blob: 30f6f27d5a59a51a4c89334b457b227318e860d7 [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/writeback.h>
26#include <linux/blkdev.h>
27#include <linux/backing-dev.h>
Dave Chinner455b2862010-07-07 13:24:06 +100028#include <linux/tracepoint.h>
David Howells07f3f052006-09-30 20:52:18 +020029#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070030
Jens Axboed0bceac2009-05-18 08:20:32 +020031/*
Wu Fengguangbc31b862012-01-07 20:41:55 -060032 * 4MB minimal write chunk size
33 */
34#define MIN_WRITEBACK_PAGES (4096UL >> (PAGE_CACHE_SHIFT - 10))
35
36/*
Jens Axboec4a77a62009-09-16 15:18:25 +020037 * Passed into wb_writeback(), essentially a subset of writeback_control
38 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +020039struct wb_writeback_work {
Jens Axboec4a77a62009-09-16 15:18:25 +020040 long nr_pages;
41 struct super_block *sb;
Wu Fengguangd46db3d2011-05-04 19:54:37 -060042 unsigned long *older_than_this;
Jens Axboec4a77a62009-09-16 15:18:25 +020043 enum writeback_sync_modes sync_mode;
Wu Fengguang6e6938b2010-06-06 10:38:15 -060044 unsigned int tagged_writepages:1;
H Hartley Sweeten52957fe2010-04-01 20:36:30 -050045 unsigned int for_kupdate:1;
46 unsigned int range_cyclic:1;
47 unsigned int for_background:1;
Dave Chinner7747bd42013-07-02 22:38:35 +100048 unsigned int for_sync:1; /* sync(2) WB_SYNC_ALL writeback */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -060049 enum wb_reason reason; /* why was writeback initiated? */
Jens Axboec4a77a62009-09-16 15:18:25 +020050
Jens Axboe8010c3b2009-09-15 20:04:57 +020051 struct list_head list; /* pending work list */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +020052 struct completion *done; /* set if the caller waits */
Jens Axboe03ba3782009-09-09 09:08:54 +020053};
54
Adrian Bunkf11b00f2008-04-29 00:58:56 -070055/**
56 * writeback_in_progress - determine whether there is writeback in progress
57 * @bdi: the device's backing_dev_info structure.
58 *
Jens Axboe03ba3782009-09-09 09:08:54 +020059 * Determine whether there is writeback waiting to be handled against a
60 * backing device.
Adrian Bunkf11b00f2008-04-29 00:58:56 -070061 */
62int writeback_in_progress(struct backing_dev_info *bdi)
63{
Jan Kara81d73a32010-08-11 14:17:44 -070064 return test_bit(BDI_writeback_running, &bdi->state);
Adrian Bunkf11b00f2008-04-29 00:58:56 -070065}
Theodore Ts'o00d4e732012-09-19 22:42:36 -040066EXPORT_SYMBOL(writeback_in_progress);
Adrian Bunkf11b00f2008-04-29 00:58:56 -070067
Jan Kara692ebd12010-09-21 11:51:01 +020068static inline struct backing_dev_info *inode_to_bdi(struct inode *inode)
69{
70 struct super_block *sb = inode->i_sb;
Jan Kara692ebd12010-09-21 11:51:01 +020071
Christoph Hellwigaaead252010-10-04 14:25:33 +020072 if (strcmp(sb->s_type->name, "bdev") == 0)
73 return inode->i_mapping->backing_dev_info;
74
75 return sb->s_bdi;
Jan Kara692ebd12010-09-21 11:51:01 +020076}
77
Nick Piggin7ccf19a2010-10-21 11:49:30 +110078static inline struct inode *wb_inode(struct list_head *head)
79{
80 return list_entry(head, struct inode, i_wb_list);
81}
82
Wu Fengguang15eb77a2012-01-17 11:18:56 -060083/*
84 * Include the creation of the trace points after defining the
85 * wb_writeback_work structure and inline functions so that the definition
86 * remains local to this file.
87 */
88#define CREATE_TRACE_POINTS
89#include <trace/events/writeback.h>
90
Jan Kara65850272011-01-13 15:45:44 -080091static void bdi_queue_work(struct backing_dev_info *bdi,
92 struct wb_writeback_work *work)
93{
94 trace_writeback_queue(bdi, work);
95
96 spin_lock_bh(&bdi->wb_lock);
97 list_add_tail(&work->list, &bdi->work_list);
Artem Bityutskiy64677162010-07-25 14:29:22 +030098 spin_unlock_bh(&bdi->wb_lock);
Tejun Heo839a8e82013-04-01 19:08:06 -070099
100 mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101}
102
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200103static void
104__bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600105 bool range_cyclic, enum wb_reason reason)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200107 struct wb_writeback_work *work;
Jens Axboe03ba3782009-09-09 09:08:54 +0200108
Jens Axboebcddc3f2009-09-13 20:07:36 +0200109 /*
110 * This is WB_SYNC_NONE writeback, so if allocation fails just
111 * wakeup the thread for old dirty data writeback
112 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200113 work = kzalloc(sizeof(*work), GFP_ATOMIC);
114 if (!work) {
Tejun Heo839a8e82013-04-01 19:08:06 -0700115 trace_writeback_nowork(bdi);
116 mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200117 return;
Jens Axboebcddc3f2009-09-13 20:07:36 +0200118 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200119
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200120 work->sync_mode = WB_SYNC_NONE;
121 work->nr_pages = nr_pages;
122 work->range_cyclic = range_cyclic;
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600123 work->reason = reason;
Christoph Hellwigf0fad8a2009-09-11 09:47:56 +0200124
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200125 bdi_queue_work(bdi, work);
Jens Axboeb6e51312009-09-16 15:13:54 +0200126}
127
128/**
129 * bdi_start_writeback - start writeback
130 * @bdi: the backing device to write from
131 * @nr_pages: the number of pages to write
Marcos Paulo de Souza786228a2011-11-23 20:56:45 +0800132 * @reason: reason why some writeback work was initiated
Jens Axboeb6e51312009-09-16 15:13:54 +0200133 *
134 * Description:
135 * This does WB_SYNC_NONE opportunistic writeback. The IO is only
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300136 * started when this function returns, we make no guarantees on
Jens Axboe0e3c9a22010-06-01 11:08:43 +0200137 * completion. Caller need not hold sb s_umount semaphore.
Jens Axboeb6e51312009-09-16 15:13:54 +0200138 *
139 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600140void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages,
141 enum wb_reason reason)
Jens Axboeb6e51312009-09-16 15:13:54 +0200142{
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600143 __bdi_start_writeback(bdi, nr_pages, true, reason);
Christoph Hellwigc5444192010-06-08 18:15:15 +0200144}
Wu Fengguangd3ddec72009-09-23 20:33:40 +0800145
Christoph Hellwigc5444192010-06-08 18:15:15 +0200146/**
147 * bdi_start_background_writeback - start background writeback
148 * @bdi: the backing device to write from
149 *
150 * Description:
Jan Kara65850272011-01-13 15:45:44 -0800151 * This makes sure WB_SYNC_NONE background writeback happens. When
152 * this function returns, it is only guaranteed that for given BDI
153 * some IO is happening if we are over background dirty threshold.
154 * Caller need not hold sb s_umount semaphore.
Christoph Hellwigc5444192010-06-08 18:15:15 +0200155 */
156void bdi_start_background_writeback(struct backing_dev_info *bdi)
157{
Jan Kara65850272011-01-13 15:45:44 -0800158 /*
159 * We just wake up the flusher thread. It will perform background
160 * writeback as soon as there is no other work to do.
161 */
Wu Fengguang71927e82011-01-13 15:45:46 -0800162 trace_writeback_wake_background(bdi);
Tejun Heo839a8e82013-04-01 19:08:06 -0700163 mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164}
165
166/*
Dave Chinnera66979a2011-03-22 22:23:41 +1100167 * Remove the inode from the writeback list it is on.
168 */
169void inode_wb_list_del(struct inode *inode)
170{
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600171 struct backing_dev_info *bdi = inode_to_bdi(inode);
Dave Chinnera66979a2011-03-22 22:23:41 +1100172
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600173 spin_lock(&bdi->wb.list_lock);
174 list_del_init(&inode->i_wb_list);
175 spin_unlock(&bdi->wb.list_lock);
176}
Dave Chinnera66979a2011-03-22 22:23:41 +1100177
178/*
Andrew Morton6610a0b2007-10-16 23:30:32 -0700179 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
180 * furthest end of its superblock's dirty-inode list.
181 *
182 * Before stamping the inode's ->dirtied_when, we check to see whether it is
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200183 * already the most-recently-dirtied inode on the b_dirty list. If that is
Andrew Morton6610a0b2007-10-16 23:30:32 -0700184 * the case then the inode must have been redirtied while it was being written
185 * out and we don't reset its dirtied_when.
186 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600187static void redirty_tail(struct inode *inode, struct bdi_writeback *wb)
Andrew Morton6610a0b2007-10-16 23:30:32 -0700188{
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600189 assert_spin_locked(&wb->list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200190 if (!list_empty(&wb->b_dirty)) {
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200191 struct inode *tail;
Andrew Morton6610a0b2007-10-16 23:30:32 -0700192
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100193 tail = wb_inode(wb->b_dirty.next);
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200194 if (time_before(inode->dirtied_when, tail->dirtied_when))
Andrew Morton6610a0b2007-10-16 23:30:32 -0700195 inode->dirtied_when = jiffies;
196 }
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100197 list_move(&inode->i_wb_list, &wb->b_dirty);
Andrew Morton6610a0b2007-10-16 23:30:32 -0700198}
199
200/*
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200201 * requeue inode for re-scanning after bdi->b_io list is exhausted.
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700202 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600203static void requeue_io(struct inode *inode, struct bdi_writeback *wb)
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700204{
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600205 assert_spin_locked(&wb->list_lock);
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100206 list_move(&inode->i_wb_list, &wb->b_more_io);
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700207}
208
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700209static void inode_sync_complete(struct inode *inode)
210{
Jan Kara365b94ae2012-05-03 14:47:55 +0200211 inode->i_state &= ~I_SYNC;
Jan Kara4eff96d2012-11-26 16:29:51 -0800212 /* If inode is clean an unused, put it into LRU now... */
213 inode_add_lru(inode);
Jan Kara365b94ae2012-05-03 14:47:55 +0200214 /* Waiters must see I_SYNC cleared before being woken up */
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700215 smp_mb();
216 wake_up_bit(&inode->i_state, __I_SYNC);
217}
218
Jeff Laytond2caa3c52009-04-02 16:56:37 -0700219static bool inode_dirtied_after(struct inode *inode, unsigned long t)
220{
221 bool ret = time_after(inode->dirtied_when, t);
222#ifndef CONFIG_64BIT
223 /*
224 * For inodes being constantly redirtied, dirtied_when can get stuck.
225 * It _appears_ to be in the future, but is actually in distant past.
226 * This test is necessary to prevent such wrapped-around relative times
Jens Axboe5b0830c2009-09-23 19:37:09 +0200227 * from permanently stopping the whole bdi writeback.
Jeff Laytond2caa3c52009-04-02 16:56:37 -0700228 */
229 ret = ret && time_before_eq(inode->dirtied_when, jiffies);
230#endif
231 return ret;
232}
233
Andrew Mortonc986d1e2007-10-16 23:30:34 -0700234/*
Wang Sheng-Hui0e2f2b22012-09-11 08:28:18 +0800235 * Move expired (dirtied before work->older_than_this) dirty inodes from
Jan Kara697e6fe2012-03-09 07:26:22 -0800236 * @delaying_queue to @dispatch_queue.
Fengguang Wu2c136572007-10-16 23:30:39 -0700237 */
Wu Fengguange84d0a42011-04-23 12:27:27 -0600238static int move_expired_inodes(struct list_head *delaying_queue,
Fengguang Wu2c136572007-10-16 23:30:39 -0700239 struct list_head *dispatch_queue,
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600240 struct wb_writeback_work *work)
Fengguang Wu2c136572007-10-16 23:30:39 -0700241{
Shaohua Li5c034492009-09-24 14:42:33 +0200242 LIST_HEAD(tmp);
243 struct list_head *pos, *node;
Jens Axboecf137302009-09-24 15:12:57 +0200244 struct super_block *sb = NULL;
Shaohua Li5c034492009-09-24 14:42:33 +0200245 struct inode *inode;
Jens Axboecf137302009-09-24 15:12:57 +0200246 int do_sb_sort = 0;
Wu Fengguange84d0a42011-04-23 12:27:27 -0600247 int moved = 0;
Shaohua Li5c034492009-09-24 14:42:33 +0200248
Fengguang Wu2c136572007-10-16 23:30:39 -0700249 while (!list_empty(delaying_queue)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100250 inode = wb_inode(delaying_queue->prev);
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600251 if (work->older_than_this &&
252 inode_dirtied_after(inode, *work->older_than_this))
Fengguang Wu2c136572007-10-16 23:30:39 -0700253 break;
Jens Axboecf137302009-09-24 15:12:57 +0200254 if (sb && sb != inode->i_sb)
255 do_sb_sort = 1;
256 sb = inode->i_sb;
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100257 list_move(&inode->i_wb_list, &tmp);
Wu Fengguange84d0a42011-04-23 12:27:27 -0600258 moved++;
Shaohua Li5c034492009-09-24 14:42:33 +0200259 }
260
Jens Axboecf137302009-09-24 15:12:57 +0200261 /* just one sb in list, splice to dispatch_queue and we're done */
262 if (!do_sb_sort) {
263 list_splice(&tmp, dispatch_queue);
Wu Fengguange84d0a42011-04-23 12:27:27 -0600264 goto out;
Jens Axboecf137302009-09-24 15:12:57 +0200265 }
266
Shaohua Li5c034492009-09-24 14:42:33 +0200267 /* Move inodes from one superblock together */
268 while (!list_empty(&tmp)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100269 sb = wb_inode(tmp.prev)->i_sb;
Shaohua Li5c034492009-09-24 14:42:33 +0200270 list_for_each_prev_safe(pos, node, &tmp) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100271 inode = wb_inode(pos);
Shaohua Li5c034492009-09-24 14:42:33 +0200272 if (inode->i_sb == sb)
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100273 list_move(&inode->i_wb_list, dispatch_queue);
Shaohua Li5c034492009-09-24 14:42:33 +0200274 }
Fengguang Wu2c136572007-10-16 23:30:39 -0700275 }
Wu Fengguange84d0a42011-04-23 12:27:27 -0600276out:
277 return moved;
Fengguang Wu2c136572007-10-16 23:30:39 -0700278}
279
280/*
281 * Queue all expired dirty inodes for io, eldest first.
Wu Fengguang4ea879b2010-08-11 14:17:42 -0700282 * Before
283 * newly dirtied b_dirty b_io b_more_io
284 * =============> gf edc BA
285 * After
286 * newly dirtied b_dirty b_io b_more_io
287 * =============> g fBAedc
288 * |
289 * +--> dequeue for IO
Fengguang Wu2c136572007-10-16 23:30:39 -0700290 */
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600291static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work)
Fengguang Wu2c136572007-10-16 23:30:39 -0700292{
Wu Fengguange84d0a42011-04-23 12:27:27 -0600293 int moved;
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600294 assert_spin_locked(&wb->list_lock);
Wu Fengguang4ea879b2010-08-11 14:17:42 -0700295 list_splice_init(&wb->b_more_io, &wb->b_io);
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600296 moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, work);
297 trace_writeback_queue_io(wb, work, moved);
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200298}
299
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100300static int write_inode(struct inode *inode, struct writeback_control *wbc)
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200301{
Tejun Heo9fb0a7d2013-01-11 13:06:37 -0800302 int ret;
303
304 if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) {
305 trace_writeback_write_inode_start(inode, wbc);
306 ret = inode->i_sb->s_op->write_inode(inode, wbc);
307 trace_writeback_write_inode(inode, wbc);
308 return ret;
309 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200310 return 0;
Fengguang Wu2c136572007-10-16 23:30:39 -0700311}
312
313/*
Jan Kara169ebd92012-05-03 14:48:03 +0200314 * Wait for writeback on an inode to complete. Called with i_lock held.
315 * Caller must make sure inode cannot go away when we drop i_lock.
Christoph Hellwig01c03192009-06-08 13:35:40 +0200316 */
Jan Kara169ebd92012-05-03 14:48:03 +0200317static void __inode_wait_for_writeback(struct inode *inode)
318 __releases(inode->i_lock)
319 __acquires(inode->i_lock)
Christoph Hellwig01c03192009-06-08 13:35:40 +0200320{
321 DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC);
322 wait_queue_head_t *wqh;
323
324 wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
Dave Chinner250df6e2011-03-22 22:23:36 +1100325 while (inode->i_state & I_SYNC) {
326 spin_unlock(&inode->i_lock);
Christoph Hellwig01c03192009-06-08 13:35:40 +0200327 __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE);
Dave Chinner250df6e2011-03-22 22:23:36 +1100328 spin_lock(&inode->i_lock);
Richard Kennedy58a9d3d82010-05-24 14:32:38 -0700329 }
Christoph Hellwig01c03192009-06-08 13:35:40 +0200330}
331
332/*
Jan Kara169ebd92012-05-03 14:48:03 +0200333 * Wait for writeback on an inode to complete. Caller must have inode pinned.
334 */
335void inode_wait_for_writeback(struct inode *inode)
336{
337 spin_lock(&inode->i_lock);
338 __inode_wait_for_writeback(inode);
339 spin_unlock(&inode->i_lock);
340}
341
342/*
343 * Sleep until I_SYNC is cleared. This function must be called with i_lock
344 * held and drops it. It is aimed for callers not holding any inode reference
345 * so once i_lock is dropped, inode can go away.
346 */
347static void inode_sleep_on_writeback(struct inode *inode)
348 __releases(inode->i_lock)
349{
350 DEFINE_WAIT(wait);
351 wait_queue_head_t *wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
352 int sleep;
353
354 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
355 sleep = inode->i_state & I_SYNC;
356 spin_unlock(&inode->i_lock);
357 if (sleep)
358 schedule();
359 finish_wait(wqh, &wait);
360}
361
362/*
Jan Karaccb26b52012-05-03 14:47:58 +0200363 * Find proper writeback list for the inode depending on its current state and
364 * possibly also change of its state while we were doing writeback. Here we
365 * handle things such as livelock prevention or fairness of writeback among
366 * inodes. This function can be called only by flusher thread - noone else
367 * processes all inodes in writeback lists and requeueing inodes behind flusher
368 * thread's back can have unexpected consequences.
369 */
370static void requeue_inode(struct inode *inode, struct bdi_writeback *wb,
371 struct writeback_control *wbc)
372{
373 if (inode->i_state & I_FREEING)
374 return;
375
376 /*
377 * Sync livelock prevention. Each inode is tagged and synced in one
378 * shot. If still dirty, it will be redirty_tail()'ed below. Update
379 * the dirty time to prevent enqueue and sync it again.
380 */
381 if ((inode->i_state & I_DIRTY) &&
382 (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages))
383 inode->dirtied_when = jiffies;
384
Jan Kara4f8ad652012-05-03 14:48:00 +0200385 if (wbc->pages_skipped) {
386 /*
387 * writeback is not making progress due to locked
388 * buffers. Skip this inode for now.
389 */
390 redirty_tail(inode, wb);
391 return;
392 }
393
Jan Karaccb26b52012-05-03 14:47:58 +0200394 if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) {
395 /*
396 * We didn't write back all the pages. nfs_writepages()
397 * sometimes bales out without doing anything.
398 */
399 if (wbc->nr_to_write <= 0) {
400 /* Slice used up. Queue for next turn. */
401 requeue_io(inode, wb);
402 } else {
403 /*
404 * Writeback blocked by something other than
405 * congestion. Delay the inode for some time to
406 * avoid spinning on the CPU (100% iowait)
407 * retrying writeback of the dirty page/inode
408 * that cannot be performed immediately.
409 */
410 redirty_tail(inode, wb);
411 }
412 } else if (inode->i_state & I_DIRTY) {
413 /*
414 * Filesystems can dirty the inode during writeback operations,
415 * such as delayed allocation during submission or metadata
416 * updates after data IO completion.
417 */
418 redirty_tail(inode, wb);
419 } else {
420 /* The inode is clean. Remove from writeback lists. */
421 list_del_init(&inode->i_wb_list);
422 }
423}
424
425/*
Jan Kara4f8ad652012-05-03 14:48:00 +0200426 * Write out an inode and its dirty pages. Do not update the writeback list
427 * linkage. That is left to the caller. The caller is also responsible for
428 * setting I_SYNC flag and calling inode_sync_complete() to clear it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429 */
430static int
Yan Hongcd8ed2a2012-10-08 16:33:45 -0700431__writeback_single_inode(struct inode *inode, struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 struct address_space *mapping = inode->i_mapping;
Wu Fengguang251d6a42010-12-01 17:33:37 -0600434 long nr_to_write = wbc->nr_to_write;
Christoph Hellwig01c03192009-06-08 13:35:40 +0200435 unsigned dirty;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436 int ret;
437
Jan Kara4f8ad652012-05-03 14:48:00 +0200438 WARN_ON(!(inode->i_state & I_SYNC));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439
Tejun Heo9fb0a7d2013-01-11 13:06:37 -0800440 trace_writeback_single_inode_start(inode, wbc, nr_to_write);
441
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 ret = do_writepages(mapping, wbc);
443
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100444 /*
445 * Make sure to wait on the data before writing out the metadata.
446 * This is important for filesystems that modify metadata on data
Dave Chinner7747bd42013-07-02 22:38:35 +1000447 * I/O completion. We don't do it for sync(2) writeback because it has a
448 * separate, external IO completion path and ->sync_fs for guaranteeing
449 * inode metadata is written back correctly.
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100450 */
Dave Chinner7747bd42013-07-02 22:38:35 +1000451 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) {
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100452 int err = filemap_fdatawait(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 if (ret == 0)
454 ret = err;
455 }
456
Dmitry Monakhov5547e8a2010-05-07 13:35:44 +0400457 /*
458 * Some filesystems may redirty the inode during the writeback
459 * due to delalloc, clear dirty metadata flags right before
460 * write_inode()
461 */
Dave Chinner250df6e2011-03-22 22:23:36 +1100462 spin_lock(&inode->i_lock);
Jan Kara6290be12012-05-03 14:47:57 +0200463 /* Clear I_DIRTY_PAGES if we've written out all dirty pages */
464 if (!mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
465 inode->i_state &= ~I_DIRTY_PAGES;
Dmitry Monakhov5547e8a2010-05-07 13:35:44 +0400466 dirty = inode->i_state & I_DIRTY;
467 inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC);
Dave Chinner250df6e2011-03-22 22:23:36 +1100468 spin_unlock(&inode->i_lock);
Christoph Hellwig26821ed2010-03-05 09:21:21 +0100469 /* Don't write the inode if only I_DIRTY_PAGES was set */
470 if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
Christoph Hellwiga9185b42010-03-05 09:21:37 +0100471 int err = write_inode(inode, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472 if (ret == 0)
473 ret = err;
474 }
Jan Kara4f8ad652012-05-03 14:48:00 +0200475 trace_writeback_single_inode(inode, wbc, nr_to_write);
476 return ret;
477}
478
479/*
480 * Write out an inode's dirty pages. Either the caller has an active reference
481 * on the inode or the inode has I_WILL_FREE set.
482 *
483 * This function is designed to be called for writing back one inode which
484 * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode()
485 * and does more profound writeback list handling in writeback_sb_inodes().
486 */
487static int
488writeback_single_inode(struct inode *inode, struct bdi_writeback *wb,
489 struct writeback_control *wbc)
490{
491 int ret = 0;
492
493 spin_lock(&inode->i_lock);
494 if (!atomic_read(&inode->i_count))
495 WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING)));
496 else
497 WARN_ON(inode->i_state & I_WILL_FREE);
498
499 if (inode->i_state & I_SYNC) {
500 if (wbc->sync_mode != WB_SYNC_ALL)
501 goto out;
502 /*
Jan Kara169ebd92012-05-03 14:48:03 +0200503 * It's a data-integrity sync. We must wait. Since callers hold
504 * inode reference or inode has I_WILL_FREE set, it cannot go
505 * away under us.
Jan Kara4f8ad652012-05-03 14:48:00 +0200506 */
Jan Kara169ebd92012-05-03 14:48:03 +0200507 __inode_wait_for_writeback(inode);
Jan Kara4f8ad652012-05-03 14:48:00 +0200508 }
509 WARN_ON(inode->i_state & I_SYNC);
510 /*
511 * Skip inode if it is clean. We don't want to mess with writeback
512 * lists in this function since flusher thread may be doing for example
513 * sync in parallel and if we move the inode, it could get skipped. So
514 * here we make sure inode is on some writeback list and leave it there
515 * unless we have completely cleaned the inode.
516 */
517 if (!(inode->i_state & I_DIRTY))
518 goto out;
519 inode->i_state |= I_SYNC;
520 spin_unlock(&inode->i_lock);
521
Yan Hongcd8ed2a2012-10-08 16:33:45 -0700522 ret = __writeback_single_inode(inode, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600524 spin_lock(&wb->list_lock);
Dave Chinner250df6e2011-03-22 22:23:36 +1100525 spin_lock(&inode->i_lock);
Jan Kara4f8ad652012-05-03 14:48:00 +0200526 /*
527 * If inode is clean, remove it from writeback lists. Otherwise don't
528 * touch it. See comment above for explanation.
529 */
530 if (!(inode->i_state & I_DIRTY))
531 list_del_init(&inode->i_wb_list);
532 spin_unlock(&wb->list_lock);
Joern Engel1c0eeaf2007-10-16 23:30:44 -0700533 inode_sync_complete(inode);
Jan Kara4f8ad652012-05-03 14:48:00 +0200534out:
535 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536 return ret;
537}
538
Wu Fengguang1a12d8b2010-08-29 13:28:09 -0600539static long writeback_chunk_size(struct backing_dev_info *bdi,
540 struct wb_writeback_work *work)
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600541{
542 long pages;
543
544 /*
545 * WB_SYNC_ALL mode does livelock avoidance by syncing dirty
546 * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX
547 * here avoids calling into writeback_inodes_wb() more than once.
548 *
549 * The intended call sequence for WB_SYNC_ALL writeback is:
550 *
551 * wb_writeback()
552 * writeback_sb_inodes() <== called only once
553 * write_cache_pages() <== called once for each inode
554 * (quickly) tag currently dirty pages
555 * (maybe slowly) sync all tagged pages
556 */
557 if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages)
558 pages = LONG_MAX;
Wu Fengguang1a12d8b2010-08-29 13:28:09 -0600559 else {
560 pages = min(bdi->avg_write_bandwidth / 2,
561 global_dirty_limit / DIRTY_SCOPE);
562 pages = min(pages, work->nr_pages);
563 pages = round_down(pages + MIN_WRITEBACK_PAGES,
564 MIN_WRITEBACK_PAGES);
565 }
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600566
567 return pages;
568}
569
Jens Axboe03ba3782009-09-09 09:08:54 +0200570/*
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800571 * Write a portion of b_io inodes which belong to @sb.
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200572 *
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600573 * Return the number of pages and/or inodes written.
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800574 */
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600575static long writeback_sb_inodes(struct super_block *sb,
576 struct bdi_writeback *wb,
577 struct wb_writeback_work *work)
Jens Axboe03ba3782009-09-09 09:08:54 +0200578{
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600579 struct writeback_control wbc = {
580 .sync_mode = work->sync_mode,
581 .tagged_writepages = work->tagged_writepages,
582 .for_kupdate = work->for_kupdate,
583 .for_background = work->for_background,
Dave Chinner7747bd42013-07-02 22:38:35 +1000584 .for_sync = work->for_sync,
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600585 .range_cyclic = work->range_cyclic,
586 .range_start = 0,
587 .range_end = LLONG_MAX,
588 };
589 unsigned long start_time = jiffies;
590 long write_chunk;
591 long wrote = 0; /* count both pages and inodes */
592
Jens Axboe03ba3782009-09-09 09:08:54 +0200593 while (!list_empty(&wb->b_io)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100594 struct inode *inode = wb_inode(wb->b_io.prev);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200595
596 if (inode->i_sb != sb) {
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600597 if (work->sb) {
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200598 /*
599 * We only want to write back data for this
600 * superblock, move all inodes not belonging
601 * to it back onto the dirty list.
602 */
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600603 redirty_tail(inode, wb);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200604 continue;
605 }
606
607 /*
608 * The inode belongs to a different superblock.
609 * Bounce back to the caller to unpin this and
610 * pin the next superblock.
611 */
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600612 break;
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200613 }
614
Christoph Hellwig9843b762010-10-24 19:40:46 +0200615 /*
Wanpeng Li331cbde2012-06-09 11:10:55 +0800616 * Don't bother with new inodes or inodes being freed, first
617 * kind does not need periodic writeout yet, and for the latter
Christoph Hellwig9843b762010-10-24 19:40:46 +0200618 * kind writeout is handled by the freer.
619 */
Dave Chinner250df6e2011-03-22 22:23:36 +1100620 spin_lock(&inode->i_lock);
Christoph Hellwig9843b762010-10-24 19:40:46 +0200621 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
Dave Chinner250df6e2011-03-22 22:23:36 +1100622 spin_unlock(&inode->i_lock);
Wu Fengguangfcc5c222011-07-11 23:08:50 -0700623 redirty_tail(inode, wb);
Nick Piggin7ef0d732009-03-12 14:31:38 -0700624 continue;
625 }
Jan Karacc1676d2012-05-03 14:47:56 +0200626 if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) {
627 /*
628 * If this inode is locked for writeback and we are not
629 * doing writeback-for-data-integrity, move it to
630 * b_more_io so that writeback can proceed with the
631 * other inodes on s_io.
632 *
633 * We'll have another go at writing back this inode
634 * when we completed a full scan of b_io.
635 */
636 spin_unlock(&inode->i_lock);
637 requeue_io(inode, wb);
638 trace_writeback_sb_inodes_requeue(inode);
639 continue;
640 }
Jan Karaf0d07b72012-05-03 14:47:59 +0200641 spin_unlock(&wb->list_lock);
642
Jan Kara4f8ad652012-05-03 14:48:00 +0200643 /*
644 * We already requeued the inode if it had I_SYNC set and we
645 * are doing WB_SYNC_NONE writeback. So this catches only the
646 * WB_SYNC_ALL case.
647 */
Jan Kara169ebd92012-05-03 14:48:03 +0200648 if (inode->i_state & I_SYNC) {
649 /* Wait for I_SYNC. This function drops i_lock... */
650 inode_sleep_on_writeback(inode);
651 /* Inode may be gone, start again */
Jan Karaead188f2012-06-08 17:07:36 +0200652 spin_lock(&wb->list_lock);
Jan Kara169ebd92012-05-03 14:48:03 +0200653 continue;
654 }
Jan Kara4f8ad652012-05-03 14:48:00 +0200655 inode->i_state |= I_SYNC;
656 spin_unlock(&inode->i_lock);
Jan Kara169ebd92012-05-03 14:48:03 +0200657
Wu Fengguang1a12d8b2010-08-29 13:28:09 -0600658 write_chunk = writeback_chunk_size(wb->bdi, work);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600659 wbc.nr_to_write = write_chunk;
660 wbc.pages_skipped = 0;
Dave Chinner250df6e2011-03-22 22:23:36 +1100661
Jan Kara169ebd92012-05-03 14:48:03 +0200662 /*
663 * We use I_SYNC to pin the inode in memory. While it is set
664 * evict_inode() will wait so the inode cannot be freed.
665 */
Yan Hongcd8ed2a2012-10-08 16:33:45 -0700666 __writeback_single_inode(inode, &wbc);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600667
668 work->nr_pages -= write_chunk - wbc.nr_to_write;
669 wrote += write_chunk - wbc.nr_to_write;
Jan Kara4f8ad652012-05-03 14:48:00 +0200670 spin_lock(&wb->list_lock);
671 spin_lock(&inode->i_lock);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600672 if (!(inode->i_state & I_DIRTY))
673 wrote++;
Jan Kara4f8ad652012-05-03 14:48:00 +0200674 requeue_inode(inode, wb, &wbc);
675 inode_sync_complete(inode);
Dave Chinner0f1b1fd2011-03-22 22:23:43 +1100676 spin_unlock(&inode->i_lock);
Jan Kara169ebd92012-05-03 14:48:03 +0200677 cond_resched_lock(&wb->list_lock);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600678 /*
679 * bail out to wb_writeback() often enough to check
680 * background threshold and other termination conditions.
681 */
682 if (wrote) {
683 if (time_is_before_jiffies(start_time + HZ / 10UL))
684 break;
685 if (work->nr_pages <= 0)
686 break;
Fengguang Wu8bc3be22008-02-04 22:29:36 -0800687 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688 }
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600689 return wrote;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800690}
Nick Piggin38f21972009-01-06 14:40:25 -0800691
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600692static long __writeback_inodes_wb(struct bdi_writeback *wb,
693 struct wb_writeback_work *work)
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800694{
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600695 unsigned long start_time = jiffies;
696 long wrote = 0;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800697
698 while (!list_empty(&wb->b_io)) {
Nick Piggin7ccf19a2010-10-21 11:49:30 +1100699 struct inode *inode = wb_inode(wb->b_io.prev);
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800700 struct super_block *sb = inode->i_sb;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800701
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000702 if (!grab_super_passive(sb)) {
Wu Fengguang0e9958162011-07-29 22:14:35 -0600703 /*
704 * grab_super_passive() may fail consistently due to
705 * s_umount being grabbed by someone else. Don't use
706 * requeue_io() to avoid busy retrying the inode/sb.
707 */
708 redirty_tail(inode, wb);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200709 continue;
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800710 }
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600711 wrote += writeback_sb_inodes(sb, wb, work);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200712 drop_super(sb);
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800713
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600714 /* refer to the same tests at the end of writeback_sb_inodes */
715 if (wrote) {
716 if (time_is_before_jiffies(start_time + HZ / 10UL))
717 break;
718 if (work->nr_pages <= 0)
719 break;
720 }
Edward Shishkinf11c9c52010-03-11 14:09:47 -0800721 }
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200722 /* Leave any unwritten inodes on b_io */
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600723 return wrote;
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200724}
725
Wanpeng Li7d9f0732013-09-11 14:22:40 -0700726static long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600727 enum wb_reason reason)
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200728{
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600729 struct wb_writeback_work work = {
730 .nr_pages = nr_pages,
731 .sync_mode = WB_SYNC_NONE,
732 .range_cyclic = 1,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600733 .reason = reason,
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600734 };
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200735
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600736 spin_lock(&wb->list_lock);
Wu Fengguang424b3512010-07-21 20:11:53 -0600737 if (list_empty(&wb->b_io))
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600738 queue_io(wb, &work);
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600739 __writeback_inodes_wb(wb, &work);
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600740 spin_unlock(&wb->list_lock);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200741
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600742 return nr_pages - work.nr_pages;
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200743}
744
Wu Fengguangb00949a2010-11-18 14:38:33 -0600745static bool over_bground_thresh(struct backing_dev_info *bdi)
Jens Axboe03ba3782009-09-09 09:08:54 +0200746{
747 unsigned long background_thresh, dirty_thresh;
748
Wu Fengguang16c40422010-08-11 14:17:39 -0700749 global_dirty_limits(&background_thresh, &dirty_thresh);
Jens Axboe03ba3782009-09-09 09:08:54 +0200750
Wu Fengguangb00949a2010-11-18 14:38:33 -0600751 if (global_page_state(NR_FILE_DIRTY) +
752 global_page_state(NR_UNSTABLE_NFS) > background_thresh)
753 return true;
754
755 if (bdi_stat(bdi, BDI_RECLAIMABLE) >
756 bdi_dirty_limit(bdi, background_thresh))
757 return true;
758
759 return false;
Jens Axboe03ba3782009-09-09 09:08:54 +0200760}
761
762/*
Wu Fengguange98be2d2010-08-29 11:22:30 -0600763 * Called under wb->list_lock. If there are multiple wb per bdi,
764 * only the flusher working on the first wb should do it.
765 */
766static void wb_update_bandwidth(struct bdi_writeback *wb,
767 unsigned long start_time)
768{
Wu Fengguangaf6a3112011-10-03 20:46:17 -0600769 __bdi_update_bandwidth(wb->bdi, 0, 0, 0, 0, 0, start_time);
Wu Fengguange98be2d2010-08-29 11:22:30 -0600770}
771
772/*
Jens Axboe03ba3782009-09-09 09:08:54 +0200773 * Explicit flushing or periodic writeback of "old" data.
774 *
775 * Define "old": the first time one of an inode's pages is dirtied, we mark the
776 * dirtying-time in the inode's address_space. So this periodic writeback code
777 * just walks the superblock inode list, writing back any inodes which are
778 * older than a specific point in time.
779 *
780 * Try to run once per dirty_writeback_interval. But if a writeback event
781 * takes longer than a dirty_writeback_interval interval, then leave a
782 * one-second gap.
783 *
784 * older_than_this takes precedence over nr_to_write. So we'll only write back
785 * all dirty pages if they are all attached to "old" mappings.
786 */
Jens Axboec4a77a62009-09-16 15:18:25 +0200787static long wb_writeback(struct bdi_writeback *wb,
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200788 struct wb_writeback_work *work)
Jens Axboe03ba3782009-09-09 09:08:54 +0200789{
Wu Fengguange98be2d2010-08-29 11:22:30 -0600790 unsigned long wb_start = jiffies;
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600791 long nr_pages = work->nr_pages;
Jens Axboe03ba3782009-09-09 09:08:54 +0200792 unsigned long oldest_jif;
Jan Karaa5989bd2009-09-16 19:22:48 +0200793 struct inode *inode;
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600794 long progress;
Jens Axboe03ba3782009-09-09 09:08:54 +0200795
Wu Fengguange185dda2011-04-23 11:26:07 -0600796 oldest_jif = jiffies;
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600797 work->older_than_this = &oldest_jif;
Jens Axboe03ba3782009-09-09 09:08:54 +0200798
Wu Fengguange8dfc302011-04-21 12:06:32 -0600799 spin_lock(&wb->list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200800 for (;;) {
801 /*
Wu Fengguangd3ddec72009-09-23 20:33:40 +0800802 * Stop writeback when nr_pages has been consumed
Jens Axboe03ba3782009-09-09 09:08:54 +0200803 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200804 if (work->nr_pages <= 0)
Jens Axboe03ba3782009-09-09 09:08:54 +0200805 break;
806
807 /*
Jan Karaaa373cf2011-01-13 15:45:47 -0800808 * Background writeout and kupdate-style writeback may
809 * run forever. Stop them if there is other work to do
810 * so that e.g. sync can proceed. They'll be restarted
811 * after the other works are all done.
812 */
813 if ((work->for_background || work->for_kupdate) &&
814 !list_empty(&wb->bdi->work_list))
815 break;
816
817 /*
Wu Fengguangd3ddec72009-09-23 20:33:40 +0800818 * For background writeout, stop when we are below the
819 * background dirty threshold
Jens Axboe03ba3782009-09-09 09:08:54 +0200820 */
Wu Fengguangb00949a2010-11-18 14:38:33 -0600821 if (work->for_background && !over_bground_thresh(wb->bdi))
Jens Axboe03ba3782009-09-09 09:08:54 +0200822 break;
823
Jan Kara1bc36b62011-10-19 11:44:41 +0200824 /*
825 * Kupdate and background works are special and we want to
826 * include all inodes that need writing. Livelock avoidance is
827 * handled by these works yielding to any other work so we are
828 * safe.
829 */
Wu Fengguangba9aa832010-07-21 20:32:30 -0600830 if (work->for_kupdate) {
831 oldest_jif = jiffies -
832 msecs_to_jiffies(dirty_expire_interval * 10);
Jan Kara1bc36b62011-10-19 11:44:41 +0200833 } else if (work->for_background)
834 oldest_jif = jiffies;
Dave Chinner028c2dd2010-07-07 13:24:07 +1000835
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600836 trace_writeback_start(wb->bdi, work);
Wu Fengguange8dfc302011-04-21 12:06:32 -0600837 if (list_empty(&wb->b_io))
Curt Wohlgemuthad4e38d2011-10-07 21:51:56 -0600838 queue_io(wb, work);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200839 if (work->sb)
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600840 progress = writeback_sb_inodes(work->sb, wb, work);
Christoph Hellwigedadfb12010-06-10 12:07:54 +0200841 else
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600842 progress = __writeback_inodes_wb(wb, work);
843 trace_writeback_written(wb->bdi, work);
Dave Chinner028c2dd2010-07-07 13:24:07 +1000844
Wu Fengguange98be2d2010-08-29 11:22:30 -0600845 wb_update_bandwidth(wb, wb_start);
Jens Axboe03ba3782009-09-09 09:08:54 +0200846
847 /*
Jens Axboe71fd05a2009-09-23 19:32:26 +0200848 * Did we write something? Try for more
Wu Fengguange6fb6da2010-07-22 10:23:44 -0600849 *
850 * Dirty inodes are moved to b_io for writeback in batches.
851 * The completion of the current batch does not necessarily
852 * mean the overall work is done. So we keep looping as long
853 * as made some progress on cleaning pages or inodes.
Jens Axboe71fd05a2009-09-23 19:32:26 +0200854 */
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600855 if (progress)
Jens Axboe71fd05a2009-09-23 19:32:26 +0200856 continue;
857 /*
Wu Fengguange6fb6da2010-07-22 10:23:44 -0600858 * No more inodes for IO, bail
Jens Axboe71fd05a2009-09-23 19:32:26 +0200859 */
Wu Fengguangb7a24412010-07-21 22:19:51 -0600860 if (list_empty(&wb->b_more_io))
Jens Axboe03ba3782009-09-09 09:08:54 +0200861 break;
862 /*
Jens Axboe8010c3b2009-09-15 20:04:57 +0200863 * Nothing written. Wait for some inode to
864 * become available for writeback. Otherwise
865 * we'll just busyloop.
Jens Axboe03ba3782009-09-09 09:08:54 +0200866 */
Jens Axboe03ba3782009-09-09 09:08:54 +0200867 if (!list_empty(&wb->b_more_io)) {
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600868 trace_writeback_wait(wb->bdi, work);
Jens Axboe03ba3782009-09-09 09:08:54 +0200869 inode = wb_inode(wb->b_more_io.prev);
Dave Chinner250df6e2011-03-22 22:23:36 +1100870 spin_lock(&inode->i_lock);
Jan Karaf0d07b72012-05-03 14:47:59 +0200871 spin_unlock(&wb->list_lock);
Jan Kara169ebd92012-05-03 14:48:03 +0200872 /* This function drops i_lock... */
873 inode_sleep_on_writeback(inode);
Jan Karaf0d07b72012-05-03 14:47:59 +0200874 spin_lock(&wb->list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200875 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200876 }
Wu Fengguange8dfc302011-04-21 12:06:32 -0600877 spin_unlock(&wb->list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200878
Wu Fengguangd46db3d2011-05-04 19:54:37 -0600879 return nr_pages - work->nr_pages;
Jens Axboe03ba3782009-09-09 09:08:54 +0200880}
881
882/*
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200883 * Return the next wb_writeback_work struct that hasn't been processed yet.
Jens Axboe03ba3782009-09-09 09:08:54 +0200884 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200885static struct wb_writeback_work *
Minchan Kim08852b62010-08-03 12:51:16 +0200886get_next_work_item(struct backing_dev_info *bdi)
Jens Axboe03ba3782009-09-09 09:08:54 +0200887{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200888 struct wb_writeback_work *work = NULL;
Jens Axboe03ba3782009-09-09 09:08:54 +0200889
Artem Bityutskiy64677162010-07-25 14:29:22 +0300890 spin_lock_bh(&bdi->wb_lock);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200891 if (!list_empty(&bdi->work_list)) {
892 work = list_entry(bdi->work_list.next,
893 struct wb_writeback_work, list);
894 list_del_init(&work->list);
Jens Axboe03ba3782009-09-09 09:08:54 +0200895 }
Artem Bityutskiy64677162010-07-25 14:29:22 +0300896 spin_unlock_bh(&bdi->wb_lock);
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200897 return work;
Jens Axboe03ba3782009-09-09 09:08:54 +0200898}
899
Linus Torvaldscdf01dd2010-10-30 08:55:52 -0700900/*
901 * Add in the number of potentially dirty inodes, because each inode
902 * write can dirty pagecache in the underlying blockdev.
903 */
904static unsigned long get_nr_dirty_pages(void)
905{
906 return global_page_state(NR_FILE_DIRTY) +
907 global_page_state(NR_UNSTABLE_NFS) +
908 get_nr_dirty_inodes();
909}
910
Jan Kara65850272011-01-13 15:45:44 -0800911static long wb_check_background_flush(struct bdi_writeback *wb)
912{
Wu Fengguangb00949a2010-11-18 14:38:33 -0600913 if (over_bground_thresh(wb->bdi)) {
Jan Kara65850272011-01-13 15:45:44 -0800914
915 struct wb_writeback_work work = {
916 .nr_pages = LONG_MAX,
917 .sync_mode = WB_SYNC_NONE,
918 .for_background = 1,
919 .range_cyclic = 1,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600920 .reason = WB_REASON_BACKGROUND,
Jan Kara65850272011-01-13 15:45:44 -0800921 };
922
923 return wb_writeback(wb, &work);
924 }
925
926 return 0;
927}
928
Jens Axboe03ba3782009-09-09 09:08:54 +0200929static long wb_check_old_data_flush(struct bdi_writeback *wb)
930{
931 unsigned long expired;
932 long nr_pages;
933
Jens Axboe69b62d02010-05-17 12:51:03 +0200934 /*
935 * When set to zero, disable periodic writeback
936 */
937 if (!dirty_writeback_interval)
938 return 0;
939
Jens Axboe03ba3782009-09-09 09:08:54 +0200940 expired = wb->last_old_flush +
941 msecs_to_jiffies(dirty_writeback_interval * 10);
942 if (time_before(jiffies, expired))
943 return 0;
944
945 wb->last_old_flush = jiffies;
Linus Torvaldscdf01dd2010-10-30 08:55:52 -0700946 nr_pages = get_nr_dirty_pages();
Jens Axboe03ba3782009-09-09 09:08:54 +0200947
Jens Axboec4a77a62009-09-16 15:18:25 +0200948 if (nr_pages) {
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200949 struct wb_writeback_work work = {
Jens Axboec4a77a62009-09-16 15:18:25 +0200950 .nr_pages = nr_pages,
951 .sync_mode = WB_SYNC_NONE,
952 .for_kupdate = 1,
953 .range_cyclic = 1,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600954 .reason = WB_REASON_PERIODIC,
Jens Axboec4a77a62009-09-16 15:18:25 +0200955 };
956
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200957 return wb_writeback(wb, &work);
Jens Axboec4a77a62009-09-16 15:18:25 +0200958 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200959
960 return 0;
961}
962
963/*
964 * Retrieve work items and do the writeback they describe
965 */
Wanpeng Li25d130b2013-07-08 16:00:14 -0700966static long wb_do_writeback(struct bdi_writeback *wb)
Jens Axboe03ba3782009-09-09 09:08:54 +0200967{
968 struct backing_dev_info *bdi = wb->bdi;
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200969 struct wb_writeback_work *work;
Jens Axboec4a77a62009-09-16 15:18:25 +0200970 long wrote = 0;
Jens Axboe03ba3782009-09-09 09:08:54 +0200971
Jan Kara81d73a32010-08-11 14:17:44 -0700972 set_bit(BDI_writeback_running, &wb->bdi->state);
Minchan Kim08852b62010-08-03 12:51:16 +0200973 while ((work = get_next_work_item(bdi)) != NULL) {
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200974
Dave Chinner455b2862010-07-07 13:24:06 +1000975 trace_writeback_exec(bdi, work);
976
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200977 wrote += wb_writeback(wb, work);
Jens Axboe03ba3782009-09-09 09:08:54 +0200978
979 /*
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200980 * Notify the caller of completion if this is a synchronous
981 * work item, otherwise just free it.
Jens Axboe03ba3782009-09-09 09:08:54 +0200982 */
Christoph Hellwig83ba7b02010-07-06 08:59:53 +0200983 if (work->done)
984 complete(work->done);
985 else
986 kfree(work);
Jens Axboe03ba3782009-09-09 09:08:54 +0200987 }
988
989 /*
990 * Check for periodic writeback, kupdated() style
991 */
992 wrote += wb_check_old_data_flush(wb);
Jan Kara65850272011-01-13 15:45:44 -0800993 wrote += wb_check_background_flush(wb);
Jan Kara81d73a32010-08-11 14:17:44 -0700994 clear_bit(BDI_writeback_running, &wb->bdi->state);
Jens Axboe03ba3782009-09-09 09:08:54 +0200995
996 return wrote;
997}
998
999/*
1000 * Handle writeback of dirty data for the device backed by this bdi. Also
Tejun Heo839a8e82013-04-01 19:08:06 -07001001 * reschedules periodically and does kupdated style flushing.
Jens Axboe03ba3782009-09-09 09:08:54 +02001002 */
Tejun Heo839a8e82013-04-01 19:08:06 -07001003void bdi_writeback_workfn(struct work_struct *work)
Jens Axboe03ba3782009-09-09 09:08:54 +02001004{
Tejun Heo839a8e82013-04-01 19:08:06 -07001005 struct bdi_writeback *wb = container_of(to_delayed_work(work),
1006 struct bdi_writeback, dwork);
Christoph Hellwig08243902010-06-19 23:08:22 +02001007 struct backing_dev_info *bdi = wb->bdi;
Jens Axboe03ba3782009-09-09 09:08:54 +02001008 long pages_written;
1009
Tejun Heoef3b1012013-04-30 15:27:24 -07001010 set_worker_desc("flush-%s", dev_name(bdi->dev));
Peter Zijlstra766f9162010-10-26 14:22:45 -07001011 current->flags |= PF_SWAPWRITE;
Christoph Hellwig08243902010-06-19 23:08:22 +02001012
Tejun Heo839a8e82013-04-01 19:08:06 -07001013 if (likely(!current_is_workqueue_rescuer() ||
1014 list_empty(&bdi->bdi_list))) {
Artem Bityutskiy64677162010-07-25 14:29:22 +03001015 /*
Tejun Heo839a8e82013-04-01 19:08:06 -07001016 * The normal path. Keep writing back @bdi until its
1017 * work_list is empty. Note that this path is also taken
1018 * if @bdi is shutting down even when we're running off the
1019 * rescuer as work_list needs to be drained.
Artem Bityutskiy64677162010-07-25 14:29:22 +03001020 */
Tejun Heo839a8e82013-04-01 19:08:06 -07001021 do {
Wanpeng Li25d130b2013-07-08 16:00:14 -07001022 pages_written = wb_do_writeback(wb);
Tejun Heo839a8e82013-04-01 19:08:06 -07001023 trace_writeback_pages_written(pages_written);
1024 } while (!list_empty(&bdi->work_list));
1025 } else {
1026 /*
1027 * bdi_wq can't get enough workers and we're running off
1028 * the emergency worker. Don't hog it. Hopefully, 1024 is
1029 * enough for efficient IO.
1030 */
1031 pages_written = writeback_inodes_wb(&bdi->wb, 1024,
1032 WB_REASON_FORKER_THREAD);
Dave Chinner455b2862010-07-07 13:24:06 +10001033 trace_writeback_pages_written(pages_written);
Jens Axboe03ba3782009-09-09 09:08:54 +02001034 }
1035
Tejun Heo839a8e82013-04-01 19:08:06 -07001036 if (!list_empty(&bdi->work_list) ||
1037 (wb_has_dirty_io(wb) && dirty_writeback_interval))
1038 queue_delayed_work(bdi_wq, &wb->dwork,
1039 msecs_to_jiffies(dirty_writeback_interval * 10));
Dave Chinner455b2862010-07-07 13:24:06 +10001040
Tejun Heo839a8e82013-04-01 19:08:06 -07001041 current->flags &= ~PF_SWAPWRITE;
Jens Axboe03ba3782009-09-09 09:08:54 +02001042}
1043
1044/*
Jens Axboe03ba3782009-09-09 09:08:54 +02001045 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
1046 * the whole world.
1047 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001048void wakeup_flusher_threads(long nr_pages, enum wb_reason reason)
Jens Axboe03ba3782009-09-09 09:08:54 +02001049{
Christoph Hellwigb8c2f342010-06-08 18:15:07 +02001050 struct backing_dev_info *bdi;
Christoph Hellwigb8c2f342010-06-08 18:15:07 +02001051
Jan Kara47df3dd2013-09-11 14:22:22 -07001052 if (!nr_pages)
1053 nr_pages = get_nr_dirty_pages();
Christoph Hellwigb8c2f342010-06-08 18:15:07 +02001054
1055 rcu_read_lock();
1056 list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) {
1057 if (!bdi_has_dirty_io(bdi))
1058 continue;
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001059 __bdi_start_writeback(bdi, nr_pages, false, reason);
Christoph Hellwigb8c2f342010-06-08 18:15:07 +02001060 }
1061 rcu_read_unlock();
Jens Axboe03ba3782009-09-09 09:08:54 +02001062}
1063
1064static noinline void block_dump___mark_inode_dirty(struct inode *inode)
1065{
1066 if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) {
1067 struct dentry *dentry;
1068 const char *name = "?";
1069
1070 dentry = d_find_alias(inode);
1071 if (dentry) {
1072 spin_lock(&dentry->d_lock);
1073 name = (const char *) dentry->d_name.name;
1074 }
1075 printk(KERN_DEBUG
1076 "%s(%d): dirtied inode %lu (%s) on %s\n",
1077 current->comm, task_pid_nr(current), inode->i_ino,
1078 name, inode->i_sb->s_id);
1079 if (dentry) {
1080 spin_unlock(&dentry->d_lock);
1081 dput(dentry);
1082 }
1083 }
1084}
1085
1086/**
1087 * __mark_inode_dirty - internal function
1088 * @inode: inode to mark
1089 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
1090 * Mark an inode as dirty. Callers should use mark_inode_dirty or
1091 * mark_inode_dirty_sync.
1092 *
1093 * Put the inode on the super block's dirty list.
1094 *
1095 * CAREFUL! We mark it dirty unconditionally, but move it onto the
1096 * dirty list only if it is hashed or if it refers to a blockdev.
1097 * If it was not hashed, it will never be added to the dirty list
1098 * even if it is later hashed, as it will have been marked dirty already.
1099 *
1100 * In short, make sure you hash any inodes _before_ you start marking
1101 * them dirty.
1102 *
Jens Axboe03ba3782009-09-09 09:08:54 +02001103 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
1104 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
1105 * the kernel-internal blockdev inode represents the dirtying time of the
1106 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
1107 * page->mapping->host, so the page-dirtying time is recorded in the internal
1108 * blockdev inode.
1109 */
1110void __mark_inode_dirty(struct inode *inode, int flags)
1111{
1112 struct super_block *sb = inode->i_sb;
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001113 struct backing_dev_info *bdi = NULL;
Jens Axboe03ba3782009-09-09 09:08:54 +02001114
1115 /*
1116 * Don't do this for I_DIRTY_PAGES - that doesn't actually
1117 * dirty the inode itself
1118 */
1119 if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
Tejun Heo9fb0a7d2013-01-11 13:06:37 -08001120 trace_writeback_dirty_inode_start(inode, flags);
1121
Jens Axboe03ba3782009-09-09 09:08:54 +02001122 if (sb->s_op->dirty_inode)
Christoph Hellwigaa385722011-05-27 06:53:02 -04001123 sb->s_op->dirty_inode(inode, flags);
Tejun Heo9fb0a7d2013-01-11 13:06:37 -08001124
1125 trace_writeback_dirty_inode(inode, flags);
Jens Axboe03ba3782009-09-09 09:08:54 +02001126 }
1127
1128 /*
1129 * make sure that changes are seen by all cpus before we test i_state
1130 * -- mikulas
1131 */
1132 smp_mb();
1133
1134 /* avoid the locking if we can */
1135 if ((inode->i_state & flags) == flags)
1136 return;
1137
1138 if (unlikely(block_dump))
1139 block_dump___mark_inode_dirty(inode);
1140
Dave Chinner250df6e2011-03-22 22:23:36 +11001141 spin_lock(&inode->i_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001142 if ((inode->i_state & flags) != flags) {
1143 const int was_dirty = inode->i_state & I_DIRTY;
1144
1145 inode->i_state |= flags;
1146
1147 /*
1148 * If the inode is being synced, just update its dirty state.
1149 * The unlocker will place the inode on the appropriate
1150 * superblock list, based upon its state.
1151 */
1152 if (inode->i_state & I_SYNC)
Dave Chinner250df6e2011-03-22 22:23:36 +11001153 goto out_unlock_inode;
Jens Axboe03ba3782009-09-09 09:08:54 +02001154
1155 /*
1156 * Only add valid (hashed) inodes to the superblock's
1157 * dirty list. Add blockdev inodes as well.
1158 */
1159 if (!S_ISBLK(inode->i_mode)) {
Al Viro1d3382cb2010-10-23 15:19:20 -04001160 if (inode_unhashed(inode))
Dave Chinner250df6e2011-03-22 22:23:36 +11001161 goto out_unlock_inode;
Jens Axboe03ba3782009-09-09 09:08:54 +02001162 }
Al Viroa4ffdde2010-06-02 17:38:30 -04001163 if (inode->i_state & I_FREEING)
Dave Chinner250df6e2011-03-22 22:23:36 +11001164 goto out_unlock_inode;
Jens Axboe03ba3782009-09-09 09:08:54 +02001165
1166 /*
1167 * If the inode was already on b_dirty/b_io/b_more_io, don't
1168 * reposition it (that would break b_dirty time-ordering).
1169 */
1170 if (!was_dirty) {
Dave Chinnera66979a2011-03-22 22:23:41 +11001171 bool wakeup_bdi = false;
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001172 bdi = inode_to_bdi(inode);
Jens Axboe500b0672009-09-09 09:10:25 +02001173
Junxiao Bi146d7002013-09-11 14:23:04 -07001174 spin_unlock(&inode->i_lock);
1175 spin_lock(&bdi->wb.list_lock);
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001176 if (bdi_cap_writeback_dirty(bdi)) {
1177 WARN(!test_bit(BDI_registered, &bdi->state),
1178 "bdi-%s not registered\n", bdi->name);
1179
1180 /*
1181 * If this is the first dirty inode for this
1182 * bdi, we have to wake-up the corresponding
1183 * bdi thread to make sure background
1184 * write-back happens later.
1185 */
1186 if (!wb_has_dirty_io(&bdi->wb))
1187 wakeup_bdi = true;
Jens Axboe500b0672009-09-09 09:10:25 +02001188 }
Jens Axboe03ba3782009-09-09 09:08:54 +02001189
1190 inode->dirtied_when = jiffies;
Nick Piggin7ccf19a2010-10-21 11:49:30 +11001191 list_move(&inode->i_wb_list, &bdi->wb.b_dirty);
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001192 spin_unlock(&bdi->wb.list_lock);
Dave Chinnera66979a2011-03-22 22:23:41 +11001193
1194 if (wakeup_bdi)
1195 bdi_wakeup_thread_delayed(bdi);
1196 return;
Jens Axboe03ba3782009-09-09 09:08:54 +02001197 }
1198 }
Dave Chinner250df6e2011-03-22 22:23:36 +11001199out_unlock_inode:
1200 spin_unlock(&inode->i_lock);
Artem Bityutskiy253c34e2010-07-25 14:29:21 +03001201
Jens Axboe03ba3782009-09-09 09:08:54 +02001202}
1203EXPORT_SYMBOL(__mark_inode_dirty);
1204
Jens Axboeb6e51312009-09-16 15:13:54 +02001205static void wait_sb_inodes(struct super_block *sb)
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001206{
Jens Axboe03ba3782009-09-09 09:08:54 +02001207 struct inode *inode, *old_inode = NULL;
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001208
Jens Axboe03ba3782009-09-09 09:08:54 +02001209 /*
1210 * We need to be protected against the filesystem going from
1211 * r/o to r/w or vice versa.
1212 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001213 WARN_ON(!rwsem_is_locked(&sb->s_umount));
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001214
Dave Chinner55fa6092011-03-22 22:23:40 +11001215 spin_lock(&inode_sb_list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001216
1217 /*
1218 * Data integrity sync. Must wait for all pages under writeback,
1219 * because there may have been pages dirtied before our sync
1220 * call, but which had writeout started before we write it out.
1221 * In which case, the inode may not be on the dirty list, but
1222 * we still have to wait for that writeout.
1223 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001224 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
Dave Chinner250df6e2011-03-22 22:23:36 +11001225 struct address_space *mapping = inode->i_mapping;
Jens Axboe03ba3782009-09-09 09:08:54 +02001226
Dave Chinner250df6e2011-03-22 22:23:36 +11001227 spin_lock(&inode->i_lock);
1228 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
1229 (mapping->nrpages == 0)) {
1230 spin_unlock(&inode->i_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001231 continue;
Dave Chinner250df6e2011-03-22 22:23:36 +11001232 }
Jens Axboe03ba3782009-09-09 09:08:54 +02001233 __iget(inode);
Dave Chinner250df6e2011-03-22 22:23:36 +11001234 spin_unlock(&inode->i_lock);
Dave Chinner55fa6092011-03-22 22:23:40 +11001235 spin_unlock(&inode_sb_list_lock);
1236
Jens Axboe03ba3782009-09-09 09:08:54 +02001237 /*
Dave Chinner55fa6092011-03-22 22:23:40 +11001238 * We hold a reference to 'inode' so it couldn't have been
1239 * removed from s_inodes list while we dropped the
1240 * inode_sb_list_lock. We cannot iput the inode now as we can
1241 * be holding the last reference and we cannot iput it under
1242 * inode_sb_list_lock. So we keep the reference and iput it
1243 * later.
Jens Axboe03ba3782009-09-09 09:08:54 +02001244 */
1245 iput(old_inode);
1246 old_inode = inode;
1247
1248 filemap_fdatawait(mapping);
1249
1250 cond_resched();
Nick Piggin38f21972009-01-06 14:40:25 -08001251
Dave Chinner55fa6092011-03-22 22:23:40 +11001252 spin_lock(&inode_sb_list_lock);
Jens Axboe66f3b8e2009-09-02 09:19:46 +02001253 }
Dave Chinner55fa6092011-03-22 22:23:40 +11001254 spin_unlock(&inode_sb_list_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +02001255 iput(old_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256}
1257
Jens Axboed8a85592009-09-02 12:34:32 +02001258/**
Chris Mason3259f8b2010-10-29 11:16:17 -04001259 * writeback_inodes_sb_nr - writeback dirty inodes from given super_block
Jens Axboed8a85592009-09-02 12:34:32 +02001260 * @sb: the superblock
Chris Mason3259f8b2010-10-29 11:16:17 -04001261 * @nr: the number of pages to write
Marcos Paulo de Souza786228a2011-11-23 20:56:45 +08001262 * @reason: reason why some writeback work initiated
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263 *
Jens Axboed8a85592009-09-02 12:34:32 +02001264 * Start writeback on some inodes on this super_block. No guarantees are made
1265 * on how many (if any) will be written, and this function does not wait
Chris Mason3259f8b2010-10-29 11:16:17 -04001266 * for IO completion of submitted IO.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001268void writeback_inodes_sb_nr(struct super_block *sb,
1269 unsigned long nr,
1270 enum wb_reason reason)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001272 DECLARE_COMPLETION_ONSTACK(done);
1273 struct wb_writeback_work work = {
Wu Fengguang6e6938b2010-06-06 10:38:15 -06001274 .sb = sb,
1275 .sync_mode = WB_SYNC_NONE,
1276 .tagged_writepages = 1,
1277 .done = &done,
1278 .nr_pages = nr,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001279 .reason = reason,
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001280 };
Jens Axboe0e3c9a22010-06-01 11:08:43 +02001281
Jan Kara6eedc702012-07-03 16:45:27 +02001282 if (sb->s_bdi == &noop_backing_dev_info)
1283 return;
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001284 WARN_ON(!rwsem_is_locked(&sb->s_umount));
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001285 bdi_queue_work(sb->s_bdi, &work);
1286 wait_for_completion(&done);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287}
Chris Mason3259f8b2010-10-29 11:16:17 -04001288EXPORT_SYMBOL(writeback_inodes_sb_nr);
1289
1290/**
1291 * writeback_inodes_sb - writeback dirty inodes from given super_block
1292 * @sb: the superblock
Marcos Paulo de Souza786228a2011-11-23 20:56:45 +08001293 * @reason: reason why some writeback work was initiated
Chris Mason3259f8b2010-10-29 11:16:17 -04001294 *
1295 * Start writeback on some inodes on this super_block. No guarantees are made
1296 * on how many (if any) will be written, and this function does not wait
1297 * for IO completion of submitted IO.
1298 */
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001299void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason)
Chris Mason3259f8b2010-10-29 11:16:17 -04001300{
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001301 return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason);
Chris Mason3259f8b2010-10-29 11:16:17 -04001302}
Jens Axboed8a85592009-09-02 12:34:32 +02001303EXPORT_SYMBOL(writeback_inodes_sb);
1304
1305/**
Miao Xie10ee27a2013-01-10 13:47:57 +08001306 * try_to_writeback_inodes_sb_nr - try to start writeback if none underway
Chris Mason3259f8b2010-10-29 11:16:17 -04001307 * @sb: the superblock
1308 * @nr: the number of pages to write
Miao Xie10ee27a2013-01-10 13:47:57 +08001309 * @reason: the reason of writeback
Chris Mason3259f8b2010-10-29 11:16:17 -04001310 *
Miao Xie10ee27a2013-01-10 13:47:57 +08001311 * Invoke writeback_inodes_sb_nr if no writeback is currently underway.
Chris Mason3259f8b2010-10-29 11:16:17 -04001312 * Returns 1 if writeback was started, 0 if not.
1313 */
Miao Xie10ee27a2013-01-10 13:47:57 +08001314int try_to_writeback_inodes_sb_nr(struct super_block *sb,
1315 unsigned long nr,
1316 enum wb_reason reason)
Chris Mason3259f8b2010-10-29 11:16:17 -04001317{
Miao Xie10ee27a2013-01-10 13:47:57 +08001318 if (writeback_in_progress(sb->s_bdi))
Chris Mason3259f8b2010-10-29 11:16:17 -04001319 return 1;
Miao Xie10ee27a2013-01-10 13:47:57 +08001320
1321 if (!down_read_trylock(&sb->s_umount))
Chris Mason3259f8b2010-10-29 11:16:17 -04001322 return 0;
Miao Xie10ee27a2013-01-10 13:47:57 +08001323
1324 writeback_inodes_sb_nr(sb, nr, reason);
1325 up_read(&sb->s_umount);
1326 return 1;
Chris Mason3259f8b2010-10-29 11:16:17 -04001327}
Miao Xie10ee27a2013-01-10 13:47:57 +08001328EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr);
1329
1330/**
1331 * try_to_writeback_inodes_sb - try to start writeback if none underway
1332 * @sb: the superblock
1333 * @reason: reason why some writeback work was initiated
1334 *
1335 * Implement by try_to_writeback_inodes_sb_nr()
1336 * Returns 1 if writeback was started, 0 if not.
1337 */
1338int try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason)
1339{
1340 return try_to_writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason);
1341}
1342EXPORT_SYMBOL(try_to_writeback_inodes_sb);
Chris Mason3259f8b2010-10-29 11:16:17 -04001343
1344/**
Jens Axboed8a85592009-09-02 12:34:32 +02001345 * sync_inodes_sb - sync sb inode pages
1346 * @sb: the superblock
1347 *
1348 * This function writes and waits on any dirty inode belonging to this
Stefan Hajnoczicb9ef8d2011-01-13 15:47:26 -08001349 * super_block.
Jens Axboed8a85592009-09-02 12:34:32 +02001350 */
Jens Axboeb6e51312009-09-16 15:13:54 +02001351void sync_inodes_sb(struct super_block *sb)
Jens Axboed8a85592009-09-02 12:34:32 +02001352{
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001353 DECLARE_COMPLETION_ONSTACK(done);
1354 struct wb_writeback_work work = {
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001355 .sb = sb,
1356 .sync_mode = WB_SYNC_ALL,
1357 .nr_pages = LONG_MAX,
1358 .range_cyclic = 0,
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001359 .done = &done,
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -06001360 .reason = WB_REASON_SYNC,
Dave Chinner7747bd42013-07-02 22:38:35 +10001361 .for_sync = 1,
Christoph Hellwig3c4d7162010-06-08 18:14:43 +02001362 };
1363
Jan Kara6eedc702012-07-03 16:45:27 +02001364 /* Nothing to do? */
1365 if (sb->s_bdi == &noop_backing_dev_info)
1366 return;
Christoph Hellwigcf37e972010-06-08 18:14:51 +02001367 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1368
Christoph Hellwig83ba7b02010-07-06 08:59:53 +02001369 bdi_queue_work(sb->s_bdi, &work);
1370 wait_for_completion(&done);
1371
Jens Axboeb6e51312009-09-16 15:13:54 +02001372 wait_sb_inodes(sb);
Jens Axboed8a85592009-09-02 12:34:32 +02001373}
1374EXPORT_SYMBOL(sync_inodes_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376/**
Andrea Arcangeli7f04c262005-10-30 15:03:05 -08001377 * write_inode_now - write an inode to disk
1378 * @inode: inode to write to disk
1379 * @sync: whether the write should be synchronous or not
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380 *
Andrea Arcangeli7f04c262005-10-30 15:03:05 -08001381 * This function commits an inode to disk immediately if it is dirty. This is
1382 * primarily needed by knfsd.
1383 *
1384 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386int write_inode_now(struct inode *inode, int sync)
1387{
Christoph Hellwigf758eea2011-04-21 18:19:44 -06001388 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389 struct writeback_control wbc = {
1390 .nr_to_write = LONG_MAX,
Mike Galbraith18914b12008-02-08 04:20:23 -08001391 .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -07001392 .range_start = 0,
1393 .range_end = LLONG_MAX,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394 };
1395
1396 if (!mapping_cap_writeback_dirty(inode->i_mapping))
Andrew Morton49364ce2005-11-07 00:59:15 -08001397 wbc.nr_to_write = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398
1399 might_sleep();
Jan Kara4f8ad652012-05-03 14:48:00 +02001400 return writeback_single_inode(inode, wb, &wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401}
1402EXPORT_SYMBOL(write_inode_now);
1403
1404/**
1405 * sync_inode - write an inode and its pages to disk.
1406 * @inode: the inode to sync
1407 * @wbc: controls the writeback mode
1408 *
1409 * sync_inode() will write an inode and its pages to disk. It will also
1410 * correctly update the inode on its superblock's dirty inode lists and will
1411 * update inode->i_state.
1412 *
1413 * The caller must have a ref on the inode.
1414 */
1415int sync_inode(struct inode *inode, struct writeback_control *wbc)
1416{
Jan Kara4f8ad652012-05-03 14:48:00 +02001417 return writeback_single_inode(inode, &inode_to_bdi(inode)->wb, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418}
1419EXPORT_SYMBOL(sync_inode);
Christoph Hellwigc37650162010-10-06 10:48:20 +02001420
1421/**
Andrew Mortonc691b9d2011-01-13 15:45:48 -08001422 * sync_inode_metadata - write an inode to disk
Christoph Hellwigc37650162010-10-06 10:48:20 +02001423 * @inode: the inode to sync
1424 * @wait: wait for I/O to complete.
1425 *
Andrew Mortonc691b9d2011-01-13 15:45:48 -08001426 * Write an inode to disk and adjust its dirty state after completion.
Christoph Hellwigc37650162010-10-06 10:48:20 +02001427 *
1428 * Note: only writes the actual inode, no associated data or other metadata.
1429 */
1430int sync_inode_metadata(struct inode *inode, int wait)
1431{
1432 struct writeback_control wbc = {
1433 .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE,
1434 .nr_to_write = 0, /* metadata-only */
1435 };
1436
1437 return sync_inode(inode, &wbc);
1438}
1439EXPORT_SYMBOL(sync_inode_metadata);