blob: 5d3de002cb8ee5b96b277b28522c258e781a7a5b [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{
Jan Kara365b94ae2012-05-03 14:47:55 +0200234 inode->i_state &= ~I_SYNC;
235 /* Waiters must see I_SYNC cleared before being woken up */
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700236 smp_mb();
237 wake_up_bit(&inode->i_state, __I_SYNC);
238}
239
Jeff Laytond2caa3c52009-04-02 16:56:37 -0700240static bool inode_dirtied_after(struct inode *inode, unsigned long t)
241{
242 bool ret = time_after(inode->dirtied_when, t);
243#ifndef CONFIG_64BIT
244 /*
245 * For inodes being constantly redirtied, dirtied_when can get stuck.
246 * It _appears_ to be in the future, but is actually in distant past.
247 * This test is necessary to prevent such wrapped-around relative times
Jens Axboe5b0830c2009-09-23 19:37:09 +0200248 * from permanently stopping the whole bdi writeback.
Jeff Laytond2caa3c52009-04-02 16:56:37 -0700249 */
250 ret = ret && time_before_eq(inode->dirtied_when, jiffies);
251#endif
252 return ret;
253}
254
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700255/*
Jan Kara697e6fe2012-03-09 07:26:22 -0800256 * Move expired (dirtied after work->older_than_this) dirty inodes from
257 * @delaying_queue to @dispatch_queue.
Fengguang Wu2c136572007-10-16 23:30:39 -0700258 */
Wu Fengguange84d0a42011-04-23 12:27:27 -0600259static int move_expired_inodes(struct list_head *delaying_queue,
Fengguang Wu2c136572007-10-16 23:30:39 -0700260 struct list_head *dispatch_queue,
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600261 struct wb_writeback_work *work)
Fengguang Wu2c136572007-10-16 23:30:39 -0700262{
Shaohua Li5c034492009-09-24 14:42:33 +0200263 LIST_HEAD(tmp);
264 struct list_head *pos, *node;
Jens Axboecf137302009-09-24 15:12:57 +0200265 struct super_block *sb = NULL;
Shaohua Li5c034492009-09-24 14:42:33 +0200266 struct inode *inode;
Jens Axboecf137302009-09-24 15:12:57 +0200267 int do_sb_sort = 0;
Wu Fengguange84d0a42011-04-23 12:27:27 -0600268 int moved = 0;
Shaohua Li5c034492009-09-24 14:42:33 +0200269
Fengguang Wu2c136572007-10-16 23:30:39 -0700270 while (!list_empty(delaying_queue)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100271 inode = wb_inode(delaying_queue->prev);
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600272 if (work->older_than_this &&
273 inode_dirtied_after(inode, *work->older_than_this))
Fengguang Wu2c136572007-10-16 23:30:39 -0700274 break;
Jens Axboecf137302009-09-24 15:12:57 +0200275 if (sb && sb != inode->i_sb)
276 do_sb_sort = 1;
277 sb = inode->i_sb;
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100278 list_move(&inode->i_wb_list, &tmp);
Wu Fengguange84d0a42011-04-23 12:27:27 -0600279 moved++;
Shaohua Li5c034492009-09-24 14:42:33 +0200280 }
281
Jens Axboecf137302009-09-24 15:12:57 +0200282 /* just one sb in list, splice to dispatch_queue and we're done */
283 if (!do_sb_sort) {
284 list_splice(&tmp, dispatch_queue);
Wu Fengguange84d0a42011-04-23 12:27:27 -0600285 goto out;
Jens Axboecf137302009-09-24 15:12:57 +0200286 }
287
Shaohua Li5c034492009-09-24 14:42:33 +0200288 /* Move inodes from one superblock together */
289 while (!list_empty(&tmp)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100290 sb = wb_inode(tmp.prev)->i_sb;
Shaohua Li5c034492009-09-24 14:42:33 +0200291 list_for_each_prev_safe(pos, node, &tmp) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100292 inode = wb_inode(pos);
Shaohua Li5c034492009-09-24 14:42:33 +0200293 if (inode->i_sb == sb)
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100294 list_move(&inode->i_wb_list, dispatch_queue);
Shaohua Li5c034492009-09-24 14:42:33 +0200295 }
Fengguang Wu2c136572007-10-16 23:30:39 -0700296 }
Wu Fengguange84d0a42011-04-23 12:27:27 -0600297out:
298 return moved;
Fengguang Wu2c136572007-10-16 23:30:39 -0700299}
300
301/*
302 * Queue all expired dirty inodes for io, eldest first.
Wu Fengguang4ea879b2010-08-11 14:17:42 -0700303 * Before
304 * newly dirtied b_dirty b_io b_more_io
305 * =============> gf edc BA
306 * After
307 * newly dirtied b_dirty b_io b_more_io
308 * =============> g fBAedc
309 * |
310 * +--> dequeue for IO
Fengguang Wu2c136572007-10-16 23:30:39 -0700311 */
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600312static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work)
Fengguang Wu2c136572007-10-16 23:30:39 -0700313{
Wu Fengguange84d0a42011-04-23 12:27:27 -0600314 int moved;
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600315 assert_spin_locked(&wb->list_lock);
Wu Fengguang4ea879b2010-08-11 14:17:42 -0700316 list_splice_init(&wb->b_more_io, &wb->b_io);
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600317 moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, work);
318 trace_writeback_queue_io(wb, work, moved);
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200319}
320
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100321static int write_inode(struct inode *inode, struct writeback_control *wbc)
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200322{
Jens Axboe03ba3782009-09-09 09:08:54 +0200323 if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode))
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100324 return inode->i_sb->s_op->write_inode(inode, wbc);
Jens Axboe03ba3782009-09-09 09:08:54 +0200325 return 0;
Fengguang Wu2c136572007-10-16 23:30:39 -0700326}
327
328/*
Christoph Hellwig01c03192009-06-08 13:35:40 +0200329 * Wait for writeback on an inode to complete.
330 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600331static void inode_wait_for_writeback(struct inode *inode,
332 struct bdi_writeback *wb)
Christoph Hellwig01c03192009-06-08 13:35:40 +0200333{
334 DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC);
335 wait_queue_head_t *wqh;
336
337 wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
Dave Chinner250df6e2011-03-22 22:23:36 +1100338 while (inode->i_state & I_SYNC) {
339 spin_unlock(&inode->i_lock);
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600340 spin_unlock(&wb->list_lock);
Christoph Hellwig01c03192009-06-08 13:35:40 +0200341 __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE);
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600342 spin_lock(&wb->list_lock);
Dave Chinner250df6e2011-03-22 22:23:36 +1100343 spin_lock(&inode->i_lock);
Richard Kennedy58a9d3d82010-05-24 14:32:38 -0700344 }
Christoph Hellwig01c03192009-06-08 13:35:40 +0200345}
346
347/*
Jan Karaccb26b52012-05-03 14:47:58 +0200348 * Find proper writeback list for the inode depending on its current state and
349 * possibly also change of its state while we were doing writeback. Here we
350 * handle things such as livelock prevention or fairness of writeback among
351 * inodes. This function can be called only by flusher thread - noone else
352 * processes all inodes in writeback lists and requeueing inodes behind flusher
353 * thread's back can have unexpected consequences.
354 */
355static void requeue_inode(struct inode *inode, struct bdi_writeback *wb,
356 struct writeback_control *wbc)
357{
358 if (inode->i_state & I_FREEING)
359 return;
360
361 /*
362 * Sync livelock prevention. Each inode is tagged and synced in one
363 * shot. If still dirty, it will be redirty_tail()'ed below. Update
364 * the dirty time to prevent enqueue and sync it again.
365 */
366 if ((inode->i_state & I_DIRTY) &&
367 (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages))
368 inode->dirtied_when = jiffies;
369
370 if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) {
371 /*
372 * We didn't write back all the pages. nfs_writepages()
373 * sometimes bales out without doing anything.
374 */
375 if (wbc->nr_to_write <= 0) {
376 /* Slice used up. Queue for next turn. */
377 requeue_io(inode, wb);
378 } else {
379 /*
380 * Writeback blocked by something other than
381 * congestion. Delay the inode for some time to
382 * avoid spinning on the CPU (100% iowait)
383 * retrying writeback of the dirty page/inode
384 * that cannot be performed immediately.
385 */
386 redirty_tail(inode, wb);
387 }
388 } else if (inode->i_state & I_DIRTY) {
389 /*
390 * Filesystems can dirty the inode during writeback operations,
391 * such as delayed allocation during submission or metadata
392 * updates after data IO completion.
393 */
394 redirty_tail(inode, wb);
395 } else {
396 /* The inode is clean. Remove from writeback lists. */
397 list_del_init(&inode->i_wb_list);
398 }
399}
400
401/*
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600402 * Write out an inode's dirty pages. Called under wb->list_lock and
Dave Chinner0f1b1fd2011-03-22 22:23:43 +1100403 * inode->i_lock. Either the caller has an active reference on the inode or
404 * the inode has I_WILL_FREE set.
Christoph Hellwig01c03192009-06-08 13:35:40 +0200405 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406 * If `wait' is set, wait on the writeout.
407 *
408 * The whole writeout design is quite complex and fragile. We want to avoid
409 * starvation of particular inodes when others are being redirtied, prevent
410 * livelocks, etc.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 */
412static int
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600413writeback_single_inode(struct inode *inode, struct bdi_writeback *wb,
414 struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416 struct address_space *mapping = inode->i_mapping;
Wu Fengguang251d6a42010-12-01 17:33:37 -0600417 long nr_to_write = wbc->nr_to_write;
Christoph Hellwig01c03192009-06-08 13:35:40 +0200418 unsigned dirty;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419 int ret;
420
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600421 assert_spin_locked(&wb->list_lock);
Dave Chinner0f1b1fd2011-03-22 22:23:43 +1100422 assert_spin_locked(&inode->i_lock);
423
Christoph Hellwig01c03192009-06-08 13:35:40 +0200424 if (!atomic_read(&inode->i_count))
425 WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING)));
426 else
427 WARN_ON(inode->i_state & I_WILL_FREE);
428
429 if (inode->i_state & I_SYNC) {
Jan Karacc1676d2012-05-03 14:47:56 +0200430 if (wbc->sync_mode != WB_SYNC_ALL)
Christoph Hellwig01c03192009-06-08 13:35:40 +0200431 return 0;
Christoph Hellwig01c03192009-06-08 13:35:40 +0200432 /*
433 * It's a data-integrity sync. We must wait.
434 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600435 inode_wait_for_writeback(inode, wb);
Christoph Hellwig01c03192009-06-08 13:35:40 +0200436 }
437
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700438 BUG_ON(inode->i_state & I_SYNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439
Dmitry Monakhov5547e8a2010-05-07 13:35:44 +0400440 /* Set I_SYNC, reset I_DIRTY_PAGES */
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700441 inode->i_state |= I_SYNC;
Dave Chinner250df6e2011-03-22 22:23:36 +1100442 spin_unlock(&inode->i_lock);
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600443 spin_unlock(&wb->list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444
445 ret = do_writepages(mapping, wbc);
446
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100447 /*
448 * Make sure to wait on the data before writing out the metadata.
449 * This is important for filesystems that modify metadata on data
450 * I/O completion.
451 */
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100452 if (wbc->sync_mode == WB_SYNC_ALL) {
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100453 int err = filemap_fdatawait(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454 if (ret == 0)
455 ret = err;
456 }
457
Dmitry Monakhov5547e8a2010-05-07 13:35:44 +0400458 /*
459 * Some filesystems may redirty the inode during the writeback
460 * due to delalloc, clear dirty metadata flags right before
461 * write_inode()
462 */
Dave Chinner250df6e2011-03-22 22:23:36 +1100463 spin_lock(&inode->i_lock);
Jan Kara6290be12012-05-03 14:47:57 +0200464 /* Clear I_DIRTY_PAGES if we've written out all dirty pages */
465 if (!mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
466 inode->i_state &= ~I_DIRTY_PAGES;
Dmitry Monakhov5547e8a2010-05-07 13:35:44 +0400467 dirty = inode->i_state & I_DIRTY;
468 inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC);
Dave Chinner250df6e2011-03-22 22:23:36 +1100469 spin_unlock(&inode->i_lock);
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100470 /* Don't write the inode if only I_DIRTY_PAGES was set */
471 if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100472 int err = write_inode(inode, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473 if (ret == 0)
474 ret = err;
475 }
476
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600477 spin_lock(&wb->list_lock);
Dave Chinner250df6e2011-03-22 22:23:36 +1100478 spin_lock(&inode->i_lock);
Jan Karaccb26b52012-05-03 14:47:58 +0200479 requeue_inode(inode, wb, wbc);
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700480 inode_sync_complete(inode);
Wu Fengguang251d6a42010-12-01 17:33:37 -0600481 trace_writeback_single_inode(inode, wbc, nr_to_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482 return ret;
483}
484
Wu Fengguang1a12d8b2010-08-29 13:28:09 -0600485static long writeback_chunk_size(struct backing_dev_info *bdi,
486 struct wb_writeback_work *work)
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600487{
488 long pages;
489
490 /*
491 * WB_SYNC_ALL mode does livelock avoidance by syncing dirty
492 * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX
493 * here avoids calling into writeback_inodes_wb() more than once.
494 *
495 * The intended call sequence for WB_SYNC_ALL writeback is:
496 *
497 * wb_writeback()
498 * writeback_sb_inodes() <== called only once
499 * write_cache_pages() <== called once for each inode
500 * (quickly) tag currently dirty pages
501 * (maybe slowly) sync all tagged pages
502 */
503 if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages)
504 pages = LONG_MAX;
Wu Fengguang1a12d8b2010-08-29 13:28:09 -0600505 else {
506 pages = min(bdi->avg_write_bandwidth / 2,
507 global_dirty_limit / DIRTY_SCOPE);
508 pages = min(pages, work->nr_pages);
509 pages = round_down(pages + MIN_WRITEBACK_PAGES,
510 MIN_WRITEBACK_PAGES);
511 }
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600512
513 return pages;
514}
515
Jens Axboe03ba3782009-09-09 09:08:54 +0200516/*
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800517 * Write a portion of b_io inodes which belong to @sb.
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200518 *
519 * If @only_this_sb is true, then find and write all such
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800520 * inodes. Otherwise write only ones which go sequentially
521 * in reverse order.
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200522 *
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600523 * Return the number of pages and/or inodes written.
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800524 */
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600525static long writeback_sb_inodes(struct super_block *sb,
526 struct bdi_writeback *wb,
527 struct wb_writeback_work *work)
Jens Axboe03ba3782009-09-09 09:08:54 +0200528{
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600529 struct writeback_control wbc = {
530 .sync_mode = work->sync_mode,
531 .tagged_writepages = work->tagged_writepages,
532 .for_kupdate = work->for_kupdate,
533 .for_background = work->for_background,
534 .range_cyclic = work->range_cyclic,
535 .range_start = 0,
536 .range_end = LLONG_MAX,
537 };
538 unsigned long start_time = jiffies;
539 long write_chunk;
540 long wrote = 0; /* count both pages and inodes */
541
Jens Axboe03ba3782009-09-09 09:08:54 +0200542 while (!list_empty(&wb->b_io)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100543 struct inode *inode = wb_inode(wb->b_io.prev);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200544
545 if (inode->i_sb != sb) {
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600546 if (work->sb) {
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200547 /*
548 * We only want to write back data for this
549 * superblock, move all inodes not belonging
550 * to it back onto the dirty list.
551 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600552 redirty_tail(inode, wb);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200553 continue;
554 }
555
556 /*
557 * The inode belongs to a different superblock.
558 * Bounce back to the caller to unpin this and
559 * pin the next superblock.
560 */
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600561 break;
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200562 }
563
Christoph Hellwig9843b762010-10-24 19:40:46 +0200564 /*
565 * Don't bother with new inodes or inodes beeing freed, first
566 * kind does not need peridic writeout yet, and for the latter
567 * kind writeout is handled by the freer.
568 */
Dave Chinner250df6e2011-03-22 22:23:36 +1100569 spin_lock(&inode->i_lock);
Christoph Hellwig9843b762010-10-24 19:40:46 +0200570 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
Dave Chinner250df6e2011-03-22 22:23:36 +1100571 spin_unlock(&inode->i_lock);
Wu Fengguangfcc5c222011-07-11 23:08:50 -0700572 redirty_tail(inode, wb);
Nick Piggin7ef0d732009-03-12 14:31:38 -0700573 continue;
574 }
Jan Karacc1676d2012-05-03 14:47:56 +0200575 if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) {
576 /*
577 * If this inode is locked for writeback and we are not
578 * doing writeback-for-data-integrity, move it to
579 * b_more_io so that writeback can proceed with the
580 * other inodes on s_io.
581 *
582 * We'll have another go at writing back this inode
583 * when we completed a full scan of b_io.
584 */
585 spin_unlock(&inode->i_lock);
586 requeue_io(inode, wb);
587 trace_writeback_sb_inodes_requeue(inode);
588 continue;
589 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590 __iget(inode);
Wu Fengguang1a12d8b2010-08-29 13:28:09 -0600591 write_chunk = writeback_chunk_size(wb->bdi, work);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600592 wbc.nr_to_write = write_chunk;
593 wbc.pages_skipped = 0;
Dave Chinner250df6e2011-03-22 22:23:36 +1100594
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600595 writeback_single_inode(inode, wb, &wbc);
596
597 work->nr_pages -= write_chunk - wbc.nr_to_write;
598 wrote += write_chunk - wbc.nr_to_write;
599 if (!(inode->i_state & I_DIRTY))
600 wrote++;
601 if (wbc.pages_skipped) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 /*
603 * writeback is not making progress due to locked
604 * buffers. Skip this inode for now.
605 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600606 redirty_tail(inode, wb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 }
Dave Chinner0f1b1fd2011-03-22 22:23:43 +1100608 spin_unlock(&inode->i_lock);
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600609 spin_unlock(&wb->list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 iput(inode);
OGAWA Hirofumi4ffc8442006-03-25 03:07:44 -0800611 cond_resched();
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600612 spin_lock(&wb->list_lock);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600613 /*
614 * bail out to wb_writeback() often enough to check
615 * background threshold and other termination conditions.
616 */
617 if (wrote) {
618 if (time_is_before_jiffies(start_time + HZ / 10UL))
619 break;
620 if (work->nr_pages <= 0)
621 break;
Fengguang Wu8bc3be22008-02-04 22:29:36 -0800622 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 }
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600624 return wrote;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800625}
Nick Piggin38f21972009-01-06 14:40:25 -0800626
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600627static long __writeback_inodes_wb(struct bdi_writeback *wb,
628 struct wb_writeback_work *work)
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800629{
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600630 unsigned long start_time = jiffies;
631 long wrote = 0;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800632
633 while (!list_empty(&wb->b_io)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100634 struct inode *inode = wb_inode(wb->b_io.prev);
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800635 struct super_block *sb = inode->i_sb;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800636
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000637 if (!grab_super_passive(sb)) {
Wu Fengguang0e9958162011-07-29 22:14:35 -0600638 /*
639 * grab_super_passive() may fail consistently due to
640 * s_umount being grabbed by someone else. Don't use
641 * requeue_io() to avoid busy retrying the inode/sb.
642 */
643 redirty_tail(inode, wb);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200644 continue;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800645 }
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600646 wrote += writeback_sb_inodes(sb, wb, work);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200647 drop_super(sb);
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800648
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600649 /* refer to the same tests at the end of writeback_sb_inodes */
650 if (wrote) {
651 if (time_is_before_jiffies(start_time + HZ / 10UL))
652 break;
653 if (work->nr_pages <= 0)
654 break;
655 }
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800656 }
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200657 /* Leave any unwritten inodes on b_io */
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600658 return wrote;
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200659}
660
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600661long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages,
662 enum wb_reason reason)
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200663{
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600664 struct wb_writeback_work work = {
665 .nr_pages = nr_pages,
666 .sync_mode = WB_SYNC_NONE,
667 .range_cyclic = 1,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600668 .reason = reason,
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600669 };
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200670
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600671 spin_lock(&wb->list_lock);
Wu Fengguang424b3512010-07-21 20:11:53 -0600672 if (list_empty(&wb->b_io))
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600673 queue_io(wb, &work);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600674 __writeback_inodes_wb(wb, &work);
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600675 spin_unlock(&wb->list_lock);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200676
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600677 return nr_pages - work.nr_pages;
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200678}
679
Wu Fengguangb00949a2010-11-18 14:38:33 -0600680static bool over_bground_thresh(struct backing_dev_info *bdi)
Jens Axboe03ba3782009-09-09 09:08:54 +0200681{
682 unsigned long background_thresh, dirty_thresh;
683
Wu Fengguang16c40422010-08-11 14:17:39 -0700684 global_dirty_limits(&background_thresh, &dirty_thresh);
Jens Axboe03ba3782009-09-09 09:08:54 +0200685
Wu Fengguangb00949a2010-11-18 14:38:33 -0600686 if (global_page_state(NR_FILE_DIRTY) +
687 global_page_state(NR_UNSTABLE_NFS) > background_thresh)
688 return true;
689
690 if (bdi_stat(bdi, BDI_RECLAIMABLE) >
691 bdi_dirty_limit(bdi, background_thresh))
692 return true;
693
694 return false;
Jens Axboe03ba3782009-09-09 09:08:54 +0200695}
696
697/*
Wu Fengguange98be2d2010-08-29 11:22:30 -0600698 * Called under wb->list_lock. If there are multiple wb per bdi,
699 * only the flusher working on the first wb should do it.
700 */
701static void wb_update_bandwidth(struct bdi_writeback *wb,
702 unsigned long start_time)
703{
Wu Fengguangaf6a3112011-10-03 20:46:17 -0600704 __bdi_update_bandwidth(wb->bdi, 0, 0, 0, 0, 0, start_time);
Wu Fengguange98be2d2010-08-29 11:22:30 -0600705}
706
707/*
Jens Axboe03ba3782009-09-09 09:08:54 +0200708 * Explicit flushing or periodic writeback of "old" data.
709 *
710 * Define "old": the first time one of an inode's pages is dirtied, we mark the
711 * dirtying-time in the inode's address_space. So this periodic writeback code
712 * just walks the superblock inode list, writing back any inodes which are
713 * older than a specific point in time.
714 *
715 * Try to run once per dirty_writeback_interval. But if a writeback event
716 * takes longer than a dirty_writeback_interval interval, then leave a
717 * one-second gap.
718 *
719 * older_than_this takes precedence over nr_to_write. So we'll only write back
720 * all dirty pages if they are all attached to "old" mappings.
721 */
Jens Axboec4a77a62009-09-16 15:18:25 +0200722static long wb_writeback(struct bdi_writeback *wb,
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200723 struct wb_writeback_work *work)
Jens Axboe03ba3782009-09-09 09:08:54 +0200724{
Wu Fengguange98be2d2010-08-29 11:22:30 -0600725 unsigned long wb_start = jiffies;
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600726 long nr_pages = work->nr_pages;
Jens Axboe03ba3782009-09-09 09:08:54 +0200727 unsigned long oldest_jif;
Jan Karaa5989bd2009-09-16 19:22:48 +0200728 struct inode *inode;
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600729 long progress;
Jens Axboe03ba3782009-09-09 09:08:54 +0200730
Wu Fengguange185dda2011-04-23 11:26:07 -0600731 oldest_jif = jiffies;
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600732 work->older_than_this = &oldest_jif;
Jens Axboe03ba3782009-09-09 09:08:54 +0200733
Wu Fengguange8dfc302011-04-21 12:06:32 -0600734 spin_lock(&wb->list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200735 for (;;) {
736 /*
Wu Fengguangd3ddec72009-09-23 20:33:40 +0800737 * Stop writeback when nr_pages has been consumed
Jens Axboe03ba3782009-09-09 09:08:54 +0200738 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200739 if (work->nr_pages <= 0)
Jens Axboe03ba3782009-09-09 09:08:54 +0200740 break;
741
742 /*
Jan Karaaa373cf2011-01-13 15:45:47 -0800743 * Background writeout and kupdate-style writeback may
744 * run forever. Stop them if there is other work to do
745 * so that e.g. sync can proceed. They'll be restarted
746 * after the other works are all done.
747 */
748 if ((work->for_background || work->for_kupdate) &&
749 !list_empty(&wb->bdi->work_list))
750 break;
751
752 /*
Wu Fengguangd3ddec72009-09-23 20:33:40 +0800753 * For background writeout, stop when we are below the
754 * background dirty threshold
Jens Axboe03ba3782009-09-09 09:08:54 +0200755 */
Wu Fengguangb00949a2010-11-18 14:38:33 -0600756 if (work->for_background && !over_bground_thresh(wb->bdi))
Jens Axboe03ba3782009-09-09 09:08:54 +0200757 break;
758
Jan Kara1bc36b62011-10-19 11:44:41 +0200759 /*
760 * Kupdate and background works are special and we want to
761 * include all inodes that need writing. Livelock avoidance is
762 * handled by these works yielding to any other work so we are
763 * safe.
764 */
Wu Fengguangba9aa832010-07-21 20:32:30 -0600765 if (work->for_kupdate) {
766 oldest_jif = jiffies -
767 msecs_to_jiffies(dirty_expire_interval * 10);
Jan Kara1bc36b62011-10-19 11:44:41 +0200768 } else if (work->for_background)
769 oldest_jif = jiffies;
Dave Chinner028c2dd2010-07-07 13:24:07 +1000770
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600771 trace_writeback_start(wb->bdi, work);
Wu Fengguange8dfc302011-04-21 12:06:32 -0600772 if (list_empty(&wb->b_io))
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600773 queue_io(wb, work);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200774 if (work->sb)
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600775 progress = writeback_sb_inodes(work->sb, wb, work);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200776 else
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600777 progress = __writeback_inodes_wb(wb, work);
778 trace_writeback_written(wb->bdi, work);
Dave Chinner028c2dd2010-07-07 13:24:07 +1000779
Wu Fengguange98be2d2010-08-29 11:22:30 -0600780 wb_update_bandwidth(wb, wb_start);
Jens Axboe03ba3782009-09-09 09:08:54 +0200781
782 /*
Jens Axboe71fd05a2009-09-23 19:32:26 +0200783 * Did we write something? Try for more
Wu Fengguange6fb6da2010-07-22 10:23:44 -0600784 *
785 * Dirty inodes are moved to b_io for writeback in batches.
786 * The completion of the current batch does not necessarily
787 * mean the overall work is done. So we keep looping as long
788 * as made some progress on cleaning pages or inodes.
Jens Axboe71fd05a2009-09-23 19:32:26 +0200789 */
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600790 if (progress)
Jens Axboe71fd05a2009-09-23 19:32:26 +0200791 continue;
792 /*
Wu Fengguange6fb6da2010-07-22 10:23:44 -0600793 * No more inodes for IO, bail
Jens Axboe71fd05a2009-09-23 19:32:26 +0200794 */
Wu Fengguangb7a24412010-07-21 22:19:51 -0600795 if (list_empty(&wb->b_more_io))
Jens Axboe03ba3782009-09-09 09:08:54 +0200796 break;
797 /*
Jens Axboe8010c3b2009-09-15 20:04:57 +0200798 * Nothing written. Wait for some inode to
799 * become available for writeback. Otherwise
800 * we'll just busyloop.
Jens Axboe03ba3782009-09-09 09:08:54 +0200801 */
Jens Axboe03ba3782009-09-09 09:08:54 +0200802 if (!list_empty(&wb->b_more_io)) {
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600803 trace_writeback_wait(wb->bdi, work);
Jens Axboe03ba3782009-09-09 09:08:54 +0200804 inode = wb_inode(wb->b_more_io.prev);
Dave Chinner250df6e2011-03-22 22:23:36 +1100805 spin_lock(&inode->i_lock);
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600806 inode_wait_for_writeback(inode, wb);
Dave Chinner250df6e2011-03-22 22:23:36 +1100807 spin_unlock(&inode->i_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200808 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200809 }
Wu Fengguange8dfc302011-04-21 12:06:32 -0600810 spin_unlock(&wb->list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200811
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600812 return nr_pages - work->nr_pages;
Jens Axboe03ba3782009-09-09 09:08:54 +0200813}
814
815/*
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200816 * Return the next wb_writeback_work struct that hasn't been processed yet.
Jens Axboe03ba3782009-09-09 09:08:54 +0200817 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200818static struct wb_writeback_work *
Minchan Kim08852b62010-08-03 12:51:16 +0200819get_next_work_item(struct backing_dev_info *bdi)
Jens Axboe03ba3782009-09-09 09:08:54 +0200820{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200821 struct wb_writeback_work *work = NULL;
Jens Axboe03ba3782009-09-09 09:08:54 +0200822
Artem Bityutskiy64677162010-07-25 14:29:22 +0300823 spin_lock_bh(&bdi->wb_lock);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200824 if (!list_empty(&bdi->work_list)) {
825 work = list_entry(bdi->work_list.next,
826 struct wb_writeback_work, list);
827 list_del_init(&work->list);
Jens Axboe03ba3782009-09-09 09:08:54 +0200828 }
Artem Bityutskiy64677162010-07-25 14:29:22 +0300829 spin_unlock_bh(&bdi->wb_lock);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200830 return work;
Jens Axboe03ba3782009-09-09 09:08:54 +0200831}
832
Linus Torvaldscdf01dd2010-10-30 08:55:52 -0700833/*
834 * Add in the number of potentially dirty inodes, because each inode
835 * write can dirty pagecache in the underlying blockdev.
836 */
837static unsigned long get_nr_dirty_pages(void)
838{
839 return global_page_state(NR_FILE_DIRTY) +
840 global_page_state(NR_UNSTABLE_NFS) +
841 get_nr_dirty_inodes();
842}
843
Jan Kara65850272011-01-13 15:45:44 -0800844static long wb_check_background_flush(struct bdi_writeback *wb)
845{
Wu Fengguangb00949a2010-11-18 14:38:33 -0600846 if (over_bground_thresh(wb->bdi)) {
Jan Kara65850272011-01-13 15:45:44 -0800847
848 struct wb_writeback_work work = {
849 .nr_pages = LONG_MAX,
850 .sync_mode = WB_SYNC_NONE,
851 .for_background = 1,
852 .range_cyclic = 1,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600853 .reason = WB_REASON_BACKGROUND,
Jan Kara65850272011-01-13 15:45:44 -0800854 };
855
856 return wb_writeback(wb, &work);
857 }
858
859 return 0;
860}
861
Jens Axboe03ba3782009-09-09 09:08:54 +0200862static long wb_check_old_data_flush(struct bdi_writeback *wb)
863{
864 unsigned long expired;
865 long nr_pages;
866
Jens Axboe69b62d02010-05-17 12:51:03 +0200867 /*
868 * When set to zero, disable periodic writeback
869 */
870 if (!dirty_writeback_interval)
871 return 0;
872
Jens Axboe03ba3782009-09-09 09:08:54 +0200873 expired = wb->last_old_flush +
874 msecs_to_jiffies(dirty_writeback_interval * 10);
875 if (time_before(jiffies, expired))
876 return 0;
877
878 wb->last_old_flush = jiffies;
Linus Torvaldscdf01dd2010-10-30 08:55:52 -0700879 nr_pages = get_nr_dirty_pages();
Jens Axboe03ba3782009-09-09 09:08:54 +0200880
Jens Axboec4a77a62009-09-16 15:18:25 +0200881 if (nr_pages) {
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200882 struct wb_writeback_work work = {
Jens Axboec4a77a62009-09-16 15:18:25 +0200883 .nr_pages = nr_pages,
884 .sync_mode = WB_SYNC_NONE,
885 .for_kupdate = 1,
886 .range_cyclic = 1,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600887 .reason = WB_REASON_PERIODIC,
Jens Axboec4a77a62009-09-16 15:18:25 +0200888 };
889
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200890 return wb_writeback(wb, &work);
Jens Axboec4a77a62009-09-16 15:18:25 +0200891 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200892
893 return 0;
894}
895
896/*
897 * Retrieve work items and do the writeback they describe
898 */
899long wb_do_writeback(struct bdi_writeback *wb, int force_wait)
900{
901 struct backing_dev_info *bdi = wb->bdi;
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200902 struct wb_writeback_work *work;
Jens Axboec4a77a62009-09-16 15:18:25 +0200903 long wrote = 0;
Jens Axboe03ba3782009-09-09 09:08:54 +0200904
Jan Kara81d73a32010-08-11 14:17:44 -0700905 set_bit(BDI_writeback_running, &wb->bdi->state);
Minchan Kim08852b62010-08-03 12:51:16 +0200906 while ((work = get_next_work_item(bdi)) != NULL) {
Jens Axboe03ba3782009-09-09 09:08:54 +0200907 /*
908 * Override sync mode, in case we must wait for completion
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200909 * because this thread is exiting now.
Jens Axboe03ba3782009-09-09 09:08:54 +0200910 */
911 if (force_wait)
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200912 work->sync_mode = WB_SYNC_ALL;
913
Dave Chinner455b2862010-07-07 13:24:06 +1000914 trace_writeback_exec(bdi, work);
915
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200916 wrote += wb_writeback(wb, work);
Jens Axboe03ba3782009-09-09 09:08:54 +0200917
918 /*
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200919 * Notify the caller of completion if this is a synchronous
920 * work item, otherwise just free it.
Jens Axboe03ba3782009-09-09 09:08:54 +0200921 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200922 if (work->done)
923 complete(work->done);
924 else
925 kfree(work);
Jens Axboe03ba3782009-09-09 09:08:54 +0200926 }
927
928 /*
929 * Check for periodic writeback, kupdated() style
930 */
931 wrote += wb_check_old_data_flush(wb);
Jan Kara65850272011-01-13 15:45:44 -0800932 wrote += wb_check_background_flush(wb);
Jan Kara81d73a32010-08-11 14:17:44 -0700933 clear_bit(BDI_writeback_running, &wb->bdi->state);
Jens Axboe03ba3782009-09-09 09:08:54 +0200934
935 return wrote;
936}
937
938/*
939 * Handle writeback of dirty data for the device backed by this bdi. Also
940 * wakes up periodically and does kupdated style flushing.
941 */
Christoph Hellwig08243902010-06-19 23:08:22 +0200942int bdi_writeback_thread(void *data)
Jens Axboe03ba3782009-09-09 09:08:54 +0200943{
Christoph Hellwig08243902010-06-19 23:08:22 +0200944 struct bdi_writeback *wb = data;
945 struct backing_dev_info *bdi = wb->bdi;
Jens Axboe03ba3782009-09-09 09:08:54 +0200946 long pages_written;
947
Peter Zijlstra766f9162010-10-26 14:22:45 -0700948 current->flags |= PF_SWAPWRITE;
Christoph Hellwig08243902010-06-19 23:08:22 +0200949 set_freezable();
Artem Bityutskiyecd58402010-07-25 14:29:18 +0300950 wb->last_active = jiffies;
Christoph Hellwig08243902010-06-19 23:08:22 +0200951
952 /*
953 * Our parent may run at a different priority, just set us to normal
954 */
955 set_user_nice(current, 0);
956
Dave Chinner455b2862010-07-07 13:24:06 +1000957 trace_writeback_thread_start(bdi);
958
Tejun Heo8a32c442011-11-21 12:32:23 -0800959 while (!kthread_freezable_should_stop(NULL)) {
Artem Bityutskiy64677162010-07-25 14:29:22 +0300960 /*
961 * Remove own delayed wake-up timer, since we are already awake
962 * and we'll take care of the preriodic write-back.
963 */
964 del_timer(&wb->wakeup_timer);
965
Jens Axboe03ba3782009-09-09 09:08:54 +0200966 pages_written = wb_do_writeback(wb, 0);
967
Dave Chinner455b2862010-07-07 13:24:06 +1000968 trace_writeback_pages_written(pages_written);
Jens Axboe03ba3782009-09-09 09:08:54 +0200969
Jens Axboe03ba3782009-09-09 09:08:54 +0200970 if (pages_written)
Artem Bityutskiyecd58402010-07-25 14:29:18 +0300971 wb->last_active = jiffies;
Jens Axboe03ba3782009-09-09 09:08:54 +0200972
Artem Bityutskiy297252c2010-07-25 14:29:15 +0300973 set_current_state(TASK_INTERRUPTIBLE);
J. Bruce Fieldsb76b4012010-08-28 08:52:10 +0200974 if (!list_empty(&bdi->work_list) || kthread_should_stop()) {
Artem Bityutskiy297252c2010-07-25 14:29:15 +0300975 __set_current_state(TASK_RUNNING);
976 continue;
Jens Axboe03ba3782009-09-09 09:08:54 +0200977 }
978
Artem Bityutskiy253c34e2010-07-25 14:29:21 +0300979 if (wb_has_dirty_io(wb) && dirty_writeback_interval)
Artem Bityutskiyfff5b852010-07-25 14:29:20 +0300980 schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10));
Artem Bityutskiy253c34e2010-07-25 14:29:21 +0300981 else {
982 /*
983 * We have nothing to do, so can go sleep without any
984 * timeout and save power. When a work is queued or
985 * something is made dirty - we will be woken up.
986 */
Artem Bityutskiy297252c2010-07-25 14:29:15 +0300987 schedule();
Jens Axboef9eadbb2010-05-18 14:31:45 +0200988 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200989 }
990
Artem Bityutskiyfff5b852010-07-25 14:29:20 +0300991 /* Flush any work that raced with us exiting */
Christoph Hellwig08243902010-06-19 23:08:22 +0200992 if (!list_empty(&bdi->work_list))
993 wb_do_writeback(wb, 1);
Dave Chinner455b2862010-07-07 13:24:06 +1000994
995 trace_writeback_thread_stop(bdi);
Jens Axboe03ba3782009-09-09 09:08:54 +0200996 return 0;
997}
998
Christoph Hellwig08243902010-06-19 23:08:22 +0200999
Jens Axboe03ba3782009-09-09 09:08:54 +02001000/*
Jens Axboe03ba3782009-09-09 09:08:54 +02001001 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
1002 * the whole world.
1003 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001004void wakeup_flusher_threads(long nr_pages, enum wb_reason reason)
Jens Axboe03ba3782009-09-09 09:08:54 +02001005{
Christoph Hellwigb8c2f342010-06-08 18:15:07 +02001006 struct backing_dev_info *bdi;
Christoph Hellwigb8c2f342010-06-08 18:15:07 +02001007
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001008 if (!nr_pages) {
1009 nr_pages = global_page_state(NR_FILE_DIRTY) +
Jens Axboe03ba3782009-09-09 09:08:54 +02001010 global_page_state(NR_UNSTABLE_NFS);
Christoph Hellwigb8c2f342010-06-08 18:15:07 +02001011 }
1012
1013 rcu_read_lock();
1014 list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) {
1015 if (!bdi_has_dirty_io(bdi))
1016 continue;
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001017 __bdi_start_writeback(bdi, nr_pages, false, reason);
Christoph Hellwigb8c2f342010-06-08 18:15:07 +02001018 }
1019 rcu_read_unlock();
Jens Axboe03ba3782009-09-09 09:08:54 +02001020}
1021
1022static noinline void block_dump___mark_inode_dirty(struct inode *inode)
1023{
1024 if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) {
1025 struct dentry *dentry;
1026 const char *name = "?";
1027
1028 dentry = d_find_alias(inode);
1029 if (dentry) {
1030 spin_lock(&dentry->d_lock);
1031 name = (const char *) dentry->d_name.name;
1032 }
1033 printk(KERN_DEBUG
1034 "%s(%d): dirtied inode %lu (%s) on %s\n",
1035 current->comm, task_pid_nr(current), inode->i_ino,
1036 name, inode->i_sb->s_id);
1037 if (dentry) {
1038 spin_unlock(&dentry->d_lock);
1039 dput(dentry);
1040 }
1041 }
1042}
1043
1044/**
1045 * __mark_inode_dirty - internal function
1046 * @inode: inode to mark
1047 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
1048 * Mark an inode as dirty. Callers should use mark_inode_dirty or
1049 * mark_inode_dirty_sync.
1050 *
1051 * Put the inode on the super block's dirty list.
1052 *
1053 * CAREFUL! We mark it dirty unconditionally, but move it onto the
1054 * dirty list only if it is hashed or if it refers to a blockdev.
1055 * If it was not hashed, it will never be added to the dirty list
1056 * even if it is later hashed, as it will have been marked dirty already.
1057 *
1058 * In short, make sure you hash any inodes _before_ you start marking
1059 * them dirty.
1060 *
Jens Axboe03ba3782009-09-09 09:08:54 +02001061 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
1062 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
1063 * the kernel-internal blockdev inode represents the dirtying time of the
1064 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
1065 * page->mapping->host, so the page-dirtying time is recorded in the internal
1066 * blockdev inode.
1067 */
1068void __mark_inode_dirty(struct inode *inode, int flags)
1069{
1070 struct super_block *sb = inode->i_sb;
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001071 struct backing_dev_info *bdi = NULL;
Jens Axboe03ba3782009-09-09 09:08:54 +02001072
1073 /*
1074 * Don't do this for I_DIRTY_PAGES - that doesn't actually
1075 * dirty the inode itself
1076 */
1077 if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
1078 if (sb->s_op->dirty_inode)
Christoph Hellwigaa385722011-05-27 06:53:02 -04001079 sb->s_op->dirty_inode(inode, flags);
Jens Axboe03ba3782009-09-09 09:08:54 +02001080 }
1081
1082 /*
1083 * make sure that changes are seen by all cpus before we test i_state
1084 * -- mikulas
1085 */
1086 smp_mb();
1087
1088 /* avoid the locking if we can */
1089 if ((inode->i_state & flags) == flags)
1090 return;
1091
1092 if (unlikely(block_dump))
1093 block_dump___mark_inode_dirty(inode);
1094
Dave Chinner250df6e2011-03-22 22:23:36 +11001095 spin_lock(&inode->i_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001096 if ((inode->i_state & flags) != flags) {
1097 const int was_dirty = inode->i_state & I_DIRTY;
1098
1099 inode->i_state |= flags;
1100
1101 /*
1102 * If the inode is being synced, just update its dirty state.
1103 * The unlocker will place the inode on the appropriate
1104 * superblock list, based upon its state.
1105 */
1106 if (inode->i_state & I_SYNC)
Dave Chinner250df6e2011-03-22 22:23:36 +11001107 goto out_unlock_inode;
Jens Axboe03ba3782009-09-09 09:08:54 +02001108
1109 /*
1110 * Only add valid (hashed) inodes to the superblock's
1111 * dirty list. Add blockdev inodes as well.
1112 */
1113 if (!S_ISBLK(inode->i_mode)) {
Al Viro1d3382cb2010-10-23 15:19:20 -04001114 if (inode_unhashed(inode))
Dave Chinner250df6e2011-03-22 22:23:36 +11001115 goto out_unlock_inode;
Jens Axboe03ba3782009-09-09 09:08:54 +02001116 }
Al Viroa4ffdde2010-06-02 17:38:30 -04001117 if (inode->i_state & I_FREEING)
Dave Chinner250df6e2011-03-22 22:23:36 +11001118 goto out_unlock_inode;
Jens Axboe03ba3782009-09-09 09:08:54 +02001119
1120 /*
1121 * If the inode was already on b_dirty/b_io/b_more_io, don't
1122 * reposition it (that would break b_dirty time-ordering).
1123 */
1124 if (!was_dirty) {
Dave Chinnera66979a2011-03-22 22:23:41 +11001125 bool wakeup_bdi = false;
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001126 bdi = inode_to_bdi(inode);
Jens Axboe500b0672009-09-09 09:10:25 +02001127
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001128 if (bdi_cap_writeback_dirty(bdi)) {
1129 WARN(!test_bit(BDI_registered, &bdi->state),
1130 "bdi-%s not registered\n", bdi->name);
1131
1132 /*
1133 * If this is the first dirty inode for this
1134 * bdi, we have to wake-up the corresponding
1135 * bdi thread to make sure background
1136 * write-back happens later.
1137 */
1138 if (!wb_has_dirty_io(&bdi->wb))
1139 wakeup_bdi = true;
Jens Axboe500b0672009-09-09 09:10:25 +02001140 }
Jens Axboe03ba3782009-09-09 09:08:54 +02001141
Dave Chinnera66979a2011-03-22 22:23:41 +11001142 spin_unlock(&inode->i_lock);
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001143 spin_lock(&bdi->wb.list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001144 inode->dirtied_when = jiffies;
Nick Piggin7ccf19a2010-10-21 11:49:30 +11001145 list_move(&inode->i_wb_list, &bdi->wb.b_dirty);
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001146 spin_unlock(&bdi->wb.list_lock);
Dave Chinnera66979a2011-03-22 22:23:41 +11001147
1148 if (wakeup_bdi)
1149 bdi_wakeup_thread_delayed(bdi);
1150 return;
Jens Axboe03ba3782009-09-09 09:08:54 +02001151 }
1152 }
Dave Chinner250df6e2011-03-22 22:23:36 +11001153out_unlock_inode:
1154 spin_unlock(&inode->i_lock);
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001155
Jens Axboe03ba3782009-09-09 09:08:54 +02001156}
1157EXPORT_SYMBOL(__mark_inode_dirty);
1158
Jens Axboeb6e51312009-09-16 15:13:54 +02001159static void wait_sb_inodes(struct super_block *sb)
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001160{
Jens Axboe03ba3782009-09-09 09:08:54 +02001161 struct inode *inode, *old_inode = NULL;
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001162
Jens Axboe03ba3782009-09-09 09:08:54 +02001163 /*
1164 * We need to be protected against the filesystem going from
1165 * r/o to r/w or vice versa.
1166 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001167 WARN_ON(!rwsem_is_locked(&sb->s_umount));
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001168
Dave Chinner55fa6092011-03-22 22:23:40 +11001169 spin_lock(&inode_sb_list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001170
1171 /*
1172 * Data integrity sync. Must wait for all pages under writeback,
1173 * because there may have been pages dirtied before our sync
1174 * call, but which had writeout started before we write it out.
1175 * In which case, the inode may not be on the dirty list, but
1176 * we still have to wait for that writeout.
1177 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001178 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
Dave Chinner250df6e2011-03-22 22:23:36 +11001179 struct address_space *mapping = inode->i_mapping;
Jens Axboe03ba3782009-09-09 09:08:54 +02001180
Dave Chinner250df6e2011-03-22 22:23:36 +11001181 spin_lock(&inode->i_lock);
1182 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
1183 (mapping->nrpages == 0)) {
1184 spin_unlock(&inode->i_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001185 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +11001186 }
Jens Axboe03ba3782009-09-09 09:08:54 +02001187 __iget(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +11001188 spin_unlock(&inode->i_lock);
Dave Chinner55fa6092011-03-22 22:23:40 +11001189 spin_unlock(&inode_sb_list_lock);
1190
Jens Axboe03ba3782009-09-09 09:08:54 +02001191 /*
Dave Chinner55fa6092011-03-22 22:23:40 +11001192 * We hold a reference to 'inode' so it couldn't have been
1193 * removed from s_inodes list while we dropped the
1194 * inode_sb_list_lock. We cannot iput the inode now as we can
1195 * be holding the last reference and we cannot iput it under
1196 * inode_sb_list_lock. So we keep the reference and iput it
1197 * later.
Jens Axboe03ba3782009-09-09 09:08:54 +02001198 */
1199 iput(old_inode);
1200 old_inode = inode;
1201
1202 filemap_fdatawait(mapping);
1203
1204 cond_resched();
Nick Piggin38f21972009-01-06 14:40:25 -08001205
Dave Chinner55fa6092011-03-22 22:23:40 +11001206 spin_lock(&inode_sb_list_lock);
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001207 }
Dave Chinner55fa6092011-03-22 22:23:40 +11001208 spin_unlock(&inode_sb_list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001209 iput(old_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210}
1211
Jens Axboed8a85592009-09-02 12:34:32 +02001212/**
Chris Mason3259f8b2010-10-29 11:16:17 -04001213 * writeback_inodes_sb_nr - writeback dirty inodes from given super_block
Jens Axboed8a85592009-09-02 12:34:32 +02001214 * @sb: the superblock
Chris Mason3259f8b2010-10-29 11:16:17 -04001215 * @nr: the number of pages to write
Marcos Paulo de Souza786228a2011-11-23 20:56:45 +08001216 * @reason: reason why some writeback work initiated
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217 *
Jens Axboed8a85592009-09-02 12:34:32 +02001218 * Start writeback on some inodes on this super_block. No guarantees are made
1219 * on how many (if any) will be written, and this function does not wait
Chris Mason3259f8b2010-10-29 11:16:17 -04001220 * for IO completion of submitted IO.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001222void writeback_inodes_sb_nr(struct super_block *sb,
1223 unsigned long nr,
1224 enum wb_reason reason)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001226 DECLARE_COMPLETION_ONSTACK(done);
1227 struct wb_writeback_work work = {
Wu Fengguang6e6938b2010-06-06 10:38:15 -06001228 .sb = sb,
1229 .sync_mode = WB_SYNC_NONE,
1230 .tagged_writepages = 1,
1231 .done = &done,
1232 .nr_pages = nr,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001233 .reason = reason,
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001234 };
Jens Axboe0e3c9a22010-06-01 11:08:43 +02001235
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001236 WARN_ON(!rwsem_is_locked(&sb->s_umount));
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001237 bdi_queue_work(sb->s_bdi, &work);
1238 wait_for_completion(&done);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239}
Chris Mason3259f8b2010-10-29 11:16:17 -04001240EXPORT_SYMBOL(writeback_inodes_sb_nr);
1241
1242/**
1243 * writeback_inodes_sb - writeback dirty inodes from given super_block
1244 * @sb: the superblock
Marcos Paulo de Souza786228a2011-11-23 20:56:45 +08001245 * @reason: reason why some writeback work was initiated
Chris Mason3259f8b2010-10-29 11:16:17 -04001246 *
1247 * Start writeback on some inodes on this super_block. No guarantees are made
1248 * on how many (if any) will be written, and this function does not wait
1249 * for IO completion of submitted IO.
1250 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001251void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason)
Chris Mason3259f8b2010-10-29 11:16:17 -04001252{
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001253 return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason);
Chris Mason3259f8b2010-10-29 11:16:17 -04001254}
Jens Axboed8a85592009-09-02 12:34:32 +02001255EXPORT_SYMBOL(writeback_inodes_sb);
1256
1257/**
Eric Sandeen17bd55d2009-12-23 07:57:07 -05001258 * writeback_inodes_sb_if_idle - start writeback if none underway
1259 * @sb: the superblock
Marcos Paulo de Souza786228a2011-11-23 20:56:45 +08001260 * @reason: reason why some writeback work was initiated
Eric Sandeen17bd55d2009-12-23 07:57:07 -05001261 *
1262 * Invoke writeback_inodes_sb if no writeback is currently underway.
1263 * Returns 1 if writeback was started, 0 if not.
1264 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001265int writeback_inodes_sb_if_idle(struct super_block *sb, enum wb_reason reason)
Eric Sandeen17bd55d2009-12-23 07:57:07 -05001266{
1267 if (!writeback_in_progress(sb->s_bdi)) {
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001268 down_read(&sb->s_umount);
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001269 writeback_inodes_sb(sb, reason);
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001270 up_read(&sb->s_umount);
Eric Sandeen17bd55d2009-12-23 07:57:07 -05001271 return 1;
1272 } else
1273 return 0;
1274}
1275EXPORT_SYMBOL(writeback_inodes_sb_if_idle);
1276
1277/**
Fengguang Wuc097b2c2012-03-05 15:08:06 -08001278 * writeback_inodes_sb_nr_if_idle - start writeback if none underway
Chris Mason3259f8b2010-10-29 11:16:17 -04001279 * @sb: the superblock
1280 * @nr: the number of pages to write
Marcos Paulo de Souza786228a2011-11-23 20:56:45 +08001281 * @reason: reason why some writeback work was initiated
Chris Mason3259f8b2010-10-29 11:16:17 -04001282 *
1283 * Invoke writeback_inodes_sb if no writeback is currently underway.
1284 * Returns 1 if writeback was started, 0 if not.
1285 */
1286int writeback_inodes_sb_nr_if_idle(struct super_block *sb,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001287 unsigned long nr,
1288 enum wb_reason reason)
Chris Mason3259f8b2010-10-29 11:16:17 -04001289{
1290 if (!writeback_in_progress(sb->s_bdi)) {
1291 down_read(&sb->s_umount);
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001292 writeback_inodes_sb_nr(sb, nr, reason);
Chris Mason3259f8b2010-10-29 11:16:17 -04001293 up_read(&sb->s_umount);
1294 return 1;
1295 } else
1296 return 0;
1297}
1298EXPORT_SYMBOL(writeback_inodes_sb_nr_if_idle);
1299
1300/**
Jens Axboed8a85592009-09-02 12:34:32 +02001301 * sync_inodes_sb - sync sb inode pages
1302 * @sb: the superblock
1303 *
1304 * This function writes and waits on any dirty inode belonging to this
Stefan Hajnoczicb9ef8d2011-01-13 15:47:26 -08001305 * super_block.
Jens Axboed8a85592009-09-02 12:34:32 +02001306 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001307void sync_inodes_sb(struct super_block *sb)
Jens Axboed8a85592009-09-02 12:34:32 +02001308{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001309 DECLARE_COMPLETION_ONSTACK(done);
1310 struct wb_writeback_work work = {
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001311 .sb = sb,
1312 .sync_mode = WB_SYNC_ALL,
1313 .nr_pages = LONG_MAX,
1314 .range_cyclic = 0,
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001315 .done = &done,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001316 .reason = WB_REASON_SYNC,
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001317 };
1318
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001319 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1320
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001321 bdi_queue_work(sb->s_bdi, &work);
1322 wait_for_completion(&done);
1323
Jens Axboeb6e51312009-09-16 15:13:54 +02001324 wait_sb_inodes(sb);
Jens Axboed8a85592009-09-02 12:34:32 +02001325}
1326EXPORT_SYMBOL(sync_inodes_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328/**
Andrea Arcangeli7f04c262005-10-30 15:03:05 -08001329 * write_inode_now - write an inode to disk
1330 * @inode: inode to write to disk
1331 * @sync: whether the write should be synchronous or not
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332 *
Andrea Arcangeli7f04c262005-10-30 15:03:05 -08001333 * This function commits an inode to disk immediately if it is dirty. This is
1334 * primarily needed by knfsd.
1335 *
1336 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338int write_inode_now(struct inode *inode, int sync)
1339{
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001340 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341 int ret;
1342 struct writeback_control wbc = {
1343 .nr_to_write = LONG_MAX,
Mike Galbraith18914b12008-02-08 04:20:23 -08001344 .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -07001345 .range_start = 0,
1346 .range_end = LLONG_MAX,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347 };
1348
1349 if (!mapping_cap_writeback_dirty(inode->i_mapping))
Andrew Morton49364ce2005-11-07 00:59:15 -08001350 wbc.nr_to_write = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351
1352 might_sleep();
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001353 spin_lock(&wb->list_lock);
Dave Chinner0f1b1fd2011-03-22 22:23:43 +11001354 spin_lock(&inode->i_lock);
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001355 ret = writeback_single_inode(inode, wb, &wbc);
Dave Chinner0f1b1fd2011-03-22 22:23:43 +11001356 spin_unlock(&inode->i_lock);
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001357 spin_unlock(&wb->list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358 return ret;
1359}
1360EXPORT_SYMBOL(write_inode_now);
1361
1362/**
1363 * sync_inode - write an inode and its pages to disk.
1364 * @inode: the inode to sync
1365 * @wbc: controls the writeback mode
1366 *
1367 * sync_inode() will write an inode and its pages to disk. It will also
1368 * correctly update the inode on its superblock's dirty inode lists and will
1369 * update inode->i_state.
1370 *
1371 * The caller must have a ref on the inode.
1372 */
1373int sync_inode(struct inode *inode, struct writeback_control *wbc)
1374{
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001375 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376 int ret;
1377
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001378 spin_lock(&wb->list_lock);
Dave Chinner0f1b1fd2011-03-22 22:23:43 +11001379 spin_lock(&inode->i_lock);
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001380 ret = writeback_single_inode(inode, wb, wbc);
Dave Chinner0f1b1fd2011-03-22 22:23:43 +11001381 spin_unlock(&inode->i_lock);
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001382 spin_unlock(&wb->list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383 return ret;
1384}
1385EXPORT_SYMBOL(sync_inode);
Christoph Hellwigc37650162010-10-06 10:48:20 +02001386
1387/**
Andrew Mortonc691b9d2011-01-13 15:45:48 -08001388 * sync_inode_metadata - write an inode to disk
Christoph Hellwigc37650162010-10-06 10:48:20 +02001389 * @inode: the inode to sync
1390 * @wait: wait for I/O to complete.
1391 *
Andrew Mortonc691b9d2011-01-13 15:45:48 -08001392 * Write an inode to disk and adjust its dirty state after completion.
Christoph Hellwigc37650162010-10-06 10:48:20 +02001393 *
1394 * Note: only writes the actual inode, no associated data or other metadata.
1395 */
1396int sync_inode_metadata(struct inode *inode, int wait)
1397{
1398 struct writeback_control wbc = {
1399 .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE,
1400 .nr_to_write = 0, /* metadata-only */
1401 };
1402
1403 return sync_inode(inode, &wbc);
1404}
1405EXPORT_SYMBOL(sync_inode_metadata);