
Jens Axboe03ba3782009-09-09 09:08:54 +0200678
Wu Fengguangb00949a2010-11-18 14:38:33 -0600679 if (global_page_state(NR_FILE_DIRTY) +
680 global_page_state(NR_UNSTABLE_NFS) > background_thresh)
681 return true;
682
683 if (bdi_stat(bdi, BDI_RECLAIMABLE) >
684 bdi_dirty_limit(bdi, background_thresh))
685 return true;
686
687 return false;
Jens Axboe03ba3782009-09-09 09:08:54 +0200688}
689
690/*
Wu Fengguange98be2d2010-08-29 11:22:30 -0600691 * Called under wb->list_lock. If there are multiple wb per bdi,
692 * only the flusher working on the first wb should do it.
693 */
694static void wb_update_bandwidth(struct bdi_writeback *wb,
695 unsigned long start_time)
696{
Wu Fengguangaf6a3112011-10-03 20:46:17 -0600697 __bdi_update_bandwidth(wb->bdi, 0, 0, 0, 0, 0, start_time);
Wu Fengguange98be2d2010-08-29 11:22:30 -0600698}
699
700/*
Jens Axboe03ba3782009-09-09 09:08:54 +0200701 * Explicit flushing or periodic writeback of "old" data.
702 *
703 * Define "old": the first time one of an inode's pages is dirtied, we mark the
704 * dirtying-time in the inode's address_space. So this periodic writeback code
705 * just walks the superblock inode list, writing back any inodes which are
706 * older than a specific point in time.
707 *
708 * Try to run once per dirty_writeback_interval. But if a writeback event
709 * takes longer than a dirty_writeback_interval interval, then leave a
710 * one-second gap.
711 *
712 * older_than_this takes precedence over nr_to_write. So we'll only write back
713 * all dirty pages if they are all attached to "old" mappings.
714 */
Jens Axboec4a77a62009-09-16 15:18:25 +0200715static long wb_writeback(struct bdi_writeback *wb,
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200716 struct wb_writeback_work *work)
Jens Axboe03ba3782009-09-09 09:08:54 +0200717{
Wu Fengguange98be2d2010-08-29 11:22:30 -0600718 unsigned long wb_start = jiffies;
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600719 long nr_pages = work->nr_pages;
Jens Axboe03ba3782009-09-09 09:08:54 +0200720 unsigned long oldest_jif;
Jan Karaa5989bd2009-09-16 19:22:48 +0200721 struct inode *inode;
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600722 long progress;
Jens Axboe03ba3782009-09-09 09:08:54 +0200723
Wu Fengguange185dda2011-04-23 11:26:07 -0600724 oldest_jif = jiffies;
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600725 work->older_than_this = &oldest_jif;
Jens Axboe03ba3782009-09-09 09:08:54 +0200726
Wu Fengguange8dfc302011-04-21 12:06:32 -0600727 spin_lock(&wb->list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200728 for (;;) {
729 /*
Wu Fengguangd3ddec72009-09-23 20:33:40 +0800730 * Stop writeback when nr_pages has been consumed
Jens Axboe03ba3782009-09-09 09:08:54 +0200731 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200732 if (work->nr_pages <= 0)
Jens Axboe03ba3782009-09-09 09:08:54 +0200733 break;
734
735 /*
Jan Karaaa373cf2011-01-13 15:45:47 -0800736 * Background writeout and kupdate-style writeback may
737 * run forever. Stop them if there is other work to do
738 * so that e.g. sync can proceed. They'll be restarted
739 * after the other works are all done.
740 */
741 if ((work->for_background || work->for_kupdate) &&
742 !list_empty(&wb->bdi->work_list))
743 break;
744
745 /*
Wu Fengguangd3ddec72009-09-23 20:33:40 +0800746 * For background writeout, stop when we are below the
747 * background dirty threshold
Jens Axboe03ba3782009-09-09 09:08:54 +0200748 */
Wu Fengguangb00949a2010-11-18 14:38:33 -0600749 if (work->for_background && !over_bground_thresh(wb->bdi))
Jens Axboe03ba3782009-09-09 09:08:54 +0200750 break;
751
Jan Kara1bc36b62011-10-19 11:44:41 +0200752 /*
753 * Kupdate and background works are special and we want to
754 * include all inodes that need writing. Livelock avoidance is
755 * handled by these works yielding to any other work so we are
756 * safe.
757 */
Wu Fengguangba9aa832010-07-21 20:32:30 -0600758 if (work->for_kupdate) {
759 oldest_jif = jiffies -
760 msecs_to_jiffies(dirty_expire_interval * 10);
Jan Kara1bc36b62011-10-19 11:44:41 +0200761 } else if (work->for_background)
762 oldest_jif = jiffies;
Dave Chinner028c2dd2010-07-07 13:24:07 +1000763
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600764 trace_writeback_start(wb->bdi, work);
Wu Fengguange8dfc302011-04-21 12:06:32 -0600765 if (list_empty(&wb->b_io))
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600766 queue_io(wb, work);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200767 if (work->sb)
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600768 progress = writeback_sb_inodes(work->sb, wb, work);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200769 else
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600770 progress = __writeback_inodes_wb(wb, work);
771 trace_writeback_written(wb->bdi, work);
Dave Chinner028c2dd2010-07-07 13:24:07 +1000772
Wu Fengguange98be2d2010-08-29 11:22:30 -0600773 wb_update_bandwidth(wb, wb_start);
Jens Axboe03ba3782009-09-09 09:08:54 +0200774
775 /*
Jens Axboe71fd05a2009-09-23 19:32:26 +0200776 * Did we write something? Try for more
Wu Fengguange6fb6da2010-07-22 10:23:44 -0600777 *
778 * Dirty inodes are moved to b_io for writeback in batches.
779 * The completion of the current batch does not necessarily
780 * mean the overall work is done. So we keep looping as long
781 * as made some progress on cleaning pages or inodes.
Jens Axboe71fd05a2009-09-23 19:32:26 +0200782 */
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600783 if (progress)
Jens Axboe71fd05a2009-09-23 19:32:26 +0200784 continue;
785 /*
Wu Fengguange6fb6da2010-07-22 10:23:44 -0600786 * No more inodes for IO, bail
Jens Axboe71fd05a2009-09-23 19:32:26 +0200787 */
Wu Fengguangb7a24412010-07-21 22:19:51 -0600788 if (list_empty(&wb->b_more_io))
Jens Axboe03ba3782009-09-09 09:08:54 +0200789 break;
790 /*
Jens Axboe8010c3b2009-09-15 20:04:57 +0200791 * Nothing written. Wait for some inode to
792 * become available for writeback. Otherwise
793 * we'll just busyloop.
Jens Axboe03ba3782009-09-09 09:08:54 +0200794 */
Jens Axboe03ba3782009-09-09 09:08:54 +0200795 if (!list_empty(&wb->b_more_io)) {
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600796 trace_writeback_wait(wb->bdi, work);
Jens Axboe03ba3782009-09-09 09:08:54 +0200797 inode = wb_inode(wb->b_more_io.prev);
Dave Chinner250df6e2011-03-22 22:23:36 +1100798 spin_lock(&inode->i_lock);
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600799 inode_wait_for_writeback(inode, wb);
Dave Chinner250df6e2011-03-22 22:23:36 +1100800 spin_unlock(&inode->i_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200801 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200802 }
Wu Fengguange8dfc302011-04-21 12:06:32 -0600803 spin_unlock(&wb->list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200804
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600805 return nr_pages - work->nr_pages;
Jens Axboe03ba3782009-09-09 09:08:54 +0200806}
807
808/*
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200809 * Return the next wb_writeback_work struct that hasn't been processed yet.
Jens Axboe03ba3782009-09-09 09:08:54 +0200810 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200811static struct wb_writeback_work *
Minchan Kim08852b62010-08-03 12:51:16 +0200812get_next_work_item(struct backing_dev_info *bdi)
Jens Axboe03ba3782009-09-09 09:08:54 +0200813{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200814 struct wb_writeback_work *work = NULL;
Jens Axboe03ba3782009-09-09 09:08:54 +0200815
Artem Bityutskiy64677162010-07-25 14:29:22 +0300816 spin_lock_bh(&bdi->wb_lock);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200817 if (!list_empty(&bdi->work_list)) {
818 work = list_entry(bdi->work_list.next,
819 struct wb_writeback_work, list);
820 list_del_init(&work->list);
Jens Axboe03ba3782009-09-09 09:08:54 +0200821 }
Artem Bityutskiy64677162010-07-25 14:29:22 +0300822 spin_unlock_bh(&bdi->wb_lock);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200823 return work;
Jens Axboe03ba3782009-09-09 09:08:54 +0200824}
825
Linus Torvaldscdf01dd2010-10-30 08:55:52 -0700826/*
827 * Add in the number of potentially dirty inodes, because each inode
828 * write can dirty pagecache in the underlying blockdev.
829 */
830static unsigned long get_nr_dirty_pages(void)
831{
832 return global_page_state(NR_FILE_DIRTY) +
833 global_page_state(NR_UNSTABLE_NFS) +
834 get_nr_dirty_inodes();
835}
836
Jan Kara65850272011-01-13 15:45:44 -0800837static long wb_check_background_flush(struct bdi_writeback *wb)
838{
Wu Fengguangb00949a2010-11-18 14:38:33 -0600839 if (over_bground_thresh(wb->bdi)) {
Jan Kara65850272011-01-13 15:45:44 -0800840
841 struct wb_writeback_work work = {
842 .nr_pages = LONG_MAX,
843 .sync_mode = WB_SYNC_NONE,
844 .for_background = 1,
845 .range_cyclic = 1,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600846 .reason = WB_REASON_BACKGROUND,
Jan Kara65850272011-01-13 15:45:44 -0800847 };
848
849 return wb_writeback(wb, &work);
850 }
851
852 return 0;
853}
854
Jens Axboe03ba3782009-09-09 09:08:54 +0200855static long wb_check_old_data_flush(struct bdi_writeback *wb)
856{
857 unsigned long expired;
858 long nr_pages;
859
Jens Axboe69b62d02010-05-17 12:51:03 +0200860 /*
861 * When set to zero, disable periodic writeback
862 */
863 if (!dirty_writeback_interval)
864 return 0;
865
Jens Axboe03ba3782009-09-09 09:08:54 +0200866 expired = wb->last_old_flush +
867 msecs_to_jiffies(dirty_writeback_interval * 10);
868 if (time_before(jiffies, expired))
869 return 0;
870
871 wb->last_old_flush = jiffies;
Linus Torvaldscdf01dd2010-10-30 08:55:52 -0700872 nr_pages = get_nr_dirty_pages();
Jens Axboe03ba3782009-09-09 09:08:54 +0200873
Jens Axboec4a77a62009-09-16 15:18:25 +0200874 if (nr_pages) {
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200875 struct wb_writeback_work work = {
Jens Axboec4a77a62009-09-16 15:18:25 +0200876 .nr_pages = nr_pages,
877 .sync_mode = WB_SYNC_NONE,
878 .for_kupdate = 1,
879 .range_cyclic = 1,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600880 .reason = WB_REASON_PERIODIC,
Jens Axboec4a77a62009-09-16 15:18:25 +0200881 };
882
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200883 return wb_writeback(wb, &work);
Jens Axboec4a77a62009-09-16 15:18:25 +0200884 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200885
886 return 0;
887}
888
889/*
890 * Retrieve work items and do the writeback they describe
891 */
892long wb_do_writeback(struct bdi_writeback *wb, int force_wait)
893{
894 struct backing_dev_info *bdi = wb->bdi;
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200895 struct wb_writeback_work *work;
Jens Axboec4a77a62009-09-16 15:18:25 +0200896 long wrote = 0;
Jens Axboe03ba3782009-09-09 09:08:54 +0200897
Jan Kara81d73a32010-08-11 14:17:44 -0700898 set_bit(BDI_writeback_running, &wb->bdi->state);
Minchan Kim08852b62010-08-03 12:51:16 +0200899 while ((work = get_next_work_item(bdi)) != NULL) {
Jens Axboe03ba3782009-09-09 09:08:54 +0200900 /*
901 * Override sync mode, in case we must wait for completion
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200902 * because this thread is exiting now.
Jens Axboe03ba3782009-09-09 09:08:54 +0200903 */
904 if (force_wait)
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200905 work->sync_mode = WB_SYNC_ALL;
906
Dave Chinner455b2862010-07-07 13:24:06 +1000907 trace_writeback_exec(bdi, work);
908
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200909 wrote += wb_writeback(wb, work);
Jens Axboe03ba3782009-09-09 09:08:54 +0200910
911 /*
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200912 * Notify the caller of completion if this is a synchronous
913 * work item, otherwise just free it.
Jens Axboe03ba3782009-09-09 09:08:54 +0200914 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200915 if (work->done)
916 complete(work->done);
917 else
918 kfree(work);
Jens Axboe03ba3782009-09-09 09:08:54 +0200919 }
920
921 /*
922 * Check for periodic writeback, kupdated() style
923 */
924 wrote += wb_check_old_data_flush(wb);
Jan Kara65850272011-01-13 15:45:44 -0800925 wrote += wb_check_background_flush(wb);
Jan Kara81d73a32010-08-11 14:17:44 -0700926 clear_bit(BDI_writeback_running, &wb->bdi->state);
Jens Axboe03ba3782009-09-09 09:08:54 +0200927
928 return wrote;
929}
930
931/*
932 * Handle writeback of dirty data for the device backed by this bdi. Also
933 * wakes up periodically and does kupdated style flushing.
934 */
Christoph Hellwig08243902010-06-19 23:08:22 +0200935int bdi_writeback_thread(void *data)
Jens Axboe03ba3782009-09-09 09:08:54 +0200936{
Christoph Hellwig08243902010-06-19 23:08:22 +0200937 struct bdi_writeback *wb = data;
938 struct backing_dev_info *bdi = wb->bdi;
Jens Axboe03ba3782009-09-09 09:08:54 +0200939 long pages_written;
940
Peter Zijlstra766f9162010-10-26 14:22:45 -0700941 current->flags |= PF_SWAPWRITE;
Christoph Hellwig08243902010-06-19 23:08:22 +0200942 set_freezable();
Artem Bityutskiyecd58402010-07-25 14:29:18 +0300943 wb->last_active = jiffies;
Christoph Hellwig08243902010-06-19 23:08:22 +0200944
945 /*
946 * Our parent may run at a different priority, just set us to normal
947 */
948 set_user_nice(current, 0);
949
Dave Chinner455b2862010-07-07 13:24:06 +1000950 trace_writeback_thread_start(bdi);
951
Tejun Heo8a32c442011-11-21 12:32:23 -0800952 while (!kthread_freezable_should_stop(NULL)) {
Artem Bityutskiy64677162010-07-25 14:29:22 +0300953 /*
954 * Remove own delayed wake-up timer, since we are already awake
955 * and we'll take care of the preriodic write-back.
956 */
957 del_timer(&wb->wakeup_timer);
958
Jens Axboe03ba3782009-09-09 09:08:54 +0200959 pages_written = wb_do_writeback(wb, 0);
960
Dave Chinner455b2862010-07-07 13:24:06 +1000961 trace_writeback_pages_written(pages_written);
Jens Axboe03ba3782009-09-09 09:08:54 +0200962
Jens Axboe03ba3782009-09-09 09:08:54 +0200963 if (pages_written)
Artem Bityutskiyecd58402010-07-25 14:29:18 +0300964 wb->last_active = jiffies;
Jens Axboe03ba3782009-09-09 09:08:54 +0200965
Artem Bityutskiy297252c2010-07-25 14:29:15 +0300966 set_current_state(TASK_INTERRUPTIBLE);
J. Bruce Fieldsb76b4012010-08-28 08:52:10 +0200967 if (!list_empty(&bdi->work_list) || kthread_should_stop()) {
Artem Bityutskiy297252c2010-07-25 14:29:15 +0300968 __set_current_state(TASK_RUNNING);
969 continue;
Jens Axboe03ba3782009-09-09 09:08:54 +0200970 }
971
Artem Bityutskiy253c34e2010-07-25 14:29:21 +0300972 if (wb_has_dirty_io(wb) && dirty_writeback_interval)
Artem Bityutskiyfff5b852010-07-25 14:29:20 +0300973 schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10));
Artem Bityutskiy253c34e2010-07-25 14:29:21 +0300974 else {
975 /*
976 * We have nothing to do, so can go sleep without any
977 * timeout and save power. When a work is queued or
978 * something is made dirty - we will be woken up.
979 */
Artem Bityutskiy297252c2010-07-25 14:29:15 +0300980 schedule();
Jens Axboef9eadbb2010-05-18 14:31:45 +0200981 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200982 }
983
Artem Bityutskiyfff5b852010-07-25 14:29:20 +0300984 /* Flush any work that raced with us exiting */
Christoph Hellwig08243902010-06-19 23:08:22 +0200985 if (!list_empty(&bdi->work_list))
986 wb_do_writeback(wb, 1);
Dave Chinner455b2862010-07-07 13:24:06 +1000987
988 trace_writeback_thread_stop(bdi);
Jens Axboe03ba3782009-09-09 09:08:54 +0200989 return 0;
990}
991
Christoph Hellwig08243902010-06-19 23:08:22 +0200992
Jens Axboe03ba3782009-09-09 09:08:54 +0200993/*
Jens Axboe03ba3782009-09-09 09:08:54 +0200994 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
995 * the whole world.
996 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600997void wakeup_flusher_threads(long nr_pages, enum wb_reason reason)
Jens Axboe03ba3782009-09-09 09:08:54 +0200998{
Christoph Hellwigb8c2f342010-06-08 18:15:07 +0200999 struct backing_dev_info *bdi;
Christoph Hellwigb8c2f342010-06-08 18:15:07 +02001000
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001001 if (!nr_pages) {
1002 nr_pages = global_page_state(NR_FILE_DIRTY) +
Jens Axboe03ba3782009-09-09 09:08:54 +02001003 global_page_state(NR_UNSTABLE_NFS);
Christoph Hellwigb8c2f342010-06-08 18:15:07 +02001004 }
1005
1006 rcu_read_lock();
1007 list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) {
1008 if (!bdi_has_dirty_io(bdi))
1009 continue;
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001010 __bdi_start_writeback(bdi, nr_pages, false, reason);
Christoph Hellwigb8c2f342010-06-08 18:15:07 +02001011 }
1012 rcu_read_unlock();
Jens Axboe03ba3782009-09-09 09:08:54 +02001013}
1014
1015static noinline void block_dump___mark_inode_dirty(struct inode *inode)
1016{
1017 if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) {
1018 struct dentry *dentry;
1019 const char *name = "?";
1020
1021 dentry = d_find_alias(inode);
1022 if (dentry) {
1023 spin_lock(&dentry->d_lock);
1024 name = (const char *) dentry->d_name.name;
1025 }
1026 printk(KERN_DEBUG
1027 "%s(%d): dirtied inode %lu (%s) on %s\n",
1028 current->comm, task_pid_nr(current), inode->i_ino,
1029 name, inode->i_sb->s_id);
1030 if (dentry) {
1031 spin_unlock(&dentry->d_lock);
1032 dput(dentry);
1033 }
1034 }
1035}
1036
1037/**
1038 * __mark_inode_dirty - internal function
1039 * @inode: inode to mark
1040 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
1041 * Mark an inode as dirty. Callers should use mark_inode_dirty or
1042 * mark_inode_dirty_sync.
1043 *
1044 * Put the inode on the super block's dirty list.
1045 *
1046 * CAREFUL! We mark it dirty unconditionally, but move it onto the
1047 * dirty list only if it is hashed or if it refers to a blockdev.
1048 * If it was not hashed, it will never be added to the dirty list
1049 * even if it is later hashed, as it will have been marked dirty already.
1050 *
1051 * In short, make sure you hash any inodes _before_ you start marking
1052 * them dirty.
1053 *
Jens Axboe03ba3782009-09-09 09:08:54 +02001054 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
1055 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
1056 * the kernel-internal blockdev inode represents the dirtying time of the
1057 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
1058 * page->mapping->host, so the page-dirtying time is recorded in the internal
1059 * blockdev inode.
1060 */
1061void __mark_inode_dirty(struct inode *inode, int flags)
1062{
1063 struct super_block *sb = inode->i_sb;
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001064 struct backing_dev_info *bdi = NULL;
Jens Axboe03ba3782009-09-09 09:08:54 +02001065
1066 /*
1067 * Don't do this for I_DIRTY_PAGES - that doesn't actually
1068 * dirty the inode itself
1069 */
1070 if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
1071 if (sb->s_op->dirty_inode)
Christoph Hellwigaa385722011-05-27 06:53:02 -04001072 sb->s_op->dirty_inode(inode, flags);
Jens Axboe03ba3782009-09-09 09:08:54 +02001073 }
1074
1075 /*
1076 * make sure that changes are seen by all cpus before we test i_state
1077 * -- mikulas
1078 */
1079 smp_mb();
1080
1081 /* avoid the locking if we can */
1082 if ((inode->i_state & flags) == flags)
1083 return;
1084
1085 if (unlikely(block_dump))
1086 block_dump___mark_inode_dirty(inode);
1087
Dave Chinner250df6e2011-03-22 22:23:36 +11001088 spin_lock(&inode->i_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001089 if ((inode->i_state & flags) != flags) {
1090 const int was_dirty = inode->i_state & I_DIRTY;
1091
1092 inode->i_state |= flags;
1093
1094 /*
1095 * If the inode is being synced, just update its dirty state.
1096 * The unlocker will place the inode on the appropriate
1097 * superblock list, based upon its state.
1098 */
1099 if (inode->i_state & I_SYNC)
Dave Chinner250df6e2011-03-22 22:23:36 +11001100 goto out_unlock_inode;
Jens Axboe03ba3782009-09-09 09:08:54 +02001101
1102 /*
1103 * Only add valid (hashed) inodes to the superblock's
1104 * dirty list. Add blockdev inodes as well.
1105 */
1106 if (!S_ISBLK(inode->i_mode)) {
Al Viro1d3382c2010-10-23 15:19:20 -04001107 if (inode_unhashed(inode))
Dave Chinner250df6e2011-03-22 22:23:36 +11001108 goto out_unlock_inode;
Jens Axboe03ba3782009-09-09 09:08:54 +02001109 }
Al Viroa4ffdde2010-06-02 17:38:30 -04001110 if (inode->i_state & I_FREEING)
Dave Chinner250df6e2011-03-22 22:23:36 +11001111 goto out_unlock_inode;
Jens Axboe03ba3782009-09-09 09:08:54 +02001112
1113 /*
1114 * If the inode was already on b_dirty/b_io/b_more_io, don't
1115 * reposition it (that would break b_dirty time-ordering).
1116 */
1117 if (!was_dirty) {
Dave Chinnera66979a2011-03-22 22:23:41 +11001118 bool wakeup_bdi = false;
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001119 bdi = inode_to_bdi(inode);
Jens Axboe500b0672009-09-09 09:10:25 +02001120
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001121 if (bdi_cap_writeback_dirty(bdi)) {
1122 WARN(!test_bit(BDI_registered, &bdi->state),
1123 "bdi-%s not registered\n", bdi->name);
1124
1125 /*
1126 * If this is the first dirty inode for this
1127 * bdi, we have to wake-up the corresponding
1128 * bdi thread to make sure background
1129 * write-back happens later.
1130 */
1131 if (!wb_has_dirty_io(&bdi->wb))
1132 wakeup_bdi = true;
Jens Axboe500b0672009-09-09 09:10:25 +02001133 }
Jens Axboe03ba3782009-09-09 09:08:54 +02001134
Dave Chinnera66979a2011-03-22 22:23:41 +11001135 spin_unlock(&inode->i_lock);
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001136 spin_lock(&bdi->wb.list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001137 inode->dirtied_when = jiffies;
Nick Piggin7ccf19a2010-10-21 11:49:30 +11001138 list_move(&inode->i_wb_list, &bdi->wb.b_dirty);
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001139 spin_unlock(&bdi->wb.list_lock);
Dave Chinnera66979a2011-03-22 22:23:41 +11001140
1141 if (wakeup_bdi)
1142 bdi_wakeup_thread_delayed(bdi);
1143 return;
Jens Axboe03ba3782009-09-09 09:08:54 +02001144 }
1145 }
Dave Chinner250df6e2011-03-22 22:23:36 +11001146out_unlock_inode:
1147 spin_unlock(&inode->i_lock);
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001148
Jens Axboe03ba3782009-09-09 09:08:54 +02001149}
1150EXPORT_SYMBOL(__mark_inode_dirty);
1151
Jens Axboeb6e51312009-09-16 15:13:54 +02001152static void wait_sb_inodes(struct super_block *sb)
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001153{
Jens Axboe03ba3782009-09-09 09:08:54 +02001154 struct inode *inode, *old_inode = NULL;
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001155
Jens Axboe03ba3782009-09-09 09:08:54 +02001156 /*
1157 * We need to be protected against the filesystem going from
1158 * r/o to r/w or vice versa.
1159 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001160 WARN_ON(!rwsem_is_locked(&sb->s_umount));
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001161
Dave Chinner55fa6092011-03-22 22:23:40 +11001162 spin_lock(&inode_sb_list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001163
1164 /*
1165 * Data integrity sync. Must wait for all pages under writeback,
1166 * because there may have been pages dirtied before our sync
1167 * call, but which had writeout started before we write it out.
1168 * In which case, the inode may not be on the dirty list, but
1169 * we still have to wait for that writeout.
1170 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001171 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
Dave Chinner250df6e2011-03-22 22:23:36 +11001172 struct address_space *mapping = inode->i_mapping;
Jens Axboe03ba3782009-09-09 09:08:54 +02001173
Dave Chinner250df6e2011-03-22 22:23:36 +11001174 spin_lock(&inode->i_lock);
1175 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
1176 (mapping->nrpages == 0)) {
1177 spin_unlock(&inode->i_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001178 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +11001179 }
Jens Axboe03ba3782009-09-09 09:08:54 +02001180 __iget(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +11001181 spin_unlock(&inode->i_lock);
Dave Chinner55fa6092011-03-22 22:23:40 +11001182 spin_unlock(&inode_sb_list_lock);
1183
Jens Axboe03ba3782009-09-09 09:08:54 +02001184 /*
Dave Chinner55fa6092011-03-22 22:23:40 +11001185 * We hold a reference to 'inode' so it couldn't have been
1186 * removed from s_inodes list while we dropped the
1187 * inode_sb_list_lock. We cannot iput the inode now as we can
1188 * be holding the last reference and we cannot iput it under
1189 * inode_sb_list_lock. So we keep the reference and iput it
1190 * later.
Jens Axboe03ba3782009-09-09 09:08:54 +02001191 */
1192 iput(old_inode);
1193 old_inode = inode;
1194
1195 filemap_fdatawait(mapping);
1196
1197 cond_resched();
Nick Piggin38f21972009-01-06 14:40:25 -08001198
Dave Chinner55fa6092011-03-22 22:23:40 +11001199 spin_lock(&inode_sb_list_lock);
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001200 }
Dave Chinner55fa6092011-03-22 22:23:40 +11001201 spin_unlock(&inode_sb_list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001202 iput(old_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203}
1204
Jens Axboed8a85592009-09-02 12:34:32 +02001205/**
Chris Mason3259f8b2010-10-29 11:16:17 -04001206 * writeback_inodes_sb_nr - writeback dirty inodes from given super_block
Jens Axboed8a85592009-09-02 12:34:32 +02001207 * @sb: the superblock
Chris Mason3259f8b2010-10-29 11:16:17 -04001208 * @nr: the number of pages to write
Marcos Paulo de Souza786228a2011-11-23 20:56:45 +08001209 * @reason: reason why some writeback work initiated
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210 *
Jens Axboed8a85592009-09-02 12:34:32 +02001211 * Start writeback on some inodes on this super_block. No guarantees are made
1212 * on how many (if any) will be written, and this function does not wait
Chris Mason3259f8b2010-10-29 11:16:17 -04001213 * for IO completion of submitted IO.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001215void writeback_inodes_sb_nr(struct super_block *sb,
1216 unsigned long nr,
1217 enum wb_reason reason)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001219 DECLARE_COMPLETION_ONSTACK(done);
1220 struct wb_writeback_work work = {
Wu Fengguang6e6938b2010-06-06 10:38:15 -06001221 .sb = sb,
1222 .sync_mode = WB_SYNC_NONE,
1223 .tagged_writepages = 1,
1224 .done = &done,
1225 .nr_pages = nr,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001226 .reason = reason,
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001227 };
Jens Axboe0e3c9a22010-06-01 11:08:43 +02001228
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001229 WARN_ON(!rwsem_is_locked(&sb->s_umount));
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001230 bdi_queue_work(sb->s_bdi, &work);
1231 wait_for_completion(&done);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232}
Chris Mason3259f8b2010-10-29 11:16:17 -04001233EXPORT_SYMBOL(writeback_inodes_sb_nr);
1234
1235/**
1236 * writeback_inodes_sb - writeback dirty inodes from given super_block
1237 * @sb: the superblock
Marcos Paulo de Souza786228a2011-11-23 20:56:45 +08001238 * @reason: reason why some writeback work was initiated
Chris Mason3259f8b2010-10-29 11:16:17 -04001239 *
1240 * Start writeback on some inodes on this super_block. No guarantees are made
1241 * on how many (if any) will be written, and this function does not wait
1242 * for IO completion of submitted IO.
1243 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001244void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason)
Chris Mason3259f8b2010-10-29 11:16:17 -04001245{
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001246 return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason);
Chris Mason3259f8b2010-10-29 11:16:17 -04001247}
Jens Axboed8a85592009-09-02 12:34:32 +02001248EXPORT_SYMBOL(writeback_inodes_sb);
1249
1250/**
Eric Sandeen17bd55d2009-12-23 07:57:07 -05001251 * writeback_inodes_sb_if_idle - start writeback if none underway
1252 * @sb: the superblock
Marcos Paulo de Souza786228a2011-11-23 20:56:45 +08001253 * @reason: reason why some writeback work was initiated
Eric Sandeen17bd55d2009-12-23 07:57:07 -05001254 *
1255 * Invoke writeback_inodes_sb if no writeback is currently underway.
1256 * Returns 1 if writeback was started, 0 if not.
1257 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001258int writeback_inodes_sb_if_idle(struct super_block *sb, enum wb_reason reason)
Eric Sandeen17bd55d2009-12-23 07:57:07 -05001259{
1260 if (!writeback_in_progress(sb->s_bdi)) {
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001261 down_read(&sb->s_umount);
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001262 writeback_inodes_sb(sb, reason);
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001263 up_read(&sb->s_umount);
Eric Sandeen17bd55d2009-12-23 07:57:07 -05001264 return 1;
1265 } else
1266 return 0;
1267}
1268EXPORT_SYMBOL(writeback_inodes_sb_if_idle);
1269
1270/**
Fengguang Wuc097b2c2012-03-05 15:08:06 -08001271 * writeback_inodes_sb_nr_if_idle - start writeback if none underway
Chris Mason3259f8b2010-10-29 11:16:17 -04001272 * @sb: the superblock
1273 * @nr: the number of pages to write
Marcos Paulo de Souza786228a2011-11-23 20:56:45 +08001274 * @reason: reason why some writeback work was initiated
Chris Mason3259f8b2010-10-29 11:16:17 -04001275 *
1276 * Invoke writeback_inodes_sb if no writeback is currently underway.
1277 * Returns 1 if writeback was started, 0 if not.
1278 */
1279int writeback_inodes_sb_nr_if_idle(struct super_block *sb,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001280 unsigned long nr,
1281 enum wb_reason reason)
Chris Mason3259f8b2010-10-29 11:16:17 -04001282{
1283 if (!writeback_in_progress(sb->s_bdi)) {
1284 down_read(&sb->s_umount);
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001285 writeback_inodes_sb_nr(sb, nr, reason);
Chris Mason3259f8b2010-10-29 11:16:17 -04001286 up_read(&sb->s_umount);
1287 return 1;
1288 } else
1289 return 0;
1290}
1291EXPORT_SYMBOL(writeback_inodes_sb_nr_if_idle);
1292
1293/**
Jens Axboed8a85592009-09-02 12:34:32 +02001294 * sync_inodes_sb - sync sb inode pages
1295 * @sb: the superblock
1296 *
1297 * This function writes and waits on any dirty inode belonging to this
Stefan Hajnoczicb9ef8d2011-01-13 15:47:26 -08001298 * super_block.
Jens Axboed8a85592009-09-02 12:34:32 +02001299 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001300void sync_inodes_sb(struct super_block *sb)
Jens Axboed8a85592009-09-02 12:34:32 +02001301{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001302 DECLARE_COMPLETION_ONSTACK(done);
1303 struct wb_writeback_work work = {
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001304 .sb = sb,
1305 .sync_mode = WB_SYNC_ALL,
1306 .nr_pages = LONG_MAX,
1307 .range_cyclic = 0,
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001308 .done = &done,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001309 .reason = WB_REASON_SYNC,
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001310 };
1311
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001312 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1313
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001314 bdi_queue_work(sb->s_bdi, &work);
1315 wait_for_completion(&done);
1316
Jens Axboeb6e51312009-09-16 15:13:54 +02001317 wait_sb_inodes(sb);
Jens Axboed8a85592009-09-02 12:34:32 +02001318}
1319EXPORT_SYMBOL(sync_inodes_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321/**
Andrea Arcangeli7f04c262005-10-30 15:03:05 -08001322 * write_inode_now - write an inode to disk
1323 * @inode: inode to write to disk
1324 * @sync: whether the write should be synchronous or not
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325 *
Andrea Arcangeli7f04c262005-10-30 15:03:05 -08001326 * This function commits an inode to disk immediately if it is dirty. This is
1327 * primarily needed by knfsd.
1328 *
1329 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331int write_inode_now(struct inode *inode, int sync)
1332{
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001333 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334 int ret;
1335 struct writeback_control wbc = {
1336 .nr_to_write = LONG_MAX,
Mike Galbraith18914b12008-02-08 04:20:23 -08001337 .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -07001338 .range_start = 0,
1339 .range_end = LLONG_MAX,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340 };
1341
1342 if (!mapping_cap_writeback_dirty(inode->i_mapping))
Andrew Morton49364ce2005-11-07 00:59:15 -08001343 wbc.nr_to_write = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344
1345 might_sleep();
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001346 spin_lock(&wb->list_lock);
Dave Chinner0f1b1fd2011-03-22 22:23:43 +11001347 spin_lock(&inode->i_lock);
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001348 ret = writeback_single_inode(inode, wb, &wbc);
Dave Chinner0f1b1fd2011-03-22 22:23:43 +11001349 spin_unlock(&inode->i_lock);
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001350 spin_unlock(&wb->list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351 return ret;
1352}
1353EXPORT_SYMBOL(write_inode_now);
1354
1355/**
1356 * sync_inode - write an inode and its pages to disk.
1357 * @inode: the inode to sync
1358 * @wbc: controls the writeback mode
1359 *
1360 * sync_inode() will write an inode and its pages to disk. It will also
1361 * correctly update the inode on its superblock's dirty inode lists and will
1362 * update inode->i_state.
1363 *
1364 * The caller must have a ref on the inode.
1365 */
1366int sync_inode(struct inode *inode, struct writeback_control *wbc)
1367{
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001368 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369 int ret;
1370
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001371 spin_lock(&wb->list_lock);
Dave Chinner0f1b1fd2011-03-22 22:23:43 +11001372 spin_lock(&inode->i_lock);
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001373 ret = writeback_single_inode(inode, wb, wbc);
Dave Chinner0f1b1fd2011-03-22 22:23:43 +11001374 spin_unlock(&inode->i_lock);
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001375 spin_unlock(&wb->list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376 return ret;
1377}
1378EXPORT_SYMBOL(sync_inode);
Christoph Hellwigc3765012010-10-06 10:48:20 +02001379
1380/**
Andrew Mortonc691b9d2011-01-13 15:45:48 -08001381 * sync_inode_metadata - write an inode to disk
Christoph Hellwigc3765012010-10-06 10:48:20 +02001382 * @inode: the inode to sync
1383 * @wait: wait for I/O to complete.
1384 *
Andrew Mortonc691b9d2011-01-13 15:45:48 -08001385 * Write an inode to disk and adjust its dirty state after completion.
Christoph Hellwigc3765012010-10-06 10:48:20 +02001386 *
1387 * Note: only writes the actual inode, no associated data or other metadata.
1388 */
1389int sync_inode_metadata(struct inode *inode, int wait)
1390{
1391 struct writeback_control wbc = {
1392 .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE,
1393 .nr_to_write = 0, /* metadata-only */
1394 };
1395
1396 return sync_inode(inode, &wbc);
1397}
1398EXPORT_SYMBOL(sync_inode_metadata);