blob: 8f660dd6137a1105ba648e1a1412ffc5900c1233 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * fs/fs-writeback.c
3 *
4 * Copyright (C) 2002, Linus Torvalds.
5 *
6 * Contains all the functions related to writing back and waiting
7 * upon dirty inodes against superblocks, and writing back dirty
8 * pages against inodes. ie: data writeback. Writeout of the
9 * inode itself is not handled here.
10 *
Francois Camie1f8e872008-10-15 22:01:59 -070011 * 10Apr2002 Andrew Morton
Linus Torvalds1da177e2005-04-16 15:20:36 -070012 * Split out of fs/inode.c
13 * Additions for address_space-based writeback
14 */
15
16#include <linux/kernel.h>
Paul Gortmaker630d9c42011-11-16 23:57:37 -050017#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include <linux/spinlock.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090019#include <linux/slab.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/sched.h>
21#include <linux/fs.h>
22#include <linux/mm.h>
Wu Fengguangbc31b862012-01-07 20:41:55 -060023#include <linux/pagemap.h>
Jens Axboe03ba3782009-09-09 09:08:54 +020024#include <linux/kthread.h>
25#include <linux/freezer.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include <linux/writeback.h>
27#include <linux/blkdev.h>
28#include <linux/backing-dev.h>
Dave Chinner455b2862010-07-07 13:24:06 +100029#include <linux/tracepoint.h>
David Howells07f3f05c2006-09-30 20:52:18 +020030#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070031
Jens Axboed0bceac2009-05-18 08:20:32 +020032/*
Wu Fengguangbc31b862012-01-07 20:41:55 -060033 * 4MB minimal write chunk size
34 */
35#define MIN_WRITEBACK_PAGES (4096UL >> (PAGE_CACHE_SHIFT - 10))
36
37/*
Jens Axboec4a77a62009-09-16 15:18:25 +020038 * Passed into wb_writeback(), essentially a subset of writeback_control
39 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +020040struct wb_writeback_work {
Jens Axboec4a77a62009-09-16 15:18:25 +020041 long nr_pages;
42 struct super_block *sb;
Wu Fengguangd46db3d2011-05-04 19:54:37 -060043 unsigned long *older_than_this;
Jens Axboec4a77a62009-09-16 15:18:25 +020044 enum writeback_sync_modes sync_mode;
Wu Fengguang6e6938b2010-06-06 10:38:15 -060045 unsigned int tagged_writepages:1;
H Hartley Sweeten52957fe2010-04-01 20:36:30 -050046 unsigned int for_kupdate:1;
47 unsigned int range_cyclic:1;
48 unsigned int for_background:1;
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -060049 enum wb_reason reason; /* why was writeback initiated? */
Jens Axboec4a77a62009-09-16 15:18:25 +020050
Jens Axboe8010c3b2009-09-15 20:04:57 +020051 struct list_head list; /* pending work list */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +020052 struct completion *done; /* set if the caller waits */
Jens Axboe03ba3782009-09-09 09:08:54 +020053};
54
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/*
Jan Kara169ebd92012-05-03 14:48:03 +0200329 * Wait for writeback on an inode to complete. Called with i_lock held.
330 * Caller must make sure inode cannot go away when we drop i_lock.
Christoph Hellwig01c03192009-06-08 13:35:40 +0200331 */
Jan Kara169ebd92012-05-03 14:48:03 +0200332static void __inode_wait_for_writeback(struct inode *inode)
333 __releases(inode->i_lock)
334 __acquires(inode->i_lock)
Christoph Hellwig01c03192009-06-08 13:35:40 +0200335{
336 DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC);
337 wait_queue_head_t *wqh;
338
339 wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
Dave Chinner250df6e2011-03-22 22:23:36 +1100340 while (inode->i_state & I_SYNC) {
341 spin_unlock(&inode->i_lock);
Christoph Hellwig01c03192009-06-08 13:35:40 +0200342 __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE);
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 Kara169ebd92012-05-03 14:48:03 +0200348 * Wait for writeback on an inode to complete. Caller must have inode pinned.
349 */
350void inode_wait_for_writeback(struct inode *inode)
351{
352 spin_lock(&inode->i_lock);
353 __inode_wait_for_writeback(inode);
354 spin_unlock(&inode->i_lock);
355}
356
357/*
358 * Sleep until I_SYNC is cleared. This function must be called with i_lock
359 * held and drops it. It is aimed for callers not holding any inode reference
360 * so once i_lock is dropped, inode can go away.
361 */
362static void inode_sleep_on_writeback(struct inode *inode)
363 __releases(inode->i_lock)
364{
365 DEFINE_WAIT(wait);
366 wait_queue_head_t *wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
367 int sleep;
368
369 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
370 sleep = inode->i_state & I_SYNC;
371 spin_unlock(&inode->i_lock);
372 if (sleep)
373 schedule();
374 finish_wait(wqh, &wait);
375}
376
377/*
Jan Karaccb26b52012-05-03 14:47:58 +0200378 * Find proper writeback list for the inode depending on its current state and
379 * possibly also change of its state while we were doing writeback. Here we
380 * handle things such as livelock prevention or fairness of writeback among
381 * inodes. This function can be called only by flusher thread - noone else
382 * processes all inodes in writeback lists and requeueing inodes behind flusher
383 * thread's back can have unexpected consequences.
384 */
385static void requeue_inode(struct inode *inode, struct bdi_writeback *wb,
386 struct writeback_control *wbc)
387{
388 if (inode->i_state & I_FREEING)
389 return;
390
391 /*
392 * Sync livelock prevention. Each inode is tagged and synced in one
393 * shot. If still dirty, it will be redirty_tail()'ed below. Update
394 * the dirty time to prevent enqueue and sync it again.
395 */
396 if ((inode->i_state & I_DIRTY) &&
397 (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages))
398 inode->dirtied_when = jiffies;
399
Jan Kara4f8ad652012-05-03 14:48:00 +0200400 if (wbc->pages_skipped) {
401 /*
402 * writeback is not making progress due to locked
403 * buffers. Skip this inode for now.
404 */
405 redirty_tail(inode, wb);
406 return;
407 }
408
Jan Karaccb26b52012-05-03 14:47:58 +0200409 if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) {
410 /*
411 * We didn't write back all the pages. nfs_writepages()
412 * sometimes bales out without doing anything.
413 */
414 if (wbc->nr_to_write <= 0) {
415 /* Slice used up. Queue for next turn. */
416 requeue_io(inode, wb);
417 } else {
418 /*
419 * Writeback blocked by something other than
420 * congestion. Delay the inode for some time to
421 * avoid spinning on the CPU (100% iowait)
422 * retrying writeback of the dirty page/inode
423 * that cannot be performed immediately.
424 */
425 redirty_tail(inode, wb);
426 }
427 } else if (inode->i_state & I_DIRTY) {
428 /*
429 * Filesystems can dirty the inode during writeback operations,
430 * such as delayed allocation during submission or metadata
431 * updates after data IO completion.
432 */
433 redirty_tail(inode, wb);
434 } else {
435 /* The inode is clean. Remove from writeback lists. */
436 list_del_init(&inode->i_wb_list);
437 }
438}
439
440/*
Jan Kara4f8ad652012-05-03 14:48:00 +0200441 * Write out an inode and its dirty pages. Do not update the writeback list
442 * linkage. That is left to the caller. The caller is also responsible for
443 * setting I_SYNC flag and calling inode_sync_complete() to clear it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444 */
445static int
Jan Kara4f8ad652012-05-03 14:48:00 +0200446__writeback_single_inode(struct inode *inode, struct bdi_writeback *wb,
447 struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449 struct address_space *mapping = inode->i_mapping;
Wu Fengguang251d6a42010-12-01 17:33:37 -0600450 long nr_to_write = wbc->nr_to_write;
Christoph Hellwig01c03192009-06-08 13:35:40 +0200451 unsigned dirty;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452 int ret;
453
Jan Kara4f8ad652012-05-03 14:48:00 +0200454 WARN_ON(!(inode->i_state & I_SYNC));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455
456 ret = do_writepages(mapping, wbc);
457
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100458 /*
459 * Make sure to wait on the data before writing out the metadata.
460 * This is important for filesystems that modify metadata on data
461 * I/O completion.
462 */
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100463 if (wbc->sync_mode == WB_SYNC_ALL) {
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100464 int err = filemap_fdatawait(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465 if (ret == 0)
466 ret = err;
467 }
468
Dmitry Monakhov5547e8a2010-05-07 13:35:44 +0400469 /*
470 * Some filesystems may redirty the inode during the writeback
471 * due to delalloc, clear dirty metadata flags right before
472 * write_inode()
473 */
Dave Chinner250df6e2011-03-22 22:23:36 +1100474 spin_lock(&inode->i_lock);
Jan Kara6290be12012-05-03 14:47:57 +0200475 /* Clear I_DIRTY_PAGES if we've written out all dirty pages */
476 if (!mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
477 inode->i_state &= ~I_DIRTY_PAGES;
Dmitry Monakhov5547e8a2010-05-07 13:35:44 +0400478 dirty = inode->i_state & I_DIRTY;
479 inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC);
Dave Chinner250df6e2011-03-22 22:23:36 +1100480 spin_unlock(&inode->i_lock);
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100481 /* Don't write the inode if only I_DIRTY_PAGES was set */
482 if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100483 int err = write_inode(inode, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484 if (ret == 0)
485 ret = err;
486 }
Jan Kara4f8ad652012-05-03 14:48:00 +0200487 trace_writeback_single_inode(inode, wbc, nr_to_write);
488 return ret;
489}
490
491/*
492 * Write out an inode's dirty pages. Either the caller has an active reference
493 * on the inode or the inode has I_WILL_FREE set.
494 *
495 * This function is designed to be called for writing back one inode which
496 * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode()
497 * and does more profound writeback list handling in writeback_sb_inodes().
498 */
499static int
500writeback_single_inode(struct inode *inode, struct bdi_writeback *wb,
501 struct writeback_control *wbc)
502{
503 int ret = 0;
504
505 spin_lock(&inode->i_lock);
506 if (!atomic_read(&inode->i_count))
507 WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING)));
508 else
509 WARN_ON(inode->i_state & I_WILL_FREE);
510
511 if (inode->i_state & I_SYNC) {
512 if (wbc->sync_mode != WB_SYNC_ALL)
513 goto out;
514 /*
Jan Kara169ebd92012-05-03 14:48:03 +0200515 * It's a data-integrity sync. We must wait. Since callers hold
516 * inode reference or inode has I_WILL_FREE set, it cannot go
517 * away under us.
Jan Kara4f8ad652012-05-03 14:48:00 +0200518 */
Jan Kara169ebd92012-05-03 14:48:03 +0200519 __inode_wait_for_writeback(inode);
Jan Kara4f8ad652012-05-03 14:48:00 +0200520 }
521 WARN_ON(inode->i_state & I_SYNC);
522 /*
523 * Skip inode if it is clean. We don't want to mess with writeback
524 * lists in this function since flusher thread may be doing for example
525 * sync in parallel and if we move the inode, it could get skipped. So
526 * here we make sure inode is on some writeback list and leave it there
527 * unless we have completely cleaned the inode.
528 */
529 if (!(inode->i_state & I_DIRTY))
530 goto out;
531 inode->i_state |= I_SYNC;
532 spin_unlock(&inode->i_lock);
533
534 ret = __writeback_single_inode(inode, wb, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600536 spin_lock(&wb->list_lock);
Dave Chinner250df6e2011-03-22 22:23:36 +1100537 spin_lock(&inode->i_lock);
Jan Kara4f8ad652012-05-03 14:48:00 +0200538 /*
539 * If inode is clean, remove it from writeback lists. Otherwise don't
540 * touch it. See comment above for explanation.
541 */
542 if (!(inode->i_state & I_DIRTY))
543 list_del_init(&inode->i_wb_list);
544 spin_unlock(&wb->list_lock);
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700545 inode_sync_complete(inode);
Jan Kara4f8ad652012-05-03 14:48:00 +0200546out:
547 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548 return ret;
549}
550
Wu Fengguang1a12d8b2010-08-29 13:28:09 -0600551static long writeback_chunk_size(struct backing_dev_info *bdi,
552 struct wb_writeback_work *work)
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600553{
554 long pages;
555
556 /*
557 * WB_SYNC_ALL mode does livelock avoidance by syncing dirty
558 * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX
559 * here avoids calling into writeback_inodes_wb() more than once.
560 *
561 * The intended call sequence for WB_SYNC_ALL writeback is:
562 *
563 * wb_writeback()
564 * writeback_sb_inodes() <== called only once
565 * write_cache_pages() <== called once for each inode
566 * (quickly) tag currently dirty pages
567 * (maybe slowly) sync all tagged pages
568 */
569 if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages)
570 pages = LONG_MAX;
Wu Fengguang1a12d8b2010-08-29 13:28:09 -0600571 else {
572 pages = min(bdi->avg_write_bandwidth / 2,
573 global_dirty_limit / DIRTY_SCOPE);
574 pages = min(pages, work->nr_pages);
575 pages = round_down(pages + MIN_WRITEBACK_PAGES,
576 MIN_WRITEBACK_PAGES);
577 }
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600578
579 return pages;
580}
581
Jens Axboe03ba3782009-09-09 09:08:54 +0200582/*
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800583 * Write a portion of b_io inodes which belong to @sb.
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200584 *
585 * If @only_this_sb is true, then find and write all such
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800586 * inodes. Otherwise write only ones which go sequentially
587 * in reverse order.
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200588 *
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600589 * Return the number of pages and/or inodes written.
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800590 */
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600591static long writeback_sb_inodes(struct super_block *sb,
592 struct bdi_writeback *wb,
593 struct wb_writeback_work *work)
Jens Axboe03ba3782009-09-09 09:08:54 +0200594{
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600595 struct writeback_control wbc = {
596 .sync_mode = work->sync_mode,
597 .tagged_writepages = work->tagged_writepages,
598 .for_kupdate = work->for_kupdate,
599 .for_background = work->for_background,
600 .range_cyclic = work->range_cyclic,
601 .range_start = 0,
602 .range_end = LLONG_MAX,
603 };
604 unsigned long start_time = jiffies;
605 long write_chunk;
606 long wrote = 0; /* count both pages and inodes */
607
Jens Axboe03ba3782009-09-09 09:08:54 +0200608 while (!list_empty(&wb->b_io)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100609 struct inode *inode = wb_inode(wb->b_io.prev);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200610
611 if (inode->i_sb != sb) {
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600612 if (work->sb) {
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200613 /*
614 * We only want to write back data for this
615 * superblock, move all inodes not belonging
616 * to it back onto the dirty list.
617 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600618 redirty_tail(inode, wb);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200619 continue;
620 }
621
622 /*
623 * The inode belongs to a different superblock.
624 * Bounce back to the caller to unpin this and
625 * pin the next superblock.
626 */
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600627 break;
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200628 }
629
Christoph Hellwig9843b762010-10-24 19:40:46 +0200630 /*
631 * Don't bother with new inodes or inodes beeing freed, first
632 * kind does not need peridic writeout yet, and for the latter
633 * kind writeout is handled by the freer.
634 */
Dave Chinner250df6e2011-03-22 22:23:36 +1100635 spin_lock(&inode->i_lock);
Christoph Hellwig9843b762010-10-24 19:40:46 +0200636 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
Dave Chinner250df6e2011-03-22 22:23:36 +1100637 spin_unlock(&inode->i_lock);
Wu Fengguangfcc5c222011-07-11 23:08:50 -0700638 redirty_tail(inode, wb);
Nick Piggin7ef0d732009-03-12 14:31:38 -0700639 continue;
640 }
Jan Karacc1676d2012-05-03 14:47:56 +0200641 if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) {
642 /*
643 * If this inode is locked for writeback and we are not
644 * doing writeback-for-data-integrity, move it to
645 * b_more_io so that writeback can proceed with the
646 * other inodes on s_io.
647 *
648 * We'll have another go at writing back this inode
649 * when we completed a full scan of b_io.
650 */
651 spin_unlock(&inode->i_lock);
652 requeue_io(inode, wb);
653 trace_writeback_sb_inodes_requeue(inode);
654 continue;
655 }
Jan Karaf0d07b72012-05-03 14:47:59 +0200656 spin_unlock(&wb->list_lock);
657
Jan Kara4f8ad652012-05-03 14:48:00 +0200658 /*
659 * We already requeued the inode if it had I_SYNC set and we
660 * are doing WB_SYNC_NONE writeback. So this catches only the
661 * WB_SYNC_ALL case.
662 */
Jan Kara169ebd92012-05-03 14:48:03 +0200663 if (inode->i_state & I_SYNC) {
664 /* Wait for I_SYNC. This function drops i_lock... */
665 inode_sleep_on_writeback(inode);
666 /* Inode may be gone, start again */
Jan Karaead188f2012-06-08 17:07:36 +0200667 spin_lock(&wb->list_lock);
Jan Kara169ebd92012-05-03 14:48:03 +0200668 continue;
669 }
Jan Kara4f8ad652012-05-03 14:48:00 +0200670 inode->i_state |= I_SYNC;
671 spin_unlock(&inode->i_lock);
Jan Kara169ebd92012-05-03 14:48:03 +0200672
Wu Fengguang1a12d8b2010-08-29 13:28:09 -0600673 write_chunk = writeback_chunk_size(wb->bdi, work);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600674 wbc.nr_to_write = write_chunk;
675 wbc.pages_skipped = 0;
Dave Chinner250df6e2011-03-22 22:23:36 +1100676
Jan Kara169ebd92012-05-03 14:48:03 +0200677 /*
678 * We use I_SYNC to pin the inode in memory. While it is set
679 * evict_inode() will wait so the inode cannot be freed.
680 */
Jan Kara4f8ad652012-05-03 14:48:00 +0200681 __writeback_single_inode(inode, wb, &wbc);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600682
683 work->nr_pages -= write_chunk - wbc.nr_to_write;
684 wrote += write_chunk - wbc.nr_to_write;
Jan Kara4f8ad652012-05-03 14:48:00 +0200685 spin_lock(&wb->list_lock);
686 spin_lock(&inode->i_lock);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600687 if (!(inode->i_state & I_DIRTY))
688 wrote++;
Jan Kara4f8ad652012-05-03 14:48:00 +0200689 requeue_inode(inode, wb, &wbc);
690 inode_sync_complete(inode);
Dave Chinner0f1b1fd2011-03-22 22:23:43 +1100691 spin_unlock(&inode->i_lock);
Jan Kara169ebd92012-05-03 14:48:03 +0200692 cond_resched_lock(&wb->list_lock);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600693 /*
694 * bail out to wb_writeback() often enough to check
695 * background threshold and other termination conditions.
696 */
697 if (wrote) {
698 if (time_is_before_jiffies(start_time + HZ / 10UL))
699 break;
700 if (work->nr_pages <= 0)
701 break;
Fengguang Wu8bc3be22008-02-04 22:29:36 -0800702 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703 }
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600704 return wrote;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800705}
Nick Piggin38f21972009-01-06 14:40:25 -0800706
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600707static long __writeback_inodes_wb(struct bdi_writeback *wb,
708 struct wb_writeback_work *work)
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800709{
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600710 unsigned long start_time = jiffies;
711 long wrote = 0;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800712
713 while (!list_empty(&wb->b_io)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100714 struct inode *inode = wb_inode(wb->b_io.prev);
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800715 struct super_block *sb = inode->i_sb;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800716
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000717 if (!grab_super_passive(sb)) {
Wu Fengguang0e995812011-07-29 22:14:35 -0600718 /*
719 * grab_super_passive() may fail consistently due to
720 * s_umount being grabbed by someone else. Don't use
721 * requeue_io() to avoid busy retrying the inode/sb.
722 */
723 redirty_tail(inode, wb);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200724 continue;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800725 }
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600726 wrote += writeback_sb_inodes(sb, wb, work);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200727 drop_super(sb);
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800728
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600729 /* refer to the same tests at the end of writeback_sb_inodes */
730 if (wrote) {
731 if (time_is_before_jiffies(start_time + HZ / 10UL))
732 break;
733 if (work->nr_pages <= 0)
734 break;
735 }
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800736 }
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200737 /* Leave any unwritten inodes on b_io */
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600738 return wrote;
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200739}
740
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600741long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages,
742 enum wb_reason reason)
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200743{
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600744 struct wb_writeback_work work = {
745 .nr_pages = nr_pages,
746 .sync_mode = WB_SYNC_NONE,
747 .range_cyclic = 1,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600748 .reason = reason,
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600749 };
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200750
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600751 spin_lock(&wb->list_lock);
Wu Fengguang424b3512010-07-21 20:11:53 -0600752 if (list_empty(&wb->b_io))
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600753 queue_io(wb, &work);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600754 __writeback_inodes_wb(wb, &work);
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600755 spin_unlock(&wb->list_lock);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200756
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600757 return nr_pages - work.nr_pages;
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200758}
759
Wu Fengguangb00949a2010-11-18 14:38:33 -0600760static bool over_bground_thresh(struct backing_dev_info *bdi)
Jens Axboe03ba3782009-09-09 09:08:54 +0200761{
762 unsigned long background_thresh, dirty_thresh;
763
Wu Fengguang16c40422010-08-11 14:17:39 -0700764 global_dirty_limits(&background_thresh, &dirty_thresh);
Jens Axboe03ba3782009-09-09 09:08:54 +0200765
Wu Fengguangb00949a2010-11-18 14:38:33 -0600766 if (global_page_state(NR_FILE_DIRTY) +
767 global_page_state(NR_UNSTABLE_NFS) > background_thresh)
768 return true;
769
770 if (bdi_stat(bdi, BDI_RECLAIMABLE) >
771 bdi_dirty_limit(bdi, background_thresh))
772 return true;
773
774 return false;
Jens Axboe03ba3782009-09-09 09:08:54 +0200775}
776
777/*
Wu Fengguange98be2d2010-08-29 11:22:30 -0600778 * Called under wb->list_lock. If there are multiple wb per bdi,
779 * only the flusher working on the first wb should do it.
780 */
781static void wb_update_bandwidth(struct bdi_writeback *wb,
782 unsigned long start_time)
783{
Wu Fengguangaf6a3112011-10-03 20:46:17 -0600784 __bdi_update_bandwidth(wb->bdi, 0, 0, 0, 0, 0, start_time);
Wu Fengguange98be2d2010-08-29 11:22:30 -0600785}
786
787/*
Jens Axboe03ba3782009-09-09 09:08:54 +0200788 * Explicit flushing or periodic writeback of "old" data.
789 *
790 * Define "old": the first time one of an inode's pages is dirtied, we mark the
791 * dirtying-time in the inode's address_space. So this periodic writeback code
792 * just walks the superblock inode list, writing back any inodes which are
793 * older than a specific point in time.
794 *
795 * Try to run once per dirty_writeback_interval. But if a writeback event
796 * takes longer than a dirty_writeback_interval interval, then leave a
797 * one-second gap.
798 *
799 * older_than_this takes precedence over nr_to_write. So we'll only write back
800 * all dirty pages if they are all attached to "old" mappings.
801 */
Jens Axboec4a77a62009-09-16 15:18:25 +0200802static long wb_writeback(struct bdi_writeback *wb,
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200803 struct wb_writeback_work *work)
Jens Axboe03ba3782009-09-09 09:08:54 +0200804{
Wu Fengguange98be2d2010-08-29 11:22:30 -0600805 unsigned long wb_start = jiffies;
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600806 long nr_pages = work->nr_pages;
Jens Axboe03ba3782009-09-09 09:08:54 +0200807 unsigned long oldest_jif;
Jan Karaa5989bd2009-09-16 19:22:48 +0200808 struct inode *inode;
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600809 long progress;
Jens Axboe03ba3782009-09-09 09:08:54 +0200810
Wu Fengguange185dda2011-04-23 11:26:07 -0600811 oldest_jif = jiffies;
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600812 work->older_than_this = &oldest_jif;
Jens Axboe03ba3782009-09-09 09:08:54 +0200813
Wu Fengguange8dfc302011-04-21 12:06:32 -0600814 spin_lock(&wb->list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200815 for (;;) {
816 /*
Wu Fengguangd3ddec72009-09-23 20:33:40 +0800817 * Stop writeback when nr_pages has been consumed
Jens Axboe03ba3782009-09-09 09:08:54 +0200818 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200819 if (work->nr_pages <= 0)
Jens Axboe03ba3782009-09-09 09:08:54 +0200820 break;
821
822 /*
Jan Karaaa373cf2011-01-13 15:45:47 -0800823 * Background writeout and kupdate-style writeback may
824 * run forever. Stop them if there is other work to do
825 * so that e.g. sync can proceed. They'll be restarted
826 * after the other works are all done.
827 */
828 if ((work->for_background || work->for_kupdate) &&
829 !list_empty(&wb->bdi->work_list))
830 break;
831
832 /*
Wu Fengguangd3ddec72009-09-23 20:33:40 +0800833 * For background writeout, stop when we are below the
834 * background dirty threshold
Jens Axboe03ba3782009-09-09 09:08:54 +0200835 */
Wu Fengguangb00949a2010-11-18 14:38:33 -0600836 if (work->for_background && !over_bground_thresh(wb->bdi))
Jens Axboe03ba3782009-09-09 09:08:54 +0200837 break;
838
Jan Kara1bc36b62011-10-19 11:44:41 +0200839 /*
840 * Kupdate and background works are special and we want to
841 * include all inodes that need writing. Livelock avoidance is
842 * handled by these works yielding to any other work so we are
843 * safe.
844 */
Wu Fengguangba9aa832010-07-21 20:32:30 -0600845 if (work->for_kupdate) {
846 oldest_jif = jiffies -
847 msecs_to_jiffies(dirty_expire_interval * 10);
Jan Kara1bc36b62011-10-19 11:44:41 +0200848 } else if (work->for_background)
849 oldest_jif = jiffies;
Dave Chinner028c2dd2010-07-07 13:24:07 +1000850
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600851 trace_writeback_start(wb->bdi, work);
Wu Fengguange8dfc302011-04-21 12:06:32 -0600852 if (list_empty(&wb->b_io))
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600853 queue_io(wb, work);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200854 if (work->sb)
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600855 progress = writeback_sb_inodes(work->sb, wb, work);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200856 else
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600857 progress = __writeback_inodes_wb(wb, work);
858 trace_writeback_written(wb->bdi, work);
Dave Chinner028c2dd2010-07-07 13:24:07 +1000859
Wu Fengguange98be2d2010-08-29 11:22:30 -0600860 wb_update_bandwidth(wb, wb_start);
Jens Axboe03ba3782009-09-09 09:08:54 +0200861
862 /*
Jens Axboe71fd05a2009-09-23 19:32:26 +0200863 * Did we write something? Try for more
Wu Fengguange6fb6da2010-07-22 10:23:44 -0600864 *
865 * Dirty inodes are moved to b_io for writeback in batches.
866 * The completion of the current batch does not necessarily
867 * mean the overall work is done. So we keep looping as long
868 * as made some progress on cleaning pages or inodes.
Jens Axboe71fd05a2009-09-23 19:32:26 +0200869 */
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600870 if (progress)
Jens Axboe71fd05a2009-09-23 19:32:26 +0200871 continue;
872 /*
Wu Fengguange6fb6da2010-07-22 10:23:44 -0600873 * No more inodes for IO, bail
Jens Axboe71fd05a2009-09-23 19:32:26 +0200874 */
Wu Fengguangb7a24412010-07-21 22:19:51 -0600875 if (list_empty(&wb->b_more_io))
Jens Axboe03ba3782009-09-09 09:08:54 +0200876 break;
877 /*
Jens Axboe8010c3b2009-09-15 20:04:57 +0200878 * Nothing written. Wait for some inode to
879 * become available for writeback. Otherwise
880 * we'll just busyloop.
Jens Axboe03ba3782009-09-09 09:08:54 +0200881 */
Jens Axboe03ba3782009-09-09 09:08:54 +0200882 if (!list_empty(&wb->b_more_io)) {
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600883 trace_writeback_wait(wb->bdi, work);
Jens Axboe03ba3782009-09-09 09:08:54 +0200884 inode = wb_inode(wb->b_more_io.prev);
Dave Chinner250df6e2011-03-22 22:23:36 +1100885 spin_lock(&inode->i_lock);
Jan Karaf0d07b72012-05-03 14:47:59 +0200886 spin_unlock(&wb->list_lock);
Jan Kara169ebd92012-05-03 14:48:03 +0200887 /* This function drops i_lock... */
888 inode_sleep_on_writeback(inode);
Jan Karaf0d07b72012-05-03 14:47:59 +0200889 spin_lock(&wb->list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200890 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200891 }
Wu Fengguange8dfc302011-04-21 12:06:32 -0600892 spin_unlock(&wb->list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200893
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600894 return nr_pages - work->nr_pages;
Jens Axboe03ba3782009-09-09 09:08:54 +0200895}
896
897/*
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200898 * Return the next wb_writeback_work struct that hasn't been processed yet.
Jens Axboe03ba3782009-09-09 09:08:54 +0200899 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200900static struct wb_writeback_work *
Minchan Kim08852b62010-08-03 12:51:16 +0200901get_next_work_item(struct backing_dev_info *bdi)
Jens Axboe03ba3782009-09-09 09:08:54 +0200902{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200903 struct wb_writeback_work *work = NULL;
Jens Axboe03ba3782009-09-09 09:08:54 +0200904
Artem Bityutskiy64677162010-07-25 14:29:22 +0300905 spin_lock_bh(&bdi->wb_lock);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200906 if (!list_empty(&bdi->work_list)) {
907 work = list_entry(bdi->work_list.next,
908 struct wb_writeback_work, list);
909 list_del_init(&work->list);
Jens Axboe03ba3782009-09-09 09:08:54 +0200910 }
Artem Bityutskiy64677162010-07-25 14:29:22 +0300911 spin_unlock_bh(&bdi->wb_lock);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200912 return work;
Jens Axboe03ba3782009-09-09 09:08:54 +0200913}
914
Linus Torvaldscdf01dd2010-10-30 08:55:52 -0700915/*
916 * Add in the number of potentially dirty inodes, because each inode
917 * write can dirty pagecache in the underlying blockdev.
918 */
919static unsigned long get_nr_dirty_pages(void)
920{
921 return global_page_state(NR_FILE_DIRTY) +
922 global_page_state(NR_UNSTABLE_NFS) +
923 get_nr_dirty_inodes();
924}
925
Jan Kara65850272011-01-13 15:45:44 -0800926static long wb_check_background_flush(struct bdi_writeback *wb)
927{
Wu Fengguangb00949a2010-11-18 14:38:33 -0600928 if (over_bground_thresh(wb->bdi)) {
Jan Kara65850272011-01-13 15:45:44 -0800929
930 struct wb_writeback_work work = {
931 .nr_pages = LONG_MAX,
932 .sync_mode = WB_SYNC_NONE,
933 .for_background = 1,
934 .range_cyclic = 1,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600935 .reason = WB_REASON_BACKGROUND,
Jan Kara65850272011-01-13 15:45:44 -0800936 };
937
938 return wb_writeback(wb, &work);
939 }
940
941 return 0;
942}
943
Jens Axboe03ba3782009-09-09 09:08:54 +0200944static long wb_check_old_data_flush(struct bdi_writeback *wb)
945{
946 unsigned long expired;
947 long nr_pages;
948
Jens Axboe69b62d02010-05-17 12:51:03 +0200949 /*
950 * When set to zero, disable periodic writeback
951 */
952 if (!dirty_writeback_interval)
953 return 0;
954
Jens Axboe03ba3782009-09-09 09:08:54 +0200955 expired = wb->last_old_flush +
956 msecs_to_jiffies(dirty_writeback_interval * 10);
957 if (time_before(jiffies, expired))
958 return 0;
959
960 wb->last_old_flush = jiffies;
Linus Torvaldscdf01dd2010-10-30 08:55:52 -0700961 nr_pages = get_nr_dirty_pages();
Jens Axboe03ba3782009-09-09 09:08:54 +0200962
Jens Axboec4a77a62009-09-16 15:18:25 +0200963 if (nr_pages) {
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200964 struct wb_writeback_work work = {
Jens Axboec4a77a62009-09-16 15:18:25 +0200965 .nr_pages = nr_pages,
966 .sync_mode = WB_SYNC_NONE,
967 .for_kupdate = 1,
968 .range_cyclic = 1,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600969 .reason = WB_REASON_PERIODIC,
Jens Axboec4a77a62009-09-16 15:18:25 +0200970 };
971
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200972 return wb_writeback(wb, &work);
Jens Axboec4a77a62009-09-16 15:18:25 +0200973 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200974
975 return 0;
976}
977
978/*
979 * Retrieve work items and do the writeback they describe
980 */
981long wb_do_writeback(struct bdi_writeback *wb, int force_wait)
982{
983 struct backing_dev_info *bdi = wb->bdi;
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200984 struct wb_writeback_work *work;
Jens Axboec4a77a62009-09-16 15:18:25 +0200985 long wrote = 0;
Jens Axboe03ba3782009-09-09 09:08:54 +0200986
Jan Kara81d73a32010-08-11 14:17:44 -0700987 set_bit(BDI_writeback_running, &wb->bdi->state);
Minchan Kim08852b62010-08-03 12:51:16 +0200988 while ((work = get_next_work_item(bdi)) != NULL) {
Jens Axboe03ba3782009-09-09 09:08:54 +0200989 /*
990 * Override sync mode, in case we must wait for completion
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200991 * because this thread is exiting now.
Jens Axboe03ba3782009-09-09 09:08:54 +0200992 */
993 if (force_wait)
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200994 work->sync_mode = WB_SYNC_ALL;
995
Dave Chinner455b2862010-07-07 13:24:06 +1000996 trace_writeback_exec(bdi, work);
997
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200998 wrote += wb_writeback(wb, work);
Jens Axboe03ba3782009-09-09 09:08:54 +0200999
1000 /*
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001001 * Notify the caller of completion if this is a synchronous
1002 * work item, otherwise just free it.
Jens Axboe03ba3782009-09-09 09:08:54 +02001003 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001004 if (work->done)
1005 complete(work->done);
1006 else
1007 kfree(work);
Jens Axboe03ba3782009-09-09 09:08:54 +02001008 }
1009
1010 /*
1011 * Check for periodic writeback, kupdated() style
1012 */
1013 wrote += wb_check_old_data_flush(wb);
Jan Kara65850272011-01-13 15:45:44 -08001014 wrote += wb_check_background_flush(wb);
Jan Kara81d73a32010-08-11 14:17:44 -07001015 clear_bit(BDI_writeback_running, &wb->bdi->state);
Jens Axboe03ba3782009-09-09 09:08:54 +02001016
1017 return wrote;
1018}
1019
1020/*
1021 * Handle writeback of dirty data for the device backed by this bdi. Also
1022 * wakes up periodically and does kupdated style flushing.
1023 */
Christoph Hellwig08243902010-06-19 23:08:22 +02001024int bdi_writeback_thread(void *data)
Jens Axboe03ba3782009-09-09 09:08:54 +02001025{
Christoph Hellwig08243902010-06-19 23:08:22 +02001026 struct bdi_writeback *wb = data;
1027 struct backing_dev_info *bdi = wb->bdi;
Jens Axboe03ba3782009-09-09 09:08:54 +02001028 long pages_written;
1029
Peter Zijlstra766f9162010-10-26 14:22:45 -07001030 current->flags |= PF_SWAPWRITE;
Christoph Hellwig08243902010-06-19 23:08:22 +02001031 set_freezable();
Artem Bityutskiyecd58402010-07-25 14:29:18 +03001032 wb->last_active = jiffies;
Christoph Hellwig08243902010-06-19 23:08:22 +02001033
1034 /*
1035 * Our parent may run at a different priority, just set us to normal
1036 */
1037 set_user_nice(current, 0);
1038
Dave Chinner455b2862010-07-07 13:24:06 +10001039 trace_writeback_thread_start(bdi);
1040
Tejun Heo8a32c442011-11-21 12:32:23 -08001041 while (!kthread_freezable_should_stop(NULL)) {
Artem Bityutskiy64677162010-07-25 14:29:22 +03001042 /*
1043 * Remove own delayed wake-up timer, since we are already awake
1044 * and we'll take care of the preriodic write-back.
1045 */
1046 del_timer(&wb->wakeup_timer);
1047
Jens Axboe03ba3782009-09-09 09:08:54 +02001048 pages_written = wb_do_writeback(wb, 0);
1049
Dave Chinner455b2862010-07-07 13:24:06 +10001050 trace_writeback_pages_written(pages_written);
Jens Axboe03ba3782009-09-09 09:08:54 +02001051
Jens Axboe03ba3782009-09-09 09:08:54 +02001052 if (pages_written)
Artem Bityutskiyecd58402010-07-25 14:29:18 +03001053 wb->last_active = jiffies;
Jens Axboe03ba3782009-09-09 09:08:54 +02001054
Artem Bityutskiy297252c2010-07-25 14:29:15 +03001055 set_current_state(TASK_INTERRUPTIBLE);
J. Bruce Fieldsb76b4012010-08-28 08:52:10 +02001056 if (!list_empty(&bdi->work_list) || kthread_should_stop()) {
Artem Bityutskiy297252c2010-07-25 14:29:15 +03001057 __set_current_state(TASK_RUNNING);
1058 continue;
Jens Axboe03ba3782009-09-09 09:08:54 +02001059 }
1060
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001061 if (wb_has_dirty_io(wb) && dirty_writeback_interval)
Artem Bityutskiyfff5b852010-07-25 14:29:20 +03001062 schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10));
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001063 else {
1064 /*
1065 * We have nothing to do, so can go sleep without any
1066 * timeout and save power. When a work is queued or
1067 * something is made dirty - we will be woken up.
1068 */
Artem Bityutskiy297252c2010-07-25 14:29:15 +03001069 schedule();
Jens Axboef9eadbb2010-05-18 14:31:45 +02001070 }
Jens Axboe03ba3782009-09-09 09:08:54 +02001071 }
1072
Artem Bityutskiyfff5b852010-07-25 14:29:20 +03001073 /* Flush any work that raced with us exiting */
Christoph Hellwig08243902010-06-19 23:08:22 +02001074 if (!list_empty(&bdi->work_list))
1075 wb_do_writeback(wb, 1);
Dave Chinner455b2862010-07-07 13:24:06 +10001076
1077 trace_writeback_thread_stop(bdi);
Jens Axboe03ba3782009-09-09 09:08:54 +02001078 return 0;
1079}
1080
Christoph Hellwig08243902010-06-19 23:08:22 +02001081
Jens Axboe03ba3782009-09-09 09:08:54 +02001082/*
Jens Axboe03ba3782009-09-09 09:08:54 +02001083 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
1084 * the whole world.
1085 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001086void wakeup_flusher_threads(long nr_pages, enum wb_reason reason)
Jens Axboe03ba3782009-09-09 09:08:54 +02001087{
Christoph Hellwigb8c2f342010-06-08 18:15:07 +02001088 struct backing_dev_info *bdi;
Christoph Hellwigb8c2f342010-06-08 18:15:07 +02001089
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001090 if (!nr_pages) {
1091 nr_pages = global_page_state(NR_FILE_DIRTY) +
Jens Axboe03ba3782009-09-09 09:08:54 +02001092 global_page_state(NR_UNSTABLE_NFS);
Christoph Hellwigb8c2f342010-06-08 18:15:07 +02001093 }
1094
1095 rcu_read_lock();
1096 list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) {
1097 if (!bdi_has_dirty_io(bdi))
1098 continue;
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001099 __bdi_start_writeback(bdi, nr_pages, false, reason);
Christoph Hellwigb8c2f342010-06-08 18:15:07 +02001100 }
1101 rcu_read_unlock();
Jens Axboe03ba3782009-09-09 09:08:54 +02001102}
1103
1104static noinline void block_dump___mark_inode_dirty(struct inode *inode)
1105{
1106 if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) {
1107 struct dentry *dentry;
1108 const char *name = "?";
1109
1110 dentry = d_find_alias(inode);
1111 if (dentry) {
1112 spin_lock(&dentry->d_lock);
1113 name = (const char *) dentry->d_name.name;
1114 }
1115 printk(KERN_DEBUG
1116 "%s(%d): dirtied inode %lu (%s) on %s\n",
1117 current->comm, task_pid_nr(current), inode->i_ino,
1118 name, inode->i_sb->s_id);
1119 if (dentry) {
1120 spin_unlock(&dentry->d_lock);
1121 dput(dentry);
1122 }
1123 }
1124}
1125
1126/**
1127 * __mark_inode_dirty - internal function
1128 * @inode: inode to mark
1129 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
1130 * Mark an inode as dirty. Callers should use mark_inode_dirty or
1131 * mark_inode_dirty_sync.
1132 *
1133 * Put the inode on the super block's dirty list.
1134 *
1135 * CAREFUL! We mark it dirty unconditionally, but move it onto the
1136 * dirty list only if it is hashed or if it refers to a blockdev.
1137 * If it was not hashed, it will never be added to the dirty list
1138 * even if it is later hashed, as it will have been marked dirty already.
1139 *
1140 * In short, make sure you hash any inodes _before_ you start marking
1141 * them dirty.
1142 *
Jens Axboe03ba3782009-09-09 09:08:54 +02001143 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
1144 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
1145 * the kernel-internal blockdev inode represents the dirtying time of the
1146 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
1147 * page->mapping->host, so the page-dirtying time is recorded in the internal
1148 * blockdev inode.
1149 */
1150void __mark_inode_dirty(struct inode *inode, int flags)
1151{
1152 struct super_block *sb = inode->i_sb;
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001153 struct backing_dev_info *bdi = NULL;
Jens Axboe03ba3782009-09-09 09:08:54 +02001154
1155 /*
1156 * Don't do this for I_DIRTY_PAGES - that doesn't actually
1157 * dirty the inode itself
1158 */
1159 if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
1160 if (sb->s_op->dirty_inode)
Christoph Hellwigaa385722011-05-27 06:53:02 -04001161 sb->s_op->dirty_inode(inode, flags);
Jens Axboe03ba3782009-09-09 09:08:54 +02001162 }
1163
1164 /*
1165 * make sure that changes are seen by all cpus before we test i_state
1166 * -- mikulas
1167 */
1168 smp_mb();
1169
1170 /* avoid the locking if we can */
1171 if ((inode->i_state & flags) == flags)
1172 return;
1173
1174 if (unlikely(block_dump))
1175 block_dump___mark_inode_dirty(inode);
1176
Dave Chinner250df6e2011-03-22 22:23:36 +11001177 spin_lock(&inode->i_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001178 if ((inode->i_state & flags) != flags) {
1179 const int was_dirty = inode->i_state & I_DIRTY;
1180
1181 inode->i_state |= flags;
1182
1183 /*
1184 * If the inode is being synced, just update its dirty state.
1185 * The unlocker will place the inode on the appropriate
1186 * superblock list, based upon its state.
1187 */
1188 if (inode->i_state & I_SYNC)
Dave Chinner250df6e2011-03-22 22:23:36 +11001189 goto out_unlock_inode;
Jens Axboe03ba3782009-09-09 09:08:54 +02001190
1191 /*
1192 * Only add valid (hashed) inodes to the superblock's
1193 * dirty list. Add blockdev inodes as well.
1194 */
1195 if (!S_ISBLK(inode->i_mode)) {
Al Viro1d3382cb2010-10-23 15:19:20 -04001196 if (inode_unhashed(inode))
Dave Chinner250df6e2011-03-22 22:23:36 +11001197 goto out_unlock_inode;
Jens Axboe03ba3782009-09-09 09:08:54 +02001198 }
Al Viroa4ffdde2010-06-02 17:38:30 -04001199 if (inode->i_state & I_FREEING)
Dave Chinner250df6e2011-03-22 22:23:36 +11001200 goto out_unlock_inode;
Jens Axboe03ba3782009-09-09 09:08:54 +02001201
1202 /*
1203 * If the inode was already on b_dirty/b_io/b_more_io, don't
1204 * reposition it (that would break b_dirty time-ordering).
1205 */
1206 if (!was_dirty) {
Dave Chinnera66979a2011-03-22 22:23:41 +11001207 bool wakeup_bdi = false;
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001208 bdi = inode_to_bdi(inode);
Jens Axboe500b0672009-09-09 09:10:25 +02001209
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001210 if (bdi_cap_writeback_dirty(bdi)) {
1211 WARN(!test_bit(BDI_registered, &bdi->state),
1212 "bdi-%s not registered\n", bdi->name);
1213
1214 /*
1215 * If this is the first dirty inode for this
1216 * bdi, we have to wake-up the corresponding
1217 * bdi thread to make sure background
1218 * write-back happens later.
1219 */
1220 if (!wb_has_dirty_io(&bdi->wb))
1221 wakeup_bdi = true;
Jens Axboe500b0672009-09-09 09:10:25 +02001222 }
Jens Axboe03ba3782009-09-09 09:08:54 +02001223
Dave Chinnera66979a2011-03-22 22:23:41 +11001224 spin_unlock(&inode->i_lock);
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001225 spin_lock(&bdi->wb.list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001226 inode->dirtied_when = jiffies;
Nick Piggin7ccf19a2010-10-21 11:49:30 +11001227 list_move(&inode->i_wb_list, &bdi->wb.b_dirty);
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001228 spin_unlock(&bdi->wb.list_lock);
Dave Chinnera66979a2011-03-22 22:23:41 +11001229
1230 if (wakeup_bdi)
1231 bdi_wakeup_thread_delayed(bdi);
1232 return;
Jens Axboe03ba3782009-09-09 09:08:54 +02001233 }
1234 }
Dave Chinner250df6e2011-03-22 22:23:36 +11001235out_unlock_inode:
1236 spin_unlock(&inode->i_lock);
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001237
Jens Axboe03ba3782009-09-09 09:08:54 +02001238}
1239EXPORT_SYMBOL(__mark_inode_dirty);
1240
Jens Axboeb6e51312009-09-16 15:13:54 +02001241static void wait_sb_inodes(struct super_block *sb)
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001242{
Jens Axboe03ba3782009-09-09 09:08:54 +02001243 struct inode *inode, *old_inode = NULL;
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001244
Jens Axboe03ba3782009-09-09 09:08:54 +02001245 /*
1246 * We need to be protected against the filesystem going from
1247 * r/o to r/w or vice versa.
1248 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001249 WARN_ON(!rwsem_is_locked(&sb->s_umount));
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001250
Dave Chinner55fa6092011-03-22 22:23:40 +11001251 spin_lock(&inode_sb_list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001252
1253 /*
1254 * Data integrity sync. Must wait for all pages under writeback,
1255 * because there may have been pages dirtied before our sync
1256 * call, but which had writeout started before we write it out.
1257 * In which case, the inode may not be on the dirty list, but
1258 * we still have to wait for that writeout.
1259 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001260 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
Dave Chinner250df6e2011-03-22 22:23:36 +11001261 struct address_space *mapping = inode->i_mapping;
Jens Axboe03ba3782009-09-09 09:08:54 +02001262
Dave Chinner250df6e2011-03-22 22:23:36 +11001263 spin_lock(&inode->i_lock);
1264 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
1265 (mapping->nrpages == 0)) {
1266 spin_unlock(&inode->i_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001267 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +11001268 }
Jens Axboe03ba3782009-09-09 09:08:54 +02001269 __iget(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +11001270 spin_unlock(&inode->i_lock);
Dave Chinner55fa6092011-03-22 22:23:40 +11001271 spin_unlock(&inode_sb_list_lock);
1272
Jens Axboe03ba3782009-09-09 09:08:54 +02001273 /*
Dave Chinner55fa6092011-03-22 22:23:40 +11001274 * We hold a reference to 'inode' so it couldn't have been
1275 * removed from s_inodes list while we dropped the
1276 * inode_sb_list_lock. We cannot iput the inode now as we can
1277 * be holding the last reference and we cannot iput it under
1278 * inode_sb_list_lock. So we keep the reference and iput it
1279 * later.
Jens Axboe03ba3782009-09-09 09:08:54 +02001280 */
1281 iput(old_inode);
1282 old_inode = inode;
1283
1284 filemap_fdatawait(mapping);
1285
1286 cond_resched();
Nick Piggin38f21972009-01-06 14:40:25 -08001287
Dave Chinner55fa6092011-03-22 22:23:40 +11001288 spin_lock(&inode_sb_list_lock);
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001289 }
Dave Chinner55fa6092011-03-22 22:23:40 +11001290 spin_unlock(&inode_sb_list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001291 iput(old_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292}
1293
Jens Axboed8a85592009-09-02 12:34:32 +02001294/**
Chris Mason3259f8b2010-10-29 11:16:17 -04001295 * writeback_inodes_sb_nr - writeback dirty inodes from given super_block
Jens Axboed8a85592009-09-02 12:34:32 +02001296 * @sb: the superblock
Chris Mason3259f8b2010-10-29 11:16:17 -04001297 * @nr: the number of pages to write
Marcos Paulo de Souza786228a2011-11-23 20:56:45 +08001298 * @reason: reason why some writeback work initiated
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299 *
Jens Axboed8a85592009-09-02 12:34:32 +02001300 * Start writeback on some inodes on this super_block. No guarantees are made
1301 * on how many (if any) will be written, and this function does not wait
Chris Mason3259f8b2010-10-29 11:16:17 -04001302 * for IO completion of submitted IO.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001304void writeback_inodes_sb_nr(struct super_block *sb,
1305 unsigned long nr,
1306 enum wb_reason reason)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001308 DECLARE_COMPLETION_ONSTACK(done);
1309 struct wb_writeback_work work = {
Wu Fengguang6e6938b2010-06-06 10:38:15 -06001310 .sb = sb,
1311 .sync_mode = WB_SYNC_NONE,
1312 .tagged_writepages = 1,
1313 .done = &done,
1314 .nr_pages = nr,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001315 .reason = reason,
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001316 };
Jens Axboe0e3c9a22010-06-01 11:08:43 +02001317
Jan Kara6eedc702012-07-03 16:45:27 +02001318 if (sb->s_bdi == &noop_backing_dev_info)
1319 return;
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001320 WARN_ON(!rwsem_is_locked(&sb->s_umount));
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001321 bdi_queue_work(sb->s_bdi, &work);
1322 wait_for_completion(&done);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323}
Chris Mason3259f8b2010-10-29 11:16:17 -04001324EXPORT_SYMBOL(writeback_inodes_sb_nr);
1325
1326/**
1327 * writeback_inodes_sb - writeback dirty inodes from given super_block
1328 * @sb: the superblock
Marcos Paulo de Souza786228a2011-11-23 20:56:45 +08001329 * @reason: reason why some writeback work was initiated
Chris Mason3259f8b2010-10-29 11:16:17 -04001330 *
1331 * Start writeback on some inodes on this super_block. No guarantees are made
1332 * on how many (if any) will be written, and this function does not wait
1333 * for IO completion of submitted IO.
1334 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001335void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason)
Chris Mason3259f8b2010-10-29 11:16:17 -04001336{
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001337 return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason);
Chris Mason3259f8b2010-10-29 11:16:17 -04001338}
Jens Axboed8a85592009-09-02 12:34:32 +02001339EXPORT_SYMBOL(writeback_inodes_sb);
1340
1341/**
Eric Sandeen17bd55d2009-12-23 07:57:07 -05001342 * writeback_inodes_sb_if_idle - start writeback if none underway
1343 * @sb: the superblock
Marcos Paulo de Souza786228a2011-11-23 20:56:45 +08001344 * @reason: reason why some writeback work was initiated
Eric Sandeen17bd55d2009-12-23 07:57:07 -05001345 *
1346 * Invoke writeback_inodes_sb if no writeback is currently underway.
1347 * Returns 1 if writeback was started, 0 if not.
1348 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001349int writeback_inodes_sb_if_idle(struct super_block *sb, enum wb_reason reason)
Eric Sandeen17bd55d2009-12-23 07:57:07 -05001350{
1351 if (!writeback_in_progress(sb->s_bdi)) {
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001352 down_read(&sb->s_umount);
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001353 writeback_inodes_sb(sb, reason);
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001354 up_read(&sb->s_umount);
Eric Sandeen17bd55d2009-12-23 07:57:07 -05001355 return 1;
1356 } else
1357 return 0;
1358}
1359EXPORT_SYMBOL(writeback_inodes_sb_if_idle);
1360
1361/**
Fengguang Wuc097b2c2012-03-05 15:08:06 -08001362 * writeback_inodes_sb_nr_if_idle - start writeback if none underway
Chris Mason3259f8b2010-10-29 11:16:17 -04001363 * @sb: the superblock
1364 * @nr: the number of pages to write
Marcos Paulo de Souza786228a2011-11-23 20:56:45 +08001365 * @reason: reason why some writeback work was initiated
Chris Mason3259f8b2010-10-29 11:16:17 -04001366 *
1367 * Invoke writeback_inodes_sb if no writeback is currently underway.
1368 * Returns 1 if writeback was started, 0 if not.
1369 */
1370int writeback_inodes_sb_nr_if_idle(struct super_block *sb,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001371 unsigned long nr,
1372 enum wb_reason reason)
Chris Mason3259f8b2010-10-29 11:16:17 -04001373{
1374 if (!writeback_in_progress(sb->s_bdi)) {
1375 down_read(&sb->s_umount);
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001376 writeback_inodes_sb_nr(sb, nr, reason);
Chris Mason3259f8b2010-10-29 11:16:17 -04001377 up_read(&sb->s_umount);
1378 return 1;
1379 } else
1380 return 0;
1381}
1382EXPORT_SYMBOL(writeback_inodes_sb_nr_if_idle);
1383
1384/**
Jens Axboed8a85592009-09-02 12:34:32 +02001385 * sync_inodes_sb - sync sb inode pages
1386 * @sb: the superblock
1387 *
1388 * This function writes and waits on any dirty inode belonging to this
Stefan Hajnoczicb9ef8d2011-01-13 15:47:26 -08001389 * super_block.
Jens Axboed8a85592009-09-02 12:34:32 +02001390 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001391void sync_inodes_sb(struct super_block *sb)
Jens Axboed8a85592009-09-02 12:34:32 +02001392{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001393 DECLARE_COMPLETION_ONSTACK(done);
1394 struct wb_writeback_work work = {
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001395 .sb = sb,
1396 .sync_mode = WB_SYNC_ALL,
1397 .nr_pages = LONG_MAX,
1398 .range_cyclic = 0,
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001399 .done = &done,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001400 .reason = WB_REASON_SYNC,
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001401 };
1402
Jan Kara6eedc702012-07-03 16:45:27 +02001403 /* Nothing to do? */
1404 if (sb->s_bdi == &noop_backing_dev_info)
1405 return;
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001406 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1407
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001408 bdi_queue_work(sb->s_bdi, &work);
1409 wait_for_completion(&done);
1410
Jens Axboeb6e51312009-09-16 15:13:54 +02001411 wait_sb_inodes(sb);
Jens Axboed8a85592009-09-02 12:34:32 +02001412}
1413EXPORT_SYMBOL(sync_inodes_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415/**
Andrea Arcangeli7f04c262005-10-30 15:03:05 -08001416 * write_inode_now - write an inode to disk
1417 * @inode: inode to write to disk
1418 * @sync: whether the write should be synchronous or not
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419 *
Andrea Arcangeli7f04c262005-10-30 15:03:05 -08001420 * This function commits an inode to disk immediately if it is dirty. This is
1421 * primarily needed by knfsd.
1422 *
1423 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425int write_inode_now(struct inode *inode, int sync)
1426{
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001427 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428 struct writeback_control wbc = {
1429 .nr_to_write = LONG_MAX,
Mike Galbraith18914b12008-02-08 04:20:23 -08001430 .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -07001431 .range_start = 0,
1432 .range_end = LLONG_MAX,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433 };
1434
1435 if (!mapping_cap_writeback_dirty(inode->i_mapping))
Andrew Morton49364ce2005-11-07 00:59:15 -08001436 wbc.nr_to_write = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437
1438 might_sleep();
Jan Kara4f8ad652012-05-03 14:48:00 +02001439 return writeback_single_inode(inode, wb, &wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440}
1441EXPORT_SYMBOL(write_inode_now);
1442
1443/**
1444 * sync_inode - write an inode and its pages to disk.
1445 * @inode: the inode to sync
1446 * @wbc: controls the writeback mode
1447 *
1448 * sync_inode() will write an inode and its pages to disk. It will also
1449 * correctly update the inode on its superblock's dirty inode lists and will
1450 * update inode->i_state.
1451 *
1452 * The caller must have a ref on the inode.
1453 */
1454int sync_inode(struct inode *inode, struct writeback_control *wbc)
1455{
Jan Kara4f8ad652012-05-03 14:48:00 +02001456 return writeback_single_inode(inode, &inode_to_bdi(inode)->wb, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457}
1458EXPORT_SYMBOL(sync_inode);
Christoph Hellwigc37650162010-10-06 10:48:20 +02001459
1460/**
Andrew Mortonc691b9d2011-01-13 15:45:48 -08001461 * sync_inode_metadata - write an inode to disk
Christoph Hellwigc37650162010-10-06 10:48:20 +02001462 * @inode: the inode to sync
1463 * @wait: wait for I/O to complete.
1464 *
Andrew Mortonc691b9d2011-01-13 15:45:48 -08001465 * Write an inode to disk and adjust its dirty state after completion.
Christoph Hellwigc37650162010-10-06 10:48:20 +02001466 *
1467 * Note: only writes the actual inode, no associated data or other metadata.
1468 */
1469int sync_inode_metadata(struct inode *inode, int wait)
1470{
1471 struct writeback_control wbc = {
1472 .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE,
1473 .nr_to_write = 0, /* metadata-only */
1474 };
1475
1476 return sync_inode(inode, &wbc);
1477}
1478EXPORT_SYMBOL(sync_inode_metadata);